]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sys/dev/cxgbe/t4_main.c
Track device's NUMA domain in ifnet & alloc ifnet from NUMA local memory
[FreeBSD/FreeBSD.git] / sys / dev / cxgbe / t4_main.c
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2011 Chelsio Communications, Inc.
5  * All rights reserved.
6  * Written by: Navdeep Parhar <np@FreeBSD.org>
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32
33 #include "opt_ddb.h"
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 #include "opt_ratelimit.h"
37 #include "opt_rss.h"
38
39 #include <sys/param.h>
40 #include <sys/conf.h>
41 #include <sys/priv.h>
42 #include <sys/kernel.h>
43 #include <sys/bus.h>
44 #include <sys/module.h>
45 #include <sys/malloc.h>
46 #include <sys/queue.h>
47 #include <sys/taskqueue.h>
48 #include <sys/pciio.h>
49 #include <dev/pci/pcireg.h>
50 #include <dev/pci/pcivar.h>
51 #include <dev/pci/pci_private.h>
52 #include <sys/firmware.h>
53 #include <sys/sbuf.h>
54 #include <sys/smp.h>
55 #include <sys/socket.h>
56 #include <sys/sockio.h>
57 #include <sys/sysctl.h>
58 #include <net/ethernet.h>
59 #include <net/if.h>
60 #include <net/if_types.h>
61 #include <net/if_dl.h>
62 #include <net/if_vlan_var.h>
63 #ifdef RSS
64 #include <net/rss_config.h>
65 #endif
66 #include <netinet/in.h>
67 #include <netinet/ip.h>
68 #if defined(__i386__) || defined(__amd64__)
69 #include <machine/md_var.h>
70 #include <machine/cputypes.h>
71 #include <vm/vm.h>
72 #include <vm/pmap.h>
73 #endif
74 #include <crypto/rijndael/rijndael.h>
75 #ifdef DDB
76 #include <ddb/ddb.h>
77 #include <ddb/db_lex.h>
78 #endif
79
80 #include "common/common.h"
81 #include "common/t4_msg.h"
82 #include "common/t4_regs.h"
83 #include "common/t4_regs_values.h"
84 #include "cudbg/cudbg.h"
85 #include "t4_clip.h"
86 #include "t4_ioctl.h"
87 #include "t4_l2t.h"
88 #include "t4_mp_ring.h"
89 #include "t4_if.h"
90 #include "t4_smt.h"
91
92 /* T4 bus driver interface */
93 static int t4_probe(device_t);
94 static int t4_attach(device_t);
95 static int t4_detach(device_t);
96 static int t4_child_location_str(device_t, device_t, char *, size_t);
97 static int t4_ready(device_t);
98 static int t4_read_port_device(device_t, int, device_t *);
99 static device_method_t t4_methods[] = {
100         DEVMETHOD(device_probe,         t4_probe),
101         DEVMETHOD(device_attach,        t4_attach),
102         DEVMETHOD(device_detach,        t4_detach),
103
104         DEVMETHOD(bus_child_location_str, t4_child_location_str),
105
106         DEVMETHOD(t4_is_main_ready,     t4_ready),
107         DEVMETHOD(t4_read_port_device,  t4_read_port_device),
108
109         DEVMETHOD_END
110 };
111 static driver_t t4_driver = {
112         "t4nex",
113         t4_methods,
114         sizeof(struct adapter)
115 };
116
117
118 /* T4 port (cxgbe) interface */
119 static int cxgbe_probe(device_t);
120 static int cxgbe_attach(device_t);
121 static int cxgbe_detach(device_t);
122 device_method_t cxgbe_methods[] = {
123         DEVMETHOD(device_probe,         cxgbe_probe),
124         DEVMETHOD(device_attach,        cxgbe_attach),
125         DEVMETHOD(device_detach,        cxgbe_detach),
126         { 0, 0 }
127 };
128 static driver_t cxgbe_driver = {
129         "cxgbe",
130         cxgbe_methods,
131         sizeof(struct port_info)
132 };
133
134 /* T4 VI (vcxgbe) interface */
135 static int vcxgbe_probe(device_t);
136 static int vcxgbe_attach(device_t);
137 static int vcxgbe_detach(device_t);
138 static device_method_t vcxgbe_methods[] = {
139         DEVMETHOD(device_probe,         vcxgbe_probe),
140         DEVMETHOD(device_attach,        vcxgbe_attach),
141         DEVMETHOD(device_detach,        vcxgbe_detach),
142         { 0, 0 }
143 };
144 static driver_t vcxgbe_driver = {
145         "vcxgbe",
146         vcxgbe_methods,
147         sizeof(struct vi_info)
148 };
149
150 static d_ioctl_t t4_ioctl;
151
152 static struct cdevsw t4_cdevsw = {
153        .d_version = D_VERSION,
154        .d_ioctl = t4_ioctl,
155        .d_name = "t4nex",
156 };
157
158 /* T5 bus driver interface */
159 static int t5_probe(device_t);
160 static device_method_t t5_methods[] = {
161         DEVMETHOD(device_probe,         t5_probe),
162         DEVMETHOD(device_attach,        t4_attach),
163         DEVMETHOD(device_detach,        t4_detach),
164
165         DEVMETHOD(bus_child_location_str, t4_child_location_str),
166
167         DEVMETHOD(t4_is_main_ready,     t4_ready),
168         DEVMETHOD(t4_read_port_device,  t4_read_port_device),
169
170         DEVMETHOD_END
171 };
172 static driver_t t5_driver = {
173         "t5nex",
174         t5_methods,
175         sizeof(struct adapter)
176 };
177
178
179 /* T5 port (cxl) interface */
180 static driver_t cxl_driver = {
181         "cxl",
182         cxgbe_methods,
183         sizeof(struct port_info)
184 };
185
186 /* T5 VI (vcxl) interface */
187 static driver_t vcxl_driver = {
188         "vcxl",
189         vcxgbe_methods,
190         sizeof(struct vi_info)
191 };
192
193 /* T6 bus driver interface */
194 static int t6_probe(device_t);
195 static device_method_t t6_methods[] = {
196         DEVMETHOD(device_probe,         t6_probe),
197         DEVMETHOD(device_attach,        t4_attach),
198         DEVMETHOD(device_detach,        t4_detach),
199
200         DEVMETHOD(bus_child_location_str, t4_child_location_str),
201
202         DEVMETHOD(t4_is_main_ready,     t4_ready),
203         DEVMETHOD(t4_read_port_device,  t4_read_port_device),
204
205         DEVMETHOD_END
206 };
207 static driver_t t6_driver = {
208         "t6nex",
209         t6_methods,
210         sizeof(struct adapter)
211 };
212
213
214 /* T6 port (cc) interface */
215 static driver_t cc_driver = {
216         "cc",
217         cxgbe_methods,
218         sizeof(struct port_info)
219 };
220
221 /* T6 VI (vcc) interface */
222 static driver_t vcc_driver = {
223         "vcc",
224         vcxgbe_methods,
225         sizeof(struct vi_info)
226 };
227
228 /* ifnet interface */
229 static void cxgbe_init(void *);
230 static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
231 static int cxgbe_transmit(struct ifnet *, struct mbuf *);
232 static void cxgbe_qflush(struct ifnet *);
233
234 MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
235
236 /*
237  * Correct lock order when you need to acquire multiple locks is t4_list_lock,
238  * then ADAPTER_LOCK, then t4_uld_list_lock.
239  */
240 static struct sx t4_list_lock;
241 SLIST_HEAD(, adapter) t4_list;
242 #ifdef TCP_OFFLOAD
243 static struct sx t4_uld_list_lock;
244 SLIST_HEAD(, uld_info) t4_uld_list;
245 #endif
246
247 /*
248  * Tunables.  See tweak_tunables() too.
249  *
250  * Each tunable is set to a default value here if it's known at compile-time.
251  * Otherwise it is set to -n as an indication to tweak_tunables() that it should
252  * provide a reasonable default (upto n) when the driver is loaded.
253  *
254  * Tunables applicable to both T4 and T5 are under hw.cxgbe.  Those specific to
255  * T5 are under hw.cxl.
256  */
257 SYSCTL_NODE(_hw, OID_AUTO, cxgbe, CTLFLAG_RD, 0, "cxgbe(4) parameters");
258 SYSCTL_NODE(_hw, OID_AUTO, cxl, CTLFLAG_RD, 0, "cxgbe(4) T5+ parameters");
259 SYSCTL_NODE(_hw_cxgbe, OID_AUTO, toe, CTLFLAG_RD, 0, "cxgbe(4) TOE parameters");
260
261 /*
262  * Number of queues for tx and rx, NIC and offload.
263  */
264 #define NTXQ 16
265 int t4_ntxq = -NTXQ;
266 SYSCTL_INT(_hw_cxgbe, OID_AUTO, ntxq, CTLFLAG_RDTUN, &t4_ntxq, 0,
267     "Number of TX queues per port");
268 TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq);      /* Old name, undocumented */
269
270 #define NRXQ 8
271 int t4_nrxq = -NRXQ;
272 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nrxq, CTLFLAG_RDTUN, &t4_nrxq, 0,
273     "Number of RX queues per port");
274 TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq);      /* Old name, undocumented */
275
276 #define NTXQ_VI 1
277 static int t4_ntxq_vi = -NTXQ_VI;
278 SYSCTL_INT(_hw_cxgbe, OID_AUTO, ntxq_vi, CTLFLAG_RDTUN, &t4_ntxq_vi, 0,
279     "Number of TX queues per VI");
280
281 #define NRXQ_VI 1
282 static int t4_nrxq_vi = -NRXQ_VI;
283 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nrxq_vi, CTLFLAG_RDTUN, &t4_nrxq_vi, 0,
284     "Number of RX queues per VI");
285
286 static int t4_rsrv_noflowq = 0;
287 SYSCTL_INT(_hw_cxgbe, OID_AUTO, rsrv_noflowq, CTLFLAG_RDTUN, &t4_rsrv_noflowq,
288     0, "Reserve TX queue 0 of each VI for non-flowid packets");
289
290 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
291 #define NOFLDTXQ 8
292 static int t4_nofldtxq = -NOFLDTXQ;
293 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldtxq, CTLFLAG_RDTUN, &t4_nofldtxq, 0,
294     "Number of offload TX queues per port");
295
296 #define NOFLDRXQ 2
297 static int t4_nofldrxq = -NOFLDRXQ;
298 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldrxq, CTLFLAG_RDTUN, &t4_nofldrxq, 0,
299     "Number of offload RX queues per port");
300
301 #define NOFLDTXQ_VI 1
302 static int t4_nofldtxq_vi = -NOFLDTXQ_VI;
303 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldtxq_vi, CTLFLAG_RDTUN, &t4_nofldtxq_vi, 0,
304     "Number of offload TX queues per VI");
305
306 #define NOFLDRXQ_VI 1
307 static int t4_nofldrxq_vi = -NOFLDRXQ_VI;
308 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldrxq_vi, CTLFLAG_RDTUN, &t4_nofldrxq_vi, 0,
309     "Number of offload RX queues per VI");
310
311 #define TMR_IDX_OFLD 1
312 int t4_tmr_idx_ofld = TMR_IDX_OFLD;
313 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_timer_idx_ofld, CTLFLAG_RDTUN,
314     &t4_tmr_idx_ofld, 0, "Holdoff timer index for offload queues");
315
316 #define PKTC_IDX_OFLD (-1)
317 int t4_pktc_idx_ofld = PKTC_IDX_OFLD;
318 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_pktc_idx_ofld, CTLFLAG_RDTUN,
319     &t4_pktc_idx_ofld, 0, "holdoff packet counter index for offload queues");
320
321 /* 0 means chip/fw default, non-zero number is value in microseconds */
322 static u_long t4_toe_keepalive_idle = 0;
323 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, keepalive_idle, CTLFLAG_RDTUN,
324     &t4_toe_keepalive_idle, 0, "TOE keepalive idle timer (us)");
325
326 /* 0 means chip/fw default, non-zero number is value in microseconds */
327 static u_long t4_toe_keepalive_interval = 0;
328 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, keepalive_interval, CTLFLAG_RDTUN,
329     &t4_toe_keepalive_interval, 0, "TOE keepalive interval timer (us)");
330
331 /* 0 means chip/fw default, non-zero number is # of keepalives before abort */
332 static int t4_toe_keepalive_count = 0;
333 SYSCTL_INT(_hw_cxgbe_toe, OID_AUTO, keepalive_count, CTLFLAG_RDTUN,
334     &t4_toe_keepalive_count, 0, "Number of TOE keepalive probes before abort");
335
336 /* 0 means chip/fw default, non-zero number is value in microseconds */
337 static u_long t4_toe_rexmt_min = 0;
338 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, rexmt_min, CTLFLAG_RDTUN,
339     &t4_toe_rexmt_min, 0, "Minimum TOE retransmit interval (us)");
340
341 /* 0 means chip/fw default, non-zero number is value in microseconds */
342 static u_long t4_toe_rexmt_max = 0;
343 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, rexmt_max, CTLFLAG_RDTUN,
344     &t4_toe_rexmt_max, 0, "Maximum TOE retransmit interval (us)");
345
346 /* 0 means chip/fw default, non-zero number is # of rexmt before abort */
347 static int t4_toe_rexmt_count = 0;
348 SYSCTL_INT(_hw_cxgbe_toe, OID_AUTO, rexmt_count, CTLFLAG_RDTUN,
349     &t4_toe_rexmt_count, 0, "Number of TOE retransmissions before abort");
350
351 /* -1 means chip/fw default, other values are raw backoff values to use */
352 static int t4_toe_rexmt_backoff[16] = {
353         -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
354 };
355 SYSCTL_NODE(_hw_cxgbe_toe, OID_AUTO, rexmt_backoff, CTLFLAG_RD, 0,
356     "cxgbe(4) TOE retransmit backoff values");
357 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 0, CTLFLAG_RDTUN,
358     &t4_toe_rexmt_backoff[0], 0, "");
359 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 1, CTLFLAG_RDTUN,
360     &t4_toe_rexmt_backoff[1], 0, "");
361 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 2, CTLFLAG_RDTUN,
362     &t4_toe_rexmt_backoff[2], 0, "");
363 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 3, CTLFLAG_RDTUN,
364     &t4_toe_rexmt_backoff[3], 0, "");
365 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 4, CTLFLAG_RDTUN,
366     &t4_toe_rexmt_backoff[4], 0, "");
367 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 5, CTLFLAG_RDTUN,
368     &t4_toe_rexmt_backoff[5], 0, "");
369 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 6, CTLFLAG_RDTUN,
370     &t4_toe_rexmt_backoff[6], 0, "");
371 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 7, CTLFLAG_RDTUN,
372     &t4_toe_rexmt_backoff[7], 0, "");
373 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 8, CTLFLAG_RDTUN,
374     &t4_toe_rexmt_backoff[8], 0, "");
375 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 9, CTLFLAG_RDTUN,
376     &t4_toe_rexmt_backoff[9], 0, "");
377 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 10, CTLFLAG_RDTUN,
378     &t4_toe_rexmt_backoff[10], 0, "");
379 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 11, CTLFLAG_RDTUN,
380     &t4_toe_rexmt_backoff[11], 0, "");
381 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 12, CTLFLAG_RDTUN,
382     &t4_toe_rexmt_backoff[12], 0, "");
383 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 13, CTLFLAG_RDTUN,
384     &t4_toe_rexmt_backoff[13], 0, "");
385 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 14, CTLFLAG_RDTUN,
386     &t4_toe_rexmt_backoff[14], 0, "");
387 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 15, CTLFLAG_RDTUN,
388     &t4_toe_rexmt_backoff[15], 0, "");
389 #endif
390
391 #ifdef DEV_NETMAP
392 #define NNMTXQ_VI 2
393 static int t4_nnmtxq_vi = -NNMTXQ_VI;
394 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nnmtxq_vi, CTLFLAG_RDTUN, &t4_nnmtxq_vi, 0,
395     "Number of netmap TX queues per VI");
396
397 #define NNMRXQ_VI 2
398 static int t4_nnmrxq_vi = -NNMRXQ_VI;
399 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nnmrxq_vi, CTLFLAG_RDTUN, &t4_nnmrxq_vi, 0,
400     "Number of netmap RX queues per VI");
401 #endif
402
403 /*
404  * Holdoff parameters for ports.
405  */
406 #define TMR_IDX 1
407 int t4_tmr_idx = TMR_IDX;
408 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_timer_idx, CTLFLAG_RDTUN, &t4_tmr_idx,
409     0, "Holdoff timer index");
410 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx);     /* Old name */
411
412 #define PKTC_IDX (-1)
413 int t4_pktc_idx = PKTC_IDX;
414 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_pktc_idx, CTLFLAG_RDTUN, &t4_pktc_idx,
415     0, "Holdoff packet counter index");
416 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx);     /* Old name */
417
418 /*
419  * Size (# of entries) of each tx and rx queue.
420  */
421 unsigned int t4_qsize_txq = TX_EQ_QSIZE;
422 SYSCTL_INT(_hw_cxgbe, OID_AUTO, qsize_txq, CTLFLAG_RDTUN, &t4_qsize_txq, 0,
423     "Number of descriptors in each TX queue");
424
425 unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
426 SYSCTL_INT(_hw_cxgbe, OID_AUTO, qsize_rxq, CTLFLAG_RDTUN, &t4_qsize_rxq, 0,
427     "Number of descriptors in each RX queue");
428
429 /*
430  * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
431  */
432 int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
433 SYSCTL_INT(_hw_cxgbe, OID_AUTO, interrupt_types, CTLFLAG_RDTUN, &t4_intr_types,
434     0, "Interrupt types allowed (bit 0 = INTx, 1 = MSI, 2 = MSI-X)");
435
436 /*
437  * Configuration file.  All the _CF names here are special.
438  */
439 #define DEFAULT_CF      "default"
440 #define BUILTIN_CF      "built-in"
441 #define FLASH_CF        "flash"
442 #define UWIRE_CF        "uwire"
443 #define FPGA_CF         "fpga"
444 static char t4_cfg_file[32] = DEFAULT_CF;
445 SYSCTL_STRING(_hw_cxgbe, OID_AUTO, config_file, CTLFLAG_RDTUN, t4_cfg_file,
446     sizeof(t4_cfg_file), "Firmware configuration file");
447
448 /*
449  * PAUSE settings (bit 0, 1, 2 = rx_pause, tx_pause, pause_autoneg respectively).
450  * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
451  * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
452  *            mark or when signalled to do so, 0 to never emit PAUSE.
453  * pause_autoneg = 1 means PAUSE will be negotiated if possible and the
454  *                 negotiated settings will override rx_pause/tx_pause.
455  *                 Otherwise rx_pause/tx_pause are applied forcibly.
456  */
457 static int t4_pause_settings = PAUSE_RX | PAUSE_TX | PAUSE_AUTONEG;
458 SYSCTL_INT(_hw_cxgbe, OID_AUTO, pause_settings, CTLFLAG_RDTUN,
459     &t4_pause_settings, 0,
460     "PAUSE settings (bit 0 = rx_pause, 1 = tx_pause, 2 = pause_autoneg)");
461
462 /*
463  * Forward Error Correction settings (bit 0, 1 = RS, BASER respectively).
464  * -1 to run with the firmware default.  Same as FEC_AUTO (bit 5)
465  *  0 to disable FEC.
466  */
467 static int t4_fec = -1;
468 SYSCTL_INT(_hw_cxgbe, OID_AUTO, fec, CTLFLAG_RDTUN, &t4_fec, 0,
469     "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
470
471 /*
472  * Link autonegotiation.
473  * -1 to run with the firmware default.
474  *  0 to disable.
475  *  1 to enable.
476  */
477 static int t4_autoneg = -1;
478 SYSCTL_INT(_hw_cxgbe, OID_AUTO, autoneg, CTLFLAG_RDTUN, &t4_autoneg, 0,
479     "Link autonegotiation");
480
481 /*
482  * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
483  * encouraged respectively).  '-n' is the same as 'n' except the firmware
484  * version used in the checks is read from the firmware bundled with the driver.
485  */
486 static int t4_fw_install = 1;
487 SYSCTL_INT(_hw_cxgbe, OID_AUTO, fw_install, CTLFLAG_RDTUN, &t4_fw_install, 0,
488     "Firmware auto-install (0 = prohibited, 1 = allowed, 2 = encouraged)");
489
490 /*
491  * ASIC features that will be used.  Disable the ones you don't want so that the
492  * chip resources aren't wasted on features that will not be used.
493  */
494 static int t4_nbmcaps_allowed = 0;
495 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nbmcaps_allowed, CTLFLAG_RDTUN,
496     &t4_nbmcaps_allowed, 0, "Default NBM capabilities");
497
498 static int t4_linkcaps_allowed = 0;     /* No DCBX, PPP, etc. by default */
499 SYSCTL_INT(_hw_cxgbe, OID_AUTO, linkcaps_allowed, CTLFLAG_RDTUN,
500     &t4_linkcaps_allowed, 0, "Default link capabilities");
501
502 static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
503     FW_CAPS_CONFIG_SWITCH_EGRESS;
504 SYSCTL_INT(_hw_cxgbe, OID_AUTO, switchcaps_allowed, CTLFLAG_RDTUN,
505     &t4_switchcaps_allowed, 0, "Default switch capabilities");
506
507 #ifdef RATELIMIT
508 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
509         FW_CAPS_CONFIG_NIC_HASHFILTER | FW_CAPS_CONFIG_NIC_ETHOFLD;
510 #else
511 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
512         FW_CAPS_CONFIG_NIC_HASHFILTER;
513 #endif
514 SYSCTL_INT(_hw_cxgbe, OID_AUTO, niccaps_allowed, CTLFLAG_RDTUN,
515     &t4_niccaps_allowed, 0, "Default NIC capabilities");
516
517 static int t4_toecaps_allowed = -1;
518 SYSCTL_INT(_hw_cxgbe, OID_AUTO, toecaps_allowed, CTLFLAG_RDTUN,
519     &t4_toecaps_allowed, 0, "Default TCP offload capabilities");
520
521 static int t4_rdmacaps_allowed = -1;
522 SYSCTL_INT(_hw_cxgbe, OID_AUTO, rdmacaps_allowed, CTLFLAG_RDTUN,
523     &t4_rdmacaps_allowed, 0, "Default RDMA capabilities");
524
525 static int t4_cryptocaps_allowed = -1;
526 SYSCTL_INT(_hw_cxgbe, OID_AUTO, cryptocaps_allowed, CTLFLAG_RDTUN,
527     &t4_cryptocaps_allowed, 0, "Default crypto capabilities");
528
529 static int t4_iscsicaps_allowed = -1;
530 SYSCTL_INT(_hw_cxgbe, OID_AUTO, iscsicaps_allowed, CTLFLAG_RDTUN,
531     &t4_iscsicaps_allowed, 0, "Default iSCSI capabilities");
532
533 static int t4_fcoecaps_allowed = 0;
534 SYSCTL_INT(_hw_cxgbe, OID_AUTO, fcoecaps_allowed, CTLFLAG_RDTUN,
535     &t4_fcoecaps_allowed, 0, "Default FCoE capabilities");
536
537 static int t5_write_combine = 0;
538 SYSCTL_INT(_hw_cxl, OID_AUTO, write_combine, CTLFLAG_RDTUN, &t5_write_combine,
539     0, "Use WC instead of UC for BAR2");
540
541 static int t4_num_vis = 1;
542 SYSCTL_INT(_hw_cxgbe, OID_AUTO, num_vis, CTLFLAG_RDTUN, &t4_num_vis, 0,
543     "Number of VIs per port");
544
545 /*
546  * PCIe Relaxed Ordering.
547  * -1: driver should figure out a good value.
548  * 0: disable RO.
549  * 1: enable RO.
550  * 2: leave RO alone.
551  */
552 static int pcie_relaxed_ordering = -1;
553 SYSCTL_INT(_hw_cxgbe, OID_AUTO, pcie_relaxed_ordering, CTLFLAG_RDTUN,
554     &pcie_relaxed_ordering, 0,
555     "PCIe Relaxed Ordering: 0 = disable, 1 = enable, 2 = leave alone");
556
557 static int t4_panic_on_fatal_err = 0;
558 SYSCTL_INT(_hw_cxgbe, OID_AUTO, panic_on_fatal_err, CTLFLAG_RDTUN,
559     &t4_panic_on_fatal_err, 0, "panic on fatal errors");
560
561 #ifdef TCP_OFFLOAD
562 /*
563  * TOE tunables.
564  */
565 static int t4_cop_managed_offloading = 0;
566 SYSCTL_INT(_hw_cxgbe, OID_AUTO, cop_managed_offloading, CTLFLAG_RDTUN,
567     &t4_cop_managed_offloading, 0,
568     "COP (Connection Offload Policy) controls all TOE offload");
569 #endif
570
571 /* Functions used by VIs to obtain unique MAC addresses for each VI. */
572 static int vi_mac_funcs[] = {
573         FW_VI_FUNC_ETH,
574         FW_VI_FUNC_OFLD,
575         FW_VI_FUNC_IWARP,
576         FW_VI_FUNC_OPENISCSI,
577         FW_VI_FUNC_OPENFCOE,
578         FW_VI_FUNC_FOISCSI,
579         FW_VI_FUNC_FOFCOE,
580 };
581
582 struct intrs_and_queues {
583         uint16_t intr_type;     /* INTx, MSI, or MSI-X */
584         uint16_t num_vis;       /* number of VIs for each port */
585         uint16_t nirq;          /* Total # of vectors */
586         uint16_t ntxq;          /* # of NIC txq's for each port */
587         uint16_t nrxq;          /* # of NIC rxq's for each port */
588         uint16_t nofldtxq;      /* # of TOE/ETHOFLD txq's for each port */
589         uint16_t nofldrxq;      /* # of TOE rxq's for each port */
590
591         /* The vcxgbe/vcxl interfaces use these and not the ones above. */
592         uint16_t ntxq_vi;       /* # of NIC txq's */
593         uint16_t nrxq_vi;       /* # of NIC rxq's */
594         uint16_t nofldtxq_vi;   /* # of TOE txq's */
595         uint16_t nofldrxq_vi;   /* # of TOE rxq's */
596         uint16_t nnmtxq_vi;     /* # of netmap txq's */
597         uint16_t nnmrxq_vi;     /* # of netmap rxq's */
598 };
599
600 static void setup_memwin(struct adapter *);
601 static void position_memwin(struct adapter *, int, uint32_t);
602 static int validate_mem_range(struct adapter *, uint32_t, uint32_t);
603 static int fwmtype_to_hwmtype(int);
604 static int validate_mt_off_len(struct adapter *, int, uint32_t, uint32_t,
605     uint32_t *);
606 static int fixup_devlog_params(struct adapter *);
607 static int cfg_itype_and_nqueues(struct adapter *, struct intrs_and_queues *);
608 static int contact_firmware(struct adapter *);
609 static int partition_resources(struct adapter *);
610 static int get_params__pre_init(struct adapter *);
611 static int set_params__pre_init(struct adapter *);
612 static int get_params__post_init(struct adapter *);
613 static int set_params__post_init(struct adapter *);
614 static void t4_set_desc(struct adapter *);
615 static bool fixed_ifmedia(struct port_info *);
616 static void build_medialist(struct port_info *);
617 static void init_link_config(struct port_info *);
618 static int fixup_link_config(struct port_info *);
619 static int apply_link_config(struct port_info *);
620 static int cxgbe_init_synchronized(struct vi_info *);
621 static int cxgbe_uninit_synchronized(struct vi_info *);
622 static void quiesce_txq(struct adapter *, struct sge_txq *);
623 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
624 static void quiesce_iq(struct adapter *, struct sge_iq *);
625 static void quiesce_fl(struct adapter *, struct sge_fl *);
626 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
627     driver_intr_t *, void *, char *);
628 static int t4_free_irq(struct adapter *, struct irq *);
629 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
630 static void vi_refresh_stats(struct adapter *, struct vi_info *);
631 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
632 static void cxgbe_tick(void *);
633 static void cxgbe_sysctls(struct port_info *);
634 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
635 static int sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS);
636 static int sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS);
637 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
638 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
639 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
640 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
641 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
642 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
643 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
644 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
645 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
646 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
647 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
648 static int sysctl_loadavg(SYSCTL_HANDLER_ARGS);
649 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
650 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
651 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
652 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
653 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
654 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
655 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
656 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
657 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
658 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
659 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
660 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
661 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
662 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
663 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
664 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
665 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
666 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
667 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
668 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
669 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
670 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
671 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
672 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
673 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
674 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
675 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
676 static int sysctl_cpus(SYSCTL_HANDLER_ARGS);
677 #ifdef TCP_OFFLOAD
678 static int sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS);
679 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
680 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
681 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
682 static int sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS);
683 static int sysctl_tp_backoff(SYSCTL_HANDLER_ARGS);
684 static int sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS);
685 static int sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS);
686 #endif
687 static int get_sge_context(struct adapter *, struct t4_sge_context *);
688 static int load_fw(struct adapter *, struct t4_data *);
689 static int load_cfg(struct adapter *, struct t4_data *);
690 static int load_boot(struct adapter *, struct t4_bootrom *);
691 static int load_bootcfg(struct adapter *, struct t4_data *);
692 static int cudbg_dump(struct adapter *, struct t4_cudbg_dump *);
693 static void free_offload_policy(struct t4_offload_policy *);
694 static int set_offload_policy(struct adapter *, struct t4_offload_policy *);
695 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
696 static int read_i2c(struct adapter *, struct t4_i2c_data *);
697 #ifdef TCP_OFFLOAD
698 static int toe_capability(struct vi_info *, int);
699 #endif
700 static int mod_event(module_t, int, void *);
701 static int notify_siblings(device_t, int);
702
703 struct {
704         uint16_t device;
705         char *desc;
706 } t4_pciids[] = {
707         {0xa000, "Chelsio Terminator 4 FPGA"},
708         {0x4400, "Chelsio T440-dbg"},
709         {0x4401, "Chelsio T420-CR"},
710         {0x4402, "Chelsio T422-CR"},
711         {0x4403, "Chelsio T440-CR"},
712         {0x4404, "Chelsio T420-BCH"},
713         {0x4405, "Chelsio T440-BCH"},
714         {0x4406, "Chelsio T440-CH"},
715         {0x4407, "Chelsio T420-SO"},
716         {0x4408, "Chelsio T420-CX"},
717         {0x4409, "Chelsio T420-BT"},
718         {0x440a, "Chelsio T404-BT"},
719         {0x440e, "Chelsio T440-LP-CR"},
720 }, t5_pciids[] = {
721         {0xb000, "Chelsio Terminator 5 FPGA"},
722         {0x5400, "Chelsio T580-dbg"},
723         {0x5401,  "Chelsio T520-CR"},           /* 2 x 10G */
724         {0x5402,  "Chelsio T522-CR"},           /* 2 x 10G, 2 X 1G */
725         {0x5403,  "Chelsio T540-CR"},           /* 4 x 10G */
726         {0x5407,  "Chelsio T520-SO"},           /* 2 x 10G, nomem */
727         {0x5409,  "Chelsio T520-BT"},           /* 2 x 10GBaseT */
728         {0x540a,  "Chelsio T504-BT"},           /* 4 x 1G */
729         {0x540d,  "Chelsio T580-CR"},           /* 2 x 40G */
730         {0x540e,  "Chelsio T540-LP-CR"},        /* 4 x 10G */
731         {0x5410,  "Chelsio T580-LP-CR"},        /* 2 x 40G */
732         {0x5411,  "Chelsio T520-LL-CR"},        /* 2 x 10G */
733         {0x5412,  "Chelsio T560-CR"},           /* 1 x 40G, 2 x 10G */
734         {0x5414,  "Chelsio T580-LP-SO-CR"},     /* 2 x 40G, nomem */
735         {0x5415,  "Chelsio T502-BT"},           /* 2 x 1G */
736         {0x5418,  "Chelsio T540-BT"},           /* 4 x 10GBaseT */
737         {0x5419,  "Chelsio T540-LP-BT"},        /* 4 x 10GBaseT */
738         {0x541a,  "Chelsio T540-SO-BT"},        /* 4 x 10GBaseT, nomem */
739         {0x541b,  "Chelsio T540-SO-CR"},        /* 4 x 10G, nomem */
740
741         /* Custom */
742         {0x5483, "Custom T540-CR"},
743         {0x5484, "Custom T540-BT"},
744 }, t6_pciids[] = {
745         {0xc006, "Chelsio Terminator 6 FPGA"},  /* T6 PE10K6 FPGA (PF0) */
746         {0x6400, "Chelsio T6-DBG-25"},          /* 2 x 10/25G, debug */
747         {0x6401, "Chelsio T6225-CR"},           /* 2 x 10/25G */
748         {0x6402, "Chelsio T6225-SO-CR"},        /* 2 x 10/25G, nomem */
749         {0x6403, "Chelsio T6425-CR"},           /* 4 x 10/25G */
750         {0x6404, "Chelsio T6425-SO-CR"},        /* 4 x 10/25G, nomem */
751         {0x6405, "Chelsio T6225-OCP-SO"},       /* 2 x 10/25G, nomem */
752         {0x6406, "Chelsio T62100-OCP-SO"},      /* 2 x 40/50/100G, nomem */
753         {0x6407, "Chelsio T62100-LP-CR"},       /* 2 x 40/50/100G */
754         {0x6408, "Chelsio T62100-SO-CR"},       /* 2 x 40/50/100G, nomem */
755         {0x6409, "Chelsio T6210-BT"},           /* 2 x 10GBASE-T */
756         {0x640d, "Chelsio T62100-CR"},          /* 2 x 40/50/100G */
757         {0x6410, "Chelsio T6-DBG-100"},         /* 2 x 40/50/100G, debug */
758         {0x6411, "Chelsio T6225-LL-CR"},        /* 2 x 10/25G */
759         {0x6414, "Chelsio T61100-OCP-SO"},      /* 1 x 40/50/100G, nomem */
760         {0x6415, "Chelsio T6201-BT"},           /* 2 x 1000BASE-T */
761
762         /* Custom */
763         {0x6480, "Custom T6225-CR"},
764         {0x6481, "Custom T62100-CR"},
765         {0x6482, "Custom T6225-CR"},
766         {0x6483, "Custom T62100-CR"},
767         {0x6484, "Custom T64100-CR"},
768         {0x6485, "Custom T6240-SO"},
769         {0x6486, "Custom T6225-SO-CR"},
770         {0x6487, "Custom T6225-CR"},
771 };
772
773 #ifdef TCP_OFFLOAD
774 /*
775  * service_iq_fl() has an iq and needs the fl.  Offset of fl from the iq should
776  * be exactly the same for both rxq and ofld_rxq.
777  */
778 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
779 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
780 #endif
781 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
782
783 static int
784 t4_probe(device_t dev)
785 {
786         int i;
787         uint16_t v = pci_get_vendor(dev);
788         uint16_t d = pci_get_device(dev);
789         uint8_t f = pci_get_function(dev);
790
791         if (v != PCI_VENDOR_ID_CHELSIO)
792                 return (ENXIO);
793
794         /* Attach only to PF0 of the FPGA */
795         if (d == 0xa000 && f != 0)
796                 return (ENXIO);
797
798         for (i = 0; i < nitems(t4_pciids); i++) {
799                 if (d == t4_pciids[i].device) {
800                         device_set_desc(dev, t4_pciids[i].desc);
801                         return (BUS_PROBE_DEFAULT);
802                 }
803         }
804
805         return (ENXIO);
806 }
807
808 static int
809 t5_probe(device_t dev)
810 {
811         int i;
812         uint16_t v = pci_get_vendor(dev);
813         uint16_t d = pci_get_device(dev);
814         uint8_t f = pci_get_function(dev);
815
816         if (v != PCI_VENDOR_ID_CHELSIO)
817                 return (ENXIO);
818
819         /* Attach only to PF0 of the FPGA */
820         if (d == 0xb000 && f != 0)
821                 return (ENXIO);
822
823         for (i = 0; i < nitems(t5_pciids); i++) {
824                 if (d == t5_pciids[i].device) {
825                         device_set_desc(dev, t5_pciids[i].desc);
826                         return (BUS_PROBE_DEFAULT);
827                 }
828         }
829
830         return (ENXIO);
831 }
832
833 static int
834 t6_probe(device_t dev)
835 {
836         int i;
837         uint16_t v = pci_get_vendor(dev);
838         uint16_t d = pci_get_device(dev);
839
840         if (v != PCI_VENDOR_ID_CHELSIO)
841                 return (ENXIO);
842
843         for (i = 0; i < nitems(t6_pciids); i++) {
844                 if (d == t6_pciids[i].device) {
845                         device_set_desc(dev, t6_pciids[i].desc);
846                         return (BUS_PROBE_DEFAULT);
847                 }
848         }
849
850         return (ENXIO);
851 }
852
853 static void
854 t5_attribute_workaround(device_t dev)
855 {
856         device_t root_port;
857         uint32_t v;
858
859         /*
860          * The T5 chips do not properly echo the No Snoop and Relaxed
861          * Ordering attributes when replying to a TLP from a Root
862          * Port.  As a workaround, find the parent Root Port and
863          * disable No Snoop and Relaxed Ordering.  Note that this
864          * affects all devices under this root port.
865          */
866         root_port = pci_find_pcie_root_port(dev);
867         if (root_port == NULL) {
868                 device_printf(dev, "Unable to find parent root port\n");
869                 return;
870         }
871
872         v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
873             PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
874         if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
875             0)
876                 device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
877                     device_get_nameunit(root_port));
878 }
879
880 static const struct devnames devnames[] = {
881         {
882                 .nexus_name = "t4nex",
883                 .ifnet_name = "cxgbe",
884                 .vi_ifnet_name = "vcxgbe",
885                 .pf03_drv_name = "t4iov",
886                 .vf_nexus_name = "t4vf",
887                 .vf_ifnet_name = "cxgbev"
888         }, {
889                 .nexus_name = "t5nex",
890                 .ifnet_name = "cxl",
891                 .vi_ifnet_name = "vcxl",
892                 .pf03_drv_name = "t5iov",
893                 .vf_nexus_name = "t5vf",
894                 .vf_ifnet_name = "cxlv"
895         }, {
896                 .nexus_name = "t6nex",
897                 .ifnet_name = "cc",
898                 .vi_ifnet_name = "vcc",
899                 .pf03_drv_name = "t6iov",
900                 .vf_nexus_name = "t6vf",
901                 .vf_ifnet_name = "ccv"
902         }
903 };
904
905 void
906 t4_init_devnames(struct adapter *sc)
907 {
908         int id;
909
910         id = chip_id(sc);
911         if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
912                 sc->names = &devnames[id - CHELSIO_T4];
913         else {
914                 device_printf(sc->dev, "chip id %d is not supported.\n", id);
915                 sc->names = NULL;
916         }
917 }
918
919 static int
920 t4_ifnet_unit(struct adapter *sc, struct port_info *pi)
921 {
922         const char *parent, *name;
923         long value;
924         int line, unit;
925
926         line = 0;
927         parent = device_get_nameunit(sc->dev);
928         name = sc->names->ifnet_name;
929         while (resource_find_dev(&line, name, &unit, "at", parent) == 0) {
930                 if (resource_long_value(name, unit, "port", &value) == 0 &&
931                     value == pi->port_id)
932                         return (unit);
933         }
934         return (-1);
935 }
936
937 static int
938 t4_attach(device_t dev)
939 {
940         struct adapter *sc;
941         int rc = 0, i, j, rqidx, tqidx, nports;
942         struct make_dev_args mda;
943         struct intrs_and_queues iaq;
944         struct sge *s;
945         uint32_t *buf;
946 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
947         int ofld_tqidx;
948 #endif
949 #ifdef TCP_OFFLOAD
950         int ofld_rqidx;
951 #endif
952 #ifdef DEV_NETMAP
953         int nm_rqidx, nm_tqidx;
954 #endif
955         int num_vis;
956
957         sc = device_get_softc(dev);
958         sc->dev = dev;
959         TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
960
961         if ((pci_get_device(dev) & 0xff00) == 0x5400)
962                 t5_attribute_workaround(dev);
963         pci_enable_busmaster(dev);
964         if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
965                 uint32_t v;
966
967                 pci_set_max_read_req(dev, 4096);
968                 v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
969                 sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
970                 if (pcie_relaxed_ordering == 0 &&
971                     (v & PCIEM_CTL_RELAXED_ORD_ENABLE) != 0) {
972                         v &= ~PCIEM_CTL_RELAXED_ORD_ENABLE;
973                         pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
974                 } else if (pcie_relaxed_ordering == 1 &&
975                     (v & PCIEM_CTL_RELAXED_ORD_ENABLE) == 0) {
976                         v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
977                         pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
978                 }
979         }
980
981         sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
982         sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
983         sc->traceq = -1;
984         mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
985         snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
986             device_get_nameunit(dev));
987
988         snprintf(sc->lockname, sizeof(sc->lockname), "%s",
989             device_get_nameunit(dev));
990         mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
991         t4_add_adapter(sc);
992
993         mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
994         TAILQ_INIT(&sc->sfl);
995         callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
996
997         mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
998
999         sc->policy = NULL;
1000         rw_init(&sc->policy_lock, "connection offload policy");
1001
1002         rc = t4_map_bars_0_and_4(sc);
1003         if (rc != 0)
1004                 goto done; /* error message displayed already */
1005
1006         memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
1007
1008         /* Prepare the adapter for operation. */
1009         buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
1010         rc = -t4_prep_adapter(sc, buf);
1011         free(buf, M_CXGBE);
1012         if (rc != 0) {
1013                 device_printf(dev, "failed to prepare adapter: %d.\n", rc);
1014                 goto done;
1015         }
1016
1017         /*
1018          * This is the real PF# to which we're attaching.  Works from within PCI
1019          * passthrough environments too, where pci_get_function() could return a
1020          * different PF# depending on the passthrough configuration.  We need to
1021          * use the real PF# in all our communication with the firmware.
1022          */
1023         j = t4_read_reg(sc, A_PL_WHOAMI);
1024         sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
1025         sc->mbox = sc->pf;
1026
1027         t4_init_devnames(sc);
1028         if (sc->names == NULL) {
1029                 rc = ENOTSUP;
1030                 goto done; /* error message displayed already */
1031         }
1032
1033         /*
1034          * Do this really early, with the memory windows set up even before the
1035          * character device.  The userland tool's register i/o and mem read
1036          * will work even in "recovery mode".
1037          */
1038         setup_memwin(sc);
1039         if (t4_init_devlog_params(sc, 0) == 0)
1040                 fixup_devlog_params(sc);
1041         make_dev_args_init(&mda);
1042         mda.mda_devsw = &t4_cdevsw;
1043         mda.mda_uid = UID_ROOT;
1044         mda.mda_gid = GID_WHEEL;
1045         mda.mda_mode = 0600;
1046         mda.mda_si_drv1 = sc;
1047         rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
1048         if (rc != 0)
1049                 device_printf(dev, "failed to create nexus char device: %d.\n",
1050                     rc);
1051
1052         /* Go no further if recovery mode has been requested. */
1053         if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
1054                 device_printf(dev, "recovery mode.\n");
1055                 goto done;
1056         }
1057
1058 #if defined(__i386__)
1059         if ((cpu_feature & CPUID_CX8) == 0) {
1060                 device_printf(dev, "64 bit atomics not available.\n");
1061                 rc = ENOTSUP;
1062                 goto done;
1063         }
1064 #endif
1065
1066         /* Contact the firmware and try to become the master driver. */
1067         rc = contact_firmware(sc);
1068         if (rc != 0)
1069                 goto done; /* error message displayed already */
1070         MPASS(sc->flags & FW_OK);
1071
1072         rc = get_params__pre_init(sc);
1073         if (rc != 0)
1074                 goto done; /* error message displayed already */
1075
1076         if (sc->flags & MASTER_PF) {
1077                 rc = partition_resources(sc);
1078                 if (rc != 0)
1079                         goto done; /* error message displayed already */
1080                 t4_intr_clear(sc);
1081         }
1082
1083         rc = get_params__post_init(sc);
1084         if (rc != 0)
1085                 goto done; /* error message displayed already */
1086
1087         rc = set_params__post_init(sc);
1088         if (rc != 0)
1089                 goto done; /* error message displayed already */
1090
1091         rc = t4_map_bar_2(sc);
1092         if (rc != 0)
1093                 goto done; /* error message displayed already */
1094
1095         rc = t4_create_dma_tag(sc);
1096         if (rc != 0)
1097                 goto done; /* error message displayed already */
1098
1099         /*
1100          * First pass over all the ports - allocate VIs and initialize some
1101          * basic parameters like mac address, port type, etc.
1102          */
1103         for_each_port(sc, i) {
1104                 struct port_info *pi;
1105
1106                 pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
1107                 sc->port[i] = pi;
1108
1109                 /* These must be set before t4_port_init */
1110                 pi->adapter = sc;
1111                 pi->port_id = i;
1112                 /*
1113                  * XXX: vi[0] is special so we can't delay this allocation until
1114                  * pi->nvi's final value is known.
1115                  */
1116                 pi->vi = malloc(sizeof(struct vi_info) * t4_num_vis, M_CXGBE,
1117                     M_ZERO | M_WAITOK);
1118
1119                 /*
1120                  * Allocate the "main" VI and initialize parameters
1121                  * like mac addr.
1122                  */
1123                 rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
1124                 if (rc != 0) {
1125                         device_printf(dev, "unable to initialize port %d: %d\n",
1126                             i, rc);
1127                         free(pi->vi, M_CXGBE);
1128                         free(pi, M_CXGBE);
1129                         sc->port[i] = NULL;
1130                         goto done;
1131                 }
1132
1133                 snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
1134                     device_get_nameunit(dev), i);
1135                 mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
1136                 sc->chan_map[pi->tx_chan] = i;
1137
1138                 /* All VIs on this port share this media. */
1139                 ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
1140                     cxgbe_media_status);
1141
1142                 PORT_LOCK(pi);
1143                 init_link_config(pi);
1144                 fixup_link_config(pi);
1145                 build_medialist(pi);
1146                 if (fixed_ifmedia(pi))
1147                         pi->flags |= FIXED_IFMEDIA;
1148                 PORT_UNLOCK(pi);
1149
1150                 pi->dev = device_add_child(dev, sc->names->ifnet_name,
1151                     t4_ifnet_unit(sc, pi));
1152                 if (pi->dev == NULL) {
1153                         device_printf(dev,
1154                             "failed to add device for port %d.\n", i);
1155                         rc = ENXIO;
1156                         goto done;
1157                 }
1158                 pi->vi[0].dev = pi->dev;
1159                 device_set_softc(pi->dev, pi);
1160         }
1161
1162         /*
1163          * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1164          */
1165         nports = sc->params.nports;
1166         rc = cfg_itype_and_nqueues(sc, &iaq);
1167         if (rc != 0)
1168                 goto done; /* error message displayed already */
1169
1170         num_vis = iaq.num_vis;
1171         sc->intr_type = iaq.intr_type;
1172         sc->intr_count = iaq.nirq;
1173
1174         s = &sc->sge;
1175         s->nrxq = nports * iaq.nrxq;
1176         s->ntxq = nports * iaq.ntxq;
1177         if (num_vis > 1) {
1178                 s->nrxq += nports * (num_vis - 1) * iaq.nrxq_vi;
1179                 s->ntxq += nports * (num_vis - 1) * iaq.ntxq_vi;
1180         }
1181         s->neq = s->ntxq + s->nrxq;     /* the free list in an rxq is an eq */
1182         s->neq += nports;               /* ctrl queues: 1 per port */
1183         s->niq = s->nrxq + 1;           /* 1 extra for firmware event queue */
1184 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1185         if (is_offload(sc) || is_ethoffload(sc)) {
1186                 s->nofldtxq = nports * iaq.nofldtxq;
1187                 if (num_vis > 1)
1188                         s->nofldtxq += nports * (num_vis - 1) * iaq.nofldtxq_vi;
1189                 s->neq += s->nofldtxq;
1190
1191                 s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1192                     M_CXGBE, M_ZERO | M_WAITOK);
1193         }
1194 #endif
1195 #ifdef TCP_OFFLOAD
1196         if (is_offload(sc)) {
1197                 s->nofldrxq = nports * iaq.nofldrxq;
1198                 if (num_vis > 1)
1199                         s->nofldrxq += nports * (num_vis - 1) * iaq.nofldrxq_vi;
1200                 s->neq += s->nofldrxq;  /* free list */
1201                 s->niq += s->nofldrxq;
1202
1203                 s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1204                     M_CXGBE, M_ZERO | M_WAITOK);
1205         }
1206 #endif
1207 #ifdef DEV_NETMAP
1208         if (num_vis > 1) {
1209                 s->nnmrxq = nports * (num_vis - 1) * iaq.nnmrxq_vi;
1210                 s->nnmtxq = nports * (num_vis - 1) * iaq.nnmtxq_vi;
1211         }
1212         s->neq += s->nnmtxq + s->nnmrxq;
1213         s->niq += s->nnmrxq;
1214
1215         s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1216             M_CXGBE, M_ZERO | M_WAITOK);
1217         s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1218             M_CXGBE, M_ZERO | M_WAITOK);
1219 #endif
1220
1221         s->ctrlq = malloc(nports * sizeof(struct sge_wrq), M_CXGBE,
1222             M_ZERO | M_WAITOK);
1223         s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1224             M_ZERO | M_WAITOK);
1225         s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1226             M_ZERO | M_WAITOK);
1227         s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1228             M_ZERO | M_WAITOK);
1229         s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1230             M_ZERO | M_WAITOK);
1231
1232         sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1233             M_ZERO | M_WAITOK);
1234
1235         t4_init_l2t(sc, M_WAITOK);
1236         t4_init_smt(sc, M_WAITOK);
1237         t4_init_tx_sched(sc);
1238 #ifdef RATELIMIT
1239         t4_init_etid_table(sc);
1240 #endif
1241 #ifdef INET6
1242         t4_init_clip_table(sc);
1243 #endif
1244         if (sc->vres.key.size != 0)
1245                 sc->key_map = vmem_create("T4TLS key map", sc->vres.key.start,
1246                     sc->vres.key.size, 32, 0, M_FIRSTFIT | M_WAITOK);
1247
1248         /*
1249          * Second pass over the ports.  This time we know the number of rx and
1250          * tx queues that each port should get.
1251          */
1252         rqidx = tqidx = 0;
1253 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1254         ofld_tqidx = 0;
1255 #endif
1256 #ifdef TCP_OFFLOAD
1257         ofld_rqidx = 0;
1258 #endif
1259 #ifdef DEV_NETMAP
1260         nm_rqidx = nm_tqidx = 0;
1261 #endif
1262         for_each_port(sc, i) {
1263                 struct port_info *pi = sc->port[i];
1264                 struct vi_info *vi;
1265
1266                 if (pi == NULL)
1267                         continue;
1268
1269                 pi->nvi = num_vis;
1270                 for_each_vi(pi, j, vi) {
1271                         vi->pi = pi;
1272                         vi->qsize_rxq = t4_qsize_rxq;
1273                         vi->qsize_txq = t4_qsize_txq;
1274
1275                         vi->first_rxq = rqidx;
1276                         vi->first_txq = tqidx;
1277                         vi->tmr_idx = t4_tmr_idx;
1278                         vi->pktc_idx = t4_pktc_idx;
1279                         vi->nrxq = j == 0 ? iaq.nrxq : iaq.nrxq_vi;
1280                         vi->ntxq = j == 0 ? iaq.ntxq : iaq.ntxq_vi;
1281
1282                         rqidx += vi->nrxq;
1283                         tqidx += vi->ntxq;
1284
1285                         if (j == 0 && vi->ntxq > 1)
1286                                 vi->rsrv_noflowq = t4_rsrv_noflowq ? 1 : 0;
1287                         else
1288                                 vi->rsrv_noflowq = 0;
1289
1290 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1291                         vi->first_ofld_txq = ofld_tqidx;
1292                         vi->nofldtxq = j == 0 ? iaq.nofldtxq : iaq.nofldtxq_vi;
1293                         ofld_tqidx += vi->nofldtxq;
1294 #endif
1295 #ifdef TCP_OFFLOAD
1296                         vi->ofld_tmr_idx = t4_tmr_idx_ofld;
1297                         vi->ofld_pktc_idx = t4_pktc_idx_ofld;
1298                         vi->first_ofld_rxq = ofld_rqidx;
1299                         vi->nofldrxq = j == 0 ? iaq.nofldrxq : iaq.nofldrxq_vi;
1300
1301                         ofld_rqidx += vi->nofldrxq;
1302 #endif
1303 #ifdef DEV_NETMAP
1304                         if (j > 0) {
1305                                 vi->first_nm_rxq = nm_rqidx;
1306                                 vi->first_nm_txq = nm_tqidx;
1307                                 vi->nnmrxq = iaq.nnmrxq_vi;
1308                                 vi->nnmtxq = iaq.nnmtxq_vi;
1309                                 nm_rqidx += vi->nnmrxq;
1310                                 nm_tqidx += vi->nnmtxq;
1311                         }
1312 #endif
1313                 }
1314         }
1315
1316         rc = t4_setup_intr_handlers(sc);
1317         if (rc != 0) {
1318                 device_printf(dev,
1319                     "failed to setup interrupt handlers: %d\n", rc);
1320                 goto done;
1321         }
1322
1323         rc = bus_generic_probe(dev);
1324         if (rc != 0) {
1325                 device_printf(dev, "failed to probe child drivers: %d\n", rc);
1326                 goto done;
1327         }
1328
1329         /*
1330          * Ensure thread-safe mailbox access (in debug builds).
1331          *
1332          * So far this was the only thread accessing the mailbox but various
1333          * ifnets and sysctls are about to be created and their handlers/ioctls
1334          * will access the mailbox from different threads.
1335          */
1336         sc->flags |= CHK_MBOX_ACCESS;
1337
1338         rc = bus_generic_attach(dev);
1339         if (rc != 0) {
1340                 device_printf(dev,
1341                     "failed to attach all child ports: %d\n", rc);
1342                 goto done;
1343         }
1344
1345         device_printf(dev,
1346             "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1347             sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1348             sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1349             (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1350             sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1351
1352         t4_set_desc(sc);
1353
1354         notify_siblings(dev, 0);
1355
1356 done:
1357         if (rc != 0 && sc->cdev) {
1358                 /* cdev was created and so cxgbetool works; recover that way. */
1359                 device_printf(dev,
1360                     "error during attach, adapter is now in recovery mode.\n");
1361                 rc = 0;
1362         }
1363
1364         if (rc != 0)
1365                 t4_detach_common(dev);
1366         else
1367                 t4_sysctls(sc);
1368
1369         return (rc);
1370 }
1371
1372 static int
1373 t4_child_location_str(device_t bus, device_t dev, char *buf, size_t buflen)
1374 {
1375         struct adapter *sc;
1376         struct port_info *pi;
1377         int i;
1378
1379         sc = device_get_softc(bus);
1380         buf[0] = '\0';
1381         for_each_port(sc, i) {
1382                 pi = sc->port[i];
1383                 if (pi != NULL && pi->dev == dev) {
1384                         snprintf(buf, buflen, "port=%d", pi->port_id);
1385                         break;
1386                 }
1387         }
1388         return (0);
1389 }
1390
1391 static int
1392 t4_ready(device_t dev)
1393 {
1394         struct adapter *sc;
1395
1396         sc = device_get_softc(dev);
1397         if (sc->flags & FW_OK)
1398                 return (0);
1399         return (ENXIO);
1400 }
1401
1402 static int
1403 t4_read_port_device(device_t dev, int port, device_t *child)
1404 {
1405         struct adapter *sc;
1406         struct port_info *pi;
1407
1408         sc = device_get_softc(dev);
1409         if (port < 0 || port >= MAX_NPORTS)
1410                 return (EINVAL);
1411         pi = sc->port[port];
1412         if (pi == NULL || pi->dev == NULL)
1413                 return (ENXIO);
1414         *child = pi->dev;
1415         return (0);
1416 }
1417
1418 static int
1419 notify_siblings(device_t dev, int detaching)
1420 {
1421         device_t sibling;
1422         int error, i;
1423
1424         error = 0;
1425         for (i = 0; i < PCI_FUNCMAX; i++) {
1426                 if (i == pci_get_function(dev))
1427                         continue;
1428                 sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1429                     pci_get_slot(dev), i);
1430                 if (sibling == NULL || !device_is_attached(sibling))
1431                         continue;
1432                 if (detaching)
1433                         error = T4_DETACH_CHILD(sibling);
1434                 else
1435                         (void)T4_ATTACH_CHILD(sibling);
1436                 if (error)
1437                         break;
1438         }
1439         return (error);
1440 }
1441
1442 /*
1443  * Idempotent
1444  */
1445 static int
1446 t4_detach(device_t dev)
1447 {
1448         struct adapter *sc;
1449         int rc;
1450
1451         sc = device_get_softc(dev);
1452
1453         rc = notify_siblings(dev, 1);
1454         if (rc) {
1455                 device_printf(dev,
1456                     "failed to detach sibling devices: %d\n", rc);
1457                 return (rc);
1458         }
1459
1460         return (t4_detach_common(dev));
1461 }
1462
1463 int
1464 t4_detach_common(device_t dev)
1465 {
1466         struct adapter *sc;
1467         struct port_info *pi;
1468         int i, rc;
1469
1470         sc = device_get_softc(dev);
1471
1472         if (sc->cdev) {
1473                 destroy_dev(sc->cdev);
1474                 sc->cdev = NULL;
1475         }
1476
1477         sc->flags &= ~CHK_MBOX_ACCESS;
1478         if (sc->flags & FULL_INIT_DONE) {
1479                 if (!(sc->flags & IS_VF))
1480                         t4_intr_disable(sc);
1481         }
1482
1483         if (device_is_attached(dev)) {
1484                 rc = bus_generic_detach(dev);
1485                 if (rc) {
1486                         device_printf(dev,
1487                             "failed to detach child devices: %d\n", rc);
1488                         return (rc);
1489                 }
1490         }
1491
1492         for (i = 0; i < sc->intr_count; i++)
1493                 t4_free_irq(sc, &sc->irq[i]);
1494
1495         if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1496                 t4_free_tx_sched(sc);
1497
1498         for (i = 0; i < MAX_NPORTS; i++) {
1499                 pi = sc->port[i];
1500                 if (pi) {
1501                         t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1502                         if (pi->dev)
1503                                 device_delete_child(dev, pi->dev);
1504
1505                         mtx_destroy(&pi->pi_lock);
1506                         free(pi->vi, M_CXGBE);
1507                         free(pi, M_CXGBE);
1508                 }
1509         }
1510
1511         device_delete_children(dev);
1512
1513         if (sc->flags & FULL_INIT_DONE)
1514                 adapter_full_uninit(sc);
1515
1516         if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1517                 t4_fw_bye(sc, sc->mbox);
1518
1519         if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1520                 pci_release_msi(dev);
1521
1522         if (sc->regs_res)
1523                 bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1524                     sc->regs_res);
1525
1526         if (sc->udbs_res)
1527                 bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1528                     sc->udbs_res);
1529
1530         if (sc->msix_res)
1531                 bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1532                     sc->msix_res);
1533
1534         if (sc->l2t)
1535                 t4_free_l2t(sc->l2t);
1536         if (sc->smt)
1537                 t4_free_smt(sc->smt);
1538 #ifdef RATELIMIT
1539         t4_free_etid_table(sc);
1540 #endif
1541         if (sc->key_map)
1542                 vmem_destroy(sc->key_map);
1543 #ifdef INET6
1544         t4_destroy_clip_table(sc);
1545 #endif
1546
1547 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1548         free(sc->sge.ofld_txq, M_CXGBE);
1549 #endif
1550 #ifdef TCP_OFFLOAD
1551         free(sc->sge.ofld_rxq, M_CXGBE);
1552 #endif
1553 #ifdef DEV_NETMAP
1554         free(sc->sge.nm_rxq, M_CXGBE);
1555         free(sc->sge.nm_txq, M_CXGBE);
1556 #endif
1557         free(sc->irq, M_CXGBE);
1558         free(sc->sge.rxq, M_CXGBE);
1559         free(sc->sge.txq, M_CXGBE);
1560         free(sc->sge.ctrlq, M_CXGBE);
1561         free(sc->sge.iqmap, M_CXGBE);
1562         free(sc->sge.eqmap, M_CXGBE);
1563         free(sc->tids.ftid_tab, M_CXGBE);
1564         free(sc->tids.hpftid_tab, M_CXGBE);
1565         free_hftid_hash(&sc->tids);
1566         free(sc->tids.atid_tab, M_CXGBE);
1567         free(sc->tids.tid_tab, M_CXGBE);
1568         free(sc->tt.tls_rx_ports, M_CXGBE);
1569         t4_destroy_dma_tag(sc);
1570         if (mtx_initialized(&sc->sc_lock)) {
1571                 sx_xlock(&t4_list_lock);
1572                 SLIST_REMOVE(&t4_list, sc, adapter, link);
1573                 sx_xunlock(&t4_list_lock);
1574                 mtx_destroy(&sc->sc_lock);
1575         }
1576
1577         callout_drain(&sc->sfl_callout);
1578         if (mtx_initialized(&sc->tids.ftid_lock)) {
1579                 mtx_destroy(&sc->tids.ftid_lock);
1580                 cv_destroy(&sc->tids.ftid_cv);
1581         }
1582         if (mtx_initialized(&sc->tids.atid_lock))
1583                 mtx_destroy(&sc->tids.atid_lock);
1584         if (mtx_initialized(&sc->sfl_lock))
1585                 mtx_destroy(&sc->sfl_lock);
1586         if (mtx_initialized(&sc->ifp_lock))
1587                 mtx_destroy(&sc->ifp_lock);
1588         if (mtx_initialized(&sc->reg_lock))
1589                 mtx_destroy(&sc->reg_lock);
1590
1591         if (rw_initialized(&sc->policy_lock)) {
1592                 rw_destroy(&sc->policy_lock);
1593 #ifdef TCP_OFFLOAD
1594                 if (sc->policy != NULL)
1595                         free_offload_policy(sc->policy);
1596 #endif
1597         }
1598
1599         for (i = 0; i < NUM_MEMWIN; i++) {
1600                 struct memwin *mw = &sc->memwin[i];
1601
1602                 if (rw_initialized(&mw->mw_lock))
1603                         rw_destroy(&mw->mw_lock);
1604         }
1605
1606         bzero(sc, sizeof(*sc));
1607
1608         return (0);
1609 }
1610
1611 static int
1612 cxgbe_probe(device_t dev)
1613 {
1614         char buf[128];
1615         struct port_info *pi = device_get_softc(dev);
1616
1617         snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1618         device_set_desc_copy(dev, buf);
1619
1620         return (BUS_PROBE_DEFAULT);
1621 }
1622
1623 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1624     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1625     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS | \
1626     IFCAP_HWRXTSTMP)
1627 #define T4_CAP_ENABLE (T4_CAP)
1628
1629 static int
1630 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1631 {
1632         struct ifnet *ifp;
1633         struct sbuf *sb;
1634
1635         vi->xact_addr_filt = -1;
1636         callout_init(&vi->tick, 1);
1637
1638         /* Allocate an ifnet and set it up */
1639         ifp = if_alloc_dev(IFT_ETHER, dev);
1640         if (ifp == NULL) {
1641                 device_printf(dev, "Cannot allocate ifnet\n");
1642                 return (ENOMEM);
1643         }
1644         vi->ifp = ifp;
1645         ifp->if_softc = vi;
1646
1647         if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1648         ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1649
1650         ifp->if_init = cxgbe_init;
1651         ifp->if_ioctl = cxgbe_ioctl;
1652         ifp->if_transmit = cxgbe_transmit;
1653         ifp->if_qflush = cxgbe_qflush;
1654         ifp->if_get_counter = cxgbe_get_counter;
1655 #ifdef RATELIMIT
1656         ifp->if_snd_tag_alloc = cxgbe_snd_tag_alloc;
1657         ifp->if_snd_tag_modify = cxgbe_snd_tag_modify;
1658         ifp->if_snd_tag_query = cxgbe_snd_tag_query;
1659         ifp->if_snd_tag_free = cxgbe_snd_tag_free;
1660 #endif
1661
1662         ifp->if_capabilities = T4_CAP;
1663         ifp->if_capenable = T4_CAP_ENABLE;
1664 #ifdef TCP_OFFLOAD
1665         if (vi->nofldrxq != 0)
1666                 ifp->if_capabilities |= IFCAP_TOE;
1667 #endif
1668 #ifdef RATELIMIT
1669         if (is_ethoffload(vi->pi->adapter) && vi->nofldtxq != 0) {
1670                 ifp->if_capabilities |= IFCAP_TXRTLMT;
1671                 ifp->if_capenable |= IFCAP_TXRTLMT;
1672         }
1673 #endif
1674         ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1675             CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1676
1677         ifp->if_hw_tsomax = IP_MAXPACKET;
1678         ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS_TSO;
1679 #ifdef RATELIMIT
1680         if (is_ethoffload(vi->pi->adapter) && vi->nofldtxq != 0)
1681                 ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS_EO_TSO;
1682 #endif
1683         ifp->if_hw_tsomaxsegsize = 65536;
1684
1685         ether_ifattach(ifp, vi->hw_addr);
1686 #ifdef DEV_NETMAP
1687         if (vi->nnmrxq != 0)
1688                 cxgbe_nm_attach(vi);
1689 #endif
1690         sb = sbuf_new_auto();
1691         sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1692 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1693         switch (ifp->if_capabilities & (IFCAP_TOE | IFCAP_TXRTLMT)) {
1694         case IFCAP_TOE:
1695                 sbuf_printf(sb, "; %d txq (TOE)", vi->nofldtxq);
1696                 break;
1697         case IFCAP_TOE | IFCAP_TXRTLMT:
1698                 sbuf_printf(sb, "; %d txq (TOE/ETHOFLD)", vi->nofldtxq);
1699                 break;
1700         case IFCAP_TXRTLMT:
1701                 sbuf_printf(sb, "; %d txq (ETHOFLD)", vi->nofldtxq);
1702                 break;
1703         }
1704 #endif
1705 #ifdef TCP_OFFLOAD
1706         if (ifp->if_capabilities & IFCAP_TOE)
1707                 sbuf_printf(sb, ", %d rxq (TOE)", vi->nofldrxq);
1708 #endif
1709 #ifdef DEV_NETMAP
1710         if (ifp->if_capabilities & IFCAP_NETMAP)
1711                 sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1712                     vi->nnmtxq, vi->nnmrxq);
1713 #endif
1714         sbuf_finish(sb);
1715         device_printf(dev, "%s\n", sbuf_data(sb));
1716         sbuf_delete(sb);
1717
1718         vi_sysctls(vi);
1719
1720         return (0);
1721 }
1722
1723 static int
1724 cxgbe_attach(device_t dev)
1725 {
1726         struct port_info *pi = device_get_softc(dev);
1727         struct adapter *sc = pi->adapter;
1728         struct vi_info *vi;
1729         int i, rc;
1730
1731         callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1732
1733         rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1734         if (rc)
1735                 return (rc);
1736
1737         for_each_vi(pi, i, vi) {
1738                 if (i == 0)
1739                         continue;
1740                 vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1741                 if (vi->dev == NULL) {
1742                         device_printf(dev, "failed to add VI %d\n", i);
1743                         continue;
1744                 }
1745                 device_set_softc(vi->dev, vi);
1746         }
1747
1748         cxgbe_sysctls(pi);
1749
1750         bus_generic_attach(dev);
1751
1752         return (0);
1753 }
1754
1755 static void
1756 cxgbe_vi_detach(struct vi_info *vi)
1757 {
1758         struct ifnet *ifp = vi->ifp;
1759
1760         ether_ifdetach(ifp);
1761
1762         /* Let detach proceed even if these fail. */
1763 #ifdef DEV_NETMAP
1764         if (ifp->if_capabilities & IFCAP_NETMAP)
1765                 cxgbe_nm_detach(vi);
1766 #endif
1767         cxgbe_uninit_synchronized(vi);
1768         callout_drain(&vi->tick);
1769         vi_full_uninit(vi);
1770
1771         if_free(vi->ifp);
1772         vi->ifp = NULL;
1773 }
1774
1775 static int
1776 cxgbe_detach(device_t dev)
1777 {
1778         struct port_info *pi = device_get_softc(dev);
1779         struct adapter *sc = pi->adapter;
1780         int rc;
1781
1782         /* Detach the extra VIs first. */
1783         rc = bus_generic_detach(dev);
1784         if (rc)
1785                 return (rc);
1786         device_delete_children(dev);
1787
1788         doom_vi(sc, &pi->vi[0]);
1789
1790         if (pi->flags & HAS_TRACEQ) {
1791                 sc->traceq = -1;        /* cloner should not create ifnet */
1792                 t4_tracer_port_detach(sc);
1793         }
1794
1795         cxgbe_vi_detach(&pi->vi[0]);
1796         callout_drain(&pi->tick);
1797         ifmedia_removeall(&pi->media);
1798
1799         end_synchronized_op(sc, 0);
1800
1801         return (0);
1802 }
1803
1804 static void
1805 cxgbe_init(void *arg)
1806 {
1807         struct vi_info *vi = arg;
1808         struct adapter *sc = vi->pi->adapter;
1809
1810         if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1811                 return;
1812         cxgbe_init_synchronized(vi);
1813         end_synchronized_op(sc, 0);
1814 }
1815
1816 static int
1817 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1818 {
1819         int rc = 0, mtu, flags;
1820         struct vi_info *vi = ifp->if_softc;
1821         struct port_info *pi = vi->pi;
1822         struct adapter *sc = pi->adapter;
1823         struct ifreq *ifr = (struct ifreq *)data;
1824         uint32_t mask;
1825
1826         switch (cmd) {
1827         case SIOCSIFMTU:
1828                 mtu = ifr->ifr_mtu;
1829                 if (mtu < ETHERMIN || mtu > MAX_MTU)
1830                         return (EINVAL);
1831
1832                 rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1833                 if (rc)
1834                         return (rc);
1835                 ifp->if_mtu = mtu;
1836                 if (vi->flags & VI_INIT_DONE) {
1837                         t4_update_fl_bufsize(ifp);
1838                         if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1839                                 rc = update_mac_settings(ifp, XGMAC_MTU);
1840                 }
1841                 end_synchronized_op(sc, 0);
1842                 break;
1843
1844         case SIOCSIFFLAGS:
1845                 rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4flg");
1846                 if (rc)
1847                         return (rc);
1848
1849                 if (ifp->if_flags & IFF_UP) {
1850                         if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1851                                 flags = vi->if_flags;
1852                                 if ((ifp->if_flags ^ flags) &
1853                                     (IFF_PROMISC | IFF_ALLMULTI)) {
1854                                         rc = update_mac_settings(ifp,
1855                                             XGMAC_PROMISC | XGMAC_ALLMULTI);
1856                                 }
1857                         } else {
1858                                 rc = cxgbe_init_synchronized(vi);
1859                         }
1860                         vi->if_flags = ifp->if_flags;
1861                 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1862                         rc = cxgbe_uninit_synchronized(vi);
1863                 }
1864                 end_synchronized_op(sc, 0);
1865                 break;
1866
1867         case SIOCADDMULTI:
1868         case SIOCDELMULTI:
1869                 rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4multi");
1870                 if (rc)
1871                         return (rc);
1872                 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1873                         rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1874                 end_synchronized_op(sc, 0);
1875                 break;
1876
1877         case SIOCSIFCAP:
1878                 rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1879                 if (rc)
1880                         return (rc);
1881
1882                 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1883                 if (mask & IFCAP_TXCSUM) {
1884                         ifp->if_capenable ^= IFCAP_TXCSUM;
1885                         ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1886
1887                         if (IFCAP_TSO4 & ifp->if_capenable &&
1888                             !(IFCAP_TXCSUM & ifp->if_capenable)) {
1889                                 ifp->if_capenable &= ~IFCAP_TSO4;
1890                                 if_printf(ifp,
1891                                     "tso4 disabled due to -txcsum.\n");
1892                         }
1893                 }
1894                 if (mask & IFCAP_TXCSUM_IPV6) {
1895                         ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1896                         ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1897
1898                         if (IFCAP_TSO6 & ifp->if_capenable &&
1899                             !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1900                                 ifp->if_capenable &= ~IFCAP_TSO6;
1901                                 if_printf(ifp,
1902                                     "tso6 disabled due to -txcsum6.\n");
1903                         }
1904                 }
1905                 if (mask & IFCAP_RXCSUM)
1906                         ifp->if_capenable ^= IFCAP_RXCSUM;
1907                 if (mask & IFCAP_RXCSUM_IPV6)
1908                         ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1909
1910                 /*
1911                  * Note that we leave CSUM_TSO alone (it is always set).  The
1912                  * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1913                  * sending a TSO request our way, so it's sufficient to toggle
1914                  * IFCAP_TSOx only.
1915                  */
1916                 if (mask & IFCAP_TSO4) {
1917                         if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1918                             !(IFCAP_TXCSUM & ifp->if_capenable)) {
1919                                 if_printf(ifp, "enable txcsum first.\n");
1920                                 rc = EAGAIN;
1921                                 goto fail;
1922                         }
1923                         ifp->if_capenable ^= IFCAP_TSO4;
1924                 }
1925                 if (mask & IFCAP_TSO6) {
1926                         if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1927                             !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1928                                 if_printf(ifp, "enable txcsum6 first.\n");
1929                                 rc = EAGAIN;
1930                                 goto fail;
1931                         }
1932                         ifp->if_capenable ^= IFCAP_TSO6;
1933                 }
1934                 if (mask & IFCAP_LRO) {
1935 #if defined(INET) || defined(INET6)
1936                         int i;
1937                         struct sge_rxq *rxq;
1938
1939                         ifp->if_capenable ^= IFCAP_LRO;
1940                         for_each_rxq(vi, i, rxq) {
1941                                 if (ifp->if_capenable & IFCAP_LRO)
1942                                         rxq->iq.flags |= IQ_LRO_ENABLED;
1943                                 else
1944                                         rxq->iq.flags &= ~IQ_LRO_ENABLED;
1945                         }
1946 #endif
1947                 }
1948 #ifdef TCP_OFFLOAD
1949                 if (mask & IFCAP_TOE) {
1950                         int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1951
1952                         rc = toe_capability(vi, enable);
1953                         if (rc != 0)
1954                                 goto fail;
1955
1956                         ifp->if_capenable ^= mask;
1957                 }
1958 #endif
1959                 if (mask & IFCAP_VLAN_HWTAGGING) {
1960                         ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1961                         if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1962                                 rc = update_mac_settings(ifp, XGMAC_VLANEX);
1963                 }
1964                 if (mask & IFCAP_VLAN_MTU) {
1965                         ifp->if_capenable ^= IFCAP_VLAN_MTU;
1966
1967                         /* Need to find out how to disable auto-mtu-inflation */
1968                 }
1969                 if (mask & IFCAP_VLAN_HWTSO)
1970                         ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1971                 if (mask & IFCAP_VLAN_HWCSUM)
1972                         ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1973 #ifdef RATELIMIT
1974                 if (mask & IFCAP_TXRTLMT)
1975                         ifp->if_capenable ^= IFCAP_TXRTLMT;
1976 #endif
1977                 if (mask & IFCAP_HWRXTSTMP) {
1978                         int i;
1979                         struct sge_rxq *rxq;
1980
1981                         ifp->if_capenable ^= IFCAP_HWRXTSTMP;
1982                         for_each_rxq(vi, i, rxq) {
1983                                 if (ifp->if_capenable & IFCAP_HWRXTSTMP)
1984                                         rxq->iq.flags |= IQ_RX_TIMESTAMP;
1985                                 else
1986                                         rxq->iq.flags &= ~IQ_RX_TIMESTAMP;
1987                         }
1988                 }
1989
1990 #ifdef VLAN_CAPABILITIES
1991                 VLAN_CAPABILITIES(ifp);
1992 #endif
1993 fail:
1994                 end_synchronized_op(sc, 0);
1995                 break;
1996
1997         case SIOCSIFMEDIA:
1998         case SIOCGIFMEDIA:
1999         case SIOCGIFXMEDIA:
2000                 ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
2001                 break;
2002
2003         case SIOCGI2C: {
2004                 struct ifi2creq i2c;
2005
2006                 rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
2007                 if (rc != 0)
2008                         break;
2009                 if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
2010                         rc = EPERM;
2011                         break;
2012                 }
2013                 if (i2c.len > sizeof(i2c.data)) {
2014                         rc = EINVAL;
2015                         break;
2016                 }
2017                 rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
2018                 if (rc)
2019                         return (rc);
2020                 rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
2021                     i2c.offset, i2c.len, &i2c.data[0]);
2022                 end_synchronized_op(sc, 0);
2023                 if (rc == 0)
2024                         rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
2025                 break;
2026         }
2027
2028         default:
2029                 rc = ether_ioctl(ifp, cmd, data);
2030         }
2031
2032         return (rc);
2033 }
2034
2035 static int
2036 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
2037 {
2038         struct vi_info *vi = ifp->if_softc;
2039         struct port_info *pi = vi->pi;
2040         struct adapter *sc = pi->adapter;
2041         struct sge_txq *txq;
2042         void *items[1];
2043         int rc;
2044
2045         M_ASSERTPKTHDR(m);
2046         MPASS(m->m_nextpkt == NULL);    /* not quite ready for this yet */
2047
2048         if (__predict_false(pi->link_cfg.link_ok == false)) {
2049                 m_freem(m);
2050                 return (ENETDOWN);
2051         }
2052
2053         rc = parse_pkt(sc, &m);
2054         if (__predict_false(rc != 0)) {
2055                 MPASS(m == NULL);                       /* was freed already */
2056                 atomic_add_int(&pi->tx_parse_error, 1); /* rare, atomic is ok */
2057                 return (rc);
2058         }
2059 #ifdef RATELIMIT
2060         if (m->m_pkthdr.snd_tag != NULL) {
2061                 /* EAGAIN tells the stack we are not the correct interface. */
2062                 if (__predict_false(ifp != m->m_pkthdr.snd_tag->ifp)) {
2063                         m_freem(m);
2064                         return (EAGAIN);
2065                 }
2066
2067                 return (ethofld_transmit(ifp, m));
2068         }
2069 #endif
2070
2071         /* Select a txq. */
2072         txq = &sc->sge.txq[vi->first_txq];
2073         if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2074                 txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
2075                     vi->rsrv_noflowq);
2076
2077         items[0] = m;
2078         rc = mp_ring_enqueue(txq->r, items, 1, 4096);
2079         if (__predict_false(rc != 0))
2080                 m_freem(m);
2081
2082         return (rc);
2083 }
2084
2085 static void
2086 cxgbe_qflush(struct ifnet *ifp)
2087 {
2088         struct vi_info *vi = ifp->if_softc;
2089         struct sge_txq *txq;
2090         int i;
2091
2092         /* queues do not exist if !VI_INIT_DONE. */
2093         if (vi->flags & VI_INIT_DONE) {
2094                 for_each_txq(vi, i, txq) {
2095                         TXQ_LOCK(txq);
2096                         txq->eq.flags |= EQ_QFLUSH;
2097                         TXQ_UNLOCK(txq);
2098                         while (!mp_ring_is_idle(txq->r)) {
2099                                 mp_ring_check_drainage(txq->r, 0);
2100                                 pause("qflush", 1);
2101                         }
2102                         TXQ_LOCK(txq);
2103                         txq->eq.flags &= ~EQ_QFLUSH;
2104                         TXQ_UNLOCK(txq);
2105                 }
2106         }
2107         if_qflush(ifp);
2108 }
2109
2110 static uint64_t
2111 vi_get_counter(struct ifnet *ifp, ift_counter c)
2112 {
2113         struct vi_info *vi = ifp->if_softc;
2114         struct fw_vi_stats_vf *s = &vi->stats;
2115
2116         vi_refresh_stats(vi->pi->adapter, vi);
2117
2118         switch (c) {
2119         case IFCOUNTER_IPACKETS:
2120                 return (s->rx_bcast_frames + s->rx_mcast_frames +
2121                     s->rx_ucast_frames);
2122         case IFCOUNTER_IERRORS:
2123                 return (s->rx_err_frames);
2124         case IFCOUNTER_OPACKETS:
2125                 return (s->tx_bcast_frames + s->tx_mcast_frames +
2126                     s->tx_ucast_frames + s->tx_offload_frames);
2127         case IFCOUNTER_OERRORS:
2128                 return (s->tx_drop_frames);
2129         case IFCOUNTER_IBYTES:
2130                 return (s->rx_bcast_bytes + s->rx_mcast_bytes +
2131                     s->rx_ucast_bytes);
2132         case IFCOUNTER_OBYTES:
2133                 return (s->tx_bcast_bytes + s->tx_mcast_bytes +
2134                     s->tx_ucast_bytes + s->tx_offload_bytes);
2135         case IFCOUNTER_IMCASTS:
2136                 return (s->rx_mcast_frames);
2137         case IFCOUNTER_OMCASTS:
2138                 return (s->tx_mcast_frames);
2139         case IFCOUNTER_OQDROPS: {
2140                 uint64_t drops;
2141
2142                 drops = 0;
2143                 if (vi->flags & VI_INIT_DONE) {
2144                         int i;
2145                         struct sge_txq *txq;
2146
2147                         for_each_txq(vi, i, txq)
2148                                 drops += counter_u64_fetch(txq->r->drops);
2149                 }
2150
2151                 return (drops);
2152
2153         }
2154
2155         default:
2156                 return (if_get_counter_default(ifp, c));
2157         }
2158 }
2159
2160 uint64_t
2161 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
2162 {
2163         struct vi_info *vi = ifp->if_softc;
2164         struct port_info *pi = vi->pi;
2165         struct adapter *sc = pi->adapter;
2166         struct port_stats *s = &pi->stats;
2167
2168         if (pi->nvi > 1 || sc->flags & IS_VF)
2169                 return (vi_get_counter(ifp, c));
2170
2171         cxgbe_refresh_stats(sc, pi);
2172
2173         switch (c) {
2174         case IFCOUNTER_IPACKETS:
2175                 return (s->rx_frames);
2176
2177         case IFCOUNTER_IERRORS:
2178                 return (s->rx_jabber + s->rx_runt + s->rx_too_long +
2179                     s->rx_fcs_err + s->rx_len_err);
2180
2181         case IFCOUNTER_OPACKETS:
2182                 return (s->tx_frames);
2183
2184         case IFCOUNTER_OERRORS:
2185                 return (s->tx_error_frames);
2186
2187         case IFCOUNTER_IBYTES:
2188                 return (s->rx_octets);
2189
2190         case IFCOUNTER_OBYTES:
2191                 return (s->tx_octets);
2192
2193         case IFCOUNTER_IMCASTS:
2194                 return (s->rx_mcast_frames);
2195
2196         case IFCOUNTER_OMCASTS:
2197                 return (s->tx_mcast_frames);
2198
2199         case IFCOUNTER_IQDROPS:
2200                 return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
2201                     s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
2202                     s->rx_trunc3 + pi->tnl_cong_drops);
2203
2204         case IFCOUNTER_OQDROPS: {
2205                 uint64_t drops;
2206
2207                 drops = s->tx_drop;
2208                 if (vi->flags & VI_INIT_DONE) {
2209                         int i;
2210                         struct sge_txq *txq;
2211
2212                         for_each_txq(vi, i, txq)
2213                                 drops += counter_u64_fetch(txq->r->drops);
2214                 }
2215
2216                 return (drops);
2217
2218         }
2219
2220         default:
2221                 return (if_get_counter_default(ifp, c));
2222         }
2223 }
2224
2225 /*
2226  * The kernel picks a media from the list we had provided but we still validate
2227  * the requeste.
2228  */
2229 int
2230 cxgbe_media_change(struct ifnet *ifp)
2231 {
2232         struct vi_info *vi = ifp->if_softc;
2233         struct port_info *pi = vi->pi;
2234         struct ifmedia *ifm = &pi->media;
2235         struct link_config *lc = &pi->link_cfg;
2236         struct adapter *sc = pi->adapter;
2237         int rc;
2238
2239         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4mec");
2240         if (rc != 0)
2241                 return (rc);
2242         PORT_LOCK(pi);
2243         if (IFM_SUBTYPE(ifm->ifm_media) == IFM_AUTO) {
2244                 /* ifconfig .. media autoselect */
2245                 if (!(lc->supported & FW_PORT_CAP32_ANEG)) {
2246                         rc = ENOTSUP; /* AN not supported by transceiver */
2247                         goto done;
2248                 }
2249                 lc->requested_aneg = AUTONEG_ENABLE;
2250                 lc->requested_speed = 0;
2251                 lc->requested_fc |= PAUSE_AUTONEG;
2252         } else {
2253                 lc->requested_aneg = AUTONEG_DISABLE;
2254                 lc->requested_speed =
2255                     ifmedia_baudrate(ifm->ifm_media) / 1000000;
2256                 lc->requested_fc = 0;
2257                 if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_RXPAUSE)
2258                         lc->requested_fc |= PAUSE_RX;
2259                 if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_TXPAUSE)
2260                         lc->requested_fc |= PAUSE_TX;
2261         }
2262         if (pi->up_vis > 0) {
2263                 fixup_link_config(pi);
2264                 rc = apply_link_config(pi);
2265         }
2266 done:
2267         PORT_UNLOCK(pi);
2268         end_synchronized_op(sc, 0);
2269         return (rc);
2270 }
2271
2272 /*
2273  * Base media word (without ETHER, pause, link active, etc.) for the port at the
2274  * given speed.
2275  */
2276 static int
2277 port_mword(struct port_info *pi, uint32_t speed)
2278 {
2279
2280         MPASS(speed & M_FW_PORT_CAP32_SPEED);
2281         MPASS(powerof2(speed));
2282
2283         switch(pi->port_type) {
2284         case FW_PORT_TYPE_BT_SGMII:
2285         case FW_PORT_TYPE_BT_XFI:
2286         case FW_PORT_TYPE_BT_XAUI:
2287                 /* BaseT */
2288                 switch (speed) {
2289                 case FW_PORT_CAP32_SPEED_100M:
2290                         return (IFM_100_T);
2291                 case FW_PORT_CAP32_SPEED_1G:
2292                         return (IFM_1000_T);
2293                 case FW_PORT_CAP32_SPEED_10G:
2294                         return (IFM_10G_T);
2295                 }
2296                 break;
2297         case FW_PORT_TYPE_KX4:
2298                 if (speed == FW_PORT_CAP32_SPEED_10G)
2299                         return (IFM_10G_KX4);
2300                 break;
2301         case FW_PORT_TYPE_CX4:
2302                 if (speed == FW_PORT_CAP32_SPEED_10G)
2303                         return (IFM_10G_CX4);
2304                 break;
2305         case FW_PORT_TYPE_KX:
2306                 if (speed == FW_PORT_CAP32_SPEED_1G)
2307                         return (IFM_1000_KX);
2308                 break;
2309         case FW_PORT_TYPE_KR:
2310         case FW_PORT_TYPE_BP_AP:
2311         case FW_PORT_TYPE_BP4_AP:
2312         case FW_PORT_TYPE_BP40_BA:
2313         case FW_PORT_TYPE_KR4_100G:
2314         case FW_PORT_TYPE_KR_SFP28:
2315         case FW_PORT_TYPE_KR_XLAUI:
2316                 switch (speed) {
2317                 case FW_PORT_CAP32_SPEED_1G:
2318                         return (IFM_1000_KX);
2319                 case FW_PORT_CAP32_SPEED_10G:
2320                         return (IFM_10G_KR);
2321                 case FW_PORT_CAP32_SPEED_25G:
2322                         return (IFM_25G_KR);
2323                 case FW_PORT_CAP32_SPEED_40G:
2324                         return (IFM_40G_KR4);
2325                 case FW_PORT_CAP32_SPEED_50G:
2326                         return (IFM_50G_KR2);
2327                 case FW_PORT_CAP32_SPEED_100G:
2328                         return (IFM_100G_KR4);
2329                 }
2330                 break;
2331         case FW_PORT_TYPE_FIBER_XFI:
2332         case FW_PORT_TYPE_FIBER_XAUI:
2333         case FW_PORT_TYPE_SFP:
2334         case FW_PORT_TYPE_QSFP_10G:
2335         case FW_PORT_TYPE_QSA:
2336         case FW_PORT_TYPE_QSFP:
2337         case FW_PORT_TYPE_CR4_QSFP:
2338         case FW_PORT_TYPE_CR_QSFP:
2339         case FW_PORT_TYPE_CR2_QSFP:
2340         case FW_PORT_TYPE_SFP28:
2341                 /* Pluggable transceiver */
2342                 switch (pi->mod_type) {
2343                 case FW_PORT_MOD_TYPE_LR:
2344                         switch (speed) {
2345                         case FW_PORT_CAP32_SPEED_1G:
2346                                 return (IFM_1000_LX);
2347                         case FW_PORT_CAP32_SPEED_10G:
2348                                 return (IFM_10G_LR);
2349                         case FW_PORT_CAP32_SPEED_25G:
2350                                 return (IFM_25G_LR);
2351                         case FW_PORT_CAP32_SPEED_40G:
2352                                 return (IFM_40G_LR4);
2353                         case FW_PORT_CAP32_SPEED_50G:
2354                                 return (IFM_50G_LR2);
2355                         case FW_PORT_CAP32_SPEED_100G:
2356                                 return (IFM_100G_LR4);
2357                         }
2358                         break;
2359                 case FW_PORT_MOD_TYPE_SR:
2360                         switch (speed) {
2361                         case FW_PORT_CAP32_SPEED_1G:
2362                                 return (IFM_1000_SX);
2363                         case FW_PORT_CAP32_SPEED_10G:
2364                                 return (IFM_10G_SR);
2365                         case FW_PORT_CAP32_SPEED_25G:
2366                                 return (IFM_25G_SR);
2367                         case FW_PORT_CAP32_SPEED_40G:
2368                                 return (IFM_40G_SR4);
2369                         case FW_PORT_CAP32_SPEED_50G:
2370                                 return (IFM_50G_SR2);
2371                         case FW_PORT_CAP32_SPEED_100G:
2372                                 return (IFM_100G_SR4);
2373                         }
2374                         break;
2375                 case FW_PORT_MOD_TYPE_ER:
2376                         if (speed == FW_PORT_CAP32_SPEED_10G)
2377                                 return (IFM_10G_ER);
2378                         break;
2379                 case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
2380                 case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
2381                         switch (speed) {
2382                         case FW_PORT_CAP32_SPEED_1G:
2383                                 return (IFM_1000_CX);
2384                         case FW_PORT_CAP32_SPEED_10G:
2385                                 return (IFM_10G_TWINAX);
2386                         case FW_PORT_CAP32_SPEED_25G:
2387                                 return (IFM_25G_CR);
2388                         case FW_PORT_CAP32_SPEED_40G:
2389                                 return (IFM_40G_CR4);
2390                         case FW_PORT_CAP32_SPEED_50G:
2391                                 return (IFM_50G_CR2);
2392                         case FW_PORT_CAP32_SPEED_100G:
2393                                 return (IFM_100G_CR4);
2394                         }
2395                         break;
2396                 case FW_PORT_MOD_TYPE_LRM:
2397                         if (speed == FW_PORT_CAP32_SPEED_10G)
2398                                 return (IFM_10G_LRM);
2399                         break;
2400                 case FW_PORT_MOD_TYPE_NA:
2401                         MPASS(0);       /* Not pluggable? */
2402                         /* fall throough */
2403                 case FW_PORT_MOD_TYPE_ERROR:
2404                 case FW_PORT_MOD_TYPE_UNKNOWN:
2405                 case FW_PORT_MOD_TYPE_NOTSUPPORTED:
2406                         break;
2407                 case FW_PORT_MOD_TYPE_NONE:
2408                         return (IFM_NONE);
2409                 }
2410                 break;
2411         case FW_PORT_TYPE_NONE:
2412                 return (IFM_NONE);
2413         }
2414
2415         return (IFM_UNKNOWN);
2416 }
2417
2418 void
2419 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2420 {
2421         struct vi_info *vi = ifp->if_softc;
2422         struct port_info *pi = vi->pi;
2423         struct adapter *sc = pi->adapter;
2424         struct link_config *lc = &pi->link_cfg;
2425
2426         if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4med") != 0)
2427                 return;
2428         PORT_LOCK(pi);
2429
2430         if (pi->up_vis == 0) {
2431                 /*
2432                  * If all the interfaces are administratively down the firmware
2433                  * does not report transceiver changes.  Refresh port info here
2434                  * so that ifconfig displays accurate ifmedia at all times.
2435                  * This is the only reason we have a synchronized op in this
2436                  * function.  Just PORT_LOCK would have been enough otherwise.
2437                  */
2438                 t4_update_port_info(pi);
2439                 build_medialist(pi);
2440         }
2441
2442         /* ifm_status */
2443         ifmr->ifm_status = IFM_AVALID;
2444         if (lc->link_ok == false)
2445                 goto done;
2446         ifmr->ifm_status |= IFM_ACTIVE;
2447
2448         /* ifm_active */
2449         ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2450         ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2451         if (lc->fc & PAUSE_RX)
2452                 ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2453         if (lc->fc & PAUSE_TX)
2454                 ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2455         ifmr->ifm_active |= port_mword(pi, speed_to_fwcap(lc->speed));
2456 done:
2457         PORT_UNLOCK(pi);
2458         end_synchronized_op(sc, 0);
2459 }
2460
2461 static int
2462 vcxgbe_probe(device_t dev)
2463 {
2464         char buf[128];
2465         struct vi_info *vi = device_get_softc(dev);
2466
2467         snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2468             vi - vi->pi->vi);
2469         device_set_desc_copy(dev, buf);
2470
2471         return (BUS_PROBE_DEFAULT);
2472 }
2473
2474 static int
2475 alloc_extra_vi(struct adapter *sc, struct port_info *pi, struct vi_info *vi)
2476 {
2477         int func, index, rc;
2478         uint32_t param, val;
2479
2480         ASSERT_SYNCHRONIZED_OP(sc);
2481
2482         index = vi - pi->vi;
2483         MPASS(index > 0);       /* This function deals with _extra_ VIs only */
2484         KASSERT(index < nitems(vi_mac_funcs),
2485             ("%s: VI %s doesn't have a MAC func", __func__,
2486             device_get_nameunit(vi->dev)));
2487         func = vi_mac_funcs[index];
2488         rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2489             vi->hw_addr, &vi->rss_size, &vi->vfvld, &vi->vin, func, 0);
2490         if (rc < 0) {
2491                 device_printf(vi->dev, "failed to allocate virtual interface %d"
2492                     "for port %d: %d\n", index, pi->port_id, -rc);
2493                 return (-rc);
2494         }
2495         vi->viid = rc;
2496
2497         if (vi->rss_size == 1) {
2498                 /*
2499                  * This VI didn't get a slice of the RSS table.  Reduce the
2500                  * number of VIs being created (hw.cxgbe.num_vis) or modify the
2501                  * configuration file (nvi, rssnvi for this PF) if this is a
2502                  * problem.
2503                  */
2504                 device_printf(vi->dev, "RSS table not available.\n");
2505                 vi->rss_base = 0xffff;
2506
2507                 return (0);
2508         }
2509
2510         param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2511             V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2512             V_FW_PARAMS_PARAM_YZ(vi->viid);
2513         rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2514         if (rc)
2515                 vi->rss_base = 0xffff;
2516         else {
2517                 MPASS((val >> 16) == vi->rss_size);
2518                 vi->rss_base = val & 0xffff;
2519         }
2520
2521         return (0);
2522 }
2523
2524 static int
2525 vcxgbe_attach(device_t dev)
2526 {
2527         struct vi_info *vi;
2528         struct port_info *pi;
2529         struct adapter *sc;
2530         int rc;
2531
2532         vi = device_get_softc(dev);
2533         pi = vi->pi;
2534         sc = pi->adapter;
2535
2536         rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4via");
2537         if (rc)
2538                 return (rc);
2539         rc = alloc_extra_vi(sc, pi, vi);
2540         end_synchronized_op(sc, 0);
2541         if (rc)
2542                 return (rc);
2543
2544         rc = cxgbe_vi_attach(dev, vi);
2545         if (rc) {
2546                 t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2547                 return (rc);
2548         }
2549         return (0);
2550 }
2551
2552 static int
2553 vcxgbe_detach(device_t dev)
2554 {
2555         struct vi_info *vi;
2556         struct adapter *sc;
2557
2558         vi = device_get_softc(dev);
2559         sc = vi->pi->adapter;
2560
2561         doom_vi(sc, vi);
2562
2563         cxgbe_vi_detach(vi);
2564         t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2565
2566         end_synchronized_op(sc, 0);
2567
2568         return (0);
2569 }
2570
2571 static struct callout fatal_callout;
2572
2573 static void
2574 delayed_panic(void *arg)
2575 {
2576         struct adapter *sc = arg;
2577
2578         panic("%s: panic on fatal error", device_get_nameunit(sc->dev));
2579 }
2580
2581 void
2582 t4_fatal_err(struct adapter *sc, bool fw_error)
2583 {
2584
2585         t4_shutdown_adapter(sc);
2586         log(LOG_ALERT, "%s: encountered fatal error, adapter stopped.\n",
2587             device_get_nameunit(sc->dev));
2588         if (fw_error) {
2589                 ASSERT_SYNCHRONIZED_OP(sc);
2590                 sc->flags |= ADAP_ERR;
2591         } else {
2592                 ADAPTER_LOCK(sc);
2593                 sc->flags |= ADAP_ERR;
2594                 ADAPTER_UNLOCK(sc);
2595         }
2596
2597         if (t4_panic_on_fatal_err) {
2598                 log(LOG_ALERT, "%s: panic on fatal error after 30s",
2599                     device_get_nameunit(sc->dev));
2600                 callout_reset(&fatal_callout, hz * 30, delayed_panic, sc);
2601         }
2602 }
2603
2604 void
2605 t4_add_adapter(struct adapter *sc)
2606 {
2607         sx_xlock(&t4_list_lock);
2608         SLIST_INSERT_HEAD(&t4_list, sc, link);
2609         sx_xunlock(&t4_list_lock);
2610 }
2611
2612 int
2613 t4_map_bars_0_and_4(struct adapter *sc)
2614 {
2615         sc->regs_rid = PCIR_BAR(0);
2616         sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2617             &sc->regs_rid, RF_ACTIVE);
2618         if (sc->regs_res == NULL) {
2619                 device_printf(sc->dev, "cannot map registers.\n");
2620                 return (ENXIO);
2621         }
2622         sc->bt = rman_get_bustag(sc->regs_res);
2623         sc->bh = rman_get_bushandle(sc->regs_res);
2624         sc->mmio_len = rman_get_size(sc->regs_res);
2625         setbit(&sc->doorbells, DOORBELL_KDB);
2626
2627         sc->msix_rid = PCIR_BAR(4);
2628         sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2629             &sc->msix_rid, RF_ACTIVE);
2630         if (sc->msix_res == NULL) {
2631                 device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2632                 return (ENXIO);
2633         }
2634
2635         return (0);
2636 }
2637
2638 int
2639 t4_map_bar_2(struct adapter *sc)
2640 {
2641
2642         /*
2643          * T4: only iWARP driver uses the userspace doorbells.  There is no need
2644          * to map it if RDMA is disabled.
2645          */
2646         if (is_t4(sc) && sc->rdmacaps == 0)
2647                 return (0);
2648
2649         sc->udbs_rid = PCIR_BAR(2);
2650         sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2651             &sc->udbs_rid, RF_ACTIVE);
2652         if (sc->udbs_res == NULL) {
2653                 device_printf(sc->dev, "cannot map doorbell BAR.\n");
2654                 return (ENXIO);
2655         }
2656         sc->udbs_base = rman_get_virtual(sc->udbs_res);
2657
2658         if (chip_id(sc) >= CHELSIO_T5) {
2659                 setbit(&sc->doorbells, DOORBELL_UDB);
2660 #if defined(__i386__) || defined(__amd64__)
2661                 if (t5_write_combine) {
2662                         int rc, mode;
2663
2664                         /*
2665                          * Enable write combining on BAR2.  This is the
2666                          * userspace doorbell BAR and is split into 128B
2667                          * (UDBS_SEG_SIZE) doorbell regions, each associated
2668                          * with an egress queue.  The first 64B has the doorbell
2669                          * and the second 64B can be used to submit a tx work
2670                          * request with an implicit doorbell.
2671                          */
2672
2673                         rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2674                             rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2675                         if (rc == 0) {
2676                                 clrbit(&sc->doorbells, DOORBELL_UDB);
2677                                 setbit(&sc->doorbells, DOORBELL_WCWR);
2678                                 setbit(&sc->doorbells, DOORBELL_UDBWC);
2679                         } else {
2680                                 device_printf(sc->dev,
2681                                     "couldn't enable write combining: %d\n",
2682                                     rc);
2683                         }
2684
2685                         mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2686                         t4_write_reg(sc, A_SGE_STAT_CFG,
2687                             V_STATSOURCE_T5(7) | mode);
2688                 }
2689 #endif
2690         }
2691         sc->iwt.wc_en = isset(&sc->doorbells, DOORBELL_UDBWC) ? 1 : 0;
2692
2693         return (0);
2694 }
2695
2696 struct memwin_init {
2697         uint32_t base;
2698         uint32_t aperture;
2699 };
2700
2701 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2702         { MEMWIN0_BASE, MEMWIN0_APERTURE },
2703         { MEMWIN1_BASE, MEMWIN1_APERTURE },
2704         { MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2705 };
2706
2707 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2708         { MEMWIN0_BASE, MEMWIN0_APERTURE },
2709         { MEMWIN1_BASE, MEMWIN1_APERTURE },
2710         { MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2711 };
2712
2713 static void
2714 setup_memwin(struct adapter *sc)
2715 {
2716         const struct memwin_init *mw_init;
2717         struct memwin *mw;
2718         int i;
2719         uint32_t bar0;
2720
2721         if (is_t4(sc)) {
2722                 /*
2723                  * Read low 32b of bar0 indirectly via the hardware backdoor
2724                  * mechanism.  Works from within PCI passthrough environments
2725                  * too, where rman_get_start() can return a different value.  We
2726                  * need to program the T4 memory window decoders with the actual
2727                  * addresses that will be coming across the PCIe link.
2728                  */
2729                 bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2730                 bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2731
2732                 mw_init = &t4_memwin[0];
2733         } else {
2734                 /* T5+ use the relative offset inside the PCIe BAR */
2735                 bar0 = 0;
2736
2737                 mw_init = &t5_memwin[0];
2738         }
2739
2740         for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2741                 rw_init(&mw->mw_lock, "memory window access");
2742                 mw->mw_base = mw_init->base;
2743                 mw->mw_aperture = mw_init->aperture;
2744                 mw->mw_curpos = 0;
2745                 t4_write_reg(sc,
2746                     PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2747                     (mw->mw_base + bar0) | V_BIR(0) |
2748                     V_WINDOW(ilog2(mw->mw_aperture) - 10));
2749                 rw_wlock(&mw->mw_lock);
2750                 position_memwin(sc, i, 0);
2751                 rw_wunlock(&mw->mw_lock);
2752         }
2753
2754         /* flush */
2755         t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2756 }
2757
2758 /*
2759  * Positions the memory window at the given address in the card's address space.
2760  * There are some alignment requirements and the actual position may be at an
2761  * address prior to the requested address.  mw->mw_curpos always has the actual
2762  * position of the window.
2763  */
2764 static void
2765 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2766 {
2767         struct memwin *mw;
2768         uint32_t pf;
2769         uint32_t reg;
2770
2771         MPASS(idx >= 0 && idx < NUM_MEMWIN);
2772         mw = &sc->memwin[idx];
2773         rw_assert(&mw->mw_lock, RA_WLOCKED);
2774
2775         if (is_t4(sc)) {
2776                 pf = 0;
2777                 mw->mw_curpos = addr & ~0xf;    /* start must be 16B aligned */
2778         } else {
2779                 pf = V_PFNUM(sc->pf);
2780                 mw->mw_curpos = addr & ~0x7f;   /* start must be 128B aligned */
2781         }
2782         reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2783         t4_write_reg(sc, reg, mw->mw_curpos | pf);
2784         t4_read_reg(sc, reg);   /* flush */
2785 }
2786
2787 int
2788 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2789     int len, int rw)
2790 {
2791         struct memwin *mw;
2792         uint32_t mw_end, v;
2793
2794         MPASS(idx >= 0 && idx < NUM_MEMWIN);
2795
2796         /* Memory can only be accessed in naturally aligned 4 byte units */
2797         if (addr & 3 || len & 3 || len <= 0)
2798                 return (EINVAL);
2799
2800         mw = &sc->memwin[idx];
2801         while (len > 0) {
2802                 rw_rlock(&mw->mw_lock);
2803                 mw_end = mw->mw_curpos + mw->mw_aperture;
2804                 if (addr >= mw_end || addr < mw->mw_curpos) {
2805                         /* Will need to reposition the window */
2806                         if (!rw_try_upgrade(&mw->mw_lock)) {
2807                                 rw_runlock(&mw->mw_lock);
2808                                 rw_wlock(&mw->mw_lock);
2809                         }
2810                         rw_assert(&mw->mw_lock, RA_WLOCKED);
2811                         position_memwin(sc, idx, addr);
2812                         rw_downgrade(&mw->mw_lock);
2813                         mw_end = mw->mw_curpos + mw->mw_aperture;
2814                 }
2815                 rw_assert(&mw->mw_lock, RA_RLOCKED);
2816                 while (addr < mw_end && len > 0) {
2817                         if (rw == 0) {
2818                                 v = t4_read_reg(sc, mw->mw_base + addr -
2819                                     mw->mw_curpos);
2820                                 *val++ = le32toh(v);
2821                         } else {
2822                                 v = *val++;
2823                                 t4_write_reg(sc, mw->mw_base + addr -
2824                                     mw->mw_curpos, htole32(v));
2825                         }
2826                         addr += 4;
2827                         len -= 4;
2828                 }
2829                 rw_runlock(&mw->mw_lock);
2830         }
2831
2832         return (0);
2833 }
2834
2835 int
2836 alloc_atid_tab(struct tid_info *t, int flags)
2837 {
2838         int i;
2839
2840         MPASS(t->natids > 0);
2841         MPASS(t->atid_tab == NULL);
2842
2843         t->atid_tab = malloc(t->natids * sizeof(*t->atid_tab), M_CXGBE,
2844             M_ZERO | flags);
2845         if (t->atid_tab == NULL)
2846                 return (ENOMEM);
2847         mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
2848         t->afree = t->atid_tab;
2849         t->atids_in_use = 0;
2850         for (i = 1; i < t->natids; i++)
2851                 t->atid_tab[i - 1].next = &t->atid_tab[i];
2852         t->atid_tab[t->natids - 1].next = NULL;
2853
2854         return (0);
2855 }
2856
2857 void
2858 free_atid_tab(struct tid_info *t)
2859 {
2860
2861         KASSERT(t->atids_in_use == 0,
2862             ("%s: %d atids still in use.", __func__, t->atids_in_use));
2863
2864         if (mtx_initialized(&t->atid_lock))
2865                 mtx_destroy(&t->atid_lock);
2866         free(t->atid_tab, M_CXGBE);
2867         t->atid_tab = NULL;
2868 }
2869
2870 int
2871 alloc_atid(struct adapter *sc, void *ctx)
2872 {
2873         struct tid_info *t = &sc->tids;
2874         int atid = -1;
2875
2876         mtx_lock(&t->atid_lock);
2877         if (t->afree) {
2878                 union aopen_entry *p = t->afree;
2879
2880                 atid = p - t->atid_tab;
2881                 MPASS(atid <= M_TID_TID);
2882                 t->afree = p->next;
2883                 p->data = ctx;
2884                 t->atids_in_use++;
2885         }
2886         mtx_unlock(&t->atid_lock);
2887         return (atid);
2888 }
2889
2890 void *
2891 lookup_atid(struct adapter *sc, int atid)
2892 {
2893         struct tid_info *t = &sc->tids;
2894
2895         return (t->atid_tab[atid].data);
2896 }
2897
2898 void
2899 free_atid(struct adapter *sc, int atid)
2900 {
2901         struct tid_info *t = &sc->tids;
2902         union aopen_entry *p = &t->atid_tab[atid];
2903
2904         mtx_lock(&t->atid_lock);
2905         p->next = t->afree;
2906         t->afree = p;
2907         t->atids_in_use--;
2908         mtx_unlock(&t->atid_lock);
2909 }
2910
2911 static void
2912 queue_tid_release(struct adapter *sc, int tid)
2913 {
2914
2915         CXGBE_UNIMPLEMENTED("deferred tid release");
2916 }
2917
2918 void
2919 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
2920 {
2921         struct wrqe *wr;
2922         struct cpl_tid_release *req;
2923
2924         wr = alloc_wrqe(sizeof(*req), ctrlq);
2925         if (wr == NULL) {
2926                 queue_tid_release(sc, tid);     /* defer */
2927                 return;
2928         }
2929         req = wrtod(wr);
2930
2931         INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
2932
2933         t4_wrq_tx(sc, wr);
2934 }
2935
2936 static int
2937 t4_range_cmp(const void *a, const void *b)
2938 {
2939         return ((const struct t4_range *)a)->start -
2940                ((const struct t4_range *)b)->start;
2941 }
2942
2943 /*
2944  * Verify that the memory range specified by the addr/len pair is valid within
2945  * the card's address space.
2946  */
2947 static int
2948 validate_mem_range(struct adapter *sc, uint32_t addr, uint32_t len)
2949 {
2950         struct t4_range mem_ranges[4], *r, *next;
2951         uint32_t em, addr_len;
2952         int i, n, remaining;
2953
2954         /* Memory can only be accessed in naturally aligned 4 byte units */
2955         if (addr & 3 || len & 3 || len == 0)
2956                 return (EINVAL);
2957
2958         /* Enabled memories */
2959         em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2960
2961         r = &mem_ranges[0];
2962         n = 0;
2963         bzero(r, sizeof(mem_ranges));
2964         if (em & F_EDRAM0_ENABLE) {
2965                 addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2966                 r->size = G_EDRAM0_SIZE(addr_len) << 20;
2967                 if (r->size > 0) {
2968                         r->start = G_EDRAM0_BASE(addr_len) << 20;
2969                         if (addr >= r->start &&
2970                             addr + len <= r->start + r->size)
2971                                 return (0);
2972                         r++;
2973                         n++;
2974                 }
2975         }
2976         if (em & F_EDRAM1_ENABLE) {
2977                 addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2978                 r->size = G_EDRAM1_SIZE(addr_len) << 20;
2979                 if (r->size > 0) {
2980                         r->start = G_EDRAM1_BASE(addr_len) << 20;
2981                         if (addr >= r->start &&
2982                             addr + len <= r->start + r->size)
2983                                 return (0);
2984                         r++;
2985                         n++;
2986                 }
2987         }
2988         if (em & F_EXT_MEM_ENABLE) {
2989                 addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2990                 r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2991                 if (r->size > 0) {
2992                         r->start = G_EXT_MEM_BASE(addr_len) << 20;
2993                         if (addr >= r->start &&
2994                             addr + len <= r->start + r->size)
2995                                 return (0);
2996                         r++;
2997                         n++;
2998                 }
2999         }
3000         if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
3001                 addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
3002                 r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
3003                 if (r->size > 0) {
3004                         r->start = G_EXT_MEM1_BASE(addr_len) << 20;
3005                         if (addr >= r->start &&
3006                             addr + len <= r->start + r->size)
3007                                 return (0);
3008                         r++;
3009                         n++;
3010                 }
3011         }
3012         MPASS(n <= nitems(mem_ranges));
3013
3014         if (n > 1) {
3015                 /* Sort and merge the ranges. */
3016                 qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
3017
3018                 /* Start from index 0 and examine the next n - 1 entries. */
3019                 r = &mem_ranges[0];
3020                 for (remaining = n - 1; remaining > 0; remaining--, r++) {
3021
3022                         MPASS(r->size > 0);     /* r is a valid entry. */
3023                         next = r + 1;
3024                         MPASS(next->size > 0);  /* and so is the next one. */
3025
3026                         while (r->start + r->size >= next->start) {
3027                                 /* Merge the next one into the current entry. */
3028                                 r->size = max(r->start + r->size,
3029                                     next->start + next->size) - r->start;
3030                                 n--;    /* One fewer entry in total. */
3031                                 if (--remaining == 0)
3032                                         goto done;      /* short circuit */
3033                                 next++;
3034                         }
3035                         if (next != r + 1) {
3036                                 /*
3037                                  * Some entries were merged into r and next
3038                                  * points to the first valid entry that couldn't
3039                                  * be merged.
3040                                  */
3041                                 MPASS(next->size > 0);  /* must be valid */
3042                                 memcpy(r + 1, next, remaining * sizeof(*r));
3043 #ifdef INVARIANTS
3044                                 /*
3045                                  * This so that the foo->size assertion in the
3046                                  * next iteration of the loop do the right
3047                                  * thing for entries that were pulled up and are
3048                                  * no longer valid.
3049                                  */
3050                                 MPASS(n < nitems(mem_ranges));
3051                                 bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
3052                                     sizeof(struct t4_range));
3053 #endif
3054                         }
3055                 }
3056 done:
3057                 /* Done merging the ranges. */
3058                 MPASS(n > 0);
3059                 r = &mem_ranges[0];
3060                 for (i = 0; i < n; i++, r++) {
3061                         if (addr >= r->start &&
3062                             addr + len <= r->start + r->size)
3063                                 return (0);
3064                 }
3065         }
3066
3067         return (EFAULT);
3068 }
3069
3070 static int
3071 fwmtype_to_hwmtype(int mtype)
3072 {
3073
3074         switch (mtype) {
3075         case FW_MEMTYPE_EDC0:
3076                 return (MEM_EDC0);
3077         case FW_MEMTYPE_EDC1:
3078                 return (MEM_EDC1);
3079         case FW_MEMTYPE_EXTMEM:
3080                 return (MEM_MC0);
3081         case FW_MEMTYPE_EXTMEM1:
3082                 return (MEM_MC1);
3083         default:
3084                 panic("%s: cannot translate fw mtype %d.", __func__, mtype);
3085         }
3086 }
3087
3088 /*
3089  * Verify that the memory range specified by the memtype/offset/len pair is
3090  * valid and lies entirely within the memtype specified.  The global address of
3091  * the start of the range is returned in addr.
3092  */
3093 static int
3094 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, uint32_t len,
3095     uint32_t *addr)
3096 {
3097         uint32_t em, addr_len, maddr;
3098
3099         /* Memory can only be accessed in naturally aligned 4 byte units */
3100         if (off & 3 || len & 3 || len == 0)
3101                 return (EINVAL);
3102
3103         em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
3104         switch (fwmtype_to_hwmtype(mtype)) {
3105         case MEM_EDC0:
3106                 if (!(em & F_EDRAM0_ENABLE))
3107                         return (EINVAL);
3108                 addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
3109                 maddr = G_EDRAM0_BASE(addr_len) << 20;
3110                 break;
3111         case MEM_EDC1:
3112                 if (!(em & F_EDRAM1_ENABLE))
3113                         return (EINVAL);
3114                 addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
3115                 maddr = G_EDRAM1_BASE(addr_len) << 20;
3116                 break;
3117         case MEM_MC:
3118                 if (!(em & F_EXT_MEM_ENABLE))
3119                         return (EINVAL);
3120                 addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
3121                 maddr = G_EXT_MEM_BASE(addr_len) << 20;
3122                 break;
3123         case MEM_MC1:
3124                 if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
3125                         return (EINVAL);
3126                 addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
3127                 maddr = G_EXT_MEM1_BASE(addr_len) << 20;
3128                 break;
3129         default:
3130                 return (EINVAL);
3131         }
3132
3133         *addr = maddr + off;    /* global address */
3134         return (validate_mem_range(sc, *addr, len));
3135 }
3136
3137 static int
3138 fixup_devlog_params(struct adapter *sc)
3139 {
3140         struct devlog_params *dparams = &sc->params.devlog;
3141         int rc;
3142
3143         rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
3144             dparams->size, &dparams->addr);
3145
3146         return (rc);
3147 }
3148
3149 static void
3150 update_nirq(struct intrs_and_queues *iaq, int nports)
3151 {
3152         int extra = T4_EXTRA_INTR;
3153
3154         iaq->nirq = extra;
3155         iaq->nirq += nports * (iaq->nrxq + iaq->nofldrxq);
3156         iaq->nirq += nports * (iaq->num_vis - 1) *
3157             max(iaq->nrxq_vi, iaq->nnmrxq_vi);
3158         iaq->nirq += nports * (iaq->num_vis - 1) * iaq->nofldrxq_vi;
3159 }
3160
3161 /*
3162  * Adjust requirements to fit the number of interrupts available.
3163  */
3164 static void
3165 calculate_iaq(struct adapter *sc, struct intrs_and_queues *iaq, int itype,
3166     int navail)
3167 {
3168         int old_nirq;
3169         const int nports = sc->params.nports;
3170
3171         MPASS(nports > 0);
3172         MPASS(navail > 0);
3173
3174         bzero(iaq, sizeof(*iaq));
3175         iaq->intr_type = itype;
3176         iaq->num_vis = t4_num_vis;
3177         iaq->ntxq = t4_ntxq;
3178         iaq->ntxq_vi = t4_ntxq_vi;
3179         iaq->nrxq = t4_nrxq;
3180         iaq->nrxq_vi = t4_nrxq_vi;
3181 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
3182         if (is_offload(sc) || is_ethoffload(sc)) {
3183                 iaq->nofldtxq = t4_nofldtxq;
3184                 iaq->nofldtxq_vi = t4_nofldtxq_vi;
3185         }
3186 #endif
3187 #ifdef TCP_OFFLOAD
3188         if (is_offload(sc)) {
3189                 iaq->nofldrxq = t4_nofldrxq;
3190                 iaq->nofldrxq_vi = t4_nofldrxq_vi;
3191         }
3192 #endif
3193 #ifdef DEV_NETMAP
3194         iaq->nnmtxq_vi = t4_nnmtxq_vi;
3195         iaq->nnmrxq_vi = t4_nnmrxq_vi;
3196 #endif
3197
3198         update_nirq(iaq, nports);
3199         if (iaq->nirq <= navail &&
3200             (itype != INTR_MSI || powerof2(iaq->nirq))) {
3201                 /*
3202                  * This is the normal case -- there are enough interrupts for
3203                  * everything.
3204                  */
3205                 goto done;
3206         }
3207
3208         /*
3209          * If extra VIs have been configured try reducing their count and see if
3210          * that works.
3211          */
3212         while (iaq->num_vis > 1) {
3213                 iaq->num_vis--;
3214                 update_nirq(iaq, nports);
3215                 if (iaq->nirq <= navail &&
3216                     (itype != INTR_MSI || powerof2(iaq->nirq))) {
3217                         device_printf(sc->dev, "virtual interfaces per port "
3218                             "reduced to %d from %d.  nrxq=%u, nofldrxq=%u, "
3219                             "nrxq_vi=%u nofldrxq_vi=%u, nnmrxq_vi=%u.  "
3220                             "itype %d, navail %u, nirq %d.\n",
3221                             iaq->num_vis, t4_num_vis, iaq->nrxq, iaq->nofldrxq,
3222                             iaq->nrxq_vi, iaq->nofldrxq_vi, iaq->nnmrxq_vi,
3223                             itype, navail, iaq->nirq);
3224                         goto done;
3225                 }
3226         }
3227
3228         /*
3229          * Extra VIs will not be created.  Log a message if they were requested.
3230          */
3231         MPASS(iaq->num_vis == 1);
3232         iaq->ntxq_vi = iaq->nrxq_vi = 0;
3233         iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
3234         iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
3235         if (iaq->num_vis != t4_num_vis) {
3236                 device_printf(sc->dev, "extra virtual interfaces disabled.  "
3237                     "nrxq=%u, nofldrxq=%u, nrxq_vi=%u nofldrxq_vi=%u, "
3238                     "nnmrxq_vi=%u.  itype %d, navail %u, nirq %d.\n",
3239                     iaq->nrxq, iaq->nofldrxq, iaq->nrxq_vi, iaq->nofldrxq_vi,
3240                     iaq->nnmrxq_vi, itype, navail, iaq->nirq);
3241         }
3242
3243         /*
3244          * Keep reducing the number of NIC rx queues to the next lower power of
3245          * 2 (for even RSS distribution) and halving the TOE rx queues and see
3246          * if that works.
3247          */
3248         do {
3249                 if (iaq->nrxq > 1) {
3250                         do {
3251                                 iaq->nrxq--;
3252                         } while (!powerof2(iaq->nrxq));
3253                 }
3254                 if (iaq->nofldrxq > 1)
3255                         iaq->nofldrxq >>= 1;
3256
3257                 old_nirq = iaq->nirq;
3258                 update_nirq(iaq, nports);
3259                 if (iaq->nirq <= navail &&
3260                     (itype != INTR_MSI || powerof2(iaq->nirq))) {
3261                         device_printf(sc->dev, "running with reduced number of "
3262                             "rx queues because of shortage of interrupts.  "
3263                             "nrxq=%u, nofldrxq=%u.  "
3264                             "itype %d, navail %u, nirq %d.\n", iaq->nrxq,
3265                             iaq->nofldrxq, itype, navail, iaq->nirq);
3266                         goto done;
3267                 }
3268         } while (old_nirq != iaq->nirq);
3269
3270         /* One interrupt for everything.  Ugh. */
3271         device_printf(sc->dev, "running with minimal number of queues.  "
3272             "itype %d, navail %u.\n", itype, navail);
3273         iaq->nirq = 1;
3274         MPASS(iaq->nrxq == 1);
3275         iaq->ntxq = 1;
3276         if (iaq->nofldrxq > 1)
3277                 iaq->nofldtxq = 1;
3278 done:
3279         MPASS(iaq->num_vis > 0);
3280         if (iaq->num_vis > 1) {
3281                 MPASS(iaq->nrxq_vi > 0);
3282                 MPASS(iaq->ntxq_vi > 0);
3283         }
3284         MPASS(iaq->nirq > 0);
3285         MPASS(iaq->nrxq > 0);
3286         MPASS(iaq->ntxq > 0);
3287         if (itype == INTR_MSI) {
3288                 MPASS(powerof2(iaq->nirq));
3289         }
3290 }
3291
3292 static int
3293 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
3294 {
3295         int rc, itype, navail, nalloc;
3296
3297         for (itype = INTR_MSIX; itype; itype >>= 1) {
3298
3299                 if ((itype & t4_intr_types) == 0)
3300                         continue;       /* not allowed */
3301
3302                 if (itype == INTR_MSIX)
3303                         navail = pci_msix_count(sc->dev);
3304                 else if (itype == INTR_MSI)
3305                         navail = pci_msi_count(sc->dev);
3306                 else
3307                         navail = 1;
3308 restart:
3309                 if (navail == 0)
3310                         continue;
3311
3312                 calculate_iaq(sc, iaq, itype, navail);
3313                 nalloc = iaq->nirq;
3314                 rc = 0;
3315                 if (itype == INTR_MSIX)
3316                         rc = pci_alloc_msix(sc->dev, &nalloc);
3317                 else if (itype == INTR_MSI)
3318                         rc = pci_alloc_msi(sc->dev, &nalloc);
3319
3320                 if (rc == 0 && nalloc > 0) {
3321                         if (nalloc == iaq->nirq)
3322                                 return (0);
3323
3324                         /*
3325                          * Didn't get the number requested.  Use whatever number
3326                          * the kernel is willing to allocate.
3327                          */
3328                         device_printf(sc->dev, "fewer vectors than requested, "
3329                             "type=%d, req=%d, rcvd=%d; will downshift req.\n",
3330                             itype, iaq->nirq, nalloc);
3331                         pci_release_msi(sc->dev);
3332                         navail = nalloc;
3333                         goto restart;
3334                 }
3335
3336                 device_printf(sc->dev,
3337                     "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
3338                     itype, rc, iaq->nirq, nalloc);
3339         }
3340
3341         device_printf(sc->dev,
3342             "failed to find a usable interrupt type.  "
3343             "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
3344             pci_msix_count(sc->dev), pci_msi_count(sc->dev));
3345
3346         return (ENXIO);
3347 }
3348
3349 #define FW_VERSION(chip) ( \
3350     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
3351     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
3352     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
3353     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
3354 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
3355
3356 /* Just enough of fw_hdr to cover all version info. */
3357 struct fw_h {
3358         __u8    ver;
3359         __u8    chip;
3360         __be16  len512;
3361         __be32  fw_ver;
3362         __be32  tp_microcode_ver;
3363         __u8    intfver_nic;
3364         __u8    intfver_vnic;
3365         __u8    intfver_ofld;
3366         __u8    intfver_ri;
3367         __u8    intfver_iscsipdu;
3368         __u8    intfver_iscsi;
3369         __u8    intfver_fcoepdu;
3370         __u8    intfver_fcoe;
3371 };
3372 /* Spot check a couple of fields. */
3373 CTASSERT(offsetof(struct fw_h, fw_ver) == offsetof(struct fw_hdr, fw_ver));
3374 CTASSERT(offsetof(struct fw_h, intfver_nic) == offsetof(struct fw_hdr, intfver_nic));
3375 CTASSERT(offsetof(struct fw_h, intfver_fcoe) == offsetof(struct fw_hdr, intfver_fcoe));
3376
3377 struct fw_info {
3378         uint8_t chip;
3379         char *kld_name;
3380         char *fw_mod_name;
3381         struct fw_h fw_h;
3382 } fw_info[] = {
3383         {
3384                 .chip = CHELSIO_T4,
3385                 .kld_name = "t4fw_cfg",
3386                 .fw_mod_name = "t4fw",
3387                 .fw_h = {
3388                         .chip = FW_HDR_CHIP_T4,
3389                         .fw_ver = htobe32(FW_VERSION(T4)),
3390                         .intfver_nic = FW_INTFVER(T4, NIC),
3391                         .intfver_vnic = FW_INTFVER(T4, VNIC),
3392                         .intfver_ofld = FW_INTFVER(T4, OFLD),
3393                         .intfver_ri = FW_INTFVER(T4, RI),
3394                         .intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
3395                         .intfver_iscsi = FW_INTFVER(T4, ISCSI),
3396                         .intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
3397                         .intfver_fcoe = FW_INTFVER(T4, FCOE),
3398                 },
3399         }, {
3400                 .chip = CHELSIO_T5,
3401                 .kld_name = "t5fw_cfg",
3402                 .fw_mod_name = "t5fw",
3403                 .fw_h = {
3404                         .chip = FW_HDR_CHIP_T5,
3405                         .fw_ver = htobe32(FW_VERSION(T5)),
3406                         .intfver_nic = FW_INTFVER(T5, NIC),
3407                         .intfver_vnic = FW_INTFVER(T5, VNIC),
3408                         .intfver_ofld = FW_INTFVER(T5, OFLD),
3409                         .intfver_ri = FW_INTFVER(T5, RI),
3410                         .intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
3411                         .intfver_iscsi = FW_INTFVER(T5, ISCSI),
3412                         .intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
3413                         .intfver_fcoe = FW_INTFVER(T5, FCOE),
3414                 },
3415         }, {
3416                 .chip = CHELSIO_T6,
3417                 .kld_name = "t6fw_cfg",
3418                 .fw_mod_name = "t6fw",
3419                 .fw_h = {
3420                         .chip = FW_HDR_CHIP_T6,
3421                         .fw_ver = htobe32(FW_VERSION(T6)),
3422                         .intfver_nic = FW_INTFVER(T6, NIC),
3423                         .intfver_vnic = FW_INTFVER(T6, VNIC),
3424                         .intfver_ofld = FW_INTFVER(T6, OFLD),
3425                         .intfver_ri = FW_INTFVER(T6, RI),
3426                         .intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3427                         .intfver_iscsi = FW_INTFVER(T6, ISCSI),
3428                         .intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3429                         .intfver_fcoe = FW_INTFVER(T6, FCOE),
3430                 },
3431         }
3432 };
3433
3434 static struct fw_info *
3435 find_fw_info(int chip)
3436 {
3437         int i;
3438
3439         for (i = 0; i < nitems(fw_info); i++) {
3440                 if (fw_info[i].chip == chip)
3441                         return (&fw_info[i]);
3442         }
3443         return (NULL);
3444 }
3445
3446 /*
3447  * Is the given firmware API compatible with the one the driver was compiled
3448  * with?
3449  */
3450 static int
3451 fw_compatible(const struct fw_h *hdr1, const struct fw_h *hdr2)
3452 {
3453
3454         /* short circuit if it's the exact same firmware version */
3455         if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
3456                 return (1);
3457
3458         /*
3459          * XXX: Is this too conservative?  Perhaps I should limit this to the
3460          * features that are supported in the driver.
3461          */
3462 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
3463         if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
3464             SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
3465             SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
3466                 return (1);
3467 #undef SAME_INTF
3468
3469         return (0);
3470 }
3471
3472 static int
3473 load_fw_module(struct adapter *sc, const struct firmware **dcfg,
3474     const struct firmware **fw)
3475 {
3476         struct fw_info *fw_info;
3477
3478         *dcfg = NULL;
3479         if (fw != NULL)
3480                 *fw = NULL;
3481
3482         fw_info = find_fw_info(chip_id(sc));
3483         if (fw_info == NULL) {
3484                 device_printf(sc->dev,
3485                     "unable to look up firmware information for chip %d.\n",
3486                     chip_id(sc));
3487                 return (EINVAL);
3488         }
3489
3490         *dcfg = firmware_get(fw_info->kld_name);
3491         if (*dcfg != NULL) {
3492                 if (fw != NULL)
3493                         *fw = firmware_get(fw_info->fw_mod_name);
3494                 return (0);
3495         }
3496
3497         return (ENOENT);
3498 }
3499
3500 static void
3501 unload_fw_module(struct adapter *sc, const struct firmware *dcfg,
3502     const struct firmware *fw)
3503 {
3504
3505         if (fw != NULL)
3506                 firmware_put(fw, FIRMWARE_UNLOAD);
3507         if (dcfg != NULL)
3508                 firmware_put(dcfg, FIRMWARE_UNLOAD);
3509 }
3510
3511 /*
3512  * Return values:
3513  * 0 means no firmware install attempted.
3514  * ERESTART means a firmware install was attempted and was successful.
3515  * +ve errno means a firmware install was attempted but failed.
3516  */
3517 static int
3518 install_kld_firmware(struct adapter *sc, struct fw_h *card_fw,
3519     const struct fw_h *drv_fw, const char *reason, int *already)
3520 {
3521         const struct firmware *cfg, *fw;
3522         const uint32_t c = be32toh(card_fw->fw_ver);
3523         uint32_t d, k;
3524         int rc, fw_install;
3525         struct fw_h bundled_fw;
3526         bool load_attempted;
3527
3528         cfg = fw = NULL;
3529         load_attempted = false;
3530         fw_install = t4_fw_install < 0 ? -t4_fw_install : t4_fw_install;
3531
3532         memcpy(&bundled_fw, drv_fw, sizeof(bundled_fw));
3533         if (t4_fw_install < 0) {
3534                 rc = load_fw_module(sc, &cfg, &fw);
3535                 if (rc != 0 || fw == NULL) {
3536                         device_printf(sc->dev,
3537                             "failed to load firmware module: %d. cfg %p, fw %p;"
3538                             " will use compiled-in firmware version for"
3539                             "hw.cxgbe.fw_install checks.\n",
3540                             rc, cfg, fw);
3541                 } else {
3542                         memcpy(&bundled_fw, fw->data, sizeof(bundled_fw));
3543                 }
3544                 load_attempted = true;
3545         }
3546         d = be32toh(bundled_fw.fw_ver);
3547
3548         if (reason != NULL)
3549                 goto install;
3550
3551         if ((sc->flags & FW_OK) == 0) {
3552
3553                 if (c == 0xffffffff) {
3554                         reason = "missing";
3555                         goto install;
3556                 }
3557
3558                 rc = 0;
3559                 goto done;
3560         }
3561
3562         if (!fw_compatible(card_fw, &bundled_fw)) {
3563                 reason = "incompatible or unusable";
3564                 goto install;
3565         }
3566
3567         if (d > c) {
3568                 reason = "older than the version bundled with this driver";
3569                 goto install;
3570         }
3571
3572         if (fw_install == 2 && d != c) {
3573                 reason = "different than the version bundled with this driver";
3574                 goto install;
3575         }
3576
3577         /* No reason to do anything to the firmware already on the card. */
3578         rc = 0;
3579         goto done;
3580
3581 install:
3582         rc = 0;
3583         if ((*already)++)
3584                 goto done;
3585
3586         if (fw_install == 0) {
3587                 device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3588                     "but the driver is prohibited from installing a firmware "
3589                     "on the card.\n",
3590                     G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3591                     G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
3592
3593                 goto done;
3594         }
3595
3596         /*
3597          * We'll attempt to install a firmware.  Load the module first (if it
3598          * hasn't been loaded already).
3599          */
3600         if (!load_attempted) {
3601                 rc = load_fw_module(sc, &cfg, &fw);
3602                 if (rc != 0 || fw == NULL) {
3603                         device_printf(sc->dev,
3604                             "failed to load firmware module: %d. cfg %p, fw %p\n",
3605                             rc, cfg, fw);
3606                         /* carry on */
3607                 }
3608         }
3609         if (fw == NULL) {
3610                 device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3611                     "but the driver cannot take corrective action because it "
3612                     "is unable to load the firmware module.\n",
3613                     G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3614                     G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
3615                 rc = sc->flags & FW_OK ? 0 : ENOENT;
3616                 goto done;
3617         }
3618         k = be32toh(((const struct fw_hdr *)fw->data)->fw_ver);
3619         if (k != d) {
3620                 MPASS(t4_fw_install > 0);
3621                 device_printf(sc->dev,
3622                     "firmware in KLD (%u.%u.%u.%u) is not what the driver was "
3623                     "expecting (%u.%u.%u.%u) and will not be used.\n",
3624                     G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3625                     G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k),
3626                     G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3627                     G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3628                 rc = sc->flags & FW_OK ? 0 : EINVAL;
3629                 goto done;
3630         }
3631
3632         device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3633             "installing firmware %u.%u.%u.%u on card.\n",
3634             G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3635             G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
3636             G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3637             G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3638
3639         rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3640         if (rc != 0) {
3641                 device_printf(sc->dev, "failed to install firmware: %d\n", rc);
3642         } else {
3643                 /* Installed successfully, update the cached header too. */
3644                 rc = ERESTART;
3645                 memcpy(card_fw, fw->data, sizeof(*card_fw));
3646         }
3647 done:
3648         unload_fw_module(sc, cfg, fw);
3649
3650         return (rc);
3651 }
3652
3653 /*
3654  * Establish contact with the firmware and attempt to become the master driver.
3655  *
3656  * A firmware will be installed to the card if needed (if the driver is allowed
3657  * to do so).
3658  */
3659 static int
3660 contact_firmware(struct adapter *sc)
3661 {
3662         int rc, already = 0;
3663         enum dev_state state;
3664         struct fw_info *fw_info;
3665         struct fw_hdr *card_fw;         /* fw on the card */
3666         const struct fw_h *drv_fw;
3667
3668         fw_info = find_fw_info(chip_id(sc));
3669         if (fw_info == NULL) {
3670                 device_printf(sc->dev,
3671                     "unable to look up firmware information for chip %d.\n",
3672                     chip_id(sc));
3673                 return (EINVAL);
3674         }
3675         drv_fw = &fw_info->fw_h;
3676
3677         /* Read the header of the firmware on the card */
3678         card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
3679 restart:
3680         rc = -t4_get_fw_hdr(sc, card_fw);
3681         if (rc != 0) {
3682                 device_printf(sc->dev,
3683                     "unable to read firmware header from card's flash: %d\n",
3684                     rc);
3685                 goto done;
3686         }
3687
3688         rc = install_kld_firmware(sc, (struct fw_h *)card_fw, drv_fw, NULL,
3689             &already);
3690         if (rc == ERESTART)
3691                 goto restart;
3692         if (rc != 0)
3693                 goto done;
3694
3695         rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
3696         if (rc < 0 || state == DEV_STATE_ERR) {
3697                 rc = -rc;
3698                 device_printf(sc->dev,
3699                     "failed to connect to the firmware: %d, %d.  "
3700                     "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
3701 #if 0
3702                 if (install_kld_firmware(sc, (struct fw_h *)card_fw, drv_fw,
3703                     "not responding properly to HELLO", &already) == ERESTART)
3704                         goto restart;
3705 #endif
3706                 goto done;
3707         }
3708         MPASS(be32toh(card_fw->flags) & FW_HDR_FLAGS_RESET_HALT);
3709         sc->flags |= FW_OK;     /* The firmware responded to the FW_HELLO. */
3710
3711         if (rc == sc->pf) {
3712                 sc->flags |= MASTER_PF;
3713                 rc = install_kld_firmware(sc, (struct fw_h *)card_fw, drv_fw,
3714                     NULL, &already);
3715                 if (rc == ERESTART)
3716                         rc = 0;
3717                 else if (rc != 0)
3718                         goto done;
3719         } else if (state == DEV_STATE_UNINIT) {
3720                 /*
3721                  * We didn't get to be the master so we definitely won't be
3722                  * configuring the chip.  It's a bug if someone else hasn't
3723                  * configured it already.
3724                  */
3725                 device_printf(sc->dev, "couldn't be master(%d), "
3726                     "device not already initialized either(%d).  "
3727                     "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
3728                 rc = EPROTO;
3729                 goto done;
3730         } else {
3731                 /*
3732                  * Some other PF is the master and has configured the chip.
3733                  * This is allowed but untested.
3734                  */
3735                 device_printf(sc->dev, "PF%d is master, device state %d.  "
3736                     "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
3737                 snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", rc);
3738                 sc->cfcsum = 0;
3739                 rc = 0;
3740         }
3741 done:
3742         if (rc != 0 && sc->flags & FW_OK) {
3743                 t4_fw_bye(sc, sc->mbox);
3744                 sc->flags &= ~FW_OK;
3745         }
3746         free(card_fw, M_CXGBE);
3747         return (rc);
3748 }
3749
3750 static int
3751 copy_cfg_file_to_card(struct adapter *sc, char *cfg_file,
3752     uint32_t mtype, uint32_t moff)
3753 {
3754         struct fw_info *fw_info;
3755         const struct firmware *dcfg, *rcfg = NULL;
3756         const uint32_t *cfdata;
3757         uint32_t cflen, addr;
3758         int rc;
3759
3760         load_fw_module(sc, &dcfg, NULL);
3761
3762         /* Card specific interpretation of "default". */
3763         if (strncmp(cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3764                 if (pci_get_device(sc->dev) == 0x440a)
3765                         snprintf(cfg_file, sizeof(t4_cfg_file), UWIRE_CF);
3766                 if (is_fpga(sc))
3767                         snprintf(cfg_file, sizeof(t4_cfg_file), FPGA_CF);
3768         }
3769
3770         if (strncmp(cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3771                 if (dcfg == NULL) {
3772                         device_printf(sc->dev,
3773                             "KLD with default config is not available.\n");
3774                         rc = ENOENT;
3775                         goto done;
3776                 }
3777                 cfdata = dcfg->data;
3778                 cflen = dcfg->datasize & ~3;
3779         } else {
3780                 char s[32];
3781
3782                 fw_info = find_fw_info(chip_id(sc));
3783                 if (fw_info == NULL) {
3784                         device_printf(sc->dev,
3785                             "unable to look up firmware information for chip %d.\n",
3786                             chip_id(sc));
3787                         rc = EINVAL;
3788                         goto done;
3789                 }
3790                 snprintf(s, sizeof(s), "%s_%s", fw_info->kld_name, cfg_file);
3791
3792                 rcfg = firmware_get(s);
3793                 if (rcfg == NULL) {
3794                         device_printf(sc->dev,
3795                             "unable to load module \"%s\" for configuration "
3796                             "profile \"%s\".\n", s, cfg_file);
3797                         rc = ENOENT;
3798                         goto done;
3799                 }
3800                 cfdata = rcfg->data;
3801                 cflen = rcfg->datasize & ~3;
3802         }
3803
3804         if (cflen > FLASH_CFG_MAX_SIZE) {
3805                 device_printf(sc->dev,
3806                     "config file too long (%d, max allowed is %d).\n",
3807                     cflen, FLASH_CFG_MAX_SIZE);
3808                 rc = EINVAL;
3809                 goto done;
3810         }
3811
3812         rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3813         if (rc != 0) {
3814                 device_printf(sc->dev,
3815                     "%s: addr (%d/0x%x) or len %d is not valid: %d.\n",
3816                     __func__, mtype, moff, cflen, rc);
3817                 rc = EINVAL;
3818                 goto done;
3819         }
3820         write_via_memwin(sc, 2, addr, cfdata, cflen);
3821 done:
3822         if (rcfg != NULL)
3823                 firmware_put(rcfg, FIRMWARE_UNLOAD);
3824         unload_fw_module(sc, dcfg, NULL);
3825         return (rc);
3826 }
3827
3828 struct caps_allowed {
3829         uint16_t nbmcaps;
3830         uint16_t linkcaps;
3831         uint16_t switchcaps;
3832         uint16_t niccaps;
3833         uint16_t toecaps;
3834         uint16_t rdmacaps;
3835         uint16_t cryptocaps;
3836         uint16_t iscsicaps;
3837         uint16_t fcoecaps;
3838 };
3839
3840 #define FW_PARAM_DEV(param) \
3841         (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3842          V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3843 #define FW_PARAM_PFVF(param) \
3844         (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3845          V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3846
3847 /*
3848  * Provide a configuration profile to the firmware and have it initialize the
3849  * chip accordingly.  This may involve uploading a configuration file to the
3850  * card.
3851  */
3852 static int
3853 apply_cfg_and_initialize(struct adapter *sc, char *cfg_file,
3854     const struct caps_allowed *caps_allowed)
3855 {
3856         int rc;
3857         struct fw_caps_config_cmd caps;
3858         uint32_t mtype, moff, finicsum, cfcsum, param, val;
3859
3860         rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST);
3861         if (rc != 0) {
3862                 device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3863                 return (rc);
3864         }
3865
3866         bzero(&caps, sizeof(caps));
3867         caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3868             F_FW_CMD_REQUEST | F_FW_CMD_READ);
3869         if (strncmp(cfg_file, BUILTIN_CF, sizeof(t4_cfg_file)) == 0) {
3870                 mtype = 0;
3871                 moff = 0;
3872                 caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3873         } else if (strncmp(cfg_file, FLASH_CF, sizeof(t4_cfg_file)) == 0) {
3874                 mtype = FW_MEMTYPE_FLASH;
3875                 moff = t4_flash_cfg_addr(sc);
3876                 caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3877                     V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3878                     V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) |
3879                     FW_LEN16(caps));
3880         } else {
3881                 /*
3882                  * Ask the firmware where it wants us to upload the config file.
3883                  */
3884                 param = FW_PARAM_DEV(CF);
3885                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3886                 if (rc != 0) {
3887                         /* No support for config file?  Shouldn't happen. */
3888                         device_printf(sc->dev,
3889                             "failed to query config file location: %d.\n", rc);
3890                         goto done;
3891                 }
3892                 mtype = G_FW_PARAMS_PARAM_Y(val);
3893                 moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3894                 caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3895                     V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3896                     V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) |
3897                     FW_LEN16(caps));
3898
3899                 rc = copy_cfg_file_to_card(sc, cfg_file, mtype, moff);
3900                 if (rc != 0) {
3901                         device_printf(sc->dev,
3902                             "failed to upload config file to card: %d.\n", rc);
3903                         goto done;
3904                 }
3905         }
3906         rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3907         if (rc != 0) {
3908                 device_printf(sc->dev, "failed to pre-process config file: %d "
3909                     "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
3910                 goto done;
3911         }
3912
3913         finicsum = be32toh(caps.finicsum);
3914         cfcsum = be32toh(caps.cfcsum);  /* actual */
3915         if (finicsum != cfcsum) {
3916                 device_printf(sc->dev,
3917                     "WARNING: config file checksum mismatch: %08x %08x\n",
3918                     finicsum, cfcsum);
3919         }
3920         sc->cfcsum = cfcsum;
3921         snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", cfg_file);
3922
3923         /*
3924          * Let the firmware know what features will (not) be used so it can tune
3925          * things accordingly.
3926          */
3927 #define LIMIT_CAPS(x) do { \
3928         caps.x##caps &= htobe16(caps_allowed->x##caps); \
3929 } while (0)
3930         LIMIT_CAPS(nbm);
3931         LIMIT_CAPS(link);
3932         LIMIT_CAPS(switch);
3933         LIMIT_CAPS(nic);
3934         LIMIT_CAPS(toe);
3935         LIMIT_CAPS(rdma);
3936         LIMIT_CAPS(crypto);
3937         LIMIT_CAPS(iscsi);
3938         LIMIT_CAPS(fcoe);
3939 #undef LIMIT_CAPS
3940         if (caps.niccaps & htobe16(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
3941                 /*
3942                  * TOE and hashfilters are mutually exclusive.  It is a config
3943                  * file or firmware bug if both are reported as available.  Try
3944                  * to cope with the situation in non-debug builds by disabling
3945                  * TOE.
3946                  */
3947                 MPASS(caps.toecaps == 0);
3948
3949                 caps.toecaps = 0;
3950                 caps.rdmacaps = 0;
3951                 caps.iscsicaps = 0;
3952         }
3953
3954         caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3955             F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3956         caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3957         rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3958         if (rc != 0) {
3959                 device_printf(sc->dev,
3960                     "failed to process config file: %d.\n", rc);
3961                 goto done;
3962         }
3963
3964         t4_tweak_chip_settings(sc);
3965         set_params__pre_init(sc);
3966
3967         /* get basic stuff going */
3968         rc = -t4_fw_initialize(sc, sc->mbox);
3969         if (rc != 0) {
3970                 device_printf(sc->dev, "fw_initialize failed: %d.\n", rc);
3971                 goto done;
3972         }
3973 done:
3974         return (rc);
3975 }
3976
3977 /*
3978  * Partition chip resources for use between various PFs, VFs, etc.
3979  */
3980 static int
3981 partition_resources(struct adapter *sc)
3982 {
3983         char cfg_file[sizeof(t4_cfg_file)];
3984         struct caps_allowed caps_allowed;
3985         int rc;
3986         bool fallback;
3987
3988         /* Only the master driver gets to configure the chip resources. */
3989         MPASS(sc->flags & MASTER_PF);
3990
3991 #define COPY_CAPS(x) do { \
3992         caps_allowed.x##caps = t4_##x##caps_allowed; \
3993 } while (0)
3994         bzero(&caps_allowed, sizeof(caps_allowed));
3995         COPY_CAPS(nbm);
3996         COPY_CAPS(link);
3997         COPY_CAPS(switch);
3998         COPY_CAPS(nic);
3999         COPY_CAPS(toe);
4000         COPY_CAPS(rdma);
4001         COPY_CAPS(crypto);
4002         COPY_CAPS(iscsi);
4003         COPY_CAPS(fcoe);
4004         fallback = sc->debug_flags & DF_DISABLE_CFG_RETRY ? false : true;
4005         snprintf(cfg_file, sizeof(cfg_file), "%s", t4_cfg_file);
4006 retry:
4007         rc = apply_cfg_and_initialize(sc, cfg_file, &caps_allowed);
4008         if (rc != 0 && fallback) {
4009                 device_printf(sc->dev,
4010                     "failed (%d) to configure card with \"%s\" profile, "
4011                     "will fall back to a basic configuration and retry.\n",
4012                     rc, cfg_file);
4013                 snprintf(cfg_file, sizeof(cfg_file), "%s", BUILTIN_CF);
4014                 bzero(&caps_allowed, sizeof(caps_allowed));
4015                 COPY_CAPS(nbm);
4016                 COPY_CAPS(link);
4017                 COPY_CAPS(switch);
4018                 COPY_CAPS(nic);
4019                 fallback = false;
4020                 goto retry;
4021         }
4022 #undef COPY_CAPS
4023         return (rc);
4024 }
4025
4026 /*
4027  * Retrieve parameters that are needed (or nice to have) very early.
4028  */
4029 static int
4030 get_params__pre_init(struct adapter *sc)
4031 {
4032         int rc;
4033         uint32_t param[2], val[2];
4034
4035         t4_get_version_info(sc);
4036
4037         snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
4038             G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
4039             G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
4040             G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
4041             G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
4042
4043         snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
4044             G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
4045             G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
4046             G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
4047             G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
4048
4049         snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
4050             G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
4051             G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
4052             G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
4053             G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
4054
4055         snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
4056             G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
4057             G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
4058             G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
4059             G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
4060
4061         param[0] = FW_PARAM_DEV(PORTVEC);
4062         param[1] = FW_PARAM_DEV(CCLK);
4063         rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4064         if (rc != 0) {
4065                 device_printf(sc->dev,
4066                     "failed to query parameters (pre_init): %d.\n", rc);
4067                 return (rc);
4068         }
4069
4070         sc->params.portvec = val[0];
4071         sc->params.nports = bitcount32(val[0]);
4072         sc->params.vpd.cclk = val[1];
4073
4074         /* Read device log parameters. */
4075         rc = -t4_init_devlog_params(sc, 1);
4076         if (rc == 0)
4077                 fixup_devlog_params(sc);
4078         else {
4079                 device_printf(sc->dev,
4080                     "failed to get devlog parameters: %d.\n", rc);
4081                 rc = 0; /* devlog isn't critical for device operation */
4082         }
4083
4084         return (rc);
4085 }
4086
4087 /*
4088  * Any params that need to be set before FW_INITIALIZE.
4089  */
4090 static int
4091 set_params__pre_init(struct adapter *sc)
4092 {
4093         int rc = 0;
4094         uint32_t param, val;
4095
4096         if (chip_id(sc) >= CHELSIO_T6) {
4097                 param = FW_PARAM_DEV(HPFILTER_REGION_SUPPORT);
4098                 val = 1;
4099                 rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4100                 /* firmwares < 1.20.1.0 do not have this param. */
4101                 if (rc == FW_EINVAL && sc->params.fw_vers <
4102                     (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4103                     V_FW_HDR_FW_VER_MICRO(1) | V_FW_HDR_FW_VER_BUILD(0))) {
4104                         rc = 0;
4105                 }
4106                 if (rc != 0) {
4107                         device_printf(sc->dev,
4108                             "failed to enable high priority filters :%d.\n",
4109                             rc);
4110                 }
4111         }
4112
4113         /* Enable opaque VIIDs with firmwares that support it. */
4114         param = FW_PARAM_DEV(OPAQUE_VIID_SMT_EXTN);
4115         val = 1;
4116         rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4117         if (rc == 0 && val == 1)
4118                 sc->params.viid_smt_extn_support = true;
4119         else
4120                 sc->params.viid_smt_extn_support = false;
4121
4122         return (rc);
4123 }
4124
4125 /*
4126  * Retrieve various parameters that are of interest to the driver.  The device
4127  * has been initialized by the firmware at this point.
4128  */
4129 static int
4130 get_params__post_init(struct adapter *sc)
4131 {
4132         int rc;
4133         uint32_t param[7], val[7];
4134         struct fw_caps_config_cmd caps;
4135
4136         param[0] = FW_PARAM_PFVF(IQFLINT_START);
4137         param[1] = FW_PARAM_PFVF(EQ_START);
4138         param[2] = FW_PARAM_PFVF(FILTER_START);
4139         param[3] = FW_PARAM_PFVF(FILTER_END);
4140         param[4] = FW_PARAM_PFVF(L2T_START);
4141         param[5] = FW_PARAM_PFVF(L2T_END);
4142         param[6] = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4143             V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
4144             V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
4145         rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 7, param, val);
4146         if (rc != 0) {
4147                 device_printf(sc->dev,
4148                     "failed to query parameters (post_init): %d.\n", rc);
4149                 return (rc);
4150         }
4151
4152         sc->sge.iq_start = val[0];
4153         sc->sge.eq_start = val[1];
4154         if ((int)val[3] > (int)val[2]) {
4155                 sc->tids.ftid_base = val[2];
4156                 sc->tids.ftid_end = val[3];
4157                 sc->tids.nftids = val[3] - val[2] + 1;
4158         }
4159         sc->vres.l2t.start = val[4];
4160         sc->vres.l2t.size = val[5] - val[4] + 1;
4161         KASSERT(sc->vres.l2t.size <= L2T_SIZE,
4162             ("%s: L2 table size (%u) larger than expected (%u)",
4163             __func__, sc->vres.l2t.size, L2T_SIZE));
4164         sc->params.core_vdd = val[6];
4165
4166         if (chip_id(sc) >= CHELSIO_T6) {
4167
4168                 sc->tids.tid_base = t4_read_reg(sc,
4169                     A_LE_DB_ACTIVE_TABLE_START_INDEX);
4170
4171                 param[0] = FW_PARAM_PFVF(HPFILTER_START);
4172                 param[1] = FW_PARAM_PFVF(HPFILTER_END);
4173                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4174                 if (rc != 0) {
4175                         device_printf(sc->dev,
4176                            "failed to query hpfilter parameters: %d.\n", rc);
4177                         return (rc);
4178                 }
4179                 if ((int)val[1] > (int)val[0]) {
4180                         sc->tids.hpftid_base = val[0];
4181                         sc->tids.hpftid_end = val[1];
4182                         sc->tids.nhpftids = val[1] - val[0] + 1;
4183
4184                         /*
4185                          * These should go off if the layout changes and the
4186                          * driver needs to catch up.
4187                          */
4188                         MPASS(sc->tids.hpftid_base == 0);
4189                         MPASS(sc->tids.tid_base == sc->tids.nhpftids);
4190                 }
4191         }
4192
4193         /*
4194          * MPSBGMAP is queried separately because only recent firmwares support
4195          * it as a parameter and we don't want the compound query above to fail
4196          * on older firmwares.
4197          */
4198         param[0] = FW_PARAM_DEV(MPSBGMAP);
4199         val[0] = 0;
4200         rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4201         if (rc == 0)
4202                 sc->params.mps_bg_map = val[0];
4203         else
4204                 sc->params.mps_bg_map = 0;
4205
4206         /*
4207          * Determine whether the firmware supports the filter2 work request.
4208          * This is queried separately for the same reason as MPSBGMAP above.
4209          */
4210         param[0] = FW_PARAM_DEV(FILTER2_WR);
4211         val[0] = 0;
4212         rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4213         if (rc == 0)
4214                 sc->params.filter2_wr_support = val[0] != 0;
4215         else
4216                 sc->params.filter2_wr_support = 0;
4217
4218         /*
4219          * Find out whether we're allowed to use the ULPTX MEMWRITE DSGL.
4220          * This is queried separately for the same reason as other params above.
4221          */
4222         param[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
4223         val[0] = 0;
4224         rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4225         if (rc == 0)
4226                 sc->params.ulptx_memwrite_dsgl = val[0] != 0;
4227         else
4228                 sc->params.ulptx_memwrite_dsgl = false;
4229
4230         /* get capabilites */
4231         bzero(&caps, sizeof(caps));
4232         caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4233             F_FW_CMD_REQUEST | F_FW_CMD_READ);
4234         caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
4235         rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
4236         if (rc != 0) {
4237                 device_printf(sc->dev,
4238                     "failed to get card capabilities: %d.\n", rc);
4239                 return (rc);
4240         }
4241
4242 #define READ_CAPS(x) do { \
4243         sc->x = htobe16(caps.x); \
4244 } while (0)
4245         READ_CAPS(nbmcaps);
4246         READ_CAPS(linkcaps);
4247         READ_CAPS(switchcaps);
4248         READ_CAPS(niccaps);
4249         READ_CAPS(toecaps);
4250         READ_CAPS(rdmacaps);
4251         READ_CAPS(cryptocaps);
4252         READ_CAPS(iscsicaps);
4253         READ_CAPS(fcoecaps);
4254
4255         if (sc->niccaps & FW_CAPS_CONFIG_NIC_HASHFILTER) {
4256                 MPASS(chip_id(sc) > CHELSIO_T4);
4257                 MPASS(sc->toecaps == 0);
4258                 sc->toecaps = 0;
4259
4260                 param[0] = FW_PARAM_DEV(NTID);
4261                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4262                 if (rc != 0) {
4263                         device_printf(sc->dev,
4264                             "failed to query HASHFILTER parameters: %d.\n", rc);
4265                         return (rc);
4266                 }
4267                 sc->tids.ntids = val[0];
4268                 if (sc->params.fw_vers <
4269                     (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4270                     V_FW_HDR_FW_VER_MICRO(5) | V_FW_HDR_FW_VER_BUILD(0))) {
4271                         MPASS(sc->tids.ntids >= sc->tids.nhpftids);
4272                         sc->tids.ntids -= sc->tids.nhpftids;
4273                 }
4274                 sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
4275                 sc->params.hash_filter = 1;
4276         }
4277         if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
4278                 param[0] = FW_PARAM_PFVF(ETHOFLD_START);
4279                 param[1] = FW_PARAM_PFVF(ETHOFLD_END);
4280                 param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4281                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
4282                 if (rc != 0) {
4283                         device_printf(sc->dev,
4284                             "failed to query NIC parameters: %d.\n", rc);
4285                         return (rc);
4286                 }
4287                 if ((int)val[1] > (int)val[0]) {
4288                         sc->tids.etid_base = val[0];
4289                         sc->tids.etid_end = val[1];
4290                         sc->tids.netids = val[1] - val[0] + 1;
4291                         sc->params.eo_wr_cred = val[2];
4292                         sc->params.ethoffload = 1;
4293                 }
4294         }
4295         if (sc->toecaps) {
4296                 /* query offload-related parameters */
4297                 param[0] = FW_PARAM_DEV(NTID);
4298                 param[1] = FW_PARAM_PFVF(SERVER_START);
4299                 param[2] = FW_PARAM_PFVF(SERVER_END);
4300                 param[3] = FW_PARAM_PFVF(TDDP_START);
4301                 param[4] = FW_PARAM_PFVF(TDDP_END);
4302                 param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4303                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4304                 if (rc != 0) {
4305                         device_printf(sc->dev,
4306                             "failed to query TOE parameters: %d.\n", rc);
4307                         return (rc);
4308                 }
4309                 sc->tids.ntids = val[0];
4310                 if (sc->params.fw_vers <
4311                     (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4312                     V_FW_HDR_FW_VER_MICRO(5) | V_FW_HDR_FW_VER_BUILD(0))) {
4313                         MPASS(sc->tids.ntids >= sc->tids.nhpftids);
4314                         sc->tids.ntids -= sc->tids.nhpftids;
4315                 }
4316                 sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
4317                 if ((int)val[2] > (int)val[1]) {
4318                         sc->tids.stid_base = val[1];
4319                         sc->tids.nstids = val[2] - val[1] + 1;
4320                 }
4321                 sc->vres.ddp.start = val[3];
4322                 sc->vres.ddp.size = val[4] - val[3] + 1;
4323                 sc->params.ofldq_wr_cred = val[5];
4324                 sc->params.offload = 1;
4325         } else {
4326                 /*
4327                  * The firmware attempts memfree TOE configuration for -SO cards
4328                  * and will report toecaps=0 if it runs out of resources (this
4329                  * depends on the config file).  It may not report 0 for other
4330                  * capabilities dependent on the TOE in this case.  Set them to
4331                  * 0 here so that the driver doesn't bother tracking resources
4332                  * that will never be used.
4333                  */
4334                 sc->iscsicaps = 0;
4335                 sc->rdmacaps = 0;
4336         }
4337         if (sc->rdmacaps) {
4338                 param[0] = FW_PARAM_PFVF(STAG_START);
4339                 param[1] = FW_PARAM_PFVF(STAG_END);
4340                 param[2] = FW_PARAM_PFVF(RQ_START);
4341                 param[3] = FW_PARAM_PFVF(RQ_END);
4342                 param[4] = FW_PARAM_PFVF(PBL_START);
4343                 param[5] = FW_PARAM_PFVF(PBL_END);
4344                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4345                 if (rc != 0) {
4346                         device_printf(sc->dev,
4347                             "failed to query RDMA parameters(1): %d.\n", rc);
4348                         return (rc);
4349                 }
4350                 sc->vres.stag.start = val[0];
4351                 sc->vres.stag.size = val[1] - val[0] + 1;
4352                 sc->vres.rq.start = val[2];
4353                 sc->vres.rq.size = val[3] - val[2] + 1;
4354                 sc->vres.pbl.start = val[4];
4355                 sc->vres.pbl.size = val[5] - val[4] + 1;
4356
4357                 param[0] = FW_PARAM_PFVF(SQRQ_START);
4358                 param[1] = FW_PARAM_PFVF(SQRQ_END);
4359                 param[2] = FW_PARAM_PFVF(CQ_START);
4360                 param[3] = FW_PARAM_PFVF(CQ_END);
4361                 param[4] = FW_PARAM_PFVF(OCQ_START);
4362                 param[5] = FW_PARAM_PFVF(OCQ_END);
4363                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4364                 if (rc != 0) {
4365                         device_printf(sc->dev,
4366                             "failed to query RDMA parameters(2): %d.\n", rc);
4367                         return (rc);
4368                 }
4369                 sc->vres.qp.start = val[0];
4370                 sc->vres.qp.size = val[1] - val[0] + 1;
4371                 sc->vres.cq.start = val[2];
4372                 sc->vres.cq.size = val[3] - val[2] + 1;
4373                 sc->vres.ocq.start = val[4];
4374                 sc->vres.ocq.size = val[5] - val[4] + 1;
4375
4376                 param[0] = FW_PARAM_PFVF(SRQ_START);
4377                 param[1] = FW_PARAM_PFVF(SRQ_END);
4378                 param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
4379                 param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
4380                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
4381                 if (rc != 0) {
4382                         device_printf(sc->dev,
4383                             "failed to query RDMA parameters(3): %d.\n", rc);
4384                         return (rc);
4385                 }
4386                 sc->vres.srq.start = val[0];
4387                 sc->vres.srq.size = val[1] - val[0] + 1;
4388                 sc->params.max_ordird_qp = val[2];
4389                 sc->params.max_ird_adapter = val[3];
4390         }
4391         if (sc->iscsicaps) {
4392                 param[0] = FW_PARAM_PFVF(ISCSI_START);
4393                 param[1] = FW_PARAM_PFVF(ISCSI_END);
4394                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4395                 if (rc != 0) {
4396                         device_printf(sc->dev,
4397                             "failed to query iSCSI parameters: %d.\n", rc);
4398                         return (rc);
4399                 }
4400                 sc->vres.iscsi.start = val[0];
4401                 sc->vres.iscsi.size = val[1] - val[0] + 1;
4402         }
4403         if (sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS) {
4404                 param[0] = FW_PARAM_PFVF(TLS_START);
4405                 param[1] = FW_PARAM_PFVF(TLS_END);
4406                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4407                 if (rc != 0) {
4408                         device_printf(sc->dev,
4409                             "failed to query TLS parameters: %d.\n", rc);
4410                         return (rc);
4411                 }
4412                 sc->vres.key.start = val[0];
4413                 sc->vres.key.size = val[1] - val[0] + 1;
4414         }
4415
4416         t4_init_sge_params(sc);
4417
4418         /*
4419          * We've got the params we wanted to query via the firmware.  Now grab
4420          * some others directly from the chip.
4421          */
4422         rc = t4_read_chip_settings(sc);
4423
4424         return (rc);
4425 }
4426
4427 static int
4428 set_params__post_init(struct adapter *sc)
4429 {
4430         uint32_t param, val;
4431 #ifdef TCP_OFFLOAD
4432         int i, v, shift;
4433 #endif
4434
4435         /* ask for encapsulated CPLs */
4436         param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
4437         val = 1;
4438         (void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4439
4440         /* Enable 32b port caps if the firmware supports it. */
4441         param = FW_PARAM_PFVF(PORT_CAPS32);
4442         val = 1;
4443         if (t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val) == 0)
4444                 sc->params.port_caps32 = 1;
4445
4446         /* Let filter + maskhash steer to a part of the VI's RSS region. */
4447         val = 1 << (G_MASKSIZE(t4_read_reg(sc, A_TP_RSS_CONFIG_TNL)) - 1);
4448         t4_set_reg_field(sc, A_TP_RSS_CONFIG_TNL, V_MASKFILTER(M_MASKFILTER),
4449             V_MASKFILTER(val - 1));
4450
4451 #ifdef TCP_OFFLOAD
4452         /*
4453          * Override the TOE timers with user provided tunables.  This is not the
4454          * recommended way to change the timers (the firmware config file is) so
4455          * these tunables are not documented.
4456          *
4457          * All the timer tunables are in microseconds.
4458          */
4459         if (t4_toe_keepalive_idle != 0) {
4460                 v = us_to_tcp_ticks(sc, t4_toe_keepalive_idle);
4461                 v &= M_KEEPALIVEIDLE;
4462                 t4_set_reg_field(sc, A_TP_KEEP_IDLE,
4463                     V_KEEPALIVEIDLE(M_KEEPALIVEIDLE), V_KEEPALIVEIDLE(v));
4464         }
4465         if (t4_toe_keepalive_interval != 0) {
4466                 v = us_to_tcp_ticks(sc, t4_toe_keepalive_interval);
4467                 v &= M_KEEPALIVEINTVL;
4468                 t4_set_reg_field(sc, A_TP_KEEP_INTVL,
4469                     V_KEEPALIVEINTVL(M_KEEPALIVEINTVL), V_KEEPALIVEINTVL(v));
4470         }
4471         if (t4_toe_keepalive_count != 0) {
4472                 v = t4_toe_keepalive_count & M_KEEPALIVEMAXR2;
4473                 t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4474                     V_KEEPALIVEMAXR1(M_KEEPALIVEMAXR1) |
4475                     V_KEEPALIVEMAXR2(M_KEEPALIVEMAXR2),
4476                     V_KEEPALIVEMAXR1(1) | V_KEEPALIVEMAXR2(v));
4477         }
4478         if (t4_toe_rexmt_min != 0) {
4479                 v = us_to_tcp_ticks(sc, t4_toe_rexmt_min);
4480                 v &= M_RXTMIN;
4481                 t4_set_reg_field(sc, A_TP_RXT_MIN,
4482                     V_RXTMIN(M_RXTMIN), V_RXTMIN(v));
4483         }
4484         if (t4_toe_rexmt_max != 0) {
4485                 v = us_to_tcp_ticks(sc, t4_toe_rexmt_max);
4486                 v &= M_RXTMAX;
4487                 t4_set_reg_field(sc, A_TP_RXT_MAX,
4488                     V_RXTMAX(M_RXTMAX), V_RXTMAX(v));
4489         }
4490         if (t4_toe_rexmt_count != 0) {
4491                 v = t4_toe_rexmt_count & M_RXTSHIFTMAXR2;
4492                 t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4493                     V_RXTSHIFTMAXR1(M_RXTSHIFTMAXR1) |
4494                     V_RXTSHIFTMAXR2(M_RXTSHIFTMAXR2),
4495                     V_RXTSHIFTMAXR1(1) | V_RXTSHIFTMAXR2(v));
4496         }
4497         for (i = 0; i < nitems(t4_toe_rexmt_backoff); i++) {
4498                 if (t4_toe_rexmt_backoff[i] != -1) {
4499                         v = t4_toe_rexmt_backoff[i] & M_TIMERBACKOFFINDEX0;
4500                         shift = (i & 3) << 3;
4501                         t4_set_reg_field(sc, A_TP_TCP_BACKOFF_REG0 + (i & ~3),
4502                             M_TIMERBACKOFFINDEX0 << shift, v << shift);
4503                 }
4504         }
4505 #endif
4506         return (0);
4507 }
4508
4509 #undef FW_PARAM_PFVF
4510 #undef FW_PARAM_DEV
4511
4512 static void
4513 t4_set_desc(struct adapter *sc)
4514 {
4515         char buf[128];
4516         struct adapter_params *p = &sc->params;
4517
4518         snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
4519
4520         device_set_desc_copy(sc->dev, buf);
4521 }
4522
4523 static inline void
4524 ifmedia_add4(struct ifmedia *ifm, int m)
4525 {
4526
4527         ifmedia_add(ifm, m, 0, NULL);
4528         ifmedia_add(ifm, m | IFM_ETH_TXPAUSE, 0, NULL);
4529         ifmedia_add(ifm, m | IFM_ETH_RXPAUSE, 0, NULL);
4530         ifmedia_add(ifm, m | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE, 0, NULL);
4531 }
4532
4533 /*
4534  * This is the selected media, which is not quite the same as the active media.
4535  * The media line in ifconfig is "media: Ethernet selected (active)" if selected
4536  * and active are not the same, and "media: Ethernet selected" otherwise.
4537  */
4538 static void
4539 set_current_media(struct port_info *pi)
4540 {
4541         struct link_config *lc;
4542         struct ifmedia *ifm;
4543         int mword;
4544         u_int speed;
4545
4546         PORT_LOCK_ASSERT_OWNED(pi);
4547
4548         /* Leave current media alone if it's already set to IFM_NONE. */
4549         ifm = &pi->media;
4550         if (ifm->ifm_cur != NULL &&
4551             IFM_SUBTYPE(ifm->ifm_cur->ifm_media) == IFM_NONE)
4552                 return;
4553
4554         lc = &pi->link_cfg;
4555         if (lc->requested_aneg != AUTONEG_DISABLE &&
4556             lc->supported & FW_PORT_CAP32_ANEG) {
4557                 ifmedia_set(ifm, IFM_ETHER | IFM_AUTO);
4558                 return;
4559         }
4560         mword = IFM_ETHER | IFM_FDX;
4561         if (lc->requested_fc & PAUSE_TX)
4562                 mword |= IFM_ETH_TXPAUSE;
4563         if (lc->requested_fc & PAUSE_RX)
4564                 mword |= IFM_ETH_RXPAUSE;
4565         if (lc->requested_speed == 0)
4566                 speed = port_top_speed(pi) * 1000;      /* Gbps -> Mbps */
4567         else
4568                 speed = lc->requested_speed;
4569         mword |= port_mword(pi, speed_to_fwcap(speed));
4570         ifmedia_set(ifm, mword);
4571 }
4572
4573 /*
4574  * Returns true if the ifmedia list for the port cannot change.
4575  */
4576 static bool
4577 fixed_ifmedia(struct port_info *pi)
4578 {
4579
4580         return (pi->port_type == FW_PORT_TYPE_BT_SGMII ||
4581             pi->port_type == FW_PORT_TYPE_BT_XFI ||
4582             pi->port_type == FW_PORT_TYPE_BT_XAUI ||
4583             pi->port_type == FW_PORT_TYPE_KX4 ||
4584             pi->port_type == FW_PORT_TYPE_KX ||
4585             pi->port_type == FW_PORT_TYPE_KR ||
4586             pi->port_type == FW_PORT_TYPE_BP_AP ||
4587             pi->port_type == FW_PORT_TYPE_BP4_AP ||
4588             pi->port_type == FW_PORT_TYPE_BP40_BA ||
4589             pi->port_type == FW_PORT_TYPE_KR4_100G ||
4590             pi->port_type == FW_PORT_TYPE_KR_SFP28 ||
4591             pi->port_type == FW_PORT_TYPE_KR_XLAUI);
4592 }
4593
4594 static void
4595 build_medialist(struct port_info *pi)
4596 {
4597         uint32_t ss, speed;
4598         int unknown, mword, bit;
4599         struct link_config *lc;
4600         struct ifmedia *ifm;
4601
4602         PORT_LOCK_ASSERT_OWNED(pi);
4603
4604         if (pi->flags & FIXED_IFMEDIA)
4605                 return;
4606
4607         /*
4608          * Rebuild the ifmedia list.
4609          */
4610         ifm = &pi->media;
4611         ifmedia_removeall(ifm);
4612         lc = &pi->link_cfg;
4613         ss = G_FW_PORT_CAP32_SPEED(lc->supported); /* Supported Speeds */
4614         if (__predict_false(ss == 0)) { /* not supposed to happen. */
4615                 MPASS(ss != 0);
4616 no_media:
4617                 MPASS(LIST_EMPTY(&ifm->ifm_list));
4618                 ifmedia_add(ifm, IFM_ETHER | IFM_NONE, 0, NULL);
4619                 ifmedia_set(ifm, IFM_ETHER | IFM_NONE);
4620                 return;
4621         }
4622
4623         unknown = 0;
4624         for (bit = S_FW_PORT_CAP32_SPEED; bit < fls(ss); bit++) {
4625                 speed = 1 << bit;
4626                 MPASS(speed & M_FW_PORT_CAP32_SPEED);
4627                 if (ss & speed) {
4628                         mword = port_mword(pi, speed);
4629                         if (mword == IFM_NONE) {
4630                                 goto no_media;
4631                         } else if (mword == IFM_UNKNOWN)
4632                                 unknown++;
4633                         else
4634                                 ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | mword);
4635                 }
4636         }
4637         if (unknown > 0) /* Add one unknown for all unknown media types. */
4638                 ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | IFM_UNKNOWN);
4639         if (lc->supported & FW_PORT_CAP32_ANEG)
4640                 ifmedia_add(ifm, IFM_ETHER | IFM_AUTO, 0, NULL);
4641
4642         set_current_media(pi);
4643 }
4644
4645 /*
4646  * Initialize the requested fields in the link config based on driver tunables.
4647  */
4648 static void
4649 init_link_config(struct port_info *pi)
4650 {
4651         struct link_config *lc = &pi->link_cfg;
4652
4653         PORT_LOCK_ASSERT_OWNED(pi);
4654
4655         lc->requested_speed = 0;
4656
4657         if (t4_autoneg == 0)
4658                 lc->requested_aneg = AUTONEG_DISABLE;
4659         else if (t4_autoneg == 1)
4660                 lc->requested_aneg = AUTONEG_ENABLE;
4661         else
4662                 lc->requested_aneg = AUTONEG_AUTO;
4663
4664         lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX |
4665             PAUSE_AUTONEG);
4666
4667         if (t4_fec == -1 || t4_fec & FEC_AUTO)
4668                 lc->requested_fec = FEC_AUTO;
4669         else {
4670                 lc->requested_fec = FEC_NONE;
4671                 if (t4_fec & FEC_RS)
4672                         lc->requested_fec |= FEC_RS;
4673                 if (t4_fec & FEC_BASER_RS)
4674                         lc->requested_fec |= FEC_BASER_RS;
4675         }
4676 }
4677
4678 /*
4679  * Makes sure that all requested settings comply with what's supported by the
4680  * port.  Returns the number of settings that were invalid and had to be fixed.
4681  */
4682 static int
4683 fixup_link_config(struct port_info *pi)
4684 {
4685         int n = 0;
4686         struct link_config *lc = &pi->link_cfg;
4687         uint32_t fwspeed;
4688
4689         PORT_LOCK_ASSERT_OWNED(pi);
4690
4691         /* Speed (when not autonegotiating) */
4692         if (lc->requested_speed != 0) {
4693                 fwspeed = speed_to_fwcap(lc->requested_speed);
4694                 if ((fwspeed & lc->supported) == 0) {
4695                         n++;
4696                         lc->requested_speed = 0;
4697                 }
4698         }
4699
4700         /* Link autonegotiation */
4701         MPASS(lc->requested_aneg == AUTONEG_ENABLE ||
4702             lc->requested_aneg == AUTONEG_DISABLE ||
4703             lc->requested_aneg == AUTONEG_AUTO);
4704         if (lc->requested_aneg == AUTONEG_ENABLE &&
4705             !(lc->supported & FW_PORT_CAP32_ANEG)) {
4706                 n++;
4707                 lc->requested_aneg = AUTONEG_AUTO;
4708         }
4709
4710         /* Flow control */
4711         MPASS((lc->requested_fc & ~(PAUSE_TX | PAUSE_RX | PAUSE_AUTONEG)) == 0);
4712         if (lc->requested_fc & PAUSE_TX &&
4713             !(lc->supported & FW_PORT_CAP32_FC_TX)) {
4714                 n++;
4715                 lc->requested_fc &= ~PAUSE_TX;
4716         }
4717         if (lc->requested_fc & PAUSE_RX &&
4718             !(lc->supported & FW_PORT_CAP32_FC_RX)) {
4719                 n++;
4720                 lc->requested_fc &= ~PAUSE_RX;
4721         }
4722         if (!(lc->requested_fc & PAUSE_AUTONEG) &&
4723             !(lc->supported & FW_PORT_CAP32_FORCE_PAUSE)) {
4724                 n++;
4725                 lc->requested_fc |= PAUSE_AUTONEG;
4726         }
4727
4728         /* FEC */
4729         if ((lc->requested_fec & FEC_RS &&
4730             !(lc->supported & FW_PORT_CAP32_FEC_RS)) ||
4731             (lc->requested_fec & FEC_BASER_RS &&
4732             !(lc->supported & FW_PORT_CAP32_FEC_BASER_RS))) {
4733                 n++;
4734                 lc->requested_fec = FEC_AUTO;
4735         }
4736
4737         return (n);
4738 }
4739
4740 /*
4741  * Apply the requested L1 settings, which are expected to be valid, to the
4742  * hardware.
4743  */
4744 static int
4745 apply_link_config(struct port_info *pi)
4746 {
4747         struct adapter *sc = pi->adapter;
4748         struct link_config *lc = &pi->link_cfg;
4749         int rc;
4750
4751 #ifdef INVARIANTS
4752         ASSERT_SYNCHRONIZED_OP(sc);
4753         PORT_LOCK_ASSERT_OWNED(pi);
4754
4755         if (lc->requested_aneg == AUTONEG_ENABLE)
4756                 MPASS(lc->supported & FW_PORT_CAP32_ANEG);
4757         if (!(lc->requested_fc & PAUSE_AUTONEG))
4758                 MPASS(lc->supported & FW_PORT_CAP32_FORCE_PAUSE);
4759         if (lc->requested_fc & PAUSE_TX)
4760                 MPASS(lc->supported & FW_PORT_CAP32_FC_TX);
4761         if (lc->requested_fc & PAUSE_RX)
4762                 MPASS(lc->supported & FW_PORT_CAP32_FC_RX);
4763         if (lc->requested_fec & FEC_RS)
4764                 MPASS(lc->supported & FW_PORT_CAP32_FEC_RS);
4765         if (lc->requested_fec & FEC_BASER_RS)
4766                 MPASS(lc->supported & FW_PORT_CAP32_FEC_BASER_RS);
4767 #endif
4768         rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
4769         if (rc != 0) {
4770                 /* Don't complain if the VF driver gets back an EPERM. */
4771                 if (!(sc->flags & IS_VF) || rc != FW_EPERM)
4772                         device_printf(pi->dev, "l1cfg failed: %d\n", rc);
4773         } else {
4774                 /*
4775                  * An L1_CFG will almost always result in a link-change event if
4776                  * the link is up, and the driver will refresh the actual
4777                  * fec/fc/etc. when the notification is processed.  If the link
4778                  * is down then the actual settings are meaningless.
4779                  *
4780                  * This takes care of the case where a change in the L1 settings
4781                  * may not result in a notification.
4782                  */
4783                 if (lc->link_ok && !(lc->requested_fc & PAUSE_AUTONEG))
4784                         lc->fc = lc->requested_fc & (PAUSE_TX | PAUSE_RX);
4785         }
4786         return (rc);
4787 }
4788
4789 #define FW_MAC_EXACT_CHUNK      7
4790
4791 /*
4792  * Program the port's XGMAC based on parameters in ifnet.  The caller also
4793  * indicates which parameters should be programmed (the rest are left alone).
4794  */
4795 int
4796 update_mac_settings(struct ifnet *ifp, int flags)
4797 {
4798         int rc = 0;
4799         struct vi_info *vi = ifp->if_softc;
4800         struct port_info *pi = vi->pi;
4801         struct adapter *sc = pi->adapter;
4802         int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
4803
4804         ASSERT_SYNCHRONIZED_OP(sc);
4805         KASSERT(flags, ("%s: not told what to update.", __func__));
4806
4807         if (flags & XGMAC_MTU)
4808                 mtu = ifp->if_mtu;
4809
4810         if (flags & XGMAC_PROMISC)
4811                 promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
4812
4813         if (flags & XGMAC_ALLMULTI)
4814                 allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
4815
4816         if (flags & XGMAC_VLANEX)
4817                 vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
4818
4819         if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
4820                 rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
4821                     allmulti, 1, vlanex, false);
4822                 if (rc) {
4823                         if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
4824                             rc);
4825                         return (rc);
4826                 }
4827         }
4828
4829         if (flags & XGMAC_UCADDR) {
4830                 uint8_t ucaddr[ETHER_ADDR_LEN];
4831
4832                 bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
4833                 rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
4834                     ucaddr, true, &vi->smt_idx);
4835                 if (rc < 0) {
4836                         rc = -rc;
4837                         if_printf(ifp, "change_mac failed: %d\n", rc);
4838                         return (rc);
4839                 } else {
4840                         vi->xact_addr_filt = rc;
4841                         rc = 0;
4842                 }
4843         }
4844
4845         if (flags & XGMAC_MCADDRS) {
4846                 const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
4847                 int del = 1;
4848                 uint64_t hash = 0;
4849                 struct ifmultiaddr *ifma;
4850                 int i = 0, j;
4851
4852                 if_maddr_rlock(ifp);
4853                 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
4854                         if (ifma->ifma_addr->sa_family != AF_LINK)
4855                                 continue;
4856                         mcaddr[i] =
4857                             LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
4858                         MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
4859                         i++;
4860
4861                         if (i == FW_MAC_EXACT_CHUNK) {
4862                                 rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
4863                                     del, i, mcaddr, NULL, &hash, 0);
4864                                 if (rc < 0) {
4865                                         rc = -rc;
4866                                         for (j = 0; j < i; j++) {
4867                                                 if_printf(ifp,
4868                                                     "failed to add mc address"
4869                                                     " %02x:%02x:%02x:"
4870                                                     "%02x:%02x:%02x rc=%d\n",
4871                                                     mcaddr[j][0], mcaddr[j][1],
4872                                                     mcaddr[j][2], mcaddr[j][3],
4873                                                     mcaddr[j][4], mcaddr[j][5],
4874                                                     rc);
4875                                         }
4876                                         goto mcfail;
4877                                 }
4878                                 del = 0;
4879                                 i = 0;
4880                         }
4881                 }
4882                 if (i > 0) {
4883                         rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
4884                             mcaddr, NULL, &hash, 0);
4885                         if (rc < 0) {
4886                                 rc = -rc;
4887                                 for (j = 0; j < i; j++) {
4888                                         if_printf(ifp,
4889                                             "failed to add mc address"
4890                                             " %02x:%02x:%02x:"
4891                                             "%02x:%02x:%02x rc=%d\n",
4892                                             mcaddr[j][0], mcaddr[j][1],
4893                                             mcaddr[j][2], mcaddr[j][3],
4894                                             mcaddr[j][4], mcaddr[j][5],
4895                                             rc);
4896                                 }
4897                                 goto mcfail;
4898                         }
4899                 }
4900
4901                 rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
4902                 if (rc != 0)
4903                         if_printf(ifp, "failed to set mc address hash: %d", rc);
4904 mcfail:
4905                 if_maddr_runlock(ifp);
4906         }
4907
4908         return (rc);
4909 }
4910
4911 /*
4912  * {begin|end}_synchronized_op must be called from the same thread.
4913  */
4914 int
4915 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
4916     char *wmesg)
4917 {
4918         int rc, pri;
4919
4920 #ifdef WITNESS
4921         /* the caller thinks it's ok to sleep, but is it really? */
4922         if (flags & SLEEP_OK)
4923                 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
4924                     "begin_synchronized_op");
4925 #endif
4926
4927         if (INTR_OK)
4928                 pri = PCATCH;
4929         else
4930                 pri = 0;
4931
4932         ADAPTER_LOCK(sc);
4933         for (;;) {
4934
4935                 if (vi && IS_DOOMED(vi)) {
4936                         rc = ENXIO;
4937                         goto done;
4938                 }
4939
4940                 if (!IS_BUSY(sc)) {
4941                         rc = 0;
4942                         break;
4943                 }
4944
4945                 if (!(flags & SLEEP_OK)) {
4946                         rc = EBUSY;
4947                         goto done;
4948                 }
4949
4950                 if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
4951                         rc = EINTR;
4952                         goto done;
4953                 }
4954         }
4955
4956         KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
4957         SET_BUSY(sc);
4958 #ifdef INVARIANTS
4959         sc->last_op = wmesg;
4960         sc->last_op_thr = curthread;
4961         sc->last_op_flags = flags;
4962 #endif
4963
4964 done:
4965         if (!(flags & HOLD_LOCK) || rc)
4966                 ADAPTER_UNLOCK(sc);
4967
4968         return (rc);
4969 }
4970
4971 /*
4972  * Tell if_ioctl and if_init that the VI is going away.  This is
4973  * special variant of begin_synchronized_op and must be paired with a
4974  * call to end_synchronized_op.
4975  */
4976 void
4977 doom_vi(struct adapter *sc, struct vi_info *vi)
4978 {
4979
4980         ADAPTER_LOCK(sc);
4981         SET_DOOMED(vi);
4982         wakeup(&sc->flags);
4983         while (IS_BUSY(sc))
4984                 mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
4985         SET_BUSY(sc);
4986 #ifdef INVARIANTS
4987         sc->last_op = "t4detach";
4988         sc->last_op_thr = curthread;
4989         sc->last_op_flags = 0;
4990 #endif
4991         ADAPTER_UNLOCK(sc);
4992 }
4993
4994 /*
4995  * {begin|end}_synchronized_op must be called from the same thread.
4996  */
4997 void
4998 end_synchronized_op(struct adapter *sc, int flags)
4999 {
5000
5001         if (flags & LOCK_HELD)
5002                 ADAPTER_LOCK_ASSERT_OWNED(sc);
5003         else
5004                 ADAPTER_LOCK(sc);
5005
5006         KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
5007         CLR_BUSY(sc);
5008         wakeup(&sc->flags);
5009         ADAPTER_UNLOCK(sc);
5010 }
5011
5012 static int
5013 cxgbe_init_synchronized(struct vi_info *vi)
5014 {
5015         struct port_info *pi = vi->pi;
5016         struct adapter *sc = pi->adapter;
5017         struct ifnet *ifp = vi->ifp;
5018         int rc = 0, i;
5019         struct sge_txq *txq;
5020
5021         ASSERT_SYNCHRONIZED_OP(sc);
5022
5023         if (ifp->if_drv_flags & IFF_DRV_RUNNING)
5024                 return (0);     /* already running */
5025
5026         if (!(sc->flags & FULL_INIT_DONE) &&
5027             ((rc = adapter_full_init(sc)) != 0))
5028                 return (rc);    /* error message displayed already */
5029
5030         if (!(vi->flags & VI_INIT_DONE) &&
5031             ((rc = vi_full_init(vi)) != 0))
5032                 return (rc); /* error message displayed already */
5033
5034         rc = update_mac_settings(ifp, XGMAC_ALL);
5035         if (rc)
5036                 goto done;      /* error message displayed already */
5037
5038         PORT_LOCK(pi);
5039         if (pi->up_vis == 0) {
5040                 t4_update_port_info(pi);
5041                 fixup_link_config(pi);
5042                 build_medialist(pi);
5043                 apply_link_config(pi);
5044         }
5045
5046         rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
5047         if (rc != 0) {
5048                 if_printf(ifp, "enable_vi failed: %d\n", rc);
5049                 PORT_UNLOCK(pi);
5050                 goto done;
5051         }
5052
5053         /*
5054          * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
5055          * if this changes.
5056          */
5057
5058         for_each_txq(vi, i, txq) {
5059                 TXQ_LOCK(txq);
5060                 txq->eq.flags |= EQ_ENABLED;
5061                 TXQ_UNLOCK(txq);
5062         }
5063
5064         /*
5065          * The first iq of the first port to come up is used for tracing.
5066          */
5067         if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
5068                 sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
5069                 t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
5070                     A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
5071                     V_QUEUENUMBER(sc->traceq));
5072                 pi->flags |= HAS_TRACEQ;
5073         }
5074
5075         /* all ok */
5076         pi->up_vis++;
5077         ifp->if_drv_flags |= IFF_DRV_RUNNING;
5078
5079         if (pi->nvi > 1 || sc->flags & IS_VF)
5080                 callout_reset(&vi->tick, hz, vi_tick, vi);
5081         else
5082                 callout_reset(&pi->tick, hz, cxgbe_tick, pi);
5083         if (pi->link_cfg.link_ok)
5084                 t4_os_link_changed(pi);
5085         PORT_UNLOCK(pi);
5086 done:
5087         if (rc != 0)
5088                 cxgbe_uninit_synchronized(vi);
5089
5090         return (rc);
5091 }
5092
5093 /*
5094  * Idempotent.
5095  */
5096 static int
5097 cxgbe_uninit_synchronized(struct vi_info *vi)
5098 {
5099         struct port_info *pi = vi->pi;
5100         struct adapter *sc = pi->adapter;
5101         struct ifnet *ifp = vi->ifp;
5102         int rc, i;
5103         struct sge_txq *txq;
5104
5105         ASSERT_SYNCHRONIZED_OP(sc);
5106
5107         if (!(vi->flags & VI_INIT_DONE)) {
5108                 if (__predict_false(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
5109                         KASSERT(0, ("uninited VI is running"));
5110                         if_printf(ifp, "uninited VI with running ifnet.  "
5111                             "vi->flags 0x%016lx, if_flags 0x%08x, "
5112                             "if_drv_flags 0x%08x\n", vi->flags, ifp->if_flags,
5113                             ifp->if_drv_flags);
5114                 }
5115                 return (0);
5116         }
5117
5118         /*
5119          * Disable the VI so that all its data in either direction is discarded
5120          * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
5121          * tick) intact as the TP can deliver negative advice or data that it's
5122          * holding in its RAM (for an offloaded connection) even after the VI is
5123          * disabled.
5124          */
5125         rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
5126         if (rc) {
5127                 if_printf(ifp, "disable_vi failed: %d\n", rc);
5128                 return (rc);
5129         }
5130
5131         for_each_txq(vi, i, txq) {
5132                 TXQ_LOCK(txq);
5133                 txq->eq.flags &= ~EQ_ENABLED;
5134                 TXQ_UNLOCK(txq);
5135         }
5136
5137         PORT_LOCK(pi);
5138         if (pi->nvi > 1 || sc->flags & IS_VF)
5139                 callout_stop(&vi->tick);
5140         else
5141                 callout_stop(&pi->tick);
5142         if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
5143                 PORT_UNLOCK(pi);
5144                 return (0);
5145         }
5146         ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
5147         pi->up_vis--;
5148         if (pi->up_vis > 0) {
5149                 PORT_UNLOCK(pi);
5150                 return (0);
5151         }
5152
5153         pi->link_cfg.link_ok = false;
5154         pi->link_cfg.speed = 0;
5155         pi->link_cfg.link_down_rc = 255;
5156         t4_os_link_changed(pi);
5157         PORT_UNLOCK(pi);
5158
5159         return (0);
5160 }
5161
5162 /*
5163  * It is ok for this function to fail midway and return right away.  t4_detach
5164  * will walk the entire sc->irq list and clean up whatever is valid.
5165  */
5166 int
5167 t4_setup_intr_handlers(struct adapter *sc)
5168 {
5169         int rc, rid, p, q, v;
5170         char s[8];
5171         struct irq *irq;
5172         struct port_info *pi;
5173         struct vi_info *vi;
5174         struct sge *sge = &sc->sge;
5175         struct sge_rxq *rxq;
5176 #ifdef TCP_OFFLOAD
5177         struct sge_ofld_rxq *ofld_rxq;
5178 #endif
5179 #ifdef DEV_NETMAP
5180         struct sge_nm_rxq *nm_rxq;
5181 #endif
5182 #ifdef RSS
5183         int nbuckets = rss_getnumbuckets();
5184 #endif
5185
5186         /*
5187          * Setup interrupts.
5188          */
5189         irq = &sc->irq[0];
5190         rid = sc->intr_type == INTR_INTX ? 0 : 1;
5191         if (forwarding_intr_to_fwq(sc))
5192                 return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
5193
5194         /* Multiple interrupts. */
5195         if (sc->flags & IS_VF)
5196                 KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
5197                     ("%s: too few intr.", __func__));
5198         else
5199                 KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
5200                     ("%s: too few intr.", __func__));
5201
5202         /* The first one is always error intr on PFs */
5203         if (!(sc->flags & IS_VF)) {
5204                 rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
5205                 if (rc != 0)
5206                         return (rc);
5207                 irq++;
5208                 rid++;
5209         }
5210
5211         /* The second one is always the firmware event queue (first on VFs) */
5212         rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
5213         if (rc != 0)
5214                 return (rc);
5215         irq++;
5216         rid++;
5217
5218         for_each_port(sc, p) {
5219                 pi = sc->port[p];
5220                 for_each_vi(pi, v, vi) {
5221                         vi->first_intr = rid - 1;
5222
5223                         if (vi->nnmrxq > 0) {
5224                                 int n = max(vi->nrxq, vi->nnmrxq);
5225
5226                                 rxq = &sge->rxq[vi->first_rxq];
5227 #ifdef DEV_NETMAP
5228                                 nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
5229 #endif
5230                                 for (q = 0; q < n; q++) {
5231                                         snprintf(s, sizeof(s), "%x%c%x", p,
5232                                             'a' + v, q);
5233                                         if (q < vi->nrxq)
5234                                                 irq->rxq = rxq++;
5235 #ifdef DEV_NETMAP
5236                                         if (q < vi->nnmrxq)
5237                                                 irq->nm_rxq = nm_rxq++;
5238
5239                                         if (irq->nm_rxq != NULL &&
5240                                             irq->rxq == NULL) {
5241                                                 /* Netmap rx only */
5242                                                 rc = t4_alloc_irq(sc, irq, rid,
5243                                                     t4_nm_intr, irq->nm_rxq, s);
5244                                         }
5245                                         if (irq->nm_rxq != NULL &&
5246                                             irq->rxq != NULL) {
5247                                                 /* NIC and Netmap rx */
5248                                                 rc = t4_alloc_irq(sc, irq, rid,
5249                                                     t4_vi_intr, irq, s);
5250                                         }
5251 #endif
5252                                         if (irq->rxq != NULL &&
5253                                             irq->nm_rxq == NULL) {
5254                                                 /* NIC rx only */
5255                                                 rc = t4_alloc_irq(sc, irq, rid,
5256                                                     t4_intr, irq->rxq, s);
5257                                         }
5258                                         if (rc != 0)
5259                                                 return (rc);
5260 #ifdef RSS
5261                                         if (q < vi->nrxq) {
5262                                                 bus_bind_intr(sc->dev, irq->res,
5263                                                     rss_getcpu(q % nbuckets));
5264                                         }
5265 #endif
5266                                         irq++;
5267                                         rid++;
5268                                         vi->nintr++;
5269                                 }
5270                         } else {
5271                                 for_each_rxq(vi, q, rxq) {
5272                                         snprintf(s, sizeof(s), "%x%c%x", p,
5273                                             'a' + v, q);
5274                                         rc = t4_alloc_irq(sc, irq, rid,
5275                                             t4_intr, rxq, s);
5276                                         if (rc != 0)
5277                                                 return (rc);
5278 #ifdef RSS
5279                                         bus_bind_intr(sc->dev, irq->res,
5280                                             rss_getcpu(q % nbuckets));
5281 #endif
5282                                         irq++;
5283                                         rid++;
5284                                         vi->nintr++;
5285                                 }
5286                         }
5287 #ifdef TCP_OFFLOAD
5288                         for_each_ofld_rxq(vi, q, ofld_rxq) {
5289                                 snprintf(s, sizeof(s), "%x%c%x", p, 'A' + v, q);
5290                                 rc = t4_alloc_irq(sc, irq, rid, t4_intr,
5291                                     ofld_rxq, s);
5292                                 if (rc != 0)
5293                                         return (rc);
5294                                 irq++;
5295                                 rid++;
5296                                 vi->nintr++;
5297                         }
5298 #endif
5299                 }
5300         }
5301         MPASS(irq == &sc->irq[sc->intr_count]);
5302
5303         return (0);
5304 }
5305
5306 int
5307 adapter_full_init(struct adapter *sc)
5308 {
5309         int rc, i;
5310 #ifdef RSS
5311         uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
5312         uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
5313 #endif
5314
5315         ASSERT_SYNCHRONIZED_OP(sc);
5316         ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
5317         KASSERT((sc->flags & FULL_INIT_DONE) == 0,
5318             ("%s: FULL_INIT_DONE already", __func__));
5319
5320         /*
5321          * queues that belong to the adapter (not any particular port).
5322          */
5323         rc = t4_setup_adapter_queues(sc);
5324         if (rc != 0)
5325                 goto done;
5326
5327         for (i = 0; i < nitems(sc->tq); i++) {
5328                 sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
5329                     taskqueue_thread_enqueue, &sc->tq[i]);
5330                 if (sc->tq[i] == NULL) {
5331                         device_printf(sc->dev,
5332                             "failed to allocate task queue %d\n", i);
5333                         rc = ENOMEM;
5334                         goto done;
5335                 }
5336                 taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
5337                     device_get_nameunit(sc->dev), i);
5338         }
5339 #ifdef RSS
5340         MPASS(RSS_KEYSIZE == 40);
5341         rss_getkey((void *)&raw_rss_key[0]);
5342         for (i = 0; i < nitems(rss_key); i++) {
5343                 rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
5344         }
5345         t4_write_rss_key(sc, &rss_key[0], -1, 1);
5346 #endif
5347
5348         if (!(sc->flags & IS_VF))
5349                 t4_intr_enable(sc);
5350         sc->flags |= FULL_INIT_DONE;
5351 done:
5352         if (rc != 0)
5353                 adapter_full_uninit(sc);
5354
5355         return (rc);
5356 }
5357
5358 int
5359 adapter_full_uninit(struct adapter *sc)
5360 {
5361         int i;
5362
5363         ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
5364
5365         t4_teardown_adapter_queues(sc);
5366
5367         for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
5368                 taskqueue_free(sc->tq[i]);
5369                 sc->tq[i] = NULL;
5370         }
5371
5372         sc->flags &= ~FULL_INIT_DONE;
5373
5374         return (0);
5375 }
5376
5377 #ifdef RSS
5378 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
5379     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
5380     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
5381     RSS_HASHTYPE_RSS_UDP_IPV6)
5382
5383 /* Translates kernel hash types to hardware. */
5384 static int
5385 hashconfig_to_hashen(int hashconfig)
5386 {
5387         int hashen = 0;
5388
5389         if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
5390                 hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
5391         if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
5392                 hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
5393         if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
5394                 hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
5395                     F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
5396         }
5397         if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
5398                 hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
5399                     F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
5400         }
5401         if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
5402                 hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
5403         if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
5404                 hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
5405
5406         return (hashen);
5407 }
5408
5409 /* Translates hardware hash types to kernel. */
5410 static int
5411 hashen_to_hashconfig(int hashen)
5412 {
5413         int hashconfig = 0;
5414
5415         if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
5416                 /*
5417                  * If UDP hashing was enabled it must have been enabled for
5418                  * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
5419                  * enabling any 4-tuple hash is nonsense configuration.
5420                  */
5421                 MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5422                     F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
5423
5424                 if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5425                         hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
5426                 if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5427                         hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
5428         }
5429         if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5430                 hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
5431         if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5432                 hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
5433         if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
5434                 hashconfig |= RSS_HASHTYPE_RSS_IPV4;
5435         if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
5436                 hashconfig |= RSS_HASHTYPE_RSS_IPV6;
5437
5438         return (hashconfig);
5439 }
5440 #endif
5441
5442 int
5443 vi_full_init(struct vi_info *vi)
5444 {
5445         struct adapter *sc = vi->pi->adapter;
5446         struct ifnet *ifp = vi->ifp;
5447         uint16_t *rss;
5448         struct sge_rxq *rxq;
5449         int rc, i, j;
5450 #ifdef RSS
5451         int nbuckets = rss_getnumbuckets();
5452         int hashconfig = rss_gethashconfig();
5453         int extra;
5454 #endif
5455
5456         ASSERT_SYNCHRONIZED_OP(sc);
5457         KASSERT((vi->flags & VI_INIT_DONE) == 0,
5458             ("%s: VI_INIT_DONE already", __func__));
5459
5460         sysctl_ctx_init(&vi->ctx);
5461         vi->flags |= VI_SYSCTL_CTX;
5462
5463         /*
5464          * Allocate tx/rx/fl queues for this VI.
5465          */
5466         rc = t4_setup_vi_queues(vi);
5467         if (rc != 0)
5468                 goto done;      /* error message displayed already */
5469
5470         /*
5471          * Setup RSS for this VI.  Save a copy of the RSS table for later use.
5472          */
5473         if (vi->nrxq > vi->rss_size) {
5474                 if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
5475                     "some queues will never receive traffic.\n", vi->nrxq,
5476                     vi->rss_size);
5477         } else if (vi->rss_size % vi->nrxq) {
5478                 if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
5479                     "expect uneven traffic distribution.\n", vi->nrxq,
5480                     vi->rss_size);
5481         }
5482 #ifdef RSS
5483         if (vi->nrxq != nbuckets) {
5484                 if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
5485                     "performance will be impacted.\n", vi->nrxq, nbuckets);
5486         }
5487 #endif
5488         rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
5489         for (i = 0; i < vi->rss_size;) {
5490 #ifdef RSS
5491                 j = rss_get_indirection_to_bucket(i);
5492                 j %= vi->nrxq;
5493                 rxq = &sc->sge.rxq[vi->first_rxq + j];
5494                 rss[i++] = rxq->iq.abs_id;
5495 #else
5496                 for_each_rxq(vi, j, rxq) {
5497                         rss[i++] = rxq->iq.abs_id;
5498                         if (i == vi->rss_size)
5499                                 break;
5500                 }
5501 #endif
5502         }
5503
5504         rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
5505             vi->rss_size);
5506         if (rc != 0) {
5507                 free(rss, M_CXGBE);
5508                 if_printf(ifp, "rss_config failed: %d\n", rc);
5509                 goto done;
5510         }
5511
5512 #ifdef RSS
5513         vi->hashen = hashconfig_to_hashen(hashconfig);
5514
5515         /*
5516          * We may have had to enable some hashes even though the global config
5517          * wants them disabled.  This is a potential problem that must be
5518          * reported to the user.
5519          */
5520         extra = hashen_to_hashconfig(vi->hashen) ^ hashconfig;
5521
5522         /*
5523          * If we consider only the supported hash types, then the enabled hashes
5524          * are a superset of the requested hashes.  In other words, there cannot
5525          * be any supported hash that was requested but not enabled, but there
5526          * can be hashes that were not requested but had to be enabled.
5527          */
5528         extra &= SUPPORTED_RSS_HASHTYPES;
5529         MPASS((extra & hashconfig) == 0);
5530
5531         if (extra) {
5532                 if_printf(ifp,
5533                     "global RSS config (0x%x) cannot be accommodated.\n",
5534                     hashconfig);
5535         }
5536         if (extra & RSS_HASHTYPE_RSS_IPV4)
5537                 if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
5538         if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
5539                 if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
5540         if (extra & RSS_HASHTYPE_RSS_IPV6)
5541                 if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
5542         if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
5543                 if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
5544         if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
5545                 if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
5546         if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
5547                 if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
5548 #else
5549         vi->hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
5550             F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
5551             F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5552             F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
5553 #endif
5554         rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, vi->hashen, rss[0], 0, 0);
5555         if (rc != 0) {
5556                 free(rss, M_CXGBE);
5557                 if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
5558                 goto done;
5559         }
5560
5561         vi->rss = rss;
5562         vi->flags |= VI_INIT_DONE;
5563 done:
5564         if (rc != 0)
5565                 vi_full_uninit(vi);
5566
5567         return (rc);
5568 }
5569
5570 /*
5571  * Idempotent.
5572  */
5573 int
5574 vi_full_uninit(struct vi_info *vi)
5575 {
5576         struct port_info *pi = vi->pi;
5577         struct adapter *sc = pi->adapter;
5578         int i;
5579         struct sge_rxq *rxq;
5580         struct sge_txq *txq;
5581 #ifdef TCP_OFFLOAD
5582         struct sge_ofld_rxq *ofld_rxq;
5583 #endif
5584 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
5585         struct sge_wrq *ofld_txq;
5586 #endif
5587
5588         if (vi->flags & VI_INIT_DONE) {
5589
5590                 /* Need to quiesce queues.  */
5591
5592                 /* XXX: Only for the first VI? */
5593                 if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
5594                         quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
5595
5596                 for_each_txq(vi, i, txq) {
5597                         quiesce_txq(sc, txq);
5598                 }
5599
5600 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
5601                 for_each_ofld_txq(vi, i, ofld_txq) {
5602                         quiesce_wrq(sc, ofld_txq);
5603                 }
5604 #endif
5605
5606                 for_each_rxq(vi, i, rxq) {
5607                         quiesce_iq(sc, &rxq->iq);
5608                         quiesce_fl(sc, &rxq->fl);
5609                 }
5610
5611 #ifdef TCP_OFFLOAD
5612                 for_each_ofld_rxq(vi, i, ofld_rxq) {
5613                         quiesce_iq(sc, &ofld_rxq->iq);
5614                         quiesce_fl(sc, &ofld_rxq->fl);
5615                 }
5616 #endif
5617                 free(vi->rss, M_CXGBE);
5618                 free(vi->nm_rss, M_CXGBE);
5619         }
5620
5621         t4_teardown_vi_queues(vi);
5622         vi->flags &= ~VI_INIT_DONE;
5623
5624         return (0);
5625 }
5626
5627 static void
5628 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
5629 {
5630         struct sge_eq *eq = &txq->eq;
5631         struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
5632
5633         (void) sc;      /* unused */
5634
5635 #ifdef INVARIANTS
5636         TXQ_LOCK(txq);
5637         MPASS((eq->flags & EQ_ENABLED) == 0);
5638         TXQ_UNLOCK(txq);
5639 #endif
5640
5641         /* Wait for the mp_ring to empty. */
5642         while (!mp_ring_is_idle(txq->r)) {
5643                 mp_ring_check_drainage(txq->r, 0);
5644                 pause("rquiesce", 1);
5645         }
5646
5647         /* Then wait for the hardware to finish. */
5648         while (spg->cidx != htobe16(eq->pidx))
5649                 pause("equiesce", 1);
5650
5651         /* Finally, wait for the driver to reclaim all descriptors. */
5652         while (eq->cidx != eq->pidx)
5653                 pause("dquiesce", 1);
5654 }
5655
5656 static void
5657 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
5658 {
5659
5660         /* XXXTX */
5661 }
5662
5663 static void
5664 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
5665 {
5666         (void) sc;      /* unused */
5667
5668         /* Synchronize with the interrupt handler */
5669         while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
5670                 pause("iqfree", 1);
5671 }
5672
5673 static void
5674 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
5675 {
5676         mtx_lock(&sc->sfl_lock);
5677         FL_LOCK(fl);
5678         fl->flags |= FL_DOOMED;
5679         FL_UNLOCK(fl);
5680         callout_stop(&sc->sfl_callout);
5681         mtx_unlock(&sc->sfl_lock);
5682
5683         KASSERT((fl->flags & FL_STARVING) == 0,
5684             ("%s: still starving", __func__));
5685 }
5686
5687 static int
5688 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
5689     driver_intr_t *handler, void *arg, char *name)
5690 {
5691         int rc;
5692
5693         irq->rid = rid;
5694         irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
5695             RF_SHAREABLE | RF_ACTIVE);
5696         if (irq->res == NULL) {
5697                 device_printf(sc->dev,
5698                     "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
5699                 return (ENOMEM);
5700         }
5701
5702         rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
5703             NULL, handler, arg, &irq->tag);
5704         if (rc != 0) {
5705                 device_printf(sc->dev,
5706                     "failed to setup interrupt for rid %d, name %s: %d\n",
5707                     rid, name, rc);
5708         } else if (name)
5709                 bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
5710
5711         return (rc);
5712 }
5713
5714 static int
5715 t4_free_irq(struct adapter *sc, struct irq *irq)
5716 {
5717         if (irq->tag)
5718                 bus_teardown_intr(sc->dev, irq->res, irq->tag);
5719         if (irq->res)
5720                 bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
5721
5722         bzero(irq, sizeof(*irq));
5723
5724         return (0);
5725 }
5726
5727 static void
5728 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
5729 {
5730
5731         regs->version = chip_id(sc) | chip_rev(sc) << 10;
5732         t4_get_regs(sc, buf, regs->len);
5733 }
5734
5735 #define A_PL_INDIR_CMD  0x1f8
5736
5737 #define S_PL_AUTOINC    31
5738 #define M_PL_AUTOINC    0x1U
5739 #define V_PL_AUTOINC(x) ((x) << S_PL_AUTOINC)
5740 #define G_PL_AUTOINC(x) (((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
5741
5742 #define S_PL_VFID       20
5743 #define M_PL_VFID       0xffU
5744 #define V_PL_VFID(x)    ((x) << S_PL_VFID)
5745 #define G_PL_VFID(x)    (((x) >> S_PL_VFID) & M_PL_VFID)
5746
5747 #define S_PL_ADDR       0
5748 #define M_PL_ADDR       0xfffffU
5749 #define V_PL_ADDR(x)    ((x) << S_PL_ADDR)
5750 #define G_PL_ADDR(x)    (((x) >> S_PL_ADDR) & M_PL_ADDR)
5751
5752 #define A_PL_INDIR_DATA 0x1fc
5753
5754 static uint64_t
5755 read_vf_stat(struct adapter *sc, u_int vin, int reg)
5756 {
5757         u32 stats[2];
5758
5759         mtx_assert(&sc->reg_lock, MA_OWNED);
5760         if (sc->flags & IS_VF) {
5761                 stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
5762                 stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
5763         } else {
5764                 t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5765                     V_PL_VFID(vin) | V_PL_ADDR(VF_MPS_REG(reg)));
5766                 stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
5767                 stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
5768         }
5769         return (((uint64_t)stats[1]) << 32 | stats[0]);
5770 }
5771
5772 static void
5773 t4_get_vi_stats(struct adapter *sc, u_int vin, struct fw_vi_stats_vf *stats)
5774 {
5775
5776 #define GET_STAT(name) \
5777         read_vf_stat(sc, vin, A_MPS_VF_STAT_##name##_L)
5778
5779         stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
5780         stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
5781         stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
5782         stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
5783         stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
5784         stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
5785         stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
5786         stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
5787         stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
5788         stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
5789         stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
5790         stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
5791         stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
5792         stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
5793         stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
5794         stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
5795
5796 #undef GET_STAT
5797 }
5798
5799 static void
5800 t4_clr_vi_stats(struct adapter *sc, u_int vin)
5801 {
5802         int reg;
5803
5804         t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) | V_PL_VFID(vin) |
5805             V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
5806         for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
5807              reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
5808                 t4_write_reg(sc, A_PL_INDIR_DATA, 0);
5809 }
5810
5811 static void
5812 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
5813 {
5814         struct timeval tv;
5815         const struct timeval interval = {0, 250000};    /* 250ms */
5816
5817         if (!(vi->flags & VI_INIT_DONE))
5818                 return;
5819
5820         getmicrotime(&tv);
5821         timevalsub(&tv, &interval);
5822         if (timevalcmp(&tv, &vi->last_refreshed, <))
5823                 return;
5824
5825         mtx_lock(&sc->reg_lock);
5826         t4_get_vi_stats(sc, vi->vin, &vi->stats);
5827         getmicrotime(&vi->last_refreshed);
5828         mtx_unlock(&sc->reg_lock);
5829 }
5830
5831 static void
5832 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
5833 {
5834         u_int i, v, tnl_cong_drops, bg_map;
5835         struct timeval tv;
5836         const struct timeval interval = {0, 250000};    /* 250ms */
5837
5838         getmicrotime(&tv);
5839         timevalsub(&tv, &interval);
5840         if (timevalcmp(&tv, &pi->last_refreshed, <))
5841                 return;
5842
5843         tnl_cong_drops = 0;
5844         t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
5845         bg_map = pi->mps_bg_map;
5846         while (bg_map) {
5847                 i = ffs(bg_map) - 1;
5848                 mtx_lock(&sc->reg_lock);
5849                 t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v, 1,
5850                     A_TP_MIB_TNL_CNG_DROP_0 + i);
5851                 mtx_unlock(&sc->reg_lock);
5852                 tnl_cong_drops += v;
5853                 bg_map &= ~(1 << i);
5854         }
5855         pi->tnl_cong_drops = tnl_cong_drops;
5856         getmicrotime(&pi->last_refreshed);
5857 }
5858
5859 static void
5860 cxgbe_tick(void *arg)
5861 {
5862         struct port_info *pi = arg;
5863         struct adapter *sc = pi->adapter;
5864
5865         PORT_LOCK_ASSERT_OWNED(pi);
5866         cxgbe_refresh_stats(sc, pi);
5867
5868         callout_schedule(&pi->tick, hz);
5869 }
5870
5871 void
5872 vi_tick(void *arg)
5873 {
5874         struct vi_info *vi = arg;
5875         struct adapter *sc = vi->pi->adapter;
5876
5877         vi_refresh_stats(sc, vi);
5878
5879         callout_schedule(&vi->tick, hz);
5880 }
5881
5882 /*
5883  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
5884  */
5885 static char *caps_decoder[] = {
5886         "\20\001IPMI\002NCSI",                          /* 0: NBM */
5887         "\20\001PPP\002QFC\003DCBX",                    /* 1: link */
5888         "\20\001INGRESS\002EGRESS",                     /* 2: switch */
5889         "\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"      /* 3: NIC */
5890             "\006HASHFILTER\007ETHOFLD",
5891         "\20\001TOE",                                   /* 4: TOE */
5892         "\20\001RDDP\002RDMAC",                         /* 5: RDMA */
5893         "\20\001INITIATOR_PDU\002TARGET_PDU"            /* 6: iSCSI */
5894             "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
5895             "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
5896             "\007T10DIF"
5897             "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
5898         "\20\001LOOKASIDE\002TLSKEYS",                  /* 7: Crypto */
5899         "\20\001INITIATOR\002TARGET\003CTRL_OFLD"       /* 8: FCoE */
5900                     "\004PO_INITIATOR\005PO_TARGET",
5901 };
5902
5903 void
5904 t4_sysctls(struct adapter *sc)
5905 {
5906         struct sysctl_ctx_list *ctx;
5907         struct sysctl_oid *oid;
5908         struct sysctl_oid_list *children, *c0;
5909         static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
5910
5911         ctx = device_get_sysctl_ctx(sc->dev);
5912
5913         /*
5914          * dev.t4nex.X.
5915          */
5916         oid = device_get_sysctl_tree(sc->dev);
5917         c0 = children = SYSCTL_CHILDREN(oid);
5918
5919         sc->sc_do_rxcopy = 1;
5920         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
5921             &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
5922
5923         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
5924             sc->params.nports, "# of ports");
5925
5926         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
5927             CTLTYPE_STRING | CTLFLAG_RD, doorbells, (uintptr_t)&sc->doorbells,
5928             sysctl_bitfield_8b, "A", "available doorbells");
5929
5930         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
5931             sc->params.vpd.cclk, "core clock frequency (in KHz)");
5932
5933         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
5934             CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
5935             sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
5936             "interrupt holdoff timer values (us)");
5937
5938         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
5939             CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
5940             sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
5941             "interrupt holdoff packet counter values");
5942
5943         t4_sge_sysctls(sc, ctx, children);
5944
5945         sc->lro_timeout = 100;
5946         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
5947             &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
5948
5949         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
5950             &sc->debug_flags, 0, "flags to enable runtime debugging");
5951
5952         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
5953             CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
5954
5955         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
5956             CTLFLAG_RD, sc->fw_version, 0, "firmware version");
5957
5958         if (sc->flags & IS_VF)
5959                 return;
5960
5961         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
5962             NULL, chip_rev(sc), "chip hardware revision");
5963
5964         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
5965             CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
5966
5967         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
5968             CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
5969
5970         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
5971             CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
5972
5973         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "md_version",
5974             CTLFLAG_RD, sc->params.vpd.md, 0, "manufacturing diags version");
5975
5976         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
5977             CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
5978
5979         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
5980             sc->er_version, 0, "expansion ROM version");
5981
5982         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
5983             sc->bs_version, 0, "bootstrap firmware version");
5984
5985         SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
5986             NULL, sc->params.scfg_vers, "serial config version");
5987
5988         SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
5989             NULL, sc->params.vpd_vers, "VPD version");
5990
5991         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
5992             CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
5993
5994         SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
5995             sc->cfcsum, "config file checksum");
5996
5997 #define SYSCTL_CAP(name, n, text) \
5998         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
5999             CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], (uintptr_t)&sc->name, \
6000             sysctl_bitfield_16b, "A", "available " text " capabilities")
6001
6002         SYSCTL_CAP(nbmcaps, 0, "NBM");
6003         SYSCTL_CAP(linkcaps, 1, "link");
6004         SYSCTL_CAP(switchcaps, 2, "switch");
6005         SYSCTL_CAP(niccaps, 3, "NIC");
6006         SYSCTL_CAP(toecaps, 4, "TCP offload");
6007         SYSCTL_CAP(rdmacaps, 5, "RDMA");
6008         SYSCTL_CAP(iscsicaps, 6, "iSCSI");
6009         SYSCTL_CAP(cryptocaps, 7, "crypto");
6010         SYSCTL_CAP(fcoecaps, 8, "FCoE");
6011 #undef SYSCTL_CAP
6012
6013         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
6014             NULL, sc->tids.nftids, "number of filters");
6015
6016         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
6017             CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
6018             "chip temperature (in Celsius)");
6019
6020         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "loadavg", CTLTYPE_STRING |
6021             CTLFLAG_RD, sc, 0, sysctl_loadavg, "A",
6022             "microprocessor load averages (debug firmwares only)");
6023
6024         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_vdd", CTLFLAG_RD,
6025             &sc->params.core_vdd, 0, "core Vdd (in mV)");
6026
6027         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "local_cpus",
6028             CTLTYPE_STRING | CTLFLAG_RD, sc, LOCAL_CPUS,
6029             sysctl_cpus, "A", "local CPUs");
6030
6031         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "intr_cpus",
6032             CTLTYPE_STRING | CTLFLAG_RD, sc, INTR_CPUS,
6033             sysctl_cpus, "A", "preferred CPUs for interrupts");
6034
6035         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "swintr", CTLFLAG_RW,
6036             &sc->swintr, 0, "software triggered interrupts");
6037
6038         /*
6039          * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
6040          */
6041         oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
6042             CTLFLAG_RD | CTLFLAG_SKIP, NULL,
6043             "logs and miscellaneous information");
6044         children = SYSCTL_CHILDREN(oid);
6045
6046         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
6047             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6048             sysctl_cctrl, "A", "congestion control");
6049
6050         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
6051             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6052             sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
6053
6054         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
6055             CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
6056             sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
6057
6058         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
6059             CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
6060             sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
6061
6062         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
6063             CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
6064             sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
6065
6066         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
6067             CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
6068             sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
6069
6070         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
6071             CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
6072             sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
6073
6074         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
6075             CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_cim_la,
6076             "A", "CIM logic analyzer");
6077
6078         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
6079             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6080             sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
6081
6082         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
6083             CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
6084             sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
6085
6086         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
6087             CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
6088             sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
6089
6090         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
6091             CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
6092             sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
6093
6094         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
6095             CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
6096             sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
6097
6098         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
6099             CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
6100             sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
6101
6102         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
6103             CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
6104             sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
6105
6106         if (chip_id(sc) > CHELSIO_T4) {
6107                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
6108                     CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
6109                     sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
6110
6111                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
6112                     CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
6113                     sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
6114         }
6115
6116         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
6117             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6118             sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
6119
6120         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
6121             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6122             sysctl_cim_qcfg, "A", "CIM queue configuration");
6123
6124         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
6125             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6126             sysctl_cpl_stats, "A", "CPL statistics");
6127
6128         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
6129             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6130             sysctl_ddp_stats, "A", "non-TCP DDP statistics");
6131
6132         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
6133             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6134             sysctl_devlog, "A", "firmware's device log");
6135
6136         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
6137             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6138             sysctl_fcoe_stats, "A", "FCoE statistics");
6139
6140         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
6141             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6142             sysctl_hw_sched, "A", "hardware scheduler ");
6143
6144         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
6145             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6146             sysctl_l2t, "A", "hardware L2 table");
6147
6148         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "smt",
6149             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6150             sysctl_smt, "A", "hardware source MAC table");
6151
6152 #ifdef INET6
6153         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "clip",
6154             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6155             sysctl_clip, "A", "active CLIP table entries");
6156 #endif
6157
6158         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
6159             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6160             sysctl_lb_stats, "A", "loopback statistics");
6161
6162         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
6163             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6164             sysctl_meminfo, "A", "memory regions");
6165
6166         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
6167             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6168             chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
6169             "A", "MPS TCAM entries");
6170
6171         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
6172             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6173             sysctl_path_mtus, "A", "path MTUs");
6174
6175         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
6176             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6177             sysctl_pm_stats, "A", "PM statistics");
6178
6179         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
6180             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6181             sysctl_rdma_stats, "A", "RDMA statistics");
6182
6183         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
6184             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6185             sysctl_tcp_stats, "A", "TCP statistics");
6186
6187         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
6188             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6189             sysctl_tids, "A", "TID information");
6190
6191         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
6192             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6193             sysctl_tp_err_stats, "A", "TP error statistics");
6194
6195         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
6196             CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
6197             "TP logic analyzer event capture mask");
6198
6199         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
6200             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6201             sysctl_tp_la, "A", "TP logic analyzer");
6202
6203         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
6204             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6205             sysctl_tx_rate, "A", "Tx rate");
6206
6207         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
6208             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6209             sysctl_ulprx_la, "A", "ULPRX logic analyzer");
6210
6211         if (chip_id(sc) >= CHELSIO_T5) {
6212                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
6213                     CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6214                     sysctl_wcwr_stats, "A", "write combined work requests");
6215         }
6216
6217 #ifdef TCP_OFFLOAD
6218         if (is_offload(sc)) {
6219                 int i;
6220                 char s[4];
6221
6222                 /*
6223                  * dev.t4nex.X.toe.
6224                  */
6225                 oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
6226                     NULL, "TOE parameters");
6227                 children = SYSCTL_CHILDREN(oid);
6228
6229                 sc->tt.cong_algorithm = -1;
6230                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "cong_algorithm",
6231                     CTLFLAG_RW, &sc->tt.cong_algorithm, 0, "congestion control "
6232                     "(-1 = default, 0 = reno, 1 = tahoe, 2 = newreno, "
6233                     "3 = highspeed)");
6234
6235                 sc->tt.sndbuf = 256 * 1024;
6236                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
6237                     &sc->tt.sndbuf, 0, "max hardware send buffer size");
6238
6239                 sc->tt.ddp = 0;
6240                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
6241                     &sc->tt.ddp, 0, "DDP allowed");
6242
6243                 sc->tt.rx_coalesce = 1;
6244                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
6245                     CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
6246
6247                 sc->tt.tls = 0;
6248                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tls", CTLFLAG_RW,
6249                     &sc->tt.tls, 0, "Inline TLS allowed");
6250
6251                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tls_rx_ports",
6252                     CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tls_rx_ports,
6253                     "I", "TCP ports that use inline TLS+TOE RX");
6254
6255                 sc->tt.tx_align = 1;
6256                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
6257                     CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
6258
6259                 sc->tt.tx_zcopy = 0;
6260                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
6261                     CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
6262                     "Enable zero-copy aio_write(2)");
6263
6264                 sc->tt.cop_managed_offloading = !!t4_cop_managed_offloading;
6265                 SYSCTL_ADD_INT(ctx, children, OID_AUTO,
6266                     "cop_managed_offloading", CTLFLAG_RW,
6267                     &sc->tt.cop_managed_offloading, 0,
6268                     "COP (Connection Offload Policy) controls all TOE offload");
6269
6270                 sc->tt.autorcvbuf_inc = 16 * 1024;
6271                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "autorcvbuf_inc",
6272                     CTLFLAG_RW, &sc->tt.autorcvbuf_inc, 0,
6273                     "autorcvbuf increment");
6274
6275                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
6276                     CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
6277                     "TP timer tick (us)");
6278
6279                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
6280                     CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
6281                     "TCP timestamp tick (us)");
6282
6283                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
6284                     CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
6285                     "DACK tick (us)");
6286
6287                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
6288                     CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
6289                     "IU", "DACK timer (us)");
6290
6291                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
6292                     CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
6293                     sysctl_tp_timer, "LU", "Minimum retransmit interval (us)");
6294
6295                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
6296                     CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
6297                     sysctl_tp_timer, "LU", "Maximum retransmit interval (us)");
6298
6299                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
6300                     CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
6301                     sysctl_tp_timer, "LU", "Persist timer min (us)");
6302
6303                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
6304                     CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
6305                     sysctl_tp_timer, "LU", "Persist timer max (us)");
6306
6307                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
6308                     CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
6309                     sysctl_tp_timer, "LU", "Keepalive idle timer (us)");
6310
6311                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_interval",
6312                     CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
6313                     sysctl_tp_timer, "LU", "Keepalive interval timer (us)");
6314
6315                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
6316                     CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
6317                     sysctl_tp_timer, "LU", "Initial SRTT (us)");
6318
6319                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
6320                     CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
6321                     sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
6322
6323                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "syn_rexmt_count",
6324                     CTLTYPE_UINT | CTLFLAG_RD, sc, S_SYNSHIFTMAX,
6325                     sysctl_tp_shift_cnt, "IU",
6326                     "Number of SYN retransmissions before abort");
6327
6328                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_count",
6329                     CTLTYPE_UINT | CTLFLAG_RD, sc, S_RXTSHIFTMAXR2,
6330                     sysctl_tp_shift_cnt, "IU",
6331                     "Number of retransmissions before abort");
6332
6333                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_count",
6334                     CTLTYPE_UINT | CTLFLAG_RD, sc, S_KEEPALIVEMAXR2,
6335                     sysctl_tp_shift_cnt, "IU",
6336                     "Number of keepalive probes before abort");
6337
6338                 oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "rexmt_backoff",
6339                     CTLFLAG_RD, NULL, "TOE retransmit backoffs");
6340                 children = SYSCTL_CHILDREN(oid);
6341                 for (i = 0; i < 16; i++) {
6342                         snprintf(s, sizeof(s), "%u", i);
6343                         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, s,
6344                             CTLTYPE_UINT | CTLFLAG_RD, sc, i, sysctl_tp_backoff,
6345                             "IU", "TOE retransmit backoff");
6346                 }
6347         }
6348 #endif
6349 }
6350
6351 void
6352 vi_sysctls(struct vi_info *vi)
6353 {
6354         struct sysctl_ctx_list *ctx;
6355         struct sysctl_oid *oid;
6356         struct sysctl_oid_list *children;
6357
6358         ctx = device_get_sysctl_ctx(vi->dev);
6359
6360         /*
6361          * dev.v?(cxgbe|cxl).X.
6362          */
6363         oid = device_get_sysctl_tree(vi->dev);
6364         children = SYSCTL_CHILDREN(oid);
6365
6366         SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
6367             vi->viid, "VI identifer");
6368         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
6369             &vi->nrxq, 0, "# of rx queues");
6370         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
6371             &vi->ntxq, 0, "# of tx queues");
6372         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
6373             &vi->first_rxq, 0, "index of first rx queue");
6374         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
6375             &vi->first_txq, 0, "index of first tx queue");
6376         SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_base", CTLFLAG_RD, NULL,
6377             vi->rss_base, "start of RSS indirection table");
6378         SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
6379             vi->rss_size, "size of RSS indirection table");
6380
6381         if (IS_MAIN_VI(vi)) {
6382                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
6383                     CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
6384                     "Reserve queue 0 for non-flowid packets");
6385         }
6386
6387 #ifdef TCP_OFFLOAD
6388         if (vi->nofldrxq != 0) {
6389                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
6390                     &vi->nofldrxq, 0,
6391                     "# of rx queues for offloaded TCP connections");
6392                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
6393                     CTLFLAG_RD, &vi->first_ofld_rxq, 0,
6394                     "index of first TOE rx queue");
6395                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx_ofld",
6396                     CTLTYPE_INT | CTLFLAG_RW, vi, 0,
6397                     sysctl_holdoff_tmr_idx_ofld, "I",
6398                     "holdoff timer index for TOE queues");
6399                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx_ofld",
6400                     CTLTYPE_INT | CTLFLAG_RW, vi, 0,
6401                     sysctl_holdoff_pktc_idx_ofld, "I",
6402                     "holdoff packet counter index for TOE queues");
6403         }
6404 #endif
6405 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
6406         if (vi->nofldtxq != 0) {
6407                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
6408                     &vi->nofldtxq, 0,
6409                     "# of tx queues for TOE/ETHOFLD");
6410                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
6411                     CTLFLAG_RD, &vi->first_ofld_txq, 0,
6412                     "index of first TOE/ETHOFLD tx queue");
6413         }
6414 #endif
6415 #ifdef DEV_NETMAP
6416         if (vi->nnmrxq != 0) {
6417                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
6418                     &vi->nnmrxq, 0, "# of netmap rx queues");
6419                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
6420                     &vi->nnmtxq, 0, "# of netmap tx queues");
6421                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
6422                     CTLFLAG_RD, &vi->first_nm_rxq, 0,
6423                     "index of first netmap rx queue");
6424                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
6425                     CTLFLAG_RD, &vi->first_nm_txq, 0,
6426                     "index of first netmap tx queue");
6427         }
6428 #endif
6429
6430         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
6431             CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
6432             "holdoff timer index");
6433         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
6434             CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
6435             "holdoff packet counter index");
6436
6437         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
6438             CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
6439             "rx queue size");
6440         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
6441             CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
6442             "tx queue size");
6443 }
6444
6445 static void
6446 cxgbe_sysctls(struct port_info *pi)
6447 {
6448         struct sysctl_ctx_list *ctx;
6449         struct sysctl_oid *oid;
6450         struct sysctl_oid_list *children, *children2;
6451         struct adapter *sc = pi->adapter;
6452         int i;
6453         char name[16];
6454         static char *tc_flags = {"\20\1USER\2SYNC\3ASYNC\4ERR"};
6455
6456         ctx = device_get_sysctl_ctx(pi->dev);
6457
6458         /*
6459          * dev.cxgbe.X.
6460          */
6461         oid = device_get_sysctl_tree(pi->dev);
6462         children = SYSCTL_CHILDREN(oid);
6463
6464         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
6465            CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
6466         if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
6467                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
6468                     CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
6469                     "PHY temperature (in Celsius)");
6470                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
6471                     CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
6472                     "PHY firmware version");
6473         }
6474
6475         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
6476             CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
6477     "PAUSE settings (bit 0 = rx_pause, 1 = tx_pause, 2 = pause_autoneg)");
6478         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
6479             CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
6480             "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
6481         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
6482             CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
6483             "autonegotiation (-1 = not supported)");
6484
6485         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
6486             port_top_speed(pi), "max speed (in Gbps)");
6487         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "mps_bg_map", CTLFLAG_RD, NULL,
6488             pi->mps_bg_map, "MPS buffer group map");
6489         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_e_chan_map", CTLFLAG_RD,
6490             NULL, pi->rx_e_chan_map, "TP rx e-channel map");
6491
6492         if (sc->flags & IS_VF)
6493                 return;
6494
6495         /*
6496          * dev.(cxgbe|cxl).X.tc.
6497          */
6498         oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
6499             "Tx scheduler traffic classes (cl_rl)");
6500         children2 = SYSCTL_CHILDREN(oid);
6501         SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "pktsize",
6502             CTLFLAG_RW, &pi->sched_params->pktsize, 0,
6503             "pktsize for per-flow cl-rl (0 means up to the driver )");
6504         SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "burstsize",
6505             CTLFLAG_RW, &pi->sched_params->burstsize, 0,
6506             "burstsize for per-flow cl-rl (0 means up to the driver)");
6507         for (i = 0; i < sc->chip_params->nsched_cls; i++) {
6508                 struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
6509
6510                 snprintf(name, sizeof(name), "%d", i);
6511                 children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
6512                     SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
6513                     "traffic class"));
6514                 SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "flags",
6515                     CTLTYPE_STRING | CTLFLAG_RD, tc_flags, (uintptr_t)&tc->flags,
6516                     sysctl_bitfield_8b, "A", "flags");
6517                 SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
6518                     CTLFLAG_RD, &tc->refcount, 0, "references to this class");
6519                 SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
6520                     CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
6521                     sysctl_tc_params, "A", "traffic class parameters");
6522         }
6523
6524         /*
6525          * dev.cxgbe.X.stats.
6526          */
6527         oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
6528             NULL, "port statistics");
6529         children = SYSCTL_CHILDREN(oid);
6530         SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
6531             &pi->tx_parse_error, 0,
6532             "# of tx packets with invalid length or # of segments");
6533
6534 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
6535         SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
6536             CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
6537             sysctl_handle_t4_reg64, "QU", desc)
6538
6539         SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
6540             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
6541         SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
6542             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
6543         SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
6544             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
6545         SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
6546             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
6547         SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
6548             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
6549         SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
6550             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
6551         SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
6552             "# of tx frames in this range",
6553             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
6554         SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
6555             "# of tx frames in this range",
6556             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
6557         SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
6558             "# of tx frames in this range",
6559             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
6560         SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
6561             "# of tx frames in this range",
6562             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
6563         SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
6564             "# of tx frames in this range",
6565             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
6566         SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
6567             "# of tx frames in this range",
6568             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
6569         SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
6570             "# of tx frames in this range",
6571             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
6572         SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
6573             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
6574         SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
6575             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
6576         SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
6577             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
6578         SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
6579             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
6580         SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
6581             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
6582         SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
6583             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
6584         SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
6585             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
6586         SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
6587             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
6588         SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
6589             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
6590         SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
6591             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
6592
6593         SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
6594             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
6595         SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
6596             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
6597         SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
6598             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
6599         SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
6600             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
6601         SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
6602             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
6603         SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
6604             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
6605         SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
6606             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
6607         SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
6608             "# of frames received with bad FCS",
6609             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
6610         SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
6611             "# of frames received with length error",
6612             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
6613         SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
6614             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
6615         SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
6616             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
6617         SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
6618             "# of rx frames in this range",
6619             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
6620         SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
6621             "# of rx frames in this range",
6622             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
6623         SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
6624             "# of rx frames in this range",
6625             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
6626         SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
6627             "# of rx frames in this range",
6628             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
6629         SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
6630             "# of rx frames in this range",
6631             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
6632         SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
6633             "# of rx frames in this range",
6634             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
6635         SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
6636             "# of rx frames in this range",
6637             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
6638         SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
6639             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
6640         SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
6641             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
6642         SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
6643             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
6644         SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
6645             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
6646         SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
6647             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
6648         SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
6649             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
6650         SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
6651             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
6652         SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
6653             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
6654         SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
6655             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
6656
6657 #undef SYSCTL_ADD_T4_REG64
6658
6659 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
6660         SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
6661             &pi->stats.name, desc)
6662
6663         /* We get these from port_stats and they may be stale by up to 1s */
6664         SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
6665             "# drops due to buffer-group 0 overflows");
6666         SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
6667             "# drops due to buffer-group 1 overflows");
6668         SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
6669             "# drops due to buffer-group 2 overflows");
6670         SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
6671             "# drops due to buffer-group 3 overflows");
6672         SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
6673             "# of buffer-group 0 truncated packets");
6674         SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
6675             "# of buffer-group 1 truncated packets");
6676         SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
6677             "# of buffer-group 2 truncated packets");
6678         SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
6679             "# of buffer-group 3 truncated packets");
6680
6681 #undef SYSCTL_ADD_T4_PORTSTAT
6682
6683         SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_records",
6684             CTLFLAG_RD, &pi->tx_tls_records,
6685             "# of TLS records transmitted");
6686         SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_octets",
6687             CTLFLAG_RD, &pi->tx_tls_octets,
6688             "# of payload octets in transmitted TLS records");
6689         SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_records",
6690             CTLFLAG_RD, &pi->rx_tls_records,
6691             "# of TLS records received");
6692         SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_octets",
6693             CTLFLAG_RD, &pi->rx_tls_octets,
6694             "# of payload octets in received TLS records");
6695 }
6696
6697 static int
6698 sysctl_int_array(SYSCTL_HANDLER_ARGS)
6699 {
6700         int rc, *i, space = 0;
6701         struct sbuf sb;
6702
6703         sbuf_new_for_sysctl(&sb, NULL, 64, req);
6704         for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
6705                 if (space)
6706                         sbuf_printf(&sb, " ");
6707                 sbuf_printf(&sb, "%d", *i);
6708                 space = 1;
6709         }
6710         rc = sbuf_finish(&sb);
6711         sbuf_delete(&sb);
6712         return (rc);
6713 }
6714
6715 static int
6716 sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS)
6717 {
6718         int rc;
6719         struct sbuf *sb;
6720
6721         rc = sysctl_wire_old_buffer(req, 0);
6722         if (rc != 0)
6723                 return(rc);
6724
6725         sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6726         if (sb == NULL)
6727                 return (ENOMEM);
6728
6729         sbuf_printf(sb, "%b", *(uint8_t *)(uintptr_t)arg2, (char *)arg1);
6730         rc = sbuf_finish(sb);
6731         sbuf_delete(sb);
6732
6733         return (rc);
6734 }
6735
6736 static int
6737 sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS)
6738 {
6739         int rc;
6740         struct sbuf *sb;
6741
6742         rc = sysctl_wire_old_buffer(req, 0);
6743         if (rc != 0)
6744                 return(rc);
6745
6746         sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6747         if (sb == NULL)
6748                 return (ENOMEM);
6749
6750         sbuf_printf(sb, "%b", *(uint16_t *)(uintptr_t)arg2, (char *)arg1);
6751         rc = sbuf_finish(sb);
6752         sbuf_delete(sb);
6753
6754         return (rc);
6755 }
6756
6757 static int
6758 sysctl_btphy(SYSCTL_HANDLER_ARGS)
6759 {
6760         struct port_info *pi = arg1;
6761         int op = arg2;
6762         struct adapter *sc = pi->adapter;
6763         u_int v;
6764         int rc;
6765
6766         rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
6767         if (rc)
6768                 return (rc);
6769         /* XXX: magic numbers */
6770         rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
6771             &v);
6772         end_synchronized_op(sc, 0);
6773         if (rc)
6774                 return (rc);
6775         if (op == 0)
6776                 v /= 256;
6777
6778         rc = sysctl_handle_int(oidp, &v, 0, req);
6779         return (rc);
6780 }
6781
6782 static int
6783 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
6784 {
6785         struct vi_info *vi = arg1;
6786         int rc, val;
6787
6788         val = vi->rsrv_noflowq;
6789         rc = sysctl_handle_int(oidp, &val, 0, req);
6790         if (rc != 0 || req->newptr == NULL)
6791                 return (rc);
6792
6793         if ((val >= 1) && (vi->ntxq > 1))
6794                 vi->rsrv_noflowq = 1;
6795         else
6796                 vi->rsrv_noflowq = 0;
6797
6798         return (rc);
6799 }
6800
6801 static int
6802 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
6803 {
6804         struct vi_info *vi = arg1;
6805         struct adapter *sc = vi->pi->adapter;
6806         int idx, rc, i;
6807         struct sge_rxq *rxq;
6808         uint8_t v;
6809
6810         idx = vi->tmr_idx;
6811
6812         rc = sysctl_handle_int(oidp, &idx, 0, req);
6813         if (rc != 0 || req->newptr == NULL)
6814                 return (rc);
6815
6816         if (idx < 0 || idx >= SGE_NTIMERS)
6817                 return (EINVAL);
6818
6819         rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6820             "t4tmr");
6821         if (rc)
6822                 return (rc);
6823
6824         v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
6825         for_each_rxq(vi, i, rxq) {
6826 #ifdef atomic_store_rel_8
6827                 atomic_store_rel_8(&rxq->iq.intr_params, v);
6828 #else
6829                 rxq->iq.intr_params = v;
6830 #endif
6831         }
6832         vi->tmr_idx = idx;
6833
6834         end_synchronized_op(sc, LOCK_HELD);
6835         return (0);
6836 }
6837
6838 static int
6839 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
6840 {
6841         struct vi_info *vi = arg1;
6842         struct adapter *sc = vi->pi->adapter;
6843         int idx, rc;
6844
6845         idx = vi->pktc_idx;
6846
6847         rc = sysctl_handle_int(oidp, &idx, 0, req);
6848         if (rc != 0 || req->newptr == NULL)
6849                 return (rc);
6850
6851         if (idx < -1 || idx >= SGE_NCOUNTERS)
6852                 return (EINVAL);
6853
6854         rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6855             "t4pktc");
6856         if (rc)
6857                 return (rc);
6858
6859         if (vi->flags & VI_INIT_DONE)
6860                 rc = EBUSY; /* cannot be changed once the queues are created */
6861         else
6862                 vi->pktc_idx = idx;
6863
6864         end_synchronized_op(sc, LOCK_HELD);
6865         return (rc);
6866 }
6867
6868 static int
6869 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
6870 {
6871         struct vi_info *vi = arg1;
6872         struct adapter *sc = vi->pi->adapter;
6873         int qsize, rc;
6874
6875         qsize = vi->qsize_rxq;
6876
6877         rc = sysctl_handle_int(oidp, &qsize, 0, req);
6878         if (rc != 0 || req->newptr == NULL)
6879                 return (rc);
6880
6881         if (qsize < 128 || (qsize & 7))
6882                 return (EINVAL);
6883
6884         rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6885             "t4rxqs");
6886         if (rc)
6887                 return (rc);
6888
6889         if (vi->flags & VI_INIT_DONE)
6890                 rc = EBUSY; /* cannot be changed once the queues are created */
6891         else
6892                 vi->qsize_rxq = qsize;
6893
6894         end_synchronized_op(sc, LOCK_HELD);
6895         return (rc);
6896 }
6897
6898 static int
6899 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
6900 {
6901         struct vi_info *vi = arg1;
6902         struct adapter *sc = vi->pi->adapter;
6903         int qsize, rc;
6904
6905         qsize = vi->qsize_txq;
6906
6907         rc = sysctl_handle_int(oidp, &qsize, 0, req);
6908         if (rc != 0 || req->newptr == NULL)
6909                 return (rc);
6910
6911         if (qsize < 128 || qsize > 65536)
6912                 return (EINVAL);
6913
6914         rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6915             "t4txqs");
6916         if (rc)
6917                 return (rc);
6918
6919         if (vi->flags & VI_INIT_DONE)
6920                 rc = EBUSY; /* cannot be changed once the queues are created */
6921         else
6922                 vi->qsize_txq = qsize;
6923
6924         end_synchronized_op(sc, LOCK_HELD);
6925         return (rc);
6926 }
6927
6928 static int
6929 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
6930 {
6931         struct port_info *pi = arg1;
6932         struct adapter *sc = pi->adapter;
6933         struct link_config *lc = &pi->link_cfg;
6934         int rc;
6935
6936         if (req->newptr == NULL) {
6937                 struct sbuf *sb;
6938                 static char *bits = "\20\1RX\2TX\3AUTO";
6939
6940                 rc = sysctl_wire_old_buffer(req, 0);
6941                 if (rc != 0)
6942                         return(rc);
6943
6944                 sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6945                 if (sb == NULL)
6946                         return (ENOMEM);
6947
6948                 if (lc->link_ok) {
6949                         sbuf_printf(sb, "%b", (lc->fc & (PAUSE_TX | PAUSE_RX)) |
6950                             (lc->requested_fc & PAUSE_AUTONEG), bits);
6951                 } else {
6952                         sbuf_printf(sb, "%b", lc->requested_fc & (PAUSE_TX |
6953                             PAUSE_RX | PAUSE_AUTONEG), bits);
6954                 }
6955                 rc = sbuf_finish(sb);
6956                 sbuf_delete(sb);
6957         } else {
6958                 char s[2];
6959                 int n;
6960
6961                 s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX |
6962                     PAUSE_AUTONEG));
6963                 s[1] = 0;
6964
6965                 rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6966                 if (rc != 0)
6967                         return(rc);
6968
6969                 if (s[1] != 0)
6970                         return (EINVAL);
6971                 if (s[0] < '0' || s[0] > '9')
6972                         return (EINVAL);        /* not a number */
6973                 n = s[0] - '0';
6974                 if (n & ~(PAUSE_TX | PAUSE_RX | PAUSE_AUTONEG))
6975                         return (EINVAL);        /* some other bit is set too */
6976
6977                 rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6978                     "t4PAUSE");
6979                 if (rc)
6980                         return (rc);
6981                 PORT_LOCK(pi);
6982                 lc->requested_fc = n;
6983                 fixup_link_config(pi);
6984                 if (pi->up_vis > 0)
6985                         rc = apply_link_config(pi);
6986                 set_current_media(pi);
6987                 PORT_UNLOCK(pi);
6988                 end_synchronized_op(sc, 0);
6989         }
6990
6991         return (rc);
6992 }
6993
6994 static int
6995 sysctl_fec(SYSCTL_HANDLER_ARGS)
6996 {
6997         struct port_info *pi = arg1;
6998         struct adapter *sc = pi->adapter;
6999         struct link_config *lc = &pi->link_cfg;
7000         int rc;
7001         int8_t old;
7002
7003         if (req->newptr == NULL) {
7004                 struct sbuf *sb;
7005                 static char *bits = "\20\1RS\2BASE-R\3RSVD1\4RSVD2\5RSVD3\6AUTO";
7006
7007                 rc = sysctl_wire_old_buffer(req, 0);
7008                 if (rc != 0)
7009                         return(rc);
7010
7011                 sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
7012                 if (sb == NULL)
7013                         return (ENOMEM);
7014
7015                 /*
7016                  * Display the requested_fec when the link is down -- the actual
7017                  * FEC makes sense only when the link is up.
7018                  */
7019                 if (lc->link_ok) {
7020                         sbuf_printf(sb, "%b", (lc->fec & M_FW_PORT_CAP32_FEC) |
7021                             (lc->requested_fec & FEC_AUTO), bits);
7022                 } else {
7023                         sbuf_printf(sb, "%b", lc->requested_fec, bits);
7024                 }
7025                 rc = sbuf_finish(sb);
7026                 sbuf_delete(sb);
7027         } else {
7028                 char s[3];
7029                 int n;
7030
7031                 snprintf(s, sizeof(s), "%d",
7032                     lc->requested_fec == FEC_AUTO ? -1 :
7033                     lc->requested_fec & M_FW_PORT_CAP32_FEC);
7034
7035                 rc = sysctl_handle_string(oidp, s, sizeof(s), req);
7036                 if (rc != 0)
7037                         return(rc);
7038
7039                 n = strtol(&s[0], NULL, 0);
7040                 if (n < 0 || n & FEC_AUTO)
7041                         n = FEC_AUTO;
7042                 else {
7043                         if (n & ~M_FW_PORT_CAP32_FEC)
7044                                 return (EINVAL);/* some other bit is set too */
7045                         if (!powerof2(n))
7046                                 return (EINVAL);/* one bit can be set at most */
7047                 }
7048
7049                 rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
7050                     "t4fec");
7051                 if (rc)
7052                         return (rc);
7053                 PORT_LOCK(pi);
7054                 old = lc->requested_fec;
7055                 if (n == FEC_AUTO)
7056                         lc->requested_fec = FEC_AUTO;
7057                 else if (n == 0)
7058                         lc->requested_fec = FEC_NONE;
7059                 else {
7060                         if ((lc->supported | V_FW_PORT_CAP32_FEC(n)) !=
7061                             lc->supported) {
7062                                 rc = ENOTSUP;
7063                                 goto done;
7064                         }
7065                         lc->requested_fec = n;
7066                 }
7067                 fixup_link_config(pi);
7068                 if (pi->up_vis > 0) {
7069                         rc = apply_link_config(pi);
7070                         if (rc != 0) {
7071                                 lc->requested_fec = old;
7072                                 if (rc == FW_EPROTO)
7073                                         rc = ENOTSUP;
7074                         }
7075                 }
7076 done:
7077                 PORT_UNLOCK(pi);
7078                 end_synchronized_op(sc, 0);
7079         }
7080
7081         return (rc);
7082 }
7083
7084 static int
7085 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
7086 {
7087         struct port_info *pi = arg1;
7088         struct adapter *sc = pi->adapter;
7089         struct link_config *lc = &pi->link_cfg;
7090         int rc, val;
7091
7092         if (lc->supported & FW_PORT_CAP32_ANEG)
7093                 val = lc->requested_aneg == AUTONEG_DISABLE ? 0 : 1;
7094         else
7095                 val = -1;
7096         rc = sysctl_handle_int(oidp, &val, 0, req);
7097         if (rc != 0 || req->newptr == NULL)
7098                 return (rc);
7099         if (val == 0)
7100                 val = AUTONEG_DISABLE;
7101         else if (val == 1)
7102                 val = AUTONEG_ENABLE;
7103         else
7104                 val = AUTONEG_AUTO;
7105
7106         rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
7107             "t4aneg");
7108         if (rc)
7109                 return (rc);
7110         PORT_LOCK(pi);
7111         if (val == AUTONEG_ENABLE && !(lc->supported & FW_PORT_CAP32_ANEG)) {
7112                 rc = ENOTSUP;
7113                 goto done;
7114         }
7115         lc->requested_aneg = val;
7116         fixup_link_config(pi);
7117         if (pi->up_vis > 0)
7118                 rc = apply_link_config(pi);
7119         set_current_media(pi);
7120 done:
7121         PORT_UNLOCK(pi);
7122         end_synchronized_op(sc, 0);
7123         return (rc);
7124 }
7125
7126 static int
7127 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
7128 {
7129         struct adapter *sc = arg1;
7130         int reg = arg2;
7131         uint64_t val;
7132
7133         val = t4_read_reg64(sc, reg);
7134
7135         return (sysctl_handle_64(oidp, &val, 0, req));
7136 }
7137
7138 static int
7139 sysctl_temperature(SYSCTL_HANDLER_ARGS)
7140 {
7141         struct adapter *sc = arg1;
7142         int rc, t;
7143         uint32_t param, val;
7144
7145         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
7146         if (rc)
7147                 return (rc);
7148         param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
7149             V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
7150             V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
7151         rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
7152         end_synchronized_op(sc, 0);
7153         if (rc)
7154                 return (rc);
7155
7156         /* unknown is returned as 0 but we display -1 in that case */
7157         t = val == 0 ? -1 : val;
7158
7159         rc = sysctl_handle_int(oidp, &t, 0, req);
7160         return (rc);
7161 }
7162
7163 static int
7164 sysctl_loadavg(SYSCTL_HANDLER_ARGS)
7165 {
7166         struct adapter *sc = arg1;
7167         struct sbuf *sb;
7168         int rc;
7169         uint32_t param, val;
7170
7171         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4lavg");
7172         if (rc)
7173                 return (rc);
7174         param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
7175             V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_LOAD);
7176         rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
7177         end_synchronized_op(sc, 0);
7178         if (rc)
7179                 return (rc);
7180
7181         rc = sysctl_wire_old_buffer(req, 0);
7182         if (rc != 0)
7183                 return (rc);
7184
7185         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7186         if (sb == NULL)
7187                 return (ENOMEM);
7188
7189         if (val == 0xffffffff) {
7190                 /* Only debug and custom firmwares report load averages. */
7191                 sbuf_printf(sb, "not available");
7192         } else {
7193                 sbuf_printf(sb, "%d %d %d", val & 0xff, (val >> 8) & 0xff,
7194                     (val >> 16) & 0xff);
7195         }
7196         rc = sbuf_finish(sb);
7197         sbuf_delete(sb);
7198
7199         return (rc);
7200 }
7201
7202 static int
7203 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
7204 {
7205         struct adapter *sc = arg1;
7206         struct sbuf *sb;
7207         int rc, i;
7208         uint16_t incr[NMTUS][NCCTRL_WIN];
7209         static const char *dec_fac[] = {
7210                 "0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
7211                 "0.9375"
7212         };
7213
7214         rc = sysctl_wire_old_buffer(req, 0);
7215         if (rc != 0)
7216                 return (rc);
7217
7218         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7219         if (sb == NULL)
7220                 return (ENOMEM);
7221
7222         t4_read_cong_tbl(sc, incr);
7223
7224         for (i = 0; i < NCCTRL_WIN; ++i) {
7225                 sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
7226                     incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
7227                     incr[5][i], incr[6][i], incr[7][i]);
7228                 sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
7229                     incr[8][i], incr[9][i], incr[10][i], incr[11][i],
7230                     incr[12][i], incr[13][i], incr[14][i], incr[15][i],
7231                     sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
7232         }
7233
7234         rc = sbuf_finish(sb);
7235         sbuf_delete(sb);
7236
7237         return (rc);
7238 }
7239
7240 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
7241         "TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",   /* ibq's */
7242         "ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI", /* obq's */
7243         "SGE0-RX", "SGE1-RX"    /* additional obq's (T5 onwards) */
7244 };
7245
7246 static int
7247 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
7248 {
7249         struct adapter *sc = arg1;
7250         struct sbuf *sb;
7251         int rc, i, n, qid = arg2;
7252         uint32_t *buf, *p;
7253         char *qtype;
7254         u_int cim_num_obq = sc->chip_params->cim_num_obq;
7255
7256         KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
7257             ("%s: bad qid %d\n", __func__, qid));
7258
7259         if (qid < CIM_NUM_IBQ) {
7260                 /* inbound queue */
7261                 qtype = "IBQ";
7262                 n = 4 * CIM_IBQ_SIZE;
7263                 buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
7264                 rc = t4_read_cim_ibq(sc, qid, buf, n);
7265         } else {
7266                 /* outbound queue */
7267                 qtype = "OBQ";
7268                 qid -= CIM_NUM_IBQ;
7269                 n = 4 * cim_num_obq * CIM_OBQ_SIZE;
7270                 buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
7271                 rc = t4_read_cim_obq(sc, qid, buf, n);
7272         }
7273
7274         if (rc < 0) {
7275                 rc = -rc;
7276                 goto done;
7277         }
7278         n = rc * sizeof(uint32_t);      /* rc has # of words actually read */
7279
7280         rc = sysctl_wire_old_buffer(req, 0);
7281         if (rc != 0)
7282                 goto done;
7283
7284         sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
7285         if (sb == NULL) {
7286                 rc = ENOMEM;
7287                 goto done;
7288         }
7289
7290         sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
7291         for (i = 0, p = buf; i < n; i += 16, p += 4)
7292                 sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
7293                     p[2], p[3]);
7294
7295         rc = sbuf_finish(sb);
7296         sbuf_delete(sb);
7297 done:
7298         free(buf, M_CXGBE);
7299         return (rc);
7300 }
7301
7302 static void
7303 sbuf_cim_la4(struct adapter *sc, struct sbuf *sb, uint32_t *buf, uint32_t cfg)
7304 {
7305         uint32_t *p;
7306
7307         sbuf_printf(sb, "Status   Data      PC%s",
7308             cfg & F_UPDBGLACAPTPCONLY ? "" :
7309             "     LS0Stat  LS0Addr             LS0Data");
7310
7311         for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
7312                 if (cfg & F_UPDBGLACAPTPCONLY) {
7313                         sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
7314                             p[6], p[7]);
7315                         sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
7316                             (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
7317                             p[4] & 0xff, p[5] >> 8);
7318                         sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
7319                             (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
7320                             p[1] & 0xf, p[2] >> 4);
7321                 } else {
7322                         sbuf_printf(sb,
7323                             "\n  %02x   %x%07x %x%07x %08x %08x "
7324                             "%08x%08x%08x%08x",
7325                             (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
7326                             p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
7327                             p[6], p[7]);
7328                 }
7329         }
7330 }
7331
7332 static void
7333 sbuf_cim_la6(struct adapter *sc, struct sbuf *sb, uint32_t *buf, uint32_t cfg)
7334 {
7335         uint32_t *p;
7336
7337         sbuf_printf(sb, "Status   Inst    Data      PC%s",
7338             cfg & F_UPDBGLACAPTPCONLY ? "" :
7339             "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
7340
7341         for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
7342                 if (cfg & F_UPDBGLACAPTPCONLY) {
7343                         sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
7344                             p[3] & 0xff, p[2], p[1], p[0]);
7345                         sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
7346                             (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
7347                             p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
7348                         sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
7349                             (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
7350                             p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
7351                             p[6] >> 16);
7352                 } else {
7353                         sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
7354                             "%08x %08x %08x %08x %08x %08x",
7355                             (p[9] >> 16) & 0xff,
7356                             p[9] & 0xffff, p[8] >> 16,
7357                             p[8] & 0xffff, p[7] >> 16,
7358                             p[7] & 0xffff, p[6] >> 16,
7359                             p[2], p[1], p[0], p[5], p[4], p[3]);
7360                 }
7361         }
7362 }
7363
7364 static int
7365 sbuf_cim_la(struct adapter *sc, struct sbuf *sb, int flags)
7366 {
7367         uint32_t cfg, *buf;
7368         int rc;
7369
7370         rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
7371         if (rc != 0)
7372                 return (rc);
7373
7374         MPASS(flags == M_WAITOK || flags == M_NOWAIT);
7375         buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
7376             M_ZERO | flags);
7377         if (buf == NULL)
7378                 return (ENOMEM);
7379
7380         rc = -t4_cim_read_la(sc, buf, NULL);
7381         if (rc != 0)
7382                 goto done;
7383         if (chip_id(sc) < CHELSIO_T6)
7384                 sbuf_cim_la4(sc, sb, buf, cfg);
7385         else
7386                 sbuf_cim_la6(sc, sb, buf, cfg);
7387
7388 done:
7389         free(buf, M_CXGBE);
7390         return (rc);
7391 }
7392
7393 static int
7394 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
7395 {
7396         struct adapter *sc = arg1;
7397         struct sbuf *sb;
7398         int rc;
7399
7400         rc = sysctl_wire_old_buffer(req, 0);
7401         if (rc != 0)
7402                 return (rc);
7403         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7404         if (sb == NULL)
7405                 return (ENOMEM);
7406
7407         rc = sbuf_cim_la(sc, sb, M_WAITOK);
7408         if (rc == 0)
7409                 rc = sbuf_finish(sb);
7410         sbuf_delete(sb);
7411         return (rc);
7412 }
7413
7414 bool
7415 t4_os_dump_cimla(struct adapter *sc, int arg, bool verbose)
7416 {
7417         struct sbuf sb;
7418         int rc;
7419
7420         if (sbuf_new(&sb, NULL, 4096, SBUF_AUTOEXTEND) != &sb)
7421                 return (false);
7422         rc = sbuf_cim_la(sc, &sb, M_NOWAIT);
7423         if (rc == 0) {
7424                 rc = sbuf_finish(&sb);
7425                 if (rc == 0) {
7426                         log(LOG_DEBUG, "%s: CIM LA dump follows.\n%s",
7427                                 device_get_nameunit(sc->dev), sbuf_data(&sb));
7428                 }
7429         }
7430         sbuf_delete(&sb);
7431         return (false);
7432 }
7433
7434 static int
7435 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
7436 {
7437         struct adapter *sc = arg1;
7438         u_int i;
7439         struct sbuf *sb;
7440         uint32_t *buf, *p;
7441         int rc;
7442
7443         rc = sysctl_wire_old_buffer(req, 0);
7444         if (rc != 0)
7445                 return (rc);
7446
7447         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7448         if (sb == NULL)
7449                 return (ENOMEM);
7450
7451         buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
7452             M_ZERO | M_WAITOK);
7453
7454         t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
7455         p = buf;
7456
7457         for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
7458                 sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
7459                     p[1], p[0]);
7460         }
7461
7462         sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
7463         for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
7464                 sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
7465                     (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
7466                     (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
7467                     (p[1] >> 2) | ((p[2] & 3) << 30),
7468                     (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
7469                     p[0] & 1);
7470         }
7471
7472         rc = sbuf_finish(sb);
7473         sbuf_delete(sb);
7474         free(buf, M_CXGBE);
7475         return (rc);
7476 }
7477
7478 static int
7479 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
7480 {
7481         struct adapter *sc = arg1;
7482         u_int i;
7483         struct sbuf *sb;
7484         uint32_t *buf, *p;
7485         int rc;
7486
7487         rc = sysctl_wire_old_buffer(req, 0);
7488         if (rc != 0)
7489                 return (rc);
7490
7491         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7492         if (sb == NULL)
7493                 return (ENOMEM);
7494
7495         buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
7496             M_ZERO | M_WAITOK);
7497
7498         t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
7499         p = buf;
7500
7501         sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
7502         for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
7503                 sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
7504                     (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
7505                     p[4], p[3], p[2], p[1], p[0]);
7506         }
7507
7508         sbuf_printf(sb, "\n\nCntl ID               Data");
7509         for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
7510                 sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
7511                     (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
7512         }
7513
7514         rc = sbuf_finish(sb);
7515         sbuf_delete(sb);
7516         free(buf, M_CXGBE);
7517         return (rc);
7518 }
7519
7520 static int
7521 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
7522 {
7523         struct adapter *sc = arg1;
7524         struct sbuf *sb;
7525         int rc, i;
7526         uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
7527         uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
7528         uint16_t thres[CIM_NUM_IBQ];
7529         uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
7530         uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
7531         u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
7532
7533         cim_num_obq = sc->chip_params->cim_num_obq;
7534         if (is_t4(sc)) {
7535                 ibq_rdaddr = A_UP_IBQ_0_RDADDR;
7536                 obq_rdaddr = A_UP_OBQ_0_REALADDR;
7537         } else {
7538                 ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
7539                 obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
7540         }
7541         nq = CIM_NUM_IBQ + cim_num_obq;
7542
7543         rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
7544         if (rc == 0)
7545                 rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
7546         if (rc != 0)
7547                 return (rc);
7548
7549         t4_read_cimq_cfg(sc, base, size, thres);
7550
7551         rc = sysctl_wire_old_buffer(req, 0);
7552         if (rc != 0)
7553                 return (rc);
7554
7555         sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
7556         if (sb == NULL)
7557                 return (ENOMEM);
7558
7559         sbuf_printf(sb,
7560             "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
7561
7562         for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
7563                 sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
7564                     qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
7565                     G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
7566                     G_QUEREMFLITS(p[2]) * 16);
7567         for ( ; i < nq; i++, p += 4, wr += 2)
7568                 sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
7569                     base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
7570                     wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
7571                     G_QUEREMFLITS(p[2]) * 16);
7572
7573         rc = sbuf_finish(sb);
7574         sbuf_delete(sb);
7575
7576         return (rc);
7577 }
7578
7579 static int
7580 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
7581 {
7582         struct adapter *sc = arg1;
7583         struct sbuf *sb;
7584         int rc;
7585         struct tp_cpl_stats stats;
7586
7587         rc = sysctl_wire_old_buffer(req, 0);
7588         if (rc != 0)
7589                 return (rc);
7590
7591         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7592         if (sb == NULL)
7593                 return (ENOMEM);
7594
7595         mtx_lock(&sc->reg_lock);
7596         t4_tp_get_cpl_stats(sc, &stats, 0);
7597         mtx_unlock(&sc->reg_lock);
7598
7599         if (sc->chip_params->nchan > 2) {
7600                 sbuf_printf(sb, "                 channel 0  channel 1"
7601                     "  channel 2  channel 3");
7602                 sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
7603                     stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
7604                 sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
7605                     stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
7606         } else {
7607                 sbuf_printf(sb, "                 channel 0  channel 1");
7608                 sbuf_printf(sb, "\nCPL requests:   %10u %10u",
7609                     stats.req[0], stats.req[1]);
7610                 sbuf_printf(sb, "\nCPL responses:   %10u %10u",
7611                     stats.rsp[0], stats.rsp[1]);
7612         }
7613
7614         rc = sbuf_finish(sb);
7615         sbuf_delete(sb);
7616
7617         return (rc);
7618 }
7619
7620 static int
7621 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
7622 {
7623         struct adapter *sc = arg1;
7624         struct sbuf *sb;
7625         int rc;
7626         struct tp_usm_stats stats;
7627
7628         rc = sysctl_wire_old_buffer(req, 0);
7629         if (rc != 0)
7630                 return(rc);
7631
7632         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7633         if (sb == NULL)
7634                 return (ENOMEM);
7635
7636         t4_get_usm_stats(sc, &stats, 1);
7637
7638         sbuf_printf(sb, "Frames: %u\n", stats.frames);
7639         sbuf_printf(sb, "Octets: %ju\n", stats.octets);
7640         sbuf_printf(sb, "Drops:  %u", stats.drops);
7641
7642         rc = sbuf_finish(sb);
7643         sbuf_delete(sb);
7644
7645         return (rc);
7646 }
7647
7648 static const char * const devlog_level_strings[] = {
7649         [FW_DEVLOG_LEVEL_EMERG]         = "EMERG",
7650         [FW_DEVLOG_LEVEL_CRIT]          = "CRIT",
7651         [FW_DEVLOG_LEVEL_ERR]           = "ERR",
7652         [FW_DEVLOG_LEVEL_NOTICE]        = "NOTICE",
7653         [FW_DEVLOG_LEVEL_INFO]          = "INFO",
7654         [FW_DEVLOG_LEVEL_DEBUG]         = "DEBUG"
7655 };
7656
7657 static const char * const devlog_facility_strings[] = {
7658         [FW_DEVLOG_FACILITY_CORE]       = "CORE",
7659         [FW_DEVLOG_FACILITY_CF]         = "CF",
7660         [FW_DEVLOG_FACILITY_SCHED]      = "SCHED",
7661         [FW_DEVLOG_FACILITY_TIMER]      = "TIMER",
7662         [FW_DEVLOG_FACILITY_RES]        = "RES",
7663         [FW_DEVLOG_FACILITY_HW]         = "HW",
7664         [FW_DEVLOG_FACILITY_FLR]        = "FLR",
7665         [FW_DEVLOG_FACILITY_DMAQ]       = "DMAQ",
7666         [FW_DEVLOG_FACILITY_PHY]        = "PHY",
7667         [FW_DEVLOG_FACILITY_MAC]        = "MAC",
7668         [FW_DEVLOG_FACILITY_PORT]       = "PORT",
7669         [FW_DEVLOG_FACILITY_VI]         = "VI",
7670         [FW_DEVLOG_FACILITY_FILTER]     = "FILTER",
7671         [FW_DEVLOG_FACILITY_ACL]        = "ACL",
7672         [FW_DEVLOG_FACILITY_TM]         = "TM",
7673         [FW_DEVLOG_FACILITY_QFC]        = "QFC",
7674         [FW_DEVLOG_FACILITY_DCB]        = "DCB",
7675         [FW_DEVLOG_FACILITY_ETH]        = "ETH",
7676         [FW_DEVLOG_FACILITY_OFLD]       = "OFLD",
7677         [FW_DEVLOG_FACILITY_RI]         = "RI",
7678         [FW_DEVLOG_FACILITY_ISCSI]      = "ISCSI",
7679         [FW_DEVLOG_FACILITY_FCOE]       = "FCOE",
7680         [FW_DEVLOG_FACILITY_FOISCSI]    = "FOISCSI",
7681         [FW_DEVLOG_FACILITY_FOFCOE]     = "FOFCOE",
7682         [FW_DEVLOG_FACILITY_CHNET]      = "CHNET",
7683 };
7684
7685 static int
7686 sbuf_devlog(struct adapter *sc, struct sbuf *sb, int flags)
7687 {
7688         int i, j, rc, nentries, first = 0;
7689         struct devlog_params *dparams = &sc->params.devlog;
7690         struct fw_devlog_e *buf, *e;
7691         uint64_t ftstamp = UINT64_MAX;
7692
7693         if (dparams->addr == 0)
7694                 return (ENXIO);
7695
7696         MPASS(flags == M_WAITOK || flags == M_NOWAIT);
7697         buf = malloc(dparams->size, M_CXGBE, M_ZERO | flags);
7698         if (buf == NULL)
7699                 return (ENOMEM);
7700
7701         rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
7702         if (rc != 0)
7703                 goto done;
7704
7705         nentries = dparams->size / sizeof(struct fw_devlog_e);
7706         for (i = 0; i < nentries; i++) {
7707                 e = &buf[i];
7708
7709                 if (e->timestamp == 0)
7710                         break;  /* end */
7711
7712                 e->timestamp = be64toh(e->timestamp);
7713                 e->seqno = be32toh(e->seqno);
7714                 for (j = 0; j < 8; j++)
7715                         e->params[j] = be32toh(e->params[j]);
7716
7717                 if (e->timestamp < ftstamp) {
7718                         ftstamp = e->timestamp;
7719                         first = i;
7720                 }
7721         }
7722
7723         if (buf[first].timestamp == 0)
7724                 goto done;      /* nothing in the log */
7725
7726         sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
7727             "Seq#", "Tstamp", "Level", "Facility", "Message");
7728
7729         i = first;
7730         do {
7731                 e = &buf[i];
7732                 if (e->timestamp == 0)
7733                         break;  /* end */
7734
7735                 sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
7736                     e->seqno, e->timestamp,
7737                     (e->level < nitems(devlog_level_strings) ?
7738                         devlog_level_strings[e->level] : "UNKNOWN"),
7739                     (e->facility < nitems(devlog_facility_strings) ?
7740                         devlog_facility_strings[e->facility] : "UNKNOWN"));
7741                 sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
7742                     e->params[2], e->params[3], e->params[4],
7743                     e->params[5], e->params[6], e->params[7]);
7744
7745                 if (++i == nentries)
7746                         i = 0;
7747         } while (i != first);
7748 done:
7749         free(buf, M_CXGBE);
7750         return (rc);
7751 }
7752
7753 static int
7754 sysctl_devlog(SYSCTL_HANDLER_ARGS)
7755 {
7756         struct adapter *sc = arg1;
7757         int rc;
7758         struct sbuf *sb;
7759
7760         rc = sysctl_wire_old_buffer(req, 0);
7761         if (rc != 0)
7762                 return (rc);
7763         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7764         if (sb == NULL)
7765                 return (ENOMEM);
7766
7767         rc = sbuf_devlog(sc, sb, M_WAITOK);
7768         if (rc == 0)
7769                 rc = sbuf_finish(sb);
7770         sbuf_delete(sb);
7771         return (rc);
7772 }
7773
7774 void
7775 t4_os_dump_devlog(struct adapter *sc)
7776 {
7777         int rc;
7778         struct sbuf sb;
7779
7780         if (sbuf_new(&sb, NULL, 4096, SBUF_AUTOEXTEND) != &sb)
7781                 return;
7782         rc = sbuf_devlog(sc, &sb, M_NOWAIT);
7783         if (rc == 0) {
7784                 rc = sbuf_finish(&sb);
7785                 if (rc == 0) {
7786                         log(LOG_DEBUG, "%s: device log follows.\n%s",
7787                                 device_get_nameunit(sc->dev), sbuf_data(&sb));
7788                 }
7789         }
7790         sbuf_delete(&sb);
7791 }
7792
7793 static int
7794 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
7795 {
7796         struct adapter *sc = arg1;
7797         struct sbuf *sb;
7798         int rc;
7799         struct tp_fcoe_stats stats[MAX_NCHAN];
7800         int i, nchan = sc->chip_params->nchan;
7801
7802         rc = sysctl_wire_old_buffer(req, 0);
7803         if (rc != 0)
7804                 return (rc);
7805
7806         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7807         if (sb == NULL)
7808                 return (ENOMEM);
7809
7810         for (i = 0; i < nchan; i++)
7811                 t4_get_fcoe_stats(sc, i, &stats[i], 1);
7812
7813         if (nchan > 2) {
7814                 sbuf_printf(sb, "                   channel 0        channel 1"
7815                     "        channel 2        channel 3");
7816                 sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
7817                     stats[0].octets_ddp, stats[1].octets_ddp,
7818                     stats[2].octets_ddp, stats[3].octets_ddp);
7819                 sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
7820                     stats[0].frames_ddp, stats[1].frames_ddp,
7821                     stats[2].frames_ddp, stats[3].frames_ddp);
7822                 sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
7823                     stats[0].frames_drop, stats[1].frames_drop,
7824                     stats[2].frames_drop, stats[3].frames_drop);
7825         } else {
7826                 sbuf_printf(sb, "                   channel 0        channel 1");
7827                 sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
7828                     stats[0].octets_ddp, stats[1].octets_ddp);
7829                 sbuf_printf(sb, "\nframesDDP:  %16u %16u",
7830                     stats[0].frames_ddp, stats[1].frames_ddp);
7831                 sbuf_printf(sb, "\nframesDrop: %16u %16u",
7832                     stats[0].frames_drop, stats[1].frames_drop);
7833         }
7834
7835         rc = sbuf_finish(sb);
7836         sbuf_delete(sb);
7837
7838         return (rc);
7839 }
7840
7841 static int
7842 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
7843 {
7844         struct adapter *sc = arg1;
7845         struct sbuf *sb;
7846         int rc, i;
7847         unsigned int map, kbps, ipg, mode;
7848         unsigned int pace_tab[NTX_SCHED];
7849
7850         rc = sysctl_wire_old_buffer(req, 0);
7851         if (rc != 0)
7852                 return (rc);
7853
7854         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7855         if (sb == NULL)
7856                 return (ENOMEM);
7857
7858         map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
7859         mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
7860         t4_read_pace_tbl(sc, pace_tab);
7861
7862         sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
7863             "Class IPG (0.1 ns)   Flow IPG (us)");
7864
7865         for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
7866                 t4_get_tx_sched(sc, i, &kbps, &ipg, 1);
7867                 sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
7868                     (mode & (1 << i)) ? "flow" : "class", map & 3);
7869                 if (kbps)
7870                         sbuf_printf(sb, "%9u     ", kbps);
7871                 else
7872                         sbuf_printf(sb, " disabled     ");
7873
7874                 if (ipg)
7875                         sbuf_printf(sb, "%13u        ", ipg);
7876                 else
7877                         sbuf_printf(sb, "     disabled        ");
7878
7879                 if (pace_tab[i])
7880                         sbuf_printf(sb, "%10u", pace_tab[i]);
7881                 else
7882                         sbuf_printf(sb, "  disabled");
7883         }
7884
7885         rc = sbuf_finish(sb);
7886         sbuf_delete(sb);
7887
7888         return (rc);
7889 }
7890
7891 static int
7892 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
7893 {
7894         struct adapter *sc = arg1;
7895         struct sbuf *sb;
7896         int rc, i, j;
7897         uint64_t *p0, *p1;
7898         struct lb_port_stats s[2];
7899         static const char *stat_name[] = {
7900                 "OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
7901                 "UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
7902                 "Frames128To255:", "Frames256To511:", "Frames512To1023:",
7903                 "Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
7904                 "BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
7905                 "BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
7906                 "BG2FramesTrunc:", "BG3FramesTrunc:"
7907         };
7908
7909         rc = sysctl_wire_old_buffer(req, 0);
7910         if (rc != 0)
7911                 return (rc);
7912
7913         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7914         if (sb == NULL)
7915                 return (ENOMEM);
7916
7917         memset(s, 0, sizeof(s));
7918
7919         for (i = 0; i < sc->chip_params->nchan; i += 2) {
7920                 t4_get_lb_stats(sc, i, &s[0]);
7921                 t4_get_lb_stats(sc, i + 1, &s[1]);
7922
7923                 p0 = &s[0].octets;
7924                 p1 = &s[1].octets;
7925                 sbuf_printf(sb, "%s                       Loopback %u"
7926                     "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
7927
7928                 for (j = 0; j < nitems(stat_name); j++)
7929                         sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
7930                                    *p0++, *p1++);
7931         }
7932
7933         rc = sbuf_finish(sb);
7934         sbuf_delete(sb);
7935
7936         return (rc);
7937 }
7938
7939 static int
7940 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
7941 {
7942         int rc = 0;
7943         struct port_info *pi = arg1;
7944         struct link_config *lc = &pi->link_cfg;
7945         struct sbuf *sb;
7946
7947         rc = sysctl_wire_old_buffer(req, 0);
7948         if (rc != 0)
7949                 return(rc);
7950         sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
7951         if (sb == NULL)
7952                 return (ENOMEM);
7953
7954         if (lc->link_ok || lc->link_down_rc == 255)
7955                 sbuf_printf(sb, "n/a");
7956         else
7957                 sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
7958
7959         rc = sbuf_finish(sb);
7960         sbuf_delete(sb);
7961
7962         return (rc);
7963 }
7964
7965 struct mem_desc {
7966         unsigned int base;
7967         unsigned int limit;
7968         unsigned int idx;
7969 };
7970
7971 static int
7972 mem_desc_cmp(const void *a, const void *b)
7973 {
7974         return ((const struct mem_desc *)a)->base -
7975                ((const struct mem_desc *)b)->base;
7976 }
7977
7978 static void
7979 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
7980     unsigned int to)
7981 {
7982         unsigned int size;
7983
7984         if (from == to)
7985                 return;
7986
7987         size = to - from + 1;
7988         if (size == 0)
7989                 return;
7990
7991         /* XXX: need humanize_number(3) in libkern for a more readable 'size' */
7992         sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
7993 }
7994
7995 static int
7996 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
7997 {
7998         struct adapter *sc = arg1;
7999         struct sbuf *sb;
8000         int rc, i, n;
8001         uint32_t lo, hi, used, alloc;
8002         static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
8003         static const char *region[] = {
8004                 "DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
8005                 "Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
8006                 "Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
8007                 "TDDP region:", "TPT region:", "STAG region:", "RQ region:",
8008                 "RQUDP region:", "PBL region:", "TXPBL region:",
8009                 "DBVFIFO region:", "ULPRX state:", "ULPTX state:",
8010                 "On-chip queues:", "TLS keys:",
8011         };
8012         struct mem_desc avail[4];
8013         struct mem_desc mem[nitems(region) + 3];        /* up to 3 holes */
8014         struct mem_desc *md = mem;
8015
8016         rc = sysctl_wire_old_buffer(req, 0);
8017         if (rc != 0)
8018                 return (rc);
8019
8020         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8021         if (sb == NULL)
8022                 return (ENOMEM);
8023
8024         for (i = 0; i < nitems(mem); i++) {
8025                 mem[i].limit = 0;
8026                 mem[i].idx = i;
8027         }
8028
8029         /* Find and sort the populated memory ranges */
8030         i = 0;
8031         lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
8032         if (lo & F_EDRAM0_ENABLE) {
8033                 hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
8034                 avail[i].base = G_EDRAM0_BASE(hi) << 20;
8035                 avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
8036                 avail[i].idx = 0;
8037                 i++;
8038         }
8039         if (lo & F_EDRAM1_ENABLE) {
8040                 hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
8041                 avail[i].base = G_EDRAM1_BASE(hi) << 20;
8042                 avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
8043                 avail[i].idx = 1;
8044                 i++;
8045         }
8046         if (lo & F_EXT_MEM_ENABLE) {
8047                 hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
8048                 avail[i].base = G_EXT_MEM_BASE(hi) << 20;
8049                 avail[i].limit = avail[i].base +
8050                     (G_EXT_MEM_SIZE(hi) << 20);
8051                 avail[i].idx = is_t5(sc) ? 3 : 2;       /* Call it MC0 for T5 */
8052                 i++;
8053         }
8054         if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
8055                 hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
8056                 avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
8057                 avail[i].limit = avail[i].base +
8058                     (G_EXT_MEM1_SIZE(hi) << 20);
8059                 avail[i].idx = 4;
8060                 i++;
8061         }
8062         if (!i)                                    /* no memory available */
8063                 return 0;
8064         qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
8065
8066         (md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
8067         (md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
8068         (md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
8069         (md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
8070         (md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
8071         (md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
8072         (md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
8073         (md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
8074         (md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
8075
8076         /* the next few have explicit upper bounds */
8077         md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
8078         md->limit = md->base - 1 +
8079                     t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
8080                     G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
8081         md++;
8082
8083         md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
8084         md->limit = md->base - 1 +
8085                     t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
8086                     G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
8087         md++;
8088
8089         if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
8090                 if (chip_id(sc) <= CHELSIO_T5)
8091                         md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
8092                 else
8093                         md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
8094                 md->limit = 0;
8095         } else {
8096                 md->base = 0;
8097                 md->idx = nitems(region);  /* hide it */
8098         }
8099         md++;
8100
8101 #define ulp_region(reg) \
8102         md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
8103         (md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
8104
8105         ulp_region(RX_ISCSI);
8106         ulp_region(RX_TDDP);
8107         ulp_region(TX_TPT);
8108         ulp_region(RX_STAG);
8109         ulp_region(RX_RQ);
8110         ulp_region(RX_RQUDP);
8111         ulp_region(RX_PBL);
8112         ulp_region(TX_PBL);
8113 #undef ulp_region
8114
8115         md->base = 0;
8116         md->idx = nitems(region);
8117         if (!is_t4(sc)) {
8118                 uint32_t size = 0;
8119                 uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
8120                 uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
8121
8122                 if (is_t5(sc)) {
8123                         if (sge_ctrl & F_VFIFO_ENABLE)
8124                                 size = G_DBVFIFO_SIZE(fifo_size);
8125                 } else
8126                         size = G_T6_DBVFIFO_SIZE(fifo_size);
8127
8128                 if (size) {
8129                         md->base = G_BASEADDR(t4_read_reg(sc,
8130                             A_SGE_DBVFIFO_BADDR));
8131                         md->limit = md->base + (size << 2) - 1;
8132                 }
8133         }
8134         md++;
8135
8136         md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
8137         md->limit = 0;
8138         md++;
8139         md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
8140         md->limit = 0;
8141         md++;
8142
8143         md->base = sc->vres.ocq.start;
8144         if (sc->vres.ocq.size)
8145                 md->limit = md->base + sc->vres.ocq.size - 1;
8146         else
8147                 md->idx = nitems(region);  /* hide it */
8148         md++;
8149
8150         md->base = sc->vres.key.start;
8151         if (sc->vres.key.size)
8152                 md->limit = md->base + sc->vres.key.size - 1;
8153         else
8154                 md->idx = nitems(region);  /* hide it */
8155         md++;
8156
8157         /* add any address-space holes, there can be up to 3 */
8158         for (n = 0; n < i - 1; n++)
8159                 if (avail[n].limit < avail[n + 1].base)
8160                         (md++)->base = avail[n].limit;
8161         if (avail[n].limit)
8162                 (md++)->base = avail[n].limit;
8163
8164         n = md - mem;
8165         qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
8166
8167         for (lo = 0; lo < i; lo++)
8168                 mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
8169                                 avail[lo].limit - 1);
8170
8171         sbuf_printf(sb, "\n");
8172         for (i = 0; i < n; i++) {
8173                 if (mem[i].idx >= nitems(region))
8174                         continue;                        /* skip holes */
8175                 if (!mem[i].limit)
8176                         mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
8177                 mem_region_show(sb, region[mem[i].idx], mem[i].base,
8178                                 mem[i].limit);
8179         }
8180
8181         sbuf_printf(sb, "\n");
8182         lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
8183         hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
8184         mem_region_show(sb, "uP RAM:", lo, hi);
8185
8186         lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
8187         hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
8188         mem_region_show(sb, "uP Extmem2:", lo, hi);
8189
8190         lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
8191         sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
8192                    G_PMRXMAXPAGE(lo),
8193                    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
8194                    (lo & F_PMRXNUMCHN) ? 2 : 1);
8195
8196         lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
8197         hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
8198         sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
8199                    G_PMTXMAXPAGE(lo),
8200                    hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
8201                    hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
8202         sbuf_printf(sb, "%u p-structs\n",
8203                    t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
8204
8205         for (i = 0; i < 4; i++) {
8206                 if (chip_id(sc) > CHELSIO_T5)
8207                         lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
8208                 else
8209                         lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
8210                 if (is_t5(sc)) {
8211                         used = G_T5_USED(lo);
8212                         alloc = G_T5_ALLOC(lo);
8213                 } else {
8214                         used = G_USED(lo);
8215                         alloc = G_ALLOC(lo);
8216                 }
8217                 /* For T6 these are MAC buffer groups */
8218                 sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
8219                     i, used, alloc);
8220         }
8221         for (i = 0; i < sc->chip_params->nchan; i++) {
8222                 if (chip_id(sc) > CHELSIO_T5)
8223                         lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
8224                 else
8225                         lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
8226                 if (is_t5(sc)) {
8227                         used = G_T5_USED(lo);
8228                         alloc = G_T5_ALLOC(lo);
8229                 } else {
8230                         used = G_USED(lo);
8231                         alloc = G_ALLOC(lo);
8232                 }
8233                 /* For T6 these are MAC buffer groups */
8234                 sbuf_printf(sb,
8235                     "\nLoopback %d using %u pages out of %u allocated",
8236                     i, used, alloc);
8237         }
8238
8239         rc = sbuf_finish(sb);
8240         sbuf_delete(sb);
8241
8242         return (rc);
8243 }
8244
8245 static inline void
8246 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
8247 {
8248         *mask = x | y;
8249         y = htobe64(y);
8250         memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
8251 }
8252
8253 static int
8254 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
8255 {
8256         struct adapter *sc = arg1;
8257         struct sbuf *sb;
8258         int rc, i;
8259
8260         MPASS(chip_id(sc) <= CHELSIO_T5);
8261
8262         rc = sysctl_wire_old_buffer(req, 0);
8263         if (rc != 0)
8264                 return (rc);
8265
8266         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8267         if (sb == NULL)
8268                 return (ENOMEM);
8269
8270         sbuf_printf(sb,
8271             "Idx  Ethernet address     Mask     Vld Ports PF"
8272             "  VF              Replication             P0 P1 P2 P3  ML");
8273         for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
8274                 uint64_t tcamx, tcamy, mask;
8275                 uint32_t cls_lo, cls_hi;
8276                 uint8_t addr[ETHER_ADDR_LEN];
8277
8278                 tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
8279                 tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
8280                 if (tcamx & tcamy)
8281                         continue;
8282                 tcamxy2valmask(tcamx, tcamy, addr, &mask);
8283                 cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
8284                 cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
8285                 sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
8286                            "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
8287                            addr[3], addr[4], addr[5], (uintmax_t)mask,
8288                            (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
8289                            G_PORTMAP(cls_hi), G_PF(cls_lo),
8290                            (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
8291
8292                 if (cls_lo & F_REPLICATE) {
8293                         struct fw_ldst_cmd ldst_cmd;
8294
8295                         memset(&ldst_cmd, 0, sizeof(ldst_cmd));
8296                         ldst_cmd.op_to_addrspace =
8297                             htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
8298                                 F_FW_CMD_REQUEST | F_FW_CMD_READ |
8299                                 V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
8300                         ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
8301                         ldst_cmd.u.mps.rplc.fid_idx =
8302                             htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
8303                                 V_FW_LDST_CMD_IDX(i));
8304
8305                         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
8306                             "t4mps");
8307                         if (rc)
8308                                 break;
8309                         rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
8310                             sizeof(ldst_cmd), &ldst_cmd);
8311                         end_synchronized_op(sc, 0);
8312
8313                         if (rc != 0) {
8314                                 sbuf_printf(sb, "%36d", rc);
8315                                 rc = 0;
8316                         } else {
8317                                 sbuf_printf(sb, " %08x %08x %08x %08x",
8318                                     be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
8319                                     be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
8320                                     be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
8321                                     be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
8322                         }
8323                 } else
8324                         sbuf_printf(sb, "%36s", "");
8325
8326                 sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
8327                     G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
8328                     G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
8329         }
8330
8331         if (rc)
8332                 (void) sbuf_finish(sb);
8333         else
8334                 rc = sbuf_finish(sb);
8335         sbuf_delete(sb);
8336
8337         return (rc);
8338 }
8339
8340 static int
8341 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
8342 {
8343         struct adapter *sc = arg1;
8344         struct sbuf *sb;
8345         int rc, i;
8346
8347         MPASS(chip_id(sc) > CHELSIO_T5);
8348
8349         rc = sysctl_wire_old_buffer(req, 0);
8350         if (rc != 0)
8351                 return (rc);
8352
8353         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8354         if (sb == NULL)
8355                 return (ENOMEM);
8356
8357         sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
8358             "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
8359             "                           Replication"
8360             "                                    P0 P1 P2 P3  ML\n");
8361
8362         for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
8363                 uint8_t dip_hit, vlan_vld, lookup_type, port_num;
8364                 uint16_t ivlan;
8365                 uint64_t tcamx, tcamy, val, mask;
8366                 uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
8367                 uint8_t addr[ETHER_ADDR_LEN];
8368
8369                 ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
8370                 if (i < 256)
8371                         ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
8372                 else
8373                         ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
8374                 t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
8375                 val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
8376                 tcamy = G_DMACH(val) << 32;
8377                 tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
8378                 data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
8379                 lookup_type = G_DATALKPTYPE(data2);
8380                 port_num = G_DATAPORTNUM(data2);
8381                 if (lookup_type && lookup_type != M_DATALKPTYPE) {
8382                         /* Inner header VNI */
8383                         vniy = ((data2 & F_DATAVIDH2) << 23) |
8384                                        (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
8385                         dip_hit = data2 & F_DATADIPHIT;
8386                         vlan_vld = 0;
8387                 } else {
8388                         vniy = 0;
8389                         dip_hit = 0;
8390                         vlan_vld = data2 & F_DATAVIDH2;
8391                         ivlan = G_VIDL(val);
8392                 }
8393
8394                 ctl |= V_CTLXYBITSEL(1);
8395                 t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
8396                 val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
8397                 tcamx = G_DMACH(val) << 32;
8398                 tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
8399                 data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
8400                 if (lookup_type && lookup_type != M_DATALKPTYPE) {
8401                         /* Inner header VNI mask */
8402                         vnix = ((data2 & F_DATAVIDH2) << 23) |
8403                                (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
8404                 } else
8405                         vnix = 0;
8406
8407                 if (tcamx & tcamy)
8408                         continue;
8409                 tcamxy2valmask(tcamx, tcamy, addr, &mask);
8410
8411                 cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
8412                 cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
8413
8414                 if (lookup_type && lookup_type != M_DATALKPTYPE) {
8415                         sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
8416                             "%012jx %06x %06x    -    -   %3c"
8417                             "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
8418                             addr[1], addr[2], addr[3], addr[4], addr[5],
8419                             (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
8420                             port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
8421                             G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
8422                             cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
8423                 } else {
8424                         sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
8425                             "%012jx    -       -   ", i, addr[0], addr[1],
8426                             addr[2], addr[3], addr[4], addr[5],
8427                             (uintmax_t)mask);
8428
8429                         if (vlan_vld)
8430                                 sbuf_printf(sb, "%4u   Y     ", ivlan);
8431                         else
8432                                 sbuf_printf(sb, "  -    N     ");
8433
8434                         sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
8435                             lookup_type ? 'I' : 'O', port_num,
8436                             cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
8437                             G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
8438                             cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
8439                 }
8440
8441
8442                 if (cls_lo & F_T6_REPLICATE) {
8443                         struct fw_ldst_cmd ldst_cmd;
8444
8445                         memset(&ldst_cmd, 0, sizeof(ldst_cmd));
8446                         ldst_cmd.op_to_addrspace =
8447                             htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
8448                                 F_FW_CMD_REQUEST | F_FW_CMD_READ |
8449                                 V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
8450                         ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
8451                         ldst_cmd.u.mps.rplc.fid_idx =
8452                             htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
8453                                 V_FW_LDST_CMD_IDX(i));
8454
8455                         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
8456                             "t6mps");
8457                         if (rc)
8458                                 break;
8459                         rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
8460                             sizeof(ldst_cmd), &ldst_cmd);
8461                         end_synchronized_op(sc, 0);
8462
8463                         if (rc != 0) {
8464                                 sbuf_printf(sb, "%72d", rc);
8465                                 rc = 0;
8466                         } else {
8467                                 sbuf_printf(sb, " %08x %08x %08x %08x"
8468                                     " %08x %08x %08x %08x",
8469                                     be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
8470                                     be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
8471                                     be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
8472                                     be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
8473                                     be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
8474                                     be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
8475                                     be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
8476                                     be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
8477                         }
8478                 } else
8479                         sbuf_printf(sb, "%72s", "");
8480
8481                 sbuf_printf(sb, "%4u%3u%3u%3u %#x",
8482                     G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
8483                     G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
8484                     (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
8485         }
8486
8487         if (rc)
8488                 (void) sbuf_finish(sb);
8489         else
8490                 rc = sbuf_finish(sb);
8491         sbuf_delete(sb);
8492
8493         return (rc);
8494 }
8495
8496 static int
8497 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
8498 {
8499         struct adapter *sc = arg1;
8500         struct sbuf *sb;
8501         int rc;
8502         uint16_t mtus[NMTUS];
8503
8504         rc = sysctl_wire_old_buffer(req, 0);
8505         if (rc != 0)
8506                 return (rc);
8507
8508         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8509         if (sb == NULL)
8510                 return (ENOMEM);
8511
8512         t4_read_mtu_tbl(sc, mtus, NULL);
8513
8514         sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
8515             mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
8516             mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
8517             mtus[14], mtus[15]);
8518
8519         rc = sbuf_finish(sb);
8520         sbuf_delete(sb);
8521
8522         return (rc);
8523 }
8524
8525 static int
8526 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
8527 {
8528         struct adapter *sc = arg1;
8529         struct sbuf *sb;
8530         int rc, i;
8531         uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
8532         uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
8533         static const char *tx_stats[MAX_PM_NSTATS] = {
8534                 "Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
8535                 "Tx FIFO wait", NULL, "Tx latency"
8536         };
8537         static const char *rx_stats[MAX_PM_NSTATS] = {
8538                 "Read:", "Write bypass:", "Write mem:", "Flush:",
8539                 "Rx FIFO wait", NULL, "Rx latency"
8540         };
8541
8542         rc = sysctl_wire_old_buffer(req, 0);
8543         if (rc != 0)
8544                 return (rc);
8545
8546         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8547         if (sb == NULL)
8548                 return (ENOMEM);
8549
8550         t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
8551         t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
8552
8553         sbuf_printf(sb, "                Tx pcmds             Tx bytes");
8554         for (i = 0; i < 4; i++) {
8555                 sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8556                     tx_cyc[i]);
8557         }
8558
8559         sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
8560         for (i = 0; i < 4; i++) {
8561                 sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8562                     rx_cyc[i]);
8563         }
8564
8565         if (chip_id(sc) > CHELSIO_T5) {
8566                 sbuf_printf(sb,
8567                     "\n              Total wait      Total occupancy");
8568                 sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8569                     tx_cyc[i]);
8570                 sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8571                     rx_cyc[i]);
8572
8573                 i += 2;
8574                 MPASS(i < nitems(tx_stats));
8575
8576                 sbuf_printf(sb,
8577                     "\n                   Reads           Total wait");
8578                 sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8579                     tx_cyc[i]);
8580                 sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8581                     rx_cyc[i]);
8582         }
8583
8584         rc = sbuf_finish(sb);
8585         sbuf_delete(sb);
8586
8587         return (rc);
8588 }
8589
8590 static int
8591 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
8592 {
8593         struct adapter *sc = arg1;
8594         struct sbuf *sb;
8595         int rc;
8596         struct tp_rdma_stats stats;
8597
8598         rc = sysctl_wire_old_buffer(req, 0);
8599         if (rc != 0)
8600                 return (rc);
8601
8602         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8603         if (sb == NULL)
8604                 return (ENOMEM);
8605
8606         mtx_lock(&sc->reg_lock);
8607         t4_tp_get_rdma_stats(sc, &stats, 0);
8608         mtx_unlock(&sc->reg_lock);
8609
8610         sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
8611         sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
8612
8613         rc = sbuf_finish(sb);
8614         sbuf_delete(sb);
8615
8616         return (rc);
8617 }
8618
8619 static int
8620 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
8621 {
8622         struct adapter *sc = arg1;
8623         struct sbuf *sb;
8624         int rc;
8625         struct tp_tcp_stats v4, v6;
8626
8627         rc = sysctl_wire_old_buffer(req, 0);
8628         if (rc != 0)
8629                 return (rc);
8630
8631         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8632         if (sb == NULL)
8633                 return (ENOMEM);
8634
8635         mtx_lock(&sc->reg_lock);
8636         t4_tp_get_tcp_stats(sc, &v4, &v6, 0);
8637         mtx_unlock(&sc->reg_lock);
8638
8639         sbuf_printf(sb,
8640             "                                IP                 IPv6\n");
8641         sbuf_printf(sb, "OutRsts:      %20u %20u\n",
8642             v4.tcp_out_rsts, v6.tcp_out_rsts);
8643         sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
8644             v4.tcp_in_segs, v6.tcp_in_segs);
8645         sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
8646             v4.tcp_out_segs, v6.tcp_out_segs);
8647         sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
8648             v4.tcp_retrans_segs, v6.tcp_retrans_segs);
8649
8650         rc = sbuf_finish(sb);
8651         sbuf_delete(sb);
8652
8653         return (rc);
8654 }
8655
8656 static int
8657 sysctl_tids(SYSCTL_HANDLER_ARGS)
8658 {
8659         struct adapter *sc = arg1;
8660         struct sbuf *sb;
8661         int rc;
8662         struct tid_info *t = &sc->tids;
8663
8664         rc = sysctl_wire_old_buffer(req, 0);
8665         if (rc != 0)
8666                 return (rc);
8667
8668         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8669         if (sb == NULL)
8670                 return (ENOMEM);
8671
8672         if (t->natids) {
8673                 sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
8674                     t->atids_in_use);
8675         }
8676
8677         if (t->nhpftids) {
8678                 sbuf_printf(sb, "HPFTID range: %u-%u, in use: %u\n",
8679                     t->hpftid_base, t->hpftid_end, t->hpftids_in_use);
8680         }
8681
8682         if (t->ntids) {
8683                 sbuf_printf(sb, "TID range: ");
8684                 if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
8685                         uint32_t b, hb;
8686
8687                         if (chip_id(sc) <= CHELSIO_T5) {
8688                                 b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
8689                                 hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
8690                         } else {
8691                                 b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
8692                                 hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
8693                         }
8694
8695                         if (b)
8696                                 sbuf_printf(sb, "%u-%u, ", t->tid_base, b - 1);
8697                         sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
8698                 } else
8699                         sbuf_printf(sb, "%u-%u", t->tid_base, t->ntids - 1);
8700                 sbuf_printf(sb, ", in use: %u\n",
8701                     atomic_load_acq_int(&t->tids_in_use));
8702         }
8703
8704         if (t->nstids) {
8705                 sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
8706                     t->stid_base + t->nstids - 1, t->stids_in_use);
8707         }
8708
8709         if (t->nftids) {
8710                 sbuf_printf(sb, "FTID range: %u-%u, in use: %u\n", t->ftid_base,
8711                     t->ftid_end, t->ftids_in_use);
8712         }
8713
8714         if (t->netids) {
8715                 sbuf_printf(sb, "ETID range: %u-%u, in use: %u\n", t->etid_base,
8716                     t->etid_base + t->netids - 1, t->etids_in_use);
8717         }
8718
8719         sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
8720             t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
8721             t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
8722
8723         rc = sbuf_finish(sb);
8724         sbuf_delete(sb);
8725
8726         return (rc);
8727 }
8728
8729 static int
8730 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
8731 {
8732         struct adapter *sc = arg1;
8733         struct sbuf *sb;
8734         int rc;
8735         struct tp_err_stats stats;
8736
8737         rc = sysctl_wire_old_buffer(req, 0);
8738         if (rc != 0)
8739                 return (rc);
8740
8741         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8742         if (sb == NULL)
8743                 return (ENOMEM);
8744
8745         mtx_lock(&sc->reg_lock);
8746         t4_tp_get_err_stats(sc, &stats, 0);
8747         mtx_unlock(&sc->reg_lock);
8748
8749         if (sc->chip_params->nchan > 2) {
8750                 sbuf_printf(sb, "                 channel 0  channel 1"
8751                     "  channel 2  channel 3\n");
8752                 sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
8753                     stats.mac_in_errs[0], stats.mac_in_errs[1],
8754                     stats.mac_in_errs[2], stats.mac_in_errs[3]);
8755                 sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
8756                     stats.hdr_in_errs[0], stats.hdr_in_errs[1],
8757                     stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
8758                 sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
8759                     stats.tcp_in_errs[0], stats.tcp_in_errs[1],
8760                     stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
8761                 sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
8762                     stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
8763                     stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
8764                 sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
8765                     stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
8766                     stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
8767                 sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
8768                     stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
8769                     stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
8770                 sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
8771                     stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
8772                     stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
8773                 sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
8774                     stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
8775                     stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
8776         } else {
8777                 sbuf_printf(sb, "                 channel 0  channel 1\n");
8778                 sbuf_printf(sb, "macInErrs:      %10u %10u\n",
8779                     stats.mac_in_errs[0], stats.mac_in_errs[1]);
8780                 sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
8781                     stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
8782                 sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
8783                     stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
8784                 sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
8785                     stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
8786                 sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
8787                     stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
8788                 sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
8789                     stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
8790                 sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
8791                     stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
8792                 sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
8793                     stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
8794         }
8795
8796         sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
8797             stats.ofld_no_neigh, stats.ofld_cong_defer);
8798
8799         rc = sbuf_finish(sb);
8800         sbuf_delete(sb);
8801
8802         return (rc);
8803 }
8804
8805 static int
8806 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
8807 {
8808         struct adapter *sc = arg1;
8809         struct tp_params *tpp = &sc->params.tp;
8810         u_int mask;
8811         int rc;
8812
8813         mask = tpp->la_mask >> 16;
8814         rc = sysctl_handle_int(oidp, &mask, 0, req);
8815         if (rc != 0 || req->newptr == NULL)
8816                 return (rc);
8817         if (mask > 0xffff)
8818                 return (EINVAL);
8819         tpp->la_mask = mask << 16;
8820         t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
8821
8822         return (0);
8823 }
8824
8825 struct field_desc {
8826         const char *name;
8827         u_int start;
8828         u_int width;
8829 };
8830
8831 static void
8832 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
8833 {
8834         char buf[32];
8835         int line_size = 0;
8836
8837         while (f->name) {
8838                 uint64_t mask = (1ULL << f->width) - 1;
8839                 int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
8840                     ((uintmax_t)v >> f->start) & mask);
8841
8842                 if (line_size + len >= 79) {
8843                         line_size = 8;
8844                         sbuf_printf(sb, "\n        ");
8845                 }
8846                 sbuf_printf(sb, "%s ", buf);
8847                 line_size += len + 1;
8848                 f++;
8849         }
8850         sbuf_printf(sb, "\n");
8851 }
8852
8853 static const struct field_desc tp_la0[] = {
8854         { "RcfOpCodeOut", 60, 4 },
8855         { "State", 56, 4 },
8856         { "WcfState", 52, 4 },
8857         { "RcfOpcSrcOut", 50, 2 },
8858         { "CRxError", 49, 1 },
8859         { "ERxError", 48, 1 },
8860         { "SanityFailed", 47, 1 },
8861         { "SpuriousMsg", 46, 1 },
8862         { "FlushInputMsg", 45, 1 },
8863         { "FlushInputCpl", 44, 1 },
8864         { "RssUpBit", 43, 1 },
8865         { "RssFilterHit", 42, 1 },
8866         { "Tid", 32, 10 },
8867         { "InitTcb", 31, 1 },
8868         { "LineNumber", 24, 7 },
8869         { "Emsg", 23, 1 },
8870         { "EdataOut", 22, 1 },
8871         { "Cmsg", 21, 1 },
8872         { "CdataOut", 20, 1 },
8873         { "EreadPdu", 19, 1 },
8874         { "CreadPdu", 18, 1 },
8875         { "TunnelPkt", 17, 1 },
8876         { "RcfPeerFin", 16, 1 },
8877         { "RcfReasonOut", 12, 4 },
8878         { "TxCchannel", 10, 2 },
8879         { "RcfTxChannel", 8, 2 },
8880         { "RxEchannel", 6, 2 },
8881         { "RcfRxChannel", 5, 1 },
8882         { "RcfDataOutSrdy", 4, 1 },
8883         { "RxDvld", 3, 1 },
8884         { "RxOoDvld", 2, 1 },
8885         { "RxCongestion", 1, 1 },
8886         { "TxCongestion", 0, 1 },
8887         { NULL }
8888 };
8889
8890 static const struct field_desc tp_la1[] = {
8891         { "CplCmdIn", 56, 8 },
8892         { "CplCmdOut", 48, 8 },
8893         { "ESynOut", 47, 1 },
8894         { "EAckOut", 46, 1 },
8895         { "EFinOut", 45, 1 },
8896         { "ERstOut", 44, 1 },
8897         { "SynIn", 43, 1 },
8898         { "AckIn", 42, 1 },
8899         { "FinIn", 41, 1 },
8900         { "RstIn", 40, 1 },
8901         { "DataIn", 39, 1 },
8902         { "DataInVld", 38, 1 },
8903         { "PadIn", 37, 1 },
8904         { "RxBufEmpty", 36, 1 },
8905         { "RxDdp", 35, 1 },
8906         { "RxFbCongestion", 34, 1 },
8907         { "TxFbCongestion", 33, 1 },
8908         { "TxPktSumSrdy", 32, 1 },
8909         { "RcfUlpType", 28, 4 },
8910         { "Eread", 27, 1 },
8911         { "Ebypass", 26, 1 },
8912         { "Esave", 25, 1 },
8913         { "Static0", 24, 1 },
8914         { "Cread", 23, 1 },
8915         { "Cbypass", 22, 1 },
8916         { "Csave", 21, 1 },
8917         { "CPktOut", 20, 1 },
8918         { "RxPagePoolFull", 18, 2 },
8919         { "RxLpbkPkt", 17, 1 },
8920         { "TxLpbkPkt", 16, 1 },
8921         { "RxVfValid", 15, 1 },
8922         { "SynLearned", 14, 1 },
8923         { "SetDelEntry", 13, 1 },
8924         { "SetInvEntry", 12, 1 },
8925         { "CpcmdDvld", 11, 1 },
8926         { "CpcmdSave", 10, 1 },
8927         { "RxPstructsFull", 8, 2 },
8928         { "EpcmdDvld", 7, 1 },
8929         { "EpcmdFlush", 6, 1 },
8930         { "EpcmdTrimPrefix", 5, 1 },
8931         { "EpcmdTrimPostfix", 4, 1 },
8932         { "ERssIp4Pkt", 3, 1 },
8933         { "ERssIp6Pkt", 2, 1 },
8934         { "ERssTcpUdpPkt", 1, 1 },
8935         { "ERssFceFipPkt", 0, 1 },
8936         { NULL }
8937 };
8938
8939 static const struct field_desc tp_la2[] = {
8940         { "CplCmdIn", 56, 8 },
8941         { "MpsVfVld", 55, 1 },
8942         { "MpsPf", 52, 3 },
8943         { "MpsVf", 44, 8 },
8944         { "SynIn", 43, 1 },
8945         { "AckIn", 42, 1 },
8946         { "FinIn", 41, 1 },
8947         { "RstIn", 40, 1 },
8948         { "DataIn", 39, 1 },
8949         { "DataInVld", 38, 1 },
8950         { "PadIn", 37, 1 },
8951         { "RxBufEmpty", 36, 1 },
8952         { "RxDdp", 35, 1 },
8953         { "RxFbCongestion", 34, 1 },
8954         { "TxFbCongestion", 33, 1 },
8955         { "TxPktSumSrdy", 32, 1 },
8956         { "RcfUlpType", 28, 4 },
8957         { "Eread", 27, 1 },
8958         { "Ebypass", 26, 1 },
8959         { "Esave", 25, 1 },
8960         { "Static0", 24, 1 },
8961         { "Cread", 23, 1 },
8962         { "Cbypass", 22, 1 },
8963         { "Csave", 21, 1 },
8964         { "CPktOut", 20, 1 },
8965         { "RxPagePoolFull", 18, 2 },
8966         { "RxLpbkPkt", 17, 1 },
8967         { "TxLpbkPkt", 16, 1 },
8968         { "RxVfValid", 15, 1 },
8969         { "SynLearned", 14, 1 },
8970         { "SetDelEntry", 13, 1 },
8971         { "SetInvEntry", 12, 1 },
8972         { "CpcmdDvld", 11, 1 },
8973         { "CpcmdSave", 10, 1 },
8974         { "RxPstructsFull", 8, 2 },
8975         { "EpcmdDvld", 7, 1 },
8976         { "EpcmdFlush", 6, 1 },
8977         { "EpcmdTrimPrefix", 5, 1 },
8978         { "EpcmdTrimPostfix", 4, 1 },
8979         { "ERssIp4Pkt", 3, 1 },
8980         { "ERssIp6Pkt", 2, 1 },
8981         { "ERssTcpUdpPkt", 1, 1 },
8982         { "ERssFceFipPkt", 0, 1 },
8983         { NULL }
8984 };
8985
8986 static void
8987 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
8988 {
8989
8990         field_desc_show(sb, *p, tp_la0);
8991 }
8992
8993 static void
8994 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
8995 {
8996
8997         if (idx)
8998                 sbuf_printf(sb, "\n");
8999         field_desc_show(sb, p[0], tp_la0);
9000         if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
9001                 field_desc_show(sb, p[1], tp_la0);
9002 }
9003
9004 static void
9005 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
9006 {
9007
9008         if (idx)
9009                 sbuf_printf(sb, "\n");
9010         field_desc_show(sb, p[0], tp_la0);
9011         if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
9012                 field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
9013 }
9014
9015 static int
9016 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
9017 {
9018         struct adapter *sc = arg1;
9019         struct sbuf *sb;
9020         uint64_t *buf, *p;
9021         int rc;
9022         u_int i, inc;
9023         void (*show_func)(struct sbuf *, uint64_t *, int);
9024
9025         rc = sysctl_wire_old_buffer(req, 0);
9026         if (rc != 0)
9027                 return (rc);
9028
9029         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9030         if (sb == NULL)
9031                 return (ENOMEM);
9032
9033         buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
9034
9035         t4_tp_read_la(sc, buf, NULL);
9036         p = buf;
9037
9038         switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
9039         case 2:
9040                 inc = 2;
9041                 show_func = tp_la_show2;
9042                 break;
9043         case 3:
9044                 inc = 2;
9045                 show_func = tp_la_show3;
9046                 break;
9047         default:
9048                 inc = 1;
9049                 show_func = tp_la_show;
9050         }
9051
9052         for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
9053                 (*show_func)(sb, p, i);
9054
9055         rc = sbuf_finish(sb);
9056         sbuf_delete(sb);
9057         free(buf, M_CXGBE);
9058         return (rc);
9059 }
9060
9061 static int
9062 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
9063 {
9064         struct adapter *sc = arg1;
9065         struct sbuf *sb;
9066         int rc;
9067         u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
9068
9069         rc = sysctl_wire_old_buffer(req, 0);
9070         if (rc != 0)
9071                 return (rc);
9072
9073         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
9074         if (sb == NULL)
9075                 return (ENOMEM);
9076
9077         t4_get_chan_txrate(sc, nrate, orate);
9078
9079         if (sc->chip_params->nchan > 2) {
9080                 sbuf_printf(sb, "              channel 0   channel 1"
9081                     "   channel 2   channel 3\n");
9082                 sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
9083                     nrate[0], nrate[1], nrate[2], nrate[3]);
9084                 sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
9085                     orate[0], orate[1], orate[2], orate[3]);
9086         } else {
9087                 sbuf_printf(sb, "              channel 0   channel 1\n");
9088                 sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
9089                     nrate[0], nrate[1]);
9090                 sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
9091                     orate[0], orate[1]);
9092         }
9093
9094         rc = sbuf_finish(sb);
9095         sbuf_delete(sb);
9096
9097         return (rc);
9098 }
9099
9100 static int
9101 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
9102 {
9103         struct adapter *sc = arg1;
9104         struct sbuf *sb;
9105         uint32_t *buf, *p;
9106         int rc, i;
9107
9108         rc = sysctl_wire_old_buffer(req, 0);
9109         if (rc != 0)
9110                 return (rc);
9111
9112         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9113         if (sb == NULL)
9114                 return (ENOMEM);
9115
9116         buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
9117             M_ZERO | M_WAITOK);
9118
9119         t4_ulprx_read_la(sc, buf);
9120         p = buf;
9121
9122         sbuf_printf(sb, "      Pcmd        Type   Message"
9123             "                Data");
9124         for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
9125                 sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
9126                     p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
9127         }
9128
9129         rc = sbuf_finish(sb);
9130         sbuf_delete(sb);
9131         free(buf, M_CXGBE);
9132         return (rc);
9133 }
9134
9135 static int
9136 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
9137 {
9138         struct adapter *sc = arg1;
9139         struct sbuf *sb;
9140         int rc, v;
9141
9142         MPASS(chip_id(sc) >= CHELSIO_T5);
9143
9144         rc = sysctl_wire_old_buffer(req, 0);
9145         if (rc != 0)
9146                 return (rc);
9147
9148         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9149         if (sb == NULL)
9150                 return (ENOMEM);
9151
9152         v = t4_read_reg(sc, A_SGE_STAT_CFG);
9153         if (G_STATSOURCE_T5(v) == 7) {
9154                 int mode;
9155
9156                 mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
9157                 if (mode == 0) {
9158                         sbuf_printf(sb, "total %d, incomplete %d",
9159                             t4_read_reg(sc, A_SGE_STAT_TOTAL),
9160                             t4_read_reg(sc, A_SGE_STAT_MATCH));
9161                 } else if (mode == 1) {
9162                         sbuf_printf(sb, "total %d, data overflow %d",
9163                             t4_read_reg(sc, A_SGE_STAT_TOTAL),
9164                             t4_read_reg(sc, A_SGE_STAT_MATCH));
9165                 } else {
9166                         sbuf_printf(sb, "unknown mode %d", mode);
9167                 }
9168         }
9169         rc = sbuf_finish(sb);
9170         sbuf_delete(sb);
9171
9172         return (rc);
9173 }
9174
9175 static int
9176 sysctl_cpus(SYSCTL_HANDLER_ARGS)
9177 {
9178         struct adapter *sc = arg1;
9179         enum cpu_sets op = arg2;
9180         cpuset_t cpuset;
9181         struct sbuf *sb;
9182         int i, rc;
9183
9184         MPASS(op == LOCAL_CPUS || op == INTR_CPUS);
9185
9186         CPU_ZERO(&cpuset);
9187         rc = bus_get_cpus(sc->dev, op, sizeof(cpuset), &cpuset);
9188         if (rc != 0)
9189                 return (rc);
9190
9191         rc = sysctl_wire_old_buffer(req, 0);
9192         if (rc != 0)
9193                 return (rc);
9194
9195         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9196         if (sb == NULL)
9197                 return (ENOMEM);
9198
9199         CPU_FOREACH(i)
9200                 sbuf_printf(sb, "%d ", i);
9201         rc = sbuf_finish(sb);
9202         sbuf_delete(sb);
9203
9204         return (rc);
9205 }
9206
9207 #ifdef TCP_OFFLOAD
9208 static int
9209 sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS)
9210 {
9211         struct adapter *sc = arg1;
9212         int *old_ports, *new_ports;
9213         int i, new_count, rc;
9214
9215         if (req->newptr == NULL && req->oldptr == NULL)
9216                 return (SYSCTL_OUT(req, NULL, imax(sc->tt.num_tls_rx_ports, 1) *
9217                     sizeof(sc->tt.tls_rx_ports[0])));
9218
9219         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tlsrx");
9220         if (rc)
9221                 return (rc);
9222
9223         if (sc->tt.num_tls_rx_ports == 0) {
9224                 i = -1;
9225                 rc = SYSCTL_OUT(req, &i, sizeof(i));
9226         } else
9227                 rc = SYSCTL_OUT(req, sc->tt.tls_rx_ports,
9228                     sc->tt.num_tls_rx_ports * sizeof(sc->tt.tls_rx_ports[0]));
9229         if (rc == 0 && req->newptr != NULL) {
9230                 new_count = req->newlen / sizeof(new_ports[0]);
9231                 new_ports = malloc(new_count * sizeof(new_ports[0]), M_CXGBE,
9232                     M_WAITOK);
9233                 rc = SYSCTL_IN(req, new_ports, new_count *
9234                     sizeof(new_ports[0]));
9235                 if (rc)
9236                         goto err;
9237
9238                 /* Allow setting to a single '-1' to clear the list. */
9239                 if (new_count == 1 && new_ports[0] == -1) {
9240                         ADAPTER_LOCK(sc);
9241                         old_ports = sc->tt.tls_rx_ports;
9242                         sc->tt.tls_rx_ports = NULL;
9243                         sc->tt.num_tls_rx_ports = 0;
9244                         ADAPTER_UNLOCK(sc);
9245                         free(old_ports, M_CXGBE);
9246                 } else {
9247                         for (i = 0; i < new_count; i++) {
9248                                 if (new_ports[i] < 1 ||
9249                                     new_ports[i] > IPPORT_MAX) {
9250                                         rc = EINVAL;
9251                                         goto err;
9252                                 }
9253                         }
9254
9255                         ADAPTER_LOCK(sc);
9256                         old_ports = sc->tt.tls_rx_ports;
9257                         sc->tt.tls_rx_ports = new_ports;
9258                         sc->tt.num_tls_rx_ports = new_count;
9259                         ADAPTER_UNLOCK(sc);
9260                         free(old_ports, M_CXGBE);
9261                         new_ports = NULL;
9262                 }
9263         err:
9264                 free(new_ports, M_CXGBE);
9265         }
9266         end_synchronized_op(sc, 0);
9267         return (rc);
9268 }
9269
9270 static void
9271 unit_conv(char *buf, size_t len, u_int val, u_int factor)
9272 {
9273         u_int rem = val % factor;
9274
9275         if (rem == 0)
9276                 snprintf(buf, len, "%u", val / factor);
9277         else {
9278                 while (rem % 10 == 0)
9279                         rem /= 10;
9280                 snprintf(buf, len, "%u.%u", val / factor, rem);
9281         }
9282 }
9283
9284 static int
9285 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
9286 {
9287         struct adapter *sc = arg1;
9288         char buf[16];
9289         u_int res, re;
9290         u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
9291
9292         res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
9293         switch (arg2) {
9294         case 0:
9295                 /* timer_tick */
9296                 re = G_TIMERRESOLUTION(res);
9297                 break;
9298         case 1:
9299                 /* TCP timestamp tick */
9300                 re = G_TIMESTAMPRESOLUTION(res);
9301                 break;
9302         case 2:
9303                 /* DACK tick */
9304                 re = G_DELAYEDACKRESOLUTION(res);
9305                 break;
9306         default:
9307                 return (EDOOFUS);
9308         }
9309
9310         unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
9311
9312         return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
9313 }
9314
9315 static int
9316 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
9317 {
9318         struct adapter *sc = arg1;
9319         u_int res, dack_re, v;
9320         u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
9321
9322         res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
9323         dack_re = G_DELAYEDACKRESOLUTION(res);
9324         v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
9325
9326         return (sysctl_handle_int(oidp, &v, 0, req));
9327 }
9328
9329 static int
9330 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
9331 {
9332         struct adapter *sc = arg1;
9333         int reg = arg2;
9334         u_int tre;
9335         u_long tp_tick_us, v;
9336         u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
9337
9338         MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
9339             reg == A_TP_PERS_MIN  || reg == A_TP_PERS_MAX ||
9340             reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
9341             reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
9342
9343         tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
9344         tp_tick_us = (cclk_ps << tre) / 1000000;
9345
9346         if (reg == A_TP_INIT_SRTT)
9347                 v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
9348         else
9349                 v = tp_tick_us * t4_read_reg(sc, reg);
9350
9351         return (sysctl_handle_long(oidp, &v, 0, req));
9352 }
9353
9354 /*
9355  * All fields in TP_SHIFT_CNT are 4b and the starting location of the field is
9356  * passed to this function.
9357  */
9358 static int
9359 sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS)
9360 {
9361         struct adapter *sc = arg1;
9362         int idx = arg2;
9363         u_int v;
9364
9365         MPASS(idx >= 0 && idx <= 24);
9366
9367         v = (t4_read_reg(sc, A_TP_SHIFT_CNT) >> idx) & 0xf;
9368
9369         return (sysctl_handle_int(oidp, &v, 0, req));
9370 }
9371
9372 static int
9373 sysctl_tp_backoff(SYSCTL_HANDLER_ARGS)
9374 {
9375         struct adapter *sc = arg1;
9376         int idx = arg2;
9377         u_int shift, v, r;
9378
9379         MPASS(idx >= 0 && idx < 16);
9380
9381         r = A_TP_TCP_BACKOFF_REG0 + (idx & ~3);
9382         shift = (idx & 3) << 3;
9383         v = (t4_read_reg(sc, r) >> shift) & M_TIMERBACKOFFINDEX0;
9384
9385         return (sysctl_handle_int(oidp, &v, 0, req));
9386 }
9387
9388 static int
9389 sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS)
9390 {
9391         struct vi_info *vi = arg1;
9392         struct adapter *sc = vi->pi->adapter;
9393         int idx, rc, i;
9394         struct sge_ofld_rxq *ofld_rxq;
9395         uint8_t v;
9396
9397         idx = vi->ofld_tmr_idx;
9398
9399         rc = sysctl_handle_int(oidp, &idx, 0, req);
9400         if (rc != 0 || req->newptr == NULL)
9401                 return (rc);
9402
9403         if (idx < 0 || idx >= SGE_NTIMERS)
9404                 return (EINVAL);
9405
9406         rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
9407             "t4otmr");
9408         if (rc)
9409                 return (rc);
9410
9411         v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->ofld_pktc_idx != -1);
9412         for_each_ofld_rxq(vi, i, ofld_rxq) {
9413 #ifdef atomic_store_rel_8
9414                 atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
9415 #else
9416                 ofld_rxq->iq.intr_params = v;
9417 #endif
9418         }
9419         vi->ofld_tmr_idx = idx;
9420
9421         end_synchronized_op(sc, LOCK_HELD);
9422         return (0);
9423 }
9424
9425 static int
9426 sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS)
9427 {
9428         struct vi_info *vi = arg1;
9429         struct adapter *sc = vi->pi->adapter;
9430         int idx, rc;
9431
9432         idx = vi->ofld_pktc_idx;
9433
9434         rc = sysctl_handle_int(oidp, &idx, 0, req);
9435         if (rc != 0 || req->newptr == NULL)
9436                 return (rc);
9437
9438         if (idx < -1 || idx >= SGE_NCOUNTERS)
9439                 return (EINVAL);
9440
9441         rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
9442             "t4opktc");
9443         if (rc)
9444                 return (rc);
9445
9446         if (vi->flags & VI_INIT_DONE)
9447                 rc = EBUSY; /* cannot be changed once the queues are created */
9448         else
9449                 vi->ofld_pktc_idx = idx;
9450
9451         end_synchronized_op(sc, LOCK_HELD);
9452         return (rc);
9453 }
9454 #endif
9455
9456 static int
9457 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
9458 {
9459         int rc;
9460
9461         if (cntxt->cid > M_CTXTQID)
9462                 return (EINVAL);
9463
9464         if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
9465             cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
9466                 return (EINVAL);
9467
9468         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
9469         if (rc)
9470                 return (rc);
9471
9472         if (sc->flags & FW_OK) {
9473                 rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
9474                     &cntxt->data[0]);
9475                 if (rc == 0)
9476                         goto done;
9477         }
9478
9479         /*
9480          * Read via firmware failed or wasn't even attempted.  Read directly via
9481          * the backdoor.
9482          */
9483         rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
9484 done:
9485         end_synchronized_op(sc, 0);
9486         return (rc);
9487 }
9488
9489 static int
9490 load_fw(struct adapter *sc, struct t4_data *fw)
9491 {
9492         int rc;
9493         uint8_t *fw_data;
9494
9495         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
9496         if (rc)
9497                 return (rc);
9498
9499         /*
9500          * The firmware, with the sole exception of the memory parity error
9501          * handler, runs from memory and not flash.  It is almost always safe to
9502          * install a new firmware on a running system.  Just set bit 1 in
9503          * hw.cxgbe.dflags or dev.<nexus>.<n>.dflags first.
9504          */
9505         if (sc->flags & FULL_INIT_DONE &&
9506             (sc->debug_flags & DF_LOAD_FW_ANYTIME) == 0) {
9507                 rc = EBUSY;
9508                 goto done;
9509         }
9510
9511         fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
9512         if (fw_data == NULL) {
9513                 rc = ENOMEM;
9514                 goto done;
9515         }
9516
9517         rc = copyin(fw->data, fw_data, fw->len);
9518         if (rc == 0)
9519                 rc = -t4_load_fw(sc, fw_data, fw->len);
9520
9521         free(fw_data, M_CXGBE);
9522 done:
9523         end_synchronized_op(sc, 0);
9524         return (rc);
9525 }
9526
9527 static int
9528 load_cfg(struct adapter *sc, struct t4_data *cfg)
9529 {
9530         int rc;
9531         uint8_t *cfg_data = NULL;
9532
9533         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9534         if (rc)
9535                 return (rc);
9536
9537         if (cfg->len == 0) {
9538                 /* clear */
9539                 rc = -t4_load_cfg(sc, NULL, 0);
9540                 goto done;
9541         }
9542
9543         cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
9544         if (cfg_data == NULL) {
9545                 rc = ENOMEM;
9546                 goto done;
9547         }
9548
9549         rc = copyin(cfg->data, cfg_data, cfg->len);
9550         if (rc == 0)
9551                 rc = -t4_load_cfg(sc, cfg_data, cfg->len);
9552
9553         free(cfg_data, M_CXGBE);
9554 done:
9555         end_synchronized_op(sc, 0);
9556         return (rc);
9557 }
9558
9559 static int
9560 load_boot(struct adapter *sc, struct t4_bootrom *br)
9561 {
9562         int rc;
9563         uint8_t *br_data = NULL;
9564         u_int offset;
9565
9566         if (br->len > 1024 * 1024)
9567                 return (EFBIG);
9568
9569         if (br->pf_offset == 0) {
9570                 /* pfidx */
9571                 if (br->pfidx_addr > 7)
9572                         return (EINVAL);
9573                 offset = G_OFFSET(t4_read_reg(sc, PF_REG(br->pfidx_addr,
9574                     A_PCIE_PF_EXPROM_OFST)));
9575         } else if (br->pf_offset == 1) {
9576                 /* offset */
9577                 offset = G_OFFSET(br->pfidx_addr);
9578         } else {
9579                 return (EINVAL);
9580         }
9581
9582         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldbr");
9583         if (rc)
9584                 return (rc);
9585
9586         if (br->len == 0) {
9587                 /* clear */
9588                 rc = -t4_load_boot(sc, NULL, offset, 0);
9589                 goto done;
9590         }
9591
9592         br_data = malloc(br->len, M_CXGBE, M_WAITOK);
9593         if (br_data == NULL) {
9594                 rc = ENOMEM;
9595                 goto done;
9596         }
9597
9598         rc = copyin(br->data, br_data, br->len);
9599         if (rc == 0)
9600                 rc = -t4_load_boot(sc, br_data, offset, br->len);
9601
9602         free(br_data, M_CXGBE);
9603 done:
9604         end_synchronized_op(sc, 0);
9605         return (rc);
9606 }
9607
9608 static int
9609 load_bootcfg(struct adapter *sc, struct t4_data *bc)
9610 {
9611         int rc;
9612         uint8_t *bc_data = NULL;
9613
9614         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9615         if (rc)
9616                 return (rc);
9617
9618         if (bc->len == 0) {
9619                 /* clear */
9620                 rc = -t4_load_bootcfg(sc, NULL, 0);
9621                 goto done;
9622         }
9623
9624         bc_data = malloc(bc->len, M_CXGBE, M_WAITOK);
9625         if (bc_data == NULL) {
9626                 rc = ENOMEM;
9627                 goto done;
9628         }
9629
9630         rc = copyin(bc->data, bc_data, bc->len);
9631         if (rc == 0)
9632                 rc = -t4_load_bootcfg(sc, bc_data, bc->len);
9633
9634         free(bc_data, M_CXGBE);
9635 done:
9636         end_synchronized_op(sc, 0);
9637         return (rc);
9638 }
9639
9640 static int
9641 cudbg_dump(struct adapter *sc, struct t4_cudbg_dump *dump)
9642 {
9643         int rc;
9644         struct cudbg_init *cudbg;
9645         void *handle, *buf;
9646
9647         /* buf is large, don't block if no memory is available */
9648         buf = malloc(dump->len, M_CXGBE, M_NOWAIT | M_ZERO);
9649         if (buf == NULL)
9650                 return (ENOMEM);
9651
9652         handle = cudbg_alloc_handle();
9653         if (handle == NULL) {
9654                 rc = ENOMEM;
9655                 goto done;
9656         }
9657
9658         cudbg = cudbg_get_init(handle);
9659         cudbg->adap = sc;
9660         cudbg->print = (cudbg_print_cb)printf;
9661
9662 #ifndef notyet
9663         device_printf(sc->dev, "%s: wr_flash %u, len %u, data %p.\n",
9664             __func__, dump->wr_flash, dump->len, dump->data);
9665 #endif
9666
9667         if (dump->wr_flash)
9668                 cudbg->use_flash = 1;
9669         MPASS(sizeof(cudbg->dbg_bitmap) == sizeof(dump->bitmap));
9670         memcpy(cudbg->dbg_bitmap, dump->bitmap, sizeof(cudbg->dbg_bitmap));
9671
9672         rc = cudbg_collect(handle, buf, &dump->len);
9673         if (rc != 0)
9674                 goto done;
9675
9676         rc = copyout(buf, dump->data, dump->len);
9677 done:
9678         cudbg_free_handle(handle);
9679         free(buf, M_CXGBE);
9680         return (rc);
9681 }
9682
9683 static void
9684 free_offload_policy(struct t4_offload_policy *op)
9685 {
9686         struct offload_rule *r;
9687         int i;
9688
9689         if (op == NULL)
9690                 return;
9691
9692         r = &op->rule[0];
9693         for (i = 0; i < op->nrules; i++, r++) {
9694                 free(r->bpf_prog.bf_insns, M_CXGBE);
9695         }
9696         free(op->rule, M_CXGBE);
9697         free(op, M_CXGBE);
9698 }
9699
9700 static int
9701 set_offload_policy(struct adapter *sc, struct t4_offload_policy *uop)
9702 {
9703         int i, rc, len;
9704         struct t4_offload_policy *op, *old;
9705         struct bpf_program *bf;
9706         const struct offload_settings *s;
9707         struct offload_rule *r;
9708         void *u;
9709
9710         if (!is_offload(sc))
9711                 return (ENODEV);
9712
9713         if (uop->nrules == 0) {
9714                 /* Delete installed policies. */
9715                 op = NULL;
9716                 goto set_policy;
9717         } if (uop->nrules > 256) { /* arbitrary */
9718                 return (E2BIG);
9719         }
9720
9721         /* Copy userspace offload policy to kernel */
9722         op = malloc(sizeof(*op), M_CXGBE, M_ZERO | M_WAITOK);
9723         op->nrules = uop->nrules;
9724         len = op->nrules * sizeof(struct offload_rule);
9725         op->rule = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9726         rc = copyin(uop->rule, op->rule, len);
9727         if (rc) {
9728                 free(op->rule, M_CXGBE);
9729                 free(op, M_CXGBE);
9730                 return (rc);
9731         }
9732
9733         r = &op->rule[0];
9734         for (i = 0; i < op->nrules; i++, r++) {
9735
9736                 /* Validate open_type */
9737                 if (r->open_type != OPEN_TYPE_LISTEN &&
9738                     r->open_type != OPEN_TYPE_ACTIVE &&
9739                     r->open_type != OPEN_TYPE_PASSIVE &&
9740                     r->open_type != OPEN_TYPE_DONTCARE) {
9741 error:
9742                         /*
9743                          * Rules 0 to i have malloc'd filters that need to be
9744                          * freed.  Rules i+1 to nrules have userspace pointers
9745                          * and should be left alone.
9746                          */
9747                         op->nrules = i;
9748                         free_offload_policy(op);
9749                         return (rc);
9750                 }
9751
9752                 /* Validate settings */
9753                 s = &r->settings;
9754                 if ((s->offload != 0 && s->offload != 1) ||
9755                     s->cong_algo < -1 || s->cong_algo > CONG_ALG_HIGHSPEED ||
9756                     s->sched_class < -1 ||
9757                     s->sched_class >= sc->chip_params->nsched_cls) {
9758                         rc = EINVAL;
9759                         goto error;
9760                 }
9761
9762                 bf = &r->bpf_prog;
9763                 u = bf->bf_insns;       /* userspace ptr */
9764                 bf->bf_insns = NULL;
9765                 if (bf->bf_len == 0) {
9766                         /* legal, matches everything */
9767                         continue;
9768                 }
9769                 len = bf->bf_len * sizeof(*bf->bf_insns);
9770                 bf->bf_insns = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9771                 rc = copyin(u, bf->bf_insns, len);
9772                 if (rc != 0)
9773                         goto error;
9774
9775                 if (!bpf_validate(bf->bf_insns, bf->bf_len)) {
9776                         rc = EINVAL;
9777                         goto error;
9778                 }
9779         }
9780 set_policy:
9781         rw_wlock(&sc->policy_lock);
9782         old = sc->policy;
9783         sc->policy = op;
9784         rw_wunlock(&sc->policy_lock);
9785         free_offload_policy(old);
9786
9787         return (0);
9788 }
9789
9790 #define MAX_READ_BUF_SIZE (128 * 1024)
9791 static int
9792 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
9793 {
9794         uint32_t addr, remaining, n;
9795         uint32_t *buf;
9796         int rc;
9797         uint8_t *dst;
9798
9799         rc = validate_mem_range(sc, mr->addr, mr->len);
9800         if (rc != 0)
9801                 return (rc);
9802
9803         buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
9804         addr = mr->addr;
9805         remaining = mr->len;
9806         dst = (void *)mr->data;
9807
9808         while (remaining) {
9809                 n = min(remaining, MAX_READ_BUF_SIZE);
9810                 read_via_memwin(sc, 2, addr, buf, n);
9811
9812                 rc = copyout(buf, dst, n);
9813                 if (rc != 0)
9814                         break;
9815
9816                 dst += n;
9817                 remaining -= n;
9818                 addr += n;
9819         }
9820
9821         free(buf, M_CXGBE);
9822         return (rc);
9823 }
9824 #undef MAX_READ_BUF_SIZE
9825
9826 static int
9827 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
9828 {
9829         int rc;
9830
9831         if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
9832                 return (EINVAL);
9833
9834         if (i2cd->len > sizeof(i2cd->data))
9835                 return (EFBIG);
9836
9837         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
9838         if (rc)
9839                 return (rc);
9840         rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
9841             i2cd->offset, i2cd->len, &i2cd->data[0]);
9842         end_synchronized_op(sc, 0);
9843
9844         return (rc);
9845 }
9846
9847 int
9848 t4_os_find_pci_capability(struct adapter *sc, int cap)
9849 {
9850         int i;
9851
9852         return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9853 }
9854
9855 int
9856 t4_os_pci_save_state(struct adapter *sc)
9857 {
9858         device_t dev;
9859         struct pci_devinfo *dinfo;
9860
9861         dev = sc->dev;
9862         dinfo = device_get_ivars(dev);
9863
9864         pci_cfg_save(dev, dinfo, 0);
9865         return (0);
9866 }
9867
9868 int
9869 t4_os_pci_restore_state(struct adapter *sc)
9870 {
9871         device_t dev;
9872         struct pci_devinfo *dinfo;
9873
9874         dev = sc->dev;
9875         dinfo = device_get_ivars(dev);
9876
9877         pci_cfg_restore(dev, dinfo);
9878         return (0);
9879 }
9880
9881 void
9882 t4_os_portmod_changed(struct port_info *pi)
9883 {
9884         struct adapter *sc = pi->adapter;
9885         struct vi_info *vi;
9886         struct ifnet *ifp;
9887         static const char *mod_str[] = {
9888                 NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9889         };
9890
9891         KASSERT((pi->flags & FIXED_IFMEDIA) == 0,
9892             ("%s: port_type %u", __func__, pi->port_type));
9893
9894         vi = &pi->vi[0];
9895         if (begin_synchronized_op(sc, vi, HOLD_LOCK, "t4mod") == 0) {
9896                 PORT_LOCK(pi);
9897                 build_medialist(pi);
9898                 if (pi->mod_type != FW_PORT_MOD_TYPE_NONE) {
9899                         fixup_link_config(pi);
9900                         apply_link_config(pi);
9901                 }
9902                 PORT_UNLOCK(pi);
9903                 end_synchronized_op(sc, LOCK_HELD);
9904         }
9905
9906         ifp = vi->ifp;
9907         if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
9908                 if_printf(ifp, "transceiver unplugged.\n");
9909         else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
9910                 if_printf(ifp, "unknown transceiver inserted.\n");
9911         else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
9912                 if_printf(ifp, "unsupported transceiver inserted.\n");
9913         else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
9914                 if_printf(ifp, "%dGbps %s transceiver inserted.\n",
9915                     port_top_speed(pi), mod_str[pi->mod_type]);
9916         } else {
9917                 if_printf(ifp, "transceiver (type %d) inserted.\n",
9918                     pi->mod_type);
9919         }
9920 }
9921
9922 void
9923 t4_os_link_changed(struct port_info *pi)
9924 {
9925         struct vi_info *vi;
9926         struct ifnet *ifp;
9927         struct link_config *lc;
9928         int v;
9929
9930         PORT_LOCK_ASSERT_OWNED(pi);
9931
9932         for_each_vi(pi, v, vi) {
9933                 ifp = vi->ifp;
9934                 if (ifp == NULL)
9935                         continue;
9936
9937                 lc = &pi->link_cfg;
9938                 if (lc->link_ok) {
9939                         ifp->if_baudrate = IF_Mbps(lc->speed);
9940                         if_link_state_change(ifp, LINK_STATE_UP);
9941                 } else {
9942                         if_link_state_change(ifp, LINK_STATE_DOWN);
9943                 }
9944         }
9945 }
9946
9947 void
9948 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
9949 {
9950         struct adapter *sc;
9951
9952         sx_slock(&t4_list_lock);
9953         SLIST_FOREACH(sc, &t4_list, link) {
9954                 /*
9955                  * func should not make any assumptions about what state sc is
9956                  * in - the only guarantee is that sc->sc_lock is a valid lock.
9957                  */
9958                 func(sc, arg);
9959         }
9960         sx_sunlock(&t4_list_lock);
9961 }
9962
9963 static int
9964 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9965     struct thread *td)
9966 {
9967         int rc;
9968         struct adapter *sc = dev->si_drv1;
9969
9970         rc = priv_check(td, PRIV_DRIVER);
9971         if (rc != 0)
9972                 return (rc);
9973
9974         switch (cmd) {
9975         case CHELSIO_T4_GETREG: {
9976                 struct t4_reg *edata = (struct t4_reg *)data;
9977
9978                 if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9979                         return (EFAULT);
9980
9981                 if (edata->size == 4)
9982                         edata->val = t4_read_reg(sc, edata->addr);
9983                 else if (edata->size == 8)
9984                         edata->val = t4_read_reg64(sc, edata->addr);
9985                 else
9986                         return (EINVAL);
9987
9988                 break;
9989         }
9990         case CHELSIO_T4_SETREG: {
9991                 struct t4_reg *edata = (struct t4_reg *)data;
9992
9993                 if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9994                         return (EFAULT);
9995
9996                 if (edata->size == 4) {
9997                         if (edata->val & 0xffffffff00000000)
9998                                 return (EINVAL);
9999                         t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
10000                 } else if (edata->size == 8)
10001                         t4_write_reg64(sc, edata->addr, edata->val);
10002                 else
10003                         return (EINVAL);
10004                 break;
10005         }
10006         case CHELSIO_T4_REGDUMP: {
10007                 struct t4_regdump *regs = (struct t4_regdump *)data;
10008                 int reglen = t4_get_regs_len(sc);
10009                 uint8_t *buf;
10010
10011                 if (regs->len < reglen) {
10012                         regs->len = reglen; /* hint to the caller */
10013                         return (ENOBUFS);
10014                 }
10015
10016                 regs->len = reglen;
10017                 buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
10018                 get_regs(sc, regs, buf);
10019                 rc = copyout(buf, regs->data, reglen);
10020                 free(buf, M_CXGBE);
10021                 break;
10022         }
10023         case CHELSIO_T4_GET_FILTER_MODE:
10024                 rc = get_filter_mode(sc, (uint32_t *)data);
10025                 break;
10026         case CHELSIO_T4_SET_FILTER_MODE:
10027                 rc = set_filter_mode(sc, *(uint32_t *)data);
10028                 break;
10029         case CHELSIO_T4_GET_FILTER:
10030                 rc = get_filter(sc, (struct t4_filter *)data);
10031                 break;
10032         case CHELSIO_T4_SET_FILTER:
10033                 rc = set_filter(sc, (struct t4_filter *)data);
10034                 break;
10035         case CHELSIO_T4_DEL_FILTER:
10036                 rc = del_filter(sc, (struct t4_filter *)data);
10037                 break;
10038         case CHELSIO_T4_GET_SGE_CONTEXT:
10039                 rc = get_sge_context(sc, (struct t4_sge_context *)data);
10040                 break;
10041         case CHELSIO_T4_LOAD_FW:
10042                 rc = load_fw(sc, (struct t4_data *)data);
10043                 break;
10044         case CHELSIO_T4_GET_MEM:
10045                 rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
10046                 break;
10047         case CHELSIO_T4_GET_I2C:
10048                 rc = read_i2c(sc, (struct t4_i2c_data *)data);
10049                 break;
10050         case CHELSIO_T4_CLEAR_STATS: {
10051                 int i, v, bg_map;
10052                 u_int port_id = *(uint32_t *)data;
10053                 struct port_info *pi;
10054                 struct vi_info *vi;
10055
10056                 if (port_id >= sc->params.nports)
10057                         return (EINVAL);
10058                 pi = sc->port[port_id];
10059                 if (pi == NULL)
10060                         return (EIO);
10061
10062                 /* MAC stats */
10063                 t4_clr_port_stats(sc, pi->tx_chan);
10064                 pi->tx_parse_error = 0;
10065                 pi->tnl_cong_drops = 0;
10066                 mtx_lock(&sc->reg_lock);
10067                 for_each_vi(pi, v, vi) {
10068                         if (vi->flags & VI_INIT_DONE)
10069                                 t4_clr_vi_stats(sc, vi->vin);
10070                 }
10071                 bg_map = pi->mps_bg_map;
10072                 v = 0;  /* reuse */
10073                 while (bg_map) {
10074                         i = ffs(bg_map) - 1;
10075                         t4_write_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
10076                             1, A_TP_MIB_TNL_CNG_DROP_0 + i);
10077                         bg_map &= ~(1 << i);
10078                 }
10079                 mtx_unlock(&sc->reg_lock);
10080
10081                 /*
10082                  * Since this command accepts a port, clear stats for
10083                  * all VIs on this port.
10084                  */
10085                 for_each_vi(pi, v, vi) {
10086                         if (vi->flags & VI_INIT_DONE) {
10087                                 struct sge_rxq *rxq;
10088                                 struct sge_txq *txq;
10089                                 struct sge_wrq *wrq;
10090
10091                                 for_each_rxq(vi, i, rxq) {
10092 #if defined(INET) || defined(INET6)
10093                                         rxq->lro.lro_queued = 0;
10094                                         rxq->lro.lro_flushed = 0;
10095 #endif
10096                                         rxq->rxcsum = 0;
10097                                         rxq->vlan_extraction = 0;
10098                                 }
10099
10100                                 for_each_txq(vi, i, txq) {
10101                                         txq->txcsum = 0;
10102                                         txq->tso_wrs = 0;
10103                                         txq->vlan_insertion = 0;
10104                                         txq->imm_wrs = 0;
10105                                         txq->sgl_wrs = 0;
10106                                         txq->txpkt_wrs = 0;
10107                                         txq->txpkts0_wrs = 0;
10108                                         txq->txpkts1_wrs = 0;
10109                                         txq->txpkts0_pkts = 0;
10110                                         txq->txpkts1_pkts = 0;
10111                                         txq->raw_wrs = 0;
10112                                         mp_ring_reset_stats(txq->r);
10113                                 }
10114
10115 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
10116                                 /* nothing to clear for each ofld_rxq */
10117
10118                                 for_each_ofld_txq(vi, i, wrq) {
10119                                         wrq->tx_wrs_direct = 0;
10120                                         wrq->tx_wrs_copied = 0;
10121                                 }
10122 #endif
10123
10124                                 if (IS_MAIN_VI(vi)) {
10125                                         wrq = &sc->sge.ctrlq[pi->port_id];
10126                                         wrq->tx_wrs_direct = 0;
10127                                         wrq->tx_wrs_copied = 0;
10128                                 }
10129                         }
10130                 }
10131                 break;
10132         }
10133         case CHELSIO_T4_SCHED_CLASS:
10134                 rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
10135                 break;
10136         case CHELSIO_T4_SCHED_QUEUE:
10137                 rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
10138                 break;
10139         case CHELSIO_T4_GET_TRACER:
10140                 rc = t4_get_tracer(sc, (struct t4_tracer *)data);
10141                 break;
10142         case CHELSIO_T4_SET_TRACER:
10143                 rc = t4_set_tracer(sc, (struct t4_tracer *)data);
10144                 break;
10145         case CHELSIO_T4_LOAD_CFG:
10146                 rc = load_cfg(sc, (struct t4_data *)data);
10147                 break;
10148         case CHELSIO_T4_LOAD_BOOT:
10149                 rc = load_boot(sc, (struct t4_bootrom *)data);
10150                 break;
10151         case CHELSIO_T4_LOAD_BOOTCFG:
10152                 rc = load_bootcfg(sc, (struct t4_data *)data);
10153                 break;
10154         case CHELSIO_T4_CUDBG_DUMP:
10155                 rc = cudbg_dump(sc, (struct t4_cudbg_dump *)data);
10156                 break;
10157         case CHELSIO_T4_SET_OFLD_POLICY:
10158                 rc = set_offload_policy(sc, (struct t4_offload_policy *)data);
10159                 break;
10160         default:
10161                 rc = ENOTTY;
10162         }
10163
10164         return (rc);
10165 }
10166
10167 #ifdef TCP_OFFLOAD
10168 static int
10169 toe_capability(struct vi_info *vi, int enable)
10170 {
10171         int rc;
10172         struct port_info *pi = vi->pi;
10173         struct adapter *sc = pi->adapter;
10174
10175         ASSERT_SYNCHRONIZED_OP(sc);
10176
10177         if (!is_offload(sc))
10178                 return (ENODEV);
10179
10180         if (enable) {
10181                 if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
10182                         /* TOE is already enabled. */
10183                         return (0);
10184                 }
10185
10186                 /*
10187                  * We need the port's queues around so that we're able to send
10188                  * and receive CPLs to/from the TOE even if the ifnet for this
10189                  * port has never been UP'd administratively.
10190                  */
10191                 if (!(vi->flags & VI_INIT_DONE)) {
10192                         rc = vi_full_init(vi);
10193                         if (rc)
10194                                 return (rc);
10195                 }
10196                 if (!(pi->vi[0].flags & VI_INIT_DONE)) {
10197                         rc = vi_full_init(&pi->vi[0]);
10198                         if (rc)
10199                                 return (rc);
10200                 }
10201
10202                 if (isset(&sc->offload_map, pi->port_id)) {
10203                         /* TOE is enabled on another VI of this port. */
10204                         pi->uld_vis++;
10205                         return (0);
10206                 }
10207
10208                 if (!uld_active(sc, ULD_TOM)) {
10209                         rc = t4_activate_uld(sc, ULD_TOM);
10210                         if (rc == EAGAIN) {
10211                                 log(LOG_WARNING,
10212                                     "You must kldload t4_tom.ko before trying "
10213                                     "to enable TOE on a cxgbe interface.\n");
10214                         }
10215                         if (rc != 0)
10216                                 return (rc);
10217                         KASSERT(sc->tom_softc != NULL,
10218                             ("%s: TOM activated but softc NULL", __func__));
10219                         KASSERT(uld_active(sc, ULD_TOM),
10220                             ("%s: TOM activated but flag not set", __func__));
10221                 }
10222
10223                 /* Activate iWARP and iSCSI too, if the modules are loaded. */
10224                 if (!uld_active(sc, ULD_IWARP))
10225                         (void) t4_activate_uld(sc, ULD_IWARP);
10226                 if (!uld_active(sc, ULD_ISCSI))
10227                         (void) t4_activate_uld(sc, ULD_ISCSI);
10228
10229                 pi->uld_vis++;
10230                 setbit(&sc->offload_map, pi->port_id);
10231         } else {
10232                 pi->uld_vis--;
10233
10234                 if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
10235                         return (0);
10236
10237                 KASSERT(uld_active(sc, ULD_TOM),
10238                     ("%s: TOM never initialized?", __func__));
10239                 clrbit(&sc->offload_map, pi->port_id);
10240         }
10241
10242         return (0);
10243 }
10244
10245 /*
10246  * Add an upper layer driver to the global list.
10247  */
10248 int
10249 t4_register_uld(struct uld_info *ui)
10250 {
10251         int rc = 0;
10252         struct uld_info *u;
10253
10254         sx_xlock(&t4_uld_list_lock);
10255         SLIST_FOREACH(u, &t4_uld_list, link) {
10256             if (u->uld_id == ui->uld_id) {
10257                     rc = EEXIST;
10258                     goto done;
10259             }
10260         }
10261
10262         SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
10263         ui->refcount = 0;
10264 done:
10265         sx_xunlock(&t4_uld_list_lock);
10266         return (rc);
10267 }
10268
10269 int
10270 t4_unregister_uld(struct uld_info *ui)
10271 {
10272         int rc = EINVAL;
10273         struct uld_info *u;
10274
10275         sx_xlock(&t4_uld_list_lock);
10276
10277         SLIST_FOREACH(u, &t4_uld_list, link) {
10278             if (u == ui) {
10279                     if (ui->refcount > 0) {
10280                             rc = EBUSY;
10281                             goto done;
10282                     }
10283
10284                     SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
10285                     rc = 0;
10286                     goto done;
10287             }
10288         }
10289 done:
10290         sx_xunlock(&t4_uld_list_lock);
10291         return (rc);
10292 }
10293
10294 int
10295 t4_activate_uld(struct adapter *sc, int id)
10296 {
10297         int rc;
10298         struct uld_info *ui;
10299
10300         ASSERT_SYNCHRONIZED_OP(sc);
10301
10302         if (id < 0 || id > ULD_MAX)
10303                 return (EINVAL);
10304         rc = EAGAIN;    /* kldoad the module with this ULD and try again. */
10305
10306         sx_slock(&t4_uld_list_lock);
10307
10308         SLIST_FOREACH(ui, &t4_uld_list, link) {
10309                 if (ui->uld_id == id) {
10310                         if (!(sc->flags & FULL_INIT_DONE)) {
10311                                 rc = adapter_full_init(sc);
10312                                 if (rc != 0)
10313                                         break;
10314                         }
10315
10316                         rc = ui->activate(sc);
10317                         if (rc == 0) {
10318                                 setbit(&sc->active_ulds, id);
10319                                 ui->refcount++;
10320                         }
10321                         break;
10322                 }
10323         }
10324
10325         sx_sunlock(&t4_uld_list_lock);
10326
10327         return (rc);
10328 }
10329
10330 int
10331 t4_deactivate_uld(struct adapter *sc, int id)
10332 {
10333         int rc;
10334         struct uld_info *ui;
10335
10336         ASSERT_SYNCHRONIZED_OP(sc);
10337
10338         if (id < 0 || id > ULD_MAX)
10339                 return (EINVAL);
10340         rc = ENXIO;
10341
10342         sx_slock(&t4_uld_list_lock);
10343
10344         SLIST_FOREACH(ui, &t4_uld_list, link) {
10345                 if (ui->uld_id == id) {
10346                         rc = ui->deactivate(sc);
10347                         if (rc == 0) {
10348                                 clrbit(&sc->active_ulds, id);
10349                                 ui->refcount--;
10350                         }
10351                         break;
10352                 }
10353         }
10354
10355         sx_sunlock(&t4_uld_list_lock);
10356
10357         return (rc);
10358 }
10359
10360 int
10361 uld_active(struct adapter *sc, int uld_id)
10362 {
10363
10364         MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
10365
10366         return (isset(&sc->active_ulds, uld_id));
10367 }
10368 #endif
10369
10370 /*
10371  * t  = ptr to tunable.
10372  * nc = number of CPUs.
10373  * c  = compiled in default for that tunable.
10374  */
10375 static void
10376 calculate_nqueues(int *t, int nc, const int c)
10377 {
10378         int nq;
10379
10380         if (*t > 0)
10381                 return;
10382         nq = *t < 0 ? -*t : c;
10383         *t = min(nc, nq);
10384 }
10385
10386 /*
10387  * Come up with reasonable defaults for some of the tunables, provided they're
10388  * not set by the user (in which case we'll use the values as is).
10389  */
10390 static void
10391 tweak_tunables(void)
10392 {
10393         int nc = mp_ncpus;      /* our snapshot of the number of CPUs */
10394
10395         if (t4_ntxq < 1) {
10396 #ifdef RSS
10397                 t4_ntxq = rss_getnumbuckets();
10398 #else
10399                 calculate_nqueues(&t4_ntxq, nc, NTXQ);
10400 #endif
10401         }
10402
10403         calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
10404
10405         if (t4_nrxq < 1) {
10406 #ifdef RSS
10407                 t4_nrxq = rss_getnumbuckets();
10408 #else
10409                 calculate_nqueues(&t4_nrxq, nc, NRXQ);
10410 #endif
10411         }
10412
10413         calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
10414
10415 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
10416         calculate_nqueues(&t4_nofldtxq, nc, NOFLDTXQ);
10417         calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
10418 #endif
10419 #ifdef TCP_OFFLOAD
10420         calculate_nqueues(&t4_nofldrxq, nc, NOFLDRXQ);
10421         calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
10422
10423         if (t4_toecaps_allowed == -1)
10424                 t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
10425
10426         if (t4_rdmacaps_allowed == -1) {
10427                 t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
10428                     FW_CAPS_CONFIG_RDMA_RDMAC;
10429         }
10430
10431         if (t4_iscsicaps_allowed == -1) {
10432                 t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
10433                     FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
10434                     FW_CAPS_CONFIG_ISCSI_T10DIF;
10435         }
10436
10437         if (t4_tmr_idx_ofld < 0 || t4_tmr_idx_ofld >= SGE_NTIMERS)
10438                 t4_tmr_idx_ofld = TMR_IDX_OFLD;
10439
10440         if (t4_pktc_idx_ofld < -1 || t4_pktc_idx_ofld >= SGE_NCOUNTERS)
10441                 t4_pktc_idx_ofld = PKTC_IDX_OFLD;
10442 #else
10443         if (t4_toecaps_allowed == -1)
10444                 t4_toecaps_allowed = 0;
10445
10446         if (t4_rdmacaps_allowed == -1)
10447                 t4_rdmacaps_allowed = 0;
10448
10449         if (t4_iscsicaps_allowed == -1)
10450                 t4_iscsicaps_allowed = 0;
10451 #endif
10452
10453 #ifdef DEV_NETMAP
10454         calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
10455         calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
10456 #endif
10457
10458         if (t4_tmr_idx < 0 || t4_tmr_idx >= SGE_NTIMERS)
10459                 t4_tmr_idx = TMR_IDX;
10460
10461         if (t4_pktc_idx < -1 || t4_pktc_idx >= SGE_NCOUNTERS)
10462                 t4_pktc_idx = PKTC_IDX;
10463
10464         if (t4_qsize_txq < 128)
10465                 t4_qsize_txq = 128;
10466
10467         if (t4_qsize_rxq < 128)
10468                 t4_qsize_rxq = 128;
10469         while (t4_qsize_rxq & 7)
10470                 t4_qsize_rxq++;
10471
10472         t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
10473
10474         /*
10475          * Number of VIs to create per-port.  The first VI is the "main" regular
10476          * VI for the port.  The rest are additional virtual interfaces on the
10477          * same physical port.  Note that the main VI does not have native
10478          * netmap support but the extra VIs do.
10479          *
10480          * Limit the number of VIs per port to the number of available
10481          * MAC addresses per port.
10482          */
10483         if (t4_num_vis < 1)
10484                 t4_num_vis = 1;
10485         if (t4_num_vis > nitems(vi_mac_funcs)) {
10486                 t4_num_vis = nitems(vi_mac_funcs);
10487                 printf("cxgbe: number of VIs limited to %d\n", t4_num_vis);
10488         }
10489
10490         if (pcie_relaxed_ordering < 0 || pcie_relaxed_ordering > 2) {
10491                 pcie_relaxed_ordering = 1;
10492 #if defined(__i386__) || defined(__amd64__)
10493                 if (cpu_vendor_id == CPU_VENDOR_INTEL)
10494                         pcie_relaxed_ordering = 0;
10495 #endif
10496         }
10497 }
10498
10499 #ifdef DDB
10500 static void
10501 t4_dump_tcb(struct adapter *sc, int tid)
10502 {
10503         uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
10504
10505         reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
10506         save = t4_read_reg(sc, reg);
10507         base = sc->memwin[2].mw_base;
10508
10509         /* Dump TCB for the tid */
10510         tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
10511         tcb_addr += tid * TCB_SIZE;
10512
10513         if (is_t4(sc)) {
10514                 pf = 0;
10515                 win_pos = tcb_addr & ~0xf;      /* start must be 16B aligned */
10516         } else {
10517                 pf = V_PFNUM(sc->pf);
10518                 win_pos = tcb_addr & ~0x7f;     /* start must be 128B aligned */
10519         }
10520         t4_write_reg(sc, reg, win_pos | pf);
10521         t4_read_reg(sc, reg);
10522
10523         off = tcb_addr - win_pos;
10524         for (i = 0; i < 4; i++) {
10525                 uint32_t buf[8];
10526                 for (j = 0; j < 8; j++, off += 4)
10527                         buf[j] = htonl(t4_read_reg(sc, base + off));
10528
10529                 db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
10530                     buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
10531                     buf[7]);
10532         }
10533
10534         t4_write_reg(sc, reg, save);
10535         t4_read_reg(sc, reg);
10536 }
10537
10538 static void
10539 t4_dump_devlog(struct adapter *sc)
10540 {
10541         struct devlog_params *dparams = &sc->params.devlog;
10542         struct fw_devlog_e e;
10543         int i, first, j, m, nentries, rc;
10544         uint64_t ftstamp = UINT64_MAX;
10545
10546         if (dparams->start == 0) {
10547                 db_printf("devlog params not valid\n");
10548                 return;
10549         }
10550
10551         nentries = dparams->size / sizeof(struct fw_devlog_e);
10552         m = fwmtype_to_hwmtype(dparams->memtype);
10553
10554         /* Find the first entry. */
10555         first = -1;
10556         for (i = 0; i < nentries && !db_pager_quit; i++) {
10557                 rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10558                     sizeof(e), (void *)&e);
10559                 if (rc != 0)
10560                         break;
10561
10562                 if (e.timestamp == 0)
10563                         break;
10564
10565                 e.timestamp = be64toh(e.timestamp);
10566                 if (e.timestamp < ftstamp) {
10567                         ftstamp = e.timestamp;
10568                         first = i;
10569                 }
10570         }
10571
10572         if (first == -1)
10573                 return;
10574
10575         i = first;
10576         do {
10577                 rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10578                     sizeof(e), (void *)&e);
10579                 if (rc != 0)
10580                         return;
10581
10582                 if (e.timestamp == 0)
10583                         return;
10584
10585                 e.timestamp = be64toh(e.timestamp);
10586                 e.seqno = be32toh(e.seqno);
10587                 for (j = 0; j < 8; j++)
10588                         e.params[j] = be32toh(e.params[j]);
10589
10590                 db_printf("%10d  %15ju  %8s  %8s  ",
10591                     e.seqno, e.timestamp,
10592                     (e.level < nitems(devlog_level_strings) ?
10593                         devlog_level_strings[e.level] : "UNKNOWN"),
10594                     (e.facility < nitems(devlog_facility_strings) ?
10595                         devlog_facility_strings[e.facility] : "UNKNOWN"));
10596                 db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
10597                     e.params[3], e.params[4], e.params[5], e.params[6],
10598                     e.params[7]);
10599
10600                 if (++i == nentries)
10601                         i = 0;
10602         } while (i != first && !db_pager_quit);
10603 }
10604
10605 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
10606 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
10607
10608 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
10609 {
10610         device_t dev;
10611         int t;
10612         bool valid;
10613
10614         valid = false;
10615         t = db_read_token();
10616         if (t == tIDENT) {
10617                 dev = device_lookup_by_name(db_tok_string);
10618                 valid = true;
10619         }
10620         db_skip_to_eol();
10621         if (!valid) {
10622                 db_printf("usage: show t4 devlog <nexus>\n");
10623                 return;
10624         }
10625
10626         if (dev == NULL) {
10627                 db_printf("device not found\n");
10628                 return;
10629         }
10630
10631         t4_dump_devlog(device_get_softc(dev));
10632 }
10633
10634 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
10635 {
10636         device_t dev;
10637         int radix, tid, t;
10638         bool valid;
10639
10640         valid = false;
10641         radix = db_radix;
10642         db_radix = 10;
10643         t = db_read_token();
10644         if (t == tIDENT) {
10645                 dev = device_lookup_by_name(db_tok_string);
10646                 t = db_read_token();
10647                 if (t == tNUMBER) {
10648                         tid = db_tok_number;
10649                         valid = true;
10650                 }
10651         }       
10652         db_radix = radix;
10653         db_skip_to_eol();
10654         if (!valid) {
10655                 db_printf("usage: show t4 tcb <nexus> <tid>\n");
10656                 return;
10657         }
10658
10659         if (dev == NULL) {
10660                 db_printf("device not found\n");
10661                 return;
10662         }
10663         if (tid < 0) {
10664                 db_printf("invalid tid\n");
10665                 return;
10666         }
10667
10668         t4_dump_tcb(device_get_softc(dev), tid);
10669 }
10670 #endif
10671
10672 /*
10673  * Borrowed from cesa_prep_aes_key().
10674  *
10675  * NB: The crypto engine wants the words in the decryption key in reverse
10676  * order.
10677  */
10678 void
10679 t4_aes_getdeckey(void *dec_key, const void *enc_key, unsigned int kbits)
10680 {
10681         uint32_t ek[4 * (RIJNDAEL_MAXNR + 1)];
10682         uint32_t *dkey;
10683         int i;
10684
10685         rijndaelKeySetupEnc(ek, enc_key, kbits);
10686         dkey = dec_key;
10687         dkey += (kbits / 8) / 4;
10688
10689         switch (kbits) {
10690         case 128:
10691                 for (i = 0; i < 4; i++)
10692                         *--dkey = htobe32(ek[4 * 10 + i]);
10693                 break;
10694         case 192:
10695                 for (i = 0; i < 2; i++)
10696                         *--dkey = htobe32(ek[4 * 11 + 2 + i]);
10697                 for (i = 0; i < 4; i++)
10698                         *--dkey = htobe32(ek[4 * 12 + i]);
10699                 break;
10700         case 256:
10701                 for (i = 0; i < 4; i++)
10702                         *--dkey = htobe32(ek[4 * 13 + i]);
10703                 for (i = 0; i < 4; i++)
10704                         *--dkey = htobe32(ek[4 * 14 + i]);
10705                 break;
10706         }
10707         MPASS(dkey == dec_key);
10708 }
10709
10710 static struct sx mlu;   /* mod load unload */
10711 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
10712
10713 static int
10714 mod_event(module_t mod, int cmd, void *arg)
10715 {
10716         int rc = 0;
10717         static int loaded = 0;
10718
10719         switch (cmd) {
10720         case MOD_LOAD:
10721                 sx_xlock(&mlu);
10722                 if (loaded++ == 0) {
10723                         t4_sge_modload();
10724                         t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10725                             t4_filter_rpl, CPL_COOKIE_FILTER);
10726                         t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL,
10727                             do_l2t_write_rpl, CPL_COOKIE_FILTER);
10728                         t4_register_shared_cpl_handler(CPL_ACT_OPEN_RPL,
10729                             t4_hashfilter_ao_rpl, CPL_COOKIE_HASHFILTER);
10730                         t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10731                             t4_hashfilter_tcb_rpl, CPL_COOKIE_HASHFILTER);
10732                         t4_register_shared_cpl_handler(CPL_ABORT_RPL_RSS,
10733                             t4_del_hashfilter_rpl, CPL_COOKIE_HASHFILTER);
10734                         t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
10735                         t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
10736                         t4_register_cpl_handler(CPL_SMT_WRITE_RPL,
10737                             do_smt_write_rpl);
10738                         sx_init(&t4_list_lock, "T4/T5 adapters");
10739                         SLIST_INIT(&t4_list);
10740                         callout_init(&fatal_callout, 1);
10741 #ifdef TCP_OFFLOAD
10742                         sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
10743                         SLIST_INIT(&t4_uld_list);
10744 #endif
10745 #ifdef INET6
10746                         t4_clip_modload();
10747 #endif
10748                         t4_tracer_modload();
10749                         tweak_tunables();
10750                 }
10751                 sx_xunlock(&mlu);
10752                 break;
10753
10754         case MOD_UNLOAD:
10755                 sx_xlock(&mlu);
10756                 if (--loaded == 0) {
10757                         int tries;
10758
10759                         sx_slock(&t4_list_lock);
10760                         if (!SLIST_EMPTY(&t4_list)) {
10761                                 rc = EBUSY;
10762                                 sx_sunlock(&t4_list_lock);
10763                                 goto done_unload;
10764                         }
10765 #ifdef TCP_OFFLOAD
10766                         sx_slock(&t4_uld_list_lock);
10767                         if (!SLIST_EMPTY(&t4_uld_list)) {
10768                                 rc = EBUSY;
10769                                 sx_sunlock(&t4_uld_list_lock);
10770                                 sx_sunlock(&t4_list_lock);
10771                                 goto done_unload;
10772                         }
10773 #endif
10774                         tries = 0;
10775                         while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
10776                                 uprintf("%ju clusters with custom free routine "
10777                                     "still is use.\n", t4_sge_extfree_refs());
10778                                 pause("t4unload", 2 * hz);
10779                         }
10780 #ifdef TCP_OFFLOAD
10781                         sx_sunlock(&t4_uld_list_lock);
10782 #endif
10783                         sx_sunlock(&t4_list_lock);
10784
10785                         if (t4_sge_extfree_refs() == 0) {
10786                                 t4_tracer_modunload();
10787 #ifdef INET6
10788                                 t4_clip_modunload();
10789 #endif
10790 #ifdef TCP_OFFLOAD
10791                                 sx_destroy(&t4_uld_list_lock);
10792 #endif
10793                                 sx_destroy(&t4_list_lock);
10794                                 t4_sge_modunload();
10795                                 loaded = 0;
10796                         } else {
10797                                 rc = EBUSY;
10798                                 loaded++;       /* undo earlier decrement */
10799                         }
10800                 }
10801 done_unload:
10802                 sx_xunlock(&mlu);
10803                 break;
10804         }
10805
10806         return (rc);
10807 }
10808
10809 static devclass_t t4_devclass, t5_devclass, t6_devclass;
10810 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
10811 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
10812
10813 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
10814 MODULE_VERSION(t4nex, 1);
10815 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
10816 #ifdef DEV_NETMAP
10817 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
10818 #endif /* DEV_NETMAP */
10819
10820 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
10821 MODULE_VERSION(t5nex, 1);
10822 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
10823 #ifdef DEV_NETMAP
10824 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
10825 #endif /* DEV_NETMAP */
10826
10827 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
10828 MODULE_VERSION(t6nex, 1);
10829 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
10830 #ifdef DEV_NETMAP
10831 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
10832 #endif /* DEV_NETMAP */
10833
10834 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
10835 MODULE_VERSION(cxgbe, 1);
10836
10837 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
10838 MODULE_VERSION(cxl, 1);
10839
10840 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
10841 MODULE_VERSION(cc, 1);
10842
10843 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
10844 MODULE_VERSION(vcxgbe, 1);
10845
10846 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
10847 MODULE_VERSION(vcxl, 1);
10848
10849 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
10850 MODULE_VERSION(vcc, 1);