2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2011 Chelsio Communications, Inc.
6 * Written by: Navdeep Parhar <np@FreeBSD.org>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
35 #include "opt_inet6.h"
36 #include "opt_ratelimit.h"
39 #include <sys/param.h>
42 #include <sys/kernel.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>
55 #include <sys/socket.h>
56 #include <sys/sockio.h>
57 #include <sys/sysctl.h>
58 #include <net/ethernet.h>
60 #include <net/if_types.h>
61 #include <net/if_dl.h>
62 #include <net/if_vlan_var.h>
64 #include <net/rss_config.h>
66 #if defined(__i386__) || defined(__amd64__)
67 #include <machine/md_var.h>
68 #include <machine/cputypes.h>
72 #include <crypto/rijndael/rijndael.h>
75 #include <ddb/db_lex.h>
78 #include "common/common.h"
79 #include "common/t4_msg.h"
80 #include "common/t4_regs.h"
81 #include "common/t4_regs_values.h"
82 #include "cudbg/cudbg.h"
85 #include "t4_mp_ring.h"
88 /* T4 bus driver interface */
89 static int t4_probe(device_t);
90 static int t4_attach(device_t);
91 static int t4_detach(device_t);
92 static int t4_ready(device_t);
93 static int t4_read_port_device(device_t, int, device_t *);
94 static device_method_t t4_methods[] = {
95 DEVMETHOD(device_probe, t4_probe),
96 DEVMETHOD(device_attach, t4_attach),
97 DEVMETHOD(device_detach, t4_detach),
99 DEVMETHOD(t4_is_main_ready, t4_ready),
100 DEVMETHOD(t4_read_port_device, t4_read_port_device),
104 static driver_t t4_driver = {
107 sizeof(struct adapter)
111 /* T4 port (cxgbe) interface */
112 static int cxgbe_probe(device_t);
113 static int cxgbe_attach(device_t);
114 static int cxgbe_detach(device_t);
115 device_method_t cxgbe_methods[] = {
116 DEVMETHOD(device_probe, cxgbe_probe),
117 DEVMETHOD(device_attach, cxgbe_attach),
118 DEVMETHOD(device_detach, cxgbe_detach),
121 static driver_t cxgbe_driver = {
124 sizeof(struct port_info)
127 /* T4 VI (vcxgbe) interface */
128 static int vcxgbe_probe(device_t);
129 static int vcxgbe_attach(device_t);
130 static int vcxgbe_detach(device_t);
131 static device_method_t vcxgbe_methods[] = {
132 DEVMETHOD(device_probe, vcxgbe_probe),
133 DEVMETHOD(device_attach, vcxgbe_attach),
134 DEVMETHOD(device_detach, vcxgbe_detach),
137 static driver_t vcxgbe_driver = {
140 sizeof(struct vi_info)
143 static d_ioctl_t t4_ioctl;
145 static struct cdevsw t4_cdevsw = {
146 .d_version = D_VERSION,
151 /* T5 bus driver interface */
152 static int t5_probe(device_t);
153 static device_method_t t5_methods[] = {
154 DEVMETHOD(device_probe, t5_probe),
155 DEVMETHOD(device_attach, t4_attach),
156 DEVMETHOD(device_detach, t4_detach),
158 DEVMETHOD(t4_is_main_ready, t4_ready),
159 DEVMETHOD(t4_read_port_device, t4_read_port_device),
163 static driver_t t5_driver = {
166 sizeof(struct adapter)
170 /* T5 port (cxl) interface */
171 static driver_t cxl_driver = {
174 sizeof(struct port_info)
177 /* T5 VI (vcxl) interface */
178 static driver_t vcxl_driver = {
181 sizeof(struct vi_info)
184 /* T6 bus driver interface */
185 static int t6_probe(device_t);
186 static device_method_t t6_methods[] = {
187 DEVMETHOD(device_probe, t6_probe),
188 DEVMETHOD(device_attach, t4_attach),
189 DEVMETHOD(device_detach, t4_detach),
191 DEVMETHOD(t4_is_main_ready, t4_ready),
192 DEVMETHOD(t4_read_port_device, t4_read_port_device),
196 static driver_t t6_driver = {
199 sizeof(struct adapter)
203 /* T6 port (cc) interface */
204 static driver_t cc_driver = {
207 sizeof(struct port_info)
210 /* T6 VI (vcc) interface */
211 static driver_t vcc_driver = {
214 sizeof(struct vi_info)
217 /* ifnet + media interface */
218 static void cxgbe_init(void *);
219 static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
220 static int cxgbe_transmit(struct ifnet *, struct mbuf *);
221 static void cxgbe_qflush(struct ifnet *);
222 static int cxgbe_media_change(struct ifnet *);
223 static void cxgbe_media_status(struct ifnet *, struct ifmediareq *);
225 MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
228 * Correct lock order when you need to acquire multiple locks is t4_list_lock,
229 * then ADAPTER_LOCK, then t4_uld_list_lock.
231 static struct sx t4_list_lock;
232 SLIST_HEAD(, adapter) t4_list;
234 static struct sx t4_uld_list_lock;
235 SLIST_HEAD(, uld_info) t4_uld_list;
239 * Tunables. See tweak_tunables() too.
241 * Each tunable is set to a default value here if it's known at compile-time.
242 * Otherwise it is set to -n as an indication to tweak_tunables() that it should
243 * provide a reasonable default (upto n) when the driver is loaded.
245 * Tunables applicable to both T4 and T5 are under hw.cxgbe. Those specific to
246 * T5 are under hw.cxl.
250 * Number of queues for tx and rx, NIC and offload.
254 TUNABLE_INT("hw.cxgbe.ntxq", &t4_ntxq);
255 TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq); /* Old name, undocumented */
259 TUNABLE_INT("hw.cxgbe.nrxq", &t4_nrxq);
260 TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq); /* Old name, undocumented */
263 static int t4_ntxq_vi = -NTXQ_VI;
264 TUNABLE_INT("hw.cxgbe.ntxq_vi", &t4_ntxq_vi);
267 static int t4_nrxq_vi = -NRXQ_VI;
268 TUNABLE_INT("hw.cxgbe.nrxq_vi", &t4_nrxq_vi);
270 static int t4_rsrv_noflowq = 0;
271 TUNABLE_INT("hw.cxgbe.rsrv_noflowq", &t4_rsrv_noflowq);
273 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
275 static int t4_nofldtxq = -NOFLDTXQ;
276 TUNABLE_INT("hw.cxgbe.nofldtxq", &t4_nofldtxq);
279 static int t4_nofldrxq = -NOFLDRXQ;
280 TUNABLE_INT("hw.cxgbe.nofldrxq", &t4_nofldrxq);
282 #define NOFLDTXQ_VI 1
283 static int t4_nofldtxq_vi = -NOFLDTXQ_VI;
284 TUNABLE_INT("hw.cxgbe.nofldtxq_vi", &t4_nofldtxq_vi);
286 #define NOFLDRXQ_VI 1
287 static int t4_nofldrxq_vi = -NOFLDRXQ_VI;
288 TUNABLE_INT("hw.cxgbe.nofldrxq_vi", &t4_nofldrxq_vi);
290 #define TMR_IDX_OFLD 1
291 int t4_tmr_idx_ofld = TMR_IDX_OFLD;
292 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_ofld", &t4_tmr_idx_ofld);
294 #define PKTC_IDX_OFLD (-1)
295 int t4_pktc_idx_ofld = PKTC_IDX_OFLD;
296 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_ofld", &t4_pktc_idx_ofld);
298 /* 0 means chip/fw default, non-zero number is value in microseconds */
299 static u_long t4_toe_keepalive_idle = 0;
300 TUNABLE_ULONG("hw.cxgbe.toe.keepalive_idle", &t4_toe_keepalive_idle);
302 /* 0 means chip/fw default, non-zero number is value in microseconds */
303 static u_long t4_toe_keepalive_interval = 0;
304 TUNABLE_ULONG("hw.cxgbe.toe.keepalive_interval", &t4_toe_keepalive_interval);
306 /* 0 means chip/fw default, non-zero number is # of keepalives before abort */
307 static int t4_toe_keepalive_count = 0;
308 TUNABLE_INT("hw.cxgbe.toe.keepalive_count", &t4_toe_keepalive_count);
310 /* 0 means chip/fw default, non-zero number is value in microseconds */
311 static u_long t4_toe_rexmt_min = 0;
312 TUNABLE_ULONG("hw.cxgbe.toe.rexmt_min", &t4_toe_rexmt_min);
314 /* 0 means chip/fw default, non-zero number is value in microseconds */
315 static u_long t4_toe_rexmt_max = 0;
316 TUNABLE_ULONG("hw.cxgbe.toe.rexmt_max", &t4_toe_rexmt_max);
318 /* 0 means chip/fw default, non-zero number is # of rexmt before abort */
319 static int t4_toe_rexmt_count = 0;
320 TUNABLE_INT("hw.cxgbe.toe.rexmt_count", &t4_toe_rexmt_count);
322 /* -1 means chip/fw default, other values are raw backoff values to use */
323 static int t4_toe_rexmt_backoff[16] = {
324 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
326 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.0", &t4_toe_rexmt_backoff[0]);
327 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.1", &t4_toe_rexmt_backoff[1]);
328 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.2", &t4_toe_rexmt_backoff[2]);
329 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.3", &t4_toe_rexmt_backoff[3]);
330 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.4", &t4_toe_rexmt_backoff[4]);
331 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.5", &t4_toe_rexmt_backoff[5]);
332 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.6", &t4_toe_rexmt_backoff[6]);
333 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.7", &t4_toe_rexmt_backoff[7]);
334 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.8", &t4_toe_rexmt_backoff[8]);
335 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.9", &t4_toe_rexmt_backoff[9]);
336 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.10", &t4_toe_rexmt_backoff[10]);
337 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.11", &t4_toe_rexmt_backoff[11]);
338 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.12", &t4_toe_rexmt_backoff[12]);
339 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.13", &t4_toe_rexmt_backoff[13]);
340 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.14", &t4_toe_rexmt_backoff[14]);
341 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.15", &t4_toe_rexmt_backoff[15]);
346 static int t4_nnmtxq_vi = -NNMTXQ_VI;
347 TUNABLE_INT("hw.cxgbe.nnmtxq_vi", &t4_nnmtxq_vi);
350 static int t4_nnmrxq_vi = -NNMRXQ_VI;
351 TUNABLE_INT("hw.cxgbe.nnmrxq_vi", &t4_nnmrxq_vi);
355 * Holdoff parameters for ports.
358 int t4_tmr_idx = TMR_IDX;
359 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx", &t4_tmr_idx);
360 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx); /* Old name */
362 #define PKTC_IDX (-1)
363 int t4_pktc_idx = PKTC_IDX;
364 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx", &t4_pktc_idx);
365 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx); /* Old name */
368 * Size (# of entries) of each tx and rx queue.
370 unsigned int t4_qsize_txq = TX_EQ_QSIZE;
371 TUNABLE_INT("hw.cxgbe.qsize_txq", &t4_qsize_txq);
373 unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
374 TUNABLE_INT("hw.cxgbe.qsize_rxq", &t4_qsize_rxq);
377 * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
379 int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
380 TUNABLE_INT("hw.cxgbe.interrupt_types", &t4_intr_types);
383 * Configuration file. All the _CF names here are special.
385 #define DEFAULT_CF "default"
386 #define BUILTIN_CF "built-in"
387 #define FLASH_CF "flash"
388 #define UWIRE_CF "uwire"
389 #define FPGA_CF "fpga"
390 static char t4_cfg_file[32] = DEFAULT_CF;
391 TUNABLE_STR("hw.cxgbe.config_file", t4_cfg_file, sizeof(t4_cfg_file));
394 * PAUSE settings (bit 0, 1 = rx_pause, tx_pause respectively).
395 * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
396 * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
397 * mark or when signalled to do so, 0 to never emit PAUSE.
399 static int t4_pause_settings = PAUSE_TX | PAUSE_RX;
400 TUNABLE_INT("hw.cxgbe.pause_settings", &t4_pause_settings);
403 * Forward Error Correction settings (bit 0, 1, 2 = FEC_RS, FEC_BASER_RS,
404 * FEC_RESERVED respectively).
405 * -1 to run with the firmware default.
408 static int t4_fec = -1;
409 TUNABLE_INT("hw.cxgbe.fec", &t4_fec);
412 * Link autonegotiation.
413 * -1 to run with the firmware default.
417 static int t4_autoneg = -1;
418 TUNABLE_INT("hw.cxgbe.autoneg", &t4_autoneg);
421 * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
422 * encouraged respectively).
424 static unsigned int t4_fw_install = 1;
425 TUNABLE_INT("hw.cxgbe.fw_install", &t4_fw_install);
428 * ASIC features that will be used. Disable the ones you don't want so that the
429 * chip resources aren't wasted on features that will not be used.
431 static int t4_nbmcaps_allowed = 0;
432 TUNABLE_INT("hw.cxgbe.nbmcaps_allowed", &t4_nbmcaps_allowed);
434 static int t4_linkcaps_allowed = 0; /* No DCBX, PPP, etc. by default */
435 TUNABLE_INT("hw.cxgbe.linkcaps_allowed", &t4_linkcaps_allowed);
437 static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
438 FW_CAPS_CONFIG_SWITCH_EGRESS;
439 TUNABLE_INT("hw.cxgbe.switchcaps_allowed", &t4_switchcaps_allowed);
441 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
442 FW_CAPS_CONFIG_NIC_HASHFILTER | FW_CAPS_CONFIG_NIC_ETHOFLD;
443 TUNABLE_INT("hw.cxgbe.niccaps_allowed", &t4_niccaps_allowed);
445 static int t4_toecaps_allowed = -1;
446 TUNABLE_INT("hw.cxgbe.toecaps_allowed", &t4_toecaps_allowed);
448 static int t4_rdmacaps_allowed = -1;
449 TUNABLE_INT("hw.cxgbe.rdmacaps_allowed", &t4_rdmacaps_allowed);
451 static int t4_cryptocaps_allowed = -1;
452 TUNABLE_INT("hw.cxgbe.cryptocaps_allowed", &t4_cryptocaps_allowed);
454 static int t4_iscsicaps_allowed = -1;
455 TUNABLE_INT("hw.cxgbe.iscsicaps_allowed", &t4_iscsicaps_allowed);
457 static int t4_fcoecaps_allowed = 0;
458 TUNABLE_INT("hw.cxgbe.fcoecaps_allowed", &t4_fcoecaps_allowed);
460 static int t5_write_combine = 0;
461 TUNABLE_INT("hw.cxl.write_combine", &t5_write_combine);
463 static int t4_num_vis = 1;
464 TUNABLE_INT("hw.cxgbe.num_vis", &t4_num_vis);
466 * PCIe Relaxed Ordering.
467 * -1: driver should figure out a good value.
472 static int pcie_relaxed_ordering = -1;
473 TUNABLE_INT("hw.cxgbe.pcie_relaxed_ordering", &pcie_relaxed_ordering);
475 static int t4_panic_on_fatal_err = 0;
476 TUNABLE_INT("hw.cxgbe.panic_on_fatal_err", &t4_panic_on_fatal_err);
482 static int t4_cop_managed_offloading = 0;
483 TUNABLE_INT("hw.cxgbe.cop_managed_offloading", &t4_cop_managed_offloading);
486 /* Functions used by VIs to obtain unique MAC addresses for each VI. */
487 static int vi_mac_funcs[] = {
491 FW_VI_FUNC_OPENISCSI,
497 struct intrs_and_queues {
498 uint16_t intr_type; /* INTx, MSI, or MSI-X */
499 uint16_t num_vis; /* number of VIs for each port */
500 uint16_t nirq; /* Total # of vectors */
501 uint16_t ntxq; /* # of NIC txq's for each port */
502 uint16_t nrxq; /* # of NIC rxq's for each port */
503 uint16_t nofldtxq; /* # of TOE/ETHOFLD txq's for each port */
504 uint16_t nofldrxq; /* # of TOE rxq's for each port */
506 /* The vcxgbe/vcxl interfaces use these and not the ones above. */
507 uint16_t ntxq_vi; /* # of NIC txq's */
508 uint16_t nrxq_vi; /* # of NIC rxq's */
509 uint16_t nofldtxq_vi; /* # of TOE txq's */
510 uint16_t nofldrxq_vi; /* # of TOE rxq's */
511 uint16_t nnmtxq_vi; /* # of netmap txq's */
512 uint16_t nnmrxq_vi; /* # of netmap rxq's */
515 static void setup_memwin(struct adapter *);
516 static void position_memwin(struct adapter *, int, uint32_t);
517 static int validate_mem_range(struct adapter *, uint32_t, int);
518 static int fwmtype_to_hwmtype(int);
519 static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
521 static int fixup_devlog_params(struct adapter *);
522 static int cfg_itype_and_nqueues(struct adapter *, struct intrs_and_queues *);
523 static int prep_firmware(struct adapter *);
524 static int partition_resources(struct adapter *, const struct firmware *,
526 static int get_params__pre_init(struct adapter *);
527 static int get_params__post_init(struct adapter *);
528 static int set_params__post_init(struct adapter *);
529 static void t4_set_desc(struct adapter *);
530 static void build_medialist(struct port_info *, struct ifmedia *);
531 static void init_l1cfg(struct port_info *);
532 static int cxgbe_init_synchronized(struct vi_info *);
533 static int cxgbe_uninit_synchronized(struct vi_info *);
534 static void quiesce_txq(struct adapter *, struct sge_txq *);
535 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
536 static void quiesce_iq(struct adapter *, struct sge_iq *);
537 static void quiesce_fl(struct adapter *, struct sge_fl *);
538 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
539 driver_intr_t *, void *, char *);
540 static int t4_free_irq(struct adapter *, struct irq *);
541 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
542 static void vi_refresh_stats(struct adapter *, struct vi_info *);
543 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
544 static void cxgbe_tick(void *);
545 static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
546 static void cxgbe_sysctls(struct port_info *);
547 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
548 static int sysctl_bitfield(SYSCTL_HANDLER_ARGS);
549 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
550 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
551 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
552 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
553 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
554 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
555 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
556 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
557 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
558 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
559 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
561 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
562 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
563 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
564 static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
565 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
566 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
567 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
568 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
569 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
570 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
571 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
572 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
573 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
574 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
575 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
576 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
577 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
578 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
579 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
580 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
581 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
582 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
583 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
584 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
585 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
586 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
587 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
588 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
589 static int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
592 static int sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS);
593 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
594 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
595 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
596 static int sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS);
597 static int sysctl_tp_backoff(SYSCTL_HANDLER_ARGS);
598 static int sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS);
599 static int sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS);
601 static int get_sge_context(struct adapter *, struct t4_sge_context *);
602 static int load_fw(struct adapter *, struct t4_data *);
603 static int load_cfg(struct adapter *, struct t4_data *);
604 static int load_boot(struct adapter *, struct t4_bootrom *);
605 static int load_bootcfg(struct adapter *, struct t4_data *);
606 static int cudbg_dump(struct adapter *, struct t4_cudbg_dump *);
607 static void free_offload_policy(struct t4_offload_policy *);
608 static int set_offload_policy(struct adapter *, struct t4_offload_policy *);
609 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
610 static int read_i2c(struct adapter *, struct t4_i2c_data *);
612 static int toe_capability(struct vi_info *, int);
614 static int mod_event(module_t, int, void *);
615 static int notify_siblings(device_t, int);
621 {0xa000, "Chelsio Terminator 4 FPGA"},
622 {0x4400, "Chelsio T440-dbg"},
623 {0x4401, "Chelsio T420-CR"},
624 {0x4402, "Chelsio T422-CR"},
625 {0x4403, "Chelsio T440-CR"},
626 {0x4404, "Chelsio T420-BCH"},
627 {0x4405, "Chelsio T440-BCH"},
628 {0x4406, "Chelsio T440-CH"},
629 {0x4407, "Chelsio T420-SO"},
630 {0x4408, "Chelsio T420-CX"},
631 {0x4409, "Chelsio T420-BT"},
632 {0x440a, "Chelsio T404-BT"},
633 {0x440e, "Chelsio T440-LP-CR"},
635 {0xb000, "Chelsio Terminator 5 FPGA"},
636 {0x5400, "Chelsio T580-dbg"},
637 {0x5401, "Chelsio T520-CR"}, /* 2 x 10G */
638 {0x5402, "Chelsio T522-CR"}, /* 2 x 10G, 2 X 1G */
639 {0x5403, "Chelsio T540-CR"}, /* 4 x 10G */
640 {0x5407, "Chelsio T520-SO"}, /* 2 x 10G, nomem */
641 {0x5409, "Chelsio T520-BT"}, /* 2 x 10GBaseT */
642 {0x540a, "Chelsio T504-BT"}, /* 4 x 1G */
643 {0x540d, "Chelsio T580-CR"}, /* 2 x 40G */
644 {0x540e, "Chelsio T540-LP-CR"}, /* 4 x 10G */
645 {0x5410, "Chelsio T580-LP-CR"}, /* 2 x 40G */
646 {0x5411, "Chelsio T520-LL-CR"}, /* 2 x 10G */
647 {0x5412, "Chelsio T560-CR"}, /* 1 x 40G, 2 x 10G */
648 {0x5414, "Chelsio T580-LP-SO-CR"}, /* 2 x 40G, nomem */
649 {0x5415, "Chelsio T502-BT"}, /* 2 x 1G */
650 {0x5418, "Chelsio T540-BT"}, /* 4 x 10GBaseT */
651 {0x5419, "Chelsio T540-LP-BT"}, /* 4 x 10GBaseT */
652 {0x541a, "Chelsio T540-SO-BT"}, /* 4 x 10GBaseT, nomem */
653 {0x541b, "Chelsio T540-SO-CR"}, /* 4 x 10G, nomem */
655 {0xc006, "Chelsio Terminator 6 FPGA"}, /* T6 PE10K6 FPGA (PF0) */
656 {0x6400, "Chelsio T6-DBG-25"}, /* 2 x 10/25G, debug */
657 {0x6401, "Chelsio T6225-CR"}, /* 2 x 10/25G */
658 {0x6402, "Chelsio T6225-SO-CR"}, /* 2 x 10/25G, nomem */
659 {0x6403, "Chelsio T6425-CR"}, /* 4 x 10/25G */
660 {0x6404, "Chelsio T6425-SO-CR"}, /* 4 x 10/25G, nomem */
661 {0x6405, "Chelsio T6225-OCP-SO"}, /* 2 x 10/25G, nomem */
662 {0x6406, "Chelsio T62100-OCP-SO"}, /* 2 x 40/50/100G, nomem */
663 {0x6407, "Chelsio T62100-LP-CR"}, /* 2 x 40/50/100G */
664 {0x6408, "Chelsio T62100-SO-CR"}, /* 2 x 40/50/100G, nomem */
665 {0x6409, "Chelsio T6210-BT"}, /* 2 x 10GBASE-T */
666 {0x640d, "Chelsio T62100-CR"}, /* 2 x 40/50/100G */
667 {0x6410, "Chelsio T6-DBG-100"}, /* 2 x 40/50/100G, debug */
668 {0x6411, "Chelsio T6225-LL-CR"}, /* 2 x 10/25G */
669 {0x6414, "Chelsio T61100-OCP-SO"}, /* 1 x 40/50/100G, nomem */
670 {0x6415, "Chelsio T6201-BT"}, /* 2 x 1000BASE-T */
673 {0x6480, "Custom T6225-CR"},
674 {0x6481, "Custom T62100-CR"},
675 {0x6482, "Custom T6225-CR"},
676 {0x6483, "Custom T62100-CR"},
677 {0x6484, "Custom T64100-CR"},
678 {0x6485, "Custom T6240-SO"},
679 {0x6486, "Custom T6225-SO-CR"},
680 {0x6487, "Custom T6225-CR"},
685 * service_iq() has an iq and needs the fl. Offset of fl from the iq should be
686 * exactly the same for both rxq and ofld_rxq.
688 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
689 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
691 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
694 t4_probe(device_t dev)
697 uint16_t v = pci_get_vendor(dev);
698 uint16_t d = pci_get_device(dev);
699 uint8_t f = pci_get_function(dev);
701 if (v != PCI_VENDOR_ID_CHELSIO)
704 /* Attach only to PF0 of the FPGA */
705 if (d == 0xa000 && f != 0)
708 for (i = 0; i < nitems(t4_pciids); i++) {
709 if (d == t4_pciids[i].device) {
710 device_set_desc(dev, t4_pciids[i].desc);
711 return (BUS_PROBE_DEFAULT);
719 t5_probe(device_t dev)
722 uint16_t v = pci_get_vendor(dev);
723 uint16_t d = pci_get_device(dev);
724 uint8_t f = pci_get_function(dev);
726 if (v != PCI_VENDOR_ID_CHELSIO)
729 /* Attach only to PF0 of the FPGA */
730 if (d == 0xb000 && f != 0)
733 for (i = 0; i < nitems(t5_pciids); i++) {
734 if (d == t5_pciids[i].device) {
735 device_set_desc(dev, t5_pciids[i].desc);
736 return (BUS_PROBE_DEFAULT);
744 t6_probe(device_t dev)
747 uint16_t v = pci_get_vendor(dev);
748 uint16_t d = pci_get_device(dev);
750 if (v != PCI_VENDOR_ID_CHELSIO)
753 for (i = 0; i < nitems(t6_pciids); i++) {
754 if (d == t6_pciids[i].device) {
755 device_set_desc(dev, t6_pciids[i].desc);
756 return (BUS_PROBE_DEFAULT);
764 t5_attribute_workaround(device_t dev)
770 * The T5 chips do not properly echo the No Snoop and Relaxed
771 * Ordering attributes when replying to a TLP from a Root
772 * Port. As a workaround, find the parent Root Port and
773 * disable No Snoop and Relaxed Ordering. Note that this
774 * affects all devices under this root port.
776 root_port = pci_find_pcie_root_port(dev);
777 if (root_port == NULL) {
778 device_printf(dev, "Unable to find parent root port\n");
782 v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
783 PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
784 if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
786 device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
787 device_get_nameunit(root_port));
790 static const struct devnames devnames[] = {
792 .nexus_name = "t4nex",
793 .ifnet_name = "cxgbe",
794 .vi_ifnet_name = "vcxgbe",
795 .pf03_drv_name = "t4iov",
796 .vf_nexus_name = "t4vf",
797 .vf_ifnet_name = "cxgbev"
799 .nexus_name = "t5nex",
801 .vi_ifnet_name = "vcxl",
802 .pf03_drv_name = "t5iov",
803 .vf_nexus_name = "t5vf",
804 .vf_ifnet_name = "cxlv"
806 .nexus_name = "t6nex",
808 .vi_ifnet_name = "vcc",
809 .pf03_drv_name = "t6iov",
810 .vf_nexus_name = "t6vf",
811 .vf_ifnet_name = "ccv"
816 t4_init_devnames(struct adapter *sc)
821 if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
822 sc->names = &devnames[id - CHELSIO_T4];
824 device_printf(sc->dev, "chip id %d is not supported.\n", id);
830 t4_attach(device_t dev)
833 int rc = 0, i, j, rqidx, tqidx, nports;
834 struct make_dev_args mda;
835 struct intrs_and_queues iaq;
838 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
845 int nm_rqidx, nm_tqidx;
849 sc = device_get_softc(dev);
851 TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
853 if ((pci_get_device(dev) & 0xff00) == 0x5400)
854 t5_attribute_workaround(dev);
855 pci_enable_busmaster(dev);
856 if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
859 pci_set_max_read_req(dev, 4096);
860 v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
861 sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
862 if (pcie_relaxed_ordering == 0 &&
863 (v | PCIEM_CTL_RELAXED_ORD_ENABLE) != 0) {
864 v &= ~PCIEM_CTL_RELAXED_ORD_ENABLE;
865 pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
866 } else if (pcie_relaxed_ordering == 1 &&
867 (v & PCIEM_CTL_RELAXED_ORD_ENABLE) == 0) {
868 v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
869 pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
873 sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
874 sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
876 mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
877 snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
878 device_get_nameunit(dev));
880 snprintf(sc->lockname, sizeof(sc->lockname), "%s",
881 device_get_nameunit(dev));
882 mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
885 mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
886 TAILQ_INIT(&sc->sfl);
887 callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
889 mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
892 rw_init(&sc->policy_lock, "connection offload policy");
894 rc = t4_map_bars_0_and_4(sc);
896 goto done; /* error message displayed already */
898 memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
900 /* Prepare the adapter for operation. */
901 buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
902 rc = -t4_prep_adapter(sc, buf);
905 device_printf(dev, "failed to prepare adapter: %d.\n", rc);
910 * This is the real PF# to which we're attaching. Works from within PCI
911 * passthrough environments too, where pci_get_function() could return a
912 * different PF# depending on the passthrough configuration. We need to
913 * use the real PF# in all our communication with the firmware.
915 j = t4_read_reg(sc, A_PL_WHOAMI);
916 sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
919 t4_init_devnames(sc);
920 if (sc->names == NULL) {
922 goto done; /* error message displayed already */
926 * Do this really early, with the memory windows set up even before the
927 * character device. The userland tool's register i/o and mem read
928 * will work even in "recovery mode".
931 if (t4_init_devlog_params(sc, 0) == 0)
932 fixup_devlog_params(sc);
933 make_dev_args_init(&mda);
934 mda.mda_devsw = &t4_cdevsw;
935 mda.mda_uid = UID_ROOT;
936 mda.mda_gid = GID_WHEEL;
938 mda.mda_si_drv1 = sc;
939 rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
941 device_printf(dev, "failed to create nexus char device: %d.\n",
944 /* Go no further if recovery mode has been requested. */
945 if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
946 device_printf(dev, "recovery mode.\n");
950 #if defined(__i386__)
951 if ((cpu_feature & CPUID_CX8) == 0) {
952 device_printf(dev, "64 bit atomics not available.\n");
958 /* Prepare the firmware for operation */
959 rc = prep_firmware(sc);
961 goto done; /* error message displayed already */
963 rc = get_params__post_init(sc);
965 goto done; /* error message displayed already */
967 rc = set_params__post_init(sc);
969 goto done; /* error message displayed already */
971 rc = t4_map_bar_2(sc);
973 goto done; /* error message displayed already */
975 rc = t4_create_dma_tag(sc);
977 goto done; /* error message displayed already */
980 * First pass over all the ports - allocate VIs and initialize some
981 * basic parameters like mac address, port type, etc.
983 for_each_port(sc, i) {
984 struct port_info *pi;
986 pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
989 /* These must be set before t4_port_init */
993 * XXX: vi[0] is special so we can't delay this allocation until
994 * pi->nvi's final value is known.
996 pi->vi = malloc(sizeof(struct vi_info) * t4_num_vis, M_CXGBE,
1000 * Allocate the "main" VI and initialize parameters
1003 rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
1005 device_printf(dev, "unable to initialize port %d: %d\n",
1007 free(pi->vi, M_CXGBE);
1013 snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
1014 device_get_nameunit(dev), i);
1015 mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
1016 sc->chan_map[pi->tx_chan] = i;
1018 /* All VIs on this port share this media. */
1019 ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
1020 cxgbe_media_status);
1022 pi->dev = device_add_child(dev, sc->names->ifnet_name, -1);
1023 if (pi->dev == NULL) {
1025 "failed to add device for port %d.\n", i);
1029 pi->vi[0].dev = pi->dev;
1030 device_set_softc(pi->dev, pi);
1034 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1036 nports = sc->params.nports;
1037 rc = cfg_itype_and_nqueues(sc, &iaq);
1039 goto done; /* error message displayed already */
1041 num_vis = iaq.num_vis;
1042 sc->intr_type = iaq.intr_type;
1043 sc->intr_count = iaq.nirq;
1046 s->nrxq = nports * iaq.nrxq;
1047 s->ntxq = nports * iaq.ntxq;
1049 s->nrxq += nports * (num_vis - 1) * iaq.nrxq_vi;
1050 s->ntxq += nports * (num_vis - 1) * iaq.ntxq_vi;
1052 s->neq = s->ntxq + s->nrxq; /* the free list in an rxq is an eq */
1053 s->neq += nports + 1;/* ctrl queues: 1 per port + 1 mgmt */
1054 s->niq = s->nrxq + 1; /* 1 extra for firmware event queue */
1055 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1056 if (is_offload(sc) || is_ethoffload(sc)) {
1057 s->nofldtxq = nports * iaq.nofldtxq;
1059 s->nofldtxq += nports * (num_vis - 1) * iaq.nofldtxq_vi;
1060 s->neq += s->nofldtxq;
1062 s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1063 M_CXGBE, M_ZERO | M_WAITOK);
1067 if (is_offload(sc)) {
1068 s->nofldrxq = nports * iaq.nofldrxq;
1070 s->nofldrxq += nports * (num_vis - 1) * iaq.nofldrxq_vi;
1071 s->neq += s->nofldrxq; /* free list */
1072 s->niq += s->nofldrxq;
1074 s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1075 M_CXGBE, M_ZERO | M_WAITOK);
1080 s->nnmrxq = nports * (num_vis - 1) * iaq.nnmrxq_vi;
1081 s->nnmtxq = nports * (num_vis - 1) * iaq.nnmtxq_vi;
1083 s->neq += s->nnmtxq + s->nnmrxq;
1084 s->niq += s->nnmrxq;
1086 s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1087 M_CXGBE, M_ZERO | M_WAITOK);
1088 s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1089 M_CXGBE, M_ZERO | M_WAITOK);
1092 s->ctrlq = malloc(nports * sizeof(struct sge_wrq), M_CXGBE,
1094 s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1096 s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1098 s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1100 s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1103 sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1106 t4_init_l2t(sc, M_WAITOK);
1107 t4_init_tx_sched(sc);
1109 t4_init_etid_table(sc);
1113 * Second pass over the ports. This time we know the number of rx and
1114 * tx queues that each port should get.
1117 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1124 nm_rqidx = nm_tqidx = 0;
1126 for_each_port(sc, i) {
1127 struct port_info *pi = sc->port[i];
1134 for_each_vi(pi, j, vi) {
1136 vi->qsize_rxq = t4_qsize_rxq;
1137 vi->qsize_txq = t4_qsize_txq;
1139 vi->first_rxq = rqidx;
1140 vi->first_txq = tqidx;
1141 vi->tmr_idx = t4_tmr_idx;
1142 vi->pktc_idx = t4_pktc_idx;
1143 vi->nrxq = j == 0 ? iaq.nrxq : iaq.nrxq_vi;
1144 vi->ntxq = j == 0 ? iaq.ntxq : iaq.ntxq_vi;
1149 if (j == 0 && vi->ntxq > 1)
1150 vi->rsrv_noflowq = t4_rsrv_noflowq ? 1 : 0;
1152 vi->rsrv_noflowq = 0;
1154 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1155 vi->first_ofld_txq = ofld_tqidx;
1156 vi->nofldtxq = j == 0 ? iaq.nofldtxq : iaq.nofldtxq_vi;
1157 ofld_tqidx += vi->nofldtxq;
1160 vi->ofld_tmr_idx = t4_tmr_idx_ofld;
1161 vi->ofld_pktc_idx = t4_pktc_idx_ofld;
1162 vi->first_ofld_rxq = ofld_rqidx;
1163 vi->nofldrxq = j == 0 ? iaq.nofldrxq : iaq.nofldrxq_vi;
1165 ofld_rqidx += vi->nofldrxq;
1169 vi->first_nm_rxq = nm_rqidx;
1170 vi->first_nm_txq = nm_tqidx;
1171 vi->nnmrxq = iaq.nnmrxq_vi;
1172 vi->nnmtxq = iaq.nnmtxq_vi;
1173 nm_rqidx += vi->nnmrxq;
1174 nm_tqidx += vi->nnmtxq;
1180 rc = t4_setup_intr_handlers(sc);
1183 "failed to setup interrupt handlers: %d\n", rc);
1187 rc = bus_generic_probe(dev);
1189 device_printf(dev, "failed to probe child drivers: %d\n", rc);
1194 * Ensure thread-safe mailbox access (in debug builds).
1196 * So far this was the only thread accessing the mailbox but various
1197 * ifnets and sysctls are about to be created and their handlers/ioctls
1198 * will access the mailbox from different threads.
1200 sc->flags |= CHK_MBOX_ACCESS;
1202 rc = bus_generic_attach(dev);
1205 "failed to attach all child ports: %d\n", rc);
1210 "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1211 sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1212 sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1213 (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1214 sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1218 notify_siblings(dev, 0);
1221 if (rc != 0 && sc->cdev) {
1222 /* cdev was created and so cxgbetool works; recover that way. */
1224 "error during attach, adapter is now in recovery mode.\n");
1229 t4_detach_common(dev);
1237 t4_ready(device_t dev)
1241 sc = device_get_softc(dev);
1242 if (sc->flags & FW_OK)
1248 t4_read_port_device(device_t dev, int port, device_t *child)
1251 struct port_info *pi;
1253 sc = device_get_softc(dev);
1254 if (port < 0 || port >= MAX_NPORTS)
1256 pi = sc->port[port];
1257 if (pi == NULL || pi->dev == NULL)
1264 notify_siblings(device_t dev, int detaching)
1270 for (i = 0; i < PCI_FUNCMAX; i++) {
1271 if (i == pci_get_function(dev))
1273 sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1274 pci_get_slot(dev), i);
1275 if (sibling == NULL || !device_is_attached(sibling))
1278 error = T4_DETACH_CHILD(sibling);
1280 (void)T4_ATTACH_CHILD(sibling);
1291 t4_detach(device_t dev)
1296 sc = device_get_softc(dev);
1298 rc = notify_siblings(dev, 1);
1301 "failed to detach sibling devices: %d\n", rc);
1305 return (t4_detach_common(dev));
1309 t4_detach_common(device_t dev)
1312 struct port_info *pi;
1315 sc = device_get_softc(dev);
1318 destroy_dev(sc->cdev);
1322 sc->flags &= ~CHK_MBOX_ACCESS;
1323 if (sc->flags & FULL_INIT_DONE) {
1324 if (!(sc->flags & IS_VF))
1325 t4_intr_disable(sc);
1328 if (device_is_attached(dev)) {
1329 rc = bus_generic_detach(dev);
1332 "failed to detach child devices: %d\n", rc);
1337 for (i = 0; i < sc->intr_count; i++)
1338 t4_free_irq(sc, &sc->irq[i]);
1340 if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1341 t4_free_tx_sched(sc);
1343 for (i = 0; i < MAX_NPORTS; i++) {
1346 t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1348 device_delete_child(dev, pi->dev);
1350 mtx_destroy(&pi->pi_lock);
1351 free(pi->vi, M_CXGBE);
1356 device_delete_children(dev);
1358 if (sc->flags & FULL_INIT_DONE)
1359 adapter_full_uninit(sc);
1361 if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1362 t4_fw_bye(sc, sc->mbox);
1364 if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1365 pci_release_msi(dev);
1368 bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1372 bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1376 bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1380 t4_free_l2t(sc->l2t);
1382 t4_free_etid_table(sc);
1385 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1386 free(sc->sge.ofld_txq, M_CXGBE);
1389 free(sc->sge.ofld_rxq, M_CXGBE);
1392 free(sc->sge.nm_rxq, M_CXGBE);
1393 free(sc->sge.nm_txq, M_CXGBE);
1395 free(sc->irq, M_CXGBE);
1396 free(sc->sge.rxq, M_CXGBE);
1397 free(sc->sge.txq, M_CXGBE);
1398 free(sc->sge.ctrlq, M_CXGBE);
1399 free(sc->sge.iqmap, M_CXGBE);
1400 free(sc->sge.eqmap, M_CXGBE);
1401 free(sc->tids.ftid_tab, M_CXGBE);
1402 free(sc->tids.hftid_tab, M_CXGBE);
1403 free(sc->tids.atid_tab, M_CXGBE);
1404 free(sc->tids.tid_tab, M_CXGBE);
1405 free(sc->tt.tls_rx_ports, M_CXGBE);
1406 t4_destroy_dma_tag(sc);
1407 if (mtx_initialized(&sc->sc_lock)) {
1408 sx_xlock(&t4_list_lock);
1409 SLIST_REMOVE(&t4_list, sc, adapter, link);
1410 sx_xunlock(&t4_list_lock);
1411 mtx_destroy(&sc->sc_lock);
1414 callout_drain(&sc->sfl_callout);
1415 if (mtx_initialized(&sc->tids.ftid_lock)) {
1416 mtx_destroy(&sc->tids.ftid_lock);
1417 cv_destroy(&sc->tids.ftid_cv);
1419 if (mtx_initialized(&sc->tids.hftid_lock)) {
1420 mtx_destroy(&sc->tids.hftid_lock);
1421 cv_destroy(&sc->tids.hftid_cv);
1423 if (mtx_initialized(&sc->tids.atid_lock))
1424 mtx_destroy(&sc->tids.atid_lock);
1425 if (mtx_initialized(&sc->sfl_lock))
1426 mtx_destroy(&sc->sfl_lock);
1427 if (mtx_initialized(&sc->ifp_lock))
1428 mtx_destroy(&sc->ifp_lock);
1429 if (mtx_initialized(&sc->reg_lock))
1430 mtx_destroy(&sc->reg_lock);
1432 if (rw_initialized(&sc->policy_lock)) {
1433 rw_destroy(&sc->policy_lock);
1435 if (sc->policy != NULL)
1436 free_offload_policy(sc->policy);
1440 for (i = 0; i < NUM_MEMWIN; i++) {
1441 struct memwin *mw = &sc->memwin[i];
1443 if (rw_initialized(&mw->mw_lock))
1444 rw_destroy(&mw->mw_lock);
1447 bzero(sc, sizeof(*sc));
1453 cxgbe_probe(device_t dev)
1456 struct port_info *pi = device_get_softc(dev);
1458 snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1459 device_set_desc_copy(dev, buf);
1461 return (BUS_PROBE_DEFAULT);
1464 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1465 IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1466 IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS)
1467 #define T4_CAP_ENABLE (T4_CAP)
1470 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1475 vi->xact_addr_filt = -1;
1476 callout_init(&vi->tick, 1);
1478 /* Allocate an ifnet and set it up */
1479 ifp = if_alloc(IFT_ETHER);
1481 device_printf(dev, "Cannot allocate ifnet\n");
1487 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1488 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1490 ifp->if_init = cxgbe_init;
1491 ifp->if_ioctl = cxgbe_ioctl;
1492 ifp->if_transmit = cxgbe_transmit;
1493 ifp->if_qflush = cxgbe_qflush;
1494 ifp->if_get_counter = cxgbe_get_counter;
1496 ifp->if_snd_tag_alloc = cxgbe_snd_tag_alloc;
1497 ifp->if_snd_tag_modify = cxgbe_snd_tag_modify;
1498 ifp->if_snd_tag_query = cxgbe_snd_tag_query;
1499 ifp->if_snd_tag_free = cxgbe_snd_tag_free;
1502 ifp->if_capabilities = T4_CAP;
1504 if (vi->nofldrxq != 0)
1505 ifp->if_capabilities |= IFCAP_TOE;
1508 if (vi->nnmrxq != 0)
1509 ifp->if_capabilities |= IFCAP_NETMAP;
1511 ifp->if_capenable = T4_CAP_ENABLE;
1512 ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1513 CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1515 ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1516 ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1517 ifp->if_hw_tsomaxsegsize = 65536;
1519 vi->vlan_c = EVENTHANDLER_REGISTER(vlan_config, cxgbe_vlan_config, ifp,
1520 EVENTHANDLER_PRI_ANY);
1522 ether_ifattach(ifp, vi->hw_addr);
1524 if (ifp->if_capabilities & IFCAP_NETMAP)
1525 cxgbe_nm_attach(vi);
1527 sb = sbuf_new_auto();
1528 sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1530 if (ifp->if_capabilities & IFCAP_TOE)
1531 sbuf_printf(sb, "; %d txq, %d rxq (TOE)",
1532 vi->nofldtxq, vi->nofldrxq);
1535 if (ifp->if_capabilities & IFCAP_NETMAP)
1536 sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1537 vi->nnmtxq, vi->nnmrxq);
1540 device_printf(dev, "%s\n", sbuf_data(sb));
1549 cxgbe_attach(device_t dev)
1551 struct port_info *pi = device_get_softc(dev);
1552 struct adapter *sc = pi->adapter;
1556 callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1558 rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1562 for_each_vi(pi, i, vi) {
1565 vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1566 if (vi->dev == NULL) {
1567 device_printf(dev, "failed to add VI %d\n", i);
1570 device_set_softc(vi->dev, vi);
1575 bus_generic_attach(dev);
1581 cxgbe_vi_detach(struct vi_info *vi)
1583 struct ifnet *ifp = vi->ifp;
1585 ether_ifdetach(ifp);
1588 EVENTHANDLER_DEREGISTER(vlan_config, vi->vlan_c);
1590 /* Let detach proceed even if these fail. */
1592 if (ifp->if_capabilities & IFCAP_NETMAP)
1593 cxgbe_nm_detach(vi);
1595 cxgbe_uninit_synchronized(vi);
1596 callout_drain(&vi->tick);
1604 cxgbe_detach(device_t dev)
1606 struct port_info *pi = device_get_softc(dev);
1607 struct adapter *sc = pi->adapter;
1610 /* Detach the extra VIs first. */
1611 rc = bus_generic_detach(dev);
1614 device_delete_children(dev);
1616 doom_vi(sc, &pi->vi[0]);
1618 if (pi->flags & HAS_TRACEQ) {
1619 sc->traceq = -1; /* cloner should not create ifnet */
1620 t4_tracer_port_detach(sc);
1623 cxgbe_vi_detach(&pi->vi[0]);
1624 callout_drain(&pi->tick);
1625 ifmedia_removeall(&pi->media);
1627 end_synchronized_op(sc, 0);
1633 cxgbe_init(void *arg)
1635 struct vi_info *vi = arg;
1636 struct adapter *sc = vi->pi->adapter;
1638 if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1640 cxgbe_init_synchronized(vi);
1641 end_synchronized_op(sc, 0);
1645 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1647 int rc = 0, mtu, flags, can_sleep;
1648 struct vi_info *vi = ifp->if_softc;
1649 struct port_info *pi = vi->pi;
1650 struct adapter *sc = pi->adapter;
1651 struct ifreq *ifr = (struct ifreq *)data;
1657 if (mtu < ETHERMIN || mtu > MAX_MTU)
1660 rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1664 if (vi->flags & VI_INIT_DONE) {
1665 t4_update_fl_bufsize(ifp);
1666 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1667 rc = update_mac_settings(ifp, XGMAC_MTU);
1669 end_synchronized_op(sc, 0);
1675 rc = begin_synchronized_op(sc, vi,
1676 can_sleep ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4flg");
1678 if_printf(ifp, "%ssleepable synch operation failed: %d."
1679 " if_flags 0x%08x, if_drv_flags 0x%08x\n",
1680 can_sleep ? "" : "non-", rc, ifp->if_flags,
1685 if (ifp->if_flags & IFF_UP) {
1686 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1687 flags = vi->if_flags;
1688 if ((ifp->if_flags ^ flags) &
1689 (IFF_PROMISC | IFF_ALLMULTI)) {
1690 if (can_sleep == 1) {
1691 end_synchronized_op(sc, 0);
1695 rc = update_mac_settings(ifp,
1696 XGMAC_PROMISC | XGMAC_ALLMULTI);
1699 if (can_sleep == 0) {
1700 end_synchronized_op(sc, LOCK_HELD);
1704 rc = cxgbe_init_synchronized(vi);
1706 vi->if_flags = ifp->if_flags;
1707 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1708 if (can_sleep == 0) {
1709 end_synchronized_op(sc, LOCK_HELD);
1713 rc = cxgbe_uninit_synchronized(vi);
1715 end_synchronized_op(sc, can_sleep ? 0 : LOCK_HELD);
1719 case SIOCDELMULTI: /* these two are called with a mutex held :-( */
1720 rc = begin_synchronized_op(sc, vi, HOLD_LOCK, "t4multi");
1723 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1724 rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1725 end_synchronized_op(sc, LOCK_HELD);
1729 rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1733 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1734 if (mask & IFCAP_TXCSUM) {
1735 ifp->if_capenable ^= IFCAP_TXCSUM;
1736 ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1738 if (IFCAP_TSO4 & ifp->if_capenable &&
1739 !(IFCAP_TXCSUM & ifp->if_capenable)) {
1740 ifp->if_capenable &= ~IFCAP_TSO4;
1742 "tso4 disabled due to -txcsum.\n");
1745 if (mask & IFCAP_TXCSUM_IPV6) {
1746 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1747 ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1749 if (IFCAP_TSO6 & ifp->if_capenable &&
1750 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1751 ifp->if_capenable &= ~IFCAP_TSO6;
1753 "tso6 disabled due to -txcsum6.\n");
1756 if (mask & IFCAP_RXCSUM)
1757 ifp->if_capenable ^= IFCAP_RXCSUM;
1758 if (mask & IFCAP_RXCSUM_IPV6)
1759 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1762 * Note that we leave CSUM_TSO alone (it is always set). The
1763 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1764 * sending a TSO request our way, so it's sufficient to toggle
1767 if (mask & IFCAP_TSO4) {
1768 if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1769 !(IFCAP_TXCSUM & ifp->if_capenable)) {
1770 if_printf(ifp, "enable txcsum first.\n");
1774 ifp->if_capenable ^= IFCAP_TSO4;
1776 if (mask & IFCAP_TSO6) {
1777 if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1778 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1779 if_printf(ifp, "enable txcsum6 first.\n");
1783 ifp->if_capenable ^= IFCAP_TSO6;
1785 if (mask & IFCAP_LRO) {
1786 #if defined(INET) || defined(INET6)
1788 struct sge_rxq *rxq;
1790 ifp->if_capenable ^= IFCAP_LRO;
1791 for_each_rxq(vi, i, rxq) {
1792 if (ifp->if_capenable & IFCAP_LRO)
1793 rxq->iq.flags |= IQ_LRO_ENABLED;
1795 rxq->iq.flags &= ~IQ_LRO_ENABLED;
1800 if (mask & IFCAP_TOE) {
1801 int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1803 rc = toe_capability(vi, enable);
1807 ifp->if_capenable ^= mask;
1810 if (mask & IFCAP_VLAN_HWTAGGING) {
1811 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1812 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1813 rc = update_mac_settings(ifp, XGMAC_VLANEX);
1815 if (mask & IFCAP_VLAN_MTU) {
1816 ifp->if_capenable ^= IFCAP_VLAN_MTU;
1818 /* Need to find out how to disable auto-mtu-inflation */
1820 if (mask & IFCAP_VLAN_HWTSO)
1821 ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1822 if (mask & IFCAP_VLAN_HWCSUM)
1823 ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1825 #ifdef VLAN_CAPABILITIES
1826 VLAN_CAPABILITIES(ifp);
1829 end_synchronized_op(sc, 0);
1835 ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
1839 struct ifi2creq i2c;
1841 rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
1844 if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1848 if (i2c.len > sizeof(i2c.data)) {
1852 rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
1855 rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
1856 i2c.offset, i2c.len, &i2c.data[0]);
1857 end_synchronized_op(sc, 0);
1859 rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
1864 rc = ether_ioctl(ifp, cmd, data);
1871 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1873 struct vi_info *vi = ifp->if_softc;
1874 struct port_info *pi = vi->pi;
1875 struct adapter *sc = pi->adapter;
1876 struct sge_txq *txq;
1881 MPASS(m->m_nextpkt == NULL); /* not quite ready for this yet */
1883 if (__predict_false(pi->link_cfg.link_ok == 0)) {
1888 rc = parse_pkt(sc, &m);
1889 if (__predict_false(rc != 0)) {
1890 MPASS(m == NULL); /* was freed already */
1891 atomic_add_int(&pi->tx_parse_error, 1); /* rare, atomic is ok */
1895 if (m->m_pkthdr.snd_tag != NULL) {
1896 /* EAGAIN tells the stack we are not the correct interface. */
1897 if (__predict_false(ifp != m->m_pkthdr.snd_tag->ifp)) {
1902 return (ethofld_transmit(ifp, m));
1907 txq = &sc->sge.txq[vi->first_txq];
1908 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1909 txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
1913 rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1914 if (__predict_false(rc != 0))
1921 cxgbe_qflush(struct ifnet *ifp)
1923 struct vi_info *vi = ifp->if_softc;
1924 struct sge_txq *txq;
1927 /* queues do not exist if !VI_INIT_DONE. */
1928 if (vi->flags & VI_INIT_DONE) {
1929 for_each_txq(vi, i, txq) {
1931 txq->eq.flags |= EQ_QFLUSH;
1933 while (!mp_ring_is_idle(txq->r)) {
1934 mp_ring_check_drainage(txq->r, 0);
1938 txq->eq.flags &= ~EQ_QFLUSH;
1946 vi_get_counter(struct ifnet *ifp, ift_counter c)
1948 struct vi_info *vi = ifp->if_softc;
1949 struct fw_vi_stats_vf *s = &vi->stats;
1951 vi_refresh_stats(vi->pi->adapter, vi);
1954 case IFCOUNTER_IPACKETS:
1955 return (s->rx_bcast_frames + s->rx_mcast_frames +
1956 s->rx_ucast_frames);
1957 case IFCOUNTER_IERRORS:
1958 return (s->rx_err_frames);
1959 case IFCOUNTER_OPACKETS:
1960 return (s->tx_bcast_frames + s->tx_mcast_frames +
1961 s->tx_ucast_frames + s->tx_offload_frames);
1962 case IFCOUNTER_OERRORS:
1963 return (s->tx_drop_frames);
1964 case IFCOUNTER_IBYTES:
1965 return (s->rx_bcast_bytes + s->rx_mcast_bytes +
1967 case IFCOUNTER_OBYTES:
1968 return (s->tx_bcast_bytes + s->tx_mcast_bytes +
1969 s->tx_ucast_bytes + s->tx_offload_bytes);
1970 case IFCOUNTER_IMCASTS:
1971 return (s->rx_mcast_frames);
1972 case IFCOUNTER_OMCASTS:
1973 return (s->tx_mcast_frames);
1974 case IFCOUNTER_OQDROPS: {
1978 if (vi->flags & VI_INIT_DONE) {
1980 struct sge_txq *txq;
1982 for_each_txq(vi, i, txq)
1983 drops += counter_u64_fetch(txq->r->drops);
1991 return (if_get_counter_default(ifp, c));
1996 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
1998 struct vi_info *vi = ifp->if_softc;
1999 struct port_info *pi = vi->pi;
2000 struct adapter *sc = pi->adapter;
2001 struct port_stats *s = &pi->stats;
2003 if (pi->nvi > 1 || sc->flags & IS_VF)
2004 return (vi_get_counter(ifp, c));
2006 cxgbe_refresh_stats(sc, pi);
2009 case IFCOUNTER_IPACKETS:
2010 return (s->rx_frames);
2012 case IFCOUNTER_IERRORS:
2013 return (s->rx_jabber + s->rx_runt + s->rx_too_long +
2014 s->rx_fcs_err + s->rx_len_err);
2016 case IFCOUNTER_OPACKETS:
2017 return (s->tx_frames);
2019 case IFCOUNTER_OERRORS:
2020 return (s->tx_error_frames);
2022 case IFCOUNTER_IBYTES:
2023 return (s->rx_octets);
2025 case IFCOUNTER_OBYTES:
2026 return (s->tx_octets);
2028 case IFCOUNTER_IMCASTS:
2029 return (s->rx_mcast_frames);
2031 case IFCOUNTER_OMCASTS:
2032 return (s->tx_mcast_frames);
2034 case IFCOUNTER_IQDROPS:
2035 return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
2036 s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
2037 s->rx_trunc3 + pi->tnl_cong_drops);
2039 case IFCOUNTER_OQDROPS: {
2043 if (vi->flags & VI_INIT_DONE) {
2045 struct sge_txq *txq;
2047 for_each_txq(vi, i, txq)
2048 drops += counter_u64_fetch(txq->r->drops);
2056 return (if_get_counter_default(ifp, c));
2061 cxgbe_media_change(struct ifnet *ifp)
2063 struct vi_info *vi = ifp->if_softc;
2065 device_printf(vi->dev, "%s unimplemented.\n", __func__);
2067 return (EOPNOTSUPP);
2071 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2073 struct vi_info *vi = ifp->if_softc;
2074 struct port_info *pi = vi->pi;
2075 struct ifmedia_entry *cur;
2076 struct link_config *lc = &pi->link_cfg;
2079 * If all the interfaces are administratively down the firmware does not
2080 * report transceiver changes. Refresh port info here so that ifconfig
2081 * displays accurate information at all times.
2083 if (begin_synchronized_op(pi->adapter, NULL, SLEEP_OK | INTR_OK,
2086 if (pi->up_vis == 0) {
2087 t4_update_port_info(pi);
2088 build_medialist(pi, &pi->media);
2091 end_synchronized_op(pi->adapter, 0);
2094 ifmr->ifm_status = IFM_AVALID;
2095 if (lc->link_ok == 0)
2098 ifmr->ifm_status |= IFM_ACTIVE;
2099 ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2100 if (lc->fc & PAUSE_RX)
2101 ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2102 if (lc->fc & PAUSE_TX)
2103 ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2105 /* active and current will differ iff current media is autoselect. */
2106 cur = pi->media.ifm_cur;
2107 if (cur != NULL && IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
2110 ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2111 if (lc->fc & PAUSE_RX)
2112 ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2113 if (lc->fc & PAUSE_TX)
2114 ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2115 switch (lc->speed) {
2117 ifmr->ifm_active |= IFM_10G_T;
2120 ifmr->ifm_active |= IFM_1000_T;
2123 ifmr->ifm_active |= IFM_100_TX;
2126 ifmr->ifm_active |= IFM_10_T;
2129 device_printf(vi->dev, "link up but speed unknown (%u)\n",
2135 vcxgbe_probe(device_t dev)
2138 struct vi_info *vi = device_get_softc(dev);
2140 snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2142 device_set_desc_copy(dev, buf);
2144 return (BUS_PROBE_DEFAULT);
2148 alloc_extra_vi(struct adapter *sc, struct port_info *pi, struct vi_info *vi)
2150 int func, index, rc;
2151 uint32_t param, val;
2153 ASSERT_SYNCHRONIZED_OP(sc);
2155 index = vi - pi->vi;
2156 MPASS(index > 0); /* This function deals with _extra_ VIs only */
2157 KASSERT(index < nitems(vi_mac_funcs),
2158 ("%s: VI %s doesn't have a MAC func", __func__,
2159 device_get_nameunit(vi->dev)));
2160 func = vi_mac_funcs[index];
2161 rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2162 vi->hw_addr, &vi->rss_size, func, 0);
2164 device_printf(vi->dev, "failed to allocate virtual interface %d"
2165 "for port %d: %d\n", index, pi->port_id, -rc);
2169 if (chip_id(sc) <= CHELSIO_T5)
2170 vi->smt_idx = (rc & 0x7f) << 1;
2172 vi->smt_idx = (rc & 0x7f);
2174 if (vi->rss_size == 1) {
2176 * This VI didn't get a slice of the RSS table. Reduce the
2177 * number of VIs being created (hw.cxgbe.num_vis) or modify the
2178 * configuration file (nvi, rssnvi for this PF) if this is a
2181 device_printf(vi->dev, "RSS table not available.\n");
2182 vi->rss_base = 0xffff;
2187 param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2188 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2189 V_FW_PARAMS_PARAM_YZ(vi->viid);
2190 rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, ¶m, &val);
2192 vi->rss_base = 0xffff;
2194 MPASS((val >> 16) == vi->rss_size);
2195 vi->rss_base = val & 0xffff;
2202 vcxgbe_attach(device_t dev)
2205 struct port_info *pi;
2209 vi = device_get_softc(dev);
2213 rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4via");
2216 rc = alloc_extra_vi(sc, pi, vi);
2217 end_synchronized_op(sc, 0);
2221 rc = cxgbe_vi_attach(dev, vi);
2223 t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2230 vcxgbe_detach(device_t dev)
2235 vi = device_get_softc(dev);
2236 sc = vi->pi->adapter;
2240 cxgbe_vi_detach(vi);
2241 t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2243 end_synchronized_op(sc, 0);
2249 t4_fatal_err(struct adapter *sc)
2251 t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2252 t4_intr_disable(sc);
2253 log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
2254 device_get_nameunit(sc->dev));
2255 if (t4_panic_on_fatal_err)
2256 panic("panic requested on fatal error");
2260 t4_add_adapter(struct adapter *sc)
2262 sx_xlock(&t4_list_lock);
2263 SLIST_INSERT_HEAD(&t4_list, sc, link);
2264 sx_xunlock(&t4_list_lock);
2268 t4_map_bars_0_and_4(struct adapter *sc)
2270 sc->regs_rid = PCIR_BAR(0);
2271 sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2272 &sc->regs_rid, RF_ACTIVE);
2273 if (sc->regs_res == NULL) {
2274 device_printf(sc->dev, "cannot map registers.\n");
2277 sc->bt = rman_get_bustag(sc->regs_res);
2278 sc->bh = rman_get_bushandle(sc->regs_res);
2279 sc->mmio_len = rman_get_size(sc->regs_res);
2280 setbit(&sc->doorbells, DOORBELL_KDB);
2282 sc->msix_rid = PCIR_BAR(4);
2283 sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2284 &sc->msix_rid, RF_ACTIVE);
2285 if (sc->msix_res == NULL) {
2286 device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2294 t4_map_bar_2(struct adapter *sc)
2298 * T4: only iWARP driver uses the userspace doorbells. There is no need
2299 * to map it if RDMA is disabled.
2301 if (is_t4(sc) && sc->rdmacaps == 0)
2304 sc->udbs_rid = PCIR_BAR(2);
2305 sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2306 &sc->udbs_rid, RF_ACTIVE);
2307 if (sc->udbs_res == NULL) {
2308 device_printf(sc->dev, "cannot map doorbell BAR.\n");
2311 sc->udbs_base = rman_get_virtual(sc->udbs_res);
2313 if (chip_id(sc) >= CHELSIO_T5) {
2314 setbit(&sc->doorbells, DOORBELL_UDB);
2315 #if defined(__i386__) || defined(__amd64__)
2316 if (t5_write_combine) {
2320 * Enable write combining on BAR2. This is the
2321 * userspace doorbell BAR and is split into 128B
2322 * (UDBS_SEG_SIZE) doorbell regions, each associated
2323 * with an egress queue. The first 64B has the doorbell
2324 * and the second 64B can be used to submit a tx work
2325 * request with an implicit doorbell.
2328 rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2329 rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2331 clrbit(&sc->doorbells, DOORBELL_UDB);
2332 setbit(&sc->doorbells, DOORBELL_WCWR);
2333 setbit(&sc->doorbells, DOORBELL_UDBWC);
2335 device_printf(sc->dev,
2336 "couldn't enable write combining: %d\n",
2340 mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2341 t4_write_reg(sc, A_SGE_STAT_CFG,
2342 V_STATSOURCE_T5(7) | mode);
2346 sc->iwt.wc_en = isset(&sc->doorbells, DOORBELL_UDBWC) ? 1 : 0;
2351 struct memwin_init {
2356 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2357 { MEMWIN0_BASE, MEMWIN0_APERTURE },
2358 { MEMWIN1_BASE, MEMWIN1_APERTURE },
2359 { MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2362 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2363 { MEMWIN0_BASE, MEMWIN0_APERTURE },
2364 { MEMWIN1_BASE, MEMWIN1_APERTURE },
2365 { MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2369 setup_memwin(struct adapter *sc)
2371 const struct memwin_init *mw_init;
2378 * Read low 32b of bar0 indirectly via the hardware backdoor
2379 * mechanism. Works from within PCI passthrough environments
2380 * too, where rman_get_start() can return a different value. We
2381 * need to program the T4 memory window decoders with the actual
2382 * addresses that will be coming across the PCIe link.
2384 bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2385 bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2387 mw_init = &t4_memwin[0];
2389 /* T5+ use the relative offset inside the PCIe BAR */
2392 mw_init = &t5_memwin[0];
2395 for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2396 rw_init(&mw->mw_lock, "memory window access");
2397 mw->mw_base = mw_init->base;
2398 mw->mw_aperture = mw_init->aperture;
2401 PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2402 (mw->mw_base + bar0) | V_BIR(0) |
2403 V_WINDOW(ilog2(mw->mw_aperture) - 10));
2404 rw_wlock(&mw->mw_lock);
2405 position_memwin(sc, i, 0);
2406 rw_wunlock(&mw->mw_lock);
2410 t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2414 * Positions the memory window at the given address in the card's address space.
2415 * There are some alignment requirements and the actual position may be at an
2416 * address prior to the requested address. mw->mw_curpos always has the actual
2417 * position of the window.
2420 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2426 MPASS(idx >= 0 && idx < NUM_MEMWIN);
2427 mw = &sc->memwin[idx];
2428 rw_assert(&mw->mw_lock, RA_WLOCKED);
2432 mw->mw_curpos = addr & ~0xf; /* start must be 16B aligned */
2434 pf = V_PFNUM(sc->pf);
2435 mw->mw_curpos = addr & ~0x7f; /* start must be 128B aligned */
2437 reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2438 t4_write_reg(sc, reg, mw->mw_curpos | pf);
2439 t4_read_reg(sc, reg); /* flush */
2443 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2449 MPASS(idx >= 0 && idx < NUM_MEMWIN);
2451 /* Memory can only be accessed in naturally aligned 4 byte units */
2452 if (addr & 3 || len & 3 || len <= 0)
2455 mw = &sc->memwin[idx];
2457 rw_rlock(&mw->mw_lock);
2458 mw_end = mw->mw_curpos + mw->mw_aperture;
2459 if (addr >= mw_end || addr < mw->mw_curpos) {
2460 /* Will need to reposition the window */
2461 if (!rw_try_upgrade(&mw->mw_lock)) {
2462 rw_runlock(&mw->mw_lock);
2463 rw_wlock(&mw->mw_lock);
2465 rw_assert(&mw->mw_lock, RA_WLOCKED);
2466 position_memwin(sc, idx, addr);
2467 rw_downgrade(&mw->mw_lock);
2468 mw_end = mw->mw_curpos + mw->mw_aperture;
2470 rw_assert(&mw->mw_lock, RA_RLOCKED);
2471 while (addr < mw_end && len > 0) {
2473 v = t4_read_reg(sc, mw->mw_base + addr -
2475 *val++ = le32toh(v);
2478 t4_write_reg(sc, mw->mw_base + addr -
2479 mw->mw_curpos, htole32(v));
2484 rw_runlock(&mw->mw_lock);
2491 alloc_atid_tab(struct tid_info *t, int flags)
2495 MPASS(t->natids > 0);
2496 MPASS(t->atid_tab == NULL);
2498 t->atid_tab = malloc(t->natids * sizeof(*t->atid_tab), M_CXGBE,
2500 if (t->atid_tab == NULL)
2502 mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
2503 t->afree = t->atid_tab;
2504 t->atids_in_use = 0;
2505 for (i = 1; i < t->natids; i++)
2506 t->atid_tab[i - 1].next = &t->atid_tab[i];
2507 t->atid_tab[t->natids - 1].next = NULL;
2513 free_atid_tab(struct tid_info *t)
2516 KASSERT(t->atids_in_use == 0,
2517 ("%s: %d atids still in use.", __func__, t->atids_in_use));
2519 if (mtx_initialized(&t->atid_lock))
2520 mtx_destroy(&t->atid_lock);
2521 free(t->atid_tab, M_CXGBE);
2526 alloc_atid(struct adapter *sc, void *ctx)
2528 struct tid_info *t = &sc->tids;
2531 mtx_lock(&t->atid_lock);
2533 union aopen_entry *p = t->afree;
2535 atid = p - t->atid_tab;
2536 MPASS(atid <= M_TID_TID);
2541 mtx_unlock(&t->atid_lock);
2546 lookup_atid(struct adapter *sc, int atid)
2548 struct tid_info *t = &sc->tids;
2550 return (t->atid_tab[atid].data);
2554 free_atid(struct adapter *sc, int atid)
2556 struct tid_info *t = &sc->tids;
2557 union aopen_entry *p = &t->atid_tab[atid];
2559 mtx_lock(&t->atid_lock);
2563 mtx_unlock(&t->atid_lock);
2567 queue_tid_release(struct adapter *sc, int tid)
2570 CXGBE_UNIMPLEMENTED("deferred tid release");
2574 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
2577 struct cpl_tid_release *req;
2579 wr = alloc_wrqe(sizeof(*req), ctrlq);
2581 queue_tid_release(sc, tid); /* defer */
2586 INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
2592 t4_range_cmp(const void *a, const void *b)
2594 return ((const struct t4_range *)a)->start -
2595 ((const struct t4_range *)b)->start;
2599 * Verify that the memory range specified by the addr/len pair is valid within
2600 * the card's address space.
2603 validate_mem_range(struct adapter *sc, uint32_t addr, int len)
2605 struct t4_range mem_ranges[4], *r, *next;
2606 uint32_t em, addr_len;
2607 int i, n, remaining;
2609 /* Memory can only be accessed in naturally aligned 4 byte units */
2610 if (addr & 3 || len & 3 || len <= 0)
2613 /* Enabled memories */
2614 em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2618 bzero(r, sizeof(mem_ranges));
2619 if (em & F_EDRAM0_ENABLE) {
2620 addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2621 r->size = G_EDRAM0_SIZE(addr_len) << 20;
2623 r->start = G_EDRAM0_BASE(addr_len) << 20;
2624 if (addr >= r->start &&
2625 addr + len <= r->start + r->size)
2631 if (em & F_EDRAM1_ENABLE) {
2632 addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2633 r->size = G_EDRAM1_SIZE(addr_len) << 20;
2635 r->start = G_EDRAM1_BASE(addr_len) << 20;
2636 if (addr >= r->start &&
2637 addr + len <= r->start + r->size)
2643 if (em & F_EXT_MEM_ENABLE) {
2644 addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2645 r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2647 r->start = G_EXT_MEM_BASE(addr_len) << 20;
2648 if (addr >= r->start &&
2649 addr + len <= r->start + r->size)
2655 if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2656 addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2657 r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2659 r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2660 if (addr >= r->start &&
2661 addr + len <= r->start + r->size)
2667 MPASS(n <= nitems(mem_ranges));
2670 /* Sort and merge the ranges. */
2671 qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2673 /* Start from index 0 and examine the next n - 1 entries. */
2675 for (remaining = n - 1; remaining > 0; remaining--, r++) {
2677 MPASS(r->size > 0); /* r is a valid entry. */
2679 MPASS(next->size > 0); /* and so is the next one. */
2681 while (r->start + r->size >= next->start) {
2682 /* Merge the next one into the current entry. */
2683 r->size = max(r->start + r->size,
2684 next->start + next->size) - r->start;
2685 n--; /* One fewer entry in total. */
2686 if (--remaining == 0)
2687 goto done; /* short circuit */
2690 if (next != r + 1) {
2692 * Some entries were merged into r and next
2693 * points to the first valid entry that couldn't
2696 MPASS(next->size > 0); /* must be valid */
2697 memcpy(r + 1, next, remaining * sizeof(*r));
2700 * This so that the foo->size assertion in the
2701 * next iteration of the loop do the right
2702 * thing for entries that were pulled up and are
2705 MPASS(n < nitems(mem_ranges));
2706 bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
2707 sizeof(struct t4_range));
2712 /* Done merging the ranges. */
2715 for (i = 0; i < n; i++, r++) {
2716 if (addr >= r->start &&
2717 addr + len <= r->start + r->size)
2726 fwmtype_to_hwmtype(int mtype)
2730 case FW_MEMTYPE_EDC0:
2732 case FW_MEMTYPE_EDC1:
2734 case FW_MEMTYPE_EXTMEM:
2736 case FW_MEMTYPE_EXTMEM1:
2739 panic("%s: cannot translate fw mtype %d.", __func__, mtype);
2744 * Verify that the memory range specified by the memtype/offset/len pair is
2745 * valid and lies entirely within the memtype specified. The global address of
2746 * the start of the range is returned in addr.
2749 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
2752 uint32_t em, addr_len, maddr;
2754 /* Memory can only be accessed in naturally aligned 4 byte units */
2755 if (off & 3 || len & 3 || len == 0)
2758 em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2759 switch (fwmtype_to_hwmtype(mtype)) {
2761 if (!(em & F_EDRAM0_ENABLE))
2763 addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2764 maddr = G_EDRAM0_BASE(addr_len) << 20;
2767 if (!(em & F_EDRAM1_ENABLE))
2769 addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2770 maddr = G_EDRAM1_BASE(addr_len) << 20;
2773 if (!(em & F_EXT_MEM_ENABLE))
2775 addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2776 maddr = G_EXT_MEM_BASE(addr_len) << 20;
2779 if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
2781 addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2782 maddr = G_EXT_MEM1_BASE(addr_len) << 20;
2788 *addr = maddr + off; /* global address */
2789 return (validate_mem_range(sc, *addr, len));
2793 fixup_devlog_params(struct adapter *sc)
2795 struct devlog_params *dparams = &sc->params.devlog;
2798 rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
2799 dparams->size, &dparams->addr);
2805 update_nirq(struct intrs_and_queues *iaq, int nports)
2807 int extra = T4_EXTRA_INTR;
2810 iaq->nirq += nports * (iaq->nrxq + iaq->nofldrxq);
2811 iaq->nirq += nports * (iaq->num_vis - 1) *
2812 max(iaq->nrxq_vi, iaq->nnmrxq_vi);
2813 iaq->nirq += nports * (iaq->num_vis - 1) * iaq->nofldrxq_vi;
2817 * Adjust requirements to fit the number of interrupts available.
2820 calculate_iaq(struct adapter *sc, struct intrs_and_queues *iaq, int itype,
2824 const int nports = sc->params.nports;
2829 bzero(iaq, sizeof(*iaq));
2830 iaq->intr_type = itype;
2831 iaq->num_vis = t4_num_vis;
2832 iaq->ntxq = t4_ntxq;
2833 iaq->ntxq_vi = t4_ntxq_vi;
2834 iaq->nrxq = t4_nrxq;
2835 iaq->nrxq_vi = t4_nrxq_vi;
2836 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
2837 if (is_offload(sc) || is_ethoffload(sc)) {
2838 iaq->nofldtxq = t4_nofldtxq;
2839 iaq->nofldtxq_vi = t4_nofldtxq_vi;
2843 if (is_offload(sc)) {
2844 iaq->nofldrxq = t4_nofldrxq;
2845 iaq->nofldrxq_vi = t4_nofldrxq_vi;
2849 iaq->nnmtxq_vi = t4_nnmtxq_vi;
2850 iaq->nnmrxq_vi = t4_nnmrxq_vi;
2853 update_nirq(iaq, nports);
2854 if (iaq->nirq <= navail &&
2855 (itype != INTR_MSI || powerof2(iaq->nirq))) {
2857 * This is the normal case -- there are enough interrupts for
2864 * If extra VIs have been configured try reducing their count and see if
2867 while (iaq->num_vis > 1) {
2869 update_nirq(iaq, nports);
2870 if (iaq->nirq <= navail &&
2871 (itype != INTR_MSI || powerof2(iaq->nirq))) {
2872 device_printf(sc->dev, "virtual interfaces per port "
2873 "reduced to %d from %d. nrxq=%u, nofldrxq=%u, "
2874 "nrxq_vi=%u nofldrxq_vi=%u, nnmrxq_vi=%u. "
2875 "itype %d, navail %u, nirq %d.\n",
2876 iaq->num_vis, t4_num_vis, iaq->nrxq, iaq->nofldrxq,
2877 iaq->nrxq_vi, iaq->nofldrxq_vi, iaq->nnmrxq_vi,
2878 itype, navail, iaq->nirq);
2884 * Extra VIs will not be created. Log a message if they were requested.
2886 MPASS(iaq->num_vis == 1);
2887 iaq->ntxq_vi = iaq->nrxq_vi = 0;
2888 iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
2889 iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
2890 if (iaq->num_vis != t4_num_vis) {
2891 device_printf(sc->dev, "extra virtual interfaces disabled. "
2892 "nrxq=%u, nofldrxq=%u, nrxq_vi=%u nofldrxq_vi=%u, "
2893 "nnmrxq_vi=%u. itype %d, navail %u, nirq %d.\n",
2894 iaq->nrxq, iaq->nofldrxq, iaq->nrxq_vi, iaq->nofldrxq_vi,
2895 iaq->nnmrxq_vi, itype, navail, iaq->nirq);
2899 * Keep reducing the number of NIC rx queues to the next lower power of
2900 * 2 (for even RSS distribution) and halving the TOE rx queues and see
2904 if (iaq->nrxq > 1) {
2907 } while (!powerof2(iaq->nrxq));
2909 if (iaq->nofldrxq > 1)
2910 iaq->nofldrxq >>= 1;
2912 old_nirq = iaq->nirq;
2913 update_nirq(iaq, nports);
2914 if (iaq->nirq <= navail &&
2915 (itype != INTR_MSI || powerof2(iaq->nirq))) {
2916 device_printf(sc->dev, "running with reduced number of "
2917 "rx queues because of shortage of interrupts. "
2918 "nrxq=%u, nofldrxq=%u. "
2919 "itype %d, navail %u, nirq %d.\n", iaq->nrxq,
2920 iaq->nofldrxq, itype, navail, iaq->nirq);
2923 } while (old_nirq != iaq->nirq);
2925 /* One interrupt for everything. Ugh. */
2926 device_printf(sc->dev, "running with minimal number of queues. "
2927 "itype %d, navail %u.\n", itype, navail);
2929 MPASS(iaq->nrxq == 1);
2931 if (iaq->nofldrxq > 1)
2934 MPASS(iaq->num_vis > 0);
2935 if (iaq->num_vis > 1) {
2936 MPASS(iaq->nrxq_vi > 0);
2937 MPASS(iaq->ntxq_vi > 0);
2939 MPASS(iaq->nirq > 0);
2940 MPASS(iaq->nrxq > 0);
2941 MPASS(iaq->ntxq > 0);
2942 if (itype == INTR_MSI) {
2943 MPASS(powerof2(iaq->nirq));
2948 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
2950 int rc, itype, navail, nalloc;
2952 for (itype = INTR_MSIX; itype; itype >>= 1) {
2954 if ((itype & t4_intr_types) == 0)
2955 continue; /* not allowed */
2957 if (itype == INTR_MSIX)
2958 navail = pci_msix_count(sc->dev);
2959 else if (itype == INTR_MSI)
2960 navail = pci_msi_count(sc->dev);
2967 calculate_iaq(sc, iaq, itype, navail);
2970 if (itype == INTR_MSIX)
2971 rc = pci_alloc_msix(sc->dev, &nalloc);
2972 else if (itype == INTR_MSI)
2973 rc = pci_alloc_msi(sc->dev, &nalloc);
2975 if (rc == 0 && nalloc > 0) {
2976 if (nalloc == iaq->nirq)
2980 * Didn't get the number requested. Use whatever number
2981 * the kernel is willing to allocate.
2983 device_printf(sc->dev, "fewer vectors than requested, "
2984 "type=%d, req=%d, rcvd=%d; will downshift req.\n",
2985 itype, iaq->nirq, nalloc);
2986 pci_release_msi(sc->dev);
2991 device_printf(sc->dev,
2992 "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
2993 itype, rc, iaq->nirq, nalloc);
2996 device_printf(sc->dev,
2997 "failed to find a usable interrupt type. "
2998 "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
2999 pci_msix_count(sc->dev), pci_msi_count(sc->dev));
3004 #define FW_VERSION(chip) ( \
3005 V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
3006 V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
3007 V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
3008 V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
3009 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
3015 struct fw_hdr fw_hdr; /* XXX: waste of space, need a sparse struct */
3019 .kld_name = "t4fw_cfg",
3020 .fw_mod_name = "t4fw",
3022 .chip = FW_HDR_CHIP_T4,
3023 .fw_ver = htobe32_const(FW_VERSION(T4)),
3024 .intfver_nic = FW_INTFVER(T4, NIC),
3025 .intfver_vnic = FW_INTFVER(T4, VNIC),
3026 .intfver_ofld = FW_INTFVER(T4, OFLD),
3027 .intfver_ri = FW_INTFVER(T4, RI),
3028 .intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
3029 .intfver_iscsi = FW_INTFVER(T4, ISCSI),
3030 .intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
3031 .intfver_fcoe = FW_INTFVER(T4, FCOE),
3035 .kld_name = "t5fw_cfg",
3036 .fw_mod_name = "t5fw",
3038 .chip = FW_HDR_CHIP_T5,
3039 .fw_ver = htobe32_const(FW_VERSION(T5)),
3040 .intfver_nic = FW_INTFVER(T5, NIC),
3041 .intfver_vnic = FW_INTFVER(T5, VNIC),
3042 .intfver_ofld = FW_INTFVER(T5, OFLD),
3043 .intfver_ri = FW_INTFVER(T5, RI),
3044 .intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
3045 .intfver_iscsi = FW_INTFVER(T5, ISCSI),
3046 .intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
3047 .intfver_fcoe = FW_INTFVER(T5, FCOE),
3051 .kld_name = "t6fw_cfg",
3052 .fw_mod_name = "t6fw",
3054 .chip = FW_HDR_CHIP_T6,
3055 .fw_ver = htobe32_const(FW_VERSION(T6)),
3056 .intfver_nic = FW_INTFVER(T6, NIC),
3057 .intfver_vnic = FW_INTFVER(T6, VNIC),
3058 .intfver_ofld = FW_INTFVER(T6, OFLD),
3059 .intfver_ri = FW_INTFVER(T6, RI),
3060 .intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3061 .intfver_iscsi = FW_INTFVER(T6, ISCSI),
3062 .intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3063 .intfver_fcoe = FW_INTFVER(T6, FCOE),
3068 static struct fw_info *
3069 find_fw_info(int chip)
3073 for (i = 0; i < nitems(fw_info); i++) {
3074 if (fw_info[i].chip == chip)
3075 return (&fw_info[i]);
3081 * Is the given firmware API compatible with the one the driver was compiled
3085 fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
3088 /* short circuit if it's the exact same firmware version */
3089 if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
3093 * XXX: Is this too conservative? Perhaps I should limit this to the
3094 * features that are supported in the driver.
3096 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
3097 if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
3098 SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
3099 SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
3107 * The firmware in the KLD is usable, but should it be installed? This routine
3108 * explains itself in detail if it indicates the KLD firmware should be
3112 should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
3116 if (!card_fw_usable) {
3117 reason = "incompatible or unusable";
3122 reason = "older than the version bundled with this driver";
3126 if (t4_fw_install == 2 && k != c) {
3127 reason = "different than the version bundled with this driver";
3134 if (t4_fw_install == 0) {
3135 device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3136 "but the driver is prohibited from installing a different "
3137 "firmware on the card.\n",
3138 G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3139 G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
3144 device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3145 "installing firmware %u.%u.%u.%u on card.\n",
3146 G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3147 G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
3148 G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3149 G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3155 * Establish contact with the firmware and determine if we are the master driver
3156 * or not, and whether we are responsible for chip initialization.
3159 prep_firmware(struct adapter *sc)
3161 const struct firmware *fw = NULL, *default_cfg;
3162 int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
3163 enum dev_state state;
3164 struct fw_info *fw_info;
3165 struct fw_hdr *card_fw; /* fw on the card */
3166 const struct fw_hdr *kld_fw; /* fw in the KLD */
3167 const struct fw_hdr *drv_fw; /* fw header the driver was compiled
3170 /* This is the firmware whose headers the driver was compiled against */
3171 fw_info = find_fw_info(chip_id(sc));
3172 if (fw_info == NULL) {
3173 device_printf(sc->dev,
3174 "unable to look up firmware information for chip %d.\n",
3178 drv_fw = &fw_info->fw_hdr;
3181 * The firmware KLD contains many modules. The KLD name is also the
3182 * name of the module that contains the default config file.
3184 default_cfg = firmware_get(fw_info->kld_name);
3186 /* This is the firmware in the KLD */
3187 fw = firmware_get(fw_info->fw_mod_name);
3189 kld_fw = (const void *)fw->data;
3190 kld_fw_usable = fw_compatible(drv_fw, kld_fw);
3196 /* Read the header of the firmware on the card */
3197 card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
3198 rc = -t4_read_flash(sc, FLASH_FW_START,
3199 sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
3201 card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
3202 if (card_fw->fw_ver == be32toh(0xffffffff)) {
3203 uint32_t d = be32toh(kld_fw->fw_ver);
3205 if (!kld_fw_usable) {
3206 device_printf(sc->dev,
3207 "no firmware on the card and no usable "
3208 "firmware bundled with the driver.\n");
3211 } else if (t4_fw_install == 0) {
3212 device_printf(sc->dev,
3213 "no firmware on the card and the driver "
3214 "is prohibited from installing new "
3220 device_printf(sc->dev, "no firmware on the card, "
3221 "installing firmware %d.%d.%d.%d\n",
3222 G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3223 G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3224 rc = t4_fw_forceinstall(sc, fw->data, fw->datasize);
3227 device_printf(sc->dev,
3228 "firmware install failed: %d.\n", rc);
3231 memcpy(card_fw, kld_fw, sizeof(*card_fw));
3236 device_printf(sc->dev,
3237 "Unable to read card's firmware header: %d\n", rc);
3241 /* Contact firmware. */
3242 rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
3243 if (rc < 0 || state == DEV_STATE_ERR) {
3245 device_printf(sc->dev,
3246 "failed to connect to the firmware: %d, %d.\n", rc, state);
3251 sc->flags |= MASTER_PF;
3252 else if (state == DEV_STATE_UNINIT) {
3254 * We didn't get to be the master so we definitely won't be
3255 * configuring the chip. It's a bug if someone else hasn't
3256 * configured it already.
3258 device_printf(sc->dev, "couldn't be master(%d), "
3259 "device not already initialized either(%d).\n", rc, state);
3264 if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
3265 (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
3267 * Common case: the firmware on the card is an exact match and
3268 * the KLD is an exact match too, or the KLD is
3269 * absent/incompatible. Note that t4_fw_install = 2 is ignored
3270 * here -- use cxgbetool loadfw if you want to reinstall the
3271 * same firmware as the one on the card.
3273 } else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
3274 should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
3275 be32toh(card_fw->fw_ver))) {
3277 rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3279 device_printf(sc->dev,
3280 "failed to install firmware: %d\n", rc);
3284 /* Installed successfully, update the cached header too. */
3285 memcpy(card_fw, kld_fw, sizeof(*card_fw));
3287 need_fw_reset = 0; /* already reset as part of load_fw */
3290 if (!card_fw_usable) {
3293 d = ntohl(drv_fw->fw_ver);
3294 c = ntohl(card_fw->fw_ver);
3295 k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
3297 device_printf(sc->dev, "Cannot find a usable firmware: "
3298 "fw_install %d, chip state %d, "
3299 "driver compiled with %d.%d.%d.%d, "
3300 "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
3301 t4_fw_install, state,
3302 G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3303 G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
3304 G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3305 G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
3306 G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3307 G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3313 if (need_fw_reset &&
3314 (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
3315 device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3316 if (rc != ETIMEDOUT && rc != EIO)
3317 t4_fw_bye(sc, sc->mbox);
3322 rc = get_params__pre_init(sc);
3324 goto done; /* error message displayed already */
3326 /* Partition adapter resources as specified in the config file. */
3327 if (state == DEV_STATE_UNINIT) {
3329 KASSERT(sc->flags & MASTER_PF,
3330 ("%s: trying to change chip settings when not master.",
3333 rc = partition_resources(sc, default_cfg, fw_info->kld_name);
3335 goto done; /* error message displayed already */
3337 t4_tweak_chip_settings(sc);
3339 /* get basic stuff going */
3340 rc = -t4_fw_initialize(sc, sc->mbox);
3342 device_printf(sc->dev, "fw init failed: %d.\n", rc);
3346 snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
3351 free(card_fw, M_CXGBE);
3353 firmware_put(fw, FIRMWARE_UNLOAD);
3354 if (default_cfg != NULL)
3355 firmware_put(default_cfg, FIRMWARE_UNLOAD);
3360 #define FW_PARAM_DEV(param) \
3361 (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3362 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3363 #define FW_PARAM_PFVF(param) \
3364 (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3365 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3368 * Partition chip resources for use between various PFs, VFs, etc.
3371 partition_resources(struct adapter *sc, const struct firmware *default_cfg,
3372 const char *name_prefix)
3374 const struct firmware *cfg = NULL;
3376 struct fw_caps_config_cmd caps;
3377 uint32_t mtype, moff, finicsum, cfcsum;
3380 * Figure out what configuration file to use. Pick the default config
3381 * file for the card if the user hasn't specified one explicitly.
3383 snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
3384 if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3385 /* Card specific overrides go here. */
3386 if (pci_get_device(sc->dev) == 0x440a)
3387 snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
3389 snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
3390 } else if (strncmp(t4_cfg_file, BUILTIN_CF, sizeof(t4_cfg_file)) == 0)
3391 goto use_built_in_config; /* go straight to config. */
3394 * We need to load another module if the profile is anything except
3395 * "default" or "flash".
3397 if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
3398 strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3401 snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
3402 cfg = firmware_get(s);
3404 if (default_cfg != NULL) {
3405 device_printf(sc->dev,
3406 "unable to load module \"%s\" for "
3407 "configuration profile \"%s\", will use "
3408 "the default config file instead.\n",
3410 snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3413 device_printf(sc->dev,
3414 "unable to load module \"%s\" for "
3415 "configuration profile \"%s\", will use "
3416 "the config file on the card's flash "
3417 "instead.\n", s, sc->cfg_file);
3418 snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3424 if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
3425 default_cfg == NULL) {
3426 device_printf(sc->dev,
3427 "default config file not available, will use the config "
3428 "file on the card's flash instead.\n");
3429 snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
3432 if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3434 const uint32_t *cfdata;
3435 uint32_t param, val, addr;
3437 KASSERT(cfg != NULL || default_cfg != NULL,
3438 ("%s: no config to upload", __func__));
3441 * Ask the firmware where it wants us to upload the config file.
3443 param = FW_PARAM_DEV(CF);
3444 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, ¶m, &val);
3446 /* No support for config file? Shouldn't happen. */
3447 device_printf(sc->dev,
3448 "failed to query config file location: %d.\n", rc);
3451 mtype = G_FW_PARAMS_PARAM_Y(val);
3452 moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3455 * XXX: sheer laziness. We deliberately added 4 bytes of
3456 * useless stuffing/comments at the end of the config file so
3457 * it's ok to simply throw away the last remaining bytes when
3458 * the config file is not an exact multiple of 4. This also
3459 * helps with the validate_mt_off_len check.
3462 cflen = cfg->datasize & ~3;
3465 cflen = default_cfg->datasize & ~3;
3466 cfdata = default_cfg->data;
3469 if (cflen > FLASH_CFG_MAX_SIZE) {
3470 device_printf(sc->dev,
3471 "config file too long (%d, max allowed is %d). "
3472 "Will try to use the config on the card, if any.\n",
3473 cflen, FLASH_CFG_MAX_SIZE);
3474 goto use_config_on_flash;
3477 rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3479 device_printf(sc->dev,
3480 "%s: addr (%d/0x%x) or len %d is not valid: %d. "
3481 "Will try to use the config on the card, if any.\n",
3482 __func__, mtype, moff, cflen, rc);
3483 goto use_config_on_flash;
3485 write_via_memwin(sc, 2, addr, cfdata, cflen);
3487 use_config_on_flash:
3488 mtype = FW_MEMTYPE_FLASH;
3489 moff = t4_flash_cfg_addr(sc);
3492 bzero(&caps, sizeof(caps));
3493 caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3494 F_FW_CMD_REQUEST | F_FW_CMD_READ);
3495 caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3496 V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3497 V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
3498 rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3500 device_printf(sc->dev,
3501 "failed to pre-process config file: %d "
3502 "(mtype %d, moff 0x%x). Will reset the firmware and retry "
3503 "with the built-in configuration.\n", rc, mtype, moff);
3505 rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST);
3507 device_printf(sc->dev,
3508 "firmware reset failed: %d.\n", rc);
3509 if (rc != ETIMEDOUT && rc != EIO) {
3510 t4_fw_bye(sc, sc->mbox);
3511 sc->flags &= ~FW_OK;
3515 snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", "built-in");
3516 use_built_in_config:
3517 bzero(&caps, sizeof(caps));
3518 caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3519 F_FW_CMD_REQUEST | F_FW_CMD_READ);
3520 caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3521 rc = t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3523 device_printf(sc->dev,
3524 "built-in configuration failed: %d.\n", rc);
3529 finicsum = be32toh(caps.finicsum);
3530 cfcsum = be32toh(caps.cfcsum);
3531 if (finicsum != cfcsum) {
3532 device_printf(sc->dev,
3533 "WARNING: config file checksum mismatch: %08x %08x\n",
3536 sc->cfcsum = cfcsum;
3538 #define LIMIT_CAPS(x) do { \
3539 caps.x &= htobe16(t4_##x##_allowed); \
3543 * Let the firmware know what features will (not) be used so it can tune
3544 * things accordingly.
3546 LIMIT_CAPS(nbmcaps);
3547 LIMIT_CAPS(linkcaps);
3548 LIMIT_CAPS(switchcaps);
3549 LIMIT_CAPS(niccaps);
3550 LIMIT_CAPS(toecaps);
3551 LIMIT_CAPS(rdmacaps);
3552 LIMIT_CAPS(cryptocaps);
3553 LIMIT_CAPS(iscsicaps);
3554 LIMIT_CAPS(fcoecaps);
3557 if (caps.niccaps & htobe16(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
3559 * TOE and hashfilters are mutually exclusive. It is a config
3560 * file or firmware bug if both are reported as available. Try
3561 * to cope with the situation in non-debug builds by disabling
3564 MPASS(caps.toecaps == 0);
3571 caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3572 F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3573 caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3574 rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3576 device_printf(sc->dev,
3577 "failed to process config file: %d.\n", rc);
3581 firmware_put(cfg, FIRMWARE_UNLOAD);
3586 * Retrieve parameters that are needed (or nice to have) very early.
3589 get_params__pre_init(struct adapter *sc)
3592 uint32_t param[2], val[2];
3594 t4_get_version_info(sc);
3596 snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
3597 G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
3598 G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
3599 G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
3600 G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
3602 snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
3603 G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
3604 G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
3605 G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
3606 G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
3608 snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
3609 G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
3610 G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
3611 G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
3612 G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
3614 snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
3615 G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
3616 G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
3617 G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
3618 G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
3620 param[0] = FW_PARAM_DEV(PORTVEC);
3621 param[1] = FW_PARAM_DEV(CCLK);
3622 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3624 device_printf(sc->dev,
3625 "failed to query parameters (pre_init): %d.\n", rc);
3629 sc->params.portvec = val[0];
3630 sc->params.nports = bitcount32(val[0]);
3631 sc->params.vpd.cclk = val[1];
3633 /* Read device log parameters. */
3634 rc = -t4_init_devlog_params(sc, 1);
3636 fixup_devlog_params(sc);
3638 device_printf(sc->dev,
3639 "failed to get devlog parameters: %d.\n", rc);
3640 rc = 0; /* devlog isn't critical for device operation */
3647 * Retrieve various parameters that are of interest to the driver. The device
3648 * has been initialized by the firmware at this point.
3651 get_params__post_init(struct adapter *sc)
3654 uint32_t param[7], val[7];
3655 struct fw_caps_config_cmd caps;
3657 param[0] = FW_PARAM_PFVF(IQFLINT_START);
3658 param[1] = FW_PARAM_PFVF(EQ_START);
3659 param[2] = FW_PARAM_PFVF(FILTER_START);
3660 param[3] = FW_PARAM_PFVF(FILTER_END);
3661 param[4] = FW_PARAM_PFVF(L2T_START);
3662 param[5] = FW_PARAM_PFVF(L2T_END);
3663 param[6] = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
3664 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
3665 V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
3666 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 7, param, val);
3668 device_printf(sc->dev,
3669 "failed to query parameters (post_init): %d.\n", rc);
3673 sc->sge.iq_start = val[0];
3674 sc->sge.eq_start = val[1];
3675 sc->tids.ftid_base = val[2];
3676 sc->tids.nftids = val[3] - val[2] + 1;
3677 sc->params.ftid_min = val[2];
3678 sc->params.ftid_max = val[3];
3679 sc->vres.l2t.start = val[4];
3680 sc->vres.l2t.size = val[5] - val[4] + 1;
3681 KASSERT(sc->vres.l2t.size <= L2T_SIZE,
3682 ("%s: L2 table size (%u) larger than expected (%u)",
3683 __func__, sc->vres.l2t.size, L2T_SIZE));
3684 sc->params.core_vdd = val[6];
3687 * MPSBGMAP is queried separately because only recent firmwares support
3688 * it as a parameter and we don't want the compound query above to fail
3689 * on older firmwares.
3691 param[0] = FW_PARAM_DEV(MPSBGMAP);
3693 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
3695 sc->params.mps_bg_map = val[0];
3697 sc->params.mps_bg_map = 0;
3700 * Determine whether the firmware supports the filter2 work request.
3701 * This is queried separately for the same reason as MPSBGMAP above.
3703 param[0] = FW_PARAM_DEV(FILTER2_WR);
3705 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
3707 sc->params.filter2_wr_support = val[0] != 0;
3709 sc->params.filter2_wr_support = 0;
3711 /* get capabilites */
3712 bzero(&caps, sizeof(caps));
3713 caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3714 F_FW_CMD_REQUEST | F_FW_CMD_READ);
3715 caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3716 rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3718 device_printf(sc->dev,
3719 "failed to get card capabilities: %d.\n", rc);
3723 #define READ_CAPS(x) do { \
3724 sc->x = htobe16(caps.x); \
3727 READ_CAPS(linkcaps);
3728 READ_CAPS(switchcaps);
3731 READ_CAPS(rdmacaps);
3732 READ_CAPS(cryptocaps);
3733 READ_CAPS(iscsicaps);
3734 READ_CAPS(fcoecaps);
3736 if (sc->niccaps & FW_CAPS_CONFIG_NIC_HASHFILTER) {
3737 MPASS(chip_id(sc) > CHELSIO_T4);
3738 MPASS(sc->toecaps == 0);
3741 param[0] = FW_PARAM_DEV(NTID);
3742 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3744 device_printf(sc->dev,
3745 "failed to query HASHFILTER parameters: %d.\n", rc);
3748 sc->tids.ntids = val[0];
3749 sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3750 sc->params.hash_filter = 1;
3752 if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
3753 param[0] = FW_PARAM_PFVF(ETHOFLD_START);
3754 param[1] = FW_PARAM_PFVF(ETHOFLD_END);
3755 param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3756 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
3758 device_printf(sc->dev,
3759 "failed to query NIC parameters: %d.\n", rc);
3762 sc->tids.etid_base = val[0];
3763 sc->params.etid_min = val[0];
3764 sc->params.etid_max = val[1];
3765 sc->tids.netids = val[1] - val[0] + 1;
3766 sc->params.eo_wr_cred = val[2];
3767 sc->params.ethoffload = 1;
3770 /* query offload-related parameters */
3771 param[0] = FW_PARAM_DEV(NTID);
3772 param[1] = FW_PARAM_PFVF(SERVER_START);
3773 param[2] = FW_PARAM_PFVF(SERVER_END);
3774 param[3] = FW_PARAM_PFVF(TDDP_START);
3775 param[4] = FW_PARAM_PFVF(TDDP_END);
3776 param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3777 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3779 device_printf(sc->dev,
3780 "failed to query TOE parameters: %d.\n", rc);
3783 sc->tids.ntids = val[0];
3784 sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3785 sc->tids.stid_base = val[1];
3786 sc->tids.nstids = val[2] - val[1] + 1;
3787 sc->vres.ddp.start = val[3];
3788 sc->vres.ddp.size = val[4] - val[3] + 1;
3789 sc->params.ofldq_wr_cred = val[5];
3790 sc->params.offload = 1;
3793 * The firmware attempts memfree TOE configuration for -SO cards
3794 * and will report toecaps=0 if it runs out of resources (this
3795 * depends on the config file). It may not report 0 for other
3796 * capabilities dependent on the TOE in this case. Set them to
3797 * 0 here so that the driver doesn't bother tracking resources
3798 * that will never be used.
3804 param[0] = FW_PARAM_PFVF(STAG_START);
3805 param[1] = FW_PARAM_PFVF(STAG_END);
3806 param[2] = FW_PARAM_PFVF(RQ_START);
3807 param[3] = FW_PARAM_PFVF(RQ_END);
3808 param[4] = FW_PARAM_PFVF(PBL_START);
3809 param[5] = FW_PARAM_PFVF(PBL_END);
3810 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3812 device_printf(sc->dev,
3813 "failed to query RDMA parameters(1): %d.\n", rc);
3816 sc->vres.stag.start = val[0];
3817 sc->vres.stag.size = val[1] - val[0] + 1;
3818 sc->vres.rq.start = val[2];
3819 sc->vres.rq.size = val[3] - val[2] + 1;
3820 sc->vres.pbl.start = val[4];
3821 sc->vres.pbl.size = val[5] - val[4] + 1;
3823 param[0] = FW_PARAM_PFVF(SQRQ_START);
3824 param[1] = FW_PARAM_PFVF(SQRQ_END);
3825 param[2] = FW_PARAM_PFVF(CQ_START);
3826 param[3] = FW_PARAM_PFVF(CQ_END);
3827 param[4] = FW_PARAM_PFVF(OCQ_START);
3828 param[5] = FW_PARAM_PFVF(OCQ_END);
3829 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3831 device_printf(sc->dev,
3832 "failed to query RDMA parameters(2): %d.\n", rc);
3835 sc->vres.qp.start = val[0];
3836 sc->vres.qp.size = val[1] - val[0] + 1;
3837 sc->vres.cq.start = val[2];
3838 sc->vres.cq.size = val[3] - val[2] + 1;
3839 sc->vres.ocq.start = val[4];
3840 sc->vres.ocq.size = val[5] - val[4] + 1;
3842 param[0] = FW_PARAM_PFVF(SRQ_START);
3843 param[1] = FW_PARAM_PFVF(SRQ_END);
3844 param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
3845 param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
3846 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
3848 device_printf(sc->dev,
3849 "failed to query RDMA parameters(3): %d.\n", rc);
3852 sc->vres.srq.start = val[0];
3853 sc->vres.srq.size = val[1] - val[0] + 1;
3854 sc->params.max_ordird_qp = val[2];
3855 sc->params.max_ird_adapter = val[3];
3857 if (sc->iscsicaps) {
3858 param[0] = FW_PARAM_PFVF(ISCSI_START);
3859 param[1] = FW_PARAM_PFVF(ISCSI_END);
3860 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3862 device_printf(sc->dev,
3863 "failed to query iSCSI parameters: %d.\n", rc);
3866 sc->vres.iscsi.start = val[0];
3867 sc->vres.iscsi.size = val[1] - val[0] + 1;
3869 if (sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS) {
3870 param[0] = FW_PARAM_PFVF(TLS_START);
3871 param[1] = FW_PARAM_PFVF(TLS_END);
3872 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3874 device_printf(sc->dev,
3875 "failed to query TLS parameters: %d.\n", rc);
3878 sc->vres.key.start = val[0];
3879 sc->vres.key.size = val[1] - val[0] + 1;
3882 t4_init_sge_params(sc);
3885 * We've got the params we wanted to query via the firmware. Now grab
3886 * some others directly from the chip.
3888 rc = t4_read_chip_settings(sc);
3894 set_params__post_init(struct adapter *sc)
3896 uint32_t param, val;
3901 /* ask for encapsulated CPLs */
3902 param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3904 (void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, ¶m, &val);
3908 * Override the TOE timers with user provided tunables. This is not the
3909 * recommended way to change the timers (the firmware config file is) so
3910 * these tunables are not documented.
3912 * All the timer tunables are in microseconds.
3914 if (t4_toe_keepalive_idle != 0) {
3915 v = us_to_tcp_ticks(sc, t4_toe_keepalive_idle);
3916 v &= M_KEEPALIVEIDLE;
3917 t4_set_reg_field(sc, A_TP_KEEP_IDLE,
3918 V_KEEPALIVEIDLE(M_KEEPALIVEIDLE), V_KEEPALIVEIDLE(v));
3920 if (t4_toe_keepalive_interval != 0) {
3921 v = us_to_tcp_ticks(sc, t4_toe_keepalive_interval);
3922 v &= M_KEEPALIVEINTVL;
3923 t4_set_reg_field(sc, A_TP_KEEP_INTVL,
3924 V_KEEPALIVEINTVL(M_KEEPALIVEINTVL), V_KEEPALIVEINTVL(v));
3926 if (t4_toe_keepalive_count != 0) {
3927 v = t4_toe_keepalive_count & M_KEEPALIVEMAXR2;
3928 t4_set_reg_field(sc, A_TP_SHIFT_CNT,
3929 V_KEEPALIVEMAXR1(M_KEEPALIVEMAXR1) |
3930 V_KEEPALIVEMAXR2(M_KEEPALIVEMAXR2),
3931 V_KEEPALIVEMAXR1(1) | V_KEEPALIVEMAXR2(v));
3933 if (t4_toe_rexmt_min != 0) {
3934 v = us_to_tcp_ticks(sc, t4_toe_rexmt_min);
3936 t4_set_reg_field(sc, A_TP_RXT_MIN,
3937 V_RXTMIN(M_RXTMIN), V_RXTMIN(v));
3939 if (t4_toe_rexmt_max != 0) {
3940 v = us_to_tcp_ticks(sc, t4_toe_rexmt_max);
3942 t4_set_reg_field(sc, A_TP_RXT_MAX,
3943 V_RXTMAX(M_RXTMAX), V_RXTMAX(v));
3945 if (t4_toe_rexmt_count != 0) {
3946 v = t4_toe_rexmt_count & M_RXTSHIFTMAXR2;
3947 t4_set_reg_field(sc, A_TP_SHIFT_CNT,
3948 V_RXTSHIFTMAXR1(M_RXTSHIFTMAXR1) |
3949 V_RXTSHIFTMAXR2(M_RXTSHIFTMAXR2),
3950 V_RXTSHIFTMAXR1(1) | V_RXTSHIFTMAXR2(v));
3952 for (i = 0; i < nitems(t4_toe_rexmt_backoff); i++) {
3953 if (t4_toe_rexmt_backoff[i] != -1) {
3954 v = t4_toe_rexmt_backoff[i] & M_TIMERBACKOFFINDEX0;
3955 shift = (i & 3) << 3;
3956 t4_set_reg_field(sc, A_TP_TCP_BACKOFF_REG0 + (i & ~3),
3957 M_TIMERBACKOFFINDEX0 << shift, v << shift);
3964 #undef FW_PARAM_PFVF
3968 t4_set_desc(struct adapter *sc)
3971 struct adapter_params *p = &sc->params;
3973 snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
3975 device_set_desc_copy(sc->dev, buf);
3979 build_medialist(struct port_info *pi, struct ifmedia *media)
3983 PORT_LOCK_ASSERT_OWNED(pi);
3985 ifmedia_removeall(media);
3988 * XXX: Would it be better to ifmedia_add all 4 combinations of pause
3989 * settings for every speed instead of just txpause|rxpause? ifconfig
3990 * media display looks much better if autoselect is the only case where
3991 * ifm_current is different from ifm_active. If the user picks anything
3992 * except txpause|rxpause the display is ugly.
3994 m = IFM_ETHER | IFM_FDX | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE;
3996 switch(pi->port_type) {
3997 case FW_PORT_TYPE_BT_XFI:
3998 case FW_PORT_TYPE_BT_XAUI:
3999 ifmedia_add(media, m | IFM_10G_T, 0, NULL);
4002 case FW_PORT_TYPE_BT_SGMII:
4003 ifmedia_add(media, m | IFM_1000_T, 0, NULL);
4004 ifmedia_add(media, m | IFM_100_TX, 0, NULL);
4005 ifmedia_add(media, IFM_ETHER | IFM_AUTO, 0, NULL);
4006 ifmedia_set(media, IFM_ETHER | IFM_AUTO);
4009 case FW_PORT_TYPE_CX4:
4010 ifmedia_add(media, m | IFM_10G_CX4, 0, NULL);
4011 ifmedia_set(media, m | IFM_10G_CX4);
4014 case FW_PORT_TYPE_QSFP_10G:
4015 case FW_PORT_TYPE_SFP:
4016 case FW_PORT_TYPE_FIBER_XFI:
4017 case FW_PORT_TYPE_FIBER_XAUI:
4018 switch (pi->mod_type) {
4020 case FW_PORT_MOD_TYPE_LR:
4021 ifmedia_add(media, m | IFM_10G_LR, 0, NULL);
4022 ifmedia_set(media, m | IFM_10G_LR);
4025 case FW_PORT_MOD_TYPE_SR:
4026 ifmedia_add(media, m | IFM_10G_SR, 0, NULL);
4027 ifmedia_set(media, m | IFM_10G_SR);
4030 case FW_PORT_MOD_TYPE_LRM:
4031 ifmedia_add(media, m | IFM_10G_LRM, 0, NULL);
4032 ifmedia_set(media, m | IFM_10G_LRM);
4035 case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
4036 case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
4037 ifmedia_add(media, m | IFM_10G_TWINAX, 0, NULL);
4038 ifmedia_set(media, m | IFM_10G_TWINAX);
4041 case FW_PORT_MOD_TYPE_NONE:
4043 ifmedia_add(media, m | IFM_NONE, 0, NULL);
4044 ifmedia_set(media, m | IFM_NONE);
4047 case FW_PORT_MOD_TYPE_NA:
4048 case FW_PORT_MOD_TYPE_ER:
4050 device_printf(pi->dev,
4051 "unknown port_type (%d), mod_type (%d)\n",
4052 pi->port_type, pi->mod_type);
4053 ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
4054 ifmedia_set(media, m | IFM_UNKNOWN);
4059 case FW_PORT_TYPE_CR_QSFP:
4060 case FW_PORT_TYPE_SFP28:
4061 case FW_PORT_TYPE_KR_SFP28:
4062 switch (pi->mod_type) {
4064 case FW_PORT_MOD_TYPE_SR:
4065 ifmedia_add(media, m | IFM_25G_SR, 0, NULL);
4066 ifmedia_set(media, m | IFM_25G_SR);
4069 case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
4070 case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
4071 ifmedia_add(media, m | IFM_25G_CR, 0, NULL);
4072 ifmedia_set(media, m | IFM_25G_CR);
4075 case FW_PORT_MOD_TYPE_NONE:
4077 ifmedia_add(media, m | IFM_NONE, 0, NULL);
4078 ifmedia_set(media, m | IFM_NONE);
4082 device_printf(pi->dev,
4083 "unknown port_type (%d), mod_type (%d)\n",
4084 pi->port_type, pi->mod_type);
4085 ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
4086 ifmedia_set(media, m | IFM_UNKNOWN);
4091 case FW_PORT_TYPE_QSFP:
4092 switch (pi->mod_type) {
4094 case FW_PORT_MOD_TYPE_LR:
4095 ifmedia_add(media, m | IFM_40G_LR4, 0, NULL);
4096 ifmedia_set(media, m | IFM_40G_LR4);
4099 case FW_PORT_MOD_TYPE_SR:
4100 ifmedia_add(media, m | IFM_40G_SR4, 0, NULL);
4101 ifmedia_set(media, m | IFM_40G_SR4);
4104 case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
4105 case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
4106 ifmedia_add(media, m | IFM_40G_CR4, 0, NULL);
4107 ifmedia_set(media, m | IFM_40G_CR4);
4110 case FW_PORT_MOD_TYPE_NONE:
4112 ifmedia_add(media, m | IFM_NONE, 0, NULL);
4113 ifmedia_set(media, m | IFM_NONE);
4117 device_printf(pi->dev,
4118 "unknown port_type (%d), mod_type (%d)\n",
4119 pi->port_type, pi->mod_type);
4120 ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
4121 ifmedia_set(media, m | IFM_UNKNOWN);
4126 case FW_PORT_TYPE_KR4_100G:
4127 case FW_PORT_TYPE_CR4_QSFP:
4128 switch (pi->mod_type) {
4130 case FW_PORT_MOD_TYPE_LR:
4131 ifmedia_add(media, m | IFM_100G_LR4, 0, NULL);
4132 ifmedia_set(media, m | IFM_100G_LR4);
4135 case FW_PORT_MOD_TYPE_SR:
4136 ifmedia_add(media, m | IFM_100G_SR4, 0, NULL);
4137 ifmedia_set(media, m | IFM_100G_SR4);
4140 case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
4141 case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
4142 ifmedia_add(media, m | IFM_100G_CR4, 0, NULL);
4143 ifmedia_set(media, m | IFM_100G_CR4);
4146 case FW_PORT_MOD_TYPE_NONE:
4148 ifmedia_add(media, m | IFM_NONE, 0, NULL);
4149 ifmedia_set(media, m | IFM_NONE);
4153 device_printf(pi->dev,
4154 "unknown port_type (%d), mod_type (%d)\n",
4155 pi->port_type, pi->mod_type);
4156 ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
4157 ifmedia_set(media, m | IFM_UNKNOWN);
4163 device_printf(pi->dev,
4164 "unknown port_type (%d), mod_type (%d)\n", pi->port_type,
4166 ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
4167 ifmedia_set(media, m | IFM_UNKNOWN);
4173 * Update all the requested_* fields in the link config and then send a mailbox
4174 * command to apply the settings.
4177 init_l1cfg(struct port_info *pi)
4179 struct adapter *sc = pi->adapter;
4180 struct link_config *lc = &pi->link_cfg;
4183 ASSERT_SYNCHRONIZED_OP(sc);
4185 lc->requested_speed = port_top_speed(pi); /* in Gbps */
4186 if (t4_autoneg != 0 && lc->supported & FW_PORT_CAP_ANEG) {
4187 lc->requested_aneg = AUTONEG_ENABLE;
4189 lc->requested_aneg = AUTONEG_DISABLE;
4192 lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX);
4195 lc->requested_fec = t4_fec & (FEC_RS | FEC_BASER_RS |
4198 /* Use the suggested value provided by the firmware in acaps */
4199 if (lc->advertising & FW_PORT_CAP_FEC_RS)
4200 lc->requested_fec = FEC_RS;
4201 else if (lc->advertising & FW_PORT_CAP_FEC_BASER_RS)
4202 lc->requested_fec = FEC_BASER_RS;
4204 lc->requested_fec = 0;
4207 rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
4209 device_printf(pi->dev, "l1cfg failed: %d\n", rc);
4211 lc->fc = lc->requested_fc;
4212 lc->fec = lc->requested_fec;
4216 #define FW_MAC_EXACT_CHUNK 7
4219 * Program the port's XGMAC based on parameters in ifnet. The caller also
4220 * indicates which parameters should be programmed (the rest are left alone).
4223 update_mac_settings(struct ifnet *ifp, int flags)
4226 struct vi_info *vi = ifp->if_softc;
4227 struct port_info *pi = vi->pi;
4228 struct adapter *sc = pi->adapter;
4229 int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
4231 ASSERT_SYNCHRONIZED_OP(sc);
4232 KASSERT(flags, ("%s: not told what to update.", __func__));
4234 if (flags & XGMAC_MTU)
4237 if (flags & XGMAC_PROMISC)
4238 promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
4240 if (flags & XGMAC_ALLMULTI)
4241 allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
4243 if (flags & XGMAC_VLANEX)
4244 vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
4246 if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
4247 rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
4248 allmulti, 1, vlanex, false);
4250 if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
4256 if (flags & XGMAC_UCADDR) {
4257 uint8_t ucaddr[ETHER_ADDR_LEN];
4259 bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
4260 rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
4261 ucaddr, true, true);
4264 if_printf(ifp, "change_mac failed: %d\n", rc);
4267 vi->xact_addr_filt = rc;
4272 if (flags & XGMAC_MCADDRS) {
4273 const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
4276 struct ifmultiaddr *ifma;
4279 if_maddr_rlock(ifp);
4280 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
4281 if (ifma->ifma_addr->sa_family != AF_LINK)
4284 LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
4285 MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
4288 if (i == FW_MAC_EXACT_CHUNK) {
4289 rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
4290 del, i, mcaddr, NULL, &hash, 0);
4293 for (j = 0; j < i; j++) {
4295 "failed to add mc address"
4297 "%02x:%02x:%02x rc=%d\n",
4298 mcaddr[j][0], mcaddr[j][1],
4299 mcaddr[j][2], mcaddr[j][3],
4300 mcaddr[j][4], mcaddr[j][5],
4310 rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
4311 mcaddr, NULL, &hash, 0);
4314 for (j = 0; j < i; j++) {
4316 "failed to add mc address"
4318 "%02x:%02x:%02x rc=%d\n",
4319 mcaddr[j][0], mcaddr[j][1],
4320 mcaddr[j][2], mcaddr[j][3],
4321 mcaddr[j][4], mcaddr[j][5],
4328 rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
4330 if_printf(ifp, "failed to set mc address hash: %d", rc);
4332 if_maddr_runlock(ifp);
4339 * {begin|end}_synchronized_op must be called from the same thread.
4342 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
4348 /* the caller thinks it's ok to sleep, but is it really? */
4349 if (flags & SLEEP_OK)
4350 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
4351 "begin_synchronized_op");
4362 if (vi && IS_DOOMED(vi)) {
4372 if (!(flags & SLEEP_OK)) {
4377 if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
4383 KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
4386 sc->last_op = wmesg;
4387 sc->last_op_thr = curthread;
4388 sc->last_op_flags = flags;
4392 if (!(flags & HOLD_LOCK) || rc)
4399 * Tell if_ioctl and if_init that the VI is going away. This is
4400 * special variant of begin_synchronized_op and must be paired with a
4401 * call to end_synchronized_op.
4404 doom_vi(struct adapter *sc, struct vi_info *vi)
4411 mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
4414 sc->last_op = "t4detach";
4415 sc->last_op_thr = curthread;
4416 sc->last_op_flags = 0;
4422 * {begin|end}_synchronized_op must be called from the same thread.
4425 end_synchronized_op(struct adapter *sc, int flags)
4428 if (flags & LOCK_HELD)
4429 ADAPTER_LOCK_ASSERT_OWNED(sc);
4433 KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
4440 cxgbe_init_synchronized(struct vi_info *vi)
4442 struct port_info *pi = vi->pi;
4443 struct adapter *sc = pi->adapter;
4444 struct ifnet *ifp = vi->ifp;
4446 struct sge_txq *txq;
4448 ASSERT_SYNCHRONIZED_OP(sc);
4450 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
4451 return (0); /* already running */
4453 if (!(sc->flags & FULL_INIT_DONE) &&
4454 ((rc = adapter_full_init(sc)) != 0))
4455 return (rc); /* error message displayed already */
4457 if (!(vi->flags & VI_INIT_DONE) &&
4458 ((rc = vi_full_init(vi)) != 0))
4459 return (rc); /* error message displayed already */
4461 rc = update_mac_settings(ifp, XGMAC_ALL);
4463 goto done; /* error message displayed already */
4465 rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
4467 if_printf(ifp, "enable_vi failed: %d\n", rc);
4472 * Can't fail from this point onwards. Review cxgbe_uninit_synchronized
4476 for_each_txq(vi, i, txq) {
4478 txq->eq.flags |= EQ_ENABLED;
4483 * The first iq of the first port to come up is used for tracing.
4485 if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
4486 sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
4487 t4_write_reg(sc, is_t4(sc) ? A_MPS_TRC_RSS_CONTROL :
4488 A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
4489 V_QUEUENUMBER(sc->traceq));
4490 pi->flags |= HAS_TRACEQ;
4495 if (pi->up_vis++ == 0) {
4496 t4_update_port_info(pi);
4497 build_medialist(pi, &pi->media);
4500 ifp->if_drv_flags |= IFF_DRV_RUNNING;
4502 if (pi->nvi > 1 || sc->flags & IS_VF)
4503 callout_reset(&vi->tick, hz, vi_tick, vi);
4505 callout_reset(&pi->tick, hz, cxgbe_tick, pi);
4509 cxgbe_uninit_synchronized(vi);
4518 cxgbe_uninit_synchronized(struct vi_info *vi)
4520 struct port_info *pi = vi->pi;
4521 struct adapter *sc = pi->adapter;
4522 struct ifnet *ifp = vi->ifp;
4524 struct sge_txq *txq;
4526 ASSERT_SYNCHRONIZED_OP(sc);
4528 if (!(vi->flags & VI_INIT_DONE)) {
4529 if (__predict_false(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4530 KASSERT(0, ("uninited VI is running"));
4531 if_printf(ifp, "uninited VI with running ifnet. "
4532 "vi->flags 0x%016lx, if_flags 0x%08x, "
4533 "if_drv_flags 0x%08x\n", vi->flags, ifp->if_flags,
4540 * Disable the VI so that all its data in either direction is discarded
4541 * by the MPS. Leave everything else (the queues, interrupts, and 1Hz
4542 * tick) intact as the TP can deliver negative advice or data that it's
4543 * holding in its RAM (for an offloaded connection) even after the VI is
4546 rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
4548 if_printf(ifp, "disable_vi failed: %d\n", rc);
4552 for_each_txq(vi, i, txq) {
4554 txq->eq.flags &= ~EQ_ENABLED;
4559 if (pi->nvi > 1 || sc->flags & IS_VF)
4560 callout_stop(&vi->tick);
4562 callout_stop(&pi->tick);
4563 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4567 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
4569 if (pi->up_vis > 0) {
4575 pi->link_cfg.link_ok = 0;
4576 pi->link_cfg.speed = 0;
4577 pi->link_cfg.link_down_rc = 255;
4578 t4_os_link_changed(pi);
4579 pi->old_link_cfg = pi->link_cfg;
4585 * It is ok for this function to fail midway and return right away. t4_detach
4586 * will walk the entire sc->irq list and clean up whatever is valid.
4589 t4_setup_intr_handlers(struct adapter *sc)
4591 int rc, rid, p, q, v;
4594 struct port_info *pi;
4596 struct sge *sge = &sc->sge;
4597 struct sge_rxq *rxq;
4599 struct sge_ofld_rxq *ofld_rxq;
4602 struct sge_nm_rxq *nm_rxq;
4605 int nbuckets = rss_getnumbuckets();
4612 rid = sc->intr_type == INTR_INTX ? 0 : 1;
4613 if (forwarding_intr_to_fwq(sc))
4614 return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
4616 /* Multiple interrupts. */
4617 if (sc->flags & IS_VF)
4618 KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
4619 ("%s: too few intr.", __func__));
4621 KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
4622 ("%s: too few intr.", __func__));
4624 /* The first one is always error intr on PFs */
4625 if (!(sc->flags & IS_VF)) {
4626 rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
4633 /* The second one is always the firmware event queue (first on VFs) */
4634 rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
4640 for_each_port(sc, p) {
4642 for_each_vi(pi, v, vi) {
4643 vi->first_intr = rid - 1;
4645 if (vi->nnmrxq > 0) {
4646 int n = max(vi->nrxq, vi->nnmrxq);
4648 rxq = &sge->rxq[vi->first_rxq];
4650 nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
4652 for (q = 0; q < n; q++) {
4653 snprintf(s, sizeof(s), "%x%c%x", p,
4659 irq->nm_rxq = nm_rxq++;
4661 rc = t4_alloc_irq(sc, irq, rid,
4662 t4_vi_intr, irq, s);
4667 bus_bind_intr(sc->dev, irq->res,
4668 rss_getcpu(q % nbuckets));
4676 for_each_rxq(vi, q, rxq) {
4677 snprintf(s, sizeof(s), "%x%c%x", p,
4679 rc = t4_alloc_irq(sc, irq, rid,
4684 bus_bind_intr(sc->dev, irq->res,
4685 rss_getcpu(q % nbuckets));
4693 for_each_ofld_rxq(vi, q, ofld_rxq) {
4694 snprintf(s, sizeof(s), "%x%c%x", p, 'A' + v, q);
4695 rc = t4_alloc_irq(sc, irq, rid, t4_intr,
4706 MPASS(irq == &sc->irq[sc->intr_count]);
4712 adapter_full_init(struct adapter *sc)
4716 uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4717 uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4720 ASSERT_SYNCHRONIZED_OP(sc);
4721 ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4722 KASSERT((sc->flags & FULL_INIT_DONE) == 0,
4723 ("%s: FULL_INIT_DONE already", __func__));
4726 * queues that belong to the adapter (not any particular port).
4728 rc = t4_setup_adapter_queues(sc);
4732 for (i = 0; i < nitems(sc->tq); i++) {
4733 sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
4734 taskqueue_thread_enqueue, &sc->tq[i]);
4735 if (sc->tq[i] == NULL) {
4736 device_printf(sc->dev,
4737 "failed to allocate task queue %d\n", i);
4741 taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
4742 device_get_nameunit(sc->dev), i);
4745 MPASS(RSS_KEYSIZE == 40);
4746 rss_getkey((void *)&raw_rss_key[0]);
4747 for (i = 0; i < nitems(rss_key); i++) {
4748 rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
4750 t4_write_rss_key(sc, &rss_key[0], -1, 1);
4753 if (!(sc->flags & IS_VF))
4755 sc->flags |= FULL_INIT_DONE;
4758 adapter_full_uninit(sc);
4764 adapter_full_uninit(struct adapter *sc)
4768 ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4770 t4_teardown_adapter_queues(sc);
4772 for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
4773 taskqueue_free(sc->tq[i]);
4777 sc->flags &= ~FULL_INIT_DONE;
4783 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
4784 RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
4785 RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
4786 RSS_HASHTYPE_RSS_UDP_IPV6)
4788 /* Translates kernel hash types to hardware. */
4790 hashconfig_to_hashen(int hashconfig)
4794 if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
4795 hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
4796 if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
4797 hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
4798 if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
4799 hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4800 F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4802 if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
4803 hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4804 F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4806 if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
4807 hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4808 if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
4809 hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4814 /* Translates hardware hash types to kernel. */
4816 hashen_to_hashconfig(int hashen)
4820 if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
4822 * If UDP hashing was enabled it must have been enabled for
4823 * either IPv4 or IPv6 (inclusive or). Enabling UDP without
4824 * enabling any 4-tuple hash is nonsense configuration.
4826 MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4827 F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
4829 if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4830 hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
4831 if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4832 hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
4834 if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4835 hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
4836 if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4837 hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
4838 if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
4839 hashconfig |= RSS_HASHTYPE_RSS_IPV4;
4840 if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
4841 hashconfig |= RSS_HASHTYPE_RSS_IPV6;
4843 return (hashconfig);
4848 vi_full_init(struct vi_info *vi)
4850 struct adapter *sc = vi->pi->adapter;
4851 struct ifnet *ifp = vi->ifp;
4853 struct sge_rxq *rxq;
4854 int rc, i, j, hashen;
4856 int nbuckets = rss_getnumbuckets();
4857 int hashconfig = rss_gethashconfig();
4861 ASSERT_SYNCHRONIZED_OP(sc);
4862 KASSERT((vi->flags & VI_INIT_DONE) == 0,
4863 ("%s: VI_INIT_DONE already", __func__));
4865 sysctl_ctx_init(&vi->ctx);
4866 vi->flags |= VI_SYSCTL_CTX;
4869 * Allocate tx/rx/fl queues for this VI.
4871 rc = t4_setup_vi_queues(vi);
4873 goto done; /* error message displayed already */
4876 * Setup RSS for this VI. Save a copy of the RSS table for later use.
4878 if (vi->nrxq > vi->rss_size) {
4879 if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
4880 "some queues will never receive traffic.\n", vi->nrxq,
4882 } else if (vi->rss_size % vi->nrxq) {
4883 if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
4884 "expect uneven traffic distribution.\n", vi->nrxq,
4888 if (vi->nrxq != nbuckets) {
4889 if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
4890 "performance will be impacted.\n", vi->nrxq, nbuckets);
4893 rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
4894 for (i = 0; i < vi->rss_size;) {
4896 j = rss_get_indirection_to_bucket(i);
4898 rxq = &sc->sge.rxq[vi->first_rxq + j];
4899 rss[i++] = rxq->iq.abs_id;
4901 for_each_rxq(vi, j, rxq) {
4902 rss[i++] = rxq->iq.abs_id;
4903 if (i == vi->rss_size)
4909 rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
4912 if_printf(ifp, "rss_config failed: %d\n", rc);
4917 hashen = hashconfig_to_hashen(hashconfig);
4920 * We may have had to enable some hashes even though the global config
4921 * wants them disabled. This is a potential problem that must be
4922 * reported to the user.
4924 extra = hashen_to_hashconfig(hashen) ^ hashconfig;
4927 * If we consider only the supported hash types, then the enabled hashes
4928 * are a superset of the requested hashes. In other words, there cannot
4929 * be any supported hash that was requested but not enabled, but there
4930 * can be hashes that were not requested but had to be enabled.
4932 extra &= SUPPORTED_RSS_HASHTYPES;
4933 MPASS((extra & hashconfig) == 0);
4937 "global RSS config (0x%x) cannot be accommodated.\n",
4940 if (extra & RSS_HASHTYPE_RSS_IPV4)
4941 if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
4942 if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
4943 if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
4944 if (extra & RSS_HASHTYPE_RSS_IPV6)
4945 if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
4946 if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
4947 if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
4948 if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
4949 if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
4950 if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
4951 if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
4953 hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
4954 F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
4955 F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4956 F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
4958 rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, hashen, rss[0], 0, 0);
4960 if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
4965 vi->flags |= VI_INIT_DONE;
4977 vi_full_uninit(struct vi_info *vi)
4979 struct port_info *pi = vi->pi;
4980 struct adapter *sc = pi->adapter;
4982 struct sge_rxq *rxq;
4983 struct sge_txq *txq;
4985 struct sge_ofld_rxq *ofld_rxq;
4986 struct sge_wrq *ofld_txq;
4989 if (vi->flags & VI_INIT_DONE) {
4991 /* Need to quiesce queues. */
4993 /* XXX: Only for the first VI? */
4994 if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
4995 quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
4997 for_each_txq(vi, i, txq) {
4998 quiesce_txq(sc, txq);
5002 for_each_ofld_txq(vi, i, ofld_txq) {
5003 quiesce_wrq(sc, ofld_txq);
5007 for_each_rxq(vi, i, rxq) {
5008 quiesce_iq(sc, &rxq->iq);
5009 quiesce_fl(sc, &rxq->fl);
5013 for_each_ofld_rxq(vi, i, ofld_rxq) {
5014 quiesce_iq(sc, &ofld_rxq->iq);
5015 quiesce_fl(sc, &ofld_rxq->fl);
5018 free(vi->rss, M_CXGBE);
5019 free(vi->nm_rss, M_CXGBE);
5022 t4_teardown_vi_queues(vi);
5023 vi->flags &= ~VI_INIT_DONE;
5029 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
5031 struct sge_eq *eq = &txq->eq;
5032 struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
5034 (void) sc; /* unused */
5038 MPASS((eq->flags & EQ_ENABLED) == 0);
5042 /* Wait for the mp_ring to empty. */
5043 while (!mp_ring_is_idle(txq->r)) {
5044 mp_ring_check_drainage(txq->r, 0);
5045 pause("rquiesce", 1);
5048 /* Then wait for the hardware to finish. */
5049 while (spg->cidx != htobe16(eq->pidx))
5050 pause("equiesce", 1);
5052 /* Finally, wait for the driver to reclaim all descriptors. */
5053 while (eq->cidx != eq->pidx)
5054 pause("dquiesce", 1);
5058 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
5065 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
5067 (void) sc; /* unused */
5069 /* Synchronize with the interrupt handler */
5070 while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
5075 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
5077 mtx_lock(&sc->sfl_lock);
5079 fl->flags |= FL_DOOMED;
5081 callout_stop(&sc->sfl_callout);
5082 mtx_unlock(&sc->sfl_lock);
5084 KASSERT((fl->flags & FL_STARVING) == 0,
5085 ("%s: still starving", __func__));
5089 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
5090 driver_intr_t *handler, void *arg, char *name)
5095 irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
5096 RF_SHAREABLE | RF_ACTIVE);
5097 if (irq->res == NULL) {
5098 device_printf(sc->dev,
5099 "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
5103 rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
5104 NULL, handler, arg, &irq->tag);
5106 device_printf(sc->dev,
5107 "failed to setup interrupt for rid %d, name %s: %d\n",
5110 bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
5116 t4_free_irq(struct adapter *sc, struct irq *irq)
5119 bus_teardown_intr(sc->dev, irq->res, irq->tag);
5121 bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
5123 bzero(irq, sizeof(*irq));
5129 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
5132 regs->version = chip_id(sc) | chip_rev(sc) << 10;
5133 t4_get_regs(sc, buf, regs->len);
5136 #define A_PL_INDIR_CMD 0x1f8
5138 #define S_PL_AUTOINC 31
5139 #define M_PL_AUTOINC 0x1U
5140 #define V_PL_AUTOINC(x) ((x) << S_PL_AUTOINC)
5141 #define G_PL_AUTOINC(x) (((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
5143 #define S_PL_VFID 20
5144 #define M_PL_VFID 0xffU
5145 #define V_PL_VFID(x) ((x) << S_PL_VFID)
5146 #define G_PL_VFID(x) (((x) >> S_PL_VFID) & M_PL_VFID)
5149 #define M_PL_ADDR 0xfffffU
5150 #define V_PL_ADDR(x) ((x) << S_PL_ADDR)
5151 #define G_PL_ADDR(x) (((x) >> S_PL_ADDR) & M_PL_ADDR)
5153 #define A_PL_INDIR_DATA 0x1fc
5156 read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
5160 mtx_assert(&sc->reg_lock, MA_OWNED);
5161 if (sc->flags & IS_VF) {
5162 stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
5163 stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
5165 t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5166 V_PL_VFID(G_FW_VIID_VIN(viid)) |
5167 V_PL_ADDR(VF_MPS_REG(reg)));
5168 stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
5169 stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
5171 return (((uint64_t)stats[1]) << 32 | stats[0]);
5175 t4_get_vi_stats(struct adapter *sc, unsigned int viid,
5176 struct fw_vi_stats_vf *stats)
5179 #define GET_STAT(name) \
5180 read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
5182 stats->tx_bcast_bytes = GET_STAT(TX_VF_BCAST_BYTES);
5183 stats->tx_bcast_frames = GET_STAT(TX_VF_BCAST_FRAMES);
5184 stats->tx_mcast_bytes = GET_STAT(TX_VF_MCAST_BYTES);
5185 stats->tx_mcast_frames = GET_STAT(TX_VF_MCAST_FRAMES);
5186 stats->tx_ucast_bytes = GET_STAT(TX_VF_UCAST_BYTES);
5187 stats->tx_ucast_frames = GET_STAT(TX_VF_UCAST_FRAMES);
5188 stats->tx_drop_frames = GET_STAT(TX_VF_DROP_FRAMES);
5189 stats->tx_offload_bytes = GET_STAT(TX_VF_OFFLOAD_BYTES);
5190 stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
5191 stats->rx_bcast_bytes = GET_STAT(RX_VF_BCAST_BYTES);
5192 stats->rx_bcast_frames = GET_STAT(RX_VF_BCAST_FRAMES);
5193 stats->rx_mcast_bytes = GET_STAT(RX_VF_MCAST_BYTES);
5194 stats->rx_mcast_frames = GET_STAT(RX_VF_MCAST_FRAMES);
5195 stats->rx_ucast_bytes = GET_STAT(RX_VF_UCAST_BYTES);
5196 stats->rx_ucast_frames = GET_STAT(RX_VF_UCAST_FRAMES);
5197 stats->rx_err_frames = GET_STAT(RX_VF_ERR_FRAMES);
5203 t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
5207 t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5208 V_PL_VFID(G_FW_VIID_VIN(viid)) |
5209 V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
5210 for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
5211 reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
5212 t4_write_reg(sc, A_PL_INDIR_DATA, 0);
5216 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
5219 const struct timeval interval = {0, 250000}; /* 250ms */
5221 if (!(vi->flags & VI_INIT_DONE))
5225 timevalsub(&tv, &interval);
5226 if (timevalcmp(&tv, &vi->last_refreshed, <))
5229 mtx_lock(&sc->reg_lock);
5230 t4_get_vi_stats(sc, vi->viid, &vi->stats);
5231 getmicrotime(&vi->last_refreshed);
5232 mtx_unlock(&sc->reg_lock);
5236 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
5238 u_int i, v, tnl_cong_drops, bg_map;
5240 const struct timeval interval = {0, 250000}; /* 250ms */
5243 timevalsub(&tv, &interval);
5244 if (timevalcmp(&tv, &pi->last_refreshed, <))
5248 t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
5249 bg_map = pi->mps_bg_map;
5251 i = ffs(bg_map) - 1;
5252 mtx_lock(&sc->reg_lock);
5253 t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v, 1,
5254 A_TP_MIB_TNL_CNG_DROP_0 + i);
5255 mtx_unlock(&sc->reg_lock);
5256 tnl_cong_drops += v;
5257 bg_map &= ~(1 << i);
5259 pi->tnl_cong_drops = tnl_cong_drops;
5260 getmicrotime(&pi->last_refreshed);
5264 cxgbe_tick(void *arg)
5266 struct port_info *pi = arg;
5267 struct adapter *sc = pi->adapter;
5269 PORT_LOCK_ASSERT_OWNED(pi);
5270 cxgbe_refresh_stats(sc, pi);
5272 callout_schedule(&pi->tick, hz);
5278 struct vi_info *vi = arg;
5279 struct adapter *sc = vi->pi->adapter;
5281 vi_refresh_stats(sc, vi);
5283 callout_schedule(&vi->tick, hz);
5287 cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
5291 if (arg != ifp || ifp->if_type != IFT_ETHER)
5294 vlan = VLAN_DEVAT(ifp, vid);
5295 VLAN_SETCOOKIE(vlan, ifp);
5299 * Should match fw_caps_config_<foo> enums in t4fw_interface.h
5301 static char *caps_decoder[] = {
5302 "\20\001IPMI\002NCSI", /* 0: NBM */
5303 "\20\001PPP\002QFC\003DCBX", /* 1: link */
5304 "\20\001INGRESS\002EGRESS", /* 2: switch */
5305 "\20\001NIC\002VM\003IDS\004UM\005UM_ISGL" /* 3: NIC */
5306 "\006HASHFILTER\007ETHOFLD",
5307 "\20\001TOE", /* 4: TOE */
5308 "\20\001RDDP\002RDMAC", /* 5: RDMA */
5309 "\20\001INITIATOR_PDU\002TARGET_PDU" /* 6: iSCSI */
5310 "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
5311 "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
5313 "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
5314 "\20\001LOOKASIDE\002TLSKEYS", /* 7: Crypto */
5315 "\20\001INITIATOR\002TARGET\003CTRL_OFLD" /* 8: FCoE */
5316 "\004PO_INITIATOR\005PO_TARGET",
5320 t4_sysctls(struct adapter *sc)
5322 struct sysctl_ctx_list *ctx;
5323 struct sysctl_oid *oid;
5324 struct sysctl_oid_list *children, *c0;
5325 static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
5327 ctx = device_get_sysctl_ctx(sc->dev);
5332 oid = device_get_sysctl_tree(sc->dev);
5333 c0 = children = SYSCTL_CHILDREN(oid);
5335 sc->sc_do_rxcopy = 1;
5336 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
5337 &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
5339 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
5340 sc->params.nports, "# of ports");
5342 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
5343 CTLTYPE_STRING | CTLFLAG_RD, doorbells, sc->doorbells,
5344 sysctl_bitfield, "A", "available doorbells");
5346 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
5347 sc->params.vpd.cclk, "core clock frequency (in KHz)");
5349 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
5350 CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
5351 sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
5352 "interrupt holdoff timer values (us)");
5354 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
5355 CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
5356 sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
5357 "interrupt holdoff packet counter values");
5359 t4_sge_sysctls(sc, ctx, children);
5361 sc->lro_timeout = 100;
5362 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
5363 &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
5365 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
5366 &sc->debug_flags, 0, "flags to enable runtime debugging");
5368 SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
5369 CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
5371 SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
5372 CTLFLAG_RD, sc->fw_version, 0, "firmware version");
5374 if (sc->flags & IS_VF)
5377 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
5378 NULL, chip_rev(sc), "chip hardware revision");
5380 SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
5381 CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
5383 SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
5384 CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
5386 SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
5387 CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
5389 SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "md_version",
5390 CTLFLAG_RD, sc->params.vpd.md, 0, "manufacturing diags version");
5392 SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
5393 CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
5395 SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
5396 sc->er_version, 0, "expansion ROM version");
5398 SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
5399 sc->bs_version, 0, "bootstrap firmware version");
5401 SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
5402 NULL, sc->params.scfg_vers, "serial config version");
5404 SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
5405 NULL, sc->params.vpd_vers, "VPD version");
5407 SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
5408 CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
5410 SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
5411 sc->cfcsum, "config file checksum");
5413 #define SYSCTL_CAP(name, n, text) \
5414 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
5415 CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], sc->name, \
5416 sysctl_bitfield, "A", "available " text " capabilities")
5418 SYSCTL_CAP(nbmcaps, 0, "NBM");
5419 SYSCTL_CAP(linkcaps, 1, "link");
5420 SYSCTL_CAP(switchcaps, 2, "switch");
5421 SYSCTL_CAP(niccaps, 3, "NIC");
5422 SYSCTL_CAP(toecaps, 4, "TCP offload");
5423 SYSCTL_CAP(rdmacaps, 5, "RDMA");
5424 SYSCTL_CAP(iscsicaps, 6, "iSCSI");
5425 SYSCTL_CAP(cryptocaps, 7, "crypto");
5426 SYSCTL_CAP(fcoecaps, 8, "FCoE");
5429 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
5430 NULL, sc->tids.nftids, "number of filters");
5432 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
5433 CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
5434 "chip temperature (in Celsius)");
5436 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_vdd", CTLFLAG_RD,
5437 &sc->params.core_vdd, 0, "core Vdd (in mV)");
5441 * dev.t4nex.X.misc. Marked CTLFLAG_SKIP to avoid information overload.
5443 oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
5444 CTLFLAG_RD | CTLFLAG_SKIP, NULL,
5445 "logs and miscellaneous information");
5446 children = SYSCTL_CHILDREN(oid);
5448 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
5449 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5450 sysctl_cctrl, "A", "congestion control");
5452 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
5453 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5454 sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
5456 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
5457 CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
5458 sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
5460 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
5461 CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
5462 sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
5464 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
5465 CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
5466 sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
5468 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
5469 CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
5470 sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
5472 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
5473 CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
5474 sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
5476 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
5477 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5478 chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
5479 "A", "CIM logic analyzer");
5481 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
5482 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5483 sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
5485 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
5486 CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
5487 sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
5489 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
5490 CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
5491 sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
5493 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
5494 CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
5495 sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
5497 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
5498 CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
5499 sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
5501 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
5502 CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
5503 sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
5505 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
5506 CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
5507 sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
5509 if (chip_id(sc) > CHELSIO_T4) {
5510 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
5511 CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
5512 sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
5514 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
5515 CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
5516 sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
5519 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
5520 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5521 sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
5523 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
5524 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5525 sysctl_cim_qcfg, "A", "CIM queue configuration");
5527 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
5528 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5529 sysctl_cpl_stats, "A", "CPL statistics");
5531 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
5532 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5533 sysctl_ddp_stats, "A", "non-TCP DDP statistics");
5535 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
5536 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5537 sysctl_devlog, "A", "firmware's device log");
5539 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
5540 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5541 sysctl_fcoe_stats, "A", "FCoE statistics");
5543 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
5544 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5545 sysctl_hw_sched, "A", "hardware scheduler ");
5547 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
5548 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5549 sysctl_l2t, "A", "hardware L2 table");
5551 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
5552 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5553 sysctl_lb_stats, "A", "loopback statistics");
5555 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
5556 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5557 sysctl_meminfo, "A", "memory regions");
5559 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
5560 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5561 chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
5562 "A", "MPS TCAM entries");
5564 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
5565 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5566 sysctl_path_mtus, "A", "path MTUs");
5568 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
5569 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5570 sysctl_pm_stats, "A", "PM statistics");
5572 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
5573 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5574 sysctl_rdma_stats, "A", "RDMA statistics");
5576 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
5577 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5578 sysctl_tcp_stats, "A", "TCP statistics");
5580 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
5581 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5582 sysctl_tids, "A", "TID information");
5584 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
5585 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5586 sysctl_tp_err_stats, "A", "TP error statistics");
5588 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
5589 CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
5590 "TP logic analyzer event capture mask");
5592 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
5593 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5594 sysctl_tp_la, "A", "TP logic analyzer");
5596 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
5597 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5598 sysctl_tx_rate, "A", "Tx rate");
5600 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
5601 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5602 sysctl_ulprx_la, "A", "ULPRX logic analyzer");
5604 if (chip_id(sc) >= CHELSIO_T5) {
5605 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
5606 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5607 sysctl_wcwr_stats, "A", "write combined work requests");
5612 if (is_offload(sc)) {
5619 oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
5620 NULL, "TOE parameters");
5621 children = SYSCTL_CHILDREN(oid);
5623 sc->tt.cong_algorithm = -1;
5624 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "cong_algorithm",
5625 CTLFLAG_RW, &sc->tt.cong_algorithm, 0, "congestion control "
5626 "(-1 = default, 0 = reno, 1 = tahoe, 2 = newreno, "
5629 sc->tt.sndbuf = 256 * 1024;
5630 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
5631 &sc->tt.sndbuf, 0, "max hardware send buffer size");
5634 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
5635 &sc->tt.ddp, 0, "DDP allowed");
5637 sc->tt.rx_coalesce = 1;
5638 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
5639 CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
5642 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tls", CTLFLAG_RW,
5643 &sc->tt.tls, 0, "Inline TLS allowed");
5645 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tls_rx_ports",
5646 CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tls_rx_ports,
5647 "I", "TCP ports that use inline TLS+TOE RX");
5649 sc->tt.tx_align = 1;
5650 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
5651 CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
5653 sc->tt.tx_zcopy = 0;
5654 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
5655 CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
5656 "Enable zero-copy aio_write(2)");
5658 sc->tt.cop_managed_offloading = !!t4_cop_managed_offloading;
5659 SYSCTL_ADD_INT(ctx, children, OID_AUTO,
5660 "cop_managed_offloading", CTLFLAG_RW,
5661 &sc->tt.cop_managed_offloading, 0,
5662 "COP (Connection Offload Policy) controls all TOE offload");
5664 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
5665 CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
5666 "TP timer tick (us)");
5668 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
5669 CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
5670 "TCP timestamp tick (us)");
5672 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
5673 CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
5676 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
5677 CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
5678 "IU", "DACK timer (us)");
5680 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
5681 CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
5682 sysctl_tp_timer, "LU", "Minimum retransmit interval (us)");
5684 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
5685 CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
5686 sysctl_tp_timer, "LU", "Maximum retransmit interval (us)");
5688 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
5689 CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
5690 sysctl_tp_timer, "LU", "Persist timer min (us)");
5692 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
5693 CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
5694 sysctl_tp_timer, "LU", "Persist timer max (us)");
5696 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
5697 CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
5698 sysctl_tp_timer, "LU", "Keepalive idle timer (us)");
5700 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_interval",
5701 CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
5702 sysctl_tp_timer, "LU", "Keepalive interval timer (us)");
5704 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
5705 CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
5706 sysctl_tp_timer, "LU", "Initial SRTT (us)");
5708 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
5709 CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
5710 sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
5712 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "syn_rexmt_count",
5713 CTLTYPE_UINT | CTLFLAG_RD, sc, S_SYNSHIFTMAX,
5714 sysctl_tp_shift_cnt, "IU",
5715 "Number of SYN retransmissions before abort");
5717 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_count",
5718 CTLTYPE_UINT | CTLFLAG_RD, sc, S_RXTSHIFTMAXR2,
5719 sysctl_tp_shift_cnt, "IU",
5720 "Number of retransmissions before abort");
5722 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_count",
5723 CTLTYPE_UINT | CTLFLAG_RD, sc, S_KEEPALIVEMAXR2,
5724 sysctl_tp_shift_cnt, "IU",
5725 "Number of keepalive probes before abort");
5727 oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "rexmt_backoff",
5728 CTLFLAG_RD, NULL, "TOE retransmit backoffs");
5729 children = SYSCTL_CHILDREN(oid);
5730 for (i = 0; i < 16; i++) {
5731 snprintf(s, sizeof(s), "%u", i);
5732 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, s,
5733 CTLTYPE_UINT | CTLFLAG_RD, sc, i, sysctl_tp_backoff,
5734 "IU", "TOE retransmit backoff");
5741 vi_sysctls(struct vi_info *vi)
5743 struct sysctl_ctx_list *ctx;
5744 struct sysctl_oid *oid;
5745 struct sysctl_oid_list *children;
5747 ctx = device_get_sysctl_ctx(vi->dev);
5750 * dev.v?(cxgbe|cxl).X.
5752 oid = device_get_sysctl_tree(vi->dev);
5753 children = SYSCTL_CHILDREN(oid);
5755 SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
5756 vi->viid, "VI identifer");
5757 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
5758 &vi->nrxq, 0, "# of rx queues");
5759 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
5760 &vi->ntxq, 0, "# of tx queues");
5761 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
5762 &vi->first_rxq, 0, "index of first rx queue");
5763 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
5764 &vi->first_txq, 0, "index of first tx queue");
5765 SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
5766 vi->rss_size, "size of RSS indirection table");
5768 if (IS_MAIN_VI(vi)) {
5769 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
5770 CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
5771 "Reserve queue 0 for non-flowid packets");
5775 if (vi->nofldrxq != 0) {
5776 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
5778 "# of rx queues for offloaded TCP connections");
5779 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
5781 "# of tx queues for offloaded TCP connections");
5782 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
5783 CTLFLAG_RD, &vi->first_ofld_rxq, 0,
5784 "index of first TOE rx queue");
5785 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
5786 CTLFLAG_RD, &vi->first_ofld_txq, 0,
5787 "index of first TOE tx queue");
5788 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx_ofld",
5789 CTLTYPE_INT | CTLFLAG_RW, vi, 0,
5790 sysctl_holdoff_tmr_idx_ofld, "I",
5791 "holdoff timer index for TOE queues");
5792 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx_ofld",
5793 CTLTYPE_INT | CTLFLAG_RW, vi, 0,
5794 sysctl_holdoff_pktc_idx_ofld, "I",
5795 "holdoff packet counter index for TOE queues");
5799 if (vi->nnmrxq != 0) {
5800 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
5801 &vi->nnmrxq, 0, "# of netmap rx queues");
5802 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
5803 &vi->nnmtxq, 0, "# of netmap tx queues");
5804 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
5805 CTLFLAG_RD, &vi->first_nm_rxq, 0,
5806 "index of first netmap rx queue");
5807 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
5808 CTLFLAG_RD, &vi->first_nm_txq, 0,
5809 "index of first netmap tx queue");
5813 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
5814 CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
5815 "holdoff timer index");
5816 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
5817 CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
5818 "holdoff packet counter index");
5820 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
5821 CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
5823 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
5824 CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
5829 cxgbe_sysctls(struct port_info *pi)
5831 struct sysctl_ctx_list *ctx;
5832 struct sysctl_oid *oid;
5833 struct sysctl_oid_list *children, *children2;
5834 struct adapter *sc = pi->adapter;
5838 ctx = device_get_sysctl_ctx(pi->dev);
5843 oid = device_get_sysctl_tree(pi->dev);
5844 children = SYSCTL_CHILDREN(oid);
5846 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
5847 CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
5848 if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
5849 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
5850 CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
5851 "PHY temperature (in Celsius)");
5852 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
5853 CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
5854 "PHY firmware version");
5857 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
5858 CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
5859 "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
5860 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
5861 CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
5862 "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
5863 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
5864 CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
5865 "autonegotiation (-1 = not supported)");
5867 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
5868 port_top_speed(pi), "max speed (in Gbps)");
5869 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "mps_bg_map", CTLFLAG_RD, NULL,
5870 pi->mps_bg_map, "MPS buffer group map");
5871 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_e_chan_map", CTLFLAG_RD,
5872 NULL, pi->rx_e_chan_map, "TP rx e-channel map");
5874 if (sc->flags & IS_VF)
5878 * dev.(cxgbe|cxl).X.tc.
5880 oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
5881 "Tx scheduler traffic classes (cl_rl)");
5882 for (i = 0; i < sc->chip_params->nsched_cls; i++) {
5883 struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
5885 snprintf(name, sizeof(name), "%d", i);
5886 children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
5887 SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
5889 SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "flags", CTLFLAG_RD,
5890 &tc->flags, 0, "flags");
5891 SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
5892 CTLFLAG_RD, &tc->refcount, 0, "references to this class");
5894 SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
5895 CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
5896 sysctl_tc_params, "A", "traffic class parameters");
5901 * dev.cxgbe.X.stats.
5903 oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
5904 NULL, "port statistics");
5905 children = SYSCTL_CHILDREN(oid);
5906 SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
5907 &pi->tx_parse_error, 0,
5908 "# of tx packets with invalid length or # of segments");
5910 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
5911 SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
5912 CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
5913 sysctl_handle_t4_reg64, "QU", desc)
5915 SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
5916 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
5917 SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
5918 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
5919 SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
5920 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
5921 SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
5922 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
5923 SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
5924 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
5925 SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
5926 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
5927 SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
5928 "# of tx frames in this range",
5929 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
5930 SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
5931 "# of tx frames in this range",
5932 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
5933 SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
5934 "# of tx frames in this range",
5935 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
5936 SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
5937 "# of tx frames in this range",
5938 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
5939 SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
5940 "# of tx frames in this range",
5941 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
5942 SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
5943 "# of tx frames in this range",
5944 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
5945 SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
5946 "# of tx frames in this range",
5947 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
5948 SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
5949 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
5950 SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
5951 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
5952 SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
5953 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
5954 SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
5955 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
5956 SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
5957 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
5958 SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
5959 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
5960 SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
5961 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
5962 SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
5963 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
5964 SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
5965 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
5966 SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
5967 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
5969 SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
5970 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
5971 SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
5972 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
5973 SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
5974 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
5975 SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
5976 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
5977 SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
5978 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
5979 SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
5980 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
5981 SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
5982 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
5983 SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
5984 "# of frames received with bad FCS",
5985 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
5986 SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
5987 "# of frames received with length error",
5988 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
5989 SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
5990 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
5991 SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
5992 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
5993 SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
5994 "# of rx frames in this range",
5995 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
5996 SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
5997 "# of rx frames in this range",
5998 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
5999 SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
6000 "# of rx frames in this range",
6001 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
6002 SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
6003 "# of rx frames in this range",
6004 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
6005 SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
6006 "# of rx frames in this range",
6007 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
6008 SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
6009 "# of rx frames in this range",
6010 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
6011 SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
6012 "# of rx frames in this range",
6013 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
6014 SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
6015 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
6016 SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
6017 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
6018 SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
6019 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
6020 SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
6021 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
6022 SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
6023 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
6024 SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
6025 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
6026 SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
6027 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
6028 SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
6029 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
6030 SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
6031 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
6033 #undef SYSCTL_ADD_T4_REG64
6035 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
6036 SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
6037 &pi->stats.name, desc)
6039 /* We get these from port_stats and they may be stale by up to 1s */
6040 SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
6041 "# drops due to buffer-group 0 overflows");
6042 SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
6043 "# drops due to buffer-group 1 overflows");
6044 SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
6045 "# drops due to buffer-group 2 overflows");
6046 SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
6047 "# drops due to buffer-group 3 overflows");
6048 SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
6049 "# of buffer-group 0 truncated packets");
6050 SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
6051 "# of buffer-group 1 truncated packets");
6052 SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
6053 "# of buffer-group 2 truncated packets");
6054 SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
6055 "# of buffer-group 3 truncated packets");
6057 #undef SYSCTL_ADD_T4_PORTSTAT
6059 SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_records",
6060 CTLFLAG_RD, &pi->tx_tls_records,
6061 "# of TLS records transmitted");
6062 SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_octets",
6063 CTLFLAG_RD, &pi->tx_tls_octets,
6064 "# of payload octets in transmitted TLS records");
6065 SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_records",
6066 CTLFLAG_RD, &pi->rx_tls_records,
6067 "# of TLS records received");
6068 SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_octets",
6069 CTLFLAG_RD, &pi->rx_tls_octets,
6070 "# of payload octets in received TLS records");
6074 sysctl_int_array(SYSCTL_HANDLER_ARGS)
6076 int rc, *i, space = 0;
6079 sbuf_new_for_sysctl(&sb, NULL, 64, req);
6080 for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
6082 sbuf_printf(&sb, " ");
6083 sbuf_printf(&sb, "%d", *i);
6086 rc = sbuf_finish(&sb);
6092 sysctl_bitfield(SYSCTL_HANDLER_ARGS)
6097 rc = sysctl_wire_old_buffer(req, 0);
6101 sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6105 sbuf_printf(sb, "%b", (int)arg2, (char *)arg1);
6106 rc = sbuf_finish(sb);
6113 sysctl_btphy(SYSCTL_HANDLER_ARGS)
6115 struct port_info *pi = arg1;
6117 struct adapter *sc = pi->adapter;
6121 rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
6124 /* XXX: magic numbers */
6125 rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
6127 end_synchronized_op(sc, 0);
6133 rc = sysctl_handle_int(oidp, &v, 0, req);
6138 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
6140 struct vi_info *vi = arg1;
6143 val = vi->rsrv_noflowq;
6144 rc = sysctl_handle_int(oidp, &val, 0, req);
6145 if (rc != 0 || req->newptr == NULL)
6148 if ((val >= 1) && (vi->ntxq > 1))
6149 vi->rsrv_noflowq = 1;
6151 vi->rsrv_noflowq = 0;
6157 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
6159 struct vi_info *vi = arg1;
6160 struct adapter *sc = vi->pi->adapter;
6162 struct sge_rxq *rxq;
6167 rc = sysctl_handle_int(oidp, &idx, 0, req);
6168 if (rc != 0 || req->newptr == NULL)
6171 if (idx < 0 || idx >= SGE_NTIMERS)
6174 rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6179 v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
6180 for_each_rxq(vi, i, rxq) {
6181 #ifdef atomic_store_rel_8
6182 atomic_store_rel_8(&rxq->iq.intr_params, v);
6184 rxq->iq.intr_params = v;
6189 end_synchronized_op(sc, LOCK_HELD);
6194 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
6196 struct vi_info *vi = arg1;
6197 struct adapter *sc = vi->pi->adapter;
6202 rc = sysctl_handle_int(oidp, &idx, 0, req);
6203 if (rc != 0 || req->newptr == NULL)
6206 if (idx < -1 || idx >= SGE_NCOUNTERS)
6209 rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6214 if (vi->flags & VI_INIT_DONE)
6215 rc = EBUSY; /* cannot be changed once the queues are created */
6219 end_synchronized_op(sc, LOCK_HELD);
6224 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
6226 struct vi_info *vi = arg1;
6227 struct adapter *sc = vi->pi->adapter;
6230 qsize = vi->qsize_rxq;
6232 rc = sysctl_handle_int(oidp, &qsize, 0, req);
6233 if (rc != 0 || req->newptr == NULL)
6236 if (qsize < 128 || (qsize & 7))
6239 rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6244 if (vi->flags & VI_INIT_DONE)
6245 rc = EBUSY; /* cannot be changed once the queues are created */
6247 vi->qsize_rxq = qsize;
6249 end_synchronized_op(sc, LOCK_HELD);
6254 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
6256 struct vi_info *vi = arg1;
6257 struct adapter *sc = vi->pi->adapter;
6260 qsize = vi->qsize_txq;
6262 rc = sysctl_handle_int(oidp, &qsize, 0, req);
6263 if (rc != 0 || req->newptr == NULL)
6266 if (qsize < 128 || qsize > 65536)
6269 rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6274 if (vi->flags & VI_INIT_DONE)
6275 rc = EBUSY; /* cannot be changed once the queues are created */
6277 vi->qsize_txq = qsize;
6279 end_synchronized_op(sc, LOCK_HELD);
6284 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
6286 struct port_info *pi = arg1;
6287 struct adapter *sc = pi->adapter;
6288 struct link_config *lc = &pi->link_cfg;
6291 if (req->newptr == NULL) {
6293 static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
6295 rc = sysctl_wire_old_buffer(req, 0);
6299 sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6303 sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
6304 rc = sbuf_finish(sb);
6310 s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
6313 rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6319 if (s[0] < '0' || s[0] > '9')
6320 return (EINVAL); /* not a number */
6322 if (n & ~(PAUSE_TX | PAUSE_RX))
6323 return (EINVAL); /* some other bit is set too */
6325 rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6329 if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
6330 lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
6331 lc->requested_fc |= n;
6332 rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6334 lc->fc = lc->requested_fc;
6337 end_synchronized_op(sc, 0);
6344 sysctl_fec(SYSCTL_HANDLER_ARGS)
6346 struct port_info *pi = arg1;
6347 struct adapter *sc = pi->adapter;
6348 struct link_config *lc = &pi->link_cfg;
6351 if (req->newptr == NULL) {
6353 static char *bits = "\20\1RS\2BASER_RS\3RESERVED";
6355 rc = sysctl_wire_old_buffer(req, 0);
6359 sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6363 sbuf_printf(sb, "%b", lc->fec & M_FW_PORT_CAP_FEC, bits);
6364 rc = sbuf_finish(sb);
6370 s[0] = '0' + (lc->requested_fec & M_FW_PORT_CAP_FEC);
6373 rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6379 if (s[0] < '0' || s[0] > '9')
6380 return (EINVAL); /* not a number */
6382 if (n & ~M_FW_PORT_CAP_FEC)
6383 return (EINVAL); /* some other bit is set too */
6385 rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6389 if ((lc->requested_fec & M_FW_PORT_CAP_FEC) != n) {
6390 lc->requested_fec = n &
6391 G_FW_PORT_CAP_FEC(lc->supported);
6392 rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6394 lc->fec = lc->requested_fec;
6397 end_synchronized_op(sc, 0);
6404 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
6406 struct port_info *pi = arg1;
6407 struct adapter *sc = pi->adapter;
6408 struct link_config *lc = &pi->link_cfg;
6411 if (lc->supported & FW_PORT_CAP_ANEG)
6412 val = lc->requested_aneg == AUTONEG_ENABLE ? 1 : 0;
6415 rc = sysctl_handle_int(oidp, &val, 0, req);
6416 if (rc != 0 || req->newptr == NULL)
6418 if ((lc->supported & FW_PORT_CAP_ANEG) == 0)
6422 val = AUTONEG_DISABLE;
6424 val = AUTONEG_ENABLE;
6427 if (lc->requested_aneg == val)
6428 return (0); /* no change */
6430 rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6434 old = lc->requested_aneg;
6435 lc->requested_aneg = val;
6436 rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6438 lc->requested_aneg = old;
6439 end_synchronized_op(sc, 0);
6444 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
6446 struct adapter *sc = arg1;
6450 val = t4_read_reg64(sc, reg);
6452 return (sysctl_handle_64(oidp, &val, 0, req));
6456 sysctl_temperature(SYSCTL_HANDLER_ARGS)
6458 struct adapter *sc = arg1;
6460 uint32_t param, val;
6462 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
6465 param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
6466 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
6467 V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
6468 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, ¶m, &val);
6469 end_synchronized_op(sc, 0);
6473 /* unknown is returned as 0 but we display -1 in that case */
6474 t = val == 0 ? -1 : val;
6476 rc = sysctl_handle_int(oidp, &t, 0, req);
6482 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
6484 struct adapter *sc = arg1;
6487 uint16_t incr[NMTUS][NCCTRL_WIN];
6488 static const char *dec_fac[] = {
6489 "0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
6493 rc = sysctl_wire_old_buffer(req, 0);
6497 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6501 t4_read_cong_tbl(sc, incr);
6503 for (i = 0; i < NCCTRL_WIN; ++i) {
6504 sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
6505 incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
6506 incr[5][i], incr[6][i], incr[7][i]);
6507 sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
6508 incr[8][i], incr[9][i], incr[10][i], incr[11][i],
6509 incr[12][i], incr[13][i], incr[14][i], incr[15][i],
6510 sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
6513 rc = sbuf_finish(sb);
6519 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
6520 "TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI", /* ibq's */
6521 "ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI", /* obq's */
6522 "SGE0-RX", "SGE1-RX" /* additional obq's (T5 onwards) */
6526 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
6528 struct adapter *sc = arg1;
6530 int rc, i, n, qid = arg2;
6533 u_int cim_num_obq = sc->chip_params->cim_num_obq;
6535 KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
6536 ("%s: bad qid %d\n", __func__, qid));
6538 if (qid < CIM_NUM_IBQ) {
6541 n = 4 * CIM_IBQ_SIZE;
6542 buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6543 rc = t4_read_cim_ibq(sc, qid, buf, n);
6545 /* outbound queue */
6548 n = 4 * cim_num_obq * CIM_OBQ_SIZE;
6549 buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6550 rc = t4_read_cim_obq(sc, qid, buf, n);
6557 n = rc * sizeof(uint32_t); /* rc has # of words actually read */
6559 rc = sysctl_wire_old_buffer(req, 0);
6563 sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6569 sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
6570 for (i = 0, p = buf; i < n; i += 16, p += 4)
6571 sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
6574 rc = sbuf_finish(sb);
6582 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
6584 struct adapter *sc = arg1;
6590 MPASS(chip_id(sc) <= CHELSIO_T5);
6592 rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6596 rc = sysctl_wire_old_buffer(req, 0);
6600 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6604 buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6607 rc = -t4_cim_read_la(sc, buf, NULL);
6611 sbuf_printf(sb, "Status Data PC%s",
6612 cfg & F_UPDBGLACAPTPCONLY ? "" :
6613 " LS0Stat LS0Addr LS0Data");
6615 for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
6616 if (cfg & F_UPDBGLACAPTPCONLY) {
6617 sbuf_printf(sb, "\n %02x %08x %08x", p[5] & 0xff,
6619 sbuf_printf(sb, "\n %02x %02x%06x %02x%06x",
6620 (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
6621 p[4] & 0xff, p[5] >> 8);
6622 sbuf_printf(sb, "\n %02x %x%07x %x%07x",
6623 (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6624 p[1] & 0xf, p[2] >> 4);
6627 "\n %02x %x%07x %x%07x %08x %08x "
6629 (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6630 p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
6635 rc = sbuf_finish(sb);
6643 sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
6645 struct adapter *sc = arg1;
6651 MPASS(chip_id(sc) > CHELSIO_T5);
6653 rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6657 rc = sysctl_wire_old_buffer(req, 0);
6661 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6665 buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6668 rc = -t4_cim_read_la(sc, buf, NULL);
6672 sbuf_printf(sb, "Status Inst Data PC%s",
6673 cfg & F_UPDBGLACAPTPCONLY ? "" :
6674 " LS0Stat LS0Addr LS0Data LS1Stat LS1Addr LS1Data");
6676 for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
6677 if (cfg & F_UPDBGLACAPTPCONLY) {
6678 sbuf_printf(sb, "\n %02x %08x %08x %08x",
6679 p[3] & 0xff, p[2], p[1], p[0]);
6680 sbuf_printf(sb, "\n %02x %02x%06x %02x%06x %02x%06x",
6681 (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
6682 p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
6683 sbuf_printf(sb, "\n %02x %04x%04x %04x%04x %04x%04x",
6684 (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
6685 p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
6688 sbuf_printf(sb, "\n %02x %04x%04x %04x%04x %04x%04x "
6689 "%08x %08x %08x %08x %08x %08x",
6690 (p[9] >> 16) & 0xff,
6691 p[9] & 0xffff, p[8] >> 16,
6692 p[8] & 0xffff, p[7] >> 16,
6693 p[7] & 0xffff, p[6] >> 16,
6694 p[2], p[1], p[0], p[5], p[4], p[3]);
6698 rc = sbuf_finish(sb);
6706 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
6708 struct adapter *sc = arg1;
6714 rc = sysctl_wire_old_buffer(req, 0);
6718 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6722 buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
6725 t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
6728 for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6729 sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
6733 sbuf_printf(sb, "\n\nCnt ID Tag UE Data RDY VLD");
6734 for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6735 sbuf_printf(sb, "\n%3u %2u %x %u %08x%08x %u %u",
6736 (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
6737 (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
6738 (p[1] >> 2) | ((p[2] & 3) << 30),
6739 (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
6743 rc = sbuf_finish(sb);
6750 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
6752 struct adapter *sc = arg1;
6758 rc = sysctl_wire_old_buffer(req, 0);
6762 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6766 buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
6769 t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
6772 sbuf_printf(sb, "Cntl ID DataBE Addr Data");
6773 for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6774 sbuf_printf(sb, "\n %02x %02x %04x %08x %08x%08x%08x%08x",
6775 (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
6776 p[4], p[3], p[2], p[1], p[0]);
6779 sbuf_printf(sb, "\n\nCntl ID Data");
6780 for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6781 sbuf_printf(sb, "\n %02x %02x %08x%08x%08x%08x",
6782 (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
6785 rc = sbuf_finish(sb);
6792 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
6794 struct adapter *sc = arg1;
6797 uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6798 uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6799 uint16_t thres[CIM_NUM_IBQ];
6800 uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
6801 uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
6802 u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
6804 cim_num_obq = sc->chip_params->cim_num_obq;
6806 ibq_rdaddr = A_UP_IBQ_0_RDADDR;
6807 obq_rdaddr = A_UP_OBQ_0_REALADDR;
6809 ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
6810 obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
6812 nq = CIM_NUM_IBQ + cim_num_obq;
6814 rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
6816 rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
6820 t4_read_cimq_cfg(sc, base, size, thres);
6822 rc = sysctl_wire_old_buffer(req, 0);
6826 sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6831 " Queue Base Size Thres RdPtr WrPtr SOP EOP Avail");
6833 for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
6834 sbuf_printf(sb, "\n%7s %5x %5u %5u %6x %4x %4u %4u %5u",
6835 qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
6836 G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6837 G_QUEREMFLITS(p[2]) * 16);
6838 for ( ; i < nq; i++, p += 4, wr += 2)
6839 sbuf_printf(sb, "\n%7s %5x %5u %12x %4x %4u %4u %5u", qname[i],
6840 base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
6841 wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6842 G_QUEREMFLITS(p[2]) * 16);
6844 rc = sbuf_finish(sb);
6851 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
6853 struct adapter *sc = arg1;
6856 struct tp_cpl_stats stats;
6858 rc = sysctl_wire_old_buffer(req, 0);
6862 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6866 mtx_lock(&sc->reg_lock);
6867 t4_tp_get_cpl_stats(sc, &stats, 0);
6868 mtx_unlock(&sc->reg_lock);
6870 if (sc->chip_params->nchan > 2) {
6871 sbuf_printf(sb, " channel 0 channel 1"
6872 " channel 2 channel 3");
6873 sbuf_printf(sb, "\nCPL requests: %10u %10u %10u %10u",
6874 stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
6875 sbuf_printf(sb, "\nCPL responses: %10u %10u %10u %10u",
6876 stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
6878 sbuf_printf(sb, " channel 0 channel 1");
6879 sbuf_printf(sb, "\nCPL requests: %10u %10u",
6880 stats.req[0], stats.req[1]);
6881 sbuf_printf(sb, "\nCPL responses: %10u %10u",
6882 stats.rsp[0], stats.rsp[1]);
6885 rc = sbuf_finish(sb);
6892 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
6894 struct adapter *sc = arg1;
6897 struct tp_usm_stats stats;
6899 rc = sysctl_wire_old_buffer(req, 0);
6903 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6907 t4_get_usm_stats(sc, &stats, 1);
6909 sbuf_printf(sb, "Frames: %u\n", stats.frames);
6910 sbuf_printf(sb, "Octets: %ju\n", stats.octets);
6911 sbuf_printf(sb, "Drops: %u", stats.drops);
6913 rc = sbuf_finish(sb);
6919 static const char * const devlog_level_strings[] = {
6920 [FW_DEVLOG_LEVEL_EMERG] = "EMERG",
6921 [FW_DEVLOG_LEVEL_CRIT] = "CRIT",
6922 [FW_DEVLOG_LEVEL_ERR] = "ERR",
6923 [FW_DEVLOG_LEVEL_NOTICE] = "NOTICE",
6924 [FW_DEVLOG_LEVEL_INFO] = "INFO",
6925 [FW_DEVLOG_LEVEL_DEBUG] = "DEBUG"
6928 static const char * const devlog_facility_strings[] = {
6929 [FW_DEVLOG_FACILITY_CORE] = "CORE",
6930 [FW_DEVLOG_FACILITY_CF] = "CF",
6931 [FW_DEVLOG_FACILITY_SCHED] = "SCHED",
6932 [FW_DEVLOG_FACILITY_TIMER] = "TIMER",
6933 [FW_DEVLOG_FACILITY_RES] = "RES",
6934 [FW_DEVLOG_FACILITY_HW] = "HW",
6935 [FW_DEVLOG_FACILITY_FLR] = "FLR",
6936 [FW_DEVLOG_FACILITY_DMAQ] = "DMAQ",
6937 [FW_DEVLOG_FACILITY_PHY] = "PHY",
6938 [FW_DEVLOG_FACILITY_MAC] = "MAC",
6939 [FW_DEVLOG_FACILITY_PORT] = "PORT",
6940 [FW_DEVLOG_FACILITY_VI] = "VI",
6941 [FW_DEVLOG_FACILITY_FILTER] = "FILTER",
6942 [FW_DEVLOG_FACILITY_ACL] = "ACL",
6943 [FW_DEVLOG_FACILITY_TM] = "TM",
6944 [FW_DEVLOG_FACILITY_QFC] = "QFC",
6945 [FW_DEVLOG_FACILITY_DCB] = "DCB",
6946 [FW_DEVLOG_FACILITY_ETH] = "ETH",
6947 [FW_DEVLOG_FACILITY_OFLD] = "OFLD",
6948 [FW_DEVLOG_FACILITY_RI] = "RI",
6949 [FW_DEVLOG_FACILITY_ISCSI] = "ISCSI",
6950 [FW_DEVLOG_FACILITY_FCOE] = "FCOE",
6951 [FW_DEVLOG_FACILITY_FOISCSI] = "FOISCSI",
6952 [FW_DEVLOG_FACILITY_FOFCOE] = "FOFCOE",
6953 [FW_DEVLOG_FACILITY_CHNET] = "CHNET",
6957 sysctl_devlog(SYSCTL_HANDLER_ARGS)
6959 struct adapter *sc = arg1;
6960 struct devlog_params *dparams = &sc->params.devlog;
6961 struct fw_devlog_e *buf, *e;
6962 int i, j, rc, nentries, first = 0;
6964 uint64_t ftstamp = UINT64_MAX;
6966 if (dparams->addr == 0)
6969 buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
6973 rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
6977 nentries = dparams->size / sizeof(struct fw_devlog_e);
6978 for (i = 0; i < nentries; i++) {
6981 if (e->timestamp == 0)
6984 e->timestamp = be64toh(e->timestamp);
6985 e->seqno = be32toh(e->seqno);
6986 for (j = 0; j < 8; j++)
6987 e->params[j] = be32toh(e->params[j]);
6989 if (e->timestamp < ftstamp) {
6990 ftstamp = e->timestamp;
6995 if (buf[first].timestamp == 0)
6996 goto done; /* nothing in the log */
6998 rc = sysctl_wire_old_buffer(req, 0);
7002 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7007 sbuf_printf(sb, "%10s %15s %8s %8s %s\n",
7008 "Seq#", "Tstamp", "Level", "Facility", "Message");
7013 if (e->timestamp == 0)
7016 sbuf_printf(sb, "%10d %15ju %8s %8s ",
7017 e->seqno, e->timestamp,
7018 (e->level < nitems(devlog_level_strings) ?
7019 devlog_level_strings[e->level] : "UNKNOWN"),
7020 (e->facility < nitems(devlog_facility_strings) ?
7021 devlog_facility_strings[e->facility] : "UNKNOWN"));
7022 sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
7023 e->params[2], e->params[3], e->params[4],
7024 e->params[5], e->params[6], e->params[7]);
7026 if (++i == nentries)
7028 } while (i != first);
7030 rc = sbuf_finish(sb);
7038 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
7040 struct adapter *sc = arg1;
7043 struct tp_fcoe_stats stats[MAX_NCHAN];
7044 int i, nchan = sc->chip_params->nchan;
7046 rc = sysctl_wire_old_buffer(req, 0);
7050 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7054 for (i = 0; i < nchan; i++)
7055 t4_get_fcoe_stats(sc, i, &stats[i], 1);
7058 sbuf_printf(sb, " channel 0 channel 1"
7059 " channel 2 channel 3");
7060 sbuf_printf(sb, "\noctetsDDP: %16ju %16ju %16ju %16ju",
7061 stats[0].octets_ddp, stats[1].octets_ddp,
7062 stats[2].octets_ddp, stats[3].octets_ddp);
7063 sbuf_printf(sb, "\nframesDDP: %16u %16u %16u %16u",
7064 stats[0].frames_ddp, stats[1].frames_ddp,
7065 stats[2].frames_ddp, stats[3].frames_ddp);
7066 sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
7067 stats[0].frames_drop, stats[1].frames_drop,
7068 stats[2].frames_drop, stats[3].frames_drop);
7070 sbuf_printf(sb, " channel 0 channel 1");
7071 sbuf_printf(sb, "\noctetsDDP: %16ju %16ju",
7072 stats[0].octets_ddp, stats[1].octets_ddp);
7073 sbuf_printf(sb, "\nframesDDP: %16u %16u",
7074 stats[0].frames_ddp, stats[1].frames_ddp);
7075 sbuf_printf(sb, "\nframesDrop: %16u %16u",
7076 stats[0].frames_drop, stats[1].frames_drop);
7079 rc = sbuf_finish(sb);
7086 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
7088 struct adapter *sc = arg1;
7091 unsigned int map, kbps, ipg, mode;
7092 unsigned int pace_tab[NTX_SCHED];
7094 rc = sysctl_wire_old_buffer(req, 0);
7098 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7102 map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
7103 mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
7104 t4_read_pace_tbl(sc, pace_tab);
7106 sbuf_printf(sb, "Scheduler Mode Channel Rate (Kbps) "
7107 "Class IPG (0.1 ns) Flow IPG (us)");
7109 for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
7110 t4_get_tx_sched(sc, i, &kbps, &ipg, 1);
7111 sbuf_printf(sb, "\n %u %-5s %u ", i,
7112 (mode & (1 << i)) ? "flow" : "class", map & 3);
7114 sbuf_printf(sb, "%9u ", kbps);
7116 sbuf_printf(sb, " disabled ");
7119 sbuf_printf(sb, "%13u ", ipg);
7121 sbuf_printf(sb, " disabled ");
7124 sbuf_printf(sb, "%10u", pace_tab[i]);
7126 sbuf_printf(sb, " disabled");
7129 rc = sbuf_finish(sb);
7136 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
7138 struct adapter *sc = arg1;
7142 struct lb_port_stats s[2];
7143 static const char *stat_name[] = {
7144 "OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
7145 "UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
7146 "Frames128To255:", "Frames256To511:", "Frames512To1023:",
7147 "Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
7148 "BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
7149 "BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
7150 "BG2FramesTrunc:", "BG3FramesTrunc:"
7153 rc = sysctl_wire_old_buffer(req, 0);
7157 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7161 memset(s, 0, sizeof(s));
7163 for (i = 0; i < sc->chip_params->nchan; i += 2) {
7164 t4_get_lb_stats(sc, i, &s[0]);
7165 t4_get_lb_stats(sc, i + 1, &s[1]);
7169 sbuf_printf(sb, "%s Loopback %u"
7170 " Loopback %u", i == 0 ? "" : "\n", i, i + 1);
7172 for (j = 0; j < nitems(stat_name); j++)
7173 sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
7177 rc = sbuf_finish(sb);
7184 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
7187 struct port_info *pi = arg1;
7188 struct link_config *lc = &pi->link_cfg;
7191 rc = sysctl_wire_old_buffer(req, 0);
7194 sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
7198 if (lc->link_ok || lc->link_down_rc == 255)
7199 sbuf_printf(sb, "n/a");
7201 sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
7203 rc = sbuf_finish(sb);
7216 mem_desc_cmp(const void *a, const void *b)
7218 return ((const struct mem_desc *)a)->base -
7219 ((const struct mem_desc *)b)->base;
7223 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
7231 size = to - from + 1;
7235 /* XXX: need humanize_number(3) in libkern for a more readable 'size' */
7236 sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
7240 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
7242 struct adapter *sc = arg1;
7245 uint32_t lo, hi, used, alloc;
7246 static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
7247 static const char *region[] = {
7248 "DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
7249 "Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
7250 "Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
7251 "TDDP region:", "TPT region:", "STAG region:", "RQ region:",
7252 "RQUDP region:", "PBL region:", "TXPBL region:",
7253 "DBVFIFO region:", "ULPRX state:", "ULPTX state:",
7254 "On-chip queues:", "TLS keys:",
7256 struct mem_desc avail[4];
7257 struct mem_desc mem[nitems(region) + 3]; /* up to 3 holes */
7258 struct mem_desc *md = mem;
7260 rc = sysctl_wire_old_buffer(req, 0);
7264 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7268 for (i = 0; i < nitems(mem); i++) {
7273 /* Find and sort the populated memory ranges */
7275 lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
7276 if (lo & F_EDRAM0_ENABLE) {
7277 hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
7278 avail[i].base = G_EDRAM0_BASE(hi) << 20;
7279 avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
7283 if (lo & F_EDRAM1_ENABLE) {
7284 hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
7285 avail[i].base = G_EDRAM1_BASE(hi) << 20;
7286 avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
7290 if (lo & F_EXT_MEM_ENABLE) {
7291 hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
7292 avail[i].base = G_EXT_MEM_BASE(hi) << 20;
7293 avail[i].limit = avail[i].base +
7294 (G_EXT_MEM_SIZE(hi) << 20);
7295 avail[i].idx = is_t5(sc) ? 3 : 2; /* Call it MC0 for T5 */
7298 if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
7299 hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
7300 avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
7301 avail[i].limit = avail[i].base +
7302 (G_EXT_MEM1_SIZE(hi) << 20);
7306 if (!i) /* no memory available */
7308 qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
7310 (md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
7311 (md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
7312 (md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
7313 (md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
7314 (md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
7315 (md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
7316 (md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
7317 (md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
7318 (md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
7320 /* the next few have explicit upper bounds */
7321 md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
7322 md->limit = md->base - 1 +
7323 t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
7324 G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
7327 md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
7328 md->limit = md->base - 1 +
7329 t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
7330 G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
7333 if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7334 if (chip_id(sc) <= CHELSIO_T5)
7335 md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
7337 md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
7341 md->idx = nitems(region); /* hide it */
7345 #define ulp_region(reg) \
7346 md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
7347 (md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
7349 ulp_region(RX_ISCSI);
7350 ulp_region(RX_TDDP);
7352 ulp_region(RX_STAG);
7354 ulp_region(RX_RQUDP);
7360 md->idx = nitems(region);
7363 uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
7364 uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
7367 if (sge_ctrl & F_VFIFO_ENABLE)
7368 size = G_DBVFIFO_SIZE(fifo_size);
7370 size = G_T6_DBVFIFO_SIZE(fifo_size);
7373 md->base = G_BASEADDR(t4_read_reg(sc,
7374 A_SGE_DBVFIFO_BADDR));
7375 md->limit = md->base + (size << 2) - 1;
7380 md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
7383 md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
7387 md->base = sc->vres.ocq.start;
7388 if (sc->vres.ocq.size)
7389 md->limit = md->base + sc->vres.ocq.size - 1;
7391 md->idx = nitems(region); /* hide it */
7394 md->base = sc->vres.key.start;
7395 if (sc->vres.key.size)
7396 md->limit = md->base + sc->vres.key.size - 1;
7398 md->idx = nitems(region); /* hide it */
7401 /* add any address-space holes, there can be up to 3 */
7402 for (n = 0; n < i - 1; n++)
7403 if (avail[n].limit < avail[n + 1].base)
7404 (md++)->base = avail[n].limit;
7406 (md++)->base = avail[n].limit;
7409 qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
7411 for (lo = 0; lo < i; lo++)
7412 mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
7413 avail[lo].limit - 1);
7415 sbuf_printf(sb, "\n");
7416 for (i = 0; i < n; i++) {
7417 if (mem[i].idx >= nitems(region))
7418 continue; /* skip holes */
7420 mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
7421 mem_region_show(sb, region[mem[i].idx], mem[i].base,
7425 sbuf_printf(sb, "\n");
7426 lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
7427 hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
7428 mem_region_show(sb, "uP RAM:", lo, hi);
7430 lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
7431 hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
7432 mem_region_show(sb, "uP Extmem2:", lo, hi);
7434 lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
7435 sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
7437 t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
7438 (lo & F_PMRXNUMCHN) ? 2 : 1);
7440 lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
7441 hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
7442 sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
7444 hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
7445 hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
7446 sbuf_printf(sb, "%u p-structs\n",
7447 t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
7449 for (i = 0; i < 4; i++) {
7450 if (chip_id(sc) > CHELSIO_T5)
7451 lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
7453 lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
7455 used = G_T5_USED(lo);
7456 alloc = G_T5_ALLOC(lo);
7459 alloc = G_ALLOC(lo);
7461 /* For T6 these are MAC buffer groups */
7462 sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
7465 for (i = 0; i < sc->chip_params->nchan; i++) {
7466 if (chip_id(sc) > CHELSIO_T5)
7467 lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
7469 lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
7471 used = G_T5_USED(lo);
7472 alloc = G_T5_ALLOC(lo);
7475 alloc = G_ALLOC(lo);
7477 /* For T6 these are MAC buffer groups */
7479 "\nLoopback %d using %u pages out of %u allocated",
7483 rc = sbuf_finish(sb);
7490 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
7494 memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
7498 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
7500 struct adapter *sc = arg1;
7504 MPASS(chip_id(sc) <= CHELSIO_T5);
7506 rc = sysctl_wire_old_buffer(req, 0);
7510 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7515 "Idx Ethernet address Mask Vld Ports PF"
7516 " VF Replication P0 P1 P2 P3 ML");
7517 for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7518 uint64_t tcamx, tcamy, mask;
7519 uint32_t cls_lo, cls_hi;
7520 uint8_t addr[ETHER_ADDR_LEN];
7522 tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
7523 tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
7526 tcamxy2valmask(tcamx, tcamy, addr, &mask);
7527 cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7528 cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7529 sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
7530 " %c %#x%4u%4d", i, addr[0], addr[1], addr[2],
7531 addr[3], addr[4], addr[5], (uintmax_t)mask,
7532 (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
7533 G_PORTMAP(cls_hi), G_PF(cls_lo),
7534 (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
7536 if (cls_lo & F_REPLICATE) {
7537 struct fw_ldst_cmd ldst_cmd;
7539 memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7540 ldst_cmd.op_to_addrspace =
7541 htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7542 F_FW_CMD_REQUEST | F_FW_CMD_READ |
7543 V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7544 ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7545 ldst_cmd.u.mps.rplc.fid_idx =
7546 htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7547 V_FW_LDST_CMD_IDX(i));
7549 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7553 rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7554 sizeof(ldst_cmd), &ldst_cmd);
7555 end_synchronized_op(sc, 0);
7558 sbuf_printf(sb, "%36d", rc);
7561 sbuf_printf(sb, " %08x %08x %08x %08x",
7562 be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7563 be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7564 be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7565 be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7568 sbuf_printf(sb, "%36s", "");
7570 sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
7571 G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
7572 G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
7576 (void) sbuf_finish(sb);
7578 rc = sbuf_finish(sb);
7585 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
7587 struct adapter *sc = arg1;
7591 MPASS(chip_id(sc) > CHELSIO_T5);
7593 rc = sysctl_wire_old_buffer(req, 0);
7597 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7601 sbuf_printf(sb, "Idx Ethernet address Mask VNI Mask"
7602 " IVLAN Vld DIP_Hit Lookup Port Vld Ports PF VF"
7604 " P0 P1 P2 P3 ML\n");
7606 for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7607 uint8_t dip_hit, vlan_vld, lookup_type, port_num;
7609 uint64_t tcamx, tcamy, val, mask;
7610 uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
7611 uint8_t addr[ETHER_ADDR_LEN];
7613 ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
7615 ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
7617 ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
7618 t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7619 val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7620 tcamy = G_DMACH(val) << 32;
7621 tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7622 data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7623 lookup_type = G_DATALKPTYPE(data2);
7624 port_num = G_DATAPORTNUM(data2);
7625 if (lookup_type && lookup_type != M_DATALKPTYPE) {
7626 /* Inner header VNI */
7627 vniy = ((data2 & F_DATAVIDH2) << 23) |
7628 (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7629 dip_hit = data2 & F_DATADIPHIT;
7634 vlan_vld = data2 & F_DATAVIDH2;
7635 ivlan = G_VIDL(val);
7638 ctl |= V_CTLXYBITSEL(1);
7639 t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7640 val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7641 tcamx = G_DMACH(val) << 32;
7642 tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7643 data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7644 if (lookup_type && lookup_type != M_DATALKPTYPE) {
7645 /* Inner header VNI mask */
7646 vnix = ((data2 & F_DATAVIDH2) << 23) |
7647 (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7653 tcamxy2valmask(tcamx, tcamy, addr, &mask);
7655 cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7656 cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7658 if (lookup_type && lookup_type != M_DATALKPTYPE) {
7659 sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7660 "%012jx %06x %06x - - %3c"
7661 " 'I' %4x %3c %#x%4u%4d", i, addr[0],
7662 addr[1], addr[2], addr[3], addr[4], addr[5],
7663 (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
7664 port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7665 G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7666 cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7668 sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7669 "%012jx - - ", i, addr[0], addr[1],
7670 addr[2], addr[3], addr[4], addr[5],
7674 sbuf_printf(sb, "%4u Y ", ivlan);
7676 sbuf_printf(sb, " - N ");
7678 sbuf_printf(sb, "- %3c %4x %3c %#x%4u%4d",
7679 lookup_type ? 'I' : 'O', port_num,
7680 cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7681 G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7682 cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7686 if (cls_lo & F_T6_REPLICATE) {
7687 struct fw_ldst_cmd ldst_cmd;
7689 memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7690 ldst_cmd.op_to_addrspace =
7691 htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7692 F_FW_CMD_REQUEST | F_FW_CMD_READ |
7693 V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7694 ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7695 ldst_cmd.u.mps.rplc.fid_idx =
7696 htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7697 V_FW_LDST_CMD_IDX(i));
7699 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7703 rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7704 sizeof(ldst_cmd), &ldst_cmd);
7705 end_synchronized_op(sc, 0);
7708 sbuf_printf(sb, "%72d", rc);
7711 sbuf_printf(sb, " %08x %08x %08x %08x"
7712 " %08x %08x %08x %08x",
7713 be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
7714 be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
7715 be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
7716 be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
7717 be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7718 be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7719 be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7720 be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7723 sbuf_printf(sb, "%72s", "");
7725 sbuf_printf(sb, "%4u%3u%3u%3u %#x",
7726 G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
7727 G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
7728 (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
7732 (void) sbuf_finish(sb);
7734 rc = sbuf_finish(sb);
7741 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
7743 struct adapter *sc = arg1;
7746 uint16_t mtus[NMTUS];
7748 rc = sysctl_wire_old_buffer(req, 0);
7752 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7756 t4_read_mtu_tbl(sc, mtus, NULL);
7758 sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
7759 mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
7760 mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
7761 mtus[14], mtus[15]);
7763 rc = sbuf_finish(sb);
7770 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
7772 struct adapter *sc = arg1;
7775 uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
7776 uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
7777 static const char *tx_stats[MAX_PM_NSTATS] = {
7778 "Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
7779 "Tx FIFO wait", NULL, "Tx latency"
7781 static const char *rx_stats[MAX_PM_NSTATS] = {
7782 "Read:", "Write bypass:", "Write mem:", "Flush:",
7783 "Rx FIFO wait", NULL, "Rx latency"
7786 rc = sysctl_wire_old_buffer(req, 0);
7790 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7794 t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
7795 t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
7797 sbuf_printf(sb, " Tx pcmds Tx bytes");
7798 for (i = 0; i < 4; i++) {
7799 sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7803 sbuf_printf(sb, "\n Rx pcmds Rx bytes");
7804 for (i = 0; i < 4; i++) {
7805 sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7809 if (chip_id(sc) > CHELSIO_T5) {
7811 "\n Total wait Total occupancy");
7812 sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7814 sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7818 MPASS(i < nitems(tx_stats));
7821 "\n Reads Total wait");
7822 sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7824 sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7828 rc = sbuf_finish(sb);
7835 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
7837 struct adapter *sc = arg1;
7840 struct tp_rdma_stats stats;
7842 rc = sysctl_wire_old_buffer(req, 0);
7846 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7850 mtx_lock(&sc->reg_lock);
7851 t4_tp_get_rdma_stats(sc, &stats, 0);
7852 mtx_unlock(&sc->reg_lock);
7854 sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
7855 sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
7857 rc = sbuf_finish(sb);
7864 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
7866 struct adapter *sc = arg1;
7869 struct tp_tcp_stats v4, v6;
7871 rc = sysctl_wire_old_buffer(req, 0);
7875 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7879 mtx_lock(&sc->reg_lock);
7880 t4_tp_get_tcp_stats(sc, &v4, &v6, 0);
7881 mtx_unlock(&sc->reg_lock);
7885 sbuf_printf(sb, "OutRsts: %20u %20u\n",
7886 v4.tcp_out_rsts, v6.tcp_out_rsts);
7887 sbuf_printf(sb, "InSegs: %20ju %20ju\n",
7888 v4.tcp_in_segs, v6.tcp_in_segs);
7889 sbuf_printf(sb, "OutSegs: %20ju %20ju\n",
7890 v4.tcp_out_segs, v6.tcp_out_segs);
7891 sbuf_printf(sb, "RetransSegs: %20ju %20ju",
7892 v4.tcp_retrans_segs, v6.tcp_retrans_segs);
7894 rc = sbuf_finish(sb);
7901 sysctl_tids(SYSCTL_HANDLER_ARGS)
7903 struct adapter *sc = arg1;
7906 struct tid_info *t = &sc->tids;
7908 rc = sysctl_wire_old_buffer(req, 0);
7912 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7917 sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
7922 sbuf_printf(sb, "TID range: ");
7923 if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7926 if (chip_id(sc) <= CHELSIO_T5) {
7927 b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
7928 hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
7930 b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
7931 hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
7935 sbuf_printf(sb, "0-%u, ", b - 1);
7936 sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
7938 sbuf_printf(sb, "0-%u", t->ntids - 1);
7939 sbuf_printf(sb, ", in use: %u\n",
7940 atomic_load_acq_int(&t->tids_in_use));
7944 sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
7945 t->stid_base + t->nstids - 1, t->stids_in_use);
7949 sbuf_printf(sb, "FTID range: %u-%u\n", t->ftid_base,
7950 t->ftid_base + t->nftids - 1);
7954 sbuf_printf(sb, "ETID range: %u-%u, in use: %u\n", t->etid_base,
7955 t->etid_base + t->netids - 1, t->etids_in_use);
7958 sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
7959 t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
7960 t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
7962 rc = sbuf_finish(sb);
7969 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
7971 struct adapter *sc = arg1;
7974 struct tp_err_stats stats;
7976 rc = sysctl_wire_old_buffer(req, 0);
7980 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7984 mtx_lock(&sc->reg_lock);
7985 t4_tp_get_err_stats(sc, &stats, 0);
7986 mtx_unlock(&sc->reg_lock);
7988 if (sc->chip_params->nchan > 2) {
7989 sbuf_printf(sb, " channel 0 channel 1"
7990 " channel 2 channel 3\n");
7991 sbuf_printf(sb, "macInErrs: %10u %10u %10u %10u\n",
7992 stats.mac_in_errs[0], stats.mac_in_errs[1],
7993 stats.mac_in_errs[2], stats.mac_in_errs[3]);
7994 sbuf_printf(sb, "hdrInErrs: %10u %10u %10u %10u\n",
7995 stats.hdr_in_errs[0], stats.hdr_in_errs[1],
7996 stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
7997 sbuf_printf(sb, "tcpInErrs: %10u %10u %10u %10u\n",
7998 stats.tcp_in_errs[0], stats.tcp_in_errs[1],
7999 stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
8000 sbuf_printf(sb, "tcp6InErrs: %10u %10u %10u %10u\n",
8001 stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
8002 stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
8003 sbuf_printf(sb, "tnlCongDrops: %10u %10u %10u %10u\n",
8004 stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
8005 stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
8006 sbuf_printf(sb, "tnlTxDrops: %10u %10u %10u %10u\n",
8007 stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
8008 stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
8009 sbuf_printf(sb, "ofldVlanDrops: %10u %10u %10u %10u\n",
8010 stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
8011 stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
8012 sbuf_printf(sb, "ofldChanDrops: %10u %10u %10u %10u\n\n",
8013 stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
8014 stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
8016 sbuf_printf(sb, " channel 0 channel 1\n");
8017 sbuf_printf(sb, "macInErrs: %10u %10u\n",
8018 stats.mac_in_errs[0], stats.mac_in_errs[1]);
8019 sbuf_printf(sb, "hdrInErrs: %10u %10u\n",
8020 stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
8021 sbuf_printf(sb, "tcpInErrs: %10u %10u\n",
8022 stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
8023 sbuf_printf(sb, "tcp6InErrs: %10u %10u\n",
8024 stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
8025 sbuf_printf(sb, "tnlCongDrops: %10u %10u\n",
8026 stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
8027 sbuf_printf(sb, "tnlTxDrops: %10u %10u\n",
8028 stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
8029 sbuf_printf(sb, "ofldVlanDrops: %10u %10u\n",
8030 stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
8031 sbuf_printf(sb, "ofldChanDrops: %10u %10u\n\n",
8032 stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
8035 sbuf_printf(sb, "ofldNoNeigh: %u\nofldCongDefer: %u",
8036 stats.ofld_no_neigh, stats.ofld_cong_defer);
8038 rc = sbuf_finish(sb);
8045 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
8047 struct adapter *sc = arg1;
8048 struct tp_params *tpp = &sc->params.tp;
8052 mask = tpp->la_mask >> 16;
8053 rc = sysctl_handle_int(oidp, &mask, 0, req);
8054 if (rc != 0 || req->newptr == NULL)
8058 tpp->la_mask = mask << 16;
8059 t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
8071 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
8077 uint64_t mask = (1ULL << f->width) - 1;
8078 int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
8079 ((uintmax_t)v >> f->start) & mask);
8081 if (line_size + len >= 79) {
8083 sbuf_printf(sb, "\n ");
8085 sbuf_printf(sb, "%s ", buf);
8086 line_size += len + 1;
8089 sbuf_printf(sb, "\n");
8092 static const struct field_desc tp_la0[] = {
8093 { "RcfOpCodeOut", 60, 4 },
8095 { "WcfState", 52, 4 },
8096 { "RcfOpcSrcOut", 50, 2 },
8097 { "CRxError", 49, 1 },
8098 { "ERxError", 48, 1 },
8099 { "SanityFailed", 47, 1 },
8100 { "SpuriousMsg", 46, 1 },
8101 { "FlushInputMsg", 45, 1 },
8102 { "FlushInputCpl", 44, 1 },
8103 { "RssUpBit", 43, 1 },
8104 { "RssFilterHit", 42, 1 },
8106 { "InitTcb", 31, 1 },
8107 { "LineNumber", 24, 7 },
8109 { "EdataOut", 22, 1 },
8111 { "CdataOut", 20, 1 },
8112 { "EreadPdu", 19, 1 },
8113 { "CreadPdu", 18, 1 },
8114 { "TunnelPkt", 17, 1 },
8115 { "RcfPeerFin", 16, 1 },
8116 { "RcfReasonOut", 12, 4 },
8117 { "TxCchannel", 10, 2 },
8118 { "RcfTxChannel", 8, 2 },
8119 { "RxEchannel", 6, 2 },
8120 { "RcfRxChannel", 5, 1 },
8121 { "RcfDataOutSrdy", 4, 1 },
8123 { "RxOoDvld", 2, 1 },
8124 { "RxCongestion", 1, 1 },
8125 { "TxCongestion", 0, 1 },
8129 static const struct field_desc tp_la1[] = {
8130 { "CplCmdIn", 56, 8 },
8131 { "CplCmdOut", 48, 8 },
8132 { "ESynOut", 47, 1 },
8133 { "EAckOut", 46, 1 },
8134 { "EFinOut", 45, 1 },
8135 { "ERstOut", 44, 1 },
8140 { "DataIn", 39, 1 },
8141 { "DataInVld", 38, 1 },
8143 { "RxBufEmpty", 36, 1 },
8145 { "RxFbCongestion", 34, 1 },
8146 { "TxFbCongestion", 33, 1 },
8147 { "TxPktSumSrdy", 32, 1 },
8148 { "RcfUlpType", 28, 4 },
8150 { "Ebypass", 26, 1 },
8152 { "Static0", 24, 1 },
8154 { "Cbypass", 22, 1 },
8156 { "CPktOut", 20, 1 },
8157 { "RxPagePoolFull", 18, 2 },
8158 { "RxLpbkPkt", 17, 1 },
8159 { "TxLpbkPkt", 16, 1 },
8160 { "RxVfValid", 15, 1 },
8161 { "SynLearned", 14, 1 },
8162 { "SetDelEntry", 13, 1 },
8163 { "SetInvEntry", 12, 1 },
8164 { "CpcmdDvld", 11, 1 },
8165 { "CpcmdSave", 10, 1 },
8166 { "RxPstructsFull", 8, 2 },
8167 { "EpcmdDvld", 7, 1 },
8168 { "EpcmdFlush", 6, 1 },
8169 { "EpcmdTrimPrefix", 5, 1 },
8170 { "EpcmdTrimPostfix", 4, 1 },
8171 { "ERssIp4Pkt", 3, 1 },
8172 { "ERssIp6Pkt", 2, 1 },
8173 { "ERssTcpUdpPkt", 1, 1 },
8174 { "ERssFceFipPkt", 0, 1 },
8178 static const struct field_desc tp_la2[] = {
8179 { "CplCmdIn", 56, 8 },
8180 { "MpsVfVld", 55, 1 },
8187 { "DataIn", 39, 1 },
8188 { "DataInVld", 38, 1 },
8190 { "RxBufEmpty", 36, 1 },
8192 { "RxFbCongestion", 34, 1 },
8193 { "TxFbCongestion", 33, 1 },
8194 { "TxPktSumSrdy", 32, 1 },
8195 { "RcfUlpType", 28, 4 },
8197 { "Ebypass", 26, 1 },
8199 { "Static0", 24, 1 },
8201 { "Cbypass", 22, 1 },
8203 { "CPktOut", 20, 1 },
8204 { "RxPagePoolFull", 18, 2 },
8205 { "RxLpbkPkt", 17, 1 },
8206 { "TxLpbkPkt", 16, 1 },
8207 { "RxVfValid", 15, 1 },
8208 { "SynLearned", 14, 1 },
8209 { "SetDelEntry", 13, 1 },
8210 { "SetInvEntry", 12, 1 },
8211 { "CpcmdDvld", 11, 1 },
8212 { "CpcmdSave", 10, 1 },
8213 { "RxPstructsFull", 8, 2 },
8214 { "EpcmdDvld", 7, 1 },
8215 { "EpcmdFlush", 6, 1 },
8216 { "EpcmdTrimPrefix", 5, 1 },
8217 { "EpcmdTrimPostfix", 4, 1 },
8218 { "ERssIp4Pkt", 3, 1 },
8219 { "ERssIp6Pkt", 2, 1 },
8220 { "ERssTcpUdpPkt", 1, 1 },
8221 { "ERssFceFipPkt", 0, 1 },
8226 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
8229 field_desc_show(sb, *p, tp_la0);
8233 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
8237 sbuf_printf(sb, "\n");
8238 field_desc_show(sb, p[0], tp_la0);
8239 if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8240 field_desc_show(sb, p[1], tp_la0);
8244 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
8248 sbuf_printf(sb, "\n");
8249 field_desc_show(sb, p[0], tp_la0);
8250 if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8251 field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
8255 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
8257 struct adapter *sc = arg1;
8262 void (*show_func)(struct sbuf *, uint64_t *, int);
8264 rc = sysctl_wire_old_buffer(req, 0);
8268 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8272 buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
8274 t4_tp_read_la(sc, buf, NULL);
8277 switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
8280 show_func = tp_la_show2;
8284 show_func = tp_la_show3;
8288 show_func = tp_la_show;
8291 for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
8292 (*show_func)(sb, p, i);
8294 rc = sbuf_finish(sb);
8301 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
8303 struct adapter *sc = arg1;
8306 u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
8308 rc = sysctl_wire_old_buffer(req, 0);
8312 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8316 t4_get_chan_txrate(sc, nrate, orate);
8318 if (sc->chip_params->nchan > 2) {
8319 sbuf_printf(sb, " channel 0 channel 1"
8320 " channel 2 channel 3\n");
8321 sbuf_printf(sb, "NIC B/s: %10ju %10ju %10ju %10ju\n",
8322 nrate[0], nrate[1], nrate[2], nrate[3]);
8323 sbuf_printf(sb, "Offload B/s: %10ju %10ju %10ju %10ju",
8324 orate[0], orate[1], orate[2], orate[3]);
8326 sbuf_printf(sb, " channel 0 channel 1\n");
8327 sbuf_printf(sb, "NIC B/s: %10ju %10ju\n",
8328 nrate[0], nrate[1]);
8329 sbuf_printf(sb, "Offload B/s: %10ju %10ju",
8330 orate[0], orate[1]);
8333 rc = sbuf_finish(sb);
8340 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
8342 struct adapter *sc = arg1;
8347 rc = sysctl_wire_old_buffer(req, 0);
8351 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8355 buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
8358 t4_ulprx_read_la(sc, buf);
8361 sbuf_printf(sb, " Pcmd Type Message"
8363 for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
8364 sbuf_printf(sb, "\n%08x%08x %4x %08x %08x%08x%08x%08x",
8365 p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
8368 rc = sbuf_finish(sb);
8375 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
8377 struct adapter *sc = arg1;
8381 MPASS(chip_id(sc) >= CHELSIO_T5);
8383 rc = sysctl_wire_old_buffer(req, 0);
8387 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8391 v = t4_read_reg(sc, A_SGE_STAT_CFG);
8392 if (G_STATSOURCE_T5(v) == 7) {
8395 mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
8397 sbuf_printf(sb, "total %d, incomplete %d",
8398 t4_read_reg(sc, A_SGE_STAT_TOTAL),
8399 t4_read_reg(sc, A_SGE_STAT_MATCH));
8400 } else if (mode == 1) {
8401 sbuf_printf(sb, "total %d, data overflow %d",
8402 t4_read_reg(sc, A_SGE_STAT_TOTAL),
8403 t4_read_reg(sc, A_SGE_STAT_MATCH));
8405 sbuf_printf(sb, "unknown mode %d", mode);
8408 rc = sbuf_finish(sb);
8415 sysctl_tc_params(SYSCTL_HANDLER_ARGS)
8417 struct adapter *sc = arg1;
8418 struct tx_cl_rl_params tc;
8420 int i, rc, port_id, mbps, gbps;
8422 rc = sysctl_wire_old_buffer(req, 0);
8426 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8430 port_id = arg2 >> 16;
8431 MPASS(port_id < sc->params.nports);
8432 MPASS(sc->port[port_id] != NULL);
8434 MPASS(i < sc->chip_params->nsched_cls);
8436 mtx_lock(&sc->tc_lock);
8437 tc = sc->port[port_id]->sched_params->cl_rl[i];
8438 mtx_unlock(&sc->tc_lock);
8440 if (tc.flags & TX_CLRL_ERROR) {
8441 sbuf_printf(sb, "error");
8445 if (tc.ratemode == SCHED_CLASS_RATEMODE_REL) {
8446 /* XXX: top speed or actual link speed? */
8447 gbps = port_top_speed(sc->port[port_id]);
8448 sbuf_printf(sb, " %u%% of %uGbps", tc.maxrate, gbps);
8449 } else if (tc.ratemode == SCHED_CLASS_RATEMODE_ABS) {
8450 switch (tc.rateunit) {
8451 case SCHED_CLASS_RATEUNIT_BITS:
8452 mbps = tc.maxrate / 1000;
8453 gbps = tc.maxrate / 1000000;
8454 if (tc.maxrate == gbps * 1000000)
8455 sbuf_printf(sb, " %uGbps", gbps);
8456 else if (tc.maxrate == mbps * 1000)
8457 sbuf_printf(sb, " %uMbps", mbps);
8459 sbuf_printf(sb, " %uKbps", tc.maxrate);
8461 case SCHED_CLASS_RATEUNIT_PKTS:
8462 sbuf_printf(sb, " %upps", tc.maxrate);
8471 case SCHED_CLASS_MODE_CLASS:
8472 sbuf_printf(sb, " aggregate");
8474 case SCHED_CLASS_MODE_FLOW:
8475 sbuf_printf(sb, " per-flow");
8484 rc = sbuf_finish(sb);
8493 sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS)
8495 struct adapter *sc = arg1;
8496 int *old_ports, *new_ports;
8497 int i, new_count, rc;
8499 if (req->newptr == NULL && req->oldptr == NULL)
8500 return (SYSCTL_OUT(req, NULL, imax(sc->tt.num_tls_rx_ports, 1) *
8501 sizeof(sc->tt.tls_rx_ports[0])));
8503 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tlsrx");
8507 if (sc->tt.num_tls_rx_ports == 0) {
8509 rc = SYSCTL_OUT(req, &i, sizeof(i));
8511 rc = SYSCTL_OUT(req, sc->tt.tls_rx_ports,
8512 sc->tt.num_tls_rx_ports * sizeof(sc->tt.tls_rx_ports[0]));
8513 if (rc == 0 && req->newptr != NULL) {
8514 new_count = req->newlen / sizeof(new_ports[0]);
8515 new_ports = malloc(new_count * sizeof(new_ports[0]), M_CXGBE,
8517 rc = SYSCTL_IN(req, new_ports, new_count *
8518 sizeof(new_ports[0]));
8522 /* Allow setting to a single '-1' to clear the list. */
8523 if (new_count == 1 && new_ports[0] == -1) {
8525 old_ports = sc->tt.tls_rx_ports;
8526 sc->tt.tls_rx_ports = NULL;
8527 sc->tt.num_tls_rx_ports = 0;
8529 free(old_ports, M_CXGBE);
8531 for (i = 0; i < new_count; i++) {
8532 if (new_ports[i] < 1 ||
8533 new_ports[i] > IPPORT_MAX) {
8540 old_ports = sc->tt.tls_rx_ports;
8541 sc->tt.tls_rx_ports = new_ports;
8542 sc->tt.num_tls_rx_ports = new_count;
8544 free(old_ports, M_CXGBE);
8548 free(new_ports, M_CXGBE);
8550 end_synchronized_op(sc, 0);
8555 unit_conv(char *buf, size_t len, u_int val, u_int factor)
8557 u_int rem = val % factor;
8560 snprintf(buf, len, "%u", val / factor);
8562 while (rem % 10 == 0)
8564 snprintf(buf, len, "%u.%u", val / factor, rem);
8569 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
8571 struct adapter *sc = arg1;
8574 u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8576 res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8580 re = G_TIMERRESOLUTION(res);
8583 /* TCP timestamp tick */
8584 re = G_TIMESTAMPRESOLUTION(res);
8588 re = G_DELAYEDACKRESOLUTION(res);
8594 unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
8596 return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
8600 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
8602 struct adapter *sc = arg1;
8603 u_int res, dack_re, v;
8604 u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8606 res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8607 dack_re = G_DELAYEDACKRESOLUTION(res);
8608 v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
8610 return (sysctl_handle_int(oidp, &v, 0, req));
8614 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
8616 struct adapter *sc = arg1;
8619 u_long tp_tick_us, v;
8620 u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8622 MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
8623 reg == A_TP_PERS_MIN || reg == A_TP_PERS_MAX ||
8624 reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
8625 reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
8627 tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
8628 tp_tick_us = (cclk_ps << tre) / 1000000;
8630 if (reg == A_TP_INIT_SRTT)
8631 v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
8633 v = tp_tick_us * t4_read_reg(sc, reg);
8635 return (sysctl_handle_long(oidp, &v, 0, req));
8639 * All fields in TP_SHIFT_CNT are 4b and the starting location of the field is
8640 * passed to this function.
8643 sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS)
8645 struct adapter *sc = arg1;
8649 MPASS(idx >= 0 && idx <= 24);
8651 v = (t4_read_reg(sc, A_TP_SHIFT_CNT) >> idx) & 0xf;
8653 return (sysctl_handle_int(oidp, &v, 0, req));
8657 sysctl_tp_backoff(SYSCTL_HANDLER_ARGS)
8659 struct adapter *sc = arg1;
8663 MPASS(idx >= 0 && idx < 16);
8665 r = A_TP_TCP_BACKOFF_REG0 + (idx & ~3);
8666 shift = (idx & 3) << 3;
8667 v = (t4_read_reg(sc, r) >> shift) & M_TIMERBACKOFFINDEX0;
8669 return (sysctl_handle_int(oidp, &v, 0, req));
8673 sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS)
8675 struct vi_info *vi = arg1;
8676 struct adapter *sc = vi->pi->adapter;
8678 struct sge_ofld_rxq *ofld_rxq;
8681 idx = vi->ofld_tmr_idx;
8683 rc = sysctl_handle_int(oidp, &idx, 0, req);
8684 if (rc != 0 || req->newptr == NULL)
8687 if (idx < 0 || idx >= SGE_NTIMERS)
8690 rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8695 v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->ofld_pktc_idx != -1);
8696 for_each_ofld_rxq(vi, i, ofld_rxq) {
8697 #ifdef atomic_store_rel_8
8698 atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
8700 ofld_rxq->iq.intr_params = v;
8703 vi->ofld_tmr_idx = idx;
8705 end_synchronized_op(sc, LOCK_HELD);
8710 sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS)
8712 struct vi_info *vi = arg1;
8713 struct adapter *sc = vi->pi->adapter;
8716 idx = vi->ofld_pktc_idx;
8718 rc = sysctl_handle_int(oidp, &idx, 0, req);
8719 if (rc != 0 || req->newptr == NULL)
8722 if (idx < -1 || idx >= SGE_NCOUNTERS)
8725 rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8730 if (vi->flags & VI_INIT_DONE)
8731 rc = EBUSY; /* cannot be changed once the queues are created */
8733 vi->ofld_pktc_idx = idx;
8735 end_synchronized_op(sc, LOCK_HELD);
8741 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
8745 if (cntxt->cid > M_CTXTQID)
8748 if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
8749 cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
8752 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
8756 if (sc->flags & FW_OK) {
8757 rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
8764 * Read via firmware failed or wasn't even attempted. Read directly via
8767 rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
8769 end_synchronized_op(sc, 0);
8774 load_fw(struct adapter *sc, struct t4_data *fw)
8779 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
8784 * The firmware, with the sole exception of the memory parity error
8785 * handler, runs from memory and not flash. It is almost always safe to
8786 * install a new firmware on a running system. Just set bit 1 in
8787 * hw.cxgbe.dflags or dev.<nexus>.<n>.dflags first.
8789 if (sc->flags & FULL_INIT_DONE &&
8790 (sc->debug_flags & DF_LOAD_FW_ANYTIME) == 0) {
8795 fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
8796 if (fw_data == NULL) {
8801 rc = copyin(fw->data, fw_data, fw->len);
8803 rc = -t4_load_fw(sc, fw_data, fw->len);
8805 free(fw_data, M_CXGBE);
8807 end_synchronized_op(sc, 0);
8812 load_cfg(struct adapter *sc, struct t4_data *cfg)
8815 uint8_t *cfg_data = NULL;
8817 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
8821 if (cfg->len == 0) {
8823 rc = -t4_load_cfg(sc, NULL, 0);
8827 cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
8828 if (cfg_data == NULL) {
8833 rc = copyin(cfg->data, cfg_data, cfg->len);
8835 rc = -t4_load_cfg(sc, cfg_data, cfg->len);
8837 free(cfg_data, M_CXGBE);
8839 end_synchronized_op(sc, 0);
8844 load_boot(struct adapter *sc, struct t4_bootrom *br)
8847 uint8_t *br_data = NULL;
8850 if (br->len > 1024 * 1024)
8853 if (br->pf_offset == 0) {
8855 if (br->pfidx_addr > 7)
8857 offset = G_OFFSET(t4_read_reg(sc, PF_REG(br->pfidx_addr,
8858 A_PCIE_PF_EXPROM_OFST)));
8859 } else if (br->pf_offset == 1) {
8861 offset = G_OFFSET(br->pfidx_addr);
8866 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldbr");
8872 rc = -t4_load_boot(sc, NULL, offset, 0);
8876 br_data = malloc(br->len, M_CXGBE, M_WAITOK);
8877 if (br_data == NULL) {
8882 rc = copyin(br->data, br_data, br->len);
8884 rc = -t4_load_boot(sc, br_data, offset, br->len);
8886 free(br_data, M_CXGBE);
8888 end_synchronized_op(sc, 0);
8893 load_bootcfg(struct adapter *sc, struct t4_data *bc)
8896 uint8_t *bc_data = NULL;
8898 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
8904 rc = -t4_load_bootcfg(sc, NULL, 0);
8908 bc_data = malloc(bc->len, M_CXGBE, M_WAITOK);
8909 if (bc_data == NULL) {
8914 rc = copyin(bc->data, bc_data, bc->len);
8916 rc = -t4_load_bootcfg(sc, bc_data, bc->len);
8918 free(bc_data, M_CXGBE);
8920 end_synchronized_op(sc, 0);
8925 cudbg_dump(struct adapter *sc, struct t4_cudbg_dump *dump)
8928 struct cudbg_init *cudbg;
8931 /* buf is large, don't block if no memory is available */
8932 buf = malloc(dump->len, M_CXGBE, M_NOWAIT | M_ZERO);
8936 handle = cudbg_alloc_handle();
8937 if (handle == NULL) {
8942 cudbg = cudbg_get_init(handle);
8944 cudbg->print = (cudbg_print_cb)printf;
8947 device_printf(sc->dev, "%s: wr_flash %u, len %u, data %p.\n",
8948 __func__, dump->wr_flash, dump->len, dump->data);
8952 cudbg->use_flash = 1;
8953 MPASS(sizeof(cudbg->dbg_bitmap) == sizeof(dump->bitmap));
8954 memcpy(cudbg->dbg_bitmap, dump->bitmap, sizeof(cudbg->dbg_bitmap));
8956 rc = cudbg_collect(handle, buf, &dump->len);
8960 rc = copyout(buf, dump->data, dump->len);
8962 cudbg_free_handle(handle);
8968 free_offload_policy(struct t4_offload_policy *op)
8970 struct offload_rule *r;
8977 for (i = 0; i < op->nrules; i++, r++) {
8978 free(r->bpf_prog.bf_insns, M_CXGBE);
8980 free(op->rule, M_CXGBE);
8985 set_offload_policy(struct adapter *sc, struct t4_offload_policy *uop)
8988 struct t4_offload_policy *op, *old;
8989 struct bpf_program *bf;
8990 const struct offload_settings *s;
8991 struct offload_rule *r;
8994 if (!is_offload(sc))
8997 if (uop->nrules == 0) {
8998 /* Delete installed policies. */
9001 } if (uop->nrules > 256) { /* arbitrary */
9005 /* Copy userspace offload policy to kernel */
9006 op = malloc(sizeof(*op), M_CXGBE, M_ZERO | M_WAITOK);
9007 op->nrules = uop->nrules;
9008 len = op->nrules * sizeof(struct offload_rule);
9009 op->rule = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9010 rc = copyin(uop->rule, op->rule, len);
9012 free(op->rule, M_CXGBE);
9018 for (i = 0; i < op->nrules; i++, r++) {
9020 /* Validate open_type */
9021 if (r->open_type != OPEN_TYPE_LISTEN &&
9022 r->open_type != OPEN_TYPE_ACTIVE &&
9023 r->open_type != OPEN_TYPE_PASSIVE &&
9024 r->open_type != OPEN_TYPE_DONTCARE) {
9027 * Rules 0 to i have malloc'd filters that need to be
9028 * freed. Rules i+1 to nrules have userspace pointers
9029 * and should be left alone.
9032 free_offload_policy(op);
9036 /* Validate settings */
9038 if ((s->offload != 0 && s->offload != 1) ||
9039 s->cong_algo < -1 || s->cong_algo > CONG_ALG_HIGHSPEED ||
9040 s->sched_class < -1 ||
9041 s->sched_class >= sc->chip_params->nsched_cls) {
9047 u = bf->bf_insns; /* userspace ptr */
9048 bf->bf_insns = NULL;
9049 if (bf->bf_len == 0) {
9050 /* legal, matches everything */
9053 len = bf->bf_len * sizeof(*bf->bf_insns);
9054 bf->bf_insns = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9055 rc = copyin(u, bf->bf_insns, len);
9059 if (!bpf_validate(bf->bf_insns, bf->bf_len)) {
9065 rw_wlock(&sc->policy_lock);
9068 rw_wunlock(&sc->policy_lock);
9069 free_offload_policy(old);
9074 #define MAX_READ_BUF_SIZE (128 * 1024)
9076 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
9078 uint32_t addr, remaining, n;
9083 rc = validate_mem_range(sc, mr->addr, mr->len);
9087 buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
9089 remaining = mr->len;
9090 dst = (void *)mr->data;
9093 n = min(remaining, MAX_READ_BUF_SIZE);
9094 read_via_memwin(sc, 2, addr, buf, n);
9096 rc = copyout(buf, dst, n);
9108 #undef MAX_READ_BUF_SIZE
9111 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
9115 if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
9118 if (i2cd->len > sizeof(i2cd->data))
9121 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
9124 rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
9125 i2cd->offset, i2cd->len, &i2cd->data[0]);
9126 end_synchronized_op(sc, 0);
9132 t4_os_find_pci_capability(struct adapter *sc, int cap)
9136 return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9140 t4_os_pci_save_state(struct adapter *sc)
9143 struct pci_devinfo *dinfo;
9146 dinfo = device_get_ivars(dev);
9148 pci_cfg_save(dev, dinfo, 0);
9153 t4_os_pci_restore_state(struct adapter *sc)
9156 struct pci_devinfo *dinfo;
9159 dinfo = device_get_ivars(dev);
9161 pci_cfg_restore(dev, dinfo);
9166 t4_os_portmod_changed(struct port_info *pi)
9168 struct adapter *sc = pi->adapter;
9171 static const char *mod_str[] = {
9172 NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9176 build_medialist(pi, &pi->media);
9179 if (begin_synchronized_op(sc, vi, HOLD_LOCK, "t4mod") == 0) {
9181 end_synchronized_op(sc, LOCK_HELD);
9185 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
9186 if_printf(ifp, "transceiver unplugged.\n");
9187 else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
9188 if_printf(ifp, "unknown transceiver inserted.\n");
9189 else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
9190 if_printf(ifp, "unsupported transceiver inserted.\n");
9191 else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
9192 if_printf(ifp, "%dGbps %s transceiver inserted.\n",
9193 port_top_speed(pi), mod_str[pi->mod_type]);
9195 if_printf(ifp, "transceiver (type %d) inserted.\n",
9201 t4_os_link_changed(struct port_info *pi)
9205 struct link_config *lc;
9208 for_each_vi(pi, v, vi) {
9215 ifp->if_baudrate = IF_Mbps(lc->speed);
9216 if_link_state_change(ifp, LINK_STATE_UP);
9218 if_link_state_change(ifp, LINK_STATE_DOWN);
9224 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
9228 sx_slock(&t4_list_lock);
9229 SLIST_FOREACH(sc, &t4_list, link) {
9231 * func should not make any assumptions about what state sc is
9232 * in - the only guarantee is that sc->sc_lock is a valid lock.
9236 sx_sunlock(&t4_list_lock);
9240 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9244 struct adapter *sc = dev->si_drv1;
9246 rc = priv_check(td, PRIV_DRIVER);
9251 case CHELSIO_T4_GETREG: {
9252 struct t4_reg *edata = (struct t4_reg *)data;
9254 if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9257 if (edata->size == 4)
9258 edata->val = t4_read_reg(sc, edata->addr);
9259 else if (edata->size == 8)
9260 edata->val = t4_read_reg64(sc, edata->addr);
9266 case CHELSIO_T4_SETREG: {
9267 struct t4_reg *edata = (struct t4_reg *)data;
9269 if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9272 if (edata->size == 4) {
9273 if (edata->val & 0xffffffff00000000)
9275 t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9276 } else if (edata->size == 8)
9277 t4_write_reg64(sc, edata->addr, edata->val);
9282 case CHELSIO_T4_REGDUMP: {
9283 struct t4_regdump *regs = (struct t4_regdump *)data;
9284 int reglen = t4_get_regs_len(sc);
9287 if (regs->len < reglen) {
9288 regs->len = reglen; /* hint to the caller */
9293 buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
9294 get_regs(sc, regs, buf);
9295 rc = copyout(buf, regs->data, reglen);
9299 case CHELSIO_T4_GET_FILTER_MODE:
9300 rc = get_filter_mode(sc, (uint32_t *)data);
9302 case CHELSIO_T4_SET_FILTER_MODE:
9303 rc = set_filter_mode(sc, *(uint32_t *)data);
9305 case CHELSIO_T4_GET_FILTER:
9306 rc = get_filter(sc, (struct t4_filter *)data);
9308 case CHELSIO_T4_SET_FILTER:
9309 rc = set_filter(sc, (struct t4_filter *)data);
9311 case CHELSIO_T4_DEL_FILTER:
9312 rc = del_filter(sc, (struct t4_filter *)data);
9314 case CHELSIO_T4_GET_SGE_CONTEXT:
9315 rc = get_sge_context(sc, (struct t4_sge_context *)data);
9317 case CHELSIO_T4_LOAD_FW:
9318 rc = load_fw(sc, (struct t4_data *)data);
9320 case CHELSIO_T4_GET_MEM:
9321 rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
9323 case CHELSIO_T4_GET_I2C:
9324 rc = read_i2c(sc, (struct t4_i2c_data *)data);
9326 case CHELSIO_T4_CLEAR_STATS: {
9328 u_int port_id = *(uint32_t *)data;
9329 struct port_info *pi;
9332 if (port_id >= sc->params.nports)
9334 pi = sc->port[port_id];
9339 t4_clr_port_stats(sc, pi->tx_chan);
9340 pi->tx_parse_error = 0;
9341 pi->tnl_cong_drops = 0;
9342 mtx_lock(&sc->reg_lock);
9343 for_each_vi(pi, v, vi) {
9344 if (vi->flags & VI_INIT_DONE)
9345 t4_clr_vi_stats(sc, vi->viid);
9347 bg_map = pi->mps_bg_map;
9350 i = ffs(bg_map) - 1;
9351 t4_write_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
9352 1, A_TP_MIB_TNL_CNG_DROP_0 + i);
9353 bg_map &= ~(1 << i);
9355 mtx_unlock(&sc->reg_lock);
9358 * Since this command accepts a port, clear stats for
9359 * all VIs on this port.
9361 for_each_vi(pi, v, vi) {
9362 if (vi->flags & VI_INIT_DONE) {
9363 struct sge_rxq *rxq;
9364 struct sge_txq *txq;
9365 struct sge_wrq *wrq;
9367 for_each_rxq(vi, i, rxq) {
9368 #if defined(INET) || defined(INET6)
9369 rxq->lro.lro_queued = 0;
9370 rxq->lro.lro_flushed = 0;
9373 rxq->vlan_extraction = 0;
9376 for_each_txq(vi, i, txq) {
9379 txq->vlan_insertion = 0;
9383 txq->txpkts0_wrs = 0;
9384 txq->txpkts1_wrs = 0;
9385 txq->txpkts0_pkts = 0;
9386 txq->txpkts1_pkts = 0;
9387 mp_ring_reset_stats(txq->r);
9391 /* nothing to clear for each ofld_rxq */
9393 for_each_ofld_txq(vi, i, wrq) {
9394 wrq->tx_wrs_direct = 0;
9395 wrq->tx_wrs_copied = 0;
9399 if (IS_MAIN_VI(vi)) {
9400 wrq = &sc->sge.ctrlq[pi->port_id];
9401 wrq->tx_wrs_direct = 0;
9402 wrq->tx_wrs_copied = 0;
9408 case CHELSIO_T4_SCHED_CLASS:
9409 rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
9411 case CHELSIO_T4_SCHED_QUEUE:
9412 rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
9414 case CHELSIO_T4_GET_TRACER:
9415 rc = t4_get_tracer(sc, (struct t4_tracer *)data);
9417 case CHELSIO_T4_SET_TRACER:
9418 rc = t4_set_tracer(sc, (struct t4_tracer *)data);
9420 case CHELSIO_T4_LOAD_CFG:
9421 rc = load_cfg(sc, (struct t4_data *)data);
9423 case CHELSIO_T4_LOAD_BOOT:
9424 rc = load_boot(sc, (struct t4_bootrom *)data);
9426 case CHELSIO_T4_LOAD_BOOTCFG:
9427 rc = load_bootcfg(sc, (struct t4_data *)data);
9429 case CHELSIO_T4_CUDBG_DUMP:
9430 rc = cudbg_dump(sc, (struct t4_cudbg_dump *)data);
9432 case CHELSIO_T4_SET_OFLD_POLICY:
9433 rc = set_offload_policy(sc, (struct t4_offload_policy *)data);
9443 t4_db_full(struct adapter *sc)
9446 CXGBE_UNIMPLEMENTED(__func__);
9450 t4_db_dropped(struct adapter *sc)
9453 CXGBE_UNIMPLEMENTED(__func__);
9458 toe_capability(struct vi_info *vi, int enable)
9461 struct port_info *pi = vi->pi;
9462 struct adapter *sc = pi->adapter;
9464 ASSERT_SYNCHRONIZED_OP(sc);
9466 if (!is_offload(sc))
9470 if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
9471 /* TOE is already enabled. */
9476 * We need the port's queues around so that we're able to send
9477 * and receive CPLs to/from the TOE even if the ifnet for this
9478 * port has never been UP'd administratively.
9480 if (!(vi->flags & VI_INIT_DONE)) {
9481 rc = vi_full_init(vi);
9485 if (!(pi->vi[0].flags & VI_INIT_DONE)) {
9486 rc = vi_full_init(&pi->vi[0]);
9491 if (isset(&sc->offload_map, pi->port_id)) {
9492 /* TOE is enabled on another VI of this port. */
9497 if (!uld_active(sc, ULD_TOM)) {
9498 rc = t4_activate_uld(sc, ULD_TOM);
9501 "You must kldload t4_tom.ko before trying "
9502 "to enable TOE on a cxgbe interface.\n");
9506 KASSERT(sc->tom_softc != NULL,
9507 ("%s: TOM activated but softc NULL", __func__));
9508 KASSERT(uld_active(sc, ULD_TOM),
9509 ("%s: TOM activated but flag not set", __func__));
9512 /* Activate iWARP and iSCSI too, if the modules are loaded. */
9513 if (!uld_active(sc, ULD_IWARP))
9514 (void) t4_activate_uld(sc, ULD_IWARP);
9515 if (!uld_active(sc, ULD_ISCSI))
9516 (void) t4_activate_uld(sc, ULD_ISCSI);
9519 setbit(&sc->offload_map, pi->port_id);
9523 if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
9526 KASSERT(uld_active(sc, ULD_TOM),
9527 ("%s: TOM never initialized?", __func__));
9528 clrbit(&sc->offload_map, pi->port_id);
9535 * Add an upper layer driver to the global list.
9538 t4_register_uld(struct uld_info *ui)
9543 sx_xlock(&t4_uld_list_lock);
9544 SLIST_FOREACH(u, &t4_uld_list, link) {
9545 if (u->uld_id == ui->uld_id) {
9551 SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
9554 sx_xunlock(&t4_uld_list_lock);
9559 t4_unregister_uld(struct uld_info *ui)
9564 sx_xlock(&t4_uld_list_lock);
9566 SLIST_FOREACH(u, &t4_uld_list, link) {
9568 if (ui->refcount > 0) {
9573 SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
9579 sx_xunlock(&t4_uld_list_lock);
9584 t4_activate_uld(struct adapter *sc, int id)
9587 struct uld_info *ui;
9589 ASSERT_SYNCHRONIZED_OP(sc);
9591 if (id < 0 || id > ULD_MAX)
9593 rc = EAGAIN; /* kldoad the module with this ULD and try again. */
9595 sx_slock(&t4_uld_list_lock);
9597 SLIST_FOREACH(ui, &t4_uld_list, link) {
9598 if (ui->uld_id == id) {
9599 if (!(sc->flags & FULL_INIT_DONE)) {
9600 rc = adapter_full_init(sc);
9605 rc = ui->activate(sc);
9607 setbit(&sc->active_ulds, id);
9614 sx_sunlock(&t4_uld_list_lock);
9620 t4_deactivate_uld(struct adapter *sc, int id)
9623 struct uld_info *ui;
9625 ASSERT_SYNCHRONIZED_OP(sc);
9627 if (id < 0 || id > ULD_MAX)
9631 sx_slock(&t4_uld_list_lock);
9633 SLIST_FOREACH(ui, &t4_uld_list, link) {
9634 if (ui->uld_id == id) {
9635 rc = ui->deactivate(sc);
9637 clrbit(&sc->active_ulds, id);
9644 sx_sunlock(&t4_uld_list_lock);
9650 uld_active(struct adapter *sc, int uld_id)
9653 MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
9655 return (isset(&sc->active_ulds, uld_id));
9660 * t = ptr to tunable.
9661 * nc = number of CPUs.
9662 * c = compiled in default for that tunable.
9665 calculate_nqueues(int *t, int nc, const int c)
9671 nq = *t < 0 ? -*t : c;
9676 * Come up with reasonable defaults for some of the tunables, provided they're
9677 * not set by the user (in which case we'll use the values as is).
9680 tweak_tunables(void)
9682 int nc = mp_ncpus; /* our snapshot of the number of CPUs */
9686 t4_ntxq = rss_getnumbuckets();
9688 calculate_nqueues(&t4_ntxq, nc, NTXQ);
9692 calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
9696 t4_nrxq = rss_getnumbuckets();
9698 calculate_nqueues(&t4_nrxq, nc, NRXQ);
9702 calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
9704 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
9705 calculate_nqueues(&t4_nofldtxq, nc, NOFLDTXQ);
9706 calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
9709 calculate_nqueues(&t4_nofldrxq, nc, NOFLDRXQ);
9710 calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
9712 if (t4_toecaps_allowed == -1)
9713 t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
9715 if (t4_rdmacaps_allowed == -1) {
9716 t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
9717 FW_CAPS_CONFIG_RDMA_RDMAC;
9720 if (t4_iscsicaps_allowed == -1) {
9721 t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
9722 FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
9723 FW_CAPS_CONFIG_ISCSI_T10DIF;
9726 if (t4_tmr_idx_ofld < 0 || t4_tmr_idx_ofld >= SGE_NTIMERS)
9727 t4_tmr_idx_ofld = TMR_IDX_OFLD;
9729 if (t4_pktc_idx_ofld < -1 || t4_pktc_idx_ofld >= SGE_NCOUNTERS)
9730 t4_pktc_idx_ofld = PKTC_IDX_OFLD;
9732 if (t4_toecaps_allowed == -1)
9733 t4_toecaps_allowed = 0;
9735 if (t4_rdmacaps_allowed == -1)
9736 t4_rdmacaps_allowed = 0;
9738 if (t4_iscsicaps_allowed == -1)
9739 t4_iscsicaps_allowed = 0;
9743 calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
9744 calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
9747 if (t4_tmr_idx < 0 || t4_tmr_idx >= SGE_NTIMERS)
9748 t4_tmr_idx = TMR_IDX;
9750 if (t4_pktc_idx < -1 || t4_pktc_idx >= SGE_NCOUNTERS)
9751 t4_pktc_idx = PKTC_IDX;
9753 if (t4_qsize_txq < 128)
9756 if (t4_qsize_rxq < 128)
9758 while (t4_qsize_rxq & 7)
9761 t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
9764 * Number of VIs to create per-port. The first VI is the "main" regular
9765 * VI for the port. The rest are additional virtual interfaces on the
9766 * same physical port. Note that the main VI does not have native
9767 * netmap support but the extra VIs do.
9769 * Limit the number of VIs per port to the number of available
9770 * MAC addresses per port.
9774 if (t4_num_vis > nitems(vi_mac_funcs)) {
9775 t4_num_vis = nitems(vi_mac_funcs);
9776 printf("cxgbe: number of VIs limited to %d\n", t4_num_vis);
9779 if (pcie_relaxed_ordering < 0 || pcie_relaxed_ordering > 2) {
9780 pcie_relaxed_ordering = 1;
9781 #if defined(__i386__) || defined(__amd64__)
9782 if (cpu_vendor_id == CPU_VENDOR_INTEL)
9783 pcie_relaxed_ordering = 0;
9790 t4_dump_tcb(struct adapter *sc, int tid)
9792 uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
9794 reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
9795 save = t4_read_reg(sc, reg);
9796 base = sc->memwin[2].mw_base;
9798 /* Dump TCB for the tid */
9799 tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
9800 tcb_addr += tid * TCB_SIZE;
9804 win_pos = tcb_addr & ~0xf; /* start must be 16B aligned */
9806 pf = V_PFNUM(sc->pf);
9807 win_pos = tcb_addr & ~0x7f; /* start must be 128B aligned */
9809 t4_write_reg(sc, reg, win_pos | pf);
9810 t4_read_reg(sc, reg);
9812 off = tcb_addr - win_pos;
9813 for (i = 0; i < 4; i++) {
9815 for (j = 0; j < 8; j++, off += 4)
9816 buf[j] = htonl(t4_read_reg(sc, base + off));
9818 db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
9819 buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
9823 t4_write_reg(sc, reg, save);
9824 t4_read_reg(sc, reg);
9828 t4_dump_devlog(struct adapter *sc)
9830 struct devlog_params *dparams = &sc->params.devlog;
9831 struct fw_devlog_e e;
9832 int i, first, j, m, nentries, rc;
9833 uint64_t ftstamp = UINT64_MAX;
9835 if (dparams->start == 0) {
9836 db_printf("devlog params not valid\n");
9840 nentries = dparams->size / sizeof(struct fw_devlog_e);
9841 m = fwmtype_to_hwmtype(dparams->memtype);
9843 /* Find the first entry. */
9845 for (i = 0; i < nentries && !db_pager_quit; i++) {
9846 rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9847 sizeof(e), (void *)&e);
9851 if (e.timestamp == 0)
9854 e.timestamp = be64toh(e.timestamp);
9855 if (e.timestamp < ftstamp) {
9856 ftstamp = e.timestamp;
9866 rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9867 sizeof(e), (void *)&e);
9871 if (e.timestamp == 0)
9874 e.timestamp = be64toh(e.timestamp);
9875 e.seqno = be32toh(e.seqno);
9876 for (j = 0; j < 8; j++)
9877 e.params[j] = be32toh(e.params[j]);
9879 db_printf("%10d %15ju %8s %8s ",
9880 e.seqno, e.timestamp,
9881 (e.level < nitems(devlog_level_strings) ?
9882 devlog_level_strings[e.level] : "UNKNOWN"),
9883 (e.facility < nitems(devlog_facility_strings) ?
9884 devlog_facility_strings[e.facility] : "UNKNOWN"));
9885 db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
9886 e.params[3], e.params[4], e.params[5], e.params[6],
9889 if (++i == nentries)
9891 } while (i != first && !db_pager_quit);
9894 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
9895 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
9897 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
9904 t = db_read_token();
9906 dev = device_lookup_by_name(db_tok_string);
9911 db_printf("usage: show t4 devlog <nexus>\n");
9916 db_printf("device not found\n");
9920 t4_dump_devlog(device_get_softc(dev));
9923 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
9932 t = db_read_token();
9934 dev = device_lookup_by_name(db_tok_string);
9935 t = db_read_token();
9937 tid = db_tok_number;
9944 db_printf("usage: show t4 tcb <nexus> <tid>\n");
9949 db_printf("device not found\n");
9953 db_printf("invalid tid\n");
9957 t4_dump_tcb(device_get_softc(dev), tid);
9962 * Borrowed from cesa_prep_aes_key().
9964 * NB: The crypto engine wants the words in the decryption key in reverse
9968 t4_aes_getdeckey(void *dec_key, const void *enc_key, unsigned int kbits)
9970 uint32_t ek[4 * (RIJNDAEL_MAXNR + 1)];
9974 rijndaelKeySetupEnc(ek, enc_key, kbits);
9976 dkey += (kbits / 8) / 4;
9980 for (i = 0; i < 4; i++)
9981 *--dkey = htobe32(ek[4 * 10 + i]);
9984 for (i = 0; i < 2; i++)
9985 *--dkey = htobe32(ek[4 * 11 + 2 + i]);
9986 for (i = 0; i < 4; i++)
9987 *--dkey = htobe32(ek[4 * 12 + i]);
9990 for (i = 0; i < 4; i++)
9991 *--dkey = htobe32(ek[4 * 13 + i]);
9992 for (i = 0; i < 4; i++)
9993 *--dkey = htobe32(ek[4 * 14 + i]);
9996 MPASS(dkey == dec_key);
9999 static struct sx mlu; /* mod load unload */
10000 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
10003 mod_event(module_t mod, int cmd, void *arg)
10006 static int loaded = 0;
10011 if (loaded++ == 0) {
10013 t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10014 t4_filter_rpl, CPL_COOKIE_FILTER);
10015 t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL,
10016 do_l2t_write_rpl, CPL_COOKIE_FILTER);
10017 t4_register_shared_cpl_handler(CPL_ACT_OPEN_RPL,
10018 t4_hashfilter_ao_rpl, CPL_COOKIE_HASHFILTER);
10019 t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10020 t4_hashfilter_tcb_rpl, CPL_COOKIE_HASHFILTER);
10021 t4_register_shared_cpl_handler(CPL_ABORT_RPL_RSS,
10022 t4_del_hashfilter_rpl, CPL_COOKIE_HASHFILTER);
10023 t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
10024 t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
10025 sx_init(&t4_list_lock, "T4/T5 adapters");
10026 SLIST_INIT(&t4_list);
10028 sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
10029 SLIST_INIT(&t4_uld_list);
10031 t4_tracer_modload();
10039 if (--loaded == 0) {
10042 sx_slock(&t4_list_lock);
10043 if (!SLIST_EMPTY(&t4_list)) {
10045 sx_sunlock(&t4_list_lock);
10049 sx_slock(&t4_uld_list_lock);
10050 if (!SLIST_EMPTY(&t4_uld_list)) {
10052 sx_sunlock(&t4_uld_list_lock);
10053 sx_sunlock(&t4_list_lock);
10058 while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
10059 uprintf("%ju clusters with custom free routine "
10060 "still is use.\n", t4_sge_extfree_refs());
10061 pause("t4unload", 2 * hz);
10064 sx_sunlock(&t4_uld_list_lock);
10066 sx_sunlock(&t4_list_lock);
10068 if (t4_sge_extfree_refs() == 0) {
10069 t4_tracer_modunload();
10071 sx_destroy(&t4_uld_list_lock);
10073 sx_destroy(&t4_list_lock);
10074 t4_sge_modunload();
10078 loaded++; /* undo earlier decrement */
10089 static devclass_t t4_devclass, t5_devclass, t6_devclass;
10090 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
10091 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
10093 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
10094 MODULE_VERSION(t4nex, 1);
10095 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
10097 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
10098 #endif /* DEV_NETMAP */
10100 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
10101 MODULE_VERSION(t5nex, 1);
10102 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
10104 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
10105 #endif /* DEV_NETMAP */
10107 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
10108 MODULE_VERSION(t6nex, 1);
10109 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
10111 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
10112 #endif /* DEV_NETMAP */
10114 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
10115 MODULE_VERSION(cxgbe, 1);
10117 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
10118 MODULE_VERSION(cxl, 1);
10120 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
10121 MODULE_VERSION(cc, 1);
10123 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
10124 MODULE_VERSION(vcxgbe, 1);
10126 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
10127 MODULE_VERSION(vcxl, 1);
10129 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
10130 MODULE_VERSION(vcc, 1);