2 * Copyright (c) 2011 Chelsio Communications, Inc.
4 * Written by: Navdeep Parhar <np@FreeBSD.org>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
32 #include "opt_inet6.h"
35 #include <sys/param.h>
38 #include <sys/kernel.h>
40 #include <sys/module.h>
41 #include <sys/malloc.h>
42 #include <sys/queue.h>
43 #include <sys/taskqueue.h>
44 #include <sys/pciio.h>
45 #include <dev/pci/pcireg.h>
46 #include <dev/pci/pcivar.h>
47 #include <dev/pci/pci_private.h>
48 #include <sys/firmware.h>
51 #include <sys/socket.h>
52 #include <sys/sockio.h>
53 #include <sys/sysctl.h>
54 #include <net/ethernet.h>
56 #include <net/if_types.h>
57 #include <net/if_dl.h>
58 #include <net/if_vlan_var.h>
60 #include <net/rss_config.h>
62 #if defined(__i386__) || defined(__amd64__)
67 #include "common/common.h"
68 #include "common/t4_msg.h"
69 #include "common/t4_regs.h"
70 #include "common/t4_regs_values.h"
73 #include "t4_mp_ring.h"
75 /* T4 bus driver interface */
76 static int t4_probe(device_t);
77 static int t4_attach(device_t);
78 static int t4_detach(device_t);
79 static device_method_t t4_methods[] = {
80 DEVMETHOD(device_probe, t4_probe),
81 DEVMETHOD(device_attach, t4_attach),
82 DEVMETHOD(device_detach, t4_detach),
86 static driver_t t4_driver = {
89 sizeof(struct adapter)
93 /* T4 port (cxgbe) interface */
94 static int cxgbe_probe(device_t);
95 static int cxgbe_attach(device_t);
96 static int cxgbe_detach(device_t);
97 static device_method_t cxgbe_methods[] = {
98 DEVMETHOD(device_probe, cxgbe_probe),
99 DEVMETHOD(device_attach, cxgbe_attach),
100 DEVMETHOD(device_detach, cxgbe_detach),
103 static driver_t cxgbe_driver = {
106 sizeof(struct port_info)
109 static d_ioctl_t t4_ioctl;
110 static d_open_t t4_open;
111 static d_close_t t4_close;
113 static struct cdevsw t4_cdevsw = {
114 .d_version = D_VERSION,
122 /* T5 bus driver interface */
123 static int t5_probe(device_t);
124 static device_method_t t5_methods[] = {
125 DEVMETHOD(device_probe, t5_probe),
126 DEVMETHOD(device_attach, t4_attach),
127 DEVMETHOD(device_detach, t4_detach),
131 static driver_t t5_driver = {
134 sizeof(struct adapter)
138 /* T5 port (cxl) interface */
139 static driver_t cxl_driver = {
142 sizeof(struct port_info)
145 static struct cdevsw t5_cdevsw = {
146 .d_version = D_VERSION,
154 /* ifnet + media interface */
155 static void cxgbe_init(void *);
156 static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
157 static int cxgbe_transmit(struct ifnet *, struct mbuf *);
158 static void cxgbe_qflush(struct ifnet *);
159 static uint64_t cxgbe_get_counter(struct ifnet *, ift_counter);
160 static int cxgbe_media_change(struct ifnet *);
161 static void cxgbe_media_status(struct ifnet *, struct ifmediareq *);
163 MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
166 * Correct lock order when you need to acquire multiple locks is t4_list_lock,
167 * then ADAPTER_LOCK, then t4_uld_list_lock.
169 static struct sx t4_list_lock;
170 SLIST_HEAD(, adapter) t4_list;
172 static struct sx t4_uld_list_lock;
173 SLIST_HEAD(, uld_info) t4_uld_list;
177 * Tunables. See tweak_tunables() too.
179 * Each tunable is set to a default value here if it's known at compile-time.
180 * Otherwise it is set to -1 as an indication to tweak_tunables() that it should
181 * provide a reasonable default when the driver is loaded.
183 * Tunables applicable to both T4 and T5 are under hw.cxgbe. Those specific to
184 * T5 are under hw.cxl.
188 * Number of queues for tx and rx, 10G and 1G, NIC and offload.
191 static int t4_ntxq10g = -1;
192 TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq10g);
195 static int t4_nrxq10g = -1;
196 TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq10g);
199 static int t4_ntxq1g = -1;
200 TUNABLE_INT("hw.cxgbe.ntxq1g", &t4_ntxq1g);
203 static int t4_nrxq1g = -1;
204 TUNABLE_INT("hw.cxgbe.nrxq1g", &t4_nrxq1g);
206 static int t4_rsrv_noflowq = 0;
207 TUNABLE_INT("hw.cxgbe.rsrv_noflowq", &t4_rsrv_noflowq);
210 #define NOFLDTXQ_10G 8
211 static int t4_nofldtxq10g = -1;
212 TUNABLE_INT("hw.cxgbe.nofldtxq10g", &t4_nofldtxq10g);
214 #define NOFLDRXQ_10G 2
215 static int t4_nofldrxq10g = -1;
216 TUNABLE_INT("hw.cxgbe.nofldrxq10g", &t4_nofldrxq10g);
218 #define NOFLDTXQ_1G 2
219 static int t4_nofldtxq1g = -1;
220 TUNABLE_INT("hw.cxgbe.nofldtxq1g", &t4_nofldtxq1g);
222 #define NOFLDRXQ_1G 1
223 static int t4_nofldrxq1g = -1;
224 TUNABLE_INT("hw.cxgbe.nofldrxq1g", &t4_nofldrxq1g);
229 static int t4_nnmtxq10g = -1;
230 TUNABLE_INT("hw.cxgbe.nnmtxq10g", &t4_nnmtxq10g);
233 static int t4_nnmrxq10g = -1;
234 TUNABLE_INT("hw.cxgbe.nnmrxq10g", &t4_nnmrxq10g);
237 static int t4_nnmtxq1g = -1;
238 TUNABLE_INT("hw.cxgbe.nnmtxq1g", &t4_nnmtxq1g);
241 static int t4_nnmrxq1g = -1;
242 TUNABLE_INT("hw.cxgbe.nnmrxq1g", &t4_nnmrxq1g);
246 * Holdoff parameters for 10G and 1G ports.
248 #define TMR_IDX_10G 1
249 static int t4_tmr_idx_10g = TMR_IDX_10G;
250 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx_10g);
252 #define PKTC_IDX_10G (-1)
253 static int t4_pktc_idx_10g = PKTC_IDX_10G;
254 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx_10g);
257 static int t4_tmr_idx_1g = TMR_IDX_1G;
258 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_1G", &t4_tmr_idx_1g);
260 #define PKTC_IDX_1G (-1)
261 static int t4_pktc_idx_1g = PKTC_IDX_1G;
262 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_1G", &t4_pktc_idx_1g);
265 * Size (# of entries) of each tx and rx queue.
267 static unsigned int t4_qsize_txq = TX_EQ_QSIZE;
268 TUNABLE_INT("hw.cxgbe.qsize_txq", &t4_qsize_txq);
270 static unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
271 TUNABLE_INT("hw.cxgbe.qsize_rxq", &t4_qsize_rxq);
274 * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
276 static int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
277 TUNABLE_INT("hw.cxgbe.interrupt_types", &t4_intr_types);
280 * Configuration file.
282 #define DEFAULT_CF "default"
283 #define FLASH_CF "flash"
284 #define UWIRE_CF "uwire"
285 #define FPGA_CF "fpga"
286 static char t4_cfg_file[32] = DEFAULT_CF;
287 TUNABLE_STR("hw.cxgbe.config_file", t4_cfg_file, sizeof(t4_cfg_file));
290 * PAUSE settings (bit 0, 1 = rx_pause, tx_pause respectively).
291 * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
292 * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
293 * mark or when signalled to do so, 0 to never emit PAUSE.
295 static int t4_pause_settings = PAUSE_TX | PAUSE_RX;
296 TUNABLE_INT("hw.cxgbe.pause_settings", &t4_pause_settings);
299 * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
300 * encouraged respectively).
302 static unsigned int t4_fw_install = 1;
303 TUNABLE_INT("hw.cxgbe.fw_install", &t4_fw_install);
306 * ASIC features that will be used. Disable the ones you don't want so that the
307 * chip resources aren't wasted on features that will not be used.
309 static int t4_linkcaps_allowed = 0; /* No DCBX, PPP, etc. by default */
310 TUNABLE_INT("hw.cxgbe.linkcaps_allowed", &t4_linkcaps_allowed);
312 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC;
313 TUNABLE_INT("hw.cxgbe.niccaps_allowed", &t4_niccaps_allowed);
315 static int t4_toecaps_allowed = -1;
316 TUNABLE_INT("hw.cxgbe.toecaps_allowed", &t4_toecaps_allowed);
318 static int t4_rdmacaps_allowed = 0;
319 TUNABLE_INT("hw.cxgbe.rdmacaps_allowed", &t4_rdmacaps_allowed);
321 static int t4_iscsicaps_allowed = 0;
322 TUNABLE_INT("hw.cxgbe.iscsicaps_allowed", &t4_iscsicaps_allowed);
324 static int t4_fcoecaps_allowed = 0;
325 TUNABLE_INT("hw.cxgbe.fcoecaps_allowed", &t4_fcoecaps_allowed);
327 static int t5_write_combine = 0;
328 TUNABLE_INT("hw.cxl.write_combine", &t5_write_combine);
330 struct intrs_and_queues {
331 uint16_t intr_type; /* INTx, MSI, or MSI-X */
332 uint16_t nirq; /* Total # of vectors */
333 uint16_t intr_flags_10g;/* Interrupt flags for each 10G port */
334 uint16_t intr_flags_1g; /* Interrupt flags for each 1G port */
335 uint16_t ntxq10g; /* # of NIC txq's for each 10G port */
336 uint16_t nrxq10g; /* # of NIC rxq's for each 10G port */
337 uint16_t ntxq1g; /* # of NIC txq's for each 1G port */
338 uint16_t nrxq1g; /* # of NIC rxq's for each 1G port */
339 uint16_t rsrv_noflowq; /* Flag whether to reserve queue 0 */
341 uint16_t nofldtxq10g; /* # of TOE txq's for each 10G port */
342 uint16_t nofldrxq10g; /* # of TOE rxq's for each 10G port */
343 uint16_t nofldtxq1g; /* # of TOE txq's for each 1G port */
344 uint16_t nofldrxq1g; /* # of TOE rxq's for each 1G port */
347 uint16_t nnmtxq10g; /* # of netmap txq's for each 10G port */
348 uint16_t nnmrxq10g; /* # of netmap rxq's for each 10G port */
349 uint16_t nnmtxq1g; /* # of netmap txq's for each 1G port */
350 uint16_t nnmrxq1g; /* # of netmap rxq's for each 1G port */
354 struct filter_entry {
355 uint32_t valid:1; /* filter allocated and valid */
356 uint32_t locked:1; /* filter is administratively locked */
357 uint32_t pending:1; /* filter action is pending firmware reply */
358 uint32_t smtidx:8; /* Source MAC Table index for smac */
359 struct l2t_entry *l2t; /* Layer Two Table entry for dmac */
361 struct t4_filter_specification fs;
364 static int map_bars_0_and_4(struct adapter *);
365 static int map_bar_2(struct adapter *);
366 static void setup_memwin(struct adapter *);
367 static int validate_mem_range(struct adapter *, uint32_t, int);
368 static int fwmtype_to_hwmtype(int);
369 static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
371 static void memwin_info(struct adapter *, int, uint32_t *, uint32_t *);
372 static uint32_t position_memwin(struct adapter *, int, uint32_t);
373 static int cfg_itype_and_nqueues(struct adapter *, int, int,
374 struct intrs_and_queues *);
375 static int prep_firmware(struct adapter *);
376 static int partition_resources(struct adapter *, const struct firmware *,
378 static int get_params__pre_init(struct adapter *);
379 static int get_params__post_init(struct adapter *);
380 static int set_params__post_init(struct adapter *);
381 static void t4_set_desc(struct adapter *);
382 static void build_medialist(struct port_info *, struct ifmedia *);
383 static int cxgbe_init_synchronized(struct port_info *);
384 static int cxgbe_uninit_synchronized(struct port_info *);
385 static int setup_intr_handlers(struct adapter *);
386 static void quiesce_txq(struct adapter *, struct sge_txq *);
387 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
388 static void quiesce_iq(struct adapter *, struct sge_iq *);
389 static void quiesce_fl(struct adapter *, struct sge_fl *);
390 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
391 driver_intr_t *, void *, char *);
392 static int t4_free_irq(struct adapter *, struct irq *);
393 static void reg_block_dump(struct adapter *, uint8_t *, unsigned int,
395 static void t4_get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
396 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
397 static void cxgbe_tick(void *);
398 static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
399 static int cpl_not_handled(struct sge_iq *, const struct rss_header *,
401 static int an_not_handled(struct sge_iq *, const struct rsp_ctrl *);
402 static int fw_msg_not_handled(struct adapter *, const __be64 *);
403 static int t4_sysctls(struct adapter *);
404 static int cxgbe_sysctls(struct port_info *);
405 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
406 static int sysctl_bitfield(SYSCTL_HANDLER_ARGS);
407 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
408 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
409 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
410 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
411 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
412 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
413 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
414 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
415 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
417 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
418 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
419 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
420 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
421 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
422 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
423 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
424 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
425 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
426 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
427 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
428 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
429 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
430 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
431 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
432 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
433 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
434 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
435 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
436 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
437 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
438 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
439 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
440 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
441 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
443 static uint32_t fconf_to_mode(uint32_t);
444 static uint32_t mode_to_fconf(uint32_t);
445 static uint32_t fspec_to_fconf(struct t4_filter_specification *);
446 static int get_filter_mode(struct adapter *, uint32_t *);
447 static int set_filter_mode(struct adapter *, uint32_t);
448 static inline uint64_t get_filter_hits(struct adapter *, uint32_t);
449 static int get_filter(struct adapter *, struct t4_filter *);
450 static int set_filter(struct adapter *, struct t4_filter *);
451 static int del_filter(struct adapter *, struct t4_filter *);
452 static void clear_filter(struct filter_entry *);
453 static int set_filter_wr(struct adapter *, int);
454 static int del_filter_wr(struct adapter *, int);
455 static int get_sge_context(struct adapter *, struct t4_sge_context *);
456 static int load_fw(struct adapter *, struct t4_data *);
457 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
458 static int read_i2c(struct adapter *, struct t4_i2c_data *);
459 static int set_sched_class(struct adapter *, struct t4_sched_params *);
460 static int set_sched_queue(struct adapter *, struct t4_sched_queue *);
462 static int toe_capability(struct port_info *, int);
464 static int mod_event(module_t, int, void *);
470 {0xa000, "Chelsio Terminator 4 FPGA"},
471 {0x4400, "Chelsio T440-dbg"},
472 {0x4401, "Chelsio T420-CR"},
473 {0x4402, "Chelsio T422-CR"},
474 {0x4403, "Chelsio T440-CR"},
475 {0x4404, "Chelsio T420-BCH"},
476 {0x4405, "Chelsio T440-BCH"},
477 {0x4406, "Chelsio T440-CH"},
478 {0x4407, "Chelsio T420-SO"},
479 {0x4408, "Chelsio T420-CX"},
480 {0x4409, "Chelsio T420-BT"},
481 {0x440a, "Chelsio T404-BT"},
482 {0x440e, "Chelsio T440-LP-CR"},
484 {0xb000, "Chelsio Terminator 5 FPGA"},
485 {0x5400, "Chelsio T580-dbg"},
486 {0x5401, "Chelsio T520-CR"}, /* 2 x 10G */
487 {0x5402, "Chelsio T522-CR"}, /* 2 x 10G, 2 X 1G */
488 {0x5403, "Chelsio T540-CR"}, /* 4 x 10G */
489 {0x5407, "Chelsio T520-SO"}, /* 2 x 10G, nomem */
490 {0x5409, "Chelsio T520-BT"}, /* 2 x 10GBaseT */
491 {0x540a, "Chelsio T504-BT"}, /* 4 x 1G */
492 {0x540d, "Chelsio T580-CR"}, /* 2 x 40G */
493 {0x540e, "Chelsio T540-LP-CR"}, /* 4 x 10G */
494 {0x5410, "Chelsio T580-LP-CR"}, /* 2 x 40G */
495 {0x5411, "Chelsio T520-LL-CR"}, /* 2 x 10G */
496 {0x5412, "Chelsio T560-CR"}, /* 1 x 40G, 2 x 10G */
497 {0x5414, "Chelsio T580-LP-SO-CR"}, /* 2 x 40G, nomem */
498 {0x5415, "Chelsio T502-BT"}, /* 2 x 1G */
500 {0x5404, "Chelsio T520-BCH"},
501 {0x5405, "Chelsio T540-BCH"},
502 {0x5406, "Chelsio T540-CH"},
503 {0x5408, "Chelsio T520-CX"},
504 {0x540b, "Chelsio B520-SR"},
505 {0x540c, "Chelsio B504-BT"},
506 {0x540f, "Chelsio Amsterdam"},
507 {0x5413, "Chelsio T580-CHR"},
513 * service_iq() has an iq and needs the fl. Offset of fl from the iq should be
514 * exactly the same for both rxq and ofld_rxq.
516 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
517 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
520 /* No easy way to include t4_msg.h before adapter.h so we check this way */
521 CTASSERT(nitems(((struct adapter *)0)->cpl_handler) == NUM_CPL_CMDS);
522 CTASSERT(nitems(((struct adapter *)0)->fw_msg_handler) == NUM_FW6_TYPES);
524 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
527 t4_probe(device_t dev)
530 uint16_t v = pci_get_vendor(dev);
531 uint16_t d = pci_get_device(dev);
532 uint8_t f = pci_get_function(dev);
534 if (v != PCI_VENDOR_ID_CHELSIO)
537 /* Attach only to PF0 of the FPGA */
538 if (d == 0xa000 && f != 0)
541 for (i = 0; i < nitems(t4_pciids); i++) {
542 if (d == t4_pciids[i].device) {
543 device_set_desc(dev, t4_pciids[i].desc);
544 return (BUS_PROBE_DEFAULT);
552 t5_probe(device_t dev)
555 uint16_t v = pci_get_vendor(dev);
556 uint16_t d = pci_get_device(dev);
557 uint8_t f = pci_get_function(dev);
559 if (v != PCI_VENDOR_ID_CHELSIO)
562 /* Attach only to PF0 of the FPGA */
563 if (d == 0xb000 && f != 0)
566 for (i = 0; i < nitems(t5_pciids); i++) {
567 if (d == t5_pciids[i].device) {
568 device_set_desc(dev, t5_pciids[i].desc);
569 return (BUS_PROBE_DEFAULT);
577 t4_attach(device_t dev)
580 int rc = 0, i, n10g, n1g, rqidx, tqidx;
581 struct intrs_and_queues iaq;
584 int ofld_rqidx, ofld_tqidx;
587 int nm_rqidx, nm_tqidx;
590 sc = device_get_softc(dev);
592 TUNABLE_INT_FETCH("hw.cxgbe.debug_flags", &sc->debug_flags);
594 pci_enable_busmaster(dev);
595 if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
598 pci_set_max_read_req(dev, 4096);
599 v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
600 v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
601 pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
603 sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
607 mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
608 snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
609 device_get_nameunit(dev));
611 snprintf(sc->lockname, sizeof(sc->lockname), "%s",
612 device_get_nameunit(dev));
613 mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
614 sx_xlock(&t4_list_lock);
615 SLIST_INSERT_HEAD(&t4_list, sc, link);
616 sx_xunlock(&t4_list_lock);
618 mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
619 TAILQ_INIT(&sc->sfl);
620 callout_init(&sc->sfl_callout, 1);
622 mtx_init(&sc->regwin_lock, "register and memory window", 0, MTX_DEF);
624 rc = map_bars_0_and_4(sc);
626 goto done; /* error message displayed already */
629 * This is the real PF# to which we're attaching. Works from within PCI
630 * passthrough environments too, where pci_get_function() could return a
631 * different PF# depending on the passthrough configuration. We need to
632 * use the real PF# in all our communication with the firmware.
634 sc->pf = G_SOURCEPF(t4_read_reg(sc, A_PL_WHOAMI));
637 memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
638 sc->an_handler = an_not_handled;
639 for (i = 0; i < nitems(sc->cpl_handler); i++)
640 sc->cpl_handler[i] = cpl_not_handled;
641 for (i = 0; i < nitems(sc->fw_msg_handler); i++)
642 sc->fw_msg_handler[i] = fw_msg_not_handled;
643 t4_register_cpl_handler(sc, CPL_SET_TCB_RPL, t4_filter_rpl);
644 t4_register_cpl_handler(sc, CPL_TRACE_PKT, t4_trace_pkt);
645 t4_register_cpl_handler(sc, CPL_TRACE_PKT_T5, t5_trace_pkt);
646 t4_init_sge_cpl_handlers(sc);
648 /* Prepare the adapter for operation */
649 rc = -t4_prep_adapter(sc);
651 device_printf(dev, "failed to prepare adapter: %d.\n", rc);
656 * Do this really early, with the memory windows set up even before the
657 * character device. The userland tool's register i/o and mem read
658 * will work even in "recovery mode".
661 sc->cdev = make_dev(is_t4(sc) ? &t4_cdevsw : &t5_cdevsw,
662 device_get_unit(dev), UID_ROOT, GID_WHEEL, 0600, "%s",
663 device_get_nameunit(dev));
664 if (sc->cdev == NULL)
665 device_printf(dev, "failed to create nexus char device.\n");
667 sc->cdev->si_drv1 = sc;
669 /* Go no further if recovery mode has been requested. */
670 if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
671 device_printf(dev, "recovery mode.\n");
675 #if defined(__i386__)
676 if ((cpu_feature & CPUID_CX8) == 0) {
677 device_printf(dev, "64 bit atomics not available.\n");
683 /* Prepare the firmware for operation */
684 rc = prep_firmware(sc);
686 goto done; /* error message displayed already */
688 rc = get_params__post_init(sc);
690 goto done; /* error message displayed already */
692 rc = set_params__post_init(sc);
694 goto done; /* error message displayed already */
698 goto done; /* error message displayed already */
700 rc = t4_create_dma_tag(sc);
702 goto done; /* error message displayed already */
705 * First pass over all the ports - allocate VIs and initialize some
706 * basic parameters like mac address, port type, etc. We also figure
707 * out whether a port is 10G or 1G and use that information when
708 * calculating how many interrupts to attempt to allocate.
711 for_each_port(sc, i) {
712 struct port_info *pi;
714 pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
717 /* These must be set before t4_port_init */
721 /* Allocate the vi and initialize parameters like mac addr */
722 rc = -t4_port_init(pi, sc->mbox, sc->pf, 0);
724 device_printf(dev, "unable to initialize port %d: %d\n",
731 pi->link_cfg.requested_fc &= ~(PAUSE_TX | PAUSE_RX);
732 pi->link_cfg.requested_fc |= t4_pause_settings;
733 pi->link_cfg.fc &= ~(PAUSE_TX | PAUSE_RX);
734 pi->link_cfg.fc |= t4_pause_settings;
736 rc = -t4_link_start(sc, sc->mbox, pi->tx_chan, &pi->link_cfg);
738 device_printf(dev, "port %d l1cfg failed: %d\n", i, rc);
744 snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
745 device_get_nameunit(dev), i);
746 mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
747 sc->chan_map[pi->tx_chan] = i;
749 if (is_10G_port(pi) || is_40G_port(pi)) {
751 pi->tmr_idx = t4_tmr_idx_10g;
752 pi->pktc_idx = t4_pktc_idx_10g;
755 pi->tmr_idx = t4_tmr_idx_1g;
756 pi->pktc_idx = t4_pktc_idx_1g;
759 pi->xact_addr_filt = -1;
762 pi->qsize_rxq = t4_qsize_rxq;
763 pi->qsize_txq = t4_qsize_txq;
765 pi->dev = device_add_child(dev, is_t4(sc) ? "cxgbe" : "cxl", -1);
766 if (pi->dev == NULL) {
768 "failed to add device for port %d.\n", i);
772 device_set_softc(pi->dev, pi);
776 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
778 rc = cfg_itype_and_nqueues(sc, n10g, n1g, &iaq);
780 goto done; /* error message displayed already */
782 sc->intr_type = iaq.intr_type;
783 sc->intr_count = iaq.nirq;
786 s->nrxq = n10g * iaq.nrxq10g + n1g * iaq.nrxq1g;
787 s->ntxq = n10g * iaq.ntxq10g + n1g * iaq.ntxq1g;
788 s->neq = s->ntxq + s->nrxq; /* the free list in an rxq is an eq */
789 s->neq += sc->params.nports + 1;/* ctrl queues: 1 per port + 1 mgmt */
790 s->niq = s->nrxq + 1; /* 1 extra for firmware event queue */
792 if (is_offload(sc)) {
793 s->nofldrxq = n10g * iaq.nofldrxq10g + n1g * iaq.nofldrxq1g;
794 s->nofldtxq = n10g * iaq.nofldtxq10g + n1g * iaq.nofldtxq1g;
795 s->neq += s->nofldtxq + s->nofldrxq;
796 s->niq += s->nofldrxq;
798 s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
799 M_CXGBE, M_ZERO | M_WAITOK);
800 s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
801 M_CXGBE, M_ZERO | M_WAITOK);
805 s->nnmrxq = n10g * iaq.nnmrxq10g + n1g * iaq.nnmrxq1g;
806 s->nnmtxq = n10g * iaq.nnmtxq10g + n1g * iaq.nnmtxq1g;
807 s->neq += s->nnmtxq + s->nnmrxq;
810 s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
811 M_CXGBE, M_ZERO | M_WAITOK);
812 s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
813 M_CXGBE, M_ZERO | M_WAITOK);
816 s->ctrlq = malloc(sc->params.nports * sizeof(struct sge_wrq), M_CXGBE,
818 s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
820 s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
822 s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
824 s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
827 sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
830 t4_init_l2t(sc, M_WAITOK);
833 * Second pass over the ports. This time we know the number of rx and
834 * tx queues that each port should get.
838 ofld_rqidx = ofld_tqidx = 0;
841 nm_rqidx = nm_tqidx = 0;
843 for_each_port(sc, i) {
844 struct port_info *pi = sc->port[i];
849 pi->first_rxq = rqidx;
850 pi->first_txq = tqidx;
851 if (is_10G_port(pi) || is_40G_port(pi)) {
852 pi->flags |= iaq.intr_flags_10g;
853 pi->nrxq = iaq.nrxq10g;
854 pi->ntxq = iaq.ntxq10g;
856 pi->flags |= iaq.intr_flags_1g;
857 pi->nrxq = iaq.nrxq1g;
858 pi->ntxq = iaq.ntxq1g;
862 pi->rsrv_noflowq = iaq.rsrv_noflowq ? 1 : 0;
864 pi->rsrv_noflowq = 0;
869 if (is_offload(sc)) {
870 pi->first_ofld_rxq = ofld_rqidx;
871 pi->first_ofld_txq = ofld_tqidx;
872 if (is_10G_port(pi) || is_40G_port(pi)) {
873 pi->nofldrxq = iaq.nofldrxq10g;
874 pi->nofldtxq = iaq.nofldtxq10g;
876 pi->nofldrxq = iaq.nofldrxq1g;
877 pi->nofldtxq = iaq.nofldtxq1g;
879 ofld_rqidx += pi->nofldrxq;
880 ofld_tqidx += pi->nofldtxq;
884 pi->first_nm_rxq = nm_rqidx;
885 pi->first_nm_txq = nm_tqidx;
886 if (is_10G_port(pi) || is_40G_port(pi)) {
887 pi->nnmrxq = iaq.nnmrxq10g;
888 pi->nnmtxq = iaq.nnmtxq10g;
890 pi->nnmrxq = iaq.nnmrxq1g;
891 pi->nnmtxq = iaq.nnmtxq1g;
893 nm_rqidx += pi->nnmrxq;
894 nm_tqidx += pi->nnmtxq;
898 rc = setup_intr_handlers(sc);
901 "failed to setup interrupt handlers: %d\n", rc);
905 rc = bus_generic_attach(dev);
908 "failed to attach all child ports: %d\n", rc);
913 "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
914 sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
915 sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
916 (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
917 sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
922 if (rc != 0 && sc->cdev) {
923 /* cdev was created and so cxgbetool works; recover that way. */
925 "error during attach, adapter is now in recovery mode.\n");
941 t4_detach(device_t dev)
944 struct port_info *pi;
947 sc = device_get_softc(dev);
949 if (sc->flags & FULL_INIT_DONE)
953 destroy_dev(sc->cdev);
957 rc = bus_generic_detach(dev);
960 "failed to detach child devices: %d\n", rc);
964 for (i = 0; i < sc->intr_count; i++)
965 t4_free_irq(sc, &sc->irq[i]);
967 for (i = 0; i < MAX_NPORTS; i++) {
970 t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->viid);
972 device_delete_child(dev, pi->dev);
974 mtx_destroy(&pi->pi_lock);
979 if (sc->flags & FULL_INIT_DONE)
980 adapter_full_uninit(sc);
982 if (sc->flags & FW_OK)
983 t4_fw_bye(sc, sc->mbox);
985 if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
986 pci_release_msi(dev);
989 bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
993 bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
997 bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1001 t4_free_l2t(sc->l2t);
1004 free(sc->sge.ofld_rxq, M_CXGBE);
1005 free(sc->sge.ofld_txq, M_CXGBE);
1008 free(sc->sge.nm_rxq, M_CXGBE);
1009 free(sc->sge.nm_txq, M_CXGBE);
1011 free(sc->irq, M_CXGBE);
1012 free(sc->sge.rxq, M_CXGBE);
1013 free(sc->sge.txq, M_CXGBE);
1014 free(sc->sge.ctrlq, M_CXGBE);
1015 free(sc->sge.iqmap, M_CXGBE);
1016 free(sc->sge.eqmap, M_CXGBE);
1017 free(sc->tids.ftid_tab, M_CXGBE);
1018 t4_destroy_dma_tag(sc);
1019 if (mtx_initialized(&sc->sc_lock)) {
1020 sx_xlock(&t4_list_lock);
1021 SLIST_REMOVE(&t4_list, sc, adapter, link);
1022 sx_xunlock(&t4_list_lock);
1023 mtx_destroy(&sc->sc_lock);
1026 if (mtx_initialized(&sc->tids.ftid_lock))
1027 mtx_destroy(&sc->tids.ftid_lock);
1028 if (mtx_initialized(&sc->sfl_lock))
1029 mtx_destroy(&sc->sfl_lock);
1030 if (mtx_initialized(&sc->ifp_lock))
1031 mtx_destroy(&sc->ifp_lock);
1032 if (mtx_initialized(&sc->regwin_lock))
1033 mtx_destroy(&sc->regwin_lock);
1035 bzero(sc, sizeof(*sc));
1041 cxgbe_probe(device_t dev)
1044 struct port_info *pi = device_get_softc(dev);
1046 snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1047 device_set_desc_copy(dev, buf);
1049 return (BUS_PROBE_DEFAULT);
1052 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1053 IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1054 IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS)
1055 #define T4_CAP_ENABLE (T4_CAP)
1058 cxgbe_attach(device_t dev)
1060 struct port_info *pi = device_get_softc(dev);
1065 /* Allocate an ifnet and set it up */
1066 ifp = if_alloc(IFT_ETHER);
1068 device_printf(dev, "Cannot allocate ifnet\n");
1074 callout_init(&pi->tick, 1);
1076 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1077 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1079 ifp->if_init = cxgbe_init;
1080 ifp->if_ioctl = cxgbe_ioctl;
1081 ifp->if_transmit = cxgbe_transmit;
1082 ifp->if_qflush = cxgbe_qflush;
1083 ifp->if_get_counter = cxgbe_get_counter;
1085 ifp->if_capabilities = T4_CAP;
1087 if (is_offload(pi->adapter))
1088 ifp->if_capabilities |= IFCAP_TOE;
1090 ifp->if_capenable = T4_CAP_ENABLE;
1091 ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1092 CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1094 ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1095 ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1096 ifp->if_hw_tsomaxsegsize = 65536;
1098 /* Initialize ifmedia for this port */
1099 ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
1100 cxgbe_media_status);
1101 build_medialist(pi, &pi->media);
1103 pi->vlan_c = EVENTHANDLER_REGISTER(vlan_config, cxgbe_vlan_config, ifp,
1104 EVENTHANDLER_PRI_ANY);
1106 ether_ifattach(ifp, pi->hw_addr);
1109 s = malloc(n, M_CXGBE, M_WAITOK);
1110 o = snprintf(s, n, "%d txq, %d rxq (NIC)", pi->ntxq, pi->nrxq);
1113 if (is_offload(pi->adapter)) {
1114 o += snprintf(s + o, n - o, "; %d txq, %d rxq (TOE)",
1115 pi->nofldtxq, pi->nofldrxq);
1120 o += snprintf(s + o, n - o, "; %d txq, %d rxq (netmap)", pi->nnmtxq,
1124 device_printf(dev, "%s\n", s);
1128 /* nm_media handled here to keep implementation private to this file */
1129 ifmedia_init(&pi->nm_media, IFM_IMASK, cxgbe_media_change,
1130 cxgbe_media_status);
1131 build_medialist(pi, &pi->nm_media);
1132 create_netmap_ifnet(pi); /* logs errors it something fails */
1140 cxgbe_detach(device_t dev)
1142 struct port_info *pi = device_get_softc(dev);
1143 struct adapter *sc = pi->adapter;
1144 struct ifnet *ifp = pi->ifp;
1146 /* Tell if_ioctl and if_init that the port is going away */
1151 mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
1154 sc->last_op = "t4detach";
1155 sc->last_op_thr = curthread;
1156 sc->last_op_flags = 0;
1160 if (pi->flags & HAS_TRACEQ) {
1161 sc->traceq = -1; /* cloner should not create ifnet */
1162 t4_tracer_port_detach(sc);
1166 EVENTHANDLER_DEREGISTER(vlan_config, pi->vlan_c);
1169 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1170 callout_stop(&pi->tick);
1172 callout_drain(&pi->tick);
1174 /* Let detach proceed even if these fail. */
1175 cxgbe_uninit_synchronized(pi);
1176 port_full_uninit(pi);
1178 ifmedia_removeall(&pi->media);
1179 ether_ifdetach(pi->ifp);
1183 /* XXXNM: equivalent of cxgbe_uninit_synchronized to ifdown nm_ifp */
1184 destroy_netmap_ifnet(pi);
1196 cxgbe_init(void *arg)
1198 struct port_info *pi = arg;
1199 struct adapter *sc = pi->adapter;
1201 if (begin_synchronized_op(sc, pi, SLEEP_OK | INTR_OK, "t4init") != 0)
1203 cxgbe_init_synchronized(pi);
1204 end_synchronized_op(sc, 0);
1208 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1210 int rc = 0, mtu, flags, can_sleep;
1211 struct port_info *pi = ifp->if_softc;
1212 struct adapter *sc = pi->adapter;
1213 struct ifreq *ifr = (struct ifreq *)data;
1219 if ((mtu < ETHERMIN) || (mtu > ETHERMTU_JUMBO))
1222 rc = begin_synchronized_op(sc, pi, SLEEP_OK | INTR_OK, "t4mtu");
1226 if (pi->flags & PORT_INIT_DONE) {
1227 t4_update_fl_bufsize(ifp);
1228 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1229 rc = update_mac_settings(ifp, XGMAC_MTU);
1231 end_synchronized_op(sc, 0);
1237 rc = begin_synchronized_op(sc, pi,
1238 can_sleep ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4flg");
1242 if (ifp->if_flags & IFF_UP) {
1243 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1244 flags = pi->if_flags;
1245 if ((ifp->if_flags ^ flags) &
1246 (IFF_PROMISC | IFF_ALLMULTI)) {
1247 if (can_sleep == 1) {
1248 end_synchronized_op(sc, 0);
1252 rc = update_mac_settings(ifp,
1253 XGMAC_PROMISC | XGMAC_ALLMULTI);
1256 if (can_sleep == 0) {
1257 end_synchronized_op(sc, LOCK_HELD);
1261 rc = cxgbe_init_synchronized(pi);
1263 pi->if_flags = ifp->if_flags;
1264 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1265 if (can_sleep == 0) {
1266 end_synchronized_op(sc, LOCK_HELD);
1270 rc = cxgbe_uninit_synchronized(pi);
1272 end_synchronized_op(sc, can_sleep ? 0 : LOCK_HELD);
1276 case SIOCDELMULTI: /* these two are called with a mutex held :-( */
1277 rc = begin_synchronized_op(sc, pi, HOLD_LOCK, "t4multi");
1280 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1281 rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1282 end_synchronized_op(sc, LOCK_HELD);
1286 rc = begin_synchronized_op(sc, pi, SLEEP_OK | INTR_OK, "t4cap");
1290 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1291 if (mask & IFCAP_TXCSUM) {
1292 ifp->if_capenable ^= IFCAP_TXCSUM;
1293 ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1295 if (IFCAP_TSO4 & ifp->if_capenable &&
1296 !(IFCAP_TXCSUM & ifp->if_capenable)) {
1297 ifp->if_capenable &= ~IFCAP_TSO4;
1299 "tso4 disabled due to -txcsum.\n");
1302 if (mask & IFCAP_TXCSUM_IPV6) {
1303 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1304 ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1306 if (IFCAP_TSO6 & ifp->if_capenable &&
1307 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1308 ifp->if_capenable &= ~IFCAP_TSO6;
1310 "tso6 disabled due to -txcsum6.\n");
1313 if (mask & IFCAP_RXCSUM)
1314 ifp->if_capenable ^= IFCAP_RXCSUM;
1315 if (mask & IFCAP_RXCSUM_IPV6)
1316 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1319 * Note that we leave CSUM_TSO alone (it is always set). The
1320 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1321 * sending a TSO request our way, so it's sufficient to toggle
1324 if (mask & IFCAP_TSO4) {
1325 if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1326 !(IFCAP_TXCSUM & ifp->if_capenable)) {
1327 if_printf(ifp, "enable txcsum first.\n");
1331 ifp->if_capenable ^= IFCAP_TSO4;
1333 if (mask & IFCAP_TSO6) {
1334 if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1335 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1336 if_printf(ifp, "enable txcsum6 first.\n");
1340 ifp->if_capenable ^= IFCAP_TSO6;
1342 if (mask & IFCAP_LRO) {
1343 #if defined(INET) || defined(INET6)
1345 struct sge_rxq *rxq;
1347 ifp->if_capenable ^= IFCAP_LRO;
1348 for_each_rxq(pi, i, rxq) {
1349 if (ifp->if_capenable & IFCAP_LRO)
1350 rxq->iq.flags |= IQ_LRO_ENABLED;
1352 rxq->iq.flags &= ~IQ_LRO_ENABLED;
1357 if (mask & IFCAP_TOE) {
1358 int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1360 rc = toe_capability(pi, enable);
1364 ifp->if_capenable ^= mask;
1367 if (mask & IFCAP_VLAN_HWTAGGING) {
1368 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1369 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1370 rc = update_mac_settings(ifp, XGMAC_VLANEX);
1372 if (mask & IFCAP_VLAN_MTU) {
1373 ifp->if_capenable ^= IFCAP_VLAN_MTU;
1375 /* Need to find out how to disable auto-mtu-inflation */
1377 if (mask & IFCAP_VLAN_HWTSO)
1378 ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1379 if (mask & IFCAP_VLAN_HWCSUM)
1380 ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1382 #ifdef VLAN_CAPABILITIES
1383 VLAN_CAPABILITIES(ifp);
1386 end_synchronized_op(sc, 0);
1391 ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
1395 struct ifi2creq i2c;
1397 rc = copyin(ifr->ifr_data, &i2c, sizeof(i2c));
1400 if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1404 if (i2c.len > sizeof(i2c.data)) {
1408 rc = begin_synchronized_op(sc, pi, SLEEP_OK | INTR_OK, "t4i2c");
1411 rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
1412 i2c.offset, i2c.len, &i2c.data[0]);
1413 end_synchronized_op(sc, 0);
1415 rc = copyout(&i2c, ifr->ifr_data, sizeof(i2c));
1420 rc = ether_ioctl(ifp, cmd, data);
1427 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1429 struct port_info *pi = ifp->if_softc;
1430 struct adapter *sc = pi->adapter;
1431 struct sge_txq *txq;
1436 MPASS(m->m_nextpkt == NULL); /* not quite ready for this yet */
1438 if (__predict_false(pi->link_cfg.link_ok == 0)) {
1444 if (__predict_false(rc != 0)) {
1445 MPASS(m == NULL); /* was freed already */
1446 atomic_add_int(&pi->tx_parse_error, 1); /* rare, atomic is ok */
1451 txq = &sc->sge.txq[pi->first_txq];
1452 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1453 txq += ((m->m_pkthdr.flowid % (pi->ntxq - pi->rsrv_noflowq)) +
1457 rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1458 if (__predict_false(rc != 0))
1465 cxgbe_qflush(struct ifnet *ifp)
1467 struct port_info *pi = ifp->if_softc;
1468 struct sge_txq *txq;
1471 /* queues do not exist if !PORT_INIT_DONE. */
1472 if (pi->flags & PORT_INIT_DONE) {
1473 for_each_txq(pi, i, txq) {
1475 txq->eq.flags &= ~EQ_ENABLED;
1477 while (!mp_ring_is_idle(txq->r)) {
1478 mp_ring_check_drainage(txq->r, 0);
1487 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
1489 struct port_info *pi = ifp->if_softc;
1490 struct adapter *sc = pi->adapter;
1491 struct port_stats *s = &pi->stats;
1493 cxgbe_refresh_stats(sc, pi);
1496 case IFCOUNTER_IPACKETS:
1497 return (s->rx_frames - s->rx_pause);
1499 case IFCOUNTER_IERRORS:
1500 return (s->rx_jabber + s->rx_runt + s->rx_too_long +
1501 s->rx_fcs_err + s->rx_len_err);
1503 case IFCOUNTER_OPACKETS:
1504 return (s->tx_frames - s->tx_pause);
1506 case IFCOUNTER_OERRORS:
1507 return (s->tx_error_frames);
1509 case IFCOUNTER_IBYTES:
1510 return (s->rx_octets - s->rx_pause * 64);
1512 case IFCOUNTER_OBYTES:
1513 return (s->tx_octets - s->tx_pause * 64);
1515 case IFCOUNTER_IMCASTS:
1516 return (s->rx_mcast_frames - s->rx_pause);
1518 case IFCOUNTER_OMCASTS:
1519 return (s->tx_mcast_frames - s->tx_pause);
1521 case IFCOUNTER_IQDROPS:
1522 return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
1523 s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
1524 s->rx_trunc3 + pi->tnl_cong_drops);
1526 case IFCOUNTER_OQDROPS: {
1530 if (pi->flags & PORT_INIT_DONE) {
1532 struct sge_txq *txq;
1534 for_each_txq(pi, i, txq)
1535 drops += counter_u64_fetch(txq->r->drops);
1543 return (if_get_counter_default(ifp, c));
1548 cxgbe_media_change(struct ifnet *ifp)
1550 struct port_info *pi = ifp->if_softc;
1552 device_printf(pi->dev, "%s unimplemented.\n", __func__);
1554 return (EOPNOTSUPP);
1558 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
1560 struct port_info *pi = ifp->if_softc;
1561 struct ifmedia *media = NULL;
1562 struct ifmedia_entry *cur;
1563 int speed = pi->link_cfg.speed;
1568 else if (ifp == pi->nm_ifp)
1569 media = &pi->nm_media;
1571 MPASS(media != NULL);
1573 cur = media->ifm_cur;
1575 ifmr->ifm_status = IFM_AVALID;
1576 if (!pi->link_cfg.link_ok)
1579 ifmr->ifm_status |= IFM_ACTIVE;
1581 /* active and current will differ iff current media is autoselect. */
1582 if (IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
1585 ifmr->ifm_active = IFM_ETHER | IFM_FDX;
1586 if (speed == SPEED_10000)
1587 ifmr->ifm_active |= IFM_10G_T;
1588 else if (speed == SPEED_1000)
1589 ifmr->ifm_active |= IFM_1000_T;
1590 else if (speed == SPEED_100)
1591 ifmr->ifm_active |= IFM_100_TX;
1592 else if (speed == SPEED_10)
1593 ifmr->ifm_active |= IFM_10_T;
1595 KASSERT(0, ("%s: link up but speed unknown (%u)", __func__,
1600 t4_fatal_err(struct adapter *sc)
1602 t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
1603 t4_intr_disable(sc);
1604 log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
1605 device_get_nameunit(sc->dev));
1609 map_bars_0_and_4(struct adapter *sc)
1611 sc->regs_rid = PCIR_BAR(0);
1612 sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
1613 &sc->regs_rid, RF_ACTIVE);
1614 if (sc->regs_res == NULL) {
1615 device_printf(sc->dev, "cannot map registers.\n");
1618 sc->bt = rman_get_bustag(sc->regs_res);
1619 sc->bh = rman_get_bushandle(sc->regs_res);
1620 sc->mmio_len = rman_get_size(sc->regs_res);
1621 setbit(&sc->doorbells, DOORBELL_KDB);
1623 sc->msix_rid = PCIR_BAR(4);
1624 sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
1625 &sc->msix_rid, RF_ACTIVE);
1626 if (sc->msix_res == NULL) {
1627 device_printf(sc->dev, "cannot map MSI-X BAR.\n");
1635 map_bar_2(struct adapter *sc)
1639 * T4: only iWARP driver uses the userspace doorbells. There is no need
1640 * to map it if RDMA is disabled.
1642 if (is_t4(sc) && sc->rdmacaps == 0)
1645 sc->udbs_rid = PCIR_BAR(2);
1646 sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
1647 &sc->udbs_rid, RF_ACTIVE);
1648 if (sc->udbs_res == NULL) {
1649 device_printf(sc->dev, "cannot map doorbell BAR.\n");
1652 sc->udbs_base = rman_get_virtual(sc->udbs_res);
1655 setbit(&sc->doorbells, DOORBELL_UDB);
1656 #if defined(__i386__) || defined(__amd64__)
1657 if (t5_write_combine) {
1661 * Enable write combining on BAR2. This is the
1662 * userspace doorbell BAR and is split into 128B
1663 * (UDBS_SEG_SIZE) doorbell regions, each associated
1664 * with an egress queue. The first 64B has the doorbell
1665 * and the second 64B can be used to submit a tx work
1666 * request with an implicit doorbell.
1669 rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
1670 rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
1672 clrbit(&sc->doorbells, DOORBELL_UDB);
1673 setbit(&sc->doorbells, DOORBELL_WCWR);
1674 setbit(&sc->doorbells, DOORBELL_UDBWC);
1676 device_printf(sc->dev,
1677 "couldn't enable write combining: %d\n",
1681 t4_write_reg(sc, A_SGE_STAT_CFG,
1682 V_STATSOURCE_T5(7) | V_STATMODE(0));
1690 static const struct memwin t4_memwin[] = {
1691 { MEMWIN0_BASE, MEMWIN0_APERTURE },
1692 { MEMWIN1_BASE, MEMWIN1_APERTURE },
1693 { MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
1696 static const struct memwin t5_memwin[] = {
1697 { MEMWIN0_BASE, MEMWIN0_APERTURE },
1698 { MEMWIN1_BASE, MEMWIN1_APERTURE },
1699 { MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
1703 setup_memwin(struct adapter *sc)
1705 const struct memwin *mw;
1711 * Read low 32b of bar0 indirectly via the hardware backdoor
1712 * mechanism. Works from within PCI passthrough environments
1713 * too, where rman_get_start() can return a different value. We
1714 * need to program the T4 memory window decoders with the actual
1715 * addresses that will be coming across the PCIe link.
1717 bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
1718 bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
1721 n = nitems(t4_memwin);
1723 /* T5 uses the relative offset inside the PCIe BAR */
1727 n = nitems(t5_memwin);
1730 for (i = 0; i < n; i++, mw++) {
1732 PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
1733 (mw->base + bar0) | V_BIR(0) |
1734 V_WINDOW(ilog2(mw->aperture) - 10));
1738 t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
1742 * Verify that the memory range specified by the addr/len pair is valid and lies
1743 * entirely within a single region (EDCx or MCx).
1746 validate_mem_range(struct adapter *sc, uint32_t addr, int len)
1748 uint32_t em, addr_len, maddr, mlen;
1750 /* Memory can only be accessed in naturally aligned 4 byte units */
1751 if (addr & 3 || len & 3 || len == 0)
1754 /* Enabled memories */
1755 em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
1756 if (em & F_EDRAM0_ENABLE) {
1757 addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
1758 maddr = G_EDRAM0_BASE(addr_len) << 20;
1759 mlen = G_EDRAM0_SIZE(addr_len) << 20;
1760 if (mlen > 0 && addr >= maddr && addr < maddr + mlen &&
1761 addr + len <= maddr + mlen)
1764 if (em & F_EDRAM1_ENABLE) {
1765 addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
1766 maddr = G_EDRAM1_BASE(addr_len) << 20;
1767 mlen = G_EDRAM1_SIZE(addr_len) << 20;
1768 if (mlen > 0 && addr >= maddr && addr < maddr + mlen &&
1769 addr + len <= maddr + mlen)
1772 if (em & F_EXT_MEM_ENABLE) {
1773 addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
1774 maddr = G_EXT_MEM_BASE(addr_len) << 20;
1775 mlen = G_EXT_MEM_SIZE(addr_len) << 20;
1776 if (mlen > 0 && addr >= maddr && addr < maddr + mlen &&
1777 addr + len <= maddr + mlen)
1780 if (!is_t4(sc) && em & F_EXT_MEM1_ENABLE) {
1781 addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
1782 maddr = G_EXT_MEM1_BASE(addr_len) << 20;
1783 mlen = G_EXT_MEM1_SIZE(addr_len) << 20;
1784 if (mlen > 0 && addr >= maddr && addr < maddr + mlen &&
1785 addr + len <= maddr + mlen)
1793 fwmtype_to_hwmtype(int mtype)
1797 case FW_MEMTYPE_EDC0:
1799 case FW_MEMTYPE_EDC1:
1801 case FW_MEMTYPE_EXTMEM:
1803 case FW_MEMTYPE_EXTMEM1:
1806 panic("%s: cannot translate fw mtype %d.", __func__, mtype);
1811 * Verify that the memory range specified by the memtype/offset/len pair is
1812 * valid and lies entirely within the memtype specified. The global address of
1813 * the start of the range is returned in addr.
1816 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
1819 uint32_t em, addr_len, maddr, mlen;
1821 /* Memory can only be accessed in naturally aligned 4 byte units */
1822 if (off & 3 || len & 3 || len == 0)
1825 em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
1826 switch (fwmtype_to_hwmtype(mtype)) {
1828 if (!(em & F_EDRAM0_ENABLE))
1830 addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
1831 maddr = G_EDRAM0_BASE(addr_len) << 20;
1832 mlen = G_EDRAM0_SIZE(addr_len) << 20;
1835 if (!(em & F_EDRAM1_ENABLE))
1837 addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
1838 maddr = G_EDRAM1_BASE(addr_len) << 20;
1839 mlen = G_EDRAM1_SIZE(addr_len) << 20;
1842 if (!(em & F_EXT_MEM_ENABLE))
1844 addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
1845 maddr = G_EXT_MEM_BASE(addr_len) << 20;
1846 mlen = G_EXT_MEM_SIZE(addr_len) << 20;
1849 if (is_t4(sc) || !(em & F_EXT_MEM1_ENABLE))
1851 addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
1852 maddr = G_EXT_MEM1_BASE(addr_len) << 20;
1853 mlen = G_EXT_MEM1_SIZE(addr_len) << 20;
1859 if (mlen > 0 && off < mlen && off + len <= mlen) {
1860 *addr = maddr + off; /* global address */
1868 memwin_info(struct adapter *sc, int win, uint32_t *base, uint32_t *aperture)
1870 const struct memwin *mw;
1873 KASSERT(win >= 0 && win < nitems(t4_memwin),
1874 ("%s: incorrect memwin# (%d)", __func__, win));
1875 mw = &t4_memwin[win];
1877 KASSERT(win >= 0 && win < nitems(t5_memwin),
1878 ("%s: incorrect memwin# (%d)", __func__, win));
1879 mw = &t5_memwin[win];
1884 if (aperture != NULL)
1885 *aperture = mw->aperture;
1889 * Positions the memory window such that it can be used to access the specified
1890 * address in the chip's address space. The return value is the offset of addr
1891 * from the start of the window.
1894 position_memwin(struct adapter *sc, int n, uint32_t addr)
1899 KASSERT(n >= 0 && n <= 3,
1900 ("%s: invalid window %d.", __func__, n));
1901 KASSERT((addr & 3) == 0,
1902 ("%s: addr (0x%x) is not at a 4B boundary.", __func__, addr));
1906 start = addr & ~0xf; /* start must be 16B aligned */
1908 pf = V_PFNUM(sc->pf);
1909 start = addr & ~0x7f; /* start must be 128B aligned */
1911 reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, n);
1913 t4_write_reg(sc, reg, start | pf);
1914 t4_read_reg(sc, reg);
1916 return (addr - start);
1920 cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g,
1921 struct intrs_and_queues *iaq)
1923 int rc, itype, navail, nrxq10g, nrxq1g, n;
1924 int nofldrxq10g = 0, nofldrxq1g = 0;
1925 int nnmrxq10g = 0, nnmrxq1g = 0;
1927 bzero(iaq, sizeof(*iaq));
1929 iaq->ntxq10g = t4_ntxq10g;
1930 iaq->ntxq1g = t4_ntxq1g;
1931 iaq->nrxq10g = nrxq10g = t4_nrxq10g;
1932 iaq->nrxq1g = nrxq1g = t4_nrxq1g;
1933 iaq->rsrv_noflowq = t4_rsrv_noflowq;
1935 if (is_offload(sc)) {
1936 iaq->nofldtxq10g = t4_nofldtxq10g;
1937 iaq->nofldtxq1g = t4_nofldtxq1g;
1938 iaq->nofldrxq10g = nofldrxq10g = t4_nofldrxq10g;
1939 iaq->nofldrxq1g = nofldrxq1g = t4_nofldrxq1g;
1943 iaq->nnmtxq10g = t4_nnmtxq10g;
1944 iaq->nnmtxq1g = t4_nnmtxq1g;
1945 iaq->nnmrxq10g = nnmrxq10g = t4_nnmrxq10g;
1946 iaq->nnmrxq1g = nnmrxq1g = t4_nnmrxq1g;
1949 for (itype = INTR_MSIX; itype; itype >>= 1) {
1951 if ((itype & t4_intr_types) == 0)
1952 continue; /* not allowed */
1954 if (itype == INTR_MSIX)
1955 navail = pci_msix_count(sc->dev);
1956 else if (itype == INTR_MSI)
1957 navail = pci_msi_count(sc->dev);
1964 iaq->intr_type = itype;
1965 iaq->intr_flags_10g = 0;
1966 iaq->intr_flags_1g = 0;
1969 * Best option: an interrupt vector for errors, one for the
1970 * firmware event queue, and one for every rxq (NIC, TOE, and
1973 iaq->nirq = T4_EXTRA_INTR;
1974 iaq->nirq += n10g * (nrxq10g + nofldrxq10g + nnmrxq10g);
1975 iaq->nirq += n1g * (nrxq1g + nofldrxq1g + nnmrxq1g);
1976 if (iaq->nirq <= navail &&
1977 (itype != INTR_MSI || powerof2(iaq->nirq))) {
1978 iaq->intr_flags_10g = INTR_ALL;
1979 iaq->intr_flags_1g = INTR_ALL;
1984 * Second best option: a vector for errors, one for the firmware
1985 * event queue, and vectors for either all the NIC rx queues or
1986 * all the TOE rx queues. The queues that don't get vectors
1987 * will forward their interrupts to those that do.
1989 * Note: netmap rx queues cannot be created early and so they
1990 * can't be setup to receive forwarded interrupts for others.
1992 iaq->nirq = T4_EXTRA_INTR;
1993 if (nrxq10g >= nofldrxq10g) {
1994 iaq->intr_flags_10g = INTR_RXQ;
1995 iaq->nirq += n10g * nrxq10g;
1997 iaq->nnmrxq10g = min(nnmrxq10g, nrxq10g);
2000 iaq->intr_flags_10g = INTR_OFLD_RXQ;
2001 iaq->nirq += n10g * nofldrxq10g;
2003 iaq->nnmrxq10g = min(nnmrxq10g, nofldrxq10g);
2006 if (nrxq1g >= nofldrxq1g) {
2007 iaq->intr_flags_1g = INTR_RXQ;
2008 iaq->nirq += n1g * nrxq1g;
2010 iaq->nnmrxq1g = min(nnmrxq1g, nrxq1g);
2013 iaq->intr_flags_1g = INTR_OFLD_RXQ;
2014 iaq->nirq += n1g * nofldrxq1g;
2016 iaq->nnmrxq1g = min(nnmrxq1g, nofldrxq1g);
2019 if (iaq->nirq <= navail &&
2020 (itype != INTR_MSI || powerof2(iaq->nirq)))
2024 * Next best option: an interrupt vector for errors, one for the
2025 * firmware event queue, and at least one per port. At this
2026 * point we know we'll have to downsize nrxq and/or nofldrxq
2027 * and/or nnmrxq to fit what's available to us.
2029 iaq->nirq = T4_EXTRA_INTR;
2030 iaq->nirq += n10g + n1g;
2031 if (iaq->nirq <= navail) {
2032 int leftover = navail - iaq->nirq;
2035 int target = max(nrxq10g, nofldrxq10g);
2037 iaq->intr_flags_10g = nrxq10g >= nofldrxq10g ?
2038 INTR_RXQ : INTR_OFLD_RXQ;
2041 while (n < target && leftover >= n10g) {
2046 iaq->nrxq10g = min(n, nrxq10g);
2048 iaq->nofldrxq10g = min(n, nofldrxq10g);
2051 iaq->nnmrxq10g = min(n, nnmrxq10g);
2056 int target = max(nrxq1g, nofldrxq1g);
2058 iaq->intr_flags_1g = nrxq1g >= nofldrxq1g ?
2059 INTR_RXQ : INTR_OFLD_RXQ;
2062 while (n < target && leftover >= n1g) {
2067 iaq->nrxq1g = min(n, nrxq1g);
2069 iaq->nofldrxq1g = min(n, nofldrxq1g);
2072 iaq->nnmrxq1g = min(n, nnmrxq1g);
2076 if (itype != INTR_MSI || powerof2(iaq->nirq))
2081 * Least desirable option: one interrupt vector for everything.
2083 iaq->nirq = iaq->nrxq10g = iaq->nrxq1g = 1;
2084 iaq->intr_flags_10g = iaq->intr_flags_1g = 0;
2087 iaq->nofldrxq10g = iaq->nofldrxq1g = 1;
2090 iaq->nnmrxq10g = iaq->nnmrxq1g = 1;
2096 if (itype == INTR_MSIX)
2097 rc = pci_alloc_msix(sc->dev, &navail);
2098 else if (itype == INTR_MSI)
2099 rc = pci_alloc_msi(sc->dev, &navail);
2102 if (navail == iaq->nirq)
2106 * Didn't get the number requested. Use whatever number
2107 * the kernel is willing to allocate (it's in navail).
2109 device_printf(sc->dev, "fewer vectors than requested, "
2110 "type=%d, req=%d, rcvd=%d; will downshift req.\n",
2111 itype, iaq->nirq, navail);
2112 pci_release_msi(sc->dev);
2116 device_printf(sc->dev,
2117 "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
2118 itype, rc, iaq->nirq, navail);
2121 device_printf(sc->dev,
2122 "failed to find a usable interrupt type. "
2123 "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
2124 pci_msix_count(sc->dev), pci_msi_count(sc->dev));
2129 #define FW_VERSION(chip) ( \
2130 V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
2131 V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
2132 V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
2133 V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
2134 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
2140 struct fw_hdr fw_hdr; /* XXX: waste of space, need a sparse struct */
2144 .kld_name = "t4fw_cfg",
2145 .fw_mod_name = "t4fw",
2147 .chip = FW_HDR_CHIP_T4,
2148 .fw_ver = htobe32_const(FW_VERSION(T4)),
2149 .intfver_nic = FW_INTFVER(T4, NIC),
2150 .intfver_vnic = FW_INTFVER(T4, VNIC),
2151 .intfver_ofld = FW_INTFVER(T4, OFLD),
2152 .intfver_ri = FW_INTFVER(T4, RI),
2153 .intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
2154 .intfver_iscsi = FW_INTFVER(T4, ISCSI),
2155 .intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
2156 .intfver_fcoe = FW_INTFVER(T4, FCOE),
2160 .kld_name = "t5fw_cfg",
2161 .fw_mod_name = "t5fw",
2163 .chip = FW_HDR_CHIP_T5,
2164 .fw_ver = htobe32_const(FW_VERSION(T5)),
2165 .intfver_nic = FW_INTFVER(T5, NIC),
2166 .intfver_vnic = FW_INTFVER(T5, VNIC),
2167 .intfver_ofld = FW_INTFVER(T5, OFLD),
2168 .intfver_ri = FW_INTFVER(T5, RI),
2169 .intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
2170 .intfver_iscsi = FW_INTFVER(T5, ISCSI),
2171 .intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
2172 .intfver_fcoe = FW_INTFVER(T5, FCOE),
2177 static struct fw_info *
2178 find_fw_info(int chip)
2182 for (i = 0; i < nitems(fw_info); i++) {
2183 if (fw_info[i].chip == chip)
2184 return (&fw_info[i]);
2190 * Is the given firmware API compatible with the one the driver was compiled
2194 fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
2197 /* short circuit if it's the exact same firmware version */
2198 if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
2202 * XXX: Is this too conservative? Perhaps I should limit this to the
2203 * features that are supported in the driver.
2205 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
2206 if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
2207 SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
2208 SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
2216 * The firmware in the KLD is usable, but should it be installed? This routine
2217 * explains itself in detail if it indicates the KLD firmware should be
2221 should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
2225 if (!card_fw_usable) {
2226 reason = "incompatible or unusable";
2231 reason = "older than the version bundled with this driver";
2235 if (t4_fw_install == 2 && k != c) {
2236 reason = "different than the version bundled with this driver";
2243 if (t4_fw_install == 0) {
2244 device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2245 "but the driver is prohibited from installing a different "
2246 "firmware on the card.\n",
2247 G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2248 G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
2253 device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2254 "installing firmware %u.%u.%u.%u on card.\n",
2255 G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2256 G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
2257 G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
2258 G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
2263 * Establish contact with the firmware and determine if we are the master driver
2264 * or not, and whether we are responsible for chip initialization.
2267 prep_firmware(struct adapter *sc)
2269 const struct firmware *fw = NULL, *default_cfg;
2270 int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
2271 enum dev_state state;
2272 struct fw_info *fw_info;
2273 struct fw_hdr *card_fw; /* fw on the card */
2274 const struct fw_hdr *kld_fw; /* fw in the KLD */
2275 const struct fw_hdr *drv_fw; /* fw header the driver was compiled
2278 /* Contact firmware. */
2279 rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
2280 if (rc < 0 || state == DEV_STATE_ERR) {
2282 device_printf(sc->dev,
2283 "failed to connect to the firmware: %d, %d.\n", rc, state);
2288 sc->flags |= MASTER_PF;
2289 else if (state == DEV_STATE_UNINIT) {
2291 * We didn't get to be the master so we definitely won't be
2292 * configuring the chip. It's a bug if someone else hasn't
2293 * configured it already.
2295 device_printf(sc->dev, "couldn't be master(%d), "
2296 "device not already initialized either(%d).\n", rc, state);
2300 /* This is the firmware whose headers the driver was compiled against */
2301 fw_info = find_fw_info(chip_id(sc));
2302 if (fw_info == NULL) {
2303 device_printf(sc->dev,
2304 "unable to look up firmware information for chip %d.\n",
2308 drv_fw = &fw_info->fw_hdr;
2311 * The firmware KLD contains many modules. The KLD name is also the
2312 * name of the module that contains the default config file.
2314 default_cfg = firmware_get(fw_info->kld_name);
2316 /* Read the header of the firmware on the card */
2317 card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
2318 rc = -t4_read_flash(sc, FLASH_FW_START,
2319 sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
2321 card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
2323 device_printf(sc->dev,
2324 "Unable to read card's firmware header: %d\n", rc);
2328 /* This is the firmware in the KLD */
2329 fw = firmware_get(fw_info->fw_mod_name);
2331 kld_fw = (const void *)fw->data;
2332 kld_fw_usable = fw_compatible(drv_fw, kld_fw);
2338 if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
2339 (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
2341 * Common case: the firmware on the card is an exact match and
2342 * the KLD is an exact match too, or the KLD is
2343 * absent/incompatible. Note that t4_fw_install = 2 is ignored
2344 * here -- use cxgbetool loadfw if you want to reinstall the
2345 * same firmware as the one on the card.
2347 } else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
2348 should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
2349 be32toh(card_fw->fw_ver))) {
2351 rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
2353 device_printf(sc->dev,
2354 "failed to install firmware: %d\n", rc);
2358 /* Installed successfully, update the cached header too. */
2359 memcpy(card_fw, kld_fw, sizeof(*card_fw));
2361 need_fw_reset = 0; /* already reset as part of load_fw */
2364 if (!card_fw_usable) {
2367 d = ntohl(drv_fw->fw_ver);
2368 c = ntohl(card_fw->fw_ver);
2369 k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
2371 device_printf(sc->dev, "Cannot find a usable firmware: "
2372 "fw_install %d, chip state %d, "
2373 "driver compiled with %d.%d.%d.%d, "
2374 "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
2375 t4_fw_install, state,
2376 G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
2377 G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
2378 G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2379 G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
2380 G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
2381 G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
2386 /* We're using whatever's on the card and it's known to be good. */
2387 sc->params.fw_vers = ntohl(card_fw->fw_ver);
2388 snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
2389 G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
2390 G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
2391 G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
2392 G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
2393 t4_get_tp_version(sc, &sc->params.tp_vers);
2396 if (need_fw_reset &&
2397 (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
2398 device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
2399 if (rc != ETIMEDOUT && rc != EIO)
2400 t4_fw_bye(sc, sc->mbox);
2405 rc = get_params__pre_init(sc);
2407 goto done; /* error message displayed already */
2409 /* Partition adapter resources as specified in the config file. */
2410 if (state == DEV_STATE_UNINIT) {
2412 KASSERT(sc->flags & MASTER_PF,
2413 ("%s: trying to change chip settings when not master.",
2416 rc = partition_resources(sc, default_cfg, fw_info->kld_name);
2418 goto done; /* error message displayed already */
2420 t4_tweak_chip_settings(sc);
2422 /* get basic stuff going */
2423 rc = -t4_fw_initialize(sc, sc->mbox);
2425 device_printf(sc->dev, "fw init failed: %d.\n", rc);
2429 snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
2434 free(card_fw, M_CXGBE);
2436 firmware_put(fw, FIRMWARE_UNLOAD);
2437 if (default_cfg != NULL)
2438 firmware_put(default_cfg, FIRMWARE_UNLOAD);
2443 #define FW_PARAM_DEV(param) \
2444 (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
2445 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
2446 #define FW_PARAM_PFVF(param) \
2447 (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
2448 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
2451 * Partition chip resources for use between various PFs, VFs, etc.
2454 partition_resources(struct adapter *sc, const struct firmware *default_cfg,
2455 const char *name_prefix)
2457 const struct firmware *cfg = NULL;
2459 struct fw_caps_config_cmd caps;
2460 uint32_t mtype, moff, finicsum, cfcsum;
2463 * Figure out what configuration file to use. Pick the default config
2464 * file for the card if the user hasn't specified one explicitly.
2466 snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
2467 if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
2468 /* Card specific overrides go here. */
2469 if (pci_get_device(sc->dev) == 0x440a)
2470 snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
2472 snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
2476 * We need to load another module if the profile is anything except
2477 * "default" or "flash".
2479 if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
2480 strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
2483 snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
2484 cfg = firmware_get(s);
2486 if (default_cfg != NULL) {
2487 device_printf(sc->dev,
2488 "unable to load module \"%s\" for "
2489 "configuration profile \"%s\", will use "
2490 "the default config file instead.\n",
2492 snprintf(sc->cfg_file, sizeof(sc->cfg_file),
2495 device_printf(sc->dev,
2496 "unable to load module \"%s\" for "
2497 "configuration profile \"%s\", will use "
2498 "the config file on the card's flash "
2499 "instead.\n", s, sc->cfg_file);
2500 snprintf(sc->cfg_file, sizeof(sc->cfg_file),
2506 if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
2507 default_cfg == NULL) {
2508 device_printf(sc->dev,
2509 "default config file not available, will use the config "
2510 "file on the card's flash instead.\n");
2511 snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
2514 if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
2516 const uint32_t *cfdata;
2517 uint32_t param, val, addr, off, mw_base, mw_aperture;
2519 KASSERT(cfg != NULL || default_cfg != NULL,
2520 ("%s: no config to upload", __func__));
2523 * Ask the firmware where it wants us to upload the config file.
2525 param = FW_PARAM_DEV(CF);
2526 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, ¶m, &val);
2528 /* No support for config file? Shouldn't happen. */
2529 device_printf(sc->dev,
2530 "failed to query config file location: %d.\n", rc);
2533 mtype = G_FW_PARAMS_PARAM_Y(val);
2534 moff = G_FW_PARAMS_PARAM_Z(val) << 16;
2537 * XXX: sheer laziness. We deliberately added 4 bytes of
2538 * useless stuffing/comments at the end of the config file so
2539 * it's ok to simply throw away the last remaining bytes when
2540 * the config file is not an exact multiple of 4. This also
2541 * helps with the validate_mt_off_len check.
2544 cflen = cfg->datasize & ~3;
2547 cflen = default_cfg->datasize & ~3;
2548 cfdata = default_cfg->data;
2551 if (cflen > FLASH_CFG_MAX_SIZE) {
2552 device_printf(sc->dev,
2553 "config file too long (%d, max allowed is %d). "
2554 "Will try to use the config on the card, if any.\n",
2555 cflen, FLASH_CFG_MAX_SIZE);
2556 goto use_config_on_flash;
2559 rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
2561 device_printf(sc->dev,
2562 "%s: addr (%d/0x%x) or len %d is not valid: %d. "
2563 "Will try to use the config on the card, if any.\n",
2564 __func__, mtype, moff, cflen, rc);
2565 goto use_config_on_flash;
2568 memwin_info(sc, 2, &mw_base, &mw_aperture);
2570 off = position_memwin(sc, 2, addr);
2571 n = min(cflen, mw_aperture - off);
2572 for (i = 0; i < n; i += 4)
2573 t4_write_reg(sc, mw_base + off + i, *cfdata++);
2578 use_config_on_flash:
2579 mtype = FW_MEMTYPE_FLASH;
2580 moff = t4_flash_cfg_addr(sc);
2583 bzero(&caps, sizeof(caps));
2584 caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2585 F_FW_CMD_REQUEST | F_FW_CMD_READ);
2586 caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
2587 V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
2588 V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
2589 rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
2591 device_printf(sc->dev,
2592 "failed to pre-process config file: %d "
2593 "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
2597 finicsum = be32toh(caps.finicsum);
2598 cfcsum = be32toh(caps.cfcsum);
2599 if (finicsum != cfcsum) {
2600 device_printf(sc->dev,
2601 "WARNING: config file checksum mismatch: %08x %08x\n",
2604 sc->cfcsum = cfcsum;
2606 #define LIMIT_CAPS(x) do { \
2607 caps.x &= htobe16(t4_##x##_allowed); \
2611 * Let the firmware know what features will (not) be used so it can tune
2612 * things accordingly.
2614 LIMIT_CAPS(linkcaps);
2615 LIMIT_CAPS(niccaps);
2616 LIMIT_CAPS(toecaps);
2617 LIMIT_CAPS(rdmacaps);
2618 LIMIT_CAPS(iscsicaps);
2619 LIMIT_CAPS(fcoecaps);
2622 caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2623 F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
2624 caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
2625 rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
2627 device_printf(sc->dev,
2628 "failed to process config file: %d.\n", rc);
2632 firmware_put(cfg, FIRMWARE_UNLOAD);
2637 * Retrieve parameters that are needed (or nice to have) very early.
2640 get_params__pre_init(struct adapter *sc)
2643 uint32_t param[2], val[2];
2644 struct fw_devlog_cmd cmd;
2645 struct devlog_params *dlog = &sc->params.devlog;
2647 param[0] = FW_PARAM_DEV(PORTVEC);
2648 param[1] = FW_PARAM_DEV(CCLK);
2649 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
2651 device_printf(sc->dev,
2652 "failed to query parameters (pre_init): %d.\n", rc);
2656 sc->params.portvec = val[0];
2657 sc->params.nports = bitcount32(val[0]);
2658 sc->params.vpd.cclk = val[1];
2660 /* Read device log parameters. */
2661 bzero(&cmd, sizeof(cmd));
2662 cmd.op_to_write = htobe32(V_FW_CMD_OP(FW_DEVLOG_CMD) |
2663 F_FW_CMD_REQUEST | F_FW_CMD_READ);
2664 cmd.retval_len16 = htobe32(FW_LEN16(cmd));
2665 rc = -t4_wr_mbox(sc, sc->mbox, &cmd, sizeof(cmd), &cmd);
2667 device_printf(sc->dev,
2668 "failed to get devlog parameters: %d.\n", rc);
2669 bzero(dlog, sizeof (*dlog));
2670 rc = 0; /* devlog isn't critical for device operation */
2672 val[0] = be32toh(cmd.memtype_devlog_memaddr16_devlog);
2673 dlog->memtype = G_FW_DEVLOG_CMD_MEMTYPE_DEVLOG(val[0]);
2674 dlog->start = G_FW_DEVLOG_CMD_MEMADDR16_DEVLOG(val[0]) << 4;
2675 dlog->size = be32toh(cmd.memsize_devlog);
2682 * Retrieve various parameters that are of interest to the driver. The device
2683 * has been initialized by the firmware at this point.
2686 get_params__post_init(struct adapter *sc)
2689 uint32_t param[7], val[7];
2690 struct fw_caps_config_cmd caps;
2692 param[0] = FW_PARAM_PFVF(IQFLINT_START);
2693 param[1] = FW_PARAM_PFVF(EQ_START);
2694 param[2] = FW_PARAM_PFVF(FILTER_START);
2695 param[3] = FW_PARAM_PFVF(FILTER_END);
2696 param[4] = FW_PARAM_PFVF(L2T_START);
2697 param[5] = FW_PARAM_PFVF(L2T_END);
2698 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
2700 device_printf(sc->dev,
2701 "failed to query parameters (post_init): %d.\n", rc);
2705 sc->sge.iq_start = val[0];
2706 sc->sge.eq_start = val[1];
2707 sc->tids.ftid_base = val[2];
2708 sc->tids.nftids = val[3] - val[2] + 1;
2709 sc->params.ftid_min = val[2];
2710 sc->params.ftid_max = val[3];
2711 sc->vres.l2t.start = val[4];
2712 sc->vres.l2t.size = val[5] - val[4] + 1;
2713 KASSERT(sc->vres.l2t.size <= L2T_SIZE,
2714 ("%s: L2 table size (%u) larger than expected (%u)",
2715 __func__, sc->vres.l2t.size, L2T_SIZE));
2717 /* get capabilites */
2718 bzero(&caps, sizeof(caps));
2719 caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2720 F_FW_CMD_REQUEST | F_FW_CMD_READ);
2721 caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
2722 rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
2724 device_printf(sc->dev,
2725 "failed to get card capabilities: %d.\n", rc);
2729 #define READ_CAPS(x) do { \
2730 sc->x = htobe16(caps.x); \
2732 READ_CAPS(linkcaps);
2735 READ_CAPS(rdmacaps);
2736 READ_CAPS(iscsicaps);
2737 READ_CAPS(fcoecaps);
2739 if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
2740 param[0] = FW_PARAM_PFVF(ETHOFLD_START);
2741 param[1] = FW_PARAM_PFVF(ETHOFLD_END);
2742 param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
2743 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
2745 device_printf(sc->dev,
2746 "failed to query NIC parameters: %d.\n", rc);
2749 sc->tids.etid_base = val[0];
2750 sc->params.etid_min = val[0];
2751 sc->tids.netids = val[1] - val[0] + 1;
2752 sc->params.netids = sc->tids.netids;
2753 sc->params.eo_wr_cred = val[2];
2754 sc->params.ethoffload = 1;
2758 /* query offload-related parameters */
2759 param[0] = FW_PARAM_DEV(NTID);
2760 param[1] = FW_PARAM_PFVF(SERVER_START);
2761 param[2] = FW_PARAM_PFVF(SERVER_END);
2762 param[3] = FW_PARAM_PFVF(TDDP_START);
2763 param[4] = FW_PARAM_PFVF(TDDP_END);
2764 param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
2765 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
2767 device_printf(sc->dev,
2768 "failed to query TOE parameters: %d.\n", rc);
2771 sc->tids.ntids = val[0];
2772 sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
2773 sc->tids.stid_base = val[1];
2774 sc->tids.nstids = val[2] - val[1] + 1;
2775 sc->vres.ddp.start = val[3];
2776 sc->vres.ddp.size = val[4] - val[3] + 1;
2777 sc->params.ofldq_wr_cred = val[5];
2778 sc->params.offload = 1;
2781 param[0] = FW_PARAM_PFVF(STAG_START);
2782 param[1] = FW_PARAM_PFVF(STAG_END);
2783 param[2] = FW_PARAM_PFVF(RQ_START);
2784 param[3] = FW_PARAM_PFVF(RQ_END);
2785 param[4] = FW_PARAM_PFVF(PBL_START);
2786 param[5] = FW_PARAM_PFVF(PBL_END);
2787 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
2789 device_printf(sc->dev,
2790 "failed to query RDMA parameters(1): %d.\n", rc);
2793 sc->vres.stag.start = val[0];
2794 sc->vres.stag.size = val[1] - val[0] + 1;
2795 sc->vres.rq.start = val[2];
2796 sc->vres.rq.size = val[3] - val[2] + 1;
2797 sc->vres.pbl.start = val[4];
2798 sc->vres.pbl.size = val[5] - val[4] + 1;
2800 param[0] = FW_PARAM_PFVF(SQRQ_START);
2801 param[1] = FW_PARAM_PFVF(SQRQ_END);
2802 param[2] = FW_PARAM_PFVF(CQ_START);
2803 param[3] = FW_PARAM_PFVF(CQ_END);
2804 param[4] = FW_PARAM_PFVF(OCQ_START);
2805 param[5] = FW_PARAM_PFVF(OCQ_END);
2806 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
2808 device_printf(sc->dev,
2809 "failed to query RDMA parameters(2): %d.\n", rc);
2812 sc->vres.qp.start = val[0];
2813 sc->vres.qp.size = val[1] - val[0] + 1;
2814 sc->vres.cq.start = val[2];
2815 sc->vres.cq.size = val[3] - val[2] + 1;
2816 sc->vres.ocq.start = val[4];
2817 sc->vres.ocq.size = val[5] - val[4] + 1;
2819 if (sc->iscsicaps) {
2820 param[0] = FW_PARAM_PFVF(ISCSI_START);
2821 param[1] = FW_PARAM_PFVF(ISCSI_END);
2822 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
2824 device_printf(sc->dev,
2825 "failed to query iSCSI parameters: %d.\n", rc);
2828 sc->vres.iscsi.start = val[0];
2829 sc->vres.iscsi.size = val[1] - val[0] + 1;
2833 * We've got the params we wanted to query via the firmware. Now grab
2834 * some others directly from the chip.
2836 rc = t4_read_chip_settings(sc);
2842 set_params__post_init(struct adapter *sc)
2844 uint32_t param, val;
2846 /* ask for encapsulated CPLs */
2847 param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
2849 (void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, ¶m, &val);
2854 #undef FW_PARAM_PFVF
2858 t4_set_desc(struct adapter *sc)
2861 struct adapter_params *p = &sc->params;
2863 snprintf(buf, sizeof(buf), "Chelsio %s %sNIC (rev %d), S/N:%s, "
2864 "P/N:%s, E/C:%s", p->vpd.id, is_offload(sc) ? "R" : "",
2865 chip_rev(sc), p->vpd.sn, p->vpd.pn, p->vpd.ec);
2867 device_set_desc_copy(sc->dev, buf);
2871 build_medialist(struct port_info *pi, struct ifmedia *media)
2877 ifmedia_removeall(media);
2879 m = IFM_ETHER | IFM_FDX;
2881 switch(pi->port_type) {
2882 case FW_PORT_TYPE_BT_XFI:
2883 case FW_PORT_TYPE_BT_XAUI:
2884 ifmedia_add(media, m | IFM_10G_T, 0, NULL);
2887 case FW_PORT_TYPE_BT_SGMII:
2888 ifmedia_add(media, m | IFM_1000_T, 0, NULL);
2889 ifmedia_add(media, m | IFM_100_TX, 0, NULL);
2890 ifmedia_add(media, IFM_ETHER | IFM_AUTO, 0, NULL);
2891 ifmedia_set(media, IFM_ETHER | IFM_AUTO);
2894 case FW_PORT_TYPE_CX4:
2895 ifmedia_add(media, m | IFM_10G_CX4, 0, NULL);
2896 ifmedia_set(media, m | IFM_10G_CX4);
2899 case FW_PORT_TYPE_QSFP_10G:
2900 case FW_PORT_TYPE_SFP:
2901 case FW_PORT_TYPE_FIBER_XFI:
2902 case FW_PORT_TYPE_FIBER_XAUI:
2903 switch (pi->mod_type) {
2905 case FW_PORT_MOD_TYPE_LR:
2906 ifmedia_add(media, m | IFM_10G_LR, 0, NULL);
2907 ifmedia_set(media, m | IFM_10G_LR);
2910 case FW_PORT_MOD_TYPE_SR:
2911 ifmedia_add(media, m | IFM_10G_SR, 0, NULL);
2912 ifmedia_set(media, m | IFM_10G_SR);
2915 case FW_PORT_MOD_TYPE_LRM:
2916 ifmedia_add(media, m | IFM_10G_LRM, 0, NULL);
2917 ifmedia_set(media, m | IFM_10G_LRM);
2920 case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
2921 case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
2922 ifmedia_add(media, m | IFM_10G_TWINAX, 0, NULL);
2923 ifmedia_set(media, m | IFM_10G_TWINAX);
2926 case FW_PORT_MOD_TYPE_NONE:
2928 ifmedia_add(media, m | IFM_NONE, 0, NULL);
2929 ifmedia_set(media, m | IFM_NONE);
2932 case FW_PORT_MOD_TYPE_NA:
2933 case FW_PORT_MOD_TYPE_ER:
2935 device_printf(pi->dev,
2936 "unknown port_type (%d), mod_type (%d)\n",
2937 pi->port_type, pi->mod_type);
2938 ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
2939 ifmedia_set(media, m | IFM_UNKNOWN);
2944 case FW_PORT_TYPE_QSFP:
2945 switch (pi->mod_type) {
2947 case FW_PORT_MOD_TYPE_LR:
2948 ifmedia_add(media, m | IFM_40G_LR4, 0, NULL);
2949 ifmedia_set(media, m | IFM_40G_LR4);
2952 case FW_PORT_MOD_TYPE_SR:
2953 ifmedia_add(media, m | IFM_40G_SR4, 0, NULL);
2954 ifmedia_set(media, m | IFM_40G_SR4);
2957 case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
2958 case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
2959 ifmedia_add(media, m | IFM_40G_CR4, 0, NULL);
2960 ifmedia_set(media, m | IFM_40G_CR4);
2963 case FW_PORT_MOD_TYPE_NONE:
2965 ifmedia_add(media, m | IFM_NONE, 0, NULL);
2966 ifmedia_set(media, m | IFM_NONE);
2970 device_printf(pi->dev,
2971 "unknown port_type (%d), mod_type (%d)\n",
2972 pi->port_type, pi->mod_type);
2973 ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
2974 ifmedia_set(media, m | IFM_UNKNOWN);
2980 device_printf(pi->dev,
2981 "unknown port_type (%d), mod_type (%d)\n", pi->port_type,
2983 ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
2984 ifmedia_set(media, m | IFM_UNKNOWN);
2991 #define FW_MAC_EXACT_CHUNK 7
2994 * Program the port's XGMAC based on parameters in ifnet. The caller also
2995 * indicates which parameters should be programmed (the rest are left alone).
2998 update_mac_settings(struct ifnet *ifp, int flags)
3001 struct port_info *pi = ifp->if_softc;
3002 struct adapter *sc = pi->adapter;
3003 int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
3004 uint16_t viid = 0xffff;
3005 int16_t *xact_addr_filt = NULL;
3007 ASSERT_SYNCHRONIZED_OP(sc);
3008 KASSERT(flags, ("%s: not told what to update.", __func__));
3010 if (ifp == pi->ifp) {
3012 xact_addr_filt = &pi->xact_addr_filt;
3015 else if (ifp == pi->nm_ifp) {
3017 xact_addr_filt = &pi->nm_xact_addr_filt;
3020 if (flags & XGMAC_MTU)
3023 if (flags & XGMAC_PROMISC)
3024 promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
3026 if (flags & XGMAC_ALLMULTI)
3027 allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
3029 if (flags & XGMAC_VLANEX)
3030 vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
3032 if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
3033 rc = -t4_set_rxmode(sc, sc->mbox, viid, mtu, promisc, allmulti,
3036 if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
3042 if (flags & XGMAC_UCADDR) {
3043 uint8_t ucaddr[ETHER_ADDR_LEN];
3045 bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
3046 rc = t4_change_mac(sc, sc->mbox, viid, *xact_addr_filt, ucaddr,
3050 if_printf(ifp, "change_mac failed: %d\n", rc);
3053 *xact_addr_filt = rc;
3058 if (flags & XGMAC_MCADDRS) {
3059 const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
3062 struct ifmultiaddr *ifma;
3065 if_maddr_rlock(ifp);
3066 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3067 if (ifma->ifma_addr->sa_family != AF_LINK)
3070 LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
3071 MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
3074 if (i == FW_MAC_EXACT_CHUNK) {
3075 rc = t4_alloc_mac_filt(sc, sc->mbox, viid, del,
3076 i, mcaddr, NULL, &hash, 0);
3079 for (j = 0; j < i; j++) {
3081 "failed to add mc address"
3083 "%02x:%02x:%02x rc=%d\n",
3084 mcaddr[j][0], mcaddr[j][1],
3085 mcaddr[j][2], mcaddr[j][3],
3086 mcaddr[j][4], mcaddr[j][5],
3096 rc = t4_alloc_mac_filt(sc, sc->mbox, viid, del, i,
3097 mcaddr, NULL, &hash, 0);
3100 for (j = 0; j < i; j++) {
3102 "failed to add mc address"
3104 "%02x:%02x:%02x rc=%d\n",
3105 mcaddr[j][0], mcaddr[j][1],
3106 mcaddr[j][2], mcaddr[j][3],
3107 mcaddr[j][4], mcaddr[j][5],
3114 rc = -t4_set_addr_hash(sc, sc->mbox, viid, 0, hash, 0);
3116 if_printf(ifp, "failed to set mc address hash: %d", rc);
3118 if_maddr_runlock(ifp);
3125 * {begin|end}_synchronized_op must be called from the same thread.
3128 begin_synchronized_op(struct adapter *sc, struct port_info *pi, int flags,
3134 /* the caller thinks it's ok to sleep, but is it really? */
3135 if (flags & SLEEP_OK)
3136 pause("t4slptst", 1);
3147 if (pi && IS_DOOMED(pi)) {
3157 if (!(flags & SLEEP_OK)) {
3162 if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
3168 KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
3171 sc->last_op = wmesg;
3172 sc->last_op_thr = curthread;
3173 sc->last_op_flags = flags;
3177 if (!(flags & HOLD_LOCK) || rc)
3184 * {begin|end}_synchronized_op must be called from the same thread.
3187 end_synchronized_op(struct adapter *sc, int flags)
3190 if (flags & LOCK_HELD)
3191 ADAPTER_LOCK_ASSERT_OWNED(sc);
3195 KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
3202 cxgbe_init_synchronized(struct port_info *pi)
3204 struct adapter *sc = pi->adapter;
3205 struct ifnet *ifp = pi->ifp;
3207 struct sge_txq *txq;
3209 ASSERT_SYNCHRONIZED_OP(sc);
3211 if (isset(&sc->open_device_map, pi->port_id)) {
3212 KASSERT(ifp->if_drv_flags & IFF_DRV_RUNNING,
3213 ("mismatch between open_device_map and if_drv_flags"));
3214 return (0); /* already running */
3217 if (!(sc->flags & FULL_INIT_DONE) &&
3218 ((rc = adapter_full_init(sc)) != 0))
3219 return (rc); /* error message displayed already */
3221 if (!(pi->flags & PORT_INIT_DONE) &&
3222 ((rc = port_full_init(pi)) != 0))
3223 return (rc); /* error message displayed already */
3225 rc = update_mac_settings(ifp, XGMAC_ALL);
3227 goto done; /* error message displayed already */
3229 rc = -t4_enable_vi(sc, sc->mbox, pi->viid, true, true);
3231 if_printf(ifp, "enable_vi failed: %d\n", rc);
3236 * Can't fail from this point onwards. Review cxgbe_uninit_synchronized
3240 for_each_txq(pi, i, txq) {
3242 txq->eq.flags |= EQ_ENABLED;
3247 * The first iq of the first port to come up is used for tracing.
3249 if (sc->traceq < 0) {
3250 sc->traceq = sc->sge.rxq[pi->first_rxq].iq.abs_id;
3251 t4_write_reg(sc, is_t4(sc) ? A_MPS_TRC_RSS_CONTROL :
3252 A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
3253 V_QUEUENUMBER(sc->traceq));
3254 pi->flags |= HAS_TRACEQ;
3258 setbit(&sc->open_device_map, pi->port_id);
3260 ifp->if_drv_flags |= IFF_DRV_RUNNING;
3263 callout_reset(&pi->tick, hz, cxgbe_tick, pi);
3266 cxgbe_uninit_synchronized(pi);
3275 cxgbe_uninit_synchronized(struct port_info *pi)
3277 struct adapter *sc = pi->adapter;
3278 struct ifnet *ifp = pi->ifp;
3280 struct sge_txq *txq;
3282 ASSERT_SYNCHRONIZED_OP(sc);
3284 if (!(pi->flags & PORT_INIT_DONE)) {
3285 KASSERT(!(ifp->if_drv_flags & IFF_DRV_RUNNING),
3286 ("uninited port is running"));
3291 * Disable the VI so that all its data in either direction is discarded
3292 * by the MPS. Leave everything else (the queues, interrupts, and 1Hz
3293 * tick) intact as the TP can deliver negative advice or data that it's
3294 * holding in its RAM (for an offloaded connection) even after the VI is
3297 rc = -t4_enable_vi(sc, sc->mbox, pi->viid, false, false);
3299 if_printf(ifp, "disable_vi failed: %d\n", rc);
3303 for_each_txq(pi, i, txq) {
3305 txq->eq.flags &= ~EQ_ENABLED;
3309 clrbit(&sc->open_device_map, pi->port_id);
3311 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3314 pi->link_cfg.link_ok = 0;
3315 pi->link_cfg.speed = 0;
3317 t4_os_link_changed(sc, pi->port_id, 0, -1);
3323 * It is ok for this function to fail midway and return right away. t4_detach
3324 * will walk the entire sc->irq list and clean up whatever is valid.
3327 setup_intr_handlers(struct adapter *sc)
3332 struct port_info *pi;
3333 struct sge_rxq *rxq;
3335 struct sge_ofld_rxq *ofld_rxq;
3338 struct sge_nm_rxq *nm_rxq;
3345 rid = sc->intr_type == INTR_INTX ? 0 : 1;
3346 if (sc->intr_count == 1)
3347 return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
3349 /* Multiple interrupts. */
3350 KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
3351 ("%s: too few intr.", __func__));
3353 /* The first one is always error intr */
3354 rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
3360 /* The second one is always the firmware event queue */
3361 rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sc->sge.fwq, "evt");
3367 for_each_port(sc, p) {
3370 if (pi->flags & INTR_RXQ) {
3371 for_each_rxq(pi, q, rxq) {
3372 snprintf(s, sizeof(s), "%d.%d", p, q);
3373 rc = t4_alloc_irq(sc, irq, rid, t4_intr, rxq,
3382 if (pi->flags & INTR_OFLD_RXQ) {
3383 for_each_ofld_rxq(pi, q, ofld_rxq) {
3384 snprintf(s, sizeof(s), "%d,%d", p, q);
3385 rc = t4_alloc_irq(sc, irq, rid, t4_intr,
3395 if (pi->flags & INTR_NM_RXQ) {
3396 for_each_nm_rxq(pi, q, nm_rxq) {
3397 snprintf(s, sizeof(s), "%d-%d", p, q);
3398 rc = t4_alloc_irq(sc, irq, rid, t4_nm_intr,
3408 MPASS(irq == &sc->irq[sc->intr_count]);
3414 adapter_full_init(struct adapter *sc)
3418 ASSERT_SYNCHRONIZED_OP(sc);
3419 ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
3420 KASSERT((sc->flags & FULL_INIT_DONE) == 0,
3421 ("%s: FULL_INIT_DONE already", __func__));
3424 * queues that belong to the adapter (not any particular port).
3426 rc = t4_setup_adapter_queues(sc);
3430 for (i = 0; i < nitems(sc->tq); i++) {
3431 sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
3432 taskqueue_thread_enqueue, &sc->tq[i]);
3433 if (sc->tq[i] == NULL) {
3434 device_printf(sc->dev,
3435 "failed to allocate task queue %d\n", i);
3439 taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
3440 device_get_nameunit(sc->dev), i);
3444 sc->flags |= FULL_INIT_DONE;
3447 adapter_full_uninit(sc);
3453 adapter_full_uninit(struct adapter *sc)
3457 ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
3459 t4_teardown_adapter_queues(sc);
3461 for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
3462 taskqueue_free(sc->tq[i]);
3466 sc->flags &= ~FULL_INIT_DONE;
3472 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
3473 RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
3474 RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
3475 RSS_HASHTYPE_RSS_UDP_IPV6)
3477 /* Translates kernel hash types to hardware. */
3479 hashconfig_to_hashen(int hashconfig)
3483 if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
3484 hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
3485 if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
3486 hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
3487 if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
3488 hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
3489 F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
3491 if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
3492 hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
3493 F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
3495 if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
3496 hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
3497 if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
3498 hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
3503 /* Translates hardware hash types to kernel. */
3505 hashen_to_hashconfig(int hashen)
3509 if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
3511 * If UDP hashing was enabled it must have been enabled for
3512 * either IPv4 or IPv6 (inclusive or). Enabling UDP without
3513 * enabling any 4-tuple hash is nonsense configuration.
3515 MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
3516 F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
3518 if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
3519 hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
3520 if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
3521 hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
3523 if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
3524 hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
3525 if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
3526 hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
3527 if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
3528 hashconfig |= RSS_HASHTYPE_RSS_IPV4;
3529 if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
3530 hashconfig |= RSS_HASHTYPE_RSS_IPV6;
3532 return (hashconfig);
3537 port_full_init(struct port_info *pi)
3539 struct adapter *sc = pi->adapter;
3540 struct ifnet *ifp = pi->ifp;
3542 struct sge_rxq *rxq;
3543 int rc, i, j, hashen;
3545 int nbuckets = rss_getnumbuckets();
3546 int hashconfig = rss_gethashconfig();
3548 uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
3549 uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
3552 ASSERT_SYNCHRONIZED_OP(sc);
3553 KASSERT((pi->flags & PORT_INIT_DONE) == 0,
3554 ("%s: PORT_INIT_DONE already", __func__));
3556 sysctl_ctx_init(&pi->ctx);
3557 pi->flags |= PORT_SYSCTL_CTX;
3560 * Allocate tx/rx/fl queues for this port.
3562 rc = t4_setup_port_queues(pi);
3564 goto done; /* error message displayed already */
3567 * Setup RSS for this port. Save a copy of the RSS table for later use.
3569 if (pi->nrxq > pi->rss_size) {
3570 if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
3571 "some queues will never receive traffic.\n", pi->nrxq,
3573 } else if (pi->rss_size % pi->nrxq) {
3574 if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
3575 "expect uneven traffic distribution.\n", pi->nrxq,
3579 MPASS(RSS_KEYSIZE == 40);
3580 if (pi->nrxq != nbuckets) {
3581 if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
3582 "performance will be impacted.\n", pi->nrxq, nbuckets);
3585 rss_getkey((void *)&raw_rss_key[0]);
3586 for (i = 0; i < nitems(rss_key); i++) {
3587 rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
3589 t4_write_rss_key(sc, (void *)&rss_key[0], -1);
3591 rss = malloc(pi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
3592 for (i = 0; i < pi->rss_size;) {
3594 j = rss_get_indirection_to_bucket(i);
3596 rxq = &sc->sge.rxq[pi->first_rxq + j];
3597 rss[i++] = rxq->iq.abs_id;
3599 for_each_rxq(pi, j, rxq) {
3600 rss[i++] = rxq->iq.abs_id;
3601 if (i == pi->rss_size)
3607 rc = -t4_config_rss_range(sc, sc->mbox, pi->viid, 0, pi->rss_size, rss,
3610 if_printf(ifp, "rss_config failed: %d\n", rc);
3615 hashen = hashconfig_to_hashen(hashconfig);
3618 * We may have had to enable some hashes even though the global config
3619 * wants them disabled. This is a potential problem that must be
3620 * reported to the user.
3622 extra = hashen_to_hashconfig(hashen) ^ hashconfig;
3625 * If we consider only the supported hash types, then the enabled hashes
3626 * are a superset of the requested hashes. In other words, there cannot
3627 * be any supported hash that was requested but not enabled, but there
3628 * can be hashes that were not requested but had to be enabled.
3630 extra &= SUPPORTED_RSS_HASHTYPES;
3631 MPASS((extra & hashconfig) == 0);
3635 "global RSS config (0x%x) cannot be accomodated.\n",
3638 if (extra & RSS_HASHTYPE_RSS_IPV4)
3639 if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
3640 if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
3641 if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
3642 if (extra & RSS_HASHTYPE_RSS_IPV6)
3643 if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
3644 if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
3645 if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
3646 if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
3647 if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
3648 if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
3649 if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
3651 hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
3652 F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
3653 F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
3654 F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
3656 rc = -t4_config_vi_rss(sc, sc->mbox, pi->viid, hashen, rss[0]);
3658 if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
3663 pi->flags |= PORT_INIT_DONE;
3666 port_full_uninit(pi);
3675 port_full_uninit(struct port_info *pi)
3677 struct adapter *sc = pi->adapter;
3679 struct sge_rxq *rxq;
3680 struct sge_txq *txq;
3682 struct sge_ofld_rxq *ofld_rxq;
3683 struct sge_wrq *ofld_txq;
3686 if (pi->flags & PORT_INIT_DONE) {
3688 /* Need to quiesce queues. */
3690 quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
3692 for_each_txq(pi, i, txq) {
3693 quiesce_txq(sc, txq);
3697 for_each_ofld_txq(pi, i, ofld_txq) {
3698 quiesce_wrq(sc, ofld_txq);
3702 for_each_rxq(pi, i, rxq) {
3703 quiesce_iq(sc, &rxq->iq);
3704 quiesce_fl(sc, &rxq->fl);
3708 for_each_ofld_rxq(pi, i, ofld_rxq) {
3709 quiesce_iq(sc, &ofld_rxq->iq);
3710 quiesce_fl(sc, &ofld_rxq->fl);
3713 free(pi->rss, M_CXGBE);
3716 t4_teardown_port_queues(pi);
3717 pi->flags &= ~PORT_INIT_DONE;
3723 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
3725 struct sge_eq *eq = &txq->eq;
3726 struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
3728 (void) sc; /* unused */
3732 MPASS((eq->flags & EQ_ENABLED) == 0);
3736 /* Wait for the mp_ring to empty. */
3737 while (!mp_ring_is_idle(txq->r)) {
3738 mp_ring_check_drainage(txq->r, 0);
3739 pause("rquiesce", 1);
3742 /* Then wait for the hardware to finish. */
3743 while (spg->cidx != htobe16(eq->pidx))
3744 pause("equiesce", 1);
3746 /* Finally, wait for the driver to reclaim all descriptors. */
3747 while (eq->cidx != eq->pidx)
3748 pause("dquiesce", 1);
3752 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
3759 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
3761 (void) sc; /* unused */
3763 /* Synchronize with the interrupt handler */
3764 while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
3769 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
3771 mtx_lock(&sc->sfl_lock);
3773 fl->flags |= FL_DOOMED;
3775 mtx_unlock(&sc->sfl_lock);
3777 callout_drain(&sc->sfl_callout);
3778 KASSERT((fl->flags & FL_STARVING) == 0,
3779 ("%s: still starving", __func__));
3783 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
3784 driver_intr_t *handler, void *arg, char *name)
3789 irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
3790 RF_SHAREABLE | RF_ACTIVE);
3791 if (irq->res == NULL) {
3792 device_printf(sc->dev,
3793 "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
3797 rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
3798 NULL, handler, arg, &irq->tag);
3800 device_printf(sc->dev,
3801 "failed to setup interrupt for rid %d, name %s: %d\n",
3804 bus_describe_intr(sc->dev, irq->res, irq->tag, name);
3810 t4_free_irq(struct adapter *sc, struct irq *irq)
3813 bus_teardown_intr(sc->dev, irq->res, irq->tag);
3815 bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
3817 bzero(irq, sizeof(*irq));
3823 reg_block_dump(struct adapter *sc, uint8_t *buf, unsigned int start,
3826 uint32_t *p = (uint32_t *)(buf + start);
3828 for ( ; start <= end; start += sizeof(uint32_t))
3829 *p++ = t4_read_reg(sc, start);
3833 t4_get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
3836 const unsigned int *reg_ranges;
3837 static const unsigned int t4_reg_ranges[] = {
4057 static const unsigned int t5_reg_ranges[] = {
4498 reg_ranges = &t4_reg_ranges[0];
4499 n = nitems(t4_reg_ranges);
4501 reg_ranges = &t5_reg_ranges[0];
4502 n = nitems(t5_reg_ranges);
4505 regs->version = chip_id(sc) | chip_rev(sc) << 10;
4506 for (i = 0; i < n; i += 2)
4507 reg_block_dump(sc, buf, reg_ranges[i], reg_ranges[i + 1]);
4511 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
4514 u_int v, tnl_cong_drops;
4516 const struct timeval interval = {0, 250000}; /* 250ms */
4519 timevalsub(&tv, &interval);
4520 if (timevalcmp(&tv, &pi->last_refreshed, <))
4524 t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
4525 for (i = 0; i < NCHAN; i++) {
4526 if (pi->rx_chan_map & (1 << i)) {
4527 mtx_lock(&sc->regwin_lock);
4528 t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
4529 1, A_TP_MIB_TNL_CNG_DROP_0 + i);
4530 mtx_unlock(&sc->regwin_lock);
4531 tnl_cong_drops += v;
4534 pi->tnl_cong_drops = tnl_cong_drops;
4535 getmicrotime(&pi->last_refreshed);
4539 cxgbe_tick(void *arg)
4541 struct port_info *pi = arg;
4542 struct adapter *sc = pi->adapter;
4543 struct ifnet *ifp = pi->ifp;
4546 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4548 return; /* without scheduling another callout */
4551 cxgbe_refresh_stats(sc, pi);
4553 callout_schedule(&pi->tick, hz);
4558 cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
4562 if (arg != ifp || ifp->if_type != IFT_ETHER)
4565 vlan = VLAN_DEVAT(ifp, vid);
4566 VLAN_SETCOOKIE(vlan, ifp);
4570 cpl_not_handled(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
4574 panic("%s: opcode 0x%02x on iq %p with payload %p",
4575 __func__, rss->opcode, iq, m);
4577 log(LOG_ERR, "%s: opcode 0x%02x on iq %p with payload %p\n",
4578 __func__, rss->opcode, iq, m);
4585 t4_register_cpl_handler(struct adapter *sc, int opcode, cpl_handler_t h)
4587 uintptr_t *loc, new;
4589 if (opcode >= nitems(sc->cpl_handler))
4592 new = h ? (uintptr_t)h : (uintptr_t)cpl_not_handled;
4593 loc = (uintptr_t *) &sc->cpl_handler[opcode];
4594 atomic_store_rel_ptr(loc, new);
4600 an_not_handled(struct sge_iq *iq, const struct rsp_ctrl *ctrl)
4604 panic("%s: async notification on iq %p (ctrl %p)", __func__, iq, ctrl);
4606 log(LOG_ERR, "%s: async notification on iq %p (ctrl %p)\n",
4607 __func__, iq, ctrl);
4613 t4_register_an_handler(struct adapter *sc, an_handler_t h)
4615 uintptr_t *loc, new;
4617 new = h ? (uintptr_t)h : (uintptr_t)an_not_handled;
4618 loc = (uintptr_t *) &sc->an_handler;
4619 atomic_store_rel_ptr(loc, new);
4625 fw_msg_not_handled(struct adapter *sc, const __be64 *rpl)
4627 const struct cpl_fw6_msg *cpl =
4628 __containerof(rpl, struct cpl_fw6_msg, data[0]);
4631 panic("%s: fw_msg type %d", __func__, cpl->type);
4633 log(LOG_ERR, "%s: fw_msg type %d\n", __func__, cpl->type);
4639 t4_register_fw_msg_handler(struct adapter *sc, int type, fw_msg_handler_t h)
4641 uintptr_t *loc, new;
4643 if (type >= nitems(sc->fw_msg_handler))
4647 * These are dispatched by the handler for FW{4|6}_CPL_MSG using the CPL
4648 * handler dispatch table. Reject any attempt to install a handler for
4651 if (type == FW_TYPE_RSSCPL || type == FW6_TYPE_RSSCPL)
4654 new = h ? (uintptr_t)h : (uintptr_t)fw_msg_not_handled;
4655 loc = (uintptr_t *) &sc->fw_msg_handler[type];
4656 atomic_store_rel_ptr(loc, new);
4662 t4_sysctls(struct adapter *sc)
4664 struct sysctl_ctx_list *ctx;
4665 struct sysctl_oid *oid;
4666 struct sysctl_oid_list *children, *c0;
4667 static char *caps[] = {
4668 "\20\1PPP\2QFC\3DCBX", /* caps[0] linkcaps */
4669 "\20\1NIC\2VM\3IDS\4UM\5UM_ISGL" /* caps[1] niccaps */
4670 "\6HASHFILTER\7ETHOFLD",
4671 "\20\1TOE", /* caps[2] toecaps */
4672 "\20\1RDDP\2RDMAC", /* caps[3] rdmacaps */
4673 "\20\1INITIATOR_PDU\2TARGET_PDU" /* caps[4] iscsicaps */
4674 "\3INITIATOR_CNXOFLD\4TARGET_CNXOFLD"
4675 "\5INITIATOR_SSNOFLD\6TARGET_SSNOFLD",
4676 "\20\1INITIATOR\2TARGET\3CTRL_OFLD" /* caps[5] fcoecaps */
4677 "\4PO_INITIAOR\5PO_TARGET"
4679 static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
4681 ctx = device_get_sysctl_ctx(sc->dev);
4686 oid = device_get_sysctl_tree(sc->dev);
4687 c0 = children = SYSCTL_CHILDREN(oid);
4689 sc->sc_do_rxcopy = 1;
4690 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
4691 &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
4693 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
4694 sc->params.nports, "# of ports");
4696 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
4697 NULL, chip_rev(sc), "chip hardware revision");
4699 SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
4700 CTLFLAG_RD, sc->fw_version, 0, "firmware version");
4702 SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
4703 CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
4705 SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
4706 sc->cfcsum, "config file checksum");
4708 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
4709 CTLTYPE_STRING | CTLFLAG_RD, doorbells, sc->doorbells,
4710 sysctl_bitfield, "A", "available doorbells");
4712 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkcaps",
4713 CTLTYPE_STRING | CTLFLAG_RD, caps[0], sc->linkcaps,
4714 sysctl_bitfield, "A", "available link capabilities");
4716 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "niccaps",
4717 CTLTYPE_STRING | CTLFLAG_RD, caps[1], sc->niccaps,
4718 sysctl_bitfield, "A", "available NIC capabilities");
4720 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "toecaps",
4721 CTLTYPE_STRING | CTLFLAG_RD, caps[2], sc->toecaps,
4722 sysctl_bitfield, "A", "available TCP offload capabilities");
4724 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdmacaps",
4725 CTLTYPE_STRING | CTLFLAG_RD, caps[3], sc->rdmacaps,
4726 sysctl_bitfield, "A", "available RDMA capabilities");
4728 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "iscsicaps",
4729 CTLTYPE_STRING | CTLFLAG_RD, caps[4], sc->iscsicaps,
4730 sysctl_bitfield, "A", "available iSCSI capabilities");
4732 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoecaps",
4733 CTLTYPE_STRING | CTLFLAG_RD, caps[5], sc->fcoecaps,
4734 sysctl_bitfield, "A", "available FCoE capabilities");
4736 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
4737 sc->params.vpd.cclk, "core clock frequency (in KHz)");
4739 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
4740 CTLTYPE_STRING | CTLFLAG_RD, sc->sge.timer_val,
4741 sizeof(sc->sge.timer_val), sysctl_int_array, "A",
4742 "interrupt holdoff timer values (us)");
4744 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
4745 CTLTYPE_STRING | CTLFLAG_RD, sc->sge.counter_val,
4746 sizeof(sc->sge.counter_val), sysctl_int_array, "A",
4747 "interrupt holdoff packet counter values");
4749 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
4750 NULL, sc->tids.nftids, "number of filters");
4752 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
4753 CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
4754 "chip temperature (in Celsius)");
4756 t4_sge_sysctls(sc, ctx, children);
4758 sc->lro_timeout = 100;
4759 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
4760 &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
4762 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "debug_flags", CTLFLAG_RW,
4763 &sc->debug_flags, 0, "flags to enable runtime debugging");
4767 * dev.t4nex.X.misc. Marked CTLFLAG_SKIP to avoid information overload.
4769 oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
4770 CTLFLAG_RD | CTLFLAG_SKIP, NULL,
4771 "logs and miscellaneous information");
4772 children = SYSCTL_CHILDREN(oid);
4774 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
4775 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4776 sysctl_cctrl, "A", "congestion control");
4778 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
4779 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4780 sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
4782 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
4783 CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
4784 sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
4786 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
4787 CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
4788 sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
4790 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
4791 CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
4792 sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
4794 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
4795 CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
4796 sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
4798 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
4799 CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
4800 sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
4802 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
4803 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4804 sysctl_cim_la, "A", "CIM logic analyzer");
4806 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
4807 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4808 sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
4810 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
4811 CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
4812 sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
4814 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
4815 CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
4816 sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
4818 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
4819 CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
4820 sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
4822 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
4823 CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
4824 sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
4826 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
4827 CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
4828 sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
4830 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
4831 CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
4832 sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
4835 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
4836 CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
4837 sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
4839 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
4840 CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
4841 sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
4844 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
4845 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4846 sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
4848 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
4849 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4850 sysctl_cim_qcfg, "A", "CIM queue configuration");
4852 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
4853 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4854 sysctl_cpl_stats, "A", "CPL statistics");
4856 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
4857 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4858 sysctl_ddp_stats, "A", "non-TCP DDP statistics");
4860 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
4861 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4862 sysctl_devlog, "A", "firmware's device log");
4864 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
4865 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4866 sysctl_fcoe_stats, "A", "FCoE statistics");
4868 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
4869 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4870 sysctl_hw_sched, "A", "hardware scheduler ");
4872 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
4873 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4874 sysctl_l2t, "A", "hardware L2 table");
4876 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
4877 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4878 sysctl_lb_stats, "A", "loopback statistics");
4880 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
4881 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4882 sysctl_meminfo, "A", "memory regions");
4884 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
4885 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4886 sysctl_mps_tcam, "A", "MPS TCAM entries");
4888 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
4889 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4890 sysctl_path_mtus, "A", "path MTUs");
4892 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
4893 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4894 sysctl_pm_stats, "A", "PM statistics");
4896 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
4897 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4898 sysctl_rdma_stats, "A", "RDMA statistics");
4900 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
4901 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4902 sysctl_tcp_stats, "A", "TCP statistics");
4904 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
4905 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4906 sysctl_tids, "A", "TID information");
4908 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
4909 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4910 sysctl_tp_err_stats, "A", "TP error statistics");
4912 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
4913 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4914 sysctl_tp_la, "A", "TP logic analyzer");
4916 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
4917 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4918 sysctl_tx_rate, "A", "Tx rate");
4920 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
4921 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4922 sysctl_ulprx_la, "A", "ULPRX logic analyzer");
4925 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
4926 CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4927 sysctl_wcwr_stats, "A", "write combined work requests");
4932 if (is_offload(sc)) {
4936 oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
4937 NULL, "TOE parameters");
4938 children = SYSCTL_CHILDREN(oid);
4940 sc->tt.sndbuf = 256 * 1024;
4941 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
4942 &sc->tt.sndbuf, 0, "max hardware send buffer size");
4945 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
4946 &sc->tt.ddp, 0, "DDP allowed");
4948 sc->tt.indsz = G_INDICATESIZE(t4_read_reg(sc, A_TP_PARA_REG5));
4949 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "indsz", CTLFLAG_RW,
4950 &sc->tt.indsz, 0, "DDP max indicate size allowed");
4953 G_RXCOALESCESIZE(t4_read_reg(sc, A_TP_PARA_REG2));
4954 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp_thres", CTLFLAG_RW,
4955 &sc->tt.ddp_thres, 0, "DDP threshold");
4957 sc->tt.rx_coalesce = 1;
4958 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
4959 CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
4961 sc->tt.tx_align = 1;
4962 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
4963 CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
4972 cxgbe_sysctls(struct port_info *pi)
4974 struct sysctl_ctx_list *ctx;
4975 struct sysctl_oid *oid;
4976 struct sysctl_oid_list *children;
4977 struct adapter *sc = pi->adapter;
4979 ctx = device_get_sysctl_ctx(pi->dev);
4984 oid = device_get_sysctl_tree(pi->dev);
4985 children = SYSCTL_CHILDREN(oid);
4987 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
4988 CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
4989 if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
4990 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
4991 CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
4992 "PHY temperature (in Celsius)");
4993 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
4994 CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
4995 "PHY firmware version");
4997 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
4998 &pi->nrxq, 0, "# of rx queues");
4999 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
5000 &pi->ntxq, 0, "# of tx queues");
5001 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
5002 &pi->first_rxq, 0, "index of first rx queue");
5003 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
5004 &pi->first_txq, 0, "index of first tx queue");
5005 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq", CTLTYPE_INT |
5006 CTLFLAG_RW, pi, 0, sysctl_noflowq, "IU",
5007 "Reserve queue 0 for non-flowid packets");
5010 if (is_offload(sc)) {
5011 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
5013 "# of rx queues for offloaded TCP connections");
5014 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
5016 "# of tx queues for offloaded TCP connections");
5017 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
5018 CTLFLAG_RD, &pi->first_ofld_rxq, 0,
5019 "index of first TOE rx queue");
5020 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
5021 CTLFLAG_RD, &pi->first_ofld_txq, 0,
5022 "index of first TOE tx queue");
5026 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
5027 &pi->nnmrxq, 0, "# of rx queues for netmap");
5028 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
5029 &pi->nnmtxq, 0, "# of tx queues for netmap");
5030 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
5031 CTLFLAG_RD, &pi->first_nm_rxq, 0,
5032 "index of first netmap rx queue");
5033 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
5034 CTLFLAG_RD, &pi->first_nm_txq, 0,
5035 "index of first netmap tx queue");
5038 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
5039 CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_holdoff_tmr_idx, "I",
5040 "holdoff timer index");
5041 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
5042 CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_holdoff_pktc_idx, "I",
5043 "holdoff packet counter index");
5045 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
5046 CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_qsize_rxq, "I",
5048 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
5049 CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_qsize_txq, "I",
5052 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
5053 CTLTYPE_STRING | CTLFLAG_RW, pi, PAUSE_TX, sysctl_pause_settings,
5054 "A", "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
5057 * dev.cxgbe.X.stats.
5059 oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
5060 NULL, "port statistics");
5061 children = SYSCTL_CHILDREN(oid);
5062 SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
5063 &pi->tx_parse_error, 0,
5064 "# of tx packets with invalid length or # of segments");
5066 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
5067 SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
5068 CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
5069 sysctl_handle_t4_reg64, "QU", desc)
5071 SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
5072 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
5073 SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
5074 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
5075 SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
5076 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
5077 SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
5078 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
5079 SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
5080 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
5081 SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
5082 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
5083 SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
5084 "# of tx frames in this range",
5085 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
5086 SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
5087 "# of tx frames in this range",
5088 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
5089 SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
5090 "# of tx frames in this range",
5091 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
5092 SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
5093 "# of tx frames in this range",
5094 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
5095 SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
5096 "# of tx frames in this range",
5097 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
5098 SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
5099 "# of tx frames in this range",
5100 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
5101 SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
5102 "# of tx frames in this range",
5103 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
5104 SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
5105 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
5106 SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
5107 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
5108 SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
5109 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
5110 SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
5111 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
5112 SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
5113 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
5114 SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
5115 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
5116 SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
5117 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
5118 SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
5119 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
5120 SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
5121 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
5122 SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
5123 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
5125 SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
5126 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
5127 SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
5128 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
5129 SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
5130 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
5131 SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
5132 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
5133 SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
5134 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
5135 SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
5136 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
5137 SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
5138 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
5139 SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
5140 "# of frames received with bad FCS",
5141 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
5142 SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
5143 "# of frames received with length error",
5144 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
5145 SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
5146 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
5147 SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
5148 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
5149 SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
5150 "# of rx frames in this range",
5151 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
5152 SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
5153 "# of rx frames in this range",
5154 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
5155 SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
5156 "# of rx frames in this range",
5157 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
5158 SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
5159 "# of rx frames in this range",
5160 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
5161 SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
5162 "# of rx frames in this range",
5163 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
5164 SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
5165 "# of rx frames in this range",
5166 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
5167 SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
5168 "# of rx frames in this range",
5169 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
5170 SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
5171 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
5172 SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
5173 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
5174 SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
5175 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
5176 SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
5177 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
5178 SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
5179 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
5180 SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
5181 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
5182 SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
5183 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
5184 SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
5185 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
5186 SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
5187 PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
5189 #undef SYSCTL_ADD_T4_REG64
5191 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
5192 SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
5193 &pi->stats.name, desc)
5195 /* We get these from port_stats and they may be stale by upto 1s */
5196 SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
5197 "# drops due to buffer-group 0 overflows");
5198 SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
5199 "# drops due to buffer-group 1 overflows");
5200 SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
5201 "# drops due to buffer-group 2 overflows");
5202 SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
5203 "# drops due to buffer-group 3 overflows");
5204 SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
5205 "# of buffer-group 0 truncated packets");
5206 SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
5207 "# of buffer-group 1 truncated packets");
5208 SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
5209 "# of buffer-group 2 truncated packets");
5210 SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
5211 "# of buffer-group 3 truncated packets");
5213 #undef SYSCTL_ADD_T4_PORTSTAT
5219 sysctl_int_array(SYSCTL_HANDLER_ARGS)
5221 int rc, *i, space = 0;
5224 sbuf_new_for_sysctl(&sb, NULL, 64, req);
5225 for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
5227 sbuf_printf(&sb, " ");
5228 sbuf_printf(&sb, "%d", *i);
5231 rc = sbuf_finish(&sb);
5237 sysctl_bitfield(SYSCTL_HANDLER_ARGS)
5242 rc = sysctl_wire_old_buffer(req, 0);
5246 sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5250 sbuf_printf(sb, "%b", (int)arg2, (char *)arg1);
5251 rc = sbuf_finish(sb);
5258 sysctl_btphy(SYSCTL_HANDLER_ARGS)
5260 struct port_info *pi = arg1;
5262 struct adapter *sc = pi->adapter;
5266 rc = begin_synchronized_op(sc, pi, SLEEP_OK | INTR_OK, "t4btt");
5269 /* XXX: magic numbers */
5270 rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
5272 end_synchronized_op(sc, 0);
5278 rc = sysctl_handle_int(oidp, &v, 0, req);
5283 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
5285 struct port_info *pi = arg1;
5288 val = pi->rsrv_noflowq;
5289 rc = sysctl_handle_int(oidp, &val, 0, req);
5290 if (rc != 0 || req->newptr == NULL)
5293 if ((val >= 1) && (pi->ntxq > 1))
5294 pi->rsrv_noflowq = 1;
5296 pi->rsrv_noflowq = 0;
5302 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
5304 struct port_info *pi = arg1;
5305 struct adapter *sc = pi->adapter;
5307 struct sge_rxq *rxq;
5309 struct sge_ofld_rxq *ofld_rxq;
5315 rc = sysctl_handle_int(oidp, &idx, 0, req);
5316 if (rc != 0 || req->newptr == NULL)
5319 if (idx < 0 || idx >= SGE_NTIMERS)
5322 rc = begin_synchronized_op(sc, pi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5327 v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(pi->pktc_idx != -1);
5328 for_each_rxq(pi, i, rxq) {
5329 #ifdef atomic_store_rel_8
5330 atomic_store_rel_8(&rxq->iq.intr_params, v);
5332 rxq->iq.intr_params = v;
5336 for_each_ofld_rxq(pi, i, ofld_rxq) {
5337 #ifdef atomic_store_rel_8
5338 atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
5340 ofld_rxq->iq.intr_params = v;
5346 end_synchronized_op(sc, LOCK_HELD);
5351 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
5353 struct port_info *pi = arg1;
5354 struct adapter *sc = pi->adapter;
5359 rc = sysctl_handle_int(oidp, &idx, 0, req);
5360 if (rc != 0 || req->newptr == NULL)
5363 if (idx < -1 || idx >= SGE_NCOUNTERS)
5366 rc = begin_synchronized_op(sc, pi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5371 if (pi->flags & PORT_INIT_DONE)
5372 rc = EBUSY; /* cannot be changed once the queues are created */
5376 end_synchronized_op(sc, LOCK_HELD);
5381 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
5383 struct port_info *pi = arg1;
5384 struct adapter *sc = pi->adapter;
5387 qsize = pi->qsize_rxq;
5389 rc = sysctl_handle_int(oidp, &qsize, 0, req);
5390 if (rc != 0 || req->newptr == NULL)
5393 if (qsize < 128 || (qsize & 7))
5396 rc = begin_synchronized_op(sc, pi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5401 if (pi->flags & PORT_INIT_DONE)
5402 rc = EBUSY; /* cannot be changed once the queues are created */
5404 pi->qsize_rxq = qsize;
5406 end_synchronized_op(sc, LOCK_HELD);
5411 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
5413 struct port_info *pi = arg1;
5414 struct adapter *sc = pi->adapter;
5417 qsize = pi->qsize_txq;
5419 rc = sysctl_handle_int(oidp, &qsize, 0, req);
5420 if (rc != 0 || req->newptr == NULL)
5423 if (qsize < 128 || qsize > 65536)
5426 rc = begin_synchronized_op(sc, pi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5431 if (pi->flags & PORT_INIT_DONE)
5432 rc = EBUSY; /* cannot be changed once the queues are created */
5434 pi->qsize_txq = qsize;
5436 end_synchronized_op(sc, LOCK_HELD);
5441 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
5443 struct port_info *pi = arg1;
5444 struct adapter *sc = pi->adapter;
5445 struct link_config *lc = &pi->link_cfg;
5448 if (req->newptr == NULL) {
5450 static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
5452 rc = sysctl_wire_old_buffer(req, 0);
5456 sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5460 sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
5461 rc = sbuf_finish(sb);
5467 s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
5470 rc = sysctl_handle_string(oidp, s, sizeof(s), req);
5476 if (s[0] < '0' || s[0] > '9')
5477 return (EINVAL); /* not a number */
5479 if (n & ~(PAUSE_TX | PAUSE_RX))
5480 return (EINVAL); /* some other bit is set too */
5482 rc = begin_synchronized_op(sc, pi, SLEEP_OK | INTR_OK, "t4PAUSE");
5485 if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
5486 int link_ok = lc->link_ok;
5488 lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
5489 lc->requested_fc |= n;
5490 rc = -t4_link_start(sc, sc->mbox, pi->tx_chan, lc);
5491 lc->link_ok = link_ok; /* restore */
5493 end_synchronized_op(sc, 0);
5500 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
5502 struct adapter *sc = arg1;
5506 val = t4_read_reg64(sc, reg);
5508 return (sysctl_handle_64(oidp, &val, 0, req));
5512 sysctl_temperature(SYSCTL_HANDLER_ARGS)
5514 struct adapter *sc = arg1;
5516 uint32_t param, val;
5518 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
5521 param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
5522 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
5523 V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
5524 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, ¶m, &val);
5525 end_synchronized_op(sc, 0);
5529 /* unknown is returned as 0 but we display -1 in that case */
5530 t = val == 0 ? -1 : val;
5532 rc = sysctl_handle_int(oidp, &t, 0, req);
5538 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
5540 struct adapter *sc = arg1;
5543 uint16_t incr[NMTUS][NCCTRL_WIN];
5544 static const char *dec_fac[] = {
5545 "0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
5549 rc = sysctl_wire_old_buffer(req, 0);
5553 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5557 t4_read_cong_tbl(sc, incr);
5559 for (i = 0; i < NCCTRL_WIN; ++i) {
5560 sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
5561 incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
5562 incr[5][i], incr[6][i], incr[7][i]);
5563 sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
5564 incr[8][i], incr[9][i], incr[10][i], incr[11][i],
5565 incr[12][i], incr[13][i], incr[14][i], incr[15][i],
5566 sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
5569 rc = sbuf_finish(sb);
5575 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
5576 "TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI", /* ibq's */
5577 "ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI", /* obq's */
5578 "SGE0-RX", "SGE1-RX" /* additional obq's (T5 onwards) */
5582 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
5584 struct adapter *sc = arg1;
5586 int rc, i, n, qid = arg2;
5589 u_int cim_num_obq = is_t4(sc) ? CIM_NUM_OBQ : CIM_NUM_OBQ_T5;
5591 KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
5592 ("%s: bad qid %d\n", __func__, qid));
5594 if (qid < CIM_NUM_IBQ) {
5597 n = 4 * CIM_IBQ_SIZE;
5598 buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
5599 rc = t4_read_cim_ibq(sc, qid, buf, n);
5601 /* outbound queue */
5604 n = 4 * cim_num_obq * CIM_OBQ_SIZE;
5605 buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
5606 rc = t4_read_cim_obq(sc, qid, buf, n);
5613 n = rc * sizeof(uint32_t); /* rc has # of words actually read */
5615 rc = sysctl_wire_old_buffer(req, 0);
5619 sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
5625 sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
5626 for (i = 0, p = buf; i < n; i += 16, p += 4)
5627 sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
5630 rc = sbuf_finish(sb);
5638 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
5640 struct adapter *sc = arg1;
5646 rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
5650 rc = sysctl_wire_old_buffer(req, 0);
5654 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5658 buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
5661 rc = -t4_cim_read_la(sc, buf, NULL);
5665 sbuf_printf(sb, "Status Data PC%s",
5666 cfg & F_UPDBGLACAPTPCONLY ? "" :
5667 " LS0Stat LS0Addr LS0Data");
5669 KASSERT((sc->params.cim_la_size & 7) == 0,
5670 ("%s: p will walk off the end of buf", __func__));
5672 for (p = buf; p < &buf[sc->params.cim_la_size]; p += 8) {
5673 if (cfg & F_UPDBGLACAPTPCONLY) {
5674 sbuf_printf(sb, "\n %02x %08x %08x", p[5] & 0xff,
5676 sbuf_printf(sb, "\n %02x %02x%06x %02x%06x",
5677 (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
5678 p[4] & 0xff, p[5] >> 8);
5679 sbuf_printf(sb, "\n %02x %x%07x %x%07x",
5680 (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
5681 p[1] & 0xf, p[2] >> 4);
5684 "\n %02x %x%07x %x%07x %08x %08x "
5686 (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
5687 p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
5692 rc = sbuf_finish(sb);
5700 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
5702 struct adapter *sc = arg1;
5708 rc = sysctl_wire_old_buffer(req, 0);
5712 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5716 buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
5719 t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
5722 for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
5723 sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
5727 sbuf_printf(sb, "\n\nCnt ID Tag UE Data RDY VLD");
5728 for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
5729 sbuf_printf(sb, "\n%3u %2u %x %u %08x%08x %u %u",
5730 (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
5731 (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
5732 (p[1] >> 2) | ((p[2] & 3) << 30),
5733 (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
5737 rc = sbuf_finish(sb);
5744 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
5746 struct adapter *sc = arg1;
5752 rc = sysctl_wire_old_buffer(req, 0);
5756 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5760 buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
5763 t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
5766 sbuf_printf(sb, "Cntl ID DataBE Addr Data");
5767 for (i = 0; i < CIM_MALA_SIZE; i++, p += 6) {
5768 sbuf_printf(sb, "\n %02x %02x %04x %08x %08x%08x%08x%08x",
5769 (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
5770 p[4], p[3], p[2], p[1], p[0]);
5773 sbuf_printf(sb, "\n\nCntl ID Data");
5774 for (i = 0; i < CIM_MALA_SIZE; i++, p += 6) {
5775 sbuf_printf(sb, "\n %02x %02x %08x%08x%08x%08x",
5776 (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
5779 rc = sbuf_finish(sb);
5786 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
5788 struct adapter *sc = arg1;
5791 uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
5792 uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
5793 uint16_t thres[CIM_NUM_IBQ];
5794 uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
5795 uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
5796 u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
5799 cim_num_obq = CIM_NUM_OBQ;
5800 ibq_rdaddr = A_UP_IBQ_0_RDADDR;
5801 obq_rdaddr = A_UP_OBQ_0_REALADDR;
5803 cim_num_obq = CIM_NUM_OBQ_T5;
5804 ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
5805 obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
5807 nq = CIM_NUM_IBQ + cim_num_obq;
5809 rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
5811 rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
5815 t4_read_cimq_cfg(sc, base, size, thres);
5817 rc = sysctl_wire_old_buffer(req, 0);
5821 sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
5825 sbuf_printf(sb, "Queue Base Size Thres RdPtr WrPtr SOP EOP Avail");
5827 for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
5828 sbuf_printf(sb, "\n%7s %5x %5u %5u %6x %4x %4u %4u %5u",
5829 qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
5830 G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
5831 G_QUEREMFLITS(p[2]) * 16);
5832 for ( ; i < nq; i++, p += 4, wr += 2)
5833 sbuf_printf(sb, "\n%7s %5x %5u %12x %4x %4u %4u %5u", qname[i],
5834 base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
5835 wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
5836 G_QUEREMFLITS(p[2]) * 16);
5838 rc = sbuf_finish(sb);
5845 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
5847 struct adapter *sc = arg1;
5850 struct tp_cpl_stats stats;
5852 rc = sysctl_wire_old_buffer(req, 0);
5856 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
5860 t4_tp_get_cpl_stats(sc, &stats);
5862 sbuf_printf(sb, " channel 0 channel 1 channel 2 "
5864 sbuf_printf(sb, "CPL requests: %10u %10u %10u %10u\n",
5865 stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
5866 sbuf_printf(sb, "CPL responses: %10u %10u %10u %10u",
5867 stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
5869 rc = sbuf_finish(sb);
5876 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
5878 struct adapter *sc = arg1;
5881 struct tp_usm_stats stats;
5883 rc = sysctl_wire_old_buffer(req, 0);
5887 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
5891 t4_get_usm_stats(sc, &stats);
5893 sbuf_printf(sb, "Frames: %u\n", stats.frames);
5894 sbuf_printf(sb, "Octets: %ju\n", stats.octets);
5895 sbuf_printf(sb, "Drops: %u", stats.drops);
5897 rc = sbuf_finish(sb);
5903 const char *devlog_level_strings[] = {
5904 [FW_DEVLOG_LEVEL_EMERG] = "EMERG",
5905 [FW_DEVLOG_LEVEL_CRIT] = "CRIT",
5906 [FW_DEVLOG_LEVEL_ERR] = "ERR",
5907 [FW_DEVLOG_LEVEL_NOTICE] = "NOTICE",
5908 [FW_DEVLOG_LEVEL_INFO] = "INFO",
5909 [FW_DEVLOG_LEVEL_DEBUG] = "DEBUG"
5912 const char *devlog_facility_strings[] = {
5913 [FW_DEVLOG_FACILITY_CORE] = "CORE",
5914 [FW_DEVLOG_FACILITY_CF] = "CF",
5915 [FW_DEVLOG_FACILITY_SCHED] = "SCHED",
5916 [FW_DEVLOG_FACILITY_TIMER] = "TIMER",
5917 [FW_DEVLOG_FACILITY_RES] = "RES",
5918 [FW_DEVLOG_FACILITY_HW] = "HW",
5919 [FW_DEVLOG_FACILITY_FLR] = "FLR",
5920 [FW_DEVLOG_FACILITY_DMAQ] = "DMAQ",
5921 [FW_DEVLOG_FACILITY_PHY] = "PHY",
5922 [FW_DEVLOG_FACILITY_MAC] = "MAC",
5923 [FW_DEVLOG_FACILITY_PORT] = "PORT",
5924 [FW_DEVLOG_FACILITY_VI] = "VI",
5925 [FW_DEVLOG_FACILITY_FILTER] = "FILTER",
5926 [FW_DEVLOG_FACILITY_ACL] = "ACL",
5927 [FW_DEVLOG_FACILITY_TM] = "TM",
5928 [FW_DEVLOG_FACILITY_QFC] = "QFC",
5929 [FW_DEVLOG_FACILITY_DCB] = "DCB",
5930 [FW_DEVLOG_FACILITY_ETH] = "ETH",
5931 [FW_DEVLOG_FACILITY_OFLD] = "OFLD",
5932 [FW_DEVLOG_FACILITY_RI] = "RI",
5933 [FW_DEVLOG_FACILITY_ISCSI] = "ISCSI",
5934 [FW_DEVLOG_FACILITY_FCOE] = "FCOE",
5935 [FW_DEVLOG_FACILITY_FOISCSI] = "FOISCSI",
5936 [FW_DEVLOG_FACILITY_FOFCOE] = "FOFCOE"
5940 sysctl_devlog(SYSCTL_HANDLER_ARGS)
5942 struct adapter *sc = arg1;
5943 struct devlog_params *dparams = &sc->params.devlog;
5944 struct fw_devlog_e *buf, *e;
5945 int i, j, rc, nentries, first = 0, m;
5947 uint64_t ftstamp = UINT64_MAX;
5949 if (dparams->start == 0) {
5950 dparams->memtype = FW_MEMTYPE_EDC0;
5951 dparams->start = 0x84000;
5952 dparams->size = 32768;
5955 nentries = dparams->size / sizeof(struct fw_devlog_e);
5957 buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
5961 m = fwmtype_to_hwmtype(dparams->memtype);
5962 rc = -t4_mem_read(sc, m, dparams->start, dparams->size, (void *)buf);
5966 for (i = 0; i < nentries; i++) {
5969 if (e->timestamp == 0)
5972 e->timestamp = be64toh(e->timestamp);
5973 e->seqno = be32toh(e->seqno);
5974 for (j = 0; j < 8; j++)
5975 e->params[j] = be32toh(e->params[j]);
5977 if (e->timestamp < ftstamp) {
5978 ftstamp = e->timestamp;
5983 if (buf[first].timestamp == 0)
5984 goto done; /* nothing in the log */
5986 rc = sysctl_wire_old_buffer(req, 0);
5990 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5995 sbuf_printf(sb, "%10s %15s %8s %8s %s\n",
5996 "Seq#", "Tstamp", "Level", "Facility", "Message");
6001 if (e->timestamp == 0)
6004 sbuf_printf(sb, "%10d %15ju %8s %8s ",
6005 e->seqno, e->timestamp,
6006 (e->level < nitems(devlog_level_strings) ?
6007 devlog_level_strings[e->level] : "UNKNOWN"),
6008 (e->facility < nitems(devlog_facility_strings) ?
6009 devlog_facility_strings[e->facility] : "UNKNOWN"));
6010 sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
6011 e->params[2], e->params[3], e->params[4],
6012 e->params[5], e->params[6], e->params[7]);
6014 if (++i == nentries)
6016 } while (i != first);
6018 rc = sbuf_finish(sb);
6026 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
6028 struct adapter *sc = arg1;
6031 struct tp_fcoe_stats stats[4];
6033 rc = sysctl_wire_old_buffer(req, 0);
6037 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6041 t4_get_fcoe_stats(sc, 0, &stats[0]);
6042 t4_get_fcoe_stats(sc, 1, &stats[1]);
6043 t4_get_fcoe_stats(sc, 2, &stats[2]);
6044 t4_get_fcoe_stats(sc, 3, &stats[3]);
6046 sbuf_printf(sb, " channel 0 channel 1 "
6047 "channel 2 channel 3\n");
6048 sbuf_printf(sb, "octetsDDP: %16ju %16ju %16ju %16ju\n",
6049 stats[0].octetsDDP, stats[1].octetsDDP, stats[2].octetsDDP,
6050 stats[3].octetsDDP);
6051 sbuf_printf(sb, "framesDDP: %16u %16u %16u %16u\n", stats[0].framesDDP,
6052 stats[1].framesDDP, stats[2].framesDDP, stats[3].framesDDP);
6053 sbuf_printf(sb, "framesDrop: %16u %16u %16u %16u",
6054 stats[0].framesDrop, stats[1].framesDrop, stats[2].framesDrop,
6055 stats[3].framesDrop);
6057 rc = sbuf_finish(sb);
6064 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
6066 struct adapter *sc = arg1;
6069 unsigned int map, kbps, ipg, mode;
6070 unsigned int pace_tab[NTX_SCHED];
6072 rc = sysctl_wire_old_buffer(req, 0);
6076 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6080 map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
6081 mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
6082 t4_read_pace_tbl(sc, pace_tab);
6084 sbuf_printf(sb, "Scheduler Mode Channel Rate (Kbps) "
6085 "Class IPG (0.1 ns) Flow IPG (us)");
6087 for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
6088 t4_get_tx_sched(sc, i, &kbps, &ipg);
6089 sbuf_printf(sb, "\n %u %-5s %u ", i,
6090 (mode & (1 << i)) ? "flow" : "class", map & 3);
6092 sbuf_printf(sb, "%9u ", kbps);
6094 sbuf_printf(sb, " disabled ");
6097 sbuf_printf(sb, "%13u ", ipg);
6099 sbuf_printf(sb, " disabled ");
6102 sbuf_printf(sb, "%10u", pace_tab[i]);
6104 sbuf_printf(sb, " disabled");
6107 rc = sbuf_finish(sb);
6114 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
6116 struct adapter *sc = arg1;
6120 struct lb_port_stats s[2];
6121 static const char *stat_name[] = {
6122 "OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
6123 "UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
6124 "Frames128To255:", "Frames256To511:", "Frames512To1023:",
6125 "Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
6126 "BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
6127 "BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
6128 "BG2FramesTrunc:", "BG3FramesTrunc:"
6131 rc = sysctl_wire_old_buffer(req, 0);
6135 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6139 memset(s, 0, sizeof(s));
6141 for (i = 0; i < 4; i += 2) {
6142 t4_get_lb_stats(sc, i, &s[0]);
6143 t4_get_lb_stats(sc, i + 1, &s[1]);
6147 sbuf_printf(sb, "%s Loopback %u"
6148 " Loopback %u", i == 0 ? "" : "\n", i, i + 1);
6150 for (j = 0; j < nitems(stat_name); j++)
6151 sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
6155 rc = sbuf_finish(sb);
6162 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
6165 struct port_info *pi = arg1;
6167 static const char *linkdnreasons[] = {
6168 "non-specific", "remote fault", "autoneg failed", "reserved3",
6169 "PHY overheated", "unknown", "rx los", "reserved7"
6172 rc = sysctl_wire_old_buffer(req, 0);
6175 sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
6179 if (pi->linkdnrc < 0)
6180 sbuf_printf(sb, "n/a");
6181 else if (pi->linkdnrc < nitems(linkdnreasons))
6182 sbuf_printf(sb, "%s", linkdnreasons[pi->linkdnrc]);
6184 sbuf_printf(sb, "%d", pi->linkdnrc);
6186 rc = sbuf_finish(sb);
6199 mem_desc_cmp(const void *a, const void *b)
6201 return ((const struct mem_desc *)a)->base -
6202 ((const struct mem_desc *)b)->base;
6206 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
6211 size = to - from + 1;
6215 /* XXX: need humanize_number(3) in libkern for a more readable 'size' */
6216 sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
6220 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
6222 struct adapter *sc = arg1;
6225 uint32_t lo, hi, used, alloc;
6226 static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
6227 static const char *region[] = {
6228 "DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
6229 "Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
6230 "Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
6231 "TDDP region:", "TPT region:", "STAG region:", "RQ region:",
6232 "RQUDP region:", "PBL region:", "TXPBL region:",
6233 "DBVFIFO region:", "ULPRX state:", "ULPTX state:",
6236 struct mem_desc avail[4];
6237 struct mem_desc mem[nitems(region) + 3]; /* up to 3 holes */
6238 struct mem_desc *md = mem;
6240 rc = sysctl_wire_old_buffer(req, 0);
6244 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6248 for (i = 0; i < nitems(mem); i++) {
6253 /* Find and sort the populated memory ranges */
6255 lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
6256 if (lo & F_EDRAM0_ENABLE) {
6257 hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
6258 avail[i].base = G_EDRAM0_BASE(hi) << 20;
6259 avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
6263 if (lo & F_EDRAM1_ENABLE) {
6264 hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
6265 avail[i].base = G_EDRAM1_BASE(hi) << 20;
6266 avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
6270 if (lo & F_EXT_MEM_ENABLE) {
6271 hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
6272 avail[i].base = G_EXT_MEM_BASE(hi) << 20;
6273 avail[i].limit = avail[i].base +
6274 (G_EXT_MEM_SIZE(hi) << 20);
6275 avail[i].idx = is_t4(sc) ? 2 : 3; /* Call it MC for T4 */
6278 if (!is_t4(sc) && lo & F_EXT_MEM1_ENABLE) {
6279 hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
6280 avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
6281 avail[i].limit = avail[i].base +
6282 (G_EXT_MEM1_SIZE(hi) << 20);
6286 if (!i) /* no memory available */
6288 qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
6290 (md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
6291 (md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
6292 (md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
6293 (md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
6294 (md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
6295 (md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
6296 (md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
6297 (md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
6298 (md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
6300 /* the next few have explicit upper bounds */
6301 md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
6302 md->limit = md->base - 1 +
6303 t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
6304 G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
6307 md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
6308 md->limit = md->base - 1 +
6309 t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
6310 G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
6313 if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
6314 hi = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
6315 md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
6316 md->limit = (sc->tids.ntids - hi) * 16 + md->base - 1;
6319 md->idx = nitems(region); /* hide it */
6323 #define ulp_region(reg) \
6324 md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
6325 (md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
6327 ulp_region(RX_ISCSI);
6328 ulp_region(RX_TDDP);
6330 ulp_region(RX_STAG);
6332 ulp_region(RX_RQUDP);
6338 md->idx = nitems(region);
6339 if (!is_t4(sc) && t4_read_reg(sc, A_SGE_CONTROL2) & F_VFIFO_ENABLE) {
6340 md->base = G_BASEADDR(t4_read_reg(sc, A_SGE_DBVFIFO_BADDR));
6341 md->limit = md->base + (G_DBVFIFO_SIZE((t4_read_reg(sc,
6342 A_SGE_DBVFIFO_SIZE))) << 2) - 1;
6346 md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
6347 md->limit = md->base + sc->tids.ntids - 1;
6349 md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
6350 md->limit = md->base + sc->tids.ntids - 1;
6353 md->base = sc->vres.ocq.start;
6354 if (sc->vres.ocq.size)
6355 md->limit = md->base + sc->vres.ocq.size - 1;
6357 md->idx = nitems(region); /* hide it */
6360 /* add any address-space holes, there can be up to 3 */
6361 for (n = 0; n < i - 1; n++)
6362 if (avail[n].limit < avail[n + 1].base)
6363 (md++)->base = avail[n].limit;
6365 (md++)->base = avail[n].limit;
6368 qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
6370 for (lo = 0; lo < i; lo++)
6371 mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
6372 avail[lo].limit - 1);
6374 sbuf_printf(sb, "\n");
6375 for (i = 0; i < n; i++) {
6376 if (mem[i].idx >= nitems(region))
6377 continue; /* skip holes */
6379 mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
6380 mem_region_show(sb, region[mem[i].idx], mem[i].base,
6384 sbuf_printf(sb, "\n");
6385 lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
6386 hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
6387 mem_region_show(sb, "uP RAM:", lo, hi);
6389 lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
6390 hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
6391 mem_region_show(sb, "uP Extmem2:", lo, hi);
6393 lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
6394 sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
6396 t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
6397 (lo & F_PMRXNUMCHN) ? 2 : 1);
6399 lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
6400 hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
6401 sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
6403 hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
6404 hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
6405 sbuf_printf(sb, "%u p-structs\n",
6406 t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
6408 for (i = 0; i < 4; i++) {
6409 lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
6412 alloc = G_ALLOC(lo);
6414 used = G_T5_USED(lo);
6415 alloc = G_T5_ALLOC(lo);
6417 sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
6420 for (i = 0; i < 4; i++) {
6421 lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
6424 alloc = G_ALLOC(lo);
6426 used = G_T5_USED(lo);
6427 alloc = G_T5_ALLOC(lo);
6430 "\nLoopback %d using %u pages out of %u allocated",
6434 rc = sbuf_finish(sb);
6441 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
6445 memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
6449 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
6451 struct adapter *sc = arg1;
6455 rc = sysctl_wire_old_buffer(req, 0);
6459 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6464 "Idx Ethernet address Mask Vld Ports PF"
6465 " VF Replication P0 P1 P2 P3 ML");
6466 n = is_t4(sc) ? NUM_MPS_CLS_SRAM_L_INSTANCES :
6467 NUM_MPS_T5_CLS_SRAM_L_INSTANCES;
6468 for (i = 0; i < n; i++) {
6469 uint64_t tcamx, tcamy, mask;
6470 uint32_t cls_lo, cls_hi;
6471 uint8_t addr[ETHER_ADDR_LEN];
6473 tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
6474 tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
6475 cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
6476 cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
6481 tcamxy2valmask(tcamx, tcamy, addr, &mask);
6482 sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
6483 " %c %#x%4u%4d", i, addr[0], addr[1], addr[2],
6484 addr[3], addr[4], addr[5], (uintmax_t)mask,
6485 (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
6486 G_PORTMAP(cls_hi), G_PF(cls_lo),
6487 (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
6489 if (cls_lo & F_REPLICATE) {
6490 struct fw_ldst_cmd ldst_cmd;
6492 memset(&ldst_cmd, 0, sizeof(ldst_cmd));
6493 ldst_cmd.op_to_addrspace =
6494 htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
6495 F_FW_CMD_REQUEST | F_FW_CMD_READ |
6496 V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
6497 ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
6498 ldst_cmd.u.mps.rplc.fid_idx =
6499 htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
6500 V_FW_LDST_CMD_IDX(i));
6502 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
6506 rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
6507 sizeof(ldst_cmd), &ldst_cmd);
6508 end_synchronized_op(sc, 0);
6512 " ------------ error %3u ------------", rc);
6515 sbuf_printf(sb, " %08x %08x %08x %08x",
6516 be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
6517 be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
6518 be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
6519 be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
6522 sbuf_printf(sb, "%36s", "");
6524 sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
6525 G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
6526 G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
6530 (void) sbuf_finish(sb);
6532 rc = sbuf_finish(sb);
6539 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
6541 struct adapter *sc = arg1;
6544 uint16_t mtus[NMTUS];
6546 rc = sysctl_wire_old_buffer(req, 0);
6550 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6554 t4_read_mtu_tbl(sc, mtus, NULL);
6556 sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
6557 mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
6558 mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
6559 mtus[14], mtus[15]);
6561 rc = sbuf_finish(sb);
6568 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
6570 struct adapter *sc = arg1;
6573 uint32_t cnt[PM_NSTATS];
6574 uint64_t cyc[PM_NSTATS];
6575 static const char *rx_stats[] = {
6576 "Read:", "Write bypass:", "Write mem:", "Flush:"
6578 static const char *tx_stats[] = {
6579 "Read:", "Write bypass:", "Write mem:", "Bypass + mem:"
6582 rc = sysctl_wire_old_buffer(req, 0);
6586 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6590 t4_pmtx_get_stats(sc, cnt, cyc);
6591 sbuf_printf(sb, " Tx pcmds Tx bytes");
6592 for (i = 0; i < ARRAY_SIZE(tx_stats); i++)
6593 sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], cnt[i],
6596 t4_pmrx_get_stats(sc, cnt, cyc);
6597 sbuf_printf(sb, "\n Rx pcmds Rx bytes");
6598 for (i = 0; i < ARRAY_SIZE(rx_stats); i++)
6599 sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], cnt[i],
6602 rc = sbuf_finish(sb);
6609 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
6611 struct adapter *sc = arg1;
6614 struct tp_rdma_stats stats;
6616 rc = sysctl_wire_old_buffer(req, 0);
6620 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6624 t4_tp_get_rdma_stats(sc, &stats);
6625 sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
6626 sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
6628 rc = sbuf_finish(sb);
6635 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
6637 struct adapter *sc = arg1;
6640 struct tp_tcp_stats v4, v6;
6642 rc = sysctl_wire_old_buffer(req, 0);
6646 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6650 t4_tp_get_tcp_stats(sc, &v4, &v6);
6653 sbuf_printf(sb, "OutRsts: %20u %20u\n",
6654 v4.tcpOutRsts, v6.tcpOutRsts);
6655 sbuf_printf(sb, "InSegs: %20ju %20ju\n",
6656 v4.tcpInSegs, v6.tcpInSegs);
6657 sbuf_printf(sb, "OutSegs: %20ju %20ju\n",
6658 v4.tcpOutSegs, v6.tcpOutSegs);
6659 sbuf_printf(sb, "RetransSegs: %20ju %20ju",
6660 v4.tcpRetransSegs, v6.tcpRetransSegs);
6662 rc = sbuf_finish(sb);
6669 sysctl_tids(SYSCTL_HANDLER_ARGS)
6671 struct adapter *sc = arg1;
6674 struct tid_info *t = &sc->tids;
6676 rc = sysctl_wire_old_buffer(req, 0);
6680 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6685 sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
6690 if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
6691 uint32_t b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
6694 sbuf_printf(sb, "TID range: 0-%u, %u-%u", b - 1,
6695 t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4,
6698 sbuf_printf(sb, "TID range: %u-%u",
6699 t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4,
6703 sbuf_printf(sb, "TID range: 0-%u", t->ntids - 1);
6704 sbuf_printf(sb, ", in use: %u\n",
6705 atomic_load_acq_int(&t->tids_in_use));
6709 sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
6710 t->stid_base + t->nstids - 1, t->stids_in_use);
6714 sbuf_printf(sb, "FTID range: %u-%u\n", t->ftid_base,
6715 t->ftid_base + t->nftids - 1);
6719 sbuf_printf(sb, "ETID range: %u-%u\n", t->etid_base,
6720 t->etid_base + t->netids - 1);
6723 sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
6724 t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
6725 t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
6727 rc = sbuf_finish(sb);
6734 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
6736 struct adapter *sc = arg1;
6739 struct tp_err_stats stats;
6741 rc = sysctl_wire_old_buffer(req, 0);
6745 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6749 t4_tp_get_err_stats(sc, &stats);
6751 sbuf_printf(sb, " channel 0 channel 1 channel 2 "
6753 sbuf_printf(sb, "macInErrs: %10u %10u %10u %10u\n",
6754 stats.macInErrs[0], stats.macInErrs[1], stats.macInErrs[2],
6755 stats.macInErrs[3]);
6756 sbuf_printf(sb, "hdrInErrs: %10u %10u %10u %10u\n",
6757 stats.hdrInErrs[0], stats.hdrInErrs[1], stats.hdrInErrs[2],
6758 stats.hdrInErrs[3]);
6759 sbuf_printf(sb, "tcpInErrs: %10u %10u %10u %10u\n",
6760 stats.tcpInErrs[0], stats.tcpInErrs[1], stats.tcpInErrs[2],
6761 stats.tcpInErrs[3]);
6762 sbuf_printf(sb, "tcp6InErrs: %10u %10u %10u %10u\n",
6763 stats.tcp6InErrs[0], stats.tcp6InErrs[1], stats.tcp6InErrs[2],
6764 stats.tcp6InErrs[3]);
6765 sbuf_printf(sb, "tnlCongDrops: %10u %10u %10u %10u\n",
6766 stats.tnlCongDrops[0], stats.tnlCongDrops[1], stats.tnlCongDrops[2],
6767 stats.tnlCongDrops[3]);
6768 sbuf_printf(sb, "tnlTxDrops: %10u %10u %10u %10u\n",
6769 stats.tnlTxDrops[0], stats.tnlTxDrops[1], stats.tnlTxDrops[2],
6770 stats.tnlTxDrops[3]);
6771 sbuf_printf(sb, "ofldVlanDrops: %10u %10u %10u %10u\n",
6772 stats.ofldVlanDrops[0], stats.ofldVlanDrops[1],
6773 stats.ofldVlanDrops[2], stats.ofldVlanDrops[3]);
6774 sbuf_printf(sb, "ofldChanDrops: %10u %10u %10u %10u\n\n",
6775 stats.ofldChanDrops[0], stats.ofldChanDrops[1],
6776 stats.ofldChanDrops[2], stats.ofldChanDrops[3]);
6777 sbuf_printf(sb, "ofldNoNeigh: %u\nofldCongDefer: %u",
6778 stats.ofldNoNeigh, stats.ofldCongDefer);
6780 rc = sbuf_finish(sb);
6793 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
6799 uint64_t mask = (1ULL << f->width) - 1;
6800 int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
6801 ((uintmax_t)v >> f->start) & mask);
6803 if (line_size + len >= 79) {
6805 sbuf_printf(sb, "\n ");
6807 sbuf_printf(sb, "%s ", buf);
6808 line_size += len + 1;
6811 sbuf_printf(sb, "\n");
6814 static struct field_desc tp_la0[] = {
6815 { "RcfOpCodeOut", 60, 4 },
6817 { "WcfState", 52, 4 },
6818 { "RcfOpcSrcOut", 50, 2 },
6819 { "CRxError", 49, 1 },
6820 { "ERxError", 48, 1 },
6821 { "SanityFailed", 47, 1 },
6822 { "SpuriousMsg", 46, 1 },
6823 { "FlushInputMsg", 45, 1 },
6824 { "FlushInputCpl", 44, 1 },
6825 { "RssUpBit", 43, 1 },
6826 { "RssFilterHit", 42, 1 },
6828 { "InitTcb", 31, 1 },
6829 { "LineNumber", 24, 7 },
6831 { "EdataOut", 22, 1 },
6833 { "CdataOut", 20, 1 },
6834 { "EreadPdu", 19, 1 },
6835 { "CreadPdu", 18, 1 },
6836 { "TunnelPkt", 17, 1 },
6837 { "RcfPeerFin", 16, 1 },
6838 { "RcfReasonOut", 12, 4 },
6839 { "TxCchannel", 10, 2 },
6840 { "RcfTxChannel", 8, 2 },
6841 { "RxEchannel", 6, 2 },
6842 { "RcfRxChannel", 5, 1 },
6843 { "RcfDataOutSrdy", 4, 1 },
6845 { "RxOoDvld", 2, 1 },
6846 { "RxCongestion", 1, 1 },
6847 { "TxCongestion", 0, 1 },
6851 static struct field_desc tp_la1[] = {
6852 { "CplCmdIn", 56, 8 },
6853 { "CplCmdOut", 48, 8 },
6854 { "ESynOut", 47, 1 },
6855 { "EAckOut", 46, 1 },
6856 { "EFinOut", 45, 1 },
6857 { "ERstOut", 44, 1 },
6862 { "DataIn", 39, 1 },
6863 { "DataInVld", 38, 1 },
6865 { "RxBufEmpty", 36, 1 },
6867 { "RxFbCongestion", 34, 1 },
6868 { "TxFbCongestion", 33, 1 },
6869 { "TxPktSumSrdy", 32, 1 },
6870 { "RcfUlpType", 28, 4 },
6872 { "Ebypass", 26, 1 },
6874 { "Static0", 24, 1 },
6876 { "Cbypass", 22, 1 },
6878 { "CPktOut", 20, 1 },
6879 { "RxPagePoolFull", 18, 2 },
6880 { "RxLpbkPkt", 17, 1 },
6881 { "TxLpbkPkt", 16, 1 },
6882 { "RxVfValid", 15, 1 },
6883 { "SynLearned", 14, 1 },
6884 { "SetDelEntry", 13, 1 },
6885 { "SetInvEntry", 12, 1 },
6886 { "CpcmdDvld", 11, 1 },
6887 { "CpcmdSave", 10, 1 },
6888 { "RxPstructsFull", 8, 2 },
6889 { "EpcmdDvld", 7, 1 },
6890 { "EpcmdFlush", 6, 1 },
6891 { "EpcmdTrimPrefix", 5, 1 },
6892 { "EpcmdTrimPostfix", 4, 1 },
6893 { "ERssIp4Pkt", 3, 1 },
6894 { "ERssIp6Pkt", 2, 1 },
6895 { "ERssTcpUdpPkt", 1, 1 },
6896 { "ERssFceFipPkt", 0, 1 },
6900 static struct field_desc tp_la2[] = {
6901 { "CplCmdIn", 56, 8 },
6902 { "MpsVfVld", 55, 1 },
6909 { "DataIn", 39, 1 },
6910 { "DataInVld", 38, 1 },
6912 { "RxBufEmpty", 36, 1 },
6914 { "RxFbCongestion", 34, 1 },
6915 { "TxFbCongestion", 33, 1 },
6916 { "TxPktSumSrdy", 32, 1 },
6917 { "RcfUlpType", 28, 4 },
6919 { "Ebypass", 26, 1 },
6921 { "Static0", 24, 1 },
6923 { "Cbypass", 22, 1 },
6925 { "CPktOut", 20, 1 },
6926 { "RxPagePoolFull", 18, 2 },
6927 { "RxLpbkPkt", 17, 1 },
6928 { "TxLpbkPkt", 16, 1 },
6929 { "RxVfValid", 15, 1 },
6930 { "SynLearned", 14, 1 },
6931 { "SetDelEntry", 13, 1 },
6932 { "SetInvEntry", 12, 1 },
6933 { "CpcmdDvld", 11, 1 },
6934 { "CpcmdSave", 10, 1 },
6935 { "RxPstructsFull", 8, 2 },
6936 { "EpcmdDvld", 7, 1 },
6937 { "EpcmdFlush", 6, 1 },
6938 { "EpcmdTrimPrefix", 5, 1 },
6939 { "EpcmdTrimPostfix", 4, 1 },
6940 { "ERssIp4Pkt", 3, 1 },
6941 { "ERssIp6Pkt", 2, 1 },
6942 { "ERssTcpUdpPkt", 1, 1 },
6943 { "ERssFceFipPkt", 0, 1 },
6948 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
6951 field_desc_show(sb, *p, tp_la0);
6955 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
6959 sbuf_printf(sb, "\n");
6960 field_desc_show(sb, p[0], tp_la0);
6961 if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
6962 field_desc_show(sb, p[1], tp_la0);
6966 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
6970 sbuf_printf(sb, "\n");
6971 field_desc_show(sb, p[0], tp_la0);
6972 if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
6973 field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
6977 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
6979 struct adapter *sc = arg1;
6984 void (*show_func)(struct sbuf *, uint64_t *, int);
6986 rc = sysctl_wire_old_buffer(req, 0);
6990 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6994 buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
6996 t4_tp_read_la(sc, buf, NULL);
6999 switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
7002 show_func = tp_la_show2;
7006 show_func = tp_la_show3;
7010 show_func = tp_la_show;
7013 for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
7014 (*show_func)(sb, p, i);
7016 rc = sbuf_finish(sb);
7023 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
7025 struct adapter *sc = arg1;
7028 u64 nrate[NCHAN], orate[NCHAN];
7030 rc = sysctl_wire_old_buffer(req, 0);
7034 sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7038 t4_get_chan_txrate(sc, nrate, orate);
7039 sbuf_printf(sb, " channel 0 channel 1 channel 2 "
7041 sbuf_printf(sb, "NIC B/s: %10ju %10ju %10ju %10ju\n",
7042 nrate[0], nrate[1], nrate[2], nrate[3]);
7043 sbuf_printf(sb, "Offload B/s: %10ju %10ju %10ju %10ju",
7044 orate[0], orate[1], orate[2], orate[3]);
7046 rc = sbuf_finish(sb);
7053 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
7055 struct adapter *sc = arg1;
7060 rc = sysctl_wire_old_buffer(req, 0);
7064 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7068 buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
7071 t4_ulprx_read_la(sc, buf);
7074 sbuf_printf(sb, " Pcmd Type Message"
7076 for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
7077 sbuf_printf(sb, "\n%08x%08x %4x %08x %08x%08x%08x%08x",
7078 p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
7081 rc = sbuf_finish(sb);
7088 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
7090 struct adapter *sc = arg1;
7094 rc = sysctl_wire_old_buffer(req, 0);
7098 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7102 v = t4_read_reg(sc, A_SGE_STAT_CFG);
7103 if (G_STATSOURCE_T5(v) == 7) {
7104 if (G_STATMODE(v) == 0) {
7105 sbuf_printf(sb, "total %d, incomplete %d",
7106 t4_read_reg(sc, A_SGE_STAT_TOTAL),
7107 t4_read_reg(sc, A_SGE_STAT_MATCH));
7108 } else if (G_STATMODE(v) == 1) {
7109 sbuf_printf(sb, "total %d, data overflow %d",
7110 t4_read_reg(sc, A_SGE_STAT_TOTAL),
7111 t4_read_reg(sc, A_SGE_STAT_MATCH));
7114 rc = sbuf_finish(sb);
7122 fconf_to_mode(uint32_t fconf)
7126 mode = T4_FILTER_IPv4 | T4_FILTER_IPv6 | T4_FILTER_IP_SADDR |
7127 T4_FILTER_IP_DADDR | T4_FILTER_IP_SPORT | T4_FILTER_IP_DPORT;
7129 if (fconf & F_FRAGMENTATION)
7130 mode |= T4_FILTER_IP_FRAGMENT;
7132 if (fconf & F_MPSHITTYPE)
7133 mode |= T4_FILTER_MPS_HIT_TYPE;
7135 if (fconf & F_MACMATCH)
7136 mode |= T4_FILTER_MAC_IDX;
7138 if (fconf & F_ETHERTYPE)
7139 mode |= T4_FILTER_ETH_TYPE;
7141 if (fconf & F_PROTOCOL)
7142 mode |= T4_FILTER_IP_PROTO;
7145 mode |= T4_FILTER_IP_TOS;
7148 mode |= T4_FILTER_VLAN;
7150 if (fconf & F_VNIC_ID)
7151 mode |= T4_FILTER_VNIC;
7154 mode |= T4_FILTER_PORT;
7157 mode |= T4_FILTER_FCoE;
7163 mode_to_fconf(uint32_t mode)
7167 if (mode & T4_FILTER_IP_FRAGMENT)
7168 fconf |= F_FRAGMENTATION;
7170 if (mode & T4_FILTER_MPS_HIT_TYPE)
7171 fconf |= F_MPSHITTYPE;
7173 if (mode & T4_FILTER_MAC_IDX)
7174 fconf |= F_MACMATCH;
7176 if (mode & T4_FILTER_ETH_TYPE)
7177 fconf |= F_ETHERTYPE;
7179 if (mode & T4_FILTER_IP_PROTO)
7180 fconf |= F_PROTOCOL;
7182 if (mode & T4_FILTER_IP_TOS)
7185 if (mode & T4_FILTER_VLAN)
7188 if (mode & T4_FILTER_VNIC)
7191 if (mode & T4_FILTER_PORT)
7194 if (mode & T4_FILTER_FCoE)
7201 fspec_to_fconf(struct t4_filter_specification *fs)
7205 if (fs->val.frag || fs->mask.frag)
7206 fconf |= F_FRAGMENTATION;
7208 if (fs->val.matchtype || fs->mask.matchtype)
7209 fconf |= F_MPSHITTYPE;
7211 if (fs->val.macidx || fs->mask.macidx)
7212 fconf |= F_MACMATCH;
7214 if (fs->val.ethtype || fs->mask.ethtype)
7215 fconf |= F_ETHERTYPE;
7217 if (fs->val.proto || fs->mask.proto)
7218 fconf |= F_PROTOCOL;
7220 if (fs->val.tos || fs->mask.tos)
7223 if (fs->val.vlan_vld || fs->mask.vlan_vld)
7226 if (fs->val.vnic_vld || fs->mask.vnic_vld)
7229 if (fs->val.iport || fs->mask.iport)
7232 if (fs->val.fcoe || fs->mask.fcoe)
7239 get_filter_mode(struct adapter *sc, uint32_t *mode)
7244 rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
7249 t4_read_indirect(sc, A_TP_PIO_ADDR, A_TP_PIO_DATA, &fconf, 1,
7252 if (sc->params.tp.vlan_pri_map != fconf) {
7253 log(LOG_WARNING, "%s: cached filter mode out of sync %x %x.\n",
7254 device_get_nameunit(sc->dev), sc->params.tp.vlan_pri_map,
7258 *mode = fconf_to_mode(fconf);
7260 end_synchronized_op(sc, LOCK_HELD);
7265 set_filter_mode(struct adapter *sc, uint32_t mode)
7270 fconf = mode_to_fconf(mode);
7272 rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
7277 if (sc->tids.ftids_in_use > 0) {
7283 if (uld_active(sc, ULD_TOM)) {
7289 rc = -t4_set_filter_mode(sc, fconf);
7291 end_synchronized_op(sc, LOCK_HELD);
7295 static inline uint64_t
7296 get_filter_hits(struct adapter *sc, uint32_t fid)
7298 uint32_t mw_base, off, tcb_base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
7301 memwin_info(sc, 0, &mw_base, NULL);
7302 off = position_memwin(sc, 0,
7303 tcb_base + (fid + sc->tids.ftid_base) * TCB_SIZE);
7305 hits = t4_read_reg64(sc, mw_base + off + 16);
7306 hits = be64toh(hits);
7308 hits = t4_read_reg(sc, mw_base + off + 24);
7309 hits = be32toh(hits);
7316 get_filter(struct adapter *sc, struct t4_filter *t)
7318 int i, rc, nfilters = sc->tids.nftids;
7319 struct filter_entry *f;
7321 rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
7326 if (sc->tids.ftids_in_use == 0 || sc->tids.ftid_tab == NULL ||
7327 t->idx >= nfilters) {
7328 t->idx = 0xffffffff;
7332 f = &sc->tids.ftid_tab[t->idx];
7333 for (i = t->idx; i < nfilters; i++, f++) {
7336 t->l2tidx = f->l2t ? f->l2t->idx : 0;
7337 t->smtidx = f->smtidx;
7339 t->hits = get_filter_hits(sc, t->idx);
7341 t->hits = UINT64_MAX;
7348 t->idx = 0xffffffff;
7350 end_synchronized_op(sc, LOCK_HELD);
7355 set_filter(struct adapter *sc, struct t4_filter *t)
7357 unsigned int nfilters, nports;
7358 struct filter_entry *f;
7361 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setf");
7365 nfilters = sc->tids.nftids;
7366 nports = sc->params.nports;
7368 if (nfilters == 0) {
7373 if (!(sc->flags & FULL_INIT_DONE)) {
7378 if (t->idx >= nfilters) {
7383 /* Validate against the global filter mode */
7384 if ((sc->params.tp.vlan_pri_map | fspec_to_fconf(&t->fs)) !=
7385 sc->params.tp.vlan_pri_map) {
7390 if (t->fs.action == FILTER_SWITCH && t->fs.eport >= nports) {
7395 if (t->fs.val.iport >= nports) {
7400 /* Can't specify an iq if not steering to it */
7401 if (!t->fs.dirsteer && t->fs.iq) {
7406 /* IPv6 filter idx must be 4 aligned */
7407 if (t->fs.type == 1 &&
7408 ((t->idx & 0x3) || t->idx + 4 >= nfilters)) {
7413 if (sc->tids.ftid_tab == NULL) {
7414 KASSERT(sc->tids.ftids_in_use == 0,
7415 ("%s: no memory allocated but filters_in_use > 0",
7418 sc->tids.ftid_tab = malloc(sizeof (struct filter_entry) *
7419 nfilters, M_CXGBE, M_NOWAIT | M_ZERO);
7420 if (sc->tids.ftid_tab == NULL) {
7424 mtx_init(&sc->tids.ftid_lock, "T4 filters", 0, MTX_DEF);
7427 for (i = 0; i < 4; i++) {
7428 f = &sc->tids.ftid_tab[t->idx + i];
7430 if (f->pending || f->valid) {
7439 if (t->fs.type == 0)
7443 f = &sc->tids.ftid_tab[t->idx];
7446 rc = set_filter_wr(sc, t->idx);
7448 end_synchronized_op(sc, 0);
7451 mtx_lock(&sc->tids.ftid_lock);
7453 if (f->pending == 0) {
7454 rc = f->valid ? 0 : EIO;
7458 if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
7459 PCATCH, "t4setfw", 0)) {
7464 mtx_unlock(&sc->tids.ftid_lock);
7470 del_filter(struct adapter *sc, struct t4_filter *t)
7472 unsigned int nfilters;
7473 struct filter_entry *f;
7476 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4delf");
7480 nfilters = sc->tids.nftids;
7482 if (nfilters == 0) {
7487 if (sc->tids.ftid_tab == NULL || sc->tids.ftids_in_use == 0 ||
7488 t->idx >= nfilters) {
7493 if (!(sc->flags & FULL_INIT_DONE)) {
7498 f = &sc->tids.ftid_tab[t->idx];
7510 t->fs = f->fs; /* extra info for the caller */
7511 rc = del_filter_wr(sc, t->idx);
7515 end_synchronized_op(sc, 0);
7518 mtx_lock(&sc->tids.ftid_lock);
7520 if (f->pending == 0) {
7521 rc = f->valid ? EIO : 0;
7525 if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
7526 PCATCH, "t4delfw", 0)) {
7531 mtx_unlock(&sc->tids.ftid_lock);
7538 clear_filter(struct filter_entry *f)
7541 t4_l2t_release(f->l2t);
7543 bzero(f, sizeof (*f));
7547 set_filter_wr(struct adapter *sc, int fidx)
7549 struct filter_entry *f = &sc->tids.ftid_tab[fidx];
7550 struct fw_filter_wr *fwr;
7552 struct wrq_cookie cookie;
7554 ASSERT_SYNCHRONIZED_OP(sc);
7556 if (f->fs.newdmac || f->fs.newvlan) {
7557 /* This filter needs an L2T entry; allocate one. */
7558 f->l2t = t4_l2t_alloc_switching(sc->l2t);
7561 if (t4_l2t_set_switching(sc, f->l2t, f->fs.vlan, f->fs.eport,
7563 t4_l2t_release(f->l2t);
7569 ftid = sc->tids.ftid_base + fidx;
7571 fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
7574 bzero(fwr, sizeof(*fwr));
7576 fwr->op_pkd = htobe32(V_FW_WR_OP(FW_FILTER_WR));
7577 fwr->len16_pkd = htobe32(FW_LEN16(*fwr));
7579 htobe32(V_FW_FILTER_WR_TID(ftid) |
7580 V_FW_FILTER_WR_RQTYPE(f->fs.type) |
7581 V_FW_FILTER_WR_NOREPLY(0) |
7582 V_FW_FILTER_WR_IQ(f->fs.iq));
7583 fwr->del_filter_to_l2tix =
7584 htobe32(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
7585 V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
7586 V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
7587 V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
7588 V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
7589 V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
7590 V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
7591 V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
7592 V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
7593 f->fs.newvlan == VLAN_REWRITE) |
7594 V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
7595 f->fs.newvlan == VLAN_REWRITE) |
7596 V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
7597 V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
7598 V_FW_FILTER_WR_PRIO(f->fs.prio) |
7599 V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
7600 fwr->ethtype = htobe16(f->fs.val.ethtype);
7601 fwr->ethtypem = htobe16(f->fs.mask.ethtype);
7602 fwr->frag_to_ovlan_vldm =
7603 (V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
7604 V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
7605 V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.vlan_vld) |
7606 V_FW_FILTER_WR_OVLAN_VLD(f->fs.val.vnic_vld) |
7607 V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.vlan_vld) |
7608 V_FW_FILTER_WR_OVLAN_VLDM(f->fs.mask.vnic_vld));
7610 fwr->rx_chan_rx_rpl_iq = htobe16(V_FW_FILTER_WR_RX_CHAN(0) |
7611 V_FW_FILTER_WR_RX_RPL_IQ(sc->sge.fwq.abs_id));
7612 fwr->maci_to_matchtypem =
7613 htobe32(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
7614 V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
7615 V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
7616 V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
7617 V_FW_FILTER_WR_PORT(f->fs.val.iport) |
7618 V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
7619 V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
7620 V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
7621 fwr->ptcl = f->fs.val.proto;
7622 fwr->ptclm = f->fs.mask.proto;
7623 fwr->ttyp = f->fs.val.tos;
7624 fwr->ttypm = f->fs.mask.tos;
7625 fwr->ivlan = htobe16(f->fs.val.vlan);
7626 fwr->ivlanm = htobe16(f->fs.mask.vlan);
7627 fwr->ovlan = htobe16(f->fs.val.vnic);
7628 fwr->ovlanm = htobe16(f->fs.mask.vnic);
7629 bcopy(f->fs.val.dip, fwr->lip, sizeof (fwr->lip));
7630 bcopy(f->fs.mask.dip, fwr->lipm, sizeof (fwr->lipm));
7631 bcopy(f->fs.val.sip, fwr->fip, sizeof (fwr->fip));
7632 bcopy(f->fs.mask.sip, fwr->fipm, sizeof (fwr->fipm));
7633 fwr->lp = htobe16(f->fs.val.dport);
7634 fwr->lpm = htobe16(f->fs.mask.dport);
7635 fwr->fp = htobe16(f->fs.val.sport);
7636 fwr->fpm = htobe16(f->fs.mask.sport);
7638 bcopy(f->fs.smac, fwr->sma, sizeof (fwr->sma));
7641 sc->tids.ftids_in_use++;
7643 commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
7648 del_filter_wr(struct adapter *sc, int fidx)
7650 struct filter_entry *f = &sc->tids.ftid_tab[fidx];
7651 struct fw_filter_wr *fwr;
7653 struct wrq_cookie cookie;
7655 ftid = sc->tids.ftid_base + fidx;
7657 fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
7660 bzero(fwr, sizeof (*fwr));
7662 t4_mk_filtdelwr(ftid, fwr, sc->sge.fwq.abs_id);
7665 commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
7670 t4_filter_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
7672 struct adapter *sc = iq->adapter;
7673 const struct cpl_set_tcb_rpl *rpl = (const void *)(rss + 1);
7674 unsigned int idx = GET_TID(rpl);
7676 struct filter_entry *f;
7678 KASSERT(m == NULL, ("%s: payload with opcode %02x", __func__,
7681 if (is_ftid(sc, idx)) {
7683 idx -= sc->tids.ftid_base;
7684 f = &sc->tids.ftid_tab[idx];
7685 rc = G_COOKIE(rpl->cookie);
7687 mtx_lock(&sc->tids.ftid_lock);
7688 if (rc == FW_FILTER_WR_FLT_ADDED) {
7689 KASSERT(f->pending, ("%s: filter[%u] isn't pending.",
7691 f->smtidx = (be64toh(rpl->oldval) >> 24) & 0xff;
7692 f->pending = 0; /* asynchronous setup completed */
7695 if (rc != FW_FILTER_WR_FLT_DELETED) {
7696 /* Add or delete failed, display an error */
7698 "filter %u setup failed with error %u\n",
7703 sc->tids.ftids_in_use--;
7705 wakeup(&sc->tids.ftid_tab);
7706 mtx_unlock(&sc->tids.ftid_lock);
7713 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
7717 if (cntxt->cid > M_CTXTQID)
7720 if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
7721 cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
7724 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
7728 if (sc->flags & FW_OK) {
7729 rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
7736 * Read via firmware failed or wasn't even attempted. Read directly via
7739 rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
7741 end_synchronized_op(sc, 0);
7746 load_fw(struct adapter *sc, struct t4_data *fw)
7751 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
7755 if (sc->flags & FULL_INIT_DONE) {
7760 fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
7761 if (fw_data == NULL) {
7766 rc = copyin(fw->data, fw_data, fw->len);
7768 rc = -t4_load_fw(sc, fw_data, fw->len);
7770 free(fw_data, M_CXGBE);
7772 end_synchronized_op(sc, 0);
7777 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
7779 uint32_t addr, off, remaining, i, n;
7781 uint32_t mw_base, mw_aperture;
7785 rc = validate_mem_range(sc, mr->addr, mr->len);
7789 memwin_info(sc, win, &mw_base, &mw_aperture);
7790 buf = b = malloc(min(mr->len, mw_aperture), M_CXGBE, M_WAITOK);
7792 remaining = mr->len;
7793 dst = (void *)mr->data;
7796 off = position_memwin(sc, win, addr);
7798 /* number of bytes that we'll copy in the inner loop */
7799 n = min(remaining, mw_aperture - off);
7800 for (i = 0; i < n; i += 4)
7801 *b++ = t4_read_reg(sc, mw_base + off + i);
7803 rc = copyout(buf, dst, n);
7818 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
7822 if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
7825 if (i2cd->len > sizeof(i2cd->data))
7828 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
7831 rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
7832 i2cd->offset, i2cd->len, &i2cd->data[0]);
7833 end_synchronized_op(sc, 0);
7839 in_range(int val, int lo, int hi)
7842 return (val < 0 || (val <= hi && val >= lo));
7846 set_sched_class(struct adapter *sc, struct t4_sched_params *p)
7848 int fw_subcmd, fw_type, rc;
7850 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setsc");
7854 if (!(sc->flags & FULL_INIT_DONE)) {
7860 * Translate the cxgbetool parameters into T4 firmware parameters. (The
7861 * sub-command and type are in common locations.)
7863 if (p->subcmd == SCHED_CLASS_SUBCMD_CONFIG)
7864 fw_subcmd = FW_SCHED_SC_CONFIG;
7865 else if (p->subcmd == SCHED_CLASS_SUBCMD_PARAMS)
7866 fw_subcmd = FW_SCHED_SC_PARAMS;
7871 if (p->type == SCHED_CLASS_TYPE_PACKET)
7872 fw_type = FW_SCHED_TYPE_PKTSCHED;
7878 if (fw_subcmd == FW_SCHED_SC_CONFIG) {
7879 /* Vet our parameters ..*/
7880 if (p->u.config.minmax < 0) {
7885 /* And pass the request to the firmware ...*/
7886 rc = -t4_sched_config(sc, fw_type, p->u.config.minmax, 1);
7890 if (fw_subcmd == FW_SCHED_SC_PARAMS) {
7896 if (p->u.params.level == SCHED_CLASS_LEVEL_CL_RL)
7897 fw_level = FW_SCHED_PARAMS_LEVEL_CL_RL;
7898 else if (p->u.params.level == SCHED_CLASS_LEVEL_CL_WRR)
7899 fw_level = FW_SCHED_PARAMS_LEVEL_CL_WRR;
7900 else if (p->u.params.level == SCHED_CLASS_LEVEL_CH_RL)
7901 fw_level = FW_SCHED_PARAMS_LEVEL_CH_RL;
7907 if (p->u.params.mode == SCHED_CLASS_MODE_CLASS)
7908 fw_mode = FW_SCHED_PARAMS_MODE_CLASS;
7909 else if (p->u.params.mode == SCHED_CLASS_MODE_FLOW)
7910 fw_mode = FW_SCHED_PARAMS_MODE_FLOW;
7916 if (p->u.params.rateunit == SCHED_CLASS_RATEUNIT_BITS)
7917 fw_rateunit = FW_SCHED_PARAMS_UNIT_BITRATE;
7918 else if (p->u.params.rateunit == SCHED_CLASS_RATEUNIT_PKTS)
7919 fw_rateunit = FW_SCHED_PARAMS_UNIT_PKTRATE;
7925 if (p->u.params.ratemode == SCHED_CLASS_RATEMODE_REL)
7926 fw_ratemode = FW_SCHED_PARAMS_RATE_REL;
7927 else if (p->u.params.ratemode == SCHED_CLASS_RATEMODE_ABS)
7928 fw_ratemode = FW_SCHED_PARAMS_RATE_ABS;
7934 /* Vet our parameters ... */
7935 if (!in_range(p->u.params.channel, 0, 3) ||
7936 !in_range(p->u.params.cl, 0, is_t4(sc) ? 15 : 16) ||
7937 !in_range(p->u.params.minrate, 0, 10000000) ||
7938 !in_range(p->u.params.maxrate, 0, 10000000) ||
7939 !in_range(p->u.params.weight, 0, 100)) {
7945 * Translate any unset parameters into the firmware's
7946 * nomenclature and/or fail the call if the parameters
7949 if (p->u.params.rateunit < 0 || p->u.params.ratemode < 0 ||
7950 p->u.params.channel < 0 || p->u.params.cl < 0) {
7954 if (p->u.params.minrate < 0)
7955 p->u.params.minrate = 0;
7956 if (p->u.params.maxrate < 0) {
7957 if (p->u.params.level == SCHED_CLASS_LEVEL_CL_RL ||
7958 p->u.params.level == SCHED_CLASS_LEVEL_CH_RL) {
7962 p->u.params.maxrate = 0;
7964 if (p->u.params.weight < 0) {
7965 if (p->u.params.level == SCHED_CLASS_LEVEL_CL_WRR) {
7969 p->u.params.weight = 0;
7971 if (p->u.params.pktsize < 0) {
7972 if (p->u.params.level == SCHED_CLASS_LEVEL_CL_RL ||
7973 p->u.params.level == SCHED_CLASS_LEVEL_CH_RL) {
7977 p->u.params.pktsize = 0;
7980 /* See what the firmware thinks of the request ... */
7981 rc = -t4_sched_params(sc, fw_type, fw_level, fw_mode,
7982 fw_rateunit, fw_ratemode, p->u.params.channel,
7983 p->u.params.cl, p->u.params.minrate, p->u.params.maxrate,
7984 p->u.params.weight, p->u.params.pktsize, 1);
7990 end_synchronized_op(sc, 0);
7995 set_sched_queue(struct adapter *sc, struct t4_sched_queue *p)
7997 struct port_info *pi = NULL;
7998 struct sge_txq *txq;
7999 uint32_t fw_mnem, fw_queue, fw_class;
8002 rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setsq");
8006 if (!(sc->flags & FULL_INIT_DONE)) {
8011 if (p->port >= sc->params.nports) {
8016 pi = sc->port[p->port];
8017 if (!in_range(p->queue, 0, pi->ntxq - 1) || !in_range(p->cl, 0, 7)) {
8023 * Create a template for the FW_PARAMS_CMD mnemonic and value (TX
8024 * Scheduling Class in this case).
8026 fw_mnem = (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
8027 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_EQ_SCHEDCLASS_ETH));
8028 fw_class = p->cl < 0 ? 0xffffffff : p->cl;
8031 * If op.queue is non-negative, then we're only changing the scheduling
8032 * on a single specified TX queue.
8034 if (p->queue >= 0) {
8035 txq = &sc->sge.txq[pi->first_txq + p->queue];
8036 fw_queue = (fw_mnem | V_FW_PARAMS_PARAM_YZ(txq->eq.cntxt_id));
8037 rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &fw_queue,
8043 * Change the scheduling on all the TX queues for the
8046 for_each_txq(pi, i, txq) {
8047 fw_queue = (fw_mnem | V_FW_PARAMS_PARAM_YZ(txq->eq.cntxt_id));
8048 rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &fw_queue,
8056 end_synchronized_op(sc, 0);
8061 t4_os_find_pci_capability(struct adapter *sc, int cap)
8065 return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
8069 t4_os_pci_save_state(struct adapter *sc)
8072 struct pci_devinfo *dinfo;
8075 dinfo = device_get_ivars(dev);
8077 pci_cfg_save(dev, dinfo, 0);
8082 t4_os_pci_restore_state(struct adapter *sc)
8085 struct pci_devinfo *dinfo;
8088 dinfo = device_get_ivars(dev);
8090 pci_cfg_restore(dev, dinfo);
8095 t4_os_portmod_changed(const struct adapter *sc, int idx)
8097 struct port_info *pi = sc->port[idx];
8098 static const char *mod_str[] = {
8099 NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
8102 build_medialist(pi, &pi->media);
8104 build_medialist(pi, &pi->nm_media);
8107 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
8108 if_printf(pi->ifp, "transceiver unplugged.\n");
8109 else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
8110 if_printf(pi->ifp, "unknown transceiver inserted.\n");
8111 else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
8112 if_printf(pi->ifp, "unsupported transceiver inserted.\n");
8113 else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
8114 if_printf(pi->ifp, "%s transceiver inserted.\n",
8115 mod_str[pi->mod_type]);
8117 if_printf(pi->ifp, "transceiver (type %d) inserted.\n",
8123 t4_os_link_changed(struct adapter *sc, int idx, int link_stat, int reason)
8125 struct port_info *pi = sc->port[idx];
8126 struct ifnet *ifp = pi->ifp;
8130 ifp->if_baudrate = IF_Mbps(pi->link_cfg.speed);
8131 if_link_state_change(ifp, LINK_STATE_UP);
8134 pi->linkdnrc = reason;
8135 if_link_state_change(ifp, LINK_STATE_DOWN);
8140 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
8144 sx_slock(&t4_list_lock);
8145 SLIST_FOREACH(sc, &t4_list, link) {
8147 * func should not make any assumptions about what state sc is
8148 * in - the only guarantee is that sc->sc_lock is a valid lock.
8152 sx_sunlock(&t4_list_lock);
8156 t4_open(struct cdev *dev, int flags, int type, struct thread *td)
8162 t4_close(struct cdev *dev, int flags, int type, struct thread *td)
8168 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
8172 struct adapter *sc = dev->si_drv1;
8174 rc = priv_check(td, PRIV_DRIVER);
8179 case CHELSIO_T4_GETREG: {
8180 struct t4_reg *edata = (struct t4_reg *)data;
8182 if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
8185 if (edata->size == 4)
8186 edata->val = t4_read_reg(sc, edata->addr);
8187 else if (edata->size == 8)
8188 edata->val = t4_read_reg64(sc, edata->addr);
8194 case CHELSIO_T4_SETREG: {
8195 struct t4_reg *edata = (struct t4_reg *)data;
8197 if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
8200 if (edata->size == 4) {
8201 if (edata->val & 0xffffffff00000000)
8203 t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
8204 } else if (edata->size == 8)
8205 t4_write_reg64(sc, edata->addr, edata->val);
8210 case CHELSIO_T4_REGDUMP: {
8211 struct t4_regdump *regs = (struct t4_regdump *)data;
8212 int reglen = is_t4(sc) ? T4_REGDUMP_SIZE : T5_REGDUMP_SIZE;
8215 if (regs->len < reglen) {
8216 regs->len = reglen; /* hint to the caller */
8221 buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
8222 t4_get_regs(sc, regs, buf);
8223 rc = copyout(buf, regs->data, reglen);
8227 case CHELSIO_T4_GET_FILTER_MODE:
8228 rc = get_filter_mode(sc, (uint32_t *)data);
8230 case CHELSIO_T4_SET_FILTER_MODE:
8231 rc = set_filter_mode(sc, *(uint32_t *)data);
8233 case CHELSIO_T4_GET_FILTER:
8234 rc = get_filter(sc, (struct t4_filter *)data);
8236 case CHELSIO_T4_SET_FILTER:
8237 rc = set_filter(sc, (struct t4_filter *)data);
8239 case CHELSIO_T4_DEL_FILTER:
8240 rc = del_filter(sc, (struct t4_filter *)data);
8242 case CHELSIO_T4_GET_SGE_CONTEXT:
8243 rc = get_sge_context(sc, (struct t4_sge_context *)data);
8245 case CHELSIO_T4_LOAD_FW:
8246 rc = load_fw(sc, (struct t4_data *)data);
8248 case CHELSIO_T4_GET_MEM:
8249 rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
8251 case CHELSIO_T4_GET_I2C:
8252 rc = read_i2c(sc, (struct t4_i2c_data *)data);
8254 case CHELSIO_T4_CLEAR_STATS: {
8256 u_int port_id = *(uint32_t *)data;
8257 struct port_info *pi;
8259 if (port_id >= sc->params.nports)
8261 pi = sc->port[port_id];
8264 t4_clr_port_stats(sc, pi->tx_chan);
8265 pi->tx_parse_error = 0;
8267 if (pi->flags & PORT_INIT_DONE) {
8268 struct sge_rxq *rxq;
8269 struct sge_txq *txq;
8270 struct sge_wrq *wrq;
8272 for_each_rxq(pi, i, rxq) {
8273 #if defined(INET) || defined(INET6)
8274 rxq->lro.lro_queued = 0;
8275 rxq->lro.lro_flushed = 0;
8278 rxq->vlan_extraction = 0;
8281 for_each_txq(pi, i, txq) {
8284 txq->vlan_insertion = 0;
8288 txq->txpkts0_wrs = 0;
8289 txq->txpkts1_wrs = 0;
8290 txq->txpkts0_pkts = 0;
8291 txq->txpkts1_pkts = 0;
8292 mp_ring_reset_stats(txq->r);
8296 /* nothing to clear for each ofld_rxq */
8298 for_each_ofld_txq(pi, i, wrq) {
8299 wrq->tx_wrs_direct = 0;
8300 wrq->tx_wrs_copied = 0;
8303 wrq = &sc->sge.ctrlq[pi->port_id];
8304 wrq->tx_wrs_direct = 0;
8305 wrq->tx_wrs_copied = 0;
8309 case CHELSIO_T4_SCHED_CLASS:
8310 rc = set_sched_class(sc, (struct t4_sched_params *)data);
8312 case CHELSIO_T4_SCHED_QUEUE:
8313 rc = set_sched_queue(sc, (struct t4_sched_queue *)data);
8315 case CHELSIO_T4_GET_TRACER:
8316 rc = t4_get_tracer(sc, (struct t4_tracer *)data);
8318 case CHELSIO_T4_SET_TRACER:
8319 rc = t4_set_tracer(sc, (struct t4_tracer *)data);
8330 t4_iscsi_init(struct ifnet *ifp, unsigned int tag_mask,
8331 const unsigned int *pgsz_order)
8333 struct port_info *pi = ifp->if_softc;
8334 struct adapter *sc = pi->adapter;
8336 t4_write_reg(sc, A_ULP_RX_ISCSI_TAGMASK, tag_mask);
8337 t4_write_reg(sc, A_ULP_RX_ISCSI_PSZ, V_HPZ0(pgsz_order[0]) |
8338 V_HPZ1(pgsz_order[1]) | V_HPZ2(pgsz_order[2]) |
8339 V_HPZ3(pgsz_order[3]));
8343 toe_capability(struct port_info *pi, int enable)
8346 struct adapter *sc = pi->adapter;
8348 ASSERT_SYNCHRONIZED_OP(sc);
8350 if (!is_offload(sc))
8355 * We need the port's queues around so that we're able to send
8356 * and receive CPLs to/from the TOE even if the ifnet for this
8357 * port has never been UP'd administratively.
8359 if (!(pi->flags & PORT_INIT_DONE)) {
8360 rc = cxgbe_init_synchronized(pi);
8365 if (isset(&sc->offload_map, pi->port_id))
8368 if (!uld_active(sc, ULD_TOM)) {
8369 rc = t4_activate_uld(sc, ULD_TOM);
8372 "You must kldload t4_tom.ko before trying "
8373 "to enable TOE on a cxgbe interface.\n");
8377 KASSERT(sc->tom_softc != NULL,
8378 ("%s: TOM activated but softc NULL", __func__));
8379 KASSERT(uld_active(sc, ULD_TOM),
8380 ("%s: TOM activated but flag not set", __func__));
8383 /* Activate iWARP and iSCSI too, if the modules are loaded. */
8384 if (!uld_active(sc, ULD_IWARP))
8385 (void) t4_activate_uld(sc, ULD_IWARP);
8386 if (!uld_active(sc, ULD_ISCSI))
8387 (void) t4_activate_uld(sc, ULD_ISCSI);
8389 setbit(&sc->offload_map, pi->port_id);
8391 if (!isset(&sc->offload_map, pi->port_id))
8394 KASSERT(uld_active(sc, ULD_TOM),
8395 ("%s: TOM never initialized?", __func__));
8396 clrbit(&sc->offload_map, pi->port_id);
8403 * Add an upper layer driver to the global list.
8406 t4_register_uld(struct uld_info *ui)
8411 sx_xlock(&t4_uld_list_lock);
8412 SLIST_FOREACH(u, &t4_uld_list, link) {
8413 if (u->uld_id == ui->uld_id) {
8419 SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
8422 sx_xunlock(&t4_uld_list_lock);
8427 t4_unregister_uld(struct uld_info *ui)
8432 sx_xlock(&t4_uld_list_lock);
8434 SLIST_FOREACH(u, &t4_uld_list, link) {
8436 if (ui->refcount > 0) {
8441 SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
8447 sx_xunlock(&t4_uld_list_lock);
8452 t4_activate_uld(struct adapter *sc, int id)
8455 struct uld_info *ui;
8457 ASSERT_SYNCHRONIZED_OP(sc);
8459 if (id < 0 || id > ULD_MAX)
8461 rc = EAGAIN; /* kldoad the module with this ULD and try again. */
8463 sx_slock(&t4_uld_list_lock);
8465 SLIST_FOREACH(ui, &t4_uld_list, link) {
8466 if (ui->uld_id == id) {
8467 if (!(sc->flags & FULL_INIT_DONE)) {
8468 rc = adapter_full_init(sc);
8473 rc = ui->activate(sc);
8475 setbit(&sc->active_ulds, id);
8482 sx_sunlock(&t4_uld_list_lock);
8488 t4_deactivate_uld(struct adapter *sc, int id)
8491 struct uld_info *ui;
8493 ASSERT_SYNCHRONIZED_OP(sc);
8495 if (id < 0 || id > ULD_MAX)
8499 sx_slock(&t4_uld_list_lock);
8501 SLIST_FOREACH(ui, &t4_uld_list, link) {
8502 if (ui->uld_id == id) {
8503 rc = ui->deactivate(sc);
8505 clrbit(&sc->active_ulds, id);
8512 sx_sunlock(&t4_uld_list_lock);
8518 uld_active(struct adapter *sc, int uld_id)
8521 MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
8523 return (isset(&sc->active_ulds, uld_id));
8528 * Come up with reasonable defaults for some of the tunables, provided they're
8529 * not set by the user (in which case we'll use the values as is).
8532 tweak_tunables(void)
8534 int nc = mp_ncpus; /* our snapshot of the number of CPUs */
8536 if (t4_ntxq10g < 1) {
8538 t4_ntxq10g = rss_getnumbuckets();
8540 t4_ntxq10g = min(nc, NTXQ_10G);
8544 if (t4_ntxq1g < 1) {
8546 /* XXX: way too many for 1GbE? */
8547 t4_ntxq1g = rss_getnumbuckets();
8549 t4_ntxq1g = min(nc, NTXQ_1G);
8553 if (t4_nrxq10g < 1) {
8555 t4_nrxq10g = rss_getnumbuckets();
8557 t4_nrxq10g = min(nc, NRXQ_10G);
8561 if (t4_nrxq1g < 1) {
8563 /* XXX: way too many for 1GbE? */
8564 t4_nrxq1g = rss_getnumbuckets();
8566 t4_nrxq1g = min(nc, NRXQ_1G);
8571 if (t4_nofldtxq10g < 1)
8572 t4_nofldtxq10g = min(nc, NOFLDTXQ_10G);
8574 if (t4_nofldtxq1g < 1)
8575 t4_nofldtxq1g = min(nc, NOFLDTXQ_1G);
8577 if (t4_nofldrxq10g < 1)
8578 t4_nofldrxq10g = min(nc, NOFLDRXQ_10G);
8580 if (t4_nofldrxq1g < 1)
8581 t4_nofldrxq1g = min(nc, NOFLDRXQ_1G);
8583 if (t4_toecaps_allowed == -1)
8584 t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
8586 if (t4_toecaps_allowed == -1)
8587 t4_toecaps_allowed = 0;
8591 if (t4_nnmtxq10g < 1)
8592 t4_nnmtxq10g = min(nc, NNMTXQ_10G);
8594 if (t4_nnmtxq1g < 1)
8595 t4_nnmtxq1g = min(nc, NNMTXQ_1G);
8597 if (t4_nnmrxq10g < 1)
8598 t4_nnmrxq10g = min(nc, NNMRXQ_10G);
8600 if (t4_nnmrxq1g < 1)
8601 t4_nnmrxq1g = min(nc, NNMRXQ_1G);
8604 if (t4_tmr_idx_10g < 0 || t4_tmr_idx_10g >= SGE_NTIMERS)
8605 t4_tmr_idx_10g = TMR_IDX_10G;
8607 if (t4_pktc_idx_10g < -1 || t4_pktc_idx_10g >= SGE_NCOUNTERS)
8608 t4_pktc_idx_10g = PKTC_IDX_10G;
8610 if (t4_tmr_idx_1g < 0 || t4_tmr_idx_1g >= SGE_NTIMERS)
8611 t4_tmr_idx_1g = TMR_IDX_1G;
8613 if (t4_pktc_idx_1g < -1 || t4_pktc_idx_1g >= SGE_NCOUNTERS)
8614 t4_pktc_idx_1g = PKTC_IDX_1G;
8616 if (t4_qsize_txq < 128)
8619 if (t4_qsize_rxq < 128)
8621 while (t4_qsize_rxq & 7)
8624 t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
8627 static struct sx mlu; /* mod load unload */
8628 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
8631 mod_event(module_t mod, int cmd, void *arg)
8634 static int loaded = 0;
8639 if (loaded++ == 0) {
8641 sx_init(&t4_list_lock, "T4/T5 adapters");
8642 SLIST_INIT(&t4_list);
8644 sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
8645 SLIST_INIT(&t4_uld_list);
8647 t4_tracer_modload();
8655 if (--loaded == 0) {
8658 sx_slock(&t4_list_lock);
8659 if (!SLIST_EMPTY(&t4_list)) {
8661 sx_sunlock(&t4_list_lock);
8665 sx_slock(&t4_uld_list_lock);
8666 if (!SLIST_EMPTY(&t4_uld_list)) {
8668 sx_sunlock(&t4_uld_list_lock);
8669 sx_sunlock(&t4_list_lock);
8674 while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
8675 uprintf("%ju clusters with custom free routine "
8676 "still is use.\n", t4_sge_extfree_refs());
8677 pause("t4unload", 2 * hz);
8680 sx_sunlock(&t4_uld_list_lock);
8682 sx_sunlock(&t4_list_lock);
8684 if (t4_sge_extfree_refs() == 0) {
8685 t4_tracer_modunload();
8687 sx_destroy(&t4_uld_list_lock);
8689 sx_destroy(&t4_list_lock);
8694 loaded++; /* undo earlier decrement */
8705 static devclass_t t4_devclass, t5_devclass;
8706 static devclass_t cxgbe_devclass, cxl_devclass;
8708 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
8709 MODULE_VERSION(t4nex, 1);
8710 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
8712 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
8713 #endif /* DEV_NETMAP */
8716 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
8717 MODULE_VERSION(t5nex, 1);
8718 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
8720 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
8721 #endif /* DEV_NETMAP */
8723 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
8724 MODULE_VERSION(cxgbe, 1);
8726 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
8727 MODULE_VERSION(cxl, 1);