sys/dev/cxgbe/t4_main.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (c) 2011 Chelsio Communications, Inc.
   5  * All rights reserved.
   6  * Written by: Navdeep Parhar <np@FreeBSD.org>
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  */
  29
  30 #include <sys/cdefs.h>
  31 __FBSDID("$FreeBSD$");
  32
  33 #include "opt_ddb.h"
  34 #include "opt_inet.h"
  35 #include "opt_inet6.h"
  36 #include "opt_ratelimit.h"
  37 #include "opt_rss.h"
  38
  39 #include <sys/param.h>
  40 #include <sys/conf.h>
  41 #include <sys/priv.h>
  42 #include <sys/kernel.h>
  43 #include <sys/bus.h>
  44 #include <sys/module.h>
  45 #include <sys/malloc.h>
  46 #include <sys/queue.h>
  47 #include <sys/taskqueue.h>
  48 #include <sys/pciio.h>
  49 #include <dev/pci/pcireg.h>
  50 #include <dev/pci/pcivar.h>
  51 #include <dev/pci/pci_private.h>
  52 #include <sys/firmware.h>
  53 #include <sys/sbuf.h>
  54 #include <sys/smp.h>
  55 #include <sys/socket.h>
  56 #include <sys/sockio.h>
  57 #include <sys/sysctl.h>
  58 #include <net/ethernet.h>
  59 #include <net/if.h>
  60 #include <net/if_types.h>
  61 #include <net/if_dl.h>
  62 #include <net/if_vlan_var.h>
  63 #ifdef RSS
  64 #include <net/rss_config.h>
  65 #endif
  66 #if defined(__i386__) || defined(__amd64__)
  67 #include <machine/md_var.h>
  68 #include <machine/cputypes.h>
  69 #include <vm/vm.h>
  70 #include <vm/pmap.h>
  71 #endif
  72 #include <crypto/rijndael/rijndael.h>
  73 #ifdef DDB
  74 #include <ddb/ddb.h>
  75 #include <ddb/db_lex.h>
  76 #endif
  77
  78 #include "common/common.h"
  79 #include "common/t4_msg.h"
  80 #include "common/t4_regs.h"
  81 #include "common/t4_regs_values.h"
  82 #include "cudbg/cudbg.h"
  83 #include "t4_ioctl.h"
  84 #include "t4_l2t.h"
  85 #include "t4_mp_ring.h"
  86 #include "t4_if.h"
  87 #include "t4_smt.h"
  88
  89 /* T4 bus driver interface */
  90 static int t4_probe(device_t);
  91 static int t4_attach(device_t);
  92 static int t4_detach(device_t);
  93 static int t4_ready(device_t);
  94 static int t4_read_port_device(device_t, int, device_t *);
  95 static device_method_t t4_methods[] = {
  96         DEVMETHOD(device_probe,         t4_probe),
  97         DEVMETHOD(device_attach,        t4_attach),
  98         DEVMETHOD(device_detach,        t4_detach),
  99
 100         DEVMETHOD(t4_is_main_ready,     t4_ready),
 101         DEVMETHOD(t4_read_port_device,  t4_read_port_device),
 102
 103         DEVMETHOD_END
 104 };
 105 static driver_t t4_driver = {
 106         "t4nex",
 107         t4_methods,
 108         sizeof(struct adapter)
 109 };
 110
 111
 112 /* T4 port (cxgbe) interface */
 113 static int cxgbe_probe(device_t);
 114 static int cxgbe_attach(device_t);
 115 static int cxgbe_detach(device_t);
 116 device_method_t cxgbe_methods[] = {
 117         DEVMETHOD(device_probe,         cxgbe_probe),
 118         DEVMETHOD(device_attach,        cxgbe_attach),
 119         DEVMETHOD(device_detach,        cxgbe_detach),
 120         { 0, 0 }
 121 };
 122 static driver_t cxgbe_driver = {
 123         "cxgbe",
 124         cxgbe_methods,
 125         sizeof(struct port_info)
 126 };
 127
 128 /* T4 VI (vcxgbe) interface */
 129 static int vcxgbe_probe(device_t);
 130 static int vcxgbe_attach(device_t);
 131 static int vcxgbe_detach(device_t);
 132 static device_method_t vcxgbe_methods[] = {
 133         DEVMETHOD(device_probe,         vcxgbe_probe),
 134         DEVMETHOD(device_attach,        vcxgbe_attach),
 135         DEVMETHOD(device_detach,        vcxgbe_detach),
 136         { 0, 0 }
 137 };
 138 static driver_t vcxgbe_driver = {
 139         "vcxgbe",
 140         vcxgbe_methods,
 141         sizeof(struct vi_info)
 142 };
 143
 144 static d_ioctl_t t4_ioctl;
 145
 146 static struct cdevsw t4_cdevsw = {
 147        .d_version = D_VERSION,
 148        .d_ioctl = t4_ioctl,
 149        .d_name = "t4nex",
 150 };
 151
 152 /* T5 bus driver interface */
 153 static int t5_probe(device_t);
 154 static device_method_t t5_methods[] = {
 155         DEVMETHOD(device_probe,         t5_probe),
 156         DEVMETHOD(device_attach,        t4_attach),
 157         DEVMETHOD(device_detach,        t4_detach),
 158
 159         DEVMETHOD(t4_is_main_ready,     t4_ready),
 160         DEVMETHOD(t4_read_port_device,  t4_read_port_device),
 161
 162         DEVMETHOD_END
 163 };
 164 static driver_t t5_driver = {
 165         "t5nex",
 166         t5_methods,
 167         sizeof(struct adapter)
 168 };
 169
 170
 171 /* T5 port (cxl) interface */
 172 static driver_t cxl_driver = {
 173         "cxl",
 174         cxgbe_methods,
 175         sizeof(struct port_info)
 176 };
 177
 178 /* T5 VI (vcxl) interface */
 179 static driver_t vcxl_driver = {
 180         "vcxl",
 181         vcxgbe_methods,
 182         sizeof(struct vi_info)
 183 };
 184
 185 /* T6 bus driver interface */
 186 static int t6_probe(device_t);
 187 static device_method_t t6_methods[] = {
 188         DEVMETHOD(device_probe,         t6_probe),
 189         DEVMETHOD(device_attach,        t4_attach),
 190         DEVMETHOD(device_detach,        t4_detach),
 191
 192         DEVMETHOD(t4_is_main_ready,     t4_ready),
 193         DEVMETHOD(t4_read_port_device,  t4_read_port_device),
 194
 195         DEVMETHOD_END
 196 };
 197 static driver_t t6_driver = {
 198         "t6nex",
 199         t6_methods,
 200         sizeof(struct adapter)
 201 };
 202
 203
 204 /* T6 port (cc) interface */
 205 static driver_t cc_driver = {
 206         "cc",
 207         cxgbe_methods,
 208         sizeof(struct port_info)
 209 };
 210
 211 /* T6 VI (vcc) interface */
 212 static driver_t vcc_driver = {
 213         "vcc",
 214         vcxgbe_methods,
 215         sizeof(struct vi_info)
 216 };
 217
 218 /* ifnet + media interface */
 219 static void cxgbe_init(void *);
 220 static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
 221 static int cxgbe_transmit(struct ifnet *, struct mbuf *);
 222 static void cxgbe_qflush(struct ifnet *);
 223 static int cxgbe_media_change(struct ifnet *);
 224 static void cxgbe_media_status(struct ifnet *, struct ifmediareq *);
 225
 226 MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
 227
 228 /*
 229  * Correct lock order when you need to acquire multiple locks is t4_list_lock,
 230  * then ADAPTER_LOCK, then t4_uld_list_lock.
 231  */
 232 static struct sx t4_list_lock;
 233 SLIST_HEAD(, adapter) t4_list;
 234 #ifdef TCP_OFFLOAD
 235 static struct sx t4_uld_list_lock;
 236 SLIST_HEAD(, uld_info) t4_uld_list;
 237 #endif
 238
 239 /*
 240  * Tunables.  See tweak_tunables() too.
 241  *
 242  * Each tunable is set to a default value here if it's known at compile-time.
 243  * Otherwise it is set to -n as an indication to tweak_tunables() that it should
 244  * provide a reasonable default (upto n) when the driver is loaded.
 245  *
 246  * Tunables applicable to both T4 and T5 are under hw.cxgbe.  Those specific to
 247  * T5 are under hw.cxl.
 248  */
 249
 250 /*
 251  * Number of queues for tx and rx, NIC and offload.
 252  */
 253 #define NTXQ 16
 254 int t4_ntxq = -NTXQ;
 255 TUNABLE_INT("hw.cxgbe.ntxq", &t4_ntxq);
 256 TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq);      /* Old name, undocumented */
 257
 258 #define NRXQ 8
 259 int t4_nrxq = -NRXQ;
 260 TUNABLE_INT("hw.cxgbe.nrxq", &t4_nrxq);
 261 TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq);      /* Old name, undocumented */
 262
 263 #define NTXQ_VI 1
 264 static int t4_ntxq_vi = -NTXQ_VI;
 265 TUNABLE_INT("hw.cxgbe.ntxq_vi", &t4_ntxq_vi);
 266
 267 #define NRXQ_VI 1
 268 static int t4_nrxq_vi = -NRXQ_VI;
 269 TUNABLE_INT("hw.cxgbe.nrxq_vi", &t4_nrxq_vi);
 270
 271 static int t4_rsrv_noflowq = 0;
 272 TUNABLE_INT("hw.cxgbe.rsrv_noflowq", &t4_rsrv_noflowq);
 273
 274 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
 275 #define NOFLDTXQ 8
 276 static int t4_nofldtxq = -NOFLDTXQ;
 277 TUNABLE_INT("hw.cxgbe.nofldtxq", &t4_nofldtxq);
 278
 279 #define NOFLDRXQ 2
 280 static int t4_nofldrxq = -NOFLDRXQ;
 281 TUNABLE_INT("hw.cxgbe.nofldrxq", &t4_nofldrxq);
 282
 283 #define NOFLDTXQ_VI 1
 284 static int t4_nofldtxq_vi = -NOFLDTXQ_VI;
 285 TUNABLE_INT("hw.cxgbe.nofldtxq_vi", &t4_nofldtxq_vi);
 286
 287 #define NOFLDRXQ_VI 1
 288 static int t4_nofldrxq_vi = -NOFLDRXQ_VI;
 289 TUNABLE_INT("hw.cxgbe.nofldrxq_vi", &t4_nofldrxq_vi);
 290
 291 #define TMR_IDX_OFLD 1
 292 int t4_tmr_idx_ofld = TMR_IDX_OFLD;
 293 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_ofld", &t4_tmr_idx_ofld);
 294
 295 #define PKTC_IDX_OFLD (-1)
 296 int t4_pktc_idx_ofld = PKTC_IDX_OFLD;
 297 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_ofld", &t4_pktc_idx_ofld);
 298
 299 /* 0 means chip/fw default, non-zero number is value in microseconds */
 300 static u_long t4_toe_keepalive_idle = 0;
 301 TUNABLE_ULONG("hw.cxgbe.toe.keepalive_idle", &t4_toe_keepalive_idle);
 302
 303 /* 0 means chip/fw default, non-zero number is value in microseconds */
 304 static u_long t4_toe_keepalive_interval = 0;
 305 TUNABLE_ULONG("hw.cxgbe.toe.keepalive_interval", &t4_toe_keepalive_interval);
 306
 307 /* 0 means chip/fw default, non-zero number is # of keepalives before abort */
 308 static int t4_toe_keepalive_count = 0;
 309 TUNABLE_INT("hw.cxgbe.toe.keepalive_count", &t4_toe_keepalive_count);
 310
 311 /* 0 means chip/fw default, non-zero number is value in microseconds */
 312 static u_long t4_toe_rexmt_min = 0;
 313 TUNABLE_ULONG("hw.cxgbe.toe.rexmt_min", &t4_toe_rexmt_min);
 314
 315 /* 0 means chip/fw default, non-zero number is value in microseconds */
 316 static u_long t4_toe_rexmt_max = 0;
 317 TUNABLE_ULONG("hw.cxgbe.toe.rexmt_max", &t4_toe_rexmt_max);
 318
 319 /* 0 means chip/fw default, non-zero number is # of rexmt before abort */
 320 static int t4_toe_rexmt_count = 0;
 321 TUNABLE_INT("hw.cxgbe.toe.rexmt_count", &t4_toe_rexmt_count);
 322
 323 /* -1 means chip/fw default, other values are raw backoff values to use */
 324 static int t4_toe_rexmt_backoff[16] = {
 325         -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
 326 };
 327 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.0", &t4_toe_rexmt_backoff[0]);
 328 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.1", &t4_toe_rexmt_backoff[1]);
 329 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.2", &t4_toe_rexmt_backoff[2]);
 330 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.3", &t4_toe_rexmt_backoff[3]);
 331 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.4", &t4_toe_rexmt_backoff[4]);
 332 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.5", &t4_toe_rexmt_backoff[5]);
 333 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.6", &t4_toe_rexmt_backoff[6]);
 334 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.7", &t4_toe_rexmt_backoff[7]);
 335 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.8", &t4_toe_rexmt_backoff[8]);
 336 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.9", &t4_toe_rexmt_backoff[9]);
 337 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.10", &t4_toe_rexmt_backoff[10]);
 338 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.11", &t4_toe_rexmt_backoff[11]);
 339 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.12", &t4_toe_rexmt_backoff[12]);
 340 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.13", &t4_toe_rexmt_backoff[13]);
 341 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.14", &t4_toe_rexmt_backoff[14]);
 342 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.15", &t4_toe_rexmt_backoff[15]);
 343 #endif
 344
 345 #ifdef DEV_NETMAP
 346 #define NNMTXQ_VI 2
 347 static int t4_nnmtxq_vi = -NNMTXQ_VI;
 348 TUNABLE_INT("hw.cxgbe.nnmtxq_vi", &t4_nnmtxq_vi);
 349
 350 #define NNMRXQ_VI 2
 351 static int t4_nnmrxq_vi = -NNMRXQ_VI;
 352 TUNABLE_INT("hw.cxgbe.nnmrxq_vi", &t4_nnmrxq_vi);
 353 #endif
 354
 355 /*
 356  * Holdoff parameters for ports.
 357  */
 358 #define TMR_IDX 1
 359 int t4_tmr_idx = TMR_IDX;
 360 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx", &t4_tmr_idx);
 361 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx);     /* Old name */
 362
 363 #define PKTC_IDX (-1)
 364 int t4_pktc_idx = PKTC_IDX;
 365 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx", &t4_pktc_idx);
 366 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx);     /* Old name */
 367
 368 /*
 369  * Size (# of entries) of each tx and rx queue.
 370  */
 371 unsigned int t4_qsize_txq = TX_EQ_QSIZE;
 372 TUNABLE_INT("hw.cxgbe.qsize_txq", &t4_qsize_txq);
 373
 374 unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
 375 TUNABLE_INT("hw.cxgbe.qsize_rxq", &t4_qsize_rxq);
 376
 377 /*
 378  * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
 379  */
 380 int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
 381 TUNABLE_INT("hw.cxgbe.interrupt_types", &t4_intr_types);
 382
 383 /*
 384  * Configuration file.  All the _CF names here are special.
 385  */
 386 #define DEFAULT_CF      "default"
 387 #define BUILTIN_CF      "built-in"
 388 #define FLASH_CF        "flash"
 389 #define UWIRE_CF        "uwire"
 390 #define FPGA_CF         "fpga"
 391 static char t4_cfg_file[32] = DEFAULT_CF;
 392 TUNABLE_STR("hw.cxgbe.config_file", t4_cfg_file, sizeof(t4_cfg_file));
 393
 394 /*
 395  * PAUSE settings (bit 0, 1 = rx_pause, tx_pause respectively).
 396  * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
 397  * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
 398  *            mark or when signalled to do so, 0 to never emit PAUSE.
 399  */
 400 static int t4_pause_settings = PAUSE_TX | PAUSE_RX;
 401 TUNABLE_INT("hw.cxgbe.pause_settings", &t4_pause_settings);
 402
 403 /*
 404  * Forward Error Correction settings (bit 0, 1, 2 = FEC_RS, FEC_BASER_RS,
 405  * FEC_RESERVED respectively).
 406  * -1 to run with the firmware default.
 407  *  0 to disable FEC.
 408  */
 409 static int t4_fec = -1;
 410 TUNABLE_INT("hw.cxgbe.fec", &t4_fec);
 411
 412 /*
 413  * Link autonegotiation.
 414  * -1 to run with the firmware default.
 415  *  0 to disable.
 416  *  1 to enable.
 417  */
 418 static int t4_autoneg = -1;
 419 TUNABLE_INT("hw.cxgbe.autoneg", &t4_autoneg);
 420
 421 /*
 422  * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
 423  * encouraged respectively).
 424  */
 425 static unsigned int t4_fw_install = 1;
 426 TUNABLE_INT("hw.cxgbe.fw_install", &t4_fw_install);
 427
 428 /*
 429  * ASIC features that will be used.  Disable the ones you don't want so that the
 430  * chip resources aren't wasted on features that will not be used.
 431  */
 432 static int t4_nbmcaps_allowed = 0;
 433 TUNABLE_INT("hw.cxgbe.nbmcaps_allowed", &t4_nbmcaps_allowed);
 434
 435 static int t4_linkcaps_allowed = 0;     /* No DCBX, PPP, etc. by default */
 436 TUNABLE_INT("hw.cxgbe.linkcaps_allowed", &t4_linkcaps_allowed);
 437
 438 static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
 439     FW_CAPS_CONFIG_SWITCH_EGRESS;
 440 TUNABLE_INT("hw.cxgbe.switchcaps_allowed", &t4_switchcaps_allowed);
 441
 442 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
 443         FW_CAPS_CONFIG_NIC_HASHFILTER | FW_CAPS_CONFIG_NIC_ETHOFLD;
 444 TUNABLE_INT("hw.cxgbe.niccaps_allowed", &t4_niccaps_allowed);
 445
 446 static int t4_toecaps_allowed = -1;
 447 TUNABLE_INT("hw.cxgbe.toecaps_allowed", &t4_toecaps_allowed);
 448
 449 static int t4_rdmacaps_allowed = -1;
 450 TUNABLE_INT("hw.cxgbe.rdmacaps_allowed", &t4_rdmacaps_allowed);
 451
 452 static int t4_cryptocaps_allowed = -1;
 453 TUNABLE_INT("hw.cxgbe.cryptocaps_allowed", &t4_cryptocaps_allowed);
 454
 455 static int t4_iscsicaps_allowed = -1;
 456 TUNABLE_INT("hw.cxgbe.iscsicaps_allowed", &t4_iscsicaps_allowed);
 457
 458 static int t4_fcoecaps_allowed = 0;
 459 TUNABLE_INT("hw.cxgbe.fcoecaps_allowed", &t4_fcoecaps_allowed);
 460
 461 static int t5_write_combine = 0;
 462 TUNABLE_INT("hw.cxl.write_combine", &t5_write_combine);
 463
 464 static int t4_num_vis = 1;
 465 TUNABLE_INT("hw.cxgbe.num_vis", &t4_num_vis);
 466 /*
 467  * PCIe Relaxed Ordering.
 468  * -1: driver should figure out a good value.
 469  * 0: disable RO.
 470  * 1: enable RO.
 471  * 2: leave RO alone.
 472  */
 473 static int pcie_relaxed_ordering = -1;
 474 TUNABLE_INT("hw.cxgbe.pcie_relaxed_ordering", &pcie_relaxed_ordering);
 475
 476 static int t4_panic_on_fatal_err = 0;
 477 TUNABLE_INT("hw.cxgbe.panic_on_fatal_err", &t4_panic_on_fatal_err);
 478
 479 #ifdef TCP_OFFLOAD
 480 /*
 481  * TOE tunables.
 482  */
 483 static int t4_cop_managed_offloading = 0;
 484 TUNABLE_INT("hw.cxgbe.cop_managed_offloading", &t4_cop_managed_offloading);
 485 #endif
 486
 487 /* Functions used by VIs to obtain unique MAC addresses for each VI. */
 488 static int vi_mac_funcs[] = {
 489         FW_VI_FUNC_ETH,
 490         FW_VI_FUNC_OFLD,
 491         FW_VI_FUNC_IWARP,
 492         FW_VI_FUNC_OPENISCSI,
 493         FW_VI_FUNC_OPENFCOE,
 494         FW_VI_FUNC_FOISCSI,
 495         FW_VI_FUNC_FOFCOE,
 496 };
 497
 498 struct intrs_and_queues {
 499         uint16_t intr_type;     /* INTx, MSI, or MSI-X */
 500         uint16_t num_vis;       /* number of VIs for each port */
 501         uint16_t nirq;          /* Total # of vectors */
 502         uint16_t ntxq;          /* # of NIC txq's for each port */
 503         uint16_t nrxq;          /* # of NIC rxq's for each port */
 504         uint16_t nofldtxq;      /* # of TOE/ETHOFLD txq's for each port */
 505         uint16_t nofldrxq;      /* # of TOE rxq's for each port */
 506
 507         /* The vcxgbe/vcxl interfaces use these and not the ones above. */
 508         uint16_t ntxq_vi;       /* # of NIC txq's */
 509         uint16_t nrxq_vi;       /* # of NIC rxq's */
 510         uint16_t nofldtxq_vi;   /* # of TOE txq's */
 511         uint16_t nofldrxq_vi;   /* # of TOE rxq's */
 512         uint16_t nnmtxq_vi;     /* # of netmap txq's */
 513         uint16_t nnmrxq_vi;     /* # of netmap rxq's */
 514 };
 515
 516 static void setup_memwin(struct adapter *);
 517 static void position_memwin(struct adapter *, int, uint32_t);
 518 static int validate_mem_range(struct adapter *, uint32_t, int);
 519 static int fwmtype_to_hwmtype(int);
 520 static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
 521     uint32_t *);
 522 static int fixup_devlog_params(struct adapter *);
 523 static int cfg_itype_and_nqueues(struct adapter *, struct intrs_and_queues *);
 524 static int prep_firmware(struct adapter *);
 525 static int partition_resources(struct adapter *, const struct firmware *,
 526     const char *);
 527 static int get_params__pre_init(struct adapter *);
 528 static int get_params__post_init(struct adapter *);
 529 static int set_params__post_init(struct adapter *);
 530 static void t4_set_desc(struct adapter *);
 531 static void build_medialist(struct port_info *, struct ifmedia *);
 532 static void init_l1cfg(struct port_info *);
 533 static int apply_l1cfg(struct port_info *);
 534 static int cxgbe_init_synchronized(struct vi_info *);
 535 static int cxgbe_uninit_synchronized(struct vi_info *);
 536 static void quiesce_txq(struct adapter *, struct sge_txq *);
 537 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
 538 static void quiesce_iq(struct adapter *, struct sge_iq *);
 539 static void quiesce_fl(struct adapter *, struct sge_fl *);
 540 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
 541     driver_intr_t *, void *, char *);
 542 static int t4_free_irq(struct adapter *, struct irq *);
 543 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
 544 static void vi_refresh_stats(struct adapter *, struct vi_info *);
 545 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
 546 static void cxgbe_tick(void *);
 547 static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
 548 static void cxgbe_sysctls(struct port_info *);
 549 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
 550 static int sysctl_bitfield(SYSCTL_HANDLER_ARGS);
 551 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
 552 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
 553 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
 554 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
 555 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
 556 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
 557 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
 558 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
 559 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
 560 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
 561 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
 562 static int sysctl_loadavg(SYSCTL_HANDLER_ARGS);
 563 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
 564 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
 565 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
 566 static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
 567 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
 568 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
 569 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
 570 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
 571 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
 572 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
 573 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
 574 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
 575 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
 576 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
 577 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
 578 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
 579 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
 580 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
 581 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
 582 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
 583 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
 584 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
 585 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
 586 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
 587 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
 588 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
 589 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
 590 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
 591 static int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
 592 static int sysctl_cpus(SYSCTL_HANDLER_ARGS);
 593 #ifdef TCP_OFFLOAD
 594 static int sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS);
 595 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
 596 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
 597 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
 598 static int sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS);
 599 static int sysctl_tp_backoff(SYSCTL_HANDLER_ARGS);
 600 static int sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS);
 601 static int sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS);
 602 #endif
 603 static int get_sge_context(struct adapter *, struct t4_sge_context *);
 604 static int load_fw(struct adapter *, struct t4_data *);
 605 static int load_cfg(struct adapter *, struct t4_data *);
 606 static int load_boot(struct adapter *, struct t4_bootrom *);
 607 static int load_bootcfg(struct adapter *, struct t4_data *);
 608 static int cudbg_dump(struct adapter *, struct t4_cudbg_dump *);
 609 static void free_offload_policy(struct t4_offload_policy *);
 610 static int set_offload_policy(struct adapter *, struct t4_offload_policy *);
 611 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
 612 static int read_i2c(struct adapter *, struct t4_i2c_data *);
 613 #ifdef TCP_OFFLOAD
 614 static int toe_capability(struct vi_info *, int);
 615 #endif
 616 static int mod_event(module_t, int, void *);
 617 static int notify_siblings(device_t, int);
 618
 619 struct {
 620         uint16_t device;
 621         char *desc;
 622 } t4_pciids[] = {
 623         {0xa000, "Chelsio Terminator 4 FPGA"},
 624         {0x4400, "Chelsio T440-dbg"},
 625         {0x4401, "Chelsio T420-CR"},
 626         {0x4402, "Chelsio T422-CR"},
 627         {0x4403, "Chelsio T440-CR"},
 628         {0x4404, "Chelsio T420-BCH"},
 629         {0x4405, "Chelsio T440-BCH"},
 630         {0x4406, "Chelsio T440-CH"},
 631         {0x4407, "Chelsio T420-SO"},
 632         {0x4408, "Chelsio T420-CX"},
 633         {0x4409, "Chelsio T420-BT"},
 634         {0x440a, "Chelsio T404-BT"},
 635         {0x440e, "Chelsio T440-LP-CR"},
 636 }, t5_pciids[] = {
 637         {0xb000, "Chelsio Terminator 5 FPGA"},
 638         {0x5400, "Chelsio T580-dbg"},
 639         {0x5401,  "Chelsio T520-CR"},           /* 2 x 10G */
 640         {0x5402,  "Chelsio T522-CR"},           /* 2 x 10G, 2 X 1G */
 641         {0x5403,  "Chelsio T540-CR"},           /* 4 x 10G */
 642         {0x5407,  "Chelsio T520-SO"},           /* 2 x 10G, nomem */
 643         {0x5409,  "Chelsio T520-BT"},           /* 2 x 10GBaseT */
 644         {0x540a,  "Chelsio T504-BT"},           /* 4 x 1G */
 645         {0x540d,  "Chelsio T580-CR"},           /* 2 x 40G */
 646         {0x540e,  "Chelsio T540-LP-CR"},        /* 4 x 10G */
 647         {0x5410,  "Chelsio T580-LP-CR"},        /* 2 x 40G */
 648         {0x5411,  "Chelsio T520-LL-CR"},        /* 2 x 10G */
 649         {0x5412,  "Chelsio T560-CR"},           /* 1 x 40G, 2 x 10G */
 650         {0x5414,  "Chelsio T580-LP-SO-CR"},     /* 2 x 40G, nomem */
 651         {0x5415,  "Chelsio T502-BT"},           /* 2 x 1G */
 652         {0x5418,  "Chelsio T540-BT"},           /* 4 x 10GBaseT */
 653         {0x5419,  "Chelsio T540-LP-BT"},        /* 4 x 10GBaseT */
 654         {0x541a,  "Chelsio T540-SO-BT"},        /* 4 x 10GBaseT, nomem */
 655         {0x541b,  "Chelsio T540-SO-CR"},        /* 4 x 10G, nomem */
 656 }, t6_pciids[] = {
 657         {0xc006, "Chelsio Terminator 6 FPGA"},  /* T6 PE10K6 FPGA (PF0) */
 658         {0x6400, "Chelsio T6-DBG-25"},          /* 2 x 10/25G, debug */
 659         {0x6401, "Chelsio T6225-CR"},           /* 2 x 10/25G */
 660         {0x6402, "Chelsio T6225-SO-CR"},        /* 2 x 10/25G, nomem */
 661         {0x6403, "Chelsio T6425-CR"},           /* 4 x 10/25G */
 662         {0x6404, "Chelsio T6425-SO-CR"},        /* 4 x 10/25G, nomem */
 663         {0x6405, "Chelsio T6225-OCP-SO"},       /* 2 x 10/25G, nomem */
 664         {0x6406, "Chelsio T62100-OCP-SO"},      /* 2 x 40/50/100G, nomem */
 665         {0x6407, "Chelsio T62100-LP-CR"},       /* 2 x 40/50/100G */
 666         {0x6408, "Chelsio T62100-SO-CR"},       /* 2 x 40/50/100G, nomem */
 667         {0x6409, "Chelsio T6210-BT"},           /* 2 x 10GBASE-T */
 668         {0x640d, "Chelsio T62100-CR"},          /* 2 x 40/50/100G */
 669         {0x6410, "Chelsio T6-DBG-100"},         /* 2 x 40/50/100G, debug */
 670         {0x6411, "Chelsio T6225-LL-CR"},        /* 2 x 10/25G */
 671         {0x6414, "Chelsio T61100-OCP-SO"},      /* 1 x 40/50/100G, nomem */
 672         {0x6415, "Chelsio T6201-BT"},           /* 2 x 1000BASE-T */
 673
 674         /* Custom */
 675         {0x6480, "Custom T6225-CR"},
 676         {0x6481, "Custom T62100-CR"},
 677         {0x6482, "Custom T6225-CR"},
 678         {0x6483, "Custom T62100-CR"},
 679         {0x6484, "Custom T64100-CR"},
 680         {0x6485, "Custom T6240-SO"},
 681         {0x6486, "Custom T6225-SO-CR"},
 682         {0x6487, "Custom T6225-CR"},
 683 };
 684
 685 #ifdef TCP_OFFLOAD
 686 /*
 687  * service_iq() has an iq and needs the fl.  Offset of fl from the iq should be
 688  * exactly the same for both rxq and ofld_rxq.
 689  */
 690 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
 691 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
 692 #endif
 693 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
 694
 695 static int
 696 t4_probe(device_t dev)
 697 {
 698         int i;
 699         uint16_t v = pci_get_vendor(dev);
 700         uint16_t d = pci_get_device(dev);
 701         uint8_t f = pci_get_function(dev);
 702
 703         if (v != PCI_VENDOR_ID_CHELSIO)
 704                 return (ENXIO);
 705
 706         /* Attach only to PF0 of the FPGA */
 707         if (d == 0xa000 && f != 0)
 708                 return (ENXIO);
 709
 710         for (i = 0; i < nitems(t4_pciids); i++) {
 711                 if (d == t4_pciids[i].device) {
 712                         device_set_desc(dev, t4_pciids[i].desc);
 713                         return (BUS_PROBE_DEFAULT);
 714                 }
 715         }
 716
 717         return (ENXIO);
 718 }
 719
 720 static int
 721 t5_probe(device_t dev)
 722 {
 723         int i;
 724         uint16_t v = pci_get_vendor(dev);
 725         uint16_t d = pci_get_device(dev);
 726         uint8_t f = pci_get_function(dev);
 727
 728         if (v != PCI_VENDOR_ID_CHELSIO)
 729                 return (ENXIO);
 730
 731         /* Attach only to PF0 of the FPGA */
 732         if (d == 0xb000 && f != 0)
 733                 return (ENXIO);
 734
 735         for (i = 0; i < nitems(t5_pciids); i++) {
 736                 if (d == t5_pciids[i].device) {
 737                         device_set_desc(dev, t5_pciids[i].desc);
 738                         return (BUS_PROBE_DEFAULT);
 739                 }
 740         }
 741
 742         return (ENXIO);
 743 }
 744
 745 static int
 746 t6_probe(device_t dev)
 747 {
 748         int i;
 749         uint16_t v = pci_get_vendor(dev);
 750         uint16_t d = pci_get_device(dev);
 751
 752         if (v != PCI_VENDOR_ID_CHELSIO)
 753                 return (ENXIO);
 754
 755         for (i = 0; i < nitems(t6_pciids); i++) {
 756                 if (d == t6_pciids[i].device) {
 757                         device_set_desc(dev, t6_pciids[i].desc);
 758                         return (BUS_PROBE_DEFAULT);
 759                 }
 760         }
 761
 762         return (ENXIO);
 763 }
 764
 765 static void
 766 t5_attribute_workaround(device_t dev)
 767 {
 768         device_t root_port;
 769         uint32_t v;
 770
 771         /*
 772          * The T5 chips do not properly echo the No Snoop and Relaxed
 773          * Ordering attributes when replying to a TLP from a Root
 774          * Port.  As a workaround, find the parent Root Port and
 775          * disable No Snoop and Relaxed Ordering.  Note that this
 776          * affects all devices under this root port.
 777          */
 778         root_port = pci_find_pcie_root_port(dev);
 779         if (root_port == NULL) {
 780                 device_printf(dev, "Unable to find parent root port\n");
 781                 return;
 782         }
 783
 784         v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
 785             PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
 786         if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
 787             0)
 788                 device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
 789                     device_get_nameunit(root_port));
 790 }
 791
 792 static const struct devnames devnames[] = {
 793         {
 794                 .nexus_name = "t4nex",
 795                 .ifnet_name = "cxgbe",
 796                 .vi_ifnet_name = "vcxgbe",
 797                 .pf03_drv_name = "t4iov",
 798                 .vf_nexus_name = "t4vf",
 799                 .vf_ifnet_name = "cxgbev"
 800         }, {
 801                 .nexus_name = "t5nex",
 802                 .ifnet_name = "cxl",
 803                 .vi_ifnet_name = "vcxl",
 804                 .pf03_drv_name = "t5iov",
 805                 .vf_nexus_name = "t5vf",
 806                 .vf_ifnet_name = "cxlv"
 807         }, {
 808                 .nexus_name = "t6nex",
 809                 .ifnet_name = "cc",
 810                 .vi_ifnet_name = "vcc",
 811                 .pf03_drv_name = "t6iov",
 812                 .vf_nexus_name = "t6vf",
 813                 .vf_ifnet_name = "ccv"
 814         }
 815 };
 816
 817 void
 818 t4_init_devnames(struct adapter *sc)
 819 {
 820         int id;
 821
 822         id = chip_id(sc);
 823         if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
 824                 sc->names = &devnames[id - CHELSIO_T4];
 825         else {
 826                 device_printf(sc->dev, "chip id %d is not supported.\n", id);
 827                 sc->names = NULL;
 828         }
 829 }
 830
 831 static int
 832 t4_attach(device_t dev)
 833 {
 834         struct adapter *sc;
 835         int rc = 0, i, j, rqidx, tqidx, nports;
 836         struct make_dev_args mda;
 837         struct intrs_and_queues iaq;
 838         struct sge *s;
 839         uint32_t *buf;
 840 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
 841         int ofld_tqidx;
 842 #endif
 843 #ifdef TCP_OFFLOAD
 844         int ofld_rqidx;
 845 #endif
 846 #ifdef DEV_NETMAP
 847         int nm_rqidx, nm_tqidx;
 848 #endif
 849         int num_vis;
 850
 851         sc = device_get_softc(dev);
 852         sc->dev = dev;
 853         TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
 854
 855         if ((pci_get_device(dev) & 0xff00) == 0x5400)
 856                 t5_attribute_workaround(dev);
 857         pci_enable_busmaster(dev);
 858         if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
 859                 uint32_t v;
 860
 861                 pci_set_max_read_req(dev, 4096);
 862                 v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
 863                 sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
 864                 if (pcie_relaxed_ordering == 0 &&
 865                     (v | PCIEM_CTL_RELAXED_ORD_ENABLE) != 0) {
 866                         v &= ~PCIEM_CTL_RELAXED_ORD_ENABLE;
 867                         pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
 868                 } else if (pcie_relaxed_ordering == 1 &&
 869                     (v & PCIEM_CTL_RELAXED_ORD_ENABLE) == 0) {
 870                         v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
 871                         pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
 872                 }
 873         }
 874
 875         sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
 876         sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
 877         sc->traceq = -1;
 878         mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
 879         snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
 880             device_get_nameunit(dev));
 881
 882         snprintf(sc->lockname, sizeof(sc->lockname), "%s",
 883             device_get_nameunit(dev));
 884         mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
 885         t4_add_adapter(sc);
 886
 887         mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
 888         TAILQ_INIT(&sc->sfl);
 889         callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
 890
 891         mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
 892
 893         sc->policy = NULL;
 894         rw_init(&sc->policy_lock, "connection offload policy");
 895
 896         rc = t4_map_bars_0_and_4(sc);
 897         if (rc != 0)
 898                 goto done; /* error message displayed already */
 899
 900         memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
 901
 902         /* Prepare the adapter for operation. */
 903         buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
 904         rc = -t4_prep_adapter(sc, buf);
 905         free(buf, M_CXGBE);
 906         if (rc != 0) {
 907                 device_printf(dev, "failed to prepare adapter: %d.\n", rc);
 908                 goto done;
 909         }
 910
 911         /*
 912          * This is the real PF# to which we're attaching.  Works from within PCI
 913          * passthrough environments too, where pci_get_function() could return a
 914          * different PF# depending on the passthrough configuration.  We need to
 915          * use the real PF# in all our communication with the firmware.
 916          */
 917         j = t4_read_reg(sc, A_PL_WHOAMI);
 918         sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
 919         sc->mbox = sc->pf;
 920
 921         t4_init_devnames(sc);
 922         if (sc->names == NULL) {
 923                 rc = ENOTSUP;
 924                 goto done; /* error message displayed already */
 925         }
 926
 927         /*
 928          * Do this really early, with the memory windows set up even before the
 929          * character device.  The userland tool's register i/o and mem read
 930          * will work even in "recovery mode".
 931          */
 932         setup_memwin(sc);
 933         if (t4_init_devlog_params(sc, 0) == 0)
 934                 fixup_devlog_params(sc);
 935         make_dev_args_init(&mda);
 936         mda.mda_devsw = &t4_cdevsw;
 937         mda.mda_uid = UID_ROOT;
 938         mda.mda_gid = GID_WHEEL;
 939         mda.mda_mode = 0600;
 940         mda.mda_si_drv1 = sc;
 941         rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
 942         if (rc != 0)
 943                 device_printf(dev, "failed to create nexus char device: %d.\n",
 944                     rc);
 945
 946         /* Go no further if recovery mode has been requested. */
 947         if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
 948                 device_printf(dev, "recovery mode.\n");
 949                 goto done;
 950         }
 951
 952 #if defined(__i386__)
 953         if ((cpu_feature & CPUID_CX8) == 0) {
 954                 device_printf(dev, "64 bit atomics not available.\n");
 955                 rc = ENOTSUP;
 956                 goto done;
 957         }
 958 #endif
 959
 960         /* Prepare the firmware for operation */
 961         rc = prep_firmware(sc);
 962         if (rc != 0)
 963                 goto done; /* error message displayed already */
 964
 965         rc = get_params__post_init(sc);
 966         if (rc != 0)
 967                 goto done; /* error message displayed already */
 968
 969         rc = set_params__post_init(sc);
 970         if (rc != 0)
 971                 goto done; /* error message displayed already */
 972
 973         rc = t4_map_bar_2(sc);
 974         if (rc != 0)
 975                 goto done; /* error message displayed already */
 976
 977         rc = t4_create_dma_tag(sc);
 978         if (rc != 0)
 979                 goto done; /* error message displayed already */
 980
 981         /*
 982          * First pass over all the ports - allocate VIs and initialize some
 983          * basic parameters like mac address, port type, etc.
 984          */
 985         for_each_port(sc, i) {
 986                 struct port_info *pi;
 987
 988                 pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
 989                 sc->port[i] = pi;
 990
 991                 /* These must be set before t4_port_init */
 992                 pi->adapter = sc;
 993                 pi->port_id = i;
 994                 /*
 995                  * XXX: vi[0] is special so we can't delay this allocation until
 996                  * pi->nvi's final value is known.
 997                  */
 998                 pi->vi = malloc(sizeof(struct vi_info) * t4_num_vis, M_CXGBE,
 999                     M_ZERO | M_WAITOK);
1000
1001                 /*
1002                  * Allocate the "main" VI and initialize parameters
1003                  * like mac addr.
1004                  */
1005                 rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
1006                 if (rc != 0) {
1007                         device_printf(dev, "unable to initialize port %d: %d\n",
1008                             i, rc);
1009                         free(pi->vi, M_CXGBE);
1010                         free(pi, M_CXGBE);
1011                         sc->port[i] = NULL;
1012                         goto done;
1013                 }
1014
1015                 snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
1016                     device_get_nameunit(dev), i);
1017                 mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
1018                 sc->chan_map[pi->tx_chan] = i;
1019
1020                 /* All VIs on this port share this media. */
1021                 ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
1022                     cxgbe_media_status);
1023
1024                 pi->dev = device_add_child(dev, sc->names->ifnet_name, -1);
1025                 if (pi->dev == NULL) {
1026                         device_printf(dev,
1027                             "failed to add device for port %d.\n", i);
1028                         rc = ENXIO;
1029                         goto done;
1030                 }
1031                 pi->vi[0].dev = pi->dev;
1032                 device_set_softc(pi->dev, pi);
1033         }
1034
1035         /*
1036          * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1037          */
1038         nports = sc->params.nports;
1039         rc = cfg_itype_and_nqueues(sc, &iaq);
1040         if (rc != 0)
1041                 goto done; /* error message displayed already */
1042
1043         num_vis = iaq.num_vis;
1044         sc->intr_type = iaq.intr_type;
1045         sc->intr_count = iaq.nirq;
1046
1047         s = &sc->sge;
1048         s->nrxq = nports * iaq.nrxq;
1049         s->ntxq = nports * iaq.ntxq;
1050         if (num_vis > 1) {
1051                 s->nrxq += nports * (num_vis - 1) * iaq.nrxq_vi;
1052                 s->ntxq += nports * (num_vis - 1) * iaq.ntxq_vi;
1053         }
1054         s->neq = s->ntxq + s->nrxq;     /* the free list in an rxq is an eq */
1055         s->neq += nports + 1;/* ctrl queues: 1 per port + 1 mgmt */
1056         s->niq = s->nrxq + 1;           /* 1 extra for firmware event queue */
1057 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1058         if (is_offload(sc) || is_ethoffload(sc)) {
1059                 s->nofldtxq = nports * iaq.nofldtxq;
1060                 if (num_vis > 1)
1061                         s->nofldtxq += nports * (num_vis - 1) * iaq.nofldtxq_vi;
1062                 s->neq += s->nofldtxq;
1063
1064                 s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1065                     M_CXGBE, M_ZERO | M_WAITOK);
1066         }
1067 #endif
1068 #ifdef TCP_OFFLOAD
1069         if (is_offload(sc)) {
1070                 s->nofldrxq = nports * iaq.nofldrxq;
1071                 if (num_vis > 1)
1072                         s->nofldrxq += nports * (num_vis - 1) * iaq.nofldrxq_vi;
1073                 s->neq += s->nofldrxq;  /* free list */
1074                 s->niq += s->nofldrxq;
1075
1076                 s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1077                     M_CXGBE, M_ZERO | M_WAITOK);
1078         }
1079 #endif
1080 #ifdef DEV_NETMAP
1081         if (num_vis > 1) {
1082                 s->nnmrxq = nports * (num_vis - 1) * iaq.nnmrxq_vi;
1083                 s->nnmtxq = nports * (num_vis - 1) * iaq.nnmtxq_vi;
1084         }
1085         s->neq += s->nnmtxq + s->nnmrxq;
1086         s->niq += s->nnmrxq;
1087
1088         s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1089             M_CXGBE, M_ZERO | M_WAITOK);
1090         s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1091             M_CXGBE, M_ZERO | M_WAITOK);
1092 #endif
1093
1094         s->ctrlq = malloc(nports * sizeof(struct sge_wrq), M_CXGBE,
1095             M_ZERO | M_WAITOK);
1096         s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1097             M_ZERO | M_WAITOK);
1098         s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1099             M_ZERO | M_WAITOK);
1100         s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1101             M_ZERO | M_WAITOK);
1102         s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1103             M_ZERO | M_WAITOK);
1104
1105         sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1106             M_ZERO | M_WAITOK);
1107
1108         t4_init_l2t(sc, M_WAITOK);
1109         t4_init_smt(sc, M_WAITOK);
1110         t4_init_tx_sched(sc);
1111 #ifdef RATELIMIT
1112         t4_init_etid_table(sc);
1113 #endif
1114
1115         /*
1116          * Second pass over the ports.  This time we know the number of rx and
1117          * tx queues that each port should get.
1118          */
1119         rqidx = tqidx = 0;
1120 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1121         ofld_tqidx = 0;
1122 #endif
1123 #ifdef TCP_OFFLOAD
1124         ofld_rqidx = 0;
1125 #endif
1126 #ifdef DEV_NETMAP
1127         nm_rqidx = nm_tqidx = 0;
1128 #endif
1129         for_each_port(sc, i) {
1130                 struct port_info *pi = sc->port[i];
1131                 struct vi_info *vi;
1132
1133                 if (pi == NULL)
1134                         continue;
1135
1136                 pi->nvi = num_vis;
1137                 for_each_vi(pi, j, vi) {
1138                         vi->pi = pi;
1139                         vi->qsize_rxq = t4_qsize_rxq;
1140                         vi->qsize_txq = t4_qsize_txq;
1141
1142                         vi->first_rxq = rqidx;
1143                         vi->first_txq = tqidx;
1144                         vi->tmr_idx = t4_tmr_idx;
1145                         vi->pktc_idx = t4_pktc_idx;
1146                         vi->nrxq = j == 0 ? iaq.nrxq : iaq.nrxq_vi;
1147                         vi->ntxq = j == 0 ? iaq.ntxq : iaq.ntxq_vi;
1148
1149                         rqidx += vi->nrxq;
1150                         tqidx += vi->ntxq;
1151
1152                         if (j == 0 && vi->ntxq > 1)
1153                                 vi->rsrv_noflowq = t4_rsrv_noflowq ? 1 : 0;
1154                         else
1155                                 vi->rsrv_noflowq = 0;
1156
1157 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1158                         vi->first_ofld_txq = ofld_tqidx;
1159                         vi->nofldtxq = j == 0 ? iaq.nofldtxq : iaq.nofldtxq_vi;
1160                         ofld_tqidx += vi->nofldtxq;
1161 #endif
1162 #ifdef TCP_OFFLOAD
1163                         vi->ofld_tmr_idx = t4_tmr_idx_ofld;
1164                         vi->ofld_pktc_idx = t4_pktc_idx_ofld;
1165                         vi->first_ofld_rxq = ofld_rqidx;
1166                         vi->nofldrxq = j == 0 ? iaq.nofldrxq : iaq.nofldrxq_vi;
1167
1168                         ofld_rqidx += vi->nofldrxq;
1169 #endif
1170 #ifdef DEV_NETMAP
1171                         if (j > 0) {
1172                                 vi->first_nm_rxq = nm_rqidx;
1173                                 vi->first_nm_txq = nm_tqidx;
1174                                 vi->nnmrxq = iaq.nnmrxq_vi;
1175                                 vi->nnmtxq = iaq.nnmtxq_vi;
1176                                 nm_rqidx += vi->nnmrxq;
1177                                 nm_tqidx += vi->nnmtxq;
1178                         }
1179 #endif
1180                 }
1181         }
1182
1183         rc = t4_setup_intr_handlers(sc);
1184         if (rc != 0) {
1185                 device_printf(dev,
1186                     "failed to setup interrupt handlers: %d\n", rc);
1187                 goto done;
1188         }
1189
1190         rc = bus_generic_probe(dev);
1191         if (rc != 0) {
1192                 device_printf(dev, "failed to probe child drivers: %d\n", rc);
1193                 goto done;
1194         }
1195
1196         /*
1197          * Ensure thread-safe mailbox access (in debug builds).
1198          *
1199          * So far this was the only thread accessing the mailbox but various
1200          * ifnets and sysctls are about to be created and their handlers/ioctls
1201          * will access the mailbox from different threads.
1202          */
1203         sc->flags |= CHK_MBOX_ACCESS;
1204
1205         rc = bus_generic_attach(dev);
1206         if (rc != 0) {
1207                 device_printf(dev,
1208                     "failed to attach all child ports: %d\n", rc);
1209                 goto done;
1210         }
1211
1212         device_printf(dev,
1213             "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1214             sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1215             sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1216             (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1217             sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1218
1219         t4_set_desc(sc);
1220
1221         notify_siblings(dev, 0);
1222
1223 done:
1224         if (rc != 0 && sc->cdev) {
1225                 /* cdev was created and so cxgbetool works; recover that way. */
1226                 device_printf(dev,
1227                     "error during attach, adapter is now in recovery mode.\n");
1228                 rc = 0;
1229         }
1230
1231         if (rc != 0)
1232                 t4_detach_common(dev);
1233         else
1234                 t4_sysctls(sc);
1235
1236         return (rc);
1237 }
1238
1239 static int
1240 t4_ready(device_t dev)
1241 {
1242         struct adapter *sc;
1243
1244         sc = device_get_softc(dev);
1245         if (sc->flags & FW_OK)
1246                 return (0);
1247         return (ENXIO);
1248 }
1249
1250 static int
1251 t4_read_port_device(device_t dev, int port, device_t *child)
1252 {
1253         struct adapter *sc;
1254         struct port_info *pi;
1255
1256         sc = device_get_softc(dev);
1257         if (port < 0 || port >= MAX_NPORTS)
1258                 return (EINVAL);
1259         pi = sc->port[port];
1260         if (pi == NULL || pi->dev == NULL)
1261                 return (ENXIO);
1262         *child = pi->dev;
1263         return (0);
1264 }
1265
1266 static int
1267 notify_siblings(device_t dev, int detaching)
1268 {
1269         device_t sibling;
1270         int error, i;
1271
1272         error = 0;
1273         for (i = 0; i < PCI_FUNCMAX; i++) {
1274                 if (i == pci_get_function(dev))
1275                         continue;
1276                 sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1277                     pci_get_slot(dev), i);
1278                 if (sibling == NULL || !device_is_attached(sibling))
1279                         continue;
1280                 if (detaching)
1281                         error = T4_DETACH_CHILD(sibling);
1282                 else
1283                         (void)T4_ATTACH_CHILD(sibling);
1284                 if (error)
1285                         break;
1286         }
1287         return (error);
1288 }
1289
1290 /*
1291  * Idempotent
1292  */
1293 static int
1294 t4_detach(device_t dev)
1295 {
1296         struct adapter *sc;
1297         int rc;
1298
1299         sc = device_get_softc(dev);
1300
1301         rc = notify_siblings(dev, 1);
1302         if (rc) {
1303                 device_printf(dev,
1304                     "failed to detach sibling devices: %d\n", rc);
1305                 return (rc);
1306         }
1307
1308         return (t4_detach_common(dev));
1309 }
1310
1311 int
1312 t4_detach_common(device_t dev)
1313 {
1314         struct adapter *sc;
1315         struct port_info *pi;
1316         int i, rc;
1317
1318         sc = device_get_softc(dev);
1319
1320         if (sc->cdev) {
1321                 destroy_dev(sc->cdev);
1322                 sc->cdev = NULL;
1323         }
1324
1325         sc->flags &= ~CHK_MBOX_ACCESS;
1326         if (sc->flags & FULL_INIT_DONE) {
1327                 if (!(sc->flags & IS_VF))
1328                         t4_intr_disable(sc);
1329         }
1330
1331         if (device_is_attached(dev)) {
1332                 rc = bus_generic_detach(dev);
1333                 if (rc) {
1334                         device_printf(dev,
1335                             "failed to detach child devices: %d\n", rc);
1336                         return (rc);
1337                 }
1338         }
1339
1340         for (i = 0; i < sc->intr_count; i++)
1341                 t4_free_irq(sc, &sc->irq[i]);
1342
1343         if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1344                 t4_free_tx_sched(sc);
1345
1346         for (i = 0; i < MAX_NPORTS; i++) {
1347                 pi = sc->port[i];
1348                 if (pi) {
1349                         t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1350                         if (pi->dev)
1351                                 device_delete_child(dev, pi->dev);
1352
1353                         mtx_destroy(&pi->pi_lock);
1354                         free(pi->vi, M_CXGBE);
1355                         free(pi, M_CXGBE);
1356                 }
1357         }
1358
1359         device_delete_children(dev);
1360
1361         if (sc->flags & FULL_INIT_DONE)
1362                 adapter_full_uninit(sc);
1363
1364         if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1365                 t4_fw_bye(sc, sc->mbox);
1366
1367         if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1368                 pci_release_msi(dev);
1369
1370         if (sc->regs_res)
1371                 bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1372                     sc->regs_res);
1373
1374         if (sc->udbs_res)
1375                 bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1376                     sc->udbs_res);
1377
1378         if (sc->msix_res)
1379                 bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1380                     sc->msix_res);
1381
1382         if (sc->l2t)
1383                 t4_free_l2t(sc->l2t);
1384         if (sc->smt)
1385                 t4_free_smt(sc->smt);
1386 #ifdef RATELIMIT
1387         t4_free_etid_table(sc);
1388 #endif
1389
1390 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1391         free(sc->sge.ofld_txq, M_CXGBE);
1392 #endif
1393 #ifdef TCP_OFFLOAD
1394         free(sc->sge.ofld_rxq, M_CXGBE);
1395 #endif
1396 #ifdef DEV_NETMAP
1397         free(sc->sge.nm_rxq, M_CXGBE);
1398         free(sc->sge.nm_txq, M_CXGBE);
1399 #endif
1400         free(sc->irq, M_CXGBE);
1401         free(sc->sge.rxq, M_CXGBE);
1402         free(sc->sge.txq, M_CXGBE);
1403         free(sc->sge.ctrlq, M_CXGBE);
1404         free(sc->sge.iqmap, M_CXGBE);
1405         free(sc->sge.eqmap, M_CXGBE);
1406         free(sc->tids.ftid_tab, M_CXGBE);
1407         if (sc->tids.hftid_tab)
1408                 free_hftid_tab(&sc->tids);
1409         free(sc->tids.atid_tab, M_CXGBE);
1410         free(sc->tids.tid_tab, M_CXGBE);
1411         free(sc->tt.tls_rx_ports, M_CXGBE);
1412         t4_destroy_dma_tag(sc);
1413         if (mtx_initialized(&sc->sc_lock)) {
1414                 sx_xlock(&t4_list_lock);
1415                 SLIST_REMOVE(&t4_list, sc, adapter, link);
1416                 sx_xunlock(&t4_list_lock);
1417                 mtx_destroy(&sc->sc_lock);
1418         }
1419
1420         callout_drain(&sc->sfl_callout);
1421         if (mtx_initialized(&sc->tids.ftid_lock)) {
1422                 mtx_destroy(&sc->tids.ftid_lock);
1423                 cv_destroy(&sc->tids.ftid_cv);
1424         }
1425         if (mtx_initialized(&sc->tids.atid_lock))
1426                 mtx_destroy(&sc->tids.atid_lock);
1427         if (mtx_initialized(&sc->sfl_lock))
1428                 mtx_destroy(&sc->sfl_lock);
1429         if (mtx_initialized(&sc->ifp_lock))
1430                 mtx_destroy(&sc->ifp_lock);
1431         if (mtx_initialized(&sc->reg_lock))
1432                 mtx_destroy(&sc->reg_lock);
1433
1434         if (rw_initialized(&sc->policy_lock)) {
1435                 rw_destroy(&sc->policy_lock);
1436 #ifdef TCP_OFFLOAD
1437                 if (sc->policy != NULL)
1438                         free_offload_policy(sc->policy);
1439 #endif
1440         }
1441
1442         for (i = 0; i < NUM_MEMWIN; i++) {
1443                 struct memwin *mw = &sc->memwin[i];
1444
1445                 if (rw_initialized(&mw->mw_lock))
1446                         rw_destroy(&mw->mw_lock);
1447         }
1448
1449         bzero(sc, sizeof(*sc));
1450
1451         return (0);
1452 }
1453
1454 static int
1455 cxgbe_probe(device_t dev)
1456 {
1457         char buf[128];
1458         struct port_info *pi = device_get_softc(dev);
1459
1460         snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1461         device_set_desc_copy(dev, buf);
1462
1463         return (BUS_PROBE_DEFAULT);
1464 }
1465
1466 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1467     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1468     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS)
1469 #define T4_CAP_ENABLE (T4_CAP)
1470
1471 static int
1472 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1473 {
1474         struct ifnet *ifp;
1475         struct sbuf *sb;
1476
1477         vi->xact_addr_filt = -1;
1478         callout_init(&vi->tick, 1);
1479
1480         /* Allocate an ifnet and set it up */
1481         ifp = if_alloc(IFT_ETHER);
1482         if (ifp == NULL) {
1483                 device_printf(dev, "Cannot allocate ifnet\n");
1484                 return (ENOMEM);
1485         }
1486         vi->ifp = ifp;
1487         ifp->if_softc = vi;
1488
1489         if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1490         ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1491
1492         ifp->if_init = cxgbe_init;
1493         ifp->if_ioctl = cxgbe_ioctl;
1494         ifp->if_transmit = cxgbe_transmit;
1495         ifp->if_qflush = cxgbe_qflush;
1496         ifp->if_get_counter = cxgbe_get_counter;
1497 #ifdef RATELIMIT
1498         ifp->if_snd_tag_alloc = cxgbe_snd_tag_alloc;
1499         ifp->if_snd_tag_modify = cxgbe_snd_tag_modify;
1500         ifp->if_snd_tag_query = cxgbe_snd_tag_query;
1501         ifp->if_snd_tag_free = cxgbe_snd_tag_free;
1502 #endif
1503
1504         ifp->if_capabilities = T4_CAP;
1505 #ifdef TCP_OFFLOAD
1506         if (vi->nofldrxq != 0)
1507                 ifp->if_capabilities |= IFCAP_TOE;
1508 #endif
1509 #ifdef DEV_NETMAP
1510         if (vi->nnmrxq != 0)
1511                 ifp->if_capabilities |= IFCAP_NETMAP;
1512 #endif
1513 #ifdef RATELIMIT
1514         if (is_ethoffload(vi->pi->adapter) && vi->nofldtxq != 0)
1515                 ifp->if_capabilities |= IFCAP_TXRTLMT;
1516 #endif
1517         ifp->if_capenable = T4_CAP_ENABLE;
1518         ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1519             CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1520
1521         ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1522         ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1523         ifp->if_hw_tsomaxsegsize = 65536;
1524
1525         vi->vlan_c = EVENTHANDLER_REGISTER(vlan_config, cxgbe_vlan_config, ifp,
1526             EVENTHANDLER_PRI_ANY);
1527
1528         ether_ifattach(ifp, vi->hw_addr);
1529 #ifdef DEV_NETMAP
1530         if (ifp->if_capabilities & IFCAP_NETMAP)
1531                 cxgbe_nm_attach(vi);
1532 #endif
1533         sb = sbuf_new_auto();
1534         sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1535 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1536         switch (ifp->if_capabilities & (IFCAP_TOE | IFCAP_TXRTLMT)) {
1537         case IFCAP_TOE:
1538                 sbuf_printf(sb, "; %d txq (TOE)", vi->nofldtxq);
1539                 break;
1540         case IFCAP_TOE | IFCAP_TXRTLMT:
1541                 sbuf_printf(sb, "; %d txq (TOE/ETHOFLD)", vi->nofldtxq);
1542                 break;
1543         case IFCAP_TXRTLMT:
1544                 sbuf_printf(sb, "; %d txq (ETHOFLD)", vi->nofldtxq);
1545                 break;
1546         }
1547 #endif
1548 #ifdef TCP_OFFLOAD
1549         if (ifp->if_capabilities & IFCAP_TOE)
1550                 sbuf_printf(sb, ", %d rxq (TOE)", vi->nofldrxq);
1551 #endif
1552 #ifdef DEV_NETMAP
1553         if (ifp->if_capabilities & IFCAP_NETMAP)
1554                 sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1555                     vi->nnmtxq, vi->nnmrxq);
1556 #endif
1557         sbuf_finish(sb);
1558         device_printf(dev, "%s\n", sbuf_data(sb));
1559         sbuf_delete(sb);
1560
1561         vi_sysctls(vi);
1562
1563         return (0);
1564 }
1565
1566 static int
1567 cxgbe_attach(device_t dev)
1568 {
1569         struct port_info *pi = device_get_softc(dev);
1570         struct adapter *sc = pi->adapter;
1571         struct vi_info *vi;
1572         int i, rc;
1573
1574         callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1575
1576         rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1577         if (rc)
1578                 return (rc);
1579
1580         for_each_vi(pi, i, vi) {
1581                 if (i == 0)
1582                         continue;
1583                 vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1584                 if (vi->dev == NULL) {
1585                         device_printf(dev, "failed to add VI %d\n", i);
1586                         continue;
1587                 }
1588                 device_set_softc(vi->dev, vi);
1589         }
1590
1591         cxgbe_sysctls(pi);
1592
1593         bus_generic_attach(dev);
1594
1595         return (0);
1596 }
1597
1598 static void
1599 cxgbe_vi_detach(struct vi_info *vi)
1600 {
1601         struct ifnet *ifp = vi->ifp;
1602
1603         ether_ifdetach(ifp);
1604
1605         if (vi->vlan_c)
1606                 EVENTHANDLER_DEREGISTER(vlan_config, vi->vlan_c);
1607
1608         /* Let detach proceed even if these fail. */
1609 #ifdef DEV_NETMAP
1610         if (ifp->if_capabilities & IFCAP_NETMAP)
1611                 cxgbe_nm_detach(vi);
1612 #endif
1613         cxgbe_uninit_synchronized(vi);
1614         callout_drain(&vi->tick);
1615         vi_full_uninit(vi);
1616
1617         if_free(vi->ifp);
1618         vi->ifp = NULL;
1619 }
1620
1621 static int
1622 cxgbe_detach(device_t dev)
1623 {
1624         struct port_info *pi = device_get_softc(dev);
1625         struct adapter *sc = pi->adapter;
1626         int rc;
1627
1628         /* Detach the extra VIs first. */
1629         rc = bus_generic_detach(dev);
1630         if (rc)
1631                 return (rc);
1632         device_delete_children(dev);
1633
1634         doom_vi(sc, &pi->vi[0]);
1635
1636         if (pi->flags & HAS_TRACEQ) {
1637                 sc->traceq = -1;        /* cloner should not create ifnet */
1638                 t4_tracer_port_detach(sc);
1639         }
1640
1641         cxgbe_vi_detach(&pi->vi[0]);
1642         callout_drain(&pi->tick);
1643         ifmedia_removeall(&pi->media);
1644
1645         end_synchronized_op(sc, 0);
1646
1647         return (0);
1648 }
1649
1650 static void
1651 cxgbe_init(void *arg)
1652 {
1653         struct vi_info *vi = arg;
1654         struct adapter *sc = vi->pi->adapter;
1655
1656         if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1657                 return;
1658         cxgbe_init_synchronized(vi);
1659         end_synchronized_op(sc, 0);
1660 }
1661
1662 static int
1663 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1664 {
1665         int rc = 0, mtu, flags;
1666         struct vi_info *vi = ifp->if_softc;
1667         struct port_info *pi = vi->pi;
1668         struct adapter *sc = pi->adapter;
1669         struct ifreq *ifr = (struct ifreq *)data;
1670         uint32_t mask;
1671
1672         switch (cmd) {
1673         case SIOCSIFMTU:
1674                 mtu = ifr->ifr_mtu;
1675                 if (mtu < ETHERMIN || mtu > MAX_MTU)
1676                         return (EINVAL);
1677
1678                 rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1679                 if (rc)
1680                         return (rc);
1681                 ifp->if_mtu = mtu;
1682                 if (vi->flags & VI_INIT_DONE) {
1683                         t4_update_fl_bufsize(ifp);
1684                         if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1685                                 rc = update_mac_settings(ifp, XGMAC_MTU);
1686                 }
1687                 end_synchronized_op(sc, 0);
1688                 break;
1689
1690         case SIOCSIFFLAGS:
1691                 rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4flg");
1692                 if (rc)
1693                         return (rc);
1694
1695                 if (ifp->if_flags & IFF_UP) {
1696                         if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1697                                 flags = vi->if_flags;
1698                                 if ((ifp->if_flags ^ flags) &
1699                                     (IFF_PROMISC | IFF_ALLMULTI)) {
1700                                         rc = update_mac_settings(ifp,
1701                                             XGMAC_PROMISC | XGMAC_ALLMULTI);
1702                                 }
1703                         } else {
1704                                 rc = cxgbe_init_synchronized(vi);
1705                         }
1706                         vi->if_flags = ifp->if_flags;
1707                 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1708                         rc = cxgbe_uninit_synchronized(vi);
1709                 }
1710                 end_synchronized_op(sc, 0);
1711                 break;
1712
1713         case SIOCADDMULTI:
1714         case SIOCDELMULTI:
1715                 rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4multi");
1716                 if (rc)
1717                         return (rc);
1718                 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1719                         rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1720                 end_synchronized_op(sc, 0);
1721                 break;
1722
1723         case SIOCSIFCAP:
1724                 rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1725                 if (rc)
1726                         return (rc);
1727
1728                 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1729                 if (mask & IFCAP_TXCSUM) {
1730                         ifp->if_capenable ^= IFCAP_TXCSUM;
1731                         ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1732
1733                         if (IFCAP_TSO4 & ifp->if_capenable &&
1734                             !(IFCAP_TXCSUM & ifp->if_capenable)) {
1735                                 ifp->if_capenable &= ~IFCAP_TSO4;
1736                                 if_printf(ifp,
1737                                     "tso4 disabled due to -txcsum.\n");
1738                         }
1739                 }
1740                 if (mask & IFCAP_TXCSUM_IPV6) {
1741                         ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1742                         ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1743
1744                         if (IFCAP_TSO6 & ifp->if_capenable &&
1745                             !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1746                                 ifp->if_capenable &= ~IFCAP_TSO6;
1747                                 if_printf(ifp,
1748                                     "tso6 disabled due to -txcsum6.\n");
1749                         }
1750                 }
1751                 if (mask & IFCAP_RXCSUM)
1752                         ifp->if_capenable ^= IFCAP_RXCSUM;
1753                 if (mask & IFCAP_RXCSUM_IPV6)
1754                         ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1755
1756                 /*
1757                  * Note that we leave CSUM_TSO alone (it is always set).  The
1758                  * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1759                  * sending a TSO request our way, so it's sufficient to toggle
1760                  * IFCAP_TSOx only.
1761                  */
1762                 if (mask & IFCAP_TSO4) {
1763                         if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1764                             !(IFCAP_TXCSUM & ifp->if_capenable)) {
1765                                 if_printf(ifp, "enable txcsum first.\n");
1766                                 rc = EAGAIN;
1767                                 goto fail;
1768                         }
1769                         ifp->if_capenable ^= IFCAP_TSO4;
1770                 }
1771                 if (mask & IFCAP_TSO6) {
1772                         if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1773                             !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1774                                 if_printf(ifp, "enable txcsum6 first.\n");
1775                                 rc = EAGAIN;
1776                                 goto fail;
1777                         }
1778                         ifp->if_capenable ^= IFCAP_TSO6;
1779                 }
1780                 if (mask & IFCAP_LRO) {
1781 #if defined(INET) || defined(INET6)
1782                         int i;
1783                         struct sge_rxq *rxq;
1784
1785                         ifp->if_capenable ^= IFCAP_LRO;
1786                         for_each_rxq(vi, i, rxq) {
1787                                 if (ifp->if_capenable & IFCAP_LRO)
1788                                         rxq->iq.flags |= IQ_LRO_ENABLED;
1789                                 else
1790                                         rxq->iq.flags &= ~IQ_LRO_ENABLED;
1791                         }
1792 #endif
1793                 }
1794 #ifdef TCP_OFFLOAD
1795                 if (mask & IFCAP_TOE) {
1796                         int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1797
1798                         rc = toe_capability(vi, enable);
1799                         if (rc != 0)
1800                                 goto fail;
1801
1802                         ifp->if_capenable ^= mask;
1803                 }
1804 #endif
1805                 if (mask & IFCAP_VLAN_HWTAGGING) {
1806                         ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1807                         if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1808                                 rc = update_mac_settings(ifp, XGMAC_VLANEX);
1809                 }
1810                 if (mask & IFCAP_VLAN_MTU) {
1811                         ifp->if_capenable ^= IFCAP_VLAN_MTU;
1812
1813                         /* Need to find out how to disable auto-mtu-inflation */
1814                 }
1815                 if (mask & IFCAP_VLAN_HWTSO)
1816                         ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1817                 if (mask & IFCAP_VLAN_HWCSUM)
1818                         ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1819 #ifdef RATELIMIT
1820                 if (mask & IFCAP_TXRTLMT)
1821                         ifp->if_capenable ^= IFCAP_TXRTLMT;
1822 #endif
1823
1824 #ifdef VLAN_CAPABILITIES
1825                 VLAN_CAPABILITIES(ifp);
1826 #endif
1827 fail:
1828                 end_synchronized_op(sc, 0);
1829                 break;
1830
1831         case SIOCSIFMEDIA:
1832         case SIOCGIFMEDIA:
1833         case SIOCGIFXMEDIA:
1834                 ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
1835                 break;
1836
1837         case SIOCGI2C: {
1838                 struct ifi2creq i2c;
1839
1840                 rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
1841                 if (rc != 0)
1842                         break;
1843                 if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1844                         rc = EPERM;
1845                         break;
1846                 }
1847                 if (i2c.len > sizeof(i2c.data)) {
1848                         rc = EINVAL;
1849                         break;
1850                 }
1851                 rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
1852                 if (rc)
1853                         return (rc);
1854                 rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
1855                     i2c.offset, i2c.len, &i2c.data[0]);
1856                 end_synchronized_op(sc, 0);
1857                 if (rc == 0)
1858                         rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
1859                 break;
1860         }
1861
1862         default:
1863                 rc = ether_ioctl(ifp, cmd, data);
1864         }
1865
1866         return (rc);
1867 }
1868
1869 static int
1870 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1871 {
1872         struct vi_info *vi = ifp->if_softc;
1873         struct port_info *pi = vi->pi;
1874         struct adapter *sc = pi->adapter;
1875         struct sge_txq *txq;
1876         void *items[1];
1877         int rc;
1878
1879         M_ASSERTPKTHDR(m);
1880         MPASS(m->m_nextpkt == NULL);    /* not quite ready for this yet */
1881
1882         if (__predict_false(pi->link_cfg.link_ok == 0)) {
1883                 m_freem(m);
1884                 return (ENETDOWN);
1885         }
1886
1887         rc = parse_pkt(sc, &m);
1888         if (__predict_false(rc != 0)) {
1889                 MPASS(m == NULL);                       /* was freed already */
1890                 atomic_add_int(&pi->tx_parse_error, 1); /* rare, atomic is ok */
1891                 return (rc);
1892         }
1893 #ifdef RATELIMIT
1894         if (m->m_pkthdr.snd_tag != NULL) {
1895                 /* EAGAIN tells the stack we are not the correct interface. */
1896                 if (__predict_false(ifp != m->m_pkthdr.snd_tag->ifp)) {
1897                         m_freem(m);
1898                         return (EAGAIN);
1899                 }
1900
1901                 return (ethofld_transmit(ifp, m));
1902         }
1903 #endif
1904
1905         /* Select a txq. */
1906         txq = &sc->sge.txq[vi->first_txq];
1907         if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1908                 txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
1909                     vi->rsrv_noflowq);
1910
1911         items[0] = m;
1912         rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1913         if (__predict_false(rc != 0))
1914                 m_freem(m);
1915
1916         return (rc);
1917 }
1918
1919 static void
1920 cxgbe_qflush(struct ifnet *ifp)
1921 {
1922         struct vi_info *vi = ifp->if_softc;
1923         struct sge_txq *txq;
1924         int i;
1925
1926         /* queues do not exist if !VI_INIT_DONE. */
1927         if (vi->flags & VI_INIT_DONE) {
1928                 for_each_txq(vi, i, txq) {
1929                         TXQ_LOCK(txq);
1930                         txq->eq.flags |= EQ_QFLUSH;
1931                         TXQ_UNLOCK(txq);
1932                         while (!mp_ring_is_idle(txq->r)) {
1933                                 mp_ring_check_drainage(txq->r, 0);
1934                                 pause("qflush", 1);
1935                         }
1936                         TXQ_LOCK(txq);
1937                         txq->eq.flags &= ~EQ_QFLUSH;
1938                         TXQ_UNLOCK(txq);
1939                 }
1940         }
1941         if_qflush(ifp);
1942 }
1943
1944 static uint64_t
1945 vi_get_counter(struct ifnet *ifp, ift_counter c)
1946 {
1947         struct vi_info *vi = ifp->if_softc;
1948         struct fw_vi_stats_vf *s = &vi->stats;
1949
1950         vi_refresh_stats(vi->pi->adapter, vi);
1951
1952         switch (c) {
1953         case IFCOUNTER_IPACKETS:
1954                 return (s->rx_bcast_frames + s->rx_mcast_frames +
1955                     s->rx_ucast_frames);
1956         case IFCOUNTER_IERRORS:
1957                 return (s->rx_err_frames);
1958         case IFCOUNTER_OPACKETS:
1959                 return (s->tx_bcast_frames + s->tx_mcast_frames +
1960                     s->tx_ucast_frames + s->tx_offload_frames);
1961         case IFCOUNTER_OERRORS:
1962                 return (s->tx_drop_frames);
1963         case IFCOUNTER_IBYTES:
1964                 return (s->rx_bcast_bytes + s->rx_mcast_bytes +
1965                     s->rx_ucast_bytes);
1966         case IFCOUNTER_OBYTES:
1967                 return (s->tx_bcast_bytes + s->tx_mcast_bytes +
1968                     s->tx_ucast_bytes + s->tx_offload_bytes);
1969         case IFCOUNTER_IMCASTS:
1970                 return (s->rx_mcast_frames);
1971         case IFCOUNTER_OMCASTS:
1972                 return (s->tx_mcast_frames);
1973         case IFCOUNTER_OQDROPS: {
1974                 uint64_t drops;
1975
1976                 drops = 0;
1977                 if (vi->flags & VI_INIT_DONE) {
1978                         int i;
1979                         struct sge_txq *txq;
1980
1981                         for_each_txq(vi, i, txq)
1982                                 drops += counter_u64_fetch(txq->r->drops);
1983                 }
1984
1985                 return (drops);
1986
1987         }
1988
1989         default:
1990                 return (if_get_counter_default(ifp, c));
1991         }
1992 }
1993
1994 uint64_t
1995 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
1996 {
1997         struct vi_info *vi = ifp->if_softc;
1998         struct port_info *pi = vi->pi;
1999         struct adapter *sc = pi->adapter;
2000         struct port_stats *s = &pi->stats;
2001
2002         if (pi->nvi > 1 || sc->flags & IS_VF)
2003                 return (vi_get_counter(ifp, c));
2004
2005         cxgbe_refresh_stats(sc, pi);
2006
2007         switch (c) {
2008         case IFCOUNTER_IPACKETS:
2009                 return (s->rx_frames);
2010
2011         case IFCOUNTER_IERRORS:
2012                 return (s->rx_jabber + s->rx_runt + s->rx_too_long +
2013                     s->rx_fcs_err + s->rx_len_err);
2014
2015         case IFCOUNTER_OPACKETS:
2016                 return (s->tx_frames);
2017
2018         case IFCOUNTER_OERRORS:
2019                 return (s->tx_error_frames);
2020
2021         case IFCOUNTER_IBYTES:
2022                 return (s->rx_octets);
2023
2024         case IFCOUNTER_OBYTES:
2025                 return (s->tx_octets);
2026
2027         case IFCOUNTER_IMCASTS:
2028                 return (s->rx_mcast_frames);
2029
2030         case IFCOUNTER_OMCASTS:
2031                 return (s->tx_mcast_frames);
2032
2033         case IFCOUNTER_IQDROPS:
2034                 return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
2035                     s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
2036                     s->rx_trunc3 + pi->tnl_cong_drops);
2037
2038         case IFCOUNTER_OQDROPS: {
2039                 uint64_t drops;
2040
2041                 drops = s->tx_drop;
2042                 if (vi->flags & VI_INIT_DONE) {
2043                         int i;
2044                         struct sge_txq *txq;
2045
2046                         for_each_txq(vi, i, txq)
2047                                 drops += counter_u64_fetch(txq->r->drops);
2048                 }
2049
2050                 return (drops);
2051
2052         }
2053
2054         default:
2055                 return (if_get_counter_default(ifp, c));
2056         }
2057 }
2058
2059 /*
2060  * The kernel picks a media from the list we had provided so we do not have to
2061  * validate the request.
2062  */
2063 static int
2064 cxgbe_media_change(struct ifnet *ifp)
2065 {
2066         struct vi_info *vi = ifp->if_softc;
2067         struct port_info *pi = vi->pi;
2068         struct ifmedia *ifm = &pi->media;
2069         struct link_config *lc = &pi->link_cfg;
2070         struct adapter *sc = pi->adapter;
2071         int rc;
2072
2073         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4mec");
2074         if (rc != 0)
2075                 return (rc);
2076         PORT_LOCK(pi);
2077         if (IFM_SUBTYPE(ifm->ifm_media) == IFM_AUTO) {
2078                 MPASS(lc->supported & FW_PORT_CAP_ANEG);
2079                 lc->requested_aneg = AUTONEG_ENABLE;
2080         } else {
2081                 lc->requested_aneg = AUTONEG_DISABLE;
2082                 lc->requested_speed =
2083                     ifmedia_baudrate(ifm->ifm_media) / 1000000;
2084                 lc->requested_fc = 0;
2085                 if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_RXPAUSE)
2086                         lc->requested_fc |= PAUSE_RX;
2087                 if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_TXPAUSE)
2088                         lc->requested_fc |= PAUSE_TX;
2089         }
2090         if (pi->up_vis > 0)
2091                 rc = apply_l1cfg(pi);
2092         PORT_UNLOCK(pi);
2093         end_synchronized_op(sc, 0);
2094         return (rc);
2095 }
2096
2097 /*
2098  * Mbps to FW_PORT_CAP_SPEED_* bit.
2099  */
2100 static uint16_t
2101 speed_to_fwspeed(int speed)
2102 {
2103
2104         switch (speed) {
2105         case 100000:
2106                 return (FW_PORT_CAP_SPEED_100G);
2107         case 40000:
2108                 return (FW_PORT_CAP_SPEED_40G);
2109         case 25000:
2110                 return (FW_PORT_CAP_SPEED_25G);
2111         case 10000:
2112                 return (FW_PORT_CAP_SPEED_10G);
2113         case 1000:
2114                 return (FW_PORT_CAP_SPEED_1G);
2115         case 100:
2116                 return (FW_PORT_CAP_SPEED_100M);
2117         }
2118
2119         return (0);
2120 }
2121
2122 /*
2123  * Base media word (without ETHER, pause, link active, etc.) for the port at the
2124  * given speed.
2125  */
2126 static int
2127 port_mword(struct port_info *pi, uint16_t speed)
2128 {
2129
2130         MPASS(speed & M_FW_PORT_CAP_SPEED);
2131         MPASS(powerof2(speed));
2132
2133         switch(pi->port_type) {
2134         case FW_PORT_TYPE_BT_SGMII:
2135         case FW_PORT_TYPE_BT_XFI:
2136         case FW_PORT_TYPE_BT_XAUI:
2137                 /* BaseT */
2138                 switch (speed) {
2139                 case FW_PORT_CAP_SPEED_100M:
2140                         return (IFM_100_T);
2141                 case FW_PORT_CAP_SPEED_1G:
2142                         return (IFM_1000_T);
2143                 case FW_PORT_CAP_SPEED_10G:
2144                         return (IFM_10G_T);
2145                 }
2146                 break;
2147         case FW_PORT_TYPE_KX4:
2148                 if (speed == FW_PORT_CAP_SPEED_10G)
2149                         return (IFM_10G_KX4);
2150                 break;
2151         case FW_PORT_TYPE_CX4:
2152                 if (speed == FW_PORT_CAP_SPEED_10G)
2153                         return (IFM_10G_CX4);
2154                 break;
2155         case FW_PORT_TYPE_KX:
2156                 if (speed == FW_PORT_CAP_SPEED_1G)
2157                         return (IFM_1000_KX);
2158                 break;
2159         case FW_PORT_TYPE_KR:
2160         case FW_PORT_TYPE_BP_AP:
2161         case FW_PORT_TYPE_BP4_AP:
2162         case FW_PORT_TYPE_BP40_BA:
2163         case FW_PORT_TYPE_KR4_100G:
2164         case FW_PORT_TYPE_KR_SFP28:
2165         case FW_PORT_TYPE_KR_XLAUI:
2166                 switch (speed) {
2167                 case FW_PORT_CAP_SPEED_1G:
2168                         return (IFM_1000_KX);
2169                 case FW_PORT_CAP_SPEED_10G:
2170                         return (IFM_10G_KR);
2171                 case FW_PORT_CAP_SPEED_25G:
2172                         return (IFM_25G_KR);
2173                 case FW_PORT_CAP_SPEED_40G:
2174                         return (IFM_40G_KR4);
2175                 case FW_PORT_CAP_SPEED_100G:
2176                         return (IFM_100G_KR4);
2177                 }
2178                 break;
2179         case FW_PORT_TYPE_FIBER_XFI:
2180         case FW_PORT_TYPE_FIBER_XAUI:
2181         case FW_PORT_TYPE_SFP:
2182         case FW_PORT_TYPE_QSFP_10G:
2183         case FW_PORT_TYPE_QSA:
2184         case FW_PORT_TYPE_QSFP:
2185         case FW_PORT_TYPE_CR4_QSFP:
2186         case FW_PORT_TYPE_CR_QSFP:
2187         case FW_PORT_TYPE_CR2_QSFP:
2188         case FW_PORT_TYPE_SFP28:
2189                 /* Pluggable transceiver */
2190                 switch (pi->mod_type) {
2191                 case FW_PORT_MOD_TYPE_LR:
2192                         switch (speed) {
2193                         case FW_PORT_CAP_SPEED_1G:
2194                                 return (IFM_1000_LX);
2195                         case FW_PORT_CAP_SPEED_10G:
2196                                 return (IFM_10G_LR);
2197                         case FW_PORT_CAP_SPEED_25G:
2198                                 return (IFM_25G_LR);
2199                         case FW_PORT_CAP_SPEED_40G:
2200                                 return (IFM_40G_LR4);
2201                         case FW_PORT_CAP_SPEED_100G:
2202                                 return (IFM_100G_LR4);
2203                         }
2204                         break;
2205                 case FW_PORT_MOD_TYPE_SR:
2206                         switch (speed) {
2207                         case FW_PORT_CAP_SPEED_1G:
2208                                 return (IFM_1000_SX);
2209                         case FW_PORT_CAP_SPEED_10G:
2210                                 return (IFM_10G_SR);
2211                         case FW_PORT_CAP_SPEED_25G:
2212                                 return (IFM_25G_SR);
2213                         case FW_PORT_CAP_SPEED_40G:
2214                                 return (IFM_40G_SR4);
2215                         case FW_PORT_CAP_SPEED_100G:
2216                                 return (IFM_100G_SR4);
2217                         }
2218                         break;
2219                 case FW_PORT_MOD_TYPE_ER:
2220                         if (speed == FW_PORT_CAP_SPEED_10G)
2221                                 return (IFM_10G_ER);
2222                         break;
2223                 case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
2224                 case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
2225                         switch (speed) {
2226                         case FW_PORT_CAP_SPEED_1G:
2227                                 return (IFM_1000_CX);
2228                         case FW_PORT_CAP_SPEED_10G:
2229                                 return (IFM_10G_TWINAX);
2230                         case FW_PORT_CAP_SPEED_25G:
2231                                 return (IFM_25G_CR);
2232                         case FW_PORT_CAP_SPEED_40G:
2233                                 return (IFM_40G_CR4);
2234                         case FW_PORT_CAP_SPEED_100G:
2235                                 return (IFM_100G_CR4);
2236                         }
2237                         break;
2238                 case FW_PORT_MOD_TYPE_LRM:
2239                         if (speed == FW_PORT_CAP_SPEED_10G)
2240                                 return (IFM_10G_LRM);
2241                         break;
2242                 case FW_PORT_MOD_TYPE_NA:
2243                         MPASS(0);       /* Not pluggable? */
2244                         /* fall throough */
2245                 case FW_PORT_MOD_TYPE_ERROR:
2246                 case FW_PORT_MOD_TYPE_UNKNOWN:
2247                 case FW_PORT_MOD_TYPE_NOTSUPPORTED:
2248                         break;
2249                 case FW_PORT_MOD_TYPE_NONE:
2250                         return (IFM_NONE);
2251                 }
2252                 break;
2253         case FW_PORT_TYPE_NONE:
2254                 return (IFM_NONE);
2255         }
2256
2257         return (IFM_UNKNOWN);
2258 }
2259
2260 static void
2261 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2262 {
2263         struct vi_info *vi = ifp->if_softc;
2264         struct port_info *pi = vi->pi;
2265         struct adapter *sc = pi->adapter;
2266         struct link_config *lc = &pi->link_cfg;
2267
2268         if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4med") != 0)
2269                 return;
2270         PORT_LOCK(pi);
2271
2272         if (pi->up_vis == 0) {
2273                 /*
2274                  * If all the interfaces are administratively down the firmware
2275                  * does not report transceiver changes.  Refresh port info here
2276                  * so that ifconfig displays accurate ifmedia at all times.
2277                  * This is the only reason we have a synchronized op in this
2278                  * function.  Just PORT_LOCK would have been enough otherwise.
2279                  */
2280                 t4_update_port_info(pi);
2281                 build_medialist(pi, &pi->media);
2282         }
2283
2284         /* ifm_status */
2285         ifmr->ifm_status = IFM_AVALID;
2286         if (lc->link_ok == 0)
2287                 goto done;
2288         ifmr->ifm_status |= IFM_ACTIVE;
2289
2290         /* ifm_active */
2291         ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2292         ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2293         if (lc->fc & PAUSE_RX)
2294                 ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2295         if (lc->fc & PAUSE_TX)
2296                 ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2297         ifmr->ifm_active |= port_mword(pi, speed_to_fwspeed(lc->speed));
2298 done:
2299         PORT_UNLOCK(pi);
2300         end_synchronized_op(sc, 0);
2301 }
2302
2303 static int
2304 vcxgbe_probe(device_t dev)
2305 {
2306         char buf[128];
2307         struct vi_info *vi = device_get_softc(dev);
2308
2309         snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2310             vi - vi->pi->vi);
2311         device_set_desc_copy(dev, buf);
2312
2313         return (BUS_PROBE_DEFAULT);
2314 }
2315
2316 static int
2317 alloc_extra_vi(struct adapter *sc, struct port_info *pi, struct vi_info *vi)
2318 {
2319         int func, index, rc;
2320         uint32_t param, val;
2321
2322         ASSERT_SYNCHRONIZED_OP(sc);
2323
2324         index = vi - pi->vi;
2325         MPASS(index > 0);       /* This function deals with _extra_ VIs only */
2326         KASSERT(index < nitems(vi_mac_funcs),
2327             ("%s: VI %s doesn't have a MAC func", __func__,
2328             device_get_nameunit(vi->dev)));
2329         func = vi_mac_funcs[index];
2330         rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2331             vi->hw_addr, &vi->rss_size, func, 0);
2332         if (rc < 0) {
2333                 device_printf(vi->dev, "failed to allocate virtual interface %d"
2334                     "for port %d: %d\n", index, pi->port_id, -rc);
2335                 return (-rc);
2336         }
2337         vi->viid = rc;
2338         if (chip_id(sc) <= CHELSIO_T5)
2339                 vi->smt_idx = (rc & 0x7f) << 1;
2340         else
2341                 vi->smt_idx = (rc & 0x7f);
2342
2343         if (vi->rss_size == 1) {
2344                 /*
2345                  * This VI didn't get a slice of the RSS table.  Reduce the
2346                  * number of VIs being created (hw.cxgbe.num_vis) or modify the
2347                  * configuration file (nvi, rssnvi for this PF) if this is a
2348                  * problem.
2349                  */
2350                 device_printf(vi->dev, "RSS table not available.\n");
2351                 vi->rss_base = 0xffff;
2352
2353                 return (0);
2354         }
2355
2356         param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2357             V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2358             V_FW_PARAMS_PARAM_YZ(vi->viid);
2359         rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2360         if (rc)
2361                 vi->rss_base = 0xffff;
2362         else {
2363                 MPASS((val >> 16) == vi->rss_size);
2364                 vi->rss_base = val & 0xffff;
2365         }
2366
2367         return (0);
2368 }
2369
2370 static int
2371 vcxgbe_attach(device_t dev)
2372 {
2373         struct vi_info *vi;
2374         struct port_info *pi;
2375         struct adapter *sc;
2376         int rc;
2377
2378         vi = device_get_softc(dev);
2379         pi = vi->pi;
2380         sc = pi->adapter;
2381
2382         rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4via");
2383         if (rc)
2384                 return (rc);
2385         rc = alloc_extra_vi(sc, pi, vi);
2386         end_synchronized_op(sc, 0);
2387         if (rc)
2388                 return (rc);
2389
2390         rc = cxgbe_vi_attach(dev, vi);
2391         if (rc) {
2392                 t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2393                 return (rc);
2394         }
2395         return (0);
2396 }
2397
2398 static int
2399 vcxgbe_detach(device_t dev)
2400 {
2401         struct vi_info *vi;
2402         struct adapter *sc;
2403
2404         vi = device_get_softc(dev);
2405         sc = vi->pi->adapter;
2406
2407         doom_vi(sc, vi);
2408
2409         cxgbe_vi_detach(vi);
2410         t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2411
2412         end_synchronized_op(sc, 0);
2413
2414         return (0);
2415 }
2416
2417 void
2418 t4_fatal_err(struct adapter *sc)
2419 {
2420         t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2421         t4_intr_disable(sc);
2422         log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
2423             device_get_nameunit(sc->dev));
2424         if (t4_panic_on_fatal_err)
2425                 panic("panic requested on fatal error");
2426 }
2427
2428 void
2429 t4_add_adapter(struct adapter *sc)
2430 {
2431         sx_xlock(&t4_list_lock);
2432         SLIST_INSERT_HEAD(&t4_list, sc, link);
2433         sx_xunlock(&t4_list_lock);
2434 }
2435
2436 int
2437 t4_map_bars_0_and_4(struct adapter *sc)
2438 {
2439         sc->regs_rid = PCIR_BAR(0);
2440         sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2441             &sc->regs_rid, RF_ACTIVE);
2442         if (sc->regs_res == NULL) {
2443                 device_printf(sc->dev, "cannot map registers.\n");
2444                 return (ENXIO);
2445         }
2446         sc->bt = rman_get_bustag(sc->regs_res);
2447         sc->bh = rman_get_bushandle(sc->regs_res);
2448         sc->mmio_len = rman_get_size(sc->regs_res);
2449         setbit(&sc->doorbells, DOORBELL_KDB);
2450
2451         sc->msix_rid = PCIR_BAR(4);
2452         sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2453             &sc->msix_rid, RF_ACTIVE);
2454         if (sc->msix_res == NULL) {
2455                 device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2456                 return (ENXIO);
2457         }
2458
2459         return (0);
2460 }
2461
2462 int
2463 t4_map_bar_2(struct adapter *sc)
2464 {
2465
2466         /*
2467          * T4: only iWARP driver uses the userspace doorbells.  There is no need
2468          * to map it if RDMA is disabled.
2469          */
2470         if (is_t4(sc) && sc->rdmacaps == 0)
2471                 return (0);
2472
2473         sc->udbs_rid = PCIR_BAR(2);
2474         sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2475             &sc->udbs_rid, RF_ACTIVE);
2476         if (sc->udbs_res == NULL) {
2477                 device_printf(sc->dev, "cannot map doorbell BAR.\n");
2478                 return (ENXIO);
2479         }
2480         sc->udbs_base = rman_get_virtual(sc->udbs_res);
2481
2482         if (chip_id(sc) >= CHELSIO_T5) {
2483                 setbit(&sc->doorbells, DOORBELL_UDB);
2484 #if defined(__i386__) || defined(__amd64__)
2485                 if (t5_write_combine) {
2486                         int rc, mode;
2487
2488                         /*
2489                          * Enable write combining on BAR2.  This is the
2490                          * userspace doorbell BAR and is split into 128B
2491                          * (UDBS_SEG_SIZE) doorbell regions, each associated
2492                          * with an egress queue.  The first 64B has the doorbell
2493                          * and the second 64B can be used to submit a tx work
2494                          * request with an implicit doorbell.
2495                          */
2496
2497                         rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2498                             rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2499                         if (rc == 0) {
2500                                 clrbit(&sc->doorbells, DOORBELL_UDB);
2501                                 setbit(&sc->doorbells, DOORBELL_WCWR);
2502                                 setbit(&sc->doorbells, DOORBELL_UDBWC);
2503                         } else {
2504                                 device_printf(sc->dev,
2505                                     "couldn't enable write combining: %d\n",
2506                                     rc);
2507                         }
2508
2509                         mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2510                         t4_write_reg(sc, A_SGE_STAT_CFG,
2511                             V_STATSOURCE_T5(7) | mode);
2512                 }
2513 #endif
2514         }
2515         sc->iwt.wc_en = isset(&sc->doorbells, DOORBELL_UDBWC) ? 1 : 0;
2516
2517         return (0);
2518 }
2519
2520 struct memwin_init {
2521         uint32_t base;
2522         uint32_t aperture;
2523 };
2524
2525 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2526         { MEMWIN0_BASE, MEMWIN0_APERTURE },
2527         { MEMWIN1_BASE, MEMWIN1_APERTURE },
2528         { MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2529 };
2530
2531 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2532         { MEMWIN0_BASE, MEMWIN0_APERTURE },
2533         { MEMWIN1_BASE, MEMWIN1_APERTURE },
2534         { MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2535 };
2536
2537 static void
2538 setup_memwin(struct adapter *sc)
2539 {
2540         const struct memwin_init *mw_init;
2541         struct memwin *mw;
2542         int i;
2543         uint32_t bar0;
2544
2545         if (is_t4(sc)) {
2546                 /*
2547                  * Read low 32b of bar0 indirectly via the hardware backdoor
2548                  * mechanism.  Works from within PCI passthrough environments
2549                  * too, where rman_get_start() can return a different value.  We
2550                  * need to program the T4 memory window decoders with the actual
2551                  * addresses that will be coming across the PCIe link.
2552                  */
2553                 bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2554                 bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2555
2556                 mw_init = &t4_memwin[0];
2557         } else {
2558                 /* T5+ use the relative offset inside the PCIe BAR */
2559                 bar0 = 0;
2560
2561                 mw_init = &t5_memwin[0];
2562         }
2563
2564         for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2565                 rw_init(&mw->mw_lock, "memory window access");
2566                 mw->mw_base = mw_init->base;
2567                 mw->mw_aperture = mw_init->aperture;
2568                 mw->mw_curpos = 0;
2569                 t4_write_reg(sc,
2570                     PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2571                     (mw->mw_base + bar0) | V_BIR(0) |
2572                     V_WINDOW(ilog2(mw->mw_aperture) - 10));
2573                 rw_wlock(&mw->mw_lock);
2574                 position_memwin(sc, i, 0);
2575                 rw_wunlock(&mw->mw_lock);
2576         }
2577
2578         /* flush */
2579         t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2580 }
2581
2582 /*
2583  * Positions the memory window at the given address in the card's address space.
2584  * There are some alignment requirements and the actual position may be at an
2585  * address prior to the requested address.  mw->mw_curpos always has the actual
2586  * position of the window.
2587  */
2588 static void
2589 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2590 {
2591         struct memwin *mw;
2592         uint32_t pf;
2593         uint32_t reg;
2594
2595         MPASS(idx >= 0 && idx < NUM_MEMWIN);
2596         mw = &sc->memwin[idx];
2597         rw_assert(&mw->mw_lock, RA_WLOCKED);
2598
2599         if (is_t4(sc)) {
2600                 pf = 0;
2601                 mw->mw_curpos = addr & ~0xf;    /* start must be 16B aligned */
2602         } else {
2603                 pf = V_PFNUM(sc->pf);
2604                 mw->mw_curpos = addr & ~0x7f;   /* start must be 128B aligned */
2605         }
2606         reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2607         t4_write_reg(sc, reg, mw->mw_curpos | pf);
2608         t4_read_reg(sc, reg);   /* flush */
2609 }
2610
2611 int
2612 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2613     int len, int rw)
2614 {
2615         struct memwin *mw;
2616         uint32_t mw_end, v;
2617
2618         MPASS(idx >= 0 && idx < NUM_MEMWIN);
2619
2620         /* Memory can only be accessed in naturally aligned 4 byte units */
2621         if (addr & 3 || len & 3 || len <= 0)
2622                 return (EINVAL);
2623
2624         mw = &sc->memwin[idx];
2625         while (len > 0) {
2626                 rw_rlock(&mw->mw_lock);
2627                 mw_end = mw->mw_curpos + mw->mw_aperture;
2628                 if (addr >= mw_end || addr < mw->mw_curpos) {
2629                         /* Will need to reposition the window */
2630                         if (!rw_try_upgrade(&mw->mw_lock)) {
2631                                 rw_runlock(&mw->mw_lock);
2632                                 rw_wlock(&mw->mw_lock);
2633                         }
2634                         rw_assert(&mw->mw_lock, RA_WLOCKED);
2635                         position_memwin(sc, idx, addr);
2636                         rw_downgrade(&mw->mw_lock);
2637                         mw_end = mw->mw_curpos + mw->mw_aperture;
2638                 }
2639                 rw_assert(&mw->mw_lock, RA_RLOCKED);
2640                 while (addr < mw_end && len > 0) {
2641                         if (rw == 0) {
2642                                 v = t4_read_reg(sc, mw->mw_base + addr -
2643                                     mw->mw_curpos);
2644                                 *val++ = le32toh(v);
2645                         } else {
2646                                 v = *val++;
2647                                 t4_write_reg(sc, mw->mw_base + addr -
2648                                     mw->mw_curpos, htole32(v));
2649                         }
2650                         addr += 4;
2651                         len -= 4;
2652                 }
2653                 rw_runlock(&mw->mw_lock);
2654         }
2655
2656         return (0);
2657 }
2658
2659 int
2660 alloc_atid_tab(struct tid_info *t, int flags)
2661 {
2662         int i;
2663
2664         MPASS(t->natids > 0);
2665         MPASS(t->atid_tab == NULL);
2666
2667         t->atid_tab = malloc(t->natids * sizeof(*t->atid_tab), M_CXGBE,
2668             M_ZERO | flags);
2669         if (t->atid_tab == NULL)
2670                 return (ENOMEM);
2671         mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
2672         t->afree = t->atid_tab;
2673         t->atids_in_use = 0;
2674         for (i = 1; i < t->natids; i++)
2675                 t->atid_tab[i - 1].next = &t->atid_tab[i];
2676         t->atid_tab[t->natids - 1].next = NULL;
2677
2678         return (0);
2679 }
2680
2681 void
2682 free_atid_tab(struct tid_info *t)
2683 {
2684
2685         KASSERT(t->atids_in_use == 0,
2686             ("%s: %d atids still in use.", __func__, t->atids_in_use));
2687
2688         if (mtx_initialized(&t->atid_lock))
2689                 mtx_destroy(&t->atid_lock);
2690         free(t->atid_tab, M_CXGBE);
2691         t->atid_tab = NULL;
2692 }
2693
2694 int
2695 alloc_atid(struct adapter *sc, void *ctx)
2696 {
2697         struct tid_info *t = &sc->tids;
2698         int atid = -1;
2699
2700         mtx_lock(&t->atid_lock);
2701         if (t->afree) {
2702                 union aopen_entry *p = t->afree;
2703
2704                 atid = p - t->atid_tab;
2705                 MPASS(atid <= M_TID_TID);
2706                 t->afree = p->next;
2707                 p->data = ctx;
2708                 t->atids_in_use++;
2709         }
2710         mtx_unlock(&t->atid_lock);
2711         return (atid);
2712 }
2713
2714 void *
2715 lookup_atid(struct adapter *sc, int atid)
2716 {
2717         struct tid_info *t = &sc->tids;
2718
2719         return (t->atid_tab[atid].data);
2720 }
2721
2722 void
2723 free_atid(struct adapter *sc, int atid)
2724 {
2725         struct tid_info *t = &sc->tids;
2726         union aopen_entry *p = &t->atid_tab[atid];
2727
2728         mtx_lock(&t->atid_lock);
2729         p->next = t->afree;
2730         t->afree = p;
2731         t->atids_in_use--;
2732         mtx_unlock(&t->atid_lock);
2733 }
2734
2735 static void
2736 queue_tid_release(struct adapter *sc, int tid)
2737 {
2738
2739         CXGBE_UNIMPLEMENTED("deferred tid release");
2740 }
2741
2742 void
2743 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
2744 {
2745         struct wrqe *wr;
2746         struct cpl_tid_release *req;
2747
2748         wr = alloc_wrqe(sizeof(*req), ctrlq);
2749         if (wr == NULL) {
2750                 queue_tid_release(sc, tid);     /* defer */
2751                 return;
2752         }
2753         req = wrtod(wr);
2754
2755         INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
2756
2757         t4_wrq_tx(sc, wr);
2758 }
2759
2760 static int
2761 t4_range_cmp(const void *a, const void *b)
2762 {
2763         return ((const struct t4_range *)a)->start -
2764                ((const struct t4_range *)b)->start;
2765 }
2766
2767 /*
2768  * Verify that the memory range specified by the addr/len pair is valid within
2769  * the card's address space.
2770  */
2771 static int
2772 validate_mem_range(struct adapter *sc, uint32_t addr, int len)
2773 {
2774         struct t4_range mem_ranges[4], *r, *next;
2775         uint32_t em, addr_len;
2776         int i, n, remaining;
2777
2778         /* Memory can only be accessed in naturally aligned 4 byte units */
2779         if (addr & 3 || len & 3 || len <= 0)
2780                 return (EINVAL);
2781
2782         /* Enabled memories */
2783         em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2784
2785         r = &mem_ranges[0];
2786         n = 0;
2787         bzero(r, sizeof(mem_ranges));
2788         if (em & F_EDRAM0_ENABLE) {
2789                 addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2790                 r->size = G_EDRAM0_SIZE(addr_len) << 20;
2791                 if (r->size > 0) {
2792                         r->start = G_EDRAM0_BASE(addr_len) << 20;
2793                         if (addr >= r->start &&
2794                             addr + len <= r->start + r->size)
2795                                 return (0);
2796                         r++;
2797                         n++;
2798                 }
2799         }
2800         if (em & F_EDRAM1_ENABLE) {
2801                 addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2802                 r->size = G_EDRAM1_SIZE(addr_len) << 20;
2803                 if (r->size > 0) {
2804                         r->start = G_EDRAM1_BASE(addr_len) << 20;
2805                         if (addr >= r->start &&
2806                             addr + len <= r->start + r->size)
2807                                 return (0);
2808                         r++;
2809                         n++;
2810                 }
2811         }
2812         if (em & F_EXT_MEM_ENABLE) {
2813                 addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2814                 r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2815                 if (r->size > 0) {
2816                         r->start = G_EXT_MEM_BASE(addr_len) << 20;
2817                         if (addr >= r->start &&
2818                             addr + len <= r->start + r->size)
2819                                 return (0);
2820                         r++;
2821                         n++;
2822                 }
2823         }
2824         if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2825                 addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2826                 r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2827                 if (r->size > 0) {
2828                         r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2829                         if (addr >= r->start &&
2830                             addr + len <= r->start + r->size)
2831                                 return (0);
2832                         r++;
2833                         n++;
2834                 }
2835         }
2836         MPASS(n <= nitems(mem_ranges));
2837
2838         if (n > 1) {
2839                 /* Sort and merge the ranges. */
2840                 qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2841
2842                 /* Start from index 0 and examine the next n - 1 entries. */
2843                 r = &mem_ranges[0];
2844                 for (remaining = n - 1; remaining > 0; remaining--, r++) {
2845
2846                         MPASS(r->size > 0);     /* r is a valid entry. */
2847                         next = r + 1;
2848                         MPASS(next->size > 0);  /* and so is the next one. */
2849
2850                         while (r->start + r->size >= next->start) {
2851                                 /* Merge the next one into the current entry. */
2852                                 r->size = max(r->start + r->size,
2853                                     next->start + next->size) - r->start;
2854                                 n--;    /* One fewer entry in total. */
2855                                 if (--remaining == 0)
2856                                         goto done;      /* short circuit */
2857                                 next++;
2858                         }
2859                         if (next != r + 1) {
2860                                 /*
2861                                  * Some entries were merged into r and next
2862                                  * points to the first valid entry that couldn't
2863                                  * be merged.
2864                                  */
2865                                 MPASS(next->size > 0);  /* must be valid */
2866                                 memcpy(r + 1, next, remaining * sizeof(*r));
2867 #ifdef INVARIANTS
2868                                 /*
2869                                  * This so that the foo->size assertion in the
2870                                  * next iteration of the loop do the right
2871                                  * thing for entries that were pulled up and are
2872                                  * no longer valid.
2873                                  */
2874                                 MPASS(n < nitems(mem_ranges));
2875                                 bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
2876                                     sizeof(struct t4_range));
2877 #endif
2878                         }
2879                 }
2880 done:
2881                 /* Done merging the ranges. */
2882                 MPASS(n > 0);
2883                 r = &mem_ranges[0];
2884                 for (i = 0; i < n; i++, r++) {
2885                         if (addr >= r->start &&
2886                             addr + len <= r->start + r->size)
2887                                 return (0);
2888                 }
2889         }
2890
2891         return (EFAULT);
2892 }
2893
2894 static int
2895 fwmtype_to_hwmtype(int mtype)
2896 {
2897
2898         switch (mtype) {
2899         case FW_MEMTYPE_EDC0:
2900                 return (MEM_EDC0);
2901         case FW_MEMTYPE_EDC1:
2902                 return (MEM_EDC1);
2903         case FW_MEMTYPE_EXTMEM:
2904                 return (MEM_MC0);
2905         case FW_MEMTYPE_EXTMEM1:
2906                 return (MEM_MC1);
2907         default:
2908                 panic("%s: cannot translate fw mtype %d.", __func__, mtype);
2909         }
2910 }
2911
2912 /*
2913  * Verify that the memory range specified by the memtype/offset/len pair is
2914  * valid and lies entirely within the memtype specified.  The global address of
2915  * the start of the range is returned in addr.
2916  */
2917 static int
2918 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
2919     uint32_t *addr)
2920 {
2921         uint32_t em, addr_len, maddr;
2922
2923         /* Memory can only be accessed in naturally aligned 4 byte units */
2924         if (off & 3 || len & 3 || len == 0)
2925                 return (EINVAL);
2926
2927         em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2928         switch (fwmtype_to_hwmtype(mtype)) {
2929         case MEM_EDC0:
2930                 if (!(em & F_EDRAM0_ENABLE))
2931                         return (EINVAL);
2932                 addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2933                 maddr = G_EDRAM0_BASE(addr_len) << 20;
2934                 break;
2935         case MEM_EDC1:
2936                 if (!(em & F_EDRAM1_ENABLE))
2937                         return (EINVAL);
2938                 addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2939                 maddr = G_EDRAM1_BASE(addr_len) << 20;
2940                 break;
2941         case MEM_MC:
2942                 if (!(em & F_EXT_MEM_ENABLE))
2943                         return (EINVAL);
2944                 addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2945                 maddr = G_EXT_MEM_BASE(addr_len) << 20;
2946                 break;
2947         case MEM_MC1:
2948                 if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
2949                         return (EINVAL);
2950                 addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2951                 maddr = G_EXT_MEM1_BASE(addr_len) << 20;
2952                 break;
2953         default:
2954                 return (EINVAL);
2955         }
2956
2957         *addr = maddr + off;    /* global address */
2958         return (validate_mem_range(sc, *addr, len));
2959 }
2960
2961 static int
2962 fixup_devlog_params(struct adapter *sc)
2963 {
2964         struct devlog_params *dparams = &sc->params.devlog;
2965         int rc;
2966
2967         rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
2968             dparams->size, &dparams->addr);
2969
2970         return (rc);
2971 }
2972
2973 static void
2974 update_nirq(struct intrs_and_queues *iaq, int nports)
2975 {
2976         int extra = T4_EXTRA_INTR;
2977
2978         iaq->nirq = extra;
2979         iaq->nirq += nports * (iaq->nrxq + iaq->nofldrxq);
2980         iaq->nirq += nports * (iaq->num_vis - 1) *
2981             max(iaq->nrxq_vi, iaq->nnmrxq_vi);
2982         iaq->nirq += nports * (iaq->num_vis - 1) * iaq->nofldrxq_vi;
2983 }
2984
2985 /*
2986  * Adjust requirements to fit the number of interrupts available.
2987  */
2988 static void
2989 calculate_iaq(struct adapter *sc, struct intrs_and_queues *iaq, int itype,
2990     int navail)
2991 {
2992         int old_nirq;
2993         const int nports = sc->params.nports;
2994
2995         MPASS(nports > 0);
2996         MPASS(navail > 0);
2997
2998         bzero(iaq, sizeof(*iaq));
2999         iaq->intr_type = itype;
3000         iaq->num_vis = t4_num_vis;
3001         iaq->ntxq = t4_ntxq;
3002         iaq->ntxq_vi = t4_ntxq_vi;
3003         iaq->nrxq = t4_nrxq;
3004         iaq->nrxq_vi = t4_nrxq_vi;
3005 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
3006         if (is_offload(sc) || is_ethoffload(sc)) {
3007                 iaq->nofldtxq = t4_nofldtxq;
3008                 iaq->nofldtxq_vi = t4_nofldtxq_vi;
3009         }
3010 #endif
3011 #ifdef TCP_OFFLOAD
3012         if (is_offload(sc)) {
3013                 iaq->nofldrxq = t4_nofldrxq;
3014                 iaq->nofldrxq_vi = t4_nofldrxq_vi;
3015         }
3016 #endif
3017 #ifdef DEV_NETMAP
3018         iaq->nnmtxq_vi = t4_nnmtxq_vi;
3019         iaq->nnmrxq_vi = t4_nnmrxq_vi;
3020 #endif
3021
3022         update_nirq(iaq, nports);
3023         if (iaq->nirq <= navail &&
3024             (itype != INTR_MSI || powerof2(iaq->nirq))) {
3025                 /*
3026                  * This is the normal case -- there are enough interrupts for
3027                  * everything.
3028                  */
3029                 goto done;
3030         }
3031
3032         /*
3033          * If extra VIs have been configured try reducing their count and see if
3034          * that works.
3035          */
3036         while (iaq->num_vis > 1) {
3037                 iaq->num_vis--;
3038                 update_nirq(iaq, nports);
3039                 if (iaq->nirq <= navail &&
3040                     (itype != INTR_MSI || powerof2(iaq->nirq))) {
3041                         device_printf(sc->dev, "virtual interfaces per port "
3042                             "reduced to %d from %d.  nrxq=%u, nofldrxq=%u, "
3043                             "nrxq_vi=%u nofldrxq_vi=%u, nnmrxq_vi=%u.  "
3044                             "itype %d, navail %u, nirq %d.\n",
3045                             iaq->num_vis, t4_num_vis, iaq->nrxq, iaq->nofldrxq,
3046                             iaq->nrxq_vi, iaq->nofldrxq_vi, iaq->nnmrxq_vi,
3047                             itype, navail, iaq->nirq);
3048                         goto done;
3049                 }
3050         }
3051
3052         /*
3053          * Extra VIs will not be created.  Log a message if they were requested.
3054          */
3055         MPASS(iaq->num_vis == 1);
3056         iaq->ntxq_vi = iaq->nrxq_vi = 0;
3057         iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
3058         iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
3059         if (iaq->num_vis != t4_num_vis) {
3060                 device_printf(sc->dev, "extra virtual interfaces disabled.  "
3061                     "nrxq=%u, nofldrxq=%u, nrxq_vi=%u nofldrxq_vi=%u, "
3062                     "nnmrxq_vi=%u.  itype %d, navail %u, nirq %d.\n",
3063                     iaq->nrxq, iaq->nofldrxq, iaq->nrxq_vi, iaq->nofldrxq_vi,
3064                     iaq->nnmrxq_vi, itype, navail, iaq->nirq);
3065         }
3066
3067         /*
3068          * Keep reducing the number of NIC rx queues to the next lower power of
3069          * 2 (for even RSS distribution) and halving the TOE rx queues and see
3070          * if that works.
3071          */
3072         do {
3073                 if (iaq->nrxq > 1) {
3074                         do {
3075                                 iaq->nrxq--;
3076                         } while (!powerof2(iaq->nrxq));
3077                 }
3078                 if (iaq->nofldrxq > 1)
3079                         iaq->nofldrxq >>= 1;
3080
3081                 old_nirq = iaq->nirq;
3082                 update_nirq(iaq, nports);
3083                 if (iaq->nirq <= navail &&
3084                     (itype != INTR_MSI || powerof2(iaq->nirq))) {
3085                         device_printf(sc->dev, "running with reduced number of "
3086                             "rx queues because of shortage of interrupts.  "
3087                             "nrxq=%u, nofldrxq=%u.  "
3088                             "itype %d, navail %u, nirq %d.\n", iaq->nrxq,
3089                             iaq->nofldrxq, itype, navail, iaq->nirq);
3090                         goto done;
3091                 }
3092         } while (old_nirq != iaq->nirq);
3093
3094         /* One interrupt for everything.  Ugh. */
3095         device_printf(sc->dev, "running with minimal number of queues.  "
3096             "itype %d, navail %u.\n", itype, navail);
3097         iaq->nirq = 1;
3098         MPASS(iaq->nrxq == 1);
3099         iaq->ntxq = 1;
3100         if (iaq->nofldrxq > 1)
3101                 iaq->nofldtxq = 1;
3102 done:
3103         MPASS(iaq->num_vis > 0);
3104         if (iaq->num_vis > 1) {
3105                 MPASS(iaq->nrxq_vi > 0);
3106                 MPASS(iaq->ntxq_vi > 0);
3107         }
3108         MPASS(iaq->nirq > 0);
3109         MPASS(iaq->nrxq > 0);
3110         MPASS(iaq->ntxq > 0);
3111         if (itype == INTR_MSI) {
3112                 MPASS(powerof2(iaq->nirq));
3113         }
3114 }
3115
3116 static int
3117 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
3118 {
3119         int rc, itype, navail, nalloc;
3120
3121         for (itype = INTR_MSIX; itype; itype >>= 1) {
3122
3123                 if ((itype & t4_intr_types) == 0)
3124                         continue;       /* not allowed */
3125
3126                 if (itype == INTR_MSIX)
3127                         navail = pci_msix_count(sc->dev);
3128                 else if (itype == INTR_MSI)
3129                         navail = pci_msi_count(sc->dev);
3130                 else
3131                         navail = 1;
3132 restart:
3133                 if (navail == 0)
3134                         continue;
3135
3136                 calculate_iaq(sc, iaq, itype, navail);
3137                 nalloc = iaq->nirq;
3138                 rc = 0;
3139                 if (itype == INTR_MSIX)
3140                         rc = pci_alloc_msix(sc->dev, &nalloc);
3141                 else if (itype == INTR_MSI)
3142                         rc = pci_alloc_msi(sc->dev, &nalloc);
3143
3144                 if (rc == 0 && nalloc > 0) {
3145                         if (nalloc == iaq->nirq)
3146                                 return (0);
3147
3148                         /*
3149                          * Didn't get the number requested.  Use whatever number
3150                          * the kernel is willing to allocate.
3151                          */
3152                         device_printf(sc->dev, "fewer vectors than requested, "
3153                             "type=%d, req=%d, rcvd=%d; will downshift req.\n",
3154                             itype, iaq->nirq, nalloc);
3155                         pci_release_msi(sc->dev);
3156                         navail = nalloc;
3157                         goto restart;
3158                 }
3159
3160                 device_printf(sc->dev,
3161                     "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
3162                     itype, rc, iaq->nirq, nalloc);
3163         }
3164
3165         device_printf(sc->dev,
3166             "failed to find a usable interrupt type.  "
3167             "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
3168             pci_msix_count(sc->dev), pci_msi_count(sc->dev));
3169
3170         return (ENXIO);
3171 }
3172
3173 #define FW_VERSION(chip) ( \
3174     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
3175     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
3176     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
3177     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
3178 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
3179
3180 struct fw_info {
3181         uint8_t chip;
3182         char *kld_name;
3183         char *fw_mod_name;
3184         struct fw_hdr fw_hdr;   /* XXX: waste of space, need a sparse struct */
3185 } fw_info[] = {
3186         {
3187                 .chip = CHELSIO_T4,
3188                 .kld_name = "t4fw_cfg",
3189                 .fw_mod_name = "t4fw",
3190                 .fw_hdr = {
3191                         .chip = FW_HDR_CHIP_T4,
3192                         .fw_ver = htobe32(FW_VERSION(T4)),
3193                         .intfver_nic = FW_INTFVER(T4, NIC),
3194                         .intfver_vnic = FW_INTFVER(T4, VNIC),
3195                         .intfver_ofld = FW_INTFVER(T4, OFLD),
3196                         .intfver_ri = FW_INTFVER(T4, RI),
3197                         .intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
3198                         .intfver_iscsi = FW_INTFVER(T4, ISCSI),
3199                         .intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
3200                         .intfver_fcoe = FW_INTFVER(T4, FCOE),
3201                 },
3202         }, {
3203                 .chip = CHELSIO_T5,
3204                 .kld_name = "t5fw_cfg",
3205                 .fw_mod_name = "t5fw",
3206                 .fw_hdr = {
3207                         .chip = FW_HDR_CHIP_T5,
3208                         .fw_ver = htobe32(FW_VERSION(T5)),
3209                         .intfver_nic = FW_INTFVER(T5, NIC),
3210                         .intfver_vnic = FW_INTFVER(T5, VNIC),
3211                         .intfver_ofld = FW_INTFVER(T5, OFLD),
3212                         .intfver_ri = FW_INTFVER(T5, RI),
3213                         .intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
3214                         .intfver_iscsi = FW_INTFVER(T5, ISCSI),
3215                         .intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
3216                         .intfver_fcoe = FW_INTFVER(T5, FCOE),
3217                 },
3218         }, {
3219                 .chip = CHELSIO_T6,
3220                 .kld_name = "t6fw_cfg",
3221                 .fw_mod_name = "t6fw",
3222                 .fw_hdr = {
3223                         .chip = FW_HDR_CHIP_T6,
3224                         .fw_ver = htobe32(FW_VERSION(T6)),
3225                         .intfver_nic = FW_INTFVER(T6, NIC),
3226                         .intfver_vnic = FW_INTFVER(T6, VNIC),
3227                         .intfver_ofld = FW_INTFVER(T6, OFLD),
3228                         .intfver_ri = FW_INTFVER(T6, RI),
3229                         .intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3230                         .intfver_iscsi = FW_INTFVER(T6, ISCSI),
3231                         .intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3232                         .intfver_fcoe = FW_INTFVER(T6, FCOE),
3233                 },
3234         }
3235 };
3236
3237 static struct fw_info *
3238 find_fw_info(int chip)
3239 {
3240         int i;
3241
3242         for (i = 0; i < nitems(fw_info); i++) {
3243                 if (fw_info[i].chip == chip)
3244                         return (&fw_info[i]);
3245         }
3246         return (NULL);
3247 }
3248
3249 /*
3250  * Is the given firmware API compatible with the one the driver was compiled
3251  * with?
3252  */
3253 static int
3254 fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
3255 {
3256
3257         /* short circuit if it's the exact same firmware version */
3258         if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
3259                 return (1);
3260
3261         /*
3262          * XXX: Is this too conservative?  Perhaps I should limit this to the
3263          * features that are supported in the driver.
3264          */
3265 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
3266         if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
3267             SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
3268             SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
3269                 return (1);
3270 #undef SAME_INTF
3271
3272         return (0);
3273 }
3274
3275 /*
3276  * The firmware in the KLD is usable, but should it be installed?  This routine
3277  * explains itself in detail if it indicates the KLD firmware should be
3278  * installed.
3279  */
3280 static int
3281 should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
3282 {
3283         const char *reason;
3284
3285         if (!card_fw_usable) {
3286                 reason = "incompatible or unusable";
3287                 goto install;
3288         }
3289
3290         if (k > c) {
3291                 reason = "older than the version bundled with this driver";
3292                 goto install;
3293         }
3294
3295         if (t4_fw_install == 2 && k != c) {
3296                 reason = "different than the version bundled with this driver";
3297                 goto install;
3298         }
3299
3300         return (0);
3301
3302 install:
3303         if (t4_fw_install == 0) {
3304                 device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3305                     "but the driver is prohibited from installing a different "
3306                     "firmware on the card.\n",
3307                     G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3308                     G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
3309
3310                 return (0);
3311         }
3312
3313         device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3314             "installing firmware %u.%u.%u.%u on card.\n",
3315             G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3316             G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
3317             G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3318             G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3319
3320         return (1);
3321 }
3322
3323 /*
3324  * Establish contact with the firmware and determine if we are the master driver
3325  * or not, and whether we are responsible for chip initialization.
3326  */
3327 static int
3328 prep_firmware(struct adapter *sc)
3329 {
3330         const struct firmware *fw = NULL, *default_cfg;
3331         int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
3332         enum dev_state state;
3333         struct fw_info *fw_info;
3334         struct fw_hdr *card_fw;         /* fw on the card */
3335         const struct fw_hdr *kld_fw;    /* fw in the KLD */
3336         const struct fw_hdr *drv_fw;    /* fw header the driver was compiled
3337                                            against */
3338
3339         /* This is the firmware whose headers the driver was compiled against */
3340         fw_info = find_fw_info(chip_id(sc));
3341         if (fw_info == NULL) {
3342                 device_printf(sc->dev,
3343                     "unable to look up firmware information for chip %d.\n",
3344                     chip_id(sc));
3345                 return (EINVAL);
3346         }
3347         drv_fw = &fw_info->fw_hdr;
3348
3349         /*
3350          * The firmware KLD contains many modules.  The KLD name is also the
3351          * name of the module that contains the default config file.
3352          */
3353         default_cfg = firmware_get(fw_info->kld_name);
3354
3355         /* This is the firmware in the KLD */
3356         fw = firmware_get(fw_info->fw_mod_name);
3357         if (fw != NULL) {
3358                 kld_fw = (const void *)fw->data;
3359                 kld_fw_usable = fw_compatible(drv_fw, kld_fw);
3360         } else {
3361                 kld_fw = NULL;
3362                 kld_fw_usable = 0;
3363         }
3364
3365         /* Read the header of the firmware on the card */
3366         card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
3367         rc = -t4_read_flash(sc, FLASH_FW_START,
3368             sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
3369         if (rc == 0) {
3370                 card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
3371                 if (card_fw->fw_ver == be32toh(0xffffffff)) {
3372                         uint32_t d = be32toh(kld_fw->fw_ver);
3373
3374                         if (!kld_fw_usable) {
3375                                 device_printf(sc->dev,
3376                                     "no firmware on the card and no usable "
3377                                     "firmware bundled with the driver.\n");
3378                                 rc = EIO;
3379                                 goto done;
3380                         } else if (t4_fw_install == 0) {
3381                                 device_printf(sc->dev,
3382                                     "no firmware on the card and the driver "
3383                                     "is prohibited from installing new "
3384                                     "firmware.\n");
3385                                 rc = EIO;
3386                                 goto done;
3387                         }
3388
3389                         device_printf(sc->dev, "no firmware on the card, "
3390                             "installing firmware %d.%d.%d.%d\n",
3391                             G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3392                             G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3393                         rc = t4_fw_forceinstall(sc, fw->data, fw->datasize);
3394                         if (rc < 0) {
3395                                 rc = -rc;
3396                                 device_printf(sc->dev,
3397                                     "firmware install failed: %d.\n", rc);
3398                                 goto done;
3399                         }
3400                         memcpy(card_fw, kld_fw, sizeof(*card_fw));
3401                         card_fw_usable = 1;
3402                         need_fw_reset = 0;
3403                 }
3404         } else {
3405                 device_printf(sc->dev,
3406                     "Unable to read card's firmware header: %d\n", rc);
3407                 card_fw_usable = 0;
3408         }
3409
3410         /* Contact firmware. */
3411         rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
3412         if (rc < 0 || state == DEV_STATE_ERR) {
3413                 rc = -rc;
3414                 device_printf(sc->dev,
3415                     "failed to connect to the firmware: %d, %d.\n", rc, state);
3416                 goto done;
3417         }
3418         pf = rc;
3419         if (pf == sc->mbox)
3420                 sc->flags |= MASTER_PF;
3421         else if (state == DEV_STATE_UNINIT) {
3422                 /*
3423                  * We didn't get to be the master so we definitely won't be
3424                  * configuring the chip.  It's a bug if someone else hasn't
3425                  * configured it already.
3426                  */
3427                 device_printf(sc->dev, "couldn't be master(%d), "
3428                     "device not already initialized either(%d).\n", rc, state);
3429                 rc = EPROTO;
3430                 goto done;
3431         }
3432
3433         if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
3434             (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
3435                 /*
3436                  * Common case: the firmware on the card is an exact match and
3437                  * the KLD is an exact match too, or the KLD is
3438                  * absent/incompatible.  Note that t4_fw_install = 2 is ignored
3439                  * here -- use cxgbetool loadfw if you want to reinstall the
3440                  * same firmware as the one on the card.
3441                  */
3442         } else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
3443             should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
3444             be32toh(card_fw->fw_ver))) {
3445
3446                 rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3447                 if (rc != 0) {
3448                         device_printf(sc->dev,
3449                             "failed to install firmware: %d\n", rc);
3450                         goto done;
3451                 }
3452
3453                 /* Installed successfully, update the cached header too. */
3454                 memcpy(card_fw, kld_fw, sizeof(*card_fw));
3455                 card_fw_usable = 1;
3456                 need_fw_reset = 0;      /* already reset as part of load_fw */
3457         }
3458
3459         if (!card_fw_usable) {
3460                 uint32_t d, c, k;
3461
3462                 d = ntohl(drv_fw->fw_ver);
3463                 c = ntohl(card_fw->fw_ver);
3464                 k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
3465
3466                 device_printf(sc->dev, "Cannot find a usable firmware: "
3467                     "fw_install %d, chip state %d, "
3468                     "driver compiled with %d.%d.%d.%d, "
3469                     "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
3470                     t4_fw_install, state,
3471                     G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3472                     G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
3473                     G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3474                     G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
3475                     G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3476                     G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3477                 rc = EINVAL;
3478                 goto done;
3479         }
3480
3481         /* Reset device */
3482         if (need_fw_reset &&
3483             (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
3484                 device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3485                 if (rc != ETIMEDOUT && rc != EIO)
3486                         t4_fw_bye(sc, sc->mbox);
3487                 goto done;
3488         }
3489         sc->flags |= FW_OK;
3490
3491         rc = get_params__pre_init(sc);
3492         if (rc != 0)
3493                 goto done; /* error message displayed already */
3494
3495         /* Partition adapter resources as specified in the config file. */
3496         if (state == DEV_STATE_UNINIT) {
3497
3498                 KASSERT(sc->flags & MASTER_PF,
3499                     ("%s: trying to change chip settings when not master.",
3500                     __func__));
3501
3502                 rc = partition_resources(sc, default_cfg, fw_info->kld_name);
3503                 if (rc != 0)
3504                         goto done;      /* error message displayed already */
3505
3506                 t4_tweak_chip_settings(sc);
3507
3508                 /* get basic stuff going */
3509                 rc = -t4_fw_initialize(sc, sc->mbox);
3510                 if (rc != 0) {
3511                         device_printf(sc->dev, "fw init failed: %d.\n", rc);
3512                         goto done;
3513                 }
3514         } else {
3515                 snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
3516                 sc->cfcsum = 0;
3517         }
3518
3519 done:
3520         free(card_fw, M_CXGBE);
3521         if (fw != NULL)
3522                 firmware_put(fw, FIRMWARE_UNLOAD);
3523         if (default_cfg != NULL)
3524                 firmware_put(default_cfg, FIRMWARE_UNLOAD);
3525
3526         return (rc);
3527 }
3528
3529 #define FW_PARAM_DEV(param) \
3530         (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3531          V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3532 #define FW_PARAM_PFVF(param) \
3533         (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3534          V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3535
3536 /*
3537  * Partition chip resources for use between various PFs, VFs, etc.
3538  */
3539 static int
3540 partition_resources(struct adapter *sc, const struct firmware *default_cfg,
3541     const char *name_prefix)
3542 {
3543         const struct firmware *cfg = NULL;
3544         int rc = 0;
3545         struct fw_caps_config_cmd caps;
3546         uint32_t mtype, moff, finicsum, cfcsum;
3547
3548         /*
3549          * Figure out what configuration file to use.  Pick the default config
3550          * file for the card if the user hasn't specified one explicitly.
3551          */
3552         snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
3553         if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3554                 /* Card specific overrides go here. */
3555                 if (pci_get_device(sc->dev) == 0x440a)
3556                         snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
3557                 if (is_fpga(sc))
3558                         snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
3559         } else if (strncmp(t4_cfg_file, BUILTIN_CF, sizeof(t4_cfg_file)) == 0)
3560                 goto use_built_in_config;       /* go straight to config. */
3561
3562         /*
3563          * We need to load another module if the profile is anything except
3564          * "default" or "flash".
3565          */
3566         if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
3567             strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3568                 char s[32];
3569
3570                 snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
3571                 cfg = firmware_get(s);
3572                 if (cfg == NULL) {
3573                         if (default_cfg != NULL) {
3574                                 device_printf(sc->dev,
3575                                     "unable to load module \"%s\" for "
3576                                     "configuration profile \"%s\", will use "
3577                                     "the default config file instead.\n",
3578                                     s, sc->cfg_file);
3579                                 snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3580                                     "%s", DEFAULT_CF);
3581                         } else {
3582                                 device_printf(sc->dev,
3583                                     "unable to load module \"%s\" for "
3584                                     "configuration profile \"%s\", will use "
3585                                     "the config file on the card's flash "
3586                                     "instead.\n", s, sc->cfg_file);
3587                                 snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3588                                     "%s", FLASH_CF);
3589                         }
3590                 }
3591         }
3592
3593         if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
3594             default_cfg == NULL) {
3595                 device_printf(sc->dev,
3596                     "default config file not available, will use the config "
3597                     "file on the card's flash instead.\n");
3598                 snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
3599         }
3600
3601         if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3602                 u_int cflen;
3603                 const uint32_t *cfdata;
3604                 uint32_t param, val, addr;
3605
3606                 KASSERT(cfg != NULL || default_cfg != NULL,
3607                     ("%s: no config to upload", __func__));
3608
3609                 /*
3610                  * Ask the firmware where it wants us to upload the config file.
3611                  */
3612                 param = FW_PARAM_DEV(CF);
3613                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3614                 if (rc != 0) {
3615                         /* No support for config file?  Shouldn't happen. */
3616                         device_printf(sc->dev,
3617                             "failed to query config file location: %d.\n", rc);
3618                         goto done;
3619                 }
3620                 mtype = G_FW_PARAMS_PARAM_Y(val);
3621                 moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3622
3623                 /*
3624                  * XXX: sheer laziness.  We deliberately added 4 bytes of
3625                  * useless stuffing/comments at the end of the config file so
3626                  * it's ok to simply throw away the last remaining bytes when
3627                  * the config file is not an exact multiple of 4.  This also
3628                  * helps with the validate_mt_off_len check.
3629                  */
3630                 if (cfg != NULL) {
3631                         cflen = cfg->datasize & ~3;
3632                         cfdata = cfg->data;
3633                 } else {
3634                         cflen = default_cfg->datasize & ~3;
3635                         cfdata = default_cfg->data;
3636                 }
3637
3638                 if (cflen > FLASH_CFG_MAX_SIZE) {
3639                         device_printf(sc->dev,
3640                             "config file too long (%d, max allowed is %d).  "
3641                             "Will try to use the config on the card, if any.\n",
3642                             cflen, FLASH_CFG_MAX_SIZE);
3643                         goto use_config_on_flash;
3644                 }
3645
3646                 rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3647                 if (rc != 0) {
3648                         device_printf(sc->dev,
3649                             "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
3650                             "Will try to use the config on the card, if any.\n",
3651                             __func__, mtype, moff, cflen, rc);
3652                         goto use_config_on_flash;
3653                 }
3654                 write_via_memwin(sc, 2, addr, cfdata, cflen);
3655         } else {
3656 use_config_on_flash:
3657                 mtype = FW_MEMTYPE_FLASH;
3658                 moff = t4_flash_cfg_addr(sc);
3659         }
3660
3661         bzero(&caps, sizeof(caps));
3662         caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3663             F_FW_CMD_REQUEST | F_FW_CMD_READ);
3664         caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3665             V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3666             V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
3667         rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3668         if (rc != 0) {
3669                 device_printf(sc->dev,
3670                     "failed to pre-process config file: %d "
3671                     "(mtype %d, moff 0x%x).  Will reset the firmware and retry "
3672                     "with the built-in configuration.\n", rc, mtype, moff);
3673
3674                 rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST);
3675                 if (rc != 0) {
3676                         device_printf(sc->dev,
3677                             "firmware reset failed: %d.\n", rc);
3678                         if (rc != ETIMEDOUT && rc != EIO) {
3679                                 t4_fw_bye(sc, sc->mbox);
3680                                 sc->flags &= ~FW_OK;
3681                         }
3682                         goto done;
3683                 }
3684                 snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", "built-in");
3685 use_built_in_config:
3686                 bzero(&caps, sizeof(caps));
3687                 caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3688                     F_FW_CMD_REQUEST | F_FW_CMD_READ);
3689                 caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3690                 rc = t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3691                 if (rc != 0) {
3692                         device_printf(sc->dev,
3693                             "built-in configuration failed: %d.\n", rc);
3694                         goto done;
3695                 }
3696         }
3697
3698         finicsum = be32toh(caps.finicsum);
3699         cfcsum = be32toh(caps.cfcsum);
3700         if (finicsum != cfcsum) {
3701                 device_printf(sc->dev,
3702                     "WARNING: config file checksum mismatch: %08x %08x\n",
3703                     finicsum, cfcsum);
3704         }
3705         sc->cfcsum = cfcsum;
3706
3707 #define LIMIT_CAPS(x) do { \
3708         caps.x &= htobe16(t4_##x##_allowed); \
3709 } while (0)
3710
3711         /*
3712          * Let the firmware know what features will (not) be used so it can tune
3713          * things accordingly.
3714          */
3715         LIMIT_CAPS(nbmcaps);
3716         LIMIT_CAPS(linkcaps);
3717         LIMIT_CAPS(switchcaps);
3718         LIMIT_CAPS(niccaps);
3719         LIMIT_CAPS(toecaps);
3720         LIMIT_CAPS(rdmacaps);
3721         LIMIT_CAPS(cryptocaps);
3722         LIMIT_CAPS(iscsicaps);
3723         LIMIT_CAPS(fcoecaps);
3724 #undef LIMIT_CAPS
3725
3726         if (caps.niccaps & htobe16(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
3727                 /*
3728                  * TOE and hashfilters are mutually exclusive.  It is a config
3729                  * file or firmware bug if both are reported as available.  Try
3730                  * to cope with the situation in non-debug builds by disabling
3731                  * TOE.
3732                  */
3733                 MPASS(caps.toecaps == 0);
3734
3735                 caps.toecaps = 0;
3736                 caps.rdmacaps = 0;
3737                 caps.iscsicaps = 0;
3738         }
3739
3740         caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3741             F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3742         caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3743         rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3744         if (rc != 0) {
3745                 device_printf(sc->dev,
3746                     "failed to process config file: %d.\n", rc);
3747         }
3748 done:
3749         if (cfg != NULL)
3750                 firmware_put(cfg, FIRMWARE_UNLOAD);
3751         return (rc);
3752 }
3753
3754 /*
3755  * Retrieve parameters that are needed (or nice to have) very early.
3756  */
3757 static int
3758 get_params__pre_init(struct adapter *sc)
3759 {
3760         int rc;
3761         uint32_t param[2], val[2];
3762
3763         t4_get_version_info(sc);
3764
3765         snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
3766             G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
3767             G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
3768             G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
3769             G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
3770
3771         snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
3772             G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
3773             G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
3774             G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
3775             G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
3776
3777         snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
3778             G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
3779             G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
3780             G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
3781             G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
3782
3783         snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
3784             G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
3785             G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
3786             G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
3787             G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
3788
3789         param[0] = FW_PARAM_DEV(PORTVEC);
3790         param[1] = FW_PARAM_DEV(CCLK);
3791         rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3792         if (rc != 0) {
3793                 device_printf(sc->dev,
3794                     "failed to query parameters (pre_init): %d.\n", rc);
3795                 return (rc);
3796         }
3797
3798         sc->params.portvec = val[0];
3799         sc->params.nports = bitcount32(val[0]);
3800         sc->params.vpd.cclk = val[1];
3801
3802         /* Read device log parameters. */
3803         rc = -t4_init_devlog_params(sc, 1);
3804         if (rc == 0)
3805                 fixup_devlog_params(sc);
3806         else {
3807                 device_printf(sc->dev,
3808                     "failed to get devlog parameters: %d.\n", rc);
3809                 rc = 0; /* devlog isn't critical for device operation */
3810         }
3811
3812         return (rc);
3813 }
3814
3815 /*
3816  * Retrieve various parameters that are of interest to the driver.  The device
3817  * has been initialized by the firmware at this point.
3818  */
3819 static int
3820 get_params__post_init(struct adapter *sc)
3821 {
3822         int rc;
3823         uint32_t param[7], val[7];
3824         struct fw_caps_config_cmd caps;
3825
3826         param[0] = FW_PARAM_PFVF(IQFLINT_START);
3827         param[1] = FW_PARAM_PFVF(EQ_START);
3828         param[2] = FW_PARAM_PFVF(FILTER_START);
3829         param[3] = FW_PARAM_PFVF(FILTER_END);
3830         param[4] = FW_PARAM_PFVF(L2T_START);
3831         param[5] = FW_PARAM_PFVF(L2T_END);
3832         param[6] = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
3833             V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
3834             V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
3835         rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 7, param, val);
3836         if (rc != 0) {
3837                 device_printf(sc->dev,
3838                     "failed to query parameters (post_init): %d.\n", rc);
3839                 return (rc);
3840         }
3841
3842         sc->sge.iq_start = val[0];
3843         sc->sge.eq_start = val[1];
3844         sc->tids.ftid_base = val[2];
3845         sc->tids.nftids = val[3] - val[2] + 1;
3846         sc->params.ftid_min = val[2];
3847         sc->params.ftid_max = val[3];
3848         sc->vres.l2t.start = val[4];
3849         sc->vres.l2t.size = val[5] - val[4] + 1;
3850         KASSERT(sc->vres.l2t.size <= L2T_SIZE,
3851             ("%s: L2 table size (%u) larger than expected (%u)",
3852             __func__, sc->vres.l2t.size, L2T_SIZE));
3853         sc->params.core_vdd = val[6];
3854
3855         /*
3856          * MPSBGMAP is queried separately because only recent firmwares support
3857          * it as a parameter and we don't want the compound query above to fail
3858          * on older firmwares.
3859          */
3860         param[0] = FW_PARAM_DEV(MPSBGMAP);
3861         val[0] = 0;
3862         rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
3863         if (rc == 0)
3864                 sc->params.mps_bg_map = val[0];
3865         else
3866                 sc->params.mps_bg_map = 0;
3867
3868         /*
3869          * Determine whether the firmware supports the filter2 work request.
3870          * This is queried separately for the same reason as MPSBGMAP above.
3871          */
3872         param[0] = FW_PARAM_DEV(FILTER2_WR);
3873         val[0] = 0;
3874         rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
3875         if (rc == 0)
3876                 sc->params.filter2_wr_support = val[0] != 0;
3877         else
3878                 sc->params.filter2_wr_support = 0;
3879
3880         /* get capabilites */
3881         bzero(&caps, sizeof(caps));
3882         caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3883             F_FW_CMD_REQUEST | F_FW_CMD_READ);
3884         caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3885         rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3886         if (rc != 0) {
3887                 device_printf(sc->dev,
3888                     "failed to get card capabilities: %d.\n", rc);
3889                 return (rc);
3890         }
3891
3892 #define READ_CAPS(x) do { \
3893         sc->x = htobe16(caps.x); \
3894 } while (0)
3895         READ_CAPS(nbmcaps);
3896         READ_CAPS(linkcaps);
3897         READ_CAPS(switchcaps);
3898         READ_CAPS(niccaps);
3899         READ_CAPS(toecaps);
3900         READ_CAPS(rdmacaps);
3901         READ_CAPS(cryptocaps);
3902         READ_CAPS(iscsicaps);
3903         READ_CAPS(fcoecaps);
3904
3905         if (sc->niccaps & FW_CAPS_CONFIG_NIC_HASHFILTER) {
3906                 MPASS(chip_id(sc) > CHELSIO_T4);
3907                 MPASS(sc->toecaps == 0);
3908                 sc->toecaps = 0;
3909
3910                 param[0] = FW_PARAM_DEV(NTID);
3911                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3912                 if (rc != 0) {
3913                         device_printf(sc->dev,
3914                             "failed to query HASHFILTER parameters: %d.\n", rc);
3915                         return (rc);
3916                 }
3917                 sc->tids.ntids = val[0];
3918                 sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3919                 sc->params.hash_filter = 1;
3920         }
3921         if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
3922                 param[0] = FW_PARAM_PFVF(ETHOFLD_START);
3923                 param[1] = FW_PARAM_PFVF(ETHOFLD_END);
3924                 param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3925                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
3926                 if (rc != 0) {
3927                         device_printf(sc->dev,
3928                             "failed to query NIC parameters: %d.\n", rc);
3929                         return (rc);
3930                 }
3931                 sc->tids.etid_base = val[0];
3932                 sc->params.etid_min = val[0];
3933                 sc->params.etid_max = val[1];
3934                 sc->tids.netids = val[1] - val[0] + 1;
3935                 sc->params.eo_wr_cred = val[2];
3936                 sc->params.ethoffload = 1;
3937         }
3938         if (sc->toecaps) {
3939                 /* query offload-related parameters */
3940                 param[0] = FW_PARAM_DEV(NTID);
3941                 param[1] = FW_PARAM_PFVF(SERVER_START);
3942                 param[2] = FW_PARAM_PFVF(SERVER_END);
3943                 param[3] = FW_PARAM_PFVF(TDDP_START);
3944                 param[4] = FW_PARAM_PFVF(TDDP_END);
3945                 param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3946                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3947                 if (rc != 0) {
3948                         device_printf(sc->dev,
3949                             "failed to query TOE parameters: %d.\n", rc);
3950                         return (rc);
3951                 }
3952                 sc->tids.ntids = val[0];
3953                 sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3954                 sc->tids.stid_base = val[1];
3955                 sc->tids.nstids = val[2] - val[1] + 1;
3956                 sc->vres.ddp.start = val[3];
3957                 sc->vres.ddp.size = val[4] - val[3] + 1;
3958                 sc->params.ofldq_wr_cred = val[5];
3959                 sc->params.offload = 1;
3960         } else {
3961                 /*
3962                  * The firmware attempts memfree TOE configuration for -SO cards
3963                  * and will report toecaps=0 if it runs out of resources (this
3964                  * depends on the config file).  It may not report 0 for other
3965                  * capabilities dependent on the TOE in this case.  Set them to
3966                  * 0 here so that the driver doesn't bother tracking resources
3967                  * that will never be used.
3968                  */
3969                 sc->iscsicaps = 0;
3970                 sc->rdmacaps = 0;
3971         }
3972         if (sc->rdmacaps) {
3973                 param[0] = FW_PARAM_PFVF(STAG_START);
3974                 param[1] = FW_PARAM_PFVF(STAG_END);
3975                 param[2] = FW_PARAM_PFVF(RQ_START);
3976                 param[3] = FW_PARAM_PFVF(RQ_END);
3977                 param[4] = FW_PARAM_PFVF(PBL_START);
3978                 param[5] = FW_PARAM_PFVF(PBL_END);
3979                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3980                 if (rc != 0) {
3981                         device_printf(sc->dev,
3982                             "failed to query RDMA parameters(1): %d.\n", rc);
3983                         return (rc);
3984                 }
3985                 sc->vres.stag.start = val[0];
3986                 sc->vres.stag.size = val[1] - val[0] + 1;
3987                 sc->vres.rq.start = val[2];
3988                 sc->vres.rq.size = val[3] - val[2] + 1;
3989                 sc->vres.pbl.start = val[4];
3990                 sc->vres.pbl.size = val[5] - val[4] + 1;
3991
3992                 param[0] = FW_PARAM_PFVF(SQRQ_START);
3993                 param[1] = FW_PARAM_PFVF(SQRQ_END);
3994                 param[2] = FW_PARAM_PFVF(CQ_START);
3995                 param[3] = FW_PARAM_PFVF(CQ_END);
3996                 param[4] = FW_PARAM_PFVF(OCQ_START);
3997                 param[5] = FW_PARAM_PFVF(OCQ_END);
3998                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3999                 if (rc != 0) {
4000                         device_printf(sc->dev,
4001                             "failed to query RDMA parameters(2): %d.\n", rc);
4002                         return (rc);
4003                 }
4004                 sc->vres.qp.start = val[0];
4005                 sc->vres.qp.size = val[1] - val[0] + 1;
4006                 sc->vres.cq.start = val[2];
4007                 sc->vres.cq.size = val[3] - val[2] + 1;
4008                 sc->vres.ocq.start = val[4];
4009                 sc->vres.ocq.size = val[5] - val[4] + 1;
4010
4011                 param[0] = FW_PARAM_PFVF(SRQ_START);
4012                 param[1] = FW_PARAM_PFVF(SRQ_END);
4013                 param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
4014                 param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
4015                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
4016                 if (rc != 0) {
4017                         device_printf(sc->dev,
4018                             "failed to query RDMA parameters(3): %d.\n", rc);
4019                         return (rc);
4020                 }
4021                 sc->vres.srq.start = val[0];
4022                 sc->vres.srq.size = val[1] - val[0] + 1;
4023                 sc->params.max_ordird_qp = val[2];
4024                 sc->params.max_ird_adapter = val[3];
4025         }
4026         if (sc->iscsicaps) {
4027                 param[0] = FW_PARAM_PFVF(ISCSI_START);
4028                 param[1] = FW_PARAM_PFVF(ISCSI_END);
4029                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4030                 if (rc != 0) {
4031                         device_printf(sc->dev,
4032                             "failed to query iSCSI parameters: %d.\n", rc);
4033                         return (rc);
4034                 }
4035                 sc->vres.iscsi.start = val[0];
4036                 sc->vres.iscsi.size = val[1] - val[0] + 1;
4037         }
4038         if (sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS) {
4039                 param[0] = FW_PARAM_PFVF(TLS_START);
4040                 param[1] = FW_PARAM_PFVF(TLS_END);
4041                 rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4042                 if (rc != 0) {
4043                         device_printf(sc->dev,
4044                             "failed to query TLS parameters: %d.\n", rc);
4045                         return (rc);
4046                 }
4047                 sc->vres.key.start = val[0];
4048                 sc->vres.key.size = val[1] - val[0] + 1;
4049         }
4050
4051         t4_init_sge_params(sc);
4052
4053         /*
4054          * We've got the params we wanted to query via the firmware.  Now grab
4055          * some others directly from the chip.
4056          */
4057         rc = t4_read_chip_settings(sc);
4058
4059         return (rc);
4060 }
4061
4062 static int
4063 set_params__post_init(struct adapter *sc)
4064 {
4065         uint32_t param, val;
4066 #ifdef TCP_OFFLOAD
4067         int i, v, shift;
4068 #endif
4069
4070         /* ask for encapsulated CPLs */
4071         param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
4072         val = 1;
4073         (void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4074
4075 #ifdef TCP_OFFLOAD
4076         /*
4077          * Override the TOE timers with user provided tunables.  This is not the
4078          * recommended way to change the timers (the firmware config file is) so
4079          * these tunables are not documented.
4080          *
4081          * All the timer tunables are in microseconds.
4082          */
4083         if (t4_toe_keepalive_idle != 0) {
4084                 v = us_to_tcp_ticks(sc, t4_toe_keepalive_idle);
4085                 v &= M_KEEPALIVEIDLE;
4086                 t4_set_reg_field(sc, A_TP_KEEP_IDLE,
4087                     V_KEEPALIVEIDLE(M_KEEPALIVEIDLE), V_KEEPALIVEIDLE(v));
4088         }
4089         if (t4_toe_keepalive_interval != 0) {
4090                 v = us_to_tcp_ticks(sc, t4_toe_keepalive_interval);
4091                 v &= M_KEEPALIVEINTVL;
4092                 t4_set_reg_field(sc, A_TP_KEEP_INTVL,
4093                     V_KEEPALIVEINTVL(M_KEEPALIVEINTVL), V_KEEPALIVEINTVL(v));
4094         }
4095         if (t4_toe_keepalive_count != 0) {
4096                 v = t4_toe_keepalive_count & M_KEEPALIVEMAXR2;
4097                 t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4098                     V_KEEPALIVEMAXR1(M_KEEPALIVEMAXR1) |
4099                     V_KEEPALIVEMAXR2(M_KEEPALIVEMAXR2),
4100                     V_KEEPALIVEMAXR1(1) | V_KEEPALIVEMAXR2(v));
4101         }
4102         if (t4_toe_rexmt_min != 0) {
4103                 v = us_to_tcp_ticks(sc, t4_toe_rexmt_min);
4104                 v &= M_RXTMIN;
4105                 t4_set_reg_field(sc, A_TP_RXT_MIN,
4106                     V_RXTMIN(M_RXTMIN), V_RXTMIN(v));
4107         }
4108         if (t4_toe_rexmt_max != 0) {
4109                 v = us_to_tcp_ticks(sc, t4_toe_rexmt_max);
4110                 v &= M_RXTMAX;
4111                 t4_set_reg_field(sc, A_TP_RXT_MAX,
4112                     V_RXTMAX(M_RXTMAX), V_RXTMAX(v));
4113         }
4114         if (t4_toe_rexmt_count != 0) {
4115                 v = t4_toe_rexmt_count & M_RXTSHIFTMAXR2;
4116                 t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4117                     V_RXTSHIFTMAXR1(M_RXTSHIFTMAXR1) |
4118                     V_RXTSHIFTMAXR2(M_RXTSHIFTMAXR2),
4119                     V_RXTSHIFTMAXR1(1) | V_RXTSHIFTMAXR2(v));
4120         }
4121         for (i = 0; i < nitems(t4_toe_rexmt_backoff); i++) {
4122                 if (t4_toe_rexmt_backoff[i] != -1) {
4123                         v = t4_toe_rexmt_backoff[i] & M_TIMERBACKOFFINDEX0;
4124                         shift = (i & 3) << 3;
4125                         t4_set_reg_field(sc, A_TP_TCP_BACKOFF_REG0 + (i & ~3),
4126                             M_TIMERBACKOFFINDEX0 << shift, v << shift);
4127                 }
4128         }
4129 #endif
4130         return (0);
4131 }
4132
4133 #undef FW_PARAM_PFVF
4134 #undef FW_PARAM_DEV
4135
4136 static void
4137 t4_set_desc(struct adapter *sc)
4138 {
4139         char buf[128];
4140         struct adapter_params *p = &sc->params;
4141
4142         snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
4143
4144         device_set_desc_copy(sc->dev, buf);
4145 }
4146
4147 static inline void
4148 ifmedia_add4(struct ifmedia *ifm, int m)
4149 {
4150
4151         ifmedia_add(ifm, m, 0, NULL);
4152         ifmedia_add(ifm, m | IFM_ETH_TXPAUSE, 0, NULL);
4153         ifmedia_add(ifm, m | IFM_ETH_RXPAUSE, 0, NULL);
4154         ifmedia_add(ifm, m | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE, 0, NULL);
4155 }
4156
4157 static void
4158 set_current_media(struct port_info *pi, struct ifmedia *ifm)
4159 {
4160         struct link_config *lc;
4161         int mword;
4162
4163         PORT_LOCK_ASSERT_OWNED(pi);
4164
4165         /* Leave current media alone if it's already set to IFM_NONE. */
4166         if (ifm->ifm_cur != NULL &&
4167             IFM_SUBTYPE(ifm->ifm_cur->ifm_media) == IFM_NONE)
4168                 return;
4169
4170         lc = &pi->link_cfg;
4171         if (lc->requested_aneg == AUTONEG_ENABLE &&
4172             lc->supported & FW_PORT_CAP_ANEG) {
4173                 ifmedia_set(ifm, IFM_ETHER | IFM_AUTO);
4174                 return;
4175         }
4176         mword = IFM_ETHER | IFM_FDX;
4177         if (lc->requested_fc & PAUSE_TX)
4178                 mword |= IFM_ETH_TXPAUSE;
4179         if (lc->requested_fc & PAUSE_RX)
4180                 mword |= IFM_ETH_RXPAUSE;
4181         mword |= port_mword(pi, speed_to_fwspeed(lc->requested_speed));
4182         ifmedia_set(ifm, mword);
4183 }
4184
4185 static void
4186 build_medialist(struct port_info *pi, struct ifmedia *ifm)
4187 {
4188         uint16_t ss, speed;
4189         int unknown, mword, bit;
4190         struct link_config *lc;
4191
4192         PORT_LOCK_ASSERT_OWNED(pi);
4193
4194         if (pi->flags & FIXED_IFMEDIA)
4195                 return;
4196
4197         /*
4198          * First setup all the requested_ fields so that they comply with what's
4199          * supported by the port + transceiver.  Note that this clobbers any
4200          * user preferences set via sysctl_pause_settings or sysctl_autoneg.
4201          */
4202         init_l1cfg(pi);
4203
4204         /*
4205          * Now (re)build the ifmedia list.
4206          */
4207         ifmedia_removeall(ifm);
4208         lc = &pi->link_cfg;
4209         ss = G_FW_PORT_CAP_SPEED(lc->supported); /* Supported Speeds */
4210         if (__predict_false(ss == 0)) { /* not supposed to happen. */
4211                 MPASS(ss != 0);
4212 no_media:
4213                 MPASS(LIST_EMPTY(&ifm->ifm_list));
4214                 ifmedia_add(ifm, IFM_ETHER | IFM_NONE, 0, NULL);
4215                 ifmedia_set(ifm, IFM_ETHER | IFM_NONE);
4216                 return;
4217         }
4218
4219         unknown = 0;
4220         for (bit = 0; bit < fls(ss); bit++) {
4221                 speed = 1 << bit;
4222                 MPASS(speed & M_FW_PORT_CAP_SPEED);
4223                 if (ss & speed) {
4224                         mword = port_mword(pi, speed);
4225                         if (mword == IFM_NONE) {
4226                                 goto no_media;
4227                         } else if (mword == IFM_UNKNOWN)
4228                                 unknown++;
4229                         else
4230                                 ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | mword);
4231                 }
4232         }
4233         if (unknown > 0) /* Add one unknown for all unknown media types. */
4234                 ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | IFM_UNKNOWN);
4235         if (lc->supported & FW_PORT_CAP_ANEG)
4236                 ifmedia_add(ifm, IFM_ETHER | IFM_AUTO, 0, NULL);
4237
4238         set_current_media(pi, ifm);
4239 }
4240
4241 /*
4242  * Update all the requested_* fields in the link config to something valid (and
4243  * reasonable).
4244  */
4245 static void
4246 init_l1cfg(struct port_info *pi)
4247 {
4248         struct link_config *lc = &pi->link_cfg;
4249
4250         PORT_LOCK_ASSERT_OWNED(pi);
4251
4252         /* Gbps -> Mbps */
4253         lc->requested_speed = port_top_speed(pi) * 1000;
4254
4255         if (t4_autoneg != 0 && lc->supported & FW_PORT_CAP_ANEG) {
4256                 lc->requested_aneg = AUTONEG_ENABLE;
4257         } else {
4258                 lc->requested_aneg = AUTONEG_DISABLE;
4259         }
4260
4261         lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX);
4262
4263         if (t4_fec != -1) {
4264                 if (t4_fec & FEC_RS && lc->supported & FW_PORT_CAP_FEC_RS) {
4265                         lc->requested_fec = FEC_RS;
4266                 } else if (t4_fec & FEC_BASER_RS &&
4267                     lc->supported & FW_PORT_CAP_FEC_BASER_RS) {
4268                         lc->requested_fec = FEC_BASER_RS;
4269                 } else {
4270                         lc->requested_fec = 0;
4271                 }
4272         } else {
4273                 /* Use the suggested value provided by the firmware in acaps */
4274                 if (lc->advertising & FW_PORT_CAP_FEC_RS &&
4275                     lc->supported & FW_PORT_CAP_FEC_RS) {
4276                         lc->requested_fec = FEC_RS;
4277                 } else if (lc->advertising & FW_PORT_CAP_FEC_BASER_RS &&
4278                     lc->supported & FW_PORT_CAP_FEC_BASER_RS) {
4279                         lc->requested_fec = FEC_BASER_RS;
4280                 } else {
4281                         lc->requested_fec = 0;
4282                 }
4283         }
4284 }
4285
4286 /*
4287  * Apply the settings in requested_* to the hardware.  The parameters are
4288  * expected to be sane.
4289  */
4290 static int
4291 apply_l1cfg(struct port_info *pi)
4292 {
4293         struct adapter *sc = pi->adapter;
4294         struct link_config *lc = &pi->link_cfg;
4295         int rc;
4296 #ifdef INVARIANTS
4297         uint16_t fwspeed;
4298
4299         ASSERT_SYNCHRONIZED_OP(sc);
4300         PORT_LOCK_ASSERT_OWNED(pi);
4301
4302         if (lc->requested_aneg == AUTONEG_ENABLE)
4303                 MPASS(lc->supported & FW_PORT_CAP_ANEG);
4304         if (lc->requested_fc & PAUSE_TX)
4305                 MPASS(lc->supported & FW_PORT_CAP_FC_TX);
4306         if (lc->requested_fc & PAUSE_RX)
4307                 MPASS(lc->supported & FW_PORT_CAP_FC_RX);
4308         if (lc->requested_fec == FEC_RS)
4309                 MPASS(lc->supported & FW_PORT_CAP_FEC_RS);
4310         if (lc->requested_fec == FEC_BASER_RS)
4311                 MPASS(lc->supported & FW_PORT_CAP_FEC_BASER_RS);
4312         fwspeed = speed_to_fwspeed(lc->requested_speed);
4313         MPASS(fwspeed != 0);
4314         MPASS(lc->supported & fwspeed);
4315 #endif
4316         rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
4317         if (rc != 0) {
4318                 device_printf(pi->dev, "l1cfg failed: %d\n", rc);
4319         } else {
4320                 lc->fc = lc->requested_fc;
4321                 lc->fec = lc->requested_fec;
4322         }
4323         return (rc);
4324 }
4325
4326 #define FW_MAC_EXACT_CHUNK      7
4327
4328 /*
4329  * Program the port's XGMAC based on parameters in ifnet.  The caller also
4330  * indicates which parameters should be programmed (the rest are left alone).
4331  */
4332 int
4333 update_mac_settings(struct ifnet *ifp, int flags)
4334 {
4335         int rc = 0;
4336         struct vi_info *vi = ifp->if_softc;
4337         struct port_info *pi = vi->pi;
4338         struct adapter *sc = pi->adapter;
4339         int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
4340
4341         ASSERT_SYNCHRONIZED_OP(sc);
4342         KASSERT(flags, ("%s: not told what to update.", __func__));
4343
4344         if (flags & XGMAC_MTU)
4345                 mtu = ifp->if_mtu;
4346
4347         if (flags & XGMAC_PROMISC)
4348                 promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
4349
4350         if (flags & XGMAC_ALLMULTI)
4351                 allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
4352
4353         if (flags & XGMAC_VLANEX)
4354                 vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
4355
4356         if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
4357                 rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
4358                     allmulti, 1, vlanex, false);
4359                 if (rc) {
4360                         if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
4361                             rc);
4362                         return (rc);
4363                 }
4364         }
4365
4366         if (flags & XGMAC_UCADDR) {
4367                 uint8_t ucaddr[ETHER_ADDR_LEN];
4368
4369                 bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
4370                 rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
4371                     ucaddr, true, true);
4372                 if (rc < 0) {
4373                         rc = -rc;
4374                         if_printf(ifp, "change_mac failed: %d\n", rc);
4375                         return (rc);
4376                 } else {
4377                         vi->xact_addr_filt = rc;
4378                         rc = 0;
4379                 }
4380         }
4381
4382         if (flags & XGMAC_MCADDRS) {
4383                 const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
4384                 int del = 1;
4385                 uint64_t hash = 0;
4386                 struct ifmultiaddr *ifma;
4387                 int i = 0, j;
4388
4389                 if_maddr_rlock(ifp);
4390                 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
4391                         if (ifma->ifma_addr->sa_family != AF_LINK)
4392                                 continue;
4393                         mcaddr[i] =
4394                             LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
4395                         MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
4396                         i++;
4397
4398                         if (i == FW_MAC_EXACT_CHUNK) {
4399                                 rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
4400                                     del, i, mcaddr, NULL, &hash, 0);
4401                                 if (rc < 0) {
4402                                         rc = -rc;
4403                                         for (j = 0; j < i; j++) {
4404                                                 if_printf(ifp,
4405                                                     "failed to add mc address"
4406                                                     " %02x:%02x:%02x:"
4407                                                     "%02x:%02x:%02x rc=%d\n",
4408                                                     mcaddr[j][0], mcaddr[j][1],
4409                                                     mcaddr[j][2], mcaddr[j][3],
4410                                                     mcaddr[j][4], mcaddr[j][5],
4411                                                     rc);
4412                                         }
4413                                         goto mcfail;
4414                                 }
4415                                 del = 0;
4416                                 i = 0;
4417                         }
4418                 }
4419                 if (i > 0) {
4420                         rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
4421                             mcaddr, NULL, &hash, 0);
4422                         if (rc < 0) {
4423                                 rc = -rc;
4424                                 for (j = 0; j < i; j++) {
4425                                         if_printf(ifp,
4426                                             "failed to add mc address"
4427                                             " %02x:%02x:%02x:"
4428                                             "%02x:%02x:%02x rc=%d\n",
4429                                             mcaddr[j][0], mcaddr[j][1],
4430                                             mcaddr[j][2], mcaddr[j][3],
4431                                             mcaddr[j][4], mcaddr[j][5],
4432                                             rc);
4433                                 }
4434                                 goto mcfail;
4435                         }
4436                 }
4437
4438                 rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
4439                 if (rc != 0)
4440                         if_printf(ifp, "failed to set mc address hash: %d", rc);
4441 mcfail:
4442                 if_maddr_runlock(ifp);
4443         }
4444
4445         return (rc);
4446 }
4447
4448 /*
4449  * {begin|end}_synchronized_op must be called from the same thread.
4450  */
4451 int
4452 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
4453     char *wmesg)
4454 {
4455         int rc, pri;
4456
4457 #ifdef WITNESS
4458         /* the caller thinks it's ok to sleep, but is it really? */
4459         if (flags & SLEEP_OK)
4460                 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
4461                     "begin_synchronized_op");
4462 #endif
4463
4464         if (INTR_OK)
4465                 pri = PCATCH;
4466         else
4467                 pri = 0;
4468
4469         ADAPTER_LOCK(sc);
4470         for (;;) {
4471
4472                 if (vi && IS_DOOMED(vi)) {
4473                         rc = ENXIO;
4474                         goto done;
4475                 }
4476
4477                 if (!IS_BUSY(sc)) {
4478                         rc = 0;
4479                         break;
4480                 }
4481
4482                 if (!(flags & SLEEP_OK)) {
4483                         rc = EBUSY;
4484                         goto done;
4485                 }
4486
4487                 if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
4488                         rc = EINTR;
4489                         goto done;
4490                 }
4491         }
4492
4493         KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
4494         SET_BUSY(sc);
4495 #ifdef INVARIANTS
4496         sc->last_op = wmesg;
4497         sc->last_op_thr = curthread;
4498         sc->last_op_flags = flags;
4499 #endif
4500
4501 done:
4502         if (!(flags & HOLD_LOCK) || rc)
4503                 ADAPTER_UNLOCK(sc);
4504
4505         return (rc);
4506 }
4507
4508 /*
4509  * Tell if_ioctl and if_init that the VI is going away.  This is
4510  * special variant of begin_synchronized_op and must be paired with a
4511  * call to end_synchronized_op.
4512  */
4513 void
4514 doom_vi(struct adapter *sc, struct vi_info *vi)
4515 {
4516
4517         ADAPTER_LOCK(sc);
4518         SET_DOOMED(vi);
4519         wakeup(&sc->flags);
4520         while (IS_BUSY(sc))
4521                 mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
4522         SET_BUSY(sc);
4523 #ifdef INVARIANTS
4524         sc->last_op = "t4detach";
4525         sc->last_op_thr = curthread;
4526         sc->last_op_flags = 0;
4527 #endif
4528         ADAPTER_UNLOCK(sc);
4529 }
4530
4531 /*
4532  * {begin|end}_synchronized_op must be called from the same thread.
4533  */
4534 void
4535 end_synchronized_op(struct adapter *sc, int flags)
4536 {
4537
4538         if (flags & LOCK_HELD)
4539                 ADAPTER_LOCK_ASSERT_OWNED(sc);
4540         else
4541                 ADAPTER_LOCK(sc);
4542
4543         KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
4544         CLR_BUSY(sc);
4545         wakeup(&sc->flags);
4546         ADAPTER_UNLOCK(sc);
4547 }
4548
4549 static int
4550 cxgbe_init_synchronized(struct vi_info *vi)
4551 {
4552         struct port_info *pi = vi->pi;
4553         struct adapter *sc = pi->adapter;
4554         struct ifnet *ifp = vi->ifp;
4555         int rc = 0, i;
4556         struct sge_txq *txq;
4557
4558         ASSERT_SYNCHRONIZED_OP(sc);
4559
4560         if (ifp->if_drv_flags & IFF_DRV_RUNNING)
4561                 return (0);     /* already running */
4562
4563         if (!(sc->flags & FULL_INIT_DONE) &&
4564             ((rc = adapter_full_init(sc)) != 0))
4565                 return (rc);    /* error message displayed already */
4566
4567         if (!(vi->flags & VI_INIT_DONE) &&
4568             ((rc = vi_full_init(vi)) != 0))
4569                 return (rc); /* error message displayed already */
4570
4571         rc = update_mac_settings(ifp, XGMAC_ALL);
4572         if (rc)
4573                 goto done;      /* error message displayed already */
4574
4575         rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
4576         if (rc != 0) {
4577                 if_printf(ifp, "enable_vi failed: %d\n", rc);
4578                 goto done;
4579         }
4580
4581         /*
4582          * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
4583          * if this changes.
4584          */
4585
4586         for_each_txq(vi, i, txq) {
4587                 TXQ_LOCK(txq);
4588                 txq->eq.flags |= EQ_ENABLED;
4589                 TXQ_UNLOCK(txq);
4590         }
4591
4592         /*
4593          * The first iq of the first port to come up is used for tracing.
4594          */
4595         if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
4596                 sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
4597                 t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
4598                     A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
4599                     V_QUEUENUMBER(sc->traceq));
4600                 pi->flags |= HAS_TRACEQ;
4601         }
4602
4603         /* all ok */
4604         PORT_LOCK(pi);
4605         if (pi->up_vis++ == 0) {
4606                 t4_update_port_info(pi);
4607                 build_medialist(pi, &pi->media);
4608                 apply_l1cfg(pi);
4609         }
4610         ifp->if_drv_flags |= IFF_DRV_RUNNING;
4611
4612         if (pi->nvi > 1 || sc->flags & IS_VF)
4613                 callout_reset(&vi->tick, hz, vi_tick, vi);
4614         else
4615                 callout_reset(&pi->tick, hz, cxgbe_tick, pi);
4616         PORT_UNLOCK(pi);
4617 done:
4618         if (rc != 0)
4619                 cxgbe_uninit_synchronized(vi);
4620
4621         return (rc);
4622 }
4623
4624 /*
4625  * Idempotent.
4626  */
4627 static int
4628 cxgbe_uninit_synchronized(struct vi_info *vi)
4629 {
4630         struct port_info *pi = vi->pi;
4631         struct adapter *sc = pi->adapter;
4632         struct ifnet *ifp = vi->ifp;
4633         int rc, i;
4634         struct sge_txq *txq;
4635
4636         ASSERT_SYNCHRONIZED_OP(sc);
4637
4638         if (!(vi->flags & VI_INIT_DONE)) {
4639                 if (__predict_false(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4640                         KASSERT(0, ("uninited VI is running"));
4641                         if_printf(ifp, "uninited VI with running ifnet.  "
4642                             "vi->flags 0x%016lx, if_flags 0x%08x, "
4643                             "if_drv_flags 0x%08x\n", vi->flags, ifp->if_flags,
4644                             ifp->if_drv_flags);
4645                 }
4646                 return (0);
4647         }
4648
4649         /*
4650          * Disable the VI so that all its data in either direction is discarded
4651          * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
4652          * tick) intact as the TP can deliver negative advice or data that it's
4653          * holding in its RAM (for an offloaded connection) even after the VI is
4654          * disabled.
4655          */
4656         rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
4657         if (rc) {
4658                 if_printf(ifp, "disable_vi failed: %d\n", rc);
4659                 return (rc);
4660         }
4661
4662         for_each_txq(vi, i, txq) {
4663                 TXQ_LOCK(txq);
4664                 txq->eq.flags &= ~EQ_ENABLED;
4665                 TXQ_UNLOCK(txq);
4666         }
4667
4668         PORT_LOCK(pi);
4669         if (pi->nvi > 1 || sc->flags & IS_VF)
4670                 callout_stop(&vi->tick);
4671         else
4672                 callout_stop(&pi->tick);
4673         if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4674                 PORT_UNLOCK(pi);
4675                 return (0);
4676         }
4677         ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
4678         pi->up_vis--;
4679         if (pi->up_vis > 0) {
4680                 PORT_UNLOCK(pi);
4681                 return (0);
4682         }
4683
4684         pi->link_cfg.link_ok = 0;
4685         pi->link_cfg.speed = 0;
4686         pi->link_cfg.link_down_rc = 255;
4687         t4_os_link_changed(pi);
4688         pi->old_link_cfg = pi->link_cfg;
4689         PORT_UNLOCK(pi);
4690
4691         return (0);
4692 }
4693
4694 /*
4695  * It is ok for this function to fail midway and return right away.  t4_detach
4696  * will walk the entire sc->irq list and clean up whatever is valid.
4697  */
4698 int
4699 t4_setup_intr_handlers(struct adapter *sc)
4700 {
4701         int rc, rid, p, q, v;
4702         char s[8];
4703         struct irq *irq;
4704         struct port_info *pi;
4705         struct vi_info *vi;
4706         struct sge *sge = &sc->sge;
4707         struct sge_rxq *rxq;
4708 #ifdef TCP_OFFLOAD
4709         struct sge_ofld_rxq *ofld_rxq;
4710 #endif
4711 #ifdef DEV_NETMAP
4712         struct sge_nm_rxq *nm_rxq;
4713 #endif
4714 #ifdef RSS
4715         int nbuckets = rss_getnumbuckets();
4716 #endif
4717
4718         /*
4719          * Setup interrupts.
4720          */
4721         irq = &sc->irq[0];
4722         rid = sc->intr_type == INTR_INTX ? 0 : 1;
4723         if (forwarding_intr_to_fwq(sc))
4724                 return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
4725
4726         /* Multiple interrupts. */
4727         if (sc->flags & IS_VF)
4728                 KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
4729                     ("%s: too few intr.", __func__));
4730         else
4731                 KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
4732                     ("%s: too few intr.", __func__));
4733
4734         /* The first one is always error intr on PFs */
4735         if (!(sc->flags & IS_VF)) {
4736                 rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
4737                 if (rc != 0)
4738                         return (rc);
4739                 irq++;
4740                 rid++;
4741         }
4742
4743         /* The second one is always the firmware event queue (first on VFs) */
4744         rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
4745         if (rc != 0)
4746                 return (rc);
4747         irq++;
4748         rid++;
4749
4750         for_each_port(sc, p) {
4751                 pi = sc->port[p];
4752                 for_each_vi(pi, v, vi) {
4753                         vi->first_intr = rid - 1;
4754
4755                         if (vi->nnmrxq > 0) {
4756                                 int n = max(vi->nrxq, vi->nnmrxq);
4757
4758                                 rxq = &sge->rxq[vi->first_rxq];
4759 #ifdef DEV_NETMAP
4760                                 nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
4761 #endif
4762                                 for (q = 0; q < n; q++) {
4763                                         snprintf(s, sizeof(s), "%x%c%x", p,
4764                                             'a' + v, q);
4765                                         if (q < vi->nrxq)
4766                                                 irq->rxq = rxq++;
4767 #ifdef DEV_NETMAP
4768                                         if (q < vi->nnmrxq)
4769                                                 irq->nm_rxq = nm_rxq++;
4770 #endif
4771                                         rc = t4_alloc_irq(sc, irq, rid,
4772                                             t4_vi_intr, irq, s);
4773                                         if (rc != 0)
4774                                                 return (rc);
4775 #ifdef RSS
4776                                         if (q < vi->nrxq) {
4777                                                 bus_bind_intr(sc->dev, irq->res,
4778                                                     rss_getcpu(q % nbuckets));
4779                                         }
4780 #endif
4781                                         irq++;
4782                                         rid++;
4783                                         vi->nintr++;
4784                                 }
4785                         } else {
4786                                 for_each_rxq(vi, q, rxq) {
4787                                         snprintf(s, sizeof(s), "%x%c%x", p,
4788                                             'a' + v, q);
4789                                         rc = t4_alloc_irq(sc, irq, rid,
4790                                             t4_intr, rxq, s);
4791                                         if (rc != 0)
4792                                                 return (rc);
4793 #ifdef RSS
4794                                         bus_bind_intr(sc->dev, irq->res,
4795                                             rss_getcpu(q % nbuckets));
4796 #endif
4797                                         irq++;
4798                                         rid++;
4799                                         vi->nintr++;
4800                                 }
4801                         }
4802 #ifdef TCP_OFFLOAD
4803                         for_each_ofld_rxq(vi, q, ofld_rxq) {
4804                                 snprintf(s, sizeof(s), "%x%c%x", p, 'A' + v, q);
4805                                 rc = t4_alloc_irq(sc, irq, rid, t4_intr,
4806                                     ofld_rxq, s);
4807                                 if (rc != 0)
4808                                         return (rc);
4809                                 irq++;
4810                                 rid++;
4811                                 vi->nintr++;
4812                         }
4813 #endif
4814                 }
4815         }
4816         MPASS(irq == &sc->irq[sc->intr_count]);
4817
4818         return (0);
4819 }
4820
4821 int
4822 adapter_full_init(struct adapter *sc)
4823 {
4824         int rc, i;
4825 #ifdef RSS
4826         uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4827         uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4828 #endif
4829
4830         ASSERT_SYNCHRONIZED_OP(sc);
4831         ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4832         KASSERT((sc->flags & FULL_INIT_DONE) == 0,
4833             ("%s: FULL_INIT_DONE already", __func__));
4834
4835         /*
4836          * queues that belong to the adapter (not any particular port).
4837          */
4838         rc = t4_setup_adapter_queues(sc);
4839         if (rc != 0)
4840                 goto done;
4841
4842         for (i = 0; i < nitems(sc->tq); i++) {
4843                 sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
4844                     taskqueue_thread_enqueue, &sc->tq[i]);
4845                 if (sc->tq[i] == NULL) {
4846                         device_printf(sc->dev,
4847                             "failed to allocate task queue %d\n", i);
4848                         rc = ENOMEM;
4849                         goto done;
4850                 }
4851                 taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
4852                     device_get_nameunit(sc->dev), i);
4853         }
4854 #ifdef RSS
4855         MPASS(RSS_KEYSIZE == 40);
4856         rss_getkey((void *)&raw_rss_key[0]);
4857         for (i = 0; i < nitems(rss_key); i++) {
4858                 rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
4859         }
4860         t4_write_rss_key(sc, &rss_key[0], -1, 1);
4861 #endif
4862
4863         if (!(sc->flags & IS_VF))
4864                 t4_intr_enable(sc);
4865         sc->flags |= FULL_INIT_DONE;
4866 done:
4867         if (rc != 0)
4868                 adapter_full_uninit(sc);
4869
4870         return (rc);
4871 }
4872
4873 int
4874 adapter_full_uninit(struct adapter *sc)
4875 {
4876         int i;
4877
4878         ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4879
4880         t4_teardown_adapter_queues(sc);
4881
4882         for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
4883                 taskqueue_free(sc->tq[i]);
4884                 sc->tq[i] = NULL;
4885         }
4886
4887         sc->flags &= ~FULL_INIT_DONE;
4888
4889         return (0);
4890 }
4891
4892 #ifdef RSS
4893 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
4894     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
4895     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
4896     RSS_HASHTYPE_RSS_UDP_IPV6)
4897
4898 /* Translates kernel hash types to hardware. */
4899 static int
4900 hashconfig_to_hashen(int hashconfig)
4901 {
4902         int hashen = 0;
4903
4904         if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
4905                 hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
4906         if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
4907                 hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
4908         if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
4909                 hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4910                     F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4911         }
4912         if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
4913                 hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4914                     F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4915         }
4916         if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
4917                 hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4918         if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
4919                 hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4920
4921         return (hashen);
4922 }
4923
4924 /* Translates hardware hash types to kernel. */
4925 static int
4926 hashen_to_hashconfig(int hashen)
4927 {
4928         int hashconfig = 0;
4929
4930         if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
4931                 /*
4932                  * If UDP hashing was enabled it must have been enabled for
4933                  * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
4934                  * enabling any 4-tuple hash is nonsense configuration.
4935                  */
4936                 MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4937                     F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
4938
4939                 if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4940                         hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
4941                 if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4942                         hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
4943         }
4944         if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4945                 hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
4946         if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4947                 hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
4948         if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
4949                 hashconfig |= RSS_HASHTYPE_RSS_IPV4;
4950         if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
4951                 hashconfig |= RSS_HASHTYPE_RSS_IPV6;
4952
4953         return (hashconfig);
4954 }
4955 #endif
4956
4957 int
4958 vi_full_init(struct vi_info *vi)
4959 {
4960         struct adapter *sc = vi->pi->adapter;
4961         struct ifnet *ifp = vi->ifp;
4962         uint16_t *rss;
4963         struct sge_rxq *rxq;
4964         int rc, i, j, hashen;
4965 #ifdef RSS
4966         int nbuckets = rss_getnumbuckets();
4967         int hashconfig = rss_gethashconfig();
4968         int extra;
4969 #endif
4970
4971         ASSERT_SYNCHRONIZED_OP(sc);
4972         KASSERT((vi->flags & VI_INIT_DONE) == 0,
4973             ("%s: VI_INIT_DONE already", __func__));
4974
4975         sysctl_ctx_init(&vi->ctx);
4976         vi->flags |= VI_SYSCTL_CTX;
4977
4978         /*
4979          * Allocate tx/rx/fl queues for this VI.
4980          */
4981         rc = t4_setup_vi_queues(vi);
4982         if (rc != 0)
4983                 goto done;      /* error message displayed already */
4984
4985         /*
4986          * Setup RSS for this VI.  Save a copy of the RSS table for later use.
4987          */
4988         if (vi->nrxq > vi->rss_size) {
4989                 if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
4990                     "some queues will never receive traffic.\n", vi->nrxq,
4991                     vi->rss_size);
4992         } else if (vi->rss_size % vi->nrxq) {
4993                 if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
4994                     "expect uneven traffic distribution.\n", vi->nrxq,
4995                     vi->rss_size);
4996         }
4997 #ifdef RSS
4998         if (vi->nrxq != nbuckets) {
4999                 if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
5000                     "performance will be impacted.\n", vi->nrxq, nbuckets);
5001         }
5002 #endif
5003         rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
5004         for (i = 0; i < vi->rss_size;) {
5005 #ifdef RSS
5006                 j = rss_get_indirection_to_bucket(i);
5007                 j %= vi->nrxq;
5008                 rxq = &sc->sge.rxq[vi->first_rxq + j];
5009                 rss[i++] = rxq->iq.abs_id;
5010 #else
5011                 for_each_rxq(vi, j, rxq) {
5012                         rss[i++] = rxq->iq.abs_id;
5013                         if (i == vi->rss_size)
5014                                 break;
5015                 }
5016 #endif
5017         }
5018
5019         rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
5020             vi->rss_size);
5021         if (rc != 0) {
5022                 if_printf(ifp, "rss_config failed: %d\n", rc);
5023                 goto done;
5024         }
5025
5026 #ifdef RSS
5027         hashen = hashconfig_to_hashen(hashconfig);
5028
5029         /*
5030          * We may have had to enable some hashes even though the global config
5031          * wants them disabled.  This is a potential problem that must be
5032          * reported to the user.
5033          */
5034         extra = hashen_to_hashconfig(hashen) ^ hashconfig;
5035
5036         /*
5037          * If we consider only the supported hash types, then the enabled hashes
5038          * are a superset of the requested hashes.  In other words, there cannot
5039          * be any supported hash that was requested but not enabled, but there
5040          * can be hashes that were not requested but had to be enabled.
5041          */
5042         extra &= SUPPORTED_RSS_HASHTYPES;
5043         MPASS((extra & hashconfig) == 0);
5044
5045         if (extra) {
5046                 if_printf(ifp,
5047                     "global RSS config (0x%x) cannot be accommodated.\n",
5048                     hashconfig);
5049         }
5050         if (extra & RSS_HASHTYPE_RSS_IPV4)
5051                 if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
5052         if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
5053                 if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
5054         if (extra & RSS_HASHTYPE_RSS_IPV6)
5055                 if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
5056         if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
5057                 if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
5058         if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
5059                 if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
5060         if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
5061                 if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
5062 #else
5063         hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
5064             F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
5065             F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5066             F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
5067 #endif
5068         rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, hashen, rss[0], 0, 0);
5069         if (rc != 0) {
5070                 if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
5071                 goto done;
5072         }
5073
5074         vi->rss = rss;
5075         vi->flags |= VI_INIT_DONE;
5076 done:
5077         if (rc != 0)
5078                 vi_full_uninit(vi);
5079
5080         return (rc);
5081 }
5082
5083 /*
5084  * Idempotent.
5085  */
5086 int
5087 vi_full_uninit(struct vi_info *vi)
5088 {
5089         struct port_info *pi = vi->pi;
5090         struct adapter *sc = pi->adapter;
5091         int i;
5092         struct sge_rxq *rxq;
5093         struct sge_txq *txq;
5094 #ifdef TCP_OFFLOAD
5095         struct sge_ofld_rxq *ofld_rxq;
5096         struct sge_wrq *ofld_txq;
5097 #endif
5098
5099         if (vi->flags & VI_INIT_DONE) {
5100
5101                 /* Need to quiesce queues.  */
5102
5103                 /* XXX: Only for the first VI? */
5104                 if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
5105                         quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
5106
5107                 for_each_txq(vi, i, txq) {
5108                         quiesce_txq(sc, txq);
5109                 }
5110
5111 #ifdef TCP_OFFLOAD
5112                 for_each_ofld_txq(vi, i, ofld_txq) {
5113                         quiesce_wrq(sc, ofld_txq);
5114                 }
5115 #endif
5116
5117                 for_each_rxq(vi, i, rxq) {
5118                         quiesce_iq(sc, &rxq->iq);
5119                         quiesce_fl(sc, &rxq->fl);
5120                 }
5121
5122 #ifdef TCP_OFFLOAD
5123                 for_each_ofld_rxq(vi, i, ofld_rxq) {
5124                         quiesce_iq(sc, &ofld_rxq->iq);
5125                         quiesce_fl(sc, &ofld_rxq->fl);
5126                 }
5127 #endif
5128                 free(vi->rss, M_CXGBE);
5129                 free(vi->nm_rss, M_CXGBE);
5130         }
5131
5132         t4_teardown_vi_queues(vi);
5133         vi->flags &= ~VI_INIT_DONE;
5134
5135         return (0);
5136 }
5137
5138 static void
5139 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
5140 {
5141         struct sge_eq *eq = &txq->eq;
5142         struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
5143
5144         (void) sc;      /* unused */
5145
5146 #ifdef INVARIANTS
5147         TXQ_LOCK(txq);
5148         MPASS((eq->flags & EQ_ENABLED) == 0);
5149         TXQ_UNLOCK(txq);
5150 #endif
5151
5152         /* Wait for the mp_ring to empty. */
5153         while (!mp_ring_is_idle(txq->r)) {
5154                 mp_ring_check_drainage(txq->r, 0);
5155                 pause("rquiesce", 1);
5156         }
5157
5158         /* Then wait for the hardware to finish. */
5159         while (spg->cidx != htobe16(eq->pidx))
5160                 pause("equiesce", 1);
5161
5162         /* Finally, wait for the driver to reclaim all descriptors. */
5163         while (eq->cidx != eq->pidx)
5164                 pause("dquiesce", 1);
5165 }
5166
5167 static void
5168 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
5169 {
5170
5171         /* XXXTX */
5172 }
5173
5174 static void
5175 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
5176 {
5177         (void) sc;      /* unused */
5178
5179         /* Synchronize with the interrupt handler */
5180         while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
5181                 pause("iqfree", 1);
5182 }
5183
5184 static void
5185 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
5186 {
5187         mtx_lock(&sc->sfl_lock);
5188         FL_LOCK(fl);
5189         fl->flags |= FL_DOOMED;
5190         FL_UNLOCK(fl);
5191         callout_stop(&sc->sfl_callout);
5192         mtx_unlock(&sc->sfl_lock);
5193
5194         KASSERT((fl->flags & FL_STARVING) == 0,
5195             ("%s: still starving", __func__));
5196 }
5197
5198 static int
5199 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
5200     driver_intr_t *handler, void *arg, char *name)
5201 {
5202         int rc;
5203
5204         irq->rid = rid;
5205         irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
5206             RF_SHAREABLE | RF_ACTIVE);
5207         if (irq->res == NULL) {
5208                 device_printf(sc->dev,
5209                     "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
5210                 return (ENOMEM);
5211         }
5212
5213         rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
5214             NULL, handler, arg, &irq->tag);
5215         if (rc != 0) {
5216                 device_printf(sc->dev,
5217                     "failed to setup interrupt for rid %d, name %s: %d\n",
5218                     rid, name, rc);
5219         } else if (name)
5220                 bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
5221
5222         return (rc);
5223 }
5224
5225 static int
5226 t4_free_irq(struct adapter *sc, struct irq *irq)
5227 {
5228         if (irq->tag)
5229                 bus_teardown_intr(sc->dev, irq->res, irq->tag);
5230         if (irq->res)
5231                 bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
5232
5233         bzero(irq, sizeof(*irq));
5234
5235         return (0);
5236 }
5237
5238 static void
5239 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
5240 {
5241
5242         regs->version = chip_id(sc) | chip_rev(sc) << 10;
5243         t4_get_regs(sc, buf, regs->len);
5244 }
5245
5246 #define A_PL_INDIR_CMD  0x1f8
5247
5248 #define S_PL_AUTOINC    31
5249 #define M_PL_AUTOINC    0x1U
5250 #define V_PL_AUTOINC(x) ((x) << S_PL_AUTOINC)
5251 #define G_PL_AUTOINC(x) (((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
5252
5253 #define S_PL_VFID       20
5254 #define M_PL_VFID       0xffU
5255 #define V_PL_VFID(x)    ((x) << S_PL_VFID)
5256 #define G_PL_VFID(x)    (((x) >> S_PL_VFID) & M_PL_VFID)
5257
5258 #define S_PL_ADDR       0
5259 #define M_PL_ADDR       0xfffffU
5260 #define V_PL_ADDR(x)    ((x) << S_PL_ADDR)
5261 #define G_PL_ADDR(x)    (((x) >> S_PL_ADDR) & M_PL_ADDR)
5262
5263 #define A_PL_INDIR_DATA 0x1fc
5264
5265 static uint64_t
5266 read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
5267 {
5268         u32 stats[2];
5269
5270         mtx_assert(&sc->reg_lock, MA_OWNED);
5271         if (sc->flags & IS_VF) {
5272                 stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
5273                 stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
5274         } else {
5275                 t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5276                     V_PL_VFID(G_FW_VIID_VIN(viid)) |
5277                     V_PL_ADDR(VF_MPS_REG(reg)));
5278                 stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
5279                 stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
5280         }
5281         return (((uint64_t)stats[1]) << 32 | stats[0]);
5282 }
5283
5284 static void
5285 t4_get_vi_stats(struct adapter *sc, unsigned int viid,
5286     struct fw_vi_stats_vf *stats)
5287 {
5288
5289 #define GET_STAT(name) \
5290         read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
5291
5292         stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
5293         stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
5294         stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
5295         stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
5296         stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
5297         stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
5298         stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
5299         stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
5300         stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
5301         stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
5302         stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
5303         stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
5304         stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
5305         stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
5306         stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
5307         stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
5308
5309 #undef GET_STAT
5310 }
5311
5312 static void
5313 t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
5314 {
5315         int reg;
5316
5317         t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5318             V_PL_VFID(G_FW_VIID_VIN(viid)) |
5319             V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
5320         for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
5321              reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
5322                 t4_write_reg(sc, A_PL_INDIR_DATA, 0);
5323 }
5324
5325 static void
5326 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
5327 {
5328         struct timeval tv;
5329         const struct timeval interval = {0, 250000};    /* 250ms */
5330
5331         if (!(vi->flags & VI_INIT_DONE))
5332                 return;
5333
5334         getmicrotime(&tv);
5335         timevalsub(&tv, &interval);
5336         if (timevalcmp(&tv, &vi->last_refreshed, <))
5337                 return;
5338
5339         mtx_lock(&sc->reg_lock);
5340         t4_get_vi_stats(sc, vi->viid, &vi->stats);
5341         getmicrotime(&vi->last_refreshed);
5342         mtx_unlock(&sc->reg_lock);
5343 }
5344
5345 static void
5346 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
5347 {
5348         u_int i, v, tnl_cong_drops, bg_map;
5349         struct timeval tv;
5350         const struct timeval interval = {0, 250000};    /* 250ms */
5351
5352         getmicrotime(&tv);
5353         timevalsub(&tv, &interval);
5354         if (timevalcmp(&tv, &pi->last_refreshed, <))
5355                 return;
5356
5357         tnl_cong_drops = 0;
5358         t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
5359         bg_map = pi->mps_bg_map;
5360         while (bg_map) {
5361                 i = ffs(bg_map) - 1;
5362                 mtx_lock(&sc->reg_lock);
5363                 t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v, 1,
5364                     A_TP_MIB_TNL_CNG_DROP_0 + i);
5365                 mtx_unlock(&sc->reg_lock);
5366                 tnl_cong_drops += v;
5367                 bg_map &= ~(1 << i);
5368         }
5369         pi->tnl_cong_drops = tnl_cong_drops;
5370         getmicrotime(&pi->last_refreshed);
5371 }
5372
5373 static void
5374 cxgbe_tick(void *arg)
5375 {
5376         struct port_info *pi = arg;
5377         struct adapter *sc = pi->adapter;
5378
5379         PORT_LOCK_ASSERT_OWNED(pi);
5380         cxgbe_refresh_stats(sc, pi);
5381
5382         callout_schedule(&pi->tick, hz);
5383 }
5384
5385 void
5386 vi_tick(void *arg)
5387 {
5388         struct vi_info *vi = arg;
5389         struct adapter *sc = vi->pi->adapter;
5390
5391         vi_refresh_stats(sc, vi);
5392
5393         callout_schedule(&vi->tick, hz);
5394 }
5395
5396 static void
5397 cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
5398 {
5399         struct ifnet *vlan;
5400
5401         if (arg != ifp || ifp->if_type != IFT_ETHER)
5402                 return;
5403
5404         vlan = VLAN_DEVAT(ifp, vid);
5405         VLAN_SETCOOKIE(vlan, ifp);
5406 }
5407
5408 /*
5409  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
5410  */
5411 static char *caps_decoder[] = {
5412         "\20\001IPMI\002NCSI",                          /* 0: NBM */
5413         "\20\001PPP\002QFC\003DCBX",                    /* 1: link */
5414         "\20\001INGRESS\002EGRESS",                     /* 2: switch */
5415         "\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"      /* 3: NIC */
5416             "\006HASHFILTER\007ETHOFLD",
5417         "\20\001TOE",                                   /* 4: TOE */
5418         "\20\001RDDP\002RDMAC",                         /* 5: RDMA */
5419         "\20\001INITIATOR_PDU\002TARGET_PDU"            /* 6: iSCSI */
5420             "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
5421             "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
5422             "\007T10DIF"
5423             "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
5424         "\20\001LOOKASIDE\002TLSKEYS",                  /* 7: Crypto */
5425         "\20\001INITIATOR\002TARGET\003CTRL_OFLD"       /* 8: FCoE */
5426                     "\004PO_INITIATOR\005PO_TARGET",
5427 };
5428
5429 void
5430 t4_sysctls(struct adapter *sc)
5431 {
5432         struct sysctl_ctx_list *ctx;
5433         struct sysctl_oid *oid;
5434         struct sysctl_oid_list *children, *c0;
5435         static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
5436
5437         ctx = device_get_sysctl_ctx(sc->dev);
5438
5439         /*
5440          * dev.t4nex.X.
5441          */
5442         oid = device_get_sysctl_tree(sc->dev);
5443         c0 = children = SYSCTL_CHILDREN(oid);
5444
5445         sc->sc_do_rxcopy = 1;
5446         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
5447             &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
5448
5449         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
5450             sc->params.nports, "# of ports");
5451
5452         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
5453             CTLTYPE_STRING | CTLFLAG_RD, doorbells, sc->doorbells,
5454             sysctl_bitfield, "A", "available doorbells");
5455
5456         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
5457             sc->params.vpd.cclk, "core clock frequency (in KHz)");
5458
5459         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
5460             CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
5461             sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
5462             "interrupt holdoff timer values (us)");
5463
5464         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
5465             CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
5466             sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
5467             "interrupt holdoff packet counter values");
5468
5469         t4_sge_sysctls(sc, ctx, children);
5470
5471         sc->lro_timeout = 100;
5472         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
5473             &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
5474
5475         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
5476             &sc->debug_flags, 0, "flags to enable runtime debugging");
5477
5478         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
5479             CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
5480
5481         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
5482             CTLFLAG_RD, sc->fw_version, 0, "firmware version");
5483
5484         if (sc->flags & IS_VF)
5485                 return;
5486
5487         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
5488             NULL, chip_rev(sc), "chip hardware revision");
5489
5490         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
5491             CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
5492
5493         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
5494             CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
5495
5496         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
5497             CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
5498
5499         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "md_version",
5500             CTLFLAG_RD, sc->params.vpd.md, 0, "manufacturing diags version");
5501
5502         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
5503             CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
5504
5505         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
5506             sc->er_version, 0, "expansion ROM version");
5507
5508         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
5509             sc->bs_version, 0, "bootstrap firmware version");
5510
5511         SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
5512             NULL, sc->params.scfg_vers, "serial config version");
5513
5514         SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
5515             NULL, sc->params.vpd_vers, "VPD version");
5516
5517         SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
5518             CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
5519
5520         SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
5521             sc->cfcsum, "config file checksum");
5522
5523 #define SYSCTL_CAP(name, n, text) \
5524         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
5525             CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], sc->name, \
5526             sysctl_bitfield, "A", "available " text " capabilities")
5527
5528         SYSCTL_CAP(nbmcaps, 0, "NBM");
5529         SYSCTL_CAP(linkcaps, 1, "link");
5530         SYSCTL_CAP(switchcaps, 2, "switch");
5531         SYSCTL_CAP(niccaps, 3, "NIC");
5532         SYSCTL_CAP(toecaps, 4, "TCP offload");
5533         SYSCTL_CAP(rdmacaps, 5, "RDMA");
5534         SYSCTL_CAP(iscsicaps, 6, "iSCSI");
5535         SYSCTL_CAP(cryptocaps, 7, "crypto");
5536         SYSCTL_CAP(fcoecaps, 8, "FCoE");
5537 #undef SYSCTL_CAP
5538
5539         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
5540             NULL, sc->tids.nftids, "number of filters");
5541
5542         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
5543             CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
5544             "chip temperature (in Celsius)");
5545
5546         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "loadavg", CTLTYPE_STRING |
5547             CTLFLAG_RD, sc, 0, sysctl_loadavg, "A",
5548             "microprocessor load averages (debug firmwares only)");
5549
5550         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_vdd", CTLFLAG_RD,
5551             &sc->params.core_vdd, 0, "core Vdd (in mV)");
5552
5553         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "local_cpus",
5554             CTLTYPE_STRING | CTLFLAG_RD, sc, LOCAL_CPUS,
5555             sysctl_cpus, "A", "local CPUs");
5556
5557         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "intr_cpus",
5558             CTLTYPE_STRING | CTLFLAG_RD, sc, INTR_CPUS,
5559             sysctl_cpus, "A", "preferred CPUs for interrupts");
5560
5561         /*
5562          * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
5563          */
5564         oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
5565             CTLFLAG_RD | CTLFLAG_SKIP, NULL,
5566             "logs and miscellaneous information");
5567         children = SYSCTL_CHILDREN(oid);
5568
5569         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
5570             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5571             sysctl_cctrl, "A", "congestion control");
5572
5573         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
5574             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5575             sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
5576
5577         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
5578             CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
5579             sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
5580
5581         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
5582             CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
5583             sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
5584
5585         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
5586             CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
5587             sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
5588
5589         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
5590             CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
5591             sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
5592
5593         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
5594             CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
5595             sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
5596
5597         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
5598             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5599             chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
5600             "A", "CIM logic analyzer");
5601
5602         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
5603             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5604             sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
5605
5606         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
5607             CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
5608             sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
5609
5610         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
5611             CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
5612             sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
5613
5614         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
5615             CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
5616             sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
5617
5618         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
5619             CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
5620             sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
5621
5622         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
5623             CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
5624             sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
5625
5626         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
5627             CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
5628             sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
5629
5630         if (chip_id(sc) > CHELSIO_T4) {
5631                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
5632                     CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
5633                     sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
5634
5635                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
5636                     CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
5637                     sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
5638         }
5639
5640         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
5641             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5642             sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
5643
5644         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
5645             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5646             sysctl_cim_qcfg, "A", "CIM queue configuration");
5647
5648         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
5649             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5650             sysctl_cpl_stats, "A", "CPL statistics");
5651
5652         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
5653             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5654             sysctl_ddp_stats, "A", "non-TCP DDP statistics");
5655
5656         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
5657             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5658             sysctl_devlog, "A", "firmware's device log");
5659
5660         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
5661             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5662             sysctl_fcoe_stats, "A", "FCoE statistics");
5663
5664         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
5665             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5666             sysctl_hw_sched, "A", "hardware scheduler ");
5667
5668         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
5669             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5670             sysctl_l2t, "A", "hardware L2 table");
5671
5672         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "smt",
5673             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5674             sysctl_smt, "A", "hardware source MAC table");
5675
5676         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
5677             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5678             sysctl_lb_stats, "A", "loopback statistics");
5679
5680         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
5681             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5682             sysctl_meminfo, "A", "memory regions");
5683
5684         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
5685             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5686             chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
5687             "A", "MPS TCAM entries");
5688
5689         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
5690             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5691             sysctl_path_mtus, "A", "path MTUs");
5692
5693         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
5694             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5695             sysctl_pm_stats, "A", "PM statistics");
5696
5697         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
5698             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5699             sysctl_rdma_stats, "A", "RDMA statistics");
5700
5701         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
5702             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5703             sysctl_tcp_stats, "A", "TCP statistics");
5704
5705         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
5706             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5707             sysctl_tids, "A", "TID information");
5708
5709         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
5710             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5711             sysctl_tp_err_stats, "A", "TP error statistics");
5712
5713         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
5714             CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
5715             "TP logic analyzer event capture mask");
5716
5717         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
5718             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5719             sysctl_tp_la, "A", "TP logic analyzer");
5720
5721         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
5722             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5723             sysctl_tx_rate, "A", "Tx rate");
5724
5725         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
5726             CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5727             sysctl_ulprx_la, "A", "ULPRX logic analyzer");
5728
5729         if (chip_id(sc) >= CHELSIO_T5) {
5730                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
5731                     CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5732                     sysctl_wcwr_stats, "A", "write combined work requests");
5733         }
5734
5735 #ifdef TCP_OFFLOAD
5736         if (is_offload(sc)) {
5737                 int i;
5738                 char s[4];
5739
5740                 /*
5741                  * dev.t4nex.X.toe.
5742                  */
5743                 oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
5744                     NULL, "TOE parameters");
5745                 children = SYSCTL_CHILDREN(oid);
5746
5747                 sc->tt.cong_algorithm = -1;
5748                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "cong_algorithm",
5749                     CTLFLAG_RW, &sc->tt.cong_algorithm, 0, "congestion control "
5750                     "(-1 = default, 0 = reno, 1 = tahoe, 2 = newreno, "
5751                     "3 = highspeed)");
5752
5753                 sc->tt.sndbuf = 256 * 1024;
5754                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
5755                     &sc->tt.sndbuf, 0, "max hardware send buffer size");
5756
5757                 sc->tt.ddp = 0;
5758                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
5759                     &sc->tt.ddp, 0, "DDP allowed");
5760
5761                 sc->tt.rx_coalesce = 1;
5762                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
5763                     CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
5764
5765                 sc->tt.tls = 0;
5766                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tls", CTLFLAG_RW,
5767                     &sc->tt.tls, 0, "Inline TLS allowed");
5768
5769                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tls_rx_ports",
5770                     CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tls_rx_ports,
5771                     "I", "TCP ports that use inline TLS+TOE RX");
5772
5773                 sc->tt.tx_align = 1;
5774                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
5775                     CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
5776
5777                 sc->tt.tx_zcopy = 0;
5778                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
5779                     CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
5780                     "Enable zero-copy aio_write(2)");
5781
5782                 sc->tt.cop_managed_offloading = !!t4_cop_managed_offloading;
5783                 SYSCTL_ADD_INT(ctx, children, OID_AUTO,
5784                     "cop_managed_offloading", CTLFLAG_RW,
5785                     &sc->tt.cop_managed_offloading, 0,
5786                     "COP (Connection Offload Policy) controls all TOE offload");
5787
5788                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
5789                     CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
5790                     "TP timer tick (us)");
5791
5792                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
5793                     CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
5794                     "TCP timestamp tick (us)");
5795
5796                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
5797                     CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
5798                     "DACK tick (us)");
5799
5800                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
5801                     CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
5802                     "IU", "DACK timer (us)");
5803
5804                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
5805                     CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
5806                     sysctl_tp_timer, "LU", "Minimum retransmit interval (us)");
5807
5808                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
5809                     CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
5810                     sysctl_tp_timer, "LU", "Maximum retransmit interval (us)");
5811
5812                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
5813                     CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
5814                     sysctl_tp_timer, "LU", "Persist timer min (us)");
5815
5816                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
5817                     CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
5818                     sysctl_tp_timer, "LU", "Persist timer max (us)");
5819
5820                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
5821                     CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
5822                     sysctl_tp_timer, "LU", "Keepalive idle timer (us)");
5823
5824                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_interval",
5825                     CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
5826                     sysctl_tp_timer, "LU", "Keepalive interval timer (us)");
5827
5828                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
5829                     CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
5830                     sysctl_tp_timer, "LU", "Initial SRTT (us)");
5831
5832                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
5833                     CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
5834                     sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
5835
5836                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "syn_rexmt_count",
5837                     CTLTYPE_UINT | CTLFLAG_RD, sc, S_SYNSHIFTMAX,
5838                     sysctl_tp_shift_cnt, "IU",
5839                     "Number of SYN retransmissions before abort");
5840
5841                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_count",
5842                     CTLTYPE_UINT | CTLFLAG_RD, sc, S_RXTSHIFTMAXR2,
5843                     sysctl_tp_shift_cnt, "IU",
5844                     "Number of retransmissions before abort");
5845
5846                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_count",
5847                     CTLTYPE_UINT | CTLFLAG_RD, sc, S_KEEPALIVEMAXR2,
5848                     sysctl_tp_shift_cnt, "IU",
5849                     "Number of keepalive probes before abort");
5850
5851                 oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "rexmt_backoff",
5852                     CTLFLAG_RD, NULL, "TOE retransmit backoffs");
5853                 children = SYSCTL_CHILDREN(oid);
5854                 for (i = 0; i < 16; i++) {
5855                         snprintf(s, sizeof(s), "%u", i);
5856                         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, s,
5857                             CTLTYPE_UINT | CTLFLAG_RD, sc, i, sysctl_tp_backoff,
5858                             "IU", "TOE retransmit backoff");
5859                 }
5860         }
5861 #endif
5862 }
5863
5864 void
5865 vi_sysctls(struct vi_info *vi)
5866 {
5867         struct sysctl_ctx_list *ctx;
5868         struct sysctl_oid *oid;
5869         struct sysctl_oid_list *children;
5870
5871         ctx = device_get_sysctl_ctx(vi->dev);
5872
5873         /*
5874          * dev.v?(cxgbe|cxl).X.
5875          */
5876         oid = device_get_sysctl_tree(vi->dev);
5877         children = SYSCTL_CHILDREN(oid);
5878
5879         SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
5880             vi->viid, "VI identifer");
5881         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
5882             &vi->nrxq, 0, "# of rx queues");
5883         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
5884             &vi->ntxq, 0, "# of tx queues");
5885         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
5886             &vi->first_rxq, 0, "index of first rx queue");
5887         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
5888             &vi->first_txq, 0, "index of first tx queue");
5889         SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
5890             vi->rss_size, "size of RSS indirection table");
5891
5892         if (IS_MAIN_VI(vi)) {
5893                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
5894                     CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
5895                     "Reserve queue 0 for non-flowid packets");
5896         }
5897
5898 #ifdef TCP_OFFLOAD
5899         if (vi->nofldrxq != 0) {
5900                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
5901                     &vi->nofldrxq, 0,
5902                     "# of rx queues for offloaded TCP connections");
5903                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
5904                     &vi->nofldtxq, 0,
5905                     "# of tx queues for offloaded TCP connections");
5906                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
5907                     CTLFLAG_RD, &vi->first_ofld_rxq, 0,
5908                     "index of first TOE rx queue");
5909                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
5910                     CTLFLAG_RD, &vi->first_ofld_txq, 0,
5911                     "index of first TOE tx queue");
5912                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx_ofld",
5913                     CTLTYPE_INT | CTLFLAG_RW, vi, 0,
5914                     sysctl_holdoff_tmr_idx_ofld, "I",
5915                     "holdoff timer index for TOE queues");
5916                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx_ofld",
5917                     CTLTYPE_INT | CTLFLAG_RW, vi, 0,
5918                     sysctl_holdoff_pktc_idx_ofld, "I",
5919                     "holdoff packet counter index for TOE queues");
5920         }
5921 #endif
5922 #ifdef DEV_NETMAP
5923         if (vi->nnmrxq != 0) {
5924                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
5925                     &vi->nnmrxq, 0, "# of netmap rx queues");
5926                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
5927                     &vi->nnmtxq, 0, "# of netmap tx queues");
5928                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
5929                     CTLFLAG_RD, &vi->first_nm_rxq, 0,
5930                     "index of first netmap rx queue");
5931                 SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
5932                     CTLFLAG_RD, &vi->first_nm_txq, 0,
5933                     "index of first netmap tx queue");
5934         }
5935 #endif
5936
5937         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
5938             CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
5939             "holdoff timer index");
5940         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
5941             CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
5942             "holdoff packet counter index");
5943
5944         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
5945             CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
5946             "rx queue size");
5947         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
5948             CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
5949             "tx queue size");
5950 }
5951
5952 static void
5953 cxgbe_sysctls(struct port_info *pi)
5954 {
5955         struct sysctl_ctx_list *ctx;
5956         struct sysctl_oid *oid;
5957         struct sysctl_oid_list *children, *children2;
5958         struct adapter *sc = pi->adapter;
5959         int i;
5960         char name[16];
5961
5962         ctx = device_get_sysctl_ctx(pi->dev);
5963
5964         /*
5965          * dev.cxgbe.X.
5966          */
5967         oid = device_get_sysctl_tree(pi->dev);
5968         children = SYSCTL_CHILDREN(oid);
5969
5970         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
5971            CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
5972         if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
5973                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
5974                     CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
5975                     "PHY temperature (in Celsius)");
5976                 SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
5977                     CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
5978                     "PHY firmware version");
5979         }
5980
5981         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
5982             CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
5983             "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
5984         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
5985             CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
5986             "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
5987         SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
5988             CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
5989             "autonegotiation (-1 = not supported)");
5990
5991         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
5992             port_top_speed(pi), "max speed (in Gbps)");
5993         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "mps_bg_map", CTLFLAG_RD, NULL,
5994             pi->mps_bg_map, "MPS buffer group map");
5995         SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_e_chan_map", CTLFLAG_RD,
5996             NULL, pi->rx_e_chan_map, "TP rx e-channel map");
5997
5998         if (sc->flags & IS_VF)
5999                 return;
6000
6001         /*
6002          * dev.(cxgbe|cxl).X.tc.
6003          */
6004         oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
6005             "Tx scheduler traffic classes (cl_rl)");
6006         for (i = 0; i < sc->chip_params->nsched_cls; i++) {
6007                 struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
6008
6009                 snprintf(name, sizeof(name), "%d", i);
6010                 children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
6011                     SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
6012                     "traffic class"));
6013                 SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "flags", CTLFLAG_RD,
6014                     &tc->flags, 0, "flags");
6015                 SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
6016                     CTLFLAG_RD, &tc->refcount, 0, "references to this class");
6017                 SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
6018                     CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
6019                     sysctl_tc_params, "A", "traffic class parameters");
6020         }
6021
6022         /*
6023          * dev.cxgbe.X.stats.
6024          */
6025         oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
6026             NULL, "port statistics");
6027         children = SYSCTL_CHILDREN(oid);
6028         SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
6029             &pi->tx_parse_error, 0,
6030             "# of tx packets with invalid length or # of segments");
6031
6032 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
6033         SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
6034             CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
6035             sysctl_handle_t4_reg64, "QU", desc)
6036
6037         SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
6038             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
6039         SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
6040             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
6041         SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
6042             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
6043         SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
6044             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
6045         SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
6046             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
6047         SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
6048             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
6049         SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
6050             "# of tx frames in this range",
6051             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
6052         SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
6053             "# of tx frames in this range",
6054             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
6055         SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
6056             "# of tx frames in this range",
6057             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
6058         SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
6059             "# of tx frames in this range",
6060             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
6061         SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
6062             "# of tx frames in this range",
6063             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
6064         SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
6065             "# of tx frames in this range",
6066             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
6067         SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
6068             "# of tx frames in this range",
6069             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
6070         SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
6071             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
6072         SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
6073             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
6074         SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
6075             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
6076         SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
6077             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
6078         SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
6079             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
6080         SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
6081             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
6082         SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
6083             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
6084         SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
6085             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
6086         SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
6087             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
6088         SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
6089             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
6090
6091         SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
6092             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
6093         SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
6094             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
6095         SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
6096             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
6097         SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
6098             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
6099         SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
6100             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
6101         SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
6102             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
6103         SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
6104             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
6105         SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
6106             "# of frames received with bad FCS",
6107             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
6108         SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
6109             "# of frames received with length error",
6110             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
6111         SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
6112             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
6113         SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
6114             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
6115         SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
6116             "# of rx frames in this range",
6117             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
6118         SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
6119             "# of rx frames in this range",
6120             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
6121         SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
6122             "# of rx frames in this range",
6123             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
6124         SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
6125             "# of rx frames in this range",
6126             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
6127         SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
6128             "# of rx frames in this range",
6129             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
6130         SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
6131             "# of rx frames in this range",
6132             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
6133         SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
6134             "# of rx frames in this range",
6135             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
6136         SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
6137             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
6138         SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
6139             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
6140         SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
6141             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
6142         SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
6143             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
6144         SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
6145             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
6146         SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
6147             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
6148         SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
6149             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
6150         SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
6151             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
6152         SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
6153             PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
6154
6155 #undef SYSCTL_ADD_T4_REG64
6156
6157 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
6158         SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
6159             &pi->stats.name, desc)
6160
6161         /* We get these from port_stats and they may be stale by up to 1s */
6162         SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
6163             "# drops due to buffer-group 0 overflows");
6164         SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
6165             "# drops due to buffer-group 1 overflows");
6166         SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
6167             "# drops due to buffer-group 2 overflows");
6168         SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
6169             "# drops due to buffer-group 3 overflows");
6170         SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
6171             "# of buffer-group 0 truncated packets");
6172         SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
6173             "# of buffer-group 1 truncated packets");
6174         SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
6175             "# of buffer-group 2 truncated packets");
6176         SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
6177             "# of buffer-group 3 truncated packets");
6178
6179 #undef SYSCTL_ADD_T4_PORTSTAT
6180
6181         SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_records",
6182             CTLFLAG_RD, &pi->tx_tls_records,
6183             "# of TLS records transmitted");
6184         SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_octets",
6185             CTLFLAG_RD, &pi->tx_tls_octets,
6186             "# of payload octets in transmitted TLS records");
6187         SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_records",
6188             CTLFLAG_RD, &pi->rx_tls_records,
6189             "# of TLS records received");
6190         SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_octets",
6191             CTLFLAG_RD, &pi->rx_tls_octets,
6192             "# of payload octets in received TLS records");
6193 }
6194
6195 static int
6196 sysctl_int_array(SYSCTL_HANDLER_ARGS)
6197 {
6198         int rc, *i, space = 0;
6199         struct sbuf sb;
6200
6201         sbuf_new_for_sysctl(&sb, NULL, 64, req);
6202         for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
6203                 if (space)
6204                         sbuf_printf(&sb, " ");
6205                 sbuf_printf(&sb, "%d", *i);
6206                 space = 1;
6207         }
6208         rc = sbuf_finish(&sb);
6209         sbuf_delete(&sb);
6210         return (rc);
6211 }
6212
6213 static int
6214 sysctl_bitfield(SYSCTL_HANDLER_ARGS)
6215 {
6216         int rc;
6217         struct sbuf *sb;
6218
6219         rc = sysctl_wire_old_buffer(req, 0);
6220         if (rc != 0)
6221                 return(rc);
6222
6223         sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6224         if (sb == NULL)
6225                 return (ENOMEM);
6226
6227         sbuf_printf(sb, "%b", (int)arg2, (char *)arg1);
6228         rc = sbuf_finish(sb);
6229         sbuf_delete(sb);
6230
6231         return (rc);
6232 }
6233
6234 static int
6235 sysctl_btphy(SYSCTL_HANDLER_ARGS)
6236 {
6237         struct port_info *pi = arg1;
6238         int op = arg2;
6239         struct adapter *sc = pi->adapter;
6240         u_int v;
6241         int rc;
6242
6243         rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
6244         if (rc)
6245                 return (rc);
6246         /* XXX: magic numbers */
6247         rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
6248             &v);
6249         end_synchronized_op(sc, 0);
6250         if (rc)
6251                 return (rc);
6252         if (op == 0)
6253                 v /= 256;
6254
6255         rc = sysctl_handle_int(oidp, &v, 0, req);
6256         return (rc);
6257 }
6258
6259 static int
6260 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
6261 {
6262         struct vi_info *vi = arg1;
6263         int rc, val;
6264
6265         val = vi->rsrv_noflowq;
6266         rc = sysctl_handle_int(oidp, &val, 0, req);
6267         if (rc != 0 || req->newptr == NULL)
6268                 return (rc);
6269
6270         if ((val >= 1) && (vi->ntxq > 1))
6271                 vi->rsrv_noflowq = 1;
6272         else
6273                 vi->rsrv_noflowq = 0;
6274
6275         return (rc);
6276 }
6277
6278 static int
6279 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
6280 {
6281         struct vi_info *vi = arg1;
6282         struct adapter *sc = vi->pi->adapter;
6283         int idx, rc, i;
6284         struct sge_rxq *rxq;
6285         uint8_t v;
6286
6287         idx = vi->tmr_idx;
6288
6289         rc = sysctl_handle_int(oidp, &idx, 0, req);
6290         if (rc != 0 || req->newptr == NULL)
6291                 return (rc);
6292
6293         if (idx < 0 || idx >= SGE_NTIMERS)
6294                 return (EINVAL);
6295
6296         rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6297             "t4tmr");
6298         if (rc)
6299                 return (rc);
6300
6301         v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
6302         for_each_rxq(vi, i, rxq) {
6303 #ifdef atomic_store_rel_8
6304                 atomic_store_rel_8(&rxq->iq.intr_params, v);
6305 #else
6306                 rxq->iq.intr_params = v;
6307 #endif
6308         }
6309         vi->tmr_idx = idx;
6310
6311         end_synchronized_op(sc, LOCK_HELD);
6312         return (0);
6313 }
6314
6315 static int
6316 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
6317 {
6318         struct vi_info *vi = arg1;
6319         struct adapter *sc = vi->pi->adapter;
6320         int idx, rc;
6321
6322         idx = vi->pktc_idx;
6323
6324         rc = sysctl_handle_int(oidp, &idx, 0, req);
6325         if (rc != 0 || req->newptr == NULL)
6326                 return (rc);
6327
6328         if (idx < -1 || idx >= SGE_NCOUNTERS)
6329                 return (EINVAL);
6330
6331         rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6332             "t4pktc");
6333         if (rc)
6334                 return (rc);
6335
6336         if (vi->flags & VI_INIT_DONE)
6337                 rc = EBUSY; /* cannot be changed once the queues are created */
6338         else
6339                 vi->pktc_idx = idx;
6340
6341         end_synchronized_op(sc, LOCK_HELD);
6342         return (rc);
6343 }
6344
6345 static int
6346 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
6347 {
6348         struct vi_info *vi = arg1;
6349         struct adapter *sc = vi->pi->adapter;
6350         int qsize, rc;
6351
6352         qsize = vi->qsize_rxq;
6353
6354         rc = sysctl_handle_int(oidp, &qsize, 0, req);
6355         if (rc != 0 || req->newptr == NULL)
6356                 return (rc);
6357
6358         if (qsize < 128 || (qsize & 7))
6359                 return (EINVAL);
6360
6361         rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6362             "t4rxqs");
6363         if (rc)
6364                 return (rc);
6365
6366         if (vi->flags & VI_INIT_DONE)
6367                 rc = EBUSY; /* cannot be changed once the queues are created */
6368         else
6369                 vi->qsize_rxq = qsize;
6370
6371         end_synchronized_op(sc, LOCK_HELD);
6372         return (rc);
6373 }
6374
6375 static int
6376 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
6377 {
6378         struct vi_info *vi = arg1;
6379         struct adapter *sc = vi->pi->adapter;
6380         int qsize, rc;
6381
6382         qsize = vi->qsize_txq;
6383
6384         rc = sysctl_handle_int(oidp, &qsize, 0, req);
6385         if (rc != 0 || req->newptr == NULL)
6386                 return (rc);
6387
6388         if (qsize < 128 || qsize > 65536)
6389                 return (EINVAL);
6390
6391         rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6392             "t4txqs");
6393         if (rc)
6394                 return (rc);
6395
6396         if (vi->flags & VI_INIT_DONE)
6397                 rc = EBUSY; /* cannot be changed once the queues are created */
6398         else
6399                 vi->qsize_txq = qsize;
6400
6401         end_synchronized_op(sc, LOCK_HELD);
6402         return (rc);
6403 }
6404
6405 static int
6406 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
6407 {
6408         struct port_info *pi = arg1;
6409         struct adapter *sc = pi->adapter;
6410         struct link_config *lc = &pi->link_cfg;
6411         int rc;
6412
6413         if (req->newptr == NULL) {
6414                 struct sbuf *sb;
6415                 static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
6416
6417                 rc = sysctl_wire_old_buffer(req, 0);
6418                 if (rc != 0)
6419                         return(rc);
6420
6421                 sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6422                 if (sb == NULL)
6423                         return (ENOMEM);
6424
6425                 sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
6426                 rc = sbuf_finish(sb);
6427                 sbuf_delete(sb);
6428         } else {
6429                 char s[2];
6430                 int n;
6431
6432                 s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
6433                 s[1] = 0;
6434
6435                 rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6436                 if (rc != 0)
6437                         return(rc);
6438
6439                 if (s[1] != 0)
6440                         return (EINVAL);
6441                 if (s[0] < '0' || s[0] > '9')
6442                         return (EINVAL);        /* not a number */
6443                 n = s[0] - '0';
6444                 if (n & ~(PAUSE_TX | PAUSE_RX))
6445                         return (EINVAL);        /* some other bit is set too */
6446
6447                 rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6448                     "t4PAUSE");
6449                 if (rc)
6450                         return (rc);
6451                 PORT_LOCK(pi);
6452                 if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
6453                         lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
6454                         lc->requested_fc |= n;
6455                         rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6456                         if (rc == 0) {
6457                                 lc->fc = lc->requested_fc;
6458                                 set_current_media(pi, &pi->media);
6459                         }
6460                 }
6461                 PORT_UNLOCK(pi);
6462                 end_synchronized_op(sc, 0);
6463         }
6464
6465         return (rc);
6466 }
6467
6468 static int
6469 sysctl_fec(SYSCTL_HANDLER_ARGS)
6470 {
6471         struct port_info *pi = arg1;
6472         struct adapter *sc = pi->adapter;
6473         struct link_config *lc = &pi->link_cfg;
6474         int rc;
6475
6476         if (req->newptr == NULL) {
6477                 struct sbuf *sb;
6478                 static char *bits = "\20\1RS\2BASER_RS\3RESERVED";
6479
6480                 rc = sysctl_wire_old_buffer(req, 0);
6481                 if (rc != 0)
6482                         return(rc);
6483
6484                 sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6485                 if (sb == NULL)
6486                         return (ENOMEM);
6487
6488                 sbuf_printf(sb, "%b", lc->fec & M_FW_PORT_CAP_FEC, bits);
6489                 rc = sbuf_finish(sb);
6490                 sbuf_delete(sb);
6491         } else {
6492                 char s[2];
6493                 int n;
6494
6495                 s[0] = '0' + (lc->requested_fec & M_FW_PORT_CAP_FEC);
6496                 s[1] = 0;
6497
6498                 rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6499                 if (rc != 0)
6500                         return(rc);
6501
6502                 if (s[1] != 0)
6503                         return (EINVAL);
6504                 if (s[0] < '0' || s[0] > '9')
6505                         return (EINVAL);        /* not a number */
6506                 n = s[0] - '0';
6507                 if (n & ~M_FW_PORT_CAP_FEC)
6508                         return (EINVAL);        /* some other bit is set too */
6509                 if (!powerof2(n))
6510                         return (EINVAL);        /* one bit can be set at most */
6511
6512                 rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6513                     "t4fec");
6514                 if (rc)
6515                         return (rc);
6516                 PORT_LOCK(pi);
6517                 if ((lc->requested_fec & M_FW_PORT_CAP_FEC) != n) {
6518                         lc->requested_fec = n &
6519                             G_FW_PORT_CAP_FEC(lc->supported);
6520                         rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6521                         if (rc == 0) {
6522                                 lc->fec = lc->requested_fec;
6523                         }
6524                 }
6525                 PORT_UNLOCK(pi);
6526                 end_synchronized_op(sc, 0);
6527         }
6528
6529         return (rc);
6530 }
6531
6532 static int
6533 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
6534 {
6535         struct port_info *pi = arg1;
6536         struct adapter *sc = pi->adapter;
6537         struct link_config *lc = &pi->link_cfg;
6538         int rc, val, old;
6539
6540         if (lc->supported & FW_PORT_CAP_ANEG)
6541                 val = lc->requested_aneg == AUTONEG_ENABLE ? 1 : 0;
6542         else
6543                 val = -1;
6544         rc = sysctl_handle_int(oidp, &val, 0, req);
6545         if (rc != 0 || req->newptr == NULL)
6546                 return (rc);
6547         if (val == 0)
6548                 val = AUTONEG_DISABLE;
6549         else if (val == 1)
6550                 val = AUTONEG_ENABLE;
6551         else
6552                 return (EINVAL);
6553
6554         rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6555             "t4aneg");
6556         if (rc)
6557                 return (rc);
6558         PORT_LOCK(pi);
6559         if ((lc->supported & FW_PORT_CAP_ANEG) == 0) {
6560                 rc = ENOTSUP;
6561                 goto done;
6562         }
6563         if (lc->requested_aneg == val) {
6564                 rc = 0; /* no change, do nothing. */
6565                 goto done;
6566         }
6567         old = lc->requested_aneg;
6568         lc->requested_aneg = val;
6569         rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6570         if (rc != 0)
6571                 lc->requested_aneg = old;
6572         else
6573                 set_current_media(pi, &pi->media);
6574 done:
6575         PORT_UNLOCK(pi);
6576         end_synchronized_op(sc, 0);
6577         return (rc);
6578 }
6579
6580 static int
6581 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
6582 {
6583         struct adapter *sc = arg1;
6584         int reg = arg2;
6585         uint64_t val;
6586
6587         val = t4_read_reg64(sc, reg);
6588
6589         return (sysctl_handle_64(oidp, &val, 0, req));
6590 }
6591
6592 static int
6593 sysctl_temperature(SYSCTL_HANDLER_ARGS)
6594 {
6595         struct adapter *sc = arg1;
6596         int rc, t;
6597         uint32_t param, val;
6598
6599         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
6600         if (rc)
6601                 return (rc);
6602         param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
6603             V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
6604             V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
6605         rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
6606         end_synchronized_op(sc, 0);
6607         if (rc)
6608                 return (rc);
6609
6610         /* unknown is returned as 0 but we display -1 in that case */
6611         t = val == 0 ? -1 : val;
6612
6613         rc = sysctl_handle_int(oidp, &t, 0, req);
6614         return (rc);
6615 }
6616
6617 static int
6618 sysctl_loadavg(SYSCTL_HANDLER_ARGS)
6619 {
6620         struct adapter *sc = arg1;
6621         struct sbuf *sb;
6622         int rc;
6623         uint32_t param, val;
6624
6625         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4lavg");
6626         if (rc)
6627                 return (rc);
6628         param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
6629             V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_LOAD);
6630         rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
6631         end_synchronized_op(sc, 0);
6632         if (rc)
6633                 return (rc);
6634
6635         rc = sysctl_wire_old_buffer(req, 0);
6636         if (rc != 0)
6637                 return (rc);
6638
6639         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6640         if (sb == NULL)
6641                 return (ENOMEM);
6642
6643         if (val == 0xffffffff) {
6644                 /* Only debug and custom firmwares report load averages. */
6645                 sbuf_printf(sb, "not available");
6646         } else {
6647                 sbuf_printf(sb, "%d %d %d", val & 0xff, (val >> 8) & 0xff,
6648                     (val >> 16) & 0xff);
6649         }
6650         rc = sbuf_finish(sb);
6651         sbuf_delete(sb);
6652
6653         return (rc);
6654 }
6655
6656 static int
6657 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
6658 {
6659         struct adapter *sc = arg1;
6660         struct sbuf *sb;
6661         int rc, i;
6662         uint16_t incr[NMTUS][NCCTRL_WIN];
6663         static const char *dec_fac[] = {
6664                 "0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
6665                 "0.9375"
6666         };
6667
6668         rc = sysctl_wire_old_buffer(req, 0);
6669         if (rc != 0)
6670                 return (rc);
6671
6672         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6673         if (sb == NULL)
6674                 return (ENOMEM);
6675
6676         t4_read_cong_tbl(sc, incr);
6677
6678         for (i = 0; i < NCCTRL_WIN; ++i) {
6679                 sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
6680                     incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
6681                     incr[5][i], incr[6][i], incr[7][i]);
6682                 sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
6683                     incr[8][i], incr[9][i], incr[10][i], incr[11][i],
6684                     incr[12][i], incr[13][i], incr[14][i], incr[15][i],
6685                     sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
6686         }
6687
6688         rc = sbuf_finish(sb);
6689         sbuf_delete(sb);
6690
6691         return (rc);
6692 }
6693
6694 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
6695         "TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",   /* ibq's */
6696         "ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI", /* obq's */
6697         "SGE0-RX", "SGE1-RX"    /* additional obq's (T5 onwards) */
6698 };
6699
6700 static int
6701 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
6702 {
6703         struct adapter *sc = arg1;
6704         struct sbuf *sb;
6705         int rc, i, n, qid = arg2;
6706         uint32_t *buf, *p;
6707         char *qtype;
6708         u_int cim_num_obq = sc->chip_params->cim_num_obq;
6709
6710         KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
6711             ("%s: bad qid %d\n", __func__, qid));
6712
6713         if (qid < CIM_NUM_IBQ) {
6714                 /* inbound queue */
6715                 qtype = "IBQ";
6716                 n = 4 * CIM_IBQ_SIZE;
6717                 buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6718                 rc = t4_read_cim_ibq(sc, qid, buf, n);
6719         } else {
6720                 /* outbound queue */
6721                 qtype = "OBQ";
6722                 qid -= CIM_NUM_IBQ;
6723                 n = 4 * cim_num_obq * CIM_OBQ_SIZE;
6724                 buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6725                 rc = t4_read_cim_obq(sc, qid, buf, n);
6726         }
6727
6728         if (rc < 0) {
6729                 rc = -rc;
6730                 goto done;
6731         }
6732         n = rc * sizeof(uint32_t);      /* rc has # of words actually read */
6733
6734         rc = sysctl_wire_old_buffer(req, 0);
6735         if (rc != 0)
6736                 goto done;
6737
6738         sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6739         if (sb == NULL) {
6740                 rc = ENOMEM;
6741                 goto done;
6742         }
6743
6744         sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
6745         for (i = 0, p = buf; i < n; i += 16, p += 4)
6746                 sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
6747                     p[2], p[3]);
6748
6749         rc = sbuf_finish(sb);
6750         sbuf_delete(sb);
6751 done:
6752         free(buf, M_CXGBE);
6753         return (rc);
6754 }
6755
6756 static int
6757 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
6758 {
6759         struct adapter *sc = arg1;
6760         u_int cfg;
6761         struct sbuf *sb;
6762         uint32_t *buf, *p;
6763         int rc;
6764
6765         MPASS(chip_id(sc) <= CHELSIO_T5);
6766
6767         rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6768         if (rc != 0)
6769                 return (rc);
6770
6771         rc = sysctl_wire_old_buffer(req, 0);
6772         if (rc != 0)
6773                 return (rc);
6774
6775         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6776         if (sb == NULL)
6777                 return (ENOMEM);
6778
6779         buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6780             M_ZERO | M_WAITOK);
6781
6782         rc = -t4_cim_read_la(sc, buf, NULL);
6783         if (rc != 0)
6784                 goto done;
6785
6786         sbuf_printf(sb, "Status   Data      PC%s",
6787             cfg & F_UPDBGLACAPTPCONLY ? "" :
6788             "     LS0Stat  LS0Addr             LS0Data");
6789
6790         for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
6791                 if (cfg & F_UPDBGLACAPTPCONLY) {
6792                         sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
6793                             p[6], p[7]);
6794                         sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
6795                             (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
6796                             p[4] & 0xff, p[5] >> 8);
6797                         sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
6798                             (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6799                             p[1] & 0xf, p[2] >> 4);
6800                 } else {
6801                         sbuf_printf(sb,
6802                             "\n  %02x   %x%07x %x%07x %08x %08x "
6803                             "%08x%08x%08x%08x",
6804                             (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6805                             p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
6806                             p[6], p[7]);
6807                 }
6808         }
6809
6810         rc = sbuf_finish(sb);
6811         sbuf_delete(sb);
6812 done:
6813         free(buf, M_CXGBE);
6814         return (rc);
6815 }
6816
6817 static int
6818 sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
6819 {
6820         struct adapter *sc = arg1;
6821         u_int cfg;
6822         struct sbuf *sb;
6823         uint32_t *buf, *p;
6824         int rc;
6825
6826         MPASS(chip_id(sc) > CHELSIO_T5);
6827
6828         rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6829         if (rc != 0)
6830                 return (rc);
6831
6832         rc = sysctl_wire_old_buffer(req, 0);
6833         if (rc != 0)
6834                 return (rc);
6835
6836         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6837         if (sb == NULL)
6838                 return (ENOMEM);
6839
6840         buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6841             M_ZERO | M_WAITOK);
6842
6843         rc = -t4_cim_read_la(sc, buf, NULL);
6844         if (rc != 0)
6845                 goto done;
6846
6847         sbuf_printf(sb, "Status   Inst    Data      PC%s",
6848             cfg & F_UPDBGLACAPTPCONLY ? "" :
6849             "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
6850
6851         for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
6852                 if (cfg & F_UPDBGLACAPTPCONLY) {
6853                         sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
6854                             p[3] & 0xff, p[2], p[1], p[0]);
6855                         sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
6856                             (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
6857                             p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
6858                         sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
6859                             (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
6860                             p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
6861                             p[6] >> 16);
6862                 } else {
6863                         sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
6864                             "%08x %08x %08x %08x %08x %08x",
6865                             (p[9] >> 16) & 0xff,
6866                             p[9] & 0xffff, p[8] >> 16,
6867                             p[8] & 0xffff, p[7] >> 16,
6868                             p[7] & 0xffff, p[6] >> 16,
6869                             p[2], p[1], p[0], p[5], p[4], p[3]);
6870                 }
6871         }
6872
6873         rc = sbuf_finish(sb);
6874         sbuf_delete(sb);
6875 done:
6876         free(buf, M_CXGBE);
6877         return (rc);
6878 }
6879
6880 static int
6881 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
6882 {
6883         struct adapter *sc = arg1;
6884         u_int i;
6885         struct sbuf *sb;
6886         uint32_t *buf, *p;
6887         int rc;
6888
6889         rc = sysctl_wire_old_buffer(req, 0);
6890         if (rc != 0)
6891                 return (rc);
6892
6893         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6894         if (sb == NULL)
6895                 return (ENOMEM);
6896
6897         buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
6898             M_ZERO | M_WAITOK);
6899
6900         t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
6901         p = buf;
6902
6903         for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6904                 sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
6905                     p[1], p[0]);
6906         }
6907
6908         sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
6909         for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6910                 sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
6911                     (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
6912                     (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
6913                     (p[1] >> 2) | ((p[2] & 3) << 30),
6914                     (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
6915                     p[0] & 1);
6916         }
6917
6918         rc = sbuf_finish(sb);
6919         sbuf_delete(sb);
6920         free(buf, M_CXGBE);
6921         return (rc);
6922 }
6923
6924 static int
6925 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
6926 {
6927         struct adapter *sc = arg1;
6928         u_int i;
6929         struct sbuf *sb;
6930         uint32_t *buf, *p;
6931         int rc;
6932
6933         rc = sysctl_wire_old_buffer(req, 0);
6934         if (rc != 0)
6935                 return (rc);
6936
6937         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6938         if (sb == NULL)
6939                 return (ENOMEM);
6940
6941         buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
6942             M_ZERO | M_WAITOK);
6943
6944         t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
6945         p = buf;
6946
6947         sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
6948         for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6949                 sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
6950                     (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
6951                     p[4], p[3], p[2], p[1], p[0]);
6952         }
6953
6954         sbuf_printf(sb, "\n\nCntl ID               Data");
6955         for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6956                 sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
6957                     (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
6958         }
6959
6960         rc = sbuf_finish(sb);
6961         sbuf_delete(sb);
6962         free(buf, M_CXGBE);
6963         return (rc);
6964 }
6965
6966 static int
6967 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
6968 {
6969         struct adapter *sc = arg1;
6970         struct sbuf *sb;
6971         int rc, i;
6972         uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6973         uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6974         uint16_t thres[CIM_NUM_IBQ];
6975         uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
6976         uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
6977         u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
6978
6979         cim_num_obq = sc->chip_params->cim_num_obq;
6980         if (is_t4(sc)) {
6981                 ibq_rdaddr = A_UP_IBQ_0_RDADDR;
6982                 obq_rdaddr = A_UP_OBQ_0_REALADDR;
6983         } else {
6984                 ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
6985                 obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
6986         }
6987         nq = CIM_NUM_IBQ + cim_num_obq;
6988
6989         rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
6990         if (rc == 0)
6991                 rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
6992         if (rc != 0)
6993                 return (rc);
6994
6995         t4_read_cimq_cfg(sc, base, size, thres);
6996
6997         rc = sysctl_wire_old_buffer(req, 0);
6998         if (rc != 0)
6999                 return (rc);
7000
7001         sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
7002         if (sb == NULL)
7003                 return (ENOMEM);
7004
7005         sbuf_printf(sb,
7006             "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
7007
7008         for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
7009                 sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
7010                     qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
7011                     G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
7012                     G_QUEREMFLITS(p[2]) * 16);
7013         for ( ; i < nq; i++, p += 4, wr += 2)
7014                 sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
7015                     base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
7016                     wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
7017                     G_QUEREMFLITS(p[2]) * 16);
7018
7019         rc = sbuf_finish(sb);
7020         sbuf_delete(sb);
7021
7022         return (rc);
7023 }
7024
7025 static int
7026 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
7027 {
7028         struct adapter *sc = arg1;
7029         struct sbuf *sb;
7030         int rc;
7031         struct tp_cpl_stats stats;
7032
7033         rc = sysctl_wire_old_buffer(req, 0);
7034         if (rc != 0)
7035                 return (rc);
7036
7037         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7038         if (sb == NULL)
7039                 return (ENOMEM);
7040
7041         mtx_lock(&sc->reg_lock);
7042         t4_tp_get_cpl_stats(sc, &stats, 0);
7043         mtx_unlock(&sc->reg_lock);
7044
7045         if (sc->chip_params->nchan > 2) {
7046                 sbuf_printf(sb, "                 channel 0  channel 1"
7047                     "  channel 2  channel 3");
7048                 sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
7049                     stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
7050                 sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
7051                     stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
7052         } else {
7053                 sbuf_printf(sb, "                 channel 0  channel 1");
7054                 sbuf_printf(sb, "\nCPL requests:   %10u %10u",
7055                     stats.req[0], stats.req[1]);
7056                 sbuf_printf(sb, "\nCPL responses:   %10u %10u",
7057                     stats.rsp[0], stats.rsp[1]);
7058         }
7059
7060         rc = sbuf_finish(sb);
7061         sbuf_delete(sb);
7062
7063         return (rc);
7064 }
7065
7066 static int
7067 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
7068 {
7069         struct adapter *sc = arg1;
7070         struct sbuf *sb;
7071         int rc;
7072         struct tp_usm_stats stats;
7073
7074         rc = sysctl_wire_old_buffer(req, 0);
7075         if (rc != 0)
7076                 return(rc);
7077
7078         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7079         if (sb == NULL)
7080                 return (ENOMEM);
7081
7082         t4_get_usm_stats(sc, &stats, 1);
7083
7084         sbuf_printf(sb, "Frames: %u\n", stats.frames);
7085         sbuf_printf(sb, "Octets: %ju\n", stats.octets);
7086         sbuf_printf(sb, "Drops:  %u", stats.drops);
7087
7088         rc = sbuf_finish(sb);
7089         sbuf_delete(sb);
7090
7091         return (rc);
7092 }
7093
7094 static const char * const devlog_level_strings[] = {
7095         [FW_DEVLOG_LEVEL_EMERG]         = "EMERG",
7096         [FW_DEVLOG_LEVEL_CRIT]          = "CRIT",
7097         [FW_DEVLOG_LEVEL_ERR]           = "ERR",
7098         [FW_DEVLOG_LEVEL_NOTICE]        = "NOTICE",
7099         [FW_DEVLOG_LEVEL_INFO]          = "INFO",
7100         [FW_DEVLOG_LEVEL_DEBUG]         = "DEBUG"
7101 };
7102
7103 static const char * const devlog_facility_strings[] = {
7104         [FW_DEVLOG_FACILITY_CORE]       = "CORE",
7105         [FW_DEVLOG_FACILITY_CF]         = "CF",
7106         [FW_DEVLOG_FACILITY_SCHED]      = "SCHED",
7107         [FW_DEVLOG_FACILITY_TIMER]      = "TIMER",
7108         [FW_DEVLOG_FACILITY_RES]        = "RES",
7109         [FW_DEVLOG_FACILITY_HW]         = "HW",
7110         [FW_DEVLOG_FACILITY_FLR]        = "FLR",
7111         [FW_DEVLOG_FACILITY_DMAQ]       = "DMAQ",
7112         [FW_DEVLOG_FACILITY_PHY]        = "PHY",
7113         [FW_DEVLOG_FACILITY_MAC]        = "MAC",
7114         [FW_DEVLOG_FACILITY_PORT]       = "PORT",
7115         [FW_DEVLOG_FACILITY_VI]         = "VI",
7116         [FW_DEVLOG_FACILITY_FILTER]     = "FILTER",
7117         [FW_DEVLOG_FACILITY_ACL]        = "ACL",
7118         [FW_DEVLOG_FACILITY_TM]         = "TM",
7119         [FW_DEVLOG_FACILITY_QFC]        = "QFC",
7120         [FW_DEVLOG_FACILITY_DCB]        = "DCB",
7121         [FW_DEVLOG_FACILITY_ETH]        = "ETH",
7122         [FW_DEVLOG_FACILITY_OFLD]       = "OFLD",
7123         [FW_DEVLOG_FACILITY_RI]         = "RI",
7124         [FW_DEVLOG_FACILITY_ISCSI]      = "ISCSI",
7125         [FW_DEVLOG_FACILITY_FCOE]       = "FCOE",
7126         [FW_DEVLOG_FACILITY_FOISCSI]    = "FOISCSI",
7127         [FW_DEVLOG_FACILITY_FOFCOE]     = "FOFCOE",
7128         [FW_DEVLOG_FACILITY_CHNET]      = "CHNET",
7129 };
7130
7131 static int
7132 sysctl_devlog(SYSCTL_HANDLER_ARGS)
7133 {
7134         struct adapter *sc = arg1;
7135         struct devlog_params *dparams = &sc->params.devlog;
7136         struct fw_devlog_e *buf, *e;
7137         int i, j, rc, nentries, first = 0;
7138         struct sbuf *sb;
7139         uint64_t ftstamp = UINT64_MAX;
7140
7141         if (dparams->addr == 0)
7142                 return (ENXIO);
7143
7144         buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
7145         if (buf == NULL)
7146                 return (ENOMEM);
7147
7148         rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
7149         if (rc != 0)
7150                 goto done;
7151
7152         nentries = dparams->size / sizeof(struct fw_devlog_e);
7153         for (i = 0; i < nentries; i++) {
7154                 e = &buf[i];
7155
7156                 if (e->timestamp == 0)
7157                         break;  /* end */
7158
7159                 e->timestamp = be64toh(e->timestamp);
7160                 e->seqno = be32toh(e->seqno);
7161                 for (j = 0; j < 8; j++)
7162                         e->params[j] = be32toh(e->params[j]);
7163
7164                 if (e->timestamp < ftstamp) {
7165                         ftstamp = e->timestamp;
7166                         first = i;
7167                 }
7168         }
7169
7170         if (buf[first].timestamp == 0)
7171                 goto done;      /* nothing in the log */
7172
7173         rc = sysctl_wire_old_buffer(req, 0);
7174         if (rc != 0)
7175                 goto done;
7176
7177         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7178         if (sb == NULL) {
7179                 rc = ENOMEM;
7180                 goto done;
7181         }
7182         sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
7183             "Seq#", "Tstamp", "Level", "Facility", "Message");
7184
7185         i = first;
7186         do {
7187                 e = &buf[i];
7188                 if (e->timestamp == 0)
7189                         break;  /* end */
7190
7191                 sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
7192                     e->seqno, e->timestamp,
7193                     (e->level < nitems(devlog_level_strings) ?
7194                         devlog_level_strings[e->level] : "UNKNOWN"),
7195                     (e->facility < nitems(devlog_facility_strings) ?
7196                         devlog_facility_strings[e->facility] : "UNKNOWN"));
7197                 sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
7198                     e->params[2], e->params[3], e->params[4],
7199                     e->params[5], e->params[6], e->params[7]);
7200
7201                 if (++i == nentries)
7202                         i = 0;
7203         } while (i != first);
7204
7205         rc = sbuf_finish(sb);
7206         sbuf_delete(sb);
7207 done:
7208         free(buf, M_CXGBE);
7209         return (rc);
7210 }
7211
7212 static int
7213 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
7214 {
7215         struct adapter *sc = arg1;
7216         struct sbuf *sb;
7217         int rc;
7218         struct tp_fcoe_stats stats[MAX_NCHAN];
7219         int i, nchan = sc->chip_params->nchan;
7220
7221         rc = sysctl_wire_old_buffer(req, 0);
7222         if (rc != 0)
7223                 return (rc);
7224
7225         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7226         if (sb == NULL)
7227                 return (ENOMEM);
7228
7229         for (i = 0; i < nchan; i++)
7230                 t4_get_fcoe_stats(sc, i, &stats[i], 1);
7231
7232         if (nchan > 2) {
7233                 sbuf_printf(sb, "                   channel 0        channel 1"
7234                     "        channel 2        channel 3");
7235                 sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
7236                     stats[0].octets_ddp, stats[1].octets_ddp,
7237                     stats[2].octets_ddp, stats[3].octets_ddp);
7238                 sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
7239                     stats[0].frames_ddp, stats[1].frames_ddp,
7240                     stats[2].frames_ddp, stats[3].frames_ddp);
7241                 sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
7242                     stats[0].frames_drop, stats[1].frames_drop,
7243                     stats[2].frames_drop, stats[3].frames_drop);
7244         } else {
7245                 sbuf_printf(sb, "                   channel 0        channel 1");
7246                 sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
7247                     stats[0].octets_ddp, stats[1].octets_ddp);
7248                 sbuf_printf(sb, "\nframesDDP:  %16u %16u",
7249                     stats[0].frames_ddp, stats[1].frames_ddp);
7250                 sbuf_printf(sb, "\nframesDrop: %16u %16u",
7251                     stats[0].frames_drop, stats[1].frames_drop);
7252         }
7253
7254         rc = sbuf_finish(sb);
7255         sbuf_delete(sb);
7256
7257         return (rc);
7258 }
7259
7260 static int
7261 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
7262 {
7263         struct adapter *sc = arg1;
7264         struct sbuf *sb;
7265         int rc, i;
7266         unsigned int map, kbps, ipg, mode;
7267         unsigned int pace_tab[NTX_SCHED];
7268
7269         rc = sysctl_wire_old_buffer(req, 0);
7270         if (rc != 0)
7271                 return (rc);
7272
7273         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7274         if (sb == NULL)
7275                 return (ENOMEM);
7276
7277         map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
7278         mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
7279         t4_read_pace_tbl(sc, pace_tab);
7280
7281         sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
7282             "Class IPG (0.1 ns)   Flow IPG (us)");
7283
7284         for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
7285                 t4_get_tx_sched(sc, i, &kbps, &ipg, 1);
7286                 sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
7287                     (mode & (1 << i)) ? "flow" : "class", map & 3);
7288                 if (kbps)
7289                         sbuf_printf(sb, "%9u     ", kbps);
7290                 else
7291                         sbuf_printf(sb, " disabled     ");
7292
7293                 if (ipg)
7294                         sbuf_printf(sb, "%13u        ", ipg);
7295                 else
7296                         sbuf_printf(sb, "     disabled        ");
7297
7298                 if (pace_tab[i])
7299                         sbuf_printf(sb, "%10u", pace_tab[i]);
7300                 else
7301                         sbuf_printf(sb, "  disabled");
7302         }
7303
7304         rc = sbuf_finish(sb);
7305         sbuf_delete(sb);
7306
7307         return (rc);
7308 }
7309
7310 static int
7311 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
7312 {
7313         struct adapter *sc = arg1;
7314         struct sbuf *sb;
7315         int rc, i, j;
7316         uint64_t *p0, *p1;
7317         struct lb_port_stats s[2];
7318         static const char *stat_name[] = {
7319                 "OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
7320                 "UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
7321                 "Frames128To255:", "Frames256To511:", "Frames512To1023:",
7322                 "Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
7323                 "BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
7324                 "BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
7325                 "BG2FramesTrunc:", "BG3FramesTrunc:"
7326         };
7327
7328         rc = sysctl_wire_old_buffer(req, 0);
7329         if (rc != 0)
7330                 return (rc);
7331
7332         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7333         if (sb == NULL)
7334                 return (ENOMEM);
7335
7336         memset(s, 0, sizeof(s));
7337
7338         for (i = 0; i < sc->chip_params->nchan; i += 2) {
7339                 t4_get_lb_stats(sc, i, &s[0]);
7340                 t4_get_lb_stats(sc, i + 1, &s[1]);
7341
7342                 p0 = &s[0].octets;
7343                 p1 = &s[1].octets;
7344                 sbuf_printf(sb, "%s                       Loopback %u"
7345                     "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
7346
7347                 for (j = 0; j < nitems(stat_name); j++)
7348                         sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
7349                                    *p0++, *p1++);
7350         }
7351
7352         rc = sbuf_finish(sb);
7353         sbuf_delete(sb);
7354
7355         return (rc);
7356 }
7357
7358 static int
7359 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
7360 {
7361         int rc = 0;
7362         struct port_info *pi = arg1;
7363         struct link_config *lc = &pi->link_cfg;
7364         struct sbuf *sb;
7365
7366         rc = sysctl_wire_old_buffer(req, 0);
7367         if (rc != 0)
7368                 return(rc);
7369         sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
7370         if (sb == NULL)
7371                 return (ENOMEM);
7372
7373         if (lc->link_ok || lc->link_down_rc == 255)
7374                 sbuf_printf(sb, "n/a");
7375         else
7376                 sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
7377
7378         rc = sbuf_finish(sb);
7379         sbuf_delete(sb);
7380
7381         return (rc);
7382 }
7383
7384 struct mem_desc {
7385         unsigned int base;
7386         unsigned int limit;
7387         unsigned int idx;
7388 };
7389
7390 static int
7391 mem_desc_cmp(const void *a, const void *b)
7392 {
7393         return ((const struct mem_desc *)a)->base -
7394                ((const struct mem_desc *)b)->base;
7395 }
7396
7397 static void
7398 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
7399     unsigned int to)
7400 {
7401         unsigned int size;
7402
7403         if (from == to)
7404                 return;
7405
7406         size = to - from + 1;
7407         if (size == 0)
7408                 return;
7409
7410         /* XXX: need humanize_number(3) in libkern for a more readable 'size' */
7411         sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
7412 }
7413
7414 static int
7415 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
7416 {
7417         struct adapter *sc = arg1;
7418         struct sbuf *sb;
7419         int rc, i, n;
7420         uint32_t lo, hi, used, alloc;
7421         static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
7422         static const char *region[] = {
7423                 "DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
7424                 "Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
7425                 "Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
7426                 "TDDP region:", "TPT region:", "STAG region:", "RQ region:",
7427                 "RQUDP region:", "PBL region:", "TXPBL region:",
7428                 "DBVFIFO region:", "ULPRX state:", "ULPTX state:",
7429                 "On-chip queues:", "TLS keys:",
7430         };
7431         struct mem_desc avail[4];
7432         struct mem_desc mem[nitems(region) + 3];        /* up to 3 holes */
7433         struct mem_desc *md = mem;
7434
7435         rc = sysctl_wire_old_buffer(req, 0);
7436         if (rc != 0)
7437                 return (rc);
7438
7439         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7440         if (sb == NULL)
7441                 return (ENOMEM);
7442
7443         for (i = 0; i < nitems(mem); i++) {
7444                 mem[i].limit = 0;
7445                 mem[i].idx = i;
7446         }
7447
7448         /* Find and sort the populated memory ranges */
7449         i = 0;
7450         lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
7451         if (lo & F_EDRAM0_ENABLE) {
7452                 hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
7453                 avail[i].base = G_EDRAM0_BASE(hi) << 20;
7454                 avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
7455                 avail[i].idx = 0;
7456                 i++;
7457         }
7458         if (lo & F_EDRAM1_ENABLE) {
7459                 hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
7460                 avail[i].base = G_EDRAM1_BASE(hi) << 20;
7461                 avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
7462                 avail[i].idx = 1;
7463                 i++;
7464         }
7465         if (lo & F_EXT_MEM_ENABLE) {
7466                 hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
7467                 avail[i].base = G_EXT_MEM_BASE(hi) << 20;
7468                 avail[i].limit = avail[i].base +
7469                     (G_EXT_MEM_SIZE(hi) << 20);
7470                 avail[i].idx = is_t5(sc) ? 3 : 2;       /* Call it MC0 for T5 */
7471                 i++;
7472         }
7473         if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
7474                 hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
7475                 avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
7476                 avail[i].limit = avail[i].base +
7477                     (G_EXT_MEM1_SIZE(hi) << 20);
7478                 avail[i].idx = 4;
7479                 i++;
7480         }
7481         if (!i)                                    /* no memory available */
7482                 return 0;
7483         qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
7484
7485         (md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
7486         (md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
7487         (md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
7488         (md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
7489         (md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
7490         (md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
7491         (md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
7492         (md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
7493         (md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
7494
7495         /* the next few have explicit upper bounds */
7496         md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
7497         md->limit = md->base - 1 +
7498                     t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
7499                     G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
7500         md++;
7501
7502         md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
7503         md->limit = md->base - 1 +
7504                     t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
7505                     G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
7506         md++;
7507
7508         if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7509                 if (chip_id(sc) <= CHELSIO_T5)
7510                         md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
7511                 else
7512                         md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
7513                 md->limit = 0;
7514         } else {
7515                 md->base = 0;
7516                 md->idx = nitems(region);  /* hide it */
7517         }
7518         md++;
7519
7520 #define ulp_region(reg) \
7521         md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
7522         (md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
7523
7524         ulp_region(RX_ISCSI);
7525         ulp_region(RX_TDDP);
7526         ulp_region(TX_TPT);
7527         ulp_region(RX_STAG);
7528         ulp_region(RX_RQ);
7529         ulp_region(RX_RQUDP);
7530         ulp_region(RX_PBL);
7531         ulp_region(TX_PBL);
7532 #undef ulp_region
7533
7534         md->base = 0;
7535         md->idx = nitems(region);
7536         if (!is_t4(sc)) {
7537                 uint32_t size = 0;
7538                 uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
7539                 uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
7540
7541                 if (is_t5(sc)) {
7542                         if (sge_ctrl & F_VFIFO_ENABLE)
7543                                 size = G_DBVFIFO_SIZE(fifo_size);
7544                 } else
7545                         size = G_T6_DBVFIFO_SIZE(fifo_size);
7546
7547                 if (size) {
7548                         md->base = G_BASEADDR(t4_read_reg(sc,
7549                             A_SGE_DBVFIFO_BADDR));
7550                         md->limit = md->base + (size << 2) - 1;
7551                 }
7552         }
7553         md++;
7554
7555         md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
7556         md->limit = 0;
7557         md++;
7558         md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
7559         md->limit = 0;
7560         md++;
7561
7562         md->base = sc->vres.ocq.start;
7563         if (sc->vres.ocq.size)
7564                 md->limit = md->base + sc->vres.ocq.size - 1;
7565         else
7566                 md->idx = nitems(region);  /* hide it */
7567         md++;
7568
7569         md->base = sc->vres.key.start;
7570         if (sc->vres.key.size)
7571                 md->limit = md->base + sc->vres.key.size - 1;
7572         else
7573                 md->idx = nitems(region);  /* hide it */
7574         md++;
7575
7576         /* add any address-space holes, there can be up to 3 */
7577         for (n = 0; n < i - 1; n++)
7578                 if (avail[n].limit < avail[n + 1].base)
7579                         (md++)->base = avail[n].limit;
7580         if (avail[n].limit)
7581                 (md++)->base = avail[n].limit;
7582
7583         n = md - mem;
7584         qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
7585
7586         for (lo = 0; lo < i; lo++)
7587                 mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
7588                                 avail[lo].limit - 1);
7589
7590         sbuf_printf(sb, "\n");
7591         for (i = 0; i < n; i++) {
7592                 if (mem[i].idx >= nitems(region))
7593                         continue;                        /* skip holes */
7594                 if (!mem[i].limit)
7595                         mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
7596                 mem_region_show(sb, region[mem[i].idx], mem[i].base,
7597                                 mem[i].limit);
7598         }
7599
7600         sbuf_printf(sb, "\n");
7601         lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
7602         hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
7603         mem_region_show(sb, "uP RAM:", lo, hi);
7604
7605         lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
7606         hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
7607         mem_region_show(sb, "uP Extmem2:", lo, hi);
7608
7609         lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
7610         sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
7611                    G_PMRXMAXPAGE(lo),
7612                    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
7613                    (lo & F_PMRXNUMCHN) ? 2 : 1);
7614
7615         lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
7616         hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
7617         sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
7618                    G_PMTXMAXPAGE(lo),
7619                    hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
7620                    hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
7621         sbuf_printf(sb, "%u p-structs\n",
7622                    t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
7623
7624         for (i = 0; i < 4; i++) {
7625                 if (chip_id(sc) > CHELSIO_T5)
7626                         lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
7627                 else
7628                         lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
7629                 if (is_t5(sc)) {
7630                         used = G_T5_USED(lo);
7631                         alloc = G_T5_ALLOC(lo);
7632                 } else {
7633                         used = G_USED(lo);
7634                         alloc = G_ALLOC(lo);
7635                 }
7636                 /* For T6 these are MAC buffer groups */
7637                 sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
7638                     i, used, alloc);
7639         }
7640         for (i = 0; i < sc->chip_params->nchan; i++) {
7641                 if (chip_id(sc) > CHELSIO_T5)
7642                         lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
7643                 else
7644                         lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
7645                 if (is_t5(sc)) {
7646                         used = G_T5_USED(lo);
7647                         alloc = G_T5_ALLOC(lo);
7648                 } else {
7649                         used = G_USED(lo);
7650                         alloc = G_ALLOC(lo);
7651                 }
7652                 /* For T6 these are MAC buffer groups */
7653                 sbuf_printf(sb,
7654                     "\nLoopback %d using %u pages out of %u allocated",
7655                     i, used, alloc);
7656         }
7657
7658         rc = sbuf_finish(sb);
7659         sbuf_delete(sb);
7660
7661         return (rc);
7662 }
7663
7664 static inline void
7665 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
7666 {
7667         *mask = x | y;
7668         y = htobe64(y);
7669         memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
7670 }
7671
7672 static int
7673 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
7674 {
7675         struct adapter *sc = arg1;
7676         struct sbuf *sb;
7677         int rc, i;
7678
7679         MPASS(chip_id(sc) <= CHELSIO_T5);
7680
7681         rc = sysctl_wire_old_buffer(req, 0);
7682         if (rc != 0)
7683                 return (rc);
7684
7685         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7686         if (sb == NULL)
7687                 return (ENOMEM);
7688
7689         sbuf_printf(sb,
7690             "Idx  Ethernet address     Mask     Vld Ports PF"
7691             "  VF              Replication             P0 P1 P2 P3  ML");
7692         for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7693                 uint64_t tcamx, tcamy, mask;
7694                 uint32_t cls_lo, cls_hi;
7695                 uint8_t addr[ETHER_ADDR_LEN];
7696
7697                 tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
7698                 tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
7699                 if (tcamx & tcamy)
7700                         continue;
7701                 tcamxy2valmask(tcamx, tcamy, addr, &mask);
7702                 cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7703                 cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7704                 sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
7705                            "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
7706                            addr[3], addr[4], addr[5], (uintmax_t)mask,
7707                            (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
7708                            G_PORTMAP(cls_hi), G_PF(cls_lo),
7709                            (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
7710
7711                 if (cls_lo & F_REPLICATE) {
7712                         struct fw_ldst_cmd ldst_cmd;
7713
7714                         memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7715                         ldst_cmd.op_to_addrspace =
7716                             htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7717                                 F_FW_CMD_REQUEST | F_FW_CMD_READ |
7718                                 V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7719                         ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7720                         ldst_cmd.u.mps.rplc.fid_idx =
7721                             htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7722                                 V_FW_LDST_CMD_IDX(i));
7723
7724                         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7725                             "t4mps");
7726                         if (rc)
7727                                 break;
7728                         rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7729                             sizeof(ldst_cmd), &ldst_cmd);
7730                         end_synchronized_op(sc, 0);
7731
7732                         if (rc != 0) {
7733                                 sbuf_printf(sb, "%36d", rc);
7734                                 rc = 0;
7735                         } else {
7736                                 sbuf_printf(sb, " %08x %08x %08x %08x",
7737                                     be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7738                                     be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7739                                     be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7740                                     be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7741                         }
7742                 } else
7743                         sbuf_printf(sb, "%36s", "");
7744
7745                 sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
7746                     G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
7747                     G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
7748         }
7749
7750         if (rc)
7751                 (void) sbuf_finish(sb);
7752         else
7753                 rc = sbuf_finish(sb);
7754         sbuf_delete(sb);
7755
7756         return (rc);
7757 }
7758
7759 static int
7760 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
7761 {
7762         struct adapter *sc = arg1;
7763         struct sbuf *sb;
7764         int rc, i;
7765
7766         MPASS(chip_id(sc) > CHELSIO_T5);
7767
7768         rc = sysctl_wire_old_buffer(req, 0);
7769         if (rc != 0)
7770                 return (rc);
7771
7772         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7773         if (sb == NULL)
7774                 return (ENOMEM);
7775
7776         sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
7777             "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
7778             "                           Replication"
7779             "                                    P0 P1 P2 P3  ML\n");
7780
7781         for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7782                 uint8_t dip_hit, vlan_vld, lookup_type, port_num;
7783                 uint16_t ivlan;
7784                 uint64_t tcamx, tcamy, val, mask;
7785                 uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
7786                 uint8_t addr[ETHER_ADDR_LEN];
7787
7788                 ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
7789                 if (i < 256)
7790                         ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
7791                 else
7792                         ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
7793                 t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7794                 val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7795                 tcamy = G_DMACH(val) << 32;
7796                 tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7797                 data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7798                 lookup_type = G_DATALKPTYPE(data2);
7799                 port_num = G_DATAPORTNUM(data2);
7800                 if (lookup_type && lookup_type != M_DATALKPTYPE) {
7801                         /* Inner header VNI */
7802                         vniy = ((data2 & F_DATAVIDH2) << 23) |
7803                                        (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7804                         dip_hit = data2 & F_DATADIPHIT;
7805                         vlan_vld = 0;
7806                 } else {
7807                         vniy = 0;
7808                         dip_hit = 0;
7809                         vlan_vld = data2 & F_DATAVIDH2;
7810                         ivlan = G_VIDL(val);
7811                 }
7812
7813                 ctl |= V_CTLXYBITSEL(1);
7814                 t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7815                 val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7816                 tcamx = G_DMACH(val) << 32;
7817                 tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7818                 data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7819                 if (lookup_type && lookup_type != M_DATALKPTYPE) {
7820                         /* Inner header VNI mask */
7821                         vnix = ((data2 & F_DATAVIDH2) << 23) |
7822                                (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7823                 } else
7824                         vnix = 0;
7825
7826                 if (tcamx & tcamy)
7827                         continue;
7828                 tcamxy2valmask(tcamx, tcamy, addr, &mask);
7829
7830                 cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7831                 cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7832
7833                 if (lookup_type && lookup_type != M_DATALKPTYPE) {
7834                         sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7835                             "%012jx %06x %06x    -    -   %3c"
7836                             "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
7837                             addr[1], addr[2], addr[3], addr[4], addr[5],
7838                             (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
7839                             port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7840                             G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7841                             cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7842                 } else {
7843                         sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7844                             "%012jx    -       -   ", i, addr[0], addr[1],
7845                             addr[2], addr[3], addr[4], addr[5],
7846                             (uintmax_t)mask);
7847
7848                         if (vlan_vld)
7849                                 sbuf_printf(sb, "%4u   Y     ", ivlan);
7850                         else
7851                                 sbuf_printf(sb, "  -    N     ");
7852
7853                         sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
7854                             lookup_type ? 'I' : 'O', port_num,
7855                             cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7856                             G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7857                             cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7858                 }
7859
7860
7861                 if (cls_lo & F_T6_REPLICATE) {
7862                         struct fw_ldst_cmd ldst_cmd;
7863
7864                         memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7865                         ldst_cmd.op_to_addrspace =
7866                             htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7867                                 F_FW_CMD_REQUEST | F_FW_CMD_READ |
7868                                 V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7869                         ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7870                         ldst_cmd.u.mps.rplc.fid_idx =
7871                             htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7872                                 V_FW_LDST_CMD_IDX(i));
7873
7874                         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7875                             "t6mps");
7876                         if (rc)
7877                                 break;
7878                         rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7879                             sizeof(ldst_cmd), &ldst_cmd);
7880                         end_synchronized_op(sc, 0);
7881
7882                         if (rc != 0) {
7883                                 sbuf_printf(sb, "%72d", rc);
7884                                 rc = 0;
7885                         } else {
7886                                 sbuf_printf(sb, " %08x %08x %08x %08x"
7887                                     " %08x %08x %08x %08x",
7888                                     be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
7889                                     be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
7890                                     be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
7891                                     be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
7892                                     be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7893                                     be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7894                                     be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7895                                     be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7896                         }
7897                 } else
7898                         sbuf_printf(sb, "%72s", "");
7899
7900                 sbuf_printf(sb, "%4u%3u%3u%3u %#x",
7901                     G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
7902                     G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
7903                     (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
7904         }
7905
7906         if (rc)
7907                 (void) sbuf_finish(sb);
7908         else
7909                 rc = sbuf_finish(sb);
7910         sbuf_delete(sb);
7911
7912         return (rc);
7913 }
7914
7915 static int
7916 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
7917 {
7918         struct adapter *sc = arg1;
7919         struct sbuf *sb;
7920         int rc;
7921         uint16_t mtus[NMTUS];
7922
7923         rc = sysctl_wire_old_buffer(req, 0);
7924         if (rc != 0)
7925                 return (rc);
7926
7927         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7928         if (sb == NULL)
7929                 return (ENOMEM);
7930
7931         t4_read_mtu_tbl(sc, mtus, NULL);
7932
7933         sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
7934             mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
7935             mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
7936             mtus[14], mtus[15]);
7937
7938         rc = sbuf_finish(sb);
7939         sbuf_delete(sb);
7940
7941         return (rc);
7942 }
7943
7944 static int
7945 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
7946 {
7947         struct adapter *sc = arg1;
7948         struct sbuf *sb;
7949         int rc, i;
7950         uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
7951         uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
7952         static const char *tx_stats[MAX_PM_NSTATS] = {
7953                 "Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
7954                 "Tx FIFO wait", NULL, "Tx latency"
7955         };
7956         static const char *rx_stats[MAX_PM_NSTATS] = {
7957                 "Read:", "Write bypass:", "Write mem:", "Flush:",
7958                 "Rx FIFO wait", NULL, "Rx latency"
7959         };
7960
7961         rc = sysctl_wire_old_buffer(req, 0);
7962         if (rc != 0)
7963                 return (rc);
7964
7965         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7966         if (sb == NULL)
7967                 return (ENOMEM);
7968
7969         t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
7970         t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
7971
7972         sbuf_printf(sb, "                Tx pcmds             Tx bytes");
7973         for (i = 0; i < 4; i++) {
7974                 sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7975                     tx_cyc[i]);
7976         }
7977
7978         sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
7979         for (i = 0; i < 4; i++) {
7980                 sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7981                     rx_cyc[i]);
7982         }
7983
7984         if (chip_id(sc) > CHELSIO_T5) {
7985                 sbuf_printf(sb,
7986                     "\n              Total wait      Total occupancy");
7987                 sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7988                     tx_cyc[i]);
7989                 sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7990                     rx_cyc[i]);
7991
7992                 i += 2;
7993                 MPASS(i < nitems(tx_stats));
7994
7995                 sbuf_printf(sb,
7996                     "\n                   Reads           Total wait");
7997                 sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7998                     tx_cyc[i]);
7999                 sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8000                     rx_cyc[i]);
8001         }
8002
8003         rc = sbuf_finish(sb);
8004         sbuf_delete(sb);
8005
8006         return (rc);
8007 }
8008
8009 static int
8010 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
8011 {
8012         struct adapter *sc = arg1;
8013         struct sbuf *sb;
8014         int rc;
8015         struct tp_rdma_stats stats;
8016
8017         rc = sysctl_wire_old_buffer(req, 0);
8018         if (rc != 0)
8019                 return (rc);
8020
8021         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8022         if (sb == NULL)
8023                 return (ENOMEM);
8024
8025         mtx_lock(&sc->reg_lock);
8026         t4_tp_get_rdma_stats(sc, &stats, 0);
8027         mtx_unlock(&sc->reg_lock);
8028
8029         sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
8030         sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
8031
8032         rc = sbuf_finish(sb);
8033         sbuf_delete(sb);
8034
8035         return (rc);
8036 }
8037
8038 static int
8039 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
8040 {
8041         struct adapter *sc = arg1;
8042         struct sbuf *sb;
8043         int rc;
8044         struct tp_tcp_stats v4, v6;
8045
8046         rc = sysctl_wire_old_buffer(req, 0);
8047         if (rc != 0)
8048                 return (rc);
8049
8050         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8051         if (sb == NULL)
8052                 return (ENOMEM);
8053
8054         mtx_lock(&sc->reg_lock);
8055         t4_tp_get_tcp_stats(sc, &v4, &v6, 0);
8056         mtx_unlock(&sc->reg_lock);
8057
8058         sbuf_printf(sb,
8059             "                                IP                 IPv6\n");
8060         sbuf_printf(sb, "OutRsts:      %20u %20u\n",
8061             v4.tcp_out_rsts, v6.tcp_out_rsts);
8062         sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
8063             v4.tcp_in_segs, v6.tcp_in_segs);
8064         sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
8065             v4.tcp_out_segs, v6.tcp_out_segs);
8066         sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
8067             v4.tcp_retrans_segs, v6.tcp_retrans_segs);
8068
8069         rc = sbuf_finish(sb);
8070         sbuf_delete(sb);
8071
8072         return (rc);
8073 }
8074
8075 static int
8076 sysctl_tids(SYSCTL_HANDLER_ARGS)
8077 {
8078         struct adapter *sc = arg1;
8079         struct sbuf *sb;
8080         int rc;
8081         struct tid_info *t = &sc->tids;
8082
8083         rc = sysctl_wire_old_buffer(req, 0);
8084         if (rc != 0)
8085                 return (rc);
8086
8087         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8088         if (sb == NULL)
8089                 return (ENOMEM);
8090
8091         if (t->natids) {
8092                 sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
8093                     t->atids_in_use);
8094         }
8095
8096         if (t->ntids) {
8097                 sbuf_printf(sb, "TID range: ");
8098                 if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
8099                         uint32_t b, hb;
8100
8101                         if (chip_id(sc) <= CHELSIO_T5) {
8102                                 b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
8103                                 hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
8104                         } else {
8105                                 b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
8106                                 hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
8107                         }
8108
8109                         if (b)
8110                                 sbuf_printf(sb, "0-%u, ", b - 1);
8111                         sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
8112                 } else
8113                         sbuf_printf(sb, "0-%u", t->ntids - 1);
8114                 sbuf_printf(sb, ", in use: %u\n",
8115                     atomic_load_acq_int(&t->tids_in_use));
8116         }
8117
8118         if (t->nstids) {
8119                 sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
8120                     t->stid_base + t->nstids - 1, t->stids_in_use);
8121         }
8122
8123         if (t->nftids) {
8124                 sbuf_printf(sb, "FTID range: %u-%u\n", t->ftid_base,
8125                     t->ftid_base + t->nftids - 1);
8126         }
8127
8128         if (t->netids) {
8129                 sbuf_printf(sb, "ETID range: %u-%u, in use: %u\n", t->etid_base,
8130                     t->etid_base + t->netids - 1, t->etids_in_use);
8131         }
8132
8133         sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
8134             t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
8135             t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
8136
8137         rc = sbuf_finish(sb);
8138         sbuf_delete(sb);
8139
8140         return (rc);
8141 }
8142
8143 static int
8144 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
8145 {
8146         struct adapter *sc = arg1;
8147         struct sbuf *sb;
8148         int rc;
8149         struct tp_err_stats stats;
8150
8151         rc = sysctl_wire_old_buffer(req, 0);
8152         if (rc != 0)
8153                 return (rc);
8154
8155         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8156         if (sb == NULL)
8157                 return (ENOMEM);
8158
8159         mtx_lock(&sc->reg_lock);
8160         t4_tp_get_err_stats(sc, &stats, 0);
8161         mtx_unlock(&sc->reg_lock);
8162
8163         if (sc->chip_params->nchan > 2) {
8164                 sbuf_printf(sb, "                 channel 0  channel 1"
8165                     "  channel 2  channel 3\n");
8166                 sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
8167                     stats.mac_in_errs[0], stats.mac_in_errs[1],
8168                     stats.mac_in_errs[2], stats.mac_in_errs[3]);
8169                 sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
8170                     stats.hdr_in_errs[0], stats.hdr_in_errs[1],
8171                     stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
8172                 sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
8173                     stats.tcp_in_errs[0], stats.tcp_in_errs[1],
8174                     stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
8175                 sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
8176                     stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
8177                     stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
8178                 sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
8179                     stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
8180                     stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
8181                 sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
8182                     stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
8183                     stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
8184                 sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
8185                     stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
8186                     stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
8187                 sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
8188                     stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
8189                     stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
8190         } else {
8191                 sbuf_printf(sb, "                 channel 0  channel 1\n");
8192                 sbuf_printf(sb, "macInErrs:      %10u %10u\n",
8193                     stats.mac_in_errs[0], stats.mac_in_errs[1]);
8194                 sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
8195                     stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
8196                 sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
8197                     stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
8198                 sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
8199                     stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
8200                 sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
8201                     stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
8202                 sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
8203                     stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
8204                 sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
8205                     stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
8206                 sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
8207                     stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
8208         }
8209
8210         sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
8211             stats.ofld_no_neigh, stats.ofld_cong_defer);
8212
8213         rc = sbuf_finish(sb);
8214         sbuf_delete(sb);
8215
8216         return (rc);
8217 }
8218
8219 static int
8220 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
8221 {
8222         struct adapter *sc = arg1;
8223         struct tp_params *tpp = &sc->params.tp;
8224         u_int mask;
8225         int rc;
8226
8227         mask = tpp->la_mask >> 16;
8228         rc = sysctl_handle_int(oidp, &mask, 0, req);
8229         if (rc != 0 || req->newptr == NULL)
8230                 return (rc);
8231         if (mask > 0xffff)
8232                 return (EINVAL);
8233         tpp->la_mask = mask << 16;
8234         t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
8235
8236         return (0);
8237 }
8238
8239 struct field_desc {
8240         const char *name;
8241         u_int start;
8242         u_int width;
8243 };
8244
8245 static void
8246 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
8247 {
8248         char buf[32];
8249         int line_size = 0;
8250
8251         while (f->name) {
8252                 uint64_t mask = (1ULL << f->width) - 1;
8253                 int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
8254                     ((uintmax_t)v >> f->start) & mask);
8255
8256                 if (line_size + len >= 79) {
8257                         line_size = 8;
8258                         sbuf_printf(sb, "\n        ");
8259                 }
8260                 sbuf_printf(sb, "%s ", buf);
8261                 line_size += len + 1;
8262                 f++;
8263         }
8264         sbuf_printf(sb, "\n");
8265 }
8266
8267 static const struct field_desc tp_la0[] = {
8268         { "RcfOpCodeOut", 60, 4 },
8269         { "State", 56, 4 },
8270         { "WcfState", 52, 4 },
8271         { "RcfOpcSrcOut", 50, 2 },
8272         { "CRxError", 49, 1 },
8273         { "ERxError", 48, 1 },
8274         { "SanityFailed", 47, 1 },
8275         { "SpuriousMsg", 46, 1 },
8276         { "FlushInputMsg", 45, 1 },
8277         { "FlushInputCpl", 44, 1 },
8278         { "RssUpBit", 43, 1 },
8279         { "RssFilterHit", 42, 1 },
8280         { "Tid", 32, 10 },
8281         { "InitTcb", 31, 1 },
8282         { "LineNumber", 24, 7 },
8283         { "Emsg", 23, 1 },
8284         { "EdataOut", 22, 1 },
8285         { "Cmsg", 21, 1 },
8286         { "CdataOut", 20, 1 },
8287         { "EreadPdu", 19, 1 },
8288         { "CreadPdu", 18, 1 },
8289         { "TunnelPkt", 17, 1 },
8290         { "RcfPeerFin", 16, 1 },
8291         { "RcfReasonOut", 12, 4 },
8292         { "TxCchannel", 10, 2 },
8293         { "RcfTxChannel", 8, 2 },
8294         { "RxEchannel", 6, 2 },
8295         { "RcfRxChannel", 5, 1 },
8296         { "RcfDataOutSrdy", 4, 1 },
8297         { "RxDvld", 3, 1 },
8298         { "RxOoDvld", 2, 1 },
8299         { "RxCongestion", 1, 1 },
8300         { "TxCongestion", 0, 1 },
8301         { NULL }
8302 };
8303
8304 static const struct field_desc tp_la1[] = {
8305         { "CplCmdIn", 56, 8 },
8306         { "CplCmdOut", 48, 8 },
8307         { "ESynOut", 47, 1 },
8308         { "EAckOut", 46, 1 },
8309         { "EFinOut", 45, 1 },
8310         { "ERstOut", 44, 1 },
8311         { "SynIn", 43, 1 },
8312         { "AckIn", 42, 1 },
8313         { "FinIn", 41, 1 },
8314         { "RstIn", 40, 1 },
8315         { "DataIn", 39, 1 },
8316         { "DataInVld", 38, 1 },
8317         { "PadIn", 37, 1 },
8318         { "RxBufEmpty", 36, 1 },
8319         { "RxDdp", 35, 1 },
8320         { "RxFbCongestion", 34, 1 },
8321         { "TxFbCongestion", 33, 1 },
8322         { "TxPktSumSrdy", 32, 1 },
8323         { "RcfUlpType", 28, 4 },
8324         { "Eread", 27, 1 },
8325         { "Ebypass", 26, 1 },
8326         { "Esave", 25, 1 },
8327         { "Static0", 24, 1 },
8328         { "Cread", 23, 1 },
8329         { "Cbypass", 22, 1 },
8330         { "Csave", 21, 1 },
8331         { "CPktOut", 20, 1 },
8332         { "RxPagePoolFull", 18, 2 },
8333         { "RxLpbkPkt", 17, 1 },
8334         { "TxLpbkPkt", 16, 1 },
8335         { "RxVfValid", 15, 1 },
8336         { "SynLearned", 14, 1 },
8337         { "SetDelEntry", 13, 1 },
8338         { "SetInvEntry", 12, 1 },
8339         { "CpcmdDvld", 11, 1 },
8340         { "CpcmdSave", 10, 1 },
8341         { "RxPstructsFull", 8, 2 },
8342         { "EpcmdDvld", 7, 1 },
8343         { "EpcmdFlush", 6, 1 },
8344         { "EpcmdTrimPrefix", 5, 1 },
8345         { "EpcmdTrimPostfix", 4, 1 },
8346         { "ERssIp4Pkt", 3, 1 },
8347         { "ERssIp6Pkt", 2, 1 },
8348         { "ERssTcpUdpPkt", 1, 1 },
8349         { "ERssFceFipPkt", 0, 1 },
8350         { NULL }
8351 };
8352
8353 static const struct field_desc tp_la2[] = {
8354         { "CplCmdIn", 56, 8 },
8355         { "MpsVfVld", 55, 1 },
8356         { "MpsPf", 52, 3 },
8357         { "MpsVf", 44, 8 },
8358         { "SynIn", 43, 1 },
8359         { "AckIn", 42, 1 },
8360         { "FinIn", 41, 1 },
8361         { "RstIn", 40, 1 },
8362         { "DataIn", 39, 1 },
8363         { "DataInVld", 38, 1 },
8364         { "PadIn", 37, 1 },
8365         { "RxBufEmpty", 36, 1 },
8366         { "RxDdp", 35, 1 },
8367         { "RxFbCongestion", 34, 1 },
8368         { "TxFbCongestion", 33, 1 },
8369         { "TxPktSumSrdy", 32, 1 },
8370         { "RcfUlpType", 28, 4 },
8371         { "Eread", 27, 1 },
8372         { "Ebypass", 26, 1 },
8373         { "Esave", 25, 1 },
8374         { "Static0", 24, 1 },
8375         { "Cread", 23, 1 },
8376         { "Cbypass", 22, 1 },
8377         { "Csave", 21, 1 },
8378         { "CPktOut", 20, 1 },
8379         { "RxPagePoolFull", 18, 2 },
8380         { "RxLpbkPkt", 17, 1 },
8381         { "TxLpbkPkt", 16, 1 },
8382         { "RxVfValid", 15, 1 },
8383         { "SynLearned", 14, 1 },
8384         { "SetDelEntry", 13, 1 },
8385         { "SetInvEntry", 12, 1 },
8386         { "CpcmdDvld", 11, 1 },
8387         { "CpcmdSave", 10, 1 },
8388         { "RxPstructsFull", 8, 2 },
8389         { "EpcmdDvld", 7, 1 },
8390         { "EpcmdFlush", 6, 1 },
8391         { "EpcmdTrimPrefix", 5, 1 },
8392         { "EpcmdTrimPostfix", 4, 1 },
8393         { "ERssIp4Pkt", 3, 1 },
8394         { "ERssIp6Pkt", 2, 1 },
8395         { "ERssTcpUdpPkt", 1, 1 },
8396         { "ERssFceFipPkt", 0, 1 },
8397         { NULL }
8398 };
8399
8400 static void
8401 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
8402 {
8403
8404         field_desc_show(sb, *p, tp_la0);
8405 }
8406
8407 static void
8408 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
8409 {
8410
8411         if (idx)
8412                 sbuf_printf(sb, "\n");
8413         field_desc_show(sb, p[0], tp_la0);
8414         if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8415                 field_desc_show(sb, p[1], tp_la0);
8416 }
8417
8418 static void
8419 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
8420 {
8421
8422         if (idx)
8423                 sbuf_printf(sb, "\n");
8424         field_desc_show(sb, p[0], tp_la0);
8425         if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8426                 field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
8427 }
8428
8429 static int
8430 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
8431 {
8432         struct adapter *sc = arg1;
8433         struct sbuf *sb;
8434         uint64_t *buf, *p;
8435         int rc;
8436         u_int i, inc;
8437         void (*show_func)(struct sbuf *, uint64_t *, int);
8438
8439         rc = sysctl_wire_old_buffer(req, 0);
8440         if (rc != 0)
8441                 return (rc);
8442
8443         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8444         if (sb == NULL)
8445                 return (ENOMEM);
8446
8447         buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
8448
8449         t4_tp_read_la(sc, buf, NULL);
8450         p = buf;
8451
8452         switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
8453         case 2:
8454                 inc = 2;
8455                 show_func = tp_la_show2;
8456                 break;
8457         case 3:
8458                 inc = 2;
8459                 show_func = tp_la_show3;
8460                 break;
8461         default:
8462                 inc = 1;
8463                 show_func = tp_la_show;
8464         }
8465
8466         for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
8467                 (*show_func)(sb, p, i);
8468
8469         rc = sbuf_finish(sb);
8470         sbuf_delete(sb);
8471         free(buf, M_CXGBE);
8472         return (rc);
8473 }
8474
8475 static int
8476 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
8477 {
8478         struct adapter *sc = arg1;
8479         struct sbuf *sb;
8480         int rc;
8481         u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
8482
8483         rc = sysctl_wire_old_buffer(req, 0);
8484         if (rc != 0)
8485                 return (rc);
8486
8487         sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8488         if (sb == NULL)
8489                 return (ENOMEM);
8490
8491         t4_get_chan_txrate(sc, nrate, orate);
8492
8493         if (sc->chip_params->nchan > 2) {
8494                 sbuf_printf(sb, "              channel 0   channel 1"
8495                     "   channel 2   channel 3\n");
8496                 sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
8497                     nrate[0], nrate[1], nrate[2], nrate[3]);
8498                 sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
8499                     orate[0], orate[1], orate[2], orate[3]);
8500         } else {
8501                 sbuf_printf(sb, "              channel 0   channel 1\n");
8502                 sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
8503                     nrate[0], nrate[1]);
8504                 sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
8505                     orate[0], orate[1]);
8506         }
8507
8508         rc = sbuf_finish(sb);
8509         sbuf_delete(sb);
8510
8511         return (rc);
8512 }
8513
8514 static int
8515 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
8516 {
8517         struct adapter *sc = arg1;
8518         struct sbuf *sb;
8519         uint32_t *buf, *p;
8520         int rc, i;
8521
8522         rc = sysctl_wire_old_buffer(req, 0);
8523         if (rc != 0)
8524                 return (rc);
8525
8526         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8527         if (sb == NULL)
8528                 return (ENOMEM);
8529
8530         buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
8531             M_ZERO | M_WAITOK);
8532
8533         t4_ulprx_read_la(sc, buf);
8534         p = buf;
8535
8536         sbuf_printf(sb, "      Pcmd        Type   Message"
8537             "                Data");
8538         for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
8539                 sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
8540                     p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
8541         }
8542
8543         rc = sbuf_finish(sb);
8544         sbuf_delete(sb);
8545         free(buf, M_CXGBE);
8546         return (rc);
8547 }
8548
8549 static int
8550 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
8551 {
8552         struct adapter *sc = arg1;
8553         struct sbuf *sb;
8554         int rc, v;
8555
8556         MPASS(chip_id(sc) >= CHELSIO_T5);
8557
8558         rc = sysctl_wire_old_buffer(req, 0);
8559         if (rc != 0)
8560                 return (rc);
8561
8562         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8563         if (sb == NULL)
8564                 return (ENOMEM);
8565
8566         v = t4_read_reg(sc, A_SGE_STAT_CFG);
8567         if (G_STATSOURCE_T5(v) == 7) {
8568                 int mode;
8569
8570                 mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
8571                 if (mode == 0) {
8572                         sbuf_printf(sb, "total %d, incomplete %d",
8573                             t4_read_reg(sc, A_SGE_STAT_TOTAL),
8574                             t4_read_reg(sc, A_SGE_STAT_MATCH));
8575                 } else if (mode == 1) {
8576                         sbuf_printf(sb, "total %d, data overflow %d",
8577                             t4_read_reg(sc, A_SGE_STAT_TOTAL),
8578                             t4_read_reg(sc, A_SGE_STAT_MATCH));
8579                 } else {
8580                         sbuf_printf(sb, "unknown mode %d", mode);
8581                 }
8582         }
8583         rc = sbuf_finish(sb);
8584         sbuf_delete(sb);
8585
8586         return (rc);
8587 }
8588
8589 static int
8590 sysctl_tc_params(SYSCTL_HANDLER_ARGS)
8591 {
8592         struct adapter *sc = arg1;
8593         struct tx_cl_rl_params tc;
8594         struct sbuf *sb;
8595         int i, rc, port_id, mbps, gbps;
8596
8597         rc = sysctl_wire_old_buffer(req, 0);
8598         if (rc != 0)
8599                 return (rc);
8600
8601         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8602         if (sb == NULL)
8603                 return (ENOMEM);
8604
8605         port_id = arg2 >> 16;
8606         MPASS(port_id < sc->params.nports);
8607         MPASS(sc->port[port_id] != NULL);
8608         i = arg2 & 0xffff;
8609         MPASS(i < sc->chip_params->nsched_cls);
8610
8611         mtx_lock(&sc->tc_lock);
8612         tc = sc->port[port_id]->sched_params->cl_rl[i];
8613         mtx_unlock(&sc->tc_lock);
8614
8615         switch (tc.rateunit) {
8616         case SCHED_CLASS_RATEUNIT_BITS:
8617                 switch (tc.ratemode) {
8618                 case SCHED_CLASS_RATEMODE_REL:
8619                         /* XXX: top speed or actual link speed? */
8620                         gbps = port_top_speed(sc->port[port_id]);
8621                         sbuf_printf(sb, "%u%% of %uGbps", tc.maxrate, gbps);
8622                         break;
8623                 case SCHED_CLASS_RATEMODE_ABS:
8624                         mbps = tc.maxrate / 1000;
8625                         gbps = tc.maxrate / 1000000;
8626                         if (tc.maxrate == gbps * 1000000)
8627                                 sbuf_printf(sb, "%uGbps", gbps);
8628                         else if (tc.maxrate == mbps * 1000)
8629                                 sbuf_printf(sb, "%uMbps", mbps);
8630                         else
8631                                 sbuf_printf(sb, "%uKbps", tc.maxrate);
8632                         break;
8633                 default:
8634                         rc = ENXIO;
8635                         goto done;
8636                 }
8637                 break;
8638         case SCHED_CLASS_RATEUNIT_PKTS:
8639                 sbuf_printf(sb, "%upps", tc.maxrate);
8640                 break;
8641         default:
8642                 rc = ENXIO;
8643                 goto done;
8644         }
8645
8646         switch (tc.mode) {
8647         case SCHED_CLASS_MODE_CLASS:
8648                 sbuf_printf(sb, " aggregate");
8649                 break;
8650         case SCHED_CLASS_MODE_FLOW:
8651                 sbuf_printf(sb, " per-flow");
8652                 break;
8653         default:
8654                 rc = ENXIO;
8655                 goto done;
8656         }
8657
8658 done:
8659         if (rc == 0)
8660                 rc = sbuf_finish(sb);
8661         sbuf_delete(sb);
8662
8663         return (rc);
8664 }
8665
8666 static int
8667 sysctl_cpus(SYSCTL_HANDLER_ARGS)
8668 {
8669         struct adapter *sc = arg1;
8670         enum cpu_sets op = arg2;
8671         cpuset_t cpuset;
8672         struct sbuf *sb;
8673         int i, rc;
8674
8675         MPASS(op == LOCAL_CPUS || op == INTR_CPUS);
8676
8677         CPU_ZERO(&cpuset);
8678         rc = bus_get_cpus(sc->dev, op, sizeof(cpuset), &cpuset);
8679         if (rc != 0)
8680                 return (rc);
8681
8682         rc = sysctl_wire_old_buffer(req, 0);
8683         if (rc != 0)
8684                 return (rc);
8685
8686         sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8687         if (sb == NULL)
8688                 return (ENOMEM);
8689
8690         CPU_FOREACH(i)
8691                 sbuf_printf(sb, "%d ", i);
8692         rc = sbuf_finish(sb);
8693         sbuf_delete(sb);
8694
8695         return (rc);
8696 }
8697
8698 #ifdef TCP_OFFLOAD
8699 static int
8700 sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS)
8701 {
8702         struct adapter *sc = arg1;
8703         int *old_ports, *new_ports;
8704         int i, new_count, rc;
8705
8706         if (req->newptr == NULL && req->oldptr == NULL)
8707                 return (SYSCTL_OUT(req, NULL, imax(sc->tt.num_tls_rx_ports, 1) *
8708                     sizeof(sc->tt.tls_rx_ports[0])));
8709
8710         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tlsrx");
8711         if (rc)
8712                 return (rc);
8713
8714         if (sc->tt.num_tls_rx_ports == 0) {
8715                 i = -1;
8716                 rc = SYSCTL_OUT(req, &i, sizeof(i));
8717         } else
8718                 rc = SYSCTL_OUT(req, sc->tt.tls_rx_ports,
8719                     sc->tt.num_tls_rx_ports * sizeof(sc->tt.tls_rx_ports[0]));
8720         if (rc == 0 && req->newptr != NULL) {
8721                 new_count = req->newlen / sizeof(new_ports[0]);
8722                 new_ports = malloc(new_count * sizeof(new_ports[0]), M_CXGBE,
8723                     M_WAITOK);
8724                 rc = SYSCTL_IN(req, new_ports, new_count *
8725                     sizeof(new_ports[0]));
8726                 if (rc)
8727                         goto err;
8728
8729                 /* Allow setting to a single '-1' to clear the list. */
8730                 if (new_count == 1 && new_ports[0] == -1) {
8731                         ADAPTER_LOCK(sc);
8732                         old_ports = sc->tt.tls_rx_ports;
8733                         sc->tt.tls_rx_ports = NULL;
8734                         sc->tt.num_tls_rx_ports = 0;
8735                         ADAPTER_UNLOCK(sc);
8736                         free(old_ports, M_CXGBE);
8737                 } else {
8738                         for (i = 0; i < new_count; i++) {
8739                                 if (new_ports[i] < 1 ||
8740                                     new_ports[i] > IPPORT_MAX) {
8741                                         rc = EINVAL;
8742                                         goto err;
8743                                 }
8744                         }
8745
8746                         ADAPTER_LOCK(sc);
8747                         old_ports = sc->tt.tls_rx_ports;
8748                         sc->tt.tls_rx_ports = new_ports;
8749                         sc->tt.num_tls_rx_ports = new_count;
8750                         ADAPTER_UNLOCK(sc);
8751                         free(old_ports, M_CXGBE);
8752                         new_ports = NULL;
8753                 }
8754         err:
8755                 free(new_ports, M_CXGBE);
8756         }
8757         end_synchronized_op(sc, 0);
8758         return (rc);
8759 }
8760
8761 static void
8762 unit_conv(char *buf, size_t len, u_int val, u_int factor)
8763 {
8764         u_int rem = val % factor;
8765
8766         if (rem == 0)
8767                 snprintf(buf, len, "%u", val / factor);
8768         else {
8769                 while (rem % 10 == 0)
8770                         rem /= 10;
8771                 snprintf(buf, len, "%u.%u", val / factor, rem);
8772         }
8773 }
8774
8775 static int
8776 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
8777 {
8778         struct adapter *sc = arg1;
8779         char buf[16];
8780         u_int res, re;
8781         u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8782
8783         res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8784         switch (arg2) {
8785         case 0:
8786                 /* timer_tick */
8787                 re = G_TIMERRESOLUTION(res);
8788                 break;
8789         case 1:
8790                 /* TCP timestamp tick */
8791                 re = G_TIMESTAMPRESOLUTION(res);
8792                 break;
8793         case 2:
8794                 /* DACK tick */
8795                 re = G_DELAYEDACKRESOLUTION(res);
8796                 break;
8797         default:
8798                 return (EDOOFUS);
8799         }
8800
8801         unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
8802
8803         return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
8804 }
8805
8806 static int
8807 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
8808 {
8809         struct adapter *sc = arg1;
8810         u_int res, dack_re, v;
8811         u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8812
8813         res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8814         dack_re = G_DELAYEDACKRESOLUTION(res);
8815         v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
8816
8817         return (sysctl_handle_int(oidp, &v, 0, req));
8818 }
8819
8820 static int
8821 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
8822 {
8823         struct adapter *sc = arg1;
8824         int reg = arg2;
8825         u_int tre;
8826         u_long tp_tick_us, v;
8827         u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8828
8829         MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
8830             reg == A_TP_PERS_MIN  || reg == A_TP_PERS_MAX ||
8831             reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
8832             reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
8833
8834         tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
8835         tp_tick_us = (cclk_ps << tre) / 1000000;
8836
8837         if (reg == A_TP_INIT_SRTT)
8838                 v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
8839         else
8840                 v = tp_tick_us * t4_read_reg(sc, reg);
8841
8842         return (sysctl_handle_long(oidp, &v, 0, req));
8843 }
8844
8845 /*
8846  * All fields in TP_SHIFT_CNT are 4b and the starting location of the field is
8847  * passed to this function.
8848  */
8849 static int
8850 sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS)
8851 {
8852         struct adapter *sc = arg1;
8853         int idx = arg2;
8854         u_int v;
8855
8856         MPASS(idx >= 0 && idx <= 24);
8857
8858         v = (t4_read_reg(sc, A_TP_SHIFT_CNT) >> idx) & 0xf;
8859
8860         return (sysctl_handle_int(oidp, &v, 0, req));
8861 }
8862
8863 static int
8864 sysctl_tp_backoff(SYSCTL_HANDLER_ARGS)
8865 {
8866         struct adapter *sc = arg1;
8867         int idx = arg2;
8868         u_int shift, v, r;
8869
8870         MPASS(idx >= 0 && idx < 16);
8871
8872         r = A_TP_TCP_BACKOFF_REG0 + (idx & ~3);
8873         shift = (idx & 3) << 3;
8874         v = (t4_read_reg(sc, r) >> shift) & M_TIMERBACKOFFINDEX0;
8875
8876         return (sysctl_handle_int(oidp, &v, 0, req));
8877 }
8878
8879 static int
8880 sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS)
8881 {
8882         struct vi_info *vi = arg1;
8883         struct adapter *sc = vi->pi->adapter;
8884         int idx, rc, i;
8885         struct sge_ofld_rxq *ofld_rxq;
8886         uint8_t v;
8887
8888         idx = vi->ofld_tmr_idx;
8889
8890         rc = sysctl_handle_int(oidp, &idx, 0, req);
8891         if (rc != 0 || req->newptr == NULL)
8892                 return (rc);
8893
8894         if (idx < 0 || idx >= SGE_NTIMERS)
8895                 return (EINVAL);
8896
8897         rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8898             "t4otmr");
8899         if (rc)
8900                 return (rc);
8901
8902         v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->ofld_pktc_idx != -1);
8903         for_each_ofld_rxq(vi, i, ofld_rxq) {
8904 #ifdef atomic_store_rel_8
8905                 atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
8906 #else
8907                 ofld_rxq->iq.intr_params = v;
8908 #endif
8909         }
8910         vi->ofld_tmr_idx = idx;
8911
8912         end_synchronized_op(sc, LOCK_HELD);
8913         return (0);
8914 }
8915
8916 static int
8917 sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS)
8918 {
8919         struct vi_info *vi = arg1;
8920         struct adapter *sc = vi->pi->adapter;
8921         int idx, rc;
8922
8923         idx = vi->ofld_pktc_idx;
8924
8925         rc = sysctl_handle_int(oidp, &idx, 0, req);
8926         if (rc != 0 || req->newptr == NULL)
8927                 return (rc);
8928
8929         if (idx < -1 || idx >= SGE_NCOUNTERS)
8930                 return (EINVAL);
8931
8932         rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8933             "t4opktc");
8934         if (rc)
8935                 return (rc);
8936
8937         if (vi->flags & VI_INIT_DONE)
8938                 rc = EBUSY; /* cannot be changed once the queues are created */
8939         else
8940                 vi->ofld_pktc_idx = idx;
8941
8942         end_synchronized_op(sc, LOCK_HELD);
8943         return (rc);
8944 }
8945 #endif
8946
8947 static int
8948 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
8949 {
8950         int rc;
8951
8952         if (cntxt->cid > M_CTXTQID)
8953                 return (EINVAL);
8954
8955         if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
8956             cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
8957                 return (EINVAL);
8958
8959         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
8960         if (rc)
8961                 return (rc);
8962
8963         if (sc->flags & FW_OK) {
8964                 rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
8965                     &cntxt->data[0]);
8966                 if (rc == 0)
8967                         goto done;
8968         }
8969
8970         /*
8971          * Read via firmware failed or wasn't even attempted.  Read directly via
8972          * the backdoor.
8973          */
8974         rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
8975 done:
8976         end_synchronized_op(sc, 0);
8977         return (rc);
8978 }
8979
8980 static int
8981 load_fw(struct adapter *sc, struct t4_data *fw)
8982 {
8983         int rc;
8984         uint8_t *fw_data;
8985
8986         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
8987         if (rc)
8988                 return (rc);
8989
8990         /*
8991          * The firmware, with the sole exception of the memory parity error
8992          * handler, runs from memory and not flash.  It is almost always safe to
8993          * install a new firmware on a running system.  Just set bit 1 in
8994          * hw.cxgbe.dflags or dev.<nexus>.<n>.dflags first.
8995          */
8996         if (sc->flags & FULL_INIT_DONE &&
8997             (sc->debug_flags & DF_LOAD_FW_ANYTIME) == 0) {
8998                 rc = EBUSY;
8999                 goto done;
9000         }
9001
9002         fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
9003         if (fw_data == NULL) {
9004                 rc = ENOMEM;
9005                 goto done;
9006         }
9007
9008         rc = copyin(fw->data, fw_data, fw->len);
9009         if (rc == 0)
9010                 rc = -t4_load_fw(sc, fw_data, fw->len);
9011
9012         free(fw_data, M_CXGBE);
9013 done:
9014         end_synchronized_op(sc, 0);
9015         return (rc);
9016 }
9017
9018 static int
9019 load_cfg(struct adapter *sc, struct t4_data *cfg)
9020 {
9021         int rc;
9022         uint8_t *cfg_data = NULL;
9023
9024         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9025         if (rc)
9026                 return (rc);
9027
9028         if (cfg->len == 0) {
9029                 /* clear */
9030                 rc = -t4_load_cfg(sc, NULL, 0);
9031                 goto done;
9032         }
9033
9034         cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
9035         if (cfg_data == NULL) {
9036                 rc = ENOMEM;
9037                 goto done;
9038         }
9039
9040         rc = copyin(cfg->data, cfg_data, cfg->len);
9041         if (rc == 0)
9042                 rc = -t4_load_cfg(sc, cfg_data, cfg->len);
9043
9044         free(cfg_data, M_CXGBE);
9045 done:
9046         end_synchronized_op(sc, 0);
9047         return (rc);
9048 }
9049
9050 static int
9051 load_boot(struct adapter *sc, struct t4_bootrom *br)
9052 {
9053         int rc;
9054         uint8_t *br_data = NULL;
9055         u_int offset;
9056
9057         if (br->len > 1024 * 1024)
9058                 return (EFBIG);
9059
9060         if (br->pf_offset == 0) {
9061                 /* pfidx */
9062                 if (br->pfidx_addr > 7)
9063                         return (EINVAL);
9064                 offset = G_OFFSET(t4_read_reg(sc, PF_REG(br->pfidx_addr,
9065                     A_PCIE_PF_EXPROM_OFST)));
9066         } else if (br->pf_offset == 1) {
9067                 /* offset */
9068                 offset = G_OFFSET(br->pfidx_addr);
9069         } else {
9070                 return (EINVAL);
9071         }
9072
9073         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldbr");
9074         if (rc)
9075                 return (rc);
9076
9077         if (br->len == 0) {
9078                 /* clear */
9079                 rc = -t4_load_boot(sc, NULL, offset, 0);
9080                 goto done;
9081         }
9082
9083         br_data = malloc(br->len, M_CXGBE, M_WAITOK);
9084         if (br_data == NULL) {
9085                 rc = ENOMEM;
9086                 goto done;
9087         }
9088
9089         rc = copyin(br->data, br_data, br->len);
9090         if (rc == 0)
9091                 rc = -t4_load_boot(sc, br_data, offset, br->len);
9092
9093         free(br_data, M_CXGBE);
9094 done:
9095         end_synchronized_op(sc, 0);
9096         return (rc);
9097 }
9098
9099 static int
9100 load_bootcfg(struct adapter *sc, struct t4_data *bc)
9101 {
9102         int rc;
9103         uint8_t *bc_data = NULL;
9104
9105         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9106         if (rc)
9107                 return (rc);
9108
9109         if (bc->len == 0) {
9110                 /* clear */
9111                 rc = -t4_load_bootcfg(sc, NULL, 0);
9112                 goto done;
9113         }
9114
9115         bc_data = malloc(bc->len, M_CXGBE, M_WAITOK);
9116         if (bc_data == NULL) {
9117                 rc = ENOMEM;
9118                 goto done;
9119         }
9120
9121         rc = copyin(bc->data, bc_data, bc->len);
9122         if (rc == 0)
9123                 rc = -t4_load_bootcfg(sc, bc_data, bc->len);
9124
9125         free(bc_data, M_CXGBE);
9126 done:
9127         end_synchronized_op(sc, 0);
9128         return (rc);
9129 }
9130
9131 static int
9132 cudbg_dump(struct adapter *sc, struct t4_cudbg_dump *dump)
9133 {
9134         int rc;
9135         struct cudbg_init *cudbg;
9136         void *handle, *buf;
9137
9138         /* buf is large, don't block if no memory is available */
9139         buf = malloc(dump->len, M_CXGBE, M_NOWAIT | M_ZERO);
9140         if (buf == NULL)
9141                 return (ENOMEM);
9142
9143         handle = cudbg_alloc_handle();
9144         if (handle == NULL) {
9145                 rc = ENOMEM;
9146                 goto done;
9147         }
9148
9149         cudbg = cudbg_get_init(handle);
9150         cudbg->adap = sc;
9151         cudbg->print = (cudbg_print_cb)printf;
9152
9153 #ifndef notyet
9154         device_printf(sc->dev, "%s: wr_flash %u, len %u, data %p.\n",
9155             __func__, dump->wr_flash, dump->len, dump->data);
9156 #endif
9157
9158         if (dump->wr_flash)
9159                 cudbg->use_flash = 1;
9160         MPASS(sizeof(cudbg->dbg_bitmap) == sizeof(dump->bitmap));
9161         memcpy(cudbg->dbg_bitmap, dump->bitmap, sizeof(cudbg->dbg_bitmap));
9162
9163         rc = cudbg_collect(handle, buf, &dump->len);
9164         if (rc != 0)
9165                 goto done;
9166
9167         rc = copyout(buf, dump->data, dump->len);
9168 done:
9169         cudbg_free_handle(handle);
9170         free(buf, M_CXGBE);
9171         return (rc);
9172 }
9173
9174 static void
9175 free_offload_policy(struct t4_offload_policy *op)
9176 {
9177         struct offload_rule *r;
9178         int i;
9179
9180         if (op == NULL)
9181                 return;
9182
9183         r = &op->rule[0];
9184         for (i = 0; i < op->nrules; i++, r++) {
9185                 free(r->bpf_prog.bf_insns, M_CXGBE);
9186         }
9187         free(op->rule, M_CXGBE);
9188         free(op, M_CXGBE);
9189 }
9190
9191 static int
9192 set_offload_policy(struct adapter *sc, struct t4_offload_policy *uop)
9193 {
9194         int i, rc, len;
9195         struct t4_offload_policy *op, *old;
9196         struct bpf_program *bf;
9197         const struct offload_settings *s;
9198         struct offload_rule *r;
9199         void *u;
9200
9201         if (!is_offload(sc))
9202                 return (ENODEV);
9203
9204         if (uop->nrules == 0) {
9205                 /* Delete installed policies. */
9206                 op = NULL;
9207                 goto set_policy;
9208         } if (uop->nrules > 256) { /* arbitrary */
9209                 return (E2BIG);
9210         }
9211
9212         /* Copy userspace offload policy to kernel */
9213         op = malloc(sizeof(*op), M_CXGBE, M_ZERO | M_WAITOK);
9214         op->nrules = uop->nrules;
9215         len = op->nrules * sizeof(struct offload_rule);
9216         op->rule = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9217         rc = copyin(uop->rule, op->rule, len);
9218         if (rc) {
9219                 free(op->rule, M_CXGBE);
9220                 free(op, M_CXGBE);
9221                 return (rc);
9222         }
9223
9224         r = &op->rule[0];
9225         for (i = 0; i < op->nrules; i++, r++) {
9226
9227                 /* Validate open_type */
9228                 if (r->open_type != OPEN_TYPE_LISTEN &&
9229                     r->open_type != OPEN_TYPE_ACTIVE &&
9230                     r->open_type != OPEN_TYPE_PASSIVE &&
9231                     r->open_type != OPEN_TYPE_DONTCARE) {
9232 error:
9233                         /*
9234                          * Rules 0 to i have malloc'd filters that need to be
9235                          * freed.  Rules i+1 to nrules have userspace pointers
9236                          * and should be left alone.
9237                          */
9238                         op->nrules = i;
9239                         free_offload_policy(op);
9240                         return (rc);
9241                 }
9242
9243                 /* Validate settings */
9244                 s = &r->settings;
9245                 if ((s->offload != 0 && s->offload != 1) ||
9246                     s->cong_algo < -1 || s->cong_algo > CONG_ALG_HIGHSPEED ||
9247                     s->sched_class < -1 ||
9248                     s->sched_class >= sc->chip_params->nsched_cls) {
9249                         rc = EINVAL;
9250                         goto error;
9251                 }
9252
9253                 bf = &r->bpf_prog;
9254                 u = bf->bf_insns;       /* userspace ptr */
9255                 bf->bf_insns = NULL;
9256                 if (bf->bf_len == 0) {
9257                         /* legal, matches everything */
9258                         continue;
9259                 }
9260                 len = bf->bf_len * sizeof(*bf->bf_insns);
9261                 bf->bf_insns = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9262                 rc = copyin(u, bf->bf_insns, len);
9263                 if (rc != 0)
9264                         goto error;
9265
9266                 if (!bpf_validate(bf->bf_insns, bf->bf_len)) {
9267                         rc = EINVAL;
9268                         goto error;
9269                 }
9270         }
9271 set_policy:
9272         rw_wlock(&sc->policy_lock);
9273         old = sc->policy;
9274         sc->policy = op;
9275         rw_wunlock(&sc->policy_lock);
9276         free_offload_policy(old);
9277
9278         return (0);
9279 }
9280
9281 #define MAX_READ_BUF_SIZE (128 * 1024)
9282 static int
9283 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
9284 {
9285         uint32_t addr, remaining, n;
9286         uint32_t *buf;
9287         int rc;
9288         uint8_t *dst;
9289
9290         rc = validate_mem_range(sc, mr->addr, mr->len);
9291         if (rc != 0)
9292                 return (rc);
9293
9294         buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
9295         addr = mr->addr;
9296         remaining = mr->len;
9297         dst = (void *)mr->data;
9298
9299         while (remaining) {
9300                 n = min(remaining, MAX_READ_BUF_SIZE);
9301                 read_via_memwin(sc, 2, addr, buf, n);
9302
9303                 rc = copyout(buf, dst, n);
9304                 if (rc != 0)
9305                         break;
9306
9307                 dst += n;
9308                 remaining -= n;
9309                 addr += n;
9310         }
9311
9312         free(buf, M_CXGBE);
9313         return (rc);
9314 }
9315 #undef MAX_READ_BUF_SIZE
9316
9317 static int
9318 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
9319 {
9320         int rc;
9321
9322         if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
9323                 return (EINVAL);
9324
9325         if (i2cd->len > sizeof(i2cd->data))
9326                 return (EFBIG);
9327
9328         rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
9329         if (rc)
9330                 return (rc);
9331         rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
9332             i2cd->offset, i2cd->len, &i2cd->data[0]);
9333         end_synchronized_op(sc, 0);
9334
9335         return (rc);
9336 }
9337
9338 int
9339 t4_os_find_pci_capability(struct adapter *sc, int cap)
9340 {
9341         int i;
9342
9343         return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9344 }
9345
9346 int
9347 t4_os_pci_save_state(struct adapter *sc)
9348 {
9349         device_t dev;
9350         struct pci_devinfo *dinfo;
9351
9352         dev = sc->dev;
9353         dinfo = device_get_ivars(dev);
9354
9355         pci_cfg_save(dev, dinfo, 0);
9356         return (0);
9357 }
9358
9359 int
9360 t4_os_pci_restore_state(struct adapter *sc)
9361 {
9362         device_t dev;
9363         struct pci_devinfo *dinfo;
9364
9365         dev = sc->dev;
9366         dinfo = device_get_ivars(dev);
9367
9368         pci_cfg_restore(dev, dinfo);
9369         return (0);
9370 }
9371
9372 void
9373 t4_os_portmod_changed(struct port_info *pi)
9374 {
9375         struct adapter *sc = pi->adapter;
9376         struct vi_info *vi;
9377         struct ifnet *ifp;
9378         static const char *mod_str[] = {
9379                 NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9380         };
9381
9382         MPASS((pi->flags & FIXED_IFMEDIA) == 0);
9383
9384         vi = &pi->vi[0];
9385         if (begin_synchronized_op(sc, vi, HOLD_LOCK, "t4mod") == 0) {
9386                 PORT_LOCK(pi);
9387                 build_medialist(pi, &pi->media);
9388                 apply_l1cfg(pi);
9389                 PORT_UNLOCK(pi);
9390                 end_synchronized_op(sc, LOCK_HELD);
9391         }
9392
9393         ifp = vi->ifp;
9394         if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
9395                 if_printf(ifp, "transceiver unplugged.\n");
9396         else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
9397                 if_printf(ifp, "unknown transceiver inserted.\n");
9398         else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
9399                 if_printf(ifp, "unsupported transceiver inserted.\n");
9400         else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
9401                 if_printf(ifp, "%dGbps %s transceiver inserted.\n",
9402                     port_top_speed(pi), mod_str[pi->mod_type]);
9403         } else {
9404                 if_printf(ifp, "transceiver (type %d) inserted.\n",
9405                     pi->mod_type);
9406         }
9407 }
9408
9409 void
9410 t4_os_link_changed(struct port_info *pi)
9411 {
9412         struct vi_info *vi;
9413         struct ifnet *ifp;
9414         struct link_config *lc;
9415         int v;
9416
9417         PORT_LOCK_ASSERT_OWNED(pi);
9418
9419         for_each_vi(pi, v, vi) {
9420                 ifp = vi->ifp;
9421                 if (ifp == NULL)
9422                         continue;
9423
9424                 lc = &pi->link_cfg;
9425                 if (lc->link_ok) {
9426                         ifp->if_baudrate = IF_Mbps(lc->speed);
9427                         if_link_state_change(ifp, LINK_STATE_UP);
9428                 } else {
9429                         if_link_state_change(ifp, LINK_STATE_DOWN);
9430                 }
9431         }
9432 }
9433
9434 void
9435 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
9436 {
9437         struct adapter *sc;
9438
9439         sx_slock(&t4_list_lock);
9440         SLIST_FOREACH(sc, &t4_list, link) {
9441                 /*
9442                  * func should not make any assumptions about what state sc is
9443                  * in - the only guarantee is that sc->sc_lock is a valid lock.
9444                  */
9445                 func(sc, arg);
9446         }
9447         sx_sunlock(&t4_list_lock);
9448 }
9449
9450 static int
9451 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9452     struct thread *td)
9453 {
9454         int rc;
9455         struct adapter *sc = dev->si_drv1;
9456
9457         rc = priv_check(td, PRIV_DRIVER);
9458         if (rc != 0)
9459                 return (rc);
9460
9461         switch (cmd) {
9462         case CHELSIO_T4_GETREG: {
9463                 struct t4_reg *edata = (struct t4_reg *)data;
9464
9465                 if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9466                         return (EFAULT);
9467
9468                 if (edata->size == 4)
9469                         edata->val = t4_read_reg(sc, edata->addr);
9470                 else if (edata->size == 8)
9471                         edata->val = t4_read_reg64(sc, edata->addr);
9472                 else
9473                         return (EINVAL);
9474
9475                 break;
9476         }
9477         case CHELSIO_T4_SETREG: {
9478                 struct t4_reg *edata = (struct t4_reg *)data;
9479
9480                 if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9481                         return (EFAULT);
9482
9483                 if (edata->size == 4) {
9484                         if (edata->val & 0xffffffff00000000)
9485                                 return (EINVAL);
9486                         t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9487                 } else if (edata->size == 8)
9488                         t4_write_reg64(sc, edata->addr, edata->val);
9489                 else
9490                         return (EINVAL);
9491                 break;
9492         }
9493         case CHELSIO_T4_REGDUMP: {
9494                 struct t4_regdump *regs = (struct t4_regdump *)data;
9495                 int reglen = t4_get_regs_len(sc);
9496                 uint8_t *buf;
9497
9498                 if (regs->len < reglen) {
9499                         regs->len = reglen; /* hint to the caller */
9500                         return (ENOBUFS);
9501                 }
9502
9503                 regs->len = reglen;
9504                 buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
9505                 get_regs(sc, regs, buf);
9506                 rc = copyout(buf, regs->data, reglen);
9507                 free(buf, M_CXGBE);
9508                 break;
9509         }
9510         case CHELSIO_T4_GET_FILTER_MODE:
9511                 rc = get_filter_mode(sc, (uint32_t *)data);
9512                 break;
9513         case CHELSIO_T4_SET_FILTER_MODE:
9514                 rc = set_filter_mode(sc, *(uint32_t *)data);
9515                 break;
9516         case CHELSIO_T4_GET_FILTER:
9517                 rc = get_filter(sc, (struct t4_filter *)data);
9518                 break;
9519         case CHELSIO_T4_SET_FILTER:
9520                 rc = set_filter(sc, (struct t4_filter *)data);
9521                 break;
9522         case CHELSIO_T4_DEL_FILTER:
9523                 rc = del_filter(sc, (struct t4_filter *)data);
9524                 break;
9525         case CHELSIO_T4_GET_SGE_CONTEXT:
9526                 rc = get_sge_context(sc, (struct t4_sge_context *)data);
9527                 break;
9528         case CHELSIO_T4_LOAD_FW:
9529                 rc = load_fw(sc, (struct t4_data *)data);
9530                 break;
9531         case CHELSIO_T4_GET_MEM:
9532                 rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
9533                 break;
9534         case CHELSIO_T4_GET_I2C:
9535                 rc = read_i2c(sc, (struct t4_i2c_data *)data);
9536                 break;
9537         case CHELSIO_T4_CLEAR_STATS: {
9538                 int i, v, bg_map;
9539                 u_int port_id = *(uint32_t *)data;
9540                 struct port_info *pi;
9541                 struct vi_info *vi;
9542
9543                 if (port_id >= sc->params.nports)
9544                         return (EINVAL);
9545                 pi = sc->port[port_id];
9546                 if (pi == NULL)
9547                         return (EIO);
9548
9549                 /* MAC stats */
9550                 t4_clr_port_stats(sc, pi->tx_chan);
9551                 pi->tx_parse_error = 0;
9552                 pi->tnl_cong_drops = 0;
9553                 mtx_lock(&sc->reg_lock);
9554                 for_each_vi(pi, v, vi) {
9555                         if (vi->flags & VI_INIT_DONE)
9556                                 t4_clr_vi_stats(sc, vi->viid);
9557                 }
9558                 bg_map = pi->mps_bg_map;
9559                 v = 0;  /* reuse */
9560                 while (bg_map) {
9561                         i = ffs(bg_map) - 1;
9562                         t4_write_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
9563                             1, A_TP_MIB_TNL_CNG_DROP_0 + i);
9564                         bg_map &= ~(1 << i);
9565                 }
9566                 mtx_unlock(&sc->reg_lock);
9567
9568                 /*
9569                  * Since this command accepts a port, clear stats for
9570                  * all VIs on this port.
9571                  */
9572                 for_each_vi(pi, v, vi) {
9573                         if (vi->flags & VI_INIT_DONE) {
9574                                 struct sge_rxq *rxq;
9575                                 struct sge_txq *txq;
9576                                 struct sge_wrq *wrq;
9577
9578                                 for_each_rxq(vi, i, rxq) {
9579 #if defined(INET) || defined(INET6)
9580                                         rxq->lro.lro_queued = 0;
9581                                         rxq->lro.lro_flushed = 0;
9582 #endif
9583                                         rxq->rxcsum = 0;
9584                                         rxq->vlan_extraction = 0;
9585                                 }
9586
9587                                 for_each_txq(vi, i, txq) {
9588                                         txq->txcsum = 0;
9589                                         txq->tso_wrs = 0;
9590                                         txq->vlan_insertion = 0;
9591                                         txq->imm_wrs = 0;
9592                                         txq->sgl_wrs = 0;
9593                                         txq->txpkt_wrs = 0;
9594                                         txq->txpkts0_wrs = 0;
9595                                         txq->txpkts1_wrs = 0;
9596                                         txq->txpkts0_pkts = 0;
9597                                         txq->txpkts1_pkts = 0;
9598                                         mp_ring_reset_stats(txq->r);
9599                                 }
9600
9601 #ifdef TCP_OFFLOAD
9602                                 /* nothing to clear for each ofld_rxq */
9603
9604                                 for_each_ofld_txq(vi, i, wrq) {
9605                                         wrq->tx_wrs_direct = 0;
9606                                         wrq->tx_wrs_copied = 0;
9607                                 }
9608 #endif
9609
9610                                 if (IS_MAIN_VI(vi)) {
9611                                         wrq = &sc->sge.ctrlq[pi->port_id];
9612                                         wrq->tx_wrs_direct = 0;
9613                                         wrq->tx_wrs_copied = 0;
9614                                 }
9615                         }
9616                 }
9617                 break;
9618         }
9619         case CHELSIO_T4_SCHED_CLASS:
9620                 rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
9621                 break;
9622         case CHELSIO_T4_SCHED_QUEUE:
9623                 rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
9624                 break;
9625         case CHELSIO_T4_GET_TRACER:
9626                 rc = t4_get_tracer(sc, (struct t4_tracer *)data);
9627                 break;
9628         case CHELSIO_T4_SET_TRACER:
9629                 rc = t4_set_tracer(sc, (struct t4_tracer *)data);
9630                 break;
9631         case CHELSIO_T4_LOAD_CFG:
9632                 rc = load_cfg(sc, (struct t4_data *)data);
9633                 break;
9634         case CHELSIO_T4_LOAD_BOOT:
9635                 rc = load_boot(sc, (struct t4_bootrom *)data);
9636                 break;
9637         case CHELSIO_T4_LOAD_BOOTCFG:
9638                 rc = load_bootcfg(sc, (struct t4_data *)data);
9639                 break;
9640         case CHELSIO_T4_CUDBG_DUMP:
9641                 rc = cudbg_dump(sc, (struct t4_cudbg_dump *)data);
9642                 break;
9643         case CHELSIO_T4_SET_OFLD_POLICY:
9644                 rc = set_offload_policy(sc, (struct t4_offload_policy *)data);
9645                 break;
9646         default:
9647                 rc = ENOTTY;
9648         }
9649
9650         return (rc);
9651 }
9652
9653 void
9654 t4_db_full(struct adapter *sc)
9655 {
9656
9657         CXGBE_UNIMPLEMENTED(__func__);
9658 }
9659
9660 void
9661 t4_db_dropped(struct adapter *sc)
9662 {
9663
9664         CXGBE_UNIMPLEMENTED(__func__);
9665 }
9666
9667 #ifdef TCP_OFFLOAD
9668 static int
9669 toe_capability(struct vi_info *vi, int enable)
9670 {
9671         int rc;
9672         struct port_info *pi = vi->pi;
9673         struct adapter *sc = pi->adapter;
9674
9675         ASSERT_SYNCHRONIZED_OP(sc);
9676
9677         if (!is_offload(sc))
9678                 return (ENODEV);
9679
9680         if (enable) {
9681                 if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
9682                         /* TOE is already enabled. */
9683                         return (0);
9684                 }
9685
9686                 /*
9687                  * We need the port's queues around so that we're able to send
9688                  * and receive CPLs to/from the TOE even if the ifnet for this
9689                  * port has never been UP'd administratively.
9690                  */
9691                 if (!(vi->flags & VI_INIT_DONE)) {
9692                         rc = vi_full_init(vi);
9693                         if (rc)
9694                                 return (rc);
9695                 }
9696                 if (!(pi->vi[0].flags & VI_INIT_DONE)) {
9697                         rc = vi_full_init(&pi->vi[0]);
9698                         if (rc)
9699                                 return (rc);
9700                 }
9701
9702                 if (isset(&sc->offload_map, pi->port_id)) {
9703                         /* TOE is enabled on another VI of this port. */
9704                         pi->uld_vis++;
9705                         return (0);
9706                 }
9707
9708                 if (!uld_active(sc, ULD_TOM)) {
9709                         rc = t4_activate_uld(sc, ULD_TOM);
9710                         if (rc == EAGAIN) {
9711                                 log(LOG_WARNING,
9712                                     "You must kldload t4_tom.ko before trying "
9713                                     "to enable TOE on a cxgbe interface.\n");
9714                         }
9715                         if (rc != 0)
9716                                 return (rc);
9717                         KASSERT(sc->tom_softc != NULL,
9718                             ("%s: TOM activated but softc NULL", __func__));
9719                         KASSERT(uld_active(sc, ULD_TOM),
9720                             ("%s: TOM activated but flag not set", __func__));
9721                 }
9722
9723                 /* Activate iWARP and iSCSI too, if the modules are loaded. */
9724                 if (!uld_active(sc, ULD_IWARP))
9725                         (void) t4_activate_uld(sc, ULD_IWARP);
9726                 if (!uld_active(sc, ULD_ISCSI))
9727                         (void) t4_activate_uld(sc, ULD_ISCSI);
9728
9729                 pi->uld_vis++;
9730                 setbit(&sc->offload_map, pi->port_id);
9731         } else {
9732                 pi->uld_vis--;
9733
9734                 if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
9735                         return (0);
9736
9737                 KASSERT(uld_active(sc, ULD_TOM),
9738                     ("%s: TOM never initialized?", __func__));
9739                 clrbit(&sc->offload_map, pi->port_id);
9740         }
9741
9742         return (0);
9743 }
9744
9745 /*
9746  * Add an upper layer driver to the global list.
9747  */
9748 int
9749 t4_register_uld(struct uld_info *ui)
9750 {
9751         int rc = 0;
9752         struct uld_info *u;
9753
9754         sx_xlock(&t4_uld_list_lock);
9755         SLIST_FOREACH(u, &t4_uld_list, link) {
9756             if (u->uld_id == ui->uld_id) {
9757                     rc = EEXIST;
9758                     goto done;
9759             }
9760         }
9761
9762         SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
9763         ui->refcount = 0;
9764 done:
9765         sx_xunlock(&t4_uld_list_lock);
9766         return (rc);
9767 }
9768
9769 int
9770 t4_unregister_uld(struct uld_info *ui)
9771 {
9772         int rc = EINVAL;
9773         struct uld_info *u;
9774
9775         sx_xlock(&t4_uld_list_lock);
9776
9777         SLIST_FOREACH(u, &t4_uld_list, link) {
9778             if (u == ui) {
9779                     if (ui->refcount > 0) {
9780                             rc = EBUSY;
9781                             goto done;
9782                     }
9783
9784                     SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
9785                     rc = 0;
9786                     goto done;
9787             }
9788         }
9789 done:
9790         sx_xunlock(&t4_uld_list_lock);
9791         return (rc);
9792 }
9793
9794 int
9795 t4_activate_uld(struct adapter *sc, int id)
9796 {
9797         int rc;
9798         struct uld_info *ui;
9799
9800         ASSERT_SYNCHRONIZED_OP(sc);
9801
9802         if (id < 0 || id > ULD_MAX)
9803                 return (EINVAL);
9804         rc = EAGAIN;    /* kldoad the module with this ULD and try again. */
9805
9806         sx_slock(&t4_uld_list_lock);
9807
9808         SLIST_FOREACH(ui, &t4_uld_list, link) {
9809                 if (ui->uld_id == id) {
9810                         if (!(sc->flags & FULL_INIT_DONE)) {
9811                                 rc = adapter_full_init(sc);
9812                                 if (rc != 0)
9813                                         break;
9814                         }
9815
9816                         rc = ui->activate(sc);
9817                         if (rc == 0) {
9818                                 setbit(&sc->active_ulds, id);
9819                                 ui->refcount++;
9820                         }
9821                         break;
9822                 }
9823         }
9824
9825         sx_sunlock(&t4_uld_list_lock);
9826
9827         return (rc);
9828 }
9829
9830 int
9831 t4_deactivate_uld(struct adapter *sc, int id)
9832 {
9833         int rc;
9834         struct uld_info *ui;
9835
9836         ASSERT_SYNCHRONIZED_OP(sc);
9837
9838         if (id < 0 || id > ULD_MAX)
9839                 return (EINVAL);
9840         rc = ENXIO;
9841
9842         sx_slock(&t4_uld_list_lock);
9843
9844         SLIST_FOREACH(ui, &t4_uld_list, link) {
9845                 if (ui->uld_id == id) {
9846                         rc = ui->deactivate(sc);
9847                         if (rc == 0) {
9848                                 clrbit(&sc->active_ulds, id);
9849                                 ui->refcount--;
9850                         }
9851                         break;
9852                 }
9853         }
9854
9855         sx_sunlock(&t4_uld_list_lock);
9856
9857         return (rc);
9858 }
9859
9860 int
9861 uld_active(struct adapter *sc, int uld_id)
9862 {
9863
9864         MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
9865
9866         return (isset(&sc->active_ulds, uld_id));
9867 }
9868 #endif
9869
9870 /*
9871  * t  = ptr to tunable.
9872  * nc = number of CPUs.
9873  * c  = compiled in default for that tunable.
9874  */
9875 static void
9876 calculate_nqueues(int *t, int nc, const int c)
9877 {
9878         int nq;
9879
9880         if (*t > 0)
9881                 return;
9882         nq = *t < 0 ? -*t : c;
9883         *t = min(nc, nq);
9884 }
9885
9886 /*
9887  * Come up with reasonable defaults for some of the tunables, provided they're
9888  * not set by the user (in which case we'll use the values as is).
9889  */
9890 static void
9891 tweak_tunables(void)
9892 {
9893         int nc = mp_ncpus;      /* our snapshot of the number of CPUs */
9894
9895         if (t4_ntxq < 1) {
9896 #ifdef RSS
9897                 t4_ntxq = rss_getnumbuckets();
9898 #else
9899                 calculate_nqueues(&t4_ntxq, nc, NTXQ);
9900 #endif
9901         }
9902
9903         calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
9904
9905         if (t4_nrxq < 1) {
9906 #ifdef RSS
9907                 t4_nrxq = rss_getnumbuckets();
9908 #else
9909                 calculate_nqueues(&t4_nrxq, nc, NRXQ);
9910 #endif
9911         }
9912
9913         calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
9914
9915 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
9916         calculate_nqueues(&t4_nofldtxq, nc, NOFLDTXQ);
9917         calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
9918 #endif
9919 #ifdef TCP_OFFLOAD
9920         calculate_nqueues(&t4_nofldrxq, nc, NOFLDRXQ);
9921         calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
9922
9923         if (t4_toecaps_allowed == -1)
9924                 t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
9925
9926         if (t4_rdmacaps_allowed == -1) {
9927                 t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
9928                     FW_CAPS_CONFIG_RDMA_RDMAC;
9929         }
9930
9931         if (t4_iscsicaps_allowed == -1) {
9932                 t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
9933                     FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
9934                     FW_CAPS_CONFIG_ISCSI_T10DIF;
9935         }
9936
9937         if (t4_tmr_idx_ofld < 0 || t4_tmr_idx_ofld >= SGE_NTIMERS)
9938                 t4_tmr_idx_ofld = TMR_IDX_OFLD;
9939
9940         if (t4_pktc_idx_ofld < -1 || t4_pktc_idx_ofld >= SGE_NCOUNTERS)
9941                 t4_pktc_idx_ofld = PKTC_IDX_OFLD;
9942 #else
9943         if (t4_toecaps_allowed == -1)
9944                 t4_toecaps_allowed = 0;
9945
9946         if (t4_rdmacaps_allowed == -1)
9947                 t4_rdmacaps_allowed = 0;
9948
9949         if (t4_iscsicaps_allowed == -1)
9950                 t4_iscsicaps_allowed = 0;
9951 #endif
9952
9953 #ifdef DEV_NETMAP
9954         calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
9955         calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
9956 #endif
9957
9958         if (t4_tmr_idx < 0 || t4_tmr_idx >= SGE_NTIMERS)
9959                 t4_tmr_idx = TMR_IDX;
9960
9961         if (t4_pktc_idx < -1 || t4_pktc_idx >= SGE_NCOUNTERS)
9962                 t4_pktc_idx = PKTC_IDX;
9963
9964         if (t4_qsize_txq < 128)
9965                 t4_qsize_txq = 128;
9966
9967         if (t4_qsize_rxq < 128)
9968                 t4_qsize_rxq = 128;
9969         while (t4_qsize_rxq & 7)
9970                 t4_qsize_rxq++;
9971
9972         t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
9973
9974         /*
9975          * Number of VIs to create per-port.  The first VI is the "main" regular
9976          * VI for the port.  The rest are additional virtual interfaces on the
9977          * same physical port.  Note that the main VI does not have native
9978          * netmap support but the extra VIs do.
9979          *
9980          * Limit the number of VIs per port to the number of available
9981          * MAC addresses per port.
9982          */
9983         if (t4_num_vis < 1)
9984                 t4_num_vis = 1;
9985         if (t4_num_vis > nitems(vi_mac_funcs)) {
9986                 t4_num_vis = nitems(vi_mac_funcs);
9987                 printf("cxgbe: number of VIs limited to %d\n", t4_num_vis);
9988         }
9989
9990         if (pcie_relaxed_ordering < 0 || pcie_relaxed_ordering > 2) {
9991                 pcie_relaxed_ordering = 1;
9992 #if defined(__i386__) || defined(__amd64__)
9993                 if (cpu_vendor_id == CPU_VENDOR_INTEL)
9994                         pcie_relaxed_ordering = 0;
9995 #endif
9996         }
9997 }
9998
9999 #ifdef DDB
10000 static void
10001 t4_dump_tcb(struct adapter *sc, int tid)
10002 {
10003         uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
10004
10005         reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
10006         save = t4_read_reg(sc, reg);
10007         base = sc->memwin[2].mw_base;
10008
10009         /* Dump TCB for the tid */
10010         tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
10011         tcb_addr += tid * TCB_SIZE;
10012
10013         if (is_t4(sc)) {
10014                 pf = 0;
10015                 win_pos = tcb_addr & ~0xf;      /* start must be 16B aligned */
10016         } else {
10017                 pf = V_PFNUM(sc->pf);
10018                 win_pos = tcb_addr & ~0x7f;     /* start must be 128B aligned */
10019         }
10020         t4_write_reg(sc, reg, win_pos | pf);
10021         t4_read_reg(sc, reg);
10022
10023         off = tcb_addr - win_pos;
10024         for (i = 0; i < 4; i++) {
10025                 uint32_t buf[8];
10026                 for (j = 0; j < 8; j++, off += 4)
10027                         buf[j] = htonl(t4_read_reg(sc, base + off));
10028
10029                 db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
10030                     buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
10031                     buf[7]);
10032         }
10033
10034         t4_write_reg(sc, reg, save);
10035         t4_read_reg(sc, reg);
10036 }
10037
10038 static void
10039 t4_dump_devlog(struct adapter *sc)
10040 {
10041         struct devlog_params *dparams = &sc->params.devlog;
10042         struct fw_devlog_e e;
10043         int i, first, j, m, nentries, rc;
10044         uint64_t ftstamp = UINT64_MAX;
10045
10046         if (dparams->start == 0) {
10047                 db_printf("devlog params not valid\n");
10048                 return;
10049         }
10050
10051         nentries = dparams->size / sizeof(struct fw_devlog_e);
10052         m = fwmtype_to_hwmtype(dparams->memtype);
10053
10054         /* Find the first entry. */
10055         first = -1;
10056         for (i = 0; i < nentries && !db_pager_quit; i++) {
10057                 rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10058                     sizeof(e), (void *)&e);
10059                 if (rc != 0)
10060                         break;
10061
10062                 if (e.timestamp == 0)
10063                         break;
10064
10065                 e.timestamp = be64toh(e.timestamp);
10066                 if (e.timestamp < ftstamp) {
10067                         ftstamp = e.timestamp;
10068                         first = i;
10069                 }
10070         }
10071
10072         if (first == -1)
10073                 return;
10074
10075         i = first;
10076         do {
10077                 rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10078                     sizeof(e), (void *)&e);
10079                 if (rc != 0)
10080                         return;
10081
10082                 if (e.timestamp == 0)
10083                         return;
10084
10085                 e.timestamp = be64toh(e.timestamp);
10086                 e.seqno = be32toh(e.seqno);
10087                 for (j = 0; j < 8; j++)
10088                         e.params[j] = be32toh(e.params[j]);
10089
10090                 db_printf("%10d  %15ju  %8s  %8s  ",
10091                     e.seqno, e.timestamp,
10092                     (e.level < nitems(devlog_level_strings) ?
10093                         devlog_level_strings[e.level] : "UNKNOWN"),
10094                     (e.facility < nitems(devlog_facility_strings) ?
10095                         devlog_facility_strings[e.facility] : "UNKNOWN"));
10096                 db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
10097                     e.params[3], e.params[4], e.params[5], e.params[6],
10098                     e.params[7]);
10099
10100                 if (++i == nentries)
10101                         i = 0;
10102         } while (i != first && !db_pager_quit);
10103 }
10104
10105 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
10106 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
10107
10108 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
10109 {
10110         device_t dev;
10111         int t;
10112         bool valid;
10113
10114         valid = false;
10115         t = db_read_token();
10116         if (t == tIDENT) {
10117                 dev = device_lookup_by_name(db_tok_string);
10118                 valid = true;
10119         }
10120         db_skip_to_eol();
10121         if (!valid) {
10122                 db_printf("usage: show t4 devlog <nexus>\n");
10123                 return;
10124         }
10125
10126         if (dev == NULL) {
10127                 db_printf("device not found\n");
10128                 return;
10129         }
10130
10131         t4_dump_devlog(device_get_softc(dev));
10132 }
10133
10134 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
10135 {
10136         device_t dev;
10137         int radix, tid, t;
10138         bool valid;
10139
10140         valid = false;
10141         radix = db_radix;
10142         db_radix = 10;
10143         t = db_read_token();
10144         if (t == tIDENT) {
10145                 dev = device_lookup_by_name(db_tok_string);
10146                 t = db_read_token();
10147                 if (t == tNUMBER) {
10148                         tid = db_tok_number;
10149                         valid = true;
10150                 }
10151         }
10152         db_radix = radix;
10153         db_skip_to_eol();
10154         if (!valid) {
10155                 db_printf("usage: show t4 tcb <nexus> <tid>\n");
10156                 return;
10157         }
10158
10159         if (dev == NULL) {
10160                 db_printf("device not found\n");
10161                 return;
10162         }
10163         if (tid < 0) {
10164                 db_printf("invalid tid\n");
10165                 return;
10166         }
10167
10168         t4_dump_tcb(device_get_softc(dev), tid);
10169 }
10170 #endif
10171
10172 /*
10173  * Borrowed from cesa_prep_aes_key().
10174  *
10175  * NB: The crypto engine wants the words in the decryption key in reverse
10176  * order.
10177  */
10178 void
10179 t4_aes_getdeckey(void *dec_key, const void *enc_key, unsigned int kbits)
10180 {
10181         uint32_t ek[4 * (RIJNDAEL_MAXNR + 1)];
10182         uint32_t *dkey;
10183         int i;
10184
10185         rijndaelKeySetupEnc(ek, enc_key, kbits);
10186         dkey = dec_key;
10187         dkey += (kbits / 8) / 4;
10188
10189         switch (kbits) {
10190         case 128:
10191                 for (i = 0; i < 4; i++)
10192                         *--dkey = htobe32(ek[4 * 10 + i]);
10193                 break;
10194         case 192:
10195                 for (i = 0; i < 2; i++)
10196                         *--dkey = htobe32(ek[4 * 11 + 2 + i]);
10197                 for (i = 0; i < 4; i++)
10198                         *--dkey = htobe32(ek[4 * 12 + i]);
10199                 break;
10200         case 256:
10201                 for (i = 0; i < 4; i++)
10202                         *--dkey = htobe32(ek[4 * 13 + i]);
10203                 for (i = 0; i < 4; i++)
10204                         *--dkey = htobe32(ek[4 * 14 + i]);
10205                 break;
10206         }
10207         MPASS(dkey == dec_key);
10208 }
10209
10210 static struct sx mlu;   /* mod load unload */
10211 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
10212
10213 static int
10214 mod_event(module_t mod, int cmd, void *arg)
10215 {
10216         int rc = 0;
10217         static int loaded = 0;
10218
10219         switch (cmd) {
10220         case MOD_LOAD:
10221                 sx_xlock(&mlu);
10222                 if (loaded++ == 0) {
10223                         t4_sge_modload();
10224                         t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10225                             t4_filter_rpl, CPL_COOKIE_FILTER);
10226                         t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL,
10227                             do_l2t_write_rpl, CPL_COOKIE_FILTER);
10228                         t4_register_shared_cpl_handler(CPL_ACT_OPEN_RPL,
10229                             t4_hashfilter_ao_rpl, CPL_COOKIE_HASHFILTER);
10230                         t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10231                             t4_hashfilter_tcb_rpl, CPL_COOKIE_HASHFILTER);
10232                         t4_register_shared_cpl_handler(CPL_ABORT_RPL_RSS,
10233                             t4_del_hashfilter_rpl, CPL_COOKIE_HASHFILTER);
10234                         t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
10235                         t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
10236                         t4_register_cpl_handler(CPL_SMT_WRITE_RPL,
10237                             do_smt_write_rpl);
10238                         sx_init(&t4_list_lock, "T4/T5 adapters");
10239                         SLIST_INIT(&t4_list);
10240 #ifdef TCP_OFFLOAD
10241                         sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
10242                         SLIST_INIT(&t4_uld_list);
10243 #endif
10244                         t4_tracer_modload();
10245                         tweak_tunables();
10246                 }
10247                 sx_xunlock(&mlu);
10248                 break;
10249
10250         case MOD_UNLOAD:
10251                 sx_xlock(&mlu);
10252                 if (--loaded == 0) {
10253                         int tries;
10254
10255                         sx_slock(&t4_list_lock);
10256                         if (!SLIST_EMPTY(&t4_list)) {
10257                                 rc = EBUSY;
10258                                 sx_sunlock(&t4_list_lock);
10259                                 goto done_unload;
10260                         }
10261 #ifdef TCP_OFFLOAD
10262                         sx_slock(&t4_uld_list_lock);
10263                         if (!SLIST_EMPTY(&t4_uld_list)) {
10264                                 rc = EBUSY;
10265                                 sx_sunlock(&t4_uld_list_lock);
10266                                 sx_sunlock(&t4_list_lock);
10267                                 goto done_unload;
10268                         }
10269 #endif
10270                         tries = 0;
10271                         while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
10272                                 uprintf("%ju clusters with custom free routine "
10273                                     "still is use.\n", t4_sge_extfree_refs());
10274                                 pause("t4unload", 2 * hz);
10275                         }
10276 #ifdef TCP_OFFLOAD
10277                         sx_sunlock(&t4_uld_list_lock);
10278 #endif
10279                         sx_sunlock(&t4_list_lock);
10280
10281                         if (t4_sge_extfree_refs() == 0) {
10282                                 t4_tracer_modunload();
10283 #ifdef TCP_OFFLOAD
10284                                 sx_destroy(&t4_uld_list_lock);
10285 #endif
10286                                 sx_destroy(&t4_list_lock);
10287                                 t4_sge_modunload();
10288                                 loaded = 0;
10289                         } else {
10290                                 rc = EBUSY;
10291                                 loaded++;       /* undo earlier decrement */
10292                         }
10293                 }
10294 done_unload:
10295                 sx_xunlock(&mlu);
10296                 break;
10297         }
10298
10299         return (rc);
10300 }
10301
10302 static devclass_t t4_devclass, t5_devclass, t6_devclass;
10303 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
10304 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
10305
10306 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
10307 MODULE_VERSION(t4nex, 1);
10308 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
10309 #ifdef DEV_NETMAP
10310 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
10311 #endif /* DEV_NETMAP */
10312
10313 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
10314 MODULE_VERSION(t5nex, 1);
10315 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
10316 #ifdef DEV_NETMAP
10317 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
10318 #endif /* DEV_NETMAP */
10319
10320 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
10321 MODULE_VERSION(t6nex, 1);
10322 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
10323 #ifdef DEV_NETMAP
10324 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
10325 #endif /* DEV_NETMAP */
10326
10327 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
10328 MODULE_VERSION(cxgbe, 1);
10329
10330 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
10331 MODULE_VERSION(cxl, 1);
10332
10333 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
10334 MODULE_VERSION(cc, 1);
10335
10336 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
10337 MODULE_VERSION(vcxgbe, 1);
10338
10339 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
10340 MODULE_VERSION(vcxl, 1);
10341
10342 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
10343 MODULE_VERSION(vcc, 1);