1 /*- 2 * Copyright (c) 2014-2016, Matthew Macy <mmacy@nextbsd.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions are met: 7 * 8 * 1. Redistributions of source code must retain the above copyright notice, 9 * this list of conditions and the following disclaimer. 10 * 11 * 2. Neither the name of Matthew Macy nor the names of its 12 * contributors may be used to endorse or promote products derived from 13 * this software without specific prior written permission. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" 16 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE 19 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 20 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 21 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 22 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 23 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 24 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 25 * POSSIBILITY OF SUCH DAMAGE. 26 */ 27 28 #include <sys/cdefs.h> 29 __FBSDID("$FreeBSD$"); 30 31 #include "opt_inet.h" 32 #include "opt_inet6.h" 33 #include "opt_acpi.h" 34 35 #include <sys/param.h> 36 #include <sys/types.h> 37 #include <sys/bus.h> 38 #include <sys/eventhandler.h> 39 #include <sys/sockio.h> 40 #include <sys/kernel.h> 41 #include <sys/lock.h> 42 #include <sys/mutex.h> 43 #include <sys/module.h> 44 #include <sys/kobj.h> 45 #include <sys/rman.h> 46 #include <sys/sbuf.h> 47 #include <sys/smp.h> 48 #include <sys/socket.h> 49 #include <sys/sysctl.h> 50 #include <sys/syslog.h> 51 #include <sys/taskqueue.h> 52 #include <sys/limits.h> 53 54 55 #include <net/if.h> 56 #include <net/if_var.h> 57 #include <net/if_types.h> 58 #include <net/if_media.h> 59 #include <net/bpf.h> 60 #include <net/ethernet.h> 61 #include <net/mp_ring.h> 62 63 #include <netinet/in.h> 64 #include <netinet/in_pcb.h> 65 #include <netinet/tcp_lro.h> 66 #include <netinet/in_systm.h> 67 #include <netinet/if_ether.h> 68 #include <netinet/ip.h> 69 #include <netinet/ip6.h> 70 #include <netinet/tcp.h> 71 72 #include <machine/bus.h> 73 #include <machine/in_cksum.h> 74 75 #include <vm/vm.h> 76 #include <vm/pmap.h> 77 78 #include <dev/led/led.h> 79 #include <dev/pci/pcireg.h> 80 #include <dev/pci/pcivar.h> 81 #include <dev/pci/pci_private.h> 82 83 #include <net/iflib.h> 84 85 #include "ifdi_if.h" 86 87 #if defined(__i386__) || defined(__amd64__) 88 #include <sys/memdesc.h> 89 #include <machine/bus.h> 90 #include <machine/md_var.h> 91 #include <machine/specialreg.h> 92 #include <x86/include/busdma_impl.h> 93 #include <x86/iommu/busdma_dmar.h> 94 #endif 95 96 97 /* 98 * enable accounting of every mbuf as it comes in to and goes out of iflib's software descriptor references 99 */ 100 #define MEMORY_LOGGING 0 101 /* 102 * Enable mbuf vectors for compressing long mbuf chains 103 */ 104 105 /* 106 * NB: 107 * - Prefetching in tx cleaning should perhaps be a tunable. The distance ahead 108 * we prefetch needs to be determined by the time spent in m_free vis a vis 109 * the cost of a prefetch. This will of course vary based on the workload: 110 * - NFLX's m_free path is dominated by vm-based M_EXT manipulation which 111 * is quite expensive, thus suggesting very little prefetch. 112 * - small packet forwarding which is just returning a single mbuf to 113 * UMA will typically be very fast vis a vis the cost of a memory 114 * access. 115 */ 116 117 118 /* 119 * File organization: 120 * - private structures 121 * - iflib private utility functions 122 * - ifnet functions 123 * - vlan registry and other exported functions 124 * - iflib public core functions 125 * 126 * 127 */ 128 static MALLOC_DEFINE(M_IFLIB, "iflib", "ifnet library"); 129 130 struct iflib_txq; 131 typedef struct iflib_txq *iflib_txq_t; 132 struct iflib_rxq; 133 typedef struct iflib_rxq *iflib_rxq_t; 134 struct iflib_fl; 135 typedef struct iflib_fl *iflib_fl_t; 136 137 typedef struct iflib_filter_info { 138 driver_filter_t *ifi_filter; 139 void *ifi_filter_arg; 140 struct grouptask *ifi_task; 141 } *iflib_filter_info_t; 142 143 struct iflib_ctx { 144 KOBJ_FIELDS; 145 /* 146 * Pointer to hardware driver's softc 147 */ 148 void *ifc_softc; 149 device_t ifc_dev; 150 if_t ifc_ifp; 151 152 cpuset_t ifc_cpus; 153 if_shared_ctx_t ifc_sctx; 154 struct if_softc_ctx ifc_softc_ctx; 155 156 struct mtx ifc_mtx; 157 158 uint16_t ifc_nhwtxqs; 159 uint16_t ifc_nhwrxqs; 160 161 iflib_txq_t ifc_txqs; 162 iflib_rxq_t ifc_rxqs; 163 uint32_t ifc_if_flags; 164 uint32_t ifc_flags; 165 uint32_t ifc_max_fl_buf_size; 166 int ifc_in_detach; 167 168 int ifc_link_state; 169 int ifc_link_irq; 170 int ifc_pause_frames; 171 int ifc_watchdog_events; 172 struct cdev *ifc_led_dev; 173 struct resource *ifc_msix_mem; 174 175 struct if_irq ifc_legacy_irq; 176 struct grouptask ifc_admin_task; 177 struct grouptask ifc_vflr_task; 178 struct iflib_filter_info ifc_filter_info; 179 struct ifmedia ifc_media; 180 181 struct sysctl_oid *ifc_sysctl_node; 182 uint16_t ifc_sysctl_ntxqs; 183 uint16_t ifc_sysctl_nrxqs; 184 uint16_t ifc_sysctl_qs_eq_override; 185 186 uint16_t ifc_sysctl_ntxds[8]; 187 uint16_t ifc_sysctl_nrxds[8]; 188 struct if_txrx ifc_txrx; 189 #define isc_txd_encap ifc_txrx.ift_txd_encap 190 #define isc_txd_flush ifc_txrx.ift_txd_flush 191 #define isc_txd_credits_update ifc_txrx.ift_txd_credits_update 192 #define isc_rxd_available ifc_txrx.ift_rxd_available 193 #define isc_rxd_pkt_get ifc_txrx.ift_rxd_pkt_get 194 #define isc_rxd_refill ifc_txrx.ift_rxd_refill 195 #define isc_rxd_flush ifc_txrx.ift_rxd_flush 196 #define isc_rxd_refill ifc_txrx.ift_rxd_refill 197 #define isc_rxd_refill ifc_txrx.ift_rxd_refill 198 #define isc_legacy_intr ifc_txrx.ift_legacy_intr 199 eventhandler_tag ifc_vlan_attach_event; 200 eventhandler_tag ifc_vlan_detach_event; 201 uint8_t ifc_mac[ETHER_ADDR_LEN]; 202 char ifc_mtx_name[16]; 203 }; 204 205 206 void * 207 iflib_get_softc(if_ctx_t ctx) 208 { 209 210 return (ctx->ifc_softc); 211 } 212 213 device_t 214 iflib_get_dev(if_ctx_t ctx) 215 { 216 217 return (ctx->ifc_dev); 218 } 219 220 if_t 221 iflib_get_ifp(if_ctx_t ctx) 222 { 223 224 return (ctx->ifc_ifp); 225 } 226 227 struct ifmedia * 228 iflib_get_media(if_ctx_t ctx) 229 { 230 231 return (&ctx->ifc_media); 232 } 233 234 void 235 iflib_set_mac(if_ctx_t ctx, uint8_t mac[ETHER_ADDR_LEN]) 236 { 237 238 bcopy(mac, ctx->ifc_mac, ETHER_ADDR_LEN); 239 } 240 241 if_softc_ctx_t 242 iflib_get_softc_ctx(if_ctx_t ctx) 243 { 244 245 return (&ctx->ifc_softc_ctx); 246 } 247 248 if_shared_ctx_t 249 iflib_get_sctx(if_ctx_t ctx) 250 { 251 252 return (ctx->ifc_sctx); 253 } 254 255 #define CACHE_PTR_INCREMENT (CACHE_LINE_SIZE/sizeof(void*)) 256 257 #define LINK_ACTIVE(ctx) ((ctx)->ifc_link_state == LINK_STATE_UP) 258 #define CTX_IS_VF(ctx) ((ctx)->ifc_sctx->isc_flags & IFLIB_IS_VF) 259 260 #define RX_SW_DESC_MAP_CREATED (1 << 0) 261 #define TX_SW_DESC_MAP_CREATED (1 << 1) 262 #define RX_SW_DESC_INUSE (1 << 3) 263 #define TX_SW_DESC_MAPPED (1 << 4) 264 265 typedef struct iflib_sw_rx_desc { 266 bus_dmamap_t ifsd_map; /* bus_dma map for packet */ 267 struct mbuf *ifsd_m; /* rx: uninitialized mbuf */ 268 caddr_t ifsd_cl; /* direct cluster pointer for rx */ 269 uint16_t ifsd_flags; 270 } *iflib_rxsd_t; 271 272 typedef struct iflib_sw_tx_desc_val { 273 bus_dmamap_t ifsd_map; /* bus_dma map for packet */ 274 struct mbuf *ifsd_m; /* pkthdr mbuf */ 275 uint8_t ifsd_flags; 276 } *iflib_txsd_val_t; 277 278 typedef struct iflib_sw_tx_desc_array { 279 bus_dmamap_t *ifsd_map; /* bus_dma maps for packet */ 280 struct mbuf **ifsd_m; /* pkthdr mbufs */ 281 uint8_t *ifsd_flags; 282 } iflib_txsd_array_t; 283 284 285 /* magic number that should be high enough for any hardware */ 286 #define IFLIB_MAX_TX_SEGS 128 287 #define IFLIB_MAX_RX_SEGS 32 288 #define IFLIB_RX_COPY_THRESH 128 289 #define IFLIB_MAX_RX_REFRESH 32 290 #define IFLIB_QUEUE_IDLE 0 291 #define IFLIB_QUEUE_HUNG 1 292 #define IFLIB_QUEUE_WORKING 2 293 294 /* this should really scale with ring size - 32 is a fairly arbitrary value for this */ 295 #define TX_BATCH_SIZE 16 296 297 #define IFLIB_RESTART_BUDGET 8 298 299 #define IFC_LEGACY 0x01 300 #define IFC_QFLUSH 0x02 301 #define IFC_MULTISEG 0x04 302 #define IFC_DMAR 0x08 303 #define IFC_SC_ALLOCATED 0x10 304 305 #define CSUM_OFFLOAD (CSUM_IP_TSO|CSUM_IP6_TSO|CSUM_IP| \ 306 CSUM_IP_UDP|CSUM_IP_TCP|CSUM_IP_SCTP| \ 307 CSUM_IP6_UDP|CSUM_IP6_TCP|CSUM_IP6_SCTP) 308 struct iflib_txq { 309 uint16_t ift_in_use; 310 uint16_t ift_cidx; 311 uint16_t ift_cidx_processed; 312 uint16_t ift_pidx; 313 uint8_t ift_gen; 314 uint8_t ift_db_pending; 315 uint8_t ift_db_pending_queued; 316 uint8_t ift_npending; 317 uint8_t ift_br_offset; 318 /* implicit pad */ 319 uint64_t ift_processed; 320 uint64_t ift_cleaned; 321 #if MEMORY_LOGGING 322 uint64_t ift_enqueued; 323 uint64_t ift_dequeued; 324 #endif 325 uint64_t ift_no_tx_dma_setup; 326 uint64_t ift_no_desc_avail; 327 uint64_t ift_mbuf_defrag_failed; 328 uint64_t ift_mbuf_defrag; 329 uint64_t ift_map_failed; 330 uint64_t ift_txd_encap_efbig; 331 uint64_t ift_pullups; 332 333 struct mtx ift_mtx; 334 struct mtx ift_db_mtx; 335 336 /* constant values */ 337 if_ctx_t ift_ctx; 338 struct ifmp_ring **ift_br; 339 struct grouptask ift_task; 340 uint16_t ift_size; 341 uint16_t ift_id; 342 struct callout ift_timer; 343 struct callout ift_db_check; 344 345 iflib_txsd_array_t ift_sds; 346 uint8_t ift_nbr; 347 uint8_t ift_qstatus; 348 uint8_t ift_active; 349 uint8_t ift_closed; 350 int ift_watchdog_time; 351 struct iflib_filter_info ift_filter_info; 352 bus_dma_tag_t ift_desc_tag; 353 bus_dma_tag_t ift_tso_desc_tag; 354 iflib_dma_info_t ift_ifdi; 355 #define MTX_NAME_LEN 16 356 char ift_mtx_name[MTX_NAME_LEN]; 357 char ift_db_mtx_name[MTX_NAME_LEN]; 358 bus_dma_segment_t ift_segs[IFLIB_MAX_TX_SEGS] __aligned(CACHE_LINE_SIZE); 359 } __aligned(CACHE_LINE_SIZE); 360 361 struct iflib_fl { 362 uint16_t ifl_cidx; 363 uint16_t ifl_pidx; 364 uint16_t ifl_credits; 365 uint8_t ifl_gen; 366 #if MEMORY_LOGGING 367 uint64_t ifl_m_enqueued; 368 uint64_t ifl_m_dequeued; 369 uint64_t ifl_cl_enqueued; 370 uint64_t ifl_cl_dequeued; 371 #endif 372 /* implicit pad */ 373 374 /* constant */ 375 uint16_t ifl_size; 376 uint16_t ifl_buf_size; 377 uint16_t ifl_cltype; 378 uma_zone_t ifl_zone; 379 iflib_rxsd_t ifl_sds; 380 iflib_rxq_t ifl_rxq; 381 uint8_t ifl_id; 382 bus_dma_tag_t ifl_desc_tag; 383 iflib_dma_info_t ifl_ifdi; 384 uint64_t ifl_bus_addrs[IFLIB_MAX_RX_REFRESH] __aligned(CACHE_LINE_SIZE); 385 caddr_t ifl_vm_addrs[IFLIB_MAX_RX_REFRESH]; 386 } __aligned(CACHE_LINE_SIZE); 387 388 static inline int 389 get_inuse(int size, int cidx, int pidx, int gen) 390 { 391 int used; 392 393 if (pidx > cidx) 394 used = pidx - cidx; 395 else if (pidx < cidx) 396 used = size - cidx + pidx; 397 else if (gen == 0 && pidx == cidx) 398 used = 0; 399 else if (gen == 1 && pidx == cidx) 400 used = size; 401 else 402 panic("bad state"); 403 404 return (used); 405 } 406 407 #define TXQ_AVAIL(txq) (txq->ift_size - get_inuse(txq->ift_size, txq->ift_cidx, txq->ift_pidx, txq->ift_gen)) 408 409 #define IDXDIFF(head, tail, wrap) \ 410 ((head) >= (tail) ? (head) - (tail) : (wrap) - (tail) + (head)) 411 412 struct iflib_rxq { 413 /* If there is a separate completion queue - 414 * these are the cq cidx and pidx. Otherwise 415 * these are unused. 416 */ 417 uint16_t ifr_size; 418 uint16_t ifr_cq_cidx; 419 uint16_t ifr_cq_pidx; 420 uint8_t ifr_cq_gen; 421 uint8_t ifr_fl_offset; 422 423 if_ctx_t ifr_ctx; 424 iflib_fl_t ifr_fl; 425 uint64_t ifr_rx_irq; 426 uint16_t ifr_id; 427 uint8_t ifr_lro_enabled; 428 uint8_t ifr_nfl; 429 struct lro_ctrl ifr_lc; 430 struct grouptask ifr_task; 431 struct iflib_filter_info ifr_filter_info; 432 iflib_dma_info_t ifr_ifdi; 433 /* dynamically allocate if any drivers need a value substantially larger than this */ 434 struct if_rxd_frag ifr_frags[IFLIB_MAX_RX_SEGS] __aligned(CACHE_LINE_SIZE); 435 } __aligned(CACHE_LINE_SIZE); 436 437 /* 438 * Only allow a single packet to take up most 1/nth of the tx ring 439 */ 440 #define MAX_SINGLE_PACKET_FRACTION 12 441 #define IF_BAD_DMA (bus_addr_t)-1 442 443 static int enable_msix = 1; 444 445 #define mtx_held(m) (((m)->mtx_lock & ~MTX_FLAGMASK) != (uintptr_t)0) 446 447 448 449 #define CTX_ACTIVE(ctx) ((if_getdrvflags((ctx)->ifc_ifp) & IFF_DRV_RUNNING)) 450 451 #define CTX_LOCK_INIT(_sc, _name) mtx_init(&(_sc)->ifc_mtx, _name, "iflib ctx lock", MTX_DEF) 452 453 #define CTX_LOCK(ctx) mtx_lock(&(ctx)->ifc_mtx) 454 #define CTX_UNLOCK(ctx) mtx_unlock(&(ctx)->ifc_mtx) 455 #define CTX_LOCK_DESTROY(ctx) mtx_destroy(&(ctx)->ifc_mtx) 456 457 458 #define TXDB_LOCK_INIT(txq) mtx_init(&(txq)->ift_db_mtx, (txq)->ift_db_mtx_name, NULL, MTX_DEF) 459 #define TXDB_TRYLOCK(txq) mtx_trylock(&(txq)->ift_db_mtx) 460 #define TXDB_LOCK(txq) mtx_lock(&(txq)->ift_db_mtx) 461 #define TXDB_UNLOCK(txq) mtx_unlock(&(txq)->ift_db_mtx) 462 #define TXDB_LOCK_DESTROY(txq) mtx_destroy(&(txq)->ift_db_mtx) 463 464 #define CALLOUT_LOCK(txq) mtx_lock(&txq->ift_mtx) 465 #define CALLOUT_UNLOCK(txq) mtx_unlock(&txq->ift_mtx) 466 467 468 /* Our boot-time initialization hook */ 469 static int iflib_module_event_handler(module_t, int, void *); 470 471 static moduledata_t iflib_moduledata = { 472 "iflib", 473 iflib_module_event_handler, 474 NULL 475 }; 476 477 DECLARE_MODULE(iflib, iflib_moduledata, SI_SUB_INIT_IF, SI_ORDER_ANY); 478 MODULE_VERSION(iflib, 1); 479 480 MODULE_DEPEND(iflib, pci, 1, 1, 1); 481 MODULE_DEPEND(iflib, ether, 1, 1, 1); 482 483 TASKQGROUP_DEFINE(if_io_tqg, mp_ncpus, 1); 484 TASKQGROUP_DEFINE(if_config_tqg, 1, 1); 485 486 #ifndef IFLIB_DEBUG_COUNTERS 487 #ifdef INVARIANTS 488 #define IFLIB_DEBUG_COUNTERS 1 489 #else 490 #define IFLIB_DEBUG_COUNTERS 0 491 #endif /* !INVARIANTS */ 492 #endif 493 494 static SYSCTL_NODE(_net, OID_AUTO, iflib, CTLFLAG_RD, 0, 495 "iflib driver parameters"); 496 497 /* 498 * XXX need to ensure that this can't accidentally cause the head to be moved backwards 499 */ 500 static int iflib_min_tx_latency = 0; 501 502 SYSCTL_INT(_net_iflib, OID_AUTO, min_tx_latency, CTLFLAG_RW, 503 &iflib_min_tx_latency, 0, "minimize transmit latency at the possibel expense of throughput"); 504 505 506 #if IFLIB_DEBUG_COUNTERS 507 508 static int iflib_tx_seen; 509 static int iflib_tx_sent; 510 static int iflib_tx_encap; 511 static int iflib_rx_allocs; 512 static int iflib_fl_refills; 513 static int iflib_fl_refills_large; 514 static int iflib_tx_frees; 515 516 SYSCTL_INT(_net_iflib, OID_AUTO, tx_seen, CTLFLAG_RD, 517 &iflib_tx_seen, 0, "# tx mbufs seen"); 518 SYSCTL_INT(_net_iflib, OID_AUTO, tx_sent, CTLFLAG_RD, 519 &iflib_tx_sent, 0, "# tx mbufs sent"); 520 SYSCTL_INT(_net_iflib, OID_AUTO, tx_encap, CTLFLAG_RD, 521 &iflib_tx_encap, 0, "# tx mbufs encapped"); 522 SYSCTL_INT(_net_iflib, OID_AUTO, tx_frees, CTLFLAG_RD, 523 &iflib_tx_frees, 0, "# tx frees"); 524 SYSCTL_INT(_net_iflib, OID_AUTO, rx_allocs, CTLFLAG_RD, 525 &iflib_rx_allocs, 0, "# rx allocations"); 526 SYSCTL_INT(_net_iflib, OID_AUTO, fl_refills, CTLFLAG_RD, 527 &iflib_fl_refills, 0, "# refills"); 528 SYSCTL_INT(_net_iflib, OID_AUTO, fl_refills_large, CTLFLAG_RD, 529 &iflib_fl_refills_large, 0, "# large refills"); 530 531 532 static int iflib_txq_drain_flushing; 533 static int iflib_txq_drain_oactive; 534 static int iflib_txq_drain_notready; 535 static int iflib_txq_drain_encapfail; 536 537 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_flushing, CTLFLAG_RD, 538 &iflib_txq_drain_flushing, 0, "# drain flushes"); 539 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_oactive, CTLFLAG_RD, 540 &iflib_txq_drain_oactive, 0, "# drain oactives"); 541 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_notready, CTLFLAG_RD, 542 &iflib_txq_drain_notready, 0, "# drain notready"); 543 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_encapfail, CTLFLAG_RD, 544 &iflib_txq_drain_encapfail, 0, "# drain encap fails"); 545 546 547 static int iflib_encap_load_mbuf_fail; 548 static int iflib_encap_txq_avail_fail; 549 static int iflib_encap_txd_encap_fail; 550 551 SYSCTL_INT(_net_iflib, OID_AUTO, encap_load_mbuf_fail, CTLFLAG_RD, 552 &iflib_encap_load_mbuf_fail, 0, "# busdma load failures"); 553 SYSCTL_INT(_net_iflib, OID_AUTO, encap_txq_avail_fail, CTLFLAG_RD, 554 &iflib_encap_txq_avail_fail, 0, "# txq avail failures"); 555 SYSCTL_INT(_net_iflib, OID_AUTO, encap_txd_encap_fail, CTLFLAG_RD, 556 &iflib_encap_txd_encap_fail, 0, "# driver encap failures"); 557 558 static int iflib_task_fn_rxs; 559 static int iflib_rx_intr_enables; 560 static int iflib_fast_intrs; 561 static int iflib_intr_link; 562 static int iflib_intr_msix; 563 static int iflib_rx_unavail; 564 static int iflib_rx_ctx_inactive; 565 static int iflib_rx_zero_len; 566 static int iflib_rx_if_input; 567 static int iflib_rx_mbuf_null; 568 static int iflib_rxd_flush; 569 570 static int iflib_verbose_debug; 571 572 SYSCTL_INT(_net_iflib, OID_AUTO, intr_link, CTLFLAG_RD, 573 &iflib_intr_link, 0, "# intr link calls"); 574 SYSCTL_INT(_net_iflib, OID_AUTO, intr_msix, CTLFLAG_RD, 575 &iflib_intr_msix, 0, "# intr msix calls"); 576 SYSCTL_INT(_net_iflib, OID_AUTO, task_fn_rx, CTLFLAG_RD, 577 &iflib_task_fn_rxs, 0, "# task_fn_rx calls"); 578 SYSCTL_INT(_net_iflib, OID_AUTO, rx_intr_enables, CTLFLAG_RD, 579 &iflib_rx_intr_enables, 0, "# rx intr enables"); 580 SYSCTL_INT(_net_iflib, OID_AUTO, fast_intrs, CTLFLAG_RD, 581 &iflib_fast_intrs, 0, "# fast_intr calls"); 582 SYSCTL_INT(_net_iflib, OID_AUTO, rx_unavail, CTLFLAG_RD, 583 &iflib_rx_unavail, 0, "# times rxeof called with no available data"); 584 SYSCTL_INT(_net_iflib, OID_AUTO, rx_ctx_inactive, CTLFLAG_RD, 585 &iflib_rx_ctx_inactive, 0, "# times rxeof called with inactive context"); 586 SYSCTL_INT(_net_iflib, OID_AUTO, rx_zero_len, CTLFLAG_RD, 587 &iflib_rx_zero_len, 0, "# times rxeof saw zero len mbuf"); 588 SYSCTL_INT(_net_iflib, OID_AUTO, rx_if_input, CTLFLAG_RD, 589 &iflib_rx_if_input, 0, "# times rxeof called if_input"); 590 SYSCTL_INT(_net_iflib, OID_AUTO, rx_mbuf_null, CTLFLAG_RD, 591 &iflib_rx_mbuf_null, 0, "# times rxeof got null mbuf"); 592 SYSCTL_INT(_net_iflib, OID_AUTO, rxd_flush, CTLFLAG_RD, 593 &iflib_rxd_flush, 0, "# times rxd_flush called"); 594 SYSCTL_INT(_net_iflib, OID_AUTO, verbose_debug, CTLFLAG_RW, 595 &iflib_verbose_debug, 0, "enable verbose debugging"); 596 597 #define DBG_COUNTER_INC(name) atomic_add_int(&(iflib_ ## name), 1) 598 599 #else 600 #define DBG_COUNTER_INC(name) 601 602 #endif 603 604 605 606 #define IFLIB_DEBUG 0 607 608 static void iflib_tx_structures_free(if_ctx_t ctx); 609 static void iflib_rx_structures_free(if_ctx_t ctx); 610 static int iflib_queues_alloc(if_ctx_t ctx); 611 static int iflib_tx_credits_update(if_ctx_t ctx, iflib_txq_t txq); 612 static int iflib_rxd_avail(if_ctx_t ctx, iflib_rxq_t rxq, int cidx, int budget); 613 static int iflib_qset_structures_setup(if_ctx_t ctx); 614 static int iflib_msix_init(if_ctx_t ctx); 615 static int iflib_legacy_setup(if_ctx_t ctx, driver_filter_t filter, void *filterarg, int *rid, char *str); 616 static void iflib_txq_check_drain(iflib_txq_t txq, int budget); 617 static uint32_t iflib_txq_can_drain(struct ifmp_ring *); 618 static int iflib_register(if_ctx_t); 619 static void iflib_init_locked(if_ctx_t ctx); 620 static void iflib_add_device_sysctl_pre(if_ctx_t ctx); 621 static void iflib_add_device_sysctl_post(if_ctx_t ctx); 622 623 624 #ifdef DEV_NETMAP 625 #include <sys/selinfo.h> 626 #include <net/netmap.h> 627 #include <dev/netmap/netmap_kern.h> 628 629 MODULE_DEPEND(iflib, netmap, 1, 1, 1); 630 631 /* 632 * device-specific sysctl variables: 633 * 634 * iflib_crcstrip: 0: keep CRC in rx frames (default), 1: strip it. 635 * During regular operations the CRC is stripped, but on some 636 * hardware reception of frames not multiple of 64 is slower, 637 * so using crcstrip=0 helps in benchmarks. 638 * 639 * iflib_rx_miss, iflib_rx_miss_bufs: 640 * count packets that might be missed due to lost interrupts. 641 */ 642 SYSCTL_DECL(_dev_netmap); 643 /* 644 * The xl driver by default strips CRCs and we do not override it. 645 */ 646 647 int iflib_crcstrip = 1; 648 SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_crcstrip, 649 CTLFLAG_RW, &iflib_crcstrip, 1, "strip CRC on rx frames"); 650 651 int iflib_rx_miss, iflib_rx_miss_bufs; 652 SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_rx_miss, 653 CTLFLAG_RW, &iflib_rx_miss, 0, "potentially missed rx intr"); 654 SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_rx_miss_bufs, 655 CTLFLAG_RW, &iflib_rx_miss_bufs, 0, "potentially missed rx intr bufs"); 656 657 /* 658 * Register/unregister. We are already under netmap lock. 659 * Only called on the first register or the last unregister. 660 */ 661 static int 662 iflib_netmap_register(struct netmap_adapter *na, int onoff) 663 { 664 struct ifnet *ifp = na->ifp; 665 if_ctx_t ctx = ifp->if_softc; 666 667 CTX_LOCK(ctx); 668 IFDI_INTR_DISABLE(ctx); 669 670 /* Tell the stack that the interface is no longer active */ 671 ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); 672 673 if (!CTX_IS_VF(ctx)) 674 IFDI_CRCSTRIP_SET(ctx, onoff); 675 676 /* enable or disable flags and callbacks in na and ifp */ 677 if (onoff) { 678 nm_set_native_flags(na); 679 } else { 680 nm_clear_native_flags(na); 681 } 682 IFDI_INIT(ctx); 683 IFDI_CRCSTRIP_SET(ctx, onoff); // XXX why twice ? 684 CTX_UNLOCK(ctx); 685 return (ifp->if_drv_flags & IFF_DRV_RUNNING ? 0 : 1); 686 } 687 688 /* 689 * Reconcile kernel and user view of the transmit ring. 690 * 691 * All information is in the kring. 692 * Userspace wants to send packets up to the one before kring->rhead, 693 * kernel knows kring->nr_hwcur is the first unsent packet. 694 * 695 * Here we push packets out (as many as possible), and possibly 696 * reclaim buffers from previously completed transmission. 697 * 698 * The caller (netmap) guarantees that there is only one instance 699 * running at any time. Any interference with other driver 700 * methods should be handled by the individual drivers. 701 */ 702 static int 703 iflib_netmap_txsync(struct netmap_kring *kring, int flags) 704 { 705 struct netmap_adapter *na = kring->na; 706 struct ifnet *ifp = na->ifp; 707 struct netmap_ring *ring = kring->ring; 708 u_int nm_i; /* index into the netmap ring */ 709 u_int nic_i; /* index into the NIC ring */ 710 u_int n; 711 u_int const lim = kring->nkr_num_slots - 1; 712 u_int const head = kring->rhead; 713 struct if_pkt_info pi; 714 715 /* 716 * interrupts on every tx packet are expensive so request 717 * them every half ring, or where NS_REPORT is set 718 */ 719 u_int report_frequency = kring->nkr_num_slots >> 1; 720 /* device-specific */ 721 if_ctx_t ctx = ifp->if_softc; 722 iflib_txq_t txq = &ctx->ifc_txqs[kring->ring_id]; 723 724 pi.ipi_segs = txq->ift_segs; 725 pi.ipi_qsidx = kring->ring_id; 726 pi.ipi_ndescs = 0; 727 728 bus_dmamap_sync(txq->ift_desc_tag, txq->ift_ifdi->idi_map, 729 BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); 730 731 732 /* 733 * First part: process new packets to send. 734 * nm_i is the current index in the netmap ring, 735 * nic_i is the corresponding index in the NIC ring. 736 * 737 * If we have packets to send (nm_i != head) 738 * iterate over the netmap ring, fetch length and update 739 * the corresponding slot in the NIC ring. Some drivers also 740 * need to update the buffer's physical address in the NIC slot 741 * even NS_BUF_CHANGED is not set (PNMB computes the addresses). 742 * 743 * The netmap_reload_map() calls is especially expensive, 744 * even when (as in this case) the tag is 0, so do only 745 * when the buffer has actually changed. 746 * 747 * If possible do not set the report/intr bit on all slots, 748 * but only a few times per ring or when NS_REPORT is set. 749 * 750 * Finally, on 10G and faster drivers, it might be useful 751 * to prefetch the next slot and txr entry. 752 */ 753 754 nm_i = kring->nr_hwcur; 755 if (nm_i != head) { /* we have new packets to send */ 756 nic_i = netmap_idx_k2n(kring, nm_i); 757 758 __builtin_prefetch(&ring->slot[nm_i]); 759 __builtin_prefetch(&txq->ift_sds.ifsd_m[nic_i]); 760 __builtin_prefetch(&txq->ift_sds.ifsd_map[nic_i]); 761 762 for (n = 0; nm_i != head; n++) { 763 struct netmap_slot *slot = &ring->slot[nm_i]; 764 u_int len = slot->len; 765 uint64_t paddr; 766 void *addr = PNMB(na, slot, &paddr); 767 int flags = (slot->flags & NS_REPORT || 768 nic_i == 0 || nic_i == report_frequency) ? 769 IPI_TX_INTR : 0; 770 771 /* device-specific */ 772 pi.ipi_pidx = nic_i; 773 pi.ipi_flags = flags; 774 775 /* Fill the slot in the NIC ring. */ 776 ctx->isc_txd_encap(ctx->ifc_softc, &pi); 777 778 /* prefetch for next round */ 779 __builtin_prefetch(&ring->slot[nm_i + 1]); 780 __builtin_prefetch(&txq->ift_sds.ifsd_m[nic_i + 1]); 781 __builtin_prefetch(&txq->ift_sds.ifsd_map[nic_i + 1]); 782 783 NM_CHECK_ADDR_LEN(na, addr, len); 784 785 if (slot->flags & NS_BUF_CHANGED) { 786 /* buffer has changed, reload map */ 787 netmap_reload_map(na, txq->ift_desc_tag, txq->ift_sds.ifsd_map[nic_i], addr); 788 } 789 slot->flags &= ~(NS_REPORT | NS_BUF_CHANGED); 790 791 /* make sure changes to the buffer are synced */ 792 bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_sds.ifsd_map[nic_i], 793 BUS_DMASYNC_PREWRITE); 794 795 nm_i = nm_next(nm_i, lim); 796 nic_i = nm_next(nic_i, lim); 797 } 798 kring->nr_hwcur = head; 799 800 /* synchronize the NIC ring */ 801 bus_dmamap_sync(txq->ift_desc_tag, txq->ift_ifdi->idi_map, 802 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); 803 804 /* (re)start the tx unit up to slot nic_i (excluded) */ 805 ctx->isc_txd_flush(ctx->ifc_softc, txq->ift_id, nic_i); 806 } 807 808 /* 809 * Second part: reclaim buffers for completed transmissions. 810 */ 811 if (iflib_tx_credits_update(ctx, txq)) { 812 /* some tx completed, increment avail */ 813 nic_i = txq->ift_cidx_processed; 814 kring->nr_hwtail = nm_prev(netmap_idx_n2k(kring, nic_i), lim); 815 } 816 return (0); 817 } 818 819 /* 820 * Reconcile kernel and user view of the receive ring. 821 * Same as for the txsync, this routine must be efficient. 822 * The caller guarantees a single invocations, but races against 823 * the rest of the driver should be handled here. 824 * 825 * On call, kring->rhead is the first packet that userspace wants 826 * to keep, and kring->rcur is the wakeup point. 827 * The kernel has previously reported packets up to kring->rtail. 828 * 829 * If (flags & NAF_FORCE_READ) also check for incoming packets irrespective 830 * of whether or not we received an interrupt. 831 */ 832 static int 833 iflib_netmap_rxsync(struct netmap_kring *kring, int flags) 834 { 835 struct netmap_adapter *na = kring->na; 836 struct ifnet *ifp = na->ifp; 837 struct netmap_ring *ring = kring->ring; 838 u_int nm_i; /* index into the netmap ring */ 839 u_int nic_i; /* index into the NIC ring */ 840 u_int i, n; 841 u_int const lim = kring->nkr_num_slots - 1; 842 u_int const head = kring->rhead; 843 int force_update = (flags & NAF_FORCE_READ) || kring->nr_kflags & NKR_PENDINTR; 844 struct if_rxd_info ri; 845 /* device-specific */ 846 if_ctx_t ctx = ifp->if_softc; 847 iflib_rxq_t rxq = &ctx->ifc_rxqs[kring->ring_id]; 848 iflib_fl_t fl = rxq->ifr_fl; 849 if (head > lim) 850 return netmap_ring_reinit(kring); 851 852 bzero(&ri, sizeof(ri)); 853 ri.iri_qsidx = kring->ring_id; 854 ri.iri_ifp = ctx->ifc_ifp; 855 /* XXX check sync modes */ 856 for (i = 0, fl = rxq->ifr_fl; i < rxq->ifr_nfl; i++, fl++) 857 bus_dmamap_sync(rxq->ifr_fl[i].ifl_desc_tag, fl->ifl_ifdi->idi_map, 858 BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); 859 860 /* 861 * First part: import newly received packets. 862 * 863 * nm_i is the index of the next free slot in the netmap ring, 864 * nic_i is the index of the next received packet in the NIC ring, 865 * and they may differ in case if_init() has been called while 866 * in netmap mode. For the receive ring we have 867 * 868 * nic_i = rxr->next_check; 869 * nm_i = kring->nr_hwtail (previous) 870 * and 871 * nm_i == (nic_i + kring->nkr_hwofs) % ring_size 872 * 873 * rxr->next_check is set to 0 on a ring reinit 874 */ 875 if (netmap_no_pendintr || force_update) { 876 int crclen = iflib_crcstrip ? 0 : 4; 877 int error, avail; 878 uint16_t slot_flags = kring->nkr_slot_flags; 879 880 for (fl = rxq->ifr_fl, i = 0; i < rxq->ifr_nfl; i++, fl++) { 881 nic_i = fl->ifl_cidx; 882 nm_i = netmap_idx_n2k(kring, nic_i); 883 avail = ctx->isc_rxd_available(ctx->ifc_softc, kring->ring_id, nic_i, INT_MAX); 884 for (n = 0; avail > 0; n++, avail--) { 885 error = ctx->isc_rxd_pkt_get(ctx->ifc_softc, &ri); 886 if (error) 887 ring->slot[nm_i].len = 0; 888 else 889 ring->slot[nm_i].len = ri.iri_len - crclen; 890 ring->slot[nm_i].flags = slot_flags; 891 bus_dmamap_sync(fl->ifl_ifdi->idi_tag, 892 fl->ifl_sds[nic_i].ifsd_map, BUS_DMASYNC_POSTREAD); 893 nm_i = nm_next(nm_i, lim); 894 nic_i = nm_next(nic_i, lim); 895 } 896 if (n) { /* update the state variables */ 897 if (netmap_no_pendintr && !force_update) { 898 /* diagnostics */ 899 iflib_rx_miss ++; 900 iflib_rx_miss_bufs += n; 901 } 902 fl->ifl_cidx = nic_i; 903 kring->nr_hwtail = nm_i; 904 } 905 kring->nr_kflags &= ~NKR_PENDINTR; 906 } 907 } 908 /* 909 * Second part: skip past packets that userspace has released. 910 * (kring->nr_hwcur to head excluded), 911 * and make the buffers available for reception. 912 * As usual nm_i is the index in the netmap ring, 913 * nic_i is the index in the NIC ring, and 914 * nm_i == (nic_i + kring->nkr_hwofs) % ring_size 915 */ 916 /* XXX not sure how this will work with multiple free lists */ 917 nm_i = kring->nr_hwcur; 918 if (nm_i != head) { 919 nic_i = netmap_idx_k2n(kring, nm_i); 920 for (n = 0; nm_i != head; n++) { 921 struct netmap_slot *slot = &ring->slot[nm_i]; 922 uint64_t paddr; 923 caddr_t vaddr; 924 void *addr = PNMB(na, slot, &paddr); 925 926 if (addr == NETMAP_BUF_BASE(na)) /* bad buf */ 927 goto ring_reset; 928 929 vaddr = addr; 930 if (slot->flags & NS_BUF_CHANGED) { 931 /* buffer has changed, reload map */ 932 netmap_reload_map(na, fl->ifl_ifdi->idi_tag, fl->ifl_sds[nic_i].ifsd_map, addr); 933 slot->flags &= ~NS_BUF_CHANGED; 934 } 935 /* 936 * XXX we should be batching this operation - TODO 937 */ 938 ctx->isc_rxd_refill(ctx->ifc_softc, rxq->ifr_id, fl->ifl_id, nic_i, &paddr, &vaddr, 1, fl->ifl_buf_size); 939 bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_sds[nic_i].ifsd_map, 940 BUS_DMASYNC_PREREAD); 941 nm_i = nm_next(nm_i, lim); 942 nic_i = nm_next(nic_i, lim); 943 } 944 kring->nr_hwcur = head; 945 946 bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map, 947 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); 948 /* 949 * IMPORTANT: we must leave one free slot in the ring, 950 * so move nic_i back by one unit 951 */ 952 nic_i = nm_prev(nic_i, lim); 953 ctx->isc_rxd_flush(ctx->ifc_softc, rxq->ifr_id, fl->ifl_id, nic_i); 954 } 955 956 return 0; 957 958 ring_reset: 959 return netmap_ring_reinit(kring); 960 } 961 962 static int 963 iflib_netmap_attach(if_ctx_t ctx) 964 { 965 struct netmap_adapter na; 966 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 967 968 bzero(&na, sizeof(na)); 969 970 na.ifp = ctx->ifc_ifp; 971 na.na_flags = NAF_BDG_MAYSLEEP; 972 MPASS(ctx->ifc_softc_ctx.isc_ntxqsets); 973 MPASS(ctx->ifc_softc_ctx.isc_nrxqsets); 974 975 na.num_tx_desc = scctx->isc_ntxd[0]; 976 na.num_rx_desc = scctx->isc_nrxd[0]; 977 na.nm_txsync = iflib_netmap_txsync; 978 na.nm_rxsync = iflib_netmap_rxsync; 979 na.nm_register = iflib_netmap_register; 980 na.num_tx_rings = ctx->ifc_softc_ctx.isc_ntxqsets; 981 na.num_rx_rings = ctx->ifc_softc_ctx.isc_nrxqsets; 982 return (netmap_attach(&na)); 983 } 984 985 static void 986 iflib_netmap_txq_init(if_ctx_t ctx, iflib_txq_t txq) 987 { 988 struct netmap_adapter *na = NA(ctx->ifc_ifp); 989 struct netmap_slot *slot; 990 991 slot = netmap_reset(na, NR_TX, txq->ift_id, 0); 992 if (slot == 0) 993 return; 994 995 for (int i = 0; i < ctx->ifc_softc_ctx.isc_ntxd[0]; i++) { 996 997 /* 998 * In netmap mode, set the map for the packet buffer. 999 * NOTE: Some drivers (not this one) also need to set 1000 * the physical buffer address in the NIC ring. 1001 * netmap_idx_n2k() maps a nic index, i, into the corresponding 1002 * netmap slot index, si 1003 */ 1004 int si = netmap_idx_n2k(&na->tx_rings[txq->ift_id], i); 1005 netmap_load_map(na, txq->ift_desc_tag, txq->ift_sds.ifsd_map[i], NMB(na, slot + si)); 1006 } 1007 } 1008 static void 1009 iflib_netmap_rxq_init(if_ctx_t ctx, iflib_rxq_t rxq) 1010 { 1011 struct netmap_adapter *na = NA(ctx->ifc_ifp); 1012 struct netmap_slot *slot; 1013 iflib_rxsd_t sd; 1014 int nrxd; 1015 1016 slot = netmap_reset(na, NR_RX, rxq->ifr_id, 0); 1017 if (slot == 0) 1018 return; 1019 sd = rxq->ifr_fl[0].ifl_sds; 1020 nrxd = ctx->ifc_softc_ctx.isc_nrxd[0]; 1021 for (int i = 0; i < nrxd; i++, sd++) { 1022 int sj = netmap_idx_n2k(&na->rx_rings[rxq->ifr_id], i); 1023 uint64_t paddr; 1024 void *addr; 1025 caddr_t vaddr; 1026 1027 vaddr = addr = PNMB(na, slot + sj, &paddr); 1028 netmap_load_map(na, rxq->ifr_fl[0].ifl_ifdi->idi_tag, sd->ifsd_map, addr); 1029 /* Update descriptor and the cached value */ 1030 ctx->isc_rxd_refill(ctx->ifc_softc, rxq->ifr_id, 0 /* fl_id */, i, &paddr, &vaddr, 1, rxq->ifr_fl[0].ifl_buf_size); 1031 } 1032 /* preserve queue */ 1033 if (ctx->ifc_ifp->if_capenable & IFCAP_NETMAP) { 1034 struct netmap_kring *kring = &na->rx_rings[rxq->ifr_id]; 1035 int t = na->num_rx_desc - 1 - nm_kr_rxspace(kring); 1036 ctx->isc_rxd_flush(ctx->ifc_softc, rxq->ifr_id, 0 /* fl_id */, t); 1037 } else 1038 ctx->isc_rxd_flush(ctx->ifc_softc, rxq->ifr_id, 0 /* fl_id */, nrxd-1); 1039 } 1040 1041 #define iflib_netmap_detach(ifp) netmap_detach(ifp) 1042 1043 #else 1044 #define iflib_netmap_txq_init(ctx, txq) 1045 #define iflib_netmap_rxq_init(ctx, rxq) 1046 #define iflib_netmap_detach(ifp) 1047 1048 #define iflib_netmap_attach(ctx) (0) 1049 #define netmap_rx_irq(ifp, qid, budget) (0) 1050 1051 #endif 1052 1053 #if defined(__i386__) || defined(__amd64__) 1054 static __inline void 1055 prefetch(void *x) 1056 { 1057 __asm volatile("prefetcht0 %0" :: "m" (*(unsigned long *)x)); 1058 } 1059 #else 1060 #define prefetch(x) 1061 #endif 1062 1063 static void 1064 _iflib_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int err) 1065 { 1066 if (err) 1067 return; 1068 *(bus_addr_t *) arg = segs[0].ds_addr; 1069 } 1070 1071 int 1072 iflib_dma_alloc(if_ctx_t ctx, int size, iflib_dma_info_t dma, int mapflags) 1073 { 1074 int err; 1075 if_shared_ctx_t sctx = ctx->ifc_sctx; 1076 device_t dev = ctx->ifc_dev; 1077 1078 KASSERT(sctx->isc_q_align != 0, ("alignment value not initialized")); 1079 1080 err = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */ 1081 sctx->isc_q_align, 0, /* alignment, bounds */ 1082 BUS_SPACE_MAXADDR, /* lowaddr */ 1083 BUS_SPACE_MAXADDR, /* highaddr */ 1084 NULL, NULL, /* filter, filterarg */ 1085 size, /* maxsize */ 1086 1, /* nsegments */ 1087 size, /* maxsegsize */ 1088 BUS_DMA_ALLOCNOW, /* flags */ 1089 NULL, /* lockfunc */ 1090 NULL, /* lockarg */ 1091 &dma->idi_tag); 1092 if (err) { 1093 device_printf(dev, 1094 "%s: bus_dma_tag_create failed: %d\n", 1095 __func__, err); 1096 goto fail_0; 1097 } 1098 1099 err = bus_dmamem_alloc(dma->idi_tag, (void**) &dma->idi_vaddr, 1100 BUS_DMA_NOWAIT | BUS_DMA_COHERENT | BUS_DMA_ZERO, &dma->idi_map); 1101 if (err) { 1102 device_printf(dev, 1103 "%s: bus_dmamem_alloc(%ju) failed: %d\n", 1104 __func__, (uintmax_t)size, err); 1105 goto fail_1; 1106 } 1107 1108 dma->idi_paddr = IF_BAD_DMA; 1109 err = bus_dmamap_load(dma->idi_tag, dma->idi_map, dma->idi_vaddr, 1110 size, _iflib_dmamap_cb, &dma->idi_paddr, mapflags | BUS_DMA_NOWAIT); 1111 if (err || dma->idi_paddr == IF_BAD_DMA) { 1112 device_printf(dev, 1113 "%s: bus_dmamap_load failed: %d\n", 1114 __func__, err); 1115 goto fail_2; 1116 } 1117 1118 dma->idi_size = size; 1119 return (0); 1120 1121 fail_2: 1122 bus_dmamem_free(dma->idi_tag, dma->idi_vaddr, dma->idi_map); 1123 fail_1: 1124 bus_dma_tag_destroy(dma->idi_tag); 1125 fail_0: 1126 dma->idi_tag = NULL; 1127 1128 return (err); 1129 } 1130 1131 int 1132 iflib_dma_alloc_multi(if_ctx_t ctx, int *sizes, iflib_dma_info_t *dmalist, int mapflags, int count) 1133 { 1134 int i, err; 1135 iflib_dma_info_t *dmaiter; 1136 1137 dmaiter = dmalist; 1138 for (i = 0; i < count; i++, dmaiter++) { 1139 if ((err = iflib_dma_alloc(ctx, sizes[i], *dmaiter, mapflags)) != 0) 1140 break; 1141 } 1142 if (err) 1143 iflib_dma_free_multi(dmalist, i); 1144 return (err); 1145 } 1146 1147 void 1148 iflib_dma_free(iflib_dma_info_t dma) 1149 { 1150 if (dma->idi_tag == NULL) 1151 return; 1152 if (dma->idi_paddr != IF_BAD_DMA) { 1153 bus_dmamap_sync(dma->idi_tag, dma->idi_map, 1154 BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); 1155 bus_dmamap_unload(dma->idi_tag, dma->idi_map); 1156 dma->idi_paddr = IF_BAD_DMA; 1157 } 1158 if (dma->idi_vaddr != NULL) { 1159 bus_dmamem_free(dma->idi_tag, dma->idi_vaddr, dma->idi_map); 1160 dma->idi_vaddr = NULL; 1161 } 1162 bus_dma_tag_destroy(dma->idi_tag); 1163 dma->idi_tag = NULL; 1164 } 1165 1166 void 1167 iflib_dma_free_multi(iflib_dma_info_t *dmalist, int count) 1168 { 1169 int i; 1170 iflib_dma_info_t *dmaiter = dmalist; 1171 1172 for (i = 0; i < count; i++, dmaiter++) 1173 iflib_dma_free(*dmaiter); 1174 } 1175 1176 static int 1177 iflib_fast_intr(void *arg) 1178 { 1179 iflib_filter_info_t info = arg; 1180 struct grouptask *gtask = info->ifi_task; 1181 1182 DBG_COUNTER_INC(fast_intrs); 1183 if (info->ifi_filter != NULL && info->ifi_filter(info->ifi_filter_arg) == FILTER_HANDLED) 1184 return (FILTER_HANDLED); 1185 1186 GROUPTASK_ENQUEUE(gtask); 1187 return (FILTER_HANDLED); 1188 } 1189 1190 static int 1191 _iflib_irq_alloc(if_ctx_t ctx, if_irq_t irq, int rid, 1192 driver_filter_t filter, driver_intr_t handler, void *arg, 1193 char *name) 1194 { 1195 int rc; 1196 struct resource *res; 1197 void *tag; 1198 device_t dev = ctx->ifc_dev; 1199 1200 MPASS(rid < 512); 1201 irq->ii_rid = rid; 1202 res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &irq->ii_rid, 1203 RF_SHAREABLE | RF_ACTIVE); 1204 if (res == NULL) { 1205 device_printf(dev, 1206 "failed to allocate IRQ for rid %d, name %s.\n", rid, name); 1207 return (ENOMEM); 1208 } 1209 irq->ii_res = res; 1210 KASSERT(filter == NULL || handler == NULL, ("filter and handler can't both be non-NULL")); 1211 rc = bus_setup_intr(dev, res, INTR_MPSAFE | INTR_TYPE_NET, 1212 filter, handler, arg, &tag); 1213 if (rc != 0) { 1214 device_printf(dev, 1215 "failed to setup interrupt for rid %d, name %s: %d\n", 1216 rid, name ? name : "unknown", rc); 1217 return (rc); 1218 } else if (name) 1219 bus_describe_intr(dev, res, tag, "%s", name); 1220 1221 irq->ii_tag = tag; 1222 return (0); 1223 } 1224 1225 1226 /********************************************************************* 1227 * 1228 * Allocate memory for tx_buffer structures. The tx_buffer stores all 1229 * the information needed to transmit a packet on the wire. This is 1230 * called only once at attach, setup is done every reset. 1231 * 1232 **********************************************************************/ 1233 1234 static int 1235 iflib_txsd_alloc(iflib_txq_t txq) 1236 { 1237 if_ctx_t ctx = txq->ift_ctx; 1238 if_shared_ctx_t sctx = ctx->ifc_sctx; 1239 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 1240 device_t dev = ctx->ifc_dev; 1241 int err, nsegments, ntsosegments; 1242 1243 nsegments = scctx->isc_tx_nsegments; 1244 ntsosegments = scctx->isc_tx_tso_segments_max; 1245 MPASS(scctx->isc_ntxd[0] > 0); 1246 MPASS(scctx->isc_ntxd[txq->ift_br_offset] > 0); 1247 MPASS(nsegments > 0); 1248 MPASS(ntsosegments > 0); 1249 /* 1250 * Setup DMA descriptor areas. 1251 */ 1252 if ((err = bus_dma_tag_create(bus_get_dma_tag(dev), 1253 1, 0, /* alignment, bounds */ 1254 BUS_SPACE_MAXADDR, /* lowaddr */ 1255 BUS_SPACE_MAXADDR, /* highaddr */ 1256 NULL, NULL, /* filter, filterarg */ 1257 sctx->isc_tx_maxsize, /* maxsize */ 1258 nsegments, /* nsegments */ 1259 sctx->isc_tx_maxsegsize, /* maxsegsize */ 1260 0, /* flags */ 1261 NULL, /* lockfunc */ 1262 NULL, /* lockfuncarg */ 1263 &txq->ift_desc_tag))) { 1264 device_printf(dev,"Unable to allocate TX DMA tag: %d\n", err); 1265 device_printf(dev,"maxsize: %zd nsegments: %d maxsegsize: %zd\n", 1266 sctx->isc_tx_maxsize, nsegments, sctx->isc_tx_maxsegsize); 1267 goto fail; 1268 } 1269 #ifdef IFLIB_DIAGNOSTICS 1270 device_printf(dev,"maxsize: %zd nsegments: %d maxsegsize: %zd\n", 1271 sctx->isc_tx_maxsize, nsegments, sctx->isc_tx_maxsegsize); 1272 1273 #endif 1274 if ((err = bus_dma_tag_create(bus_get_dma_tag(dev), 1275 1, 0, /* alignment, bounds */ 1276 BUS_SPACE_MAXADDR, /* lowaddr */ 1277 BUS_SPACE_MAXADDR, /* highaddr */ 1278 NULL, NULL, /* filter, filterarg */ 1279 scctx->isc_tx_tso_size_max, /* maxsize */ 1280 ntsosegments, /* nsegments */ 1281 scctx->isc_tx_tso_segsize_max, /* maxsegsize */ 1282 0, /* flags */ 1283 NULL, /* lockfunc */ 1284 NULL, /* lockfuncarg */ 1285 &txq->ift_tso_desc_tag))) { 1286 device_printf(dev,"Unable to allocate TX TSO DMA tag: %d\n", err); 1287 1288 goto fail; 1289 } 1290 #ifdef IFLIB_DIAGNOSTICS 1291 device_printf(dev,"TSO maxsize: %d ntsosegments: %d maxsegsize: %d\n", 1292 scctx->isc_tx_tso_size_max, ntsosegments, 1293 scctx->isc_tx_tso_segsize_max); 1294 #endif 1295 if (!(txq->ift_sds.ifsd_flags = 1296 (uint8_t *) malloc(sizeof(uint8_t) * 1297 scctx->isc_ntxd[txq->ift_br_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { 1298 device_printf(dev, "Unable to allocate tx_buffer memory\n"); 1299 err = ENOMEM; 1300 goto fail; 1301 } 1302 if (!(txq->ift_sds.ifsd_m = 1303 (struct mbuf **) malloc(sizeof(struct mbuf *) * 1304 scctx->isc_ntxd[txq->ift_br_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { 1305 device_printf(dev, "Unable to allocate tx_buffer memory\n"); 1306 err = ENOMEM; 1307 goto fail; 1308 } 1309 1310 /* Create the descriptor buffer dma maps */ 1311 #if defined(ACPI_DMAR) || (!(defined(__i386__) && !defined(__amd64__))) 1312 if ((ctx->ifc_flags & IFC_DMAR) == 0) 1313 return (0); 1314 1315 if (!(txq->ift_sds.ifsd_map = 1316 (bus_dmamap_t *) malloc(sizeof(bus_dmamap_t) * scctx->isc_ntxd[txq->ift_br_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { 1317 device_printf(dev, "Unable to allocate tx_buffer map memory\n"); 1318 err = ENOMEM; 1319 goto fail; 1320 } 1321 1322 for (int i = 0; i < scctx->isc_ntxd[txq->ift_br_offset]; i++) { 1323 err = bus_dmamap_create(txq->ift_desc_tag, 0, &txq->ift_sds.ifsd_map[i]); 1324 if (err != 0) { 1325 device_printf(dev, "Unable to create TX DMA map\n"); 1326 goto fail; 1327 } 1328 } 1329 #endif 1330 return (0); 1331 fail: 1332 /* We free all, it handles case where we are in the middle */ 1333 iflib_tx_structures_free(ctx); 1334 return (err); 1335 } 1336 1337 static void 1338 iflib_txsd_destroy(if_ctx_t ctx, iflib_txq_t txq, int i) 1339 { 1340 bus_dmamap_t map; 1341 1342 map = NULL; 1343 if (txq->ift_sds.ifsd_map != NULL) 1344 map = txq->ift_sds.ifsd_map[i]; 1345 if (map != NULL) { 1346 bus_dmamap_unload(txq->ift_desc_tag, map); 1347 bus_dmamap_destroy(txq->ift_desc_tag, map); 1348 txq->ift_sds.ifsd_map[i] = NULL; 1349 } 1350 } 1351 1352 static void 1353 iflib_txq_destroy(iflib_txq_t txq) 1354 { 1355 if_ctx_t ctx = txq->ift_ctx; 1356 1357 for (int i = 0; i < txq->ift_size; i++) 1358 iflib_txsd_destroy(ctx, txq, i); 1359 if (txq->ift_sds.ifsd_map != NULL) { 1360 free(txq->ift_sds.ifsd_map, M_IFLIB); 1361 txq->ift_sds.ifsd_map = NULL; 1362 } 1363 if (txq->ift_sds.ifsd_m != NULL) { 1364 free(txq->ift_sds.ifsd_m, M_IFLIB); 1365 txq->ift_sds.ifsd_m = NULL; 1366 } 1367 if (txq->ift_sds.ifsd_flags != NULL) { 1368 free(txq->ift_sds.ifsd_flags, M_IFLIB); 1369 txq->ift_sds.ifsd_flags = NULL; 1370 } 1371 if (txq->ift_desc_tag != NULL) { 1372 bus_dma_tag_destroy(txq->ift_desc_tag); 1373 txq->ift_desc_tag = NULL; 1374 } 1375 if (txq->ift_tso_desc_tag != NULL) { 1376 bus_dma_tag_destroy(txq->ift_tso_desc_tag); 1377 txq->ift_tso_desc_tag = NULL; 1378 } 1379 } 1380 1381 static void 1382 iflib_txsd_free(if_ctx_t ctx, iflib_txq_t txq, int i) 1383 { 1384 struct mbuf **mp; 1385 1386 mp = &txq->ift_sds.ifsd_m[i]; 1387 if (*mp == NULL) 1388 return; 1389 1390 if (txq->ift_sds.ifsd_map != NULL) { 1391 bus_dmamap_sync(txq->ift_desc_tag, 1392 txq->ift_sds.ifsd_map[i], 1393 BUS_DMASYNC_POSTWRITE); 1394 bus_dmamap_unload(txq->ift_desc_tag, 1395 txq->ift_sds.ifsd_map[i]); 1396 } 1397 m_free(*mp); 1398 DBG_COUNTER_INC(tx_frees); 1399 *mp = NULL; 1400 } 1401 1402 static int 1403 iflib_txq_setup(iflib_txq_t txq) 1404 { 1405 if_ctx_t ctx = txq->ift_ctx; 1406 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 1407 iflib_dma_info_t di; 1408 int i; 1409 1410 /* Set number of descriptors available */ 1411 txq->ift_qstatus = IFLIB_QUEUE_IDLE; 1412 1413 /* Reset indices */ 1414 txq->ift_cidx_processed = txq->ift_pidx = txq->ift_cidx = txq->ift_npending = 0; 1415 txq->ift_size = scctx->isc_ntxd[txq->ift_br_offset]; 1416 1417 for (i = 0, di = txq->ift_ifdi; i < ctx->ifc_nhwtxqs; i++, di++) 1418 bzero((void *)di->idi_vaddr, di->idi_size); 1419 1420 IFDI_TXQ_SETUP(ctx, txq->ift_id); 1421 for (i = 0, di = txq->ift_ifdi; i < ctx->ifc_nhwtxqs; i++, di++) 1422 bus_dmamap_sync(di->idi_tag, di->idi_map, 1423 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); 1424 return (0); 1425 } 1426 1427 /********************************************************************* 1428 * 1429 * Allocate memory for rx_buffer structures. Since we use one 1430 * rx_buffer per received packet, the maximum number of rx_buffer's 1431 * that we'll need is equal to the number of receive descriptors 1432 * that we've allocated. 1433 * 1434 **********************************************************************/ 1435 static int 1436 iflib_rxsd_alloc(iflib_rxq_t rxq) 1437 { 1438 if_ctx_t ctx = rxq->ifr_ctx; 1439 if_shared_ctx_t sctx = ctx->ifc_sctx; 1440 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 1441 device_t dev = ctx->ifc_dev; 1442 iflib_fl_t fl; 1443 iflib_rxsd_t rxsd; 1444 int err; 1445 1446 MPASS(scctx->isc_nrxd[0] > 0); 1447 MPASS(scctx->isc_nrxd[rxq->ifr_fl_offset] > 0); 1448 1449 fl = rxq->ifr_fl; 1450 for (int i = 0; i < rxq->ifr_nfl; i++, fl++) { 1451 fl->ifl_sds = malloc(sizeof(struct iflib_sw_rx_desc) * 1452 scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, 1453 M_WAITOK | M_ZERO); 1454 if (fl->ifl_sds == NULL) { 1455 device_printf(dev, "Unable to allocate rx sw desc memory\n"); 1456 return (ENOMEM); 1457 } 1458 fl->ifl_size = scctx->isc_nrxd[rxq->ifr_fl_offset]; /* this isn't necessarily the same */ 1459 err = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */ 1460 1, 0, /* alignment, bounds */ 1461 BUS_SPACE_MAXADDR, /* lowaddr */ 1462 BUS_SPACE_MAXADDR, /* highaddr */ 1463 NULL, NULL, /* filter, filterarg */ 1464 sctx->isc_rx_maxsize, /* maxsize */ 1465 sctx->isc_rx_nsegments, /* nsegments */ 1466 sctx->isc_rx_maxsegsize, /* maxsegsize */ 1467 0, /* flags */ 1468 NULL, /* lockfunc */ 1469 NULL, /* lockarg */ 1470 &fl->ifl_desc_tag); 1471 if (err) { 1472 device_printf(dev, "%s: bus_dma_tag_create failed %d\n", 1473 __func__, err); 1474 goto fail; 1475 } 1476 1477 rxsd = fl->ifl_sds; 1478 for (int i = 0; i < scctx->isc_nrxd[rxq->ifr_fl_offset]; i++, rxsd++) { 1479 err = bus_dmamap_create(fl->ifl_desc_tag, 0, &rxsd->ifsd_map); 1480 if (err) { 1481 device_printf(dev, "%s: bus_dmamap_create failed: %d\n", 1482 __func__, err); 1483 goto fail; 1484 } 1485 } 1486 } 1487 return (0); 1488 1489 fail: 1490 iflib_rx_structures_free(ctx); 1491 return (err); 1492 } 1493 1494 1495 /* 1496 * Internal service routines 1497 */ 1498 1499 struct rxq_refill_cb_arg { 1500 int error; 1501 bus_dma_segment_t seg; 1502 int nseg; 1503 }; 1504 1505 static void 1506 _rxq_refill_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error) 1507 { 1508 struct rxq_refill_cb_arg *cb_arg = arg; 1509 1510 cb_arg->error = error; 1511 cb_arg->seg = segs[0]; 1512 cb_arg->nseg = nseg; 1513 } 1514 1515 1516 #ifdef ACPI_DMAR 1517 #define IS_DMAR(ctx) (ctx->ifc_flags & IFC_DMAR) 1518 #else 1519 #define IS_DMAR(ctx) (0) 1520 #endif 1521 1522 /** 1523 * rxq_refill - refill an rxq free-buffer list 1524 * @ctx: the iflib context 1525 * @rxq: the free-list to refill 1526 * @n: the number of new buffers to allocate 1527 * 1528 * (Re)populate an rxq free-buffer list with up to @n new packet buffers. 1529 * The caller must assure that @n does not exceed the queue's capacity. 1530 */ 1531 static void 1532 _iflib_fl_refill(if_ctx_t ctx, iflib_fl_t fl, int count) 1533 { 1534 struct mbuf *m; 1535 int pidx = fl->ifl_pidx; 1536 iflib_rxsd_t rxsd = &fl->ifl_sds[pidx]; 1537 caddr_t cl; 1538 int n, i = 0; 1539 uint64_t bus_addr; 1540 int err; 1541 1542 n = count; 1543 MPASS(n > 0); 1544 MPASS(fl->ifl_credits + n <= fl->ifl_size); 1545 1546 if (pidx < fl->ifl_cidx) 1547 MPASS(pidx + n <= fl->ifl_cidx); 1548 if (pidx == fl->ifl_cidx && (fl->ifl_credits < fl->ifl_size)) 1549 MPASS(fl->ifl_gen == 0); 1550 if (pidx > fl->ifl_cidx) 1551 MPASS(n <= fl->ifl_size - pidx + fl->ifl_cidx); 1552 1553 DBG_COUNTER_INC(fl_refills); 1554 if (n > 8) 1555 DBG_COUNTER_INC(fl_refills_large); 1556 1557 while (n--) { 1558 /* 1559 * We allocate an uninitialized mbuf + cluster, mbuf is 1560 * initialized after rx. 1561 * 1562 * If the cluster is still set then we know a minimum sized packet was received 1563 */ 1564 if ((cl = rxsd->ifsd_cl) == NULL) { 1565 if ((cl = rxsd->ifsd_cl = m_cljget(NULL, M_NOWAIT, fl->ifl_buf_size)) == NULL) 1566 break; 1567 #if MEMORY_LOGGING 1568 fl->ifl_cl_enqueued++; 1569 #endif 1570 } 1571 if ((m = m_gethdr(M_NOWAIT, MT_NOINIT)) == NULL) { 1572 break; 1573 } 1574 #if MEMORY_LOGGING 1575 fl->ifl_m_enqueued++; 1576 #endif 1577 1578 DBG_COUNTER_INC(rx_allocs); 1579 #ifdef notyet 1580 if ((rxsd->ifsd_flags & RX_SW_DESC_MAP_CREATED) == 0) { 1581 int err; 1582 1583 if ((err = bus_dmamap_create(fl->ifl_ifdi->idi_tag, 0, &rxsd->ifsd_map))) { 1584 log(LOG_WARNING, "bus_dmamap_create failed %d\n", err); 1585 uma_zfree(fl->ifl_zone, cl); 1586 n = 0; 1587 goto done; 1588 } 1589 rxsd->ifsd_flags |= RX_SW_DESC_MAP_CREATED; 1590 } 1591 #endif 1592 #if defined(__i386__) || defined(__amd64__) 1593 if (!IS_DMAR(ctx)) { 1594 bus_addr = pmap_kextract((vm_offset_t)cl); 1595 } else 1596 #endif 1597 { 1598 struct rxq_refill_cb_arg cb_arg; 1599 iflib_rxq_t q; 1600 1601 cb_arg.error = 0; 1602 q = fl->ifl_rxq; 1603 err = bus_dmamap_load(fl->ifl_desc_tag, rxsd->ifsd_map, 1604 cl, fl->ifl_buf_size, _rxq_refill_cb, &cb_arg, 0); 1605 1606 if (err != 0 || cb_arg.error) { 1607 /* 1608 * !zone_pack ? 1609 */ 1610 if (fl->ifl_zone == zone_pack) 1611 uma_zfree(fl->ifl_zone, cl); 1612 m_free(m); 1613 n = 0; 1614 goto done; 1615 } 1616 bus_addr = cb_arg.seg.ds_addr; 1617 } 1618 rxsd->ifsd_flags |= RX_SW_DESC_INUSE; 1619 1620 MPASS(rxsd->ifsd_m == NULL); 1621 rxsd->ifsd_cl = cl; 1622 rxsd->ifsd_m = m; 1623 fl->ifl_bus_addrs[i] = bus_addr; 1624 fl->ifl_vm_addrs[i] = cl; 1625 rxsd++; 1626 fl->ifl_credits++; 1627 i++; 1628 MPASS(fl->ifl_credits <= fl->ifl_size); 1629 if (++fl->ifl_pidx == fl->ifl_size) { 1630 fl->ifl_pidx = 0; 1631 fl->ifl_gen = 1; 1632 rxsd = fl->ifl_sds; 1633 } 1634 if (n == 0 || i == IFLIB_MAX_RX_REFRESH) { 1635 ctx->isc_rxd_refill(ctx->ifc_softc, fl->ifl_rxq->ifr_id, fl->ifl_id, pidx, 1636 fl->ifl_bus_addrs, fl->ifl_vm_addrs, i, fl->ifl_buf_size); 1637 i = 0; 1638 pidx = fl->ifl_pidx; 1639 } 1640 } 1641 done: 1642 DBG_COUNTER_INC(rxd_flush); 1643 if (fl->ifl_pidx == 0) 1644 pidx = fl->ifl_size - 1; 1645 else 1646 pidx = fl->ifl_pidx - 1; 1647 ctx->isc_rxd_flush(ctx->ifc_softc, fl->ifl_rxq->ifr_id, fl->ifl_id, pidx); 1648 } 1649 1650 static __inline void 1651 __iflib_fl_refill_lt(if_ctx_t ctx, iflib_fl_t fl, int max) 1652 { 1653 /* we avoid allowing pidx to catch up with cidx as it confuses ixl */ 1654 int32_t reclaimable = fl->ifl_size - fl->ifl_credits - 1; 1655 #ifdef INVARIANTS 1656 int32_t delta = fl->ifl_size - get_inuse(fl->ifl_size, fl->ifl_cidx, fl->ifl_pidx, fl->ifl_gen) - 1; 1657 #endif 1658 1659 MPASS(fl->ifl_credits <= fl->ifl_size); 1660 MPASS(reclaimable == delta); 1661 1662 if (reclaimable > 0) 1663 _iflib_fl_refill(ctx, fl, min(max, reclaimable)); 1664 } 1665 1666 static void 1667 iflib_fl_bufs_free(iflib_fl_t fl) 1668 { 1669 iflib_dma_info_t idi = fl->ifl_ifdi; 1670 uint32_t i; 1671 1672 for (i = 0; i < fl->ifl_size; i++) { 1673 iflib_rxsd_t d = &fl->ifl_sds[i]; 1674 1675 if (d->ifsd_flags & RX_SW_DESC_INUSE) { 1676 bus_dmamap_unload(fl->ifl_desc_tag, d->ifsd_map); 1677 bus_dmamap_destroy(fl->ifl_desc_tag, d->ifsd_map); 1678 if (d->ifsd_m != NULL) { 1679 m_init(d->ifsd_m, M_NOWAIT, MT_DATA, 0); 1680 uma_zfree(zone_mbuf, d->ifsd_m); 1681 } 1682 if (d->ifsd_cl != NULL) 1683 uma_zfree(fl->ifl_zone, d->ifsd_cl); 1684 d->ifsd_flags = 0; 1685 } else { 1686 MPASS(d->ifsd_cl == NULL); 1687 MPASS(d->ifsd_m == NULL); 1688 } 1689 #if MEMORY_LOGGING 1690 fl->ifl_m_dequeued++; 1691 fl->ifl_cl_dequeued++; 1692 #endif 1693 d->ifsd_cl = NULL; 1694 d->ifsd_m = NULL; 1695 } 1696 /* 1697 * Reset free list values 1698 */ 1699 fl->ifl_credits = fl->ifl_cidx = fl->ifl_pidx = fl->ifl_gen = 0;; 1700 bzero(idi->idi_vaddr, idi->idi_size); 1701 } 1702 1703 /********************************************************************* 1704 * 1705 * Initialize a receive ring and its buffers. 1706 * 1707 **********************************************************************/ 1708 static int 1709 iflib_fl_setup(iflib_fl_t fl) 1710 { 1711 iflib_rxq_t rxq = fl->ifl_rxq; 1712 if_ctx_t ctx = rxq->ifr_ctx; 1713 if_softc_ctx_t sctx = &ctx->ifc_softc_ctx; 1714 1715 /* 1716 ** Free current RX buffer structs and their mbufs 1717 */ 1718 iflib_fl_bufs_free(fl); 1719 /* Now replenish the mbufs */ 1720 MPASS(fl->ifl_credits == 0); 1721 /* 1722 * XXX don't set the max_frame_size to larger 1723 * than the hardware can handle 1724 */ 1725 if (sctx->isc_max_frame_size <= 2048) 1726 fl->ifl_buf_size = MCLBYTES; 1727 else if (sctx->isc_max_frame_size <= 4096) 1728 fl->ifl_buf_size = MJUMPAGESIZE; 1729 else if (sctx->isc_max_frame_size <= 9216) 1730 fl->ifl_buf_size = MJUM9BYTES; 1731 else 1732 fl->ifl_buf_size = MJUM16BYTES; 1733 if (fl->ifl_buf_size > ctx->ifc_max_fl_buf_size) 1734 ctx->ifc_max_fl_buf_size = fl->ifl_buf_size; 1735 fl->ifl_cltype = m_gettype(fl->ifl_buf_size); 1736 fl->ifl_zone = m_getzone(fl->ifl_buf_size); 1737 1738 1739 /* avoid pre-allocating zillions of clusters to an idle card 1740 * potentially speeding up attach 1741 */ 1742 _iflib_fl_refill(ctx, fl, min(128, fl->ifl_size)); 1743 MPASS(min(128, fl->ifl_size) == fl->ifl_credits); 1744 if (min(128, fl->ifl_size) != fl->ifl_credits) 1745 return (ENOBUFS); 1746 /* 1747 * handle failure 1748 */ 1749 MPASS(rxq != NULL); 1750 MPASS(fl->ifl_ifdi != NULL); 1751 bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map, 1752 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); 1753 return (0); 1754 } 1755 1756 /********************************************************************* 1757 * 1758 * Free receive ring data structures 1759 * 1760 **********************************************************************/ 1761 static void 1762 iflib_rx_sds_free(iflib_rxq_t rxq) 1763 { 1764 iflib_fl_t fl; 1765 int i; 1766 1767 if (rxq->ifr_fl != NULL) { 1768 for (i = 0; i < rxq->ifr_nfl; i++) { 1769 fl = &rxq->ifr_fl[i]; 1770 if (fl->ifl_desc_tag != NULL) { 1771 bus_dma_tag_destroy(fl->ifl_desc_tag); 1772 fl->ifl_desc_tag = NULL; 1773 } 1774 } 1775 if (rxq->ifr_fl->ifl_sds != NULL) 1776 free(rxq->ifr_fl->ifl_sds, M_IFLIB); 1777 1778 free(rxq->ifr_fl, M_IFLIB); 1779 rxq->ifr_fl = NULL; 1780 rxq->ifr_cq_gen = rxq->ifr_cq_cidx = rxq->ifr_cq_pidx = 0; 1781 } 1782 } 1783 1784 /* 1785 * MI independent logic 1786 * 1787 */ 1788 static void 1789 iflib_timer(void *arg) 1790 { 1791 iflib_txq_t txq = arg; 1792 if_ctx_t ctx = txq->ift_ctx; 1793 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 1794 1795 if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)) 1796 return; 1797 /* 1798 ** Check on the state of the TX queue(s), this 1799 ** can be done without the lock because its RO 1800 ** and the HUNG state will be static if set. 1801 */ 1802 IFDI_TIMER(ctx, txq->ift_id); 1803 if ((txq->ift_qstatus == IFLIB_QUEUE_HUNG) && 1804 (ctx->ifc_pause_frames == 0)) 1805 goto hung; 1806 1807 if (TXQ_AVAIL(txq) <= 2*scctx->isc_tx_nsegments || 1808 ifmp_ring_is_stalled(txq->ift_br[0])) 1809 GROUPTASK_ENQUEUE(&txq->ift_task); 1810 1811 ctx->ifc_pause_frames = 0; 1812 if (if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING) 1813 callout_reset_on(&txq->ift_timer, hz/2, iflib_timer, txq, txq->ift_timer.c_cpu); 1814 return; 1815 hung: 1816 CTX_LOCK(ctx); 1817 if_setdrvflagbits(ctx->ifc_ifp, 0, IFF_DRV_RUNNING); 1818 device_printf(ctx->ifc_dev, "TX(%d) desc avail = %d, pidx = %d\n", 1819 txq->ift_id, TXQ_AVAIL(txq), txq->ift_pidx); 1820 1821 IFDI_WATCHDOG_RESET(ctx); 1822 ctx->ifc_watchdog_events++; 1823 ctx->ifc_pause_frames = 0; 1824 1825 iflib_init_locked(ctx); 1826 CTX_UNLOCK(ctx); 1827 } 1828 1829 static void 1830 iflib_init_locked(if_ctx_t ctx) 1831 { 1832 if_softc_ctx_t sctx = &ctx->ifc_softc_ctx; 1833 if_t ifp = ctx->ifc_ifp; 1834 iflib_fl_t fl; 1835 iflib_txq_t txq; 1836 iflib_rxq_t rxq; 1837 int i, j; 1838 1839 1840 if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING); 1841 IFDI_INTR_DISABLE(ctx); 1842 1843 /* Set hardware offload abilities */ 1844 if_clearhwassist(ifp); 1845 if (if_getcapenable(ifp) & IFCAP_TXCSUM) 1846 if_sethwassistbits(ifp, CSUM_IP | CSUM_TCP | CSUM_UDP, 0); 1847 if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6) 1848 if_sethwassistbits(ifp, (CSUM_TCP_IPV6 | CSUM_UDP_IPV6), 0); 1849 if (if_getcapenable(ifp) & IFCAP_TSO4) 1850 if_sethwassistbits(ifp, CSUM_IP_TSO, 0); 1851 if (if_getcapenable(ifp) & IFCAP_TSO6) 1852 if_sethwassistbits(ifp, CSUM_IP6_TSO, 0); 1853 1854 for (i = 0, txq = ctx->ifc_txqs; i < sctx->isc_ntxqsets; i++, txq++) { 1855 CALLOUT_LOCK(txq); 1856 callout_stop(&txq->ift_timer); 1857 callout_stop(&txq->ift_db_check); 1858 CALLOUT_UNLOCK(txq); 1859 iflib_netmap_txq_init(ctx, txq); 1860 } 1861 for (i = 0, rxq = ctx->ifc_rxqs; i < sctx->isc_nrxqsets; i++, rxq++) { 1862 iflib_netmap_rxq_init(ctx, rxq); 1863 } 1864 #ifdef INVARIANTS 1865 i = if_getdrvflags(ifp); 1866 #endif 1867 IFDI_INIT(ctx); 1868 MPASS(if_getdrvflags(ifp) == i); 1869 for (i = 0, rxq = ctx->ifc_rxqs; i < sctx->isc_nrxqsets; i++, rxq++) { 1870 for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) { 1871 if (iflib_fl_setup(fl)) { 1872 device_printf(ctx->ifc_dev, "freelist setup failed - check cluster settings\n"); 1873 goto done; 1874 } 1875 } 1876 } 1877 done: 1878 if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_RUNNING, IFF_DRV_OACTIVE); 1879 IFDI_INTR_ENABLE(ctx); 1880 txq = ctx->ifc_txqs; 1881 for (i = 0; i < sctx->isc_ntxqsets; i++, txq++) 1882 callout_reset_on(&txq->ift_timer, hz/2, iflib_timer, txq, 1883 txq->ift_timer.c_cpu); 1884 } 1885 1886 static int 1887 iflib_media_change(if_t ifp) 1888 { 1889 if_ctx_t ctx = if_getsoftc(ifp); 1890 int err; 1891 1892 CTX_LOCK(ctx); 1893 if ((err = IFDI_MEDIA_CHANGE(ctx)) == 0) 1894 iflib_init_locked(ctx); 1895 CTX_UNLOCK(ctx); 1896 return (err); 1897 } 1898 1899 static void 1900 iflib_media_status(if_t ifp, struct ifmediareq *ifmr) 1901 { 1902 if_ctx_t ctx = if_getsoftc(ifp); 1903 1904 CTX_LOCK(ctx); 1905 IFDI_UPDATE_ADMIN_STATUS(ctx); 1906 IFDI_MEDIA_STATUS(ctx, ifmr); 1907 CTX_UNLOCK(ctx); 1908 } 1909 1910 static void 1911 iflib_stop(if_ctx_t ctx) 1912 { 1913 iflib_txq_t txq = ctx->ifc_txqs; 1914 iflib_rxq_t rxq = ctx->ifc_rxqs; 1915 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 1916 iflib_dma_info_t di; 1917 iflib_fl_t fl; 1918 int i, j; 1919 1920 /* Tell the stack that the interface is no longer active */ 1921 if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING); 1922 1923 IFDI_INTR_DISABLE(ctx); 1924 msleep(ctx, &ctx->ifc_mtx, PUSER, "iflib_init", hz); 1925 1926 /* Wait for current tx queue users to exit to disarm watchdog timer. */ 1927 for (i = 0; i < scctx->isc_ntxqsets; i++, txq++) { 1928 /* make sure all transmitters have completed before proceeding XXX */ 1929 1930 /* clean any enqueued buffers */ 1931 iflib_txq_check_drain(txq, 0); 1932 /* Free any existing tx buffers. */ 1933 for (j = 0; j < txq->ift_size; j++) { 1934 iflib_txsd_free(ctx, txq, j); 1935 } 1936 txq->ift_processed = txq->ift_cleaned = txq->ift_cidx_processed = 0; 1937 txq->ift_in_use = txq->ift_cidx = txq->ift_pidx = txq->ift_no_desc_avail = 0; 1938 txq->ift_closed = txq->ift_mbuf_defrag = txq->ift_mbuf_defrag_failed = 0; 1939 txq->ift_no_tx_dma_setup = txq->ift_txd_encap_efbig = txq->ift_map_failed = 0; 1940 txq->ift_pullups = 0; 1941 ifmp_ring_reset_stats(txq->ift_br[0]); 1942 for (j = 0, di = txq->ift_ifdi; j < ctx->ifc_nhwtxqs; j++, di++) 1943 bzero((void *)di->idi_vaddr, di->idi_size); 1944 } 1945 for (i = 0; i < scctx->isc_nrxqsets; i++, rxq++) { 1946 /* make sure all transmitters have completed before proceeding XXX */ 1947 1948 for (j = 0, di = txq->ift_ifdi; j < ctx->ifc_nhwrxqs; j++, di++) 1949 bzero((void *)di->idi_vaddr, di->idi_size); 1950 /* also resets the free lists pidx/cidx */ 1951 for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) 1952 iflib_fl_bufs_free(fl); 1953 } 1954 IFDI_STOP(ctx); 1955 } 1956 1957 static iflib_rxsd_t 1958 rxd_frag_to_sd(iflib_rxq_t rxq, if_rxd_frag_t irf, int *cltype, int unload) 1959 { 1960 int flid, cidx; 1961 iflib_rxsd_t sd; 1962 iflib_fl_t fl; 1963 iflib_dma_info_t di; 1964 1965 flid = irf->irf_flid; 1966 cidx = irf->irf_idx; 1967 fl = &rxq->ifr_fl[flid]; 1968 fl->ifl_credits--; 1969 #if MEMORY_LOGGING 1970 fl->ifl_m_dequeued++; 1971 if (cltype) 1972 fl->ifl_cl_dequeued++; 1973 #endif 1974 sd = &fl->ifl_sds[cidx]; 1975 di = fl->ifl_ifdi; 1976 bus_dmamap_sync(di->idi_tag, di->idi_map, 1977 BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); 1978 1979 /* not valid assert if bxe really does SGE from non-contiguous elements */ 1980 MPASS(fl->ifl_cidx == cidx); 1981 if (unload) 1982 bus_dmamap_unload(fl->ifl_desc_tag, sd->ifsd_map); 1983 1984 if (__predict_false(++fl->ifl_cidx == fl->ifl_size)) { 1985 fl->ifl_cidx = 0; 1986 fl->ifl_gen = 0; 1987 } 1988 /* YES ick */ 1989 if (cltype) 1990 *cltype = fl->ifl_cltype; 1991 return (sd); 1992 } 1993 1994 static struct mbuf * 1995 assemble_segments(iflib_rxq_t rxq, if_rxd_info_t ri) 1996 { 1997 int i, padlen , flags, cltype; 1998 struct mbuf *m, *mh, *mt; 1999 iflib_rxsd_t sd; 2000 caddr_t cl; 2001 2002 i = 0; 2003 mh = NULL; 2004 do { 2005 sd = rxd_frag_to_sd(rxq, &ri->iri_frags[i], &cltype, TRUE); 2006 2007 MPASS(sd->ifsd_cl != NULL); 2008 MPASS(sd->ifsd_m != NULL); 2009 2010 /* Don't include zero-length frags */ 2011 if (ri->iri_frags[i].irf_len == 0) { 2012 /* XXX we can save the cluster here, but not the mbuf */ 2013 m_init(sd->ifsd_m, M_NOWAIT, MT_DATA, 0); 2014 m_free(sd->ifsd_m); 2015 sd->ifsd_m = NULL; 2016 continue; 2017 } 2018 2019 m = sd->ifsd_m; 2020 if (mh == NULL) { 2021 flags = M_PKTHDR|M_EXT; 2022 mh = mt = m; 2023 padlen = ri->iri_pad; 2024 } else { 2025 flags = M_EXT; 2026 mt->m_next = m; 2027 mt = m; 2028 /* assuming padding is only on the first fragment */ 2029 padlen = 0; 2030 } 2031 sd->ifsd_m = NULL; 2032 cl = sd->ifsd_cl; 2033 sd->ifsd_cl = NULL; 2034 2035 /* Can these two be made one ? */ 2036 m_init(m, M_NOWAIT, MT_DATA, flags); 2037 m_cljset(m, cl, cltype); 2038 /* 2039 * These must follow m_init and m_cljset 2040 */ 2041 m->m_data += padlen; 2042 ri->iri_len -= padlen; 2043 m->m_len = ri->iri_frags[i].irf_len; 2044 } while (++i < ri->iri_nfrags); 2045 2046 return (mh); 2047 } 2048 2049 /* 2050 * Process one software descriptor 2051 */ 2052 static struct mbuf * 2053 iflib_rxd_pkt_get(iflib_rxq_t rxq, if_rxd_info_t ri) 2054 { 2055 struct mbuf *m; 2056 iflib_rxsd_t sd; 2057 2058 /* should I merge this back in now that the two paths are basically duplicated? */ 2059 if (ri->iri_nfrags == 1 && 2060 ri->iri_frags[0].irf_len <= IFLIB_RX_COPY_THRESH) { 2061 sd = rxd_frag_to_sd(rxq, &ri->iri_frags[0], NULL, FALSE); 2062 m = sd->ifsd_m; 2063 sd->ifsd_m = NULL; 2064 m_init(m, M_NOWAIT, MT_DATA, M_PKTHDR); 2065 memcpy(m->m_data, sd->ifsd_cl, ri->iri_len); 2066 m->m_len = ri->iri_frags[0].irf_len; 2067 } else { 2068 m = assemble_segments(rxq, ri); 2069 } 2070 m->m_pkthdr.len = ri->iri_len; 2071 m->m_pkthdr.rcvif = ri->iri_ifp; 2072 m->m_flags |= ri->iri_flags; 2073 m->m_pkthdr.ether_vtag = ri->iri_vtag; 2074 m->m_pkthdr.flowid = ri->iri_flowid; 2075 M_HASHTYPE_SET(m, ri->iri_rsstype); 2076 m->m_pkthdr.csum_flags = ri->iri_csum_flags; 2077 m->m_pkthdr.csum_data = ri->iri_csum_data; 2078 return (m); 2079 } 2080 2081 static bool 2082 iflib_rxeof(iflib_rxq_t rxq, int budget) 2083 { 2084 if_ctx_t ctx = rxq->ifr_ctx; 2085 if_shared_ctx_t sctx = ctx->ifc_sctx; 2086 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 2087 int avail, i; 2088 uint16_t *cidxp; 2089 struct if_rxd_info ri; 2090 int err, budget_left, rx_bytes, rx_pkts; 2091 iflib_fl_t fl; 2092 struct ifnet *ifp; 2093 int lro_enabled; 2094 /* 2095 * XXX early demux data packets so that if_input processing only handles 2096 * acks in interrupt context 2097 */ 2098 struct mbuf *m, *mh, *mt; 2099 2100 if (netmap_rx_irq(ctx->ifc_ifp, rxq->ifr_id, &budget)) { 2101 return (FALSE); 2102 } 2103 2104 mh = mt = NULL; 2105 MPASS(budget > 0); 2106 rx_pkts = rx_bytes = 0; 2107 if (sctx->isc_flags & IFLIB_HAS_RXCQ) 2108 cidxp = &rxq->ifr_cq_cidx; 2109 else 2110 cidxp = &rxq->ifr_fl[0].ifl_cidx; 2111 if ((avail = iflib_rxd_avail(ctx, rxq, *cidxp, budget)) == 0) { 2112 for (i = 0, fl = &rxq->ifr_fl[0]; i < sctx->isc_nfl; i++, fl++) 2113 __iflib_fl_refill_lt(ctx, fl, budget + 8); 2114 DBG_COUNTER_INC(rx_unavail); 2115 return (false); 2116 } 2117 2118 for (budget_left = budget; (budget_left > 0) && (avail > 0); budget_left--, avail--) { 2119 if (__predict_false(!CTX_ACTIVE(ctx))) { 2120 DBG_COUNTER_INC(rx_ctx_inactive); 2121 break; 2122 } 2123 /* 2124 * Reset client set fields to their default values 2125 */ 2126 bzero(&ri, sizeof(ri)); 2127 ri.iri_qsidx = rxq->ifr_id; 2128 ri.iri_cidx = *cidxp; 2129 ri.iri_ifp = ctx->ifc_ifp; 2130 ri.iri_frags = rxq->ifr_frags; 2131 err = ctx->isc_rxd_pkt_get(ctx->ifc_softc, &ri); 2132 2133 /* in lieu of handling correctly - make sure it isn't being unhandled */ 2134 MPASS(err == 0); 2135 if (sctx->isc_flags & IFLIB_HAS_RXCQ) { 2136 *cidxp = ri.iri_cidx; 2137 /* Update our consumer index */ 2138 while (rxq->ifr_cq_cidx >= scctx->isc_nrxd[0]) { 2139 rxq->ifr_cq_cidx -= scctx->isc_nrxd[0]; 2140 rxq->ifr_cq_gen = 0; 2141 } 2142 /* was this only a completion queue message? */ 2143 if (__predict_false(ri.iri_nfrags == 0)) 2144 continue; 2145 } 2146 MPASS(ri.iri_nfrags != 0); 2147 MPASS(ri.iri_len != 0); 2148 2149 /* will advance the cidx on the corresponding free lists */ 2150 m = iflib_rxd_pkt_get(rxq, &ri); 2151 if (avail == 0 && budget_left) 2152 avail = iflib_rxd_avail(ctx, rxq, *cidxp, budget_left); 2153 2154 if (__predict_false(m == NULL)) { 2155 DBG_COUNTER_INC(rx_mbuf_null); 2156 continue; 2157 } 2158 /* imm_pkt: -- cxgb */ 2159 if (mh == NULL) 2160 mh = mt = m; 2161 else { 2162 mt->m_nextpkt = m; 2163 mt = m; 2164 } 2165 } 2166 /* make sure that we can refill faster than drain */ 2167 for (i = 0, fl = &rxq->ifr_fl[0]; i < sctx->isc_nfl; i++, fl++) 2168 __iflib_fl_refill_lt(ctx, fl, budget + 8); 2169 2170 ifp = ctx->ifc_ifp; 2171 lro_enabled = (if_getcapenable(ifp) & IFCAP_LRO); 2172 while (mh != NULL) { 2173 m = mh; 2174 mh = mh->m_nextpkt; 2175 m->m_nextpkt = NULL; 2176 rx_bytes += m->m_pkthdr.len; 2177 rx_pkts++; 2178 #if defined(INET6) || defined(INET) 2179 if (lro_enabled && tcp_lro_rx(&rxq->ifr_lc, m, 0) == 0) 2180 continue; 2181 #endif 2182 DBG_COUNTER_INC(rx_if_input); 2183 ifp->if_input(ifp, m); 2184 } 2185 2186 if_inc_counter(ifp, IFCOUNTER_IBYTES, rx_bytes); 2187 if_inc_counter(ifp, IFCOUNTER_IPACKETS, rx_pkts); 2188 2189 /* 2190 * Flush any outstanding LRO work 2191 */ 2192 #if defined(INET6) || defined(INET) 2193 tcp_lro_flush_all(&rxq->ifr_lc); 2194 #endif 2195 if (avail) 2196 return true; 2197 return (iflib_rxd_avail(ctx, rxq, *cidxp, 1)); 2198 } 2199 2200 #define M_CSUM_FLAGS(m) ((m)->m_pkthdr.csum_flags) 2201 #define M_HAS_VLANTAG(m) (m->m_flags & M_VLANTAG) 2202 #define TXQ_MAX_DB_DEFERRED(size) (size >> 5) 2203 #define TXQ_MAX_DB_CONSUMED(size) (size >> 4) 2204 2205 static __inline void 2206 iflib_txd_db_check(if_ctx_t ctx, iflib_txq_t txq, int ring) 2207 { 2208 uint32_t dbval; 2209 2210 if (ring || txq->ift_db_pending >= 2211 TXQ_MAX_DB_DEFERRED(txq->ift_size)) { 2212 2213 /* the lock will only ever be contended in the !min_latency case */ 2214 if (!TXDB_TRYLOCK(txq)) 2215 return; 2216 dbval = txq->ift_npending ? txq->ift_npending : txq->ift_pidx; 2217 ctx->isc_txd_flush(ctx->ifc_softc, txq->ift_id, dbval); 2218 txq->ift_db_pending = txq->ift_npending = 0; 2219 TXDB_UNLOCK(txq); 2220 } 2221 } 2222 2223 static void 2224 iflib_txd_deferred_db_check(void * arg) 2225 { 2226 iflib_txq_t txq = arg; 2227 2228 /* simple non-zero boolean so use bitwise OR */ 2229 if ((txq->ift_db_pending | txq->ift_npending) && 2230 txq->ift_db_pending >= txq->ift_db_pending_queued) 2231 iflib_txd_db_check(txq->ift_ctx, txq, TRUE); 2232 txq->ift_db_pending_queued = 0; 2233 if (ifmp_ring_is_stalled(txq->ift_br[0])) 2234 iflib_txq_check_drain(txq, 4); 2235 } 2236 2237 #ifdef PKT_DEBUG 2238 static void 2239 print_pkt(if_pkt_info_t pi) 2240 { 2241 printf("pi len: %d qsidx: %d nsegs: %d ndescs: %d flags: %x pidx: %d\n", 2242 pi->ipi_len, pi->ipi_qsidx, pi->ipi_nsegs, pi->ipi_ndescs, pi->ipi_flags, pi->ipi_pidx); 2243 printf("pi new_pidx: %d csum_flags: %lx tso_segsz: %d mflags: %x vtag: %d\n", 2244 pi->ipi_new_pidx, pi->ipi_csum_flags, pi->ipi_tso_segsz, pi->ipi_mflags, pi->ipi_vtag); 2245 printf("pi etype: %d ehdrlen: %d ip_hlen: %d ipproto: %d\n", 2246 pi->ipi_etype, pi->ipi_ehdrlen, pi->ipi_ip_hlen, pi->ipi_ipproto); 2247 } 2248 #endif 2249 2250 #define IS_TSO4(pi) ((pi)->ipi_csum_flags & CSUM_IP_TSO) 2251 #define IS_TSO6(pi) ((pi)->ipi_csum_flags & CSUM_IP6_TSO) 2252 2253 static int 2254 iflib_parse_header(iflib_txq_t txq, if_pkt_info_t pi, struct mbuf **mp) 2255 { 2256 struct ether_vlan_header *eh; 2257 struct mbuf *m, *n; 2258 2259 n = m = *mp; 2260 /* 2261 * Determine where frame payload starts. 2262 * Jump over vlan headers if already present, 2263 * helpful for QinQ too. 2264 */ 2265 if (__predict_false(m->m_len < sizeof(*eh))) { 2266 txq->ift_pullups++; 2267 if (__predict_false((m = m_pullup(m, sizeof(*eh))) == NULL)) 2268 return (ENOMEM); 2269 } 2270 eh = mtod(m, struct ether_vlan_header *); 2271 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) { 2272 pi->ipi_etype = ntohs(eh->evl_proto); 2273 pi->ipi_ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN; 2274 } else { 2275 pi->ipi_etype = ntohs(eh->evl_encap_proto); 2276 pi->ipi_ehdrlen = ETHER_HDR_LEN; 2277 } 2278 2279 switch (pi->ipi_etype) { 2280 #ifdef INET 2281 case ETHERTYPE_IP: 2282 { 2283 struct ip *ip = NULL; 2284 struct tcphdr *th = NULL; 2285 int minthlen; 2286 2287 minthlen = min(m->m_pkthdr.len, pi->ipi_ehdrlen + sizeof(*ip) + sizeof(*th)); 2288 if (__predict_false(m->m_len < minthlen)) { 2289 /* 2290 * if this code bloat is causing too much of a hit 2291 * move it to a separate function and mark it noinline 2292 */ 2293 if (m->m_len == pi->ipi_ehdrlen) { 2294 n = m->m_next; 2295 MPASS(n); 2296 if (n->m_len >= sizeof(*ip)) { 2297 ip = (struct ip *)n->m_data; 2298 if (n->m_len >= (ip->ip_hl << 2) + sizeof(*th)) 2299 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2300 } else { 2301 txq->ift_pullups++; 2302 if (__predict_false((m = m_pullup(m, minthlen)) == NULL)) 2303 return (ENOMEM); 2304 ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen); 2305 } 2306 } else { 2307 txq->ift_pullups++; 2308 if (__predict_false((m = m_pullup(m, minthlen)) == NULL)) 2309 return (ENOMEM); 2310 ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen); 2311 if (m->m_len >= (ip->ip_hl << 2) + sizeof(*th)) 2312 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2313 } 2314 } else { 2315 ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen); 2316 if (m->m_len >= (ip->ip_hl << 2) + sizeof(*th)) 2317 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2318 } 2319 pi->ipi_ip_hlen = ip->ip_hl << 2; 2320 pi->ipi_ipproto = ip->ip_p; 2321 pi->ipi_flags |= IPI_TX_IPV4; 2322 2323 if (pi->ipi_csum_flags & CSUM_IP) 2324 ip->ip_sum = 0; 2325 2326 if (pi->ipi_ipproto == IPPROTO_TCP) { 2327 if (__predict_false(th == NULL)) { 2328 txq->ift_pullups++; 2329 if (__predict_false((m = m_pullup(m, (ip->ip_hl << 2) + sizeof(*th))) == NULL)) 2330 return (ENOMEM); 2331 th = (struct tcphdr *)((caddr_t)ip + pi->ipi_ip_hlen); 2332 } 2333 pi->ipi_tcp_hflags = th->th_flags; 2334 pi->ipi_tcp_hlen = th->th_off << 2; 2335 pi->ipi_tcp_seq = th->th_seq; 2336 } 2337 if (IS_TSO4(pi)) { 2338 if (__predict_false(ip->ip_p != IPPROTO_TCP)) 2339 return (ENXIO); 2340 th->th_sum = in_pseudo(ip->ip_src.s_addr, 2341 ip->ip_dst.s_addr, htons(IPPROTO_TCP)); 2342 pi->ipi_tso_segsz = m->m_pkthdr.tso_segsz; 2343 } 2344 break; 2345 } 2346 #endif 2347 #ifdef INET6 2348 case ETHERTYPE_IPV6: 2349 { 2350 struct ip6_hdr *ip6 = (struct ip6_hdr *)(m->m_data + pi->ipi_ehdrlen); 2351 struct tcphdr *th; 2352 pi->ipi_ip_hlen = sizeof(struct ip6_hdr); 2353 2354 if (__predict_false(m->m_len < pi->ipi_ehdrlen + sizeof(struct ip6_hdr))) { 2355 if (__predict_false((m = m_pullup(m, pi->ipi_ehdrlen + sizeof(struct ip6_hdr))) == NULL)) 2356 return (ENOMEM); 2357 } 2358 th = (struct tcphdr *)((caddr_t)ip6 + pi->ipi_ip_hlen); 2359 2360 /* XXX-BZ this will go badly in case of ext hdrs. */ 2361 pi->ipi_ipproto = ip6->ip6_nxt; 2362 pi->ipi_flags |= IPI_TX_IPV6; 2363 2364 if (pi->ipi_ipproto == IPPROTO_TCP) { 2365 if (__predict_false(m->m_len < pi->ipi_ehdrlen + sizeof(struct ip6_hdr) + sizeof(struct tcphdr))) { 2366 if (__predict_false((m = m_pullup(m, pi->ipi_ehdrlen + sizeof(struct ip6_hdr) + sizeof(struct tcphdr))) == NULL)) 2367 return (ENOMEM); 2368 } 2369 pi->ipi_tcp_hflags = th->th_flags; 2370 pi->ipi_tcp_hlen = th->th_off << 2; 2371 } 2372 if (IS_TSO6(pi)) { 2373 2374 if (__predict_false(ip6->ip6_nxt != IPPROTO_TCP)) 2375 return (ENXIO); 2376 /* 2377 * The corresponding flag is set by the stack in the IPv4 2378 * TSO case, but not in IPv6 (at least in FreeBSD 10.2). 2379 * So, set it here because the rest of the flow requires it. 2380 */ 2381 pi->ipi_csum_flags |= CSUM_TCP_IPV6; 2382 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0); 2383 pi->ipi_tso_segsz = m->m_pkthdr.tso_segsz; 2384 } 2385 break; 2386 } 2387 #endif 2388 default: 2389 pi->ipi_csum_flags &= ~CSUM_OFFLOAD; 2390 pi->ipi_ip_hlen = 0; 2391 break; 2392 } 2393 *mp = m; 2394 return (0); 2395 } 2396 2397 2398 static __noinline struct mbuf * 2399 collapse_pkthdr(struct mbuf *m0) 2400 { 2401 struct mbuf *m, *m_next, *tmp; 2402 2403 m = m0; 2404 m_next = m->m_next; 2405 while (m_next != NULL && m_next->m_len == 0) { 2406 m = m_next; 2407 m->m_next = NULL; 2408 m_free(m); 2409 m_next = m_next->m_next; 2410 } 2411 m = m0; 2412 m->m_next = m_next; 2413 if ((m_next->m_flags & M_EXT) == 0) { 2414 m = m_defrag(m, M_NOWAIT); 2415 } else { 2416 tmp = m_next->m_next; 2417 memcpy(m_next, m, MPKTHSIZE); 2418 m = m_next; 2419 m->m_next = tmp; 2420 } 2421 return (m); 2422 } 2423 2424 /* 2425 * If dodgy hardware rejects the scatter gather chain we've handed it 2426 * we'll need to remove the mbuf chain from ifsg_m[] before we can add the 2427 * m_defrag'd mbufs 2428 */ 2429 static __noinline struct mbuf * 2430 iflib_remove_mbuf(iflib_txq_t txq) 2431 { 2432 int ntxd, i, pidx; 2433 struct mbuf *m, *mh, **ifsd_m; 2434 2435 pidx = txq->ift_pidx; 2436 ifsd_m = txq->ift_sds.ifsd_m; 2437 ntxd = txq->ift_size; 2438 mh = m = ifsd_m[pidx]; 2439 ifsd_m[pidx] = NULL; 2440 #if MEMORY_LOGGING 2441 txq->ift_dequeued++; 2442 #endif 2443 i = 1; 2444 2445 while (m) { 2446 ifsd_m[(pidx + i) & (ntxd -1)] = NULL; 2447 #if MEMORY_LOGGING 2448 txq->ift_dequeued++; 2449 #endif 2450 m = m->m_next; 2451 i++; 2452 } 2453 return (mh); 2454 } 2455 2456 static int 2457 iflib_busdma_load_mbuf_sg(iflib_txq_t txq, bus_dma_tag_t tag, bus_dmamap_t map, 2458 struct mbuf **m0, bus_dma_segment_t *segs, int *nsegs, 2459 int max_segs, int flags) 2460 { 2461 if_ctx_t ctx; 2462 if_shared_ctx_t sctx; 2463 if_softc_ctx_t scctx; 2464 int i, next, pidx, mask, err, maxsegsz, ntxd, count; 2465 struct mbuf *m, *tmp, **ifsd_m, **mp; 2466 2467 m = *m0; 2468 2469 /* 2470 * Please don't ever do this 2471 */ 2472 if (__predict_false(m->m_len == 0)) 2473 *m0 = m = collapse_pkthdr(m); 2474 2475 ctx = txq->ift_ctx; 2476 sctx = ctx->ifc_sctx; 2477 scctx = &ctx->ifc_softc_ctx; 2478 ifsd_m = txq->ift_sds.ifsd_m; 2479 ntxd = txq->ift_size; 2480 pidx = txq->ift_pidx; 2481 if (map != NULL) { 2482 uint8_t *ifsd_flags = txq->ift_sds.ifsd_flags; 2483 2484 err = bus_dmamap_load_mbuf_sg(tag, map, 2485 *m0, segs, nsegs, BUS_DMA_NOWAIT); 2486 if (err) 2487 return (err); 2488 ifsd_flags[pidx] |= TX_SW_DESC_MAPPED; 2489 i = 0; 2490 next = pidx; 2491 mask = (txq->ift_size-1); 2492 m = *m0; 2493 do { 2494 mp = &ifsd_m[next]; 2495 *mp = m; 2496 m = m->m_next; 2497 if (__predict_false((*mp)->m_len == 0)) { 2498 m_free(*mp); 2499 *mp = NULL; 2500 } else 2501 next = (pidx + i) & (ntxd-1); 2502 } while (m != NULL); 2503 } else { 2504 int buflen, sgsize, max_sgsize; 2505 vm_offset_t vaddr; 2506 vm_paddr_t curaddr; 2507 2508 count = i = 0; 2509 maxsegsz = sctx->isc_tx_maxsize; 2510 m = *m0; 2511 do { 2512 if (__predict_false(m->m_len <= 0)) { 2513 tmp = m; 2514 m = m->m_next; 2515 tmp->m_next = NULL; 2516 m_free(tmp); 2517 continue; 2518 } 2519 buflen = m->m_len; 2520 vaddr = (vm_offset_t)m->m_data; 2521 /* 2522 * see if we can't be smarter about physically 2523 * contiguous mappings 2524 */ 2525 next = (pidx + count) & (ntxd-1); 2526 MPASS(ifsd_m[next] == NULL); 2527 #if MEMORY_LOGGING 2528 txq->ift_enqueued++; 2529 #endif 2530 ifsd_m[next] = m; 2531 while (buflen > 0) { 2532 max_sgsize = MIN(buflen, maxsegsz); 2533 curaddr = pmap_kextract(vaddr); 2534 sgsize = PAGE_SIZE - (curaddr & PAGE_MASK); 2535 sgsize = MIN(sgsize, max_sgsize); 2536 segs[i].ds_addr = curaddr; 2537 segs[i].ds_len = sgsize; 2538 vaddr += sgsize; 2539 buflen -= sgsize; 2540 i++; 2541 if (i >= max_segs) 2542 goto err; 2543 } 2544 count++; 2545 tmp = m; 2546 m = m->m_next; 2547 } while (m != NULL); 2548 *nsegs = i; 2549 } 2550 return (0); 2551 err: 2552 *m0 = iflib_remove_mbuf(txq); 2553 return (EFBIG); 2554 } 2555 2556 static int 2557 iflib_encap(iflib_txq_t txq, struct mbuf **m_headp) 2558 { 2559 if_ctx_t ctx; 2560 if_shared_ctx_t sctx; 2561 if_softc_ctx_t scctx; 2562 bus_dma_segment_t *segs; 2563 struct mbuf *m_head; 2564 bus_dmamap_t map; 2565 struct if_pkt_info pi; 2566 int remap = 0; 2567 int err, nsegs, ndesc, max_segs, pidx, cidx, next, ntxd; 2568 bus_dma_tag_t desc_tag; 2569 2570 segs = txq->ift_segs; 2571 ctx = txq->ift_ctx; 2572 sctx = ctx->ifc_sctx; 2573 scctx = &ctx->ifc_softc_ctx; 2574 segs = txq->ift_segs; 2575 ntxd = txq->ift_size; 2576 m_head = *m_headp; 2577 map = NULL; 2578 2579 /* 2580 * If we're doing TSO the next descriptor to clean may be quite far ahead 2581 */ 2582 cidx = txq->ift_cidx; 2583 pidx = txq->ift_pidx; 2584 next = (cidx + CACHE_PTR_INCREMENT) & (ntxd-1); 2585 2586 /* prefetch the next cache line of mbuf pointers and flags */ 2587 prefetch(&txq->ift_sds.ifsd_m[next]); 2588 if (txq->ift_sds.ifsd_map != NULL) { 2589 prefetch(&txq->ift_sds.ifsd_map[next]); 2590 map = txq->ift_sds.ifsd_map[pidx]; 2591 next = (cidx + CACHE_LINE_SIZE) & (ntxd-1); 2592 prefetch(&txq->ift_sds.ifsd_flags[next]); 2593 } 2594 2595 2596 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) { 2597 desc_tag = txq->ift_tso_desc_tag; 2598 max_segs = scctx->isc_tx_tso_segments_max; 2599 } else { 2600 desc_tag = txq->ift_desc_tag; 2601 max_segs = scctx->isc_tx_nsegments; 2602 } 2603 m_head = *m_headp; 2604 bzero(&pi, sizeof(pi)); 2605 pi.ipi_len = m_head->m_pkthdr.len; 2606 pi.ipi_mflags = (m_head->m_flags & (M_VLANTAG|M_BCAST|M_MCAST)); 2607 pi.ipi_csum_flags = m_head->m_pkthdr.csum_flags; 2608 pi.ipi_vtag = (m_head->m_flags & M_VLANTAG) ? m_head->m_pkthdr.ether_vtag : 0; 2609 pi.ipi_pidx = pidx; 2610 pi.ipi_qsidx = txq->ift_id; 2611 2612 /* deliberate bitwise OR to make one condition */ 2613 if (__predict_true((pi.ipi_csum_flags | pi.ipi_vtag))) { 2614 if (__predict_false((err = iflib_parse_header(txq, &pi, m_headp)) != 0)) 2615 return (err); 2616 m_head = *m_headp; 2617 } 2618 2619 retry: 2620 err = iflib_busdma_load_mbuf_sg(txq, desc_tag, map, m_headp, segs, &nsegs, max_segs, BUS_DMA_NOWAIT); 2621 defrag: 2622 if (__predict_false(err)) { 2623 switch (err) { 2624 case EFBIG: 2625 /* try collapse once and defrag once */ 2626 if (remap == 0) 2627 m_head = m_collapse(*m_headp, M_NOWAIT, max_segs); 2628 if (remap == 1) 2629 m_head = m_defrag(*m_headp, M_NOWAIT); 2630 remap++; 2631 if (__predict_false(m_head == NULL)) 2632 goto defrag_failed; 2633 txq->ift_mbuf_defrag++; 2634 *m_headp = m_head; 2635 goto retry; 2636 break; 2637 case ENOMEM: 2638 txq->ift_no_tx_dma_setup++; 2639 break; 2640 default: 2641 txq->ift_no_tx_dma_setup++; 2642 m_freem(*m_headp); 2643 DBG_COUNTER_INC(tx_frees); 2644 *m_headp = NULL; 2645 break; 2646 } 2647 txq->ift_map_failed++; 2648 DBG_COUNTER_INC(encap_load_mbuf_fail); 2649 return (err); 2650 } 2651 2652 /* 2653 * XXX assumes a 1 to 1 relationship between segments and 2654 * descriptors - this does not hold true on all drivers, e.g. 2655 * cxgb 2656 */ 2657 if (__predict_false(nsegs + 2 > TXQ_AVAIL(txq))) { 2658 txq->ift_no_desc_avail++; 2659 if (map != NULL) 2660 bus_dmamap_unload(desc_tag, map); 2661 DBG_COUNTER_INC(encap_txq_avail_fail); 2662 if ((txq->ift_task.gt_task.ta_flags & TASK_ENQUEUED) == 0) 2663 GROUPTASK_ENQUEUE(&txq->ift_task); 2664 return (ENOBUFS); 2665 } 2666 pi.ipi_segs = segs; 2667 pi.ipi_nsegs = nsegs; 2668 2669 MPASS(pidx >= 0 && pidx < txq->ift_size); 2670 #ifdef PKT_DEBUG 2671 print_pkt(&pi); 2672 #endif 2673 if ((err = ctx->isc_txd_encap(ctx->ifc_softc, &pi)) == 0) { 2674 bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map, 2675 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); 2676 2677 DBG_COUNTER_INC(tx_encap); 2678 MPASS(pi.ipi_new_pidx >= 0 && 2679 pi.ipi_new_pidx < txq->ift_size); 2680 2681 ndesc = pi.ipi_new_pidx - pi.ipi_pidx; 2682 if (pi.ipi_new_pidx < pi.ipi_pidx) { 2683 ndesc += txq->ift_size; 2684 txq->ift_gen = 1; 2685 } 2686 MPASS(pi.ipi_new_pidx != pidx); 2687 MPASS(ndesc > 0); 2688 txq->ift_in_use += ndesc; 2689 /* 2690 * We update the last software descriptor again here because there may 2691 * be a sentinel and/or there may be more mbufs than segments 2692 */ 2693 txq->ift_pidx = pi.ipi_new_pidx; 2694 txq->ift_npending += pi.ipi_ndescs; 2695 } else if (__predict_false(err == EFBIG && remap < 2)) { 2696 *m_headp = m_head = iflib_remove_mbuf(txq); 2697 remap = 1; 2698 txq->ift_txd_encap_efbig++; 2699 goto defrag; 2700 } else 2701 DBG_COUNTER_INC(encap_txd_encap_fail); 2702 return (err); 2703 2704 defrag_failed: 2705 txq->ift_mbuf_defrag_failed++; 2706 txq->ift_map_failed++; 2707 m_freem(*m_headp); 2708 DBG_COUNTER_INC(tx_frees); 2709 *m_headp = NULL; 2710 return (ENOMEM); 2711 } 2712 2713 /* forward compatibility for cxgb */ 2714 #define FIRST_QSET(ctx) 0 2715 2716 #define NTXQSETS(ctx) ((ctx)->ifc_softc_ctx.isc_ntxqsets) 2717 #define NRXQSETS(ctx) ((ctx)->ifc_softc_ctx.isc_nrxqsets) 2718 #define QIDX(ctx, m) ((((m)->m_pkthdr.flowid & ctx->ifc_softc_ctx.isc_rss_table_mask) % NTXQSETS(ctx)) + FIRST_QSET(ctx)) 2719 #define DESC_RECLAIMABLE(q) ((int)((q)->ift_processed - (q)->ift_cleaned - (q)->ift_ctx->ifc_softc_ctx.isc_tx_nsegments)) 2720 #define RECLAIM_THRESH(ctx) ((ctx)->ifc_sctx->isc_tx_reclaim_thresh) 2721 #define MAX_TX_DESC(ctx) ((ctx)->ifc_softc_ctx.isc_tx_tso_segments_max) 2722 2723 2724 2725 /* if there are more than TXQ_MIN_OCCUPANCY packets pending we consider deferring 2726 * doorbell writes 2727 * 2728 * ORing with 2 assures that min occupancy is never less than 2 without any conditional logic 2729 */ 2730 #define TXQ_MIN_OCCUPANCY(size) ((size >> 6)| 0x2) 2731 2732 static inline int 2733 iflib_txq_min_occupancy(iflib_txq_t txq) 2734 { 2735 if_ctx_t ctx; 2736 2737 ctx = txq->ift_ctx; 2738 return (get_inuse(txq->ift_size, txq->ift_cidx, txq->ift_pidx, 2739 txq->ift_gen) < TXQ_MIN_OCCUPANCY(txq->ift_size) + 2740 MAX_TX_DESC(ctx)); 2741 } 2742 2743 static void 2744 iflib_tx_desc_free(iflib_txq_t txq, int n) 2745 { 2746 int hasmap; 2747 uint32_t qsize, cidx, mask, gen; 2748 struct mbuf *m, **ifsd_m; 2749 uint8_t *ifsd_flags; 2750 bus_dmamap_t *ifsd_map; 2751 2752 cidx = txq->ift_cidx; 2753 gen = txq->ift_gen; 2754 qsize = txq->ift_size; 2755 mask = qsize-1; 2756 hasmap = txq->ift_sds.ifsd_map != NULL; 2757 ifsd_flags = txq->ift_sds.ifsd_flags; 2758 ifsd_m = txq->ift_sds.ifsd_m; 2759 ifsd_map = txq->ift_sds.ifsd_map; 2760 2761 while (n--) { 2762 prefetch(ifsd_m[(cidx + 3) & mask]); 2763 prefetch(ifsd_m[(cidx + 4) & mask]); 2764 2765 if (ifsd_m[cidx] != NULL) { 2766 prefetch(&ifsd_m[(cidx + CACHE_PTR_INCREMENT) & mask]); 2767 prefetch(&ifsd_flags[(cidx + CACHE_PTR_INCREMENT) & mask]); 2768 if (hasmap && (ifsd_flags[cidx] & TX_SW_DESC_MAPPED)) { 2769 /* 2770 * does it matter if it's not the TSO tag? If so we'll 2771 * have to add the type to flags 2772 */ 2773 bus_dmamap_unload(txq->ift_desc_tag, ifsd_map[cidx]); 2774 ifsd_flags[cidx] &= ~TX_SW_DESC_MAPPED; 2775 } 2776 if ((m = ifsd_m[cidx]) != NULL) { 2777 /* XXX we don't support any drivers that batch packets yet */ 2778 MPASS(m->m_nextpkt == NULL); 2779 2780 m_free(m); 2781 ifsd_m[cidx] = NULL; 2782 #if MEMORY_LOGGING 2783 txq->ift_dequeued++; 2784 #endif 2785 DBG_COUNTER_INC(tx_frees); 2786 } 2787 } 2788 if (__predict_false(++cidx == qsize)) { 2789 cidx = 0; 2790 gen = 0; 2791 } 2792 } 2793 txq->ift_cidx = cidx; 2794 txq->ift_gen = gen; 2795 } 2796 2797 static __inline int 2798 iflib_completed_tx_reclaim(iflib_txq_t txq, int thresh) 2799 { 2800 int reclaim; 2801 if_ctx_t ctx = txq->ift_ctx; 2802 2803 KASSERT(thresh >= 0, ("invalid threshold to reclaim")); 2804 MPASS(thresh /*+ MAX_TX_DESC(txq->ift_ctx) */ < txq->ift_size); 2805 2806 /* 2807 * Need a rate-limiting check so that this isn't called every time 2808 */ 2809 iflib_tx_credits_update(ctx, txq); 2810 reclaim = DESC_RECLAIMABLE(txq); 2811 2812 if (reclaim <= thresh /* + MAX_TX_DESC(txq->ift_ctx) */) { 2813 #ifdef INVARIANTS 2814 if (iflib_verbose_debug) { 2815 printf("%s processed=%ju cleaned=%ju tx_nsegments=%d reclaim=%d thresh=%d\n", __FUNCTION__, 2816 txq->ift_processed, txq->ift_cleaned, txq->ift_ctx->ifc_softc_ctx.isc_tx_nsegments, 2817 reclaim, thresh); 2818 2819 } 2820 #endif 2821 return (0); 2822 } 2823 iflib_tx_desc_free(txq, reclaim); 2824 txq->ift_cleaned += reclaim; 2825 txq->ift_in_use -= reclaim; 2826 2827 if (txq->ift_active == FALSE) 2828 txq->ift_active = TRUE; 2829 2830 return (reclaim); 2831 } 2832 2833 static struct mbuf ** 2834 _ring_peek_one(struct ifmp_ring *r, int cidx, int offset) 2835 { 2836 2837 return (__DEVOLATILE(struct mbuf **, &r->items[(cidx + offset) & (r->size-1)])); 2838 } 2839 2840 static void 2841 iflib_txq_check_drain(iflib_txq_t txq, int budget) 2842 { 2843 2844 ifmp_ring_check_drainage(txq->ift_br[0], budget); 2845 } 2846 2847 static uint32_t 2848 iflib_txq_can_drain(struct ifmp_ring *r) 2849 { 2850 iflib_txq_t txq = r->cookie; 2851 if_ctx_t ctx = txq->ift_ctx; 2852 2853 return ((TXQ_AVAIL(txq) >= MAX_TX_DESC(ctx)) || 2854 ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, txq->ift_cidx_processed, false)); 2855 } 2856 2857 static uint32_t 2858 iflib_txq_drain(struct ifmp_ring *r, uint32_t cidx, uint32_t pidx) 2859 { 2860 iflib_txq_t txq = r->cookie; 2861 if_ctx_t ctx = txq->ift_ctx; 2862 if_t ifp = ctx->ifc_ifp; 2863 struct mbuf **mp, *m; 2864 int i, count, consumed, pkt_sent, bytes_sent, mcast_sent, avail, err, in_use_prev, desc_used; 2865 2866 if (__predict_false(!(if_getdrvflags(ifp) & IFF_DRV_RUNNING) || 2867 !LINK_ACTIVE(ctx))) { 2868 DBG_COUNTER_INC(txq_drain_notready); 2869 return (0); 2870 } 2871 2872 avail = IDXDIFF(pidx, cidx, r->size); 2873 if (__predict_false(ctx->ifc_flags & IFC_QFLUSH)) { 2874 DBG_COUNTER_INC(txq_drain_flushing); 2875 for (i = 0; i < avail; i++) { 2876 m_free(r->items[(cidx + i) & (r->size-1)]); 2877 r->items[(cidx + i) & (r->size-1)] = NULL; 2878 } 2879 return (avail); 2880 } 2881 iflib_completed_tx_reclaim(txq, RECLAIM_THRESH(ctx)); 2882 if (__predict_false(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_OACTIVE)) { 2883 txq->ift_qstatus = IFLIB_QUEUE_IDLE; 2884 CALLOUT_LOCK(txq); 2885 callout_stop(&txq->ift_timer); 2886 callout_stop(&txq->ift_db_check); 2887 CALLOUT_UNLOCK(txq); 2888 DBG_COUNTER_INC(txq_drain_oactive); 2889 return (0); 2890 } 2891 consumed = mcast_sent = bytes_sent = pkt_sent = 0; 2892 count = MIN(avail, TX_BATCH_SIZE); 2893 2894 for (desc_used = i = 0; i < count && TXQ_AVAIL(txq) > MAX_TX_DESC(ctx) + 2; i++) { 2895 mp = _ring_peek_one(r, cidx, i); 2896 in_use_prev = txq->ift_in_use; 2897 err = iflib_encap(txq, mp); 2898 /* 2899 * What other errors should we bail out for? 2900 */ 2901 if (err == ENOBUFS) { 2902 DBG_COUNTER_INC(txq_drain_encapfail); 2903 break; 2904 } 2905 consumed++; 2906 if (err) 2907 continue; 2908 2909 pkt_sent++; 2910 m = *mp; 2911 DBG_COUNTER_INC(tx_sent); 2912 bytes_sent += m->m_pkthdr.len; 2913 if (m->m_flags & M_MCAST) 2914 mcast_sent++; 2915 2916 txq->ift_db_pending += (txq->ift_in_use - in_use_prev); 2917 desc_used += (txq->ift_in_use - in_use_prev); 2918 iflib_txd_db_check(ctx, txq, FALSE); 2919 ETHER_BPF_MTAP(ifp, m); 2920 if (__predict_false(!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING))) 2921 break; 2922 2923 if (desc_used > TXQ_MAX_DB_CONSUMED(txq->ift_size)) 2924 break; 2925 } 2926 2927 if ((iflib_min_tx_latency || iflib_txq_min_occupancy(txq)) && txq->ift_db_pending) 2928 iflib_txd_db_check(ctx, txq, TRUE); 2929 else if ((txq->ift_db_pending || TXQ_AVAIL(txq) < MAX_TX_DESC(ctx)) && 2930 (callout_pending(&txq->ift_db_check) == 0)) { 2931 txq->ift_db_pending_queued = txq->ift_db_pending; 2932 callout_reset_on(&txq->ift_db_check, 1, iflib_txd_deferred_db_check, 2933 txq, txq->ift_db_check.c_cpu); 2934 } 2935 if_inc_counter(ifp, IFCOUNTER_OBYTES, bytes_sent); 2936 if_inc_counter(ifp, IFCOUNTER_OPACKETS, pkt_sent); 2937 if (mcast_sent) 2938 if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast_sent); 2939 2940 return (consumed); 2941 } 2942 2943 static void 2944 _task_fn_tx(void *context) 2945 { 2946 iflib_txq_t txq = context; 2947 if_ctx_t ctx = txq->ift_ctx; 2948 2949 if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)) 2950 return; 2951 ifmp_ring_check_drainage(txq->ift_br[0], TX_BATCH_SIZE); 2952 } 2953 2954 static void 2955 _task_fn_rx(void *context) 2956 { 2957 iflib_rxq_t rxq = context; 2958 if_ctx_t ctx = rxq->ifr_ctx; 2959 bool more; 2960 int rc; 2961 2962 DBG_COUNTER_INC(task_fn_rxs); 2963 if (__predict_false(!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING))) 2964 return; 2965 2966 if ((more = iflib_rxeof(rxq, 16 /* XXX */)) == false) { 2967 if (ctx->ifc_flags & IFC_LEGACY) 2968 IFDI_INTR_ENABLE(ctx); 2969 else { 2970 DBG_COUNTER_INC(rx_intr_enables); 2971 rc = IFDI_QUEUE_INTR_ENABLE(ctx, rxq->ifr_id); 2972 KASSERT(rc != ENOTSUP, ("MSI-X support requires queue_intr_enable, but not implemented in driver")); 2973 } 2974 } 2975 if (__predict_false(!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING))) 2976 return; 2977 if (more) 2978 GROUPTASK_ENQUEUE(&rxq->ifr_task); 2979 } 2980 2981 static void 2982 _task_fn_admin(void *context) 2983 { 2984 if_ctx_t ctx = context; 2985 if_softc_ctx_t sctx = &ctx->ifc_softc_ctx; 2986 iflib_txq_t txq; 2987 int i; 2988 2989 if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)) 2990 return; 2991 2992 CTX_LOCK(ctx); 2993 for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++) { 2994 CALLOUT_LOCK(txq); 2995 callout_stop(&txq->ift_timer); 2996 CALLOUT_UNLOCK(txq); 2997 } 2998 IFDI_UPDATE_ADMIN_STATUS(ctx); 2999 for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++) 3000 callout_reset_on(&txq->ift_timer, hz/2, iflib_timer, txq, txq->ift_timer.c_cpu); 3001 IFDI_LINK_INTR_ENABLE(ctx); 3002 CTX_UNLOCK(ctx); 3003 3004 if (LINK_ACTIVE(ctx) == 0) 3005 return; 3006 for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++) 3007 iflib_txq_check_drain(txq, IFLIB_RESTART_BUDGET); 3008 } 3009 3010 3011 static void 3012 _task_fn_iov(void *context) 3013 { 3014 if_ctx_t ctx = context; 3015 3016 if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)) 3017 return; 3018 3019 CTX_LOCK(ctx); 3020 IFDI_VFLR_HANDLE(ctx); 3021 CTX_UNLOCK(ctx); 3022 } 3023 3024 static int 3025 iflib_sysctl_int_delay(SYSCTL_HANDLER_ARGS) 3026 { 3027 int err; 3028 if_int_delay_info_t info; 3029 if_ctx_t ctx; 3030 3031 info = (if_int_delay_info_t)arg1; 3032 ctx = info->iidi_ctx; 3033 info->iidi_req = req; 3034 info->iidi_oidp = oidp; 3035 CTX_LOCK(ctx); 3036 err = IFDI_SYSCTL_INT_DELAY(ctx, info); 3037 CTX_UNLOCK(ctx); 3038 return (err); 3039 } 3040 3041 /********************************************************************* 3042 * 3043 * IFNET FUNCTIONS 3044 * 3045 **********************************************************************/ 3046 3047 static void 3048 iflib_if_init_locked(if_ctx_t ctx) 3049 { 3050 iflib_stop(ctx); 3051 iflib_init_locked(ctx); 3052 } 3053 3054 3055 static void 3056 iflib_if_init(void *arg) 3057 { 3058 if_ctx_t ctx = arg; 3059 3060 CTX_LOCK(ctx); 3061 iflib_if_init_locked(ctx); 3062 CTX_UNLOCK(ctx); 3063 } 3064 3065 static int 3066 iflib_if_transmit(if_t ifp, struct mbuf *m) 3067 { 3068 if_ctx_t ctx = if_getsoftc(ifp); 3069 3070 iflib_txq_t txq; 3071 int err, qidx; 3072 3073 if (__predict_false((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || !LINK_ACTIVE(ctx))) { 3074 DBG_COUNTER_INC(tx_frees); 3075 m_freem(m); 3076 return (0); 3077 } 3078 3079 MPASS(m->m_nextpkt == NULL); 3080 qidx = 0; 3081 if ((NTXQSETS(ctx) > 1) && M_HASHTYPE_GET(m)) 3082 qidx = QIDX(ctx, m); 3083 /* 3084 * XXX calculate buf_ring based on flowid (divvy up bits?) 3085 */ 3086 txq = &ctx->ifc_txqs[qidx]; 3087 3088 #ifdef DRIVER_BACKPRESSURE 3089 if (txq->ift_closed) { 3090 while (m != NULL) { 3091 next = m->m_nextpkt; 3092 m->m_nextpkt = NULL; 3093 m_freem(m); 3094 m = next; 3095 } 3096 return (ENOBUFS); 3097 } 3098 #endif 3099 #ifdef notyet 3100 qidx = count = 0; 3101 mp = marr; 3102 next = m; 3103 do { 3104 count++; 3105 next = next->m_nextpkt; 3106 } while (next != NULL); 3107 3108 if (count > nitems(marr)) 3109 if ((mp = malloc(count*sizeof(struct mbuf *), M_IFLIB, M_NOWAIT)) == NULL) { 3110 /* XXX check nextpkt */ 3111 m_freem(m); 3112 /* XXX simplify for now */ 3113 DBG_COUNTER_INC(tx_frees); 3114 return (ENOBUFS); 3115 } 3116 for (next = m, i = 0; next != NULL; i++) { 3117 mp[i] = next; 3118 next = next->m_nextpkt; 3119 mp[i]->m_nextpkt = NULL; 3120 } 3121 #endif 3122 DBG_COUNTER_INC(tx_seen); 3123 err = ifmp_ring_enqueue(txq->ift_br[0], (void **)&m, 1, TX_BATCH_SIZE); 3124 3125 if (err) { 3126 GROUPTASK_ENQUEUE(&txq->ift_task); 3127 /* support forthcoming later */ 3128 #ifdef DRIVER_BACKPRESSURE 3129 txq->ift_closed = TRUE; 3130 #endif 3131 ifmp_ring_check_drainage(txq->ift_br[0], TX_BATCH_SIZE); 3132 m_freem(m); 3133 } else if (TXQ_AVAIL(txq) < (txq->ift_size >> 1)) { 3134 GROUPTASK_ENQUEUE(&txq->ift_task); 3135 } 3136 3137 return (err); 3138 } 3139 3140 static void 3141 iflib_if_qflush(if_t ifp) 3142 { 3143 if_ctx_t ctx = if_getsoftc(ifp); 3144 iflib_txq_t txq = ctx->ifc_txqs; 3145 int i; 3146 3147 CTX_LOCK(ctx); 3148 ctx->ifc_flags |= IFC_QFLUSH; 3149 CTX_UNLOCK(ctx); 3150 for (i = 0; i < NTXQSETS(ctx); i++, txq++) 3151 while (!(ifmp_ring_is_idle(txq->ift_br[0]) || ifmp_ring_is_stalled(txq->ift_br[0]))) 3152 iflib_txq_check_drain(txq, 0); 3153 CTX_LOCK(ctx); 3154 ctx->ifc_flags &= ~IFC_QFLUSH; 3155 CTX_UNLOCK(ctx); 3156 3157 if_qflush(ifp); 3158 } 3159 3160 3161 #define IFCAP_FLAGS (IFCAP_TXCSUM_IPV6 | IFCAP_RXCSUM_IPV6 | IFCAP_HWCSUM | IFCAP_LRO | \ 3162 IFCAP_TSO4 | IFCAP_TSO6 | IFCAP_VLAN_HWTAGGING | \ 3163 IFCAP_VLAN_MTU | IFCAP_VLAN_HWFILTER | IFCAP_VLAN_HWTSO) 3164 3165 #define IFCAP_REINIT IFCAP_FLAGS 3166 3167 static int 3168 iflib_if_ioctl(if_t ifp, u_long command, caddr_t data) 3169 { 3170 if_ctx_t ctx = if_getsoftc(ifp); 3171 struct ifreq *ifr = (struct ifreq *)data; 3172 #if defined(INET) || defined(INET6) 3173 struct ifaddr *ifa = (struct ifaddr *)data; 3174 #endif 3175 bool avoid_reset = FALSE; 3176 int err = 0, reinit = 0, bits; 3177 3178 switch (command) { 3179 case SIOCSIFADDR: 3180 #ifdef INET 3181 if (ifa->ifa_addr->sa_family == AF_INET) 3182 avoid_reset = TRUE; 3183 #endif 3184 #ifdef INET6 3185 if (ifa->ifa_addr->sa_family == AF_INET6) 3186 avoid_reset = TRUE; 3187 #endif 3188 /* 3189 ** Calling init results in link renegotiation, 3190 ** so we avoid doing it when possible. 3191 */ 3192 if (avoid_reset) { 3193 if_setflagbits(ifp, IFF_UP,0); 3194 if (!(if_getdrvflags(ifp)& IFF_DRV_RUNNING)) 3195 reinit = 1; 3196 #ifdef INET 3197 if (!(if_getflags(ifp) & IFF_NOARP)) 3198 arp_ifinit(ifp, ifa); 3199 #endif 3200 } else 3201 err = ether_ioctl(ifp, command, data); 3202 break; 3203 case SIOCSIFMTU: 3204 CTX_LOCK(ctx); 3205 if (ifr->ifr_mtu == if_getmtu(ifp)) { 3206 CTX_UNLOCK(ctx); 3207 break; 3208 } 3209 bits = if_getdrvflags(ifp); 3210 /* stop the driver and free any clusters before proceeding */ 3211 iflib_stop(ctx); 3212 3213 if ((err = IFDI_MTU_SET(ctx, ifr->ifr_mtu)) == 0) { 3214 if (ifr->ifr_mtu > ctx->ifc_max_fl_buf_size) 3215 ctx->ifc_flags |= IFC_MULTISEG; 3216 else 3217 ctx->ifc_flags &= ~IFC_MULTISEG; 3218 err = if_setmtu(ifp, ifr->ifr_mtu); 3219 } 3220 iflib_init_locked(ctx); 3221 if_setdrvflags(ifp, bits); 3222 CTX_UNLOCK(ctx); 3223 break; 3224 case SIOCSIFFLAGS: 3225 CTX_LOCK(ctx); 3226 if (if_getflags(ifp) & IFF_UP) { 3227 if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) { 3228 if ((if_getflags(ifp) ^ ctx->ifc_if_flags) & 3229 (IFF_PROMISC | IFF_ALLMULTI)) { 3230 err = IFDI_PROMISC_SET(ctx, if_getflags(ifp)); 3231 } 3232 } else 3233 reinit = 1; 3234 } else if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) { 3235 iflib_stop(ctx); 3236 } 3237 ctx->ifc_if_flags = if_getflags(ifp); 3238 CTX_UNLOCK(ctx); 3239 break; 3240 3241 break; 3242 case SIOCADDMULTI: 3243 case SIOCDELMULTI: 3244 if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) { 3245 CTX_LOCK(ctx); 3246 IFDI_INTR_DISABLE(ctx); 3247 IFDI_MULTI_SET(ctx); 3248 IFDI_INTR_ENABLE(ctx); 3249 CTX_UNLOCK(ctx); 3250 } 3251 break; 3252 case SIOCSIFMEDIA: 3253 CTX_LOCK(ctx); 3254 IFDI_MEDIA_SET(ctx); 3255 CTX_UNLOCK(ctx); 3256 /* falls thru */ 3257 case SIOCGIFMEDIA: 3258 err = ifmedia_ioctl(ifp, ifr, &ctx->ifc_media, command); 3259 break; 3260 case SIOCGI2C: 3261 { 3262 struct ifi2creq i2c; 3263 3264 err = copyin(ifr->ifr_data, &i2c, sizeof(i2c)); 3265 if (err != 0) 3266 break; 3267 if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) { 3268 err = EINVAL; 3269 break; 3270 } 3271 if (i2c.len > sizeof(i2c.data)) { 3272 err = EINVAL; 3273 break; 3274 } 3275 3276 if ((err = IFDI_I2C_REQ(ctx, &i2c)) == 0) 3277 err = copyout(&i2c, ifr->ifr_data, sizeof(i2c)); 3278 break; 3279 } 3280 case SIOCSIFCAP: 3281 { 3282 int mask, setmask; 3283 3284 mask = ifr->ifr_reqcap ^ if_getcapenable(ifp); 3285 setmask = 0; 3286 #ifdef TCP_OFFLOAD 3287 setmask |= mask & (IFCAP_TOE4|IFCAP_TOE6); 3288 #endif 3289 setmask |= (mask & IFCAP_FLAGS); 3290 3291 if ((mask & IFCAP_WOL) && 3292 (if_getcapabilities(ifp) & IFCAP_WOL) != 0) 3293 setmask |= (mask & (IFCAP_WOL_MCAST|IFCAP_WOL_MAGIC)); 3294 if_vlancap(ifp); 3295 /* 3296 * want to ensure that traffic has stopped before we change any of the flags 3297 */ 3298 if (setmask) { 3299 CTX_LOCK(ctx); 3300 bits = if_getdrvflags(ifp); 3301 if (setmask & IFCAP_REINIT) 3302 iflib_stop(ctx); 3303 if_togglecapenable(ifp, setmask); 3304 if (setmask & IFCAP_REINIT) 3305 iflib_init_locked(ctx); 3306 if_setdrvflags(ifp, bits); 3307 CTX_UNLOCK(ctx); 3308 } 3309 break; 3310 } 3311 case SIOCGPRIVATE_0: 3312 case SIOCSDRVSPEC: 3313 case SIOCGDRVSPEC: 3314 CTX_LOCK(ctx); 3315 err = IFDI_PRIV_IOCTL(ctx, command, data); 3316 CTX_UNLOCK(ctx); 3317 break; 3318 default: 3319 err = ether_ioctl(ifp, command, data); 3320 break; 3321 } 3322 if (reinit) 3323 iflib_if_init(ctx); 3324 return (err); 3325 } 3326 3327 static uint64_t 3328 iflib_if_get_counter(if_t ifp, ift_counter cnt) 3329 { 3330 if_ctx_t ctx = if_getsoftc(ifp); 3331 3332 return (IFDI_GET_COUNTER(ctx, cnt)); 3333 } 3334 3335 /********************************************************************* 3336 * 3337 * OTHER FUNCTIONS EXPORTED TO THE STACK 3338 * 3339 **********************************************************************/ 3340 3341 static void 3342 iflib_vlan_register(void *arg, if_t ifp, uint16_t vtag) 3343 { 3344 if_ctx_t ctx = if_getsoftc(ifp); 3345 3346 if ((void *)ctx != arg) 3347 return; 3348 3349 if ((vtag == 0) || (vtag > 4095)) 3350 return; 3351 3352 CTX_LOCK(ctx); 3353 IFDI_VLAN_REGISTER(ctx, vtag); 3354 /* Re-init to load the changes */ 3355 if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) 3356 iflib_init_locked(ctx); 3357 CTX_UNLOCK(ctx); 3358 } 3359 3360 static void 3361 iflib_vlan_unregister(void *arg, if_t ifp, uint16_t vtag) 3362 { 3363 if_ctx_t ctx = if_getsoftc(ifp); 3364 3365 if ((void *)ctx != arg) 3366 return; 3367 3368 if ((vtag == 0) || (vtag > 4095)) 3369 return; 3370 3371 CTX_LOCK(ctx); 3372 IFDI_VLAN_UNREGISTER(ctx, vtag); 3373 /* Re-init to load the changes */ 3374 if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) 3375 iflib_init_locked(ctx); 3376 CTX_UNLOCK(ctx); 3377 } 3378 3379 static void 3380 iflib_led_func(void *arg, int onoff) 3381 { 3382 if_ctx_t ctx = arg; 3383 3384 CTX_LOCK(ctx); 3385 IFDI_LED_FUNC(ctx, onoff); 3386 CTX_UNLOCK(ctx); 3387 } 3388 3389 /********************************************************************* 3390 * 3391 * BUS FUNCTION DEFINITIONS 3392 * 3393 **********************************************************************/ 3394 3395 int 3396 iflib_device_probe(device_t dev) 3397 { 3398 pci_vendor_info_t *ent; 3399 3400 uint16_t pci_vendor_id, pci_device_id; 3401 uint16_t pci_subvendor_id, pci_subdevice_id; 3402 uint16_t pci_rev_id; 3403 if_shared_ctx_t sctx; 3404 3405 if ((sctx = DEVICE_REGISTER(dev)) == NULL || sctx->isc_magic != IFLIB_MAGIC) 3406 return (ENOTSUP); 3407 3408 pci_vendor_id = pci_get_vendor(dev); 3409 pci_device_id = pci_get_device(dev); 3410 pci_subvendor_id = pci_get_subvendor(dev); 3411 pci_subdevice_id = pci_get_subdevice(dev); 3412 pci_rev_id = pci_get_revid(dev); 3413 if (sctx->isc_parse_devinfo != NULL) 3414 sctx->isc_parse_devinfo(&pci_device_id, &pci_subvendor_id, &pci_subdevice_id, &pci_rev_id); 3415 3416 ent = sctx->isc_vendor_info; 3417 while (ent->pvi_vendor_id != 0) { 3418 if (pci_vendor_id != ent->pvi_vendor_id) { 3419 ent++; 3420 continue; 3421 } 3422 if ((pci_device_id == ent->pvi_device_id) && 3423 ((pci_subvendor_id == ent->pvi_subvendor_id) || 3424 (ent->pvi_subvendor_id == 0)) && 3425 ((pci_subdevice_id == ent->pvi_subdevice_id) || 3426 (ent->pvi_subdevice_id == 0)) && 3427 ((pci_rev_id == ent->pvi_rev_id) || 3428 (ent->pvi_rev_id == 0))) { 3429 3430 device_set_desc_copy(dev, ent->pvi_name); 3431 /* this needs to be changed to zero if the bus probing code 3432 * ever stops re-probing on best match because the sctx 3433 * may have its values over written by register calls 3434 * in subsequent probes 3435 */ 3436 return (BUS_PROBE_DEFAULT); 3437 } 3438 ent++; 3439 } 3440 return (ENXIO); 3441 } 3442 3443 int 3444 iflib_device_register(device_t dev, void *sc, if_shared_ctx_t sctx, if_ctx_t *ctxp) 3445 { 3446 int err, rid, msix, msix_bar; 3447 if_ctx_t ctx; 3448 if_t ifp; 3449 if_softc_ctx_t scctx; 3450 int i; 3451 uint16_t main_txq; 3452 uint16_t main_rxq; 3453 3454 3455 ctx = malloc(sizeof(* ctx), M_IFLIB, M_WAITOK|M_ZERO); 3456 3457 if (sc == NULL) { 3458 sc = malloc(sctx->isc_driver->size, M_IFLIB, M_WAITOK|M_ZERO); 3459 device_set_softc(dev, ctx); 3460 ctx->ifc_flags |= IFC_SC_ALLOCATED; 3461 } 3462 3463 ctx->ifc_sctx = sctx; 3464 ctx->ifc_dev = dev; 3465 ctx->ifc_txrx = *sctx->isc_txrx; 3466 ctx->ifc_softc = sc; 3467 3468 if ((err = iflib_register(ctx)) != 0) { 3469 device_printf(dev, "iflib_register failed %d\n", err); 3470 return (err); 3471 } 3472 iflib_add_device_sysctl_pre(ctx); 3473 3474 scctx = &ctx->ifc_softc_ctx; 3475 /* 3476 * XXX sanity check that ntxd & nrxd are a power of 2 3477 */ 3478 if (ctx->ifc_sysctl_ntxqs != 0) 3479 scctx->isc_ntxqsets = ctx->ifc_sysctl_ntxqs; 3480 if (ctx->ifc_sysctl_nrxqs != 0) 3481 scctx->isc_nrxqsets = ctx->ifc_sysctl_nrxqs; 3482 3483 for (i = 0; i < sctx->isc_ntxqs; i++) { 3484 if (ctx->ifc_sysctl_ntxds[i] != 0) 3485 scctx->isc_ntxd[i] = ctx->ifc_sysctl_ntxds[i]; 3486 else 3487 scctx->isc_ntxd[i] = sctx->isc_ntxd_default[i]; 3488 } 3489 3490 for (i = 0; i < sctx->isc_nrxqs; i++) { 3491 if (ctx->ifc_sysctl_nrxds[i] != 0) 3492 scctx->isc_nrxd[i] = ctx->ifc_sysctl_nrxds[i]; 3493 else 3494 scctx->isc_nrxd[i] = sctx->isc_nrxd_default[i]; 3495 } 3496 3497 for (i = 0; i < sctx->isc_nrxqs; i++) { 3498 if (scctx->isc_nrxd[i] < sctx->isc_nrxd_min[i]) { 3499 device_printf(dev, "nrxd%d: %d less than nrxd_min %d - resetting to min\n", 3500 i, scctx->isc_nrxd[i], sctx->isc_nrxd_min[i]); 3501 scctx->isc_nrxd[i] = sctx->isc_nrxd_min[i]; 3502 } 3503 if (scctx->isc_nrxd[i] > sctx->isc_nrxd_max[i]) { 3504 device_printf(dev, "nrxd%d: %d greater than nrxd_max %d - resetting to max\n", 3505 i, scctx->isc_nrxd[i], sctx->isc_nrxd_max[i]); 3506 scctx->isc_nrxd[i] = sctx->isc_nrxd_max[i]; 3507 } 3508 } 3509 3510 for (i = 0; i < sctx->isc_ntxqs; i++) { 3511 if (scctx->isc_ntxd[i] < sctx->isc_ntxd_min[i]) { 3512 device_printf(dev, "ntxd%d: %d less than ntxd_min %d - resetting to min\n", 3513 i, scctx->isc_ntxd[i], sctx->isc_ntxd_min[i]); 3514 scctx->isc_ntxd[i] = sctx->isc_ntxd_min[i]; 3515 } 3516 if (scctx->isc_ntxd[i] > sctx->isc_ntxd_max[i]) { 3517 device_printf(dev, "ntxd%d: %d greater than ntxd_max %d - resetting to max\n", 3518 i, scctx->isc_ntxd[i], sctx->isc_ntxd_max[i]); 3519 scctx->isc_ntxd[i] = sctx->isc_ntxd_max[i]; 3520 } 3521 } 3522 3523 if ((err = IFDI_ATTACH_PRE(ctx)) != 0) { 3524 device_printf(dev, "IFDI_ATTACH_PRE failed %d\n", err); 3525 return (err); 3526 } 3527 if (scctx->isc_ntxqsets_max) 3528 scctx->isc_ntxqsets = min(scctx->isc_ntxqsets, scctx->isc_ntxqsets_max); 3529 if (scctx->isc_nrxqsets_max) 3530 scctx->isc_nrxqsets = min(scctx->isc_nrxqsets, scctx->isc_nrxqsets_max); 3531 3532 #ifdef ACPI_DMAR 3533 if (dmar_get_dma_tag(device_get_parent(dev), dev) != NULL) 3534 ctx->ifc_flags |= IFC_DMAR; 3535 #endif 3536 3537 msix_bar = scctx->isc_msix_bar; 3538 3539 ifp = ctx->ifc_ifp; 3540 3541 if(sctx->isc_flags & IFLIB_HAS_TXCQ) 3542 main_txq = 1; 3543 else 3544 main_txq = 0; 3545 3546 if(sctx->isc_flags & IFLIB_HAS_RXCQ) 3547 main_rxq = 1; 3548 else 3549 main_rxq = 0; 3550 3551 /* XXX change for per-queue sizes */ 3552 device_printf(dev, "using %d tx descriptors and %d rx descriptors\n", 3553 scctx->isc_ntxd[main_txq], scctx->isc_nrxd[main_rxq]); 3554 for (i = 0; i < sctx->isc_nrxqs; i++) { 3555 if (!powerof2(scctx->isc_nrxd[i])) { 3556 /* round down instead? */ 3557 device_printf(dev, "# rx descriptors must be a power of 2\n"); 3558 err = EINVAL; 3559 goto fail; 3560 } 3561 } 3562 for (i = 0; i < sctx->isc_ntxqs; i++) { 3563 if (!powerof2(scctx->isc_ntxd[i])) { 3564 device_printf(dev, 3565 "# tx descriptors must be a power of 2"); 3566 err = EINVAL; 3567 goto fail; 3568 } 3569 } 3570 3571 if (scctx->isc_tx_nsegments > scctx->isc_ntxd[main_txq] / 3572 MAX_SINGLE_PACKET_FRACTION) 3573 scctx->isc_tx_nsegments = max(1, scctx->isc_ntxd[main_txq] / 3574 MAX_SINGLE_PACKET_FRACTION); 3575 if (scctx->isc_tx_tso_segments_max > scctx->isc_ntxd[main_txq] / 3576 MAX_SINGLE_PACKET_FRACTION) 3577 scctx->isc_tx_tso_segments_max = max(1, 3578 scctx->isc_ntxd[main_txq] / MAX_SINGLE_PACKET_FRACTION); 3579 3580 /* 3581 * Protect the stack against modern hardware 3582 */ 3583 if (scctx->isc_tx_tso_size_max > FREEBSD_TSO_SIZE_MAX) 3584 scctx->isc_tx_tso_size_max = FREEBSD_TSO_SIZE_MAX; 3585 3586 /* TSO parameters - dig these out of the data sheet - simply correspond to tag setup */ 3587 ifp->if_hw_tsomaxsegcount = scctx->isc_tx_tso_segments_max; 3588 ifp->if_hw_tsomax = scctx->isc_tx_tso_size_max; 3589 ifp->if_hw_tsomaxsegsize = scctx->isc_tx_tso_segsize_max; 3590 if (scctx->isc_rss_table_size == 0) 3591 scctx->isc_rss_table_size = 64; 3592 scctx->isc_rss_table_mask = scctx->isc_rss_table_size-1; 3593 /* 3594 ** Now setup MSI or MSI/X, should 3595 ** return us the number of supported 3596 ** vectors. (Will be 1 for MSI) 3597 */ 3598 if (sctx->isc_flags & IFLIB_SKIP_MSIX) { 3599 msix = scctx->isc_vectors; 3600 } else if (scctx->isc_msix_bar != 0) 3601 msix = iflib_msix_init(ctx); 3602 else { 3603 scctx->isc_vectors = 1; 3604 scctx->isc_ntxqsets = 1; 3605 scctx->isc_nrxqsets = 1; 3606 scctx->isc_intr = IFLIB_INTR_LEGACY; 3607 msix = 0; 3608 } 3609 /* Get memory for the station queues */ 3610 if ((err = iflib_queues_alloc(ctx))) { 3611 device_printf(dev, "Unable to allocate queue memory\n"); 3612 goto fail; 3613 } 3614 3615 if ((err = iflib_qset_structures_setup(ctx))) { 3616 device_printf(dev, "qset structure setup failed %d\n", err); 3617 goto fail_queues; 3618 } 3619 3620 if (msix > 1 && (err = IFDI_MSIX_INTR_ASSIGN(ctx, msix)) != 0) { 3621 device_printf(dev, "IFDI_MSIX_INTR_ASSIGN failed %d\n", err); 3622 goto fail_intr_free; 3623 } 3624 if (msix <= 1) { 3625 rid = 0; 3626 if (scctx->isc_intr == IFLIB_INTR_MSI) { 3627 MPASS(msix == 1); 3628 rid = 1; 3629 } 3630 if ((err = iflib_legacy_setup(ctx, ctx->isc_legacy_intr, ctx->ifc_softc, &rid, "irq0")) != 0) { 3631 device_printf(dev, "iflib_legacy_setup failed %d\n", err); 3632 goto fail_intr_free; 3633 } 3634 } 3635 ether_ifattach(ctx->ifc_ifp, ctx->ifc_mac); 3636 if ((err = IFDI_ATTACH_POST(ctx)) != 0) { 3637 device_printf(dev, "IFDI_ATTACH_POST failed %d\n", err); 3638 goto fail_detach; 3639 } 3640 if ((err = iflib_netmap_attach(ctx))) { 3641 device_printf(ctx->ifc_dev, "netmap attach failed: %d\n", err); 3642 goto fail_detach; 3643 } 3644 *ctxp = ctx; 3645 3646 if_setgetcounterfn(ctx->ifc_ifp, iflib_if_get_counter); 3647 iflib_add_device_sysctl_post(ctx); 3648 return (0); 3649 fail_detach: 3650 ether_ifdetach(ctx->ifc_ifp); 3651 fail_intr_free: 3652 if (scctx->isc_intr == IFLIB_INTR_MSIX || scctx->isc_intr == IFLIB_INTR_MSI) 3653 pci_release_msi(ctx->ifc_dev); 3654 fail_queues: 3655 /* XXX free queues */ 3656 fail: 3657 IFDI_DETACH(ctx); 3658 return (err); 3659 } 3660 3661 int 3662 iflib_device_attach(device_t dev) 3663 { 3664 if_ctx_t ctx; 3665 if_shared_ctx_t sctx; 3666 3667 if ((sctx = DEVICE_REGISTER(dev)) == NULL || sctx->isc_magic != IFLIB_MAGIC) 3668 return (ENOTSUP); 3669 3670 pci_enable_busmaster(dev); 3671 3672 return (iflib_device_register(dev, NULL, sctx, &ctx)); 3673 } 3674 3675 int 3676 iflib_device_deregister(if_ctx_t ctx) 3677 { 3678 if_t ifp = ctx->ifc_ifp; 3679 iflib_txq_t txq; 3680 iflib_rxq_t rxq; 3681 device_t dev = ctx->ifc_dev; 3682 int i; 3683 struct taskqgroup *tqg; 3684 3685 /* Make sure VLANS are not using driver */ 3686 if (if_vlantrunkinuse(ifp)) { 3687 device_printf(dev,"Vlan in use, detach first\n"); 3688 return (EBUSY); 3689 } 3690 3691 CTX_LOCK(ctx); 3692 ctx->ifc_in_detach = 1; 3693 iflib_stop(ctx); 3694 CTX_UNLOCK(ctx); 3695 3696 /* Unregister VLAN events */ 3697 if (ctx->ifc_vlan_attach_event != NULL) 3698 EVENTHANDLER_DEREGISTER(vlan_config, ctx->ifc_vlan_attach_event); 3699 if (ctx->ifc_vlan_detach_event != NULL) 3700 EVENTHANDLER_DEREGISTER(vlan_unconfig, ctx->ifc_vlan_detach_event); 3701 3702 iflib_netmap_detach(ifp); 3703 ether_ifdetach(ifp); 3704 /* ether_ifdetach calls if_qflush - lock must be destroy afterwards*/ 3705 CTX_LOCK_DESTROY(ctx); 3706 if (ctx->ifc_led_dev != NULL) 3707 led_destroy(ctx->ifc_led_dev); 3708 /* XXX drain any dependent tasks */ 3709 tqg = qgroup_if_io_tqg; 3710 for (txq = ctx->ifc_txqs, i = 0; i < NTXQSETS(ctx); i++, txq++) { 3711 callout_drain(&txq->ift_timer); 3712 callout_drain(&txq->ift_db_check); 3713 if (txq->ift_task.gt_uniq != NULL) 3714 taskqgroup_detach(tqg, &txq->ift_task); 3715 } 3716 for (i = 0, rxq = ctx->ifc_rxqs; i < NRXQSETS(ctx); i++, rxq++) { 3717 if (rxq->ifr_task.gt_uniq != NULL) 3718 taskqgroup_detach(tqg, &rxq->ifr_task); 3719 } 3720 tqg = qgroup_if_config_tqg; 3721 if (ctx->ifc_admin_task.gt_uniq != NULL) 3722 taskqgroup_detach(tqg, &ctx->ifc_admin_task); 3723 if (ctx->ifc_vflr_task.gt_uniq != NULL) 3724 taskqgroup_detach(tqg, &ctx->ifc_vflr_task); 3725 3726 IFDI_DETACH(ctx); 3727 device_set_softc(ctx->ifc_dev, NULL); 3728 if (ctx->ifc_softc_ctx.isc_intr != IFLIB_INTR_LEGACY) { 3729 pci_release_msi(dev); 3730 } 3731 if (ctx->ifc_softc_ctx.isc_intr != IFLIB_INTR_MSIX) { 3732 iflib_irq_free(ctx, &ctx->ifc_legacy_irq); 3733 } 3734 if (ctx->ifc_msix_mem != NULL) { 3735 bus_release_resource(ctx->ifc_dev, SYS_RES_MEMORY, 3736 ctx->ifc_softc_ctx.isc_msix_bar, ctx->ifc_msix_mem); 3737 ctx->ifc_msix_mem = NULL; 3738 } 3739 3740 bus_generic_detach(dev); 3741 if_free(ifp); 3742 3743 iflib_tx_structures_free(ctx); 3744 iflib_rx_structures_free(ctx); 3745 if (ctx->ifc_flags & IFC_SC_ALLOCATED) 3746 free(ctx->ifc_softc, M_IFLIB); 3747 free(ctx, M_IFLIB); 3748 return (0); 3749 } 3750 3751 3752 int 3753 iflib_device_detach(device_t dev) 3754 { 3755 if_ctx_t ctx = device_get_softc(dev); 3756 3757 return (iflib_device_deregister(ctx)); 3758 } 3759 3760 int 3761 iflib_device_suspend(device_t dev) 3762 { 3763 if_ctx_t ctx = device_get_softc(dev); 3764 3765 CTX_LOCK(ctx); 3766 IFDI_SUSPEND(ctx); 3767 CTX_UNLOCK(ctx); 3768 3769 return bus_generic_suspend(dev); 3770 } 3771 int 3772 iflib_device_shutdown(device_t dev) 3773 { 3774 if_ctx_t ctx = device_get_softc(dev); 3775 3776 CTX_LOCK(ctx); 3777 IFDI_SHUTDOWN(ctx); 3778 CTX_UNLOCK(ctx); 3779 3780 return bus_generic_suspend(dev); 3781 } 3782 3783 3784 int 3785 iflib_device_resume(device_t dev) 3786 { 3787 if_ctx_t ctx = device_get_softc(dev); 3788 iflib_txq_t txq = ctx->ifc_txqs; 3789 3790 CTX_LOCK(ctx); 3791 IFDI_RESUME(ctx); 3792 iflib_init_locked(ctx); 3793 CTX_UNLOCK(ctx); 3794 for (int i = 0; i < NTXQSETS(ctx); i++, txq++) 3795 iflib_txq_check_drain(txq, IFLIB_RESTART_BUDGET); 3796 3797 return (bus_generic_resume(dev)); 3798 } 3799 3800 int 3801 iflib_device_iov_init(device_t dev, uint16_t num_vfs, const nvlist_t *params) 3802 { 3803 int error; 3804 if_ctx_t ctx = device_get_softc(dev); 3805 3806 CTX_LOCK(ctx); 3807 error = IFDI_IOV_INIT(ctx, num_vfs, params); 3808 CTX_UNLOCK(ctx); 3809 3810 return (error); 3811 } 3812 3813 void 3814 iflib_device_iov_uninit(device_t dev) 3815 { 3816 if_ctx_t ctx = device_get_softc(dev); 3817 3818 CTX_LOCK(ctx); 3819 IFDI_IOV_UNINIT(ctx); 3820 CTX_UNLOCK(ctx); 3821 } 3822 3823 int 3824 iflib_device_iov_add_vf(device_t dev, uint16_t vfnum, const nvlist_t *params) 3825 { 3826 int error; 3827 if_ctx_t ctx = device_get_softc(dev); 3828 3829 CTX_LOCK(ctx); 3830 error = IFDI_IOV_VF_ADD(ctx, vfnum, params); 3831 CTX_UNLOCK(ctx); 3832 3833 return (error); 3834 } 3835 3836 /********************************************************************* 3837 * 3838 * MODULE FUNCTION DEFINITIONS 3839 * 3840 **********************************************************************/ 3841 3842 /* 3843 * - Start a fast taskqueue thread for each core 3844 * - Start a taskqueue for control operations 3845 */ 3846 static int 3847 iflib_module_init(void) 3848 { 3849 return (0); 3850 } 3851 3852 static int 3853 iflib_module_event_handler(module_t mod, int what, void *arg) 3854 { 3855 int err; 3856 3857 switch (what) { 3858 case MOD_LOAD: 3859 if ((err = iflib_module_init()) != 0) 3860 return (err); 3861 break; 3862 case MOD_UNLOAD: 3863 return (EBUSY); 3864 default: 3865 return (EOPNOTSUPP); 3866 } 3867 3868 return (0); 3869 } 3870 3871 /********************************************************************* 3872 * 3873 * PUBLIC FUNCTION DEFINITIONS 3874 * ordered as in iflib.h 3875 * 3876 **********************************************************************/ 3877 3878 3879 static void 3880 _iflib_assert(if_shared_ctx_t sctx) 3881 { 3882 MPASS(sctx->isc_tx_maxsize); 3883 MPASS(sctx->isc_tx_maxsegsize); 3884 3885 MPASS(sctx->isc_rx_maxsize); 3886 MPASS(sctx->isc_rx_nsegments); 3887 MPASS(sctx->isc_rx_maxsegsize); 3888 3889 3890 MPASS(sctx->isc_txrx->ift_txd_encap); 3891 MPASS(sctx->isc_txrx->ift_txd_flush); 3892 MPASS(sctx->isc_txrx->ift_txd_credits_update); 3893 MPASS(sctx->isc_txrx->ift_rxd_available); 3894 MPASS(sctx->isc_txrx->ift_rxd_pkt_get); 3895 MPASS(sctx->isc_txrx->ift_rxd_refill); 3896 MPASS(sctx->isc_txrx->ift_rxd_flush); 3897 3898 MPASS(sctx->isc_nrxd_min[0]); 3899 MPASS(sctx->isc_nrxd_max[0]); 3900 MPASS(sctx->isc_nrxd_default[0]); 3901 MPASS(sctx->isc_ntxd_min[0]); 3902 MPASS(sctx->isc_ntxd_max[0]); 3903 MPASS(sctx->isc_ntxd_default[0]); 3904 } 3905 3906 static int 3907 iflib_register(if_ctx_t ctx) 3908 { 3909 if_shared_ctx_t sctx = ctx->ifc_sctx; 3910 driver_t *driver = sctx->isc_driver; 3911 device_t dev = ctx->ifc_dev; 3912 if_t ifp; 3913 3914 _iflib_assert(sctx); 3915 3916 CTX_LOCK_INIT(ctx, device_get_nameunit(ctx->ifc_dev)); 3917 3918 ifp = ctx->ifc_ifp = if_gethandle(IFT_ETHER); 3919 if (ifp == NULL) { 3920 device_printf(dev, "can not allocate ifnet structure\n"); 3921 return (ENOMEM); 3922 } 3923 3924 /* 3925 * Initialize our context's device specific methods 3926 */ 3927 kobj_init((kobj_t) ctx, (kobj_class_t) driver); 3928 kobj_class_compile((kobj_class_t) driver); 3929 driver->refs++; 3930 3931 if_initname(ifp, device_get_name(dev), device_get_unit(dev)); 3932 if_setsoftc(ifp, ctx); 3933 if_setdev(ifp, dev); 3934 if_setinitfn(ifp, iflib_if_init); 3935 if_setioctlfn(ifp, iflib_if_ioctl); 3936 if_settransmitfn(ifp, iflib_if_transmit); 3937 if_setqflushfn(ifp, iflib_if_qflush); 3938 if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST); 3939 3940 if_setcapabilities(ifp, 0); 3941 if_setcapenable(ifp, 0); 3942 3943 ctx->ifc_vlan_attach_event = 3944 EVENTHANDLER_REGISTER(vlan_config, iflib_vlan_register, ctx, 3945 EVENTHANDLER_PRI_FIRST); 3946 ctx->ifc_vlan_detach_event = 3947 EVENTHANDLER_REGISTER(vlan_unconfig, iflib_vlan_unregister, ctx, 3948 EVENTHANDLER_PRI_FIRST); 3949 3950 ifmedia_init(&ctx->ifc_media, IFM_IMASK, 3951 iflib_media_change, iflib_media_status); 3952 3953 return (0); 3954 } 3955 3956 3957 static int 3958 iflib_queues_alloc(if_ctx_t ctx) 3959 { 3960 if_shared_ctx_t sctx = ctx->ifc_sctx; 3961 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 3962 device_t dev = ctx->ifc_dev; 3963 int nrxqsets = scctx->isc_nrxqsets; 3964 int ntxqsets = scctx->isc_ntxqsets; 3965 iflib_txq_t txq; 3966 iflib_rxq_t rxq; 3967 iflib_fl_t fl = NULL; 3968 int i, j, cpu, err, txconf, rxconf; 3969 iflib_dma_info_t ifdip; 3970 uint32_t *rxqsizes = scctx->isc_rxqsizes; 3971 uint32_t *txqsizes = scctx->isc_txqsizes; 3972 uint8_t nrxqs = sctx->isc_nrxqs; 3973 uint8_t ntxqs = sctx->isc_ntxqs; 3974 int nfree_lists = sctx->isc_nfl ? sctx->isc_nfl : 1; 3975 caddr_t *vaddrs; 3976 uint64_t *paddrs; 3977 struct ifmp_ring **brscp; 3978 int nbuf_rings = 1; /* XXX determine dynamically */ 3979 3980 KASSERT(ntxqs > 0, ("number of queues per qset must be at least 1")); 3981 KASSERT(nrxqs > 0, ("number of queues per qset must be at least 1")); 3982 3983 brscp = NULL; 3984 txq = NULL; 3985 rxq = NULL; 3986 3987 /* Allocate the TX ring struct memory */ 3988 if (!(txq = 3989 (iflib_txq_t) malloc(sizeof(struct iflib_txq) * 3990 ntxqsets, M_IFLIB, M_NOWAIT | M_ZERO))) { 3991 device_printf(dev, "Unable to allocate TX ring memory\n"); 3992 err = ENOMEM; 3993 goto fail; 3994 } 3995 3996 /* Now allocate the RX */ 3997 if (!(rxq = 3998 (iflib_rxq_t) malloc(sizeof(struct iflib_rxq) * 3999 nrxqsets, M_IFLIB, M_NOWAIT | M_ZERO))) { 4000 device_printf(dev, "Unable to allocate RX ring memory\n"); 4001 err = ENOMEM; 4002 goto rx_fail; 4003 } 4004 if (!(brscp = malloc(sizeof(void *) * nbuf_rings * nrxqsets, M_IFLIB, M_NOWAIT | M_ZERO))) { 4005 device_printf(dev, "Unable to buf_ring_sc * memory\n"); 4006 err = ENOMEM; 4007 goto rx_fail; 4008 } 4009 4010 ctx->ifc_txqs = txq; 4011 ctx->ifc_rxqs = rxq; 4012 4013 /* 4014 * XXX handle allocation failure 4015 */ 4016 for (txconf = i = 0, cpu = CPU_FIRST(); i < ntxqsets; i++, txconf++, txq++, cpu = CPU_NEXT(cpu)) { 4017 /* Set up some basics */ 4018 4019 if ((ifdip = malloc(sizeof(struct iflib_dma_info) * ntxqs, M_IFLIB, M_WAITOK|M_ZERO)) == NULL) { 4020 device_printf(dev, "failed to allocate iflib_dma_info\n"); 4021 err = ENOMEM; 4022 goto err_tx_desc; 4023 } 4024 txq->ift_ifdi = ifdip; 4025 for (j = 0; j < ntxqs; j++, ifdip++) { 4026 if (iflib_dma_alloc(ctx, txqsizes[j], ifdip, BUS_DMA_NOWAIT)) { 4027 device_printf(dev, "Unable to allocate Descriptor memory\n"); 4028 err = ENOMEM; 4029 goto err_tx_desc; 4030 } 4031 bzero((void *)ifdip->idi_vaddr, txqsizes[j]); 4032 } 4033 txq->ift_ctx = ctx; 4034 txq->ift_id = i; 4035 if (sctx->isc_flags & IFLIB_HAS_TXCQ) { 4036 txq->ift_br_offset = 1; 4037 } else { 4038 txq->ift_br_offset = 0; 4039 } 4040 /* XXX fix this */ 4041 txq->ift_timer.c_cpu = cpu; 4042 txq->ift_db_check.c_cpu = cpu; 4043 txq->ift_nbr = nbuf_rings; 4044 4045 if (iflib_txsd_alloc(txq)) { 4046 device_printf(dev, "Critical Failure setting up TX buffers\n"); 4047 err = ENOMEM; 4048 goto err_tx_desc; 4049 } 4050 4051 /* Initialize the TX lock */ 4052 snprintf(txq->ift_mtx_name, MTX_NAME_LEN, "%s:tx(%d):callout", 4053 device_get_nameunit(dev), txq->ift_id); 4054 mtx_init(&txq->ift_mtx, txq->ift_mtx_name, NULL, MTX_DEF); 4055 callout_init_mtx(&txq->ift_timer, &txq->ift_mtx, 0); 4056 callout_init_mtx(&txq->ift_db_check, &txq->ift_mtx, 0); 4057 4058 snprintf(txq->ift_db_mtx_name, MTX_NAME_LEN, "%s:tx(%d):db", 4059 device_get_nameunit(dev), txq->ift_id); 4060 TXDB_LOCK_INIT(txq); 4061 4062 txq->ift_br = brscp + i*nbuf_rings; 4063 for (j = 0; j < nbuf_rings; j++) { 4064 err = ifmp_ring_alloc(&txq->ift_br[j], 2048, txq, iflib_txq_drain, 4065 iflib_txq_can_drain, M_IFLIB, M_WAITOK); 4066 if (err) { 4067 /* XXX free any allocated rings */ 4068 device_printf(dev, "Unable to allocate buf_ring\n"); 4069 goto err_tx_desc; 4070 } 4071 } 4072 } 4073 4074 for (rxconf = i = 0; i < nrxqsets; i++, rxconf++, rxq++) { 4075 /* Set up some basics */ 4076 4077 if ((ifdip = malloc(sizeof(struct iflib_dma_info) * nrxqs, M_IFLIB, M_WAITOK|M_ZERO)) == NULL) { 4078 device_printf(dev, "failed to allocate iflib_dma_info\n"); 4079 err = ENOMEM; 4080 goto err_tx_desc; 4081 } 4082 4083 rxq->ifr_ifdi = ifdip; 4084 for (j = 0; j < nrxqs; j++, ifdip++) { 4085 if (iflib_dma_alloc(ctx, rxqsizes[j], ifdip, BUS_DMA_NOWAIT)) { 4086 device_printf(dev, "Unable to allocate Descriptor memory\n"); 4087 err = ENOMEM; 4088 goto err_tx_desc; 4089 } 4090 bzero((void *)ifdip->idi_vaddr, rxqsizes[j]); 4091 } 4092 rxq->ifr_ctx = ctx; 4093 rxq->ifr_id = i; 4094 if (sctx->isc_flags & IFLIB_HAS_RXCQ) { 4095 rxq->ifr_fl_offset = 1; 4096 } else { 4097 rxq->ifr_fl_offset = 0; 4098 } 4099 rxq->ifr_nfl = nfree_lists; 4100 if (!(fl = 4101 (iflib_fl_t) malloc(sizeof(struct iflib_fl) * nfree_lists, M_IFLIB, M_NOWAIT | M_ZERO))) { 4102 device_printf(dev, "Unable to allocate free list memory\n"); 4103 err = ENOMEM; 4104 goto err_tx_desc; 4105 } 4106 rxq->ifr_fl = fl; 4107 for (j = 0; j < nfree_lists; j++) { 4108 rxq->ifr_fl[j].ifl_rxq = rxq; 4109 rxq->ifr_fl[j].ifl_id = j; 4110 rxq->ifr_fl[j].ifl_ifdi = 4111 &rxq->ifr_ifdi[j + rxq->ifr_fl_offset]; 4112 } 4113 /* Allocate receive buffers for the ring*/ 4114 if (iflib_rxsd_alloc(rxq)) { 4115 device_printf(dev, 4116 "Critical Failure setting up receive buffers\n"); 4117 err = ENOMEM; 4118 goto err_rx_desc; 4119 } 4120 } 4121 4122 /* TXQs */ 4123 vaddrs = malloc(sizeof(caddr_t)*ntxqsets*ntxqs, M_IFLIB, M_WAITOK); 4124 paddrs = malloc(sizeof(uint64_t)*ntxqsets*ntxqs, M_IFLIB, M_WAITOK); 4125 for (i = 0; i < ntxqsets; i++) { 4126 iflib_dma_info_t di = ctx->ifc_txqs[i].ift_ifdi; 4127 4128 for (j = 0; j < ntxqs; j++, di++) { 4129 vaddrs[i*ntxqs + j] = di->idi_vaddr; 4130 paddrs[i*ntxqs + j] = di->idi_paddr; 4131 } 4132 } 4133 if ((err = IFDI_TX_QUEUES_ALLOC(ctx, vaddrs, paddrs, ntxqs, ntxqsets)) != 0) { 4134 device_printf(ctx->ifc_dev, "device queue allocation failed\n"); 4135 iflib_tx_structures_free(ctx); 4136 free(vaddrs, M_IFLIB); 4137 free(paddrs, M_IFLIB); 4138 goto err_rx_desc; 4139 } 4140 free(vaddrs, M_IFLIB); 4141 free(paddrs, M_IFLIB); 4142 4143 /* RXQs */ 4144 vaddrs = malloc(sizeof(caddr_t)*nrxqsets*nrxqs, M_IFLIB, M_WAITOK); 4145 paddrs = malloc(sizeof(uint64_t)*nrxqsets*nrxqs, M_IFLIB, M_WAITOK); 4146 for (i = 0; i < nrxqsets; i++) { 4147 iflib_dma_info_t di = ctx->ifc_rxqs[i].ifr_ifdi; 4148 4149 for (j = 0; j < nrxqs; j++, di++) { 4150 vaddrs[i*nrxqs + j] = di->idi_vaddr; 4151 paddrs[i*nrxqs + j] = di->idi_paddr; 4152 } 4153 } 4154 if ((err = IFDI_RX_QUEUES_ALLOC(ctx, vaddrs, paddrs, nrxqs, nrxqsets)) != 0) { 4155 device_printf(ctx->ifc_dev, "device queue allocation failed\n"); 4156 iflib_tx_structures_free(ctx); 4157 free(vaddrs, M_IFLIB); 4158 free(paddrs, M_IFLIB); 4159 goto err_rx_desc; 4160 } 4161 free(vaddrs, M_IFLIB); 4162 free(paddrs, M_IFLIB); 4163 4164 return (0); 4165 4166 /* XXX handle allocation failure changes */ 4167 err_rx_desc: 4168 err_tx_desc: 4169 if (ctx->ifc_rxqs != NULL) 4170 free(ctx->ifc_rxqs, M_IFLIB); 4171 ctx->ifc_rxqs = NULL; 4172 if (ctx->ifc_txqs != NULL) 4173 free(ctx->ifc_txqs, M_IFLIB); 4174 ctx->ifc_txqs = NULL; 4175 rx_fail: 4176 if (brscp != NULL) 4177 free(brscp, M_IFLIB); 4178 if (rxq != NULL) 4179 free(rxq, M_IFLIB); 4180 if (txq != NULL) 4181 free(txq, M_IFLIB); 4182 fail: 4183 return (err); 4184 } 4185 4186 static int 4187 iflib_tx_structures_setup(if_ctx_t ctx) 4188 { 4189 iflib_txq_t txq = ctx->ifc_txqs; 4190 int i; 4191 4192 for (i = 0; i < NTXQSETS(ctx); i++, txq++) 4193 iflib_txq_setup(txq); 4194 4195 return (0); 4196 } 4197 4198 static void 4199 iflib_tx_structures_free(if_ctx_t ctx) 4200 { 4201 iflib_txq_t txq = ctx->ifc_txqs; 4202 int i, j; 4203 4204 for (i = 0; i < NTXQSETS(ctx); i++, txq++) { 4205 iflib_txq_destroy(txq); 4206 for (j = 0; j < ctx->ifc_nhwtxqs; j++) 4207 iflib_dma_free(&txq->ift_ifdi[j]); 4208 } 4209 free(ctx->ifc_txqs, M_IFLIB); 4210 ctx->ifc_txqs = NULL; 4211 IFDI_QUEUES_FREE(ctx); 4212 } 4213 4214 /********************************************************************* 4215 * 4216 * Initialize all receive rings. 4217 * 4218 **********************************************************************/ 4219 static int 4220 iflib_rx_structures_setup(if_ctx_t ctx) 4221 { 4222 iflib_rxq_t rxq = ctx->ifc_rxqs; 4223 int q; 4224 #if defined(INET6) || defined(INET) 4225 int i, err; 4226 #endif 4227 4228 for (q = 0; q < ctx->ifc_softc_ctx.isc_nrxqsets; q++, rxq++) { 4229 #if defined(INET6) || defined(INET) 4230 tcp_lro_free(&rxq->ifr_lc); 4231 if ((err = tcp_lro_init_args(&rxq->ifr_lc, ctx->ifc_ifp, 4232 TCP_LRO_ENTRIES, min(1024, 4233 ctx->ifc_softc_ctx.isc_nrxd[rxq->ifr_fl_offset]))) != 0) { 4234 device_printf(ctx->ifc_dev, "LRO Initialization failed!\n"); 4235 goto fail; 4236 } 4237 rxq->ifr_lro_enabled = TRUE; 4238 #endif 4239 IFDI_RXQ_SETUP(ctx, rxq->ifr_id); 4240 } 4241 return (0); 4242 #if defined(INET6) || defined(INET) 4243 fail: 4244 /* 4245 * Free RX software descriptors allocated so far, we will only handle 4246 * the rings that completed, the failing case will have 4247 * cleaned up for itself. 'q' failed, so its the terminus. 4248 */ 4249 rxq = ctx->ifc_rxqs; 4250 for (i = 0; i < q; ++i, rxq++) { 4251 iflib_rx_sds_free(rxq); 4252 rxq->ifr_cq_gen = rxq->ifr_cq_cidx = rxq->ifr_cq_pidx = 0; 4253 } 4254 return (err); 4255 #endif 4256 } 4257 4258 /********************************************************************* 4259 * 4260 * Free all receive rings. 4261 * 4262 **********************************************************************/ 4263 static void 4264 iflib_rx_structures_free(if_ctx_t ctx) 4265 { 4266 iflib_rxq_t rxq = ctx->ifc_rxqs; 4267 4268 for (int i = 0; i < ctx->ifc_softc_ctx.isc_nrxqsets; i++, rxq++) { 4269 iflib_rx_sds_free(rxq); 4270 } 4271 } 4272 4273 static int 4274 iflib_qset_structures_setup(if_ctx_t ctx) 4275 { 4276 int err; 4277 4278 if ((err = iflib_tx_structures_setup(ctx)) != 0) 4279 return (err); 4280 4281 if ((err = iflib_rx_structures_setup(ctx)) != 0) { 4282 device_printf(ctx->ifc_dev, "iflib_rx_structures_setup failed: %d\n", err); 4283 iflib_tx_structures_free(ctx); 4284 iflib_rx_structures_free(ctx); 4285 } 4286 return (err); 4287 } 4288 4289 int 4290 iflib_irq_alloc(if_ctx_t ctx, if_irq_t irq, int rid, 4291 driver_filter_t filter, void *filter_arg, driver_intr_t handler, void *arg, char *name) 4292 { 4293 4294 return (_iflib_irq_alloc(ctx, irq, rid, filter, handler, arg, name)); 4295 } 4296 4297 static void 4298 find_nth(if_ctx_t ctx, cpuset_t *cpus, int qid) 4299 { 4300 int i, cpuid; 4301 4302 CPU_COPY(&ctx->ifc_cpus, cpus); 4303 /* clear up to the qid'th bit */ 4304 for (i = 0; i < qid; i++) { 4305 cpuid = CPU_FFS(cpus); 4306 CPU_CLR(cpuid, cpus); 4307 } 4308 } 4309 4310 int 4311 iflib_irq_alloc_generic(if_ctx_t ctx, if_irq_t irq, int rid, 4312 iflib_intr_type_t type, driver_filter_t *filter, 4313 void *filter_arg, int qid, char *name) 4314 { 4315 struct grouptask *gtask; 4316 struct taskqgroup *tqg; 4317 iflib_filter_info_t info; 4318 cpuset_t cpus; 4319 gtask_fn_t *fn; 4320 int tqrid, err; 4321 void *q; 4322 4323 info = &ctx->ifc_filter_info; 4324 4325 switch (type) { 4326 /* XXX merge tx/rx for netmap? */ 4327 case IFLIB_INTR_TX: 4328 q = &ctx->ifc_txqs[qid]; 4329 info = &ctx->ifc_txqs[qid].ift_filter_info; 4330 gtask = &ctx->ifc_txqs[qid].ift_task; 4331 tqg = qgroup_if_io_tqg; 4332 tqrid = irq->ii_rid; 4333 fn = _task_fn_tx; 4334 break; 4335 case IFLIB_INTR_RX: 4336 q = &ctx->ifc_rxqs[qid]; 4337 info = &ctx->ifc_rxqs[qid].ifr_filter_info; 4338 gtask = &ctx->ifc_rxqs[qid].ifr_task; 4339 tqg = qgroup_if_io_tqg; 4340 tqrid = irq->ii_rid; 4341 fn = _task_fn_rx; 4342 break; 4343 case IFLIB_INTR_ADMIN: 4344 q = ctx; 4345 info = &ctx->ifc_filter_info; 4346 gtask = &ctx->ifc_admin_task; 4347 tqg = qgroup_if_config_tqg; 4348 tqrid = -1; 4349 fn = _task_fn_admin; 4350 break; 4351 default: 4352 panic("unknown net intr type"); 4353 } 4354 GROUPTASK_INIT(gtask, 0, fn, q); 4355 4356 info->ifi_filter = filter; 4357 info->ifi_filter_arg = filter_arg; 4358 info->ifi_task = gtask; 4359 4360 /* XXX query cpu that rid belongs to */ 4361 4362 err = _iflib_irq_alloc(ctx, irq, rid, iflib_fast_intr, NULL, info, name); 4363 if (err != 0) 4364 return (err); 4365 if (tqrid != -1) { 4366 find_nth(ctx, &cpus, qid); 4367 taskqgroup_attach_cpu(tqg, gtask, q, CPU_FFS(&cpus), irq->ii_rid, name); 4368 } else 4369 taskqgroup_attach(tqg, gtask, q, tqrid, name); 4370 4371 4372 return (0); 4373 } 4374 4375 void 4376 iflib_softirq_alloc_generic(if_ctx_t ctx, int rid, iflib_intr_type_t type, void *arg, int qid, char *name) 4377 { 4378 struct grouptask *gtask; 4379 struct taskqgroup *tqg; 4380 gtask_fn_t *fn; 4381 void *q; 4382 4383 switch (type) { 4384 case IFLIB_INTR_TX: 4385 q = &ctx->ifc_txqs[qid]; 4386 gtask = &ctx->ifc_txqs[qid].ift_task; 4387 tqg = qgroup_if_io_tqg; 4388 fn = _task_fn_tx; 4389 break; 4390 case IFLIB_INTR_RX: 4391 q = &ctx->ifc_rxqs[qid]; 4392 gtask = &ctx->ifc_rxqs[qid].ifr_task; 4393 tqg = qgroup_if_io_tqg; 4394 fn = _task_fn_rx; 4395 break; 4396 case IFLIB_INTR_ADMIN: 4397 q = ctx; 4398 gtask = &ctx->ifc_admin_task; 4399 tqg = qgroup_if_config_tqg; 4400 rid = -1; 4401 fn = _task_fn_admin; 4402 break; 4403 case IFLIB_INTR_IOV: 4404 q = ctx; 4405 gtask = &ctx->ifc_vflr_task; 4406 tqg = qgroup_if_config_tqg; 4407 rid = -1; 4408 fn = _task_fn_iov; 4409 break; 4410 default: 4411 panic("unknown net intr type"); 4412 } 4413 GROUPTASK_INIT(gtask, 0, fn, q); 4414 taskqgroup_attach(tqg, gtask, q, rid, name); 4415 } 4416 4417 void 4418 iflib_irq_free(if_ctx_t ctx, if_irq_t irq) 4419 { 4420 if (irq->ii_tag) 4421 bus_teardown_intr(ctx->ifc_dev, irq->ii_res, irq->ii_tag); 4422 4423 if (irq->ii_res) 4424 bus_release_resource(ctx->ifc_dev, SYS_RES_IRQ, irq->ii_rid, irq->ii_res); 4425 } 4426 4427 static int 4428 iflib_legacy_setup(if_ctx_t ctx, driver_filter_t filter, void *filter_arg, int *rid, char *name) 4429 { 4430 iflib_txq_t txq = ctx->ifc_txqs; 4431 iflib_rxq_t rxq = ctx->ifc_rxqs; 4432 if_irq_t irq = &ctx->ifc_legacy_irq; 4433 iflib_filter_info_t info; 4434 struct grouptask *gtask; 4435 struct taskqgroup *tqg; 4436 gtask_fn_t *fn; 4437 int tqrid; 4438 void *q; 4439 int err; 4440 4441 q = &ctx->ifc_rxqs[0]; 4442 info = &rxq[0].ifr_filter_info; 4443 gtask = &rxq[0].ifr_task; 4444 tqg = qgroup_if_io_tqg; 4445 tqrid = irq->ii_rid = *rid; 4446 fn = _task_fn_rx; 4447 4448 ctx->ifc_flags |= IFC_LEGACY; 4449 info->ifi_filter = filter; 4450 info->ifi_filter_arg = filter_arg; 4451 info->ifi_task = gtask; 4452 4453 /* We allocate a single interrupt resource */ 4454 if ((err = _iflib_irq_alloc(ctx, irq, tqrid, iflib_fast_intr, NULL, info, name)) != 0) 4455 return (err); 4456 GROUPTASK_INIT(gtask, 0, fn, q); 4457 taskqgroup_attach(tqg, gtask, q, tqrid, name); 4458 4459 GROUPTASK_INIT(&txq->ift_task, 0, _task_fn_tx, txq); 4460 taskqgroup_attach(qgroup_if_io_tqg, &txq->ift_task, txq, tqrid, "tx"); 4461 GROUPTASK_INIT(&ctx->ifc_admin_task, 0, _task_fn_admin, ctx); 4462 taskqgroup_attach(qgroup_if_config_tqg, &ctx->ifc_admin_task, ctx, -1, "admin/link"); 4463 4464 return (0); 4465 } 4466 4467 void 4468 iflib_led_create(if_ctx_t ctx) 4469 { 4470 4471 ctx->ifc_led_dev = led_create(iflib_led_func, ctx, 4472 device_get_nameunit(ctx->ifc_dev)); 4473 } 4474 4475 void 4476 iflib_tx_intr_deferred(if_ctx_t ctx, int txqid) 4477 { 4478 4479 GROUPTASK_ENQUEUE(&ctx->ifc_txqs[txqid].ift_task); 4480 } 4481 4482 void 4483 iflib_rx_intr_deferred(if_ctx_t ctx, int rxqid) 4484 { 4485 4486 GROUPTASK_ENQUEUE(&ctx->ifc_rxqs[rxqid].ifr_task); 4487 } 4488 4489 void 4490 iflib_admin_intr_deferred(if_ctx_t ctx) 4491 { 4492 4493 GROUPTASK_ENQUEUE(&ctx->ifc_admin_task); 4494 } 4495 4496 void 4497 iflib_iov_intr_deferred(if_ctx_t ctx) 4498 { 4499 4500 GROUPTASK_ENQUEUE(&ctx->ifc_vflr_task); 4501 } 4502 4503 void 4504 iflib_io_tqg_attach(struct grouptask *gt, void *uniq, int cpu, char *name) 4505 { 4506 4507 taskqgroup_attach_cpu(qgroup_if_io_tqg, gt, uniq, cpu, -1, name); 4508 } 4509 4510 void 4511 iflib_config_gtask_init(if_ctx_t ctx, struct grouptask *gtask, gtask_fn_t *fn, 4512 char *name) 4513 { 4514 4515 GROUPTASK_INIT(gtask, 0, fn, ctx); 4516 taskqgroup_attach(qgroup_if_config_tqg, gtask, gtask, -1, name); 4517 } 4518 4519 void 4520 iflib_config_gtask_deinit(struct grouptask *gtask) 4521 { 4522 4523 taskqgroup_detach(qgroup_if_config_tqg, gtask); 4524 } 4525 4526 void 4527 iflib_link_state_change(if_ctx_t ctx, int link_state, uint64_t baudrate) 4528 { 4529 if_t ifp = ctx->ifc_ifp; 4530 iflib_txq_t txq = ctx->ifc_txqs; 4531 4532 4533 if_setbaudrate(ifp, baudrate); 4534 4535 /* If link down, disable watchdog */ 4536 if ((ctx->ifc_link_state == LINK_STATE_UP) && (link_state == LINK_STATE_DOWN)) { 4537 for (int i = 0; i < ctx->ifc_softc_ctx.isc_ntxqsets; i++, txq++) 4538 txq->ift_qstatus = IFLIB_QUEUE_IDLE; 4539 } 4540 ctx->ifc_link_state = link_state; 4541 if_link_state_change(ifp, link_state); 4542 } 4543 4544 static int 4545 iflib_tx_credits_update(if_ctx_t ctx, iflib_txq_t txq) 4546 { 4547 int credits; 4548 4549 if (ctx->isc_txd_credits_update == NULL) 4550 return (0); 4551 4552 if ((credits = ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, txq->ift_cidx_processed, true)) == 0) 4553 return (0); 4554 4555 txq->ift_processed += credits; 4556 txq->ift_cidx_processed += credits; 4557 4558 if (txq->ift_cidx_processed >= txq->ift_size) 4559 txq->ift_cidx_processed -= txq->ift_size; 4560 return (credits); 4561 } 4562 4563 static int 4564 iflib_rxd_avail(if_ctx_t ctx, iflib_rxq_t rxq, int cidx, int budget) 4565 { 4566 4567 return (ctx->isc_rxd_available(ctx->ifc_softc, rxq->ifr_id, cidx, 4568 budget)); 4569 } 4570 4571 void 4572 iflib_add_int_delay_sysctl(if_ctx_t ctx, const char *name, 4573 const char *description, if_int_delay_info_t info, 4574 int offset, int value) 4575 { 4576 info->iidi_ctx = ctx; 4577 info->iidi_offset = offset; 4578 info->iidi_value = value; 4579 SYSCTL_ADD_PROC(device_get_sysctl_ctx(ctx->ifc_dev), 4580 SYSCTL_CHILDREN(device_get_sysctl_tree(ctx->ifc_dev)), 4581 OID_AUTO, name, CTLTYPE_INT|CTLFLAG_RW, 4582 info, 0, iflib_sysctl_int_delay, "I", description); 4583 } 4584 4585 struct mtx * 4586 iflib_ctx_lock_get(if_ctx_t ctx) 4587 { 4588 4589 return (&ctx->ifc_mtx); 4590 } 4591 4592 static int 4593 iflib_msix_init(if_ctx_t ctx) 4594 { 4595 device_t dev = ctx->ifc_dev; 4596 if_shared_ctx_t sctx = ctx->ifc_sctx; 4597 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 4598 int vectors, queues, rx_queues, tx_queues, queuemsgs, msgs; 4599 int iflib_num_tx_queues, iflib_num_rx_queues; 4600 int err, admincnt, bar; 4601 4602 iflib_num_tx_queues = scctx->isc_ntxqsets; 4603 iflib_num_rx_queues = scctx->isc_nrxqsets; 4604 4605 bar = ctx->ifc_softc_ctx.isc_msix_bar; 4606 admincnt = sctx->isc_admin_intrcnt; 4607 /* Override by tuneable */ 4608 if (enable_msix == 0) 4609 goto msi; 4610 4611 /* 4612 ** When used in a virtualized environment 4613 ** PCI BUSMASTER capability may not be set 4614 ** so explicity set it here and rewrite 4615 ** the ENABLE in the MSIX control register 4616 ** at this point to cause the host to 4617 ** successfully initialize us. 4618 */ 4619 { 4620 uint16_t pci_cmd_word; 4621 int msix_ctrl, rid; 4622 4623 rid = 0; 4624 pci_cmd_word = pci_read_config(dev, PCIR_COMMAND, 2); 4625 pci_cmd_word |= PCIM_CMD_BUSMASTEREN; 4626 pci_write_config(dev, PCIR_COMMAND, pci_cmd_word, 2); 4627 pci_find_cap(dev, PCIY_MSIX, &rid); 4628 rid += PCIR_MSIX_CTRL; 4629 msix_ctrl = pci_read_config(dev, rid, 2); 4630 msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE; 4631 pci_write_config(dev, rid, msix_ctrl, 2); 4632 } 4633 4634 /* 4635 * bar == -1 => "trust me I know what I'm doing" 4636 * https://www.youtube.com/watch?v=nnwWKkNau4I 4637 * Some drivers are for hardware that is so shoddily 4638 * documented that no one knows which bars are which 4639 * so the developer has to map all bars. This hack 4640 * allows shoddy garbage to use msix in this framework. 4641 */ 4642 if (bar != -1) { 4643 ctx->ifc_msix_mem = bus_alloc_resource_any(dev, 4644 SYS_RES_MEMORY, &bar, RF_ACTIVE); 4645 if (ctx->ifc_msix_mem == NULL) { 4646 /* May not be enabled */ 4647 device_printf(dev, "Unable to map MSIX table \n"); 4648 goto msi; 4649 } 4650 } 4651 /* First try MSI/X */ 4652 if ((msgs = pci_msix_count(dev)) == 0) { /* system has msix disabled */ 4653 device_printf(dev, "System has MSIX disabled \n"); 4654 bus_release_resource(dev, SYS_RES_MEMORY, 4655 bar, ctx->ifc_msix_mem); 4656 ctx->ifc_msix_mem = NULL; 4657 goto msi; 4658 } 4659 #if IFLIB_DEBUG 4660 /* use only 1 qset in debug mode */ 4661 queuemsgs = min(msgs - admincnt, 1); 4662 #else 4663 queuemsgs = msgs - admincnt; 4664 #endif 4665 if (bus_get_cpus(dev, INTR_CPUS, sizeof(ctx->ifc_cpus), &ctx->ifc_cpus) == 0) { 4666 #ifdef RSS 4667 queues = imin(queuemsgs, rss_getnumbuckets()); 4668 #else 4669 queues = queuemsgs; 4670 #endif 4671 queues = imin(CPU_COUNT(&ctx->ifc_cpus), queues); 4672 device_printf(dev, "pxm cpus: %d queue msgs: %d admincnt: %d\n", 4673 CPU_COUNT(&ctx->ifc_cpus), queuemsgs, admincnt); 4674 } else { 4675 device_printf(dev, "Unable to fetch CPU list\n"); 4676 /* Figure out a reasonable auto config value */ 4677 queues = min(queuemsgs, mp_ncpus); 4678 } 4679 #ifdef RSS 4680 /* If we're doing RSS, clamp at the number of RSS buckets */ 4681 if (queues > rss_getnumbuckets()) 4682 queues = rss_getnumbuckets(); 4683 #endif 4684 if (iflib_num_rx_queues > 0 && iflib_num_rx_queues < queuemsgs - admincnt) 4685 rx_queues = iflib_num_rx_queues; 4686 else 4687 rx_queues = queues; 4688 /* 4689 * We want this to be all logical CPUs by default 4690 */ 4691 if (iflib_num_tx_queues > 0 && iflib_num_tx_queues < queues) 4692 tx_queues = iflib_num_tx_queues; 4693 else 4694 tx_queues = mp_ncpus; 4695 4696 if (ctx->ifc_sysctl_qs_eq_override == 0) { 4697 #ifdef INVARIANTS 4698 if (tx_queues != rx_queues) 4699 device_printf(dev, "queue equality override not set, capping rx_queues at %d and tx_queues at %d\n", 4700 min(rx_queues, tx_queues), min(rx_queues, tx_queues)); 4701 #endif 4702 tx_queues = min(rx_queues, tx_queues); 4703 rx_queues = min(rx_queues, tx_queues); 4704 } 4705 4706 device_printf(dev, "using %d rx queues %d tx queues \n", rx_queues, tx_queues); 4707 4708 vectors = rx_queues + admincnt; 4709 if ((err = pci_alloc_msix(dev, &vectors)) == 0) { 4710 device_printf(dev, 4711 "Using MSIX interrupts with %d vectors\n", vectors); 4712 scctx->isc_vectors = vectors; 4713 scctx->isc_nrxqsets = rx_queues; 4714 scctx->isc_ntxqsets = tx_queues; 4715 scctx->isc_intr = IFLIB_INTR_MSIX; 4716 4717 return (vectors); 4718 } else { 4719 device_printf(dev, "failed to allocate %d msix vectors, err: %d - using MSI\n", vectors, err); 4720 } 4721 msi: 4722 vectors = pci_msi_count(dev); 4723 scctx->isc_nrxqsets = 1; 4724 scctx->isc_ntxqsets = 1; 4725 scctx->isc_vectors = vectors; 4726 if (vectors == 1 && pci_alloc_msi(dev, &vectors) == 0) { 4727 device_printf(dev,"Using an MSI interrupt\n"); 4728 scctx->isc_intr = IFLIB_INTR_MSI; 4729 } else { 4730 device_printf(dev,"Using a Legacy interrupt\n"); 4731 scctx->isc_intr = IFLIB_INTR_LEGACY; 4732 } 4733 4734 return (vectors); 4735 } 4736 4737 char * ring_states[] = { "IDLE", "BUSY", "STALLED", "ABDICATED" }; 4738 4739 static int 4740 mp_ring_state_handler(SYSCTL_HANDLER_ARGS) 4741 { 4742 int rc; 4743 uint16_t *state = ((uint16_t *)oidp->oid_arg1); 4744 struct sbuf *sb; 4745 char *ring_state = "UNKNOWN"; 4746 4747 /* XXX needed ? */ 4748 rc = sysctl_wire_old_buffer(req, 0); 4749 MPASS(rc == 0); 4750 if (rc != 0) 4751 return (rc); 4752 sb = sbuf_new_for_sysctl(NULL, NULL, 80, req); 4753 MPASS(sb != NULL); 4754 if (sb == NULL) 4755 return (ENOMEM); 4756 if (state[3] <= 3) 4757 ring_state = ring_states[state[3]]; 4758 4759 sbuf_printf(sb, "pidx_head: %04hd pidx_tail: %04hd cidx: %04hd state: %s", 4760 state[0], state[1], state[2], ring_state); 4761 rc = sbuf_finish(sb); 4762 sbuf_delete(sb); 4763 return(rc); 4764 } 4765 4766 enum iflib_ndesc_handler { 4767 IFLIB_NTXD_HANDLER, 4768 IFLIB_NRXD_HANDLER, 4769 }; 4770 4771 static int 4772 mp_ndesc_handler(SYSCTL_HANDLER_ARGS) 4773 { 4774 if_ctx_t ctx = (void *)arg1; 4775 enum iflib_ndesc_handler type = arg2; 4776 char buf[256] = {0}; 4777 uint16_t *ndesc; 4778 char *p, *next; 4779 int nqs, rc, i; 4780 4781 MPASS(type == IFLIB_NTXD_HANDLER || type == IFLIB_NRXD_HANDLER); 4782 4783 nqs = 8; 4784 switch(type) { 4785 case IFLIB_NTXD_HANDLER: 4786 ndesc = ctx->ifc_sysctl_ntxds; 4787 if (ctx->ifc_sctx) 4788 nqs = ctx->ifc_sctx->isc_ntxqs; 4789 break; 4790 case IFLIB_NRXD_HANDLER: 4791 ndesc = ctx->ifc_sysctl_nrxds; 4792 if (ctx->ifc_sctx) 4793 nqs = ctx->ifc_sctx->isc_nrxqs; 4794 break; 4795 } 4796 if (nqs == 0) 4797 nqs = 8; 4798 4799 for (i=0; i<8; i++) { 4800 if (i >= nqs) 4801 break; 4802 if (i) 4803 strcat(buf, ","); 4804 sprintf(strchr(buf, 0), "%d", ndesc[i]); 4805 } 4806 4807 rc = sysctl_handle_string(oidp, buf, sizeof(buf), req); 4808 if (rc || req->newptr == NULL) 4809 return rc; 4810 4811 for (i = 0, next = buf, p = strsep(&next, " ,"); i < 8 && p; 4812 i++, p = strsep(&next, " ,")) { 4813 ndesc[i] = strtoul(p, NULL, 10); 4814 } 4815 4816 return(rc); 4817 } 4818 4819 #define NAME_BUFLEN 32 4820 static void 4821 iflib_add_device_sysctl_pre(if_ctx_t ctx) 4822 { 4823 device_t dev = iflib_get_dev(ctx); 4824 struct sysctl_oid_list *child, *oid_list; 4825 struct sysctl_ctx_list *ctx_list; 4826 struct sysctl_oid *node; 4827 4828 ctx_list = device_get_sysctl_ctx(dev); 4829 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev)); 4830 ctx->ifc_sysctl_node = node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, "iflib", 4831 CTLFLAG_RD, NULL, "IFLIB fields"); 4832 oid_list = SYSCTL_CHILDREN(node); 4833 4834 SYSCTL_ADD_STRING(ctx_list, oid_list, OID_AUTO, "driver_version", 4835 CTLFLAG_RD, ctx->ifc_sctx->isc_driver_version, 0, 4836 "driver version"); 4837 4838 SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_ntxqs", 4839 CTLFLAG_RWTUN, &ctx->ifc_sysctl_ntxqs, 0, 4840 "# of txqs to use, 0 => use default #"); 4841 SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_nrxqs", 4842 CTLFLAG_RWTUN, &ctx->ifc_sysctl_nrxqs, 0, 4843 "# of rxqs to use, 0 => use default #"); 4844 SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_qs_enable", 4845 CTLFLAG_RWTUN, &ctx->ifc_sysctl_qs_eq_override, 0, 4846 "permit #txq != #rxq"); 4847 4848 /* XXX change for per-queue sizes */ 4849 SYSCTL_ADD_PROC(ctx_list, oid_list, OID_AUTO, "override_ntxds", 4850 CTLTYPE_STRING|CTLFLAG_RWTUN, ctx, IFLIB_NTXD_HANDLER, 4851 mp_ndesc_handler, "A", 4852 "list of # of tx descriptors to use, 0 = use default #"); 4853 SYSCTL_ADD_PROC(ctx_list, oid_list, OID_AUTO, "override_nrxds", 4854 CTLTYPE_STRING|CTLFLAG_RWTUN, ctx, IFLIB_NRXD_HANDLER, 4855 mp_ndesc_handler, "A", 4856 "list of # of rx descriptors to use, 0 = use default #"); 4857 } 4858 4859 static void 4860 iflib_add_device_sysctl_post(if_ctx_t ctx) 4861 { 4862 if_shared_ctx_t sctx = ctx->ifc_sctx; 4863 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 4864 device_t dev = iflib_get_dev(ctx); 4865 struct sysctl_oid_list *child; 4866 struct sysctl_ctx_list *ctx_list; 4867 iflib_fl_t fl; 4868 iflib_txq_t txq; 4869 iflib_rxq_t rxq; 4870 int i, j; 4871 char namebuf[NAME_BUFLEN]; 4872 char *qfmt; 4873 struct sysctl_oid *queue_node, *fl_node, *node; 4874 struct sysctl_oid_list *queue_list, *fl_list; 4875 ctx_list = device_get_sysctl_ctx(dev); 4876 4877 node = ctx->ifc_sysctl_node; 4878 child = SYSCTL_CHILDREN(node); 4879 4880 if (scctx->isc_ntxqsets > 100) 4881 qfmt = "txq%03d"; 4882 else if (scctx->isc_ntxqsets > 10) 4883 qfmt = "txq%02d"; 4884 else 4885 qfmt = "txq%d"; 4886 for (i = 0, txq = ctx->ifc_txqs; i < scctx->isc_ntxqsets; i++, txq++) { 4887 snprintf(namebuf, NAME_BUFLEN, qfmt, i); 4888 queue_node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, namebuf, 4889 CTLFLAG_RD, NULL, "Queue Name"); 4890 queue_list = SYSCTL_CHILDREN(queue_node); 4891 #if MEMORY_LOGGING 4892 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_dequeued", 4893 CTLFLAG_RD, 4894 &txq->ift_dequeued, "total mbufs freed"); 4895 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_enqueued", 4896 CTLFLAG_RD, 4897 &txq->ift_enqueued, "total mbufs enqueued"); 4898 #endif 4899 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "mbuf_defrag", 4900 CTLFLAG_RD, 4901 &txq->ift_mbuf_defrag, "# of times m_defrag was called"); 4902 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "m_pullups", 4903 CTLFLAG_RD, 4904 &txq->ift_pullups, "# of times m_pullup was called"); 4905 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "mbuf_defrag_failed", 4906 CTLFLAG_RD, 4907 &txq->ift_mbuf_defrag_failed, "# of times m_defrag failed"); 4908 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "no_desc_avail", 4909 CTLFLAG_RD, 4910 &txq->ift_no_desc_avail, "# of times no descriptors were available"); 4911 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "tx_map_failed", 4912 CTLFLAG_RD, 4913 &txq->ift_map_failed, "# of times dma map failed"); 4914 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txd_encap_efbig", 4915 CTLFLAG_RD, 4916 &txq->ift_txd_encap_efbig, "# of times txd_encap returned EFBIG"); 4917 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "no_tx_dma_setup", 4918 CTLFLAG_RD, 4919 &txq->ift_no_tx_dma_setup, "# of times map failed for other than EFBIG"); 4920 SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_pidx", 4921 CTLFLAG_RD, 4922 &txq->ift_pidx, 1, "Producer Index"); 4923 SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_cidx", 4924 CTLFLAG_RD, 4925 &txq->ift_cidx, 1, "Consumer Index"); 4926 SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_cidx_processed", 4927 CTLFLAG_RD, 4928 &txq->ift_cidx_processed, 1, "Consumer Index seen by credit update"); 4929 SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_in_use", 4930 CTLFLAG_RD, 4931 &txq->ift_in_use, 1, "descriptors in use"); 4932 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_processed", 4933 CTLFLAG_RD, 4934 &txq->ift_processed, "descriptors procesed for clean"); 4935 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_cleaned", 4936 CTLFLAG_RD, 4937 &txq->ift_cleaned, "total cleaned"); 4938 SYSCTL_ADD_PROC(ctx_list, queue_list, OID_AUTO, "ring_state", 4939 CTLTYPE_STRING | CTLFLAG_RD, __DEVOLATILE(uint64_t *, &txq->ift_br[0]->state), 4940 0, mp_ring_state_handler, "A", "soft ring state"); 4941 SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_enqueues", 4942 CTLFLAG_RD, &txq->ift_br[0]->enqueues, 4943 "# of enqueues to the mp_ring for this queue"); 4944 SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_drops", 4945 CTLFLAG_RD, &txq->ift_br[0]->drops, 4946 "# of drops in the mp_ring for this queue"); 4947 SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_starts", 4948 CTLFLAG_RD, &txq->ift_br[0]->starts, 4949 "# of normal consumer starts in the mp_ring for this queue"); 4950 SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_stalls", 4951 CTLFLAG_RD, &txq->ift_br[0]->stalls, 4952 "# of consumer stalls in the mp_ring for this queue"); 4953 SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_restarts", 4954 CTLFLAG_RD, &txq->ift_br[0]->restarts, 4955 "# of consumer restarts in the mp_ring for this queue"); 4956 SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_abdications", 4957 CTLFLAG_RD, &txq->ift_br[0]->abdications, 4958 "# of consumer abdications in the mp_ring for this queue"); 4959 4960 } 4961 4962 if (scctx->isc_nrxqsets > 100) 4963 qfmt = "rxq%03d"; 4964 else if (scctx->isc_nrxqsets > 10) 4965 qfmt = "rxq%02d"; 4966 else 4967 qfmt = "rxq%d"; 4968 for (i = 0, rxq = ctx->ifc_rxqs; i < scctx->isc_nrxqsets; i++, rxq++) { 4969 snprintf(namebuf, NAME_BUFLEN, qfmt, i); 4970 queue_node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, namebuf, 4971 CTLFLAG_RD, NULL, "Queue Name"); 4972 queue_list = SYSCTL_CHILDREN(queue_node); 4973 if (sctx->isc_flags & IFLIB_HAS_RXCQ) { 4974 SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "rxq_cq_pidx", 4975 CTLFLAG_RD, 4976 &rxq->ifr_cq_pidx, 1, "Producer Index"); 4977 SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "rxq_cq_cidx", 4978 CTLFLAG_RD, 4979 &rxq->ifr_cq_cidx, 1, "Consumer Index"); 4980 } 4981 for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) { 4982 snprintf(namebuf, NAME_BUFLEN, "rxq_fl%d", j); 4983 fl_node = SYSCTL_ADD_NODE(ctx_list, queue_list, OID_AUTO, namebuf, 4984 CTLFLAG_RD, NULL, "freelist Name"); 4985 fl_list = SYSCTL_CHILDREN(fl_node); 4986 SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "pidx", 4987 CTLFLAG_RD, 4988 &fl->ifl_pidx, 1, "Producer Index"); 4989 SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "cidx", 4990 CTLFLAG_RD, 4991 &fl->ifl_cidx, 1, "Consumer Index"); 4992 SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "credits", 4993 CTLFLAG_RD, 4994 &fl->ifl_credits, 1, "credits available"); 4995 #if MEMORY_LOGGING 4996 SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_m_enqueued", 4997 CTLFLAG_RD, 4998 &fl->ifl_m_enqueued, "mbufs allocated"); 4999 SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_m_dequeued", 5000 CTLFLAG_RD, 5001 &fl->ifl_m_dequeued, "mbufs freed"); 5002 SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_cl_enqueued", 5003 CTLFLAG_RD, 5004 &fl->ifl_cl_enqueued, "clusters allocated"); 5005 SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_cl_dequeued", 5006 CTLFLAG_RD, 5007 &fl->ifl_cl_dequeued, "clusters freed"); 5008 #endif 5009 5010 } 5011 } 5012 5013 } 5014