1 /*- 2 * Copyright (c) 2014-2018, Matthew Macy <mmacy@mattmacy.io> 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 #include "opt_sched.h" 35 36 #include <sys/param.h> 37 #include <sys/types.h> 38 #include <sys/bus.h> 39 #include <sys/eventhandler.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/sockio.h> 50 #include <sys/sysctl.h> 51 #include <sys/syslog.h> 52 #include <sys/taskqueue.h> 53 #include <sys/limits.h> 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 #include <net/pfil.h> 63 #include <net/vnet.h> 64 65 #include <netinet/in.h> 66 #include <netinet/in_pcb.h> 67 #include <netinet/tcp_lro.h> 68 #include <netinet/in_systm.h> 69 #include <netinet/if_ether.h> 70 #include <netinet/ip.h> 71 #include <netinet/ip6.h> 72 #include <netinet/tcp.h> 73 #include <netinet/ip_var.h> 74 #include <netinet/netdump/netdump.h> 75 #include <netinet6/ip6_var.h> 76 77 #include <machine/bus.h> 78 #include <machine/in_cksum.h> 79 80 #include <vm/vm.h> 81 #include <vm/pmap.h> 82 83 #include <dev/led/led.h> 84 #include <dev/pci/pcireg.h> 85 #include <dev/pci/pcivar.h> 86 #include <dev/pci/pci_private.h> 87 88 #include <net/iflib.h> 89 #include <net/iflib_private.h> 90 91 #include "ifdi_if.h" 92 93 #ifdef PCI_IOV 94 #include <dev/pci/pci_iov.h> 95 #endif 96 97 #include <sys/bitstring.h> 98 /* 99 * enable accounting of every mbuf as it comes in to and goes out of 100 * iflib's software descriptor references 101 */ 102 #define MEMORY_LOGGING 0 103 /* 104 * Enable mbuf vectors for compressing long mbuf chains 105 */ 106 107 /* 108 * NB: 109 * - Prefetching in tx cleaning should perhaps be a tunable. The distance ahead 110 * we prefetch needs to be determined by the time spent in m_free vis a vis 111 * the cost of a prefetch. This will of course vary based on the workload: 112 * - NFLX's m_free path is dominated by vm-based M_EXT manipulation which 113 * is quite expensive, thus suggesting very little prefetch. 114 * - small packet forwarding which is just returning a single mbuf to 115 * UMA will typically be very fast vis a vis the cost of a memory 116 * access. 117 */ 118 119 120 /* 121 * File organization: 122 * - private structures 123 * - iflib private utility functions 124 * - ifnet functions 125 * - vlan registry and other exported functions 126 * - iflib public core functions 127 * 128 * 129 */ 130 MALLOC_DEFINE(M_IFLIB, "iflib", "ifnet library"); 131 132 struct iflib_txq; 133 typedef struct iflib_txq *iflib_txq_t; 134 struct iflib_rxq; 135 typedef struct iflib_rxq *iflib_rxq_t; 136 struct iflib_fl; 137 typedef struct iflib_fl *iflib_fl_t; 138 139 struct iflib_ctx; 140 141 static void iru_init(if_rxd_update_t iru, iflib_rxq_t rxq, uint8_t flid); 142 static void iflib_timer(void *arg); 143 144 typedef struct iflib_filter_info { 145 driver_filter_t *ifi_filter; 146 void *ifi_filter_arg; 147 struct grouptask *ifi_task; 148 void *ifi_ctx; 149 } *iflib_filter_info_t; 150 151 struct iflib_ctx { 152 KOBJ_FIELDS; 153 /* 154 * Pointer to hardware driver's softc 155 */ 156 void *ifc_softc; 157 device_t ifc_dev; 158 if_t ifc_ifp; 159 160 cpuset_t ifc_cpus; 161 if_shared_ctx_t ifc_sctx; 162 struct if_softc_ctx ifc_softc_ctx; 163 164 struct sx ifc_ctx_sx; 165 struct mtx ifc_state_mtx; 166 167 iflib_txq_t ifc_txqs; 168 iflib_rxq_t ifc_rxqs; 169 uint32_t ifc_if_flags; 170 uint32_t ifc_flags; 171 uint32_t ifc_max_fl_buf_size; 172 uint32_t ifc_rx_mbuf_sz; 173 174 int ifc_link_state; 175 int ifc_watchdog_events; 176 struct cdev *ifc_led_dev; 177 struct resource *ifc_msix_mem; 178 179 struct if_irq ifc_legacy_irq; 180 struct grouptask ifc_admin_task; 181 struct grouptask ifc_vflr_task; 182 struct iflib_filter_info ifc_filter_info; 183 struct ifmedia ifc_media; 184 struct ifmedia *ifc_mediap; 185 186 struct sysctl_oid *ifc_sysctl_node; 187 uint16_t ifc_sysctl_ntxqs; 188 uint16_t ifc_sysctl_nrxqs; 189 uint16_t ifc_sysctl_qs_eq_override; 190 uint16_t ifc_sysctl_rx_budget; 191 uint16_t ifc_sysctl_tx_abdicate; 192 uint16_t ifc_sysctl_core_offset; 193 #define CORE_OFFSET_UNSPECIFIED 0xffff 194 uint8_t ifc_sysctl_separate_txrx; 195 196 qidx_t ifc_sysctl_ntxds[8]; 197 qidx_t ifc_sysctl_nrxds[8]; 198 struct if_txrx ifc_txrx; 199 #define isc_txd_encap ifc_txrx.ift_txd_encap 200 #define isc_txd_flush ifc_txrx.ift_txd_flush 201 #define isc_txd_credits_update ifc_txrx.ift_txd_credits_update 202 #define isc_rxd_available ifc_txrx.ift_rxd_available 203 #define isc_rxd_pkt_get ifc_txrx.ift_rxd_pkt_get 204 #define isc_rxd_refill ifc_txrx.ift_rxd_refill 205 #define isc_rxd_flush ifc_txrx.ift_rxd_flush 206 #define isc_rxd_refill ifc_txrx.ift_rxd_refill 207 #define isc_rxd_refill ifc_txrx.ift_rxd_refill 208 #define isc_legacy_intr ifc_txrx.ift_legacy_intr 209 eventhandler_tag ifc_vlan_attach_event; 210 eventhandler_tag ifc_vlan_detach_event; 211 struct ether_addr ifc_mac; 212 }; 213 214 void * 215 iflib_get_softc(if_ctx_t ctx) 216 { 217 218 return (ctx->ifc_softc); 219 } 220 221 device_t 222 iflib_get_dev(if_ctx_t ctx) 223 { 224 225 return (ctx->ifc_dev); 226 } 227 228 if_t 229 iflib_get_ifp(if_ctx_t ctx) 230 { 231 232 return (ctx->ifc_ifp); 233 } 234 235 struct ifmedia * 236 iflib_get_media(if_ctx_t ctx) 237 { 238 239 return (ctx->ifc_mediap); 240 } 241 242 uint32_t 243 iflib_get_flags(if_ctx_t ctx) 244 { 245 return (ctx->ifc_flags); 246 } 247 248 void 249 iflib_set_mac(if_ctx_t ctx, uint8_t mac[ETHER_ADDR_LEN]) 250 { 251 252 bcopy(mac, ctx->ifc_mac.octet, ETHER_ADDR_LEN); 253 } 254 255 if_softc_ctx_t 256 iflib_get_softc_ctx(if_ctx_t ctx) 257 { 258 259 return (&ctx->ifc_softc_ctx); 260 } 261 262 if_shared_ctx_t 263 iflib_get_sctx(if_ctx_t ctx) 264 { 265 266 return (ctx->ifc_sctx); 267 } 268 269 #define IP_ALIGNED(m) ((((uintptr_t)(m)->m_data) & 0x3) == 0x2) 270 #define CACHE_PTR_INCREMENT (CACHE_LINE_SIZE/sizeof(void*)) 271 #define CACHE_PTR_NEXT(ptr) ((void *)(((uintptr_t)(ptr)+CACHE_LINE_SIZE-1) & (CACHE_LINE_SIZE-1))) 272 273 #define LINK_ACTIVE(ctx) ((ctx)->ifc_link_state == LINK_STATE_UP) 274 #define CTX_IS_VF(ctx) ((ctx)->ifc_sctx->isc_flags & IFLIB_IS_VF) 275 276 typedef struct iflib_sw_rx_desc_array { 277 bus_dmamap_t *ifsd_map; /* bus_dma maps for packet */ 278 struct mbuf **ifsd_m; /* pkthdr mbufs */ 279 caddr_t *ifsd_cl; /* direct cluster pointer for rx */ 280 bus_addr_t *ifsd_ba; /* bus addr of cluster for rx */ 281 } iflib_rxsd_array_t; 282 283 typedef struct iflib_sw_tx_desc_array { 284 bus_dmamap_t *ifsd_map; /* bus_dma maps for packet */ 285 bus_dmamap_t *ifsd_tso_map; /* bus_dma maps for TSO packet */ 286 struct mbuf **ifsd_m; /* pkthdr mbufs */ 287 } if_txsd_vec_t; 288 289 /* magic number that should be high enough for any hardware */ 290 #define IFLIB_MAX_TX_SEGS 128 291 #define IFLIB_RX_COPY_THRESH 128 292 #define IFLIB_MAX_RX_REFRESH 32 293 /* The minimum descriptors per second before we start coalescing */ 294 #define IFLIB_MIN_DESC_SEC 16384 295 #define IFLIB_DEFAULT_TX_UPDATE_FREQ 16 296 #define IFLIB_QUEUE_IDLE 0 297 #define IFLIB_QUEUE_HUNG 1 298 #define IFLIB_QUEUE_WORKING 2 299 /* maximum number of txqs that can share an rx interrupt */ 300 #define IFLIB_MAX_TX_SHARED_INTR 4 301 302 /* this should really scale with ring size - this is a fairly arbitrary value */ 303 #define TX_BATCH_SIZE 32 304 305 #define IFLIB_RESTART_BUDGET 8 306 307 #define CSUM_OFFLOAD (CSUM_IP_TSO|CSUM_IP6_TSO|CSUM_IP| \ 308 CSUM_IP_UDP|CSUM_IP_TCP|CSUM_IP_SCTP| \ 309 CSUM_IP6_UDP|CSUM_IP6_TCP|CSUM_IP6_SCTP) 310 311 struct iflib_txq { 312 qidx_t ift_in_use; 313 qidx_t ift_cidx; 314 qidx_t ift_cidx_processed; 315 qidx_t ift_pidx; 316 uint8_t ift_gen; 317 uint8_t ift_br_offset; 318 uint16_t ift_npending; 319 uint16_t ift_db_pending; 320 uint16_t ift_rs_pending; 321 /* implicit pad */ 322 uint8_t ift_txd_size[8]; 323 uint64_t ift_processed; 324 uint64_t ift_cleaned; 325 uint64_t ift_cleaned_prev; 326 #if MEMORY_LOGGING 327 uint64_t ift_enqueued; 328 uint64_t ift_dequeued; 329 #endif 330 uint64_t ift_no_tx_dma_setup; 331 uint64_t ift_no_desc_avail; 332 uint64_t ift_mbuf_defrag_failed; 333 uint64_t ift_mbuf_defrag; 334 uint64_t ift_map_failed; 335 uint64_t ift_txd_encap_efbig; 336 uint64_t ift_pullups; 337 uint64_t ift_last_timer_tick; 338 339 struct mtx ift_mtx; 340 struct mtx ift_db_mtx; 341 342 /* constant values */ 343 if_ctx_t ift_ctx; 344 struct ifmp_ring *ift_br; 345 struct grouptask ift_task; 346 qidx_t ift_size; 347 uint16_t ift_id; 348 struct callout ift_timer; 349 350 if_txsd_vec_t ift_sds; 351 uint8_t ift_qstatus; 352 uint8_t ift_closed; 353 uint8_t ift_update_freq; 354 struct iflib_filter_info ift_filter_info; 355 bus_dma_tag_t ift_buf_tag; 356 bus_dma_tag_t ift_tso_buf_tag; 357 iflib_dma_info_t ift_ifdi; 358 #define MTX_NAME_LEN 16 359 char ift_mtx_name[MTX_NAME_LEN]; 360 bus_dma_segment_t ift_segs[IFLIB_MAX_TX_SEGS] __aligned(CACHE_LINE_SIZE); 361 #ifdef IFLIB_DIAGNOSTICS 362 uint64_t ift_cpu_exec_count[256]; 363 #endif 364 } __aligned(CACHE_LINE_SIZE); 365 366 struct iflib_fl { 367 qidx_t ifl_cidx; 368 qidx_t ifl_pidx; 369 qidx_t ifl_credits; 370 uint8_t ifl_gen; 371 uint8_t ifl_rxd_size; 372 #if MEMORY_LOGGING 373 uint64_t ifl_m_enqueued; 374 uint64_t ifl_m_dequeued; 375 uint64_t ifl_cl_enqueued; 376 uint64_t ifl_cl_dequeued; 377 #endif 378 /* implicit pad */ 379 bitstr_t *ifl_rx_bitmap; 380 qidx_t ifl_fragidx; 381 /* constant */ 382 qidx_t ifl_size; 383 uint16_t ifl_buf_size; 384 uint16_t ifl_cltype; 385 uma_zone_t ifl_zone; 386 iflib_rxsd_array_t ifl_sds; 387 iflib_rxq_t ifl_rxq; 388 uint8_t ifl_id; 389 bus_dma_tag_t ifl_buf_tag; 390 iflib_dma_info_t ifl_ifdi; 391 uint64_t ifl_bus_addrs[IFLIB_MAX_RX_REFRESH] __aligned(CACHE_LINE_SIZE); 392 caddr_t ifl_vm_addrs[IFLIB_MAX_RX_REFRESH]; 393 qidx_t ifl_rxd_idxs[IFLIB_MAX_RX_REFRESH]; 394 } __aligned(CACHE_LINE_SIZE); 395 396 static inline qidx_t 397 get_inuse(int size, qidx_t cidx, qidx_t pidx, uint8_t gen) 398 { 399 qidx_t used; 400 401 if (pidx > cidx) 402 used = pidx - cidx; 403 else if (pidx < cidx) 404 used = size - cidx + pidx; 405 else if (gen == 0 && pidx == cidx) 406 used = 0; 407 else if (gen == 1 && pidx == cidx) 408 used = size; 409 else 410 panic("bad state"); 411 412 return (used); 413 } 414 415 #define TXQ_AVAIL(txq) (txq->ift_size - get_inuse(txq->ift_size, txq->ift_cidx, txq->ift_pidx, txq->ift_gen)) 416 417 #define IDXDIFF(head, tail, wrap) \ 418 ((head) >= (tail) ? (head) - (tail) : (wrap) - (tail) + (head)) 419 420 struct iflib_rxq { 421 if_ctx_t ifr_ctx; 422 iflib_fl_t ifr_fl; 423 uint64_t ifr_rx_irq; 424 struct pfil_head *pfil; 425 /* 426 * If there is a separate completion queue (IFLIB_HAS_RXCQ), this is 427 * the command queue consumer index. Otherwise it's unused. 428 */ 429 qidx_t ifr_cq_cidx; 430 uint16_t ifr_id; 431 uint8_t ifr_nfl; 432 uint8_t ifr_ntxqirq; 433 uint8_t ifr_txqid[IFLIB_MAX_TX_SHARED_INTR]; 434 uint8_t ifr_fl_offset; 435 struct lro_ctrl ifr_lc; 436 struct grouptask ifr_task; 437 struct iflib_filter_info ifr_filter_info; 438 iflib_dma_info_t ifr_ifdi; 439 440 /* dynamically allocate if any drivers need a value substantially larger than this */ 441 struct if_rxd_frag ifr_frags[IFLIB_MAX_RX_SEGS] __aligned(CACHE_LINE_SIZE); 442 #ifdef IFLIB_DIAGNOSTICS 443 uint64_t ifr_cpu_exec_count[256]; 444 #endif 445 } __aligned(CACHE_LINE_SIZE); 446 447 typedef struct if_rxsd { 448 caddr_t *ifsd_cl; 449 iflib_fl_t ifsd_fl; 450 qidx_t ifsd_cidx; 451 } *if_rxsd_t; 452 453 /* multiple of word size */ 454 #ifdef __LP64__ 455 #define PKT_INFO_SIZE 6 456 #define RXD_INFO_SIZE 5 457 #define PKT_TYPE uint64_t 458 #else 459 #define PKT_INFO_SIZE 11 460 #define RXD_INFO_SIZE 8 461 #define PKT_TYPE uint32_t 462 #endif 463 #define PKT_LOOP_BOUND ((PKT_INFO_SIZE/3)*3) 464 #define RXD_LOOP_BOUND ((RXD_INFO_SIZE/4)*4) 465 466 typedef struct if_pkt_info_pad { 467 PKT_TYPE pkt_val[PKT_INFO_SIZE]; 468 } *if_pkt_info_pad_t; 469 typedef struct if_rxd_info_pad { 470 PKT_TYPE rxd_val[RXD_INFO_SIZE]; 471 } *if_rxd_info_pad_t; 472 473 CTASSERT(sizeof(struct if_pkt_info_pad) == sizeof(struct if_pkt_info)); 474 CTASSERT(sizeof(struct if_rxd_info_pad) == sizeof(struct if_rxd_info)); 475 476 477 static inline void 478 pkt_info_zero(if_pkt_info_t pi) 479 { 480 if_pkt_info_pad_t pi_pad; 481 482 pi_pad = (if_pkt_info_pad_t)pi; 483 pi_pad->pkt_val[0] = 0; pi_pad->pkt_val[1] = 0; pi_pad->pkt_val[2] = 0; 484 pi_pad->pkt_val[3] = 0; pi_pad->pkt_val[4] = 0; pi_pad->pkt_val[5] = 0; 485 #ifndef __LP64__ 486 pi_pad->pkt_val[6] = 0; pi_pad->pkt_val[7] = 0; pi_pad->pkt_val[8] = 0; 487 pi_pad->pkt_val[9] = 0; pi_pad->pkt_val[10] = 0; 488 #endif 489 } 490 491 static device_method_t iflib_pseudo_methods[] = { 492 DEVMETHOD(device_attach, noop_attach), 493 DEVMETHOD(device_detach, iflib_pseudo_detach), 494 DEVMETHOD_END 495 }; 496 497 driver_t iflib_pseudodriver = { 498 "iflib_pseudo", iflib_pseudo_methods, sizeof(struct iflib_ctx), 499 }; 500 501 static inline void 502 rxd_info_zero(if_rxd_info_t ri) 503 { 504 if_rxd_info_pad_t ri_pad; 505 int i; 506 507 ri_pad = (if_rxd_info_pad_t)ri; 508 for (i = 0; i < RXD_LOOP_BOUND; i += 4) { 509 ri_pad->rxd_val[i] = 0; 510 ri_pad->rxd_val[i+1] = 0; 511 ri_pad->rxd_val[i+2] = 0; 512 ri_pad->rxd_val[i+3] = 0; 513 } 514 #ifdef __LP64__ 515 ri_pad->rxd_val[RXD_INFO_SIZE-1] = 0; 516 #endif 517 } 518 519 /* 520 * Only allow a single packet to take up most 1/nth of the tx ring 521 */ 522 #define MAX_SINGLE_PACKET_FRACTION 12 523 #define IF_BAD_DMA (bus_addr_t)-1 524 525 #define CTX_ACTIVE(ctx) ((if_getdrvflags((ctx)->ifc_ifp) & IFF_DRV_RUNNING)) 526 527 #define CTX_LOCK_INIT(_sc) sx_init(&(_sc)->ifc_ctx_sx, "iflib ctx lock") 528 #define CTX_LOCK(ctx) sx_xlock(&(ctx)->ifc_ctx_sx) 529 #define CTX_UNLOCK(ctx) sx_xunlock(&(ctx)->ifc_ctx_sx) 530 #define CTX_LOCK_DESTROY(ctx) sx_destroy(&(ctx)->ifc_ctx_sx) 531 532 #define STATE_LOCK_INIT(_sc, _name) mtx_init(&(_sc)->ifc_state_mtx, _name, "iflib state lock", MTX_DEF) 533 #define STATE_LOCK(ctx) mtx_lock(&(ctx)->ifc_state_mtx) 534 #define STATE_UNLOCK(ctx) mtx_unlock(&(ctx)->ifc_state_mtx) 535 #define STATE_LOCK_DESTROY(ctx) mtx_destroy(&(ctx)->ifc_state_mtx) 536 537 #define CALLOUT_LOCK(txq) mtx_lock(&txq->ift_mtx) 538 #define CALLOUT_UNLOCK(txq) mtx_unlock(&txq->ift_mtx) 539 540 void 541 iflib_set_detach(if_ctx_t ctx) 542 { 543 STATE_LOCK(ctx); 544 ctx->ifc_flags |= IFC_IN_DETACH; 545 STATE_UNLOCK(ctx); 546 } 547 548 /* Our boot-time initialization hook */ 549 static int iflib_module_event_handler(module_t, int, void *); 550 551 static moduledata_t iflib_moduledata = { 552 "iflib", 553 iflib_module_event_handler, 554 NULL 555 }; 556 557 DECLARE_MODULE(iflib, iflib_moduledata, SI_SUB_INIT_IF, SI_ORDER_ANY); 558 MODULE_VERSION(iflib, 1); 559 560 MODULE_DEPEND(iflib, pci, 1, 1, 1); 561 MODULE_DEPEND(iflib, ether, 1, 1, 1); 562 563 TASKQGROUP_DEFINE(if_io_tqg, mp_ncpus, 1); 564 TASKQGROUP_DEFINE(if_config_tqg, 1, 1); 565 566 #ifndef IFLIB_DEBUG_COUNTERS 567 #ifdef INVARIANTS 568 #define IFLIB_DEBUG_COUNTERS 1 569 #else 570 #define IFLIB_DEBUG_COUNTERS 0 571 #endif /* !INVARIANTS */ 572 #endif 573 574 static SYSCTL_NODE(_net, OID_AUTO, iflib, CTLFLAG_RD, 0, 575 "iflib driver parameters"); 576 577 /* 578 * XXX need to ensure that this can't accidentally cause the head to be moved backwards 579 */ 580 static int iflib_min_tx_latency = 0; 581 SYSCTL_INT(_net_iflib, OID_AUTO, min_tx_latency, CTLFLAG_RW, 582 &iflib_min_tx_latency, 0, "minimize transmit latency at the possible expense of throughput"); 583 static int iflib_no_tx_batch = 0; 584 SYSCTL_INT(_net_iflib, OID_AUTO, no_tx_batch, CTLFLAG_RW, 585 &iflib_no_tx_batch, 0, "minimize transmit latency at the possible expense of throughput"); 586 587 588 #if IFLIB_DEBUG_COUNTERS 589 590 static int iflib_tx_seen; 591 static int iflib_tx_sent; 592 static int iflib_tx_encap; 593 static int iflib_rx_allocs; 594 static int iflib_fl_refills; 595 static int iflib_fl_refills_large; 596 static int iflib_tx_frees; 597 598 SYSCTL_INT(_net_iflib, OID_AUTO, tx_seen, CTLFLAG_RD, 599 &iflib_tx_seen, 0, "# TX mbufs seen"); 600 SYSCTL_INT(_net_iflib, OID_AUTO, tx_sent, CTLFLAG_RD, 601 &iflib_tx_sent, 0, "# TX mbufs sent"); 602 SYSCTL_INT(_net_iflib, OID_AUTO, tx_encap, CTLFLAG_RD, 603 &iflib_tx_encap, 0, "# TX mbufs encapped"); 604 SYSCTL_INT(_net_iflib, OID_AUTO, tx_frees, CTLFLAG_RD, 605 &iflib_tx_frees, 0, "# TX frees"); 606 SYSCTL_INT(_net_iflib, OID_AUTO, rx_allocs, CTLFLAG_RD, 607 &iflib_rx_allocs, 0, "# RX allocations"); 608 SYSCTL_INT(_net_iflib, OID_AUTO, fl_refills, CTLFLAG_RD, 609 &iflib_fl_refills, 0, "# refills"); 610 SYSCTL_INT(_net_iflib, OID_AUTO, fl_refills_large, CTLFLAG_RD, 611 &iflib_fl_refills_large, 0, "# large refills"); 612 613 614 static int iflib_txq_drain_flushing; 615 static int iflib_txq_drain_oactive; 616 static int iflib_txq_drain_notready; 617 618 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_flushing, CTLFLAG_RD, 619 &iflib_txq_drain_flushing, 0, "# drain flushes"); 620 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_oactive, CTLFLAG_RD, 621 &iflib_txq_drain_oactive, 0, "# drain oactives"); 622 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_notready, CTLFLAG_RD, 623 &iflib_txq_drain_notready, 0, "# drain notready"); 624 625 626 static int iflib_encap_load_mbuf_fail; 627 static int iflib_encap_pad_mbuf_fail; 628 static int iflib_encap_txq_avail_fail; 629 static int iflib_encap_txd_encap_fail; 630 631 SYSCTL_INT(_net_iflib, OID_AUTO, encap_load_mbuf_fail, CTLFLAG_RD, 632 &iflib_encap_load_mbuf_fail, 0, "# busdma load failures"); 633 SYSCTL_INT(_net_iflib, OID_AUTO, encap_pad_mbuf_fail, CTLFLAG_RD, 634 &iflib_encap_pad_mbuf_fail, 0, "# runt frame pad failures"); 635 SYSCTL_INT(_net_iflib, OID_AUTO, encap_txq_avail_fail, CTLFLAG_RD, 636 &iflib_encap_txq_avail_fail, 0, "# txq avail failures"); 637 SYSCTL_INT(_net_iflib, OID_AUTO, encap_txd_encap_fail, CTLFLAG_RD, 638 &iflib_encap_txd_encap_fail, 0, "# driver encap failures"); 639 640 static int iflib_task_fn_rxs; 641 static int iflib_rx_intr_enables; 642 static int iflib_fast_intrs; 643 static int iflib_rx_unavail; 644 static int iflib_rx_ctx_inactive; 645 static int iflib_rx_if_input; 646 static int iflib_rxd_flush; 647 648 static int iflib_verbose_debug; 649 650 SYSCTL_INT(_net_iflib, OID_AUTO, task_fn_rx, CTLFLAG_RD, 651 &iflib_task_fn_rxs, 0, "# task_fn_rx calls"); 652 SYSCTL_INT(_net_iflib, OID_AUTO, rx_intr_enables, CTLFLAG_RD, 653 &iflib_rx_intr_enables, 0, "# RX intr enables"); 654 SYSCTL_INT(_net_iflib, OID_AUTO, fast_intrs, CTLFLAG_RD, 655 &iflib_fast_intrs, 0, "# fast_intr calls"); 656 SYSCTL_INT(_net_iflib, OID_AUTO, rx_unavail, CTLFLAG_RD, 657 &iflib_rx_unavail, 0, "# times rxeof called with no available data"); 658 SYSCTL_INT(_net_iflib, OID_AUTO, rx_ctx_inactive, CTLFLAG_RD, 659 &iflib_rx_ctx_inactive, 0, "# times rxeof called with inactive context"); 660 SYSCTL_INT(_net_iflib, OID_AUTO, rx_if_input, CTLFLAG_RD, 661 &iflib_rx_if_input, 0, "# times rxeof called if_input"); 662 SYSCTL_INT(_net_iflib, OID_AUTO, rxd_flush, CTLFLAG_RD, 663 &iflib_rxd_flush, 0, "# times rxd_flush called"); 664 SYSCTL_INT(_net_iflib, OID_AUTO, verbose_debug, CTLFLAG_RW, 665 &iflib_verbose_debug, 0, "enable verbose debugging"); 666 667 #define DBG_COUNTER_INC(name) atomic_add_int(&(iflib_ ## name), 1) 668 static void 669 iflib_debug_reset(void) 670 { 671 iflib_tx_seen = iflib_tx_sent = iflib_tx_encap = iflib_rx_allocs = 672 iflib_fl_refills = iflib_fl_refills_large = iflib_tx_frees = 673 iflib_txq_drain_flushing = iflib_txq_drain_oactive = 674 iflib_txq_drain_notready = 675 iflib_encap_load_mbuf_fail = iflib_encap_pad_mbuf_fail = 676 iflib_encap_txq_avail_fail = iflib_encap_txd_encap_fail = 677 iflib_task_fn_rxs = iflib_rx_intr_enables = iflib_fast_intrs = 678 iflib_rx_unavail = 679 iflib_rx_ctx_inactive = iflib_rx_if_input = 680 iflib_rxd_flush = 0; 681 } 682 683 #else 684 #define DBG_COUNTER_INC(name) 685 static void iflib_debug_reset(void) {} 686 #endif 687 688 #define IFLIB_DEBUG 0 689 690 static void iflib_tx_structures_free(if_ctx_t ctx); 691 static void iflib_rx_structures_free(if_ctx_t ctx); 692 static int iflib_queues_alloc(if_ctx_t ctx); 693 static int iflib_tx_credits_update(if_ctx_t ctx, iflib_txq_t txq); 694 static int iflib_rxd_avail(if_ctx_t ctx, iflib_rxq_t rxq, qidx_t cidx, qidx_t budget); 695 static int iflib_qset_structures_setup(if_ctx_t ctx); 696 static int iflib_msix_init(if_ctx_t ctx); 697 static int iflib_legacy_setup(if_ctx_t ctx, driver_filter_t filter, void *filterarg, int *rid, const char *str); 698 static void iflib_txq_check_drain(iflib_txq_t txq, int budget); 699 static uint32_t iflib_txq_can_drain(struct ifmp_ring *); 700 #ifdef ALTQ 701 static void iflib_altq_if_start(if_t ifp); 702 static int iflib_altq_if_transmit(if_t ifp, struct mbuf *m); 703 #endif 704 static int iflib_register(if_ctx_t); 705 static void iflib_init_locked(if_ctx_t ctx); 706 static void iflib_add_device_sysctl_pre(if_ctx_t ctx); 707 static void iflib_add_device_sysctl_post(if_ctx_t ctx); 708 static void iflib_ifmp_purge(iflib_txq_t txq); 709 static void _iflib_pre_assert(if_softc_ctx_t scctx); 710 static void iflib_if_init_locked(if_ctx_t ctx); 711 static void iflib_free_intr_mem(if_ctx_t ctx); 712 #ifndef __NO_STRICT_ALIGNMENT 713 static struct mbuf * iflib_fixup_rx(struct mbuf *m); 714 #endif 715 716 static SLIST_HEAD(cpu_offset_list, cpu_offset) cpu_offsets = 717 SLIST_HEAD_INITIALIZER(cpu_offsets); 718 struct cpu_offset { 719 SLIST_ENTRY(cpu_offset) entries; 720 cpuset_t set; 721 unsigned int refcount; 722 uint16_t offset; 723 }; 724 static struct mtx cpu_offset_mtx; 725 MTX_SYSINIT(iflib_cpu_offset, &cpu_offset_mtx, "iflib_cpu_offset lock", 726 MTX_DEF); 727 728 NETDUMP_DEFINE(iflib); 729 730 #ifdef DEV_NETMAP 731 #include <sys/selinfo.h> 732 #include <net/netmap.h> 733 #include <dev/netmap/netmap_kern.h> 734 735 MODULE_DEPEND(iflib, netmap, 1, 1, 1); 736 737 static int netmap_fl_refill(iflib_rxq_t rxq, struct netmap_kring *kring, uint32_t nm_i, bool init); 738 739 /* 740 * device-specific sysctl variables: 741 * 742 * iflib_crcstrip: 0: keep CRC in rx frames (default), 1: strip it. 743 * During regular operations the CRC is stripped, but on some 744 * hardware reception of frames not multiple of 64 is slower, 745 * so using crcstrip=0 helps in benchmarks. 746 * 747 * iflib_rx_miss, iflib_rx_miss_bufs: 748 * count packets that might be missed due to lost interrupts. 749 */ 750 SYSCTL_DECL(_dev_netmap); 751 /* 752 * The xl driver by default strips CRCs and we do not override it. 753 */ 754 755 int iflib_crcstrip = 1; 756 SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_crcstrip, 757 CTLFLAG_RW, &iflib_crcstrip, 1, "strip CRC on RX frames"); 758 759 int iflib_rx_miss, iflib_rx_miss_bufs; 760 SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_rx_miss, 761 CTLFLAG_RW, &iflib_rx_miss, 0, "potentially missed RX intr"); 762 SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_rx_miss_bufs, 763 CTLFLAG_RW, &iflib_rx_miss_bufs, 0, "potentially missed RX intr bufs"); 764 765 /* 766 * Register/unregister. We are already under netmap lock. 767 * Only called on the first register or the last unregister. 768 */ 769 static int 770 iflib_netmap_register(struct netmap_adapter *na, int onoff) 771 { 772 if_t ifp = na->ifp; 773 if_ctx_t ctx = ifp->if_softc; 774 int status; 775 776 CTX_LOCK(ctx); 777 IFDI_INTR_DISABLE(ctx); 778 779 /* Tell the stack that the interface is no longer active */ 780 ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); 781 782 if (!CTX_IS_VF(ctx)) 783 IFDI_CRCSTRIP_SET(ctx, onoff, iflib_crcstrip); 784 785 /* enable or disable flags and callbacks in na and ifp */ 786 if (onoff) { 787 nm_set_native_flags(na); 788 } else { 789 nm_clear_native_flags(na); 790 } 791 iflib_stop(ctx); 792 iflib_init_locked(ctx); 793 IFDI_CRCSTRIP_SET(ctx, onoff, iflib_crcstrip); // XXX why twice ? 794 status = ifp->if_drv_flags & IFF_DRV_RUNNING ? 0 : 1; 795 if (status) 796 nm_clear_native_flags(na); 797 CTX_UNLOCK(ctx); 798 return (status); 799 } 800 801 static int 802 netmap_fl_refill(iflib_rxq_t rxq, struct netmap_kring *kring, uint32_t nm_i, bool init) 803 { 804 struct netmap_adapter *na = kring->na; 805 u_int const lim = kring->nkr_num_slots - 1; 806 u_int head = kring->rhead; 807 struct netmap_ring *ring = kring->ring; 808 bus_dmamap_t *map; 809 struct if_rxd_update iru; 810 if_ctx_t ctx = rxq->ifr_ctx; 811 iflib_fl_t fl = &rxq->ifr_fl[0]; 812 uint32_t refill_pidx, nic_i; 813 #if IFLIB_DEBUG_COUNTERS 814 int rf_count = 0; 815 #endif 816 817 if (nm_i == head && __predict_true(!init)) 818 return 0; 819 iru_init(&iru, rxq, 0 /* flid */); 820 map = fl->ifl_sds.ifsd_map; 821 refill_pidx = netmap_idx_k2n(kring, nm_i); 822 /* 823 * IMPORTANT: we must leave one free slot in the ring, 824 * so move head back by one unit 825 */ 826 head = nm_prev(head, lim); 827 nic_i = UINT_MAX; 828 DBG_COUNTER_INC(fl_refills); 829 while (nm_i != head) { 830 #if IFLIB_DEBUG_COUNTERS 831 if (++rf_count == 9) 832 DBG_COUNTER_INC(fl_refills_large); 833 #endif 834 for (int tmp_pidx = 0; tmp_pidx < IFLIB_MAX_RX_REFRESH && nm_i != head; tmp_pidx++) { 835 struct netmap_slot *slot = &ring->slot[nm_i]; 836 void *addr = PNMB(na, slot, &fl->ifl_bus_addrs[tmp_pidx]); 837 uint32_t nic_i_dma = refill_pidx; 838 nic_i = netmap_idx_k2n(kring, nm_i); 839 840 MPASS(tmp_pidx < IFLIB_MAX_RX_REFRESH); 841 842 if (addr == NETMAP_BUF_BASE(na)) /* bad buf */ 843 return netmap_ring_reinit(kring); 844 845 fl->ifl_vm_addrs[tmp_pidx] = addr; 846 if (__predict_false(init)) { 847 netmap_load_map(na, fl->ifl_buf_tag, 848 map[nic_i], addr); 849 } else if (slot->flags & NS_BUF_CHANGED) { 850 /* buffer has changed, reload map */ 851 netmap_reload_map(na, fl->ifl_buf_tag, 852 map[nic_i], addr); 853 } 854 slot->flags &= ~NS_BUF_CHANGED; 855 856 nm_i = nm_next(nm_i, lim); 857 fl->ifl_rxd_idxs[tmp_pidx] = nic_i = nm_next(nic_i, lim); 858 if (nm_i != head && tmp_pidx < IFLIB_MAX_RX_REFRESH-1) 859 continue; 860 861 iru.iru_pidx = refill_pidx; 862 iru.iru_count = tmp_pidx+1; 863 ctx->isc_rxd_refill(ctx->ifc_softc, &iru); 864 refill_pidx = nic_i; 865 for (int n = 0; n < iru.iru_count; n++) { 866 bus_dmamap_sync(fl->ifl_buf_tag, map[nic_i_dma], 867 BUS_DMASYNC_PREREAD); 868 /* XXX - change this to not use the netmap func*/ 869 nic_i_dma = nm_next(nic_i_dma, lim); 870 } 871 } 872 } 873 kring->nr_hwcur = head; 874 875 bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map, 876 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); 877 if (__predict_true(nic_i != UINT_MAX)) { 878 ctx->isc_rxd_flush(ctx->ifc_softc, rxq->ifr_id, fl->ifl_id, nic_i); 879 DBG_COUNTER_INC(rxd_flush); 880 } 881 return (0); 882 } 883 884 /* 885 * Reconcile kernel and user view of the transmit ring. 886 * 887 * All information is in the kring. 888 * Userspace wants to send packets up to the one before kring->rhead, 889 * kernel knows kring->nr_hwcur is the first unsent packet. 890 * 891 * Here we push packets out (as many as possible), and possibly 892 * reclaim buffers from previously completed transmission. 893 * 894 * The caller (netmap) guarantees that there is only one instance 895 * running at any time. Any interference with other driver 896 * methods should be handled by the individual drivers. 897 */ 898 static int 899 iflib_netmap_txsync(struct netmap_kring *kring, int flags) 900 { 901 struct netmap_adapter *na = kring->na; 902 if_t ifp = na->ifp; 903 struct netmap_ring *ring = kring->ring; 904 u_int nm_i; /* index into the netmap kring */ 905 u_int nic_i; /* index into the NIC ring */ 906 u_int n; 907 u_int const lim = kring->nkr_num_slots - 1; 908 u_int const head = kring->rhead; 909 struct if_pkt_info pi; 910 911 /* 912 * interrupts on every tx packet are expensive so request 913 * them every half ring, or where NS_REPORT is set 914 */ 915 u_int report_frequency = kring->nkr_num_slots >> 1; 916 /* device-specific */ 917 if_ctx_t ctx = ifp->if_softc; 918 iflib_txq_t txq = &ctx->ifc_txqs[kring->ring_id]; 919 920 bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map, 921 BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); 922 923 /* 924 * First part: process new packets to send. 925 * nm_i is the current index in the netmap kring, 926 * nic_i is the corresponding index in the NIC ring. 927 * 928 * If we have packets to send (nm_i != head) 929 * iterate over the netmap ring, fetch length and update 930 * the corresponding slot in the NIC ring. Some drivers also 931 * need to update the buffer's physical address in the NIC slot 932 * even NS_BUF_CHANGED is not set (PNMB computes the addresses). 933 * 934 * The netmap_reload_map() calls is especially expensive, 935 * even when (as in this case) the tag is 0, so do only 936 * when the buffer has actually changed. 937 * 938 * If possible do not set the report/intr bit on all slots, 939 * but only a few times per ring or when NS_REPORT is set. 940 * 941 * Finally, on 10G and faster drivers, it might be useful 942 * to prefetch the next slot and txr entry. 943 */ 944 945 nm_i = kring->nr_hwcur; 946 if (nm_i != head) { /* we have new packets to send */ 947 pkt_info_zero(&pi); 948 pi.ipi_segs = txq->ift_segs; 949 pi.ipi_qsidx = kring->ring_id; 950 nic_i = netmap_idx_k2n(kring, nm_i); 951 952 __builtin_prefetch(&ring->slot[nm_i]); 953 __builtin_prefetch(&txq->ift_sds.ifsd_m[nic_i]); 954 __builtin_prefetch(&txq->ift_sds.ifsd_map[nic_i]); 955 956 for (n = 0; nm_i != head; n++) { 957 struct netmap_slot *slot = &ring->slot[nm_i]; 958 u_int len = slot->len; 959 uint64_t paddr; 960 void *addr = PNMB(na, slot, &paddr); 961 int flags = (slot->flags & NS_REPORT || 962 nic_i == 0 || nic_i == report_frequency) ? 963 IPI_TX_INTR : 0; 964 965 /* device-specific */ 966 pi.ipi_len = len; 967 pi.ipi_segs[0].ds_addr = paddr; 968 pi.ipi_segs[0].ds_len = len; 969 pi.ipi_nsegs = 1; 970 pi.ipi_ndescs = 0; 971 pi.ipi_pidx = nic_i; 972 pi.ipi_flags = flags; 973 974 /* Fill the slot in the NIC ring. */ 975 ctx->isc_txd_encap(ctx->ifc_softc, &pi); 976 DBG_COUNTER_INC(tx_encap); 977 978 /* prefetch for next round */ 979 __builtin_prefetch(&ring->slot[nm_i + 1]); 980 __builtin_prefetch(&txq->ift_sds.ifsd_m[nic_i + 1]); 981 __builtin_prefetch(&txq->ift_sds.ifsd_map[nic_i + 1]); 982 983 NM_CHECK_ADDR_LEN(na, addr, len); 984 985 if (slot->flags & NS_BUF_CHANGED) { 986 /* buffer has changed, reload map */ 987 netmap_reload_map(na, txq->ift_buf_tag, 988 txq->ift_sds.ifsd_map[nic_i], addr); 989 } 990 /* make sure changes to the buffer are synced */ 991 bus_dmamap_sync(txq->ift_buf_tag, 992 txq->ift_sds.ifsd_map[nic_i], 993 BUS_DMASYNC_PREWRITE); 994 995 slot->flags &= ~(NS_REPORT | NS_BUF_CHANGED); 996 nm_i = nm_next(nm_i, lim); 997 nic_i = nm_next(nic_i, lim); 998 } 999 kring->nr_hwcur = nm_i; 1000 1001 /* synchronize the NIC ring */ 1002 bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map, 1003 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); 1004 1005 /* (re)start the tx unit up to slot nic_i (excluded) */ 1006 ctx->isc_txd_flush(ctx->ifc_softc, txq->ift_id, nic_i); 1007 } 1008 1009 /* 1010 * Second part: reclaim buffers for completed transmissions. 1011 * 1012 * If there are unclaimed buffers, attempt to reclaim them. 1013 * If none are reclaimed, and TX IRQs are not in use, do an initial 1014 * minimal delay, then trigger the tx handler which will spin in the 1015 * group task queue. 1016 */ 1017 if (kring->nr_hwtail != nm_prev(kring->nr_hwcur, lim)) { 1018 if (iflib_tx_credits_update(ctx, txq)) { 1019 /* some tx completed, increment avail */ 1020 nic_i = txq->ift_cidx_processed; 1021 kring->nr_hwtail = nm_prev(netmap_idx_n2k(kring, nic_i), lim); 1022 } 1023 } 1024 if (!(ctx->ifc_flags & IFC_NETMAP_TX_IRQ)) 1025 if (kring->nr_hwtail != nm_prev(kring->nr_hwcur, lim)) { 1026 callout_reset_on(&txq->ift_timer, hz < 2000 ? 1 : hz / 1000, 1027 iflib_timer, txq, txq->ift_timer.c_cpu); 1028 } 1029 return (0); 1030 } 1031 1032 /* 1033 * Reconcile kernel and user view of the receive ring. 1034 * Same as for the txsync, this routine must be efficient. 1035 * The caller guarantees a single invocations, but races against 1036 * the rest of the driver should be handled here. 1037 * 1038 * On call, kring->rhead is the first packet that userspace wants 1039 * to keep, and kring->rcur is the wakeup point. 1040 * The kernel has previously reported packets up to kring->rtail. 1041 * 1042 * If (flags & NAF_FORCE_READ) also check for incoming packets irrespective 1043 * of whether or not we received an interrupt. 1044 */ 1045 static int 1046 iflib_netmap_rxsync(struct netmap_kring *kring, int flags) 1047 { 1048 struct netmap_adapter *na = kring->na; 1049 struct netmap_ring *ring = kring->ring; 1050 if_t ifp = na->ifp; 1051 iflib_fl_t fl; 1052 uint32_t nm_i; /* index into the netmap ring */ 1053 uint32_t nic_i; /* index into the NIC ring */ 1054 u_int i, n; 1055 u_int const lim = kring->nkr_num_slots - 1; 1056 u_int const head = kring->rhead; 1057 int force_update = (flags & NAF_FORCE_READ) || kring->nr_kflags & NKR_PENDINTR; 1058 struct if_rxd_info ri; 1059 1060 if_ctx_t ctx = ifp->if_softc; 1061 iflib_rxq_t rxq = &ctx->ifc_rxqs[kring->ring_id]; 1062 if (head > lim) 1063 return netmap_ring_reinit(kring); 1064 1065 /* 1066 * XXX netmap_fl_refill() only ever (re)fills free list 0 so far. 1067 */ 1068 1069 for (i = 0, fl = rxq->ifr_fl; i < rxq->ifr_nfl; i++, fl++) { 1070 bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map, 1071 BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); 1072 } 1073 1074 /* 1075 * First part: import newly received packets. 1076 * 1077 * nm_i is the index of the next free slot in the netmap ring, 1078 * nic_i is the index of the next received packet in the NIC ring, 1079 * and they may differ in case if_init() has been called while 1080 * in netmap mode. For the receive ring we have 1081 * 1082 * nic_i = rxr->next_check; 1083 * nm_i = kring->nr_hwtail (previous) 1084 * and 1085 * nm_i == (nic_i + kring->nkr_hwofs) % ring_size 1086 * 1087 * rxr->next_check is set to 0 on a ring reinit 1088 */ 1089 if (netmap_no_pendintr || force_update) { 1090 int crclen = iflib_crcstrip ? 0 : 4; 1091 int error, avail; 1092 1093 for (i = 0; i < rxq->ifr_nfl; i++) { 1094 fl = &rxq->ifr_fl[i]; 1095 nic_i = fl->ifl_cidx; 1096 nm_i = netmap_idx_n2k(kring, nic_i); 1097 avail = ctx->isc_rxd_available(ctx->ifc_softc, 1098 rxq->ifr_id, nic_i, USHRT_MAX); 1099 for (n = 0; avail > 0; n++, avail--) { 1100 rxd_info_zero(&ri); 1101 ri.iri_frags = rxq->ifr_frags; 1102 ri.iri_qsidx = kring->ring_id; 1103 ri.iri_ifp = ctx->ifc_ifp; 1104 ri.iri_cidx = nic_i; 1105 1106 error = ctx->isc_rxd_pkt_get(ctx->ifc_softc, &ri); 1107 ring->slot[nm_i].len = error ? 0 : ri.iri_len - crclen; 1108 ring->slot[nm_i].flags = 0; 1109 bus_dmamap_sync(fl->ifl_buf_tag, 1110 fl->ifl_sds.ifsd_map[nic_i], BUS_DMASYNC_POSTREAD); 1111 nm_i = nm_next(nm_i, lim); 1112 nic_i = nm_next(nic_i, lim); 1113 } 1114 if (n) { /* update the state variables */ 1115 if (netmap_no_pendintr && !force_update) { 1116 /* diagnostics */ 1117 iflib_rx_miss ++; 1118 iflib_rx_miss_bufs += n; 1119 } 1120 fl->ifl_cidx = nic_i; 1121 kring->nr_hwtail = nm_i; 1122 } 1123 kring->nr_kflags &= ~NKR_PENDINTR; 1124 } 1125 } 1126 /* 1127 * Second part: skip past packets that userspace has released. 1128 * (kring->nr_hwcur to head excluded), 1129 * and make the buffers available for reception. 1130 * As usual nm_i is the index in the netmap ring, 1131 * nic_i is the index in the NIC ring, and 1132 * nm_i == (nic_i + kring->nkr_hwofs) % ring_size 1133 */ 1134 /* XXX not sure how this will work with multiple free lists */ 1135 nm_i = kring->nr_hwcur; 1136 1137 return (netmap_fl_refill(rxq, kring, nm_i, false)); 1138 } 1139 1140 static void 1141 iflib_netmap_intr(struct netmap_adapter *na, int onoff) 1142 { 1143 if_ctx_t ctx = na->ifp->if_softc; 1144 1145 CTX_LOCK(ctx); 1146 if (onoff) { 1147 IFDI_INTR_ENABLE(ctx); 1148 } else { 1149 IFDI_INTR_DISABLE(ctx); 1150 } 1151 CTX_UNLOCK(ctx); 1152 } 1153 1154 1155 static int 1156 iflib_netmap_attach(if_ctx_t ctx) 1157 { 1158 struct netmap_adapter na; 1159 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 1160 1161 bzero(&na, sizeof(na)); 1162 1163 na.ifp = ctx->ifc_ifp; 1164 na.na_flags = NAF_BDG_MAYSLEEP; 1165 MPASS(ctx->ifc_softc_ctx.isc_ntxqsets); 1166 MPASS(ctx->ifc_softc_ctx.isc_nrxqsets); 1167 1168 na.num_tx_desc = scctx->isc_ntxd[0]; 1169 na.num_rx_desc = scctx->isc_nrxd[0]; 1170 na.nm_txsync = iflib_netmap_txsync; 1171 na.nm_rxsync = iflib_netmap_rxsync; 1172 na.nm_register = iflib_netmap_register; 1173 na.nm_intr = iflib_netmap_intr; 1174 na.num_tx_rings = ctx->ifc_softc_ctx.isc_ntxqsets; 1175 na.num_rx_rings = ctx->ifc_softc_ctx.isc_nrxqsets; 1176 return (netmap_attach(&na)); 1177 } 1178 1179 static void 1180 iflib_netmap_txq_init(if_ctx_t ctx, iflib_txq_t txq) 1181 { 1182 struct netmap_adapter *na = NA(ctx->ifc_ifp); 1183 struct netmap_slot *slot; 1184 1185 slot = netmap_reset(na, NR_TX, txq->ift_id, 0); 1186 if (slot == NULL) 1187 return; 1188 for (int i = 0; i < ctx->ifc_softc_ctx.isc_ntxd[0]; i++) { 1189 1190 /* 1191 * In netmap mode, set the map for the packet buffer. 1192 * NOTE: Some drivers (not this one) also need to set 1193 * the physical buffer address in the NIC ring. 1194 * netmap_idx_n2k() maps a nic index, i, into the corresponding 1195 * netmap slot index, si 1196 */ 1197 int si = netmap_idx_n2k(na->tx_rings[txq->ift_id], i); 1198 netmap_load_map(na, txq->ift_buf_tag, txq->ift_sds.ifsd_map[i], 1199 NMB(na, slot + si)); 1200 } 1201 } 1202 1203 static void 1204 iflib_netmap_rxq_init(if_ctx_t ctx, iflib_rxq_t rxq) 1205 { 1206 struct netmap_adapter *na = NA(ctx->ifc_ifp); 1207 struct netmap_kring *kring = na->rx_rings[rxq->ifr_id]; 1208 struct netmap_slot *slot; 1209 uint32_t nm_i; 1210 1211 slot = netmap_reset(na, NR_RX, rxq->ifr_id, 0); 1212 if (slot == NULL) 1213 return; 1214 nm_i = netmap_idx_n2k(kring, 0); 1215 netmap_fl_refill(rxq, kring, nm_i, true); 1216 } 1217 1218 static void 1219 iflib_netmap_timer_adjust(if_ctx_t ctx, iflib_txq_t txq, uint32_t *reset_on) 1220 { 1221 struct netmap_kring *kring; 1222 uint16_t txqid; 1223 1224 txqid = txq->ift_id; 1225 kring = NA(ctx->ifc_ifp)->tx_rings[txqid]; 1226 1227 if (kring->nr_hwcur != nm_next(kring->nr_hwtail, kring->nkr_num_slots - 1)) { 1228 bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map, 1229 BUS_DMASYNC_POSTREAD); 1230 if (ctx->isc_txd_credits_update(ctx->ifc_softc, txqid, false)) 1231 netmap_tx_irq(ctx->ifc_ifp, txqid); 1232 if (!(ctx->ifc_flags & IFC_NETMAP_TX_IRQ)) { 1233 if (hz < 2000) 1234 *reset_on = 1; 1235 else 1236 *reset_on = hz / 1000; 1237 } 1238 } 1239 } 1240 1241 #define iflib_netmap_detach(ifp) netmap_detach(ifp) 1242 1243 #else 1244 #define iflib_netmap_txq_init(ctx, txq) 1245 #define iflib_netmap_rxq_init(ctx, rxq) 1246 #define iflib_netmap_detach(ifp) 1247 1248 #define iflib_netmap_attach(ctx) (0) 1249 #define netmap_rx_irq(ifp, qid, budget) (0) 1250 #define netmap_tx_irq(ifp, qid) do {} while (0) 1251 #define iflib_netmap_timer_adjust(ctx, txq, reset_on) 1252 #endif 1253 1254 #if defined(__i386__) || defined(__amd64__) 1255 static __inline void 1256 prefetch(void *x) 1257 { 1258 __asm volatile("prefetcht0 %0" :: "m" (*(unsigned long *)x)); 1259 } 1260 static __inline void 1261 prefetch2cachelines(void *x) 1262 { 1263 __asm volatile("prefetcht0 %0" :: "m" (*(unsigned long *)x)); 1264 #if (CACHE_LINE_SIZE < 128) 1265 __asm volatile("prefetcht0 %0" :: "m" (*(((unsigned long *)x)+CACHE_LINE_SIZE/(sizeof(unsigned long))))); 1266 #endif 1267 } 1268 #else 1269 #define prefetch(x) 1270 #define prefetch2cachelines(x) 1271 #endif 1272 1273 static void 1274 iru_init(if_rxd_update_t iru, iflib_rxq_t rxq, uint8_t flid) 1275 { 1276 iflib_fl_t fl; 1277 1278 fl = &rxq->ifr_fl[flid]; 1279 iru->iru_paddrs = fl->ifl_bus_addrs; 1280 iru->iru_vaddrs = &fl->ifl_vm_addrs[0]; 1281 iru->iru_idxs = fl->ifl_rxd_idxs; 1282 iru->iru_qsidx = rxq->ifr_id; 1283 iru->iru_buf_size = fl->ifl_buf_size; 1284 iru->iru_flidx = fl->ifl_id; 1285 } 1286 1287 static void 1288 _iflib_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int err) 1289 { 1290 if (err) 1291 return; 1292 *(bus_addr_t *) arg = segs[0].ds_addr; 1293 } 1294 1295 int 1296 iflib_dma_alloc_align(if_ctx_t ctx, int size, int align, iflib_dma_info_t dma, int mapflags) 1297 { 1298 int err; 1299 device_t dev = ctx->ifc_dev; 1300 1301 err = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */ 1302 align, 0, /* alignment, bounds */ 1303 BUS_SPACE_MAXADDR, /* lowaddr */ 1304 BUS_SPACE_MAXADDR, /* highaddr */ 1305 NULL, NULL, /* filter, filterarg */ 1306 size, /* maxsize */ 1307 1, /* nsegments */ 1308 size, /* maxsegsize */ 1309 BUS_DMA_ALLOCNOW, /* flags */ 1310 NULL, /* lockfunc */ 1311 NULL, /* lockarg */ 1312 &dma->idi_tag); 1313 if (err) { 1314 device_printf(dev, 1315 "%s: bus_dma_tag_create failed: %d\n", 1316 __func__, err); 1317 goto fail_0; 1318 } 1319 1320 err = bus_dmamem_alloc(dma->idi_tag, (void**) &dma->idi_vaddr, 1321 BUS_DMA_NOWAIT | BUS_DMA_COHERENT | BUS_DMA_ZERO, &dma->idi_map); 1322 if (err) { 1323 device_printf(dev, 1324 "%s: bus_dmamem_alloc(%ju) failed: %d\n", 1325 __func__, (uintmax_t)size, err); 1326 goto fail_1; 1327 } 1328 1329 dma->idi_paddr = IF_BAD_DMA; 1330 err = bus_dmamap_load(dma->idi_tag, dma->idi_map, dma->idi_vaddr, 1331 size, _iflib_dmamap_cb, &dma->idi_paddr, mapflags | BUS_DMA_NOWAIT); 1332 if (err || dma->idi_paddr == IF_BAD_DMA) { 1333 device_printf(dev, 1334 "%s: bus_dmamap_load failed: %d\n", 1335 __func__, err); 1336 goto fail_2; 1337 } 1338 1339 dma->idi_size = size; 1340 return (0); 1341 1342 fail_2: 1343 bus_dmamem_free(dma->idi_tag, dma->idi_vaddr, dma->idi_map); 1344 fail_1: 1345 bus_dma_tag_destroy(dma->idi_tag); 1346 fail_0: 1347 dma->idi_tag = NULL; 1348 1349 return (err); 1350 } 1351 1352 int 1353 iflib_dma_alloc(if_ctx_t ctx, int size, iflib_dma_info_t dma, int mapflags) 1354 { 1355 if_shared_ctx_t sctx = ctx->ifc_sctx; 1356 1357 KASSERT(sctx->isc_q_align != 0, ("alignment value not initialized")); 1358 1359 return (iflib_dma_alloc_align(ctx, size, sctx->isc_q_align, dma, mapflags)); 1360 } 1361 1362 int 1363 iflib_dma_alloc_multi(if_ctx_t ctx, int *sizes, iflib_dma_info_t *dmalist, int mapflags, int count) 1364 { 1365 int i, err; 1366 iflib_dma_info_t *dmaiter; 1367 1368 dmaiter = dmalist; 1369 for (i = 0; i < count; i++, dmaiter++) { 1370 if ((err = iflib_dma_alloc(ctx, sizes[i], *dmaiter, mapflags)) != 0) 1371 break; 1372 } 1373 if (err) 1374 iflib_dma_free_multi(dmalist, i); 1375 return (err); 1376 } 1377 1378 void 1379 iflib_dma_free(iflib_dma_info_t dma) 1380 { 1381 if (dma->idi_tag == NULL) 1382 return; 1383 if (dma->idi_paddr != IF_BAD_DMA) { 1384 bus_dmamap_sync(dma->idi_tag, dma->idi_map, 1385 BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); 1386 bus_dmamap_unload(dma->idi_tag, dma->idi_map); 1387 dma->idi_paddr = IF_BAD_DMA; 1388 } 1389 if (dma->idi_vaddr != NULL) { 1390 bus_dmamem_free(dma->idi_tag, dma->idi_vaddr, dma->idi_map); 1391 dma->idi_vaddr = NULL; 1392 } 1393 bus_dma_tag_destroy(dma->idi_tag); 1394 dma->idi_tag = NULL; 1395 } 1396 1397 void 1398 iflib_dma_free_multi(iflib_dma_info_t *dmalist, int count) 1399 { 1400 int i; 1401 iflib_dma_info_t *dmaiter = dmalist; 1402 1403 for (i = 0; i < count; i++, dmaiter++) 1404 iflib_dma_free(*dmaiter); 1405 } 1406 1407 #ifdef EARLY_AP_STARTUP 1408 static const int iflib_started = 1; 1409 #else 1410 /* 1411 * We used to abuse the smp_started flag to decide if the queues have been 1412 * fully initialized (by late taskqgroup_adjust() calls in a SYSINIT()). 1413 * That gave bad races, since the SYSINIT() runs strictly after smp_started 1414 * is set. Run a SYSINIT() strictly after that to just set a usable 1415 * completion flag. 1416 */ 1417 1418 static int iflib_started; 1419 1420 static void 1421 iflib_record_started(void *arg) 1422 { 1423 iflib_started = 1; 1424 } 1425 1426 SYSINIT(iflib_record_started, SI_SUB_SMP + 1, SI_ORDER_FIRST, 1427 iflib_record_started, NULL); 1428 #endif 1429 1430 static int 1431 iflib_fast_intr(void *arg) 1432 { 1433 iflib_filter_info_t info = arg; 1434 struct grouptask *gtask = info->ifi_task; 1435 int result; 1436 1437 if (!iflib_started) 1438 return (FILTER_STRAY); 1439 1440 DBG_COUNTER_INC(fast_intrs); 1441 if (info->ifi_filter != NULL) { 1442 result = info->ifi_filter(info->ifi_filter_arg); 1443 if ((result & FILTER_SCHEDULE_THREAD) == 0) 1444 return (result); 1445 } 1446 1447 GROUPTASK_ENQUEUE(gtask); 1448 return (FILTER_HANDLED); 1449 } 1450 1451 static int 1452 iflib_fast_intr_rxtx(void *arg) 1453 { 1454 iflib_filter_info_t info = arg; 1455 struct grouptask *gtask = info->ifi_task; 1456 if_ctx_t ctx; 1457 iflib_rxq_t rxq = (iflib_rxq_t)info->ifi_ctx; 1458 iflib_txq_t txq; 1459 void *sc; 1460 int i, cidx, result; 1461 qidx_t txqid; 1462 bool intr_enable, intr_legacy; 1463 1464 if (!iflib_started) 1465 return (FILTER_STRAY); 1466 1467 DBG_COUNTER_INC(fast_intrs); 1468 if (info->ifi_filter != NULL) { 1469 result = info->ifi_filter(info->ifi_filter_arg); 1470 if ((result & FILTER_SCHEDULE_THREAD) == 0) 1471 return (result); 1472 } 1473 1474 ctx = rxq->ifr_ctx; 1475 sc = ctx->ifc_softc; 1476 intr_enable = false; 1477 intr_legacy = !!(ctx->ifc_flags & IFC_LEGACY); 1478 MPASS(rxq->ifr_ntxqirq); 1479 for (i = 0; i < rxq->ifr_ntxqirq; i++) { 1480 txqid = rxq->ifr_txqid[i]; 1481 txq = &ctx->ifc_txqs[txqid]; 1482 bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map, 1483 BUS_DMASYNC_POSTREAD); 1484 if (!ctx->isc_txd_credits_update(sc, txqid, false)) { 1485 if (intr_legacy) 1486 intr_enable = true; 1487 else 1488 IFDI_TX_QUEUE_INTR_ENABLE(ctx, txqid); 1489 continue; 1490 } 1491 GROUPTASK_ENQUEUE(&txq->ift_task); 1492 } 1493 if (ctx->ifc_sctx->isc_flags & IFLIB_HAS_RXCQ) 1494 cidx = rxq->ifr_cq_cidx; 1495 else 1496 cidx = rxq->ifr_fl[0].ifl_cidx; 1497 if (iflib_rxd_avail(ctx, rxq, cidx, 1)) 1498 GROUPTASK_ENQUEUE(gtask); 1499 else { 1500 if (intr_legacy) 1501 intr_enable = true; 1502 else 1503 IFDI_RX_QUEUE_INTR_ENABLE(ctx, rxq->ifr_id); 1504 DBG_COUNTER_INC(rx_intr_enables); 1505 } 1506 if (intr_enable) 1507 IFDI_INTR_ENABLE(ctx); 1508 return (FILTER_HANDLED); 1509 } 1510 1511 1512 static int 1513 iflib_fast_intr_ctx(void *arg) 1514 { 1515 iflib_filter_info_t info = arg; 1516 struct grouptask *gtask = info->ifi_task; 1517 int result; 1518 1519 if (!iflib_started) 1520 return (FILTER_STRAY); 1521 1522 DBG_COUNTER_INC(fast_intrs); 1523 if (info->ifi_filter != NULL) { 1524 result = info->ifi_filter(info->ifi_filter_arg); 1525 if ((result & FILTER_SCHEDULE_THREAD) == 0) 1526 return (result); 1527 } 1528 1529 GROUPTASK_ENQUEUE(gtask); 1530 return (FILTER_HANDLED); 1531 } 1532 1533 static int 1534 _iflib_irq_alloc(if_ctx_t ctx, if_irq_t irq, int rid, 1535 driver_filter_t filter, driver_intr_t handler, void *arg, 1536 const char *name) 1537 { 1538 struct resource *res; 1539 void *tag = NULL; 1540 device_t dev = ctx->ifc_dev; 1541 int flags, i, rc; 1542 1543 flags = RF_ACTIVE; 1544 if (ctx->ifc_flags & IFC_LEGACY) 1545 flags |= RF_SHAREABLE; 1546 MPASS(rid < 512); 1547 i = rid; 1548 res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &i, flags); 1549 if (res == NULL) { 1550 device_printf(dev, 1551 "failed to allocate IRQ for rid %d, name %s.\n", rid, name); 1552 return (ENOMEM); 1553 } 1554 irq->ii_res = res; 1555 KASSERT(filter == NULL || handler == NULL, ("filter and handler can't both be non-NULL")); 1556 rc = bus_setup_intr(dev, res, INTR_MPSAFE | INTR_TYPE_NET, 1557 filter, handler, arg, &tag); 1558 if (rc != 0) { 1559 device_printf(dev, 1560 "failed to setup interrupt for rid %d, name %s: %d\n", 1561 rid, name ? name : "unknown", rc); 1562 return (rc); 1563 } else if (name) 1564 bus_describe_intr(dev, res, tag, "%s", name); 1565 1566 irq->ii_tag = tag; 1567 return (0); 1568 } 1569 1570 /********************************************************************* 1571 * 1572 * Allocate DMA resources for TX buffers as well as memory for the TX 1573 * mbuf map. TX DMA maps (non-TSO/TSO) and TX mbuf map are kept in a 1574 * iflib_sw_tx_desc_array structure, storing all the information that 1575 * is needed to transmit a packet on the wire. This is called only 1576 * once at attach, setup is done every reset. 1577 * 1578 **********************************************************************/ 1579 static int 1580 iflib_txsd_alloc(iflib_txq_t txq) 1581 { 1582 if_ctx_t ctx = txq->ift_ctx; 1583 if_shared_ctx_t sctx = ctx->ifc_sctx; 1584 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 1585 device_t dev = ctx->ifc_dev; 1586 bus_size_t tsomaxsize; 1587 int err, nsegments, ntsosegments; 1588 bool tso; 1589 1590 nsegments = scctx->isc_tx_nsegments; 1591 ntsosegments = scctx->isc_tx_tso_segments_max; 1592 tsomaxsize = scctx->isc_tx_tso_size_max; 1593 if (if_getcapabilities(ctx->ifc_ifp) & IFCAP_VLAN_MTU) 1594 tsomaxsize += sizeof(struct ether_vlan_header); 1595 MPASS(scctx->isc_ntxd[0] > 0); 1596 MPASS(scctx->isc_ntxd[txq->ift_br_offset] > 0); 1597 MPASS(nsegments > 0); 1598 if (if_getcapabilities(ctx->ifc_ifp) & IFCAP_TSO) { 1599 MPASS(ntsosegments > 0); 1600 MPASS(sctx->isc_tso_maxsize >= tsomaxsize); 1601 } 1602 1603 /* 1604 * Set up DMA tags for TX buffers. 1605 */ 1606 if ((err = bus_dma_tag_create(bus_get_dma_tag(dev), 1607 1, 0, /* alignment, bounds */ 1608 BUS_SPACE_MAXADDR, /* lowaddr */ 1609 BUS_SPACE_MAXADDR, /* highaddr */ 1610 NULL, NULL, /* filter, filterarg */ 1611 sctx->isc_tx_maxsize, /* maxsize */ 1612 nsegments, /* nsegments */ 1613 sctx->isc_tx_maxsegsize, /* maxsegsize */ 1614 0, /* flags */ 1615 NULL, /* lockfunc */ 1616 NULL, /* lockfuncarg */ 1617 &txq->ift_buf_tag))) { 1618 device_printf(dev,"Unable to allocate TX DMA tag: %d\n", err); 1619 device_printf(dev,"maxsize: %ju nsegments: %d maxsegsize: %ju\n", 1620 (uintmax_t)sctx->isc_tx_maxsize, nsegments, (uintmax_t)sctx->isc_tx_maxsegsize); 1621 goto fail; 1622 } 1623 tso = (if_getcapabilities(ctx->ifc_ifp) & IFCAP_TSO) != 0; 1624 if (tso && (err = bus_dma_tag_create(bus_get_dma_tag(dev), 1625 1, 0, /* alignment, bounds */ 1626 BUS_SPACE_MAXADDR, /* lowaddr */ 1627 BUS_SPACE_MAXADDR, /* highaddr */ 1628 NULL, NULL, /* filter, filterarg */ 1629 tsomaxsize, /* maxsize */ 1630 ntsosegments, /* nsegments */ 1631 sctx->isc_tso_maxsegsize,/* maxsegsize */ 1632 0, /* flags */ 1633 NULL, /* lockfunc */ 1634 NULL, /* lockfuncarg */ 1635 &txq->ift_tso_buf_tag))) { 1636 device_printf(dev, "Unable to allocate TSO TX DMA tag: %d\n", 1637 err); 1638 goto fail; 1639 } 1640 1641 /* Allocate memory for the TX mbuf map. */ 1642 if (!(txq->ift_sds.ifsd_m = 1643 (struct mbuf **) malloc(sizeof(struct mbuf *) * 1644 scctx->isc_ntxd[txq->ift_br_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { 1645 device_printf(dev, "Unable to allocate TX mbuf map memory\n"); 1646 err = ENOMEM; 1647 goto fail; 1648 } 1649 1650 /* 1651 * Create the DMA maps for TX buffers. 1652 */ 1653 if ((txq->ift_sds.ifsd_map = (bus_dmamap_t *)malloc( 1654 sizeof(bus_dmamap_t) * scctx->isc_ntxd[txq->ift_br_offset], 1655 M_IFLIB, M_NOWAIT | M_ZERO)) == NULL) { 1656 device_printf(dev, 1657 "Unable to allocate TX buffer DMA map memory\n"); 1658 err = ENOMEM; 1659 goto fail; 1660 } 1661 if (tso && (txq->ift_sds.ifsd_tso_map = (bus_dmamap_t *)malloc( 1662 sizeof(bus_dmamap_t) * scctx->isc_ntxd[txq->ift_br_offset], 1663 M_IFLIB, M_NOWAIT | M_ZERO)) == NULL) { 1664 device_printf(dev, 1665 "Unable to allocate TSO TX buffer map memory\n"); 1666 err = ENOMEM; 1667 goto fail; 1668 } 1669 for (int i = 0; i < scctx->isc_ntxd[txq->ift_br_offset]; i++) { 1670 err = bus_dmamap_create(txq->ift_buf_tag, 0, 1671 &txq->ift_sds.ifsd_map[i]); 1672 if (err != 0) { 1673 device_printf(dev, "Unable to create TX DMA map\n"); 1674 goto fail; 1675 } 1676 if (!tso) 1677 continue; 1678 err = bus_dmamap_create(txq->ift_tso_buf_tag, 0, 1679 &txq->ift_sds.ifsd_tso_map[i]); 1680 if (err != 0) { 1681 device_printf(dev, "Unable to create TSO TX DMA map\n"); 1682 goto fail; 1683 } 1684 } 1685 return (0); 1686 fail: 1687 /* We free all, it handles case where we are in the middle */ 1688 iflib_tx_structures_free(ctx); 1689 return (err); 1690 } 1691 1692 static void 1693 iflib_txsd_destroy(if_ctx_t ctx, iflib_txq_t txq, int i) 1694 { 1695 bus_dmamap_t map; 1696 1697 map = NULL; 1698 if (txq->ift_sds.ifsd_map != NULL) 1699 map = txq->ift_sds.ifsd_map[i]; 1700 if (map != NULL) { 1701 bus_dmamap_sync(txq->ift_buf_tag, map, BUS_DMASYNC_POSTWRITE); 1702 bus_dmamap_unload(txq->ift_buf_tag, map); 1703 bus_dmamap_destroy(txq->ift_buf_tag, map); 1704 txq->ift_sds.ifsd_map[i] = NULL; 1705 } 1706 1707 map = NULL; 1708 if (txq->ift_sds.ifsd_tso_map != NULL) 1709 map = txq->ift_sds.ifsd_tso_map[i]; 1710 if (map != NULL) { 1711 bus_dmamap_sync(txq->ift_tso_buf_tag, map, 1712 BUS_DMASYNC_POSTWRITE); 1713 bus_dmamap_unload(txq->ift_tso_buf_tag, map); 1714 bus_dmamap_destroy(txq->ift_tso_buf_tag, map); 1715 txq->ift_sds.ifsd_tso_map[i] = NULL; 1716 } 1717 } 1718 1719 static void 1720 iflib_txq_destroy(iflib_txq_t txq) 1721 { 1722 if_ctx_t ctx = txq->ift_ctx; 1723 1724 for (int i = 0; i < txq->ift_size; i++) 1725 iflib_txsd_destroy(ctx, txq, i); 1726 if (txq->ift_sds.ifsd_map != NULL) { 1727 free(txq->ift_sds.ifsd_map, M_IFLIB); 1728 txq->ift_sds.ifsd_map = NULL; 1729 } 1730 if (txq->ift_sds.ifsd_tso_map != NULL) { 1731 free(txq->ift_sds.ifsd_tso_map, M_IFLIB); 1732 txq->ift_sds.ifsd_tso_map = NULL; 1733 } 1734 if (txq->ift_sds.ifsd_m != NULL) { 1735 free(txq->ift_sds.ifsd_m, M_IFLIB); 1736 txq->ift_sds.ifsd_m = NULL; 1737 } 1738 if (txq->ift_buf_tag != NULL) { 1739 bus_dma_tag_destroy(txq->ift_buf_tag); 1740 txq->ift_buf_tag = NULL; 1741 } 1742 if (txq->ift_tso_buf_tag != NULL) { 1743 bus_dma_tag_destroy(txq->ift_tso_buf_tag); 1744 txq->ift_tso_buf_tag = NULL; 1745 } 1746 } 1747 1748 static void 1749 iflib_txsd_free(if_ctx_t ctx, iflib_txq_t txq, int i) 1750 { 1751 struct mbuf **mp; 1752 1753 mp = &txq->ift_sds.ifsd_m[i]; 1754 if (*mp == NULL) 1755 return; 1756 1757 if (txq->ift_sds.ifsd_map != NULL) { 1758 bus_dmamap_sync(txq->ift_buf_tag, 1759 txq->ift_sds.ifsd_map[i], BUS_DMASYNC_POSTWRITE); 1760 bus_dmamap_unload(txq->ift_buf_tag, txq->ift_sds.ifsd_map[i]); 1761 } 1762 if (txq->ift_sds.ifsd_tso_map != NULL) { 1763 bus_dmamap_sync(txq->ift_tso_buf_tag, 1764 txq->ift_sds.ifsd_tso_map[i], BUS_DMASYNC_POSTWRITE); 1765 bus_dmamap_unload(txq->ift_tso_buf_tag, 1766 txq->ift_sds.ifsd_tso_map[i]); 1767 } 1768 m_free(*mp); 1769 DBG_COUNTER_INC(tx_frees); 1770 *mp = NULL; 1771 } 1772 1773 static int 1774 iflib_txq_setup(iflib_txq_t txq) 1775 { 1776 if_ctx_t ctx = txq->ift_ctx; 1777 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 1778 if_shared_ctx_t sctx = ctx->ifc_sctx; 1779 iflib_dma_info_t di; 1780 int i; 1781 1782 /* Set number of descriptors available */ 1783 txq->ift_qstatus = IFLIB_QUEUE_IDLE; 1784 /* XXX make configurable */ 1785 txq->ift_update_freq = IFLIB_DEFAULT_TX_UPDATE_FREQ; 1786 1787 /* Reset indices */ 1788 txq->ift_cidx_processed = 0; 1789 txq->ift_pidx = txq->ift_cidx = txq->ift_npending = 0; 1790 txq->ift_size = scctx->isc_ntxd[txq->ift_br_offset]; 1791 1792 for (i = 0, di = txq->ift_ifdi; i < sctx->isc_ntxqs; i++, di++) 1793 bzero((void *)di->idi_vaddr, di->idi_size); 1794 1795 IFDI_TXQ_SETUP(ctx, txq->ift_id); 1796 for (i = 0, di = txq->ift_ifdi; i < sctx->isc_ntxqs; i++, di++) 1797 bus_dmamap_sync(di->idi_tag, di->idi_map, 1798 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); 1799 return (0); 1800 } 1801 1802 /********************************************************************* 1803 * 1804 * Allocate DMA resources for RX buffers as well as memory for the RX 1805 * mbuf map, direct RX cluster pointer map and RX cluster bus address 1806 * map. RX DMA map, RX mbuf map, direct RX cluster pointer map and 1807 * RX cluster map are kept in a iflib_sw_rx_desc_array structure. 1808 * Since we use use one entry in iflib_sw_rx_desc_array per received 1809 * packet, the maximum number of entries we'll need is equal to the 1810 * number of hardware receive descriptors that we've allocated. 1811 * 1812 **********************************************************************/ 1813 static int 1814 iflib_rxsd_alloc(iflib_rxq_t rxq) 1815 { 1816 if_ctx_t ctx = rxq->ifr_ctx; 1817 if_shared_ctx_t sctx = ctx->ifc_sctx; 1818 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 1819 device_t dev = ctx->ifc_dev; 1820 iflib_fl_t fl; 1821 int err; 1822 1823 MPASS(scctx->isc_nrxd[0] > 0); 1824 MPASS(scctx->isc_nrxd[rxq->ifr_fl_offset] > 0); 1825 1826 fl = rxq->ifr_fl; 1827 for (int i = 0; i < rxq->ifr_nfl; i++, fl++) { 1828 fl->ifl_size = scctx->isc_nrxd[rxq->ifr_fl_offset]; /* this isn't necessarily the same */ 1829 /* Set up DMA tag for RX buffers. */ 1830 err = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */ 1831 1, 0, /* alignment, bounds */ 1832 BUS_SPACE_MAXADDR, /* lowaddr */ 1833 BUS_SPACE_MAXADDR, /* highaddr */ 1834 NULL, NULL, /* filter, filterarg */ 1835 sctx->isc_rx_maxsize, /* maxsize */ 1836 sctx->isc_rx_nsegments, /* nsegments */ 1837 sctx->isc_rx_maxsegsize, /* maxsegsize */ 1838 0, /* flags */ 1839 NULL, /* lockfunc */ 1840 NULL, /* lockarg */ 1841 &fl->ifl_buf_tag); 1842 if (err) { 1843 device_printf(dev, 1844 "Unable to allocate RX DMA tag: %d\n", err); 1845 goto fail; 1846 } 1847 1848 /* Allocate memory for the RX mbuf map. */ 1849 if (!(fl->ifl_sds.ifsd_m = 1850 (struct mbuf **) malloc(sizeof(struct mbuf *) * 1851 scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { 1852 device_printf(dev, 1853 "Unable to allocate RX mbuf map memory\n"); 1854 err = ENOMEM; 1855 goto fail; 1856 } 1857 1858 /* Allocate memory for the direct RX cluster pointer map. */ 1859 if (!(fl->ifl_sds.ifsd_cl = 1860 (caddr_t *) malloc(sizeof(caddr_t) * 1861 scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { 1862 device_printf(dev, 1863 "Unable to allocate RX cluster map memory\n"); 1864 err = ENOMEM; 1865 goto fail; 1866 } 1867 1868 /* Allocate memory for the RX cluster bus address map. */ 1869 if (!(fl->ifl_sds.ifsd_ba = 1870 (bus_addr_t *) malloc(sizeof(bus_addr_t) * 1871 scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { 1872 device_printf(dev, 1873 "Unable to allocate RX bus address map memory\n"); 1874 err = ENOMEM; 1875 goto fail; 1876 } 1877 1878 /* 1879 * Create the DMA maps for RX buffers. 1880 */ 1881 if (!(fl->ifl_sds.ifsd_map = 1882 (bus_dmamap_t *) malloc(sizeof(bus_dmamap_t) * scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { 1883 device_printf(dev, 1884 "Unable to allocate RX buffer DMA map memory\n"); 1885 err = ENOMEM; 1886 goto fail; 1887 } 1888 for (int i = 0; i < scctx->isc_nrxd[rxq->ifr_fl_offset]; i++) { 1889 err = bus_dmamap_create(fl->ifl_buf_tag, 0, 1890 &fl->ifl_sds.ifsd_map[i]); 1891 if (err != 0) { 1892 device_printf(dev, "Unable to create RX buffer DMA map\n"); 1893 goto fail; 1894 } 1895 } 1896 } 1897 return (0); 1898 1899 fail: 1900 iflib_rx_structures_free(ctx); 1901 return (err); 1902 } 1903 1904 1905 /* 1906 * Internal service routines 1907 */ 1908 1909 struct rxq_refill_cb_arg { 1910 int error; 1911 bus_dma_segment_t seg; 1912 int nseg; 1913 }; 1914 1915 static void 1916 _rxq_refill_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error) 1917 { 1918 struct rxq_refill_cb_arg *cb_arg = arg; 1919 1920 cb_arg->error = error; 1921 cb_arg->seg = segs[0]; 1922 cb_arg->nseg = nseg; 1923 } 1924 1925 /** 1926 * _iflib_fl_refill - refill an rxq free-buffer list 1927 * @ctx: the iflib context 1928 * @fl: the free list to refill 1929 * @count: the number of new buffers to allocate 1930 * 1931 * (Re)populate an rxq free-buffer list with up to @count new packet buffers. 1932 * The caller must assure that @count does not exceed the queue's capacity. 1933 */ 1934 static void 1935 _iflib_fl_refill(if_ctx_t ctx, iflib_fl_t fl, int count) 1936 { 1937 struct if_rxd_update iru; 1938 struct rxq_refill_cb_arg cb_arg; 1939 struct mbuf *m; 1940 caddr_t cl, *sd_cl; 1941 struct mbuf **sd_m; 1942 bus_dmamap_t *sd_map; 1943 bus_addr_t bus_addr, *sd_ba; 1944 int err, frag_idx, i, idx, n, pidx; 1945 qidx_t credits; 1946 1947 sd_m = fl->ifl_sds.ifsd_m; 1948 sd_map = fl->ifl_sds.ifsd_map; 1949 sd_cl = fl->ifl_sds.ifsd_cl; 1950 sd_ba = fl->ifl_sds.ifsd_ba; 1951 pidx = fl->ifl_pidx; 1952 idx = pidx; 1953 frag_idx = fl->ifl_fragidx; 1954 credits = fl->ifl_credits; 1955 1956 i = 0; 1957 n = count; 1958 MPASS(n > 0); 1959 MPASS(credits + n <= fl->ifl_size); 1960 1961 if (pidx < fl->ifl_cidx) 1962 MPASS(pidx + n <= fl->ifl_cidx); 1963 if (pidx == fl->ifl_cidx && (credits < fl->ifl_size)) 1964 MPASS(fl->ifl_gen == 0); 1965 if (pidx > fl->ifl_cidx) 1966 MPASS(n <= fl->ifl_size - pidx + fl->ifl_cidx); 1967 1968 DBG_COUNTER_INC(fl_refills); 1969 if (n > 8) 1970 DBG_COUNTER_INC(fl_refills_large); 1971 iru_init(&iru, fl->ifl_rxq, fl->ifl_id); 1972 while (n--) { 1973 /* 1974 * We allocate an uninitialized mbuf + cluster, mbuf is 1975 * initialized after rx. 1976 * 1977 * If the cluster is still set then we know a minimum sized packet was received 1978 */ 1979 bit_ffc_at(fl->ifl_rx_bitmap, frag_idx, fl->ifl_size, 1980 &frag_idx); 1981 if (frag_idx < 0) 1982 bit_ffc(fl->ifl_rx_bitmap, fl->ifl_size, &frag_idx); 1983 MPASS(frag_idx >= 0); 1984 if ((cl = sd_cl[frag_idx]) == NULL) { 1985 if ((cl = m_cljget(NULL, M_NOWAIT, fl->ifl_buf_size)) == NULL) 1986 break; 1987 1988 cb_arg.error = 0; 1989 MPASS(sd_map != NULL); 1990 err = bus_dmamap_load(fl->ifl_buf_tag, sd_map[frag_idx], 1991 cl, fl->ifl_buf_size, _rxq_refill_cb, &cb_arg, 1992 BUS_DMA_NOWAIT); 1993 if (err != 0 || cb_arg.error) { 1994 /* 1995 * !zone_pack ? 1996 */ 1997 if (fl->ifl_zone == zone_pack) 1998 uma_zfree(fl->ifl_zone, cl); 1999 break; 2000 } 2001 2002 sd_ba[frag_idx] = bus_addr = cb_arg.seg.ds_addr; 2003 sd_cl[frag_idx] = cl; 2004 #if MEMORY_LOGGING 2005 fl->ifl_cl_enqueued++; 2006 #endif 2007 } else { 2008 bus_addr = sd_ba[frag_idx]; 2009 } 2010 bus_dmamap_sync(fl->ifl_buf_tag, sd_map[frag_idx], 2011 BUS_DMASYNC_PREREAD); 2012 2013 if (sd_m[frag_idx] == NULL) { 2014 if ((m = m_gethdr(M_NOWAIT, MT_NOINIT)) == NULL) { 2015 break; 2016 } 2017 sd_m[frag_idx] = m; 2018 } 2019 bit_set(fl->ifl_rx_bitmap, frag_idx); 2020 #if MEMORY_LOGGING 2021 fl->ifl_m_enqueued++; 2022 #endif 2023 2024 DBG_COUNTER_INC(rx_allocs); 2025 fl->ifl_rxd_idxs[i] = frag_idx; 2026 fl->ifl_bus_addrs[i] = bus_addr; 2027 fl->ifl_vm_addrs[i] = cl; 2028 credits++; 2029 i++; 2030 MPASS(credits <= fl->ifl_size); 2031 if (++idx == fl->ifl_size) { 2032 fl->ifl_gen = 1; 2033 idx = 0; 2034 } 2035 if (n == 0 || i == IFLIB_MAX_RX_REFRESH) { 2036 iru.iru_pidx = pidx; 2037 iru.iru_count = i; 2038 ctx->isc_rxd_refill(ctx->ifc_softc, &iru); 2039 i = 0; 2040 pidx = idx; 2041 fl->ifl_pidx = idx; 2042 fl->ifl_credits = credits; 2043 } 2044 } 2045 2046 if (i) { 2047 iru.iru_pidx = pidx; 2048 iru.iru_count = i; 2049 ctx->isc_rxd_refill(ctx->ifc_softc, &iru); 2050 fl->ifl_pidx = idx; 2051 fl->ifl_credits = credits; 2052 } 2053 DBG_COUNTER_INC(rxd_flush); 2054 if (fl->ifl_pidx == 0) 2055 pidx = fl->ifl_size - 1; 2056 else 2057 pidx = fl->ifl_pidx - 1; 2058 2059 bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map, 2060 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); 2061 ctx->isc_rxd_flush(ctx->ifc_softc, fl->ifl_rxq->ifr_id, fl->ifl_id, pidx); 2062 fl->ifl_fragidx = frag_idx; 2063 } 2064 2065 static __inline void 2066 __iflib_fl_refill_lt(if_ctx_t ctx, iflib_fl_t fl, int max) 2067 { 2068 /* we avoid allowing pidx to catch up with cidx as it confuses ixl */ 2069 int32_t reclaimable = fl->ifl_size - fl->ifl_credits - 1; 2070 #ifdef INVARIANTS 2071 int32_t delta = fl->ifl_size - get_inuse(fl->ifl_size, fl->ifl_cidx, fl->ifl_pidx, fl->ifl_gen) - 1; 2072 #endif 2073 2074 MPASS(fl->ifl_credits <= fl->ifl_size); 2075 MPASS(reclaimable == delta); 2076 2077 if (reclaimable > 0) 2078 _iflib_fl_refill(ctx, fl, min(max, reclaimable)); 2079 } 2080 2081 uint8_t 2082 iflib_in_detach(if_ctx_t ctx) 2083 { 2084 bool in_detach; 2085 2086 STATE_LOCK(ctx); 2087 in_detach = !!(ctx->ifc_flags & IFC_IN_DETACH); 2088 STATE_UNLOCK(ctx); 2089 return (in_detach); 2090 } 2091 2092 static void 2093 iflib_fl_bufs_free(iflib_fl_t fl) 2094 { 2095 iflib_dma_info_t idi = fl->ifl_ifdi; 2096 bus_dmamap_t sd_map; 2097 uint32_t i; 2098 2099 for (i = 0; i < fl->ifl_size; i++) { 2100 struct mbuf **sd_m = &fl->ifl_sds.ifsd_m[i]; 2101 caddr_t *sd_cl = &fl->ifl_sds.ifsd_cl[i]; 2102 2103 if (*sd_cl != NULL) { 2104 sd_map = fl->ifl_sds.ifsd_map[i]; 2105 bus_dmamap_sync(fl->ifl_buf_tag, sd_map, 2106 BUS_DMASYNC_POSTREAD); 2107 bus_dmamap_unload(fl->ifl_buf_tag, sd_map); 2108 if (*sd_cl != NULL) 2109 uma_zfree(fl->ifl_zone, *sd_cl); 2110 // XXX: Should this get moved out? 2111 if (iflib_in_detach(fl->ifl_rxq->ifr_ctx)) 2112 bus_dmamap_destroy(fl->ifl_buf_tag, sd_map); 2113 if (*sd_m != NULL) { 2114 m_init(*sd_m, M_NOWAIT, MT_DATA, 0); 2115 uma_zfree(zone_mbuf, *sd_m); 2116 } 2117 } else { 2118 MPASS(*sd_cl == NULL); 2119 MPASS(*sd_m == NULL); 2120 } 2121 #if MEMORY_LOGGING 2122 fl->ifl_m_dequeued++; 2123 fl->ifl_cl_dequeued++; 2124 #endif 2125 *sd_cl = NULL; 2126 *sd_m = NULL; 2127 } 2128 #ifdef INVARIANTS 2129 for (i = 0; i < fl->ifl_size; i++) { 2130 MPASS(fl->ifl_sds.ifsd_cl[i] == NULL); 2131 MPASS(fl->ifl_sds.ifsd_m[i] == NULL); 2132 } 2133 #endif 2134 /* 2135 * Reset free list values 2136 */ 2137 fl->ifl_credits = fl->ifl_cidx = fl->ifl_pidx = fl->ifl_gen = fl->ifl_fragidx = 0; 2138 bzero(idi->idi_vaddr, idi->idi_size); 2139 } 2140 2141 /********************************************************************* 2142 * 2143 * Initialize a free list and its buffers. 2144 * 2145 **********************************************************************/ 2146 static int 2147 iflib_fl_setup(iflib_fl_t fl) 2148 { 2149 iflib_rxq_t rxq = fl->ifl_rxq; 2150 if_ctx_t ctx = rxq->ifr_ctx; 2151 2152 bit_nclear(fl->ifl_rx_bitmap, 0, fl->ifl_size - 1); 2153 /* 2154 ** Free current RX buffer structs and their mbufs 2155 */ 2156 iflib_fl_bufs_free(fl); 2157 /* Now replenish the mbufs */ 2158 MPASS(fl->ifl_credits == 0); 2159 fl->ifl_buf_size = ctx->ifc_rx_mbuf_sz; 2160 if (fl->ifl_buf_size > ctx->ifc_max_fl_buf_size) 2161 ctx->ifc_max_fl_buf_size = fl->ifl_buf_size; 2162 fl->ifl_cltype = m_gettype(fl->ifl_buf_size); 2163 fl->ifl_zone = m_getzone(fl->ifl_buf_size); 2164 2165 2166 /* avoid pre-allocating zillions of clusters to an idle card 2167 * potentially speeding up attach 2168 */ 2169 _iflib_fl_refill(ctx, fl, min(128, fl->ifl_size)); 2170 MPASS(min(128, fl->ifl_size) == fl->ifl_credits); 2171 if (min(128, fl->ifl_size) != fl->ifl_credits) 2172 return (ENOBUFS); 2173 /* 2174 * handle failure 2175 */ 2176 MPASS(rxq != NULL); 2177 MPASS(fl->ifl_ifdi != NULL); 2178 bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map, 2179 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); 2180 return (0); 2181 } 2182 2183 /********************************************************************* 2184 * 2185 * Free receive ring data structures 2186 * 2187 **********************************************************************/ 2188 static void 2189 iflib_rx_sds_free(iflib_rxq_t rxq) 2190 { 2191 iflib_fl_t fl; 2192 int i, j; 2193 2194 if (rxq->ifr_fl != NULL) { 2195 for (i = 0; i < rxq->ifr_nfl; i++) { 2196 fl = &rxq->ifr_fl[i]; 2197 if (fl->ifl_buf_tag != NULL) { 2198 if (fl->ifl_sds.ifsd_map != NULL) { 2199 for (j = 0; j < fl->ifl_size; j++) { 2200 if (fl->ifl_sds.ifsd_map[j] == 2201 NULL) 2202 continue; 2203 bus_dmamap_sync( 2204 fl->ifl_buf_tag, 2205 fl->ifl_sds.ifsd_map[j], 2206 BUS_DMASYNC_POSTREAD); 2207 bus_dmamap_unload( 2208 fl->ifl_buf_tag, 2209 fl->ifl_sds.ifsd_map[j]); 2210 } 2211 } 2212 bus_dma_tag_destroy(fl->ifl_buf_tag); 2213 fl->ifl_buf_tag = NULL; 2214 } 2215 free(fl->ifl_sds.ifsd_m, M_IFLIB); 2216 free(fl->ifl_sds.ifsd_cl, M_IFLIB); 2217 free(fl->ifl_sds.ifsd_ba, M_IFLIB); 2218 free(fl->ifl_sds.ifsd_map, M_IFLIB); 2219 fl->ifl_sds.ifsd_m = NULL; 2220 fl->ifl_sds.ifsd_cl = NULL; 2221 fl->ifl_sds.ifsd_ba = NULL; 2222 fl->ifl_sds.ifsd_map = NULL; 2223 } 2224 free(rxq->ifr_fl, M_IFLIB); 2225 rxq->ifr_fl = NULL; 2226 rxq->ifr_cq_cidx = 0; 2227 } 2228 } 2229 2230 /* 2231 * Timer routine 2232 */ 2233 static void 2234 iflib_timer(void *arg) 2235 { 2236 iflib_txq_t txq = arg; 2237 if_ctx_t ctx = txq->ift_ctx; 2238 if_softc_ctx_t sctx = &ctx->ifc_softc_ctx; 2239 uint64_t this_tick = ticks; 2240 uint32_t reset_on = hz / 2; 2241 2242 if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)) 2243 return; 2244 2245 /* 2246 ** Check on the state of the TX queue(s), this 2247 ** can be done without the lock because its RO 2248 ** and the HUNG state will be static if set. 2249 */ 2250 if (this_tick - txq->ift_last_timer_tick >= hz / 2) { 2251 txq->ift_last_timer_tick = this_tick; 2252 IFDI_TIMER(ctx, txq->ift_id); 2253 if ((txq->ift_qstatus == IFLIB_QUEUE_HUNG) && 2254 ((txq->ift_cleaned_prev == txq->ift_cleaned) || 2255 (sctx->isc_pause_frames == 0))) 2256 goto hung; 2257 2258 if (ifmp_ring_is_stalled(txq->ift_br)) 2259 txq->ift_qstatus = IFLIB_QUEUE_HUNG; 2260 txq->ift_cleaned_prev = txq->ift_cleaned; 2261 } 2262 #ifdef DEV_NETMAP 2263 if (if_getcapenable(ctx->ifc_ifp) & IFCAP_NETMAP) 2264 iflib_netmap_timer_adjust(ctx, txq, &reset_on); 2265 #endif 2266 /* handle any laggards */ 2267 if (txq->ift_db_pending) 2268 GROUPTASK_ENQUEUE(&txq->ift_task); 2269 2270 sctx->isc_pause_frames = 0; 2271 if (if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING) 2272 callout_reset_on(&txq->ift_timer, reset_on, iflib_timer, txq, txq->ift_timer.c_cpu); 2273 return; 2274 2275 hung: 2276 device_printf(ctx->ifc_dev, 2277 "Watchdog timeout (TX: %d desc avail: %d pidx: %d) -- resetting\n", 2278 txq->ift_id, TXQ_AVAIL(txq), txq->ift_pidx); 2279 STATE_LOCK(ctx); 2280 if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING); 2281 ctx->ifc_flags |= (IFC_DO_WATCHDOG|IFC_DO_RESET); 2282 iflib_admin_intr_deferred(ctx); 2283 STATE_UNLOCK(ctx); 2284 } 2285 2286 static void 2287 iflib_calc_rx_mbuf_sz(if_ctx_t ctx) 2288 { 2289 if_softc_ctx_t sctx = &ctx->ifc_softc_ctx; 2290 2291 /* 2292 * XXX don't set the max_frame_size to larger 2293 * than the hardware can handle 2294 */ 2295 if (sctx->isc_max_frame_size <= MCLBYTES) 2296 ctx->ifc_rx_mbuf_sz = MCLBYTES; 2297 else 2298 ctx->ifc_rx_mbuf_sz = MJUMPAGESIZE; 2299 } 2300 2301 uint32_t 2302 iflib_get_rx_mbuf_sz(if_ctx_t ctx) 2303 { 2304 2305 return (ctx->ifc_rx_mbuf_sz); 2306 } 2307 2308 static void 2309 iflib_init_locked(if_ctx_t ctx) 2310 { 2311 if_softc_ctx_t sctx = &ctx->ifc_softc_ctx; 2312 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 2313 if_t ifp = ctx->ifc_ifp; 2314 iflib_fl_t fl; 2315 iflib_txq_t txq; 2316 iflib_rxq_t rxq; 2317 int i, j, tx_ip_csum_flags, tx_ip6_csum_flags; 2318 2319 if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING); 2320 IFDI_INTR_DISABLE(ctx); 2321 2322 tx_ip_csum_flags = scctx->isc_tx_csum_flags & (CSUM_IP | CSUM_TCP | CSUM_UDP | CSUM_SCTP); 2323 tx_ip6_csum_flags = scctx->isc_tx_csum_flags & (CSUM_IP6_TCP | CSUM_IP6_UDP | CSUM_IP6_SCTP); 2324 /* Set hardware offload abilities */ 2325 if_clearhwassist(ifp); 2326 if (if_getcapenable(ifp) & IFCAP_TXCSUM) 2327 if_sethwassistbits(ifp, tx_ip_csum_flags, 0); 2328 if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6) 2329 if_sethwassistbits(ifp, tx_ip6_csum_flags, 0); 2330 if (if_getcapenable(ifp) & IFCAP_TSO4) 2331 if_sethwassistbits(ifp, CSUM_IP_TSO, 0); 2332 if (if_getcapenable(ifp) & IFCAP_TSO6) 2333 if_sethwassistbits(ifp, CSUM_IP6_TSO, 0); 2334 2335 for (i = 0, txq = ctx->ifc_txqs; i < sctx->isc_ntxqsets; i++, txq++) { 2336 CALLOUT_LOCK(txq); 2337 callout_stop(&txq->ift_timer); 2338 CALLOUT_UNLOCK(txq); 2339 iflib_netmap_txq_init(ctx, txq); 2340 } 2341 2342 /* 2343 * Calculate a suitable Rx mbuf size prior to calling IFDI_INIT, so 2344 * that drivers can use the value when setting up the hardware receive 2345 * buffers. 2346 */ 2347 iflib_calc_rx_mbuf_sz(ctx); 2348 2349 #ifdef INVARIANTS 2350 i = if_getdrvflags(ifp); 2351 #endif 2352 IFDI_INIT(ctx); 2353 MPASS(if_getdrvflags(ifp) == i); 2354 for (i = 0, rxq = ctx->ifc_rxqs; i < sctx->isc_nrxqsets; i++, rxq++) { 2355 /* XXX this should really be done on a per-queue basis */ 2356 if (if_getcapenable(ifp) & IFCAP_NETMAP) { 2357 MPASS(rxq->ifr_id == i); 2358 iflib_netmap_rxq_init(ctx, rxq); 2359 continue; 2360 } 2361 for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) { 2362 if (iflib_fl_setup(fl)) { 2363 device_printf(ctx->ifc_dev, 2364 "setting up free list %d failed - " 2365 "check cluster settings\n", j); 2366 goto done; 2367 } 2368 } 2369 } 2370 done: 2371 if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_RUNNING, IFF_DRV_OACTIVE); 2372 IFDI_INTR_ENABLE(ctx); 2373 txq = ctx->ifc_txqs; 2374 for (i = 0; i < sctx->isc_ntxqsets; i++, txq++) 2375 callout_reset_on(&txq->ift_timer, hz/2, iflib_timer, txq, 2376 txq->ift_timer.c_cpu); 2377 } 2378 2379 static int 2380 iflib_media_change(if_t ifp) 2381 { 2382 if_ctx_t ctx = if_getsoftc(ifp); 2383 int err; 2384 2385 CTX_LOCK(ctx); 2386 if ((err = IFDI_MEDIA_CHANGE(ctx)) == 0) 2387 iflib_init_locked(ctx); 2388 CTX_UNLOCK(ctx); 2389 return (err); 2390 } 2391 2392 static void 2393 iflib_media_status(if_t ifp, struct ifmediareq *ifmr) 2394 { 2395 if_ctx_t ctx = if_getsoftc(ifp); 2396 2397 CTX_LOCK(ctx); 2398 IFDI_UPDATE_ADMIN_STATUS(ctx); 2399 IFDI_MEDIA_STATUS(ctx, ifmr); 2400 CTX_UNLOCK(ctx); 2401 } 2402 2403 void 2404 iflib_stop(if_ctx_t ctx) 2405 { 2406 iflib_txq_t txq = ctx->ifc_txqs; 2407 iflib_rxq_t rxq = ctx->ifc_rxqs; 2408 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 2409 if_shared_ctx_t sctx = ctx->ifc_sctx; 2410 iflib_dma_info_t di; 2411 iflib_fl_t fl; 2412 int i, j; 2413 2414 /* Tell the stack that the interface is no longer active */ 2415 if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING); 2416 2417 IFDI_INTR_DISABLE(ctx); 2418 DELAY(1000); 2419 IFDI_STOP(ctx); 2420 DELAY(1000); 2421 2422 iflib_debug_reset(); 2423 /* Wait for current tx queue users to exit to disarm watchdog timer. */ 2424 for (i = 0; i < scctx->isc_ntxqsets; i++, txq++) { 2425 /* make sure all transmitters have completed before proceeding XXX */ 2426 2427 CALLOUT_LOCK(txq); 2428 callout_stop(&txq->ift_timer); 2429 CALLOUT_UNLOCK(txq); 2430 2431 /* clean any enqueued buffers */ 2432 iflib_ifmp_purge(txq); 2433 /* Free any existing tx buffers. */ 2434 for (j = 0; j < txq->ift_size; j++) { 2435 iflib_txsd_free(ctx, txq, j); 2436 } 2437 txq->ift_processed = txq->ift_cleaned = txq->ift_cidx_processed = 0; 2438 txq->ift_in_use = txq->ift_gen = txq->ift_cidx = txq->ift_pidx = txq->ift_no_desc_avail = 0; 2439 txq->ift_closed = txq->ift_mbuf_defrag = txq->ift_mbuf_defrag_failed = 0; 2440 txq->ift_no_tx_dma_setup = txq->ift_txd_encap_efbig = txq->ift_map_failed = 0; 2441 txq->ift_pullups = 0; 2442 ifmp_ring_reset_stats(txq->ift_br); 2443 for (j = 0, di = txq->ift_ifdi; j < sctx->isc_ntxqs; j++, di++) 2444 bzero((void *)di->idi_vaddr, di->idi_size); 2445 } 2446 for (i = 0; i < scctx->isc_nrxqsets; i++, rxq++) { 2447 /* make sure all transmitters have completed before proceeding XXX */ 2448 2449 rxq->ifr_cq_cidx = 0; 2450 for (j = 0, di = rxq->ifr_ifdi; j < sctx->isc_nrxqs; j++, di++) 2451 bzero((void *)di->idi_vaddr, di->idi_size); 2452 /* also resets the free lists pidx/cidx */ 2453 for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) 2454 iflib_fl_bufs_free(fl); 2455 } 2456 } 2457 2458 static inline caddr_t 2459 calc_next_rxd(iflib_fl_t fl, int cidx) 2460 { 2461 qidx_t size; 2462 int nrxd; 2463 caddr_t start, end, cur, next; 2464 2465 nrxd = fl->ifl_size; 2466 size = fl->ifl_rxd_size; 2467 start = fl->ifl_ifdi->idi_vaddr; 2468 2469 if (__predict_false(size == 0)) 2470 return (start); 2471 cur = start + size*cidx; 2472 end = start + size*nrxd; 2473 next = CACHE_PTR_NEXT(cur); 2474 return (next < end ? next : start); 2475 } 2476 2477 static inline void 2478 prefetch_pkts(iflib_fl_t fl, int cidx) 2479 { 2480 int nextptr; 2481 int nrxd = fl->ifl_size; 2482 caddr_t next_rxd; 2483 2484 2485 nextptr = (cidx + CACHE_PTR_INCREMENT) & (nrxd-1); 2486 prefetch(&fl->ifl_sds.ifsd_m[nextptr]); 2487 prefetch(&fl->ifl_sds.ifsd_cl[nextptr]); 2488 next_rxd = calc_next_rxd(fl, cidx); 2489 prefetch(next_rxd); 2490 prefetch(fl->ifl_sds.ifsd_m[(cidx + 1) & (nrxd-1)]); 2491 prefetch(fl->ifl_sds.ifsd_m[(cidx + 2) & (nrxd-1)]); 2492 prefetch(fl->ifl_sds.ifsd_m[(cidx + 3) & (nrxd-1)]); 2493 prefetch(fl->ifl_sds.ifsd_m[(cidx + 4) & (nrxd-1)]); 2494 prefetch(fl->ifl_sds.ifsd_cl[(cidx + 1) & (nrxd-1)]); 2495 prefetch(fl->ifl_sds.ifsd_cl[(cidx + 2) & (nrxd-1)]); 2496 prefetch(fl->ifl_sds.ifsd_cl[(cidx + 3) & (nrxd-1)]); 2497 prefetch(fl->ifl_sds.ifsd_cl[(cidx + 4) & (nrxd-1)]); 2498 } 2499 2500 static struct mbuf * 2501 rxd_frag_to_sd(iflib_rxq_t rxq, if_rxd_frag_t irf, bool unload, if_rxsd_t sd, 2502 int *pf_rv, if_rxd_info_t ri) 2503 { 2504 bus_dmamap_t map; 2505 iflib_fl_t fl; 2506 caddr_t payload; 2507 struct mbuf *m; 2508 int flid, cidx, len, next; 2509 2510 map = NULL; 2511 flid = irf->irf_flid; 2512 cidx = irf->irf_idx; 2513 fl = &rxq->ifr_fl[flid]; 2514 sd->ifsd_fl = fl; 2515 sd->ifsd_cidx = cidx; 2516 m = fl->ifl_sds.ifsd_m[cidx]; 2517 sd->ifsd_cl = &fl->ifl_sds.ifsd_cl[cidx]; 2518 fl->ifl_credits--; 2519 #if MEMORY_LOGGING 2520 fl->ifl_m_dequeued++; 2521 #endif 2522 if (rxq->ifr_ctx->ifc_flags & IFC_PREFETCH) 2523 prefetch_pkts(fl, cidx); 2524 next = (cidx + CACHE_PTR_INCREMENT) & (fl->ifl_size-1); 2525 prefetch(&fl->ifl_sds.ifsd_map[next]); 2526 map = fl->ifl_sds.ifsd_map[cidx]; 2527 next = (cidx + CACHE_LINE_SIZE) & (fl->ifl_size-1); 2528 2529 /* not valid assert if bxe really does SGE from non-contiguous elements */ 2530 MPASS(fl->ifl_cidx == cidx); 2531 bus_dmamap_sync(fl->ifl_buf_tag, map, BUS_DMASYNC_POSTREAD); 2532 2533 if (rxq->pfil != NULL && PFIL_HOOKED_IN(rxq->pfil) && pf_rv != NULL) { 2534 payload = *sd->ifsd_cl; 2535 payload += ri->iri_pad; 2536 len = ri->iri_len - ri->iri_pad; 2537 *pf_rv = pfil_run_hooks(rxq->pfil, payload, ri->iri_ifp, 2538 len | PFIL_MEMPTR | PFIL_IN, NULL); 2539 switch (*pf_rv) { 2540 case PFIL_DROPPED: 2541 case PFIL_CONSUMED: 2542 /* 2543 * The filter ate it. Everything is recycled. 2544 */ 2545 m = NULL; 2546 unload = 0; 2547 break; 2548 case PFIL_REALLOCED: 2549 /* 2550 * The filter copied it. Everything is recycled. 2551 */ 2552 m = pfil_mem2mbuf(payload); 2553 unload = 0; 2554 break; 2555 case PFIL_PASS: 2556 /* 2557 * Filter said it was OK, so receive like 2558 * normal 2559 */ 2560 fl->ifl_sds.ifsd_m[cidx] = NULL; 2561 break; 2562 default: 2563 MPASS(0); 2564 } 2565 } else { 2566 fl->ifl_sds.ifsd_m[cidx] = NULL; 2567 *pf_rv = PFIL_PASS; 2568 } 2569 2570 if (unload) 2571 bus_dmamap_unload(fl->ifl_buf_tag, map); 2572 fl->ifl_cidx = (fl->ifl_cidx + 1) & (fl->ifl_size-1); 2573 if (__predict_false(fl->ifl_cidx == 0)) 2574 fl->ifl_gen = 0; 2575 bit_clear(fl->ifl_rx_bitmap, cidx); 2576 return (m); 2577 } 2578 2579 static struct mbuf * 2580 assemble_segments(iflib_rxq_t rxq, if_rxd_info_t ri, if_rxsd_t sd, int *pf_rv) 2581 { 2582 struct mbuf *m, *mh, *mt; 2583 caddr_t cl; 2584 int *pf_rv_ptr, flags, i, padlen; 2585 bool consumed; 2586 2587 i = 0; 2588 mh = NULL; 2589 consumed = false; 2590 *pf_rv = PFIL_PASS; 2591 pf_rv_ptr = pf_rv; 2592 do { 2593 m = rxd_frag_to_sd(rxq, &ri->iri_frags[i], !consumed, sd, 2594 pf_rv_ptr, ri); 2595 2596 MPASS(*sd->ifsd_cl != NULL); 2597 2598 /* 2599 * Exclude zero-length frags & frags from 2600 * packets the filter has consumed or dropped 2601 */ 2602 if (ri->iri_frags[i].irf_len == 0 || consumed || 2603 *pf_rv == PFIL_CONSUMED || *pf_rv == PFIL_DROPPED) { 2604 if (mh == NULL) { 2605 /* everything saved here */ 2606 consumed = true; 2607 pf_rv_ptr = NULL; 2608 continue; 2609 } 2610 /* XXX we can save the cluster here, but not the mbuf */ 2611 m_init(m, M_NOWAIT, MT_DATA, 0); 2612 m_free(m); 2613 continue; 2614 } 2615 if (mh == NULL) { 2616 flags = M_PKTHDR|M_EXT; 2617 mh = mt = m; 2618 padlen = ri->iri_pad; 2619 } else { 2620 flags = M_EXT; 2621 mt->m_next = m; 2622 mt = m; 2623 /* assuming padding is only on the first fragment */ 2624 padlen = 0; 2625 } 2626 cl = *sd->ifsd_cl; 2627 *sd->ifsd_cl = NULL; 2628 2629 /* Can these two be made one ? */ 2630 m_init(m, M_NOWAIT, MT_DATA, flags); 2631 m_cljset(m, cl, sd->ifsd_fl->ifl_cltype); 2632 /* 2633 * These must follow m_init and m_cljset 2634 */ 2635 m->m_data += padlen; 2636 ri->iri_len -= padlen; 2637 m->m_len = ri->iri_frags[i].irf_len; 2638 } while (++i < ri->iri_nfrags); 2639 2640 return (mh); 2641 } 2642 2643 /* 2644 * Process one software descriptor 2645 */ 2646 static struct mbuf * 2647 iflib_rxd_pkt_get(iflib_rxq_t rxq, if_rxd_info_t ri) 2648 { 2649 struct if_rxsd sd; 2650 struct mbuf *m; 2651 int pf_rv; 2652 2653 /* should I merge this back in now that the two paths are basically duplicated? */ 2654 if (ri->iri_nfrags == 1 && 2655 ri->iri_frags[0].irf_len <= MIN(IFLIB_RX_COPY_THRESH, MHLEN)) { 2656 m = rxd_frag_to_sd(rxq, &ri->iri_frags[0], false, &sd, 2657 &pf_rv, ri); 2658 if (pf_rv != PFIL_PASS && pf_rv != PFIL_REALLOCED) 2659 return (m); 2660 if (pf_rv == PFIL_PASS) { 2661 m_init(m, M_NOWAIT, MT_DATA, M_PKTHDR); 2662 #ifndef __NO_STRICT_ALIGNMENT 2663 if (!IP_ALIGNED(m)) 2664 m->m_data += 2; 2665 #endif 2666 memcpy(m->m_data, *sd.ifsd_cl, ri->iri_len); 2667 m->m_len = ri->iri_frags[0].irf_len; 2668 } 2669 } else { 2670 m = assemble_segments(rxq, ri, &sd, &pf_rv); 2671 if (pf_rv != PFIL_PASS && pf_rv != PFIL_REALLOCED) 2672 return (m); 2673 } 2674 m->m_pkthdr.len = ri->iri_len; 2675 m->m_pkthdr.rcvif = ri->iri_ifp; 2676 m->m_flags |= ri->iri_flags; 2677 m->m_pkthdr.ether_vtag = ri->iri_vtag; 2678 m->m_pkthdr.flowid = ri->iri_flowid; 2679 M_HASHTYPE_SET(m, ri->iri_rsstype); 2680 m->m_pkthdr.csum_flags = ri->iri_csum_flags; 2681 m->m_pkthdr.csum_data = ri->iri_csum_data; 2682 return (m); 2683 } 2684 2685 #if defined(INET6) || defined(INET) 2686 static void 2687 iflib_get_ip_forwarding(struct lro_ctrl *lc, bool *v4, bool *v6) 2688 { 2689 CURVNET_SET(lc->ifp->if_vnet); 2690 #if defined(INET6) 2691 *v6 = V_ip6_forwarding; 2692 #endif 2693 #if defined(INET) 2694 *v4 = V_ipforwarding; 2695 #endif 2696 CURVNET_RESTORE(); 2697 } 2698 2699 /* 2700 * Returns true if it's possible this packet could be LROed. 2701 * if it returns false, it is guaranteed that tcp_lro_rx() 2702 * would not return zero. 2703 */ 2704 static bool 2705 iflib_check_lro_possible(struct mbuf *m, bool v4_forwarding, bool v6_forwarding) 2706 { 2707 struct ether_header *eh; 2708 2709 eh = mtod(m, struct ether_header *); 2710 switch (eh->ether_type) { 2711 #if defined(INET6) 2712 case htons(ETHERTYPE_IPV6): 2713 return (!v6_forwarding); 2714 #endif 2715 #if defined (INET) 2716 case htons(ETHERTYPE_IP): 2717 return (!v4_forwarding); 2718 #endif 2719 } 2720 2721 return false; 2722 } 2723 #else 2724 static void 2725 iflib_get_ip_forwarding(struct lro_ctrl *lc __unused, bool *v4 __unused, bool *v6 __unused) 2726 { 2727 } 2728 #endif 2729 2730 static bool 2731 iflib_rxeof(iflib_rxq_t rxq, qidx_t budget) 2732 { 2733 if_t ifp; 2734 if_ctx_t ctx = rxq->ifr_ctx; 2735 if_shared_ctx_t sctx = ctx->ifc_sctx; 2736 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 2737 int avail, i; 2738 qidx_t *cidxp; 2739 struct if_rxd_info ri; 2740 int err, budget_left, rx_bytes, rx_pkts; 2741 iflib_fl_t fl; 2742 int lro_enabled; 2743 bool v4_forwarding, v6_forwarding, lro_possible; 2744 2745 /* 2746 * XXX early demux data packets so that if_input processing only handles 2747 * acks in interrupt context 2748 */ 2749 struct mbuf *m, *mh, *mt, *mf; 2750 2751 lro_possible = v4_forwarding = v6_forwarding = false; 2752 ifp = ctx->ifc_ifp; 2753 mh = mt = NULL; 2754 MPASS(budget > 0); 2755 rx_pkts = rx_bytes = 0; 2756 if (sctx->isc_flags & IFLIB_HAS_RXCQ) 2757 cidxp = &rxq->ifr_cq_cidx; 2758 else 2759 cidxp = &rxq->ifr_fl[0].ifl_cidx; 2760 if ((avail = iflib_rxd_avail(ctx, rxq, *cidxp, budget)) == 0) { 2761 for (i = 0, fl = &rxq->ifr_fl[0]; i < sctx->isc_nfl; i++, fl++) 2762 __iflib_fl_refill_lt(ctx, fl, budget + 8); 2763 DBG_COUNTER_INC(rx_unavail); 2764 return (false); 2765 } 2766 2767 /* pfil needs the vnet to be set */ 2768 CURVNET_SET_QUIET(ifp->if_vnet); 2769 for (budget_left = budget; budget_left > 0 && avail > 0;) { 2770 if (__predict_false(!CTX_ACTIVE(ctx))) { 2771 DBG_COUNTER_INC(rx_ctx_inactive); 2772 break; 2773 } 2774 /* 2775 * Reset client set fields to their default values 2776 */ 2777 rxd_info_zero(&ri); 2778 ri.iri_qsidx = rxq->ifr_id; 2779 ri.iri_cidx = *cidxp; 2780 ri.iri_ifp = ifp; 2781 ri.iri_frags = rxq->ifr_frags; 2782 err = ctx->isc_rxd_pkt_get(ctx->ifc_softc, &ri); 2783 2784 if (err) 2785 goto err; 2786 rx_pkts += 1; 2787 rx_bytes += ri.iri_len; 2788 if (sctx->isc_flags & IFLIB_HAS_RXCQ) { 2789 *cidxp = ri.iri_cidx; 2790 /* Update our consumer index */ 2791 /* XXX NB: shurd - check if this is still safe */ 2792 while (rxq->ifr_cq_cidx >= scctx->isc_nrxd[0]) 2793 rxq->ifr_cq_cidx -= scctx->isc_nrxd[0]; 2794 /* was this only a completion queue message? */ 2795 if (__predict_false(ri.iri_nfrags == 0)) 2796 continue; 2797 } 2798 MPASS(ri.iri_nfrags != 0); 2799 MPASS(ri.iri_len != 0); 2800 2801 /* will advance the cidx on the corresponding free lists */ 2802 m = iflib_rxd_pkt_get(rxq, &ri); 2803 avail--; 2804 budget_left--; 2805 if (avail == 0 && budget_left) 2806 avail = iflib_rxd_avail(ctx, rxq, *cidxp, budget_left); 2807 2808 if (__predict_false(m == NULL)) 2809 continue; 2810 2811 /* imm_pkt: -- cxgb */ 2812 if (mh == NULL) 2813 mh = mt = m; 2814 else { 2815 mt->m_nextpkt = m; 2816 mt = m; 2817 } 2818 } 2819 CURVNET_RESTORE(); 2820 /* make sure that we can refill faster than drain */ 2821 for (i = 0, fl = &rxq->ifr_fl[0]; i < sctx->isc_nfl; i++, fl++) 2822 __iflib_fl_refill_lt(ctx, fl, budget + 8); 2823 2824 lro_enabled = (if_getcapenable(ifp) & IFCAP_LRO); 2825 if (lro_enabled) 2826 iflib_get_ip_forwarding(&rxq->ifr_lc, &v4_forwarding, &v6_forwarding); 2827 mt = mf = NULL; 2828 while (mh != NULL) { 2829 m = mh; 2830 mh = mh->m_nextpkt; 2831 m->m_nextpkt = NULL; 2832 #ifndef __NO_STRICT_ALIGNMENT 2833 if (!IP_ALIGNED(m) && (m = iflib_fixup_rx(m)) == NULL) 2834 continue; 2835 #endif 2836 rx_bytes += m->m_pkthdr.len; 2837 rx_pkts++; 2838 #if defined(INET6) || defined(INET) 2839 if (lro_enabled) { 2840 if (!lro_possible) { 2841 lro_possible = iflib_check_lro_possible(m, v4_forwarding, v6_forwarding); 2842 if (lro_possible && mf != NULL) { 2843 ifp->if_input(ifp, mf); 2844 DBG_COUNTER_INC(rx_if_input); 2845 mt = mf = NULL; 2846 } 2847 } 2848 if ((m->m_pkthdr.csum_flags & (CSUM_L4_CALC|CSUM_L4_VALID)) == 2849 (CSUM_L4_CALC|CSUM_L4_VALID)) { 2850 if (lro_possible && tcp_lro_rx(&rxq->ifr_lc, m, 0) == 0) 2851 continue; 2852 } 2853 } 2854 #endif 2855 if (lro_possible) { 2856 ifp->if_input(ifp, m); 2857 DBG_COUNTER_INC(rx_if_input); 2858 continue; 2859 } 2860 2861 if (mf == NULL) 2862 mf = m; 2863 if (mt != NULL) 2864 mt->m_nextpkt = m; 2865 mt = m; 2866 } 2867 if (mf != NULL) { 2868 ifp->if_input(ifp, mf); 2869 DBG_COUNTER_INC(rx_if_input); 2870 } 2871 2872 if_inc_counter(ifp, IFCOUNTER_IBYTES, rx_bytes); 2873 if_inc_counter(ifp, IFCOUNTER_IPACKETS, rx_pkts); 2874 2875 /* 2876 * Flush any outstanding LRO work 2877 */ 2878 #if defined(INET6) || defined(INET) 2879 tcp_lro_flush_all(&rxq->ifr_lc); 2880 #endif 2881 if (avail) 2882 return true; 2883 return (iflib_rxd_avail(ctx, rxq, *cidxp, 1)); 2884 err: 2885 STATE_LOCK(ctx); 2886 ctx->ifc_flags |= IFC_DO_RESET; 2887 iflib_admin_intr_deferred(ctx); 2888 STATE_UNLOCK(ctx); 2889 return (false); 2890 } 2891 2892 #define TXD_NOTIFY_COUNT(txq) (((txq)->ift_size / (txq)->ift_update_freq)-1) 2893 static inline qidx_t 2894 txq_max_db_deferred(iflib_txq_t txq, qidx_t in_use) 2895 { 2896 qidx_t notify_count = TXD_NOTIFY_COUNT(txq); 2897 qidx_t minthresh = txq->ift_size / 8; 2898 if (in_use > 4*minthresh) 2899 return (notify_count); 2900 if (in_use > 2*minthresh) 2901 return (notify_count >> 1); 2902 if (in_use > minthresh) 2903 return (notify_count >> 3); 2904 return (0); 2905 } 2906 2907 static inline qidx_t 2908 txq_max_rs_deferred(iflib_txq_t txq) 2909 { 2910 qidx_t notify_count = TXD_NOTIFY_COUNT(txq); 2911 qidx_t minthresh = txq->ift_size / 8; 2912 if (txq->ift_in_use > 4*minthresh) 2913 return (notify_count); 2914 if (txq->ift_in_use > 2*minthresh) 2915 return (notify_count >> 1); 2916 if (txq->ift_in_use > minthresh) 2917 return (notify_count >> 2); 2918 return (2); 2919 } 2920 2921 #define M_CSUM_FLAGS(m) ((m)->m_pkthdr.csum_flags) 2922 #define M_HAS_VLANTAG(m) (m->m_flags & M_VLANTAG) 2923 2924 #define TXQ_MAX_DB_DEFERRED(txq, in_use) txq_max_db_deferred((txq), (in_use)) 2925 #define TXQ_MAX_RS_DEFERRED(txq) txq_max_rs_deferred(txq) 2926 #define TXQ_MAX_DB_CONSUMED(size) (size >> 4) 2927 2928 /* forward compatibility for cxgb */ 2929 #define FIRST_QSET(ctx) 0 2930 #define NTXQSETS(ctx) ((ctx)->ifc_softc_ctx.isc_ntxqsets) 2931 #define NRXQSETS(ctx) ((ctx)->ifc_softc_ctx.isc_nrxqsets) 2932 #define QIDX(ctx, m) ((((m)->m_pkthdr.flowid & ctx->ifc_softc_ctx.isc_rss_table_mask) % NTXQSETS(ctx)) + FIRST_QSET(ctx)) 2933 #define DESC_RECLAIMABLE(q) ((int)((q)->ift_processed - (q)->ift_cleaned - (q)->ift_ctx->ifc_softc_ctx.isc_tx_nsegments)) 2934 2935 /* XXX we should be setting this to something other than zero */ 2936 #define RECLAIM_THRESH(ctx) ((ctx)->ifc_sctx->isc_tx_reclaim_thresh) 2937 #define MAX_TX_DESC(ctx) max((ctx)->ifc_softc_ctx.isc_tx_tso_segments_max, \ 2938 (ctx)->ifc_softc_ctx.isc_tx_nsegments) 2939 2940 static inline bool 2941 iflib_txd_db_check(if_ctx_t ctx, iflib_txq_t txq, int ring, qidx_t in_use) 2942 { 2943 qidx_t dbval, max; 2944 bool rang; 2945 2946 rang = false; 2947 max = TXQ_MAX_DB_DEFERRED(txq, in_use); 2948 if (ring || txq->ift_db_pending >= max) { 2949 dbval = txq->ift_npending ? txq->ift_npending : txq->ift_pidx; 2950 bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map, 2951 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); 2952 ctx->isc_txd_flush(ctx->ifc_softc, txq->ift_id, dbval); 2953 txq->ift_db_pending = txq->ift_npending = 0; 2954 rang = true; 2955 } 2956 return (rang); 2957 } 2958 2959 #ifdef PKT_DEBUG 2960 static void 2961 print_pkt(if_pkt_info_t pi) 2962 { 2963 printf("pi len: %d qsidx: %d nsegs: %d ndescs: %d flags: %x pidx: %d\n", 2964 pi->ipi_len, pi->ipi_qsidx, pi->ipi_nsegs, pi->ipi_ndescs, pi->ipi_flags, pi->ipi_pidx); 2965 printf("pi new_pidx: %d csum_flags: %lx tso_segsz: %d mflags: %x vtag: %d\n", 2966 pi->ipi_new_pidx, pi->ipi_csum_flags, pi->ipi_tso_segsz, pi->ipi_mflags, pi->ipi_vtag); 2967 printf("pi etype: %d ehdrlen: %d ip_hlen: %d ipproto: %d\n", 2968 pi->ipi_etype, pi->ipi_ehdrlen, pi->ipi_ip_hlen, pi->ipi_ipproto); 2969 } 2970 #endif 2971 2972 #define IS_TSO4(pi) ((pi)->ipi_csum_flags & CSUM_IP_TSO) 2973 #define IS_TX_OFFLOAD4(pi) ((pi)->ipi_csum_flags & (CSUM_IP_TCP | CSUM_IP_TSO)) 2974 #define IS_TSO6(pi) ((pi)->ipi_csum_flags & CSUM_IP6_TSO) 2975 #define IS_TX_OFFLOAD6(pi) ((pi)->ipi_csum_flags & (CSUM_IP6_TCP | CSUM_IP6_TSO)) 2976 2977 static int 2978 iflib_parse_header(iflib_txq_t txq, if_pkt_info_t pi, struct mbuf **mp) 2979 { 2980 if_shared_ctx_t sctx = txq->ift_ctx->ifc_sctx; 2981 struct ether_vlan_header *eh; 2982 struct mbuf *m; 2983 2984 m = *mp; 2985 if ((sctx->isc_flags & IFLIB_NEED_SCRATCH) && 2986 M_WRITABLE(m) == 0) { 2987 if ((m = m_dup(m, M_NOWAIT)) == NULL) { 2988 return (ENOMEM); 2989 } else { 2990 m_freem(*mp); 2991 DBG_COUNTER_INC(tx_frees); 2992 *mp = m; 2993 } 2994 } 2995 2996 /* 2997 * Determine where frame payload starts. 2998 * Jump over vlan headers if already present, 2999 * helpful for QinQ too. 3000 */ 3001 if (__predict_false(m->m_len < sizeof(*eh))) { 3002 txq->ift_pullups++; 3003 if (__predict_false((m = m_pullup(m, sizeof(*eh))) == NULL)) 3004 return (ENOMEM); 3005 } 3006 eh = mtod(m, struct ether_vlan_header *); 3007 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) { 3008 pi->ipi_etype = ntohs(eh->evl_proto); 3009 pi->ipi_ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN; 3010 } else { 3011 pi->ipi_etype = ntohs(eh->evl_encap_proto); 3012 pi->ipi_ehdrlen = ETHER_HDR_LEN; 3013 } 3014 3015 switch (pi->ipi_etype) { 3016 #ifdef INET 3017 case ETHERTYPE_IP: 3018 { 3019 struct mbuf *n; 3020 struct ip *ip = NULL; 3021 struct tcphdr *th = NULL; 3022 int minthlen; 3023 3024 minthlen = min(m->m_pkthdr.len, pi->ipi_ehdrlen + sizeof(*ip) + sizeof(*th)); 3025 if (__predict_false(m->m_len < minthlen)) { 3026 /* 3027 * if this code bloat is causing too much of a hit 3028 * move it to a separate function and mark it noinline 3029 */ 3030 if (m->m_len == pi->ipi_ehdrlen) { 3031 n = m->m_next; 3032 MPASS(n); 3033 if (n->m_len >= sizeof(*ip)) { 3034 ip = (struct ip *)n->m_data; 3035 if (n->m_len >= (ip->ip_hl << 2) + sizeof(*th)) 3036 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 3037 } else { 3038 txq->ift_pullups++; 3039 if (__predict_false((m = m_pullup(m, minthlen)) == NULL)) 3040 return (ENOMEM); 3041 ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen); 3042 } 3043 } else { 3044 txq->ift_pullups++; 3045 if (__predict_false((m = m_pullup(m, minthlen)) == NULL)) 3046 return (ENOMEM); 3047 ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen); 3048 if (m->m_len >= (ip->ip_hl << 2) + sizeof(*th)) 3049 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 3050 } 3051 } else { 3052 ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen); 3053 if (m->m_len >= (ip->ip_hl << 2) + sizeof(*th)) 3054 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 3055 } 3056 pi->ipi_ip_hlen = ip->ip_hl << 2; 3057 pi->ipi_ipproto = ip->ip_p; 3058 pi->ipi_flags |= IPI_TX_IPV4; 3059 3060 /* TCP checksum offload may require TCP header length */ 3061 if (IS_TX_OFFLOAD4(pi)) { 3062 if (__predict_true(pi->ipi_ipproto == IPPROTO_TCP)) { 3063 if (__predict_false(th == NULL)) { 3064 txq->ift_pullups++; 3065 if (__predict_false((m = m_pullup(m, (ip->ip_hl << 2) + sizeof(*th))) == NULL)) 3066 return (ENOMEM); 3067 th = (struct tcphdr *)((caddr_t)ip + pi->ipi_ip_hlen); 3068 } 3069 pi->ipi_tcp_hflags = th->th_flags; 3070 pi->ipi_tcp_hlen = th->th_off << 2; 3071 pi->ipi_tcp_seq = th->th_seq; 3072 } 3073 if (IS_TSO4(pi)) { 3074 if (__predict_false(ip->ip_p != IPPROTO_TCP)) 3075 return (ENXIO); 3076 /* 3077 * TSO always requires hardware checksum offload. 3078 */ 3079 pi->ipi_csum_flags |= (CSUM_IP_TCP | CSUM_IP); 3080 th->th_sum = in_pseudo(ip->ip_src.s_addr, 3081 ip->ip_dst.s_addr, htons(IPPROTO_TCP)); 3082 pi->ipi_tso_segsz = m->m_pkthdr.tso_segsz; 3083 if (sctx->isc_flags & IFLIB_TSO_INIT_IP) { 3084 ip->ip_sum = 0; 3085 ip->ip_len = htons(pi->ipi_ip_hlen + pi->ipi_tcp_hlen + pi->ipi_tso_segsz); 3086 } 3087 } 3088 } 3089 if ((sctx->isc_flags & IFLIB_NEED_ZERO_CSUM) && (pi->ipi_csum_flags & CSUM_IP)) 3090 ip->ip_sum = 0; 3091 3092 break; 3093 } 3094 #endif 3095 #ifdef INET6 3096 case ETHERTYPE_IPV6: 3097 { 3098 struct ip6_hdr *ip6 = (struct ip6_hdr *)(m->m_data + pi->ipi_ehdrlen); 3099 struct tcphdr *th; 3100 pi->ipi_ip_hlen = sizeof(struct ip6_hdr); 3101 3102 if (__predict_false(m->m_len < pi->ipi_ehdrlen + sizeof(struct ip6_hdr))) { 3103 txq->ift_pullups++; 3104 if (__predict_false((m = m_pullup(m, pi->ipi_ehdrlen + sizeof(struct ip6_hdr))) == NULL)) 3105 return (ENOMEM); 3106 } 3107 th = (struct tcphdr *)((caddr_t)ip6 + pi->ipi_ip_hlen); 3108 3109 /* XXX-BZ this will go badly in case of ext hdrs. */ 3110 pi->ipi_ipproto = ip6->ip6_nxt; 3111 pi->ipi_flags |= IPI_TX_IPV6; 3112 3113 /* TCP checksum offload may require TCP header length */ 3114 if (IS_TX_OFFLOAD6(pi)) { 3115 if (pi->ipi_ipproto == IPPROTO_TCP) { 3116 if (__predict_false(m->m_len < pi->ipi_ehdrlen + sizeof(struct ip6_hdr) + sizeof(struct tcphdr))) { 3117 txq->ift_pullups++; 3118 if (__predict_false((m = m_pullup(m, pi->ipi_ehdrlen + sizeof(struct ip6_hdr) + sizeof(struct tcphdr))) == NULL)) 3119 return (ENOMEM); 3120 } 3121 pi->ipi_tcp_hflags = th->th_flags; 3122 pi->ipi_tcp_hlen = th->th_off << 2; 3123 pi->ipi_tcp_seq = th->th_seq; 3124 } 3125 if (IS_TSO6(pi)) { 3126 if (__predict_false(ip6->ip6_nxt != IPPROTO_TCP)) 3127 return (ENXIO); 3128 /* 3129 * TSO always requires hardware checksum offload. 3130 */ 3131 pi->ipi_csum_flags |= CSUM_IP6_TCP; 3132 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0); 3133 pi->ipi_tso_segsz = m->m_pkthdr.tso_segsz; 3134 } 3135 } 3136 break; 3137 } 3138 #endif 3139 default: 3140 pi->ipi_csum_flags &= ~CSUM_OFFLOAD; 3141 pi->ipi_ip_hlen = 0; 3142 break; 3143 } 3144 *mp = m; 3145 3146 return (0); 3147 } 3148 3149 /* 3150 * If dodgy hardware rejects the scatter gather chain we've handed it 3151 * we'll need to remove the mbuf chain from ifsg_m[] before we can add the 3152 * m_defrag'd mbufs 3153 */ 3154 static __noinline struct mbuf * 3155 iflib_remove_mbuf(iflib_txq_t txq) 3156 { 3157 int ntxd, pidx; 3158 struct mbuf *m, **ifsd_m; 3159 3160 ifsd_m = txq->ift_sds.ifsd_m; 3161 ntxd = txq->ift_size; 3162 pidx = txq->ift_pidx & (ntxd - 1); 3163 ifsd_m = txq->ift_sds.ifsd_m; 3164 m = ifsd_m[pidx]; 3165 ifsd_m[pidx] = NULL; 3166 bus_dmamap_unload(txq->ift_buf_tag, txq->ift_sds.ifsd_map[pidx]); 3167 if (txq->ift_sds.ifsd_tso_map != NULL) 3168 bus_dmamap_unload(txq->ift_tso_buf_tag, 3169 txq->ift_sds.ifsd_tso_map[pidx]); 3170 #if MEMORY_LOGGING 3171 txq->ift_dequeued++; 3172 #endif 3173 return (m); 3174 } 3175 3176 static inline caddr_t 3177 calc_next_txd(iflib_txq_t txq, int cidx, uint8_t qid) 3178 { 3179 qidx_t size; 3180 int ntxd; 3181 caddr_t start, end, cur, next; 3182 3183 ntxd = txq->ift_size; 3184 size = txq->ift_txd_size[qid]; 3185 start = txq->ift_ifdi[qid].idi_vaddr; 3186 3187 if (__predict_false(size == 0)) 3188 return (start); 3189 cur = start + size*cidx; 3190 end = start + size*ntxd; 3191 next = CACHE_PTR_NEXT(cur); 3192 return (next < end ? next : start); 3193 } 3194 3195 /* 3196 * Pad an mbuf to ensure a minimum ethernet frame size. 3197 * min_frame_size is the frame size (less CRC) to pad the mbuf to 3198 */ 3199 static __noinline int 3200 iflib_ether_pad(device_t dev, struct mbuf **m_head, uint16_t min_frame_size) 3201 { 3202 /* 3203 * 18 is enough bytes to pad an ARP packet to 46 bytes, and 3204 * and ARP message is the smallest common payload I can think of 3205 */ 3206 static char pad[18]; /* just zeros */ 3207 int n; 3208 struct mbuf *new_head; 3209 3210 if (!M_WRITABLE(*m_head)) { 3211 new_head = m_dup(*m_head, M_NOWAIT); 3212 if (new_head == NULL) { 3213 m_freem(*m_head); 3214 device_printf(dev, "cannot pad short frame, m_dup() failed"); 3215 DBG_COUNTER_INC(encap_pad_mbuf_fail); 3216 DBG_COUNTER_INC(tx_frees); 3217 return ENOMEM; 3218 } 3219 m_freem(*m_head); 3220 *m_head = new_head; 3221 } 3222 3223 for (n = min_frame_size - (*m_head)->m_pkthdr.len; 3224 n > 0; n -= sizeof(pad)) 3225 if (!m_append(*m_head, min(n, sizeof(pad)), pad)) 3226 break; 3227 3228 if (n > 0) { 3229 m_freem(*m_head); 3230 device_printf(dev, "cannot pad short frame\n"); 3231 DBG_COUNTER_INC(encap_pad_mbuf_fail); 3232 DBG_COUNTER_INC(tx_frees); 3233 return (ENOBUFS); 3234 } 3235 3236 return 0; 3237 } 3238 3239 static int 3240 iflib_encap(iflib_txq_t txq, struct mbuf **m_headp) 3241 { 3242 if_ctx_t ctx; 3243 if_shared_ctx_t sctx; 3244 if_softc_ctx_t scctx; 3245 bus_dma_tag_t buf_tag; 3246 bus_dma_segment_t *segs; 3247 struct mbuf *m_head, **ifsd_m; 3248 void *next_txd; 3249 bus_dmamap_t map; 3250 struct if_pkt_info pi; 3251 int remap = 0; 3252 int err, nsegs, ndesc, max_segs, pidx, cidx, next, ntxd; 3253 3254 ctx = txq->ift_ctx; 3255 sctx = ctx->ifc_sctx; 3256 scctx = &ctx->ifc_softc_ctx; 3257 segs = txq->ift_segs; 3258 ntxd = txq->ift_size; 3259 m_head = *m_headp; 3260 map = NULL; 3261 3262 /* 3263 * If we're doing TSO the next descriptor to clean may be quite far ahead 3264 */ 3265 cidx = txq->ift_cidx; 3266 pidx = txq->ift_pidx; 3267 if (ctx->ifc_flags & IFC_PREFETCH) { 3268 next = (cidx + CACHE_PTR_INCREMENT) & (ntxd-1); 3269 if (!(ctx->ifc_flags & IFLIB_HAS_TXCQ)) { 3270 next_txd = calc_next_txd(txq, cidx, 0); 3271 prefetch(next_txd); 3272 } 3273 3274 /* prefetch the next cache line of mbuf pointers and flags */ 3275 prefetch(&txq->ift_sds.ifsd_m[next]); 3276 prefetch(&txq->ift_sds.ifsd_map[next]); 3277 next = (cidx + CACHE_LINE_SIZE) & (ntxd-1); 3278 } 3279 map = txq->ift_sds.ifsd_map[pidx]; 3280 ifsd_m = txq->ift_sds.ifsd_m; 3281 3282 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) { 3283 buf_tag = txq->ift_tso_buf_tag; 3284 max_segs = scctx->isc_tx_tso_segments_max; 3285 map = txq->ift_sds.ifsd_tso_map[pidx]; 3286 MPASS(buf_tag != NULL); 3287 MPASS(max_segs > 0); 3288 } else { 3289 buf_tag = txq->ift_buf_tag; 3290 max_segs = scctx->isc_tx_nsegments; 3291 map = txq->ift_sds.ifsd_map[pidx]; 3292 } 3293 if ((sctx->isc_flags & IFLIB_NEED_ETHER_PAD) && 3294 __predict_false(m_head->m_pkthdr.len < scctx->isc_min_frame_size)) { 3295 err = iflib_ether_pad(ctx->ifc_dev, m_headp, scctx->isc_min_frame_size); 3296 if (err) { 3297 DBG_COUNTER_INC(encap_txd_encap_fail); 3298 return err; 3299 } 3300 } 3301 m_head = *m_headp; 3302 3303 pkt_info_zero(&pi); 3304 pi.ipi_mflags = (m_head->m_flags & (M_VLANTAG|M_BCAST|M_MCAST)); 3305 pi.ipi_pidx = pidx; 3306 pi.ipi_qsidx = txq->ift_id; 3307 pi.ipi_len = m_head->m_pkthdr.len; 3308 pi.ipi_csum_flags = m_head->m_pkthdr.csum_flags; 3309 pi.ipi_vtag = M_HAS_VLANTAG(m_head) ? m_head->m_pkthdr.ether_vtag : 0; 3310 3311 /* deliberate bitwise OR to make one condition */ 3312 if (__predict_true((pi.ipi_csum_flags | pi.ipi_vtag))) { 3313 if (__predict_false((err = iflib_parse_header(txq, &pi, m_headp)) != 0)) { 3314 DBG_COUNTER_INC(encap_txd_encap_fail); 3315 return (err); 3316 } 3317 m_head = *m_headp; 3318 } 3319 3320 retry: 3321 err = bus_dmamap_load_mbuf_sg(buf_tag, map, m_head, segs, &nsegs, 3322 BUS_DMA_NOWAIT); 3323 defrag: 3324 if (__predict_false(err)) { 3325 switch (err) { 3326 case EFBIG: 3327 /* try collapse once and defrag once */ 3328 if (remap == 0) { 3329 m_head = m_collapse(*m_headp, M_NOWAIT, max_segs); 3330 /* try defrag if collapsing fails */ 3331 if (m_head == NULL) 3332 remap++; 3333 } 3334 if (remap == 1) { 3335 txq->ift_mbuf_defrag++; 3336 m_head = m_defrag(*m_headp, M_NOWAIT); 3337 } 3338 /* 3339 * remap should never be >1 unless bus_dmamap_load_mbuf_sg 3340 * failed to map an mbuf that was run through m_defrag 3341 */ 3342 MPASS(remap <= 1); 3343 if (__predict_false(m_head == NULL || remap > 1)) 3344 goto defrag_failed; 3345 remap++; 3346 *m_headp = m_head; 3347 goto retry; 3348 break; 3349 case ENOMEM: 3350 txq->ift_no_tx_dma_setup++; 3351 break; 3352 default: 3353 txq->ift_no_tx_dma_setup++; 3354 m_freem(*m_headp); 3355 DBG_COUNTER_INC(tx_frees); 3356 *m_headp = NULL; 3357 break; 3358 } 3359 txq->ift_map_failed++; 3360 DBG_COUNTER_INC(encap_load_mbuf_fail); 3361 DBG_COUNTER_INC(encap_txd_encap_fail); 3362 return (err); 3363 } 3364 ifsd_m[pidx] = m_head; 3365 /* 3366 * XXX assumes a 1 to 1 relationship between segments and 3367 * descriptors - this does not hold true on all drivers, e.g. 3368 * cxgb 3369 */ 3370 if (__predict_false(nsegs + 2 > TXQ_AVAIL(txq))) { 3371 txq->ift_no_desc_avail++; 3372 bus_dmamap_unload(buf_tag, map); 3373 DBG_COUNTER_INC(encap_txq_avail_fail); 3374 DBG_COUNTER_INC(encap_txd_encap_fail); 3375 if ((txq->ift_task.gt_task.ta_flags & TASK_ENQUEUED) == 0) 3376 GROUPTASK_ENQUEUE(&txq->ift_task); 3377 return (ENOBUFS); 3378 } 3379 /* 3380 * On Intel cards we can greatly reduce the number of TX interrupts 3381 * we see by only setting report status on every Nth descriptor. 3382 * However, this also means that the driver will need to keep track 3383 * of the descriptors that RS was set on to check them for the DD bit. 3384 */ 3385 txq->ift_rs_pending += nsegs + 1; 3386 if (txq->ift_rs_pending > TXQ_MAX_RS_DEFERRED(txq) || 3387 iflib_no_tx_batch || (TXQ_AVAIL(txq) - nsegs) <= MAX_TX_DESC(ctx) + 2) { 3388 pi.ipi_flags |= IPI_TX_INTR; 3389 txq->ift_rs_pending = 0; 3390 } 3391 3392 pi.ipi_segs = segs; 3393 pi.ipi_nsegs = nsegs; 3394 3395 MPASS(pidx >= 0 && pidx < txq->ift_size); 3396 #ifdef PKT_DEBUG 3397 print_pkt(&pi); 3398 #endif 3399 if ((err = ctx->isc_txd_encap(ctx->ifc_softc, &pi)) == 0) { 3400 bus_dmamap_sync(buf_tag, map, BUS_DMASYNC_PREWRITE); 3401 DBG_COUNTER_INC(tx_encap); 3402 MPASS(pi.ipi_new_pidx < txq->ift_size); 3403 3404 ndesc = pi.ipi_new_pidx - pi.ipi_pidx; 3405 if (pi.ipi_new_pidx < pi.ipi_pidx) { 3406 ndesc += txq->ift_size; 3407 txq->ift_gen = 1; 3408 } 3409 /* 3410 * drivers can need as many as 3411 * two sentinels 3412 */ 3413 MPASS(ndesc <= pi.ipi_nsegs + 2); 3414 MPASS(pi.ipi_new_pidx != pidx); 3415 MPASS(ndesc > 0); 3416 txq->ift_in_use += ndesc; 3417 3418 /* 3419 * We update the last software descriptor again here because there may 3420 * be a sentinel and/or there may be more mbufs than segments 3421 */ 3422 txq->ift_pidx = pi.ipi_new_pidx; 3423 txq->ift_npending += pi.ipi_ndescs; 3424 } else { 3425 *m_headp = m_head = iflib_remove_mbuf(txq); 3426 if (err == EFBIG) { 3427 txq->ift_txd_encap_efbig++; 3428 if (remap < 2) { 3429 remap = 1; 3430 goto defrag; 3431 } 3432 } 3433 goto defrag_failed; 3434 } 3435 /* 3436 * err can't possibly be non-zero here, so we don't neet to test it 3437 * to see if we need to DBG_COUNTER_INC(encap_txd_encap_fail). 3438 */ 3439 return (err); 3440 3441 defrag_failed: 3442 txq->ift_mbuf_defrag_failed++; 3443 txq->ift_map_failed++; 3444 m_freem(*m_headp); 3445 DBG_COUNTER_INC(tx_frees); 3446 *m_headp = NULL; 3447 DBG_COUNTER_INC(encap_txd_encap_fail); 3448 return (ENOMEM); 3449 } 3450 3451 static void 3452 iflib_tx_desc_free(iflib_txq_t txq, int n) 3453 { 3454 uint32_t qsize, cidx, mask, gen; 3455 struct mbuf *m, **ifsd_m; 3456 bool do_prefetch; 3457 3458 cidx = txq->ift_cidx; 3459 gen = txq->ift_gen; 3460 qsize = txq->ift_size; 3461 mask = qsize-1; 3462 ifsd_m = txq->ift_sds.ifsd_m; 3463 do_prefetch = (txq->ift_ctx->ifc_flags & IFC_PREFETCH); 3464 3465 while (n-- > 0) { 3466 if (do_prefetch) { 3467 prefetch(ifsd_m[(cidx + 3) & mask]); 3468 prefetch(ifsd_m[(cidx + 4) & mask]); 3469 } 3470 if ((m = ifsd_m[cidx]) != NULL) { 3471 prefetch(&ifsd_m[(cidx + CACHE_PTR_INCREMENT) & mask]); 3472 if (m->m_pkthdr.csum_flags & CSUM_TSO) { 3473 bus_dmamap_sync(txq->ift_tso_buf_tag, 3474 txq->ift_sds.ifsd_tso_map[cidx], 3475 BUS_DMASYNC_POSTWRITE); 3476 bus_dmamap_unload(txq->ift_tso_buf_tag, 3477 txq->ift_sds.ifsd_tso_map[cidx]); 3478 } else { 3479 bus_dmamap_sync(txq->ift_buf_tag, 3480 txq->ift_sds.ifsd_map[cidx], 3481 BUS_DMASYNC_POSTWRITE); 3482 bus_dmamap_unload(txq->ift_buf_tag, 3483 txq->ift_sds.ifsd_map[cidx]); 3484 } 3485 /* XXX we don't support any drivers that batch packets yet */ 3486 MPASS(m->m_nextpkt == NULL); 3487 m_freem(m); 3488 ifsd_m[cidx] = NULL; 3489 #if MEMORY_LOGGING 3490 txq->ift_dequeued++; 3491 #endif 3492 DBG_COUNTER_INC(tx_frees); 3493 } 3494 if (__predict_false(++cidx == qsize)) { 3495 cidx = 0; 3496 gen = 0; 3497 } 3498 } 3499 txq->ift_cidx = cidx; 3500 txq->ift_gen = gen; 3501 } 3502 3503 static __inline int 3504 iflib_completed_tx_reclaim(iflib_txq_t txq, int thresh) 3505 { 3506 int reclaim; 3507 if_ctx_t ctx = txq->ift_ctx; 3508 3509 KASSERT(thresh >= 0, ("invalid threshold to reclaim")); 3510 MPASS(thresh /*+ MAX_TX_DESC(txq->ift_ctx) */ < txq->ift_size); 3511 3512 /* 3513 * Need a rate-limiting check so that this isn't called every time 3514 */ 3515 iflib_tx_credits_update(ctx, txq); 3516 reclaim = DESC_RECLAIMABLE(txq); 3517 3518 if (reclaim <= thresh /* + MAX_TX_DESC(txq->ift_ctx) */) { 3519 #ifdef INVARIANTS 3520 if (iflib_verbose_debug) { 3521 printf("%s processed=%ju cleaned=%ju tx_nsegments=%d reclaim=%d thresh=%d\n", __FUNCTION__, 3522 txq->ift_processed, txq->ift_cleaned, txq->ift_ctx->ifc_softc_ctx.isc_tx_nsegments, 3523 reclaim, thresh); 3524 3525 } 3526 #endif 3527 return (0); 3528 } 3529 iflib_tx_desc_free(txq, reclaim); 3530 txq->ift_cleaned += reclaim; 3531 txq->ift_in_use -= reclaim; 3532 3533 return (reclaim); 3534 } 3535 3536 static struct mbuf ** 3537 _ring_peek_one(struct ifmp_ring *r, int cidx, int offset, int remaining) 3538 { 3539 int next, size; 3540 struct mbuf **items; 3541 3542 size = r->size; 3543 next = (cidx + CACHE_PTR_INCREMENT) & (size-1); 3544 items = __DEVOLATILE(struct mbuf **, &r->items[0]); 3545 3546 prefetch(items[(cidx + offset) & (size-1)]); 3547 if (remaining > 1) { 3548 prefetch2cachelines(&items[next]); 3549 prefetch2cachelines(items[(cidx + offset + 1) & (size-1)]); 3550 prefetch2cachelines(items[(cidx + offset + 2) & (size-1)]); 3551 prefetch2cachelines(items[(cidx + offset + 3) & (size-1)]); 3552 } 3553 return (__DEVOLATILE(struct mbuf **, &r->items[(cidx + offset) & (size-1)])); 3554 } 3555 3556 static void 3557 iflib_txq_check_drain(iflib_txq_t txq, int budget) 3558 { 3559 3560 ifmp_ring_check_drainage(txq->ift_br, budget); 3561 } 3562 3563 static uint32_t 3564 iflib_txq_can_drain(struct ifmp_ring *r) 3565 { 3566 iflib_txq_t txq = r->cookie; 3567 if_ctx_t ctx = txq->ift_ctx; 3568 3569 if (TXQ_AVAIL(txq) > MAX_TX_DESC(ctx) + 2) 3570 return (1); 3571 bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map, 3572 BUS_DMASYNC_POSTREAD); 3573 return (ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, 3574 false)); 3575 } 3576 3577 static uint32_t 3578 iflib_txq_drain(struct ifmp_ring *r, uint32_t cidx, uint32_t pidx) 3579 { 3580 iflib_txq_t txq = r->cookie; 3581 if_ctx_t ctx = txq->ift_ctx; 3582 if_t ifp = ctx->ifc_ifp; 3583 struct mbuf **mp, *m; 3584 int i, count, consumed, pkt_sent, bytes_sent, mcast_sent, avail; 3585 int reclaimed, err, in_use_prev, desc_used; 3586 bool do_prefetch, ring, rang; 3587 3588 if (__predict_false(!(if_getdrvflags(ifp) & IFF_DRV_RUNNING) || 3589 !LINK_ACTIVE(ctx))) { 3590 DBG_COUNTER_INC(txq_drain_notready); 3591 return (0); 3592 } 3593 reclaimed = iflib_completed_tx_reclaim(txq, RECLAIM_THRESH(ctx)); 3594 rang = iflib_txd_db_check(ctx, txq, reclaimed, txq->ift_in_use); 3595 avail = IDXDIFF(pidx, cidx, r->size); 3596 if (__predict_false(ctx->ifc_flags & IFC_QFLUSH)) { 3597 DBG_COUNTER_INC(txq_drain_flushing); 3598 for (i = 0; i < avail; i++) { 3599 if (__predict_true(r->items[(cidx + i) & (r->size-1)] != (void *)txq)) 3600 m_free(r->items[(cidx + i) & (r->size-1)]); 3601 r->items[(cidx + i) & (r->size-1)] = NULL; 3602 } 3603 return (avail); 3604 } 3605 3606 if (__predict_false(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_OACTIVE)) { 3607 txq->ift_qstatus = IFLIB_QUEUE_IDLE; 3608 CALLOUT_LOCK(txq); 3609 callout_stop(&txq->ift_timer); 3610 CALLOUT_UNLOCK(txq); 3611 DBG_COUNTER_INC(txq_drain_oactive); 3612 return (0); 3613 } 3614 if (reclaimed) 3615 txq->ift_qstatus = IFLIB_QUEUE_IDLE; 3616 consumed = mcast_sent = bytes_sent = pkt_sent = 0; 3617 count = MIN(avail, TX_BATCH_SIZE); 3618 #ifdef INVARIANTS 3619 if (iflib_verbose_debug) 3620 printf("%s avail=%d ifc_flags=%x txq_avail=%d ", __FUNCTION__, 3621 avail, ctx->ifc_flags, TXQ_AVAIL(txq)); 3622 #endif 3623 do_prefetch = (ctx->ifc_flags & IFC_PREFETCH); 3624 avail = TXQ_AVAIL(txq); 3625 err = 0; 3626 for (desc_used = i = 0; i < count && avail > MAX_TX_DESC(ctx) + 2; i++) { 3627 int rem = do_prefetch ? count - i : 0; 3628 3629 mp = _ring_peek_one(r, cidx, i, rem); 3630 MPASS(mp != NULL && *mp != NULL); 3631 if (__predict_false(*mp == (struct mbuf *)txq)) { 3632 consumed++; 3633 reclaimed++; 3634 continue; 3635 } 3636 in_use_prev = txq->ift_in_use; 3637 err = iflib_encap(txq, mp); 3638 if (__predict_false(err)) { 3639 /* no room - bail out */ 3640 if (err == ENOBUFS) 3641 break; 3642 consumed++; 3643 /* we can't send this packet - skip it */ 3644 continue; 3645 } 3646 consumed++; 3647 pkt_sent++; 3648 m = *mp; 3649 DBG_COUNTER_INC(tx_sent); 3650 bytes_sent += m->m_pkthdr.len; 3651 mcast_sent += !!(m->m_flags & M_MCAST); 3652 avail = TXQ_AVAIL(txq); 3653 3654 txq->ift_db_pending += (txq->ift_in_use - in_use_prev); 3655 desc_used += (txq->ift_in_use - in_use_prev); 3656 ETHER_BPF_MTAP(ifp, m); 3657 if (__predict_false(!(ifp->if_drv_flags & IFF_DRV_RUNNING))) 3658 break; 3659 rang = iflib_txd_db_check(ctx, txq, false, in_use_prev); 3660 } 3661 3662 /* deliberate use of bitwise or to avoid gratuitous short-circuit */ 3663 ring = rang ? false : (iflib_min_tx_latency | err) || (TXQ_AVAIL(txq) < MAX_TX_DESC(ctx)); 3664 iflib_txd_db_check(ctx, txq, ring, txq->ift_in_use); 3665 if_inc_counter(ifp, IFCOUNTER_OBYTES, bytes_sent); 3666 if_inc_counter(ifp, IFCOUNTER_OPACKETS, pkt_sent); 3667 if (mcast_sent) 3668 if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast_sent); 3669 #ifdef INVARIANTS 3670 if (iflib_verbose_debug) 3671 printf("consumed=%d\n", consumed); 3672 #endif 3673 return (consumed); 3674 } 3675 3676 static uint32_t 3677 iflib_txq_drain_always(struct ifmp_ring *r) 3678 { 3679 return (1); 3680 } 3681 3682 static uint32_t 3683 iflib_txq_drain_free(struct ifmp_ring *r, uint32_t cidx, uint32_t pidx) 3684 { 3685 int i, avail; 3686 struct mbuf **mp; 3687 iflib_txq_t txq; 3688 3689 txq = r->cookie; 3690 3691 txq->ift_qstatus = IFLIB_QUEUE_IDLE; 3692 CALLOUT_LOCK(txq); 3693 callout_stop(&txq->ift_timer); 3694 CALLOUT_UNLOCK(txq); 3695 3696 avail = IDXDIFF(pidx, cidx, r->size); 3697 for (i = 0; i < avail; i++) { 3698 mp = _ring_peek_one(r, cidx, i, avail - i); 3699 if (__predict_false(*mp == (struct mbuf *)txq)) 3700 continue; 3701 m_freem(*mp); 3702 DBG_COUNTER_INC(tx_frees); 3703 } 3704 MPASS(ifmp_ring_is_stalled(r) == 0); 3705 return (avail); 3706 } 3707 3708 static void 3709 iflib_ifmp_purge(iflib_txq_t txq) 3710 { 3711 struct ifmp_ring *r; 3712 3713 r = txq->ift_br; 3714 r->drain = iflib_txq_drain_free; 3715 r->can_drain = iflib_txq_drain_always; 3716 3717 ifmp_ring_check_drainage(r, r->size); 3718 3719 r->drain = iflib_txq_drain; 3720 r->can_drain = iflib_txq_can_drain; 3721 } 3722 3723 static void 3724 _task_fn_tx(void *context) 3725 { 3726 iflib_txq_t txq = context; 3727 if_ctx_t ctx = txq->ift_ctx; 3728 #if defined(ALTQ) || defined(DEV_NETMAP) 3729 if_t ifp = ctx->ifc_ifp; 3730 #endif 3731 int abdicate = ctx->ifc_sysctl_tx_abdicate; 3732 3733 #ifdef IFLIB_DIAGNOSTICS 3734 txq->ift_cpu_exec_count[curcpu]++; 3735 #endif 3736 if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)) 3737 return; 3738 #ifdef DEV_NETMAP 3739 if (if_getcapenable(ifp) & IFCAP_NETMAP) { 3740 bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map, 3741 BUS_DMASYNC_POSTREAD); 3742 if (ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, false)) 3743 netmap_tx_irq(ifp, txq->ift_id); 3744 if (ctx->ifc_flags & IFC_LEGACY) 3745 IFDI_INTR_ENABLE(ctx); 3746 else 3747 IFDI_TX_QUEUE_INTR_ENABLE(ctx, txq->ift_id); 3748 return; 3749 } 3750 #endif 3751 #ifdef ALTQ 3752 if (ALTQ_IS_ENABLED(&ifp->if_snd)) 3753 iflib_altq_if_start(ifp); 3754 #endif 3755 if (txq->ift_db_pending) 3756 ifmp_ring_enqueue(txq->ift_br, (void **)&txq, 1, TX_BATCH_SIZE, abdicate); 3757 else if (!abdicate) 3758 ifmp_ring_check_drainage(txq->ift_br, TX_BATCH_SIZE); 3759 /* 3760 * When abdicating, we always need to check drainage, not just when we don't enqueue 3761 */ 3762 if (abdicate) 3763 ifmp_ring_check_drainage(txq->ift_br, TX_BATCH_SIZE); 3764 if (ctx->ifc_flags & IFC_LEGACY) 3765 IFDI_INTR_ENABLE(ctx); 3766 else 3767 IFDI_TX_QUEUE_INTR_ENABLE(ctx, txq->ift_id); 3768 } 3769 3770 static void 3771 _task_fn_rx(void *context) 3772 { 3773 iflib_rxq_t rxq = context; 3774 if_ctx_t ctx = rxq->ifr_ctx; 3775 bool more; 3776 uint16_t budget; 3777 3778 #ifdef IFLIB_DIAGNOSTICS 3779 rxq->ifr_cpu_exec_count[curcpu]++; 3780 #endif 3781 DBG_COUNTER_INC(task_fn_rxs); 3782 if (__predict_false(!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING))) 3783 return; 3784 more = true; 3785 #ifdef DEV_NETMAP 3786 if (if_getcapenable(ctx->ifc_ifp) & IFCAP_NETMAP) { 3787 u_int work = 0; 3788 if (netmap_rx_irq(ctx->ifc_ifp, rxq->ifr_id, &work)) { 3789 more = false; 3790 } 3791 } 3792 #endif 3793 budget = ctx->ifc_sysctl_rx_budget; 3794 if (budget == 0) 3795 budget = 16; /* XXX */ 3796 if (more == false || (more = iflib_rxeof(rxq, budget)) == false) { 3797 if (ctx->ifc_flags & IFC_LEGACY) 3798 IFDI_INTR_ENABLE(ctx); 3799 else 3800 IFDI_RX_QUEUE_INTR_ENABLE(ctx, rxq->ifr_id); 3801 DBG_COUNTER_INC(rx_intr_enables); 3802 } 3803 if (__predict_false(!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING))) 3804 return; 3805 if (more) 3806 GROUPTASK_ENQUEUE(&rxq->ifr_task); 3807 } 3808 3809 static void 3810 _task_fn_admin(void *context) 3811 { 3812 if_ctx_t ctx = context; 3813 if_softc_ctx_t sctx = &ctx->ifc_softc_ctx; 3814 iflib_txq_t txq; 3815 int i; 3816 bool oactive, running, do_reset, do_watchdog, in_detach; 3817 uint32_t reset_on = hz / 2; 3818 3819 STATE_LOCK(ctx); 3820 running = (if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING); 3821 oactive = (if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_OACTIVE); 3822 do_reset = (ctx->ifc_flags & IFC_DO_RESET); 3823 do_watchdog = (ctx->ifc_flags & IFC_DO_WATCHDOG); 3824 in_detach = (ctx->ifc_flags & IFC_IN_DETACH); 3825 ctx->ifc_flags &= ~(IFC_DO_RESET|IFC_DO_WATCHDOG); 3826 STATE_UNLOCK(ctx); 3827 3828 if ((!running && !oactive) && !(ctx->ifc_sctx->isc_flags & IFLIB_ADMIN_ALWAYS_RUN)) 3829 return; 3830 if (in_detach) 3831 return; 3832 3833 CTX_LOCK(ctx); 3834 for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++) { 3835 CALLOUT_LOCK(txq); 3836 callout_stop(&txq->ift_timer); 3837 CALLOUT_UNLOCK(txq); 3838 } 3839 if (do_watchdog) { 3840 ctx->ifc_watchdog_events++; 3841 IFDI_WATCHDOG_RESET(ctx); 3842 } 3843 IFDI_UPDATE_ADMIN_STATUS(ctx); 3844 for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++) { 3845 #ifdef DEV_NETMAP 3846 reset_on = hz / 2; 3847 if (if_getcapenable(ctx->ifc_ifp) & IFCAP_NETMAP) 3848 iflib_netmap_timer_adjust(ctx, txq, &reset_on); 3849 #endif 3850 callout_reset_on(&txq->ift_timer, reset_on, iflib_timer, txq, txq->ift_timer.c_cpu); 3851 } 3852 IFDI_LINK_INTR_ENABLE(ctx); 3853 if (do_reset) 3854 iflib_if_init_locked(ctx); 3855 CTX_UNLOCK(ctx); 3856 3857 if (LINK_ACTIVE(ctx) == 0) 3858 return; 3859 for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++) 3860 iflib_txq_check_drain(txq, IFLIB_RESTART_BUDGET); 3861 } 3862 3863 3864 static void 3865 _task_fn_iov(void *context) 3866 { 3867 if_ctx_t ctx = context; 3868 3869 if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING) && 3870 !(ctx->ifc_sctx->isc_flags & IFLIB_ADMIN_ALWAYS_RUN)) 3871 return; 3872 3873 CTX_LOCK(ctx); 3874 IFDI_VFLR_HANDLE(ctx); 3875 CTX_UNLOCK(ctx); 3876 } 3877 3878 static int 3879 iflib_sysctl_int_delay(SYSCTL_HANDLER_ARGS) 3880 { 3881 int err; 3882 if_int_delay_info_t info; 3883 if_ctx_t ctx; 3884 3885 info = (if_int_delay_info_t)arg1; 3886 ctx = info->iidi_ctx; 3887 info->iidi_req = req; 3888 info->iidi_oidp = oidp; 3889 CTX_LOCK(ctx); 3890 err = IFDI_SYSCTL_INT_DELAY(ctx, info); 3891 CTX_UNLOCK(ctx); 3892 return (err); 3893 } 3894 3895 /********************************************************************* 3896 * 3897 * IFNET FUNCTIONS 3898 * 3899 **********************************************************************/ 3900 3901 static void 3902 iflib_if_init_locked(if_ctx_t ctx) 3903 { 3904 iflib_stop(ctx); 3905 iflib_init_locked(ctx); 3906 } 3907 3908 3909 static void 3910 iflib_if_init(void *arg) 3911 { 3912 if_ctx_t ctx = arg; 3913 3914 CTX_LOCK(ctx); 3915 iflib_if_init_locked(ctx); 3916 CTX_UNLOCK(ctx); 3917 } 3918 3919 static int 3920 iflib_if_transmit(if_t ifp, struct mbuf *m) 3921 { 3922 if_ctx_t ctx = if_getsoftc(ifp); 3923 3924 iflib_txq_t txq; 3925 int err, qidx; 3926 int abdicate = ctx->ifc_sysctl_tx_abdicate; 3927 3928 if (__predict_false((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || !LINK_ACTIVE(ctx))) { 3929 DBG_COUNTER_INC(tx_frees); 3930 m_freem(m); 3931 return (ENETDOWN); 3932 } 3933 3934 MPASS(m->m_nextpkt == NULL); 3935 /* ALTQ-enabled interfaces always use queue 0. */ 3936 qidx = 0; 3937 if ((NTXQSETS(ctx) > 1) && M_HASHTYPE_GET(m) && !ALTQ_IS_ENABLED(&ifp->if_snd)) 3938 qidx = QIDX(ctx, m); 3939 /* 3940 * XXX calculate buf_ring based on flowid (divvy up bits?) 3941 */ 3942 txq = &ctx->ifc_txqs[qidx]; 3943 3944 #ifdef DRIVER_BACKPRESSURE 3945 if (txq->ift_closed) { 3946 while (m != NULL) { 3947 next = m->m_nextpkt; 3948 m->m_nextpkt = NULL; 3949 m_freem(m); 3950 DBG_COUNTER_INC(tx_frees); 3951 m = next; 3952 } 3953 return (ENOBUFS); 3954 } 3955 #endif 3956 #ifdef notyet 3957 qidx = count = 0; 3958 mp = marr; 3959 next = m; 3960 do { 3961 count++; 3962 next = next->m_nextpkt; 3963 } while (next != NULL); 3964 3965 if (count > nitems(marr)) 3966 if ((mp = malloc(count*sizeof(struct mbuf *), M_IFLIB, M_NOWAIT)) == NULL) { 3967 /* XXX check nextpkt */ 3968 m_freem(m); 3969 /* XXX simplify for now */ 3970 DBG_COUNTER_INC(tx_frees); 3971 return (ENOBUFS); 3972 } 3973 for (next = m, i = 0; next != NULL; i++) { 3974 mp[i] = next; 3975 next = next->m_nextpkt; 3976 mp[i]->m_nextpkt = NULL; 3977 } 3978 #endif 3979 DBG_COUNTER_INC(tx_seen); 3980 err = ifmp_ring_enqueue(txq->ift_br, (void **)&m, 1, TX_BATCH_SIZE, abdicate); 3981 3982 if (abdicate) 3983 GROUPTASK_ENQUEUE(&txq->ift_task); 3984 if (err) { 3985 if (!abdicate) 3986 GROUPTASK_ENQUEUE(&txq->ift_task); 3987 /* support forthcoming later */ 3988 #ifdef DRIVER_BACKPRESSURE 3989 txq->ift_closed = TRUE; 3990 #endif 3991 ifmp_ring_check_drainage(txq->ift_br, TX_BATCH_SIZE); 3992 m_freem(m); 3993 DBG_COUNTER_INC(tx_frees); 3994 } 3995 3996 return (err); 3997 } 3998 3999 #ifdef ALTQ 4000 /* 4001 * The overall approach to integrating iflib with ALTQ is to continue to use 4002 * the iflib mp_ring machinery between the ALTQ queue(s) and the hardware 4003 * ring. Technically, when using ALTQ, queueing to an intermediate mp_ring 4004 * is redundant/unnecessary, but doing so minimizes the amount of 4005 * ALTQ-specific code required in iflib. It is assumed that the overhead of 4006 * redundantly queueing to an intermediate mp_ring is swamped by the 4007 * performance limitations inherent in using ALTQ. 4008 * 4009 * When ALTQ support is compiled in, all iflib drivers will use a transmit 4010 * routine, iflib_altq_if_transmit(), that checks if ALTQ is enabled for the 4011 * given interface. If ALTQ is enabled for an interface, then all 4012 * transmitted packets for that interface will be submitted to the ALTQ 4013 * subsystem via IFQ_ENQUEUE(). We don't use the legacy if_transmit() 4014 * implementation because it uses IFQ_HANDOFF(), which will duplicatively 4015 * update stats that the iflib machinery handles, and which is sensitve to 4016 * the disused IFF_DRV_OACTIVE flag. Additionally, iflib_altq_if_start() 4017 * will be installed as the start routine for use by ALTQ facilities that 4018 * need to trigger queue drains on a scheduled basis. 4019 * 4020 */ 4021 static void 4022 iflib_altq_if_start(if_t ifp) 4023 { 4024 struct ifaltq *ifq = &ifp->if_snd; 4025 struct mbuf *m; 4026 4027 IFQ_LOCK(ifq); 4028 IFQ_DEQUEUE_NOLOCK(ifq, m); 4029 while (m != NULL) { 4030 iflib_if_transmit(ifp, m); 4031 IFQ_DEQUEUE_NOLOCK(ifq, m); 4032 } 4033 IFQ_UNLOCK(ifq); 4034 } 4035 4036 static int 4037 iflib_altq_if_transmit(if_t ifp, struct mbuf *m) 4038 { 4039 int err; 4040 4041 if (ALTQ_IS_ENABLED(&ifp->if_snd)) { 4042 IFQ_ENQUEUE(&ifp->if_snd, m, err); 4043 if (err == 0) 4044 iflib_altq_if_start(ifp); 4045 } else 4046 err = iflib_if_transmit(ifp, m); 4047 4048 return (err); 4049 } 4050 #endif /* ALTQ */ 4051 4052 static void 4053 iflib_if_qflush(if_t ifp) 4054 { 4055 if_ctx_t ctx = if_getsoftc(ifp); 4056 iflib_txq_t txq = ctx->ifc_txqs; 4057 int i; 4058 4059 STATE_LOCK(ctx); 4060 ctx->ifc_flags |= IFC_QFLUSH; 4061 STATE_UNLOCK(ctx); 4062 for (i = 0; i < NTXQSETS(ctx); i++, txq++) 4063 while (!(ifmp_ring_is_idle(txq->ift_br) || ifmp_ring_is_stalled(txq->ift_br))) 4064 iflib_txq_check_drain(txq, 0); 4065 STATE_LOCK(ctx); 4066 ctx->ifc_flags &= ~IFC_QFLUSH; 4067 STATE_UNLOCK(ctx); 4068 4069 /* 4070 * When ALTQ is enabled, this will also take care of purging the 4071 * ALTQ queue(s). 4072 */ 4073 if_qflush(ifp); 4074 } 4075 4076 4077 #define IFCAP_FLAGS (IFCAP_HWCSUM_IPV6 | IFCAP_HWCSUM | IFCAP_LRO | \ 4078 IFCAP_TSO | IFCAP_VLAN_HWTAGGING | IFCAP_HWSTATS | \ 4079 IFCAP_VLAN_MTU | IFCAP_VLAN_HWFILTER | \ 4080 IFCAP_VLAN_HWTSO | IFCAP_VLAN_HWCSUM) 4081 4082 static int 4083 iflib_if_ioctl(if_t ifp, u_long command, caddr_t data) 4084 { 4085 if_ctx_t ctx = if_getsoftc(ifp); 4086 struct ifreq *ifr = (struct ifreq *)data; 4087 #if defined(INET) || defined(INET6) 4088 struct ifaddr *ifa = (struct ifaddr *)data; 4089 #endif 4090 bool avoid_reset = false; 4091 int err = 0, reinit = 0, bits; 4092 4093 switch (command) { 4094 case SIOCSIFADDR: 4095 #ifdef INET 4096 if (ifa->ifa_addr->sa_family == AF_INET) 4097 avoid_reset = true; 4098 #endif 4099 #ifdef INET6 4100 if (ifa->ifa_addr->sa_family == AF_INET6) 4101 avoid_reset = true; 4102 #endif 4103 /* 4104 ** Calling init results in link renegotiation, 4105 ** so we avoid doing it when possible. 4106 */ 4107 if (avoid_reset) { 4108 if_setflagbits(ifp, IFF_UP,0); 4109 if (!(if_getdrvflags(ifp) & IFF_DRV_RUNNING)) 4110 reinit = 1; 4111 #ifdef INET 4112 if (!(if_getflags(ifp) & IFF_NOARP)) 4113 arp_ifinit(ifp, ifa); 4114 #endif 4115 } else 4116 err = ether_ioctl(ifp, command, data); 4117 break; 4118 case SIOCSIFMTU: 4119 CTX_LOCK(ctx); 4120 if (ifr->ifr_mtu == if_getmtu(ifp)) { 4121 CTX_UNLOCK(ctx); 4122 break; 4123 } 4124 bits = if_getdrvflags(ifp); 4125 /* stop the driver and free any clusters before proceeding */ 4126 iflib_stop(ctx); 4127 4128 if ((err = IFDI_MTU_SET(ctx, ifr->ifr_mtu)) == 0) { 4129 STATE_LOCK(ctx); 4130 if (ifr->ifr_mtu > ctx->ifc_max_fl_buf_size) 4131 ctx->ifc_flags |= IFC_MULTISEG; 4132 else 4133 ctx->ifc_flags &= ~IFC_MULTISEG; 4134 STATE_UNLOCK(ctx); 4135 err = if_setmtu(ifp, ifr->ifr_mtu); 4136 } 4137 iflib_init_locked(ctx); 4138 STATE_LOCK(ctx); 4139 if_setdrvflags(ifp, bits); 4140 STATE_UNLOCK(ctx); 4141 CTX_UNLOCK(ctx); 4142 break; 4143 case SIOCSIFFLAGS: 4144 CTX_LOCK(ctx); 4145 if (if_getflags(ifp) & IFF_UP) { 4146 if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) { 4147 if ((if_getflags(ifp) ^ ctx->ifc_if_flags) & 4148 (IFF_PROMISC | IFF_ALLMULTI)) { 4149 err = IFDI_PROMISC_SET(ctx, if_getflags(ifp)); 4150 } 4151 } else 4152 reinit = 1; 4153 } else if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) { 4154 iflib_stop(ctx); 4155 } 4156 ctx->ifc_if_flags = if_getflags(ifp); 4157 CTX_UNLOCK(ctx); 4158 break; 4159 case SIOCADDMULTI: 4160 case SIOCDELMULTI: 4161 if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) { 4162 CTX_LOCK(ctx); 4163 IFDI_INTR_DISABLE(ctx); 4164 IFDI_MULTI_SET(ctx); 4165 IFDI_INTR_ENABLE(ctx); 4166 CTX_UNLOCK(ctx); 4167 } 4168 break; 4169 case SIOCSIFMEDIA: 4170 CTX_LOCK(ctx); 4171 IFDI_MEDIA_SET(ctx); 4172 CTX_UNLOCK(ctx); 4173 /* FALLTHROUGH */ 4174 case SIOCGIFMEDIA: 4175 case SIOCGIFXMEDIA: 4176 err = ifmedia_ioctl(ifp, ifr, ctx->ifc_mediap, command); 4177 break; 4178 case SIOCGI2C: 4179 { 4180 struct ifi2creq i2c; 4181 4182 err = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c)); 4183 if (err != 0) 4184 break; 4185 if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) { 4186 err = EINVAL; 4187 break; 4188 } 4189 if (i2c.len > sizeof(i2c.data)) { 4190 err = EINVAL; 4191 break; 4192 } 4193 4194 if ((err = IFDI_I2C_REQ(ctx, &i2c)) == 0) 4195 err = copyout(&i2c, ifr_data_get_ptr(ifr), 4196 sizeof(i2c)); 4197 break; 4198 } 4199 case SIOCSIFCAP: 4200 { 4201 int mask, setmask, oldmask; 4202 4203 oldmask = if_getcapenable(ifp); 4204 mask = ifr->ifr_reqcap ^ oldmask; 4205 mask &= ctx->ifc_softc_ctx.isc_capabilities; 4206 setmask = 0; 4207 #ifdef TCP_OFFLOAD 4208 setmask |= mask & (IFCAP_TOE4|IFCAP_TOE6); 4209 #endif 4210 setmask |= (mask & IFCAP_FLAGS); 4211 setmask |= (mask & IFCAP_WOL); 4212 4213 /* 4214 * If any RX csum has changed, change all the ones that 4215 * are supported by the driver. 4216 */ 4217 if (setmask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) { 4218 setmask |= ctx->ifc_softc_ctx.isc_capabilities & 4219 (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6); 4220 } 4221 4222 /* 4223 * want to ensure that traffic has stopped before we change any of the flags 4224 */ 4225 if (setmask) { 4226 CTX_LOCK(ctx); 4227 bits = if_getdrvflags(ifp); 4228 if (bits & IFF_DRV_RUNNING && setmask & ~IFCAP_WOL) 4229 iflib_stop(ctx); 4230 STATE_LOCK(ctx); 4231 if_togglecapenable(ifp, setmask); 4232 STATE_UNLOCK(ctx); 4233 if (bits & IFF_DRV_RUNNING && setmask & ~IFCAP_WOL) 4234 iflib_init_locked(ctx); 4235 STATE_LOCK(ctx); 4236 if_setdrvflags(ifp, bits); 4237 STATE_UNLOCK(ctx); 4238 CTX_UNLOCK(ctx); 4239 } 4240 if_vlancap(ifp); 4241 break; 4242 } 4243 case SIOCGPRIVATE_0: 4244 case SIOCSDRVSPEC: 4245 case SIOCGDRVSPEC: 4246 CTX_LOCK(ctx); 4247 err = IFDI_PRIV_IOCTL(ctx, command, data); 4248 CTX_UNLOCK(ctx); 4249 break; 4250 default: 4251 err = ether_ioctl(ifp, command, data); 4252 break; 4253 } 4254 if (reinit) 4255 iflib_if_init(ctx); 4256 return (err); 4257 } 4258 4259 static uint64_t 4260 iflib_if_get_counter(if_t ifp, ift_counter cnt) 4261 { 4262 if_ctx_t ctx = if_getsoftc(ifp); 4263 4264 return (IFDI_GET_COUNTER(ctx, cnt)); 4265 } 4266 4267 /********************************************************************* 4268 * 4269 * OTHER FUNCTIONS EXPORTED TO THE STACK 4270 * 4271 **********************************************************************/ 4272 4273 static void 4274 iflib_vlan_register(void *arg, if_t ifp, uint16_t vtag) 4275 { 4276 if_ctx_t ctx = if_getsoftc(ifp); 4277 4278 if ((void *)ctx != arg) 4279 return; 4280 4281 if ((vtag == 0) || (vtag > 4095)) 4282 return; 4283 4284 CTX_LOCK(ctx); 4285 IFDI_VLAN_REGISTER(ctx, vtag); 4286 /* Re-init to load the changes */ 4287 if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) 4288 iflib_if_init_locked(ctx); 4289 CTX_UNLOCK(ctx); 4290 } 4291 4292 static void 4293 iflib_vlan_unregister(void *arg, if_t ifp, uint16_t vtag) 4294 { 4295 if_ctx_t ctx = if_getsoftc(ifp); 4296 4297 if ((void *)ctx != arg) 4298 return; 4299 4300 if ((vtag == 0) || (vtag > 4095)) 4301 return; 4302 4303 CTX_LOCK(ctx); 4304 IFDI_VLAN_UNREGISTER(ctx, vtag); 4305 /* Re-init to load the changes */ 4306 if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) 4307 iflib_if_init_locked(ctx); 4308 CTX_UNLOCK(ctx); 4309 } 4310 4311 static void 4312 iflib_led_func(void *arg, int onoff) 4313 { 4314 if_ctx_t ctx = arg; 4315 4316 CTX_LOCK(ctx); 4317 IFDI_LED_FUNC(ctx, onoff); 4318 CTX_UNLOCK(ctx); 4319 } 4320 4321 /********************************************************************* 4322 * 4323 * BUS FUNCTION DEFINITIONS 4324 * 4325 **********************************************************************/ 4326 4327 int 4328 iflib_device_probe(device_t dev) 4329 { 4330 const pci_vendor_info_t *ent; 4331 if_shared_ctx_t sctx; 4332 uint16_t pci_device_id, pci_rev_id, pci_subdevice_id, pci_subvendor_id; 4333 uint16_t pci_vendor_id; 4334 4335 if ((sctx = DEVICE_REGISTER(dev)) == NULL || sctx->isc_magic != IFLIB_MAGIC) 4336 return (ENOTSUP); 4337 4338 pci_vendor_id = pci_get_vendor(dev); 4339 pci_device_id = pci_get_device(dev); 4340 pci_subvendor_id = pci_get_subvendor(dev); 4341 pci_subdevice_id = pci_get_subdevice(dev); 4342 pci_rev_id = pci_get_revid(dev); 4343 if (sctx->isc_parse_devinfo != NULL) 4344 sctx->isc_parse_devinfo(&pci_device_id, &pci_subvendor_id, &pci_subdevice_id, &pci_rev_id); 4345 4346 ent = sctx->isc_vendor_info; 4347 while (ent->pvi_vendor_id != 0) { 4348 if (pci_vendor_id != ent->pvi_vendor_id) { 4349 ent++; 4350 continue; 4351 } 4352 if ((pci_device_id == ent->pvi_device_id) && 4353 ((pci_subvendor_id == ent->pvi_subvendor_id) || 4354 (ent->pvi_subvendor_id == 0)) && 4355 ((pci_subdevice_id == ent->pvi_subdevice_id) || 4356 (ent->pvi_subdevice_id == 0)) && 4357 ((pci_rev_id == ent->pvi_rev_id) || 4358 (ent->pvi_rev_id == 0))) { 4359 4360 device_set_desc_copy(dev, ent->pvi_name); 4361 /* this needs to be changed to zero if the bus probing code 4362 * ever stops re-probing on best match because the sctx 4363 * may have its values over written by register calls 4364 * in subsequent probes 4365 */ 4366 return (BUS_PROBE_DEFAULT); 4367 } 4368 ent++; 4369 } 4370 return (ENXIO); 4371 } 4372 4373 int 4374 iflib_device_probe_vendor(device_t dev) 4375 { 4376 int probe; 4377 4378 probe = iflib_device_probe(dev); 4379 if (probe == BUS_PROBE_DEFAULT) 4380 return (BUS_PROBE_VENDOR); 4381 else 4382 return (probe); 4383 } 4384 4385 static void 4386 iflib_reset_qvalues(if_ctx_t ctx) 4387 { 4388 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 4389 if_shared_ctx_t sctx = ctx->ifc_sctx; 4390 device_t dev = ctx->ifc_dev; 4391 int i; 4392 4393 if (ctx->ifc_sysctl_ntxqs != 0) 4394 scctx->isc_ntxqsets = ctx->ifc_sysctl_ntxqs; 4395 if (ctx->ifc_sysctl_nrxqs != 0) 4396 scctx->isc_nrxqsets = ctx->ifc_sysctl_nrxqs; 4397 4398 for (i = 0; i < sctx->isc_ntxqs; i++) { 4399 if (ctx->ifc_sysctl_ntxds[i] != 0) 4400 scctx->isc_ntxd[i] = ctx->ifc_sysctl_ntxds[i]; 4401 else 4402 scctx->isc_ntxd[i] = sctx->isc_ntxd_default[i]; 4403 } 4404 4405 for (i = 0; i < sctx->isc_nrxqs; i++) { 4406 if (ctx->ifc_sysctl_nrxds[i] != 0) 4407 scctx->isc_nrxd[i] = ctx->ifc_sysctl_nrxds[i]; 4408 else 4409 scctx->isc_nrxd[i] = sctx->isc_nrxd_default[i]; 4410 } 4411 4412 for (i = 0; i < sctx->isc_nrxqs; i++) { 4413 if (scctx->isc_nrxd[i] < sctx->isc_nrxd_min[i]) { 4414 device_printf(dev, "nrxd%d: %d less than nrxd_min %d - resetting to min\n", 4415 i, scctx->isc_nrxd[i], sctx->isc_nrxd_min[i]); 4416 scctx->isc_nrxd[i] = sctx->isc_nrxd_min[i]; 4417 } 4418 if (scctx->isc_nrxd[i] > sctx->isc_nrxd_max[i]) { 4419 device_printf(dev, "nrxd%d: %d greater than nrxd_max %d - resetting to max\n", 4420 i, scctx->isc_nrxd[i], sctx->isc_nrxd_max[i]); 4421 scctx->isc_nrxd[i] = sctx->isc_nrxd_max[i]; 4422 } 4423 if (!powerof2(scctx->isc_nrxd[i])) { 4424 device_printf(dev, "nrxd%d: %d is not a power of 2 - using default value of %d\n", 4425 i, scctx->isc_nrxd[i], sctx->isc_nrxd_default[i]); 4426 scctx->isc_nrxd[i] = sctx->isc_nrxd_default[i]; 4427 } 4428 } 4429 4430 for (i = 0; i < sctx->isc_ntxqs; i++) { 4431 if (scctx->isc_ntxd[i] < sctx->isc_ntxd_min[i]) { 4432 device_printf(dev, "ntxd%d: %d less than ntxd_min %d - resetting to min\n", 4433 i, scctx->isc_ntxd[i], sctx->isc_ntxd_min[i]); 4434 scctx->isc_ntxd[i] = sctx->isc_ntxd_min[i]; 4435 } 4436 if (scctx->isc_ntxd[i] > sctx->isc_ntxd_max[i]) { 4437 device_printf(dev, "ntxd%d: %d greater than ntxd_max %d - resetting to max\n", 4438 i, scctx->isc_ntxd[i], sctx->isc_ntxd_max[i]); 4439 scctx->isc_ntxd[i] = sctx->isc_ntxd_max[i]; 4440 } 4441 if (!powerof2(scctx->isc_ntxd[i])) { 4442 device_printf(dev, "ntxd%d: %d is not a power of 2 - using default value of %d\n", 4443 i, scctx->isc_ntxd[i], sctx->isc_ntxd_default[i]); 4444 scctx->isc_ntxd[i] = sctx->isc_ntxd_default[i]; 4445 } 4446 } 4447 } 4448 4449 static void 4450 iflib_add_pfil(if_ctx_t ctx) 4451 { 4452 struct pfil_head *pfil; 4453 struct pfil_head_args pa; 4454 iflib_rxq_t rxq; 4455 int i; 4456 4457 pa.pa_version = PFIL_VERSION; 4458 pa.pa_flags = PFIL_IN; 4459 pa.pa_type = PFIL_TYPE_ETHERNET; 4460 pa.pa_headname = ctx->ifc_ifp->if_xname; 4461 pfil = pfil_head_register(&pa); 4462 4463 for (i = 0, rxq = ctx->ifc_rxqs; i < NRXQSETS(ctx); i++, rxq++) { 4464 rxq->pfil = pfil; 4465 } 4466 } 4467 4468 static void 4469 iflib_rem_pfil(if_ctx_t ctx) 4470 { 4471 struct pfil_head *pfil; 4472 iflib_rxq_t rxq; 4473 int i; 4474 4475 rxq = ctx->ifc_rxqs; 4476 pfil = rxq->pfil; 4477 for (i = 0; i < NRXQSETS(ctx); i++, rxq++) { 4478 rxq->pfil = NULL; 4479 } 4480 pfil_head_unregister(pfil); 4481 } 4482 4483 static uint16_t 4484 get_ctx_core_offset(if_ctx_t ctx) 4485 { 4486 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 4487 struct cpu_offset *op; 4488 uint16_t qc; 4489 uint16_t ret = ctx->ifc_sysctl_core_offset; 4490 4491 if (ret != CORE_OFFSET_UNSPECIFIED) 4492 return (ret); 4493 4494 if (ctx->ifc_sysctl_separate_txrx) 4495 qc = scctx->isc_ntxqsets + scctx->isc_nrxqsets; 4496 else 4497 qc = max(scctx->isc_ntxqsets, scctx->isc_nrxqsets); 4498 4499 mtx_lock(&cpu_offset_mtx); 4500 SLIST_FOREACH(op, &cpu_offsets, entries) { 4501 if (CPU_CMP(&ctx->ifc_cpus, &op->set) == 0) { 4502 ret = op->offset; 4503 op->offset += qc; 4504 MPASS(op->refcount < UINT_MAX); 4505 op->refcount++; 4506 break; 4507 } 4508 } 4509 if (ret == CORE_OFFSET_UNSPECIFIED) { 4510 ret = 0; 4511 op = malloc(sizeof(struct cpu_offset), M_IFLIB, 4512 M_NOWAIT | M_ZERO); 4513 if (op == NULL) { 4514 device_printf(ctx->ifc_dev, 4515 "allocation for cpu offset failed.\n"); 4516 } else { 4517 op->offset = qc; 4518 op->refcount = 1; 4519 CPU_COPY(&ctx->ifc_cpus, &op->set); 4520 SLIST_INSERT_HEAD(&cpu_offsets, op, entries); 4521 } 4522 } 4523 mtx_unlock(&cpu_offset_mtx); 4524 4525 return (ret); 4526 } 4527 4528 static void 4529 unref_ctx_core_offset(if_ctx_t ctx) 4530 { 4531 struct cpu_offset *op, *top; 4532 4533 mtx_lock(&cpu_offset_mtx); 4534 SLIST_FOREACH_SAFE(op, &cpu_offsets, entries, top) { 4535 if (CPU_CMP(&ctx->ifc_cpus, &op->set) == 0) { 4536 MPASS(op->refcount > 0); 4537 op->refcount--; 4538 if (op->refcount == 0) { 4539 SLIST_REMOVE(&cpu_offsets, op, cpu_offset, entries); 4540 free(op, M_IFLIB); 4541 } 4542 break; 4543 } 4544 } 4545 mtx_unlock(&cpu_offset_mtx); 4546 } 4547 4548 int 4549 iflib_device_register(device_t dev, void *sc, if_shared_ctx_t sctx, if_ctx_t *ctxp) 4550 { 4551 if_ctx_t ctx; 4552 if_t ifp; 4553 if_softc_ctx_t scctx; 4554 kobjop_desc_t kobj_desc; 4555 kobj_method_t *kobj_method; 4556 int err, msix, rid; 4557 uint16_t main_rxq, main_txq; 4558 4559 ctx = malloc(sizeof(* ctx), M_IFLIB, M_WAITOK|M_ZERO); 4560 4561 if (sc == NULL) { 4562 sc = malloc(sctx->isc_driver->size, M_IFLIB, M_WAITOK|M_ZERO); 4563 device_set_softc(dev, ctx); 4564 ctx->ifc_flags |= IFC_SC_ALLOCATED; 4565 } 4566 4567 ctx->ifc_sctx = sctx; 4568 ctx->ifc_dev = dev; 4569 ctx->ifc_softc = sc; 4570 4571 if ((err = iflib_register(ctx)) != 0) { 4572 device_printf(dev, "iflib_register failed %d\n", err); 4573 goto fail_ctx_free; 4574 } 4575 iflib_add_device_sysctl_pre(ctx); 4576 4577 scctx = &ctx->ifc_softc_ctx; 4578 ifp = ctx->ifc_ifp; 4579 4580 iflib_reset_qvalues(ctx); 4581 CTX_LOCK(ctx); 4582 if ((err = IFDI_ATTACH_PRE(ctx)) != 0) { 4583 device_printf(dev, "IFDI_ATTACH_PRE failed %d\n", err); 4584 goto fail_unlock; 4585 } 4586 _iflib_pre_assert(scctx); 4587 ctx->ifc_txrx = *scctx->isc_txrx; 4588 4589 if (sctx->isc_flags & IFLIB_DRIVER_MEDIA) 4590 ctx->ifc_mediap = scctx->isc_media; 4591 4592 #ifdef INVARIANTS 4593 if (scctx->isc_capabilities & IFCAP_TXCSUM) 4594 MPASS(scctx->isc_tx_csum_flags); 4595 #endif 4596 4597 if_setcapabilities(ifp, scctx->isc_capabilities | IFCAP_HWSTATS); 4598 if_setcapenable(ifp, scctx->isc_capenable | IFCAP_HWSTATS); 4599 4600 if (scctx->isc_ntxqsets == 0 || (scctx->isc_ntxqsets_max && scctx->isc_ntxqsets_max < scctx->isc_ntxqsets)) 4601 scctx->isc_ntxqsets = scctx->isc_ntxqsets_max; 4602 if (scctx->isc_nrxqsets == 0 || (scctx->isc_nrxqsets_max && scctx->isc_nrxqsets_max < scctx->isc_nrxqsets)) 4603 scctx->isc_nrxqsets = scctx->isc_nrxqsets_max; 4604 4605 main_txq = (sctx->isc_flags & IFLIB_HAS_TXCQ) ? 1 : 0; 4606 main_rxq = (sctx->isc_flags & IFLIB_HAS_RXCQ) ? 1 : 0; 4607 4608 /* XXX change for per-queue sizes */ 4609 device_printf(dev, "Using %d TX descriptors and %d RX descriptors\n", 4610 scctx->isc_ntxd[main_txq], scctx->isc_nrxd[main_rxq]); 4611 4612 if (scctx->isc_tx_nsegments > scctx->isc_ntxd[main_txq] / 4613 MAX_SINGLE_PACKET_FRACTION) 4614 scctx->isc_tx_nsegments = max(1, scctx->isc_ntxd[main_txq] / 4615 MAX_SINGLE_PACKET_FRACTION); 4616 if (scctx->isc_tx_tso_segments_max > scctx->isc_ntxd[main_txq] / 4617 MAX_SINGLE_PACKET_FRACTION) 4618 scctx->isc_tx_tso_segments_max = max(1, 4619 scctx->isc_ntxd[main_txq] / MAX_SINGLE_PACKET_FRACTION); 4620 4621 /* TSO parameters - dig these out of the data sheet - simply correspond to tag setup */ 4622 if (if_getcapabilities(ifp) & IFCAP_TSO) { 4623 /* 4624 * The stack can't handle a TSO size larger than IP_MAXPACKET, 4625 * but some MACs do. 4626 */ 4627 if_sethwtsomax(ifp, min(scctx->isc_tx_tso_size_max, 4628 IP_MAXPACKET)); 4629 /* 4630 * Take maximum number of m_pullup(9)'s in iflib_parse_header() 4631 * into account. In the worst case, each of these calls will 4632 * add another mbuf and, thus, the requirement for another DMA 4633 * segment. So for best performance, it doesn't make sense to 4634 * advertize a maximum of TSO segments that typically will 4635 * require defragmentation in iflib_encap(). 4636 */ 4637 if_sethwtsomaxsegcount(ifp, scctx->isc_tx_tso_segments_max - 3); 4638 if_sethwtsomaxsegsize(ifp, scctx->isc_tx_tso_segsize_max); 4639 } 4640 if (scctx->isc_rss_table_size == 0) 4641 scctx->isc_rss_table_size = 64; 4642 scctx->isc_rss_table_mask = scctx->isc_rss_table_size-1; 4643 4644 GROUPTASK_INIT(&ctx->ifc_admin_task, 0, _task_fn_admin, ctx); 4645 /* XXX format name */ 4646 taskqgroup_attach(qgroup_if_config_tqg, &ctx->ifc_admin_task, ctx, 4647 NULL, NULL, "admin"); 4648 4649 /* Set up cpu set. If it fails, use the set of all CPUs. */ 4650 if (bus_get_cpus(dev, INTR_CPUS, sizeof(ctx->ifc_cpus), &ctx->ifc_cpus) != 0) { 4651 device_printf(dev, "Unable to fetch CPU list\n"); 4652 CPU_COPY(&all_cpus, &ctx->ifc_cpus); 4653 } 4654 MPASS(CPU_COUNT(&ctx->ifc_cpus) > 0); 4655 4656 /* 4657 ** Now set up MSI or MSI-X, should return us the number of supported 4658 ** vectors (will be 1 for a legacy interrupt and MSI). 4659 */ 4660 if (sctx->isc_flags & IFLIB_SKIP_MSIX) { 4661 msix = scctx->isc_vectors; 4662 } else if (scctx->isc_msix_bar != 0) 4663 /* 4664 * The simple fact that isc_msix_bar is not 0 does not mean we 4665 * we have a good value there that is known to work. 4666 */ 4667 msix = iflib_msix_init(ctx); 4668 else { 4669 scctx->isc_vectors = 1; 4670 scctx->isc_ntxqsets = 1; 4671 scctx->isc_nrxqsets = 1; 4672 scctx->isc_intr = IFLIB_INTR_LEGACY; 4673 msix = 0; 4674 } 4675 /* Get memory for the station queues */ 4676 if ((err = iflib_queues_alloc(ctx))) { 4677 device_printf(dev, "Unable to allocate queue memory\n"); 4678 goto fail_intr_free; 4679 } 4680 4681 if ((err = iflib_qset_structures_setup(ctx))) 4682 goto fail_queues; 4683 4684 /* 4685 * Now that we know how many queues there are, get the core offset. 4686 */ 4687 ctx->ifc_sysctl_core_offset = get_ctx_core_offset(ctx); 4688 4689 /* 4690 * Group taskqueues aren't properly set up until SMP is started, 4691 * so we disable interrupts until we can handle them post 4692 * SI_SUB_SMP. 4693 * 4694 * XXX: disabling interrupts doesn't actually work, at least for 4695 * the non-MSI case. When they occur before SI_SUB_SMP completes, 4696 * we do null handling and depend on this not causing too large an 4697 * interrupt storm. 4698 */ 4699 IFDI_INTR_DISABLE(ctx); 4700 4701 if (msix > 1) { 4702 /* 4703 * When using MSI-X, ensure that ifdi_{r,t}x_queue_intr_enable 4704 * aren't the default NULL implementation. 4705 */ 4706 kobj_desc = &ifdi_rx_queue_intr_enable_desc; 4707 kobj_method = kobj_lookup_method(((kobj_t)ctx)->ops->cls, NULL, 4708 kobj_desc); 4709 if (kobj_method == &kobj_desc->deflt) { 4710 device_printf(dev, 4711 "MSI-X requires ifdi_rx_queue_intr_enable method"); 4712 err = EOPNOTSUPP; 4713 goto fail_queues; 4714 } 4715 kobj_desc = &ifdi_tx_queue_intr_enable_desc; 4716 kobj_method = kobj_lookup_method(((kobj_t)ctx)->ops->cls, NULL, 4717 kobj_desc); 4718 if (kobj_method == &kobj_desc->deflt) { 4719 device_printf(dev, 4720 "MSI-X requires ifdi_tx_queue_intr_enable method"); 4721 err = EOPNOTSUPP; 4722 goto fail_queues; 4723 } 4724 4725 /* 4726 * Assign the MSI-X vectors. 4727 * Note that the default NULL ifdi_msix_intr_assign method will 4728 * fail here, too. 4729 */ 4730 err = IFDI_MSIX_INTR_ASSIGN(ctx, msix); 4731 if (err != 0) { 4732 device_printf(dev, "IFDI_MSIX_INTR_ASSIGN failed %d\n", 4733 err); 4734 goto fail_queues; 4735 } 4736 } else { 4737 rid = 0; 4738 if (scctx->isc_intr == IFLIB_INTR_MSI) { 4739 MPASS(msix == 1); 4740 rid = 1; 4741 } 4742 if ((err = iflib_legacy_setup(ctx, ctx->isc_legacy_intr, ctx->ifc_softc, &rid, "irq0")) != 0) { 4743 device_printf(dev, "iflib_legacy_setup failed %d\n", err); 4744 goto fail_queues; 4745 } 4746 } 4747 4748 ether_ifattach(ctx->ifc_ifp, ctx->ifc_mac.octet); 4749 4750 if ((err = IFDI_ATTACH_POST(ctx)) != 0) { 4751 device_printf(dev, "IFDI_ATTACH_POST failed %d\n", err); 4752 goto fail_detach; 4753 } 4754 4755 /* 4756 * Tell the upper layer(s) if IFCAP_VLAN_MTU is supported. 4757 * This must appear after the call to ether_ifattach() because 4758 * ether_ifattach() sets if_hdrlen to the default value. 4759 */ 4760 if (if_getcapabilities(ifp) & IFCAP_VLAN_MTU) 4761 if_setifheaderlen(ifp, sizeof(struct ether_vlan_header)); 4762 4763 if ((err = iflib_netmap_attach(ctx))) { 4764 device_printf(ctx->ifc_dev, "netmap attach failed: %d\n", err); 4765 goto fail_detach; 4766 } 4767 *ctxp = ctx; 4768 4769 NETDUMP_SET(ctx->ifc_ifp, iflib); 4770 4771 if_setgetcounterfn(ctx->ifc_ifp, iflib_if_get_counter); 4772 iflib_add_device_sysctl_post(ctx); 4773 iflib_add_pfil(ctx); 4774 ctx->ifc_flags |= IFC_INIT_DONE; 4775 CTX_UNLOCK(ctx); 4776 4777 return (0); 4778 4779 fail_detach: 4780 ether_ifdetach(ctx->ifc_ifp); 4781 fail_intr_free: 4782 iflib_free_intr_mem(ctx); 4783 fail_queues: 4784 iflib_tx_structures_free(ctx); 4785 iflib_rx_structures_free(ctx); 4786 IFDI_DETACH(ctx); 4787 fail_unlock: 4788 CTX_UNLOCK(ctx); 4789 fail_ctx_free: 4790 if (ctx->ifc_flags & IFC_SC_ALLOCATED) 4791 free(ctx->ifc_softc, M_IFLIB); 4792 free(ctx, M_IFLIB); 4793 return (err); 4794 } 4795 4796 int 4797 iflib_pseudo_register(device_t dev, if_shared_ctx_t sctx, if_ctx_t *ctxp, 4798 struct iflib_cloneattach_ctx *clctx) 4799 { 4800 int err; 4801 if_ctx_t ctx; 4802 if_t ifp; 4803 if_softc_ctx_t scctx; 4804 int i; 4805 void *sc; 4806 uint16_t main_txq; 4807 uint16_t main_rxq; 4808 4809 ctx = malloc(sizeof(*ctx), M_IFLIB, M_WAITOK|M_ZERO); 4810 sc = malloc(sctx->isc_driver->size, M_IFLIB, M_WAITOK|M_ZERO); 4811 ctx->ifc_flags |= IFC_SC_ALLOCATED; 4812 if (sctx->isc_flags & (IFLIB_PSEUDO|IFLIB_VIRTUAL)) 4813 ctx->ifc_flags |= IFC_PSEUDO; 4814 4815 ctx->ifc_sctx = sctx; 4816 ctx->ifc_softc = sc; 4817 ctx->ifc_dev = dev; 4818 4819 if ((err = iflib_register(ctx)) != 0) { 4820 device_printf(dev, "%s: iflib_register failed %d\n", __func__, err); 4821 goto fail_ctx_free; 4822 } 4823 iflib_add_device_sysctl_pre(ctx); 4824 4825 scctx = &ctx->ifc_softc_ctx; 4826 ifp = ctx->ifc_ifp; 4827 4828 iflib_reset_qvalues(ctx); 4829 CTX_LOCK(ctx); 4830 if ((err = IFDI_ATTACH_PRE(ctx)) != 0) { 4831 device_printf(dev, "IFDI_ATTACH_PRE failed %d\n", err); 4832 goto fail_unlock; 4833 } 4834 if (sctx->isc_flags & IFLIB_GEN_MAC) 4835 ether_gen_addr(ifp, &ctx->ifc_mac); 4836 if ((err = IFDI_CLONEATTACH(ctx, clctx->cc_ifc, clctx->cc_name, 4837 clctx->cc_params)) != 0) { 4838 device_printf(dev, "IFDI_CLONEATTACH failed %d\n", err); 4839 goto fail_ctx_free; 4840 } 4841 ifmedia_add(ctx->ifc_mediap, IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL); 4842 ifmedia_add(ctx->ifc_mediap, IFM_ETHER | IFM_AUTO, 0, NULL); 4843 ifmedia_set(ctx->ifc_mediap, IFM_ETHER | IFM_AUTO); 4844 4845 #ifdef INVARIANTS 4846 if (scctx->isc_capabilities & IFCAP_TXCSUM) 4847 MPASS(scctx->isc_tx_csum_flags); 4848 #endif 4849 4850 if_setcapabilities(ifp, scctx->isc_capabilities | IFCAP_HWSTATS | IFCAP_LINKSTATE); 4851 if_setcapenable(ifp, scctx->isc_capenable | IFCAP_HWSTATS | IFCAP_LINKSTATE); 4852 4853 ifp->if_flags |= IFF_NOGROUP; 4854 if (sctx->isc_flags & IFLIB_PSEUDO) { 4855 ether_ifattach(ctx->ifc_ifp, ctx->ifc_mac.octet); 4856 4857 if ((err = IFDI_ATTACH_POST(ctx)) != 0) { 4858 device_printf(dev, "IFDI_ATTACH_POST failed %d\n", err); 4859 goto fail_detach; 4860 } 4861 *ctxp = ctx; 4862 4863 /* 4864 * Tell the upper layer(s) if IFCAP_VLAN_MTU is supported. 4865 * This must appear after the call to ether_ifattach() because 4866 * ether_ifattach() sets if_hdrlen to the default value. 4867 */ 4868 if (if_getcapabilities(ifp) & IFCAP_VLAN_MTU) 4869 if_setifheaderlen(ifp, 4870 sizeof(struct ether_vlan_header)); 4871 4872 if_setgetcounterfn(ctx->ifc_ifp, iflib_if_get_counter); 4873 iflib_add_device_sysctl_post(ctx); 4874 ctx->ifc_flags |= IFC_INIT_DONE; 4875 return (0); 4876 } 4877 _iflib_pre_assert(scctx); 4878 ctx->ifc_txrx = *scctx->isc_txrx; 4879 4880 if (scctx->isc_ntxqsets == 0 || (scctx->isc_ntxqsets_max && scctx->isc_ntxqsets_max < scctx->isc_ntxqsets)) 4881 scctx->isc_ntxqsets = scctx->isc_ntxqsets_max; 4882 if (scctx->isc_nrxqsets == 0 || (scctx->isc_nrxqsets_max && scctx->isc_nrxqsets_max < scctx->isc_nrxqsets)) 4883 scctx->isc_nrxqsets = scctx->isc_nrxqsets_max; 4884 4885 main_txq = (sctx->isc_flags & IFLIB_HAS_TXCQ) ? 1 : 0; 4886 main_rxq = (sctx->isc_flags & IFLIB_HAS_RXCQ) ? 1 : 0; 4887 4888 /* XXX change for per-queue sizes */ 4889 device_printf(dev, "Using %d TX descriptors and %d RX descriptors\n", 4890 scctx->isc_ntxd[main_txq], scctx->isc_nrxd[main_rxq]); 4891 4892 if (scctx->isc_tx_nsegments > scctx->isc_ntxd[main_txq] / 4893 MAX_SINGLE_PACKET_FRACTION) 4894 scctx->isc_tx_nsegments = max(1, scctx->isc_ntxd[main_txq] / 4895 MAX_SINGLE_PACKET_FRACTION); 4896 if (scctx->isc_tx_tso_segments_max > scctx->isc_ntxd[main_txq] / 4897 MAX_SINGLE_PACKET_FRACTION) 4898 scctx->isc_tx_tso_segments_max = max(1, 4899 scctx->isc_ntxd[main_txq] / MAX_SINGLE_PACKET_FRACTION); 4900 4901 /* TSO parameters - dig these out of the data sheet - simply correspond to tag setup */ 4902 if (if_getcapabilities(ifp) & IFCAP_TSO) { 4903 /* 4904 * The stack can't handle a TSO size larger than IP_MAXPACKET, 4905 * but some MACs do. 4906 */ 4907 if_sethwtsomax(ifp, min(scctx->isc_tx_tso_size_max, 4908 IP_MAXPACKET)); 4909 /* 4910 * Take maximum number of m_pullup(9)'s in iflib_parse_header() 4911 * into account. In the worst case, each of these calls will 4912 * add another mbuf and, thus, the requirement for another DMA 4913 * segment. So for best performance, it doesn't make sense to 4914 * advertize a maximum of TSO segments that typically will 4915 * require defragmentation in iflib_encap(). 4916 */ 4917 if_sethwtsomaxsegcount(ifp, scctx->isc_tx_tso_segments_max - 3); 4918 if_sethwtsomaxsegsize(ifp, scctx->isc_tx_tso_segsize_max); 4919 } 4920 if (scctx->isc_rss_table_size == 0) 4921 scctx->isc_rss_table_size = 64; 4922 scctx->isc_rss_table_mask = scctx->isc_rss_table_size-1; 4923 4924 GROUPTASK_INIT(&ctx->ifc_admin_task, 0, _task_fn_admin, ctx); 4925 /* XXX format name */ 4926 taskqgroup_attach(qgroup_if_config_tqg, &ctx->ifc_admin_task, ctx, 4927 NULL, NULL, "admin"); 4928 4929 /* XXX --- can support > 1 -- but keep it simple for now */ 4930 scctx->isc_intr = IFLIB_INTR_LEGACY; 4931 4932 /* Get memory for the station queues */ 4933 if ((err = iflib_queues_alloc(ctx))) { 4934 device_printf(dev, "Unable to allocate queue memory\n"); 4935 goto fail_iflib_detach; 4936 } 4937 4938 if ((err = iflib_qset_structures_setup(ctx))) { 4939 device_printf(dev, "qset structure setup failed %d\n", err); 4940 goto fail_queues; 4941 } 4942 4943 /* 4944 * XXX What if anything do we want to do about interrupts? 4945 */ 4946 ether_ifattach(ctx->ifc_ifp, ctx->ifc_mac.octet); 4947 if ((err = IFDI_ATTACH_POST(ctx)) != 0) { 4948 device_printf(dev, "IFDI_ATTACH_POST failed %d\n", err); 4949 goto fail_detach; 4950 } 4951 4952 /* 4953 * Tell the upper layer(s) if IFCAP_VLAN_MTU is supported. 4954 * This must appear after the call to ether_ifattach() because 4955 * ether_ifattach() sets if_hdrlen to the default value. 4956 */ 4957 if (if_getcapabilities(ifp) & IFCAP_VLAN_MTU) 4958 if_setifheaderlen(ifp, sizeof(struct ether_vlan_header)); 4959 4960 /* XXX handle more than one queue */ 4961 for (i = 0; i < scctx->isc_nrxqsets; i++) 4962 IFDI_RX_CLSET(ctx, 0, i, ctx->ifc_rxqs[i].ifr_fl[0].ifl_sds.ifsd_cl); 4963 4964 *ctxp = ctx; 4965 4966 if_setgetcounterfn(ctx->ifc_ifp, iflib_if_get_counter); 4967 iflib_add_device_sysctl_post(ctx); 4968 ctx->ifc_flags |= IFC_INIT_DONE; 4969 CTX_UNLOCK(ctx); 4970 4971 return (0); 4972 fail_detach: 4973 ether_ifdetach(ctx->ifc_ifp); 4974 fail_queues: 4975 iflib_tx_structures_free(ctx); 4976 iflib_rx_structures_free(ctx); 4977 fail_iflib_detach: 4978 IFDI_DETACH(ctx); 4979 fail_unlock: 4980 CTX_UNLOCK(ctx); 4981 fail_ctx_free: 4982 free(ctx->ifc_softc, M_IFLIB); 4983 free(ctx, M_IFLIB); 4984 return (err); 4985 } 4986 4987 int 4988 iflib_pseudo_deregister(if_ctx_t ctx) 4989 { 4990 if_t ifp = ctx->ifc_ifp; 4991 iflib_txq_t txq; 4992 iflib_rxq_t rxq; 4993 int i, j; 4994 struct taskqgroup *tqg; 4995 iflib_fl_t fl; 4996 4997 /* Unregister VLAN events */ 4998 if (ctx->ifc_vlan_attach_event != NULL) 4999 EVENTHANDLER_DEREGISTER(vlan_config, ctx->ifc_vlan_attach_event); 5000 if (ctx->ifc_vlan_detach_event != NULL) 5001 EVENTHANDLER_DEREGISTER(vlan_unconfig, ctx->ifc_vlan_detach_event); 5002 5003 ether_ifdetach(ifp); 5004 /* ether_ifdetach calls if_qflush - lock must be destroy afterwards*/ 5005 CTX_LOCK_DESTROY(ctx); 5006 /* XXX drain any dependent tasks */ 5007 tqg = qgroup_if_io_tqg; 5008 for (txq = ctx->ifc_txqs, i = 0; i < NTXQSETS(ctx); i++, txq++) { 5009 callout_drain(&txq->ift_timer); 5010 if (txq->ift_task.gt_uniq != NULL) 5011 taskqgroup_detach(tqg, &txq->ift_task); 5012 } 5013 for (i = 0, rxq = ctx->ifc_rxqs; i < NRXQSETS(ctx); i++, rxq++) { 5014 if (rxq->ifr_task.gt_uniq != NULL) 5015 taskqgroup_detach(tqg, &rxq->ifr_task); 5016 5017 for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) 5018 free(fl->ifl_rx_bitmap, M_IFLIB); 5019 } 5020 tqg = qgroup_if_config_tqg; 5021 if (ctx->ifc_admin_task.gt_uniq != NULL) 5022 taskqgroup_detach(tqg, &ctx->ifc_admin_task); 5023 if (ctx->ifc_vflr_task.gt_uniq != NULL) 5024 taskqgroup_detach(tqg, &ctx->ifc_vflr_task); 5025 5026 if_free(ifp); 5027 5028 iflib_tx_structures_free(ctx); 5029 iflib_rx_structures_free(ctx); 5030 if (ctx->ifc_flags & IFC_SC_ALLOCATED) 5031 free(ctx->ifc_softc, M_IFLIB); 5032 free(ctx, M_IFLIB); 5033 return (0); 5034 } 5035 5036 int 5037 iflib_device_attach(device_t dev) 5038 { 5039 if_ctx_t ctx; 5040 if_shared_ctx_t sctx; 5041 5042 if ((sctx = DEVICE_REGISTER(dev)) == NULL || sctx->isc_magic != IFLIB_MAGIC) 5043 return (ENOTSUP); 5044 5045 pci_enable_busmaster(dev); 5046 5047 return (iflib_device_register(dev, NULL, sctx, &ctx)); 5048 } 5049 5050 int 5051 iflib_device_deregister(if_ctx_t ctx) 5052 { 5053 if_t ifp = ctx->ifc_ifp; 5054 iflib_txq_t txq; 5055 iflib_rxq_t rxq; 5056 device_t dev = ctx->ifc_dev; 5057 int i, j; 5058 struct taskqgroup *tqg; 5059 iflib_fl_t fl; 5060 5061 /* Make sure VLANS are not using driver */ 5062 if (if_vlantrunkinuse(ifp)) { 5063 device_printf(dev, "Vlan in use, detach first\n"); 5064 return (EBUSY); 5065 } 5066 #ifdef PCI_IOV 5067 if (!CTX_IS_VF(ctx) && pci_iov_detach(dev) != 0) { 5068 device_printf(dev, "SR-IOV in use; detach first.\n"); 5069 return (EBUSY); 5070 } 5071 #endif 5072 5073 STATE_LOCK(ctx); 5074 ctx->ifc_flags |= IFC_IN_DETACH; 5075 STATE_UNLOCK(ctx); 5076 5077 CTX_LOCK(ctx); 5078 iflib_stop(ctx); 5079 CTX_UNLOCK(ctx); 5080 5081 /* Unregister VLAN events */ 5082 if (ctx->ifc_vlan_attach_event != NULL) 5083 EVENTHANDLER_DEREGISTER(vlan_config, ctx->ifc_vlan_attach_event); 5084 if (ctx->ifc_vlan_detach_event != NULL) 5085 EVENTHANDLER_DEREGISTER(vlan_unconfig, ctx->ifc_vlan_detach_event); 5086 5087 iflib_netmap_detach(ifp); 5088 ether_ifdetach(ifp); 5089 iflib_rem_pfil(ctx); 5090 if (ctx->ifc_led_dev != NULL) 5091 led_destroy(ctx->ifc_led_dev); 5092 /* XXX drain any dependent tasks */ 5093 tqg = qgroup_if_io_tqg; 5094 for (txq = ctx->ifc_txqs, i = 0; i < NTXQSETS(ctx); i++, txq++) { 5095 callout_drain(&txq->ift_timer); 5096 if (txq->ift_task.gt_uniq != NULL) 5097 taskqgroup_detach(tqg, &txq->ift_task); 5098 } 5099 for (i = 0, rxq = ctx->ifc_rxqs; i < NRXQSETS(ctx); i++, rxq++) { 5100 if (rxq->ifr_task.gt_uniq != NULL) 5101 taskqgroup_detach(tqg, &rxq->ifr_task); 5102 5103 for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) 5104 free(fl->ifl_rx_bitmap, M_IFLIB); 5105 } 5106 tqg = qgroup_if_config_tqg; 5107 if (ctx->ifc_admin_task.gt_uniq != NULL) 5108 taskqgroup_detach(tqg, &ctx->ifc_admin_task); 5109 if (ctx->ifc_vflr_task.gt_uniq != NULL) 5110 taskqgroup_detach(tqg, &ctx->ifc_vflr_task); 5111 CTX_LOCK(ctx); 5112 IFDI_DETACH(ctx); 5113 CTX_UNLOCK(ctx); 5114 5115 /* ether_ifdetach calls if_qflush - lock must be destroy afterwards*/ 5116 CTX_LOCK_DESTROY(ctx); 5117 device_set_softc(ctx->ifc_dev, NULL); 5118 iflib_free_intr_mem(ctx); 5119 5120 bus_generic_detach(dev); 5121 if_free(ifp); 5122 5123 iflib_tx_structures_free(ctx); 5124 iflib_rx_structures_free(ctx); 5125 if (ctx->ifc_flags & IFC_SC_ALLOCATED) 5126 free(ctx->ifc_softc, M_IFLIB); 5127 unref_ctx_core_offset(ctx); 5128 STATE_LOCK_DESTROY(ctx); 5129 free(ctx, M_IFLIB); 5130 return (0); 5131 } 5132 5133 static void 5134 iflib_free_intr_mem(if_ctx_t ctx) 5135 { 5136 5137 if (ctx->ifc_softc_ctx.isc_intr != IFLIB_INTR_MSIX) { 5138 iflib_irq_free(ctx, &ctx->ifc_legacy_irq); 5139 } 5140 if (ctx->ifc_softc_ctx.isc_intr != IFLIB_INTR_LEGACY) { 5141 pci_release_msi(ctx->ifc_dev); 5142 } 5143 if (ctx->ifc_msix_mem != NULL) { 5144 bus_release_resource(ctx->ifc_dev, SYS_RES_MEMORY, 5145 rman_get_rid(ctx->ifc_msix_mem), ctx->ifc_msix_mem); 5146 ctx->ifc_msix_mem = NULL; 5147 } 5148 } 5149 5150 int 5151 iflib_device_detach(device_t dev) 5152 { 5153 if_ctx_t ctx = device_get_softc(dev); 5154 5155 return (iflib_device_deregister(ctx)); 5156 } 5157 5158 int 5159 iflib_device_suspend(device_t dev) 5160 { 5161 if_ctx_t ctx = device_get_softc(dev); 5162 5163 CTX_LOCK(ctx); 5164 IFDI_SUSPEND(ctx); 5165 CTX_UNLOCK(ctx); 5166 5167 return bus_generic_suspend(dev); 5168 } 5169 int 5170 iflib_device_shutdown(device_t dev) 5171 { 5172 if_ctx_t ctx = device_get_softc(dev); 5173 5174 CTX_LOCK(ctx); 5175 IFDI_SHUTDOWN(ctx); 5176 CTX_UNLOCK(ctx); 5177 5178 return bus_generic_suspend(dev); 5179 } 5180 5181 5182 int 5183 iflib_device_resume(device_t dev) 5184 { 5185 if_ctx_t ctx = device_get_softc(dev); 5186 iflib_txq_t txq = ctx->ifc_txqs; 5187 5188 CTX_LOCK(ctx); 5189 IFDI_RESUME(ctx); 5190 iflib_if_init_locked(ctx); 5191 CTX_UNLOCK(ctx); 5192 for (int i = 0; i < NTXQSETS(ctx); i++, txq++) 5193 iflib_txq_check_drain(txq, IFLIB_RESTART_BUDGET); 5194 5195 return (bus_generic_resume(dev)); 5196 } 5197 5198 int 5199 iflib_device_iov_init(device_t dev, uint16_t num_vfs, const nvlist_t *params) 5200 { 5201 int error; 5202 if_ctx_t ctx = device_get_softc(dev); 5203 5204 CTX_LOCK(ctx); 5205 error = IFDI_IOV_INIT(ctx, num_vfs, params); 5206 CTX_UNLOCK(ctx); 5207 5208 return (error); 5209 } 5210 5211 void 5212 iflib_device_iov_uninit(device_t dev) 5213 { 5214 if_ctx_t ctx = device_get_softc(dev); 5215 5216 CTX_LOCK(ctx); 5217 IFDI_IOV_UNINIT(ctx); 5218 CTX_UNLOCK(ctx); 5219 } 5220 5221 int 5222 iflib_device_iov_add_vf(device_t dev, uint16_t vfnum, const nvlist_t *params) 5223 { 5224 int error; 5225 if_ctx_t ctx = device_get_softc(dev); 5226 5227 CTX_LOCK(ctx); 5228 error = IFDI_IOV_VF_ADD(ctx, vfnum, params); 5229 CTX_UNLOCK(ctx); 5230 5231 return (error); 5232 } 5233 5234 /********************************************************************* 5235 * 5236 * MODULE FUNCTION DEFINITIONS 5237 * 5238 **********************************************************************/ 5239 5240 /* 5241 * - Start a fast taskqueue thread for each core 5242 * - Start a taskqueue for control operations 5243 */ 5244 static int 5245 iflib_module_init(void) 5246 { 5247 return (0); 5248 } 5249 5250 static int 5251 iflib_module_event_handler(module_t mod, int what, void *arg) 5252 { 5253 int err; 5254 5255 switch (what) { 5256 case MOD_LOAD: 5257 if ((err = iflib_module_init()) != 0) 5258 return (err); 5259 break; 5260 case MOD_UNLOAD: 5261 return (EBUSY); 5262 default: 5263 return (EOPNOTSUPP); 5264 } 5265 5266 return (0); 5267 } 5268 5269 /********************************************************************* 5270 * 5271 * PUBLIC FUNCTION DEFINITIONS 5272 * ordered as in iflib.h 5273 * 5274 **********************************************************************/ 5275 5276 5277 static void 5278 _iflib_assert(if_shared_ctx_t sctx) 5279 { 5280 int i; 5281 5282 MPASS(sctx->isc_tx_maxsize); 5283 MPASS(sctx->isc_tx_maxsegsize); 5284 5285 MPASS(sctx->isc_rx_maxsize); 5286 MPASS(sctx->isc_rx_nsegments); 5287 MPASS(sctx->isc_rx_maxsegsize); 5288 5289 MPASS(sctx->isc_nrxqs >= 1 && sctx->isc_nrxqs <= 8); 5290 for (i = 0; i < sctx->isc_nrxqs; i++) { 5291 MPASS(sctx->isc_nrxd_min[i]); 5292 MPASS(powerof2(sctx->isc_nrxd_min[i])); 5293 MPASS(sctx->isc_nrxd_max[i]); 5294 MPASS(powerof2(sctx->isc_nrxd_max[i])); 5295 MPASS(sctx->isc_nrxd_default[i]); 5296 MPASS(powerof2(sctx->isc_nrxd_default[i])); 5297 } 5298 5299 MPASS(sctx->isc_ntxqs >= 1 && sctx->isc_ntxqs <= 8); 5300 for (i = 0; i < sctx->isc_ntxqs; i++) { 5301 MPASS(sctx->isc_ntxd_min[i]); 5302 MPASS(powerof2(sctx->isc_ntxd_min[i])); 5303 MPASS(sctx->isc_ntxd_max[i]); 5304 MPASS(powerof2(sctx->isc_ntxd_max[i])); 5305 MPASS(sctx->isc_ntxd_default[i]); 5306 MPASS(powerof2(sctx->isc_ntxd_default[i])); 5307 } 5308 } 5309 5310 static void 5311 _iflib_pre_assert(if_softc_ctx_t scctx) 5312 { 5313 5314 MPASS(scctx->isc_txrx->ift_txd_encap); 5315 MPASS(scctx->isc_txrx->ift_txd_flush); 5316 MPASS(scctx->isc_txrx->ift_txd_credits_update); 5317 MPASS(scctx->isc_txrx->ift_rxd_available); 5318 MPASS(scctx->isc_txrx->ift_rxd_pkt_get); 5319 MPASS(scctx->isc_txrx->ift_rxd_refill); 5320 MPASS(scctx->isc_txrx->ift_rxd_flush); 5321 } 5322 5323 static int 5324 iflib_register(if_ctx_t ctx) 5325 { 5326 if_shared_ctx_t sctx = ctx->ifc_sctx; 5327 driver_t *driver = sctx->isc_driver; 5328 device_t dev = ctx->ifc_dev; 5329 if_t ifp; 5330 5331 _iflib_assert(sctx); 5332 5333 CTX_LOCK_INIT(ctx); 5334 STATE_LOCK_INIT(ctx, device_get_nameunit(ctx->ifc_dev)); 5335 ifp = ctx->ifc_ifp = if_alloc(IFT_ETHER); 5336 if (ifp == NULL) { 5337 device_printf(dev, "can not allocate ifnet structure\n"); 5338 return (ENOMEM); 5339 } 5340 5341 /* 5342 * Initialize our context's device specific methods 5343 */ 5344 kobj_init((kobj_t) ctx, (kobj_class_t) driver); 5345 kobj_class_compile((kobj_class_t) driver); 5346 driver->refs++; 5347 5348 if_initname(ifp, device_get_name(dev), device_get_unit(dev)); 5349 if_setsoftc(ifp, ctx); 5350 if_setdev(ifp, dev); 5351 if_setinitfn(ifp, iflib_if_init); 5352 if_setioctlfn(ifp, iflib_if_ioctl); 5353 #ifdef ALTQ 5354 if_setstartfn(ifp, iflib_altq_if_start); 5355 if_settransmitfn(ifp, iflib_altq_if_transmit); 5356 if_setsendqready(ifp); 5357 #else 5358 if_settransmitfn(ifp, iflib_if_transmit); 5359 #endif 5360 if_setqflushfn(ifp, iflib_if_qflush); 5361 if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST); 5362 5363 ctx->ifc_vlan_attach_event = 5364 EVENTHANDLER_REGISTER(vlan_config, iflib_vlan_register, ctx, 5365 EVENTHANDLER_PRI_FIRST); 5366 ctx->ifc_vlan_detach_event = 5367 EVENTHANDLER_REGISTER(vlan_unconfig, iflib_vlan_unregister, ctx, 5368 EVENTHANDLER_PRI_FIRST); 5369 5370 if ((sctx->isc_flags & IFLIB_DRIVER_MEDIA) == 0) { 5371 ctx->ifc_mediap = &ctx->ifc_media; 5372 ifmedia_init(ctx->ifc_mediap, IFM_IMASK, 5373 iflib_media_change, iflib_media_status); 5374 } 5375 return (0); 5376 } 5377 5378 static int 5379 iflib_queues_alloc(if_ctx_t ctx) 5380 { 5381 if_shared_ctx_t sctx = ctx->ifc_sctx; 5382 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 5383 device_t dev = ctx->ifc_dev; 5384 int nrxqsets = scctx->isc_nrxqsets; 5385 int ntxqsets = scctx->isc_ntxqsets; 5386 iflib_txq_t txq; 5387 iflib_rxq_t rxq; 5388 iflib_fl_t fl = NULL; 5389 int i, j, cpu, err, txconf, rxconf; 5390 iflib_dma_info_t ifdip; 5391 uint32_t *rxqsizes = scctx->isc_rxqsizes; 5392 uint32_t *txqsizes = scctx->isc_txqsizes; 5393 uint8_t nrxqs = sctx->isc_nrxqs; 5394 uint8_t ntxqs = sctx->isc_ntxqs; 5395 int nfree_lists = sctx->isc_nfl ? sctx->isc_nfl : 1; 5396 caddr_t *vaddrs; 5397 uint64_t *paddrs; 5398 5399 KASSERT(ntxqs > 0, ("number of queues per qset must be at least 1")); 5400 KASSERT(nrxqs > 0, ("number of queues per qset must be at least 1")); 5401 5402 /* Allocate the TX ring struct memory */ 5403 if (!(ctx->ifc_txqs = 5404 (iflib_txq_t) malloc(sizeof(struct iflib_txq) * 5405 ntxqsets, M_IFLIB, M_NOWAIT | M_ZERO))) { 5406 device_printf(dev, "Unable to allocate TX ring memory\n"); 5407 err = ENOMEM; 5408 goto fail; 5409 } 5410 5411 /* Now allocate the RX */ 5412 if (!(ctx->ifc_rxqs = 5413 (iflib_rxq_t) malloc(sizeof(struct iflib_rxq) * 5414 nrxqsets, M_IFLIB, M_NOWAIT | M_ZERO))) { 5415 device_printf(dev, "Unable to allocate RX ring memory\n"); 5416 err = ENOMEM; 5417 goto rx_fail; 5418 } 5419 5420 txq = ctx->ifc_txqs; 5421 rxq = ctx->ifc_rxqs; 5422 5423 /* 5424 * XXX handle allocation failure 5425 */ 5426 for (txconf = i = 0, cpu = CPU_FIRST(); i < ntxqsets; i++, txconf++, txq++, cpu = CPU_NEXT(cpu)) { 5427 /* Set up some basics */ 5428 5429 if ((ifdip = malloc(sizeof(struct iflib_dma_info) * ntxqs, 5430 M_IFLIB, M_NOWAIT | M_ZERO)) == NULL) { 5431 device_printf(dev, 5432 "Unable to allocate TX DMA info memory\n"); 5433 err = ENOMEM; 5434 goto err_tx_desc; 5435 } 5436 txq->ift_ifdi = ifdip; 5437 for (j = 0; j < ntxqs; j++, ifdip++) { 5438 if (iflib_dma_alloc(ctx, txqsizes[j], ifdip, 0)) { 5439 device_printf(dev, 5440 "Unable to allocate TX descriptors\n"); 5441 err = ENOMEM; 5442 goto err_tx_desc; 5443 } 5444 txq->ift_txd_size[j] = scctx->isc_txd_size[j]; 5445 bzero((void *)ifdip->idi_vaddr, txqsizes[j]); 5446 } 5447 txq->ift_ctx = ctx; 5448 txq->ift_id = i; 5449 if (sctx->isc_flags & IFLIB_HAS_TXCQ) { 5450 txq->ift_br_offset = 1; 5451 } else { 5452 txq->ift_br_offset = 0; 5453 } 5454 /* XXX fix this */ 5455 txq->ift_timer.c_cpu = cpu; 5456 5457 if (iflib_txsd_alloc(txq)) { 5458 device_printf(dev, "Critical Failure setting up TX buffers\n"); 5459 err = ENOMEM; 5460 goto err_tx_desc; 5461 } 5462 5463 /* Initialize the TX lock */ 5464 snprintf(txq->ift_mtx_name, MTX_NAME_LEN, "%s:TX(%d):callout", 5465 device_get_nameunit(dev), txq->ift_id); 5466 mtx_init(&txq->ift_mtx, txq->ift_mtx_name, NULL, MTX_DEF); 5467 callout_init_mtx(&txq->ift_timer, &txq->ift_mtx, 0); 5468 5469 err = ifmp_ring_alloc(&txq->ift_br, 2048, txq, iflib_txq_drain, 5470 iflib_txq_can_drain, M_IFLIB, M_WAITOK); 5471 if (err) { 5472 /* XXX free any allocated rings */ 5473 device_printf(dev, "Unable to allocate buf_ring\n"); 5474 goto err_tx_desc; 5475 } 5476 } 5477 5478 for (rxconf = i = 0; i < nrxqsets; i++, rxconf++, rxq++) { 5479 /* Set up some basics */ 5480 5481 if ((ifdip = malloc(sizeof(struct iflib_dma_info) * nrxqs, 5482 M_IFLIB, M_NOWAIT | M_ZERO)) == NULL) { 5483 device_printf(dev, 5484 "Unable to allocate RX DMA info memory\n"); 5485 err = ENOMEM; 5486 goto err_tx_desc; 5487 } 5488 5489 rxq->ifr_ifdi = ifdip; 5490 /* XXX this needs to be changed if #rx queues != #tx queues */ 5491 rxq->ifr_ntxqirq = 1; 5492 rxq->ifr_txqid[0] = i; 5493 for (j = 0; j < nrxqs; j++, ifdip++) { 5494 if (iflib_dma_alloc(ctx, rxqsizes[j], ifdip, 0)) { 5495 device_printf(dev, 5496 "Unable to allocate RX descriptors\n"); 5497 err = ENOMEM; 5498 goto err_tx_desc; 5499 } 5500 bzero((void *)ifdip->idi_vaddr, rxqsizes[j]); 5501 } 5502 rxq->ifr_ctx = ctx; 5503 rxq->ifr_id = i; 5504 if (sctx->isc_flags & IFLIB_HAS_RXCQ) { 5505 rxq->ifr_fl_offset = 1; 5506 } else { 5507 rxq->ifr_fl_offset = 0; 5508 } 5509 rxq->ifr_nfl = nfree_lists; 5510 if (!(fl = 5511 (iflib_fl_t) malloc(sizeof(struct iflib_fl) * nfree_lists, M_IFLIB, M_NOWAIT | M_ZERO))) { 5512 device_printf(dev, "Unable to allocate free list memory\n"); 5513 err = ENOMEM; 5514 goto err_tx_desc; 5515 } 5516 rxq->ifr_fl = fl; 5517 for (j = 0; j < nfree_lists; j++) { 5518 fl[j].ifl_rxq = rxq; 5519 fl[j].ifl_id = j; 5520 fl[j].ifl_ifdi = &rxq->ifr_ifdi[j + rxq->ifr_fl_offset]; 5521 fl[j].ifl_rxd_size = scctx->isc_rxd_size[j]; 5522 } 5523 /* Allocate receive buffers for the ring */ 5524 if (iflib_rxsd_alloc(rxq)) { 5525 device_printf(dev, 5526 "Critical Failure setting up receive buffers\n"); 5527 err = ENOMEM; 5528 goto err_rx_desc; 5529 } 5530 5531 for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) 5532 fl->ifl_rx_bitmap = bit_alloc(fl->ifl_size, M_IFLIB, 5533 M_WAITOK); 5534 } 5535 5536 /* TXQs */ 5537 vaddrs = malloc(sizeof(caddr_t)*ntxqsets*ntxqs, M_IFLIB, M_WAITOK); 5538 paddrs = malloc(sizeof(uint64_t)*ntxqsets*ntxqs, M_IFLIB, M_WAITOK); 5539 for (i = 0; i < ntxqsets; i++) { 5540 iflib_dma_info_t di = ctx->ifc_txqs[i].ift_ifdi; 5541 5542 for (j = 0; j < ntxqs; j++, di++) { 5543 vaddrs[i*ntxqs + j] = di->idi_vaddr; 5544 paddrs[i*ntxqs + j] = di->idi_paddr; 5545 } 5546 } 5547 if ((err = IFDI_TX_QUEUES_ALLOC(ctx, vaddrs, paddrs, ntxqs, ntxqsets)) != 0) { 5548 device_printf(ctx->ifc_dev, 5549 "Unable to allocate device TX queue\n"); 5550 iflib_tx_structures_free(ctx); 5551 free(vaddrs, M_IFLIB); 5552 free(paddrs, M_IFLIB); 5553 goto err_rx_desc; 5554 } 5555 free(vaddrs, M_IFLIB); 5556 free(paddrs, M_IFLIB); 5557 5558 /* RXQs */ 5559 vaddrs = malloc(sizeof(caddr_t)*nrxqsets*nrxqs, M_IFLIB, M_WAITOK); 5560 paddrs = malloc(sizeof(uint64_t)*nrxqsets*nrxqs, M_IFLIB, M_WAITOK); 5561 for (i = 0; i < nrxqsets; i++) { 5562 iflib_dma_info_t di = ctx->ifc_rxqs[i].ifr_ifdi; 5563 5564 for (j = 0; j < nrxqs; j++, di++) { 5565 vaddrs[i*nrxqs + j] = di->idi_vaddr; 5566 paddrs[i*nrxqs + j] = di->idi_paddr; 5567 } 5568 } 5569 if ((err = IFDI_RX_QUEUES_ALLOC(ctx, vaddrs, paddrs, nrxqs, nrxqsets)) != 0) { 5570 device_printf(ctx->ifc_dev, 5571 "Unable to allocate device RX queue\n"); 5572 iflib_tx_structures_free(ctx); 5573 free(vaddrs, M_IFLIB); 5574 free(paddrs, M_IFLIB); 5575 goto err_rx_desc; 5576 } 5577 free(vaddrs, M_IFLIB); 5578 free(paddrs, M_IFLIB); 5579 5580 return (0); 5581 5582 /* XXX handle allocation failure changes */ 5583 err_rx_desc: 5584 err_tx_desc: 5585 rx_fail: 5586 if (ctx->ifc_rxqs != NULL) 5587 free(ctx->ifc_rxqs, M_IFLIB); 5588 ctx->ifc_rxqs = NULL; 5589 if (ctx->ifc_txqs != NULL) 5590 free(ctx->ifc_txqs, M_IFLIB); 5591 ctx->ifc_txqs = NULL; 5592 fail: 5593 return (err); 5594 } 5595 5596 static int 5597 iflib_tx_structures_setup(if_ctx_t ctx) 5598 { 5599 iflib_txq_t txq = ctx->ifc_txqs; 5600 int i; 5601 5602 for (i = 0; i < NTXQSETS(ctx); i++, txq++) 5603 iflib_txq_setup(txq); 5604 5605 return (0); 5606 } 5607 5608 static void 5609 iflib_tx_structures_free(if_ctx_t ctx) 5610 { 5611 iflib_txq_t txq = ctx->ifc_txqs; 5612 if_shared_ctx_t sctx = ctx->ifc_sctx; 5613 int i, j; 5614 5615 for (i = 0; i < NTXQSETS(ctx); i++, txq++) { 5616 iflib_txq_destroy(txq); 5617 for (j = 0; j < sctx->isc_ntxqs; j++) 5618 iflib_dma_free(&txq->ift_ifdi[j]); 5619 } 5620 free(ctx->ifc_txqs, M_IFLIB); 5621 ctx->ifc_txqs = NULL; 5622 IFDI_QUEUES_FREE(ctx); 5623 } 5624 5625 /********************************************************************* 5626 * 5627 * Initialize all receive rings. 5628 * 5629 **********************************************************************/ 5630 static int 5631 iflib_rx_structures_setup(if_ctx_t ctx) 5632 { 5633 iflib_rxq_t rxq = ctx->ifc_rxqs; 5634 int q; 5635 #if defined(INET6) || defined(INET) 5636 int err, i; 5637 #endif 5638 5639 for (q = 0; q < ctx->ifc_softc_ctx.isc_nrxqsets; q++, rxq++) { 5640 #if defined(INET6) || defined(INET) 5641 if (if_getcapabilities(ctx->ifc_ifp) & IFCAP_LRO) { 5642 err = tcp_lro_init_args(&rxq->ifr_lc, ctx->ifc_ifp, 5643 TCP_LRO_ENTRIES, min(1024, 5644 ctx->ifc_softc_ctx.isc_nrxd[rxq->ifr_fl_offset])); 5645 if (err != 0) { 5646 device_printf(ctx->ifc_dev, 5647 "LRO Initialization failed!\n"); 5648 goto fail; 5649 } 5650 } 5651 #endif 5652 IFDI_RXQ_SETUP(ctx, rxq->ifr_id); 5653 } 5654 return (0); 5655 #if defined(INET6) || defined(INET) 5656 fail: 5657 /* 5658 * Free LRO resources allocated so far, we will only handle 5659 * the rings that completed, the failing case will have 5660 * cleaned up for itself. 'q' failed, so its the terminus. 5661 */ 5662 rxq = ctx->ifc_rxqs; 5663 for (i = 0; i < q; ++i, rxq++) { 5664 if (if_getcapabilities(ctx->ifc_ifp) & IFCAP_LRO) 5665 tcp_lro_free(&rxq->ifr_lc); 5666 } 5667 return (err); 5668 #endif 5669 } 5670 5671 /********************************************************************* 5672 * 5673 * Free all receive rings. 5674 * 5675 **********************************************************************/ 5676 static void 5677 iflib_rx_structures_free(if_ctx_t ctx) 5678 { 5679 iflib_rxq_t rxq = ctx->ifc_rxqs; 5680 int i; 5681 5682 for (i = 0; i < ctx->ifc_softc_ctx.isc_nrxqsets; i++, rxq++) { 5683 iflib_rx_sds_free(rxq); 5684 #if defined(INET6) || defined(INET) 5685 if (if_getcapabilities(ctx->ifc_ifp) & IFCAP_LRO) 5686 tcp_lro_free(&rxq->ifr_lc); 5687 #endif 5688 } 5689 free(ctx->ifc_rxqs, M_IFLIB); 5690 ctx->ifc_rxqs = NULL; 5691 } 5692 5693 static int 5694 iflib_qset_structures_setup(if_ctx_t ctx) 5695 { 5696 int err; 5697 5698 /* 5699 * It is expected that the caller takes care of freeing queues if this 5700 * fails. 5701 */ 5702 if ((err = iflib_tx_structures_setup(ctx)) != 0) { 5703 device_printf(ctx->ifc_dev, "iflib_tx_structures_setup failed: %d\n", err); 5704 return (err); 5705 } 5706 5707 if ((err = iflib_rx_structures_setup(ctx)) != 0) 5708 device_printf(ctx->ifc_dev, "iflib_rx_structures_setup failed: %d\n", err); 5709 5710 return (err); 5711 } 5712 5713 int 5714 iflib_irq_alloc(if_ctx_t ctx, if_irq_t irq, int rid, 5715 driver_filter_t filter, void *filter_arg, driver_intr_t handler, void *arg, const char *name) 5716 { 5717 5718 return (_iflib_irq_alloc(ctx, irq, rid, filter, handler, arg, name)); 5719 } 5720 5721 #ifdef SMP 5722 static int 5723 find_nth(if_ctx_t ctx, int qid) 5724 { 5725 cpuset_t cpus; 5726 int i, cpuid, eqid, count; 5727 5728 CPU_COPY(&ctx->ifc_cpus, &cpus); 5729 count = CPU_COUNT(&cpus); 5730 eqid = qid % count; 5731 /* clear up to the qid'th bit */ 5732 for (i = 0; i < eqid; i++) { 5733 cpuid = CPU_FFS(&cpus); 5734 MPASS(cpuid != 0); 5735 CPU_CLR(cpuid-1, &cpus); 5736 } 5737 cpuid = CPU_FFS(&cpus); 5738 MPASS(cpuid != 0); 5739 return (cpuid-1); 5740 } 5741 5742 #ifdef SCHED_ULE 5743 extern struct cpu_group *cpu_top; /* CPU topology */ 5744 5745 static int 5746 find_child_with_core(int cpu, struct cpu_group *grp) 5747 { 5748 int i; 5749 5750 if (grp->cg_children == 0) 5751 return -1; 5752 5753 MPASS(grp->cg_child); 5754 for (i = 0; i < grp->cg_children; i++) { 5755 if (CPU_ISSET(cpu, &grp->cg_child[i].cg_mask)) 5756 return i; 5757 } 5758 5759 return -1; 5760 } 5761 5762 /* 5763 * Find the nth "close" core to the specified core 5764 * "close" is defined as the deepest level that shares 5765 * at least an L2 cache. With threads, this will be 5766 * threads on the same core. If the shared cache is L3 5767 * or higher, simply returns the same core. 5768 */ 5769 static int 5770 find_close_core(int cpu, int core_offset) 5771 { 5772 struct cpu_group *grp; 5773 int i; 5774 int fcpu; 5775 cpuset_t cs; 5776 5777 grp = cpu_top; 5778 if (grp == NULL) 5779 return cpu; 5780 i = 0; 5781 while ((i = find_child_with_core(cpu, grp)) != -1) { 5782 /* If the child only has one cpu, don't descend */ 5783 if (grp->cg_child[i].cg_count <= 1) 5784 break; 5785 grp = &grp->cg_child[i]; 5786 } 5787 5788 /* If they don't share at least an L2 cache, use the same CPU */ 5789 if (grp->cg_level > CG_SHARE_L2 || grp->cg_level == CG_SHARE_NONE) 5790 return cpu; 5791 5792 /* Now pick one */ 5793 CPU_COPY(&grp->cg_mask, &cs); 5794 5795 /* Add the selected CPU offset to core offset. */ 5796 for (i = 0; (fcpu = CPU_FFS(&cs)) != 0; i++) { 5797 if (fcpu - 1 == cpu) 5798 break; 5799 CPU_CLR(fcpu - 1, &cs); 5800 } 5801 MPASS(fcpu); 5802 5803 core_offset += i; 5804 5805 CPU_COPY(&grp->cg_mask, &cs); 5806 for (i = core_offset % grp->cg_count; i > 0; i--) { 5807 MPASS(CPU_FFS(&cs)); 5808 CPU_CLR(CPU_FFS(&cs) - 1, &cs); 5809 } 5810 MPASS(CPU_FFS(&cs)); 5811 return CPU_FFS(&cs) - 1; 5812 } 5813 #else 5814 static int 5815 find_close_core(int cpu, int core_offset __unused) 5816 { 5817 return cpu; 5818 } 5819 #endif 5820 5821 static int 5822 get_core_offset(if_ctx_t ctx, iflib_intr_type_t type, int qid) 5823 { 5824 switch (type) { 5825 case IFLIB_INTR_TX: 5826 /* TX queues get cores which share at least an L2 cache with the corresponding RX queue */ 5827 /* XXX handle multiple RX threads per core and more than two core per L2 group */ 5828 return qid / CPU_COUNT(&ctx->ifc_cpus) + 1; 5829 case IFLIB_INTR_RX: 5830 case IFLIB_INTR_RXTX: 5831 /* RX queues get the specified core */ 5832 return qid / CPU_COUNT(&ctx->ifc_cpus); 5833 default: 5834 return -1; 5835 } 5836 } 5837 #else 5838 #define get_core_offset(ctx, type, qid) CPU_FIRST() 5839 #define find_close_core(cpuid, tid) CPU_FIRST() 5840 #define find_nth(ctx, gid) CPU_FIRST() 5841 #endif 5842 5843 /* Just to avoid copy/paste */ 5844 static inline int 5845 iflib_irq_set_affinity(if_ctx_t ctx, if_irq_t irq, iflib_intr_type_t type, 5846 int qid, struct grouptask *gtask, struct taskqgroup *tqg, void *uniq, 5847 const char *name) 5848 { 5849 device_t dev; 5850 int co, cpuid, err, tid; 5851 5852 dev = ctx->ifc_dev; 5853 co = ctx->ifc_sysctl_core_offset; 5854 if (ctx->ifc_sysctl_separate_txrx && type == IFLIB_INTR_TX) 5855 co += ctx->ifc_softc_ctx.isc_nrxqsets; 5856 cpuid = find_nth(ctx, qid + co); 5857 tid = get_core_offset(ctx, type, qid); 5858 if (tid < 0) { 5859 device_printf(dev, "get_core_offset failed\n"); 5860 return (EOPNOTSUPP); 5861 } 5862 cpuid = find_close_core(cpuid, tid); 5863 err = taskqgroup_attach_cpu(tqg, gtask, uniq, cpuid, dev, irq->ii_res, 5864 name); 5865 if (err) { 5866 device_printf(dev, "taskqgroup_attach_cpu failed %d\n", err); 5867 return (err); 5868 } 5869 #ifdef notyet 5870 if (cpuid > ctx->ifc_cpuid_highest) 5871 ctx->ifc_cpuid_highest = cpuid; 5872 #endif 5873 return (0); 5874 } 5875 5876 int 5877 iflib_irq_alloc_generic(if_ctx_t ctx, if_irq_t irq, int rid, 5878 iflib_intr_type_t type, driver_filter_t *filter, 5879 void *filter_arg, int qid, const char *name) 5880 { 5881 device_t dev; 5882 struct grouptask *gtask; 5883 struct taskqgroup *tqg; 5884 iflib_filter_info_t info; 5885 gtask_fn_t *fn; 5886 int tqrid, err; 5887 driver_filter_t *intr_fast; 5888 void *q; 5889 5890 info = &ctx->ifc_filter_info; 5891 tqrid = rid; 5892 5893 switch (type) { 5894 /* XXX merge tx/rx for netmap? */ 5895 case IFLIB_INTR_TX: 5896 q = &ctx->ifc_txqs[qid]; 5897 info = &ctx->ifc_txqs[qid].ift_filter_info; 5898 gtask = &ctx->ifc_txqs[qid].ift_task; 5899 tqg = qgroup_if_io_tqg; 5900 fn = _task_fn_tx; 5901 intr_fast = iflib_fast_intr; 5902 GROUPTASK_INIT(gtask, 0, fn, q); 5903 ctx->ifc_flags |= IFC_NETMAP_TX_IRQ; 5904 break; 5905 case IFLIB_INTR_RX: 5906 q = &ctx->ifc_rxqs[qid]; 5907 info = &ctx->ifc_rxqs[qid].ifr_filter_info; 5908 gtask = &ctx->ifc_rxqs[qid].ifr_task; 5909 tqg = qgroup_if_io_tqg; 5910 fn = _task_fn_rx; 5911 intr_fast = iflib_fast_intr; 5912 GROUPTASK_INIT(gtask, 0, fn, q); 5913 break; 5914 case IFLIB_INTR_RXTX: 5915 q = &ctx->ifc_rxqs[qid]; 5916 info = &ctx->ifc_rxqs[qid].ifr_filter_info; 5917 gtask = &ctx->ifc_rxqs[qid].ifr_task; 5918 tqg = qgroup_if_io_tqg; 5919 fn = _task_fn_rx; 5920 intr_fast = iflib_fast_intr_rxtx; 5921 GROUPTASK_INIT(gtask, 0, fn, q); 5922 break; 5923 case IFLIB_INTR_ADMIN: 5924 q = ctx; 5925 tqrid = -1; 5926 info = &ctx->ifc_filter_info; 5927 gtask = &ctx->ifc_admin_task; 5928 tqg = qgroup_if_config_tqg; 5929 fn = _task_fn_admin; 5930 intr_fast = iflib_fast_intr_ctx; 5931 break; 5932 default: 5933 device_printf(ctx->ifc_dev, "%s: unknown net intr type\n", 5934 __func__); 5935 return (EINVAL); 5936 } 5937 5938 info->ifi_filter = filter; 5939 info->ifi_filter_arg = filter_arg; 5940 info->ifi_task = gtask; 5941 info->ifi_ctx = q; 5942 5943 dev = ctx->ifc_dev; 5944 err = _iflib_irq_alloc(ctx, irq, rid, intr_fast, NULL, info, name); 5945 if (err != 0) { 5946 device_printf(dev, "_iflib_irq_alloc failed %d\n", err); 5947 return (err); 5948 } 5949 if (type == IFLIB_INTR_ADMIN) 5950 return (0); 5951 5952 if (tqrid != -1) { 5953 err = iflib_irq_set_affinity(ctx, irq, type, qid, gtask, tqg, 5954 q, name); 5955 if (err) 5956 return (err); 5957 } else { 5958 taskqgroup_attach(tqg, gtask, q, dev, irq->ii_res, name); 5959 } 5960 5961 return (0); 5962 } 5963 5964 void 5965 iflib_softirq_alloc_generic(if_ctx_t ctx, if_irq_t irq, iflib_intr_type_t type, void *arg, int qid, const char *name) 5966 { 5967 struct grouptask *gtask; 5968 struct taskqgroup *tqg; 5969 gtask_fn_t *fn; 5970 void *q; 5971 int err; 5972 5973 switch (type) { 5974 case IFLIB_INTR_TX: 5975 q = &ctx->ifc_txqs[qid]; 5976 gtask = &ctx->ifc_txqs[qid].ift_task; 5977 tqg = qgroup_if_io_tqg; 5978 fn = _task_fn_tx; 5979 break; 5980 case IFLIB_INTR_RX: 5981 q = &ctx->ifc_rxqs[qid]; 5982 gtask = &ctx->ifc_rxqs[qid].ifr_task; 5983 tqg = qgroup_if_io_tqg; 5984 fn = _task_fn_rx; 5985 break; 5986 case IFLIB_INTR_IOV: 5987 q = ctx; 5988 gtask = &ctx->ifc_vflr_task; 5989 tqg = qgroup_if_config_tqg; 5990 fn = _task_fn_iov; 5991 break; 5992 default: 5993 panic("unknown net intr type"); 5994 } 5995 GROUPTASK_INIT(gtask, 0, fn, q); 5996 if (irq != NULL) { 5997 err = iflib_irq_set_affinity(ctx, irq, type, qid, gtask, tqg, 5998 q, name); 5999 if (err) 6000 taskqgroup_attach(tqg, gtask, q, ctx->ifc_dev, 6001 irq->ii_res, name); 6002 } else { 6003 taskqgroup_attach(tqg, gtask, q, NULL, NULL, name); 6004 } 6005 } 6006 6007 void 6008 iflib_irq_free(if_ctx_t ctx, if_irq_t irq) 6009 { 6010 6011 if (irq->ii_tag) 6012 bus_teardown_intr(ctx->ifc_dev, irq->ii_res, irq->ii_tag); 6013 6014 if (irq->ii_res) 6015 bus_release_resource(ctx->ifc_dev, SYS_RES_IRQ, 6016 rman_get_rid(irq->ii_res), irq->ii_res); 6017 } 6018 6019 static int 6020 iflib_legacy_setup(if_ctx_t ctx, driver_filter_t filter, void *filter_arg, int *rid, const char *name) 6021 { 6022 iflib_txq_t txq = ctx->ifc_txqs; 6023 iflib_rxq_t rxq = ctx->ifc_rxqs; 6024 if_irq_t irq = &ctx->ifc_legacy_irq; 6025 iflib_filter_info_t info; 6026 device_t dev; 6027 struct grouptask *gtask; 6028 struct resource *res; 6029 struct taskqgroup *tqg; 6030 gtask_fn_t *fn; 6031 void *q; 6032 int err, tqrid; 6033 6034 q = &ctx->ifc_rxqs[0]; 6035 info = &rxq[0].ifr_filter_info; 6036 gtask = &rxq[0].ifr_task; 6037 tqg = qgroup_if_io_tqg; 6038 tqrid = *rid; 6039 fn = _task_fn_rx; 6040 6041 ctx->ifc_flags |= IFC_LEGACY; 6042 info->ifi_filter = filter; 6043 info->ifi_filter_arg = filter_arg; 6044 info->ifi_task = gtask; 6045 info->ifi_ctx = q; 6046 6047 dev = ctx->ifc_dev; 6048 /* We allocate a single interrupt resource */ 6049 if ((err = _iflib_irq_alloc(ctx, irq, tqrid, iflib_fast_intr_rxtx, 6050 NULL, info, name)) != 0) 6051 return (err); 6052 GROUPTASK_INIT(gtask, 0, fn, q); 6053 res = irq->ii_res; 6054 taskqgroup_attach(tqg, gtask, q, dev, res, name); 6055 6056 GROUPTASK_INIT(&txq->ift_task, 0, _task_fn_tx, txq); 6057 taskqgroup_attach(qgroup_if_io_tqg, &txq->ift_task, txq, dev, res, 6058 "tx"); 6059 return (0); 6060 } 6061 6062 void 6063 iflib_led_create(if_ctx_t ctx) 6064 { 6065 6066 ctx->ifc_led_dev = led_create(iflib_led_func, ctx, 6067 device_get_nameunit(ctx->ifc_dev)); 6068 } 6069 6070 void 6071 iflib_tx_intr_deferred(if_ctx_t ctx, int txqid) 6072 { 6073 6074 GROUPTASK_ENQUEUE(&ctx->ifc_txqs[txqid].ift_task); 6075 } 6076 6077 void 6078 iflib_rx_intr_deferred(if_ctx_t ctx, int rxqid) 6079 { 6080 6081 GROUPTASK_ENQUEUE(&ctx->ifc_rxqs[rxqid].ifr_task); 6082 } 6083 6084 void 6085 iflib_admin_intr_deferred(if_ctx_t ctx) 6086 { 6087 #ifdef INVARIANTS 6088 struct grouptask *gtask; 6089 6090 gtask = &ctx->ifc_admin_task; 6091 MPASS(gtask != NULL && gtask->gt_taskqueue != NULL); 6092 #endif 6093 6094 GROUPTASK_ENQUEUE(&ctx->ifc_admin_task); 6095 } 6096 6097 void 6098 iflib_iov_intr_deferred(if_ctx_t ctx) 6099 { 6100 6101 GROUPTASK_ENQUEUE(&ctx->ifc_vflr_task); 6102 } 6103 6104 void 6105 iflib_io_tqg_attach(struct grouptask *gt, void *uniq, int cpu, const char *name) 6106 { 6107 6108 taskqgroup_attach_cpu(qgroup_if_io_tqg, gt, uniq, cpu, NULL, NULL, 6109 name); 6110 } 6111 6112 void 6113 iflib_config_gtask_init(void *ctx, struct grouptask *gtask, gtask_fn_t *fn, 6114 const char *name) 6115 { 6116 6117 GROUPTASK_INIT(gtask, 0, fn, ctx); 6118 taskqgroup_attach(qgroup_if_config_tqg, gtask, gtask, NULL, NULL, 6119 name); 6120 } 6121 6122 void 6123 iflib_config_gtask_deinit(struct grouptask *gtask) 6124 { 6125 6126 taskqgroup_detach(qgroup_if_config_tqg, gtask); 6127 } 6128 6129 void 6130 iflib_link_state_change(if_ctx_t ctx, int link_state, uint64_t baudrate) 6131 { 6132 if_t ifp = ctx->ifc_ifp; 6133 iflib_txq_t txq = ctx->ifc_txqs; 6134 6135 if_setbaudrate(ifp, baudrate); 6136 if (baudrate >= IF_Gbps(10)) { 6137 STATE_LOCK(ctx); 6138 ctx->ifc_flags |= IFC_PREFETCH; 6139 STATE_UNLOCK(ctx); 6140 } 6141 /* If link down, disable watchdog */ 6142 if ((ctx->ifc_link_state == LINK_STATE_UP) && (link_state == LINK_STATE_DOWN)) { 6143 for (int i = 0; i < ctx->ifc_softc_ctx.isc_ntxqsets; i++, txq++) 6144 txq->ift_qstatus = IFLIB_QUEUE_IDLE; 6145 } 6146 ctx->ifc_link_state = link_state; 6147 if_link_state_change(ifp, link_state); 6148 } 6149 6150 static int 6151 iflib_tx_credits_update(if_ctx_t ctx, iflib_txq_t txq) 6152 { 6153 int credits; 6154 #ifdef INVARIANTS 6155 int credits_pre = txq->ift_cidx_processed; 6156 #endif 6157 6158 if (ctx->isc_txd_credits_update == NULL) 6159 return (0); 6160 6161 bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map, 6162 BUS_DMASYNC_POSTREAD); 6163 if ((credits = ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, true)) == 0) 6164 return (0); 6165 6166 txq->ift_processed += credits; 6167 txq->ift_cidx_processed += credits; 6168 6169 MPASS(credits_pre + credits == txq->ift_cidx_processed); 6170 if (txq->ift_cidx_processed >= txq->ift_size) 6171 txq->ift_cidx_processed -= txq->ift_size; 6172 return (credits); 6173 } 6174 6175 static int 6176 iflib_rxd_avail(if_ctx_t ctx, iflib_rxq_t rxq, qidx_t cidx, qidx_t budget) 6177 { 6178 iflib_fl_t fl; 6179 u_int i; 6180 6181 for (i = 0, fl = &rxq->ifr_fl[0]; i < rxq->ifr_nfl; i++, fl++) 6182 bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map, 6183 BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); 6184 return (ctx->isc_rxd_available(ctx->ifc_softc, rxq->ifr_id, cidx, 6185 budget)); 6186 } 6187 6188 void 6189 iflib_add_int_delay_sysctl(if_ctx_t ctx, const char *name, 6190 const char *description, if_int_delay_info_t info, 6191 int offset, int value) 6192 { 6193 info->iidi_ctx = ctx; 6194 info->iidi_offset = offset; 6195 info->iidi_value = value; 6196 SYSCTL_ADD_PROC(device_get_sysctl_ctx(ctx->ifc_dev), 6197 SYSCTL_CHILDREN(device_get_sysctl_tree(ctx->ifc_dev)), 6198 OID_AUTO, name, CTLTYPE_INT|CTLFLAG_RW, 6199 info, 0, iflib_sysctl_int_delay, "I", description); 6200 } 6201 6202 struct sx * 6203 iflib_ctx_lock_get(if_ctx_t ctx) 6204 { 6205 6206 return (&ctx->ifc_ctx_sx); 6207 } 6208 6209 static int 6210 iflib_msix_init(if_ctx_t ctx) 6211 { 6212 device_t dev = ctx->ifc_dev; 6213 if_shared_ctx_t sctx = ctx->ifc_sctx; 6214 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 6215 int admincnt, bar, err, iflib_num_rx_queues, iflib_num_tx_queues; 6216 int msgs, queuemsgs, queues, rx_queues, tx_queues, vectors; 6217 6218 iflib_num_tx_queues = ctx->ifc_sysctl_ntxqs; 6219 iflib_num_rx_queues = ctx->ifc_sysctl_nrxqs; 6220 6221 if (bootverbose) 6222 device_printf(dev, "msix_init qsets capped at %d\n", 6223 imax(scctx->isc_ntxqsets, scctx->isc_nrxqsets)); 6224 6225 /* Override by tuneable */ 6226 if (scctx->isc_disable_msix) 6227 goto msi; 6228 6229 /* First try MSI-X */ 6230 if ((msgs = pci_msix_count(dev)) == 0) { 6231 if (bootverbose) 6232 device_printf(dev, "MSI-X not supported or disabled\n"); 6233 goto msi; 6234 } 6235 6236 bar = ctx->ifc_softc_ctx.isc_msix_bar; 6237 /* 6238 * bar == -1 => "trust me I know what I'm doing" 6239 * Some drivers are for hardware that is so shoddily 6240 * documented that no one knows which bars are which 6241 * so the developer has to map all bars. This hack 6242 * allows shoddy garbage to use MSI-X in this framework. 6243 */ 6244 if (bar != -1) { 6245 ctx->ifc_msix_mem = bus_alloc_resource_any(dev, 6246 SYS_RES_MEMORY, &bar, RF_ACTIVE); 6247 if (ctx->ifc_msix_mem == NULL) { 6248 device_printf(dev, "Unable to map MSI-X table\n"); 6249 goto msi; 6250 } 6251 } 6252 6253 admincnt = sctx->isc_admin_intrcnt; 6254 #if IFLIB_DEBUG 6255 /* use only 1 qset in debug mode */ 6256 queuemsgs = min(msgs - admincnt, 1); 6257 #else 6258 queuemsgs = msgs - admincnt; 6259 #endif 6260 #ifdef RSS 6261 queues = imin(queuemsgs, rss_getnumbuckets()); 6262 #else 6263 queues = queuemsgs; 6264 #endif 6265 queues = imin(CPU_COUNT(&ctx->ifc_cpus), queues); 6266 if (bootverbose) 6267 device_printf(dev, 6268 "intr CPUs: %d queue msgs: %d admincnt: %d\n", 6269 CPU_COUNT(&ctx->ifc_cpus), queuemsgs, admincnt); 6270 #ifdef RSS 6271 /* If we're doing RSS, clamp at the number of RSS buckets */ 6272 if (queues > rss_getnumbuckets()) 6273 queues = rss_getnumbuckets(); 6274 #endif 6275 if (iflib_num_rx_queues > 0 && iflib_num_rx_queues < queuemsgs - admincnt) 6276 rx_queues = iflib_num_rx_queues; 6277 else 6278 rx_queues = queues; 6279 6280 if (rx_queues > scctx->isc_nrxqsets) 6281 rx_queues = scctx->isc_nrxqsets; 6282 6283 /* 6284 * We want this to be all logical CPUs by default 6285 */ 6286 if (iflib_num_tx_queues > 0 && iflib_num_tx_queues < queues) 6287 tx_queues = iflib_num_tx_queues; 6288 else 6289 tx_queues = mp_ncpus; 6290 6291 if (tx_queues > scctx->isc_ntxqsets) 6292 tx_queues = scctx->isc_ntxqsets; 6293 6294 if (ctx->ifc_sysctl_qs_eq_override == 0) { 6295 #ifdef INVARIANTS 6296 if (tx_queues != rx_queues) 6297 device_printf(dev, 6298 "queue equality override not set, capping rx_queues at %d and tx_queues at %d\n", 6299 min(rx_queues, tx_queues), min(rx_queues, tx_queues)); 6300 #endif 6301 tx_queues = min(rx_queues, tx_queues); 6302 rx_queues = min(rx_queues, tx_queues); 6303 } 6304 6305 vectors = rx_queues + admincnt; 6306 if (msgs < vectors) { 6307 device_printf(dev, 6308 "insufficient number of MSI-X vectors " 6309 "(supported %d, need %d)\n", msgs, vectors); 6310 goto msi; 6311 } 6312 6313 device_printf(dev, "Using %d RX queues %d TX queues\n", rx_queues, 6314 tx_queues); 6315 msgs = vectors; 6316 if ((err = pci_alloc_msix(dev, &vectors)) == 0) { 6317 if (vectors != msgs) { 6318 device_printf(dev, 6319 "Unable to allocate sufficient MSI-X vectors " 6320 "(got %d, need %d)\n", vectors, msgs); 6321 pci_release_msi(dev); 6322 if (bar != -1) { 6323 bus_release_resource(dev, SYS_RES_MEMORY, bar, 6324 ctx->ifc_msix_mem); 6325 ctx->ifc_msix_mem = NULL; 6326 } 6327 goto msi; 6328 } 6329 device_printf(dev, "Using MSI-X interrupts with %d vectors\n", 6330 vectors); 6331 scctx->isc_vectors = vectors; 6332 scctx->isc_nrxqsets = rx_queues; 6333 scctx->isc_ntxqsets = tx_queues; 6334 scctx->isc_intr = IFLIB_INTR_MSIX; 6335 6336 return (vectors); 6337 } else { 6338 device_printf(dev, 6339 "failed to allocate %d MSI-X vectors, err: %d\n", vectors, 6340 err); 6341 if (bar != -1) { 6342 bus_release_resource(dev, SYS_RES_MEMORY, bar, 6343 ctx->ifc_msix_mem); 6344 ctx->ifc_msix_mem = NULL; 6345 } 6346 } 6347 6348 msi: 6349 vectors = pci_msi_count(dev); 6350 scctx->isc_nrxqsets = 1; 6351 scctx->isc_ntxqsets = 1; 6352 scctx->isc_vectors = vectors; 6353 if (vectors == 1 && pci_alloc_msi(dev, &vectors) == 0) { 6354 device_printf(dev,"Using an MSI interrupt\n"); 6355 scctx->isc_intr = IFLIB_INTR_MSI; 6356 } else { 6357 scctx->isc_vectors = 1; 6358 device_printf(dev,"Using a Legacy interrupt\n"); 6359 scctx->isc_intr = IFLIB_INTR_LEGACY; 6360 } 6361 6362 return (vectors); 6363 } 6364 6365 static const char *ring_states[] = { "IDLE", "BUSY", "STALLED", "ABDICATED" }; 6366 6367 static int 6368 mp_ring_state_handler(SYSCTL_HANDLER_ARGS) 6369 { 6370 int rc; 6371 uint16_t *state = ((uint16_t *)oidp->oid_arg1); 6372 struct sbuf *sb; 6373 const char *ring_state = "UNKNOWN"; 6374 6375 /* XXX needed ? */ 6376 rc = sysctl_wire_old_buffer(req, 0); 6377 MPASS(rc == 0); 6378 if (rc != 0) 6379 return (rc); 6380 sb = sbuf_new_for_sysctl(NULL, NULL, 80, req); 6381 MPASS(sb != NULL); 6382 if (sb == NULL) 6383 return (ENOMEM); 6384 if (state[3] <= 3) 6385 ring_state = ring_states[state[3]]; 6386 6387 sbuf_printf(sb, "pidx_head: %04hd pidx_tail: %04hd cidx: %04hd state: %s", 6388 state[0], state[1], state[2], ring_state); 6389 rc = sbuf_finish(sb); 6390 sbuf_delete(sb); 6391 return(rc); 6392 } 6393 6394 enum iflib_ndesc_handler { 6395 IFLIB_NTXD_HANDLER, 6396 IFLIB_NRXD_HANDLER, 6397 }; 6398 6399 static int 6400 mp_ndesc_handler(SYSCTL_HANDLER_ARGS) 6401 { 6402 if_ctx_t ctx = (void *)arg1; 6403 enum iflib_ndesc_handler type = arg2; 6404 char buf[256] = {0}; 6405 qidx_t *ndesc; 6406 char *p, *next; 6407 int nqs, rc, i; 6408 6409 nqs = 8; 6410 switch(type) { 6411 case IFLIB_NTXD_HANDLER: 6412 ndesc = ctx->ifc_sysctl_ntxds; 6413 if (ctx->ifc_sctx) 6414 nqs = ctx->ifc_sctx->isc_ntxqs; 6415 break; 6416 case IFLIB_NRXD_HANDLER: 6417 ndesc = ctx->ifc_sysctl_nrxds; 6418 if (ctx->ifc_sctx) 6419 nqs = ctx->ifc_sctx->isc_nrxqs; 6420 break; 6421 default: 6422 printf("%s: unhandled type\n", __func__); 6423 return (EINVAL); 6424 } 6425 if (nqs == 0) 6426 nqs = 8; 6427 6428 for (i=0; i<8; i++) { 6429 if (i >= nqs) 6430 break; 6431 if (i) 6432 strcat(buf, ","); 6433 sprintf(strchr(buf, 0), "%d", ndesc[i]); 6434 } 6435 6436 rc = sysctl_handle_string(oidp, buf, sizeof(buf), req); 6437 if (rc || req->newptr == NULL) 6438 return rc; 6439 6440 for (i = 0, next = buf, p = strsep(&next, " ,"); i < 8 && p; 6441 i++, p = strsep(&next, " ,")) { 6442 ndesc[i] = strtoul(p, NULL, 10); 6443 } 6444 6445 return(rc); 6446 } 6447 6448 #define NAME_BUFLEN 32 6449 static void 6450 iflib_add_device_sysctl_pre(if_ctx_t ctx) 6451 { 6452 device_t dev = iflib_get_dev(ctx); 6453 struct sysctl_oid_list *child, *oid_list; 6454 struct sysctl_ctx_list *ctx_list; 6455 struct sysctl_oid *node; 6456 6457 ctx_list = device_get_sysctl_ctx(dev); 6458 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev)); 6459 ctx->ifc_sysctl_node = node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, "iflib", 6460 CTLFLAG_RD, NULL, "IFLIB fields"); 6461 oid_list = SYSCTL_CHILDREN(node); 6462 6463 SYSCTL_ADD_CONST_STRING(ctx_list, oid_list, OID_AUTO, "driver_version", 6464 CTLFLAG_RD, ctx->ifc_sctx->isc_driver_version, 6465 "driver version"); 6466 6467 SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_ntxqs", 6468 CTLFLAG_RWTUN, &ctx->ifc_sysctl_ntxqs, 0, 6469 "# of txqs to use, 0 => use default #"); 6470 SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_nrxqs", 6471 CTLFLAG_RWTUN, &ctx->ifc_sysctl_nrxqs, 0, 6472 "# of rxqs to use, 0 => use default #"); 6473 SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_qs_enable", 6474 CTLFLAG_RWTUN, &ctx->ifc_sysctl_qs_eq_override, 0, 6475 "permit #txq != #rxq"); 6476 SYSCTL_ADD_INT(ctx_list, oid_list, OID_AUTO, "disable_msix", 6477 CTLFLAG_RWTUN, &ctx->ifc_softc_ctx.isc_disable_msix, 0, 6478 "disable MSI-X (default 0)"); 6479 SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "rx_budget", 6480 CTLFLAG_RWTUN, &ctx->ifc_sysctl_rx_budget, 0, 6481 "set the RX budget"); 6482 SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "tx_abdicate", 6483 CTLFLAG_RWTUN, &ctx->ifc_sysctl_tx_abdicate, 0, 6484 "cause TX to abdicate instead of running to completion"); 6485 ctx->ifc_sysctl_core_offset = CORE_OFFSET_UNSPECIFIED; 6486 SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "core_offset", 6487 CTLFLAG_RDTUN, &ctx->ifc_sysctl_core_offset, 0, 6488 "offset to start using cores at"); 6489 SYSCTL_ADD_U8(ctx_list, oid_list, OID_AUTO, "separate_txrx", 6490 CTLFLAG_RDTUN, &ctx->ifc_sysctl_separate_txrx, 0, 6491 "use separate cores for TX and RX"); 6492 6493 /* XXX change for per-queue sizes */ 6494 SYSCTL_ADD_PROC(ctx_list, oid_list, OID_AUTO, "override_ntxds", 6495 CTLTYPE_STRING|CTLFLAG_RWTUN, ctx, IFLIB_NTXD_HANDLER, 6496 mp_ndesc_handler, "A", 6497 "list of # of TX descriptors to use, 0 = use default #"); 6498 SYSCTL_ADD_PROC(ctx_list, oid_list, OID_AUTO, "override_nrxds", 6499 CTLTYPE_STRING|CTLFLAG_RWTUN, ctx, IFLIB_NRXD_HANDLER, 6500 mp_ndesc_handler, "A", 6501 "list of # of RX descriptors to use, 0 = use default #"); 6502 } 6503 6504 static void 6505 iflib_add_device_sysctl_post(if_ctx_t ctx) 6506 { 6507 if_shared_ctx_t sctx = ctx->ifc_sctx; 6508 if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; 6509 device_t dev = iflib_get_dev(ctx); 6510 struct sysctl_oid_list *child; 6511 struct sysctl_ctx_list *ctx_list; 6512 iflib_fl_t fl; 6513 iflib_txq_t txq; 6514 iflib_rxq_t rxq; 6515 int i, j; 6516 char namebuf[NAME_BUFLEN]; 6517 char *qfmt; 6518 struct sysctl_oid *queue_node, *fl_node, *node; 6519 struct sysctl_oid_list *queue_list, *fl_list; 6520 ctx_list = device_get_sysctl_ctx(dev); 6521 6522 node = ctx->ifc_sysctl_node; 6523 child = SYSCTL_CHILDREN(node); 6524 6525 if (scctx->isc_ntxqsets > 100) 6526 qfmt = "txq%03d"; 6527 else if (scctx->isc_ntxqsets > 10) 6528 qfmt = "txq%02d"; 6529 else 6530 qfmt = "txq%d"; 6531 for (i = 0, txq = ctx->ifc_txqs; i < scctx->isc_ntxqsets; i++, txq++) { 6532 snprintf(namebuf, NAME_BUFLEN, qfmt, i); 6533 queue_node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, namebuf, 6534 CTLFLAG_RD, NULL, "Queue Name"); 6535 queue_list = SYSCTL_CHILDREN(queue_node); 6536 #if MEMORY_LOGGING 6537 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_dequeued", 6538 CTLFLAG_RD, 6539 &txq->ift_dequeued, "total mbufs freed"); 6540 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_enqueued", 6541 CTLFLAG_RD, 6542 &txq->ift_enqueued, "total mbufs enqueued"); 6543 #endif 6544 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "mbuf_defrag", 6545 CTLFLAG_RD, 6546 &txq->ift_mbuf_defrag, "# of times m_defrag was called"); 6547 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "m_pullups", 6548 CTLFLAG_RD, 6549 &txq->ift_pullups, "# of times m_pullup was called"); 6550 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "mbuf_defrag_failed", 6551 CTLFLAG_RD, 6552 &txq->ift_mbuf_defrag_failed, "# of times m_defrag failed"); 6553 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "no_desc_avail", 6554 CTLFLAG_RD, 6555 &txq->ift_no_desc_avail, "# of times no descriptors were available"); 6556 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "tx_map_failed", 6557 CTLFLAG_RD, 6558 &txq->ift_map_failed, "# of times DMA map failed"); 6559 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txd_encap_efbig", 6560 CTLFLAG_RD, 6561 &txq->ift_txd_encap_efbig, "# of times txd_encap returned EFBIG"); 6562 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "no_tx_dma_setup", 6563 CTLFLAG_RD, 6564 &txq->ift_no_tx_dma_setup, "# of times map failed for other than EFBIG"); 6565 SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_pidx", 6566 CTLFLAG_RD, 6567 &txq->ift_pidx, 1, "Producer Index"); 6568 SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_cidx", 6569 CTLFLAG_RD, 6570 &txq->ift_cidx, 1, "Consumer Index"); 6571 SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_cidx_processed", 6572 CTLFLAG_RD, 6573 &txq->ift_cidx_processed, 1, "Consumer Index seen by credit update"); 6574 SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_in_use", 6575 CTLFLAG_RD, 6576 &txq->ift_in_use, 1, "descriptors in use"); 6577 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_processed", 6578 CTLFLAG_RD, 6579 &txq->ift_processed, "descriptors procesed for clean"); 6580 SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_cleaned", 6581 CTLFLAG_RD, 6582 &txq->ift_cleaned, "total cleaned"); 6583 SYSCTL_ADD_PROC(ctx_list, queue_list, OID_AUTO, "ring_state", 6584 CTLTYPE_STRING | CTLFLAG_RD, __DEVOLATILE(uint64_t *, &txq->ift_br->state), 6585 0, mp_ring_state_handler, "A", "soft ring state"); 6586 SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_enqueues", 6587 CTLFLAG_RD, &txq->ift_br->enqueues, 6588 "# of enqueues to the mp_ring for this queue"); 6589 SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_drops", 6590 CTLFLAG_RD, &txq->ift_br->drops, 6591 "# of drops in the mp_ring for this queue"); 6592 SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_starts", 6593 CTLFLAG_RD, &txq->ift_br->starts, 6594 "# of normal consumer starts in the mp_ring for this queue"); 6595 SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_stalls", 6596 CTLFLAG_RD, &txq->ift_br->stalls, 6597 "# of consumer stalls in the mp_ring for this queue"); 6598 SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_restarts", 6599 CTLFLAG_RD, &txq->ift_br->restarts, 6600 "# of consumer restarts in the mp_ring for this queue"); 6601 SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_abdications", 6602 CTLFLAG_RD, &txq->ift_br->abdications, 6603 "# of consumer abdications in the mp_ring for this queue"); 6604 } 6605 6606 if (scctx->isc_nrxqsets > 100) 6607 qfmt = "rxq%03d"; 6608 else if (scctx->isc_nrxqsets > 10) 6609 qfmt = "rxq%02d"; 6610 else 6611 qfmt = "rxq%d"; 6612 for (i = 0, rxq = ctx->ifc_rxqs; i < scctx->isc_nrxqsets; i++, rxq++) { 6613 snprintf(namebuf, NAME_BUFLEN, qfmt, i); 6614 queue_node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, namebuf, 6615 CTLFLAG_RD, NULL, "Queue Name"); 6616 queue_list = SYSCTL_CHILDREN(queue_node); 6617 if (sctx->isc_flags & IFLIB_HAS_RXCQ) { 6618 SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "rxq_cq_cidx", 6619 CTLFLAG_RD, 6620 &rxq->ifr_cq_cidx, 1, "Consumer Index"); 6621 } 6622 6623 for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) { 6624 snprintf(namebuf, NAME_BUFLEN, "rxq_fl%d", j); 6625 fl_node = SYSCTL_ADD_NODE(ctx_list, queue_list, OID_AUTO, namebuf, 6626 CTLFLAG_RD, NULL, "freelist Name"); 6627 fl_list = SYSCTL_CHILDREN(fl_node); 6628 SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "pidx", 6629 CTLFLAG_RD, 6630 &fl->ifl_pidx, 1, "Producer Index"); 6631 SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "cidx", 6632 CTLFLAG_RD, 6633 &fl->ifl_cidx, 1, "Consumer Index"); 6634 SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "credits", 6635 CTLFLAG_RD, 6636 &fl->ifl_credits, 1, "credits available"); 6637 #if MEMORY_LOGGING 6638 SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_m_enqueued", 6639 CTLFLAG_RD, 6640 &fl->ifl_m_enqueued, "mbufs allocated"); 6641 SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_m_dequeued", 6642 CTLFLAG_RD, 6643 &fl->ifl_m_dequeued, "mbufs freed"); 6644 SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_cl_enqueued", 6645 CTLFLAG_RD, 6646 &fl->ifl_cl_enqueued, "clusters allocated"); 6647 SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_cl_dequeued", 6648 CTLFLAG_RD, 6649 &fl->ifl_cl_dequeued, "clusters freed"); 6650 #endif 6651 6652 } 6653 } 6654 6655 } 6656 6657 void 6658 iflib_request_reset(if_ctx_t ctx) 6659 { 6660 6661 STATE_LOCK(ctx); 6662 ctx->ifc_flags |= IFC_DO_RESET; 6663 STATE_UNLOCK(ctx); 6664 } 6665 6666 #ifndef __NO_STRICT_ALIGNMENT 6667 static struct mbuf * 6668 iflib_fixup_rx(struct mbuf *m) 6669 { 6670 struct mbuf *n; 6671 6672 if (m->m_len <= (MCLBYTES - ETHER_HDR_LEN)) { 6673 bcopy(m->m_data, m->m_data + ETHER_HDR_LEN, m->m_len); 6674 m->m_data += ETHER_HDR_LEN; 6675 n = m; 6676 } else { 6677 MGETHDR(n, M_NOWAIT, MT_DATA); 6678 if (n == NULL) { 6679 m_freem(m); 6680 return (NULL); 6681 } 6682 bcopy(m->m_data, n->m_data, ETHER_HDR_LEN); 6683 m->m_data += ETHER_HDR_LEN; 6684 m->m_len -= ETHER_HDR_LEN; 6685 n->m_len = ETHER_HDR_LEN; 6686 M_MOVE_PKTHDR(n, m); 6687 n->m_next = m; 6688 } 6689 return (n); 6690 } 6691 #endif 6692 6693 #ifdef NETDUMP 6694 static void 6695 iflib_netdump_init(if_t ifp, int *nrxr, int *ncl, int *clsize) 6696 { 6697 if_ctx_t ctx; 6698 6699 ctx = if_getsoftc(ifp); 6700 CTX_LOCK(ctx); 6701 *nrxr = NRXQSETS(ctx); 6702 *ncl = ctx->ifc_rxqs[0].ifr_fl->ifl_size; 6703 *clsize = ctx->ifc_rxqs[0].ifr_fl->ifl_buf_size; 6704 CTX_UNLOCK(ctx); 6705 } 6706 6707 static void 6708 iflib_netdump_event(if_t ifp, enum netdump_ev event) 6709 { 6710 if_ctx_t ctx; 6711 if_softc_ctx_t scctx; 6712 iflib_fl_t fl; 6713 iflib_rxq_t rxq; 6714 int i, j; 6715 6716 ctx = if_getsoftc(ifp); 6717 scctx = &ctx->ifc_softc_ctx; 6718 6719 switch (event) { 6720 case NETDUMP_START: 6721 for (i = 0; i < scctx->isc_nrxqsets; i++) { 6722 rxq = &ctx->ifc_rxqs[i]; 6723 for (j = 0; j < rxq->ifr_nfl; j++) { 6724 fl = rxq->ifr_fl; 6725 fl->ifl_zone = m_getzone(fl->ifl_buf_size); 6726 } 6727 } 6728 iflib_no_tx_batch = 1; 6729 break; 6730 default: 6731 break; 6732 } 6733 } 6734 6735 static int 6736 iflib_netdump_transmit(if_t ifp, struct mbuf *m) 6737 { 6738 if_ctx_t ctx; 6739 iflib_txq_t txq; 6740 int error; 6741 6742 ctx = if_getsoftc(ifp); 6743 if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != 6744 IFF_DRV_RUNNING) 6745 return (EBUSY); 6746 6747 txq = &ctx->ifc_txqs[0]; 6748 error = iflib_encap(txq, &m); 6749 if (error == 0) 6750 (void)iflib_txd_db_check(ctx, txq, true, txq->ift_in_use); 6751 return (error); 6752 } 6753 6754 static int 6755 iflib_netdump_poll(if_t ifp, int count) 6756 { 6757 if_ctx_t ctx; 6758 if_softc_ctx_t scctx; 6759 iflib_txq_t txq; 6760 int i; 6761 6762 ctx = if_getsoftc(ifp); 6763 scctx = &ctx->ifc_softc_ctx; 6764 6765 if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != 6766 IFF_DRV_RUNNING) 6767 return (EBUSY); 6768 6769 txq = &ctx->ifc_txqs[0]; 6770 (void)iflib_completed_tx_reclaim(txq, RECLAIM_THRESH(ctx)); 6771 6772 for (i = 0; i < scctx->isc_nrxqsets; i++) 6773 (void)iflib_rxeof(&ctx->ifc_rxqs[i], 16 /* XXX */); 6774 return (0); 6775 } 6776 #endif /* NETDUMP */ 6777