1 /*- 2 * Copyright (c) 2011 Chelsio Communications, Inc. 3 * All rights reserved. 4 * Written by: Navdeep Parhar <np@FreeBSD.org> 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 * 27 * $FreeBSD$ 28 * 29 */ 30 31 #ifndef __T4_ADAPTER_H__ 32 #define __T4_ADAPTER_H__ 33 34 #include <sys/kernel.h> 35 #include <sys/bus.h> 36 #include <sys/rman.h> 37 #include <sys/types.h> 38 #include <sys/lock.h> 39 #include <sys/malloc.h> 40 #include <sys/rwlock.h> 41 #include <sys/sx.h> 42 #include <vm/uma.h> 43 44 #include <dev/pci/pcivar.h> 45 #include <dev/pci/pcireg.h> 46 #include <machine/bus.h> 47 #include <sys/socket.h> 48 #include <sys/sysctl.h> 49 #include <net/ethernet.h> 50 #include <net/if.h> 51 #include <net/if_var.h> 52 #include <net/if_media.h> 53 #include <netinet/in.h> 54 #include <netinet/tcp_lro.h> 55 56 #include "offload.h" 57 #include "common/t4_msg.h" 58 #include "firmware/t4fw_interface.h" 59 60 MALLOC_DECLARE(M_CXGBE); 61 #define CXGBE_UNIMPLEMENTED(s) \ 62 panic("%s (%s, line %d) not implemented yet.", s, __FILE__, __LINE__) 63 64 #if defined(__i386__) || defined(__amd64__) 65 static __inline void 66 prefetch(void *x) 67 { 68 __asm volatile("prefetcht0 %0" :: "m" (*(unsigned long *)x)); 69 } 70 #else 71 #define prefetch(x) 72 #endif 73 74 #ifndef SYSCTL_ADD_UQUAD 75 #define SYSCTL_ADD_UQUAD SYSCTL_ADD_QUAD 76 #define sysctl_handle_64 sysctl_handle_quad 77 #define CTLTYPE_U64 CTLTYPE_QUAD 78 #endif 79 80 #if (__FreeBSD_version >= 900030) || \ 81 ((__FreeBSD_version >= 802507) && (__FreeBSD_version < 900000)) 82 #define SBUF_DRAIN 1 83 #endif 84 85 #ifdef __amd64__ 86 /* XXX: need systemwide bus_space_read_8/bus_space_write_8 */ 87 static __inline uint64_t 88 t4_bus_space_read_8(bus_space_tag_t tag, bus_space_handle_t handle, 89 bus_size_t offset) 90 { 91 KASSERT(tag == X86_BUS_SPACE_MEM, 92 ("%s: can only handle mem space", __func__)); 93 94 return (*(volatile uint64_t *)(handle + offset)); 95 } 96 97 static __inline void 98 t4_bus_space_write_8(bus_space_tag_t tag, bus_space_handle_t bsh, 99 bus_size_t offset, uint64_t value) 100 { 101 KASSERT(tag == X86_BUS_SPACE_MEM, 102 ("%s: can only handle mem space", __func__)); 103 104 *(volatile uint64_t *)(bsh + offset) = value; 105 } 106 #else 107 static __inline uint64_t 108 t4_bus_space_read_8(bus_space_tag_t tag, bus_space_handle_t handle, 109 bus_size_t offset) 110 { 111 return (uint64_t)bus_space_read_4(tag, handle, offset) + 112 ((uint64_t)bus_space_read_4(tag, handle, offset + 4) << 32); 113 } 114 115 static __inline void 116 t4_bus_space_write_8(bus_space_tag_t tag, bus_space_handle_t bsh, 117 bus_size_t offset, uint64_t value) 118 { 119 bus_space_write_4(tag, bsh, offset, value); 120 bus_space_write_4(tag, bsh, offset + 4, value >> 32); 121 } 122 #endif 123 124 struct adapter; 125 typedef struct adapter adapter_t; 126 127 enum { 128 /* 129 * All ingress queues use this entry size. Note that the firmware event 130 * queue and any iq expecting CPL_RX_PKT in the descriptor needs this to 131 * be at least 64. 132 */ 133 IQ_ESIZE = 64, 134 135 /* Default queue sizes for all kinds of ingress queues */ 136 FW_IQ_QSIZE = 256, 137 RX_IQ_QSIZE = 1024, 138 139 /* All egress queues use this entry size */ 140 EQ_ESIZE = 64, 141 142 /* Default queue sizes for all kinds of egress queues */ 143 CTRL_EQ_QSIZE = 128, 144 TX_EQ_QSIZE = 1024, 145 146 #if MJUMPAGESIZE != MCLBYTES 147 SW_ZONE_SIZES = 4, /* cluster, jumbop, jumbo9k, jumbo16k */ 148 #else 149 SW_ZONE_SIZES = 3, /* cluster, jumbo9k, jumbo16k */ 150 #endif 151 CL_METADATA_SIZE = CACHE_LINE_SIZE, 152 153 SGE_MAX_WR_NDESC = SGE_MAX_WR_LEN / EQ_ESIZE, /* max WR size in desc */ 154 TX_SGL_SEGS = 36, 155 TX_WR_FLITS = SGE_MAX_WR_LEN / 8 156 }; 157 158 enum { 159 /* adapter intr_type */ 160 INTR_INTX = (1 << 0), 161 INTR_MSI = (1 << 1), 162 INTR_MSIX = (1 << 2) 163 }; 164 165 enum { 166 XGMAC_MTU = (1 << 0), 167 XGMAC_PROMISC = (1 << 1), 168 XGMAC_ALLMULTI = (1 << 2), 169 XGMAC_VLANEX = (1 << 3), 170 XGMAC_UCADDR = (1 << 4), 171 XGMAC_MCADDRS = (1 << 5), 172 173 XGMAC_ALL = 0xffff 174 }; 175 176 enum { 177 /* flags understood by begin_synchronized_op */ 178 HOLD_LOCK = (1 << 0), 179 SLEEP_OK = (1 << 1), 180 INTR_OK = (1 << 2), 181 182 /* flags understood by end_synchronized_op */ 183 LOCK_HELD = HOLD_LOCK, 184 }; 185 186 enum { 187 /* adapter flags */ 188 FULL_INIT_DONE = (1 << 0), 189 FW_OK = (1 << 1), 190 /* INTR_DIRECT = (1 << 2), No longer used. */ 191 MASTER_PF = (1 << 3), 192 ADAP_SYSCTL_CTX = (1 << 4), 193 TOM_INIT_DONE = (1 << 5), 194 BUF_PACKING_OK = (1 << 6), 195 196 CXGBE_BUSY = (1 << 9), 197 198 /* port flags */ 199 DOOMED = (1 << 0), 200 PORT_INIT_DONE = (1 << 1), 201 PORT_SYSCTL_CTX = (1 << 2), 202 HAS_TRACEQ = (1 << 3), 203 INTR_RXQ = (1 << 4), /* All NIC rxq's take interrupts */ 204 INTR_OFLD_RXQ = (1 << 5), /* All TOE rxq's take interrupts */ 205 INTR_NM_RXQ = (1 << 6), /* All netmap rxq's take interrupts */ 206 INTR_ALL = (INTR_RXQ | INTR_OFLD_RXQ | INTR_NM_RXQ), 207 }; 208 209 #define IS_DOOMED(pi) ((pi)->flags & DOOMED) 210 #define SET_DOOMED(pi) do {(pi)->flags |= DOOMED;} while (0) 211 #define IS_BUSY(sc) ((sc)->flags & CXGBE_BUSY) 212 #define SET_BUSY(sc) do {(sc)->flags |= CXGBE_BUSY;} while (0) 213 #define CLR_BUSY(sc) do {(sc)->flags &= ~CXGBE_BUSY;} while (0) 214 215 struct port_info { 216 device_t dev; 217 struct adapter *adapter; 218 219 struct ifnet *ifp; 220 struct ifmedia media; 221 222 struct mtx pi_lock; 223 char lockname[16]; 224 unsigned long flags; 225 int if_flags; 226 227 uint16_t *rss; 228 uint16_t viid; 229 int16_t xact_addr_filt;/* index of exact MAC address filter */ 230 uint16_t rss_size; /* size of VI's RSS table slice */ 231 uint8_t lport; /* associated offload logical port */ 232 int8_t mdio_addr; 233 uint8_t port_type; 234 uint8_t mod_type; 235 uint8_t port_id; 236 uint8_t tx_chan; 237 uint8_t rx_chan_map; /* rx MPS channel bitmap */ 238 239 /* These need to be int as they are used in sysctl */ 240 int ntxq; /* # of tx queues */ 241 int first_txq; /* index of first tx queue */ 242 int rsrv_noflowq; /* Reserve queue 0 for non-flowid packets */ 243 int nrxq; /* # of rx queues */ 244 int first_rxq; /* index of first rx queue */ 245 #ifdef TCP_OFFLOAD 246 int nofldtxq; /* # of offload tx queues */ 247 int first_ofld_txq; /* index of first offload tx queue */ 248 int nofldrxq; /* # of offload rx queues */ 249 int first_ofld_rxq; /* index of first offload rx queue */ 250 #endif 251 #ifdef DEV_NETMAP 252 int nnmtxq; /* # of netmap tx queues */ 253 int first_nm_txq; /* index of first netmap tx queue */ 254 int nnmrxq; /* # of netmap rx queues */ 255 int first_nm_rxq; /* index of first netmap rx queue */ 256 257 struct ifnet *nm_ifp; 258 struct ifmedia nm_media; 259 int nmif_flags; 260 uint16_t nm_viid; 261 int16_t nm_xact_addr_filt; 262 uint16_t nm_rss_size; /* size of netmap VI's RSS table slice */ 263 #endif 264 int tmr_idx; 265 int pktc_idx; 266 int qsize_rxq; 267 int qsize_txq; 268 269 int linkdnrc; 270 struct link_config link_cfg; 271 272 struct timeval last_refreshed; 273 struct port_stats stats; 274 u_int tnl_cong_drops; 275 276 eventhandler_tag vlan_c; 277 278 struct callout tick; 279 struct sysctl_ctx_list ctx; /* from ifconfig up to driver detach */ 280 281 uint8_t hw_addr[ETHER_ADDR_LEN]; /* factory MAC address, won't change */ 282 }; 283 284 /* Where the cluster came from, how it has been carved up. */ 285 struct cluster_layout { 286 int8_t zidx; 287 int8_t hwidx; 288 uint16_t region1; /* mbufs laid out within this region */ 289 /* region2 is the DMA region */ 290 uint16_t region3; /* cluster_metadata within this region */ 291 }; 292 293 struct cluster_metadata { 294 u_int refcount; 295 #ifdef INVARIANTS 296 struct fl_sdesc *sd; /* For debug only. Could easily be stale */ 297 #endif 298 }; 299 300 struct fl_sdesc { 301 caddr_t cl; 302 uint16_t nmbuf; /* # of driver originated mbufs with ref on cluster */ 303 struct cluster_layout cll; 304 }; 305 306 struct tx_desc { 307 __be64 flit[8]; 308 }; 309 310 struct tx_map { 311 struct mbuf *m; 312 bus_dmamap_t map; 313 }; 314 315 /* DMA maps used for tx */ 316 struct tx_maps { 317 struct tx_map *maps; 318 uint32_t map_total; /* # of DMA maps */ 319 uint32_t map_pidx; /* next map to be used */ 320 uint32_t map_cidx; /* reclaimed up to this index */ 321 uint32_t map_avail; /* # of available maps */ 322 }; 323 324 struct tx_sdesc { 325 uint8_t desc_used; /* # of hardware descriptors used by the WR */ 326 uint8_t credits; /* NIC txq: # of frames sent out in the WR */ 327 }; 328 329 330 #define IQ_PAD (IQ_ESIZE - sizeof(struct rsp_ctrl) - sizeof(struct rss_header)) 331 struct iq_desc { 332 struct rss_header rss; 333 uint8_t cpl[IQ_PAD]; 334 struct rsp_ctrl rsp; 335 }; 336 #undef IQ_PAD 337 CTASSERT(sizeof(struct iq_desc) == IQ_ESIZE); 338 339 enum { 340 /* iq flags */ 341 IQ_ALLOCATED = (1 << 0), /* firmware resources allocated */ 342 IQ_HAS_FL = (1 << 1), /* iq associated with a freelist */ 343 IQ_INTR = (1 << 2), /* iq takes direct interrupt */ 344 IQ_LRO_ENABLED = (1 << 3), /* iq is an eth rxq with LRO enabled */ 345 346 /* iq state */ 347 IQS_DISABLED = 0, 348 IQS_BUSY = 1, 349 IQS_IDLE = 2, 350 }; 351 352 /* 353 * Ingress Queue: T4 is producer, driver is consumer. 354 */ 355 struct sge_iq { 356 uint32_t flags; 357 volatile int state; 358 struct adapter *adapter; 359 struct iq_desc *desc; /* KVA of descriptor ring */ 360 int8_t intr_pktc_idx; /* packet count threshold index */ 361 uint8_t gen; /* generation bit */ 362 uint8_t intr_params; /* interrupt holdoff parameters */ 363 uint8_t intr_next; /* XXX: holdoff for next interrupt */ 364 uint16_t qsize; /* size (# of entries) of the queue */ 365 uint16_t sidx; /* index of the entry with the status page */ 366 uint16_t cidx; /* consumer index */ 367 uint16_t cntxt_id; /* SGE context id for the iq */ 368 uint16_t abs_id; /* absolute SGE id for the iq */ 369 370 STAILQ_ENTRY(sge_iq) link; 371 372 bus_dma_tag_t desc_tag; 373 bus_dmamap_t desc_map; 374 bus_addr_t ba; /* bus address of descriptor ring */ 375 }; 376 377 enum { 378 EQ_CTRL = 1, 379 EQ_ETH = 2, 380 #ifdef TCP_OFFLOAD 381 EQ_OFLD = 3, 382 #endif 383 384 /* eq flags */ 385 EQ_TYPEMASK = 7, /* 3 lsbits hold the type */ 386 EQ_ALLOCATED = (1 << 3), /* firmware resources allocated */ 387 EQ_DOOMED = (1 << 4), /* about to be destroyed */ 388 EQ_CRFLUSHED = (1 << 5), /* expecting an update from SGE */ 389 EQ_STALLED = (1 << 6), /* out of hw descriptors or dmamaps */ 390 }; 391 392 /* Listed in order of preference. Update t4_sysctls too if you change these */ 393 enum {DOORBELL_UDB, DOORBELL_WCWR, DOORBELL_UDBWC, DOORBELL_KDB}; 394 395 /* 396 * Egress Queue: driver is producer, T4 is consumer. 397 * 398 * Note: A free list is an egress queue (driver produces the buffers and T4 399 * consumes them) but it's special enough to have its own struct (see sge_fl). 400 */ 401 struct sge_eq { 402 unsigned int flags; /* MUST be first */ 403 unsigned int cntxt_id; /* SGE context id for the eq */ 404 bus_dma_tag_t desc_tag; 405 bus_dmamap_t desc_map; 406 char lockname[16]; 407 struct mtx eq_lock; 408 409 struct tx_desc *desc; /* KVA of descriptor ring */ 410 bus_addr_t ba; /* bus address of descriptor ring */ 411 struct sge_qstat *spg; /* status page, for convenience */ 412 uint16_t doorbells; 413 volatile uint32_t *udb; /* KVA of doorbell (lies within BAR2) */ 414 u_int udb_qid; /* relative qid within the doorbell page */ 415 uint16_t cap; /* max # of desc, for convenience */ 416 uint16_t avail; /* available descriptors, for convenience */ 417 uint16_t qsize; /* size (# of entries) of the queue */ 418 uint16_t cidx; /* consumer idx (desc idx) */ 419 uint16_t pidx; /* producer idx (desc idx) */ 420 uint16_t pending; /* # of descriptors used since last doorbell */ 421 uint16_t iqid; /* iq that gets egr_update for the eq */ 422 uint8_t tx_chan; /* tx channel used by the eq */ 423 struct task tx_task; 424 struct callout tx_callout; 425 426 /* stats */ 427 428 uint32_t egr_update; /* # of SGE_EGR_UPDATE notifications for eq */ 429 uint32_t unstalled; /* recovered from stall */ 430 }; 431 432 struct sw_zone_info { 433 uma_zone_t zone; /* zone that this cluster comes from */ 434 int size; /* size of cluster: 2K, 4K, 9K, 16K, etc. */ 435 int type; /* EXT_xxx type of the cluster */ 436 int8_t head_hwidx; 437 int8_t tail_hwidx; 438 }; 439 440 struct hw_buf_info { 441 int8_t zidx; /* backpointer to zone; -ve means unused */ 442 int8_t next; /* next hwidx for this zone; -1 means no more */ 443 int size; 444 }; 445 446 enum { 447 FL_STARVING = (1 << 0), /* on the adapter's list of starving fl's */ 448 FL_DOOMED = (1 << 1), /* about to be destroyed */ 449 FL_BUF_PACKING = (1 << 2), /* buffer packing enabled */ 450 FL_BUF_RESUME = (1 << 3), /* resume from the middle of the frame */ 451 }; 452 453 #define FL_RUNNING_LOW(fl) \ 454 (IDXDIFF(fl->dbidx * 8, fl->cidx, fl->sidx * 8) <= fl->lowat) 455 #define FL_NOT_RUNNING_LOW(fl) \ 456 (IDXDIFF(fl->dbidx * 8, fl->cidx, fl->sidx * 8) >= 2 * fl->lowat) 457 458 struct sge_fl { 459 struct mtx fl_lock; 460 __be64 *desc; /* KVA of descriptor ring, ptr to addresses */ 461 struct fl_sdesc *sdesc; /* KVA of software descriptor ring */ 462 struct cluster_layout cll_def; /* default refill zone, layout */ 463 uint16_t lowat; /* # of buffers <= this means fl needs help */ 464 int flags; 465 uint16_t buf_boundary; 466 467 /* The 16b idx all deal with hw descriptors */ 468 uint16_t dbidx; /* hw pidx after last doorbell */ 469 uint16_t sidx; /* index of status page */ 470 volatile uint16_t hw_cidx; 471 472 /* The 32b idx are all buffer idx, not hardware descriptor idx */ 473 uint32_t cidx; /* consumer index */ 474 uint32_t pidx; /* producer index */ 475 476 uint32_t dbval; 477 u_int rx_offset; /* offset in fl buf (when buffer packing) */ 478 volatile uint32_t *udb; 479 480 uint64_t mbuf_allocated;/* # of mbuf allocated from zone_mbuf */ 481 uint64_t mbuf_inlined; /* # of mbuf created within clusters */ 482 uint64_t cl_allocated; /* # of clusters allocated */ 483 uint64_t cl_recycled; /* # of clusters recycled */ 484 uint64_t cl_fast_recycled; /* # of clusters recycled (fast) */ 485 486 /* These 3 are valid when FL_BUF_RESUME is set, stale otherwise. */ 487 struct mbuf *m0; 488 struct mbuf **pnext; 489 u_int remaining; 490 491 uint16_t qsize; /* # of hw descriptors (status page included) */ 492 uint16_t cntxt_id; /* SGE context id for the freelist */ 493 TAILQ_ENTRY(sge_fl) link; /* All starving freelists */ 494 bus_dma_tag_t desc_tag; 495 bus_dmamap_t desc_map; 496 char lockname[16]; 497 bus_addr_t ba; /* bus address of descriptor ring */ 498 struct cluster_layout cll_alt; /* alternate refill zone, layout */ 499 }; 500 501 /* txq: SGE egress queue + what's needed for Ethernet NIC */ 502 struct sge_txq { 503 struct sge_eq eq; /* MUST be first */ 504 505 struct ifnet *ifp; /* the interface this txq belongs to */ 506 bus_dma_tag_t tx_tag; /* tag for transmit buffers */ 507 struct buf_ring *br; /* tx buffer ring */ 508 struct tx_sdesc *sdesc; /* KVA of software descriptor ring */ 509 struct mbuf *m; /* held up due to temporary resource shortage */ 510 511 struct tx_maps txmaps; 512 513 /* stats for common events first */ 514 515 uint64_t txcsum; /* # of times hardware assisted with checksum */ 516 uint64_t tso_wrs; /* # of TSO work requests */ 517 uint64_t vlan_insertion;/* # of times VLAN tag was inserted */ 518 uint64_t imm_wrs; /* # of work requests with immediate data */ 519 uint64_t sgl_wrs; /* # of work requests with direct SGL */ 520 uint64_t txpkt_wrs; /* # of txpkt work requests (not coalesced) */ 521 uint64_t txpkts_wrs; /* # of coalesced tx work requests */ 522 uint64_t txpkts_pkts; /* # of frames in coalesced tx work requests */ 523 524 /* stats for not-that-common events */ 525 526 uint32_t no_dmamap; /* no DMA map to load the mbuf */ 527 uint32_t no_desc; /* out of hardware descriptors */ 528 } __aligned(CACHE_LINE_SIZE); 529 530 /* rxq: SGE ingress queue + SGE free list + miscellaneous items */ 531 struct sge_rxq { 532 struct sge_iq iq; /* MUST be first */ 533 struct sge_fl fl; /* MUST follow iq */ 534 535 struct ifnet *ifp; /* the interface this rxq belongs to */ 536 #if defined(INET) || defined(INET6) 537 struct lro_ctrl lro; /* LRO state */ 538 #endif 539 540 /* stats for common events first */ 541 542 uint64_t rxcsum; /* # of times hardware assisted with checksum */ 543 uint64_t vlan_extraction;/* # of times VLAN tag was extracted */ 544 545 /* stats for not-that-common events */ 546 547 } __aligned(CACHE_LINE_SIZE); 548 549 static inline struct sge_rxq * 550 iq_to_rxq(struct sge_iq *iq) 551 { 552 553 return (__containerof(iq, struct sge_rxq, iq)); 554 } 555 556 557 #ifdef TCP_OFFLOAD 558 /* ofld_rxq: SGE ingress queue + SGE free list + miscellaneous items */ 559 struct sge_ofld_rxq { 560 struct sge_iq iq; /* MUST be first */ 561 struct sge_fl fl; /* MUST follow iq */ 562 } __aligned(CACHE_LINE_SIZE); 563 564 static inline struct sge_ofld_rxq * 565 iq_to_ofld_rxq(struct sge_iq *iq) 566 { 567 568 return (__containerof(iq, struct sge_ofld_rxq, iq)); 569 } 570 #endif 571 572 struct wrqe { 573 STAILQ_ENTRY(wrqe) link; 574 struct sge_wrq *wrq; 575 int wr_len; 576 uint64_t wr[] __aligned(16); 577 }; 578 579 /* 580 * wrq: SGE egress queue that is given prebuilt work requests. Both the control 581 * and offload tx queues are of this type. 582 */ 583 struct sge_wrq { 584 struct sge_eq eq; /* MUST be first */ 585 586 struct adapter *adapter; 587 588 /* List of WRs held up due to lack of tx descriptors */ 589 STAILQ_HEAD(, wrqe) wr_list; 590 591 /* stats for common events first */ 592 593 uint64_t tx_wrs; /* # of tx work requests */ 594 595 /* stats for not-that-common events */ 596 597 uint32_t no_desc; /* out of hardware descriptors */ 598 } __aligned(CACHE_LINE_SIZE); 599 600 601 #ifdef DEV_NETMAP 602 struct sge_nm_rxq { 603 struct port_info *pi; 604 605 struct iq_desc *iq_desc; 606 uint16_t iq_abs_id; 607 uint16_t iq_cntxt_id; 608 uint16_t iq_cidx; 609 uint16_t iq_sidx; 610 uint8_t iq_gen; 611 612 __be64 *fl_desc; 613 uint16_t fl_cntxt_id; 614 uint32_t fl_cidx; 615 uint32_t fl_pidx; 616 uint32_t fl_sidx; 617 uint32_t fl_db_val; 618 u_int fl_hwidx:4; 619 620 u_int nid; /* netmap ring # for this queue */ 621 622 /* infrequently used items after this */ 623 624 bus_dma_tag_t iq_desc_tag; 625 bus_dmamap_t iq_desc_map; 626 bus_addr_t iq_ba; 627 int intr_idx; 628 629 bus_dma_tag_t fl_desc_tag; 630 bus_dmamap_t fl_desc_map; 631 bus_addr_t fl_ba; 632 } __aligned(CACHE_LINE_SIZE); 633 634 struct sge_nm_txq { 635 struct tx_desc *desc; 636 uint16_t cidx; 637 uint16_t pidx; 638 uint16_t sidx; 639 uint16_t equiqidx; /* EQUIQ last requested at this pidx */ 640 uint16_t equeqidx; /* EQUEQ last requested at this pidx */ 641 uint16_t dbidx; /* pidx of the most recent doorbell */ 642 uint16_t doorbells; 643 volatile uint32_t *udb; 644 u_int udb_qid; 645 u_int cntxt_id; 646 __be32 cpl_ctrl0; /* for convenience */ 647 u_int nid; /* netmap ring # for this queue */ 648 649 /* infrequently used items after this */ 650 651 bus_dma_tag_t desc_tag; 652 bus_dmamap_t desc_map; 653 bus_addr_t ba; 654 int iqidx; 655 } __aligned(CACHE_LINE_SIZE); 656 #endif 657 658 struct sge { 659 int timer_val[SGE_NTIMERS]; 660 int counter_val[SGE_NCOUNTERS]; 661 int fl_starve_threshold; 662 int fl_starve_threshold2; 663 int eq_s_qpp; 664 int iq_s_qpp; 665 666 int nrxq; /* total # of Ethernet rx queues */ 667 int ntxq; /* total # of Ethernet tx tx queues */ 668 #ifdef TCP_OFFLOAD 669 int nofldrxq; /* total # of TOE rx queues */ 670 int nofldtxq; /* total # of TOE tx queues */ 671 #endif 672 #ifdef DEV_NETMAP 673 int nnmrxq; /* total # of netmap rx queues */ 674 int nnmtxq; /* total # of netmap tx queues */ 675 #endif 676 int niq; /* total # of ingress queues */ 677 int neq; /* total # of egress queues */ 678 679 struct sge_iq fwq; /* Firmware event queue */ 680 struct sge_wrq mgmtq; /* Management queue (control queue) */ 681 struct sge_wrq *ctrlq; /* Control queues */ 682 struct sge_txq *txq; /* NIC tx queues */ 683 struct sge_rxq *rxq; /* NIC rx queues */ 684 #ifdef TCP_OFFLOAD 685 struct sge_wrq *ofld_txq; /* TOE tx queues */ 686 struct sge_ofld_rxq *ofld_rxq; /* TOE rx queues */ 687 #endif 688 #ifdef DEV_NETMAP 689 struct sge_nm_txq *nm_txq; /* netmap tx queues */ 690 struct sge_nm_rxq *nm_rxq; /* netmap rx queues */ 691 #endif 692 693 uint16_t iq_start; 694 int eq_start; 695 struct sge_iq **iqmap; /* iq->cntxt_id to iq mapping */ 696 struct sge_eq **eqmap; /* eq->cntxt_id to eq mapping */ 697 698 int pack_boundary; 699 int8_t safe_hwidx1; /* may not have room for metadata */ 700 int8_t safe_hwidx2; /* with room for metadata and maybe more */ 701 struct sw_zone_info sw_zone_info[SW_ZONE_SIZES]; 702 struct hw_buf_info hw_buf_info[SGE_FLBUF_SIZES]; 703 }; 704 705 struct rss_header; 706 typedef int (*cpl_handler_t)(struct sge_iq *, const struct rss_header *, 707 struct mbuf *); 708 typedef int (*an_handler_t)(struct sge_iq *, const struct rsp_ctrl *); 709 typedef int (*fw_msg_handler_t)(struct adapter *, const __be64 *); 710 711 struct adapter { 712 SLIST_ENTRY(adapter) link; 713 device_t dev; 714 struct cdev *cdev; 715 716 /* PCIe register resources */ 717 int regs_rid; 718 struct resource *regs_res; 719 int msix_rid; 720 struct resource *msix_res; 721 bus_space_handle_t bh; 722 bus_space_tag_t bt; 723 bus_size_t mmio_len; 724 int udbs_rid; 725 struct resource *udbs_res; 726 volatile uint8_t *udbs_base; 727 728 unsigned int pf; 729 unsigned int mbox; 730 731 /* Interrupt information */ 732 int intr_type; 733 int intr_count; 734 struct irq { 735 struct resource *res; 736 int rid; 737 void *tag; 738 } *irq; 739 740 bus_dma_tag_t dmat; /* Parent DMA tag */ 741 742 struct sge sge; 743 int lro_timeout; 744 745 struct taskqueue *tq[NCHAN]; /* taskqueues that flush data out */ 746 struct port_info *port[MAX_NPORTS]; 747 uint8_t chan_map[NCHAN]; 748 749 #ifdef TCP_OFFLOAD 750 void *tom_softc; /* (struct tom_data *) */ 751 struct tom_tunables tt; 752 void *iwarp_softc; /* (struct c4iw_dev *) */ 753 void *iscsi_softc; 754 #endif 755 struct l2t_data *l2t; /* L2 table */ 756 struct tid_info tids; 757 758 uint16_t doorbells; 759 int open_device_map; 760 #ifdef TCP_OFFLOAD 761 int offload_map; 762 #endif 763 int flags; 764 765 char ifp_lockname[16]; 766 struct mtx ifp_lock; 767 struct ifnet *ifp; /* tracer ifp */ 768 struct ifmedia media; 769 int traceq; /* iq used by all tracers, -1 if none */ 770 int tracer_valid; /* bitmap of valid tracers */ 771 int tracer_enabled; /* bitmap of enabled tracers */ 772 773 char fw_version[32]; 774 char cfg_file[32]; 775 u_int cfcsum; 776 struct adapter_params params; 777 struct t4_virt_res vres; 778 779 uint16_t linkcaps; 780 uint16_t niccaps; 781 uint16_t toecaps; 782 uint16_t rdmacaps; 783 uint16_t iscsicaps; 784 uint16_t fcoecaps; 785 786 struct sysctl_ctx_list ctx; /* from adapter_full_init to full_uninit */ 787 788 struct mtx sc_lock; 789 char lockname[16]; 790 791 /* Starving free lists */ 792 struct mtx sfl_lock; /* same cache-line as sc_lock? but that's ok */ 793 TAILQ_HEAD(, sge_fl) sfl; 794 struct callout sfl_callout; 795 796 struct mtx regwin_lock; /* for indirect reads and memory windows */ 797 798 an_handler_t an_handler __aligned(CACHE_LINE_SIZE); 799 fw_msg_handler_t fw_msg_handler[5]; /* NUM_FW6_TYPES */ 800 cpl_handler_t cpl_handler[0xef]; /* NUM_CPL_CMDS */ 801 802 #ifdef INVARIANTS 803 const char *last_op; 804 const void *last_op_thr; 805 #endif 806 807 int sc_do_rxcopy; 808 }; 809 810 #define ADAPTER_LOCK(sc) mtx_lock(&(sc)->sc_lock) 811 #define ADAPTER_UNLOCK(sc) mtx_unlock(&(sc)->sc_lock) 812 #define ADAPTER_LOCK_ASSERT_OWNED(sc) mtx_assert(&(sc)->sc_lock, MA_OWNED) 813 #define ADAPTER_LOCK_ASSERT_NOTOWNED(sc) mtx_assert(&(sc)->sc_lock, MA_NOTOWNED) 814 815 /* XXX: not bulletproof, but much better than nothing */ 816 #define ASSERT_SYNCHRONIZED_OP(sc) \ 817 KASSERT(IS_BUSY(sc) && \ 818 (mtx_owned(&(sc)->sc_lock) || sc->last_op_thr == curthread), \ 819 ("%s: operation not synchronized.", __func__)) 820 821 #define PORT_LOCK(pi) mtx_lock(&(pi)->pi_lock) 822 #define PORT_UNLOCK(pi) mtx_unlock(&(pi)->pi_lock) 823 #define PORT_LOCK_ASSERT_OWNED(pi) mtx_assert(&(pi)->pi_lock, MA_OWNED) 824 #define PORT_LOCK_ASSERT_NOTOWNED(pi) mtx_assert(&(pi)->pi_lock, MA_NOTOWNED) 825 826 #define FL_LOCK(fl) mtx_lock(&(fl)->fl_lock) 827 #define FL_TRYLOCK(fl) mtx_trylock(&(fl)->fl_lock) 828 #define FL_UNLOCK(fl) mtx_unlock(&(fl)->fl_lock) 829 #define FL_LOCK_ASSERT_OWNED(fl) mtx_assert(&(fl)->fl_lock, MA_OWNED) 830 #define FL_LOCK_ASSERT_NOTOWNED(fl) mtx_assert(&(fl)->fl_lock, MA_NOTOWNED) 831 832 #define RXQ_FL_LOCK(rxq) FL_LOCK(&(rxq)->fl) 833 #define RXQ_FL_UNLOCK(rxq) FL_UNLOCK(&(rxq)->fl) 834 #define RXQ_FL_LOCK_ASSERT_OWNED(rxq) FL_LOCK_ASSERT_OWNED(&(rxq)->fl) 835 #define RXQ_FL_LOCK_ASSERT_NOTOWNED(rxq) FL_LOCK_ASSERT_NOTOWNED(&(rxq)->fl) 836 837 #define EQ_LOCK(eq) mtx_lock(&(eq)->eq_lock) 838 #define EQ_TRYLOCK(eq) mtx_trylock(&(eq)->eq_lock) 839 #define EQ_UNLOCK(eq) mtx_unlock(&(eq)->eq_lock) 840 #define EQ_LOCK_ASSERT_OWNED(eq) mtx_assert(&(eq)->eq_lock, MA_OWNED) 841 #define EQ_LOCK_ASSERT_NOTOWNED(eq) mtx_assert(&(eq)->eq_lock, MA_NOTOWNED) 842 843 #define TXQ_LOCK(txq) EQ_LOCK(&(txq)->eq) 844 #define TXQ_TRYLOCK(txq) EQ_TRYLOCK(&(txq)->eq) 845 #define TXQ_UNLOCK(txq) EQ_UNLOCK(&(txq)->eq) 846 #define TXQ_LOCK_ASSERT_OWNED(txq) EQ_LOCK_ASSERT_OWNED(&(txq)->eq) 847 #define TXQ_LOCK_ASSERT_NOTOWNED(txq) EQ_LOCK_ASSERT_NOTOWNED(&(txq)->eq) 848 849 #define for_each_txq(pi, iter, q) \ 850 for (q = &pi->adapter->sge.txq[pi->first_txq], iter = 0; \ 851 iter < pi->ntxq; ++iter, ++q) 852 #define for_each_rxq(pi, iter, q) \ 853 for (q = &pi->adapter->sge.rxq[pi->first_rxq], iter = 0; \ 854 iter < pi->nrxq; ++iter, ++q) 855 #define for_each_ofld_txq(pi, iter, q) \ 856 for (q = &pi->adapter->sge.ofld_txq[pi->first_ofld_txq], iter = 0; \ 857 iter < pi->nofldtxq; ++iter, ++q) 858 #define for_each_ofld_rxq(pi, iter, q) \ 859 for (q = &pi->adapter->sge.ofld_rxq[pi->first_ofld_rxq], iter = 0; \ 860 iter < pi->nofldrxq; ++iter, ++q) 861 #define for_each_nm_txq(pi, iter, q) \ 862 for (q = &pi->adapter->sge.nm_txq[pi->first_nm_txq], iter = 0; \ 863 iter < pi->nnmtxq; ++iter, ++q) 864 #define for_each_nm_rxq(pi, iter, q) \ 865 for (q = &pi->adapter->sge.nm_rxq[pi->first_nm_rxq], iter = 0; \ 866 iter < pi->nnmrxq; ++iter, ++q) 867 868 #define IDXINCR(idx, incr, wrap) do { \ 869 idx = wrap - idx > incr ? idx + incr : incr - (wrap - idx); \ 870 } while (0) 871 #define IDXDIFF(head, tail, wrap) \ 872 ((head) >= (tail) ? (head) - (tail) : (wrap) - (tail) + (head)) 873 874 /* One for errors, one for firmware events */ 875 #define T4_EXTRA_INTR 2 876 877 static inline uint32_t 878 t4_read_reg(struct adapter *sc, uint32_t reg) 879 { 880 881 return bus_space_read_4(sc->bt, sc->bh, reg); 882 } 883 884 static inline void 885 t4_write_reg(struct adapter *sc, uint32_t reg, uint32_t val) 886 { 887 888 bus_space_write_4(sc->bt, sc->bh, reg, val); 889 } 890 891 static inline uint64_t 892 t4_read_reg64(struct adapter *sc, uint32_t reg) 893 { 894 895 return t4_bus_space_read_8(sc->bt, sc->bh, reg); 896 } 897 898 static inline void 899 t4_write_reg64(struct adapter *sc, uint32_t reg, uint64_t val) 900 { 901 902 t4_bus_space_write_8(sc->bt, sc->bh, reg, val); 903 } 904 905 static inline void 906 t4_os_pci_read_cfg1(struct adapter *sc, int reg, uint8_t *val) 907 { 908 909 *val = pci_read_config(sc->dev, reg, 1); 910 } 911 912 static inline void 913 t4_os_pci_write_cfg1(struct adapter *sc, int reg, uint8_t val) 914 { 915 916 pci_write_config(sc->dev, reg, val, 1); 917 } 918 919 static inline void 920 t4_os_pci_read_cfg2(struct adapter *sc, int reg, uint16_t *val) 921 { 922 923 *val = pci_read_config(sc->dev, reg, 2); 924 } 925 926 static inline void 927 t4_os_pci_write_cfg2(struct adapter *sc, int reg, uint16_t val) 928 { 929 930 pci_write_config(sc->dev, reg, val, 2); 931 } 932 933 static inline void 934 t4_os_pci_read_cfg4(struct adapter *sc, int reg, uint32_t *val) 935 { 936 937 *val = pci_read_config(sc->dev, reg, 4); 938 } 939 940 static inline void 941 t4_os_pci_write_cfg4(struct adapter *sc, int reg, uint32_t val) 942 { 943 944 pci_write_config(sc->dev, reg, val, 4); 945 } 946 947 static inline struct port_info * 948 adap2pinfo(struct adapter *sc, int idx) 949 { 950 951 return (sc->port[idx]); 952 } 953 954 static inline void 955 t4_os_set_hw_addr(struct adapter *sc, int idx, uint8_t hw_addr[]) 956 { 957 958 bcopy(hw_addr, sc->port[idx]->hw_addr, ETHER_ADDR_LEN); 959 } 960 961 static inline bool 962 is_10G_port(const struct port_info *pi) 963 { 964 965 return ((pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G) != 0); 966 } 967 968 static inline bool 969 is_40G_port(const struct port_info *pi) 970 { 971 972 return ((pi->link_cfg.supported & FW_PORT_CAP_SPEED_40G) != 0); 973 } 974 975 static inline int 976 tx_resume_threshold(struct sge_eq *eq) 977 { 978 979 return (eq->qsize / 4); 980 } 981 982 /* t4_main.c */ 983 void t4_tx_task(void *, int); 984 void t4_tx_callout(void *); 985 int t4_os_find_pci_capability(struct adapter *, int); 986 int t4_os_pci_save_state(struct adapter *); 987 int t4_os_pci_restore_state(struct adapter *); 988 void t4_os_portmod_changed(const struct adapter *, int); 989 void t4_os_link_changed(struct adapter *, int, int, int); 990 void t4_iterate(void (*)(struct adapter *, void *), void *); 991 int t4_register_cpl_handler(struct adapter *, int, cpl_handler_t); 992 int t4_register_an_handler(struct adapter *, an_handler_t); 993 int t4_register_fw_msg_handler(struct adapter *, int, fw_msg_handler_t); 994 int t4_filter_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *); 995 int begin_synchronized_op(struct adapter *, struct port_info *, int, char *); 996 void end_synchronized_op(struct adapter *, int); 997 int update_mac_settings(struct ifnet *, int); 998 int adapter_full_init(struct adapter *); 999 int adapter_full_uninit(struct adapter *); 1000 int port_full_init(struct port_info *); 1001 int port_full_uninit(struct port_info *); 1002 1003 #ifdef DEV_NETMAP 1004 /* t4_netmap.c */ 1005 int create_netmap_ifnet(struct port_info *); 1006 int destroy_netmap_ifnet(struct port_info *); 1007 void t4_nm_intr(void *); 1008 #endif 1009 1010 /* t4_sge.c */ 1011 void t4_sge_modload(void); 1012 void t4_sge_modunload(void); 1013 uint64_t t4_sge_extfree_refs(void); 1014 void t4_init_sge_cpl_handlers(struct adapter *); 1015 void t4_tweak_chip_settings(struct adapter *); 1016 int t4_read_chip_settings(struct adapter *); 1017 int t4_create_dma_tag(struct adapter *); 1018 void t4_sge_sysctls(struct adapter *, struct sysctl_ctx_list *, 1019 struct sysctl_oid_list *); 1020 int t4_destroy_dma_tag(struct adapter *); 1021 int t4_setup_adapter_queues(struct adapter *); 1022 int t4_teardown_adapter_queues(struct adapter *); 1023 int t4_setup_port_queues(struct port_info *); 1024 int t4_teardown_port_queues(struct port_info *); 1025 int t4_alloc_tx_maps(struct tx_maps *, bus_dma_tag_t, int, int); 1026 void t4_free_tx_maps(struct tx_maps *, bus_dma_tag_t); 1027 void t4_intr_all(void *); 1028 void t4_intr(void *); 1029 void t4_intr_err(void *); 1030 void t4_intr_evt(void *); 1031 void t4_wrq_tx_locked(struct adapter *, struct sge_wrq *, struct wrqe *); 1032 int t4_eth_tx(struct ifnet *, struct sge_txq *, struct mbuf *); 1033 void t4_update_fl_bufsize(struct ifnet *); 1034 int can_resume_tx(struct sge_eq *); 1035 1036 /* t4_tracer.c */ 1037 struct t4_tracer; 1038 void t4_tracer_modload(void); 1039 void t4_tracer_modunload(void); 1040 void t4_tracer_port_detach(struct adapter *); 1041 int t4_get_tracer(struct adapter *, struct t4_tracer *); 1042 int t4_set_tracer(struct adapter *, struct t4_tracer *); 1043 int t4_trace_pkt(struct sge_iq *, const struct rss_header *, struct mbuf *); 1044 int t5_trace_pkt(struct sge_iq *, const struct rss_header *, struct mbuf *); 1045 1046 static inline struct wrqe * 1047 alloc_wrqe(int wr_len, struct sge_wrq *wrq) 1048 { 1049 int len = offsetof(struct wrqe, wr) + wr_len; 1050 struct wrqe *wr; 1051 1052 wr = malloc(len, M_CXGBE, M_NOWAIT); 1053 if (__predict_false(wr == NULL)) 1054 return (NULL); 1055 wr->wr_len = wr_len; 1056 wr->wrq = wrq; 1057 return (wr); 1058 } 1059 1060 static inline void * 1061 wrtod(struct wrqe *wr) 1062 { 1063 return (&wr->wr[0]); 1064 } 1065 1066 static inline void 1067 free_wrqe(struct wrqe *wr) 1068 { 1069 free(wr, M_CXGBE); 1070 } 1071 1072 static inline void 1073 t4_wrq_tx(struct adapter *sc, struct wrqe *wr) 1074 { 1075 struct sge_wrq *wrq = wr->wrq; 1076 1077 TXQ_LOCK(wrq); 1078 t4_wrq_tx_locked(sc, wrq, wr); 1079 TXQ_UNLOCK(wrq); 1080 } 1081 1082 #endif 1083