1 /* 2 * Copyright (c) 2012, 2013 Intel Corporation. All rights reserved. 3 * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved. 4 * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved. 5 * 6 * This software is available to you under a choice of one of two 7 * licenses. You may choose to be licensed under the terms of the GNU 8 * General Public License (GPL) Version 2, available from the file 9 * COPYING in the main directory of this source tree, or the 10 * OpenIB.org BSD license below: 11 * 12 * Redistribution and use in source and binary forms, with or 13 * without modification, are permitted provided that the following 14 * conditions are met: 15 * 16 * - Redistributions of source code must retain the above 17 * copyright notice, this list of conditions and the following 18 * disclaimer. 19 * 20 * - Redistributions in binary form must reproduce the above 21 * copyright notice, this list of conditions and the following 22 * disclaimer in the documentation and/or other materials 23 * provided with the distribution. 24 * 25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 32 * SOFTWARE. 33 */ 34 35 #ifndef QIB_VERBS_H 36 #define QIB_VERBS_H 37 38 #include <linux/types.h> 39 #include <linux/spinlock.h> 40 #include <linux/kernel.h> 41 #include <linux/interrupt.h> 42 #include <linux/kref.h> 43 #include <linux/workqueue.h> 44 #include <linux/kthread.h> 45 #include <linux/completion.h> 46 #include <rdma/ib_pack.h> 47 #include <rdma/ib_user_verbs.h> 48 49 struct qib_ctxtdata; 50 struct qib_pportdata; 51 struct qib_devdata; 52 struct qib_verbs_txreq; 53 54 #define QIB_MAX_RDMA_ATOMIC 16 55 #define QIB_GUIDS_PER_PORT 5 56 57 #define QPN_MAX (1 << 24) 58 #define QPNMAP_ENTRIES (QPN_MAX / PAGE_SIZE / BITS_PER_BYTE) 59 60 /* 61 * Increment this value if any changes that break userspace ABI 62 * compatibility are made. 63 */ 64 #define QIB_UVERBS_ABI_VERSION 2 65 66 /* 67 * Define an ib_cq_notify value that is not valid so we know when CQ 68 * notifications are armed. 69 */ 70 #define IB_CQ_NONE (IB_CQ_NEXT_COMP + 1) 71 72 #define IB_SEQ_NAK (3 << 29) 73 74 /* AETH NAK opcode values */ 75 #define IB_RNR_NAK 0x20 76 #define IB_NAK_PSN_ERROR 0x60 77 #define IB_NAK_INVALID_REQUEST 0x61 78 #define IB_NAK_REMOTE_ACCESS_ERROR 0x62 79 #define IB_NAK_REMOTE_OPERATIONAL_ERROR 0x63 80 #define IB_NAK_INVALID_RD_REQUEST 0x64 81 82 /* Flags for checking QP state (see ib_qib_state_ops[]) */ 83 #define QIB_POST_SEND_OK 0x01 84 #define QIB_POST_RECV_OK 0x02 85 #define QIB_PROCESS_RECV_OK 0x04 86 #define QIB_PROCESS_SEND_OK 0x08 87 #define QIB_PROCESS_NEXT_SEND_OK 0x10 88 #define QIB_FLUSH_SEND 0x20 89 #define QIB_FLUSH_RECV 0x40 90 #define QIB_PROCESS_OR_FLUSH_SEND \ 91 (QIB_PROCESS_SEND_OK | QIB_FLUSH_SEND) 92 93 /* IB Performance Manager status values */ 94 #define IB_PMA_SAMPLE_STATUS_DONE 0x00 95 #define IB_PMA_SAMPLE_STATUS_STARTED 0x01 96 #define IB_PMA_SAMPLE_STATUS_RUNNING 0x02 97 98 /* Mandatory IB performance counter select values. */ 99 #define IB_PMA_PORT_XMIT_DATA cpu_to_be16(0x0001) 100 #define IB_PMA_PORT_RCV_DATA cpu_to_be16(0x0002) 101 #define IB_PMA_PORT_XMIT_PKTS cpu_to_be16(0x0003) 102 #define IB_PMA_PORT_RCV_PKTS cpu_to_be16(0x0004) 103 #define IB_PMA_PORT_XMIT_WAIT cpu_to_be16(0x0005) 104 105 #define QIB_VENDOR_IPG cpu_to_be16(0xFFA0) 106 107 #define IB_BTH_REQ_ACK (1 << 31) 108 #define IB_BTH_SOLICITED (1 << 23) 109 #define IB_BTH_MIG_REQ (1 << 22) 110 111 /* XXX Should be defined in ib_verbs.h enum ib_port_cap_flags */ 112 #define IB_PORT_OTHER_LOCAL_CHANGES_SUP (1 << 26) 113 114 #define IB_GRH_VERSION 6 115 #define IB_GRH_VERSION_MASK 0xF 116 #define IB_GRH_VERSION_SHIFT 28 117 #define IB_GRH_TCLASS_MASK 0xFF 118 #define IB_GRH_TCLASS_SHIFT 20 119 #define IB_GRH_FLOW_MASK 0xFFFFF 120 #define IB_GRH_FLOW_SHIFT 0 121 #define IB_GRH_NEXT_HDR 0x1B 122 123 #define IB_DEFAULT_GID_PREFIX cpu_to_be64(0xfe80000000000000ULL) 124 125 /* Values for set/get portinfo VLCap OperationalVLs */ 126 #define IB_VL_VL0 1 127 #define IB_VL_VL0_1 2 128 #define IB_VL_VL0_3 3 129 #define IB_VL_VL0_7 4 130 #define IB_VL_VL0_14 5 131 132 static inline int qib_num_vls(int vls) 133 { 134 switch (vls) { 135 default: 136 case IB_VL_VL0: 137 return 1; 138 case IB_VL_VL0_1: 139 return 2; 140 case IB_VL_VL0_3: 141 return 4; 142 case IB_VL_VL0_7: 143 return 8; 144 case IB_VL_VL0_14: 145 return 15; 146 } 147 } 148 149 struct ib_reth { 150 __be64 vaddr; 151 __be32 rkey; 152 __be32 length; 153 } __packed; 154 155 struct ib_atomic_eth { 156 __be32 vaddr[2]; /* unaligned so access as 2 32-bit words */ 157 __be32 rkey; 158 __be64 swap_data; 159 __be64 compare_data; 160 } __packed; 161 162 struct qib_other_headers { 163 __be32 bth[3]; 164 union { 165 struct { 166 __be32 deth[2]; 167 __be32 imm_data; 168 } ud; 169 struct { 170 struct ib_reth reth; 171 __be32 imm_data; 172 } rc; 173 struct { 174 __be32 aeth; 175 __be32 atomic_ack_eth[2]; 176 } at; 177 __be32 imm_data; 178 __be32 aeth; 179 struct ib_atomic_eth atomic_eth; 180 } u; 181 } __packed; 182 183 /* 184 * Note that UD packets with a GRH header are 8+40+12+8 = 68 bytes 185 * long (72 w/ imm_data). Only the first 56 bytes of the IB header 186 * will be in the eager header buffer. The remaining 12 or 16 bytes 187 * are in the data buffer. 188 */ 189 struct qib_ib_header { 190 __be16 lrh[4]; 191 union { 192 struct { 193 struct ib_grh grh; 194 struct qib_other_headers oth; 195 } l; 196 struct qib_other_headers oth; 197 } u; 198 } __packed; 199 200 struct qib_pio_header { 201 __le32 pbc[2]; 202 struct qib_ib_header hdr; 203 } __packed; 204 205 /* 206 * There is one struct qib_mcast for each multicast GID. 207 * All attached QPs are then stored as a list of 208 * struct qib_mcast_qp. 209 */ 210 struct qib_mcast_qp { 211 struct list_head list; 212 struct qib_qp *qp; 213 }; 214 215 struct qib_mcast { 216 struct rb_node rb_node; 217 union ib_gid mgid; 218 struct list_head qp_list; 219 wait_queue_head_t wait; 220 atomic_t refcount; 221 int n_attached; 222 }; 223 224 /* Protection domain */ 225 struct qib_pd { 226 struct ib_pd ibpd; 227 int user; /* non-zero if created from user space */ 228 }; 229 230 /* Address Handle */ 231 struct qib_ah { 232 struct ib_ah ibah; 233 struct ib_ah_attr attr; 234 atomic_t refcount; 235 }; 236 237 /* 238 * This structure is used by qib_mmap() to validate an offset 239 * when an mmap() request is made. The vm_area_struct then uses 240 * this as its vm_private_data. 241 */ 242 struct qib_mmap_info { 243 struct list_head pending_mmaps; 244 struct ib_ucontext *context; 245 void *obj; 246 __u64 offset; 247 struct kref ref; 248 unsigned size; 249 }; 250 251 /* 252 * This structure is used to contain the head pointer, tail pointer, 253 * and completion queue entries as a single memory allocation so 254 * it can be mmap'ed into user space. 255 */ 256 struct qib_cq_wc { 257 u32 head; /* index of next entry to fill */ 258 u32 tail; /* index of next ib_poll_cq() entry */ 259 union { 260 /* these are actually size ibcq.cqe + 1 */ 261 struct ib_uverbs_wc uqueue[0]; 262 struct ib_wc kqueue[0]; 263 }; 264 }; 265 266 /* 267 * The completion queue structure. 268 */ 269 struct qib_cq { 270 struct ib_cq ibcq; 271 struct kthread_work comptask; 272 struct qib_devdata *dd; 273 spinlock_t lock; /* protect changes in this struct */ 274 u8 notify; 275 u8 triggered; 276 struct qib_cq_wc *queue; 277 struct qib_mmap_info *ip; 278 }; 279 280 /* 281 * A segment is a linear region of low physical memory. 282 * XXX Maybe we should use phys addr here and kmap()/kunmap(). 283 * Used by the verbs layer. 284 */ 285 struct qib_seg { 286 void *vaddr; 287 size_t length; 288 }; 289 290 /* The number of qib_segs that fit in a page. */ 291 #define QIB_SEGSZ (PAGE_SIZE / sizeof(struct qib_seg)) 292 293 struct qib_segarray { 294 struct qib_seg segs[QIB_SEGSZ]; 295 }; 296 297 struct qib_mregion { 298 struct ib_pd *pd; /* shares refcnt of ibmr.pd */ 299 u64 user_base; /* User's address for this region */ 300 u64 iova; /* IB start address of this region */ 301 size_t length; 302 u32 lkey; 303 u32 offset; /* offset (bytes) to start of region */ 304 int access_flags; 305 u32 max_segs; /* number of qib_segs in all the arrays */ 306 u32 mapsz; /* size of the map array */ 307 u8 page_shift; /* 0 - non unform/non powerof2 sizes */ 308 u8 lkey_published; /* in global table */ 309 struct completion comp; /* complete when refcount goes to zero */ 310 struct rcu_head list; 311 atomic_t refcount; 312 struct qib_segarray *map[0]; /* the segments */ 313 }; 314 315 /* 316 * These keep track of the copy progress within a memory region. 317 * Used by the verbs layer. 318 */ 319 struct qib_sge { 320 struct qib_mregion *mr; 321 void *vaddr; /* kernel virtual address of segment */ 322 u32 sge_length; /* length of the SGE */ 323 u32 length; /* remaining length of the segment */ 324 u16 m; /* current index: mr->map[m] */ 325 u16 n; /* current index: mr->map[m]->segs[n] */ 326 }; 327 328 /* Memory region */ 329 struct qib_mr { 330 struct ib_mr ibmr; 331 struct ib_umem *umem; 332 struct qib_mregion mr; /* must be last */ 333 }; 334 335 /* 336 * Send work request queue entry. 337 * The size of the sg_list is determined when the QP is created and stored 338 * in qp->s_max_sge. 339 */ 340 struct qib_swqe { 341 struct ib_send_wr wr; /* don't use wr.sg_list */ 342 u32 psn; /* first packet sequence number */ 343 u32 lpsn; /* last packet sequence number */ 344 u32 ssn; /* send sequence number */ 345 u32 length; /* total length of data in sg_list */ 346 struct qib_sge sg_list[0]; 347 }; 348 349 /* 350 * Receive work request queue entry. 351 * The size of the sg_list is determined when the QP (or SRQ) is created 352 * and stored in qp->r_rq.max_sge (or srq->rq.max_sge). 353 */ 354 struct qib_rwqe { 355 u64 wr_id; 356 u8 num_sge; 357 struct ib_sge sg_list[0]; 358 }; 359 360 /* 361 * This structure is used to contain the head pointer, tail pointer, 362 * and receive work queue entries as a single memory allocation so 363 * it can be mmap'ed into user space. 364 * Note that the wq array elements are variable size so you can't 365 * just index into the array to get the N'th element; 366 * use get_rwqe_ptr() instead. 367 */ 368 struct qib_rwq { 369 u32 head; /* new work requests posted to the head */ 370 u32 tail; /* receives pull requests from here. */ 371 struct qib_rwqe wq[0]; 372 }; 373 374 struct qib_rq { 375 struct qib_rwq *wq; 376 u32 size; /* size of RWQE array */ 377 u8 max_sge; 378 spinlock_t lock /* protect changes in this struct */ 379 ____cacheline_aligned_in_smp; 380 }; 381 382 struct qib_srq { 383 struct ib_srq ibsrq; 384 struct qib_rq rq; 385 struct qib_mmap_info *ip; 386 /* send signal when number of RWQEs < limit */ 387 u32 limit; 388 }; 389 390 struct qib_sge_state { 391 struct qib_sge *sg_list; /* next SGE to be used if any */ 392 struct qib_sge sge; /* progress state for the current SGE */ 393 u32 total_len; 394 u8 num_sge; 395 }; 396 397 /* 398 * This structure holds the information that the send tasklet needs 399 * to send a RDMA read response or atomic operation. 400 */ 401 struct qib_ack_entry { 402 u8 opcode; 403 u8 sent; 404 u32 psn; 405 u32 lpsn; 406 union { 407 struct qib_sge rdma_sge; 408 u64 atomic_data; 409 }; 410 }; 411 412 /* 413 * Variables prefixed with s_ are for the requester (sender). 414 * Variables prefixed with r_ are for the responder (receiver). 415 * Variables prefixed with ack_ are for responder replies. 416 * 417 * Common variables are protected by both r_rq.lock and s_lock in that order 418 * which only happens in modify_qp() or changing the QP 'state'. 419 */ 420 struct qib_qp { 421 struct ib_qp ibqp; 422 /* read mostly fields above and below */ 423 struct ib_ah_attr remote_ah_attr; 424 struct ib_ah_attr alt_ah_attr; 425 struct qib_qp __rcu *next; /* link list for QPN hash table */ 426 struct qib_swqe *s_wq; /* send work queue */ 427 struct qib_mmap_info *ip; 428 struct qib_ib_header *s_hdr; /* next packet header to send */ 429 unsigned long timeout_jiffies; /* computed from timeout */ 430 431 enum ib_mtu path_mtu; 432 u32 remote_qpn; 433 u32 pmtu; /* decoded from path_mtu */ 434 u32 qkey; /* QKEY for this QP (for UD or RD) */ 435 u32 s_size; /* send work queue size */ 436 u32 s_rnr_timeout; /* number of milliseconds for RNR timeout */ 437 438 u8 state; /* QP state */ 439 u8 qp_access_flags; 440 u8 alt_timeout; /* Alternate path timeout for this QP */ 441 u8 timeout; /* Timeout for this QP */ 442 u8 s_srate; 443 u8 s_mig_state; 444 u8 port_num; 445 u8 s_pkey_index; /* PKEY index to use */ 446 u8 s_alt_pkey_index; /* Alternate path PKEY index to use */ 447 u8 r_max_rd_atomic; /* max number of RDMA read/atomic to receive */ 448 u8 s_max_rd_atomic; /* max number of RDMA read/atomic to send */ 449 u8 s_retry_cnt; /* number of times to retry */ 450 u8 s_rnr_retry_cnt; 451 u8 r_min_rnr_timer; /* retry timeout value for RNR NAKs */ 452 u8 s_max_sge; /* size of s_wq->sg_list */ 453 u8 s_draining; 454 455 /* start of read/write fields */ 456 457 atomic_t refcount ____cacheline_aligned_in_smp; 458 wait_queue_head_t wait; 459 460 461 struct qib_ack_entry s_ack_queue[QIB_MAX_RDMA_ATOMIC + 1] 462 ____cacheline_aligned_in_smp; 463 struct qib_sge_state s_rdma_read_sge; 464 465 spinlock_t r_lock ____cacheline_aligned_in_smp; /* used for APM */ 466 unsigned long r_aflags; 467 u64 r_wr_id; /* ID for current receive WQE */ 468 u32 r_ack_psn; /* PSN for next ACK or atomic ACK */ 469 u32 r_len; /* total length of r_sge */ 470 u32 r_rcv_len; /* receive data len processed */ 471 u32 r_psn; /* expected rcv packet sequence number */ 472 u32 r_msn; /* message sequence number */ 473 474 u8 r_state; /* opcode of last packet received */ 475 u8 r_flags; 476 u8 r_head_ack_queue; /* index into s_ack_queue[] */ 477 478 struct list_head rspwait; /* link for waititing to respond */ 479 480 struct qib_sge_state r_sge; /* current receive data */ 481 struct qib_rq r_rq; /* receive work queue */ 482 483 spinlock_t s_lock ____cacheline_aligned_in_smp; 484 struct qib_sge_state *s_cur_sge; 485 u32 s_flags; 486 struct qib_verbs_txreq *s_tx; 487 struct qib_swqe *s_wqe; 488 struct qib_sge_state s_sge; /* current send request data */ 489 struct qib_mregion *s_rdma_mr; 490 atomic_t s_dma_busy; 491 u32 s_cur_size; /* size of send packet in bytes */ 492 u32 s_len; /* total length of s_sge */ 493 u32 s_rdma_read_len; /* total length of s_rdma_read_sge */ 494 u32 s_next_psn; /* PSN for next request */ 495 u32 s_last_psn; /* last response PSN processed */ 496 u32 s_sending_psn; /* lowest PSN that is being sent */ 497 u32 s_sending_hpsn; /* highest PSN that is being sent */ 498 u32 s_psn; /* current packet sequence number */ 499 u32 s_ack_rdma_psn; /* PSN for sending RDMA read responses */ 500 u32 s_ack_psn; /* PSN for acking sends and RDMA writes */ 501 u32 s_head; /* new entries added here */ 502 u32 s_tail; /* next entry to process */ 503 u32 s_cur; /* current work queue entry */ 504 u32 s_acked; /* last un-ACK'ed entry */ 505 u32 s_last; /* last completed entry */ 506 u32 s_ssn; /* SSN of tail entry */ 507 u32 s_lsn; /* limit sequence number (credit) */ 508 u16 s_hdrwords; /* size of s_hdr in 32 bit words */ 509 u16 s_rdma_ack_cnt; 510 u8 s_state; /* opcode of last packet sent */ 511 u8 s_ack_state; /* opcode of packet to ACK */ 512 u8 s_nak_state; /* non-zero if NAK is pending */ 513 u8 r_nak_state; /* non-zero if NAK is pending */ 514 u8 s_retry; /* requester retry counter */ 515 u8 s_rnr_retry; /* requester RNR retry counter */ 516 u8 s_num_rd_atomic; /* number of RDMA read/atomic pending */ 517 u8 s_tail_ack_queue; /* index into s_ack_queue[] */ 518 519 struct qib_sge_state s_ack_rdma_sge; 520 struct timer_list s_timer; 521 struct list_head iowait; /* link for wait PIO buf */ 522 523 struct work_struct s_work; 524 525 wait_queue_head_t wait_dma; 526 527 struct qib_sge r_sg_list[0] /* verified SGEs */ 528 ____cacheline_aligned_in_smp; 529 }; 530 531 /* 532 * Atomic bit definitions for r_aflags. 533 */ 534 #define QIB_R_WRID_VALID 0 535 #define QIB_R_REWIND_SGE 1 536 537 /* 538 * Bit definitions for r_flags. 539 */ 540 #define QIB_R_REUSE_SGE 0x01 541 #define QIB_R_RDMAR_SEQ 0x02 542 #define QIB_R_RSP_NAK 0x04 543 #define QIB_R_RSP_SEND 0x08 544 #define QIB_R_COMM_EST 0x10 545 546 /* 547 * Bit definitions for s_flags. 548 * 549 * QIB_S_SIGNAL_REQ_WR - set if QP send WRs contain completion signaled 550 * QIB_S_BUSY - send tasklet is processing the QP 551 * QIB_S_TIMER - the RC retry timer is active 552 * QIB_S_ACK_PENDING - an ACK is waiting to be sent after RDMA read/atomics 553 * QIB_S_WAIT_FENCE - waiting for all prior RDMA read or atomic SWQEs 554 * before processing the next SWQE 555 * QIB_S_WAIT_RDMAR - waiting for a RDMA read or atomic SWQE to complete 556 * before processing the next SWQE 557 * QIB_S_WAIT_RNR - waiting for RNR timeout 558 * QIB_S_WAIT_SSN_CREDIT - waiting for RC credits to process next SWQE 559 * QIB_S_WAIT_DMA - waiting for send DMA queue to drain before generating 560 * next send completion entry not via send DMA 561 * QIB_S_WAIT_PIO - waiting for a send buffer to be available 562 * QIB_S_WAIT_TX - waiting for a struct qib_verbs_txreq to be available 563 * QIB_S_WAIT_DMA_DESC - waiting for DMA descriptors to be available 564 * QIB_S_WAIT_KMEM - waiting for kernel memory to be available 565 * QIB_S_WAIT_PSN - waiting for a packet to exit the send DMA queue 566 * QIB_S_WAIT_ACK - waiting for an ACK packet before sending more requests 567 * QIB_S_SEND_ONE - send one packet, request ACK, then wait for ACK 568 */ 569 #define QIB_S_SIGNAL_REQ_WR 0x0001 570 #define QIB_S_BUSY 0x0002 571 #define QIB_S_TIMER 0x0004 572 #define QIB_S_RESP_PENDING 0x0008 573 #define QIB_S_ACK_PENDING 0x0010 574 #define QIB_S_WAIT_FENCE 0x0020 575 #define QIB_S_WAIT_RDMAR 0x0040 576 #define QIB_S_WAIT_RNR 0x0080 577 #define QIB_S_WAIT_SSN_CREDIT 0x0100 578 #define QIB_S_WAIT_DMA 0x0200 579 #define QIB_S_WAIT_PIO 0x0400 580 #define QIB_S_WAIT_TX 0x0800 581 #define QIB_S_WAIT_DMA_DESC 0x1000 582 #define QIB_S_WAIT_KMEM 0x2000 583 #define QIB_S_WAIT_PSN 0x4000 584 #define QIB_S_WAIT_ACK 0x8000 585 #define QIB_S_SEND_ONE 0x10000 586 #define QIB_S_UNLIMITED_CREDIT 0x20000 587 588 /* 589 * Wait flags that would prevent any packet type from being sent. 590 */ 591 #define QIB_S_ANY_WAIT_IO (QIB_S_WAIT_PIO | QIB_S_WAIT_TX | \ 592 QIB_S_WAIT_DMA_DESC | QIB_S_WAIT_KMEM) 593 594 /* 595 * Wait flags that would prevent send work requests from making progress. 596 */ 597 #define QIB_S_ANY_WAIT_SEND (QIB_S_WAIT_FENCE | QIB_S_WAIT_RDMAR | \ 598 QIB_S_WAIT_RNR | QIB_S_WAIT_SSN_CREDIT | QIB_S_WAIT_DMA | \ 599 QIB_S_WAIT_PSN | QIB_S_WAIT_ACK) 600 601 #define QIB_S_ANY_WAIT (QIB_S_ANY_WAIT_IO | QIB_S_ANY_WAIT_SEND) 602 603 #define QIB_PSN_CREDIT 16 604 605 /* 606 * Since struct qib_swqe is not a fixed size, we can't simply index into 607 * struct qib_qp.s_wq. This function does the array index computation. 608 */ 609 static inline struct qib_swqe *get_swqe_ptr(struct qib_qp *qp, 610 unsigned n) 611 { 612 return (struct qib_swqe *)((char *)qp->s_wq + 613 (sizeof(struct qib_swqe) + 614 qp->s_max_sge * 615 sizeof(struct qib_sge)) * n); 616 } 617 618 /* 619 * Since struct qib_rwqe is not a fixed size, we can't simply index into 620 * struct qib_rwq.wq. This function does the array index computation. 621 */ 622 static inline struct qib_rwqe *get_rwqe_ptr(struct qib_rq *rq, unsigned n) 623 { 624 return (struct qib_rwqe *) 625 ((char *) rq->wq->wq + 626 (sizeof(struct qib_rwqe) + 627 rq->max_sge * sizeof(struct ib_sge)) * n); 628 } 629 630 /* 631 * QPN-map pages start out as NULL, they get allocated upon 632 * first use and are never deallocated. This way, 633 * large bitmaps are not allocated unless large numbers of QPs are used. 634 */ 635 struct qpn_map { 636 void *page; 637 }; 638 639 struct qib_qpn_table { 640 spinlock_t lock; /* protect changes in this struct */ 641 unsigned flags; /* flags for QP0/1 allocated for each port */ 642 u32 last; /* last QP number allocated */ 643 u32 nmaps; /* size of the map table */ 644 u16 limit; 645 u16 mask; 646 /* bit map of free QP numbers other than 0/1 */ 647 struct qpn_map map[QPNMAP_ENTRIES]; 648 }; 649 650 struct qib_lkey_table { 651 spinlock_t lock; /* protect changes in this struct */ 652 u32 next; /* next unused index (speeds search) */ 653 u32 gen; /* generation count */ 654 u32 max; /* size of the table */ 655 struct qib_mregion __rcu **table; 656 }; 657 658 struct qib_opcode_stats { 659 u64 n_packets; /* number of packets */ 660 u64 n_bytes; /* total number of bytes */ 661 }; 662 663 struct qib_opcode_stats_perctx { 664 struct qib_opcode_stats stats[128]; 665 }; 666 667 struct qib_pma_counters { 668 u64 n_unicast_xmit; /* total unicast packets sent */ 669 u64 n_unicast_rcv; /* total unicast packets received */ 670 u64 n_multicast_xmit; /* total multicast packets sent */ 671 u64 n_multicast_rcv; /* total multicast packets received */ 672 }; 673 674 struct qib_ibport { 675 struct qib_qp __rcu *qp0; 676 struct qib_qp __rcu *qp1; 677 struct ib_mad_agent *send_agent; /* agent for SMI (traps) */ 678 struct qib_ah *sm_ah; 679 struct qib_ah *smi_ah; 680 struct rb_root mcast_tree; 681 spinlock_t lock; /* protect changes in this struct */ 682 683 /* non-zero when timer is set */ 684 unsigned long mkey_lease_timeout; 685 unsigned long trap_timeout; 686 __be64 gid_prefix; /* in network order */ 687 __be64 mkey; 688 __be64 guids[QIB_GUIDS_PER_PORT - 1]; /* writable GUIDs */ 689 u64 tid; /* TID for traps */ 690 struct qib_pma_counters __percpu *pmastats; 691 u64 z_unicast_xmit; /* starting count for PMA */ 692 u64 z_unicast_rcv; /* starting count for PMA */ 693 u64 z_multicast_xmit; /* starting count for PMA */ 694 u64 z_multicast_rcv; /* starting count for PMA */ 695 u64 z_symbol_error_counter; /* starting count for PMA */ 696 u64 z_link_error_recovery_counter; /* starting count for PMA */ 697 u64 z_link_downed_counter; /* starting count for PMA */ 698 u64 z_port_rcv_errors; /* starting count for PMA */ 699 u64 z_port_rcv_remphys_errors; /* starting count for PMA */ 700 u64 z_port_xmit_discards; /* starting count for PMA */ 701 u64 z_port_xmit_data; /* starting count for PMA */ 702 u64 z_port_rcv_data; /* starting count for PMA */ 703 u64 z_port_xmit_packets; /* starting count for PMA */ 704 u64 z_port_rcv_packets; /* starting count for PMA */ 705 u32 z_local_link_integrity_errors; /* starting count for PMA */ 706 u32 z_excessive_buffer_overrun_errors; /* starting count for PMA */ 707 u32 z_vl15_dropped; /* starting count for PMA */ 708 u32 n_rc_resends; 709 u32 n_rc_acks; 710 u32 n_rc_qacks; 711 u32 n_rc_delayed_comp; 712 u32 n_seq_naks; 713 u32 n_rdma_seq; 714 u32 n_rnr_naks; 715 u32 n_other_naks; 716 u32 n_loop_pkts; 717 u32 n_pkt_drops; 718 u32 n_vl15_dropped; 719 u32 n_rc_timeouts; 720 u32 n_dmawait; 721 u32 n_unaligned; 722 u32 n_rc_dupreq; 723 u32 n_rc_seqnak; 724 u32 port_cap_flags; 725 u32 pma_sample_start; 726 u32 pma_sample_interval; 727 __be16 pma_counter_select[5]; 728 u16 pma_tag; 729 u16 pkey_violations; 730 u16 qkey_violations; 731 u16 mkey_violations; 732 u16 mkey_lease_period; 733 u16 sm_lid; 734 u16 repress_traps; 735 u8 sm_sl; 736 u8 mkeyprot; 737 u8 subnet_timeout; 738 u8 vl_high_limit; 739 u8 sl_to_vl[16]; 740 741 }; 742 743 744 struct qib_ibdev { 745 struct ib_device ibdev; 746 struct list_head pending_mmaps; 747 spinlock_t mmap_offset_lock; /* protect mmap_offset */ 748 u32 mmap_offset; 749 struct qib_mregion __rcu *dma_mr; 750 751 /* QP numbers are shared by all IB ports */ 752 struct qib_qpn_table qpn_table; 753 struct qib_lkey_table lk_table; 754 struct list_head piowait; /* list for wait PIO buf */ 755 struct list_head dmawait; /* list for wait DMA */ 756 struct list_head txwait; /* list for wait qib_verbs_txreq */ 757 struct list_head memwait; /* list for wait kernel memory */ 758 struct list_head txreq_free; 759 struct timer_list mem_timer; 760 struct qib_qp __rcu **qp_table; 761 struct qib_pio_header *pio_hdrs; 762 dma_addr_t pio_hdrs_phys; 763 /* list of QPs waiting for RNR timer */ 764 spinlock_t pending_lock; /* protect wait lists, PMA counters, etc. */ 765 u32 qp_table_size; /* size of the hash table */ 766 u32 qp_rnd; /* random bytes for hash */ 767 spinlock_t qpt_lock; 768 769 u32 n_piowait; 770 u32 n_txwait; 771 772 u32 n_pds_allocated; /* number of PDs allocated for device */ 773 spinlock_t n_pds_lock; 774 u32 n_ahs_allocated; /* number of AHs allocated for device */ 775 spinlock_t n_ahs_lock; 776 u32 n_cqs_allocated; /* number of CQs allocated for device */ 777 spinlock_t n_cqs_lock; 778 u32 n_qps_allocated; /* number of QPs allocated for device */ 779 spinlock_t n_qps_lock; 780 u32 n_srqs_allocated; /* number of SRQs allocated for device */ 781 spinlock_t n_srqs_lock; 782 u32 n_mcast_grps_allocated; /* number of mcast groups allocated */ 783 spinlock_t n_mcast_grps_lock; 784 #ifdef CONFIG_DEBUG_FS 785 /* per HCA debugfs */ 786 struct dentry *qib_ibdev_dbg; 787 #endif 788 }; 789 790 struct qib_verbs_counters { 791 u64 symbol_error_counter; 792 u64 link_error_recovery_counter; 793 u64 link_downed_counter; 794 u64 port_rcv_errors; 795 u64 port_rcv_remphys_errors; 796 u64 port_xmit_discards; 797 u64 port_xmit_data; 798 u64 port_rcv_data; 799 u64 port_xmit_packets; 800 u64 port_rcv_packets; 801 u32 local_link_integrity_errors; 802 u32 excessive_buffer_overrun_errors; 803 u32 vl15_dropped; 804 }; 805 806 static inline struct qib_mr *to_imr(struct ib_mr *ibmr) 807 { 808 return container_of(ibmr, struct qib_mr, ibmr); 809 } 810 811 static inline struct qib_pd *to_ipd(struct ib_pd *ibpd) 812 { 813 return container_of(ibpd, struct qib_pd, ibpd); 814 } 815 816 static inline struct qib_ah *to_iah(struct ib_ah *ibah) 817 { 818 return container_of(ibah, struct qib_ah, ibah); 819 } 820 821 static inline struct qib_cq *to_icq(struct ib_cq *ibcq) 822 { 823 return container_of(ibcq, struct qib_cq, ibcq); 824 } 825 826 static inline struct qib_srq *to_isrq(struct ib_srq *ibsrq) 827 { 828 return container_of(ibsrq, struct qib_srq, ibsrq); 829 } 830 831 static inline struct qib_qp *to_iqp(struct ib_qp *ibqp) 832 { 833 return container_of(ibqp, struct qib_qp, ibqp); 834 } 835 836 static inline struct qib_ibdev *to_idev(struct ib_device *ibdev) 837 { 838 return container_of(ibdev, struct qib_ibdev, ibdev); 839 } 840 841 /* 842 * Send if not busy or waiting for I/O and either 843 * a RC response is pending or we can process send work requests. 844 */ 845 static inline int qib_send_ok(struct qib_qp *qp) 846 { 847 return !(qp->s_flags & (QIB_S_BUSY | QIB_S_ANY_WAIT_IO)) && 848 (qp->s_hdrwords || (qp->s_flags & QIB_S_RESP_PENDING) || 849 !(qp->s_flags & QIB_S_ANY_WAIT_SEND)); 850 } 851 852 /* 853 * This must be called with s_lock held. 854 */ 855 void qib_schedule_send(struct qib_qp *qp); 856 857 static inline int qib_pkey_ok(u16 pkey1, u16 pkey2) 858 { 859 u16 p1 = pkey1 & 0x7FFF; 860 u16 p2 = pkey2 & 0x7FFF; 861 862 /* 863 * Low 15 bits must be non-zero and match, and 864 * one of the two must be a full member. 865 */ 866 return p1 && p1 == p2 && ((__s16)pkey1 < 0 || (__s16)pkey2 < 0); 867 } 868 869 void qib_bad_pqkey(struct qib_ibport *ibp, __be16 trap_num, u32 key, u32 sl, 870 u32 qp1, u32 qp2, __be16 lid1, __be16 lid2); 871 void qib_cap_mask_chg(struct qib_ibport *ibp); 872 void qib_sys_guid_chg(struct qib_ibport *ibp); 873 void qib_node_desc_chg(struct qib_ibport *ibp); 874 int qib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num, 875 struct ib_wc *in_wc, struct ib_grh *in_grh, 876 struct ib_mad *in_mad, struct ib_mad *out_mad); 877 int qib_create_agents(struct qib_ibdev *dev); 878 void qib_free_agents(struct qib_ibdev *dev); 879 880 /* 881 * Compare the lower 24 bits of the two values. 882 * Returns an integer <, ==, or > than zero. 883 */ 884 static inline int qib_cmp24(u32 a, u32 b) 885 { 886 return (((int) a) - ((int) b)) << 8; 887 } 888 889 struct qib_mcast *qib_mcast_find(struct qib_ibport *ibp, union ib_gid *mgid); 890 891 int qib_snapshot_counters(struct qib_pportdata *ppd, u64 *swords, 892 u64 *rwords, u64 *spkts, u64 *rpkts, 893 u64 *xmit_wait); 894 895 int qib_get_counters(struct qib_pportdata *ppd, 896 struct qib_verbs_counters *cntrs); 897 898 int qib_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid); 899 900 int qib_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid); 901 902 int qib_mcast_tree_empty(struct qib_ibport *ibp); 903 904 __be32 qib_compute_aeth(struct qib_qp *qp); 905 906 struct qib_qp *qib_lookup_qpn(struct qib_ibport *ibp, u32 qpn); 907 908 struct ib_qp *qib_create_qp(struct ib_pd *ibpd, 909 struct ib_qp_init_attr *init_attr, 910 struct ib_udata *udata); 911 912 int qib_destroy_qp(struct ib_qp *ibqp); 913 914 int qib_error_qp(struct qib_qp *qp, enum ib_wc_status err); 915 916 int qib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, 917 int attr_mask, struct ib_udata *udata); 918 919 int qib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, 920 int attr_mask, struct ib_qp_init_attr *init_attr); 921 922 unsigned qib_free_all_qps(struct qib_devdata *dd); 923 924 void qib_init_qpn_table(struct qib_devdata *dd, struct qib_qpn_table *qpt); 925 926 void qib_free_qpn_table(struct qib_qpn_table *qpt); 927 928 #ifdef CONFIG_DEBUG_FS 929 930 struct qib_qp_iter; 931 932 struct qib_qp_iter *qib_qp_iter_init(struct qib_ibdev *dev); 933 934 int qib_qp_iter_next(struct qib_qp_iter *iter); 935 936 void qib_qp_iter_print(struct seq_file *s, struct qib_qp_iter *iter); 937 938 #endif 939 940 void qib_get_credit(struct qib_qp *qp, u32 aeth); 941 942 unsigned qib_pkt_delay(u32 plen, u8 snd_mult, u8 rcv_mult); 943 944 void qib_verbs_sdma_desc_avail(struct qib_pportdata *ppd, unsigned avail); 945 946 void qib_put_txreq(struct qib_verbs_txreq *tx); 947 948 int qib_verbs_send(struct qib_qp *qp, struct qib_ib_header *hdr, 949 u32 hdrwords, struct qib_sge_state *ss, u32 len); 950 951 void qib_copy_sge(struct qib_sge_state *ss, void *data, u32 length, 952 int release); 953 954 void qib_skip_sge(struct qib_sge_state *ss, u32 length, int release); 955 956 void qib_uc_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr, 957 int has_grh, void *data, u32 tlen, struct qib_qp *qp); 958 959 void qib_rc_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr, 960 int has_grh, void *data, u32 tlen, struct qib_qp *qp); 961 962 int qib_check_ah(struct ib_device *ibdev, struct ib_ah_attr *ah_attr); 963 964 struct ib_ah *qib_create_qp0_ah(struct qib_ibport *ibp, u16 dlid); 965 966 void qib_rc_rnr_retry(unsigned long arg); 967 968 void qib_rc_send_complete(struct qib_qp *qp, struct qib_ib_header *hdr); 969 970 void qib_rc_error(struct qib_qp *qp, enum ib_wc_status err); 971 972 int qib_post_ud_send(struct qib_qp *qp, struct ib_send_wr *wr); 973 974 void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr, 975 int has_grh, void *data, u32 tlen, struct qib_qp *qp); 976 977 int qib_alloc_lkey(struct qib_mregion *mr, int dma_region); 978 979 void qib_free_lkey(struct qib_mregion *mr); 980 981 int qib_lkey_ok(struct qib_lkey_table *rkt, struct qib_pd *pd, 982 struct qib_sge *isge, struct ib_sge *sge, int acc); 983 984 int qib_rkey_ok(struct qib_qp *qp, struct qib_sge *sge, 985 u32 len, u64 vaddr, u32 rkey, int acc); 986 987 int qib_post_srq_receive(struct ib_srq *ibsrq, struct ib_recv_wr *wr, 988 struct ib_recv_wr **bad_wr); 989 990 struct ib_srq *qib_create_srq(struct ib_pd *ibpd, 991 struct ib_srq_init_attr *srq_init_attr, 992 struct ib_udata *udata); 993 994 int qib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, 995 enum ib_srq_attr_mask attr_mask, 996 struct ib_udata *udata); 997 998 int qib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr); 999 1000 int qib_destroy_srq(struct ib_srq *ibsrq); 1001 1002 int qib_cq_init(struct qib_devdata *dd); 1003 1004 void qib_cq_exit(struct qib_devdata *dd); 1005 1006 void qib_cq_enter(struct qib_cq *cq, struct ib_wc *entry, int sig); 1007 1008 int qib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry); 1009 1010 struct ib_cq *qib_create_cq(struct ib_device *ibdev, int entries, 1011 int comp_vector, struct ib_ucontext *context, 1012 struct ib_udata *udata); 1013 1014 int qib_destroy_cq(struct ib_cq *ibcq); 1015 1016 int qib_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags notify_flags); 1017 1018 int qib_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata); 1019 1020 struct ib_mr *qib_get_dma_mr(struct ib_pd *pd, int acc); 1021 1022 struct ib_mr *qib_reg_phys_mr(struct ib_pd *pd, 1023 struct ib_phys_buf *buffer_list, 1024 int num_phys_buf, int acc, u64 *iova_start); 1025 1026 struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, 1027 u64 virt_addr, int mr_access_flags, 1028 struct ib_udata *udata); 1029 1030 int qib_dereg_mr(struct ib_mr *ibmr); 1031 1032 struct ib_mr *qib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len); 1033 1034 struct ib_fast_reg_page_list *qib_alloc_fast_reg_page_list( 1035 struct ib_device *ibdev, int page_list_len); 1036 1037 void qib_free_fast_reg_page_list(struct ib_fast_reg_page_list *pl); 1038 1039 int qib_fast_reg_mr(struct qib_qp *qp, struct ib_send_wr *wr); 1040 1041 struct ib_fmr *qib_alloc_fmr(struct ib_pd *pd, int mr_access_flags, 1042 struct ib_fmr_attr *fmr_attr); 1043 1044 int qib_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list, 1045 int list_len, u64 iova); 1046 1047 int qib_unmap_fmr(struct list_head *fmr_list); 1048 1049 int qib_dealloc_fmr(struct ib_fmr *ibfmr); 1050 1051 static inline void qib_get_mr(struct qib_mregion *mr) 1052 { 1053 atomic_inc(&mr->refcount); 1054 } 1055 1056 void mr_rcu_callback(struct rcu_head *list); 1057 1058 static inline void qib_put_mr(struct qib_mregion *mr) 1059 { 1060 if (unlikely(atomic_dec_and_test(&mr->refcount))) 1061 call_rcu(&mr->list, mr_rcu_callback); 1062 } 1063 1064 static inline void qib_put_ss(struct qib_sge_state *ss) 1065 { 1066 while (ss->num_sge) { 1067 qib_put_mr(ss->sge.mr); 1068 if (--ss->num_sge) 1069 ss->sge = *ss->sg_list++; 1070 } 1071 } 1072 1073 1074 void qib_release_mmap_info(struct kref *ref); 1075 1076 struct qib_mmap_info *qib_create_mmap_info(struct qib_ibdev *dev, u32 size, 1077 struct ib_ucontext *context, 1078 void *obj); 1079 1080 void qib_update_mmap_info(struct qib_ibdev *dev, struct qib_mmap_info *ip, 1081 u32 size, void *obj); 1082 1083 int qib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma); 1084 1085 int qib_get_rwqe(struct qib_qp *qp, int wr_id_only); 1086 1087 void qib_migrate_qp(struct qib_qp *qp); 1088 1089 int qib_ruc_check_hdr(struct qib_ibport *ibp, struct qib_ib_header *hdr, 1090 int has_grh, struct qib_qp *qp, u32 bth0); 1091 1092 u32 qib_make_grh(struct qib_ibport *ibp, struct ib_grh *hdr, 1093 struct ib_global_route *grh, u32 hwords, u32 nwords); 1094 1095 void qib_make_ruc_header(struct qib_qp *qp, struct qib_other_headers *ohdr, 1096 u32 bth0, u32 bth2); 1097 1098 void qib_do_send(struct work_struct *work); 1099 1100 void qib_send_complete(struct qib_qp *qp, struct qib_swqe *wqe, 1101 enum ib_wc_status status); 1102 1103 void qib_send_rc_ack(struct qib_qp *qp); 1104 1105 int qib_make_rc_req(struct qib_qp *qp); 1106 1107 int qib_make_uc_req(struct qib_qp *qp); 1108 1109 int qib_make_ud_req(struct qib_qp *qp); 1110 1111 int qib_register_ib_device(struct qib_devdata *); 1112 1113 void qib_unregister_ib_device(struct qib_devdata *); 1114 1115 void qib_ib_rcv(struct qib_ctxtdata *, void *, void *, u32); 1116 1117 void qib_ib_piobufavail(struct qib_devdata *); 1118 1119 unsigned qib_get_npkeys(struct qib_devdata *); 1120 1121 unsigned qib_get_pkey(struct qib_ibport *, unsigned); 1122 1123 extern const enum ib_wc_opcode ib_qib_wc_opcode[]; 1124 1125 /* 1126 * Below HCA-independent IB PhysPortState values, returned 1127 * by the f_ibphys_portstate() routine. 1128 */ 1129 #define IB_PHYSPORTSTATE_SLEEP 1 1130 #define IB_PHYSPORTSTATE_POLL 2 1131 #define IB_PHYSPORTSTATE_DISABLED 3 1132 #define IB_PHYSPORTSTATE_CFG_TRAIN 4 1133 #define IB_PHYSPORTSTATE_LINKUP 5 1134 #define IB_PHYSPORTSTATE_LINK_ERR_RECOVER 6 1135 #define IB_PHYSPORTSTATE_CFG_DEBOUNCE 8 1136 #define IB_PHYSPORTSTATE_CFG_IDLE 0xB 1137 #define IB_PHYSPORTSTATE_RECOVERY_RETRAIN 0xC 1138 #define IB_PHYSPORTSTATE_RECOVERY_WAITRMT 0xE 1139 #define IB_PHYSPORTSTATE_RECOVERY_IDLE 0xF 1140 #define IB_PHYSPORTSTATE_CFG_ENH 0x10 1141 #define IB_PHYSPORTSTATE_CFG_WAIT_ENH 0x13 1142 1143 extern const int ib_qib_state_ops[]; 1144 1145 extern __be64 ib_qib_sys_image_guid; /* in network order */ 1146 1147 extern unsigned int ib_qib_lkey_table_size; 1148 1149 extern unsigned int ib_qib_max_cqes; 1150 1151 extern unsigned int ib_qib_max_cqs; 1152 1153 extern unsigned int ib_qib_max_qp_wrs; 1154 1155 extern unsigned int ib_qib_max_qps; 1156 1157 extern unsigned int ib_qib_max_sges; 1158 1159 extern unsigned int ib_qib_max_mcast_grps; 1160 1161 extern unsigned int ib_qib_max_mcast_qp_attached; 1162 1163 extern unsigned int ib_qib_max_srqs; 1164 1165 extern unsigned int ib_qib_max_srq_sges; 1166 1167 extern unsigned int ib_qib_max_srq_wrs; 1168 1169 extern const u32 ib_qib_rnr_table[]; 1170 1171 extern struct ib_dma_mapping_ops qib_dma_mapping_ops; 1172 1173 #endif /* QIB_VERBS_H */ 1174