2 * Broadcom NetXtreme-E RoCE driver.
4 * Copyright (c) 2016 - 2017, Broadcom. All rights reserved. The term
5 * Broadcom refers to Broadcom Limited and/or its subsidiaries.
7 * This software is available to you under a choice of one of two
8 * licenses. You may choose to be licensed under the terms of the GNU
9 * General Public License (GPL) Version 2, available from the file
10 * COPYING in the main directory of this source tree, or the
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
17 * 1. Redistributions of source code must retain the above copyright
18 * notice, this list of conditions and the following disclaimer.
19 * 2. Redistributions in binary form must reproduce the above copyright
20 * notice, this list of conditions and the following disclaimer in
21 * the documentation and/or other materials provided with the
24 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS''
25 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
26 * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
27 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS
28 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
32 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
33 * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN
34 * IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36 * Description: Fast Path Operators (header)
39 #ifndef __BNXT_QPLIB_FP_H__
40 #define __BNXT_QPLIB_FP_H__
42 #include <rdma/bnxt_re-abi.h>
44 /* Few helper structures temporarily defined here
45 * should get rid of these when roce_hsi.h is updated
46 * in original code base
48 struct sq_ud_ext_hdr {
54 struct sq_raw_ext_hdr {
60 struct sq_rdma_ext_hdr {
66 struct sq_atomic_ext_hdr {
71 struct sq_fr_pmr_ext_hdr {
76 struct sq_bind_ext_hdr {
87 /* Helper structures end */
89 struct bnxt_qplib_srq {
90 struct bnxt_qplib_pd *pd;
91 struct bnxt_qplib_dpi *dpi;
92 struct bnxt_qplib_db_info dbinfo;
100 struct bnxt_qplib_cq *cq;
101 struct bnxt_qplib_hwq hwq;
102 struct bnxt_qplib_swq *swq;
105 struct bnxt_qplib_sg_info sg_info;
106 u16 eventq_hw_ring_id;
107 spinlock_t lock; /* protect SRQE link list */
110 struct bnxt_qplib_sge {
116 #define BNXT_QPLIB_QP_MAX_SGL 6
117 struct bnxt_qplib_swq {
126 struct sq_psn_search *psn_search;
127 struct sq_psn_search_ext *psn_ext;
130 struct bnxt_qplib_swqe {
132 #define BNXT_QPLIB_FENCE_WRID 0x46454E43 /* "FENC" */
136 #define BNXT_QPLIB_SWQE_TYPE_SEND 0
137 #define BNXT_QPLIB_SWQE_TYPE_SEND_WITH_IMM 1
138 #define BNXT_QPLIB_SWQE_TYPE_SEND_WITH_INV 2
139 #define BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE 4
140 #define BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE_WITH_IMM 5
141 #define BNXT_QPLIB_SWQE_TYPE_RDMA_READ 6
142 #define BNXT_QPLIB_SWQE_TYPE_ATOMIC_CMP_AND_SWP 8
143 #define BNXT_QPLIB_SWQE_TYPE_ATOMIC_FETCH_AND_ADD 11
144 #define BNXT_QPLIB_SWQE_TYPE_LOCAL_INV 12
145 #define BNXT_QPLIB_SWQE_TYPE_FAST_REG_MR 13
146 #define BNXT_QPLIB_SWQE_TYPE_REG_MR 13
147 #define BNXT_QPLIB_SWQE_TYPE_BIND_MW 14
148 #define BNXT_QPLIB_SWQE_TYPE_RECV 128
149 #define BNXT_QPLIB_SWQE_TYPE_RECV_RDMA_IMM 129
151 #define BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP BIT(0)
152 #define BNXT_QPLIB_SWQE_FLAGS_RD_ATOMIC_FENCE BIT(1)
153 #define BNXT_QPLIB_SWQE_FLAGS_UC_FENCE BIT(2)
154 #define BNXT_QPLIB_SWQE_FLAGS_SOLICIT_EVENT BIT(3)
155 #define BNXT_QPLIB_SWQE_FLAGS_INLINE BIT(4)
156 struct bnxt_qplib_sge sg_list[BNXT_QPLIB_QP_MAX_SGL];
158 /* Max inline data is 96 bytes */
160 #define BNXT_QPLIB_SWQE_MAX_INLINE_LENGTH 96
161 u8 inline_data[BNXT_QPLIB_SWQE_MAX_INLINE_LENGTH];
164 /* Send, with imm, inval key */
175 /* Send Raw Ethernet and QP1 */
182 /* RDMA write, with imm, read */
192 /* Atomic cmp/swap, fetch/add */
200 /* Local Invalidate */
213 #define BNXT_QPLIB_SWQE_PAGE_SIZE_4K 0
214 #define BNXT_QPLIB_SWQE_PAGE_SIZE_8K 1
215 #define BNXT_QPLIB_SWQE_PAGE_SIZE_64K 4
216 #define BNXT_QPLIB_SWQE_PAGE_SIZE_256K 6
217 #define BNXT_QPLIB_SWQE_PAGE_SIZE_1M 8
218 #define BNXT_QPLIB_SWQE_PAGE_SIZE_2M 9
219 #define BNXT_QPLIB_SWQE_PAGE_SIZE_4M 10
220 #define BNXT_QPLIB_SWQE_PAGE_SIZE_1G 18
222 #define PAGE_SHIFT_4K 12
224 dma_addr_t pbl_dma_ptr;
233 #define BNXT_QPLIB_BIND_SWQE_ACCESS_LOCAL_WRITE BIT(0)
234 #define BNXT_QPLIB_BIND_SWQE_ACCESS_REMOTE_READ BIT(1)
235 #define BNXT_QPLIB_BIND_SWQE_ACCESS_REMOTE_WRITE BIT(2)
236 #define BNXT_QPLIB_BIND_SWQE_ACCESS_REMOTE_ATOMIC BIT(3)
237 #define BNXT_QPLIB_BIND_SWQE_ACCESS_WINDOW_BIND BIT(4)
248 struct bnxt_qplib_q {
249 struct bnxt_qplib_hwq hwq;
250 struct bnxt_qplib_swq *swq;
251 struct bnxt_qplib_db_info dbinfo;
252 struct bnxt_qplib_sg_info sg_info;
269 struct bnxt_qplib_qp {
270 struct bnxt_qplib_pd *pd;
271 struct bnxt_qplib_dpi *dpi;
272 struct bnxt_qplib_chip_ctx *cctx;
274 #define BNXT_QPLIB_QP_ID_INVALID 0xFFFFFFFF
285 bool en_sqd_async_notify;
296 u32 max_dest_rd_atomic;
301 struct bnxt_qplib_ah ah;
303 #define BTH_PSN_MASK ((1 << 24) - 1)
305 struct bnxt_qplib_q sq;
307 struct bnxt_qplib_q rq;
309 struct bnxt_qplib_srq *srq;
311 struct bnxt_qplib_cq *scq;
312 struct bnxt_qplib_cq *rcq;
314 struct bnxt_qplib_hwq irrq;
315 struct bnxt_qplib_hwq orrq;
316 /* Header buffer for QP1 */
320 * Buffer space for ETH(14), IP or GRH(40), UDP header(8)
321 * and ib_bth + ib_deth (20).
322 * Max required is 82 when RoCE V2 is enabled
324 #define BNXT_QPLIB_MAX_QP1_SQ_HDR_SIZE_V2 86
325 /* Ethernet header = 14 */
326 /* ib_grh = 40 (provided by MAD) */
327 /* ib_bth + ib_deth = 20 */
328 /* MAD = 256 (provided by MAD) */
330 #define BNXT_QPLIB_MAX_QP1_RQ_ETH_HDR_SIZE 14
331 #define BNXT_QPLIB_MAX_QP1_RQ_HDR_SIZE_V2 512
332 #define BNXT_QPLIB_MAX_GRH_HDR_SIZE_IPV4 20
333 #define BNXT_QPLIB_MAX_GRH_HDR_SIZE_IPV6 40
334 #define BNXT_QPLIB_MAX_QP1_RQ_BDETH_HDR_SIZE 20
336 dma_addr_t sq_hdr_buf_map;
338 dma_addr_t rq_hdr_buf_map;
339 struct list_head sq_flush;
340 struct list_head rq_flush;
343 #define BNXT_QPLIB_MAX_CQE_ENTRY_SIZE sizeof(struct cq_base)
345 #define CQE_CNT_PER_PG (PAGE_SIZE / BNXT_QPLIB_MAX_CQE_ENTRY_SIZE)
346 #define CQE_MAX_IDX_PER_PG (CQE_CNT_PER_PG - 1)
347 #define CQE_PG(x) (((x) & ~CQE_MAX_IDX_PER_PG) / CQE_CNT_PER_PG)
348 #define CQE_IDX(x) ((x) & CQE_MAX_IDX_PER_PG)
350 #define ROCE_CQE_CMP_V 0
351 #define CQE_CMP_VALID(hdr, raw_cons, cp_bit) \
352 (!!((hdr)->cqe_type_toggle & CQ_BASE_TOGGLE) == \
353 !((raw_cons) & (cp_bit)))
355 static inline bool bnxt_qplib_queue_full(struct bnxt_qplib_q *que,
358 struct bnxt_qplib_hwq *hwq;
362 /* False full is possible, retrying post-send makes sense */
363 avail = hwq->cons - hwq->prod;
364 if (hwq->cons <= hwq->prod)
366 return avail <= slots;
369 struct bnxt_qplib_cqe {
385 u16 raweth_qp1_flags;
386 u16 raweth_qp1_errors;
387 u16 raweth_qp1_cfa_code;
388 u32 raweth_qp1_flags2;
389 u32 raweth_qp1_metadata;
390 u8 raweth_qp1_payload_offset;
394 #define BNXT_QPLIB_QUEUE_START_PERIOD 0x01
395 struct bnxt_qplib_cq {
396 struct bnxt_qplib_dpi *dpi;
397 struct bnxt_qplib_db_info dbinfo;
402 struct bnxt_qplib_hwq hwq;
403 struct bnxt_qplib_hwq resize_hwq;
405 struct bnxt_qplib_nq *nq;
406 bool resize_in_progress;
407 struct bnxt_qplib_sg_info sg_info;
410 #define CQ_RESIZE_WAIT_TIME_MS 500
412 #define CQ_FLAGS_RESIZE_IN_PROG 1
413 wait_queue_head_t waitq;
414 struct list_head sqf_head, rqf_head;
416 spinlock_t compl_lock; /* synch CQ handlers */
418 * QP can move to error state from modify_qp, async error event or error
419 * CQE as part of poll_cq. When QP is moved to error state, it gets added
420 * to two flush lists, one each for SQ and RQ.
421 * Each flush list is protected by qplib_cq->flush_lock. Both scq and rcq
422 * flush_locks should be acquired when QP is moved to error. The control path
423 * operations(modify_qp and async error events) are synchronized with poll_cq
424 * using upper level CQ locks (bnxt_re_cq->cq_lock) of both SCQ and RCQ.
425 * The qplib_cq->flush_lock is required to synchronize two instances of poll_cq
426 * of the same QP while manipulating the flush list.
428 spinlock_t flush_lock; /* QP flush management */
432 #define BNXT_QPLIB_MAX_IRRQE_ENTRY_SIZE sizeof(struct xrrq_irrq)
433 #define BNXT_QPLIB_MAX_ORRQE_ENTRY_SIZE sizeof(struct xrrq_orrq)
434 #define IRD_LIMIT_TO_IRRQ_SLOTS(x) (2 * (x) + 2)
435 #define IRRQ_SLOTS_TO_IRD_LIMIT(s) (((s) >> 1) - 1)
436 #define ORD_LIMIT_TO_ORRQ_SLOTS(x) ((x) + 1)
437 #define ORRQ_SLOTS_TO_ORD_LIMIT(s) ((s) - 1)
439 #define BNXT_QPLIB_MAX_NQE_ENTRY_SIZE sizeof(struct nq_base)
441 #define NQE_CNT_PER_PG (PAGE_SIZE / BNXT_QPLIB_MAX_NQE_ENTRY_SIZE)
442 #define NQE_MAX_IDX_PER_PG (NQE_CNT_PER_PG - 1)
443 #define NQE_PG(x) (((x) & ~NQE_MAX_IDX_PER_PG) / NQE_CNT_PER_PG)
444 #define NQE_IDX(x) ((x) & NQE_MAX_IDX_PER_PG)
446 #define NQE_CMP_VALID(hdr, raw_cons, cp_bit) \
447 (!!(le32_to_cpu((hdr)->info63_v[0]) & NQ_BASE_V) == \
448 !((raw_cons) & (cp_bit)))
450 #define BNXT_QPLIB_NQE_MAX_CNT (128 * 1024)
452 #define NQ_CONS_PCI_BAR_REGION 2
453 #define NQ_DB_KEY_CP (0x2 << CMPL_DOORBELL_KEY_SFT)
454 #define NQ_DB_IDX_VALID CMPL_DOORBELL_IDX_VALID
455 #define NQ_DB_IRQ_DIS CMPL_DOORBELL_MASK
456 #define NQ_DB_CP_FLAGS_REARM (NQ_DB_KEY_CP | \
458 #define NQ_DB_CP_FLAGS (NQ_DB_KEY_CP | \
462 struct bnxt_qplib_nq_db {
463 struct bnxt_qplib_reg_desc reg;
464 struct bnxt_qplib_db_info dbinfo;
467 typedef int (*cqn_handler_t)(struct bnxt_qplib_nq *nq,
468 struct bnxt_qplib_cq *cq);
469 typedef int (*srqn_handler_t)(struct bnxt_qplib_nq *nq,
470 struct bnxt_qplib_srq *srq, u8 event);
472 struct bnxt_qplib_nq {
473 struct pci_dev *pdev;
474 struct bnxt_qplib_res *res;
476 struct bnxt_qplib_hwq hwq;
477 struct bnxt_qplib_nq_db nq_db;
481 struct tasklet_struct nq_tasklet;
485 cqn_handler_t cqn_handler;
486 srqn_handler_t srqn_handler;
487 struct workqueue_struct *cqn_wq;
490 struct bnxt_qplib_nq_work {
491 struct work_struct work;
492 struct bnxt_qplib_nq *nq;
493 struct bnxt_qplib_cq *cq;
496 void bnxt_qplib_nq_stop_irq(struct bnxt_qplib_nq *nq, bool kill);
497 void bnxt_qplib_disable_nq(struct bnxt_qplib_nq *nq);
498 int bnxt_qplib_nq_start_irq(struct bnxt_qplib_nq *nq, int nq_indx,
499 int msix_vector, bool need_init);
500 int bnxt_qplib_enable_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq,
501 int nq_idx, int msix_vector, int bar_reg_offset,
502 cqn_handler_t cqn_handler,
503 srqn_handler_t srq_handler);
504 int bnxt_qplib_create_srq(struct bnxt_qplib_res *res,
505 struct bnxt_qplib_srq *srq);
506 int bnxt_qplib_modify_srq(struct bnxt_qplib_res *res,
507 struct bnxt_qplib_srq *srq);
508 int bnxt_qplib_query_srq(struct bnxt_qplib_res *res,
509 struct bnxt_qplib_srq *srq);
510 void bnxt_qplib_destroy_srq(struct bnxt_qplib_res *res,
511 struct bnxt_qplib_srq *srq);
512 int bnxt_qplib_post_srq_recv(struct bnxt_qplib_srq *srq,
513 struct bnxt_qplib_swqe *wqe);
514 int bnxt_qplib_create_qp1(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp);
515 int bnxt_qplib_create_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp);
516 int bnxt_qplib_modify_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp);
517 int bnxt_qplib_query_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp);
518 int bnxt_qplib_destroy_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp);
519 void bnxt_qplib_clean_qp(struct bnxt_qplib_qp *qp);
520 void bnxt_qplib_free_qp_res(struct bnxt_qplib_res *res,
521 struct bnxt_qplib_qp *qp);
522 void *bnxt_qplib_get_qp1_sq_buf(struct bnxt_qplib_qp *qp,
523 struct bnxt_qplib_sge *sge);
524 void *bnxt_qplib_get_qp1_rq_buf(struct bnxt_qplib_qp *qp,
525 struct bnxt_qplib_sge *sge);
526 u32 bnxt_qplib_get_rq_prod_index(struct bnxt_qplib_qp *qp);
527 dma_addr_t bnxt_qplib_get_qp_buf_from_index(struct bnxt_qplib_qp *qp,
529 void bnxt_qplib_post_send_db(struct bnxt_qplib_qp *qp);
530 int bnxt_qplib_post_send(struct bnxt_qplib_qp *qp,
531 struct bnxt_qplib_swqe *wqe);
532 void bnxt_qplib_post_recv_db(struct bnxt_qplib_qp *qp);
533 int bnxt_qplib_post_recv(struct bnxt_qplib_qp *qp,
534 struct bnxt_qplib_swqe *wqe);
535 int bnxt_qplib_create_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq);
536 int bnxt_qplib_resize_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq,
538 void bnxt_qplib_resize_cq_complete(struct bnxt_qplib_res *res,
539 struct bnxt_qplib_cq *cq);
540 int bnxt_qplib_destroy_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq);
541 int bnxt_qplib_poll_cq(struct bnxt_qplib_cq *cq, struct bnxt_qplib_cqe *cqe,
542 int num, struct bnxt_qplib_qp **qp);
543 bool bnxt_qplib_is_cq_empty(struct bnxt_qplib_cq *cq);
544 void bnxt_qplib_req_notify_cq(struct bnxt_qplib_cq *cq, u32 arm_type);
545 void bnxt_qplib_free_nq(struct bnxt_qplib_nq *nq);
546 int bnxt_qplib_alloc_nq(struct bnxt_qplib_res *res, struct bnxt_qplib_nq *nq);
547 void bnxt_qplib_add_flush_qp(struct bnxt_qplib_qp *qp);
548 void bnxt_qplib_acquire_cq_locks(struct bnxt_qplib_qp *qp,
549 unsigned long *flags);
550 void bnxt_qplib_release_cq_locks(struct bnxt_qplib_qp *qp,
551 unsigned long *flags);
552 int bnxt_qplib_process_flush_list(struct bnxt_qplib_cq *cq,
553 struct bnxt_qplib_cqe *cqe,
555 void bnxt_qplib_flush_cqn_wq(struct bnxt_qplib_qp *qp);
557 static inline void *bnxt_qplib_get_swqe(struct bnxt_qplib_q *que, u32 *swq_idx)
561 idx = que->swq_start;
564 return &que->swq[idx];
567 static inline void bnxt_qplib_swq_mod_start(struct bnxt_qplib_q *que, u32 idx)
569 que->swq_start = que->swq[idx].next_idx;
572 static inline u32 bnxt_qplib_get_depth(struct bnxt_qplib_q *que)
574 return (que->wqe_size * que->max_wqe) / sizeof(struct sq_sge);
577 static inline u32 bnxt_qplib_set_sq_size(struct bnxt_qplib_q *que, u8 wqe_mode)
579 return (wqe_mode == BNXT_QPLIB_WQE_MODE_STATIC) ?
580 que->max_wqe : bnxt_qplib_get_depth(que);
583 static inline u32 bnxt_qplib_set_sq_max_slot(u8 wqe_mode)
585 return (wqe_mode == BNXT_QPLIB_WQE_MODE_STATIC) ?
586 sizeof(struct sq_send) / sizeof(struct sq_sge) : 1;
589 static inline u32 bnxt_qplib_set_rq_max_slot(u32 wqe_size)
591 return (wqe_size / sizeof(struct sq_sge));
594 static inline u16 __xlate_qfd(u16 delta, u16 wqe_bytes)
596 /* For Cu/Wh delta = 128, stride = 16, wqe_bytes = 128
597 * For Gen-p5 B/C mode delta = 0, stride = 16, wqe_bytes = 128.
598 * For Gen-p5 delta = 0, stride = 16, 32 <= wqe_bytes <= 512.
599 * when 8916 is disabled.
601 return (delta * wqe_bytes) / sizeof(struct sq_sge);
604 static inline u16 bnxt_qplib_calc_ilsize(struct bnxt_qplib_swqe *wqe, u16 max)
609 for (indx = 0; indx < wqe->num_sge; indx++)
610 size += wqe->sg_list[indx].size;
616 #endif /* __BNXT_QPLIB_FP_H__ */