1 /* This file is part of the Emulex RoCE Device Driver for
2 * RoCE (RDMA over Converged Ethernet) adapters.
3 * Copyright (C) 2012-2015 Emulex. All rights reserved.
4 * EMULEX and SLI are trademarks of Emulex.
7 * This software is available to you under a choice of one of two licenses.
8 * You may choose to be licensed under the terms of the GNU General Public
9 * License (GPL) Version 2, available from the file COPYING in the main
10 * directory of this source tree, or the BSD license below:
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
16 * - Redistributions of source code must retain the above copyright notice,
17 * this list of conditions and the following disclaimer.
19 * - 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 distribution.
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
24 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
27 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
31 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
32 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
33 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 * Contact Information:
36 * linux-drivers@emulex.com
40 * Costa Mesa, CA 92626
46 #include <linux/mutex.h>
47 #include <linux/list.h>
48 #include <linux/spinlock.h>
49 #include <linux/pci.h>
51 #include <rdma/ib_verbs.h>
52 #include <rdma/ib_user_verbs.h>
53 #include <rdma/ib_addr.h>
56 #include "ocrdma_sli.h"
58 #define OCRDMA_ROCE_DRV_VERSION "11.0.0.0"
60 #define OCRDMA_ROCE_DRV_DESC "Emulex OneConnect RoCE Driver"
61 #define OCRDMA_NODE_DESC "Emulex OneConnect RoCE HCA"
63 #define OC_NAME_SH OCRDMA_NODE_DESC "(Skyhawk)"
64 #define OC_NAME_UNKNOWN OCRDMA_NODE_DESC "(Unknown)"
66 #define OC_SKH_DEVICE_PF 0x720
67 #define OC_SKH_DEVICE_VF 0x728
68 #define OCRDMA_MAX_AH 512
70 #define OCRDMA_UVERBS(CMD_NAME) (1ull << IB_USER_VERBS_CMD_##CMD_NAME)
72 #define convert_to_64bit(lo, hi) ((u64)hi << 32 | (u64)lo)
73 #define EQ_INTR_PER_SEC_THRSH_HI 150000
74 #define EQ_INTR_PER_SEC_THRSH_LOW 100000
75 #define EQ_AIC_MAX_EQD 20
76 #define EQ_AIC_MIN_EQD 0
78 void ocrdma_eqd_set_task(struct work_struct *work);
80 struct ocrdma_dev_attr {
102 int max_pages_per_frmr;
106 int device_cap_flags;
107 u8 cq_overflow_detect;
113 u8 local_ca_ack_delay;
119 struct ocrdma_dma_mem {
130 struct ocrdma_queue_info {
135 u16 entry_size; /* Size of an element in the queue */
136 u16 id; /* qid, where to ring the doorbell. */
141 struct ocrdma_aic_obj { /* Adaptive interrupt coalescing (AIC) info */
144 u64 prev_eq_intr_cnt;
148 struct ocrdma_queue_info q;
151 struct ocrdma_dev *dev;
153 struct ocrdma_aic_obj aic_obj;
157 struct ocrdma_queue_info sq;
158 struct ocrdma_queue_info cq;
163 struct mutex lock; /* for serializing mailbox commands on MQ */
164 wait_queue_head_t cmd_wait;
172 struct ocrdma_hw_mr {
183 struct ocrdma_pbl *pbl_table;
193 struct ib_umem *umem;
194 struct ocrdma_hw_mr hwmr;
199 struct ocrdma_stats {
201 struct ocrdma_dev *dev;
204 struct ocrdma_pd_resource_mgr {
214 unsigned long *pd_norm_bitmap;
215 unsigned long *pd_dpp_bitmap;
216 bool pd_prealloc_valid;
220 struct ocrdma_mqe mqe;
228 u16 auto_speeds_supported;
229 u16 fixed_speeds_supported;
235 OCRDMA_FLAGS_LINK_STATUS_INIT = 0x01
239 struct ib_device ibdev;
240 struct ocrdma_dev_attr attr;
242 struct mutex dev_lock; /* provides syncronise access to device data */
243 spinlock_t flush_q_lock ____cacheline_aligned;
245 struct ocrdma_cq **cq_tbl;
246 struct ocrdma_qp **qp_tbl;
248 struct ocrdma_eq *eq_tbl;
250 struct delayed_work eqd_work;
254 /* provided synchronization to sgid table for
255 * updating gid entries triggered by notifier.
257 spinlock_t sgid_lock;
260 struct ocrdma_cq *gsi_sqcq;
261 struct ocrdma_cq *gsi_rqcq;
264 struct ocrdma_av *va;
268 /* provide synchronization for av
273 struct ocrdma_pbl pbl;
278 struct mqe_ctx mqe_ctx;
280 struct be_dev_info nic_info;
282 char model_number[32];
285 struct list_head entry;
288 u8 sl; /* service level */
295 ulong last_stats_time;
296 struct mutex stats_lock; /* provide synch for debugfs operations */
297 struct stats_mem stats_mem;
298 struct ocrdma_stats rsrc_stats;
299 struct ocrdma_stats rx_stats;
300 struct ocrdma_stats wqe_stats;
301 struct ocrdma_stats tx_stats;
302 struct ocrdma_stats db_err_stats;
303 struct ocrdma_stats tx_qp_err_stats;
304 struct ocrdma_stats rx_qp_err_stats;
305 struct ocrdma_stats tx_dbg_stats;
306 struct ocrdma_stats rx_dbg_stats;
307 struct ocrdma_stats driver_stats;
308 struct ocrdma_stats reset_stats;
310 atomic_t async_err_stats[OCRDMA_MAX_ASYNC_ERRORS];
311 atomic_t cqe_err_stats[OCRDMA_MAX_CQE_ERR];
312 struct ocrdma_pd_resource_mgr *pd_mgr;
317 struct ocrdma_cqe *va;
319 u32 getp; /* pointer to pending wrs to
320 * return to stack, wrap arounds
325 spinlock_t cq_lock ____cacheline_aligned; /* provide synchronization
328 /* syncronizes cq completion handler invoked from multiple context */
329 spinlock_t comp_handler_lock ____cacheline_aligned;
333 struct ocrdma_ucontext *ucontext;
338 /* head of all qp's sq and rq for which cqes need to be flushed
341 struct list_head sq_head, rq_head;
346 struct ocrdma_ucontext *uctx;
355 struct ocrdma_av *av;
361 struct ocrdma_qp_hwq_info {
362 u8 *va; /* virtual address */
368 u16 dbid; /* qid, where to ring the doorbell. */
376 struct ocrdma_qp_hwq_info rq;
381 /* provide synchronization to multiple context(s) posting rqe */
382 spinlock_t q_lock ____cacheline_aligned;
384 struct ocrdma_pd *pd;
392 struct ocrdma_qp_hwq_info sq;
395 uint16_t dpp_wqe_idx;
402 /* provide synchronization to multiple context(s) posting wqe, rqe */
403 spinlock_t q_lock ____cacheline_aligned;
404 struct ocrdma_cq *sq_cq;
405 /* list maintained per CQ to flush SQ errors */
406 struct list_head sq_entry;
409 struct ocrdma_qp_hwq_info rq;
411 struct ocrdma_cq *rq_cq;
412 struct ocrdma_srq *srq;
413 /* list maintained per CQ to flush RQ errors */
414 struct list_head rq_entry;
416 enum ocrdma_qp_state state; /* QP state */
418 u32 max_ord, max_ird;
421 struct ocrdma_pd *pd;
423 enum ib_qp_type qp_type;
432 struct ocrdma_ucontext {
433 struct ib_ucontext ibucontext;
435 struct list_head mm_head;
436 struct mutex mm_list_lock; /* protects list entries of mm type */
437 struct ocrdma_pd *cntxt_pd;
452 struct list_head entry;
455 static inline struct ocrdma_dev *get_ocrdma_dev(struct ib_device *ibdev)
457 return container_of(ibdev, struct ocrdma_dev, ibdev);
460 static inline struct ocrdma_ucontext *get_ocrdma_ucontext(struct ib_ucontext
463 return container_of(ibucontext, struct ocrdma_ucontext, ibucontext);
466 static inline struct ocrdma_pd *get_ocrdma_pd(struct ib_pd *ibpd)
468 return container_of(ibpd, struct ocrdma_pd, ibpd);
471 static inline struct ocrdma_cq *get_ocrdma_cq(struct ib_cq *ibcq)
473 return container_of(ibcq, struct ocrdma_cq, ibcq);
476 static inline struct ocrdma_qp *get_ocrdma_qp(struct ib_qp *ibqp)
478 return container_of(ibqp, struct ocrdma_qp, ibqp);
481 static inline struct ocrdma_mr *get_ocrdma_mr(struct ib_mr *ibmr)
483 return container_of(ibmr, struct ocrdma_mr, ibmr);
486 static inline struct ocrdma_ah *get_ocrdma_ah(struct ib_ah *ibah)
488 return container_of(ibah, struct ocrdma_ah, ibah);
491 static inline struct ocrdma_srq *get_ocrdma_srq(struct ib_srq *ibsrq)
493 return container_of(ibsrq, struct ocrdma_srq, ibsrq);
496 static inline int is_cqe_valid(struct ocrdma_cq *cq, struct ocrdma_cqe *cqe)
499 cqe_valid = le32_to_cpu(cqe->flags_status_srcqpn) & OCRDMA_CQE_VALID;
500 return (cqe_valid == cq->phase);
503 static inline int is_cqe_for_sq(struct ocrdma_cqe *cqe)
505 return (le32_to_cpu(cqe->flags_status_srcqpn) &
506 OCRDMA_CQE_QTYPE) ? 0 : 1;
509 static inline int is_cqe_invalidated(struct ocrdma_cqe *cqe)
511 return (le32_to_cpu(cqe->flags_status_srcqpn) &
512 OCRDMA_CQE_INVALIDATE) ? 1 : 0;
515 static inline int is_cqe_imm(struct ocrdma_cqe *cqe)
517 return (le32_to_cpu(cqe->flags_status_srcqpn) &
518 OCRDMA_CQE_IMM) ? 1 : 0;
521 static inline int is_cqe_wr_imm(struct ocrdma_cqe *cqe)
523 return (le32_to_cpu(cqe->flags_status_srcqpn) &
524 OCRDMA_CQE_WRITE_IMM) ? 1 : 0;
527 static inline int ocrdma_resolve_dmac(struct ocrdma_dev *dev,
528 struct rdma_ah_attr *ah_attr, u8 *mac_addr)
532 memcpy(&in6, rdma_ah_read_grh(ah_attr)->dgid.raw, sizeof(in6));
533 if (rdma_is_multicast_addr(&in6))
534 rdma_get_mcast_mac(&in6, mac_addr);
535 else if (rdma_link_local_addr(&in6))
536 rdma_get_ll_mac(&in6, mac_addr);
538 memcpy(mac_addr, ah_attr->roce.dmac, ETH_ALEN);
542 static inline char *hca_name(struct ocrdma_dev *dev)
544 switch (dev->nic_info.pdev->device) {
545 case OC_SKH_DEVICE_PF:
546 case OC_SKH_DEVICE_VF:
549 return OC_NAME_UNKNOWN;
553 static inline int ocrdma_get_eq_table_index(struct ocrdma_dev *dev,
558 for (indx = 0; indx < dev->eq_cnt; indx++) {
559 if (dev->eq_tbl[indx].q.id == eqid)
566 static inline u8 ocrdma_get_asic_type(struct ocrdma_dev *dev)
568 if (dev->nic_info.dev_family == 0xF && !dev->asic_id) {
569 pci_read_config_dword(
571 OCRDMA_SLI_ASIC_ID_OFFSET, &dev->asic_id);
574 return (dev->asic_id & OCRDMA_SLI_ASIC_GEN_NUM_MASK) >>
575 OCRDMA_SLI_ASIC_GEN_NUM_SHIFT;
578 static inline u8 ocrdma_get_pfc_prio(u8 *pfc, u8 prio)
580 return *(pfc + prio);
583 static inline u8 ocrdma_get_app_prio(u8 *app_prio, u8 prio)
585 return *(app_prio + prio);
588 static inline u8 ocrdma_is_enabled_and_synced(u32 state)
589 { /* May also be used to interpret TC-state, QCN-state
590 * Appl-state and Logical-link-state in future.
592 return (state & OCRDMA_STATE_FLAG_ENABLED) &&
593 (state & OCRDMA_STATE_FLAG_SYNC);
596 static inline u8 ocrdma_get_ae_link_state(u32 ae_state)
598 return ((ae_state & OCRDMA_AE_LSC_LS_MASK) >> OCRDMA_AE_LSC_LS_SHIFT);
601 static inline bool ocrdma_is_udp_encap_supported(struct ocrdma_dev *dev)
603 return (dev->attr.udp_encap & OCRDMA_L3_TYPE_IPV4) ||
604 (dev->attr.udp_encap & OCRDMA_L3_TYPE_IPV6);