1 /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
3 * Copyright (c) 2013-2020, Mellanox Technologies inc. All rights reserved.
4 * Copyright (c) 2020, Intel Corporation. All rights reserved.
10 #include <linux/kernel.h>
11 #include <linux/sched.h>
12 #include <rdma/ib_verbs.h>
13 #include <rdma/ib_umem.h>
14 #include <rdma/ib_smi.h>
15 #include <linux/mlx5/driver.h>
16 #include <linux/mlx5/cq.h>
17 #include <linux/mlx5/fs.h>
18 #include <linux/mlx5/qp.h>
19 #include <linux/types.h>
20 #include <linux/mlx5/transobj.h>
21 #include <rdma/ib_user_verbs.h>
22 #include <rdma/mlx5-abi.h>
23 #include <rdma/uverbs_ioctl.h>
24 #include <rdma/mlx5_user_ioctl_cmds.h>
25 #include <rdma/mlx5_user_ioctl_verbs.h>
29 #define mlx5_ib_dbg(_dev, format, arg...) \
30 dev_dbg(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__, \
31 __LINE__, current->pid, ##arg)
33 #define mlx5_ib_err(_dev, format, arg...) \
34 dev_err(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__, \
35 __LINE__, current->pid, ##arg)
37 #define mlx5_ib_warn(_dev, format, arg...) \
38 dev_warn(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__, \
39 __LINE__, current->pid, ##arg)
41 #define MLX5_IB_DEFAULT_UIDX 0xffffff
42 #define MLX5_USER_ASSIGNED_UIDX_MASK __mlx5_mask(qpc, user_index)
44 static __always_inline unsigned long
45 __mlx5_log_page_size_to_bitmap(unsigned int log_pgsz_bits,
46 unsigned int pgsz_shift)
48 unsigned int largest_pg_shift =
49 min_t(unsigned long, (1ULL << log_pgsz_bits) - 1 + pgsz_shift,
53 * Despite a command allowing it, the device does not support lower than
56 pgsz_shift = max_t(unsigned int, MLX5_ADAPTER_PAGE_SHIFT, pgsz_shift);
57 return GENMASK(largest_pg_shift, pgsz_shift);
61 * For mkc users, instead of a page_offset the command has a start_iova which
62 * specifies both the page_offset and the on-the-wire IOVA
64 #define mlx5_umem_find_best_pgsz(umem, typ, log_pgsz_fld, pgsz_shift, iova) \
65 ib_umem_find_best_pgsz(umem, \
66 __mlx5_log_page_size_to_bitmap( \
67 __mlx5_bit_sz(typ, log_pgsz_fld), \
71 static __always_inline unsigned long
72 __mlx5_page_offset_to_bitmask(unsigned int page_offset_bits,
73 unsigned int offset_shift)
75 unsigned int largest_offset_shift =
76 min_t(unsigned long, page_offset_bits - 1 + offset_shift,
79 return GENMASK(largest_offset_shift, offset_shift);
83 * QP/CQ/WQ/etc type commands take a page offset that satisifies:
84 * page_offset_quantized * (page_size/scale) = page_offset
85 * Which restricts allowed page sizes to ones that satisify the above.
87 unsigned long __mlx5_umem_find_best_quantized_pgoff(
88 struct ib_umem *umem, unsigned long pgsz_bitmap,
89 unsigned int page_offset_bits, u64 pgoff_bitmask, unsigned int scale,
90 unsigned int *page_offset_quantized);
91 #define mlx5_umem_find_best_quantized_pgoff(umem, typ, log_pgsz_fld, \
92 pgsz_shift, page_offset_fld, \
93 scale, page_offset_quantized) \
94 __mlx5_umem_find_best_quantized_pgoff( \
96 __mlx5_log_page_size_to_bitmap( \
97 __mlx5_bit_sz(typ, log_pgsz_fld), pgsz_shift), \
98 __mlx5_bit_sz(typ, page_offset_fld), \
99 GENMASK(31, order_base_2(scale)), scale, \
100 page_offset_quantized)
102 #define mlx5_umem_find_best_cq_quantized_pgoff(umem, typ, log_pgsz_fld, \
103 pgsz_shift, page_offset_fld, \
104 scale, page_offset_quantized) \
105 __mlx5_umem_find_best_quantized_pgoff( \
107 __mlx5_log_page_size_to_bitmap( \
108 __mlx5_bit_sz(typ, log_pgsz_fld), pgsz_shift), \
109 __mlx5_bit_sz(typ, page_offset_fld), 0, scale, \
110 page_offset_quantized)
113 MLX5_IB_MMAP_OFFSET_START = 9,
114 MLX5_IB_MMAP_OFFSET_END = 255,
118 MLX5_IB_MMAP_CMD_SHIFT = 8,
119 MLX5_IB_MMAP_CMD_MASK = 0xff,
123 MLX5_RES_SCAT_DATA32_CQE = 0x1,
124 MLX5_RES_SCAT_DATA64_CQE = 0x2,
125 MLX5_REQ_SCAT_DATA32_CQE = 0x11,
126 MLX5_REQ_SCAT_DATA64_CQE = 0x22,
129 enum mlx5_ib_mad_ifc_flags {
130 MLX5_MAD_IFC_IGNORE_MKEY = 1,
131 MLX5_MAD_IFC_IGNORE_BKEY = 2,
132 MLX5_MAD_IFC_NET_VIEW = 4,
136 MLX5_CROSS_CHANNEL_BFREG = 0,
145 MLX5_TM_MAX_RNDV_MSG_SIZE = 64,
150 MLX5_IB_INVALID_UAR_INDEX = BIT(31),
151 MLX5_IB_INVALID_BFREG = BIT(31),
155 MLX5_MAX_MEMIC_PAGES = 0x100,
156 MLX5_MEMIC_ALLOC_SIZE_MASK = 0x3f,
160 MLX5_MEMIC_BASE_ALIGN = 6,
161 MLX5_MEMIC_BASE_SIZE = 1 << MLX5_MEMIC_BASE_ALIGN,
164 enum mlx5_ib_mmap_type {
165 MLX5_IB_MMAP_TYPE_MEMIC = 1,
166 MLX5_IB_MMAP_TYPE_VAR = 2,
167 MLX5_IB_MMAP_TYPE_UAR_WC = 3,
168 MLX5_IB_MMAP_TYPE_UAR_NC = 4,
171 struct mlx5_bfreg_info {
173 int num_low_latency_bfregs;
177 * protect bfreg allocation data structs
184 u32 num_static_sys_pages;
185 u32 total_num_bfregs;
189 struct mlx5_ib_ucontext {
190 struct ib_ucontext ibucontext;
191 struct list_head db_page_list;
193 /* protect doorbell record alloc/free
195 struct mutex db_page_mutex;
196 struct mlx5_bfreg_info bfregi;
198 /* Transport Domain number */
203 /* For RoCE LAG TX affinity */
204 atomic_t tx_port_affinity;
207 static inline struct mlx5_ib_ucontext *to_mucontext(struct ib_ucontext *ibucontext)
209 return container_of(ibucontext, struct mlx5_ib_ucontext, ibucontext);
219 MLX5_IB_FLOW_ACTION_MODIFY_HEADER,
220 MLX5_IB_FLOW_ACTION_PACKET_REFORMAT,
221 MLX5_IB_FLOW_ACTION_DECAP,
224 #define MLX5_IB_FLOW_MCAST_PRIO (MLX5_BY_PASS_NUM_PRIOS - 1)
225 #define MLX5_IB_FLOW_LAST_PRIO (MLX5_BY_PASS_NUM_REGULAR_PRIOS - 1)
226 #if (MLX5_IB_FLOW_LAST_PRIO <= 0)
227 #error "Invalid number of bypass priorities"
229 #define MLX5_IB_FLOW_LEFTOVERS_PRIO (MLX5_IB_FLOW_MCAST_PRIO + 1)
231 #define MLX5_IB_NUM_FLOW_FT (MLX5_IB_FLOW_LEFTOVERS_PRIO + 1)
232 #define MLX5_IB_NUM_SNIFFER_FTS 2
233 #define MLX5_IB_NUM_EGRESS_FTS 1
234 struct mlx5_ib_flow_prio {
235 struct mlx5_flow_table *flow_table;
236 unsigned int refcount;
239 struct mlx5_ib_flow_handler {
240 struct list_head list;
241 struct ib_flow ibflow;
242 struct mlx5_ib_flow_prio *prio;
243 struct mlx5_flow_handle *rule;
244 struct ib_counters *ibcounters;
245 struct mlx5_ib_dev *dev;
246 struct mlx5_ib_flow_matcher *flow_matcher;
249 struct mlx5_ib_flow_matcher {
250 struct mlx5_ib_match_params matcher_mask;
252 enum mlx5_ib_flow_type flow_type;
253 enum mlx5_flow_namespace_type ns_type;
255 struct mlx5_core_dev *mdev;
257 u8 match_criteria_enable;
262 struct mlx5_core_dev *mdev;
265 struct mlx5_ib_flow_db {
266 struct mlx5_ib_flow_prio prios[MLX5_IB_NUM_FLOW_FT];
267 struct mlx5_ib_flow_prio egress_prios[MLX5_IB_NUM_FLOW_FT];
268 struct mlx5_ib_flow_prio sniffer[MLX5_IB_NUM_SNIFFER_FTS];
269 struct mlx5_ib_flow_prio egress[MLX5_IB_NUM_EGRESS_FTS];
270 struct mlx5_ib_flow_prio fdb;
271 struct mlx5_ib_flow_prio rdma_rx[MLX5_IB_NUM_FLOW_FT];
272 struct mlx5_ib_flow_prio rdma_tx[MLX5_IB_NUM_FLOW_FT];
273 struct mlx5_flow_table *lag_demux_ft;
274 /* Protect flow steering bypass flow tables
275 * when add/del flow rules.
276 * only single add/removal of flow steering rule could be done
282 /* Use macros here so that don't have to duplicate
283 * enum ib_send_flags and enum ib_qp_type for low-level driver
286 #define MLX5_IB_SEND_UMR_ENABLE_MR (IB_SEND_RESERVED_START << 0)
287 #define MLX5_IB_SEND_UMR_DISABLE_MR (IB_SEND_RESERVED_START << 1)
288 #define MLX5_IB_SEND_UMR_FAIL_IF_FREE (IB_SEND_RESERVED_START << 2)
289 #define MLX5_IB_SEND_UMR_UPDATE_XLT (IB_SEND_RESERVED_START << 3)
290 #define MLX5_IB_SEND_UMR_UPDATE_TRANSLATION (IB_SEND_RESERVED_START << 4)
291 #define MLX5_IB_SEND_UMR_UPDATE_PD_ACCESS IB_SEND_RESERVED_END
293 #define MLX5_IB_QPT_REG_UMR IB_QPT_RESERVED1
295 * IB_QPT_GSI creates the software wrapper around GSI, and MLX5_IB_QPT_HW_GSI
296 * creates the actual hardware QP.
298 #define MLX5_IB_QPT_HW_GSI IB_QPT_RESERVED2
299 #define MLX5_IB_QPT_DCI IB_QPT_RESERVED3
300 #define MLX5_IB_QPT_DCT IB_QPT_RESERVED4
301 #define MLX5_IB_WR_UMR IB_WR_RESERVED1
303 #define MLX5_IB_UMR_OCTOWORD 16
304 #define MLX5_IB_UMR_XLT_ALIGNMENT 64
306 #define MLX5_IB_UPD_XLT_ZAP BIT(0)
307 #define MLX5_IB_UPD_XLT_ENABLE BIT(1)
308 #define MLX5_IB_UPD_XLT_ATOMIC BIT(2)
309 #define MLX5_IB_UPD_XLT_ADDR BIT(3)
310 #define MLX5_IB_UPD_XLT_PD BIT(4)
311 #define MLX5_IB_UPD_XLT_ACCESS BIT(5)
312 #define MLX5_IB_UPD_XLT_INDIRECT BIT(6)
314 /* Private QP creation flags to be passed in ib_qp_init_attr.create_flags.
316 * These flags are intended for internal use by the mlx5_ib driver, and they
317 * rely on the range reserved for that use in the ib_qp_create_flags enum.
319 #define MLX5_IB_QP_CREATE_SQPN_QP1 IB_QP_CREATE_RESERVED_START
320 #define MLX5_IB_QP_CREATE_WC_TEST (IB_QP_CREATE_RESERVED_START << 1)
327 enum mlx5_ib_rq_flags {
328 MLX5_IB_RQ_CVLAN_STRIPPING = 1 << 0,
329 MLX5_IB_RQ_PCI_WRITE_END_PADDING = 1 << 1,
333 struct mlx5_frag_buf_ctrl fbc;
336 struct wr_list *w_list;
340 /* serialize post to the work queue
355 enum mlx5_ib_wq_flags {
356 MLX5_IB_WQ_FLAGS_DELAY_DROP = 0x1,
357 MLX5_IB_WQ_FLAGS_STRIDING_RQ = 0x2,
360 #define MLX5_MIN_SINGLE_WQE_LOG_NUM_STRIDES 9
361 #define MLX5_MAX_SINGLE_WQE_LOG_NUM_STRIDES 16
362 #define MLX5_MIN_SINGLE_STRIDE_LOG_NUM_BYTES 6
363 #define MLX5_MAX_SINGLE_STRIDE_LOG_NUM_BYTES 13
364 #define MLX5_EXT_MIN_SINGLE_WQE_LOG_NUM_STRIDES 3
368 struct mlx5_core_qp core_qp;
375 u32 two_byte_shift_en;
376 u32 single_stride_log_num_of_bytes;
377 struct ib_umem *umem;
379 unsigned int page_shift;
385 u32 create_flags; /* Use enum mlx5_ib_wq_flags */
388 struct mlx5_ib_rwq_ind_table {
389 struct ib_rwq_ind_table ib_rwq_ind_tbl;
394 struct mlx5_ib_ubuffer {
395 struct ib_umem *umem;
400 struct mlx5_ib_qp_base {
401 struct mlx5_ib_qp *container_mibqp;
402 struct mlx5_core_qp mqp;
403 struct mlx5_ib_ubuffer ubuffer;
406 struct mlx5_ib_qp_trans {
407 struct mlx5_ib_qp_base base;
414 struct mlx5_ib_rss_qp {
419 struct mlx5_ib_qp_base base;
420 struct mlx5_ib_wq *rq;
421 struct mlx5_ib_ubuffer ubuffer;
422 struct mlx5_db *doorbell;
429 struct mlx5_ib_qp_base base;
430 struct mlx5_ib_wq *sq;
431 struct mlx5_ib_ubuffer ubuffer;
432 struct mlx5_db *doorbell;
433 struct mlx5_flow_handle *flow_rule;
438 struct mlx5_ib_raw_packet_qp {
439 struct mlx5_ib_sq sq;
440 struct mlx5_ib_rq rq;
445 unsigned long offset;
446 struct mlx5_sq_bfreg *bfreg;
450 struct mlx5_core_dct mdct;
454 struct mlx5_ib_gsi_qp {
457 struct ib_qp_cap cap;
459 struct mlx5_ib_gsi_wr *outstanding_wrs;
460 u32 outstanding_pi, outstanding_ci;
462 /* Protects access to the tx_qps. Post send operations synchronize
463 * with tx_qp creation in setup_qp(). Also protects the
464 * outstanding_wrs array and indices.
467 struct ib_qp **tx_qps;
473 struct mlx5_ib_qp_trans trans_qp;
474 struct mlx5_ib_raw_packet_qp raw_packet_qp;
475 struct mlx5_ib_rss_qp rss_qp;
476 struct mlx5_ib_dct dct;
477 struct mlx5_ib_gsi_qp gsi;
479 struct mlx5_frag_buf buf;
482 struct mlx5_ib_wq rq;
486 struct mlx5_ib_wq sq;
488 /* serialize qp state modifications
491 /* cached variant of create_flags from struct ib_qp_init_attr */
500 /* only for user space QPs. For kernel
501 * we have it from the bf object
505 struct list_head qps_list;
506 struct list_head cq_recv_list;
507 struct list_head cq_send_list;
508 struct mlx5_rate_limit rl;
512 * IB/core doesn't store low-level QP types, so
513 * store both MLX and IBTA types in the field below.
514 * IB_QPT_DRIVER will be break to DCI/DCT subtypes.
516 enum ib_qp_type type;
517 /* A flag to indicate if there's a new counter is configured
518 * but not take effective
524 struct mlx5_ib_cq_buf {
525 struct mlx5_frag_buf_ctrl fbc;
526 struct mlx5_frag_buf frag_buf;
527 struct ib_umem *umem;
533 struct ib_send_wr wr;
537 unsigned int page_shift;
538 unsigned int xlt_size;
542 u8 ignore_free_state:1;
545 static inline const struct mlx5_umr_wr *umr_wr(const struct ib_send_wr *wr)
547 return container_of(wr, struct mlx5_umr_wr, wr);
550 enum mlx5_ib_cq_pr_flags {
551 MLX5_IB_CQ_PR_FLAGS_CQE_128_PAD = 1 << 0,
556 struct mlx5_core_cq mcq;
557 struct mlx5_ib_cq_buf buf;
560 /* serialize access to the CQ
566 struct mutex resize_mutex;
567 struct mlx5_ib_cq_buf *resize_buf;
568 struct ib_umem *resize_umem;
570 struct list_head list_send_qp;
571 struct list_head list_recv_qp;
573 struct list_head wc_list;
574 enum ib_cq_notify_flags notify_flags;
575 struct work_struct notify_work;
576 u16 private_flags; /* Use mlx5_ib_cq_pr_flags */
581 struct list_head list;
586 struct mlx5_core_srq msrq;
587 struct mlx5_frag_buf buf;
589 struct mlx5_frag_buf_ctrl fbc;
591 /* protect SRQ hanlding
597 struct ib_umem *umem;
598 /* serialize arming a SRQ
604 struct mlx5_ib_xrcd {
605 struct ib_xrcd ibxrcd;
609 enum mlx5_ib_mtt_access_flags {
610 MLX5_IB_MTT_READ = (1 << 0),
611 MLX5_IB_MTT_WRITE = (1 << 1),
614 struct mlx5_user_mmap_entry {
615 struct rdma_user_mmap_entry rdma_entry;
623 phys_addr_t dev_addr;
630 /* other dm types specific params should be added here */
632 struct mlx5_user_mmap_entry mentry;
635 #define MLX5_IB_MTT_PRESENT (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE)
637 #define MLX5_IB_DM_MEMIC_ALLOWED_ACCESS (IB_ACCESS_LOCAL_WRITE |\
638 IB_ACCESS_REMOTE_WRITE |\
639 IB_ACCESS_REMOTE_READ |\
640 IB_ACCESS_REMOTE_ATOMIC |\
643 #define MLX5_IB_DM_SW_ICM_ALLOWED_ACCESS (IB_ACCESS_LOCAL_WRITE |\
644 IB_ACCESS_REMOTE_WRITE |\
645 IB_ACCESS_REMOTE_READ |\
648 #define mlx5_update_odp_stats(mr, counter_name, value) \
649 atomic64_add(value, &((mr)->odp_stats.counter_name))
653 struct mlx5_core_mkey mmkey;
656 struct mlx5_cache_ent *cache_ent;
657 struct ib_umem *umem;
659 /* This is zero'd when the MR is allocated */
661 /* Used only while the MR is in the cache */
663 u32 out[MLX5_ST_SZ_DW(create_mkey_out)];
664 struct mlx5_async_work cb_work;
665 /* Cache list element */
666 struct list_head list;
669 /* Used only by kernel MRs (umem == NULL) */
679 /* For Kernel IB_MR_TYPE_INTEGRITY */
680 struct mlx5_core_sig_ctx *sig;
681 struct mlx5_ib_mr *pi_mr;
682 struct mlx5_ib_mr *klm_mr;
683 struct mlx5_ib_mr *mtt_mr;
691 /* Used only by User MRs (umem != NULL) */
693 unsigned int page_shift;
694 /* Current access_flags */
698 struct mlx5_ib_mr *parent;
699 struct xarray implicit_children;
701 struct work_struct work;
703 struct ib_odp_counters odp_stats;
704 bool is_odp_implicit;
709 /* Zero the fields in the mr that are variant depending on usage */
710 static inline void mlx5_clear_mr(struct mlx5_ib_mr *mr)
712 memset(mr->out, 0, sizeof(*mr) - offsetof(struct mlx5_ib_mr, out));
715 static inline bool is_odp_mr(struct mlx5_ib_mr *mr)
717 return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem &&
721 static inline bool is_dmabuf_mr(struct mlx5_ib_mr *mr)
723 return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem &&
729 struct mlx5_core_mkey mmkey;
733 struct mlx5_ib_devx_mr {
734 struct mlx5_core_mkey mmkey;
738 struct mlx5_ib_umr_context {
740 enum ib_wc_status status;
741 struct completion done;
748 /* control access to UMR QP
750 struct semaphore sem;
753 struct mlx5_cache_ent {
754 struct list_head head;
755 /* sync access to the cahce entry
767 u8 fill_to_high_water:1;
770 * - available_mrs is the length of list head, ie the number of MRs
771 * available for immediate allocation.
772 * - total_mrs is available_mrs plus all in use MRs that could be
773 * returned to the cache.
774 * - limit is the low water mark for available_mrs, 2* limit is the
776 * - pending is the number of MRs currently being created
786 struct mlx5_ib_dev *dev;
787 struct work_struct work;
788 struct delayed_work dwork;
791 struct mlx5_mr_cache {
792 struct workqueue_struct *wq;
793 struct mlx5_cache_ent ent[MAX_MR_CACHE_ENTRIES];
795 unsigned long last_add;
798 struct mlx5_ib_port_resources {
799 struct mlx5_ib_gsi_qp *gsi;
800 struct work_struct pkey_change_work;
803 struct mlx5_ib_resources {
810 struct mlx5_ib_port_resources ports[2];
811 /* Protects changes to the port resources */
815 struct mlx5_ib_counters {
819 u32 num_cong_counters;
820 u32 num_ext_ppcnt_counters;
824 struct mlx5_ib_multiport_info;
826 struct mlx5_ib_multiport {
827 struct mlx5_ib_multiport_info *mpi;
828 /* To be held when accessing the multiport info */
833 /* Protect mlx5_ib_get_netdev from invoking dev_hold() with a NULL
836 rwlock_t netdev_lock;
837 struct net_device *netdev;
838 struct notifier_block nb;
839 atomic_t tx_port_affinity;
840 enum ib_port_state last_port_state;
841 struct mlx5_ib_dev *dev;
845 struct mlx5_ib_port {
846 struct mlx5_ib_counters cnts;
847 struct mlx5_ib_multiport mp;
848 struct mlx5_ib_dbg_cc_params *dbg_cc_params;
849 struct mlx5_roce roce;
850 struct mlx5_eswitch_rep *rep;
853 struct mlx5_ib_dbg_param {
855 struct mlx5_ib_dev *dev;
856 struct dentry *dentry;
860 enum mlx5_ib_dbg_cc_types {
861 MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE,
862 MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE_ATI,
863 MLX5_IB_DBG_CC_RP_TIME_RESET,
864 MLX5_IB_DBG_CC_RP_BYTE_RESET,
865 MLX5_IB_DBG_CC_RP_THRESHOLD,
866 MLX5_IB_DBG_CC_RP_AI_RATE,
867 MLX5_IB_DBG_CC_RP_MAX_RATE,
868 MLX5_IB_DBG_CC_RP_HAI_RATE,
869 MLX5_IB_DBG_CC_RP_MIN_DEC_FAC,
870 MLX5_IB_DBG_CC_RP_MIN_RATE,
871 MLX5_IB_DBG_CC_RP_RATE_TO_SET_ON_FIRST_CNP,
872 MLX5_IB_DBG_CC_RP_DCE_TCP_G,
873 MLX5_IB_DBG_CC_RP_DCE_TCP_RTT,
874 MLX5_IB_DBG_CC_RP_RATE_REDUCE_MONITOR_PERIOD,
875 MLX5_IB_DBG_CC_RP_INITIAL_ALPHA_VALUE,
876 MLX5_IB_DBG_CC_RP_GD,
877 MLX5_IB_DBG_CC_NP_MIN_TIME_BETWEEN_CNPS,
878 MLX5_IB_DBG_CC_NP_CNP_DSCP,
879 MLX5_IB_DBG_CC_NP_CNP_PRIO_MODE,
880 MLX5_IB_DBG_CC_NP_CNP_PRIO,
884 struct mlx5_ib_dbg_cc_params {
886 struct mlx5_ib_dbg_param params[MLX5_IB_DBG_CC_MAX];
890 MLX5_MAX_DELAY_DROP_TIMEOUT_MS = 100,
893 struct mlx5_ib_delay_drop {
894 struct mlx5_ib_dev *dev;
895 struct work_struct delay_drop_work;
896 /* serialize setting of delay drop */
902 struct dentry *dir_debugfs;
905 enum mlx5_ib_stages {
909 MLX5_IB_STAGE_NON_DEFAULT_CB,
913 MLX5_IB_STAGE_DEVICE_RESOURCES,
914 MLX5_IB_STAGE_DEVICE_NOTIFIER,
916 MLX5_IB_STAGE_COUNTERS,
917 MLX5_IB_STAGE_CONG_DEBUGFS,
920 MLX5_IB_STAGE_PRE_IB_REG_UMR,
921 MLX5_IB_STAGE_WHITELIST_UID,
922 MLX5_IB_STAGE_IB_REG,
923 MLX5_IB_STAGE_POST_IB_REG_UMR,
924 MLX5_IB_STAGE_DELAY_DROP,
925 MLX5_IB_STAGE_RESTRACK,
929 struct mlx5_ib_stage {
930 int (*init)(struct mlx5_ib_dev *dev);
931 void (*cleanup)(struct mlx5_ib_dev *dev);
934 #define STAGE_CREATE(_stage, _init, _cleanup) \
935 .stage[_stage] = {.init = _init, .cleanup = _cleanup}
937 struct mlx5_ib_profile {
938 struct mlx5_ib_stage stage[MLX5_IB_STAGE_MAX];
941 struct mlx5_ib_multiport_info {
942 struct list_head list;
943 struct mlx5_ib_dev *ibdev;
944 struct mlx5_core_dev *mdev;
945 struct notifier_block mdev_events;
946 struct completion unref_comp;
953 struct mlx5_ib_flow_action {
954 struct ib_flow_action ib_action;
958 struct mlx5_accel_esp_xfrm *ctx;
961 struct mlx5_ib_dev *dev;
964 struct mlx5_modify_hdr *modify_hdr;
965 struct mlx5_pkt_reformat *pkt_reformat;
972 struct mlx5_core_dev *dev;
973 /* This lock is used to protect the access to the shared
974 * allocation map when concurrent requests by different
975 * processes are handled.
978 DECLARE_BITMAP(memic_alloc_pages, MLX5_MAX_MEMIC_PAGES);
981 struct mlx5_read_counters_attr {
982 struct mlx5_fc *hw_cntrs_hndl;
987 enum mlx5_ib_counters_type {
988 MLX5_IB_COUNTERS_FLOW,
991 struct mlx5_ib_mcounters {
992 struct ib_counters ibcntrs;
993 enum mlx5_ib_counters_type type;
994 /* number of counters supported for this counters type */
996 struct mlx5_fc *hw_cntrs_hndl;
997 /* read function for this counters type */
998 int (*read_counters)(struct ib_device *ibdev,
999 struct mlx5_read_counters_attr *read_attr);
1000 /* max index set as part of create_flow */
1001 u32 cntrs_max_index;
1002 /* number of counters data entries (<description,index> pair) */
1004 /* counters data array for descriptions and indexes */
1005 struct mlx5_ib_flow_counters_desc *counters_data;
1006 /* protects access to mcounters internal data */
1007 struct mutex mcntrs_mutex;
1010 static inline struct mlx5_ib_mcounters *
1011 to_mcounters(struct ib_counters *ibcntrs)
1013 return container_of(ibcntrs, struct mlx5_ib_mcounters, ibcntrs);
1016 int parse_flow_flow_action(struct mlx5_ib_flow_action *maction,
1018 struct mlx5_flow_act *action);
1019 struct mlx5_ib_lb_state {
1020 /* protect the user_td */
1027 struct mlx5_ib_pf_eq {
1028 struct notifier_block irq_nb;
1029 struct mlx5_ib_dev *dev;
1030 struct mlx5_eq *core;
1031 struct work_struct work;
1032 spinlock_t lock; /* Pagefaults spinlock */
1033 struct workqueue_struct *wq;
1037 struct mlx5_devx_event_table {
1038 struct mlx5_nb devx_nb;
1039 /* serialize updating the event_xa */
1040 struct mutex event_xa_lock;
1041 struct xarray event_xa;
1044 struct mlx5_var_table {
1045 /* serialize updating the bitmap */
1046 struct mutex bitmap_lock;
1047 unsigned long *bitmap;
1050 u64 num_var_hw_entries;
1053 struct mlx5_port_caps {
1058 struct mlx5_ib_dev {
1059 struct ib_device ib_dev;
1060 struct mlx5_core_dev *mdev;
1061 struct notifier_block mdev_events;
1063 /* serialize update of capability mask
1065 struct mutex cap_mask_mutex;
1071 struct umr_common umrc;
1072 /* sync used page count stats
1074 struct mlx5_ib_resources devr;
1077 struct mlx5_mr_cache cache;
1078 struct timer_list delay_timer;
1079 /* Prevents soft lock on massive reg MRs */
1080 struct mutex slow_path_mutex;
1081 struct ib_odp_caps odp_caps;
1083 struct mlx5_ib_pf_eq odp_pf_eq;
1085 struct xarray odp_mkeys;
1088 struct mlx5_ib_flow_db *flow_db;
1089 /* protect resources needed as part of reset flow */
1090 spinlock_t reset_flow_resource_lock;
1091 struct list_head qp_list;
1092 /* Array with num_ports elements */
1093 struct mlx5_ib_port *port;
1094 struct mlx5_sq_bfreg bfreg;
1095 struct mlx5_sq_bfreg wc_bfreg;
1096 struct mlx5_sq_bfreg fp_bfreg;
1097 struct mlx5_ib_delay_drop delay_drop;
1098 const struct mlx5_ib_profile *profile;
1100 struct mlx5_ib_lb_state lb;
1102 struct list_head ib_dev_list;
1105 u16 devx_whitelist_uid;
1106 struct mlx5_srq_table srq_table;
1107 struct mlx5_qp_table qp_table;
1108 struct mlx5_async_ctx async_ctx;
1109 struct mlx5_devx_event_table devx_event_table;
1110 struct mlx5_var_table var_table;
1112 struct xarray sig_mrs;
1113 struct mlx5_port_caps port_caps[MLX5_MAX_PORTS];
1117 static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq)
1119 return container_of(mcq, struct mlx5_ib_cq, mcq);
1122 static inline struct mlx5_ib_xrcd *to_mxrcd(struct ib_xrcd *ibxrcd)
1124 return container_of(ibxrcd, struct mlx5_ib_xrcd, ibxrcd);
1127 static inline struct mlx5_ib_dev *to_mdev(struct ib_device *ibdev)
1129 return container_of(ibdev, struct mlx5_ib_dev, ib_dev);
1132 static inline struct mlx5_ib_dev *mr_to_mdev(struct mlx5_ib_mr *mr)
1134 return to_mdev(mr->ibmr.device);
1137 static inline struct mlx5_ib_dev *mlx5_udata_to_mdev(struct ib_udata *udata)
1139 struct mlx5_ib_ucontext *context = rdma_udata_to_drv_context(
1140 udata, struct mlx5_ib_ucontext, ibucontext);
1142 return to_mdev(context->ibucontext.device);
1145 static inline struct mlx5_ib_cq *to_mcq(struct ib_cq *ibcq)
1147 return container_of(ibcq, struct mlx5_ib_cq, ibcq);
1150 static inline struct mlx5_ib_qp *to_mibqp(struct mlx5_core_qp *mqp)
1152 return container_of(mqp, struct mlx5_ib_qp_base, mqp)->container_mibqp;
1155 static inline struct mlx5_ib_rwq *to_mibrwq(struct mlx5_core_qp *core_qp)
1157 return container_of(core_qp, struct mlx5_ib_rwq, core_qp);
1160 static inline struct mlx5_ib_pd *to_mpd(struct ib_pd *ibpd)
1162 return container_of(ibpd, struct mlx5_ib_pd, ibpd);
1165 static inline struct mlx5_ib_srq *to_msrq(struct ib_srq *ibsrq)
1167 return container_of(ibsrq, struct mlx5_ib_srq, ibsrq);
1170 static inline struct mlx5_ib_qp *to_mqp(struct ib_qp *ibqp)
1172 return container_of(ibqp, struct mlx5_ib_qp, ibqp);
1175 static inline struct mlx5_ib_rwq *to_mrwq(struct ib_wq *ibwq)
1177 return container_of(ibwq, struct mlx5_ib_rwq, ibwq);
1180 static inline struct mlx5_ib_rwq_ind_table *to_mrwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_tbl)
1182 return container_of(ib_rwq_ind_tbl, struct mlx5_ib_rwq_ind_table, ib_rwq_ind_tbl);
1185 static inline struct mlx5_ib_srq *to_mibsrq(struct mlx5_core_srq *msrq)
1187 return container_of(msrq, struct mlx5_ib_srq, msrq);
1190 static inline struct mlx5_ib_dm *to_mdm(struct ib_dm *ibdm)
1192 return container_of(ibdm, struct mlx5_ib_dm, ibdm);
1195 static inline struct mlx5_ib_mr *to_mmr(struct ib_mr *ibmr)
1197 return container_of(ibmr, struct mlx5_ib_mr, ibmr);
1200 static inline struct mlx5_ib_mw *to_mmw(struct ib_mw *ibmw)
1202 return container_of(ibmw, struct mlx5_ib_mw, ibmw);
1205 static inline struct mlx5_ib_flow_action *
1206 to_mflow_act(struct ib_flow_action *ibact)
1208 return container_of(ibact, struct mlx5_ib_flow_action, ib_action);
1211 static inline struct mlx5_user_mmap_entry *
1212 to_mmmap(struct rdma_user_mmap_entry *rdma_entry)
1214 return container_of(rdma_entry,
1215 struct mlx5_user_mmap_entry, rdma_entry);
1218 int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context,
1219 struct ib_udata *udata, unsigned long virt,
1220 struct mlx5_db *db);
1221 void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db);
1222 void __mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
1223 void mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
1224 void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index);
1225 int mlx5_ib_create_ah(struct ib_ah *ah, struct rdma_ah_init_attr *init_attr,
1226 struct ib_udata *udata);
1227 int mlx5_ib_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr);
1228 static inline int mlx5_ib_destroy_ah(struct ib_ah *ah, u32 flags)
1232 int mlx5_ib_create_srq(struct ib_srq *srq, struct ib_srq_init_attr *init_attr,
1233 struct ib_udata *udata);
1234 int mlx5_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
1235 enum ib_srq_attr_mask attr_mask, struct ib_udata *udata);
1236 int mlx5_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr);
1237 int mlx5_ib_destroy_srq(struct ib_srq *srq, struct ib_udata *udata);
1238 int mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr,
1239 const struct ib_recv_wr **bad_wr);
1240 int mlx5_ib_enable_lb(struct mlx5_ib_dev *dev, bool td, bool qp);
1241 void mlx5_ib_disable_lb(struct mlx5_ib_dev *dev, bool td, bool qp);
1242 struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
1243 struct ib_qp_init_attr *init_attr,
1244 struct ib_udata *udata);
1245 int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1246 int attr_mask, struct ib_udata *udata);
1247 int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask,
1248 struct ib_qp_init_attr *qp_init_attr);
1249 int mlx5_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata);
1250 void mlx5_ib_drain_sq(struct ib_qp *qp);
1251 void mlx5_ib_drain_rq(struct ib_qp *qp);
1252 int mlx5_ib_read_wqe_sq(struct mlx5_ib_qp *qp, int wqe_index, void *buffer,
1253 size_t buflen, size_t *bc);
1254 int mlx5_ib_read_wqe_rq(struct mlx5_ib_qp *qp, int wqe_index, void *buffer,
1255 size_t buflen, size_t *bc);
1256 int mlx5_ib_read_wqe_srq(struct mlx5_ib_srq *srq, int wqe_index, void *buffer,
1257 size_t buflen, size_t *bc);
1258 int mlx5_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
1259 struct ib_udata *udata);
1260 int mlx5_ib_destroy_cq(struct ib_cq *cq, struct ib_udata *udata);
1261 int mlx5_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
1262 int mlx5_ib_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags);
1263 int mlx5_ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1264 int mlx5_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata);
1265 struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc);
1266 struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
1267 u64 virt_addr, int access_flags,
1268 struct ib_udata *udata);
1269 struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 start,
1270 u64 length, u64 virt_addr,
1271 int fd, int access_flags,
1272 struct ib_udata *udata);
1273 int mlx5_ib_advise_mr(struct ib_pd *pd,
1274 enum ib_uverbs_advise_mr_advice advice,
1276 struct ib_sge *sg_list,
1278 struct uverbs_attr_bundle *attrs);
1279 int mlx5_ib_alloc_mw(struct ib_mw *mw, struct ib_udata *udata);
1280 int mlx5_ib_dealloc_mw(struct ib_mw *mw);
1281 int mlx5_ib_update_xlt(struct mlx5_ib_mr *mr, u64 idx, int npages,
1282 int page_shift, int flags);
1283 int mlx5_ib_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags);
1284 struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd,
1285 struct ib_udata *udata,
1287 void mlx5_ib_free_implicit_mr(struct mlx5_ib_mr *mr);
1288 void mlx5_ib_fence_odp_mr(struct mlx5_ib_mr *mr);
1289 void mlx5_ib_fence_dmabuf_mr(struct mlx5_ib_mr *mr);
1290 struct ib_mr *mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start,
1291 u64 length, u64 virt_addr, int access_flags,
1292 struct ib_pd *pd, struct ib_udata *udata);
1293 int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata);
1294 struct ib_mr *mlx5_ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
1296 struct ib_mr *mlx5_ib_alloc_mr_integrity(struct ib_pd *pd,
1298 u32 max_num_meta_sg);
1299 int mlx5_ib_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, int sg_nents,
1300 unsigned int *sg_offset);
1301 int mlx5_ib_map_mr_sg_pi(struct ib_mr *ibmr, struct scatterlist *data_sg,
1302 int data_sg_nents, unsigned int *data_sg_offset,
1303 struct scatterlist *meta_sg, int meta_sg_nents,
1304 unsigned int *meta_sg_offset);
1305 int mlx5_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
1306 const struct ib_wc *in_wc, const struct ib_grh *in_grh,
1307 const struct ib_mad *in, struct ib_mad *out,
1308 size_t *out_mad_size, u16 *out_mad_pkey_index);
1309 int mlx5_ib_alloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
1310 int mlx5_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
1311 int mlx5_query_ext_port_caps(struct mlx5_ib_dev *dev, unsigned int port);
1312 int mlx5_query_mad_ifc_system_image_guid(struct ib_device *ibdev,
1313 __be64 *sys_image_guid);
1314 int mlx5_query_mad_ifc_max_pkeys(struct ib_device *ibdev,
1316 int mlx5_query_mad_ifc_vendor_id(struct ib_device *ibdev,
1318 int mlx5_query_mad_ifc_node_desc(struct mlx5_ib_dev *dev, char *node_desc);
1319 int mlx5_query_mad_ifc_node_guid(struct mlx5_ib_dev *dev, __be64 *node_guid);
1320 int mlx5_query_mad_ifc_pkey(struct ib_device *ibdev, u8 port, u16 index,
1322 int mlx5_query_mad_ifc_gids(struct ib_device *ibdev, u8 port, int index,
1324 int mlx5_query_mad_ifc_port(struct ib_device *ibdev, u8 port,
1325 struct ib_port_attr *props);
1326 int mlx5_ib_query_port(struct ib_device *ibdev, u8 port,
1327 struct ib_port_attr *props);
1328 void mlx5_ib_populate_pas(struct ib_umem *umem, size_t page_size, __be64 *pas,
1330 void mlx5_ib_copy_pas(u64 *old, u64 *new, int step, int num);
1331 int mlx5_ib_get_cqe_size(struct ib_cq *ibcq);
1332 int mlx5_mr_cache_init(struct mlx5_ib_dev *dev);
1333 int mlx5_mr_cache_cleanup(struct mlx5_ib_dev *dev);
1335 struct mlx5_ib_mr *mlx5_mr_cache_alloc(struct mlx5_ib_dev *dev,
1336 unsigned int entry, int access_flags);
1337 void mlx5_mr_cache_free(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr);
1338 int mlx5_mr_cache_invalidate(struct mlx5_ib_mr *mr);
1340 int mlx5_ib_check_mr_status(struct ib_mr *ibmr, u32 check_mask,
1341 struct ib_mr_status *mr_status);
1342 struct ib_wq *mlx5_ib_create_wq(struct ib_pd *pd,
1343 struct ib_wq_init_attr *init_attr,
1344 struct ib_udata *udata);
1345 int mlx5_ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
1346 int mlx5_ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *wq_attr,
1347 u32 wq_attr_mask, struct ib_udata *udata);
1348 int mlx5_ib_create_rwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_table,
1349 struct ib_rwq_ind_table_init_attr *init_attr,
1350 struct ib_udata *udata);
1351 int mlx5_ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
1352 struct ib_dm *mlx5_ib_alloc_dm(struct ib_device *ibdev,
1353 struct ib_ucontext *context,
1354 struct ib_dm_alloc_attr *attr,
1355 struct uverbs_attr_bundle *attrs);
1356 int mlx5_ib_dealloc_dm(struct ib_dm *ibdm, struct uverbs_attr_bundle *attrs);
1357 struct ib_mr *mlx5_ib_reg_dm_mr(struct ib_pd *pd, struct ib_dm *dm,
1358 struct ib_dm_mr_attr *attr,
1359 struct uverbs_attr_bundle *attrs);
1361 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1362 void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev);
1363 int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev);
1364 void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev);
1365 int __init mlx5_ib_odp_init(void);
1366 void mlx5_ib_odp_cleanup(void);
1367 void mlx5_odp_init_mr_cache_entry(struct mlx5_cache_ent *ent);
1368 void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries,
1369 struct mlx5_ib_mr *mr, int flags);
1371 int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd,
1372 enum ib_uverbs_advise_mr_advice advice,
1373 u32 flags, struct ib_sge *sg_list, u32 num_sge);
1374 int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr);
1375 int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr);
1376 #else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
1377 static inline void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev)
1382 static inline int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev) { return 0; }
1383 static inline void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev) {}
1384 static inline int mlx5_ib_odp_init(void) { return 0; }
1385 static inline void mlx5_ib_odp_cleanup(void) {}
1386 static inline void mlx5_odp_init_mr_cache_entry(struct mlx5_cache_ent *ent) {}
1387 static inline void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries,
1388 struct mlx5_ib_mr *mr, int flags) {}
1391 mlx5_ib_advise_mr_prefetch(struct ib_pd *pd,
1392 enum ib_uverbs_advise_mr_advice advice, u32 flags,
1393 struct ib_sge *sg_list, u32 num_sge)
1397 static inline int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr)
1401 static inline int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr)
1405 #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
1407 extern const struct mmu_interval_notifier_ops mlx5_mn_ops;
1409 /* Needed for rep profile */
1410 void __mlx5_ib_remove(struct mlx5_ib_dev *dev,
1411 const struct mlx5_ib_profile *profile,
1413 int __mlx5_ib_add(struct mlx5_ib_dev *dev,
1414 const struct mlx5_ib_profile *profile);
1416 int mlx5_ib_get_vf_config(struct ib_device *device, int vf,
1417 u8 port, struct ifla_vf_info *info);
1418 int mlx5_ib_set_vf_link_state(struct ib_device *device, int vf,
1419 u8 port, int state);
1420 int mlx5_ib_get_vf_stats(struct ib_device *device, int vf,
1421 u8 port, struct ifla_vf_stats *stats);
1422 int mlx5_ib_get_vf_guid(struct ib_device *device, int vf, u8 port,
1423 struct ifla_vf_guid *node_guid,
1424 struct ifla_vf_guid *port_guid);
1425 int mlx5_ib_set_vf_guid(struct ib_device *device, int vf, u8 port,
1426 u64 guid, int type);
1428 __be16 mlx5_get_roce_udp_sport_min(const struct mlx5_ib_dev *dev,
1429 const struct ib_gid_attr *attr);
1431 void mlx5_ib_cleanup_cong_debugfs(struct mlx5_ib_dev *dev, u8 port_num);
1432 void mlx5_ib_init_cong_debugfs(struct mlx5_ib_dev *dev, u8 port_num);
1434 /* GSI QP helper functions */
1435 int mlx5_ib_create_gsi(struct ib_pd *pd, struct mlx5_ib_qp *mqp,
1436 struct ib_qp_init_attr *attr);
1437 int mlx5_ib_destroy_gsi(struct mlx5_ib_qp *mqp);
1438 int mlx5_ib_gsi_modify_qp(struct ib_qp *qp, struct ib_qp_attr *attr,
1440 int mlx5_ib_gsi_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
1442 struct ib_qp_init_attr *qp_init_attr);
1443 int mlx5_ib_gsi_post_send(struct ib_qp *qp, const struct ib_send_wr *wr,
1444 const struct ib_send_wr **bad_wr);
1445 int mlx5_ib_gsi_post_recv(struct ib_qp *qp, const struct ib_recv_wr *wr,
1446 const struct ib_recv_wr **bad_wr);
1447 void mlx5_ib_gsi_pkey_change(struct mlx5_ib_gsi_qp *gsi);
1449 int mlx5_ib_generate_wc(struct ib_cq *ibcq, struct ib_wc *wc);
1451 void mlx5_ib_free_bfreg(struct mlx5_ib_dev *dev, struct mlx5_bfreg_info *bfregi,
1453 struct mlx5_ib_dev *mlx5_ib_get_ibdev_from_mpi(struct mlx5_ib_multiport_info *mpi);
1454 struct mlx5_core_dev *mlx5_ib_get_native_port_mdev(struct mlx5_ib_dev *dev,
1456 u8 *native_port_num);
1457 void mlx5_ib_put_native_port_mdev(struct mlx5_ib_dev *dev,
1460 extern const struct uapi_definition mlx5_ib_devx_defs[];
1461 extern const struct uapi_definition mlx5_ib_flow_defs[];
1462 extern const struct uapi_definition mlx5_ib_qos_defs[];
1463 extern const struct uapi_definition mlx5_ib_std_types_defs[];
1465 static inline void init_query_mad(struct ib_smp *mad)
1467 mad->base_version = 1;
1468 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
1469 mad->class_version = 1;
1470 mad->method = IB_MGMT_METHOD_GET;
1473 static inline int is_qp1(enum ib_qp_type qp_type)
1475 return qp_type == MLX5_IB_QPT_HW_GSI || qp_type == IB_QPT_GSI;
1478 #define MLX5_MAX_UMR_SHIFT 16
1479 #define MLX5_MAX_UMR_PAGES (1 << MLX5_MAX_UMR_SHIFT)
1481 static inline u32 check_cq_create_flags(u32 flags)
1484 * It returns non-zero value for unsupported CQ
1485 * create flags, otherwise it returns zero.
1487 return (flags & ~(IB_UVERBS_CQ_FLAGS_IGNORE_OVERRUN |
1488 IB_UVERBS_CQ_FLAGS_TIMESTAMP_COMPLETION));
1491 static inline int verify_assign_uidx(u8 cqe_version, u32 cmd_uidx,
1495 if ((cmd_uidx == MLX5_IB_DEFAULT_UIDX) ||
1496 (cmd_uidx & ~MLX5_USER_ASSIGNED_UIDX_MASK))
1498 *user_index = cmd_uidx;
1500 *user_index = MLX5_IB_DEFAULT_UIDX;
1506 static inline int get_qp_user_index(struct mlx5_ib_ucontext *ucontext,
1507 struct mlx5_ib_create_qp *ucmd,
1511 u8 cqe_version = ucontext->cqe_version;
1513 if ((offsetofend(typeof(*ucmd), uidx) <= inlen) && !cqe_version &&
1514 (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1517 if ((offsetofend(typeof(*ucmd), uidx) <= inlen) != !!cqe_version)
1520 return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1523 static inline int get_srq_user_index(struct mlx5_ib_ucontext *ucontext,
1524 struct mlx5_ib_create_srq *ucmd,
1528 u8 cqe_version = ucontext->cqe_version;
1530 if ((offsetofend(typeof(*ucmd), uidx) <= inlen) && !cqe_version &&
1531 (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1534 if ((offsetofend(typeof(*ucmd), uidx) <= inlen) != !!cqe_version)
1537 return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1540 static inline int get_uars_per_sys_page(struct mlx5_ib_dev *dev, bool lib_support)
1542 return lib_support && MLX5_CAP_GEN(dev->mdev, uar_4k) ?
1543 MLX5_UARS_IN_PAGE : 1;
1546 static inline int get_num_static_uars(struct mlx5_ib_dev *dev,
1547 struct mlx5_bfreg_info *bfregi)
1549 return get_uars_per_sys_page(dev, bfregi->lib_uar_4k) * bfregi->num_static_sys_pages;
1552 extern void *xlt_emergency_page;
1554 int bfregn_to_uar_index(struct mlx5_ib_dev *dev,
1555 struct mlx5_bfreg_info *bfregi, u32 bfregn,
1558 static inline bool mlx5_ib_can_load_pas_with_umr(struct mlx5_ib_dev *dev,
1562 * umr_check_mkey_mask() rejects MLX5_MKEY_MASK_PAGE_SIZE which is
1563 * always set if MLX5_IB_SEND_UMR_UPDATE_TRANSLATION (aka
1564 * MLX5_IB_UPD_XLT_ADDR and MLX5_IB_UPD_XLT_ENABLE) is set. Thus, a mkey
1565 * can never be enabled without this capability. Simplify this weird
1566 * quirky hardware by just saying it can't use PAS lists with UMR at
1569 if (MLX5_CAP_GEN(dev->mdev, umr_modify_entity_size_disabled))
1573 * length is the size of the MR in bytes when mlx5_ib_update_xlt() is
1576 if (!MLX5_CAP_GEN(dev->mdev, umr_extended_translation_offset) &&
1577 length >= MLX5_MAX_UMR_PAGES * PAGE_SIZE)
1583 * true if an existing MR can be reconfigured to new access_flags using UMR.
1584 * Older HW cannot use UMR to update certain elements of the MKC. See
1585 * umr_check_mkey_mask(), get_umr_update_access_mask() and umr_check_mkey_mask()
1587 static inline bool mlx5_ib_can_reconfig_with_umr(struct mlx5_ib_dev *dev,
1588 unsigned int current_access_flags,
1589 unsigned int target_access_flags)
1591 unsigned int diffs = current_access_flags ^ target_access_flags;
1593 if ((diffs & IB_ACCESS_REMOTE_ATOMIC) &&
1594 MLX5_CAP_GEN(dev->mdev, atomic) &&
1595 MLX5_CAP_GEN(dev->mdev, umr_modify_atomic_disabled))
1598 if ((diffs & IB_ACCESS_RELAXED_ORDERING) &&
1599 MLX5_CAP_GEN(dev->mdev, relaxed_ordering_write) &&
1600 !MLX5_CAP_GEN(dev->mdev, relaxed_ordering_write_umr))
1603 if ((diffs & IB_ACCESS_RELAXED_ORDERING) &&
1604 MLX5_CAP_GEN(dev->mdev, relaxed_ordering_read) &&
1605 !MLX5_CAP_GEN(dev->mdev, relaxed_ordering_read_umr))
1611 static inline int mlx5r_store_odp_mkey(struct mlx5_ib_dev *dev,
1612 struct mlx5_core_mkey *mmkey)
1614 refcount_set(&mmkey->usecount, 1);
1616 return xa_err(xa_store(&dev->odp_mkeys, mlx5_base_mkey(mmkey->key),
1617 mmkey, GFP_KERNEL));
1620 /* deref an mkey that can participate in ODP flow */
1621 static inline void mlx5r_deref_odp_mkey(struct mlx5_core_mkey *mmkey)
1623 if (refcount_dec_and_test(&mmkey->usecount))
1624 wake_up(&mmkey->wait);
1627 /* deref an mkey that can participate in ODP flow and wait for relese */
1628 static inline void mlx5r_deref_wait_odp_mkey(struct mlx5_core_mkey *mmkey)
1630 mlx5r_deref_odp_mkey(mmkey);
1631 wait_event(mmkey->wait, refcount_read(&mmkey->usecount) == 0);
1634 int mlx5_ib_test_wc(struct mlx5_ib_dev *dev);
1636 static inline bool mlx5_ib_lag_should_assign_affinity(struct mlx5_ib_dev *dev)
1638 return dev->lag_active ||
1639 (MLX5_CAP_GEN(dev->mdev, num_lag_ports) > 1 &&
1640 MLX5_CAP_GEN(dev->mdev, lag_tx_port_affinity));
1642 #endif /* MLX5_IB_H */