2 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
3 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
4 * Copyright (c) 2004 Intel Corporation. All rights reserved.
5 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
6 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
10 * This software is available to you under a choice of one of two
11 * licenses. You may choose to be licensed under the terms of the GNU
12 * General Public License (GPL) Version 2, available from the file
13 * COPYING in the main directory of this source tree, or the
14 * OpenIB.org BSD license below:
16 * Redistribution and use in source and binary forms, with or
17 * without modification, are permitted provided that the following
20 * - Redistributions of source code must retain the above
21 * copyright notice, this list of conditions and the following
24 * - Redistributions in binary form must reproduce the above
25 * copyright notice, this list of conditions and the following
26 * disclaimer in the documentation and/or other materials
27 * provided with the distribution.
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
39 #if !defined(IB_VERBS_H)
42 #include <linux/types.h>
43 #include <linux/device.h>
44 #include <linux/dma-mapping.h>
45 #include <linux/kref.h>
46 #include <linux/list.h>
47 #include <linux/rwsem.h>
48 #include <linux/workqueue.h>
49 #include <linux/irq_poll.h>
50 #include <uapi/linux/if_ether.h>
53 #include <linux/string.h>
54 #include <linux/slab.h>
55 #include <linux/netdevice.h>
56 #include <linux/refcount.h>
57 #include <linux/if_link.h>
58 #include <linux/atomic.h>
59 #include <linux/mmu_notifier.h>
60 #include <linux/uaccess.h>
61 #include <linux/cgroup_rdma.h>
62 #include <linux/irqflags.h>
63 #include <linux/preempt.h>
64 #include <linux/dim.h>
65 #include <uapi/rdma/ib_user_verbs.h>
66 #include <rdma/rdma_counter.h>
67 #include <rdma/restrack.h>
68 #include <rdma/signature.h>
69 #include <uapi/rdma/rdma_user_ioctl.h>
70 #include <uapi/rdma/ib_user_ioctl_verbs.h>
72 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
76 struct ib_usrq_object;
79 extern struct workqueue_struct *ib_wq;
80 extern struct workqueue_struct *ib_comp_wq;
81 extern struct workqueue_struct *ib_comp_unbound_wq;
86 void ibdev_printk(const char *level, const struct ib_device *ibdev,
87 const char *format, ...);
89 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
91 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
93 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
95 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
97 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
99 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
100 __printf(2, 3) __cold
101 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
103 #if defined(CONFIG_DYNAMIC_DEBUG)
104 #define ibdev_dbg(__dev, format, args...) \
105 dynamic_ibdev_dbg(__dev, format, ##args)
107 __printf(2, 3) __cold
109 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
112 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \
114 static DEFINE_RATELIMIT_STATE(_rs, \
115 DEFAULT_RATELIMIT_INTERVAL, \
116 DEFAULT_RATELIMIT_BURST); \
117 if (__ratelimit(&_rs)) \
118 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \
121 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
122 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
123 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
124 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
125 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
126 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
127 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
128 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
129 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
130 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
131 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
132 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
133 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
134 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
136 #if defined(CONFIG_DYNAMIC_DEBUG)
137 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
138 #define ibdev_dbg_ratelimited(ibdev, fmt, ...) \
140 static DEFINE_RATELIMIT_STATE(_rs, \
141 DEFAULT_RATELIMIT_INTERVAL, \
142 DEFAULT_RATELIMIT_BURST); \
143 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \
144 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \
145 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \
149 __printf(2, 3) __cold
151 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
157 __be64 subnet_prefix;
162 extern union ib_gid zgid;
165 /* If link layer is Ethernet, this is RoCE V1 */
167 IB_GID_TYPE_ROCE = 0,
168 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
172 #define ROCE_V2_UDP_DPORT 4791
174 struct net_device __rcu *ndev;
175 struct ib_device *device;
177 enum ib_gid_type gid_type;
183 /* set the local administered indication */
184 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
187 enum rdma_transport_type {
189 RDMA_TRANSPORT_IWARP,
190 RDMA_TRANSPORT_USNIC,
191 RDMA_TRANSPORT_USNIC_UDP,
192 RDMA_TRANSPORT_UNSPECIFIED,
195 enum rdma_protocol_type {
199 RDMA_PROTOCOL_USNIC_UDP
202 __attribute_const__ enum rdma_transport_type
203 rdma_node_get_transport(unsigned int node_type);
205 enum rdma_network_type {
207 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
212 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
214 if (network_type == RDMA_NETWORK_IPV4 ||
215 network_type == RDMA_NETWORK_IPV6)
216 return IB_GID_TYPE_ROCE_UDP_ENCAP;
218 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
219 return IB_GID_TYPE_IB;
222 static inline enum rdma_network_type
223 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
225 if (attr->gid_type == IB_GID_TYPE_IB)
226 return RDMA_NETWORK_IB;
228 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
229 return RDMA_NETWORK_IPV4;
231 return RDMA_NETWORK_IPV6;
234 enum rdma_link_layer {
235 IB_LINK_LAYER_UNSPECIFIED,
236 IB_LINK_LAYER_INFINIBAND,
237 IB_LINK_LAYER_ETHERNET,
240 enum ib_device_cap_flags {
241 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
242 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
243 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
244 IB_DEVICE_RAW_MULTI = (1 << 3),
245 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
246 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
247 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
248 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
249 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
250 /* Not in use, former INIT_TYPE = (1 << 9),*/
251 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
252 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
253 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
254 IB_DEVICE_SRQ_RESIZE = (1 << 13),
255 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
258 * This device supports a per-device lkey or stag that can be
259 * used without performing a memory registration for the local
260 * memory. Note that ULPs should never check this flag, but
261 * instead of use the local_dma_lkey flag in the ib_pd structure,
262 * which will always contain a usable lkey.
264 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
265 /* Reserved, old SEND_W_INV = (1 << 16),*/
266 IB_DEVICE_MEM_WINDOW = (1 << 17),
268 * Devices should set IB_DEVICE_UD_IP_SUM if they support
269 * insertion of UDP and TCP checksum on outgoing UD IPoIB
270 * messages and can verify the validity of checksum for
271 * incoming messages. Setting this flag implies that the
272 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
274 IB_DEVICE_UD_IP_CSUM = (1 << 18),
275 IB_DEVICE_UD_TSO = (1 << 19),
276 IB_DEVICE_XRC = (1 << 20),
279 * This device supports the IB "base memory management extension",
280 * which includes support for fast registrations (IB_WR_REG_MR,
281 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
282 * also be set by any iWarp device which must support FRs to comply
283 * to the iWarp verbs spec. iWarp devices also support the
284 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
287 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
288 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
289 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
290 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
291 IB_DEVICE_RC_IP_CSUM = (1 << 25),
292 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
293 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
295 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
296 * support execution of WQEs that involve synchronization
297 * of I/O operations with single completion queue managed
300 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
301 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
302 IB_DEVICE_INTEGRITY_HANDOVER = (1 << 30),
303 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
304 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
305 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
306 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
307 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
308 IB_DEVICE_RDMA_NETDEV_OPA = (1ULL << 35),
309 /* The device supports padding incoming writes to cacheline. */
310 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
311 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37),
320 enum ib_odp_general_cap_bits {
321 IB_ODP_SUPPORT = 1 << 0,
322 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
325 enum ib_odp_transport_cap_bits {
326 IB_ODP_SUPPORT_SEND = 1 << 0,
327 IB_ODP_SUPPORT_RECV = 1 << 1,
328 IB_ODP_SUPPORT_WRITE = 1 << 2,
329 IB_ODP_SUPPORT_READ = 1 << 3,
330 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
331 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
335 uint64_t general_caps;
337 uint32_t rc_odp_caps;
338 uint32_t uc_odp_caps;
339 uint32_t ud_odp_caps;
340 uint32_t xrc_odp_caps;
341 } per_transport_caps;
345 /* Corresponding bit will be set if qp type from
346 * 'enum ib_qp_type' is supported, e.g.
347 * supported_qpts |= 1 << IB_QPT_UD
350 u32 max_rwq_indirection_tables;
351 u32 max_rwq_indirection_table_size;
354 enum ib_tm_cap_flags {
355 /* Support tag matching with rendezvous offload for RC transport */
356 IB_TM_CAP_RNDV_RC = 1 << 0,
360 /* Max size of RNDV header */
361 u32 max_rndv_hdr_size;
362 /* Max number of entries in tag matching list */
364 /* From enum ib_tm_cap_flags */
366 /* Max number of outstanding list operations */
368 /* Max number of SGE in tag matching entry */
372 struct ib_cq_init_attr {
378 enum ib_cq_attr_mask {
379 IB_CQ_MODERATE = 1 << 0,
383 u16 max_cq_moderation_count;
384 u16 max_cq_moderation_period;
387 struct ib_dm_mr_attr {
393 struct ib_dm_alloc_attr {
399 struct ib_device_attr {
401 __be64 sys_image_guid;
409 u64 device_cap_flags;
420 int max_qp_init_rd_atom;
421 int max_ee_init_rd_atom;
422 enum ib_atomic_cap atomic_cap;
423 enum ib_atomic_cap masked_atomic_cap;
430 int max_mcast_qp_attach;
431 int max_total_mcast_qp_attach;
438 unsigned int max_fast_reg_page_list_len;
439 unsigned int max_pi_fast_reg_page_list_len;
441 u8 local_ca_ack_delay;
444 struct ib_odp_caps odp_caps;
445 uint64_t timestamp_mask;
446 uint64_t hca_core_clock; /* in KHZ */
447 struct ib_rss_caps rss_caps;
449 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
450 struct ib_tm_caps tm_caps;
451 struct ib_cq_caps cq_caps;
453 /* Max entries for sgl for optimized performance per READ */
470 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
473 case IB_MTU_256: return 256;
474 case IB_MTU_512: return 512;
475 case IB_MTU_1024: return 1024;
476 case IB_MTU_2048: return 2048;
477 case IB_MTU_4096: return 4096;
482 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
486 else if (mtu >= 2048)
488 else if (mtu >= 1024)
496 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
504 return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
508 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
511 return OPA_MTU_10240;
512 else if (mtu >= 8192)
515 return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
524 IB_PORT_ACTIVE_DEFER = 5
527 enum ib_port_phys_state {
528 IB_PORT_PHYS_STATE_SLEEP = 1,
529 IB_PORT_PHYS_STATE_POLLING = 2,
530 IB_PORT_PHYS_STATE_DISABLED = 3,
531 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
532 IB_PORT_PHYS_STATE_LINK_UP = 5,
533 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
534 IB_PORT_PHYS_STATE_PHY_TEST = 7,
545 static inline int ib_width_enum_to_int(enum ib_port_width width)
548 case IB_WIDTH_1X: return 1;
549 case IB_WIDTH_2X: return 2;
550 case IB_WIDTH_4X: return 4;
551 case IB_WIDTH_8X: return 8;
552 case IB_WIDTH_12X: return 12;
568 * struct rdma_hw_stats
569 * @lock - Mutex to protect parallel write access to lifespan and values
570 * of counters, which are 64bits and not guaranteeed to be written
571 * atomicaly on 32bits systems.
572 * @timestamp - Used by the core code to track when the last update was
573 * @lifespan - Used by the core code to determine how old the counters
574 * should be before being updated again. Stored in jiffies, defaults
575 * to 10 milliseconds, drivers can override the default be specifying
576 * their own value during their allocation routine.
577 * @name - Array of pointers to static names used for the counters in
579 * @num_counters - How many hardware counters there are. If name is
580 * shorter than this number, a kernel oops will result. Driver authors
581 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
582 * in their code to prevent this.
583 * @value - Array of u64 counters that are accessed by the sysfs code and
584 * filled in by the drivers get_stats routine
586 struct rdma_hw_stats {
587 struct mutex lock; /* Protect lifespan and values[] */
588 unsigned long timestamp;
589 unsigned long lifespan;
590 const char * const *names;
595 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
597 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
599 * @names - Array of static const char *
600 * @num_counters - How many elements in array
601 * @lifespan - How many milliseconds between updates
603 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
604 const char * const *names, int num_counters,
605 unsigned long lifespan)
607 struct rdma_hw_stats *stats;
609 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
613 stats->names = names;
614 stats->num_counters = num_counters;
615 stats->lifespan = msecs_to_jiffies(lifespan);
621 /* Define bits for the various functionality this port needs to be supported by
624 /* Management 0x00000FFF */
625 #define RDMA_CORE_CAP_IB_MAD 0x00000001
626 #define RDMA_CORE_CAP_IB_SMI 0x00000002
627 #define RDMA_CORE_CAP_IB_CM 0x00000004
628 #define RDMA_CORE_CAP_IW_CM 0x00000008
629 #define RDMA_CORE_CAP_IB_SA 0x00000010
630 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
632 /* Address format 0x000FF000 */
633 #define RDMA_CORE_CAP_AF_IB 0x00001000
634 #define RDMA_CORE_CAP_ETH_AH 0x00002000
635 #define RDMA_CORE_CAP_OPA_AH 0x00004000
636 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
638 /* Protocol 0xFFF00000 */
639 #define RDMA_CORE_CAP_PROT_IB 0x00100000
640 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
641 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
642 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
643 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
644 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
646 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
647 | RDMA_CORE_CAP_PROT_ROCE \
648 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
650 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
651 | RDMA_CORE_CAP_IB_MAD \
652 | RDMA_CORE_CAP_IB_SMI \
653 | RDMA_CORE_CAP_IB_CM \
654 | RDMA_CORE_CAP_IB_SA \
655 | RDMA_CORE_CAP_AF_IB)
656 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
657 | RDMA_CORE_CAP_IB_MAD \
658 | RDMA_CORE_CAP_IB_CM \
659 | RDMA_CORE_CAP_AF_IB \
660 | RDMA_CORE_CAP_ETH_AH)
661 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
662 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
663 | RDMA_CORE_CAP_IB_MAD \
664 | RDMA_CORE_CAP_IB_CM \
665 | RDMA_CORE_CAP_AF_IB \
666 | RDMA_CORE_CAP_ETH_AH)
667 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
668 | RDMA_CORE_CAP_IW_CM)
669 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
670 | RDMA_CORE_CAP_OPA_MAD)
672 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
674 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
676 struct ib_port_attr {
678 enum ib_port_state state;
680 enum ib_mtu active_mtu;
683 unsigned int ip_gids:1;
684 /* This is the value from PortInfo CapabilityMask, defined by IBA */
703 enum ib_device_modify_flags {
704 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
705 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
708 #define IB_DEVICE_NODE_DESC_MAX 64
710 struct ib_device_modify {
712 char node_desc[IB_DEVICE_NODE_DESC_MAX];
715 enum ib_port_modify_flags {
716 IB_PORT_SHUTDOWN = 1,
717 IB_PORT_INIT_TYPE = (1<<2),
718 IB_PORT_RESET_QKEY_CNTR = (1<<3),
719 IB_PORT_OPA_MASK_CHG = (1<<4)
722 struct ib_port_modify {
723 u32 set_port_cap_mask;
724 u32 clr_port_cap_mask;
732 IB_EVENT_QP_ACCESS_ERR,
736 IB_EVENT_PATH_MIG_ERR,
737 IB_EVENT_DEVICE_FATAL,
738 IB_EVENT_PORT_ACTIVE,
741 IB_EVENT_PKEY_CHANGE,
744 IB_EVENT_SRQ_LIMIT_REACHED,
745 IB_EVENT_QP_LAST_WQE_REACHED,
746 IB_EVENT_CLIENT_REREGISTER,
751 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
754 struct ib_device *device;
762 enum ib_event_type event;
765 struct ib_event_handler {
766 struct ib_device *device;
767 void (*handler)(struct ib_event_handler *, struct ib_event *);
768 struct list_head list;
771 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
773 (_ptr)->device = _device; \
774 (_ptr)->handler = _handler; \
775 INIT_LIST_HEAD(&(_ptr)->list); \
778 struct ib_global_route {
779 const struct ib_gid_attr *sgid_attr;
788 __be32 version_tclass_flow;
796 union rdma_network_hdr {
799 /* The IB spec states that if it's IPv4, the header
800 * is located in the last 20 bytes of the header.
803 struct iphdr roce4grh;
807 #define IB_QPN_MASK 0xFFFFFF
810 IB_MULTICAST_QPN = 0xffffff
813 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
814 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
821 IB_RATE_PORT_CURRENT = 0,
822 IB_RATE_2_5_GBPS = 2,
830 IB_RATE_120_GBPS = 10,
831 IB_RATE_14_GBPS = 11,
832 IB_RATE_56_GBPS = 12,
833 IB_RATE_112_GBPS = 13,
834 IB_RATE_168_GBPS = 14,
835 IB_RATE_25_GBPS = 15,
836 IB_RATE_100_GBPS = 16,
837 IB_RATE_200_GBPS = 17,
838 IB_RATE_300_GBPS = 18,
839 IB_RATE_28_GBPS = 19,
840 IB_RATE_50_GBPS = 20,
841 IB_RATE_400_GBPS = 21,
842 IB_RATE_600_GBPS = 22,
846 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
847 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
848 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
849 * @rate: rate to convert.
851 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
854 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
855 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
856 * @rate: rate to convert.
858 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
862 * enum ib_mr_type - memory region type
863 * @IB_MR_TYPE_MEM_REG: memory region that is used for
864 * normal registration
865 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
866 * register any arbitrary sg lists (without
867 * the normal mr constraints - see
869 * @IB_MR_TYPE_DM: memory region that is used for device
870 * memory registration
871 * @IB_MR_TYPE_USER: memory region that is used for the user-space
873 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
874 * without address translations (VA=PA)
875 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
876 * data integrity operations
884 IB_MR_TYPE_INTEGRITY,
887 enum ib_mr_status_check {
888 IB_MR_CHECK_SIG_STATUS = 1,
892 * struct ib_mr_status - Memory region status container
894 * @fail_status: Bitmask of MR checks status. For each
895 * failed check a corresponding status bit is set.
896 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
899 struct ib_mr_status {
901 struct ib_sig_err sig_err;
905 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
907 * @mult: multiple to convert.
909 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
911 struct rdma_ah_init_attr {
912 struct rdma_ah_attr *ah_attr;
914 struct net_device *xmit_slave;
917 enum rdma_ah_attr_type {
918 RDMA_AH_ATTR_TYPE_UNDEFINED,
919 RDMA_AH_ATTR_TYPE_IB,
920 RDMA_AH_ATTR_TYPE_ROCE,
921 RDMA_AH_ATTR_TYPE_OPA,
929 struct roce_ah_attr {
939 struct rdma_ah_attr {
940 struct ib_global_route grh;
945 enum rdma_ah_attr_type type;
947 struct ib_ah_attr ib;
948 struct roce_ah_attr roce;
949 struct opa_ah_attr opa;
957 IB_WC_LOC_EEC_OP_ERR,
962 IB_WC_LOC_ACCESS_ERR,
963 IB_WC_REM_INV_REQ_ERR,
964 IB_WC_REM_ACCESS_ERR,
967 IB_WC_RNR_RETRY_EXC_ERR,
968 IB_WC_LOC_RDD_VIOL_ERR,
969 IB_WC_REM_INV_RD_REQ_ERR,
972 IB_WC_INV_EEC_STATE_ERR,
974 IB_WC_RESP_TIMEOUT_ERR,
978 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
989 IB_WC_MASKED_COMP_SWAP,
990 IB_WC_MASKED_FETCH_ADD,
992 * Set value of IB_WC_RECV so consumers can test if a completion is a
993 * receive by testing (opcode & IB_WC_RECV).
996 IB_WC_RECV_RDMA_WITH_IMM
1001 IB_WC_WITH_IMM = (1<<1),
1002 IB_WC_WITH_INVALIDATE = (1<<2),
1003 IB_WC_IP_CSUM_OK = (1<<3),
1004 IB_WC_WITH_SMAC = (1<<4),
1005 IB_WC_WITH_VLAN = (1<<5),
1006 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
1012 struct ib_cqe *wr_cqe;
1014 enum ib_wc_status status;
1015 enum ib_wc_opcode opcode;
1021 u32 invalidate_rkey;
1029 u8 port_num; /* valid only for DR SMPs on switches */
1032 u8 network_hdr_type;
1035 enum ib_cq_notify_flags {
1036 IB_CQ_SOLICITED = 1 << 0,
1037 IB_CQ_NEXT_COMP = 1 << 1,
1038 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1039 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1043 IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1044 IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1045 IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1048 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1050 return srq_type == IB_SRQT_XRC ||
1051 srq_type == IB_SRQT_TM;
1054 enum ib_srq_attr_mask {
1055 IB_SRQ_MAX_WR = 1 << 0,
1056 IB_SRQ_LIMIT = 1 << 1,
1059 struct ib_srq_attr {
1065 struct ib_srq_init_attr {
1066 void (*event_handler)(struct ib_event *, void *);
1068 struct ib_srq_attr attr;
1069 enum ib_srq_type srq_type;
1075 struct ib_xrcd *xrcd;
1090 u32 max_inline_data;
1093 * Maximum number of rdma_rw_ctx structures in flight at a time.
1094 * ib_create_qp() will calculate the right amount of neededed WRs
1095 * and MRs based on this.
1107 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1108 * here (and in that order) since the MAD layer uses them as
1109 * indices into a 2-entry table.
1114 IB_QPT_RC = IB_UVERBS_QPT_RC,
1115 IB_QPT_UC = IB_UVERBS_QPT_UC,
1116 IB_QPT_UD = IB_UVERBS_QPT_UD,
1118 IB_QPT_RAW_ETHERTYPE,
1119 IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1120 IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1121 IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1123 IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1124 /* Reserve a range for qp types internal to the low level driver.
1125 * These qp types will not be visible at the IB core layer, so the
1126 * IB_QPT_MAX usages should not be affected in the core layer
1128 IB_QPT_RESERVED1 = 0x1000,
1140 enum ib_qp_create_flags {
1141 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1142 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK =
1143 IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1144 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1145 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1146 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1147 IB_QP_CREATE_NETIF_QP = 1 << 5,
1148 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
1149 IB_QP_CREATE_NETDEV_USE = 1 << 7,
1150 IB_QP_CREATE_SCATTER_FCS =
1151 IB_UVERBS_QP_CREATE_SCATTER_FCS,
1152 IB_QP_CREATE_CVLAN_STRIPPING =
1153 IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1154 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1155 IB_QP_CREATE_PCI_WRITE_END_PADDING =
1156 IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1157 /* reserve bits 26-31 for low level drivers' internal use */
1158 IB_QP_CREATE_RESERVED_START = 1 << 26,
1159 IB_QP_CREATE_RESERVED_END = 1 << 31,
1163 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1164 * callback to destroy the passed in QP.
1167 struct ib_qp_init_attr {
1168 /* Consumer's event_handler callback must not block */
1169 void (*event_handler)(struct ib_event *, void *);
1172 struct ib_cq *send_cq;
1173 struct ib_cq *recv_cq;
1175 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1176 struct ib_qp_cap cap;
1177 enum ib_sig_type sq_sig_type;
1178 enum ib_qp_type qp_type;
1182 * Only needed for special QP types, or when using the RW API.
1185 struct ib_rwq_ind_table *rwq_ind_tbl;
1189 struct ib_qp_open_attr {
1190 void (*event_handler)(struct ib_event *, void *);
1193 enum ib_qp_type qp_type;
1196 enum ib_rnr_timeout {
1197 IB_RNR_TIMER_655_36 = 0,
1198 IB_RNR_TIMER_000_01 = 1,
1199 IB_RNR_TIMER_000_02 = 2,
1200 IB_RNR_TIMER_000_03 = 3,
1201 IB_RNR_TIMER_000_04 = 4,
1202 IB_RNR_TIMER_000_06 = 5,
1203 IB_RNR_TIMER_000_08 = 6,
1204 IB_RNR_TIMER_000_12 = 7,
1205 IB_RNR_TIMER_000_16 = 8,
1206 IB_RNR_TIMER_000_24 = 9,
1207 IB_RNR_TIMER_000_32 = 10,
1208 IB_RNR_TIMER_000_48 = 11,
1209 IB_RNR_TIMER_000_64 = 12,
1210 IB_RNR_TIMER_000_96 = 13,
1211 IB_RNR_TIMER_001_28 = 14,
1212 IB_RNR_TIMER_001_92 = 15,
1213 IB_RNR_TIMER_002_56 = 16,
1214 IB_RNR_TIMER_003_84 = 17,
1215 IB_RNR_TIMER_005_12 = 18,
1216 IB_RNR_TIMER_007_68 = 19,
1217 IB_RNR_TIMER_010_24 = 20,
1218 IB_RNR_TIMER_015_36 = 21,
1219 IB_RNR_TIMER_020_48 = 22,
1220 IB_RNR_TIMER_030_72 = 23,
1221 IB_RNR_TIMER_040_96 = 24,
1222 IB_RNR_TIMER_061_44 = 25,
1223 IB_RNR_TIMER_081_92 = 26,
1224 IB_RNR_TIMER_122_88 = 27,
1225 IB_RNR_TIMER_163_84 = 28,
1226 IB_RNR_TIMER_245_76 = 29,
1227 IB_RNR_TIMER_327_68 = 30,
1228 IB_RNR_TIMER_491_52 = 31
1231 enum ib_qp_attr_mask {
1233 IB_QP_CUR_STATE = (1<<1),
1234 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1235 IB_QP_ACCESS_FLAGS = (1<<3),
1236 IB_QP_PKEY_INDEX = (1<<4),
1237 IB_QP_PORT = (1<<5),
1238 IB_QP_QKEY = (1<<6),
1240 IB_QP_PATH_MTU = (1<<8),
1241 IB_QP_TIMEOUT = (1<<9),
1242 IB_QP_RETRY_CNT = (1<<10),
1243 IB_QP_RNR_RETRY = (1<<11),
1244 IB_QP_RQ_PSN = (1<<12),
1245 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1246 IB_QP_ALT_PATH = (1<<14),
1247 IB_QP_MIN_RNR_TIMER = (1<<15),
1248 IB_QP_SQ_PSN = (1<<16),
1249 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1250 IB_QP_PATH_MIG_STATE = (1<<18),
1251 IB_QP_CAP = (1<<19),
1252 IB_QP_DEST_QPN = (1<<20),
1253 IB_QP_RESERVED1 = (1<<21),
1254 IB_QP_RESERVED2 = (1<<22),
1255 IB_QP_RESERVED3 = (1<<23),
1256 IB_QP_RESERVED4 = (1<<24),
1257 IB_QP_RATE_LIMIT = (1<<25),
1282 enum ib_qp_state qp_state;
1283 enum ib_qp_state cur_qp_state;
1284 enum ib_mtu path_mtu;
1285 enum ib_mig_state path_mig_state;
1290 int qp_access_flags;
1291 struct ib_qp_cap cap;
1292 struct rdma_ah_attr ah_attr;
1293 struct rdma_ah_attr alt_ah_attr;
1296 u8 en_sqd_async_notify;
1299 u8 max_dest_rd_atomic;
1308 struct net_device *xmit_slave;
1312 /* These are shared with userspace */
1313 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1314 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1315 IB_WR_SEND = IB_UVERBS_WR_SEND,
1316 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1317 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1318 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1319 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1320 IB_WR_LSO = IB_UVERBS_WR_TSO,
1321 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1322 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1323 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1324 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1325 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1326 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1327 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1329 /* These are kernel only and can not be issued by userspace */
1330 IB_WR_REG_MR = 0x20,
1331 IB_WR_REG_MR_INTEGRITY,
1333 /* reserve values for low level drivers' internal use.
1334 * These values will not be used at all in the ib core layer.
1336 IB_WR_RESERVED1 = 0xf0,
1348 enum ib_send_flags {
1350 IB_SEND_SIGNALED = (1<<1),
1351 IB_SEND_SOLICITED = (1<<2),
1352 IB_SEND_INLINE = (1<<3),
1353 IB_SEND_IP_CSUM = (1<<4),
1355 /* reserve bits 26-31 for low level drivers' internal use */
1356 IB_SEND_RESERVED_START = (1 << 26),
1357 IB_SEND_RESERVED_END = (1 << 31),
1367 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1371 struct ib_send_wr *next;
1374 struct ib_cqe *wr_cqe;
1376 struct ib_sge *sg_list;
1378 enum ib_wr_opcode opcode;
1382 u32 invalidate_rkey;
1387 struct ib_send_wr wr;
1392 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1394 return container_of(wr, struct ib_rdma_wr, wr);
1397 struct ib_atomic_wr {
1398 struct ib_send_wr wr;
1402 u64 compare_add_mask;
1407 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1409 return container_of(wr, struct ib_atomic_wr, wr);
1413 struct ib_send_wr wr;
1420 u16 pkey_index; /* valid for GSI only */
1421 u8 port_num; /* valid for DR SMPs on switch only */
1424 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1426 return container_of(wr, struct ib_ud_wr, wr);
1430 struct ib_send_wr wr;
1436 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1438 return container_of(wr, struct ib_reg_wr, wr);
1442 struct ib_recv_wr *next;
1445 struct ib_cqe *wr_cqe;
1447 struct ib_sge *sg_list;
1451 enum ib_access_flags {
1452 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1453 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1454 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1455 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1456 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1457 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1458 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1459 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1460 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1462 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1463 IB_ACCESS_SUPPORTED =
1464 ((IB_ACCESS_HUGETLB << 1) - 1) | IB_ACCESS_OPTIONAL,
1468 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1469 * are hidden here instead of a uapi header!
1471 enum ib_mr_rereg_flags {
1472 IB_MR_REREG_TRANS = 1,
1473 IB_MR_REREG_PD = (1<<1),
1474 IB_MR_REREG_ACCESS = (1<<2),
1475 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1478 struct ib_fmr_attr {
1486 enum rdma_remove_reason {
1488 * Userspace requested uobject deletion or initial try
1489 * to remove uobject via cleanup. Call could fail
1491 RDMA_REMOVE_DESTROY,
1492 /* Context deletion. This call should delete the actual object itself */
1494 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1495 RDMA_REMOVE_DRIVER_REMOVE,
1496 /* uobj is being cleaned-up before being committed */
1499 * uobj has been fully created, with the uobj->object set, but is being
1500 * cleaned up before being comitted
1502 RDMA_REMOVE_ABORT_HWOBJ,
1505 struct ib_rdmacg_object {
1506 #ifdef CONFIG_CGROUP_RDMA
1507 struct rdma_cgroup *cg; /* owner rdma cgroup */
1511 struct ib_ucontext {
1512 struct ib_device *device;
1513 struct ib_uverbs_file *ufile;
1515 * 'closing' can be read by the driver only during a destroy callback,
1516 * it is set when we are closing the file descriptor and indicates
1517 * that mm_sem may be locked.
1521 bool cleanup_retryable;
1523 struct ib_rdmacg_object cg_obj;
1525 * Implementation details of the RDMA core, don't use in drivers:
1527 struct rdma_restrack_entry res;
1528 struct xarray mmap_xa;
1532 u64 user_handle; /* handle given to us by userspace */
1533 /* ufile & ucontext owning this object */
1534 struct ib_uverbs_file *ufile;
1535 /* FIXME, save memory: ufile->context == context */
1536 struct ib_ucontext *context; /* associated user context */
1537 void *object; /* containing object */
1538 struct list_head list; /* link to context's list */
1539 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1540 int id; /* index into kernel idr */
1542 atomic_t usecnt; /* protects exclusive access */
1543 struct rcu_head rcu; /* kfree_rcu() overhead */
1545 const struct uverbs_api_object *uapi_object;
1549 const void __user *inbuf;
1550 void __user *outbuf;
1558 struct ib_device *device;
1559 struct ib_uobject *uobject;
1560 atomic_t usecnt; /* count all resources */
1562 u32 unsafe_global_rkey;
1565 * Implementation details of the RDMA core, don't use in drivers:
1567 struct ib_mr *__internal_mr;
1568 struct rdma_restrack_entry res;
1572 struct ib_device *device;
1573 atomic_t usecnt; /* count all exposed resources */
1574 struct inode *inode;
1576 struct mutex tgt_qp_mutex;
1577 struct list_head tgt_qp_list;
1581 struct ib_device *device;
1583 struct ib_uobject *uobject;
1584 const struct ib_gid_attr *sgid_attr;
1585 enum rdma_ah_attr_type type;
1588 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1590 enum ib_poll_context {
1591 IB_POLL_DIRECT, /* caller context, no hw completions */
1592 IB_POLL_SOFTIRQ, /* poll from softirq context */
1593 IB_POLL_WORKQUEUE, /* poll from workqueue */
1594 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1598 struct ib_device *device;
1599 struct ib_ucq_object *uobject;
1600 ib_comp_handler comp_handler;
1601 void (*event_handler)(struct ib_event *, void *);
1604 atomic_t usecnt; /* count number of work queues */
1605 enum ib_poll_context poll_ctx;
1608 struct irq_poll iop;
1609 struct work_struct work;
1611 struct workqueue_struct *comp_wq;
1614 /* updated only by trace points */
1619 * Implementation details of the RDMA core, don't use in drivers:
1621 struct rdma_restrack_entry res;
1625 struct ib_device *device;
1627 struct ib_usrq_object *uobject;
1628 void (*event_handler)(struct ib_event *, void *);
1630 enum ib_srq_type srq_type;
1637 struct ib_xrcd *xrcd;
1644 enum ib_raw_packet_caps {
1645 /* Strip cvlan from incoming packet and report it in the matching work
1646 * completion is supported.
1648 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1649 /* Scatter FCS field of an incoming packet to host memory is supported.
1651 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1652 /* Checksum offloads are supported (for both send and receive). */
1653 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1654 /* When a packet is received for an RQ with no receive WQEs, the
1655 * packet processing is delayed.
1657 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1661 IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1671 struct ib_device *device;
1672 struct ib_uwq_object *uobject;
1674 void (*event_handler)(struct ib_event *, void *);
1678 enum ib_wq_state state;
1679 enum ib_wq_type wq_type;
1684 IB_WQ_FLAGS_CVLAN_STRIPPING = IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1685 IB_WQ_FLAGS_SCATTER_FCS = IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1686 IB_WQ_FLAGS_DELAY_DROP = IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1687 IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1688 IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1691 struct ib_wq_init_attr {
1693 enum ib_wq_type wq_type;
1697 void (*event_handler)(struct ib_event *, void *);
1698 u32 create_flags; /* Use enum ib_wq_flags */
1701 enum ib_wq_attr_mask {
1702 IB_WQ_STATE = 1 << 0,
1703 IB_WQ_CUR_STATE = 1 << 1,
1704 IB_WQ_FLAGS = 1 << 2,
1708 enum ib_wq_state wq_state;
1709 enum ib_wq_state curr_wq_state;
1710 u32 flags; /* Use enum ib_wq_flags */
1711 u32 flags_mask; /* Use enum ib_wq_flags */
1714 struct ib_rwq_ind_table {
1715 struct ib_device *device;
1716 struct ib_uobject *uobject;
1719 u32 log_ind_tbl_size;
1720 struct ib_wq **ind_tbl;
1723 struct ib_rwq_ind_table_init_attr {
1724 u32 log_ind_tbl_size;
1725 /* Each entry is a pointer to Receive Work Queue */
1726 struct ib_wq **ind_tbl;
1729 enum port_pkey_state {
1730 IB_PORT_PKEY_NOT_VALID = 0,
1731 IB_PORT_PKEY_VALID = 1,
1732 IB_PORT_PKEY_LISTED = 2,
1735 struct ib_qp_security;
1737 struct ib_port_pkey {
1738 enum port_pkey_state state;
1741 struct list_head qp_list;
1742 struct list_head to_error_list;
1743 struct ib_qp_security *sec;
1746 struct ib_ports_pkeys {
1747 struct ib_port_pkey main;
1748 struct ib_port_pkey alt;
1751 struct ib_qp_security {
1753 struct ib_device *dev;
1754 /* Hold this mutex when changing port and pkey settings. */
1756 struct ib_ports_pkeys *ports_pkeys;
1757 /* A list of all open shared QP handles. Required to enforce security
1758 * properly for all users of a shared QP.
1760 struct list_head shared_qp_list;
1763 atomic_t error_list_count;
1764 struct completion error_complete;
1765 int error_comps_pending;
1769 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1770 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1773 struct ib_device *device;
1775 struct ib_cq *send_cq;
1776 struct ib_cq *recv_cq;
1779 struct list_head rdma_mrs;
1780 struct list_head sig_mrs;
1782 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1783 struct list_head xrcd_list;
1785 /* count times opened, mcast attaches, flow attaches */
1787 struct list_head open_list;
1788 struct ib_qp *real_qp;
1789 struct ib_uqp_object *uobject;
1790 void (*event_handler)(struct ib_event *, void *);
1792 /* sgid_attrs associated with the AV's */
1793 const struct ib_gid_attr *av_sgid_attr;
1794 const struct ib_gid_attr *alt_path_sgid_attr;
1798 enum ib_qp_type qp_type;
1799 struct ib_rwq_ind_table *rwq_ind_tbl;
1800 struct ib_qp_security *qp_sec;
1805 * Implementation details of the RDMA core, don't use in drivers:
1807 struct rdma_restrack_entry res;
1809 /* The counter the qp is bind to */
1810 struct rdma_counter *counter;
1814 struct ib_device *device;
1817 struct ib_uobject *uobject;
1822 struct ib_device *device;
1828 unsigned int page_size;
1829 enum ib_mr_type type;
1832 struct ib_uobject *uobject; /* user */
1833 struct list_head qp_entry; /* FR */
1837 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1839 * Implementation details of the RDMA core, don't use in drivers:
1841 struct rdma_restrack_entry res;
1845 struct ib_device *device;
1847 struct ib_uobject *uobject;
1849 enum ib_mw_type type;
1853 struct ib_device *device;
1855 struct list_head list;
1860 /* Supported steering options */
1861 enum ib_flow_attr_type {
1862 /* steering according to rule specifications */
1863 IB_FLOW_ATTR_NORMAL = 0x0,
1864 /* default unicast and multicast rule -
1865 * receive all Eth traffic which isn't steered to any QP
1867 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1868 /* default multicast rule -
1869 * receive all Eth multicast traffic which isn't steered to any QP
1871 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1872 /* sniffer rule - receive all port traffic */
1873 IB_FLOW_ATTR_SNIFFER = 0x3
1876 /* Supported steering header types */
1877 enum ib_flow_spec_type {
1879 IB_FLOW_SPEC_ETH = 0x20,
1880 IB_FLOW_SPEC_IB = 0x22,
1882 IB_FLOW_SPEC_IPV4 = 0x30,
1883 IB_FLOW_SPEC_IPV6 = 0x31,
1884 IB_FLOW_SPEC_ESP = 0x34,
1886 IB_FLOW_SPEC_TCP = 0x40,
1887 IB_FLOW_SPEC_UDP = 0x41,
1888 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1889 IB_FLOW_SPEC_GRE = 0x51,
1890 IB_FLOW_SPEC_MPLS = 0x60,
1891 IB_FLOW_SPEC_INNER = 0x100,
1893 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1894 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1895 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1896 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1898 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1899 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1901 /* Flow steering rule priority is set according to it's domain.
1902 * Lower domain value means higher priority.
1904 enum ib_flow_domain {
1905 IB_FLOW_DOMAIN_USER,
1906 IB_FLOW_DOMAIN_ETHTOOL,
1909 IB_FLOW_DOMAIN_NUM /* Must be last */
1912 enum ib_flow_flags {
1913 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1914 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1915 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1918 struct ib_flow_eth_filter {
1927 struct ib_flow_spec_eth {
1930 struct ib_flow_eth_filter val;
1931 struct ib_flow_eth_filter mask;
1934 struct ib_flow_ib_filter {
1941 struct ib_flow_spec_ib {
1944 struct ib_flow_ib_filter val;
1945 struct ib_flow_ib_filter mask;
1948 /* IPv4 header flags */
1949 enum ib_ipv4_flags {
1950 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1951 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1952 last have this flag set */
1955 struct ib_flow_ipv4_filter {
1966 struct ib_flow_spec_ipv4 {
1969 struct ib_flow_ipv4_filter val;
1970 struct ib_flow_ipv4_filter mask;
1973 struct ib_flow_ipv6_filter {
1984 struct ib_flow_spec_ipv6 {
1987 struct ib_flow_ipv6_filter val;
1988 struct ib_flow_ipv6_filter mask;
1991 struct ib_flow_tcp_udp_filter {
1998 struct ib_flow_spec_tcp_udp {
2001 struct ib_flow_tcp_udp_filter val;
2002 struct ib_flow_tcp_udp_filter mask;
2005 struct ib_flow_tunnel_filter {
2010 /* ib_flow_spec_tunnel describes the Vxlan tunnel
2011 * the tunnel_id from val has the vni value
2013 struct ib_flow_spec_tunnel {
2016 struct ib_flow_tunnel_filter val;
2017 struct ib_flow_tunnel_filter mask;
2020 struct ib_flow_esp_filter {
2027 struct ib_flow_spec_esp {
2030 struct ib_flow_esp_filter val;
2031 struct ib_flow_esp_filter mask;
2034 struct ib_flow_gre_filter {
2035 __be16 c_ks_res0_ver;
2042 struct ib_flow_spec_gre {
2045 struct ib_flow_gre_filter val;
2046 struct ib_flow_gre_filter mask;
2049 struct ib_flow_mpls_filter {
2055 struct ib_flow_spec_mpls {
2058 struct ib_flow_mpls_filter val;
2059 struct ib_flow_mpls_filter mask;
2062 struct ib_flow_spec_action_tag {
2063 enum ib_flow_spec_type type;
2068 struct ib_flow_spec_action_drop {
2069 enum ib_flow_spec_type type;
2073 struct ib_flow_spec_action_handle {
2074 enum ib_flow_spec_type type;
2076 struct ib_flow_action *act;
2079 enum ib_counters_description {
2084 struct ib_flow_spec_action_count {
2085 enum ib_flow_spec_type type;
2087 struct ib_counters *counters;
2090 union ib_flow_spec {
2095 struct ib_flow_spec_eth eth;
2096 struct ib_flow_spec_ib ib;
2097 struct ib_flow_spec_ipv4 ipv4;
2098 struct ib_flow_spec_tcp_udp tcp_udp;
2099 struct ib_flow_spec_ipv6 ipv6;
2100 struct ib_flow_spec_tunnel tunnel;
2101 struct ib_flow_spec_esp esp;
2102 struct ib_flow_spec_gre gre;
2103 struct ib_flow_spec_mpls mpls;
2104 struct ib_flow_spec_action_tag flow_tag;
2105 struct ib_flow_spec_action_drop drop;
2106 struct ib_flow_spec_action_handle action;
2107 struct ib_flow_spec_action_count flow_count;
2110 struct ib_flow_attr {
2111 enum ib_flow_attr_type type;
2117 union ib_flow_spec flows[];
2122 struct ib_device *device;
2123 struct ib_uobject *uobject;
2126 enum ib_flow_action_type {
2127 IB_FLOW_ACTION_UNSPECIFIED,
2128 IB_FLOW_ACTION_ESP = 1,
2131 struct ib_flow_action_attrs_esp_keymats {
2132 enum ib_uverbs_flow_action_esp_keymat protocol;
2134 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2138 struct ib_flow_action_attrs_esp_replays {
2139 enum ib_uverbs_flow_action_esp_replay protocol;
2141 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2145 enum ib_flow_action_attrs_esp_flags {
2146 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2147 * This is done in order to share the same flags between user-space and
2148 * kernel and spare an unnecessary translation.
2152 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2153 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2156 struct ib_flow_spec_list {
2157 struct ib_flow_spec_list *next;
2158 union ib_flow_spec spec;
2161 struct ib_flow_action_attrs_esp {
2162 struct ib_flow_action_attrs_esp_keymats *keymat;
2163 struct ib_flow_action_attrs_esp_replays *replay;
2164 struct ib_flow_spec_list *encap;
2165 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2166 * Value of 0 is a valid value.
2172 /* Use enum ib_flow_action_attrs_esp_flags */
2174 u64 hard_limit_pkts;
2177 struct ib_flow_action {
2178 struct ib_device *device;
2179 struct ib_uobject *uobject;
2180 enum ib_flow_action_type type;
2187 enum ib_process_mad_flags {
2188 IB_MAD_IGNORE_MKEY = 1,
2189 IB_MAD_IGNORE_BKEY = 2,
2190 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2193 enum ib_mad_result {
2194 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2195 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2196 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2197 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2200 struct ib_port_cache {
2202 struct ib_pkey_cache *pkey;
2203 struct ib_gid_table *gid;
2205 enum ib_port_state port_state;
2208 struct ib_port_immutable {
2215 struct ib_port_data {
2216 struct ib_device *ib_dev;
2218 struct ib_port_immutable immutable;
2220 spinlock_t pkey_list_lock;
2221 struct list_head pkey_list;
2223 struct ib_port_cache cache;
2225 spinlock_t netdev_lock;
2226 struct net_device __rcu *netdev;
2227 struct hlist_node ndev_hash_link;
2228 struct rdma_port_counter port_counter;
2229 struct rdma_hw_stats *hw_stats;
2232 /* rdma netdev type - specifies protocol type */
2233 enum rdma_netdev_t {
2234 RDMA_NETDEV_OPA_VNIC,
2239 * struct rdma_netdev - rdma netdev
2240 * For cases where netstack interfacing is required.
2242 struct rdma_netdev {
2244 struct ib_device *hca;
2249 * cleanup function must be specified.
2250 * FIXME: This is only used for OPA_VNIC and that usage should be
2253 void (*free_rdma_netdev)(struct net_device *netdev);
2255 /* control functions */
2256 void (*set_id)(struct net_device *netdev, int id);
2258 int (*send)(struct net_device *dev, struct sk_buff *skb,
2259 struct ib_ah *address, u32 dqpn);
2261 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2262 union ib_gid *gid, u16 mlid,
2263 int set_qkey, u32 qkey);
2264 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2265 union ib_gid *gid, u16 mlid);
2268 struct rdma_netdev_alloc_params {
2274 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2275 struct net_device *netdev, void *param);
2278 struct ib_odp_counters {
2280 atomic64_t invalidations;
2283 struct ib_counters {
2284 struct ib_device *device;
2285 struct ib_uobject *uobject;
2286 /* num of objects attached */
2290 struct ib_counters_read_attr {
2293 u32 flags; /* use enum ib_read_counters_flags */
2296 struct uverbs_attr_bundle;
2298 struct iw_cm_conn_param;
2300 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2301 .size_##ib_struct = \
2302 (sizeof(struct drv_struct) + \
2303 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2304 BUILD_BUG_ON_ZERO( \
2305 !__same_type(((struct drv_struct *)NULL)->member, \
2308 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2309 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2311 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2312 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2314 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2316 struct rdma_user_mmap_entry {
2318 struct ib_ucontext *ucontext;
2319 unsigned long start_pgoff;
2321 bool driver_removed;
2324 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2326 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2328 return (u64)entry->start_pgoff << PAGE_SHIFT;
2332 * struct ib_device_ops - InfiniBand device operations
2333 * This structure defines all the InfiniBand device operations, providers will
2334 * need to define the supported operations, otherwise they will be set to null.
2336 struct ib_device_ops {
2337 struct module *owner;
2338 enum rdma_driver_id driver_id;
2340 unsigned int uverbs_no_driver_id_binding:1;
2342 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2343 const struct ib_send_wr **bad_send_wr);
2344 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2345 const struct ib_recv_wr **bad_recv_wr);
2346 void (*drain_rq)(struct ib_qp *qp);
2347 void (*drain_sq)(struct ib_qp *qp);
2348 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2349 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2350 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2351 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2352 int (*post_srq_recv)(struct ib_srq *srq,
2353 const struct ib_recv_wr *recv_wr,
2354 const struct ib_recv_wr **bad_recv_wr);
2355 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2356 u8 port_num, const struct ib_wc *in_wc,
2357 const struct ib_grh *in_grh,
2358 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2359 size_t *out_mad_size, u16 *out_mad_pkey_index);
2360 int (*query_device)(struct ib_device *device,
2361 struct ib_device_attr *device_attr,
2362 struct ib_udata *udata);
2363 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2364 struct ib_device_modify *device_modify);
2365 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2366 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2368 int (*query_port)(struct ib_device *device, u8 port_num,
2369 struct ib_port_attr *port_attr);
2370 int (*modify_port)(struct ib_device *device, u8 port_num,
2371 int port_modify_mask,
2372 struct ib_port_modify *port_modify);
2374 * The following mandatory functions are used only at device
2375 * registration. Keep functions such as these at the end of this
2376 * structure to avoid cache line misses when accessing struct ib_device
2379 int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2380 struct ib_port_immutable *immutable);
2381 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2384 * When calling get_netdev, the HW vendor's driver should return the
2385 * net device of device @device at port @port_num or NULL if such
2386 * a net device doesn't exist. The vendor driver should call dev_hold
2387 * on this net device. The HW vendor's device driver must guarantee
2388 * that this function returns NULL before the net device has finished
2389 * NETDEV_UNREGISTER state.
2391 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2393 * rdma netdev operation
2395 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2396 * must return -EOPNOTSUPP if it doesn't support the specified type.
2398 struct net_device *(*alloc_rdma_netdev)(
2399 struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2400 const char *name, unsigned char name_assign_type,
2401 void (*setup)(struct net_device *));
2403 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2404 enum rdma_netdev_t type,
2405 struct rdma_netdev_alloc_params *params);
2407 * query_gid should be return GID value for @device, when @port_num
2408 * link layer is either IB or iWarp. It is no-op if @port_num port
2409 * is RoCE link layer.
2411 int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2414 * When calling add_gid, the HW vendor's driver should add the gid
2415 * of device of port at gid index available at @attr. Meta-info of
2416 * that gid (for example, the network device related to this gid) is
2417 * available at @attr. @context allows the HW vendor driver to store
2418 * extra information together with a GID entry. The HW vendor driver may
2419 * allocate memory to contain this information and store it in @context
2420 * when a new GID entry is written to. Params are consistent until the
2421 * next call of add_gid or delete_gid. The function should return 0 on
2422 * success or error otherwise. The function could be called
2423 * concurrently for different ports. This function is only called when
2424 * roce_gid_table is used.
2426 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2428 * When calling del_gid, the HW vendor's driver should delete the
2429 * gid of device @device at gid index gid_index of port port_num
2430 * available in @attr.
2431 * Upon the deletion of a GID entry, the HW vendor must free any
2432 * allocated memory. The caller will clear @context afterwards.
2433 * This function is only called when roce_gid_table is used.
2435 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2436 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2438 int (*alloc_ucontext)(struct ib_ucontext *context,
2439 struct ib_udata *udata);
2440 void (*dealloc_ucontext)(struct ib_ucontext *context);
2441 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2443 * This will be called once refcount of an entry in mmap_xa reaches
2444 * zero. The type of the memory that was mapped may differ between
2445 * entries and is opaque to the rdma_user_mmap interface.
2446 * Therefore needs to be implemented by the driver in mmap_free.
2448 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2449 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2450 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2451 void (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2452 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2453 struct ib_udata *udata);
2454 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2455 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2456 void (*destroy_ah)(struct ib_ah *ah, u32 flags);
2457 int (*create_srq)(struct ib_srq *srq,
2458 struct ib_srq_init_attr *srq_init_attr,
2459 struct ib_udata *udata);
2460 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2461 enum ib_srq_attr_mask srq_attr_mask,
2462 struct ib_udata *udata);
2463 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2464 void (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2465 struct ib_qp *(*create_qp)(struct ib_pd *pd,
2466 struct ib_qp_init_attr *qp_init_attr,
2467 struct ib_udata *udata);
2468 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2469 int qp_attr_mask, struct ib_udata *udata);
2470 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2471 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2472 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2473 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2474 struct ib_udata *udata);
2475 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2476 void (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2477 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2478 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2479 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2480 u64 virt_addr, int mr_access_flags,
2481 struct ib_udata *udata);
2482 int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2483 u64 virt_addr, int mr_access_flags,
2484 struct ib_pd *pd, struct ib_udata *udata);
2485 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2486 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2487 u32 max_num_sg, struct ib_udata *udata);
2488 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2489 u32 max_num_data_sg,
2490 u32 max_num_meta_sg);
2491 int (*advise_mr)(struct ib_pd *pd,
2492 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2493 struct ib_sge *sg_list, u32 num_sge,
2494 struct uverbs_attr_bundle *attrs);
2495 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2496 unsigned int *sg_offset);
2497 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2498 struct ib_mr_status *mr_status);
2499 struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2500 struct ib_udata *udata);
2501 int (*dealloc_mw)(struct ib_mw *mw);
2502 struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags,
2503 struct ib_fmr_attr *fmr_attr);
2504 int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len,
2506 int (*unmap_fmr)(struct list_head *fmr_list);
2507 int (*dealloc_fmr)(struct ib_fmr *fmr);
2508 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2509 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2510 struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device,
2511 struct ib_udata *udata);
2512 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2513 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2514 struct ib_flow_attr *flow_attr,
2515 int domain, struct ib_udata *udata);
2516 int (*destroy_flow)(struct ib_flow *flow_id);
2517 struct ib_flow_action *(*create_flow_action_esp)(
2518 struct ib_device *device,
2519 const struct ib_flow_action_attrs_esp *attr,
2520 struct uverbs_attr_bundle *attrs);
2521 int (*destroy_flow_action)(struct ib_flow_action *action);
2522 int (*modify_flow_action_esp)(
2523 struct ib_flow_action *action,
2524 const struct ib_flow_action_attrs_esp *attr,
2525 struct uverbs_attr_bundle *attrs);
2526 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2528 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2529 struct ifla_vf_info *ivf);
2530 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2531 struct ifla_vf_stats *stats);
2532 int (*get_vf_guid)(struct ib_device *device, int vf, u8 port,
2533 struct ifla_vf_guid *node_guid,
2534 struct ifla_vf_guid *port_guid);
2535 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2537 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2538 struct ib_wq_init_attr *init_attr,
2539 struct ib_udata *udata);
2540 void (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2541 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2542 u32 wq_attr_mask, struct ib_udata *udata);
2543 struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2544 struct ib_device *device,
2545 struct ib_rwq_ind_table_init_attr *init_attr,
2546 struct ib_udata *udata);
2547 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2548 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2549 struct ib_ucontext *context,
2550 struct ib_dm_alloc_attr *attr,
2551 struct uverbs_attr_bundle *attrs);
2552 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2553 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2554 struct ib_dm_mr_attr *attr,
2555 struct uverbs_attr_bundle *attrs);
2556 struct ib_counters *(*create_counters)(
2557 struct ib_device *device, struct uverbs_attr_bundle *attrs);
2558 int (*destroy_counters)(struct ib_counters *counters);
2559 int (*read_counters)(struct ib_counters *counters,
2560 struct ib_counters_read_attr *counters_read_attr,
2561 struct uverbs_attr_bundle *attrs);
2562 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2563 int data_sg_nents, unsigned int *data_sg_offset,
2564 struct scatterlist *meta_sg, int meta_sg_nents,
2565 unsigned int *meta_sg_offset);
2568 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2569 * driver initialized data. The struct is kfree()'ed by the sysfs
2570 * core when the device is removed. A lifespan of -1 in the return
2571 * struct tells the core to set a default lifespan.
2573 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2576 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2577 * @index - The index in the value array we wish to have updated, or
2578 * num_counters if we want all stats updated
2580 * < 0 - Error, no counters updated
2581 * index - Updated the single counter pointed to by index
2582 * num_counters - Updated all counters (will reset the timestamp
2583 * and prevent further calls for lifespan milliseconds)
2584 * Drivers are allowed to update all counters in leiu of just the
2585 * one given in index at their option
2587 int (*get_hw_stats)(struct ib_device *device,
2588 struct rdma_hw_stats *stats, u8 port, int index);
2590 * This function is called once for each port when a ib device is
2593 int (*init_port)(struct ib_device *device, u8 port_num,
2594 struct kobject *port_sysfs);
2596 * Allows rdma drivers to add their own restrack attributes.
2598 int (*fill_res_entry)(struct sk_buff *msg,
2599 struct rdma_restrack_entry *entry);
2601 /* Device lifecycle callbacks */
2603 * Called after the device becomes registered, before clients are
2606 int (*enable_driver)(struct ib_device *dev);
2608 * This is called as part of ib_dealloc_device().
2610 void (*dealloc_driver)(struct ib_device *dev);
2612 /* iWarp CM callbacks */
2613 void (*iw_add_ref)(struct ib_qp *qp);
2614 void (*iw_rem_ref)(struct ib_qp *qp);
2615 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2616 int (*iw_connect)(struct iw_cm_id *cm_id,
2617 struct iw_cm_conn_param *conn_param);
2618 int (*iw_accept)(struct iw_cm_id *cm_id,
2619 struct iw_cm_conn_param *conn_param);
2620 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2622 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2623 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2625 * counter_bind_qp - Bind a QP to a counter.
2626 * @counter - The counter to be bound. If counter->id is zero then
2627 * the driver needs to allocate a new counter and set counter->id
2629 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2631 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2632 * counter and bind it onto the default one
2634 int (*counter_unbind_qp)(struct ib_qp *qp);
2636 * counter_dealloc -De-allocate the hw counter
2638 int (*counter_dealloc)(struct rdma_counter *counter);
2640 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2641 * the driver initialized data.
2643 struct rdma_hw_stats *(*counter_alloc_stats)(
2644 struct rdma_counter *counter);
2646 * counter_update_stats - Query the stats value of this counter
2648 int (*counter_update_stats)(struct rdma_counter *counter);
2651 * Allows rdma drivers to add their own restrack attributes
2652 * dumped via 'rdma stat' iproute2 command.
2654 int (*fill_stat_entry)(struct sk_buff *msg,
2655 struct rdma_restrack_entry *entry);
2657 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2658 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2659 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2660 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2661 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2664 struct ib_core_device {
2665 /* device must be the first element in structure until,
2666 * union of ib_core_device and device exists in ib_device.
2669 possible_net_t rdma_net;
2670 struct kobject *ports_kobj;
2671 struct list_head port_list;
2672 struct ib_device *owner; /* reach back to owner ib_device */
2675 struct rdma_restrack_root;
2677 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2678 struct device *dma_device;
2679 struct ib_device_ops ops;
2680 char name[IB_DEVICE_NAME_MAX];
2681 struct rcu_head rcu_head;
2683 struct list_head event_handler_list;
2684 /* Protects event_handler_list */
2685 struct rw_semaphore event_handler_rwsem;
2687 /* Protects QP's event_handler calls and open_qp list */
2688 spinlock_t qp_open_list_lock;
2690 struct rw_semaphore client_data_rwsem;
2691 struct xarray client_data;
2692 struct mutex unregistration_lock;
2694 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2695 rwlock_t cache_lock;
2697 * port_data is indexed by port number
2699 struct ib_port_data *port_data;
2701 int num_comp_vectors;
2705 struct ib_core_device coredev;
2708 /* First group for device attributes,
2709 * Second group for driver provided attributes (optional).
2710 * It is NULL terminated array.
2712 const struct attribute_group *groups[3];
2714 u64 uverbs_cmd_mask;
2715 u64 uverbs_ex_cmd_mask;
2717 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2721 /* Indicates kernel verbs support, should not be used in drivers */
2722 u16 kverbs_provider:1;
2723 /* CQ adaptive moderation (RDMA DIM) */
2727 struct ib_device_attr attrs;
2728 struct attribute_group *hw_stats_ag;
2729 struct rdma_hw_stats *hw_stats;
2731 #ifdef CONFIG_CGROUP_RDMA
2732 struct rdmacg_device cg_device;
2736 struct rdma_restrack_root *res;
2738 const struct uapi_definition *driver_def;
2741 * Positive refcount indicates that the device is currently
2742 * registered and cannot be unregistered.
2744 refcount_t refcount;
2745 struct completion unreg_completion;
2746 struct work_struct unregistration_work;
2748 const struct rdma_link_ops *link_ops;
2750 /* Protects compat_devs xarray modifications */
2751 struct mutex compat_devs_mutex;
2752 /* Maintains compat devices for each net namespace */
2753 struct xarray compat_devs;
2755 /* Used by iWarp CM */
2756 char iw_ifname[IFNAMSIZ];
2757 u32 iw_driver_flags;
2761 struct ib_client_nl_info;
2764 int (*add)(struct ib_device *ibdev);
2765 void (*remove)(struct ib_device *, void *client_data);
2766 void (*rename)(struct ib_device *dev, void *client_data);
2767 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2768 struct ib_client_nl_info *res);
2769 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2771 /* Returns the net_dev belonging to this ib_client and matching the
2773 * @dev: An RDMA device that the net_dev use for communication.
2774 * @port: A physical port number on the RDMA device.
2775 * @pkey: P_Key that the net_dev uses if applicable.
2776 * @gid: A GID that the net_dev uses to communicate.
2777 * @addr: An IP address the net_dev is configured with.
2778 * @client_data: The device's client data set by ib_set_client_data().
2780 * An ib_client that implements a net_dev on top of RDMA devices
2781 * (such as IP over IB) should implement this callback, allowing the
2782 * rdma_cm module to find the right net_dev for a given request.
2784 * The caller is responsible for calling dev_put on the returned
2786 struct net_device *(*get_net_dev_by_params)(
2787 struct ib_device *dev,
2790 const union ib_gid *gid,
2791 const struct sockaddr *addr,
2795 struct completion uses_zero;
2798 /* kverbs are not required by the client */
2803 * IB block DMA iterator
2805 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2806 * to a HW supported page size.
2808 struct ib_block_iter {
2809 /* internal states */
2810 struct scatterlist *__sg; /* sg holding the current aligned block */
2811 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2812 unsigned int __sg_nents; /* number of SG entries */
2813 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2814 unsigned int __pg_bit; /* alignment of current block */
2817 struct ib_device *_ib_alloc_device(size_t size);
2818 #define ib_alloc_device(drv_struct, member) \
2819 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2820 BUILD_BUG_ON_ZERO(offsetof( \
2821 struct drv_struct, member))), \
2822 struct drv_struct, member)
2824 void ib_dealloc_device(struct ib_device *device);
2826 void ib_get_device_fw_str(struct ib_device *device, char *str);
2828 int ib_register_device(struct ib_device *device, const char *name);
2829 void ib_unregister_device(struct ib_device *device);
2830 void ib_unregister_driver(enum rdma_driver_id driver_id);
2831 void ib_unregister_device_and_put(struct ib_device *device);
2832 void ib_unregister_device_queued(struct ib_device *ib_dev);
2834 int ib_register_client (struct ib_client *client);
2835 void ib_unregister_client(struct ib_client *client);
2837 void __rdma_block_iter_start(struct ib_block_iter *biter,
2838 struct scatterlist *sglist,
2840 unsigned long pgsz);
2841 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2844 * rdma_block_iter_dma_address - get the aligned dma address of the current
2845 * block held by the block iterator.
2846 * @biter: block iterator holding the memory block
2848 static inline dma_addr_t
2849 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2851 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2855 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2856 * @sglist: sglist to iterate over
2857 * @biter: block iterator holding the memory block
2858 * @nents: maximum number of sg entries to iterate over
2859 * @pgsz: best HW supported page size to use
2861 * Callers may use rdma_block_iter_dma_address() to get each
2862 * blocks aligned DMA address.
2864 #define rdma_for_each_block(sglist, biter, nents, pgsz) \
2865 for (__rdma_block_iter_start(biter, sglist, nents, \
2867 __rdma_block_iter_next(biter);)
2870 * ib_get_client_data - Get IB client context
2871 * @device:Device to get context for
2872 * @client:Client to get context for
2874 * ib_get_client_data() returns the client context data set with
2875 * ib_set_client_data(). This can only be called while the client is
2876 * registered to the device, once the ib_client remove() callback returns this
2879 static inline void *ib_get_client_data(struct ib_device *device,
2880 struct ib_client *client)
2882 return xa_load(&device->client_data, client->client_id);
2884 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2886 void ib_set_device_ops(struct ib_device *device,
2887 const struct ib_device_ops *ops);
2889 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2890 unsigned long pfn, unsigned long size, pgprot_t prot,
2891 struct rdma_user_mmap_entry *entry);
2892 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2893 struct rdma_user_mmap_entry *entry,
2895 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2896 struct rdma_user_mmap_entry *entry,
2897 size_t length, u32 min_pgoff,
2900 struct rdma_user_mmap_entry *
2901 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2902 unsigned long pgoff);
2903 struct rdma_user_mmap_entry *
2904 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2905 struct vm_area_struct *vma);
2906 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2908 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2910 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2912 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2915 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2917 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2920 static inline bool ib_is_buffer_cleared(const void __user *p,
2926 if (len > USHRT_MAX)
2929 buf = memdup_user(p, len);
2933 ret = !memchr_inv(buf, 0, len);
2938 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2942 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2946 * ib_is_destroy_retryable - Check whether the uobject destruction
2948 * @ret: The initial destruction return code
2949 * @why: remove reason
2950 * @uobj: The uobject that is destroyed
2952 * This function is a helper function that IB layer and low-level drivers
2953 * can use to consider whether the destruction of the given uobject is
2955 * It checks the original return code, if it wasn't success the destruction
2956 * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2957 * the remove reason. (i.e. why).
2958 * Must be called with the object locked for destroy.
2960 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2961 struct ib_uobject *uobj)
2963 return ret && (why == RDMA_REMOVE_DESTROY ||
2964 uobj->context->cleanup_retryable);
2968 * ib_destroy_usecnt - Called during destruction to check the usecnt
2969 * @usecnt: The usecnt atomic
2970 * @why: remove reason
2971 * @uobj: The uobject that is destroyed
2973 * Non-zero usecnts will block destruction unless destruction was triggered by
2974 * a ucontext cleanup.
2976 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2977 enum rdma_remove_reason why,
2978 struct ib_uobject *uobj)
2980 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2986 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2987 * contains all required attributes and no attributes not allowed for
2988 * the given QP state transition.
2989 * @cur_state: Current QP state
2990 * @next_state: Next QP state
2992 * @mask: Mask of supplied QP attributes
2994 * This function is a helper function that a low-level driver's
2995 * modify_qp method can use to validate the consumer's input. It
2996 * checks that cur_state and next_state are valid QP states, that a
2997 * transition from cur_state to next_state is allowed by the IB spec,
2998 * and that the attribute mask supplied is allowed for the transition.
3000 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
3001 enum ib_qp_type type, enum ib_qp_attr_mask mask);
3003 void ib_register_event_handler(struct ib_event_handler *event_handler);
3004 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
3005 void ib_dispatch_event(const struct ib_event *event);
3007 int ib_query_port(struct ib_device *device,
3008 u8 port_num, struct ib_port_attr *port_attr);
3010 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
3014 * rdma_cap_ib_switch - Check if the device is IB switch
3015 * @device: Device to check
3017 * Device driver is responsible for setting is_switch bit on
3018 * in ib_device structure at init time.
3020 * Return: true if the device is IB switch.
3022 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
3024 return device->is_switch;
3028 * rdma_start_port - Return the first valid port number for the device
3031 * @device: Device to be checked
3033 * Return start port number
3035 static inline u8 rdma_start_port(const struct ib_device *device)
3037 return rdma_cap_ib_switch(device) ? 0 : 1;
3041 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3042 * @device - The struct ib_device * to iterate over
3043 * @iter - The unsigned int to store the port number
3045 #define rdma_for_each_port(device, iter) \
3046 for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type( \
3047 unsigned int, iter))); \
3048 iter <= rdma_end_port(device); (iter)++)
3051 * rdma_end_port - Return the last valid port number for the device
3054 * @device: Device to be checked
3056 * Return last port number
3058 static inline u8 rdma_end_port(const struct ib_device *device)
3060 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3063 static inline int rdma_is_port_valid(const struct ib_device *device,
3066 return (port >= rdma_start_port(device) &&
3067 port <= rdma_end_port(device));
3070 static inline bool rdma_is_grh_required(const struct ib_device *device,
3073 return device->port_data[port_num].immutable.core_cap_flags &
3074 RDMA_CORE_PORT_IB_GRH_REQUIRED;
3077 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
3079 return device->port_data[port_num].immutable.core_cap_flags &
3080 RDMA_CORE_CAP_PROT_IB;
3083 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
3085 return device->port_data[port_num].immutable.core_cap_flags &
3086 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3089 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
3091 return device->port_data[port_num].immutable.core_cap_flags &
3092 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3095 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
3097 return device->port_data[port_num].immutable.core_cap_flags &
3098 RDMA_CORE_CAP_PROT_ROCE;
3101 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
3103 return device->port_data[port_num].immutable.core_cap_flags &
3104 RDMA_CORE_CAP_PROT_IWARP;
3107 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
3109 return rdma_protocol_ib(device, port_num) ||
3110 rdma_protocol_roce(device, port_num);
3113 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
3115 return device->port_data[port_num].immutable.core_cap_flags &
3116 RDMA_CORE_CAP_PROT_RAW_PACKET;
3119 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
3121 return device->port_data[port_num].immutable.core_cap_flags &
3122 RDMA_CORE_CAP_PROT_USNIC;
3126 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3127 * Management Datagrams.
3128 * @device: Device to check
3129 * @port_num: Port number to check
3131 * Management Datagrams (MAD) are a required part of the InfiniBand
3132 * specification and are supported on all InfiniBand devices. A slightly
3133 * extended version are also supported on OPA interfaces.
3135 * Return: true if the port supports sending/receiving of MAD packets.
3137 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
3139 return device->port_data[port_num].immutable.core_cap_flags &
3140 RDMA_CORE_CAP_IB_MAD;
3144 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3145 * Management Datagrams.
3146 * @device: Device to check
3147 * @port_num: Port number to check
3149 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3150 * datagrams with their own versions. These OPA MADs share many but not all of
3151 * the characteristics of InfiniBand MADs.
3153 * OPA MADs differ in the following ways:
3155 * 1) MADs are variable size up to 2K
3156 * IBTA defined MADs remain fixed at 256 bytes
3157 * 2) OPA SMPs must carry valid PKeys
3158 * 3) OPA SMP packets are a different format
3160 * Return: true if the port supports OPA MAD packet formats.
3162 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
3164 return device->port_data[port_num].immutable.core_cap_flags &
3165 RDMA_CORE_CAP_OPA_MAD;
3169 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3170 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3171 * @device: Device to check
3172 * @port_num: Port number to check
3174 * Each InfiniBand node is required to provide a Subnet Management Agent
3175 * that the subnet manager can access. Prior to the fabric being fully
3176 * configured by the subnet manager, the SMA is accessed via a well known
3177 * interface called the Subnet Management Interface (SMI). This interface
3178 * uses directed route packets to communicate with the SM to get around the
3179 * chicken and egg problem of the SM needing to know what's on the fabric
3180 * in order to configure the fabric, and needing to configure the fabric in
3181 * order to send packets to the devices on the fabric. These directed
3182 * route packets do not need the fabric fully configured in order to reach
3183 * their destination. The SMI is the only method allowed to send
3184 * directed route packets on an InfiniBand fabric.
3186 * Return: true if the port provides an SMI.
3188 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
3190 return device->port_data[port_num].immutable.core_cap_flags &
3191 RDMA_CORE_CAP_IB_SMI;
3195 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3196 * Communication Manager.
3197 * @device: Device to check
3198 * @port_num: Port number to check
3200 * The InfiniBand Communication Manager is one of many pre-defined General
3201 * Service Agents (GSA) that are accessed via the General Service
3202 * Interface (GSI). It's role is to facilitate establishment of connections
3203 * between nodes as well as other management related tasks for established
3206 * Return: true if the port supports an IB CM (this does not guarantee that
3207 * a CM is actually running however).
3209 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
3211 return device->port_data[port_num].immutable.core_cap_flags &
3212 RDMA_CORE_CAP_IB_CM;
3216 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3217 * Communication Manager.
3218 * @device: Device to check
3219 * @port_num: Port number to check
3221 * Similar to above, but specific to iWARP connections which have a different
3222 * managment protocol than InfiniBand.
3224 * Return: true if the port supports an iWARP CM (this does not guarantee that
3225 * a CM is actually running however).
3227 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
3229 return device->port_data[port_num].immutable.core_cap_flags &
3230 RDMA_CORE_CAP_IW_CM;
3234 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3235 * Subnet Administration.
3236 * @device: Device to check
3237 * @port_num: Port number to check
3239 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3240 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3241 * fabrics, devices should resolve routes to other hosts by contacting the
3242 * SA to query the proper route.
3244 * Return: true if the port should act as a client to the fabric Subnet
3245 * Administration interface. This does not imply that the SA service is
3248 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3250 return device->port_data[port_num].immutable.core_cap_flags &
3251 RDMA_CORE_CAP_IB_SA;
3255 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3257 * @device: Device to check
3258 * @port_num: Port number to check
3260 * InfiniBand multicast registration is more complex than normal IPv4 or
3261 * IPv6 multicast registration. Each Host Channel Adapter must register
3262 * with the Subnet Manager when it wishes to join a multicast group. It
3263 * should do so only once regardless of how many queue pairs it subscribes
3264 * to this group. And it should leave the group only after all queue pairs
3265 * attached to the group have been detached.
3267 * Return: true if the port must undertake the additional adminstrative
3268 * overhead of registering/unregistering with the SM and tracking of the
3269 * total number of queue pairs attached to the multicast group.
3271 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3273 return rdma_cap_ib_sa(device, port_num);
3277 * rdma_cap_af_ib - Check if the port of device has the capability
3278 * Native Infiniband Address.
3279 * @device: Device to check
3280 * @port_num: Port number to check
3282 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3283 * GID. RoCE uses a different mechanism, but still generates a GID via
3284 * a prescribed mechanism and port specific data.
3286 * Return: true if the port uses a GID address to identify devices on the
3289 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3291 return device->port_data[port_num].immutable.core_cap_flags &
3292 RDMA_CORE_CAP_AF_IB;
3296 * rdma_cap_eth_ah - Check if the port of device has the capability
3297 * Ethernet Address Handle.
3298 * @device: Device to check
3299 * @port_num: Port number to check
3301 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3302 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3303 * port. Normally, packet headers are generated by the sending host
3304 * adapter, but when sending connectionless datagrams, we must manually
3305 * inject the proper headers for the fabric we are communicating over.
3307 * Return: true if we are running as a RoCE port and must force the
3308 * addition of a Global Route Header built from our Ethernet Address
3309 * Handle into our header list for connectionless packets.
3311 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3313 return device->port_data[port_num].immutable.core_cap_flags &
3314 RDMA_CORE_CAP_ETH_AH;
3318 * rdma_cap_opa_ah - Check if the port of device supports
3319 * OPA Address handles
3320 * @device: Device to check
3321 * @port_num: Port number to check
3323 * Return: true if we are running on an OPA device which supports
3324 * the extended OPA addressing.
3326 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3328 return (device->port_data[port_num].immutable.core_cap_flags &
3329 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3333 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3336 * @port_num: Port number
3338 * This MAD size includes the MAD headers and MAD payload. No other headers
3341 * Return the max MAD size required by the Port. Will return 0 if the port
3342 * does not support MADs
3344 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3346 return device->port_data[port_num].immutable.max_mad_size;
3350 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3351 * @device: Device to check
3352 * @port_num: Port number to check
3354 * RoCE GID table mechanism manages the various GIDs for a device.
3356 * NOTE: if allocating the port's GID table has failed, this call will still
3357 * return true, but any RoCE GID table API will fail.
3359 * Return: true if the port uses RoCE GID table mechanism in order to manage
3362 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3365 return rdma_protocol_roce(device, port_num) &&
3366 device->ops.add_gid && device->ops.del_gid;
3370 * Check if the device supports READ W/ INVALIDATE.
3372 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3375 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3376 * has support for it yet.
3378 return rdma_protocol_iwarp(dev, port_num);
3382 * rdma_find_pg_bit - Find page bit given address and HW supported page sizes
3385 * @pgsz_bitmap: bitmap of HW supported page sizes
3387 static inline unsigned int rdma_find_pg_bit(unsigned long addr,
3388 unsigned long pgsz_bitmap)
3390 unsigned long align;
3393 align = addr & -addr;
3395 /* Find page bit such that addr is aligned to the highest supported
3398 pgsz = pgsz_bitmap & ~(-align << 1);
3400 return __ffs(pgsz_bitmap);
3406 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3408 * @port_num: 1 based Port number
3410 * Return true if port is an Intel OPA port , false if not
3412 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3415 return (device->port_data[port_num].immutable.core_cap_flags &
3416 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3420 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3422 * @port_num: Port number
3423 * @mtu: enum value of MTU
3425 * Return the MTU size supported by the port as an integer value. Will return
3426 * -1 if enum value of mtu is not supported.
3428 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u8 port,
3431 if (rdma_core_cap_opa_port(device, port))
3432 return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3434 return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3438 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3440 * @port_num: Port number
3441 * @attr: port attribute
3443 * Return the MTU size supported by the port as an integer value.
3445 static inline int rdma_mtu_from_attr(struct ib_device *device, u8 port,
3446 struct ib_port_attr *attr)
3448 if (rdma_core_cap_opa_port(device, port))
3449 return attr->phys_mtu;
3451 return ib_mtu_enum_to_int(attr->max_mtu);
3454 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3456 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3457 struct ifla_vf_info *info);
3458 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3459 struct ifla_vf_stats *stats);
3460 int ib_get_vf_guid(struct ib_device *device, int vf, u8 port,
3461 struct ifla_vf_guid *node_guid,
3462 struct ifla_vf_guid *port_guid);
3463 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3466 int ib_query_pkey(struct ib_device *device,
3467 u8 port_num, u16 index, u16 *pkey);
3469 int ib_modify_device(struct ib_device *device,
3470 int device_modify_mask,
3471 struct ib_device_modify *device_modify);
3473 int ib_modify_port(struct ib_device *device,
3474 u8 port_num, int port_modify_mask,
3475 struct ib_port_modify *port_modify);
3477 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3478 u8 *port_num, u16 *index);
3480 int ib_find_pkey(struct ib_device *device,
3481 u8 port_num, u16 pkey, u16 *index);
3485 * Create a memory registration for all memory in the system and place
3486 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3487 * ULPs to avoid the overhead of dynamic MRs.
3489 * This flag is generally considered unsafe and must only be used in
3490 * extremly trusted environments. Every use of it will log a warning
3491 * in the kernel log.
3493 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3496 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3497 const char *caller);
3499 #define ib_alloc_pd(device, flags) \
3500 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3503 * ib_dealloc_pd_user - Deallocate kernel/user PD
3504 * @pd: The protection domain
3505 * @udata: Valid user data or NULL for kernel objects
3507 void ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3510 * ib_dealloc_pd - Deallocate kernel PD
3511 * @pd: The protection domain
3513 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3515 static inline void ib_dealloc_pd(struct ib_pd *pd)
3517 ib_dealloc_pd_user(pd, NULL);
3520 enum rdma_create_ah_flags {
3521 /* In a sleepable context */
3522 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3526 * rdma_create_ah - Creates an address handle for the given address vector.
3527 * @pd: The protection domain associated with the address handle.
3528 * @ah_attr: The attributes of the address vector.
3529 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3531 * The address handle is used to reference a local or global destination
3532 * in all UD QP post sends.
3534 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3538 * rdma_create_user_ah - Creates an address handle for the given address vector.
3539 * It resolves destination mac address for ah attribute of RoCE type.
3540 * @pd: The protection domain associated with the address handle.
3541 * @ah_attr: The attributes of the address vector.
3542 * @udata: pointer to user's input output buffer information need by
3545 * It returns 0 on success and returns appropriate error code on error.
3546 * The address handle is used to reference a local or global destination
3547 * in all UD QP post sends.
3549 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3550 struct rdma_ah_attr *ah_attr,
3551 struct ib_udata *udata);
3553 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3555 * @hdr: the L3 header to parse
3556 * @net_type: type of header to parse
3557 * @sgid: place to store source gid
3558 * @dgid: place to store destination gid
3560 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3561 enum rdma_network_type net_type,
3562 union ib_gid *sgid, union ib_gid *dgid);
3565 * ib_get_rdma_header_version - Get the header version
3566 * @hdr: the L3 header to parse
3568 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3571 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3573 * @device: Device on which the received message arrived.
3574 * @port_num: Port on which the received message arrived.
3575 * @wc: Work completion associated with the received message.
3576 * @grh: References the received global route header. This parameter is
3577 * ignored unless the work completion indicates that the GRH is valid.
3578 * @ah_attr: Returned attributes that can be used when creating an address
3579 * handle for replying to the message.
3580 * When ib_init_ah_attr_from_wc() returns success,
3581 * (a) for IB link layer it optionally contains a reference to SGID attribute
3582 * when GRH is present for IB link layer.
3583 * (b) for RoCE link layer it contains a reference to SGID attribute.
3584 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3585 * attributes which are initialized using ib_init_ah_attr_from_wc().
3588 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3589 const struct ib_wc *wc, const struct ib_grh *grh,
3590 struct rdma_ah_attr *ah_attr);
3593 * ib_create_ah_from_wc - Creates an address handle associated with the
3594 * sender of the specified work completion.
3595 * @pd: The protection domain associated with the address handle.
3596 * @wc: Work completion information associated with a received message.
3597 * @grh: References the received global route header. This parameter is
3598 * ignored unless the work completion indicates that the GRH is valid.
3599 * @port_num: The outbound port number to associate with the address.
3601 * The address handle is used to reference a local or global destination
3602 * in all UD QP post sends.
3604 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3605 const struct ib_grh *grh, u8 port_num);
3608 * rdma_modify_ah - Modifies the address vector associated with an address
3610 * @ah: The address handle to modify.
3611 * @ah_attr: The new address vector attributes to associate with the
3614 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3617 * rdma_query_ah - Queries the address vector associated with an address
3619 * @ah: The address handle to query.
3620 * @ah_attr: The address vector attributes associated with the address
3623 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3625 enum rdma_destroy_ah_flags {
3626 /* In a sleepable context */
3627 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3631 * rdma_destroy_ah_user - Destroys an address handle.
3632 * @ah: The address handle to destroy.
3633 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3634 * @udata: Valid user data or NULL for kernel objects
3636 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3639 * rdma_destroy_ah - Destroys an kernel address handle.
3640 * @ah: The address handle to destroy.
3641 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3643 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3645 static inline int rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3647 return rdma_destroy_ah_user(ah, flags, NULL);
3650 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3651 struct ib_srq_init_attr *srq_init_attr,
3652 struct ib_usrq_object *uobject,
3653 struct ib_udata *udata);
3654 static inline struct ib_srq *
3655 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3657 if (!pd->device->ops.create_srq)
3658 return ERR_PTR(-EOPNOTSUPP);
3660 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3664 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3665 * @srq: The SRQ to modify.
3666 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3667 * the current values of selected SRQ attributes are returned.
3668 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3669 * are being modified.
3671 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3672 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3673 * the number of receives queued drops below the limit.
3675 int ib_modify_srq(struct ib_srq *srq,
3676 struct ib_srq_attr *srq_attr,
3677 enum ib_srq_attr_mask srq_attr_mask);
3680 * ib_query_srq - Returns the attribute list and current values for the
3682 * @srq: The SRQ to query.
3683 * @srq_attr: The attributes of the specified SRQ.
3685 int ib_query_srq(struct ib_srq *srq,
3686 struct ib_srq_attr *srq_attr);
3689 * ib_destroy_srq_user - Destroys the specified SRQ.
3690 * @srq: The SRQ to destroy.
3691 * @udata: Valid user data or NULL for kernel objects
3693 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3696 * ib_destroy_srq - Destroys the specified kernel SRQ.
3697 * @srq: The SRQ to destroy.
3699 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3701 static inline int ib_destroy_srq(struct ib_srq *srq)
3703 return ib_destroy_srq_user(srq, NULL);
3707 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3708 * @srq: The SRQ to post the work request on.
3709 * @recv_wr: A list of work requests to post on the receive queue.
3710 * @bad_recv_wr: On an immediate failure, this parameter will reference
3711 * the work request that failed to be posted on the QP.
3713 static inline int ib_post_srq_recv(struct ib_srq *srq,
3714 const struct ib_recv_wr *recv_wr,
3715 const struct ib_recv_wr **bad_recv_wr)
3717 const struct ib_recv_wr *dummy;
3719 return srq->device->ops.post_srq_recv(srq, recv_wr,
3720 bad_recv_wr ? : &dummy);
3723 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3724 struct ib_qp_init_attr *qp_init_attr);
3727 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3728 * @qp: The QP to modify.
3729 * @attr: On input, specifies the QP attributes to modify. On output,
3730 * the current values of selected QP attributes are returned.
3731 * @attr_mask: A bit-mask used to specify which attributes of the QP
3732 * are being modified.
3733 * @udata: pointer to user's input output buffer information
3734 * are being modified.
3735 * It returns 0 on success and returns appropriate error code on error.
3737 int ib_modify_qp_with_udata(struct ib_qp *qp,
3738 struct ib_qp_attr *attr,
3740 struct ib_udata *udata);
3743 * ib_modify_qp - Modifies the attributes for the specified QP and then
3744 * transitions the QP to the given state.
3745 * @qp: The QP to modify.
3746 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3747 * the current values of selected QP attributes are returned.
3748 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3749 * are being modified.
3751 int ib_modify_qp(struct ib_qp *qp,
3752 struct ib_qp_attr *qp_attr,
3756 * ib_query_qp - Returns the attribute list and current values for the
3758 * @qp: The QP to query.
3759 * @qp_attr: The attributes of the specified QP.
3760 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3761 * @qp_init_attr: Additional attributes of the selected QP.
3763 * The qp_attr_mask may be used to limit the query to gathering only the
3764 * selected attributes.
3766 int ib_query_qp(struct ib_qp *qp,
3767 struct ib_qp_attr *qp_attr,
3769 struct ib_qp_init_attr *qp_init_attr);
3772 * ib_destroy_qp - Destroys the specified QP.
3773 * @qp: The QP to destroy.
3774 * @udata: Valid udata or NULL for kernel objects
3776 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3779 * ib_destroy_qp - Destroys the specified kernel QP.
3780 * @qp: The QP to destroy.
3782 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3784 static inline int ib_destroy_qp(struct ib_qp *qp)
3786 return ib_destroy_qp_user(qp, NULL);
3790 * ib_open_qp - Obtain a reference to an existing sharable QP.
3791 * @xrcd - XRC domain
3792 * @qp_open_attr: Attributes identifying the QP to open.
3794 * Returns a reference to a sharable QP.
3796 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3797 struct ib_qp_open_attr *qp_open_attr);
3800 * ib_close_qp - Release an external reference to a QP.
3801 * @qp: The QP handle to release
3803 * The opened QP handle is released by the caller. The underlying
3804 * shared QP is not destroyed until all internal references are released.
3806 int ib_close_qp(struct ib_qp *qp);
3809 * ib_post_send - Posts a list of work requests to the send queue of
3811 * @qp: The QP to post the work request on.
3812 * @send_wr: A list of work requests to post on the send queue.
3813 * @bad_send_wr: On an immediate failure, this parameter will reference
3814 * the work request that failed to be posted on the QP.
3816 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3817 * error is returned, the QP state shall not be affected,
3818 * ib_post_send() will return an immediate error after queueing any
3819 * earlier work requests in the list.
3821 static inline int ib_post_send(struct ib_qp *qp,
3822 const struct ib_send_wr *send_wr,
3823 const struct ib_send_wr **bad_send_wr)
3825 const struct ib_send_wr *dummy;
3827 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3831 * ib_post_recv - Posts a list of work requests to the receive queue of
3833 * @qp: The QP to post the work request on.
3834 * @recv_wr: A list of work requests to post on the receive queue.
3835 * @bad_recv_wr: On an immediate failure, this parameter will reference
3836 * the work request that failed to be posted on the QP.
3838 static inline int ib_post_recv(struct ib_qp *qp,
3839 const struct ib_recv_wr *recv_wr,
3840 const struct ib_recv_wr **bad_recv_wr)
3842 const struct ib_recv_wr *dummy;
3844 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3847 struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private,
3848 int nr_cqe, int comp_vector,
3849 enum ib_poll_context poll_ctx,
3850 const char *caller, struct ib_udata *udata);
3853 * ib_alloc_cq_user: Allocate kernel/user CQ
3854 * @dev: The IB device
3855 * @private: Private data attached to the CQE
3856 * @nr_cqe: Number of CQEs in the CQ
3857 * @comp_vector: Completion vector used for the IRQs
3858 * @poll_ctx: Context used for polling the CQ
3859 * @udata: Valid user data or NULL for kernel objects
3861 static inline struct ib_cq *ib_alloc_cq_user(struct ib_device *dev,
3862 void *private, int nr_cqe,
3864 enum ib_poll_context poll_ctx,
3865 struct ib_udata *udata)
3867 return __ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3868 KBUILD_MODNAME, udata);
3872 * ib_alloc_cq: Allocate kernel CQ
3873 * @dev: The IB device
3874 * @private: Private data attached to the CQE
3875 * @nr_cqe: Number of CQEs in the CQ
3876 * @comp_vector: Completion vector used for the IRQs
3877 * @poll_ctx: Context used for polling the CQ
3879 * NOTE: for user cq use ib_alloc_cq_user with valid udata!
3881 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3882 int nr_cqe, int comp_vector,
3883 enum ib_poll_context poll_ctx)
3885 return ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3889 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3890 int nr_cqe, enum ib_poll_context poll_ctx,
3891 const char *caller);
3894 * ib_alloc_cq_any: Allocate kernel CQ
3895 * @dev: The IB device
3896 * @private: Private data attached to the CQE
3897 * @nr_cqe: Number of CQEs in the CQ
3898 * @poll_ctx: Context used for polling the CQ
3900 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3901 void *private, int nr_cqe,
3902 enum ib_poll_context poll_ctx)
3904 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3909 * ib_free_cq_user - Free kernel/user CQ
3910 * @cq: The CQ to free
3911 * @udata: Valid user data or NULL for kernel objects
3913 void ib_free_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3916 * ib_free_cq - Free kernel CQ
3917 * @cq: The CQ to free
3919 * NOTE: for user cq use ib_free_cq_user with valid udata!
3921 static inline void ib_free_cq(struct ib_cq *cq)
3923 ib_free_cq_user(cq, NULL);
3926 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3929 * ib_create_cq - Creates a CQ on the specified device.
3930 * @device: The device on which to create the CQ.
3931 * @comp_handler: A user-specified callback that is invoked when a
3932 * completion event occurs on the CQ.
3933 * @event_handler: A user-specified callback that is invoked when an
3934 * asynchronous event not associated with a completion occurs on the CQ.
3935 * @cq_context: Context associated with the CQ returned to the user via
3936 * the associated completion and event handlers.
3937 * @cq_attr: The attributes the CQ should be created upon.
3939 * Users can examine the cq structure to determine the actual CQ size.
3941 struct ib_cq *__ib_create_cq(struct ib_device *device,
3942 ib_comp_handler comp_handler,
3943 void (*event_handler)(struct ib_event *, void *),
3945 const struct ib_cq_init_attr *cq_attr,
3946 const char *caller);
3947 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3948 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3951 * ib_resize_cq - Modifies the capacity of the CQ.
3952 * @cq: The CQ to resize.
3953 * @cqe: The minimum size of the CQ.
3955 * Users can examine the cq structure to determine the actual CQ size.
3957 int ib_resize_cq(struct ib_cq *cq, int cqe);
3960 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3961 * @cq: The CQ to modify.
3962 * @cq_count: number of CQEs that will trigger an event
3963 * @cq_period: max period of time in usec before triggering an event
3966 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3969 * ib_destroy_cq_user - Destroys the specified CQ.
3970 * @cq: The CQ to destroy.
3971 * @udata: Valid user data or NULL for kernel objects
3973 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3976 * ib_destroy_cq - Destroys the specified kernel CQ.
3977 * @cq: The CQ to destroy.
3979 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3981 static inline void ib_destroy_cq(struct ib_cq *cq)
3983 ib_destroy_cq_user(cq, NULL);
3987 * ib_poll_cq - poll a CQ for completion(s)
3988 * @cq:the CQ being polled
3989 * @num_entries:maximum number of completions to return
3990 * @wc:array of at least @num_entries &struct ib_wc where completions
3993 * Poll a CQ for (possibly multiple) completions. If the return value
3994 * is < 0, an error occurred. If the return value is >= 0, it is the
3995 * number of completions returned. If the return value is
3996 * non-negative and < num_entries, then the CQ was emptied.
3998 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
4001 return cq->device->ops.poll_cq(cq, num_entries, wc);
4005 * ib_req_notify_cq - Request completion notification on a CQ.
4006 * @cq: The CQ to generate an event for.
4008 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
4009 * to request an event on the next solicited event or next work
4010 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
4011 * may also be |ed in to request a hint about missed events, as
4015 * < 0 means an error occurred while requesting notification
4016 * == 0 means notification was requested successfully, and if
4017 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
4018 * were missed and it is safe to wait for another event. In
4019 * this case is it guaranteed that any work completions added
4020 * to the CQ since the last CQ poll will trigger a completion
4021 * notification event.
4022 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
4023 * in. It means that the consumer must poll the CQ again to
4024 * make sure it is empty to avoid missing an event because of a
4025 * race between requesting notification and an entry being
4026 * added to the CQ. This return value means it is possible
4027 * (but not guaranteed) that a work completion has been added
4028 * to the CQ since the last poll without triggering a
4029 * completion notification event.
4031 static inline int ib_req_notify_cq(struct ib_cq *cq,
4032 enum ib_cq_notify_flags flags)
4034 return cq->device->ops.req_notify_cq(cq, flags);
4038 * ib_req_ncomp_notif - Request completion notification when there are
4039 * at least the specified number of unreaped completions on the CQ.
4040 * @cq: The CQ to generate an event for.
4041 * @wc_cnt: The number of unreaped completions that should be on the
4042 * CQ before an event is generated.
4044 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
4046 return cq->device->ops.req_ncomp_notif ?
4047 cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
4052 * ib_dma_mapping_error - check a DMA addr for error
4053 * @dev: The device for which the dma_addr was created
4054 * @dma_addr: The DMA address to check
4056 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
4058 return dma_mapping_error(dev->dma_device, dma_addr);
4062 * ib_dma_map_single - Map a kernel virtual address to DMA address
4063 * @dev: The device for which the dma_addr is to be created
4064 * @cpu_addr: The kernel virtual address
4065 * @size: The size of the region in bytes
4066 * @direction: The direction of the DMA
4068 static inline u64 ib_dma_map_single(struct ib_device *dev,
4069 void *cpu_addr, size_t size,
4070 enum dma_data_direction direction)
4072 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
4076 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4077 * @dev: The device for which the DMA address was created
4078 * @addr: The DMA address
4079 * @size: The size of the region in bytes
4080 * @direction: The direction of the DMA
4082 static inline void ib_dma_unmap_single(struct ib_device *dev,
4083 u64 addr, size_t size,
4084 enum dma_data_direction direction)
4086 dma_unmap_single(dev->dma_device, addr, size, direction);
4090 * ib_dma_map_page - Map a physical page to DMA address
4091 * @dev: The device for which the dma_addr is to be created
4092 * @page: The page to be mapped
4093 * @offset: The offset within the page
4094 * @size: The size of the region in bytes
4095 * @direction: The direction of the DMA
4097 static inline u64 ib_dma_map_page(struct ib_device *dev,
4099 unsigned long offset,
4101 enum dma_data_direction direction)
4103 return dma_map_page(dev->dma_device, page, offset, size, direction);
4107 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4108 * @dev: The device for which the DMA address was created
4109 * @addr: The DMA address
4110 * @size: The size of the region in bytes
4111 * @direction: The direction of the DMA
4113 static inline void ib_dma_unmap_page(struct ib_device *dev,
4114 u64 addr, size_t size,
4115 enum dma_data_direction direction)
4117 dma_unmap_page(dev->dma_device, addr, size, direction);
4121 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4122 * @dev: The device for which the DMA addresses are to be created
4123 * @sg: The array of scatter/gather entries
4124 * @nents: The number of scatter/gather entries
4125 * @direction: The direction of the DMA
4127 static inline int ib_dma_map_sg(struct ib_device *dev,
4128 struct scatterlist *sg, int nents,
4129 enum dma_data_direction direction)
4131 return dma_map_sg(dev->dma_device, sg, nents, direction);
4135 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4136 * @dev: The device for which the DMA addresses were created
4137 * @sg: The array of scatter/gather entries
4138 * @nents: The number of scatter/gather entries
4139 * @direction: The direction of the DMA
4141 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4142 struct scatterlist *sg, int nents,
4143 enum dma_data_direction direction)
4145 dma_unmap_sg(dev->dma_device, sg, nents, direction);
4148 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4149 struct scatterlist *sg, int nents,
4150 enum dma_data_direction direction,
4151 unsigned long dma_attrs)
4153 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4157 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4158 struct scatterlist *sg, int nents,
4159 enum dma_data_direction direction,
4160 unsigned long dma_attrs)
4162 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
4166 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4167 * @dev: The device to query
4169 * The returned value represents a size in bytes.
4171 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4173 return dma_get_max_seg_size(dev->dma_device);
4177 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4178 * @dev: The device for which the DMA address was created
4179 * @addr: The DMA address
4180 * @size: The size of the region in bytes
4181 * @dir: The direction of the DMA
4183 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4186 enum dma_data_direction dir)
4188 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4192 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4193 * @dev: The device for which the DMA address was created
4194 * @addr: The DMA address
4195 * @size: The size of the region in bytes
4196 * @dir: The direction of the DMA
4198 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4201 enum dma_data_direction dir)
4203 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4207 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
4208 * @dev: The device for which the DMA address is requested
4209 * @size: The size of the region to allocate in bytes
4210 * @dma_handle: A pointer for returning the DMA address of the region
4211 * @flag: memory allocator flags
4213 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
4215 dma_addr_t *dma_handle,
4218 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
4222 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
4223 * @dev: The device for which the DMA addresses were allocated
4224 * @size: The size of the region
4225 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
4226 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
4228 static inline void ib_dma_free_coherent(struct ib_device *dev,
4229 size_t size, void *cpu_addr,
4230 dma_addr_t dma_handle)
4232 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
4235 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4236 * space. This function should be called when 'current' is the owning MM.
4238 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4239 u64 virt_addr, int mr_access_flags);
4241 /* ib_advise_mr - give an advice about an address range in a memory region */
4242 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4243 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4245 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4246 * HCA translation table.
4247 * @mr: The memory region to deregister.
4248 * @udata: Valid user data or NULL for kernel object
4250 * This function can fail, if the memory region has memory windows bound to it.
4252 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4255 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4256 * HCA translation table.
4257 * @mr: The memory region to deregister.
4259 * This function can fail, if the memory region has memory windows bound to it.
4261 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4263 static inline int ib_dereg_mr(struct ib_mr *mr)
4265 return ib_dereg_mr_user(mr, NULL);
4268 struct ib_mr *ib_alloc_mr_user(struct ib_pd *pd, enum ib_mr_type mr_type,
4269 u32 max_num_sg, struct ib_udata *udata);
4271 static inline struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
4272 enum ib_mr_type mr_type, u32 max_num_sg)
4274 return ib_alloc_mr_user(pd, mr_type, max_num_sg, NULL);
4277 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4278 u32 max_num_data_sg,
4279 u32 max_num_meta_sg);
4282 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4284 * @mr - struct ib_mr pointer to be updated.
4285 * @newkey - new key to be used.
4287 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4289 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4290 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4294 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4295 * for calculating a new rkey for type 2 memory windows.
4296 * @rkey - the rkey to increment.
4298 static inline u32 ib_inc_rkey(u32 rkey)
4300 const u32 mask = 0x000000ff;
4301 return ((rkey + 1) & mask) | (rkey & ~mask);
4305 * ib_alloc_fmr - Allocates a unmapped fast memory region.
4306 * @pd: The protection domain associated with the unmapped region.
4307 * @mr_access_flags: Specifies the memory access rights.
4308 * @fmr_attr: Attributes of the unmapped region.
4310 * A fast memory region must be mapped before it can be used as part of
4313 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
4314 int mr_access_flags,
4315 struct ib_fmr_attr *fmr_attr);
4318 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
4319 * @fmr: The fast memory region to associate with the pages.
4320 * @page_list: An array of physical pages to map to the fast memory region.
4321 * @list_len: The number of pages in page_list.
4322 * @iova: The I/O virtual address to use with the mapped region.
4324 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
4325 u64 *page_list, int list_len,
4328 return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova);
4332 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
4333 * @fmr_list: A linked list of fast memory regions to unmap.
4335 int ib_unmap_fmr(struct list_head *fmr_list);
4338 * ib_dealloc_fmr - Deallocates a fast memory region.
4339 * @fmr: The fast memory region to deallocate.
4341 int ib_dealloc_fmr(struct ib_fmr *fmr);
4344 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4345 * @qp: QP to attach to the multicast group. The QP must be type
4347 * @gid: Multicast group GID.
4348 * @lid: Multicast group LID in host byte order.
4350 * In order to send and receive multicast packets, subnet
4351 * administration must have created the multicast group and configured
4352 * the fabric appropriately. The port associated with the specified
4353 * QP must also be a member of the multicast group.
4355 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4358 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4359 * @qp: QP to detach from the multicast group.
4360 * @gid: Multicast group GID.
4361 * @lid: Multicast group LID in host byte order.
4363 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4366 * ib_alloc_xrcd - Allocates an XRC domain.
4367 * @device: The device on which to allocate the XRC domain.
4368 * @caller: Module name for kernel consumers
4370 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
4371 #define ib_alloc_xrcd(device) \
4372 __ib_alloc_xrcd((device), KBUILD_MODNAME)
4375 * ib_dealloc_xrcd - Deallocates an XRC domain.
4376 * @xrcd: The XRC domain to deallocate.
4377 * @udata: Valid user data or NULL for kernel object
4379 int ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
4381 static inline int ib_check_mr_access(int flags)
4384 * Local write permission is required if remote write or
4385 * remote atomic permission is also requested.
4387 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4388 !(flags & IB_ACCESS_LOCAL_WRITE))
4391 if (flags & ~IB_ACCESS_SUPPORTED)
4397 static inline bool ib_access_writable(int access_flags)
4400 * We have writable memory backing the MR if any of the following
4401 * access flags are set. "Local write" and "remote write" obviously
4402 * require write access. "Remote atomic" can do things like fetch and
4403 * add, which will modify memory, and "MW bind" can change permissions
4404 * by binding a window.
4406 return access_flags &
4407 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4408 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4412 * ib_check_mr_status: lightweight check of MR status.
4413 * This routine may provide status checks on a selected
4414 * ib_mr. first use is for signature status check.
4416 * @mr: A memory region.
4417 * @check_mask: Bitmask of which checks to perform from
4418 * ib_mr_status_check enumeration.
4419 * @mr_status: The container of relevant status checks.
4420 * failed checks will be indicated in the status bitmask
4421 * and the relevant info shall be in the error item.
4423 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4424 struct ib_mr_status *mr_status);
4427 * ib_device_try_get: Hold a registration lock
4428 * device: The device to lock
4430 * A device under an active registration lock cannot become unregistered. It
4431 * is only possible to obtain a registration lock on a device that is fully
4432 * registered, otherwise this function returns false.
4434 * The registration lock is only necessary for actions which require the
4435 * device to still be registered. Uses that only require the device pointer to
4436 * be valid should use get_device(&ibdev->dev) to hold the memory.
4439 static inline bool ib_device_try_get(struct ib_device *dev)
4441 return refcount_inc_not_zero(&dev->refcount);
4444 void ib_device_put(struct ib_device *device);
4445 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4446 enum rdma_driver_id driver_id);
4447 struct ib_device *ib_device_get_by_name(const char *name,
4448 enum rdma_driver_id driver_id);
4449 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4450 u16 pkey, const union ib_gid *gid,
4451 const struct sockaddr *addr);
4452 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4454 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4456 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4457 struct ib_wq_init_attr *init_attr);
4458 int ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
4459 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4461 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
4462 struct ib_rwq_ind_table_init_attr*
4463 wq_ind_table_init_attr);
4464 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
4466 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4467 unsigned int *sg_offset, unsigned int page_size);
4468 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4469 int data_sg_nents, unsigned int *data_sg_offset,
4470 struct scatterlist *meta_sg, int meta_sg_nents,
4471 unsigned int *meta_sg_offset, unsigned int page_size);
4474 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4475 unsigned int *sg_offset, unsigned int page_size)
4479 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4485 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4486 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4488 void ib_drain_rq(struct ib_qp *qp);
4489 void ib_drain_sq(struct ib_qp *qp);
4490 void ib_drain_qp(struct ib_qp *qp);
4492 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
4494 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4496 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4497 return attr->roce.dmac;
4501 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4503 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4504 attr->ib.dlid = (u16)dlid;
4505 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4506 attr->opa.dlid = dlid;
4509 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4511 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4512 return attr->ib.dlid;
4513 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4514 return attr->opa.dlid;
4518 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4523 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4528 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4531 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4532 attr->ib.src_path_bits = src_path_bits;
4533 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4534 attr->opa.src_path_bits = src_path_bits;
4537 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4539 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4540 return attr->ib.src_path_bits;
4541 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4542 return attr->opa.src_path_bits;
4546 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4549 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4550 attr->opa.make_grd = make_grd;
4553 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4555 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4556 return attr->opa.make_grd;
4560 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4562 attr->port_num = port_num;
4565 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4567 return attr->port_num;
4570 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4573 attr->static_rate = static_rate;
4576 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4578 return attr->static_rate;
4581 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4582 enum ib_ah_flags flag)
4584 attr->ah_flags = flag;
4587 static inline enum ib_ah_flags
4588 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4590 return attr->ah_flags;
4593 static inline const struct ib_global_route
4594 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4599 /*To retrieve and modify the grh */
4600 static inline struct ib_global_route
4601 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4606 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4608 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4610 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4613 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4616 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4618 grh->dgid.global.subnet_prefix = prefix;
4621 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4624 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4626 grh->dgid.global.interface_id = if_id;
4629 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4630 union ib_gid *dgid, u32 flow_label,
4631 u8 sgid_index, u8 hop_limit,
4634 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4636 attr->ah_flags = IB_AH_GRH;
4639 grh->flow_label = flow_label;
4640 grh->sgid_index = sgid_index;
4641 grh->hop_limit = hop_limit;
4642 grh->traffic_class = traffic_class;
4643 grh->sgid_attr = NULL;
4646 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4647 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4648 u32 flow_label, u8 hop_limit, u8 traffic_class,
4649 const struct ib_gid_attr *sgid_attr);
4650 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4651 const struct rdma_ah_attr *src);
4652 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4653 const struct rdma_ah_attr *new);
4654 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4657 * rdma_ah_find_type - Return address handle type.
4659 * @dev: Device to be checked
4660 * @port_num: Port number
4662 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4665 if (rdma_protocol_roce(dev, port_num))
4666 return RDMA_AH_ATTR_TYPE_ROCE;
4667 if (rdma_protocol_ib(dev, port_num)) {
4668 if (rdma_cap_opa_ah(dev, port_num))
4669 return RDMA_AH_ATTR_TYPE_OPA;
4670 return RDMA_AH_ATTR_TYPE_IB;
4673 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4677 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4678 * In the current implementation the only way to get
4679 * get the 32bit lid is from other sources for OPA.
4680 * For IB, lids will always be 16bits so cast the
4681 * value accordingly.
4685 static inline u16 ib_lid_cpu16(u32 lid)
4687 WARN_ON_ONCE(lid & 0xFFFF0000);
4692 * ib_lid_be16 - Return lid in 16bit BE encoding.
4696 static inline __be16 ib_lid_be16(u32 lid)
4698 WARN_ON_ONCE(lid & 0xFFFF0000);
4699 return cpu_to_be16((u16)lid);
4703 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4705 * @device: the rdma device
4706 * @comp_vector: index of completion vector
4708 * Returns NULL on failure, otherwise a corresponding cpu map of the
4709 * completion vector (returns all-cpus map if the device driver doesn't
4710 * implement get_vector_affinity).
4712 static inline const struct cpumask *
4713 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4715 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4716 !device->ops.get_vector_affinity)
4719 return device->ops.get_vector_affinity(device, comp_vector);
4724 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4725 * and add their gids, as needed, to the relevant RoCE devices.
4727 * @device: the rdma device
4729 void rdma_roce_rescan_device(struct ib_device *ibdev);
4731 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4733 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4735 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4736 enum rdma_netdev_t type, const char *name,
4737 unsigned char name_assign_type,
4738 void (*setup)(struct net_device *));
4740 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4741 enum rdma_netdev_t type, const char *name,
4742 unsigned char name_assign_type,
4743 void (*setup)(struct net_device *),
4744 struct net_device *netdev);
4747 * rdma_set_device_sysfs_group - Set device attributes group to have
4748 * driver specific sysfs entries at
4749 * for infiniband class.
4751 * @device: device pointer for which attributes to be created
4752 * @group: Pointer to group which should be added when device
4753 * is registered with sysfs.
4754 * rdma_set_device_sysfs_group() allows existing drivers to expose one
4755 * group per device to have sysfs attributes.
4757 * NOTE: New drivers should not make use of this API; instead new device
4758 * parameter should be exposed via netlink command. This API and mechanism
4759 * exist only for existing drivers.
4762 rdma_set_device_sysfs_group(struct ib_device *dev,
4763 const struct attribute_group *group)
4765 dev->groups[1] = group;
4769 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4771 * @device: device pointer for which ib_device pointer to retrieve
4773 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4776 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4778 struct ib_core_device *coredev =
4779 container_of(device, struct ib_core_device, dev);
4781 return coredev->owner;
4785 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4786 * ib_device holder structure from device pointer.
4788 * NOTE: New drivers should not make use of this API; This API is only for
4789 * existing drivers who have exposed sysfs entries using
4790 * rdma_set_device_sysfs_group().
4792 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4793 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4795 bool rdma_dev_access_netns(const struct ib_device *device,
4796 const struct net *net);
4798 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4799 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4802 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4805 * This function will convert the 20 bit flow_label input to a valid RoCE v2
4806 * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4809 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4811 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4813 fl_low ^= fl_high >> 14;
4814 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4818 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4819 * local and remote qpn values
4821 * This function folded the multiplication results of two qpns, 24 bit each,
4822 * fields, and converts it to a 20 bit results.
4824 * This function will create symmetric flow_label value based on the local
4825 * and remote qpn values. this will allow both the requester and responder
4826 * to calculate the same flow_label for a given connection.
4828 * This helper function should be used by driver in case the upper layer
4829 * provide a zero flow_label value. This is to improve entropy of RDMA
4830 * traffic in the network.
4832 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4834 u64 v = (u64)lqpn * rqpn;
4839 return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4841 #endif /* IB_VERBS_H */