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 <uapi/rdma/ib_user_verbs.h>
65 #include <rdma/restrack.h>
66 #include <uapi/rdma/rdma_user_ioctl.h>
67 #include <uapi/rdma/ib_user_ioctl_verbs.h>
69 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
73 extern struct workqueue_struct *ib_wq;
74 extern struct workqueue_struct *ib_comp_wq;
75 extern struct workqueue_struct *ib_comp_unbound_wq;
78 void ibdev_printk(const char *level, const struct ib_device *ibdev,
79 const char *format, ...);
81 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
83 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
85 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
87 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
89 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
91 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
93 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
95 #if defined(CONFIG_DYNAMIC_DEBUG)
96 #define ibdev_dbg(__dev, format, args...) \
97 dynamic_ibdev_dbg(__dev, format, ##args)
99 #define ibdev_dbg(__dev, format, args...) \
100 ibdev_printk(KERN_DEBUG, __dev, format, ##args)
102 __printf(2, 3) __cold
104 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
110 __be64 subnet_prefix;
115 extern union ib_gid zgid;
118 /* If link layer is Ethernet, this is RoCE V1 */
120 IB_GID_TYPE_ROCE = 0,
121 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
125 #define ROCE_V2_UDP_DPORT 4791
127 struct net_device *ndev;
128 struct ib_device *device;
130 enum ib_gid_type gid_type;
135 enum rdma_node_type {
136 /* IB values map to NodeInfo:NodeType. */
146 /* set the local administered indication */
147 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
150 enum rdma_transport_type {
152 RDMA_TRANSPORT_IWARP,
153 RDMA_TRANSPORT_USNIC,
154 RDMA_TRANSPORT_USNIC_UDP
157 enum rdma_protocol_type {
161 RDMA_PROTOCOL_USNIC_UDP
164 __attribute_const__ enum rdma_transport_type
165 rdma_node_get_transport(enum rdma_node_type node_type);
167 enum rdma_network_type {
169 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
174 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
176 if (network_type == RDMA_NETWORK_IPV4 ||
177 network_type == RDMA_NETWORK_IPV6)
178 return IB_GID_TYPE_ROCE_UDP_ENCAP;
180 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
181 return IB_GID_TYPE_IB;
184 static inline enum rdma_network_type
185 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
187 if (attr->gid_type == IB_GID_TYPE_IB)
188 return RDMA_NETWORK_IB;
190 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
191 return RDMA_NETWORK_IPV4;
193 return RDMA_NETWORK_IPV6;
196 enum rdma_link_layer {
197 IB_LINK_LAYER_UNSPECIFIED,
198 IB_LINK_LAYER_INFINIBAND,
199 IB_LINK_LAYER_ETHERNET,
202 enum ib_device_cap_flags {
203 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
204 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
205 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
206 IB_DEVICE_RAW_MULTI = (1 << 3),
207 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
208 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
209 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
210 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
211 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
212 /* Not in use, former INIT_TYPE = (1 << 9),*/
213 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
214 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
215 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
216 IB_DEVICE_SRQ_RESIZE = (1 << 13),
217 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
220 * This device supports a per-device lkey or stag that can be
221 * used without performing a memory registration for the local
222 * memory. Note that ULPs should never check this flag, but
223 * instead of use the local_dma_lkey flag in the ib_pd structure,
224 * which will always contain a usable lkey.
226 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
227 /* Reserved, old SEND_W_INV = (1 << 16),*/
228 IB_DEVICE_MEM_WINDOW = (1 << 17),
230 * Devices should set IB_DEVICE_UD_IP_SUM if they support
231 * insertion of UDP and TCP checksum on outgoing UD IPoIB
232 * messages and can verify the validity of checksum for
233 * incoming messages. Setting this flag implies that the
234 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
236 IB_DEVICE_UD_IP_CSUM = (1 << 18),
237 IB_DEVICE_UD_TSO = (1 << 19),
238 IB_DEVICE_XRC = (1 << 20),
241 * This device supports the IB "base memory management extension",
242 * which includes support for fast registrations (IB_WR_REG_MR,
243 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
244 * also be set by any iWarp device which must support FRs to comply
245 * to the iWarp verbs spec. iWarp devices also support the
246 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
249 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
250 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
251 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
252 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
253 IB_DEVICE_RC_IP_CSUM = (1 << 25),
254 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
255 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
257 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
258 * support execution of WQEs that involve synchronization
259 * of I/O operations with single completion queue managed
262 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
263 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
264 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30),
265 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
266 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
267 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
268 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
269 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
270 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35),
271 /* The device supports padding incoming writes to cacheline. */
272 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
273 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37),
276 enum ib_signature_prot_cap {
277 IB_PROT_T10DIF_TYPE_1 = 1,
278 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
279 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
282 enum ib_signature_guard_cap {
283 IB_GUARD_T10DIF_CRC = 1,
284 IB_GUARD_T10DIF_CSUM = 1 << 1,
293 enum ib_odp_general_cap_bits {
294 IB_ODP_SUPPORT = 1 << 0,
295 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
298 enum ib_odp_transport_cap_bits {
299 IB_ODP_SUPPORT_SEND = 1 << 0,
300 IB_ODP_SUPPORT_RECV = 1 << 1,
301 IB_ODP_SUPPORT_WRITE = 1 << 2,
302 IB_ODP_SUPPORT_READ = 1 << 3,
303 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
304 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
308 uint64_t general_caps;
310 uint32_t rc_odp_caps;
311 uint32_t uc_odp_caps;
312 uint32_t ud_odp_caps;
313 uint32_t xrc_odp_caps;
314 } per_transport_caps;
318 /* Corresponding bit will be set if qp type from
319 * 'enum ib_qp_type' is supported, e.g.
320 * supported_qpts |= 1 << IB_QPT_UD
323 u32 max_rwq_indirection_tables;
324 u32 max_rwq_indirection_table_size;
327 enum ib_tm_cap_flags {
328 /* Support tag matching on RC transport */
329 IB_TM_CAP_RC = 1 << 0,
333 /* Max size of RNDV header */
334 u32 max_rndv_hdr_size;
335 /* Max number of entries in tag matching list */
337 /* From enum ib_tm_cap_flags */
339 /* Max number of outstanding list operations */
341 /* Max number of SGE in tag matching entry */
345 struct ib_cq_init_attr {
351 enum ib_cq_attr_mask {
352 IB_CQ_MODERATE = 1 << 0,
356 u16 max_cq_moderation_count;
357 u16 max_cq_moderation_period;
360 struct ib_dm_mr_attr {
366 struct ib_dm_alloc_attr {
372 struct ib_device_attr {
374 __be64 sys_image_guid;
382 u64 device_cap_flags;
393 int max_qp_init_rd_atom;
394 int max_ee_init_rd_atom;
395 enum ib_atomic_cap atomic_cap;
396 enum ib_atomic_cap masked_atomic_cap;
403 int max_mcast_qp_attach;
404 int max_total_mcast_qp_attach;
411 unsigned int max_fast_reg_page_list_len;
413 u8 local_ca_ack_delay;
416 struct ib_odp_caps odp_caps;
417 uint64_t timestamp_mask;
418 uint64_t hca_core_clock; /* in KHZ */
419 struct ib_rss_caps rss_caps;
421 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
422 struct ib_tm_caps tm_caps;
423 struct ib_cq_caps cq_caps;
435 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
438 case IB_MTU_256: return 256;
439 case IB_MTU_512: return 512;
440 case IB_MTU_1024: return 1024;
441 case IB_MTU_2048: return 2048;
442 case IB_MTU_4096: return 4096;
447 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
451 else if (mtu >= 2048)
453 else if (mtu >= 1024)
467 IB_PORT_ACTIVE_DEFER = 5
478 static inline int ib_width_enum_to_int(enum ib_port_width width)
481 case IB_WIDTH_1X: return 1;
482 case IB_WIDTH_2X: return 2;
483 case IB_WIDTH_4X: return 4;
484 case IB_WIDTH_8X: return 8;
485 case IB_WIDTH_12X: return 12;
501 * struct rdma_hw_stats
502 * @lock - Mutex to protect parallel write access to lifespan and values
503 * of counters, which are 64bits and not guaranteeed to be written
504 * atomicaly on 32bits systems.
505 * @timestamp - Used by the core code to track when the last update was
506 * @lifespan - Used by the core code to determine how old the counters
507 * should be before being updated again. Stored in jiffies, defaults
508 * to 10 milliseconds, drivers can override the default be specifying
509 * their own value during their allocation routine.
510 * @name - Array of pointers to static names used for the counters in
512 * @num_counters - How many hardware counters there are. If name is
513 * shorter than this number, a kernel oops will result. Driver authors
514 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
515 * in their code to prevent this.
516 * @value - Array of u64 counters that are accessed by the sysfs code and
517 * filled in by the drivers get_stats routine
519 struct rdma_hw_stats {
520 struct mutex lock; /* Protect lifespan and values[] */
521 unsigned long timestamp;
522 unsigned long lifespan;
523 const char * const *names;
528 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
530 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
532 * @names - Array of static const char *
533 * @num_counters - How many elements in array
534 * @lifespan - How many milliseconds between updates
536 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
537 const char * const *names, int num_counters,
538 unsigned long lifespan)
540 struct rdma_hw_stats *stats;
542 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
546 stats->names = names;
547 stats->num_counters = num_counters;
548 stats->lifespan = msecs_to_jiffies(lifespan);
554 /* Define bits for the various functionality this port needs to be supported by
557 /* Management 0x00000FFF */
558 #define RDMA_CORE_CAP_IB_MAD 0x00000001
559 #define RDMA_CORE_CAP_IB_SMI 0x00000002
560 #define RDMA_CORE_CAP_IB_CM 0x00000004
561 #define RDMA_CORE_CAP_IW_CM 0x00000008
562 #define RDMA_CORE_CAP_IB_SA 0x00000010
563 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
565 /* Address format 0x000FF000 */
566 #define RDMA_CORE_CAP_AF_IB 0x00001000
567 #define RDMA_CORE_CAP_ETH_AH 0x00002000
568 #define RDMA_CORE_CAP_OPA_AH 0x00004000
569 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
571 /* Protocol 0xFFF00000 */
572 #define RDMA_CORE_CAP_PROT_IB 0x00100000
573 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
574 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
575 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
576 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
577 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
579 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
580 | RDMA_CORE_CAP_PROT_ROCE \
581 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
583 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
584 | RDMA_CORE_CAP_IB_MAD \
585 | RDMA_CORE_CAP_IB_SMI \
586 | RDMA_CORE_CAP_IB_CM \
587 | RDMA_CORE_CAP_IB_SA \
588 | RDMA_CORE_CAP_AF_IB)
589 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
590 | RDMA_CORE_CAP_IB_MAD \
591 | RDMA_CORE_CAP_IB_CM \
592 | RDMA_CORE_CAP_AF_IB \
593 | RDMA_CORE_CAP_ETH_AH)
594 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
595 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
596 | RDMA_CORE_CAP_IB_MAD \
597 | RDMA_CORE_CAP_IB_CM \
598 | RDMA_CORE_CAP_AF_IB \
599 | RDMA_CORE_CAP_ETH_AH)
600 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
601 | RDMA_CORE_CAP_IW_CM)
602 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
603 | RDMA_CORE_CAP_OPA_MAD)
605 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
607 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
609 struct ib_port_attr {
611 enum ib_port_state state;
613 enum ib_mtu active_mtu;
615 unsigned int ip_gids:1;
616 /* This is the value from PortInfo CapabilityMask, defined by IBA */
635 enum ib_device_modify_flags {
636 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
637 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
640 #define IB_DEVICE_NODE_DESC_MAX 64
642 struct ib_device_modify {
644 char node_desc[IB_DEVICE_NODE_DESC_MAX];
647 enum ib_port_modify_flags {
648 IB_PORT_SHUTDOWN = 1,
649 IB_PORT_INIT_TYPE = (1<<2),
650 IB_PORT_RESET_QKEY_CNTR = (1<<3),
651 IB_PORT_OPA_MASK_CHG = (1<<4)
654 struct ib_port_modify {
655 u32 set_port_cap_mask;
656 u32 clr_port_cap_mask;
664 IB_EVENT_QP_ACCESS_ERR,
668 IB_EVENT_PATH_MIG_ERR,
669 IB_EVENT_DEVICE_FATAL,
670 IB_EVENT_PORT_ACTIVE,
673 IB_EVENT_PKEY_CHANGE,
676 IB_EVENT_SRQ_LIMIT_REACHED,
677 IB_EVENT_QP_LAST_WQE_REACHED,
678 IB_EVENT_CLIENT_REREGISTER,
683 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
686 struct ib_device *device;
694 enum ib_event_type event;
697 struct ib_event_handler {
698 struct ib_device *device;
699 void (*handler)(struct ib_event_handler *, struct ib_event *);
700 struct list_head list;
703 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
705 (_ptr)->device = _device; \
706 (_ptr)->handler = _handler; \
707 INIT_LIST_HEAD(&(_ptr)->list); \
710 struct ib_global_route {
711 const struct ib_gid_attr *sgid_attr;
720 __be32 version_tclass_flow;
728 union rdma_network_hdr {
731 /* The IB spec states that if it's IPv4, the header
732 * is located in the last 20 bytes of the header.
735 struct iphdr roce4grh;
739 #define IB_QPN_MASK 0xFFFFFF
742 IB_MULTICAST_QPN = 0xffffff
745 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
746 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
753 IB_RATE_PORT_CURRENT = 0,
754 IB_RATE_2_5_GBPS = 2,
762 IB_RATE_120_GBPS = 10,
763 IB_RATE_14_GBPS = 11,
764 IB_RATE_56_GBPS = 12,
765 IB_RATE_112_GBPS = 13,
766 IB_RATE_168_GBPS = 14,
767 IB_RATE_25_GBPS = 15,
768 IB_RATE_100_GBPS = 16,
769 IB_RATE_200_GBPS = 17,
770 IB_RATE_300_GBPS = 18,
771 IB_RATE_28_GBPS = 19,
772 IB_RATE_50_GBPS = 20,
773 IB_RATE_400_GBPS = 21,
774 IB_RATE_600_GBPS = 22,
778 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
779 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
780 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
781 * @rate: rate to convert.
783 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
786 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
787 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
788 * @rate: rate to convert.
790 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
794 * enum ib_mr_type - memory region type
795 * @IB_MR_TYPE_MEM_REG: memory region that is used for
796 * normal registration
797 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
798 * signature operations (data-integrity
800 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
801 * register any arbitrary sg lists (without
802 * the normal mr constraints - see
807 IB_MR_TYPE_SIGNATURE,
813 * IB_SIG_TYPE_NONE: Unprotected.
814 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
816 enum ib_signature_type {
822 * Signature T10-DIF block-guard types
823 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
824 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
826 enum ib_t10_dif_bg_type {
832 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
834 * @bg_type: T10-DIF block guard type (CRC|CSUM)
835 * @pi_interval: protection information interval.
836 * @bg: seed of guard computation.
837 * @app_tag: application tag of guard block
838 * @ref_tag: initial guard block reference tag.
839 * @ref_remap: Indicate wethear the reftag increments each block
840 * @app_escape: Indicate to skip block check if apptag=0xffff
841 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
842 * @apptag_check_mask: check bitmask of application tag.
844 struct ib_t10_dif_domain {
845 enum ib_t10_dif_bg_type bg_type;
853 u16 apptag_check_mask;
857 * struct ib_sig_domain - Parameters for signature domain
858 * @sig_type: specific signauture type
859 * @sig: union of all signature domain attributes that may
860 * be used to set domain layout.
862 struct ib_sig_domain {
863 enum ib_signature_type sig_type;
865 struct ib_t10_dif_domain dif;
870 * struct ib_sig_attrs - Parameters for signature handover operation
871 * @check_mask: bitmask for signature byte check (8 bytes)
872 * @mem: memory domain layout desciptor.
873 * @wire: wire domain layout desciptor.
875 struct ib_sig_attrs {
877 struct ib_sig_domain mem;
878 struct ib_sig_domain wire;
881 enum ib_sig_err_type {
888 * Signature check masks (8 bytes in total) according to the T10-PI standard:
889 * -------- -------- ------------
890 * | GUARD | APPTAG | REFTAG |
892 * -------- -------- ------------
895 IB_SIG_CHECK_GUARD = 0xc0,
896 IB_SIG_CHECK_APPTAG = 0x30,
897 IB_SIG_CHECK_REFTAG = 0x0f,
901 * struct ib_sig_err - signature error descriptor
904 enum ib_sig_err_type err_type;
911 enum ib_mr_status_check {
912 IB_MR_CHECK_SIG_STATUS = 1,
916 * struct ib_mr_status - Memory region status container
918 * @fail_status: Bitmask of MR checks status. For each
919 * failed check a corresponding status bit is set.
920 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
923 struct ib_mr_status {
925 struct ib_sig_err sig_err;
929 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
931 * @mult: multiple to convert.
933 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
935 enum rdma_ah_attr_type {
936 RDMA_AH_ATTR_TYPE_UNDEFINED,
937 RDMA_AH_ATTR_TYPE_IB,
938 RDMA_AH_ATTR_TYPE_ROCE,
939 RDMA_AH_ATTR_TYPE_OPA,
947 struct roce_ah_attr {
957 struct rdma_ah_attr {
958 struct ib_global_route grh;
963 enum rdma_ah_attr_type type;
965 struct ib_ah_attr ib;
966 struct roce_ah_attr roce;
967 struct opa_ah_attr opa;
975 IB_WC_LOC_EEC_OP_ERR,
980 IB_WC_LOC_ACCESS_ERR,
981 IB_WC_REM_INV_REQ_ERR,
982 IB_WC_REM_ACCESS_ERR,
985 IB_WC_RNR_RETRY_EXC_ERR,
986 IB_WC_LOC_RDD_VIOL_ERR,
987 IB_WC_REM_INV_RD_REQ_ERR,
990 IB_WC_INV_EEC_STATE_ERR,
992 IB_WC_RESP_TIMEOUT_ERR,
996 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
1007 IB_WC_MASKED_COMP_SWAP,
1008 IB_WC_MASKED_FETCH_ADD,
1010 * Set value of IB_WC_RECV so consumers can test if a completion is a
1011 * receive by testing (opcode & IB_WC_RECV).
1013 IB_WC_RECV = 1 << 7,
1014 IB_WC_RECV_RDMA_WITH_IMM
1019 IB_WC_WITH_IMM = (1<<1),
1020 IB_WC_WITH_INVALIDATE = (1<<2),
1021 IB_WC_IP_CSUM_OK = (1<<3),
1022 IB_WC_WITH_SMAC = (1<<4),
1023 IB_WC_WITH_VLAN = (1<<5),
1024 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
1030 struct ib_cqe *wr_cqe;
1032 enum ib_wc_status status;
1033 enum ib_wc_opcode opcode;
1039 u32 invalidate_rkey;
1047 u8 port_num; /* valid only for DR SMPs on switches */
1050 u8 network_hdr_type;
1053 enum ib_cq_notify_flags {
1054 IB_CQ_SOLICITED = 1 << 0,
1055 IB_CQ_NEXT_COMP = 1 << 1,
1056 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1057 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1066 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1068 return srq_type == IB_SRQT_XRC ||
1069 srq_type == IB_SRQT_TM;
1072 enum ib_srq_attr_mask {
1073 IB_SRQ_MAX_WR = 1 << 0,
1074 IB_SRQ_LIMIT = 1 << 1,
1077 struct ib_srq_attr {
1083 struct ib_srq_init_attr {
1084 void (*event_handler)(struct ib_event *, void *);
1086 struct ib_srq_attr attr;
1087 enum ib_srq_type srq_type;
1093 struct ib_xrcd *xrcd;
1108 u32 max_inline_data;
1111 * Maximum number of rdma_rw_ctx structures in flight at a time.
1112 * ib_create_qp() will calculate the right amount of neededed WRs
1113 * and MRs based on this.
1125 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1126 * here (and in that order) since the MAD layer uses them as
1127 * indices into a 2-entry table.
1136 IB_QPT_RAW_ETHERTYPE,
1137 IB_QPT_RAW_PACKET = 8,
1141 IB_QPT_DRIVER = 0xFF,
1142 /* Reserve a range for qp types internal to the low level driver.
1143 * These qp types will not be visible at the IB core layer, so the
1144 * IB_QPT_MAX usages should not be affected in the core layer
1146 IB_QPT_RESERVED1 = 0x1000,
1158 enum ib_qp_create_flags {
1159 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1160 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1161 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1162 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1163 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1164 IB_QP_CREATE_NETIF_QP = 1 << 5,
1165 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
1166 /* FREE = 1 << 7, */
1167 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1168 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
1169 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1170 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11,
1171 /* reserve bits 26-31 for low level drivers' internal use */
1172 IB_QP_CREATE_RESERVED_START = 1 << 26,
1173 IB_QP_CREATE_RESERVED_END = 1 << 31,
1177 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1178 * callback to destroy the passed in QP.
1181 struct ib_qp_init_attr {
1182 /* Consumer's event_handler callback must not block */
1183 void (*event_handler)(struct ib_event *, void *);
1186 struct ib_cq *send_cq;
1187 struct ib_cq *recv_cq;
1189 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1190 struct ib_qp_cap cap;
1191 enum ib_sig_type sq_sig_type;
1192 enum ib_qp_type qp_type;
1196 * Only needed for special QP types, or when using the RW API.
1199 struct ib_rwq_ind_table *rwq_ind_tbl;
1203 struct ib_qp_open_attr {
1204 void (*event_handler)(struct ib_event *, void *);
1207 enum ib_qp_type qp_type;
1210 enum ib_rnr_timeout {
1211 IB_RNR_TIMER_655_36 = 0,
1212 IB_RNR_TIMER_000_01 = 1,
1213 IB_RNR_TIMER_000_02 = 2,
1214 IB_RNR_TIMER_000_03 = 3,
1215 IB_RNR_TIMER_000_04 = 4,
1216 IB_RNR_TIMER_000_06 = 5,
1217 IB_RNR_TIMER_000_08 = 6,
1218 IB_RNR_TIMER_000_12 = 7,
1219 IB_RNR_TIMER_000_16 = 8,
1220 IB_RNR_TIMER_000_24 = 9,
1221 IB_RNR_TIMER_000_32 = 10,
1222 IB_RNR_TIMER_000_48 = 11,
1223 IB_RNR_TIMER_000_64 = 12,
1224 IB_RNR_TIMER_000_96 = 13,
1225 IB_RNR_TIMER_001_28 = 14,
1226 IB_RNR_TIMER_001_92 = 15,
1227 IB_RNR_TIMER_002_56 = 16,
1228 IB_RNR_TIMER_003_84 = 17,
1229 IB_RNR_TIMER_005_12 = 18,
1230 IB_RNR_TIMER_007_68 = 19,
1231 IB_RNR_TIMER_010_24 = 20,
1232 IB_RNR_TIMER_015_36 = 21,
1233 IB_RNR_TIMER_020_48 = 22,
1234 IB_RNR_TIMER_030_72 = 23,
1235 IB_RNR_TIMER_040_96 = 24,
1236 IB_RNR_TIMER_061_44 = 25,
1237 IB_RNR_TIMER_081_92 = 26,
1238 IB_RNR_TIMER_122_88 = 27,
1239 IB_RNR_TIMER_163_84 = 28,
1240 IB_RNR_TIMER_245_76 = 29,
1241 IB_RNR_TIMER_327_68 = 30,
1242 IB_RNR_TIMER_491_52 = 31
1245 enum ib_qp_attr_mask {
1247 IB_QP_CUR_STATE = (1<<1),
1248 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1249 IB_QP_ACCESS_FLAGS = (1<<3),
1250 IB_QP_PKEY_INDEX = (1<<4),
1251 IB_QP_PORT = (1<<5),
1252 IB_QP_QKEY = (1<<6),
1254 IB_QP_PATH_MTU = (1<<8),
1255 IB_QP_TIMEOUT = (1<<9),
1256 IB_QP_RETRY_CNT = (1<<10),
1257 IB_QP_RNR_RETRY = (1<<11),
1258 IB_QP_RQ_PSN = (1<<12),
1259 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1260 IB_QP_ALT_PATH = (1<<14),
1261 IB_QP_MIN_RNR_TIMER = (1<<15),
1262 IB_QP_SQ_PSN = (1<<16),
1263 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1264 IB_QP_PATH_MIG_STATE = (1<<18),
1265 IB_QP_CAP = (1<<19),
1266 IB_QP_DEST_QPN = (1<<20),
1267 IB_QP_RESERVED1 = (1<<21),
1268 IB_QP_RESERVED2 = (1<<22),
1269 IB_QP_RESERVED3 = (1<<23),
1270 IB_QP_RESERVED4 = (1<<24),
1271 IB_QP_RATE_LIMIT = (1<<25),
1296 enum ib_qp_state qp_state;
1297 enum ib_qp_state cur_qp_state;
1298 enum ib_mtu path_mtu;
1299 enum ib_mig_state path_mig_state;
1304 int qp_access_flags;
1305 struct ib_qp_cap cap;
1306 struct rdma_ah_attr ah_attr;
1307 struct rdma_ah_attr alt_ah_attr;
1310 u8 en_sqd_async_notify;
1313 u8 max_dest_rd_atomic;
1325 /* These are shared with userspace */
1326 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1327 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1328 IB_WR_SEND = IB_UVERBS_WR_SEND,
1329 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1330 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1331 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1332 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1333 IB_WR_LSO = IB_UVERBS_WR_TSO,
1334 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1335 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1336 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1337 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1338 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1339 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1340 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1342 /* These are kernel only and can not be issued by userspace */
1343 IB_WR_REG_MR = 0x20,
1346 /* reserve values for low level drivers' internal use.
1347 * These values will not be used at all in the ib core layer.
1349 IB_WR_RESERVED1 = 0xf0,
1361 enum ib_send_flags {
1363 IB_SEND_SIGNALED = (1<<1),
1364 IB_SEND_SOLICITED = (1<<2),
1365 IB_SEND_INLINE = (1<<3),
1366 IB_SEND_IP_CSUM = (1<<4),
1368 /* reserve bits 26-31 for low level drivers' internal use */
1369 IB_SEND_RESERVED_START = (1 << 26),
1370 IB_SEND_RESERVED_END = (1 << 31),
1380 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1384 struct ib_send_wr *next;
1387 struct ib_cqe *wr_cqe;
1389 struct ib_sge *sg_list;
1391 enum ib_wr_opcode opcode;
1395 u32 invalidate_rkey;
1400 struct ib_send_wr wr;
1405 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1407 return container_of(wr, struct ib_rdma_wr, wr);
1410 struct ib_atomic_wr {
1411 struct ib_send_wr wr;
1415 u64 compare_add_mask;
1420 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1422 return container_of(wr, struct ib_atomic_wr, wr);
1426 struct ib_send_wr wr;
1433 u16 pkey_index; /* valid for GSI only */
1434 u8 port_num; /* valid for DR SMPs on switch only */
1437 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1439 return container_of(wr, struct ib_ud_wr, wr);
1443 struct ib_send_wr wr;
1449 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1451 return container_of(wr, struct ib_reg_wr, wr);
1454 struct ib_sig_handover_wr {
1455 struct ib_send_wr wr;
1456 struct ib_sig_attrs *sig_attrs;
1457 struct ib_mr *sig_mr;
1459 struct ib_sge *prot;
1462 static inline const struct ib_sig_handover_wr *
1463 sig_handover_wr(const struct ib_send_wr *wr)
1465 return container_of(wr, struct ib_sig_handover_wr, wr);
1469 struct ib_recv_wr *next;
1472 struct ib_cqe *wr_cqe;
1474 struct ib_sge *sg_list;
1478 enum ib_access_flags {
1479 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1480 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1481 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1482 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1483 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1484 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1485 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1486 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1488 IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1)
1492 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1493 * are hidden here instead of a uapi header!
1495 enum ib_mr_rereg_flags {
1496 IB_MR_REREG_TRANS = 1,
1497 IB_MR_REREG_PD = (1<<1),
1498 IB_MR_REREG_ACCESS = (1<<2),
1499 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1502 struct ib_fmr_attr {
1510 enum rdma_remove_reason {
1512 * Userspace requested uobject deletion or initial try
1513 * to remove uobject via cleanup. Call could fail
1515 RDMA_REMOVE_DESTROY,
1516 /* Context deletion. This call should delete the actual object itself */
1518 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1519 RDMA_REMOVE_DRIVER_REMOVE,
1520 /* uobj is being cleaned-up before being committed */
1524 struct ib_rdmacg_object {
1525 #ifdef CONFIG_CGROUP_RDMA
1526 struct rdma_cgroup *cg; /* owner rdma cgroup */
1530 struct ib_ucontext {
1531 struct ib_device *device;
1532 struct ib_uverbs_file *ufile;
1534 * 'closing' can be read by the driver only during a destroy callback,
1535 * it is set when we are closing the file descriptor and indicates
1536 * that mm_sem may be locked.
1540 bool cleanup_retryable;
1542 void (*invalidate_range)(struct ib_umem_odp *umem_odp,
1543 unsigned long start, unsigned long end);
1544 struct mutex per_mm_list_lock;
1545 struct list_head per_mm_list;
1547 struct ib_rdmacg_object cg_obj;
1549 * Implementation details of the RDMA core, don't use in drivers:
1551 struct rdma_restrack_entry res;
1555 u64 user_handle; /* handle given to us by userspace */
1556 /* ufile & ucontext owning this object */
1557 struct ib_uverbs_file *ufile;
1558 /* FIXME, save memory: ufile->context == context */
1559 struct ib_ucontext *context; /* associated user context */
1560 void *object; /* containing object */
1561 struct list_head list; /* link to context's list */
1562 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1563 int id; /* index into kernel idr */
1565 atomic_t usecnt; /* protects exclusive access */
1566 struct rcu_head rcu; /* kfree_rcu() overhead */
1568 const struct uverbs_api_object *uapi_object;
1572 const void __user *inbuf;
1573 void __user *outbuf;
1581 struct ib_device *device;
1582 struct ib_uobject *uobject;
1583 atomic_t usecnt; /* count all resources */
1585 u32 unsafe_global_rkey;
1588 * Implementation details of the RDMA core, don't use in drivers:
1590 struct ib_mr *__internal_mr;
1591 struct rdma_restrack_entry res;
1595 struct ib_device *device;
1596 atomic_t usecnt; /* count all exposed resources */
1597 struct inode *inode;
1599 struct mutex tgt_qp_mutex;
1600 struct list_head tgt_qp_list;
1604 struct ib_device *device;
1606 struct ib_uobject *uobject;
1607 const struct ib_gid_attr *sgid_attr;
1608 enum rdma_ah_attr_type type;
1611 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1613 enum ib_poll_context {
1614 IB_POLL_DIRECT, /* caller context, no hw completions */
1615 IB_POLL_SOFTIRQ, /* poll from softirq context */
1616 IB_POLL_WORKQUEUE, /* poll from workqueue */
1617 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1621 struct ib_device *device;
1622 struct ib_uobject *uobject;
1623 ib_comp_handler comp_handler;
1624 void (*event_handler)(struct ib_event *, void *);
1627 atomic_t usecnt; /* count number of work queues */
1628 enum ib_poll_context poll_ctx;
1631 struct irq_poll iop;
1632 struct work_struct work;
1634 struct workqueue_struct *comp_wq;
1636 * Implementation details of the RDMA core, don't use in drivers:
1638 struct rdma_restrack_entry res;
1642 struct ib_device *device;
1644 struct ib_uobject *uobject;
1645 void (*event_handler)(struct ib_event *, void *);
1647 enum ib_srq_type srq_type;
1654 struct ib_xrcd *xrcd;
1661 enum ib_raw_packet_caps {
1662 /* Strip cvlan from incoming packet and report it in the matching work
1663 * completion is supported.
1665 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1666 /* Scatter FCS field of an incoming packet to host memory is supported.
1668 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1669 /* Checksum offloads are supported (for both send and receive). */
1670 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1671 /* When a packet is received for an RQ with no receive WQEs, the
1672 * packet processing is delayed.
1674 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1688 struct ib_device *device;
1689 struct ib_uobject *uobject;
1691 void (*event_handler)(struct ib_event *, void *);
1695 enum ib_wq_state state;
1696 enum ib_wq_type wq_type;
1701 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1702 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1703 IB_WQ_FLAGS_DELAY_DROP = 1 << 2,
1704 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1707 struct ib_wq_init_attr {
1709 enum ib_wq_type wq_type;
1713 void (*event_handler)(struct ib_event *, void *);
1714 u32 create_flags; /* Use enum ib_wq_flags */
1717 enum ib_wq_attr_mask {
1718 IB_WQ_STATE = 1 << 0,
1719 IB_WQ_CUR_STATE = 1 << 1,
1720 IB_WQ_FLAGS = 1 << 2,
1724 enum ib_wq_state wq_state;
1725 enum ib_wq_state curr_wq_state;
1726 u32 flags; /* Use enum ib_wq_flags */
1727 u32 flags_mask; /* Use enum ib_wq_flags */
1730 struct ib_rwq_ind_table {
1731 struct ib_device *device;
1732 struct ib_uobject *uobject;
1735 u32 log_ind_tbl_size;
1736 struct ib_wq **ind_tbl;
1739 struct ib_rwq_ind_table_init_attr {
1740 u32 log_ind_tbl_size;
1741 /* Each entry is a pointer to Receive Work Queue */
1742 struct ib_wq **ind_tbl;
1745 enum port_pkey_state {
1746 IB_PORT_PKEY_NOT_VALID = 0,
1747 IB_PORT_PKEY_VALID = 1,
1748 IB_PORT_PKEY_LISTED = 2,
1751 struct ib_qp_security;
1753 struct ib_port_pkey {
1754 enum port_pkey_state state;
1757 struct list_head qp_list;
1758 struct list_head to_error_list;
1759 struct ib_qp_security *sec;
1762 struct ib_ports_pkeys {
1763 struct ib_port_pkey main;
1764 struct ib_port_pkey alt;
1767 struct ib_qp_security {
1769 struct ib_device *dev;
1770 /* Hold this mutex when changing port and pkey settings. */
1772 struct ib_ports_pkeys *ports_pkeys;
1773 /* A list of all open shared QP handles. Required to enforce security
1774 * properly for all users of a shared QP.
1776 struct list_head shared_qp_list;
1779 atomic_t error_list_count;
1780 struct completion error_complete;
1781 int error_comps_pending;
1785 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1786 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1789 struct ib_device *device;
1791 struct ib_cq *send_cq;
1792 struct ib_cq *recv_cq;
1795 struct list_head rdma_mrs;
1796 struct list_head sig_mrs;
1798 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1799 struct list_head xrcd_list;
1801 /* count times opened, mcast attaches, flow attaches */
1803 struct list_head open_list;
1804 struct ib_qp *real_qp;
1805 struct ib_uobject *uobject;
1806 void (*event_handler)(struct ib_event *, void *);
1808 /* sgid_attrs associated with the AV's */
1809 const struct ib_gid_attr *av_sgid_attr;
1810 const struct ib_gid_attr *alt_path_sgid_attr;
1814 enum ib_qp_type qp_type;
1815 struct ib_rwq_ind_table *rwq_ind_tbl;
1816 struct ib_qp_security *qp_sec;
1820 * Implementation details of the RDMA core, don't use in drivers:
1822 struct rdma_restrack_entry res;
1826 struct ib_device *device;
1829 struct ib_uobject *uobject;
1834 struct ib_device *device;
1840 unsigned int page_size;
1843 struct ib_uobject *uobject; /* user */
1844 struct list_head qp_entry; /* FR */
1850 * Implementation details of the RDMA core, don't use in drivers:
1852 struct rdma_restrack_entry res;
1856 struct ib_device *device;
1858 struct ib_uobject *uobject;
1860 enum ib_mw_type type;
1864 struct ib_device *device;
1866 struct list_head list;
1871 /* Supported steering options */
1872 enum ib_flow_attr_type {
1873 /* steering according to rule specifications */
1874 IB_FLOW_ATTR_NORMAL = 0x0,
1875 /* default unicast and multicast rule -
1876 * receive all Eth traffic which isn't steered to any QP
1878 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1879 /* default multicast rule -
1880 * receive all Eth multicast traffic which isn't steered to any QP
1882 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1883 /* sniffer rule - receive all port traffic */
1884 IB_FLOW_ATTR_SNIFFER = 0x3
1887 /* Supported steering header types */
1888 enum ib_flow_spec_type {
1890 IB_FLOW_SPEC_ETH = 0x20,
1891 IB_FLOW_SPEC_IB = 0x22,
1893 IB_FLOW_SPEC_IPV4 = 0x30,
1894 IB_FLOW_SPEC_IPV6 = 0x31,
1895 IB_FLOW_SPEC_ESP = 0x34,
1897 IB_FLOW_SPEC_TCP = 0x40,
1898 IB_FLOW_SPEC_UDP = 0x41,
1899 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1900 IB_FLOW_SPEC_GRE = 0x51,
1901 IB_FLOW_SPEC_MPLS = 0x60,
1902 IB_FLOW_SPEC_INNER = 0x100,
1904 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1905 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1906 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1907 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1909 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1910 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1912 /* Flow steering rule priority is set according to it's domain.
1913 * Lower domain value means higher priority.
1915 enum ib_flow_domain {
1916 IB_FLOW_DOMAIN_USER,
1917 IB_FLOW_DOMAIN_ETHTOOL,
1920 IB_FLOW_DOMAIN_NUM /* Must be last */
1923 enum ib_flow_flags {
1924 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1925 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1926 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1929 struct ib_flow_eth_filter {
1938 struct ib_flow_spec_eth {
1941 struct ib_flow_eth_filter val;
1942 struct ib_flow_eth_filter mask;
1945 struct ib_flow_ib_filter {
1952 struct ib_flow_spec_ib {
1955 struct ib_flow_ib_filter val;
1956 struct ib_flow_ib_filter mask;
1959 /* IPv4 header flags */
1960 enum ib_ipv4_flags {
1961 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1962 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1963 last have this flag set */
1966 struct ib_flow_ipv4_filter {
1977 struct ib_flow_spec_ipv4 {
1980 struct ib_flow_ipv4_filter val;
1981 struct ib_flow_ipv4_filter mask;
1984 struct ib_flow_ipv6_filter {
1995 struct ib_flow_spec_ipv6 {
1998 struct ib_flow_ipv6_filter val;
1999 struct ib_flow_ipv6_filter mask;
2002 struct ib_flow_tcp_udp_filter {
2009 struct ib_flow_spec_tcp_udp {
2012 struct ib_flow_tcp_udp_filter val;
2013 struct ib_flow_tcp_udp_filter mask;
2016 struct ib_flow_tunnel_filter {
2021 /* ib_flow_spec_tunnel describes the Vxlan tunnel
2022 * the tunnel_id from val has the vni value
2024 struct ib_flow_spec_tunnel {
2027 struct ib_flow_tunnel_filter val;
2028 struct ib_flow_tunnel_filter mask;
2031 struct ib_flow_esp_filter {
2038 struct ib_flow_spec_esp {
2041 struct ib_flow_esp_filter val;
2042 struct ib_flow_esp_filter mask;
2045 struct ib_flow_gre_filter {
2046 __be16 c_ks_res0_ver;
2053 struct ib_flow_spec_gre {
2056 struct ib_flow_gre_filter val;
2057 struct ib_flow_gre_filter mask;
2060 struct ib_flow_mpls_filter {
2066 struct ib_flow_spec_mpls {
2069 struct ib_flow_mpls_filter val;
2070 struct ib_flow_mpls_filter mask;
2073 struct ib_flow_spec_action_tag {
2074 enum ib_flow_spec_type type;
2079 struct ib_flow_spec_action_drop {
2080 enum ib_flow_spec_type type;
2084 struct ib_flow_spec_action_handle {
2085 enum ib_flow_spec_type type;
2087 struct ib_flow_action *act;
2090 enum ib_counters_description {
2095 struct ib_flow_spec_action_count {
2096 enum ib_flow_spec_type type;
2098 struct ib_counters *counters;
2101 union ib_flow_spec {
2106 struct ib_flow_spec_eth eth;
2107 struct ib_flow_spec_ib ib;
2108 struct ib_flow_spec_ipv4 ipv4;
2109 struct ib_flow_spec_tcp_udp tcp_udp;
2110 struct ib_flow_spec_ipv6 ipv6;
2111 struct ib_flow_spec_tunnel tunnel;
2112 struct ib_flow_spec_esp esp;
2113 struct ib_flow_spec_gre gre;
2114 struct ib_flow_spec_mpls mpls;
2115 struct ib_flow_spec_action_tag flow_tag;
2116 struct ib_flow_spec_action_drop drop;
2117 struct ib_flow_spec_action_handle action;
2118 struct ib_flow_spec_action_count flow_count;
2121 struct ib_flow_attr {
2122 enum ib_flow_attr_type type;
2128 union ib_flow_spec flows[];
2133 struct ib_device *device;
2134 struct ib_uobject *uobject;
2137 enum ib_flow_action_type {
2138 IB_FLOW_ACTION_UNSPECIFIED,
2139 IB_FLOW_ACTION_ESP = 1,
2142 struct ib_flow_action_attrs_esp_keymats {
2143 enum ib_uverbs_flow_action_esp_keymat protocol;
2145 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2149 struct ib_flow_action_attrs_esp_replays {
2150 enum ib_uverbs_flow_action_esp_replay protocol;
2152 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2156 enum ib_flow_action_attrs_esp_flags {
2157 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2158 * This is done in order to share the same flags between user-space and
2159 * kernel and spare an unnecessary translation.
2163 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2164 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2167 struct ib_flow_spec_list {
2168 struct ib_flow_spec_list *next;
2169 union ib_flow_spec spec;
2172 struct ib_flow_action_attrs_esp {
2173 struct ib_flow_action_attrs_esp_keymats *keymat;
2174 struct ib_flow_action_attrs_esp_replays *replay;
2175 struct ib_flow_spec_list *encap;
2176 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2177 * Value of 0 is a valid value.
2183 /* Use enum ib_flow_action_attrs_esp_flags */
2185 u64 hard_limit_pkts;
2188 struct ib_flow_action {
2189 struct ib_device *device;
2190 struct ib_uobject *uobject;
2191 enum ib_flow_action_type type;
2198 enum ib_process_mad_flags {
2199 IB_MAD_IGNORE_MKEY = 1,
2200 IB_MAD_IGNORE_BKEY = 2,
2201 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2204 enum ib_mad_result {
2205 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2206 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2207 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2208 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2211 struct ib_port_cache {
2213 struct ib_pkey_cache *pkey;
2214 struct ib_gid_table *gid;
2216 enum ib_port_state port_state;
2221 struct ib_event_handler event_handler;
2226 struct ib_port_immutable {
2233 struct ib_port_data {
2234 struct ib_device *ib_dev;
2236 struct ib_port_immutable immutable;
2238 spinlock_t pkey_list_lock;
2239 struct list_head pkey_list;
2241 struct ib_port_cache cache;
2243 spinlock_t netdev_lock;
2244 struct net_device __rcu *netdev;
2245 struct hlist_node ndev_hash_link;
2248 /* rdma netdev type - specifies protocol type */
2249 enum rdma_netdev_t {
2250 RDMA_NETDEV_OPA_VNIC,
2255 * struct rdma_netdev - rdma netdev
2256 * For cases where netstack interfacing is required.
2258 struct rdma_netdev {
2260 struct ib_device *hca;
2264 * cleanup function must be specified.
2265 * FIXME: This is only used for OPA_VNIC and that usage should be
2268 void (*free_rdma_netdev)(struct net_device *netdev);
2270 /* control functions */
2271 void (*set_id)(struct net_device *netdev, int id);
2273 int (*send)(struct net_device *dev, struct sk_buff *skb,
2274 struct ib_ah *address, u32 dqpn);
2276 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2277 union ib_gid *gid, u16 mlid,
2278 int set_qkey, u32 qkey);
2279 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2280 union ib_gid *gid, u16 mlid);
2283 struct rdma_netdev_alloc_params {
2289 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2290 struct net_device *netdev, void *param);
2293 struct ib_counters {
2294 struct ib_device *device;
2295 struct ib_uobject *uobject;
2296 /* num of objects attached */
2300 struct ib_counters_read_attr {
2303 u32 flags; /* use enum ib_read_counters_flags */
2306 struct uverbs_attr_bundle;
2308 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2309 .size_##ib_struct = \
2310 (sizeof(struct drv_struct) + \
2311 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2312 BUILD_BUG_ON_ZERO( \
2313 !__same_type(((struct drv_struct *)NULL)->member, \
2316 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2317 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2319 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2320 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2322 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2325 * struct ib_device_ops - InfiniBand device operations
2326 * This structure defines all the InfiniBand device operations, providers will
2327 * need to define the supported operations, otherwise they will be set to null.
2329 struct ib_device_ops {
2330 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2331 const struct ib_send_wr **bad_send_wr);
2332 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2333 const struct ib_recv_wr **bad_recv_wr);
2334 void (*drain_rq)(struct ib_qp *qp);
2335 void (*drain_sq)(struct ib_qp *qp);
2336 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2337 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2338 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2339 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2340 int (*post_srq_recv)(struct ib_srq *srq,
2341 const struct ib_recv_wr *recv_wr,
2342 const struct ib_recv_wr **bad_recv_wr);
2343 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2344 u8 port_num, const struct ib_wc *in_wc,
2345 const struct ib_grh *in_grh,
2346 const struct ib_mad_hdr *in_mad, size_t in_mad_size,
2347 struct ib_mad_hdr *out_mad, size_t *out_mad_size,
2348 u16 *out_mad_pkey_index);
2349 int (*query_device)(struct ib_device *device,
2350 struct ib_device_attr *device_attr,
2351 struct ib_udata *udata);
2352 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2353 struct ib_device_modify *device_modify);
2354 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2355 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2357 int (*query_port)(struct ib_device *device, u8 port_num,
2358 struct ib_port_attr *port_attr);
2359 int (*modify_port)(struct ib_device *device, u8 port_num,
2360 int port_modify_mask,
2361 struct ib_port_modify *port_modify);
2363 * The following mandatory functions are used only at device
2364 * registration. Keep functions such as these at the end of this
2365 * structure to avoid cache line misses when accessing struct ib_device
2368 int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2369 struct ib_port_immutable *immutable);
2370 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2373 * When calling get_netdev, the HW vendor's driver should return the
2374 * net device of device @device at port @port_num or NULL if such
2375 * a net device doesn't exist. The vendor driver should call dev_hold
2376 * on this net device. The HW vendor's device driver must guarantee
2377 * that this function returns NULL before the net device has finished
2378 * NETDEV_UNREGISTER state.
2380 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2382 * rdma netdev operation
2384 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2385 * must return -EOPNOTSUPP if it doesn't support the specified type.
2387 struct net_device *(*alloc_rdma_netdev)(
2388 struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2389 const char *name, unsigned char name_assign_type,
2390 void (*setup)(struct net_device *));
2392 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2393 enum rdma_netdev_t type,
2394 struct rdma_netdev_alloc_params *params);
2396 * query_gid should be return GID value for @device, when @port_num
2397 * link layer is either IB or iWarp. It is no-op if @port_num port
2398 * is RoCE link layer.
2400 int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2403 * When calling add_gid, the HW vendor's driver should add the gid
2404 * of device of port at gid index available at @attr. Meta-info of
2405 * that gid (for example, the network device related to this gid) is
2406 * available at @attr. @context allows the HW vendor driver to store
2407 * extra information together with a GID entry. The HW vendor driver may
2408 * allocate memory to contain this information and store it in @context
2409 * when a new GID entry is written to. Params are consistent until the
2410 * next call of add_gid or delete_gid. The function should return 0 on
2411 * success or error otherwise. The function could be called
2412 * concurrently for different ports. This function is only called when
2413 * roce_gid_table is used.
2415 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2417 * When calling del_gid, the HW vendor's driver should delete the
2418 * gid of device @device at gid index gid_index of port port_num
2419 * available in @attr.
2420 * Upon the deletion of a GID entry, the HW vendor must free any
2421 * allocated memory. The caller will clear @context afterwards.
2422 * This function is only called when roce_gid_table is used.
2424 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2425 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2427 int (*alloc_ucontext)(struct ib_ucontext *context,
2428 struct ib_udata *udata);
2429 void (*dealloc_ucontext)(struct ib_ucontext *context);
2430 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2431 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2432 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2433 void (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2434 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr,
2435 u32 flags, struct ib_udata *udata);
2436 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2437 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2438 void (*destroy_ah)(struct ib_ah *ah, u32 flags);
2439 int (*create_srq)(struct ib_srq *srq,
2440 struct ib_srq_init_attr *srq_init_attr,
2441 struct ib_udata *udata);
2442 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2443 enum ib_srq_attr_mask srq_attr_mask,
2444 struct ib_udata *udata);
2445 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2446 void (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2447 struct ib_qp *(*create_qp)(struct ib_pd *pd,
2448 struct ib_qp_init_attr *qp_init_attr,
2449 struct ib_udata *udata);
2450 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2451 int qp_attr_mask, struct ib_udata *udata);
2452 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2453 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2454 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2455 struct ib_cq *(*create_cq)(struct ib_device *device,
2456 const struct ib_cq_init_attr *attr,
2457 struct ib_udata *udata);
2458 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2459 int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2460 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2461 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2462 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2463 u64 virt_addr, int mr_access_flags,
2464 struct ib_udata *udata);
2465 int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2466 u64 virt_addr, int mr_access_flags,
2467 struct ib_pd *pd, struct ib_udata *udata);
2468 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2469 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2470 u32 max_num_sg, struct ib_udata *udata);
2471 int (*advise_mr)(struct ib_pd *pd,
2472 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2473 struct ib_sge *sg_list, u32 num_sge,
2474 struct uverbs_attr_bundle *attrs);
2475 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2476 unsigned int *sg_offset);
2477 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2478 struct ib_mr_status *mr_status);
2479 struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2480 struct ib_udata *udata);
2481 int (*dealloc_mw)(struct ib_mw *mw);
2482 struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags,
2483 struct ib_fmr_attr *fmr_attr);
2484 int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len,
2486 int (*unmap_fmr)(struct list_head *fmr_list);
2487 int (*dealloc_fmr)(struct ib_fmr *fmr);
2488 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2489 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2490 struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device,
2491 struct ib_udata *udata);
2492 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2493 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2494 struct ib_flow_attr *flow_attr,
2495 int domain, struct ib_udata *udata);
2496 int (*destroy_flow)(struct ib_flow *flow_id);
2497 struct ib_flow_action *(*create_flow_action_esp)(
2498 struct ib_device *device,
2499 const struct ib_flow_action_attrs_esp *attr,
2500 struct uverbs_attr_bundle *attrs);
2501 int (*destroy_flow_action)(struct ib_flow_action *action);
2502 int (*modify_flow_action_esp)(
2503 struct ib_flow_action *action,
2504 const struct ib_flow_action_attrs_esp *attr,
2505 struct uverbs_attr_bundle *attrs);
2506 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2508 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2509 struct ifla_vf_info *ivf);
2510 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2511 struct ifla_vf_stats *stats);
2512 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2514 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2515 struct ib_wq_init_attr *init_attr,
2516 struct ib_udata *udata);
2517 int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2518 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2519 u32 wq_attr_mask, struct ib_udata *udata);
2520 struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2521 struct ib_device *device,
2522 struct ib_rwq_ind_table_init_attr *init_attr,
2523 struct ib_udata *udata);
2524 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2525 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2526 struct ib_ucontext *context,
2527 struct ib_dm_alloc_attr *attr,
2528 struct uverbs_attr_bundle *attrs);
2529 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2530 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2531 struct ib_dm_mr_attr *attr,
2532 struct uverbs_attr_bundle *attrs);
2533 struct ib_counters *(*create_counters)(
2534 struct ib_device *device, struct uverbs_attr_bundle *attrs);
2535 int (*destroy_counters)(struct ib_counters *counters);
2536 int (*read_counters)(struct ib_counters *counters,
2537 struct ib_counters_read_attr *counters_read_attr,
2538 struct uverbs_attr_bundle *attrs);
2540 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2541 * driver initialized data. The struct is kfree()'ed by the sysfs
2542 * core when the device is removed. A lifespan of -1 in the return
2543 * struct tells the core to set a default lifespan.
2545 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2548 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2549 * @index - The index in the value array we wish to have updated, or
2550 * num_counters if we want all stats updated
2552 * < 0 - Error, no counters updated
2553 * index - Updated the single counter pointed to by index
2554 * num_counters - Updated all counters (will reset the timestamp
2555 * and prevent further calls for lifespan milliseconds)
2556 * Drivers are allowed to update all counters in leiu of just the
2557 * one given in index at their option
2559 int (*get_hw_stats)(struct ib_device *device,
2560 struct rdma_hw_stats *stats, u8 port, int index);
2562 * This function is called once for each port when a ib device is
2565 int (*init_port)(struct ib_device *device, u8 port_num,
2566 struct kobject *port_sysfs);
2568 * Allows rdma drivers to add their own restrack attributes.
2570 int (*fill_res_entry)(struct sk_buff *msg,
2571 struct rdma_restrack_entry *entry);
2573 /* Device lifecycle callbacks */
2575 * Called after the device becomes registered, before clients are
2578 int (*enable_driver)(struct ib_device *dev);
2580 * This is called as part of ib_dealloc_device().
2582 void (*dealloc_driver)(struct ib_device *dev);
2584 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2585 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2586 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2587 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2590 struct ib_core_device {
2591 /* device must be the first element in structure until,
2592 * union of ib_core_device and device exists in ib_device.
2595 possible_net_t rdma_net;
2596 struct kobject *ports_kobj;
2597 struct list_head port_list;
2598 struct ib_device *owner; /* reach back to owner ib_device */
2601 struct rdma_restrack_root;
2603 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2604 struct device *dma_device;
2605 struct ib_device_ops ops;
2606 char name[IB_DEVICE_NAME_MAX];
2607 struct rcu_head rcu_head;
2609 struct list_head event_handler_list;
2610 spinlock_t event_handler_lock;
2612 struct rw_semaphore client_data_rwsem;
2613 struct xarray client_data;
2614 struct mutex unregistration_lock;
2616 struct ib_cache cache;
2618 * port_data is indexed by port number
2620 struct ib_port_data *port_data;
2622 int num_comp_vectors;
2624 struct iw_cm_verbs *iwcm;
2626 struct module *owner;
2629 struct ib_core_device coredev;
2632 /* First group for device attributes,
2633 * Second group for driver provided attributes (optional).
2634 * It is NULL terminated array.
2636 const struct attribute_group *groups[3];
2639 u64 uverbs_cmd_mask;
2640 u64 uverbs_ex_cmd_mask;
2642 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2646 /* Indicates kernel verbs support, should not be used in drivers */
2647 u16 kverbs_provider:1;
2650 struct ib_device_attr attrs;
2651 struct attribute_group *hw_stats_ag;
2652 struct rdma_hw_stats *hw_stats;
2654 #ifdef CONFIG_CGROUP_RDMA
2655 struct rdmacg_device cg_device;
2659 struct rdma_restrack_root *res;
2661 const struct uapi_definition *driver_def;
2662 enum rdma_driver_id driver_id;
2665 * Positive refcount indicates that the device is currently
2666 * registered and cannot be unregistered.
2668 refcount_t refcount;
2669 struct completion unreg_completion;
2670 struct work_struct unregistration_work;
2672 const struct rdma_link_ops *link_ops;
2674 /* Protects compat_devs xarray modifications */
2675 struct mutex compat_devs_mutex;
2676 /* Maintains compat devices for each net namespace */
2677 struct xarray compat_devs;
2682 void (*add) (struct ib_device *);
2683 void (*remove)(struct ib_device *, void *client_data);
2685 /* Returns the net_dev belonging to this ib_client and matching the
2687 * @dev: An RDMA device that the net_dev use for communication.
2688 * @port: A physical port number on the RDMA device.
2689 * @pkey: P_Key that the net_dev uses if applicable.
2690 * @gid: A GID that the net_dev uses to communicate.
2691 * @addr: An IP address the net_dev is configured with.
2692 * @client_data: The device's client data set by ib_set_client_data().
2694 * An ib_client that implements a net_dev on top of RDMA devices
2695 * (such as IP over IB) should implement this callback, allowing the
2696 * rdma_cm module to find the right net_dev for a given request.
2698 * The caller is responsible for calling dev_put on the returned
2700 struct net_device *(*get_net_dev_by_params)(
2701 struct ib_device *dev,
2704 const union ib_gid *gid,
2705 const struct sockaddr *addr,
2707 struct list_head list;
2710 /* kverbs are not required by the client */
2714 struct ib_device *_ib_alloc_device(size_t size);
2715 #define ib_alloc_device(drv_struct, member) \
2716 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2717 BUILD_BUG_ON_ZERO(offsetof( \
2718 struct drv_struct, member))), \
2719 struct drv_struct, member)
2721 void ib_dealloc_device(struct ib_device *device);
2723 void ib_get_device_fw_str(struct ib_device *device, char *str);
2725 int ib_register_device(struct ib_device *device, const char *name);
2726 void ib_unregister_device(struct ib_device *device);
2727 void ib_unregister_driver(enum rdma_driver_id driver_id);
2728 void ib_unregister_device_and_put(struct ib_device *device);
2729 void ib_unregister_device_queued(struct ib_device *ib_dev);
2731 int ib_register_client (struct ib_client *client);
2732 void ib_unregister_client(struct ib_client *client);
2735 * ib_get_client_data - Get IB client context
2736 * @device:Device to get context for
2737 * @client:Client to get context for
2739 * ib_get_client_data() returns the client context data set with
2740 * ib_set_client_data(). This can only be called while the client is
2741 * registered to the device, once the ib_client remove() callback returns this
2744 static inline void *ib_get_client_data(struct ib_device *device,
2745 struct ib_client *client)
2747 return xa_load(&device->client_data, client->client_id);
2749 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2751 void ib_set_device_ops(struct ib_device *device,
2752 const struct ib_device_ops *ops);
2754 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
2755 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2756 unsigned long pfn, unsigned long size, pgprot_t prot);
2758 static inline int rdma_user_mmap_io(struct ib_ucontext *ucontext,
2759 struct vm_area_struct *vma,
2760 unsigned long pfn, unsigned long size,
2767 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2769 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2772 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2774 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2777 static inline bool ib_is_buffer_cleared(const void __user *p,
2783 if (len > USHRT_MAX)
2786 buf = memdup_user(p, len);
2790 ret = !memchr_inv(buf, 0, len);
2795 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2799 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2803 * ib_is_destroy_retryable - Check whether the uobject destruction
2805 * @ret: The initial destruction return code
2806 * @why: remove reason
2807 * @uobj: The uobject that is destroyed
2809 * This function is a helper function that IB layer and low-level drivers
2810 * can use to consider whether the destruction of the given uobject is
2812 * It checks the original return code, if it wasn't success the destruction
2813 * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2814 * the remove reason. (i.e. why).
2815 * Must be called with the object locked for destroy.
2817 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2818 struct ib_uobject *uobj)
2820 return ret && (why == RDMA_REMOVE_DESTROY ||
2821 uobj->context->cleanup_retryable);
2825 * ib_destroy_usecnt - Called during destruction to check the usecnt
2826 * @usecnt: The usecnt atomic
2827 * @why: remove reason
2828 * @uobj: The uobject that is destroyed
2830 * Non-zero usecnts will block destruction unless destruction was triggered by
2831 * a ucontext cleanup.
2833 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2834 enum rdma_remove_reason why,
2835 struct ib_uobject *uobj)
2837 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2843 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2844 * contains all required attributes and no attributes not allowed for
2845 * the given QP state transition.
2846 * @cur_state: Current QP state
2847 * @next_state: Next QP state
2849 * @mask: Mask of supplied QP attributes
2851 * This function is a helper function that a low-level driver's
2852 * modify_qp method can use to validate the consumer's input. It
2853 * checks that cur_state and next_state are valid QP states, that a
2854 * transition from cur_state to next_state is allowed by the IB spec,
2855 * and that the attribute mask supplied is allowed for the transition.
2857 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2858 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2860 void ib_register_event_handler(struct ib_event_handler *event_handler);
2861 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2862 void ib_dispatch_event(struct ib_event *event);
2864 int ib_query_port(struct ib_device *device,
2865 u8 port_num, struct ib_port_attr *port_attr);
2867 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2871 * rdma_cap_ib_switch - Check if the device is IB switch
2872 * @device: Device to check
2874 * Device driver is responsible for setting is_switch bit on
2875 * in ib_device structure at init time.
2877 * Return: true if the device is IB switch.
2879 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2881 return device->is_switch;
2885 * rdma_start_port - Return the first valid port number for the device
2888 * @device: Device to be checked
2890 * Return start port number
2892 static inline u8 rdma_start_port(const struct ib_device *device)
2894 return rdma_cap_ib_switch(device) ? 0 : 1;
2898 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
2899 * @device - The struct ib_device * to iterate over
2900 * @iter - The unsigned int to store the port number
2902 #define rdma_for_each_port(device, iter) \
2903 for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type( \
2904 unsigned int, iter))); \
2905 iter <= rdma_end_port(device); (iter)++)
2908 * rdma_end_port - Return the last valid port number for the device
2911 * @device: Device to be checked
2913 * Return last port number
2915 static inline u8 rdma_end_port(const struct ib_device *device)
2917 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2920 static inline int rdma_is_port_valid(const struct ib_device *device,
2923 return (port >= rdma_start_port(device) &&
2924 port <= rdma_end_port(device));
2927 static inline bool rdma_is_grh_required(const struct ib_device *device,
2930 return device->port_data[port_num].immutable.core_cap_flags &
2931 RDMA_CORE_PORT_IB_GRH_REQUIRED;
2934 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2936 return device->port_data[port_num].immutable.core_cap_flags &
2937 RDMA_CORE_CAP_PROT_IB;
2940 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2942 return device->port_data[port_num].immutable.core_cap_flags &
2943 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2946 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2948 return device->port_data[port_num].immutable.core_cap_flags &
2949 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2952 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2954 return device->port_data[port_num].immutable.core_cap_flags &
2955 RDMA_CORE_CAP_PROT_ROCE;
2958 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2960 return device->port_data[port_num].immutable.core_cap_flags &
2961 RDMA_CORE_CAP_PROT_IWARP;
2964 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2966 return rdma_protocol_ib(device, port_num) ||
2967 rdma_protocol_roce(device, port_num);
2970 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2972 return device->port_data[port_num].immutable.core_cap_flags &
2973 RDMA_CORE_CAP_PROT_RAW_PACKET;
2976 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2978 return device->port_data[port_num].immutable.core_cap_flags &
2979 RDMA_CORE_CAP_PROT_USNIC;
2983 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2984 * Management Datagrams.
2985 * @device: Device to check
2986 * @port_num: Port number to check
2988 * Management Datagrams (MAD) are a required part of the InfiniBand
2989 * specification and are supported on all InfiniBand devices. A slightly
2990 * extended version are also supported on OPA interfaces.
2992 * Return: true if the port supports sending/receiving of MAD packets.
2994 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2996 return device->port_data[port_num].immutable.core_cap_flags &
2997 RDMA_CORE_CAP_IB_MAD;
3001 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3002 * Management Datagrams.
3003 * @device: Device to check
3004 * @port_num: Port number to check
3006 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3007 * datagrams with their own versions. These OPA MADs share many but not all of
3008 * the characteristics of InfiniBand MADs.
3010 * OPA MADs differ in the following ways:
3012 * 1) MADs are variable size up to 2K
3013 * IBTA defined MADs remain fixed at 256 bytes
3014 * 2) OPA SMPs must carry valid PKeys
3015 * 3) OPA SMP packets are a different format
3017 * Return: true if the port supports OPA MAD packet formats.
3019 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
3021 return device->port_data[port_num].immutable.core_cap_flags &
3022 RDMA_CORE_CAP_OPA_MAD;
3026 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3027 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3028 * @device: Device to check
3029 * @port_num: Port number to check
3031 * Each InfiniBand node is required to provide a Subnet Management Agent
3032 * that the subnet manager can access. Prior to the fabric being fully
3033 * configured by the subnet manager, the SMA is accessed via a well known
3034 * interface called the Subnet Management Interface (SMI). This interface
3035 * uses directed route packets to communicate with the SM to get around the
3036 * chicken and egg problem of the SM needing to know what's on the fabric
3037 * in order to configure the fabric, and needing to configure the fabric in
3038 * order to send packets to the devices on the fabric. These directed
3039 * route packets do not need the fabric fully configured in order to reach
3040 * their destination. The SMI is the only method allowed to send
3041 * directed route packets on an InfiniBand fabric.
3043 * Return: true if the port provides an SMI.
3045 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
3047 return device->port_data[port_num].immutable.core_cap_flags &
3048 RDMA_CORE_CAP_IB_SMI;
3052 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3053 * Communication Manager.
3054 * @device: Device to check
3055 * @port_num: Port number to check
3057 * The InfiniBand Communication Manager is one of many pre-defined General
3058 * Service Agents (GSA) that are accessed via the General Service
3059 * Interface (GSI). It's role is to facilitate establishment of connections
3060 * between nodes as well as other management related tasks for established
3063 * Return: true if the port supports an IB CM (this does not guarantee that
3064 * a CM is actually running however).
3066 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
3068 return device->port_data[port_num].immutable.core_cap_flags &
3069 RDMA_CORE_CAP_IB_CM;
3073 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3074 * Communication Manager.
3075 * @device: Device to check
3076 * @port_num: Port number to check
3078 * Similar to above, but specific to iWARP connections which have a different
3079 * managment protocol than InfiniBand.
3081 * Return: true if the port supports an iWARP CM (this does not guarantee that
3082 * a CM is actually running however).
3084 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
3086 return device->port_data[port_num].immutable.core_cap_flags &
3087 RDMA_CORE_CAP_IW_CM;
3091 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3092 * Subnet Administration.
3093 * @device: Device to check
3094 * @port_num: Port number to check
3096 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3097 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3098 * fabrics, devices should resolve routes to other hosts by contacting the
3099 * SA to query the proper route.
3101 * Return: true if the port should act as a client to the fabric Subnet
3102 * Administration interface. This does not imply that the SA service is
3105 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3107 return device->port_data[port_num].immutable.core_cap_flags &
3108 RDMA_CORE_CAP_IB_SA;
3112 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3114 * @device: Device to check
3115 * @port_num: Port number to check
3117 * InfiniBand multicast registration is more complex than normal IPv4 or
3118 * IPv6 multicast registration. Each Host Channel Adapter must register
3119 * with the Subnet Manager when it wishes to join a multicast group. It
3120 * should do so only once regardless of how many queue pairs it subscribes
3121 * to this group. And it should leave the group only after all queue pairs
3122 * attached to the group have been detached.
3124 * Return: true if the port must undertake the additional adminstrative
3125 * overhead of registering/unregistering with the SM and tracking of the
3126 * total number of queue pairs attached to the multicast group.
3128 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3130 return rdma_cap_ib_sa(device, port_num);
3134 * rdma_cap_af_ib - Check if the port of device has the capability
3135 * Native Infiniband Address.
3136 * @device: Device to check
3137 * @port_num: Port number to check
3139 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3140 * GID. RoCE uses a different mechanism, but still generates a GID via
3141 * a prescribed mechanism and port specific data.
3143 * Return: true if the port uses a GID address to identify devices on the
3146 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3148 return device->port_data[port_num].immutable.core_cap_flags &
3149 RDMA_CORE_CAP_AF_IB;
3153 * rdma_cap_eth_ah - Check if the port of device has the capability
3154 * Ethernet Address Handle.
3155 * @device: Device to check
3156 * @port_num: Port number to check
3158 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3159 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3160 * port. Normally, packet headers are generated by the sending host
3161 * adapter, but when sending connectionless datagrams, we must manually
3162 * inject the proper headers for the fabric we are communicating over.
3164 * Return: true if we are running as a RoCE port and must force the
3165 * addition of a Global Route Header built from our Ethernet Address
3166 * Handle into our header list for connectionless packets.
3168 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3170 return device->port_data[port_num].immutable.core_cap_flags &
3171 RDMA_CORE_CAP_ETH_AH;
3175 * rdma_cap_opa_ah - Check if the port of device supports
3176 * OPA Address handles
3177 * @device: Device to check
3178 * @port_num: Port number to check
3180 * Return: true if we are running on an OPA device which supports
3181 * the extended OPA addressing.
3183 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3185 return (device->port_data[port_num].immutable.core_cap_flags &
3186 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3190 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3193 * @port_num: Port number
3195 * This MAD size includes the MAD headers and MAD payload. No other headers
3198 * Return the max MAD size required by the Port. Will return 0 if the port
3199 * does not support MADs
3201 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3203 return device->port_data[port_num].immutable.max_mad_size;
3207 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3208 * @device: Device to check
3209 * @port_num: Port number to check
3211 * RoCE GID table mechanism manages the various GIDs for a device.
3213 * NOTE: if allocating the port's GID table has failed, this call will still
3214 * return true, but any RoCE GID table API will fail.
3216 * Return: true if the port uses RoCE GID table mechanism in order to manage
3219 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3222 return rdma_protocol_roce(device, port_num) &&
3223 device->ops.add_gid && device->ops.del_gid;
3227 * Check if the device supports READ W/ INVALIDATE.
3229 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3232 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3233 * has support for it yet.
3235 return rdma_protocol_iwarp(dev, port_num);
3238 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3240 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3241 struct ifla_vf_info *info);
3242 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3243 struct ifla_vf_stats *stats);
3244 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3247 int ib_query_pkey(struct ib_device *device,
3248 u8 port_num, u16 index, u16 *pkey);
3250 int ib_modify_device(struct ib_device *device,
3251 int device_modify_mask,
3252 struct ib_device_modify *device_modify);
3254 int ib_modify_port(struct ib_device *device,
3255 u8 port_num, int port_modify_mask,
3256 struct ib_port_modify *port_modify);
3258 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3259 u8 *port_num, u16 *index);
3261 int ib_find_pkey(struct ib_device *device,
3262 u8 port_num, u16 pkey, u16 *index);
3266 * Create a memory registration for all memory in the system and place
3267 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3268 * ULPs to avoid the overhead of dynamic MRs.
3270 * This flag is generally considered unsafe and must only be used in
3271 * extremly trusted environments. Every use of it will log a warning
3272 * in the kernel log.
3274 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3277 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3278 const char *caller);
3280 #define ib_alloc_pd(device, flags) \
3281 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3284 * ib_dealloc_pd_user - Deallocate kernel/user PD
3285 * @pd: The protection domain
3286 * @udata: Valid user data or NULL for kernel objects
3288 void ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3291 * ib_dealloc_pd - Deallocate kernel PD
3292 * @pd: The protection domain
3294 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3296 static inline void ib_dealloc_pd(struct ib_pd *pd)
3298 ib_dealloc_pd_user(pd, NULL);
3301 enum rdma_create_ah_flags {
3302 /* In a sleepable context */
3303 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3307 * rdma_create_ah - Creates an address handle for the given address vector.
3308 * @pd: The protection domain associated with the address handle.
3309 * @ah_attr: The attributes of the address vector.
3310 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3312 * The address handle is used to reference a local or global destination
3313 * in all UD QP post sends.
3315 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3319 * rdma_create_user_ah - Creates an address handle for the given address vector.
3320 * It resolves destination mac address for ah attribute of RoCE type.
3321 * @pd: The protection domain associated with the address handle.
3322 * @ah_attr: The attributes of the address vector.
3323 * @udata: pointer to user's input output buffer information need by
3326 * It returns 0 on success and returns appropriate error code on error.
3327 * The address handle is used to reference a local or global destination
3328 * in all UD QP post sends.
3330 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3331 struct rdma_ah_attr *ah_attr,
3332 struct ib_udata *udata);
3334 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3336 * @hdr: the L3 header to parse
3337 * @net_type: type of header to parse
3338 * @sgid: place to store source gid
3339 * @dgid: place to store destination gid
3341 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3342 enum rdma_network_type net_type,
3343 union ib_gid *sgid, union ib_gid *dgid);
3346 * ib_get_rdma_header_version - Get the header version
3347 * @hdr: the L3 header to parse
3349 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3352 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3354 * @device: Device on which the received message arrived.
3355 * @port_num: Port on which the received message arrived.
3356 * @wc: Work completion associated with the received message.
3357 * @grh: References the received global route header. This parameter is
3358 * ignored unless the work completion indicates that the GRH is valid.
3359 * @ah_attr: Returned attributes that can be used when creating an address
3360 * handle for replying to the message.
3361 * When ib_init_ah_attr_from_wc() returns success,
3362 * (a) for IB link layer it optionally contains a reference to SGID attribute
3363 * when GRH is present for IB link layer.
3364 * (b) for RoCE link layer it contains a reference to SGID attribute.
3365 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3366 * attributes which are initialized using ib_init_ah_attr_from_wc().
3369 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3370 const struct ib_wc *wc, const struct ib_grh *grh,
3371 struct rdma_ah_attr *ah_attr);
3374 * ib_create_ah_from_wc - Creates an address handle associated with the
3375 * sender of the specified work completion.
3376 * @pd: The protection domain associated with the address handle.
3377 * @wc: Work completion information associated with a received message.
3378 * @grh: References the received global route header. This parameter is
3379 * ignored unless the work completion indicates that the GRH is valid.
3380 * @port_num: The outbound port number to associate with the address.
3382 * The address handle is used to reference a local or global destination
3383 * in all UD QP post sends.
3385 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3386 const struct ib_grh *grh, u8 port_num);
3389 * rdma_modify_ah - Modifies the address vector associated with an address
3391 * @ah: The address handle to modify.
3392 * @ah_attr: The new address vector attributes to associate with the
3395 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3398 * rdma_query_ah - Queries the address vector associated with an address
3400 * @ah: The address handle to query.
3401 * @ah_attr: The address vector attributes associated with the address
3404 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3406 enum rdma_destroy_ah_flags {
3407 /* In a sleepable context */
3408 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3412 * rdma_destroy_ah_user - Destroys an address handle.
3413 * @ah: The address handle to destroy.
3414 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3415 * @udata: Valid user data or NULL for kernel objects
3417 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3420 * rdma_destroy_ah - Destroys an kernel address handle.
3421 * @ah: The address handle to destroy.
3422 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3424 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3426 static inline int rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3428 return rdma_destroy_ah_user(ah, flags, NULL);
3432 * ib_create_srq - Creates a SRQ associated with the specified protection
3434 * @pd: The protection domain associated with the SRQ.
3435 * @srq_init_attr: A list of initial attributes required to create the
3436 * SRQ. If SRQ creation succeeds, then the attributes are updated to
3437 * the actual capabilities of the created SRQ.
3439 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3440 * requested size of the SRQ, and set to the actual values allocated
3441 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
3442 * will always be at least as large as the requested values.
3444 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3445 struct ib_srq_init_attr *srq_init_attr);
3448 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3449 * @srq: The SRQ to modify.
3450 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3451 * the current values of selected SRQ attributes are returned.
3452 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3453 * are being modified.
3455 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3456 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3457 * the number of receives queued drops below the limit.
3459 int ib_modify_srq(struct ib_srq *srq,
3460 struct ib_srq_attr *srq_attr,
3461 enum ib_srq_attr_mask srq_attr_mask);
3464 * ib_query_srq - Returns the attribute list and current values for the
3466 * @srq: The SRQ to query.
3467 * @srq_attr: The attributes of the specified SRQ.
3469 int ib_query_srq(struct ib_srq *srq,
3470 struct ib_srq_attr *srq_attr);
3473 * ib_destroy_srq_user - Destroys the specified SRQ.
3474 * @srq: The SRQ to destroy.
3475 * @udata: Valid user data or NULL for kernel objects
3477 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3480 * ib_destroy_srq - Destroys the specified kernel SRQ.
3481 * @srq: The SRQ to destroy.
3483 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3485 static inline int ib_destroy_srq(struct ib_srq *srq)
3487 return ib_destroy_srq_user(srq, NULL);
3491 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3492 * @srq: The SRQ to post the work request on.
3493 * @recv_wr: A list of work requests to post on the receive queue.
3494 * @bad_recv_wr: On an immediate failure, this parameter will reference
3495 * the work request that failed to be posted on the QP.
3497 static inline int ib_post_srq_recv(struct ib_srq *srq,
3498 const struct ib_recv_wr *recv_wr,
3499 const struct ib_recv_wr **bad_recv_wr)
3501 const struct ib_recv_wr *dummy;
3503 return srq->device->ops.post_srq_recv(srq, recv_wr,
3504 bad_recv_wr ? : &dummy);
3508 * ib_create_qp_user - Creates a QP associated with the specified protection
3510 * @pd: The protection domain associated with the QP.
3511 * @qp_init_attr: A list of initial attributes required to create the
3512 * QP. If QP creation succeeds, then the attributes are updated to
3513 * the actual capabilities of the created QP.
3514 * @udata: Valid user data or NULL for kernel objects
3516 struct ib_qp *ib_create_qp_user(struct ib_pd *pd,
3517 struct ib_qp_init_attr *qp_init_attr,
3518 struct ib_udata *udata);
3521 * ib_create_qp - Creates a kernel QP associated with the specified protection
3523 * @pd: The protection domain associated with the QP.
3524 * @qp_init_attr: A list of initial attributes required to create the
3525 * QP. If QP creation succeeds, then the attributes are updated to
3526 * the actual capabilities of the created QP.
3527 * @udata: Valid user data or NULL for kernel objects
3529 * NOTE: for user qp use ib_create_qp_user with valid udata!
3531 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3532 struct ib_qp_init_attr *qp_init_attr)
3534 return ib_create_qp_user(pd, qp_init_attr, NULL);
3538 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3539 * @qp: The QP to modify.
3540 * @attr: On input, specifies the QP attributes to modify. On output,
3541 * the current values of selected QP attributes are returned.
3542 * @attr_mask: A bit-mask used to specify which attributes of the QP
3543 * are being modified.
3544 * @udata: pointer to user's input output buffer information
3545 * are being modified.
3546 * It returns 0 on success and returns appropriate error code on error.
3548 int ib_modify_qp_with_udata(struct ib_qp *qp,
3549 struct ib_qp_attr *attr,
3551 struct ib_udata *udata);
3554 * ib_modify_qp - Modifies the attributes for the specified QP and then
3555 * transitions the QP to the given state.
3556 * @qp: The QP to modify.
3557 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3558 * the current values of selected QP attributes are returned.
3559 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3560 * are being modified.
3562 int ib_modify_qp(struct ib_qp *qp,
3563 struct ib_qp_attr *qp_attr,
3567 * ib_query_qp - Returns the attribute list and current values for the
3569 * @qp: The QP to query.
3570 * @qp_attr: The attributes of the specified QP.
3571 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3572 * @qp_init_attr: Additional attributes of the selected QP.
3574 * The qp_attr_mask may be used to limit the query to gathering only the
3575 * selected attributes.
3577 int ib_query_qp(struct ib_qp *qp,
3578 struct ib_qp_attr *qp_attr,
3580 struct ib_qp_init_attr *qp_init_attr);
3583 * ib_destroy_qp - Destroys the specified QP.
3584 * @qp: The QP to destroy.
3585 * @udata: Valid udata or NULL for kernel objects
3587 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3590 * ib_destroy_qp - Destroys the specified kernel QP.
3591 * @qp: The QP to destroy.
3593 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3595 static inline int ib_destroy_qp(struct ib_qp *qp)
3597 return ib_destroy_qp_user(qp, NULL);
3601 * ib_open_qp - Obtain a reference to an existing sharable QP.
3602 * @xrcd - XRC domain
3603 * @qp_open_attr: Attributes identifying the QP to open.
3605 * Returns a reference to a sharable QP.
3607 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3608 struct ib_qp_open_attr *qp_open_attr);
3611 * ib_close_qp - Release an external reference to a QP.
3612 * @qp: The QP handle to release
3614 * The opened QP handle is released by the caller. The underlying
3615 * shared QP is not destroyed until all internal references are released.
3617 int ib_close_qp(struct ib_qp *qp);
3620 * ib_post_send - Posts a list of work requests to the send queue of
3622 * @qp: The QP to post the work request on.
3623 * @send_wr: A list of work requests to post on the send queue.
3624 * @bad_send_wr: On an immediate failure, this parameter will reference
3625 * the work request that failed to be posted on the QP.
3627 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3628 * error is returned, the QP state shall not be affected,
3629 * ib_post_send() will return an immediate error after queueing any
3630 * earlier work requests in the list.
3632 static inline int ib_post_send(struct ib_qp *qp,
3633 const struct ib_send_wr *send_wr,
3634 const struct ib_send_wr **bad_send_wr)
3636 const struct ib_send_wr *dummy;
3638 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3642 * ib_post_recv - Posts a list of work requests to the receive queue of
3644 * @qp: The QP to post the work request on.
3645 * @recv_wr: A list of work requests to post on the receive queue.
3646 * @bad_recv_wr: On an immediate failure, this parameter will reference
3647 * the work request that failed to be posted on the QP.
3649 static inline int ib_post_recv(struct ib_qp *qp,
3650 const struct ib_recv_wr *recv_wr,
3651 const struct ib_recv_wr **bad_recv_wr)
3653 const struct ib_recv_wr *dummy;
3655 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3658 struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private,
3659 int nr_cqe, int comp_vector,
3660 enum ib_poll_context poll_ctx,
3661 const char *caller, struct ib_udata *udata);
3664 * ib_alloc_cq_user: Allocate kernel/user CQ
3665 * @dev: The IB device
3666 * @private: Private data attached to the CQE
3667 * @nr_cqe: Number of CQEs in the CQ
3668 * @comp_vector: Completion vector used for the IRQs
3669 * @poll_ctx: Context used for polling the CQ
3670 * @udata: Valid user data or NULL for kernel objects
3672 static inline struct ib_cq *ib_alloc_cq_user(struct ib_device *dev,
3673 void *private, int nr_cqe,
3675 enum ib_poll_context poll_ctx,
3676 struct ib_udata *udata)
3678 return __ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3679 KBUILD_MODNAME, udata);
3683 * ib_alloc_cq: Allocate kernel CQ
3684 * @dev: The IB device
3685 * @private: Private data attached to the CQE
3686 * @nr_cqe: Number of CQEs in the CQ
3687 * @comp_vector: Completion vector used for the IRQs
3688 * @poll_ctx: Context used for polling the CQ
3690 * NOTE: for user cq use ib_alloc_cq_user with valid udata!
3692 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3693 int nr_cqe, int comp_vector,
3694 enum ib_poll_context poll_ctx)
3696 return ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3701 * ib_free_cq_user - Free kernel/user CQ
3702 * @cq: The CQ to free
3703 * @udata: Valid user data or NULL for kernel objects
3705 void ib_free_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3708 * ib_free_cq - Free kernel CQ
3709 * @cq: The CQ to free
3711 * NOTE: for user cq use ib_free_cq_user with valid udata!
3713 static inline void ib_free_cq(struct ib_cq *cq)
3715 ib_free_cq_user(cq, NULL);
3718 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3721 * ib_create_cq - Creates a CQ on the specified device.
3722 * @device: The device on which to create the CQ.
3723 * @comp_handler: A user-specified callback that is invoked when a
3724 * completion event occurs on the CQ.
3725 * @event_handler: A user-specified callback that is invoked when an
3726 * asynchronous event not associated with a completion occurs on the CQ.
3727 * @cq_context: Context associated with the CQ returned to the user via
3728 * the associated completion and event handlers.
3729 * @cq_attr: The attributes the CQ should be created upon.
3731 * Users can examine the cq structure to determine the actual CQ size.
3733 struct ib_cq *__ib_create_cq(struct ib_device *device,
3734 ib_comp_handler comp_handler,
3735 void (*event_handler)(struct ib_event *, void *),
3737 const struct ib_cq_init_attr *cq_attr,
3738 const char *caller);
3739 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3740 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3743 * ib_resize_cq - Modifies the capacity of the CQ.
3744 * @cq: The CQ to resize.
3745 * @cqe: The minimum size of the CQ.
3747 * Users can examine the cq structure to determine the actual CQ size.
3749 int ib_resize_cq(struct ib_cq *cq, int cqe);
3752 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3753 * @cq: The CQ to modify.
3754 * @cq_count: number of CQEs that will trigger an event
3755 * @cq_period: max period of time in usec before triggering an event
3758 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3761 * ib_destroy_cq_user - Destroys the specified CQ.
3762 * @cq: The CQ to destroy.
3763 * @udata: Valid user data or NULL for kernel objects
3765 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3768 * ib_destroy_cq - Destroys the specified kernel CQ.
3769 * @cq: The CQ to destroy.
3771 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3773 static inline int ib_destroy_cq(struct ib_cq *cq)
3775 return ib_destroy_cq_user(cq, NULL);
3779 * ib_poll_cq - poll a CQ for completion(s)
3780 * @cq:the CQ being polled
3781 * @num_entries:maximum number of completions to return
3782 * @wc:array of at least @num_entries &struct ib_wc where completions
3785 * Poll a CQ for (possibly multiple) completions. If the return value
3786 * is < 0, an error occurred. If the return value is >= 0, it is the
3787 * number of completions returned. If the return value is
3788 * non-negative and < num_entries, then the CQ was emptied.
3790 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3793 return cq->device->ops.poll_cq(cq, num_entries, wc);
3797 * ib_req_notify_cq - Request completion notification on a CQ.
3798 * @cq: The CQ to generate an event for.
3800 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3801 * to request an event on the next solicited event or next work
3802 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3803 * may also be |ed in to request a hint about missed events, as
3807 * < 0 means an error occurred while requesting notification
3808 * == 0 means notification was requested successfully, and if
3809 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3810 * were missed and it is safe to wait for another event. In
3811 * this case is it guaranteed that any work completions added
3812 * to the CQ since the last CQ poll will trigger a completion
3813 * notification event.
3814 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3815 * in. It means that the consumer must poll the CQ again to
3816 * make sure it is empty to avoid missing an event because of a
3817 * race between requesting notification and an entry being
3818 * added to the CQ. This return value means it is possible
3819 * (but not guaranteed) that a work completion has been added
3820 * to the CQ since the last poll without triggering a
3821 * completion notification event.
3823 static inline int ib_req_notify_cq(struct ib_cq *cq,
3824 enum ib_cq_notify_flags flags)
3826 return cq->device->ops.req_notify_cq(cq, flags);
3830 * ib_req_ncomp_notif - Request completion notification when there are
3831 * at least the specified number of unreaped completions on the CQ.
3832 * @cq: The CQ to generate an event for.
3833 * @wc_cnt: The number of unreaped completions that should be on the
3834 * CQ before an event is generated.
3836 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3838 return cq->device->ops.req_ncomp_notif ?
3839 cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
3844 * ib_dma_mapping_error - check a DMA addr for error
3845 * @dev: The device for which the dma_addr was created
3846 * @dma_addr: The DMA address to check
3848 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3850 return dma_mapping_error(dev->dma_device, dma_addr);
3854 * ib_dma_map_single - Map a kernel virtual address to DMA address
3855 * @dev: The device for which the dma_addr is to be created
3856 * @cpu_addr: The kernel virtual address
3857 * @size: The size of the region in bytes
3858 * @direction: The direction of the DMA
3860 static inline u64 ib_dma_map_single(struct ib_device *dev,
3861 void *cpu_addr, size_t size,
3862 enum dma_data_direction direction)
3864 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3868 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3869 * @dev: The device for which the DMA address was created
3870 * @addr: The DMA address
3871 * @size: The size of the region in bytes
3872 * @direction: The direction of the DMA
3874 static inline void ib_dma_unmap_single(struct ib_device *dev,
3875 u64 addr, size_t size,
3876 enum dma_data_direction direction)
3878 dma_unmap_single(dev->dma_device, addr, size, direction);
3882 * ib_dma_map_page - Map a physical page to DMA address
3883 * @dev: The device for which the dma_addr is to be created
3884 * @page: The page to be mapped
3885 * @offset: The offset within the page
3886 * @size: The size of the region in bytes
3887 * @direction: The direction of the DMA
3889 static inline u64 ib_dma_map_page(struct ib_device *dev,
3891 unsigned long offset,
3893 enum dma_data_direction direction)
3895 return dma_map_page(dev->dma_device, page, offset, size, direction);
3899 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3900 * @dev: The device for which the DMA address was created
3901 * @addr: The DMA address
3902 * @size: The size of the region in bytes
3903 * @direction: The direction of the DMA
3905 static inline void ib_dma_unmap_page(struct ib_device *dev,
3906 u64 addr, size_t size,
3907 enum dma_data_direction direction)
3909 dma_unmap_page(dev->dma_device, addr, size, direction);
3913 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3914 * @dev: The device for which the DMA addresses are to be created
3915 * @sg: The array of scatter/gather entries
3916 * @nents: The number of scatter/gather entries
3917 * @direction: The direction of the DMA
3919 static inline int ib_dma_map_sg(struct ib_device *dev,
3920 struct scatterlist *sg, int nents,
3921 enum dma_data_direction direction)
3923 return dma_map_sg(dev->dma_device, sg, nents, direction);
3927 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3928 * @dev: The device for which the DMA addresses were created
3929 * @sg: The array of scatter/gather entries
3930 * @nents: The number of scatter/gather entries
3931 * @direction: The direction of the DMA
3933 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3934 struct scatterlist *sg, int nents,
3935 enum dma_data_direction direction)
3937 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3940 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3941 struct scatterlist *sg, int nents,
3942 enum dma_data_direction direction,
3943 unsigned long dma_attrs)
3945 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3949 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3950 struct scatterlist *sg, int nents,
3951 enum dma_data_direction direction,
3952 unsigned long dma_attrs)
3954 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3958 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
3959 * @dev: The device to query
3961 * The returned value represents a size in bytes.
3963 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
3965 struct device_dma_parameters *p = dev->dma_device->dma_parms;
3967 return p ? p->max_segment_size : UINT_MAX;
3971 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3972 * @dev: The device for which the DMA address was created
3973 * @addr: The DMA address
3974 * @size: The size of the region in bytes
3975 * @dir: The direction of the DMA
3977 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3980 enum dma_data_direction dir)
3982 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3986 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3987 * @dev: The device for which the DMA address was created
3988 * @addr: The DMA address
3989 * @size: The size of the region in bytes
3990 * @dir: The direction of the DMA
3992 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3995 enum dma_data_direction dir)
3997 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4001 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
4002 * @dev: The device for which the DMA address is requested
4003 * @size: The size of the region to allocate in bytes
4004 * @dma_handle: A pointer for returning the DMA address of the region
4005 * @flag: memory allocator flags
4007 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
4009 dma_addr_t *dma_handle,
4012 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
4016 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
4017 * @dev: The device for which the DMA addresses were allocated
4018 * @size: The size of the region
4019 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
4020 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
4022 static inline void ib_dma_free_coherent(struct ib_device *dev,
4023 size_t size, void *cpu_addr,
4024 dma_addr_t dma_handle)
4026 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
4030 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4031 * HCA translation table.
4032 * @mr: The memory region to deregister.
4033 * @udata: Valid user data or NULL for kernel object
4035 * This function can fail, if the memory region has memory windows bound to it.
4037 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4040 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4041 * HCA translation table.
4042 * @mr: The memory region to deregister.
4044 * This function can fail, if the memory region has memory windows bound to it.
4046 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4048 static inline int ib_dereg_mr(struct ib_mr *mr)
4050 return ib_dereg_mr_user(mr, NULL);
4053 struct ib_mr *ib_alloc_mr_user(struct ib_pd *pd, enum ib_mr_type mr_type,
4054 u32 max_num_sg, struct ib_udata *udata);
4056 static inline struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
4057 enum ib_mr_type mr_type, u32 max_num_sg)
4059 return ib_alloc_mr_user(pd, mr_type, max_num_sg, NULL);
4063 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4065 * @mr - struct ib_mr pointer to be updated.
4066 * @newkey - new key to be used.
4068 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4070 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4071 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4075 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4076 * for calculating a new rkey for type 2 memory windows.
4077 * @rkey - the rkey to increment.
4079 static inline u32 ib_inc_rkey(u32 rkey)
4081 const u32 mask = 0x000000ff;
4082 return ((rkey + 1) & mask) | (rkey & ~mask);
4086 * ib_alloc_fmr - Allocates a unmapped fast memory region.
4087 * @pd: The protection domain associated with the unmapped region.
4088 * @mr_access_flags: Specifies the memory access rights.
4089 * @fmr_attr: Attributes of the unmapped region.
4091 * A fast memory region must be mapped before it can be used as part of
4094 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
4095 int mr_access_flags,
4096 struct ib_fmr_attr *fmr_attr);
4099 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
4100 * @fmr: The fast memory region to associate with the pages.
4101 * @page_list: An array of physical pages to map to the fast memory region.
4102 * @list_len: The number of pages in page_list.
4103 * @iova: The I/O virtual address to use with the mapped region.
4105 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
4106 u64 *page_list, int list_len,
4109 return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova);
4113 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
4114 * @fmr_list: A linked list of fast memory regions to unmap.
4116 int ib_unmap_fmr(struct list_head *fmr_list);
4119 * ib_dealloc_fmr - Deallocates a fast memory region.
4120 * @fmr: The fast memory region to deallocate.
4122 int ib_dealloc_fmr(struct ib_fmr *fmr);
4125 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4126 * @qp: QP to attach to the multicast group. The QP must be type
4128 * @gid: Multicast group GID.
4129 * @lid: Multicast group LID in host byte order.
4131 * In order to send and receive multicast packets, subnet
4132 * administration must have created the multicast group and configured
4133 * the fabric appropriately. The port associated with the specified
4134 * QP must also be a member of the multicast group.
4136 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4139 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4140 * @qp: QP to detach from the multicast group.
4141 * @gid: Multicast group GID.
4142 * @lid: Multicast group LID in host byte order.
4144 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4147 * ib_alloc_xrcd - Allocates an XRC domain.
4148 * @device: The device on which to allocate the XRC domain.
4149 * @caller: Module name for kernel consumers
4151 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
4152 #define ib_alloc_xrcd(device) \
4153 __ib_alloc_xrcd((device), KBUILD_MODNAME)
4156 * ib_dealloc_xrcd - Deallocates an XRC domain.
4157 * @xrcd: The XRC domain to deallocate.
4158 * @udata: Valid user data or NULL for kernel object
4160 int ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
4162 static inline int ib_check_mr_access(int flags)
4165 * Local write permission is required if remote write or
4166 * remote atomic permission is also requested.
4168 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4169 !(flags & IB_ACCESS_LOCAL_WRITE))
4175 static inline bool ib_access_writable(int access_flags)
4178 * We have writable memory backing the MR if any of the following
4179 * access flags are set. "Local write" and "remote write" obviously
4180 * require write access. "Remote atomic" can do things like fetch and
4181 * add, which will modify memory, and "MW bind" can change permissions
4182 * by binding a window.
4184 return access_flags &
4185 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4186 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4190 * ib_check_mr_status: lightweight check of MR status.
4191 * This routine may provide status checks on a selected
4192 * ib_mr. first use is for signature status check.
4194 * @mr: A memory region.
4195 * @check_mask: Bitmask of which checks to perform from
4196 * ib_mr_status_check enumeration.
4197 * @mr_status: The container of relevant status checks.
4198 * failed checks will be indicated in the status bitmask
4199 * and the relevant info shall be in the error item.
4201 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4202 struct ib_mr_status *mr_status);
4205 * ib_device_try_get: Hold a registration lock
4206 * device: The device to lock
4208 * A device under an active registration lock cannot become unregistered. It
4209 * is only possible to obtain a registration lock on a device that is fully
4210 * registered, otherwise this function returns false.
4212 * The registration lock is only necessary for actions which require the
4213 * device to still be registered. Uses that only require the device pointer to
4214 * be valid should use get_device(&ibdev->dev) to hold the memory.
4217 static inline bool ib_device_try_get(struct ib_device *dev)
4219 return refcount_inc_not_zero(&dev->refcount);
4222 void ib_device_put(struct ib_device *device);
4223 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4224 enum rdma_driver_id driver_id);
4225 struct ib_device *ib_device_get_by_name(const char *name,
4226 enum rdma_driver_id driver_id);
4227 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4228 u16 pkey, const union ib_gid *gid,
4229 const struct sockaddr *addr);
4230 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4232 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4234 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4235 struct ib_wq_init_attr *init_attr);
4236 int ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
4237 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4239 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
4240 struct ib_rwq_ind_table_init_attr*
4241 wq_ind_table_init_attr);
4242 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
4244 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4245 unsigned int *sg_offset, unsigned int page_size);
4248 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4249 unsigned int *sg_offset, unsigned int page_size)
4253 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4259 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4260 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4262 void ib_drain_rq(struct ib_qp *qp);
4263 void ib_drain_sq(struct ib_qp *qp);
4264 void ib_drain_qp(struct ib_qp *qp);
4266 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
4268 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4270 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4271 return attr->roce.dmac;
4275 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4277 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4278 attr->ib.dlid = (u16)dlid;
4279 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4280 attr->opa.dlid = dlid;
4283 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4285 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4286 return attr->ib.dlid;
4287 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4288 return attr->opa.dlid;
4292 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4297 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4302 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4305 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4306 attr->ib.src_path_bits = src_path_bits;
4307 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4308 attr->opa.src_path_bits = src_path_bits;
4311 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4313 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4314 return attr->ib.src_path_bits;
4315 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4316 return attr->opa.src_path_bits;
4320 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4323 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4324 attr->opa.make_grd = make_grd;
4327 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4329 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4330 return attr->opa.make_grd;
4334 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4336 attr->port_num = port_num;
4339 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4341 return attr->port_num;
4344 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4347 attr->static_rate = static_rate;
4350 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4352 return attr->static_rate;
4355 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4356 enum ib_ah_flags flag)
4358 attr->ah_flags = flag;
4361 static inline enum ib_ah_flags
4362 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4364 return attr->ah_flags;
4367 static inline const struct ib_global_route
4368 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4373 /*To retrieve and modify the grh */
4374 static inline struct ib_global_route
4375 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4380 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4382 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4384 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4387 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4390 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4392 grh->dgid.global.subnet_prefix = prefix;
4395 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4398 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4400 grh->dgid.global.interface_id = if_id;
4403 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4404 union ib_gid *dgid, u32 flow_label,
4405 u8 sgid_index, u8 hop_limit,
4408 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4410 attr->ah_flags = IB_AH_GRH;
4413 grh->flow_label = flow_label;
4414 grh->sgid_index = sgid_index;
4415 grh->hop_limit = hop_limit;
4416 grh->traffic_class = traffic_class;
4417 grh->sgid_attr = NULL;
4420 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4421 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4422 u32 flow_label, u8 hop_limit, u8 traffic_class,
4423 const struct ib_gid_attr *sgid_attr);
4424 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4425 const struct rdma_ah_attr *src);
4426 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4427 const struct rdma_ah_attr *new);
4428 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4431 * rdma_ah_find_type - Return address handle type.
4433 * @dev: Device to be checked
4434 * @port_num: Port number
4436 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4439 if (rdma_protocol_roce(dev, port_num))
4440 return RDMA_AH_ATTR_TYPE_ROCE;
4441 if (rdma_protocol_ib(dev, port_num)) {
4442 if (rdma_cap_opa_ah(dev, port_num))
4443 return RDMA_AH_ATTR_TYPE_OPA;
4444 return RDMA_AH_ATTR_TYPE_IB;
4447 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4451 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4452 * In the current implementation the only way to get
4453 * get the 32bit lid is from other sources for OPA.
4454 * For IB, lids will always be 16bits so cast the
4455 * value accordingly.
4459 static inline u16 ib_lid_cpu16(u32 lid)
4461 WARN_ON_ONCE(lid & 0xFFFF0000);
4466 * ib_lid_be16 - Return lid in 16bit BE encoding.
4470 static inline __be16 ib_lid_be16(u32 lid)
4472 WARN_ON_ONCE(lid & 0xFFFF0000);
4473 return cpu_to_be16((u16)lid);
4477 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4479 * @device: the rdma device
4480 * @comp_vector: index of completion vector
4482 * Returns NULL on failure, otherwise a corresponding cpu map of the
4483 * completion vector (returns all-cpus map if the device driver doesn't
4484 * implement get_vector_affinity).
4486 static inline const struct cpumask *
4487 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4489 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4490 !device->ops.get_vector_affinity)
4493 return device->ops.get_vector_affinity(device, comp_vector);
4498 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4499 * and add their gids, as needed, to the relevant RoCE devices.
4501 * @device: the rdma device
4503 void rdma_roce_rescan_device(struct ib_device *ibdev);
4505 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4507 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4509 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4510 enum rdma_netdev_t type, const char *name,
4511 unsigned char name_assign_type,
4512 void (*setup)(struct net_device *));
4514 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4515 enum rdma_netdev_t type, const char *name,
4516 unsigned char name_assign_type,
4517 void (*setup)(struct net_device *),
4518 struct net_device *netdev);
4521 * rdma_set_device_sysfs_group - Set device attributes group to have
4522 * driver specific sysfs entries at
4523 * for infiniband class.
4525 * @device: device pointer for which attributes to be created
4526 * @group: Pointer to group which should be added when device
4527 * is registered with sysfs.
4528 * rdma_set_device_sysfs_group() allows existing drivers to expose one
4529 * group per device to have sysfs attributes.
4531 * NOTE: New drivers should not make use of this API; instead new device
4532 * parameter should be exposed via netlink command. This API and mechanism
4533 * exist only for existing drivers.
4536 rdma_set_device_sysfs_group(struct ib_device *dev,
4537 const struct attribute_group *group)
4539 dev->groups[1] = group;
4543 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4545 * @device: device pointer for which ib_device pointer to retrieve
4547 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4550 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4552 struct ib_core_device *coredev =
4553 container_of(device, struct ib_core_device, dev);
4555 return coredev->owner;
4559 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4560 * ib_device holder structure from device pointer.
4562 * NOTE: New drivers should not make use of this API; This API is only for
4563 * existing drivers who have exposed sysfs entries using
4564 * rdma_set_device_sysfs_group().
4566 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4567 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4569 bool rdma_dev_access_netns(const struct ib_device *device,
4570 const struct net *net);
4571 #endif /* IB_VERBS_H */