1 /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
3 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
4 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
5 * Copyright (c) 2004, 2020 Intel Corporation. All rights reserved.
6 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
7 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
8 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
9 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
15 #include <linux/ethtool.h>
16 #include <linux/types.h>
17 #include <linux/device.h>
18 #include <linux/dma-mapping.h>
19 #include <linux/kref.h>
20 #include <linux/list.h>
21 #include <linux/rwsem.h>
22 #include <linux/workqueue.h>
23 #include <linux/irq_poll.h>
24 #include <uapi/linux/if_ether.h>
27 #include <linux/string.h>
28 #include <linux/slab.h>
29 #include <linux/netdevice.h>
30 #include <linux/refcount.h>
31 #include <linux/if_link.h>
32 #include <linux/atomic.h>
33 #include <linux/mmu_notifier.h>
34 #include <linux/uaccess.h>
35 #include <linux/cgroup_rdma.h>
36 #include <linux/irqflags.h>
37 #include <linux/preempt.h>
38 #include <linux/dim.h>
39 #include <uapi/rdma/ib_user_verbs.h>
40 #include <rdma/rdma_counter.h>
41 #include <rdma/restrack.h>
42 #include <rdma/signature.h>
43 #include <uapi/rdma/rdma_user_ioctl.h>
44 #include <uapi/rdma/ib_user_ioctl_verbs.h>
46 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
50 struct ib_usrq_object;
54 struct hw_stats_device_data;
56 extern struct workqueue_struct *ib_wq;
57 extern struct workqueue_struct *ib_comp_wq;
58 extern struct workqueue_struct *ib_comp_unbound_wq;
63 void ibdev_printk(const char *level, const struct ib_device *ibdev,
64 const char *format, ...);
66 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
68 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
70 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
72 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
74 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
76 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
78 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
80 #if defined(CONFIG_DYNAMIC_DEBUG) || \
81 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
82 #define ibdev_dbg(__dev, format, args...) \
83 dynamic_ibdev_dbg(__dev, format, ##args)
87 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
90 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \
92 static DEFINE_RATELIMIT_STATE(_rs, \
93 DEFAULT_RATELIMIT_INTERVAL, \
94 DEFAULT_RATELIMIT_BURST); \
95 if (__ratelimit(&_rs)) \
96 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \
99 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
100 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
101 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
102 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
103 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
104 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
105 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
106 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
107 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
108 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
109 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
110 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
111 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
112 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
114 #if defined(CONFIG_DYNAMIC_DEBUG) || \
115 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
116 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
117 #define ibdev_dbg_ratelimited(ibdev, fmt, ...) \
119 static DEFINE_RATELIMIT_STATE(_rs, \
120 DEFAULT_RATELIMIT_INTERVAL, \
121 DEFAULT_RATELIMIT_BURST); \
122 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \
123 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \
124 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \
128 __printf(2, 3) __cold
130 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
136 __be64 subnet_prefix;
141 extern union ib_gid zgid;
144 IB_GID_TYPE_IB = IB_UVERBS_GID_TYPE_IB,
145 IB_GID_TYPE_ROCE = IB_UVERBS_GID_TYPE_ROCE_V1,
146 IB_GID_TYPE_ROCE_UDP_ENCAP = IB_UVERBS_GID_TYPE_ROCE_V2,
150 #define ROCE_V2_UDP_DPORT 4791
152 struct net_device __rcu *ndev;
153 struct ib_device *device;
155 enum ib_gid_type gid_type;
161 /* set the local administered indication */
162 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
165 enum rdma_transport_type {
167 RDMA_TRANSPORT_IWARP,
168 RDMA_TRANSPORT_USNIC,
169 RDMA_TRANSPORT_USNIC_UDP,
170 RDMA_TRANSPORT_UNSPECIFIED,
173 enum rdma_protocol_type {
177 RDMA_PROTOCOL_USNIC_UDP
180 __attribute_const__ enum rdma_transport_type
181 rdma_node_get_transport(unsigned int node_type);
183 enum rdma_network_type {
185 RDMA_NETWORK_ROCE_V1,
190 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
192 if (network_type == RDMA_NETWORK_IPV4 ||
193 network_type == RDMA_NETWORK_IPV6)
194 return IB_GID_TYPE_ROCE_UDP_ENCAP;
195 else if (network_type == RDMA_NETWORK_ROCE_V1)
196 return IB_GID_TYPE_ROCE;
198 return IB_GID_TYPE_IB;
201 static inline enum rdma_network_type
202 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
204 if (attr->gid_type == IB_GID_TYPE_IB)
205 return RDMA_NETWORK_IB;
207 if (attr->gid_type == IB_GID_TYPE_ROCE)
208 return RDMA_NETWORK_ROCE_V1;
210 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
211 return RDMA_NETWORK_IPV4;
213 return RDMA_NETWORK_IPV6;
216 enum rdma_link_layer {
217 IB_LINK_LAYER_UNSPECIFIED,
218 IB_LINK_LAYER_INFINIBAND,
219 IB_LINK_LAYER_ETHERNET,
222 enum ib_device_cap_flags {
223 IB_DEVICE_RESIZE_MAX_WR = IB_UVERBS_DEVICE_RESIZE_MAX_WR,
224 IB_DEVICE_BAD_PKEY_CNTR = IB_UVERBS_DEVICE_BAD_PKEY_CNTR,
225 IB_DEVICE_BAD_QKEY_CNTR = IB_UVERBS_DEVICE_BAD_QKEY_CNTR,
226 IB_DEVICE_RAW_MULTI = IB_UVERBS_DEVICE_RAW_MULTI,
227 IB_DEVICE_AUTO_PATH_MIG = IB_UVERBS_DEVICE_AUTO_PATH_MIG,
228 IB_DEVICE_CHANGE_PHY_PORT = IB_UVERBS_DEVICE_CHANGE_PHY_PORT,
229 IB_DEVICE_UD_AV_PORT_ENFORCE = IB_UVERBS_DEVICE_UD_AV_PORT_ENFORCE,
230 IB_DEVICE_CURR_QP_STATE_MOD = IB_UVERBS_DEVICE_CURR_QP_STATE_MOD,
231 IB_DEVICE_SHUTDOWN_PORT = IB_UVERBS_DEVICE_SHUTDOWN_PORT,
232 /* IB_DEVICE_INIT_TYPE = IB_UVERBS_DEVICE_INIT_TYPE, (not in use) */
233 IB_DEVICE_PORT_ACTIVE_EVENT = IB_UVERBS_DEVICE_PORT_ACTIVE_EVENT,
234 IB_DEVICE_SYS_IMAGE_GUID = IB_UVERBS_DEVICE_SYS_IMAGE_GUID,
235 IB_DEVICE_RC_RNR_NAK_GEN = IB_UVERBS_DEVICE_RC_RNR_NAK_GEN,
236 IB_DEVICE_SRQ_RESIZE = IB_UVERBS_DEVICE_SRQ_RESIZE,
237 IB_DEVICE_N_NOTIFY_CQ = IB_UVERBS_DEVICE_N_NOTIFY_CQ,
239 /* Reserved, old SEND_W_INV = 1 << 16,*/
240 IB_DEVICE_MEM_WINDOW = IB_UVERBS_DEVICE_MEM_WINDOW,
242 * Devices should set IB_DEVICE_UD_IP_SUM if they support
243 * insertion of UDP and TCP checksum on outgoing UD IPoIB
244 * messages and can verify the validity of checksum for
245 * incoming messages. Setting this flag implies that the
246 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
248 IB_DEVICE_UD_IP_CSUM = IB_UVERBS_DEVICE_UD_IP_CSUM,
249 IB_DEVICE_XRC = IB_UVERBS_DEVICE_XRC,
252 * This device supports the IB "base memory management extension",
253 * which includes support for fast registrations (IB_WR_REG_MR,
254 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
255 * also be set by any iWarp device which must support FRs to comply
256 * to the iWarp verbs spec. iWarp devices also support the
257 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
260 IB_DEVICE_MEM_MGT_EXTENSIONS = IB_UVERBS_DEVICE_MEM_MGT_EXTENSIONS,
261 IB_DEVICE_MEM_WINDOW_TYPE_2A = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2A,
262 IB_DEVICE_MEM_WINDOW_TYPE_2B = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2B,
263 IB_DEVICE_RC_IP_CSUM = IB_UVERBS_DEVICE_RC_IP_CSUM,
264 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
265 IB_DEVICE_RAW_IP_CSUM = IB_UVERBS_DEVICE_RAW_IP_CSUM,
266 IB_DEVICE_MANAGED_FLOW_STEERING =
267 IB_UVERBS_DEVICE_MANAGED_FLOW_STEERING,
268 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
269 IB_DEVICE_RAW_SCATTER_FCS = IB_UVERBS_DEVICE_RAW_SCATTER_FCS,
270 /* The device supports padding incoming writes to cacheline. */
271 IB_DEVICE_PCI_WRITE_END_PADDING =
272 IB_UVERBS_DEVICE_PCI_WRITE_END_PADDING,
273 /* Placement type attributes */
274 IB_DEVICE_FLUSH_GLOBAL = IB_UVERBS_DEVICE_FLUSH_GLOBAL,
275 IB_DEVICE_FLUSH_PERSISTENT = IB_UVERBS_DEVICE_FLUSH_PERSISTENT,
276 IB_DEVICE_ATOMIC_WRITE = IB_UVERBS_DEVICE_ATOMIC_WRITE,
279 enum ib_kernel_cap_flags {
281 * This device supports a per-device lkey or stag that can be
282 * used without performing a memory registration for the local
283 * memory. Note that ULPs should never check this flag, but
284 * instead of use the local_dma_lkey flag in the ib_pd structure,
285 * which will always contain a usable lkey.
287 IBK_LOCAL_DMA_LKEY = 1 << 0,
288 /* IB_QP_CREATE_INTEGRITY_EN is supported to implement T10-PI */
289 IBK_INTEGRITY_HANDOVER = 1 << 1,
290 /* IB_ACCESS_ON_DEMAND is supported during reg_user_mr() */
291 IBK_ON_DEMAND_PAGING = 1 << 2,
292 /* IB_MR_TYPE_SG_GAPS is supported */
293 IBK_SG_GAPS_REG = 1 << 3,
294 /* Driver supports RDMA_NLDEV_CMD_DELLINK */
295 IBK_ALLOW_USER_UNREG = 1 << 4,
297 /* ipoib will use IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK */
298 IBK_BLOCK_MULTICAST_LOOPBACK = 1 << 5,
299 /* iopib will use IB_QP_CREATE_IPOIB_UD_LSO for its QPs */
301 /* iopib will use the device ops:
308 IBK_VIRTUAL_FUNCTION = 1 << 7,
309 /* ipoib will use IB_QP_CREATE_NETDEV_USE for its QPs */
310 IBK_RDMA_NETDEV_OPA = 1 << 8,
319 enum ib_odp_general_cap_bits {
320 IB_ODP_SUPPORT = 1 << 0,
321 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
324 enum ib_odp_transport_cap_bits {
325 IB_ODP_SUPPORT_SEND = 1 << 0,
326 IB_ODP_SUPPORT_RECV = 1 << 1,
327 IB_ODP_SUPPORT_WRITE = 1 << 2,
328 IB_ODP_SUPPORT_READ = 1 << 3,
329 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
330 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
334 uint64_t general_caps;
336 uint32_t rc_odp_caps;
337 uint32_t uc_odp_caps;
338 uint32_t ud_odp_caps;
339 uint32_t xrc_odp_caps;
340 } per_transport_caps;
344 /* Corresponding bit will be set if qp type from
345 * 'enum ib_qp_type' is supported, e.g.
346 * supported_qpts |= 1 << IB_QPT_UD
349 u32 max_rwq_indirection_tables;
350 u32 max_rwq_indirection_table_size;
353 enum ib_tm_cap_flags {
354 /* Support tag matching with rendezvous offload for RC transport */
355 IB_TM_CAP_RNDV_RC = 1 << 0,
359 /* Max size of RNDV header */
360 u32 max_rndv_hdr_size;
361 /* Max number of entries in tag matching list */
363 /* From enum ib_tm_cap_flags */
365 /* Max number of outstanding list operations */
367 /* Max number of SGE in tag matching entry */
371 struct ib_cq_init_attr {
377 enum ib_cq_attr_mask {
378 IB_CQ_MODERATE = 1 << 0,
382 u16 max_cq_moderation_count;
383 u16 max_cq_moderation_period;
386 struct ib_dm_mr_attr {
392 struct ib_dm_alloc_attr {
398 struct ib_device_attr {
400 __be64 sys_image_guid;
408 u64 device_cap_flags;
409 u64 kernel_cap_flags;
420 int max_qp_init_rd_atom;
421 int max_ee_init_rd_atom;
422 enum ib_atomic_cap atomic_cap;
423 enum ib_atomic_cap masked_atomic_cap;
430 int max_mcast_qp_attach;
431 int max_total_mcast_qp_attach;
436 unsigned int max_fast_reg_page_list_len;
437 unsigned int max_pi_fast_reg_page_list_len;
439 u8 local_ca_ack_delay;
442 struct ib_odp_caps odp_caps;
443 uint64_t timestamp_mask;
444 uint64_t hca_core_clock; /* in KHZ */
445 struct ib_rss_caps rss_caps;
447 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
448 struct ib_tm_caps tm_caps;
449 struct ib_cq_caps cq_caps;
451 /* Max entries for sgl for optimized performance per READ */
468 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
471 case IB_MTU_256: return 256;
472 case IB_MTU_512: return 512;
473 case IB_MTU_1024: return 1024;
474 case IB_MTU_2048: return 2048;
475 case IB_MTU_4096: return 4096;
480 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
484 else if (mtu >= 2048)
486 else if (mtu >= 1024)
494 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
502 return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
506 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
509 return OPA_MTU_10240;
510 else if (mtu >= 8192)
513 return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
522 IB_PORT_ACTIVE_DEFER = 5
525 enum ib_port_phys_state {
526 IB_PORT_PHYS_STATE_SLEEP = 1,
527 IB_PORT_PHYS_STATE_POLLING = 2,
528 IB_PORT_PHYS_STATE_DISABLED = 3,
529 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
530 IB_PORT_PHYS_STATE_LINK_UP = 5,
531 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
532 IB_PORT_PHYS_STATE_PHY_TEST = 7,
543 static inline int ib_width_enum_to_int(enum ib_port_width width)
546 case IB_WIDTH_1X: return 1;
547 case IB_WIDTH_2X: return 2;
548 case IB_WIDTH_4X: return 4;
549 case IB_WIDTH_8X: return 8;
550 case IB_WIDTH_12X: return 12;
568 IB_STAT_FLAG_OPTIONAL = 1 << 0,
572 * struct rdma_stat_desc
573 * @name - The name of the counter
574 * @flags - Flags of the counter; For example, IB_STAT_FLAG_OPTIONAL
575 * @priv - Driver private information; Core code should not use
577 struct rdma_stat_desc {
584 * struct rdma_hw_stats
585 * @lock - Mutex to protect parallel write access to lifespan and values
586 * of counters, which are 64bits and not guaranteed to be written
587 * atomicaly on 32bits systems.
588 * @timestamp - Used by the core code to track when the last update was
589 * @lifespan - Used by the core code to determine how old the counters
590 * should be before being updated again. Stored in jiffies, defaults
591 * to 10 milliseconds, drivers can override the default be specifying
592 * their own value during their allocation routine.
593 * @descs - Array of pointers to static descriptors used for the counters
595 * @is_disabled - A bitmap to indicate each counter is currently disabled
597 * @num_counters - How many hardware counters there are. If name is
598 * shorter than this number, a kernel oops will result. Driver authors
599 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
600 * in their code to prevent this.
601 * @value - Array of u64 counters that are accessed by the sysfs code and
602 * filled in by the drivers get_stats routine
604 struct rdma_hw_stats {
605 struct mutex lock; /* Protect lifespan and values[] */
606 unsigned long timestamp;
607 unsigned long lifespan;
608 const struct rdma_stat_desc *descs;
609 unsigned long *is_disabled;
611 u64 value[] __counted_by(num_counters);
614 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
616 struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
617 const struct rdma_stat_desc *descs, int num_counters,
618 unsigned long lifespan);
620 void rdma_free_hw_stats_struct(struct rdma_hw_stats *stats);
622 /* Define bits for the various functionality this port needs to be supported by
625 /* Management 0x00000FFF */
626 #define RDMA_CORE_CAP_IB_MAD 0x00000001
627 #define RDMA_CORE_CAP_IB_SMI 0x00000002
628 #define RDMA_CORE_CAP_IB_CM 0x00000004
629 #define RDMA_CORE_CAP_IW_CM 0x00000008
630 #define RDMA_CORE_CAP_IB_SA 0x00000010
631 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
633 /* Address format 0x000FF000 */
634 #define RDMA_CORE_CAP_AF_IB 0x00001000
635 #define RDMA_CORE_CAP_ETH_AH 0x00002000
636 #define RDMA_CORE_CAP_OPA_AH 0x00004000
637 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
639 /* Protocol 0xFFF00000 */
640 #define RDMA_CORE_CAP_PROT_IB 0x00100000
641 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
642 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
643 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
644 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
645 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
647 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
648 | RDMA_CORE_CAP_PROT_ROCE \
649 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
651 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
652 | RDMA_CORE_CAP_IB_MAD \
653 | RDMA_CORE_CAP_IB_SMI \
654 | RDMA_CORE_CAP_IB_CM \
655 | RDMA_CORE_CAP_IB_SA \
656 | RDMA_CORE_CAP_AF_IB)
657 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
658 | RDMA_CORE_CAP_IB_MAD \
659 | RDMA_CORE_CAP_IB_CM \
660 | RDMA_CORE_CAP_AF_IB \
661 | RDMA_CORE_CAP_ETH_AH)
662 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
663 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
664 | RDMA_CORE_CAP_IB_MAD \
665 | RDMA_CORE_CAP_IB_CM \
666 | RDMA_CORE_CAP_AF_IB \
667 | RDMA_CORE_CAP_ETH_AH)
668 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
669 | RDMA_CORE_CAP_IW_CM)
670 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
671 | RDMA_CORE_CAP_OPA_MAD)
673 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
675 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
677 struct ib_port_attr {
679 enum ib_port_state state;
681 enum ib_mtu active_mtu;
684 unsigned int ip_gids:1;
685 /* This is the value from PortInfo CapabilityMask, defined by IBA */
704 enum ib_device_modify_flags {
705 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
706 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
709 #define IB_DEVICE_NODE_DESC_MAX 64
711 struct ib_device_modify {
713 char node_desc[IB_DEVICE_NODE_DESC_MAX];
716 enum ib_port_modify_flags {
717 IB_PORT_SHUTDOWN = 1,
718 IB_PORT_INIT_TYPE = (1<<2),
719 IB_PORT_RESET_QKEY_CNTR = (1<<3),
720 IB_PORT_OPA_MASK_CHG = (1<<4)
723 struct ib_port_modify {
724 u32 set_port_cap_mask;
725 u32 clr_port_cap_mask;
733 IB_EVENT_QP_ACCESS_ERR,
737 IB_EVENT_PATH_MIG_ERR,
738 IB_EVENT_DEVICE_FATAL,
739 IB_EVENT_PORT_ACTIVE,
742 IB_EVENT_PKEY_CHANGE,
745 IB_EVENT_SRQ_LIMIT_REACHED,
746 IB_EVENT_QP_LAST_WQE_REACHED,
747 IB_EVENT_CLIENT_REREGISTER,
752 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
755 struct ib_device *device;
763 enum ib_event_type event;
766 struct ib_event_handler {
767 struct ib_device *device;
768 void (*handler)(struct ib_event_handler *, struct ib_event *);
769 struct list_head list;
772 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
774 (_ptr)->device = _device; \
775 (_ptr)->handler = _handler; \
776 INIT_LIST_HEAD(&(_ptr)->list); \
779 struct ib_global_route {
780 const struct ib_gid_attr *sgid_attr;
789 __be32 version_tclass_flow;
797 union rdma_network_hdr {
800 /* The IB spec states that if it's IPv4, the header
801 * is located in the last 20 bytes of the header.
804 struct iphdr roce4grh;
808 #define IB_QPN_MASK 0xFFFFFF
811 IB_MULTICAST_QPN = 0xffffff
814 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
815 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
822 IB_RATE_PORT_CURRENT = 0,
823 IB_RATE_2_5_GBPS = 2,
831 IB_RATE_120_GBPS = 10,
832 IB_RATE_14_GBPS = 11,
833 IB_RATE_56_GBPS = 12,
834 IB_RATE_112_GBPS = 13,
835 IB_RATE_168_GBPS = 14,
836 IB_RATE_25_GBPS = 15,
837 IB_RATE_100_GBPS = 16,
838 IB_RATE_200_GBPS = 17,
839 IB_RATE_300_GBPS = 18,
840 IB_RATE_28_GBPS = 19,
841 IB_RATE_50_GBPS = 20,
842 IB_RATE_400_GBPS = 21,
843 IB_RATE_600_GBPS = 22,
844 IB_RATE_800_GBPS = 23,
848 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
849 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
850 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
851 * @rate: rate to convert.
853 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
856 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
857 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
858 * @rate: rate to convert.
860 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
864 * enum ib_mr_type - memory region type
865 * @IB_MR_TYPE_MEM_REG: memory region that is used for
866 * normal registration
867 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
868 * register any arbitrary sg lists (without
869 * the normal mr constraints - see
871 * @IB_MR_TYPE_DM: memory region that is used for device
872 * memory registration
873 * @IB_MR_TYPE_USER: memory region that is used for the user-space
875 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
876 * without address translations (VA=PA)
877 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
878 * data integrity operations
886 IB_MR_TYPE_INTEGRITY,
889 enum ib_mr_status_check {
890 IB_MR_CHECK_SIG_STATUS = 1,
894 * struct ib_mr_status - Memory region status container
896 * @fail_status: Bitmask of MR checks status. For each
897 * failed check a corresponding status bit is set.
898 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
901 struct ib_mr_status {
903 struct ib_sig_err sig_err;
907 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
909 * @mult: multiple to convert.
911 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
913 struct rdma_ah_init_attr {
914 struct rdma_ah_attr *ah_attr;
916 struct net_device *xmit_slave;
919 enum rdma_ah_attr_type {
920 RDMA_AH_ATTR_TYPE_UNDEFINED,
921 RDMA_AH_ATTR_TYPE_IB,
922 RDMA_AH_ATTR_TYPE_ROCE,
923 RDMA_AH_ATTR_TYPE_OPA,
931 struct roce_ah_attr {
941 struct rdma_ah_attr {
942 struct ib_global_route grh;
947 enum rdma_ah_attr_type type;
949 struct ib_ah_attr ib;
950 struct roce_ah_attr roce;
951 struct opa_ah_attr opa;
959 IB_WC_LOC_EEC_OP_ERR,
964 IB_WC_LOC_ACCESS_ERR,
965 IB_WC_REM_INV_REQ_ERR,
966 IB_WC_REM_ACCESS_ERR,
969 IB_WC_RNR_RETRY_EXC_ERR,
970 IB_WC_LOC_RDD_VIOL_ERR,
971 IB_WC_REM_INV_RD_REQ_ERR,
974 IB_WC_INV_EEC_STATE_ERR,
976 IB_WC_RESP_TIMEOUT_ERR,
980 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
983 IB_WC_SEND = IB_UVERBS_WC_SEND,
984 IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE,
985 IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ,
986 IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP,
987 IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD,
988 IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW,
989 IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV,
990 IB_WC_LSO = IB_UVERBS_WC_TSO,
991 IB_WC_ATOMIC_WRITE = IB_UVERBS_WC_ATOMIC_WRITE,
993 IB_WC_MASKED_COMP_SWAP,
994 IB_WC_MASKED_FETCH_ADD,
995 IB_WC_FLUSH = IB_UVERBS_WC_FLUSH,
997 * Set value of IB_WC_RECV so consumers can test if a completion is a
998 * receive by testing (opcode & IB_WC_RECV).
1000 IB_WC_RECV = 1 << 7,
1001 IB_WC_RECV_RDMA_WITH_IMM
1006 IB_WC_WITH_IMM = (1<<1),
1007 IB_WC_WITH_INVALIDATE = (1<<2),
1008 IB_WC_IP_CSUM_OK = (1<<3),
1009 IB_WC_WITH_SMAC = (1<<4),
1010 IB_WC_WITH_VLAN = (1<<5),
1011 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
1017 struct ib_cqe *wr_cqe;
1019 enum ib_wc_status status;
1020 enum ib_wc_opcode opcode;
1026 u32 invalidate_rkey;
1034 u32 port_num; /* valid only for DR SMPs on switches */
1037 u8 network_hdr_type;
1040 enum ib_cq_notify_flags {
1041 IB_CQ_SOLICITED = 1 << 0,
1042 IB_CQ_NEXT_COMP = 1 << 1,
1043 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1044 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1048 IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1049 IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1050 IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1053 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1055 return srq_type == IB_SRQT_XRC ||
1056 srq_type == IB_SRQT_TM;
1059 enum ib_srq_attr_mask {
1060 IB_SRQ_MAX_WR = 1 << 0,
1061 IB_SRQ_LIMIT = 1 << 1,
1064 struct ib_srq_attr {
1070 struct ib_srq_init_attr {
1071 void (*event_handler)(struct ib_event *, void *);
1073 struct ib_srq_attr attr;
1074 enum ib_srq_type srq_type;
1080 struct ib_xrcd *xrcd;
1095 u32 max_inline_data;
1098 * Maximum number of rdma_rw_ctx structures in flight at a time.
1099 * ib_create_qp() will calculate the right amount of needed WRs
1100 * and MRs based on this.
1112 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1113 * here (and in that order) since the MAD layer uses them as
1114 * indices into a 2-entry table.
1119 IB_QPT_RC = IB_UVERBS_QPT_RC,
1120 IB_QPT_UC = IB_UVERBS_QPT_UC,
1121 IB_QPT_UD = IB_UVERBS_QPT_UD,
1123 IB_QPT_RAW_ETHERTYPE,
1124 IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1125 IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1126 IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1128 IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1129 /* Reserve a range for qp types internal to the low level driver.
1130 * These qp types will not be visible at the IB core layer, so the
1131 * IB_QPT_MAX usages should not be affected in the core layer
1133 IB_QPT_RESERVED1 = 0x1000,
1145 enum ib_qp_create_flags {
1146 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1147 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK =
1148 IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1149 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1150 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1151 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1152 IB_QP_CREATE_NETIF_QP = 1 << 5,
1153 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
1154 IB_QP_CREATE_NETDEV_USE = 1 << 7,
1155 IB_QP_CREATE_SCATTER_FCS =
1156 IB_UVERBS_QP_CREATE_SCATTER_FCS,
1157 IB_QP_CREATE_CVLAN_STRIPPING =
1158 IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1159 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1160 IB_QP_CREATE_PCI_WRITE_END_PADDING =
1161 IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1162 /* reserve bits 26-31 for low level drivers' internal use */
1163 IB_QP_CREATE_RESERVED_START = 1 << 26,
1164 IB_QP_CREATE_RESERVED_END = 1 << 31,
1168 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1169 * callback to destroy the passed in QP.
1172 struct ib_qp_init_attr {
1173 /* This callback occurs in workqueue context */
1174 void (*event_handler)(struct ib_event *, void *);
1177 struct ib_cq *send_cq;
1178 struct ib_cq *recv_cq;
1180 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1181 struct ib_qp_cap cap;
1182 enum ib_sig_type sq_sig_type;
1183 enum ib_qp_type qp_type;
1187 * Only needed for special QP types, or when using the RW API.
1190 struct ib_rwq_ind_table *rwq_ind_tbl;
1194 struct ib_qp_open_attr {
1195 void (*event_handler)(struct ib_event *, void *);
1198 enum ib_qp_type qp_type;
1201 enum ib_rnr_timeout {
1202 IB_RNR_TIMER_655_36 = 0,
1203 IB_RNR_TIMER_000_01 = 1,
1204 IB_RNR_TIMER_000_02 = 2,
1205 IB_RNR_TIMER_000_03 = 3,
1206 IB_RNR_TIMER_000_04 = 4,
1207 IB_RNR_TIMER_000_06 = 5,
1208 IB_RNR_TIMER_000_08 = 6,
1209 IB_RNR_TIMER_000_12 = 7,
1210 IB_RNR_TIMER_000_16 = 8,
1211 IB_RNR_TIMER_000_24 = 9,
1212 IB_RNR_TIMER_000_32 = 10,
1213 IB_RNR_TIMER_000_48 = 11,
1214 IB_RNR_TIMER_000_64 = 12,
1215 IB_RNR_TIMER_000_96 = 13,
1216 IB_RNR_TIMER_001_28 = 14,
1217 IB_RNR_TIMER_001_92 = 15,
1218 IB_RNR_TIMER_002_56 = 16,
1219 IB_RNR_TIMER_003_84 = 17,
1220 IB_RNR_TIMER_005_12 = 18,
1221 IB_RNR_TIMER_007_68 = 19,
1222 IB_RNR_TIMER_010_24 = 20,
1223 IB_RNR_TIMER_015_36 = 21,
1224 IB_RNR_TIMER_020_48 = 22,
1225 IB_RNR_TIMER_030_72 = 23,
1226 IB_RNR_TIMER_040_96 = 24,
1227 IB_RNR_TIMER_061_44 = 25,
1228 IB_RNR_TIMER_081_92 = 26,
1229 IB_RNR_TIMER_122_88 = 27,
1230 IB_RNR_TIMER_163_84 = 28,
1231 IB_RNR_TIMER_245_76 = 29,
1232 IB_RNR_TIMER_327_68 = 30,
1233 IB_RNR_TIMER_491_52 = 31
1236 enum ib_qp_attr_mask {
1238 IB_QP_CUR_STATE = (1<<1),
1239 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1240 IB_QP_ACCESS_FLAGS = (1<<3),
1241 IB_QP_PKEY_INDEX = (1<<4),
1242 IB_QP_PORT = (1<<5),
1243 IB_QP_QKEY = (1<<6),
1245 IB_QP_PATH_MTU = (1<<8),
1246 IB_QP_TIMEOUT = (1<<9),
1247 IB_QP_RETRY_CNT = (1<<10),
1248 IB_QP_RNR_RETRY = (1<<11),
1249 IB_QP_RQ_PSN = (1<<12),
1250 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1251 IB_QP_ALT_PATH = (1<<14),
1252 IB_QP_MIN_RNR_TIMER = (1<<15),
1253 IB_QP_SQ_PSN = (1<<16),
1254 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1255 IB_QP_PATH_MIG_STATE = (1<<18),
1256 IB_QP_CAP = (1<<19),
1257 IB_QP_DEST_QPN = (1<<20),
1258 IB_QP_RESERVED1 = (1<<21),
1259 IB_QP_RESERVED2 = (1<<22),
1260 IB_QP_RESERVED3 = (1<<23),
1261 IB_QP_RESERVED4 = (1<<24),
1262 IB_QP_RATE_LIMIT = (1<<25),
1264 IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0),
1289 enum ib_qp_state qp_state;
1290 enum ib_qp_state cur_qp_state;
1291 enum ib_mtu path_mtu;
1292 enum ib_mig_state path_mig_state;
1297 int qp_access_flags;
1298 struct ib_qp_cap cap;
1299 struct rdma_ah_attr ah_attr;
1300 struct rdma_ah_attr alt_ah_attr;
1303 u8 en_sqd_async_notify;
1306 u8 max_dest_rd_atomic;
1315 struct net_device *xmit_slave;
1319 /* These are shared with userspace */
1320 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1321 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1322 IB_WR_SEND = IB_UVERBS_WR_SEND,
1323 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1324 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1325 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1326 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1327 IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW,
1328 IB_WR_LSO = IB_UVERBS_WR_TSO,
1329 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1330 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1331 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1332 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1333 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1334 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1335 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1336 IB_WR_FLUSH = IB_UVERBS_WR_FLUSH,
1337 IB_WR_ATOMIC_WRITE = IB_UVERBS_WR_ATOMIC_WRITE,
1339 /* These are kernel only and can not be issued by userspace */
1340 IB_WR_REG_MR = 0x20,
1341 IB_WR_REG_MR_INTEGRITY,
1343 /* reserve values for low level drivers' internal use.
1344 * These values will not be used at all in the ib core layer.
1346 IB_WR_RESERVED1 = 0xf0,
1358 enum ib_send_flags {
1360 IB_SEND_SIGNALED = (1<<1),
1361 IB_SEND_SOLICITED = (1<<2),
1362 IB_SEND_INLINE = (1<<3),
1363 IB_SEND_IP_CSUM = (1<<4),
1365 /* reserve bits 26-31 for low level drivers' internal use */
1366 IB_SEND_RESERVED_START = (1 << 26),
1367 IB_SEND_RESERVED_END = (1 << 31),
1377 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1381 struct ib_send_wr *next;
1384 struct ib_cqe *wr_cqe;
1386 struct ib_sge *sg_list;
1388 enum ib_wr_opcode opcode;
1392 u32 invalidate_rkey;
1397 struct ib_send_wr wr;
1402 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1404 return container_of(wr, struct ib_rdma_wr, wr);
1407 struct ib_atomic_wr {
1408 struct ib_send_wr wr;
1412 u64 compare_add_mask;
1417 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1419 return container_of(wr, struct ib_atomic_wr, wr);
1423 struct ib_send_wr wr;
1430 u16 pkey_index; /* valid for GSI only */
1431 u32 port_num; /* valid for DR SMPs on switch only */
1434 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1436 return container_of(wr, struct ib_ud_wr, wr);
1440 struct ib_send_wr wr;
1446 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1448 return container_of(wr, struct ib_reg_wr, wr);
1452 struct ib_recv_wr *next;
1455 struct ib_cqe *wr_cqe;
1457 struct ib_sge *sg_list;
1461 enum ib_access_flags {
1462 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1463 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1464 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1465 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1466 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1467 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1468 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1469 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1470 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1471 IB_ACCESS_FLUSH_GLOBAL = IB_UVERBS_ACCESS_FLUSH_GLOBAL,
1472 IB_ACCESS_FLUSH_PERSISTENT = IB_UVERBS_ACCESS_FLUSH_PERSISTENT,
1474 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1475 IB_ACCESS_SUPPORTED =
1476 ((IB_ACCESS_FLUSH_PERSISTENT << 1) - 1) | IB_ACCESS_OPTIONAL,
1480 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1481 * are hidden here instead of a uapi header!
1483 enum ib_mr_rereg_flags {
1484 IB_MR_REREG_TRANS = 1,
1485 IB_MR_REREG_PD = (1<<1),
1486 IB_MR_REREG_ACCESS = (1<<2),
1487 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1492 enum rdma_remove_reason {
1494 * Userspace requested uobject deletion or initial try
1495 * to remove uobject via cleanup. Call could fail
1497 RDMA_REMOVE_DESTROY,
1498 /* Context deletion. This call should delete the actual object itself */
1500 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1501 RDMA_REMOVE_DRIVER_REMOVE,
1502 /* uobj is being cleaned-up before being committed */
1504 /* The driver failed to destroy the uobject and is being disconnected */
1505 RDMA_REMOVE_DRIVER_FAILURE,
1508 struct ib_rdmacg_object {
1509 #ifdef CONFIG_CGROUP_RDMA
1510 struct rdma_cgroup *cg; /* owner rdma cgroup */
1514 struct ib_ucontext {
1515 struct ib_device *device;
1516 struct ib_uverbs_file *ufile;
1518 struct ib_rdmacg_object cg_obj;
1520 * Implementation details of the RDMA core, don't use in drivers:
1522 struct rdma_restrack_entry res;
1523 struct xarray mmap_xa;
1527 u64 user_handle; /* handle given to us by userspace */
1528 /* ufile & ucontext owning this object */
1529 struct ib_uverbs_file *ufile;
1530 /* FIXME, save memory: ufile->context == context */
1531 struct ib_ucontext *context; /* associated user context */
1532 void *object; /* containing object */
1533 struct list_head list; /* link to context's list */
1534 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1535 int id; /* index into kernel idr */
1537 atomic_t usecnt; /* protects exclusive access */
1538 struct rcu_head rcu; /* kfree_rcu() overhead */
1540 const struct uverbs_api_object *uapi_object;
1544 const void __user *inbuf;
1545 void __user *outbuf;
1553 struct ib_device *device;
1554 struct ib_uobject *uobject;
1555 atomic_t usecnt; /* count all resources */
1557 u32 unsafe_global_rkey;
1560 * Implementation details of the RDMA core, don't use in drivers:
1562 struct ib_mr *__internal_mr;
1563 struct rdma_restrack_entry res;
1567 struct ib_device *device;
1568 atomic_t usecnt; /* count all exposed resources */
1569 struct inode *inode;
1570 struct rw_semaphore tgt_qps_rwsem;
1571 struct xarray tgt_qps;
1575 struct ib_device *device;
1577 struct ib_uobject *uobject;
1578 const struct ib_gid_attr *sgid_attr;
1579 enum rdma_ah_attr_type type;
1582 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1584 enum ib_poll_context {
1585 IB_POLL_SOFTIRQ, /* poll from softirq context */
1586 IB_POLL_WORKQUEUE, /* poll from workqueue */
1587 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1588 IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1590 IB_POLL_DIRECT, /* caller context, no hw completions */
1594 struct ib_device *device;
1595 struct ib_ucq_object *uobject;
1596 ib_comp_handler comp_handler;
1597 void (*event_handler)(struct ib_event *, void *);
1600 unsigned int cqe_used;
1601 atomic_t usecnt; /* count number of work queues */
1602 enum ib_poll_context poll_ctx;
1604 struct list_head pool_entry;
1606 struct irq_poll iop;
1607 struct work_struct work;
1609 struct workqueue_struct *comp_wq;
1612 /* updated only by trace points */
1616 unsigned int comp_vector;
1619 * Implementation details of the RDMA core, don't use in drivers:
1621 struct rdma_restrack_entry res;
1625 struct ib_device *device;
1627 struct ib_usrq_object *uobject;
1628 void (*event_handler)(struct ib_event *, void *);
1630 enum ib_srq_type srq_type;
1637 struct ib_xrcd *xrcd;
1644 * Implementation details of the RDMA core, don't use in drivers:
1646 struct rdma_restrack_entry res;
1649 enum ib_raw_packet_caps {
1651 * Strip cvlan from incoming packet and report it in the matching work
1652 * completion is supported.
1654 IB_RAW_PACKET_CAP_CVLAN_STRIPPING =
1655 IB_UVERBS_RAW_PACKET_CAP_CVLAN_STRIPPING,
1657 * Scatter FCS field of an incoming packet to host memory is supported.
1659 IB_RAW_PACKET_CAP_SCATTER_FCS = IB_UVERBS_RAW_PACKET_CAP_SCATTER_FCS,
1660 /* Checksum offloads are supported (for both send and receive). */
1661 IB_RAW_PACKET_CAP_IP_CSUM = IB_UVERBS_RAW_PACKET_CAP_IP_CSUM,
1663 * When a packet is received for an RQ with no receive WQEs, the
1664 * packet processing is delayed.
1666 IB_RAW_PACKET_CAP_DELAY_DROP = IB_UVERBS_RAW_PACKET_CAP_DELAY_DROP,
1670 IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1680 struct ib_device *device;
1681 struct ib_uwq_object *uobject;
1683 void (*event_handler)(struct ib_event *, void *);
1687 enum ib_wq_state state;
1688 enum ib_wq_type wq_type;
1693 IB_WQ_FLAGS_CVLAN_STRIPPING = IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1694 IB_WQ_FLAGS_SCATTER_FCS = IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1695 IB_WQ_FLAGS_DELAY_DROP = IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1696 IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1697 IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1700 struct ib_wq_init_attr {
1702 enum ib_wq_type wq_type;
1706 void (*event_handler)(struct ib_event *, void *);
1707 u32 create_flags; /* Use enum ib_wq_flags */
1710 enum ib_wq_attr_mask {
1711 IB_WQ_STATE = 1 << 0,
1712 IB_WQ_CUR_STATE = 1 << 1,
1713 IB_WQ_FLAGS = 1 << 2,
1717 enum ib_wq_state wq_state;
1718 enum ib_wq_state curr_wq_state;
1719 u32 flags; /* Use enum ib_wq_flags */
1720 u32 flags_mask; /* Use enum ib_wq_flags */
1723 struct ib_rwq_ind_table {
1724 struct ib_device *device;
1725 struct ib_uobject *uobject;
1728 u32 log_ind_tbl_size;
1729 struct ib_wq **ind_tbl;
1732 struct ib_rwq_ind_table_init_attr {
1733 u32 log_ind_tbl_size;
1734 /* Each entry is a pointer to Receive Work Queue */
1735 struct ib_wq **ind_tbl;
1738 enum port_pkey_state {
1739 IB_PORT_PKEY_NOT_VALID = 0,
1740 IB_PORT_PKEY_VALID = 1,
1741 IB_PORT_PKEY_LISTED = 2,
1744 struct ib_qp_security;
1746 struct ib_port_pkey {
1747 enum port_pkey_state state;
1750 struct list_head qp_list;
1751 struct list_head to_error_list;
1752 struct ib_qp_security *sec;
1755 struct ib_ports_pkeys {
1756 struct ib_port_pkey main;
1757 struct ib_port_pkey alt;
1760 struct ib_qp_security {
1762 struct ib_device *dev;
1763 /* Hold this mutex when changing port and pkey settings. */
1765 struct ib_ports_pkeys *ports_pkeys;
1766 /* A list of all open shared QP handles. Required to enforce security
1767 * properly for all users of a shared QP.
1769 struct list_head shared_qp_list;
1772 atomic_t error_list_count;
1773 struct completion error_complete;
1774 int error_comps_pending;
1778 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1779 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1782 struct ib_device *device;
1784 struct ib_cq *send_cq;
1785 struct ib_cq *recv_cq;
1788 struct list_head rdma_mrs;
1789 struct list_head sig_mrs;
1791 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1792 struct list_head xrcd_list;
1794 /* count times opened, mcast attaches, flow attaches */
1796 struct list_head open_list;
1797 struct ib_qp *real_qp;
1798 struct ib_uqp_object *uobject;
1799 void (*event_handler)(struct ib_event *, void *);
1801 /* sgid_attrs associated with the AV's */
1802 const struct ib_gid_attr *av_sgid_attr;
1803 const struct ib_gid_attr *alt_path_sgid_attr;
1807 enum ib_qp_type qp_type;
1808 struct ib_rwq_ind_table *rwq_ind_tbl;
1809 struct ib_qp_security *qp_sec;
1814 * Implementation details of the RDMA core, don't use in drivers:
1816 struct rdma_restrack_entry res;
1818 /* The counter the qp is bind to */
1819 struct rdma_counter *counter;
1823 struct ib_device *device;
1826 struct ib_uobject *uobject;
1831 struct ib_device *device;
1837 unsigned int page_size;
1838 enum ib_mr_type type;
1841 struct ib_uobject *uobject; /* user */
1842 struct list_head qp_entry; /* FR */
1846 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1848 * Implementation details of the RDMA core, don't use in drivers:
1850 struct rdma_restrack_entry res;
1854 struct ib_device *device;
1856 struct ib_uobject *uobject;
1858 enum ib_mw_type type;
1861 /* Supported steering options */
1862 enum ib_flow_attr_type {
1863 /* steering according to rule specifications */
1864 IB_FLOW_ATTR_NORMAL = 0x0,
1865 /* default unicast and multicast rule -
1866 * receive all Eth traffic which isn't steered to any QP
1868 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1869 /* default multicast rule -
1870 * receive all Eth multicast traffic which isn't steered to any QP
1872 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1873 /* sniffer rule - receive all port traffic */
1874 IB_FLOW_ATTR_SNIFFER = 0x3
1877 /* Supported steering header types */
1878 enum ib_flow_spec_type {
1880 IB_FLOW_SPEC_ETH = 0x20,
1881 IB_FLOW_SPEC_IB = 0x22,
1883 IB_FLOW_SPEC_IPV4 = 0x30,
1884 IB_FLOW_SPEC_IPV6 = 0x31,
1885 IB_FLOW_SPEC_ESP = 0x34,
1887 IB_FLOW_SPEC_TCP = 0x40,
1888 IB_FLOW_SPEC_UDP = 0x41,
1889 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1890 IB_FLOW_SPEC_GRE = 0x51,
1891 IB_FLOW_SPEC_MPLS = 0x60,
1892 IB_FLOW_SPEC_INNER = 0x100,
1894 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1895 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1896 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1897 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1899 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1900 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1902 enum ib_flow_flags {
1903 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1904 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1905 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1908 struct ib_flow_eth_filter {
1915 struct ib_flow_spec_eth {
1918 struct ib_flow_eth_filter val;
1919 struct ib_flow_eth_filter mask;
1922 struct ib_flow_ib_filter {
1927 struct ib_flow_spec_ib {
1930 struct ib_flow_ib_filter val;
1931 struct ib_flow_ib_filter mask;
1934 /* IPv4 header flags */
1935 enum ib_ipv4_flags {
1936 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1937 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1938 last have this flag set */
1941 struct ib_flow_ipv4_filter {
1950 struct ib_flow_spec_ipv4 {
1953 struct ib_flow_ipv4_filter val;
1954 struct ib_flow_ipv4_filter mask;
1957 struct ib_flow_ipv6_filter {
1966 struct ib_flow_spec_ipv6 {
1969 struct ib_flow_ipv6_filter val;
1970 struct ib_flow_ipv6_filter mask;
1973 struct ib_flow_tcp_udp_filter {
1978 struct ib_flow_spec_tcp_udp {
1981 struct ib_flow_tcp_udp_filter val;
1982 struct ib_flow_tcp_udp_filter mask;
1985 struct ib_flow_tunnel_filter {
1989 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1990 * the tunnel_id from val has the vni value
1992 struct ib_flow_spec_tunnel {
1995 struct ib_flow_tunnel_filter val;
1996 struct ib_flow_tunnel_filter mask;
1999 struct ib_flow_esp_filter {
2004 struct ib_flow_spec_esp {
2007 struct ib_flow_esp_filter val;
2008 struct ib_flow_esp_filter mask;
2011 struct ib_flow_gre_filter {
2012 __be16 c_ks_res0_ver;
2017 struct ib_flow_spec_gre {
2020 struct ib_flow_gre_filter val;
2021 struct ib_flow_gre_filter mask;
2024 struct ib_flow_mpls_filter {
2028 struct ib_flow_spec_mpls {
2031 struct ib_flow_mpls_filter val;
2032 struct ib_flow_mpls_filter mask;
2035 struct ib_flow_spec_action_tag {
2036 enum ib_flow_spec_type type;
2041 struct ib_flow_spec_action_drop {
2042 enum ib_flow_spec_type type;
2046 struct ib_flow_spec_action_handle {
2047 enum ib_flow_spec_type type;
2049 struct ib_flow_action *act;
2052 enum ib_counters_description {
2057 struct ib_flow_spec_action_count {
2058 enum ib_flow_spec_type type;
2060 struct ib_counters *counters;
2063 union ib_flow_spec {
2068 struct ib_flow_spec_eth eth;
2069 struct ib_flow_spec_ib ib;
2070 struct ib_flow_spec_ipv4 ipv4;
2071 struct ib_flow_spec_tcp_udp tcp_udp;
2072 struct ib_flow_spec_ipv6 ipv6;
2073 struct ib_flow_spec_tunnel tunnel;
2074 struct ib_flow_spec_esp esp;
2075 struct ib_flow_spec_gre gre;
2076 struct ib_flow_spec_mpls mpls;
2077 struct ib_flow_spec_action_tag flow_tag;
2078 struct ib_flow_spec_action_drop drop;
2079 struct ib_flow_spec_action_handle action;
2080 struct ib_flow_spec_action_count flow_count;
2083 struct ib_flow_attr {
2084 enum ib_flow_attr_type type;
2090 union ib_flow_spec flows[];
2095 struct ib_device *device;
2096 struct ib_uobject *uobject;
2099 enum ib_flow_action_type {
2100 IB_FLOW_ACTION_UNSPECIFIED,
2101 IB_FLOW_ACTION_ESP = 1,
2104 struct ib_flow_action_attrs_esp_keymats {
2105 enum ib_uverbs_flow_action_esp_keymat protocol;
2107 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2111 struct ib_flow_action_attrs_esp_replays {
2112 enum ib_uverbs_flow_action_esp_replay protocol;
2114 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2118 enum ib_flow_action_attrs_esp_flags {
2119 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2120 * This is done in order to share the same flags between user-space and
2121 * kernel and spare an unnecessary translation.
2125 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2126 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2129 struct ib_flow_spec_list {
2130 struct ib_flow_spec_list *next;
2131 union ib_flow_spec spec;
2134 struct ib_flow_action_attrs_esp {
2135 struct ib_flow_action_attrs_esp_keymats *keymat;
2136 struct ib_flow_action_attrs_esp_replays *replay;
2137 struct ib_flow_spec_list *encap;
2138 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2139 * Value of 0 is a valid value.
2145 /* Use enum ib_flow_action_attrs_esp_flags */
2147 u64 hard_limit_pkts;
2150 struct ib_flow_action {
2151 struct ib_device *device;
2152 struct ib_uobject *uobject;
2153 enum ib_flow_action_type type;
2159 enum ib_process_mad_flags {
2160 IB_MAD_IGNORE_MKEY = 1,
2161 IB_MAD_IGNORE_BKEY = 2,
2162 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2165 enum ib_mad_result {
2166 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2167 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2168 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2169 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2172 struct ib_port_cache {
2174 struct ib_pkey_cache *pkey;
2175 struct ib_gid_table *gid;
2177 enum ib_port_state port_state;
2180 struct ib_port_immutable {
2187 struct ib_port_data {
2188 struct ib_device *ib_dev;
2190 struct ib_port_immutable immutable;
2192 spinlock_t pkey_list_lock;
2194 spinlock_t netdev_lock;
2196 struct list_head pkey_list;
2198 struct ib_port_cache cache;
2200 struct net_device __rcu *netdev;
2201 netdevice_tracker netdev_tracker;
2202 struct hlist_node ndev_hash_link;
2203 struct rdma_port_counter port_counter;
2204 struct ib_port *sysfs;
2207 /* rdma netdev type - specifies protocol type */
2208 enum rdma_netdev_t {
2209 RDMA_NETDEV_OPA_VNIC,
2214 * struct rdma_netdev - rdma netdev
2215 * For cases where netstack interfacing is required.
2217 struct rdma_netdev {
2219 struct ib_device *hca;
2224 * cleanup function must be specified.
2225 * FIXME: This is only used for OPA_VNIC and that usage should be
2228 void (*free_rdma_netdev)(struct net_device *netdev);
2230 /* control functions */
2231 void (*set_id)(struct net_device *netdev, int id);
2233 int (*send)(struct net_device *dev, struct sk_buff *skb,
2234 struct ib_ah *address, u32 dqpn);
2236 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2237 union ib_gid *gid, u16 mlid,
2238 int set_qkey, u32 qkey);
2239 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2240 union ib_gid *gid, u16 mlid);
2242 void (*tx_timeout)(struct net_device *dev, unsigned int txqueue);
2245 struct rdma_netdev_alloc_params {
2251 int (*initialize_rdma_netdev)(struct ib_device *device, u32 port_num,
2252 struct net_device *netdev, void *param);
2255 struct ib_odp_counters {
2257 atomic64_t invalidations;
2258 atomic64_t prefetch;
2261 struct ib_counters {
2262 struct ib_device *device;
2263 struct ib_uobject *uobject;
2264 /* num of objects attached */
2268 struct ib_counters_read_attr {
2271 u32 flags; /* use enum ib_read_counters_flags */
2274 struct uverbs_attr_bundle;
2276 struct iw_cm_conn_param;
2278 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2279 .size_##ib_struct = \
2280 (sizeof(struct drv_struct) + \
2281 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2282 BUILD_BUG_ON_ZERO( \
2283 !__same_type(((struct drv_struct *)NULL)->member, \
2286 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2287 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2290 #define rdma_zalloc_drv_obj_numa(ib_dev, ib_type) \
2291 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2294 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2295 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2297 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2299 struct rdma_user_mmap_entry {
2301 struct ib_ucontext *ucontext;
2302 unsigned long start_pgoff;
2304 bool driver_removed;
2307 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2309 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2311 return (u64)entry->start_pgoff << PAGE_SHIFT;
2315 * struct ib_device_ops - InfiniBand device operations
2316 * This structure defines all the InfiniBand device operations, providers will
2317 * need to define the supported operations, otherwise they will be set to null.
2319 struct ib_device_ops {
2320 struct module *owner;
2321 enum rdma_driver_id driver_id;
2323 unsigned int uverbs_no_driver_id_binding:1;
2326 * NOTE: New drivers should not make use of device_group; instead new
2327 * device parameter should be exposed via netlink command. This
2328 * mechanism exists only for existing drivers.
2330 const struct attribute_group *device_group;
2331 const struct attribute_group **port_groups;
2333 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2334 const struct ib_send_wr **bad_send_wr);
2335 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2336 const struct ib_recv_wr **bad_recv_wr);
2337 void (*drain_rq)(struct ib_qp *qp);
2338 void (*drain_sq)(struct ib_qp *qp);
2339 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2340 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2341 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2342 int (*post_srq_recv)(struct ib_srq *srq,
2343 const struct ib_recv_wr *recv_wr,
2344 const struct ib_recv_wr **bad_recv_wr);
2345 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2346 u32 port_num, const struct ib_wc *in_wc,
2347 const struct ib_grh *in_grh,
2348 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2349 size_t *out_mad_size, u16 *out_mad_pkey_index);
2350 int (*query_device)(struct ib_device *device,
2351 struct ib_device_attr *device_attr,
2352 struct ib_udata *udata);
2353 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2354 struct ib_device_modify *device_modify);
2355 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2356 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2358 int (*query_port)(struct ib_device *device, u32 port_num,
2359 struct ib_port_attr *port_attr);
2360 int (*modify_port)(struct ib_device *device, u32 port_num,
2361 int port_modify_mask,
2362 struct ib_port_modify *port_modify);
2364 * The following mandatory functions are used only at device
2365 * registration. Keep functions such as these at the end of this
2366 * structure to avoid cache line misses when accessing struct ib_device
2369 int (*get_port_immutable)(struct ib_device *device, u32 port_num,
2370 struct ib_port_immutable *immutable);
2371 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2374 * When calling get_netdev, the HW vendor's driver should return the
2375 * net device of device @device at port @port_num or NULL if such
2376 * a net device doesn't exist. The vendor driver should call dev_hold
2377 * on this net device. The HW vendor's device driver must guarantee
2378 * that this function returns NULL before the net device has finished
2379 * NETDEV_UNREGISTER state.
2381 struct net_device *(*get_netdev)(struct ib_device *device,
2384 * rdma netdev operation
2386 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2387 * must return -EOPNOTSUPP if it doesn't support the specified type.
2389 struct net_device *(*alloc_rdma_netdev)(
2390 struct ib_device *device, u32 port_num, enum rdma_netdev_t type,
2391 const char *name, unsigned char name_assign_type,
2392 void (*setup)(struct net_device *));
2394 int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num,
2395 enum rdma_netdev_t type,
2396 struct rdma_netdev_alloc_params *params);
2398 * query_gid should be return GID value for @device, when @port_num
2399 * link layer is either IB or iWarp. It is no-op if @port_num port
2400 * is RoCE link layer.
2402 int (*query_gid)(struct ib_device *device, u32 port_num, int index,
2405 * When calling add_gid, the HW vendor's driver should add the gid
2406 * of device of port at gid index available at @attr. Meta-info of
2407 * that gid (for example, the network device related to this gid) is
2408 * available at @attr. @context allows the HW vendor driver to store
2409 * extra information together with a GID entry. The HW vendor driver may
2410 * allocate memory to contain this information and store it in @context
2411 * when a new GID entry is written to. Params are consistent until the
2412 * next call of add_gid or delete_gid. The function should return 0 on
2413 * success or error otherwise. The function could be called
2414 * concurrently for different ports. This function is only called when
2415 * roce_gid_table is used.
2417 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2419 * When calling del_gid, the HW vendor's driver should delete the
2420 * gid of device @device at gid index gid_index of port port_num
2421 * available in @attr.
2422 * Upon the deletion of a GID entry, the HW vendor must free any
2423 * allocated memory. The caller will clear @context afterwards.
2424 * This function is only called when roce_gid_table is used.
2426 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2427 int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index,
2429 int (*alloc_ucontext)(struct ib_ucontext *context,
2430 struct ib_udata *udata);
2431 void (*dealloc_ucontext)(struct ib_ucontext *context);
2432 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2434 * This will be called once refcount of an entry in mmap_xa reaches
2435 * zero. The type of the memory that was mapped may differ between
2436 * entries and is opaque to the rdma_user_mmap interface.
2437 * Therefore needs to be implemented by the driver in mmap_free.
2439 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2440 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2441 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2442 int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2443 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2444 struct ib_udata *udata);
2445 int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2446 struct ib_udata *udata);
2447 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2448 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2449 int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2450 int (*create_srq)(struct ib_srq *srq,
2451 struct ib_srq_init_attr *srq_init_attr,
2452 struct ib_udata *udata);
2453 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2454 enum ib_srq_attr_mask srq_attr_mask,
2455 struct ib_udata *udata);
2456 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2457 int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2458 int (*create_qp)(struct ib_qp *qp, struct ib_qp_init_attr *qp_init_attr,
2459 struct ib_udata *udata);
2460 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2461 int qp_attr_mask, struct ib_udata *udata);
2462 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2463 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2464 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2465 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2466 struct ib_udata *udata);
2467 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2468 int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2469 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2470 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2471 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2472 u64 virt_addr, int mr_access_flags,
2473 struct ib_udata *udata);
2474 struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset,
2475 u64 length, u64 virt_addr, int fd,
2476 int mr_access_flags,
2477 struct ib_udata *udata);
2478 struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
2479 u64 length, u64 virt_addr,
2480 int mr_access_flags, struct ib_pd *pd,
2481 struct ib_udata *udata);
2482 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2483 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2485 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2486 u32 max_num_data_sg,
2487 u32 max_num_meta_sg);
2488 int (*advise_mr)(struct ib_pd *pd,
2489 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2490 struct ib_sge *sg_list, u32 num_sge,
2491 struct uverbs_attr_bundle *attrs);
2494 * Kernel users should universally support relaxed ordering (RO), as
2495 * they are designed to read data only after observing the CQE and use
2496 * the DMA API correctly.
2498 * Some drivers implicitly enable RO if platform supports it.
2500 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2501 unsigned int *sg_offset);
2502 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2503 struct ib_mr_status *mr_status);
2504 int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2505 int (*dealloc_mw)(struct ib_mw *mw);
2506 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2507 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2508 int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2509 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2510 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2511 struct ib_flow_attr *flow_attr,
2512 struct ib_udata *udata);
2513 int (*destroy_flow)(struct ib_flow *flow_id);
2514 int (*destroy_flow_action)(struct ib_flow_action *action);
2515 int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port,
2517 int (*get_vf_config)(struct ib_device *device, int vf, u32 port,
2518 struct ifla_vf_info *ivf);
2519 int (*get_vf_stats)(struct ib_device *device, int vf, u32 port,
2520 struct ifla_vf_stats *stats);
2521 int (*get_vf_guid)(struct ib_device *device, int vf, u32 port,
2522 struct ifla_vf_guid *node_guid,
2523 struct ifla_vf_guid *port_guid);
2524 int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid,
2526 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2527 struct ib_wq_init_attr *init_attr,
2528 struct ib_udata *udata);
2529 int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2530 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2531 u32 wq_attr_mask, struct ib_udata *udata);
2532 int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2533 struct ib_rwq_ind_table_init_attr *init_attr,
2534 struct ib_udata *udata);
2535 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2536 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2537 struct ib_ucontext *context,
2538 struct ib_dm_alloc_attr *attr,
2539 struct uverbs_attr_bundle *attrs);
2540 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2541 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2542 struct ib_dm_mr_attr *attr,
2543 struct uverbs_attr_bundle *attrs);
2544 int (*create_counters)(struct ib_counters *counters,
2545 struct uverbs_attr_bundle *attrs);
2546 int (*destroy_counters)(struct ib_counters *counters);
2547 int (*read_counters)(struct ib_counters *counters,
2548 struct ib_counters_read_attr *counters_read_attr,
2549 struct uverbs_attr_bundle *attrs);
2550 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2551 int data_sg_nents, unsigned int *data_sg_offset,
2552 struct scatterlist *meta_sg, int meta_sg_nents,
2553 unsigned int *meta_sg_offset);
2556 * alloc_hw_[device,port]_stats - Allocate a struct rdma_hw_stats and
2557 * fill in the driver initialized data. The struct is kfree()'ed by
2558 * the sysfs core when the device is removed. A lifespan of -1 in the
2559 * return struct tells the core to set a default lifespan.
2561 struct rdma_hw_stats *(*alloc_hw_device_stats)(struct ib_device *device);
2562 struct rdma_hw_stats *(*alloc_hw_port_stats)(struct ib_device *device,
2565 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2566 * @index - The index in the value array we wish to have updated, or
2567 * num_counters if we want all stats updated
2569 * < 0 - Error, no counters updated
2570 * index - Updated the single counter pointed to by index
2571 * num_counters - Updated all counters (will reset the timestamp
2572 * and prevent further calls for lifespan milliseconds)
2573 * Drivers are allowed to update all counters in leiu of just the
2574 * one given in index at their option
2576 int (*get_hw_stats)(struct ib_device *device,
2577 struct rdma_hw_stats *stats, u32 port, int index);
2580 * modify_hw_stat - Modify the counter configuration
2581 * @enable: true/false when enable/disable a counter
2582 * Return codes - 0 on success or error code otherwise.
2584 int (*modify_hw_stat)(struct ib_device *device, u32 port,
2585 unsigned int counter_index, bool enable);
2587 * Allows rdma drivers to add their own restrack attributes.
2589 int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2590 int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2591 int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2592 int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2593 int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2594 int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2595 int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2596 int (*fill_res_srq_entry)(struct sk_buff *msg, struct ib_srq *ib_srq);
2597 int (*fill_res_srq_entry_raw)(struct sk_buff *msg, struct ib_srq *ib_srq);
2599 /* Device lifecycle callbacks */
2601 * Called after the device becomes registered, before clients are
2604 int (*enable_driver)(struct ib_device *dev);
2606 * This is called as part of ib_dealloc_device().
2608 void (*dealloc_driver)(struct ib_device *dev);
2610 /* iWarp CM callbacks */
2611 void (*iw_add_ref)(struct ib_qp *qp);
2612 void (*iw_rem_ref)(struct ib_qp *qp);
2613 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2614 int (*iw_connect)(struct iw_cm_id *cm_id,
2615 struct iw_cm_conn_param *conn_param);
2616 int (*iw_accept)(struct iw_cm_id *cm_id,
2617 struct iw_cm_conn_param *conn_param);
2618 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2620 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2621 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2623 * counter_bind_qp - Bind a QP to a counter.
2624 * @counter - The counter to be bound. If counter->id is zero then
2625 * the driver needs to allocate a new counter and set counter->id
2627 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2629 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2630 * counter and bind it onto the default one
2632 int (*counter_unbind_qp)(struct ib_qp *qp);
2634 * counter_dealloc -De-allocate the hw counter
2636 int (*counter_dealloc)(struct rdma_counter *counter);
2638 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2639 * the driver initialized data.
2641 struct rdma_hw_stats *(*counter_alloc_stats)(
2642 struct rdma_counter *counter);
2644 * counter_update_stats - Query the stats value of this counter
2646 int (*counter_update_stats)(struct rdma_counter *counter);
2649 * Allows rdma drivers to add their own restrack attributes
2650 * dumped via 'rdma stat' iproute2 command.
2652 int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2654 /* query driver for its ucontext properties */
2655 int (*query_ucontext)(struct ib_ucontext *context,
2656 struct uverbs_attr_bundle *attrs);
2659 * Provide NUMA node. This API exists for rdmavt/hfi1 only.
2660 * Everyone else relies on Linux memory management model.
2662 int (*get_numa_node)(struct ib_device *dev);
2664 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2665 DECLARE_RDMA_OBJ_SIZE(ib_counters);
2666 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2667 DECLARE_RDMA_OBJ_SIZE(ib_mw);
2668 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2669 DECLARE_RDMA_OBJ_SIZE(ib_qp);
2670 DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2671 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2672 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2673 DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2676 struct ib_core_device {
2677 /* device must be the first element in structure until,
2678 * union of ib_core_device and device exists in ib_device.
2681 possible_net_t rdma_net;
2682 struct kobject *ports_kobj;
2683 struct list_head port_list;
2684 struct ib_device *owner; /* reach back to owner ib_device */
2687 struct rdma_restrack_root;
2689 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2690 struct device *dma_device;
2691 struct ib_device_ops ops;
2692 char name[IB_DEVICE_NAME_MAX];
2693 struct rcu_head rcu_head;
2695 struct list_head event_handler_list;
2696 /* Protects event_handler_list */
2697 struct rw_semaphore event_handler_rwsem;
2699 /* Protects QP's event_handler calls and open_qp list */
2700 spinlock_t qp_open_list_lock;
2702 struct rw_semaphore client_data_rwsem;
2703 struct xarray client_data;
2704 struct mutex unregistration_lock;
2706 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2707 rwlock_t cache_lock;
2709 * port_data is indexed by port number
2711 struct ib_port_data *port_data;
2713 int num_comp_vectors;
2717 struct ib_core_device coredev;
2720 /* First group is for device attributes,
2721 * Second group is for driver provided attributes (optional).
2722 * Third group is for the hw_stats
2723 * It is a NULL terminated array.
2725 const struct attribute_group *groups[4];
2727 u64 uverbs_cmd_mask;
2729 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2733 /* Indicates kernel verbs support, should not be used in drivers */
2734 u16 kverbs_provider:1;
2735 /* CQ adaptive moderation (RDMA DIM) */
2739 struct ib_device_attr attrs;
2740 struct hw_stats_device_data *hw_stats_data;
2742 #ifdef CONFIG_CGROUP_RDMA
2743 struct rdmacg_device cg_device;
2748 spinlock_t cq_pools_lock;
2749 struct list_head cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2751 struct rdma_restrack_root *res;
2753 const struct uapi_definition *driver_def;
2756 * Positive refcount indicates that the device is currently
2757 * registered and cannot be unregistered.
2759 refcount_t refcount;
2760 struct completion unreg_completion;
2761 struct work_struct unregistration_work;
2763 const struct rdma_link_ops *link_ops;
2765 /* Protects compat_devs xarray modifications */
2766 struct mutex compat_devs_mutex;
2767 /* Maintains compat devices for each net namespace */
2768 struct xarray compat_devs;
2770 /* Used by iWarp CM */
2771 char iw_ifname[IFNAMSIZ];
2772 u32 iw_driver_flags;
2776 static inline void *rdma_zalloc_obj(struct ib_device *dev, size_t size,
2777 gfp_t gfp, bool is_numa_aware)
2779 if (is_numa_aware && dev->ops.get_numa_node)
2780 return kzalloc_node(size, gfp, dev->ops.get_numa_node(dev));
2782 return kzalloc(size, gfp);
2785 struct ib_client_nl_info;
2788 int (*add)(struct ib_device *ibdev);
2789 void (*remove)(struct ib_device *, void *client_data);
2790 void (*rename)(struct ib_device *dev, void *client_data);
2791 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2792 struct ib_client_nl_info *res);
2793 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2795 /* Returns the net_dev belonging to this ib_client and matching the
2797 * @dev: An RDMA device that the net_dev use for communication.
2798 * @port: A physical port number on the RDMA device.
2799 * @pkey: P_Key that the net_dev uses if applicable.
2800 * @gid: A GID that the net_dev uses to communicate.
2801 * @addr: An IP address the net_dev is configured with.
2802 * @client_data: The device's client data set by ib_set_client_data().
2804 * An ib_client that implements a net_dev on top of RDMA devices
2805 * (such as IP over IB) should implement this callback, allowing the
2806 * rdma_cm module to find the right net_dev for a given request.
2808 * The caller is responsible for calling dev_put on the returned
2810 struct net_device *(*get_net_dev_by_params)(
2811 struct ib_device *dev,
2814 const union ib_gid *gid,
2815 const struct sockaddr *addr,
2819 struct completion uses_zero;
2822 /* kverbs are not required by the client */
2827 * IB block DMA iterator
2829 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2830 * to a HW supported page size.
2832 struct ib_block_iter {
2833 /* internal states */
2834 struct scatterlist *__sg; /* sg holding the current aligned block */
2835 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2836 size_t __sg_numblocks; /* ib_umem_num_dma_blocks() */
2837 unsigned int __sg_nents; /* number of SG entries */
2838 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2839 unsigned int __pg_bit; /* alignment of current block */
2842 struct ib_device *_ib_alloc_device(size_t size);
2843 #define ib_alloc_device(drv_struct, member) \
2844 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2845 BUILD_BUG_ON_ZERO(offsetof( \
2846 struct drv_struct, member))), \
2847 struct drv_struct, member)
2849 void ib_dealloc_device(struct ib_device *device);
2851 void ib_get_device_fw_str(struct ib_device *device, char *str);
2853 int ib_register_device(struct ib_device *device, const char *name,
2854 struct device *dma_device);
2855 void ib_unregister_device(struct ib_device *device);
2856 void ib_unregister_driver(enum rdma_driver_id driver_id);
2857 void ib_unregister_device_and_put(struct ib_device *device);
2858 void ib_unregister_device_queued(struct ib_device *ib_dev);
2860 int ib_register_client (struct ib_client *client);
2861 void ib_unregister_client(struct ib_client *client);
2863 void __rdma_block_iter_start(struct ib_block_iter *biter,
2864 struct scatterlist *sglist,
2866 unsigned long pgsz);
2867 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2870 * rdma_block_iter_dma_address - get the aligned dma address of the current
2871 * block held by the block iterator.
2872 * @biter: block iterator holding the memory block
2874 static inline dma_addr_t
2875 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2877 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2881 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2882 * @sglist: sglist to iterate over
2883 * @biter: block iterator holding the memory block
2884 * @nents: maximum number of sg entries to iterate over
2885 * @pgsz: best HW supported page size to use
2887 * Callers may use rdma_block_iter_dma_address() to get each
2888 * blocks aligned DMA address.
2890 #define rdma_for_each_block(sglist, biter, nents, pgsz) \
2891 for (__rdma_block_iter_start(biter, sglist, nents, \
2893 __rdma_block_iter_next(biter);)
2896 * ib_get_client_data - Get IB client context
2897 * @device:Device to get context for
2898 * @client:Client to get context for
2900 * ib_get_client_data() returns the client context data set with
2901 * ib_set_client_data(). This can only be called while the client is
2902 * registered to the device, once the ib_client remove() callback returns this
2905 static inline void *ib_get_client_data(struct ib_device *device,
2906 struct ib_client *client)
2908 return xa_load(&device->client_data, client->client_id);
2910 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2912 void ib_set_device_ops(struct ib_device *device,
2913 const struct ib_device_ops *ops);
2915 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2916 unsigned long pfn, unsigned long size, pgprot_t prot,
2917 struct rdma_user_mmap_entry *entry);
2918 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2919 struct rdma_user_mmap_entry *entry,
2921 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2922 struct rdma_user_mmap_entry *entry,
2923 size_t length, u32 min_pgoff,
2927 rdma_user_mmap_entry_insert_exact(struct ib_ucontext *ucontext,
2928 struct rdma_user_mmap_entry *entry,
2929 size_t length, u32 pgoff)
2931 return rdma_user_mmap_entry_insert_range(ucontext, entry, length, pgoff,
2935 struct rdma_user_mmap_entry *
2936 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2937 unsigned long pgoff);
2938 struct rdma_user_mmap_entry *
2939 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2940 struct vm_area_struct *vma);
2941 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2943 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2945 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2947 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2950 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2952 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2955 static inline bool ib_is_buffer_cleared(const void __user *p,
2961 if (len > USHRT_MAX)
2964 buf = memdup_user(p, len);
2968 ret = !memchr_inv(buf, 0, len);
2973 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2977 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2981 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2982 * contains all required attributes and no attributes not allowed for
2983 * the given QP state transition.
2984 * @cur_state: Current QP state
2985 * @next_state: Next QP state
2987 * @mask: Mask of supplied QP attributes
2989 * This function is a helper function that a low-level driver's
2990 * modify_qp method can use to validate the consumer's input. It
2991 * checks that cur_state and next_state are valid QP states, that a
2992 * transition from cur_state to next_state is allowed by the IB spec,
2993 * and that the attribute mask supplied is allowed for the transition.
2995 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2996 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2998 void ib_register_event_handler(struct ib_event_handler *event_handler);
2999 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
3000 void ib_dispatch_event(const struct ib_event *event);
3002 int ib_query_port(struct ib_device *device,
3003 u32 port_num, struct ib_port_attr *port_attr);
3005 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
3009 * rdma_cap_ib_switch - Check if the device is IB switch
3010 * @device: Device to check
3012 * Device driver is responsible for setting is_switch bit on
3013 * in ib_device structure at init time.
3015 * Return: true if the device is IB switch.
3017 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
3019 return device->is_switch;
3023 * rdma_start_port - Return the first valid port number for the device
3026 * @device: Device to be checked
3028 * Return start port number
3030 static inline u32 rdma_start_port(const struct ib_device *device)
3032 return rdma_cap_ib_switch(device) ? 0 : 1;
3036 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3037 * @device - The struct ib_device * to iterate over
3038 * @iter - The unsigned int to store the port number
3040 #define rdma_for_each_port(device, iter) \
3041 for (iter = rdma_start_port(device + \
3042 BUILD_BUG_ON_ZERO(!__same_type(u32, \
3044 iter <= rdma_end_port(device); iter++)
3047 * rdma_end_port - Return the last valid port number for the device
3050 * @device: Device to be checked
3052 * Return last port number
3054 static inline u32 rdma_end_port(const struct ib_device *device)
3056 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3059 static inline int rdma_is_port_valid(const struct ib_device *device,
3062 return (port >= rdma_start_port(device) &&
3063 port <= rdma_end_port(device));
3066 static inline bool rdma_is_grh_required(const struct ib_device *device,
3069 return device->port_data[port_num].immutable.core_cap_flags &
3070 RDMA_CORE_PORT_IB_GRH_REQUIRED;
3073 static inline bool rdma_protocol_ib(const struct ib_device *device,
3076 return device->port_data[port_num].immutable.core_cap_flags &
3077 RDMA_CORE_CAP_PROT_IB;
3080 static inline bool rdma_protocol_roce(const struct ib_device *device,
3083 return device->port_data[port_num].immutable.core_cap_flags &
3084 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3087 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device,
3090 return device->port_data[port_num].immutable.core_cap_flags &
3091 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3094 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device,
3097 return device->port_data[port_num].immutable.core_cap_flags &
3098 RDMA_CORE_CAP_PROT_ROCE;
3101 static inline bool rdma_protocol_iwarp(const struct ib_device *device,
3104 return device->port_data[port_num].immutable.core_cap_flags &
3105 RDMA_CORE_CAP_PROT_IWARP;
3108 static inline bool rdma_ib_or_roce(const struct ib_device *device,
3111 return rdma_protocol_ib(device, port_num) ||
3112 rdma_protocol_roce(device, port_num);
3115 static inline bool rdma_protocol_raw_packet(const struct ib_device *device,
3118 return device->port_data[port_num].immutable.core_cap_flags &
3119 RDMA_CORE_CAP_PROT_RAW_PACKET;
3122 static inline bool rdma_protocol_usnic(const struct ib_device *device,
3125 return device->port_data[port_num].immutable.core_cap_flags &
3126 RDMA_CORE_CAP_PROT_USNIC;
3130 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3131 * Management Datagrams.
3132 * @device: Device to check
3133 * @port_num: Port number to check
3135 * Management Datagrams (MAD) are a required part of the InfiniBand
3136 * specification and are supported on all InfiniBand devices. A slightly
3137 * extended version are also supported on OPA interfaces.
3139 * Return: true if the port supports sending/receiving of MAD packets.
3141 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num)
3143 return device->port_data[port_num].immutable.core_cap_flags &
3144 RDMA_CORE_CAP_IB_MAD;
3148 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3149 * Management Datagrams.
3150 * @device: Device to check
3151 * @port_num: Port number to check
3153 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3154 * datagrams with their own versions. These OPA MADs share many but not all of
3155 * the characteristics of InfiniBand MADs.
3157 * OPA MADs differ in the following ways:
3159 * 1) MADs are variable size up to 2K
3160 * IBTA defined MADs remain fixed at 256 bytes
3161 * 2) OPA SMPs must carry valid PKeys
3162 * 3) OPA SMP packets are a different format
3164 * Return: true if the port supports OPA MAD packet formats.
3166 static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num)
3168 return device->port_data[port_num].immutable.core_cap_flags &
3169 RDMA_CORE_CAP_OPA_MAD;
3173 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3174 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3175 * @device: Device to check
3176 * @port_num: Port number to check
3178 * Each InfiniBand node is required to provide a Subnet Management Agent
3179 * that the subnet manager can access. Prior to the fabric being fully
3180 * configured by the subnet manager, the SMA is accessed via a well known
3181 * interface called the Subnet Management Interface (SMI). This interface
3182 * uses directed route packets to communicate with the SM to get around the
3183 * chicken and egg problem of the SM needing to know what's on the fabric
3184 * in order to configure the fabric, and needing to configure the fabric in
3185 * order to send packets to the devices on the fabric. These directed
3186 * route packets do not need the fabric fully configured in order to reach
3187 * their destination. The SMI is the only method allowed to send
3188 * directed route packets on an InfiniBand fabric.
3190 * Return: true if the port provides an SMI.
3192 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num)
3194 return device->port_data[port_num].immutable.core_cap_flags &
3195 RDMA_CORE_CAP_IB_SMI;
3199 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3200 * Communication Manager.
3201 * @device: Device to check
3202 * @port_num: Port number to check
3204 * The InfiniBand Communication Manager is one of many pre-defined General
3205 * Service Agents (GSA) that are accessed via the General Service
3206 * Interface (GSI). It's role is to facilitate establishment of connections
3207 * between nodes as well as other management related tasks for established
3210 * Return: true if the port supports an IB CM (this does not guarantee that
3211 * a CM is actually running however).
3213 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num)
3215 return device->port_data[port_num].immutable.core_cap_flags &
3216 RDMA_CORE_CAP_IB_CM;
3220 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3221 * Communication Manager.
3222 * @device: Device to check
3223 * @port_num: Port number to check
3225 * Similar to above, but specific to iWARP connections which have a different
3226 * managment protocol than InfiniBand.
3228 * Return: true if the port supports an iWARP CM (this does not guarantee that
3229 * a CM is actually running however).
3231 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num)
3233 return device->port_data[port_num].immutable.core_cap_flags &
3234 RDMA_CORE_CAP_IW_CM;
3238 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3239 * Subnet Administration.
3240 * @device: Device to check
3241 * @port_num: Port number to check
3243 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3244 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3245 * fabrics, devices should resolve routes to other hosts by contacting the
3246 * SA to query the proper route.
3248 * Return: true if the port should act as a client to the fabric Subnet
3249 * Administration interface. This does not imply that the SA service is
3252 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num)
3254 return device->port_data[port_num].immutable.core_cap_flags &
3255 RDMA_CORE_CAP_IB_SA;
3259 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3261 * @device: Device to check
3262 * @port_num: Port number to check
3264 * InfiniBand multicast registration is more complex than normal IPv4 or
3265 * IPv6 multicast registration. Each Host Channel Adapter must register
3266 * with the Subnet Manager when it wishes to join a multicast group. It
3267 * should do so only once regardless of how many queue pairs it subscribes
3268 * to this group. And it should leave the group only after all queue pairs
3269 * attached to the group have been detached.
3271 * Return: true if the port must undertake the additional adminstrative
3272 * overhead of registering/unregistering with the SM and tracking of the
3273 * total number of queue pairs attached to the multicast group.
3275 static inline bool rdma_cap_ib_mcast(const struct ib_device *device,
3278 return rdma_cap_ib_sa(device, port_num);
3282 * rdma_cap_af_ib - Check if the port of device has the capability
3283 * Native Infiniband Address.
3284 * @device: Device to check
3285 * @port_num: Port number to check
3287 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3288 * GID. RoCE uses a different mechanism, but still generates a GID via
3289 * a prescribed mechanism and port specific data.
3291 * Return: true if the port uses a GID address to identify devices on the
3294 static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num)
3296 return device->port_data[port_num].immutable.core_cap_flags &
3297 RDMA_CORE_CAP_AF_IB;
3301 * rdma_cap_eth_ah - Check if the port of device has the capability
3302 * Ethernet Address Handle.
3303 * @device: Device to check
3304 * @port_num: Port number to check
3306 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3307 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3308 * port. Normally, packet headers are generated by the sending host
3309 * adapter, but when sending connectionless datagrams, we must manually
3310 * inject the proper headers for the fabric we are communicating over.
3312 * Return: true if we are running as a RoCE port and must force the
3313 * addition of a Global Route Header built from our Ethernet Address
3314 * Handle into our header list for connectionless packets.
3316 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num)
3318 return device->port_data[port_num].immutable.core_cap_flags &
3319 RDMA_CORE_CAP_ETH_AH;
3323 * rdma_cap_opa_ah - Check if the port of device supports
3324 * OPA Address handles
3325 * @device: Device to check
3326 * @port_num: Port number to check
3328 * Return: true if we are running on an OPA device which supports
3329 * the extended OPA addressing.
3331 static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num)
3333 return (device->port_data[port_num].immutable.core_cap_flags &
3334 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3338 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3341 * @port_num: Port number
3343 * This MAD size includes the MAD headers and MAD payload. No other headers
3346 * Return the max MAD size required by the Port. Will return 0 if the port
3347 * does not support MADs
3349 static inline size_t rdma_max_mad_size(const struct ib_device *device,
3352 return device->port_data[port_num].immutable.max_mad_size;
3356 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3357 * @device: Device to check
3358 * @port_num: Port number to check
3360 * RoCE GID table mechanism manages the various GIDs for a device.
3362 * NOTE: if allocating the port's GID table has failed, this call will still
3363 * return true, but any RoCE GID table API will fail.
3365 * Return: true if the port uses RoCE GID table mechanism in order to manage
3368 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3371 return rdma_protocol_roce(device, port_num) &&
3372 device->ops.add_gid && device->ops.del_gid;
3376 * Check if the device supports READ W/ INVALIDATE.
3378 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3381 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3382 * has support for it yet.
3384 return rdma_protocol_iwarp(dev, port_num);
3388 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3390 * @port_num: 1 based Port number
3392 * Return true if port is an Intel OPA port , false if not
3394 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3397 return (device->port_data[port_num].immutable.core_cap_flags &
3398 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3402 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3404 * @port_num: Port number
3405 * @mtu: enum value of MTU
3407 * Return the MTU size supported by the port as an integer value. Will return
3408 * -1 if enum value of mtu is not supported.
3410 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port,
3413 if (rdma_core_cap_opa_port(device, port))
3414 return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3416 return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3420 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3422 * @port_num: Port number
3423 * @attr: port attribute
3425 * Return the MTU size supported by the port as an integer value.
3427 static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port,
3428 struct ib_port_attr *attr)
3430 if (rdma_core_cap_opa_port(device, port))
3431 return attr->phys_mtu;
3433 return ib_mtu_enum_to_int(attr->max_mtu);
3436 int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port,
3438 int ib_get_vf_config(struct ib_device *device, int vf, u32 port,
3439 struct ifla_vf_info *info);
3440 int ib_get_vf_stats(struct ib_device *device, int vf, u32 port,
3441 struct ifla_vf_stats *stats);
3442 int ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
3443 struct ifla_vf_guid *node_guid,
3444 struct ifla_vf_guid *port_guid);
3445 int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid,
3448 int ib_query_pkey(struct ib_device *device,
3449 u32 port_num, u16 index, u16 *pkey);
3451 int ib_modify_device(struct ib_device *device,
3452 int device_modify_mask,
3453 struct ib_device_modify *device_modify);
3455 int ib_modify_port(struct ib_device *device,
3456 u32 port_num, int port_modify_mask,
3457 struct ib_port_modify *port_modify);
3459 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3460 u32 *port_num, u16 *index);
3462 int ib_find_pkey(struct ib_device *device,
3463 u32 port_num, u16 pkey, u16 *index);
3467 * Create a memory registration for all memory in the system and place
3468 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3469 * ULPs to avoid the overhead of dynamic MRs.
3471 * This flag is generally considered unsafe and must only be used in
3472 * extremly trusted environments. Every use of it will log a warning
3473 * in the kernel log.
3475 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3478 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3479 const char *caller);
3482 * ib_alloc_pd - Allocates an unused protection domain.
3483 * @device: The device on which to allocate the protection domain.
3484 * @flags: protection domain flags
3486 * A protection domain object provides an association between QPs, shared
3487 * receive queues, address handles, memory regions, and memory windows.
3489 * Every PD has a local_dma_lkey which can be used as the lkey value for local
3490 * memory operations.
3492 #define ib_alloc_pd(device, flags) \
3493 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3495 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3498 * ib_dealloc_pd - Deallocate kernel PD
3499 * @pd: The protection domain
3501 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3503 static inline void ib_dealloc_pd(struct ib_pd *pd)
3505 int ret = ib_dealloc_pd_user(pd, NULL);
3507 WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3510 enum rdma_create_ah_flags {
3511 /* In a sleepable context */
3512 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3516 * rdma_create_ah - Creates an address handle for the given address vector.
3517 * @pd: The protection domain associated with the address handle.
3518 * @ah_attr: The attributes of the address vector.
3519 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3521 * The address handle is used to reference a local or global destination
3522 * in all UD QP post sends.
3524 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3528 * rdma_create_user_ah - Creates an address handle for the given address vector.
3529 * It resolves destination mac address for ah attribute of RoCE type.
3530 * @pd: The protection domain associated with the address handle.
3531 * @ah_attr: The attributes of the address vector.
3532 * @udata: pointer to user's input output buffer information need by
3535 * It returns 0 on success and returns appropriate error code on error.
3536 * The address handle is used to reference a local or global destination
3537 * in all UD QP post sends.
3539 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3540 struct rdma_ah_attr *ah_attr,
3541 struct ib_udata *udata);
3543 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3545 * @hdr: the L3 header to parse
3546 * @net_type: type of header to parse
3547 * @sgid: place to store source gid
3548 * @dgid: place to store destination gid
3550 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3551 enum rdma_network_type net_type,
3552 union ib_gid *sgid, union ib_gid *dgid);
3555 * ib_get_rdma_header_version - Get the header version
3556 * @hdr: the L3 header to parse
3558 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3561 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3563 * @device: Device on which the received message arrived.
3564 * @port_num: Port on which the received message arrived.
3565 * @wc: Work completion associated with the received message.
3566 * @grh: References the received global route header. This parameter is
3567 * ignored unless the work completion indicates that the GRH is valid.
3568 * @ah_attr: Returned attributes that can be used when creating an address
3569 * handle for replying to the message.
3570 * When ib_init_ah_attr_from_wc() returns success,
3571 * (a) for IB link layer it optionally contains a reference to SGID attribute
3572 * when GRH is present for IB link layer.
3573 * (b) for RoCE link layer it contains a reference to SGID attribute.
3574 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3575 * attributes which are initialized using ib_init_ah_attr_from_wc().
3578 int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num,
3579 const struct ib_wc *wc, const struct ib_grh *grh,
3580 struct rdma_ah_attr *ah_attr);
3583 * ib_create_ah_from_wc - Creates an address handle associated with the
3584 * sender of the specified work completion.
3585 * @pd: The protection domain associated with the address handle.
3586 * @wc: Work completion information associated with a received message.
3587 * @grh: References the received global route header. This parameter is
3588 * ignored unless the work completion indicates that the GRH is valid.
3589 * @port_num: The outbound port number to associate with the address.
3591 * The address handle is used to reference a local or global destination
3592 * in all UD QP post sends.
3594 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3595 const struct ib_grh *grh, u32 port_num);
3598 * rdma_modify_ah - Modifies the address vector associated with an address
3600 * @ah: The address handle to modify.
3601 * @ah_attr: The new address vector attributes to associate with the
3604 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3607 * rdma_query_ah - Queries the address vector associated with an address
3609 * @ah: The address handle to query.
3610 * @ah_attr: The address vector attributes associated with the address
3613 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3615 enum rdma_destroy_ah_flags {
3616 /* In a sleepable context */
3617 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3621 * rdma_destroy_ah_user - Destroys an address handle.
3622 * @ah: The address handle to destroy.
3623 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3624 * @udata: Valid user data or NULL for kernel objects
3626 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3629 * rdma_destroy_ah - Destroys an kernel address handle.
3630 * @ah: The address handle to destroy.
3631 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3633 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3635 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3637 int ret = rdma_destroy_ah_user(ah, flags, NULL);
3639 WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3642 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3643 struct ib_srq_init_attr *srq_init_attr,
3644 struct ib_usrq_object *uobject,
3645 struct ib_udata *udata);
3646 static inline struct ib_srq *
3647 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3649 if (!pd->device->ops.create_srq)
3650 return ERR_PTR(-EOPNOTSUPP);
3652 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3656 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3657 * @srq: The SRQ to modify.
3658 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3659 * the current values of selected SRQ attributes are returned.
3660 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3661 * are being modified.
3663 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3664 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3665 * the number of receives queued drops below the limit.
3667 int ib_modify_srq(struct ib_srq *srq,
3668 struct ib_srq_attr *srq_attr,
3669 enum ib_srq_attr_mask srq_attr_mask);
3672 * ib_query_srq - Returns the attribute list and current values for the
3674 * @srq: The SRQ to query.
3675 * @srq_attr: The attributes of the specified SRQ.
3677 int ib_query_srq(struct ib_srq *srq,
3678 struct ib_srq_attr *srq_attr);
3681 * ib_destroy_srq_user - Destroys the specified SRQ.
3682 * @srq: The SRQ to destroy.
3683 * @udata: Valid user data or NULL for kernel objects
3685 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3688 * ib_destroy_srq - Destroys the specified kernel SRQ.
3689 * @srq: The SRQ to destroy.
3691 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3693 static inline void ib_destroy_srq(struct ib_srq *srq)
3695 int ret = ib_destroy_srq_user(srq, NULL);
3697 WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3701 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3702 * @srq: The SRQ to post the work request on.
3703 * @recv_wr: A list of work requests to post on the receive queue.
3704 * @bad_recv_wr: On an immediate failure, this parameter will reference
3705 * the work request that failed to be posted on the QP.
3707 static inline int ib_post_srq_recv(struct ib_srq *srq,
3708 const struct ib_recv_wr *recv_wr,
3709 const struct ib_recv_wr **bad_recv_wr)
3711 const struct ib_recv_wr *dummy;
3713 return srq->device->ops.post_srq_recv(srq, recv_wr,
3714 bad_recv_wr ? : &dummy);
3717 struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd,
3718 struct ib_qp_init_attr *qp_init_attr,
3719 const char *caller);
3721 * ib_create_qp - Creates a kernel QP associated with the specific protection
3723 * @pd: The protection domain associated with the QP.
3724 * @init_attr: A list of initial attributes required to create the
3725 * QP. If QP creation succeeds, then the attributes are updated to
3726 * the actual capabilities of the created QP.
3728 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3729 struct ib_qp_init_attr *init_attr)
3731 return ib_create_qp_kernel(pd, init_attr, KBUILD_MODNAME);
3735 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3736 * @qp: The QP to modify.
3737 * @attr: On input, specifies the QP attributes to modify. On output,
3738 * the current values of selected QP attributes are returned.
3739 * @attr_mask: A bit-mask used to specify which attributes of the QP
3740 * are being modified.
3741 * @udata: pointer to user's input output buffer information
3742 * are being modified.
3743 * It returns 0 on success and returns appropriate error code on error.
3745 int ib_modify_qp_with_udata(struct ib_qp *qp,
3746 struct ib_qp_attr *attr,
3748 struct ib_udata *udata);
3751 * ib_modify_qp - Modifies the attributes for the specified QP and then
3752 * transitions the QP to the given state.
3753 * @qp: The QP to modify.
3754 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3755 * the current values of selected QP attributes are returned.
3756 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3757 * are being modified.
3759 int ib_modify_qp(struct ib_qp *qp,
3760 struct ib_qp_attr *qp_attr,
3764 * ib_query_qp - Returns the attribute list and current values for the
3766 * @qp: The QP to query.
3767 * @qp_attr: The attributes of the specified QP.
3768 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3769 * @qp_init_attr: Additional attributes of the selected QP.
3771 * The qp_attr_mask may be used to limit the query to gathering only the
3772 * selected attributes.
3774 int ib_query_qp(struct ib_qp *qp,
3775 struct ib_qp_attr *qp_attr,
3777 struct ib_qp_init_attr *qp_init_attr);
3780 * ib_destroy_qp - Destroys the specified QP.
3781 * @qp: The QP to destroy.
3782 * @udata: Valid udata or NULL for kernel objects
3784 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3787 * ib_destroy_qp - Destroys the specified kernel QP.
3788 * @qp: The QP to destroy.
3790 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3792 static inline int ib_destroy_qp(struct ib_qp *qp)
3794 return ib_destroy_qp_user(qp, NULL);
3798 * ib_open_qp - Obtain a reference to an existing sharable QP.
3799 * @xrcd - XRC domain
3800 * @qp_open_attr: Attributes identifying the QP to open.
3802 * Returns a reference to a sharable QP.
3804 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3805 struct ib_qp_open_attr *qp_open_attr);
3808 * ib_close_qp - Release an external reference to a QP.
3809 * @qp: The QP handle to release
3811 * The opened QP handle is released by the caller. The underlying
3812 * shared QP is not destroyed until all internal references are released.
3814 int ib_close_qp(struct ib_qp *qp);
3817 * ib_post_send - Posts a list of work requests to the send queue of
3819 * @qp: The QP to post the work request on.
3820 * @send_wr: A list of work requests to post on the send queue.
3821 * @bad_send_wr: On an immediate failure, this parameter will reference
3822 * the work request that failed to be posted on the QP.
3824 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3825 * error is returned, the QP state shall not be affected,
3826 * ib_post_send() will return an immediate error after queueing any
3827 * earlier work requests in the list.
3829 static inline int ib_post_send(struct ib_qp *qp,
3830 const struct ib_send_wr *send_wr,
3831 const struct ib_send_wr **bad_send_wr)
3833 const struct ib_send_wr *dummy;
3835 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3839 * ib_post_recv - Posts a list of work requests to the receive queue of
3841 * @qp: The QP to post the work request on.
3842 * @recv_wr: A list of work requests to post on the receive queue.
3843 * @bad_recv_wr: On an immediate failure, this parameter will reference
3844 * the work request that failed to be posted on the QP.
3846 static inline int ib_post_recv(struct ib_qp *qp,
3847 const struct ib_recv_wr *recv_wr,
3848 const struct ib_recv_wr **bad_recv_wr)
3850 const struct ib_recv_wr *dummy;
3852 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3855 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3856 int comp_vector, enum ib_poll_context poll_ctx,
3857 const char *caller);
3858 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3859 int nr_cqe, int comp_vector,
3860 enum ib_poll_context poll_ctx)
3862 return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3866 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3867 int nr_cqe, enum ib_poll_context poll_ctx,
3868 const char *caller);
3871 * ib_alloc_cq_any: Allocate kernel CQ
3872 * @dev: The IB device
3873 * @private: Private data attached to the CQE
3874 * @nr_cqe: Number of CQEs in the CQ
3875 * @poll_ctx: Context used for polling the CQ
3877 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3878 void *private, int nr_cqe,
3879 enum ib_poll_context poll_ctx)
3881 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3885 void ib_free_cq(struct ib_cq *cq);
3886 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3889 * ib_create_cq - Creates a CQ on the specified device.
3890 * @device: The device on which to create the CQ.
3891 * @comp_handler: A user-specified callback that is invoked when a
3892 * completion event occurs on the CQ.
3893 * @event_handler: A user-specified callback that is invoked when an
3894 * asynchronous event not associated with a completion occurs on the CQ.
3895 * @cq_context: Context associated with the CQ returned to the user via
3896 * the associated completion and event handlers.
3897 * @cq_attr: The attributes the CQ should be created upon.
3899 * Users can examine the cq structure to determine the actual CQ size.
3901 struct ib_cq *__ib_create_cq(struct ib_device *device,
3902 ib_comp_handler comp_handler,
3903 void (*event_handler)(struct ib_event *, void *),
3905 const struct ib_cq_init_attr *cq_attr,
3906 const char *caller);
3907 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3908 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3911 * ib_resize_cq - Modifies the capacity of the CQ.
3912 * @cq: The CQ to resize.
3913 * @cqe: The minimum size of the CQ.
3915 * Users can examine the cq structure to determine the actual CQ size.
3917 int ib_resize_cq(struct ib_cq *cq, int cqe);
3920 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3921 * @cq: The CQ to modify.
3922 * @cq_count: number of CQEs that will trigger an event
3923 * @cq_period: max period of time in usec before triggering an event
3926 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3929 * ib_destroy_cq_user - Destroys the specified CQ.
3930 * @cq: The CQ to destroy.
3931 * @udata: Valid user data or NULL for kernel objects
3933 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3936 * ib_destroy_cq - Destroys the specified kernel CQ.
3937 * @cq: The CQ to destroy.
3939 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3941 static inline void ib_destroy_cq(struct ib_cq *cq)
3943 int ret = ib_destroy_cq_user(cq, NULL);
3945 WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
3949 * ib_poll_cq - poll a CQ for completion(s)
3950 * @cq:the CQ being polled
3951 * @num_entries:maximum number of completions to return
3952 * @wc:array of at least @num_entries &struct ib_wc where completions
3955 * Poll a CQ for (possibly multiple) completions. If the return value
3956 * is < 0, an error occurred. If the return value is >= 0, it is the
3957 * number of completions returned. If the return value is
3958 * non-negative and < num_entries, then the CQ was emptied.
3960 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3963 return cq->device->ops.poll_cq(cq, num_entries, wc);
3967 * ib_req_notify_cq - Request completion notification on a CQ.
3968 * @cq: The CQ to generate an event for.
3970 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3971 * to request an event on the next solicited event or next work
3972 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3973 * may also be |ed in to request a hint about missed events, as
3977 * < 0 means an error occurred while requesting notification
3978 * == 0 means notification was requested successfully, and if
3979 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3980 * were missed and it is safe to wait for another event. In
3981 * this case is it guaranteed that any work completions added
3982 * to the CQ since the last CQ poll will trigger a completion
3983 * notification event.
3984 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3985 * in. It means that the consumer must poll the CQ again to
3986 * make sure it is empty to avoid missing an event because of a
3987 * race between requesting notification and an entry being
3988 * added to the CQ. This return value means it is possible
3989 * (but not guaranteed) that a work completion has been added
3990 * to the CQ since the last poll without triggering a
3991 * completion notification event.
3993 static inline int ib_req_notify_cq(struct ib_cq *cq,
3994 enum ib_cq_notify_flags flags)
3996 return cq->device->ops.req_notify_cq(cq, flags);
3999 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
4000 int comp_vector_hint,
4001 enum ib_poll_context poll_ctx);
4003 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
4006 * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
4007 * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
4008 * address into the dma address.
4010 static inline bool ib_uses_virt_dma(struct ib_device *dev)
4012 return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
4016 * Check if a IB device's underlying DMA mapping supports P2PDMA transfers.
4018 static inline bool ib_dma_pci_p2p_dma_supported(struct ib_device *dev)
4020 if (ib_uses_virt_dma(dev))
4023 return dma_pci_p2pdma_supported(dev->dma_device);
4027 * ib_virt_dma_to_ptr - Convert a dma_addr to a kernel pointer
4028 * @dma_addr: The DMA address
4030 * Used by ib_uses_virt_dma() devices to get back to the kernel pointer after
4031 * going through the dma_addr marshalling.
4033 static inline void *ib_virt_dma_to_ptr(u64 dma_addr)
4035 /* virt_dma mode maps the kvs's directly into the dma addr */
4036 return (void *)(uintptr_t)dma_addr;
4040 * ib_virt_dma_to_page - Convert a dma_addr to a struct page
4041 * @dma_addr: The DMA address
4043 * Used by ib_uses_virt_dma() device to get back to the struct page after going
4044 * through the dma_addr marshalling.
4046 static inline struct page *ib_virt_dma_to_page(u64 dma_addr)
4048 return virt_to_page(ib_virt_dma_to_ptr(dma_addr));
4052 * ib_dma_mapping_error - check a DMA addr for error
4053 * @dev: The device for which the dma_addr was created
4054 * @dma_addr: The DMA address to check
4056 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
4058 if (ib_uses_virt_dma(dev))
4060 return dma_mapping_error(dev->dma_device, dma_addr);
4064 * ib_dma_map_single - Map a kernel virtual address to DMA address
4065 * @dev: The device for which the dma_addr is to be created
4066 * @cpu_addr: The kernel virtual address
4067 * @size: The size of the region in bytes
4068 * @direction: The direction of the DMA
4070 static inline u64 ib_dma_map_single(struct ib_device *dev,
4071 void *cpu_addr, size_t size,
4072 enum dma_data_direction direction)
4074 if (ib_uses_virt_dma(dev))
4075 return (uintptr_t)cpu_addr;
4076 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
4080 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4081 * @dev: The device for which the DMA address was created
4082 * @addr: The DMA address
4083 * @size: The size of the region in bytes
4084 * @direction: The direction of the DMA
4086 static inline void ib_dma_unmap_single(struct ib_device *dev,
4087 u64 addr, size_t size,
4088 enum dma_data_direction direction)
4090 if (!ib_uses_virt_dma(dev))
4091 dma_unmap_single(dev->dma_device, addr, size, direction);
4095 * ib_dma_map_page - Map a physical page to DMA address
4096 * @dev: The device for which the dma_addr is to be created
4097 * @page: The page to be mapped
4098 * @offset: The offset within the page
4099 * @size: The size of the region in bytes
4100 * @direction: The direction of the DMA
4102 static inline u64 ib_dma_map_page(struct ib_device *dev,
4104 unsigned long offset,
4106 enum dma_data_direction direction)
4108 if (ib_uses_virt_dma(dev))
4109 return (uintptr_t)(page_address(page) + offset);
4110 return dma_map_page(dev->dma_device, page, offset, size, direction);
4114 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4115 * @dev: The device for which the DMA address was created
4116 * @addr: The DMA address
4117 * @size: The size of the region in bytes
4118 * @direction: The direction of the DMA
4120 static inline void ib_dma_unmap_page(struct ib_device *dev,
4121 u64 addr, size_t size,
4122 enum dma_data_direction direction)
4124 if (!ib_uses_virt_dma(dev))
4125 dma_unmap_page(dev->dma_device, addr, size, direction);
4128 int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4129 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4130 struct scatterlist *sg, int nents,
4131 enum dma_data_direction direction,
4132 unsigned long dma_attrs)
4134 if (ib_uses_virt_dma(dev))
4135 return ib_dma_virt_map_sg(dev, sg, nents);
4136 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4140 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4141 struct scatterlist *sg, int nents,
4142 enum dma_data_direction direction,
4143 unsigned long dma_attrs)
4145 if (!ib_uses_virt_dma(dev))
4146 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4151 * ib_dma_map_sgtable_attrs - Map a scatter/gather table to DMA addresses
4152 * @dev: The device for which the DMA addresses are to be created
4153 * @sg: The sg_table object describing the buffer
4154 * @direction: The direction of the DMA
4155 * @attrs: Optional DMA attributes for the map operation
4157 static inline int ib_dma_map_sgtable_attrs(struct ib_device *dev,
4158 struct sg_table *sgt,
4159 enum dma_data_direction direction,
4160 unsigned long dma_attrs)
4164 if (ib_uses_virt_dma(dev)) {
4165 nents = ib_dma_virt_map_sg(dev, sgt->sgl, sgt->orig_nents);
4171 return dma_map_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4174 static inline void ib_dma_unmap_sgtable_attrs(struct ib_device *dev,
4175 struct sg_table *sgt,
4176 enum dma_data_direction direction,
4177 unsigned long dma_attrs)
4179 if (!ib_uses_virt_dma(dev))
4180 dma_unmap_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4184 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4185 * @dev: The device for which the DMA addresses are to be created
4186 * @sg: The array of scatter/gather entries
4187 * @nents: The number of scatter/gather entries
4188 * @direction: The direction of the DMA
4190 static inline int ib_dma_map_sg(struct ib_device *dev,
4191 struct scatterlist *sg, int nents,
4192 enum dma_data_direction direction)
4194 return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4198 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4199 * @dev: The device for which the DMA addresses were created
4200 * @sg: The array of scatter/gather entries
4201 * @nents: The number of scatter/gather entries
4202 * @direction: The direction of the DMA
4204 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4205 struct scatterlist *sg, int nents,
4206 enum dma_data_direction direction)
4208 ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4212 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4213 * @dev: The device to query
4215 * The returned value represents a size in bytes.
4217 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4219 if (ib_uses_virt_dma(dev))
4221 return dma_get_max_seg_size(dev->dma_device);
4225 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4226 * @dev: The device for which the DMA address was created
4227 * @addr: The DMA address
4228 * @size: The size of the region in bytes
4229 * @dir: The direction of the DMA
4231 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4234 enum dma_data_direction dir)
4236 if (!ib_uses_virt_dma(dev))
4237 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4241 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4242 * @dev: The device for which the DMA address was created
4243 * @addr: The DMA address
4244 * @size: The size of the region in bytes
4245 * @dir: The direction of the DMA
4247 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4250 enum dma_data_direction dir)
4252 if (!ib_uses_virt_dma(dev))
4253 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4256 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4257 * space. This function should be called when 'current' is the owning MM.
4259 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4260 u64 virt_addr, int mr_access_flags);
4262 /* ib_advise_mr - give an advice about an address range in a memory region */
4263 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4264 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4266 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4267 * HCA translation table.
4268 * @mr: The memory region to deregister.
4269 * @udata: Valid user data or NULL for kernel object
4271 * This function can fail, if the memory region has memory windows bound to it.
4273 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4276 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4277 * HCA translation table.
4278 * @mr: The memory region to deregister.
4280 * This function can fail, if the memory region has memory windows bound to it.
4282 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4284 static inline int ib_dereg_mr(struct ib_mr *mr)
4286 return ib_dereg_mr_user(mr, NULL);
4289 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4292 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4293 u32 max_num_data_sg,
4294 u32 max_num_meta_sg);
4297 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4299 * @mr - struct ib_mr pointer to be updated.
4300 * @newkey - new key to be used.
4302 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4304 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4305 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4309 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4310 * for calculating a new rkey for type 2 memory windows.
4311 * @rkey - the rkey to increment.
4313 static inline u32 ib_inc_rkey(u32 rkey)
4315 const u32 mask = 0x000000ff;
4316 return ((rkey + 1) & mask) | (rkey & ~mask);
4320 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4321 * @qp: QP to attach to the multicast group. The QP must be type
4323 * @gid: Multicast group GID.
4324 * @lid: Multicast group LID in host byte order.
4326 * In order to send and receive multicast packets, subnet
4327 * administration must have created the multicast group and configured
4328 * the fabric appropriately. The port associated with the specified
4329 * QP must also be a member of the multicast group.
4331 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4334 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4335 * @qp: QP to detach from the multicast group.
4336 * @gid: Multicast group GID.
4337 * @lid: Multicast group LID in host byte order.
4339 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4341 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4342 struct inode *inode, struct ib_udata *udata);
4343 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4345 static inline int ib_check_mr_access(struct ib_device *ib_dev,
4348 u64 device_cap = ib_dev->attrs.device_cap_flags;
4351 * Local write permission is required if remote write or
4352 * remote atomic permission is also requested.
4354 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4355 !(flags & IB_ACCESS_LOCAL_WRITE))
4358 if (flags & ~IB_ACCESS_SUPPORTED)
4361 if (flags & IB_ACCESS_ON_DEMAND &&
4362 !(ib_dev->attrs.kernel_cap_flags & IBK_ON_DEMAND_PAGING))
4365 if ((flags & IB_ACCESS_FLUSH_GLOBAL &&
4366 !(device_cap & IB_DEVICE_FLUSH_GLOBAL)) ||
4367 (flags & IB_ACCESS_FLUSH_PERSISTENT &&
4368 !(device_cap & IB_DEVICE_FLUSH_PERSISTENT)))
4374 static inline bool ib_access_writable(int access_flags)
4377 * We have writable memory backing the MR if any of the following
4378 * access flags are set. "Local write" and "remote write" obviously
4379 * require write access. "Remote atomic" can do things like fetch and
4380 * add, which will modify memory, and "MW bind" can change permissions
4381 * by binding a window.
4383 return access_flags &
4384 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4385 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4389 * ib_check_mr_status: lightweight check of MR status.
4390 * This routine may provide status checks on a selected
4391 * ib_mr. first use is for signature status check.
4393 * @mr: A memory region.
4394 * @check_mask: Bitmask of which checks to perform from
4395 * ib_mr_status_check enumeration.
4396 * @mr_status: The container of relevant status checks.
4397 * failed checks will be indicated in the status bitmask
4398 * and the relevant info shall be in the error item.
4400 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4401 struct ib_mr_status *mr_status);
4404 * ib_device_try_get: Hold a registration lock
4405 * device: The device to lock
4407 * A device under an active registration lock cannot become unregistered. It
4408 * is only possible to obtain a registration lock on a device that is fully
4409 * registered, otherwise this function returns false.
4411 * The registration lock is only necessary for actions which require the
4412 * device to still be registered. Uses that only require the device pointer to
4413 * be valid should use get_device(&ibdev->dev) to hold the memory.
4416 static inline bool ib_device_try_get(struct ib_device *dev)
4418 return refcount_inc_not_zero(&dev->refcount);
4421 void ib_device_put(struct ib_device *device);
4422 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4423 enum rdma_driver_id driver_id);
4424 struct ib_device *ib_device_get_by_name(const char *name,
4425 enum rdma_driver_id driver_id);
4426 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port,
4427 u16 pkey, const union ib_gid *gid,
4428 const struct sockaddr *addr);
4429 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4431 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4432 struct ib_wq_init_attr *init_attr);
4433 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4435 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4436 unsigned int *sg_offset, unsigned int page_size);
4437 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4438 int data_sg_nents, unsigned int *data_sg_offset,
4439 struct scatterlist *meta_sg, int meta_sg_nents,
4440 unsigned int *meta_sg_offset, unsigned int page_size);
4443 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4444 unsigned int *sg_offset, unsigned int page_size)
4448 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4454 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4455 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4457 void ib_drain_rq(struct ib_qp *qp);
4458 void ib_drain_sq(struct ib_qp *qp);
4459 void ib_drain_qp(struct ib_qp *qp);
4461 int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed,
4464 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4466 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4467 return attr->roce.dmac;
4471 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4473 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4474 attr->ib.dlid = (u16)dlid;
4475 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4476 attr->opa.dlid = dlid;
4479 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4481 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4482 return attr->ib.dlid;
4483 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4484 return attr->opa.dlid;
4488 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4493 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4498 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4501 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4502 attr->ib.src_path_bits = src_path_bits;
4503 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4504 attr->opa.src_path_bits = src_path_bits;
4507 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4509 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4510 return attr->ib.src_path_bits;
4511 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4512 return attr->opa.src_path_bits;
4516 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4519 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4520 attr->opa.make_grd = make_grd;
4523 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4525 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4526 return attr->opa.make_grd;
4530 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num)
4532 attr->port_num = port_num;
4535 static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4537 return attr->port_num;
4540 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4543 attr->static_rate = static_rate;
4546 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4548 return attr->static_rate;
4551 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4552 enum ib_ah_flags flag)
4554 attr->ah_flags = flag;
4557 static inline enum ib_ah_flags
4558 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4560 return attr->ah_flags;
4563 static inline const struct ib_global_route
4564 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4569 /*To retrieve and modify the grh */
4570 static inline struct ib_global_route
4571 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4576 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4578 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4580 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4583 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4586 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4588 grh->dgid.global.subnet_prefix = prefix;
4591 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4594 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4596 grh->dgid.global.interface_id = if_id;
4599 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4600 union ib_gid *dgid, u32 flow_label,
4601 u8 sgid_index, u8 hop_limit,
4604 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4606 attr->ah_flags = IB_AH_GRH;
4609 grh->flow_label = flow_label;
4610 grh->sgid_index = sgid_index;
4611 grh->hop_limit = hop_limit;
4612 grh->traffic_class = traffic_class;
4613 grh->sgid_attr = NULL;
4616 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4617 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4618 u32 flow_label, u8 hop_limit, u8 traffic_class,
4619 const struct ib_gid_attr *sgid_attr);
4620 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4621 const struct rdma_ah_attr *src);
4622 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4623 const struct rdma_ah_attr *new);
4624 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4627 * rdma_ah_find_type - Return address handle type.
4629 * @dev: Device to be checked
4630 * @port_num: Port number
4632 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4635 if (rdma_protocol_roce(dev, port_num))
4636 return RDMA_AH_ATTR_TYPE_ROCE;
4637 if (rdma_protocol_ib(dev, port_num)) {
4638 if (rdma_cap_opa_ah(dev, port_num))
4639 return RDMA_AH_ATTR_TYPE_OPA;
4640 return RDMA_AH_ATTR_TYPE_IB;
4643 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4647 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4648 * In the current implementation the only way to
4649 * get the 32bit lid is from other sources for OPA.
4650 * For IB, lids will always be 16bits so cast the
4651 * value accordingly.
4655 static inline u16 ib_lid_cpu16(u32 lid)
4657 WARN_ON_ONCE(lid & 0xFFFF0000);
4662 * ib_lid_be16 - Return lid in 16bit BE encoding.
4666 static inline __be16 ib_lid_be16(u32 lid)
4668 WARN_ON_ONCE(lid & 0xFFFF0000);
4669 return cpu_to_be16((u16)lid);
4673 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4675 * @device: the rdma device
4676 * @comp_vector: index of completion vector
4678 * Returns NULL on failure, otherwise a corresponding cpu map of the
4679 * completion vector (returns all-cpus map if the device driver doesn't
4680 * implement get_vector_affinity).
4682 static inline const struct cpumask *
4683 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4685 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4686 !device->ops.get_vector_affinity)
4689 return device->ops.get_vector_affinity(device, comp_vector);
4694 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4695 * and add their gids, as needed, to the relevant RoCE devices.
4697 * @device: the rdma device
4699 void rdma_roce_rescan_device(struct ib_device *ibdev);
4701 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4703 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4705 struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num,
4706 enum rdma_netdev_t type, const char *name,
4707 unsigned char name_assign_type,
4708 void (*setup)(struct net_device *));
4710 int rdma_init_netdev(struct ib_device *device, u32 port_num,
4711 enum rdma_netdev_t type, const char *name,
4712 unsigned char name_assign_type,
4713 void (*setup)(struct net_device *),
4714 struct net_device *netdev);
4717 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4719 * @device: device pointer for which ib_device pointer to retrieve
4721 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4724 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4726 struct ib_core_device *coredev =
4727 container_of(device, struct ib_core_device, dev);
4729 return coredev->owner;
4733 * ibdev_to_node - return the NUMA node for a given ib_device
4734 * @dev: device to get the NUMA node for.
4736 static inline int ibdev_to_node(struct ib_device *ibdev)
4738 struct device *parent = ibdev->dev.parent;
4741 return NUMA_NO_NODE;
4742 return dev_to_node(parent);
4746 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4747 * ib_device holder structure from device pointer.
4749 * NOTE: New drivers should not make use of this API; This API is only for
4750 * existing drivers who have exposed sysfs entries using
4751 * ops->device_group.
4753 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4754 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4756 bool rdma_dev_access_netns(const struct ib_device *device,
4757 const struct net *net);
4759 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4760 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4761 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4764 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4767 * This function will convert the 20 bit flow_label input to a valid RoCE v2
4768 * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4771 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4773 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4775 fl_low ^= fl_high >> 14;
4776 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4780 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4781 * local and remote qpn values
4783 * This function folded the multiplication results of two qpns, 24 bit each,
4784 * fields, and converts it to a 20 bit results.
4786 * This function will create symmetric flow_label value based on the local
4787 * and remote qpn values. this will allow both the requester and responder
4788 * to calculate the same flow_label for a given connection.
4790 * This helper function should be used by driver in case the upper layer
4791 * provide a zero flow_label value. This is to improve entropy of RDMA
4792 * traffic in the network.
4794 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4796 u64 v = (u64)lqpn * rqpn;
4801 return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4805 * rdma_get_udp_sport - Calculate and set UDP source port based on the flow
4806 * label. If flow label is not defined in GRH then
4807 * calculate it based on lqpn/rqpn.
4809 * @fl: flow label from GRH
4810 * @lqpn: local qp number
4811 * @rqpn: remote qp number
4813 static inline u16 rdma_get_udp_sport(u32 fl, u32 lqpn, u32 rqpn)
4816 fl = rdma_calc_flow_label(lqpn, rqpn);
4818 return rdma_flow_label_to_udp_sport(fl);
4821 const struct ib_port_immutable*
4822 ib_port_immutable_read(struct ib_device *dev, unsigned int port);
4823 #endif /* IB_VERBS_H */