Loading...
1/* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
2/*
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.
10 */
11
12#ifndef IB_VERBS_H
13#define IB_VERBS_H
14
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>
25#include <net/ipv6.h>
26#include <net/ip.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>
45
46#define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
47
48struct ib_umem_odp;
49struct ib_uqp_object;
50struct ib_usrq_object;
51struct ib_uwq_object;
52struct rdma_cm_id;
53struct ib_port;
54struct hw_stats_device_data;
55
56extern struct workqueue_struct *ib_wq;
57extern struct workqueue_struct *ib_comp_wq;
58extern struct workqueue_struct *ib_comp_unbound_wq;
59
60struct ib_ucq_object;
61
62__printf(3, 4) __cold
63void ibdev_printk(const char *level, const struct ib_device *ibdev,
64 const char *format, ...);
65__printf(2, 3) __cold
66void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
67__printf(2, 3) __cold
68void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
69__printf(2, 3) __cold
70void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
71__printf(2, 3) __cold
72void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
73__printf(2, 3) __cold
74void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
75__printf(2, 3) __cold
76void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
77__printf(2, 3) __cold
78void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
79
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)
84#else
85__printf(2, 3) __cold
86static inline
87void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
88#endif
89
90#define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \
91do { \
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__); \
97} while (0)
98
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__)
113
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, ...) \
118do { \
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, \
125 ##__VA_ARGS__); \
126} while (0)
127#else
128__printf(2, 3) __cold
129static inline
130void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
131#endif
132
133union ib_gid {
134 u8 raw[16];
135 struct {
136 __be64 subnet_prefix;
137 __be64 interface_id;
138 } global;
139};
140
141extern union ib_gid zgid;
142
143enum ib_gid_type {
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,
147 IB_GID_TYPE_SIZE
148};
149
150#define ROCE_V2_UDP_DPORT 4791
151struct ib_gid_attr {
152 struct net_device __rcu *ndev;
153 struct ib_device *device;
154 union ib_gid gid;
155 enum ib_gid_type gid_type;
156 u16 index;
157 u32 port_num;
158};
159
160enum {
161 /* set the local administered indication */
162 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
163};
164
165enum rdma_transport_type {
166 RDMA_TRANSPORT_IB,
167 RDMA_TRANSPORT_IWARP,
168 RDMA_TRANSPORT_USNIC,
169 RDMA_TRANSPORT_USNIC_UDP,
170 RDMA_TRANSPORT_UNSPECIFIED,
171};
172
173enum rdma_protocol_type {
174 RDMA_PROTOCOL_IB,
175 RDMA_PROTOCOL_IBOE,
176 RDMA_PROTOCOL_IWARP,
177 RDMA_PROTOCOL_USNIC_UDP
178};
179
180__attribute_const__ enum rdma_transport_type
181rdma_node_get_transport(unsigned int node_type);
182
183enum rdma_network_type {
184 RDMA_NETWORK_IB,
185 RDMA_NETWORK_ROCE_V1,
186 RDMA_NETWORK_IPV4,
187 RDMA_NETWORK_IPV6
188};
189
190static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
191{
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;
197 else
198 return IB_GID_TYPE_IB;
199}
200
201static inline enum rdma_network_type
202rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
203{
204 if (attr->gid_type == IB_GID_TYPE_IB)
205 return RDMA_NETWORK_IB;
206
207 if (attr->gid_type == IB_GID_TYPE_ROCE)
208 return RDMA_NETWORK_ROCE_V1;
209
210 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
211 return RDMA_NETWORK_IPV4;
212 else
213 return RDMA_NETWORK_IPV6;
214}
215
216enum rdma_link_layer {
217 IB_LINK_LAYER_UNSPECIFIED,
218 IB_LINK_LAYER_INFINIBAND,
219 IB_LINK_LAYER_ETHERNET,
220};
221
222enum 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,
238
239 /* Reserved, old SEND_W_INV = 1 << 16,*/
240 IB_DEVICE_MEM_WINDOW = IB_UVERBS_DEVICE_MEM_WINDOW,
241 /*
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.
247 */
248 IB_DEVICE_UD_IP_CSUM = IB_UVERBS_DEVICE_UD_IP_CSUM,
249 IB_DEVICE_XRC = IB_UVERBS_DEVICE_XRC,
250
251 /*
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
258 * stag.
259 */
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,
277};
278
279enum ib_kernel_cap_flags {
280 /*
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.
286 */
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,
296
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 */
300 IBK_UD_TSO = 1 << 6,
301 /* iopib will use the device ops:
302 * get_vf_config
303 * get_vf_guid
304 * get_vf_stats
305 * set_vf_guid
306 * set_vf_link_state
307 */
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,
311};
312
313enum ib_atomic_cap {
314 IB_ATOMIC_NONE,
315 IB_ATOMIC_HCA,
316 IB_ATOMIC_GLOB
317};
318
319enum ib_odp_general_cap_bits {
320 IB_ODP_SUPPORT = 1 << 0,
321 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
322};
323
324enum 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,
331};
332
333struct ib_odp_caps {
334 uint64_t general_caps;
335 struct {
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;
341};
342
343struct ib_rss_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
347 */
348 u32 supported_qpts;
349 u32 max_rwq_indirection_tables;
350 u32 max_rwq_indirection_table_size;
351};
352
353enum ib_tm_cap_flags {
354 /* Support tag matching with rendezvous offload for RC transport */
355 IB_TM_CAP_RNDV_RC = 1 << 0,
356};
357
358struct ib_tm_caps {
359 /* Max size of RNDV header */
360 u32 max_rndv_hdr_size;
361 /* Max number of entries in tag matching list */
362 u32 max_num_tags;
363 /* From enum ib_tm_cap_flags */
364 u32 flags;
365 /* Max number of outstanding list operations */
366 u32 max_ops;
367 /* Max number of SGE in tag matching entry */
368 u32 max_sge;
369};
370
371struct ib_cq_init_attr {
372 unsigned int cqe;
373 u32 comp_vector;
374 u32 flags;
375};
376
377enum ib_cq_attr_mask {
378 IB_CQ_MODERATE = 1 << 0,
379};
380
381struct ib_cq_caps {
382 u16 max_cq_moderation_count;
383 u16 max_cq_moderation_period;
384};
385
386struct ib_dm_mr_attr {
387 u64 length;
388 u64 offset;
389 u32 access_flags;
390};
391
392struct ib_dm_alloc_attr {
393 u64 length;
394 u32 alignment;
395 u32 flags;
396};
397
398struct ib_device_attr {
399 u64 fw_ver;
400 __be64 sys_image_guid;
401 u64 max_mr_size;
402 u64 page_size_cap;
403 u32 vendor_id;
404 u32 vendor_part_id;
405 u32 hw_ver;
406 int max_qp;
407 int max_qp_wr;
408 u64 device_cap_flags;
409 u64 kernel_cap_flags;
410 int max_send_sge;
411 int max_recv_sge;
412 int max_sge_rd;
413 int max_cq;
414 int max_cqe;
415 int max_mr;
416 int max_pd;
417 int max_qp_rd_atom;
418 int max_ee_rd_atom;
419 int max_res_rd_atom;
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;
424 int max_ee;
425 int max_rdd;
426 int max_mw;
427 int max_raw_ipv6_qp;
428 int max_raw_ethy_qp;
429 int max_mcast_grp;
430 int max_mcast_qp_attach;
431 int max_total_mcast_qp_attach;
432 int max_ah;
433 int max_srq;
434 int max_srq_wr;
435 int max_srq_sge;
436 unsigned int max_fast_reg_page_list_len;
437 unsigned int max_pi_fast_reg_page_list_len;
438 u16 max_pkeys;
439 u8 local_ca_ack_delay;
440 int sig_prot_cap;
441 int sig_guard_cap;
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;
446 u32 max_wq_type_rq;
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;
450 u64 max_dm_size;
451 /* Max entries for sgl for optimized performance per READ */
452 u32 max_sgl_rd;
453};
454
455enum ib_mtu {
456 IB_MTU_256 = 1,
457 IB_MTU_512 = 2,
458 IB_MTU_1024 = 3,
459 IB_MTU_2048 = 4,
460 IB_MTU_4096 = 5
461};
462
463enum opa_mtu {
464 OPA_MTU_8192 = 6,
465 OPA_MTU_10240 = 7
466};
467
468static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
469{
470 switch (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;
476 default: return -1;
477 }
478}
479
480static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
481{
482 if (mtu >= 4096)
483 return IB_MTU_4096;
484 else if (mtu >= 2048)
485 return IB_MTU_2048;
486 else if (mtu >= 1024)
487 return IB_MTU_1024;
488 else if (mtu >= 512)
489 return IB_MTU_512;
490 else
491 return IB_MTU_256;
492}
493
494static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
495{
496 switch (mtu) {
497 case OPA_MTU_8192:
498 return 8192;
499 case OPA_MTU_10240:
500 return 10240;
501 default:
502 return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
503 }
504}
505
506static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
507{
508 if (mtu >= 10240)
509 return OPA_MTU_10240;
510 else if (mtu >= 8192)
511 return OPA_MTU_8192;
512 else
513 return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
514}
515
516enum ib_port_state {
517 IB_PORT_NOP = 0,
518 IB_PORT_DOWN = 1,
519 IB_PORT_INIT = 2,
520 IB_PORT_ARMED = 3,
521 IB_PORT_ACTIVE = 4,
522 IB_PORT_ACTIVE_DEFER = 5
523};
524
525enum 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,
533};
534
535enum ib_port_width {
536 IB_WIDTH_1X = 1,
537 IB_WIDTH_2X = 16,
538 IB_WIDTH_4X = 2,
539 IB_WIDTH_8X = 4,
540 IB_WIDTH_12X = 8
541};
542
543static inline int ib_width_enum_to_int(enum ib_port_width width)
544{
545 switch (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;
551 default: return -1;
552 }
553}
554
555enum ib_port_speed {
556 IB_SPEED_SDR = 1,
557 IB_SPEED_DDR = 2,
558 IB_SPEED_QDR = 4,
559 IB_SPEED_FDR10 = 8,
560 IB_SPEED_FDR = 16,
561 IB_SPEED_EDR = 32,
562 IB_SPEED_HDR = 64,
563 IB_SPEED_NDR = 128,
564 IB_SPEED_XDR = 256,
565};
566
567enum ib_stat_flag {
568 IB_STAT_FLAG_OPTIONAL = 1 << 0,
569};
570
571/**
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
576 */
577struct rdma_stat_desc {
578 const char *name;
579 unsigned int flags;
580 const void *priv;
581};
582
583/**
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
594 * in directory.
595 * @is_disabled - A bitmap to indicate each counter is currently disabled
596 * or not.
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
603 */
604struct 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;
610 int num_counters;
611 u64 value[] __counted_by(num_counters);
612};
613
614#define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
615
616struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
617 const struct rdma_stat_desc *descs, int num_counters,
618 unsigned long lifespan);
619
620void rdma_free_hw_stats_struct(struct rdma_hw_stats *stats);
621
622/* Define bits for the various functionality this port needs to be supported by
623 * the core.
624 */
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
632
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
638
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
646
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)
650
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)
672
673#define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
674
675#define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
676
677struct ib_port_attr {
678 u64 subnet_prefix;
679 enum ib_port_state state;
680 enum ib_mtu max_mtu;
681 enum ib_mtu active_mtu;
682 u32 phys_mtu;
683 int gid_tbl_len;
684 unsigned int ip_gids:1;
685 /* This is the value from PortInfo CapabilityMask, defined by IBA */
686 u32 port_cap_flags;
687 u32 max_msg_sz;
688 u32 bad_pkey_cntr;
689 u32 qkey_viol_cntr;
690 u16 pkey_tbl_len;
691 u32 sm_lid;
692 u32 lid;
693 u8 lmc;
694 u8 max_vl_num;
695 u8 sm_sl;
696 u8 subnet_timeout;
697 u8 init_type_reply;
698 u8 active_width;
699 u16 active_speed;
700 u8 phys_state;
701 u16 port_cap_flags2;
702};
703
704enum ib_device_modify_flags {
705 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
706 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
707};
708
709#define IB_DEVICE_NODE_DESC_MAX 64
710
711struct ib_device_modify {
712 u64 sys_image_guid;
713 char node_desc[IB_DEVICE_NODE_DESC_MAX];
714};
715
716enum 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)
721};
722
723struct ib_port_modify {
724 u32 set_port_cap_mask;
725 u32 clr_port_cap_mask;
726 u8 init_type;
727};
728
729enum ib_event_type {
730 IB_EVENT_CQ_ERR,
731 IB_EVENT_QP_FATAL,
732 IB_EVENT_QP_REQ_ERR,
733 IB_EVENT_QP_ACCESS_ERR,
734 IB_EVENT_COMM_EST,
735 IB_EVENT_SQ_DRAINED,
736 IB_EVENT_PATH_MIG,
737 IB_EVENT_PATH_MIG_ERR,
738 IB_EVENT_DEVICE_FATAL,
739 IB_EVENT_PORT_ACTIVE,
740 IB_EVENT_PORT_ERR,
741 IB_EVENT_LID_CHANGE,
742 IB_EVENT_PKEY_CHANGE,
743 IB_EVENT_SM_CHANGE,
744 IB_EVENT_SRQ_ERR,
745 IB_EVENT_SRQ_LIMIT_REACHED,
746 IB_EVENT_QP_LAST_WQE_REACHED,
747 IB_EVENT_CLIENT_REREGISTER,
748 IB_EVENT_GID_CHANGE,
749 IB_EVENT_WQ_FATAL,
750};
751
752const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
753
754struct ib_event {
755 struct ib_device *device;
756 union {
757 struct ib_cq *cq;
758 struct ib_qp *qp;
759 struct ib_srq *srq;
760 struct ib_wq *wq;
761 u32 port_num;
762 } element;
763 enum ib_event_type event;
764};
765
766struct ib_event_handler {
767 struct ib_device *device;
768 void (*handler)(struct ib_event_handler *, struct ib_event *);
769 struct list_head list;
770};
771
772#define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
773 do { \
774 (_ptr)->device = _device; \
775 (_ptr)->handler = _handler; \
776 INIT_LIST_HEAD(&(_ptr)->list); \
777 } while (0)
778
779struct ib_global_route {
780 const struct ib_gid_attr *sgid_attr;
781 union ib_gid dgid;
782 u32 flow_label;
783 u8 sgid_index;
784 u8 hop_limit;
785 u8 traffic_class;
786};
787
788struct ib_grh {
789 __be32 version_tclass_flow;
790 __be16 paylen;
791 u8 next_hdr;
792 u8 hop_limit;
793 union ib_gid sgid;
794 union ib_gid dgid;
795};
796
797union rdma_network_hdr {
798 struct ib_grh ibgrh;
799 struct {
800 /* The IB spec states that if it's IPv4, the header
801 * is located in the last 20 bytes of the header.
802 */
803 u8 reserved[20];
804 struct iphdr roce4grh;
805 };
806};
807
808#define IB_QPN_MASK 0xFFFFFF
809
810enum {
811 IB_MULTICAST_QPN = 0xffffff
812};
813
814#define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
815#define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
816
817enum ib_ah_flags {
818 IB_AH_GRH = 1
819};
820
821enum ib_rate {
822 IB_RATE_PORT_CURRENT = 0,
823 IB_RATE_2_5_GBPS = 2,
824 IB_RATE_5_GBPS = 5,
825 IB_RATE_10_GBPS = 3,
826 IB_RATE_20_GBPS = 6,
827 IB_RATE_30_GBPS = 4,
828 IB_RATE_40_GBPS = 7,
829 IB_RATE_60_GBPS = 8,
830 IB_RATE_80_GBPS = 9,
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,
845};
846
847/**
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.
852 */
853__attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
854
855/**
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.
859 */
860__attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
861
862
863/**
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
870 * ib_map_mr_sg)
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
874 * application
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
879 */
880enum ib_mr_type {
881 IB_MR_TYPE_MEM_REG,
882 IB_MR_TYPE_SG_GAPS,
883 IB_MR_TYPE_DM,
884 IB_MR_TYPE_USER,
885 IB_MR_TYPE_DMA,
886 IB_MR_TYPE_INTEGRITY,
887};
888
889enum ib_mr_status_check {
890 IB_MR_CHECK_SIG_STATUS = 1,
891};
892
893/**
894 * struct ib_mr_status - Memory region status container
895 *
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
899 * failure.
900 */
901struct ib_mr_status {
902 u32 fail_status;
903 struct ib_sig_err sig_err;
904};
905
906/**
907 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
908 * enum.
909 * @mult: multiple to convert.
910 */
911__attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
912
913struct rdma_ah_init_attr {
914 struct rdma_ah_attr *ah_attr;
915 u32 flags;
916 struct net_device *xmit_slave;
917};
918
919enum 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,
924};
925
926struct ib_ah_attr {
927 u16 dlid;
928 u8 src_path_bits;
929};
930
931struct roce_ah_attr {
932 u8 dmac[ETH_ALEN];
933};
934
935struct opa_ah_attr {
936 u32 dlid;
937 u8 src_path_bits;
938 bool make_grd;
939};
940
941struct rdma_ah_attr {
942 struct ib_global_route grh;
943 u8 sl;
944 u8 static_rate;
945 u32 port_num;
946 u8 ah_flags;
947 enum rdma_ah_attr_type type;
948 union {
949 struct ib_ah_attr ib;
950 struct roce_ah_attr roce;
951 struct opa_ah_attr opa;
952 };
953};
954
955enum ib_wc_status {
956 IB_WC_SUCCESS,
957 IB_WC_LOC_LEN_ERR,
958 IB_WC_LOC_QP_OP_ERR,
959 IB_WC_LOC_EEC_OP_ERR,
960 IB_WC_LOC_PROT_ERR,
961 IB_WC_WR_FLUSH_ERR,
962 IB_WC_MW_BIND_ERR,
963 IB_WC_BAD_RESP_ERR,
964 IB_WC_LOC_ACCESS_ERR,
965 IB_WC_REM_INV_REQ_ERR,
966 IB_WC_REM_ACCESS_ERR,
967 IB_WC_REM_OP_ERR,
968 IB_WC_RETRY_EXC_ERR,
969 IB_WC_RNR_RETRY_EXC_ERR,
970 IB_WC_LOC_RDD_VIOL_ERR,
971 IB_WC_REM_INV_RD_REQ_ERR,
972 IB_WC_REM_ABORT_ERR,
973 IB_WC_INV_EECN_ERR,
974 IB_WC_INV_EEC_STATE_ERR,
975 IB_WC_FATAL_ERR,
976 IB_WC_RESP_TIMEOUT_ERR,
977 IB_WC_GENERAL_ERR
978};
979
980const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
981
982enum ib_wc_opcode {
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,
992 IB_WC_REG_MR,
993 IB_WC_MASKED_COMP_SWAP,
994 IB_WC_MASKED_FETCH_ADD,
995 IB_WC_FLUSH = IB_UVERBS_WC_FLUSH,
996/*
997 * Set value of IB_WC_RECV so consumers can test if a completion is a
998 * receive by testing (opcode & IB_WC_RECV).
999 */
1000 IB_WC_RECV = 1 << 7,
1001 IB_WC_RECV_RDMA_WITH_IMM
1002};
1003
1004enum ib_wc_flags {
1005 IB_WC_GRH = 1,
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),
1012};
1013
1014struct ib_wc {
1015 union {
1016 u64 wr_id;
1017 struct ib_cqe *wr_cqe;
1018 };
1019 enum ib_wc_status status;
1020 enum ib_wc_opcode opcode;
1021 u32 vendor_err;
1022 u32 byte_len;
1023 struct ib_qp *qp;
1024 union {
1025 __be32 imm_data;
1026 u32 invalidate_rkey;
1027 } ex;
1028 u32 src_qp;
1029 u32 slid;
1030 int wc_flags;
1031 u16 pkey_index;
1032 u8 sl;
1033 u8 dlid_path_bits;
1034 u32 port_num; /* valid only for DR SMPs on switches */
1035 u8 smac[ETH_ALEN];
1036 u16 vlan_id;
1037 u8 network_hdr_type;
1038};
1039
1040enum 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,
1045};
1046
1047enum ib_srq_type {
1048 IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1049 IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1050 IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1051};
1052
1053static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1054{
1055 return srq_type == IB_SRQT_XRC ||
1056 srq_type == IB_SRQT_TM;
1057}
1058
1059enum ib_srq_attr_mask {
1060 IB_SRQ_MAX_WR = 1 << 0,
1061 IB_SRQ_LIMIT = 1 << 1,
1062};
1063
1064struct ib_srq_attr {
1065 u32 max_wr;
1066 u32 max_sge;
1067 u32 srq_limit;
1068};
1069
1070struct ib_srq_init_attr {
1071 void (*event_handler)(struct ib_event *, void *);
1072 void *srq_context;
1073 struct ib_srq_attr attr;
1074 enum ib_srq_type srq_type;
1075
1076 struct {
1077 struct ib_cq *cq;
1078 union {
1079 struct {
1080 struct ib_xrcd *xrcd;
1081 } xrc;
1082
1083 struct {
1084 u32 max_num_tags;
1085 } tag_matching;
1086 };
1087 } ext;
1088};
1089
1090struct ib_qp_cap {
1091 u32 max_send_wr;
1092 u32 max_recv_wr;
1093 u32 max_send_sge;
1094 u32 max_recv_sge;
1095 u32 max_inline_data;
1096
1097 /*
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.
1101 */
1102 u32 max_rdma_ctxs;
1103};
1104
1105enum ib_sig_type {
1106 IB_SIGNAL_ALL_WR,
1107 IB_SIGNAL_REQ_WR
1108};
1109
1110enum ib_qp_type {
1111 /*
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.
1115 */
1116 IB_QPT_SMI,
1117 IB_QPT_GSI,
1118
1119 IB_QPT_RC = IB_UVERBS_QPT_RC,
1120 IB_QPT_UC = IB_UVERBS_QPT_UC,
1121 IB_QPT_UD = IB_UVERBS_QPT_UD,
1122 IB_QPT_RAW_IPV6,
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,
1127 IB_QPT_MAX,
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
1132 */
1133 IB_QPT_RESERVED1 = 0x1000,
1134 IB_QPT_RESERVED2,
1135 IB_QPT_RESERVED3,
1136 IB_QPT_RESERVED4,
1137 IB_QPT_RESERVED5,
1138 IB_QPT_RESERVED6,
1139 IB_QPT_RESERVED7,
1140 IB_QPT_RESERVED8,
1141 IB_QPT_RESERVED9,
1142 IB_QPT_RESERVED10,
1143};
1144
1145enum 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,
1165};
1166
1167/*
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.
1170 */
1171
1172struct ib_qp_init_attr {
1173 /* This callback occurs in workqueue context */
1174 void (*event_handler)(struct ib_event *, void *);
1175
1176 void *qp_context;
1177 struct ib_cq *send_cq;
1178 struct ib_cq *recv_cq;
1179 struct ib_srq *srq;
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;
1184 u32 create_flags;
1185
1186 /*
1187 * Only needed for special QP types, or when using the RW API.
1188 */
1189 u32 port_num;
1190 struct ib_rwq_ind_table *rwq_ind_tbl;
1191 u32 source_qpn;
1192};
1193
1194struct ib_qp_open_attr {
1195 void (*event_handler)(struct ib_event *, void *);
1196 void *qp_context;
1197 u32 qp_num;
1198 enum ib_qp_type qp_type;
1199};
1200
1201enum 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
1234};
1235
1236enum ib_qp_attr_mask {
1237 IB_QP_STATE = 1,
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),
1244 IB_QP_AV = (1<<7),
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),
1263
1264 IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0),
1265};
1266
1267enum ib_qp_state {
1268 IB_QPS_RESET,
1269 IB_QPS_INIT,
1270 IB_QPS_RTR,
1271 IB_QPS_RTS,
1272 IB_QPS_SQD,
1273 IB_QPS_SQE,
1274 IB_QPS_ERR
1275};
1276
1277enum ib_mig_state {
1278 IB_MIG_MIGRATED,
1279 IB_MIG_REARM,
1280 IB_MIG_ARMED
1281};
1282
1283enum ib_mw_type {
1284 IB_MW_TYPE_1 = 1,
1285 IB_MW_TYPE_2 = 2
1286};
1287
1288struct ib_qp_attr {
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;
1293 u32 qkey;
1294 u32 rq_psn;
1295 u32 sq_psn;
1296 u32 dest_qp_num;
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;
1301 u16 pkey_index;
1302 u16 alt_pkey_index;
1303 u8 en_sqd_async_notify;
1304 u8 sq_draining;
1305 u8 max_rd_atomic;
1306 u8 max_dest_rd_atomic;
1307 u8 min_rnr_timer;
1308 u32 port_num;
1309 u8 timeout;
1310 u8 retry_cnt;
1311 u8 rnr_retry;
1312 u32 alt_port_num;
1313 u8 alt_timeout;
1314 u32 rate_limit;
1315 struct net_device *xmit_slave;
1316};
1317
1318enum ib_wr_opcode {
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,
1338
1339 /* These are kernel only and can not be issued by userspace */
1340 IB_WR_REG_MR = 0x20,
1341 IB_WR_REG_MR_INTEGRITY,
1342
1343 /* reserve values for low level drivers' internal use.
1344 * These values will not be used at all in the ib core layer.
1345 */
1346 IB_WR_RESERVED1 = 0xf0,
1347 IB_WR_RESERVED2,
1348 IB_WR_RESERVED3,
1349 IB_WR_RESERVED4,
1350 IB_WR_RESERVED5,
1351 IB_WR_RESERVED6,
1352 IB_WR_RESERVED7,
1353 IB_WR_RESERVED8,
1354 IB_WR_RESERVED9,
1355 IB_WR_RESERVED10,
1356};
1357
1358enum ib_send_flags {
1359 IB_SEND_FENCE = 1,
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),
1364
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),
1368};
1369
1370struct ib_sge {
1371 u64 addr;
1372 u32 length;
1373 u32 lkey;
1374};
1375
1376struct ib_cqe {
1377 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1378};
1379
1380struct ib_send_wr {
1381 struct ib_send_wr *next;
1382 union {
1383 u64 wr_id;
1384 struct ib_cqe *wr_cqe;
1385 };
1386 struct ib_sge *sg_list;
1387 int num_sge;
1388 enum ib_wr_opcode opcode;
1389 int send_flags;
1390 union {
1391 __be32 imm_data;
1392 u32 invalidate_rkey;
1393 } ex;
1394};
1395
1396struct ib_rdma_wr {
1397 struct ib_send_wr wr;
1398 u64 remote_addr;
1399 u32 rkey;
1400};
1401
1402static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1403{
1404 return container_of(wr, struct ib_rdma_wr, wr);
1405}
1406
1407struct ib_atomic_wr {
1408 struct ib_send_wr wr;
1409 u64 remote_addr;
1410 u64 compare_add;
1411 u64 swap;
1412 u64 compare_add_mask;
1413 u64 swap_mask;
1414 u32 rkey;
1415};
1416
1417static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1418{
1419 return container_of(wr, struct ib_atomic_wr, wr);
1420}
1421
1422struct ib_ud_wr {
1423 struct ib_send_wr wr;
1424 struct ib_ah *ah;
1425 void *header;
1426 int hlen;
1427 int mss;
1428 u32 remote_qpn;
1429 u32 remote_qkey;
1430 u16 pkey_index; /* valid for GSI only */
1431 u32 port_num; /* valid for DR SMPs on switch only */
1432};
1433
1434static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1435{
1436 return container_of(wr, struct ib_ud_wr, wr);
1437}
1438
1439struct ib_reg_wr {
1440 struct ib_send_wr wr;
1441 struct ib_mr *mr;
1442 u32 key;
1443 int access;
1444};
1445
1446static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1447{
1448 return container_of(wr, struct ib_reg_wr, wr);
1449}
1450
1451struct ib_recv_wr {
1452 struct ib_recv_wr *next;
1453 union {
1454 u64 wr_id;
1455 struct ib_cqe *wr_cqe;
1456 };
1457 struct ib_sge *sg_list;
1458 int num_sge;
1459};
1460
1461enum 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,
1473
1474 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1475 IB_ACCESS_SUPPORTED =
1476 ((IB_ACCESS_FLUSH_PERSISTENT << 1) - 1) | IB_ACCESS_OPTIONAL,
1477};
1478
1479/*
1480 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1481 * are hidden here instead of a uapi header!
1482 */
1483enum 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)
1488};
1489
1490struct ib_umem;
1491
1492enum rdma_remove_reason {
1493 /*
1494 * Userspace requested uobject deletion or initial try
1495 * to remove uobject via cleanup. Call could fail
1496 */
1497 RDMA_REMOVE_DESTROY,
1498 /* Context deletion. This call should delete the actual object itself */
1499 RDMA_REMOVE_CLOSE,
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 */
1503 RDMA_REMOVE_ABORT,
1504 /* The driver failed to destroy the uobject and is being disconnected */
1505 RDMA_REMOVE_DRIVER_FAILURE,
1506};
1507
1508struct ib_rdmacg_object {
1509#ifdef CONFIG_CGROUP_RDMA
1510 struct rdma_cgroup *cg; /* owner rdma cgroup */
1511#endif
1512};
1513
1514struct ib_ucontext {
1515 struct ib_device *device;
1516 struct ib_uverbs_file *ufile;
1517
1518 struct ib_rdmacg_object cg_obj;
1519 /*
1520 * Implementation details of the RDMA core, don't use in drivers:
1521 */
1522 struct rdma_restrack_entry res;
1523 struct xarray mmap_xa;
1524};
1525
1526struct ib_uobject {
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 */
1536 struct kref ref;
1537 atomic_t usecnt; /* protects exclusive access */
1538 struct rcu_head rcu; /* kfree_rcu() overhead */
1539
1540 const struct uverbs_api_object *uapi_object;
1541};
1542
1543struct ib_udata {
1544 const void __user *inbuf;
1545 void __user *outbuf;
1546 size_t inlen;
1547 size_t outlen;
1548};
1549
1550struct ib_pd {
1551 u32 local_dma_lkey;
1552 u32 flags;
1553 struct ib_device *device;
1554 struct ib_uobject *uobject;
1555 atomic_t usecnt; /* count all resources */
1556
1557 u32 unsafe_global_rkey;
1558
1559 /*
1560 * Implementation details of the RDMA core, don't use in drivers:
1561 */
1562 struct ib_mr *__internal_mr;
1563 struct rdma_restrack_entry res;
1564};
1565
1566struct ib_xrcd {
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;
1572};
1573
1574struct ib_ah {
1575 struct ib_device *device;
1576 struct ib_pd *pd;
1577 struct ib_uobject *uobject;
1578 const struct ib_gid_attr *sgid_attr;
1579 enum rdma_ah_attr_type type;
1580};
1581
1582typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1583
1584enum 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,
1589
1590 IB_POLL_DIRECT, /* caller context, no hw completions */
1591};
1592
1593struct ib_cq {
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 *);
1598 void *cq_context;
1599 int cqe;
1600 unsigned int cqe_used;
1601 atomic_t usecnt; /* count number of work queues */
1602 enum ib_poll_context poll_ctx;
1603 struct ib_wc *wc;
1604 struct list_head pool_entry;
1605 union {
1606 struct irq_poll iop;
1607 struct work_struct work;
1608 };
1609 struct workqueue_struct *comp_wq;
1610 struct dim *dim;
1611
1612 /* updated only by trace points */
1613 ktime_t timestamp;
1614 u8 interrupt:1;
1615 u8 shared:1;
1616 unsigned int comp_vector;
1617
1618 /*
1619 * Implementation details of the RDMA core, don't use in drivers:
1620 */
1621 struct rdma_restrack_entry res;
1622};
1623
1624struct ib_srq {
1625 struct ib_device *device;
1626 struct ib_pd *pd;
1627 struct ib_usrq_object *uobject;
1628 void (*event_handler)(struct ib_event *, void *);
1629 void *srq_context;
1630 enum ib_srq_type srq_type;
1631 atomic_t usecnt;
1632
1633 struct {
1634 struct ib_cq *cq;
1635 union {
1636 struct {
1637 struct ib_xrcd *xrcd;
1638 u32 srq_num;
1639 } xrc;
1640 };
1641 } ext;
1642
1643 /*
1644 * Implementation details of the RDMA core, don't use in drivers:
1645 */
1646 struct rdma_restrack_entry res;
1647};
1648
1649enum ib_raw_packet_caps {
1650 /*
1651 * Strip cvlan from incoming packet and report it in the matching work
1652 * completion is supported.
1653 */
1654 IB_RAW_PACKET_CAP_CVLAN_STRIPPING =
1655 IB_UVERBS_RAW_PACKET_CAP_CVLAN_STRIPPING,
1656 /*
1657 * Scatter FCS field of an incoming packet to host memory is supported.
1658 */
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,
1662 /*
1663 * When a packet is received for an RQ with no receive WQEs, the
1664 * packet processing is delayed.
1665 */
1666 IB_RAW_PACKET_CAP_DELAY_DROP = IB_UVERBS_RAW_PACKET_CAP_DELAY_DROP,
1667};
1668
1669enum ib_wq_type {
1670 IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1671};
1672
1673enum ib_wq_state {
1674 IB_WQS_RESET,
1675 IB_WQS_RDY,
1676 IB_WQS_ERR
1677};
1678
1679struct ib_wq {
1680 struct ib_device *device;
1681 struct ib_uwq_object *uobject;
1682 void *wq_context;
1683 void (*event_handler)(struct ib_event *, void *);
1684 struct ib_pd *pd;
1685 struct ib_cq *cq;
1686 u32 wq_num;
1687 enum ib_wq_state state;
1688 enum ib_wq_type wq_type;
1689 atomic_t usecnt;
1690};
1691
1692enum ib_wq_flags {
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,
1698};
1699
1700struct ib_wq_init_attr {
1701 void *wq_context;
1702 enum ib_wq_type wq_type;
1703 u32 max_wr;
1704 u32 max_sge;
1705 struct ib_cq *cq;
1706 void (*event_handler)(struct ib_event *, void *);
1707 u32 create_flags; /* Use enum ib_wq_flags */
1708};
1709
1710enum ib_wq_attr_mask {
1711 IB_WQ_STATE = 1 << 0,
1712 IB_WQ_CUR_STATE = 1 << 1,
1713 IB_WQ_FLAGS = 1 << 2,
1714};
1715
1716struct ib_wq_attr {
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 */
1721};
1722
1723struct ib_rwq_ind_table {
1724 struct ib_device *device;
1725 struct ib_uobject *uobject;
1726 atomic_t usecnt;
1727 u32 ind_tbl_num;
1728 u32 log_ind_tbl_size;
1729 struct ib_wq **ind_tbl;
1730};
1731
1732struct 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;
1736};
1737
1738enum port_pkey_state {
1739 IB_PORT_PKEY_NOT_VALID = 0,
1740 IB_PORT_PKEY_VALID = 1,
1741 IB_PORT_PKEY_LISTED = 2,
1742};
1743
1744struct ib_qp_security;
1745
1746struct ib_port_pkey {
1747 enum port_pkey_state state;
1748 u16 pkey_index;
1749 u32 port_num;
1750 struct list_head qp_list;
1751 struct list_head to_error_list;
1752 struct ib_qp_security *sec;
1753};
1754
1755struct ib_ports_pkeys {
1756 struct ib_port_pkey main;
1757 struct ib_port_pkey alt;
1758};
1759
1760struct ib_qp_security {
1761 struct ib_qp *qp;
1762 struct ib_device *dev;
1763 /* Hold this mutex when changing port and pkey settings. */
1764 struct mutex mutex;
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.
1768 */
1769 struct list_head shared_qp_list;
1770 void *security;
1771 bool destroying;
1772 atomic_t error_list_count;
1773 struct completion error_complete;
1774 int error_comps_pending;
1775};
1776
1777/*
1778 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1779 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1780 */
1781struct ib_qp {
1782 struct ib_device *device;
1783 struct ib_pd *pd;
1784 struct ib_cq *send_cq;
1785 struct ib_cq *recv_cq;
1786 spinlock_t mr_lock;
1787 int mrs_used;
1788 struct list_head rdma_mrs;
1789 struct list_head sig_mrs;
1790 struct ib_srq *srq;
1791 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1792 struct list_head xrcd_list;
1793
1794 /* count times opened, mcast attaches, flow attaches */
1795 atomic_t usecnt;
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 *);
1800 void *qp_context;
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;
1804 u32 qp_num;
1805 u32 max_write_sge;
1806 u32 max_read_sge;
1807 enum ib_qp_type qp_type;
1808 struct ib_rwq_ind_table *rwq_ind_tbl;
1809 struct ib_qp_security *qp_sec;
1810 u32 port;
1811
1812 bool integrity_en;
1813 /*
1814 * Implementation details of the RDMA core, don't use in drivers:
1815 */
1816 struct rdma_restrack_entry res;
1817
1818 /* The counter the qp is bind to */
1819 struct rdma_counter *counter;
1820};
1821
1822struct ib_dm {
1823 struct ib_device *device;
1824 u32 length;
1825 u32 flags;
1826 struct ib_uobject *uobject;
1827 atomic_t usecnt;
1828};
1829
1830struct ib_mr {
1831 struct ib_device *device;
1832 struct ib_pd *pd;
1833 u32 lkey;
1834 u32 rkey;
1835 u64 iova;
1836 u64 length;
1837 unsigned int page_size;
1838 enum ib_mr_type type;
1839 bool need_inval;
1840 union {
1841 struct ib_uobject *uobject; /* user */
1842 struct list_head qp_entry; /* FR */
1843 };
1844
1845 struct ib_dm *dm;
1846 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1847 /*
1848 * Implementation details of the RDMA core, don't use in drivers:
1849 */
1850 struct rdma_restrack_entry res;
1851};
1852
1853struct ib_mw {
1854 struct ib_device *device;
1855 struct ib_pd *pd;
1856 struct ib_uobject *uobject;
1857 u32 rkey;
1858 enum ib_mw_type type;
1859};
1860
1861/* Supported steering options */
1862enum 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
1867 */
1868 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1869 /* default multicast rule -
1870 * receive all Eth multicast traffic which isn't steered to any QP
1871 */
1872 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1873 /* sniffer rule - receive all port traffic */
1874 IB_FLOW_ATTR_SNIFFER = 0x3
1875};
1876
1877/* Supported steering header types */
1878enum ib_flow_spec_type {
1879 /* L2 headers*/
1880 IB_FLOW_SPEC_ETH = 0x20,
1881 IB_FLOW_SPEC_IB = 0x22,
1882 /* L3 header*/
1883 IB_FLOW_SPEC_IPV4 = 0x30,
1884 IB_FLOW_SPEC_IPV6 = 0x31,
1885 IB_FLOW_SPEC_ESP = 0x34,
1886 /* L4 headers*/
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,
1893 /* Actions */
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,
1898};
1899#define IB_FLOW_SPEC_LAYER_MASK 0xF0
1900#define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1901
1902enum 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 */
1906};
1907
1908struct ib_flow_eth_filter {
1909 u8 dst_mac[6];
1910 u8 src_mac[6];
1911 __be16 ether_type;
1912 __be16 vlan_tag;
1913 /* Must be last */
1914 u8 real_sz[];
1915};
1916
1917struct ib_flow_spec_eth {
1918 u32 type;
1919 u16 size;
1920 struct ib_flow_eth_filter val;
1921 struct ib_flow_eth_filter mask;
1922};
1923
1924struct ib_flow_ib_filter {
1925 __be16 dlid;
1926 __u8 sl;
1927 /* Must be last */
1928 u8 real_sz[];
1929};
1930
1931struct ib_flow_spec_ib {
1932 u32 type;
1933 u16 size;
1934 struct ib_flow_ib_filter val;
1935 struct ib_flow_ib_filter mask;
1936};
1937
1938/* IPv4 header flags */
1939enum ib_ipv4_flags {
1940 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1941 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1942 last have this flag set */
1943};
1944
1945struct ib_flow_ipv4_filter {
1946 __be32 src_ip;
1947 __be32 dst_ip;
1948 u8 proto;
1949 u8 tos;
1950 u8 ttl;
1951 u8 flags;
1952 /* Must be last */
1953 u8 real_sz[];
1954};
1955
1956struct ib_flow_spec_ipv4 {
1957 u32 type;
1958 u16 size;
1959 struct ib_flow_ipv4_filter val;
1960 struct ib_flow_ipv4_filter mask;
1961};
1962
1963struct ib_flow_ipv6_filter {
1964 u8 src_ip[16];
1965 u8 dst_ip[16];
1966 __be32 flow_label;
1967 u8 next_hdr;
1968 u8 traffic_class;
1969 u8 hop_limit;
1970 /* Must be last */
1971 u8 real_sz[];
1972};
1973
1974struct ib_flow_spec_ipv6 {
1975 u32 type;
1976 u16 size;
1977 struct ib_flow_ipv6_filter val;
1978 struct ib_flow_ipv6_filter mask;
1979};
1980
1981struct ib_flow_tcp_udp_filter {
1982 __be16 dst_port;
1983 __be16 src_port;
1984 /* Must be last */
1985 u8 real_sz[];
1986};
1987
1988struct ib_flow_spec_tcp_udp {
1989 u32 type;
1990 u16 size;
1991 struct ib_flow_tcp_udp_filter val;
1992 struct ib_flow_tcp_udp_filter mask;
1993};
1994
1995struct ib_flow_tunnel_filter {
1996 __be32 tunnel_id;
1997 u8 real_sz[];
1998};
1999
2000/* ib_flow_spec_tunnel describes the Vxlan tunnel
2001 * the tunnel_id from val has the vni value
2002 */
2003struct ib_flow_spec_tunnel {
2004 u32 type;
2005 u16 size;
2006 struct ib_flow_tunnel_filter val;
2007 struct ib_flow_tunnel_filter mask;
2008};
2009
2010struct ib_flow_esp_filter {
2011 __be32 spi;
2012 __be32 seq;
2013 /* Must be last */
2014 u8 real_sz[];
2015};
2016
2017struct ib_flow_spec_esp {
2018 u32 type;
2019 u16 size;
2020 struct ib_flow_esp_filter val;
2021 struct ib_flow_esp_filter mask;
2022};
2023
2024struct ib_flow_gre_filter {
2025 __be16 c_ks_res0_ver;
2026 __be16 protocol;
2027 __be32 key;
2028 /* Must be last */
2029 u8 real_sz[];
2030};
2031
2032struct ib_flow_spec_gre {
2033 u32 type;
2034 u16 size;
2035 struct ib_flow_gre_filter val;
2036 struct ib_flow_gre_filter mask;
2037};
2038
2039struct ib_flow_mpls_filter {
2040 __be32 tag;
2041 /* Must be last */
2042 u8 real_sz[];
2043};
2044
2045struct ib_flow_spec_mpls {
2046 u32 type;
2047 u16 size;
2048 struct ib_flow_mpls_filter val;
2049 struct ib_flow_mpls_filter mask;
2050};
2051
2052struct ib_flow_spec_action_tag {
2053 enum ib_flow_spec_type type;
2054 u16 size;
2055 u32 tag_id;
2056};
2057
2058struct ib_flow_spec_action_drop {
2059 enum ib_flow_spec_type type;
2060 u16 size;
2061};
2062
2063struct ib_flow_spec_action_handle {
2064 enum ib_flow_spec_type type;
2065 u16 size;
2066 struct ib_flow_action *act;
2067};
2068
2069enum ib_counters_description {
2070 IB_COUNTER_PACKETS,
2071 IB_COUNTER_BYTES,
2072};
2073
2074struct ib_flow_spec_action_count {
2075 enum ib_flow_spec_type type;
2076 u16 size;
2077 struct ib_counters *counters;
2078};
2079
2080union ib_flow_spec {
2081 struct {
2082 u32 type;
2083 u16 size;
2084 };
2085 struct ib_flow_spec_eth eth;
2086 struct ib_flow_spec_ib ib;
2087 struct ib_flow_spec_ipv4 ipv4;
2088 struct ib_flow_spec_tcp_udp tcp_udp;
2089 struct ib_flow_spec_ipv6 ipv6;
2090 struct ib_flow_spec_tunnel tunnel;
2091 struct ib_flow_spec_esp esp;
2092 struct ib_flow_spec_gre gre;
2093 struct ib_flow_spec_mpls mpls;
2094 struct ib_flow_spec_action_tag flow_tag;
2095 struct ib_flow_spec_action_drop drop;
2096 struct ib_flow_spec_action_handle action;
2097 struct ib_flow_spec_action_count flow_count;
2098};
2099
2100struct ib_flow_attr {
2101 enum ib_flow_attr_type type;
2102 u16 size;
2103 u16 priority;
2104 u32 flags;
2105 u8 num_of_specs;
2106 u32 port;
2107 union ib_flow_spec flows[];
2108};
2109
2110struct ib_flow {
2111 struct ib_qp *qp;
2112 struct ib_device *device;
2113 struct ib_uobject *uobject;
2114};
2115
2116enum ib_flow_action_type {
2117 IB_FLOW_ACTION_UNSPECIFIED,
2118 IB_FLOW_ACTION_ESP = 1,
2119};
2120
2121struct ib_flow_action_attrs_esp_keymats {
2122 enum ib_uverbs_flow_action_esp_keymat protocol;
2123 union {
2124 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2125 } keymat;
2126};
2127
2128struct ib_flow_action_attrs_esp_replays {
2129 enum ib_uverbs_flow_action_esp_replay protocol;
2130 union {
2131 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2132 } replay;
2133};
2134
2135enum ib_flow_action_attrs_esp_flags {
2136 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2137 * This is done in order to share the same flags between user-space and
2138 * kernel and spare an unnecessary translation.
2139 */
2140
2141 /* Kernel flags */
2142 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2143 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2144};
2145
2146struct ib_flow_spec_list {
2147 struct ib_flow_spec_list *next;
2148 union ib_flow_spec spec;
2149};
2150
2151struct ib_flow_action_attrs_esp {
2152 struct ib_flow_action_attrs_esp_keymats *keymat;
2153 struct ib_flow_action_attrs_esp_replays *replay;
2154 struct ib_flow_spec_list *encap;
2155 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2156 * Value of 0 is a valid value.
2157 */
2158 u32 esn;
2159 u32 spi;
2160 u32 seq;
2161 u32 tfc_pad;
2162 /* Use enum ib_flow_action_attrs_esp_flags */
2163 u64 flags;
2164 u64 hard_limit_pkts;
2165};
2166
2167struct ib_flow_action {
2168 struct ib_device *device;
2169 struct ib_uobject *uobject;
2170 enum ib_flow_action_type type;
2171 atomic_t usecnt;
2172};
2173
2174struct ib_mad;
2175
2176enum ib_process_mad_flags {
2177 IB_MAD_IGNORE_MKEY = 1,
2178 IB_MAD_IGNORE_BKEY = 2,
2179 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2180};
2181
2182enum ib_mad_result {
2183 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2184 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2185 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2186 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2187};
2188
2189struct ib_port_cache {
2190 u64 subnet_prefix;
2191 struct ib_pkey_cache *pkey;
2192 struct ib_gid_table *gid;
2193 u8 lmc;
2194 enum ib_port_state port_state;
2195};
2196
2197struct ib_port_immutable {
2198 int pkey_tbl_len;
2199 int gid_tbl_len;
2200 u32 core_cap_flags;
2201 u32 max_mad_size;
2202};
2203
2204struct ib_port_data {
2205 struct ib_device *ib_dev;
2206
2207 struct ib_port_immutable immutable;
2208
2209 spinlock_t pkey_list_lock;
2210
2211 spinlock_t netdev_lock;
2212
2213 struct list_head pkey_list;
2214
2215 struct ib_port_cache cache;
2216
2217 struct net_device __rcu *netdev;
2218 netdevice_tracker netdev_tracker;
2219 struct hlist_node ndev_hash_link;
2220 struct rdma_port_counter port_counter;
2221 struct ib_port *sysfs;
2222};
2223
2224/* rdma netdev type - specifies protocol type */
2225enum rdma_netdev_t {
2226 RDMA_NETDEV_OPA_VNIC,
2227 RDMA_NETDEV_IPOIB,
2228};
2229
2230/**
2231 * struct rdma_netdev - rdma netdev
2232 * For cases where netstack interfacing is required.
2233 */
2234struct rdma_netdev {
2235 void *clnt_priv;
2236 struct ib_device *hca;
2237 u32 port_num;
2238 int mtu;
2239
2240 /*
2241 * cleanup function must be specified.
2242 * FIXME: This is only used for OPA_VNIC and that usage should be
2243 * removed too.
2244 */
2245 void (*free_rdma_netdev)(struct net_device *netdev);
2246
2247 /* control functions */
2248 void (*set_id)(struct net_device *netdev, int id);
2249 /* send packet */
2250 int (*send)(struct net_device *dev, struct sk_buff *skb,
2251 struct ib_ah *address, u32 dqpn);
2252 /* multicast */
2253 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2254 union ib_gid *gid, u16 mlid,
2255 int set_qkey, u32 qkey);
2256 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2257 union ib_gid *gid, u16 mlid);
2258 /* timeout */
2259 void (*tx_timeout)(struct net_device *dev, unsigned int txqueue);
2260};
2261
2262struct rdma_netdev_alloc_params {
2263 size_t sizeof_priv;
2264 unsigned int txqs;
2265 unsigned int rxqs;
2266 void *param;
2267
2268 int (*initialize_rdma_netdev)(struct ib_device *device, u32 port_num,
2269 struct net_device *netdev, void *param);
2270};
2271
2272struct ib_odp_counters {
2273 atomic64_t faults;
2274 atomic64_t invalidations;
2275 atomic64_t prefetch;
2276};
2277
2278struct ib_counters {
2279 struct ib_device *device;
2280 struct ib_uobject *uobject;
2281 /* num of objects attached */
2282 atomic_t usecnt;
2283};
2284
2285struct ib_counters_read_attr {
2286 u64 *counters_buff;
2287 u32 ncounters;
2288 u32 flags; /* use enum ib_read_counters_flags */
2289};
2290
2291struct uverbs_attr_bundle;
2292struct iw_cm_id;
2293struct iw_cm_conn_param;
2294
2295#define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2296 .size_##ib_struct = \
2297 (sizeof(struct drv_struct) + \
2298 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2299 BUILD_BUG_ON_ZERO( \
2300 !__same_type(((struct drv_struct *)NULL)->member, \
2301 struct ib_struct)))
2302
2303#define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2304 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2305 gfp, false))
2306
2307#define rdma_zalloc_drv_obj_numa(ib_dev, ib_type) \
2308 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2309 GFP_KERNEL, true))
2310
2311#define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2312 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2313
2314#define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2315
2316struct rdma_user_mmap_entry {
2317 struct kref ref;
2318 struct ib_ucontext *ucontext;
2319 unsigned long start_pgoff;
2320 size_t npages;
2321 bool driver_removed;
2322};
2323
2324/* Return the offset (in bytes) the user should pass to libc's mmap() */
2325static inline u64
2326rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2327{
2328 return (u64)entry->start_pgoff << PAGE_SHIFT;
2329}
2330
2331/**
2332 * struct ib_device_ops - InfiniBand device operations
2333 * This structure defines all the InfiniBand device operations, providers will
2334 * need to define the supported operations, otherwise they will be set to null.
2335 */
2336struct ib_device_ops {
2337 struct module *owner;
2338 enum rdma_driver_id driver_id;
2339 u32 uverbs_abi_ver;
2340 unsigned int uverbs_no_driver_id_binding:1;
2341
2342 /*
2343 * NOTE: New drivers should not make use of device_group; instead new
2344 * device parameter should be exposed via netlink command. This
2345 * mechanism exists only for existing drivers.
2346 */
2347 const struct attribute_group *device_group;
2348 const struct attribute_group **port_groups;
2349
2350 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2351 const struct ib_send_wr **bad_send_wr);
2352 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2353 const struct ib_recv_wr **bad_recv_wr);
2354 void (*drain_rq)(struct ib_qp *qp);
2355 void (*drain_sq)(struct ib_qp *qp);
2356 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2357 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2358 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2359 int (*post_srq_recv)(struct ib_srq *srq,
2360 const struct ib_recv_wr *recv_wr,
2361 const struct ib_recv_wr **bad_recv_wr);
2362 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2363 u32 port_num, const struct ib_wc *in_wc,
2364 const struct ib_grh *in_grh,
2365 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2366 size_t *out_mad_size, u16 *out_mad_pkey_index);
2367 int (*query_device)(struct ib_device *device,
2368 struct ib_device_attr *device_attr,
2369 struct ib_udata *udata);
2370 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2371 struct ib_device_modify *device_modify);
2372 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2373 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2374 int comp_vector);
2375 int (*query_port)(struct ib_device *device, u32 port_num,
2376 struct ib_port_attr *port_attr);
2377 int (*modify_port)(struct ib_device *device, u32 port_num,
2378 int port_modify_mask,
2379 struct ib_port_modify *port_modify);
2380 /**
2381 * The following mandatory functions are used only at device
2382 * registration. Keep functions such as these at the end of this
2383 * structure to avoid cache line misses when accessing struct ib_device
2384 * in fast paths.
2385 */
2386 int (*get_port_immutable)(struct ib_device *device, u32 port_num,
2387 struct ib_port_immutable *immutable);
2388 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2389 u32 port_num);
2390 /**
2391 * When calling get_netdev, the HW vendor's driver should return the
2392 * net device of device @device at port @port_num or NULL if such
2393 * a net device doesn't exist. The vendor driver should call dev_hold
2394 * on this net device. The HW vendor's device driver must guarantee
2395 * that this function returns NULL before the net device has finished
2396 * NETDEV_UNREGISTER state.
2397 */
2398 struct net_device *(*get_netdev)(struct ib_device *device,
2399 u32 port_num);
2400 /**
2401 * rdma netdev operation
2402 *
2403 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2404 * must return -EOPNOTSUPP if it doesn't support the specified type.
2405 */
2406 struct net_device *(*alloc_rdma_netdev)(
2407 struct ib_device *device, u32 port_num, enum rdma_netdev_t type,
2408 const char *name, unsigned char name_assign_type,
2409 void (*setup)(struct net_device *));
2410
2411 int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num,
2412 enum rdma_netdev_t type,
2413 struct rdma_netdev_alloc_params *params);
2414 /**
2415 * query_gid should be return GID value for @device, when @port_num
2416 * link layer is either IB or iWarp. It is no-op if @port_num port
2417 * is RoCE link layer.
2418 */
2419 int (*query_gid)(struct ib_device *device, u32 port_num, int index,
2420 union ib_gid *gid);
2421 /**
2422 * When calling add_gid, the HW vendor's driver should add the gid
2423 * of device of port at gid index available at @attr. Meta-info of
2424 * that gid (for example, the network device related to this gid) is
2425 * available at @attr. @context allows the HW vendor driver to store
2426 * extra information together with a GID entry. The HW vendor driver may
2427 * allocate memory to contain this information and store it in @context
2428 * when a new GID entry is written to. Params are consistent until the
2429 * next call of add_gid or delete_gid. The function should return 0 on
2430 * success or error otherwise. The function could be called
2431 * concurrently for different ports. This function is only called when
2432 * roce_gid_table is used.
2433 */
2434 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2435 /**
2436 * When calling del_gid, the HW vendor's driver should delete the
2437 * gid of device @device at gid index gid_index of port port_num
2438 * available in @attr.
2439 * Upon the deletion of a GID entry, the HW vendor must free any
2440 * allocated memory. The caller will clear @context afterwards.
2441 * This function is only called when roce_gid_table is used.
2442 */
2443 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2444 int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index,
2445 u16 *pkey);
2446 int (*alloc_ucontext)(struct ib_ucontext *context,
2447 struct ib_udata *udata);
2448 void (*dealloc_ucontext)(struct ib_ucontext *context);
2449 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2450 /**
2451 * This will be called once refcount of an entry in mmap_xa reaches
2452 * zero. The type of the memory that was mapped may differ between
2453 * entries and is opaque to the rdma_user_mmap interface.
2454 * Therefore needs to be implemented by the driver in mmap_free.
2455 */
2456 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2457 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2458 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2459 int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2460 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2461 struct ib_udata *udata);
2462 int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2463 struct ib_udata *udata);
2464 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2465 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2466 int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2467 int (*create_srq)(struct ib_srq *srq,
2468 struct ib_srq_init_attr *srq_init_attr,
2469 struct ib_udata *udata);
2470 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2471 enum ib_srq_attr_mask srq_attr_mask,
2472 struct ib_udata *udata);
2473 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2474 int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2475 int (*create_qp)(struct ib_qp *qp, struct ib_qp_init_attr *qp_init_attr,
2476 struct ib_udata *udata);
2477 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2478 int qp_attr_mask, struct ib_udata *udata);
2479 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2480 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2481 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2482 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2483 struct ib_udata *udata);
2484 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2485 int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2486 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2487 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2488 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2489 u64 virt_addr, int mr_access_flags,
2490 struct ib_udata *udata);
2491 struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset,
2492 u64 length, u64 virt_addr, int fd,
2493 int mr_access_flags,
2494 struct ib_udata *udata);
2495 struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
2496 u64 length, u64 virt_addr,
2497 int mr_access_flags, struct ib_pd *pd,
2498 struct ib_udata *udata);
2499 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2500 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2501 u32 max_num_sg);
2502 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2503 u32 max_num_data_sg,
2504 u32 max_num_meta_sg);
2505 int (*advise_mr)(struct ib_pd *pd,
2506 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2507 struct ib_sge *sg_list, u32 num_sge,
2508 struct uverbs_attr_bundle *attrs);
2509
2510 /*
2511 * Kernel users should universally support relaxed ordering (RO), as
2512 * they are designed to read data only after observing the CQE and use
2513 * the DMA API correctly.
2514 *
2515 * Some drivers implicitly enable RO if platform supports it.
2516 */
2517 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2518 unsigned int *sg_offset);
2519 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2520 struct ib_mr_status *mr_status);
2521 int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2522 int (*dealloc_mw)(struct ib_mw *mw);
2523 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2524 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2525 int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2526 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2527 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2528 struct ib_flow_attr *flow_attr,
2529 struct ib_udata *udata);
2530 int (*destroy_flow)(struct ib_flow *flow_id);
2531 int (*destroy_flow_action)(struct ib_flow_action *action);
2532 int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port,
2533 int state);
2534 int (*get_vf_config)(struct ib_device *device, int vf, u32 port,
2535 struct ifla_vf_info *ivf);
2536 int (*get_vf_stats)(struct ib_device *device, int vf, u32 port,
2537 struct ifla_vf_stats *stats);
2538 int (*get_vf_guid)(struct ib_device *device, int vf, u32 port,
2539 struct ifla_vf_guid *node_guid,
2540 struct ifla_vf_guid *port_guid);
2541 int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid,
2542 int type);
2543 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2544 struct ib_wq_init_attr *init_attr,
2545 struct ib_udata *udata);
2546 int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2547 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2548 u32 wq_attr_mask, struct ib_udata *udata);
2549 int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2550 struct ib_rwq_ind_table_init_attr *init_attr,
2551 struct ib_udata *udata);
2552 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2553 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2554 struct ib_ucontext *context,
2555 struct ib_dm_alloc_attr *attr,
2556 struct uverbs_attr_bundle *attrs);
2557 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2558 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2559 struct ib_dm_mr_attr *attr,
2560 struct uverbs_attr_bundle *attrs);
2561 int (*create_counters)(struct ib_counters *counters,
2562 struct uverbs_attr_bundle *attrs);
2563 int (*destroy_counters)(struct ib_counters *counters);
2564 int (*read_counters)(struct ib_counters *counters,
2565 struct ib_counters_read_attr *counters_read_attr,
2566 struct uverbs_attr_bundle *attrs);
2567 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2568 int data_sg_nents, unsigned int *data_sg_offset,
2569 struct scatterlist *meta_sg, int meta_sg_nents,
2570 unsigned int *meta_sg_offset);
2571
2572 /**
2573 * alloc_hw_[device,port]_stats - Allocate a struct rdma_hw_stats and
2574 * fill in the driver initialized data. The struct is kfree()'ed by
2575 * the sysfs core when the device is removed. A lifespan of -1 in the
2576 * return struct tells the core to set a default lifespan.
2577 */
2578 struct rdma_hw_stats *(*alloc_hw_device_stats)(struct ib_device *device);
2579 struct rdma_hw_stats *(*alloc_hw_port_stats)(struct ib_device *device,
2580 u32 port_num);
2581 /**
2582 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2583 * @index - The index in the value array we wish to have updated, or
2584 * num_counters if we want all stats updated
2585 * Return codes -
2586 * < 0 - Error, no counters updated
2587 * index - Updated the single counter pointed to by index
2588 * num_counters - Updated all counters (will reset the timestamp
2589 * and prevent further calls for lifespan milliseconds)
2590 * Drivers are allowed to update all counters in leiu of just the
2591 * one given in index at their option
2592 */
2593 int (*get_hw_stats)(struct ib_device *device,
2594 struct rdma_hw_stats *stats, u32 port, int index);
2595
2596 /**
2597 * modify_hw_stat - Modify the counter configuration
2598 * @enable: true/false when enable/disable a counter
2599 * Return codes - 0 on success or error code otherwise.
2600 */
2601 int (*modify_hw_stat)(struct ib_device *device, u32 port,
2602 unsigned int counter_index, bool enable);
2603 /**
2604 * Allows rdma drivers to add their own restrack attributes.
2605 */
2606 int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2607 int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2608 int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2609 int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2610 int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2611 int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2612 int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2613 int (*fill_res_srq_entry)(struct sk_buff *msg, struct ib_srq *ib_srq);
2614 int (*fill_res_srq_entry_raw)(struct sk_buff *msg, struct ib_srq *ib_srq);
2615
2616 /* Device lifecycle callbacks */
2617 /*
2618 * Called after the device becomes registered, before clients are
2619 * attached
2620 */
2621 int (*enable_driver)(struct ib_device *dev);
2622 /*
2623 * This is called as part of ib_dealloc_device().
2624 */
2625 void (*dealloc_driver)(struct ib_device *dev);
2626
2627 /* iWarp CM callbacks */
2628 void (*iw_add_ref)(struct ib_qp *qp);
2629 void (*iw_rem_ref)(struct ib_qp *qp);
2630 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2631 int (*iw_connect)(struct iw_cm_id *cm_id,
2632 struct iw_cm_conn_param *conn_param);
2633 int (*iw_accept)(struct iw_cm_id *cm_id,
2634 struct iw_cm_conn_param *conn_param);
2635 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2636 u8 pdata_len);
2637 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2638 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2639 /**
2640 * counter_bind_qp - Bind a QP to a counter.
2641 * @counter - The counter to be bound. If counter->id is zero then
2642 * the driver needs to allocate a new counter and set counter->id
2643 */
2644 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2645 /**
2646 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2647 * counter and bind it onto the default one
2648 */
2649 int (*counter_unbind_qp)(struct ib_qp *qp);
2650 /**
2651 * counter_dealloc -De-allocate the hw counter
2652 */
2653 int (*counter_dealloc)(struct rdma_counter *counter);
2654 /**
2655 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2656 * the driver initialized data.
2657 */
2658 struct rdma_hw_stats *(*counter_alloc_stats)(
2659 struct rdma_counter *counter);
2660 /**
2661 * counter_update_stats - Query the stats value of this counter
2662 */
2663 int (*counter_update_stats)(struct rdma_counter *counter);
2664
2665 /**
2666 * Allows rdma drivers to add their own restrack attributes
2667 * dumped via 'rdma stat' iproute2 command.
2668 */
2669 int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2670
2671 /* query driver for its ucontext properties */
2672 int (*query_ucontext)(struct ib_ucontext *context,
2673 struct uverbs_attr_bundle *attrs);
2674
2675 /*
2676 * Provide NUMA node. This API exists for rdmavt/hfi1 only.
2677 * Everyone else relies on Linux memory management model.
2678 */
2679 int (*get_numa_node)(struct ib_device *dev);
2680
2681 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2682 DECLARE_RDMA_OBJ_SIZE(ib_counters);
2683 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2684 DECLARE_RDMA_OBJ_SIZE(ib_mw);
2685 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2686 DECLARE_RDMA_OBJ_SIZE(ib_qp);
2687 DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2688 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2689 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2690 DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2691};
2692
2693struct ib_core_device {
2694 /* device must be the first element in structure until,
2695 * union of ib_core_device and device exists in ib_device.
2696 */
2697 struct device dev;
2698 possible_net_t rdma_net;
2699 struct kobject *ports_kobj;
2700 struct list_head port_list;
2701 struct ib_device *owner; /* reach back to owner ib_device */
2702};
2703
2704struct rdma_restrack_root;
2705struct ib_device {
2706 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2707 struct device *dma_device;
2708 struct ib_device_ops ops;
2709 char name[IB_DEVICE_NAME_MAX];
2710 struct rcu_head rcu_head;
2711
2712 struct list_head event_handler_list;
2713 /* Protects event_handler_list */
2714 struct rw_semaphore event_handler_rwsem;
2715
2716 /* Protects QP's event_handler calls and open_qp list */
2717 spinlock_t qp_open_list_lock;
2718
2719 struct rw_semaphore client_data_rwsem;
2720 struct xarray client_data;
2721 struct mutex unregistration_lock;
2722
2723 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2724 rwlock_t cache_lock;
2725 /**
2726 * port_data is indexed by port number
2727 */
2728 struct ib_port_data *port_data;
2729
2730 int num_comp_vectors;
2731
2732 union {
2733 struct device dev;
2734 struct ib_core_device coredev;
2735 };
2736
2737 /* First group is for device attributes,
2738 * Second group is for driver provided attributes (optional).
2739 * Third group is for the hw_stats
2740 * It is a NULL terminated array.
2741 */
2742 const struct attribute_group *groups[4];
2743
2744 u64 uverbs_cmd_mask;
2745
2746 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2747 __be64 node_guid;
2748 u32 local_dma_lkey;
2749 u16 is_switch:1;
2750 /* Indicates kernel verbs support, should not be used in drivers */
2751 u16 kverbs_provider:1;
2752 /* CQ adaptive moderation (RDMA DIM) */
2753 u16 use_cq_dim:1;
2754 u8 node_type;
2755 u32 phys_port_cnt;
2756 struct ib_device_attr attrs;
2757 struct hw_stats_device_data *hw_stats_data;
2758
2759#ifdef CONFIG_CGROUP_RDMA
2760 struct rdmacg_device cg_device;
2761#endif
2762
2763 u32 index;
2764
2765 spinlock_t cq_pools_lock;
2766 struct list_head cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2767
2768 struct rdma_restrack_root *res;
2769
2770 const struct uapi_definition *driver_def;
2771
2772 /*
2773 * Positive refcount indicates that the device is currently
2774 * registered and cannot be unregistered.
2775 */
2776 refcount_t refcount;
2777 struct completion unreg_completion;
2778 struct work_struct unregistration_work;
2779
2780 const struct rdma_link_ops *link_ops;
2781
2782 /* Protects compat_devs xarray modifications */
2783 struct mutex compat_devs_mutex;
2784 /* Maintains compat devices for each net namespace */
2785 struct xarray compat_devs;
2786
2787 /* Used by iWarp CM */
2788 char iw_ifname[IFNAMSIZ];
2789 u32 iw_driver_flags;
2790 u32 lag_flags;
2791};
2792
2793static inline void *rdma_zalloc_obj(struct ib_device *dev, size_t size,
2794 gfp_t gfp, bool is_numa_aware)
2795{
2796 if (is_numa_aware && dev->ops.get_numa_node)
2797 return kzalloc_node(size, gfp, dev->ops.get_numa_node(dev));
2798
2799 return kzalloc(size, gfp);
2800}
2801
2802struct ib_client_nl_info;
2803struct ib_client {
2804 const char *name;
2805 int (*add)(struct ib_device *ibdev);
2806 void (*remove)(struct ib_device *, void *client_data);
2807 void (*rename)(struct ib_device *dev, void *client_data);
2808 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2809 struct ib_client_nl_info *res);
2810 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2811
2812 /* Returns the net_dev belonging to this ib_client and matching the
2813 * given parameters.
2814 * @dev: An RDMA device that the net_dev use for communication.
2815 * @port: A physical port number on the RDMA device.
2816 * @pkey: P_Key that the net_dev uses if applicable.
2817 * @gid: A GID that the net_dev uses to communicate.
2818 * @addr: An IP address the net_dev is configured with.
2819 * @client_data: The device's client data set by ib_set_client_data().
2820 *
2821 * An ib_client that implements a net_dev on top of RDMA devices
2822 * (such as IP over IB) should implement this callback, allowing the
2823 * rdma_cm module to find the right net_dev for a given request.
2824 *
2825 * The caller is responsible for calling dev_put on the returned
2826 * netdev. */
2827 struct net_device *(*get_net_dev_by_params)(
2828 struct ib_device *dev,
2829 u32 port,
2830 u16 pkey,
2831 const union ib_gid *gid,
2832 const struct sockaddr *addr,
2833 void *client_data);
2834
2835 refcount_t uses;
2836 struct completion uses_zero;
2837 u32 client_id;
2838
2839 /* kverbs are not required by the client */
2840 u8 no_kverbs_req:1;
2841};
2842
2843/*
2844 * IB block DMA iterator
2845 *
2846 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2847 * to a HW supported page size.
2848 */
2849struct ib_block_iter {
2850 /* internal states */
2851 struct scatterlist *__sg; /* sg holding the current aligned block */
2852 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2853 size_t __sg_numblocks; /* ib_umem_num_dma_blocks() */
2854 unsigned int __sg_nents; /* number of SG entries */
2855 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2856 unsigned int __pg_bit; /* alignment of current block */
2857};
2858
2859struct ib_device *_ib_alloc_device(size_t size);
2860#define ib_alloc_device(drv_struct, member) \
2861 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2862 BUILD_BUG_ON_ZERO(offsetof( \
2863 struct drv_struct, member))), \
2864 struct drv_struct, member)
2865
2866void ib_dealloc_device(struct ib_device *device);
2867
2868void ib_get_device_fw_str(struct ib_device *device, char *str);
2869
2870int ib_register_device(struct ib_device *device, const char *name,
2871 struct device *dma_device);
2872void ib_unregister_device(struct ib_device *device);
2873void ib_unregister_driver(enum rdma_driver_id driver_id);
2874void ib_unregister_device_and_put(struct ib_device *device);
2875void ib_unregister_device_queued(struct ib_device *ib_dev);
2876
2877int ib_register_client (struct ib_client *client);
2878void ib_unregister_client(struct ib_client *client);
2879
2880void __rdma_block_iter_start(struct ib_block_iter *biter,
2881 struct scatterlist *sglist,
2882 unsigned int nents,
2883 unsigned long pgsz);
2884bool __rdma_block_iter_next(struct ib_block_iter *biter);
2885
2886/**
2887 * rdma_block_iter_dma_address - get the aligned dma address of the current
2888 * block held by the block iterator.
2889 * @biter: block iterator holding the memory block
2890 */
2891static inline dma_addr_t
2892rdma_block_iter_dma_address(struct ib_block_iter *biter)
2893{
2894 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2895}
2896
2897/**
2898 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2899 * @sglist: sglist to iterate over
2900 * @biter: block iterator holding the memory block
2901 * @nents: maximum number of sg entries to iterate over
2902 * @pgsz: best HW supported page size to use
2903 *
2904 * Callers may use rdma_block_iter_dma_address() to get each
2905 * blocks aligned DMA address.
2906 */
2907#define rdma_for_each_block(sglist, biter, nents, pgsz) \
2908 for (__rdma_block_iter_start(biter, sglist, nents, \
2909 pgsz); \
2910 __rdma_block_iter_next(biter);)
2911
2912/**
2913 * ib_get_client_data - Get IB client context
2914 * @device:Device to get context for
2915 * @client:Client to get context for
2916 *
2917 * ib_get_client_data() returns the client context data set with
2918 * ib_set_client_data(). This can only be called while the client is
2919 * registered to the device, once the ib_client remove() callback returns this
2920 * cannot be called.
2921 */
2922static inline void *ib_get_client_data(struct ib_device *device,
2923 struct ib_client *client)
2924{
2925 return xa_load(&device->client_data, client->client_id);
2926}
2927void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2928 void *data);
2929void ib_set_device_ops(struct ib_device *device,
2930 const struct ib_device_ops *ops);
2931
2932int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2933 unsigned long pfn, unsigned long size, pgprot_t prot,
2934 struct rdma_user_mmap_entry *entry);
2935int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2936 struct rdma_user_mmap_entry *entry,
2937 size_t length);
2938int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2939 struct rdma_user_mmap_entry *entry,
2940 size_t length, u32 min_pgoff,
2941 u32 max_pgoff);
2942
2943static inline int
2944rdma_user_mmap_entry_insert_exact(struct ib_ucontext *ucontext,
2945 struct rdma_user_mmap_entry *entry,
2946 size_t length, u32 pgoff)
2947{
2948 return rdma_user_mmap_entry_insert_range(ucontext, entry, length, pgoff,
2949 pgoff);
2950}
2951
2952struct rdma_user_mmap_entry *
2953rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2954 unsigned long pgoff);
2955struct rdma_user_mmap_entry *
2956rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2957 struct vm_area_struct *vma);
2958void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2959
2960void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2961
2962static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2963{
2964 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2965}
2966
2967static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2968{
2969 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2970}
2971
2972static inline bool ib_is_buffer_cleared(const void __user *p,
2973 size_t len)
2974{
2975 bool ret;
2976 u8 *buf;
2977
2978 if (len > USHRT_MAX)
2979 return false;
2980
2981 buf = memdup_user(p, len);
2982 if (IS_ERR(buf))
2983 return false;
2984
2985 ret = !memchr_inv(buf, 0, len);
2986 kfree(buf);
2987 return ret;
2988}
2989
2990static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2991 size_t offset,
2992 size_t len)
2993{
2994 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2995}
2996
2997/**
2998 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2999 * contains all required attributes and no attributes not allowed for
3000 * the given QP state transition.
3001 * @cur_state: Current QP state
3002 * @next_state: Next QP state
3003 * @type: QP type
3004 * @mask: Mask of supplied QP attributes
3005 *
3006 * This function is a helper function that a low-level driver's
3007 * modify_qp method can use to validate the consumer's input. It
3008 * checks that cur_state and next_state are valid QP states, that a
3009 * transition from cur_state to next_state is allowed by the IB spec,
3010 * and that the attribute mask supplied is allowed for the transition.
3011 */
3012bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
3013 enum ib_qp_type type, enum ib_qp_attr_mask mask);
3014
3015void ib_register_event_handler(struct ib_event_handler *event_handler);
3016void ib_unregister_event_handler(struct ib_event_handler *event_handler);
3017void ib_dispatch_event(const struct ib_event *event);
3018
3019int ib_query_port(struct ib_device *device,
3020 u32 port_num, struct ib_port_attr *port_attr);
3021
3022enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
3023 u32 port_num);
3024
3025/**
3026 * rdma_cap_ib_switch - Check if the device is IB switch
3027 * @device: Device to check
3028 *
3029 * Device driver is responsible for setting is_switch bit on
3030 * in ib_device structure at init time.
3031 *
3032 * Return: true if the device is IB switch.
3033 */
3034static inline bool rdma_cap_ib_switch(const struct ib_device *device)
3035{
3036 return device->is_switch;
3037}
3038
3039/**
3040 * rdma_start_port - Return the first valid port number for the device
3041 * specified
3042 *
3043 * @device: Device to be checked
3044 *
3045 * Return start port number
3046 */
3047static inline u32 rdma_start_port(const struct ib_device *device)
3048{
3049 return rdma_cap_ib_switch(device) ? 0 : 1;
3050}
3051
3052/**
3053 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3054 * @device - The struct ib_device * to iterate over
3055 * @iter - The unsigned int to store the port number
3056 */
3057#define rdma_for_each_port(device, iter) \
3058 for (iter = rdma_start_port(device + \
3059 BUILD_BUG_ON_ZERO(!__same_type(u32, \
3060 iter))); \
3061 iter <= rdma_end_port(device); iter++)
3062
3063/**
3064 * rdma_end_port - Return the last valid port number for the device
3065 * specified
3066 *
3067 * @device: Device to be checked
3068 *
3069 * Return last port number
3070 */
3071static inline u32 rdma_end_port(const struct ib_device *device)
3072{
3073 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3074}
3075
3076static inline int rdma_is_port_valid(const struct ib_device *device,
3077 unsigned int port)
3078{
3079 return (port >= rdma_start_port(device) &&
3080 port <= rdma_end_port(device));
3081}
3082
3083static inline bool rdma_is_grh_required(const struct ib_device *device,
3084 u32 port_num)
3085{
3086 return device->port_data[port_num].immutable.core_cap_flags &
3087 RDMA_CORE_PORT_IB_GRH_REQUIRED;
3088}
3089
3090static inline bool rdma_protocol_ib(const struct ib_device *device,
3091 u32 port_num)
3092{
3093 return device->port_data[port_num].immutable.core_cap_flags &
3094 RDMA_CORE_CAP_PROT_IB;
3095}
3096
3097static inline bool rdma_protocol_roce(const struct ib_device *device,
3098 u32 port_num)
3099{
3100 return device->port_data[port_num].immutable.core_cap_flags &
3101 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3102}
3103
3104static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device,
3105 u32 port_num)
3106{
3107 return device->port_data[port_num].immutable.core_cap_flags &
3108 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3109}
3110
3111static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device,
3112 u32 port_num)
3113{
3114 return device->port_data[port_num].immutable.core_cap_flags &
3115 RDMA_CORE_CAP_PROT_ROCE;
3116}
3117
3118static inline bool rdma_protocol_iwarp(const struct ib_device *device,
3119 u32 port_num)
3120{
3121 return device->port_data[port_num].immutable.core_cap_flags &
3122 RDMA_CORE_CAP_PROT_IWARP;
3123}
3124
3125static inline bool rdma_ib_or_roce(const struct ib_device *device,
3126 u32 port_num)
3127{
3128 return rdma_protocol_ib(device, port_num) ||
3129 rdma_protocol_roce(device, port_num);
3130}
3131
3132static inline bool rdma_protocol_raw_packet(const struct ib_device *device,
3133 u32 port_num)
3134{
3135 return device->port_data[port_num].immutable.core_cap_flags &
3136 RDMA_CORE_CAP_PROT_RAW_PACKET;
3137}
3138
3139static inline bool rdma_protocol_usnic(const struct ib_device *device,
3140 u32 port_num)
3141{
3142 return device->port_data[port_num].immutable.core_cap_flags &
3143 RDMA_CORE_CAP_PROT_USNIC;
3144}
3145
3146/**
3147 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3148 * Management Datagrams.
3149 * @device: Device to check
3150 * @port_num: Port number to check
3151 *
3152 * Management Datagrams (MAD) are a required part of the InfiniBand
3153 * specification and are supported on all InfiniBand devices. A slightly
3154 * extended version are also supported on OPA interfaces.
3155 *
3156 * Return: true if the port supports sending/receiving of MAD packets.
3157 */
3158static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num)
3159{
3160 return device->port_data[port_num].immutable.core_cap_flags &
3161 RDMA_CORE_CAP_IB_MAD;
3162}
3163
3164/**
3165 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3166 * Management Datagrams.
3167 * @device: Device to check
3168 * @port_num: Port number to check
3169 *
3170 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3171 * datagrams with their own versions. These OPA MADs share many but not all of
3172 * the characteristics of InfiniBand MADs.
3173 *
3174 * OPA MADs differ in the following ways:
3175 *
3176 * 1) MADs are variable size up to 2K
3177 * IBTA defined MADs remain fixed at 256 bytes
3178 * 2) OPA SMPs must carry valid PKeys
3179 * 3) OPA SMP packets are a different format
3180 *
3181 * Return: true if the port supports OPA MAD packet formats.
3182 */
3183static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num)
3184{
3185 return device->port_data[port_num].immutable.core_cap_flags &
3186 RDMA_CORE_CAP_OPA_MAD;
3187}
3188
3189/**
3190 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3191 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3192 * @device: Device to check
3193 * @port_num: Port number to check
3194 *
3195 * Each InfiniBand node is required to provide a Subnet Management Agent
3196 * that the subnet manager can access. Prior to the fabric being fully
3197 * configured by the subnet manager, the SMA is accessed via a well known
3198 * interface called the Subnet Management Interface (SMI). This interface
3199 * uses directed route packets to communicate with the SM to get around the
3200 * chicken and egg problem of the SM needing to know what's on the fabric
3201 * in order to configure the fabric, and needing to configure the fabric in
3202 * order to send packets to the devices on the fabric. These directed
3203 * route packets do not need the fabric fully configured in order to reach
3204 * their destination. The SMI is the only method allowed to send
3205 * directed route packets on an InfiniBand fabric.
3206 *
3207 * Return: true if the port provides an SMI.
3208 */
3209static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num)
3210{
3211 return device->port_data[port_num].immutable.core_cap_flags &
3212 RDMA_CORE_CAP_IB_SMI;
3213}
3214
3215/**
3216 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3217 * Communication Manager.
3218 * @device: Device to check
3219 * @port_num: Port number to check
3220 *
3221 * The InfiniBand Communication Manager is one of many pre-defined General
3222 * Service Agents (GSA) that are accessed via the General Service
3223 * Interface (GSI). It's role is to facilitate establishment of connections
3224 * between nodes as well as other management related tasks for established
3225 * connections.
3226 *
3227 * Return: true if the port supports an IB CM (this does not guarantee that
3228 * a CM is actually running however).
3229 */
3230static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num)
3231{
3232 return device->port_data[port_num].immutable.core_cap_flags &
3233 RDMA_CORE_CAP_IB_CM;
3234}
3235
3236/**
3237 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3238 * Communication Manager.
3239 * @device: Device to check
3240 * @port_num: Port number to check
3241 *
3242 * Similar to above, but specific to iWARP connections which have a different
3243 * managment protocol than InfiniBand.
3244 *
3245 * Return: true if the port supports an iWARP CM (this does not guarantee that
3246 * a CM is actually running however).
3247 */
3248static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num)
3249{
3250 return device->port_data[port_num].immutable.core_cap_flags &
3251 RDMA_CORE_CAP_IW_CM;
3252}
3253
3254/**
3255 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3256 * Subnet Administration.
3257 * @device: Device to check
3258 * @port_num: Port number to check
3259 *
3260 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3261 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3262 * fabrics, devices should resolve routes to other hosts by contacting the
3263 * SA to query the proper route.
3264 *
3265 * Return: true if the port should act as a client to the fabric Subnet
3266 * Administration interface. This does not imply that the SA service is
3267 * running locally.
3268 */
3269static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num)
3270{
3271 return device->port_data[port_num].immutable.core_cap_flags &
3272 RDMA_CORE_CAP_IB_SA;
3273}
3274
3275/**
3276 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3277 * Multicast.
3278 * @device: Device to check
3279 * @port_num: Port number to check
3280 *
3281 * InfiniBand multicast registration is more complex than normal IPv4 or
3282 * IPv6 multicast registration. Each Host Channel Adapter must register
3283 * with the Subnet Manager when it wishes to join a multicast group. It
3284 * should do so only once regardless of how many queue pairs it subscribes
3285 * to this group. And it should leave the group only after all queue pairs
3286 * attached to the group have been detached.
3287 *
3288 * Return: true if the port must undertake the additional adminstrative
3289 * overhead of registering/unregistering with the SM and tracking of the
3290 * total number of queue pairs attached to the multicast group.
3291 */
3292static inline bool rdma_cap_ib_mcast(const struct ib_device *device,
3293 u32 port_num)
3294{
3295 return rdma_cap_ib_sa(device, port_num);
3296}
3297
3298/**
3299 * rdma_cap_af_ib - Check if the port of device has the capability
3300 * Native Infiniband Address.
3301 * @device: Device to check
3302 * @port_num: Port number to check
3303 *
3304 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3305 * GID. RoCE uses a different mechanism, but still generates a GID via
3306 * a prescribed mechanism and port specific data.
3307 *
3308 * Return: true if the port uses a GID address to identify devices on the
3309 * network.
3310 */
3311static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num)
3312{
3313 return device->port_data[port_num].immutable.core_cap_flags &
3314 RDMA_CORE_CAP_AF_IB;
3315}
3316
3317/**
3318 * rdma_cap_eth_ah - Check if the port of device has the capability
3319 * Ethernet Address Handle.
3320 * @device: Device to check
3321 * @port_num: Port number to check
3322 *
3323 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3324 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3325 * port. Normally, packet headers are generated by the sending host
3326 * adapter, but when sending connectionless datagrams, we must manually
3327 * inject the proper headers for the fabric we are communicating over.
3328 *
3329 * Return: true if we are running as a RoCE port and must force the
3330 * addition of a Global Route Header built from our Ethernet Address
3331 * Handle into our header list for connectionless packets.
3332 */
3333static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num)
3334{
3335 return device->port_data[port_num].immutable.core_cap_flags &
3336 RDMA_CORE_CAP_ETH_AH;
3337}
3338
3339/**
3340 * rdma_cap_opa_ah - Check if the port of device supports
3341 * OPA Address handles
3342 * @device: Device to check
3343 * @port_num: Port number to check
3344 *
3345 * Return: true if we are running on an OPA device which supports
3346 * the extended OPA addressing.
3347 */
3348static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num)
3349{
3350 return (device->port_data[port_num].immutable.core_cap_flags &
3351 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3352}
3353
3354/**
3355 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3356 *
3357 * @device: Device
3358 * @port_num: Port number
3359 *
3360 * This MAD size includes the MAD headers and MAD payload. No other headers
3361 * are included.
3362 *
3363 * Return the max MAD size required by the Port. Will return 0 if the port
3364 * does not support MADs
3365 */
3366static inline size_t rdma_max_mad_size(const struct ib_device *device,
3367 u32 port_num)
3368{
3369 return device->port_data[port_num].immutable.max_mad_size;
3370}
3371
3372/**
3373 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3374 * @device: Device to check
3375 * @port_num: Port number to check
3376 *
3377 * RoCE GID table mechanism manages the various GIDs for a device.
3378 *
3379 * NOTE: if allocating the port's GID table has failed, this call will still
3380 * return true, but any RoCE GID table API will fail.
3381 *
3382 * Return: true if the port uses RoCE GID table mechanism in order to manage
3383 * its GIDs.
3384 */
3385static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3386 u32 port_num)
3387{
3388 return rdma_protocol_roce(device, port_num) &&
3389 device->ops.add_gid && device->ops.del_gid;
3390}
3391
3392/*
3393 * Check if the device supports READ W/ INVALIDATE.
3394 */
3395static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3396{
3397 /*
3398 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3399 * has support for it yet.
3400 */
3401 return rdma_protocol_iwarp(dev, port_num);
3402}
3403
3404/**
3405 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3406 * @device: Device
3407 * @port_num: 1 based Port number
3408 *
3409 * Return true if port is an Intel OPA port , false if not
3410 */
3411static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3412 u32 port_num)
3413{
3414 return (device->port_data[port_num].immutable.core_cap_flags &
3415 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3416}
3417
3418/**
3419 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3420 * @device: Device
3421 * @port_num: Port number
3422 * @mtu: enum value of MTU
3423 *
3424 * Return the MTU size supported by the port as an integer value. Will return
3425 * -1 if enum value of mtu is not supported.
3426 */
3427static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port,
3428 int mtu)
3429{
3430 if (rdma_core_cap_opa_port(device, port))
3431 return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3432 else
3433 return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3434}
3435
3436/**
3437 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3438 * @device: Device
3439 * @port_num: Port number
3440 * @attr: port attribute
3441 *
3442 * Return the MTU size supported by the port as an integer value.
3443 */
3444static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port,
3445 struct ib_port_attr *attr)
3446{
3447 if (rdma_core_cap_opa_port(device, port))
3448 return attr->phys_mtu;
3449 else
3450 return ib_mtu_enum_to_int(attr->max_mtu);
3451}
3452
3453int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port,
3454 int state);
3455int ib_get_vf_config(struct ib_device *device, int vf, u32 port,
3456 struct ifla_vf_info *info);
3457int ib_get_vf_stats(struct ib_device *device, int vf, u32 port,
3458 struct ifla_vf_stats *stats);
3459int ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
3460 struct ifla_vf_guid *node_guid,
3461 struct ifla_vf_guid *port_guid);
3462int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid,
3463 int type);
3464
3465int ib_query_pkey(struct ib_device *device,
3466 u32 port_num, u16 index, u16 *pkey);
3467
3468int ib_modify_device(struct ib_device *device,
3469 int device_modify_mask,
3470 struct ib_device_modify *device_modify);
3471
3472int ib_modify_port(struct ib_device *device,
3473 u32 port_num, int port_modify_mask,
3474 struct ib_port_modify *port_modify);
3475
3476int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3477 u32 *port_num, u16 *index);
3478
3479int ib_find_pkey(struct ib_device *device,
3480 u32 port_num, u16 pkey, u16 *index);
3481
3482enum ib_pd_flags {
3483 /*
3484 * Create a memory registration for all memory in the system and place
3485 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3486 * ULPs to avoid the overhead of dynamic MRs.
3487 *
3488 * This flag is generally considered unsafe and must only be used in
3489 * extremly trusted environments. Every use of it will log a warning
3490 * in the kernel log.
3491 */
3492 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3493};
3494
3495struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3496 const char *caller);
3497
3498/**
3499 * ib_alloc_pd - Allocates an unused protection domain.
3500 * @device: The device on which to allocate the protection domain.
3501 * @flags: protection domain flags
3502 *
3503 * A protection domain object provides an association between QPs, shared
3504 * receive queues, address handles, memory regions, and memory windows.
3505 *
3506 * Every PD has a local_dma_lkey which can be used as the lkey value for local
3507 * memory operations.
3508 */
3509#define ib_alloc_pd(device, flags) \
3510 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3511
3512int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3513
3514/**
3515 * ib_dealloc_pd - Deallocate kernel PD
3516 * @pd: The protection domain
3517 *
3518 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3519 */
3520static inline void ib_dealloc_pd(struct ib_pd *pd)
3521{
3522 int ret = ib_dealloc_pd_user(pd, NULL);
3523
3524 WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3525}
3526
3527enum rdma_create_ah_flags {
3528 /* In a sleepable context */
3529 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3530};
3531
3532/**
3533 * rdma_create_ah - Creates an address handle for the given address vector.
3534 * @pd: The protection domain associated with the address handle.
3535 * @ah_attr: The attributes of the address vector.
3536 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3537 *
3538 * The address handle is used to reference a local or global destination
3539 * in all UD QP post sends.
3540 */
3541struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3542 u32 flags);
3543
3544/**
3545 * rdma_create_user_ah - Creates an address handle for the given address vector.
3546 * It resolves destination mac address for ah attribute of RoCE type.
3547 * @pd: The protection domain associated with the address handle.
3548 * @ah_attr: The attributes of the address vector.
3549 * @udata: pointer to user's input output buffer information need by
3550 * provider driver.
3551 *
3552 * It returns 0 on success and returns appropriate error code on error.
3553 * The address handle is used to reference a local or global destination
3554 * in all UD QP post sends.
3555 */
3556struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3557 struct rdma_ah_attr *ah_attr,
3558 struct ib_udata *udata);
3559/**
3560 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3561 * work completion.
3562 * @hdr: the L3 header to parse
3563 * @net_type: type of header to parse
3564 * @sgid: place to store source gid
3565 * @dgid: place to store destination gid
3566 */
3567int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3568 enum rdma_network_type net_type,
3569 union ib_gid *sgid, union ib_gid *dgid);
3570
3571/**
3572 * ib_get_rdma_header_version - Get the header version
3573 * @hdr: the L3 header to parse
3574 */
3575int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3576
3577/**
3578 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3579 * work completion.
3580 * @device: Device on which the received message arrived.
3581 * @port_num: Port on which the received message arrived.
3582 * @wc: Work completion associated with the received message.
3583 * @grh: References the received global route header. This parameter is
3584 * ignored unless the work completion indicates that the GRH is valid.
3585 * @ah_attr: Returned attributes that can be used when creating an address
3586 * handle for replying to the message.
3587 * When ib_init_ah_attr_from_wc() returns success,
3588 * (a) for IB link layer it optionally contains a reference to SGID attribute
3589 * when GRH is present for IB link layer.
3590 * (b) for RoCE link layer it contains a reference to SGID attribute.
3591 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3592 * attributes which are initialized using ib_init_ah_attr_from_wc().
3593 *
3594 */
3595int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num,
3596 const struct ib_wc *wc, const struct ib_grh *grh,
3597 struct rdma_ah_attr *ah_attr);
3598
3599/**
3600 * ib_create_ah_from_wc - Creates an address handle associated with the
3601 * sender of the specified work completion.
3602 * @pd: The protection domain associated with the address handle.
3603 * @wc: Work completion information associated with a received message.
3604 * @grh: References the received global route header. This parameter is
3605 * ignored unless the work completion indicates that the GRH is valid.
3606 * @port_num: The outbound port number to associate with the address.
3607 *
3608 * The address handle is used to reference a local or global destination
3609 * in all UD QP post sends.
3610 */
3611struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3612 const struct ib_grh *grh, u32 port_num);
3613
3614/**
3615 * rdma_modify_ah - Modifies the address vector associated with an address
3616 * handle.
3617 * @ah: The address handle to modify.
3618 * @ah_attr: The new address vector attributes to associate with the
3619 * address handle.
3620 */
3621int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3622
3623/**
3624 * rdma_query_ah - Queries the address vector associated with an address
3625 * handle.
3626 * @ah: The address handle to query.
3627 * @ah_attr: The address vector attributes associated with the address
3628 * handle.
3629 */
3630int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3631
3632enum rdma_destroy_ah_flags {
3633 /* In a sleepable context */
3634 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3635};
3636
3637/**
3638 * rdma_destroy_ah_user - Destroys an address handle.
3639 * @ah: The address handle to destroy.
3640 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3641 * @udata: Valid user data or NULL for kernel objects
3642 */
3643int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3644
3645/**
3646 * rdma_destroy_ah - Destroys an kernel address handle.
3647 * @ah: The address handle to destroy.
3648 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3649 *
3650 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3651 */
3652static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3653{
3654 int ret = rdma_destroy_ah_user(ah, flags, NULL);
3655
3656 WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3657}
3658
3659struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3660 struct ib_srq_init_attr *srq_init_attr,
3661 struct ib_usrq_object *uobject,
3662 struct ib_udata *udata);
3663static inline struct ib_srq *
3664ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3665{
3666 if (!pd->device->ops.create_srq)
3667 return ERR_PTR(-EOPNOTSUPP);
3668
3669 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3670}
3671
3672/**
3673 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3674 * @srq: The SRQ to modify.
3675 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3676 * the current values of selected SRQ attributes are returned.
3677 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3678 * are being modified.
3679 *
3680 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3681 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3682 * the number of receives queued drops below the limit.
3683 */
3684int ib_modify_srq(struct ib_srq *srq,
3685 struct ib_srq_attr *srq_attr,
3686 enum ib_srq_attr_mask srq_attr_mask);
3687
3688/**
3689 * ib_query_srq - Returns the attribute list and current values for the
3690 * specified SRQ.
3691 * @srq: The SRQ to query.
3692 * @srq_attr: The attributes of the specified SRQ.
3693 */
3694int ib_query_srq(struct ib_srq *srq,
3695 struct ib_srq_attr *srq_attr);
3696
3697/**
3698 * ib_destroy_srq_user - Destroys the specified SRQ.
3699 * @srq: The SRQ to destroy.
3700 * @udata: Valid user data or NULL for kernel objects
3701 */
3702int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3703
3704/**
3705 * ib_destroy_srq - Destroys the specified kernel SRQ.
3706 * @srq: The SRQ to destroy.
3707 *
3708 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3709 */
3710static inline void ib_destroy_srq(struct ib_srq *srq)
3711{
3712 int ret = ib_destroy_srq_user(srq, NULL);
3713
3714 WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3715}
3716
3717/**
3718 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3719 * @srq: The SRQ to post the work request on.
3720 * @recv_wr: A list of work requests to post on the receive queue.
3721 * @bad_recv_wr: On an immediate failure, this parameter will reference
3722 * the work request that failed to be posted on the QP.
3723 */
3724static inline int ib_post_srq_recv(struct ib_srq *srq,
3725 const struct ib_recv_wr *recv_wr,
3726 const struct ib_recv_wr **bad_recv_wr)
3727{
3728 const struct ib_recv_wr *dummy;
3729
3730 return srq->device->ops.post_srq_recv(srq, recv_wr,
3731 bad_recv_wr ? : &dummy);
3732}
3733
3734struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd,
3735 struct ib_qp_init_attr *qp_init_attr,
3736 const char *caller);
3737/**
3738 * ib_create_qp - Creates a kernel QP associated with the specific protection
3739 * domain.
3740 * @pd: The protection domain associated with the QP.
3741 * @init_attr: A list of initial attributes required to create the
3742 * QP. If QP creation succeeds, then the attributes are updated to
3743 * the actual capabilities of the created QP.
3744 */
3745static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3746 struct ib_qp_init_attr *init_attr)
3747{
3748 return ib_create_qp_kernel(pd, init_attr, KBUILD_MODNAME);
3749}
3750
3751/**
3752 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3753 * @qp: The QP to modify.
3754 * @attr: On input, specifies the QP attributes to modify. On output,
3755 * the current values of selected QP attributes are returned.
3756 * @attr_mask: A bit-mask used to specify which attributes of the QP
3757 * are being modified.
3758 * @udata: pointer to user's input output buffer information
3759 * are being modified.
3760 * It returns 0 on success and returns appropriate error code on error.
3761 */
3762int ib_modify_qp_with_udata(struct ib_qp *qp,
3763 struct ib_qp_attr *attr,
3764 int attr_mask,
3765 struct ib_udata *udata);
3766
3767/**
3768 * ib_modify_qp - Modifies the attributes for the specified QP and then
3769 * transitions the QP to the given state.
3770 * @qp: The QP to modify.
3771 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3772 * the current values of selected QP attributes are returned.
3773 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3774 * are being modified.
3775 */
3776int ib_modify_qp(struct ib_qp *qp,
3777 struct ib_qp_attr *qp_attr,
3778 int qp_attr_mask);
3779
3780/**
3781 * ib_query_qp - Returns the attribute list and current values for the
3782 * specified QP.
3783 * @qp: The QP to query.
3784 * @qp_attr: The attributes of the specified QP.
3785 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3786 * @qp_init_attr: Additional attributes of the selected QP.
3787 *
3788 * The qp_attr_mask may be used to limit the query to gathering only the
3789 * selected attributes.
3790 */
3791int ib_query_qp(struct ib_qp *qp,
3792 struct ib_qp_attr *qp_attr,
3793 int qp_attr_mask,
3794 struct ib_qp_init_attr *qp_init_attr);
3795
3796/**
3797 * ib_destroy_qp - Destroys the specified QP.
3798 * @qp: The QP to destroy.
3799 * @udata: Valid udata or NULL for kernel objects
3800 */
3801int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3802
3803/**
3804 * ib_destroy_qp - Destroys the specified kernel QP.
3805 * @qp: The QP to destroy.
3806 *
3807 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3808 */
3809static inline int ib_destroy_qp(struct ib_qp *qp)
3810{
3811 return ib_destroy_qp_user(qp, NULL);
3812}
3813
3814/**
3815 * ib_open_qp - Obtain a reference to an existing sharable QP.
3816 * @xrcd - XRC domain
3817 * @qp_open_attr: Attributes identifying the QP to open.
3818 *
3819 * Returns a reference to a sharable QP.
3820 */
3821struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3822 struct ib_qp_open_attr *qp_open_attr);
3823
3824/**
3825 * ib_close_qp - Release an external reference to a QP.
3826 * @qp: The QP handle to release
3827 *
3828 * The opened QP handle is released by the caller. The underlying
3829 * shared QP is not destroyed until all internal references are released.
3830 */
3831int ib_close_qp(struct ib_qp *qp);
3832
3833/**
3834 * ib_post_send - Posts a list of work requests to the send queue of
3835 * the specified QP.
3836 * @qp: The QP to post the work request on.
3837 * @send_wr: A list of work requests to post on the send queue.
3838 * @bad_send_wr: On an immediate failure, this parameter will reference
3839 * the work request that failed to be posted on the QP.
3840 *
3841 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3842 * error is returned, the QP state shall not be affected,
3843 * ib_post_send() will return an immediate error after queueing any
3844 * earlier work requests in the list.
3845 */
3846static inline int ib_post_send(struct ib_qp *qp,
3847 const struct ib_send_wr *send_wr,
3848 const struct ib_send_wr **bad_send_wr)
3849{
3850 const struct ib_send_wr *dummy;
3851
3852 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3853}
3854
3855/**
3856 * ib_post_recv - Posts a list of work requests to the receive queue of
3857 * the specified QP.
3858 * @qp: The QP to post the work request on.
3859 * @recv_wr: A list of work requests to post on the receive queue.
3860 * @bad_recv_wr: On an immediate failure, this parameter will reference
3861 * the work request that failed to be posted on the QP.
3862 */
3863static inline int ib_post_recv(struct ib_qp *qp,
3864 const struct ib_recv_wr *recv_wr,
3865 const struct ib_recv_wr **bad_recv_wr)
3866{
3867 const struct ib_recv_wr *dummy;
3868
3869 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3870}
3871
3872struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3873 int comp_vector, enum ib_poll_context poll_ctx,
3874 const char *caller);
3875static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3876 int nr_cqe, int comp_vector,
3877 enum ib_poll_context poll_ctx)
3878{
3879 return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3880 KBUILD_MODNAME);
3881}
3882
3883struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3884 int nr_cqe, enum ib_poll_context poll_ctx,
3885 const char *caller);
3886
3887/**
3888 * ib_alloc_cq_any: Allocate kernel CQ
3889 * @dev: The IB device
3890 * @private: Private data attached to the CQE
3891 * @nr_cqe: Number of CQEs in the CQ
3892 * @poll_ctx: Context used for polling the CQ
3893 */
3894static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3895 void *private, int nr_cqe,
3896 enum ib_poll_context poll_ctx)
3897{
3898 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3899 KBUILD_MODNAME);
3900}
3901
3902void ib_free_cq(struct ib_cq *cq);
3903int ib_process_cq_direct(struct ib_cq *cq, int budget);
3904
3905/**
3906 * ib_create_cq - Creates a CQ on the specified device.
3907 * @device: The device on which to create the CQ.
3908 * @comp_handler: A user-specified callback that is invoked when a
3909 * completion event occurs on the CQ.
3910 * @event_handler: A user-specified callback that is invoked when an
3911 * asynchronous event not associated with a completion occurs on the CQ.
3912 * @cq_context: Context associated with the CQ returned to the user via
3913 * the associated completion and event handlers.
3914 * @cq_attr: The attributes the CQ should be created upon.
3915 *
3916 * Users can examine the cq structure to determine the actual CQ size.
3917 */
3918struct ib_cq *__ib_create_cq(struct ib_device *device,
3919 ib_comp_handler comp_handler,
3920 void (*event_handler)(struct ib_event *, void *),
3921 void *cq_context,
3922 const struct ib_cq_init_attr *cq_attr,
3923 const char *caller);
3924#define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3925 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3926
3927/**
3928 * ib_resize_cq - Modifies the capacity of the CQ.
3929 * @cq: The CQ to resize.
3930 * @cqe: The minimum size of the CQ.
3931 *
3932 * Users can examine the cq structure to determine the actual CQ size.
3933 */
3934int ib_resize_cq(struct ib_cq *cq, int cqe);
3935
3936/**
3937 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3938 * @cq: The CQ to modify.
3939 * @cq_count: number of CQEs that will trigger an event
3940 * @cq_period: max period of time in usec before triggering an event
3941 *
3942 */
3943int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3944
3945/**
3946 * ib_destroy_cq_user - Destroys the specified CQ.
3947 * @cq: The CQ to destroy.
3948 * @udata: Valid user data or NULL for kernel objects
3949 */
3950int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3951
3952/**
3953 * ib_destroy_cq - Destroys the specified kernel CQ.
3954 * @cq: The CQ to destroy.
3955 *
3956 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3957 */
3958static inline void ib_destroy_cq(struct ib_cq *cq)
3959{
3960 int ret = ib_destroy_cq_user(cq, NULL);
3961
3962 WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
3963}
3964
3965/**
3966 * ib_poll_cq - poll a CQ for completion(s)
3967 * @cq:the CQ being polled
3968 * @num_entries:maximum number of completions to return
3969 * @wc:array of at least @num_entries &struct ib_wc where completions
3970 * will be returned
3971 *
3972 * Poll a CQ for (possibly multiple) completions. If the return value
3973 * is < 0, an error occurred. If the return value is >= 0, it is the
3974 * number of completions returned. If the return value is
3975 * non-negative and < num_entries, then the CQ was emptied.
3976 */
3977static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3978 struct ib_wc *wc)
3979{
3980 return cq->device->ops.poll_cq(cq, num_entries, wc);
3981}
3982
3983/**
3984 * ib_req_notify_cq - Request completion notification on a CQ.
3985 * @cq: The CQ to generate an event for.
3986 * @flags:
3987 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3988 * to request an event on the next solicited event or next work
3989 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3990 * may also be |ed in to request a hint about missed events, as
3991 * described below.
3992 *
3993 * Return Value:
3994 * < 0 means an error occurred while requesting notification
3995 * == 0 means notification was requested successfully, and if
3996 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3997 * were missed and it is safe to wait for another event. In
3998 * this case is it guaranteed that any work completions added
3999 * to the CQ since the last CQ poll will trigger a completion
4000 * notification event.
4001 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
4002 * in. It means that the consumer must poll the CQ again to
4003 * make sure it is empty to avoid missing an event because of a
4004 * race between requesting notification and an entry being
4005 * added to the CQ. This return value means it is possible
4006 * (but not guaranteed) that a work completion has been added
4007 * to the CQ since the last poll without triggering a
4008 * completion notification event.
4009 */
4010static inline int ib_req_notify_cq(struct ib_cq *cq,
4011 enum ib_cq_notify_flags flags)
4012{
4013 return cq->device->ops.req_notify_cq(cq, flags);
4014}
4015
4016struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
4017 int comp_vector_hint,
4018 enum ib_poll_context poll_ctx);
4019
4020void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
4021
4022/*
4023 * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
4024 * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
4025 * address into the dma address.
4026 */
4027static inline bool ib_uses_virt_dma(struct ib_device *dev)
4028{
4029 return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
4030}
4031
4032/*
4033 * Check if a IB device's underlying DMA mapping supports P2PDMA transfers.
4034 */
4035static inline bool ib_dma_pci_p2p_dma_supported(struct ib_device *dev)
4036{
4037 if (ib_uses_virt_dma(dev))
4038 return false;
4039
4040 return dma_pci_p2pdma_supported(dev->dma_device);
4041}
4042
4043/**
4044 * ib_virt_dma_to_ptr - Convert a dma_addr to a kernel pointer
4045 * @dma_addr: The DMA address
4046 *
4047 * Used by ib_uses_virt_dma() devices to get back to the kernel pointer after
4048 * going through the dma_addr marshalling.
4049 */
4050static inline void *ib_virt_dma_to_ptr(u64 dma_addr)
4051{
4052 /* virt_dma mode maps the kvs's directly into the dma addr */
4053 return (void *)(uintptr_t)dma_addr;
4054}
4055
4056/**
4057 * ib_virt_dma_to_page - Convert a dma_addr to a struct page
4058 * @dma_addr: The DMA address
4059 *
4060 * Used by ib_uses_virt_dma() device to get back to the struct page after going
4061 * through the dma_addr marshalling.
4062 */
4063static inline struct page *ib_virt_dma_to_page(u64 dma_addr)
4064{
4065 return virt_to_page(ib_virt_dma_to_ptr(dma_addr));
4066}
4067
4068/**
4069 * ib_dma_mapping_error - check a DMA addr for error
4070 * @dev: The device for which the dma_addr was created
4071 * @dma_addr: The DMA address to check
4072 */
4073static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
4074{
4075 if (ib_uses_virt_dma(dev))
4076 return 0;
4077 return dma_mapping_error(dev->dma_device, dma_addr);
4078}
4079
4080/**
4081 * ib_dma_map_single - Map a kernel virtual address to DMA address
4082 * @dev: The device for which the dma_addr is to be created
4083 * @cpu_addr: The kernel virtual address
4084 * @size: The size of the region in bytes
4085 * @direction: The direction of the DMA
4086 */
4087static inline u64 ib_dma_map_single(struct ib_device *dev,
4088 void *cpu_addr, size_t size,
4089 enum dma_data_direction direction)
4090{
4091 if (ib_uses_virt_dma(dev))
4092 return (uintptr_t)cpu_addr;
4093 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
4094}
4095
4096/**
4097 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4098 * @dev: The device for which the DMA address was created
4099 * @addr: The DMA address
4100 * @size: The size of the region in bytes
4101 * @direction: The direction of the DMA
4102 */
4103static inline void ib_dma_unmap_single(struct ib_device *dev,
4104 u64 addr, size_t size,
4105 enum dma_data_direction direction)
4106{
4107 if (!ib_uses_virt_dma(dev))
4108 dma_unmap_single(dev->dma_device, addr, size, direction);
4109}
4110
4111/**
4112 * ib_dma_map_page - Map a physical page to DMA address
4113 * @dev: The device for which the dma_addr is to be created
4114 * @page: The page to be mapped
4115 * @offset: The offset within the page
4116 * @size: The size of the region in bytes
4117 * @direction: The direction of the DMA
4118 */
4119static inline u64 ib_dma_map_page(struct ib_device *dev,
4120 struct page *page,
4121 unsigned long offset,
4122 size_t size,
4123 enum dma_data_direction direction)
4124{
4125 if (ib_uses_virt_dma(dev))
4126 return (uintptr_t)(page_address(page) + offset);
4127 return dma_map_page(dev->dma_device, page, offset, size, direction);
4128}
4129
4130/**
4131 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4132 * @dev: The device for which the DMA address was created
4133 * @addr: The DMA address
4134 * @size: The size of the region in bytes
4135 * @direction: The direction of the DMA
4136 */
4137static inline void ib_dma_unmap_page(struct ib_device *dev,
4138 u64 addr, size_t size,
4139 enum dma_data_direction direction)
4140{
4141 if (!ib_uses_virt_dma(dev))
4142 dma_unmap_page(dev->dma_device, addr, size, direction);
4143}
4144
4145int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4146static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4147 struct scatterlist *sg, int nents,
4148 enum dma_data_direction direction,
4149 unsigned long dma_attrs)
4150{
4151 if (ib_uses_virt_dma(dev))
4152 return ib_dma_virt_map_sg(dev, sg, nents);
4153 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4154 dma_attrs);
4155}
4156
4157static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4158 struct scatterlist *sg, int nents,
4159 enum dma_data_direction direction,
4160 unsigned long dma_attrs)
4161{
4162 if (!ib_uses_virt_dma(dev))
4163 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4164 dma_attrs);
4165}
4166
4167/**
4168 * ib_dma_map_sgtable_attrs - Map a scatter/gather table to DMA addresses
4169 * @dev: The device for which the DMA addresses are to be created
4170 * @sg: The sg_table object describing the buffer
4171 * @direction: The direction of the DMA
4172 * @attrs: Optional DMA attributes for the map operation
4173 */
4174static inline int ib_dma_map_sgtable_attrs(struct ib_device *dev,
4175 struct sg_table *sgt,
4176 enum dma_data_direction direction,
4177 unsigned long dma_attrs)
4178{
4179 int nents;
4180
4181 if (ib_uses_virt_dma(dev)) {
4182 nents = ib_dma_virt_map_sg(dev, sgt->sgl, sgt->orig_nents);
4183 if (!nents)
4184 return -EIO;
4185 sgt->nents = nents;
4186 return 0;
4187 }
4188 return dma_map_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4189}
4190
4191static inline void ib_dma_unmap_sgtable_attrs(struct ib_device *dev,
4192 struct sg_table *sgt,
4193 enum dma_data_direction direction,
4194 unsigned long dma_attrs)
4195{
4196 if (!ib_uses_virt_dma(dev))
4197 dma_unmap_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4198}
4199
4200/**
4201 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4202 * @dev: The device for which the DMA addresses are to be created
4203 * @sg: The array of scatter/gather entries
4204 * @nents: The number of scatter/gather entries
4205 * @direction: The direction of the DMA
4206 */
4207static inline int ib_dma_map_sg(struct ib_device *dev,
4208 struct scatterlist *sg, int nents,
4209 enum dma_data_direction direction)
4210{
4211 return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4212}
4213
4214/**
4215 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4216 * @dev: The device for which the DMA addresses were created
4217 * @sg: The array of scatter/gather entries
4218 * @nents: The number of scatter/gather entries
4219 * @direction: The direction of the DMA
4220 */
4221static inline void ib_dma_unmap_sg(struct ib_device *dev,
4222 struct scatterlist *sg, int nents,
4223 enum dma_data_direction direction)
4224{
4225 ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4226}
4227
4228/**
4229 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4230 * @dev: The device to query
4231 *
4232 * The returned value represents a size in bytes.
4233 */
4234static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4235{
4236 if (ib_uses_virt_dma(dev))
4237 return UINT_MAX;
4238 return dma_get_max_seg_size(dev->dma_device);
4239}
4240
4241/**
4242 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4243 * @dev: The device for which the DMA address was created
4244 * @addr: The DMA address
4245 * @size: The size of the region in bytes
4246 * @dir: The direction of the DMA
4247 */
4248static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4249 u64 addr,
4250 size_t size,
4251 enum dma_data_direction dir)
4252{
4253 if (!ib_uses_virt_dma(dev))
4254 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4255}
4256
4257/**
4258 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4259 * @dev: The device for which the DMA address was created
4260 * @addr: The DMA address
4261 * @size: The size of the region in bytes
4262 * @dir: The direction of the DMA
4263 */
4264static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4265 u64 addr,
4266 size_t size,
4267 enum dma_data_direction dir)
4268{
4269 if (!ib_uses_virt_dma(dev))
4270 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4271}
4272
4273/* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4274 * space. This function should be called when 'current' is the owning MM.
4275 */
4276struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4277 u64 virt_addr, int mr_access_flags);
4278
4279/* ib_advise_mr - give an advice about an address range in a memory region */
4280int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4281 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4282/**
4283 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4284 * HCA translation table.
4285 * @mr: The memory region to deregister.
4286 * @udata: Valid user data or NULL for kernel object
4287 *
4288 * This function can fail, if the memory region has memory windows bound to it.
4289 */
4290int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4291
4292/**
4293 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4294 * HCA translation table.
4295 * @mr: The memory region to deregister.
4296 *
4297 * This function can fail, if the memory region has memory windows bound to it.
4298 *
4299 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4300 */
4301static inline int ib_dereg_mr(struct ib_mr *mr)
4302{
4303 return ib_dereg_mr_user(mr, NULL);
4304}
4305
4306struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4307 u32 max_num_sg);
4308
4309struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4310 u32 max_num_data_sg,
4311 u32 max_num_meta_sg);
4312
4313/**
4314 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4315 * R_Key and L_Key.
4316 * @mr - struct ib_mr pointer to be updated.
4317 * @newkey - new key to be used.
4318 */
4319static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4320{
4321 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4322 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4323}
4324
4325/**
4326 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4327 * for calculating a new rkey for type 2 memory windows.
4328 * @rkey - the rkey to increment.
4329 */
4330static inline u32 ib_inc_rkey(u32 rkey)
4331{
4332 const u32 mask = 0x000000ff;
4333 return ((rkey + 1) & mask) | (rkey & ~mask);
4334}
4335
4336/**
4337 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4338 * @qp: QP to attach to the multicast group. The QP must be type
4339 * IB_QPT_UD.
4340 * @gid: Multicast group GID.
4341 * @lid: Multicast group LID in host byte order.
4342 *
4343 * In order to send and receive multicast packets, subnet
4344 * administration must have created the multicast group and configured
4345 * the fabric appropriately. The port associated with the specified
4346 * QP must also be a member of the multicast group.
4347 */
4348int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4349
4350/**
4351 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4352 * @qp: QP to detach from the multicast group.
4353 * @gid: Multicast group GID.
4354 * @lid: Multicast group LID in host byte order.
4355 */
4356int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4357
4358struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4359 struct inode *inode, struct ib_udata *udata);
4360int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4361
4362static inline int ib_check_mr_access(struct ib_device *ib_dev,
4363 unsigned int flags)
4364{
4365 u64 device_cap = ib_dev->attrs.device_cap_flags;
4366
4367 /*
4368 * Local write permission is required if remote write or
4369 * remote atomic permission is also requested.
4370 */
4371 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4372 !(flags & IB_ACCESS_LOCAL_WRITE))
4373 return -EINVAL;
4374
4375 if (flags & ~IB_ACCESS_SUPPORTED)
4376 return -EINVAL;
4377
4378 if (flags & IB_ACCESS_ON_DEMAND &&
4379 !(ib_dev->attrs.kernel_cap_flags & IBK_ON_DEMAND_PAGING))
4380 return -EOPNOTSUPP;
4381
4382 if ((flags & IB_ACCESS_FLUSH_GLOBAL &&
4383 !(device_cap & IB_DEVICE_FLUSH_GLOBAL)) ||
4384 (flags & IB_ACCESS_FLUSH_PERSISTENT &&
4385 !(device_cap & IB_DEVICE_FLUSH_PERSISTENT)))
4386 return -EOPNOTSUPP;
4387
4388 return 0;
4389}
4390
4391static inline bool ib_access_writable(int access_flags)
4392{
4393 /*
4394 * We have writable memory backing the MR if any of the following
4395 * access flags are set. "Local write" and "remote write" obviously
4396 * require write access. "Remote atomic" can do things like fetch and
4397 * add, which will modify memory, and "MW bind" can change permissions
4398 * by binding a window.
4399 */
4400 return access_flags &
4401 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4402 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4403}
4404
4405/**
4406 * ib_check_mr_status: lightweight check of MR status.
4407 * This routine may provide status checks on a selected
4408 * ib_mr. first use is for signature status check.
4409 *
4410 * @mr: A memory region.
4411 * @check_mask: Bitmask of which checks to perform from
4412 * ib_mr_status_check enumeration.
4413 * @mr_status: The container of relevant status checks.
4414 * failed checks will be indicated in the status bitmask
4415 * and the relevant info shall be in the error item.
4416 */
4417int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4418 struct ib_mr_status *mr_status);
4419
4420/**
4421 * ib_device_try_get: Hold a registration lock
4422 * device: The device to lock
4423 *
4424 * A device under an active registration lock cannot become unregistered. It
4425 * is only possible to obtain a registration lock on a device that is fully
4426 * registered, otherwise this function returns false.
4427 *
4428 * The registration lock is only necessary for actions which require the
4429 * device to still be registered. Uses that only require the device pointer to
4430 * be valid should use get_device(&ibdev->dev) to hold the memory.
4431 *
4432 */
4433static inline bool ib_device_try_get(struct ib_device *dev)
4434{
4435 return refcount_inc_not_zero(&dev->refcount);
4436}
4437
4438void ib_device_put(struct ib_device *device);
4439struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4440 enum rdma_driver_id driver_id);
4441struct ib_device *ib_device_get_by_name(const char *name,
4442 enum rdma_driver_id driver_id);
4443struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port,
4444 u16 pkey, const union ib_gid *gid,
4445 const struct sockaddr *addr);
4446int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4447 unsigned int port);
4448struct ib_wq *ib_create_wq(struct ib_pd *pd,
4449 struct ib_wq_init_attr *init_attr);
4450int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4451
4452int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4453 unsigned int *sg_offset, unsigned int page_size);
4454int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4455 int data_sg_nents, unsigned int *data_sg_offset,
4456 struct scatterlist *meta_sg, int meta_sg_nents,
4457 unsigned int *meta_sg_offset, unsigned int page_size);
4458
4459static inline int
4460ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4461 unsigned int *sg_offset, unsigned int page_size)
4462{
4463 int n;
4464
4465 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4466 mr->iova = 0;
4467
4468 return n;
4469}
4470
4471int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4472 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4473
4474void ib_drain_rq(struct ib_qp *qp);
4475void ib_drain_sq(struct ib_qp *qp);
4476void ib_drain_qp(struct ib_qp *qp);
4477
4478int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed,
4479 u8 *width);
4480
4481static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4482{
4483 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4484 return attr->roce.dmac;
4485 return NULL;
4486}
4487
4488static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4489{
4490 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4491 attr->ib.dlid = (u16)dlid;
4492 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4493 attr->opa.dlid = dlid;
4494}
4495
4496static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4497{
4498 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4499 return attr->ib.dlid;
4500 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4501 return attr->opa.dlid;
4502 return 0;
4503}
4504
4505static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4506{
4507 attr->sl = sl;
4508}
4509
4510static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4511{
4512 return attr->sl;
4513}
4514
4515static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4516 u8 src_path_bits)
4517{
4518 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4519 attr->ib.src_path_bits = src_path_bits;
4520 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4521 attr->opa.src_path_bits = src_path_bits;
4522}
4523
4524static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4525{
4526 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4527 return attr->ib.src_path_bits;
4528 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4529 return attr->opa.src_path_bits;
4530 return 0;
4531}
4532
4533static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4534 bool make_grd)
4535{
4536 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4537 attr->opa.make_grd = make_grd;
4538}
4539
4540static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4541{
4542 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4543 return attr->opa.make_grd;
4544 return false;
4545}
4546
4547static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num)
4548{
4549 attr->port_num = port_num;
4550}
4551
4552static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4553{
4554 return attr->port_num;
4555}
4556
4557static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4558 u8 static_rate)
4559{
4560 attr->static_rate = static_rate;
4561}
4562
4563static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4564{
4565 return attr->static_rate;
4566}
4567
4568static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4569 enum ib_ah_flags flag)
4570{
4571 attr->ah_flags = flag;
4572}
4573
4574static inline enum ib_ah_flags
4575 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4576{
4577 return attr->ah_flags;
4578}
4579
4580static inline const struct ib_global_route
4581 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4582{
4583 return &attr->grh;
4584}
4585
4586/*To retrieve and modify the grh */
4587static inline struct ib_global_route
4588 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4589{
4590 return &attr->grh;
4591}
4592
4593static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4594{
4595 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4596
4597 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4598}
4599
4600static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4601 __be64 prefix)
4602{
4603 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4604
4605 grh->dgid.global.subnet_prefix = prefix;
4606}
4607
4608static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4609 __be64 if_id)
4610{
4611 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4612
4613 grh->dgid.global.interface_id = if_id;
4614}
4615
4616static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4617 union ib_gid *dgid, u32 flow_label,
4618 u8 sgid_index, u8 hop_limit,
4619 u8 traffic_class)
4620{
4621 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4622
4623 attr->ah_flags = IB_AH_GRH;
4624 if (dgid)
4625 grh->dgid = *dgid;
4626 grh->flow_label = flow_label;
4627 grh->sgid_index = sgid_index;
4628 grh->hop_limit = hop_limit;
4629 grh->traffic_class = traffic_class;
4630 grh->sgid_attr = NULL;
4631}
4632
4633void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4634void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4635 u32 flow_label, u8 hop_limit, u8 traffic_class,
4636 const struct ib_gid_attr *sgid_attr);
4637void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4638 const struct rdma_ah_attr *src);
4639void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4640 const struct rdma_ah_attr *new);
4641void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4642
4643/**
4644 * rdma_ah_find_type - Return address handle type.
4645 *
4646 * @dev: Device to be checked
4647 * @port_num: Port number
4648 */
4649static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4650 u32 port_num)
4651{
4652 if (rdma_protocol_roce(dev, port_num))
4653 return RDMA_AH_ATTR_TYPE_ROCE;
4654 if (rdma_protocol_ib(dev, port_num)) {
4655 if (rdma_cap_opa_ah(dev, port_num))
4656 return RDMA_AH_ATTR_TYPE_OPA;
4657 return RDMA_AH_ATTR_TYPE_IB;
4658 }
4659
4660 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4661}
4662
4663/**
4664 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4665 * In the current implementation the only way to
4666 * get the 32bit lid is from other sources for OPA.
4667 * For IB, lids will always be 16bits so cast the
4668 * value accordingly.
4669 *
4670 * @lid: A 32bit LID
4671 */
4672static inline u16 ib_lid_cpu16(u32 lid)
4673{
4674 WARN_ON_ONCE(lid & 0xFFFF0000);
4675 return (u16)lid;
4676}
4677
4678/**
4679 * ib_lid_be16 - Return lid in 16bit BE encoding.
4680 *
4681 * @lid: A 32bit LID
4682 */
4683static inline __be16 ib_lid_be16(u32 lid)
4684{
4685 WARN_ON_ONCE(lid & 0xFFFF0000);
4686 return cpu_to_be16((u16)lid);
4687}
4688
4689/**
4690 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4691 * vector
4692 * @device: the rdma device
4693 * @comp_vector: index of completion vector
4694 *
4695 * Returns NULL on failure, otherwise a corresponding cpu map of the
4696 * completion vector (returns all-cpus map if the device driver doesn't
4697 * implement get_vector_affinity).
4698 */
4699static inline const struct cpumask *
4700ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4701{
4702 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4703 !device->ops.get_vector_affinity)
4704 return NULL;
4705
4706 return device->ops.get_vector_affinity(device, comp_vector);
4707
4708}
4709
4710/**
4711 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4712 * and add their gids, as needed, to the relevant RoCE devices.
4713 *
4714 * @device: the rdma device
4715 */
4716void rdma_roce_rescan_device(struct ib_device *ibdev);
4717
4718struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4719
4720int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4721
4722struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num,
4723 enum rdma_netdev_t type, const char *name,
4724 unsigned char name_assign_type,
4725 void (*setup)(struct net_device *));
4726
4727int rdma_init_netdev(struct ib_device *device, u32 port_num,
4728 enum rdma_netdev_t type, const char *name,
4729 unsigned char name_assign_type,
4730 void (*setup)(struct net_device *),
4731 struct net_device *netdev);
4732
4733/**
4734 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4735 *
4736 * @device: device pointer for which ib_device pointer to retrieve
4737 *
4738 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4739 *
4740 */
4741static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4742{
4743 struct ib_core_device *coredev =
4744 container_of(device, struct ib_core_device, dev);
4745
4746 return coredev->owner;
4747}
4748
4749/**
4750 * ibdev_to_node - return the NUMA node for a given ib_device
4751 * @dev: device to get the NUMA node for.
4752 */
4753static inline int ibdev_to_node(struct ib_device *ibdev)
4754{
4755 struct device *parent = ibdev->dev.parent;
4756
4757 if (!parent)
4758 return NUMA_NO_NODE;
4759 return dev_to_node(parent);
4760}
4761
4762/**
4763 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4764 * ib_device holder structure from device pointer.
4765 *
4766 * NOTE: New drivers should not make use of this API; This API is only for
4767 * existing drivers who have exposed sysfs entries using
4768 * ops->device_group.
4769 */
4770#define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4771 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4772
4773bool rdma_dev_access_netns(const struct ib_device *device,
4774 const struct net *net);
4775
4776#define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4777#define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4778#define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4779
4780/**
4781 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4782 * on the flow_label
4783 *
4784 * This function will convert the 20 bit flow_label input to a valid RoCE v2
4785 * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4786 * convention.
4787 */
4788static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4789{
4790 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4791
4792 fl_low ^= fl_high >> 14;
4793 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4794}
4795
4796/**
4797 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4798 * local and remote qpn values
4799 *
4800 * This function folded the multiplication results of two qpns, 24 bit each,
4801 * fields, and converts it to a 20 bit results.
4802 *
4803 * This function will create symmetric flow_label value based on the local
4804 * and remote qpn values. this will allow both the requester and responder
4805 * to calculate the same flow_label for a given connection.
4806 *
4807 * This helper function should be used by driver in case the upper layer
4808 * provide a zero flow_label value. This is to improve entropy of RDMA
4809 * traffic in the network.
4810 */
4811static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4812{
4813 u64 v = (u64)lqpn * rqpn;
4814
4815 v ^= v >> 20;
4816 v ^= v >> 40;
4817
4818 return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4819}
4820
4821/**
4822 * rdma_get_udp_sport - Calculate and set UDP source port based on the flow
4823 * label. If flow label is not defined in GRH then
4824 * calculate it based on lqpn/rqpn.
4825 *
4826 * @fl: flow label from GRH
4827 * @lqpn: local qp number
4828 * @rqpn: remote qp number
4829 */
4830static inline u16 rdma_get_udp_sport(u32 fl, u32 lqpn, u32 rqpn)
4831{
4832 if (!fl)
4833 fl = rdma_calc_flow_label(lqpn, rqpn);
4834
4835 return rdma_flow_label_to_udp_sport(fl);
4836}
4837
4838const struct ib_port_immutable*
4839ib_port_immutable_read(struct ib_device *dev, unsigned int port);
4840#endif /* IB_VERBS_H */
1/*
2 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
3 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
4 * Copyright (c) 2004 Intel Corporation. All rights reserved.
5 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
6 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
9 *
10 * This software is available to you under a choice of one of two
11 * licenses. You may choose to be licensed under the terms of the GNU
12 * General Public License (GPL) Version 2, available from the file
13 * COPYING in the main directory of this source tree, or the
14 * OpenIB.org BSD license below:
15 *
16 * Redistribution and use in source and binary forms, with or
17 * without modification, are permitted provided that the following
18 * conditions are met:
19 *
20 * - Redistributions of source code must retain the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer.
23 *
24 * - Redistributions in binary form must reproduce the above
25 * copyright notice, this list of conditions and the following
26 * disclaimer in the documentation and/or other materials
27 * provided with the distribution.
28 *
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36 * SOFTWARE.
37 */
38
39#if !defined(IB_VERBS_H)
40#define IB_VERBS_H
41
42#include <linux/types.h>
43#include <linux/device.h>
44#include <linux/mm.h>
45#include <linux/dma-mapping.h>
46#include <linux/kref.h>
47#include <linux/list.h>
48#include <linux/rwsem.h>
49#include <linux/scatterlist.h>
50#include <linux/workqueue.h>
51
52#include <linux/atomic.h>
53#include <asm/uaccess.h>
54
55extern struct workqueue_struct *ib_wq;
56
57union ib_gid {
58 u8 raw[16];
59 struct {
60 __be64 subnet_prefix;
61 __be64 interface_id;
62 } global;
63};
64
65enum rdma_node_type {
66 /* IB values map to NodeInfo:NodeType. */
67 RDMA_NODE_IB_CA = 1,
68 RDMA_NODE_IB_SWITCH,
69 RDMA_NODE_IB_ROUTER,
70 RDMA_NODE_RNIC
71};
72
73enum rdma_transport_type {
74 RDMA_TRANSPORT_IB,
75 RDMA_TRANSPORT_IWARP
76};
77
78enum rdma_transport_type
79rdma_node_get_transport(enum rdma_node_type node_type) __attribute_const__;
80
81enum rdma_link_layer {
82 IB_LINK_LAYER_UNSPECIFIED,
83 IB_LINK_LAYER_INFINIBAND,
84 IB_LINK_LAYER_ETHERNET,
85};
86
87enum ib_device_cap_flags {
88 IB_DEVICE_RESIZE_MAX_WR = 1,
89 IB_DEVICE_BAD_PKEY_CNTR = (1<<1),
90 IB_DEVICE_BAD_QKEY_CNTR = (1<<2),
91 IB_DEVICE_RAW_MULTI = (1<<3),
92 IB_DEVICE_AUTO_PATH_MIG = (1<<4),
93 IB_DEVICE_CHANGE_PHY_PORT = (1<<5),
94 IB_DEVICE_UD_AV_PORT_ENFORCE = (1<<6),
95 IB_DEVICE_CURR_QP_STATE_MOD = (1<<7),
96 IB_DEVICE_SHUTDOWN_PORT = (1<<8),
97 IB_DEVICE_INIT_TYPE = (1<<9),
98 IB_DEVICE_PORT_ACTIVE_EVENT = (1<<10),
99 IB_DEVICE_SYS_IMAGE_GUID = (1<<11),
100 IB_DEVICE_RC_RNR_NAK_GEN = (1<<12),
101 IB_DEVICE_SRQ_RESIZE = (1<<13),
102 IB_DEVICE_N_NOTIFY_CQ = (1<<14),
103 IB_DEVICE_LOCAL_DMA_LKEY = (1<<15),
104 IB_DEVICE_RESERVED = (1<<16), /* old SEND_W_INV */
105 IB_DEVICE_MEM_WINDOW = (1<<17),
106 /*
107 * Devices should set IB_DEVICE_UD_IP_SUM if they support
108 * insertion of UDP and TCP checksum on outgoing UD IPoIB
109 * messages and can verify the validity of checksum for
110 * incoming messages. Setting this flag implies that the
111 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
112 */
113 IB_DEVICE_UD_IP_CSUM = (1<<18),
114 IB_DEVICE_UD_TSO = (1<<19),
115 IB_DEVICE_XRC = (1<<20),
116 IB_DEVICE_MEM_MGT_EXTENSIONS = (1<<21),
117 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22),
118};
119
120enum ib_atomic_cap {
121 IB_ATOMIC_NONE,
122 IB_ATOMIC_HCA,
123 IB_ATOMIC_GLOB
124};
125
126struct ib_device_attr {
127 u64 fw_ver;
128 __be64 sys_image_guid;
129 u64 max_mr_size;
130 u64 page_size_cap;
131 u32 vendor_id;
132 u32 vendor_part_id;
133 u32 hw_ver;
134 int max_qp;
135 int max_qp_wr;
136 int device_cap_flags;
137 int max_sge;
138 int max_sge_rd;
139 int max_cq;
140 int max_cqe;
141 int max_mr;
142 int max_pd;
143 int max_qp_rd_atom;
144 int max_ee_rd_atom;
145 int max_res_rd_atom;
146 int max_qp_init_rd_atom;
147 int max_ee_init_rd_atom;
148 enum ib_atomic_cap atomic_cap;
149 enum ib_atomic_cap masked_atomic_cap;
150 int max_ee;
151 int max_rdd;
152 int max_mw;
153 int max_raw_ipv6_qp;
154 int max_raw_ethy_qp;
155 int max_mcast_grp;
156 int max_mcast_qp_attach;
157 int max_total_mcast_qp_attach;
158 int max_ah;
159 int max_fmr;
160 int max_map_per_fmr;
161 int max_srq;
162 int max_srq_wr;
163 int max_srq_sge;
164 unsigned int max_fast_reg_page_list_len;
165 u16 max_pkeys;
166 u8 local_ca_ack_delay;
167};
168
169enum ib_mtu {
170 IB_MTU_256 = 1,
171 IB_MTU_512 = 2,
172 IB_MTU_1024 = 3,
173 IB_MTU_2048 = 4,
174 IB_MTU_4096 = 5
175};
176
177static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
178{
179 switch (mtu) {
180 case IB_MTU_256: return 256;
181 case IB_MTU_512: return 512;
182 case IB_MTU_1024: return 1024;
183 case IB_MTU_2048: return 2048;
184 case IB_MTU_4096: return 4096;
185 default: return -1;
186 }
187}
188
189enum ib_port_state {
190 IB_PORT_NOP = 0,
191 IB_PORT_DOWN = 1,
192 IB_PORT_INIT = 2,
193 IB_PORT_ARMED = 3,
194 IB_PORT_ACTIVE = 4,
195 IB_PORT_ACTIVE_DEFER = 5
196};
197
198enum ib_port_cap_flags {
199 IB_PORT_SM = 1 << 1,
200 IB_PORT_NOTICE_SUP = 1 << 2,
201 IB_PORT_TRAP_SUP = 1 << 3,
202 IB_PORT_OPT_IPD_SUP = 1 << 4,
203 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
204 IB_PORT_SL_MAP_SUP = 1 << 6,
205 IB_PORT_MKEY_NVRAM = 1 << 7,
206 IB_PORT_PKEY_NVRAM = 1 << 8,
207 IB_PORT_LED_INFO_SUP = 1 << 9,
208 IB_PORT_SM_DISABLED = 1 << 10,
209 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
210 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
211 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
212 IB_PORT_CM_SUP = 1 << 16,
213 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
214 IB_PORT_REINIT_SUP = 1 << 18,
215 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
216 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
217 IB_PORT_DR_NOTICE_SUP = 1 << 21,
218 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
219 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
220 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
221 IB_PORT_CLIENT_REG_SUP = 1 << 25
222};
223
224enum ib_port_width {
225 IB_WIDTH_1X = 1,
226 IB_WIDTH_4X = 2,
227 IB_WIDTH_8X = 4,
228 IB_WIDTH_12X = 8
229};
230
231static inline int ib_width_enum_to_int(enum ib_port_width width)
232{
233 switch (width) {
234 case IB_WIDTH_1X: return 1;
235 case IB_WIDTH_4X: return 4;
236 case IB_WIDTH_8X: return 8;
237 case IB_WIDTH_12X: return 12;
238 default: return -1;
239 }
240}
241
242enum ib_port_speed {
243 IB_SPEED_SDR = 1,
244 IB_SPEED_DDR = 2,
245 IB_SPEED_QDR = 4,
246 IB_SPEED_FDR10 = 8,
247 IB_SPEED_FDR = 16,
248 IB_SPEED_EDR = 32
249};
250
251struct ib_protocol_stats {
252 /* TBD... */
253};
254
255struct iw_protocol_stats {
256 u64 ipInReceives;
257 u64 ipInHdrErrors;
258 u64 ipInTooBigErrors;
259 u64 ipInNoRoutes;
260 u64 ipInAddrErrors;
261 u64 ipInUnknownProtos;
262 u64 ipInTruncatedPkts;
263 u64 ipInDiscards;
264 u64 ipInDelivers;
265 u64 ipOutForwDatagrams;
266 u64 ipOutRequests;
267 u64 ipOutDiscards;
268 u64 ipOutNoRoutes;
269 u64 ipReasmTimeout;
270 u64 ipReasmReqds;
271 u64 ipReasmOKs;
272 u64 ipReasmFails;
273 u64 ipFragOKs;
274 u64 ipFragFails;
275 u64 ipFragCreates;
276 u64 ipInMcastPkts;
277 u64 ipOutMcastPkts;
278 u64 ipInBcastPkts;
279 u64 ipOutBcastPkts;
280
281 u64 tcpRtoAlgorithm;
282 u64 tcpRtoMin;
283 u64 tcpRtoMax;
284 u64 tcpMaxConn;
285 u64 tcpActiveOpens;
286 u64 tcpPassiveOpens;
287 u64 tcpAttemptFails;
288 u64 tcpEstabResets;
289 u64 tcpCurrEstab;
290 u64 tcpInSegs;
291 u64 tcpOutSegs;
292 u64 tcpRetransSegs;
293 u64 tcpInErrs;
294 u64 tcpOutRsts;
295};
296
297union rdma_protocol_stats {
298 struct ib_protocol_stats ib;
299 struct iw_protocol_stats iw;
300};
301
302struct ib_port_attr {
303 enum ib_port_state state;
304 enum ib_mtu max_mtu;
305 enum ib_mtu active_mtu;
306 int gid_tbl_len;
307 u32 port_cap_flags;
308 u32 max_msg_sz;
309 u32 bad_pkey_cntr;
310 u32 qkey_viol_cntr;
311 u16 pkey_tbl_len;
312 u16 lid;
313 u16 sm_lid;
314 u8 lmc;
315 u8 max_vl_num;
316 u8 sm_sl;
317 u8 subnet_timeout;
318 u8 init_type_reply;
319 u8 active_width;
320 u8 active_speed;
321 u8 phys_state;
322};
323
324enum ib_device_modify_flags {
325 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
326 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
327};
328
329struct ib_device_modify {
330 u64 sys_image_guid;
331 char node_desc[64];
332};
333
334enum ib_port_modify_flags {
335 IB_PORT_SHUTDOWN = 1,
336 IB_PORT_INIT_TYPE = (1<<2),
337 IB_PORT_RESET_QKEY_CNTR = (1<<3)
338};
339
340struct ib_port_modify {
341 u32 set_port_cap_mask;
342 u32 clr_port_cap_mask;
343 u8 init_type;
344};
345
346enum ib_event_type {
347 IB_EVENT_CQ_ERR,
348 IB_EVENT_QP_FATAL,
349 IB_EVENT_QP_REQ_ERR,
350 IB_EVENT_QP_ACCESS_ERR,
351 IB_EVENT_COMM_EST,
352 IB_EVENT_SQ_DRAINED,
353 IB_EVENT_PATH_MIG,
354 IB_EVENT_PATH_MIG_ERR,
355 IB_EVENT_DEVICE_FATAL,
356 IB_EVENT_PORT_ACTIVE,
357 IB_EVENT_PORT_ERR,
358 IB_EVENT_LID_CHANGE,
359 IB_EVENT_PKEY_CHANGE,
360 IB_EVENT_SM_CHANGE,
361 IB_EVENT_SRQ_ERR,
362 IB_EVENT_SRQ_LIMIT_REACHED,
363 IB_EVENT_QP_LAST_WQE_REACHED,
364 IB_EVENT_CLIENT_REREGISTER,
365 IB_EVENT_GID_CHANGE,
366};
367
368struct ib_event {
369 struct ib_device *device;
370 union {
371 struct ib_cq *cq;
372 struct ib_qp *qp;
373 struct ib_srq *srq;
374 u8 port_num;
375 } element;
376 enum ib_event_type event;
377};
378
379struct ib_event_handler {
380 struct ib_device *device;
381 void (*handler)(struct ib_event_handler *, struct ib_event *);
382 struct list_head list;
383};
384
385#define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
386 do { \
387 (_ptr)->device = _device; \
388 (_ptr)->handler = _handler; \
389 INIT_LIST_HEAD(&(_ptr)->list); \
390 } while (0)
391
392struct ib_global_route {
393 union ib_gid dgid;
394 u32 flow_label;
395 u8 sgid_index;
396 u8 hop_limit;
397 u8 traffic_class;
398};
399
400struct ib_grh {
401 __be32 version_tclass_flow;
402 __be16 paylen;
403 u8 next_hdr;
404 u8 hop_limit;
405 union ib_gid sgid;
406 union ib_gid dgid;
407};
408
409enum {
410 IB_MULTICAST_QPN = 0xffffff
411};
412
413#define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
414
415enum ib_ah_flags {
416 IB_AH_GRH = 1
417};
418
419enum ib_rate {
420 IB_RATE_PORT_CURRENT = 0,
421 IB_RATE_2_5_GBPS = 2,
422 IB_RATE_5_GBPS = 5,
423 IB_RATE_10_GBPS = 3,
424 IB_RATE_20_GBPS = 6,
425 IB_RATE_30_GBPS = 4,
426 IB_RATE_40_GBPS = 7,
427 IB_RATE_60_GBPS = 8,
428 IB_RATE_80_GBPS = 9,
429 IB_RATE_120_GBPS = 10,
430 IB_RATE_14_GBPS = 11,
431 IB_RATE_56_GBPS = 12,
432 IB_RATE_112_GBPS = 13,
433 IB_RATE_168_GBPS = 14,
434 IB_RATE_25_GBPS = 15,
435 IB_RATE_100_GBPS = 16,
436 IB_RATE_200_GBPS = 17,
437 IB_RATE_300_GBPS = 18
438};
439
440/**
441 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
442 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
443 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
444 * @rate: rate to convert.
445 */
446int ib_rate_to_mult(enum ib_rate rate) __attribute_const__;
447
448/**
449 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
450 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
451 * @rate: rate to convert.
452 */
453int ib_rate_to_mbps(enum ib_rate rate) __attribute_const__;
454
455/**
456 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
457 * enum.
458 * @mult: multiple to convert.
459 */
460enum ib_rate mult_to_ib_rate(int mult) __attribute_const__;
461
462struct ib_ah_attr {
463 struct ib_global_route grh;
464 u16 dlid;
465 u8 sl;
466 u8 src_path_bits;
467 u8 static_rate;
468 u8 ah_flags;
469 u8 port_num;
470};
471
472enum ib_wc_status {
473 IB_WC_SUCCESS,
474 IB_WC_LOC_LEN_ERR,
475 IB_WC_LOC_QP_OP_ERR,
476 IB_WC_LOC_EEC_OP_ERR,
477 IB_WC_LOC_PROT_ERR,
478 IB_WC_WR_FLUSH_ERR,
479 IB_WC_MW_BIND_ERR,
480 IB_WC_BAD_RESP_ERR,
481 IB_WC_LOC_ACCESS_ERR,
482 IB_WC_REM_INV_REQ_ERR,
483 IB_WC_REM_ACCESS_ERR,
484 IB_WC_REM_OP_ERR,
485 IB_WC_RETRY_EXC_ERR,
486 IB_WC_RNR_RETRY_EXC_ERR,
487 IB_WC_LOC_RDD_VIOL_ERR,
488 IB_WC_REM_INV_RD_REQ_ERR,
489 IB_WC_REM_ABORT_ERR,
490 IB_WC_INV_EECN_ERR,
491 IB_WC_INV_EEC_STATE_ERR,
492 IB_WC_FATAL_ERR,
493 IB_WC_RESP_TIMEOUT_ERR,
494 IB_WC_GENERAL_ERR
495};
496
497enum ib_wc_opcode {
498 IB_WC_SEND,
499 IB_WC_RDMA_WRITE,
500 IB_WC_RDMA_READ,
501 IB_WC_COMP_SWAP,
502 IB_WC_FETCH_ADD,
503 IB_WC_BIND_MW,
504 IB_WC_LSO,
505 IB_WC_LOCAL_INV,
506 IB_WC_FAST_REG_MR,
507 IB_WC_MASKED_COMP_SWAP,
508 IB_WC_MASKED_FETCH_ADD,
509/*
510 * Set value of IB_WC_RECV so consumers can test if a completion is a
511 * receive by testing (opcode & IB_WC_RECV).
512 */
513 IB_WC_RECV = 1 << 7,
514 IB_WC_RECV_RDMA_WITH_IMM
515};
516
517enum ib_wc_flags {
518 IB_WC_GRH = 1,
519 IB_WC_WITH_IMM = (1<<1),
520 IB_WC_WITH_INVALIDATE = (1<<2),
521 IB_WC_IP_CSUM_OK = (1<<3),
522};
523
524struct ib_wc {
525 u64 wr_id;
526 enum ib_wc_status status;
527 enum ib_wc_opcode opcode;
528 u32 vendor_err;
529 u32 byte_len;
530 struct ib_qp *qp;
531 union {
532 __be32 imm_data;
533 u32 invalidate_rkey;
534 } ex;
535 u32 src_qp;
536 int wc_flags;
537 u16 pkey_index;
538 u16 slid;
539 u8 sl;
540 u8 dlid_path_bits;
541 u8 port_num; /* valid only for DR SMPs on switches */
542};
543
544enum ib_cq_notify_flags {
545 IB_CQ_SOLICITED = 1 << 0,
546 IB_CQ_NEXT_COMP = 1 << 1,
547 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
548 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
549};
550
551enum ib_srq_type {
552 IB_SRQT_BASIC,
553 IB_SRQT_XRC
554};
555
556enum ib_srq_attr_mask {
557 IB_SRQ_MAX_WR = 1 << 0,
558 IB_SRQ_LIMIT = 1 << 1,
559};
560
561struct ib_srq_attr {
562 u32 max_wr;
563 u32 max_sge;
564 u32 srq_limit;
565};
566
567struct ib_srq_init_attr {
568 void (*event_handler)(struct ib_event *, void *);
569 void *srq_context;
570 struct ib_srq_attr attr;
571 enum ib_srq_type srq_type;
572
573 union {
574 struct {
575 struct ib_xrcd *xrcd;
576 struct ib_cq *cq;
577 } xrc;
578 } ext;
579};
580
581struct ib_qp_cap {
582 u32 max_send_wr;
583 u32 max_recv_wr;
584 u32 max_send_sge;
585 u32 max_recv_sge;
586 u32 max_inline_data;
587};
588
589enum ib_sig_type {
590 IB_SIGNAL_ALL_WR,
591 IB_SIGNAL_REQ_WR
592};
593
594enum ib_qp_type {
595 /*
596 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
597 * here (and in that order) since the MAD layer uses them as
598 * indices into a 2-entry table.
599 */
600 IB_QPT_SMI,
601 IB_QPT_GSI,
602
603 IB_QPT_RC,
604 IB_QPT_UC,
605 IB_QPT_UD,
606 IB_QPT_RAW_IPV6,
607 IB_QPT_RAW_ETHERTYPE,
608 IB_QPT_RAW_PACKET = 8,
609 IB_QPT_XRC_INI = 9,
610 IB_QPT_XRC_TGT,
611 IB_QPT_MAX
612};
613
614enum ib_qp_create_flags {
615 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
616 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
617};
618
619struct ib_qp_init_attr {
620 void (*event_handler)(struct ib_event *, void *);
621 void *qp_context;
622 struct ib_cq *send_cq;
623 struct ib_cq *recv_cq;
624 struct ib_srq *srq;
625 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
626 struct ib_qp_cap cap;
627 enum ib_sig_type sq_sig_type;
628 enum ib_qp_type qp_type;
629 enum ib_qp_create_flags create_flags;
630 u8 port_num; /* special QP types only */
631};
632
633struct ib_qp_open_attr {
634 void (*event_handler)(struct ib_event *, void *);
635 void *qp_context;
636 u32 qp_num;
637 enum ib_qp_type qp_type;
638};
639
640enum ib_rnr_timeout {
641 IB_RNR_TIMER_655_36 = 0,
642 IB_RNR_TIMER_000_01 = 1,
643 IB_RNR_TIMER_000_02 = 2,
644 IB_RNR_TIMER_000_03 = 3,
645 IB_RNR_TIMER_000_04 = 4,
646 IB_RNR_TIMER_000_06 = 5,
647 IB_RNR_TIMER_000_08 = 6,
648 IB_RNR_TIMER_000_12 = 7,
649 IB_RNR_TIMER_000_16 = 8,
650 IB_RNR_TIMER_000_24 = 9,
651 IB_RNR_TIMER_000_32 = 10,
652 IB_RNR_TIMER_000_48 = 11,
653 IB_RNR_TIMER_000_64 = 12,
654 IB_RNR_TIMER_000_96 = 13,
655 IB_RNR_TIMER_001_28 = 14,
656 IB_RNR_TIMER_001_92 = 15,
657 IB_RNR_TIMER_002_56 = 16,
658 IB_RNR_TIMER_003_84 = 17,
659 IB_RNR_TIMER_005_12 = 18,
660 IB_RNR_TIMER_007_68 = 19,
661 IB_RNR_TIMER_010_24 = 20,
662 IB_RNR_TIMER_015_36 = 21,
663 IB_RNR_TIMER_020_48 = 22,
664 IB_RNR_TIMER_030_72 = 23,
665 IB_RNR_TIMER_040_96 = 24,
666 IB_RNR_TIMER_061_44 = 25,
667 IB_RNR_TIMER_081_92 = 26,
668 IB_RNR_TIMER_122_88 = 27,
669 IB_RNR_TIMER_163_84 = 28,
670 IB_RNR_TIMER_245_76 = 29,
671 IB_RNR_TIMER_327_68 = 30,
672 IB_RNR_TIMER_491_52 = 31
673};
674
675enum ib_qp_attr_mask {
676 IB_QP_STATE = 1,
677 IB_QP_CUR_STATE = (1<<1),
678 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
679 IB_QP_ACCESS_FLAGS = (1<<3),
680 IB_QP_PKEY_INDEX = (1<<4),
681 IB_QP_PORT = (1<<5),
682 IB_QP_QKEY = (1<<6),
683 IB_QP_AV = (1<<7),
684 IB_QP_PATH_MTU = (1<<8),
685 IB_QP_TIMEOUT = (1<<9),
686 IB_QP_RETRY_CNT = (1<<10),
687 IB_QP_RNR_RETRY = (1<<11),
688 IB_QP_RQ_PSN = (1<<12),
689 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
690 IB_QP_ALT_PATH = (1<<14),
691 IB_QP_MIN_RNR_TIMER = (1<<15),
692 IB_QP_SQ_PSN = (1<<16),
693 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
694 IB_QP_PATH_MIG_STATE = (1<<18),
695 IB_QP_CAP = (1<<19),
696 IB_QP_DEST_QPN = (1<<20)
697};
698
699enum ib_qp_state {
700 IB_QPS_RESET,
701 IB_QPS_INIT,
702 IB_QPS_RTR,
703 IB_QPS_RTS,
704 IB_QPS_SQD,
705 IB_QPS_SQE,
706 IB_QPS_ERR
707};
708
709enum ib_mig_state {
710 IB_MIG_MIGRATED,
711 IB_MIG_REARM,
712 IB_MIG_ARMED
713};
714
715struct ib_qp_attr {
716 enum ib_qp_state qp_state;
717 enum ib_qp_state cur_qp_state;
718 enum ib_mtu path_mtu;
719 enum ib_mig_state path_mig_state;
720 u32 qkey;
721 u32 rq_psn;
722 u32 sq_psn;
723 u32 dest_qp_num;
724 int qp_access_flags;
725 struct ib_qp_cap cap;
726 struct ib_ah_attr ah_attr;
727 struct ib_ah_attr alt_ah_attr;
728 u16 pkey_index;
729 u16 alt_pkey_index;
730 u8 en_sqd_async_notify;
731 u8 sq_draining;
732 u8 max_rd_atomic;
733 u8 max_dest_rd_atomic;
734 u8 min_rnr_timer;
735 u8 port_num;
736 u8 timeout;
737 u8 retry_cnt;
738 u8 rnr_retry;
739 u8 alt_port_num;
740 u8 alt_timeout;
741};
742
743enum ib_wr_opcode {
744 IB_WR_RDMA_WRITE,
745 IB_WR_RDMA_WRITE_WITH_IMM,
746 IB_WR_SEND,
747 IB_WR_SEND_WITH_IMM,
748 IB_WR_RDMA_READ,
749 IB_WR_ATOMIC_CMP_AND_SWP,
750 IB_WR_ATOMIC_FETCH_AND_ADD,
751 IB_WR_LSO,
752 IB_WR_SEND_WITH_INV,
753 IB_WR_RDMA_READ_WITH_INV,
754 IB_WR_LOCAL_INV,
755 IB_WR_FAST_REG_MR,
756 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
757 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
758};
759
760enum ib_send_flags {
761 IB_SEND_FENCE = 1,
762 IB_SEND_SIGNALED = (1<<1),
763 IB_SEND_SOLICITED = (1<<2),
764 IB_SEND_INLINE = (1<<3),
765 IB_SEND_IP_CSUM = (1<<4)
766};
767
768struct ib_sge {
769 u64 addr;
770 u32 length;
771 u32 lkey;
772};
773
774struct ib_fast_reg_page_list {
775 struct ib_device *device;
776 u64 *page_list;
777 unsigned int max_page_list_len;
778};
779
780struct ib_send_wr {
781 struct ib_send_wr *next;
782 u64 wr_id;
783 struct ib_sge *sg_list;
784 int num_sge;
785 enum ib_wr_opcode opcode;
786 int send_flags;
787 union {
788 __be32 imm_data;
789 u32 invalidate_rkey;
790 } ex;
791 union {
792 struct {
793 u64 remote_addr;
794 u32 rkey;
795 } rdma;
796 struct {
797 u64 remote_addr;
798 u64 compare_add;
799 u64 swap;
800 u64 compare_add_mask;
801 u64 swap_mask;
802 u32 rkey;
803 } atomic;
804 struct {
805 struct ib_ah *ah;
806 void *header;
807 int hlen;
808 int mss;
809 u32 remote_qpn;
810 u32 remote_qkey;
811 u16 pkey_index; /* valid for GSI only */
812 u8 port_num; /* valid for DR SMPs on switch only */
813 } ud;
814 struct {
815 u64 iova_start;
816 struct ib_fast_reg_page_list *page_list;
817 unsigned int page_shift;
818 unsigned int page_list_len;
819 u32 length;
820 int access_flags;
821 u32 rkey;
822 } fast_reg;
823 } wr;
824 u32 xrc_remote_srq_num; /* XRC TGT QPs only */
825};
826
827struct ib_recv_wr {
828 struct ib_recv_wr *next;
829 u64 wr_id;
830 struct ib_sge *sg_list;
831 int num_sge;
832};
833
834enum ib_access_flags {
835 IB_ACCESS_LOCAL_WRITE = 1,
836 IB_ACCESS_REMOTE_WRITE = (1<<1),
837 IB_ACCESS_REMOTE_READ = (1<<2),
838 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
839 IB_ACCESS_MW_BIND = (1<<4)
840};
841
842struct ib_phys_buf {
843 u64 addr;
844 u64 size;
845};
846
847struct ib_mr_attr {
848 struct ib_pd *pd;
849 u64 device_virt_addr;
850 u64 size;
851 int mr_access_flags;
852 u32 lkey;
853 u32 rkey;
854};
855
856enum ib_mr_rereg_flags {
857 IB_MR_REREG_TRANS = 1,
858 IB_MR_REREG_PD = (1<<1),
859 IB_MR_REREG_ACCESS = (1<<2)
860};
861
862struct ib_mw_bind {
863 struct ib_mr *mr;
864 u64 wr_id;
865 u64 addr;
866 u32 length;
867 int send_flags;
868 int mw_access_flags;
869};
870
871struct ib_fmr_attr {
872 int max_pages;
873 int max_maps;
874 u8 page_shift;
875};
876
877struct ib_ucontext {
878 struct ib_device *device;
879 struct list_head pd_list;
880 struct list_head mr_list;
881 struct list_head mw_list;
882 struct list_head cq_list;
883 struct list_head qp_list;
884 struct list_head srq_list;
885 struct list_head ah_list;
886 struct list_head xrcd_list;
887 int closing;
888};
889
890struct ib_uobject {
891 u64 user_handle; /* handle given to us by userspace */
892 struct ib_ucontext *context; /* associated user context */
893 void *object; /* containing object */
894 struct list_head list; /* link to context's list */
895 int id; /* index into kernel idr */
896 struct kref ref;
897 struct rw_semaphore mutex; /* protects .live */
898 int live;
899};
900
901struct ib_udata {
902 void __user *inbuf;
903 void __user *outbuf;
904 size_t inlen;
905 size_t outlen;
906};
907
908struct ib_pd {
909 struct ib_device *device;
910 struct ib_uobject *uobject;
911 atomic_t usecnt; /* count all resources */
912};
913
914struct ib_xrcd {
915 struct ib_device *device;
916 atomic_t usecnt; /* count all exposed resources */
917 struct inode *inode;
918
919 struct mutex tgt_qp_mutex;
920 struct list_head tgt_qp_list;
921};
922
923struct ib_ah {
924 struct ib_device *device;
925 struct ib_pd *pd;
926 struct ib_uobject *uobject;
927};
928
929typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
930
931struct ib_cq {
932 struct ib_device *device;
933 struct ib_uobject *uobject;
934 ib_comp_handler comp_handler;
935 void (*event_handler)(struct ib_event *, void *);
936 void *cq_context;
937 int cqe;
938 atomic_t usecnt; /* count number of work queues */
939};
940
941struct ib_srq {
942 struct ib_device *device;
943 struct ib_pd *pd;
944 struct ib_uobject *uobject;
945 void (*event_handler)(struct ib_event *, void *);
946 void *srq_context;
947 enum ib_srq_type srq_type;
948 atomic_t usecnt;
949
950 union {
951 struct {
952 struct ib_xrcd *xrcd;
953 struct ib_cq *cq;
954 u32 srq_num;
955 } xrc;
956 } ext;
957};
958
959struct ib_qp {
960 struct ib_device *device;
961 struct ib_pd *pd;
962 struct ib_cq *send_cq;
963 struct ib_cq *recv_cq;
964 struct ib_srq *srq;
965 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
966 struct list_head xrcd_list;
967 atomic_t usecnt; /* count times opened, mcast attaches */
968 struct list_head open_list;
969 struct ib_qp *real_qp;
970 struct ib_uobject *uobject;
971 void (*event_handler)(struct ib_event *, void *);
972 void *qp_context;
973 u32 qp_num;
974 enum ib_qp_type qp_type;
975};
976
977struct ib_mr {
978 struct ib_device *device;
979 struct ib_pd *pd;
980 struct ib_uobject *uobject;
981 u32 lkey;
982 u32 rkey;
983 atomic_t usecnt; /* count number of MWs */
984};
985
986struct ib_mw {
987 struct ib_device *device;
988 struct ib_pd *pd;
989 struct ib_uobject *uobject;
990 u32 rkey;
991};
992
993struct ib_fmr {
994 struct ib_device *device;
995 struct ib_pd *pd;
996 struct list_head list;
997 u32 lkey;
998 u32 rkey;
999};
1000
1001struct ib_mad;
1002struct ib_grh;
1003
1004enum ib_process_mad_flags {
1005 IB_MAD_IGNORE_MKEY = 1,
1006 IB_MAD_IGNORE_BKEY = 2,
1007 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1008};
1009
1010enum ib_mad_result {
1011 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1012 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1013 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1014 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1015};
1016
1017#define IB_DEVICE_NAME_MAX 64
1018
1019struct ib_cache {
1020 rwlock_t lock;
1021 struct ib_event_handler event_handler;
1022 struct ib_pkey_cache **pkey_cache;
1023 struct ib_gid_cache **gid_cache;
1024 u8 *lmc_cache;
1025};
1026
1027struct ib_dma_mapping_ops {
1028 int (*mapping_error)(struct ib_device *dev,
1029 u64 dma_addr);
1030 u64 (*map_single)(struct ib_device *dev,
1031 void *ptr, size_t size,
1032 enum dma_data_direction direction);
1033 void (*unmap_single)(struct ib_device *dev,
1034 u64 addr, size_t size,
1035 enum dma_data_direction direction);
1036 u64 (*map_page)(struct ib_device *dev,
1037 struct page *page, unsigned long offset,
1038 size_t size,
1039 enum dma_data_direction direction);
1040 void (*unmap_page)(struct ib_device *dev,
1041 u64 addr, size_t size,
1042 enum dma_data_direction direction);
1043 int (*map_sg)(struct ib_device *dev,
1044 struct scatterlist *sg, int nents,
1045 enum dma_data_direction direction);
1046 void (*unmap_sg)(struct ib_device *dev,
1047 struct scatterlist *sg, int nents,
1048 enum dma_data_direction direction);
1049 u64 (*dma_address)(struct ib_device *dev,
1050 struct scatterlist *sg);
1051 unsigned int (*dma_len)(struct ib_device *dev,
1052 struct scatterlist *sg);
1053 void (*sync_single_for_cpu)(struct ib_device *dev,
1054 u64 dma_handle,
1055 size_t size,
1056 enum dma_data_direction dir);
1057 void (*sync_single_for_device)(struct ib_device *dev,
1058 u64 dma_handle,
1059 size_t size,
1060 enum dma_data_direction dir);
1061 void *(*alloc_coherent)(struct ib_device *dev,
1062 size_t size,
1063 u64 *dma_handle,
1064 gfp_t flag);
1065 void (*free_coherent)(struct ib_device *dev,
1066 size_t size, void *cpu_addr,
1067 u64 dma_handle);
1068};
1069
1070struct iw_cm_verbs;
1071
1072struct ib_device {
1073 struct device *dma_device;
1074
1075 char name[IB_DEVICE_NAME_MAX];
1076
1077 struct list_head event_handler_list;
1078 spinlock_t event_handler_lock;
1079
1080 spinlock_t client_data_lock;
1081 struct list_head core_list;
1082 struct list_head client_data_list;
1083
1084 struct ib_cache cache;
1085 int *pkey_tbl_len;
1086 int *gid_tbl_len;
1087
1088 int num_comp_vectors;
1089
1090 struct iw_cm_verbs *iwcm;
1091
1092 int (*get_protocol_stats)(struct ib_device *device,
1093 union rdma_protocol_stats *stats);
1094 int (*query_device)(struct ib_device *device,
1095 struct ib_device_attr *device_attr);
1096 int (*query_port)(struct ib_device *device,
1097 u8 port_num,
1098 struct ib_port_attr *port_attr);
1099 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1100 u8 port_num);
1101 int (*query_gid)(struct ib_device *device,
1102 u8 port_num, int index,
1103 union ib_gid *gid);
1104 int (*query_pkey)(struct ib_device *device,
1105 u8 port_num, u16 index, u16 *pkey);
1106 int (*modify_device)(struct ib_device *device,
1107 int device_modify_mask,
1108 struct ib_device_modify *device_modify);
1109 int (*modify_port)(struct ib_device *device,
1110 u8 port_num, int port_modify_mask,
1111 struct ib_port_modify *port_modify);
1112 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
1113 struct ib_udata *udata);
1114 int (*dealloc_ucontext)(struct ib_ucontext *context);
1115 int (*mmap)(struct ib_ucontext *context,
1116 struct vm_area_struct *vma);
1117 struct ib_pd * (*alloc_pd)(struct ib_device *device,
1118 struct ib_ucontext *context,
1119 struct ib_udata *udata);
1120 int (*dealloc_pd)(struct ib_pd *pd);
1121 struct ib_ah * (*create_ah)(struct ib_pd *pd,
1122 struct ib_ah_attr *ah_attr);
1123 int (*modify_ah)(struct ib_ah *ah,
1124 struct ib_ah_attr *ah_attr);
1125 int (*query_ah)(struct ib_ah *ah,
1126 struct ib_ah_attr *ah_attr);
1127 int (*destroy_ah)(struct ib_ah *ah);
1128 struct ib_srq * (*create_srq)(struct ib_pd *pd,
1129 struct ib_srq_init_attr *srq_init_attr,
1130 struct ib_udata *udata);
1131 int (*modify_srq)(struct ib_srq *srq,
1132 struct ib_srq_attr *srq_attr,
1133 enum ib_srq_attr_mask srq_attr_mask,
1134 struct ib_udata *udata);
1135 int (*query_srq)(struct ib_srq *srq,
1136 struct ib_srq_attr *srq_attr);
1137 int (*destroy_srq)(struct ib_srq *srq);
1138 int (*post_srq_recv)(struct ib_srq *srq,
1139 struct ib_recv_wr *recv_wr,
1140 struct ib_recv_wr **bad_recv_wr);
1141 struct ib_qp * (*create_qp)(struct ib_pd *pd,
1142 struct ib_qp_init_attr *qp_init_attr,
1143 struct ib_udata *udata);
1144 int (*modify_qp)(struct ib_qp *qp,
1145 struct ib_qp_attr *qp_attr,
1146 int qp_attr_mask,
1147 struct ib_udata *udata);
1148 int (*query_qp)(struct ib_qp *qp,
1149 struct ib_qp_attr *qp_attr,
1150 int qp_attr_mask,
1151 struct ib_qp_init_attr *qp_init_attr);
1152 int (*destroy_qp)(struct ib_qp *qp);
1153 int (*post_send)(struct ib_qp *qp,
1154 struct ib_send_wr *send_wr,
1155 struct ib_send_wr **bad_send_wr);
1156 int (*post_recv)(struct ib_qp *qp,
1157 struct ib_recv_wr *recv_wr,
1158 struct ib_recv_wr **bad_recv_wr);
1159 struct ib_cq * (*create_cq)(struct ib_device *device, int cqe,
1160 int comp_vector,
1161 struct ib_ucontext *context,
1162 struct ib_udata *udata);
1163 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1164 u16 cq_period);
1165 int (*destroy_cq)(struct ib_cq *cq);
1166 int (*resize_cq)(struct ib_cq *cq, int cqe,
1167 struct ib_udata *udata);
1168 int (*poll_cq)(struct ib_cq *cq, int num_entries,
1169 struct ib_wc *wc);
1170 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1171 int (*req_notify_cq)(struct ib_cq *cq,
1172 enum ib_cq_notify_flags flags);
1173 int (*req_ncomp_notif)(struct ib_cq *cq,
1174 int wc_cnt);
1175 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
1176 int mr_access_flags);
1177 struct ib_mr * (*reg_phys_mr)(struct ib_pd *pd,
1178 struct ib_phys_buf *phys_buf_array,
1179 int num_phys_buf,
1180 int mr_access_flags,
1181 u64 *iova_start);
1182 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
1183 u64 start, u64 length,
1184 u64 virt_addr,
1185 int mr_access_flags,
1186 struct ib_udata *udata);
1187 int (*query_mr)(struct ib_mr *mr,
1188 struct ib_mr_attr *mr_attr);
1189 int (*dereg_mr)(struct ib_mr *mr);
1190 struct ib_mr * (*alloc_fast_reg_mr)(struct ib_pd *pd,
1191 int max_page_list_len);
1192 struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1193 int page_list_len);
1194 void (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1195 int (*rereg_phys_mr)(struct ib_mr *mr,
1196 int mr_rereg_mask,
1197 struct ib_pd *pd,
1198 struct ib_phys_buf *phys_buf_array,
1199 int num_phys_buf,
1200 int mr_access_flags,
1201 u64 *iova_start);
1202 struct ib_mw * (*alloc_mw)(struct ib_pd *pd);
1203 int (*bind_mw)(struct ib_qp *qp,
1204 struct ib_mw *mw,
1205 struct ib_mw_bind *mw_bind);
1206 int (*dealloc_mw)(struct ib_mw *mw);
1207 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
1208 int mr_access_flags,
1209 struct ib_fmr_attr *fmr_attr);
1210 int (*map_phys_fmr)(struct ib_fmr *fmr,
1211 u64 *page_list, int list_len,
1212 u64 iova);
1213 int (*unmap_fmr)(struct list_head *fmr_list);
1214 int (*dealloc_fmr)(struct ib_fmr *fmr);
1215 int (*attach_mcast)(struct ib_qp *qp,
1216 union ib_gid *gid,
1217 u16 lid);
1218 int (*detach_mcast)(struct ib_qp *qp,
1219 union ib_gid *gid,
1220 u16 lid);
1221 int (*process_mad)(struct ib_device *device,
1222 int process_mad_flags,
1223 u8 port_num,
1224 struct ib_wc *in_wc,
1225 struct ib_grh *in_grh,
1226 struct ib_mad *in_mad,
1227 struct ib_mad *out_mad);
1228 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
1229 struct ib_ucontext *ucontext,
1230 struct ib_udata *udata);
1231 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1232
1233 struct ib_dma_mapping_ops *dma_ops;
1234
1235 struct module *owner;
1236 struct device dev;
1237 struct kobject *ports_parent;
1238 struct list_head port_list;
1239
1240 enum {
1241 IB_DEV_UNINITIALIZED,
1242 IB_DEV_REGISTERED,
1243 IB_DEV_UNREGISTERED
1244 } reg_state;
1245
1246 int uverbs_abi_ver;
1247 u64 uverbs_cmd_mask;
1248
1249 char node_desc[64];
1250 __be64 node_guid;
1251 u32 local_dma_lkey;
1252 u8 node_type;
1253 u8 phys_port_cnt;
1254};
1255
1256struct ib_client {
1257 char *name;
1258 void (*add) (struct ib_device *);
1259 void (*remove)(struct ib_device *);
1260
1261 struct list_head list;
1262};
1263
1264struct ib_device *ib_alloc_device(size_t size);
1265void ib_dealloc_device(struct ib_device *device);
1266
1267int ib_register_device(struct ib_device *device,
1268 int (*port_callback)(struct ib_device *,
1269 u8, struct kobject *));
1270void ib_unregister_device(struct ib_device *device);
1271
1272int ib_register_client (struct ib_client *client);
1273void ib_unregister_client(struct ib_client *client);
1274
1275void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1276void ib_set_client_data(struct ib_device *device, struct ib_client *client,
1277 void *data);
1278
1279static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1280{
1281 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1282}
1283
1284static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1285{
1286 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1287}
1288
1289/**
1290 * ib_modify_qp_is_ok - Check that the supplied attribute mask
1291 * contains all required attributes and no attributes not allowed for
1292 * the given QP state transition.
1293 * @cur_state: Current QP state
1294 * @next_state: Next QP state
1295 * @type: QP type
1296 * @mask: Mask of supplied QP attributes
1297 *
1298 * This function is a helper function that a low-level driver's
1299 * modify_qp method can use to validate the consumer's input. It
1300 * checks that cur_state and next_state are valid QP states, that a
1301 * transition from cur_state to next_state is allowed by the IB spec,
1302 * and that the attribute mask supplied is allowed for the transition.
1303 */
1304int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1305 enum ib_qp_type type, enum ib_qp_attr_mask mask);
1306
1307int ib_register_event_handler (struct ib_event_handler *event_handler);
1308int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1309void ib_dispatch_event(struct ib_event *event);
1310
1311int ib_query_device(struct ib_device *device,
1312 struct ib_device_attr *device_attr);
1313
1314int ib_query_port(struct ib_device *device,
1315 u8 port_num, struct ib_port_attr *port_attr);
1316
1317enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1318 u8 port_num);
1319
1320int ib_query_gid(struct ib_device *device,
1321 u8 port_num, int index, union ib_gid *gid);
1322
1323int ib_query_pkey(struct ib_device *device,
1324 u8 port_num, u16 index, u16 *pkey);
1325
1326int ib_modify_device(struct ib_device *device,
1327 int device_modify_mask,
1328 struct ib_device_modify *device_modify);
1329
1330int ib_modify_port(struct ib_device *device,
1331 u8 port_num, int port_modify_mask,
1332 struct ib_port_modify *port_modify);
1333
1334int ib_find_gid(struct ib_device *device, union ib_gid *gid,
1335 u8 *port_num, u16 *index);
1336
1337int ib_find_pkey(struct ib_device *device,
1338 u8 port_num, u16 pkey, u16 *index);
1339
1340/**
1341 * ib_alloc_pd - Allocates an unused protection domain.
1342 * @device: The device on which to allocate the protection domain.
1343 *
1344 * A protection domain object provides an association between QPs, shared
1345 * receive queues, address handles, memory regions, and memory windows.
1346 */
1347struct ib_pd *ib_alloc_pd(struct ib_device *device);
1348
1349/**
1350 * ib_dealloc_pd - Deallocates a protection domain.
1351 * @pd: The protection domain to deallocate.
1352 */
1353int ib_dealloc_pd(struct ib_pd *pd);
1354
1355/**
1356 * ib_create_ah - Creates an address handle for the given address vector.
1357 * @pd: The protection domain associated with the address handle.
1358 * @ah_attr: The attributes of the address vector.
1359 *
1360 * The address handle is used to reference a local or global destination
1361 * in all UD QP post sends.
1362 */
1363struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
1364
1365/**
1366 * ib_init_ah_from_wc - Initializes address handle attributes from a
1367 * work completion.
1368 * @device: Device on which the received message arrived.
1369 * @port_num: Port on which the received message arrived.
1370 * @wc: Work completion associated with the received message.
1371 * @grh: References the received global route header. This parameter is
1372 * ignored unless the work completion indicates that the GRH is valid.
1373 * @ah_attr: Returned attributes that can be used when creating an address
1374 * handle for replying to the message.
1375 */
1376int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, struct ib_wc *wc,
1377 struct ib_grh *grh, struct ib_ah_attr *ah_attr);
1378
1379/**
1380 * ib_create_ah_from_wc - Creates an address handle associated with the
1381 * sender of the specified work completion.
1382 * @pd: The protection domain associated with the address handle.
1383 * @wc: Work completion information associated with a received message.
1384 * @grh: References the received global route header. This parameter is
1385 * ignored unless the work completion indicates that the GRH is valid.
1386 * @port_num: The outbound port number to associate with the address.
1387 *
1388 * The address handle is used to reference a local or global destination
1389 * in all UD QP post sends.
1390 */
1391struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc,
1392 struct ib_grh *grh, u8 port_num);
1393
1394/**
1395 * ib_modify_ah - Modifies the address vector associated with an address
1396 * handle.
1397 * @ah: The address handle to modify.
1398 * @ah_attr: The new address vector attributes to associate with the
1399 * address handle.
1400 */
1401int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1402
1403/**
1404 * ib_query_ah - Queries the address vector associated with an address
1405 * handle.
1406 * @ah: The address handle to query.
1407 * @ah_attr: The address vector attributes associated with the address
1408 * handle.
1409 */
1410int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1411
1412/**
1413 * ib_destroy_ah - Destroys an address handle.
1414 * @ah: The address handle to destroy.
1415 */
1416int ib_destroy_ah(struct ib_ah *ah);
1417
1418/**
1419 * ib_create_srq - Creates a SRQ associated with the specified protection
1420 * domain.
1421 * @pd: The protection domain associated with the SRQ.
1422 * @srq_init_attr: A list of initial attributes required to create the
1423 * SRQ. If SRQ creation succeeds, then the attributes are updated to
1424 * the actual capabilities of the created SRQ.
1425 *
1426 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
1427 * requested size of the SRQ, and set to the actual values allocated
1428 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
1429 * will always be at least as large as the requested values.
1430 */
1431struct ib_srq *ib_create_srq(struct ib_pd *pd,
1432 struct ib_srq_init_attr *srq_init_attr);
1433
1434/**
1435 * ib_modify_srq - Modifies the attributes for the specified SRQ.
1436 * @srq: The SRQ to modify.
1437 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
1438 * the current values of selected SRQ attributes are returned.
1439 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
1440 * are being modified.
1441 *
1442 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
1443 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
1444 * the number of receives queued drops below the limit.
1445 */
1446int ib_modify_srq(struct ib_srq *srq,
1447 struct ib_srq_attr *srq_attr,
1448 enum ib_srq_attr_mask srq_attr_mask);
1449
1450/**
1451 * ib_query_srq - Returns the attribute list and current values for the
1452 * specified SRQ.
1453 * @srq: The SRQ to query.
1454 * @srq_attr: The attributes of the specified SRQ.
1455 */
1456int ib_query_srq(struct ib_srq *srq,
1457 struct ib_srq_attr *srq_attr);
1458
1459/**
1460 * ib_destroy_srq - Destroys the specified SRQ.
1461 * @srq: The SRQ to destroy.
1462 */
1463int ib_destroy_srq(struct ib_srq *srq);
1464
1465/**
1466 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
1467 * @srq: The SRQ to post the work request on.
1468 * @recv_wr: A list of work requests to post on the receive queue.
1469 * @bad_recv_wr: On an immediate failure, this parameter will reference
1470 * the work request that failed to be posted on the QP.
1471 */
1472static inline int ib_post_srq_recv(struct ib_srq *srq,
1473 struct ib_recv_wr *recv_wr,
1474 struct ib_recv_wr **bad_recv_wr)
1475{
1476 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
1477}
1478
1479/**
1480 * ib_create_qp - Creates a QP associated with the specified protection
1481 * domain.
1482 * @pd: The protection domain associated with the QP.
1483 * @qp_init_attr: A list of initial attributes required to create the
1484 * QP. If QP creation succeeds, then the attributes are updated to
1485 * the actual capabilities of the created QP.
1486 */
1487struct ib_qp *ib_create_qp(struct ib_pd *pd,
1488 struct ib_qp_init_attr *qp_init_attr);
1489
1490/**
1491 * ib_modify_qp - Modifies the attributes for the specified QP and then
1492 * transitions the QP to the given state.
1493 * @qp: The QP to modify.
1494 * @qp_attr: On input, specifies the QP attributes to modify. On output,
1495 * the current values of selected QP attributes are returned.
1496 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
1497 * are being modified.
1498 */
1499int ib_modify_qp(struct ib_qp *qp,
1500 struct ib_qp_attr *qp_attr,
1501 int qp_attr_mask);
1502
1503/**
1504 * ib_query_qp - Returns the attribute list and current values for the
1505 * specified QP.
1506 * @qp: The QP to query.
1507 * @qp_attr: The attributes of the specified QP.
1508 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
1509 * @qp_init_attr: Additional attributes of the selected QP.
1510 *
1511 * The qp_attr_mask may be used to limit the query to gathering only the
1512 * selected attributes.
1513 */
1514int ib_query_qp(struct ib_qp *qp,
1515 struct ib_qp_attr *qp_attr,
1516 int qp_attr_mask,
1517 struct ib_qp_init_attr *qp_init_attr);
1518
1519/**
1520 * ib_destroy_qp - Destroys the specified QP.
1521 * @qp: The QP to destroy.
1522 */
1523int ib_destroy_qp(struct ib_qp *qp);
1524
1525/**
1526 * ib_open_qp - Obtain a reference to an existing sharable QP.
1527 * @xrcd - XRC domain
1528 * @qp_open_attr: Attributes identifying the QP to open.
1529 *
1530 * Returns a reference to a sharable QP.
1531 */
1532struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
1533 struct ib_qp_open_attr *qp_open_attr);
1534
1535/**
1536 * ib_close_qp - Release an external reference to a QP.
1537 * @qp: The QP handle to release
1538 *
1539 * The opened QP handle is released by the caller. The underlying
1540 * shared QP is not destroyed until all internal references are released.
1541 */
1542int ib_close_qp(struct ib_qp *qp);
1543
1544/**
1545 * ib_post_send - Posts a list of work requests to the send queue of
1546 * the specified QP.
1547 * @qp: The QP to post the work request on.
1548 * @send_wr: A list of work requests to post on the send queue.
1549 * @bad_send_wr: On an immediate failure, this parameter will reference
1550 * the work request that failed to be posted on the QP.
1551 *
1552 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
1553 * error is returned, the QP state shall not be affected,
1554 * ib_post_send() will return an immediate error after queueing any
1555 * earlier work requests in the list.
1556 */
1557static inline int ib_post_send(struct ib_qp *qp,
1558 struct ib_send_wr *send_wr,
1559 struct ib_send_wr **bad_send_wr)
1560{
1561 return qp->device->post_send(qp, send_wr, bad_send_wr);
1562}
1563
1564/**
1565 * ib_post_recv - Posts a list of work requests to the receive queue of
1566 * the specified QP.
1567 * @qp: The QP to post the work request on.
1568 * @recv_wr: A list of work requests to post on the receive queue.
1569 * @bad_recv_wr: On an immediate failure, this parameter will reference
1570 * the work request that failed to be posted on the QP.
1571 */
1572static inline int ib_post_recv(struct ib_qp *qp,
1573 struct ib_recv_wr *recv_wr,
1574 struct ib_recv_wr **bad_recv_wr)
1575{
1576 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
1577}
1578
1579/**
1580 * ib_create_cq - Creates a CQ on the specified device.
1581 * @device: The device on which to create the CQ.
1582 * @comp_handler: A user-specified callback that is invoked when a
1583 * completion event occurs on the CQ.
1584 * @event_handler: A user-specified callback that is invoked when an
1585 * asynchronous event not associated with a completion occurs on the CQ.
1586 * @cq_context: Context associated with the CQ returned to the user via
1587 * the associated completion and event handlers.
1588 * @cqe: The minimum size of the CQ.
1589 * @comp_vector - Completion vector used to signal completion events.
1590 * Must be >= 0 and < context->num_comp_vectors.
1591 *
1592 * Users can examine the cq structure to determine the actual CQ size.
1593 */
1594struct ib_cq *ib_create_cq(struct ib_device *device,
1595 ib_comp_handler comp_handler,
1596 void (*event_handler)(struct ib_event *, void *),
1597 void *cq_context, int cqe, int comp_vector);
1598
1599/**
1600 * ib_resize_cq - Modifies the capacity of the CQ.
1601 * @cq: The CQ to resize.
1602 * @cqe: The minimum size of the CQ.
1603 *
1604 * Users can examine the cq structure to determine the actual CQ size.
1605 */
1606int ib_resize_cq(struct ib_cq *cq, int cqe);
1607
1608/**
1609 * ib_modify_cq - Modifies moderation params of the CQ
1610 * @cq: The CQ to modify.
1611 * @cq_count: number of CQEs that will trigger an event
1612 * @cq_period: max period of time in usec before triggering an event
1613 *
1614 */
1615int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1616
1617/**
1618 * ib_destroy_cq - Destroys the specified CQ.
1619 * @cq: The CQ to destroy.
1620 */
1621int ib_destroy_cq(struct ib_cq *cq);
1622
1623/**
1624 * ib_poll_cq - poll a CQ for completion(s)
1625 * @cq:the CQ being polled
1626 * @num_entries:maximum number of completions to return
1627 * @wc:array of at least @num_entries &struct ib_wc where completions
1628 * will be returned
1629 *
1630 * Poll a CQ for (possibly multiple) completions. If the return value
1631 * is < 0, an error occurred. If the return value is >= 0, it is the
1632 * number of completions returned. If the return value is
1633 * non-negative and < num_entries, then the CQ was emptied.
1634 */
1635static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
1636 struct ib_wc *wc)
1637{
1638 return cq->device->poll_cq(cq, num_entries, wc);
1639}
1640
1641/**
1642 * ib_peek_cq - Returns the number of unreaped completions currently
1643 * on the specified CQ.
1644 * @cq: The CQ to peek.
1645 * @wc_cnt: A minimum number of unreaped completions to check for.
1646 *
1647 * If the number of unreaped completions is greater than or equal to wc_cnt,
1648 * this function returns wc_cnt, otherwise, it returns the actual number of
1649 * unreaped completions.
1650 */
1651int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
1652
1653/**
1654 * ib_req_notify_cq - Request completion notification on a CQ.
1655 * @cq: The CQ to generate an event for.
1656 * @flags:
1657 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
1658 * to request an event on the next solicited event or next work
1659 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
1660 * may also be |ed in to request a hint about missed events, as
1661 * described below.
1662 *
1663 * Return Value:
1664 * < 0 means an error occurred while requesting notification
1665 * == 0 means notification was requested successfully, and if
1666 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
1667 * were missed and it is safe to wait for another event. In
1668 * this case is it guaranteed that any work completions added
1669 * to the CQ since the last CQ poll will trigger a completion
1670 * notification event.
1671 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
1672 * in. It means that the consumer must poll the CQ again to
1673 * make sure it is empty to avoid missing an event because of a
1674 * race between requesting notification and an entry being
1675 * added to the CQ. This return value means it is possible
1676 * (but not guaranteed) that a work completion has been added
1677 * to the CQ since the last poll without triggering a
1678 * completion notification event.
1679 */
1680static inline int ib_req_notify_cq(struct ib_cq *cq,
1681 enum ib_cq_notify_flags flags)
1682{
1683 return cq->device->req_notify_cq(cq, flags);
1684}
1685
1686/**
1687 * ib_req_ncomp_notif - Request completion notification when there are
1688 * at least the specified number of unreaped completions on the CQ.
1689 * @cq: The CQ to generate an event for.
1690 * @wc_cnt: The number of unreaped completions that should be on the
1691 * CQ before an event is generated.
1692 */
1693static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
1694{
1695 return cq->device->req_ncomp_notif ?
1696 cq->device->req_ncomp_notif(cq, wc_cnt) :
1697 -ENOSYS;
1698}
1699
1700/**
1701 * ib_get_dma_mr - Returns a memory region for system memory that is
1702 * usable for DMA.
1703 * @pd: The protection domain associated with the memory region.
1704 * @mr_access_flags: Specifies the memory access rights.
1705 *
1706 * Note that the ib_dma_*() functions defined below must be used
1707 * to create/destroy addresses used with the Lkey or Rkey returned
1708 * by ib_get_dma_mr().
1709 */
1710struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
1711
1712/**
1713 * ib_dma_mapping_error - check a DMA addr for error
1714 * @dev: The device for which the dma_addr was created
1715 * @dma_addr: The DMA address to check
1716 */
1717static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
1718{
1719 if (dev->dma_ops)
1720 return dev->dma_ops->mapping_error(dev, dma_addr);
1721 return dma_mapping_error(dev->dma_device, dma_addr);
1722}
1723
1724/**
1725 * ib_dma_map_single - Map a kernel virtual address to DMA address
1726 * @dev: The device for which the dma_addr is to be created
1727 * @cpu_addr: The kernel virtual address
1728 * @size: The size of the region in bytes
1729 * @direction: The direction of the DMA
1730 */
1731static inline u64 ib_dma_map_single(struct ib_device *dev,
1732 void *cpu_addr, size_t size,
1733 enum dma_data_direction direction)
1734{
1735 if (dev->dma_ops)
1736 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
1737 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
1738}
1739
1740/**
1741 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
1742 * @dev: The device for which the DMA address was created
1743 * @addr: The DMA address
1744 * @size: The size of the region in bytes
1745 * @direction: The direction of the DMA
1746 */
1747static inline void ib_dma_unmap_single(struct ib_device *dev,
1748 u64 addr, size_t size,
1749 enum dma_data_direction direction)
1750{
1751 if (dev->dma_ops)
1752 dev->dma_ops->unmap_single(dev, addr, size, direction);
1753 else
1754 dma_unmap_single(dev->dma_device, addr, size, direction);
1755}
1756
1757static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
1758 void *cpu_addr, size_t size,
1759 enum dma_data_direction direction,
1760 struct dma_attrs *attrs)
1761{
1762 return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
1763 direction, attrs);
1764}
1765
1766static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
1767 u64 addr, size_t size,
1768 enum dma_data_direction direction,
1769 struct dma_attrs *attrs)
1770{
1771 return dma_unmap_single_attrs(dev->dma_device, addr, size,
1772 direction, attrs);
1773}
1774
1775/**
1776 * ib_dma_map_page - Map a physical page to DMA address
1777 * @dev: The device for which the dma_addr is to be created
1778 * @page: The page to be mapped
1779 * @offset: The offset within the page
1780 * @size: The size of the region in bytes
1781 * @direction: The direction of the DMA
1782 */
1783static inline u64 ib_dma_map_page(struct ib_device *dev,
1784 struct page *page,
1785 unsigned long offset,
1786 size_t size,
1787 enum dma_data_direction direction)
1788{
1789 if (dev->dma_ops)
1790 return dev->dma_ops->map_page(dev, page, offset, size, direction);
1791 return dma_map_page(dev->dma_device, page, offset, size, direction);
1792}
1793
1794/**
1795 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
1796 * @dev: The device for which the DMA address was created
1797 * @addr: The DMA address
1798 * @size: The size of the region in bytes
1799 * @direction: The direction of the DMA
1800 */
1801static inline void ib_dma_unmap_page(struct ib_device *dev,
1802 u64 addr, size_t size,
1803 enum dma_data_direction direction)
1804{
1805 if (dev->dma_ops)
1806 dev->dma_ops->unmap_page(dev, addr, size, direction);
1807 else
1808 dma_unmap_page(dev->dma_device, addr, size, direction);
1809}
1810
1811/**
1812 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
1813 * @dev: The device for which the DMA addresses are to be created
1814 * @sg: The array of scatter/gather entries
1815 * @nents: The number of scatter/gather entries
1816 * @direction: The direction of the DMA
1817 */
1818static inline int ib_dma_map_sg(struct ib_device *dev,
1819 struct scatterlist *sg, int nents,
1820 enum dma_data_direction direction)
1821{
1822 if (dev->dma_ops)
1823 return dev->dma_ops->map_sg(dev, sg, nents, direction);
1824 return dma_map_sg(dev->dma_device, sg, nents, direction);
1825}
1826
1827/**
1828 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
1829 * @dev: The device for which the DMA addresses were created
1830 * @sg: The array of scatter/gather entries
1831 * @nents: The number of scatter/gather entries
1832 * @direction: The direction of the DMA
1833 */
1834static inline void ib_dma_unmap_sg(struct ib_device *dev,
1835 struct scatterlist *sg, int nents,
1836 enum dma_data_direction direction)
1837{
1838 if (dev->dma_ops)
1839 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
1840 else
1841 dma_unmap_sg(dev->dma_device, sg, nents, direction);
1842}
1843
1844static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
1845 struct scatterlist *sg, int nents,
1846 enum dma_data_direction direction,
1847 struct dma_attrs *attrs)
1848{
1849 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
1850}
1851
1852static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
1853 struct scatterlist *sg, int nents,
1854 enum dma_data_direction direction,
1855 struct dma_attrs *attrs)
1856{
1857 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
1858}
1859/**
1860 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
1861 * @dev: The device for which the DMA addresses were created
1862 * @sg: The scatter/gather entry
1863 */
1864static inline u64 ib_sg_dma_address(struct ib_device *dev,
1865 struct scatterlist *sg)
1866{
1867 if (dev->dma_ops)
1868 return dev->dma_ops->dma_address(dev, sg);
1869 return sg_dma_address(sg);
1870}
1871
1872/**
1873 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
1874 * @dev: The device for which the DMA addresses were created
1875 * @sg: The scatter/gather entry
1876 */
1877static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
1878 struct scatterlist *sg)
1879{
1880 if (dev->dma_ops)
1881 return dev->dma_ops->dma_len(dev, sg);
1882 return sg_dma_len(sg);
1883}
1884
1885/**
1886 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
1887 * @dev: The device for which the DMA address was created
1888 * @addr: The DMA address
1889 * @size: The size of the region in bytes
1890 * @dir: The direction of the DMA
1891 */
1892static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
1893 u64 addr,
1894 size_t size,
1895 enum dma_data_direction dir)
1896{
1897 if (dev->dma_ops)
1898 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
1899 else
1900 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
1901}
1902
1903/**
1904 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
1905 * @dev: The device for which the DMA address was created
1906 * @addr: The DMA address
1907 * @size: The size of the region in bytes
1908 * @dir: The direction of the DMA
1909 */
1910static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
1911 u64 addr,
1912 size_t size,
1913 enum dma_data_direction dir)
1914{
1915 if (dev->dma_ops)
1916 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
1917 else
1918 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
1919}
1920
1921/**
1922 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
1923 * @dev: The device for which the DMA address is requested
1924 * @size: The size of the region to allocate in bytes
1925 * @dma_handle: A pointer for returning the DMA address of the region
1926 * @flag: memory allocator flags
1927 */
1928static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
1929 size_t size,
1930 u64 *dma_handle,
1931 gfp_t flag)
1932{
1933 if (dev->dma_ops)
1934 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
1935 else {
1936 dma_addr_t handle;
1937 void *ret;
1938
1939 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
1940 *dma_handle = handle;
1941 return ret;
1942 }
1943}
1944
1945/**
1946 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
1947 * @dev: The device for which the DMA addresses were allocated
1948 * @size: The size of the region
1949 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
1950 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
1951 */
1952static inline void ib_dma_free_coherent(struct ib_device *dev,
1953 size_t size, void *cpu_addr,
1954 u64 dma_handle)
1955{
1956 if (dev->dma_ops)
1957 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
1958 else
1959 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
1960}
1961
1962/**
1963 * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
1964 * by an HCA.
1965 * @pd: The protection domain associated assigned to the registered region.
1966 * @phys_buf_array: Specifies a list of physical buffers to use in the
1967 * memory region.
1968 * @num_phys_buf: Specifies the size of the phys_buf_array.
1969 * @mr_access_flags: Specifies the memory access rights.
1970 * @iova_start: The offset of the region's starting I/O virtual address.
1971 */
1972struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
1973 struct ib_phys_buf *phys_buf_array,
1974 int num_phys_buf,
1975 int mr_access_flags,
1976 u64 *iova_start);
1977
1978/**
1979 * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
1980 * Conceptually, this call performs the functions deregister memory region
1981 * followed by register physical memory region. Where possible,
1982 * resources are reused instead of deallocated and reallocated.
1983 * @mr: The memory region to modify.
1984 * @mr_rereg_mask: A bit-mask used to indicate which of the following
1985 * properties of the memory region are being modified.
1986 * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
1987 * the new protection domain to associated with the memory region,
1988 * otherwise, this parameter is ignored.
1989 * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
1990 * field specifies a list of physical buffers to use in the new
1991 * translation, otherwise, this parameter is ignored.
1992 * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
1993 * field specifies the size of the phys_buf_array, otherwise, this
1994 * parameter is ignored.
1995 * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
1996 * field specifies the new memory access rights, otherwise, this
1997 * parameter is ignored.
1998 * @iova_start: The offset of the region's starting I/O virtual address.
1999 */
2000int ib_rereg_phys_mr(struct ib_mr *mr,
2001 int mr_rereg_mask,
2002 struct ib_pd *pd,
2003 struct ib_phys_buf *phys_buf_array,
2004 int num_phys_buf,
2005 int mr_access_flags,
2006 u64 *iova_start);
2007
2008/**
2009 * ib_query_mr - Retrieves information about a specific memory region.
2010 * @mr: The memory region to retrieve information about.
2011 * @mr_attr: The attributes of the specified memory region.
2012 */
2013int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
2014
2015/**
2016 * ib_dereg_mr - Deregisters a memory region and removes it from the
2017 * HCA translation table.
2018 * @mr: The memory region to deregister.
2019 */
2020int ib_dereg_mr(struct ib_mr *mr);
2021
2022/**
2023 * ib_alloc_fast_reg_mr - Allocates memory region usable with the
2024 * IB_WR_FAST_REG_MR send work request.
2025 * @pd: The protection domain associated with the region.
2026 * @max_page_list_len: requested max physical buffer list length to be
2027 * used with fast register work requests for this MR.
2028 */
2029struct ib_mr *ib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len);
2030
2031/**
2032 * ib_alloc_fast_reg_page_list - Allocates a page list array
2033 * @device - ib device pointer.
2034 * @page_list_len - size of the page list array to be allocated.
2035 *
2036 * This allocates and returns a struct ib_fast_reg_page_list * and a
2037 * page_list array that is at least page_list_len in size. The actual
2038 * size is returned in max_page_list_len. The caller is responsible
2039 * for initializing the contents of the page_list array before posting
2040 * a send work request with the IB_WC_FAST_REG_MR opcode.
2041 *
2042 * The page_list array entries must be translated using one of the
2043 * ib_dma_*() functions just like the addresses passed to
2044 * ib_map_phys_fmr(). Once the ib_post_send() is issued, the struct
2045 * ib_fast_reg_page_list must not be modified by the caller until the
2046 * IB_WC_FAST_REG_MR work request completes.
2047 */
2048struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
2049 struct ib_device *device, int page_list_len);
2050
2051/**
2052 * ib_free_fast_reg_page_list - Deallocates a previously allocated
2053 * page list array.
2054 * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
2055 */
2056void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
2057
2058/**
2059 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2060 * R_Key and L_Key.
2061 * @mr - struct ib_mr pointer to be updated.
2062 * @newkey - new key to be used.
2063 */
2064static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2065{
2066 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2067 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2068}
2069
2070/**
2071 * ib_alloc_mw - Allocates a memory window.
2072 * @pd: The protection domain associated with the memory window.
2073 */
2074struct ib_mw *ib_alloc_mw(struct ib_pd *pd);
2075
2076/**
2077 * ib_bind_mw - Posts a work request to the send queue of the specified
2078 * QP, which binds the memory window to the given address range and
2079 * remote access attributes.
2080 * @qp: QP to post the bind work request on.
2081 * @mw: The memory window to bind.
2082 * @mw_bind: Specifies information about the memory window, including
2083 * its address range, remote access rights, and associated memory region.
2084 */
2085static inline int ib_bind_mw(struct ib_qp *qp,
2086 struct ib_mw *mw,
2087 struct ib_mw_bind *mw_bind)
2088{
2089 /* XXX reference counting in corresponding MR? */
2090 return mw->device->bind_mw ?
2091 mw->device->bind_mw(qp, mw, mw_bind) :
2092 -ENOSYS;
2093}
2094
2095/**
2096 * ib_dealloc_mw - Deallocates a memory window.
2097 * @mw: The memory window to deallocate.
2098 */
2099int ib_dealloc_mw(struct ib_mw *mw);
2100
2101/**
2102 * ib_alloc_fmr - Allocates a unmapped fast memory region.
2103 * @pd: The protection domain associated with the unmapped region.
2104 * @mr_access_flags: Specifies the memory access rights.
2105 * @fmr_attr: Attributes of the unmapped region.
2106 *
2107 * A fast memory region must be mapped before it can be used as part of
2108 * a work request.
2109 */
2110struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2111 int mr_access_flags,
2112 struct ib_fmr_attr *fmr_attr);
2113
2114/**
2115 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2116 * @fmr: The fast memory region to associate with the pages.
2117 * @page_list: An array of physical pages to map to the fast memory region.
2118 * @list_len: The number of pages in page_list.
2119 * @iova: The I/O virtual address to use with the mapped region.
2120 */
2121static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2122 u64 *page_list, int list_len,
2123 u64 iova)
2124{
2125 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2126}
2127
2128/**
2129 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2130 * @fmr_list: A linked list of fast memory regions to unmap.
2131 */
2132int ib_unmap_fmr(struct list_head *fmr_list);
2133
2134/**
2135 * ib_dealloc_fmr - Deallocates a fast memory region.
2136 * @fmr: The fast memory region to deallocate.
2137 */
2138int ib_dealloc_fmr(struct ib_fmr *fmr);
2139
2140/**
2141 * ib_attach_mcast - Attaches the specified QP to a multicast group.
2142 * @qp: QP to attach to the multicast group. The QP must be type
2143 * IB_QPT_UD.
2144 * @gid: Multicast group GID.
2145 * @lid: Multicast group LID in host byte order.
2146 *
2147 * In order to send and receive multicast packets, subnet
2148 * administration must have created the multicast group and configured
2149 * the fabric appropriately. The port associated with the specified
2150 * QP must also be a member of the multicast group.
2151 */
2152int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2153
2154/**
2155 * ib_detach_mcast - Detaches the specified QP from a multicast group.
2156 * @qp: QP to detach from the multicast group.
2157 * @gid: Multicast group GID.
2158 * @lid: Multicast group LID in host byte order.
2159 */
2160int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2161
2162/**
2163 * ib_alloc_xrcd - Allocates an XRC domain.
2164 * @device: The device on which to allocate the XRC domain.
2165 */
2166struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
2167
2168/**
2169 * ib_dealloc_xrcd - Deallocates an XRC domain.
2170 * @xrcd: The XRC domain to deallocate.
2171 */
2172int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
2173
2174#endif /* IB_VERBS_H */