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#include <linux/pci-tph.h>
46
47#define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
48
49struct ib_umem_odp;
50struct ib_uqp_object;
51struct ib_usrq_object;
52struct ib_uwq_object;
53struct rdma_cm_id;
54struct ib_port;
55struct hw_stats_device_data;
56
57extern struct workqueue_struct *ib_wq;
58extern struct workqueue_struct *ib_comp_wq;
59extern struct workqueue_struct *ib_comp_unbound_wq;
60
61struct ib_ucq_object;
62
63__printf(2, 3) __cold
64void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
65__printf(2, 3) __cold
66void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
67__printf(2, 3) __cold
68void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
69__printf(2, 3) __cold
70void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
71__printf(2, 3) __cold
72void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
73__printf(2, 3) __cold
74void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
75__printf(2, 3) __cold
76void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
77
78#if defined(CONFIG_DYNAMIC_DEBUG) || \
79 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
80#define ibdev_dbg(__dev, format, args...) \
81 dynamic_ibdev_dbg(__dev, format, ##args)
82#else
83__printf(2, 3) __cold
84static inline
85void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
86#endif
87
88#define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \
89do { \
90 static DEFINE_RATELIMIT_STATE(_rs, \
91 DEFAULT_RATELIMIT_INTERVAL, \
92 DEFAULT_RATELIMIT_BURST); \
93 if (__ratelimit(&_rs)) \
94 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \
95} while (0)
96
97#define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
98 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
99#define ibdev_alert_ratelimited(ibdev, fmt, ...) \
100 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
101#define ibdev_crit_ratelimited(ibdev, fmt, ...) \
102 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
103#define ibdev_err_ratelimited(ibdev, fmt, ...) \
104 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
105#define ibdev_warn_ratelimited(ibdev, fmt, ...) \
106 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
107#define ibdev_notice_ratelimited(ibdev, fmt, ...) \
108 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
109#define ibdev_info_ratelimited(ibdev, fmt, ...) \
110 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
111
112#if defined(CONFIG_DYNAMIC_DEBUG) || \
113 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
114/* descriptor check is first to prevent flooding with "callbacks suppressed" */
115#define ibdev_dbg_ratelimited(ibdev, fmt, ...) \
116do { \
117 static DEFINE_RATELIMIT_STATE(_rs, \
118 DEFAULT_RATELIMIT_INTERVAL, \
119 DEFAULT_RATELIMIT_BURST); \
120 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \
121 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \
122 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \
123 ##__VA_ARGS__); \
124} while (0)
125#else
126__printf(2, 3) __cold
127static inline
128void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
129#endif
130
131union ib_gid {
132 u8 raw[16];
133 struct {
134 __be64 subnet_prefix;
135 __be64 interface_id;
136 } global;
137};
138
139extern union ib_gid zgid;
140
141enum ib_gid_type {
142 IB_GID_TYPE_IB = IB_UVERBS_GID_TYPE_IB,
143 IB_GID_TYPE_ROCE = IB_UVERBS_GID_TYPE_ROCE_V1,
144 IB_GID_TYPE_ROCE_UDP_ENCAP = IB_UVERBS_GID_TYPE_ROCE_V2,
145 IB_GID_TYPE_SIZE
146};
147
148#define ROCE_V2_UDP_DPORT 4791
149struct ib_gid_attr {
150 struct net_device __rcu *ndev;
151 struct ib_device *device;
152 union ib_gid gid;
153 enum ib_gid_type gid_type;
154 u16 index;
155 u32 port_num;
156};
157
158enum {
159 /* set the local administered indication */
160 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
161};
162
163enum rdma_transport_type {
164 RDMA_TRANSPORT_IB,
165 RDMA_TRANSPORT_IWARP,
166 RDMA_TRANSPORT_USNIC,
167 RDMA_TRANSPORT_USNIC_UDP,
168 RDMA_TRANSPORT_UNSPECIFIED,
169};
170
171enum rdma_protocol_type {
172 RDMA_PROTOCOL_IB,
173 RDMA_PROTOCOL_IBOE,
174 RDMA_PROTOCOL_IWARP,
175 RDMA_PROTOCOL_USNIC_UDP
176};
177
178__attribute_const__ enum rdma_transport_type
179rdma_node_get_transport(unsigned int node_type);
180
181enum rdma_network_type {
182 RDMA_NETWORK_IB,
183 RDMA_NETWORK_ROCE_V1,
184 RDMA_NETWORK_IPV4,
185 RDMA_NETWORK_IPV6
186};
187
188static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
189{
190 if (network_type == RDMA_NETWORK_IPV4 ||
191 network_type == RDMA_NETWORK_IPV6)
192 return IB_GID_TYPE_ROCE_UDP_ENCAP;
193 else if (network_type == RDMA_NETWORK_ROCE_V1)
194 return IB_GID_TYPE_ROCE;
195 else
196 return IB_GID_TYPE_IB;
197}
198
199static inline enum rdma_network_type
200rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
201{
202 if (attr->gid_type == IB_GID_TYPE_IB)
203 return RDMA_NETWORK_IB;
204
205 if (attr->gid_type == IB_GID_TYPE_ROCE)
206 return RDMA_NETWORK_ROCE_V1;
207
208 if (ipv6_addr_v4mapped(a: (struct in6_addr *)&attr->gid))
209 return RDMA_NETWORK_IPV4;
210 else
211 return RDMA_NETWORK_IPV6;
212}
213
214enum rdma_link_layer {
215 IB_LINK_LAYER_UNSPECIFIED,
216 IB_LINK_LAYER_INFINIBAND,
217 IB_LINK_LAYER_ETHERNET,
218};
219
220enum ib_device_cap_flags {
221 IB_DEVICE_RESIZE_MAX_WR = IB_UVERBS_DEVICE_RESIZE_MAX_WR,
222 IB_DEVICE_BAD_PKEY_CNTR = IB_UVERBS_DEVICE_BAD_PKEY_CNTR,
223 IB_DEVICE_BAD_QKEY_CNTR = IB_UVERBS_DEVICE_BAD_QKEY_CNTR,
224 IB_DEVICE_RAW_MULTI = IB_UVERBS_DEVICE_RAW_MULTI,
225 IB_DEVICE_AUTO_PATH_MIG = IB_UVERBS_DEVICE_AUTO_PATH_MIG,
226 IB_DEVICE_CHANGE_PHY_PORT = IB_UVERBS_DEVICE_CHANGE_PHY_PORT,
227 IB_DEVICE_UD_AV_PORT_ENFORCE = IB_UVERBS_DEVICE_UD_AV_PORT_ENFORCE,
228 IB_DEVICE_CURR_QP_STATE_MOD = IB_UVERBS_DEVICE_CURR_QP_STATE_MOD,
229 IB_DEVICE_SHUTDOWN_PORT = IB_UVERBS_DEVICE_SHUTDOWN_PORT,
230 /* IB_DEVICE_INIT_TYPE = IB_UVERBS_DEVICE_INIT_TYPE, (not in use) */
231 IB_DEVICE_PORT_ACTIVE_EVENT = IB_UVERBS_DEVICE_PORT_ACTIVE_EVENT,
232 IB_DEVICE_SYS_IMAGE_GUID = IB_UVERBS_DEVICE_SYS_IMAGE_GUID,
233 IB_DEVICE_RC_RNR_NAK_GEN = IB_UVERBS_DEVICE_RC_RNR_NAK_GEN,
234 IB_DEVICE_SRQ_RESIZE = IB_UVERBS_DEVICE_SRQ_RESIZE,
235 IB_DEVICE_N_NOTIFY_CQ = IB_UVERBS_DEVICE_N_NOTIFY_CQ,
236
237 /* Reserved, old SEND_W_INV = 1 << 16,*/
238 IB_DEVICE_MEM_WINDOW = IB_UVERBS_DEVICE_MEM_WINDOW,
239 /*
240 * Devices should set IB_DEVICE_UD_IP_SUM if they support
241 * insertion of UDP and TCP checksum on outgoing UD IPoIB
242 * messages and can verify the validity of checksum for
243 * incoming messages. Setting this flag implies that the
244 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
245 */
246 IB_DEVICE_UD_IP_CSUM = IB_UVERBS_DEVICE_UD_IP_CSUM,
247 IB_DEVICE_XRC = IB_UVERBS_DEVICE_XRC,
248
249 /*
250 * This device supports the IB "base memory management extension",
251 * which includes support for fast registrations (IB_WR_REG_MR,
252 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
253 * also be set by any iWarp device which must support FRs to comply
254 * to the iWarp verbs spec. iWarp devices also support the
255 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
256 * stag.
257 */
258 IB_DEVICE_MEM_MGT_EXTENSIONS = IB_UVERBS_DEVICE_MEM_MGT_EXTENSIONS,
259 IB_DEVICE_MEM_WINDOW_TYPE_2A = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2A,
260 IB_DEVICE_MEM_WINDOW_TYPE_2B = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2B,
261 IB_DEVICE_RC_IP_CSUM = IB_UVERBS_DEVICE_RC_IP_CSUM,
262 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
263 IB_DEVICE_RAW_IP_CSUM = IB_UVERBS_DEVICE_RAW_IP_CSUM,
264 IB_DEVICE_MANAGED_FLOW_STEERING =
265 IB_UVERBS_DEVICE_MANAGED_FLOW_STEERING,
266 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
267 IB_DEVICE_RAW_SCATTER_FCS = IB_UVERBS_DEVICE_RAW_SCATTER_FCS,
268 /* The device supports padding incoming writes to cacheline. */
269 IB_DEVICE_PCI_WRITE_END_PADDING =
270 IB_UVERBS_DEVICE_PCI_WRITE_END_PADDING,
271 /* Placement type attributes */
272 IB_DEVICE_FLUSH_GLOBAL = IB_UVERBS_DEVICE_FLUSH_GLOBAL,
273 IB_DEVICE_FLUSH_PERSISTENT = IB_UVERBS_DEVICE_FLUSH_PERSISTENT,
274 IB_DEVICE_ATOMIC_WRITE = IB_UVERBS_DEVICE_ATOMIC_WRITE,
275};
276
277enum ib_kernel_cap_flags {
278 /*
279 * This device supports a per-device lkey or stag that can be
280 * used without performing a memory registration for the local
281 * memory. Note that ULPs should never check this flag, but
282 * instead of use the local_dma_lkey flag in the ib_pd structure,
283 * which will always contain a usable lkey.
284 */
285 IBK_LOCAL_DMA_LKEY = 1 << 0,
286 /* IB_QP_CREATE_INTEGRITY_EN is supported to implement T10-PI */
287 IBK_INTEGRITY_HANDOVER = 1 << 1,
288 /* IB_ACCESS_ON_DEMAND is supported during reg_user_mr() */
289 IBK_ON_DEMAND_PAGING = 1 << 2,
290 /* IB_MR_TYPE_SG_GAPS is supported */
291 IBK_SG_GAPS_REG = 1 << 3,
292 /* Driver supports RDMA_NLDEV_CMD_DELLINK */
293 IBK_ALLOW_USER_UNREG = 1 << 4,
294
295 /* ipoib will use IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK */
296 IBK_BLOCK_MULTICAST_LOOPBACK = 1 << 5,
297 /* iopib will use IB_QP_CREATE_IPOIB_UD_LSO for its QPs */
298 IBK_UD_TSO = 1 << 6,
299 /* iopib will use the device ops:
300 * get_vf_config
301 * get_vf_guid
302 * get_vf_stats
303 * set_vf_guid
304 * set_vf_link_state
305 */
306 IBK_VIRTUAL_FUNCTION = 1 << 7,
307 /* ipoib will use IB_QP_CREATE_NETDEV_USE for its QPs */
308 IBK_RDMA_NETDEV_OPA = 1 << 8,
309};
310
311enum ib_atomic_cap {
312 IB_ATOMIC_NONE,
313 IB_ATOMIC_HCA,
314 IB_ATOMIC_GLOB
315};
316
317enum ib_odp_general_cap_bits {
318 IB_ODP_SUPPORT = IB_UVERBS_ODP_SUPPORT,
319 IB_ODP_SUPPORT_IMPLICIT = IB_UVERBS_ODP_SUPPORT_IMPLICIT,
320};
321
322enum ib_odp_transport_cap_bits {
323 IB_ODP_SUPPORT_SEND = IB_UVERBS_ODP_SUPPORT_SEND,
324 IB_ODP_SUPPORT_RECV = IB_UVERBS_ODP_SUPPORT_RECV,
325 IB_ODP_SUPPORT_WRITE = IB_UVERBS_ODP_SUPPORT_WRITE,
326 IB_ODP_SUPPORT_READ = IB_UVERBS_ODP_SUPPORT_READ,
327 IB_ODP_SUPPORT_ATOMIC = IB_UVERBS_ODP_SUPPORT_ATOMIC,
328 IB_ODP_SUPPORT_SRQ_RECV = IB_UVERBS_ODP_SUPPORT_SRQ_RECV,
329 IB_ODP_SUPPORT_FLUSH = IB_UVERBS_ODP_SUPPORT_FLUSH,
330 IB_ODP_SUPPORT_ATOMIC_WRITE = IB_UVERBS_ODP_SUPPORT_ATOMIC_WRITE,
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(mtu: (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
525static inline const char *__attribute_const__
526ib_port_state_to_str(enum ib_port_state state)
527{
528 const char * const states[] = {
529 [IB_PORT_NOP] = "NOP",
530 [IB_PORT_DOWN] = "DOWN",
531 [IB_PORT_INIT] = "INIT",
532 [IB_PORT_ARMED] = "ARMED",
533 [IB_PORT_ACTIVE] = "ACTIVE",
534 [IB_PORT_ACTIVE_DEFER] = "ACTIVE_DEFER",
535 };
536
537 if (state < ARRAY_SIZE(states))
538 return states[state];
539 return "UNKNOWN";
540}
541
542enum ib_port_phys_state {
543 IB_PORT_PHYS_STATE_SLEEP = 1,
544 IB_PORT_PHYS_STATE_POLLING = 2,
545 IB_PORT_PHYS_STATE_DISABLED = 3,
546 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
547 IB_PORT_PHYS_STATE_LINK_UP = 5,
548 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
549 IB_PORT_PHYS_STATE_PHY_TEST = 7,
550};
551
552enum ib_port_width {
553 IB_WIDTH_1X = 1,
554 IB_WIDTH_2X = 16,
555 IB_WIDTH_4X = 2,
556 IB_WIDTH_8X = 4,
557 IB_WIDTH_12X = 8
558};
559
560static inline int ib_width_enum_to_int(enum ib_port_width width)
561{
562 switch (width) {
563 case IB_WIDTH_1X: return 1;
564 case IB_WIDTH_2X: return 2;
565 case IB_WIDTH_4X: return 4;
566 case IB_WIDTH_8X: return 8;
567 case IB_WIDTH_12X: return 12;
568 default: return -1;
569 }
570}
571
572enum ib_port_speed {
573 IB_SPEED_SDR = 1,
574 IB_SPEED_DDR = 2,
575 IB_SPEED_QDR = 4,
576 IB_SPEED_FDR10 = 8,
577 IB_SPEED_FDR = 16,
578 IB_SPEED_EDR = 32,
579 IB_SPEED_HDR = 64,
580 IB_SPEED_NDR = 128,
581 IB_SPEED_XDR = 256,
582};
583
584enum ib_stat_flag {
585 IB_STAT_FLAG_OPTIONAL = 1 << 0,
586};
587
588/**
589 * struct rdma_stat_desc
590 * @name - The name of the counter
591 * @flags - Flags of the counter; For example, IB_STAT_FLAG_OPTIONAL
592 * @priv - Driver private information; Core code should not use
593 */
594struct rdma_stat_desc {
595 const char *name;
596 unsigned int flags;
597 const void *priv;
598};
599
600/**
601 * struct rdma_hw_stats
602 * @lock - Mutex to protect parallel write access to lifespan and values
603 * of counters, which are 64bits and not guaranteed to be written
604 * atomicaly on 32bits systems.
605 * @timestamp - Used by the core code to track when the last update was
606 * @lifespan - Used by the core code to determine how old the counters
607 * should be before being updated again. Stored in jiffies, defaults
608 * to 10 milliseconds, drivers can override the default be specifying
609 * their own value during their allocation routine.
610 * @descs - Array of pointers to static descriptors used for the counters
611 * in directory.
612 * @is_disabled - A bitmap to indicate each counter is currently disabled
613 * or not.
614 * @num_counters - How many hardware counters there are. If name is
615 * shorter than this number, a kernel oops will result. Driver authors
616 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
617 * in their code to prevent this.
618 * @value - Array of u64 counters that are accessed by the sysfs code and
619 * filled in by the drivers get_stats routine
620 */
621struct rdma_hw_stats {
622 struct mutex lock; /* Protect lifespan and values[] */
623 unsigned long timestamp;
624 unsigned long lifespan;
625 const struct rdma_stat_desc *descs;
626 unsigned long *is_disabled;
627 int num_counters;
628 u64 value[] __counted_by(num_counters);
629};
630
631#define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
632
633struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
634 const struct rdma_stat_desc *descs, int num_counters,
635 unsigned long lifespan);
636
637void rdma_free_hw_stats_struct(struct rdma_hw_stats *stats);
638
639/* Define bits for the various functionality this port needs to be supported by
640 * the core.
641 */
642/* Management 0x00000FFF */
643#define RDMA_CORE_CAP_IB_MAD 0x00000001
644#define RDMA_CORE_CAP_IB_SMI 0x00000002
645#define RDMA_CORE_CAP_IB_CM 0x00000004
646#define RDMA_CORE_CAP_IW_CM 0x00000008
647#define RDMA_CORE_CAP_IB_SA 0x00000010
648#define RDMA_CORE_CAP_OPA_MAD 0x00000020
649
650/* Address format 0x000FF000 */
651#define RDMA_CORE_CAP_AF_IB 0x00001000
652#define RDMA_CORE_CAP_ETH_AH 0x00002000
653#define RDMA_CORE_CAP_OPA_AH 0x00004000
654#define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
655
656/* Protocol 0xFFF00000 */
657#define RDMA_CORE_CAP_PROT_IB 0x00100000
658#define RDMA_CORE_CAP_PROT_ROCE 0x00200000
659#define RDMA_CORE_CAP_PROT_IWARP 0x00400000
660#define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
661#define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
662#define RDMA_CORE_CAP_PROT_USNIC 0x02000000
663
664#define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
665 | RDMA_CORE_CAP_PROT_ROCE \
666 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
667
668#define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
669 | RDMA_CORE_CAP_IB_MAD \
670 | RDMA_CORE_CAP_IB_SMI \
671 | RDMA_CORE_CAP_IB_CM \
672 | RDMA_CORE_CAP_IB_SA \
673 | RDMA_CORE_CAP_AF_IB)
674#define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
675 | RDMA_CORE_CAP_IB_MAD \
676 | RDMA_CORE_CAP_IB_CM \
677 | RDMA_CORE_CAP_AF_IB \
678 | RDMA_CORE_CAP_ETH_AH)
679#define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
680 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
681 | RDMA_CORE_CAP_IB_MAD \
682 | RDMA_CORE_CAP_IB_CM \
683 | RDMA_CORE_CAP_AF_IB \
684 | RDMA_CORE_CAP_ETH_AH)
685#define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
686 | RDMA_CORE_CAP_IW_CM)
687#define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
688 | RDMA_CORE_CAP_OPA_MAD)
689
690#define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
691
692#define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
693
694struct ib_port_attr {
695 u64 subnet_prefix;
696 enum ib_port_state state;
697 enum ib_mtu max_mtu;
698 enum ib_mtu active_mtu;
699 u32 phys_mtu;
700 int gid_tbl_len;
701 unsigned int ip_gids:1;
702 /* This is the value from PortInfo CapabilityMask, defined by IBA */
703 u32 port_cap_flags;
704 u32 max_msg_sz;
705 u32 bad_pkey_cntr;
706 u32 qkey_viol_cntr;
707 u16 pkey_tbl_len;
708 u32 sm_lid;
709 u32 lid;
710 u8 lmc;
711 u8 max_vl_num;
712 u8 sm_sl;
713 u8 subnet_timeout;
714 u8 init_type_reply;
715 u8 active_width;
716 u16 active_speed;
717 u8 phys_state;
718 u16 port_cap_flags2;
719};
720
721enum ib_device_modify_flags {
722 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
723 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
724};
725
726#define IB_DEVICE_NODE_DESC_MAX 64
727
728struct ib_device_modify {
729 u64 sys_image_guid;
730 char node_desc[IB_DEVICE_NODE_DESC_MAX];
731};
732
733enum ib_port_modify_flags {
734 IB_PORT_SHUTDOWN = 1,
735 IB_PORT_INIT_TYPE = (1<<2),
736 IB_PORT_RESET_QKEY_CNTR = (1<<3),
737 IB_PORT_OPA_MASK_CHG = (1<<4)
738};
739
740struct ib_port_modify {
741 u32 set_port_cap_mask;
742 u32 clr_port_cap_mask;
743 u8 init_type;
744};
745
746enum ib_event_type {
747 IB_EVENT_CQ_ERR,
748 IB_EVENT_QP_FATAL,
749 IB_EVENT_QP_REQ_ERR,
750 IB_EVENT_QP_ACCESS_ERR,
751 IB_EVENT_COMM_EST,
752 IB_EVENT_SQ_DRAINED,
753 IB_EVENT_PATH_MIG,
754 IB_EVENT_PATH_MIG_ERR,
755 IB_EVENT_DEVICE_FATAL,
756 IB_EVENT_PORT_ACTIVE,
757 IB_EVENT_PORT_ERR,
758 IB_EVENT_LID_CHANGE,
759 IB_EVENT_PKEY_CHANGE,
760 IB_EVENT_SM_CHANGE,
761 IB_EVENT_SRQ_ERR,
762 IB_EVENT_SRQ_LIMIT_REACHED,
763 IB_EVENT_QP_LAST_WQE_REACHED,
764 IB_EVENT_CLIENT_REREGISTER,
765 IB_EVENT_GID_CHANGE,
766 IB_EVENT_WQ_FATAL,
767};
768
769const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
770
771struct ib_event {
772 struct ib_device *device;
773 union {
774 struct ib_cq *cq;
775 struct ib_qp *qp;
776 struct ib_srq *srq;
777 struct ib_wq *wq;
778 u32 port_num;
779 } element;
780 enum ib_event_type event;
781};
782
783struct ib_event_handler {
784 struct ib_device *device;
785 void (*handler)(struct ib_event_handler *, struct ib_event *);
786 struct list_head list;
787};
788
789#define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
790 do { \
791 (_ptr)->device = _device; \
792 (_ptr)->handler = _handler; \
793 INIT_LIST_HEAD(&(_ptr)->list); \
794 } while (0)
795
796struct ib_global_route {
797 const struct ib_gid_attr *sgid_attr;
798 union ib_gid dgid;
799 u32 flow_label;
800 u8 sgid_index;
801 u8 hop_limit;
802 u8 traffic_class;
803};
804
805struct ib_grh {
806 __be32 version_tclass_flow;
807 __be16 paylen;
808 u8 next_hdr;
809 u8 hop_limit;
810 union ib_gid sgid;
811 union ib_gid dgid;
812};
813
814union rdma_network_hdr {
815 struct ib_grh ibgrh;
816 struct {
817 /* The IB spec states that if it's IPv4, the header
818 * is located in the last 20 bytes of the header.
819 */
820 u8 reserved[20];
821 struct iphdr roce4grh;
822 };
823};
824
825#define IB_QPN_MASK 0xFFFFFF
826
827enum {
828 IB_MULTICAST_QPN = 0xffffff
829};
830
831#define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
832#define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
833
834enum ib_ah_flags {
835 IB_AH_GRH = 1
836};
837
838enum ib_rate {
839 IB_RATE_PORT_CURRENT = 0,
840 IB_RATE_2_5_GBPS = 2,
841 IB_RATE_5_GBPS = 5,
842 IB_RATE_10_GBPS = 3,
843 IB_RATE_20_GBPS = 6,
844 IB_RATE_30_GBPS = 4,
845 IB_RATE_40_GBPS = 7,
846 IB_RATE_60_GBPS = 8,
847 IB_RATE_80_GBPS = 9,
848 IB_RATE_120_GBPS = 10,
849 IB_RATE_14_GBPS = 11,
850 IB_RATE_56_GBPS = 12,
851 IB_RATE_112_GBPS = 13,
852 IB_RATE_168_GBPS = 14,
853 IB_RATE_25_GBPS = 15,
854 IB_RATE_100_GBPS = 16,
855 IB_RATE_200_GBPS = 17,
856 IB_RATE_300_GBPS = 18,
857 IB_RATE_28_GBPS = 19,
858 IB_RATE_50_GBPS = 20,
859 IB_RATE_400_GBPS = 21,
860 IB_RATE_600_GBPS = 22,
861 IB_RATE_800_GBPS = 23,
862};
863
864/**
865 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
866 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
867 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
868 * @rate: rate to convert.
869 */
870__attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
871
872/**
873 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
874 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
875 * @rate: rate to convert.
876 */
877__attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
878
879
880/**
881 * enum ib_mr_type - memory region type
882 * @IB_MR_TYPE_MEM_REG: memory region that is used for
883 * normal registration
884 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
885 * register any arbitrary sg lists (without
886 * the normal mr constraints - see
887 * ib_map_mr_sg)
888 * @IB_MR_TYPE_DM: memory region that is used for device
889 * memory registration
890 * @IB_MR_TYPE_USER: memory region that is used for the user-space
891 * application
892 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
893 * without address translations (VA=PA)
894 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
895 * data integrity operations
896 */
897enum ib_mr_type {
898 IB_MR_TYPE_MEM_REG,
899 IB_MR_TYPE_SG_GAPS,
900 IB_MR_TYPE_DM,
901 IB_MR_TYPE_USER,
902 IB_MR_TYPE_DMA,
903 IB_MR_TYPE_INTEGRITY,
904};
905
906enum ib_mr_status_check {
907 IB_MR_CHECK_SIG_STATUS = 1,
908};
909
910/**
911 * struct ib_mr_status - Memory region status container
912 *
913 * @fail_status: Bitmask of MR checks status. For each
914 * failed check a corresponding status bit is set.
915 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
916 * failure.
917 */
918struct ib_mr_status {
919 u32 fail_status;
920 struct ib_sig_err sig_err;
921};
922
923/**
924 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
925 * enum.
926 * @mult: multiple to convert.
927 */
928__attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
929
930struct rdma_ah_init_attr {
931 struct rdma_ah_attr *ah_attr;
932 u32 flags;
933 struct net_device *xmit_slave;
934};
935
936enum rdma_ah_attr_type {
937 RDMA_AH_ATTR_TYPE_UNDEFINED,
938 RDMA_AH_ATTR_TYPE_IB,
939 RDMA_AH_ATTR_TYPE_ROCE,
940 RDMA_AH_ATTR_TYPE_OPA,
941};
942
943struct ib_ah_attr {
944 u16 dlid;
945 u8 src_path_bits;
946};
947
948struct roce_ah_attr {
949 u8 dmac[ETH_ALEN];
950};
951
952struct opa_ah_attr {
953 u32 dlid;
954 u8 src_path_bits;
955 bool make_grd;
956};
957
958struct rdma_ah_attr {
959 struct ib_global_route grh;
960 u8 sl;
961 u8 static_rate;
962 u32 port_num;
963 u8 ah_flags;
964 enum rdma_ah_attr_type type;
965 union {
966 struct ib_ah_attr ib;
967 struct roce_ah_attr roce;
968 struct opa_ah_attr opa;
969 };
970};
971
972enum ib_wc_status {
973 IB_WC_SUCCESS,
974 IB_WC_LOC_LEN_ERR,
975 IB_WC_LOC_QP_OP_ERR,
976 IB_WC_LOC_EEC_OP_ERR,
977 IB_WC_LOC_PROT_ERR,
978 IB_WC_WR_FLUSH_ERR,
979 IB_WC_MW_BIND_ERR,
980 IB_WC_BAD_RESP_ERR,
981 IB_WC_LOC_ACCESS_ERR,
982 IB_WC_REM_INV_REQ_ERR,
983 IB_WC_REM_ACCESS_ERR,
984 IB_WC_REM_OP_ERR,
985 IB_WC_RETRY_EXC_ERR,
986 IB_WC_RNR_RETRY_EXC_ERR,
987 IB_WC_LOC_RDD_VIOL_ERR,
988 IB_WC_REM_INV_RD_REQ_ERR,
989 IB_WC_REM_ABORT_ERR,
990 IB_WC_INV_EECN_ERR,
991 IB_WC_INV_EEC_STATE_ERR,
992 IB_WC_FATAL_ERR,
993 IB_WC_RESP_TIMEOUT_ERR,
994 IB_WC_GENERAL_ERR
995};
996
997const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
998
999enum ib_wc_opcode {
1000 IB_WC_SEND = IB_UVERBS_WC_SEND,
1001 IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE,
1002 IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ,
1003 IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP,
1004 IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD,
1005 IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW,
1006 IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV,
1007 IB_WC_LSO = IB_UVERBS_WC_TSO,
1008 IB_WC_ATOMIC_WRITE = IB_UVERBS_WC_ATOMIC_WRITE,
1009 IB_WC_REG_MR,
1010 IB_WC_MASKED_COMP_SWAP,
1011 IB_WC_MASKED_FETCH_ADD,
1012 IB_WC_FLUSH = IB_UVERBS_WC_FLUSH,
1013/*
1014 * Set value of IB_WC_RECV so consumers can test if a completion is a
1015 * receive by testing (opcode & IB_WC_RECV).
1016 */
1017 IB_WC_RECV = 1 << 7,
1018 IB_WC_RECV_RDMA_WITH_IMM
1019};
1020
1021enum ib_wc_flags {
1022 IB_WC_GRH = 1,
1023 IB_WC_WITH_IMM = (1<<1),
1024 IB_WC_WITH_INVALIDATE = (1<<2),
1025 IB_WC_IP_CSUM_OK = (1<<3),
1026 IB_WC_WITH_SMAC = (1<<4),
1027 IB_WC_WITH_VLAN = (1<<5),
1028 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
1029};
1030
1031struct ib_wc {
1032 union {
1033 u64 wr_id;
1034 struct ib_cqe *wr_cqe;
1035 };
1036 enum ib_wc_status status;
1037 enum ib_wc_opcode opcode;
1038 u32 vendor_err;
1039 u32 byte_len;
1040 struct ib_qp *qp;
1041 union {
1042 __be32 imm_data;
1043 u32 invalidate_rkey;
1044 } ex;
1045 u32 src_qp;
1046 u32 slid;
1047 int wc_flags;
1048 u16 pkey_index;
1049 u8 sl;
1050 u8 dlid_path_bits;
1051 u32 port_num; /* valid only for DR SMPs on switches */
1052 u8 smac[ETH_ALEN];
1053 u16 vlan_id;
1054 u8 network_hdr_type;
1055};
1056
1057enum ib_cq_notify_flags {
1058 IB_CQ_SOLICITED = 1 << 0,
1059 IB_CQ_NEXT_COMP = 1 << 1,
1060 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1061 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1062};
1063
1064enum ib_srq_type {
1065 IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1066 IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1067 IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1068};
1069
1070static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1071{
1072 return srq_type == IB_SRQT_XRC ||
1073 srq_type == IB_SRQT_TM;
1074}
1075
1076enum ib_srq_attr_mask {
1077 IB_SRQ_MAX_WR = 1 << 0,
1078 IB_SRQ_LIMIT = 1 << 1,
1079};
1080
1081struct ib_srq_attr {
1082 u32 max_wr;
1083 u32 max_sge;
1084 u32 srq_limit;
1085};
1086
1087struct ib_srq_init_attr {
1088 void (*event_handler)(struct ib_event *, void *);
1089 void *srq_context;
1090 struct ib_srq_attr attr;
1091 enum ib_srq_type srq_type;
1092
1093 struct {
1094 struct ib_cq *cq;
1095 union {
1096 struct {
1097 struct ib_xrcd *xrcd;
1098 } xrc;
1099
1100 struct {
1101 u32 max_num_tags;
1102 } tag_matching;
1103 };
1104 } ext;
1105};
1106
1107struct ib_qp_cap {
1108 u32 max_send_wr;
1109 u32 max_recv_wr;
1110 u32 max_send_sge;
1111 u32 max_recv_sge;
1112 u32 max_inline_data;
1113
1114 /*
1115 * Maximum number of rdma_rw_ctx structures in flight at a time.
1116 * ib_create_qp() will calculate the right amount of needed WRs
1117 * and MRs based on this.
1118 */
1119 u32 max_rdma_ctxs;
1120};
1121
1122enum ib_sig_type {
1123 IB_SIGNAL_ALL_WR,
1124 IB_SIGNAL_REQ_WR
1125};
1126
1127enum ib_qp_type {
1128 /*
1129 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1130 * here (and in that order) since the MAD layer uses them as
1131 * indices into a 2-entry table.
1132 */
1133 IB_QPT_SMI,
1134 IB_QPT_GSI,
1135
1136 IB_QPT_RC = IB_UVERBS_QPT_RC,
1137 IB_QPT_UC = IB_UVERBS_QPT_UC,
1138 IB_QPT_UD = IB_UVERBS_QPT_UD,
1139 IB_QPT_RAW_IPV6,
1140 IB_QPT_RAW_ETHERTYPE,
1141 IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1142 IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1143 IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1144 IB_QPT_MAX,
1145 IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1146 /* Reserve a range for qp types internal to the low level driver.
1147 * These qp types will not be visible at the IB core layer, so the
1148 * IB_QPT_MAX usages should not be affected in the core layer
1149 */
1150 IB_QPT_RESERVED1 = 0x1000,
1151 IB_QPT_RESERVED2,
1152 IB_QPT_RESERVED3,
1153 IB_QPT_RESERVED4,
1154 IB_QPT_RESERVED5,
1155 IB_QPT_RESERVED6,
1156 IB_QPT_RESERVED7,
1157 IB_QPT_RESERVED8,
1158 IB_QPT_RESERVED9,
1159 IB_QPT_RESERVED10,
1160};
1161
1162enum ib_qp_create_flags {
1163 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1164 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK =
1165 IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1166 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1167 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1168 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1169 IB_QP_CREATE_NETIF_QP = 1 << 5,
1170 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
1171 IB_QP_CREATE_NETDEV_USE = 1 << 7,
1172 IB_QP_CREATE_SCATTER_FCS =
1173 IB_UVERBS_QP_CREATE_SCATTER_FCS,
1174 IB_QP_CREATE_CVLAN_STRIPPING =
1175 IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1176 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1177 IB_QP_CREATE_PCI_WRITE_END_PADDING =
1178 IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1179 /* reserve bits 26-31 for low level drivers' internal use */
1180 IB_QP_CREATE_RESERVED_START = 1 << 26,
1181 IB_QP_CREATE_RESERVED_END = 1 << 31,
1182};
1183
1184/*
1185 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1186 * callback to destroy the passed in QP.
1187 */
1188
1189struct ib_qp_init_attr {
1190 /* This callback occurs in workqueue context */
1191 void (*event_handler)(struct ib_event *, void *);
1192
1193 void *qp_context;
1194 struct ib_cq *send_cq;
1195 struct ib_cq *recv_cq;
1196 struct ib_srq *srq;
1197 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1198 struct ib_qp_cap cap;
1199 enum ib_sig_type sq_sig_type;
1200 enum ib_qp_type qp_type;
1201 u32 create_flags;
1202
1203 /*
1204 * Only needed for special QP types, or when using the RW API.
1205 */
1206 u32 port_num;
1207 struct ib_rwq_ind_table *rwq_ind_tbl;
1208 u32 source_qpn;
1209};
1210
1211struct ib_qp_open_attr {
1212 void (*event_handler)(struct ib_event *, void *);
1213 void *qp_context;
1214 u32 qp_num;
1215 enum ib_qp_type qp_type;
1216};
1217
1218enum ib_rnr_timeout {
1219 IB_RNR_TIMER_655_36 = 0,
1220 IB_RNR_TIMER_000_01 = 1,
1221 IB_RNR_TIMER_000_02 = 2,
1222 IB_RNR_TIMER_000_03 = 3,
1223 IB_RNR_TIMER_000_04 = 4,
1224 IB_RNR_TIMER_000_06 = 5,
1225 IB_RNR_TIMER_000_08 = 6,
1226 IB_RNR_TIMER_000_12 = 7,
1227 IB_RNR_TIMER_000_16 = 8,
1228 IB_RNR_TIMER_000_24 = 9,
1229 IB_RNR_TIMER_000_32 = 10,
1230 IB_RNR_TIMER_000_48 = 11,
1231 IB_RNR_TIMER_000_64 = 12,
1232 IB_RNR_TIMER_000_96 = 13,
1233 IB_RNR_TIMER_001_28 = 14,
1234 IB_RNR_TIMER_001_92 = 15,
1235 IB_RNR_TIMER_002_56 = 16,
1236 IB_RNR_TIMER_003_84 = 17,
1237 IB_RNR_TIMER_005_12 = 18,
1238 IB_RNR_TIMER_007_68 = 19,
1239 IB_RNR_TIMER_010_24 = 20,
1240 IB_RNR_TIMER_015_36 = 21,
1241 IB_RNR_TIMER_020_48 = 22,
1242 IB_RNR_TIMER_030_72 = 23,
1243 IB_RNR_TIMER_040_96 = 24,
1244 IB_RNR_TIMER_061_44 = 25,
1245 IB_RNR_TIMER_081_92 = 26,
1246 IB_RNR_TIMER_122_88 = 27,
1247 IB_RNR_TIMER_163_84 = 28,
1248 IB_RNR_TIMER_245_76 = 29,
1249 IB_RNR_TIMER_327_68 = 30,
1250 IB_RNR_TIMER_491_52 = 31
1251};
1252
1253enum ib_qp_attr_mask {
1254 IB_QP_STATE = 1,
1255 IB_QP_CUR_STATE = (1<<1),
1256 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1257 IB_QP_ACCESS_FLAGS = (1<<3),
1258 IB_QP_PKEY_INDEX = (1<<4),
1259 IB_QP_PORT = (1<<5),
1260 IB_QP_QKEY = (1<<6),
1261 IB_QP_AV = (1<<7),
1262 IB_QP_PATH_MTU = (1<<8),
1263 IB_QP_TIMEOUT = (1<<9),
1264 IB_QP_RETRY_CNT = (1<<10),
1265 IB_QP_RNR_RETRY = (1<<11),
1266 IB_QP_RQ_PSN = (1<<12),
1267 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1268 IB_QP_ALT_PATH = (1<<14),
1269 IB_QP_MIN_RNR_TIMER = (1<<15),
1270 IB_QP_SQ_PSN = (1<<16),
1271 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1272 IB_QP_PATH_MIG_STATE = (1<<18),
1273 IB_QP_CAP = (1<<19),
1274 IB_QP_DEST_QPN = (1<<20),
1275 IB_QP_RESERVED1 = (1<<21),
1276 IB_QP_RESERVED2 = (1<<22),
1277 IB_QP_RESERVED3 = (1<<23),
1278 IB_QP_RESERVED4 = (1<<24),
1279 IB_QP_RATE_LIMIT = (1<<25),
1280
1281 IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0),
1282};
1283
1284enum ib_qp_state {
1285 IB_QPS_RESET,
1286 IB_QPS_INIT,
1287 IB_QPS_RTR,
1288 IB_QPS_RTS,
1289 IB_QPS_SQD,
1290 IB_QPS_SQE,
1291 IB_QPS_ERR
1292};
1293
1294enum ib_mig_state {
1295 IB_MIG_MIGRATED,
1296 IB_MIG_REARM,
1297 IB_MIG_ARMED
1298};
1299
1300enum ib_mw_type {
1301 IB_MW_TYPE_1 = 1,
1302 IB_MW_TYPE_2 = 2
1303};
1304
1305struct ib_qp_attr {
1306 enum ib_qp_state qp_state;
1307 enum ib_qp_state cur_qp_state;
1308 enum ib_mtu path_mtu;
1309 enum ib_mig_state path_mig_state;
1310 u32 qkey;
1311 u32 rq_psn;
1312 u32 sq_psn;
1313 u32 dest_qp_num;
1314 int qp_access_flags;
1315 struct ib_qp_cap cap;
1316 struct rdma_ah_attr ah_attr;
1317 struct rdma_ah_attr alt_ah_attr;
1318 u16 pkey_index;
1319 u16 alt_pkey_index;
1320 u8 en_sqd_async_notify;
1321 u8 sq_draining;
1322 u8 max_rd_atomic;
1323 u8 max_dest_rd_atomic;
1324 u8 min_rnr_timer;
1325 u32 port_num;
1326 u8 timeout;
1327 u8 retry_cnt;
1328 u8 rnr_retry;
1329 u32 alt_port_num;
1330 u8 alt_timeout;
1331 u32 rate_limit;
1332 struct net_device *xmit_slave;
1333};
1334
1335enum ib_wr_opcode {
1336 /* These are shared with userspace */
1337 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1338 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1339 IB_WR_SEND = IB_UVERBS_WR_SEND,
1340 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1341 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1342 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1343 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1344 IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW,
1345 IB_WR_LSO = IB_UVERBS_WR_TSO,
1346 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1347 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1348 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1349 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1350 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1351 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1352 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1353 IB_WR_FLUSH = IB_UVERBS_WR_FLUSH,
1354 IB_WR_ATOMIC_WRITE = IB_UVERBS_WR_ATOMIC_WRITE,
1355
1356 /* These are kernel only and can not be issued by userspace */
1357 IB_WR_REG_MR = 0x20,
1358 IB_WR_REG_MR_INTEGRITY,
1359
1360 /* reserve values for low level drivers' internal use.
1361 * These values will not be used at all in the ib core layer.
1362 */
1363 IB_WR_RESERVED1 = 0xf0,
1364 IB_WR_RESERVED2,
1365 IB_WR_RESERVED3,
1366 IB_WR_RESERVED4,
1367 IB_WR_RESERVED5,
1368 IB_WR_RESERVED6,
1369 IB_WR_RESERVED7,
1370 IB_WR_RESERVED8,
1371 IB_WR_RESERVED9,
1372 IB_WR_RESERVED10,
1373};
1374
1375enum ib_send_flags {
1376 IB_SEND_FENCE = 1,
1377 IB_SEND_SIGNALED = (1<<1),
1378 IB_SEND_SOLICITED = (1<<2),
1379 IB_SEND_INLINE = (1<<3),
1380 IB_SEND_IP_CSUM = (1<<4),
1381
1382 /* reserve bits 26-31 for low level drivers' internal use */
1383 IB_SEND_RESERVED_START = (1 << 26),
1384 IB_SEND_RESERVED_END = (1 << 31),
1385};
1386
1387struct ib_sge {
1388 u64 addr;
1389 u32 length;
1390 u32 lkey;
1391};
1392
1393struct ib_cqe {
1394 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1395};
1396
1397struct ib_send_wr {
1398 struct ib_send_wr *next;
1399 union {
1400 u64 wr_id;
1401 struct ib_cqe *wr_cqe;
1402 };
1403 struct ib_sge *sg_list;
1404 int num_sge;
1405 enum ib_wr_opcode opcode;
1406 int send_flags;
1407 union {
1408 __be32 imm_data;
1409 u32 invalidate_rkey;
1410 } ex;
1411};
1412
1413struct ib_rdma_wr {
1414 struct ib_send_wr wr;
1415 u64 remote_addr;
1416 u32 rkey;
1417};
1418
1419static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1420{
1421 return container_of(wr, struct ib_rdma_wr, wr);
1422}
1423
1424struct ib_atomic_wr {
1425 struct ib_send_wr wr;
1426 u64 remote_addr;
1427 u64 compare_add;
1428 u64 swap;
1429 u64 compare_add_mask;
1430 u64 swap_mask;
1431 u32 rkey;
1432};
1433
1434static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1435{
1436 return container_of(wr, struct ib_atomic_wr, wr);
1437}
1438
1439struct ib_ud_wr {
1440 struct ib_send_wr wr;
1441 struct ib_ah *ah;
1442 void *header;
1443 int hlen;
1444 int mss;
1445 u32 remote_qpn;
1446 u32 remote_qkey;
1447 u16 pkey_index; /* valid for GSI only */
1448 u32 port_num; /* valid for DR SMPs on switch only */
1449};
1450
1451static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1452{
1453 return container_of(wr, struct ib_ud_wr, wr);
1454}
1455
1456struct ib_reg_wr {
1457 struct ib_send_wr wr;
1458 struct ib_mr *mr;
1459 u32 key;
1460 int access;
1461};
1462
1463static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1464{
1465 return container_of(wr, struct ib_reg_wr, wr);
1466}
1467
1468struct ib_recv_wr {
1469 struct ib_recv_wr *next;
1470 union {
1471 u64 wr_id;
1472 struct ib_cqe *wr_cqe;
1473 };
1474 struct ib_sge *sg_list;
1475 int num_sge;
1476};
1477
1478enum ib_access_flags {
1479 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1480 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1481 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1482 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1483 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1484 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1485 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1486 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1487 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1488 IB_ACCESS_FLUSH_GLOBAL = IB_UVERBS_ACCESS_FLUSH_GLOBAL,
1489 IB_ACCESS_FLUSH_PERSISTENT = IB_UVERBS_ACCESS_FLUSH_PERSISTENT,
1490
1491 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1492 IB_ACCESS_SUPPORTED =
1493 ((IB_ACCESS_FLUSH_PERSISTENT << 1) - 1) | IB_ACCESS_OPTIONAL,
1494};
1495
1496/*
1497 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1498 * are hidden here instead of a uapi header!
1499 */
1500enum ib_mr_rereg_flags {
1501 IB_MR_REREG_TRANS = 1,
1502 IB_MR_REREG_PD = (1<<1),
1503 IB_MR_REREG_ACCESS = (1<<2),
1504 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1505};
1506
1507struct ib_umem;
1508
1509enum rdma_remove_reason {
1510 /*
1511 * Userspace requested uobject deletion or initial try
1512 * to remove uobject via cleanup. Call could fail
1513 */
1514 RDMA_REMOVE_DESTROY,
1515 /* Context deletion. This call should delete the actual object itself */
1516 RDMA_REMOVE_CLOSE,
1517 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1518 RDMA_REMOVE_DRIVER_REMOVE,
1519 /* uobj is being cleaned-up before being committed */
1520 RDMA_REMOVE_ABORT,
1521 /* The driver failed to destroy the uobject and is being disconnected */
1522 RDMA_REMOVE_DRIVER_FAILURE,
1523};
1524
1525struct ib_rdmacg_object {
1526#ifdef CONFIG_CGROUP_RDMA
1527 struct rdma_cgroup *cg; /* owner rdma cgroup */
1528#endif
1529};
1530
1531struct ib_ucontext {
1532 struct ib_device *device;
1533 struct ib_uverbs_file *ufile;
1534
1535 struct ib_rdmacg_object cg_obj;
1536 u64 enabled_caps;
1537 /*
1538 * Implementation details of the RDMA core, don't use in drivers:
1539 */
1540 struct rdma_restrack_entry res;
1541 struct xarray mmap_xa;
1542};
1543
1544struct ib_uobject {
1545 u64 user_handle; /* handle given to us by userspace */
1546 /* ufile & ucontext owning this object */
1547 struct ib_uverbs_file *ufile;
1548 /* FIXME, save memory: ufile->context == context */
1549 struct ib_ucontext *context; /* associated user context */
1550 void *object; /* containing object */
1551 struct list_head list; /* link to context's list */
1552 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1553 int id; /* index into kernel idr */
1554 struct kref ref;
1555 atomic_t usecnt; /* protects exclusive access */
1556 struct rcu_head rcu; /* kfree_rcu() overhead */
1557
1558 const struct uverbs_api_object *uapi_object;
1559};
1560
1561struct ib_udata {
1562 const void __user *inbuf;
1563 void __user *outbuf;
1564 size_t inlen;
1565 size_t outlen;
1566};
1567
1568struct ib_pd {
1569 u32 local_dma_lkey;
1570 u32 flags;
1571 struct ib_device *device;
1572 struct ib_uobject *uobject;
1573 atomic_t usecnt; /* count all resources */
1574
1575 u32 unsafe_global_rkey;
1576
1577 /*
1578 * Implementation details of the RDMA core, don't use in drivers:
1579 */
1580 struct ib_mr *__internal_mr;
1581 struct rdma_restrack_entry res;
1582};
1583
1584struct ib_xrcd {
1585 struct ib_device *device;
1586 atomic_t usecnt; /* count all exposed resources */
1587 struct inode *inode;
1588 struct rw_semaphore tgt_qps_rwsem;
1589 struct xarray tgt_qps;
1590};
1591
1592struct ib_ah {
1593 struct ib_device *device;
1594 struct ib_pd *pd;
1595 struct ib_uobject *uobject;
1596 const struct ib_gid_attr *sgid_attr;
1597 enum rdma_ah_attr_type type;
1598};
1599
1600typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1601
1602enum ib_poll_context {
1603 IB_POLL_SOFTIRQ, /* poll from softirq context */
1604 IB_POLL_WORKQUEUE, /* poll from workqueue */
1605 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1606 IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1607
1608 IB_POLL_DIRECT, /* caller context, no hw completions */
1609};
1610
1611struct ib_cq {
1612 struct ib_device *device;
1613 struct ib_ucq_object *uobject;
1614 ib_comp_handler comp_handler;
1615 void (*event_handler)(struct ib_event *, void *);
1616 void *cq_context;
1617 int cqe;
1618 unsigned int cqe_used;
1619 atomic_t usecnt; /* count number of work queues */
1620 enum ib_poll_context poll_ctx;
1621 struct ib_wc *wc;
1622 struct list_head pool_entry;
1623 union {
1624 struct irq_poll iop;
1625 struct work_struct work;
1626 };
1627 struct workqueue_struct *comp_wq;
1628 struct dim *dim;
1629
1630 /* updated only by trace points */
1631 ktime_t timestamp;
1632 u8 interrupt:1;
1633 u8 shared:1;
1634 unsigned int comp_vector;
1635
1636 /*
1637 * Implementation details of the RDMA core, don't use in drivers:
1638 */
1639 struct rdma_restrack_entry res;
1640};
1641
1642struct ib_srq {
1643 struct ib_device *device;
1644 struct ib_pd *pd;
1645 struct ib_usrq_object *uobject;
1646 void (*event_handler)(struct ib_event *, void *);
1647 void *srq_context;
1648 enum ib_srq_type srq_type;
1649 atomic_t usecnt;
1650
1651 struct {
1652 struct ib_cq *cq;
1653 union {
1654 struct {
1655 struct ib_xrcd *xrcd;
1656 u32 srq_num;
1657 } xrc;
1658 };
1659 } ext;
1660
1661 /*
1662 * Implementation details of the RDMA core, don't use in drivers:
1663 */
1664 struct rdma_restrack_entry res;
1665};
1666
1667enum ib_raw_packet_caps {
1668 /*
1669 * Strip cvlan from incoming packet and report it in the matching work
1670 * completion is supported.
1671 */
1672 IB_RAW_PACKET_CAP_CVLAN_STRIPPING =
1673 IB_UVERBS_RAW_PACKET_CAP_CVLAN_STRIPPING,
1674 /*
1675 * Scatter FCS field of an incoming packet to host memory is supported.
1676 */
1677 IB_RAW_PACKET_CAP_SCATTER_FCS = IB_UVERBS_RAW_PACKET_CAP_SCATTER_FCS,
1678 /* Checksum offloads are supported (for both send and receive). */
1679 IB_RAW_PACKET_CAP_IP_CSUM = IB_UVERBS_RAW_PACKET_CAP_IP_CSUM,
1680 /*
1681 * When a packet is received for an RQ with no receive WQEs, the
1682 * packet processing is delayed.
1683 */
1684 IB_RAW_PACKET_CAP_DELAY_DROP = IB_UVERBS_RAW_PACKET_CAP_DELAY_DROP,
1685};
1686
1687enum ib_wq_type {
1688 IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1689};
1690
1691enum ib_wq_state {
1692 IB_WQS_RESET,
1693 IB_WQS_RDY,
1694 IB_WQS_ERR
1695};
1696
1697struct ib_wq {
1698 struct ib_device *device;
1699 struct ib_uwq_object *uobject;
1700 void *wq_context;
1701 void (*event_handler)(struct ib_event *, void *);
1702 struct ib_pd *pd;
1703 struct ib_cq *cq;
1704 u32 wq_num;
1705 enum ib_wq_state state;
1706 enum ib_wq_type wq_type;
1707 atomic_t usecnt;
1708};
1709
1710enum ib_wq_flags {
1711 IB_WQ_FLAGS_CVLAN_STRIPPING = IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1712 IB_WQ_FLAGS_SCATTER_FCS = IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1713 IB_WQ_FLAGS_DELAY_DROP = IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1714 IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1715 IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1716};
1717
1718struct ib_wq_init_attr {
1719 void *wq_context;
1720 enum ib_wq_type wq_type;
1721 u32 max_wr;
1722 u32 max_sge;
1723 struct ib_cq *cq;
1724 void (*event_handler)(struct ib_event *, void *);
1725 u32 create_flags; /* Use enum ib_wq_flags */
1726};
1727
1728enum ib_wq_attr_mask {
1729 IB_WQ_STATE = 1 << 0,
1730 IB_WQ_CUR_STATE = 1 << 1,
1731 IB_WQ_FLAGS = 1 << 2,
1732};
1733
1734struct ib_wq_attr {
1735 enum ib_wq_state wq_state;
1736 enum ib_wq_state curr_wq_state;
1737 u32 flags; /* Use enum ib_wq_flags */
1738 u32 flags_mask; /* Use enum ib_wq_flags */
1739};
1740
1741struct ib_rwq_ind_table {
1742 struct ib_device *device;
1743 struct ib_uobject *uobject;
1744 atomic_t usecnt;
1745 u32 ind_tbl_num;
1746 u32 log_ind_tbl_size;
1747 struct ib_wq **ind_tbl;
1748};
1749
1750struct ib_rwq_ind_table_init_attr {
1751 u32 log_ind_tbl_size;
1752 /* Each entry is a pointer to Receive Work Queue */
1753 struct ib_wq **ind_tbl;
1754};
1755
1756enum port_pkey_state {
1757 IB_PORT_PKEY_NOT_VALID = 0,
1758 IB_PORT_PKEY_VALID = 1,
1759 IB_PORT_PKEY_LISTED = 2,
1760};
1761
1762struct ib_qp_security;
1763
1764struct ib_port_pkey {
1765 enum port_pkey_state state;
1766 u16 pkey_index;
1767 u32 port_num;
1768 struct list_head qp_list;
1769 struct list_head to_error_list;
1770 struct ib_qp_security *sec;
1771};
1772
1773struct ib_ports_pkeys {
1774 struct ib_port_pkey main;
1775 struct ib_port_pkey alt;
1776};
1777
1778struct ib_qp_security {
1779 struct ib_qp *qp;
1780 struct ib_device *dev;
1781 /* Hold this mutex when changing port and pkey settings. */
1782 struct mutex mutex;
1783 struct ib_ports_pkeys *ports_pkeys;
1784 /* A list of all open shared QP handles. Required to enforce security
1785 * properly for all users of a shared QP.
1786 */
1787 struct list_head shared_qp_list;
1788 void *security;
1789 bool destroying;
1790 atomic_t error_list_count;
1791 struct completion error_complete;
1792 int error_comps_pending;
1793};
1794
1795/*
1796 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1797 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1798 */
1799struct ib_qp {
1800 struct ib_device *device;
1801 struct ib_pd *pd;
1802 struct ib_cq *send_cq;
1803 struct ib_cq *recv_cq;
1804 spinlock_t mr_lock;
1805 int mrs_used;
1806 struct list_head rdma_mrs;
1807 struct list_head sig_mrs;
1808 struct ib_srq *srq;
1809 struct completion srq_completion;
1810 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1811 struct list_head xrcd_list;
1812
1813 /* count times opened, mcast attaches, flow attaches */
1814 atomic_t usecnt;
1815 struct list_head open_list;
1816 struct ib_qp *real_qp;
1817 struct ib_uqp_object *uobject;
1818 void (*event_handler)(struct ib_event *, void *);
1819 void (*registered_event_handler)(struct ib_event *, void *);
1820 void *qp_context;
1821 /* sgid_attrs associated with the AV's */
1822 const struct ib_gid_attr *av_sgid_attr;
1823 const struct ib_gid_attr *alt_path_sgid_attr;
1824 u32 qp_num;
1825 u32 max_write_sge;
1826 u32 max_read_sge;
1827 enum ib_qp_type qp_type;
1828 struct ib_rwq_ind_table *rwq_ind_tbl;
1829 struct ib_qp_security *qp_sec;
1830 u32 port;
1831
1832 bool integrity_en;
1833 /*
1834 * Implementation details of the RDMA core, don't use in drivers:
1835 */
1836 struct rdma_restrack_entry res;
1837
1838 /* The counter the qp is bind to */
1839 struct rdma_counter *counter;
1840};
1841
1842struct ib_dm {
1843 struct ib_device *device;
1844 u32 length;
1845 u32 flags;
1846 struct ib_uobject *uobject;
1847 atomic_t usecnt;
1848};
1849
1850/* bit values to mark existence of ib_dmah fields */
1851enum {
1852 IB_DMAH_CPU_ID_EXISTS,
1853 IB_DMAH_MEM_TYPE_EXISTS,
1854 IB_DMAH_PH_EXISTS,
1855};
1856
1857struct ib_dmah {
1858 struct ib_device *device;
1859 struct ib_uobject *uobject;
1860 /*
1861 * Implementation details of the RDMA core, don't use in drivers:
1862 */
1863 struct rdma_restrack_entry res;
1864 u32 cpu_id;
1865 enum tph_mem_type mem_type;
1866 atomic_t usecnt;
1867 u8 ph;
1868 u8 valid_fields; /* use IB_DMAH_XXX_EXISTS */
1869};
1870
1871struct ib_mr {
1872 struct ib_device *device;
1873 struct ib_pd *pd;
1874 u32 lkey;
1875 u32 rkey;
1876 u64 iova;
1877 u64 length;
1878 unsigned int page_size;
1879 enum ib_mr_type type;
1880 bool need_inval;
1881 union {
1882 struct ib_uobject *uobject; /* user */
1883 struct list_head qp_entry; /* FR */
1884 };
1885
1886 struct ib_dm *dm;
1887 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1888 struct ib_dmah *dmah;
1889 /*
1890 * Implementation details of the RDMA core, don't use in drivers:
1891 */
1892 struct rdma_restrack_entry res;
1893};
1894
1895struct ib_mw {
1896 struct ib_device *device;
1897 struct ib_pd *pd;
1898 struct ib_uobject *uobject;
1899 u32 rkey;
1900 enum ib_mw_type type;
1901};
1902
1903/* Supported steering options */
1904enum ib_flow_attr_type {
1905 /* steering according to rule specifications */
1906 IB_FLOW_ATTR_NORMAL = 0x0,
1907 /* default unicast and multicast rule -
1908 * receive all Eth traffic which isn't steered to any QP
1909 */
1910 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1911 /* default multicast rule -
1912 * receive all Eth multicast traffic which isn't steered to any QP
1913 */
1914 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1915 /* sniffer rule - receive all port traffic */
1916 IB_FLOW_ATTR_SNIFFER = 0x3
1917};
1918
1919/* Supported steering header types */
1920enum ib_flow_spec_type {
1921 /* L2 headers*/
1922 IB_FLOW_SPEC_ETH = 0x20,
1923 IB_FLOW_SPEC_IB = 0x22,
1924 /* L3 header*/
1925 IB_FLOW_SPEC_IPV4 = 0x30,
1926 IB_FLOW_SPEC_IPV6 = 0x31,
1927 IB_FLOW_SPEC_ESP = 0x34,
1928 /* L4 headers*/
1929 IB_FLOW_SPEC_TCP = 0x40,
1930 IB_FLOW_SPEC_UDP = 0x41,
1931 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1932 IB_FLOW_SPEC_GRE = 0x51,
1933 IB_FLOW_SPEC_MPLS = 0x60,
1934 IB_FLOW_SPEC_INNER = 0x100,
1935 /* Actions */
1936 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1937 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1938 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1939 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1940};
1941#define IB_FLOW_SPEC_LAYER_MASK 0xF0
1942#define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1943
1944enum ib_flow_flags {
1945 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1946 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1947 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1948};
1949
1950struct ib_flow_eth_filter {
1951 u8 dst_mac[6];
1952 u8 src_mac[6];
1953 __be16 ether_type;
1954 __be16 vlan_tag;
1955};
1956
1957struct ib_flow_spec_eth {
1958 u32 type;
1959 u16 size;
1960 struct ib_flow_eth_filter val;
1961 struct ib_flow_eth_filter mask;
1962};
1963
1964struct ib_flow_ib_filter {
1965 __be16 dlid;
1966 __u8 sl;
1967};
1968
1969struct ib_flow_spec_ib {
1970 u32 type;
1971 u16 size;
1972 struct ib_flow_ib_filter val;
1973 struct ib_flow_ib_filter mask;
1974};
1975
1976/* IPv4 header flags */
1977enum ib_ipv4_flags {
1978 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1979 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1980 last have this flag set */
1981};
1982
1983struct ib_flow_ipv4_filter {
1984 __be32 src_ip;
1985 __be32 dst_ip;
1986 u8 proto;
1987 u8 tos;
1988 u8 ttl;
1989 u8 flags;
1990};
1991
1992struct ib_flow_spec_ipv4 {
1993 u32 type;
1994 u16 size;
1995 struct ib_flow_ipv4_filter val;
1996 struct ib_flow_ipv4_filter mask;
1997};
1998
1999struct ib_flow_ipv6_filter {
2000 u8 src_ip[16];
2001 u8 dst_ip[16];
2002 __be32 flow_label;
2003 u8 next_hdr;
2004 u8 traffic_class;
2005 u8 hop_limit;
2006} __packed;
2007
2008struct ib_flow_spec_ipv6 {
2009 u32 type;
2010 u16 size;
2011 struct ib_flow_ipv6_filter val;
2012 struct ib_flow_ipv6_filter mask;
2013};
2014
2015struct ib_flow_tcp_udp_filter {
2016 __be16 dst_port;
2017 __be16 src_port;
2018};
2019
2020struct ib_flow_spec_tcp_udp {
2021 u32 type;
2022 u16 size;
2023 struct ib_flow_tcp_udp_filter val;
2024 struct ib_flow_tcp_udp_filter mask;
2025};
2026
2027struct ib_flow_tunnel_filter {
2028 __be32 tunnel_id;
2029};
2030
2031/* ib_flow_spec_tunnel describes the Vxlan tunnel
2032 * the tunnel_id from val has the vni value
2033 */
2034struct ib_flow_spec_tunnel {
2035 u32 type;
2036 u16 size;
2037 struct ib_flow_tunnel_filter val;
2038 struct ib_flow_tunnel_filter mask;
2039};
2040
2041struct ib_flow_esp_filter {
2042 __be32 spi;
2043 __be32 seq;
2044};
2045
2046struct ib_flow_spec_esp {
2047 u32 type;
2048 u16 size;
2049 struct ib_flow_esp_filter val;
2050 struct ib_flow_esp_filter mask;
2051};
2052
2053struct ib_flow_gre_filter {
2054 __be16 c_ks_res0_ver;
2055 __be16 protocol;
2056 __be32 key;
2057};
2058
2059struct ib_flow_spec_gre {
2060 u32 type;
2061 u16 size;
2062 struct ib_flow_gre_filter val;
2063 struct ib_flow_gre_filter mask;
2064};
2065
2066struct ib_flow_mpls_filter {
2067 __be32 tag;
2068};
2069
2070struct ib_flow_spec_mpls {
2071 u32 type;
2072 u16 size;
2073 struct ib_flow_mpls_filter val;
2074 struct ib_flow_mpls_filter mask;
2075};
2076
2077struct ib_flow_spec_action_tag {
2078 enum ib_flow_spec_type type;
2079 u16 size;
2080 u32 tag_id;
2081};
2082
2083struct ib_flow_spec_action_drop {
2084 enum ib_flow_spec_type type;
2085 u16 size;
2086};
2087
2088struct ib_flow_spec_action_handle {
2089 enum ib_flow_spec_type type;
2090 u16 size;
2091 struct ib_flow_action *act;
2092};
2093
2094enum ib_counters_description {
2095 IB_COUNTER_PACKETS,
2096 IB_COUNTER_BYTES,
2097};
2098
2099struct ib_flow_spec_action_count {
2100 enum ib_flow_spec_type type;
2101 u16 size;
2102 struct ib_counters *counters;
2103};
2104
2105union ib_flow_spec {
2106 struct {
2107 u32 type;
2108 u16 size;
2109 };
2110 struct ib_flow_spec_eth eth;
2111 struct ib_flow_spec_ib ib;
2112 struct ib_flow_spec_ipv4 ipv4;
2113 struct ib_flow_spec_tcp_udp tcp_udp;
2114 struct ib_flow_spec_ipv6 ipv6;
2115 struct ib_flow_spec_tunnel tunnel;
2116 struct ib_flow_spec_esp esp;
2117 struct ib_flow_spec_gre gre;
2118 struct ib_flow_spec_mpls mpls;
2119 struct ib_flow_spec_action_tag flow_tag;
2120 struct ib_flow_spec_action_drop drop;
2121 struct ib_flow_spec_action_handle action;
2122 struct ib_flow_spec_action_count flow_count;
2123};
2124
2125struct ib_flow_attr {
2126 enum ib_flow_attr_type type;
2127 u16 size;
2128 u16 priority;
2129 u32 flags;
2130 u8 num_of_specs;
2131 u32 port;
2132 union ib_flow_spec flows[];
2133};
2134
2135struct ib_flow {
2136 struct ib_qp *qp;
2137 struct ib_device *device;
2138 struct ib_uobject *uobject;
2139};
2140
2141enum ib_flow_action_type {
2142 IB_FLOW_ACTION_UNSPECIFIED,
2143 IB_FLOW_ACTION_ESP = 1,
2144};
2145
2146struct ib_flow_action_attrs_esp_keymats {
2147 enum ib_uverbs_flow_action_esp_keymat protocol;
2148 union {
2149 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2150 } keymat;
2151};
2152
2153struct ib_flow_action_attrs_esp_replays {
2154 enum ib_uverbs_flow_action_esp_replay protocol;
2155 union {
2156 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2157 } replay;
2158};
2159
2160enum ib_flow_action_attrs_esp_flags {
2161 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2162 * This is done in order to share the same flags between user-space and
2163 * kernel and spare an unnecessary translation.
2164 */
2165
2166 /* Kernel flags */
2167 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2168 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2169};
2170
2171struct ib_flow_spec_list {
2172 struct ib_flow_spec_list *next;
2173 union ib_flow_spec spec;
2174};
2175
2176struct ib_flow_action_attrs_esp {
2177 struct ib_flow_action_attrs_esp_keymats *keymat;
2178 struct ib_flow_action_attrs_esp_replays *replay;
2179 struct ib_flow_spec_list *encap;
2180 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2181 * Value of 0 is a valid value.
2182 */
2183 u32 esn;
2184 u32 spi;
2185 u32 seq;
2186 u32 tfc_pad;
2187 /* Use enum ib_flow_action_attrs_esp_flags */
2188 u64 flags;
2189 u64 hard_limit_pkts;
2190};
2191
2192struct ib_flow_action {
2193 struct ib_device *device;
2194 struct ib_uobject *uobject;
2195 enum ib_flow_action_type type;
2196 atomic_t usecnt;
2197};
2198
2199struct ib_mad;
2200
2201enum ib_process_mad_flags {
2202 IB_MAD_IGNORE_MKEY = 1,
2203 IB_MAD_IGNORE_BKEY = 2,
2204 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2205};
2206
2207enum ib_mad_result {
2208 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2209 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2210 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2211 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2212};
2213
2214struct ib_port_cache {
2215 u64 subnet_prefix;
2216 struct ib_pkey_cache *pkey;
2217 struct ib_gid_table *gid;
2218 u8 lmc;
2219 enum ib_port_state port_state;
2220 enum ib_port_state last_port_state;
2221};
2222
2223struct ib_port_immutable {
2224 int pkey_tbl_len;
2225 int gid_tbl_len;
2226 u32 core_cap_flags;
2227 u32 max_mad_size;
2228};
2229
2230struct ib_port_data {
2231 struct ib_device *ib_dev;
2232
2233 struct ib_port_immutable immutable;
2234
2235 spinlock_t pkey_list_lock;
2236
2237 spinlock_t netdev_lock;
2238
2239 struct list_head pkey_list;
2240
2241 struct ib_port_cache cache;
2242
2243 struct net_device __rcu *netdev;
2244 netdevice_tracker netdev_tracker;
2245 struct hlist_node ndev_hash_link;
2246 struct rdma_port_counter port_counter;
2247 struct ib_port *sysfs;
2248};
2249
2250/* rdma netdev type - specifies protocol type */
2251enum rdma_netdev_t {
2252 RDMA_NETDEV_OPA_VNIC,
2253 RDMA_NETDEV_IPOIB,
2254};
2255
2256/**
2257 * struct rdma_netdev - rdma netdev
2258 * For cases where netstack interfacing is required.
2259 */
2260struct rdma_netdev {
2261 void *clnt_priv;
2262 struct ib_device *hca;
2263 u32 port_num;
2264 int mtu;
2265
2266 /*
2267 * cleanup function must be specified.
2268 * FIXME: This is only used for OPA_VNIC and that usage should be
2269 * removed too.
2270 */
2271 void (*free_rdma_netdev)(struct net_device *netdev);
2272
2273 /* control functions */
2274 void (*set_id)(struct net_device *netdev, int id);
2275 /* send packet */
2276 int (*send)(struct net_device *dev, struct sk_buff *skb,
2277 struct ib_ah *address, u32 dqpn);
2278 /* multicast */
2279 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2280 union ib_gid *gid, u16 mlid,
2281 int set_qkey, u32 qkey);
2282 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2283 union ib_gid *gid, u16 mlid);
2284 /* timeout */
2285 void (*tx_timeout)(struct net_device *dev, unsigned int txqueue);
2286};
2287
2288struct rdma_netdev_alloc_params {
2289 size_t sizeof_priv;
2290 unsigned int txqs;
2291 unsigned int rxqs;
2292 void *param;
2293
2294 int (*initialize_rdma_netdev)(struct ib_device *device, u32 port_num,
2295 struct net_device *netdev, void *param);
2296};
2297
2298struct ib_odp_counters {
2299 atomic64_t faults;
2300 atomic64_t faults_handled;
2301 atomic64_t invalidations;
2302 atomic64_t invalidations_handled;
2303 atomic64_t prefetch;
2304};
2305
2306struct ib_counters {
2307 struct ib_device *device;
2308 struct ib_uobject *uobject;
2309 /* num of objects attached */
2310 atomic_t usecnt;
2311};
2312
2313struct ib_counters_read_attr {
2314 u64 *counters_buff;
2315 u32 ncounters;
2316 u32 flags; /* use enum ib_read_counters_flags */
2317};
2318
2319struct uverbs_attr_bundle;
2320struct iw_cm_id;
2321struct iw_cm_conn_param;
2322
2323#define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2324 .size_##ib_struct = \
2325 (sizeof(struct drv_struct) + \
2326 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2327 BUILD_BUG_ON_ZERO( \
2328 !__same_type(((struct drv_struct *)NULL)->member, \
2329 struct ib_struct)))
2330
2331#define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2332 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2333 gfp, false))
2334
2335#define rdma_zalloc_drv_obj_numa(ib_dev, ib_type) \
2336 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2337 GFP_KERNEL, true))
2338
2339#define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2340 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2341
2342#define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2343
2344struct rdma_user_mmap_entry {
2345 struct kref ref;
2346 struct ib_ucontext *ucontext;
2347 unsigned long start_pgoff;
2348 size_t npages;
2349 bool driver_removed;
2350};
2351
2352/* Return the offset (in bytes) the user should pass to libc's mmap() */
2353static inline u64
2354rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2355{
2356 return (u64)entry->start_pgoff << PAGE_SHIFT;
2357}
2358
2359/**
2360 * struct ib_device_ops - InfiniBand device operations
2361 * This structure defines all the InfiniBand device operations, providers will
2362 * need to define the supported operations, otherwise they will be set to null.
2363 */
2364struct ib_device_ops {
2365 struct module *owner;
2366 enum rdma_driver_id driver_id;
2367 u32 uverbs_abi_ver;
2368 unsigned int uverbs_no_driver_id_binding:1;
2369
2370 /*
2371 * NOTE: New drivers should not make use of device_group; instead new
2372 * device parameter should be exposed via netlink command. This
2373 * mechanism exists only for existing drivers.
2374 */
2375 const struct attribute_group *device_group;
2376 const struct attribute_group **port_groups;
2377
2378 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2379 const struct ib_send_wr **bad_send_wr);
2380 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2381 const struct ib_recv_wr **bad_recv_wr);
2382 void (*drain_rq)(struct ib_qp *qp);
2383 void (*drain_sq)(struct ib_qp *qp);
2384 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2385 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2386 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2387 int (*post_srq_recv)(struct ib_srq *srq,
2388 const struct ib_recv_wr *recv_wr,
2389 const struct ib_recv_wr **bad_recv_wr);
2390 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2391 u32 port_num, const struct ib_wc *in_wc,
2392 const struct ib_grh *in_grh,
2393 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2394 size_t *out_mad_size, u16 *out_mad_pkey_index);
2395 int (*query_device)(struct ib_device *device,
2396 struct ib_device_attr *device_attr,
2397 struct ib_udata *udata);
2398 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2399 struct ib_device_modify *device_modify);
2400 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2401 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2402 int comp_vector);
2403 int (*query_port)(struct ib_device *device, u32 port_num,
2404 struct ib_port_attr *port_attr);
2405 int (*modify_port)(struct ib_device *device, u32 port_num,
2406 int port_modify_mask,
2407 struct ib_port_modify *port_modify);
2408 /**
2409 * The following mandatory functions are used only at device
2410 * registration. Keep functions such as these at the end of this
2411 * structure to avoid cache line misses when accessing struct ib_device
2412 * in fast paths.
2413 */
2414 int (*get_port_immutable)(struct ib_device *device, u32 port_num,
2415 struct ib_port_immutable *immutable);
2416 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2417 u32 port_num);
2418 /**
2419 * When calling get_netdev, the HW vendor's driver should return the
2420 * net device of device @device at port @port_num or NULL if such
2421 * a net device doesn't exist. The vendor driver should call dev_hold
2422 * on this net device. The HW vendor's device driver must guarantee
2423 * that this function returns NULL before the net device has finished
2424 * NETDEV_UNREGISTER state.
2425 */
2426 struct net_device *(*get_netdev)(struct ib_device *device,
2427 u32 port_num);
2428 /**
2429 * rdma netdev operation
2430 *
2431 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2432 * must return -EOPNOTSUPP if it doesn't support the specified type.
2433 */
2434 struct net_device *(*alloc_rdma_netdev)(
2435 struct ib_device *device, u32 port_num, enum rdma_netdev_t type,
2436 const char *name, unsigned char name_assign_type,
2437 void (*setup)(struct net_device *));
2438
2439 int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num,
2440 enum rdma_netdev_t type,
2441 struct rdma_netdev_alloc_params *params);
2442 /**
2443 * query_gid should be return GID value for @device, when @port_num
2444 * link layer is either IB or iWarp. It is no-op if @port_num port
2445 * is RoCE link layer.
2446 */
2447 int (*query_gid)(struct ib_device *device, u32 port_num, int index,
2448 union ib_gid *gid);
2449 /**
2450 * When calling add_gid, the HW vendor's driver should add the gid
2451 * of device of port at gid index available at @attr. Meta-info of
2452 * that gid (for example, the network device related to this gid) is
2453 * available at @attr. @context allows the HW vendor driver to store
2454 * extra information together with a GID entry. The HW vendor driver may
2455 * allocate memory to contain this information and store it in @context
2456 * when a new GID entry is written to. Params are consistent until the
2457 * next call of add_gid or delete_gid. The function should return 0 on
2458 * success or error otherwise. The function could be called
2459 * concurrently for different ports. This function is only called when
2460 * roce_gid_table is used.
2461 */
2462 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2463 /**
2464 * When calling del_gid, the HW vendor's driver should delete the
2465 * gid of device @device at gid index gid_index of port port_num
2466 * available in @attr.
2467 * Upon the deletion of a GID entry, the HW vendor must free any
2468 * allocated memory. The caller will clear @context afterwards.
2469 * This function is only called when roce_gid_table is used.
2470 */
2471 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2472 int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index,
2473 u16 *pkey);
2474 int (*alloc_ucontext)(struct ib_ucontext *context,
2475 struct ib_udata *udata);
2476 void (*dealloc_ucontext)(struct ib_ucontext *context);
2477 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2478 /**
2479 * This will be called once refcount of an entry in mmap_xa reaches
2480 * zero. The type of the memory that was mapped may differ between
2481 * entries and is opaque to the rdma_user_mmap interface.
2482 * Therefore needs to be implemented by the driver in mmap_free.
2483 */
2484 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2485 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2486 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2487 int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2488 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2489 struct ib_udata *udata);
2490 int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2491 struct ib_udata *udata);
2492 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2493 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2494 int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2495 int (*create_srq)(struct ib_srq *srq,
2496 struct ib_srq_init_attr *srq_init_attr,
2497 struct ib_udata *udata);
2498 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2499 enum ib_srq_attr_mask srq_attr_mask,
2500 struct ib_udata *udata);
2501 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2502 int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2503 int (*create_qp)(struct ib_qp *qp, struct ib_qp_init_attr *qp_init_attr,
2504 struct ib_udata *udata);
2505 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2506 int qp_attr_mask, struct ib_udata *udata);
2507 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2508 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2509 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2510 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2511 struct uverbs_attr_bundle *attrs);
2512 int (*create_cq_umem)(struct ib_cq *cq,
2513 const struct ib_cq_init_attr *attr,
2514 struct ib_umem *umem,
2515 struct uverbs_attr_bundle *attrs);
2516 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2517 int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2518 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2519 /**
2520 * pre_destroy_cq - Prevent a cq from generating any new work
2521 * completions, but not free any kernel resources
2522 */
2523 int (*pre_destroy_cq)(struct ib_cq *cq);
2524 /**
2525 * post_destroy_cq - Free all kernel resources
2526 */
2527 void (*post_destroy_cq)(struct ib_cq *cq);
2528 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2529 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2530 u64 virt_addr, int mr_access_flags,
2531 struct ib_dmah *dmah,
2532 struct ib_udata *udata);
2533 struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset,
2534 u64 length, u64 virt_addr, int fd,
2535 int mr_access_flags,
2536 struct ib_dmah *dmah,
2537 struct uverbs_attr_bundle *attrs);
2538 struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
2539 u64 length, u64 virt_addr,
2540 int mr_access_flags, struct ib_pd *pd,
2541 struct ib_udata *udata);
2542 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2543 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2544 u32 max_num_sg);
2545 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2546 u32 max_num_data_sg,
2547 u32 max_num_meta_sg);
2548 int (*advise_mr)(struct ib_pd *pd,
2549 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2550 struct ib_sge *sg_list, u32 num_sge,
2551 struct uverbs_attr_bundle *attrs);
2552
2553 /*
2554 * Kernel users should universally support relaxed ordering (RO), as
2555 * they are designed to read data only after observing the CQE and use
2556 * the DMA API correctly.
2557 *
2558 * Some drivers implicitly enable RO if platform supports it.
2559 */
2560 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2561 unsigned int *sg_offset);
2562 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2563 struct ib_mr_status *mr_status);
2564 int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2565 int (*dealloc_mw)(struct ib_mw *mw);
2566 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2567 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2568 int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2569 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2570 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2571 struct ib_flow_attr *flow_attr,
2572 struct ib_udata *udata);
2573 int (*destroy_flow)(struct ib_flow *flow_id);
2574 int (*destroy_flow_action)(struct ib_flow_action *action);
2575 int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port,
2576 int state);
2577 int (*get_vf_config)(struct ib_device *device, int vf, u32 port,
2578 struct ifla_vf_info *ivf);
2579 int (*get_vf_stats)(struct ib_device *device, int vf, u32 port,
2580 struct ifla_vf_stats *stats);
2581 int (*get_vf_guid)(struct ib_device *device, int vf, u32 port,
2582 struct ifla_vf_guid *node_guid,
2583 struct ifla_vf_guid *port_guid);
2584 int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid,
2585 int type);
2586 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2587 struct ib_wq_init_attr *init_attr,
2588 struct ib_udata *udata);
2589 int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2590 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2591 u32 wq_attr_mask, struct ib_udata *udata);
2592 int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2593 struct ib_rwq_ind_table_init_attr *init_attr,
2594 struct ib_udata *udata);
2595 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2596 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2597 struct ib_ucontext *context,
2598 struct ib_dm_alloc_attr *attr,
2599 struct uverbs_attr_bundle *attrs);
2600 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2601 int (*alloc_dmah)(struct ib_dmah *ibdmah,
2602 struct uverbs_attr_bundle *attrs);
2603 int (*dealloc_dmah)(struct ib_dmah *dmah, struct uverbs_attr_bundle *attrs);
2604 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2605 struct ib_dm_mr_attr *attr,
2606 struct uverbs_attr_bundle *attrs);
2607 int (*create_counters)(struct ib_counters *counters,
2608 struct uverbs_attr_bundle *attrs);
2609 int (*destroy_counters)(struct ib_counters *counters);
2610 int (*read_counters)(struct ib_counters *counters,
2611 struct ib_counters_read_attr *counters_read_attr,
2612 struct uverbs_attr_bundle *attrs);
2613 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2614 int data_sg_nents, unsigned int *data_sg_offset,
2615 struct scatterlist *meta_sg, int meta_sg_nents,
2616 unsigned int *meta_sg_offset);
2617
2618 /**
2619 * alloc_hw_[device,port]_stats - Allocate a struct rdma_hw_stats and
2620 * fill in the driver initialized data. The struct is kfree()'ed by
2621 * the sysfs core when the device is removed. A lifespan of -1 in the
2622 * return struct tells the core to set a default lifespan.
2623 */
2624 struct rdma_hw_stats *(*alloc_hw_device_stats)(struct ib_device *device);
2625 struct rdma_hw_stats *(*alloc_hw_port_stats)(struct ib_device *device,
2626 u32 port_num);
2627 /**
2628 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2629 * @index - The index in the value array we wish to have updated, or
2630 * num_counters if we want all stats updated
2631 * Return codes -
2632 * < 0 - Error, no counters updated
2633 * index - Updated the single counter pointed to by index
2634 * num_counters - Updated all counters (will reset the timestamp
2635 * and prevent further calls for lifespan milliseconds)
2636 * Drivers are allowed to update all counters in leiu of just the
2637 * one given in index at their option
2638 */
2639 int (*get_hw_stats)(struct ib_device *device,
2640 struct rdma_hw_stats *stats, u32 port, int index);
2641
2642 /**
2643 * modify_hw_stat - Modify the counter configuration
2644 * @enable: true/false when enable/disable a counter
2645 * Return codes - 0 on success or error code otherwise.
2646 */
2647 int (*modify_hw_stat)(struct ib_device *device, u32 port,
2648 unsigned int counter_index, bool enable);
2649 /**
2650 * Allows rdma drivers to add their own restrack attributes.
2651 */
2652 int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2653 int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2654 int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2655 int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2656 int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2657 int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2658 int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2659 int (*fill_res_srq_entry)(struct sk_buff *msg, struct ib_srq *ib_srq);
2660 int (*fill_res_srq_entry_raw)(struct sk_buff *msg, struct ib_srq *ib_srq);
2661
2662 /* Device lifecycle callbacks */
2663 /*
2664 * Called after the device becomes registered, before clients are
2665 * attached
2666 */
2667 int (*enable_driver)(struct ib_device *dev);
2668 /*
2669 * This is called as part of ib_dealloc_device().
2670 */
2671 void (*dealloc_driver)(struct ib_device *dev);
2672
2673 /* iWarp CM callbacks */
2674 void (*iw_add_ref)(struct ib_qp *qp);
2675 void (*iw_rem_ref)(struct ib_qp *qp);
2676 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2677 int (*iw_connect)(struct iw_cm_id *cm_id,
2678 struct iw_cm_conn_param *conn_param);
2679 int (*iw_accept)(struct iw_cm_id *cm_id,
2680 struct iw_cm_conn_param *conn_param);
2681 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2682 u8 pdata_len);
2683 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2684 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2685 /**
2686 * counter_bind_qp - Bind a QP to a counter.
2687 * @counter - The counter to be bound. If counter->id is zero then
2688 * the driver needs to allocate a new counter and set counter->id
2689 */
2690 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp,
2691 u32 port);
2692 /**
2693 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2694 * counter and bind it onto the default one
2695 */
2696 int (*counter_unbind_qp)(struct ib_qp *qp, u32 port);
2697 /**
2698 * counter_dealloc -De-allocate the hw counter
2699 */
2700 int (*counter_dealloc)(struct rdma_counter *counter);
2701 /**
2702 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2703 * the driver initialized data.
2704 */
2705 struct rdma_hw_stats *(*counter_alloc_stats)(
2706 struct rdma_counter *counter);
2707 /**
2708 * counter_update_stats - Query the stats value of this counter
2709 */
2710 int (*counter_update_stats)(struct rdma_counter *counter);
2711
2712 /**
2713 * counter_init - Initialize the driver specific rdma counter struct.
2714 */
2715 void (*counter_init)(struct rdma_counter *counter);
2716
2717 /**
2718 * Allows rdma drivers to add their own restrack attributes
2719 * dumped via 'rdma stat' iproute2 command.
2720 */
2721 int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2722
2723 /* query driver for its ucontext properties */
2724 int (*query_ucontext)(struct ib_ucontext *context,
2725 struct uverbs_attr_bundle *attrs);
2726
2727 /*
2728 * Provide NUMA node. This API exists for rdmavt/hfi1 only.
2729 * Everyone else relies on Linux memory management model.
2730 */
2731 int (*get_numa_node)(struct ib_device *dev);
2732
2733 /**
2734 * add_sub_dev - Add a sub IB device
2735 */
2736 struct ib_device *(*add_sub_dev)(struct ib_device *parent,
2737 enum rdma_nl_dev_type type,
2738 const char *name);
2739
2740 /**
2741 * del_sub_dev - Delete a sub IB device
2742 */
2743 void (*del_sub_dev)(struct ib_device *sub_dev);
2744
2745 /**
2746 * ufile_cleanup - Attempt to cleanup ubojects HW resources inside
2747 * the ufile.
2748 */
2749 void (*ufile_hw_cleanup)(struct ib_uverbs_file *ufile);
2750
2751 /**
2752 * report_port_event - Drivers need to implement this if they have
2753 * some private stuff to handle when link status changes.
2754 */
2755 void (*report_port_event)(struct ib_device *ibdev,
2756 struct net_device *ndev, unsigned long event);
2757
2758 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2759 DECLARE_RDMA_OBJ_SIZE(ib_counters);
2760 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2761 DECLARE_RDMA_OBJ_SIZE(ib_dmah);
2762 DECLARE_RDMA_OBJ_SIZE(ib_mw);
2763 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2764 DECLARE_RDMA_OBJ_SIZE(ib_qp);
2765 DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2766 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2767 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2768 DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2769 DECLARE_RDMA_OBJ_SIZE(rdma_counter);
2770};
2771
2772struct ib_core_device {
2773 /* device must be the first element in structure until,
2774 * union of ib_core_device and device exists in ib_device.
2775 */
2776 struct device dev;
2777 possible_net_t rdma_net;
2778 struct kobject *ports_kobj;
2779 struct list_head port_list;
2780 struct ib_device *owner; /* reach back to owner ib_device */
2781};
2782
2783struct rdma_restrack_root;
2784struct ib_device {
2785 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2786 struct device *dma_device;
2787 struct ib_device_ops ops;
2788 char name[IB_DEVICE_NAME_MAX];
2789 struct rcu_head rcu_head;
2790
2791 struct list_head event_handler_list;
2792 /* Protects event_handler_list */
2793 struct rw_semaphore event_handler_rwsem;
2794
2795 /* Protects QP's event_handler calls and open_qp list */
2796 spinlock_t qp_open_list_lock;
2797
2798 struct rw_semaphore client_data_rwsem;
2799 struct xarray client_data;
2800 struct mutex unregistration_lock;
2801
2802 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2803 rwlock_t cache_lock;
2804 /**
2805 * port_data is indexed by port number
2806 */
2807 struct ib_port_data *port_data;
2808
2809 int num_comp_vectors;
2810
2811 union {
2812 struct device dev;
2813 struct ib_core_device coredev;
2814 };
2815
2816 /* First group is for device attributes,
2817 * Second group is for driver provided attributes (optional).
2818 * Third group is for the hw_stats
2819 * It is a NULL terminated array.
2820 */
2821 const struct attribute_group *groups[4];
2822 u8 hw_stats_attr_index;
2823
2824 u64 uverbs_cmd_mask;
2825
2826 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2827 __be64 node_guid;
2828 u32 local_dma_lkey;
2829 u16 is_switch:1;
2830 /* Indicates kernel verbs support, should not be used in drivers */
2831 u16 kverbs_provider:1;
2832 /* CQ adaptive moderation (RDMA DIM) */
2833 u16 use_cq_dim:1;
2834 u8 node_type;
2835 u32 phys_port_cnt;
2836 struct ib_device_attr attrs;
2837 struct hw_stats_device_data *hw_stats_data;
2838
2839#ifdef CONFIG_CGROUP_RDMA
2840 struct rdmacg_device cg_device;
2841#endif
2842
2843 u32 index;
2844
2845 spinlock_t cq_pools_lock;
2846 struct list_head cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2847
2848 struct rdma_restrack_root *res;
2849
2850 const struct uapi_definition *driver_def;
2851
2852 /*
2853 * Positive refcount indicates that the device is currently
2854 * registered and cannot be unregistered.
2855 */
2856 refcount_t refcount;
2857 struct completion unreg_completion;
2858 struct work_struct unregistration_work;
2859
2860 const struct rdma_link_ops *link_ops;
2861
2862 /* Protects compat_devs xarray modifications */
2863 struct mutex compat_devs_mutex;
2864 /* Maintains compat devices for each net namespace */
2865 struct xarray compat_devs;
2866
2867 /* Used by iWarp CM */
2868 char iw_ifname[IFNAMSIZ];
2869 u32 iw_driver_flags;
2870 u32 lag_flags;
2871
2872 /* A parent device has a list of sub-devices */
2873 struct mutex subdev_lock;
2874 struct list_head subdev_list_head;
2875
2876 /* A sub device has a type and a parent */
2877 enum rdma_nl_dev_type type;
2878 struct ib_device *parent;
2879 struct list_head subdev_list;
2880
2881 enum rdma_nl_name_assign_type name_assign_type;
2882};
2883
2884static inline void *rdma_zalloc_obj(struct ib_device *dev, size_t size,
2885 gfp_t gfp, bool is_numa_aware)
2886{
2887 if (is_numa_aware && dev->ops.get_numa_node)
2888 return kzalloc_node(size, gfp, dev->ops.get_numa_node(dev));
2889
2890 return kzalloc(size, gfp);
2891}
2892
2893struct ib_client_nl_info;
2894struct ib_client {
2895 const char *name;
2896 int (*add)(struct ib_device *ibdev);
2897 void (*remove)(struct ib_device *, void *client_data);
2898 void (*rename)(struct ib_device *dev, void *client_data);
2899 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2900 struct ib_client_nl_info *res);
2901 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2902
2903 /* Returns the net_dev belonging to this ib_client and matching the
2904 * given parameters.
2905 * @dev: An RDMA device that the net_dev use for communication.
2906 * @port: A physical port number on the RDMA device.
2907 * @pkey: P_Key that the net_dev uses if applicable.
2908 * @gid: A GID that the net_dev uses to communicate.
2909 * @addr: An IP address the net_dev is configured with.
2910 * @client_data: The device's client data set by ib_set_client_data().
2911 *
2912 * An ib_client that implements a net_dev on top of RDMA devices
2913 * (such as IP over IB) should implement this callback, allowing the
2914 * rdma_cm module to find the right net_dev for a given request.
2915 *
2916 * The caller is responsible for calling dev_put on the returned
2917 * netdev. */
2918 struct net_device *(*get_net_dev_by_params)(
2919 struct ib_device *dev,
2920 u32 port,
2921 u16 pkey,
2922 const union ib_gid *gid,
2923 const struct sockaddr *addr,
2924 void *client_data);
2925
2926 refcount_t uses;
2927 struct completion uses_zero;
2928 u32 client_id;
2929
2930 /* kverbs are not required by the client */
2931 u8 no_kverbs_req:1;
2932};
2933
2934/*
2935 * IB block DMA iterator
2936 *
2937 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2938 * to a HW supported page size.
2939 */
2940struct ib_block_iter {
2941 /* internal states */
2942 struct scatterlist *__sg; /* sg holding the current aligned block */
2943 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2944 size_t __sg_numblocks; /* ib_umem_num_dma_blocks() */
2945 unsigned int __sg_nents; /* number of SG entries */
2946 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2947 unsigned int __pg_bit; /* alignment of current block */
2948};
2949
2950struct ib_device *_ib_alloc_device(size_t size, struct net *net);
2951#define ib_alloc_device(drv_struct, member) \
2952 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2953 BUILD_BUG_ON_ZERO(offsetof( \
2954 struct drv_struct, member)), \
2955 &init_net), \
2956 struct drv_struct, member)
2957
2958#define ib_alloc_device_with_net(drv_struct, member, net) \
2959 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2960 BUILD_BUG_ON_ZERO(offsetof( \
2961 struct drv_struct, member)), net), \
2962 struct drv_struct, member)
2963
2964void ib_dealloc_device(struct ib_device *device);
2965
2966void ib_get_device_fw_str(struct ib_device *device, char *str);
2967
2968int ib_register_device(struct ib_device *device, const char *name,
2969 struct device *dma_device);
2970void ib_unregister_device(struct ib_device *device);
2971void ib_unregister_driver(enum rdma_driver_id driver_id);
2972void ib_unregister_device_and_put(struct ib_device *device);
2973void ib_unregister_device_queued(struct ib_device *ib_dev);
2974
2975int ib_register_client (struct ib_client *client);
2976void ib_unregister_client(struct ib_client *client);
2977
2978void __rdma_block_iter_start(struct ib_block_iter *biter,
2979 struct scatterlist *sglist,
2980 unsigned int nents,
2981 unsigned long pgsz);
2982bool __rdma_block_iter_next(struct ib_block_iter *biter);
2983
2984/**
2985 * rdma_block_iter_dma_address - get the aligned dma address of the current
2986 * block held by the block iterator.
2987 * @biter: block iterator holding the memory block
2988 */
2989static inline dma_addr_t
2990rdma_block_iter_dma_address(struct ib_block_iter *biter)
2991{
2992 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2993}
2994
2995/**
2996 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2997 * @sglist: sglist to iterate over
2998 * @biter: block iterator holding the memory block
2999 * @nents: maximum number of sg entries to iterate over
3000 * @pgsz: best HW supported page size to use
3001 *
3002 * Callers may use rdma_block_iter_dma_address() to get each
3003 * blocks aligned DMA address.
3004 */
3005#define rdma_for_each_block(sglist, biter, nents, pgsz) \
3006 for (__rdma_block_iter_start(biter, sglist, nents, \
3007 pgsz); \
3008 __rdma_block_iter_next(biter);)
3009
3010/**
3011 * ib_get_client_data - Get IB client context
3012 * @device:Device to get context for
3013 * @client:Client to get context for
3014 *
3015 * ib_get_client_data() returns the client context data set with
3016 * ib_set_client_data(). This can only be called while the client is
3017 * registered to the device, once the ib_client remove() callback returns this
3018 * cannot be called.
3019 */
3020static inline void *ib_get_client_data(struct ib_device *device,
3021 struct ib_client *client)
3022{
3023 return xa_load(&device->client_data, index: client->client_id);
3024}
3025void ib_set_client_data(struct ib_device *device, struct ib_client *client,
3026 void *data);
3027void ib_set_device_ops(struct ib_device *device,
3028 const struct ib_device_ops *ops);
3029
3030int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
3031 unsigned long pfn, unsigned long size, pgprot_t prot,
3032 struct rdma_user_mmap_entry *entry);
3033int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
3034 struct rdma_user_mmap_entry *entry,
3035 size_t length);
3036int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
3037 struct rdma_user_mmap_entry *entry,
3038 size_t length, u32 min_pgoff,
3039 u32 max_pgoff);
3040
3041#if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
3042void rdma_user_mmap_disassociate(struct ib_device *device);
3043#else
3044static inline void rdma_user_mmap_disassociate(struct ib_device *device)
3045{
3046}
3047#endif
3048
3049static inline int
3050rdma_user_mmap_entry_insert_exact(struct ib_ucontext *ucontext,
3051 struct rdma_user_mmap_entry *entry,
3052 size_t length, u32 pgoff)
3053{
3054 return rdma_user_mmap_entry_insert_range(ucontext, entry, length, min_pgoff: pgoff,
3055 max_pgoff: pgoff);
3056}
3057
3058struct rdma_user_mmap_entry *
3059rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
3060 unsigned long pgoff);
3061struct rdma_user_mmap_entry *
3062rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
3063 struct vm_area_struct *vma);
3064void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
3065
3066void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
3067
3068static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
3069{
3070 return copy_from_user(to: dest, from: udata->inbuf, n: len) ? -EFAULT : 0;
3071}
3072
3073static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
3074{
3075 return copy_to_user(to: udata->outbuf, from: src, n: len) ? -EFAULT : 0;
3076}
3077
3078static inline bool ib_is_buffer_cleared(const void __user *p,
3079 size_t len)
3080{
3081 bool ret;
3082 u8 *buf;
3083
3084 if (len > USHRT_MAX)
3085 return false;
3086
3087 buf = memdup_user(p, len);
3088 if (IS_ERR(ptr: buf))
3089 return false;
3090
3091 ret = !memchr_inv(s: buf, c: 0, n: len);
3092 kfree(objp: buf);
3093 return ret;
3094}
3095
3096static inline bool ib_is_udata_cleared(struct ib_udata *udata,
3097 size_t offset,
3098 size_t len)
3099{
3100 return ib_is_buffer_cleared(p: udata->inbuf + offset, len);
3101}
3102
3103/**
3104 * ib_modify_qp_is_ok - Check that the supplied attribute mask
3105 * contains all required attributes and no attributes not allowed for
3106 * the given QP state transition.
3107 * @cur_state: Current QP state
3108 * @next_state: Next QP state
3109 * @type: QP type
3110 * @mask: Mask of supplied QP attributes
3111 *
3112 * This function is a helper function that a low-level driver's
3113 * modify_qp method can use to validate the consumer's input. It
3114 * checks that cur_state and next_state are valid QP states, that a
3115 * transition from cur_state to next_state is allowed by the IB spec,
3116 * and that the attribute mask supplied is allowed for the transition.
3117 */
3118bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
3119 enum ib_qp_type type, enum ib_qp_attr_mask mask);
3120
3121void ib_register_event_handler(struct ib_event_handler *event_handler);
3122void ib_unregister_event_handler(struct ib_event_handler *event_handler);
3123void ib_dispatch_event(const struct ib_event *event);
3124
3125int ib_query_port(struct ib_device *device,
3126 u32 port_num, struct ib_port_attr *port_attr);
3127
3128enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
3129 u32 port_num);
3130
3131/**
3132 * rdma_cap_ib_switch - Check if the device is IB switch
3133 * @device: Device to check
3134 *
3135 * Device driver is responsible for setting is_switch bit on
3136 * in ib_device structure at init time.
3137 *
3138 * Return: true if the device is IB switch.
3139 */
3140static inline bool rdma_cap_ib_switch(const struct ib_device *device)
3141{
3142 return device->is_switch;
3143}
3144
3145/**
3146 * rdma_start_port - Return the first valid port number for the device
3147 * specified
3148 *
3149 * @device: Device to be checked
3150 *
3151 * Return start port number
3152 */
3153static inline u32 rdma_start_port(const struct ib_device *device)
3154{
3155 return rdma_cap_ib_switch(device) ? 0 : 1;
3156}
3157
3158/**
3159 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3160 * @device - The struct ib_device * to iterate over
3161 * @iter - The unsigned int to store the port number
3162 */
3163#define rdma_for_each_port(device, iter) \
3164 for (iter = rdma_start_port(device + \
3165 BUILD_BUG_ON_ZERO(!__same_type(u32, \
3166 iter))); \
3167 iter <= rdma_end_port(device); iter++)
3168
3169/**
3170 * rdma_end_port - Return the last valid port number for the device
3171 * specified
3172 *
3173 * @device: Device to be checked
3174 *
3175 * Return last port number
3176 */
3177static inline u32 rdma_end_port(const struct ib_device *device)
3178{
3179 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3180}
3181
3182static inline int rdma_is_port_valid(const struct ib_device *device,
3183 unsigned int port)
3184{
3185 return (port >= rdma_start_port(device) &&
3186 port <= rdma_end_port(device));
3187}
3188
3189static inline bool rdma_is_grh_required(const struct ib_device *device,
3190 u32 port_num)
3191{
3192 return device->port_data[port_num].immutable.core_cap_flags &
3193 RDMA_CORE_PORT_IB_GRH_REQUIRED;
3194}
3195
3196static inline bool rdma_protocol_ib(const struct ib_device *device,
3197 u32 port_num)
3198{
3199 return device->port_data[port_num].immutable.core_cap_flags &
3200 RDMA_CORE_CAP_PROT_IB;
3201}
3202
3203static inline bool rdma_protocol_roce(const struct ib_device *device,
3204 u32 port_num)
3205{
3206 return device->port_data[port_num].immutable.core_cap_flags &
3207 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3208}
3209
3210static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device,
3211 u32 port_num)
3212{
3213 return device->port_data[port_num].immutable.core_cap_flags &
3214 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3215}
3216
3217static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device,
3218 u32 port_num)
3219{
3220 return device->port_data[port_num].immutable.core_cap_flags &
3221 RDMA_CORE_CAP_PROT_ROCE;
3222}
3223
3224static inline bool rdma_protocol_iwarp(const struct ib_device *device,
3225 u32 port_num)
3226{
3227 return device->port_data[port_num].immutable.core_cap_flags &
3228 RDMA_CORE_CAP_PROT_IWARP;
3229}
3230
3231static inline bool rdma_ib_or_roce(const struct ib_device *device,
3232 u32 port_num)
3233{
3234 return rdma_protocol_ib(device, port_num) ||
3235 rdma_protocol_roce(device, port_num);
3236}
3237
3238static inline bool rdma_protocol_raw_packet(const struct ib_device *device,
3239 u32 port_num)
3240{
3241 return device->port_data[port_num].immutable.core_cap_flags &
3242 RDMA_CORE_CAP_PROT_RAW_PACKET;
3243}
3244
3245static inline bool rdma_protocol_usnic(const struct ib_device *device,
3246 u32 port_num)
3247{
3248 return device->port_data[port_num].immutable.core_cap_flags &
3249 RDMA_CORE_CAP_PROT_USNIC;
3250}
3251
3252/**
3253 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3254 * Management Datagrams.
3255 * @device: Device to check
3256 * @port_num: Port number to check
3257 *
3258 * Management Datagrams (MAD) are a required part of the InfiniBand
3259 * specification and are supported on all InfiniBand devices. A slightly
3260 * extended version are also supported on OPA interfaces.
3261 *
3262 * Return: true if the port supports sending/receiving of MAD packets.
3263 */
3264static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num)
3265{
3266 return device->port_data[port_num].immutable.core_cap_flags &
3267 RDMA_CORE_CAP_IB_MAD;
3268}
3269
3270/**
3271 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3272 * Management Datagrams.
3273 * @device: Device to check
3274 * @port_num: Port number to check
3275 *
3276 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3277 * datagrams with their own versions. These OPA MADs share many but not all of
3278 * the characteristics of InfiniBand MADs.
3279 *
3280 * OPA MADs differ in the following ways:
3281 *
3282 * 1) MADs are variable size up to 2K
3283 * IBTA defined MADs remain fixed at 256 bytes
3284 * 2) OPA SMPs must carry valid PKeys
3285 * 3) OPA SMP packets are a different format
3286 *
3287 * Return: true if the port supports OPA MAD packet formats.
3288 */
3289static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num)
3290{
3291 return device->port_data[port_num].immutable.core_cap_flags &
3292 RDMA_CORE_CAP_OPA_MAD;
3293}
3294
3295/**
3296 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3297 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3298 * @device: Device to check
3299 * @port_num: Port number to check
3300 *
3301 * Each InfiniBand node is required to provide a Subnet Management Agent
3302 * that the subnet manager can access. Prior to the fabric being fully
3303 * configured by the subnet manager, the SMA is accessed via a well known
3304 * interface called the Subnet Management Interface (SMI). This interface
3305 * uses directed route packets to communicate with the SM to get around the
3306 * chicken and egg problem of the SM needing to know what's on the fabric
3307 * in order to configure the fabric, and needing to configure the fabric in
3308 * order to send packets to the devices on the fabric. These directed
3309 * route packets do not need the fabric fully configured in order to reach
3310 * their destination. The SMI is the only method allowed to send
3311 * directed route packets on an InfiniBand fabric.
3312 *
3313 * Return: true if the port provides an SMI.
3314 */
3315static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num)
3316{
3317 return device->port_data[port_num].immutable.core_cap_flags &
3318 RDMA_CORE_CAP_IB_SMI;
3319}
3320
3321/**
3322 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3323 * Communication Manager.
3324 * @device: Device to check
3325 * @port_num: Port number to check
3326 *
3327 * The InfiniBand Communication Manager is one of many pre-defined General
3328 * Service Agents (GSA) that are accessed via the General Service
3329 * Interface (GSI). It's role is to facilitate establishment of connections
3330 * between nodes as well as other management related tasks for established
3331 * connections.
3332 *
3333 * Return: true if the port supports an IB CM (this does not guarantee that
3334 * a CM is actually running however).
3335 */
3336static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num)
3337{
3338 return device->port_data[port_num].immutable.core_cap_flags &
3339 RDMA_CORE_CAP_IB_CM;
3340}
3341
3342/**
3343 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3344 * Communication Manager.
3345 * @device: Device to check
3346 * @port_num: Port number to check
3347 *
3348 * Similar to above, but specific to iWARP connections which have a different
3349 * managment protocol than InfiniBand.
3350 *
3351 * Return: true if the port supports an iWARP CM (this does not guarantee that
3352 * a CM is actually running however).
3353 */
3354static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num)
3355{
3356 return device->port_data[port_num].immutable.core_cap_flags &
3357 RDMA_CORE_CAP_IW_CM;
3358}
3359
3360/**
3361 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3362 * Subnet Administration.
3363 * @device: Device to check
3364 * @port_num: Port number to check
3365 *
3366 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3367 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3368 * fabrics, devices should resolve routes to other hosts by contacting the
3369 * SA to query the proper route.
3370 *
3371 * Return: true if the port should act as a client to the fabric Subnet
3372 * Administration interface. This does not imply that the SA service is
3373 * running locally.
3374 */
3375static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num)
3376{
3377 return device->port_data[port_num].immutable.core_cap_flags &
3378 RDMA_CORE_CAP_IB_SA;
3379}
3380
3381/**
3382 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3383 * Multicast.
3384 * @device: Device to check
3385 * @port_num: Port number to check
3386 *
3387 * InfiniBand multicast registration is more complex than normal IPv4 or
3388 * IPv6 multicast registration. Each Host Channel Adapter must register
3389 * with the Subnet Manager when it wishes to join a multicast group. It
3390 * should do so only once regardless of how many queue pairs it subscribes
3391 * to this group. And it should leave the group only after all queue pairs
3392 * attached to the group have been detached.
3393 *
3394 * Return: true if the port must undertake the additional adminstrative
3395 * overhead of registering/unregistering with the SM and tracking of the
3396 * total number of queue pairs attached to the multicast group.
3397 */
3398static inline bool rdma_cap_ib_mcast(const struct ib_device *device,
3399 u32 port_num)
3400{
3401 return rdma_cap_ib_sa(device, port_num);
3402}
3403
3404/**
3405 * rdma_cap_af_ib - Check if the port of device has the capability
3406 * Native Infiniband Address.
3407 * @device: Device to check
3408 * @port_num: Port number to check
3409 *
3410 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3411 * GID. RoCE uses a different mechanism, but still generates a GID via
3412 * a prescribed mechanism and port specific data.
3413 *
3414 * Return: true if the port uses a GID address to identify devices on the
3415 * network.
3416 */
3417static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num)
3418{
3419 return device->port_data[port_num].immutable.core_cap_flags &
3420 RDMA_CORE_CAP_AF_IB;
3421}
3422
3423/**
3424 * rdma_cap_eth_ah - Check if the port of device has the capability
3425 * Ethernet Address Handle.
3426 * @device: Device to check
3427 * @port_num: Port number to check
3428 *
3429 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3430 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3431 * port. Normally, packet headers are generated by the sending host
3432 * adapter, but when sending connectionless datagrams, we must manually
3433 * inject the proper headers for the fabric we are communicating over.
3434 *
3435 * Return: true if we are running as a RoCE port and must force the
3436 * addition of a Global Route Header built from our Ethernet Address
3437 * Handle into our header list for connectionless packets.
3438 */
3439static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num)
3440{
3441 return device->port_data[port_num].immutable.core_cap_flags &
3442 RDMA_CORE_CAP_ETH_AH;
3443}
3444
3445/**
3446 * rdma_cap_opa_ah - Check if the port of device supports
3447 * OPA Address handles
3448 * @device: Device to check
3449 * @port_num: Port number to check
3450 *
3451 * Return: true if we are running on an OPA device which supports
3452 * the extended OPA addressing.
3453 */
3454static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num)
3455{
3456 return (device->port_data[port_num].immutable.core_cap_flags &
3457 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3458}
3459
3460/**
3461 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3462 *
3463 * @device: Device
3464 * @port_num: Port number
3465 *
3466 * This MAD size includes the MAD headers and MAD payload. No other headers
3467 * are included.
3468 *
3469 * Return the max MAD size required by the Port. Will return 0 if the port
3470 * does not support MADs
3471 */
3472static inline size_t rdma_max_mad_size(const struct ib_device *device,
3473 u32 port_num)
3474{
3475 return device->port_data[port_num].immutable.max_mad_size;
3476}
3477
3478/**
3479 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3480 * @device: Device to check
3481 * @port_num: Port number to check
3482 *
3483 * RoCE GID table mechanism manages the various GIDs for a device.
3484 *
3485 * NOTE: if allocating the port's GID table has failed, this call will still
3486 * return true, but any RoCE GID table API will fail.
3487 *
3488 * Return: true if the port uses RoCE GID table mechanism in order to manage
3489 * its GIDs.
3490 */
3491static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3492 u32 port_num)
3493{
3494 return rdma_protocol_roce(device, port_num) &&
3495 device->ops.add_gid && device->ops.del_gid;
3496}
3497
3498/*
3499 * Check if the device supports READ W/ INVALIDATE.
3500 */
3501static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3502{
3503 /*
3504 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3505 * has support for it yet.
3506 */
3507 return rdma_protocol_iwarp(device: dev, port_num);
3508}
3509
3510/**
3511 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3512 * @device: Device
3513 * @port_num: 1 based Port number
3514 *
3515 * Return true if port is an Intel OPA port , false if not
3516 */
3517static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3518 u32 port_num)
3519{
3520 return (device->port_data[port_num].immutable.core_cap_flags &
3521 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3522}
3523
3524/**
3525 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3526 * @device: Device
3527 * @port_num: Port number
3528 * @mtu: enum value of MTU
3529 *
3530 * Return the MTU size supported by the port as an integer value. Will return
3531 * -1 if enum value of mtu is not supported.
3532 */
3533static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port,
3534 int mtu)
3535{
3536 if (rdma_core_cap_opa_port(device, port_num: port))
3537 return opa_mtu_enum_to_int(mtu: (enum opa_mtu)mtu);
3538 else
3539 return ib_mtu_enum_to_int(mtu: (enum ib_mtu)mtu);
3540}
3541
3542/**
3543 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3544 * @device: Device
3545 * @port_num: Port number
3546 * @attr: port attribute
3547 *
3548 * Return the MTU size supported by the port as an integer value.
3549 */
3550static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port,
3551 struct ib_port_attr *attr)
3552{
3553 if (rdma_core_cap_opa_port(device, port_num: port))
3554 return attr->phys_mtu;
3555 else
3556 return ib_mtu_enum_to_int(mtu: attr->max_mtu);
3557}
3558
3559int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port,
3560 int state);
3561int ib_get_vf_config(struct ib_device *device, int vf, u32 port,
3562 struct ifla_vf_info *info);
3563int ib_get_vf_stats(struct ib_device *device, int vf, u32 port,
3564 struct ifla_vf_stats *stats);
3565int ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
3566 struct ifla_vf_guid *node_guid,
3567 struct ifla_vf_guid *port_guid);
3568int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid,
3569 int type);
3570
3571int ib_query_pkey(struct ib_device *device,
3572 u32 port_num, u16 index, u16 *pkey);
3573
3574int ib_modify_device(struct ib_device *device,
3575 int device_modify_mask,
3576 struct ib_device_modify *device_modify);
3577
3578int ib_modify_port(struct ib_device *device,
3579 u32 port_num, int port_modify_mask,
3580 struct ib_port_modify *port_modify);
3581
3582int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3583 u32 *port_num, u16 *index);
3584
3585int ib_find_pkey(struct ib_device *device,
3586 u32 port_num, u16 pkey, u16 *index);
3587
3588enum ib_pd_flags {
3589 /*
3590 * Create a memory registration for all memory in the system and place
3591 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3592 * ULPs to avoid the overhead of dynamic MRs.
3593 *
3594 * This flag is generally considered unsafe and must only be used in
3595 * extremly trusted environments. Every use of it will log a warning
3596 * in the kernel log.
3597 */
3598 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3599};
3600
3601struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3602 const char *caller);
3603
3604/**
3605 * ib_alloc_pd - Allocates an unused protection domain.
3606 * @device: The device on which to allocate the protection domain.
3607 * @flags: protection domain flags
3608 *
3609 * A protection domain object provides an association between QPs, shared
3610 * receive queues, address handles, memory regions, and memory windows.
3611 *
3612 * Every PD has a local_dma_lkey which can be used as the lkey value for local
3613 * memory operations.
3614 */
3615#define ib_alloc_pd(device, flags) \
3616 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3617
3618int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3619
3620/**
3621 * ib_dealloc_pd - Deallocate kernel PD
3622 * @pd: The protection domain
3623 *
3624 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3625 */
3626static inline void ib_dealloc_pd(struct ib_pd *pd)
3627{
3628 int ret = ib_dealloc_pd_user(pd, NULL);
3629
3630 WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3631}
3632
3633enum rdma_create_ah_flags {
3634 /* In a sleepable context */
3635 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3636};
3637
3638/**
3639 * rdma_create_ah - Creates an address handle for the given address vector.
3640 * @pd: The protection domain associated with the address handle.
3641 * @ah_attr: The attributes of the address vector.
3642 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3643 *
3644 * The address handle is used to reference a local or global destination
3645 * in all UD QP post sends.
3646 */
3647struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3648 u32 flags);
3649
3650/**
3651 * rdma_create_user_ah - Creates an address handle for the given address vector.
3652 * It resolves destination mac address for ah attribute of RoCE type.
3653 * @pd: The protection domain associated with the address handle.
3654 * @ah_attr: The attributes of the address vector.
3655 * @udata: pointer to user's input output buffer information need by
3656 * provider driver.
3657 *
3658 * It returns 0 on success and returns appropriate error code on error.
3659 * The address handle is used to reference a local or global destination
3660 * in all UD QP post sends.
3661 */
3662struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3663 struct rdma_ah_attr *ah_attr,
3664 struct ib_udata *udata);
3665/**
3666 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3667 * work completion.
3668 * @hdr: the L3 header to parse
3669 * @net_type: type of header to parse
3670 * @sgid: place to store source gid
3671 * @dgid: place to store destination gid
3672 */
3673int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3674 enum rdma_network_type net_type,
3675 union ib_gid *sgid, union ib_gid *dgid);
3676
3677/**
3678 * ib_get_rdma_header_version - Get the header version
3679 * @hdr: the L3 header to parse
3680 */
3681int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3682
3683/**
3684 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3685 * work completion.
3686 * @device: Device on which the received message arrived.
3687 * @port_num: Port on which the received message arrived.
3688 * @wc: Work completion associated with the received message.
3689 * @grh: References the received global route header. This parameter is
3690 * ignored unless the work completion indicates that the GRH is valid.
3691 * @ah_attr: Returned attributes that can be used when creating an address
3692 * handle for replying to the message.
3693 * When ib_init_ah_attr_from_wc() returns success,
3694 * (a) for IB link layer it optionally contains a reference to SGID attribute
3695 * when GRH is present for IB link layer.
3696 * (b) for RoCE link layer it contains a reference to SGID attribute.
3697 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3698 * attributes which are initialized using ib_init_ah_attr_from_wc().
3699 *
3700 */
3701int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num,
3702 const struct ib_wc *wc, const struct ib_grh *grh,
3703 struct rdma_ah_attr *ah_attr);
3704
3705/**
3706 * ib_create_ah_from_wc - Creates an address handle associated with the
3707 * sender of the specified work completion.
3708 * @pd: The protection domain associated with the address handle.
3709 * @wc: Work completion information associated with a received message.
3710 * @grh: References the received global route header. This parameter is
3711 * ignored unless the work completion indicates that the GRH is valid.
3712 * @port_num: The outbound port number to associate with the address.
3713 *
3714 * The address handle is used to reference a local or global destination
3715 * in all UD QP post sends.
3716 */
3717struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3718 const struct ib_grh *grh, u32 port_num);
3719
3720/**
3721 * rdma_modify_ah - Modifies the address vector associated with an address
3722 * handle.
3723 * @ah: The address handle to modify.
3724 * @ah_attr: The new address vector attributes to associate with the
3725 * address handle.
3726 */
3727int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3728
3729/**
3730 * rdma_query_ah - Queries the address vector associated with an address
3731 * handle.
3732 * @ah: The address handle to query.
3733 * @ah_attr: The address vector attributes associated with the address
3734 * handle.
3735 */
3736int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3737
3738enum rdma_destroy_ah_flags {
3739 /* In a sleepable context */
3740 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3741};
3742
3743/**
3744 * rdma_destroy_ah_user - Destroys an address handle.
3745 * @ah: The address handle to destroy.
3746 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3747 * @udata: Valid user data or NULL for kernel objects
3748 */
3749int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3750
3751/**
3752 * rdma_destroy_ah - Destroys an kernel address handle.
3753 * @ah: The address handle to destroy.
3754 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3755 *
3756 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3757 */
3758static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3759{
3760 int ret = rdma_destroy_ah_user(ah, flags, NULL);
3761
3762 WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3763}
3764
3765struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3766 struct ib_srq_init_attr *srq_init_attr,
3767 struct ib_usrq_object *uobject,
3768 struct ib_udata *udata);
3769static inline struct ib_srq *
3770ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3771{
3772 if (!pd->device->ops.create_srq)
3773 return ERR_PTR(error: -EOPNOTSUPP);
3774
3775 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3776}
3777
3778/**
3779 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3780 * @srq: The SRQ to modify.
3781 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3782 * the current values of selected SRQ attributes are returned.
3783 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3784 * are being modified.
3785 *
3786 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3787 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3788 * the number of receives queued drops below the limit.
3789 */
3790int ib_modify_srq(struct ib_srq *srq,
3791 struct ib_srq_attr *srq_attr,
3792 enum ib_srq_attr_mask srq_attr_mask);
3793
3794/**
3795 * ib_query_srq - Returns the attribute list and current values for the
3796 * specified SRQ.
3797 * @srq: The SRQ to query.
3798 * @srq_attr: The attributes of the specified SRQ.
3799 */
3800int ib_query_srq(struct ib_srq *srq,
3801 struct ib_srq_attr *srq_attr);
3802
3803/**
3804 * ib_destroy_srq_user - Destroys the specified SRQ.
3805 * @srq: The SRQ to destroy.
3806 * @udata: Valid user data or NULL for kernel objects
3807 */
3808int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3809
3810/**
3811 * ib_destroy_srq - Destroys the specified kernel SRQ.
3812 * @srq: The SRQ to destroy.
3813 *
3814 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3815 */
3816static inline void ib_destroy_srq(struct ib_srq *srq)
3817{
3818 int ret = ib_destroy_srq_user(srq, NULL);
3819
3820 WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3821}
3822
3823/**
3824 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3825 * @srq: The SRQ to post the work request on.
3826 * @recv_wr: A list of work requests to post on the receive queue.
3827 * @bad_recv_wr: On an immediate failure, this parameter will reference
3828 * the work request that failed to be posted on the QP.
3829 */
3830static inline int ib_post_srq_recv(struct ib_srq *srq,
3831 const struct ib_recv_wr *recv_wr,
3832 const struct ib_recv_wr **bad_recv_wr)
3833{
3834 const struct ib_recv_wr *dummy;
3835
3836 return srq->device->ops.post_srq_recv(srq, recv_wr,
3837 bad_recv_wr ? : &dummy);
3838}
3839
3840struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd,
3841 struct ib_qp_init_attr *qp_init_attr,
3842 const char *caller);
3843/**
3844 * ib_create_qp - Creates a kernel QP associated with the specific protection
3845 * domain.
3846 * @pd: The protection domain associated with the QP.
3847 * @init_attr: A list of initial attributes required to create the
3848 * QP. If QP creation succeeds, then the attributes are updated to
3849 * the actual capabilities of the created QP.
3850 */
3851static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3852 struct ib_qp_init_attr *init_attr)
3853{
3854 return ib_create_qp_kernel(pd, qp_init_attr: init_attr, KBUILD_MODNAME);
3855}
3856
3857/**
3858 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3859 * @qp: The QP to modify.
3860 * @attr: On input, specifies the QP attributes to modify. On output,
3861 * the current values of selected QP attributes are returned.
3862 * @attr_mask: A bit-mask used to specify which attributes of the QP
3863 * are being modified.
3864 * @udata: pointer to user's input output buffer information
3865 * are being modified.
3866 * It returns 0 on success and returns appropriate error code on error.
3867 */
3868int ib_modify_qp_with_udata(struct ib_qp *qp,
3869 struct ib_qp_attr *attr,
3870 int attr_mask,
3871 struct ib_udata *udata);
3872
3873/**
3874 * ib_modify_qp - Modifies the attributes for the specified QP and then
3875 * transitions the QP to the given state.
3876 * @qp: The QP to modify.
3877 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3878 * the current values of selected QP attributes are returned.
3879 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3880 * are being modified.
3881 */
3882int ib_modify_qp(struct ib_qp *qp,
3883 struct ib_qp_attr *qp_attr,
3884 int qp_attr_mask);
3885
3886/**
3887 * ib_query_qp - Returns the attribute list and current values for the
3888 * specified QP.
3889 * @qp: The QP to query.
3890 * @qp_attr: The attributes of the specified QP.
3891 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3892 * @qp_init_attr: Additional attributes of the selected QP.
3893 *
3894 * The qp_attr_mask may be used to limit the query to gathering only the
3895 * selected attributes.
3896 */
3897int ib_query_qp(struct ib_qp *qp,
3898 struct ib_qp_attr *qp_attr,
3899 int qp_attr_mask,
3900 struct ib_qp_init_attr *qp_init_attr);
3901
3902/**
3903 * ib_destroy_qp - Destroys the specified QP.
3904 * @qp: The QP to destroy.
3905 * @udata: Valid udata or NULL for kernel objects
3906 */
3907int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3908
3909/**
3910 * ib_destroy_qp - Destroys the specified kernel QP.
3911 * @qp: The QP to destroy.
3912 *
3913 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3914 */
3915static inline int ib_destroy_qp(struct ib_qp *qp)
3916{
3917 return ib_destroy_qp_user(qp, NULL);
3918}
3919
3920/**
3921 * ib_open_qp - Obtain a reference to an existing sharable QP.
3922 * @xrcd - XRC domain
3923 * @qp_open_attr: Attributes identifying the QP to open.
3924 *
3925 * Returns a reference to a sharable QP.
3926 */
3927struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3928 struct ib_qp_open_attr *qp_open_attr);
3929
3930/**
3931 * ib_close_qp - Release an external reference to a QP.
3932 * @qp: The QP handle to release
3933 *
3934 * The opened QP handle is released by the caller. The underlying
3935 * shared QP is not destroyed until all internal references are released.
3936 */
3937int ib_close_qp(struct ib_qp *qp);
3938
3939/**
3940 * ib_post_send - Posts a list of work requests to the send queue of
3941 * the specified QP.
3942 * @qp: The QP to post the work request on.
3943 * @send_wr: A list of work requests to post on the send queue.
3944 * @bad_send_wr: On an immediate failure, this parameter will reference
3945 * the work request that failed to be posted on the QP.
3946 *
3947 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3948 * error is returned, the QP state shall not be affected,
3949 * ib_post_send() will return an immediate error after queueing any
3950 * earlier work requests in the list.
3951 */
3952static inline int ib_post_send(struct ib_qp *qp,
3953 const struct ib_send_wr *send_wr,
3954 const struct ib_send_wr **bad_send_wr)
3955{
3956 const struct ib_send_wr *dummy;
3957
3958 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3959}
3960
3961/**
3962 * ib_post_recv - Posts a list of work requests to the receive queue of
3963 * the specified QP.
3964 * @qp: The QP to post the work request on.
3965 * @recv_wr: A list of work requests to post on the receive queue.
3966 * @bad_recv_wr: On an immediate failure, this parameter will reference
3967 * the work request that failed to be posted on the QP.
3968 */
3969static inline int ib_post_recv(struct ib_qp *qp,
3970 const struct ib_recv_wr *recv_wr,
3971 const struct ib_recv_wr **bad_recv_wr)
3972{
3973 const struct ib_recv_wr *dummy;
3974
3975 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3976}
3977
3978struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3979 int comp_vector, enum ib_poll_context poll_ctx,
3980 const char *caller);
3981static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3982 int nr_cqe, int comp_vector,
3983 enum ib_poll_context poll_ctx)
3984{
3985 return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3986 KBUILD_MODNAME);
3987}
3988
3989struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3990 int nr_cqe, enum ib_poll_context poll_ctx,
3991 const char *caller);
3992
3993/**
3994 * ib_alloc_cq_any: Allocate kernel CQ
3995 * @dev: The IB device
3996 * @private: Private data attached to the CQE
3997 * @nr_cqe: Number of CQEs in the CQ
3998 * @poll_ctx: Context used for polling the CQ
3999 */
4000static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
4001 void *private, int nr_cqe,
4002 enum ib_poll_context poll_ctx)
4003{
4004 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
4005 KBUILD_MODNAME);
4006}
4007
4008void ib_free_cq(struct ib_cq *cq);
4009int ib_process_cq_direct(struct ib_cq *cq, int budget);
4010
4011/**
4012 * ib_create_cq - Creates a CQ on the specified device.
4013 * @device: The device on which to create the CQ.
4014 * @comp_handler: A user-specified callback that is invoked when a
4015 * completion event occurs on the CQ.
4016 * @event_handler: A user-specified callback that is invoked when an
4017 * asynchronous event not associated with a completion occurs on the CQ.
4018 * @cq_context: Context associated with the CQ returned to the user via
4019 * the associated completion and event handlers.
4020 * @cq_attr: The attributes the CQ should be created upon.
4021 *
4022 * Users can examine the cq structure to determine the actual CQ size.
4023 */
4024struct ib_cq *__ib_create_cq(struct ib_device *device,
4025 ib_comp_handler comp_handler,
4026 void (*event_handler)(struct ib_event *, void *),
4027 void *cq_context,
4028 const struct ib_cq_init_attr *cq_attr,
4029 const char *caller);
4030#define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
4031 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
4032
4033/**
4034 * ib_resize_cq - Modifies the capacity of the CQ.
4035 * @cq: The CQ to resize.
4036 * @cqe: The minimum size of the CQ.
4037 *
4038 * Users can examine the cq structure to determine the actual CQ size.
4039 */
4040int ib_resize_cq(struct ib_cq *cq, int cqe);
4041
4042/**
4043 * rdma_set_cq_moderation - Modifies moderation params of the CQ
4044 * @cq: The CQ to modify.
4045 * @cq_count: number of CQEs that will trigger an event
4046 * @cq_period: max period of time in usec before triggering an event
4047 *
4048 */
4049int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
4050
4051/**
4052 * ib_destroy_cq_user - Destroys the specified CQ.
4053 * @cq: The CQ to destroy.
4054 * @udata: Valid user data or NULL for kernel objects
4055 */
4056int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
4057
4058/**
4059 * ib_destroy_cq - Destroys the specified kernel CQ.
4060 * @cq: The CQ to destroy.
4061 *
4062 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
4063 */
4064static inline void ib_destroy_cq(struct ib_cq *cq)
4065{
4066 int ret = ib_destroy_cq_user(cq, NULL);
4067
4068 WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
4069}
4070
4071/**
4072 * ib_poll_cq - poll a CQ for completion(s)
4073 * @cq:the CQ being polled
4074 * @num_entries:maximum number of completions to return
4075 * @wc:array of at least @num_entries &struct ib_wc where completions
4076 * will be returned
4077 *
4078 * Poll a CQ for (possibly multiple) completions. If the return value
4079 * is < 0, an error occurred. If the return value is >= 0, it is the
4080 * number of completions returned. If the return value is
4081 * non-negative and < num_entries, then the CQ was emptied.
4082 */
4083static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
4084 struct ib_wc *wc)
4085{
4086 return cq->device->ops.poll_cq(cq, num_entries, wc);
4087}
4088
4089/**
4090 * ib_req_notify_cq - Request completion notification on a CQ.
4091 * @cq: The CQ to generate an event for.
4092 * @flags:
4093 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
4094 * to request an event on the next solicited event or next work
4095 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
4096 * may also be |ed in to request a hint about missed events, as
4097 * described below.
4098 *
4099 * Return Value:
4100 * < 0 means an error occurred while requesting notification
4101 * == 0 means notification was requested successfully, and if
4102 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
4103 * were missed and it is safe to wait for another event. In
4104 * this case is it guaranteed that any work completions added
4105 * to the CQ since the last CQ poll will trigger a completion
4106 * notification event.
4107 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
4108 * in. It means that the consumer must poll the CQ again to
4109 * make sure it is empty to avoid missing an event because of a
4110 * race between requesting notification and an entry being
4111 * added to the CQ. This return value means it is possible
4112 * (but not guaranteed) that a work completion has been added
4113 * to the CQ since the last poll without triggering a
4114 * completion notification event.
4115 */
4116static inline int ib_req_notify_cq(struct ib_cq *cq,
4117 enum ib_cq_notify_flags flags)
4118{
4119 return cq->device->ops.req_notify_cq(cq, flags);
4120}
4121
4122struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
4123 int comp_vector_hint,
4124 enum ib_poll_context poll_ctx);
4125
4126void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
4127
4128/*
4129 * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
4130 * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
4131 * address into the dma address.
4132 */
4133static inline bool ib_uses_virt_dma(struct ib_device *dev)
4134{
4135 return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
4136}
4137
4138/*
4139 * Check if a IB device's underlying DMA mapping supports P2PDMA transfers.
4140 */
4141static inline bool ib_dma_pci_p2p_dma_supported(struct ib_device *dev)
4142{
4143 if (ib_uses_virt_dma(dev))
4144 return false;
4145
4146 return dma_pci_p2pdma_supported(dev: dev->dma_device);
4147}
4148
4149/**
4150 * ib_virt_dma_to_ptr - Convert a dma_addr to a kernel pointer
4151 * @dma_addr: The DMA address
4152 *
4153 * Used by ib_uses_virt_dma() devices to get back to the kernel pointer after
4154 * going through the dma_addr marshalling.
4155 */
4156static inline void *ib_virt_dma_to_ptr(u64 dma_addr)
4157{
4158 /* virt_dma mode maps the kvs's directly into the dma addr */
4159 return (void *)(uintptr_t)dma_addr;
4160}
4161
4162/**
4163 * ib_virt_dma_to_page - Convert a dma_addr to a struct page
4164 * @dma_addr: The DMA address
4165 *
4166 * Used by ib_uses_virt_dma() device to get back to the struct page after going
4167 * through the dma_addr marshalling.
4168 */
4169static inline struct page *ib_virt_dma_to_page(u64 dma_addr)
4170{
4171 return virt_to_page(ib_virt_dma_to_ptr(dma_addr));
4172}
4173
4174/**
4175 * ib_dma_mapping_error - check a DMA addr for error
4176 * @dev: The device for which the dma_addr was created
4177 * @dma_addr: The DMA address to check
4178 */
4179static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
4180{
4181 if (ib_uses_virt_dma(dev))
4182 return 0;
4183 return dma_mapping_error(dev: dev->dma_device, dma_addr);
4184}
4185
4186/**
4187 * ib_dma_map_single - Map a kernel virtual address to DMA address
4188 * @dev: The device for which the dma_addr is to be created
4189 * @cpu_addr: The kernel virtual address
4190 * @size: The size of the region in bytes
4191 * @direction: The direction of the DMA
4192 */
4193static inline u64 ib_dma_map_single(struct ib_device *dev,
4194 void *cpu_addr, size_t size,
4195 enum dma_data_direction direction)
4196{
4197 if (ib_uses_virt_dma(dev))
4198 return (uintptr_t)cpu_addr;
4199 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
4200}
4201
4202/**
4203 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4204 * @dev: The device for which the DMA address was created
4205 * @addr: The DMA address
4206 * @size: The size of the region in bytes
4207 * @direction: The direction of the DMA
4208 */
4209static inline void ib_dma_unmap_single(struct ib_device *dev,
4210 u64 addr, size_t size,
4211 enum dma_data_direction direction)
4212{
4213 if (!ib_uses_virt_dma(dev))
4214 dma_unmap_single(dev->dma_device, addr, size, direction);
4215}
4216
4217/**
4218 * ib_dma_map_page - Map a physical page to DMA address
4219 * @dev: The device for which the dma_addr is to be created
4220 * @page: The page to be mapped
4221 * @offset: The offset within the page
4222 * @size: The size of the region in bytes
4223 * @direction: The direction of the DMA
4224 */
4225static inline u64 ib_dma_map_page(struct ib_device *dev,
4226 struct page *page,
4227 unsigned long offset,
4228 size_t size,
4229 enum dma_data_direction direction)
4230{
4231 if (ib_uses_virt_dma(dev))
4232 return (uintptr_t)(page_address(page) + offset);
4233 return dma_map_page(dev->dma_device, page, offset, size, direction);
4234}
4235
4236/**
4237 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4238 * @dev: The device for which the DMA address was created
4239 * @addr: The DMA address
4240 * @size: The size of the region in bytes
4241 * @direction: The direction of the DMA
4242 */
4243static inline void ib_dma_unmap_page(struct ib_device *dev,
4244 u64 addr, size_t size,
4245 enum dma_data_direction direction)
4246{
4247 if (!ib_uses_virt_dma(dev))
4248 dma_unmap_page(dev->dma_device, addr, size, direction);
4249}
4250
4251int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4252static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4253 struct scatterlist *sg, int nents,
4254 enum dma_data_direction direction,
4255 unsigned long dma_attrs)
4256{
4257 if (ib_uses_virt_dma(dev))
4258 return ib_dma_virt_map_sg(dev, sg, nents);
4259 return dma_map_sg_attrs(dev: dev->dma_device, sg, nents, dir: direction,
4260 attrs: dma_attrs);
4261}
4262
4263static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4264 struct scatterlist *sg, int nents,
4265 enum dma_data_direction direction,
4266 unsigned long dma_attrs)
4267{
4268 if (!ib_uses_virt_dma(dev))
4269 dma_unmap_sg_attrs(dev: dev->dma_device, sg, nents, dir: direction,
4270 attrs: dma_attrs);
4271}
4272
4273/**
4274 * ib_dma_map_sgtable_attrs - Map a scatter/gather table to DMA addresses
4275 * @dev: The device for which the DMA addresses are to be created
4276 * @sg: The sg_table object describing the buffer
4277 * @direction: The direction of the DMA
4278 * @attrs: Optional DMA attributes for the map operation
4279 */
4280static inline int ib_dma_map_sgtable_attrs(struct ib_device *dev,
4281 struct sg_table *sgt,
4282 enum dma_data_direction direction,
4283 unsigned long dma_attrs)
4284{
4285 int nents;
4286
4287 if (ib_uses_virt_dma(dev)) {
4288 nents = ib_dma_virt_map_sg(dev, sg: sgt->sgl, nents: sgt->orig_nents);
4289 if (!nents)
4290 return -EIO;
4291 sgt->nents = nents;
4292 return 0;
4293 }
4294 return dma_map_sgtable(dev: dev->dma_device, sgt, dir: direction, attrs: dma_attrs);
4295}
4296
4297static inline void ib_dma_unmap_sgtable_attrs(struct ib_device *dev,
4298 struct sg_table *sgt,
4299 enum dma_data_direction direction,
4300 unsigned long dma_attrs)
4301{
4302 if (!ib_uses_virt_dma(dev))
4303 dma_unmap_sgtable(dev: dev->dma_device, sgt, dir: direction, attrs: dma_attrs);
4304}
4305
4306/**
4307 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4308 * @dev: The device for which the DMA addresses are to be created
4309 * @sg: The array of scatter/gather entries
4310 * @nents: The number of scatter/gather entries
4311 * @direction: The direction of the DMA
4312 */
4313static inline int ib_dma_map_sg(struct ib_device *dev,
4314 struct scatterlist *sg, int nents,
4315 enum dma_data_direction direction)
4316{
4317 return ib_dma_map_sg_attrs(dev, sg, nents, direction, dma_attrs: 0);
4318}
4319
4320/**
4321 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4322 * @dev: The device for which the DMA addresses were created
4323 * @sg: The array of scatter/gather entries
4324 * @nents: The number of scatter/gather entries
4325 * @direction: The direction of the DMA
4326 */
4327static inline void ib_dma_unmap_sg(struct ib_device *dev,
4328 struct scatterlist *sg, int nents,
4329 enum dma_data_direction direction)
4330{
4331 ib_dma_unmap_sg_attrs(dev, sg, nents, direction, dma_attrs: 0);
4332}
4333
4334/**
4335 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4336 * @dev: The device to query
4337 *
4338 * The returned value represents a size in bytes.
4339 */
4340static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4341{
4342 if (ib_uses_virt_dma(dev))
4343 return UINT_MAX;
4344 return dma_get_max_seg_size(dev: dev->dma_device);
4345}
4346
4347/**
4348 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4349 * @dev: The device for which the DMA address was created
4350 * @addr: The DMA address
4351 * @size: The size of the region in bytes
4352 * @dir: The direction of the DMA
4353 */
4354static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4355 u64 addr,
4356 size_t size,
4357 enum dma_data_direction dir)
4358{
4359 if (!ib_uses_virt_dma(dev))
4360 dma_sync_single_for_cpu(dev: dev->dma_device, addr, size, dir);
4361}
4362
4363/**
4364 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4365 * @dev: The device for which the DMA address was created
4366 * @addr: The DMA address
4367 * @size: The size of the region in bytes
4368 * @dir: The direction of the DMA
4369 */
4370static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4371 u64 addr,
4372 size_t size,
4373 enum dma_data_direction dir)
4374{
4375 if (!ib_uses_virt_dma(dev))
4376 dma_sync_single_for_device(dev: dev->dma_device, addr, size, dir);
4377}
4378
4379/* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4380 * space. This function should be called when 'current' is the owning MM.
4381 */
4382struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4383 u64 virt_addr, int mr_access_flags);
4384
4385/* ib_advise_mr - give an advice about an address range in a memory region */
4386int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4387 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4388/**
4389 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4390 * HCA translation table.
4391 * @mr: The memory region to deregister.
4392 * @udata: Valid user data or NULL for kernel object
4393 *
4394 * This function can fail, if the memory region has memory windows bound to it.
4395 */
4396int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4397
4398/**
4399 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4400 * HCA translation table.
4401 * @mr: The memory region to deregister.
4402 *
4403 * This function can fail, if the memory region has memory windows bound to it.
4404 *
4405 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4406 */
4407static inline int ib_dereg_mr(struct ib_mr *mr)
4408{
4409 return ib_dereg_mr_user(mr, NULL);
4410}
4411
4412struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4413 u32 max_num_sg);
4414
4415struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4416 u32 max_num_data_sg,
4417 u32 max_num_meta_sg);
4418
4419/**
4420 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4421 * R_Key and L_Key.
4422 * @mr - struct ib_mr pointer to be updated.
4423 * @newkey - new key to be used.
4424 */
4425static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4426{
4427 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4428 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4429}
4430
4431/**
4432 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4433 * for calculating a new rkey for type 2 memory windows.
4434 * @rkey - the rkey to increment.
4435 */
4436static inline u32 ib_inc_rkey(u32 rkey)
4437{
4438 const u32 mask = 0x000000ff;
4439 return ((rkey + 1) & mask) | (rkey & ~mask);
4440}
4441
4442/**
4443 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4444 * @qp: QP to attach to the multicast group. The QP must be type
4445 * IB_QPT_UD.
4446 * @gid: Multicast group GID.
4447 * @lid: Multicast group LID in host byte order.
4448 *
4449 * In order to send and receive multicast packets, subnet
4450 * administration must have created the multicast group and configured
4451 * the fabric appropriately. The port associated with the specified
4452 * QP must also be a member of the multicast group.
4453 */
4454int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4455
4456/**
4457 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4458 * @qp: QP to detach from the multicast group.
4459 * @gid: Multicast group GID.
4460 * @lid: Multicast group LID in host byte order.
4461 */
4462int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4463
4464struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4465 struct inode *inode, struct ib_udata *udata);
4466int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4467
4468static inline int ib_check_mr_access(struct ib_device *ib_dev,
4469 unsigned int flags)
4470{
4471 u64 device_cap = ib_dev->attrs.device_cap_flags;
4472
4473 /*
4474 * Local write permission is required if remote write or
4475 * remote atomic permission is also requested.
4476 */
4477 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4478 !(flags & IB_ACCESS_LOCAL_WRITE))
4479 return -EINVAL;
4480
4481 if (flags & ~IB_ACCESS_SUPPORTED)
4482 return -EINVAL;
4483
4484 if (flags & IB_ACCESS_ON_DEMAND &&
4485 !(ib_dev->attrs.kernel_cap_flags & IBK_ON_DEMAND_PAGING))
4486 return -EOPNOTSUPP;
4487
4488 if ((flags & IB_ACCESS_FLUSH_GLOBAL &&
4489 !(device_cap & IB_DEVICE_FLUSH_GLOBAL)) ||
4490 (flags & IB_ACCESS_FLUSH_PERSISTENT &&
4491 !(device_cap & IB_DEVICE_FLUSH_PERSISTENT)))
4492 return -EOPNOTSUPP;
4493
4494 return 0;
4495}
4496
4497static inline bool ib_access_writable(int access_flags)
4498{
4499 /*
4500 * We have writable memory backing the MR if any of the following
4501 * access flags are set. "Local write" and "remote write" obviously
4502 * require write access. "Remote atomic" can do things like fetch and
4503 * add, which will modify memory, and "MW bind" can change permissions
4504 * by binding a window.
4505 */
4506 return access_flags &
4507 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4508 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4509}
4510
4511/**
4512 * ib_check_mr_status: lightweight check of MR status.
4513 * This routine may provide status checks on a selected
4514 * ib_mr. first use is for signature status check.
4515 *
4516 * @mr: A memory region.
4517 * @check_mask: Bitmask of which checks to perform from
4518 * ib_mr_status_check enumeration.
4519 * @mr_status: The container of relevant status checks.
4520 * failed checks will be indicated in the status bitmask
4521 * and the relevant info shall be in the error item.
4522 */
4523int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4524 struct ib_mr_status *mr_status);
4525
4526/**
4527 * ib_device_try_get: Hold a registration lock
4528 * device: The device to lock
4529 *
4530 * A device under an active registration lock cannot become unregistered. It
4531 * is only possible to obtain a registration lock on a device that is fully
4532 * registered, otherwise this function returns false.
4533 *
4534 * The registration lock is only necessary for actions which require the
4535 * device to still be registered. Uses that only require the device pointer to
4536 * be valid should use get_device(&ibdev->dev) to hold the memory.
4537 *
4538 */
4539static inline bool ib_device_try_get(struct ib_device *dev)
4540{
4541 return refcount_inc_not_zero(r: &dev->refcount);
4542}
4543
4544void ib_device_put(struct ib_device *device);
4545struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4546 enum rdma_driver_id driver_id);
4547struct ib_device *ib_device_get_by_name(const char *name,
4548 enum rdma_driver_id driver_id);
4549struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port,
4550 u16 pkey, const union ib_gid *gid,
4551 const struct sockaddr *addr);
4552int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4553 unsigned int port);
4554struct net_device *ib_device_get_netdev(struct ib_device *ib_dev,
4555 u32 port);
4556int ib_query_netdev_port(struct ib_device *ibdev, struct net_device *ndev,
4557 u32 *port);
4558
4559static inline enum ib_port_state ib_get_curr_port_state(struct net_device *net_dev)
4560{
4561 return (netif_running(dev: net_dev) && netif_carrier_ok(dev: net_dev)) ?
4562 IB_PORT_ACTIVE : IB_PORT_DOWN;
4563}
4564
4565void ib_dispatch_port_state_event(struct ib_device *ibdev,
4566 struct net_device *ndev);
4567struct ib_wq *ib_create_wq(struct ib_pd *pd,
4568 struct ib_wq_init_attr *init_attr);
4569int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4570
4571int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4572 unsigned int *sg_offset, unsigned int page_size);
4573int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4574 int data_sg_nents, unsigned int *data_sg_offset,
4575 struct scatterlist *meta_sg, int meta_sg_nents,
4576 unsigned int *meta_sg_offset, unsigned int page_size);
4577
4578static inline int
4579ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4580 unsigned int *sg_offset, unsigned int page_size)
4581{
4582 int n;
4583
4584 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4585 mr->iova = 0;
4586
4587 return n;
4588}
4589
4590int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4591 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4592
4593void ib_drain_rq(struct ib_qp *qp);
4594void ib_drain_sq(struct ib_qp *qp);
4595void ib_drain_qp(struct ib_qp *qp);
4596
4597int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed,
4598 u8 *width);
4599
4600static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4601{
4602 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4603 return attr->roce.dmac;
4604 return NULL;
4605}
4606
4607static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4608{
4609 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4610 attr->ib.dlid = (u16)dlid;
4611 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4612 attr->opa.dlid = dlid;
4613}
4614
4615static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4616{
4617 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4618 return attr->ib.dlid;
4619 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4620 return attr->opa.dlid;
4621 return 0;
4622}
4623
4624static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4625{
4626 attr->sl = sl;
4627}
4628
4629static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4630{
4631 return attr->sl;
4632}
4633
4634static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4635 u8 src_path_bits)
4636{
4637 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4638 attr->ib.src_path_bits = src_path_bits;
4639 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4640 attr->opa.src_path_bits = src_path_bits;
4641}
4642
4643static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4644{
4645 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4646 return attr->ib.src_path_bits;
4647 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4648 return attr->opa.src_path_bits;
4649 return 0;
4650}
4651
4652static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4653 bool make_grd)
4654{
4655 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4656 attr->opa.make_grd = make_grd;
4657}
4658
4659static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4660{
4661 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4662 return attr->opa.make_grd;
4663 return false;
4664}
4665
4666static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num)
4667{
4668 attr->port_num = port_num;
4669}
4670
4671static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4672{
4673 return attr->port_num;
4674}
4675
4676static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4677 u8 static_rate)
4678{
4679 attr->static_rate = static_rate;
4680}
4681
4682static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4683{
4684 return attr->static_rate;
4685}
4686
4687static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4688 enum ib_ah_flags flag)
4689{
4690 attr->ah_flags = flag;
4691}
4692
4693static inline enum ib_ah_flags
4694 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4695{
4696 return attr->ah_flags;
4697}
4698
4699static inline const struct ib_global_route
4700 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4701{
4702 return &attr->grh;
4703}
4704
4705/*To retrieve and modify the grh */
4706static inline struct ib_global_route
4707 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4708{
4709 return &attr->grh;
4710}
4711
4712static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4713{
4714 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4715
4716 memcpy(to: grh->dgid.raw, from: dgid, len: sizeof(grh->dgid));
4717}
4718
4719static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4720 __be64 prefix)
4721{
4722 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4723
4724 grh->dgid.global.subnet_prefix = prefix;
4725}
4726
4727static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4728 __be64 if_id)
4729{
4730 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4731
4732 grh->dgid.global.interface_id = if_id;
4733}
4734
4735static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4736 union ib_gid *dgid, u32 flow_label,
4737 u8 sgid_index, u8 hop_limit,
4738 u8 traffic_class)
4739{
4740 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4741
4742 attr->ah_flags = IB_AH_GRH;
4743 if (dgid)
4744 grh->dgid = *dgid;
4745 grh->flow_label = flow_label;
4746 grh->sgid_index = sgid_index;
4747 grh->hop_limit = hop_limit;
4748 grh->traffic_class = traffic_class;
4749 grh->sgid_attr = NULL;
4750}
4751
4752void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4753void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4754 u32 flow_label, u8 hop_limit, u8 traffic_class,
4755 const struct ib_gid_attr *sgid_attr);
4756void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4757 const struct rdma_ah_attr *src);
4758void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4759 const struct rdma_ah_attr *new);
4760void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4761
4762/**
4763 * rdma_ah_find_type - Return address handle type.
4764 *
4765 * @dev: Device to be checked
4766 * @port_num: Port number
4767 */
4768static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4769 u32 port_num)
4770{
4771 if (rdma_protocol_roce(device: dev, port_num))
4772 return RDMA_AH_ATTR_TYPE_ROCE;
4773 if (rdma_protocol_ib(device: dev, port_num)) {
4774 if (rdma_cap_opa_ah(device: dev, port_num))
4775 return RDMA_AH_ATTR_TYPE_OPA;
4776 return RDMA_AH_ATTR_TYPE_IB;
4777 }
4778 if (dev->type == RDMA_DEVICE_TYPE_SMI)
4779 return RDMA_AH_ATTR_TYPE_IB;
4780
4781 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4782}
4783
4784/**
4785 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4786 * In the current implementation the only way to
4787 * get the 32bit lid is from other sources for OPA.
4788 * For IB, lids will always be 16bits so cast the
4789 * value accordingly.
4790 *
4791 * @lid: A 32bit LID
4792 */
4793static inline u16 ib_lid_cpu16(u32 lid)
4794{
4795 WARN_ON_ONCE(lid & 0xFFFF0000);
4796 return (u16)lid;
4797}
4798
4799/**
4800 * ib_lid_be16 - Return lid in 16bit BE encoding.
4801 *
4802 * @lid: A 32bit LID
4803 */
4804static inline __be16 ib_lid_be16(u32 lid)
4805{
4806 WARN_ON_ONCE(lid & 0xFFFF0000);
4807 return cpu_to_be16((u16)lid);
4808}
4809
4810/**
4811 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4812 * vector
4813 * @device: the rdma device
4814 * @comp_vector: index of completion vector
4815 *
4816 * Returns NULL on failure, otherwise a corresponding cpu map of the
4817 * completion vector (returns all-cpus map if the device driver doesn't
4818 * implement get_vector_affinity).
4819 */
4820static inline const struct cpumask *
4821ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4822{
4823 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4824 !device->ops.get_vector_affinity)
4825 return NULL;
4826
4827 return device->ops.get_vector_affinity(device, comp_vector);
4828
4829}
4830
4831/**
4832 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4833 * and add their gids, as needed, to the relevant RoCE devices.
4834 *
4835 * @device: the rdma device
4836 */
4837void rdma_roce_rescan_device(struct ib_device *ibdev);
4838void rdma_roce_rescan_port(struct ib_device *ib_dev, u32 port);
4839void roce_del_all_netdev_gids(struct ib_device *ib_dev,
4840 u32 port, struct net_device *ndev);
4841
4842struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4843
4844#if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
4845int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4846bool rdma_uattrs_has_raw_cap(const struct uverbs_attr_bundle *attrs);
4847#else
4848static inline int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs)
4849{
4850 return 0;
4851}
4852static inline bool
4853rdma_uattrs_has_raw_cap(const struct uverbs_attr_bundle *attrs)
4854{
4855 return false;
4856}
4857#endif
4858
4859struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num,
4860 enum rdma_netdev_t type, const char *name,
4861 unsigned char name_assign_type,
4862 void (*setup)(struct net_device *));
4863
4864int rdma_init_netdev(struct ib_device *device, u32 port_num,
4865 enum rdma_netdev_t type, const char *name,
4866 unsigned char name_assign_type,
4867 void (*setup)(struct net_device *),
4868 struct net_device *netdev);
4869
4870/**
4871 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4872 *
4873 * @device: device pointer for which ib_device pointer to retrieve
4874 *
4875 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4876 *
4877 */
4878static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4879{
4880 struct ib_core_device *coredev =
4881 container_of(device, struct ib_core_device, dev);
4882
4883 return coredev->owner;
4884}
4885
4886/**
4887 * ibdev_to_node - return the NUMA node for a given ib_device
4888 * @dev: device to get the NUMA node for.
4889 */
4890static inline int ibdev_to_node(struct ib_device *ibdev)
4891{
4892 struct device *parent = ibdev->dev.parent;
4893
4894 if (!parent)
4895 return NUMA_NO_NODE;
4896 return dev_to_node(dev: parent);
4897}
4898
4899/**
4900 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4901 * ib_device holder structure from device pointer.
4902 *
4903 * NOTE: New drivers should not make use of this API; This API is only for
4904 * existing drivers who have exposed sysfs entries using
4905 * ops->device_group.
4906 */
4907#define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4908 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4909
4910bool rdma_dev_access_netns(const struct ib_device *device,
4911 const struct net *net);
4912
4913bool rdma_dev_has_raw_cap(const struct ib_device *dev);
4914static inline struct net *rdma_dev_net(struct ib_device *device)
4915{
4916 return read_pnet(pnet: &device->coredev.rdma_net);
4917}
4918
4919#define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4920#define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4921#define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4922
4923/**
4924 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4925 * on the flow_label
4926 *
4927 * This function will convert the 20 bit flow_label input to a valid RoCE v2
4928 * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4929 * convention.
4930 */
4931static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4932{
4933 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4934
4935 fl_low ^= fl_high >> 14;
4936 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4937}
4938
4939/**
4940 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4941 * local and remote qpn values
4942 *
4943 * This function folded the multiplication results of two qpns, 24 bit each,
4944 * fields, and converts it to a 20 bit results.
4945 *
4946 * This function will create symmetric flow_label value based on the local
4947 * and remote qpn values. this will allow both the requester and responder
4948 * to calculate the same flow_label for a given connection.
4949 *
4950 * This helper function should be used by driver in case the upper layer
4951 * provide a zero flow_label value. This is to improve entropy of RDMA
4952 * traffic in the network.
4953 */
4954static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4955{
4956 u64 v = (u64)lqpn * rqpn;
4957
4958 v ^= v >> 20;
4959 v ^= v >> 40;
4960
4961 return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4962}
4963
4964/**
4965 * rdma_get_udp_sport - Calculate and set UDP source port based on the flow
4966 * label. If flow label is not defined in GRH then
4967 * calculate it based on lqpn/rqpn.
4968 *
4969 * @fl: flow label from GRH
4970 * @lqpn: local qp number
4971 * @rqpn: remote qp number
4972 */
4973static inline u16 rdma_get_udp_sport(u32 fl, u32 lqpn, u32 rqpn)
4974{
4975 if (!fl)
4976 fl = rdma_calc_flow_label(lqpn, rqpn);
4977
4978 return rdma_flow_label_to_udp_sport(fl);
4979}
4980
4981const struct ib_port_immutable*
4982ib_port_immutable_read(struct ib_device *dev, unsigned int port);
4983
4984/** ib_add_sub_device - Add a sub IB device on an existing one
4985 *
4986 * @parent: The IB device that needs to add a sub device
4987 * @type: The type of the new sub device
4988 * @name: The name of the new sub device
4989 *
4990 *
4991 * Return 0 on success, an error code otherwise
4992 */
4993int ib_add_sub_device(struct ib_device *parent,
4994 enum rdma_nl_dev_type type,
4995 const char *name);
4996
4997
4998/** ib_del_sub_device_and_put - Delect an IB sub device while holding a 'get'
4999 *
5000 * @sub: The sub device that is going to be deleted
5001 *
5002 * Return 0 on success, an error code otherwise
5003 */
5004int ib_del_sub_device_and_put(struct ib_device *sub);
5005
5006static inline void ib_mark_name_assigned_by_user(struct ib_device *ibdev)
5007{
5008 ibdev->name_assign_type = RDMA_NAME_ASSIGN_TYPE_USER;
5009}
5010
5011#endif /* IB_VERBS_H */
5012