projects/linux·blob master · refs

5207 lines · 153.8 KiB

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/bvec.h>
19#include <linux/dma-mapping.h>
20#include <linux/kref.h>
21#include <linux/list.h>
22#include <linux/rwsem.h>
23#include <linux/workqueue.h>
24#include <linux/irq_poll.h>
25#include <uapi/linux/if_ether.h>
26#include <net/ipv6.h>
27#include <net/ip.h>
28#include <linux/string.h>
29#include <linux/slab.h>
30#include <linux/netdevice.h>
31#include <linux/refcount.h>
32#include <linux/if_link.h>
33#include <linux/atomic.h>
34#include <linux/mmu_notifier.h>
35#include <linux/uaccess.h>
36#include <linux/cgroup_rdma.h>
37#include <linux/irqflags.h>
38#include <linux/preempt.h>
39#include <linux/dim.h>
40#include <uapi/rdma/ib_user_verbs.h>
41#include <rdma/rdma_counter.h>
42#include <rdma/restrack.h>
43#include <rdma/signature.h>
44#include <uapi/rdma/rdma_user_ioctl.h>
45#include <uapi/rdma/ib_user_ioctl_verbs.h>
46#include <linux/pci-tph.h>
47#include <rdma/frmr_pools.h>
48#include <linux/dma-buf.h>
49
50#define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
51
52struct ib_umem_odp;
53struct ib_uqp_object;
54struct ib_usrq_object;
55struct ib_uwq_object;
56struct rdma_cm_id;
57struct ib_port;
58struct hw_stats_device_data;
59
60extern struct workqueue_struct *ib_wq;
61extern struct workqueue_struct *ib_comp_wq;
62extern struct workqueue_struct *ib_comp_unbound_wq;
63
64struct ib_ucq_object;
65
66__printf(2, 3) __cold
67void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
68__printf(2, 3) __cold
69void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
70__printf(2, 3) __cold
71void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
72__printf(2, 3) __cold
73void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
74__printf(2, 3) __cold
75void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
76__printf(2, 3) __cold
77void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
78__printf(2, 3) __cold
79void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
80
81#if defined(CONFIG_DYNAMIC_DEBUG) || \
82 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
83#define ibdev_dbg(__dev, format, args...) \
84 dynamic_ibdev_dbg(__dev, format, ##args)
85#else
86__printf(2, 3) __cold
87static inline
88void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
89#endif
90
91#define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \
92do { \
93 static DEFINE_RATELIMIT_STATE(_rs, \
94 DEFAULT_RATELIMIT_INTERVAL, \
95 DEFAULT_RATELIMIT_BURST); \
96 if (__ratelimit(&_rs)) \
97 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \
98} while (0)
99
100#define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
101 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
102#define ibdev_alert_ratelimited(ibdev, fmt, ...) \
103 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
104#define ibdev_crit_ratelimited(ibdev, fmt, ...) \
105 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
106#define ibdev_err_ratelimited(ibdev, fmt, ...) \
107 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
108#define ibdev_warn_ratelimited(ibdev, fmt, ...) \
109 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
110#define ibdev_notice_ratelimited(ibdev, fmt, ...) \
111 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
112#define ibdev_info_ratelimited(ibdev, fmt, ...) \
113 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
114
115#if defined(CONFIG_DYNAMIC_DEBUG) || \
116 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
117/* descriptor check is first to prevent flooding with "callbacks suppressed" */
118#define ibdev_dbg_ratelimited(ibdev, fmt, ...) \
119do { \
120 static DEFINE_RATELIMIT_STATE(_rs, \
121 DEFAULT_RATELIMIT_INTERVAL, \
122 DEFAULT_RATELIMIT_BURST); \
123 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \
124 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \
125 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \
126 ##__VA_ARGS__); \
127} while (0)
128#else
129__printf(2, 3) __cold
130static inline
131void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
132#endif
133
134union ib_gid {
135 u8 raw[16];
136 struct {
137 __be64 subnet_prefix;
138 __be64 interface_id;
139 } global;
140};
141
142extern union ib_gid zgid;
143
144enum ib_gid_type {
145 IB_GID_TYPE_IB = IB_UVERBS_GID_TYPE_IB,
146 IB_GID_TYPE_ROCE = IB_UVERBS_GID_TYPE_ROCE_V1,
147 IB_GID_TYPE_ROCE_UDP_ENCAP = IB_UVERBS_GID_TYPE_ROCE_V2,
148 IB_GID_TYPE_SIZE
149};
150
151#define ROCE_V2_UDP_DPORT 4791
152struct ib_gid_attr {
153 struct net_device __rcu *ndev;
154 struct ib_device *device;
155 union ib_gid gid;
156 enum ib_gid_type gid_type;
157 u16 index;
158 u32 port_num;
159};
160
161enum {
162 /* set the local administered indication */
163 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
164};
165
166enum rdma_transport_type {
167 RDMA_TRANSPORT_IB,
168 RDMA_TRANSPORT_IWARP,
169 RDMA_TRANSPORT_USNIC,
170 RDMA_TRANSPORT_USNIC_UDP,
171 RDMA_TRANSPORT_UNSPECIFIED,
172};
173
174enum rdma_protocol_type {
175 RDMA_PROTOCOL_IB,
176 RDMA_PROTOCOL_IBOE,
177 RDMA_PROTOCOL_IWARP,
178 RDMA_PROTOCOL_USNIC_UDP
179};
180
181__attribute_const__ enum rdma_transport_type
182rdma_node_get_transport(unsigned int node_type);
183
184enum rdma_network_type {
185 RDMA_NETWORK_IB,
186 RDMA_NETWORK_ROCE_V1,
187 RDMA_NETWORK_IPV4,
188 RDMA_NETWORK_IPV6
189};
190
191static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
192{
193 if (network_type == RDMA_NETWORK_IPV4 ||
194 network_type == RDMA_NETWORK_IPV6)
195 return IB_GID_TYPE_ROCE_UDP_ENCAP;
196 else if (network_type == RDMA_NETWORK_ROCE_V1)
197 return IB_GID_TYPE_ROCE;
198 else
199 return IB_GID_TYPE_IB;
200}
201
202static inline enum rdma_network_type
203rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
204{
205 if (attr->gid_type == IB_GID_TYPE_IB)
206 return RDMA_NETWORK_IB;
207
208 if (attr->gid_type == IB_GID_TYPE_ROCE)
209 return RDMA_NETWORK_ROCE_V1;
210
211 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
212 return RDMA_NETWORK_IPV4;
213 else
214 return RDMA_NETWORK_IPV6;
215}
216
217enum rdma_link_layer {
218 IB_LINK_LAYER_UNSPECIFIED,
219 IB_LINK_LAYER_INFINIBAND,
220 IB_LINK_LAYER_ETHERNET,
221};
222
223enum ib_device_cap_flags {
224 IB_DEVICE_RESIZE_MAX_WR = IB_UVERBS_DEVICE_RESIZE_MAX_WR,
225 IB_DEVICE_BAD_PKEY_CNTR = IB_UVERBS_DEVICE_BAD_PKEY_CNTR,
226 IB_DEVICE_BAD_QKEY_CNTR = IB_UVERBS_DEVICE_BAD_QKEY_CNTR,
227 IB_DEVICE_RAW_MULTI = IB_UVERBS_DEVICE_RAW_MULTI,
228 IB_DEVICE_AUTO_PATH_MIG = IB_UVERBS_DEVICE_AUTO_PATH_MIG,
229 IB_DEVICE_CHANGE_PHY_PORT = IB_UVERBS_DEVICE_CHANGE_PHY_PORT,
230 IB_DEVICE_UD_AV_PORT_ENFORCE = IB_UVERBS_DEVICE_UD_AV_PORT_ENFORCE,
231 IB_DEVICE_CURR_QP_STATE_MOD = IB_UVERBS_DEVICE_CURR_QP_STATE_MOD,
232 IB_DEVICE_SHUTDOWN_PORT = IB_UVERBS_DEVICE_SHUTDOWN_PORT,
233 /* IB_DEVICE_INIT_TYPE = IB_UVERBS_DEVICE_INIT_TYPE, (not in use) */
234 IB_DEVICE_PORT_ACTIVE_EVENT = IB_UVERBS_DEVICE_PORT_ACTIVE_EVENT,
235 IB_DEVICE_SYS_IMAGE_GUID = IB_UVERBS_DEVICE_SYS_IMAGE_GUID,
236 IB_DEVICE_RC_RNR_NAK_GEN = IB_UVERBS_DEVICE_RC_RNR_NAK_GEN,
237 IB_DEVICE_SRQ_RESIZE = IB_UVERBS_DEVICE_SRQ_RESIZE,
238 IB_DEVICE_N_NOTIFY_CQ = IB_UVERBS_DEVICE_N_NOTIFY_CQ,
239
240 /* Reserved, old SEND_W_INV = 1 << 16,*/
241 IB_DEVICE_MEM_WINDOW = IB_UVERBS_DEVICE_MEM_WINDOW,
242 /*
243 * Devices should set IB_DEVICE_UD_IP_SUM if they support
244 * insertion of UDP and TCP checksum on outgoing UD IPoIB
245 * messages and can verify the validity of checksum for
246 * incoming messages. Setting this flag implies that the
247 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
248 */
249 IB_DEVICE_UD_IP_CSUM = IB_UVERBS_DEVICE_UD_IP_CSUM,
250 IB_DEVICE_XRC = IB_UVERBS_DEVICE_XRC,
251
252 /*
253 * This device supports the IB "base memory management extension",
254 * which includes support for fast registrations (IB_WR_REG_MR,
255 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
256 * also be set by any iWarp device which must support FRs to comply
257 * to the iWarp verbs spec. iWarp devices also support the
258 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
259 * stag.
260 */
261 IB_DEVICE_MEM_MGT_EXTENSIONS = IB_UVERBS_DEVICE_MEM_MGT_EXTENSIONS,
262 IB_DEVICE_MEM_WINDOW_TYPE_2A = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2A,
263 IB_DEVICE_MEM_WINDOW_TYPE_2B = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2B,
264 IB_DEVICE_RC_IP_CSUM = IB_UVERBS_DEVICE_RC_IP_CSUM,
265 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
266 IB_DEVICE_RAW_IP_CSUM = IB_UVERBS_DEVICE_RAW_IP_CSUM,
267 IB_DEVICE_MANAGED_FLOW_STEERING =
268 IB_UVERBS_DEVICE_MANAGED_FLOW_STEERING,
269 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
270 IB_DEVICE_RAW_SCATTER_FCS = IB_UVERBS_DEVICE_RAW_SCATTER_FCS,
271 /* The device supports padding incoming writes to cacheline. */
272 IB_DEVICE_PCI_WRITE_END_PADDING =
273 IB_UVERBS_DEVICE_PCI_WRITE_END_PADDING,
274 /* Placement type attributes */
275 IB_DEVICE_FLUSH_GLOBAL = IB_UVERBS_DEVICE_FLUSH_GLOBAL,
276 IB_DEVICE_FLUSH_PERSISTENT = IB_UVERBS_DEVICE_FLUSH_PERSISTENT,
277 IB_DEVICE_ATOMIC_WRITE = IB_UVERBS_DEVICE_ATOMIC_WRITE,
278 IB_DEVICE_CC_DMA_BOUNCE = IB_UVERBS_DEVICE_CC_DMA_BOUNCE,
279};
280
281enum ib_kernel_cap_flags {
282 /*
283 * This device supports a per-device lkey or stag that can be
284 * used without performing a memory registration for the local
285 * memory. Note that ULPs should never check this flag, but
286 * instead of use the local_dma_lkey flag in the ib_pd structure,
287 * which will always contain a usable lkey.
288 */
289 IBK_LOCAL_DMA_LKEY = 1 << 0,
290 /* IB_QP_CREATE_INTEGRITY_EN is supported to implement T10-PI */
291 IBK_INTEGRITY_HANDOVER = 1 << 1,
292 /* IB_ACCESS_ON_DEMAND is supported during reg_user_mr() */
293 IBK_ON_DEMAND_PAGING = 1 << 2,
294 /* IB_MR_TYPE_SG_GAPS is supported */
295 IBK_SG_GAPS_REG = 1 << 3,
296 /* Driver supports RDMA_NLDEV_CMD_DELLINK */
297 IBK_ALLOW_USER_UNREG = 1 << 4,
298
299 /* ipoib will use IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK */
300 IBK_BLOCK_MULTICAST_LOOPBACK = 1 << 5,
301 /* iopib will use IB_QP_CREATE_IPOIB_UD_LSO for its QPs */
302 IBK_UD_TSO = 1 << 6,
303 /* iopib will use the device ops:
304 * get_vf_config
305 * get_vf_guid
306 * get_vf_stats
307 * set_vf_guid
308 * set_vf_link_state
309 */
310 IBK_VIRTUAL_FUNCTION = 1 << 7,
311 /* ipoib will use IB_QP_CREATE_NETDEV_USE for its QPs */
312 IBK_RDMA_NETDEV_OPA = 1 << 8,
313};
314
315enum ib_atomic_cap {
316 IB_ATOMIC_NONE,
317 IB_ATOMIC_HCA,
318 IB_ATOMIC_GLOB
319};
320
321enum ib_odp_general_cap_bits {
322 IB_ODP_SUPPORT = IB_UVERBS_ODP_SUPPORT,
323 IB_ODP_SUPPORT_IMPLICIT = IB_UVERBS_ODP_SUPPORT_IMPLICIT,
324};
325
326enum ib_odp_transport_cap_bits {
327 IB_ODP_SUPPORT_SEND = IB_UVERBS_ODP_SUPPORT_SEND,
328 IB_ODP_SUPPORT_RECV = IB_UVERBS_ODP_SUPPORT_RECV,
329 IB_ODP_SUPPORT_WRITE = IB_UVERBS_ODP_SUPPORT_WRITE,
330 IB_ODP_SUPPORT_READ = IB_UVERBS_ODP_SUPPORT_READ,
331 IB_ODP_SUPPORT_ATOMIC = IB_UVERBS_ODP_SUPPORT_ATOMIC,
332 IB_ODP_SUPPORT_SRQ_RECV = IB_UVERBS_ODP_SUPPORT_SRQ_RECV,
333 IB_ODP_SUPPORT_FLUSH = IB_UVERBS_ODP_SUPPORT_FLUSH,
334 IB_ODP_SUPPORT_ATOMIC_WRITE = IB_UVERBS_ODP_SUPPORT_ATOMIC_WRITE,
335};
336
337struct ib_odp_caps {
338 uint64_t general_caps;
339 struct {
340 uint32_t rc_odp_caps;
341 uint32_t uc_odp_caps;
342 uint32_t ud_odp_caps;
343 uint32_t xrc_odp_caps;
344 } per_transport_caps;
345};
346
347struct ib_rss_caps {
348 /* Corresponding bit will be set if qp type from
349 * 'enum ib_qp_type' is supported, e.g.
350 * supported_qpts |= 1 << IB_QPT_UD
351 */
352 u32 supported_qpts;
353 u32 max_rwq_indirection_tables;
354 u32 max_rwq_indirection_table_size;
355};
356
357enum ib_tm_cap_flags {
358 /* Support tag matching with rendezvous offload for RC transport */
359 IB_TM_CAP_RNDV_RC = 1 << 0,
360};
361
362struct ib_tm_caps {
363 /* Max size of RNDV header */
364 u32 max_rndv_hdr_size;
365 /* Max number of entries in tag matching list */
366 u32 max_num_tags;
367 /* From enum ib_tm_cap_flags */
368 u32 flags;
369 /* Max number of outstanding list operations */
370 u32 max_ops;
371 /* Max number of SGE in tag matching entry */
372 u32 max_sge;
373};
374
375struct ib_cq_init_attr {
376 unsigned int cqe;
377 u32 comp_vector;
378 u32 flags;
379};
380
381enum ib_cq_attr_mask {
382 IB_CQ_MODERATE = 1 << 0,
383};
384
385struct ib_cq_caps {
386 u16 max_cq_moderation_count;
387 u16 max_cq_moderation_period;
388};
389
390struct ib_dm_mr_attr {
391 u64 length;
392 u64 offset;
393 u32 access_flags;
394};
395
396struct ib_dm_alloc_attr {
397 u64 length;
398 u32 alignment;
399 u32 flags;
400};
401
402struct ib_device_attr {
403 u64 fw_ver;
404 __be64 sys_image_guid;
405 u64 max_mr_size;
406 u64 page_size_cap;
407 u32 vendor_id;
408 u32 vendor_part_id;
409 u32 hw_ver;
410 u32 max_qp;
411 u32 max_qp_wr;
412 u64 device_cap_flags;
413 u64 kernel_cap_flags;
414 u32 max_send_sge;
415 u32 max_recv_sge;
416 u32 max_sge_rd;
417 u32 max_cq;
418 u32 max_cqe;
419 u32 max_mr;
420 u32 max_pd;
421 u32 max_qp_rd_atom;
422 u32 max_ee_rd_atom;
423 u32 max_res_rd_atom;
424 u32 max_qp_init_rd_atom;
425 u32 max_ee_init_rd_atom;
426 enum ib_atomic_cap atomic_cap;
427 enum ib_atomic_cap masked_atomic_cap;
428 u32 max_ee;
429 u32 max_rdd;
430 u32 max_mw;
431 u32 max_raw_ipv6_qp;
432 u32 max_raw_ethy_qp;
433 u32 max_mcast_grp;
434 u32 max_mcast_qp_attach;
435 u32 max_total_mcast_qp_attach;
436 u32 max_ah;
437 u32 max_srq;
438 u32 max_srq_wr;
439 u32 max_srq_sge;
440 unsigned int max_fast_reg_page_list_len;
441 unsigned int max_pi_fast_reg_page_list_len;
442 u16 max_pkeys;
443 u8 local_ca_ack_delay;
444 int sig_prot_cap;
445 int sig_guard_cap;
446 struct ib_odp_caps odp_caps;
447 uint64_t timestamp_mask;
448 uint64_t hca_core_clock; /* in KHZ */
449 struct ib_rss_caps rss_caps;
450 u32 max_wq_type_rq;
451 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
452 struct ib_tm_caps tm_caps;
453 struct ib_cq_caps cq_caps;
454 u64 max_dm_size;
455 /* Max entries for sgl for optimized performance per READ */
456 u32 max_sgl_rd;
457};
458
459enum ib_mtu {
460 IB_MTU_256 = 1,
461 IB_MTU_512 = 2,
462 IB_MTU_1024 = 3,
463 IB_MTU_2048 = 4,
464 IB_MTU_4096 = 5
465};
466
467enum opa_mtu {
468 OPA_MTU_8192 = 6,
469 OPA_MTU_10240 = 7
470};
471
472static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
473{
474 switch (mtu) {
475 case IB_MTU_256: return 256;
476 case IB_MTU_512: return 512;
477 case IB_MTU_1024: return 1024;
478 case IB_MTU_2048: return 2048;
479 case IB_MTU_4096: return 4096;
480 default: return -1;
481 }
482}
483
484static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
485{
486 if (mtu >= 4096)
487 return IB_MTU_4096;
488 else if (mtu >= 2048)
489 return IB_MTU_2048;
490 else if (mtu >= 1024)
491 return IB_MTU_1024;
492 else if (mtu >= 512)
493 return IB_MTU_512;
494 else
495 return IB_MTU_256;
496}
497
498static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
499{
500 switch (mtu) {
501 case OPA_MTU_8192:
502 return 8192;
503 case OPA_MTU_10240:
504 return 10240;
505 default:
506 return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
507 }
508}
509
510static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
511{
512 if (mtu >= 10240)
513 return OPA_MTU_10240;
514 else if (mtu >= 8192)
515 return OPA_MTU_8192;
516 else
517 return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
518}
519
520enum ib_port_state {
521 IB_PORT_NOP = 0,
522 IB_PORT_DOWN = 1,
523 IB_PORT_INIT = 2,
524 IB_PORT_ARMED = 3,
525 IB_PORT_ACTIVE = 4,
526 IB_PORT_ACTIVE_DEFER = 5
527};
528
529static inline const char *__attribute_const__
530ib_port_state_to_str(enum ib_port_state state)
531{
532 const char * const states[] = {
533 [IB_PORT_NOP] = "NOP",
534 [IB_PORT_DOWN] = "DOWN",
535 [IB_PORT_INIT] = "INIT",
536 [IB_PORT_ARMED] = "ARMED",
537 [IB_PORT_ACTIVE] = "ACTIVE",
538 [IB_PORT_ACTIVE_DEFER] = "ACTIVE_DEFER",
539 };
540
541 if (state < ARRAY_SIZE(states))
542 return states[state];
543 return "UNKNOWN";
544}
545
546enum ib_port_phys_state {
547 IB_PORT_PHYS_STATE_SLEEP = 1,
548 IB_PORT_PHYS_STATE_POLLING = 2,
549 IB_PORT_PHYS_STATE_DISABLED = 3,
550 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
551 IB_PORT_PHYS_STATE_LINK_UP = 5,
552 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
553 IB_PORT_PHYS_STATE_PHY_TEST = 7,
554};
555
556enum ib_port_width {
557 IB_WIDTH_1X = 1,
558 IB_WIDTH_2X = 16,
559 IB_WIDTH_4X = 2,
560 IB_WIDTH_8X = 4,
561 IB_WIDTH_12X = 8
562};
563
564static inline int ib_width_enum_to_int(enum ib_port_width width)
565{
566 switch (width) {
567 case IB_WIDTH_1X: return 1;
568 case IB_WIDTH_2X: return 2;
569 case IB_WIDTH_4X: return 4;
570 case IB_WIDTH_8X: return 8;
571 case IB_WIDTH_12X: return 12;
572 default: return -1;
573 }
574}
575
576enum ib_port_speed {
577 IB_SPEED_SDR = 1,
578 IB_SPEED_DDR = 2,
579 IB_SPEED_QDR = 4,
580 IB_SPEED_FDR10 = 8,
581 IB_SPEED_FDR = 16,
582 IB_SPEED_EDR = 32,
583 IB_SPEED_HDR = 64,
584 IB_SPEED_NDR = 128,
585 IB_SPEED_XDR = 256,
586};
587
588enum ib_stat_flag {
589 IB_STAT_FLAG_OPTIONAL = 1 << 0,
590};
591
592/**
593 * struct rdma_stat_desc - description of one rdma stat/counter
594 * @name: The name of the counter
595 * @flags: Flags of the counter; For example, IB_STAT_FLAG_OPTIONAL
596 * @priv: Driver private information; Core code should not use
597 */
598struct rdma_stat_desc {
599 const char *name;
600 unsigned int flags;
601 const void *priv;
602};
603
604/**
605 * struct rdma_hw_stats - collection of hardware stats and their management
606 * @lock: Mutex to protect parallel write access to lifespan and values
607 * of counters, which are 64bits and not guaranteed to be written
608 * atomicaly on 32bits systems.
609 * @timestamp: Used by the core code to track when the last update was
610 * @lifespan: Used by the core code to determine how old the counters
611 * should be before being updated again. Stored in jiffies, defaults
612 * to 10 milliseconds, drivers can override the default be specifying
613 * their own value during their allocation routine.
614 * @descs: Array of pointers to static descriptors used for the counters
615 * in directory.
616 * @is_disabled: A bitmap to indicate each counter is currently disabled
617 * or not.
618 * @num_counters: How many hardware counters there are. If name is
619 * shorter than this number, a kernel oops will result. Driver authors
620 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
621 * in their code to prevent this.
622 * @value: Array of u64 counters that are accessed by the sysfs code and
623 * filled in by the drivers get_stats routine
624 */
625struct rdma_hw_stats {
626 struct mutex lock; /* Protect lifespan and values[] */
627 unsigned long timestamp;
628 unsigned long lifespan;
629 const struct rdma_stat_desc *descs;
630 unsigned long *is_disabled;
631 int num_counters;
632 u64 value[] __counted_by(num_counters);
633};
634
635#define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
636
637struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
638 const struct rdma_stat_desc *descs, int num_counters,
639 unsigned long lifespan);
640
641void rdma_free_hw_stats_struct(struct rdma_hw_stats *stats);
642
643/* Define bits for the various functionality this port needs to be supported by
644 * the core.
645 */
646/* Management 0x00000FFF */
647#define RDMA_CORE_CAP_IB_MAD 0x00000001
648#define RDMA_CORE_CAP_IB_SMI 0x00000002
649#define RDMA_CORE_CAP_IB_CM 0x00000004
650#define RDMA_CORE_CAP_IW_CM 0x00000008
651#define RDMA_CORE_CAP_IB_SA 0x00000010
652#define RDMA_CORE_CAP_OPA_MAD 0x00000020
653
654/* Address format 0x000FF000 */
655#define RDMA_CORE_CAP_AF_IB 0x00001000
656#define RDMA_CORE_CAP_ETH_AH 0x00002000
657#define RDMA_CORE_CAP_OPA_AH 0x00004000
658#define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
659
660/* Protocol 0xFFF00000 */
661#define RDMA_CORE_CAP_PROT_IB 0x00100000
662#define RDMA_CORE_CAP_PROT_ROCE 0x00200000
663#define RDMA_CORE_CAP_PROT_IWARP 0x00400000
664#define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
665#define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
666#define RDMA_CORE_CAP_PROT_USNIC 0x02000000
667
668#define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
669 | RDMA_CORE_CAP_PROT_ROCE \
670 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
671
672#define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
673 | RDMA_CORE_CAP_IB_MAD \
674 | RDMA_CORE_CAP_IB_SMI \
675 | RDMA_CORE_CAP_IB_CM \
676 | RDMA_CORE_CAP_IB_SA \
677 | RDMA_CORE_CAP_AF_IB)
678#define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
679 | RDMA_CORE_CAP_IB_MAD \
680 | RDMA_CORE_CAP_IB_CM \
681 | RDMA_CORE_CAP_AF_IB \
682 | RDMA_CORE_CAP_ETH_AH)
683#define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
684 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
685 | RDMA_CORE_CAP_IB_MAD \
686 | RDMA_CORE_CAP_IB_CM \
687 | RDMA_CORE_CAP_AF_IB \
688 | RDMA_CORE_CAP_ETH_AH)
689#define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
690 | RDMA_CORE_CAP_IW_CM)
691#define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
692 | RDMA_CORE_CAP_OPA_MAD)
693
694#define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
695
696#define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
697
698struct ib_port_attr {
699 u64 subnet_prefix;
700 enum ib_port_state state;
701 enum ib_mtu max_mtu;
702 enum ib_mtu active_mtu;
703 u32 phys_mtu;
704 int gid_tbl_len;
705 unsigned int ip_gids:1;
706 /* This is the value from PortInfo CapabilityMask, defined by IBA */
707 u32 port_cap_flags;
708 u32 max_msg_sz;
709 u32 bad_pkey_cntr;
710 u32 qkey_viol_cntr;
711 u16 pkey_tbl_len;
712 u32 sm_lid;
713 u32 lid;
714 u8 lmc;
715 u8 max_vl_num;
716 u8 sm_sl;
717 u8 subnet_timeout;
718 u8 init_type_reply;
719 u8 active_width;
720 u16 active_speed;
721 u8 phys_state;
722 u16 port_cap_flags2;
723};
724
725enum ib_device_modify_flags {
726 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
727 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
728};
729
730#define IB_DEVICE_NODE_DESC_MAX 64
731
732struct ib_device_modify {
733 u64 sys_image_guid;
734 char node_desc[IB_DEVICE_NODE_DESC_MAX];
735};
736
737enum ib_port_modify_flags {
738 IB_PORT_SHUTDOWN = 1,
739 IB_PORT_INIT_TYPE = (1<<2),
740 IB_PORT_RESET_QKEY_CNTR = (1<<3),
741 IB_PORT_OPA_MASK_CHG = (1<<4)
742};
743
744struct ib_port_modify {
745 u32 set_port_cap_mask;
746 u32 clr_port_cap_mask;
747 u8 init_type;
748};
749
750enum ib_event_type {
751 IB_EVENT_CQ_ERR,
752 IB_EVENT_QP_FATAL,
753 IB_EVENT_QP_REQ_ERR,
754 IB_EVENT_QP_ACCESS_ERR,
755 IB_EVENT_COMM_EST,
756 IB_EVENT_SQ_DRAINED,
757 IB_EVENT_PATH_MIG,
758 IB_EVENT_PATH_MIG_ERR,
759 IB_EVENT_DEVICE_FATAL,
760 IB_EVENT_PORT_ACTIVE,
761 IB_EVENT_PORT_ERR,
762 IB_EVENT_LID_CHANGE,
763 IB_EVENT_PKEY_CHANGE,
764 IB_EVENT_SM_CHANGE,
765 IB_EVENT_SRQ_ERR,
766 IB_EVENT_SRQ_LIMIT_REACHED,
767 IB_EVENT_QP_LAST_WQE_REACHED,
768 IB_EVENT_CLIENT_REREGISTER,
769 IB_EVENT_GID_CHANGE,
770 IB_EVENT_WQ_FATAL,
771 IB_EVENT_DEVICE_SPEED_CHANGE,
772};
773
774const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
775
776struct ib_event {
777 struct ib_device *device;
778 union {
779 struct ib_cq *cq;
780 struct ib_qp *qp;
781 struct ib_srq *srq;
782 struct ib_wq *wq;
783 u32 port_num;
784 } element;
785 enum ib_event_type event;
786};
787
788struct ib_event_handler {
789 struct ib_device *device;
790 void (*handler)(struct ib_event_handler *, struct ib_event *);
791 struct list_head list;
792};
793
794#define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
795 do { \
796 (_ptr)->device = _device; \
797 (_ptr)->handler = _handler; \
798 INIT_LIST_HEAD(&(_ptr)->list); \
799 } while (0)
800
801struct ib_global_route {
802 const struct ib_gid_attr *sgid_attr;
803 union ib_gid dgid;
804 u32 flow_label;
805 u8 sgid_index;
806 u8 hop_limit;
807 u8 traffic_class;
808};
809
810struct ib_grh {
811 __be32 version_tclass_flow;
812 __be16 paylen;
813 u8 next_hdr;
814 u8 hop_limit;
815 union ib_gid sgid;
816 union ib_gid dgid;
817};
818
819union rdma_network_hdr {
820 struct ib_grh ibgrh;
821 struct {
822 /* The IB spec states that if it's IPv4, the header
823 * is located in the last 20 bytes of the header.
824 */
825 u8 reserved[20];
826 struct iphdr roce4grh;
827 };
828};
829
830#define IB_QPN_MASK 0xFFFFFF
831
832enum {
833 IB_MULTICAST_QPN = 0xffffff
834};
835
836#define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
837#define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
838
839enum ib_ah_flags {
840 IB_AH_GRH = 1
841};
842
843enum ib_rate {
844 IB_RATE_PORT_CURRENT = 0,
845 IB_RATE_2_5_GBPS = 2,
846 IB_RATE_5_GBPS = 5,
847 IB_RATE_10_GBPS = 3,
848 IB_RATE_20_GBPS = 6,
849 IB_RATE_30_GBPS = 4,
850 IB_RATE_40_GBPS = 7,
851 IB_RATE_60_GBPS = 8,
852 IB_RATE_80_GBPS = 9,
853 IB_RATE_120_GBPS = 10,
854 IB_RATE_14_GBPS = 11,
855 IB_RATE_56_GBPS = 12,
856 IB_RATE_112_GBPS = 13,
857 IB_RATE_168_GBPS = 14,
858 IB_RATE_25_GBPS = 15,
859 IB_RATE_100_GBPS = 16,
860 IB_RATE_200_GBPS = 17,
861 IB_RATE_300_GBPS = 18,
862 IB_RATE_28_GBPS = 19,
863 IB_RATE_50_GBPS = 20,
864 IB_RATE_400_GBPS = 21,
865 IB_RATE_600_GBPS = 22,
866 IB_RATE_800_GBPS = 23,
867 IB_RATE_1600_GBPS = 25,
868};
869
870/**
871 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
872 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
873 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
874 * @rate: rate to convert.
875 */
876__attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
877
878/**
879 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
880 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
881 * @rate: rate to convert.
882 */
883__attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
884
885struct ib_port_speed_info {
886 const char *str;
887 int rate; /* in deci-Gb/sec (100 MBps units) */
888};
889
890/**
891 * ib_port_attr_to_speed_info - Convert port attributes to speed information
892 * @attr: Port attributes containing active_speed and active_width
893 * @speed_info: Speed information to return
894 *
895 * Returns 0 on success, -EINVAL on error.
896 */
897int ib_port_attr_to_speed_info(struct ib_port_attr *attr,
898 struct ib_port_speed_info *speed_info);
899
900/**
901 * enum ib_mr_type - memory region type
902 * @IB_MR_TYPE_MEM_REG: memory region that is used for
903 * normal registration
904 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
905 * register any arbitrary sg lists (without
906 * the normal mr constraints - see
907 * ib_map_mr_sg)
908 * @IB_MR_TYPE_DM: memory region that is used for device
909 * memory registration
910 * @IB_MR_TYPE_USER: memory region that is used for the user-space
911 * application
912 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
913 * without address translations (VA=PA)
914 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
915 * data integrity operations
916 */
917enum ib_mr_type {
918 IB_MR_TYPE_MEM_REG,
919 IB_MR_TYPE_SG_GAPS,
920 IB_MR_TYPE_DM,
921 IB_MR_TYPE_USER,
922 IB_MR_TYPE_DMA,
923 IB_MR_TYPE_INTEGRITY,
924};
925
926enum ib_mr_status_check {
927 IB_MR_CHECK_SIG_STATUS = 1,
928};
929
930/**
931 * struct ib_mr_status - Memory region status container
932 *
933 * @fail_status: Bitmask of MR checks status. For each
934 * failed check a corresponding status bit is set.
935 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
936 * failure.
937 */
938struct ib_mr_status {
939 u32 fail_status;
940 struct ib_sig_err sig_err;
941};
942
943/**
944 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
945 * enum.
946 * @mult: multiple to convert.
947 */
948__attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
949
950struct rdma_ah_init_attr {
951 struct rdma_ah_attr *ah_attr;
952 u32 flags;
953 struct net_device *xmit_slave;
954};
955
956enum rdma_ah_attr_type {
957 RDMA_AH_ATTR_TYPE_UNDEFINED,
958 RDMA_AH_ATTR_TYPE_IB,
959 RDMA_AH_ATTR_TYPE_ROCE,
960 RDMA_AH_ATTR_TYPE_OPA,
961};
962
963struct ib_ah_attr {
964 u16 dlid;
965 u8 src_path_bits;
966};
967
968struct roce_ah_attr {
969 u8 dmac[ETH_ALEN];
970};
971
972struct opa_ah_attr {
973 u32 dlid;
974 u8 src_path_bits;
975 bool make_grd;
976};
977
978struct rdma_ah_attr {
979 struct ib_global_route grh;
980 u8 sl;
981 u8 static_rate;
982 u32 port_num;
983 u8 ah_flags;
984 enum rdma_ah_attr_type type;
985 union {
986 struct ib_ah_attr ib;
987 struct roce_ah_attr roce;
988 struct opa_ah_attr opa;
989 };
990};
991
992enum ib_wc_status {
993 IB_WC_SUCCESS,
994 IB_WC_LOC_LEN_ERR,
995 IB_WC_LOC_QP_OP_ERR,
996 IB_WC_LOC_EEC_OP_ERR,
997 IB_WC_LOC_PROT_ERR,
998 IB_WC_WR_FLUSH_ERR,
999 IB_WC_MW_BIND_ERR,
1000 IB_WC_BAD_RESP_ERR,
1001 IB_WC_LOC_ACCESS_ERR,
1002 IB_WC_REM_INV_REQ_ERR,
1003 IB_WC_REM_ACCESS_ERR,
1004 IB_WC_REM_OP_ERR,
1005 IB_WC_RETRY_EXC_ERR,
1006 IB_WC_RNR_RETRY_EXC_ERR,
1007 IB_WC_LOC_RDD_VIOL_ERR,
1008 IB_WC_REM_INV_RD_REQ_ERR,
1009 IB_WC_REM_ABORT_ERR,
1010 IB_WC_INV_EECN_ERR,
1011 IB_WC_INV_EEC_STATE_ERR,
1012 IB_WC_FATAL_ERR,
1013 IB_WC_RESP_TIMEOUT_ERR,
1014 IB_WC_GENERAL_ERR
1015};
1016
1017const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
1018
1019enum ib_wc_opcode {
1020 IB_WC_SEND = IB_UVERBS_WC_SEND,
1021 IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE,
1022 IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ,
1023 IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP,
1024 IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD,
1025 IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW,
1026 IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV,
1027 IB_WC_LSO = IB_UVERBS_WC_TSO,
1028 IB_WC_ATOMIC_WRITE = IB_UVERBS_WC_ATOMIC_WRITE,
1029 IB_WC_REG_MR,
1030 IB_WC_MASKED_COMP_SWAP,
1031 IB_WC_MASKED_FETCH_ADD,
1032 IB_WC_FLUSH = IB_UVERBS_WC_FLUSH,
1033/*
1034 * Set value of IB_WC_RECV so consumers can test if a completion is a
1035 * receive by testing (opcode & IB_WC_RECV).
1036 */
1037 IB_WC_RECV = 1 << 7,
1038 IB_WC_RECV_RDMA_WITH_IMM
1039};
1040
1041enum ib_wc_flags {
1042 IB_WC_GRH = 1,
1043 IB_WC_WITH_IMM = (1<<1),
1044 IB_WC_WITH_INVALIDATE = (1<<2),
1045 IB_WC_IP_CSUM_OK = (1<<3),
1046 IB_WC_WITH_SMAC = (1<<4),
1047 IB_WC_WITH_VLAN = (1<<5),
1048 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
1049};
1050
1051struct ib_wc {
1052 union {
1053 u64 wr_id;
1054 struct ib_cqe *wr_cqe;
1055 };
1056 enum ib_wc_status status;
1057 enum ib_wc_opcode opcode;
1058 u32 vendor_err;
1059 u32 byte_len;
1060 struct ib_qp *qp;
1061 union {
1062 __be32 imm_data;
1063 u32 invalidate_rkey;
1064 } ex;
1065 u32 src_qp;
1066 u32 slid;
1067 int wc_flags;
1068 u16 pkey_index;
1069 u8 sl;
1070 u8 dlid_path_bits;
1071 u32 port_num; /* valid only for DR SMPs on switches */
1072 u8 smac[ETH_ALEN];
1073 u16 vlan_id;
1074 u8 network_hdr_type;
1075};
1076
1077enum ib_cq_notify_flags {
1078 IB_CQ_SOLICITED = 1 << 0,
1079 IB_CQ_NEXT_COMP = 1 << 1,
1080 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1081 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1082};
1083
1084enum ib_srq_type {
1085 IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1086 IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1087 IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1088};
1089
1090static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1091{
1092 return srq_type == IB_SRQT_XRC ||
1093 srq_type == IB_SRQT_TM;
1094}
1095
1096enum ib_srq_attr_mask {
1097 IB_SRQ_MAX_WR = 1 << 0,
1098 IB_SRQ_LIMIT = 1 << 1,
1099};
1100
1101struct ib_srq_attr {
1102 u32 max_wr;
1103 u32 max_sge;
1104 u32 srq_limit;
1105};
1106
1107struct ib_srq_init_attr {
1108 void (*event_handler)(struct ib_event *, void *);
1109 void *srq_context;
1110 struct ib_srq_attr attr;
1111 enum ib_srq_type srq_type;
1112
1113 struct {
1114 struct ib_cq *cq;
1115 union {
1116 struct {
1117 struct ib_xrcd *xrcd;
1118 } xrc;
1119
1120 struct {
1121 u32 max_num_tags;
1122 } tag_matching;
1123 };
1124 } ext;
1125};
1126
1127struct ib_qp_cap {
1128 u32 max_send_wr;
1129 u32 max_recv_wr;
1130 u32 max_send_sge;
1131 u32 max_recv_sge;
1132 u32 max_inline_data;
1133
1134 /*
1135 * Maximum number of rdma_rw_ctx structures in flight at a time.
1136 * ib_create_qp() will calculate the right amount of needed WRs
1137 * and MRs based on this.
1138 */
1139 u32 max_rdma_ctxs;
1140};
1141
1142enum ib_sig_type {
1143 IB_SIGNAL_ALL_WR,
1144 IB_SIGNAL_REQ_WR
1145};
1146
1147enum ib_qp_type {
1148 /*
1149 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1150 * here (and in that order) since the MAD layer uses them as
1151 * indices into a 2-entry table.
1152 */
1153 IB_QPT_SMI,
1154 IB_QPT_GSI,
1155
1156 IB_QPT_RC = IB_UVERBS_QPT_RC,
1157 IB_QPT_UC = IB_UVERBS_QPT_UC,
1158 IB_QPT_UD = IB_UVERBS_QPT_UD,
1159 IB_QPT_RAW_IPV6,
1160 IB_QPT_RAW_ETHERTYPE,
1161 IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1162 IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1163 IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1164 IB_QPT_MAX,
1165 IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1166 /* Reserve a range for qp types internal to the low level driver.
1167 * These qp types will not be visible at the IB core layer, so the
1168 * IB_QPT_MAX usages should not be affected in the core layer
1169 */
1170 IB_QPT_RESERVED1 = 0x1000,
1171 IB_QPT_RESERVED2,
1172 IB_QPT_RESERVED3,
1173 IB_QPT_RESERVED4,
1174 IB_QPT_RESERVED5,
1175 IB_QPT_RESERVED6,
1176 IB_QPT_RESERVED7,
1177 IB_QPT_RESERVED8,
1178 IB_QPT_RESERVED9,
1179 IB_QPT_RESERVED10,
1180};
1181
1182enum ib_qp_create_flags {
1183 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1184 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK =
1185 IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1186 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1187 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1188 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1189 IB_QP_CREATE_NETIF_QP = 1 << 5,
1190 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
1191 IB_QP_CREATE_NETDEV_USE = 1 << 7,
1192 IB_QP_CREATE_SCATTER_FCS =
1193 IB_UVERBS_QP_CREATE_SCATTER_FCS,
1194 IB_QP_CREATE_CVLAN_STRIPPING =
1195 IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1196 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1197 IB_QP_CREATE_PCI_WRITE_END_PADDING =
1198 IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1199 /* reserve bits 26-31 for low level drivers' internal use */
1200 IB_QP_CREATE_RESERVED_START = 1 << 26,
1201 IB_QP_CREATE_RESERVED_END = 1 << 31,
1202};
1203
1204/*
1205 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1206 * callback to destroy the passed in QP.
1207 */
1208
1209struct ib_qp_init_attr {
1210 /* This callback occurs in workqueue context */
1211 void (*event_handler)(struct ib_event *, void *);
1212
1213 void *qp_context;
1214 struct ib_cq *send_cq;
1215 struct ib_cq *recv_cq;
1216 struct ib_srq *srq;
1217 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1218 struct ib_qp_cap cap;
1219 enum ib_sig_type sq_sig_type;
1220 enum ib_qp_type qp_type;
1221 u32 create_flags;
1222
1223 /*
1224 * Only needed for special QP types, or when using the RW API.
1225 */
1226 u32 port_num;
1227 struct ib_rwq_ind_table *rwq_ind_tbl;
1228 u32 source_qpn;
1229};
1230
1231struct ib_qp_open_attr {
1232 void (*event_handler)(struct ib_event *, void *);
1233 void *qp_context;
1234 u32 qp_num;
1235 enum ib_qp_type qp_type;
1236};
1237
1238enum ib_rnr_timeout {
1239 IB_RNR_TIMER_655_36 = 0,
1240 IB_RNR_TIMER_000_01 = 1,
1241 IB_RNR_TIMER_000_02 = 2,
1242 IB_RNR_TIMER_000_03 = 3,
1243 IB_RNR_TIMER_000_04 = 4,
1244 IB_RNR_TIMER_000_06 = 5,
1245 IB_RNR_TIMER_000_08 = 6,
1246 IB_RNR_TIMER_000_12 = 7,
1247 IB_RNR_TIMER_000_16 = 8,
1248 IB_RNR_TIMER_000_24 = 9,
1249 IB_RNR_TIMER_000_32 = 10,
1250 IB_RNR_TIMER_000_48 = 11,
1251 IB_RNR_TIMER_000_64 = 12,
1252 IB_RNR_TIMER_000_96 = 13,
1253 IB_RNR_TIMER_001_28 = 14,
1254 IB_RNR_TIMER_001_92 = 15,
1255 IB_RNR_TIMER_002_56 = 16,
1256 IB_RNR_TIMER_003_84 = 17,
1257 IB_RNR_TIMER_005_12 = 18,
1258 IB_RNR_TIMER_007_68 = 19,
1259 IB_RNR_TIMER_010_24 = 20,
1260 IB_RNR_TIMER_015_36 = 21,
1261 IB_RNR_TIMER_020_48 = 22,
1262 IB_RNR_TIMER_030_72 = 23,
1263 IB_RNR_TIMER_040_96 = 24,
1264 IB_RNR_TIMER_061_44 = 25,
1265 IB_RNR_TIMER_081_92 = 26,
1266 IB_RNR_TIMER_122_88 = 27,
1267 IB_RNR_TIMER_163_84 = 28,
1268 IB_RNR_TIMER_245_76 = 29,
1269 IB_RNR_TIMER_327_68 = 30,
1270 IB_RNR_TIMER_491_52 = 31
1271};
1272
1273enum ib_qp_attr_mask {
1274 IB_QP_STATE = 1,
1275 IB_QP_CUR_STATE = (1<<1),
1276 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1277 IB_QP_ACCESS_FLAGS = (1<<3),
1278 IB_QP_PKEY_INDEX = (1<<4),
1279 IB_QP_PORT = (1<<5),
1280 IB_QP_QKEY = (1<<6),
1281 IB_QP_AV = (1<<7),
1282 IB_QP_PATH_MTU = (1<<8),
1283 IB_QP_TIMEOUT = (1<<9),
1284 IB_QP_RETRY_CNT = (1<<10),
1285 IB_QP_RNR_RETRY = (1<<11),
1286 IB_QP_RQ_PSN = (1<<12),
1287 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1288 IB_QP_ALT_PATH = (1<<14),
1289 IB_QP_MIN_RNR_TIMER = (1<<15),
1290 IB_QP_SQ_PSN = (1<<16),
1291 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1292 IB_QP_PATH_MIG_STATE = (1<<18),
1293 IB_QP_CAP = (1<<19),
1294 IB_QP_DEST_QPN = (1<<20),
1295 IB_QP_RESERVED1 = (1<<21),
1296 IB_QP_RESERVED2 = (1<<22),
1297 IB_QP_RESERVED3 = (1<<23),
1298 IB_QP_RESERVED4 = (1<<24),
1299 IB_QP_RATE_LIMIT = (1<<25),
1300
1301 IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0),
1302};
1303
1304enum ib_qp_state {
1305 IB_QPS_RESET,
1306 IB_QPS_INIT,
1307 IB_QPS_RTR,
1308 IB_QPS_RTS,
1309 IB_QPS_SQD,
1310 IB_QPS_SQE,
1311 IB_QPS_ERR
1312};
1313
1314enum ib_mig_state {
1315 IB_MIG_MIGRATED,
1316 IB_MIG_REARM,
1317 IB_MIG_ARMED
1318};
1319
1320enum ib_mw_type {
1321 IB_MW_TYPE_1 = 1,
1322 IB_MW_TYPE_2 = 2
1323};
1324
1325struct ib_qp_attr {
1326 enum ib_qp_state qp_state;
1327 enum ib_qp_state cur_qp_state;
1328 enum ib_mtu path_mtu;
1329 enum ib_mig_state path_mig_state;
1330 u32 qkey;
1331 u32 rq_psn;
1332 u32 sq_psn;
1333 u32 dest_qp_num;
1334 int qp_access_flags;
1335 struct ib_qp_cap cap;
1336 struct rdma_ah_attr ah_attr;
1337 struct rdma_ah_attr alt_ah_attr;
1338 u16 pkey_index;
1339 u16 alt_pkey_index;
1340 u8 en_sqd_async_notify;
1341 u8 sq_draining;
1342 u8 max_rd_atomic;
1343 u8 max_dest_rd_atomic;
1344 u8 min_rnr_timer;
1345 u32 port_num;
1346 u8 timeout;
1347 u8 retry_cnt;
1348 u8 rnr_retry;
1349 u32 alt_port_num;
1350 u8 alt_timeout;
1351 u32 rate_limit;
1352 struct net_device *xmit_slave;
1353};
1354
1355enum ib_wr_opcode {
1356 /* These are shared with userspace */
1357 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1358 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1359 IB_WR_SEND = IB_UVERBS_WR_SEND,
1360 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1361 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1362 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1363 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1364 IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW,
1365 IB_WR_LSO = IB_UVERBS_WR_TSO,
1366 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1367 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1368 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1369 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1370 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1371 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1372 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1373 IB_WR_FLUSH = IB_UVERBS_WR_FLUSH,
1374 IB_WR_ATOMIC_WRITE = IB_UVERBS_WR_ATOMIC_WRITE,
1375
1376 /* These are kernel only and can not be issued by userspace */
1377 IB_WR_REG_MR = 0x20,
1378 IB_WR_REG_MR_INTEGRITY,
1379
1380 /* reserve values for low level drivers' internal use.
1381 * These values will not be used at all in the ib core layer.
1382 */
1383 IB_WR_RESERVED1 = 0xf0,
1384 IB_WR_RESERVED2,
1385 IB_WR_RESERVED3,
1386 IB_WR_RESERVED4,
1387 IB_WR_RESERVED5,
1388 IB_WR_RESERVED6,
1389 IB_WR_RESERVED7,
1390 IB_WR_RESERVED8,
1391 IB_WR_RESERVED9,
1392 IB_WR_RESERVED10,
1393};
1394
1395enum ib_send_flags {
1396 IB_SEND_FENCE = 1,
1397 IB_SEND_SIGNALED = (1<<1),
1398 IB_SEND_SOLICITED = (1<<2),
1399 IB_SEND_INLINE = (1<<3),
1400 IB_SEND_IP_CSUM = (1<<4),
1401
1402 /* reserve bits 26-31 for low level drivers' internal use */
1403 IB_SEND_RESERVED_START = (1 << 26),
1404 IB_SEND_RESERVED_END = (1 << 31),
1405};
1406
1407struct ib_sge {
1408 u64 addr;
1409 u32 length;
1410 u32 lkey;
1411};
1412
1413struct ib_cqe {
1414 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1415};
1416
1417struct ib_send_wr {
1418 struct ib_send_wr *next;
1419 union {
1420 u64 wr_id;
1421 struct ib_cqe *wr_cqe;
1422 };
1423 struct ib_sge *sg_list;
1424 int num_sge;
1425 enum ib_wr_opcode opcode;
1426 int send_flags;
1427 union {
1428 __be32 imm_data;
1429 u32 invalidate_rkey;
1430 } ex;
1431};
1432
1433struct ib_rdma_wr {
1434 struct ib_send_wr wr;
1435 u64 remote_addr;
1436 u32 rkey;
1437};
1438
1439static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1440{
1441 return container_of(wr, struct ib_rdma_wr, wr);
1442}
1443
1444struct ib_atomic_wr {
1445 struct ib_send_wr wr;
1446 u64 remote_addr;
1447 u64 compare_add;
1448 u64 swap;
1449 u64 compare_add_mask;
1450 u64 swap_mask;
1451 u32 rkey;
1452};
1453
1454static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1455{
1456 return container_of(wr, struct ib_atomic_wr, wr);
1457}
1458
1459struct ib_ud_wr {
1460 struct ib_send_wr wr;
1461 struct ib_ah *ah;
1462 void *header;
1463 int hlen;
1464 int mss;
1465 u32 remote_qpn;
1466 u32 remote_qkey;
1467 u16 pkey_index; /* valid for GSI only */
1468 u32 port_num; /* valid for DR SMPs on switch only */
1469};
1470
1471static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1472{
1473 return container_of(wr, struct ib_ud_wr, wr);
1474}
1475
1476struct ib_reg_wr {
1477 struct ib_send_wr wr;
1478 struct ib_mr *mr;
1479 u32 key;
1480 int access;
1481};
1482
1483static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1484{
1485 return container_of(wr, struct ib_reg_wr, wr);
1486}
1487
1488struct ib_recv_wr {
1489 struct ib_recv_wr *next;
1490 union {
1491 u64 wr_id;
1492 struct ib_cqe *wr_cqe;
1493 };
1494 struct ib_sge *sg_list;
1495 int num_sge;
1496};
1497
1498enum ib_access_flags {
1499 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1500 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1501 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1502 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1503 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1504 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1505 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1506 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1507 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1508 IB_ACCESS_FLUSH_GLOBAL = IB_UVERBS_ACCESS_FLUSH_GLOBAL,
1509 IB_ACCESS_FLUSH_PERSISTENT = IB_UVERBS_ACCESS_FLUSH_PERSISTENT,
1510
1511 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1512 IB_ACCESS_SUPPORTED =
1513 ((IB_ACCESS_FLUSH_PERSISTENT << 1) - 1) | IB_ACCESS_OPTIONAL,
1514};
1515
1516/*
1517 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1518 * are hidden here instead of a uapi header!
1519 */
1520enum ib_mr_rereg_flags {
1521 IB_MR_REREG_TRANS = 1,
1522 IB_MR_REREG_PD = (1<<1),
1523 IB_MR_REREG_ACCESS = (1<<2),
1524 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1525};
1526
1527struct ib_umem;
1528
1529enum rdma_remove_reason {
1530 /*
1531 * Userspace requested uobject deletion or initial try
1532 * to remove uobject via cleanup. Call could fail
1533 */
1534 RDMA_REMOVE_DESTROY,
1535 /* Context deletion. This call should delete the actual object itself */
1536 RDMA_REMOVE_CLOSE,
1537 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1538 RDMA_REMOVE_DRIVER_REMOVE,
1539 /* uobj is being cleaned-up before being committed */
1540 RDMA_REMOVE_ABORT,
1541 /* The driver failed to destroy the uobject and is being disconnected */
1542 RDMA_REMOVE_DRIVER_FAILURE,
1543};
1544
1545struct ib_rdmacg_object {
1546#ifdef CONFIG_CGROUP_RDMA
1547 struct rdma_cgroup *cg; /* owner rdma cgroup */
1548#endif
1549};
1550
1551struct ib_ucontext {
1552 struct ib_device *device;
1553 struct ib_uverbs_file *ufile;
1554
1555 struct ib_rdmacg_object cg_obj;
1556 u64 enabled_caps;
1557 /*
1558 * Implementation details of the RDMA core, don't use in drivers:
1559 */
1560 struct rdma_restrack_entry res;
1561 struct xarray mmap_xa;
1562};
1563
1564struct ib_uobject {
1565 u64 user_handle; /* handle given to us by userspace */
1566 /* ufile & ucontext owning this object */
1567 struct ib_uverbs_file *ufile;
1568 /* FIXME, save memory: ufile->context == context */
1569 struct ib_ucontext *context; /* associated user context */
1570 void *object; /* containing object */
1571 struct list_head list; /* link to context's list */
1572 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1573 int id; /* index into kernel idr */
1574 struct kref ref;
1575 atomic_t usecnt; /* protects exclusive access */
1576 struct rcu_head rcu; /* kfree_rcu() overhead */
1577
1578 const struct uverbs_api_object *uapi_object;
1579};
1580
1581/**
1582 * struct ib_udata - Driver request/response data from userspace
1583 * @inbuf: Pointer to request data from userspace
1584 * @outbuf: Pointer to response buffer in userspace
1585 * @inlen: Length of request data
1586 * @outlen: Length of response buffer
1587 *
1588 * struct ib_udata is used to hold the driver data request and response
1589 * structures defined in the uapi. They follow these rules for forwards and
1590 * backwards compatibility:
1591 *
1592 * 1) Userspace can provide a longer request so long as the trailing part the
1593 * kernel doesn't understand is all zeros.
1594 *
1595 * This provides a degree of safety if userspace wrongly tries to use a new
1596 * feature the kernel does not understand with some non-zero value.
1597 *
1598 * It allows a simpler rdma-core implementation because the library can
1599 * simply always use the latest structs for the request, even if they are
1600 * bigger. It simply has to avoid using the new members if they are not
1601 * supported/required.
1602 *
1603 * 2) Userspace can provide a shorter request; the kernel will zero-pad it out
1604 * to fill the storage. The newer kernel should understand that older
1605 * userspace will provide 0 to new fields. The kernel has three options to
1606 * enable new request fields:
1607 *
1608 * - Input comp_mask that says the field is supported
1609 * - Look for non-zero values
1610 * - Check if the udata->inlen size covers the field
1611 *
1612 * This also corrects any bugs related to not filling in request structures
1613 * as the new helper always fully writes to the struct.
1614 *
1615 * 3) Userspace can provide a shorter or longer response struct. If shorter,
1616 * the kernel reply is truncated. The kernel should be designed to not write
1617 * to new reply fields unless userspace has affirmatively requested them.
1618 *
1619 * If the user buffer is longer, the kernel will zero-fill it.
1620 *
1621 * Userspace has three options to enable new response fields:
1622 *
1623 * - Output comp_mask that says the field is supported
1624 * - Look for non-zero values
1625 * - Infer the output must be valid because the request contents demand it
1626 * and old kernels will fail the request
1627 *
1628 * The following helper functions implement these semantics:
1629 *
1630 * ib_copy_validate_udata_in() - Checks the minimum length, and zero trailing::
1631 *
1632 * struct driver_create_cq_req req;
1633 * int err;
1634 *
1635 * err = ib_copy_validate_udata_in(udata, req, end_member);
1636 * if (err)
1637 * return err;
1638 *
1639 * The third argument specifies the last member of the struct in the first
1640 * kernel version that introduced it, establishing the minimum required size.
1641 *
1642 * ib_copy_validate_udata_in_cm() - The above but also validate a
1643 * comp_mask member only has supported bits set::
1644 *
1645 * err = ib_copy_validate_udata_in_cm(udata, req, first_version_last_member,
1646 * DRIVER_CREATE_CQ_MASK_FEATURE_A |
1647 * DRIVER_CREATE_CQ_MASK_FEATURE_B);
1648 *
1649 * ib_respond_udata() - Implements the response rules::
1650 *
1651 * struct driver_create_cq_resp resp = {};
1652 *
1653 * resp.some_field = value;
1654 * return ib_respond_udata(udata, resp);
1655 *
1656 * ib_is_udata_in_empty() - Used instead of ib_copy_validate_udata_in() if the
1657 * driver does not have a request structure::
1658 *
1659 * ret = ib_is_udata_in_empty(udata);
1660 * if (ret)
1661 * return ret;
1662 *
1663 * Similarly ib_respond_empty_udata() is used instead of ib_respond_udata() if
1664 * the driver does not have a response structure::
1665 *
1666 * return ib_respond_empty_udata(udata);
1667 */
1668struct ib_udata {
1669 const void __user *inbuf;
1670 void __user *outbuf;
1671 size_t inlen;
1672 size_t outlen;
1673};
1674
1675struct ib_pd {
1676 u32 local_dma_lkey;
1677 u32 flags;
1678 struct ib_device *device;
1679 struct ib_uobject *uobject;
1680 atomic_t usecnt; /* count all resources */
1681
1682 u32 unsafe_global_rkey;
1683
1684 /*
1685 * Implementation details of the RDMA core, don't use in drivers:
1686 */
1687 struct ib_mr *__internal_mr;
1688 struct rdma_restrack_entry res;
1689};
1690
1691struct ib_xrcd {
1692 struct ib_device *device;
1693 atomic_t usecnt; /* count all exposed resources */
1694 struct inode *inode;
1695 struct rw_semaphore tgt_qps_rwsem;
1696 struct xarray tgt_qps;
1697};
1698
1699struct ib_ah {
1700 struct ib_device *device;
1701 struct ib_pd *pd;
1702 struct ib_uobject *uobject;
1703 const struct ib_gid_attr *sgid_attr;
1704 enum rdma_ah_attr_type type;
1705};
1706
1707typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1708
1709enum ib_poll_context {
1710 IB_POLL_SOFTIRQ, /* poll from softirq context */
1711 IB_POLL_WORKQUEUE, /* poll from workqueue */
1712 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1713 IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1714
1715 IB_POLL_DIRECT, /* caller context, no hw completions */
1716};
1717
1718struct ib_cq {
1719 struct ib_device *device;
1720 struct ib_ucq_object *uobject;
1721 ib_comp_handler comp_handler;
1722 void (*event_handler)(struct ib_event *, void *);
1723 void *cq_context;
1724 int cqe;
1725 unsigned int cqe_used;
1726 atomic_t usecnt; /* count number of work queues */
1727 enum ib_poll_context poll_ctx;
1728 struct ib_wc *wc;
1729 struct list_head pool_entry;
1730 union {
1731 struct irq_poll iop;
1732 struct work_struct work;
1733 };
1734 struct workqueue_struct *comp_wq;
1735 struct dim *dim;
1736
1737 /* updated only by trace points */
1738 ktime_t timestamp;
1739 u8 interrupt:1;
1740 u8 shared:1;
1741 unsigned int comp_vector;
1742
1743 /*
1744 * Implementation details of the RDMA core, don't use in drivers:
1745 */
1746 struct rdma_restrack_entry res;
1747};
1748
1749enum ib_qp_attach_comp_cntr_op {
1750 IB_QP_ATTACH_COMP_CNTR_OP_SEND = IB_UVERBS_QP_ATTACH_COMP_CNTR_OP_SEND,
1751 IB_QP_ATTACH_COMP_CNTR_OP_RECV = IB_UVERBS_QP_ATTACH_COMP_CNTR_OP_RECV,
1752 IB_QP_ATTACH_COMP_CNTR_OP_RDMA_READ = IB_UVERBS_QP_ATTACH_COMP_CNTR_OP_RDMA_READ,
1753 IB_QP_ATTACH_COMP_CNTR_OP_REMOTE_RDMA_READ = IB_UVERBS_QP_ATTACH_COMP_CNTR_OP_REMOTE_RDMA_READ,
1754 IB_QP_ATTACH_COMP_CNTR_OP_RDMA_WRITE = IB_UVERBS_QP_ATTACH_COMP_CNTR_OP_RDMA_WRITE,
1755 IB_QP_ATTACH_COMP_CNTR_OP_REMOTE_RDMA_WRITE = IB_UVERBS_QP_ATTACH_COMP_CNTR_OP_REMOTE_RDMA_WRITE,
1756};
1757
1758struct ib_comp_cntr_caps {
1759 u64 max_value;
1760 u32 max_counters;
1761 u32 supported_qp_attach_ops; /* Bitmask of enum ib_qp_attach_comp_cntr_op */
1762};
1763
1764struct ib_comp_cntr {
1765 struct ib_device *device;
1766 struct ib_uobject *uobject;
1767 atomic_t usecnt;
1768 struct rdma_restrack_entry res;
1769};
1770
1771enum ib_comp_cntr_entry {
1772 IB_COMP_CNTR_ENTRY_COMP = IB_UVERBS_COMP_CNTR_ENTRY_COMP,
1773 IB_COMP_CNTR_ENTRY_ERR = IB_UVERBS_COMP_CNTR_ENTRY_ERR,
1774};
1775
1776enum ib_comp_cntr_modify_op {
1777 IB_COMP_CNTR_MODIFY_OP_SET = IB_UVERBS_COMP_CNTR_MODIFY_OP_SET,
1778 IB_COMP_CNTR_MODIFY_OP_INC = IB_UVERBS_COMP_CNTR_MODIFY_OP_INC,
1779};
1780
1781struct ib_qp_attach_comp_cntr_attr {
1782 u32 op_mask; /* Bitmask of enum ib_qp_attach_comp_cntr_op */
1783};
1784
1785struct ib_srq {
1786 struct ib_device *device;
1787 struct ib_pd *pd;
1788 struct ib_usrq_object *uobject;
1789 void (*event_handler)(struct ib_event *, void *);
1790 void *srq_context;
1791 enum ib_srq_type srq_type;
1792 atomic_t usecnt;
1793
1794 struct {
1795 struct ib_cq *cq;
1796 union {
1797 struct {
1798 struct ib_xrcd *xrcd;
1799 u32 srq_num;
1800 } xrc;
1801 };
1802 } ext;
1803
1804 /*
1805 * Implementation details of the RDMA core, don't use in drivers:
1806 */
1807 struct rdma_restrack_entry res;
1808};
1809
1810enum ib_raw_packet_caps {
1811 /*
1812 * Strip cvlan from incoming packet and report it in the matching work
1813 * completion is supported.
1814 */
1815 IB_RAW_PACKET_CAP_CVLAN_STRIPPING =
1816 IB_UVERBS_RAW_PACKET_CAP_CVLAN_STRIPPING,
1817 /*
1818 * Scatter FCS field of an incoming packet to host memory is supported.
1819 */
1820 IB_RAW_PACKET_CAP_SCATTER_FCS = IB_UVERBS_RAW_PACKET_CAP_SCATTER_FCS,
1821 /* Checksum offloads are supported (for both send and receive). */
1822 IB_RAW_PACKET_CAP_IP_CSUM = IB_UVERBS_RAW_PACKET_CAP_IP_CSUM,
1823 /*
1824 * When a packet is received for an RQ with no receive WQEs, the
1825 * packet processing is delayed.
1826 */
1827 IB_RAW_PACKET_CAP_DELAY_DROP = IB_UVERBS_RAW_PACKET_CAP_DELAY_DROP,
1828};
1829
1830enum ib_wq_type {
1831 IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1832};
1833
1834enum ib_wq_state {
1835 IB_WQS_RESET,
1836 IB_WQS_RDY,
1837 IB_WQS_ERR
1838};
1839
1840struct ib_wq {
1841 struct ib_device *device;
1842 struct ib_uwq_object *uobject;
1843 void *wq_context;
1844 void (*event_handler)(struct ib_event *, void *);
1845 struct ib_pd *pd;
1846 struct ib_cq *cq;
1847 u32 wq_num;
1848 enum ib_wq_state state;
1849 enum ib_wq_type wq_type;
1850 atomic_t usecnt;
1851};
1852
1853enum ib_wq_flags {
1854 IB_WQ_FLAGS_CVLAN_STRIPPING = IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1855 IB_WQ_FLAGS_SCATTER_FCS = IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1856 IB_WQ_FLAGS_DELAY_DROP = IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1857 IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1858 IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1859};
1860
1861struct ib_wq_init_attr {
1862 void *wq_context;
1863 enum ib_wq_type wq_type;
1864 u32 max_wr;
1865 u32 max_sge;
1866 struct ib_cq *cq;
1867 void (*event_handler)(struct ib_event *, void *);
1868 u32 create_flags; /* Use enum ib_wq_flags */
1869};
1870
1871enum ib_wq_attr_mask {
1872 IB_WQ_STATE = 1 << 0,
1873 IB_WQ_CUR_STATE = 1 << 1,
1874 IB_WQ_FLAGS = 1 << 2,
1875};
1876
1877struct ib_wq_attr {
1878 enum ib_wq_state wq_state;
1879 enum ib_wq_state curr_wq_state;
1880 u32 flags; /* Use enum ib_wq_flags */
1881 u32 flags_mask; /* Use enum ib_wq_flags */
1882};
1883
1884struct ib_rwq_ind_table {
1885 struct ib_device *device;
1886 struct ib_uobject *uobject;
1887 atomic_t usecnt;
1888 u32 ind_tbl_num;
1889 u32 log_ind_tbl_size;
1890 struct ib_wq **ind_tbl;
1891};
1892
1893struct ib_rwq_ind_table_init_attr {
1894 u32 log_ind_tbl_size;
1895 /* Each entry is a pointer to Receive Work Queue */
1896 struct ib_wq **ind_tbl;
1897};
1898
1899enum port_pkey_state {
1900 IB_PORT_PKEY_NOT_VALID = 0,
1901 IB_PORT_PKEY_VALID = 1,
1902 IB_PORT_PKEY_LISTED = 2,
1903};
1904
1905struct ib_qp_security;
1906
1907struct ib_port_pkey {
1908 enum port_pkey_state state;
1909 u16 pkey_index;
1910 u32 port_num;
1911 struct list_head qp_list;
1912 struct list_head to_error_list;
1913 struct ib_qp_security *sec;
1914};
1915
1916struct ib_ports_pkeys {
1917 struct ib_port_pkey main;
1918 struct ib_port_pkey alt;
1919};
1920
1921struct ib_qp_security {
1922 struct ib_qp *qp;
1923 struct ib_device *dev;
1924 /* Hold this mutex when changing port and pkey settings. */
1925 struct mutex mutex;
1926 struct ib_ports_pkeys *ports_pkeys;
1927 /* A list of all open shared QP handles. Required to enforce security
1928 * properly for all users of a shared QP.
1929 */
1930 struct list_head shared_qp_list;
1931 void *security;
1932 bool destroying;
1933 atomic_t error_list_count;
1934 struct completion error_complete;
1935 int error_comps_pending;
1936};
1937
1938/*
1939 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1940 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1941 */
1942struct ib_qp {
1943 struct ib_device *device;
1944 struct ib_pd *pd;
1945 struct ib_cq *send_cq;
1946 struct ib_cq *recv_cq;
1947 spinlock_t mr_lock;
1948 int mrs_used;
1949 struct list_head rdma_mrs;
1950 struct list_head sig_mrs;
1951 struct ib_srq *srq;
1952 struct completion srq_completion;
1953 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1954 struct list_head xrcd_list;
1955 struct xarray comp_cntrs; /* op_mask -> comp_cntr */
1956 u32 comp_cntr_op_mask;
1957
1958 /* count times opened, mcast attaches, flow attaches */
1959 atomic_t usecnt;
1960 struct list_head open_list;
1961 struct ib_qp *real_qp;
1962 struct ib_uqp_object *uobject;
1963 void (*event_handler)(struct ib_event *, void *);
1964 void (*registered_event_handler)(struct ib_event *, void *);
1965 void *qp_context;
1966 /* sgid_attrs associated with the AV's */
1967 const struct ib_gid_attr *av_sgid_attr;
1968 const struct ib_gid_attr *alt_path_sgid_attr;
1969 u32 qp_num;
1970 u32 max_write_sge;
1971 u32 max_read_sge;
1972 enum ib_qp_type qp_type;
1973 struct ib_rwq_ind_table *rwq_ind_tbl;
1974 struct ib_qp_security *qp_sec;
1975 u32 port;
1976
1977 bool integrity_en;
1978 /*
1979 * Implementation details of the RDMA core, don't use in drivers:
1980 */
1981 struct rdma_restrack_entry res;
1982
1983 /* The counter the qp is bind to */
1984 struct rdma_counter *counter;
1985};
1986
1987struct ib_dm {
1988 struct ib_device *device;
1989 u32 length;
1990 u32 flags;
1991 struct ib_uobject *uobject;
1992 atomic_t usecnt;
1993};
1994
1995/* bit values to mark existence of ib_dmah fields */
1996enum {
1997 IB_DMAH_CPU_ID_EXISTS,
1998 IB_DMAH_MEM_TYPE_EXISTS,
1999 IB_DMAH_PH_EXISTS,
2000};
2001
2002struct ib_dmah {
2003 struct ib_device *device;
2004 struct ib_uobject *uobject;
2005 /*
2006 * Implementation details of the RDMA core, don't use in drivers:
2007 */
2008 struct rdma_restrack_entry res;
2009 u32 cpu_id;
2010 enum tph_mem_type mem_type;
2011 atomic_t usecnt;
2012 u8 ph;
2013 u8 valid_fields; /* use IB_DMAH_XXX_EXISTS */
2014};
2015
2016struct ib_mr {
2017 struct ib_device *device;
2018 /*
2019 * Due to IB_MR_REREG_PD pd is not a fixed pointer and can change. For a
2020 * user MR, this value should only be read from a system call that holds
2021 * the uobject lock, or the driver should disable in-place REREG_PD.
2022 */
2023 struct ib_pd *pd;
2024 u32 lkey;
2025 u32 rkey;
2026 u64 iova;
2027 u64 length;
2028 unsigned int page_size;
2029 enum ib_mr_type type;
2030 bool need_inval;
2031 union {
2032 struct ib_uobject *uobject; /* user */
2033 struct list_head qp_entry; /* FR */
2034 };
2035
2036 struct ib_dm *dm;
2037 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
2038 struct ib_dmah *dmah;
2039 struct {
2040 struct ib_frmr_pool *pool;
2041 struct ib_frmr_key key;
2042 u32 handle;
2043 } frmr;
2044 /*
2045 * Implementation details of the RDMA core, don't use in drivers:
2046 */
2047 struct rdma_restrack_entry res;
2048};
2049
2050struct ib_mw {
2051 struct ib_device *device;
2052 struct ib_pd *pd;
2053 struct ib_uobject *uobject;
2054 u32 rkey;
2055 enum ib_mw_type type;
2056};
2057
2058/* Supported steering options */
2059enum ib_flow_attr_type {
2060 /* steering according to rule specifications */
2061 IB_FLOW_ATTR_NORMAL = 0x0,
2062 /* default unicast and multicast rule -
2063 * receive all Eth traffic which isn't steered to any QP
2064 */
2065 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
2066 /* default multicast rule -
2067 * receive all Eth multicast traffic which isn't steered to any QP
2068 */
2069 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
2070 /* sniffer rule - receive all port traffic */
2071 IB_FLOW_ATTR_SNIFFER = 0x3
2072};
2073
2074/* Supported steering header types */
2075enum ib_flow_spec_type {
2076 /* L2 headers*/
2077 IB_FLOW_SPEC_ETH = 0x20,
2078 IB_FLOW_SPEC_IB = 0x22,
2079 /* L3 header*/
2080 IB_FLOW_SPEC_IPV4 = 0x30,
2081 IB_FLOW_SPEC_IPV6 = 0x31,
2082 IB_FLOW_SPEC_ESP = 0x34,
2083 /* L4 headers*/
2084 IB_FLOW_SPEC_TCP = 0x40,
2085 IB_FLOW_SPEC_UDP = 0x41,
2086 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
2087 IB_FLOW_SPEC_GRE = 0x51,
2088 IB_FLOW_SPEC_MPLS = 0x60,
2089 IB_FLOW_SPEC_INNER = 0x100,
2090 /* Actions */
2091 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
2092 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
2093 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
2094 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
2095};
2096#define IB_FLOW_SPEC_LAYER_MASK 0xF0
2097#define IB_FLOW_SPEC_SUPPORT_LAYERS 10
2098
2099enum ib_flow_flags {
2100 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
2101 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
2102 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
2103};
2104
2105struct ib_flow_eth_filter {
2106 u8 dst_mac[6];
2107 u8 src_mac[6];
2108 __be16 ether_type;
2109 __be16 vlan_tag;
2110};
2111
2112struct ib_flow_spec_eth {
2113 u32 type;
2114 u16 size;
2115 struct ib_flow_eth_filter val;
2116 struct ib_flow_eth_filter mask;
2117};
2118
2119struct ib_flow_ib_filter {
2120 __be16 dlid;
2121 __u8 sl;
2122};
2123
2124struct ib_flow_spec_ib {
2125 u32 type;
2126 u16 size;
2127 struct ib_flow_ib_filter val;
2128 struct ib_flow_ib_filter mask;
2129};
2130
2131/* IPv4 header flags */
2132enum ib_ipv4_flags {
2133 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
2134 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
2135 last have this flag set */
2136};
2137
2138struct ib_flow_ipv4_filter {
2139 __be32 src_ip;
2140 __be32 dst_ip;
2141 u8 proto;
2142 u8 tos;
2143 u8 ttl;
2144 u8 flags;
2145};
2146
2147struct ib_flow_spec_ipv4 {
2148 u32 type;
2149 u16 size;
2150 struct ib_flow_ipv4_filter val;
2151 struct ib_flow_ipv4_filter mask;
2152};
2153
2154struct ib_flow_ipv6_filter {
2155 u8 src_ip[16];
2156 u8 dst_ip[16];
2157 __be32 flow_label;
2158 u8 next_hdr;
2159 u8 traffic_class;
2160 u8 hop_limit;
2161} __packed;
2162
2163struct ib_flow_spec_ipv6 {
2164 u32 type;
2165 u16 size;
2166 struct ib_flow_ipv6_filter val;
2167 struct ib_flow_ipv6_filter mask;
2168};
2169
2170struct ib_flow_tcp_udp_filter {
2171 __be16 dst_port;
2172 __be16 src_port;
2173};
2174
2175struct ib_flow_spec_tcp_udp {
2176 u32 type;
2177 u16 size;
2178 struct ib_flow_tcp_udp_filter val;
2179 struct ib_flow_tcp_udp_filter mask;
2180};
2181
2182struct ib_flow_tunnel_filter {
2183 __be32 tunnel_id;
2184};
2185
2186/* ib_flow_spec_tunnel describes the Vxlan tunnel
2187 * the tunnel_id from val has the vni value
2188 */
2189struct ib_flow_spec_tunnel {
2190 u32 type;
2191 u16 size;
2192 struct ib_flow_tunnel_filter val;
2193 struct ib_flow_tunnel_filter mask;
2194};
2195
2196struct ib_flow_esp_filter {
2197 __be32 spi;
2198 __be32 seq;
2199};
2200
2201struct ib_flow_spec_esp {
2202 u32 type;
2203 u16 size;
2204 struct ib_flow_esp_filter val;
2205 struct ib_flow_esp_filter mask;
2206};
2207
2208struct ib_flow_gre_filter {
2209 __be16 c_ks_res0_ver;
2210 __be16 protocol;
2211 __be32 key;
2212};
2213
2214struct ib_flow_spec_gre {
2215 u32 type;
2216 u16 size;
2217 struct ib_flow_gre_filter val;
2218 struct ib_flow_gre_filter mask;
2219};
2220
2221struct ib_flow_mpls_filter {
2222 __be32 tag;
2223};
2224
2225struct ib_flow_spec_mpls {
2226 u32 type;
2227 u16 size;
2228 struct ib_flow_mpls_filter val;
2229 struct ib_flow_mpls_filter mask;
2230};
2231
2232struct ib_flow_spec_action_tag {
2233 enum ib_flow_spec_type type;
2234 u16 size;
2235 u32 tag_id;
2236};
2237
2238struct ib_flow_spec_action_drop {
2239 enum ib_flow_spec_type type;
2240 u16 size;
2241};
2242
2243struct ib_flow_spec_action_handle {
2244 enum ib_flow_spec_type type;
2245 u16 size;
2246 struct ib_flow_action *act;
2247};
2248
2249enum ib_counters_description {
2250 IB_COUNTER_PACKETS,
2251 IB_COUNTER_BYTES,
2252};
2253
2254struct ib_flow_spec_action_count {
2255 enum ib_flow_spec_type type;
2256 u16 size;
2257 struct ib_counters *counters;
2258};
2259
2260union ib_flow_spec {
2261 struct {
2262 u32 type;
2263 u16 size;
2264 };
2265 struct ib_flow_spec_eth eth;
2266 struct ib_flow_spec_ib ib;
2267 struct ib_flow_spec_ipv4 ipv4;
2268 struct ib_flow_spec_tcp_udp tcp_udp;
2269 struct ib_flow_spec_ipv6 ipv6;
2270 struct ib_flow_spec_tunnel tunnel;
2271 struct ib_flow_spec_esp esp;
2272 struct ib_flow_spec_gre gre;
2273 struct ib_flow_spec_mpls mpls;
2274 struct ib_flow_spec_action_tag flow_tag;
2275 struct ib_flow_spec_action_drop drop;
2276 struct ib_flow_spec_action_handle action;
2277 struct ib_flow_spec_action_count flow_count;
2278};
2279
2280struct ib_flow_attr {
2281 enum ib_flow_attr_type type;
2282 u16 size;
2283 u16 priority;
2284 u32 flags;
2285 u8 num_of_specs;
2286 u32 port;
2287 union ib_flow_spec flows[];
2288};
2289
2290struct ib_flow {
2291 struct ib_qp *qp;
2292 struct ib_device *device;
2293 struct ib_uobject *uobject;
2294};
2295
2296enum ib_flow_action_type {
2297 IB_FLOW_ACTION_UNSPECIFIED,
2298 IB_FLOW_ACTION_ESP = 1,
2299};
2300
2301struct ib_flow_action_attrs_esp_keymats {
2302 enum ib_uverbs_flow_action_esp_keymat protocol;
2303 union {
2304 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2305 } keymat;
2306};
2307
2308struct ib_flow_action_attrs_esp_replays {
2309 enum ib_uverbs_flow_action_esp_replay protocol;
2310 union {
2311 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2312 } replay;
2313};
2314
2315enum ib_flow_action_attrs_esp_flags {
2316 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2317 * This is done in order to share the same flags between user-space and
2318 * kernel and spare an unnecessary translation.
2319 */
2320
2321 /* Kernel flags */
2322 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2323 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2324};
2325
2326struct ib_flow_spec_list {
2327 struct ib_flow_spec_list *next;
2328 union ib_flow_spec spec;
2329};
2330
2331struct ib_flow_action_attrs_esp {
2332 struct ib_flow_action_attrs_esp_keymats *keymat;
2333 struct ib_flow_action_attrs_esp_replays *replay;
2334 struct ib_flow_spec_list *encap;
2335 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2336 * Value of 0 is a valid value.
2337 */
2338 u32 esn;
2339 u32 spi;
2340 u32 seq;
2341 u32 tfc_pad;
2342 /* Use enum ib_flow_action_attrs_esp_flags */
2343 u64 flags;
2344 u64 hard_limit_pkts;
2345};
2346
2347struct ib_flow_action {
2348 struct ib_device *device;
2349 struct ib_uobject *uobject;
2350 enum ib_flow_action_type type;
2351 atomic_t usecnt;
2352};
2353
2354struct ib_mad;
2355
2356enum ib_process_mad_flags {
2357 IB_MAD_IGNORE_MKEY = 1,
2358 IB_MAD_IGNORE_BKEY = 2,
2359 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2360};
2361
2362enum ib_mad_result {
2363 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2364 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2365 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2366 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2367};
2368
2369struct ib_port_cache {
2370 u64 subnet_prefix;
2371 struct ib_pkey_cache *pkey;
2372 struct ib_gid_table *gid;
2373 u8 lmc;
2374 enum ib_port_state port_state;
2375 enum ib_port_state last_port_state;
2376};
2377
2378struct ib_port_immutable {
2379 int pkey_tbl_len;
2380 int gid_tbl_len;
2381 u32 core_cap_flags;
2382 u32 max_mad_size;
2383};
2384
2385struct ib_port_data {
2386 struct ib_device *ib_dev;
2387
2388 struct ib_port_immutable immutable;
2389
2390 spinlock_t pkey_list_lock;
2391
2392 spinlock_t netdev_lock;
2393
2394 struct list_head pkey_list;
2395
2396 struct ib_port_cache cache;
2397
2398 struct net_device __rcu *netdev;
2399 netdevice_tracker netdev_tracker;
2400 struct hlist_node ndev_hash_link;
2401 struct rdma_port_counter port_counter;
2402 struct ib_port *sysfs;
2403};
2404
2405/* rdma netdev type - specifies protocol type */
2406enum rdma_netdev_t {
2407 RDMA_NETDEV_IPOIB,
2408};
2409
2410/**
2411 * struct rdma_netdev - rdma netdev
2412 * For cases where netstack interfacing is required.
2413 */
2414struct rdma_netdev {
2415 void *clnt_priv;
2416 struct ib_device *hca;
2417 u32 port_num;
2418 int mtu;
2419
2420 void (*free_rdma_netdev)(struct net_device *netdev);
2421
2422 /* control functions */
2423 void (*set_id)(struct net_device *netdev, int id);
2424 /* send packet */
2425 int (*send)(struct net_device *dev, struct sk_buff *skb,
2426 struct ib_ah *address, u32 dqpn);
2427 /* multicast */
2428 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2429 union ib_gid *gid, u16 mlid,
2430 int set_qkey, u32 qkey);
2431 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2432 union ib_gid *gid, u16 mlid);
2433 /* timeout */
2434 void (*tx_timeout)(struct net_device *dev, unsigned int txqueue);
2435};
2436
2437struct rdma_netdev_alloc_params {
2438 size_t sizeof_priv;
2439 unsigned int txqs;
2440 unsigned int rxqs;
2441 void *param;
2442
2443 int (*initialize_rdma_netdev)(struct ib_device *device, u32 port_num,
2444 struct net_device *netdev, void *param);
2445};
2446
2447struct ib_odp_counters {
2448 atomic64_t faults;
2449 atomic64_t faults_handled;
2450 atomic64_t invalidations;
2451 atomic64_t invalidations_handled;
2452 atomic64_t prefetch;
2453};
2454
2455struct ib_counters {
2456 struct ib_device *device;
2457 struct ib_uobject *uobject;
2458 /* num of objects attached */
2459 atomic_t usecnt;
2460};
2461
2462struct ib_counters_read_attr {
2463 u64 *counters_buff;
2464 u32 ncounters;
2465 u32 flags; /* use enum ib_read_counters_flags */
2466};
2467
2468struct uverbs_attr_bundle;
2469struct iw_cm_id;
2470struct iw_cm_conn_param;
2471
2472#define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2473 .size_##ib_struct = \
2474 (sizeof(struct drv_struct) + \
2475 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2476 BUILD_BUG_ON_ZERO( \
2477 !__same_type(((struct drv_struct *)NULL)->member, \
2478 struct ib_struct)))
2479
2480#define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2481 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2482 gfp, false))
2483
2484#define rdma_zalloc_drv_obj_numa(ib_dev, ib_type) \
2485 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2486 GFP_KERNEL, true))
2487
2488#define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2489 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2490
2491#define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2492
2493struct rdma_user_mmap_entry {
2494 struct kref ref;
2495 struct ib_ucontext *ucontext;
2496 unsigned long start_pgoff;
2497 size_t npages;
2498 bool driver_removed;
2499 /* protects access to dmabufs */
2500 struct mutex dmabufs_lock;
2501 struct list_head dmabufs;
2502};
2503
2504/* Return the offset (in bytes) the user should pass to libc's mmap() */
2505static inline u64
2506rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2507{
2508 return (u64)entry->start_pgoff << PAGE_SHIFT;
2509}
2510
2511/**
2512 * struct ib_device_ops - InfiniBand device operations
2513 * This structure defines all the InfiniBand device operations, providers will
2514 * need to define the supported operations, otherwise they will be set to null.
2515 */
2516struct ib_device_ops {
2517 struct module *owner;
2518 enum rdma_driver_id driver_id;
2519 u32 uverbs_abi_ver;
2520 unsigned int uverbs_no_driver_id_binding:1;
2521 /*
2522 * Indicates the driver checks every op accepting a udata for the
2523 * correct size on input and always handles the output using the udata
2524 * helpers.
2525 */
2526 unsigned int uverbs_robust_udata:1;
2527
2528 /*
2529 * NOTE: New drivers should not make use of device_group; instead new
2530 * device parameter should be exposed via netlink command. This
2531 * mechanism exists only for existing drivers.
2532 */
2533 const struct attribute_group *device_group;
2534 const struct attribute_group **port_groups;
2535
2536 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2537 const struct ib_send_wr **bad_send_wr);
2538 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2539 const struct ib_recv_wr **bad_recv_wr);
2540 void (*drain_rq)(struct ib_qp *qp);
2541 void (*drain_sq)(struct ib_qp *qp);
2542 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2543 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2544 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2545 int (*post_srq_recv)(struct ib_srq *srq,
2546 const struct ib_recv_wr *recv_wr,
2547 const struct ib_recv_wr **bad_recv_wr);
2548 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2549 u32 port_num, const struct ib_wc *in_wc,
2550 const struct ib_grh *in_grh,
2551 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2552 size_t *out_mad_size, u16 *out_mad_pkey_index);
2553 int (*query_device)(struct ib_device *device,
2554 struct ib_device_attr *device_attr,
2555 struct ib_udata *udata);
2556 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2557 struct ib_device_modify *device_modify);
2558 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2559 int (*query_port)(struct ib_device *device, u32 port_num,
2560 struct ib_port_attr *port_attr);
2561 int (*query_port_speed)(struct ib_device *device, u32 port_num,
2562 u64 *speed);
2563 int (*modify_port)(struct ib_device *device, u32 port_num,
2564 int port_modify_mask,
2565 struct ib_port_modify *port_modify);
2566 /*
2567 * The following mandatory functions are used only at device
2568 * registration. Keep functions such as these at the end of this
2569 * structure to avoid cache line misses when accessing struct ib_device
2570 * in fast paths.
2571 */
2572 int (*get_port_immutable)(struct ib_device *device, u32 port_num,
2573 struct ib_port_immutable *immutable);
2574 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2575 u32 port_num);
2576 /*
2577 * When calling get_netdev, the HW vendor's driver should return the
2578 * net device of device @device at port @port_num or NULL if such
2579 * a net device doesn't exist. The vendor driver should call dev_hold
2580 * on this net device. The HW vendor's device driver must guarantee
2581 * that this function returns NULL before the net device has finished
2582 * NETDEV_UNREGISTER state.
2583 */
2584 struct net_device *(*get_netdev)(struct ib_device *device,
2585 u32 port_num);
2586 /*
2587 * rdma netdev operation
2588 *
2589 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2590 * must return -EOPNOTSUPP if it doesn't support the specified type.
2591 */
2592 struct net_device *(*alloc_rdma_netdev)(
2593 struct ib_device *device, u32 port_num, enum rdma_netdev_t type,
2594 const char *name, unsigned char name_assign_type,
2595 void (*setup)(struct net_device *));
2596
2597 int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num,
2598 enum rdma_netdev_t type,
2599 struct rdma_netdev_alloc_params *params);
2600 /*
2601 * query_gid should be return GID value for @device, when @port_num
2602 * link layer is either IB or iWarp. It is no-op if @port_num port
2603 * is RoCE link layer.
2604 */
2605 int (*query_gid)(struct ib_device *device, u32 port_num, int index,
2606 union ib_gid *gid);
2607 /*
2608 * When calling add_gid, the HW vendor's driver should add the gid
2609 * of device of port at gid index available at @attr. Meta-info of
2610 * that gid (for example, the network device related to this gid) is
2611 * available at @attr. @context allows the HW vendor driver to store
2612 * extra information together with a GID entry. The HW vendor driver may
2613 * allocate memory to contain this information and store it in @context
2614 * when a new GID entry is written to. Params are consistent until the
2615 * next call of add_gid or delete_gid. The function should return 0 on
2616 * success or error otherwise. The function could be called
2617 * concurrently for different ports. This function is only called when
2618 * roce_gid_table is used.
2619 */
2620 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2621 /*
2622 * When calling del_gid, the HW vendor's driver should delete the
2623 * gid of device @device at gid index gid_index of port port_num
2624 * available in @attr.
2625 * Upon the deletion of a GID entry, the HW vendor must free any
2626 * allocated memory. The caller will clear @context afterwards.
2627 * This function is only called when roce_gid_table is used.
2628 */
2629 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2630 int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index,
2631 u16 *pkey);
2632 int (*alloc_ucontext)(struct ib_ucontext *context,
2633 struct ib_udata *udata);
2634 void (*dealloc_ucontext)(struct ib_ucontext *context);
2635 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2636 /*
2637 * This will be called once refcount of an entry in mmap_xa reaches
2638 * zero. The type of the memory that was mapped may differ between
2639 * entries and is opaque to the rdma_user_mmap interface.
2640 * Therefore needs to be implemented by the driver in mmap_free.
2641 */
2642 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2643 int (*mmap_get_pfns)(struct rdma_user_mmap_entry *entry,
2644 struct phys_vec *phys_vec,
2645 struct p2pdma_provider **provider);
2646 struct rdma_user_mmap_entry *(*pgoff_to_mmap_entry)(struct ib_ucontext *ucontext,
2647 off_t pg_off);
2648 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2649 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2650 int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2651 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2652 struct ib_udata *udata);
2653 int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2654 struct ib_udata *udata);
2655 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2656 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2657 int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2658 int (*create_srq)(struct ib_srq *srq,
2659 struct ib_srq_init_attr *srq_init_attr,
2660 struct ib_udata *udata);
2661 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2662 enum ib_srq_attr_mask srq_attr_mask,
2663 struct ib_udata *udata);
2664 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2665 int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2666 int (*create_qp)(struct ib_qp *qp, struct ib_qp_init_attr *qp_init_attr,
2667 struct ib_udata *udata);
2668 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2669 int qp_attr_mask, struct ib_udata *udata);
2670 int (*qp_attach_comp_cntr)(struct ib_qp *qp, struct ib_comp_cntr *cc,
2671 struct ib_qp_attach_comp_cntr_attr *attr);
2672 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2673 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2674 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2675 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2676 struct uverbs_attr_bundle *attrs);
2677 int (*create_user_cq)(struct ib_cq *cq,
2678 const struct ib_cq_init_attr *attr,
2679 struct uverbs_attr_bundle *attrs);
2680 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2681 int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2682 int (*resize_user_cq)(struct ib_cq *cq, unsigned int cqe,
2683 struct ib_udata *udata);
2684 /*
2685 * pre_destroy_cq - Prevent a cq from generating any new work
2686 * completions, but not free any kernel resources
2687 */
2688 int (*pre_destroy_cq)(struct ib_cq *cq);
2689 /*
2690 * post_destroy_cq - Free all kernel resources
2691 */
2692 void (*post_destroy_cq)(struct ib_cq *cq);
2693 int (*create_comp_cntr)(struct ib_comp_cntr *cc,
2694 struct uverbs_attr_bundle *attrs);
2695 int (*destroy_comp_cntr)(struct ib_comp_cntr *cc);
2696 int (*modify_comp_cntr)(struct ib_comp_cntr *cc, enum ib_comp_cntr_entry entry,
2697 enum ib_comp_cntr_modify_op op, u64 value);
2698 int (*read_comp_cntr)(struct ib_comp_cntr *cc, enum ib_comp_cntr_entry entry, u64 *value);
2699 int (*query_comp_cntr_caps)(struct ib_device *dev,
2700 struct ib_comp_cntr_caps *caps,
2701 struct uverbs_attr_bundle *attrs);
2702 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2703 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2704 u64 virt_addr, int mr_access_flags,
2705 struct ib_dmah *dmah,
2706 struct ib_udata *udata);
2707 struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset,
2708 u64 length, u64 virt_addr, int fd,
2709 int mr_access_flags,
2710 struct ib_dmah *dmah,
2711 struct uverbs_attr_bundle *attrs);
2712 struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
2713 u64 length, u64 virt_addr,
2714 int mr_access_flags, struct ib_pd *pd,
2715 struct ib_udata *udata);
2716 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2717 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2718 u32 max_num_sg);
2719 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2720 u32 max_num_data_sg,
2721 u32 max_num_meta_sg);
2722 int (*advise_mr)(struct ib_pd *pd,
2723 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2724 struct ib_sge *sg_list, u32 num_sge,
2725 struct uverbs_attr_bundle *attrs);
2726
2727 /*
2728 * Kernel users should universally support relaxed ordering (RO), as
2729 * they are designed to read data only after observing the CQE and use
2730 * the DMA API correctly.
2731 *
2732 * Some drivers implicitly enable RO if platform supports it.
2733 */
2734 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2735 unsigned int *sg_offset);
2736 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2737 struct ib_mr_status *mr_status);
2738 int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2739 int (*dealloc_mw)(struct ib_mw *mw);
2740 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2741 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2742 int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2743 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2744 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2745 struct ib_flow_attr *flow_attr,
2746 struct ib_udata *udata);
2747 int (*destroy_flow)(struct ib_flow *flow_id);
2748 int (*destroy_flow_action)(struct ib_flow_action *action);
2749 int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port,
2750 int state);
2751 int (*get_vf_config)(struct ib_device *device, int vf, u32 port,
2752 struct ifla_vf_info *ivf);
2753 int (*get_vf_stats)(struct ib_device *device, int vf, u32 port,
2754 struct ifla_vf_stats *stats);
2755 int (*get_vf_guid)(struct ib_device *device, int vf, u32 port,
2756 struct ifla_vf_guid *node_guid,
2757 struct ifla_vf_guid *port_guid);
2758 int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid,
2759 int type);
2760 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2761 struct ib_wq_init_attr *init_attr,
2762 struct ib_udata *udata);
2763 int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2764 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2765 u32 wq_attr_mask, struct ib_udata *udata);
2766 int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2767 struct ib_rwq_ind_table_init_attr *init_attr,
2768 struct ib_udata *udata);
2769 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2770 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2771 struct ib_ucontext *context,
2772 struct ib_dm_alloc_attr *attr,
2773 struct uverbs_attr_bundle *attrs);
2774 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2775 int (*alloc_dmah)(struct ib_dmah *ibdmah,
2776 struct uverbs_attr_bundle *attrs);
2777 int (*dealloc_dmah)(struct ib_dmah *dmah, struct uverbs_attr_bundle *attrs);
2778 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2779 struct ib_dm_mr_attr *attr,
2780 struct uverbs_attr_bundle *attrs);
2781 int (*create_counters)(struct ib_counters *counters,
2782 struct uverbs_attr_bundle *attrs);
2783 int (*destroy_counters)(struct ib_counters *counters);
2784 int (*read_counters)(struct ib_counters *counters,
2785 struct ib_counters_read_attr *counters_read_attr,
2786 struct uverbs_attr_bundle *attrs);
2787 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2788 int data_sg_nents, unsigned int *data_sg_offset,
2789 struct scatterlist *meta_sg, int meta_sg_nents,
2790 unsigned int *meta_sg_offset);
2791
2792 /*
2793 * alloc_hw_[device,port]_stats - Allocate a struct rdma_hw_stats and
2794 * fill in the driver initialized data. The struct is kfree()'ed by
2795 * the sysfs core when the device is removed. A lifespan of -1 in the
2796 * return struct tells the core to set a default lifespan.
2797 */
2798 struct rdma_hw_stats *(*alloc_hw_device_stats)(struct ib_device *device);
2799 struct rdma_hw_stats *(*alloc_hw_port_stats)(struct ib_device *device,
2800 u32 port_num);
2801 /*
2802 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2803 * @index - The index in the value array we wish to have updated, or
2804 * num_counters if we want all stats updated
2805 * Return codes -
2806 * < 0 - Error, no counters updated
2807 * index - Updated the single counter pointed to by index
2808 * num_counters - Updated all counters (will reset the timestamp
2809 * and prevent further calls for lifespan milliseconds)
2810 * Drivers are allowed to update all counters in leiu of just the
2811 * one given in index at their option
2812 */
2813 int (*get_hw_stats)(struct ib_device *device,
2814 struct rdma_hw_stats *stats, u32 port, int index);
2815
2816 /*
2817 * modify_hw_stat - Modify the counter configuration
2818 * @enable: true/false when enable/disable a counter
2819 * Return codes - 0 on success or error code otherwise.
2820 */
2821 int (*modify_hw_stat)(struct ib_device *device, u32 port,
2822 unsigned int counter_index, bool enable);
2823 /*
2824 * Allows rdma drivers to add their own restrack attributes.
2825 */
2826 int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2827 int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2828 int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2829 int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2830 int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2831 int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2832 int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2833 int (*fill_res_srq_entry)(struct sk_buff *msg, struct ib_srq *ib_srq);
2834 int (*fill_res_srq_entry_raw)(struct sk_buff *msg, struct ib_srq *ib_srq);
2835
2836 /* Device lifecycle callbacks */
2837 /*
2838 * Called after the device becomes registered, before clients are
2839 * attached
2840 */
2841 int (*enable_driver)(struct ib_device *dev);
2842 /*
2843 * This is called as part of ib_dealloc_device().
2844 */
2845 void (*dealloc_driver)(struct ib_device *dev);
2846
2847 /* iWarp CM callbacks */
2848 void (*iw_add_ref)(struct ib_qp *qp);
2849 void (*iw_rem_ref)(struct ib_qp *qp);
2850 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2851 int (*iw_connect)(struct iw_cm_id *cm_id,
2852 struct iw_cm_conn_param *conn_param);
2853 int (*iw_accept)(struct iw_cm_id *cm_id,
2854 struct iw_cm_conn_param *conn_param);
2855 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2856 u8 pdata_len);
2857 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2858 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2859 /*
2860 * counter_bind_qp - Bind a QP to a counter.
2861 * @counter - The counter to be bound. If counter->id is zero then
2862 * the driver needs to allocate a new counter and set counter->id
2863 */
2864 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp,
2865 u32 port);
2866 /*
2867 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2868 * counter and bind it onto the default one
2869 */
2870 int (*counter_unbind_qp)(struct ib_qp *qp, u32 port);
2871 /*
2872 * counter_dealloc -De-allocate the hw counter
2873 */
2874 int (*counter_dealloc)(struct rdma_counter *counter);
2875 /*
2876 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2877 * the driver initialized data.
2878 */
2879 struct rdma_hw_stats *(*counter_alloc_stats)(
2880 struct rdma_counter *counter);
2881 /*
2882 * counter_update_stats - Query the stats value of this counter
2883 */
2884 int (*counter_update_stats)(struct rdma_counter *counter);
2885
2886 /*
2887 * counter_init - Initialize the driver specific rdma counter struct.
2888 */
2889 void (*counter_init)(struct rdma_counter *counter);
2890
2891 /*
2892 * Allows rdma drivers to add their own restrack attributes
2893 * dumped via 'rdma stat' iproute2 command.
2894 */
2895 int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2896
2897 /* query driver for its ucontext properties */
2898 int (*query_ucontext)(struct ib_ucontext *context,
2899 struct uverbs_attr_bundle *attrs);
2900
2901 /*
2902 * Provide NUMA node. This API exists for rdmavt/hfi1 only.
2903 * Everyone else relies on Linux memory management model.
2904 */
2905 int (*get_numa_node)(struct ib_device *dev);
2906
2907 /*
2908 * add_sub_dev - Add a sub IB device
2909 */
2910 struct ib_device *(*add_sub_dev)(struct ib_device *parent,
2911 enum rdma_nl_dev_type type,
2912 const char *name);
2913
2914 /*
2915 * del_sub_dev - Delete a sub IB device
2916 */
2917 void (*del_sub_dev)(struct ib_device *sub_dev);
2918
2919 /*
2920 * ufile_cleanup - Attempt to cleanup ubojects HW resources inside
2921 * the ufile.
2922 */
2923 void (*ufile_hw_cleanup)(struct ib_uverbs_file *ufile);
2924
2925 /*
2926 * report_port_event - Drivers need to implement this if they have
2927 * some private stuff to handle when link status changes.
2928 */
2929 void (*report_port_event)(struct ib_device *ibdev,
2930 struct net_device *ndev, unsigned long event);
2931
2932 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2933 DECLARE_RDMA_OBJ_SIZE(ib_counters);
2934 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2935 DECLARE_RDMA_OBJ_SIZE(ib_comp_cntr);
2936 DECLARE_RDMA_OBJ_SIZE(ib_dmah);
2937 DECLARE_RDMA_OBJ_SIZE(ib_mw);
2938 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2939 DECLARE_RDMA_OBJ_SIZE(ib_qp);
2940 DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2941 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2942 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2943 DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2944 DECLARE_RDMA_OBJ_SIZE(rdma_counter);
2945};
2946
2947struct ib_core_device {
2948 /* device must be the first element in structure until,
2949 * union of ib_core_device and device exists in ib_device.
2950 */
2951 struct device dev;
2952 possible_net_t rdma_net;
2953 struct kobject *ports_kobj;
2954 struct list_head port_list;
2955 struct ib_device *owner; /* reach back to owner ib_device */
2956};
2957
2958struct rdma_restrack_root;
2959struct ib_device {
2960 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2961 struct device *dma_device;
2962 struct ib_device_ops ops;
2963 char name[IB_DEVICE_NAME_MAX];
2964 struct rcu_head rcu_head;
2965
2966 struct list_head event_handler_list;
2967 /* Protects event_handler_list */
2968 struct rw_semaphore event_handler_rwsem;
2969
2970 /* Protects QP's event_handler calls and open_qp list */
2971 spinlock_t qp_open_list_lock;
2972
2973 struct rw_semaphore client_data_rwsem;
2974 struct xarray client_data;
2975 struct mutex unregistration_lock;
2976
2977 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2978 rwlock_t cache_lock;
2979 /**
2980 * port_data is indexed by port number
2981 */
2982 struct ib_port_data *port_data;
2983
2984 int num_comp_vectors;
2985
2986 union {
2987 struct device dev;
2988 struct ib_core_device coredev;
2989 };
2990
2991 /* First group is for device attributes,
2992 * Second group is for driver provided attributes (optional).
2993 * Third group is for the hw_stats
2994 * It is a NULL terminated array.
2995 */
2996 const struct attribute_group *groups[4];
2997 u8 hw_stats_attr_index;
2998
2999 u64 uverbs_cmd_mask;
3000
3001 char node_desc[IB_DEVICE_NODE_DESC_MAX];
3002 __be64 node_guid;
3003 u32 local_dma_lkey;
3004 u16 is_switch:1;
3005 /* Indicates kernel verbs support, should not be used in drivers */
3006 u16 kverbs_provider:1;
3007 /* CQ adaptive moderation (RDMA DIM) */
3008 u16 use_cq_dim:1;
3009 /* CoCo guest with DMA bounce buffering required */
3010 u16 cc_dma_bounce:1;
3011 u8 node_type;
3012 u32 phys_port_cnt;
3013 struct ib_device_attr attrs;
3014 struct hw_stats_device_data *hw_stats_data;
3015
3016#ifdef CONFIG_CGROUP_RDMA
3017 struct rdmacg_device cg_device;
3018#endif
3019
3020 u32 index;
3021
3022 spinlock_t cq_pools_lock;
3023 struct list_head cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
3024
3025 struct rdma_restrack_root *res;
3026
3027 const struct uapi_definition *driver_def;
3028
3029 /*
3030 * Positive refcount indicates that the device is currently
3031 * registered and cannot be unregistered.
3032 */
3033 refcount_t refcount;
3034 struct completion unreg_completion;
3035 struct work_struct unregistration_work;
3036
3037 const struct rdma_link_ops *link_ops;
3038
3039 /* Protects compat_devs xarray modifications */
3040 struct mutex compat_devs_mutex;
3041 /* Maintains compat devices for each net namespace */
3042 struct xarray compat_devs;
3043
3044 /* Used by iWarp CM */
3045 char iw_ifname[IFNAMSIZ];
3046 u32 iw_driver_flags;
3047 u32 lag_flags;
3048
3049 /* A parent device has a list of sub-devices */
3050 struct mutex subdev_lock;
3051 struct list_head subdev_list_head;
3052
3053 /* A sub device has a type and a parent */
3054 enum rdma_nl_dev_type type;
3055 struct ib_device *parent;
3056 struct list_head subdev_list;
3057
3058 enum rdma_nl_name_assign_type name_assign_type;
3059
3060 struct ib_frmr_pools *frmr_pools;
3061};
3062
3063static inline void *rdma_zalloc_obj(struct ib_device *dev, size_t size,
3064 gfp_t gfp, bool is_numa_aware)
3065{
3066 if (is_numa_aware && dev->ops.get_numa_node)
3067 return kzalloc_node(size, gfp, dev->ops.get_numa_node(dev));
3068
3069 return kzalloc(size, gfp);
3070}
3071
3072struct ib_client_nl_info;
3073struct ib_client {
3074 const char *name;
3075 int (*add)(struct ib_device *ibdev);
3076 void (*remove)(struct ib_device *, void *client_data);
3077 void (*rename)(struct ib_device *dev, void *client_data);
3078 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
3079 struct ib_client_nl_info *res);
3080 int (*get_global_nl_info)(struct ib_client_nl_info *res);
3081
3082 /* Returns the net_dev belonging to this ib_client and matching the
3083 * given parameters.
3084 * @dev: An RDMA device that the net_dev use for communication.
3085 * @port: A physical port number on the RDMA device.
3086 * @pkey: P_Key that the net_dev uses if applicable.
3087 * @gid: A GID that the net_dev uses to communicate.
3088 * @addr: An IP address the net_dev is configured with.
3089 * @client_data: The device's client data set by ib_set_client_data().
3090 *
3091 * An ib_client that implements a net_dev on top of RDMA devices
3092 * (such as IP over IB) should implement this callback, allowing the
3093 * rdma_cm module to find the right net_dev for a given request.
3094 *
3095 * The caller is responsible for calling dev_put on the returned
3096 * netdev. */
3097 struct net_device *(*get_net_dev_by_params)(
3098 struct ib_device *dev,
3099 u32 port,
3100 u16 pkey,
3101 const union ib_gid *gid,
3102 const struct sockaddr *addr,
3103 void *client_data);
3104
3105 refcount_t uses;
3106 struct completion uses_zero;
3107 u32 client_id;
3108
3109 /* kverbs are not required by the client */
3110 u8 no_kverbs_req:1;
3111};
3112
3113struct ib_device *_ib_alloc_device(size_t size, struct net *net);
3114#define ib_alloc_device(drv_struct, member) \
3115 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
3116 BUILD_BUG_ON_ZERO(offsetof( \
3117 struct drv_struct, member)), \
3118 &init_net), \
3119 struct drv_struct, member)
3120
3121#define ib_alloc_device_with_net(drv_struct, member, net) \
3122 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
3123 BUILD_BUG_ON_ZERO(offsetof( \
3124 struct drv_struct, member)), net), \
3125 struct drv_struct, member)
3126
3127void ib_dealloc_device(struct ib_device *device);
3128
3129void ib_get_device_fw_str(struct ib_device *device, char *str);
3130
3131int ib_register_device(struct ib_device *device, const char *name,
3132 struct device *dma_device);
3133void ib_unregister_device(struct ib_device *device);
3134void ib_unregister_driver(enum rdma_driver_id driver_id);
3135void ib_unregister_device_and_put(struct ib_device *device);
3136void ib_unregister_device_queued(struct ib_device *ib_dev);
3137
3138int ib_register_client (struct ib_client *client);
3139void ib_unregister_client(struct ib_client *client);
3140
3141/**
3142 * ib_get_client_data - Get IB client context
3143 * @device:Device to get context for
3144 * @client:Client to get context for
3145 *
3146 * ib_get_client_data() returns the client context data set with
3147 * ib_set_client_data(). This can only be called while the client is
3148 * registered to the device, once the ib_client remove() callback returns this
3149 * cannot be called.
3150 */
3151static inline void *ib_get_client_data(struct ib_device *device,
3152 struct ib_client *client)
3153{
3154 return xa_load(&device->client_data, client->client_id);
3155}
3156void ib_set_client_data(struct ib_device *device, struct ib_client *client,
3157 void *data);
3158void ib_set_device_ops(struct ib_device *device,
3159 const struct ib_device_ops *ops);
3160
3161#if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
3162int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
3163 unsigned long pfn, unsigned long size, pgprot_t prot,
3164 struct rdma_user_mmap_entry *entry);
3165int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
3166 struct rdma_user_mmap_entry *entry,
3167 size_t length);
3168int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
3169 struct rdma_user_mmap_entry *entry,
3170 size_t length, u32 min_pgoff,
3171 u32 max_pgoff);
3172
3173void rdma_user_mmap_disassociate(struct ib_device *device);
3174
3175static inline int
3176rdma_user_mmap_entry_insert_exact(struct ib_ucontext *ucontext,
3177 struct rdma_user_mmap_entry *entry,
3178 size_t length, u32 pgoff)
3179{
3180 return rdma_user_mmap_entry_insert_range(ucontext, entry, length, pgoff,
3181 pgoff);
3182}
3183
3184struct rdma_user_mmap_entry *
3185rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
3186 unsigned long pgoff);
3187struct rdma_user_mmap_entry *
3188rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
3189 struct vm_area_struct *vma);
3190void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
3191
3192void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
3193#else
3194static inline int rdma_user_mmap_io(struct ib_ucontext *ucontext,
3195 struct vm_area_struct *vma,
3196 unsigned long pfn, unsigned long size,
3197 pgprot_t prot,
3198 struct rdma_user_mmap_entry *entry)
3199{
3200 return -EINVAL;
3201}
3202
3203static inline int
3204rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
3205 struct rdma_user_mmap_entry *entry, size_t length)
3206{
3207 return -EINVAL;
3208}
3209
3210static inline int
3211rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
3212 struct rdma_user_mmap_entry *entry,
3213 size_t length, u32 min_pgoff, u32 max_pgoff)
3214{
3215 return -EINVAL;
3216}
3217
3218static inline void rdma_user_mmap_disassociate(struct ib_device *device)
3219{
3220}
3221
3222static inline int
3223rdma_user_mmap_entry_insert_exact(struct ib_ucontext *ucontext,
3224 struct rdma_user_mmap_entry *entry,
3225 size_t length, u32 pgoff)
3226{
3227 return -EINVAL;
3228}
3229
3230static inline struct rdma_user_mmap_entry *
3231rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
3232 unsigned long pgoff)
3233{
3234 return NULL;
3235}
3236
3237static inline struct rdma_user_mmap_entry *
3238rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
3239 struct vm_area_struct *vma)
3240{
3241 return NULL;
3242}
3243
3244static inline void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry)
3245{
3246}
3247
3248static inline void
3249rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry)
3250{
3251}
3252#endif
3253
3254static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
3255{
3256 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
3257}
3258
3259static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
3260{
3261 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
3262}
3263
3264static inline bool ib_is_buffer_cleared(const void __user *p,
3265 size_t len)
3266{
3267 bool ret;
3268 u8 *buf;
3269
3270 if (len > USHRT_MAX)
3271 return false;
3272
3273 buf = memdup_user(p, len);
3274 if (IS_ERR(buf))
3275 return false;
3276
3277 ret = !memchr_inv(buf, 0, len);
3278 kfree(buf);
3279 return ret;
3280}
3281
3282static inline bool ib_is_udata_cleared(struct ib_udata *udata,
3283 size_t offset,
3284 size_t len)
3285{
3286 return ib_is_buffer_cleared(udata->inbuf + offset, len);
3287}
3288
3289/**
3290 * ib_modify_qp_is_ok - Check that the supplied attribute mask
3291 * contains all required attributes and no attributes not allowed for
3292 * the given QP state transition.
3293 * @cur_state: Current QP state
3294 * @next_state: Next QP state
3295 * @type: QP type
3296 * @mask: Mask of supplied QP attributes
3297 *
3298 * This function is a helper function that a low-level driver's
3299 * modify_qp method can use to validate the consumer's input. It
3300 * checks that cur_state and next_state are valid QP states, that a
3301 * transition from cur_state to next_state is allowed by the IB spec,
3302 * and that the attribute mask supplied is allowed for the transition.
3303 */
3304bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
3305 enum ib_qp_type type, enum ib_qp_attr_mask mask);
3306
3307void ib_register_event_handler(struct ib_event_handler *event_handler);
3308void ib_unregister_event_handler(struct ib_event_handler *event_handler);
3309void ib_dispatch_event(const struct ib_event *event);
3310
3311int ib_query_port(struct ib_device *device,
3312 u32 port_num, struct ib_port_attr *port_attr);
3313
3314enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
3315 u32 port_num);
3316
3317/**
3318 * rdma_cap_ib_switch - Check if the device is IB switch
3319 * @device: Device to check
3320 *
3321 * Device driver is responsible for setting is_switch bit on
3322 * in ib_device structure at init time.
3323 *
3324 * Return: true if the device is IB switch.
3325 */
3326static inline bool rdma_cap_ib_switch(const struct ib_device *device)
3327{
3328 return device->is_switch;
3329}
3330
3331/**
3332 * rdma_start_port - Return the first valid port number for the device
3333 * specified
3334 *
3335 * @device: Device to be checked
3336 *
3337 * Return start port number
3338 */
3339static inline u32 rdma_start_port(const struct ib_device *device)
3340{
3341 return rdma_cap_ib_switch(device) ? 0 : 1;
3342}
3343
3344/**
3345 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3346 * @device: The struct ib_device * to iterate over
3347 * @iter: The unsigned int to store the port number
3348 */
3349#define rdma_for_each_port(device, iter) \
3350 for (iter = rdma_start_port(device + \
3351 BUILD_BUG_ON_ZERO(!__same_type(u32, \
3352 iter))); \
3353 iter <= rdma_end_port(device); iter++)
3354
3355/**
3356 * rdma_end_port - Return the last valid port number for the device
3357 * specified
3358 *
3359 * @device: Device to be checked
3360 *
3361 * Return last port number
3362 */
3363static inline u32 rdma_end_port(const struct ib_device *device)
3364{
3365 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3366}
3367
3368static inline int rdma_is_port_valid(const struct ib_device *device,
3369 unsigned int port)
3370{
3371 return (port >= rdma_start_port(device) &&
3372 port <= rdma_end_port(device));
3373}
3374
3375static inline bool rdma_is_grh_required(const struct ib_device *device,
3376 u32 port_num)
3377{
3378 return device->port_data[port_num].immutable.core_cap_flags &
3379 RDMA_CORE_PORT_IB_GRH_REQUIRED;
3380}
3381
3382static inline bool rdma_protocol_ib(const struct ib_device *device,
3383 u32 port_num)
3384{
3385 return device->port_data[port_num].immutable.core_cap_flags &
3386 RDMA_CORE_CAP_PROT_IB;
3387}
3388
3389static inline bool rdma_protocol_roce(const struct ib_device *device,
3390 u32 port_num)
3391{
3392 return device->port_data[port_num].immutable.core_cap_flags &
3393 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3394}
3395
3396static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device,
3397 u32 port_num)
3398{
3399 return device->port_data[port_num].immutable.core_cap_flags &
3400 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3401}
3402
3403static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device,
3404 u32 port_num)
3405{
3406 return device->port_data[port_num].immutable.core_cap_flags &
3407 RDMA_CORE_CAP_PROT_ROCE;
3408}
3409
3410static inline bool rdma_protocol_iwarp(const struct ib_device *device,
3411 u32 port_num)
3412{
3413 return device->port_data[port_num].immutable.core_cap_flags &
3414 RDMA_CORE_CAP_PROT_IWARP;
3415}
3416
3417static inline bool rdma_ib_or_roce(const struct ib_device *device,
3418 u32 port_num)
3419{
3420 return rdma_protocol_ib(device, port_num) ||
3421 rdma_protocol_roce(device, port_num);
3422}
3423
3424static inline bool rdma_protocol_raw_packet(const struct ib_device *device,
3425 u32 port_num)
3426{
3427 return device->port_data[port_num].immutable.core_cap_flags &
3428 RDMA_CORE_CAP_PROT_RAW_PACKET;
3429}
3430
3431static inline bool rdma_protocol_usnic(const struct ib_device *device,
3432 u32 port_num)
3433{
3434 return device->port_data[port_num].immutable.core_cap_flags &
3435 RDMA_CORE_CAP_PROT_USNIC;
3436}
3437
3438/**
3439 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3440 * Management Datagrams.
3441 * @device: Device to check
3442 * @port_num: Port number to check
3443 *
3444 * Management Datagrams (MAD) are a required part of the InfiniBand
3445 * specification and are supported on all InfiniBand devices. A slightly
3446 * extended version are also supported on OPA interfaces.
3447 *
3448 * Return: true if the port supports sending/receiving of MAD packets.
3449 */
3450static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num)
3451{
3452 return device->port_data[port_num].immutable.core_cap_flags &
3453 RDMA_CORE_CAP_IB_MAD;
3454}
3455
3456/**
3457 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3458 * Management Datagrams.
3459 * @device: Device to check
3460 * @port_num: Port number to check
3461 *
3462 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3463 * datagrams with their own versions. These OPA MADs share many but not all of
3464 * the characteristics of InfiniBand MADs.
3465 *
3466 * OPA MADs differ in the following ways:
3467 *
3468 * 1) MADs are variable size up to 2K
3469 * IBTA defined MADs remain fixed at 256 bytes
3470 * 2) OPA SMPs must carry valid PKeys
3471 * 3) OPA SMP packets are a different format
3472 *
3473 * Return: true if the port supports OPA MAD packet formats.
3474 */
3475static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num)
3476{
3477 return device->port_data[port_num].immutable.core_cap_flags &
3478 RDMA_CORE_CAP_OPA_MAD;
3479}
3480
3481/**
3482 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3483 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3484 * @device: Device to check
3485 * @port_num: Port number to check
3486 *
3487 * Each InfiniBand node is required to provide a Subnet Management Agent
3488 * that the subnet manager can access. Prior to the fabric being fully
3489 * configured by the subnet manager, the SMA is accessed via a well known
3490 * interface called the Subnet Management Interface (SMI). This interface
3491 * uses directed route packets to communicate with the SM to get around the
3492 * chicken and egg problem of the SM needing to know what's on the fabric
3493 * in order to configure the fabric, and needing to configure the fabric in
3494 * order to send packets to the devices on the fabric. These directed
3495 * route packets do not need the fabric fully configured in order to reach
3496 * their destination. The SMI is the only method allowed to send
3497 * directed route packets on an InfiniBand fabric.
3498 *
3499 * Return: true if the port provides an SMI.
3500 */
3501static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num)
3502{
3503 return device->port_data[port_num].immutable.core_cap_flags &
3504 RDMA_CORE_CAP_IB_SMI;
3505}
3506
3507/**
3508 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3509 * Communication Manager.
3510 * @device: Device to check
3511 * @port_num: Port number to check
3512 *
3513 * The InfiniBand Communication Manager is one of many pre-defined General
3514 * Service Agents (GSA) that are accessed via the General Service
3515 * Interface (GSI). It's role is to facilitate establishment of connections
3516 * between nodes as well as other management related tasks for established
3517 * connections.
3518 *
3519 * Return: true if the port supports an IB CM (this does not guarantee that
3520 * a CM is actually running however).
3521 */
3522static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num)
3523{
3524 return device->port_data[port_num].immutable.core_cap_flags &
3525 RDMA_CORE_CAP_IB_CM;
3526}
3527
3528/**
3529 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3530 * Communication Manager.
3531 * @device: Device to check
3532 * @port_num: Port number to check
3533 *
3534 * Similar to above, but specific to iWARP connections which have a different
3535 * managment protocol than InfiniBand.
3536 *
3537 * Return: true if the port supports an iWARP CM (this does not guarantee that
3538 * a CM is actually running however).
3539 */
3540static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num)
3541{
3542 return device->port_data[port_num].immutable.core_cap_flags &
3543 RDMA_CORE_CAP_IW_CM;
3544}
3545
3546/**
3547 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3548 * Subnet Administration.
3549 * @device: Device to check
3550 * @port_num: Port number to check
3551 *
3552 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3553 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3554 * fabrics, devices should resolve routes to other hosts by contacting the
3555 * SA to query the proper route.
3556 *
3557 * Return: true if the port should act as a client to the fabric Subnet
3558 * Administration interface. This does not imply that the SA service is
3559 * running locally.
3560 */
3561static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num)
3562{
3563 return device->port_data[port_num].immutable.core_cap_flags &
3564 RDMA_CORE_CAP_IB_SA;
3565}
3566
3567/**
3568 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3569 * Multicast.
3570 * @device: Device to check
3571 * @port_num: Port number to check
3572 *
3573 * InfiniBand multicast registration is more complex than normal IPv4 or
3574 * IPv6 multicast registration. Each Host Channel Adapter must register
3575 * with the Subnet Manager when it wishes to join a multicast group. It
3576 * should do so only once regardless of how many queue pairs it subscribes
3577 * to this group. And it should leave the group only after all queue pairs
3578 * attached to the group have been detached.
3579 *
3580 * Return: true if the port must undertake the additional adminstrative
3581 * overhead of registering/unregistering with the SM and tracking of the
3582 * total number of queue pairs attached to the multicast group.
3583 */
3584static inline bool rdma_cap_ib_mcast(const struct ib_device *device,
3585 u32 port_num)
3586{
3587 return rdma_cap_ib_sa(device, port_num);
3588}
3589
3590/**
3591 * rdma_cap_af_ib - Check if the port of device has the capability
3592 * Native Infiniband Address.
3593 * @device: Device to check
3594 * @port_num: Port number to check
3595 *
3596 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3597 * GID. RoCE uses a different mechanism, but still generates a GID via
3598 * a prescribed mechanism and port specific data.
3599 *
3600 * Return: true if the port uses a GID address to identify devices on the
3601 * network.
3602 */
3603static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num)
3604{
3605 return device->port_data[port_num].immutable.core_cap_flags &
3606 RDMA_CORE_CAP_AF_IB;
3607}
3608
3609/**
3610 * rdma_cap_eth_ah - Check if the port of device has the capability
3611 * Ethernet Address Handle.
3612 * @device: Device to check
3613 * @port_num: Port number to check
3614 *
3615 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3616 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3617 * port. Normally, packet headers are generated by the sending host
3618 * adapter, but when sending connectionless datagrams, we must manually
3619 * inject the proper headers for the fabric we are communicating over.
3620 *
3621 * Return: true if we are running as a RoCE port and must force the
3622 * addition of a Global Route Header built from our Ethernet Address
3623 * Handle into our header list for connectionless packets.
3624 */
3625static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num)
3626{
3627 return device->port_data[port_num].immutable.core_cap_flags &
3628 RDMA_CORE_CAP_ETH_AH;
3629}
3630
3631/**
3632 * rdma_cap_opa_ah - Check if the port of device supports
3633 * OPA Address handles
3634 * @device: Device to check
3635 * @port_num: Port number to check
3636 *
3637 * Return: true if we are running on an OPA device which supports
3638 * the extended OPA addressing.
3639 */
3640static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num)
3641{
3642 return (device->port_data[port_num].immutable.core_cap_flags &
3643 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3644}
3645
3646/**
3647 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3648 *
3649 * @device: Device
3650 * @port_num: Port number
3651 *
3652 * This MAD size includes the MAD headers and MAD payload. No other headers
3653 * are included.
3654 *
3655 * Return the max MAD size required by the Port. Will return 0 if the port
3656 * does not support MADs
3657 */
3658static inline size_t rdma_max_mad_size(const struct ib_device *device,
3659 u32 port_num)
3660{
3661 return device->port_data[port_num].immutable.max_mad_size;
3662}
3663
3664/**
3665 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3666 * @device: Device to check
3667 * @port_num: Port number to check
3668 *
3669 * RoCE GID table mechanism manages the various GIDs for a device.
3670 *
3671 * NOTE: if allocating the port's GID table has failed, this call will still
3672 * return true, but any RoCE GID table API will fail.
3673 *
3674 * Return: true if the port uses RoCE GID table mechanism in order to manage
3675 * its GIDs.
3676 */
3677static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3678 u32 port_num)
3679{
3680 return rdma_protocol_roce(device, port_num) &&
3681 device->ops.add_gid && device->ops.del_gid;
3682}
3683
3684/*
3685 * Check if the device supports READ W/ INVALIDATE.
3686 */
3687static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3688{
3689 /*
3690 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3691 * has support for it yet.
3692 */
3693 return rdma_protocol_iwarp(dev, port_num);
3694}
3695
3696/**
3697 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3698 * @device: Device
3699 * @port_num: 1 based Port number
3700 *
3701 * Return true if port is an Intel OPA port , false if not
3702 */
3703static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3704 u32 port_num)
3705{
3706 return (device->port_data[port_num].immutable.core_cap_flags &
3707 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3708}
3709
3710/**
3711 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3712 * @device: Device
3713 * @port: Port number
3714 * @mtu: enum value of MTU
3715 *
3716 * Return the MTU size supported by the port as an integer value. Will return
3717 * -1 if enum value of mtu is not supported.
3718 */
3719static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port,
3720 int mtu)
3721{
3722 if (rdma_core_cap_opa_port(device, port))
3723 return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3724 else
3725 return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3726}
3727
3728/**
3729 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3730 * @device: Device
3731 * @port: Port number
3732 * @attr: port attribute
3733 *
3734 * Return the MTU size supported by the port as an integer value.
3735 */
3736static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port,
3737 struct ib_port_attr *attr)
3738{
3739 if (rdma_core_cap_opa_port(device, port))
3740 return attr->phys_mtu;
3741 else
3742 return ib_mtu_enum_to_int(attr->max_mtu);
3743}
3744
3745int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port,
3746 int state);
3747int ib_get_vf_config(struct ib_device *device, int vf, u32 port,
3748 struct ifla_vf_info *info);
3749int ib_get_vf_stats(struct ib_device *device, int vf, u32 port,
3750 struct ifla_vf_stats *stats);
3751int ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
3752 struct ifla_vf_guid *node_guid,
3753 struct ifla_vf_guid *port_guid);
3754int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid,
3755 int type);
3756
3757int ib_query_pkey(struct ib_device *device,
3758 u32 port_num, u16 index, u16 *pkey);
3759
3760int ib_modify_device(struct ib_device *device,
3761 int device_modify_mask,
3762 struct ib_device_modify *device_modify);
3763
3764int ib_modify_port(struct ib_device *device,
3765 u32 port_num, int port_modify_mask,
3766 struct ib_port_modify *port_modify);
3767
3768int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3769 u32 *port_num, u16 *index);
3770
3771int ib_find_pkey(struct ib_device *device,
3772 u32 port_num, u16 pkey, u16 *index);
3773
3774enum ib_pd_flags {
3775 /*
3776 * Create a memory registration for all memory in the system and place
3777 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3778 * ULPs to avoid the overhead of dynamic MRs.
3779 *
3780 * This flag is generally considered unsafe and must only be used in
3781 * extremly trusted environments. Every use of it will log a warning
3782 * in the kernel log.
3783 */
3784 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3785};
3786
3787struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3788 const char *caller);
3789
3790/**
3791 * ib_alloc_pd - Allocates an unused protection domain.
3792 * @device: The device on which to allocate the protection domain.
3793 * @flags: protection domain flags
3794 *
3795 * A protection domain object provides an association between QPs, shared
3796 * receive queues, address handles, memory regions, and memory windows.
3797 *
3798 * Every PD has a local_dma_lkey which can be used as the lkey value for local
3799 * memory operations.
3800 */
3801#define ib_alloc_pd(device, flags) \
3802 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3803
3804int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3805
3806/**
3807 * ib_dealloc_pd - Deallocate kernel PD
3808 * @pd: The protection domain
3809 *
3810 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3811 */
3812static inline void ib_dealloc_pd(struct ib_pd *pd)
3813{
3814 int ret = ib_dealloc_pd_user(pd, NULL);
3815
3816 WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3817}
3818
3819enum rdma_create_ah_flags {
3820 /* In a sleepable context */
3821 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3822};
3823
3824/**
3825 * rdma_create_ah - Creates an address handle for the given address vector.
3826 * @pd: The protection domain associated with the address handle.
3827 * @ah_attr: The attributes of the address vector.
3828 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3829 *
3830 * The address handle is used to reference a local or global destination
3831 * in all UD QP post sends.
3832 */
3833struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3834 u32 flags);
3835
3836/**
3837 * rdma_create_user_ah - Creates an address handle for the given address vector.
3838 * It resolves destination mac address for ah attribute of RoCE type.
3839 * @pd: The protection domain associated with the address handle.
3840 * @ah_attr: The attributes of the address vector.
3841 * @udata: pointer to user's input output buffer information need by
3842 * provider driver.
3843 *
3844 * It returns 0 on success and returns appropriate error code on error.
3845 * The address handle is used to reference a local or global destination
3846 * in all UD QP post sends.
3847 */
3848struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3849 struct rdma_ah_attr *ah_attr,
3850 struct ib_udata *udata);
3851/**
3852 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3853 * work completion.
3854 * @hdr: the L3 header to parse
3855 * @net_type: type of header to parse
3856 * @sgid: place to store source gid
3857 * @dgid: place to store destination gid
3858 */
3859int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3860 enum rdma_network_type net_type,
3861 union ib_gid *sgid, union ib_gid *dgid);
3862
3863/**
3864 * ib_get_rdma_header_version - Get the header version
3865 * @hdr: the L3 header to parse
3866 */
3867int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3868
3869/**
3870 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3871 * work completion.
3872 * @device: Device on which the received message arrived.
3873 * @port_num: Port on which the received message arrived.
3874 * @wc: Work completion associated with the received message.
3875 * @grh: References the received global route header. This parameter is
3876 * ignored unless the work completion indicates that the GRH is valid.
3877 * @ah_attr: Returned attributes that can be used when creating an address
3878 * handle for replying to the message.
3879 * When ib_init_ah_attr_from_wc() returns success,
3880 * (a) for IB link layer it optionally contains a reference to SGID attribute
3881 * when GRH is present for IB link layer.
3882 * (b) for RoCE link layer it contains a reference to SGID attribute.
3883 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3884 * attributes which are initialized using ib_init_ah_attr_from_wc().
3885 *
3886 */
3887int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num,
3888 const struct ib_wc *wc, const struct ib_grh *grh,
3889 struct rdma_ah_attr *ah_attr);
3890
3891/**
3892 * ib_create_ah_from_wc - Creates an address handle associated with the
3893 * sender of the specified work completion.
3894 * @pd: The protection domain associated with the address handle.
3895 * @wc: Work completion information associated with a received message.
3896 * @grh: References the received global route header. This parameter is
3897 * ignored unless the work completion indicates that the GRH is valid.
3898 * @port_num: The outbound port number to associate with the address.
3899 *
3900 * The address handle is used to reference a local or global destination
3901 * in all UD QP post sends.
3902 */
3903struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3904 const struct ib_grh *grh, u32 port_num);
3905
3906/**
3907 * rdma_modify_ah - Modifies the address vector associated with an address
3908 * handle.
3909 * @ah: The address handle to modify.
3910 * @ah_attr: The new address vector attributes to associate with the
3911 * address handle.
3912 */
3913int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3914
3915/**
3916 * rdma_query_ah - Queries the address vector associated with an address
3917 * handle.
3918 * @ah: The address handle to query.
3919 * @ah_attr: The address vector attributes associated with the address
3920 * handle.
3921 */
3922int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3923
3924enum rdma_destroy_ah_flags {
3925 /* In a sleepable context */
3926 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3927};
3928
3929/**
3930 * rdma_destroy_ah_user - Destroys an address handle.
3931 * @ah: The address handle to destroy.
3932 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3933 * @udata: Valid user data or NULL for kernel objects
3934 */
3935int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3936
3937/**
3938 * rdma_destroy_ah - Destroys an kernel address handle.
3939 * @ah: The address handle to destroy.
3940 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3941 *
3942 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3943 */
3944static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3945{
3946 int ret = rdma_destroy_ah_user(ah, flags, NULL);
3947
3948 WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3949}
3950
3951struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3952 struct ib_srq_init_attr *srq_init_attr,
3953 struct ib_usrq_object *uobject,
3954 struct ib_udata *udata);
3955static inline struct ib_srq *
3956ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3957{
3958 if (!pd->device->ops.create_srq)
3959 return ERR_PTR(-EOPNOTSUPP);
3960
3961 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3962}
3963
3964/**
3965 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3966 * @srq: The SRQ to modify.
3967 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3968 * the current values of selected SRQ attributes are returned.
3969 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3970 * are being modified.
3971 *
3972 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3973 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3974 * the number of receives queued drops below the limit.
3975 */
3976int ib_modify_srq(struct ib_srq *srq,
3977 struct ib_srq_attr *srq_attr,
3978 enum ib_srq_attr_mask srq_attr_mask);
3979
3980/**
3981 * ib_query_srq - Returns the attribute list and current values for the
3982 * specified SRQ.
3983 * @srq: The SRQ to query.
3984 * @srq_attr: The attributes of the specified SRQ.
3985 */
3986int ib_query_srq(struct ib_srq *srq,
3987 struct ib_srq_attr *srq_attr);
3988
3989/**
3990 * ib_destroy_srq_user - Destroys the specified SRQ.
3991 * @srq: The SRQ to destroy.
3992 * @udata: Valid user data or NULL for kernel objects
3993 */
3994int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3995
3996/**
3997 * ib_destroy_srq - Destroys the specified kernel SRQ.
3998 * @srq: The SRQ to destroy.
3999 *
4000 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
4001 */
4002static inline void ib_destroy_srq(struct ib_srq *srq)
4003{
4004 int ret = ib_destroy_srq_user(srq, NULL);
4005
4006 WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
4007}
4008
4009/**
4010 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
4011 * @srq: The SRQ to post the work request on.
4012 * @recv_wr: A list of work requests to post on the receive queue.
4013 * @bad_recv_wr: On an immediate failure, this parameter will reference
4014 * the work request that failed to be posted on the QP.
4015 */
4016static inline int ib_post_srq_recv(struct ib_srq *srq,
4017 const struct ib_recv_wr *recv_wr,
4018 const struct ib_recv_wr **bad_recv_wr)
4019{
4020 const struct ib_recv_wr *dummy;
4021
4022 return srq->device->ops.post_srq_recv(srq, recv_wr,
4023 bad_recv_wr ? : &dummy);
4024}
4025
4026struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd,
4027 struct ib_qp_init_attr *qp_init_attr,
4028 const char *caller);
4029/**
4030 * ib_create_qp - Creates a kernel QP associated with the specific protection
4031 * domain.
4032 * @pd: The protection domain associated with the QP.
4033 * @init_attr: A list of initial attributes required to create the
4034 * QP. If QP creation succeeds, then the attributes are updated to
4035 * the actual capabilities of the created QP.
4036 */
4037static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
4038 struct ib_qp_init_attr *init_attr)
4039{
4040 return ib_create_qp_kernel(pd, init_attr, KBUILD_MODNAME);
4041}
4042
4043/**
4044 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
4045 * @qp: The QP to modify.
4046 * @attr: On input, specifies the QP attributes to modify. On output,
4047 * the current values of selected QP attributes are returned.
4048 * @attr_mask: A bit-mask used to specify which attributes of the QP
4049 * are being modified.
4050 * @udata: pointer to user's input output buffer information
4051 * are being modified.
4052 * It returns 0 on success and returns appropriate error code on error.
4053 */
4054int ib_modify_qp_with_udata(struct ib_qp *qp,
4055 struct ib_qp_attr *attr,
4056 int attr_mask,
4057 struct ib_udata *udata);
4058
4059/**
4060 * ib_modify_qp - Modifies the attributes for the specified QP and then
4061 * transitions the QP to the given state.
4062 * @qp: The QP to modify.
4063 * @qp_attr: On input, specifies the QP attributes to modify. On output,
4064 * the current values of selected QP attributes are returned.
4065 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
4066 * are being modified.
4067 */
4068int ib_modify_qp(struct ib_qp *qp,
4069 struct ib_qp_attr *qp_attr,
4070 int qp_attr_mask);
4071
4072/**
4073 * ib_query_qp - Returns the attribute list and current values for the
4074 * specified QP.
4075 * @qp: The QP to query.
4076 * @qp_attr: The attributes of the specified QP.
4077 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
4078 * @qp_init_attr: Additional attributes of the selected QP.
4079 *
4080 * The qp_attr_mask may be used to limit the query to gathering only the
4081 * selected attributes.
4082 */
4083int ib_query_qp(struct ib_qp *qp,
4084 struct ib_qp_attr *qp_attr,
4085 int qp_attr_mask,
4086 struct ib_qp_init_attr *qp_init_attr);
4087
4088/**
4089 * ib_destroy_qp - Destroys the specified QP.
4090 * @qp: The QP to destroy.
4091 * @udata: Valid udata or NULL for kernel objects
4092 */
4093int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
4094
4095/**
4096 * ib_destroy_qp - Destroys the specified kernel QP.
4097 * @qp: The QP to destroy.
4098 *
4099 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
4100 */
4101static inline int ib_destroy_qp(struct ib_qp *qp)
4102{
4103 return ib_destroy_qp_user(qp, NULL);
4104}
4105
4106/**
4107 * ib_open_qp - Obtain a reference to an existing sharable QP.
4108 * @xrcd: XRC domain
4109 * @qp_open_attr: Attributes identifying the QP to open.
4110 *
4111 * Returns a reference to a sharable QP.
4112 */
4113struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
4114 struct ib_qp_open_attr *qp_open_attr);
4115
4116/**
4117 * ib_close_qp - Release an external reference to a QP.
4118 * @qp: The QP handle to release
4119 *
4120 * The opened QP handle is released by the caller. The underlying
4121 * shared QP is not destroyed until all internal references are released.
4122 */
4123int ib_close_qp(struct ib_qp *qp);
4124
4125/**
4126 * ib_post_send - Posts a list of work requests to the send queue of
4127 * the specified QP.
4128 * @qp: The QP to post the work request on.
4129 * @send_wr: A list of work requests to post on the send queue.
4130 * @bad_send_wr: On an immediate failure, this parameter will reference
4131 * the work request that failed to be posted on the QP.
4132 *
4133 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
4134 * error is returned, the QP state shall not be affected,
4135 * ib_post_send() will return an immediate error after queueing any
4136 * earlier work requests in the list.
4137 */
4138static inline int ib_post_send(struct ib_qp *qp,
4139 const struct ib_send_wr *send_wr,
4140 const struct ib_send_wr **bad_send_wr)
4141{
4142 const struct ib_send_wr *dummy;
4143
4144 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
4145}
4146
4147/**
4148 * ib_post_recv - Posts a list of work requests to the receive queue of
4149 * the specified QP.
4150 * @qp: The QP to post the work request on.
4151 * @recv_wr: A list of work requests to post on the receive queue.
4152 * @bad_recv_wr: On an immediate failure, this parameter will reference
4153 * the work request that failed to be posted on the QP.
4154 */
4155static inline int ib_post_recv(struct ib_qp *qp,
4156 const struct ib_recv_wr *recv_wr,
4157 const struct ib_recv_wr **bad_recv_wr)
4158{
4159 const struct ib_recv_wr *dummy;
4160
4161 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
4162}
4163
4164struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
4165 int comp_vector, enum ib_poll_context poll_ctx,
4166 const char *caller);
4167static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
4168 int nr_cqe, int comp_vector,
4169 enum ib_poll_context poll_ctx)
4170{
4171 return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
4172 KBUILD_MODNAME);
4173}
4174
4175struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
4176 int nr_cqe, enum ib_poll_context poll_ctx,
4177 const char *caller);
4178
4179/**
4180 * ib_alloc_cq_any: Allocate kernel CQ
4181 * @dev: The IB device
4182 * @private: Private data attached to the CQE
4183 * @nr_cqe: Number of CQEs in the CQ
4184 * @poll_ctx: Context used for polling the CQ
4185 */
4186static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
4187 void *private, int nr_cqe,
4188 enum ib_poll_context poll_ctx)
4189{
4190 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
4191 KBUILD_MODNAME);
4192}
4193
4194void ib_free_cq(struct ib_cq *cq);
4195int ib_process_cq_direct(struct ib_cq *cq, int budget);
4196
4197/**
4198 * ib_create_cq - Creates a CQ on the specified device.
4199 * @device: The device on which to create the CQ.
4200 * @comp_handler: A user-specified callback that is invoked when a
4201 * completion event occurs on the CQ.
4202 * @event_handler: A user-specified callback that is invoked when an
4203 * asynchronous event not associated with a completion occurs on the CQ.
4204 * @cq_context: Context associated with the CQ returned to the user via
4205 * the associated completion and event handlers.
4206 * @cq_attr: The attributes the CQ should be created upon.
4207 *
4208 * Users can examine the cq structure to determine the actual CQ size.
4209 */
4210struct ib_cq *__ib_create_cq(struct ib_device *device,
4211 ib_comp_handler comp_handler,
4212 void (*event_handler)(struct ib_event *, void *),
4213 void *cq_context,
4214 const struct ib_cq_init_attr *cq_attr,
4215 const char *caller);
4216#define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
4217 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
4218
4219/**
4220 * rdma_set_cq_moderation - Modifies moderation params of the CQ
4221 * @cq: The CQ to modify.
4222 * @cq_count: number of CQEs that will trigger an event
4223 * @cq_period: max period of time in usec before triggering an event
4224 *
4225 */
4226int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
4227
4228/**
4229 * ib_destroy_cq_user - Destroys the specified CQ.
4230 * @cq: The CQ to destroy.
4231 * @udata: Valid user data or NULL for kernel objects
4232 */
4233int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
4234
4235/**
4236 * ib_destroy_cq - Destroys the specified kernel CQ.
4237 * @cq: The CQ to destroy.
4238 *
4239 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
4240 */
4241static inline void ib_destroy_cq(struct ib_cq *cq)
4242{
4243 int ret = ib_destroy_cq_user(cq, NULL);
4244
4245 WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
4246}
4247
4248/**
4249 * ib_poll_cq - poll a CQ for completion(s)
4250 * @cq:the CQ being polled
4251 * @num_entries:maximum number of completions to return
4252 * @wc:array of at least @num_entries &struct ib_wc where completions
4253 * will be returned
4254 *
4255 * Poll a CQ for (possibly multiple) completions. If the return value
4256 * is < 0, an error occurred. If the return value is >= 0, it is the
4257 * number of completions returned. If the return value is
4258 * non-negative and < num_entries, then the CQ was emptied.
4259 */
4260static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
4261 struct ib_wc *wc)
4262{
4263 return cq->device->ops.poll_cq(cq, num_entries, wc);
4264}
4265
4266/**
4267 * ib_req_notify_cq - Request completion notification on a CQ.
4268 * @cq: The CQ to generate an event for.
4269 * @flags:
4270 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
4271 * to request an event on the next solicited event or next work
4272 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
4273 * may also be |ed in to request a hint about missed events, as
4274 * described below.
4275 *
4276 * Return Value:
4277 * < 0 means an error occurred while requesting notification
4278 * == 0 means notification was requested successfully, and if
4279 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
4280 * were missed and it is safe to wait for another event. In
4281 * this case is it guaranteed that any work completions added
4282 * to the CQ since the last CQ poll will trigger a completion
4283 * notification event.
4284 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
4285 * in. It means that the consumer must poll the CQ again to
4286 * make sure it is empty to avoid missing an event because of a
4287 * race between requesting notification and an entry being
4288 * added to the CQ. This return value means it is possible
4289 * (but not guaranteed) that a work completion has been added
4290 * to the CQ since the last poll without triggering a
4291 * completion notification event.
4292 */
4293static inline int ib_req_notify_cq(struct ib_cq *cq,
4294 enum ib_cq_notify_flags flags)
4295{
4296 return cq->device->ops.req_notify_cq(cq, flags);
4297}
4298
4299struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
4300 int comp_vector_hint,
4301 enum ib_poll_context poll_ctx);
4302
4303void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
4304
4305/*
4306 * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
4307 * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
4308 * address into the dma address.
4309 */
4310static inline bool ib_uses_virt_dma(struct ib_device *dev)
4311{
4312 return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
4313}
4314
4315/*
4316 * Check if a IB device's underlying DMA mapping supports P2PDMA transfers.
4317 */
4318static inline bool ib_dma_pci_p2p_dma_supported(struct ib_device *dev)
4319{
4320 if (ib_uses_virt_dma(dev))
4321 return false;
4322
4323 return dma_pci_p2pdma_supported(dev->dma_device);
4324}
4325
4326/**
4327 * ib_virt_dma_to_ptr - Convert a dma_addr to a kernel pointer
4328 * @dma_addr: The DMA address
4329 *
4330 * Used by ib_uses_virt_dma() devices to get back to the kernel pointer after
4331 * going through the dma_addr marshalling.
4332 */
4333static inline void *ib_virt_dma_to_ptr(u64 dma_addr)
4334{
4335 /* virt_dma mode maps the kvs's directly into the dma addr */
4336 return (void *)(uintptr_t)dma_addr;
4337}
4338
4339/**
4340 * ib_virt_dma_to_page - Convert a dma_addr to a struct page
4341 * @dma_addr: The DMA address
4342 *
4343 * Used by ib_uses_virt_dma() device to get back to the struct page after going
4344 * through the dma_addr marshalling.
4345 */
4346static inline struct page *ib_virt_dma_to_page(u64 dma_addr)
4347{
4348 return virt_to_page(ib_virt_dma_to_ptr(dma_addr));
4349}
4350
4351/**
4352 * ib_dma_mapping_error - check a DMA addr for error
4353 * @dev: The device for which the dma_addr was created
4354 * @dma_addr: The DMA address to check
4355 */
4356static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
4357{
4358 if (ib_uses_virt_dma(dev))
4359 return 0;
4360 return dma_mapping_error(dev->dma_device, dma_addr);
4361}
4362
4363/**
4364 * ib_dma_map_single - Map a kernel virtual address to DMA address
4365 * @dev: The device for which the dma_addr is to be created
4366 * @cpu_addr: The kernel virtual address
4367 * @size: The size of the region in bytes
4368 * @direction: The direction of the DMA
4369 */
4370static inline u64 ib_dma_map_single(struct ib_device *dev,
4371 void *cpu_addr, size_t size,
4372 enum dma_data_direction direction)
4373{
4374 if (ib_uses_virt_dma(dev))
4375 return (uintptr_t)cpu_addr;
4376 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
4377}
4378
4379/**
4380 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4381 * @dev: The device for which the DMA address was created
4382 * @addr: The DMA address
4383 * @size: The size of the region in bytes
4384 * @direction: The direction of the DMA
4385 */
4386static inline void ib_dma_unmap_single(struct ib_device *dev,
4387 u64 addr, size_t size,
4388 enum dma_data_direction direction)
4389{
4390 if (!ib_uses_virt_dma(dev))
4391 dma_unmap_single(dev->dma_device, addr, size, direction);
4392}
4393
4394/**
4395 * ib_dma_map_page - Map a physical page to DMA address
4396 * @dev: The device for which the dma_addr is to be created
4397 * @page: The page to be mapped
4398 * @offset: The offset within the page
4399 * @size: The size of the region in bytes
4400 * @direction: The direction of the DMA
4401 */
4402static inline u64 ib_dma_map_page(struct ib_device *dev,
4403 struct page *page,
4404 unsigned long offset,
4405 size_t size,
4406 enum dma_data_direction direction)
4407{
4408 if (ib_uses_virt_dma(dev))
4409 return (uintptr_t)(page_address(page) + offset);
4410 return dma_map_page(dev->dma_device, page, offset, size, direction);
4411}
4412
4413/**
4414 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4415 * @dev: The device for which the DMA address was created
4416 * @addr: The DMA address
4417 * @size: The size of the region in bytes
4418 * @direction: The direction of the DMA
4419 */
4420static inline void ib_dma_unmap_page(struct ib_device *dev,
4421 u64 addr, size_t size,
4422 enum dma_data_direction direction)
4423{
4424 if (!ib_uses_virt_dma(dev))
4425 dma_unmap_page(dev->dma_device, addr, size, direction);
4426}
4427
4428/**
4429 * ib_dma_map_bvec - Map a bio_vec to DMA address
4430 * @dev: The device for which the dma_addr is to be created
4431 * @bvec: The bio_vec to map
4432 * @direction: The direction of the DMA
4433 *
4434 * Returns a DMA address for the bio_vec. The caller must check the
4435 * result with ib_dma_mapping_error() before use; a failed mapping
4436 * must not be passed to ib_dma_unmap_bvec().
4437 *
4438 * For software RDMA devices (rxe, siw), returns a virtual address
4439 * and no actual DMA mapping occurs.
4440 */
4441static inline u64 ib_dma_map_bvec(struct ib_device *dev,
4442 struct bio_vec *bvec,
4443 enum dma_data_direction direction)
4444{
4445 if (ib_uses_virt_dma(dev))
4446 return (uintptr_t)bvec_virt(bvec);
4447 return dma_map_phys(dev->dma_device, bvec_phys(bvec),
4448 bvec->bv_len, direction, 0);
4449}
4450
4451/**
4452 * ib_dma_unmap_bvec - Unmap a bio_vec DMA mapping
4453 * @dev: The device for which the DMA address was created
4454 * @addr: The DMA address returned by ib_dma_map_bvec()
4455 * @size: The size of the region in bytes
4456 * @direction: The direction of the DMA
4457 *
4458 * Releases a DMA mapping created by ib_dma_map_bvec(). For software
4459 * RDMA devices this is a no-op since no actual mapping occurred.
4460 */
4461static inline void ib_dma_unmap_bvec(struct ib_device *dev,
4462 u64 addr, size_t size,
4463 enum dma_data_direction direction)
4464{
4465 if (!ib_uses_virt_dma(dev))
4466 dma_unmap_phys(dev->dma_device, addr, size, direction, 0);
4467}
4468
4469int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4470static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4471 struct scatterlist *sg, int nents,
4472 enum dma_data_direction direction,
4473 unsigned long dma_attrs)
4474{
4475 if (ib_uses_virt_dma(dev))
4476 return ib_dma_virt_map_sg(dev, sg, nents);
4477 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4478 dma_attrs);
4479}
4480
4481static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4482 struct scatterlist *sg, int nents,
4483 enum dma_data_direction direction,
4484 unsigned long dma_attrs)
4485{
4486 if (!ib_uses_virt_dma(dev))
4487 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4488 dma_attrs);
4489}
4490
4491/**
4492 * ib_dma_map_sgtable_attrs - Map a scatter/gather table to DMA addresses
4493 * @dev: The device for which the DMA addresses are to be created
4494 * @sgt: The sg_table object describing the buffer
4495 * @direction: The direction of the DMA
4496 * @dma_attrs: Optional DMA attributes for the map operation
4497 */
4498static inline int ib_dma_map_sgtable_attrs(struct ib_device *dev,
4499 struct sg_table *sgt,
4500 enum dma_data_direction direction,
4501 unsigned long dma_attrs)
4502{
4503 int nents;
4504
4505 if (ib_uses_virt_dma(dev)) {
4506 nents = ib_dma_virt_map_sg(dev, sgt->sgl, sgt->orig_nents);
4507 if (!nents)
4508 return -EIO;
4509 sgt->nents = nents;
4510 return 0;
4511 }
4512 return dma_map_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4513}
4514
4515static inline void ib_dma_unmap_sgtable_attrs(struct ib_device *dev,
4516 struct sg_table *sgt,
4517 enum dma_data_direction direction,
4518 unsigned long dma_attrs)
4519{
4520 if (!ib_uses_virt_dma(dev))
4521 dma_unmap_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4522}
4523
4524/**
4525 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4526 * @dev: The device for which the DMA addresses are to be created
4527 * @sg: The array of scatter/gather entries
4528 * @nents: The number of scatter/gather entries
4529 * @direction: The direction of the DMA
4530 */
4531static inline int ib_dma_map_sg(struct ib_device *dev,
4532 struct scatterlist *sg, int nents,
4533 enum dma_data_direction direction)
4534{
4535 return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4536}
4537
4538/**
4539 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4540 * @dev: The device for which the DMA addresses were created
4541 * @sg: The array of scatter/gather entries
4542 * @nents: The number of scatter/gather entries
4543 * @direction: The direction of the DMA
4544 */
4545static inline void ib_dma_unmap_sg(struct ib_device *dev,
4546 struct scatterlist *sg, int nents,
4547 enum dma_data_direction direction)
4548{
4549 ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4550}
4551
4552/**
4553 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4554 * @dev: The device to query
4555 *
4556 * The returned value represents a size in bytes.
4557 */
4558static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4559{
4560 if (ib_uses_virt_dma(dev))
4561 return UINT_MAX;
4562 return dma_get_max_seg_size(dev->dma_device);
4563}
4564
4565/**
4566 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4567 * @dev: The device for which the DMA address was created
4568 * @addr: The DMA address
4569 * @size: The size of the region in bytes
4570 * @dir: The direction of the DMA
4571 */
4572static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4573 u64 addr,
4574 size_t size,
4575 enum dma_data_direction dir)
4576{
4577 if (!ib_uses_virt_dma(dev))
4578 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4579}
4580
4581/**
4582 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4583 * @dev: The device for which the DMA address was created
4584 * @addr: The DMA address
4585 * @size: The size of the region in bytes
4586 * @dir: The direction of the DMA
4587 */
4588static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4589 u64 addr,
4590 size_t size,
4591 enum dma_data_direction dir)
4592{
4593 if (!ib_uses_virt_dma(dev))
4594 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4595}
4596
4597/* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4598 * space. This function should be called when 'current' is the owning MM.
4599 */
4600struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4601 u64 virt_addr, int mr_access_flags);
4602
4603/* ib_advise_mr - give an advice about an address range in a memory region */
4604int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4605 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4606/**
4607 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4608 * HCA translation table.
4609 * @mr: The memory region to deregister.
4610 * @udata: Valid user data or NULL for kernel object
4611 *
4612 * This function can fail, if the memory region has memory windows bound to it.
4613 */
4614int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4615
4616/**
4617 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4618 * HCA translation table.
4619 * @mr: The memory region to deregister.
4620 *
4621 * This function can fail, if the memory region has memory windows bound to it.
4622 *
4623 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4624 */
4625static inline int ib_dereg_mr(struct ib_mr *mr)
4626{
4627 return ib_dereg_mr_user(mr, NULL);
4628}
4629
4630struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4631 u32 max_num_sg);
4632
4633struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4634 u32 max_num_data_sg,
4635 u32 max_num_meta_sg);
4636
4637/**
4638 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4639 * R_Key and L_Key.
4640 * @mr: struct ib_mr pointer to be updated.
4641 * @newkey: new key to be used.
4642 */
4643static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4644{
4645 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4646 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4647}
4648
4649/**
4650 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4651 * for calculating a new rkey for type 2 memory windows.
4652 * @rkey: the rkey to increment.
4653 */
4654static inline u32 ib_inc_rkey(u32 rkey)
4655{
4656 const u32 mask = 0x000000ff;
4657 return ((rkey + 1) & mask) | (rkey & ~mask);
4658}
4659
4660/**
4661 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4662 * @qp: QP to attach to the multicast group. The QP must be type
4663 * IB_QPT_UD.
4664 * @gid: Multicast group GID.
4665 * @lid: Multicast group LID in host byte order.
4666 *
4667 * In order to send and receive multicast packets, subnet
4668 * administration must have created the multicast group and configured
4669 * the fabric appropriately. The port associated with the specified
4670 * QP must also be a member of the multicast group.
4671 */
4672int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4673
4674/**
4675 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4676 * @qp: QP to detach from the multicast group.
4677 * @gid: Multicast group GID.
4678 * @lid: Multicast group LID in host byte order.
4679 */
4680int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4681
4682struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4683 struct inode *inode, struct ib_udata *udata);
4684int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4685
4686static inline int ib_check_mr_access(struct ib_device *ib_dev,
4687 unsigned int flags)
4688{
4689 u64 device_cap = ib_dev->attrs.device_cap_flags;
4690
4691 /*
4692 * Local write permission is required if remote write or
4693 * remote atomic permission is also requested.
4694 */
4695 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4696 !(flags & IB_ACCESS_LOCAL_WRITE))
4697 return -EINVAL;
4698
4699 if (flags & ~IB_ACCESS_SUPPORTED)
4700 return -EINVAL;
4701
4702 if (flags & IB_ACCESS_ON_DEMAND &&
4703 !(ib_dev->attrs.kernel_cap_flags & IBK_ON_DEMAND_PAGING))
4704 return -EOPNOTSUPP;
4705
4706 if ((flags & IB_ACCESS_FLUSH_GLOBAL &&
4707 !(device_cap & IB_DEVICE_FLUSH_GLOBAL)) ||
4708 (flags & IB_ACCESS_FLUSH_PERSISTENT &&
4709 !(device_cap & IB_DEVICE_FLUSH_PERSISTENT)))
4710 return -EOPNOTSUPP;
4711
4712 return 0;
4713}
4714
4715static inline bool ib_access_writable(int access_flags)
4716{
4717 /*
4718 * We have writable memory backing the MR if any of the following
4719 * access flags are set. "Local write" and "remote write" obviously
4720 * require write access. "Remote atomic" can do things like fetch and
4721 * add, which will modify memory, and "MW bind" can change permissions
4722 * by binding a window.
4723 */
4724 return access_flags &
4725 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4726 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4727}
4728
4729/**
4730 * ib_check_mr_status: lightweight check of MR status.
4731 * This routine may provide status checks on a selected
4732 * ib_mr. first use is for signature status check.
4733 *
4734 * @mr: A memory region.
4735 * @check_mask: Bitmask of which checks to perform from
4736 * ib_mr_status_check enumeration.
4737 * @mr_status: The container of relevant status checks.
4738 * failed checks will be indicated in the status bitmask
4739 * and the relevant info shall be in the error item.
4740 */
4741int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4742 struct ib_mr_status *mr_status);
4743
4744/**
4745 * ib_device_try_get: Hold a registration lock
4746 * @dev: The device to lock
4747 *
4748 * A device under an active registration lock cannot become unregistered. It
4749 * is only possible to obtain a registration lock on a device that is fully
4750 * registered, otherwise this function returns false.
4751 *
4752 * The registration lock is only necessary for actions which require the
4753 * device to still be registered. Uses that only require the device pointer to
4754 * be valid should use get_device(&ibdev->dev) to hold the memory.
4755 *
4756 */
4757static inline bool ib_device_try_get(struct ib_device *dev)
4758{
4759 return refcount_inc_not_zero(&dev->refcount);
4760}
4761
4762void ib_device_put(struct ib_device *device);
4763struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4764 enum rdma_driver_id driver_id);
4765struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port,
4766 u16 pkey, const union ib_gid *gid,
4767 const struct sockaddr *addr);
4768int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4769 unsigned int port);
4770struct net_device *ib_device_get_netdev(struct ib_device *ib_dev,
4771 u32 port);
4772int ib_query_netdev_port(struct ib_device *ibdev, struct net_device *ndev,
4773 u32 *port);
4774
4775static inline enum ib_port_state ib_get_curr_port_state(struct net_device *net_dev)
4776{
4777 return (netif_running(net_dev) && netif_carrier_ok(net_dev)) ?
4778 IB_PORT_ACTIVE : IB_PORT_DOWN;
4779}
4780
4781void ib_dispatch_port_state_event(struct ib_device *ibdev,
4782 struct net_device *ndev);
4783struct ib_wq *ib_create_wq(struct ib_pd *pd,
4784 struct ib_wq_init_attr *init_attr);
4785int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4786
4787int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4788 unsigned int *sg_offset, unsigned int page_size);
4789int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4790 int data_sg_nents, unsigned int *data_sg_offset,
4791 struct scatterlist *meta_sg, int meta_sg_nents,
4792 unsigned int *meta_sg_offset, unsigned int page_size);
4793
4794static inline int
4795ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4796 unsigned int *sg_offset, unsigned int page_size)
4797{
4798 int n;
4799
4800 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4801 mr->iova = 0;
4802
4803 return n;
4804}
4805
4806int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4807 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4808
4809void ib_drain_rq(struct ib_qp *qp);
4810void ib_drain_sq(struct ib_qp *qp);
4811void ib_drain_qp(struct ib_qp *qp);
4812
4813int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed,
4814 u8 *width);
4815
4816static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4817{
4818 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4819 return attr->roce.dmac;
4820 return NULL;
4821}
4822
4823static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4824{
4825 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4826 attr->ib.dlid = (u16)dlid;
4827 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4828 attr->opa.dlid = dlid;
4829}
4830
4831static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4832{
4833 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4834 return attr->ib.dlid;
4835 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4836 return attr->opa.dlid;
4837 return 0;
4838}
4839
4840static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4841{
4842 attr->sl = sl;
4843}
4844
4845static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4846{
4847 return attr->sl;
4848}
4849
4850static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4851 u8 src_path_bits)
4852{
4853 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4854 attr->ib.src_path_bits = src_path_bits;
4855 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4856 attr->opa.src_path_bits = src_path_bits;
4857}
4858
4859static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4860{
4861 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4862 return attr->ib.src_path_bits;
4863 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4864 return attr->opa.src_path_bits;
4865 return 0;
4866}
4867
4868static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4869 bool make_grd)
4870{
4871 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4872 attr->opa.make_grd = make_grd;
4873}
4874
4875static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4876{
4877 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4878 return attr->opa.make_grd;
4879 return false;
4880}
4881
4882static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num)
4883{
4884 attr->port_num = port_num;
4885}
4886
4887static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4888{
4889 return attr->port_num;
4890}
4891
4892static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4893 u8 static_rate)
4894{
4895 attr->static_rate = static_rate;
4896}
4897
4898static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4899{
4900 return attr->static_rate;
4901}
4902
4903static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4904 enum ib_ah_flags flag)
4905{
4906 attr->ah_flags = flag;
4907}
4908
4909static inline enum ib_ah_flags
4910 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4911{
4912 return attr->ah_flags;
4913}
4914
4915static inline const struct ib_global_route
4916 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4917{
4918 return &attr->grh;
4919}
4920
4921/*To retrieve and modify the grh */
4922static inline struct ib_global_route
4923 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4924{
4925 return &attr->grh;
4926}
4927
4928static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4929{
4930 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4931
4932 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4933}
4934
4935static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4936 __be64 prefix)
4937{
4938 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4939
4940 grh->dgid.global.subnet_prefix = prefix;
4941}
4942
4943static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4944 __be64 if_id)
4945{
4946 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4947
4948 grh->dgid.global.interface_id = if_id;
4949}
4950
4951static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4952 union ib_gid *dgid, u32 flow_label,
4953 u8 sgid_index, u8 hop_limit,
4954 u8 traffic_class)
4955{
4956 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4957
4958 attr->ah_flags = IB_AH_GRH;
4959 if (dgid)
4960 grh->dgid = *dgid;
4961 grh->flow_label = flow_label;
4962 grh->sgid_index = sgid_index;
4963 grh->hop_limit = hop_limit;
4964 grh->traffic_class = traffic_class;
4965 grh->sgid_attr = NULL;
4966}
4967
4968void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4969void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4970 u32 flow_label, u8 hop_limit, u8 traffic_class,
4971 const struct ib_gid_attr *sgid_attr);
4972void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4973 const struct rdma_ah_attr *src);
4974void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4975 const struct rdma_ah_attr *new);
4976void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4977
4978/**
4979 * rdma_ah_find_type - Return address handle type.
4980 *
4981 * @dev: Device to be checked
4982 * @port_num: Port number
4983 */
4984static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4985 u32 port_num)
4986{
4987 if (rdma_protocol_roce(dev, port_num))
4988 return RDMA_AH_ATTR_TYPE_ROCE;
4989 if (rdma_protocol_ib(dev, port_num)) {
4990 if (rdma_cap_opa_ah(dev, port_num))
4991 return RDMA_AH_ATTR_TYPE_OPA;
4992 return RDMA_AH_ATTR_TYPE_IB;
4993 }
4994 if (dev->type == RDMA_DEVICE_TYPE_SMI)
4995 return RDMA_AH_ATTR_TYPE_IB;
4996
4997 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4998}
4999
5000/**
5001 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
5002 * In the current implementation the only way to
5003 * get the 32bit lid is from other sources for OPA.
5004 * For IB, lids will always be 16bits so cast the
5005 * value accordingly.
5006 *
5007 * @lid: A 32bit LID
5008 */
5009static inline u16 ib_lid_cpu16(u32 lid)
5010{
5011 WARN_ON_ONCE(lid & 0xFFFF0000);
5012 return (u16)lid;
5013}
5014
5015/**
5016 * ib_lid_be16 - Return lid in 16bit BE encoding.
5017 *
5018 * @lid: A 32bit LID
5019 */
5020static inline __be16 ib_lid_be16(u32 lid)
5021{
5022 WARN_ON_ONCE(lid & 0xFFFF0000);
5023 return cpu_to_be16((u16)lid);
5024}
5025
5026/**
5027 * rdma_roce_rescan_device - Rescan all of the network devices in the system
5028 * and add their gids, as needed, to the relevant RoCE devices.
5029 *
5030 * @ibdev: the rdma device
5031 */
5032void rdma_roce_rescan_device(struct ib_device *ibdev);
5033void rdma_roce_rescan_port(struct ib_device *ib_dev, u32 port);
5034void roce_del_all_netdev_gids(struct ib_device *ib_dev,
5035 u32 port, struct net_device *ndev);
5036
5037struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
5038
5039#if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
5040int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
5041bool rdma_uattrs_has_raw_cap(const struct uverbs_attr_bundle *attrs);
5042#else
5043static inline int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs)
5044{
5045 return 0;
5046}
5047static inline bool
5048rdma_uattrs_has_raw_cap(const struct uverbs_attr_bundle *attrs)
5049{
5050 return false;
5051}
5052#endif
5053
5054struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num,
5055 enum rdma_netdev_t type, const char *name,
5056 unsigned char name_assign_type,
5057 void (*setup)(struct net_device *));
5058
5059int rdma_init_netdev(struct ib_device *device, u32 port_num,
5060 enum rdma_netdev_t type, const char *name,
5061 unsigned char name_assign_type,
5062 void (*setup)(struct net_device *),
5063 struct net_device *netdev);
5064
5065/**
5066 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
5067 *
5068 * @device: device pointer for which ib_device pointer to retrieve
5069 *
5070 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
5071 *
5072 */
5073static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
5074{
5075 struct ib_core_device *coredev =
5076 container_of(device, struct ib_core_device, dev);
5077
5078 return coredev->owner;
5079}
5080
5081/**
5082 * ibdev_to_node - return the NUMA node for a given ib_device
5083 * @ibdev: device to get the NUMA node for.
5084 */
5085static inline int ibdev_to_node(struct ib_device *ibdev)
5086{
5087 struct device *parent = ibdev->dev.parent;
5088
5089 if (!parent)
5090 return NUMA_NO_NODE;
5091 return dev_to_node(parent);
5092}
5093
5094/**
5095 * rdma_device_to_drv_device - Helper macro to reach back to driver's
5096 * ib_device holder structure from device pointer.
5097 *
5098 * NOTE: New drivers should not make use of this API; This API is only for
5099 * existing drivers who have exposed sysfs entries using
5100 * ops->device_group.
5101 */
5102#define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
5103 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
5104
5105bool rdma_dev_access_netns(const struct ib_device *device,
5106 const struct net *net);
5107
5108bool rdma_dev_has_raw_cap(const struct ib_device *dev);
5109static inline struct net *rdma_dev_net(struct ib_device *device)
5110{
5111 return read_pnet(&device->coredev.rdma_net);
5112}
5113
5114#define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
5115#define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
5116#define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
5117
5118/**
5119 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
5120 * on the flow_label
5121 * @fl: flow_label value
5122 *
5123 * This function will convert the 20 bit flow_label input to a valid RoCE v2
5124 * UDP src port 14 bit value. All RoCE V2 drivers should use this same
5125 * convention.
5126 */
5127static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
5128{
5129 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
5130
5131 fl_low ^= fl_high >> 14;
5132 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
5133}
5134
5135/**
5136 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
5137 * local and remote qpn values
5138 *
5139 * This function folded the multiplication results of two qpns, 24 bit each,
5140 * fields, and converts it to a 20 bit results.
5141 *
5142 * This function will create symmetric flow_label value based on the local
5143 * and remote qpn values. this will allow both the requester and responder
5144 * to calculate the same flow_label for a given connection.
5145 *
5146 * This helper function should be used by driver in case the upper layer
5147 * provide a zero flow_label value. This is to improve entropy of RDMA
5148 * traffic in the network.
5149 */
5150static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
5151{
5152 u64 v = (u64)lqpn * rqpn;
5153
5154 v ^= v >> 20;
5155 v ^= v >> 40;
5156
5157 return (u32)(v & IB_GRH_FLOWLABEL_MASK);
5158}
5159
5160/**
5161 * rdma_get_udp_sport - Calculate and set UDP source port based on the flow
5162 * label. If flow label is not defined in GRH then
5163 * calculate it based on lqpn/rqpn.
5164 *
5165 * @fl: flow label from GRH
5166 * @lqpn: local qp number
5167 * @rqpn: remote qp number
5168 */
5169static inline u16 rdma_get_udp_sport(u32 fl, u32 lqpn, u32 rqpn)
5170{
5171 if (!fl)
5172 fl = rdma_calc_flow_label(lqpn, rqpn);
5173
5174 return rdma_flow_label_to_udp_sport(fl);
5175}
5176
5177const struct ib_port_immutable*
5178ib_port_immutable_read(struct ib_device *dev, unsigned int port);
5179
5180/** ib_add_sub_device - Add a sub IB device on an existing one
5181 *
5182 * @parent: The IB device that needs to add a sub device
5183 * @type: The type of the new sub device
5184 * @name: The name of the new sub device
5185 *
5186 *
5187 * Return 0 on success, an error code otherwise
5188 */
5189int ib_add_sub_device(struct ib_device *parent,
5190 enum rdma_nl_dev_type type,
5191 const char *name);
5192
5193
5194/** ib_del_sub_device_and_put - Delect an IB sub device while holding a 'get'
5195 *
5196 * @sub: The sub device that is going to be deleted
5197 *
5198 * Return 0 on success, an error code otherwise
5199 */
5200int ib_del_sub_device_and_put(struct ib_device *sub);
5201
5202static inline void ib_mark_name_assigned_by_user(struct ib_device *ibdev)
5203{
5204 ibdev->name_assign_type = RDMA_NAME_ASSIGN_TYPE_USER;
5205}
5206
5207#endif /* IB_VERBS_H */