blob: 432bed510369e7b19a4773ec5a8bb09b8d9bb1e6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
3 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
4 * Copyright (c) 2004 Intel Corporation. All rights reserved.
5 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
6 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
Roland Dreier2a1d9b72005-08-10 23:03:10 -07007 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
Roland Dreierf7c6a7b2007-03-04 16:15:11 -08008 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 *
10 * This software is available to you under a choice of one of two
11 * licenses. You may choose to be licensed under the terms of the GNU
12 * General Public License (GPL) Version 2, available from the file
13 * COPYING in the main directory of this source tree, or the
14 * OpenIB.org BSD license below:
15 *
16 * Redistribution and use in source and binary forms, with or
17 * without modification, are permitted provided that the following
18 * conditions are met:
19 *
20 * - Redistributions of source code must retain the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer.
23 *
24 * - Redistributions in binary form must reproduce the above
25 * copyright notice, this list of conditions and the following
26 * disclaimer in the documentation and/or other materials
27 * provided with the distribution.
28 *
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36 * SOFTWARE.
Linus Torvalds1da177e2005-04-16 15:20:36 -070037 */
38
39#if !defined(IB_VERBS_H)
40#define IB_VERBS_H
41
42#include <linux/types.h>
43#include <linux/device.h>
Ralph Campbell9b513092006-12-12 14:27:41 -080044#include <linux/mm.h>
45#include <linux/dma-mapping.h>
Michael S. Tsirkin459d6e22007-02-04 14:11:55 -080046#include <linux/kref.h>
Dotan Barakbfb3ea12007-07-31 16:49:15 +030047#include <linux/list.h>
48#include <linux/rwsem.h>
Adrian Bunk87ae9af2007-10-30 10:35:04 +010049#include <linux/scatterlist.h>
Tejun Heof0626712010-10-19 15:24:36 +000050#include <linux/workqueue.h>
Yotam Kenneth9268f722015-07-30 17:50:15 +030051#include <linux/socket.h>
Christoph Hellwig14d3a3b2015-12-11 11:53:03 -080052#include <linux/irq_poll.h>
Matan Barakdd5f03b2013-12-12 18:03:11 +020053#include <uapi/linux/if_ether.h>
Somnath Koturc865f242015-12-23 14:56:51 +020054#include <net/ipv6.h>
55#include <net/ip.h>
Matan Barak301a7212015-12-15 20:30:10 +020056#include <linux/string.h>
57#include <linux/slab.h>
Roland Dreiere2773c02005-07-07 17:57:10 -070058
Eli Cohen50174a72016-03-11 22:58:38 +020059#include <linux/if_link.h>
Arun Sharma600634972011-07-26 16:09:06 -070060#include <linux/atomic.h>
Haggai Eran882214e2014-12-11 17:04:18 +020061#include <linux/mmu_notifier.h>
Roland Dreiere2773c02005-07-07 17:57:10 -070062#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
Tejun Heof0626712010-10-19 15:24:36 +000064extern struct workqueue_struct *ib_wq;
Christoph Hellwig14d3a3b2015-12-11 11:53:03 -080065extern struct workqueue_struct *ib_comp_wq;
Tejun Heof0626712010-10-19 15:24:36 +000066
Linus Torvalds1da177e2005-04-16 15:20:36 -070067union ib_gid {
68 u8 raw[16];
69 struct {
Sean Hefty97f52eb2005-08-13 21:05:57 -070070 __be64 subnet_prefix;
71 __be64 interface_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 } global;
73};
74
Moni Shouae26be1b2015-07-30 18:33:29 +030075extern union ib_gid zgid;
76
Matan Barakb39ffa12015-12-23 14:56:47 +020077enum ib_gid_type {
78 /* If link layer is Ethernet, this is RoCE V1 */
79 IB_GID_TYPE_IB = 0,
80 IB_GID_TYPE_ROCE = 0,
Matan Barak7766a992015-12-23 14:56:50 +020081 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
Matan Barakb39ffa12015-12-23 14:56:47 +020082 IB_GID_TYPE_SIZE
83};
84
Moni Shoua7ead4bc2016-01-14 17:50:38 +020085#define ROCE_V2_UDP_DPORT 4791
Matan Barak03db3a22015-07-30 18:33:26 +030086struct ib_gid_attr {
Matan Barakb39ffa12015-12-23 14:56:47 +020087 enum ib_gid_type gid_type;
Matan Barak03db3a22015-07-30 18:33:26 +030088 struct net_device *ndev;
89};
90
Tom Tucker07ebafb2006-08-03 16:02:42 -050091enum rdma_node_type {
92 /* IB values map to NodeInfo:NodeType. */
93 RDMA_NODE_IB_CA = 1,
94 RDMA_NODE_IB_SWITCH,
95 RDMA_NODE_IB_ROUTER,
Upinder Malhi \(umalhi\)180771a2013-09-10 03:36:59 +000096 RDMA_NODE_RNIC,
97 RDMA_NODE_USNIC,
Upinder Malhi5db57652014-01-15 17:02:36 -080098 RDMA_NODE_USNIC_UDP,
Linus Torvalds1da177e2005-04-16 15:20:36 -070099};
100
Eli Cohena0c1b2a2016-03-11 22:58:37 +0200101enum {
102 /* set the local administered indication */
103 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
104};
105
Tom Tucker07ebafb2006-08-03 16:02:42 -0500106enum rdma_transport_type {
107 RDMA_TRANSPORT_IB,
Upinder Malhi \(umalhi\)180771a2013-09-10 03:36:59 +0000108 RDMA_TRANSPORT_IWARP,
Upinder Malhi248567f2014-01-09 14:48:19 -0800109 RDMA_TRANSPORT_USNIC,
110 RDMA_TRANSPORT_USNIC_UDP
Tom Tucker07ebafb2006-08-03 16:02:42 -0500111};
112
Michael Wang6b90a6d2015-05-05 14:50:18 +0200113enum rdma_protocol_type {
114 RDMA_PROTOCOL_IB,
115 RDMA_PROTOCOL_IBOE,
116 RDMA_PROTOCOL_IWARP,
117 RDMA_PROTOCOL_USNIC_UDP
118};
119
Roland Dreier8385fd82014-06-04 10:00:16 -0700120__attribute_const__ enum rdma_transport_type
121rdma_node_get_transport(enum rdma_node_type node_type);
Tom Tucker07ebafb2006-08-03 16:02:42 -0500122
Somnath Koturc865f242015-12-23 14:56:51 +0200123enum rdma_network_type {
124 RDMA_NETWORK_IB,
125 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
126 RDMA_NETWORK_IPV4,
127 RDMA_NETWORK_IPV6
128};
129
130static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
131{
132 if (network_type == RDMA_NETWORK_IPV4 ||
133 network_type == RDMA_NETWORK_IPV6)
134 return IB_GID_TYPE_ROCE_UDP_ENCAP;
135
136 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
137 return IB_GID_TYPE_IB;
138}
139
140static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
141 union ib_gid *gid)
142{
143 if (gid_type == IB_GID_TYPE_IB)
144 return RDMA_NETWORK_IB;
145
146 if (ipv6_addr_v4mapped((struct in6_addr *)gid))
147 return RDMA_NETWORK_IPV4;
148 else
149 return RDMA_NETWORK_IPV6;
150}
151
Eli Cohena3f5ada2010-09-27 17:51:10 -0700152enum rdma_link_layer {
153 IB_LINK_LAYER_UNSPECIFIED,
154 IB_LINK_LAYER_INFINIBAND,
155 IB_LINK_LAYER_ETHERNET,
156};
157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158enum ib_device_cap_flags {
Leon Romanovsky7ca0bc52015-12-20 12:16:09 +0200159 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
160 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
161 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
162 IB_DEVICE_RAW_MULTI = (1 << 3),
163 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
164 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
165 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
166 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
167 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
168 IB_DEVICE_INIT_TYPE = (1 << 9),
169 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
170 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
171 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
172 IB_DEVICE_SRQ_RESIZE = (1 << 13),
173 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
Christoph Hellwigb1adc712015-12-23 19:12:45 +0100174
175 /*
176 * This device supports a per-device lkey or stag that can be
177 * used without performing a memory registration for the local
178 * memory. Note that ULPs should never check this flag, but
179 * instead of use the local_dma_lkey flag in the ib_pd structure,
180 * which will always contain a usable lkey.
181 */
Leon Romanovsky7ca0bc52015-12-20 12:16:09 +0200182 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
183 IB_DEVICE_RESERVED /* old SEND_W_INV */ = (1 << 16),
184 IB_DEVICE_MEM_WINDOW = (1 << 17),
Eli Cohene0605d92008-01-30 18:30:57 +0200185 /*
186 * Devices should set IB_DEVICE_UD_IP_SUM if they support
187 * insertion of UDP and TCP checksum on outgoing UD IPoIB
188 * messages and can verify the validity of checksum for
189 * incoming messages. Setting this flag implies that the
190 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
191 */
Leon Romanovsky7ca0bc52015-12-20 12:16:09 +0200192 IB_DEVICE_UD_IP_CSUM = (1 << 18),
193 IB_DEVICE_UD_TSO = (1 << 19),
194 IB_DEVICE_XRC = (1 << 20),
Christoph Hellwigb1adc712015-12-23 19:12:45 +0100195
196 /*
197 * This device supports the IB "base memory management extension",
198 * which includes support for fast registrations (IB_WR_REG_MR,
199 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
200 * also be set by any iWarp device which must support FRs to comply
201 * to the iWarp verbs spec. iWarp devices also support the
202 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
203 * stag.
204 */
Leon Romanovsky7ca0bc52015-12-20 12:16:09 +0200205 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
206 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
207 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
208 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
209 IB_DEVICE_RC_IP_CSUM = (1 << 25),
210 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
Leon Romanovsky8a06ce52015-12-20 12:16:10 +0200211 /*
212 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
213 * support execution of WQEs that involve synchronization
214 * of I/O operations with single completion queue managed
215 * by hardware.
216 */
217 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
Leon Romanovsky7ca0bc52015-12-20 12:16:09 +0200218 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
219 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30),
220 IB_DEVICE_ON_DEMAND_PAGING = (1 << 31),
Sagi Grimbergf5aa9152016-02-29 19:07:32 +0200221 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
Eli Cohen50174a72016-03-11 22:58:38 +0200222 IB_DEVICE_VIRTUAL_FUNCTION = ((u64)1 << 33),
Majd Dibbiny45686f22016-04-17 17:19:35 +0300223 IB_DEVICE_RAW_SCATTER_FCS = ((u64)1 << 34),
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200224};
225
226enum ib_signature_prot_cap {
227 IB_PROT_T10DIF_TYPE_1 = 1,
228 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
229 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
230};
231
232enum ib_signature_guard_cap {
233 IB_GUARD_T10DIF_CRC = 1,
234 IB_GUARD_T10DIF_CSUM = 1 << 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235};
236
237enum ib_atomic_cap {
238 IB_ATOMIC_NONE,
239 IB_ATOMIC_HCA,
240 IB_ATOMIC_GLOB
241};
242
Sagi Grimberg860f10a2014-12-11 17:04:16 +0200243enum ib_odp_general_cap_bits {
244 IB_ODP_SUPPORT = 1 << 0,
245};
246
247enum ib_odp_transport_cap_bits {
248 IB_ODP_SUPPORT_SEND = 1 << 0,
249 IB_ODP_SUPPORT_RECV = 1 << 1,
250 IB_ODP_SUPPORT_WRITE = 1 << 2,
251 IB_ODP_SUPPORT_READ = 1 << 3,
252 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
253};
254
255struct ib_odp_caps {
256 uint64_t general_caps;
257 struct {
258 uint32_t rc_odp_caps;
259 uint32_t uc_odp_caps;
260 uint32_t ud_odp_caps;
261 } per_transport_caps;
262};
263
Matan Barakb9926b92015-06-11 16:35:22 +0300264enum ib_cq_creation_flags {
265 IB_CQ_FLAGS_TIMESTAMP_COMPLETION = 1 << 0,
Leon Romanovsky8a06ce52015-12-20 12:16:10 +0200266 IB_CQ_FLAGS_IGNORE_OVERRUN = 1 << 1,
Matan Barakb9926b92015-06-11 16:35:22 +0300267};
268
Matan Barakbcf4c1e2015-06-11 16:35:20 +0300269struct ib_cq_init_attr {
270 unsigned int cqe;
271 int comp_vector;
272 u32 flags;
273};
274
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275struct ib_device_attr {
276 u64 fw_ver;
Sean Hefty97f52eb2005-08-13 21:05:57 -0700277 __be64 sys_image_guid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 u64 max_mr_size;
279 u64 page_size_cap;
280 u32 vendor_id;
281 u32 vendor_part_id;
282 u32 hw_ver;
283 int max_qp;
284 int max_qp_wr;
Leon Romanovskyfb532d62016-02-23 10:25:25 +0200285 u64 device_cap_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 int max_sge;
287 int max_sge_rd;
288 int max_cq;
289 int max_cqe;
290 int max_mr;
291 int max_pd;
292 int max_qp_rd_atom;
293 int max_ee_rd_atom;
294 int max_res_rd_atom;
295 int max_qp_init_rd_atom;
296 int max_ee_init_rd_atom;
297 enum ib_atomic_cap atomic_cap;
Vladimir Sokolovsky5e80ba82010-04-14 17:23:01 +0300298 enum ib_atomic_cap masked_atomic_cap;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 int max_ee;
300 int max_rdd;
301 int max_mw;
302 int max_raw_ipv6_qp;
303 int max_raw_ethy_qp;
304 int max_mcast_grp;
305 int max_mcast_qp_attach;
306 int max_total_mcast_qp_attach;
307 int max_ah;
308 int max_fmr;
309 int max_map_per_fmr;
310 int max_srq;
311 int max_srq_wr;
312 int max_srq_sge;
Steve Wise00f7ec32008-07-14 23:48:45 -0700313 unsigned int max_fast_reg_page_list_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 u16 max_pkeys;
315 u8 local_ca_ack_delay;
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200316 int sig_prot_cap;
317 int sig_guard_cap;
Sagi Grimberg860f10a2014-12-11 17:04:16 +0200318 struct ib_odp_caps odp_caps;
Matan Barak24306dc2015-06-11 16:35:24 +0300319 uint64_t timestamp_mask;
320 uint64_t hca_core_clock; /* in KHZ */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321};
322
323enum ib_mtu {
324 IB_MTU_256 = 1,
325 IB_MTU_512 = 2,
326 IB_MTU_1024 = 3,
327 IB_MTU_2048 = 4,
328 IB_MTU_4096 = 5
329};
330
331static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
332{
333 switch (mtu) {
334 case IB_MTU_256: return 256;
335 case IB_MTU_512: return 512;
336 case IB_MTU_1024: return 1024;
337 case IB_MTU_2048: return 2048;
338 case IB_MTU_4096: return 4096;
339 default: return -1;
340 }
341}
342
343enum ib_port_state {
344 IB_PORT_NOP = 0,
345 IB_PORT_DOWN = 1,
346 IB_PORT_INIT = 2,
347 IB_PORT_ARMED = 3,
348 IB_PORT_ACTIVE = 4,
349 IB_PORT_ACTIVE_DEFER = 5
350};
351
352enum ib_port_cap_flags {
353 IB_PORT_SM = 1 << 1,
354 IB_PORT_NOTICE_SUP = 1 << 2,
355 IB_PORT_TRAP_SUP = 1 << 3,
356 IB_PORT_OPT_IPD_SUP = 1 << 4,
357 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
358 IB_PORT_SL_MAP_SUP = 1 << 6,
359 IB_PORT_MKEY_NVRAM = 1 << 7,
360 IB_PORT_PKEY_NVRAM = 1 << 8,
361 IB_PORT_LED_INFO_SUP = 1 << 9,
362 IB_PORT_SM_DISABLED = 1 << 10,
363 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
364 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
Marcel Apfelbaum71eeba12011-10-05 14:21:47 +0300365 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 IB_PORT_CM_SUP = 1 << 16,
367 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
368 IB_PORT_REINIT_SUP = 1 << 18,
369 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
370 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
371 IB_PORT_DR_NOTICE_SUP = 1 << 21,
372 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
373 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
374 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
Moni Shouab4a26a22014-02-09 11:54:34 +0200375 IB_PORT_CLIENT_REG_SUP = 1 << 25,
Matan Barak03db3a22015-07-30 18:33:26 +0300376 IB_PORT_IP_BASED_GIDS = 1 << 26,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377};
378
379enum ib_port_width {
380 IB_WIDTH_1X = 1,
381 IB_WIDTH_4X = 2,
382 IB_WIDTH_8X = 4,
383 IB_WIDTH_12X = 8
384};
385
386static inline int ib_width_enum_to_int(enum ib_port_width width)
387{
388 switch (width) {
389 case IB_WIDTH_1X: return 1;
390 case IB_WIDTH_4X: return 4;
391 case IB_WIDTH_8X: return 8;
392 case IB_WIDTH_12X: return 12;
393 default: return -1;
394 }
395}
396
Or Gerlitz2e966912012-02-28 18:49:50 +0200397enum ib_port_speed {
398 IB_SPEED_SDR = 1,
399 IB_SPEED_DDR = 2,
400 IB_SPEED_QDR = 4,
401 IB_SPEED_FDR10 = 8,
402 IB_SPEED_FDR = 16,
403 IB_SPEED_EDR = 32
404};
405
Christoph Lameterb40f4752016-05-16 12:49:33 -0500406/**
407 * struct rdma_hw_stats
408 * @timestamp - Used by the core code to track when the last update was
409 * @lifespan - Used by the core code to determine how old the counters
410 * should be before being updated again. Stored in jiffies, defaults
411 * to 10 milliseconds, drivers can override the default be specifying
412 * their own value during their allocation routine.
413 * @name - Array of pointers to static names used for the counters in
414 * directory.
415 * @num_counters - How many hardware counters there are. If name is
416 * shorter than this number, a kernel oops will result. Driver authors
417 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
418 * in their code to prevent this.
419 * @value - Array of u64 counters that are accessed by the sysfs code and
420 * filled in by the drivers get_stats routine
421 */
422struct rdma_hw_stats {
423 unsigned long timestamp;
424 unsigned long lifespan;
425 const char * const *names;
426 int num_counters;
427 u64 value[];
Steve Wise7f624d02008-07-14 23:48:48 -0700428};
429
Christoph Lameterb40f4752016-05-16 12:49:33 -0500430#define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
431/**
432 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
433 * for drivers.
434 * @names - Array of static const char *
435 * @num_counters - How many elements in array
436 * @lifespan - How many milliseconds between updates
437 */
438static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
439 const char * const *names, int num_counters,
440 unsigned long lifespan)
441{
442 struct rdma_hw_stats *stats;
Steve Wise7f624d02008-07-14 23:48:48 -0700443
Christoph Lameterb40f4752016-05-16 12:49:33 -0500444 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
445 GFP_KERNEL);
446 if (!stats)
447 return NULL;
448 stats->names = names;
449 stats->num_counters = num_counters;
450 stats->lifespan = msecs_to_jiffies(lifespan);
Steve Wise7f624d02008-07-14 23:48:48 -0700451
Christoph Lameterb40f4752016-05-16 12:49:33 -0500452 return stats;
453}
454
Steve Wise7f624d02008-07-14 23:48:48 -0700455
Ira Weinyf9b22e32015-05-13 20:02:59 -0400456/* Define bits for the various functionality this port needs to be supported by
457 * the core.
458 */
459/* Management 0x00000FFF */
460#define RDMA_CORE_CAP_IB_MAD 0x00000001
461#define RDMA_CORE_CAP_IB_SMI 0x00000002
462#define RDMA_CORE_CAP_IB_CM 0x00000004
463#define RDMA_CORE_CAP_IW_CM 0x00000008
464#define RDMA_CORE_CAP_IB_SA 0x00000010
Ira Weiny65995fe2015-06-06 14:38:32 -0400465#define RDMA_CORE_CAP_OPA_MAD 0x00000020
Ira Weinyf9b22e32015-05-13 20:02:59 -0400466
467/* Address format 0x000FF000 */
468#define RDMA_CORE_CAP_AF_IB 0x00001000
469#define RDMA_CORE_CAP_ETH_AH 0x00002000
470
471/* Protocol 0xFFF00000 */
472#define RDMA_CORE_CAP_PROT_IB 0x00100000
473#define RDMA_CORE_CAP_PROT_ROCE 0x00200000
474#define RDMA_CORE_CAP_PROT_IWARP 0x00400000
Matan Barak7766a992015-12-23 14:56:50 +0200475#define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
Ira Weinyf9b22e32015-05-13 20:02:59 -0400476
477#define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
478 | RDMA_CORE_CAP_IB_MAD \
479 | RDMA_CORE_CAP_IB_SMI \
480 | RDMA_CORE_CAP_IB_CM \
481 | RDMA_CORE_CAP_IB_SA \
482 | RDMA_CORE_CAP_AF_IB)
483#define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
484 | RDMA_CORE_CAP_IB_MAD \
485 | RDMA_CORE_CAP_IB_CM \
Ira Weinyf9b22e32015-05-13 20:02:59 -0400486 | RDMA_CORE_CAP_AF_IB \
487 | RDMA_CORE_CAP_ETH_AH)
Matan Barak7766a992015-12-23 14:56:50 +0200488#define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
489 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
490 | RDMA_CORE_CAP_IB_MAD \
491 | RDMA_CORE_CAP_IB_CM \
492 | RDMA_CORE_CAP_AF_IB \
493 | RDMA_CORE_CAP_ETH_AH)
Ira Weinyf9b22e32015-05-13 20:02:59 -0400494#define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
495 | RDMA_CORE_CAP_IW_CM)
Ira Weiny65995fe2015-06-06 14:38:32 -0400496#define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
497 | RDMA_CORE_CAP_OPA_MAD)
Ira Weinyf9b22e32015-05-13 20:02:59 -0400498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499struct ib_port_attr {
Eli Cohenfad61ad2016-03-11 22:58:36 +0200500 u64 subnet_prefix;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 enum ib_port_state state;
502 enum ib_mtu max_mtu;
503 enum ib_mtu active_mtu;
504 int gid_tbl_len;
505 u32 port_cap_flags;
506 u32 max_msg_sz;
507 u32 bad_pkey_cntr;
508 u32 qkey_viol_cntr;
509 u16 pkey_tbl_len;
510 u16 lid;
511 u16 sm_lid;
512 u8 lmc;
513 u8 max_vl_num;
514 u8 sm_sl;
515 u8 subnet_timeout;
516 u8 init_type_reply;
517 u8 active_width;
518 u8 active_speed;
519 u8 phys_state;
Eli Cohena0c1b2a2016-03-11 22:58:37 +0200520 bool grh_required;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521};
522
523enum ib_device_modify_flags {
Roland Dreierc5bcbbb2006-02-02 09:47:14 -0800524 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
525 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526};
527
528struct ib_device_modify {
529 u64 sys_image_guid;
Roland Dreierc5bcbbb2006-02-02 09:47:14 -0800530 char node_desc[64];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531};
532
533enum ib_port_modify_flags {
534 IB_PORT_SHUTDOWN = 1,
535 IB_PORT_INIT_TYPE = (1<<2),
536 IB_PORT_RESET_QKEY_CNTR = (1<<3)
537};
538
539struct ib_port_modify {
540 u32 set_port_cap_mask;
541 u32 clr_port_cap_mask;
542 u8 init_type;
543};
544
545enum ib_event_type {
546 IB_EVENT_CQ_ERR,
547 IB_EVENT_QP_FATAL,
548 IB_EVENT_QP_REQ_ERR,
549 IB_EVENT_QP_ACCESS_ERR,
550 IB_EVENT_COMM_EST,
551 IB_EVENT_SQ_DRAINED,
552 IB_EVENT_PATH_MIG,
553 IB_EVENT_PATH_MIG_ERR,
554 IB_EVENT_DEVICE_FATAL,
555 IB_EVENT_PORT_ACTIVE,
556 IB_EVENT_PORT_ERR,
557 IB_EVENT_LID_CHANGE,
558 IB_EVENT_PKEY_CHANGE,
Roland Dreierd41fcc62005-08-18 12:23:08 -0700559 IB_EVENT_SM_CHANGE,
560 IB_EVENT_SRQ_ERR,
561 IB_EVENT_SRQ_LIMIT_REACHED,
Leonid Arsh63942c92006-06-17 20:37:35 -0700562 IB_EVENT_QP_LAST_WQE_REACHED,
Or Gerlitz761d90e2011-06-15 14:39:29 +0000563 IB_EVENT_CLIENT_REREGISTER,
564 IB_EVENT_GID_CHANGE,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565};
566
Bart Van Asschedb7489e2015-08-03 10:01:52 -0700567const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
Sagi Grimberg2b1b5b62015-05-18 13:40:28 +0300568
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569struct ib_event {
570 struct ib_device *device;
571 union {
572 struct ib_cq *cq;
573 struct ib_qp *qp;
Roland Dreierd41fcc62005-08-18 12:23:08 -0700574 struct ib_srq *srq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 u8 port_num;
576 } element;
577 enum ib_event_type event;
578};
579
580struct ib_event_handler {
581 struct ib_device *device;
582 void (*handler)(struct ib_event_handler *, struct ib_event *);
583 struct list_head list;
584};
585
586#define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
587 do { \
588 (_ptr)->device = _device; \
589 (_ptr)->handler = _handler; \
590 INIT_LIST_HEAD(&(_ptr)->list); \
591 } while (0)
592
593struct ib_global_route {
594 union ib_gid dgid;
595 u32 flow_label;
596 u8 sgid_index;
597 u8 hop_limit;
598 u8 traffic_class;
599};
600
Hal Rosenstock513789e2005-07-27 11:45:34 -0700601struct ib_grh {
Sean Hefty97f52eb2005-08-13 21:05:57 -0700602 __be32 version_tclass_flow;
603 __be16 paylen;
Hal Rosenstock513789e2005-07-27 11:45:34 -0700604 u8 next_hdr;
605 u8 hop_limit;
606 union ib_gid sgid;
607 union ib_gid dgid;
608};
609
Somnath Koturc865f242015-12-23 14:56:51 +0200610union rdma_network_hdr {
611 struct ib_grh ibgrh;
612 struct {
613 /* The IB spec states that if it's IPv4, the header
614 * is located in the last 20 bytes of the header.
615 */
616 u8 reserved[20];
617 struct iphdr roce4grh;
618 };
619};
620
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621enum {
622 IB_MULTICAST_QPN = 0xffffff
623};
624
Harvey Harrisonf3a7c662009-02-14 22:58:35 -0800625#define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
Dennis Dalessandrob4e64392016-01-06 10:04:31 -0800626#define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
Sean Hefty97f52eb2005-08-13 21:05:57 -0700627
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628enum ib_ah_flags {
629 IB_AH_GRH = 1
630};
631
Jack Morgensteinbf6a9e32006-04-10 09:43:47 -0700632enum ib_rate {
633 IB_RATE_PORT_CURRENT = 0,
634 IB_RATE_2_5_GBPS = 2,
635 IB_RATE_5_GBPS = 5,
636 IB_RATE_10_GBPS = 3,
637 IB_RATE_20_GBPS = 6,
638 IB_RATE_30_GBPS = 4,
639 IB_RATE_40_GBPS = 7,
640 IB_RATE_60_GBPS = 8,
641 IB_RATE_80_GBPS = 9,
Marcel Apfelbaum71eeba12011-10-05 14:21:47 +0300642 IB_RATE_120_GBPS = 10,
643 IB_RATE_14_GBPS = 11,
644 IB_RATE_56_GBPS = 12,
645 IB_RATE_112_GBPS = 13,
646 IB_RATE_168_GBPS = 14,
647 IB_RATE_25_GBPS = 15,
648 IB_RATE_100_GBPS = 16,
649 IB_RATE_200_GBPS = 17,
650 IB_RATE_300_GBPS = 18
Jack Morgensteinbf6a9e32006-04-10 09:43:47 -0700651};
652
653/**
654 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
655 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
656 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
657 * @rate: rate to convert.
658 */
Roland Dreier8385fd82014-06-04 10:00:16 -0700659__attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
Jack Morgensteinbf6a9e32006-04-10 09:43:47 -0700660
661/**
Marcel Apfelbaum71eeba12011-10-05 14:21:47 +0300662 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
663 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
664 * @rate: rate to convert.
665 */
Roland Dreier8385fd82014-06-04 10:00:16 -0700666__attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
Marcel Apfelbaum71eeba12011-10-05 14:21:47 +0300667
Sagi Grimberg17cd3a22014-02-23 14:19:04 +0200668
669/**
Sagi Grimberg9bee1782015-07-30 10:32:35 +0300670 * enum ib_mr_type - memory region type
671 * @IB_MR_TYPE_MEM_REG: memory region that is used for
672 * normal registration
673 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
674 * signature operations (data-integrity
675 * capable regions)
Sagi Grimbergf5aa9152016-02-29 19:07:32 +0200676 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
677 * register any arbitrary sg lists (without
678 * the normal mr constraints - see
679 * ib_map_mr_sg)
Sagi Grimberg17cd3a22014-02-23 14:19:04 +0200680 */
Sagi Grimberg9bee1782015-07-30 10:32:35 +0300681enum ib_mr_type {
682 IB_MR_TYPE_MEM_REG,
683 IB_MR_TYPE_SIGNATURE,
Sagi Grimbergf5aa9152016-02-29 19:07:32 +0200684 IB_MR_TYPE_SG_GAPS,
Sagi Grimberg17cd3a22014-02-23 14:19:04 +0200685};
686
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200687/**
Sagi Grimberg78eda2b2014-08-13 19:54:35 +0300688 * Signature types
689 * IB_SIG_TYPE_NONE: Unprotected.
690 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200691 */
Sagi Grimberg78eda2b2014-08-13 19:54:35 +0300692enum ib_signature_type {
693 IB_SIG_TYPE_NONE,
694 IB_SIG_TYPE_T10_DIF,
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200695};
696
697/**
698 * Signature T10-DIF block-guard types
699 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
700 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
701 */
702enum ib_t10_dif_bg_type {
703 IB_T10DIF_CRC,
704 IB_T10DIF_CSUM
705};
706
707/**
708 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
709 * domain.
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200710 * @bg_type: T10-DIF block guard type (CRC|CSUM)
711 * @pi_interval: protection information interval.
712 * @bg: seed of guard computation.
713 * @app_tag: application tag of guard block
714 * @ref_tag: initial guard block reference tag.
Sagi Grimberg78eda2b2014-08-13 19:54:35 +0300715 * @ref_remap: Indicate wethear the reftag increments each block
716 * @app_escape: Indicate to skip block check if apptag=0xffff
717 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
718 * @apptag_check_mask: check bitmask of application tag.
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200719 */
720struct ib_t10_dif_domain {
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200721 enum ib_t10_dif_bg_type bg_type;
722 u16 pi_interval;
723 u16 bg;
724 u16 app_tag;
725 u32 ref_tag;
Sagi Grimberg78eda2b2014-08-13 19:54:35 +0300726 bool ref_remap;
727 bool app_escape;
728 bool ref_escape;
729 u16 apptag_check_mask;
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200730};
731
732/**
733 * struct ib_sig_domain - Parameters for signature domain
734 * @sig_type: specific signauture type
735 * @sig: union of all signature domain attributes that may
736 * be used to set domain layout.
737 */
738struct ib_sig_domain {
739 enum ib_signature_type sig_type;
740 union {
741 struct ib_t10_dif_domain dif;
742 } sig;
743};
744
745/**
746 * struct ib_sig_attrs - Parameters for signature handover operation
747 * @check_mask: bitmask for signature byte check (8 bytes)
748 * @mem: memory domain layout desciptor.
749 * @wire: wire domain layout desciptor.
750 */
751struct ib_sig_attrs {
752 u8 check_mask;
753 struct ib_sig_domain mem;
754 struct ib_sig_domain wire;
755};
756
757enum ib_sig_err_type {
758 IB_SIG_BAD_GUARD,
759 IB_SIG_BAD_REFTAG,
760 IB_SIG_BAD_APPTAG,
761};
762
763/**
764 * struct ib_sig_err - signature error descriptor
765 */
766struct ib_sig_err {
767 enum ib_sig_err_type err_type;
768 u32 expected;
769 u32 actual;
770 u64 sig_err_offset;
771 u32 key;
772};
773
774enum ib_mr_status_check {
775 IB_MR_CHECK_SIG_STATUS = 1,
776};
777
778/**
779 * struct ib_mr_status - Memory region status container
780 *
781 * @fail_status: Bitmask of MR checks status. For each
782 * failed check a corresponding status bit is set.
783 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
784 * failure.
785 */
786struct ib_mr_status {
787 u32 fail_status;
788 struct ib_sig_err sig_err;
789};
790
Marcel Apfelbaum71eeba12011-10-05 14:21:47 +0300791/**
Jack Morgensteinbf6a9e32006-04-10 09:43:47 -0700792 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
793 * enum.
794 * @mult: multiple to convert.
795 */
Roland Dreier8385fd82014-06-04 10:00:16 -0700796__attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
Jack Morgensteinbf6a9e32006-04-10 09:43:47 -0700797
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798struct ib_ah_attr {
799 struct ib_global_route grh;
800 u16 dlid;
801 u8 sl;
802 u8 src_path_bits;
803 u8 static_rate;
804 u8 ah_flags;
805 u8 port_num;
Matan Barakdd5f03b2013-12-12 18:03:11 +0200806 u8 dmac[ETH_ALEN];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807};
808
809enum ib_wc_status {
810 IB_WC_SUCCESS,
811 IB_WC_LOC_LEN_ERR,
812 IB_WC_LOC_QP_OP_ERR,
813 IB_WC_LOC_EEC_OP_ERR,
814 IB_WC_LOC_PROT_ERR,
815 IB_WC_WR_FLUSH_ERR,
816 IB_WC_MW_BIND_ERR,
817 IB_WC_BAD_RESP_ERR,
818 IB_WC_LOC_ACCESS_ERR,
819 IB_WC_REM_INV_REQ_ERR,
820 IB_WC_REM_ACCESS_ERR,
821 IB_WC_REM_OP_ERR,
822 IB_WC_RETRY_EXC_ERR,
823 IB_WC_RNR_RETRY_EXC_ERR,
824 IB_WC_LOC_RDD_VIOL_ERR,
825 IB_WC_REM_INV_RD_REQ_ERR,
826 IB_WC_REM_ABORT_ERR,
827 IB_WC_INV_EECN_ERR,
828 IB_WC_INV_EEC_STATE_ERR,
829 IB_WC_FATAL_ERR,
830 IB_WC_RESP_TIMEOUT_ERR,
831 IB_WC_GENERAL_ERR
832};
833
Bart Van Asschedb7489e2015-08-03 10:01:52 -0700834const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
Sagi Grimberg2b1b5b62015-05-18 13:40:28 +0300835
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836enum ib_wc_opcode {
837 IB_WC_SEND,
838 IB_WC_RDMA_WRITE,
839 IB_WC_RDMA_READ,
840 IB_WC_COMP_SWAP,
841 IB_WC_FETCH_ADD,
Eli Cohenc93570f2008-04-16 21:09:27 -0700842 IB_WC_LSO,
Steve Wise00f7ec32008-07-14 23:48:45 -0700843 IB_WC_LOCAL_INV,
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +0300844 IB_WC_REG_MR,
Vladimir Sokolovsky5e80ba82010-04-14 17:23:01 +0300845 IB_WC_MASKED_COMP_SWAP,
846 IB_WC_MASKED_FETCH_ADD,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847/*
848 * Set value of IB_WC_RECV so consumers can test if a completion is a
849 * receive by testing (opcode & IB_WC_RECV).
850 */
851 IB_WC_RECV = 1 << 7,
852 IB_WC_RECV_RDMA_WITH_IMM
853};
854
855enum ib_wc_flags {
856 IB_WC_GRH = 1,
Steve Wise00f7ec32008-07-14 23:48:45 -0700857 IB_WC_WITH_IMM = (1<<1),
858 IB_WC_WITH_INVALIDATE = (1<<2),
Or Gerlitzd927d502012-01-11 19:03:51 +0200859 IB_WC_IP_CSUM_OK = (1<<3),
Matan Barakdd5f03b2013-12-12 18:03:11 +0200860 IB_WC_WITH_SMAC = (1<<4),
861 IB_WC_WITH_VLAN = (1<<5),
Somnath Koturc865f242015-12-23 14:56:51 +0200862 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863};
864
865struct ib_wc {
Christoph Hellwig14d3a3b2015-12-11 11:53:03 -0800866 union {
867 u64 wr_id;
868 struct ib_cqe *wr_cqe;
869 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 enum ib_wc_status status;
871 enum ib_wc_opcode opcode;
872 u32 vendor_err;
873 u32 byte_len;
Michael S. Tsirkin062dbb62006-12-31 21:09:42 +0200874 struct ib_qp *qp;
Steve Wise00f7ec32008-07-14 23:48:45 -0700875 union {
876 __be32 imm_data;
877 u32 invalidate_rkey;
878 } ex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 u32 src_qp;
880 int wc_flags;
881 u16 pkey_index;
882 u16 slid;
883 u8 sl;
884 u8 dlid_path_bits;
885 u8 port_num; /* valid only for DR SMPs on switches */
Matan Barakdd5f03b2013-12-12 18:03:11 +0200886 u8 smac[ETH_ALEN];
887 u16 vlan_id;
Somnath Koturc865f242015-12-23 14:56:51 +0200888 u8 network_hdr_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889};
890
Roland Dreiered23a722007-05-06 21:02:48 -0700891enum ib_cq_notify_flags {
892 IB_CQ_SOLICITED = 1 << 0,
893 IB_CQ_NEXT_COMP = 1 << 1,
894 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
895 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896};
897
Sean Hefty96104ed2011-05-23 16:31:36 -0700898enum ib_srq_type {
Sean Hefty418d5132011-05-23 19:42:29 -0700899 IB_SRQT_BASIC,
900 IB_SRQT_XRC
Sean Hefty96104ed2011-05-23 16:31:36 -0700901};
902
Roland Dreierd41fcc62005-08-18 12:23:08 -0700903enum ib_srq_attr_mask {
904 IB_SRQ_MAX_WR = 1 << 0,
905 IB_SRQ_LIMIT = 1 << 1,
906};
907
908struct ib_srq_attr {
909 u32 max_wr;
910 u32 max_sge;
911 u32 srq_limit;
912};
913
914struct ib_srq_init_attr {
915 void (*event_handler)(struct ib_event *, void *);
916 void *srq_context;
917 struct ib_srq_attr attr;
Sean Hefty96104ed2011-05-23 16:31:36 -0700918 enum ib_srq_type srq_type;
Sean Hefty418d5132011-05-23 19:42:29 -0700919
920 union {
921 struct {
922 struct ib_xrcd *xrcd;
923 struct ib_cq *cq;
924 } xrc;
925 } ext;
Roland Dreierd41fcc62005-08-18 12:23:08 -0700926};
927
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928struct ib_qp_cap {
929 u32 max_send_wr;
930 u32 max_recv_wr;
931 u32 max_send_sge;
932 u32 max_recv_sge;
933 u32 max_inline_data;
Christoph Hellwiga060b562016-05-03 18:01:09 +0200934
935 /*
936 * Maximum number of rdma_rw_ctx structures in flight at a time.
937 * ib_create_qp() will calculate the right amount of neededed WRs
938 * and MRs based on this.
939 */
940 u32 max_rdma_ctxs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941};
942
943enum ib_sig_type {
944 IB_SIGNAL_ALL_WR,
945 IB_SIGNAL_REQ_WR
946};
947
948enum ib_qp_type {
949 /*
950 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
951 * here (and in that order) since the MAD layer uses them as
952 * indices into a 2-entry table.
953 */
954 IB_QPT_SMI,
955 IB_QPT_GSI,
956
957 IB_QPT_RC,
958 IB_QPT_UC,
959 IB_QPT_UD,
960 IB_QPT_RAW_IPV6,
Sean Heftyb42b63c2011-05-23 19:59:25 -0700961 IB_QPT_RAW_ETHERTYPE,
Or Gerlitzc938a612012-03-01 12:17:51 +0200962 IB_QPT_RAW_PACKET = 8,
Sean Heftyb42b63c2011-05-23 19:59:25 -0700963 IB_QPT_XRC_INI = 9,
964 IB_QPT_XRC_TGT,
Jack Morgenstein0134f162013-07-07 17:25:52 +0300965 IB_QPT_MAX,
966 /* Reserve a range for qp types internal to the low level driver.
967 * These qp types will not be visible at the IB core layer, so the
968 * IB_QPT_MAX usages should not be affected in the core layer
969 */
970 IB_QPT_RESERVED1 = 0x1000,
971 IB_QPT_RESERVED2,
972 IB_QPT_RESERVED3,
973 IB_QPT_RESERVED4,
974 IB_QPT_RESERVED5,
975 IB_QPT_RESERVED6,
976 IB_QPT_RESERVED7,
977 IB_QPT_RESERVED8,
978 IB_QPT_RESERVED9,
979 IB_QPT_RESERVED10,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980};
981
Eli Cohenb846f252008-04-16 21:09:27 -0700982enum ib_qp_create_flags {
Ron Livne47ee1b92008-07-14 23:48:48 -0700983 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
984 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
Leon Romanovsky8a06ce52015-12-20 12:16:10 +0200985 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
986 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
987 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
Matan Barak90f1d1b2013-11-07 15:25:12 +0200988 IB_QP_CREATE_NETIF_QP = 1 << 5,
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200989 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
Or Gerlitz09b93082014-05-11 15:15:11 +0300990 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7,
Majd Dibbinyb531b902016-04-17 17:19:36 +0300991 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
Jack Morgensteind2b57062012-08-03 08:40:37 +0000992 /* reserve bits 26-31 for low level drivers' internal use */
993 IB_QP_CREATE_RESERVED_START = 1 << 26,
994 IB_QP_CREATE_RESERVED_END = 1 << 31,
Eli Cohenb846f252008-04-16 21:09:27 -0700995};
996
Yishai Hadas73c40c62013-08-01 18:49:53 +0300997/*
998 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
999 * callback to destroy the passed in QP.
1000 */
1001
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002struct ib_qp_init_attr {
1003 void (*event_handler)(struct ib_event *, void *);
1004 void *qp_context;
1005 struct ib_cq *send_cq;
1006 struct ib_cq *recv_cq;
1007 struct ib_srq *srq;
Sean Heftyb42b63c2011-05-23 19:59:25 -07001008 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 struct ib_qp_cap cap;
1010 enum ib_sig_type sq_sig_type;
1011 enum ib_qp_type qp_type;
Eli Cohenb846f252008-04-16 21:09:27 -07001012 enum ib_qp_create_flags create_flags;
Christoph Hellwiga060b562016-05-03 18:01:09 +02001013
1014 /*
1015 * Only needed for special QP types, or when using the RW API.
1016 */
1017 u8 port_num;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018};
1019
Sean Hefty0e0ec7e2011-08-08 15:31:51 -07001020struct ib_qp_open_attr {
1021 void (*event_handler)(struct ib_event *, void *);
1022 void *qp_context;
1023 u32 qp_num;
1024 enum ib_qp_type qp_type;
1025};
1026
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027enum ib_rnr_timeout {
1028 IB_RNR_TIMER_655_36 = 0,
1029 IB_RNR_TIMER_000_01 = 1,
1030 IB_RNR_TIMER_000_02 = 2,
1031 IB_RNR_TIMER_000_03 = 3,
1032 IB_RNR_TIMER_000_04 = 4,
1033 IB_RNR_TIMER_000_06 = 5,
1034 IB_RNR_TIMER_000_08 = 6,
1035 IB_RNR_TIMER_000_12 = 7,
1036 IB_RNR_TIMER_000_16 = 8,
1037 IB_RNR_TIMER_000_24 = 9,
1038 IB_RNR_TIMER_000_32 = 10,
1039 IB_RNR_TIMER_000_48 = 11,
1040 IB_RNR_TIMER_000_64 = 12,
1041 IB_RNR_TIMER_000_96 = 13,
1042 IB_RNR_TIMER_001_28 = 14,
1043 IB_RNR_TIMER_001_92 = 15,
1044 IB_RNR_TIMER_002_56 = 16,
1045 IB_RNR_TIMER_003_84 = 17,
1046 IB_RNR_TIMER_005_12 = 18,
1047 IB_RNR_TIMER_007_68 = 19,
1048 IB_RNR_TIMER_010_24 = 20,
1049 IB_RNR_TIMER_015_36 = 21,
1050 IB_RNR_TIMER_020_48 = 22,
1051 IB_RNR_TIMER_030_72 = 23,
1052 IB_RNR_TIMER_040_96 = 24,
1053 IB_RNR_TIMER_061_44 = 25,
1054 IB_RNR_TIMER_081_92 = 26,
1055 IB_RNR_TIMER_122_88 = 27,
1056 IB_RNR_TIMER_163_84 = 28,
1057 IB_RNR_TIMER_245_76 = 29,
1058 IB_RNR_TIMER_327_68 = 30,
1059 IB_RNR_TIMER_491_52 = 31
1060};
1061
1062enum ib_qp_attr_mask {
1063 IB_QP_STATE = 1,
1064 IB_QP_CUR_STATE = (1<<1),
1065 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1066 IB_QP_ACCESS_FLAGS = (1<<3),
1067 IB_QP_PKEY_INDEX = (1<<4),
1068 IB_QP_PORT = (1<<5),
1069 IB_QP_QKEY = (1<<6),
1070 IB_QP_AV = (1<<7),
1071 IB_QP_PATH_MTU = (1<<8),
1072 IB_QP_TIMEOUT = (1<<9),
1073 IB_QP_RETRY_CNT = (1<<10),
1074 IB_QP_RNR_RETRY = (1<<11),
1075 IB_QP_RQ_PSN = (1<<12),
1076 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1077 IB_QP_ALT_PATH = (1<<14),
1078 IB_QP_MIN_RNR_TIMER = (1<<15),
1079 IB_QP_SQ_PSN = (1<<16),
1080 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1081 IB_QP_PATH_MIG_STATE = (1<<18),
1082 IB_QP_CAP = (1<<19),
Matan Barakdd5f03b2013-12-12 18:03:11 +02001083 IB_QP_DEST_QPN = (1<<20),
Matan Barakaa744cc2015-10-15 18:38:53 +03001084 IB_QP_RESERVED1 = (1<<21),
1085 IB_QP_RESERVED2 = (1<<22),
1086 IB_QP_RESERVED3 = (1<<23),
1087 IB_QP_RESERVED4 = (1<<24),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088};
1089
1090enum ib_qp_state {
1091 IB_QPS_RESET,
1092 IB_QPS_INIT,
1093 IB_QPS_RTR,
1094 IB_QPS_RTS,
1095 IB_QPS_SQD,
1096 IB_QPS_SQE,
1097 IB_QPS_ERR
1098};
1099
1100enum ib_mig_state {
1101 IB_MIG_MIGRATED,
1102 IB_MIG_REARM,
1103 IB_MIG_ARMED
1104};
1105
Shani Michaeli7083e422013-02-06 16:19:12 +00001106enum ib_mw_type {
1107 IB_MW_TYPE_1 = 1,
1108 IB_MW_TYPE_2 = 2
1109};
1110
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111struct ib_qp_attr {
1112 enum ib_qp_state qp_state;
1113 enum ib_qp_state cur_qp_state;
1114 enum ib_mtu path_mtu;
1115 enum ib_mig_state path_mig_state;
1116 u32 qkey;
1117 u32 rq_psn;
1118 u32 sq_psn;
1119 u32 dest_qp_num;
1120 int qp_access_flags;
1121 struct ib_qp_cap cap;
1122 struct ib_ah_attr ah_attr;
1123 struct ib_ah_attr alt_ah_attr;
1124 u16 pkey_index;
1125 u16 alt_pkey_index;
1126 u8 en_sqd_async_notify;
1127 u8 sq_draining;
1128 u8 max_rd_atomic;
1129 u8 max_dest_rd_atomic;
1130 u8 min_rnr_timer;
1131 u8 port_num;
1132 u8 timeout;
1133 u8 retry_cnt;
1134 u8 rnr_retry;
1135 u8 alt_port_num;
1136 u8 alt_timeout;
1137};
1138
1139enum ib_wr_opcode {
1140 IB_WR_RDMA_WRITE,
1141 IB_WR_RDMA_WRITE_WITH_IMM,
1142 IB_WR_SEND,
1143 IB_WR_SEND_WITH_IMM,
1144 IB_WR_RDMA_READ,
1145 IB_WR_ATOMIC_CMP_AND_SWP,
Eli Cohenc93570f2008-04-16 21:09:27 -07001146 IB_WR_ATOMIC_FETCH_AND_ADD,
Roland Dreier0f39cf32008-04-16 21:09:32 -07001147 IB_WR_LSO,
1148 IB_WR_SEND_WITH_INV,
Steve Wise00f7ec32008-07-14 23:48:45 -07001149 IB_WR_RDMA_READ_WITH_INV,
1150 IB_WR_LOCAL_INV,
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +03001151 IB_WR_REG_MR,
Vladimir Sokolovsky5e80ba82010-04-14 17:23:01 +03001152 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1153 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
Sagi Grimberg1b01d332014-02-23 14:19:05 +02001154 IB_WR_REG_SIG_MR,
Jack Morgenstein0134f162013-07-07 17:25:52 +03001155 /* reserve values for low level drivers' internal use.
1156 * These values will not be used at all in the ib core layer.
1157 */
1158 IB_WR_RESERVED1 = 0xf0,
1159 IB_WR_RESERVED2,
1160 IB_WR_RESERVED3,
1161 IB_WR_RESERVED4,
1162 IB_WR_RESERVED5,
1163 IB_WR_RESERVED6,
1164 IB_WR_RESERVED7,
1165 IB_WR_RESERVED8,
1166 IB_WR_RESERVED9,
1167 IB_WR_RESERVED10,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168};
1169
1170enum ib_send_flags {
1171 IB_SEND_FENCE = 1,
1172 IB_SEND_SIGNALED = (1<<1),
1173 IB_SEND_SOLICITED = (1<<2),
Eli Cohene0605d92008-01-30 18:30:57 +02001174 IB_SEND_INLINE = (1<<3),
Jack Morgenstein0134f162013-07-07 17:25:52 +03001175 IB_SEND_IP_CSUM = (1<<4),
1176
1177 /* reserve bits 26-31 for low level drivers' internal use */
1178 IB_SEND_RESERVED_START = (1 << 26),
1179 IB_SEND_RESERVED_END = (1 << 31),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180};
1181
1182struct ib_sge {
1183 u64 addr;
1184 u32 length;
1185 u32 lkey;
1186};
1187
Christoph Hellwig14d3a3b2015-12-11 11:53:03 -08001188struct ib_cqe {
1189 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1190};
1191
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192struct ib_send_wr {
1193 struct ib_send_wr *next;
Christoph Hellwig14d3a3b2015-12-11 11:53:03 -08001194 union {
1195 u64 wr_id;
1196 struct ib_cqe *wr_cqe;
1197 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 struct ib_sge *sg_list;
1199 int num_sge;
1200 enum ib_wr_opcode opcode;
1201 int send_flags;
Roland Dreier0f39cf32008-04-16 21:09:32 -07001202 union {
1203 __be32 imm_data;
1204 u32 invalidate_rkey;
1205 } ex;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206};
1207
Christoph Hellwige622f2f2015-10-08 09:16:33 +01001208struct ib_rdma_wr {
1209 struct ib_send_wr wr;
1210 u64 remote_addr;
1211 u32 rkey;
1212};
1213
1214static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1215{
1216 return container_of(wr, struct ib_rdma_wr, wr);
1217}
1218
1219struct ib_atomic_wr {
1220 struct ib_send_wr wr;
1221 u64 remote_addr;
1222 u64 compare_add;
1223 u64 swap;
1224 u64 compare_add_mask;
1225 u64 swap_mask;
1226 u32 rkey;
1227};
1228
1229static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1230{
1231 return container_of(wr, struct ib_atomic_wr, wr);
1232}
1233
1234struct ib_ud_wr {
1235 struct ib_send_wr wr;
1236 struct ib_ah *ah;
1237 void *header;
1238 int hlen;
1239 int mss;
1240 u32 remote_qpn;
1241 u32 remote_qkey;
1242 u16 pkey_index; /* valid for GSI only */
1243 u8 port_num; /* valid for DR SMPs on switch only */
1244};
1245
1246static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1247{
1248 return container_of(wr, struct ib_ud_wr, wr);
1249}
1250
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +03001251struct ib_reg_wr {
1252 struct ib_send_wr wr;
1253 struct ib_mr *mr;
1254 u32 key;
1255 int access;
1256};
1257
1258static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1259{
1260 return container_of(wr, struct ib_reg_wr, wr);
1261}
1262
Christoph Hellwige622f2f2015-10-08 09:16:33 +01001263struct ib_sig_handover_wr {
1264 struct ib_send_wr wr;
1265 struct ib_sig_attrs *sig_attrs;
1266 struct ib_mr *sig_mr;
1267 int access_flags;
1268 struct ib_sge *prot;
1269};
1270
1271static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1272{
1273 return container_of(wr, struct ib_sig_handover_wr, wr);
1274}
1275
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276struct ib_recv_wr {
1277 struct ib_recv_wr *next;
Christoph Hellwig14d3a3b2015-12-11 11:53:03 -08001278 union {
1279 u64 wr_id;
1280 struct ib_cqe *wr_cqe;
1281 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 struct ib_sge *sg_list;
1283 int num_sge;
1284};
1285
1286enum ib_access_flags {
1287 IB_ACCESS_LOCAL_WRITE = 1,
1288 IB_ACCESS_REMOTE_WRITE = (1<<1),
1289 IB_ACCESS_REMOTE_READ = (1<<2),
1290 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
Shani Michaeli7083e422013-02-06 16:19:12 +00001291 IB_ACCESS_MW_BIND = (1<<4),
Sagi Grimberg860f10a2014-12-11 17:04:16 +02001292 IB_ZERO_BASED = (1<<5),
1293 IB_ACCESS_ON_DEMAND = (1<<6),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294};
1295
Christoph Hellwigb7d3e0a2015-12-23 19:12:47 +01001296/*
1297 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1298 * are hidden here instead of a uapi header!
1299 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300enum ib_mr_rereg_flags {
1301 IB_MR_REREG_TRANS = 1,
1302 IB_MR_REREG_PD = (1<<1),
Matan Barak7e6edb92014-07-31 11:01:28 +03001303 IB_MR_REREG_ACCESS = (1<<2),
1304 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305};
1306
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307struct ib_fmr_attr {
1308 int max_pages;
1309 int max_maps;
Or Gerlitzd36f34a2006-02-02 10:43:45 -08001310 u8 page_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311};
1312
Haggai Eran882214e2014-12-11 17:04:18 +02001313struct ib_umem;
1314
Roland Dreiere2773c02005-07-07 17:57:10 -07001315struct ib_ucontext {
1316 struct ib_device *device;
1317 struct list_head pd_list;
1318 struct list_head mr_list;
1319 struct list_head mw_list;
1320 struct list_head cq_list;
1321 struct list_head qp_list;
1322 struct list_head srq_list;
1323 struct list_head ah_list;
Sean Hefty53d0bd12011-05-24 08:33:46 -07001324 struct list_head xrcd_list;
Hadar Hen Zion436f2ad2013-08-14 13:58:30 +03001325 struct list_head rule_list;
Roland Dreierf7c6a7b2007-03-04 16:15:11 -08001326 int closing;
Shachar Raindel8ada2c12014-12-11 17:04:17 +02001327
1328 struct pid *tgid;
Haggai Eran882214e2014-12-11 17:04:18 +02001329#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1330 struct rb_root umem_tree;
1331 /*
1332 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1333 * mmu notifiers registration.
1334 */
1335 struct rw_semaphore umem_rwsem;
1336 void (*invalidate_range)(struct ib_umem *umem,
1337 unsigned long start, unsigned long end);
1338
1339 struct mmu_notifier mn;
1340 atomic_t notifier_count;
1341 /* A list of umems that don't have private mmu notifier counters yet. */
1342 struct list_head no_private_counters;
1343 int odp_mrs_count;
1344#endif
Roland Dreiere2773c02005-07-07 17:57:10 -07001345};
1346
1347struct ib_uobject {
1348 u64 user_handle; /* handle given to us by userspace */
1349 struct ib_ucontext *context; /* associated user context */
Roland Dreier9ead1902006-06-17 20:44:49 -07001350 void *object; /* containing object */
Roland Dreiere2773c02005-07-07 17:57:10 -07001351 struct list_head list; /* link to context's list */
Roland Dreierb3d636b2008-04-16 21:01:06 -07001352 int id; /* index into kernel idr */
Roland Dreier9ead1902006-06-17 20:44:49 -07001353 struct kref ref;
1354 struct rw_semaphore mutex; /* protects .live */
Mike Marciniszynd144da82015-11-02 12:13:25 -05001355 struct rcu_head rcu; /* kfree_rcu() overhead */
Roland Dreier9ead1902006-06-17 20:44:49 -07001356 int live;
Roland Dreiere2773c02005-07-07 17:57:10 -07001357};
1358
Roland Dreiere2773c02005-07-07 17:57:10 -07001359struct ib_udata {
Yann Droneaud309243e2013-12-11 23:01:44 +01001360 const void __user *inbuf;
Roland Dreiere2773c02005-07-07 17:57:10 -07001361 void __user *outbuf;
1362 size_t inlen;
1363 size_t outlen;
1364};
1365
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366struct ib_pd {
Jason Gunthorpe96249d72015-08-05 14:14:45 -06001367 u32 local_dma_lkey;
Roland Dreiere2773c02005-07-07 17:57:10 -07001368 struct ib_device *device;
1369 struct ib_uobject *uobject;
1370 atomic_t usecnt; /* count all resources */
Jason Gunthorpe96249d72015-08-05 14:14:45 -06001371 struct ib_mr *local_mr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372};
1373
Sean Hefty59991f92011-05-23 17:52:46 -07001374struct ib_xrcd {
1375 struct ib_device *device;
Sean Heftyd3d72d92011-05-26 23:06:44 -07001376 atomic_t usecnt; /* count all exposed resources */
Sean Hefty53d0bd12011-05-24 08:33:46 -07001377 struct inode *inode;
Sean Heftyd3d72d92011-05-26 23:06:44 -07001378
1379 struct mutex tgt_qp_mutex;
1380 struct list_head tgt_qp_list;
Sean Hefty59991f92011-05-23 17:52:46 -07001381};
1382
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383struct ib_ah {
1384 struct ib_device *device;
1385 struct ib_pd *pd;
Roland Dreiere2773c02005-07-07 17:57:10 -07001386 struct ib_uobject *uobject;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387};
1388
1389typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1390
Christoph Hellwig14d3a3b2015-12-11 11:53:03 -08001391enum ib_poll_context {
1392 IB_POLL_DIRECT, /* caller context, no hw completions */
1393 IB_POLL_SOFTIRQ, /* poll from softirq context */
1394 IB_POLL_WORKQUEUE, /* poll from workqueue */
1395};
1396
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397struct ib_cq {
Roland Dreiere2773c02005-07-07 17:57:10 -07001398 struct ib_device *device;
1399 struct ib_uobject *uobject;
1400 ib_comp_handler comp_handler;
1401 void (*event_handler)(struct ib_event *, void *);
Dotan Barak4deccd62008-07-14 23:48:44 -07001402 void *cq_context;
Roland Dreiere2773c02005-07-07 17:57:10 -07001403 int cqe;
1404 atomic_t usecnt; /* count number of work queues */
Christoph Hellwig14d3a3b2015-12-11 11:53:03 -08001405 enum ib_poll_context poll_ctx;
1406 struct ib_wc *wc;
1407 union {
1408 struct irq_poll iop;
1409 struct work_struct work;
1410 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411};
1412
1413struct ib_srq {
Roland Dreierd41fcc62005-08-18 12:23:08 -07001414 struct ib_device *device;
1415 struct ib_pd *pd;
1416 struct ib_uobject *uobject;
1417 void (*event_handler)(struct ib_event *, void *);
1418 void *srq_context;
Sean Hefty96104ed2011-05-23 16:31:36 -07001419 enum ib_srq_type srq_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 atomic_t usecnt;
Sean Hefty418d5132011-05-23 19:42:29 -07001421
1422 union {
1423 struct {
1424 struct ib_xrcd *xrcd;
1425 struct ib_cq *cq;
1426 u32 srq_num;
1427 } xrc;
1428 } ext;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429};
1430
1431struct ib_qp {
1432 struct ib_device *device;
1433 struct ib_pd *pd;
1434 struct ib_cq *send_cq;
1435 struct ib_cq *recv_cq;
Christoph Hellwigfffb0382016-05-03 18:01:07 +02001436 spinlock_t mr_lock;
1437 int mrs_used;
Christoph Hellwiga060b562016-05-03 18:01:09 +02001438 struct list_head rdma_mrs;
Christoph Hellwig0e353e32016-05-03 18:01:12 +02001439 struct list_head sig_mrs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 struct ib_srq *srq;
Sean Heftyb42b63c2011-05-23 19:59:25 -07001441 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
Sean Heftyd3d72d92011-05-26 23:06:44 -07001442 struct list_head xrcd_list;
Christoph Hellwigfffb0382016-05-03 18:01:07 +02001443
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001444 /* count times opened, mcast attaches, flow attaches */
1445 atomic_t usecnt;
Sean Hefty0e0ec7e2011-08-08 15:31:51 -07001446 struct list_head open_list;
1447 struct ib_qp *real_qp;
Roland Dreiere2773c02005-07-07 17:57:10 -07001448 struct ib_uobject *uobject;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 void (*event_handler)(struct ib_event *, void *);
1450 void *qp_context;
1451 u32 qp_num;
1452 enum ib_qp_type qp_type;
1453};
1454
1455struct ib_mr {
Roland Dreiere2773c02005-07-07 17:57:10 -07001456 struct ib_device *device;
1457 struct ib_pd *pd;
Roland Dreiere2773c02005-07-07 17:57:10 -07001458 u32 lkey;
1459 u32 rkey;
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +03001460 u64 iova;
1461 u32 length;
1462 unsigned int page_size;
Steve Wised4a85c32016-05-03 18:01:08 +02001463 bool need_inval;
Christoph Hellwigfffb0382016-05-03 18:01:07 +02001464 union {
1465 struct ib_uobject *uobject; /* user */
1466 struct list_head qp_entry; /* FR */
1467 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468};
1469
1470struct ib_mw {
1471 struct ib_device *device;
1472 struct ib_pd *pd;
Roland Dreiere2773c02005-07-07 17:57:10 -07001473 struct ib_uobject *uobject;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 u32 rkey;
Shani Michaeli7083e422013-02-06 16:19:12 +00001475 enum ib_mw_type type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476};
1477
1478struct ib_fmr {
1479 struct ib_device *device;
1480 struct ib_pd *pd;
1481 struct list_head list;
1482 u32 lkey;
1483 u32 rkey;
1484};
1485
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001486/* Supported steering options */
1487enum ib_flow_attr_type {
1488 /* steering according to rule specifications */
1489 IB_FLOW_ATTR_NORMAL = 0x0,
1490 /* default unicast and multicast rule -
1491 * receive all Eth traffic which isn't steered to any QP
1492 */
1493 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1494 /* default multicast rule -
1495 * receive all Eth multicast traffic which isn't steered to any QP
1496 */
1497 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1498 /* sniffer rule - receive all port traffic */
1499 IB_FLOW_ATTR_SNIFFER = 0x3
1500};
1501
1502/* Supported steering header types */
1503enum ib_flow_spec_type {
1504 /* L2 headers*/
1505 IB_FLOW_SPEC_ETH = 0x20,
Matan Barak240ae002013-11-07 15:25:13 +02001506 IB_FLOW_SPEC_IB = 0x22,
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001507 /* L3 header*/
1508 IB_FLOW_SPEC_IPV4 = 0x30,
1509 /* L4 headers*/
1510 IB_FLOW_SPEC_TCP = 0x40,
1511 IB_FLOW_SPEC_UDP = 0x41
1512};
Matan Barak240ae002013-11-07 15:25:13 +02001513#define IB_FLOW_SPEC_LAYER_MASK 0xF0
Matan Barak22878db2013-09-01 18:39:52 +03001514#define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1515
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001516/* Flow steering rule priority is set according to it's domain.
1517 * Lower domain value means higher priority.
1518 */
1519enum ib_flow_domain {
1520 IB_FLOW_DOMAIN_USER,
1521 IB_FLOW_DOMAIN_ETHTOOL,
1522 IB_FLOW_DOMAIN_RFS,
1523 IB_FLOW_DOMAIN_NIC,
1524 IB_FLOW_DOMAIN_NUM /* Must be last */
1525};
1526
Marina Varshavera3100a72016-02-18 18:31:05 +02001527enum ib_flow_flags {
1528 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1529 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 2 /* Must be last */
1530};
1531
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001532struct ib_flow_eth_filter {
1533 u8 dst_mac[6];
1534 u8 src_mac[6];
1535 __be16 ether_type;
1536 __be16 vlan_tag;
1537};
1538
1539struct ib_flow_spec_eth {
1540 enum ib_flow_spec_type type;
1541 u16 size;
1542 struct ib_flow_eth_filter val;
1543 struct ib_flow_eth_filter mask;
1544};
1545
Matan Barak240ae002013-11-07 15:25:13 +02001546struct ib_flow_ib_filter {
1547 __be16 dlid;
1548 __u8 sl;
1549};
1550
1551struct ib_flow_spec_ib {
1552 enum ib_flow_spec_type type;
1553 u16 size;
1554 struct ib_flow_ib_filter val;
1555 struct ib_flow_ib_filter mask;
1556};
1557
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001558struct ib_flow_ipv4_filter {
1559 __be32 src_ip;
1560 __be32 dst_ip;
1561};
1562
1563struct ib_flow_spec_ipv4 {
1564 enum ib_flow_spec_type type;
1565 u16 size;
1566 struct ib_flow_ipv4_filter val;
1567 struct ib_flow_ipv4_filter mask;
1568};
1569
1570struct ib_flow_tcp_udp_filter {
1571 __be16 dst_port;
1572 __be16 src_port;
1573};
1574
1575struct ib_flow_spec_tcp_udp {
1576 enum ib_flow_spec_type type;
1577 u16 size;
1578 struct ib_flow_tcp_udp_filter val;
1579 struct ib_flow_tcp_udp_filter mask;
1580};
1581
1582union ib_flow_spec {
1583 struct {
1584 enum ib_flow_spec_type type;
1585 u16 size;
1586 };
1587 struct ib_flow_spec_eth eth;
Matan Barak240ae002013-11-07 15:25:13 +02001588 struct ib_flow_spec_ib ib;
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001589 struct ib_flow_spec_ipv4 ipv4;
1590 struct ib_flow_spec_tcp_udp tcp_udp;
1591};
1592
1593struct ib_flow_attr {
1594 enum ib_flow_attr_type type;
1595 u16 size;
1596 u16 priority;
1597 u32 flags;
1598 u8 num_of_specs;
1599 u8 port;
1600 /* Following are the optional layers according to user request
1601 * struct ib_flow_spec_xxx
1602 * struct ib_flow_spec_yyy
1603 */
1604};
1605
1606struct ib_flow {
1607 struct ib_qp *qp;
1608 struct ib_uobject *uobject;
1609};
1610
Ira Weiny4cd7c942015-06-06 14:38:31 -04001611struct ib_mad_hdr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612struct ib_grh;
1613
1614enum ib_process_mad_flags {
1615 IB_MAD_IGNORE_MKEY = 1,
1616 IB_MAD_IGNORE_BKEY = 2,
1617 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1618};
1619
1620enum ib_mad_result {
1621 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1622 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1623 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1624 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1625};
1626
1627#define IB_DEVICE_NAME_MAX 64
1628
1629struct ib_cache {
1630 rwlock_t lock;
1631 struct ib_event_handler event_handler;
1632 struct ib_pkey_cache **pkey_cache;
Matan Barak03db3a22015-07-30 18:33:26 +03001633 struct ib_gid_table **gid_cache;
Jack Morgenstein6fb9cdb2006-06-17 20:37:34 -07001634 u8 *lmc_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635};
1636
Ralph Campbell9b513092006-12-12 14:27:41 -08001637struct ib_dma_mapping_ops {
1638 int (*mapping_error)(struct ib_device *dev,
1639 u64 dma_addr);
1640 u64 (*map_single)(struct ib_device *dev,
1641 void *ptr, size_t size,
1642 enum dma_data_direction direction);
1643 void (*unmap_single)(struct ib_device *dev,
1644 u64 addr, size_t size,
1645 enum dma_data_direction direction);
1646 u64 (*map_page)(struct ib_device *dev,
1647 struct page *page, unsigned long offset,
1648 size_t size,
1649 enum dma_data_direction direction);
1650 void (*unmap_page)(struct ib_device *dev,
1651 u64 addr, size_t size,
1652 enum dma_data_direction direction);
1653 int (*map_sg)(struct ib_device *dev,
1654 struct scatterlist *sg, int nents,
1655 enum dma_data_direction direction);
1656 void (*unmap_sg)(struct ib_device *dev,
1657 struct scatterlist *sg, int nents,
1658 enum dma_data_direction direction);
Ralph Campbell9b513092006-12-12 14:27:41 -08001659 void (*sync_single_for_cpu)(struct ib_device *dev,
1660 u64 dma_handle,
1661 size_t size,
Dotan Barak4deccd62008-07-14 23:48:44 -07001662 enum dma_data_direction dir);
Ralph Campbell9b513092006-12-12 14:27:41 -08001663 void (*sync_single_for_device)(struct ib_device *dev,
1664 u64 dma_handle,
1665 size_t size,
1666 enum dma_data_direction dir);
1667 void *(*alloc_coherent)(struct ib_device *dev,
1668 size_t size,
1669 u64 *dma_handle,
1670 gfp_t flag);
1671 void (*free_coherent)(struct ib_device *dev,
1672 size_t size, void *cpu_addr,
1673 u64 dma_handle);
1674};
1675
Tom Tucker07ebafb2006-08-03 16:02:42 -05001676struct iw_cm_verbs;
1677
Ira Weiny77386132015-05-13 20:02:58 -04001678struct ib_port_immutable {
1679 int pkey_tbl_len;
1680 int gid_tbl_len;
Ira Weinyf9b22e32015-05-13 20:02:59 -04001681 u32 core_cap_flags;
Ira Weiny337877a2015-06-06 14:38:29 -04001682 u32 max_mad_size;
Ira Weiny77386132015-05-13 20:02:58 -04001683};
1684
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685struct ib_device {
1686 struct device *dma_device;
1687
1688 char name[IB_DEVICE_NAME_MAX];
1689
1690 struct list_head event_handler_list;
1691 spinlock_t event_handler_lock;
1692
Alexander Chiang17a55f72010-02-02 19:09:16 +00001693 spinlock_t client_data_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 struct list_head core_list;
Haggai Eran7c1eb452015-07-30 17:50:14 +03001695 /* Access to the client_data_list is protected by the client_data_lock
1696 * spinlock and the lists_rwsem read-write semaphore */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 struct list_head client_data_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698
1699 struct ib_cache cache;
Ira Weiny77386132015-05-13 20:02:58 -04001700 /**
1701 * port_immutable is indexed by port number
1702 */
1703 struct ib_port_immutable *port_immutable;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
Michael S. Tsirkinf4fd0b22007-05-03 13:48:47 +03001705 int num_comp_vectors;
1706
Tom Tucker07ebafb2006-08-03 16:02:42 -05001707 struct iw_cm_verbs *iwcm;
1708
Christoph Lameterb40f4752016-05-16 12:49:33 -05001709 /**
1710 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1711 * driver initialized data. The struct is kfree()'ed by the sysfs
1712 * core when the device is removed. A lifespan of -1 in the return
1713 * struct tells the core to set a default lifespan.
1714 */
1715 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
1716 u8 port_num);
1717 /**
1718 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1719 * @index - The index in the value array we wish to have updated, or
1720 * num_counters if we want all stats updated
1721 * Return codes -
1722 * < 0 - Error, no counters updated
1723 * index - Updated the single counter pointed to by index
1724 * num_counters - Updated all counters (will reset the timestamp
1725 * and prevent further calls for lifespan milliseconds)
1726 * Drivers are allowed to update all counters in leiu of just the
1727 * one given in index at their option
1728 */
1729 int (*get_hw_stats)(struct ib_device *device,
1730 struct rdma_hw_stats *stats,
1731 u8 port, int index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 int (*query_device)(struct ib_device *device,
Matan Barak2528e332015-06-11 16:35:25 +03001733 struct ib_device_attr *device_attr,
1734 struct ib_udata *udata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 int (*query_port)(struct ib_device *device,
1736 u8 port_num,
1737 struct ib_port_attr *port_attr);
Eli Cohena3f5ada2010-09-27 17:51:10 -07001738 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1739 u8 port_num);
Matan Barak03db3a22015-07-30 18:33:26 +03001740 /* When calling get_netdev, the HW vendor's driver should return the
1741 * net device of device @device at port @port_num or NULL if such
1742 * a net device doesn't exist. The vendor driver should call dev_hold
1743 * on this net device. The HW vendor's device driver must guarantee
1744 * that this function returns NULL before the net device reaches
1745 * NETDEV_UNREGISTER_FINAL state.
1746 */
1747 struct net_device *(*get_netdev)(struct ib_device *device,
1748 u8 port_num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 int (*query_gid)(struct ib_device *device,
1750 u8 port_num, int index,
1751 union ib_gid *gid);
Matan Barak03db3a22015-07-30 18:33:26 +03001752 /* When calling add_gid, the HW vendor's driver should
1753 * add the gid of device @device at gid index @index of
1754 * port @port_num to be @gid. Meta-info of that gid (for example,
1755 * the network device related to this gid is available
1756 * at @attr. @context allows the HW vendor driver to store extra
1757 * information together with a GID entry. The HW vendor may allocate
1758 * memory to contain this information and store it in @context when a
1759 * new GID entry is written to. Params are consistent until the next
1760 * call of add_gid or delete_gid. The function should return 0 on
1761 * success or error otherwise. The function could be called
1762 * concurrently for different ports. This function is only called
1763 * when roce_gid_table is used.
1764 */
1765 int (*add_gid)(struct ib_device *device,
1766 u8 port_num,
1767 unsigned int index,
1768 const union ib_gid *gid,
1769 const struct ib_gid_attr *attr,
1770 void **context);
1771 /* When calling del_gid, the HW vendor's driver should delete the
1772 * gid of device @device at gid index @index of port @port_num.
1773 * Upon the deletion of a GID entry, the HW vendor must free any
1774 * allocated memory. The caller will clear @context afterwards.
1775 * This function is only called when roce_gid_table is used.
1776 */
1777 int (*del_gid)(struct ib_device *device,
1778 u8 port_num,
1779 unsigned int index,
1780 void **context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 int (*query_pkey)(struct ib_device *device,
1782 u8 port_num, u16 index, u16 *pkey);
1783 int (*modify_device)(struct ib_device *device,
1784 int device_modify_mask,
1785 struct ib_device_modify *device_modify);
1786 int (*modify_port)(struct ib_device *device,
1787 u8 port_num, int port_modify_mask,
1788 struct ib_port_modify *port_modify);
Roland Dreiere2773c02005-07-07 17:57:10 -07001789 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
1790 struct ib_udata *udata);
1791 int (*dealloc_ucontext)(struct ib_ucontext *context);
1792 int (*mmap)(struct ib_ucontext *context,
1793 struct vm_area_struct *vma);
1794 struct ib_pd * (*alloc_pd)(struct ib_device *device,
1795 struct ib_ucontext *context,
1796 struct ib_udata *udata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 int (*dealloc_pd)(struct ib_pd *pd);
1798 struct ib_ah * (*create_ah)(struct ib_pd *pd,
1799 struct ib_ah_attr *ah_attr);
1800 int (*modify_ah)(struct ib_ah *ah,
1801 struct ib_ah_attr *ah_attr);
1802 int (*query_ah)(struct ib_ah *ah,
1803 struct ib_ah_attr *ah_attr);
1804 int (*destroy_ah)(struct ib_ah *ah);
Roland Dreierd41fcc62005-08-18 12:23:08 -07001805 struct ib_srq * (*create_srq)(struct ib_pd *pd,
1806 struct ib_srq_init_attr *srq_init_attr,
1807 struct ib_udata *udata);
1808 int (*modify_srq)(struct ib_srq *srq,
1809 struct ib_srq_attr *srq_attr,
Ralph Campbell9bc57e22006-08-11 14:58:09 -07001810 enum ib_srq_attr_mask srq_attr_mask,
1811 struct ib_udata *udata);
Roland Dreierd41fcc62005-08-18 12:23:08 -07001812 int (*query_srq)(struct ib_srq *srq,
1813 struct ib_srq_attr *srq_attr);
1814 int (*destroy_srq)(struct ib_srq *srq);
1815 int (*post_srq_recv)(struct ib_srq *srq,
1816 struct ib_recv_wr *recv_wr,
1817 struct ib_recv_wr **bad_recv_wr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 struct ib_qp * (*create_qp)(struct ib_pd *pd,
Roland Dreiere2773c02005-07-07 17:57:10 -07001819 struct ib_qp_init_attr *qp_init_attr,
1820 struct ib_udata *udata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 int (*modify_qp)(struct ib_qp *qp,
1822 struct ib_qp_attr *qp_attr,
Ralph Campbell9bc57e22006-08-11 14:58:09 -07001823 int qp_attr_mask,
1824 struct ib_udata *udata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 int (*query_qp)(struct ib_qp *qp,
1826 struct ib_qp_attr *qp_attr,
1827 int qp_attr_mask,
1828 struct ib_qp_init_attr *qp_init_attr);
1829 int (*destroy_qp)(struct ib_qp *qp);
1830 int (*post_send)(struct ib_qp *qp,
1831 struct ib_send_wr *send_wr,
1832 struct ib_send_wr **bad_send_wr);
1833 int (*post_recv)(struct ib_qp *qp,
1834 struct ib_recv_wr *recv_wr,
1835 struct ib_recv_wr **bad_recv_wr);
Matan Barakbcf4c1e2015-06-11 16:35:20 +03001836 struct ib_cq * (*create_cq)(struct ib_device *device,
1837 const struct ib_cq_init_attr *attr,
Roland Dreiere2773c02005-07-07 17:57:10 -07001838 struct ib_ucontext *context,
1839 struct ib_udata *udata);
Eli Cohen2dd57162008-04-16 21:09:33 -07001840 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1841 u16 cq_period);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 int (*destroy_cq)(struct ib_cq *cq);
Roland Dreier33b9b3e2006-01-30 14:29:21 -08001843 int (*resize_cq)(struct ib_cq *cq, int cqe,
1844 struct ib_udata *udata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 int (*poll_cq)(struct ib_cq *cq, int num_entries,
1846 struct ib_wc *wc);
1847 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1848 int (*req_notify_cq)(struct ib_cq *cq,
Roland Dreiered23a722007-05-06 21:02:48 -07001849 enum ib_cq_notify_flags flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 int (*req_ncomp_notif)(struct ib_cq *cq,
1851 int wc_cnt);
1852 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
1853 int mr_access_flags);
Roland Dreiere2773c02005-07-07 17:57:10 -07001854 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
Roland Dreierf7c6a7b2007-03-04 16:15:11 -08001855 u64 start, u64 length,
1856 u64 virt_addr,
Roland Dreiere2773c02005-07-07 17:57:10 -07001857 int mr_access_flags,
1858 struct ib_udata *udata);
Matan Barak7e6edb92014-07-31 11:01:28 +03001859 int (*rereg_user_mr)(struct ib_mr *mr,
1860 int flags,
1861 u64 start, u64 length,
1862 u64 virt_addr,
1863 int mr_access_flags,
1864 struct ib_pd *pd,
1865 struct ib_udata *udata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 int (*dereg_mr)(struct ib_mr *mr);
Sagi Grimberg9bee1782015-07-30 10:32:35 +03001867 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
1868 enum ib_mr_type mr_type,
1869 u32 max_num_sg);
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +03001870 int (*map_mr_sg)(struct ib_mr *mr,
1871 struct scatterlist *sg,
Christoph Hellwigff2ba992016-05-03 18:01:04 +02001872 int sg_nents,
Bart Van Assche9aa8b322016-05-12 10:49:15 -07001873 unsigned int *sg_offset);
Shani Michaeli7083e422013-02-06 16:19:12 +00001874 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
Matan Barakb2a239d2016-02-29 18:05:29 +02001875 enum ib_mw_type type,
1876 struct ib_udata *udata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 int (*dealloc_mw)(struct ib_mw *mw);
1878 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
1879 int mr_access_flags,
1880 struct ib_fmr_attr *fmr_attr);
1881 int (*map_phys_fmr)(struct ib_fmr *fmr,
1882 u64 *page_list, int list_len,
1883 u64 iova);
1884 int (*unmap_fmr)(struct list_head *fmr_list);
1885 int (*dealloc_fmr)(struct ib_fmr *fmr);
1886 int (*attach_mcast)(struct ib_qp *qp,
1887 union ib_gid *gid,
1888 u16 lid);
1889 int (*detach_mcast)(struct ib_qp *qp,
1890 union ib_gid *gid,
1891 u16 lid);
1892 int (*process_mad)(struct ib_device *device,
1893 int process_mad_flags,
1894 u8 port_num,
Ira Weinya97e2d82015-05-31 17:15:30 -04001895 const struct ib_wc *in_wc,
1896 const struct ib_grh *in_grh,
Ira Weiny4cd7c942015-06-06 14:38:31 -04001897 const struct ib_mad_hdr *in_mad,
1898 size_t in_mad_size,
1899 struct ib_mad_hdr *out_mad,
1900 size_t *out_mad_size,
1901 u16 *out_mad_pkey_index);
Sean Hefty59991f92011-05-23 17:52:46 -07001902 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
1903 struct ib_ucontext *ucontext,
1904 struct ib_udata *udata);
1905 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001906 struct ib_flow * (*create_flow)(struct ib_qp *qp,
1907 struct ib_flow_attr
1908 *flow_attr,
1909 int domain);
1910 int (*destroy_flow)(struct ib_flow *flow_id);
Sagi Grimberg1b01d332014-02-23 14:19:05 +02001911 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
1912 struct ib_mr_status *mr_status);
Yishai Hadas036b1062015-08-13 18:32:05 +03001913 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
Steve Wise765d6772016-02-17 08:15:41 -08001914 void (*drain_rq)(struct ib_qp *qp);
1915 void (*drain_sq)(struct ib_qp *qp);
Eli Cohen50174a72016-03-11 22:58:38 +02001916 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
1917 int state);
1918 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
1919 struct ifla_vf_info *ivf);
1920 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
1921 struct ifla_vf_stats *stats);
1922 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
1923 int type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924
Ralph Campbell9b513092006-12-12 14:27:41 -08001925 struct ib_dma_mapping_ops *dma_ops;
1926
Roland Dreiere2773c02005-07-07 17:57:10 -07001927 struct module *owner;
Tony Jonesf4e91eb2008-02-22 00:13:36 +01001928 struct device dev;
Greg Kroah-Hartman35be0682007-12-17 15:54:39 -04001929 struct kobject *ports_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 struct list_head port_list;
1931
1932 enum {
1933 IB_DEV_UNINITIALIZED,
1934 IB_DEV_REGISTERED,
1935 IB_DEV_UNREGISTERED
1936 } reg_state;
1937
Roland Dreier274c0892005-09-29 14:17:48 -07001938 int uverbs_abi_ver;
Alexander Chiang17a55f72010-02-02 19:09:16 +00001939 u64 uverbs_cmd_mask;
Yann Droneaudf21519b2013-11-06 23:21:49 +01001940 u64 uverbs_ex_cmd_mask;
Roland Dreier274c0892005-09-29 14:17:48 -07001941
Roland Dreierc5bcbbb2006-02-02 09:47:14 -08001942 char node_desc[64];
Sean Heftycf311cd2006-01-10 07:39:34 -08001943 __be64 node_guid;
Steve Wise96f15c02008-07-14 23:48:53 -07001944 u32 local_dma_lkey;
Hal Rosenstock41390322015-06-29 09:57:00 -04001945 u16 is_switch:1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 u8 node_type;
1947 u8 phys_port_cnt;
Ira Weiny3e153a92015-12-18 10:59:44 +02001948 struct ib_device_attr attrs;
Christoph Lameterb40f4752016-05-16 12:49:33 -05001949 struct attribute_group *hw_stats_ag;
1950 struct rdma_hw_stats *hw_stats;
Ira Weiny77386132015-05-13 20:02:58 -04001951
1952 /**
1953 * The following mandatory functions are used only at device
1954 * registration. Keep functions such as these at the end of this
1955 * structure to avoid cache line misses when accessing struct ib_device
1956 * in fast paths.
1957 */
1958 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959};
1960
1961struct ib_client {
1962 char *name;
1963 void (*add) (struct ib_device *);
Haggai Eran7c1eb452015-07-30 17:50:14 +03001964 void (*remove)(struct ib_device *, void *client_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965
Yotam Kenneth9268f722015-07-30 17:50:15 +03001966 /* Returns the net_dev belonging to this ib_client and matching the
1967 * given parameters.
1968 * @dev: An RDMA device that the net_dev use for communication.
1969 * @port: A physical port number on the RDMA device.
1970 * @pkey: P_Key that the net_dev uses if applicable.
1971 * @gid: A GID that the net_dev uses to communicate.
1972 * @addr: An IP address the net_dev is configured with.
1973 * @client_data: The device's client data set by ib_set_client_data().
1974 *
1975 * An ib_client that implements a net_dev on top of RDMA devices
1976 * (such as IP over IB) should implement this callback, allowing the
1977 * rdma_cm module to find the right net_dev for a given request.
1978 *
1979 * The caller is responsible for calling dev_put on the returned
1980 * netdev. */
1981 struct net_device *(*get_net_dev_by_params)(
1982 struct ib_device *dev,
1983 u8 port,
1984 u16 pkey,
1985 const union ib_gid *gid,
1986 const struct sockaddr *addr,
1987 void *client_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 struct list_head list;
1989};
1990
1991struct ib_device *ib_alloc_device(size_t size);
1992void ib_dealloc_device(struct ib_device *device);
1993
Ralph Campbell9a6edb62010-05-06 17:03:25 -07001994int ib_register_device(struct ib_device *device,
1995 int (*port_callback)(struct ib_device *,
1996 u8, struct kobject *));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997void ib_unregister_device(struct ib_device *device);
1998
1999int ib_register_client (struct ib_client *client);
2000void ib_unregister_client(struct ib_client *client);
2001
2002void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2003void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2004 void *data);
2005
Roland Dreiere2773c02005-07-07 17:57:10 -07002006static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2007{
2008 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2009}
2010
2011static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2012{
Yann Droneaud43c611652015-02-05 22:10:18 +01002013 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
Roland Dreiere2773c02005-07-07 17:57:10 -07002014}
2015
Matan Barak301a7212015-12-15 20:30:10 +02002016static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2017 size_t offset,
2018 size_t len)
2019{
2020 const void __user *p = udata->inbuf + offset;
2021 bool ret = false;
2022 u8 *buf;
2023
2024 if (len > USHRT_MAX)
2025 return false;
2026
2027 buf = kmalloc(len, GFP_KERNEL);
2028 if (!buf)
2029 return false;
2030
2031 if (copy_from_user(buf, p, len))
2032 goto free;
2033
2034 ret = !memchr_inv(buf, 0, len);
2035
2036free:
2037 kfree(buf);
2038 return ret;
2039}
2040
Roland Dreier8a518662006-02-13 12:48:12 -08002041/**
2042 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2043 * contains all required attributes and no attributes not allowed for
2044 * the given QP state transition.
2045 * @cur_state: Current QP state
2046 * @next_state: Next QP state
2047 * @type: QP type
2048 * @mask: Mask of supplied QP attributes
Matan Barakdd5f03b2013-12-12 18:03:11 +02002049 * @ll : link layer of port
Roland Dreier8a518662006-02-13 12:48:12 -08002050 *
2051 * This function is a helper function that a low-level driver's
2052 * modify_qp method can use to validate the consumer's input. It
2053 * checks that cur_state and next_state are valid QP states, that a
2054 * transition from cur_state to next_state is allowed by the IB spec,
2055 * and that the attribute mask supplied is allowed for the transition.
2056 */
2057int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
Matan Barakdd5f03b2013-12-12 18:03:11 +02002058 enum ib_qp_type type, enum ib_qp_attr_mask mask,
2059 enum rdma_link_layer ll);
Roland Dreier8a518662006-02-13 12:48:12 -08002060
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061int ib_register_event_handler (struct ib_event_handler *event_handler);
2062int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2063void ib_dispatch_event(struct ib_event *event);
2064
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065int ib_query_port(struct ib_device *device,
2066 u8 port_num, struct ib_port_attr *port_attr);
2067
Eli Cohena3f5ada2010-09-27 17:51:10 -07002068enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2069 u8 port_num);
2070
Ira Weiny0cf18d72015-05-13 20:02:55 -04002071/**
Hal Rosenstock41390322015-06-29 09:57:00 -04002072 * rdma_cap_ib_switch - Check if the device is IB switch
2073 * @device: Device to check
2074 *
2075 * Device driver is responsible for setting is_switch bit on
2076 * in ib_device structure at init time.
2077 *
2078 * Return: true if the device is IB switch.
2079 */
2080static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2081{
2082 return device->is_switch;
2083}
2084
2085/**
Ira Weiny0cf18d72015-05-13 20:02:55 -04002086 * rdma_start_port - Return the first valid port number for the device
2087 * specified
2088 *
2089 * @device: Device to be checked
2090 *
2091 * Return start port number
2092 */
2093static inline u8 rdma_start_port(const struct ib_device *device)
2094{
Hal Rosenstock41390322015-06-29 09:57:00 -04002095 return rdma_cap_ib_switch(device) ? 0 : 1;
Ira Weiny0cf18d72015-05-13 20:02:55 -04002096}
2097
2098/**
2099 * rdma_end_port - Return the last valid port number for the device
2100 * specified
2101 *
2102 * @device: Device to be checked
2103 *
2104 * Return last port number
2105 */
2106static inline u8 rdma_end_port(const struct ib_device *device)
2107{
Hal Rosenstock41390322015-06-29 09:57:00 -04002108 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
Ira Weiny0cf18d72015-05-13 20:02:55 -04002109}
2110
Ira Weiny5ede9282015-05-31 17:15:29 -04002111static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
Michael Wangde66be92015-05-05 14:50:19 +02002112{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002113 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
Michael Wangde66be92015-05-05 14:50:19 +02002114}
2115
Ira Weiny5ede9282015-05-31 17:15:29 -04002116static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
Michael Wangde66be92015-05-05 14:50:19 +02002117{
Matan Barak7766a992015-12-23 14:56:50 +02002118 return device->port_immutable[port_num].core_cap_flags &
2119 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2120}
2121
2122static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2123{
2124 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2125}
2126
2127static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2128{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002129 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
Michael Wangde66be92015-05-05 14:50:19 +02002130}
2131
Ira Weiny5ede9282015-05-31 17:15:29 -04002132static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
Michael Wangde66be92015-05-05 14:50:19 +02002133{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002134 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
Michael Wangde66be92015-05-05 14:50:19 +02002135}
2136
Ira Weiny5ede9282015-05-31 17:15:29 -04002137static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
Michael Wangde66be92015-05-05 14:50:19 +02002138{
Matan Barak7766a992015-12-23 14:56:50 +02002139 return rdma_protocol_ib(device, port_num) ||
2140 rdma_protocol_roce(device, port_num);
Michael Wangde66be92015-05-05 14:50:19 +02002141}
2142
Michael Wangc757dea2015-05-05 14:50:32 +02002143/**
Michael Wang296ec002015-05-18 10:41:45 +02002144 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
Michael Wangc757dea2015-05-05 14:50:32 +02002145 * Management Datagrams.
Michael Wang296ec002015-05-18 10:41:45 +02002146 * @device: Device to check
2147 * @port_num: Port number to check
Michael Wangc757dea2015-05-05 14:50:32 +02002148 *
Michael Wang296ec002015-05-18 10:41:45 +02002149 * Management Datagrams (MAD) are a required part of the InfiniBand
2150 * specification and are supported on all InfiniBand devices. A slightly
2151 * extended version are also supported on OPA interfaces.
Michael Wangc757dea2015-05-05 14:50:32 +02002152 *
Michael Wang296ec002015-05-18 10:41:45 +02002153 * Return: true if the port supports sending/receiving of MAD packets.
Michael Wangc757dea2015-05-05 14:50:32 +02002154 */
Ira Weiny5ede9282015-05-31 17:15:29 -04002155static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
Michael Wangc757dea2015-05-05 14:50:32 +02002156{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002157 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
Michael Wangc757dea2015-05-05 14:50:32 +02002158}
2159
Michael Wang29541e32015-05-05 14:50:33 +02002160/**
Ira Weiny65995fe2015-06-06 14:38:32 -04002161 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2162 * Management Datagrams.
2163 * @device: Device to check
2164 * @port_num: Port number to check
2165 *
2166 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2167 * datagrams with their own versions. These OPA MADs share many but not all of
2168 * the characteristics of InfiniBand MADs.
2169 *
2170 * OPA MADs differ in the following ways:
2171 *
2172 * 1) MADs are variable size up to 2K
2173 * IBTA defined MADs remain fixed at 256 bytes
2174 * 2) OPA SMPs must carry valid PKeys
2175 * 3) OPA SMP packets are a different format
2176 *
2177 * Return: true if the port supports OPA MAD packet formats.
2178 */
2179static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2180{
2181 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2182 == RDMA_CORE_CAP_OPA_MAD;
2183}
2184
2185/**
Michael Wang296ec002015-05-18 10:41:45 +02002186 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2187 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2188 * @device: Device to check
2189 * @port_num: Port number to check
Michael Wang29541e32015-05-05 14:50:33 +02002190 *
Michael Wang296ec002015-05-18 10:41:45 +02002191 * Each InfiniBand node is required to provide a Subnet Management Agent
2192 * that the subnet manager can access. Prior to the fabric being fully
2193 * configured by the subnet manager, the SMA is accessed via a well known
2194 * interface called the Subnet Management Interface (SMI). This interface
2195 * uses directed route packets to communicate with the SM to get around the
2196 * chicken and egg problem of the SM needing to know what's on the fabric
2197 * in order to configure the fabric, and needing to configure the fabric in
2198 * order to send packets to the devices on the fabric. These directed
2199 * route packets do not need the fabric fully configured in order to reach
2200 * their destination. The SMI is the only method allowed to send
2201 * directed route packets on an InfiniBand fabric.
Michael Wang29541e32015-05-05 14:50:33 +02002202 *
Michael Wang296ec002015-05-18 10:41:45 +02002203 * Return: true if the port provides an SMI.
Michael Wang29541e32015-05-05 14:50:33 +02002204 */
Ira Weiny5ede9282015-05-31 17:15:29 -04002205static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
Michael Wang29541e32015-05-05 14:50:33 +02002206{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002207 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
Michael Wang29541e32015-05-05 14:50:33 +02002208}
2209
Michael Wang72219cea2015-05-05 14:50:34 +02002210/**
2211 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2212 * Communication Manager.
Michael Wang296ec002015-05-18 10:41:45 +02002213 * @device: Device to check
2214 * @port_num: Port number to check
Michael Wang72219cea2015-05-05 14:50:34 +02002215 *
Michael Wang296ec002015-05-18 10:41:45 +02002216 * The InfiniBand Communication Manager is one of many pre-defined General
2217 * Service Agents (GSA) that are accessed via the General Service
2218 * Interface (GSI). It's role is to facilitate establishment of connections
2219 * between nodes as well as other management related tasks for established
2220 * connections.
Michael Wang72219cea2015-05-05 14:50:34 +02002221 *
Michael Wang296ec002015-05-18 10:41:45 +02002222 * Return: true if the port supports an IB CM (this does not guarantee that
2223 * a CM is actually running however).
Michael Wang72219cea2015-05-05 14:50:34 +02002224 */
Ira Weiny5ede9282015-05-31 17:15:29 -04002225static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
Michael Wang72219cea2015-05-05 14:50:34 +02002226{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002227 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
Michael Wang72219cea2015-05-05 14:50:34 +02002228}
2229
Michael Wang04215332015-05-05 14:50:35 +02002230/**
2231 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2232 * Communication Manager.
Michael Wang296ec002015-05-18 10:41:45 +02002233 * @device: Device to check
2234 * @port_num: Port number to check
Michael Wang04215332015-05-05 14:50:35 +02002235 *
Michael Wang296ec002015-05-18 10:41:45 +02002236 * Similar to above, but specific to iWARP connections which have a different
2237 * managment protocol than InfiniBand.
Michael Wang04215332015-05-05 14:50:35 +02002238 *
Michael Wang296ec002015-05-18 10:41:45 +02002239 * Return: true if the port supports an iWARP CM (this does not guarantee that
2240 * a CM is actually running however).
Michael Wang04215332015-05-05 14:50:35 +02002241 */
Ira Weiny5ede9282015-05-31 17:15:29 -04002242static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
Michael Wang04215332015-05-05 14:50:35 +02002243{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002244 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
Michael Wang04215332015-05-05 14:50:35 +02002245}
2246
Michael Wangfe53ba22015-05-05 14:50:36 +02002247/**
2248 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2249 * Subnet Administration.
Michael Wang296ec002015-05-18 10:41:45 +02002250 * @device: Device to check
2251 * @port_num: Port number to check
Michael Wangfe53ba22015-05-05 14:50:36 +02002252 *
Michael Wang296ec002015-05-18 10:41:45 +02002253 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2254 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2255 * fabrics, devices should resolve routes to other hosts by contacting the
2256 * SA to query the proper route.
Michael Wangfe53ba22015-05-05 14:50:36 +02002257 *
Michael Wang296ec002015-05-18 10:41:45 +02002258 * Return: true if the port should act as a client to the fabric Subnet
2259 * Administration interface. This does not imply that the SA service is
2260 * running locally.
Michael Wangfe53ba22015-05-05 14:50:36 +02002261 */
Ira Weiny5ede9282015-05-31 17:15:29 -04002262static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
Michael Wangfe53ba22015-05-05 14:50:36 +02002263{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002264 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
Michael Wangfe53ba22015-05-05 14:50:36 +02002265}
2266
Michael Wanga31ad3b2015-05-05 14:50:37 +02002267/**
2268 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2269 * Multicast.
Michael Wang296ec002015-05-18 10:41:45 +02002270 * @device: Device to check
2271 * @port_num: Port number to check
Michael Wanga31ad3b2015-05-05 14:50:37 +02002272 *
Michael Wang296ec002015-05-18 10:41:45 +02002273 * InfiniBand multicast registration is more complex than normal IPv4 or
2274 * IPv6 multicast registration. Each Host Channel Adapter must register
2275 * with the Subnet Manager when it wishes to join a multicast group. It
2276 * should do so only once regardless of how many queue pairs it subscribes
2277 * to this group. And it should leave the group only after all queue pairs
2278 * attached to the group have been detached.
Michael Wanga31ad3b2015-05-05 14:50:37 +02002279 *
Michael Wang296ec002015-05-18 10:41:45 +02002280 * Return: true if the port must undertake the additional adminstrative
2281 * overhead of registering/unregistering with the SM and tracking of the
2282 * total number of queue pairs attached to the multicast group.
Michael Wanga31ad3b2015-05-05 14:50:37 +02002283 */
Ira Weiny5ede9282015-05-31 17:15:29 -04002284static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
Michael Wanga31ad3b2015-05-05 14:50:37 +02002285{
2286 return rdma_cap_ib_sa(device, port_num);
2287}
2288
Michael Wangbc0f1d72015-05-05 14:50:38 +02002289/**
Michael Wang30a74ef2015-05-05 14:50:39 +02002290 * rdma_cap_af_ib - Check if the port of device has the capability
2291 * Native Infiniband Address.
Michael Wang296ec002015-05-18 10:41:45 +02002292 * @device: Device to check
2293 * @port_num: Port number to check
Michael Wang30a74ef2015-05-05 14:50:39 +02002294 *
Michael Wang296ec002015-05-18 10:41:45 +02002295 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2296 * GID. RoCE uses a different mechanism, but still generates a GID via
2297 * a prescribed mechanism and port specific data.
Michael Wang30a74ef2015-05-05 14:50:39 +02002298 *
Michael Wang296ec002015-05-18 10:41:45 +02002299 * Return: true if the port uses a GID address to identify devices on the
2300 * network.
Michael Wang30a74ef2015-05-05 14:50:39 +02002301 */
Ira Weiny5ede9282015-05-31 17:15:29 -04002302static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
Michael Wang30a74ef2015-05-05 14:50:39 +02002303{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002304 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
Michael Wang30a74ef2015-05-05 14:50:39 +02002305}
2306
2307/**
Michael Wang227128f2015-05-05 14:50:40 +02002308 * rdma_cap_eth_ah - Check if the port of device has the capability
Michael Wang296ec002015-05-18 10:41:45 +02002309 * Ethernet Address Handle.
2310 * @device: Device to check
2311 * @port_num: Port number to check
Michael Wang227128f2015-05-05 14:50:40 +02002312 *
Michael Wang296ec002015-05-18 10:41:45 +02002313 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2314 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2315 * port. Normally, packet headers are generated by the sending host
2316 * adapter, but when sending connectionless datagrams, we must manually
2317 * inject the proper headers for the fabric we are communicating over.
Michael Wang227128f2015-05-05 14:50:40 +02002318 *
Michael Wang296ec002015-05-18 10:41:45 +02002319 * Return: true if we are running as a RoCE port and must force the
2320 * addition of a Global Route Header built from our Ethernet Address
2321 * Handle into our header list for connectionless packets.
Michael Wang227128f2015-05-05 14:50:40 +02002322 */
Ira Weiny5ede9282015-05-31 17:15:29 -04002323static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
Michael Wang227128f2015-05-05 14:50:40 +02002324{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002325 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
Michael Wang227128f2015-05-05 14:50:40 +02002326}
2327
2328/**
Ira Weiny337877a2015-06-06 14:38:29 -04002329 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2330 *
2331 * @device: Device
2332 * @port_num: Port number
2333 *
2334 * This MAD size includes the MAD headers and MAD payload. No other headers
2335 * are included.
2336 *
2337 * Return the max MAD size required by the Port. Will return 0 if the port
2338 * does not support MADs
2339 */
2340static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2341{
2342 return device->port_immutable[port_num].max_mad_size;
2343}
2344
Matan Barak03db3a22015-07-30 18:33:26 +03002345/**
2346 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2347 * @device: Device to check
2348 * @port_num: Port number to check
2349 *
2350 * RoCE GID table mechanism manages the various GIDs for a device.
2351 *
2352 * NOTE: if allocating the port's GID table has failed, this call will still
2353 * return true, but any RoCE GID table API will fail.
2354 *
2355 * Return: true if the port uses RoCE GID table mechanism in order to manage
2356 * its GIDs.
2357 */
2358static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2359 u8 port_num)
2360{
2361 return rdma_protocol_roce(device, port_num) &&
2362 device->add_gid && device->del_gid;
2363}
2364
Christoph Hellwig002516e2016-05-03 18:01:05 +02002365/*
2366 * Check if the device supports READ W/ INVALIDATE.
2367 */
2368static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2369{
2370 /*
2371 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
2372 * has support for it yet.
2373 */
2374 return rdma_protocol_iwarp(dev, port_num);
2375}
2376
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377int ib_query_gid(struct ib_device *device,
Matan Barak55ee3ab2015-10-15 18:38:45 +03002378 u8 port_num, int index, union ib_gid *gid,
2379 struct ib_gid_attr *attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380
Eli Cohen50174a72016-03-11 22:58:38 +02002381int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2382 int state);
2383int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2384 struct ifla_vf_info *info);
2385int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2386 struct ifla_vf_stats *stats);
2387int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2388 int type);
2389
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390int ib_query_pkey(struct ib_device *device,
2391 u8 port_num, u16 index, u16 *pkey);
2392
2393int ib_modify_device(struct ib_device *device,
2394 int device_modify_mask,
2395 struct ib_device_modify *device_modify);
2396
2397int ib_modify_port(struct ib_device *device,
2398 u8 port_num, int port_modify_mask,
2399 struct ib_port_modify *port_modify);
2400
Yosef Etigin5eb620c2007-05-14 07:26:51 +03002401int ib_find_gid(struct ib_device *device, union ib_gid *gid,
Matan Barakb39ffa12015-12-23 14:56:47 +02002402 enum ib_gid_type gid_type, struct net_device *ndev,
2403 u8 *port_num, u16 *index);
Yosef Etigin5eb620c2007-05-14 07:26:51 +03002404
2405int ib_find_pkey(struct ib_device *device,
2406 u8 port_num, u16 pkey, u16 *index);
2407
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408struct ib_pd *ib_alloc_pd(struct ib_device *device);
2409
Jason Gunthorpe7dd78642015-08-05 14:34:31 -06002410void ib_dealloc_pd(struct ib_pd *pd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411
2412/**
2413 * ib_create_ah - Creates an address handle for the given address vector.
2414 * @pd: The protection domain associated with the address handle.
2415 * @ah_attr: The attributes of the address vector.
2416 *
2417 * The address handle is used to reference a local or global destination
2418 * in all UD QP post sends.
2419 */
2420struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2421
2422/**
Sean Hefty4e00d692006-06-17 20:37:39 -07002423 * ib_init_ah_from_wc - Initializes address handle attributes from a
2424 * work completion.
2425 * @device: Device on which the received message arrived.
2426 * @port_num: Port on which the received message arrived.
2427 * @wc: Work completion associated with the received message.
2428 * @grh: References the received global route header. This parameter is
2429 * ignored unless the work completion indicates that the GRH is valid.
2430 * @ah_attr: Returned attributes that can be used when creating an address
2431 * handle for replying to the message.
2432 */
Ira Weiny73cdaae2015-05-31 17:15:31 -04002433int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2434 const struct ib_wc *wc, const struct ib_grh *grh,
2435 struct ib_ah_attr *ah_attr);
Sean Hefty4e00d692006-06-17 20:37:39 -07002436
2437/**
Hal Rosenstock513789e2005-07-27 11:45:34 -07002438 * ib_create_ah_from_wc - Creates an address handle associated with the
2439 * sender of the specified work completion.
2440 * @pd: The protection domain associated with the address handle.
2441 * @wc: Work completion information associated with a received message.
2442 * @grh: References the received global route header. This parameter is
2443 * ignored unless the work completion indicates that the GRH is valid.
2444 * @port_num: The outbound port number to associate with the address.
2445 *
2446 * The address handle is used to reference a local or global destination
2447 * in all UD QP post sends.
2448 */
Ira Weiny73cdaae2015-05-31 17:15:31 -04002449struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2450 const struct ib_grh *grh, u8 port_num);
Hal Rosenstock513789e2005-07-27 11:45:34 -07002451
2452/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 * ib_modify_ah - Modifies the address vector associated with an address
2454 * handle.
2455 * @ah: The address handle to modify.
2456 * @ah_attr: The new address vector attributes to associate with the
2457 * address handle.
2458 */
2459int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2460
2461/**
2462 * ib_query_ah - Queries the address vector associated with an address
2463 * handle.
2464 * @ah: The address handle to query.
2465 * @ah_attr: The address vector attributes associated with the address
2466 * handle.
2467 */
2468int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2469
2470/**
2471 * ib_destroy_ah - Destroys an address handle.
2472 * @ah: The address handle to destroy.
2473 */
2474int ib_destroy_ah(struct ib_ah *ah);
2475
2476/**
Roland Dreierd41fcc62005-08-18 12:23:08 -07002477 * ib_create_srq - Creates a SRQ associated with the specified protection
2478 * domain.
2479 * @pd: The protection domain associated with the SRQ.
Dotan Barakabb6e9b2006-02-23 12:13:51 -08002480 * @srq_init_attr: A list of initial attributes required to create the
2481 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2482 * the actual capabilities of the created SRQ.
Roland Dreierd41fcc62005-08-18 12:23:08 -07002483 *
2484 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2485 * requested size of the SRQ, and set to the actual values allocated
2486 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2487 * will always be at least as large as the requested values.
2488 */
2489struct ib_srq *ib_create_srq(struct ib_pd *pd,
2490 struct ib_srq_init_attr *srq_init_attr);
2491
2492/**
2493 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2494 * @srq: The SRQ to modify.
2495 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2496 * the current values of selected SRQ attributes are returned.
2497 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2498 * are being modified.
2499 *
2500 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2501 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2502 * the number of receives queued drops below the limit.
2503 */
2504int ib_modify_srq(struct ib_srq *srq,
2505 struct ib_srq_attr *srq_attr,
2506 enum ib_srq_attr_mask srq_attr_mask);
2507
2508/**
2509 * ib_query_srq - Returns the attribute list and current values for the
2510 * specified SRQ.
2511 * @srq: The SRQ to query.
2512 * @srq_attr: The attributes of the specified SRQ.
2513 */
2514int ib_query_srq(struct ib_srq *srq,
2515 struct ib_srq_attr *srq_attr);
2516
2517/**
2518 * ib_destroy_srq - Destroys the specified SRQ.
2519 * @srq: The SRQ to destroy.
2520 */
2521int ib_destroy_srq(struct ib_srq *srq);
2522
2523/**
2524 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2525 * @srq: The SRQ to post the work request on.
2526 * @recv_wr: A list of work requests to post on the receive queue.
2527 * @bad_recv_wr: On an immediate failure, this parameter will reference
2528 * the work request that failed to be posted on the QP.
2529 */
2530static inline int ib_post_srq_recv(struct ib_srq *srq,
2531 struct ib_recv_wr *recv_wr,
2532 struct ib_recv_wr **bad_recv_wr)
2533{
2534 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2535}
2536
2537/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 * ib_create_qp - Creates a QP associated with the specified protection
2539 * domain.
2540 * @pd: The protection domain associated with the QP.
Dotan Barakabb6e9b2006-02-23 12:13:51 -08002541 * @qp_init_attr: A list of initial attributes required to create the
2542 * QP. If QP creation succeeds, then the attributes are updated to
2543 * the actual capabilities of the created QP.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 */
2545struct ib_qp *ib_create_qp(struct ib_pd *pd,
2546 struct ib_qp_init_attr *qp_init_attr);
2547
2548/**
2549 * ib_modify_qp - Modifies the attributes for the specified QP and then
2550 * transitions the QP to the given state.
2551 * @qp: The QP to modify.
2552 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2553 * the current values of selected QP attributes are returned.
2554 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2555 * are being modified.
2556 */
2557int ib_modify_qp(struct ib_qp *qp,
2558 struct ib_qp_attr *qp_attr,
2559 int qp_attr_mask);
2560
2561/**
2562 * ib_query_qp - Returns the attribute list and current values for the
2563 * specified QP.
2564 * @qp: The QP to query.
2565 * @qp_attr: The attributes of the specified QP.
2566 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2567 * @qp_init_attr: Additional attributes of the selected QP.
2568 *
2569 * The qp_attr_mask may be used to limit the query to gathering only the
2570 * selected attributes.
2571 */
2572int ib_query_qp(struct ib_qp *qp,
2573 struct ib_qp_attr *qp_attr,
2574 int qp_attr_mask,
2575 struct ib_qp_init_attr *qp_init_attr);
2576
2577/**
2578 * ib_destroy_qp - Destroys the specified QP.
2579 * @qp: The QP to destroy.
2580 */
2581int ib_destroy_qp(struct ib_qp *qp);
2582
2583/**
Sean Hefty0e0ec7e2011-08-08 15:31:51 -07002584 * ib_open_qp - Obtain a reference to an existing sharable QP.
2585 * @xrcd - XRC domain
2586 * @qp_open_attr: Attributes identifying the QP to open.
2587 *
2588 * Returns a reference to a sharable QP.
2589 */
2590struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2591 struct ib_qp_open_attr *qp_open_attr);
2592
2593/**
2594 * ib_close_qp - Release an external reference to a QP.
Sean Heftyd3d72d92011-05-26 23:06:44 -07002595 * @qp: The QP handle to release
2596 *
Sean Hefty0e0ec7e2011-08-08 15:31:51 -07002597 * The opened QP handle is released by the caller. The underlying
2598 * shared QP is not destroyed until all internal references are released.
Sean Heftyd3d72d92011-05-26 23:06:44 -07002599 */
Sean Hefty0e0ec7e2011-08-08 15:31:51 -07002600int ib_close_qp(struct ib_qp *qp);
Sean Heftyd3d72d92011-05-26 23:06:44 -07002601
2602/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 * ib_post_send - Posts a list of work requests to the send queue of
2604 * the specified QP.
2605 * @qp: The QP to post the work request on.
2606 * @send_wr: A list of work requests to post on the send queue.
2607 * @bad_send_wr: On an immediate failure, this parameter will reference
2608 * the work request that failed to be posted on the QP.
Bart Van Assche55464d42009-12-09 14:20:04 -08002609 *
2610 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2611 * error is returned, the QP state shall not be affected,
2612 * ib_post_send() will return an immediate error after queueing any
2613 * earlier work requests in the list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614 */
2615static inline int ib_post_send(struct ib_qp *qp,
2616 struct ib_send_wr *send_wr,
2617 struct ib_send_wr **bad_send_wr)
2618{
2619 return qp->device->post_send(qp, send_wr, bad_send_wr);
2620}
2621
2622/**
2623 * ib_post_recv - Posts a list of work requests to the receive queue of
2624 * the specified QP.
2625 * @qp: The QP to post the work request on.
2626 * @recv_wr: A list of work requests to post on the receive queue.
2627 * @bad_recv_wr: On an immediate failure, this parameter will reference
2628 * the work request that failed to be posted on the QP.
2629 */
2630static inline int ib_post_recv(struct ib_qp *qp,
2631 struct ib_recv_wr *recv_wr,
2632 struct ib_recv_wr **bad_recv_wr)
2633{
2634 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2635}
2636
Christoph Hellwig14d3a3b2015-12-11 11:53:03 -08002637struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2638 int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2639void ib_free_cq(struct ib_cq *cq);
2640int ib_process_cq_direct(struct ib_cq *cq, int budget);
2641
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642/**
2643 * ib_create_cq - Creates a CQ on the specified device.
2644 * @device: The device on which to create the CQ.
2645 * @comp_handler: A user-specified callback that is invoked when a
2646 * completion event occurs on the CQ.
2647 * @event_handler: A user-specified callback that is invoked when an
2648 * asynchronous event not associated with a completion occurs on the CQ.
2649 * @cq_context: Context associated with the CQ returned to the user via
2650 * the associated completion and event handlers.
Matan Barak8e372102015-06-11 16:35:21 +03002651 * @cq_attr: The attributes the CQ should be created upon.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652 *
2653 * Users can examine the cq structure to determine the actual CQ size.
2654 */
2655struct ib_cq *ib_create_cq(struct ib_device *device,
2656 ib_comp_handler comp_handler,
2657 void (*event_handler)(struct ib_event *, void *),
Matan Barak8e372102015-06-11 16:35:21 +03002658 void *cq_context,
2659 const struct ib_cq_init_attr *cq_attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660
2661/**
2662 * ib_resize_cq - Modifies the capacity of the CQ.
2663 * @cq: The CQ to resize.
2664 * @cqe: The minimum size of the CQ.
2665 *
2666 * Users can examine the cq structure to determine the actual CQ size.
2667 */
2668int ib_resize_cq(struct ib_cq *cq, int cqe);
2669
2670/**
Eli Cohen2dd57162008-04-16 21:09:33 -07002671 * ib_modify_cq - Modifies moderation params of the CQ
2672 * @cq: The CQ to modify.
2673 * @cq_count: number of CQEs that will trigger an event
2674 * @cq_period: max period of time in usec before triggering an event
2675 *
2676 */
2677int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2678
2679/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 * ib_destroy_cq - Destroys the specified CQ.
2681 * @cq: The CQ to destroy.
2682 */
2683int ib_destroy_cq(struct ib_cq *cq);
2684
2685/**
2686 * ib_poll_cq - poll a CQ for completion(s)
2687 * @cq:the CQ being polled
2688 * @num_entries:maximum number of completions to return
2689 * @wc:array of at least @num_entries &struct ib_wc where completions
2690 * will be returned
2691 *
2692 * Poll a CQ for (possibly multiple) completions. If the return value
2693 * is < 0, an error occurred. If the return value is >= 0, it is the
2694 * number of completions returned. If the return value is
2695 * non-negative and < num_entries, then the CQ was emptied.
2696 */
2697static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2698 struct ib_wc *wc)
2699{
2700 return cq->device->poll_cq(cq, num_entries, wc);
2701}
2702
2703/**
2704 * ib_peek_cq - Returns the number of unreaped completions currently
2705 * on the specified CQ.
2706 * @cq: The CQ to peek.
2707 * @wc_cnt: A minimum number of unreaped completions to check for.
2708 *
2709 * If the number of unreaped completions is greater than or equal to wc_cnt,
2710 * this function returns wc_cnt, otherwise, it returns the actual number of
2711 * unreaped completions.
2712 */
2713int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2714
2715/**
2716 * ib_req_notify_cq - Request completion notification on a CQ.
2717 * @cq: The CQ to generate an event for.
Roland Dreiered23a722007-05-06 21:02:48 -07002718 * @flags:
2719 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2720 * to request an event on the next solicited event or next work
2721 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2722 * may also be |ed in to request a hint about missed events, as
2723 * described below.
2724 *
2725 * Return Value:
2726 * < 0 means an error occurred while requesting notification
2727 * == 0 means notification was requested successfully, and if
2728 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2729 * were missed and it is safe to wait for another event. In
2730 * this case is it guaranteed that any work completions added
2731 * to the CQ since the last CQ poll will trigger a completion
2732 * notification event.
2733 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2734 * in. It means that the consumer must poll the CQ again to
2735 * make sure it is empty to avoid missing an event because of a
2736 * race between requesting notification and an entry being
2737 * added to the CQ. This return value means it is possible
2738 * (but not guaranteed) that a work completion has been added
2739 * to the CQ since the last poll without triggering a
2740 * completion notification event.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741 */
2742static inline int ib_req_notify_cq(struct ib_cq *cq,
Roland Dreiered23a722007-05-06 21:02:48 -07002743 enum ib_cq_notify_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744{
Roland Dreiered23a722007-05-06 21:02:48 -07002745 return cq->device->req_notify_cq(cq, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746}
2747
2748/**
2749 * ib_req_ncomp_notif - Request completion notification when there are
2750 * at least the specified number of unreaped completions on the CQ.
2751 * @cq: The CQ to generate an event for.
2752 * @wc_cnt: The number of unreaped completions that should be on the
2753 * CQ before an event is generated.
2754 */
2755static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2756{
2757 return cq->device->req_ncomp_notif ?
2758 cq->device->req_ncomp_notif(cq, wc_cnt) :
2759 -ENOSYS;
2760}
2761
2762/**
2763 * ib_get_dma_mr - Returns a memory region for system memory that is
2764 * usable for DMA.
2765 * @pd: The protection domain associated with the memory region.
2766 * @mr_access_flags: Specifies the memory access rights.
Ralph Campbell9b513092006-12-12 14:27:41 -08002767 *
2768 * Note that the ib_dma_*() functions defined below must be used
2769 * to create/destroy addresses used with the Lkey or Rkey returned
2770 * by ib_get_dma_mr().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 */
2772struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
2773
2774/**
Ralph Campbell9b513092006-12-12 14:27:41 -08002775 * ib_dma_mapping_error - check a DMA addr for error
2776 * @dev: The device for which the dma_addr was created
2777 * @dma_addr: The DMA address to check
2778 */
2779static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2780{
Ben Collinsd1998ef2006-12-13 22:10:05 -05002781 if (dev->dma_ops)
2782 return dev->dma_ops->mapping_error(dev, dma_addr);
FUJITA Tomonori8d8bb392008-07-25 19:44:49 -07002783 return dma_mapping_error(dev->dma_device, dma_addr);
Ralph Campbell9b513092006-12-12 14:27:41 -08002784}
2785
2786/**
2787 * ib_dma_map_single - Map a kernel virtual address to DMA address
2788 * @dev: The device for which the dma_addr is to be created
2789 * @cpu_addr: The kernel virtual address
2790 * @size: The size of the region in bytes
2791 * @direction: The direction of the DMA
2792 */
2793static inline u64 ib_dma_map_single(struct ib_device *dev,
2794 void *cpu_addr, size_t size,
2795 enum dma_data_direction direction)
2796{
Ben Collinsd1998ef2006-12-13 22:10:05 -05002797 if (dev->dma_ops)
2798 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2799 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
Ralph Campbell9b513092006-12-12 14:27:41 -08002800}
2801
2802/**
2803 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2804 * @dev: The device for which the DMA address was created
2805 * @addr: The DMA address
2806 * @size: The size of the region in bytes
2807 * @direction: The direction of the DMA
2808 */
2809static inline void ib_dma_unmap_single(struct ib_device *dev,
2810 u64 addr, size_t size,
2811 enum dma_data_direction direction)
2812{
Ben Collinsd1998ef2006-12-13 22:10:05 -05002813 if (dev->dma_ops)
2814 dev->dma_ops->unmap_single(dev, addr, size, direction);
2815 else
Ralph Campbell9b513092006-12-12 14:27:41 -08002816 dma_unmap_single(dev->dma_device, addr, size, direction);
2817}
2818
Arthur Kepnercb9fbc52008-04-29 01:00:34 -07002819static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2820 void *cpu_addr, size_t size,
2821 enum dma_data_direction direction,
2822 struct dma_attrs *attrs)
2823{
2824 return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2825 direction, attrs);
2826}
2827
2828static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2829 u64 addr, size_t size,
2830 enum dma_data_direction direction,
2831 struct dma_attrs *attrs)
2832{
2833 return dma_unmap_single_attrs(dev->dma_device, addr, size,
2834 direction, attrs);
2835}
2836
Ralph Campbell9b513092006-12-12 14:27:41 -08002837/**
2838 * ib_dma_map_page - Map a physical page to DMA address
2839 * @dev: The device for which the dma_addr is to be created
2840 * @page: The page to be mapped
2841 * @offset: The offset within the page
2842 * @size: The size of the region in bytes
2843 * @direction: The direction of the DMA
2844 */
2845static inline u64 ib_dma_map_page(struct ib_device *dev,
2846 struct page *page,
2847 unsigned long offset,
2848 size_t size,
2849 enum dma_data_direction direction)
2850{
Ben Collinsd1998ef2006-12-13 22:10:05 -05002851 if (dev->dma_ops)
2852 return dev->dma_ops->map_page(dev, page, offset, size, direction);
2853 return dma_map_page(dev->dma_device, page, offset, size, direction);
Ralph Campbell9b513092006-12-12 14:27:41 -08002854}
2855
2856/**
2857 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2858 * @dev: The device for which the DMA address was created
2859 * @addr: The DMA address
2860 * @size: The size of the region in bytes
2861 * @direction: The direction of the DMA
2862 */
2863static inline void ib_dma_unmap_page(struct ib_device *dev,
2864 u64 addr, size_t size,
2865 enum dma_data_direction direction)
2866{
Ben Collinsd1998ef2006-12-13 22:10:05 -05002867 if (dev->dma_ops)
2868 dev->dma_ops->unmap_page(dev, addr, size, direction);
2869 else
Ralph Campbell9b513092006-12-12 14:27:41 -08002870 dma_unmap_page(dev->dma_device, addr, size, direction);
2871}
2872
2873/**
2874 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2875 * @dev: The device for which the DMA addresses are to be created
2876 * @sg: The array of scatter/gather entries
2877 * @nents: The number of scatter/gather entries
2878 * @direction: The direction of the DMA
2879 */
2880static inline int ib_dma_map_sg(struct ib_device *dev,
2881 struct scatterlist *sg, int nents,
2882 enum dma_data_direction direction)
2883{
Ben Collinsd1998ef2006-12-13 22:10:05 -05002884 if (dev->dma_ops)
2885 return dev->dma_ops->map_sg(dev, sg, nents, direction);
2886 return dma_map_sg(dev->dma_device, sg, nents, direction);
Ralph Campbell9b513092006-12-12 14:27:41 -08002887}
2888
2889/**
2890 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2891 * @dev: The device for which the DMA addresses were created
2892 * @sg: The array of scatter/gather entries
2893 * @nents: The number of scatter/gather entries
2894 * @direction: The direction of the DMA
2895 */
2896static inline void ib_dma_unmap_sg(struct ib_device *dev,
2897 struct scatterlist *sg, int nents,
2898 enum dma_data_direction direction)
2899{
Ben Collinsd1998ef2006-12-13 22:10:05 -05002900 if (dev->dma_ops)
2901 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2902 else
Ralph Campbell9b513092006-12-12 14:27:41 -08002903 dma_unmap_sg(dev->dma_device, sg, nents, direction);
2904}
2905
Arthur Kepnercb9fbc52008-04-29 01:00:34 -07002906static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
2907 struct scatterlist *sg, int nents,
2908 enum dma_data_direction direction,
2909 struct dma_attrs *attrs)
2910{
2911 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2912}
2913
2914static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
2915 struct scatterlist *sg, int nents,
2916 enum dma_data_direction direction,
2917 struct dma_attrs *attrs)
2918{
2919 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2920}
Ralph Campbell9b513092006-12-12 14:27:41 -08002921/**
2922 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2923 * @dev: The device for which the DMA addresses were created
2924 * @sg: The scatter/gather entry
Mike Marciniszynea58a592014-03-28 13:26:59 -04002925 *
2926 * Note: this function is obsolete. To do: change all occurrences of
2927 * ib_sg_dma_address() into sg_dma_address().
Ralph Campbell9b513092006-12-12 14:27:41 -08002928 */
2929static inline u64 ib_sg_dma_address(struct ib_device *dev,
2930 struct scatterlist *sg)
2931{
Ben Collinsd1998ef2006-12-13 22:10:05 -05002932 return sg_dma_address(sg);
Ralph Campbell9b513092006-12-12 14:27:41 -08002933}
2934
2935/**
2936 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2937 * @dev: The device for which the DMA addresses were created
2938 * @sg: The scatter/gather entry
Mike Marciniszynea58a592014-03-28 13:26:59 -04002939 *
2940 * Note: this function is obsolete. To do: change all occurrences of
2941 * ib_sg_dma_len() into sg_dma_len().
Ralph Campbell9b513092006-12-12 14:27:41 -08002942 */
2943static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
2944 struct scatterlist *sg)
2945{
Ben Collinsd1998ef2006-12-13 22:10:05 -05002946 return sg_dma_len(sg);
Ralph Campbell9b513092006-12-12 14:27:41 -08002947}
2948
2949/**
2950 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2951 * @dev: The device for which the DMA address was created
2952 * @addr: The DMA address
2953 * @size: The size of the region in bytes
2954 * @dir: The direction of the DMA
2955 */
2956static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
2957 u64 addr,
2958 size_t size,
2959 enum dma_data_direction dir)
2960{
Ben Collinsd1998ef2006-12-13 22:10:05 -05002961 if (dev->dma_ops)
2962 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
2963 else
Ralph Campbell9b513092006-12-12 14:27:41 -08002964 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2965}
2966
2967/**
2968 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2969 * @dev: The device for which the DMA address was created
2970 * @addr: The DMA address
2971 * @size: The size of the region in bytes
2972 * @dir: The direction of the DMA
2973 */
2974static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
2975 u64 addr,
2976 size_t size,
2977 enum dma_data_direction dir)
2978{
Ben Collinsd1998ef2006-12-13 22:10:05 -05002979 if (dev->dma_ops)
2980 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
2981 else
Ralph Campbell9b513092006-12-12 14:27:41 -08002982 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2983}
2984
2985/**
2986 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2987 * @dev: The device for which the DMA address is requested
2988 * @size: The size of the region to allocate in bytes
2989 * @dma_handle: A pointer for returning the DMA address of the region
2990 * @flag: memory allocator flags
2991 */
2992static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2993 size_t size,
2994 u64 *dma_handle,
2995 gfp_t flag)
2996{
Ben Collinsd1998ef2006-12-13 22:10:05 -05002997 if (dev->dma_ops)
2998 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
Roland Dreierc59a3da2006-12-15 13:57:26 -08002999 else {
3000 dma_addr_t handle;
3001 void *ret;
3002
3003 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
3004 *dma_handle = handle;
3005 return ret;
3006 }
Ralph Campbell9b513092006-12-12 14:27:41 -08003007}
3008
3009/**
3010 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3011 * @dev: The device for which the DMA addresses were allocated
3012 * @size: The size of the region
3013 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3014 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3015 */
3016static inline void ib_dma_free_coherent(struct ib_device *dev,
3017 size_t size, void *cpu_addr,
3018 u64 dma_handle)
3019{
Ben Collinsd1998ef2006-12-13 22:10:05 -05003020 if (dev->dma_ops)
3021 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
3022 else
Ralph Campbell9b513092006-12-12 14:27:41 -08003023 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3024}
3025
3026/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003027 * ib_dereg_mr - Deregisters a memory region and removes it from the
3028 * HCA translation table.
3029 * @mr: The memory region to deregister.
Shani Michaeli7083e422013-02-06 16:19:12 +00003030 *
3031 * This function can fail, if the memory region has memory windows bound to it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032 */
3033int ib_dereg_mr(struct ib_mr *mr);
3034
Sagi Grimberg9bee1782015-07-30 10:32:35 +03003035struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3036 enum ib_mr_type mr_type,
3037 u32 max_num_sg);
Steve Wise00f7ec32008-07-14 23:48:45 -07003038
3039/**
Steve Wise00f7ec32008-07-14 23:48:45 -07003040 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3041 * R_Key and L_Key.
3042 * @mr - struct ib_mr pointer to be updated.
3043 * @newkey - new key to be used.
3044 */
3045static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3046{
3047 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3048 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3049}
3050
3051/**
Shani Michaeli7083e422013-02-06 16:19:12 +00003052 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3053 * for calculating a new rkey for type 2 memory windows.
3054 * @rkey - the rkey to increment.
3055 */
3056static inline u32 ib_inc_rkey(u32 rkey)
3057{
3058 const u32 mask = 0x000000ff;
3059 return ((rkey + 1) & mask) | (rkey & ~mask);
3060}
3061
3062/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3064 * @pd: The protection domain associated with the unmapped region.
3065 * @mr_access_flags: Specifies the memory access rights.
3066 * @fmr_attr: Attributes of the unmapped region.
3067 *
3068 * A fast memory region must be mapped before it can be used as part of
3069 * a work request.
3070 */
3071struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3072 int mr_access_flags,
3073 struct ib_fmr_attr *fmr_attr);
3074
3075/**
3076 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3077 * @fmr: The fast memory region to associate with the pages.
3078 * @page_list: An array of physical pages to map to the fast memory region.
3079 * @list_len: The number of pages in page_list.
3080 * @iova: The I/O virtual address to use with the mapped region.
3081 */
3082static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3083 u64 *page_list, int list_len,
3084 u64 iova)
3085{
3086 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3087}
3088
3089/**
3090 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3091 * @fmr_list: A linked list of fast memory regions to unmap.
3092 */
3093int ib_unmap_fmr(struct list_head *fmr_list);
3094
3095/**
3096 * ib_dealloc_fmr - Deallocates a fast memory region.
3097 * @fmr: The fast memory region to deallocate.
3098 */
3099int ib_dealloc_fmr(struct ib_fmr *fmr);
3100
3101/**
3102 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3103 * @qp: QP to attach to the multicast group. The QP must be type
3104 * IB_QPT_UD.
3105 * @gid: Multicast group GID.
3106 * @lid: Multicast group LID in host byte order.
3107 *
3108 * In order to send and receive multicast packets, subnet
3109 * administration must have created the multicast group and configured
3110 * the fabric appropriately. The port associated with the specified
3111 * QP must also be a member of the multicast group.
3112 */
3113int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3114
3115/**
3116 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3117 * @qp: QP to detach from the multicast group.
3118 * @gid: Multicast group GID.
3119 * @lid: Multicast group LID in host byte order.
3120 */
3121int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3122
Sean Hefty59991f92011-05-23 17:52:46 -07003123/**
3124 * ib_alloc_xrcd - Allocates an XRC domain.
3125 * @device: The device on which to allocate the XRC domain.
3126 */
3127struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3128
3129/**
3130 * ib_dealloc_xrcd - Deallocates an XRC domain.
3131 * @xrcd: The XRC domain to deallocate.
3132 */
3133int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3134
Hadar Hen Zion319a4412013-08-07 14:01:59 +03003135struct ib_flow *ib_create_flow(struct ib_qp *qp,
3136 struct ib_flow_attr *flow_attr, int domain);
3137int ib_destroy_flow(struct ib_flow *flow_id);
3138
Eli Cohen1c636f82013-10-31 15:26:32 +02003139static inline int ib_check_mr_access(int flags)
3140{
3141 /*
3142 * Local write permission is required if remote write or
3143 * remote atomic permission is also requested.
3144 */
3145 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3146 !(flags & IB_ACCESS_LOCAL_WRITE))
3147 return -EINVAL;
3148
3149 return 0;
3150}
3151
Sagi Grimberg1b01d332014-02-23 14:19:05 +02003152/**
3153 * ib_check_mr_status: lightweight check of MR status.
3154 * This routine may provide status checks on a selected
3155 * ib_mr. first use is for signature status check.
3156 *
3157 * @mr: A memory region.
3158 * @check_mask: Bitmask of which checks to perform from
3159 * ib_mr_status_check enumeration.
3160 * @mr_status: The container of relevant status checks.
3161 * failed checks will be indicated in the status bitmask
3162 * and the relevant info shall be in the error item.
3163 */
3164int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3165 struct ib_mr_status *mr_status);
3166
Yotam Kenneth9268f722015-07-30 17:50:15 +03003167struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3168 u16 pkey, const union ib_gid *gid,
3169 const struct sockaddr *addr);
3170
Christoph Hellwigff2ba992016-05-03 18:01:04 +02003171int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
Bart Van Assche9aa8b322016-05-12 10:49:15 -07003172 unsigned int *sg_offset, unsigned int page_size);
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +03003173
3174static inline int
Christoph Hellwigff2ba992016-05-03 18:01:04 +02003175ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
Bart Van Assche9aa8b322016-05-12 10:49:15 -07003176 unsigned int *sg_offset, unsigned int page_size)
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +03003177{
3178 int n;
3179
Christoph Hellwigff2ba992016-05-03 18:01:04 +02003180 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +03003181 mr->iova = 0;
3182
3183 return n;
3184}
3185
Christoph Hellwigff2ba992016-05-03 18:01:04 +02003186int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
Bart Van Assche9aa8b322016-05-12 10:49:15 -07003187 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +03003188
Steve Wise765d6772016-02-17 08:15:41 -08003189void ib_drain_rq(struct ib_qp *qp);
3190void ib_drain_sq(struct ib_qp *qp);
3191void ib_drain_qp(struct ib_qp *qp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192#endif /* IB_VERBS_H */