blob: 5ad43a487745d93103c27b68333ab507eab652e1 [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),
Max Gurtovoy47355b32016-06-06 19:34:39 +0300220 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
Sagi Grimbergf5aa9152016-02-29 19:07:32 +0200221 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
Max Gurtovoyc7e162a2016-06-06 19:34:40 +0300222 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
223 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 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
Yishai Hadasccf20562016-08-28 11:28:43 +0300264struct ib_rss_caps {
265 /* Corresponding bit will be set if qp type from
266 * 'enum ib_qp_type' is supported, e.g.
267 * supported_qpts |= 1 << IB_QPT_UD
268 */
269 u32 supported_qpts;
270 u32 max_rwq_indirection_tables;
271 u32 max_rwq_indirection_table_size;
272};
273
Matan Barakb9926b92015-06-11 16:35:22 +0300274enum ib_cq_creation_flags {
275 IB_CQ_FLAGS_TIMESTAMP_COMPLETION = 1 << 0,
Leon Romanovsky8a06ce52015-12-20 12:16:10 +0200276 IB_CQ_FLAGS_IGNORE_OVERRUN = 1 << 1,
Matan Barakb9926b92015-06-11 16:35:22 +0300277};
278
Matan Barakbcf4c1e2015-06-11 16:35:20 +0300279struct ib_cq_init_attr {
280 unsigned int cqe;
281 int comp_vector;
282 u32 flags;
283};
284
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285struct ib_device_attr {
286 u64 fw_ver;
Sean Hefty97f52eb2005-08-13 21:05:57 -0700287 __be64 sys_image_guid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 u64 max_mr_size;
289 u64 page_size_cap;
290 u32 vendor_id;
291 u32 vendor_part_id;
292 u32 hw_ver;
293 int max_qp;
294 int max_qp_wr;
Leon Romanovskyfb532d62016-02-23 10:25:25 +0200295 u64 device_cap_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 int max_sge;
297 int max_sge_rd;
298 int max_cq;
299 int max_cqe;
300 int max_mr;
301 int max_pd;
302 int max_qp_rd_atom;
303 int max_ee_rd_atom;
304 int max_res_rd_atom;
305 int max_qp_init_rd_atom;
306 int max_ee_init_rd_atom;
307 enum ib_atomic_cap atomic_cap;
Vladimir Sokolovsky5e80ba82010-04-14 17:23:01 +0300308 enum ib_atomic_cap masked_atomic_cap;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 int max_ee;
310 int max_rdd;
311 int max_mw;
312 int max_raw_ipv6_qp;
313 int max_raw_ethy_qp;
314 int max_mcast_grp;
315 int max_mcast_qp_attach;
316 int max_total_mcast_qp_attach;
317 int max_ah;
318 int max_fmr;
319 int max_map_per_fmr;
320 int max_srq;
321 int max_srq_wr;
322 int max_srq_sge;
Steve Wise00f7ec32008-07-14 23:48:45 -0700323 unsigned int max_fast_reg_page_list_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 u16 max_pkeys;
325 u8 local_ca_ack_delay;
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200326 int sig_prot_cap;
327 int sig_guard_cap;
Sagi Grimberg860f10a2014-12-11 17:04:16 +0200328 struct ib_odp_caps odp_caps;
Matan Barak24306dc2015-06-11 16:35:24 +0300329 uint64_t timestamp_mask;
330 uint64_t hca_core_clock; /* in KHZ */
Yishai Hadasccf20562016-08-28 11:28:43 +0300331 struct ib_rss_caps rss_caps;
332 u32 max_wq_type_rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333};
334
335enum ib_mtu {
336 IB_MTU_256 = 1,
337 IB_MTU_512 = 2,
338 IB_MTU_1024 = 3,
339 IB_MTU_2048 = 4,
340 IB_MTU_4096 = 5
341};
342
343static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
344{
345 switch (mtu) {
346 case IB_MTU_256: return 256;
347 case IB_MTU_512: return 512;
348 case IB_MTU_1024: return 1024;
349 case IB_MTU_2048: return 2048;
350 case IB_MTU_4096: return 4096;
351 default: return -1;
352 }
353}
354
355enum ib_port_state {
356 IB_PORT_NOP = 0,
357 IB_PORT_DOWN = 1,
358 IB_PORT_INIT = 2,
359 IB_PORT_ARMED = 3,
360 IB_PORT_ACTIVE = 4,
361 IB_PORT_ACTIVE_DEFER = 5
362};
363
364enum ib_port_cap_flags {
365 IB_PORT_SM = 1 << 1,
366 IB_PORT_NOTICE_SUP = 1 << 2,
367 IB_PORT_TRAP_SUP = 1 << 3,
368 IB_PORT_OPT_IPD_SUP = 1 << 4,
369 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
370 IB_PORT_SL_MAP_SUP = 1 << 6,
371 IB_PORT_MKEY_NVRAM = 1 << 7,
372 IB_PORT_PKEY_NVRAM = 1 << 8,
373 IB_PORT_LED_INFO_SUP = 1 << 9,
374 IB_PORT_SM_DISABLED = 1 << 10,
375 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
376 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
Marcel Apfelbaum71eeba12011-10-05 14:21:47 +0300377 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 IB_PORT_CM_SUP = 1 << 16,
379 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
380 IB_PORT_REINIT_SUP = 1 << 18,
381 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
382 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
383 IB_PORT_DR_NOTICE_SUP = 1 << 21,
384 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
385 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
386 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
Moni Shouab4a26a22014-02-09 11:54:34 +0200387 IB_PORT_CLIENT_REG_SUP = 1 << 25,
Matan Barak03db3a22015-07-30 18:33:26 +0300388 IB_PORT_IP_BASED_GIDS = 1 << 26,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389};
390
391enum ib_port_width {
392 IB_WIDTH_1X = 1,
393 IB_WIDTH_4X = 2,
394 IB_WIDTH_8X = 4,
395 IB_WIDTH_12X = 8
396};
397
398static inline int ib_width_enum_to_int(enum ib_port_width width)
399{
400 switch (width) {
401 case IB_WIDTH_1X: return 1;
402 case IB_WIDTH_4X: return 4;
403 case IB_WIDTH_8X: return 8;
404 case IB_WIDTH_12X: return 12;
405 default: return -1;
406 }
407}
408
Or Gerlitz2e966912012-02-28 18:49:50 +0200409enum ib_port_speed {
410 IB_SPEED_SDR = 1,
411 IB_SPEED_DDR = 2,
412 IB_SPEED_QDR = 4,
413 IB_SPEED_FDR10 = 8,
414 IB_SPEED_FDR = 16,
415 IB_SPEED_EDR = 32
416};
417
Christoph Lameterb40f4752016-05-16 12:49:33 -0500418/**
419 * struct rdma_hw_stats
420 * @timestamp - Used by the core code to track when the last update was
421 * @lifespan - Used by the core code to determine how old the counters
422 * should be before being updated again. Stored in jiffies, defaults
423 * to 10 milliseconds, drivers can override the default be specifying
424 * their own value during their allocation routine.
425 * @name - Array of pointers to static names used for the counters in
426 * directory.
427 * @num_counters - How many hardware counters there are. If name is
428 * shorter than this number, a kernel oops will result. Driver authors
429 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
430 * in their code to prevent this.
431 * @value - Array of u64 counters that are accessed by the sysfs code and
432 * filled in by the drivers get_stats routine
433 */
434struct rdma_hw_stats {
435 unsigned long timestamp;
436 unsigned long lifespan;
437 const char * const *names;
438 int num_counters;
439 u64 value[];
Steve Wise7f624d02008-07-14 23:48:48 -0700440};
441
Christoph Lameterb40f4752016-05-16 12:49:33 -0500442#define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
443/**
444 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
445 * for drivers.
446 * @names - Array of static const char *
447 * @num_counters - How many elements in array
448 * @lifespan - How many milliseconds between updates
449 */
450static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
451 const char * const *names, int num_counters,
452 unsigned long lifespan)
453{
454 struct rdma_hw_stats *stats;
Steve Wise7f624d02008-07-14 23:48:48 -0700455
Christoph Lameterb40f4752016-05-16 12:49:33 -0500456 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
457 GFP_KERNEL);
458 if (!stats)
459 return NULL;
460 stats->names = names;
461 stats->num_counters = num_counters;
462 stats->lifespan = msecs_to_jiffies(lifespan);
Steve Wise7f624d02008-07-14 23:48:48 -0700463
Christoph Lameterb40f4752016-05-16 12:49:33 -0500464 return stats;
465}
466
Steve Wise7f624d02008-07-14 23:48:48 -0700467
Ira Weinyf9b22e32015-05-13 20:02:59 -0400468/* Define bits for the various functionality this port needs to be supported by
469 * the core.
470 */
471/* Management 0x00000FFF */
472#define RDMA_CORE_CAP_IB_MAD 0x00000001
473#define RDMA_CORE_CAP_IB_SMI 0x00000002
474#define RDMA_CORE_CAP_IB_CM 0x00000004
475#define RDMA_CORE_CAP_IW_CM 0x00000008
476#define RDMA_CORE_CAP_IB_SA 0x00000010
Ira Weiny65995fe2015-06-06 14:38:32 -0400477#define RDMA_CORE_CAP_OPA_MAD 0x00000020
Ira Weinyf9b22e32015-05-13 20:02:59 -0400478
479/* Address format 0x000FF000 */
480#define RDMA_CORE_CAP_AF_IB 0x00001000
481#define RDMA_CORE_CAP_ETH_AH 0x00002000
482
483/* Protocol 0xFFF00000 */
484#define RDMA_CORE_CAP_PROT_IB 0x00100000
485#define RDMA_CORE_CAP_PROT_ROCE 0x00200000
486#define RDMA_CORE_CAP_PROT_IWARP 0x00400000
Matan Barak7766a992015-12-23 14:56:50 +0200487#define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
Ira Weinyf9b22e32015-05-13 20:02:59 -0400488
489#define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
490 | RDMA_CORE_CAP_IB_MAD \
491 | RDMA_CORE_CAP_IB_SMI \
492 | RDMA_CORE_CAP_IB_CM \
493 | RDMA_CORE_CAP_IB_SA \
494 | RDMA_CORE_CAP_AF_IB)
495#define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
496 | RDMA_CORE_CAP_IB_MAD \
497 | RDMA_CORE_CAP_IB_CM \
Ira Weinyf9b22e32015-05-13 20:02:59 -0400498 | RDMA_CORE_CAP_AF_IB \
499 | RDMA_CORE_CAP_ETH_AH)
Matan Barak7766a992015-12-23 14:56:50 +0200500#define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
501 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
502 | RDMA_CORE_CAP_IB_MAD \
503 | RDMA_CORE_CAP_IB_CM \
504 | RDMA_CORE_CAP_AF_IB \
505 | RDMA_CORE_CAP_ETH_AH)
Ira Weinyf9b22e32015-05-13 20:02:59 -0400506#define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
507 | RDMA_CORE_CAP_IW_CM)
Ira Weiny65995fe2015-06-06 14:38:32 -0400508#define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
509 | RDMA_CORE_CAP_OPA_MAD)
Ira Weinyf9b22e32015-05-13 20:02:59 -0400510
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511struct ib_port_attr {
Eli Cohenfad61ad2016-03-11 22:58:36 +0200512 u64 subnet_prefix;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 enum ib_port_state state;
514 enum ib_mtu max_mtu;
515 enum ib_mtu active_mtu;
516 int gid_tbl_len;
517 u32 port_cap_flags;
518 u32 max_msg_sz;
519 u32 bad_pkey_cntr;
520 u32 qkey_viol_cntr;
521 u16 pkey_tbl_len;
522 u16 lid;
523 u16 sm_lid;
524 u8 lmc;
525 u8 max_vl_num;
526 u8 sm_sl;
527 u8 subnet_timeout;
528 u8 init_type_reply;
529 u8 active_width;
530 u8 active_speed;
531 u8 phys_state;
Eli Cohena0c1b2a2016-03-11 22:58:37 +0200532 bool grh_required;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533};
534
535enum ib_device_modify_flags {
Roland Dreierc5bcbbb2006-02-02 09:47:14 -0800536 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
537 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538};
539
Yuval Shaiabd99fde2016-08-25 10:57:07 -0700540#define IB_DEVICE_NODE_DESC_MAX 64
541
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542struct ib_device_modify {
543 u64 sys_image_guid;
Yuval Shaiabd99fde2016-08-25 10:57:07 -0700544 char node_desc[IB_DEVICE_NODE_DESC_MAX];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545};
546
547enum ib_port_modify_flags {
548 IB_PORT_SHUTDOWN = 1,
549 IB_PORT_INIT_TYPE = (1<<2),
550 IB_PORT_RESET_QKEY_CNTR = (1<<3)
551};
552
553struct ib_port_modify {
554 u32 set_port_cap_mask;
555 u32 clr_port_cap_mask;
556 u8 init_type;
557};
558
559enum ib_event_type {
560 IB_EVENT_CQ_ERR,
561 IB_EVENT_QP_FATAL,
562 IB_EVENT_QP_REQ_ERR,
563 IB_EVENT_QP_ACCESS_ERR,
564 IB_EVENT_COMM_EST,
565 IB_EVENT_SQ_DRAINED,
566 IB_EVENT_PATH_MIG,
567 IB_EVENT_PATH_MIG_ERR,
568 IB_EVENT_DEVICE_FATAL,
569 IB_EVENT_PORT_ACTIVE,
570 IB_EVENT_PORT_ERR,
571 IB_EVENT_LID_CHANGE,
572 IB_EVENT_PKEY_CHANGE,
Roland Dreierd41fcc62005-08-18 12:23:08 -0700573 IB_EVENT_SM_CHANGE,
574 IB_EVENT_SRQ_ERR,
575 IB_EVENT_SRQ_LIMIT_REACHED,
Leonid Arsh63942c92006-06-17 20:37:35 -0700576 IB_EVENT_QP_LAST_WQE_REACHED,
Or Gerlitz761d90e2011-06-15 14:39:29 +0000577 IB_EVENT_CLIENT_REREGISTER,
578 IB_EVENT_GID_CHANGE,
Yishai Hadasf213c052016-05-23 15:20:49 +0300579 IB_EVENT_WQ_FATAL,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580};
581
Bart Van Asschedb7489e2015-08-03 10:01:52 -0700582const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
Sagi Grimberg2b1b5b62015-05-18 13:40:28 +0300583
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584struct ib_event {
585 struct ib_device *device;
586 union {
587 struct ib_cq *cq;
588 struct ib_qp *qp;
Roland Dreierd41fcc62005-08-18 12:23:08 -0700589 struct ib_srq *srq;
Yishai Hadasf213c052016-05-23 15:20:49 +0300590 struct ib_wq *wq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 u8 port_num;
592 } element;
593 enum ib_event_type event;
594};
595
596struct ib_event_handler {
597 struct ib_device *device;
598 void (*handler)(struct ib_event_handler *, struct ib_event *);
599 struct list_head list;
600};
601
602#define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
603 do { \
604 (_ptr)->device = _device; \
605 (_ptr)->handler = _handler; \
606 INIT_LIST_HEAD(&(_ptr)->list); \
607 } while (0)
608
609struct ib_global_route {
610 union ib_gid dgid;
611 u32 flow_label;
612 u8 sgid_index;
613 u8 hop_limit;
614 u8 traffic_class;
615};
616
Hal Rosenstock513789e2005-07-27 11:45:34 -0700617struct ib_grh {
Sean Hefty97f52eb2005-08-13 21:05:57 -0700618 __be32 version_tclass_flow;
619 __be16 paylen;
Hal Rosenstock513789e2005-07-27 11:45:34 -0700620 u8 next_hdr;
621 u8 hop_limit;
622 union ib_gid sgid;
623 union ib_gid dgid;
624};
625
Somnath Koturc865f242015-12-23 14:56:51 +0200626union rdma_network_hdr {
627 struct ib_grh ibgrh;
628 struct {
629 /* The IB spec states that if it's IPv4, the header
630 * is located in the last 20 bytes of the header.
631 */
632 u8 reserved[20];
633 struct iphdr roce4grh;
634 };
635};
636
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637enum {
638 IB_MULTICAST_QPN = 0xffffff
639};
640
Harvey Harrisonf3a7c662009-02-14 22:58:35 -0800641#define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
Dennis Dalessandrob4e64392016-01-06 10:04:31 -0800642#define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
Sean Hefty97f52eb2005-08-13 21:05:57 -0700643
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644enum ib_ah_flags {
645 IB_AH_GRH = 1
646};
647
Jack Morgensteinbf6a9e32006-04-10 09:43:47 -0700648enum ib_rate {
649 IB_RATE_PORT_CURRENT = 0,
650 IB_RATE_2_5_GBPS = 2,
651 IB_RATE_5_GBPS = 5,
652 IB_RATE_10_GBPS = 3,
653 IB_RATE_20_GBPS = 6,
654 IB_RATE_30_GBPS = 4,
655 IB_RATE_40_GBPS = 7,
656 IB_RATE_60_GBPS = 8,
657 IB_RATE_80_GBPS = 9,
Marcel Apfelbaum71eeba12011-10-05 14:21:47 +0300658 IB_RATE_120_GBPS = 10,
659 IB_RATE_14_GBPS = 11,
660 IB_RATE_56_GBPS = 12,
661 IB_RATE_112_GBPS = 13,
662 IB_RATE_168_GBPS = 14,
663 IB_RATE_25_GBPS = 15,
664 IB_RATE_100_GBPS = 16,
665 IB_RATE_200_GBPS = 17,
666 IB_RATE_300_GBPS = 18
Jack Morgensteinbf6a9e32006-04-10 09:43:47 -0700667};
668
669/**
670 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
671 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
672 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
673 * @rate: rate to convert.
674 */
Roland Dreier8385fd82014-06-04 10:00:16 -0700675__attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
Jack Morgensteinbf6a9e32006-04-10 09:43:47 -0700676
677/**
Marcel Apfelbaum71eeba12011-10-05 14:21:47 +0300678 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
679 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
680 * @rate: rate to convert.
681 */
Roland Dreier8385fd82014-06-04 10:00:16 -0700682__attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
Marcel Apfelbaum71eeba12011-10-05 14:21:47 +0300683
Sagi Grimberg17cd3a22014-02-23 14:19:04 +0200684
685/**
Sagi Grimberg9bee1782015-07-30 10:32:35 +0300686 * enum ib_mr_type - memory region type
687 * @IB_MR_TYPE_MEM_REG: memory region that is used for
688 * normal registration
689 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
690 * signature operations (data-integrity
691 * capable regions)
Sagi Grimbergf5aa9152016-02-29 19:07:32 +0200692 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
693 * register any arbitrary sg lists (without
694 * the normal mr constraints - see
695 * ib_map_mr_sg)
Sagi Grimberg17cd3a22014-02-23 14:19:04 +0200696 */
Sagi Grimberg9bee1782015-07-30 10:32:35 +0300697enum ib_mr_type {
698 IB_MR_TYPE_MEM_REG,
699 IB_MR_TYPE_SIGNATURE,
Sagi Grimbergf5aa9152016-02-29 19:07:32 +0200700 IB_MR_TYPE_SG_GAPS,
Sagi Grimberg17cd3a22014-02-23 14:19:04 +0200701};
702
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200703/**
Sagi Grimberg78eda2b2014-08-13 19:54:35 +0300704 * Signature types
705 * IB_SIG_TYPE_NONE: Unprotected.
706 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200707 */
Sagi Grimberg78eda2b2014-08-13 19:54:35 +0300708enum ib_signature_type {
709 IB_SIG_TYPE_NONE,
710 IB_SIG_TYPE_T10_DIF,
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200711};
712
713/**
714 * Signature T10-DIF block-guard types
715 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
716 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
717 */
718enum ib_t10_dif_bg_type {
719 IB_T10DIF_CRC,
720 IB_T10DIF_CSUM
721};
722
723/**
724 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
725 * domain.
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200726 * @bg_type: T10-DIF block guard type (CRC|CSUM)
727 * @pi_interval: protection information interval.
728 * @bg: seed of guard computation.
729 * @app_tag: application tag of guard block
730 * @ref_tag: initial guard block reference tag.
Sagi Grimberg78eda2b2014-08-13 19:54:35 +0300731 * @ref_remap: Indicate wethear the reftag increments each block
732 * @app_escape: Indicate to skip block check if apptag=0xffff
733 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
734 * @apptag_check_mask: check bitmask of application tag.
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200735 */
736struct ib_t10_dif_domain {
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200737 enum ib_t10_dif_bg_type bg_type;
738 u16 pi_interval;
739 u16 bg;
740 u16 app_tag;
741 u32 ref_tag;
Sagi Grimberg78eda2b2014-08-13 19:54:35 +0300742 bool ref_remap;
743 bool app_escape;
744 bool ref_escape;
745 u16 apptag_check_mask;
Sagi Grimberg1b01d332014-02-23 14:19:05 +0200746};
747
748/**
749 * struct ib_sig_domain - Parameters for signature domain
750 * @sig_type: specific signauture type
751 * @sig: union of all signature domain attributes that may
752 * be used to set domain layout.
753 */
754struct ib_sig_domain {
755 enum ib_signature_type sig_type;
756 union {
757 struct ib_t10_dif_domain dif;
758 } sig;
759};
760
761/**
762 * struct ib_sig_attrs - Parameters for signature handover operation
763 * @check_mask: bitmask for signature byte check (8 bytes)
764 * @mem: memory domain layout desciptor.
765 * @wire: wire domain layout desciptor.
766 */
767struct ib_sig_attrs {
768 u8 check_mask;
769 struct ib_sig_domain mem;
770 struct ib_sig_domain wire;
771};
772
773enum ib_sig_err_type {
774 IB_SIG_BAD_GUARD,
775 IB_SIG_BAD_REFTAG,
776 IB_SIG_BAD_APPTAG,
777};
778
779/**
780 * struct ib_sig_err - signature error descriptor
781 */
782struct ib_sig_err {
783 enum ib_sig_err_type err_type;
784 u32 expected;
785 u32 actual;
786 u64 sig_err_offset;
787 u32 key;
788};
789
790enum ib_mr_status_check {
791 IB_MR_CHECK_SIG_STATUS = 1,
792};
793
794/**
795 * struct ib_mr_status - Memory region status container
796 *
797 * @fail_status: Bitmask of MR checks status. For each
798 * failed check a corresponding status bit is set.
799 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
800 * failure.
801 */
802struct ib_mr_status {
803 u32 fail_status;
804 struct ib_sig_err sig_err;
805};
806
Marcel Apfelbaum71eeba12011-10-05 14:21:47 +0300807/**
Jack Morgensteinbf6a9e32006-04-10 09:43:47 -0700808 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
809 * enum.
810 * @mult: multiple to convert.
811 */
Roland Dreier8385fd82014-06-04 10:00:16 -0700812__attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
Jack Morgensteinbf6a9e32006-04-10 09:43:47 -0700813
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814struct ib_ah_attr {
815 struct ib_global_route grh;
816 u16 dlid;
817 u8 sl;
818 u8 src_path_bits;
819 u8 static_rate;
820 u8 ah_flags;
821 u8 port_num;
Matan Barakdd5f03b2013-12-12 18:03:11 +0200822 u8 dmac[ETH_ALEN];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823};
824
825enum ib_wc_status {
826 IB_WC_SUCCESS,
827 IB_WC_LOC_LEN_ERR,
828 IB_WC_LOC_QP_OP_ERR,
829 IB_WC_LOC_EEC_OP_ERR,
830 IB_WC_LOC_PROT_ERR,
831 IB_WC_WR_FLUSH_ERR,
832 IB_WC_MW_BIND_ERR,
833 IB_WC_BAD_RESP_ERR,
834 IB_WC_LOC_ACCESS_ERR,
835 IB_WC_REM_INV_REQ_ERR,
836 IB_WC_REM_ACCESS_ERR,
837 IB_WC_REM_OP_ERR,
838 IB_WC_RETRY_EXC_ERR,
839 IB_WC_RNR_RETRY_EXC_ERR,
840 IB_WC_LOC_RDD_VIOL_ERR,
841 IB_WC_REM_INV_RD_REQ_ERR,
842 IB_WC_REM_ABORT_ERR,
843 IB_WC_INV_EECN_ERR,
844 IB_WC_INV_EEC_STATE_ERR,
845 IB_WC_FATAL_ERR,
846 IB_WC_RESP_TIMEOUT_ERR,
847 IB_WC_GENERAL_ERR
848};
849
Bart Van Asschedb7489e2015-08-03 10:01:52 -0700850const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
Sagi Grimberg2b1b5b62015-05-18 13:40:28 +0300851
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852enum ib_wc_opcode {
853 IB_WC_SEND,
854 IB_WC_RDMA_WRITE,
855 IB_WC_RDMA_READ,
856 IB_WC_COMP_SWAP,
857 IB_WC_FETCH_ADD,
Eli Cohenc93570f2008-04-16 21:09:27 -0700858 IB_WC_LSO,
Steve Wise00f7ec32008-07-14 23:48:45 -0700859 IB_WC_LOCAL_INV,
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +0300860 IB_WC_REG_MR,
Vladimir Sokolovsky5e80ba82010-04-14 17:23:01 +0300861 IB_WC_MASKED_COMP_SWAP,
862 IB_WC_MASKED_FETCH_ADD,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863/*
864 * Set value of IB_WC_RECV so consumers can test if a completion is a
865 * receive by testing (opcode & IB_WC_RECV).
866 */
867 IB_WC_RECV = 1 << 7,
868 IB_WC_RECV_RDMA_WITH_IMM
869};
870
871enum ib_wc_flags {
872 IB_WC_GRH = 1,
Steve Wise00f7ec32008-07-14 23:48:45 -0700873 IB_WC_WITH_IMM = (1<<1),
874 IB_WC_WITH_INVALIDATE = (1<<2),
Or Gerlitzd927d502012-01-11 19:03:51 +0200875 IB_WC_IP_CSUM_OK = (1<<3),
Matan Barakdd5f03b2013-12-12 18:03:11 +0200876 IB_WC_WITH_SMAC = (1<<4),
877 IB_WC_WITH_VLAN = (1<<5),
Somnath Koturc865f242015-12-23 14:56:51 +0200878 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879};
880
881struct ib_wc {
Christoph Hellwig14d3a3b2015-12-11 11:53:03 -0800882 union {
883 u64 wr_id;
884 struct ib_cqe *wr_cqe;
885 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 enum ib_wc_status status;
887 enum ib_wc_opcode opcode;
888 u32 vendor_err;
889 u32 byte_len;
Michael S. Tsirkin062dbb62006-12-31 21:09:42 +0200890 struct ib_qp *qp;
Steve Wise00f7ec32008-07-14 23:48:45 -0700891 union {
892 __be32 imm_data;
893 u32 invalidate_rkey;
894 } ex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 u32 src_qp;
896 int wc_flags;
897 u16 pkey_index;
898 u16 slid;
899 u8 sl;
900 u8 dlid_path_bits;
901 u8 port_num; /* valid only for DR SMPs on switches */
Matan Barakdd5f03b2013-12-12 18:03:11 +0200902 u8 smac[ETH_ALEN];
903 u16 vlan_id;
Somnath Koturc865f242015-12-23 14:56:51 +0200904 u8 network_hdr_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905};
906
Roland Dreiered23a722007-05-06 21:02:48 -0700907enum ib_cq_notify_flags {
908 IB_CQ_SOLICITED = 1 << 0,
909 IB_CQ_NEXT_COMP = 1 << 1,
910 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
911 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912};
913
Sean Hefty96104ed2011-05-23 16:31:36 -0700914enum ib_srq_type {
Sean Hefty418d5132011-05-23 19:42:29 -0700915 IB_SRQT_BASIC,
916 IB_SRQT_XRC
Sean Hefty96104ed2011-05-23 16:31:36 -0700917};
918
Roland Dreierd41fcc62005-08-18 12:23:08 -0700919enum ib_srq_attr_mask {
920 IB_SRQ_MAX_WR = 1 << 0,
921 IB_SRQ_LIMIT = 1 << 1,
922};
923
924struct ib_srq_attr {
925 u32 max_wr;
926 u32 max_sge;
927 u32 srq_limit;
928};
929
930struct ib_srq_init_attr {
931 void (*event_handler)(struct ib_event *, void *);
932 void *srq_context;
933 struct ib_srq_attr attr;
Sean Hefty96104ed2011-05-23 16:31:36 -0700934 enum ib_srq_type srq_type;
Sean Hefty418d5132011-05-23 19:42:29 -0700935
936 union {
937 struct {
938 struct ib_xrcd *xrcd;
939 struct ib_cq *cq;
940 } xrc;
941 } ext;
Roland Dreierd41fcc62005-08-18 12:23:08 -0700942};
943
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944struct ib_qp_cap {
945 u32 max_send_wr;
946 u32 max_recv_wr;
947 u32 max_send_sge;
948 u32 max_recv_sge;
949 u32 max_inline_data;
Christoph Hellwiga060b562016-05-03 18:01:09 +0200950
951 /*
952 * Maximum number of rdma_rw_ctx structures in flight at a time.
953 * ib_create_qp() will calculate the right amount of neededed WRs
954 * and MRs based on this.
955 */
956 u32 max_rdma_ctxs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957};
958
959enum ib_sig_type {
960 IB_SIGNAL_ALL_WR,
961 IB_SIGNAL_REQ_WR
962};
963
964enum ib_qp_type {
965 /*
966 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
967 * here (and in that order) since the MAD layer uses them as
968 * indices into a 2-entry table.
969 */
970 IB_QPT_SMI,
971 IB_QPT_GSI,
972
973 IB_QPT_RC,
974 IB_QPT_UC,
975 IB_QPT_UD,
976 IB_QPT_RAW_IPV6,
Sean Heftyb42b63c2011-05-23 19:59:25 -0700977 IB_QPT_RAW_ETHERTYPE,
Or Gerlitzc938a612012-03-01 12:17:51 +0200978 IB_QPT_RAW_PACKET = 8,
Sean Heftyb42b63c2011-05-23 19:59:25 -0700979 IB_QPT_XRC_INI = 9,
980 IB_QPT_XRC_TGT,
Jack Morgenstein0134f162013-07-07 17:25:52 +0300981 IB_QPT_MAX,
982 /* Reserve a range for qp types internal to the low level driver.
983 * These qp types will not be visible at the IB core layer, so the
984 * IB_QPT_MAX usages should not be affected in the core layer
985 */
986 IB_QPT_RESERVED1 = 0x1000,
987 IB_QPT_RESERVED2,
988 IB_QPT_RESERVED3,
989 IB_QPT_RESERVED4,
990 IB_QPT_RESERVED5,
991 IB_QPT_RESERVED6,
992 IB_QPT_RESERVED7,
993 IB_QPT_RESERVED8,
994 IB_QPT_RESERVED9,
995 IB_QPT_RESERVED10,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996};
997
Eli Cohenb846f252008-04-16 21:09:27 -0700998enum ib_qp_create_flags {
Ron Livne47ee1b92008-07-14 23:48:48 -0700999 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1000 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
Leon Romanovsky8a06ce52015-12-20 12:16:10 +02001001 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1002 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1003 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
Matan Barak90f1d1b2013-11-07 15:25:12 +02001004 IB_QP_CREATE_NETIF_QP = 1 << 5,
Sagi Grimberg1b01d332014-02-23 14:19:05 +02001005 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
Or Gerlitz09b93082014-05-11 15:15:11 +03001006 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7,
Majd Dibbinyb531b902016-04-17 17:19:36 +03001007 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
Jack Morgensteind2b57062012-08-03 08:40:37 +00001008 /* reserve bits 26-31 for low level drivers' internal use */
1009 IB_QP_CREATE_RESERVED_START = 1 << 26,
1010 IB_QP_CREATE_RESERVED_END = 1 << 31,
Eli Cohenb846f252008-04-16 21:09:27 -07001011};
1012
Yishai Hadas73c40c62013-08-01 18:49:53 +03001013/*
1014 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1015 * callback to destroy the passed in QP.
1016 */
1017
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018struct ib_qp_init_attr {
1019 void (*event_handler)(struct ib_event *, void *);
1020 void *qp_context;
1021 struct ib_cq *send_cq;
1022 struct ib_cq *recv_cq;
1023 struct ib_srq *srq;
Sean Heftyb42b63c2011-05-23 19:59:25 -07001024 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 struct ib_qp_cap cap;
1026 enum ib_sig_type sq_sig_type;
1027 enum ib_qp_type qp_type;
Eli Cohenb846f252008-04-16 21:09:27 -07001028 enum ib_qp_create_flags create_flags;
Christoph Hellwiga060b562016-05-03 18:01:09 +02001029
1030 /*
1031 * Only needed for special QP types, or when using the RW API.
1032 */
1033 u8 port_num;
Yishai Hadasa9017e22016-05-23 15:20:54 +03001034 struct ib_rwq_ind_table *rwq_ind_tbl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035};
1036
Sean Hefty0e0ec7e2011-08-08 15:31:51 -07001037struct ib_qp_open_attr {
1038 void (*event_handler)(struct ib_event *, void *);
1039 void *qp_context;
1040 u32 qp_num;
1041 enum ib_qp_type qp_type;
1042};
1043
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044enum ib_rnr_timeout {
1045 IB_RNR_TIMER_655_36 = 0,
1046 IB_RNR_TIMER_000_01 = 1,
1047 IB_RNR_TIMER_000_02 = 2,
1048 IB_RNR_TIMER_000_03 = 3,
1049 IB_RNR_TIMER_000_04 = 4,
1050 IB_RNR_TIMER_000_06 = 5,
1051 IB_RNR_TIMER_000_08 = 6,
1052 IB_RNR_TIMER_000_12 = 7,
1053 IB_RNR_TIMER_000_16 = 8,
1054 IB_RNR_TIMER_000_24 = 9,
1055 IB_RNR_TIMER_000_32 = 10,
1056 IB_RNR_TIMER_000_48 = 11,
1057 IB_RNR_TIMER_000_64 = 12,
1058 IB_RNR_TIMER_000_96 = 13,
1059 IB_RNR_TIMER_001_28 = 14,
1060 IB_RNR_TIMER_001_92 = 15,
1061 IB_RNR_TIMER_002_56 = 16,
1062 IB_RNR_TIMER_003_84 = 17,
1063 IB_RNR_TIMER_005_12 = 18,
1064 IB_RNR_TIMER_007_68 = 19,
1065 IB_RNR_TIMER_010_24 = 20,
1066 IB_RNR_TIMER_015_36 = 21,
1067 IB_RNR_TIMER_020_48 = 22,
1068 IB_RNR_TIMER_030_72 = 23,
1069 IB_RNR_TIMER_040_96 = 24,
1070 IB_RNR_TIMER_061_44 = 25,
1071 IB_RNR_TIMER_081_92 = 26,
1072 IB_RNR_TIMER_122_88 = 27,
1073 IB_RNR_TIMER_163_84 = 28,
1074 IB_RNR_TIMER_245_76 = 29,
1075 IB_RNR_TIMER_327_68 = 30,
1076 IB_RNR_TIMER_491_52 = 31
1077};
1078
1079enum ib_qp_attr_mask {
1080 IB_QP_STATE = 1,
1081 IB_QP_CUR_STATE = (1<<1),
1082 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1083 IB_QP_ACCESS_FLAGS = (1<<3),
1084 IB_QP_PKEY_INDEX = (1<<4),
1085 IB_QP_PORT = (1<<5),
1086 IB_QP_QKEY = (1<<6),
1087 IB_QP_AV = (1<<7),
1088 IB_QP_PATH_MTU = (1<<8),
1089 IB_QP_TIMEOUT = (1<<9),
1090 IB_QP_RETRY_CNT = (1<<10),
1091 IB_QP_RNR_RETRY = (1<<11),
1092 IB_QP_RQ_PSN = (1<<12),
1093 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1094 IB_QP_ALT_PATH = (1<<14),
1095 IB_QP_MIN_RNR_TIMER = (1<<15),
1096 IB_QP_SQ_PSN = (1<<16),
1097 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1098 IB_QP_PATH_MIG_STATE = (1<<18),
1099 IB_QP_CAP = (1<<19),
Matan Barakdd5f03b2013-12-12 18:03:11 +02001100 IB_QP_DEST_QPN = (1<<20),
Matan Barakaa744cc2015-10-15 18:38:53 +03001101 IB_QP_RESERVED1 = (1<<21),
1102 IB_QP_RESERVED2 = (1<<22),
1103 IB_QP_RESERVED3 = (1<<23),
1104 IB_QP_RESERVED4 = (1<<24),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105};
1106
1107enum ib_qp_state {
1108 IB_QPS_RESET,
1109 IB_QPS_INIT,
1110 IB_QPS_RTR,
1111 IB_QPS_RTS,
1112 IB_QPS_SQD,
1113 IB_QPS_SQE,
1114 IB_QPS_ERR
1115};
1116
1117enum ib_mig_state {
1118 IB_MIG_MIGRATED,
1119 IB_MIG_REARM,
1120 IB_MIG_ARMED
1121};
1122
Shani Michaeli7083e422013-02-06 16:19:12 +00001123enum ib_mw_type {
1124 IB_MW_TYPE_1 = 1,
1125 IB_MW_TYPE_2 = 2
1126};
1127
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128struct ib_qp_attr {
1129 enum ib_qp_state qp_state;
1130 enum ib_qp_state cur_qp_state;
1131 enum ib_mtu path_mtu;
1132 enum ib_mig_state path_mig_state;
1133 u32 qkey;
1134 u32 rq_psn;
1135 u32 sq_psn;
1136 u32 dest_qp_num;
1137 int qp_access_flags;
1138 struct ib_qp_cap cap;
1139 struct ib_ah_attr ah_attr;
1140 struct ib_ah_attr alt_ah_attr;
1141 u16 pkey_index;
1142 u16 alt_pkey_index;
1143 u8 en_sqd_async_notify;
1144 u8 sq_draining;
1145 u8 max_rd_atomic;
1146 u8 max_dest_rd_atomic;
1147 u8 min_rnr_timer;
1148 u8 port_num;
1149 u8 timeout;
1150 u8 retry_cnt;
1151 u8 rnr_retry;
1152 u8 alt_port_num;
1153 u8 alt_timeout;
1154};
1155
1156enum ib_wr_opcode {
1157 IB_WR_RDMA_WRITE,
1158 IB_WR_RDMA_WRITE_WITH_IMM,
1159 IB_WR_SEND,
1160 IB_WR_SEND_WITH_IMM,
1161 IB_WR_RDMA_READ,
1162 IB_WR_ATOMIC_CMP_AND_SWP,
Eli Cohenc93570f2008-04-16 21:09:27 -07001163 IB_WR_ATOMIC_FETCH_AND_ADD,
Roland Dreier0f39cf32008-04-16 21:09:32 -07001164 IB_WR_LSO,
1165 IB_WR_SEND_WITH_INV,
Steve Wise00f7ec32008-07-14 23:48:45 -07001166 IB_WR_RDMA_READ_WITH_INV,
1167 IB_WR_LOCAL_INV,
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +03001168 IB_WR_REG_MR,
Vladimir Sokolovsky5e80ba82010-04-14 17:23:01 +03001169 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1170 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
Sagi Grimberg1b01d332014-02-23 14:19:05 +02001171 IB_WR_REG_SIG_MR,
Jack Morgenstein0134f162013-07-07 17:25:52 +03001172 /* reserve values for low level drivers' internal use.
1173 * These values will not be used at all in the ib core layer.
1174 */
1175 IB_WR_RESERVED1 = 0xf0,
1176 IB_WR_RESERVED2,
1177 IB_WR_RESERVED3,
1178 IB_WR_RESERVED4,
1179 IB_WR_RESERVED5,
1180 IB_WR_RESERVED6,
1181 IB_WR_RESERVED7,
1182 IB_WR_RESERVED8,
1183 IB_WR_RESERVED9,
1184 IB_WR_RESERVED10,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185};
1186
1187enum ib_send_flags {
1188 IB_SEND_FENCE = 1,
1189 IB_SEND_SIGNALED = (1<<1),
1190 IB_SEND_SOLICITED = (1<<2),
Eli Cohene0605d92008-01-30 18:30:57 +02001191 IB_SEND_INLINE = (1<<3),
Jack Morgenstein0134f162013-07-07 17:25:52 +03001192 IB_SEND_IP_CSUM = (1<<4),
1193
1194 /* reserve bits 26-31 for low level drivers' internal use */
1195 IB_SEND_RESERVED_START = (1 << 26),
1196 IB_SEND_RESERVED_END = (1 << 31),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197};
1198
1199struct ib_sge {
1200 u64 addr;
1201 u32 length;
1202 u32 lkey;
1203};
1204
Christoph Hellwig14d3a3b2015-12-11 11:53:03 -08001205struct ib_cqe {
1206 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1207};
1208
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209struct ib_send_wr {
1210 struct ib_send_wr *next;
Christoph Hellwig14d3a3b2015-12-11 11:53:03 -08001211 union {
1212 u64 wr_id;
1213 struct ib_cqe *wr_cqe;
1214 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 struct ib_sge *sg_list;
1216 int num_sge;
1217 enum ib_wr_opcode opcode;
1218 int send_flags;
Roland Dreier0f39cf32008-04-16 21:09:32 -07001219 union {
1220 __be32 imm_data;
1221 u32 invalidate_rkey;
1222 } ex;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223};
1224
Christoph Hellwige622f2f2015-10-08 09:16:33 +01001225struct ib_rdma_wr {
1226 struct ib_send_wr wr;
1227 u64 remote_addr;
1228 u32 rkey;
1229};
1230
1231static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1232{
1233 return container_of(wr, struct ib_rdma_wr, wr);
1234}
1235
1236struct ib_atomic_wr {
1237 struct ib_send_wr wr;
1238 u64 remote_addr;
1239 u64 compare_add;
1240 u64 swap;
1241 u64 compare_add_mask;
1242 u64 swap_mask;
1243 u32 rkey;
1244};
1245
1246static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1247{
1248 return container_of(wr, struct ib_atomic_wr, wr);
1249}
1250
1251struct ib_ud_wr {
1252 struct ib_send_wr wr;
1253 struct ib_ah *ah;
1254 void *header;
1255 int hlen;
1256 int mss;
1257 u32 remote_qpn;
1258 u32 remote_qkey;
1259 u16 pkey_index; /* valid for GSI only */
1260 u8 port_num; /* valid for DR SMPs on switch only */
1261};
1262
1263static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1264{
1265 return container_of(wr, struct ib_ud_wr, wr);
1266}
1267
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +03001268struct ib_reg_wr {
1269 struct ib_send_wr wr;
1270 struct ib_mr *mr;
1271 u32 key;
1272 int access;
1273};
1274
1275static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1276{
1277 return container_of(wr, struct ib_reg_wr, wr);
1278}
1279
Christoph Hellwige622f2f2015-10-08 09:16:33 +01001280struct ib_sig_handover_wr {
1281 struct ib_send_wr wr;
1282 struct ib_sig_attrs *sig_attrs;
1283 struct ib_mr *sig_mr;
1284 int access_flags;
1285 struct ib_sge *prot;
1286};
1287
1288static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1289{
1290 return container_of(wr, struct ib_sig_handover_wr, wr);
1291}
1292
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293struct ib_recv_wr {
1294 struct ib_recv_wr *next;
Christoph Hellwig14d3a3b2015-12-11 11:53:03 -08001295 union {
1296 u64 wr_id;
1297 struct ib_cqe *wr_cqe;
1298 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 struct ib_sge *sg_list;
1300 int num_sge;
1301};
1302
1303enum ib_access_flags {
1304 IB_ACCESS_LOCAL_WRITE = 1,
1305 IB_ACCESS_REMOTE_WRITE = (1<<1),
1306 IB_ACCESS_REMOTE_READ = (1<<2),
1307 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
Shani Michaeli7083e422013-02-06 16:19:12 +00001308 IB_ACCESS_MW_BIND = (1<<4),
Sagi Grimberg860f10a2014-12-11 17:04:16 +02001309 IB_ZERO_BASED = (1<<5),
1310 IB_ACCESS_ON_DEMAND = (1<<6),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311};
1312
Christoph Hellwigb7d3e0a2015-12-23 19:12:47 +01001313/*
1314 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1315 * are hidden here instead of a uapi header!
1316 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317enum ib_mr_rereg_flags {
1318 IB_MR_REREG_TRANS = 1,
1319 IB_MR_REREG_PD = (1<<1),
Matan Barak7e6edb92014-07-31 11:01:28 +03001320 IB_MR_REREG_ACCESS = (1<<2),
1321 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322};
1323
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324struct ib_fmr_attr {
1325 int max_pages;
1326 int max_maps;
Or Gerlitzd36f34a2006-02-02 10:43:45 -08001327 u8 page_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328};
1329
Haggai Eran882214e2014-12-11 17:04:18 +02001330struct ib_umem;
1331
Roland Dreiere2773c02005-07-07 17:57:10 -07001332struct ib_ucontext {
1333 struct ib_device *device;
1334 struct list_head pd_list;
1335 struct list_head mr_list;
1336 struct list_head mw_list;
1337 struct list_head cq_list;
1338 struct list_head qp_list;
1339 struct list_head srq_list;
1340 struct list_head ah_list;
Sean Hefty53d0bd12011-05-24 08:33:46 -07001341 struct list_head xrcd_list;
Hadar Hen Zion436f2ad2013-08-14 13:58:30 +03001342 struct list_head rule_list;
Yishai Hadasf213c052016-05-23 15:20:49 +03001343 struct list_head wq_list;
Yishai Hadasde019a92016-05-23 15:20:52 +03001344 struct list_head rwq_ind_tbl_list;
Roland Dreierf7c6a7b2007-03-04 16:15:11 -08001345 int closing;
Shachar Raindel8ada2c12014-12-11 17:04:17 +02001346
1347 struct pid *tgid;
Haggai Eran882214e2014-12-11 17:04:18 +02001348#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1349 struct rb_root umem_tree;
1350 /*
1351 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1352 * mmu notifiers registration.
1353 */
1354 struct rw_semaphore umem_rwsem;
1355 void (*invalidate_range)(struct ib_umem *umem,
1356 unsigned long start, unsigned long end);
1357
1358 struct mmu_notifier mn;
1359 atomic_t notifier_count;
1360 /* A list of umems that don't have private mmu notifier counters yet. */
1361 struct list_head no_private_counters;
1362 int odp_mrs_count;
1363#endif
Roland Dreiere2773c02005-07-07 17:57:10 -07001364};
1365
1366struct ib_uobject {
1367 u64 user_handle; /* handle given to us by userspace */
1368 struct ib_ucontext *context; /* associated user context */
Roland Dreier9ead1902006-06-17 20:44:49 -07001369 void *object; /* containing object */
Roland Dreiere2773c02005-07-07 17:57:10 -07001370 struct list_head list; /* link to context's list */
Roland Dreierb3d636b2008-04-16 21:01:06 -07001371 int id; /* index into kernel idr */
Roland Dreier9ead1902006-06-17 20:44:49 -07001372 struct kref ref;
1373 struct rw_semaphore mutex; /* protects .live */
Mike Marciniszynd144da82015-11-02 12:13:25 -05001374 struct rcu_head rcu; /* kfree_rcu() overhead */
Roland Dreier9ead1902006-06-17 20:44:49 -07001375 int live;
Roland Dreiere2773c02005-07-07 17:57:10 -07001376};
1377
Roland Dreiere2773c02005-07-07 17:57:10 -07001378struct ib_udata {
Yann Droneaud309243e2013-12-11 23:01:44 +01001379 const void __user *inbuf;
Roland Dreiere2773c02005-07-07 17:57:10 -07001380 void __user *outbuf;
1381 size_t inlen;
1382 size_t outlen;
1383};
1384
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385struct ib_pd {
Jason Gunthorpe96249d72015-08-05 14:14:45 -06001386 u32 local_dma_lkey;
Christoph Hellwiged082d32016-09-05 12:56:17 +02001387 u32 flags;
Roland Dreiere2773c02005-07-07 17:57:10 -07001388 struct ib_device *device;
1389 struct ib_uobject *uobject;
1390 atomic_t usecnt; /* count all resources */
Christoph Hellwig50d46332016-09-05 12:56:16 +02001391
Christoph Hellwiged082d32016-09-05 12:56:17 +02001392 u32 unsafe_global_rkey;
1393
Christoph Hellwig50d46332016-09-05 12:56:16 +02001394 /*
1395 * Implementation details of the RDMA core, don't use in drivers:
1396 */
1397 struct ib_mr *__internal_mr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398};
1399
Sean Hefty59991f92011-05-23 17:52:46 -07001400struct ib_xrcd {
1401 struct ib_device *device;
Sean Heftyd3d72d92011-05-26 23:06:44 -07001402 atomic_t usecnt; /* count all exposed resources */
Sean Hefty53d0bd12011-05-24 08:33:46 -07001403 struct inode *inode;
Sean Heftyd3d72d92011-05-26 23:06:44 -07001404
1405 struct mutex tgt_qp_mutex;
1406 struct list_head tgt_qp_list;
Sean Hefty59991f92011-05-23 17:52:46 -07001407};
1408
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409struct ib_ah {
1410 struct ib_device *device;
1411 struct ib_pd *pd;
Roland Dreiere2773c02005-07-07 17:57:10 -07001412 struct ib_uobject *uobject;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413};
1414
1415typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1416
Christoph Hellwig14d3a3b2015-12-11 11:53:03 -08001417enum ib_poll_context {
1418 IB_POLL_DIRECT, /* caller context, no hw completions */
1419 IB_POLL_SOFTIRQ, /* poll from softirq context */
1420 IB_POLL_WORKQUEUE, /* poll from workqueue */
1421};
1422
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423struct ib_cq {
Roland Dreiere2773c02005-07-07 17:57:10 -07001424 struct ib_device *device;
1425 struct ib_uobject *uobject;
1426 ib_comp_handler comp_handler;
1427 void (*event_handler)(struct ib_event *, void *);
Dotan Barak4deccd62008-07-14 23:48:44 -07001428 void *cq_context;
Roland Dreiere2773c02005-07-07 17:57:10 -07001429 int cqe;
1430 atomic_t usecnt; /* count number of work queues */
Christoph Hellwig14d3a3b2015-12-11 11:53:03 -08001431 enum ib_poll_context poll_ctx;
1432 struct ib_wc *wc;
1433 union {
1434 struct irq_poll iop;
1435 struct work_struct work;
1436 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437};
1438
1439struct ib_srq {
Roland Dreierd41fcc62005-08-18 12:23:08 -07001440 struct ib_device *device;
1441 struct ib_pd *pd;
1442 struct ib_uobject *uobject;
1443 void (*event_handler)(struct ib_event *, void *);
1444 void *srq_context;
Sean Hefty96104ed2011-05-23 16:31:36 -07001445 enum ib_srq_type srq_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 atomic_t usecnt;
Sean Hefty418d5132011-05-23 19:42:29 -07001447
1448 union {
1449 struct {
1450 struct ib_xrcd *xrcd;
1451 struct ib_cq *cq;
1452 u32 srq_num;
1453 } xrc;
1454 } ext;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455};
1456
Yishai Hadas5fd251c2016-05-23 15:20:48 +03001457enum ib_wq_type {
1458 IB_WQT_RQ
1459};
1460
1461enum ib_wq_state {
1462 IB_WQS_RESET,
1463 IB_WQS_RDY,
1464 IB_WQS_ERR
1465};
1466
1467struct ib_wq {
1468 struct ib_device *device;
1469 struct ib_uobject *uobject;
1470 void *wq_context;
1471 void (*event_handler)(struct ib_event *, void *);
1472 struct ib_pd *pd;
1473 struct ib_cq *cq;
1474 u32 wq_num;
1475 enum ib_wq_state state;
1476 enum ib_wq_type wq_type;
1477 atomic_t usecnt;
1478};
1479
1480struct ib_wq_init_attr {
1481 void *wq_context;
1482 enum ib_wq_type wq_type;
1483 u32 max_wr;
1484 u32 max_sge;
1485 struct ib_cq *cq;
1486 void (*event_handler)(struct ib_event *, void *);
1487};
1488
1489enum ib_wq_attr_mask {
1490 IB_WQ_STATE = 1 << 0,
1491 IB_WQ_CUR_STATE = 1 << 1,
1492};
1493
1494struct ib_wq_attr {
1495 enum ib_wq_state wq_state;
1496 enum ib_wq_state curr_wq_state;
1497};
1498
Yishai Hadas6d397862016-05-23 15:20:51 +03001499struct ib_rwq_ind_table {
1500 struct ib_device *device;
1501 struct ib_uobject *uobject;
1502 atomic_t usecnt;
1503 u32 ind_tbl_num;
1504 u32 log_ind_tbl_size;
1505 struct ib_wq **ind_tbl;
1506};
1507
1508struct ib_rwq_ind_table_init_attr {
1509 u32 log_ind_tbl_size;
1510 /* Each entry is a pointer to Receive Work Queue */
1511 struct ib_wq **ind_tbl;
1512};
1513
Bart Van Assche632bc3f2016-07-21 13:03:30 -07001514/*
1515 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1516 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1517 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518struct ib_qp {
1519 struct ib_device *device;
1520 struct ib_pd *pd;
1521 struct ib_cq *send_cq;
1522 struct ib_cq *recv_cq;
Christoph Hellwigfffb0382016-05-03 18:01:07 +02001523 spinlock_t mr_lock;
1524 int mrs_used;
Christoph Hellwiga060b562016-05-03 18:01:09 +02001525 struct list_head rdma_mrs;
Christoph Hellwig0e353e32016-05-03 18:01:12 +02001526 struct list_head sig_mrs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 struct ib_srq *srq;
Sean Heftyb42b63c2011-05-23 19:59:25 -07001528 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
Sean Heftyd3d72d92011-05-26 23:06:44 -07001529 struct list_head xrcd_list;
Christoph Hellwigfffb0382016-05-03 18:01:07 +02001530
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001531 /* count times opened, mcast attaches, flow attaches */
1532 atomic_t usecnt;
Sean Hefty0e0ec7e2011-08-08 15:31:51 -07001533 struct list_head open_list;
1534 struct ib_qp *real_qp;
Roland Dreiere2773c02005-07-07 17:57:10 -07001535 struct ib_uobject *uobject;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 void (*event_handler)(struct ib_event *, void *);
1537 void *qp_context;
1538 u32 qp_num;
Bart Van Assche632bc3f2016-07-21 13:03:30 -07001539 u32 max_write_sge;
1540 u32 max_read_sge;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 enum ib_qp_type qp_type;
Yishai Hadasa9017e22016-05-23 15:20:54 +03001542 struct ib_rwq_ind_table *rwq_ind_tbl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543};
1544
1545struct ib_mr {
Roland Dreiere2773c02005-07-07 17:57:10 -07001546 struct ib_device *device;
1547 struct ib_pd *pd;
Roland Dreiere2773c02005-07-07 17:57:10 -07001548 u32 lkey;
1549 u32 rkey;
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +03001550 u64 iova;
1551 u32 length;
1552 unsigned int page_size;
Steve Wised4a85c32016-05-03 18:01:08 +02001553 bool need_inval;
Christoph Hellwigfffb0382016-05-03 18:01:07 +02001554 union {
1555 struct ib_uobject *uobject; /* user */
1556 struct list_head qp_entry; /* FR */
1557 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558};
1559
1560struct ib_mw {
1561 struct ib_device *device;
1562 struct ib_pd *pd;
Roland Dreiere2773c02005-07-07 17:57:10 -07001563 struct ib_uobject *uobject;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 u32 rkey;
Shani Michaeli7083e422013-02-06 16:19:12 +00001565 enum ib_mw_type type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566};
1567
1568struct ib_fmr {
1569 struct ib_device *device;
1570 struct ib_pd *pd;
1571 struct list_head list;
1572 u32 lkey;
1573 u32 rkey;
1574};
1575
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001576/* Supported steering options */
1577enum ib_flow_attr_type {
1578 /* steering according to rule specifications */
1579 IB_FLOW_ATTR_NORMAL = 0x0,
1580 /* default unicast and multicast rule -
1581 * receive all Eth traffic which isn't steered to any QP
1582 */
1583 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1584 /* default multicast rule -
1585 * receive all Eth multicast traffic which isn't steered to any QP
1586 */
1587 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1588 /* sniffer rule - receive all port traffic */
1589 IB_FLOW_ATTR_SNIFFER = 0x3
1590};
1591
1592/* Supported steering header types */
1593enum ib_flow_spec_type {
1594 /* L2 headers*/
1595 IB_FLOW_SPEC_ETH = 0x20,
Matan Barak240ae002013-11-07 15:25:13 +02001596 IB_FLOW_SPEC_IB = 0x22,
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001597 /* L3 header*/
1598 IB_FLOW_SPEC_IPV4 = 0x30,
Maor Gottlieb4c2aae72016-06-17 15:14:50 +03001599 IB_FLOW_SPEC_IPV6 = 0x31,
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001600 /* L4 headers*/
1601 IB_FLOW_SPEC_TCP = 0x40,
1602 IB_FLOW_SPEC_UDP = 0x41
1603};
Matan Barak240ae002013-11-07 15:25:13 +02001604#define IB_FLOW_SPEC_LAYER_MASK 0xF0
Matan Barak22878db2013-09-01 18:39:52 +03001605#define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1606
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001607/* Flow steering rule priority is set according to it's domain.
1608 * Lower domain value means higher priority.
1609 */
1610enum ib_flow_domain {
1611 IB_FLOW_DOMAIN_USER,
1612 IB_FLOW_DOMAIN_ETHTOOL,
1613 IB_FLOW_DOMAIN_RFS,
1614 IB_FLOW_DOMAIN_NIC,
1615 IB_FLOW_DOMAIN_NUM /* Must be last */
1616};
1617
Marina Varshavera3100a72016-02-18 18:31:05 +02001618enum ib_flow_flags {
1619 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1620 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 2 /* Must be last */
1621};
1622
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001623struct ib_flow_eth_filter {
1624 u8 dst_mac[6];
1625 u8 src_mac[6];
1626 __be16 ether_type;
1627 __be16 vlan_tag;
Maor Gottlieb15dfbd62016-08-30 16:58:32 +03001628 /* Must be last */
1629 u8 real_sz[0];
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001630};
1631
1632struct ib_flow_spec_eth {
1633 enum ib_flow_spec_type type;
1634 u16 size;
1635 struct ib_flow_eth_filter val;
1636 struct ib_flow_eth_filter mask;
1637};
1638
Matan Barak240ae002013-11-07 15:25:13 +02001639struct ib_flow_ib_filter {
1640 __be16 dlid;
1641 __u8 sl;
Maor Gottlieb15dfbd62016-08-30 16:58:32 +03001642 /* Must be last */
1643 u8 real_sz[0];
Matan Barak240ae002013-11-07 15:25:13 +02001644};
1645
1646struct ib_flow_spec_ib {
1647 enum ib_flow_spec_type type;
1648 u16 size;
1649 struct ib_flow_ib_filter val;
1650 struct ib_flow_ib_filter mask;
1651};
1652
Maor Gottlieb989a3a82016-08-30 16:58:33 +03001653/* IPv4 header flags */
1654enum ib_ipv4_flags {
1655 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1656 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1657 last have this flag set */
1658};
1659
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001660struct ib_flow_ipv4_filter {
1661 __be32 src_ip;
1662 __be32 dst_ip;
Maor Gottlieb989a3a82016-08-30 16:58:33 +03001663 u8 proto;
1664 u8 tos;
1665 u8 ttl;
1666 u8 flags;
Maor Gottlieb15dfbd62016-08-30 16:58:32 +03001667 /* Must be last */
1668 u8 real_sz[0];
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001669};
1670
1671struct ib_flow_spec_ipv4 {
1672 enum ib_flow_spec_type type;
1673 u16 size;
1674 struct ib_flow_ipv4_filter val;
1675 struct ib_flow_ipv4_filter mask;
1676};
1677
Maor Gottlieb4c2aae72016-06-17 15:14:50 +03001678struct ib_flow_ipv6_filter {
1679 u8 src_ip[16];
1680 u8 dst_ip[16];
Maor Gottlieba72c6a22016-08-30 16:58:34 +03001681 __be32 flow_label;
1682 u8 next_hdr;
1683 u8 traffic_class;
1684 u8 hop_limit;
Maor Gottlieb15dfbd62016-08-30 16:58:32 +03001685 /* Must be last */
1686 u8 real_sz[0];
Maor Gottlieb4c2aae72016-06-17 15:14:50 +03001687};
1688
1689struct ib_flow_spec_ipv6 {
1690 enum ib_flow_spec_type type;
1691 u16 size;
1692 struct ib_flow_ipv6_filter val;
1693 struct ib_flow_ipv6_filter mask;
1694};
1695
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001696struct ib_flow_tcp_udp_filter {
1697 __be16 dst_port;
1698 __be16 src_port;
Maor Gottlieb15dfbd62016-08-30 16:58:32 +03001699 /* Must be last */
1700 u8 real_sz[0];
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001701};
1702
1703struct ib_flow_spec_tcp_udp {
1704 enum ib_flow_spec_type type;
1705 u16 size;
1706 struct ib_flow_tcp_udp_filter val;
1707 struct ib_flow_tcp_udp_filter mask;
1708};
1709
1710union ib_flow_spec {
1711 struct {
1712 enum ib_flow_spec_type type;
1713 u16 size;
1714 };
1715 struct ib_flow_spec_eth eth;
Matan Barak240ae002013-11-07 15:25:13 +02001716 struct ib_flow_spec_ib ib;
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001717 struct ib_flow_spec_ipv4 ipv4;
1718 struct ib_flow_spec_tcp_udp tcp_udp;
Maor Gottlieb4c2aae72016-06-17 15:14:50 +03001719 struct ib_flow_spec_ipv6 ipv6;
Hadar Hen Zion319a4412013-08-07 14:01:59 +03001720};
1721
1722struct ib_flow_attr {
1723 enum ib_flow_attr_type type;
1724 u16 size;
1725 u16 priority;
1726 u32 flags;
1727 u8 num_of_specs;
1728 u8 port;
1729 /* Following are the optional layers according to user request
1730 * struct ib_flow_spec_xxx
1731 * struct ib_flow_spec_yyy
1732 */
1733};
1734
1735struct ib_flow {
1736 struct ib_qp *qp;
1737 struct ib_uobject *uobject;
1738};
1739
Ira Weiny4cd7c942015-06-06 14:38:31 -04001740struct ib_mad_hdr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741struct ib_grh;
1742
1743enum ib_process_mad_flags {
1744 IB_MAD_IGNORE_MKEY = 1,
1745 IB_MAD_IGNORE_BKEY = 2,
1746 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1747};
1748
1749enum ib_mad_result {
1750 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1751 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1752 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1753 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1754};
1755
1756#define IB_DEVICE_NAME_MAX 64
1757
1758struct ib_cache {
1759 rwlock_t lock;
1760 struct ib_event_handler event_handler;
1761 struct ib_pkey_cache **pkey_cache;
Matan Barak03db3a22015-07-30 18:33:26 +03001762 struct ib_gid_table **gid_cache;
Jack Morgenstein6fb9cdb2006-06-17 20:37:34 -07001763 u8 *lmc_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764};
1765
Ralph Campbell9b513092006-12-12 14:27:41 -08001766struct ib_dma_mapping_ops {
1767 int (*mapping_error)(struct ib_device *dev,
1768 u64 dma_addr);
1769 u64 (*map_single)(struct ib_device *dev,
1770 void *ptr, size_t size,
1771 enum dma_data_direction direction);
1772 void (*unmap_single)(struct ib_device *dev,
1773 u64 addr, size_t size,
1774 enum dma_data_direction direction);
1775 u64 (*map_page)(struct ib_device *dev,
1776 struct page *page, unsigned long offset,
1777 size_t size,
1778 enum dma_data_direction direction);
1779 void (*unmap_page)(struct ib_device *dev,
1780 u64 addr, size_t size,
1781 enum dma_data_direction direction);
1782 int (*map_sg)(struct ib_device *dev,
1783 struct scatterlist *sg, int nents,
1784 enum dma_data_direction direction);
1785 void (*unmap_sg)(struct ib_device *dev,
1786 struct scatterlist *sg, int nents,
1787 enum dma_data_direction direction);
Parav Panditd9703652016-09-28 20:25:47 +00001788 int (*map_sg_attrs)(struct ib_device *dev,
1789 struct scatterlist *sg, int nents,
1790 enum dma_data_direction direction,
1791 unsigned long attrs);
1792 void (*unmap_sg_attrs)(struct ib_device *dev,
1793 struct scatterlist *sg, int nents,
1794 enum dma_data_direction direction,
1795 unsigned long attrs);
Ralph Campbell9b513092006-12-12 14:27:41 -08001796 void (*sync_single_for_cpu)(struct ib_device *dev,
1797 u64 dma_handle,
1798 size_t size,
Dotan Barak4deccd62008-07-14 23:48:44 -07001799 enum dma_data_direction dir);
Ralph Campbell9b513092006-12-12 14:27:41 -08001800 void (*sync_single_for_device)(struct ib_device *dev,
1801 u64 dma_handle,
1802 size_t size,
1803 enum dma_data_direction dir);
1804 void *(*alloc_coherent)(struct ib_device *dev,
1805 size_t size,
1806 u64 *dma_handle,
1807 gfp_t flag);
1808 void (*free_coherent)(struct ib_device *dev,
1809 size_t size, void *cpu_addr,
1810 u64 dma_handle);
1811};
1812
Tom Tucker07ebafb2006-08-03 16:02:42 -05001813struct iw_cm_verbs;
1814
Ira Weiny77386132015-05-13 20:02:58 -04001815struct ib_port_immutable {
1816 int pkey_tbl_len;
1817 int gid_tbl_len;
Ira Weinyf9b22e32015-05-13 20:02:59 -04001818 u32 core_cap_flags;
Ira Weiny337877a2015-06-06 14:38:29 -04001819 u32 max_mad_size;
Ira Weiny77386132015-05-13 20:02:58 -04001820};
1821
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822struct ib_device {
1823 struct device *dma_device;
1824
1825 char name[IB_DEVICE_NAME_MAX];
1826
1827 struct list_head event_handler_list;
1828 spinlock_t event_handler_lock;
1829
Alexander Chiang17a55f72010-02-02 19:09:16 +00001830 spinlock_t client_data_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 struct list_head core_list;
Haggai Eran7c1eb452015-07-30 17:50:14 +03001832 /* Access to the client_data_list is protected by the client_data_lock
1833 * spinlock and the lists_rwsem read-write semaphore */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 struct list_head client_data_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835
1836 struct ib_cache cache;
Ira Weiny77386132015-05-13 20:02:58 -04001837 /**
1838 * port_immutable is indexed by port number
1839 */
1840 struct ib_port_immutable *port_immutable;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841
Michael S. Tsirkinf4fd0b22007-05-03 13:48:47 +03001842 int num_comp_vectors;
1843
Tom Tucker07ebafb2006-08-03 16:02:42 -05001844 struct iw_cm_verbs *iwcm;
1845
Christoph Lameterb40f4752016-05-16 12:49:33 -05001846 /**
1847 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1848 * driver initialized data. The struct is kfree()'ed by the sysfs
1849 * core when the device is removed. A lifespan of -1 in the return
1850 * struct tells the core to set a default lifespan.
1851 */
1852 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
1853 u8 port_num);
1854 /**
1855 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1856 * @index - The index in the value array we wish to have updated, or
1857 * num_counters if we want all stats updated
1858 * Return codes -
1859 * < 0 - Error, no counters updated
1860 * index - Updated the single counter pointed to by index
1861 * num_counters - Updated all counters (will reset the timestamp
1862 * and prevent further calls for lifespan milliseconds)
1863 * Drivers are allowed to update all counters in leiu of just the
1864 * one given in index at their option
1865 */
1866 int (*get_hw_stats)(struct ib_device *device,
1867 struct rdma_hw_stats *stats,
1868 u8 port, int index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 int (*query_device)(struct ib_device *device,
Matan Barak2528e332015-06-11 16:35:25 +03001870 struct ib_device_attr *device_attr,
1871 struct ib_udata *udata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 int (*query_port)(struct ib_device *device,
1873 u8 port_num,
1874 struct ib_port_attr *port_attr);
Eli Cohena3f5ada2010-09-27 17:51:10 -07001875 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1876 u8 port_num);
Matan Barak03db3a22015-07-30 18:33:26 +03001877 /* When calling get_netdev, the HW vendor's driver should return the
1878 * net device of device @device at port @port_num or NULL if such
1879 * a net device doesn't exist. The vendor driver should call dev_hold
1880 * on this net device. The HW vendor's device driver must guarantee
1881 * that this function returns NULL before the net device reaches
1882 * NETDEV_UNREGISTER_FINAL state.
1883 */
1884 struct net_device *(*get_netdev)(struct ib_device *device,
1885 u8 port_num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 int (*query_gid)(struct ib_device *device,
1887 u8 port_num, int index,
1888 union ib_gid *gid);
Matan Barak03db3a22015-07-30 18:33:26 +03001889 /* When calling add_gid, the HW vendor's driver should
1890 * add the gid of device @device at gid index @index of
1891 * port @port_num to be @gid. Meta-info of that gid (for example,
1892 * the network device related to this gid is available
1893 * at @attr. @context allows the HW vendor driver to store extra
1894 * information together with a GID entry. The HW vendor may allocate
1895 * memory to contain this information and store it in @context when a
1896 * new GID entry is written to. Params are consistent until the next
1897 * call of add_gid or delete_gid. The function should return 0 on
1898 * success or error otherwise. The function could be called
1899 * concurrently for different ports. This function is only called
1900 * when roce_gid_table is used.
1901 */
1902 int (*add_gid)(struct ib_device *device,
1903 u8 port_num,
1904 unsigned int index,
1905 const union ib_gid *gid,
1906 const struct ib_gid_attr *attr,
1907 void **context);
1908 /* When calling del_gid, the HW vendor's driver should delete the
1909 * gid of device @device at gid index @index of port @port_num.
1910 * Upon the deletion of a GID entry, the HW vendor must free any
1911 * allocated memory. The caller will clear @context afterwards.
1912 * This function is only called when roce_gid_table is used.
1913 */
1914 int (*del_gid)(struct ib_device *device,
1915 u8 port_num,
1916 unsigned int index,
1917 void **context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 int (*query_pkey)(struct ib_device *device,
1919 u8 port_num, u16 index, u16 *pkey);
1920 int (*modify_device)(struct ib_device *device,
1921 int device_modify_mask,
1922 struct ib_device_modify *device_modify);
1923 int (*modify_port)(struct ib_device *device,
1924 u8 port_num, int port_modify_mask,
1925 struct ib_port_modify *port_modify);
Roland Dreiere2773c02005-07-07 17:57:10 -07001926 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
1927 struct ib_udata *udata);
1928 int (*dealloc_ucontext)(struct ib_ucontext *context);
1929 int (*mmap)(struct ib_ucontext *context,
1930 struct vm_area_struct *vma);
1931 struct ib_pd * (*alloc_pd)(struct ib_device *device,
1932 struct ib_ucontext *context,
1933 struct ib_udata *udata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 int (*dealloc_pd)(struct ib_pd *pd);
1935 struct ib_ah * (*create_ah)(struct ib_pd *pd,
1936 struct ib_ah_attr *ah_attr);
1937 int (*modify_ah)(struct ib_ah *ah,
1938 struct ib_ah_attr *ah_attr);
1939 int (*query_ah)(struct ib_ah *ah,
1940 struct ib_ah_attr *ah_attr);
1941 int (*destroy_ah)(struct ib_ah *ah);
Roland Dreierd41fcc62005-08-18 12:23:08 -07001942 struct ib_srq * (*create_srq)(struct ib_pd *pd,
1943 struct ib_srq_init_attr *srq_init_attr,
1944 struct ib_udata *udata);
1945 int (*modify_srq)(struct ib_srq *srq,
1946 struct ib_srq_attr *srq_attr,
Ralph Campbell9bc57e22006-08-11 14:58:09 -07001947 enum ib_srq_attr_mask srq_attr_mask,
1948 struct ib_udata *udata);
Roland Dreierd41fcc62005-08-18 12:23:08 -07001949 int (*query_srq)(struct ib_srq *srq,
1950 struct ib_srq_attr *srq_attr);
1951 int (*destroy_srq)(struct ib_srq *srq);
1952 int (*post_srq_recv)(struct ib_srq *srq,
1953 struct ib_recv_wr *recv_wr,
1954 struct ib_recv_wr **bad_recv_wr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 struct ib_qp * (*create_qp)(struct ib_pd *pd,
Roland Dreiere2773c02005-07-07 17:57:10 -07001956 struct ib_qp_init_attr *qp_init_attr,
1957 struct ib_udata *udata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 int (*modify_qp)(struct ib_qp *qp,
1959 struct ib_qp_attr *qp_attr,
Ralph Campbell9bc57e22006-08-11 14:58:09 -07001960 int qp_attr_mask,
1961 struct ib_udata *udata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 int (*query_qp)(struct ib_qp *qp,
1963 struct ib_qp_attr *qp_attr,
1964 int qp_attr_mask,
1965 struct ib_qp_init_attr *qp_init_attr);
1966 int (*destroy_qp)(struct ib_qp *qp);
1967 int (*post_send)(struct ib_qp *qp,
1968 struct ib_send_wr *send_wr,
1969 struct ib_send_wr **bad_send_wr);
1970 int (*post_recv)(struct ib_qp *qp,
1971 struct ib_recv_wr *recv_wr,
1972 struct ib_recv_wr **bad_recv_wr);
Matan Barakbcf4c1e2015-06-11 16:35:20 +03001973 struct ib_cq * (*create_cq)(struct ib_device *device,
1974 const struct ib_cq_init_attr *attr,
Roland Dreiere2773c02005-07-07 17:57:10 -07001975 struct ib_ucontext *context,
1976 struct ib_udata *udata);
Eli Cohen2dd57162008-04-16 21:09:33 -07001977 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1978 u16 cq_period);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 int (*destroy_cq)(struct ib_cq *cq);
Roland Dreier33b9b3e2006-01-30 14:29:21 -08001980 int (*resize_cq)(struct ib_cq *cq, int cqe,
1981 struct ib_udata *udata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 int (*poll_cq)(struct ib_cq *cq, int num_entries,
1983 struct ib_wc *wc);
1984 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1985 int (*req_notify_cq)(struct ib_cq *cq,
Roland Dreiered23a722007-05-06 21:02:48 -07001986 enum ib_cq_notify_flags flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 int (*req_ncomp_notif)(struct ib_cq *cq,
1988 int wc_cnt);
1989 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
1990 int mr_access_flags);
Roland Dreiere2773c02005-07-07 17:57:10 -07001991 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
Roland Dreierf7c6a7b2007-03-04 16:15:11 -08001992 u64 start, u64 length,
1993 u64 virt_addr,
Roland Dreiere2773c02005-07-07 17:57:10 -07001994 int mr_access_flags,
1995 struct ib_udata *udata);
Matan Barak7e6edb92014-07-31 11:01:28 +03001996 int (*rereg_user_mr)(struct ib_mr *mr,
1997 int flags,
1998 u64 start, u64 length,
1999 u64 virt_addr,
2000 int mr_access_flags,
2001 struct ib_pd *pd,
2002 struct ib_udata *udata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 int (*dereg_mr)(struct ib_mr *mr);
Sagi Grimberg9bee1782015-07-30 10:32:35 +03002004 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
2005 enum ib_mr_type mr_type,
2006 u32 max_num_sg);
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +03002007 int (*map_mr_sg)(struct ib_mr *mr,
2008 struct scatterlist *sg,
Christoph Hellwigff2ba992016-05-03 18:01:04 +02002009 int sg_nents,
Bart Van Assche9aa8b322016-05-12 10:49:15 -07002010 unsigned int *sg_offset);
Shani Michaeli7083e422013-02-06 16:19:12 +00002011 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
Matan Barakb2a239d2016-02-29 18:05:29 +02002012 enum ib_mw_type type,
2013 struct ib_udata *udata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 int (*dealloc_mw)(struct ib_mw *mw);
2015 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
2016 int mr_access_flags,
2017 struct ib_fmr_attr *fmr_attr);
2018 int (*map_phys_fmr)(struct ib_fmr *fmr,
2019 u64 *page_list, int list_len,
2020 u64 iova);
2021 int (*unmap_fmr)(struct list_head *fmr_list);
2022 int (*dealloc_fmr)(struct ib_fmr *fmr);
2023 int (*attach_mcast)(struct ib_qp *qp,
2024 union ib_gid *gid,
2025 u16 lid);
2026 int (*detach_mcast)(struct ib_qp *qp,
2027 union ib_gid *gid,
2028 u16 lid);
2029 int (*process_mad)(struct ib_device *device,
2030 int process_mad_flags,
2031 u8 port_num,
Ira Weinya97e2d82015-05-31 17:15:30 -04002032 const struct ib_wc *in_wc,
2033 const struct ib_grh *in_grh,
Ira Weiny4cd7c942015-06-06 14:38:31 -04002034 const struct ib_mad_hdr *in_mad,
2035 size_t in_mad_size,
2036 struct ib_mad_hdr *out_mad,
2037 size_t *out_mad_size,
2038 u16 *out_mad_pkey_index);
Sean Hefty59991f92011-05-23 17:52:46 -07002039 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
2040 struct ib_ucontext *ucontext,
2041 struct ib_udata *udata);
2042 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
Hadar Hen Zion319a4412013-08-07 14:01:59 +03002043 struct ib_flow * (*create_flow)(struct ib_qp *qp,
2044 struct ib_flow_attr
2045 *flow_attr,
2046 int domain);
2047 int (*destroy_flow)(struct ib_flow *flow_id);
Sagi Grimberg1b01d332014-02-23 14:19:05 +02002048 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2049 struct ib_mr_status *mr_status);
Yishai Hadas036b1062015-08-13 18:32:05 +03002050 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
Steve Wise765d6772016-02-17 08:15:41 -08002051 void (*drain_rq)(struct ib_qp *qp);
2052 void (*drain_sq)(struct ib_qp *qp);
Eli Cohen50174a72016-03-11 22:58:38 +02002053 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2054 int state);
2055 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2056 struct ifla_vf_info *ivf);
2057 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2058 struct ifla_vf_stats *stats);
2059 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2060 int type);
Yishai Hadas5fd251c2016-05-23 15:20:48 +03002061 struct ib_wq * (*create_wq)(struct ib_pd *pd,
2062 struct ib_wq_init_attr *init_attr,
2063 struct ib_udata *udata);
2064 int (*destroy_wq)(struct ib_wq *wq);
2065 int (*modify_wq)(struct ib_wq *wq,
2066 struct ib_wq_attr *attr,
2067 u32 wq_attr_mask,
2068 struct ib_udata *udata);
Yishai Hadas6d397862016-05-23 15:20:51 +03002069 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device,
2070 struct ib_rwq_ind_table_init_attr *init_attr,
2071 struct ib_udata *udata);
2072 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
Ralph Campbell9b513092006-12-12 14:27:41 -08002073 struct ib_dma_mapping_ops *dma_ops;
2074
Roland Dreiere2773c02005-07-07 17:57:10 -07002075 struct module *owner;
Tony Jonesf4e91eb2008-02-22 00:13:36 +01002076 struct device dev;
Greg Kroah-Hartman35be0682007-12-17 15:54:39 -04002077 struct kobject *ports_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 struct list_head port_list;
2079
2080 enum {
2081 IB_DEV_UNINITIALIZED,
2082 IB_DEV_REGISTERED,
2083 IB_DEV_UNREGISTERED
2084 } reg_state;
2085
Roland Dreier274c0892005-09-29 14:17:48 -07002086 int uverbs_abi_ver;
Alexander Chiang17a55f72010-02-02 19:09:16 +00002087 u64 uverbs_cmd_mask;
Yann Droneaudf21519b2013-11-06 23:21:49 +01002088 u64 uverbs_ex_cmd_mask;
Roland Dreier274c0892005-09-29 14:17:48 -07002089
Yuval Shaiabd99fde2016-08-25 10:57:07 -07002090 char node_desc[IB_DEVICE_NODE_DESC_MAX];
Sean Heftycf311cd2006-01-10 07:39:34 -08002091 __be64 node_guid;
Steve Wise96f15c02008-07-14 23:48:53 -07002092 u32 local_dma_lkey;
Hal Rosenstock41390322015-06-29 09:57:00 -04002093 u16 is_switch:1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 u8 node_type;
2095 u8 phys_port_cnt;
Ira Weiny3e153a92015-12-18 10:59:44 +02002096 struct ib_device_attr attrs;
Christoph Lameterb40f4752016-05-16 12:49:33 -05002097 struct attribute_group *hw_stats_ag;
2098 struct rdma_hw_stats *hw_stats;
Ira Weiny77386132015-05-13 20:02:58 -04002099
2100 /**
2101 * The following mandatory functions are used only at device
2102 * registration. Keep functions such as these at the end of this
2103 * structure to avoid cache line misses when accessing struct ib_device
2104 * in fast paths.
2105 */
2106 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
Ira Weiny5fa76c22016-06-15 02:21:56 -04002107 void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108};
2109
2110struct ib_client {
2111 char *name;
2112 void (*add) (struct ib_device *);
Haggai Eran7c1eb452015-07-30 17:50:14 +03002113 void (*remove)(struct ib_device *, void *client_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114
Yotam Kenneth9268f722015-07-30 17:50:15 +03002115 /* Returns the net_dev belonging to this ib_client and matching the
2116 * given parameters.
2117 * @dev: An RDMA device that the net_dev use for communication.
2118 * @port: A physical port number on the RDMA device.
2119 * @pkey: P_Key that the net_dev uses if applicable.
2120 * @gid: A GID that the net_dev uses to communicate.
2121 * @addr: An IP address the net_dev is configured with.
2122 * @client_data: The device's client data set by ib_set_client_data().
2123 *
2124 * An ib_client that implements a net_dev on top of RDMA devices
2125 * (such as IP over IB) should implement this callback, allowing the
2126 * rdma_cm module to find the right net_dev for a given request.
2127 *
2128 * The caller is responsible for calling dev_put on the returned
2129 * netdev. */
2130 struct net_device *(*get_net_dev_by_params)(
2131 struct ib_device *dev,
2132 u8 port,
2133 u16 pkey,
2134 const union ib_gid *gid,
2135 const struct sockaddr *addr,
2136 void *client_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 struct list_head list;
2138};
2139
2140struct ib_device *ib_alloc_device(size_t size);
2141void ib_dealloc_device(struct ib_device *device);
2142
Ira Weiny5fa76c22016-06-15 02:21:56 -04002143void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len);
2144
Ralph Campbell9a6edb62010-05-06 17:03:25 -07002145int ib_register_device(struct ib_device *device,
2146 int (*port_callback)(struct ib_device *,
2147 u8, struct kobject *));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148void ib_unregister_device(struct ib_device *device);
2149
2150int ib_register_client (struct ib_client *client);
2151void ib_unregister_client(struct ib_client *client);
2152
2153void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2154void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2155 void *data);
2156
Roland Dreiere2773c02005-07-07 17:57:10 -07002157static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2158{
2159 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2160}
2161
2162static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2163{
Yann Droneaud43c611652015-02-05 22:10:18 +01002164 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
Roland Dreiere2773c02005-07-07 17:57:10 -07002165}
2166
Matan Barak301a7212015-12-15 20:30:10 +02002167static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2168 size_t offset,
2169 size_t len)
2170{
2171 const void __user *p = udata->inbuf + offset;
Markus Elfring92d27ae2016-08-22 18:23:24 +02002172 bool ret;
Matan Barak301a7212015-12-15 20:30:10 +02002173 u8 *buf;
2174
2175 if (len > USHRT_MAX)
2176 return false;
2177
Markus Elfring92d27ae2016-08-22 18:23:24 +02002178 buf = memdup_user(p, len);
2179 if (IS_ERR(buf))
Matan Barak301a7212015-12-15 20:30:10 +02002180 return false;
2181
Matan Barak301a7212015-12-15 20:30:10 +02002182 ret = !memchr_inv(buf, 0, len);
Matan Barak301a7212015-12-15 20:30:10 +02002183 kfree(buf);
2184 return ret;
2185}
2186
Roland Dreier8a518662006-02-13 12:48:12 -08002187/**
2188 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2189 * contains all required attributes and no attributes not allowed for
2190 * the given QP state transition.
2191 * @cur_state: Current QP state
2192 * @next_state: Next QP state
2193 * @type: QP type
2194 * @mask: Mask of supplied QP attributes
Matan Barakdd5f03b2013-12-12 18:03:11 +02002195 * @ll : link layer of port
Roland Dreier8a518662006-02-13 12:48:12 -08002196 *
2197 * This function is a helper function that a low-level driver's
2198 * modify_qp method can use to validate the consumer's input. It
2199 * checks that cur_state and next_state are valid QP states, that a
2200 * transition from cur_state to next_state is allowed by the IB spec,
2201 * and that the attribute mask supplied is allowed for the transition.
2202 */
2203int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
Matan Barakdd5f03b2013-12-12 18:03:11 +02002204 enum ib_qp_type type, enum ib_qp_attr_mask mask,
2205 enum rdma_link_layer ll);
Roland Dreier8a518662006-02-13 12:48:12 -08002206
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207int ib_register_event_handler (struct ib_event_handler *event_handler);
2208int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2209void ib_dispatch_event(struct ib_event *event);
2210
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211int ib_query_port(struct ib_device *device,
2212 u8 port_num, struct ib_port_attr *port_attr);
2213
Eli Cohena3f5ada2010-09-27 17:51:10 -07002214enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2215 u8 port_num);
2216
Ira Weiny0cf18d72015-05-13 20:02:55 -04002217/**
Hal Rosenstock41390322015-06-29 09:57:00 -04002218 * rdma_cap_ib_switch - Check if the device is IB switch
2219 * @device: Device to check
2220 *
2221 * Device driver is responsible for setting is_switch bit on
2222 * in ib_device structure at init time.
2223 *
2224 * Return: true if the device is IB switch.
2225 */
2226static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2227{
2228 return device->is_switch;
2229}
2230
2231/**
Ira Weiny0cf18d72015-05-13 20:02:55 -04002232 * rdma_start_port - Return the first valid port number for the device
2233 * specified
2234 *
2235 * @device: Device to be checked
2236 *
2237 * Return start port number
2238 */
2239static inline u8 rdma_start_port(const struct ib_device *device)
2240{
Hal Rosenstock41390322015-06-29 09:57:00 -04002241 return rdma_cap_ib_switch(device) ? 0 : 1;
Ira Weiny0cf18d72015-05-13 20:02:55 -04002242}
2243
2244/**
2245 * rdma_end_port - Return the last valid port number for the device
2246 * specified
2247 *
2248 * @device: Device to be checked
2249 *
2250 * Return last port number
2251 */
2252static inline u8 rdma_end_port(const struct ib_device *device)
2253{
Hal Rosenstock41390322015-06-29 09:57:00 -04002254 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
Ira Weiny0cf18d72015-05-13 20:02:55 -04002255}
2256
Ira Weiny5ede9282015-05-31 17:15:29 -04002257static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
Michael Wangde66be92015-05-05 14:50:19 +02002258{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002259 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
Michael Wangde66be92015-05-05 14:50:19 +02002260}
2261
Ira Weiny5ede9282015-05-31 17:15:29 -04002262static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
Michael Wangde66be92015-05-05 14:50:19 +02002263{
Matan Barak7766a992015-12-23 14:56:50 +02002264 return device->port_immutable[port_num].core_cap_flags &
2265 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2266}
2267
2268static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2269{
2270 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2271}
2272
2273static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2274{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002275 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
Michael Wangde66be92015-05-05 14:50:19 +02002276}
2277
Ira Weiny5ede9282015-05-31 17:15:29 -04002278static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
Michael Wangde66be92015-05-05 14:50:19 +02002279{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002280 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
Michael Wangde66be92015-05-05 14:50:19 +02002281}
2282
Ira Weiny5ede9282015-05-31 17:15:29 -04002283static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
Michael Wangde66be92015-05-05 14:50:19 +02002284{
Matan Barak7766a992015-12-23 14:56:50 +02002285 return rdma_protocol_ib(device, port_num) ||
2286 rdma_protocol_roce(device, port_num);
Michael Wangde66be92015-05-05 14:50:19 +02002287}
2288
Michael Wangc757dea2015-05-05 14:50:32 +02002289/**
Michael Wang296ec002015-05-18 10:41:45 +02002290 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
Michael Wangc757dea2015-05-05 14:50:32 +02002291 * Management Datagrams.
Michael Wang296ec002015-05-18 10:41:45 +02002292 * @device: Device to check
2293 * @port_num: Port number to check
Michael Wangc757dea2015-05-05 14:50:32 +02002294 *
Michael Wang296ec002015-05-18 10:41:45 +02002295 * Management Datagrams (MAD) are a required part of the InfiniBand
2296 * specification and are supported on all InfiniBand devices. A slightly
2297 * extended version are also supported on OPA interfaces.
Michael Wangc757dea2015-05-05 14:50:32 +02002298 *
Michael Wang296ec002015-05-18 10:41:45 +02002299 * Return: true if the port supports sending/receiving of MAD packets.
Michael Wangc757dea2015-05-05 14:50:32 +02002300 */
Ira Weiny5ede9282015-05-31 17:15:29 -04002301static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
Michael Wangc757dea2015-05-05 14:50:32 +02002302{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002303 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
Michael Wangc757dea2015-05-05 14:50:32 +02002304}
2305
Michael Wang29541e32015-05-05 14:50:33 +02002306/**
Ira Weiny65995fe2015-06-06 14:38:32 -04002307 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2308 * Management Datagrams.
2309 * @device: Device to check
2310 * @port_num: Port number to check
2311 *
2312 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2313 * datagrams with their own versions. These OPA MADs share many but not all of
2314 * the characteristics of InfiniBand MADs.
2315 *
2316 * OPA MADs differ in the following ways:
2317 *
2318 * 1) MADs are variable size up to 2K
2319 * IBTA defined MADs remain fixed at 256 bytes
2320 * 2) OPA SMPs must carry valid PKeys
2321 * 3) OPA SMP packets are a different format
2322 *
2323 * Return: true if the port supports OPA MAD packet formats.
2324 */
2325static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2326{
2327 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2328 == RDMA_CORE_CAP_OPA_MAD;
2329}
2330
2331/**
Michael Wang296ec002015-05-18 10:41:45 +02002332 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2333 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2334 * @device: Device to check
2335 * @port_num: Port number to check
Michael Wang29541e32015-05-05 14:50:33 +02002336 *
Michael Wang296ec002015-05-18 10:41:45 +02002337 * Each InfiniBand node is required to provide a Subnet Management Agent
2338 * that the subnet manager can access. Prior to the fabric being fully
2339 * configured by the subnet manager, the SMA is accessed via a well known
2340 * interface called the Subnet Management Interface (SMI). This interface
2341 * uses directed route packets to communicate with the SM to get around the
2342 * chicken and egg problem of the SM needing to know what's on the fabric
2343 * in order to configure the fabric, and needing to configure the fabric in
2344 * order to send packets to the devices on the fabric. These directed
2345 * route packets do not need the fabric fully configured in order to reach
2346 * their destination. The SMI is the only method allowed to send
2347 * directed route packets on an InfiniBand fabric.
Michael Wang29541e32015-05-05 14:50:33 +02002348 *
Michael Wang296ec002015-05-18 10:41:45 +02002349 * Return: true if the port provides an SMI.
Michael Wang29541e32015-05-05 14:50:33 +02002350 */
Ira Weiny5ede9282015-05-31 17:15:29 -04002351static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
Michael Wang29541e32015-05-05 14:50:33 +02002352{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002353 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
Michael Wang29541e32015-05-05 14:50:33 +02002354}
2355
Michael Wang72219cea2015-05-05 14:50:34 +02002356/**
2357 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2358 * Communication Manager.
Michael Wang296ec002015-05-18 10:41:45 +02002359 * @device: Device to check
2360 * @port_num: Port number to check
Michael Wang72219cea2015-05-05 14:50:34 +02002361 *
Michael Wang296ec002015-05-18 10:41:45 +02002362 * The InfiniBand Communication Manager is one of many pre-defined General
2363 * Service Agents (GSA) that are accessed via the General Service
2364 * Interface (GSI). It's role is to facilitate establishment of connections
2365 * between nodes as well as other management related tasks for established
2366 * connections.
Michael Wang72219cea2015-05-05 14:50:34 +02002367 *
Michael Wang296ec002015-05-18 10:41:45 +02002368 * Return: true if the port supports an IB CM (this does not guarantee that
2369 * a CM is actually running however).
Michael Wang72219cea2015-05-05 14:50:34 +02002370 */
Ira Weiny5ede9282015-05-31 17:15:29 -04002371static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
Michael Wang72219cea2015-05-05 14:50:34 +02002372{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002373 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
Michael Wang72219cea2015-05-05 14:50:34 +02002374}
2375
Michael Wang04215332015-05-05 14:50:35 +02002376/**
2377 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2378 * Communication Manager.
Michael Wang296ec002015-05-18 10:41:45 +02002379 * @device: Device to check
2380 * @port_num: Port number to check
Michael Wang04215332015-05-05 14:50:35 +02002381 *
Michael Wang296ec002015-05-18 10:41:45 +02002382 * Similar to above, but specific to iWARP connections which have a different
2383 * managment protocol than InfiniBand.
Michael Wang04215332015-05-05 14:50:35 +02002384 *
Michael Wang296ec002015-05-18 10:41:45 +02002385 * Return: true if the port supports an iWARP CM (this does not guarantee that
2386 * a CM is actually running however).
Michael Wang04215332015-05-05 14:50:35 +02002387 */
Ira Weiny5ede9282015-05-31 17:15:29 -04002388static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
Michael Wang04215332015-05-05 14:50:35 +02002389{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002390 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
Michael Wang04215332015-05-05 14:50:35 +02002391}
2392
Michael Wangfe53ba22015-05-05 14:50:36 +02002393/**
2394 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2395 * Subnet Administration.
Michael Wang296ec002015-05-18 10:41:45 +02002396 * @device: Device to check
2397 * @port_num: Port number to check
Michael Wangfe53ba22015-05-05 14:50:36 +02002398 *
Michael Wang296ec002015-05-18 10:41:45 +02002399 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2400 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2401 * fabrics, devices should resolve routes to other hosts by contacting the
2402 * SA to query the proper route.
Michael Wangfe53ba22015-05-05 14:50:36 +02002403 *
Michael Wang296ec002015-05-18 10:41:45 +02002404 * Return: true if the port should act as a client to the fabric Subnet
2405 * Administration interface. This does not imply that the SA service is
2406 * running locally.
Michael Wangfe53ba22015-05-05 14:50:36 +02002407 */
Ira Weiny5ede9282015-05-31 17:15:29 -04002408static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
Michael Wangfe53ba22015-05-05 14:50:36 +02002409{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002410 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
Michael Wangfe53ba22015-05-05 14:50:36 +02002411}
2412
Michael Wanga31ad3b2015-05-05 14:50:37 +02002413/**
2414 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2415 * Multicast.
Michael Wang296ec002015-05-18 10:41:45 +02002416 * @device: Device to check
2417 * @port_num: Port number to check
Michael Wanga31ad3b2015-05-05 14:50:37 +02002418 *
Michael Wang296ec002015-05-18 10:41:45 +02002419 * InfiniBand multicast registration is more complex than normal IPv4 or
2420 * IPv6 multicast registration. Each Host Channel Adapter must register
2421 * with the Subnet Manager when it wishes to join a multicast group. It
2422 * should do so only once regardless of how many queue pairs it subscribes
2423 * to this group. And it should leave the group only after all queue pairs
2424 * attached to the group have been detached.
Michael Wanga31ad3b2015-05-05 14:50:37 +02002425 *
Michael Wang296ec002015-05-18 10:41:45 +02002426 * Return: true if the port must undertake the additional adminstrative
2427 * overhead of registering/unregistering with the SM and tracking of the
2428 * total number of queue pairs attached to the multicast group.
Michael Wanga31ad3b2015-05-05 14:50:37 +02002429 */
Ira Weiny5ede9282015-05-31 17:15:29 -04002430static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
Michael Wanga31ad3b2015-05-05 14:50:37 +02002431{
2432 return rdma_cap_ib_sa(device, port_num);
2433}
2434
Michael Wangbc0f1d72015-05-05 14:50:38 +02002435/**
Michael Wang30a74ef2015-05-05 14:50:39 +02002436 * rdma_cap_af_ib - Check if the port of device has the capability
2437 * Native Infiniband Address.
Michael Wang296ec002015-05-18 10:41:45 +02002438 * @device: Device to check
2439 * @port_num: Port number to check
Michael Wang30a74ef2015-05-05 14:50:39 +02002440 *
Michael Wang296ec002015-05-18 10:41:45 +02002441 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2442 * GID. RoCE uses a different mechanism, but still generates a GID via
2443 * a prescribed mechanism and port specific data.
Michael Wang30a74ef2015-05-05 14:50:39 +02002444 *
Michael Wang296ec002015-05-18 10:41:45 +02002445 * Return: true if the port uses a GID address to identify devices on the
2446 * network.
Michael Wang30a74ef2015-05-05 14:50:39 +02002447 */
Ira Weiny5ede9282015-05-31 17:15:29 -04002448static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
Michael Wang30a74ef2015-05-05 14:50:39 +02002449{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002450 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
Michael Wang30a74ef2015-05-05 14:50:39 +02002451}
2452
2453/**
Michael Wang227128f2015-05-05 14:50:40 +02002454 * rdma_cap_eth_ah - Check if the port of device has the capability
Michael Wang296ec002015-05-18 10:41:45 +02002455 * Ethernet Address Handle.
2456 * @device: Device to check
2457 * @port_num: Port number to check
Michael Wang227128f2015-05-05 14:50:40 +02002458 *
Michael Wang296ec002015-05-18 10:41:45 +02002459 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2460 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2461 * port. Normally, packet headers are generated by the sending host
2462 * adapter, but when sending connectionless datagrams, we must manually
2463 * inject the proper headers for the fabric we are communicating over.
Michael Wang227128f2015-05-05 14:50:40 +02002464 *
Michael Wang296ec002015-05-18 10:41:45 +02002465 * Return: true if we are running as a RoCE port and must force the
2466 * addition of a Global Route Header built from our Ethernet Address
2467 * Handle into our header list for connectionless packets.
Michael Wang227128f2015-05-05 14:50:40 +02002468 */
Ira Weiny5ede9282015-05-31 17:15:29 -04002469static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
Michael Wang227128f2015-05-05 14:50:40 +02002470{
Ira Weinyf9b22e32015-05-13 20:02:59 -04002471 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
Michael Wang227128f2015-05-05 14:50:40 +02002472}
2473
2474/**
Ira Weiny337877a2015-06-06 14:38:29 -04002475 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2476 *
2477 * @device: Device
2478 * @port_num: Port number
2479 *
2480 * This MAD size includes the MAD headers and MAD payload. No other headers
2481 * are included.
2482 *
2483 * Return the max MAD size required by the Port. Will return 0 if the port
2484 * does not support MADs
2485 */
2486static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2487{
2488 return device->port_immutable[port_num].max_mad_size;
2489}
2490
Matan Barak03db3a22015-07-30 18:33:26 +03002491/**
2492 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2493 * @device: Device to check
2494 * @port_num: Port number to check
2495 *
2496 * RoCE GID table mechanism manages the various GIDs for a device.
2497 *
2498 * NOTE: if allocating the port's GID table has failed, this call will still
2499 * return true, but any RoCE GID table API will fail.
2500 *
2501 * Return: true if the port uses RoCE GID table mechanism in order to manage
2502 * its GIDs.
2503 */
2504static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2505 u8 port_num)
2506{
2507 return rdma_protocol_roce(device, port_num) &&
2508 device->add_gid && device->del_gid;
2509}
2510
Christoph Hellwig002516e2016-05-03 18:01:05 +02002511/*
2512 * Check if the device supports READ W/ INVALIDATE.
2513 */
2514static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2515{
2516 /*
2517 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
2518 * has support for it yet.
2519 */
2520 return rdma_protocol_iwarp(dev, port_num);
2521}
2522
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523int ib_query_gid(struct ib_device *device,
Matan Barak55ee3ab2015-10-15 18:38:45 +03002524 u8 port_num, int index, union ib_gid *gid,
2525 struct ib_gid_attr *attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526
Eli Cohen50174a72016-03-11 22:58:38 +02002527int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2528 int state);
2529int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2530 struct ifla_vf_info *info);
2531int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2532 struct ifla_vf_stats *stats);
2533int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2534 int type);
2535
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536int ib_query_pkey(struct ib_device *device,
2537 u8 port_num, u16 index, u16 *pkey);
2538
2539int ib_modify_device(struct ib_device *device,
2540 int device_modify_mask,
2541 struct ib_device_modify *device_modify);
2542
2543int ib_modify_port(struct ib_device *device,
2544 u8 port_num, int port_modify_mask,
2545 struct ib_port_modify *port_modify);
2546
Yosef Etigin5eb620c2007-05-14 07:26:51 +03002547int ib_find_gid(struct ib_device *device, union ib_gid *gid,
Matan Barakb39ffa12015-12-23 14:56:47 +02002548 enum ib_gid_type gid_type, struct net_device *ndev,
2549 u8 *port_num, u16 *index);
Yosef Etigin5eb620c2007-05-14 07:26:51 +03002550
2551int ib_find_pkey(struct ib_device *device,
2552 u8 port_num, u16 pkey, u16 *index);
2553
Christoph Hellwiged082d32016-09-05 12:56:17 +02002554enum ib_pd_flags {
2555 /*
2556 * Create a memory registration for all memory in the system and place
2557 * the rkey for it into pd->unsafe_global_rkey. This can be used by
2558 * ULPs to avoid the overhead of dynamic MRs.
2559 *
2560 * This flag is generally considered unsafe and must only be used in
2561 * extremly trusted environments. Every use of it will log a warning
2562 * in the kernel log.
2563 */
2564 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
2565};
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566
Christoph Hellwiged082d32016-09-05 12:56:17 +02002567struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2568 const char *caller);
2569#define ib_alloc_pd(device, flags) \
2570 __ib_alloc_pd((device), (flags), __func__)
Jason Gunthorpe7dd78642015-08-05 14:34:31 -06002571void ib_dealloc_pd(struct ib_pd *pd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572
2573/**
2574 * ib_create_ah - Creates an address handle for the given address vector.
2575 * @pd: The protection domain associated with the address handle.
2576 * @ah_attr: The attributes of the address vector.
2577 *
2578 * The address handle is used to reference a local or global destination
2579 * in all UD QP post sends.
2580 */
2581struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2582
2583/**
Sean Hefty4e00d692006-06-17 20:37:39 -07002584 * ib_init_ah_from_wc - Initializes address handle attributes from a
2585 * work completion.
2586 * @device: Device on which the received message arrived.
2587 * @port_num: Port on which the received message arrived.
2588 * @wc: Work completion associated with the received message.
2589 * @grh: References the received global route header. This parameter is
2590 * ignored unless the work completion indicates that the GRH is valid.
2591 * @ah_attr: Returned attributes that can be used when creating an address
2592 * handle for replying to the message.
2593 */
Ira Weiny73cdaae2015-05-31 17:15:31 -04002594int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2595 const struct ib_wc *wc, const struct ib_grh *grh,
2596 struct ib_ah_attr *ah_attr);
Sean Hefty4e00d692006-06-17 20:37:39 -07002597
2598/**
Hal Rosenstock513789e2005-07-27 11:45:34 -07002599 * ib_create_ah_from_wc - Creates an address handle associated with the
2600 * sender of the specified work completion.
2601 * @pd: The protection domain associated with the address handle.
2602 * @wc: Work completion information associated with a received message.
2603 * @grh: References the received global route header. This parameter is
2604 * ignored unless the work completion indicates that the GRH is valid.
2605 * @port_num: The outbound port number to associate with the address.
2606 *
2607 * The address handle is used to reference a local or global destination
2608 * in all UD QP post sends.
2609 */
Ira Weiny73cdaae2015-05-31 17:15:31 -04002610struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2611 const struct ib_grh *grh, u8 port_num);
Hal Rosenstock513789e2005-07-27 11:45:34 -07002612
2613/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614 * ib_modify_ah - Modifies the address vector associated with an address
2615 * handle.
2616 * @ah: The address handle to modify.
2617 * @ah_attr: The new address vector attributes to associate with the
2618 * address handle.
2619 */
2620int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2621
2622/**
2623 * ib_query_ah - Queries the address vector associated with an address
2624 * handle.
2625 * @ah: The address handle to query.
2626 * @ah_attr: The address vector attributes associated with the address
2627 * handle.
2628 */
2629int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2630
2631/**
2632 * ib_destroy_ah - Destroys an address handle.
2633 * @ah: The address handle to destroy.
2634 */
2635int ib_destroy_ah(struct ib_ah *ah);
2636
2637/**
Roland Dreierd41fcc62005-08-18 12:23:08 -07002638 * ib_create_srq - Creates a SRQ associated with the specified protection
2639 * domain.
2640 * @pd: The protection domain associated with the SRQ.
Dotan Barakabb6e9b2006-02-23 12:13:51 -08002641 * @srq_init_attr: A list of initial attributes required to create the
2642 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2643 * the actual capabilities of the created SRQ.
Roland Dreierd41fcc62005-08-18 12:23:08 -07002644 *
2645 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2646 * requested size of the SRQ, and set to the actual values allocated
2647 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2648 * will always be at least as large as the requested values.
2649 */
2650struct ib_srq *ib_create_srq(struct ib_pd *pd,
2651 struct ib_srq_init_attr *srq_init_attr);
2652
2653/**
2654 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2655 * @srq: The SRQ to modify.
2656 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2657 * the current values of selected SRQ attributes are returned.
2658 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2659 * are being modified.
2660 *
2661 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2662 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2663 * the number of receives queued drops below the limit.
2664 */
2665int ib_modify_srq(struct ib_srq *srq,
2666 struct ib_srq_attr *srq_attr,
2667 enum ib_srq_attr_mask srq_attr_mask);
2668
2669/**
2670 * ib_query_srq - Returns the attribute list and current values for the
2671 * specified SRQ.
2672 * @srq: The SRQ to query.
2673 * @srq_attr: The attributes of the specified SRQ.
2674 */
2675int ib_query_srq(struct ib_srq *srq,
2676 struct ib_srq_attr *srq_attr);
2677
2678/**
2679 * ib_destroy_srq - Destroys the specified SRQ.
2680 * @srq: The SRQ to destroy.
2681 */
2682int ib_destroy_srq(struct ib_srq *srq);
2683
2684/**
2685 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2686 * @srq: The SRQ to post the work request on.
2687 * @recv_wr: A list of work requests to post on the receive queue.
2688 * @bad_recv_wr: On an immediate failure, this parameter will reference
2689 * the work request that failed to be posted on the QP.
2690 */
2691static inline int ib_post_srq_recv(struct ib_srq *srq,
2692 struct ib_recv_wr *recv_wr,
2693 struct ib_recv_wr **bad_recv_wr)
2694{
2695 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2696}
2697
2698/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699 * ib_create_qp - Creates a QP associated with the specified protection
2700 * domain.
2701 * @pd: The protection domain associated with the QP.
Dotan Barakabb6e9b2006-02-23 12:13:51 -08002702 * @qp_init_attr: A list of initial attributes required to create the
2703 * QP. If QP creation succeeds, then the attributes are updated to
2704 * the actual capabilities of the created QP.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 */
2706struct ib_qp *ib_create_qp(struct ib_pd *pd,
2707 struct ib_qp_init_attr *qp_init_attr);
2708
2709/**
2710 * ib_modify_qp - Modifies the attributes for the specified QP and then
2711 * transitions the QP to the given state.
2712 * @qp: The QP to modify.
2713 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2714 * the current values of selected QP attributes are returned.
2715 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2716 * are being modified.
2717 */
2718int ib_modify_qp(struct ib_qp *qp,
2719 struct ib_qp_attr *qp_attr,
2720 int qp_attr_mask);
2721
2722/**
2723 * ib_query_qp - Returns the attribute list and current values for the
2724 * specified QP.
2725 * @qp: The QP to query.
2726 * @qp_attr: The attributes of the specified QP.
2727 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2728 * @qp_init_attr: Additional attributes of the selected QP.
2729 *
2730 * The qp_attr_mask may be used to limit the query to gathering only the
2731 * selected attributes.
2732 */
2733int ib_query_qp(struct ib_qp *qp,
2734 struct ib_qp_attr *qp_attr,
2735 int qp_attr_mask,
2736 struct ib_qp_init_attr *qp_init_attr);
2737
2738/**
2739 * ib_destroy_qp - Destroys the specified QP.
2740 * @qp: The QP to destroy.
2741 */
2742int ib_destroy_qp(struct ib_qp *qp);
2743
2744/**
Sean Hefty0e0ec7e2011-08-08 15:31:51 -07002745 * ib_open_qp - Obtain a reference to an existing sharable QP.
2746 * @xrcd - XRC domain
2747 * @qp_open_attr: Attributes identifying the QP to open.
2748 *
2749 * Returns a reference to a sharable QP.
2750 */
2751struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2752 struct ib_qp_open_attr *qp_open_attr);
2753
2754/**
2755 * ib_close_qp - Release an external reference to a QP.
Sean Heftyd3d72d92011-05-26 23:06:44 -07002756 * @qp: The QP handle to release
2757 *
Sean Hefty0e0ec7e2011-08-08 15:31:51 -07002758 * The opened QP handle is released by the caller. The underlying
2759 * shared QP is not destroyed until all internal references are released.
Sean Heftyd3d72d92011-05-26 23:06:44 -07002760 */
Sean Hefty0e0ec7e2011-08-08 15:31:51 -07002761int ib_close_qp(struct ib_qp *qp);
Sean Heftyd3d72d92011-05-26 23:06:44 -07002762
2763/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764 * ib_post_send - Posts a list of work requests to the send queue of
2765 * the specified QP.
2766 * @qp: The QP to post the work request on.
2767 * @send_wr: A list of work requests to post on the send queue.
2768 * @bad_send_wr: On an immediate failure, this parameter will reference
2769 * the work request that failed to be posted on the QP.
Bart Van Assche55464d42009-12-09 14:20:04 -08002770 *
2771 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2772 * error is returned, the QP state shall not be affected,
2773 * ib_post_send() will return an immediate error after queueing any
2774 * earlier work requests in the list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775 */
2776static inline int ib_post_send(struct ib_qp *qp,
2777 struct ib_send_wr *send_wr,
2778 struct ib_send_wr **bad_send_wr)
2779{
2780 return qp->device->post_send(qp, send_wr, bad_send_wr);
2781}
2782
2783/**
2784 * ib_post_recv - Posts a list of work requests to the receive queue of
2785 * the specified QP.
2786 * @qp: The QP to post the work request on.
2787 * @recv_wr: A list of work requests to post on the receive queue.
2788 * @bad_recv_wr: On an immediate failure, this parameter will reference
2789 * the work request that failed to be posted on the QP.
2790 */
2791static inline int ib_post_recv(struct ib_qp *qp,
2792 struct ib_recv_wr *recv_wr,
2793 struct ib_recv_wr **bad_recv_wr)
2794{
2795 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2796}
2797
Christoph Hellwig14d3a3b2015-12-11 11:53:03 -08002798struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2799 int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2800void ib_free_cq(struct ib_cq *cq);
2801int ib_process_cq_direct(struct ib_cq *cq, int budget);
2802
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803/**
2804 * ib_create_cq - Creates a CQ on the specified device.
2805 * @device: The device on which to create the CQ.
2806 * @comp_handler: A user-specified callback that is invoked when a
2807 * completion event occurs on the CQ.
2808 * @event_handler: A user-specified callback that is invoked when an
2809 * asynchronous event not associated with a completion occurs on the CQ.
2810 * @cq_context: Context associated with the CQ returned to the user via
2811 * the associated completion and event handlers.
Matan Barak8e372102015-06-11 16:35:21 +03002812 * @cq_attr: The attributes the CQ should be created upon.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813 *
2814 * Users can examine the cq structure to determine the actual CQ size.
2815 */
2816struct ib_cq *ib_create_cq(struct ib_device *device,
2817 ib_comp_handler comp_handler,
2818 void (*event_handler)(struct ib_event *, void *),
Matan Barak8e372102015-06-11 16:35:21 +03002819 void *cq_context,
2820 const struct ib_cq_init_attr *cq_attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821
2822/**
2823 * ib_resize_cq - Modifies the capacity of the CQ.
2824 * @cq: The CQ to resize.
2825 * @cqe: The minimum size of the CQ.
2826 *
2827 * Users can examine the cq structure to determine the actual CQ size.
2828 */
2829int ib_resize_cq(struct ib_cq *cq, int cqe);
2830
2831/**
Eli Cohen2dd57162008-04-16 21:09:33 -07002832 * ib_modify_cq - Modifies moderation params of the CQ
2833 * @cq: The CQ to modify.
2834 * @cq_count: number of CQEs that will trigger an event
2835 * @cq_period: max period of time in usec before triggering an event
2836 *
2837 */
2838int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2839
2840/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 * ib_destroy_cq - Destroys the specified CQ.
2842 * @cq: The CQ to destroy.
2843 */
2844int ib_destroy_cq(struct ib_cq *cq);
2845
2846/**
2847 * ib_poll_cq - poll a CQ for completion(s)
2848 * @cq:the CQ being polled
2849 * @num_entries:maximum number of completions to return
2850 * @wc:array of at least @num_entries &struct ib_wc where completions
2851 * will be returned
2852 *
2853 * Poll a CQ for (possibly multiple) completions. If the return value
2854 * is < 0, an error occurred. If the return value is >= 0, it is the
2855 * number of completions returned. If the return value is
2856 * non-negative and < num_entries, then the CQ was emptied.
2857 */
2858static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2859 struct ib_wc *wc)
2860{
2861 return cq->device->poll_cq(cq, num_entries, wc);
2862}
2863
2864/**
2865 * ib_peek_cq - Returns the number of unreaped completions currently
2866 * on the specified CQ.
2867 * @cq: The CQ to peek.
2868 * @wc_cnt: A minimum number of unreaped completions to check for.
2869 *
2870 * If the number of unreaped completions is greater than or equal to wc_cnt,
2871 * this function returns wc_cnt, otherwise, it returns the actual number of
2872 * unreaped completions.
2873 */
2874int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2875
2876/**
2877 * ib_req_notify_cq - Request completion notification on a CQ.
2878 * @cq: The CQ to generate an event for.
Roland Dreiered23a722007-05-06 21:02:48 -07002879 * @flags:
2880 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2881 * to request an event on the next solicited event or next work
2882 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2883 * may also be |ed in to request a hint about missed events, as
2884 * described below.
2885 *
2886 * Return Value:
2887 * < 0 means an error occurred while requesting notification
2888 * == 0 means notification was requested successfully, and if
2889 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2890 * were missed and it is safe to wait for another event. In
2891 * this case is it guaranteed that any work completions added
2892 * to the CQ since the last CQ poll will trigger a completion
2893 * notification event.
2894 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2895 * in. It means that the consumer must poll the CQ again to
2896 * make sure it is empty to avoid missing an event because of a
2897 * race between requesting notification and an entry being
2898 * added to the CQ. This return value means it is possible
2899 * (but not guaranteed) that a work completion has been added
2900 * to the CQ since the last poll without triggering a
2901 * completion notification event.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 */
2903static inline int ib_req_notify_cq(struct ib_cq *cq,
Roland Dreiered23a722007-05-06 21:02:48 -07002904 enum ib_cq_notify_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905{
Roland Dreiered23a722007-05-06 21:02:48 -07002906 return cq->device->req_notify_cq(cq, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907}
2908
2909/**
2910 * ib_req_ncomp_notif - Request completion notification when there are
2911 * at least the specified number of unreaped completions on the CQ.
2912 * @cq: The CQ to generate an event for.
2913 * @wc_cnt: The number of unreaped completions that should be on the
2914 * CQ before an event is generated.
2915 */
2916static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2917{
2918 return cq->device->req_ncomp_notif ?
2919 cq->device->req_ncomp_notif(cq, wc_cnt) :
2920 -ENOSYS;
2921}
2922
2923/**
Ralph Campbell9b513092006-12-12 14:27:41 -08002924 * ib_dma_mapping_error - check a DMA addr for error
2925 * @dev: The device for which the dma_addr was created
2926 * @dma_addr: The DMA address to check
2927 */
2928static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2929{
Ben Collinsd1998ef2006-12-13 22:10:05 -05002930 if (dev->dma_ops)
2931 return dev->dma_ops->mapping_error(dev, dma_addr);
FUJITA Tomonori8d8bb392008-07-25 19:44:49 -07002932 return dma_mapping_error(dev->dma_device, dma_addr);
Ralph Campbell9b513092006-12-12 14:27:41 -08002933}
2934
2935/**
2936 * ib_dma_map_single - Map a kernel virtual address to DMA address
2937 * @dev: The device for which the dma_addr is to be created
2938 * @cpu_addr: The kernel virtual address
2939 * @size: The size of the region in bytes
2940 * @direction: The direction of the DMA
2941 */
2942static inline u64 ib_dma_map_single(struct ib_device *dev,
2943 void *cpu_addr, size_t size,
2944 enum dma_data_direction direction)
2945{
Ben Collinsd1998ef2006-12-13 22:10:05 -05002946 if (dev->dma_ops)
2947 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2948 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
Ralph Campbell9b513092006-12-12 14:27:41 -08002949}
2950
2951/**
2952 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2953 * @dev: The device for which the DMA address was created
2954 * @addr: The DMA address
2955 * @size: The size of the region in bytes
2956 * @direction: The direction of the DMA
2957 */
2958static inline void ib_dma_unmap_single(struct ib_device *dev,
2959 u64 addr, size_t size,
2960 enum dma_data_direction direction)
2961{
Ben Collinsd1998ef2006-12-13 22:10:05 -05002962 if (dev->dma_ops)
2963 dev->dma_ops->unmap_single(dev, addr, size, direction);
2964 else
Ralph Campbell9b513092006-12-12 14:27:41 -08002965 dma_unmap_single(dev->dma_device, addr, size, direction);
2966}
2967
Arthur Kepnercb9fbc52008-04-29 01:00:34 -07002968static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2969 void *cpu_addr, size_t size,
2970 enum dma_data_direction direction,
Krzysztof Kozlowski00085f12016-08-03 13:46:00 -07002971 unsigned long dma_attrs)
Arthur Kepnercb9fbc52008-04-29 01:00:34 -07002972{
2973 return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
Krzysztof Kozlowski00085f12016-08-03 13:46:00 -07002974 direction, dma_attrs);
Arthur Kepnercb9fbc52008-04-29 01:00:34 -07002975}
2976
2977static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2978 u64 addr, size_t size,
2979 enum dma_data_direction direction,
Krzysztof Kozlowski00085f12016-08-03 13:46:00 -07002980 unsigned long dma_attrs)
Arthur Kepnercb9fbc52008-04-29 01:00:34 -07002981{
2982 return dma_unmap_single_attrs(dev->dma_device, addr, size,
Krzysztof Kozlowski00085f12016-08-03 13:46:00 -07002983 direction, dma_attrs);
Arthur Kepnercb9fbc52008-04-29 01:00:34 -07002984}
2985
Ralph Campbell9b513092006-12-12 14:27:41 -08002986/**
2987 * ib_dma_map_page - Map a physical page to DMA address
2988 * @dev: The device for which the dma_addr is to be created
2989 * @page: The page to be mapped
2990 * @offset: The offset within the page
2991 * @size: The size of the region in bytes
2992 * @direction: The direction of the DMA
2993 */
2994static inline u64 ib_dma_map_page(struct ib_device *dev,
2995 struct page *page,
2996 unsigned long offset,
2997 size_t size,
2998 enum dma_data_direction direction)
2999{
Ben Collinsd1998ef2006-12-13 22:10:05 -05003000 if (dev->dma_ops)
3001 return dev->dma_ops->map_page(dev, page, offset, size, direction);
3002 return dma_map_page(dev->dma_device, page, offset, size, direction);
Ralph Campbell9b513092006-12-12 14:27:41 -08003003}
3004
3005/**
3006 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3007 * @dev: The device for which the DMA address was created
3008 * @addr: The DMA address
3009 * @size: The size of the region in bytes
3010 * @direction: The direction of the DMA
3011 */
3012static inline void ib_dma_unmap_page(struct ib_device *dev,
3013 u64 addr, size_t size,
3014 enum dma_data_direction direction)
3015{
Ben Collinsd1998ef2006-12-13 22:10:05 -05003016 if (dev->dma_ops)
3017 dev->dma_ops->unmap_page(dev, addr, size, direction);
3018 else
Ralph Campbell9b513092006-12-12 14:27:41 -08003019 dma_unmap_page(dev->dma_device, addr, size, direction);
3020}
3021
3022/**
3023 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3024 * @dev: The device for which the DMA addresses are to be created
3025 * @sg: The array of scatter/gather entries
3026 * @nents: The number of scatter/gather entries
3027 * @direction: The direction of the DMA
3028 */
3029static inline int ib_dma_map_sg(struct ib_device *dev,
3030 struct scatterlist *sg, int nents,
3031 enum dma_data_direction direction)
3032{
Ben Collinsd1998ef2006-12-13 22:10:05 -05003033 if (dev->dma_ops)
3034 return dev->dma_ops->map_sg(dev, sg, nents, direction);
3035 return dma_map_sg(dev->dma_device, sg, nents, direction);
Ralph Campbell9b513092006-12-12 14:27:41 -08003036}
3037
3038/**
3039 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3040 * @dev: The device for which the DMA addresses were created
3041 * @sg: The array of scatter/gather entries
3042 * @nents: The number of scatter/gather entries
3043 * @direction: The direction of the DMA
3044 */
3045static inline void ib_dma_unmap_sg(struct ib_device *dev,
3046 struct scatterlist *sg, int nents,
3047 enum dma_data_direction direction)
3048{
Ben Collinsd1998ef2006-12-13 22:10:05 -05003049 if (dev->dma_ops)
3050 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
3051 else
Ralph Campbell9b513092006-12-12 14:27:41 -08003052 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3053}
3054
Arthur Kepnercb9fbc52008-04-29 01:00:34 -07003055static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3056 struct scatterlist *sg, int nents,
3057 enum dma_data_direction direction,
Krzysztof Kozlowski00085f12016-08-03 13:46:00 -07003058 unsigned long dma_attrs)
Arthur Kepnercb9fbc52008-04-29 01:00:34 -07003059{
Parav Panditd9703652016-09-28 20:25:47 +00003060 if (dev->dma_ops)
3061 return dev->dma_ops->map_sg_attrs(dev, sg, nents, direction,
3062 dma_attrs);
3063 else
3064 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3065 dma_attrs);
Arthur Kepnercb9fbc52008-04-29 01:00:34 -07003066}
3067
3068static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3069 struct scatterlist *sg, int nents,
3070 enum dma_data_direction direction,
Krzysztof Kozlowski00085f12016-08-03 13:46:00 -07003071 unsigned long dma_attrs)
Arthur Kepnercb9fbc52008-04-29 01:00:34 -07003072{
Parav Panditd9703652016-09-28 20:25:47 +00003073 if (dev->dma_ops)
3074 return dev->dma_ops->unmap_sg_attrs(dev, sg, nents, direction,
3075 dma_attrs);
3076 else
3077 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
3078 dma_attrs);
Arthur Kepnercb9fbc52008-04-29 01:00:34 -07003079}
Ralph Campbell9b513092006-12-12 14:27:41 -08003080/**
3081 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3082 * @dev: The device for which the DMA addresses were created
3083 * @sg: The scatter/gather entry
Mike Marciniszynea58a592014-03-28 13:26:59 -04003084 *
3085 * Note: this function is obsolete. To do: change all occurrences of
3086 * ib_sg_dma_address() into sg_dma_address().
Ralph Campbell9b513092006-12-12 14:27:41 -08003087 */
3088static inline u64 ib_sg_dma_address(struct ib_device *dev,
3089 struct scatterlist *sg)
3090{
Ben Collinsd1998ef2006-12-13 22:10:05 -05003091 return sg_dma_address(sg);
Ralph Campbell9b513092006-12-12 14:27:41 -08003092}
3093
3094/**
3095 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3096 * @dev: The device for which the DMA addresses were created
3097 * @sg: The scatter/gather entry
Mike Marciniszynea58a592014-03-28 13:26:59 -04003098 *
3099 * Note: this function is obsolete. To do: change all occurrences of
3100 * ib_sg_dma_len() into sg_dma_len().
Ralph Campbell9b513092006-12-12 14:27:41 -08003101 */
3102static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3103 struct scatterlist *sg)
3104{
Ben Collinsd1998ef2006-12-13 22:10:05 -05003105 return sg_dma_len(sg);
Ralph Campbell9b513092006-12-12 14:27:41 -08003106}
3107
3108/**
3109 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3110 * @dev: The device for which the DMA address was created
3111 * @addr: The DMA address
3112 * @size: The size of the region in bytes
3113 * @dir: The direction of the DMA
3114 */
3115static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3116 u64 addr,
3117 size_t size,
3118 enum dma_data_direction dir)
3119{
Ben Collinsd1998ef2006-12-13 22:10:05 -05003120 if (dev->dma_ops)
3121 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
3122 else
Ralph Campbell9b513092006-12-12 14:27:41 -08003123 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3124}
3125
3126/**
3127 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3128 * @dev: The device for which the DMA address was created
3129 * @addr: The DMA address
3130 * @size: The size of the region in bytes
3131 * @dir: The direction of the DMA
3132 */
3133static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3134 u64 addr,
3135 size_t size,
3136 enum dma_data_direction dir)
3137{
Ben Collinsd1998ef2006-12-13 22:10:05 -05003138 if (dev->dma_ops)
3139 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
3140 else
Ralph Campbell9b513092006-12-12 14:27:41 -08003141 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3142}
3143
3144/**
3145 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3146 * @dev: The device for which the DMA address is requested
3147 * @size: The size of the region to allocate in bytes
3148 * @dma_handle: A pointer for returning the DMA address of the region
3149 * @flag: memory allocator flags
3150 */
3151static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3152 size_t size,
3153 u64 *dma_handle,
3154 gfp_t flag)
3155{
Ben Collinsd1998ef2006-12-13 22:10:05 -05003156 if (dev->dma_ops)
3157 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
Roland Dreierc59a3da2006-12-15 13:57:26 -08003158 else {
3159 dma_addr_t handle;
3160 void *ret;
3161
3162 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
3163 *dma_handle = handle;
3164 return ret;
3165 }
Ralph Campbell9b513092006-12-12 14:27:41 -08003166}
3167
3168/**
3169 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3170 * @dev: The device for which the DMA addresses were allocated
3171 * @size: The size of the region
3172 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3173 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3174 */
3175static inline void ib_dma_free_coherent(struct ib_device *dev,
3176 size_t size, void *cpu_addr,
3177 u64 dma_handle)
3178{
Ben Collinsd1998ef2006-12-13 22:10:05 -05003179 if (dev->dma_ops)
3180 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
3181 else
Ralph Campbell9b513092006-12-12 14:27:41 -08003182 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3183}
3184
3185/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003186 * ib_dereg_mr - Deregisters a memory region and removes it from the
3187 * HCA translation table.
3188 * @mr: The memory region to deregister.
Shani Michaeli7083e422013-02-06 16:19:12 +00003189 *
3190 * This function can fail, if the memory region has memory windows bound to it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191 */
3192int ib_dereg_mr(struct ib_mr *mr);
3193
Sagi Grimberg9bee1782015-07-30 10:32:35 +03003194struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3195 enum ib_mr_type mr_type,
3196 u32 max_num_sg);
Steve Wise00f7ec32008-07-14 23:48:45 -07003197
3198/**
Steve Wise00f7ec32008-07-14 23:48:45 -07003199 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3200 * R_Key and L_Key.
3201 * @mr - struct ib_mr pointer to be updated.
3202 * @newkey - new key to be used.
3203 */
3204static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3205{
3206 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3207 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3208}
3209
3210/**
Shani Michaeli7083e422013-02-06 16:19:12 +00003211 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3212 * for calculating a new rkey for type 2 memory windows.
3213 * @rkey - the rkey to increment.
3214 */
3215static inline u32 ib_inc_rkey(u32 rkey)
3216{
3217 const u32 mask = 0x000000ff;
3218 return ((rkey + 1) & mask) | (rkey & ~mask);
3219}
3220
3221/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003222 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3223 * @pd: The protection domain associated with the unmapped region.
3224 * @mr_access_flags: Specifies the memory access rights.
3225 * @fmr_attr: Attributes of the unmapped region.
3226 *
3227 * A fast memory region must be mapped before it can be used as part of
3228 * a work request.
3229 */
3230struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3231 int mr_access_flags,
3232 struct ib_fmr_attr *fmr_attr);
3233
3234/**
3235 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3236 * @fmr: The fast memory region to associate with the pages.
3237 * @page_list: An array of physical pages to map to the fast memory region.
3238 * @list_len: The number of pages in page_list.
3239 * @iova: The I/O virtual address to use with the mapped region.
3240 */
3241static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3242 u64 *page_list, int list_len,
3243 u64 iova)
3244{
3245 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3246}
3247
3248/**
3249 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3250 * @fmr_list: A linked list of fast memory regions to unmap.
3251 */
3252int ib_unmap_fmr(struct list_head *fmr_list);
3253
3254/**
3255 * ib_dealloc_fmr - Deallocates a fast memory region.
3256 * @fmr: The fast memory region to deallocate.
3257 */
3258int ib_dealloc_fmr(struct ib_fmr *fmr);
3259
3260/**
3261 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3262 * @qp: QP to attach to the multicast group. The QP must be type
3263 * IB_QPT_UD.
3264 * @gid: Multicast group GID.
3265 * @lid: Multicast group LID in host byte order.
3266 *
3267 * In order to send and receive multicast packets, subnet
3268 * administration must have created the multicast group and configured
3269 * the fabric appropriately. The port associated with the specified
3270 * QP must also be a member of the multicast group.
3271 */
3272int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3273
3274/**
3275 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3276 * @qp: QP to detach from the multicast group.
3277 * @gid: Multicast group GID.
3278 * @lid: Multicast group LID in host byte order.
3279 */
3280int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3281
Sean Hefty59991f92011-05-23 17:52:46 -07003282/**
3283 * ib_alloc_xrcd - Allocates an XRC domain.
3284 * @device: The device on which to allocate the XRC domain.
3285 */
3286struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3287
3288/**
3289 * ib_dealloc_xrcd - Deallocates an XRC domain.
3290 * @xrcd: The XRC domain to deallocate.
3291 */
3292int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3293
Hadar Hen Zion319a4412013-08-07 14:01:59 +03003294struct ib_flow *ib_create_flow(struct ib_qp *qp,
3295 struct ib_flow_attr *flow_attr, int domain);
3296int ib_destroy_flow(struct ib_flow *flow_id);
3297
Eli Cohen1c636f82013-10-31 15:26:32 +02003298static inline int ib_check_mr_access(int flags)
3299{
3300 /*
3301 * Local write permission is required if remote write or
3302 * remote atomic permission is also requested.
3303 */
3304 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3305 !(flags & IB_ACCESS_LOCAL_WRITE))
3306 return -EINVAL;
3307
3308 return 0;
3309}
3310
Sagi Grimberg1b01d332014-02-23 14:19:05 +02003311/**
3312 * ib_check_mr_status: lightweight check of MR status.
3313 * This routine may provide status checks on a selected
3314 * ib_mr. first use is for signature status check.
3315 *
3316 * @mr: A memory region.
3317 * @check_mask: Bitmask of which checks to perform from
3318 * ib_mr_status_check enumeration.
3319 * @mr_status: The container of relevant status checks.
3320 * failed checks will be indicated in the status bitmask
3321 * and the relevant info shall be in the error item.
3322 */
3323int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3324 struct ib_mr_status *mr_status);
3325
Yotam Kenneth9268f722015-07-30 17:50:15 +03003326struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3327 u16 pkey, const union ib_gid *gid,
3328 const struct sockaddr *addr);
Yishai Hadas5fd251c2016-05-23 15:20:48 +03003329struct ib_wq *ib_create_wq(struct ib_pd *pd,
3330 struct ib_wq_init_attr *init_attr);
3331int ib_destroy_wq(struct ib_wq *wq);
3332int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3333 u32 wq_attr_mask);
Yishai Hadas6d397862016-05-23 15:20:51 +03003334struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3335 struct ib_rwq_ind_table_init_attr*
3336 wq_ind_table_init_attr);
3337int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
Yotam Kenneth9268f722015-07-30 17:50:15 +03003338
Christoph Hellwigff2ba992016-05-03 18:01:04 +02003339int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
Bart Van Assche9aa8b322016-05-12 10:49:15 -07003340 unsigned int *sg_offset, unsigned int page_size);
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +03003341
3342static inline int
Christoph Hellwigff2ba992016-05-03 18:01:04 +02003343ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
Bart Van Assche9aa8b322016-05-12 10:49:15 -07003344 unsigned int *sg_offset, unsigned int page_size)
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +03003345{
3346 int n;
3347
Christoph Hellwigff2ba992016-05-03 18:01:04 +02003348 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +03003349 mr->iova = 0;
3350
3351 return n;
3352}
3353
Christoph Hellwigff2ba992016-05-03 18:01:04 +02003354int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
Bart Van Assche9aa8b322016-05-12 10:49:15 -07003355 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
Sagi Grimberg4c67e2b2015-10-13 19:11:24 +03003356
Steve Wise765d6772016-02-17 08:15:41 -08003357void ib_drain_rq(struct ib_qp *qp);
3358void ib_drain_sq(struct ib_qp *qp);
3359void ib_drain_qp(struct ib_qp *qp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360#endif /* IB_VERBS_H */