Blame view
net/rds/ib.c
12.2 KB
ec16227e1 RDS/IB: Infiniban... |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 |
/* * Copyright (c) 2006 Oracle. All rights reserved. * * This software is available to you under a choice of one of two * licenses. You may choose to be licensed under the terms of the GNU * General Public License (GPL) Version 2, available from the file * COPYING in the main directory of this source tree, or the * OpenIB.org BSD license below: * * Redistribution and use in source and binary forms, with or * without modification, are permitted provided that the following * conditions are met: * * - Redistributions of source code must retain the above * copyright notice, this list of conditions and the following * disclaimer. * * - Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following * disclaimer in the documentation and/or other materials * provided with the distribution. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. * */ #include <linux/kernel.h> #include <linux/in.h> #include <linux/if.h> #include <linux/netdevice.h> #include <linux/inetdevice.h> #include <linux/if_arp.h> #include <linux/delay.h> |
5a0e3ad6a include cleanup: ... |
40 |
#include <linux/slab.h> |
3a9a231d9 net: Fix files ex... |
41 |
#include <linux/module.h> |
ec16227e1 RDS/IB: Infiniban... |
42 43 44 |
#include "rds.h" #include "ib.h" |
ff51bf841 rds: make local f... |
45 |
static unsigned int fmr_pool_size = RDS_FMR_POOL_SIZE; |
ec16227e1 RDS/IB: Infiniban... |
46 |
unsigned int fmr_message_size = RDS_FMR_SIZE + 1; /* +1 allows for unaligned MRs */ |
3ba23ade4 RDS: Set retry_co... |
47 |
unsigned int rds_ib_retry_count = RDS_IB_DEFAULT_RETRY_COUNT; |
ec16227e1 RDS/IB: Infiniban... |
48 49 50 51 52 |
module_param(fmr_pool_size, int, 0444); MODULE_PARM_DESC(fmr_pool_size, " Max number of fmr per HCA"); module_param(fmr_message_size, int, 0444); MODULE_PARM_DESC(fmr_message_size, " Max size of a RDMA transfer"); |
3ba23ade4 RDS: Set retry_co... |
53 54 |
module_param(rds_ib_retry_count, int, 0444); MODULE_PARM_DESC(rds_ib_retry_count, " Number of hw retries before reporting an error"); |
ec16227e1 RDS/IB: Infiniban... |
55 |
|
ea819867b RDS/IB: protect t... |
56 57 58 59 60 61 |
/* * we have a clumsy combination of RCU and a rwsem protecting this list * because it is used both in the get_mr fast path and while blocking in * the FMR flushing path. */ DECLARE_RWSEM(rds_ib_devices_lock); |
ec16227e1 RDS/IB: Infiniban... |
62 |
struct list_head rds_ib_devices; |
745cbccac RDS: Rewrite conn... |
63 |
/* NOTE: if also grabbing ibdev lock, grab this first */ |
ec16227e1 RDS/IB: Infiniban... |
64 65 |
DEFINE_SPINLOCK(ib_nodev_conns_lock); LIST_HEAD(ib_nodev_conns); |
ff51bf841 rds: make local f... |
66 |
static void rds_ib_nodev_connect(void) |
fc19de38b RDS/IB: disconnec... |
67 68 69 70 71 72 73 74 |
{ struct rds_ib_connection *ic; spin_lock(&ib_nodev_conns_lock); list_for_each_entry(ic, &ib_nodev_conns, ib_node) rds_conn_connect_if_down(ic->conn); spin_unlock(&ib_nodev_conns_lock); } |
ff51bf841 rds: make local f... |
75 |
static void rds_ib_dev_shutdown(struct rds_ib_device *rds_ibdev) |
fc19de38b RDS/IB: disconnec... |
76 77 78 79 80 81 82 83 84 |
{ struct rds_ib_connection *ic; unsigned long flags; spin_lock_irqsave(&rds_ibdev->spinlock, flags); list_for_each_entry(ic, &rds_ibdev->conn_list, ib_node) rds_conn_drop(ic->conn); spin_unlock_irqrestore(&rds_ibdev->spinlock, flags); } |
3e0249f9c RDS/IB: add refco... |
85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 |
/* * rds_ib_destroy_mr_pool() blocks on a few things and mrs drop references * from interrupt context so we push freing off into a work struct in krdsd. */ static void rds_ib_dev_free(struct work_struct *work) { struct rds_ib_ipaddr *i_ipaddr, *i_next; struct rds_ib_device *rds_ibdev = container_of(work, struct rds_ib_device, free_work); if (rds_ibdev->mr_pool) rds_ib_destroy_mr_pool(rds_ibdev->mr_pool); if (rds_ibdev->mr) ib_dereg_mr(rds_ibdev->mr); if (rds_ibdev->pd) ib_dealloc_pd(rds_ibdev->pd); list_for_each_entry_safe(i_ipaddr, i_next, &rds_ibdev->ipaddr_list, list) { list_del(&i_ipaddr->list); kfree(i_ipaddr); } kfree(rds_ibdev); } void rds_ib_dev_put(struct rds_ib_device *rds_ibdev) { BUG_ON(atomic_read(&rds_ibdev->refcount) <= 0); if (atomic_dec_and_test(&rds_ibdev->refcount)) queue_work(rds_wq, &rds_ibdev->free_work); } |
ff51bf841 rds: make local f... |
116 |
static void rds_ib_add_one(struct ib_device *device) |
ec16227e1 RDS/IB: Infiniban... |
117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 |
{ struct rds_ib_device *rds_ibdev; struct ib_device_attr *dev_attr; /* Only handle IB (no iWARP) devices */ if (device->node_type != RDMA_NODE_IB_CA) return; dev_attr = kmalloc(sizeof *dev_attr, GFP_KERNEL); if (!dev_attr) return; if (ib_query_device(device, dev_attr)) { rdsdebug("Query device failed for %s ", device->name); goto free_attr; } |
3e0249f9c RDS/IB: add refco... |
134 135 |
rds_ibdev = kzalloc_node(sizeof(struct rds_ib_device), GFP_KERNEL, ibdev_to_node(device)); |
ec16227e1 RDS/IB: Infiniban... |
136 137 138 139 |
if (!rds_ibdev) goto free_attr; spin_lock_init(&rds_ibdev->spinlock); |
3e0249f9c RDS/IB: add refco... |
140 141 |
atomic_set(&rds_ibdev->refcount, 1); INIT_WORK(&rds_ibdev->free_work, rds_ib_dev_free); |
ec16227e1 RDS/IB: Infiniban... |
142 143 144 |
rds_ibdev->max_wrs = dev_attr->max_qp_wr; rds_ibdev->max_sge = min(dev_attr->max_sge, RDS_IB_MAX_SGE); |
ec16227e1 RDS/IB: Infiniban... |
145 146 147 148 |
rds_ibdev->fmr_max_remaps = dev_attr->max_map_per_fmr?: 32; rds_ibdev->max_fmrs = dev_attr->max_fmr ? min_t(unsigned int, dev_attr->max_fmr, fmr_pool_size) : fmr_pool_size; |
40589e74f RDS: Base init_de... |
149 150 |
rds_ibdev->max_initiator_depth = dev_attr->max_qp_init_rd_atom; rds_ibdev->max_responder_resources = dev_attr->max_qp_rd_atom; |
ec16227e1 RDS/IB: Infiniban... |
151 152 |
rds_ibdev->dev = device; rds_ibdev->pd = ib_alloc_pd(device); |
3e0249f9c RDS/IB: add refco... |
153 154 155 156 |
if (IS_ERR(rds_ibdev->pd)) { rds_ibdev->pd = NULL; goto put_dev; } |
ec16227e1 RDS/IB: Infiniban... |
157 |
|
3e0249f9c RDS/IB: add refco... |
158 159 160 161 162 |
rds_ibdev->mr = ib_get_dma_mr(rds_ibdev->pd, IB_ACCESS_LOCAL_WRITE); if (IS_ERR(rds_ibdev->mr)) { rds_ibdev->mr = NULL; goto put_dev; } |
ec16227e1 RDS/IB: Infiniban... |
163 164 165 166 |
rds_ibdev->mr_pool = rds_ib_create_mr_pool(rds_ibdev); if (IS_ERR(rds_ibdev->mr_pool)) { rds_ibdev->mr_pool = NULL; |
3e0249f9c RDS/IB: add refco... |
167 |
goto put_dev; |
ec16227e1 RDS/IB: Infiniban... |
168 169 170 171 |
} INIT_LIST_HEAD(&rds_ibdev->ipaddr_list); INIT_LIST_HEAD(&rds_ibdev->conn_list); |
ea819867b RDS/IB: protect t... |
172 173 174 175 |
down_write(&rds_ib_devices_lock); list_add_tail_rcu(&rds_ibdev->list, &rds_ib_devices); up_write(&rds_ib_devices_lock); |
3e0249f9c RDS/IB: add refco... |
176 |
atomic_inc(&rds_ibdev->refcount); |
ec16227e1 RDS/IB: Infiniban... |
177 178 |
ib_set_client_data(device, &rds_ib_client, rds_ibdev); |
3e0249f9c RDS/IB: add refco... |
179 |
atomic_inc(&rds_ibdev->refcount); |
ec16227e1 RDS/IB: Infiniban... |
180 |
|
fc19de38b RDS/IB: disconnec... |
181 |
rds_ib_nodev_connect(); |
3e0249f9c RDS/IB: add refco... |
182 183 |
put_dev: rds_ib_dev_put(rds_ibdev); |
ec16227e1 RDS/IB: Infiniban... |
184 185 186 |
free_attr: kfree(dev_attr); } |
3e0249f9c RDS/IB: add refco... |
187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 |
/* * New connections use this to find the device to associate with the * connection. It's not in the fast path so we're not concerned about the * performance of the IB call. (As of this writing, it uses an interrupt * blocking spinlock to serialize walking a per-device list of all registered * clients.) * * RCU is used to handle incoming connections racing with device teardown. * Rather than use a lock to serialize removal from the client_data and * getting a new reference, we use an RCU grace period. The destruction * path removes the device from client_data and then waits for all RCU * readers to finish. * * A new connection can get NULL from this if its arriving on a * device that is in the process of being removed. */ struct rds_ib_device *rds_ib_get_client_data(struct ib_device *device) { struct rds_ib_device *rds_ibdev; rcu_read_lock(); rds_ibdev = ib_get_client_data(device, &rds_ib_client); if (rds_ibdev) atomic_inc(&rds_ibdev->refcount); rcu_read_unlock(); return rds_ibdev; } /* * The IB stack is letting us know that a device is going away. This can * happen if the underlying HCA driver is removed or if PCI hotplug is removing * the pci function, for example. * * This can be called at any time and can be racing with any other RDS path. */ |
ff51bf841 rds: make local f... |
222 |
static void rds_ib_remove_one(struct ib_device *device) |
ec16227e1 RDS/IB: Infiniban... |
223 224 |
{ struct rds_ib_device *rds_ibdev; |
ec16227e1 RDS/IB: Infiniban... |
225 226 227 228 |
rds_ibdev = ib_get_client_data(device, &rds_ib_client); if (!rds_ibdev) return; |
fc19de38b RDS/IB: disconnec... |
229 |
rds_ib_dev_shutdown(rds_ibdev); |
ec16227e1 RDS/IB: Infiniban... |
230 |
|
ea819867b RDS/IB: protect t... |
231 232 233 234 235 236 |
/* stop connection attempts from getting a reference to this device. */ ib_set_client_data(device, &rds_ib_client, NULL); down_write(&rds_ib_devices_lock); list_del_rcu(&rds_ibdev->list); up_write(&rds_ib_devices_lock); |
3e0249f9c RDS/IB: add refco... |
237 |
/* |
ea819867b RDS/IB: protect t... |
238 239 240 |
* This synchronize rcu is waiting for readers of both the ib * client data and the devices list to finish before we drop * both of those references. |
3e0249f9c RDS/IB: add refco... |
241 |
*/ |
3e0249f9c RDS/IB: add refco... |
242 243 |
synchronize_rcu(); rds_ib_dev_put(rds_ibdev); |
3e0249f9c RDS/IB: add refco... |
244 |
rds_ib_dev_put(rds_ibdev); |
ec16227e1 RDS/IB: Infiniban... |
245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 |
} struct ib_client rds_ib_client = { .name = "rds_ib", .add = rds_ib_add_one, .remove = rds_ib_remove_one }; static int rds_ib_conn_info_visitor(struct rds_connection *conn, void *buffer) { struct rds_info_rdma_connection *iinfo = buffer; struct rds_ib_connection *ic; /* We will only ever look at IB transports */ if (conn->c_trans != &rds_ib_transport) return 0; iinfo->src_addr = conn->c_laddr; iinfo->dst_addr = conn->c_faddr; memset(&iinfo->src_gid, 0, sizeof(iinfo->src_gid)); memset(&iinfo->dst_gid, 0, sizeof(iinfo->dst_gid)); if (rds_conn_state(conn) == RDS_CONN_UP) { struct rds_ib_device *rds_ibdev; struct rdma_dev_addr *dev_addr; ic = conn->c_transport_data; dev_addr = &ic->i_cm_id->route.addr.dev_addr; |
6f8372b69 RDMA/cm: fix loop... |
274 275 |
rdma_addr_get_sgid(dev_addr, (union ib_gid *) &iinfo->src_gid); rdma_addr_get_dgid(dev_addr, (union ib_gid *) &iinfo->dst_gid); |
ec16227e1 RDS/IB: Infiniban... |
276 |
|
3e0249f9c RDS/IB: add refco... |
277 |
rds_ibdev = ic->rds_ibdev; |
ec16227e1 RDS/IB: Infiniban... |
278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 |
iinfo->max_send_wr = ic->i_send_ring.w_nr; iinfo->max_recv_wr = ic->i_recv_ring.w_nr; iinfo->max_send_sge = rds_ibdev->max_sge; rds_ib_get_mr_info(rds_ibdev, iinfo); } return 1; } static void rds_ib_ic_info(struct socket *sock, unsigned int len, struct rds_info_iterator *iter, struct rds_info_lengths *lens) { rds_for_each_conn_info(sock, len, iter, lens, rds_ib_conn_info_visitor, sizeof(struct rds_info_rdma_connection)); } /* * Early RDS/IB was built to only bind to an address if there is an IPoIB * device with that address set. * * If it were me, I'd advocate for something more flexible. Sending and * receiving should be device-agnostic. Transports would try and maintain * connections between peers who have messages queued. Userspace would be * allowed to influence which paths have priority. We could call userspace * asserting this policy "routing". */ static int rds_ib_laddr_check(__be32 addr) { int ret; struct rdma_cm_id *cm_id; struct sockaddr_in sin; /* Create a CMA ID and try to bind it. This catches both * IB and iWARP capable NICs. */ |
b26f9b994 RDMA/cma: Pass QP... |
315 |
cm_id = rdma_create_id(NULL, NULL, RDMA_PS_TCP, IB_QPT_RC); |
94713bab6 ERR_PTR() derefer... |
316 317 |
if (IS_ERR(cm_id)) return PTR_ERR(cm_id); |
ec16227e1 RDS/IB: Infiniban... |
318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 |
memset(&sin, 0, sizeof(sin)); sin.sin_family = AF_INET; sin.sin_addr.s_addr = addr; /* rdma_bind_addr will only succeed for IB & iWARP devices */ ret = rdma_bind_addr(cm_id, (struct sockaddr *)&sin); /* due to this, we will claim to support iWARP devices unless we check node_type. */ if (ret || cm_id->device->node_type != RDMA_NODE_IB_CA) ret = -EADDRNOTAVAIL; rdsdebug("addr %pI4 ret %d node type %d ", &addr, ret, cm_id->device ? cm_id->device->node_type : -1); rdma_destroy_id(cm_id); return ret; } |
24fa163a4 RDS/IB: wait for ... |
339 340 341 342 343 344 |
static void rds_ib_unregister_client(void) { ib_unregister_client(&rds_ib_client); /* wait for rds_ib_dev_free() to complete */ flush_workqueue(rds_wq); } |
ec16227e1 RDS/IB: Infiniban... |
345 346 347 |
void rds_ib_exit(void) { rds_info_deregister_func(RDS_INFO_IB_CONNECTIONS, rds_ib_ic_info); |
24fa163a4 RDS/IB: wait for ... |
348 |
rds_ib_unregister_client(); |
8aeb1ba66 RDS/IB: destroy c... |
349 |
rds_ib_destroy_nodev_conns(); |
ec16227e1 RDS/IB: Infiniban... |
350 351 352 353 354 355 356 357 358 |
rds_ib_sysctl_exit(); rds_ib_recv_exit(); rds_trans_unregister(&rds_ib_transport); } struct rds_transport rds_ib_transport = { .laddr_check = rds_ib_laddr_check, .xmit_complete = rds_ib_xmit_complete, .xmit = rds_ib_xmit, |
ec16227e1 RDS/IB: Infiniban... |
359 |
.xmit_rdma = rds_ib_xmit_rdma, |
15133f6e6 RDS: Implement at... |
360 |
.xmit_atomic = rds_ib_xmit_atomic, |
ec16227e1 RDS/IB: Infiniban... |
361 362 363 364 365 366 |
.recv = rds_ib_recv, .conn_alloc = rds_ib_conn_alloc, .conn_free = rds_ib_conn_free, .conn_connect = rds_ib_conn_connect, .conn_shutdown = rds_ib_conn_shutdown, .inc_copy_to_user = rds_ib_inc_copy_to_user, |
ec16227e1 RDS/IB: Infiniban... |
367 368 369 370 371 372 373 374 375 376 377 378 |
.inc_free = rds_ib_inc_free, .cm_initiate_connect = rds_ib_cm_initiate_connect, .cm_handle_connect = rds_ib_cm_handle_connect, .cm_connect_complete = rds_ib_cm_connect_complete, .stats_info_copy = rds_ib_stats_info_copy, .exit = rds_ib_exit, .get_mr = rds_ib_get_mr, .sync_mr = rds_ib_sync_mr, .free_mr = rds_ib_free_mr, .flush_mrs = rds_ib_flush_mrs, .t_owner = THIS_MODULE, .t_name = "infiniband", |
335776bd6 RDS: Track transp... |
379 |
.t_type = RDS_TRANS_IB |
ec16227e1 RDS/IB: Infiniban... |
380 |
}; |
ef87b7ea3 RDS: remove __ini... |
381 |
int rds_ib_init(void) |
ec16227e1 RDS/IB: Infiniban... |
382 383 384 385 |
{ int ret; INIT_LIST_HEAD(&rds_ib_devices); |
515e079da RDS/IB: create a ... |
386 387 |
ret = ib_register_client(&rds_ib_client); if (ret) |
c534a107e rds/ib: use syste... |
388 |
goto out; |
515e079da RDS/IB: create a ... |
389 |
|
ec16227e1 RDS/IB: Infiniban... |
390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 |
ret = rds_ib_sysctl_init(); if (ret) goto out_ibreg; ret = rds_ib_recv_init(); if (ret) goto out_sysctl; ret = rds_trans_register(&rds_ib_transport); if (ret) goto out_recv; rds_info_register_func(RDS_INFO_IB_CONNECTIONS, rds_ib_ic_info); goto out; out_recv: rds_ib_recv_exit(); out_sysctl: rds_ib_sysctl_exit(); out_ibreg: |
24fa163a4 RDS/IB: wait for ... |
411 |
rds_ib_unregister_client(); |
ec16227e1 RDS/IB: Infiniban... |
412 413 414 415 416 |
out: return ret; } MODULE_LICENSE("GPL"); |