Commit | Line | Data |
---|---|---|
d7e09d03 PT |
1 | /* |
2 | * GPL HEADER START | |
3 | * | |
4 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License version 2 only, | |
8 | * as published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, but | |
11 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
13 | * General Public License version 2 for more details (a copy is included | |
14 | * in the LICENSE file that accompanied this code). | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * version 2 along with this program; If not, see | |
18 | * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf | |
19 | * | |
20 | * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
21 | * CA 95054 USA or visit www.sun.com if you need additional information or | |
22 | * have any questions. | |
23 | * | |
24 | * GPL HEADER END | |
25 | */ | |
26 | /* | |
27 | * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. | |
28 | * Use is subject to license terms. | |
29 | * | |
30 | * Copyright (c) 2011, 2012, Intel Corporation. | |
31 | */ | |
32 | /* | |
33 | * This file is part of Lustre, http://www.lustre.org/ | |
34 | * Lustre is a trademark of Sun Microsystems, Inc. | |
35 | * | |
36 | * lnet/klnds/o2iblnd/o2iblnd.h | |
37 | * | |
38 | * Author: Eric Barton <eric@bartonsoftware.com> | |
39 | */ | |
40 | ||
41 | #include <linux/module.h> | |
42 | #include <linux/kernel.h> | |
43 | #include <linux/mm.h> | |
44 | #include <linux/string.h> | |
45 | #include <linux/stat.h> | |
46 | #include <linux/errno.h> | |
47 | #include <linux/unistd.h> | |
48 | #include <linux/uio.h> | |
49 | ||
50 | #include <asm/uaccess.h> | |
51 | #include <asm/io.h> | |
52 | ||
d7e09d03 PT |
53 | #include <linux/fs.h> |
54 | #include <linux/file.h> | |
d7e09d03 PT |
55 | #include <linux/list.h> |
56 | #include <linux/kmod.h> | |
57 | #include <linux/sysctl.h> | |
58 | #include <linux/pci.h> | |
59 | ||
60 | #include <net/sock.h> | |
61 | #include <linux/in.h> | |
62 | ||
63 | #define DEBUG_SUBSYSTEM S_LND | |
64 | ||
65 | #include <linux/libcfs/libcfs.h> | |
66 | #include <linux/lnet/lnet.h> | |
67 | #include <linux/lnet/lib-lnet.h> | |
68 | #include <linux/lnet/lnet-sysctl.h> | |
69 | ||
70 | #include <rdma/rdma_cm.h> | |
71 | #include <rdma/ib_cm.h> | |
72 | #include <rdma/ib_verbs.h> | |
73 | #include <rdma/ib_fmr_pool.h> | |
74 | ||
75 | #define IBLND_PEER_HASH_SIZE 101 /* # peer lists */ | |
76 | /* # scheduler loops before reschedule */ | |
77 | #define IBLND_RESCHED 100 | |
78 | ||
79 | #define IBLND_N_SCHED 2 | |
80 | #define IBLND_N_SCHED_HIGH 4 | |
81 | ||
82 | typedef struct | |
83 | { | |
84 | int *kib_dev_failover; /* HCA failover */ | |
85 | unsigned int *kib_service; /* IB service number */ | |
86 | int *kib_min_reconnect_interval; /* first failed connection retry... */ | |
87 | int *kib_max_reconnect_interval; /* ...exponentially increasing to this */ | |
88 | int *kib_cksum; /* checksum kib_msg_t? */ | |
89 | int *kib_timeout; /* comms timeout (seconds) */ | |
90 | int *kib_keepalive; /* keepalive timeout (seconds) */ | |
91 | int *kib_ntx; /* # tx descs */ | |
92 | int *kib_credits; /* # concurrent sends */ | |
93 | int *kib_peertxcredits; /* # concurrent sends to 1 peer */ | |
94 | int *kib_peerrtrcredits; /* # per-peer router buffer credits */ | |
95 | int *kib_peercredits_hiw; /* # when eagerly to return credits */ | |
96 | int *kib_peertimeout; /* seconds to consider peer dead */ | |
97 | char **kib_default_ipif; /* default IPoIB interface */ | |
98 | int *kib_retry_count; | |
99 | int *kib_rnr_retry_count; | |
100 | int *kib_concurrent_sends; /* send work queue sizing */ | |
101 | int *kib_ib_mtu; /* IB MTU */ | |
102 | int *kib_map_on_demand; /* map-on-demand if RD has more fragments | |
103 | * than this value, 0 disable map-on-demand */ | |
104 | int *kib_pmr_pool_size; /* # physical MR in pool */ | |
105 | int *kib_fmr_pool_size; /* # FMRs in pool */ | |
106 | int *kib_fmr_flush_trigger; /* When to trigger FMR flush */ | |
107 | int *kib_fmr_cache; /* enable FMR pool cache? */ | |
d7e09d03 PT |
108 | int *kib_require_priv_port;/* accept only privileged ports */ |
109 | int *kib_use_priv_port; /* use privileged port for active connect */ | |
110 | /* # threads on each CPT */ | |
111 | int *kib_nscheds; | |
112 | } kib_tunables_t; | |
113 | ||
114 | extern kib_tunables_t kiblnd_tunables; | |
115 | ||
116 | #define IBLND_MSG_QUEUE_SIZE_V1 8 /* V1 only : # messages/RDMAs in-flight */ | |
117 | #define IBLND_CREDIT_HIGHWATER_V1 7 /* V1 only : when eagerly to return credits */ | |
118 | ||
119 | #define IBLND_CREDITS_DEFAULT 8 /* default # of peer credits */ | |
120 | #define IBLND_CREDITS_MAX ((typeof(((kib_msg_t*) 0)->ibm_credits)) - 1) /* Max # of peer credits */ | |
121 | ||
122 | #define IBLND_MSG_QUEUE_SIZE(v) ((v) == IBLND_MSG_VERSION_1 ? \ | |
123 | IBLND_MSG_QUEUE_SIZE_V1 : \ | |
124 | *kiblnd_tunables.kib_peertxcredits) /* # messages/RDMAs in-flight */ | |
125 | #define IBLND_CREDITS_HIGHWATER(v) ((v) == IBLND_MSG_VERSION_1 ? \ | |
126 | IBLND_CREDIT_HIGHWATER_V1 : \ | |
127 | *kiblnd_tunables.kib_peercredits_hiw) /* when eagerly to return credits */ | |
128 | ||
129 | #define kiblnd_rdma_create_id(cb, dev, ps, qpt) rdma_create_id(cb, dev, ps, qpt) | |
130 | ||
131 | static inline int | |
132 | kiblnd_concurrent_sends_v1(void) | |
133 | { | |
134 | if (*kiblnd_tunables.kib_concurrent_sends > IBLND_MSG_QUEUE_SIZE_V1 * 2) | |
135 | return IBLND_MSG_QUEUE_SIZE_V1 * 2; | |
136 | ||
137 | if (*kiblnd_tunables.kib_concurrent_sends < IBLND_MSG_QUEUE_SIZE_V1 / 2) | |
138 | return IBLND_MSG_QUEUE_SIZE_V1 / 2; | |
139 | ||
140 | return *kiblnd_tunables.kib_concurrent_sends; | |
141 | } | |
142 | ||
143 | #define IBLND_CONCURRENT_SENDS(v) ((v) == IBLND_MSG_VERSION_1 ? \ | |
144 | kiblnd_concurrent_sends_v1() : \ | |
145 | *kiblnd_tunables.kib_concurrent_sends) | |
146 | /* 2 OOB shall suffice for 1 keepalive and 1 returning credits */ | |
147 | #define IBLND_OOB_CAPABLE(v) ((v) != IBLND_MSG_VERSION_1) | |
148 | #define IBLND_OOB_MSGS(v) (IBLND_OOB_CAPABLE(v) ? 2 : 0) | |
149 | ||
150 | #define IBLND_MSG_SIZE (4<<10) /* max size of queued messages (inc hdr) */ | |
151 | #define IBLND_MAX_RDMA_FRAGS LNET_MAX_IOV /* max # of fragments supported */ | |
152 | #define IBLND_CFG_RDMA_FRAGS (*kiblnd_tunables.kib_map_on_demand != 0 ? \ | |
153 | *kiblnd_tunables.kib_map_on_demand : \ | |
154 | IBLND_MAX_RDMA_FRAGS) /* max # of fragments configured by user */ | |
155 | #define IBLND_RDMA_FRAGS(v) ((v) == IBLND_MSG_VERSION_1 ? \ | |
156 | IBLND_MAX_RDMA_FRAGS : IBLND_CFG_RDMA_FRAGS) | |
157 | ||
158 | /************************/ | |
159 | /* derived constants... */ | |
160 | /* Pools (shared by connections on each CPT) */ | |
161 | /* These pools can grow at runtime, so don't need give a very large value */ | |
162 | #define IBLND_TX_POOL 256 | |
163 | #define IBLND_PMR_POOL 256 | |
164 | #define IBLND_FMR_POOL 256 | |
165 | #define IBLND_FMR_POOL_FLUSH 192 | |
166 | ||
167 | /* TX messages (shared by all connections) */ | |
168 | #define IBLND_TX_MSGS() (*kiblnd_tunables.kib_ntx) | |
169 | ||
170 | /* RX messages (per connection) */ | |
171 | #define IBLND_RX_MSGS(v) (IBLND_MSG_QUEUE_SIZE(v) * 2 + IBLND_OOB_MSGS(v)) | |
172 | #define IBLND_RX_MSG_BYTES(v) (IBLND_RX_MSGS(v) * IBLND_MSG_SIZE) | |
173 | #define IBLND_RX_MSG_PAGES(v) ((IBLND_RX_MSG_BYTES(v) + PAGE_SIZE - 1) / PAGE_SIZE) | |
174 | ||
175 | /* WRs and CQEs (per connection) */ | |
176 | #define IBLND_RECV_WRS(v) IBLND_RX_MSGS(v) | |
177 | #define IBLND_SEND_WRS(v) ((IBLND_RDMA_FRAGS(v) + 1) * IBLND_CONCURRENT_SENDS(v)) | |
178 | #define IBLND_CQ_ENTRIES(v) (IBLND_RECV_WRS(v) + IBLND_SEND_WRS(v)) | |
179 | ||
180 | struct kib_hca_dev; | |
181 | ||
182 | /* o2iblnd can run over aliased interface */ | |
183 | #ifdef IFALIASZ | |
184 | #define KIB_IFNAME_SIZE IFALIASZ | |
185 | #else | |
186 | #define KIB_IFNAME_SIZE 256 | |
187 | #endif | |
188 | ||
189 | typedef struct | |
190 | { | |
191 | struct list_head ibd_list; /* chain on kib_devs */ | |
192 | struct list_head ibd_fail_list; /* chain on kib_failed_devs */ | |
193 | __u32 ibd_ifip; /* IPoIB interface IP */ | |
194 | /** IPoIB interface name */ | |
195 | char ibd_ifname[KIB_IFNAME_SIZE]; | |
196 | int ibd_nnets; /* # nets extant */ | |
197 | ||
198 | cfs_time_t ibd_next_failover; | |
199 | int ibd_failed_failover; /* # failover failures */ | |
200 | unsigned int ibd_failover; /* failover in progress */ | |
201 | unsigned int ibd_can_failover; /* IPoIB interface is a bonding master */ | |
202 | struct list_head ibd_nets; | |
203 | struct kib_hca_dev *ibd_hdev; | |
204 | } kib_dev_t; | |
205 | ||
206 | typedef struct kib_hca_dev | |
207 | { | |
208 | struct rdma_cm_id *ibh_cmid; /* listener cmid */ | |
209 | struct ib_device *ibh_ibdev; /* IB device */ | |
210 | int ibh_page_shift; /* page shift of current HCA */ | |
211 | int ibh_page_size; /* page size of current HCA */ | |
212 | __u64 ibh_page_mask; /* page mask of current HCA */ | |
213 | int ibh_mr_shift; /* bits shift of max MR size */ | |
214 | __u64 ibh_mr_size; /* size of MR */ | |
215 | int ibh_nmrs; /* # of global MRs */ | |
216 | struct ib_mr **ibh_mrs; /* global MR */ | |
217 | struct ib_pd *ibh_pd; /* PD */ | |
218 | kib_dev_t *ibh_dev; /* owner */ | |
219 | atomic_t ibh_ref; /* refcount */ | |
220 | } kib_hca_dev_t; | |
221 | ||
222 | /** # of seconds to keep pool alive */ | |
223 | #define IBLND_POOL_DEADLINE 300 | |
224 | /** # of seconds to retry if allocation failed */ | |
225 | #define IBLND_POOL_RETRY 1 | |
226 | ||
227 | typedef struct | |
228 | { | |
229 | int ibp_npages; /* # pages */ | |
230 | struct page *ibp_pages[0]; /* page array */ | |
231 | } kib_pages_t; | |
232 | ||
233 | struct kib_pmr_pool; | |
234 | ||
235 | typedef struct { | |
236 | struct list_head pmr_list; /* chain node */ | |
237 | struct ib_phys_buf *pmr_ipb; /* physical buffer */ | |
238 | struct ib_mr *pmr_mr; /* IB MR */ | |
239 | struct kib_pmr_pool *pmr_pool; /* owner of this MR */ | |
240 | __u64 pmr_iova; /* Virtual I/O address */ | |
241 | int pmr_refcount; /* reference count */ | |
242 | } kib_phys_mr_t; | |
243 | ||
244 | struct kib_pool; | |
245 | struct kib_poolset; | |
246 | ||
247 | typedef int (*kib_ps_pool_create_t)(struct kib_poolset *ps, | |
248 | int inc, struct kib_pool **pp_po); | |
249 | typedef void (*kib_ps_pool_destroy_t)(struct kib_pool *po); | |
250 | typedef void (*kib_ps_node_init_t)(struct kib_pool *po, struct list_head *node); | |
251 | typedef void (*kib_ps_node_fini_t)(struct kib_pool *po, struct list_head *node); | |
252 | ||
253 | struct kib_net; | |
254 | ||
255 | #define IBLND_POOL_NAME_LEN 32 | |
256 | ||
257 | typedef struct kib_poolset | |
258 | { | |
259 | spinlock_t ps_lock; /* serialize */ | |
260 | struct kib_net *ps_net; /* network it belongs to */ | |
261 | char ps_name[IBLND_POOL_NAME_LEN]; /* pool set name */ | |
262 | struct list_head ps_pool_list; /* list of pools */ | |
263 | struct list_head ps_failed_pool_list; /* failed pool list */ | |
264 | cfs_time_t ps_next_retry; /* time stamp for retry if failed to allocate */ | |
265 | int ps_increasing; /* is allocating new pool */ | |
266 | int ps_pool_size; /* new pool size */ | |
267 | int ps_cpt; /* CPT id */ | |
268 | ||
269 | kib_ps_pool_create_t ps_pool_create; /* create a new pool */ | |
270 | kib_ps_pool_destroy_t ps_pool_destroy; /* destroy a pool */ | |
271 | kib_ps_node_init_t ps_node_init; /* initialize new allocated node */ | |
272 | kib_ps_node_fini_t ps_node_fini; /* finalize node */ | |
273 | } kib_poolset_t; | |
274 | ||
275 | typedef struct kib_pool | |
276 | { | |
277 | struct list_head po_list; /* chain on pool list */ | |
278 | struct list_head po_free_list; /* pre-allocated node */ | |
279 | kib_poolset_t *po_owner; /* pool_set of this pool */ | |
280 | cfs_time_t po_deadline; /* deadline of this pool */ | |
281 | int po_allocated; /* # of elements in use */ | |
282 | int po_failed; /* pool is created on failed HCA */ | |
283 | int po_size; /* # of pre-allocated elements */ | |
284 | } kib_pool_t; | |
285 | ||
286 | typedef struct { | |
287 | kib_poolset_t tps_poolset; /* pool-set */ | |
288 | __u64 tps_next_tx_cookie; /* cookie of TX */ | |
289 | } kib_tx_poolset_t; | |
290 | ||
291 | typedef struct { | |
292 | kib_pool_t tpo_pool; /* pool */ | |
293 | struct kib_hca_dev *tpo_hdev; /* device for this pool */ | |
294 | struct kib_tx *tpo_tx_descs; /* all the tx descriptors */ | |
295 | kib_pages_t *tpo_tx_pages; /* premapped tx msg pages */ | |
296 | } kib_tx_pool_t; | |
297 | ||
298 | typedef struct { | |
299 | kib_poolset_t pps_poolset; /* pool-set */ | |
300 | } kib_pmr_poolset_t; | |
301 | ||
302 | typedef struct kib_pmr_pool { | |
303 | struct kib_hca_dev *ppo_hdev; /* device for this pool */ | |
304 | kib_pool_t ppo_pool; /* pool */ | |
305 | } kib_pmr_pool_t; | |
306 | ||
307 | typedef struct | |
308 | { | |
309 | spinlock_t fps_lock; /* serialize */ | |
310 | struct kib_net *fps_net; /* IB network */ | |
311 | struct list_head fps_pool_list; /* FMR pool list */ | |
312 | struct list_head fps_failed_pool_list; /* FMR pool list */ | |
313 | __u64 fps_version; /* validity stamp */ | |
314 | int fps_cpt; /* CPT id */ | |
315 | int fps_pool_size; | |
316 | int fps_flush_trigger; | |
317 | /* is allocating new pool */ | |
318 | int fps_increasing; | |
319 | /* time stamp for retry if failed to allocate */ | |
320 | cfs_time_t fps_next_retry; | |
321 | } kib_fmr_poolset_t; | |
322 | ||
323 | typedef struct | |
324 | { | |
325 | struct list_head fpo_list; /* chain on pool list */ | |
326 | struct kib_hca_dev *fpo_hdev; /* device for this pool */ | |
327 | kib_fmr_poolset_t *fpo_owner; /* owner of this pool */ | |
328 | struct ib_fmr_pool *fpo_fmr_pool; /* IB FMR pool */ | |
329 | cfs_time_t fpo_deadline; /* deadline of this pool */ | |
330 | int fpo_failed; /* fmr pool is failed */ | |
331 | int fpo_map_count; /* # of mapped FMR */ | |
332 | } kib_fmr_pool_t; | |
333 | ||
334 | typedef struct { | |
335 | struct ib_pool_fmr *fmr_pfmr; /* IB pool fmr */ | |
336 | kib_fmr_pool_t *fmr_pool; /* pool of FMR */ | |
337 | } kib_fmr_t; | |
338 | ||
339 | typedef struct kib_net | |
340 | { | |
341 | struct list_head ibn_list; /* chain on kib_dev_t::ibd_nets */ | |
342 | __u64 ibn_incarnation; /* my epoch */ | |
343 | int ibn_init; /* initialisation state */ | |
344 | int ibn_shutdown; /* shutting down? */ | |
345 | ||
346 | atomic_t ibn_npeers; /* # peers extant */ | |
347 | atomic_t ibn_nconns; /* # connections extant */ | |
348 | ||
349 | kib_tx_poolset_t **ibn_tx_ps; /* tx pool-set */ | |
350 | kib_fmr_poolset_t **ibn_fmr_ps; /* fmr pool-set */ | |
351 | kib_pmr_poolset_t **ibn_pmr_ps; /* pmr pool-set */ | |
352 | ||
353 | kib_dev_t *ibn_dev; /* underlying IB device */ | |
354 | } kib_net_t; | |
355 | ||
356 | #define KIB_THREAD_SHIFT 16 | |
357 | #define KIB_THREAD_ID(cpt, tid) ((cpt) << KIB_THREAD_SHIFT | (tid)) | |
358 | #define KIB_THREAD_CPT(id) ((id) >> KIB_THREAD_SHIFT) | |
359 | #define KIB_THREAD_TID(id) ((id) & ((1UL << KIB_THREAD_SHIFT) - 1)) | |
360 | ||
361 | struct kib_sched_info { | |
362 | /* serialise */ | |
363 | spinlock_t ibs_lock; | |
364 | /* schedulers sleep here */ | |
365 | wait_queue_head_t ibs_waitq; | |
366 | /* conns to check for rx completions */ | |
367 | struct list_head ibs_conns; | |
368 | /* number of scheduler threads */ | |
369 | int ibs_nthreads; | |
370 | /* max allowed scheduler threads */ | |
371 | int ibs_nthreads_max; | |
372 | int ibs_cpt; /* CPT id */ | |
373 | }; | |
374 | ||
375 | typedef struct | |
376 | { | |
377 | int kib_init; /* initialisation state */ | |
378 | int kib_shutdown; /* shut down? */ | |
379 | struct list_head kib_devs; /* IB devices extant */ | |
380 | /* list head of failed devices */ | |
381 | struct list_head kib_failed_devs; | |
382 | /* schedulers sleep here */ | |
383 | wait_queue_head_t kib_failover_waitq; | |
384 | atomic_t kib_nthreads; /* # live threads */ | |
385 | /* stabilize net/dev/peer/conn ops */ | |
386 | rwlock_t kib_global_lock; | |
387 | /* hash table of all my known peers */ | |
388 | struct list_head *kib_peers; | |
389 | /* size of kib_peers */ | |
390 | int kib_peer_hash_size; | |
391 | /* the connd task (serialisation assertions) */ | |
392 | void *kib_connd; | |
393 | /* connections to setup/teardown */ | |
394 | struct list_head kib_connd_conns; | |
395 | /* connections with zero refcount */ | |
396 | struct list_head kib_connd_zombies; | |
397 | /* connection daemon sleeps here */ | |
398 | wait_queue_head_t kib_connd_waitq; | |
399 | spinlock_t kib_connd_lock; /* serialise */ | |
400 | struct ib_qp_attr kib_error_qpa; /* QP->ERROR */ | |
401 | /* percpt data for schedulers */ | |
402 | struct kib_sched_info **kib_scheds; | |
403 | } kib_data_t; | |
404 | ||
405 | #define IBLND_INIT_NOTHING 0 | |
406 | #define IBLND_INIT_DATA 1 | |
407 | #define IBLND_INIT_ALL 2 | |
408 | ||
409 | /************************************************************************ | |
410 | * IB Wire message format. | |
411 | * These are sent in sender's byte order (i.e. receiver flips). | |
412 | */ | |
413 | ||
414 | typedef struct kib_connparams | |
415 | { | |
416 | __u16 ibcp_queue_depth; | |
417 | __u16 ibcp_max_frags; | |
418 | __u32 ibcp_max_msg_size; | |
419 | } WIRE_ATTR kib_connparams_t; | |
420 | ||
421 | typedef struct | |
422 | { | |
423 | lnet_hdr_t ibim_hdr; /* portals header */ | |
424 | char ibim_payload[0]; /* piggy-backed payload */ | |
425 | } WIRE_ATTR kib_immediate_msg_t; | |
426 | ||
427 | typedef struct | |
428 | { | |
429 | __u32 rf_nob; /* # bytes this frag */ | |
430 | __u64 rf_addr; /* CAVEAT EMPTOR: misaligned!! */ | |
431 | } WIRE_ATTR kib_rdma_frag_t; | |
432 | ||
433 | typedef struct | |
434 | { | |
435 | __u32 rd_key; /* local/remote key */ | |
436 | __u32 rd_nfrags; /* # fragments */ | |
437 | kib_rdma_frag_t rd_frags[0]; /* buffer frags */ | |
438 | } WIRE_ATTR kib_rdma_desc_t; | |
439 | ||
440 | typedef struct | |
441 | { | |
442 | lnet_hdr_t ibprm_hdr; /* portals header */ | |
443 | __u64 ibprm_cookie; /* opaque completion cookie */ | |
444 | } WIRE_ATTR kib_putreq_msg_t; | |
445 | ||
446 | typedef struct | |
447 | { | |
448 | __u64 ibpam_src_cookie; /* reflected completion cookie */ | |
449 | __u64 ibpam_dst_cookie; /* opaque completion cookie */ | |
450 | kib_rdma_desc_t ibpam_rd; /* sender's sink buffer */ | |
451 | } WIRE_ATTR kib_putack_msg_t; | |
452 | ||
453 | typedef struct | |
454 | { | |
455 | lnet_hdr_t ibgm_hdr; /* portals header */ | |
456 | __u64 ibgm_cookie; /* opaque completion cookie */ | |
457 | kib_rdma_desc_t ibgm_rd; /* rdma descriptor */ | |
458 | } WIRE_ATTR kib_get_msg_t; | |
459 | ||
460 | typedef struct | |
461 | { | |
462 | __u64 ibcm_cookie; /* opaque completion cookie */ | |
463 | __s32 ibcm_status; /* < 0 failure: >= 0 length */ | |
464 | } WIRE_ATTR kib_completion_msg_t; | |
465 | ||
466 | typedef struct | |
467 | { | |
468 | /* First 2 fields fixed FOR ALL TIME */ | |
469 | __u32 ibm_magic; /* I'm an ibnal message */ | |
470 | __u16 ibm_version; /* this is my version number */ | |
471 | ||
472 | __u8 ibm_type; /* msg type */ | |
473 | __u8 ibm_credits; /* returned credits */ | |
474 | __u32 ibm_nob; /* # bytes in whole message */ | |
475 | __u32 ibm_cksum; /* checksum (0 == no checksum) */ | |
476 | __u64 ibm_srcnid; /* sender's NID */ | |
477 | __u64 ibm_srcstamp; /* sender's incarnation */ | |
478 | __u64 ibm_dstnid; /* destination's NID */ | |
479 | __u64 ibm_dststamp; /* destination's incarnation */ | |
480 | ||
481 | union { | |
482 | kib_connparams_t connparams; | |
483 | kib_immediate_msg_t immediate; | |
484 | kib_putreq_msg_t putreq; | |
485 | kib_putack_msg_t putack; | |
486 | kib_get_msg_t get; | |
487 | kib_completion_msg_t completion; | |
488 | } WIRE_ATTR ibm_u; | |
489 | } WIRE_ATTR kib_msg_t; | |
490 | ||
491 | #define IBLND_MSG_MAGIC LNET_PROTO_IB_MAGIC /* unique magic */ | |
492 | ||
493 | #define IBLND_MSG_VERSION_1 0x11 | |
494 | #define IBLND_MSG_VERSION_2 0x12 | |
495 | #define IBLND_MSG_VERSION IBLND_MSG_VERSION_2 | |
496 | ||
497 | #define IBLND_MSG_CONNREQ 0xc0 /* connection request */ | |
498 | #define IBLND_MSG_CONNACK 0xc1 /* connection acknowledge */ | |
499 | #define IBLND_MSG_NOOP 0xd0 /* nothing (just credits) */ | |
500 | #define IBLND_MSG_IMMEDIATE 0xd1 /* immediate */ | |
501 | #define IBLND_MSG_PUT_REQ 0xd2 /* putreq (src->sink) */ | |
502 | #define IBLND_MSG_PUT_NAK 0xd3 /* completion (sink->src) */ | |
503 | #define IBLND_MSG_PUT_ACK 0xd4 /* putack (sink->src) */ | |
504 | #define IBLND_MSG_PUT_DONE 0xd5 /* completion (src->sink) */ | |
505 | #define IBLND_MSG_GET_REQ 0xd6 /* getreq (sink->src) */ | |
506 | #define IBLND_MSG_GET_DONE 0xd7 /* completion (src->sink: all OK) */ | |
507 | ||
508 | typedef struct { | |
509 | __u32 ibr_magic; /* sender's magic */ | |
510 | __u16 ibr_version; /* sender's version */ | |
511 | __u8 ibr_why; /* reject reason */ | |
512 | __u8 ibr_padding; /* padding */ | |
513 | __u64 ibr_incarnation; /* incarnation of peer */ | |
514 | kib_connparams_t ibr_cp; /* connection parameters */ | |
515 | } WIRE_ATTR kib_rej_t; | |
516 | ||
517 | /* connection rejection reasons */ | |
518 | #define IBLND_REJECT_CONN_RACE 1 /* You lost connection race */ | |
519 | #define IBLND_REJECT_NO_RESOURCES 2 /* Out of memory/conns etc */ | |
520 | #define IBLND_REJECT_FATAL 3 /* Anything else */ | |
521 | ||
522 | #define IBLND_REJECT_CONN_UNCOMPAT 4 /* incompatible version peer */ | |
523 | #define IBLND_REJECT_CONN_STALE 5 /* stale peer */ | |
524 | ||
525 | #define IBLND_REJECT_RDMA_FRAGS 6 /* Fatal: peer's rdma frags can't match mine */ | |
526 | #define IBLND_REJECT_MSG_QUEUE_SIZE 7 /* Fatal: peer's msg queue size can't match mine */ | |
527 | ||
528 | /***********************************************************************/ | |
529 | ||
530 | typedef struct kib_rx /* receive message */ | |
531 | { | |
532 | struct list_head rx_list; /* queue for attention */ | |
533 | struct kib_conn *rx_conn; /* owning conn */ | |
534 | int rx_nob; /* # bytes received (-1 while posted) */ | |
535 | enum ib_wc_status rx_status; /* completion status */ | |
536 | kib_msg_t *rx_msg; /* message buffer (host vaddr) */ | |
537 | __u64 rx_msgaddr; /* message buffer (I/O addr) */ | |
538 | DECLARE_PCI_UNMAP_ADDR (rx_msgunmap); /* for dma_unmap_single() */ | |
539 | struct ib_recv_wr rx_wrq; /* receive work item... */ | |
540 | struct ib_sge rx_sge; /* ...and its memory */ | |
541 | } kib_rx_t; | |
542 | ||
543 | #define IBLND_POSTRX_DONT_POST 0 /* don't post */ | |
544 | #define IBLND_POSTRX_NO_CREDIT 1 /* post: no credits */ | |
545 | #define IBLND_POSTRX_PEER_CREDIT 2 /* post: give peer back 1 credit */ | |
546 | #define IBLND_POSTRX_RSRVD_CREDIT 3 /* post: give myself back 1 reserved credit */ | |
547 | ||
548 | typedef struct kib_tx /* transmit message */ | |
549 | { | |
550 | struct list_head tx_list; /* queue on idle_txs ibc_tx_queue etc. */ | |
551 | kib_tx_pool_t *tx_pool; /* pool I'm from */ | |
552 | struct kib_conn *tx_conn; /* owning conn */ | |
553 | short tx_sending; /* # tx callbacks outstanding */ | |
554 | short tx_queued; /* queued for sending */ | |
555 | short tx_waiting; /* waiting for peer */ | |
556 | int tx_status; /* LNET completion status */ | |
557 | unsigned long tx_deadline; /* completion deadline */ | |
558 | __u64 tx_cookie; /* completion cookie */ | |
559 | lnet_msg_t *tx_lntmsg[2]; /* lnet msgs to finalize on completion */ | |
560 | kib_msg_t *tx_msg; /* message buffer (host vaddr) */ | |
561 | __u64 tx_msgaddr; /* message buffer (I/O addr) */ | |
562 | DECLARE_PCI_UNMAP_ADDR (tx_msgunmap); /* for dma_unmap_single() */ | |
563 | int tx_nwrq; /* # send work items */ | |
564 | struct ib_send_wr *tx_wrq; /* send work items... */ | |
565 | struct ib_sge *tx_sge; /* ...and their memory */ | |
566 | kib_rdma_desc_t *tx_rd; /* rdma descriptor */ | |
567 | int tx_nfrags; /* # entries in... */ | |
568 | struct scatterlist *tx_frags; /* dma_map_sg descriptor */ | |
569 | __u64 *tx_pages; /* rdma phys page addrs */ | |
570 | union { | |
571 | kib_phys_mr_t *pmr; /* MR for physical buffer */ | |
572 | kib_fmr_t fmr; /* FMR */ | |
573 | } tx_u; | |
574 | int tx_dmadir; /* dma direction */ | |
575 | } kib_tx_t; | |
576 | ||
577 | typedef struct kib_connvars | |
578 | { | |
579 | /* connection-in-progress variables */ | |
580 | kib_msg_t cv_msg; | |
581 | } kib_connvars_t; | |
582 | ||
583 | typedef struct kib_conn | |
584 | { | |
585 | struct kib_sched_info *ibc_sched; /* scheduler information */ | |
586 | struct kib_peer *ibc_peer; /* owning peer */ | |
587 | kib_hca_dev_t *ibc_hdev; /* HCA bound on */ | |
588 | struct list_head ibc_list; /* stash on peer's conn list */ | |
589 | struct list_head ibc_sched_list; /* schedule for attention */ | |
590 | __u16 ibc_version; /* version of connection */ | |
591 | __u64 ibc_incarnation; /* which instance of the peer */ | |
592 | atomic_t ibc_refcount; /* # users */ | |
593 | int ibc_state; /* what's happening */ | |
594 | int ibc_nsends_posted; /* # uncompleted sends */ | |
595 | int ibc_noops_posted; /* # uncompleted NOOPs */ | |
596 | int ibc_credits; /* # credits I have */ | |
597 | int ibc_outstanding_credits; /* # credits to return */ | |
598 | int ibc_reserved_credits;/* # ACK/DONE msg credits */ | |
599 | int ibc_comms_error; /* set on comms error */ | |
600 | unsigned int ibc_nrx:16; /* receive buffers owned */ | |
601 | unsigned int ibc_scheduled:1; /* scheduled for attention */ | |
602 | unsigned int ibc_ready:1; /* CQ callback fired */ | |
603 | /* time of last send */ | |
604 | unsigned long ibc_last_send; | |
605 | /** link chain for kiblnd_check_conns only */ | |
606 | struct list_head ibc_connd_list; | |
607 | /** rxs completed before ESTABLISHED */ | |
608 | struct list_head ibc_early_rxs; | |
609 | /** IBLND_MSG_NOOPs for IBLND_MSG_VERSION_1 */ | |
610 | struct list_head ibc_tx_noops; | |
611 | struct list_head ibc_tx_queue; /* sends that need a credit */ | |
612 | struct list_head ibc_tx_queue_nocred;/* sends that don't need a credit */ | |
613 | struct list_head ibc_tx_queue_rsrvd; /* sends that need to reserve an ACK/DONE msg */ | |
614 | struct list_head ibc_active_txs; /* active tx awaiting completion */ | |
615 | spinlock_t ibc_lock; /* serialise */ | |
616 | kib_rx_t *ibc_rxs; /* the rx descs */ | |
617 | kib_pages_t *ibc_rx_pages; /* premapped rx msg pages */ | |
618 | ||
619 | struct rdma_cm_id *ibc_cmid; /* CM id */ | |
620 | struct ib_cq *ibc_cq; /* completion queue */ | |
621 | ||
622 | kib_connvars_t *ibc_connvars; /* in-progress connection state */ | |
623 | } kib_conn_t; | |
624 | ||
625 | #define IBLND_CONN_INIT 0 /* being initialised */ | |
626 | #define IBLND_CONN_ACTIVE_CONNECT 1 /* active sending req */ | |
627 | #define IBLND_CONN_PASSIVE_WAIT 2 /* passive waiting for rtu */ | |
628 | #define IBLND_CONN_ESTABLISHED 3 /* connection established */ | |
629 | #define IBLND_CONN_CLOSING 4 /* being closed */ | |
630 | #define IBLND_CONN_DISCONNECTED 5 /* disconnected */ | |
631 | ||
632 | typedef struct kib_peer | |
633 | { | |
634 | struct list_head ibp_list; /* stash on global peer list */ | |
635 | lnet_nid_t ibp_nid; /* who's on the other end(s) */ | |
636 | lnet_ni_t *ibp_ni; /* LNet interface */ | |
637 | atomic_t ibp_refcount; /* # users */ | |
638 | struct list_head ibp_conns; /* all active connections */ | |
639 | struct list_head ibp_tx_queue; /* msgs waiting for a conn */ | |
640 | __u16 ibp_version; /* version of peer */ | |
641 | __u64 ibp_incarnation; /* incarnation of peer */ | |
642 | int ibp_connecting; /* current active connection attempts */ | |
643 | int ibp_accepting; /* current passive connection attempts */ | |
644 | int ibp_error; /* errno on closing this peer */ | |
645 | cfs_time_t ibp_last_alive; /* when (in jiffies) I was last alive */ | |
646 | } kib_peer_t; | |
647 | ||
648 | extern kib_data_t kiblnd_data; | |
649 | ||
650 | extern void kiblnd_hdev_destroy(kib_hca_dev_t *hdev); | |
651 | ||
652 | static inline void | |
653 | kiblnd_hdev_addref_locked(kib_hca_dev_t *hdev) | |
654 | { | |
655 | LASSERT (atomic_read(&hdev->ibh_ref) > 0); | |
656 | atomic_inc(&hdev->ibh_ref); | |
657 | } | |
658 | ||
659 | static inline void | |
660 | kiblnd_hdev_decref(kib_hca_dev_t *hdev) | |
661 | { | |
662 | LASSERT (atomic_read(&hdev->ibh_ref) > 0); | |
663 | if (atomic_dec_and_test(&hdev->ibh_ref)) | |
664 | kiblnd_hdev_destroy(hdev); | |
665 | } | |
666 | ||
667 | static inline int | |
668 | kiblnd_dev_can_failover(kib_dev_t *dev) | |
669 | { | |
670 | if (!list_empty(&dev->ibd_fail_list)) /* already scheduled */ | |
671 | return 0; | |
672 | ||
673 | if (*kiblnd_tunables.kib_dev_failover == 0) /* disabled */ | |
674 | return 0; | |
675 | ||
676 | if (*kiblnd_tunables.kib_dev_failover > 1) /* force failover */ | |
677 | return 1; | |
678 | ||
679 | return dev->ibd_can_failover; | |
680 | } | |
681 | ||
682 | #define kiblnd_conn_addref(conn) \ | |
683 | do { \ | |
684 | CDEBUG(D_NET, "conn[%p] (%d)++\n", \ | |
685 | (conn), atomic_read(&(conn)->ibc_refcount)); \ | |
686 | atomic_inc(&(conn)->ibc_refcount); \ | |
687 | } while (0) | |
688 | ||
689 | #define kiblnd_conn_decref(conn) \ | |
690 | do { \ | |
691 | unsigned long flags; \ | |
692 | \ | |
693 | CDEBUG(D_NET, "conn[%p] (%d)--\n", \ | |
694 | (conn), atomic_read(&(conn)->ibc_refcount)); \ | |
695 | LASSERT_ATOMIC_POS(&(conn)->ibc_refcount); \ | |
696 | if (atomic_dec_and_test(&(conn)->ibc_refcount)) { \ | |
697 | spin_lock_irqsave(&kiblnd_data.kib_connd_lock, flags); \ | |
698 | list_add_tail(&(conn)->ibc_list, \ | |
699 | &kiblnd_data.kib_connd_zombies); \ | |
700 | wake_up(&kiblnd_data.kib_connd_waitq); \ | |
701 | spin_unlock_irqrestore(&kiblnd_data.kib_connd_lock, flags);\ | |
702 | } \ | |
703 | } while (0) | |
704 | ||
705 | #define kiblnd_peer_addref(peer) \ | |
706 | do { \ | |
707 | CDEBUG(D_NET, "peer[%p] -> %s (%d)++\n", \ | |
708 | (peer), libcfs_nid2str((peer)->ibp_nid), \ | |
709 | atomic_read (&(peer)->ibp_refcount)); \ | |
710 | atomic_inc(&(peer)->ibp_refcount); \ | |
711 | } while (0) | |
712 | ||
713 | #define kiblnd_peer_decref(peer) \ | |
714 | do { \ | |
715 | CDEBUG(D_NET, "peer[%p] -> %s (%d)--\n", \ | |
716 | (peer), libcfs_nid2str((peer)->ibp_nid), \ | |
717 | atomic_read (&(peer)->ibp_refcount)); \ | |
718 | LASSERT_ATOMIC_POS(&(peer)->ibp_refcount); \ | |
719 | if (atomic_dec_and_test(&(peer)->ibp_refcount)) \ | |
720 | kiblnd_destroy_peer(peer); \ | |
721 | } while (0) | |
722 | ||
723 | static inline struct list_head * | |
724 | kiblnd_nid2peerlist (lnet_nid_t nid) | |
725 | { | |
726 | unsigned int hash = | |
727 | ((unsigned int)nid) % kiblnd_data.kib_peer_hash_size; | |
728 | ||
729 | return (&kiblnd_data.kib_peers [hash]); | |
730 | } | |
731 | ||
732 | static inline int | |
733 | kiblnd_peer_active (kib_peer_t *peer) | |
734 | { | |
735 | /* Am I in the peer hash table? */ | |
736 | return (!list_empty(&peer->ibp_list)); | |
737 | } | |
738 | ||
739 | static inline kib_conn_t * | |
740 | kiblnd_get_conn_locked (kib_peer_t *peer) | |
741 | { | |
742 | LASSERT (!list_empty(&peer->ibp_conns)); | |
743 | ||
744 | /* just return the first connection */ | |
745 | return list_entry(peer->ibp_conns.next, kib_conn_t, ibc_list); | |
746 | } | |
747 | ||
748 | static inline int | |
749 | kiblnd_send_keepalive(kib_conn_t *conn) | |
750 | { | |
751 | return (*kiblnd_tunables.kib_keepalive > 0) && | |
752 | cfs_time_after(jiffies, conn->ibc_last_send + | |
753 | *kiblnd_tunables.kib_keepalive*HZ); | |
754 | } | |
755 | ||
756 | static inline int | |
757 | kiblnd_need_noop(kib_conn_t *conn) | |
758 | { | |
759 | LASSERT (conn->ibc_state >= IBLND_CONN_ESTABLISHED); | |
760 | ||
761 | if (conn->ibc_outstanding_credits < | |
762 | IBLND_CREDITS_HIGHWATER(conn->ibc_version) && | |
763 | !kiblnd_send_keepalive(conn)) | |
764 | return 0; /* No need to send NOOP */ | |
765 | ||
766 | if (IBLND_OOB_CAPABLE(conn->ibc_version)) { | |
767 | if (!list_empty(&conn->ibc_tx_queue_nocred)) | |
768 | return 0; /* NOOP can be piggybacked */ | |
769 | ||
770 | /* No tx to piggyback NOOP onto or no credit to send a tx */ | |
771 | return (list_empty(&conn->ibc_tx_queue) || | |
772 | conn->ibc_credits == 0); | |
773 | } | |
774 | ||
775 | if (!list_empty(&conn->ibc_tx_noops) || /* NOOP already queued */ | |
776 | !list_empty(&conn->ibc_tx_queue_nocred) || /* piggyback NOOP */ | |
777 | conn->ibc_credits == 0) /* no credit */ | |
778 | return 0; | |
779 | ||
780 | if (conn->ibc_credits == 1 && /* last credit reserved for */ | |
781 | conn->ibc_outstanding_credits == 0) /* giving back credits */ | |
782 | return 0; | |
783 | ||
784 | /* No tx to piggyback NOOP onto or no credit to send a tx */ | |
785 | return (list_empty(&conn->ibc_tx_queue) || conn->ibc_credits == 1); | |
786 | } | |
787 | ||
788 | static inline void | |
789 | kiblnd_abort_receives(kib_conn_t *conn) | |
790 | { | |
791 | ib_modify_qp(conn->ibc_cmid->qp, | |
792 | &kiblnd_data.kib_error_qpa, IB_QP_STATE); | |
793 | } | |
794 | ||
795 | static inline const char * | |
796 | kiblnd_queue2str (kib_conn_t *conn, struct list_head *q) | |
797 | { | |
798 | if (q == &conn->ibc_tx_queue) | |
799 | return "tx_queue"; | |
800 | ||
801 | if (q == &conn->ibc_tx_queue_rsrvd) | |
802 | return "tx_queue_rsrvd"; | |
803 | ||
804 | if (q == &conn->ibc_tx_queue_nocred) | |
805 | return "tx_queue_nocred"; | |
806 | ||
807 | if (q == &conn->ibc_active_txs) | |
808 | return "active_txs"; | |
809 | ||
810 | LBUG(); | |
811 | return NULL; | |
812 | } | |
813 | ||
814 | /* CAVEAT EMPTOR: We rely on descriptor alignment to allow us to use the | |
815 | * lowest bits of the work request id to stash the work item type. */ | |
816 | ||
817 | #define IBLND_WID_TX 0 | |
818 | #define IBLND_WID_RDMA 1 | |
819 | #define IBLND_WID_RX 2 | |
820 | #define IBLND_WID_MASK 3UL | |
821 | ||
822 | static inline __u64 | |
823 | kiblnd_ptr2wreqid (void *ptr, int type) | |
824 | { | |
825 | unsigned long lptr = (unsigned long)ptr; | |
826 | ||
827 | LASSERT ((lptr & IBLND_WID_MASK) == 0); | |
828 | LASSERT ((type & ~IBLND_WID_MASK) == 0); | |
829 | return (__u64)(lptr | type); | |
830 | } | |
831 | ||
832 | static inline void * | |
833 | kiblnd_wreqid2ptr (__u64 wreqid) | |
834 | { | |
835 | return (void *)(((unsigned long)wreqid) & ~IBLND_WID_MASK); | |
836 | } | |
837 | ||
838 | static inline int | |
839 | kiblnd_wreqid2type (__u64 wreqid) | |
840 | { | |
841 | return (wreqid & IBLND_WID_MASK); | |
842 | } | |
843 | ||
844 | static inline void | |
845 | kiblnd_set_conn_state (kib_conn_t *conn, int state) | |
846 | { | |
847 | conn->ibc_state = state; | |
848 | mb(); | |
849 | } | |
850 | ||
851 | static inline void | |
852 | kiblnd_init_msg (kib_msg_t *msg, int type, int body_nob) | |
853 | { | |
854 | msg->ibm_type = type; | |
855 | msg->ibm_nob = offsetof(kib_msg_t, ibm_u) + body_nob; | |
856 | } | |
857 | ||
858 | static inline int | |
859 | kiblnd_rd_size (kib_rdma_desc_t *rd) | |
860 | { | |
861 | int i; | |
862 | int size; | |
863 | ||
864 | for (i = size = 0; i < rd->rd_nfrags; i++) | |
865 | size += rd->rd_frags[i].rf_nob; | |
866 | ||
867 | return size; | |
868 | } | |
869 | ||
870 | static inline __u64 | |
871 | kiblnd_rd_frag_addr(kib_rdma_desc_t *rd, int index) | |
872 | { | |
873 | return rd->rd_frags[index].rf_addr; | |
874 | } | |
875 | ||
876 | static inline __u32 | |
877 | kiblnd_rd_frag_size(kib_rdma_desc_t *rd, int index) | |
878 | { | |
879 | return rd->rd_frags[index].rf_nob; | |
880 | } | |
881 | ||
882 | static inline __u32 | |
883 | kiblnd_rd_frag_key(kib_rdma_desc_t *rd, int index) | |
884 | { | |
885 | return rd->rd_key; | |
886 | } | |
887 | ||
888 | static inline int | |
889 | kiblnd_rd_consume_frag(kib_rdma_desc_t *rd, int index, __u32 nob) | |
890 | { | |
891 | if (nob < rd->rd_frags[index].rf_nob) { | |
892 | rd->rd_frags[index].rf_addr += nob; | |
893 | rd->rd_frags[index].rf_nob -= nob; | |
894 | } else { | |
895 | index ++; | |
896 | } | |
897 | ||
898 | return index; | |
899 | } | |
900 | ||
901 | static inline int | |
902 | kiblnd_rd_msg_size(kib_rdma_desc_t *rd, int msgtype, int n) | |
903 | { | |
904 | LASSERT (msgtype == IBLND_MSG_GET_REQ || | |
905 | msgtype == IBLND_MSG_PUT_ACK); | |
906 | ||
907 | return msgtype == IBLND_MSG_GET_REQ ? | |
908 | offsetof(kib_get_msg_t, ibgm_rd.rd_frags[n]) : | |
909 | offsetof(kib_putack_msg_t, ibpam_rd.rd_frags[n]); | |
910 | } | |
911 | ||
912 | ||
913 | static inline __u64 | |
914 | kiblnd_dma_mapping_error(struct ib_device *dev, u64 dma_addr) | |
915 | { | |
916 | return ib_dma_mapping_error(dev, dma_addr); | |
917 | } | |
918 | ||
919 | static inline __u64 kiblnd_dma_map_single(struct ib_device *dev, | |
920 | void *msg, size_t size, | |
921 | enum dma_data_direction direction) | |
922 | { | |
923 | return ib_dma_map_single(dev, msg, size, direction); | |
924 | } | |
925 | ||
926 | static inline void kiblnd_dma_unmap_single(struct ib_device *dev, | |
927 | __u64 addr, size_t size, | |
928 | enum dma_data_direction direction) | |
929 | { | |
930 | ib_dma_unmap_single(dev, addr, size, direction); | |
931 | } | |
932 | ||
933 | #define KIBLND_UNMAP_ADDR_SET(p, m, a) do {} while (0) | |
934 | #define KIBLND_UNMAP_ADDR(p, m, a) (a) | |
935 | ||
936 | static inline int kiblnd_dma_map_sg(struct ib_device *dev, | |
937 | struct scatterlist *sg, int nents, | |
938 | enum dma_data_direction direction) | |
939 | { | |
940 | return ib_dma_map_sg(dev, sg, nents, direction); | |
941 | } | |
942 | ||
943 | static inline void kiblnd_dma_unmap_sg(struct ib_device *dev, | |
944 | struct scatterlist *sg, int nents, | |
945 | enum dma_data_direction direction) | |
946 | { | |
947 | ib_dma_unmap_sg(dev, sg, nents, direction); | |
948 | } | |
949 | ||
950 | static inline __u64 kiblnd_sg_dma_address(struct ib_device *dev, | |
951 | struct scatterlist *sg) | |
952 | { | |
953 | return ib_sg_dma_address(dev, sg); | |
954 | } | |
955 | ||
956 | static inline unsigned int kiblnd_sg_dma_len(struct ib_device *dev, | |
957 | struct scatterlist *sg) | |
958 | { | |
959 | return ib_sg_dma_len(dev, sg); | |
960 | } | |
961 | ||
962 | /* XXX We use KIBLND_CONN_PARAM(e) as writable buffer, it's not strictly | |
963 | * right because OFED1.2 defines it as const, to use it we have to add | |
964 | * (void *) cast to overcome "const" */ | |
965 | ||
966 | #define KIBLND_CONN_PARAM(e) ((e)->param.conn.private_data) | |
967 | #define KIBLND_CONN_PARAM_LEN(e) ((e)->param.conn.private_data_len) | |
968 | ||
969 | ||
970 | struct ib_mr *kiblnd_find_rd_dma_mr(kib_hca_dev_t *hdev, | |
971 | kib_rdma_desc_t *rd); | |
972 | struct ib_mr *kiblnd_find_dma_mr(kib_hca_dev_t *hdev, | |
973 | __u64 addr, __u64 size); | |
974 | void kiblnd_map_rx_descs(kib_conn_t *conn); | |
975 | void kiblnd_unmap_rx_descs(kib_conn_t *conn); | |
976 | int kiblnd_map_tx(lnet_ni_t *ni, kib_tx_t *tx, | |
977 | kib_rdma_desc_t *rd, int nfrags); | |
978 | void kiblnd_unmap_tx(lnet_ni_t *ni, kib_tx_t *tx); | |
979 | void kiblnd_pool_free_node(kib_pool_t *pool, struct list_head *node); | |
980 | struct list_head *kiblnd_pool_alloc_node(kib_poolset_t *ps); | |
981 | ||
982 | int kiblnd_fmr_pool_map(kib_fmr_poolset_t *fps, __u64 *pages, | |
983 | int npages, __u64 iov, kib_fmr_t *fmr); | |
984 | void kiblnd_fmr_pool_unmap(kib_fmr_t *fmr, int status); | |
985 | ||
986 | int kiblnd_pmr_pool_map(kib_pmr_poolset_t *pps, kib_hca_dev_t *hdev, | |
987 | kib_rdma_desc_t *rd, __u64 *iova, kib_phys_mr_t **pp_pmr); | |
988 | void kiblnd_pmr_pool_unmap(kib_phys_mr_t *pmr); | |
989 | ||
990 | int kiblnd_startup (lnet_ni_t *ni); | |
991 | void kiblnd_shutdown (lnet_ni_t *ni); | |
992 | int kiblnd_ctl (lnet_ni_t *ni, unsigned int cmd, void *arg); | |
993 | void kiblnd_query (struct lnet_ni *ni, lnet_nid_t nid, cfs_time_t *when); | |
994 | ||
995 | int kiblnd_tunables_init(void); | |
996 | void kiblnd_tunables_fini(void); | |
997 | ||
998 | int kiblnd_connd (void *arg); | |
999 | int kiblnd_scheduler(void *arg); | |
1000 | int kiblnd_thread_start(int (*fn)(void *arg), void *arg, char *name); | |
1001 | int kiblnd_failover_thread (void *arg); | |
1002 | ||
1003 | int kiblnd_alloc_pages(kib_pages_t **pp, int cpt, int npages); | |
1004 | void kiblnd_free_pages (kib_pages_t *p); | |
1005 | ||
1006 | int kiblnd_cm_callback(struct rdma_cm_id *cmid, | |
1007 | struct rdma_cm_event *event); | |
1008 | int kiblnd_translate_mtu(int value); | |
1009 | ||
1010 | int kiblnd_dev_failover(kib_dev_t *dev); | |
1011 | int kiblnd_create_peer (lnet_ni_t *ni, kib_peer_t **peerp, lnet_nid_t nid); | |
1012 | void kiblnd_destroy_peer (kib_peer_t *peer); | |
1013 | void kiblnd_destroy_dev (kib_dev_t *dev); | |
1014 | void kiblnd_unlink_peer_locked (kib_peer_t *peer); | |
1015 | void kiblnd_peer_alive (kib_peer_t *peer); | |
1016 | kib_peer_t *kiblnd_find_peer_locked (lnet_nid_t nid); | |
1017 | void kiblnd_peer_connect_failed (kib_peer_t *peer, int active, int error); | |
1018 | int kiblnd_close_stale_conns_locked (kib_peer_t *peer, | |
1019 | int version, __u64 incarnation); | |
1020 | int kiblnd_close_peer_conns_locked (kib_peer_t *peer, int why); | |
1021 | ||
1022 | void kiblnd_connreq_done(kib_conn_t *conn, int status); | |
1023 | kib_conn_t *kiblnd_create_conn (kib_peer_t *peer, struct rdma_cm_id *cmid, | |
1024 | int state, int version); | |
1025 | void kiblnd_destroy_conn (kib_conn_t *conn); | |
1026 | void kiblnd_close_conn (kib_conn_t *conn, int error); | |
1027 | void kiblnd_close_conn_locked (kib_conn_t *conn, int error); | |
1028 | ||
1029 | int kiblnd_init_rdma (kib_conn_t *conn, kib_tx_t *tx, int type, | |
1030 | int nob, kib_rdma_desc_t *dstrd, __u64 dstcookie); | |
1031 | ||
1032 | void kiblnd_launch_tx (lnet_ni_t *ni, kib_tx_t *tx, lnet_nid_t nid); | |
1033 | void kiblnd_queue_tx_locked (kib_tx_t *tx, kib_conn_t *conn); | |
1034 | void kiblnd_queue_tx (kib_tx_t *tx, kib_conn_t *conn); | |
1035 | void kiblnd_init_tx_msg (lnet_ni_t *ni, kib_tx_t *tx, int type, int body_nob); | |
1036 | void kiblnd_txlist_done (lnet_ni_t *ni, struct list_head *txlist, | |
1037 | int status); | |
1038 | void kiblnd_check_sends (kib_conn_t *conn); | |
1039 | ||
1040 | void kiblnd_qp_event(struct ib_event *event, void *arg); | |
1041 | void kiblnd_cq_event(struct ib_event *event, void *arg); | |
1042 | void kiblnd_cq_completion(struct ib_cq *cq, void *arg); | |
1043 | ||
1044 | void kiblnd_pack_msg (lnet_ni_t *ni, kib_msg_t *msg, int version, | |
1045 | int credits, lnet_nid_t dstnid, __u64 dststamp); | |
1046 | int kiblnd_unpack_msg(kib_msg_t *msg, int nob); | |
1047 | int kiblnd_post_rx (kib_rx_t *rx, int credit); | |
1048 | ||
1049 | int kiblnd_send(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg); | |
1050 | int kiblnd_recv(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg, int delayed, | |
1051 | unsigned int niov, struct iovec *iov, lnet_kiov_t *kiov, | |
1052 | unsigned int offset, unsigned int mlen, unsigned int rlen); |