1 /*
2 * Copyright (c) 2006, 2019 Oracle and/or its affiliates. All rights reserved.
3 *
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE.
31 *
32 */
33 #include <linux/kernel.h>
34 #include <linux/sched/clock.h>
35 #include <linux/slab.h>
36 #include <linux/pci.h>
37 #include <linux/dma-mapping.h>
38 #include <rdma/rdma_cm.h>
39
40 #include "rds_single_path.h"
41 #include "rds.h"
42 #include "ib.h"
43
44 static struct kmem_cache *rds_ib_incoming_slab;
45 static struct kmem_cache *rds_ib_frag_slab;
46 static atomic_t rds_ib_allocation = ATOMIC_INIT(0);
47
48 void rds_ib_recv_init_ring(struct rds_ib_connection *ic)
49 {
50 struct rds_ib_recv_work *recv;
51 u32 i;
52
53 for (i = 0, recv = ic->i_recvs; i < ic->i_recv_ring.w_nr; i++, recv++) {
54 struct ib_sge *sge;
55
56 recv->r_ibinc = NULL;
57 recv->r_frag = NULL;
58
59 recv->r_wr.next = NULL;
60 recv->r_wr.wr_id = i;
61 recv->r_wr.sg_list = recv->r_sge;
62 recv->r_wr.num_sge = RDS_IB_RECV_SGE;
63
64 sge = &recv->r_sge[0];
65 sge->addr = ic->i_recv_hdrs_dma[i];
66 sge->length = sizeof(struct rds_header);
67 sge->lkey = ic->i_pd->local_dma_lkey;
68
69 sge = &recv->r_sge[1];
70 sge->addr = 0;
71 sge->length = RDS_FRAG_SIZE;
72 sge->lkey = ic->i_pd->local_dma_lkey;
73 }
74 }
75
76 /*
77 * The entire 'from' list, including the from element itself, is put on
78 * to the tail of the 'to' list.
79 */
80 static void list_splice_entire_tail(struct list_head *from,
81 struct list_head *to)
82 {
83 struct list_head *from_last = from->prev;
84
85 list_splice_tail(from_last, to);
86 list_add_tail(from_last, to);
87 }
88
89 static void rds_ib_cache_xfer_to_ready(struct rds_ib_refill_cache *cache)
90 {
91 struct list_head *tmp;
92
93 tmp = xchg(&cache->xfer, NULL);
94 if (tmp) {
95 if (cache->ready)
96 list_splice_entire_tail(tmp, cache->ready);
97 else
98 cache->ready = tmp;
99 }
100 }
101
102 static int rds_ib_recv_alloc_cache(struct rds_ib_refill_cache *cache, gfp_t gfp)
103 {
104 struct rds_ib_cache_head *head;
105 int cpu;
106
107 cache->percpu = alloc_percpu_gfp(struct rds_ib_cache_head, gfp);
108 if (!cache->percpu)
109 return -ENOMEM;
110
111 for_each_possible_cpu(cpu) {
112 head = per_cpu_ptr(cache->percpu, cpu);
113 head->first = NULL;
114 head->count = 0;
115 }
116 cache->xfer = NULL;
117 cache->ready = NULL;
118
119 return 0;
120 }
121
122 int rds_ib_recv_alloc_caches(struct rds_ib_connection *ic, gfp_t gfp)
123 {
124 int ret;
125
126 ret = rds_ib_recv_alloc_cache(&ic->i_cache_incs, gfp);
127 if (!ret) {
128 ret = rds_ib_recv_alloc_cache(&ic->i_cache_frags, gfp);
129 if (ret)
130 free_percpu(ic->i_cache_incs.percpu);
131 }
132
133 return ret;
134 }
135
136 static void rds_ib_cache_splice_all_lists(struct rds_ib_refill_cache *cache,
137 struct list_head *caller_list)
138 {
139 struct rds_ib_cache_head *head;
140 int cpu;
141
142 for_each_possible_cpu(cpu) {
143 head = per_cpu_ptr(cache->percpu, cpu);
144 if (head->first) {
145 list_splice_entire_tail(head->first, caller_list);
146 head->first = NULL;
147 }
148 }
149
150 if (cache->ready) {
151 list_splice_entire_tail(cache->ready, caller_list);
152 cache->ready = NULL;
153 }
154 }
155
156 void rds_ib_recv_free_caches(struct rds_ib_connection *ic)
157 {
158 struct rds_ib_incoming *inc;
159 struct rds_ib_incoming *inc_tmp;
160 struct rds_page_frag *frag;
161 struct rds_page_frag *frag_tmp;
162 LIST_HEAD(list);
163
164 rds_ib_cache_xfer_to_ready(&ic->i_cache_incs);
165 rds_ib_cache_splice_all_lists(&ic->i_cache_incs, &list);
166 free_percpu(ic->i_cache_incs.percpu);
167
168 list_for_each_entry_safe(inc, inc_tmp, &list, ii_cache_entry) {
169 list_del(&inc->ii_cache_entry);
170 WARN_ON(!list_empty(&inc->ii_frags));
171 kmem_cache_free(rds_ib_incoming_slab, inc);
172 atomic_dec(&rds_ib_allocation);
173 }
174
175 rds_ib_cache_xfer_to_ready(&ic->i_cache_frags);
176 rds_ib_cache_splice_all_lists(&ic->i_cache_frags, &list);
177 free_percpu(ic->i_cache_frags.percpu);
178
179 list_for_each_entry_safe(frag, frag_tmp, &list, f_cache_entry) {
180 list_del(&frag->f_cache_entry);
181 WARN_ON(!list_empty(&frag->f_item));
182 kmem_cache_free(rds_ib_frag_slab, frag);
183 }
184 }
185
186 /* fwd decl */
187 static void rds_ib_recv_cache_put(struct list_head *new_item,
188 struct rds_ib_refill_cache *cache);
189 static struct list_head *rds_ib_recv_cache_get(struct rds_ib_refill_cache *cache);
190
191
192 /* Recycle frag and attached recv buffer f_sg */
193 static void rds_ib_frag_free(struct rds_ib_connection *ic,
194 struct rds_page_frag *frag)
195 {
196 rdsdebug("frag %p page %p\n", frag, sg_page(&frag->f_sg));
197
198 rds_ib_recv_cache_put(&frag->f_cache_entry, &ic->i_cache_frags);
199 atomic_add(RDS_FRAG_SIZE / SZ_1K, &ic->i_cache_allocs);
200 rds_ib_stats_add(s_ib_recv_added_to_cache, RDS_FRAG_SIZE);
201 }
202
203 /* Recycle inc after freeing attached frags */
204 void rds_ib_inc_free(struct rds_incoming *inc)
205 {
206 struct rds_ib_incoming *ibinc;
207 struct rds_page_frag *frag;
208 struct rds_page_frag *pos;
209 struct rds_ib_connection *ic = inc->i_conn->c_transport_data;
210
211 ibinc = container_of(inc, struct rds_ib_incoming, ii_inc);
212
213 /* Free attached frags */
214 list_for_each_entry_safe(frag, pos, &ibinc->ii_frags, f_item) {
215 list_del_init(&frag->f_item);
216 rds_ib_frag_free(ic, frag);
217 }
218 BUG_ON(!list_empty(&ibinc->ii_frags));
219
220 rdsdebug("freeing ibinc %p inc %p\n", ibinc, inc);
221 rds_ib_recv_cache_put(&ibinc->ii_cache_entry, &ic->i_cache_incs);
222 }
223
224 static void rds_ib_recv_clear_one(struct rds_ib_connection *ic,
225 struct rds_ib_recv_work *recv)
226 {
227 if (recv->r_ibinc) {
228 rds_inc_put(&recv->r_ibinc->ii_inc);
229 recv->r_ibinc = NULL;
230 }
231 if (recv->r_frag) {
232 ib_dma_unmap_sg(ic->i_cm_id->device, &recv->r_frag->f_sg, 1, DMA_FROM_DEVICE);
233 rds_ib_frag_free(ic, recv->r_frag);
234 recv->r_frag = NULL;
235 }
236 }
237
238 void rds_ib_recv_clear_ring(struct rds_ib_connection *ic)
239 {
240 u32 i;
241
242 for (i = 0; i < ic->i_recv_ring.w_nr; i++)
243 rds_ib_recv_clear_one(ic, &ic->i_recvs[i]);
244 }
245
246 static struct rds_ib_incoming *rds_ib_refill_one_inc(struct rds_ib_connection *ic,
247 gfp_t slab_mask)
248 {
249 struct rds_ib_incoming *ibinc;
250 struct list_head *cache_item;
251 int avail_allocs;
252
253 cache_item = rds_ib_recv_cache_get(&ic->i_cache_incs);
254 if (cache_item) {
255 ibinc = container_of(cache_item, struct rds_ib_incoming, ii_cache_entry);
256 } else {
257 avail_allocs = atomic_add_unless(&rds_ib_allocation,
258 1, rds_ib_sysctl_max_recv_allocation);
259 if (!avail_allocs) {
260 rds_ib_stats_inc(s_ib_rx_alloc_limit);
261 return NULL;
262 }
263 ibinc = kmem_cache_alloc(rds_ib_incoming_slab, slab_mask);
264 if (!ibinc) {
265 atomic_dec(&rds_ib_allocation);
266 return NULL;
267 }
268 rds_ib_stats_inc(s_ib_rx_total_incs);
269 }
270 INIT_LIST_HEAD(&ibinc->ii_frags);
271 rds_inc_init(&ibinc->ii_inc, ic->conn, &ic->conn->c_faddr);
272
273 return ibinc;
274 }
275
276 static struct rds_page_frag *rds_ib_refill_one_frag(struct rds_ib_connection *ic,
277 gfp_t slab_mask, gfp_t page_mask)
278 {
279 struct rds_page_frag *frag;
280 struct list_head *cache_item;
281 int ret;
282
283 cache_item = rds_ib_recv_cache_get(&ic->i_cache_frags);
284 if (cache_item) {
285 frag = container_of(cache_item, struct rds_page_frag, f_cache_entry);
286 atomic_sub(RDS_FRAG_SIZE / SZ_1K, &ic->i_cache_allocs);
287 rds_ib_stats_add(s_ib_recv_added_to_cache, RDS_FRAG_SIZE);
288 } else {
289 frag = kmem_cache_alloc(rds_ib_frag_slab, slab_mask);
290 if (!frag)
291 return NULL;
292
293 sg_init_table(&frag->f_sg, 1);
294 ret = rds_page_remainder_alloc(&frag->f_sg,
295 RDS_FRAG_SIZE, page_mask);
296 if (ret) {
297 kmem_cache_free(rds_ib_frag_slab, frag);
298 return NULL;
299 }
300 rds_ib_stats_inc(s_ib_rx_total_frags);
301 }
302
303 INIT_LIST_HEAD(&frag->f_item);
304
305 return frag;
306 }
307
308 static int rds_ib_recv_refill_one(struct rds_connection *conn,
309 struct rds_ib_recv_work *recv, gfp_t gfp)
310 {
311 struct rds_ib_connection *ic = conn->c_transport_data;
312 struct ib_sge *sge;
313 int ret = -ENOMEM;
314 gfp_t slab_mask = gfp;
315 gfp_t page_mask = gfp;
316
317 if (gfp & __GFP_DIRECT_RECLAIM) {
318 slab_mask = GFP_KERNEL;
319 page_mask = GFP_HIGHUSER;
320 }
321
322 if (!ic->i_cache_incs.ready)
323 rds_ib_cache_xfer_to_ready(&ic->i_cache_incs);
324 if (!ic->i_cache_frags.ready)
325 rds_ib_cache_xfer_to_ready(&ic->i_cache_frags);
326
327 /*
328 * ibinc was taken from recv if recv contained the start of a message.
329 * recvs that were continuations will still have this allocated.
330 */
331 if (!recv->r_ibinc) {
332 recv->r_ibinc = rds_ib_refill_one_inc(ic, slab_mask);
333 if (!recv->r_ibinc)
334 goto out;
335 }
336
337 WARN_ON(recv->r_frag); /* leak! */
338 recv->r_frag = rds_ib_refill_one_frag(ic, slab_mask, page_mask);
339 if (!recv->r_frag)
340 goto out;
341
342 ret = ib_dma_map_sg(ic->i_cm_id->device, &recv->r_frag->f_sg,
343 1, DMA_FROM_DEVICE);
344 WARN_ON(ret != 1);
345
346 sge = &recv->r_sge[0];
347 sge->addr = ic->i_recv_hdrs_dma[recv - ic->i_recvs];
348 sge->length = sizeof(struct rds_header);
349
350 sge = &recv->r_sge[1];
351 sge->addr = sg_dma_address(&recv->r_frag->f_sg);
352 sge->length = sg_dma_len(&recv->r_frag->f_sg);
353
354 ret = 0;
355 out:
356 return ret;
357 }
358
359 static int acquire_refill(struct rds_connection *conn)
360 {
361 return test_and_set_bit(RDS_RECV_REFILL, &conn->c_flags) == 0;
362 }
363
364 static void release_refill(struct rds_connection *conn)
365 {
366 clear_bit(RDS_RECV_REFILL, &conn->c_flags);
367 smp_mb__after_atomic();
368
369 /* We don't use wait_on_bit()/wake_up_bit() because our waking is in a
370 * hot path and finding waiters is very rare. We don't want to walk
371 * the system-wide hashed waitqueue buckets in the fast path only to
372 * almost never find waiters.
373 */
374 if (waitqueue_active(&conn->c_waitq))
375 wake_up_all(&conn->c_waitq);
376 }
377
378 /*
379 * This tries to allocate and post unused work requests after making sure that
380 * they have all the allocations they need to queue received fragments into
381 * sockets.
382 */
383 void rds_ib_recv_refill(struct rds_connection *conn, int prefill, gfp_t gfp)
384 {
385 struct rds_ib_connection *ic = conn->c_transport_data;
386 struct rds_ib_recv_work *recv;
387 unsigned int posted = 0;
388 int ret = 0;
389 bool can_wait = !!(gfp & __GFP_DIRECT_RECLAIM);
390 bool must_wake = false;
391 u32 pos;
392
393 /* the goal here is to just make sure that someone, somewhere
394 * is posting buffers. If we can't get the refill lock,
395 * let them do their thing
396 */
397 if (!acquire_refill(conn))
398 return;
399
400 while ((prefill || rds_conn_up(conn)) &&
401 rds_ib_ring_alloc(&ic->i_recv_ring, 1, &pos)) {
402 if (pos >= ic->i_recv_ring.w_nr) {
403 printk(KERN_NOTICE "Argh - ring alloc returned pos=%u\n",
404 pos);
405 break;
406 }
407
408 recv = &ic->i_recvs[pos];
409 ret = rds_ib_recv_refill_one(conn, recv, gfp);
410 if (ret) {
411 must_wake = true;
412 break;
413 }
414
415 rdsdebug("recv %p ibinc %p page %p addr %lu\n", recv,
416 recv->r_ibinc, sg_page(&recv->r_frag->f_sg),
417 (long)sg_dma_address(&recv->r_frag->f_sg));
418
419 /* XXX when can this fail? */
420 ret = ib_post_recv(ic->i_cm_id->qp, &recv->r_wr, NULL);
421 if (ret) {
422 rds_ib_conn_error(conn, "recv post on "
423 "%pI6c returned %d, disconnecting and "
424 "reconnecting\n", &conn->c_faddr,
425 ret);
426 break;
427 }
428
429 posted++;
430
431 if ((posted > 128 && need_resched()) || posted > 8192) {
432 must_wake = true;
433 break;
434 }
435 }
436
437 /* We're doing flow control - update the window. */
438 if (ic->i_flowctl && posted)
439 rds_ib_advertise_credits(conn, posted);
440
441 if (ret)
442 rds_ib_ring_unalloc(&ic->i_recv_ring, 1);
443
444 release_refill(conn);
445
446 /* if we're called from the softirq handler, we'll be GFP_NOWAIT.
447 * in this case the ring being low is going to lead to more interrupts
448 * and we can safely let the softirq code take care of it unless the
449 * ring is completely empty.
450 *
451 * if we're called from krdsd, we'll be GFP_KERNEL. In this case
452 * we might have raced with the softirq code while we had the refill
453 * lock held. Use rds_ib_ring_low() instead of ring_empty to decide
454 * if we should requeue.
455 */
456 if (rds_conn_up(conn) &&
457 (must_wake ||
458 (can_wait && rds_ib_ring_low(&ic->i_recv_ring)) ||
459 rds_ib_ring_empty(&ic->i_recv_ring))) {
460 queue_delayed_work(rds_wq, &conn->c_recv_w, 1);
461 }
462 if (can_wait)
463 cond_resched();
464 }
465
466 /*
467 * We want to recycle several types of recv allocations, like incs and frags.
468 * To use this, the *_free() function passes in the ptr to a list_head within
469 * the recyclee, as well as the cache to put it on.
470 *
471 * First, we put the memory on a percpu list. When this reaches a certain size,
472 * We move it to an intermediate non-percpu list in a lockless manner, with some
473 * xchg/compxchg wizardry.
474 *
475 * N.B. Instead of a list_head as the anchor, we use a single pointer, which can
476 * be NULL and xchg'd. The list is actually empty when the pointer is NULL, and
477 * list_empty() will return true with one element is actually present.
478 */
479 static void rds_ib_recv_cache_put(struct list_head *new_item,
480 struct rds_ib_refill_cache *cache)
481 {
482 unsigned long flags;
483 struct list_head *old, *chpfirst;
484
485 local_irq_save(flags);
486
487 chpfirst = __this_cpu_read(cache->percpu->first);
488 if (!chpfirst)
489 INIT_LIST_HEAD(new_item);
490 else /* put on front */
491 list_add_tail(new_item, chpfirst);
492
493 __this_cpu_write(cache->percpu->first, new_item);
494 __this_cpu_inc(cache->percpu->count);
495
496 if (__this_cpu_read(cache->percpu->count) < RDS_IB_RECYCLE_BATCH_COUNT)
497 goto end;
498
499 /*
500 * Return our per-cpu first list to the cache's xfer by atomically
501 * grabbing the current xfer list, appending it to our per-cpu list,
502 * and then atomically returning that entire list back to the
503 * cache's xfer list as long as it's still empty.
504 */
505 do {
506 old = xchg(&cache->xfer, NULL);
507 if (old)
508 list_splice_entire_tail(old, chpfirst);
509 old = cmpxchg(&cache->xfer, NULL, chpfirst);
510 } while (old);
511
512
513 __this_cpu_write(cache->percpu->first, NULL);
514 __this_cpu_write(cache->percpu->count, 0);
515 end:
516 local_irq_restore(flags);
517 }
518
519 static struct list_head *rds_ib_recv_cache_get(struct rds_ib_refill_cache *cache)
520 {
521 struct list_head *head = cache->ready;
522
523 if (head) {
524 if (!list_empty(head)) {
525 cache->ready = head->next;
526 list_del_init(head);
527 } else
528 cache->ready = NULL;
529 }
530
531 return head;
532 }
533
534 int rds_ib_inc_copy_to_user(struct rds_incoming *inc, struct iov_iter *to)
535 {
536 struct rds_ib_incoming *ibinc;
537 struct rds_page_frag *frag;
538 unsigned long to_copy;
539 unsigned long frag_off = 0;
540 int copied = 0;
541 int ret;
542 u32 len;
543
544 ibinc = container_of(inc, struct rds_ib_incoming, ii_inc);
545 frag = list_entry(ibinc->ii_frags.next, struct rds_page_frag, f_item);
546 len = be32_to_cpu(inc->i_hdr.h_len);
547
548 while (iov_iter_count(to) && copied < len) {
549 if (frag_off == RDS_FRAG_SIZE) {
550 frag = list_entry(frag->f_item.next,
551 struct rds_page_frag, f_item);
552 frag_off = 0;
553 }
554 to_copy = min_t(unsigned long, iov_iter_count(to),
555 RDS_FRAG_SIZE - frag_off);
556 to_copy = min_t(unsigned long, to_copy, len - copied);
557
558 /* XXX needs + offset for multiple recvs per page */
559 rds_stats_add(s_copy_to_user, to_copy);
560 ret = copy_page_to_iter(sg_page(&frag->f_sg),
561 frag->f_sg.offset + frag_off,
562 to_copy,
563 to);
564 if (ret != to_copy)
565 return -EFAULT;
566
567 frag_off += to_copy;
568 copied += to_copy;
569 }
570
571 return copied;
572 }
573
574 /* ic starts out kzalloc()ed */
575 void rds_ib_recv_init_ack(struct rds_ib_connection *ic)
576 {
577 struct ib_send_wr *wr = &ic->i_ack_wr;
578 struct ib_sge *sge = &ic->i_ack_sge;
579
580 sge->addr = ic->i_ack_dma;
581 sge->length = sizeof(struct rds_header);
582 sge->lkey = ic->i_pd->local_dma_lkey;
583
584 wr->sg_list = sge;
585 wr->num_sge = 1;
586 wr->opcode = IB_WR_SEND;
587 wr->wr_id = RDS_IB_ACK_WR_ID;
588 wr->send_flags = IB_SEND_SIGNALED | IB_SEND_SOLICITED;
589 }
590
591 /*
592 * You'd think that with reliable IB connections you wouldn't need to ack
593 * messages that have been received. The problem is that IB hardware generates
594 * an ack message before it has DMAed the message into memory. This creates a
595 * potential message loss if the HCA is disabled for any reason between when it
596 * sends the ack and before the message is DMAed and processed. This is only a
597 * potential issue if another HCA is available for fail-over.
598 *
599 * When the remote host receives our ack they'll free the sent message from
600 * their send queue. To decrease the latency of this we always send an ack
601 * immediately after we've received messages.
602 *
603 * For simplicity, we only have one ack in flight at a time. This puts
604 * pressure on senders to have deep enough send queues to absorb the latency of
605 * a single ack frame being in flight. This might not be good enough.
606 *
607 * This is implemented by have a long-lived send_wr and sge which point to a
608 * statically allocated ack frame. This ack wr does not fall under the ring
609 * accounting that the tx and rx wrs do. The QP attribute specifically makes
610 * room for it beyond the ring size. Send completion notices its special
611 * wr_id and avoids working with the ring in that case.
612 */
613 #ifndef KERNEL_HAS_ATOMIC64
614 void rds_ib_set_ack(struct rds_ib_connection *ic, u64 seq, int ack_required)
615 {
616 unsigned long flags;
617
618 spin_lock_irqsave(&ic->i_ack_lock, flags);
619 ic->i_ack_next = seq;
620 if (ack_required)
621 set_bit(IB_ACK_REQUESTED, &ic->i_ack_flags);
622 spin_unlock_irqrestore(&ic->i_ack_lock, flags);
623 }
624
625 static u64 rds_ib_get_ack(struct rds_ib_connection *ic)
626 {
627 unsigned long flags;
628 u64 seq;
629
630 clear_bit(IB_ACK_REQUESTED, &ic->i_ack_flags);
631
632 spin_lock_irqsave(&ic->i_ack_lock, flags);
633 seq = ic->i_ack_next;
634 spin_unlock_irqrestore(&ic->i_ack_lock, flags);
635
636 return seq;
637 }
638 #else
639 void rds_ib_set_ack(struct rds_ib_connection *ic, u64 seq, int ack_required)
640 {
641 atomic64_set(&ic->i_ack_next, seq);
642 if (ack_required) {
643 smp_mb__before_atomic();
644 set_bit(IB_ACK_REQUESTED, &ic->i_ack_flags);
645 }
646 }
647
648 static u64 rds_ib_get_ack(struct rds_ib_connection *ic)
649 {
650 clear_bit(IB_ACK_REQUESTED, &ic->i_ack_flags);
651 smp_mb__after_atomic();
652
653 return atomic64_read(&ic->i_ack_next);
654 }
655 #endif
656
657
658 static void rds_ib_send_ack(struct rds_ib_connection *ic, unsigned int adv_credits)
659 {
660 struct rds_header *hdr = ic->i_ack;
661 u64 seq;
662 int ret;
663
664 seq = rds_ib_get_ack(ic);
665
666 rdsdebug("send_ack: ic %p ack %llu\n", ic, (unsigned long long) seq);
667
668 ib_dma_sync_single_for_cpu(ic->rds_ibdev->dev, ic->i_ack_dma,
669 sizeof(*hdr), DMA_TO_DEVICE);
670 rds_message_populate_header(hdr, 0, 0, 0);
671 hdr->h_ack = cpu_to_be64(seq);
672 hdr->h_credit = adv_credits;
673 rds_message_make_checksum(hdr);
674 ib_dma_sync_single_for_device(ic->rds_ibdev->dev, ic->i_ack_dma,
675 sizeof(*hdr), DMA_TO_DEVICE);
676
677 ic->i_ack_queued = jiffies;
678
679 ret = ib_post_send(ic->i_cm_id->qp, &ic->i_ack_wr, NULL);
680 if (unlikely(ret)) {
681 /* Failed to send. Release the WR, and
682 * force another ACK.
683 */
684 clear_bit(IB_ACK_IN_FLIGHT, &ic->i_ack_flags);
685 set_bit(IB_ACK_REQUESTED, &ic->i_ack_flags);
686
687 rds_ib_stats_inc(s_ib_ack_send_failure);
688
689 rds_ib_conn_error(ic->conn, "sending ack failed\n");
690 } else
691 rds_ib_stats_inc(s_ib_ack_sent);
692 }
693
694 /*
695 * There are 3 ways of getting acknowledgements to the peer:
696 * 1. We call rds_ib_attempt_ack from the recv completion handler
697 * to send an ACK-only frame.
698 * However, there can be only one such frame in the send queue
699 * at any time, so we may have to postpone it.
700 * 2. When another (data) packet is transmitted while there's
701 * an ACK in the queue, we piggyback the ACK sequence number
702 * on the data packet.
703 * 3. If the ACK WR is done sending, we get called from the
704 * send queue completion handler, and check whether there's
705 * another ACK pending (postponed because the WR was on the
706 * queue). If so, we transmit it.
707 *
708 * We maintain 2 variables:
709 * - i_ack_flags, which keeps track of whether the ACK WR
710 * is currently in the send queue or not (IB_ACK_IN_FLIGHT)
711 * - i_ack_next, which is the last sequence number we received
712 *
713 * Potentially, send queue and receive queue handlers can run concurrently.
714 * It would be nice to not have to use a spinlock to synchronize things,
715 * but the one problem that rules this out is that 64bit updates are
716 * not atomic on all platforms. Things would be a lot simpler if
717 * we had atomic64 or maybe cmpxchg64 everywhere.
718 *
719 * Reconnecting complicates this picture just slightly. When we
720 * reconnect, we may be seeing duplicate packets. The peer
721 * is retransmitting them, because it hasn't seen an ACK for
722 * them. It is important that we ACK these.
723 *
724 * ACK mitigation adds a header flag "ACK_REQUIRED"; any packet with
725 * this flag set *MUST* be acknowledged immediately.
726 */
727
728 /*
729 * When we get here, we're called from the recv queue handler.
730 * Check whether we ought to transmit an ACK.
731 */
732 void rds_ib_attempt_ack(struct rds_ib_connection *ic)
733 {
734 unsigned int adv_credits;
735
736 if (!test_bit(IB_ACK_REQUESTED, &ic->i_ack_flags))
737 return;
738
739 if (test_and_set_bit(IB_ACK_IN_FLIGHT, &ic->i_ack_flags)) {
740 rds_ib_stats_inc(s_ib_ack_send_delayed);
741 return;
742 }
743
744 /* Can we get a send credit? */
745 if (!rds_ib_send_grab_credits(ic, 1, &adv_credits, 0, RDS_MAX_ADV_CREDIT)) {
746 rds_ib_stats_inc(s_ib_tx_throttle);
747 clear_bit(IB_ACK_IN_FLIGHT, &ic->i_ack_flags);
748 return;
749 }
750
751 clear_bit(IB_ACK_REQUESTED, &ic->i_ack_flags);
752 rds_ib_send_ack(ic, adv_credits);
753 }
754
755 /*
756 * We get here from the send completion handler, when the
757 * adapter tells us the ACK frame was sent.
758 */
759 void rds_ib_ack_send_complete(struct rds_ib_connection *ic)
760 {
761 clear_bit(IB_ACK_IN_FLIGHT, &ic->i_ack_flags);
762 rds_ib_attempt_ack(ic);
763 }
764
765 /*
766 * This is called by the regular xmit code when it wants to piggyback
767 * an ACK on an outgoing frame.
768 */
769 u64 rds_ib_piggyb_ack(struct rds_ib_connection *ic)
770 {
771 if (test_and_clear_bit(IB_ACK_REQUESTED, &ic->i_ack_flags))
772 rds_ib_stats_inc(s_ib_ack_send_piggybacked);
773 return rds_ib_get_ack(ic);
774 }
775
776 /*
777 * It's kind of lame that we're copying from the posted receive pages into
778 * long-lived bitmaps. We could have posted the bitmaps and rdma written into
779 * them. But receiving new congestion bitmaps should be a *rare* event, so
780 * hopefully we won't need to invest that complexity in making it more
781 * efficient. By copying we can share a simpler core with TCP which has to
782 * copy.
783 */
784 static void rds_ib_cong_recv(struct rds_connection *conn,
785 struct rds_ib_incoming *ibinc)
786 {
787 struct rds_cong_map *map;
788 unsigned int map_off;
789 unsigned int map_page;
790 struct rds_page_frag *frag;
791 unsigned long frag_off;
792 unsigned long to_copy;
793 unsigned long copied;
794 __le64 uncongested = 0;
795 void *addr;
796
797 /* catch completely corrupt packets */
798 if (be32_to_cpu(ibinc->ii_inc.i_hdr.h_len) != RDS_CONG_MAP_BYTES)
799 return;
800
801 map = conn->c_fcong;
802 map_page = 0;
803 map_off = 0;
804
805 frag = list_entry(ibinc->ii_frags.next, struct rds_page_frag, f_item);
806 frag_off = 0;
807
808 copied = 0;
809
810 while (copied < RDS_CONG_MAP_BYTES) {
811 __le64 *src, *dst;
812 unsigned int k;
813
814 to_copy = min(RDS_FRAG_SIZE - frag_off, PAGE_SIZE - map_off);
815 BUG_ON(to_copy & 7); /* Must be 64bit aligned. */
816
817 addr = kmap_atomic(sg_page(&frag->f_sg));
818
819 src = addr + frag->f_sg.offset + frag_off;
820 dst = (void *)map->m_page_addrs[map_page] + map_off;
821 for (k = 0; k < to_copy; k += 8) {
822 /* Record ports that became uncongested, ie
823 * bits that changed from 0 to 1. */
824 uncongested |= ~(*src) & *dst;
825 *dst++ = *src++;
826 }
827 kunmap_atomic(addr);
828
829 copied += to_copy;
830
831 map_off += to_copy;
832 if (map_off == PAGE_SIZE) {
833 map_off = 0;
834 map_page++;
835 }
836
837 frag_off += to_copy;
838 if (frag_off == RDS_FRAG_SIZE) {
839 frag = list_entry(frag->f_item.next,
840 struct rds_page_frag, f_item);
841 frag_off = 0;
842 }
843 }
844
845 /* the congestion map is in little endian order */
846 rds_cong_map_updated(map, le64_to_cpu(uncongested));
847 }
848
849 static void rds_ib_process_recv(struct rds_connection *conn,
850 struct rds_ib_recv_work *recv, u32 data_len,
851 struct rds_ib_ack_state *state)
852 {
853 struct rds_ib_connection *ic = conn->c_transport_data;
854 struct rds_ib_incoming *ibinc = ic->i_ibinc;
855 struct rds_header *ihdr, *hdr;
856 dma_addr_t dma_addr = ic->i_recv_hdrs_dma[recv - ic->i_recvs];
857
858 /* XXX shut down the connection if port 0,0 are seen? */
859
860 rdsdebug("ic %p ibinc %p recv %p byte len %u\n", ic, ibinc, recv,
861 data_len);
862
863 if (data_len < sizeof(struct rds_header)) {
864 rds_ib_conn_error(conn, "incoming message "
865 "from %pI6c didn't include a "
866 "header, disconnecting and "
867 "reconnecting\n",
868 &conn->c_faddr);
869 return;
870 }
871 data_len -= sizeof(struct rds_header);
872
873 ihdr = ic->i_recv_hdrs[recv - ic->i_recvs];
874
875 ib_dma_sync_single_for_cpu(ic->rds_ibdev->dev, dma_addr,
876 sizeof(*ihdr), DMA_FROM_DEVICE);
877 /* Validate the checksum. */
878 if (!rds_message_verify_checksum(ihdr)) {
879 rds_ib_conn_error(conn, "incoming message "
880 "from %pI6c has corrupted header - "
881 "forcing a reconnect\n",
882 &conn->c_faddr);
883 rds_stats_inc(s_recv_drop_bad_checksum);
884 goto done;
885 }
886
887 /* Process the ACK sequence which comes with every packet */
888 state->ack_recv = be64_to_cpu(ihdr->h_ack);
889 state->ack_recv_valid = 1;
890
891 /* Process the credits update if there was one */
892 if (ihdr->h_credit)
893 rds_ib_send_add_credits(conn, ihdr->h_credit);
894
895 if (ihdr->h_sport == 0 && ihdr->h_dport == 0 && data_len == 0) {
896 /* This is an ACK-only packet. The fact that it gets
897 * special treatment here is that historically, ACKs
898 * were rather special beasts.
899 */
900 rds_ib_stats_inc(s_ib_ack_received);
901
902 /*
903 * Usually the frags make their way on to incs and are then freed as
904 * the inc is freed. We don't go that route, so we have to drop the
905 * page ref ourselves. We can't just leave the page on the recv
906 * because that confuses the dma mapping of pages and each recv's use
907 * of a partial page.
908 *
909 * FIXME: Fold this into the code path below.
910 */
911 rds_ib_frag_free(ic, recv->r_frag);
912 recv->r_frag = NULL;
913 goto done;
914 }
915
916 /*
917 * If we don't already have an inc on the connection then this
918 * fragment has a header and starts a message.. copy its header
919 * into the inc and save the inc so we can hang upcoming fragments
920 * off its list.
921 */
922 if (!ibinc) {
923 ibinc = recv->r_ibinc;
924 recv->r_ibinc = NULL;
925 ic->i_ibinc = ibinc;
926
927 hdr = &ibinc->ii_inc.i_hdr;
928 ibinc->ii_inc.i_rx_lat_trace[RDS_MSG_RX_HDR] =
929 local_clock();
930 memcpy(hdr, ihdr, sizeof(*hdr));
931 ic->i_recv_data_rem = be32_to_cpu(hdr->h_len);
932 ibinc->ii_inc.i_rx_lat_trace[RDS_MSG_RX_START] =
933 local_clock();
934
935 rdsdebug("ic %p ibinc %p rem %u flag 0x%x\n", ic, ibinc,
936 ic->i_recv_data_rem, hdr->h_flags);
937 } else {
938 hdr = &ibinc->ii_inc.i_hdr;
939 /* We can't just use memcmp here; fragments of a
940 * single message may carry different ACKs */
941 if (hdr->h_sequence != ihdr->h_sequence ||
942 hdr->h_len != ihdr->h_len ||
943 hdr->h_sport != ihdr->h_sport ||
944 hdr->h_dport != ihdr->h_dport) {
945 rds_ib_conn_error(conn,
946 "fragment header mismatch; forcing reconnect\n");
947 goto done;
948 }
949 }
950
951 list_add_tail(&recv->r_frag->f_item, &ibinc->ii_frags);
952 recv->r_frag = NULL;
953
954 if (ic->i_recv_data_rem > RDS_FRAG_SIZE)
955 ic->i_recv_data_rem -= RDS_FRAG_SIZE;
956 else {
957 ic->i_recv_data_rem = 0;
958 ic->i_ibinc = NULL;
959
960 if (ibinc->ii_inc.i_hdr.h_flags == RDS_FLAG_CONG_BITMAP) {
961 rds_ib_cong_recv(conn, ibinc);
962 } else {
963 rds_recv_incoming(conn, &conn->c_faddr, &conn->c_laddr,
964 &ibinc->ii_inc, GFP_ATOMIC);
965 state->ack_next = be64_to_cpu(hdr->h_sequence);
966 state->ack_next_valid = 1;
967 }
968
969 /* Evaluate the ACK_REQUIRED flag *after* we received
970 * the complete frame, and after bumping the next_rx
971 * sequence. */
972 if (hdr->h_flags & RDS_FLAG_ACK_REQUIRED) {
973 rds_stats_inc(s_recv_ack_required);
974 state->ack_required = 1;
975 }
976
977 rds_inc_put(&ibinc->ii_inc);
978 }
979 done:
980 ib_dma_sync_single_for_device(ic->rds_ibdev->dev, dma_addr,
981 sizeof(*ihdr), DMA_FROM_DEVICE);
982 }
983
984 void rds_ib_recv_cqe_handler(struct rds_ib_connection *ic,
985 struct ib_wc *wc,
986 struct rds_ib_ack_state *state)
987 {
988 struct rds_connection *conn = ic->conn;
989 struct rds_ib_recv_work *recv;
990
991 rdsdebug("wc wr_id 0x%llx status %u (%s) byte_len %u imm_data %u\n",
992 (unsigned long long)wc->wr_id, wc->status,
993 ib_wc_status_msg(wc->status), wc->byte_len,
994 be32_to_cpu(wc->ex.imm_data));
995
996 rds_ib_stats_inc(s_ib_rx_cq_event);
997 recv = &ic->i_recvs[rds_ib_ring_oldest(&ic->i_recv_ring)];
998 ib_dma_unmap_sg(ic->i_cm_id->device, &recv->r_frag->f_sg, 1,
999 DMA_FROM_DEVICE);
1000
1001 /* Also process recvs in connecting state because it is possible
1002 * to get a recv completion _before_ the rdmacm ESTABLISHED
1003 * event is processed.
1004 */
1005 if (wc->status == IB_WC_SUCCESS) {
1006 rds_ib_process_recv(conn, recv, wc->byte_len, state);
1007 } else {
1008 /* We expect errors as the qp is drained during shutdown */
1009 if (rds_conn_up(conn) || rds_conn_connecting(conn))
1010 rds_ib_conn_error(conn, "recv completion on <%pI6c,%pI6c, %d> had status %u (%s), vendor err 0x%x, disconnecting and reconnecting\n",
1011 &conn->c_laddr, &conn->c_faddr,
1012 conn->c_tos, wc->status,
1013 ib_wc_status_msg(wc->status),
1014 wc->vendor_err);
1015 }
1016
1017 /* rds_ib_process_recv() doesn't always consume the frag, and
1018 * we might not have called it at all if the wc didn't indicate
1019 * success. We already unmapped the frag's pages, though, and
1020 * the following rds_ib_ring_free() call tells the refill path
1021 * that it will not find an allocated frag here. Make sure we
1022 * keep that promise by freeing a frag that's still on the ring.
1023 */
1024 if (recv->r_frag) {
1025 rds_ib_frag_free(ic, recv->r_frag);
1026 recv->r_frag = NULL;
1027 }
1028 rds_ib_ring_free(&ic->i_recv_ring, 1);
1029
1030 /* If we ever end up with a really empty receive ring, we're
1031 * in deep trouble, as the sender will definitely see RNR
1032 * timeouts. */
1033 if (rds_ib_ring_empty(&ic->i_recv_ring))
1034 rds_ib_stats_inc(s_ib_rx_ring_empty);
1035
1036 if (rds_ib_ring_low(&ic->i_recv_ring)) {
1037 rds_ib_recv_refill(conn, 0, GFP_NOWAIT | __GFP_NOWARN);
1038 rds_ib_stats_inc(s_ib_rx_refill_from_cq);
1039 }
1040 }
1041
1042 int rds_ib_recv_path(struct rds_conn_path *cp)
1043 {
1044 struct rds_connection *conn = cp->cp_conn;
1045 struct rds_ib_connection *ic = conn->c_transport_data;
1046
1047 rdsdebug("conn %p\n", conn);
1048 if (rds_conn_up(conn)) {
1049 rds_ib_attempt_ack(ic);
1050 rds_ib_recv_refill(conn, 0, GFP_KERNEL);
1051 rds_ib_stats_inc(s_ib_rx_refill_from_thread);
1052 }
1053
1054 return 0;
1055 }
1056
1057 int rds_ib_recv_init(void)
1058 {
1059 struct sysinfo si;
1060 int ret = -ENOMEM;
1061
1062 /* Default to 30% of all available RAM for recv memory */
1063 si_meminfo(&si);
1064 rds_ib_sysctl_max_recv_allocation = si.totalram / 3 * PAGE_SIZE / RDS_FRAG_SIZE;
1065
1066 rds_ib_incoming_slab =
1067 kmem_cache_create_usercopy("rds_ib_incoming",
1068 sizeof(struct rds_ib_incoming),
1069 0, SLAB_HWCACHE_ALIGN,
1070 offsetof(struct rds_ib_incoming,
1071 ii_inc.i_usercopy),
1072 sizeof(struct rds_inc_usercopy),
1073 NULL);
1074 if (!rds_ib_incoming_slab)
1075 goto out;
1076
1077 rds_ib_frag_slab = kmem_cache_create("rds_ib_frag",
1078 sizeof(struct rds_page_frag),
1079 0, SLAB_HWCACHE_ALIGN, NULL);
1080 if (!rds_ib_frag_slab) {
1081 kmem_cache_destroy(rds_ib_incoming_slab);
1082 rds_ib_incoming_slab = NULL;
1083 } else
1084 ret = 0;
1085 out:
1086 return ret;
1087 }
1088
1089 void rds_ib_recv_exit(void)
1090 {
1091 WARN_ON(atomic_read(&rds_ib_allocation));
1092
1093 kmem_cache_destroy(rds_ib_incoming_slab);
1094 kmem_cache_destroy(rds_ib_frag_slab);
1095 }
1096
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.