xref: /openbmc/linux/net/rds/ib_recv.c (revision b627b4ed)
1 /*
2  * Copyright (c) 2006 Oracle.  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/pci.h>
35 #include <linux/dma-mapping.h>
36 #include <rdma/rdma_cm.h>
37 
38 #include "rds.h"
39 #include "ib.h"
40 
41 static struct kmem_cache *rds_ib_incoming_slab;
42 static struct kmem_cache *rds_ib_frag_slab;
43 static atomic_t	rds_ib_allocation = ATOMIC_INIT(0);
44 
45 static void rds_ib_frag_drop_page(struct rds_page_frag *frag)
46 {
47 	rdsdebug("frag %p page %p\n", frag, frag->f_page);
48 	__free_page(frag->f_page);
49 	frag->f_page = NULL;
50 }
51 
52 static void rds_ib_frag_free(struct rds_page_frag *frag)
53 {
54 	rdsdebug("frag %p page %p\n", frag, frag->f_page);
55 	BUG_ON(frag->f_page != NULL);
56 	kmem_cache_free(rds_ib_frag_slab, frag);
57 }
58 
59 /*
60  * We map a page at a time.  Its fragments are posted in order.  This
61  * is called in fragment order as the fragments get send completion events.
62  * Only the last frag in the page performs the unmapping.
63  *
64  * It's OK for ring cleanup to call this in whatever order it likes because
65  * DMA is not in flight and so we can unmap while other ring entries still
66  * hold page references in their frags.
67  */
68 static void rds_ib_recv_unmap_page(struct rds_ib_connection *ic,
69 				   struct rds_ib_recv_work *recv)
70 {
71 	struct rds_page_frag *frag = recv->r_frag;
72 
73 	rdsdebug("recv %p frag %p page %p\n", recv, frag, frag->f_page);
74 	if (frag->f_mapped)
75 		ib_dma_unmap_page(ic->i_cm_id->device,
76 			       frag->f_mapped,
77 			       RDS_FRAG_SIZE, DMA_FROM_DEVICE);
78 	frag->f_mapped = 0;
79 }
80 
81 void rds_ib_recv_init_ring(struct rds_ib_connection *ic)
82 {
83 	struct rds_ib_recv_work *recv;
84 	u32 i;
85 
86 	for (i = 0, recv = ic->i_recvs; i < ic->i_recv_ring.w_nr; i++, recv++) {
87 		struct ib_sge *sge;
88 
89 		recv->r_ibinc = NULL;
90 		recv->r_frag = NULL;
91 
92 		recv->r_wr.next = NULL;
93 		recv->r_wr.wr_id = i;
94 		recv->r_wr.sg_list = recv->r_sge;
95 		recv->r_wr.num_sge = RDS_IB_RECV_SGE;
96 
97 		sge = rds_ib_data_sge(ic, recv->r_sge);
98 		sge->addr = 0;
99 		sge->length = RDS_FRAG_SIZE;
100 		sge->lkey = ic->i_mr->lkey;
101 
102 		sge = rds_ib_header_sge(ic, recv->r_sge);
103 		sge->addr = ic->i_recv_hdrs_dma + (i * sizeof(struct rds_header));
104 		sge->length = sizeof(struct rds_header);
105 		sge->lkey = ic->i_mr->lkey;
106 	}
107 }
108 
109 static void rds_ib_recv_clear_one(struct rds_ib_connection *ic,
110 				  struct rds_ib_recv_work *recv)
111 {
112 	if (recv->r_ibinc) {
113 		rds_inc_put(&recv->r_ibinc->ii_inc);
114 		recv->r_ibinc = NULL;
115 	}
116 	if (recv->r_frag) {
117 		rds_ib_recv_unmap_page(ic, recv);
118 		if (recv->r_frag->f_page)
119 			rds_ib_frag_drop_page(recv->r_frag);
120 		rds_ib_frag_free(recv->r_frag);
121 		recv->r_frag = NULL;
122 	}
123 }
124 
125 void rds_ib_recv_clear_ring(struct rds_ib_connection *ic)
126 {
127 	u32 i;
128 
129 	for (i = 0; i < ic->i_recv_ring.w_nr; i++)
130 		rds_ib_recv_clear_one(ic, &ic->i_recvs[i]);
131 
132 	if (ic->i_frag.f_page)
133 		rds_ib_frag_drop_page(&ic->i_frag);
134 }
135 
136 static int rds_ib_recv_refill_one(struct rds_connection *conn,
137 				  struct rds_ib_recv_work *recv,
138 				  gfp_t kptr_gfp, gfp_t page_gfp)
139 {
140 	struct rds_ib_connection *ic = conn->c_transport_data;
141 	dma_addr_t dma_addr;
142 	struct ib_sge *sge;
143 	int ret = -ENOMEM;
144 
145 	if (recv->r_ibinc == NULL) {
146 		if (atomic_read(&rds_ib_allocation) >= rds_ib_sysctl_max_recv_allocation) {
147 			rds_ib_stats_inc(s_ib_rx_alloc_limit);
148 			goto out;
149 		}
150 		recv->r_ibinc = kmem_cache_alloc(rds_ib_incoming_slab,
151 						 kptr_gfp);
152 		if (recv->r_ibinc == NULL)
153 			goto out;
154 		atomic_inc(&rds_ib_allocation);
155 		INIT_LIST_HEAD(&recv->r_ibinc->ii_frags);
156 		rds_inc_init(&recv->r_ibinc->ii_inc, conn, conn->c_faddr);
157 	}
158 
159 	if (recv->r_frag == NULL) {
160 		recv->r_frag = kmem_cache_alloc(rds_ib_frag_slab, kptr_gfp);
161 		if (recv->r_frag == NULL)
162 			goto out;
163 		INIT_LIST_HEAD(&recv->r_frag->f_item);
164 		recv->r_frag->f_page = NULL;
165 	}
166 
167 	if (ic->i_frag.f_page == NULL) {
168 		ic->i_frag.f_page = alloc_page(page_gfp);
169 		if (ic->i_frag.f_page == NULL)
170 			goto out;
171 		ic->i_frag.f_offset = 0;
172 	}
173 
174 	dma_addr = ib_dma_map_page(ic->i_cm_id->device,
175 				  ic->i_frag.f_page,
176 				  ic->i_frag.f_offset,
177 				  RDS_FRAG_SIZE,
178 				  DMA_FROM_DEVICE);
179 	if (ib_dma_mapping_error(ic->i_cm_id->device, dma_addr))
180 		goto out;
181 
182 	/*
183 	 * Once we get the RDS_PAGE_LAST_OFF frag then rds_ib_frag_unmap()
184 	 * must be called on this recv.  This happens as completions hit
185 	 * in order or on connection shutdown.
186 	 */
187 	recv->r_frag->f_page = ic->i_frag.f_page;
188 	recv->r_frag->f_offset = ic->i_frag.f_offset;
189 	recv->r_frag->f_mapped = dma_addr;
190 
191 	sge = rds_ib_data_sge(ic, recv->r_sge);
192 	sge->addr = dma_addr;
193 	sge->length = RDS_FRAG_SIZE;
194 
195 	sge = rds_ib_header_sge(ic, recv->r_sge);
196 	sge->addr = ic->i_recv_hdrs_dma + (recv - ic->i_recvs) * sizeof(struct rds_header);
197 	sge->length = sizeof(struct rds_header);
198 
199 	get_page(recv->r_frag->f_page);
200 
201 	if (ic->i_frag.f_offset < RDS_PAGE_LAST_OFF) {
202 		ic->i_frag.f_offset += RDS_FRAG_SIZE;
203 	} else {
204 		put_page(ic->i_frag.f_page);
205 		ic->i_frag.f_page = NULL;
206 		ic->i_frag.f_offset = 0;
207 	}
208 
209 	ret = 0;
210 out:
211 	return ret;
212 }
213 
214 /*
215  * This tries to allocate and post unused work requests after making sure that
216  * they have all the allocations they need to queue received fragments into
217  * sockets.  The i_recv_mutex is held here so that ring_alloc and _unalloc
218  * pairs don't go unmatched.
219  *
220  * -1 is returned if posting fails due to temporary resource exhaustion.
221  */
222 int rds_ib_recv_refill(struct rds_connection *conn, gfp_t kptr_gfp,
223 		       gfp_t page_gfp, int prefill)
224 {
225 	struct rds_ib_connection *ic = conn->c_transport_data;
226 	struct rds_ib_recv_work *recv;
227 	struct ib_recv_wr *failed_wr;
228 	unsigned int posted = 0;
229 	int ret = 0;
230 	u32 pos;
231 
232 	while ((prefill || rds_conn_up(conn))
233 			&& rds_ib_ring_alloc(&ic->i_recv_ring, 1, &pos)) {
234 		if (pos >= ic->i_recv_ring.w_nr) {
235 			printk(KERN_NOTICE "Argh - ring alloc returned pos=%u\n",
236 					pos);
237 			ret = -EINVAL;
238 			break;
239 		}
240 
241 		recv = &ic->i_recvs[pos];
242 		ret = rds_ib_recv_refill_one(conn, recv, kptr_gfp, page_gfp);
243 		if (ret) {
244 			ret = -1;
245 			break;
246 		}
247 
248 		/* XXX when can this fail? */
249 		ret = ib_post_recv(ic->i_cm_id->qp, &recv->r_wr, &failed_wr);
250 		rdsdebug("recv %p ibinc %p page %p addr %lu ret %d\n", recv,
251 			 recv->r_ibinc, recv->r_frag->f_page,
252 			 (long) recv->r_frag->f_mapped, ret);
253 		if (ret) {
254 			rds_ib_conn_error(conn, "recv post on "
255 			       "%pI4 returned %d, disconnecting and "
256 			       "reconnecting\n", &conn->c_faddr,
257 			       ret);
258 			ret = -1;
259 			break;
260 		}
261 
262 		posted++;
263 	}
264 
265 	/* We're doing flow control - update the window. */
266 	if (ic->i_flowctl && posted)
267 		rds_ib_advertise_credits(conn, posted);
268 
269 	if (ret)
270 		rds_ib_ring_unalloc(&ic->i_recv_ring, 1);
271 	return ret;
272 }
273 
274 void rds_ib_inc_purge(struct rds_incoming *inc)
275 {
276 	struct rds_ib_incoming *ibinc;
277 	struct rds_page_frag *frag;
278 	struct rds_page_frag *pos;
279 
280 	ibinc = container_of(inc, struct rds_ib_incoming, ii_inc);
281 	rdsdebug("purging ibinc %p inc %p\n", ibinc, inc);
282 
283 	list_for_each_entry_safe(frag, pos, &ibinc->ii_frags, f_item) {
284 		list_del_init(&frag->f_item);
285 		rds_ib_frag_drop_page(frag);
286 		rds_ib_frag_free(frag);
287 	}
288 }
289 
290 void rds_ib_inc_free(struct rds_incoming *inc)
291 {
292 	struct rds_ib_incoming *ibinc;
293 
294 	ibinc = container_of(inc, struct rds_ib_incoming, ii_inc);
295 
296 	rds_ib_inc_purge(inc);
297 	rdsdebug("freeing ibinc %p inc %p\n", ibinc, inc);
298 	BUG_ON(!list_empty(&ibinc->ii_frags));
299 	kmem_cache_free(rds_ib_incoming_slab, ibinc);
300 	atomic_dec(&rds_ib_allocation);
301 	BUG_ON(atomic_read(&rds_ib_allocation) < 0);
302 }
303 
304 int rds_ib_inc_copy_to_user(struct rds_incoming *inc, struct iovec *first_iov,
305 			    size_t size)
306 {
307 	struct rds_ib_incoming *ibinc;
308 	struct rds_page_frag *frag;
309 	struct iovec *iov = first_iov;
310 	unsigned long to_copy;
311 	unsigned long frag_off = 0;
312 	unsigned long iov_off = 0;
313 	int copied = 0;
314 	int ret;
315 	u32 len;
316 
317 	ibinc = container_of(inc, struct rds_ib_incoming, ii_inc);
318 	frag = list_entry(ibinc->ii_frags.next, struct rds_page_frag, f_item);
319 	len = be32_to_cpu(inc->i_hdr.h_len);
320 
321 	while (copied < size && copied < len) {
322 		if (frag_off == RDS_FRAG_SIZE) {
323 			frag = list_entry(frag->f_item.next,
324 					  struct rds_page_frag, f_item);
325 			frag_off = 0;
326 		}
327 		while (iov_off == iov->iov_len) {
328 			iov_off = 0;
329 			iov++;
330 		}
331 
332 		to_copy = min(iov->iov_len - iov_off, RDS_FRAG_SIZE - frag_off);
333 		to_copy = min_t(size_t, to_copy, size - copied);
334 		to_copy = min_t(unsigned long, to_copy, len - copied);
335 
336 		rdsdebug("%lu bytes to user [%p, %zu] + %lu from frag "
337 			 "[%p, %lu] + %lu\n",
338 			 to_copy, iov->iov_base, iov->iov_len, iov_off,
339 			 frag->f_page, frag->f_offset, frag_off);
340 
341 		/* XXX needs + offset for multiple recvs per page */
342 		ret = rds_page_copy_to_user(frag->f_page,
343 					    frag->f_offset + frag_off,
344 					    iov->iov_base + iov_off,
345 					    to_copy);
346 		if (ret) {
347 			copied = ret;
348 			break;
349 		}
350 
351 		iov_off += to_copy;
352 		frag_off += to_copy;
353 		copied += to_copy;
354 	}
355 
356 	return copied;
357 }
358 
359 /* ic starts out kzalloc()ed */
360 void rds_ib_recv_init_ack(struct rds_ib_connection *ic)
361 {
362 	struct ib_send_wr *wr = &ic->i_ack_wr;
363 	struct ib_sge *sge = &ic->i_ack_sge;
364 
365 	sge->addr = ic->i_ack_dma;
366 	sge->length = sizeof(struct rds_header);
367 	sge->lkey = ic->i_mr->lkey;
368 
369 	wr->sg_list = sge;
370 	wr->num_sge = 1;
371 	wr->opcode = IB_WR_SEND;
372 	wr->wr_id = RDS_IB_ACK_WR_ID;
373 	wr->send_flags = IB_SEND_SIGNALED | IB_SEND_SOLICITED;
374 }
375 
376 /*
377  * You'd think that with reliable IB connections you wouldn't need to ack
378  * messages that have been received.  The problem is that IB hardware generates
379  * an ack message before it has DMAed the message into memory.  This creates a
380  * potential message loss if the HCA is disabled for any reason between when it
381  * sends the ack and before the message is DMAed and processed.  This is only a
382  * potential issue if another HCA is available for fail-over.
383  *
384  * When the remote host receives our ack they'll free the sent message from
385  * their send queue.  To decrease the latency of this we always send an ack
386  * immediately after we've received messages.
387  *
388  * For simplicity, we only have one ack in flight at a time.  This puts
389  * pressure on senders to have deep enough send queues to absorb the latency of
390  * a single ack frame being in flight.  This might not be good enough.
391  *
392  * This is implemented by have a long-lived send_wr and sge which point to a
393  * statically allocated ack frame.  This ack wr does not fall under the ring
394  * accounting that the tx and rx wrs do.  The QP attribute specifically makes
395  * room for it beyond the ring size.  Send completion notices its special
396  * wr_id and avoids working with the ring in that case.
397  */
398 #ifndef KERNEL_HAS_ATOMIC64
399 static void rds_ib_set_ack(struct rds_ib_connection *ic, u64 seq,
400 				int ack_required)
401 {
402 	unsigned long flags;
403 
404 	spin_lock_irqsave(&ic->i_ack_lock, flags);
405 	ic->i_ack_next = seq;
406 	if (ack_required)
407 		set_bit(IB_ACK_REQUESTED, &ic->i_ack_flags);
408 	spin_unlock_irqrestore(&ic->i_ack_lock, flags);
409 }
410 
411 static u64 rds_ib_get_ack(struct rds_ib_connection *ic)
412 {
413 	unsigned long flags;
414 	u64 seq;
415 
416 	clear_bit(IB_ACK_REQUESTED, &ic->i_ack_flags);
417 
418 	spin_lock_irqsave(&ic->i_ack_lock, flags);
419 	seq = ic->i_ack_next;
420 	spin_unlock_irqrestore(&ic->i_ack_lock, flags);
421 
422 	return seq;
423 }
424 #else
425 static void rds_ib_set_ack(struct rds_ib_connection *ic, u64 seq,
426 				int ack_required)
427 {
428 	atomic64_set(&ic->i_ack_next, seq);
429 	if (ack_required) {
430 		smp_mb__before_clear_bit();
431 		set_bit(IB_ACK_REQUESTED, &ic->i_ack_flags);
432 	}
433 }
434 
435 static u64 rds_ib_get_ack(struct rds_ib_connection *ic)
436 {
437 	clear_bit(IB_ACK_REQUESTED, &ic->i_ack_flags);
438 	smp_mb__after_clear_bit();
439 
440 	return atomic64_read(&ic->i_ack_next);
441 }
442 #endif
443 
444 
445 static void rds_ib_send_ack(struct rds_ib_connection *ic, unsigned int adv_credits)
446 {
447 	struct rds_header *hdr = ic->i_ack;
448 	struct ib_send_wr *failed_wr;
449 	u64 seq;
450 	int ret;
451 
452 	seq = rds_ib_get_ack(ic);
453 
454 	rdsdebug("send_ack: ic %p ack %llu\n", ic, (unsigned long long) seq);
455 	rds_message_populate_header(hdr, 0, 0, 0);
456 	hdr->h_ack = cpu_to_be64(seq);
457 	hdr->h_credit = adv_credits;
458 	rds_message_make_checksum(hdr);
459 	ic->i_ack_queued = jiffies;
460 
461 	ret = ib_post_send(ic->i_cm_id->qp, &ic->i_ack_wr, &failed_wr);
462 	if (unlikely(ret)) {
463 		/* Failed to send. Release the WR, and
464 		 * force another ACK.
465 		 */
466 		clear_bit(IB_ACK_IN_FLIGHT, &ic->i_ack_flags);
467 		set_bit(IB_ACK_REQUESTED, &ic->i_ack_flags);
468 
469 		rds_ib_stats_inc(s_ib_ack_send_failure);
470 		/* Need to finesse this later. */
471 		BUG();
472 	} else
473 		rds_ib_stats_inc(s_ib_ack_sent);
474 }
475 
476 /*
477  * There are 3 ways of getting acknowledgements to the peer:
478  *  1.	We call rds_ib_attempt_ack from the recv completion handler
479  *	to send an ACK-only frame.
480  *	However, there can be only one such frame in the send queue
481  *	at any time, so we may have to postpone it.
482  *  2.	When another (data) packet is transmitted while there's
483  *	an ACK in the queue, we piggyback the ACK sequence number
484  *	on the data packet.
485  *  3.	If the ACK WR is done sending, we get called from the
486  *	send queue completion handler, and check whether there's
487  *	another ACK pending (postponed because the WR was on the
488  *	queue). If so, we transmit it.
489  *
490  * We maintain 2 variables:
491  *  -	i_ack_flags, which keeps track of whether the ACK WR
492  *	is currently in the send queue or not (IB_ACK_IN_FLIGHT)
493  *  -	i_ack_next, which is the last sequence number we received
494  *
495  * Potentially, send queue and receive queue handlers can run concurrently.
496  * It would be nice to not have to use a spinlock to synchronize things,
497  * but the one problem that rules this out is that 64bit updates are
498  * not atomic on all platforms. Things would be a lot simpler if
499  * we had atomic64 or maybe cmpxchg64 everywhere.
500  *
501  * Reconnecting complicates this picture just slightly. When we
502  * reconnect, we may be seeing duplicate packets. The peer
503  * is retransmitting them, because it hasn't seen an ACK for
504  * them. It is important that we ACK these.
505  *
506  * ACK mitigation adds a header flag "ACK_REQUIRED"; any packet with
507  * this flag set *MUST* be acknowledged immediately.
508  */
509 
510 /*
511  * When we get here, we're called from the recv queue handler.
512  * Check whether we ought to transmit an ACK.
513  */
514 void rds_ib_attempt_ack(struct rds_ib_connection *ic)
515 {
516 	unsigned int adv_credits;
517 
518 	if (!test_bit(IB_ACK_REQUESTED, &ic->i_ack_flags))
519 		return;
520 
521 	if (test_and_set_bit(IB_ACK_IN_FLIGHT, &ic->i_ack_flags)) {
522 		rds_ib_stats_inc(s_ib_ack_send_delayed);
523 		return;
524 	}
525 
526 	/* Can we get a send credit? */
527 	if (!rds_ib_send_grab_credits(ic, 1, &adv_credits, 0)) {
528 		rds_ib_stats_inc(s_ib_tx_throttle);
529 		clear_bit(IB_ACK_IN_FLIGHT, &ic->i_ack_flags);
530 		return;
531 	}
532 
533 	clear_bit(IB_ACK_REQUESTED, &ic->i_ack_flags);
534 	rds_ib_send_ack(ic, adv_credits);
535 }
536 
537 /*
538  * We get here from the send completion handler, when the
539  * adapter tells us the ACK frame was sent.
540  */
541 void rds_ib_ack_send_complete(struct rds_ib_connection *ic)
542 {
543 	clear_bit(IB_ACK_IN_FLIGHT, &ic->i_ack_flags);
544 	rds_ib_attempt_ack(ic);
545 }
546 
547 /*
548  * This is called by the regular xmit code when it wants to piggyback
549  * an ACK on an outgoing frame.
550  */
551 u64 rds_ib_piggyb_ack(struct rds_ib_connection *ic)
552 {
553 	if (test_and_clear_bit(IB_ACK_REQUESTED, &ic->i_ack_flags))
554 		rds_ib_stats_inc(s_ib_ack_send_piggybacked);
555 	return rds_ib_get_ack(ic);
556 }
557 
558 /*
559  * It's kind of lame that we're copying from the posted receive pages into
560  * long-lived bitmaps.  We could have posted the bitmaps and rdma written into
561  * them.  But receiving new congestion bitmaps should be a *rare* event, so
562  * hopefully we won't need to invest that complexity in making it more
563  * efficient.  By copying we can share a simpler core with TCP which has to
564  * copy.
565  */
566 static void rds_ib_cong_recv(struct rds_connection *conn,
567 			      struct rds_ib_incoming *ibinc)
568 {
569 	struct rds_cong_map *map;
570 	unsigned int map_off;
571 	unsigned int map_page;
572 	struct rds_page_frag *frag;
573 	unsigned long frag_off;
574 	unsigned long to_copy;
575 	unsigned long copied;
576 	uint64_t uncongested = 0;
577 	void *addr;
578 
579 	/* catch completely corrupt packets */
580 	if (be32_to_cpu(ibinc->ii_inc.i_hdr.h_len) != RDS_CONG_MAP_BYTES)
581 		return;
582 
583 	map = conn->c_fcong;
584 	map_page = 0;
585 	map_off = 0;
586 
587 	frag = list_entry(ibinc->ii_frags.next, struct rds_page_frag, f_item);
588 	frag_off = 0;
589 
590 	copied = 0;
591 
592 	while (copied < RDS_CONG_MAP_BYTES) {
593 		uint64_t *src, *dst;
594 		unsigned int k;
595 
596 		to_copy = min(RDS_FRAG_SIZE - frag_off, PAGE_SIZE - map_off);
597 		BUG_ON(to_copy & 7); /* Must be 64bit aligned. */
598 
599 		addr = kmap_atomic(frag->f_page, KM_SOFTIRQ0);
600 
601 		src = addr + frag_off;
602 		dst = (void *)map->m_page_addrs[map_page] + map_off;
603 		for (k = 0; k < to_copy; k += 8) {
604 			/* Record ports that became uncongested, ie
605 			 * bits that changed from 0 to 1. */
606 			uncongested |= ~(*src) & *dst;
607 			*dst++ = *src++;
608 		}
609 		kunmap_atomic(addr, KM_SOFTIRQ0);
610 
611 		copied += to_copy;
612 
613 		map_off += to_copy;
614 		if (map_off == PAGE_SIZE) {
615 			map_off = 0;
616 			map_page++;
617 		}
618 
619 		frag_off += to_copy;
620 		if (frag_off == RDS_FRAG_SIZE) {
621 			frag = list_entry(frag->f_item.next,
622 					  struct rds_page_frag, f_item);
623 			frag_off = 0;
624 		}
625 	}
626 
627 	/* the congestion map is in little endian order */
628 	uncongested = le64_to_cpu(uncongested);
629 
630 	rds_cong_map_updated(map, uncongested);
631 }
632 
633 /*
634  * Rings are posted with all the allocations they'll need to queue the
635  * incoming message to the receiving socket so this can't fail.
636  * All fragments start with a header, so we can make sure we're not receiving
637  * garbage, and we can tell a small 8 byte fragment from an ACK frame.
638  */
639 struct rds_ib_ack_state {
640 	u64		ack_next;
641 	u64		ack_recv;
642 	unsigned int	ack_required:1;
643 	unsigned int	ack_next_valid:1;
644 	unsigned int	ack_recv_valid:1;
645 };
646 
647 static void rds_ib_process_recv(struct rds_connection *conn,
648 				struct rds_ib_recv_work *recv, u32 byte_len,
649 				struct rds_ib_ack_state *state)
650 {
651 	struct rds_ib_connection *ic = conn->c_transport_data;
652 	struct rds_ib_incoming *ibinc = ic->i_ibinc;
653 	struct rds_header *ihdr, *hdr;
654 
655 	/* XXX shut down the connection if port 0,0 are seen? */
656 
657 	rdsdebug("ic %p ibinc %p recv %p byte len %u\n", ic, ibinc, recv,
658 		 byte_len);
659 
660 	if (byte_len < sizeof(struct rds_header)) {
661 		rds_ib_conn_error(conn, "incoming message "
662 		       "from %pI4 didn't inclue a "
663 		       "header, disconnecting and "
664 		       "reconnecting\n",
665 		       &conn->c_faddr);
666 		return;
667 	}
668 	byte_len -= sizeof(struct rds_header);
669 
670 	ihdr = &ic->i_recv_hdrs[recv - ic->i_recvs];
671 
672 	/* Validate the checksum. */
673 	if (!rds_message_verify_checksum(ihdr)) {
674 		rds_ib_conn_error(conn, "incoming message "
675 		       "from %pI4 has corrupted header - "
676 		       "forcing a reconnect\n",
677 		       &conn->c_faddr);
678 		rds_stats_inc(s_recv_drop_bad_checksum);
679 		return;
680 	}
681 
682 	/* Process the ACK sequence which comes with every packet */
683 	state->ack_recv = be64_to_cpu(ihdr->h_ack);
684 	state->ack_recv_valid = 1;
685 
686 	/* Process the credits update if there was one */
687 	if (ihdr->h_credit)
688 		rds_ib_send_add_credits(conn, ihdr->h_credit);
689 
690 	if (ihdr->h_sport == 0 && ihdr->h_dport == 0 && byte_len == 0) {
691 		/* This is an ACK-only packet. The fact that it gets
692 		 * special treatment here is that historically, ACKs
693 		 * were rather special beasts.
694 		 */
695 		rds_ib_stats_inc(s_ib_ack_received);
696 
697 		/*
698 		 * Usually the frags make their way on to incs and are then freed as
699 		 * the inc is freed.  We don't go that route, so we have to drop the
700 		 * page ref ourselves.  We can't just leave the page on the recv
701 		 * because that confuses the dma mapping of pages and each recv's use
702 		 * of a partial page.  We can leave the frag, though, it will be
703 		 * reused.
704 		 *
705 		 * FIXME: Fold this into the code path below.
706 		 */
707 		rds_ib_frag_drop_page(recv->r_frag);
708 		return;
709 	}
710 
711 	/*
712 	 * If we don't already have an inc on the connection then this
713 	 * fragment has a header and starts a message.. copy its header
714 	 * into the inc and save the inc so we can hang upcoming fragments
715 	 * off its list.
716 	 */
717 	if (ibinc == NULL) {
718 		ibinc = recv->r_ibinc;
719 		recv->r_ibinc = NULL;
720 		ic->i_ibinc = ibinc;
721 
722 		hdr = &ibinc->ii_inc.i_hdr;
723 		memcpy(hdr, ihdr, sizeof(*hdr));
724 		ic->i_recv_data_rem = be32_to_cpu(hdr->h_len);
725 
726 		rdsdebug("ic %p ibinc %p rem %u flag 0x%x\n", ic, ibinc,
727 			 ic->i_recv_data_rem, hdr->h_flags);
728 	} else {
729 		hdr = &ibinc->ii_inc.i_hdr;
730 		/* We can't just use memcmp here; fragments of a
731 		 * single message may carry different ACKs */
732 		if (hdr->h_sequence != ihdr->h_sequence
733 		 || hdr->h_len != ihdr->h_len
734 		 || hdr->h_sport != ihdr->h_sport
735 		 || hdr->h_dport != ihdr->h_dport) {
736 			rds_ib_conn_error(conn,
737 				"fragment header mismatch; forcing reconnect\n");
738 			return;
739 		}
740 	}
741 
742 	list_add_tail(&recv->r_frag->f_item, &ibinc->ii_frags);
743 	recv->r_frag = NULL;
744 
745 	if (ic->i_recv_data_rem > RDS_FRAG_SIZE)
746 		ic->i_recv_data_rem -= RDS_FRAG_SIZE;
747 	else {
748 		ic->i_recv_data_rem = 0;
749 		ic->i_ibinc = NULL;
750 
751 		if (ibinc->ii_inc.i_hdr.h_flags == RDS_FLAG_CONG_BITMAP)
752 			rds_ib_cong_recv(conn, ibinc);
753 		else {
754 			rds_recv_incoming(conn, conn->c_faddr, conn->c_laddr,
755 					  &ibinc->ii_inc, GFP_ATOMIC,
756 					  KM_SOFTIRQ0);
757 			state->ack_next = be64_to_cpu(hdr->h_sequence);
758 			state->ack_next_valid = 1;
759 		}
760 
761 		/* Evaluate the ACK_REQUIRED flag *after* we received
762 		 * the complete frame, and after bumping the next_rx
763 		 * sequence. */
764 		if (hdr->h_flags & RDS_FLAG_ACK_REQUIRED) {
765 			rds_stats_inc(s_recv_ack_required);
766 			state->ack_required = 1;
767 		}
768 
769 		rds_inc_put(&ibinc->ii_inc);
770 	}
771 }
772 
773 /*
774  * Plucking the oldest entry from the ring can be done concurrently with
775  * the thread refilling the ring.  Each ring operation is protected by
776  * spinlocks and the transient state of refilling doesn't change the
777  * recording of which entry is oldest.
778  *
779  * This relies on IB only calling one cq comp_handler for each cq so that
780  * there will only be one caller of rds_recv_incoming() per RDS connection.
781  */
782 void rds_ib_recv_cq_comp_handler(struct ib_cq *cq, void *context)
783 {
784 	struct rds_connection *conn = context;
785 	struct rds_ib_connection *ic = conn->c_transport_data;
786 	struct ib_wc wc;
787 	struct rds_ib_ack_state state = { 0, };
788 	struct rds_ib_recv_work *recv;
789 
790 	rdsdebug("conn %p cq %p\n", conn, cq);
791 
792 	rds_ib_stats_inc(s_ib_rx_cq_call);
793 
794 	ib_req_notify_cq(cq, IB_CQ_SOLICITED);
795 
796 	while (ib_poll_cq(cq, 1, &wc) > 0) {
797 		rdsdebug("wc wr_id 0x%llx status %u byte_len %u imm_data %u\n",
798 			 (unsigned long long)wc.wr_id, wc.status, wc.byte_len,
799 			 be32_to_cpu(wc.ex.imm_data));
800 		rds_ib_stats_inc(s_ib_rx_cq_event);
801 
802 		recv = &ic->i_recvs[rds_ib_ring_oldest(&ic->i_recv_ring)];
803 
804 		rds_ib_recv_unmap_page(ic, recv);
805 
806 		/*
807 		 * Also process recvs in connecting state because it is possible
808 		 * to get a recv completion _before_ the rdmacm ESTABLISHED
809 		 * event is processed.
810 		 */
811 		if (rds_conn_up(conn) || rds_conn_connecting(conn)) {
812 			/* We expect errors as the qp is drained during shutdown */
813 			if (wc.status == IB_WC_SUCCESS) {
814 				rds_ib_process_recv(conn, recv, wc.byte_len, &state);
815 			} else {
816 				rds_ib_conn_error(conn, "recv completion on "
817 				       "%pI4 had status %u, disconnecting and "
818 				       "reconnecting\n", &conn->c_faddr,
819 				       wc.status);
820 			}
821 		}
822 
823 		rds_ib_ring_free(&ic->i_recv_ring, 1);
824 	}
825 
826 	if (state.ack_next_valid)
827 		rds_ib_set_ack(ic, state.ack_next, state.ack_required);
828 	if (state.ack_recv_valid && state.ack_recv > ic->i_ack_recv) {
829 		rds_send_drop_acked(conn, state.ack_recv, NULL);
830 		ic->i_ack_recv = state.ack_recv;
831 	}
832 	if (rds_conn_up(conn))
833 		rds_ib_attempt_ack(ic);
834 
835 	/* If we ever end up with a really empty receive ring, we're
836 	 * in deep trouble, as the sender will definitely see RNR
837 	 * timeouts. */
838 	if (rds_ib_ring_empty(&ic->i_recv_ring))
839 		rds_ib_stats_inc(s_ib_rx_ring_empty);
840 
841 	/*
842 	 * If the ring is running low, then schedule the thread to refill.
843 	 */
844 	if (rds_ib_ring_low(&ic->i_recv_ring))
845 		queue_delayed_work(rds_wq, &conn->c_recv_w, 0);
846 }
847 
848 int rds_ib_recv(struct rds_connection *conn)
849 {
850 	struct rds_ib_connection *ic = conn->c_transport_data;
851 	int ret = 0;
852 
853 	rdsdebug("conn %p\n", conn);
854 
855 	/*
856 	 * If we get a temporary posting failure in this context then
857 	 * we're really low and we want the caller to back off for a bit.
858 	 */
859 	mutex_lock(&ic->i_recv_mutex);
860 	if (rds_ib_recv_refill(conn, GFP_KERNEL, GFP_HIGHUSER, 0))
861 		ret = -ENOMEM;
862 	else
863 		rds_ib_stats_inc(s_ib_rx_refill_from_thread);
864 	mutex_unlock(&ic->i_recv_mutex);
865 
866 	if (rds_conn_up(conn))
867 		rds_ib_attempt_ack(ic);
868 
869 	return ret;
870 }
871 
872 int __init rds_ib_recv_init(void)
873 {
874 	struct sysinfo si;
875 	int ret = -ENOMEM;
876 
877 	/* Default to 30% of all available RAM for recv memory */
878 	si_meminfo(&si);
879 	rds_ib_sysctl_max_recv_allocation = si.totalram / 3 * PAGE_SIZE / RDS_FRAG_SIZE;
880 
881 	rds_ib_incoming_slab = kmem_cache_create("rds_ib_incoming",
882 					sizeof(struct rds_ib_incoming),
883 					0, 0, NULL);
884 	if (rds_ib_incoming_slab == NULL)
885 		goto out;
886 
887 	rds_ib_frag_slab = kmem_cache_create("rds_ib_frag",
888 					sizeof(struct rds_page_frag),
889 					0, 0, NULL);
890 	if (rds_ib_frag_slab == NULL)
891 		kmem_cache_destroy(rds_ib_incoming_slab);
892 	else
893 		ret = 0;
894 out:
895 	return ret;
896 }
897 
898 void rds_ib_recv_exit(void)
899 {
900 	kmem_cache_destroy(rds_ib_incoming_slab);
901 	kmem_cache_destroy(rds_ib_frag_slab);
902 }
903