2 * Copyright (c) 2006 Oracle. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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.
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
33 #include <linux/kernel.h>
35 #include <linux/device.h>
36 #include <linux/dmapool.h>
37 #include <linux/ratelimit.h>
42 static char *rds_ib_wc_status_strings[] = {
43 #define RDS_IB_WC_STATUS_STR(foo) \
44 [IB_WC_##foo] = __stringify(IB_WC_##foo)
45 RDS_IB_WC_STATUS_STR(SUCCESS),
46 RDS_IB_WC_STATUS_STR(LOC_LEN_ERR),
47 RDS_IB_WC_STATUS_STR(LOC_QP_OP_ERR),
48 RDS_IB_WC_STATUS_STR(LOC_EEC_OP_ERR),
49 RDS_IB_WC_STATUS_STR(LOC_PROT_ERR),
50 RDS_IB_WC_STATUS_STR(WR_FLUSH_ERR),
51 RDS_IB_WC_STATUS_STR(MW_BIND_ERR),
52 RDS_IB_WC_STATUS_STR(BAD_RESP_ERR),
53 RDS_IB_WC_STATUS_STR(LOC_ACCESS_ERR),
54 RDS_IB_WC_STATUS_STR(REM_INV_REQ_ERR),
55 RDS_IB_WC_STATUS_STR(REM_ACCESS_ERR),
56 RDS_IB_WC_STATUS_STR(REM_OP_ERR),
57 RDS_IB_WC_STATUS_STR(RETRY_EXC_ERR),
58 RDS_IB_WC_STATUS_STR(RNR_RETRY_EXC_ERR),
59 RDS_IB_WC_STATUS_STR(LOC_RDD_VIOL_ERR),
60 RDS_IB_WC_STATUS_STR(REM_INV_RD_REQ_ERR),
61 RDS_IB_WC_STATUS_STR(REM_ABORT_ERR),
62 RDS_IB_WC_STATUS_STR(INV_EECN_ERR),
63 RDS_IB_WC_STATUS_STR(INV_EEC_STATE_ERR),
64 RDS_IB_WC_STATUS_STR(FATAL_ERR),
65 RDS_IB_WC_STATUS_STR(RESP_TIMEOUT_ERR),
66 RDS_IB_WC_STATUS_STR(GENERAL_ERR),
67 #undef RDS_IB_WC_STATUS_STR
70 char *rds_ib_wc_status_str(enum ib_wc_status status)
72 return rds_str_array(rds_ib_wc_status_strings,
73 ARRAY_SIZE(rds_ib_wc_status_strings), status);
77 * Convert IB-specific error message to RDS error message and call core
80 static void rds_ib_send_complete(struct rds_message *rm,
82 void (*complete)(struct rds_message *rm, int status))
87 case IB_WC_WR_FLUSH_ERR:
91 notify_status = RDS_RDMA_SUCCESS;
94 case IB_WC_REM_ACCESS_ERR:
95 notify_status = RDS_RDMA_REMOTE_ERROR;
99 notify_status = RDS_RDMA_OTHER_ERROR;
102 complete(rm, notify_status);
105 static void rds_ib_send_unmap_data(struct rds_ib_connection *ic,
106 struct rm_data_op *op,
110 ib_dma_unmap_sg(ic->i_cm_id->device,
111 op->op_sg, op->op_nents,
115 static void rds_ib_send_unmap_rdma(struct rds_ib_connection *ic,
116 struct rm_rdma_op *op,
120 ib_dma_unmap_sg(ic->i_cm_id->device,
121 op->op_sg, op->op_nents,
122 op->op_write ? DMA_TO_DEVICE : DMA_FROM_DEVICE);
126 /* If the user asked for a completion notification on this
127 * message, we can implement three different semantics:
128 * 1. Notify when we received the ACK on the RDS message
129 * that was queued with the RDMA. This provides reliable
130 * notification of RDMA status at the expense of a one-way
132 * 2. Notify when the IB stack gives us the completion event for
133 * the RDMA operation.
134 * 3. Notify when the IB stack gives us the completion event for
135 * the accompanying RDS messages.
136 * Here, we implement approach #3. To implement approach #2,
137 * we would need to take an event for the rdma WR. To implement #1,
138 * don't call rds_rdma_send_complete at all, and fall back to the notify
139 * handling in the ACK processing code.
141 * Note: There's no need to explicitly sync any RDMA buffers using
142 * ib_dma_sync_sg_for_cpu - the completion for the RDMA
143 * operation itself unmapped the RDMA buffers, which takes care
146 rds_ib_send_complete(container_of(op, struct rds_message, rdma),
147 wc_status, rds_rdma_send_complete);
150 rds_stats_add(s_send_rdma_bytes, op->op_bytes);
152 rds_stats_add(s_recv_rdma_bytes, op->op_bytes);
155 static void rds_ib_send_unmap_atomic(struct rds_ib_connection *ic,
156 struct rm_atomic_op *op,
159 /* unmap atomic recvbuf */
161 ib_dma_unmap_sg(ic->i_cm_id->device, op->op_sg, 1,
166 rds_ib_send_complete(container_of(op, struct rds_message, atomic),
167 wc_status, rds_atomic_send_complete);
169 if (op->op_type == RDS_ATOMIC_TYPE_CSWP)
170 rds_ib_stats_inc(s_ib_atomic_cswp);
172 rds_ib_stats_inc(s_ib_atomic_fadd);
176 * Unmap the resources associated with a struct send_work.
178 * Returns the rm for no good reason other than it is unobtainable
179 * other than by switching on wr.opcode, currently, and the caller,
180 * the event handler, needs it.
182 static struct rds_message *rds_ib_send_unmap_op(struct rds_ib_connection *ic,
183 struct rds_ib_send_work *send,
186 struct rds_message *rm = NULL;
188 /* In the error case, wc.opcode sometimes contains garbage */
189 switch (send->s_wr.opcode) {
192 rm = container_of(send->s_op, struct rds_message, data);
193 rds_ib_send_unmap_data(ic, send->s_op, wc_status);
196 case IB_WR_RDMA_WRITE:
197 case IB_WR_RDMA_READ:
199 rm = container_of(send->s_op, struct rds_message, rdma);
200 rds_ib_send_unmap_rdma(ic, send->s_op, wc_status);
203 case IB_WR_ATOMIC_FETCH_AND_ADD:
204 case IB_WR_ATOMIC_CMP_AND_SWP:
206 rm = container_of(send->s_op, struct rds_message, atomic);
207 rds_ib_send_unmap_atomic(ic, send->s_op, wc_status);
211 printk_ratelimited(KERN_NOTICE
212 "RDS/IB: %s: unexpected opcode 0x%x in WR!\n",
213 __func__, send->s_wr.opcode);
217 send->s_wr.opcode = 0xdead;
222 void rds_ib_send_init_ring(struct rds_ib_connection *ic)
224 struct rds_ib_send_work *send;
227 for (i = 0, send = ic->i_sends; i < ic->i_send_ring.w_nr; i++, send++) {
232 send->s_wr.wr_id = i;
233 send->s_wr.sg_list = send->s_sge;
234 send->s_wr.ex.imm_data = 0;
236 sge = &send->s_sge[0];
237 sge->addr = ic->i_send_hdrs_dma + (i * sizeof(struct rds_header));
238 sge->length = sizeof(struct rds_header);
239 sge->lkey = ic->i_mr->lkey;
241 send->s_sge[1].lkey = ic->i_mr->lkey;
245 void rds_ib_send_clear_ring(struct rds_ib_connection *ic)
247 struct rds_ib_send_work *send;
250 for (i = 0, send = ic->i_sends; i < ic->i_send_ring.w_nr; i++, send++) {
251 if (send->s_op && send->s_wr.opcode != 0xdead)
252 rds_ib_send_unmap_op(ic, send, IB_WC_WR_FLUSH_ERR);
257 * The only fast path caller always has a non-zero nr, so we don't
258 * bother testing nr before performing the atomic sub.
260 static void rds_ib_sub_signaled(struct rds_ib_connection *ic, int nr)
262 if ((atomic_sub_return(nr, &ic->i_signaled_sends) == 0) &&
263 waitqueue_active(&rds_ib_ring_empty_wait))
264 wake_up(&rds_ib_ring_empty_wait);
265 BUG_ON(atomic_read(&ic->i_signaled_sends) < 0);
269 * The _oldest/_free ring operations here race cleanly with the alloc/unalloc
270 * operations performed in the send path. As the sender allocs and potentially
271 * unallocs the next free entry in the ring it doesn't alter which is
272 * the next to be freed, which is what this is concerned with.
274 void rds_ib_send_cq_comp_handler(struct ib_cq *cq, void *context)
276 struct rds_connection *conn = context;
277 struct rds_ib_connection *ic = conn->c_transport_data;
278 struct rds_message *rm = NULL;
280 struct rds_ib_send_work *send;
287 rdsdebug("cq %p conn %p\n", cq, conn);
288 rds_ib_stats_inc(s_ib_tx_cq_call);
289 ret = ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
291 rdsdebug("ib_req_notify_cq send failed: %d\n", ret);
293 while (ib_poll_cq(cq, 1, &wc) > 0) {
294 rdsdebug("wc wr_id 0x%llx status %u (%s) byte_len %u imm_data %u\n",
295 (unsigned long long)wc.wr_id, wc.status,
296 rds_ib_wc_status_str(wc.status), wc.byte_len,
297 be32_to_cpu(wc.ex.imm_data));
298 rds_ib_stats_inc(s_ib_tx_cq_event);
300 if (wc.wr_id == RDS_IB_ACK_WR_ID) {
301 if (ic->i_ack_queued + HZ/2 < jiffies)
302 rds_ib_stats_inc(s_ib_tx_stalled);
303 rds_ib_ack_send_complete(ic);
307 oldest = rds_ib_ring_oldest(&ic->i_send_ring);
309 completed = rds_ib_ring_completed(&ic->i_send_ring, wc.wr_id, oldest);
311 for (i = 0; i < completed; i++) {
312 send = &ic->i_sends[oldest];
313 if (send->s_wr.send_flags & IB_SEND_SIGNALED)
316 rm = rds_ib_send_unmap_op(ic, send, wc.status);
318 if (send->s_queued + HZ/2 < jiffies)
319 rds_ib_stats_inc(s_ib_tx_stalled);
322 if (send->s_op == rm->m_final_op) {
323 /* If anyone waited for this message to get flushed out, wake
325 rds_message_unmapped(rm);
331 oldest = (oldest + 1) % ic->i_send_ring.w_nr;
334 rds_ib_ring_free(&ic->i_send_ring, completed);
335 rds_ib_sub_signaled(ic, nr_sig);
338 if (test_and_clear_bit(RDS_LL_SEND_FULL, &conn->c_flags) ||
339 test_bit(0, &conn->c_map_queued))
340 queue_delayed_work(rds_wq, &conn->c_send_w, 0);
342 /* We expect errors as the qp is drained during shutdown */
343 if (wc.status != IB_WC_SUCCESS && rds_conn_up(conn)) {
344 rds_ib_conn_error(conn, "send completion on %pI4 had status "
345 "%u (%s), disconnecting and reconnecting\n",
346 &conn->c_faddr, wc.status,
347 rds_ib_wc_status_str(wc.status));
353 * This is the main function for allocating credits when sending
356 * Conceptually, we have two counters:
357 * - send credits: this tells us how many WRs we're allowed
358 * to submit without overruning the receiver's queue. For
359 * each SEND WR we post, we decrement this by one.
361 * - posted credits: this tells us how many WRs we recently
362 * posted to the receive queue. This value is transferred
363 * to the peer as a "credit update" in a RDS header field.
364 * Every time we transmit credits to the peer, we subtract
365 * the amount of transferred credits from this counter.
367 * It is essential that we avoid situations where both sides have
368 * exhausted their send credits, and are unable to send new credits
369 * to the peer. We achieve this by requiring that we send at least
370 * one credit update to the peer before exhausting our credits.
371 * When new credits arrive, we subtract one credit that is withheld
372 * until we've posted new buffers and are ready to transmit these
373 * credits (see rds_ib_send_add_credits below).
375 * The RDS send code is essentially single-threaded; rds_send_xmit
376 * sets RDS_IN_XMIT to ensure exclusive access to the send ring.
377 * However, the ACK sending code is independent and can race with
380 * In the send path, we need to update the counters for send credits
381 * and the counter of posted buffers atomically - when we use the
382 * last available credit, we cannot allow another thread to race us
383 * and grab the posted credits counter. Hence, we have to use a
384 * spinlock to protect the credit counter, or use atomics.
386 * Spinlocks shared between the send and the receive path are bad,
387 * because they create unnecessary delays. An early implementation
388 * using a spinlock showed a 5% degradation in throughput at some
391 * This implementation avoids spinlocks completely, putting both
392 * counters into a single atomic, and updating that atomic using
393 * atomic_add (in the receive path, when receiving fresh credits),
394 * and using atomic_cmpxchg when updating the two counters.
396 int rds_ib_send_grab_credits(struct rds_ib_connection *ic,
397 u32 wanted, u32 *adv_credits, int need_posted, int max_posted)
399 unsigned int avail, posted, got = 0, advertise;
408 oldval = newval = atomic_read(&ic->i_credits);
409 posted = IB_GET_POST_CREDITS(oldval);
410 avail = IB_GET_SEND_CREDITS(oldval);
412 rdsdebug("rds_ib_send_grab_credits(%u): credits=%u posted=%u\n",
413 wanted, avail, posted);
415 /* The last credit must be used to send a credit update. */
416 if (avail && !posted)
419 if (avail < wanted) {
420 struct rds_connection *conn = ic->i_cm_id->context;
422 /* Oops, there aren't that many credits left! */
423 set_bit(RDS_LL_SEND_FULL, &conn->c_flags);
426 /* Sometimes you get what you want, lalala. */
429 newval -= IB_SET_SEND_CREDITS(got);
432 * If need_posted is non-zero, then the caller wants
433 * the posted regardless of whether any send credits are
436 if (posted && (got || need_posted)) {
437 advertise = min_t(unsigned int, posted, max_posted);
438 newval -= IB_SET_POST_CREDITS(advertise);
441 /* Finally bill everything */
442 if (atomic_cmpxchg(&ic->i_credits, oldval, newval) != oldval)
445 *adv_credits = advertise;
449 void rds_ib_send_add_credits(struct rds_connection *conn, unsigned int credits)
451 struct rds_ib_connection *ic = conn->c_transport_data;
456 rdsdebug("rds_ib_send_add_credits(%u): current=%u%s\n",
458 IB_GET_SEND_CREDITS(atomic_read(&ic->i_credits)),
459 test_bit(RDS_LL_SEND_FULL, &conn->c_flags) ? ", ll_send_full" : "");
461 atomic_add(IB_SET_SEND_CREDITS(credits), &ic->i_credits);
462 if (test_and_clear_bit(RDS_LL_SEND_FULL, &conn->c_flags))
463 queue_delayed_work(rds_wq, &conn->c_send_w, 0);
465 WARN_ON(IB_GET_SEND_CREDITS(credits) >= 16384);
467 rds_ib_stats_inc(s_ib_rx_credit_updates);
470 void rds_ib_advertise_credits(struct rds_connection *conn, unsigned int posted)
472 struct rds_ib_connection *ic = conn->c_transport_data;
477 atomic_add(IB_SET_POST_CREDITS(posted), &ic->i_credits);
479 /* Decide whether to send an update to the peer now.
480 * If we would send a credit update for every single buffer we
481 * post, we would end up with an ACK storm (ACK arrives,
482 * consumes buffer, we refill the ring, send ACK to remote
483 * advertising the newly posted buffer... ad inf)
485 * Performance pretty much depends on how often we send
486 * credit updates - too frequent updates mean lots of ACKs.
487 * Too infrequent updates, and the peer will run out of
488 * credits and has to throttle.
489 * For the time being, 16 seems to be a good compromise.
491 if (IB_GET_POST_CREDITS(atomic_read(&ic->i_credits)) >= 16)
492 set_bit(IB_ACK_REQUESTED, &ic->i_ack_flags);
495 static inline int rds_ib_set_wr_signal_state(struct rds_ib_connection *ic,
496 struct rds_ib_send_work *send,
500 * We want to delay signaling completions just enough to get
501 * the batching benefits but not so much that we create dead time
504 if (ic->i_unsignaled_wrs-- == 0 || notify) {
505 ic->i_unsignaled_wrs = rds_ib_sysctl_max_unsig_wrs;
506 send->s_wr.send_flags |= IB_SEND_SIGNALED;
513 * This can be called multiple times for a given message. The first time
514 * we see a message we map its scatterlist into the IB device so that
515 * we can provide that mapped address to the IB scatter gather entries
516 * in the IB work requests. We translate the scatterlist into a series
517 * of work requests that fragment the message. These work requests complete
518 * in order so we pass ownership of the message to the completion handler
519 * once we send the final fragment.
521 * The RDS core uses the c_send_lock to only enter this function once
522 * per connection. This makes sure that the tx ring alloc/unalloc pairs
523 * don't get out of sync and confuse the ring.
525 int rds_ib_xmit(struct rds_connection *conn, struct rds_message *rm,
526 unsigned int hdr_off, unsigned int sg, unsigned int off)
528 struct rds_ib_connection *ic = conn->c_transport_data;
529 struct ib_device *dev = ic->i_cm_id->device;
530 struct rds_ib_send_work *send = NULL;
531 struct rds_ib_send_work *first;
532 struct rds_ib_send_work *prev;
533 struct ib_send_wr *failed_wr;
534 struct scatterlist *scat;
538 u32 credit_alloc = 0;
544 int flow_controlled = 0;
547 BUG_ON(off % RDS_FRAG_SIZE);
548 BUG_ON(hdr_off != 0 && hdr_off != sizeof(struct rds_header));
550 /* Do not send cong updates to IB loopback */
552 && rm->m_inc.i_hdr.h_flags & RDS_FLAG_CONG_BITMAP) {
553 rds_cong_map_updated(conn->c_fcong, ~(u64) 0);
554 scat = &rm->data.op_sg[sg];
555 ret = sizeof(struct rds_header) + RDS_CONG_MAP_BYTES;
556 ret = min_t(int, ret, scat->length - conn->c_xmit_data_off);
560 /* FIXME we may overallocate here */
561 if (be32_to_cpu(rm->m_inc.i_hdr.h_len) == 0)
564 i = ceil(be32_to_cpu(rm->m_inc.i_hdr.h_len), RDS_FRAG_SIZE);
566 work_alloc = rds_ib_ring_alloc(&ic->i_send_ring, i, &pos);
567 if (work_alloc == 0) {
568 set_bit(RDS_LL_SEND_FULL, &conn->c_flags);
569 rds_ib_stats_inc(s_ib_tx_ring_full);
575 credit_alloc = rds_ib_send_grab_credits(ic, work_alloc, &posted, 0, RDS_MAX_ADV_CREDIT);
576 adv_credits += posted;
577 if (credit_alloc < work_alloc) {
578 rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc - credit_alloc);
579 work_alloc = credit_alloc;
582 if (work_alloc == 0) {
583 set_bit(RDS_LL_SEND_FULL, &conn->c_flags);
584 rds_ib_stats_inc(s_ib_tx_throttle);
590 /* map the message the first time we see it */
591 if (!ic->i_data_op) {
592 if (rm->data.op_nents) {
593 rm->data.op_count = ib_dma_map_sg(dev,
597 rdsdebug("ic %p mapping rm %p: %d\n", ic, rm, rm->data.op_count);
598 if (rm->data.op_count == 0) {
599 rds_ib_stats_inc(s_ib_tx_sg_mapping_failure);
600 rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc);
601 ret = -ENOMEM; /* XXX ? */
605 rm->data.op_count = 0;
608 rds_message_addref(rm);
609 ic->i_data_op = &rm->data;
611 /* Finalize the header */
612 if (test_bit(RDS_MSG_ACK_REQUIRED, &rm->m_flags))
613 rm->m_inc.i_hdr.h_flags |= RDS_FLAG_ACK_REQUIRED;
614 if (test_bit(RDS_MSG_RETRANSMITTED, &rm->m_flags))
615 rm->m_inc.i_hdr.h_flags |= RDS_FLAG_RETRANSMITTED;
617 /* If it has a RDMA op, tell the peer we did it. This is
618 * used by the peer to release use-once RDMA MRs. */
619 if (rm->rdma.op_active) {
620 struct rds_ext_header_rdma ext_hdr;
622 ext_hdr.h_rdma_rkey = cpu_to_be32(rm->rdma.op_rkey);
623 rds_message_add_extension(&rm->m_inc.i_hdr,
624 RDS_EXTHDR_RDMA, &ext_hdr, sizeof(ext_hdr));
626 if (rm->m_rdma_cookie) {
627 rds_message_add_rdma_dest_extension(&rm->m_inc.i_hdr,
628 rds_rdma_cookie_key(rm->m_rdma_cookie),
629 rds_rdma_cookie_offset(rm->m_rdma_cookie));
632 /* Note - rds_ib_piggyb_ack clears the ACK_REQUIRED bit, so
633 * we should not do this unless we have a chance of at least
634 * sticking the header into the send ring. Which is why we
635 * should call rds_ib_ring_alloc first. */
636 rm->m_inc.i_hdr.h_ack = cpu_to_be64(rds_ib_piggyb_ack(ic));
637 rds_message_make_checksum(&rm->m_inc.i_hdr);
640 * Update adv_credits since we reset the ACK_REQUIRED bit.
643 rds_ib_send_grab_credits(ic, 0, &posted, 1, RDS_MAX_ADV_CREDIT - adv_credits);
644 adv_credits += posted;
645 BUG_ON(adv_credits > 255);
649 /* Sometimes you want to put a fence between an RDMA
650 * READ and the following SEND.
651 * We could either do this all the time
652 * or when requested by the user. Right now, we let
653 * the application choose.
655 if (rm->rdma.op_active && rm->rdma.op_fence)
656 send_flags = IB_SEND_FENCE;
658 /* Each frag gets a header. Msgs may be 0 bytes */
659 send = &ic->i_sends[pos];
662 scat = &ic->i_data_op->op_sg[sg];
665 unsigned int len = 0;
667 /* Set up the header */
668 send->s_wr.send_flags = send_flags;
669 send->s_wr.opcode = IB_WR_SEND;
670 send->s_wr.num_sge = 1;
671 send->s_wr.next = NULL;
672 send->s_queued = jiffies;
675 send->s_sge[0].addr = ic->i_send_hdrs_dma
676 + (pos * sizeof(struct rds_header));
677 send->s_sge[0].length = sizeof(struct rds_header);
679 memcpy(&ic->i_send_hdrs[pos], &rm->m_inc.i_hdr, sizeof(struct rds_header));
681 /* Set up the data, if present */
683 && scat != &rm->data.op_sg[rm->data.op_count]) {
684 len = min(RDS_FRAG_SIZE, ib_sg_dma_len(dev, scat) - off);
685 send->s_wr.num_sge = 2;
687 send->s_sge[1].addr = ib_sg_dma_address(dev, scat) + off;
688 send->s_sge[1].length = len;
692 if (off == ib_sg_dma_len(dev, scat)) {
698 rds_ib_set_wr_signal_state(ic, send, 0);
701 * Always signal the last one if we're stopping due to flow control.
703 if (ic->i_flowctl && flow_controlled && i == (work_alloc-1))
704 send->s_wr.send_flags |= IB_SEND_SIGNALED | IB_SEND_SOLICITED;
706 if (send->s_wr.send_flags & IB_SEND_SIGNALED)
709 rdsdebug("send %p wr %p num_sge %u next %p\n", send,
710 &send->s_wr, send->s_wr.num_sge, send->s_wr.next);
712 if (ic->i_flowctl && adv_credits) {
713 struct rds_header *hdr = &ic->i_send_hdrs[pos];
715 /* add credit and redo the header checksum */
716 hdr->h_credit = adv_credits;
717 rds_message_make_checksum(hdr);
719 rds_ib_stats_inc(s_ib_tx_credit_updates);
723 prev->s_wr.next = &send->s_wr;
726 pos = (pos + 1) % ic->i_send_ring.w_nr;
727 send = &ic->i_sends[pos];
730 } while (i < work_alloc
731 && scat != &rm->data.op_sg[rm->data.op_count]);
733 /* Account the RDS header in the number of bytes we sent, but just once.
734 * The caller has no concept of fragmentation. */
736 bytes_sent += sizeof(struct rds_header);
738 /* if we finished the message then send completion owns it */
739 if (scat == &rm->data.op_sg[rm->data.op_count]) {
740 prev->s_op = ic->i_data_op;
741 prev->s_wr.send_flags |= IB_SEND_SOLICITED;
742 ic->i_data_op = NULL;
745 /* Put back wrs & credits we didn't use */
746 if (i < work_alloc) {
747 rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc - i);
750 if (ic->i_flowctl && i < credit_alloc)
751 rds_ib_send_add_credits(conn, credit_alloc - i);
754 atomic_add(nr_sig, &ic->i_signaled_sends);
756 /* XXX need to worry about failed_wr and partial sends. */
757 failed_wr = &first->s_wr;
758 ret = ib_post_send(ic->i_cm_id->qp, &first->s_wr, &failed_wr);
759 rdsdebug("ic %p first %p (wr %p) ret %d wr %p\n", ic,
760 first, &first->s_wr, ret, failed_wr);
761 BUG_ON(failed_wr != &first->s_wr);
763 printk(KERN_WARNING "RDS/IB: ib_post_send to %pI4 "
764 "returned %d\n", &conn->c_faddr, ret);
765 rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc);
766 rds_ib_sub_signaled(ic, nr_sig);
768 ic->i_data_op = prev->s_op;
772 rds_ib_conn_error(ic->conn, "ib_post_send failed\n");
783 * Issue atomic operation.
784 * A simplified version of the rdma case, we always map 1 SG, and
785 * only 8 bytes, for the return value from the atomic operation.
787 int rds_ib_xmit_atomic(struct rds_connection *conn, struct rm_atomic_op *op)
789 struct rds_ib_connection *ic = conn->c_transport_data;
790 struct rds_ib_send_work *send = NULL;
791 struct ib_send_wr *failed_wr;
792 struct rds_ib_device *rds_ibdev;
798 rds_ibdev = ib_get_client_data(ic->i_cm_id->device, &rds_ib_client);
800 work_alloc = rds_ib_ring_alloc(&ic->i_send_ring, 1, &pos);
801 if (work_alloc != 1) {
802 rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc);
803 rds_ib_stats_inc(s_ib_tx_ring_full);
808 /* address of send request in ring */
809 send = &ic->i_sends[pos];
810 send->s_queued = jiffies;
812 if (op->op_type == RDS_ATOMIC_TYPE_CSWP) {
813 send->s_wr.opcode = IB_WR_MASKED_ATOMIC_CMP_AND_SWP;
814 send->s_wr.wr.atomic.compare_add = op->op_m_cswp.compare;
815 send->s_wr.wr.atomic.swap = op->op_m_cswp.swap;
816 send->s_wr.wr.atomic.compare_add_mask = op->op_m_cswp.compare_mask;
817 send->s_wr.wr.atomic.swap_mask = op->op_m_cswp.swap_mask;
819 send->s_wr.opcode = IB_WR_MASKED_ATOMIC_FETCH_AND_ADD;
820 send->s_wr.wr.atomic.compare_add = op->op_m_fadd.add;
821 send->s_wr.wr.atomic.swap = 0;
822 send->s_wr.wr.atomic.compare_add_mask = op->op_m_fadd.nocarry_mask;
823 send->s_wr.wr.atomic.swap_mask = 0;
825 nr_sig = rds_ib_set_wr_signal_state(ic, send, op->op_notify);
826 send->s_wr.num_sge = 1;
827 send->s_wr.next = NULL;
828 send->s_wr.wr.atomic.remote_addr = op->op_remote_addr;
829 send->s_wr.wr.atomic.rkey = op->op_rkey;
831 rds_message_addref(container_of(send->s_op, struct rds_message, atomic));
833 /* map 8 byte retval buffer to the device */
834 ret = ib_dma_map_sg(ic->i_cm_id->device, op->op_sg, 1, DMA_FROM_DEVICE);
835 rdsdebug("ic %p mapping atomic op %p. mapped %d pg\n", ic, op, ret);
837 rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc);
838 rds_ib_stats_inc(s_ib_tx_sg_mapping_failure);
839 ret = -ENOMEM; /* XXX ? */
843 /* Convert our struct scatterlist to struct ib_sge */
844 send->s_sge[0].addr = ib_sg_dma_address(ic->i_cm_id->device, op->op_sg);
845 send->s_sge[0].length = ib_sg_dma_len(ic->i_cm_id->device, op->op_sg);
846 send->s_sge[0].lkey = ic->i_mr->lkey;
848 rdsdebug("rva %Lx rpa %Lx len %u\n", op->op_remote_addr,
849 send->s_sge[0].addr, send->s_sge[0].length);
852 atomic_add(nr_sig, &ic->i_signaled_sends);
854 failed_wr = &send->s_wr;
855 ret = ib_post_send(ic->i_cm_id->qp, &send->s_wr, &failed_wr);
856 rdsdebug("ic %p send %p (wr %p) ret %d wr %p\n", ic,
857 send, &send->s_wr, ret, failed_wr);
858 BUG_ON(failed_wr != &send->s_wr);
860 printk(KERN_WARNING "RDS/IB: atomic ib_post_send to %pI4 "
861 "returned %d\n", &conn->c_faddr, ret);
862 rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc);
863 rds_ib_sub_signaled(ic, nr_sig);
867 if (unlikely(failed_wr != &send->s_wr)) {
868 printk(KERN_WARNING "RDS/IB: atomic ib_post_send() rc=%d, but failed_wqe updated!\n", ret);
869 BUG_ON(failed_wr != &send->s_wr);
876 int rds_ib_xmit_rdma(struct rds_connection *conn, struct rm_rdma_op *op)
878 struct rds_ib_connection *ic = conn->c_transport_data;
879 struct rds_ib_send_work *send = NULL;
880 struct rds_ib_send_work *first;
881 struct rds_ib_send_work *prev;
882 struct ib_send_wr *failed_wr;
883 struct scatterlist *scat;
885 u64 remote_addr = op->op_remote_addr;
886 u32 max_sge = ic->rds_ibdev->max_sge;
896 /* map the op the first time we see it */
897 if (!op->op_mapped) {
898 op->op_count = ib_dma_map_sg(ic->i_cm_id->device,
899 op->op_sg, op->op_nents, (op->op_write) ?
900 DMA_TO_DEVICE : DMA_FROM_DEVICE);
901 rdsdebug("ic %p mapping op %p: %d\n", ic, op, op->op_count);
902 if (op->op_count == 0) {
903 rds_ib_stats_inc(s_ib_tx_sg_mapping_failure);
904 ret = -ENOMEM; /* XXX ? */
912 * Instead of knowing how to return a partial rdma read/write we insist that there
913 * be enough work requests to send the entire message.
915 i = ceil(op->op_count, max_sge);
917 work_alloc = rds_ib_ring_alloc(&ic->i_send_ring, i, &pos);
918 if (work_alloc != i) {
919 rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc);
920 rds_ib_stats_inc(s_ib_tx_ring_full);
925 send = &ic->i_sends[pos];
928 scat = &op->op_sg[0];
930 num_sge = op->op_count;
932 for (i = 0; i < work_alloc && scat != &op->op_sg[op->op_count]; i++) {
933 send->s_wr.send_flags = 0;
934 send->s_queued = jiffies;
937 nr_sig += rds_ib_set_wr_signal_state(ic, send, op->op_notify);
939 send->s_wr.opcode = op->op_write ? IB_WR_RDMA_WRITE : IB_WR_RDMA_READ;
940 send->s_wr.wr.rdma.remote_addr = remote_addr;
941 send->s_wr.wr.rdma.rkey = op->op_rkey;
943 if (num_sge > max_sge) {
944 send->s_wr.num_sge = max_sge;
947 send->s_wr.num_sge = num_sge;
950 send->s_wr.next = NULL;
953 prev->s_wr.next = &send->s_wr;
955 for (j = 0; j < send->s_wr.num_sge && scat != &op->op_sg[op->op_count]; j++) {
956 len = ib_sg_dma_len(ic->i_cm_id->device, scat);
957 send->s_sge[j].addr =
958 ib_sg_dma_address(ic->i_cm_id->device, scat);
959 send->s_sge[j].length = len;
960 send->s_sge[j].lkey = ic->i_mr->lkey;
963 rdsdebug("ic %p sent %d remote_addr %llu\n", ic, sent, remote_addr);
969 rdsdebug("send %p wr %p num_sge %u next %p\n", send,
970 &send->s_wr, send->s_wr.num_sge, send->s_wr.next);
973 if (++send == &ic->i_sends[ic->i_send_ring.w_nr])
977 /* give a reference to the last op */
978 if (scat == &op->op_sg[op->op_count]) {
980 rds_message_addref(container_of(op, struct rds_message, rdma));
983 if (i < work_alloc) {
984 rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc - i);
989 atomic_add(nr_sig, &ic->i_signaled_sends);
991 failed_wr = &first->s_wr;
992 ret = ib_post_send(ic->i_cm_id->qp, &first->s_wr, &failed_wr);
993 rdsdebug("ic %p first %p (wr %p) ret %d wr %p\n", ic,
994 first, &first->s_wr, ret, failed_wr);
995 BUG_ON(failed_wr != &first->s_wr);
997 printk(KERN_WARNING "RDS/IB: rdma ib_post_send to %pI4 "
998 "returned %d\n", &conn->c_faddr, ret);
999 rds_ib_ring_unalloc(&ic->i_send_ring, work_alloc);
1000 rds_ib_sub_signaled(ic, nr_sig);
1004 if (unlikely(failed_wr != &first->s_wr)) {
1005 printk(KERN_WARNING "RDS/IB: ib_post_send() rc=%d, but failed_wqe updated!\n", ret);
1006 BUG_ON(failed_wr != &first->s_wr);
1014 void rds_ib_xmit_complete(struct rds_connection *conn)
1016 struct rds_ib_connection *ic = conn->c_transport_data;
1018 /* We may have a pending ACK or window update we were unable
1019 * to send previously (due to flow control). Try again. */
1020 rds_ib_attempt_ack(ic);