Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.1.
  1/*
  2 * Copyright (c) 2015 Oracle.  All rights reserved.
  3 * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
  4 */
  5
  6/* Lightweight memory registration using Fast Registration Work
  7 * Requests (FRWR). Also referred to sometimes as FRMR mode.
  8 *
  9 * FRWR features ordered asynchronous registration and deregistration
 10 * of arbitrarily sized memory regions. This is the fastest and safest
 11 * but most complex memory registration mode.
 12 */
 13
 14/* Normal operation
 15 *
 16 * A Memory Region is prepared for RDMA READ or WRITE using a FAST_REG
 17 * Work Request (frmr_op_map). When the RDMA operation is finished, this
 18 * Memory Region is invalidated using a LOCAL_INV Work Request
 19 * (frmr_op_unmap).
 20 *
 21 * Typically these Work Requests are not signaled, and neither are RDMA
 22 * SEND Work Requests (with the exception of signaling occasionally to
 23 * prevent provider work queue overflows). This greatly reduces HCA
 24 * interrupt workload.
 25 *
 26 * As an optimization, frwr_op_unmap marks MRs INVALID before the
 27 * LOCAL_INV WR is posted. If posting succeeds, the MR is placed on
 28 * rb_mws immediately so that no work (like managing a linked list
 29 * under a spinlock) is needed in the completion upcall.
 30 *
 31 * But this means that frwr_op_map() can occasionally encounter an MR
 32 * that is INVALID but the LOCAL_INV WR has not completed. Work Queue
 33 * ordering prevents a subsequent FAST_REG WR from executing against
 34 * that MR while it is still being invalidated.
 35 */
 36
 37/* Transport recovery
 38 *
 39 * ->op_map and the transport connect worker cannot run at the same
 40 * time, but ->op_unmap can fire while the transport connect worker
 41 * is running. Thus MR recovery is handled in ->op_map, to guarantee
 42 * that recovered MRs are owned by a sending RPC, and not one where
 43 * ->op_unmap could fire at the same time transport reconnect is
 44 * being done.
 45 *
 46 * When the underlying transport disconnects, MRs are left in one of
 47 * three states:
 48 *
 49 * INVALID:	The MR was not in use before the QP entered ERROR state.
 50 *		(Or, the LOCAL_INV WR has not completed or flushed yet).
 51 *
 52 * STALE:	The MR was being registered or unregistered when the QP
 53 *		entered ERROR state, and the pending WR was flushed.
 54 *
 55 * VALID:	The MR was registered before the QP entered ERROR state.
 56 *
 57 * When frwr_op_map encounters STALE and VALID MRs, they are recovered
 58 * with ib_dereg_mr and then are re-initialized. Beause MR recovery
 59 * allocates fresh resources, it is deferred to a workqueue, and the
 60 * recovered MRs are placed back on the rb_mws list when recovery is
 61 * complete. frwr_op_map allocates another MR for the current RPC while
 62 * the broken MR is reset.
 63 *
 64 * To ensure that frwr_op_map doesn't encounter an MR that is marked
 65 * INVALID but that is about to be flushed due to a previous transport
 66 * disconnect, the transport connect worker attempts to drain all
 67 * pending send queue WRs before the transport is reconnected.
 68 */
 69
 70#include "xprt_rdma.h"
 71
 72#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 73# define RPCDBG_FACILITY	RPCDBG_TRANS
 74#endif
 75
 76static struct workqueue_struct *frwr_recovery_wq;
 77
 78#define FRWR_RECOVERY_WQ_FLAGS		(WQ_UNBOUND | WQ_MEM_RECLAIM)
 79
 80int
 81frwr_alloc_recovery_wq(void)
 82{
 83	frwr_recovery_wq = alloc_workqueue("frwr_recovery",
 84					   FRWR_RECOVERY_WQ_FLAGS, 0);
 85	return !frwr_recovery_wq ? -ENOMEM : 0;
 86}
 87
 88void
 89frwr_destroy_recovery_wq(void)
 90{
 91	struct workqueue_struct *wq;
 92
 93	if (!frwr_recovery_wq)
 94		return;
 95
 96	wq = frwr_recovery_wq;
 97	frwr_recovery_wq = NULL;
 98	destroy_workqueue(wq);
 99}
100
101/* Deferred reset of a single FRMR. Generate a fresh rkey by
102 * replacing the MR.
103 *
104 * There's no recovery if this fails. The FRMR is abandoned, but
105 * remains in rb_all. It will be cleaned up when the transport is
106 * destroyed.
107 */
108static void
109__frwr_recovery_worker(struct work_struct *work)
110{
111	struct rpcrdma_mw *r = container_of(work, struct rpcrdma_mw,
112					    frmr.fr_work);
113	struct rpcrdma_xprt *r_xprt = r->frmr.fr_xprt;
114	unsigned int depth = r_xprt->rx_ia.ri_max_frmr_depth;
115	struct ib_pd *pd = r_xprt->rx_ia.ri_pd;
116
117	if (ib_dereg_mr(r->frmr.fr_mr))
118		goto out_fail;
119
120	r->frmr.fr_mr = ib_alloc_mr(pd, IB_MR_TYPE_MEM_REG, depth);
121	if (IS_ERR(r->frmr.fr_mr))
122		goto out_fail;
123
124	dprintk("RPC:       %s: recovered FRMR %p\n", __func__, r);
125	r->frmr.fr_state = FRMR_IS_INVALID;
126	rpcrdma_put_mw(r_xprt, r);
127	return;
128
129out_fail:
130	pr_warn("RPC:       %s: FRMR %p unrecovered\n",
131		__func__, r);
132}
133
134/* A broken MR was discovered in a context that can't sleep.
135 * Defer recovery to the recovery worker.
136 */
137static void
138__frwr_queue_recovery(struct rpcrdma_mw *r)
139{
140	INIT_WORK(&r->frmr.fr_work, __frwr_recovery_worker);
141	queue_work(frwr_recovery_wq, &r->frmr.fr_work);
142}
143
144static int
145__frwr_init(struct rpcrdma_mw *r, struct ib_pd *pd, struct ib_device *device,
146	    unsigned int depth)
147{
148	struct rpcrdma_frmr *f = &r->frmr;
149	int rc;
150
151	f->fr_mr = ib_alloc_mr(pd, IB_MR_TYPE_MEM_REG, depth);
152	if (IS_ERR(f->fr_mr))
153		goto out_mr_err;
154
155	f->sg = kcalloc(depth, sizeof(*f->sg), GFP_KERNEL);
156	if (!f->sg)
157		goto out_list_err;
158
159	sg_init_table(f->sg, depth);
160
161	init_completion(&f->fr_linv_done);
162
163	return 0;
164
165out_mr_err:
166	rc = PTR_ERR(f->fr_mr);
167	dprintk("RPC:       %s: ib_alloc_mr status %i\n",
168		__func__, rc);
169	return rc;
170
171out_list_err:
172	rc = -ENOMEM;
173	dprintk("RPC:       %s: sg allocation failure\n",
174		__func__);
175	ib_dereg_mr(f->fr_mr);
176	return rc;
177}
178
179static void
180__frwr_release(struct rpcrdma_mw *r)
181{
182	int rc;
183
184	rc = ib_dereg_mr(r->frmr.fr_mr);
185	if (rc)
186		dprintk("RPC:       %s: ib_dereg_mr status %i\n",
187			__func__, rc);
188	kfree(r->frmr.sg);
189}
190
191static int
192frwr_op_open(struct rpcrdma_ia *ia, struct rpcrdma_ep *ep,
193	     struct rpcrdma_create_data_internal *cdata)
194{
195	int depth, delta;
196
197	ia->ri_max_frmr_depth =
198			min_t(unsigned int, RPCRDMA_MAX_DATA_SEGS,
199			      ia->ri_device->attrs.max_fast_reg_page_list_len);
200	dprintk("RPC:       %s: device's max FR page list len = %u\n",
201		__func__, ia->ri_max_frmr_depth);
202
203	/* Add room for frmr register and invalidate WRs.
204	 * 1. FRMR reg WR for head
205	 * 2. FRMR invalidate WR for head
206	 * 3. N FRMR reg WRs for pagelist
207	 * 4. N FRMR invalidate WRs for pagelist
208	 * 5. FRMR reg WR for tail
209	 * 6. FRMR invalidate WR for tail
210	 * 7. The RDMA_SEND WR
211	 */
212	depth = 7;
213
214	/* Calculate N if the device max FRMR depth is smaller than
215	 * RPCRDMA_MAX_DATA_SEGS.
216	 */
217	if (ia->ri_max_frmr_depth < RPCRDMA_MAX_DATA_SEGS) {
218		delta = RPCRDMA_MAX_DATA_SEGS - ia->ri_max_frmr_depth;
219		do {
220			depth += 2; /* FRMR reg + invalidate */
221			delta -= ia->ri_max_frmr_depth;
222		} while (delta > 0);
223	}
224
225	ep->rep_attr.cap.max_send_wr *= depth;
226	if (ep->rep_attr.cap.max_send_wr > ia->ri_device->attrs.max_qp_wr) {
227		cdata->max_requests = ia->ri_device->attrs.max_qp_wr / depth;
228		if (!cdata->max_requests)
229			return -EINVAL;
230		ep->rep_attr.cap.max_send_wr = cdata->max_requests *
231					       depth;
232	}
233
234	return 0;
235}
236
237/* FRWR mode conveys a list of pages per chunk segment. The
238 * maximum length of that list is the FRWR page list depth.
239 */
240static size_t
241frwr_op_maxpages(struct rpcrdma_xprt *r_xprt)
242{
243	struct rpcrdma_ia *ia = &r_xprt->rx_ia;
244
245	return min_t(unsigned int, RPCRDMA_MAX_DATA_SEGS,
246		     rpcrdma_max_segments(r_xprt) * ia->ri_max_frmr_depth);
247}
248
249static void
250__frwr_sendcompletion_flush(struct ib_wc *wc, struct rpcrdma_frmr *frmr,
251			    const char *wr)
252{
253	frmr->fr_state = FRMR_IS_STALE;
254	if (wc->status != IB_WC_WR_FLUSH_ERR)
255		pr_err("rpcrdma: %s: %s (%u/0x%x)\n",
256		       wr, ib_wc_status_msg(wc->status),
257		       wc->status, wc->vendor_err);
258}
259
260/**
261 * frwr_wc_fastreg - Invoked by RDMA provider for each polled FastReg WC
262 * @cq:	completion queue (ignored)
263 * @wc:	completed WR
264 *
265 */
266static void
267frwr_wc_fastreg(struct ib_cq *cq, struct ib_wc *wc)
268{
269	struct rpcrdma_frmr *frmr;
270	struct ib_cqe *cqe;
271
272	/* WARNING: Only wr_cqe and status are reliable at this point */
273	if (wc->status != IB_WC_SUCCESS) {
274		cqe = wc->wr_cqe;
275		frmr = container_of(cqe, struct rpcrdma_frmr, fr_cqe);
276		__frwr_sendcompletion_flush(wc, frmr, "fastreg");
277	}
278}
279
280/**
281 * frwr_wc_localinv - Invoked by RDMA provider for each polled LocalInv WC
282 * @cq:	completion queue (ignored)
283 * @wc:	completed WR
284 *
285 */
286static void
287frwr_wc_localinv(struct ib_cq *cq, struct ib_wc *wc)
288{
289	struct rpcrdma_frmr *frmr;
290	struct ib_cqe *cqe;
291
292	/* WARNING: Only wr_cqe and status are reliable at this point */
293	if (wc->status != IB_WC_SUCCESS) {
294		cqe = wc->wr_cqe;
295		frmr = container_of(cqe, struct rpcrdma_frmr, fr_cqe);
296		__frwr_sendcompletion_flush(wc, frmr, "localinv");
297	}
298}
299
300/**
301 * frwr_wc_localinv - Invoked by RDMA provider for each polled LocalInv WC
302 * @cq:	completion queue (ignored)
303 * @wc:	completed WR
304 *
305 * Awaken anyone waiting for an MR to finish being fenced.
306 */
307static void
308frwr_wc_localinv_wake(struct ib_cq *cq, struct ib_wc *wc)
309{
310	struct rpcrdma_frmr *frmr;
311	struct ib_cqe *cqe;
312
313	/* WARNING: Only wr_cqe and status are reliable at this point */
314	cqe = wc->wr_cqe;
315	frmr = container_of(cqe, struct rpcrdma_frmr, fr_cqe);
316	if (wc->status != IB_WC_SUCCESS)
317		__frwr_sendcompletion_flush(wc, frmr, "localinv");
318	complete_all(&frmr->fr_linv_done);
319}
320
321static int
322frwr_op_init(struct rpcrdma_xprt *r_xprt)
323{
324	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
325	struct ib_device *device = r_xprt->rx_ia.ri_device;
326	unsigned int depth = r_xprt->rx_ia.ri_max_frmr_depth;
327	struct ib_pd *pd = r_xprt->rx_ia.ri_pd;
328	int i;
329
330	spin_lock_init(&buf->rb_mwlock);
331	INIT_LIST_HEAD(&buf->rb_mws);
332	INIT_LIST_HEAD(&buf->rb_all);
333
334	i = max_t(int, RPCRDMA_MAX_DATA_SEGS / depth, 1);
335	i += 2;				/* head + tail */
336	i *= buf->rb_max_requests;	/* one set for each RPC slot */
337	dprintk("RPC:       %s: initalizing %d FRMRs\n", __func__, i);
338
339	while (i--) {
340		struct rpcrdma_mw *r;
341		int rc;
342
343		r = kzalloc(sizeof(*r), GFP_KERNEL);
344		if (!r)
345			return -ENOMEM;
346
347		rc = __frwr_init(r, pd, device, depth);
348		if (rc) {
349			kfree(r);
350			return rc;
351		}
352
353		list_add(&r->mw_list, &buf->rb_mws);
354		list_add(&r->mw_all, &buf->rb_all);
355		r->frmr.fr_xprt = r_xprt;
356	}
357
358	return 0;
359}
360
361/* Post a FAST_REG Work Request to register a memory region
362 * for remote access via RDMA READ or RDMA WRITE.
363 */
364static int
365frwr_op_map(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg,
366	    int nsegs, bool writing)
367{
368	struct rpcrdma_ia *ia = &r_xprt->rx_ia;
369	struct ib_device *device = ia->ri_device;
370	enum dma_data_direction direction = rpcrdma_data_dir(writing);
371	struct rpcrdma_mr_seg *seg1 = seg;
372	struct rpcrdma_mw *mw;
373	struct rpcrdma_frmr *frmr;
374	struct ib_mr *mr;
375	struct ib_reg_wr *reg_wr;
376	struct ib_send_wr *bad_wr;
377	int rc, i, n, dma_nents;
378	u8 key;
379
380	mw = seg1->rl_mw;
381	seg1->rl_mw = NULL;
382	do {
383		if (mw)
384			__frwr_queue_recovery(mw);
385		mw = rpcrdma_get_mw(r_xprt);
386		if (!mw)
387			return -ENOMEM;
388	} while (mw->frmr.fr_state != FRMR_IS_INVALID);
389	frmr = &mw->frmr;
390	frmr->fr_state = FRMR_IS_VALID;
391	mr = frmr->fr_mr;
392	reg_wr = &frmr->fr_regwr;
393
394	if (nsegs > ia->ri_max_frmr_depth)
395		nsegs = ia->ri_max_frmr_depth;
396
397	for (i = 0; i < nsegs;) {
398		if (seg->mr_page)
399			sg_set_page(&frmr->sg[i],
400				    seg->mr_page,
401				    seg->mr_len,
402				    offset_in_page(seg->mr_offset));
403		else
404			sg_set_buf(&frmr->sg[i], seg->mr_offset,
405				   seg->mr_len);
406
407		++seg;
408		++i;
409
410		/* Check for holes */
411		if ((i < nsegs && offset_in_page(seg->mr_offset)) ||
412		    offset_in_page((seg-1)->mr_offset + (seg-1)->mr_len))
413			break;
414	}
415	frmr->sg_nents = i;
416
417	dma_nents = ib_dma_map_sg(device, frmr->sg, frmr->sg_nents, direction);
418	if (!dma_nents) {
419		pr_err("RPC:       %s: failed to dma map sg %p sg_nents %u\n",
420		       __func__, frmr->sg, frmr->sg_nents);
421		return -ENOMEM;
422	}
423
424	n = ib_map_mr_sg(mr, frmr->sg, frmr->sg_nents, PAGE_SIZE);
425	if (unlikely(n != frmr->sg_nents)) {
426		pr_err("RPC:       %s: failed to map mr %p (%u/%u)\n",
427		       __func__, frmr->fr_mr, n, frmr->sg_nents);
428		rc = n < 0 ? n : -EINVAL;
429		goto out_senderr;
430	}
431
432	dprintk("RPC:       %s: Using frmr %p to map %u segments (%u bytes)\n",
433		__func__, mw, frmr->sg_nents, mr->length);
434
435	key = (u8)(mr->rkey & 0x000000FF);
436	ib_update_fast_reg_key(mr, ++key);
437
438	reg_wr->wr.next = NULL;
439	reg_wr->wr.opcode = IB_WR_REG_MR;
440	frmr->fr_cqe.done = frwr_wc_fastreg;
441	reg_wr->wr.wr_cqe = &frmr->fr_cqe;
442	reg_wr->wr.num_sge = 0;
443	reg_wr->wr.send_flags = 0;
444	reg_wr->mr = mr;
445	reg_wr->key = mr->rkey;
446	reg_wr->access = writing ?
447			 IB_ACCESS_REMOTE_WRITE | IB_ACCESS_LOCAL_WRITE :
448			 IB_ACCESS_REMOTE_READ;
449
450	DECR_CQCOUNT(&r_xprt->rx_ep);
451	rc = ib_post_send(ia->ri_id->qp, &reg_wr->wr, &bad_wr);
452	if (rc)
453		goto out_senderr;
454
455	seg1->mr_dir = direction;
456	seg1->rl_mw = mw;
457	seg1->mr_rkey = mr->rkey;
458	seg1->mr_base = mr->iova;
459	seg1->mr_nsegs = frmr->sg_nents;
460	seg1->mr_len = mr->length;
461
462	return frmr->sg_nents;
463
464out_senderr:
465	dprintk("RPC:       %s: ib_post_send status %i\n", __func__, rc);
466	ib_dma_unmap_sg(device, frmr->sg, dma_nents, direction);
467	__frwr_queue_recovery(mw);
468	return rc;
469}
470
471static struct ib_send_wr *
472__frwr_prepare_linv_wr(struct rpcrdma_mr_seg *seg)
473{
474	struct rpcrdma_mw *mw = seg->rl_mw;
475	struct rpcrdma_frmr *f = &mw->frmr;
476	struct ib_send_wr *invalidate_wr;
477
478	f->fr_state = FRMR_IS_INVALID;
479	invalidate_wr = &f->fr_invwr;
480
481	memset(invalidate_wr, 0, sizeof(*invalidate_wr));
482	f->fr_cqe.done = frwr_wc_localinv;
483	invalidate_wr->wr_cqe = &f->fr_cqe;
484	invalidate_wr->opcode = IB_WR_LOCAL_INV;
485	invalidate_wr->ex.invalidate_rkey = f->fr_mr->rkey;
486
487	return invalidate_wr;
488}
489
490static void
491__frwr_dma_unmap(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg,
492		 int rc)
493{
494	struct ib_device *device = r_xprt->rx_ia.ri_device;
495	struct rpcrdma_mw *mw = seg->rl_mw;
496	struct rpcrdma_frmr *f = &mw->frmr;
497
498	seg->rl_mw = NULL;
499
500	ib_dma_unmap_sg(device, f->sg, f->sg_nents, seg->mr_dir);
501
502	if (!rc)
503		rpcrdma_put_mw(r_xprt, mw);
504	else
505		__frwr_queue_recovery(mw);
506}
507
508/* Invalidate all memory regions that were registered for "req".
509 *
510 * Sleeps until it is safe for the host CPU to access the
511 * previously mapped memory regions.
512 */
513static void
514frwr_op_unmap_sync(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req)
515{
516	struct ib_send_wr *invalidate_wrs, *pos, *prev, *bad_wr;
517	struct rpcrdma_ia *ia = &r_xprt->rx_ia;
518	struct rpcrdma_mr_seg *seg;
519	unsigned int i, nchunks;
520	struct rpcrdma_frmr *f;
521	int rc;
522
523	dprintk("RPC:       %s: req %p\n", __func__, req);
524
525	/* ORDER: Invalidate all of the req's MRs first
526	 *
527	 * Chain the LOCAL_INV Work Requests and post them with
528	 * a single ib_post_send() call.
529	 */
530	invalidate_wrs = pos = prev = NULL;
531	seg = NULL;
532	for (i = 0, nchunks = req->rl_nchunks; nchunks; nchunks--) {
533		seg = &req->rl_segments[i];
534
535		pos = __frwr_prepare_linv_wr(seg);
536
537		if (!invalidate_wrs)
538			invalidate_wrs = pos;
539		else
540			prev->next = pos;
541		prev = pos;
542
543		i += seg->mr_nsegs;
544	}
545	f = &seg->rl_mw->frmr;
546
547	/* Strong send queue ordering guarantees that when the
548	 * last WR in the chain completes, all WRs in the chain
549	 * are complete.
550	 */
551	f->fr_invwr.send_flags = IB_SEND_SIGNALED;
552	f->fr_cqe.done = frwr_wc_localinv_wake;
553	reinit_completion(&f->fr_linv_done);
554	INIT_CQCOUNT(&r_xprt->rx_ep);
555
556	/* Transport disconnect drains the receive CQ before it
557	 * replaces the QP. The RPC reply handler won't call us
558	 * unless ri_id->qp is a valid pointer.
559	 */
560	rc = ib_post_send(ia->ri_id->qp, invalidate_wrs, &bad_wr);
561	if (rc) {
562		pr_warn("%s: ib_post_send failed %i\n", __func__, rc);
563		rdma_disconnect(ia->ri_id);
564		goto unmap;
565	}
566
567	wait_for_completion(&f->fr_linv_done);
568
569	/* ORDER: Now DMA unmap all of the req's MRs, and return
570	 * them to the free MW list.
571	 */
572unmap:
573	for (i = 0, nchunks = req->rl_nchunks; nchunks; nchunks--) {
574		seg = &req->rl_segments[i];
575
576		__frwr_dma_unmap(r_xprt, seg, rc);
577
578		i += seg->mr_nsegs;
579		seg->mr_nsegs = 0;
580	}
581
582	req->rl_nchunks = 0;
583}
584
585/* Post a LOCAL_INV Work Request to prevent further remote access
586 * via RDMA READ or RDMA WRITE.
587 */
588static int
589frwr_op_unmap(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg)
590{
591	struct rpcrdma_mr_seg *seg1 = seg;
592	struct rpcrdma_ia *ia = &r_xprt->rx_ia;
593	struct rpcrdma_mw *mw = seg1->rl_mw;
594	struct rpcrdma_frmr *frmr = &mw->frmr;
595	struct ib_send_wr *invalidate_wr, *bad_wr;
596	int rc, nsegs = seg->mr_nsegs;
597
598	dprintk("RPC:       %s: FRMR %p\n", __func__, mw);
599
600	seg1->rl_mw = NULL;
601	frmr->fr_state = FRMR_IS_INVALID;
602	invalidate_wr = &mw->frmr.fr_invwr;
603
604	memset(invalidate_wr, 0, sizeof(*invalidate_wr));
605	frmr->fr_cqe.done = frwr_wc_localinv;
606	invalidate_wr->wr_cqe = &frmr->fr_cqe;
607	invalidate_wr->opcode = IB_WR_LOCAL_INV;
608	invalidate_wr->ex.invalidate_rkey = frmr->fr_mr->rkey;
609	DECR_CQCOUNT(&r_xprt->rx_ep);
610
611	ib_dma_unmap_sg(ia->ri_device, frmr->sg, frmr->sg_nents, seg1->mr_dir);
612	read_lock(&ia->ri_qplock);
613	rc = ib_post_send(ia->ri_id->qp, invalidate_wr, &bad_wr);
614	read_unlock(&ia->ri_qplock);
615	if (rc)
616		goto out_err;
617
618	rpcrdma_put_mw(r_xprt, mw);
619	return nsegs;
620
621out_err:
622	dprintk("RPC:       %s: ib_post_send status %i\n", __func__, rc);
623	__frwr_queue_recovery(mw);
624	return nsegs;
625}
626
627static void
628frwr_op_destroy(struct rpcrdma_buffer *buf)
629{
630	struct rpcrdma_mw *r;
631
632	/* Ensure stale MWs for "buf" are no longer in flight */
633	flush_workqueue(frwr_recovery_wq);
634
635	while (!list_empty(&buf->rb_all)) {
636		r = list_entry(buf->rb_all.next, struct rpcrdma_mw, mw_all);
637		list_del(&r->mw_all);
638		__frwr_release(r);
639		kfree(r);
640	}
641}
642
643const struct rpcrdma_memreg_ops rpcrdma_frwr_memreg_ops = {
644	.ro_map				= frwr_op_map,
645	.ro_unmap_sync			= frwr_op_unmap_sync,
646	.ro_unmap			= frwr_op_unmap,
647	.ro_open			= frwr_op_open,
648	.ro_maxpages			= frwr_op_maxpages,
649	.ro_init			= frwr_op_init,
650	.ro_destroy			= frwr_op_destroy,
651	.ro_displayname			= "frwr",
652};