cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

verbs.c (36604B)


      1// SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
      2/*
      3 * Copyright (c) 2014-2017 Oracle.  All rights reserved.
      4 * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
      5 *
      6 * This software is available to you under a choice of one of two
      7 * licenses.  You may choose to be licensed under the terms of the GNU
      8 * General Public License (GPL) Version 2, available from the file
      9 * COPYING in the main directory of this source tree, or the BSD-type
     10 * license below:
     11 *
     12 * Redistribution and use in source and binary forms, with or without
     13 * modification, are permitted provided that the following conditions
     14 * are met:
     15 *
     16 *      Redistributions of source code must retain the above copyright
     17 *      notice, this list of conditions and the following disclaimer.
     18 *
     19 *      Redistributions in binary form must reproduce the above
     20 *      copyright notice, this list of conditions and the following
     21 *      disclaimer in the documentation and/or other materials provided
     22 *      with the distribution.
     23 *
     24 *      Neither the name of the Network Appliance, Inc. nor the names of
     25 *      its contributors may be used to endorse or promote products
     26 *      derived from this software without specific prior written
     27 *      permission.
     28 *
     29 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
     30 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
     31 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
     32 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
     33 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
     34 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
     35 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     36 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     37 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     38 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
     39 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     40 */
     41
     42/*
     43 * verbs.c
     44 *
     45 * Encapsulates the major functions managing:
     46 *  o adapters
     47 *  o endpoints
     48 *  o connections
     49 *  o buffer memory
     50 */
     51
     52#include <linux/interrupt.h>
     53#include <linux/slab.h>
     54#include <linux/sunrpc/addr.h>
     55#include <linux/sunrpc/svc_rdma.h>
     56#include <linux/log2.h>
     57
     58#include <asm-generic/barrier.h>
     59#include <asm/bitops.h>
     60
     61#include <rdma/ib_cm.h>
     62
     63#include "xprt_rdma.h"
     64#include <trace/events/rpcrdma.h>
     65
     66static int rpcrdma_sendctxs_create(struct rpcrdma_xprt *r_xprt);
     67static void rpcrdma_sendctxs_destroy(struct rpcrdma_xprt *r_xprt);
     68static void rpcrdma_sendctx_put_locked(struct rpcrdma_xprt *r_xprt,
     69				       struct rpcrdma_sendctx *sc);
     70static int rpcrdma_reqs_setup(struct rpcrdma_xprt *r_xprt);
     71static void rpcrdma_reqs_reset(struct rpcrdma_xprt *r_xprt);
     72static void rpcrdma_rep_destroy(struct rpcrdma_rep *rep);
     73static void rpcrdma_reps_unmap(struct rpcrdma_xprt *r_xprt);
     74static void rpcrdma_mrs_create(struct rpcrdma_xprt *r_xprt);
     75static void rpcrdma_mrs_destroy(struct rpcrdma_xprt *r_xprt);
     76static void rpcrdma_ep_get(struct rpcrdma_ep *ep);
     77static int rpcrdma_ep_put(struct rpcrdma_ep *ep);
     78static struct rpcrdma_regbuf *
     79rpcrdma_regbuf_alloc(size_t size, enum dma_data_direction direction,
     80		     gfp_t flags);
     81static void rpcrdma_regbuf_dma_unmap(struct rpcrdma_regbuf *rb);
     82static void rpcrdma_regbuf_free(struct rpcrdma_regbuf *rb);
     83
     84/* Wait for outstanding transport work to finish. ib_drain_qp
     85 * handles the drains in the wrong order for us, so open code
     86 * them here.
     87 */
     88static void rpcrdma_xprt_drain(struct rpcrdma_xprt *r_xprt)
     89{
     90	struct rpcrdma_ep *ep = r_xprt->rx_ep;
     91	struct rdma_cm_id *id = ep->re_id;
     92
     93	/* Wait for rpcrdma_post_recvs() to leave its critical
     94	 * section.
     95	 */
     96	if (atomic_inc_return(&ep->re_receiving) > 1)
     97		wait_for_completion(&ep->re_done);
     98
     99	/* Flush Receives, then wait for deferred Reply work
    100	 * to complete.
    101	 */
    102	ib_drain_rq(id->qp);
    103
    104	/* Deferred Reply processing might have scheduled
    105	 * local invalidations.
    106	 */
    107	ib_drain_sq(id->qp);
    108
    109	rpcrdma_ep_put(ep);
    110}
    111
    112/* Ensure xprt_force_disconnect() is invoked exactly once when a
    113 * connection is closed or lost. (The important thing is it needs
    114 * to be invoked "at least" once).
    115 */
    116void rpcrdma_force_disconnect(struct rpcrdma_ep *ep)
    117{
    118	if (atomic_add_unless(&ep->re_force_disconnect, 1, 1))
    119		xprt_force_disconnect(ep->re_xprt);
    120}
    121
    122/**
    123 * rpcrdma_flush_disconnect - Disconnect on flushed completion
    124 * @r_xprt: transport to disconnect
    125 * @wc: work completion entry
    126 *
    127 * Must be called in process context.
    128 */
    129void rpcrdma_flush_disconnect(struct rpcrdma_xprt *r_xprt, struct ib_wc *wc)
    130{
    131	if (wc->status != IB_WC_SUCCESS)
    132		rpcrdma_force_disconnect(r_xprt->rx_ep);
    133}
    134
    135/**
    136 * rpcrdma_wc_send - Invoked by RDMA provider for each polled Send WC
    137 * @cq:	completion queue
    138 * @wc:	WCE for a completed Send WR
    139 *
    140 */
    141static void rpcrdma_wc_send(struct ib_cq *cq, struct ib_wc *wc)
    142{
    143	struct ib_cqe *cqe = wc->wr_cqe;
    144	struct rpcrdma_sendctx *sc =
    145		container_of(cqe, struct rpcrdma_sendctx, sc_cqe);
    146	struct rpcrdma_xprt *r_xprt = cq->cq_context;
    147
    148	/* WARNING: Only wr_cqe and status are reliable at this point */
    149	trace_xprtrdma_wc_send(wc, &sc->sc_cid);
    150	rpcrdma_sendctx_put_locked(r_xprt, sc);
    151	rpcrdma_flush_disconnect(r_xprt, wc);
    152}
    153
    154/**
    155 * rpcrdma_wc_receive - Invoked by RDMA provider for each polled Receive WC
    156 * @cq:	completion queue
    157 * @wc:	WCE for a completed Receive WR
    158 *
    159 */
    160static void rpcrdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc)
    161{
    162	struct ib_cqe *cqe = wc->wr_cqe;
    163	struct rpcrdma_rep *rep = container_of(cqe, struct rpcrdma_rep,
    164					       rr_cqe);
    165	struct rpcrdma_xprt *r_xprt = cq->cq_context;
    166
    167	/* WARNING: Only wr_cqe and status are reliable at this point */
    168	trace_xprtrdma_wc_receive(wc, &rep->rr_cid);
    169	--r_xprt->rx_ep->re_receive_count;
    170	if (wc->status != IB_WC_SUCCESS)
    171		goto out_flushed;
    172
    173	/* status == SUCCESS means all fields in wc are trustworthy */
    174	rpcrdma_set_xdrlen(&rep->rr_hdrbuf, wc->byte_len);
    175	rep->rr_wc_flags = wc->wc_flags;
    176	rep->rr_inv_rkey = wc->ex.invalidate_rkey;
    177
    178	ib_dma_sync_single_for_cpu(rdmab_device(rep->rr_rdmabuf),
    179				   rdmab_addr(rep->rr_rdmabuf),
    180				   wc->byte_len, DMA_FROM_DEVICE);
    181
    182	rpcrdma_reply_handler(rep);
    183	return;
    184
    185out_flushed:
    186	rpcrdma_flush_disconnect(r_xprt, wc);
    187	rpcrdma_rep_put(&r_xprt->rx_buf, rep);
    188}
    189
    190static void rpcrdma_update_cm_private(struct rpcrdma_ep *ep,
    191				      struct rdma_conn_param *param)
    192{
    193	const struct rpcrdma_connect_private *pmsg = param->private_data;
    194	unsigned int rsize, wsize;
    195
    196	/* Default settings for RPC-over-RDMA Version One */
    197	rsize = RPCRDMA_V1_DEF_INLINE_SIZE;
    198	wsize = RPCRDMA_V1_DEF_INLINE_SIZE;
    199
    200	if (pmsg &&
    201	    pmsg->cp_magic == rpcrdma_cmp_magic &&
    202	    pmsg->cp_version == RPCRDMA_CMP_VERSION) {
    203		rsize = rpcrdma_decode_buffer_size(pmsg->cp_send_size);
    204		wsize = rpcrdma_decode_buffer_size(pmsg->cp_recv_size);
    205	}
    206
    207	if (rsize < ep->re_inline_recv)
    208		ep->re_inline_recv = rsize;
    209	if (wsize < ep->re_inline_send)
    210		ep->re_inline_send = wsize;
    211
    212	rpcrdma_set_max_header_sizes(ep);
    213}
    214
    215/**
    216 * rpcrdma_cm_event_handler - Handle RDMA CM events
    217 * @id: rdma_cm_id on which an event has occurred
    218 * @event: details of the event
    219 *
    220 * Called with @id's mutex held. Returns 1 if caller should
    221 * destroy @id, otherwise 0.
    222 */
    223static int
    224rpcrdma_cm_event_handler(struct rdma_cm_id *id, struct rdma_cm_event *event)
    225{
    226	struct sockaddr *sap = (struct sockaddr *)&id->route.addr.dst_addr;
    227	struct rpcrdma_ep *ep = id->context;
    228
    229	might_sleep();
    230
    231	switch (event->event) {
    232	case RDMA_CM_EVENT_ADDR_RESOLVED:
    233	case RDMA_CM_EVENT_ROUTE_RESOLVED:
    234		ep->re_async_rc = 0;
    235		complete(&ep->re_done);
    236		return 0;
    237	case RDMA_CM_EVENT_ADDR_ERROR:
    238		ep->re_async_rc = -EPROTO;
    239		complete(&ep->re_done);
    240		return 0;
    241	case RDMA_CM_EVENT_ROUTE_ERROR:
    242		ep->re_async_rc = -ENETUNREACH;
    243		complete(&ep->re_done);
    244		return 0;
    245	case RDMA_CM_EVENT_DEVICE_REMOVAL:
    246		pr_info("rpcrdma: removing device %s for %pISpc\n",
    247			ep->re_id->device->name, sap);
    248		fallthrough;
    249	case RDMA_CM_EVENT_ADDR_CHANGE:
    250		ep->re_connect_status = -ENODEV;
    251		goto disconnected;
    252	case RDMA_CM_EVENT_ESTABLISHED:
    253		rpcrdma_ep_get(ep);
    254		ep->re_connect_status = 1;
    255		rpcrdma_update_cm_private(ep, &event->param.conn);
    256		trace_xprtrdma_inline_thresh(ep);
    257		wake_up_all(&ep->re_connect_wait);
    258		break;
    259	case RDMA_CM_EVENT_CONNECT_ERROR:
    260		ep->re_connect_status = -ENOTCONN;
    261		goto wake_connect_worker;
    262	case RDMA_CM_EVENT_UNREACHABLE:
    263		ep->re_connect_status = -ENETUNREACH;
    264		goto wake_connect_worker;
    265	case RDMA_CM_EVENT_REJECTED:
    266		ep->re_connect_status = -ECONNREFUSED;
    267		if (event->status == IB_CM_REJ_STALE_CONN)
    268			ep->re_connect_status = -ENOTCONN;
    269wake_connect_worker:
    270		wake_up_all(&ep->re_connect_wait);
    271		return 0;
    272	case RDMA_CM_EVENT_DISCONNECTED:
    273		ep->re_connect_status = -ECONNABORTED;
    274disconnected:
    275		rpcrdma_force_disconnect(ep);
    276		return rpcrdma_ep_put(ep);
    277	default:
    278		break;
    279	}
    280
    281	return 0;
    282}
    283
    284static struct rdma_cm_id *rpcrdma_create_id(struct rpcrdma_xprt *r_xprt,
    285					    struct rpcrdma_ep *ep)
    286{
    287	unsigned long wtimeout = msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT) + 1;
    288	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
    289	struct rdma_cm_id *id;
    290	int rc;
    291
    292	init_completion(&ep->re_done);
    293
    294	id = rdma_create_id(xprt->xprt_net, rpcrdma_cm_event_handler, ep,
    295			    RDMA_PS_TCP, IB_QPT_RC);
    296	if (IS_ERR(id))
    297		return id;
    298
    299	ep->re_async_rc = -ETIMEDOUT;
    300	rc = rdma_resolve_addr(id, NULL, (struct sockaddr *)&xprt->addr,
    301			       RDMA_RESOLVE_TIMEOUT);
    302	if (rc)
    303		goto out;
    304	rc = wait_for_completion_interruptible_timeout(&ep->re_done, wtimeout);
    305	if (rc < 0)
    306		goto out;
    307
    308	rc = ep->re_async_rc;
    309	if (rc)
    310		goto out;
    311
    312	ep->re_async_rc = -ETIMEDOUT;
    313	rc = rdma_resolve_route(id, RDMA_RESOLVE_TIMEOUT);
    314	if (rc)
    315		goto out;
    316	rc = wait_for_completion_interruptible_timeout(&ep->re_done, wtimeout);
    317	if (rc < 0)
    318		goto out;
    319	rc = ep->re_async_rc;
    320	if (rc)
    321		goto out;
    322
    323	return id;
    324
    325out:
    326	rdma_destroy_id(id);
    327	return ERR_PTR(rc);
    328}
    329
    330static void rpcrdma_ep_destroy(struct kref *kref)
    331{
    332	struct rpcrdma_ep *ep = container_of(kref, struct rpcrdma_ep, re_kref);
    333
    334	if (ep->re_id->qp) {
    335		rdma_destroy_qp(ep->re_id);
    336		ep->re_id->qp = NULL;
    337	}
    338
    339	if (ep->re_attr.recv_cq)
    340		ib_free_cq(ep->re_attr.recv_cq);
    341	ep->re_attr.recv_cq = NULL;
    342	if (ep->re_attr.send_cq)
    343		ib_free_cq(ep->re_attr.send_cq);
    344	ep->re_attr.send_cq = NULL;
    345
    346	if (ep->re_pd)
    347		ib_dealloc_pd(ep->re_pd);
    348	ep->re_pd = NULL;
    349
    350	kfree(ep);
    351	module_put(THIS_MODULE);
    352}
    353
    354static noinline void rpcrdma_ep_get(struct rpcrdma_ep *ep)
    355{
    356	kref_get(&ep->re_kref);
    357}
    358
    359/* Returns:
    360 *     %0 if @ep still has a positive kref count, or
    361 *     %1 if @ep was destroyed successfully.
    362 */
    363static noinline int rpcrdma_ep_put(struct rpcrdma_ep *ep)
    364{
    365	return kref_put(&ep->re_kref, rpcrdma_ep_destroy);
    366}
    367
    368static int rpcrdma_ep_create(struct rpcrdma_xprt *r_xprt)
    369{
    370	struct rpcrdma_connect_private *pmsg;
    371	struct ib_device *device;
    372	struct rdma_cm_id *id;
    373	struct rpcrdma_ep *ep;
    374	int rc;
    375
    376	ep = kzalloc(sizeof(*ep), GFP_KERNEL);
    377	if (!ep)
    378		return -ENOTCONN;
    379	ep->re_xprt = &r_xprt->rx_xprt;
    380	kref_init(&ep->re_kref);
    381
    382	id = rpcrdma_create_id(r_xprt, ep);
    383	if (IS_ERR(id)) {
    384		kfree(ep);
    385		return PTR_ERR(id);
    386	}
    387	__module_get(THIS_MODULE);
    388	device = id->device;
    389	ep->re_id = id;
    390	reinit_completion(&ep->re_done);
    391
    392	ep->re_max_requests = r_xprt->rx_xprt.max_reqs;
    393	ep->re_inline_send = xprt_rdma_max_inline_write;
    394	ep->re_inline_recv = xprt_rdma_max_inline_read;
    395	rc = frwr_query_device(ep, device);
    396	if (rc)
    397		goto out_destroy;
    398
    399	r_xprt->rx_buf.rb_max_requests = cpu_to_be32(ep->re_max_requests);
    400
    401	ep->re_attr.srq = NULL;
    402	ep->re_attr.cap.max_inline_data = 0;
    403	ep->re_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
    404	ep->re_attr.qp_type = IB_QPT_RC;
    405	ep->re_attr.port_num = ~0;
    406
    407	ep->re_send_batch = ep->re_max_requests >> 3;
    408	ep->re_send_count = ep->re_send_batch;
    409	init_waitqueue_head(&ep->re_connect_wait);
    410
    411	ep->re_attr.send_cq = ib_alloc_cq_any(device, r_xprt,
    412					      ep->re_attr.cap.max_send_wr,
    413					      IB_POLL_WORKQUEUE);
    414	if (IS_ERR(ep->re_attr.send_cq)) {
    415		rc = PTR_ERR(ep->re_attr.send_cq);
    416		ep->re_attr.send_cq = NULL;
    417		goto out_destroy;
    418	}
    419
    420	ep->re_attr.recv_cq = ib_alloc_cq_any(device, r_xprt,
    421					      ep->re_attr.cap.max_recv_wr,
    422					      IB_POLL_WORKQUEUE);
    423	if (IS_ERR(ep->re_attr.recv_cq)) {
    424		rc = PTR_ERR(ep->re_attr.recv_cq);
    425		ep->re_attr.recv_cq = NULL;
    426		goto out_destroy;
    427	}
    428	ep->re_receive_count = 0;
    429
    430	/* Initialize cma parameters */
    431	memset(&ep->re_remote_cma, 0, sizeof(ep->re_remote_cma));
    432
    433	/* Prepare RDMA-CM private message */
    434	pmsg = &ep->re_cm_private;
    435	pmsg->cp_magic = rpcrdma_cmp_magic;
    436	pmsg->cp_version = RPCRDMA_CMP_VERSION;
    437	pmsg->cp_flags |= RPCRDMA_CMP_F_SND_W_INV_OK;
    438	pmsg->cp_send_size = rpcrdma_encode_buffer_size(ep->re_inline_send);
    439	pmsg->cp_recv_size = rpcrdma_encode_buffer_size(ep->re_inline_recv);
    440	ep->re_remote_cma.private_data = pmsg;
    441	ep->re_remote_cma.private_data_len = sizeof(*pmsg);
    442
    443	/* Client offers RDMA Read but does not initiate */
    444	ep->re_remote_cma.initiator_depth = 0;
    445	ep->re_remote_cma.responder_resources =
    446		min_t(int, U8_MAX, device->attrs.max_qp_rd_atom);
    447
    448	/* Limit transport retries so client can detect server
    449	 * GID changes quickly. RPC layer handles re-establishing
    450	 * transport connection and retransmission.
    451	 */
    452	ep->re_remote_cma.retry_count = 6;
    453
    454	/* RPC-over-RDMA handles its own flow control. In addition,
    455	 * make all RNR NAKs visible so we know that RPC-over-RDMA
    456	 * flow control is working correctly (no NAKs should be seen).
    457	 */
    458	ep->re_remote_cma.flow_control = 0;
    459	ep->re_remote_cma.rnr_retry_count = 0;
    460
    461	ep->re_pd = ib_alloc_pd(device, 0);
    462	if (IS_ERR(ep->re_pd)) {
    463		rc = PTR_ERR(ep->re_pd);
    464		ep->re_pd = NULL;
    465		goto out_destroy;
    466	}
    467
    468	rc = rdma_create_qp(id, ep->re_pd, &ep->re_attr);
    469	if (rc)
    470		goto out_destroy;
    471
    472	r_xprt->rx_ep = ep;
    473	return 0;
    474
    475out_destroy:
    476	rpcrdma_ep_put(ep);
    477	rdma_destroy_id(id);
    478	return rc;
    479}
    480
    481/**
    482 * rpcrdma_xprt_connect - Connect an unconnected transport
    483 * @r_xprt: controlling transport instance
    484 *
    485 * Returns 0 on success or a negative errno.
    486 */
    487int rpcrdma_xprt_connect(struct rpcrdma_xprt *r_xprt)
    488{
    489	struct rpc_xprt *xprt = &r_xprt->rx_xprt;
    490	struct rpcrdma_ep *ep;
    491	int rc;
    492
    493	rc = rpcrdma_ep_create(r_xprt);
    494	if (rc)
    495		return rc;
    496	ep = r_xprt->rx_ep;
    497
    498	xprt_clear_connected(xprt);
    499	rpcrdma_reset_cwnd(r_xprt);
    500
    501	/* Bump the ep's reference count while there are
    502	 * outstanding Receives.
    503	 */
    504	rpcrdma_ep_get(ep);
    505	rpcrdma_post_recvs(r_xprt, 1, true);
    506
    507	rc = rdma_connect(ep->re_id, &ep->re_remote_cma);
    508	if (rc)
    509		goto out;
    510
    511	if (xprt->reestablish_timeout < RPCRDMA_INIT_REEST_TO)
    512		xprt->reestablish_timeout = RPCRDMA_INIT_REEST_TO;
    513	wait_event_interruptible(ep->re_connect_wait,
    514				 ep->re_connect_status != 0);
    515	if (ep->re_connect_status <= 0) {
    516		rc = ep->re_connect_status;
    517		goto out;
    518	}
    519
    520	rc = rpcrdma_sendctxs_create(r_xprt);
    521	if (rc) {
    522		rc = -ENOTCONN;
    523		goto out;
    524	}
    525
    526	rc = rpcrdma_reqs_setup(r_xprt);
    527	if (rc) {
    528		rc = -ENOTCONN;
    529		goto out;
    530	}
    531	rpcrdma_mrs_create(r_xprt);
    532	frwr_wp_create(r_xprt);
    533
    534out:
    535	trace_xprtrdma_connect(r_xprt, rc);
    536	return rc;
    537}
    538
    539/**
    540 * rpcrdma_xprt_disconnect - Disconnect underlying transport
    541 * @r_xprt: controlling transport instance
    542 *
    543 * Caller serializes. Either the transport send lock is held,
    544 * or we're being called to destroy the transport.
    545 *
    546 * On return, @r_xprt is completely divested of all hardware
    547 * resources and prepared for the next ->connect operation.
    548 */
    549void rpcrdma_xprt_disconnect(struct rpcrdma_xprt *r_xprt)
    550{
    551	struct rpcrdma_ep *ep = r_xprt->rx_ep;
    552	struct rdma_cm_id *id;
    553	int rc;
    554
    555	if (!ep)
    556		return;
    557
    558	id = ep->re_id;
    559	rc = rdma_disconnect(id);
    560	trace_xprtrdma_disconnect(r_xprt, rc);
    561
    562	rpcrdma_xprt_drain(r_xprt);
    563	rpcrdma_reps_unmap(r_xprt);
    564	rpcrdma_reqs_reset(r_xprt);
    565	rpcrdma_mrs_destroy(r_xprt);
    566	rpcrdma_sendctxs_destroy(r_xprt);
    567
    568	if (rpcrdma_ep_put(ep))
    569		rdma_destroy_id(id);
    570
    571	r_xprt->rx_ep = NULL;
    572}
    573
    574/* Fixed-size circular FIFO queue. This implementation is wait-free and
    575 * lock-free.
    576 *
    577 * Consumer is the code path that posts Sends. This path dequeues a
    578 * sendctx for use by a Send operation. Multiple consumer threads
    579 * are serialized by the RPC transport lock, which allows only one
    580 * ->send_request call at a time.
    581 *
    582 * Producer is the code path that handles Send completions. This path
    583 * enqueues a sendctx that has been completed. Multiple producer
    584 * threads are serialized by the ib_poll_cq() function.
    585 */
    586
    587/* rpcrdma_sendctxs_destroy() assumes caller has already quiesced
    588 * queue activity, and rpcrdma_xprt_drain has flushed all remaining
    589 * Send requests.
    590 */
    591static void rpcrdma_sendctxs_destroy(struct rpcrdma_xprt *r_xprt)
    592{
    593	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
    594	unsigned long i;
    595
    596	if (!buf->rb_sc_ctxs)
    597		return;
    598	for (i = 0; i <= buf->rb_sc_last; i++)
    599		kfree(buf->rb_sc_ctxs[i]);
    600	kfree(buf->rb_sc_ctxs);
    601	buf->rb_sc_ctxs = NULL;
    602}
    603
    604static struct rpcrdma_sendctx *rpcrdma_sendctx_create(struct rpcrdma_ep *ep)
    605{
    606	struct rpcrdma_sendctx *sc;
    607
    608	sc = kzalloc(struct_size(sc, sc_sges, ep->re_attr.cap.max_send_sge),
    609		     GFP_KERNEL);
    610	if (!sc)
    611		return NULL;
    612
    613	sc->sc_cqe.done = rpcrdma_wc_send;
    614	sc->sc_cid.ci_queue_id = ep->re_attr.send_cq->res.id;
    615	sc->sc_cid.ci_completion_id =
    616		atomic_inc_return(&ep->re_completion_ids);
    617	return sc;
    618}
    619
    620static int rpcrdma_sendctxs_create(struct rpcrdma_xprt *r_xprt)
    621{
    622	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
    623	struct rpcrdma_sendctx *sc;
    624	unsigned long i;
    625
    626	/* Maximum number of concurrent outstanding Send WRs. Capping
    627	 * the circular queue size stops Send Queue overflow by causing
    628	 * the ->send_request call to fail temporarily before too many
    629	 * Sends are posted.
    630	 */
    631	i = r_xprt->rx_ep->re_max_requests + RPCRDMA_MAX_BC_REQUESTS;
    632	buf->rb_sc_ctxs = kcalloc(i, sizeof(sc), GFP_KERNEL);
    633	if (!buf->rb_sc_ctxs)
    634		return -ENOMEM;
    635
    636	buf->rb_sc_last = i - 1;
    637	for (i = 0; i <= buf->rb_sc_last; i++) {
    638		sc = rpcrdma_sendctx_create(r_xprt->rx_ep);
    639		if (!sc)
    640			return -ENOMEM;
    641
    642		buf->rb_sc_ctxs[i] = sc;
    643	}
    644
    645	buf->rb_sc_head = 0;
    646	buf->rb_sc_tail = 0;
    647	return 0;
    648}
    649
    650/* The sendctx queue is not guaranteed to have a size that is a
    651 * power of two, thus the helpers in circ_buf.h cannot be used.
    652 * The other option is to use modulus (%), which can be expensive.
    653 */
    654static unsigned long rpcrdma_sendctx_next(struct rpcrdma_buffer *buf,
    655					  unsigned long item)
    656{
    657	return likely(item < buf->rb_sc_last) ? item + 1 : 0;
    658}
    659
    660/**
    661 * rpcrdma_sendctx_get_locked - Acquire a send context
    662 * @r_xprt: controlling transport instance
    663 *
    664 * Returns pointer to a free send completion context; or NULL if
    665 * the queue is empty.
    666 *
    667 * Usage: Called to acquire an SGE array before preparing a Send WR.
    668 *
    669 * The caller serializes calls to this function (per transport), and
    670 * provides an effective memory barrier that flushes the new value
    671 * of rb_sc_head.
    672 */
    673struct rpcrdma_sendctx *rpcrdma_sendctx_get_locked(struct rpcrdma_xprt *r_xprt)
    674{
    675	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
    676	struct rpcrdma_sendctx *sc;
    677	unsigned long next_head;
    678
    679	next_head = rpcrdma_sendctx_next(buf, buf->rb_sc_head);
    680
    681	if (next_head == READ_ONCE(buf->rb_sc_tail))
    682		goto out_emptyq;
    683
    684	/* ORDER: item must be accessed _before_ head is updated */
    685	sc = buf->rb_sc_ctxs[next_head];
    686
    687	/* Releasing the lock in the caller acts as a memory
    688	 * barrier that flushes rb_sc_head.
    689	 */
    690	buf->rb_sc_head = next_head;
    691
    692	return sc;
    693
    694out_emptyq:
    695	/* The queue is "empty" if there have not been enough Send
    696	 * completions recently. This is a sign the Send Queue is
    697	 * backing up. Cause the caller to pause and try again.
    698	 */
    699	xprt_wait_for_buffer_space(&r_xprt->rx_xprt);
    700	r_xprt->rx_stats.empty_sendctx_q++;
    701	return NULL;
    702}
    703
    704/**
    705 * rpcrdma_sendctx_put_locked - Release a send context
    706 * @r_xprt: controlling transport instance
    707 * @sc: send context to release
    708 *
    709 * Usage: Called from Send completion to return a sendctxt
    710 * to the queue.
    711 *
    712 * The caller serializes calls to this function (per transport).
    713 */
    714static void rpcrdma_sendctx_put_locked(struct rpcrdma_xprt *r_xprt,
    715				       struct rpcrdma_sendctx *sc)
    716{
    717	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
    718	unsigned long next_tail;
    719
    720	/* Unmap SGEs of previously completed but unsignaled
    721	 * Sends by walking up the queue until @sc is found.
    722	 */
    723	next_tail = buf->rb_sc_tail;
    724	do {
    725		next_tail = rpcrdma_sendctx_next(buf, next_tail);
    726
    727		/* ORDER: item must be accessed _before_ tail is updated */
    728		rpcrdma_sendctx_unmap(buf->rb_sc_ctxs[next_tail]);
    729
    730	} while (buf->rb_sc_ctxs[next_tail] != sc);
    731
    732	/* Paired with READ_ONCE */
    733	smp_store_release(&buf->rb_sc_tail, next_tail);
    734
    735	xprt_write_space(&r_xprt->rx_xprt);
    736}
    737
    738static void
    739rpcrdma_mrs_create(struct rpcrdma_xprt *r_xprt)
    740{
    741	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
    742	struct rpcrdma_ep *ep = r_xprt->rx_ep;
    743	unsigned int count;
    744
    745	for (count = 0; count < ep->re_max_rdma_segs; count++) {
    746		struct rpcrdma_mr *mr;
    747		int rc;
    748
    749		mr = kzalloc(sizeof(*mr), GFP_KERNEL);
    750		if (!mr)
    751			break;
    752
    753		rc = frwr_mr_init(r_xprt, mr);
    754		if (rc) {
    755			kfree(mr);
    756			break;
    757		}
    758
    759		spin_lock(&buf->rb_lock);
    760		rpcrdma_mr_push(mr, &buf->rb_mrs);
    761		list_add(&mr->mr_all, &buf->rb_all_mrs);
    762		spin_unlock(&buf->rb_lock);
    763	}
    764
    765	r_xprt->rx_stats.mrs_allocated += count;
    766	trace_xprtrdma_createmrs(r_xprt, count);
    767}
    768
    769static void
    770rpcrdma_mr_refresh_worker(struct work_struct *work)
    771{
    772	struct rpcrdma_buffer *buf = container_of(work, struct rpcrdma_buffer,
    773						  rb_refresh_worker);
    774	struct rpcrdma_xprt *r_xprt = container_of(buf, struct rpcrdma_xprt,
    775						   rx_buf);
    776
    777	rpcrdma_mrs_create(r_xprt);
    778	xprt_write_space(&r_xprt->rx_xprt);
    779}
    780
    781/**
    782 * rpcrdma_mrs_refresh - Wake the MR refresh worker
    783 * @r_xprt: controlling transport instance
    784 *
    785 */
    786void rpcrdma_mrs_refresh(struct rpcrdma_xprt *r_xprt)
    787{
    788	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
    789	struct rpcrdma_ep *ep = r_xprt->rx_ep;
    790
    791	/* If there is no underlying connection, it's no use
    792	 * to wake the refresh worker.
    793	 */
    794	if (ep->re_connect_status == 1) {
    795		/* The work is scheduled on a WQ_MEM_RECLAIM
    796		 * workqueue in order to prevent MR allocation
    797		 * from recursing into NFS during direct reclaim.
    798		 */
    799		queue_work(xprtiod_workqueue, &buf->rb_refresh_worker);
    800	}
    801}
    802
    803/**
    804 * rpcrdma_req_create - Allocate an rpcrdma_req object
    805 * @r_xprt: controlling r_xprt
    806 * @size: initial size, in bytes, of send and receive buffers
    807 * @flags: GFP flags passed to memory allocators
    808 *
    809 * Returns an allocated and fully initialized rpcrdma_req or NULL.
    810 */
    811struct rpcrdma_req *rpcrdma_req_create(struct rpcrdma_xprt *r_xprt, size_t size,
    812				       gfp_t flags)
    813{
    814	struct rpcrdma_buffer *buffer = &r_xprt->rx_buf;
    815	struct rpcrdma_req *req;
    816
    817	req = kzalloc(sizeof(*req), flags);
    818	if (req == NULL)
    819		goto out1;
    820
    821	req->rl_sendbuf = rpcrdma_regbuf_alloc(size, DMA_TO_DEVICE, flags);
    822	if (!req->rl_sendbuf)
    823		goto out2;
    824
    825	req->rl_recvbuf = rpcrdma_regbuf_alloc(size, DMA_NONE, flags);
    826	if (!req->rl_recvbuf)
    827		goto out3;
    828
    829	INIT_LIST_HEAD(&req->rl_free_mrs);
    830	INIT_LIST_HEAD(&req->rl_registered);
    831	spin_lock(&buffer->rb_lock);
    832	list_add(&req->rl_all, &buffer->rb_allreqs);
    833	spin_unlock(&buffer->rb_lock);
    834	return req;
    835
    836out3:
    837	kfree(req->rl_sendbuf);
    838out2:
    839	kfree(req);
    840out1:
    841	return NULL;
    842}
    843
    844/**
    845 * rpcrdma_req_setup - Per-connection instance setup of an rpcrdma_req object
    846 * @r_xprt: controlling transport instance
    847 * @req: rpcrdma_req object to set up
    848 *
    849 * Returns zero on success, and a negative errno on failure.
    850 */
    851int rpcrdma_req_setup(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req)
    852{
    853	struct rpcrdma_regbuf *rb;
    854	size_t maxhdrsize;
    855
    856	/* Compute maximum header buffer size in bytes */
    857	maxhdrsize = rpcrdma_fixed_maxsz + 3 +
    858		     r_xprt->rx_ep->re_max_rdma_segs * rpcrdma_readchunk_maxsz;
    859	maxhdrsize *= sizeof(__be32);
    860	rb = rpcrdma_regbuf_alloc(__roundup_pow_of_two(maxhdrsize),
    861				  DMA_TO_DEVICE, GFP_KERNEL);
    862	if (!rb)
    863		goto out;
    864
    865	if (!__rpcrdma_regbuf_dma_map(r_xprt, rb))
    866		goto out_free;
    867
    868	req->rl_rdmabuf = rb;
    869	xdr_buf_init(&req->rl_hdrbuf, rdmab_data(rb), rdmab_length(rb));
    870	return 0;
    871
    872out_free:
    873	rpcrdma_regbuf_free(rb);
    874out:
    875	return -ENOMEM;
    876}
    877
    878/* ASSUMPTION: the rb_allreqs list is stable for the duration,
    879 * and thus can be walked without holding rb_lock. Eg. the
    880 * caller is holding the transport send lock to exclude
    881 * device removal or disconnection.
    882 */
    883static int rpcrdma_reqs_setup(struct rpcrdma_xprt *r_xprt)
    884{
    885	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
    886	struct rpcrdma_req *req;
    887	int rc;
    888
    889	list_for_each_entry(req, &buf->rb_allreqs, rl_all) {
    890		rc = rpcrdma_req_setup(r_xprt, req);
    891		if (rc)
    892			return rc;
    893	}
    894	return 0;
    895}
    896
    897static void rpcrdma_req_reset(struct rpcrdma_req *req)
    898{
    899	/* Credits are valid for only one connection */
    900	req->rl_slot.rq_cong = 0;
    901
    902	rpcrdma_regbuf_free(req->rl_rdmabuf);
    903	req->rl_rdmabuf = NULL;
    904
    905	rpcrdma_regbuf_dma_unmap(req->rl_sendbuf);
    906	rpcrdma_regbuf_dma_unmap(req->rl_recvbuf);
    907
    908	frwr_reset(req);
    909}
    910
    911/* ASSUMPTION: the rb_allreqs list is stable for the duration,
    912 * and thus can be walked without holding rb_lock. Eg. the
    913 * caller is holding the transport send lock to exclude
    914 * device removal or disconnection.
    915 */
    916static void rpcrdma_reqs_reset(struct rpcrdma_xprt *r_xprt)
    917{
    918	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
    919	struct rpcrdma_req *req;
    920
    921	list_for_each_entry(req, &buf->rb_allreqs, rl_all)
    922		rpcrdma_req_reset(req);
    923}
    924
    925static noinline
    926struct rpcrdma_rep *rpcrdma_rep_create(struct rpcrdma_xprt *r_xprt,
    927				       bool temp)
    928{
    929	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
    930	struct rpcrdma_rep *rep;
    931
    932	rep = kzalloc(sizeof(*rep), GFP_KERNEL);
    933	if (rep == NULL)
    934		goto out;
    935
    936	rep->rr_rdmabuf = rpcrdma_regbuf_alloc(r_xprt->rx_ep->re_inline_recv,
    937					       DMA_FROM_DEVICE, GFP_KERNEL);
    938	if (!rep->rr_rdmabuf)
    939		goto out_free;
    940
    941	if (!rpcrdma_regbuf_dma_map(r_xprt, rep->rr_rdmabuf))
    942		goto out_free_regbuf;
    943
    944	rep->rr_cid.ci_completion_id =
    945		atomic_inc_return(&r_xprt->rx_ep->re_completion_ids);
    946
    947	xdr_buf_init(&rep->rr_hdrbuf, rdmab_data(rep->rr_rdmabuf),
    948		     rdmab_length(rep->rr_rdmabuf));
    949	rep->rr_cqe.done = rpcrdma_wc_receive;
    950	rep->rr_rxprt = r_xprt;
    951	rep->rr_recv_wr.next = NULL;
    952	rep->rr_recv_wr.wr_cqe = &rep->rr_cqe;
    953	rep->rr_recv_wr.sg_list = &rep->rr_rdmabuf->rg_iov;
    954	rep->rr_recv_wr.num_sge = 1;
    955	rep->rr_temp = temp;
    956
    957	spin_lock(&buf->rb_lock);
    958	list_add(&rep->rr_all, &buf->rb_all_reps);
    959	spin_unlock(&buf->rb_lock);
    960	return rep;
    961
    962out_free_regbuf:
    963	rpcrdma_regbuf_free(rep->rr_rdmabuf);
    964out_free:
    965	kfree(rep);
    966out:
    967	return NULL;
    968}
    969
    970static void rpcrdma_rep_free(struct rpcrdma_rep *rep)
    971{
    972	rpcrdma_regbuf_free(rep->rr_rdmabuf);
    973	kfree(rep);
    974}
    975
    976static void rpcrdma_rep_destroy(struct rpcrdma_rep *rep)
    977{
    978	struct rpcrdma_buffer *buf = &rep->rr_rxprt->rx_buf;
    979
    980	spin_lock(&buf->rb_lock);
    981	list_del(&rep->rr_all);
    982	spin_unlock(&buf->rb_lock);
    983
    984	rpcrdma_rep_free(rep);
    985}
    986
    987static struct rpcrdma_rep *rpcrdma_rep_get_locked(struct rpcrdma_buffer *buf)
    988{
    989	struct llist_node *node;
    990
    991	/* Calls to llist_del_first are required to be serialized */
    992	node = llist_del_first(&buf->rb_free_reps);
    993	if (!node)
    994		return NULL;
    995	return llist_entry(node, struct rpcrdma_rep, rr_node);
    996}
    997
    998/**
    999 * rpcrdma_rep_put - Release rpcrdma_rep back to free list
   1000 * @buf: buffer pool
   1001 * @rep: rep to release
   1002 *
   1003 */
   1004void rpcrdma_rep_put(struct rpcrdma_buffer *buf, struct rpcrdma_rep *rep)
   1005{
   1006	llist_add(&rep->rr_node, &buf->rb_free_reps);
   1007}
   1008
   1009/* Caller must ensure the QP is quiescent (RQ is drained) before
   1010 * invoking this function, to guarantee rb_all_reps is not
   1011 * changing.
   1012 */
   1013static void rpcrdma_reps_unmap(struct rpcrdma_xprt *r_xprt)
   1014{
   1015	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
   1016	struct rpcrdma_rep *rep;
   1017
   1018	list_for_each_entry(rep, &buf->rb_all_reps, rr_all) {
   1019		rpcrdma_regbuf_dma_unmap(rep->rr_rdmabuf);
   1020		rep->rr_temp = true;	/* Mark this rep for destruction */
   1021	}
   1022}
   1023
   1024static void rpcrdma_reps_destroy(struct rpcrdma_buffer *buf)
   1025{
   1026	struct rpcrdma_rep *rep;
   1027
   1028	spin_lock(&buf->rb_lock);
   1029	while ((rep = list_first_entry_or_null(&buf->rb_all_reps,
   1030					       struct rpcrdma_rep,
   1031					       rr_all)) != NULL) {
   1032		list_del(&rep->rr_all);
   1033		spin_unlock(&buf->rb_lock);
   1034
   1035		rpcrdma_rep_free(rep);
   1036
   1037		spin_lock(&buf->rb_lock);
   1038	}
   1039	spin_unlock(&buf->rb_lock);
   1040}
   1041
   1042/**
   1043 * rpcrdma_buffer_create - Create initial set of req/rep objects
   1044 * @r_xprt: transport instance to (re)initialize
   1045 *
   1046 * Returns zero on success, otherwise a negative errno.
   1047 */
   1048int rpcrdma_buffer_create(struct rpcrdma_xprt *r_xprt)
   1049{
   1050	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
   1051	int i, rc;
   1052
   1053	buf->rb_bc_srv_max_requests = 0;
   1054	spin_lock_init(&buf->rb_lock);
   1055	INIT_LIST_HEAD(&buf->rb_mrs);
   1056	INIT_LIST_HEAD(&buf->rb_all_mrs);
   1057	INIT_WORK(&buf->rb_refresh_worker, rpcrdma_mr_refresh_worker);
   1058
   1059	INIT_LIST_HEAD(&buf->rb_send_bufs);
   1060	INIT_LIST_HEAD(&buf->rb_allreqs);
   1061	INIT_LIST_HEAD(&buf->rb_all_reps);
   1062
   1063	rc = -ENOMEM;
   1064	for (i = 0; i < r_xprt->rx_xprt.max_reqs; i++) {
   1065		struct rpcrdma_req *req;
   1066
   1067		req = rpcrdma_req_create(r_xprt, RPCRDMA_V1_DEF_INLINE_SIZE * 2,
   1068					 GFP_KERNEL);
   1069		if (!req)
   1070			goto out;
   1071		list_add(&req->rl_list, &buf->rb_send_bufs);
   1072	}
   1073
   1074	init_llist_head(&buf->rb_free_reps);
   1075
   1076	return 0;
   1077out:
   1078	rpcrdma_buffer_destroy(buf);
   1079	return rc;
   1080}
   1081
   1082/**
   1083 * rpcrdma_req_destroy - Destroy an rpcrdma_req object
   1084 * @req: unused object to be destroyed
   1085 *
   1086 * Relies on caller holding the transport send lock to protect
   1087 * removing req->rl_all from buf->rb_all_reqs safely.
   1088 */
   1089void rpcrdma_req_destroy(struct rpcrdma_req *req)
   1090{
   1091	struct rpcrdma_mr *mr;
   1092
   1093	list_del(&req->rl_all);
   1094
   1095	while ((mr = rpcrdma_mr_pop(&req->rl_free_mrs))) {
   1096		struct rpcrdma_buffer *buf = &mr->mr_xprt->rx_buf;
   1097
   1098		spin_lock(&buf->rb_lock);
   1099		list_del(&mr->mr_all);
   1100		spin_unlock(&buf->rb_lock);
   1101
   1102		frwr_mr_release(mr);
   1103	}
   1104
   1105	rpcrdma_regbuf_free(req->rl_recvbuf);
   1106	rpcrdma_regbuf_free(req->rl_sendbuf);
   1107	rpcrdma_regbuf_free(req->rl_rdmabuf);
   1108	kfree(req);
   1109}
   1110
   1111/**
   1112 * rpcrdma_mrs_destroy - Release all of a transport's MRs
   1113 * @r_xprt: controlling transport instance
   1114 *
   1115 * Relies on caller holding the transport send lock to protect
   1116 * removing mr->mr_list from req->rl_free_mrs safely.
   1117 */
   1118static void rpcrdma_mrs_destroy(struct rpcrdma_xprt *r_xprt)
   1119{
   1120	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
   1121	struct rpcrdma_mr *mr;
   1122
   1123	cancel_work_sync(&buf->rb_refresh_worker);
   1124
   1125	spin_lock(&buf->rb_lock);
   1126	while ((mr = list_first_entry_or_null(&buf->rb_all_mrs,
   1127					      struct rpcrdma_mr,
   1128					      mr_all)) != NULL) {
   1129		list_del(&mr->mr_list);
   1130		list_del(&mr->mr_all);
   1131		spin_unlock(&buf->rb_lock);
   1132
   1133		frwr_mr_release(mr);
   1134
   1135		spin_lock(&buf->rb_lock);
   1136	}
   1137	spin_unlock(&buf->rb_lock);
   1138}
   1139
   1140/**
   1141 * rpcrdma_buffer_destroy - Release all hw resources
   1142 * @buf: root control block for resources
   1143 *
   1144 * ORDERING: relies on a prior rpcrdma_xprt_drain :
   1145 * - No more Send or Receive completions can occur
   1146 * - All MRs, reps, and reqs are returned to their free lists
   1147 */
   1148void
   1149rpcrdma_buffer_destroy(struct rpcrdma_buffer *buf)
   1150{
   1151	rpcrdma_reps_destroy(buf);
   1152
   1153	while (!list_empty(&buf->rb_send_bufs)) {
   1154		struct rpcrdma_req *req;
   1155
   1156		req = list_first_entry(&buf->rb_send_bufs,
   1157				       struct rpcrdma_req, rl_list);
   1158		list_del(&req->rl_list);
   1159		rpcrdma_req_destroy(req);
   1160	}
   1161}
   1162
   1163/**
   1164 * rpcrdma_mr_get - Allocate an rpcrdma_mr object
   1165 * @r_xprt: controlling transport
   1166 *
   1167 * Returns an initialized rpcrdma_mr or NULL if no free
   1168 * rpcrdma_mr objects are available.
   1169 */
   1170struct rpcrdma_mr *
   1171rpcrdma_mr_get(struct rpcrdma_xprt *r_xprt)
   1172{
   1173	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
   1174	struct rpcrdma_mr *mr;
   1175
   1176	spin_lock(&buf->rb_lock);
   1177	mr = rpcrdma_mr_pop(&buf->rb_mrs);
   1178	spin_unlock(&buf->rb_lock);
   1179	return mr;
   1180}
   1181
   1182/**
   1183 * rpcrdma_reply_put - Put reply buffers back into pool
   1184 * @buffers: buffer pool
   1185 * @req: object to return
   1186 *
   1187 */
   1188void rpcrdma_reply_put(struct rpcrdma_buffer *buffers, struct rpcrdma_req *req)
   1189{
   1190	if (req->rl_reply) {
   1191		rpcrdma_rep_put(buffers, req->rl_reply);
   1192		req->rl_reply = NULL;
   1193	}
   1194}
   1195
   1196/**
   1197 * rpcrdma_buffer_get - Get a request buffer
   1198 * @buffers: Buffer pool from which to obtain a buffer
   1199 *
   1200 * Returns a fresh rpcrdma_req, or NULL if none are available.
   1201 */
   1202struct rpcrdma_req *
   1203rpcrdma_buffer_get(struct rpcrdma_buffer *buffers)
   1204{
   1205	struct rpcrdma_req *req;
   1206
   1207	spin_lock(&buffers->rb_lock);
   1208	req = list_first_entry_or_null(&buffers->rb_send_bufs,
   1209				       struct rpcrdma_req, rl_list);
   1210	if (req)
   1211		list_del_init(&req->rl_list);
   1212	spin_unlock(&buffers->rb_lock);
   1213	return req;
   1214}
   1215
   1216/**
   1217 * rpcrdma_buffer_put - Put request/reply buffers back into pool
   1218 * @buffers: buffer pool
   1219 * @req: object to return
   1220 *
   1221 */
   1222void rpcrdma_buffer_put(struct rpcrdma_buffer *buffers, struct rpcrdma_req *req)
   1223{
   1224	rpcrdma_reply_put(buffers, req);
   1225
   1226	spin_lock(&buffers->rb_lock);
   1227	list_add(&req->rl_list, &buffers->rb_send_bufs);
   1228	spin_unlock(&buffers->rb_lock);
   1229}
   1230
   1231/* Returns a pointer to a rpcrdma_regbuf object, or NULL.
   1232 *
   1233 * xprtrdma uses a regbuf for posting an outgoing RDMA SEND, or for
   1234 * receiving the payload of RDMA RECV operations. During Long Calls
   1235 * or Replies they may be registered externally via frwr_map.
   1236 */
   1237static struct rpcrdma_regbuf *
   1238rpcrdma_regbuf_alloc(size_t size, enum dma_data_direction direction,
   1239		     gfp_t flags)
   1240{
   1241	struct rpcrdma_regbuf *rb;
   1242
   1243	rb = kmalloc(sizeof(*rb), flags);
   1244	if (!rb)
   1245		return NULL;
   1246	rb->rg_data = kmalloc(size, flags);
   1247	if (!rb->rg_data) {
   1248		kfree(rb);
   1249		return NULL;
   1250	}
   1251
   1252	rb->rg_device = NULL;
   1253	rb->rg_direction = direction;
   1254	rb->rg_iov.length = size;
   1255	return rb;
   1256}
   1257
   1258/**
   1259 * rpcrdma_regbuf_realloc - re-allocate a SEND/RECV buffer
   1260 * @rb: regbuf to reallocate
   1261 * @size: size of buffer to be allocated, in bytes
   1262 * @flags: GFP flags
   1263 *
   1264 * Returns true if reallocation was successful. If false is
   1265 * returned, @rb is left untouched.
   1266 */
   1267bool rpcrdma_regbuf_realloc(struct rpcrdma_regbuf *rb, size_t size, gfp_t flags)
   1268{
   1269	void *buf;
   1270
   1271	buf = kmalloc(size, flags);
   1272	if (!buf)
   1273		return false;
   1274
   1275	rpcrdma_regbuf_dma_unmap(rb);
   1276	kfree(rb->rg_data);
   1277
   1278	rb->rg_data = buf;
   1279	rb->rg_iov.length = size;
   1280	return true;
   1281}
   1282
   1283/**
   1284 * __rpcrdma_regbuf_dma_map - DMA-map a regbuf
   1285 * @r_xprt: controlling transport instance
   1286 * @rb: regbuf to be mapped
   1287 *
   1288 * Returns true if the buffer is now DMA mapped to @r_xprt's device
   1289 */
   1290bool __rpcrdma_regbuf_dma_map(struct rpcrdma_xprt *r_xprt,
   1291			      struct rpcrdma_regbuf *rb)
   1292{
   1293	struct ib_device *device = r_xprt->rx_ep->re_id->device;
   1294
   1295	if (rb->rg_direction == DMA_NONE)
   1296		return false;
   1297
   1298	rb->rg_iov.addr = ib_dma_map_single(device, rdmab_data(rb),
   1299					    rdmab_length(rb), rb->rg_direction);
   1300	if (ib_dma_mapping_error(device, rdmab_addr(rb))) {
   1301		trace_xprtrdma_dma_maperr(rdmab_addr(rb));
   1302		return false;
   1303	}
   1304
   1305	rb->rg_device = device;
   1306	rb->rg_iov.lkey = r_xprt->rx_ep->re_pd->local_dma_lkey;
   1307	return true;
   1308}
   1309
   1310static void rpcrdma_regbuf_dma_unmap(struct rpcrdma_regbuf *rb)
   1311{
   1312	if (!rb)
   1313		return;
   1314
   1315	if (!rpcrdma_regbuf_is_mapped(rb))
   1316		return;
   1317
   1318	ib_dma_unmap_single(rb->rg_device, rdmab_addr(rb), rdmab_length(rb),
   1319			    rb->rg_direction);
   1320	rb->rg_device = NULL;
   1321}
   1322
   1323static void rpcrdma_regbuf_free(struct rpcrdma_regbuf *rb)
   1324{
   1325	rpcrdma_regbuf_dma_unmap(rb);
   1326	if (rb)
   1327		kfree(rb->rg_data);
   1328	kfree(rb);
   1329}
   1330
   1331/**
   1332 * rpcrdma_post_recvs - Refill the Receive Queue
   1333 * @r_xprt: controlling transport instance
   1334 * @needed: current credit grant
   1335 * @temp: mark Receive buffers to be deleted after one use
   1336 *
   1337 */
   1338void rpcrdma_post_recvs(struct rpcrdma_xprt *r_xprt, int needed, bool temp)
   1339{
   1340	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
   1341	struct rpcrdma_ep *ep = r_xprt->rx_ep;
   1342	struct ib_recv_wr *wr, *bad_wr;
   1343	struct rpcrdma_rep *rep;
   1344	int count, rc;
   1345
   1346	rc = 0;
   1347	count = 0;
   1348
   1349	if (likely(ep->re_receive_count > needed))
   1350		goto out;
   1351	needed -= ep->re_receive_count;
   1352	if (!temp)
   1353		needed += RPCRDMA_MAX_RECV_BATCH;
   1354
   1355	if (atomic_inc_return(&ep->re_receiving) > 1)
   1356		goto out;
   1357
   1358	/* fast path: all needed reps can be found on the free list */
   1359	wr = NULL;
   1360	while (needed) {
   1361		rep = rpcrdma_rep_get_locked(buf);
   1362		if (rep && rep->rr_temp) {
   1363			rpcrdma_rep_destroy(rep);
   1364			continue;
   1365		}
   1366		if (!rep)
   1367			rep = rpcrdma_rep_create(r_xprt, temp);
   1368		if (!rep)
   1369			break;
   1370
   1371		rep->rr_cid.ci_queue_id = ep->re_attr.recv_cq->res.id;
   1372		trace_xprtrdma_post_recv(rep);
   1373		rep->rr_recv_wr.next = wr;
   1374		wr = &rep->rr_recv_wr;
   1375		--needed;
   1376		++count;
   1377	}
   1378	if (!wr)
   1379		goto out;
   1380
   1381	rc = ib_post_recv(ep->re_id->qp, wr,
   1382			  (const struct ib_recv_wr **)&bad_wr);
   1383	if (rc) {
   1384		trace_xprtrdma_post_recvs_err(r_xprt, rc);
   1385		for (wr = bad_wr; wr;) {
   1386			struct rpcrdma_rep *rep;
   1387
   1388			rep = container_of(wr, struct rpcrdma_rep, rr_recv_wr);
   1389			wr = wr->next;
   1390			rpcrdma_rep_put(buf, rep);
   1391			--count;
   1392		}
   1393	}
   1394	if (atomic_dec_return(&ep->re_receiving) > 0)
   1395		complete(&ep->re_done);
   1396
   1397out:
   1398	trace_xprtrdma_post_recvs(r_xprt, count);
   1399	ep->re_receive_count += count;
   1400	return;
   1401}