linux-sg2042/net/9p/trans_rdma.c

766 lines
19 KiB
C
Raw Normal View History

/*
* linux/fs/9p/trans_rdma.c
*
* RDMA transport layer based on the trans_fd.c implementation.
*
* Copyright (C) 2008 by Tom Tucker <tom@opengridcomputing.com>
* Copyright (C) 2006 by Russ Cox <rsc@swtch.com>
* Copyright (C) 2004-2005 by Latchesar Ionkov <lucho@ionkov.net>
* Copyright (C) 2004-2008 by Eric Van Hensbergen <ericvh@gmail.com>
* Copyright (C) 1997-2002 by Ron Minnich <rminnich@sarnoff.com>
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2
* as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to:
* Free Software Foundation
* 51 Franklin Street, Fifth Floor
* Boston, MA 02111-1301 USA
*
*/
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
#include <linux/in.h>
#include <linux/module.h>
#include <linux/net.h>
#include <linux/ipv6.h>
#include <linux/kthread.h>
#include <linux/errno.h>
#include <linux/kernel.h>
#include <linux/un.h>
#include <linux/uaccess.h>
#include <linux/inet.h>
#include <linux/idr.h>
#include <linux/file.h>
#include <linux/parser.h>
#include <linux/semaphore.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 16:04:11 +08:00
#include <linux/slab.h>
#include <net/9p/9p.h>
#include <net/9p/client.h>
#include <net/9p/transport.h>
#include <rdma/ib_verbs.h>
#include <rdma/rdma_cm.h>
#define P9_PORT 5640
#define P9_RDMA_SQ_DEPTH 32
#define P9_RDMA_RQ_DEPTH 32
#define P9_RDMA_SEND_SGE 4
#define P9_RDMA_RECV_SGE 4
#define P9_RDMA_IRD 0
#define P9_RDMA_ORD 0
#define P9_RDMA_TIMEOUT 30000 /* 30 seconds */
#define P9_RDMA_MAXSIZE (1024*1024) /* 1MB */
/**
* struct p9_trans_rdma - RDMA transport instance
*
* @state: tracks the transport state machine for connection setup and tear down
* @cm_id: The RDMA CM ID
* @pd: Protection Domain pointer
* @qp: Queue Pair pointer
* @cq: Completion Queue pointer
* @dm_mr: DMA Memory Region pointer
* @lkey: The local access only memory region key
* @timeout: Number of uSecs to wait for connection management events
* @sq_depth: The depth of the Send Queue
* @sq_sem: Semaphore for the SQ
* @rq_depth: The depth of the Receive Queue.
* @rq_sem: Semaphore for the RQ
* @excess_rc : Amount of posted Receive Contexts without a pending request.
* See rdma_request()
* @addr: The remote peer's address
* @req_lock: Protects the active request list
* @cm_done: Completion event for connection management tracking
*/
struct p9_trans_rdma {
enum {
P9_RDMA_INIT,
P9_RDMA_ADDR_RESOLVED,
P9_RDMA_ROUTE_RESOLVED,
P9_RDMA_CONNECTED,
P9_RDMA_FLUSHING,
P9_RDMA_CLOSING,
P9_RDMA_CLOSED,
} state;
struct rdma_cm_id *cm_id;
struct ib_pd *pd;
struct ib_qp *qp;
struct ib_cq *cq;
struct ib_mr *dma_mr;
u32 lkey;
long timeout;
int sq_depth;
struct semaphore sq_sem;
int rq_depth;
struct semaphore rq_sem;
atomic_t excess_rc;
struct sockaddr_in addr;
spinlock_t req_lock;
struct completion cm_done;
};
/**
* p9_rdma_context - Keeps track of in-process WR
*
* @wc_op: The original WR op for when the CQE completes in error.
* @busa: Bus address to unmap when the WR completes
* @req: Keeps track of requests (send)
* @rc: Keepts track of replies (receive)
*/
struct p9_rdma_req;
struct p9_rdma_context {
enum ib_wc_opcode wc_op;
dma_addr_t busa;
union {
struct p9_req_t *req;
struct p9_fcall *rc;
};
};
/**
* p9_rdma_opts - Collection of mount options
* @port: port of connection
* @sq_depth: The requested depth of the SQ. This really doesn't need
* to be any deeper than the number of threads used in the client
* @rq_depth: The depth of the RQ. Should be greater than or equal to SQ depth
* @timeout: Time to wait in msecs for CM events
*/
struct p9_rdma_opts {
short port;
int sq_depth;
int rq_depth;
long timeout;
};
/*
* Option Parsing (code inspired by NFS code)
*/
enum {
/* Options that take integer arguments */
Opt_port, Opt_rq_depth, Opt_sq_depth, Opt_timeout, Opt_err,
};
static match_table_t tokens = {
{Opt_port, "port=%u"},
{Opt_sq_depth, "sq=%u"},
{Opt_rq_depth, "rq=%u"},
{Opt_timeout, "timeout=%u"},
{Opt_err, NULL},
};
/**
* parse_opts - parse mount options into rdma options structure
* @params: options string passed from mount
* @opts: rdma transport-specific structure to parse options into
*
* Returns 0 upon success, -ERRNO upon failure
*/
static int parse_opts(char *params, struct p9_rdma_opts *opts)
{
char *p;
substring_t args[MAX_OPT_ARGS];
int option;
char *options, *tmp_options;
opts->port = P9_PORT;
opts->sq_depth = P9_RDMA_SQ_DEPTH;
opts->rq_depth = P9_RDMA_RQ_DEPTH;
opts->timeout = P9_RDMA_TIMEOUT;
if (!params)
return 0;
tmp_options = kstrdup(params, GFP_KERNEL);
if (!tmp_options) {
p9_debug(P9_DEBUG_ERROR,
"failed to allocate copy of option string\n");
return -ENOMEM;
}
options = tmp_options;
while ((p = strsep(&options, ",")) != NULL) {
int token;
int r;
if (!*p)
continue;
token = match_token(p, tokens, args);
if (token == Opt_err)
continue;
r = match_int(&args[0], &option);
if (r < 0) {
p9_debug(P9_DEBUG_ERROR,
"integer field, but no integer?\n");
continue;
}
switch (token) {
case Opt_port:
opts->port = option;
break;
case Opt_sq_depth:
opts->sq_depth = option;
break;
case Opt_rq_depth:
opts->rq_depth = option;
break;
case Opt_timeout:
opts->timeout = option;
break;
default:
continue;
}
}
/* RQ must be at least as large as the SQ */
opts->rq_depth = max(opts->rq_depth, opts->sq_depth);
kfree(tmp_options);
return 0;
}
static int
p9_cm_event_handler(struct rdma_cm_id *id, struct rdma_cm_event *event)
{
struct p9_client *c = id->context;
struct p9_trans_rdma *rdma = c->trans;
switch (event->event) {
case RDMA_CM_EVENT_ADDR_RESOLVED:
BUG_ON(rdma->state != P9_RDMA_INIT);
rdma->state = P9_RDMA_ADDR_RESOLVED;
break;
case RDMA_CM_EVENT_ROUTE_RESOLVED:
BUG_ON(rdma->state != P9_RDMA_ADDR_RESOLVED);
rdma->state = P9_RDMA_ROUTE_RESOLVED;
break;
case RDMA_CM_EVENT_ESTABLISHED:
BUG_ON(rdma->state != P9_RDMA_ROUTE_RESOLVED);
rdma->state = P9_RDMA_CONNECTED;
break;
case RDMA_CM_EVENT_DISCONNECTED:
if (rdma)
rdma->state = P9_RDMA_CLOSED;
if (c)
c->status = Disconnected;
break;
case RDMA_CM_EVENT_TIMEWAIT_EXIT:
break;
case RDMA_CM_EVENT_ADDR_CHANGE:
case RDMA_CM_EVENT_ROUTE_ERROR:
case RDMA_CM_EVENT_DEVICE_REMOVAL:
case RDMA_CM_EVENT_MULTICAST_JOIN:
case RDMA_CM_EVENT_MULTICAST_ERROR:
case RDMA_CM_EVENT_REJECTED:
case RDMA_CM_EVENT_CONNECT_REQUEST:
case RDMA_CM_EVENT_CONNECT_RESPONSE:
case RDMA_CM_EVENT_CONNECT_ERROR:
case RDMA_CM_EVENT_ADDR_ERROR:
case RDMA_CM_EVENT_UNREACHABLE:
c->status = Disconnected;
rdma_disconnect(rdma->cm_id);
break;
default:
BUG();
}
complete(&rdma->cm_done);
return 0;
}
static void
handle_recv(struct p9_client *client, struct p9_trans_rdma *rdma,
struct p9_rdma_context *c, enum ib_wc_status status, u32 byte_len)
{
struct p9_req_t *req;
int err = 0;
int16_t tag;
req = NULL;
ib_dma_unmap_single(rdma->cm_id->device, c->busa, client->msize,
DMA_FROM_DEVICE);
if (status != IB_WC_SUCCESS)
goto err_out;
err = p9_parse_header(c->rc, NULL, NULL, &tag, 1);
if (err)
goto err_out;
req = p9_tag_lookup(client, tag);
if (!req)
goto err_out;
/* Check that we have not yet received a reply for this request.
*/
if (unlikely(req->rc)) {
pr_err("Duplicate reply for request %d", tag);
goto err_out;
}
req->rc = c->rc;
p9_client_cb(client, req, REQ_STATUS_RCVD);
return;
err_out:
p9_debug(P9_DEBUG_ERROR, "req %p err %d status %d\n", req, err, status);
rdma->state = P9_RDMA_FLUSHING;
client->status = Disconnected;
}
static void
handle_send(struct p9_client *client, struct p9_trans_rdma *rdma,
struct p9_rdma_context *c, enum ib_wc_status status, u32 byte_len)
{
ib_dma_unmap_single(rdma->cm_id->device,
c->busa, c->req->tc->size,
DMA_TO_DEVICE);
}
static void qp_event_handler(struct ib_event *event, void *context)
{
p9_debug(P9_DEBUG_ERROR, "QP event %d context %p\n",
event->event, context);
}
static void cq_comp_handler(struct ib_cq *cq, void *cq_context)
{
struct p9_client *client = cq_context;
struct p9_trans_rdma *rdma = client->trans;
int ret;
struct ib_wc wc;
ib_req_notify_cq(rdma->cq, IB_CQ_NEXT_COMP);
while ((ret = ib_poll_cq(cq, 1, &wc)) > 0) {
struct p9_rdma_context *c = (void *) (unsigned long) wc.wr_id;
switch (c->wc_op) {
case IB_WC_RECV:
handle_recv(client, rdma, c, wc.status, wc.byte_len);
up(&rdma->rq_sem);
break;
case IB_WC_SEND:
handle_send(client, rdma, c, wc.status, wc.byte_len);
up(&rdma->sq_sem);
break;
default:
pr_err("unexpected completion type, c->wc_op=%d, wc.opcode=%d, status=%d\n",
c->wc_op, wc.opcode, wc.status);
break;
}
kfree(c);
}
}
static void cq_event_handler(struct ib_event *e, void *v)
{
p9_debug(P9_DEBUG_ERROR, "CQ event %d context %p\n", e->event, v);
}
static void rdma_destroy_trans(struct p9_trans_rdma *rdma)
{
if (!rdma)
return;
if (rdma->dma_mr && !IS_ERR(rdma->dma_mr))
ib_dereg_mr(rdma->dma_mr);
if (rdma->qp && !IS_ERR(rdma->qp))
ib_destroy_qp(rdma->qp);
if (rdma->pd && !IS_ERR(rdma->pd))
ib_dealloc_pd(rdma->pd);
if (rdma->cq && !IS_ERR(rdma->cq))
ib_destroy_cq(rdma->cq);
if (rdma->cm_id && !IS_ERR(rdma->cm_id))
rdma_destroy_id(rdma->cm_id);
kfree(rdma);
}
static int
post_recv(struct p9_client *client, struct p9_rdma_context *c)
{
struct p9_trans_rdma *rdma = client->trans;
struct ib_recv_wr wr, *bad_wr;
struct ib_sge sge;
c->busa = ib_dma_map_single(rdma->cm_id->device,
c->rc->sdata, client->msize,
DMA_FROM_DEVICE);
if (ib_dma_mapping_error(rdma->cm_id->device, c->busa))
goto error;
sge.addr = c->busa;
sge.length = client->msize;
sge.lkey = rdma->lkey;
wr.next = NULL;
c->wc_op = IB_WC_RECV;
wr.wr_id = (unsigned long) c;
wr.sg_list = &sge;
wr.num_sge = 1;
return ib_post_recv(rdma->qp, &wr, &bad_wr);
error:
p9_debug(P9_DEBUG_ERROR, "EIO\n");
return -EIO;
}
static int rdma_request(struct p9_client *client, struct p9_req_t *req)
{
struct p9_trans_rdma *rdma = client->trans;
struct ib_send_wr wr, *bad_wr;
struct ib_sge sge;
int err = 0;
unsigned long flags;
struct p9_rdma_context *c = NULL;
struct p9_rdma_context *rpl_context = NULL;
/* When an error occurs between posting the recv and the send,
* there will be a receive context posted without a pending request.
* Since there is no way to "un-post" it, we remember it and skip
* post_recv() for the next request.
* So here,
* see if we are this `next request' and need to absorb an excess rc.
* If yes, then drop and free our own, and do not recv_post().
**/
if (unlikely(atomic_read(&rdma->excess_rc) > 0)) {
if ((atomic_sub_return(1, &rdma->excess_rc) >= 0)) {
/* Got one ! */
kfree(req->rc);
req->rc = NULL;
goto dont_need_post_recv;
} else {
/* We raced and lost. */
atomic_inc(&rdma->excess_rc);
}
}
/* Allocate an fcall for the reply */
net/9p: Convert the in the 9p rpc call path to GFP_NOFS Without this we can cause reclaim allocation in writepage. [ 3433.448430] ================================= [ 3433.449117] [ INFO: inconsistent lock state ] [ 3433.449117] 2.6.38-rc5+ #84 [ 3433.449117] --------------------------------- [ 3433.449117] inconsistent {RECLAIM_FS-ON-W} -> {IN-RECLAIM_FS-R} usage. [ 3433.449117] kswapd0/505 [HC0[0]:SC0[0]:HE1:SE1] takes: [ 3433.449117] (iprune_sem){+++++-}, at: [<ffffffff810ebbab>] shrink_icache_memory+0x45/0x2b1 [ 3433.449117] {RECLAIM_FS-ON-W} state was registered at: [ 3433.449117] [<ffffffff8107fe5f>] mark_held_locks+0x52/0x70 [ 3433.449117] [<ffffffff8107ff02>] lockdep_trace_alloc+0x85/0x9f [ 3433.449117] [<ffffffff810d353d>] slab_pre_alloc_hook+0x18/0x3c [ 3433.449117] [<ffffffff810d3fd5>] kmem_cache_alloc+0x23/0xa2 [ 3433.449117] [<ffffffff8127be77>] idr_pre_get+0x2d/0x6f [ 3433.449117] [<ffffffff815434eb>] p9_idpool_get+0x30/0xae [ 3433.449117] [<ffffffff81540123>] p9_client_rpc+0xd7/0x9b0 [ 3433.449117] [<ffffffff815427b0>] p9_client_clunk+0x88/0xdb [ 3433.449117] [<ffffffff811d56e5>] v9fs_evict_inode+0x3c/0x48 [ 3433.449117] [<ffffffff810eb511>] evict+0x1f/0x87 [ 3433.449117] [<ffffffff810eb5c0>] dispose_list+0x47/0xe3 [ 3433.449117] [<ffffffff810eb8da>] evict_inodes+0x138/0x14f [ 3433.449117] [<ffffffff810d90e2>] generic_shutdown_super+0x57/0xe8 [ 3433.449117] [<ffffffff810d91e8>] kill_anon_super+0x11/0x50 [ 3433.449117] [<ffffffff811d4951>] v9fs_kill_super+0x49/0xab [ 3433.449117] [<ffffffff810d926e>] deactivate_locked_super+0x21/0x46 [ 3433.449117] [<ffffffff810d9e84>] deactivate_super+0x40/0x44 [ 3433.449117] [<ffffffff810ef848>] mntput_no_expire+0x100/0x109 [ 3433.449117] [<ffffffff810f0aeb>] sys_umount+0x2f1/0x31c [ 3433.449117] [<ffffffff8102c87b>] system_call_fastpath+0x16/0x1b [ 3433.449117] irq event stamp: 192941 [ 3433.449117] hardirqs last enabled at (192941): [<ffffffff81568dcf>] _raw_spin_unlock_irq+0x2b/0x30 [ 3433.449117] hardirqs last disabled at (192940): [<ffffffff810b5f97>] shrink_inactive_list+0x290/0x2f5 [ 3433.449117] softirqs last enabled at (188470): [<ffffffff8105fd65>] __do_softirq+0x133/0x152 [ 3433.449117] softirqs last disabled at (188455): [<ffffffff8102d7cc>] call_softirq+0x1c/0x28 [ 3433.449117] [ 3433.449117] other info that might help us debug this: [ 3433.449117] 1 lock held by kswapd0/505: [ 3433.449117] #0: (shrinker_rwsem){++++..}, at: [<ffffffff810b52e2>] shrink_slab+0x38/0x15f [ 3433.449117] [ 3433.449117] stack backtrace: [ 3433.449117] Pid: 505, comm: kswapd0 Not tainted 2.6.38-rc5+ #84 [ 3433.449117] Call Trace: [ 3433.449117] [<ffffffff8107fbce>] ? valid_state+0x17e/0x191 [ 3433.449117] [<ffffffff81036896>] ? save_stack_trace+0x28/0x45 [ 3433.449117] [<ffffffff81080426>] ? check_usage_forwards+0x0/0x87 [ 3433.449117] [<ffffffff8107fcf4>] ? mark_lock+0x113/0x22c [ 3433.449117] [<ffffffff8108105f>] ? __lock_acquire+0x37a/0xcf7 [ 3433.449117] [<ffffffff8107fc0e>] ? mark_lock+0x2d/0x22c [ 3433.449117] [<ffffffff81081077>] ? __lock_acquire+0x392/0xcf7 [ 3433.449117] [<ffffffff810b14d2>] ? determine_dirtyable_memory+0x15/0x28 [ 3433.449117] [<ffffffff81081a33>] ? lock_acquire+0x57/0x6d [ 3433.449117] [<ffffffff810ebbab>] ? shrink_icache_memory+0x45/0x2b1 [ 3433.449117] [<ffffffff81567d85>] ? down_read+0x47/0x5c [ 3433.449117] [<ffffffff810ebbab>] ? shrink_icache_memory+0x45/0x2b1 [ 3433.449117] [<ffffffff810ebbab>] ? shrink_icache_memory+0x45/0x2b1 [ 3433.449117] [<ffffffff810b5385>] ? shrink_slab+0xdb/0x15f [ 3433.449117] [<ffffffff810b69bc>] ? kswapd+0x574/0x96a [ 3433.449117] [<ffffffff810b6448>] ? kswapd+0x0/0x96a [ 3433.449117] [<ffffffff810714e2>] ? kthread+0x7d/0x85 [ 3433.449117] [<ffffffff8102d6d4>] ? kernel_thread_helper+0x4/0x10 [ 3433.449117] [<ffffffff81569200>] ? restore_args+0x0/0x30 [ 3433.449117] [<ffffffff81071465>] ? kthread+0x0/0x85 [ 3433.449117] [<ffffffff8102d6d0>] ? kernel_thread_helper+0x0/0x10 Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> Signed-off-by: Venkateswararao Jujjuri <jvrao@linux.vnet.ibm.com> Signed-off-by: Eric Van Hensbergen <ericvh@gmail.com>
2011-03-08 19:09:47 +08:00
rpl_context = kmalloc(sizeof *rpl_context, GFP_NOFS);
if (!rpl_context) {
err = -ENOMEM;
goto recv_error;
}
rpl_context->rc = req->rc;
/*
* Post a receive buffer for this request. We need to ensure
* there is a reply buffer available for every outstanding
* request. A flushed request can result in no reply for an
* outstanding request, so we must keep a count to avoid
* overflowing the RQ.
*/
if (down_interruptible(&rdma->rq_sem)) {
err = -EINTR;
goto recv_error;
}
err = post_recv(client, rpl_context);
if (err) {
p9_debug(P9_DEBUG_FCALL, "POST RECV failed\n");
goto recv_error;
}
/* remove posted receive buffer from request structure */
req->rc = NULL;
dont_need_post_recv:
/* Post the request */
net/9p: Convert the in the 9p rpc call path to GFP_NOFS Without this we can cause reclaim allocation in writepage. [ 3433.448430] ================================= [ 3433.449117] [ INFO: inconsistent lock state ] [ 3433.449117] 2.6.38-rc5+ #84 [ 3433.449117] --------------------------------- [ 3433.449117] inconsistent {RECLAIM_FS-ON-W} -> {IN-RECLAIM_FS-R} usage. [ 3433.449117] kswapd0/505 [HC0[0]:SC0[0]:HE1:SE1] takes: [ 3433.449117] (iprune_sem){+++++-}, at: [<ffffffff810ebbab>] shrink_icache_memory+0x45/0x2b1 [ 3433.449117] {RECLAIM_FS-ON-W} state was registered at: [ 3433.449117] [<ffffffff8107fe5f>] mark_held_locks+0x52/0x70 [ 3433.449117] [<ffffffff8107ff02>] lockdep_trace_alloc+0x85/0x9f [ 3433.449117] [<ffffffff810d353d>] slab_pre_alloc_hook+0x18/0x3c [ 3433.449117] [<ffffffff810d3fd5>] kmem_cache_alloc+0x23/0xa2 [ 3433.449117] [<ffffffff8127be77>] idr_pre_get+0x2d/0x6f [ 3433.449117] [<ffffffff815434eb>] p9_idpool_get+0x30/0xae [ 3433.449117] [<ffffffff81540123>] p9_client_rpc+0xd7/0x9b0 [ 3433.449117] [<ffffffff815427b0>] p9_client_clunk+0x88/0xdb [ 3433.449117] [<ffffffff811d56e5>] v9fs_evict_inode+0x3c/0x48 [ 3433.449117] [<ffffffff810eb511>] evict+0x1f/0x87 [ 3433.449117] [<ffffffff810eb5c0>] dispose_list+0x47/0xe3 [ 3433.449117] [<ffffffff810eb8da>] evict_inodes+0x138/0x14f [ 3433.449117] [<ffffffff810d90e2>] generic_shutdown_super+0x57/0xe8 [ 3433.449117] [<ffffffff810d91e8>] kill_anon_super+0x11/0x50 [ 3433.449117] [<ffffffff811d4951>] v9fs_kill_super+0x49/0xab [ 3433.449117] [<ffffffff810d926e>] deactivate_locked_super+0x21/0x46 [ 3433.449117] [<ffffffff810d9e84>] deactivate_super+0x40/0x44 [ 3433.449117] [<ffffffff810ef848>] mntput_no_expire+0x100/0x109 [ 3433.449117] [<ffffffff810f0aeb>] sys_umount+0x2f1/0x31c [ 3433.449117] [<ffffffff8102c87b>] system_call_fastpath+0x16/0x1b [ 3433.449117] irq event stamp: 192941 [ 3433.449117] hardirqs last enabled at (192941): [<ffffffff81568dcf>] _raw_spin_unlock_irq+0x2b/0x30 [ 3433.449117] hardirqs last disabled at (192940): [<ffffffff810b5f97>] shrink_inactive_list+0x290/0x2f5 [ 3433.449117] softirqs last enabled at (188470): [<ffffffff8105fd65>] __do_softirq+0x133/0x152 [ 3433.449117] softirqs last disabled at (188455): [<ffffffff8102d7cc>] call_softirq+0x1c/0x28 [ 3433.449117] [ 3433.449117] other info that might help us debug this: [ 3433.449117] 1 lock held by kswapd0/505: [ 3433.449117] #0: (shrinker_rwsem){++++..}, at: [<ffffffff810b52e2>] shrink_slab+0x38/0x15f [ 3433.449117] [ 3433.449117] stack backtrace: [ 3433.449117] Pid: 505, comm: kswapd0 Not tainted 2.6.38-rc5+ #84 [ 3433.449117] Call Trace: [ 3433.449117] [<ffffffff8107fbce>] ? valid_state+0x17e/0x191 [ 3433.449117] [<ffffffff81036896>] ? save_stack_trace+0x28/0x45 [ 3433.449117] [<ffffffff81080426>] ? check_usage_forwards+0x0/0x87 [ 3433.449117] [<ffffffff8107fcf4>] ? mark_lock+0x113/0x22c [ 3433.449117] [<ffffffff8108105f>] ? __lock_acquire+0x37a/0xcf7 [ 3433.449117] [<ffffffff8107fc0e>] ? mark_lock+0x2d/0x22c [ 3433.449117] [<ffffffff81081077>] ? __lock_acquire+0x392/0xcf7 [ 3433.449117] [<ffffffff810b14d2>] ? determine_dirtyable_memory+0x15/0x28 [ 3433.449117] [<ffffffff81081a33>] ? lock_acquire+0x57/0x6d [ 3433.449117] [<ffffffff810ebbab>] ? shrink_icache_memory+0x45/0x2b1 [ 3433.449117] [<ffffffff81567d85>] ? down_read+0x47/0x5c [ 3433.449117] [<ffffffff810ebbab>] ? shrink_icache_memory+0x45/0x2b1 [ 3433.449117] [<ffffffff810ebbab>] ? shrink_icache_memory+0x45/0x2b1 [ 3433.449117] [<ffffffff810b5385>] ? shrink_slab+0xdb/0x15f [ 3433.449117] [<ffffffff810b69bc>] ? kswapd+0x574/0x96a [ 3433.449117] [<ffffffff810b6448>] ? kswapd+0x0/0x96a [ 3433.449117] [<ffffffff810714e2>] ? kthread+0x7d/0x85 [ 3433.449117] [<ffffffff8102d6d4>] ? kernel_thread_helper+0x4/0x10 [ 3433.449117] [<ffffffff81569200>] ? restore_args+0x0/0x30 [ 3433.449117] [<ffffffff81071465>] ? kthread+0x0/0x85 [ 3433.449117] [<ffffffff8102d6d0>] ? kernel_thread_helper+0x0/0x10 Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> Signed-off-by: Venkateswararao Jujjuri <jvrao@linux.vnet.ibm.com> Signed-off-by: Eric Van Hensbergen <ericvh@gmail.com>
2011-03-08 19:09:47 +08:00
c = kmalloc(sizeof *c, GFP_NOFS);
if (!c) {
err = -ENOMEM;
goto send_error;
}
c->req = req;
c->busa = ib_dma_map_single(rdma->cm_id->device,
c->req->tc->sdata, c->req->tc->size,
DMA_TO_DEVICE);
if (ib_dma_mapping_error(rdma->cm_id->device, c->busa)) {
err = -EIO;
goto send_error;
}
sge.addr = c->busa;
sge.length = c->req->tc->size;
sge.lkey = rdma->lkey;
wr.next = NULL;
c->wc_op = IB_WC_SEND;
wr.wr_id = (unsigned long) c;
wr.opcode = IB_WR_SEND;
wr.send_flags = IB_SEND_SIGNALED;
wr.sg_list = &sge;
wr.num_sge = 1;
if (down_interruptible(&rdma->sq_sem)) {
err = -EINTR;
goto send_error;
}
/* Mark request as `sent' *before* we actually send it,
* because doing if after could erase the REQ_STATUS_RCVD
* status in case of a very fast reply.
*/
req->status = REQ_STATUS_SENT;
err = ib_post_send(rdma->qp, &wr, &bad_wr);
if (err)
goto send_error;
/* Success */
return 0;
/* Handle errors that happened during or while preparing the send: */
send_error:
req->status = REQ_STATUS_ERROR;
kfree(c);
p9_debug(P9_DEBUG_ERROR, "Error %d in rdma_request()\n", err);
/* Ach.
* We did recv_post(), but not send. We have one recv_post in excess.
*/
atomic_inc(&rdma->excess_rc);
return err;
/* Handle errors that happened during or while preparing post_recv(): */
recv_error:
kfree(rpl_context);
spin_lock_irqsave(&rdma->req_lock, flags);
if (rdma->state < P9_RDMA_CLOSING) {
rdma->state = P9_RDMA_CLOSING;
spin_unlock_irqrestore(&rdma->req_lock, flags);
rdma_disconnect(rdma->cm_id);
} else
spin_unlock_irqrestore(&rdma->req_lock, flags);
return err;
}
static void rdma_close(struct p9_client *client)
{
struct p9_trans_rdma *rdma;
if (!client)
return;
rdma = client->trans;
if (!rdma)
return;
client->status = Disconnected;
rdma_disconnect(rdma->cm_id);
rdma_destroy_trans(rdma);
}
/**
* alloc_rdma - Allocate and initialize the rdma transport structure
* @opts: Mount options structure
*/
static struct p9_trans_rdma *alloc_rdma(struct p9_rdma_opts *opts)
{
struct p9_trans_rdma *rdma;
rdma = kzalloc(sizeof(struct p9_trans_rdma), GFP_KERNEL);
if (!rdma)
return NULL;
rdma->sq_depth = opts->sq_depth;
rdma->rq_depth = opts->rq_depth;
rdma->timeout = opts->timeout;
spin_lock_init(&rdma->req_lock);
init_completion(&rdma->cm_done);
sema_init(&rdma->sq_sem, rdma->sq_depth);
sema_init(&rdma->rq_sem, rdma->rq_depth);
atomic_set(&rdma->excess_rc, 0);
return rdma;
}
static int rdma_cancel(struct p9_client *client, struct p9_req_t *req)
{
/* Nothing to do here.
* We will take care of it (if we have to) in rdma_cancelled()
*/
return 1;
}
/* A request has been fully flushed without a reply.
* That means we have posted one buffer in excess.
*/
static int rdma_cancelled(struct p9_client *client, struct p9_req_t *req)
{
struct p9_trans_rdma *rdma = client->trans;
atomic_inc(&rdma->excess_rc);
return 0;
}
/**
* trans_create_rdma - Transport method for creating atransport instance
* @client: client instance
* @addr: IP address string
* @args: Mount options string
*/
static int
rdma_create_trans(struct p9_client *client, const char *addr, char *args)
{
int err;
struct p9_rdma_opts opts;
struct p9_trans_rdma *rdma;
struct rdma_conn_param conn_param;
struct ib_qp_init_attr qp_attr;
struct ib_device_attr devattr;
/* Parse the transport specific mount options */
err = parse_opts(args, &opts);
if (err < 0)
return err;
/* Create and initialize the RDMA transport structure */
rdma = alloc_rdma(&opts);
if (!rdma)
return -ENOMEM;
/* Create the RDMA CM ID */
rdma->cm_id = rdma_create_id(p9_cm_event_handler, client, RDMA_PS_TCP,
IB_QPT_RC);
if (IS_ERR(rdma->cm_id))
goto error;
/* Associate the client with the transport */
client->trans = rdma;
/* Resolve the server's address */
rdma->addr.sin_family = AF_INET;
rdma->addr.sin_addr.s_addr = in_aton(addr);
rdma->addr.sin_port = htons(opts.port);
err = rdma_resolve_addr(rdma->cm_id, NULL,
(struct sockaddr *)&rdma->addr,
rdma->timeout);
if (err)
goto error;
err = wait_for_completion_interruptible(&rdma->cm_done);
if (err || (rdma->state != P9_RDMA_ADDR_RESOLVED))
goto error;
/* Resolve the route to the server */
err = rdma_resolve_route(rdma->cm_id, rdma->timeout);
if (err)
goto error;
err = wait_for_completion_interruptible(&rdma->cm_done);
if (err || (rdma->state != P9_RDMA_ROUTE_RESOLVED))
goto error;
/* Query the device attributes */
err = ib_query_device(rdma->cm_id->device, &devattr);
if (err)
goto error;
/* Create the Completion Queue */
rdma->cq = ib_create_cq(rdma->cm_id->device, cq_comp_handler,
cq_event_handler, client,
opts.sq_depth + opts.rq_depth + 1, 0);
if (IS_ERR(rdma->cq))
goto error;
ib_req_notify_cq(rdma->cq, IB_CQ_NEXT_COMP);
/* Create the Protection Domain */
rdma->pd = ib_alloc_pd(rdma->cm_id->device);
if (IS_ERR(rdma->pd))
goto error;
/* Cache the DMA lkey in the transport */
rdma->dma_mr = NULL;
if (devattr.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)
rdma->lkey = rdma->cm_id->device->local_dma_lkey;
else {
rdma->dma_mr = ib_get_dma_mr(rdma->pd, IB_ACCESS_LOCAL_WRITE);
if (IS_ERR(rdma->dma_mr))
goto error;
rdma->lkey = rdma->dma_mr->lkey;
}
/* Create the Queue Pair */
memset(&qp_attr, 0, sizeof qp_attr);
qp_attr.event_handler = qp_event_handler;
qp_attr.qp_context = client;
qp_attr.cap.max_send_wr = opts.sq_depth;
qp_attr.cap.max_recv_wr = opts.rq_depth;
qp_attr.cap.max_send_sge = P9_RDMA_SEND_SGE;
qp_attr.cap.max_recv_sge = P9_RDMA_RECV_SGE;
qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
qp_attr.qp_type = IB_QPT_RC;
qp_attr.send_cq = rdma->cq;
qp_attr.recv_cq = rdma->cq;
err = rdma_create_qp(rdma->cm_id, rdma->pd, &qp_attr);
if (err)
goto error;
rdma->qp = rdma->cm_id->qp;
/* Request a connection */
memset(&conn_param, 0, sizeof(conn_param));
conn_param.private_data = NULL;
conn_param.private_data_len = 0;
conn_param.responder_resources = P9_RDMA_IRD;
conn_param.initiator_depth = P9_RDMA_ORD;
err = rdma_connect(rdma->cm_id, &conn_param);
if (err)
goto error;
err = wait_for_completion_interruptible(&rdma->cm_done);
if (err || (rdma->state != P9_RDMA_CONNECTED))
goto error;
client->status = Connected;
return 0;
error:
rdma_destroy_trans(rdma);
return -ENOTCONN;
}
static struct p9_trans_module p9_rdma_trans = {
.name = "rdma",
.maxsize = P9_RDMA_MAXSIZE,
.def = 0,
.owner = THIS_MODULE,
.create = rdma_create_trans,
.close = rdma_close,
.request = rdma_request,
.cancel = rdma_cancel,
.cancelled = rdma_cancelled,
};
/**
* p9_trans_rdma_init - Register the 9P RDMA transport driver
*/
static int __init p9_trans_rdma_init(void)
{
v9fs_register_trans(&p9_rdma_trans);
return 0;
}
static void __exit p9_trans_rdma_exit(void)
{
v9fs_unregister_trans(&p9_rdma_trans);
}
module_init(p9_trans_rdma_init);
module_exit(p9_trans_rdma_exit);
MODULE_AUTHOR("Tom Tucker <tom@opengridcomputing.com>");
MODULE_DESCRIPTION("RDMA Transport for 9P");
MODULE_LICENSE("Dual BSD/GPL");