From: Oleg Drokin <green@linuxhacker.ru>
To: Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
linux-kernel@vger.kernel.org, devel@driverdev.osuosl.org
Cc: Dmitry Eremin <dmitry.eremin@intel.com>,
Oleg Drokin <oleg.drokin@intel.com>
Subject: [PATCH 36/47] staging/lustre/clio: replace semaphore with mutex
Date: Sun, 27 Apr 2014 13:07:00 -0400 [thread overview]
Message-ID: <1398618431-29757-37-git-send-email-green@linuxhacker.ru> (raw)
In-Reply-To: <1398618431-29757-1-git-send-email-green@linuxhacker.ru>
From: Dmitry Eremin <dmitry.eremin@intel.com>
According https://www.kernel.org/doc/Documentation/mutex-design.txt:
- the mutex subsystem is slightly faster and has better scalability
for contended workloads. In terms of 'ops per CPU cycle', the
semaphore kernel performed 551 ops/sec per 1% of CPU time used,
while the mutex kernel performed 3825 ops/sec per 1% of CPU time
used - it was 6.9 times more efficient.
- there are no fastpath tradeoffs, the mutex fastpath is just as
tight as the semaphore fastpath. On x86, the locking fastpath is
2 instructions.
- 'struct mutex' semantics are well-defined and are enforced if
CONFIG_DEBUG_MUTEXES is turned on. Semaphores on the other hand
have virtually no debugging code or instrumentation.
One more benefit of mutex is optimistic spinning. It try to spin for
acquisition when there are no pending waiters and the lock owner is
currently running on a (different) CPU. The rationale is that if the
lock owner is running, it is likely to release the lock soon.
This significantly reduce amount of context switches when locked
region is small and we have high contention.
Signed-off-by: Dmitry Eremin <dmitry.eremin@intel.com>
Reviewed-on: http://review.whamcloud.com/9095
Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-4257
Reviewed-by: Andreas Dilger <andreas.dilger@intel.com>
Reviewed-by: James Simmons <uja.ornl@gmail.com>
Signed-off-by: Oleg Drokin <oleg.drokin@intel.com>
---
drivers/staging/lustre/lustre/llite/llite_internal.h | 10 ++++------
drivers/staging/lustre/lustre/llite/llite_lib.c | 12 +++---------
2 files changed, 7 insertions(+), 15 deletions(-)
diff --git a/drivers/staging/lustre/lustre/llite/llite_internal.h b/drivers/staging/lustre/lustre/llite/llite_internal.h
index d2f8250..f4b15bf 100644
--- a/drivers/staging/lustre/lustre/llite/llite_internal.h
+++ b/drivers/staging/lustre/lustre/llite/llite_internal.h
@@ -211,8 +211,7 @@ struct ll_inode_info {
/* for non-directory */
struct {
- struct semaphore f_size_sem;
- void *f_size_sem_owner;
+ struct mutex f_size_mutex;
char *f_symlink_name;
__u64 f_maxbytes;
/*
@@ -249,8 +248,7 @@ struct ll_inode_info {
char f_jobid[JOBSTATS_JOBID_SIZE];
} f;
-#define lli_size_sem u.f.f_size_sem
-#define lli_size_sem_owner u.f.f_size_sem_owner
+#define lli_size_mutex u.f.f_size_mutex
#define lli_symlink_name u.f.f_symlink_name
#define lli_maxbytes u.f.f_maxbytes
#define lli_trunc_sem u.f.f_trunc_sem
@@ -319,7 +317,7 @@ int ll_xattr_cache_get(struct inode *inode,
* Locking to guarantee consistency of non-atomic updates to long long i_size,
* consistency between file size and KMS.
*
- * Implemented by ->lli_size_sem and ->lsm_lock, nested in that order.
+ * Implemented by ->lli_size_mutex and ->lsm_lock, nested in that order.
*/
void ll_inode_size_lock(struct inode *inode);
@@ -1448,7 +1446,7 @@ static inline void cl_isize_unlock(struct inode *inode)
static inline void cl_isize_write_nolock(struct inode *inode, loff_t kms)
{
- LASSERT(down_trylock(&ll_i2info(inode)->lli_size_sem) != 0);
+ LASSERT(mutex_is_locked(&ll_i2info(inode)->lli_size_mutex));
i_size_write(inode, kms);
}
diff --git a/drivers/staging/lustre/lustre/llite/llite_lib.c b/drivers/staging/lustre/lustre/llite/llite_lib.c
index befc30b..34df5d9 100644
--- a/drivers/staging/lustre/lustre/llite/llite_lib.c
+++ b/drivers/staging/lustre/lustre/llite/llite_lib.c
@@ -982,8 +982,7 @@ void ll_lli_init(struct ll_inode_info *lli)
spin_lock_init(&lli->lli_sa_lock);
lli->lli_opendir_pid = 0;
} else {
- sema_init(&lli->lli_size_sem, 1);
- lli->lli_size_sem_owner = NULL;
+ mutex_init(&lli->lli_size_mutex);
lli->lli_symlink_name = NULL;
init_rwsem(&lli->lli_trunc_sem);
mutex_init(&lli->lli_write_mutex);
@@ -1700,10 +1699,7 @@ void ll_inode_size_lock(struct inode *inode)
LASSERT(!S_ISDIR(inode->i_mode));
lli = ll_i2info(inode);
- LASSERT(lli->lli_size_sem_owner != current);
- down(&lli->lli_size_sem);
- LASSERT(lli->lli_size_sem_owner == NULL);
- lli->lli_size_sem_owner = current;
+ mutex_lock(&lli->lli_size_mutex);
}
void ll_inode_size_unlock(struct inode *inode)
@@ -1711,9 +1707,7 @@ void ll_inode_size_unlock(struct inode *inode)
struct ll_inode_info *lli;
lli = ll_i2info(inode);
- LASSERT(lli->lli_size_sem_owner == current);
- lli->lli_size_sem_owner = NULL;
- up(&lli->lli_size_sem);
+ mutex_unlock(&lli->lli_size_mutex);
}
void ll_update_inode(struct inode *inode, struct lustre_md *md)
--
1.8.5.3
next prev parent reply other threads:[~2014-04-27 17:08 UTC|newest]
Thread overview: 65+ messages / expand[flat|nested] mbox.gz Atom feed top
2014-04-27 17:06 [PATCH 00/47] Lustre fixes and cleanups Oleg Drokin
2014-04-27 17:06 ` [PATCH 01/47] staging/lustre/ptlrpc: Fix assertion failure of null_alloc_rs() Oleg Drokin
2014-04-27 17:06 ` [PATCH 02/47] staging/lustre/ptlrpc: Remove log message about export timer update Oleg Drokin
2014-04-27 17:06 ` [PATCH 03/47] staging/lustre/gss: gssnull security flavor Oleg Drokin
2014-04-27 17:06 ` [PATCH 04/47] staging/lustre/gss: Shared key mechanism & flavors Oleg Drokin
2014-04-27 17:20 ` Greg Kroah-Hartman
2014-04-27 17:06 ` [PATCH 05/47] staging/lustre/osc: don't activate deactivated obd_import Oleg Drokin
2014-04-27 17:06 ` [PATCH 06/47] staging/lustre/lnet: Dropped messages are not accounted correctly Oleg Drokin
2014-04-27 17:06 ` [PATCH 07/47] staging/lustre/ldlm: Hold lock when clearing flag Oleg Drokin
2014-04-27 17:06 ` [PATCH 08/47] staging/lustre/clio: clear nowait flag agl lock re-enqueue Oleg Drokin
2014-04-27 17:06 ` [PATCH 09/47] staging/lustre/ptlrpc: don't try to recover no_recov connection Oleg Drokin
2014-04-27 17:06 ` [PATCH 10/47] staging/lustre/gss: fix few issues found by Klocwork Insight tool Oleg Drokin
2014-04-27 17:06 ` [PATCH 11/47] staging/lustre/ptlrpc: add rpc_cache Oleg Drokin
2014-04-29 9:46 ` Dan Carpenter
2014-04-30 3:22 ` Oleg Drokin
2014-04-27 17:06 ` [PATCH 12/47] staging/lustre: restore __GFP_WAIT flag to memalloc calls Oleg Drokin
2014-04-27 17:06 ` [PATCH 13/47] staging/lustre/gss: fix uninitialized variable Oleg Drokin
2014-04-27 17:06 ` [PATCH 14/47] staging/lustre: quiet console permission error messages Oleg Drokin
2014-04-27 17:06 ` [PATCH 15/47] staging/lustre/lov: remove unused lov llog code Oleg Drokin
2014-04-27 17:06 ` [PATCH 16/47] staging/lustre/obdclass: remove uses of lov_stripe_md Oleg Drokin
2014-04-27 17:06 ` [PATCH 17/47] staging/lustre/hsm: count NULL terminator in hai_zero/hal_size Oleg Drokin
2014-04-27 17:06 ` [PATCH 18/47] staging/lustre/hsm: HSM requests not delivered Oleg Drokin
2014-04-29 9:08 ` Dan Carpenter
2014-04-30 3:31 ` Oleg Drokin
2014-04-27 17:06 ` [PATCH 19/47] staging/lustre: fix permission problem of setfacl Oleg Drokin
2014-04-27 17:06 ` [PATCH 20/47] staging/lustre/llite: issue OST_SYNC for fsync() Oleg Drokin
2014-04-27 17:06 ` [PATCH 21/47] staging/lustre/llite: deadlock taking lli_trunc_sem during file write Oleg Drokin
2014-04-27 17:06 ` [PATCH 22/47] staging/lustre/lov: to not hold sub locks at initialization Oleg Drokin
2014-04-27 17:06 ` [PATCH 23/47] staging/lustre: Limit reply buffer size Oleg Drokin
2014-04-27 17:06 ` [PATCH 24/47] staging/lustre/llite: Avoid statahead thread start/stop deadlocks Oleg Drokin
2014-04-27 17:06 ` [PATCH 25/47] stagaing/lustre: Improve statahead debug messages Oleg Drokin
2014-04-27 17:06 ` [PATCH 26/47] staging/lustre/llite: access layout version under a lock Oleg Drokin
2014-04-27 17:06 ` [PATCH 27/47] staging/lustre: shrink lu_object_header by 8 bytes on x86_64 Oleg Drokin
2014-04-27 17:06 ` [PATCH 28/47] staging/lustre/ldlm: fix NULL pointer dereference Oleg Drokin
2014-04-27 17:06 ` [PATCH 29/47] staging/lustre/lnet: lnet: fix issues found by Klocwork Insight tool Oleg Drokin
2014-04-27 17:25 ` Greg Kroah-Hartman
2014-04-27 17:06 ` [PATCH 30/47] staging/lustre/mdc: fix issue " Oleg Drokin
2014-04-29 10:20 ` Dan Carpenter
2014-04-27 17:06 ` [PATCH 31/47] staging/lustre/libcfs: fix issues " Oleg Drokin
2014-04-27 17:06 ` [PATCH 32/47] staging/lustre/lnet: NI shutdown may loop forever Oleg Drokin
2014-04-27 17:06 ` [PATCH 33/47] staging/lustre: remove lustre/include/ioctl.h Oleg Drokin
2014-04-27 17:06 ` [PATCH 34/47] staging/lustre/libcfs: add CPU table functions for uniprocessor Oleg Drokin
2014-04-29 10:35 ` Dan Carpenter
2014-04-27 17:06 ` [PATCH 35/47] staging/lustre: replace semaphores with mutexes Oleg Drokin
2014-04-27 17:07 ` Oleg Drokin [this message]
2014-04-27 17:07 ` [PATCH 37/47] staging/lustre/llite: Do not rate limit dirty page discard warning Oleg Drokin
2014-04-27 17:07 ` [PATCH 38/47] staging/lustre/lloop: avoid panic during blockdev_info Oleg Drokin
2014-04-27 17:07 ` [PATCH 39/47] staging/lustre/clio: Solve a race in cl_lock_put Oleg Drokin
2014-04-27 17:07 ` [PATCH 40/47] staging/lustre/mdc: use cl_max_mds_md to pack getattr RPC Oleg Drokin
2014-04-27 17:07 ` [PATCH 41/47] staging/lustre/llite: remove dead code Oleg Drokin
2014-04-29 11:02 ` Dan Carpenter
2014-04-29 19:16 ` Hammond, John
2014-04-29 20:17 ` Dan Carpenter
2014-04-30 3:21 ` Oleg Drokin
2014-04-30 8:01 ` Dan Carpenter
2014-04-29 11:12 ` Richard Weinberger
2014-04-27 17:07 ` [PATCH 42/47] staging/lustre: remove assertion of spin_is_locked() Oleg Drokin
2014-04-27 17:07 ` [PATCH 43/47] staging/lustre/osc: Update inode timestamp for lockless IO as well Oleg Drokin
2014-04-27 17:07 ` [PATCH 44/47] staging/lustre: Always clamp cdls_delay between min and max Oleg Drokin
2014-04-27 17:07 ` [PATCH 45/47] staging/lustre: pass fsync() range through RPC/IO stack Oleg Drokin
2014-04-27 17:07 ` [PATCH 46/47] staging/lustre: Fix unsafe userspace access in many proc files Oleg Drokin
2014-04-27 17:30 ` Greg Kroah-Hartman
2014-04-27 17:07 ` [PATCH 47/47] staging/lustre/llite: prevent buffer overflow in fiemap Oleg Drokin
2014-04-27 17:33 ` [PATCH 00/47] Lustre fixes and cleanups Greg Kroah-Hartman
2014-04-27 18:28 ` Oleg Drokin
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1398618431-29757-37-git-send-email-green@linuxhacker.ru \
--to=green@linuxhacker.ru \
--cc=devel@driverdev.osuosl.org \
--cc=dmitry.eremin@intel.com \
--cc=gregkh@linuxfoundation.org \
--cc=linux-kernel@vger.kernel.org \
--cc=oleg.drokin@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).