reiserfs_write_end() is a hot path in reiserfs.
We have two wasteful write lock lock/release inside that can be gathered
without changing the code logic.
This patch factorizes them out in a single protected section, reducing the
number of contentions inside.
[ Impact: reduce lock contention in a reiserfs hotpath ]
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Alexander Beregalov <a.beregalov@gmail.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
search_by_key() is a central function in reiserfs which searches
the patch in the fs tree from the root to a node given its key.
It is the function that is most requesting the write lock
because it's a path very often used.
Also we forget to release the lock while reading the next tree node,
making us holding the lock in a wasteful way.
Then we release the lock while reading the current node and its childs,
all-in-one. It should be safe because we have a reference to these
blocks and even if we read a block that will be concurrently changed,
we have an fs_changed check later that will make us retry the path from
the root.
[ Impact: release the write lock while unused in a hot path ]
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Alexander Beregalov <a.beregalov@gmail.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
The write lock can be acquired recursively in reiserfs_lookup(). But we may
want to *really* release the lock before possible rescheduling from a
reiserfs_lookup() callee.
Hence we want to only acquire the lock once (ie: not recursively).
[ Impact: prevent from possible false unreleased write lock on sleeping ]
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Alexander Beregalov <a.beregalov@gmail.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
reiserfs_get_block() is one of these sites where the write lock might
be acquired recursively.
It's a particular problem because this function is called very often.
It's a hot spot which needs to reschedule() periodically while converting
direct items to indirect ones because it can take some time.
Then if we are applying the write lock release/reacquire pattern on
schedule() here, it may not produce the desired effect since we may have
locked in more than one depth.
The solution is to use reiserfs_write_lock_once() which won't try
to reacquire the lock recursively. Then the lock will be *really*
released before schedule().
Also, we only release the lock if TIF_NEED_RESCHED is set to not
create wasteful numerous contentions.
[ Impact: fix a too long holded lock case in reiserfs_get_block() ]
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Alexander Beregalov <a.beregalov@gmail.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
The goal of fs_changed() is to check whether the tree changed during a
schedule(). This is a BKL legacy.
A recent patch added an explicit unconditional release/reacquire of the
write lock around the cond_resched() called inside fs_changed.
But it's wasteful to unconditionally do that, we are creating superfluous
lock contention in !TIF_NEED_RESCHED case.
This patch manage that by calling reiserfs_cond_resched() from fs_changed()
which only releases the lock if we are going to reschedule.
[ Impact: inject less lock contention and tree job retries ]
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Alexander Beregalov <a.beregalov@gmail.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Usually, when we call cond_resched(), we want the write lock
to be released and then reacquired once we return from scheduling.
Not only does it follow the previous bkl based locking scheme, but
it also let other waiters to get the lock.
But if we aren't going to reschedule(), such as in !TIF_NEED_RESCHED
case, it's useless to release the lock. Worse, if we release and reacquire
the lock whereas it is not needed, we create useless contentions. Also
if someone takes the lock while we are modifying or reading the tree,
there are good chances we'll have to retry our operation, eg if the
block we were seeeking has moved.
So this patch introduces a helper which only unlock the write lock
if we are going to schedule.
[ Impact: prepare to inject less lock contention and less tree operation attempts ]
Reported-by: Andi Kleen <andi@firstfloor.org>
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Alexander Beregalov <a.beregalov@gmail.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
flush_commit_list() uses ll_rw_block() to commit the pending log blocks.
ll_rw_block() might sleep, and the bkl was released at this point. Then
we can also relax the write lock at this point.
[ Impact: release the reiserfs write lock when it is not needed ]
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Alexander Beregalov <a.beregalov@gmail.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
reiserfs_read_bitmap_block() uses sb_bread() to read the bitmap block. This
helper might sleep.
Then, when the bkl was used, it was released at this point. We can then
relax the write lock too here.
[ Impact: release the reiserfs write lock when it is not needed ]
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Alexander Beregalov <a.beregalov@gmail.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
get_neighbors() is used to get the left and/or right blocks
against a given one in order to balance a tree.
sb_bread() is used to read the buffer of these neighors blocks and
while it waits for this operation, it might sleep.
The bkl was released at this point, and then we can also release
the write lock before calling sb_bread().
This is safe because if the filesystem is changed after this
lock release, the function returns REPEAT_SEARCH (aka SCHEDULE_OCCURRED
in the function header comments) in order to repeat the neighbhor
research.
[ Impact: release the reiserfs write lock when it is not needed ]
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Alexander Beregalov <a.beregalov@gmail.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
prepare_for_delete_or_cut() can process several types of items, including
indirect items, ie: items which contain no file data but pointers to
unformatted nodes scattering the datas of a file.
In this case it has to zero out these pointers to block numbers of
unformatted nodes and release the bitmap from these block numbers.
It can take some time, so a rescheduling() is performed between each
block processed. We can safely release the write lock while
rescheduling(), like the bkl did, because the code checks just after
if the item has moved after sleeping.
[ Impact: release the reiserfs write lock when it is not needed ]
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Alexander Beregalov <a.beregalov@gmail.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
When do_journal_end() copies data to the journal blocks buffers in memory,
it reschedules if needed between each block copied and dirtyfied.
We can also release the write lock at this rescheduling stage,
like did the bkl implicitly.
[ Impact: release the reiserfs write lock when it is not needed ]
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Alexander Beregalov <a.beregalov@gmail.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
fs_changed() is a macro used by reiserfs to check whether its tree has been
rebalanced. It has been designed to check parallel changes on the tree after
calling a sleeping function, which released the Bkl.
fs_changed() also calls cond_resched(), so that if rescheduling is needed,
we are in the best place to do that, since we check if the tree has changed
just after (because of the bkl release on schedule()).
Even if we are not anymore using the Bkl, we still want to release the lock
while we reschedule, so that other waiters for the lock can acquire it safely,
because of the following __fs_changed() check.
[ Impact: release the reiserfs write lock when it is not needed ]
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Alexander Beregalov <a.beregalov@gmail.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Impact: fix a deadlock
reiserfs_dirty_inode() is the super_operations::dirty_inode() callback
of reiserfs. It can be called from different contexts where the write
lock can be already held.
But this function also grab the write lock (possibly recursively).
Subsequent release of the lock before sleep will actually not release
the lock if the caller of mark_inode_dirty() (which in turn calls
reiserfs_dirty_inode()) already owns the lock.
A typical case:
reiserfs_write_end() {
acquire_write_lock()
mark_inode_dirty() {
reiserfs_dirty_inode() {
reacquire_write_lock() {
journal_begin() {
do_journal_begin_r() {
/*
* fail to release, still
* one depth of lock
*/
release_write_lock()
reiserfs_wait_on_write_block() {
wait_event()
The event is usually provided by something which needs the write lock but
it hasn't been released.
We use reiserfs_write_lock_once() here to ensure we only grab the
write lock in one level.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Alessio Igor Bogani <abogani@texware.it>
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Chris Mason <chris.mason@oracle.com>
LKML-Reference: <1239680065-25013-4-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: fix a deadlock
reiserfs_truncate_file() can be called from multiple context where
the write lock can be already hold or not.
This function also acquire (possibly recursively) the write
lock. Subsequent releases before sleeping will not actually release
the lock because we may be in more than one lock depth degree.
A typical case is:
reiserfs_file_release {
acquire_the_lock()
reiserfs_truncate_file()
reacquire_the_lock()
journal_begin() {
do_journal_begin_r() {
reiserfs_wait_on_write_block() {
/*
* Not released because still one
* depth owned
*/
release_lock()
wait_for_event()
At this stage the event never happen because the one which provides
it needs the write lock.
We use reiserfs_write_lock_once() here to ensure that we don't acquire the
write lock recursively.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Alessio Igor Bogani <abogani@texware.it>
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Alexander Beregalov <a.beregalov@gmail.com>
Cc: Chris Mason <chris.mason@oracle.com>
LKML-Reference: <1239680065-25013-3-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Sometimes we don't want to recursively hold the per superblock write
lock because we want to be sure it is actually released when we come
to sleep.
This patch introduces the necessary tools for that.
reiserfs_write_lock_once() does the same job than reiserfs_write_lock()
except that it won't try to acquire recursively the lock if the current
task already owns it. Also the lock_depth before the call of this function
is returned.
reiserfs_write_unlock_once() unlock only if reiserfs_write_lock_once()
returned a depth equal to -1, ie: only if it actually locked.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Alessio Igor Bogani <abogani@texware.it>
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Alexander Beregalov <a.beregalov@gmail.com>
Cc: Chris Mason <chris.mason@oracle.com>
LKML-Reference: <1239680065-25013-2-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: fix a deadlock
The j_flush_mutex is acquired safely in journal.c:
if we can't take it, we free the reiserfs per superblock lock
and wait a bit.
But we have a remaining place in kupdate_transactions() where
j_flush_mutex is still acquired traditionnaly. Thus the following
scenario (warned by lockdep) can happen:
A B
mutex_lock(&write_lock) mutex_lock(&write_lock)
mutex_lock(&j_flush_mutex) mutex_lock(&j_flush_mutex) //block
mutex_unlock(&write_lock)
sleep...
mutex_lock(&write_lock) //deadlock
Fix this by using reiserfs_mutex_lock_safe() in kupdate_transactions().
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Alessio Igor Bogani <abogani@texware.it>
Cc: Jeff Mahoney <jeffm@suse.com>
LKML-Reference: <1239660635-12940-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch is an attempt to remove the Bkl based locking scheme from
reiserfs and is intended.
It is a bit inspired from an old attempt by Peter Zijlstra:
http://lkml.indiana.edu/hypermail/linux/kernel/0704.2/2174.html
The bkl is heavily used in this filesystem to prevent from
concurrent write accesses on the filesystem.
Reiserfs makes a deep use of the specific properties of the Bkl:
- It can be acqquired recursively by a same task
- It is released on the schedule() calls and reacquired when schedule() returns
The two properties above are a roadmap for the reiserfs write locking so it's
very hard to simply replace it with a common mutex.
- We need a recursive-able locking unless we want to restructure several blocks
of the code.
- We need to identify the sites where the bkl was implictly relaxed
(schedule, wait, sync, etc...) so that we can in turn release and
reacquire our new lock explicitly.
Such implicit releases of the lock are often required to let other
resources producer/consumer do their job or we can suffer unexpected
starvations or deadlocks.
So the new lock that replaces the bkl here is a per superblock mutex with a
specific property: it can be acquired recursively by a same task, like the
bkl.
For such purpose, we integrate a lock owner and a lock depth field on the
superblock information structure.
The first axis on this patch is to turn reiserfs_write_(un)lock() function
into a wrapper to manage this mutex. Also some explicit calls to
lock_kernel() have been converted to reiserfs_write_lock() helpers.
The second axis is to find the important blocking sites (schedule...(),
wait_on_buffer(), sync_dirty_buffer(), etc...) and then apply an explicit
release of the write lock on these locations before blocking. Then we can
safely wait for those who can give us resources or those who need some.
Typically this is a fight between the current writer, the reiserfs workqueue
(aka the async commiter) and the pdflush threads.
The third axis is a consequence of the second. The write lock is usually
on top of a lock dependency chain which can include the journal lock, the
flush lock or the commit lock. So it's dangerous to release and trying to
reacquire the write lock while we still hold other locks.
This is fine with the bkl:
T1 T2
lock_kernel()
mutex_lock(A)
unlock_kernel()
// do something
lock_kernel()
mutex_lock(A) -> already locked by T1
schedule() (and then unlock_kernel())
lock_kernel()
mutex_unlock(A)
....
This is not fine with a mutex:
T1 T2
mutex_lock(write)
mutex_lock(A)
mutex_unlock(write)
// do something
mutex_lock(write)
mutex_lock(A) -> already locked by T1
schedule()
mutex_lock(write) -> already locked by T2
deadlock
The solution in this patch is to provide a helper which releases the write
lock and sleep a bit if we can't lock a mutex that depend on it. It's another
simulation of the bkl behaviour.
The last axis is to locate the fs callbacks that are called with the bkl held,
according to Documentation/filesystem/Locking.
Those are:
- reiserfs_remount
- reiserfs_fill_super
- reiserfs_put_super
Reiserfs didn't need to explicitly lock because of the context of these callbacks.
But now we must take care of that with the new locking.
After this patch, reiserfs suffers from a slight performance regression (for now).
On UP, a high volume write with dd reports an average of 27 MB/s instead
of 30 MB/s without the patch applied.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Reviewed-by: Ingo Molnar <mingo@elte.hu>
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Bron Gondwana <brong@fastmail.fm>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
LKML-Reference: <1239070789-13354-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Andy Whitcroft reported an oops in aoe triggered by use of an
incorrectly initialised request_queue object:
[ 2645.959090] kobject '<NULL>' (ffff880059ca22c0): tried to add
an uninitialized object, something is seriously wrong.
[ 2645.959104] Pid: 6, comm: events/0 Not tainted 2.6.31-5-generic #24-Ubuntu
[ 2645.959107] Call Trace:
[ 2645.959139] [<ffffffff8126ca2f>] kobject_add+0x5f/0x70
[ 2645.959151] [<ffffffff8125b4ab>] blk_register_queue+0x8b/0xf0
[ 2645.959155] [<ffffffff8126043f>] add_disk+0x8f/0x160
[ 2645.959161] [<ffffffffa01673c4>] aoeblk_gdalloc+0x164/0x1c0 [aoe]
The request queue of an aoe device is not used but can be allocated in
code that does not sleep.
Bruno bisected this regression down to
cd43e26f07
block: Expose stacked device queues in sysfs
"This seems to generate /sys/block/$device/queue and its contents for
everyone who is using queues, not just for those queues that have a
non-NULL queue->request_fn."
Addresses http://bugs.launchpad.net/bugs/410198
Addresses http://bugzilla.kernel.org/show_bug.cgi?id=13942
Note that embedding a queue inside another object has always been
an illegal construct, since the queues are reference counted and
must persist until the last reference is dropped. So aoe was
always buggy in this respect (Jens).
Signed-off-by: Ed Cashin <ecashin@coraid.com>
Cc: Andy Whitcroft <apw@canonical.com>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Bruno Premont <bonbons@linux-vserver.org>
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Reinette Chatre reports a frozen system (with blinking keyboard LEDs)
when switching from graphics mode to the text console, or when
suspending (which does the same thing). With netconsole, the oops
turned out to be
BUG: unable to handle kernel NULL pointer dereference at 0000000000000084
IP: [<ffffffffa03ecaab>] i915_driver_irq_handler+0x26b/0xd20 [i915]
and it's due to the i915_gem.c code doing drm_irq_uninstall() after
having done i915_gem_idle(). And the i915_gem_idle() path will do
i915_gem_idle() ->
i915_gem_cleanup_ringbuffer() ->
i915_gem_cleanup_hws() ->
dev_priv->hw_status_page = NULL;
but if an i915 interrupt comes in after this stage, it may want to
access that hw_status_page, and gets the above NULL pointer dereference.
And since the NULL pointer dereference happens from within an interrupt,
and with the screen still in graphics mode, the common end result is
simply a silently hung machine.
Fix it by simply uninstalling the irq handler before idling rather than
after. Fixes
http://bugzilla.kernel.org/show_bug.cgi?id=13819
Reported-and-tested-by: Reinette Chatre <reinette.chatre@intel.com>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Frederic Weisbecker
Linus Torvalds
Ed Cashin
Zhenyu Wang