jbd2_clear_buffer_revoked_flags() takes a long time

jbd2_clear_buffer_revoked_flags() takes a long time

[Adrian Hunter]

Hi

I have a case on a v4.14 kernel where the EXT4 journal commit disables
preemption for 30ms due to jbd2_clear_buffer_revoked_flags().  That in turn
disables preemption on other CPUs as they come to spin waiting for the same
lock.  The side-effect of that is that it periodically blocks high priority
tasks from running.

I see jbd2_clear_buffer_revoked_flags() iterating 32768 times calling
__find_get_block().

Is there any way to make jbd2_clear_buffer_revoked_flags() take less time,
or move its work out from under write_lock(&journal->j_state_lock)?

Or do you have any other suggestions?

Regards
Adrian

Re: jbd2_clear_buffer_revoked_flags() takes a long time

[Theodore Y. Ts'o]

On Wed, Oct 10, 2018 at 04:43:27PM +0300, Adrian Hunter wrote:
> Hi
> 
> I have a case on a v4.14 kernel where the EXT4 journal commit disables
> preemption for 30ms due to jbd2_clear_buffer_revoked_flags().  That in turn
> disables preemption on other CPUs as they come to spin waiting for the same
> lock.  The side-effect of that is that it periodically blocks high priority
> tasks from running.
> 
> I see jbd2_clear_buffer_revoked_flags() iterating 32768 times calling
> __find_get_block().
> 
> Is there any way to make jbd2_clear_buffer_revoked_flags() take less time,
> or move its work out from under write_lock(&journal->j_state_lock)?

Hmm.... I'd have to look a bit more carefully and then run some tests,
but I *think* we can drop the j_state_lock at the beginning of JBD2
commit phase 1, and then grab it again right before we set
commit_transaction->t_state to T_FLUSH.

That should be safe because while the transaction state is T_LOCKED,
we can't start any new handles, so there can't be any new blocks added
to the revoke table.

Can you give that a try and see whether that solves your priority
inversion problem?

Cheers,

						- Ted

Re: jbd2_clear_buffer_revoked_flags() takes a long time

[Jan Kara]

[-- Attachment #1: Type: text/plain, Size: 1357 bytes --]

On Wed 10-10-18 13:49:34, Theodore Y. Ts'o wrote:
> On Wed, Oct 10, 2018 at 04:43:27PM +0300, Adrian Hunter wrote:
> > Hi
> > 
> > I have a case on a v4.14 kernel where the EXT4 journal commit disables
> > preemption for 30ms due to jbd2_clear_buffer_revoked_flags().  That in turn
> > disables preemption on other CPUs as they come to spin waiting for the same
> > lock.  The side-effect of that is that it periodically blocks high priority
> > tasks from running.
> > 
> > I see jbd2_clear_buffer_revoked_flags() iterating 32768 times calling
> > __find_get_block().
> > 
> > Is there any way to make jbd2_clear_buffer_revoked_flags() take less time,
> > or move its work out from under write_lock(&journal->j_state_lock)?
> 
> Hmm.... I'd have to look a bit more carefully and then run some tests,
> but I *think* we can drop the j_state_lock at the beginning of JBD2
> commit phase 1, and then grab it again right before we set
> commit_transaction->t_state to T_FLUSH.
> 
> That should be safe because while the transaction state is T_LOCKED,
> we can't start any new handles, so there can't be any new blocks added
> to the revoke table.
> 
> Can you give that a try and see whether that solves your priority
> inversion problem?

Agreed. Something like attached patch (compile-tested only)?

								Honza

-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

[-- Attachment #2: 0001-jbd2-Avoid-long-hold-times-of-j_state_lock-while-com.patch --]
[-- Type: text/x-patch, Size: 2286 bytes --]

>From 3627b5f30996504019cd84f326402fccbb9a298b Mon Sep 17 00:00:00 2001
From: Jan Kara <jack@suse.cz>
Date: Thu, 11 Oct 2018 13:04:44 +0200
Subject: [PATCH] jbd2: Avoid long hold times of j_state_lock while committing
 a transaction

We can hold j_state_lock for writing at the beginning of
jbd2_journal_commit_transaction() for a rather long time (reportedly for
30 ms) due cleaning revoke bits of all revoked buffers under it. The
handling of revoke tables as well as cleaning of t_reserved_list, and
checkpoint lists does not need j_state_lock for anything. Furthermore
the transaction is in T_LOCKED state and we waited for all outstanding
handles so nobody is going to be adding anything to the transaction.

Just drop the lock for unnecessary operations.

Reported-by: Adrian Hunter <adrian.hunter@intel.com>
Suggested-by: "Theodore Y. Ts'o" <tytso@mit.edu>
Signed-off-by: Jan Kara <jack@suse.cz>
---
 fs/jbd2/commit.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/fs/jbd2/commit.c b/fs/jbd2/commit.c
index 150cc030b4d7..356b75fa3101 100644
--- a/fs/jbd2/commit.c
+++ b/fs/jbd2/commit.c
@@ -422,6 +422,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
 				       stats.run.rs_locked);
 	stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start,
 					      stats.run.rs_locked);
+	write_unlock(&journal->j_state_lock);
 
 	spin_lock(&commit_transaction->t_handle_lock);
 	while (atomic_read(&commit_transaction->t_updates)) {
@@ -431,9 +432,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
 					TASK_UNINTERRUPTIBLE);
 		if (atomic_read(&commit_transaction->t_updates)) {
 			spin_unlock(&commit_transaction->t_handle_lock);
-			write_unlock(&journal->j_state_lock);
 			schedule();
-			write_lock(&journal->j_state_lock);
 			spin_lock(&commit_transaction->t_handle_lock);
 		}
 		finish_wait(&journal->j_wait_updates, &wait);
@@ -505,6 +504,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
 	atomic_sub(atomic_read(&journal->j_reserved_credits),
 		   &commit_transaction->t_outstanding_credits);
 
+	write_lock(&journal->j_state_lock);
 	trace_jbd2_commit_flushing(journal, commit_transaction);
 	stats.run.rs_flushing = jiffies;
 	stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked,
-- 
2.16.4

Re: jbd2_clear_buffer_revoked_flags() takes a long time

[Adrian Hunter]

On 11/10/18 2:12 PM, Jan Kara wrote:
> On Wed 10-10-18 13:49:34, Theodore Y. Ts'o wrote:
>> On Wed, Oct 10, 2018 at 04:43:27PM +0300, Adrian Hunter wrote:
>>> Hi
>>>
>>> I have a case on a v4.14 kernel where the EXT4 journal commit disables
>>> preemption for 30ms due to jbd2_clear_buffer_revoked_flags().  That in turn
>>> disables preemption on other CPUs as they come to spin waiting for the same
>>> lock.  The side-effect of that is that it periodically blocks high priority
>>> tasks from running.
>>>
>>> I see jbd2_clear_buffer_revoked_flags() iterating 32768 times calling
>>> __find_get_block().
>>>
>>> Is there any way to make jbd2_clear_buffer_revoked_flags() take less time,
>>> or move its work out from under write_lock(&journal->j_state_lock)?
>> Hmm.... I'd have to look a bit more carefully and then run some tests,
>> but I *think* we can drop the j_state_lock at the beginning of JBD2
>> commit phase 1, and then grab it again right before we set
>> commit_transaction->t_state to T_FLUSH.
>>
>> That should be safe because while the transaction state is T_LOCKED,
>> we can't start any new handles, so there can't be any new blocks added
>> to the revoke table.
>>
>> Can you give that a try and see whether that solves your priority
>> inversion problem?
> Agreed. Something like attached patch (compile-tested only)?

I have been testing a patch with the unlock/lock at slightly different
positions, and it definitely helps.  The incidence of my problem drops from
nearly every writeback, to a few an hour.  I haven't had time to find out
what is causing the remaining cases yet - it may not be related to EXT4.  I
should be able to test this patch tomorrow.

> 
> 								Honza
> 
> -- Jan Kara <jack@suse.com> SUSE Labs, CR
> 
> 
> 0001-jbd2-Avoid-long-hold-times-of-j_state_lock-while-com.patch
> 
>>From 3627b5f30996504019cd84f326402fccbb9a298b Mon Sep 17 00:00:00 2001
> From: Jan Kara <jack@suse.cz>
> Date: Thu, 11 Oct 2018 13:04:44 +0200
> Subject: [PATCH] jbd2: Avoid long hold times of j_state_lock while committing
>  a transaction
> 
> We can hold j_state_lock for writing at the beginning of
> jbd2_journal_commit_transaction() for a rather long time (reportedly for
> 30 ms) due cleaning revoke bits of all revoked buffers under it. The
> handling of revoke tables as well as cleaning of t_reserved_list, and
> checkpoint lists does not need j_state_lock for anything. Furthermore
> the transaction is in T_LOCKED state and we waited for all outstanding
> handles so nobody is going to be adding anything to the transaction.
> 
> Just drop the lock for unnecessary operations.
> 
> Reported-by: Adrian Hunter <adrian.hunter@intel.com>
> Suggested-by: "Theodore Y. Ts'o" <tytso@mit.edu>
> Signed-off-by: Jan Kara <jack@suse.cz>
> ---
>  fs/jbd2/commit.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/jbd2/commit.c b/fs/jbd2/commit.c
> index 150cc030b4d7..356b75fa3101 100644
> --- a/fs/jbd2/commit.c
> +++ b/fs/jbd2/commit.c
> @@ -422,6 +422,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
>  				       stats.run.rs_locked);
>  	stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start,
>  					      stats.run.rs_locked);
> +	write_unlock(&journal->j_state_lock);
>  
>  	spin_lock(&commit_transaction->t_handle_lock);
>  	while (atomic_read(&commit_transaction->t_updates)) {
> @@ -431,9 +432,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
>  					TASK_UNINTERRUPTIBLE);
>  		if (atomic_read(&commit_transaction->t_updates)) {
>  			spin_unlock(&commit_transaction->t_handle_lock);
> -			write_unlock(&journal->j_state_lock);
>  			schedule();
> -			write_lock(&journal->j_state_lock);
>  			spin_lock(&commit_transaction->t_handle_lock);
>  		}
>  		finish_wait(&journal->j_wait_updates, &wait);
> @@ -505,6 +504,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
>  	atomic_sub(atomic_read(&journal->j_reserved_credits),
>  		   &commit_transaction->t_outstanding_credits);
>  
> +	write_lock(&journal->j_state_lock);
>  	trace_jbd2_commit_flushing(journal, commit_transaction);
>  	stats.run.rs_flushing = jiffies;
>  	stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked,
> -- 2.16.4
> 

Re: jbd2_clear_buffer_revoked_flags() takes a long time

[Adrian Hunter]

On 11/10/18 3:38 PM, Adrian Hunter wrote:
> On 11/10/18 2:12 PM, Jan Kara wrote:
>> On Wed 10-10-18 13:49:34, Theodore Y. Ts'o wrote:
>>> On Wed, Oct 10, 2018 at 04:43:27PM +0300, Adrian Hunter wrote:
>>>> Hi
>>>>
>>>> I have a case on a v4.14 kernel where the EXT4 journal commit disables
>>>> preemption for 30ms due to jbd2_clear_buffer_revoked_flags().  That in turn
>>>> disables preemption on other CPUs as they come to spin waiting for the same
>>>> lock.  The side-effect of that is that it periodically blocks high priority
>>>> tasks from running.
>>>>
>>>> I see jbd2_clear_buffer_revoked_flags() iterating 32768 times calling
>>>> __find_get_block().
>>>>
>>>> Is there any way to make jbd2_clear_buffer_revoked_flags() take less time,
>>>> or move its work out from under write_lock(&journal->j_state_lock)?
>>> Hmm.... I'd have to look a bit more carefully and then run some tests,
>>> but I *think* we can drop the j_state_lock at the beginning of JBD2
>>> commit phase 1, and then grab it again right before we set
>>> commit_transaction->t_state to T_FLUSH.
>>>
>>> That should be safe because while the transaction state is T_LOCKED,
>>> we can't start any new handles, so there can't be any new blocks added
>>> to the revoke table.
>>>
>>> Can you give that a try and see whether that solves your priority
>>> inversion problem?
>> Agreed. Something like attached patch (compile-tested only)?
> 
> I have been testing a patch with the unlock/lock at slightly different
> positions, and it definitely helps.  The incidence of my problem drops from
> nearly every writeback, to a few an hour.  I haven't had time to find out
> what is causing the remaining cases yet - it may not be related to EXT4.  I
> should be able to test this patch tomorrow.

Thanks very much for the quick response and patch!

Tested-by: Adrian Hunter <adrian.hunter@intel.com>

With more stress I also found move_expired_inodes()
(wb_writeback->queue_io->move_expired_inodes) to take up to 16ms using
230,000 branches while under spin lock.  AFAICT we weren't hitting that in
practice so I am not following it up at this stage.

> 
>>
>> 								Honza
>>
>> -- Jan Kara <jack@suse.com> SUSE Labs, CR
>>
>>
>> 0001-jbd2-Avoid-long-hold-times-of-j_state_lock-while-com.patch
>>
>> >From 3627b5f30996504019cd84f326402fccbb9a298b Mon Sep 17 00:00:00 2001
>> From: Jan Kara <jack@suse.cz>
>> Date: Thu, 11 Oct 2018 13:04:44 +0200
>> Subject: [PATCH] jbd2: Avoid long hold times of j_state_lock while committing
>>  a transaction
>>
>> We can hold j_state_lock for writing at the beginning of
>> jbd2_journal_commit_transaction() for a rather long time (reportedly for
>> 30 ms) due cleaning revoke bits of all revoked buffers under it. The
>> handling of revoke tables as well as cleaning of t_reserved_list, and
>> checkpoint lists does not need j_state_lock for anything. Furthermore
>> the transaction is in T_LOCKED state and we waited for all outstanding
>> handles so nobody is going to be adding anything to the transaction.
>>
>> Just drop the lock for unnecessary operations.
>>
>> Reported-by: Adrian Hunter <adrian.hunter@intel.com>
>> Suggested-by: "Theodore Y. Ts'o" <tytso@mit.edu>
>> Signed-off-by: Jan Kara <jack@suse.cz>
>> ---
>>  fs/jbd2/commit.c | 4 ++--
>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/fs/jbd2/commit.c b/fs/jbd2/commit.c
>> index 150cc030b4d7..356b75fa3101 100644
>> --- a/fs/jbd2/commit.c
>> +++ b/fs/jbd2/commit.c
>> @@ -422,6 +422,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
>>  				       stats.run.rs_locked);
>>  	stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start,
>>  					      stats.run.rs_locked);
>> +	write_unlock(&journal->j_state_lock);
>>  
>>  	spin_lock(&commit_transaction->t_handle_lock);
>>  	while (atomic_read(&commit_transaction->t_updates)) {
>> @@ -431,9 +432,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
>>  					TASK_UNINTERRUPTIBLE);
>>  		if (atomic_read(&commit_transaction->t_updates)) {
>>  			spin_unlock(&commit_transaction->t_handle_lock);
>> -			write_unlock(&journal->j_state_lock);
>>  			schedule();
>> -			write_lock(&journal->j_state_lock);
>>  			spin_lock(&commit_transaction->t_handle_lock);
>>  		}
>>  		finish_wait(&journal->j_wait_updates, &wait);
>> @@ -505,6 +504,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
>>  	atomic_sub(atomic_read(&journal->j_reserved_credits),
>>  		   &commit_transaction->t_outstanding_credits);
>>  
>> +	write_lock(&journal->j_state_lock);
>>  	trace_jbd2_commit_flushing(journal, commit_transaction);
>>  	stats.run.rs_flushing = jiffies;
>>  	stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked,
>> -- 2.16.4
>>
> 
> 

Re: jbd2_clear_buffer_revoked_flags() takes a long time

[Jan Kara]

On Tue 16-10-18 11:49:45, Adrian Hunter wrote:
> On 11/10/18 3:38 PM, Adrian Hunter wrote:
> > On 11/10/18 2:12 PM, Jan Kara wrote:
> >> On Wed 10-10-18 13:49:34, Theodore Y. Ts'o wrote:
> >>> On Wed, Oct 10, 2018 at 04:43:27PM +0300, Adrian Hunter wrote:
> >>>> Hi
> >>>>
> >>>> I have a case on a v4.14 kernel where the EXT4 journal commit disables
> >>>> preemption for 30ms due to jbd2_clear_buffer_revoked_flags().  That in turn
> >>>> disables preemption on other CPUs as they come to spin waiting for the same
> >>>> lock.  The side-effect of that is that it periodically blocks high priority
> >>>> tasks from running.
> >>>>
> >>>> I see jbd2_clear_buffer_revoked_flags() iterating 32768 times calling
> >>>> __find_get_block().
> >>>>
> >>>> Is there any way to make jbd2_clear_buffer_revoked_flags() take less time,
> >>>> or move its work out from under write_lock(&journal->j_state_lock)?
> >>> Hmm.... I'd have to look a bit more carefully and then run some tests,
> >>> but I *think* we can drop the j_state_lock at the beginning of JBD2
> >>> commit phase 1, and then grab it again right before we set
> >>> commit_transaction->t_state to T_FLUSH.
> >>>
> >>> That should be safe because while the transaction state is T_LOCKED,
> >>> we can't start any new handles, so there can't be any new blocks added
> >>> to the revoke table.
> >>>
> >>> Can you give that a try and see whether that solves your priority
> >>> inversion problem?
> >> Agreed. Something like attached patch (compile-tested only)?
> > 
> > I have been testing a patch with the unlock/lock at slightly different
> > positions, and it definitely helps.  The incidence of my problem drops from
> > nearly every writeback, to a few an hour.  I haven't had time to find out
> > what is causing the remaining cases yet - it may not be related to EXT4.  I
> > should be able to test this patch tomorrow.
> 
> Thanks very much for the quick response and patch!
> 
> Tested-by: Adrian Hunter <adrian.hunter@intel.com>

Thanks. I've officially posted the patch.

> With more stress I also found move_expired_inodes()
> (wb_writeback->queue_io->move_expired_inodes) to take up to 16ms using
> 230,000 branches while under spin lock.  AFAICT we weren't hitting that in
> practice so I am not following it up at this stage.

Interesting. I actually have a patch simplifying that area as well sitting
in some branch in my tree. So I can dust it off if you are interested.

								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR