cpu scaling fixes

cpu scaling fixes

[Christoph Hellwig]

Hi all,

this series fully disables the mkfs automatic CPU count scaling for two
tests, where that would lead to differences in the gold output.

[PATCH 1/2] xfs/078: disable all concurrency scaling

[Christoph Hellwig]

Disable concurrency scaling to avoid mismatching output on systems with
a large CPU count.

Reported-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 tests/xfs/078 | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/tests/xfs/078 b/tests/xfs/078
index 6057aeea12ab..8ef52614b125 100755
--- a/tests/xfs/078
+++ b/tests/xfs/078
@@ -57,9 +57,11 @@ _grow_loop()
 	$XFS_IO_PROG -f -c "truncate $original" $LOOP_IMG
 	loop_dev=`_create_loop_device $LOOP_IMG $bsize`
 
-	dparam=""
+	# Disable concurrency scaling to avoid output differences due to large
+	# CPU counts.
+	mkfs_opts="-d concurrency=0 -l concurrency=0 -r concurrency=0"
 	if [ -n "$agsize" ]; then
-		dparam="-d agsize=$agsize"
+		mkfs_opts="$mkfs_opts -d agsize=$agsize"
 	fi
 
 	echo
@@ -67,7 +69,7 @@ _grow_loop()
 	echo
 
 	echo "*** mkfs loop file (size=$original)"
-	$MKFS_XFS_PROG -b size=$bsize $dparam $loop_dev | \
+	$MKFS_XFS_PROG -b size=$bsize $mkfs_opts $loop_dev | \
 		_filter_mkfs 2>/dev/null
 
 	echo "*** extend loop file"
-- 
2.53.0

[PATCH 2/2] xfs/216: disable all concurrency scaling

[Christoph Hellwig]

This test currently disables log concurrency scaling, but even the
data device concurrency scaling can create mismatching output on
systems with a large CPU count.

Reported-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 tests/xfs/216 | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tests/xfs/216 b/tests/xfs/216
index 1749647c11f7..ce8bb528410b 100755
--- a/tests/xfs/216
+++ b/tests/xfs/216
@@ -23,7 +23,7 @@ _cleanup()
 _require_scratch
 _scratch_mkfs_xfs >/dev/null 2>&1
 if _scratch_mkfs_xfs_supports_concurrency -l >> $seqres.full 2>&1; then
-	loop_mkfs_opts="-l concurrency=0"
+	loop_mkfs_opts="-d concurrency=0 -l concurrency=0 -r concurrency=0"
 else
 	loop_mkfs_opts=""
 fi
-- 
2.53.0

Re: [PATCH 1/2] xfs/078: disable all concurrency scaling

[Eric Sandeen]

On 6/18/26 8:48 AM, Christoph Hellwig wrote:
> Disable concurrency scaling to avoid mismatching output on systems with
> a large CPU count.
> 
> Reported-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
> Signed-off-by: Christoph Hellwig <hch@lst.de>

I think this papers over what I still feel is a problem in mkfs.xfs
behavior, but that's a different issue and making tests pass is good, so:

Reviewed-by: Eric Sandeen <sandeen@redhat.com>

> ---
>  tests/xfs/078 | 8 +++++---
>  1 file changed, 5 insertions(+), 3 deletions(-)
> 
> diff --git a/tests/xfs/078 b/tests/xfs/078
> index 6057aeea12ab..8ef52614b125 100755
> --- a/tests/xfs/078
> +++ b/tests/xfs/078
> @@ -57,9 +57,11 @@ _grow_loop()
>  	$XFS_IO_PROG -f -c "truncate $original" $LOOP_IMG
>  	loop_dev=`_create_loop_device $LOOP_IMG $bsize`
>  
> -	dparam=""
> +	# Disable concurrency scaling to avoid output differences due to large
> +	# CPU counts.
> +	mkfs_opts="-d concurrency=0 -l concurrency=0 -r concurrency=0"
>  	if [ -n "$agsize" ]; then
> -		dparam="-d agsize=$agsize"
> +		mkfs_opts="$mkfs_opts -d agsize=$agsize"
>  	fi
>  
>  	echo
> @@ -67,7 +69,7 @@ _grow_loop()
>  	echo
>  
>  	echo "*** mkfs loop file (size=$original)"
> -	$MKFS_XFS_PROG -b size=$bsize $dparam $loop_dev | \
> +	$MKFS_XFS_PROG -b size=$bsize $mkfs_opts $loop_dev | \
>  		_filter_mkfs 2>/dev/null
>  
>  	echo "*** extend loop file"

Re: [PATCH 1/2] xfs/078: disable all concurrency scaling

[Eric Sandeen]

On 6/18/26 8:54 AM, Eric Sandeen wrote:
> On 6/18/26 8:48 AM, Christoph Hellwig wrote:
>> Disable concurrency scaling to avoid mismatching output on systems with
>> a large CPU count.
>>
>> Reported-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
>> Signed-off-by: Christoph Hellwig <hch@lst.de>
> 
> I think this papers over what I still feel is a problem in mkfs.xfs
> behavior, but that's a different issue and making tests pass is good, so:
> 
> Reviewed-by: Eric Sandeen <sandeen@redhat.com>

Er, hang on - let me retract that. Lukas sent a series to address
this problem too, right, and one of the things he added in his series
is a helper to determine whether the concurrency options even exist,
so that this can continue to run on older xfsprogs, which seems like
a good idea, no?

https://git.kernel.org/pub/scm/fs/xfs/xfstests-dev.git/commit/?h=for-next&id=7f162f5bcf50fc32c7e3b432e7ff189571a3151f

which added _scratch_mkfs_xfs_supports_concurrency()

>> ---
>>  tests/xfs/078 | 8 +++++---
>>  1 file changed, 5 insertions(+), 3 deletions(-)
>>
>> diff --git a/tests/xfs/078 b/tests/xfs/078
>> index 6057aeea12ab..8ef52614b125 100755
>> --- a/tests/xfs/078
>> +++ b/tests/xfs/078
>> @@ -57,9 +57,11 @@ _grow_loop()
>>  	$XFS_IO_PROG -f -c "truncate $original" $LOOP_IMG
>>  	loop_dev=`_create_loop_device $LOOP_IMG $bsize`
>>  
>> -	dparam=""
>> +	# Disable concurrency scaling to avoid output differences due to large
>> +	# CPU counts.
>> +	mkfs_opts="-d concurrency=0 -l concurrency=0 -r concurrency=0"
>>  	if [ -n "$agsize" ]; then
>> -		dparam="-d agsize=$agsize"
>> +		mkfs_opts="$mkfs_opts -d agsize=$agsize"
>>  	fi
>>  
>>  	echo
>> @@ -67,7 +69,7 @@ _grow_loop()
>>  	echo
>>  
>>  	echo "*** mkfs loop file (size=$original)"
>> -	$MKFS_XFS_PROG -b size=$bsize $dparam $loop_dev | \
>> +	$MKFS_XFS_PROG -b size=$bsize $mkfs_opts $loop_dev | \
>>  		_filter_mkfs 2>/dev/null
>>  
>>  	echo "*** extend loop file"
> 
> 

Re: [PATCH 1/2] xfs/078: disable all concurrency scaling

[Shin'ichiro Kawasaki]

On Jun 18, 2026 / 09:23, Eric Sandeen wrote:
> On 6/18/26 8:54 AM, Eric Sandeen wrote:
> > On 6/18/26 8:48 AM, Christoph Hellwig wrote:
> >> Disable concurrency scaling to avoid mismatching output on systems with
> >> a large CPU count.
> >>
> >> Reported-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
> >> Signed-off-by: Christoph Hellwig <hch@lst.de>
> > 
> > I think this papers over what I still feel is a problem in mkfs.xfs
> > behavior, but that's a different issue and making tests pass is good, so:
> > 
> > Reviewed-by: Eric Sandeen <sandeen@redhat.com>
> 
> Er, hang on - let me retract that. Lukas sent a series to address
> this problem too, right, and one of the things he added in his series
> is a helper to determine whether the concurrency options even exist,
> so that this can continue to run on older xfsprogs, which seems like
> a good idea, no?
> 
> https://git.kernel.org/pub/scm/fs/xfs/xfstests-dev.git/commit/?h=for-next&id=7f162f5bcf50fc32c7e3b432e7ff189571a3151f
> 
> which added _scratch_mkfs_xfs_supports_concurrency()

I agree, I think we can add the change below on top of this patch.
Regardless of the change, I confirmed that this patch avoids the failure
that I reported. Thanks!

Tested-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>

diff --git a/tests/xfs/078 b/tests/xfs/078
index 8ef5261..dc0668e 100755
--- a/tests/xfs/078
+++ b/tests/xfs/078
@@ -59,7 +59,11 @@ _grow_loop()
 
 	# Disable concurrency scaling to avoid output differences due to large
 	# CPU counts.
-	mkfs_opts="-d concurrency=0 -l concurrency=0 -r concurrency=0"
+	if _scratch_mkfs_xfs_supports_concurrency -l >> $seqres.full 2>&1; then
+		mkfs_opts="-d concurrency=0 -l concurrency=0 -r concurrency=0"
+	else
+		mkfs_opts=""
+	fi
 	if [ -n "$agsize" ]; then
 		mkfs_opts="$mkfs_opts -d agsize=$agsize"
 	fi

Re: [PATCH 2/2] xfs/216: disable all concurrency scaling

[Shin'ichiro Kawasaki]

On Jun 18, 2026 / 15:48, Christoph Hellwig wrote:
> This test currently disables log concurrency scaling, but even the
> data device concurrency scaling can create mismatching output on
> systems with a large CPU count.
> 
> Reported-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
> Signed-off-by: Christoph Hellwig <hch@lst.de>

Thanks, I confirmed this patch avoid the failure that I reported.

Tested-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>

Re: [PATCH 1/2] xfs/078: disable all concurrency scaling

[Christoph Hellwig]

On Fri, Jun 19, 2026 at 10:55:21AM +0900, Shin'ichiro Kawasaki wrote:
> On Jun 18, 2026 / 09:23, Eric Sandeen wrote:
> > On 6/18/26 8:54 AM, Eric Sandeen wrote:
> > > On 6/18/26 8:48 AM, Christoph Hellwig wrote:
> > >> Disable concurrency scaling to avoid mismatching output on systems with
> > >> a large CPU count.
> > >>
> > >> Reported-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
> > >> Signed-off-by: Christoph Hellwig <hch@lst.de>
> > > 
> > > I think this papers over what I still feel is a problem in mkfs.xfs
> > > behavior, but that's a different issue and making tests pass is good, so:
> > > 
> > > Reviewed-by: Eric Sandeen <sandeen@redhat.com>
> > 
> > Er, hang on - let me retract that. Lukas sent a series to address
> > this problem too, right, and one of the things he added in his series
> > is a helper to determine whether the concurrency options even exist,
> > so that this can continue to run on older xfsprogs, which seems like
> > a good idea, no?
> > 
> > https://git.kernel.org/pub/scm/fs/xfs/xfstests-dev.git/commit/?h=for-next&id=7f162f5bcf50fc32c7e3b432e7ff189571a3151f
> > 
> > which added _scratch_mkfs_xfs_supports_concurrency()
> 
> I agree, I think we can add the change below on top of this patch.
> Regardless of the change, I confirmed that this patch avoids the failure
> that I reported. Thanks!

Thanks, I'll add that.

[PATCH 2/2] xfs/216: disable all concurrency scaling

[Christoph Hellwig]

This test currently disables log concurrency scaling, but even the
data device concurrency scaling can create mismatching output on
systems with a large CPU count.

Reported-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Tested-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
---
 tests/xfs/216 | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tests/xfs/216 b/tests/xfs/216
index 1749647c11f7..ce8bb528410b 100755
--- a/tests/xfs/216
+++ b/tests/xfs/216
@@ -23,7 +23,7 @@ _cleanup()
 _require_scratch
 _scratch_mkfs_xfs >/dev/null 2>&1
 if _scratch_mkfs_xfs_supports_concurrency -l >> $seqres.full 2>&1; then
-	loop_mkfs_opts="-l concurrency=0"
+	loop_mkfs_opts="-d concurrency=0 -l concurrency=0 -r concurrency=0"
 else
 	loop_mkfs_opts=""
 fi
-- 
2.53.0

Re: [PATCH 2/2] xfs/216: disable all concurrency scaling

[Eric Sandeen]

On 6/19/26 12:09 AM, Christoph Hellwig wrote:
> This test currently disables log concurrency scaling, but even the
> data device concurrency scaling can create mismatching output on
> systems with a large CPU count.
> 
> Reported-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> Tested-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>

Reviewed-by: Eric Sandeen <sandeen@redhat.com>

> ---
>  tests/xfs/216 | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/tests/xfs/216 b/tests/xfs/216
> index 1749647c11f7..ce8bb528410b 100755
> --- a/tests/xfs/216
> +++ b/tests/xfs/216
> @@ -23,7 +23,7 @@ _cleanup()
>  _require_scratch
>  _scratch_mkfs_xfs >/dev/null 2>&1
>  if _scratch_mkfs_xfs_supports_concurrency -l >> $seqres.full 2>&1; then
> -	loop_mkfs_opts="-l concurrency=0"
> +	loop_mkfs_opts="-d concurrency=0 -l concurrency=0 -r concurrency=0"
>  else
>  	loop_mkfs_opts=""
>  fi

Re: [PATCH 2/2] xfs/216: disable all concurrency scaling

[Darrick J. Wong]

On Fri, Jun 19, 2026 at 07:09:29AM +0200, Christoph Hellwig wrote:
> This test currently disables log concurrency scaling, but even the
> data device concurrency scaling can create mismatching output on
> systems with a large CPU count.
> 
> Reported-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> Tested-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
> ---
>  tests/xfs/216 | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/tests/xfs/216 b/tests/xfs/216
> index 1749647c11f7..ce8bb528410b 100755
> --- a/tests/xfs/216
> +++ b/tests/xfs/216
> @@ -23,7 +23,7 @@ _cleanup()
>  _require_scratch
>  _scratch_mkfs_xfs >/dev/null 2>&1
>  if _scratch_mkfs_xfs_supports_concurrency -l >> $seqres.full 2>&1; then
> -	loop_mkfs_opts="-l concurrency=0"
> +	loop_mkfs_opts="-d concurrency=0 -l concurrency=0 -r concurrency=0"

/me notes that -lconcurrency is not compatible with -llogdev and the
loopdev is not formatted with SCRATCH_MKFS_OPTIONS, so this won't work
to disable the concurrency= mkfs options if fstests is being run with
SCRATCH_LOGDEV set.

--D

>  else
>  	loop_mkfs_opts=""
>  fi
> -- 
> 2.53.0
> 
> 

Re: [PATCH 2/2] xfs/216: disable all concurrency scaling

[Christoph Hellwig]

On Wed, Jun 24, 2026 at 10:44:04AM -0700, Darrick J. Wong wrote:
> On Fri, Jun 19, 2026 at 07:09:29AM +0200, Christoph Hellwig wrote:
> > This test currently disables log concurrency scaling, but even the
> > data device concurrency scaling can create mismatching output on
> > systems with a large CPU count.
> > 
> > Reported-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
> > Signed-off-by: Christoph Hellwig <hch@lst.de>
> > Tested-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
> > ---
> >  tests/xfs/216 | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/tests/xfs/216 b/tests/xfs/216
> > index 1749647c11f7..ce8bb528410b 100755
> > --- a/tests/xfs/216
> > +++ b/tests/xfs/216
> > @@ -23,7 +23,7 @@ _cleanup()
> >  _require_scratch
> >  _scratch_mkfs_xfs >/dev/null 2>&1
> >  if _scratch_mkfs_xfs_supports_concurrency -l >> $seqres.full 2>&1; then
> > -	loop_mkfs_opts="-l concurrency=0"
> > +	loop_mkfs_opts="-d concurrency=0 -l concurrency=0 -r concurrency=0"
> 
> /me notes that -lconcurrency is not compatible with -llogdev and the
> loopdev is not formatted with SCRATCH_MKFS_OPTIONS, so this won't work
> to disable the concurrency= mkfs options if fstests is being run with
> SCRATCH_LOGDEV set.

Arrg.  I hate our option parsing mess.  Any good idea how we'd get
the desired result?  Unless there's something easy I'd be tempted to
leave it to the first person using an external log device and large
cpu counts..

Re: [PATCH 2/2] xfs/216: disable all concurrency scaling

[Darrick J. Wong]

On Thu, Jun 25, 2026 at 03:22:51PM +0200, Christoph Hellwig wrote:
> On Wed, Jun 24, 2026 at 10:44:04AM -0700, Darrick J. Wong wrote:
> > On Fri, Jun 19, 2026 at 07:09:29AM +0200, Christoph Hellwig wrote:
> > > This test currently disables log concurrency scaling, but even the
> > > data device concurrency scaling can create mismatching output on
> > > systems with a large CPU count.
> > > 
> > > Reported-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
> > > Signed-off-by: Christoph Hellwig <hch@lst.de>
> > > Tested-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
> > > ---
> > >  tests/xfs/216 | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > 
> > > diff --git a/tests/xfs/216 b/tests/xfs/216
> > > index 1749647c11f7..ce8bb528410b 100755
> > > --- a/tests/xfs/216
> > > +++ b/tests/xfs/216
> > > @@ -23,7 +23,7 @@ _cleanup()
> > >  _require_scratch
> > >  _scratch_mkfs_xfs >/dev/null 2>&1
> > >  if _scratch_mkfs_xfs_supports_concurrency -l >> $seqres.full 2>&1; then
> > > -	loop_mkfs_opts="-l concurrency=0"
> > > +	loop_mkfs_opts="-d concurrency=0 -l concurrency=0 -r concurrency=0"
> > 
> > /me notes that -lconcurrency is not compatible with -llogdev and the
> > loopdev is not formatted with SCRATCH_MKFS_OPTIONS, so this won't work
> > to disable the concurrency= mkfs options if fstests is being run with
> > SCRATCH_LOGDEV set.
> 
> Arrg.  I hate our option parsing mess.  Any good idea how we'd get
> the desired result?  Unless there's something easy I'd be tempted to
> leave it to the first person using an external log device and large
> cpu counts..

Yeah, that's me. :)

I just sent my latest batch of random fixes:
https://lore.kernel.org/fstests/178240619625.1665857.3954516357764795627.stgit@frogsfrogsfrogs/

I don't know that it's necessary to add -dconcurrency=0 for xfs/216
after that (it seemed to work for me) but I don't have any objections to
adding more:

if _mkfs_xfs_supported $loop_mkfs_opts -l concurrency=0 $loop_dev &>> $seqres.full; then
	loop_mkfs_opts="$loop_mkfs_opts -l concurrency=0"
fi
if _mkfs_xfs_supported $loop_mkfs_opts -d concurrency=0 $loop_dev &>> $seqres.full; then
	loop_mkfs_opts="$loop_mkfs_opts -d concurrency=0"
fi

--D

Re: [PATCH 2/2] xfs/216: disable all concurrency scaling

[Christoph Hellwig]

I'll retract this patch in favour of the btrfs fix from Darrick.
We'll still need patch 1, though.

Re: [PATCH 2/2] xfs/216: disable all concurrency scaling

[Zorro Lang]

On Fri, Jun 26, 2026 at 06:53:53AM +0200, Christoph Hellwig wrote:
> I'll retract this patch in favour of the btrfs fix from Darrick.
> We'll still need patch 1, though.

Let me make sure I have the current status mapped out correctly. So we
should apply/merge the following patches:

[PATCH 1/2] common/xfs: fix _scratch_mkfs_xfs_supports_concurrency
[PATCH 2/2] xfs/21[67]: fix mkfs log concurrency detection
[PATCH 1/2] xfs/078: disable all concurrency scaling

And revert:
[PATCH 2/2] xfs/216: disable all concurrency scaling

Is that correct?

> 

Re: [PATCH 2/2] xfs/216: disable all concurrency scaling

[Darrick J. Wong]

On Tue, Jun 30, 2026 at 02:00:10AM +0800, Zorro Lang wrote:
> On Fri, Jun 26, 2026 at 06:53:53AM +0200, Christoph Hellwig wrote:
> > I'll retract this patch in favour of the btrfs fix from Darrick.
> > We'll still need patch 1, though.
> 
> Let me make sure I have the current status mapped out correctly. So we
> should apply/merge the following patches:
> 
> [PATCH 1/2] common/xfs: fix _scratch_mkfs_xfs_supports_concurrency
> [PATCH 2/2] xfs/21[67]: fix mkfs log concurrency detection
> [PATCH 1/2] xfs/078: disable all concurrency scaling
> 
> And revert:
> [PATCH 2/2] xfs/216: disable all concurrency scaling
> 
> Is that correct?

Sounds fine to me.

--D

Re: [PATCH 2/2] xfs/216: disable all concurrency scaling

[Christoph Hellwig]

On Mon, Jun 29, 2026 at 03:58:57PM -0700, Darrick J. Wong wrote:
> > Let me make sure I have the current status mapped out correctly. So we
> > should apply/merge the following patches:
> > 
> > [PATCH 1/2] common/xfs: fix _scratch_mkfs_xfs_supports_concurrency
> > [PATCH 2/2] xfs/21[67]: fix mkfs log concurrency detection
> > [PATCH 1/2] xfs/078: disable all concurrency scaling
> > 
> > And revert:
> > [PATCH 2/2] xfs/216: disable all concurrency scaling
> > 
> > Is that correct?
> 
> Sounds fine to me.

Same here.