Re: [PATCH] mm: get rid of WARN if failed to cow user pages

Re: [PATCH] mm: get rid of WARN if failed to cow user pages

[Dan Williams]

[ drop Ross, add Kirill, linux-mm, and lkml ]

On Tue, Dec 24, 2019 at 9:42 PM Murphy Zhou <jencce.kernel@gmail.com> wrote:
>
> By running xfstests with fsdax enabled, generic/437 always hits this
> warning[1] since this commit:
>
> commit 83d116c53058d505ddef051e90ab27f57015b025
> Author: Jia He <justin.he@arm.com>
> Date:   Fri Oct 11 22:09:39 2019 +0800
>
>     mm: fix double page fault on arm64 if PTE_AF is cleared
>
> Looking at the test program[2] generic/437 uses, it's pretty easy
> to hit this warning. Remove this WARN as it seems not necessary.

This is not sufficient justification. Does this same test fail without
DAX? If not, why not? At a minimum you need to explain why this is not
indicating a problem.

Re: [PATCH] mm: get rid of WARN if failed to cow user pages

[Jeff Moyer]

Dan Williams <dan.j.williams@intel.com> writes:

> [ drop Ross, add Kirill, linux-mm, and lkml ]
>
> On Tue, Dec 24, 2019 at 9:42 PM Murphy Zhou <jencce.kernel@gmail.com> wrote:
>>
>> By running xfstests with fsdax enabled, generic/437 always hits this
>> warning[1] since this commit:
>>
>> commit 83d116c53058d505ddef051e90ab27f57015b025
>> Author: Jia He <justin.he@arm.com>
>> Date:   Fri Oct 11 22:09:39 2019 +0800
>>
>>     mm: fix double page fault on arm64 if PTE_AF is cleared
>>
>> Looking at the test program[2] generic/437 uses, it's pretty easy
>> to hit this warning. Remove this WARN as it seems not necessary.
>
> This is not sufficient justification. Does this same test fail without
> DAX? If not, why not? At a minimum you need to explain why this is not
> indicating a problem.

I ran into this, too, and Kirill has posted a patch[1] to fix the issue.
Note that it's a potential data corrupter, so just removing the warning
is NOT the right approach.  :)

-Jeff

[1] https://lore.kernel.org/linux-mm/20200218154151.13349-1-kirill.shutemov@linux.intel.com/T/#u

Re: [PATCH] mm: get rid of WARN if failed to cow user pages

[Murphy Zhou]

On Tue, Feb 18, 2020 at 04:09:30PM -0500, Jeff Moyer wrote:
> Dan Williams <dan.j.williams@intel.com> writes:
> 
> > [ drop Ross, add Kirill, linux-mm, and lkml ]
> >
> > On Tue, Dec 24, 2019 at 9:42 PM Murphy Zhou <jencce.kernel@gmail.com> wrote:
> >>
> >> By running xfstests with fsdax enabled, generic/437 always hits this
> >> warning[1] since this commit:
> >>
> >> commit 83d116c53058d505ddef051e90ab27f57015b025
> >> Author: Jia He <justin.he@arm.com>
> >> Date:   Fri Oct 11 22:09:39 2019 +0800
> >>
> >>     mm: fix double page fault on arm64 if PTE_AF is cleared
> >>
> >> Looking at the test program[2] generic/437 uses, it's pretty easy
> >> to hit this warning. Remove this WARN as it seems not necessary.
> >
> > This is not sufficient justification. Does this same test fail without
> > DAX? If not, why not? At a minimum you need to explain why this is not
> > indicating a problem.
> 
> I ran into this, too, and Kirill has posted a patch[1] to fix the issue.
> Note that it's a potential data corrupter, so just removing the warning
> is NOT the right approach.  :)

Agree :) Thanks!

> 
> -Jeff
> 
> [1] https://lore.kernel.org/linux-mm/20200218154151.13349-1-kirill.shutemov@linux.intel.com/T/#u
>