Re: [PATCH] mm/page_alloc: Work around a pahole limitation with zero-sized struct pagesets

[Andrii Nakryiko]

On Wed, May 26, 2021 at 11:13 AM Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx> wrote:
>
> On Wed, May 26, 2021 at 09:57:31AM -0700, Andrii Nakryiko wrote:
> > > This patch checks for older versions of pahole and forces struct pagesets
> > > to be non-zero sized as a workaround when CONFIG_DEBUG_INFO_BTF is set. A
> > > warning is omitted so that distributions can update pahole when 1.22
> >
> > s/omitted/emitted/ ?
> >
>
> Yes.
>
> > > is released.
> > >
> > > Reported-by: Michal Suchanek <msuchanek@xxxxxxx>
> > > Reported-by: Hritik Vijay <hritikxx8@xxxxxxxxx>
> > > Debugged-by: Andrii Nakryiko <andrii.nakryiko@xxxxxxxxx>
> > > Signed-off-by: Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx>
> > > ---
> >
> > Looks good! I verified that this does fix the issue on the latest
> > linux-next tree, thanks!
> >
>
> Excellent
>
> > One question, should
> >
> > Fixes: 5716a627517d ("mm/page_alloc: convert per-cpu list protection
> > to local_lock")
> >
> > be added to facilitate backporting?
> >
>
> The git commit is not stable because the patch "mm/page_alloc: convert
> per-cpu list protection to local_lock" is in Andrew's mmotm tree which is

Oh, I didn't know about this instability.

> quilt based. I decided not to treat the patch as a fix because the patch is
> not wrong as such, it's a limitation of an external tool.  However, I would
> expect both the problematic patch and the BTF workaround to go in during
> the same merge window so backports to -stable should not be required.

Yep, makes sense.

>
> > Either way:
> >
> > Acked-by: Andrii Nakryiko <andrii@xxxxxxxxxx>
> > Tested-by: Andrii Nakryiko <andrii@xxxxxxxxxx>
> >
>
> Thanks!
>
> --
> Mel Gorman
> SUSE Labs