[Regression] hangs caused by commit 3202fa62fb (slub: relocate freelist pointer to middle of object)

[Regression] hangs caused by commit 3202fa62fb (slub: relocate freelist pointer to middle of object)

[Paul Menzel]

Dear Kees, dear Andrew,


No idea, if you are aware of it yet, but three people verified that 
commit 3202fa62fb (slub: relocate freelist pointer to middle of object) 
causes a regression on AMD hardware [1].

It’d be great, if you took a look, and advised if this commit (and 
follow-ups) should be reverted, until the issue is analyzed.


Kind regards,

Paul


[1]: https://bugzilla.kernel.org/show_bug.cgi?id=207383
      "[Regression] 5.7 amdgpu/polaris11 gpf: amdgpu_atomic_commit_tail"

Re: [Regression] hangs caused by commit 3202fa62fb (slub: relocate freelist pointer to middle of object)

[Kees Cook]

On Tue, Jul 21, 2020 at 04:55:12PM +0200, Paul Menzel wrote:
> No idea, if you are aware of it yet, but three people verified that commit
> 3202fa62fb (slub: relocate freelist pointer to middle of object) causes a
> regression on AMD hardware [1].

Hi, thanks for emailing; I don't get bugzilla notifications, so I hadn't
seen this yet.

> It’d be great, if you took a look, and advised if this commit (and
> follow-ups) should be reverted, until the issue is analyzed.

There have been a number of fixes to that commit (which I see are
mentioned in a quick skim of the bug), but they've mostly been around
additional slab debugging features. If it's causing a problem outside
of that, my instinct would be there might be a use-after-free happening,
but I'll go read the bug more closely now, and comment there (or here,
if needed).

Thanks!

-- 
Kees Cook