Messages in this thread | | | Date | Sun, 18 Mar 2012 17:24:44 -0400 | From | Dave Jones <> | Subject | Re: [3.3-rc7] sys_poll use after free (hibernate) |
| |
On Sun, Mar 18, 2012 at 08:47:19PM +0100, richard -rw- weinberger wrote: > On Tue, Mar 13, 2012 at 1:58 AM, Dave Jones <davej@redhat.com> wrote: > > While trying to reproduce the i915 memory corruption problem with hibernate, > > If you cannot reproduce the problem you can send me debug/test-patches. > I have two machines where I can reproduce the issue within an hour.
At the time I was hoping to bisect it. But it looks like it only happens on ironlake era graphics and newer, which I don't have.
With that info, I'm not sure it's really bisectable. The bug has probably been there since day 1 when ironlake support was added. See the thread 'Subject: Re: inode->i_wb_list corruption.' for some further thoughts, where it's theorised that the GTT contains stale entries after when we thaw. I wouldn't be surprised if no-one had even tried hibernate (or at least noticed the memory corruption immediately) before that was merged.
My thinking is that some kind of GTT teardown in the ->hibernate routine is probably what's needed.
Dave
| |