[PATCH i-g-t v2] Fix memory access issue due to variable block scope

[Peter Senna Tschudin]

On Wed, Mar 27, 2024 at 5:59 PM Kamil Konieczny
<kamil.konieczny at linux.intel.com> wrote:
>
> Hi Peter,
>
> On 2024-03-26 at 22:41:27 +0100, Peter Senna Tschudin wrote:
> > Hi Kamil,
> >
> > Thank you for your email!
> >
> > On Tue, Mar 26, 2024 at 7:20 PM Kamil Konieczny
> > <kamil.konieczny at linux.intel.com> wrote:
> > >
> > > Hi peter.senna,
> > > On 2024-03-26 at 14:18:43 +0100, Peter Senna Tschudin wrote:
> > > > Hi Andi,
> > > >
> > > > Thank you for your reply.
> > > >
> > > > On Tue, Mar 26, 2024 at 1:05 PM Andi Shyti <andi.shyti at linux.intel.com> wrote:
> > > > >
> > > > > Hey Peter,
> > > > >
> > > > > On Mon, Mar 25, 2024 at 10:35:48PM +0100, Peter Senna Tschudin wrote:
> > > > > > This patch fixes the tests gem_exec_capture at many-4k-incremental and
> > > > > > gem_exec_capture at many-4k-zero that are currently failing with an invalid file
> > > > > > descriptor error.
> > > > >
> > > > > where is gem_exec_capture calling for_each_ctx_cfg_engine()?
> > > >
> > > > many(), userptr(), capture_invisible()
> > > >     find_first_available_engine()
> > > >         for_each_ctx_engine()
> > > >
> > > > When called by many(), 'e' gets corrupted when configure_hangs() tries
> > > > to assign 'e' to another variable. Then after 'e' is corrupted, the
> > > > call __captureN() will fail because it expects 'e' to be valid.
> > > >
> > >
> > > imho first step whould be to replace macro with a function:
> > >
> > > saved = find_first_available_engine(fd, ctx, e);
> >
> > When I was done with the replacement function, I found a simple
> > solution that seems to be adequate. I made a small change to
> > find_first_available_engine() to reassign e to an element of saved. My
> > patch v3 has the details. What do you think?
>
> I will reply to your v4 patch, it looks promising.

Thanks!

>
> >
> > May I ask you to increase the time the CI will waits for the tests to complete:
> >  - many-4k-zero: 2 minutes
> >  - many-4k-incremental: 5 minutes (It takes 4 minutes on my N3150)
>
> It is not posibble, sorry, as there are many of them and time adds up...

Should I send a patch to reduce the test sizes by half?

[...]