Re: [PATCH v5] drm/test: add a test suite for GEM objects backed by shmem

[Guenter Roeck]

Hi Marco,

On 2/22/24 07:32, Marco Pagani wrote:
> On 2024-02-18 16:49, Guenter Roeck wrote:
> > Hi,
> >
> > On Thu, Nov 30, 2023 at 06:14:16PM +0100, Marco Pagani wrote:
> > > This patch introduces an initial KUnit test suite for GEM objects
> > > backed by shmem buffers.
> > >
> > > Suggested-by: Javier Martinez Canillas <javierm@xxxxxxxxxx>
> > > Signed-off-by: Marco Pagani <marpagan@xxxxxxxxxx>
> >
> > When running this in qemu, I get lots of warnings backtraces in the drm
> > core.
> >
> > WARNING: CPU: 0 PID: 1341 at drivers/gpu/drm/drm_gem_shmem_helper.c:327
> > WARNING: CPU: 0 PID: 1341 at drivers/gpu/drm/drm_gem_shmem_helper.c:173
> > WARNING: CPU: 0 PID: 1341 at drivers/gpu/drm/drm_gem_shmem_helper.c:385
> > WARNING: CPU: 0 PID: 1341 at drivers/gpu/drm/drm_gem_shmem_helper.c:211
> > WARNING: CPU: 0 PID: 1345 at kernel/dma/mapping.c:194
> > WARNING: CPU: 0 PID: 1347 at drivers/gpu/drm/drm_gem_shmem_helper.c:429
> > WARNING: CPU: 0 PID: 1349 at drivers/gpu/drm/drm_gem_shmem_helper.c:445
> >
> > It looks like dma_resv_assert_held() asserts each time it is executed.
> > The backtrace in kernel/dma/mapping.c is triggered by
> > 	if (WARN_ON_ONCE(!dev->dma_mask))
> > 		return 0;
> > in __dma_map_sg_attrs().
> >
> > Is this a possible problem in the test code, or can it be caused by
> > some limitations or bugs in the qemu emulation ? If so, do you have any
> > thoughts or ideas what those limitations / bugs might be ?
>
> Hi Guenter,
>
> Thanks for reporting this issue. As you correctly noted, the warnings appear to
> be caused by the dma_mask in the mock device being uninitialized. I'll send a
> patch to fix it soon.

Thanks a lot for the update.

In this context, the TTM unit tests fail as well in qemu, with worse result:
It seems there is some bad cleanup after a failed test case, causing list
corruptions in the drm core and ultimately a crash. I don't know if this
is also caused by the missing dma_mask initialization.

Thanks,
Guenter