Re: [PATCH v2] gpu: Fix uninitialized buddy for built-in drivers

["Danilo Krummrich" <dakr@kernel.org>]

On Thu Feb 19, 2026 at 1:44 PM CET, Matthew Auld wrote:
> On 19/02/2026 11:14, Danilo Krummrich wrote:
>> On Thu Feb 19, 2026 at 11:38 AM CET, Matthew Auld wrote:
>>> On 19/02/2026 10:16, Danilo Krummrich wrote:
>>>> On Mon Feb 16, 2026 at 12:19 PM CET, Koen Koning wrote:
>>>>> Use subsys_initcall instead of module_init for the GPU buddy allocator,
>>>>> so its initialization code runs before any gpu drivers.
>>>>> Otherwise, a built-in driver that tries to use the buddy allocator will
>>>>> run into a kernel NULL pointer dereference because slab_blocks is
>>>>> uninitialized.
>>>>>
>>>>> Specifically, this fixes drm/xe (as built-in) running into a kernel
>>>>> panic during boot, because it uses buddy during device probe.
>>>>
>>>> I just noticed that this patch was sent twice, and I posted my feedback on [1]
>>>> -- pasting it here as well.
>>>>
>>>>> Fixes: ba110db8e1bc ("gpu: Move DRM buddy allocator one level up (part two)")
>>>>
>>>> This Fixes: tag seems wrong. How is this code move related to this problem?
>>>
>>> This popped up as a very recent regression for us internally. It looks
>>> like it worked before since link order ensured drm_buddy came before all
>>> the driver code. With above commit the link order changed and became
>>> drm/ and then buddy. See [1] also, which is maybe clearer to see this.
>> 
>> I see, I think this would be a good hint for the commit message. :)
>> 
>> However, I think it was never meant to rely on a build system implementation
>> detail, nor would this be correct. So, I think this should add both Fixes: tags.
>
> Yeah, I'm really not sure tbh. From a quick grep there do seem to be 
> other users relying on this:
>
> drm/drm_drv.c:1274:module_init(drm_core_init);
> drm/scheduler/sched_fence.c:238:module_init(drm_sched_fence_slab_init);
>
> The sched one looks identical with the slab thing. Do these need to be 
> fixed also?

Yes, those should be fixed as well.

Also note that module_init() compiles down to device_initcall() when built-in,
i.e. the initcall stage that is mainly for drivers, not for subsystem code.

Do you want to send a fix for thise as well?

>> 
>> Whether it should be backported is a different question though, as it seems to
>> work by accident in previous versions, i.e. it is only a "potential bug".
>> 
>> My personal opinion is that it should be backported either way, however that's
>> ultimately up to the stable team.
>> 
>> - Danilo
>> 
>>>
>>> [1]
>>> https://lore.kernel.org/intel-xe/20260213152047.179628-1-koen.koning@linux.intel.com/
>>>
>>>>
>>>> This should rather be:
>>>>
>>>> Fixes: 6387a3c4b0c4 ("drm: move the buddy allocator from i915 into common drm")
>>>>
>>>> Also, please add:
>>>>
>>>> Cc: stable@vger.kernel.org
>>>>
>>>>> Cc: Joel Fernandes <joelagnelf@nvidia.com>
>>>>> Cc: Dave Airlie <airlied@redhat.com>
>>>>> Cc: intel-xe@lists.freedesktop.org
>>>>> Cc: Peter Senna Tschudin <peter.senna@linux.intel.com>
>>>>> Cc: Matthew Auld <matthew.auld@intel.com>
>>>>> Signed-off-by: Koen Koning <koen.koning@linux.intel.com>
>>>>
>>>> [1] https://lore.kernel.org/all/DGIUUFLC31D5.2OZBF5FWQJWMZ@kernel.org/
>>