Re: [PATCH] rust: drm: fix unsound initialization in drm::Device::new

["Eliot Courtney" <ecourtney@nvidia.com>]

On Tue Apr 28, 2026 at 9:50 PM JST, Gary Guo wrote:
> On Tue Apr 28, 2026 at 1:43 PM BST, Danilo Krummrich wrote:
>> On Tue Apr 28, 2026 at 2:20 PM CEST, Eliot Courtney wrote:
>>> If pinned initialization of drm::Device::Data fails, it calls
>>> drm::Device::release via drm_dev_put. This materializes a reference to
>>> &drm::Device, but it's not fully constructed yet, because initializing
>>> `data` failed. It should not be dropped either. Instead, if pinned
>>> initialization fails, make sure drm::Device::release isn't called.
>>>
>>> Fixes: 2e9fdbe5ec7a ("rust: drm: device: drop_in_place() the drm::Device in release()")
>>> Signed-off-by: Eliot Courtney <ecourtney@nvidia.com>
>>
>> There's already a patch from Lyude for this [1].
>>
>> That said, I like the approach with the ALLOC_VTABLE.
>>
>> @Lyude: Do you mind if we pick Eliot's patch?

Sorry, I should have checked before sending this!

>>
>> Thanks,
>> Danilo
>>
>> [1] https://lore.kernel.org/lkml/20260320233645.950190-2-lyude@redhat.com/
>
> I have to second this and I think this solution is very clean. It does mean
> that we're always duplicating vtable though, for just one pointer of difference.
>
> Is it possible to have a shared vtable for drm devices that's in the
> allocated-but-not-initialized state?

It looks like __drm_dev_alloc reads `driver_features` from `drm_driver`
so we'd have to have multiple shared ones if we ever have anything other
than FEAT_GEM for that. And I guess it relies on nothing ever changing
to read more from `drm_driver` during alloc. So I feel it's a bit
fragile, personally.

>
> Best,
> Gary