Re: [PATCH 6.12 00/24] 6.12.72-rc1 review

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

On Mon Feb 16, 2026 at 4:47 PM CET, Danilo Krummrich wrote:
> On Mon Feb 16, 2026 at 3:27 PM CET, Mark Brown wrote:
>> On Fri, Feb 13, 2026 at 02:48:19PM +0100, Greg Kroah-Hartman wrote:
>>> Gui-Dong Han <hanguidong02@gmail.com>
>>>     driver core: enforce device_lock for driver_match_device()
>>
>> This breaks boot on at least the Arm Juno platform, upstream it
>> introduced regressions on quite a few systems due to drivers registering
>> in the probe of other devices.  That's obviously not a great pattern but
>> a regreession is a regression.
>
> Just for reference, I've also sent the following to the stable patch apply
> notice:
>
> "This commit reveals a few driver bugs resulting in deadlocks without the
> following fixes:
>
>   - 1. ed1ac3c977dd ("iommu/arm-smmu-qcom: do not register driver in probe()")

One additional note, we want this commit backported regardless, as it also fixes
commit 0b4eeee2876f ("iommu/arm-smmu-qcom: Register the TBU driver in
qcom_smmu_impl_init").

I.e. the current code is racy in terms of async probe and the driver is never
unregistered even if built as module and the module is unloaded, which is a
potential UAF.

>   - 2. 730e5ebff40c ("gpio: omap: do not register driver in probe()")
>   - 3. https://lore.kernel.org/lkml/20260212235842.85934-1-dakr@kernel.org/
>
> The third one will hopefully be picked up by the clk folks soon.
>
> (1) should be required since v6.11, (2) since (basically forever) v2.6.22 and
> (3) since v5.11.
>
> We should also consider that we do not know if (especially older) stable trees
> have similar cases that we did not catch in linux-next."
>
> - Danilo