Re: [PATCH] drm/imagination: Use dev_pm_domain_attach_list()

[Matt Coster]

On 27/02/2026 08:10, Geert Uytterhoeven wrote:
> Hi Mark,
> 
> On Thu, 26 Feb 2026 at 18:24, Mark Brown <broonie@xxxxxxxxxx> wrote:
>> On Fri, Jan 23, 2026 at 02:44:50PM +0000, Matt Coster wrote:
>>> This helper handles the attaching and linking of the entire list of power
>>> domains. Besides making pvr_power_domains_init() simpler, this also lays
>>> the groundwork to simplify supporting the varied power domain names used in
>>> Volcanic GPU cores.
>>
>> I'm seeing oopses in -next on at least k3-am625-verdin-wifi-mallow which
>> bisect to this patch.  We get:
>>
>> [   10.820056] powervr fd00000.gpu: Direct firmware load for powervr/rogue_33.15.11.3_v1.fw failed with error -2
>> [   10.831903] powervr fd00000.gpu: [drm] *ERROR* failed to load firmware powervr/rogue_33.15.11.3_v1.fw (err=-2)
>> ...
>> [   10.844023] Unable to handle kernel NULL pointer dereference at virtual address 0000000000000018
>> ...
>> [   11.090162] Call trace:
>> [   11.092600]  pvr_power_domains_fini+0x18/0xa0 [powervr] (P)
>> [   11.098218]  pvr_probe+0x100/0x14c [powervr]
>> [   11.102505]  platform_probe+0x5c/0xa4
>>
>> which does seem relevant to the changed code.
>>
>> Full log:
>>
>>    https://lava.sirena.org.uk/scheduler/job/2499326*L865
>>
>> bisect log (with links to additional runtime logs):
> 
>> # first bad commit: [e19cc5ab347e3cdcc21c97ea5d11af8da7f1358d] drm/imagination: Use dev_pm_domain_attach_list()
> 
> Thanks for your report!

+1 from us :)

> 
> I can confirm this crash on e.g. R-Car H3 ES2.0 (after adding a
> GPU node).  Actually I had seen it earlier this week, but didn't pay
> enough attention.  Hence  I just assumed it was the known race condition
> when removing multiple PM Domains, and thus missed it is a new bug.
> 
> The issue is that pvr_power->domains is a NULL pointer, thus causing
> a crash when dereferencing that pointer:
> 
>     int i = (int)pvr_power->domains->num_pds - 1;

We've confirmed the same on AM625; there's patches coming today for this
fix and another possible foot-gun we spotted in the process.

Cheers,
Matt

> 
> Gr{oetje,eeting}s,
> 
>                         Geert
> 


-- 
Matt Coster
E: matt.coster@xxxxxxxxxx

Attachment: OpenPGP_signature.asc
Description: OpenPGP digital signature