On Sun, Aug 30, 2026 at 11:14:34PM +0200, Lukas Wunner wrote:
> The "no active error status is read" is the only scenario where
> this *could* happen. However if no error bits are set in the
> status register, why did the device send an error message upstream
> in the first place? That would be a broken device.
>
> Another theoretical possibility would be that the device becomes
> inaccessible by the time aer_get_device_error_info() reads the
> status/mask registers. They'd both be read as "all ones",
> hence info->status & ~info->mask == 0.
Hi Lukas,
Thanks for the review and clarifying that.
The underlying reference count leak is genuine, but the additional topology
scenarios listed in v1 were inadvertently incorporated from an automated AI
review bot report, which introduced those inaccurate/hallucinated cases.
I have updated this in v2:
- Corrected the Fixes tag to 60271ab044a5 ("PCI/AER: Take reference on error
devices").
- Cleaned up the commit description to remove the extra scenarios and focus
strictly on the actual failure path (device becoming inaccessible / register
read returning no active unmasked status).
Thanks,
Priyank
