Re: Stall at page allocations with __GFP_RETRY_MAYFAIL (Re: [PATCH v1] ALSA: memalloc: Fix indefinite hang in non-iommu case)

[Sven van Ashbrook <svenva@xxxxxxxxxxxx>]

Hi Takashi,

On Wed, Feb 21, 2024 at 4:21 AM Takashi Iwai <tiwai@xxxxxxx> wrote:
>
> Both look like the code path via async PM resume.
> Were both from the runtime PM resume? Or the system resume?

The large firmware allocation that triggers the stall happens in
runtime resume.

This means runtime resume and system resume are affected.

Due to the way runtime PM works, the system suspend path is also affected.
Because system suspend first wakes up any runtime suspended devices,
before executing their system suspend callback.

So we get:
system active, SOF runtime suspended (audio not active)
  -> system suspend request
    -> calls SOF runtime resume callback (stall happens here)
    -> calls SOF system suspend callback