Re: [PATCH] bus: mhi: host: pci_generic: Add per-device ACPI reset

From: Loic Poulain

Date: Wed Nov 18 2026 - 05:41:06 EST


On Wed, Nov 18, 2026 at 10:29 AM Hongxing Xu <xing1989093@xxxxxxx> wrote:
>
> Hi Loic,
>
> Thanks for the review. Let me check that I read your two points correctly:
>
> - The reset method should not live in mhi_pci_dev_info, because whether an
> ACPI method exists depends on the host the device is plugged into, not on
> the device itself.
> - The driver should instead walk a list of methods and take the first one
> that is present, from the most generic to the most platform specific.
>
> I agree with the first point, and with the idea of a list. But I would like
> to suggest the opposite order: platform specific first.
>
> The two entries are not two ways of doing the same thing, they are two
> different kinds of reset, and the spec keeps them apart on purpose:
>
> - _RST under a device scope is a function level reset. It must affect only
> that device, and it must leave the device visible on the bus, so OSPM does
> not re-enumerate it.
> - MRST, referenced through _PRR, is a platform level reset. It power cycles
> the device, so the device comes back as if it had been removed and
> re-inserted.
>
> The spec says a device should use _PRR when its platform reset cannot satisfy
> the function level and bus requirements, and that a device may expose both.
> That is exactly the situation here.
>
> On our current board, both are present. The function level _RST succeeds, so a
> generic-first walk stops there. But it only drives the link reset signals; it
> never removes power from the module, so the part is not really reset. There
> is nothing on the host side that can tell the difference afterwards, so the
> walk never reaches MRST. A method that exists but does not do the job is
> worse than no method, because it hides the failure.
>
> So: would you be OK with trying the platform specific entry first? That is
> my suggestion, but if you prefer to keep generic-first I would rather hear
> your reasoning now than write v2 the wrong way.

Yes, from specific to generic method sounds good.

Regards,
Loic