Re: Re: [PATCH 2/3 v6] ACPI: allow longer device IDs

[Ard Biesheuvel <ardb@kernel.org>]

On Mon, 28 Feb 2022 at 23:01, Alexander Graf <graf@amazon.com> wrote:
>
> Hi Andy,
>
> On 28.02.22 22:27, Andy Shevchenko wrote:
> > On Mon, Feb 28, 2022 at 10:02:43PM +0100, Ard Biesheuvel wrote:
> >> On Mon, 28 Feb 2022 at 21:47, Andy Shevchenko <andy.shevchenko@gmail.com> wrote:
> >>> On Mon, Feb 28, 2022 at 9:28 PM Jason A. Donenfeld <Jason@zx2c4.com> wrote:
> >>>> From: Alexander Graf <graf@amazon.com>
> >>>>
> >>>> We create a list of ACPI "PNP" IDs which contains _HID, _CID, and CLS
> >>>> entries of the respective devices. However, when making structs for
> >>>> matching, we squeeze those IDs into acpi_device_id, which only has 9
> >>>> bytes space to store the identifier. The subsystem actually captures the
> >>>> full length of the IDs, and the modalias has the full length, but this
> >>>> struct we use for matching is limited. It originally had 16 bytes, but
> >>>> was changed to only have 9 in 6543becf26ff ("mod/file2alias: make
> >>>> modalias generation safe for cross compiling"), presumably on the theory
> >>>> that it would match the ACPI spec so it didn't matter.
> >>>> Unfortunately, while most people adhere to the ACPI specs, Microsoft
> >>>> decided that its VM Generation Counter device [1] should only be
> >>>> identifiable by _CID with a value of "VM_Gen_Counter", which is longer
> >>>> than 9 characters.
> >>> Then why do we not see the ECR from somebody to update the spec or to
> >>> fix MS' abuse of it?
> >>> I believe _this_ should be the prerequisite to the proposed change.
> >> What exactly are you suggesting here? That the contributor of this
> >> patch joins the UEFI forum as an individual adopter in order to get
> >> the ACPI spec updated before we can advance with this patch? Or that
> >> he works with Microsoft to get them to refrain from violating it?
> >>
> >> I don't think that is reasonable or realistic. The kernel is already
> >> riddled with UEFI and ACPI quirks that are only there because some
> >> teams at MS don't take the ACPI spec too literally (which is why they
> >> have their own AML compiler, for one), and PC vendors only care about
> >> the Windows sticker, so they don't care about the ACPI spec either.
> >>
> >> So I don't think this is the right time to get pedantic about this.
> >> Our ACPI subsystem already deals with CIDs that are longer than 8
> >> characters (which are btw permitted by the ACPI spec for bus topology
> >> related metadata), the only thing being changed here is the ability to
> >> actually match against such identifiers.
> > My point is that this is clear abuse of the spec and:
> > 1) we have to enable the broken, because it is already in the wild with
> >     the comment that this is an issue
> >
> > AND
> >
> > 2) issue an ECR / work with MS to make sure they understand the problem.
> >
> > This can be done in parallel. What I meant as a prerequisite is to start doing
> > 2) while we have 1) on table.
>
>
> While trying to revalidate whether this really is breaking the spec,
> I've tried to reread the respective section in it and I'm afraid that it
> may be valid use of the _CID identifier:
>
>
> """
>
> 6.1.2 _CID (Compatible ID)
>
> This optional object is used to supply OSPM with a device’s Plug and
> Play-Compatible Device ID. Use _CID objects when a device has no other
> defined hardware standard method to report its compatible IDs. The _CID
> object is valid only within a Full Device Descriptor. An _HID object
> must also be present.
>
> Arguments:
>
> None
>
> Return Value:
> An Integer or String containing a single CID or a Package containing a
> list of CIDs A _CID object evaluates to either:
>
>   *
>
>     A single Compatible Device ID
>
>   *
>
>     A package of Compatible Device IDs for the device – in the order of
>     preference, highest preference first.
>
> Each Compatible Device ID must be either:
>
>   *
>
>     A valid HID value (a 32-bit compressed EISA type ID or a string such
>     as “ACPI0004”).
>
>   *
>
>     A string that uses a bus-specific nomenclature. For example, _CID
>     can be used to specify the PCI ID. The format of a PCI ID string is
>     one of the following:
>
> "PCI\CC_ccss" "PCI\CC_ccsspp"
> "PCI\VEN_vvvv&DEV_dddd&SUBSYS_ssssssss&REV_rr"
> "PCI\VEN_vvvv&DEV_dddd&SUBSYS_ssssssss" "PCI\VEN_vvvv&DEV_dddd&REV_rr"
> "PCI\VEN_vvvv&DEV_dddd"
>
> """
>
> In this case, you could interpret things as looking at "bus-specific
> nomenclature" case which even in the examples mentioned in the spec
> exceeds the 8 character limit we impose on the matching logic today.
>

This is what I was referring to when I mentioned bus topology related metadata.

This is why those uses of ACPI_ID_LEN outside of struct acpi_device_id
may potentially be dangerous, given that _CIDs are apparently
effectively unbounded in size. But relying on this to justify the
"VM_GEN_COUNTER" CID is a bit of a stretch, IMO.

> There still is spec violation in Hyper-V's VMGenID device's _HID value
> which doesn't follow the PNP format, but that's not relevant here. _CID
> doesn't seem to have the same restrictions?
>