Re: [RFC PATCH] Revert "arm64: PCI: Exclude ACPI "consumer" resources from host bridge windows"

[Bjorn Helgaas]

On Thu, May 27, 2021 at 10:32:00AM +0100, Lorenzo Pieralisi wrote:
> On Wed, May 26, 2021 at 09:58:36PM +0100, Will Deacon wrote:
> > On Tue, May 11, 2021 at 01:40:20AM +0200, Maximilian Luz wrote:
> > > The Microsoft Surface Pro X has host bridges defined as
> > > 
> > >     Name (_HID, EisaId ("PNP0A08") /* PCI Express Bus */)  // _HID: Hardware ID
> > >     Name (_CID, EisaId ("PNP0A03") /* PCI Bus */)  // _CID: Compatible ID
> > > 
> > >     Method (_CRS, 0, NotSerialized)  // _CRS: Current Resource Settings
> > >     {
> > >         Name (RBUF, ResourceTemplate ()
> > >         {
> > >             Memory32Fixed (ReadWrite,
> > >                 0x60200000,         // Address Base
> > >                 0x01DF0000,         // Address Length
> > >                 )
> > >             WordBusNumber (ResourceProducer, MinFixed, MaxFixed, PosDecode,
> > >                 0x0000,             // Granularity
> > >                 0x0000,             // Range Minimum
> > >                 0x0001,             // Range Maximum
> > >                 0x0000,             // Translation Offset
> > >                 0x0002,             // Length
> > >                 ,, )
> > >         })
> > >         Return (RBUF) /* \_SB_.PCI0._CRS.RBUF */
> > >     }
> > > 
> > > meaning that the memory resources aren't (explicitly) defined as
> > > "producers", i.e. host bridge windows.
> > > 
> > > Commit 8fd4391ee717 ("arm64: PCI: Exclude ACPI "consumer" resources from
> > > host bridge windows") introduced a check that removes such resources,
> > > causing BAR allocation failures later on:
> > > 
> > >     [ 0.150731] pci 0002:00:00.0: BAR 14: no space for [mem size 0x00100000]
> > >     [ 0.150744] pci 0002:00:00.0: BAR 14: failed to assign [mem size 0x00100000]
> > >     [ 0.150758] pci 0002:01:00.0: BAR 0: no space for [mem size 0x00004000 64bit]
> > >     [ 0.150769] pci 0002:01:00.0: BAR 0: failed to assign [mem size 0x00004000 64bit]
> > > 
> > > This eventually prevents the PCIe NVME drive from being accessible.
> > > 
> > > On x86 we already skip the check for producer/window due to some history
> > > with negligent firmware. It seems that Microsoft is intent on continuing
> > > that history on their ARM devices, so let's drop that check here too.
> > > 
> > > Signed-off-by: Maximilian Luz <luzmaximilian@xxxxxxxxx>
> > > ---
> > > 
> > > Please note: I am not sure if this is the right way to fix that, e.g. I
> > > don't know if any additional checks like on IA64 or x86 might be
> > > required instead, or if this might break things on other devices. So
> > > please consider this more as a bug report rather than a fix.
> > > 
> > > Apologies for the re-send, I seem to have unintentionally added a blank
> > > line before the subject.
> > > 
> > > ---
> > >  arch/arm64/kernel/pci.c | 14 --------------
> > >  1 file changed, 14 deletions(-)
> > 
> > Adding Lorenzo to cc, as he'll have a much better idea about this than me.
> > 
> > This is:
> > 
> > https://lore.kernel.org/r/20210510234020.1330087-1-luzmaximilian@xxxxxxxxx
> 
> Sigh. We can't apply this patch since it would trigger regressions on
> other platforms (IIUC the root complex registers would end up in the
> host bridge memory windows).
> 
> I am not keen on reverting commit 8fd4391ee717 because it does the
> right thing.
> 
> I think this requires a quirk and immediate reporting to Microsoft.
> 
> Bjorn, what are your thoughts on this ?

In retrospect, I think 8fd4391ee717 (which I wrote), was probably a
mistake.

Sure, it's a nice idea to have PNP0A03 _CRS methods that work nicely
as designed, by describing host bridge registers as "consumer"
resources and host bridge windows as "producer" registers, instead of
having the bridge registers in _CRS of an unrelated PNP0C02 device.

But realistically, the PNP0A03/PNP0C02 issue is a solved problem, even
though it's ugly, and I'm not sure why I thought Microsoft would see
value in doing this differently on arm64 than on x86 and ia64.

What would break if we reverted 8fd4391ee717?  I guess any arm64
platforms that described host bridge register space in PNP0A03 _CRS
"consumer" resources?  And Windows probably doesn't work or isn't
supported on those platforms?

Bjorn