Re: [PATCH] ACPI: SPCR: Use access width to determine mmio usage

[Jon Mason]

On Mon, May 8, 2017 at 3:57 PM, Loc Ho <lho@xxxxxxx> wrote:
> Hi Jon,
>
>>>
>>>>> The current SPCR code does not check the access width of the mmio, and
>>>>> uses a default of 8bit register accesses.  This prevents devices that
>>>>> only do 16 or 32bit register accesses from working.  By simply checking
>>>>> this field and setting the mmio string appropriately, this issue can be
>>>>> corrected.  To prevent any legacy issues, the code will default to 8bit
>>>>> accesses if the value is anything but 16 or 32.
>>>>
>>>> Thanks for this. Just as an FYI I've a running discussion with Microsoft
>>>> about defining additional UART subtypes in the DBG2 for special case
>>>> UARTs. Specifically, I want to address AppliedMicro's special 8250 dw IP
>>>> that also has a non-standard clock. At this time, there is general
>>>> agreement to use the access width for some cases rather than defining
>>>> yet more subtypes - so your patch is good.
>>>>
>>>> Loc/Applied: please track this thread, incorporate feedback, and also
>>>> track the other general recent discussions of 8250 dw from this week.
>>>
>>> Thanks for forward me this patch. This patch does not work with X-Gene
>>> v1 and v2 SoC's. As BIOS SPCR encodes these info as:
>>>
>>> Bit Width: 32
>>> Bit Offset: 0
>>> Encoded Access Width: 01 (Byte Access)
>>>
>>> With this patch, it would use the "mmio" instead the "mmio32" as with
>>> this patch - https://patchwork.kernel.org/patch/9460959
>>
>> I think this is why we need the DBG2 subtype for Applied X-Gene1. I'm
>> hoping the update to the SPCR/DBG2 spec is done soon.
>
> We can't rely on the BIOS change to support this new subtype as we
> have system that is already in production deployment. When these
> system upgrade to new version of the OS (stock, RHELSA, or whatever),
> they will break. We need the patch from
> https://patchwork.kernel.org/patch/9460959/ rolled upstream.

There is no reason why the patch you reference cannot co-exist with
the one I am submitting here.  In this case, my patch would set it to
mmio, then the patch you link above would reset it to mmio32.
Personally, I would recommend a big, fat comment on why this extra
step is necessary, but it should work as desired.  Alternatively, we
could add some kind of quirk library (similar to
qdf2400_erratum_44_present) where the OEM/OEM Table ID is referenced
and workaround applied.  Thoughts?

Thanks,
Jon Mas(on)


>
> -Loc