[PATCH libdrm] xf86drm: Parse the separate files to retrieve the vendor/device info

[Emil Velikov]

On 7 November 2016 at 09:14, Michel Dänzer <michel at daenzer.net> wrote:
> On 05/11/16 03:14 AM, Emil Velikov wrote:
>> On 2 November 2016 at 03:07, Michel Dänzer <michel at daenzer.net> wrote:
>>>
>>> The first attached patch will result in drmParsePciDeviceInfo always
>>> reporting revision 0 on kernels without the second attached patch. Will
>>> that be an issue for the amdgpu-pro stack?
>>>
>>> Please follow up directly to the patch e-mails with any comments on the
>>> patches.
>>>
>> Fleshing out the question from the actual patches:
>>
>> Do the AMDGPU-PRO or the AMD stack [as a whole] depend on the revision
>> field as returned by the drmDevice API ?
>
> One answer is that https://patchwork.freedesktop.org/patch/120132/ uses
Nice, I really like this ! Wonder if others will do the same, rather
than duplicating it throughout ddx/mesa/other drivers. That aside,

> the revision ID. In this case a wrong revision ID would only cause a
> cosmetic issue, but I can imagine that other code in the amdgpu-pro
> stack really needs the correct revision ID to accurately identify the GPU.
>
Don't mean to sound rude, but I was hoping for a definite answer.
Regardless, do you/fellow AMD devs, any preference on how to go with
this bug [1] ?

Add an override to force use of the revision file - be that envvar,
new API {drmDeviceUseRevisionFile, drmDevice...v2}, or revert the 12 +
commits (pulling only the offending one won't cut it). Obviously I'm
not a huge fan of the last one :-\

[1] https://bugs.freedesktop.org/show_bug.cgi?id=98502

>
>> Since we have a lovely bug in libdrm and might roll out a release
>> soonish it'll be great to have this squashed/merged as well.
>
> I hope the release can wait for the patch above to land as well.
>
Atm, we crash for anyone using !pci devices, so I'd like to spare
that. So it'll be great if this lands in the next days/week.

Thanks
Emil