On Mon, Jun 20, 2016 at 9:27 AM, Christian König
<deathsim...@vodafone.de> wrote:
> Am 20.06.2016 um 16:13 schrieb Rob Herring:
>>
>> On Mon, Jun 20, 2016 at 8:31 AM, Nicolai Hähnle <nhaeh...@gmail.com>
>> wrote:
>>>
>>> On 17.06.2016 21:05, Rob Herring wrote:
>>>>
>>>> On Fri, Jun 17, 2016 at 1:45 PM, Emil Velikov <emil.l.veli...@gmail.com>
>>>> wrote:
>>>>>
>>>>> On 17 June 2016 at 18:45, Rob Herring <r...@kernel.org> wrote:
>>>>>>
>>>>>> Now that the pipe-loader is reference counting the screen creation, it
>>>>>> is unnecessary to do in it the winsys/driver.
>>>>
>>>>
>>>> [...]
>>>>
>>>>>> -static unsigned hash_dev(void *key)
>>>>>> -{
>>>>>> -#if defined(PIPE_ARCH_X86_64)
>>>>>> - return pointer_to_intptr(key) ^ (pointer_to_intptr(key) >> 32);
>>>>>> -#else
>>>>>> - return pointer_to_intptr(key);
>>>>>> -#endif
>>>>>> -}
>>>>>> -
>>>>>
>>>>> As you can see above the hashing algo is different for AMDGPU. Not
>>>>> familiar with the story behind any of this, so hopefully the AMD folk
>>>>> will give you some insights.
>>>>
>>>>
>>>> They are also hashing the fd in libdrm amdgpu_device_initialize(), so
>>>> I thought this was redundant (unless you have an old libdrm).
>>>
>>>
>>> Correct. The idea is that there may be non-Mesa users of libdrm_amdgpu
>>> (in
>>> particular, the amdgpu-pro stack).
>>>
>>> The hashing in libdrm is different as well, though, and apparently on
>>> purpose, see commit c68d58aa14b7f of libdrm. This might mean the series
>>> doesn't work as is. Christian or Marek might know more.
>>
>> Okay, so the screen needs to be re-used across card and render nodes.
>> Seems like other drivers could need that behavior as well and I should
>> use the libdrm version.
>
>
> Well not necessarily. Reusing the screen between cards and render nodes only
> works if the driver can deal with that.
>
> E.g. for amdgpu we ran into the situation that you can create the screen
> initially only with a render node and then get a DRI2 authenticated fd later
> on.
>
> What we do in libdrm is to remember the DRI2 authenticated fd separately to
> make use of it for flink import/exports. I fear that your cleanup would
> break that behavior.
Okay, I'll leave the amdgpu hashing code in place. AFAICT, the
pipe-loader hashing should not impact it. It will only mean that
amdgpu_device_initialize is called once per node.
Rob
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
