[PATCH] drm/radeon: Always flush VM again on < CIK

[Alex Deucher]

On Thu, Aug 7, 2014 at 11:38 AM, Marek Olšák <maraeo at gmail.com> wrote:
> So what's difference between WRITE_DATA with PFP vs ME? Would it also
> be preferable for DMA_DATA and COPY_DATA?

The PFP comes before the ME in the pipeline.  Note that there is no
PFP (or CE) on the compute queues so we can't use PFP (or CE) for
compute.  According to the internal gfx teams, we should use PFP
whenever possible since the PFP is rarely as busy as the ME.  Note
also that the engine bit is not always consistent (for some packets 0
= ME, 1 = PFP and for others 1= ME and 0 = PFP).

Alex

>
> Marek
>
> On Thu, Aug 7, 2014 at 3:59 PM, Alex Deucher <alexdeucher at gmail.com> wrote:
>> On Thu, Aug 7, 2014 at 3:46 AM, Michel Dänzer <michel at daenzer.net> wrote:
>>> From: Michel Dänzer <michel.daenzer at amd.com>
>>>
>>> Not doing this causes piglit hangs[0] on my Cape Verde card. No issues on
>>> Bonaire and Kaveri though.
>>>
>>> [0] Same symptoms as those fixed on CIK by 'drm/radeon: set VM base addr
>>> using the PFP v2'.
>>>
>>> Signed-off-by: Michel Dänzer <michel.daenzer at amd.com>
>>
>> We should be using PFP as much as possible.  Does the attached patch help?
>>
>> Alex
>>
>>> ---
>>>  drivers/gpu/drm/radeon/radeon_vm.c | 4 +++-
>>>  1 file changed, 3 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/gpu/drm/radeon/radeon_vm.c b/drivers/gpu/drm/radeon/radeon_vm.c
>>> index ccae4d9..898cbb7 100644
>>> --- a/drivers/gpu/drm/radeon/radeon_vm.c
>>> +++ b/drivers/gpu/drm/radeon/radeon_vm.c
>>> @@ -238,7 +238,9 @@ void radeon_vm_flush(struct radeon_device *rdev,
>>>         uint64_t pd_addr = radeon_bo_gpu_offset(vm->page_directory);
>>>
>>>         /* if we can't remember our last VM flush then flush now! */
>>> -       if (!vm->last_flush || pd_addr != vm->pd_gpu_addr) {
>>> +       /* XXX figure out why we have to flush all the time before CIK */
>>> +       if (rdev->family < CHIP_BONAIRE ||
>>> +           !vm->last_flush || pd_addr != vm->pd_gpu_addr) {
>>>                 trace_radeon_vm_flush(pd_addr, ring, vm->id);
>>>                 vm->pd_gpu_addr = pd_addr;
>>>                 radeon_ring_vm_flush(rdev, ring, vm);
>>> --
>>> 2.0.1
>>>
>>> _______________________________________________
>>> dri-devel mailing list
>>> dri-devel at lists.freedesktop.org
>>> http://lists.freedesktop.org/mailman/listinfo/dri-devel
>>
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel at lists.freedesktop.org
>> http://lists.freedesktop.org/mailman/listinfo/dri-devel
>>