[PATCH] drm/radeon: Sync ME and PFP after CP semaphore waits v2

[Christian König]

Am 18.08.2014 um 17:02 schrieb Alex Deucher:
> On Mon, Aug 18, 2014 at 10:30 AM, Christian König
> <deathsimple at vodafone.de> wrote:
>> From: Christian König <christian.koenig at amd.com>
>>
>> Fixes lockups due to CP read GPUVM faults when running piglit on Cape
>> Verde.
>>
>> v2 (chk): apply the fix to R600+ as well, on CIK only the GFX CP has
>>            a PFP, add more comments to R600 code, enable flushing again
>>
>> Signed-off-by: Michel Dänzer <michel.daenzer at amd.com>
>> Signed-off-by: Christian König <christian.koenig at amd.com>
> I just checked the ucode.  This packet only exists on 7xx and newer.
> How about the attached change?

Works for me as well.

On a side note: Where did you find the documentation if that's present 
on R600? I could only confirm that it works on RV710 and so just assumed 
that it works on R6XX the same way.

Thanks for double checking that,
Christian.

>
> Alex
>
>> ---
>>   drivers/gpu/drm/radeon/cik.c       | 17 +++++++++++++++++
>>   drivers/gpu/drm/radeon/r600.c      | 17 +++++++++++++++++
>>   drivers/gpu/drm/radeon/r600d.h     |  1 +
>>   drivers/gpu/drm/radeon/radeon_vm.c |  4 +---
>>   4 files changed, 36 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/radeon/cik.c b/drivers/gpu/drm/radeon/cik.c
>> index e7d99e1..8b556f2 100644
>> --- a/drivers/gpu/drm/radeon/cik.c
>> +++ b/drivers/gpu/drm/radeon/cik.c
>> @@ -3920,6 +3920,17 @@ void cik_fence_compute_ring_emit(struct radeon_device *rdev,
>>          radeon_ring_write(ring, 0);
>>   }
>>
>> +/**
>> + * cik_semaphore_ring_emit - emit a semaphore on the CP ring
>> + *
>> + * @rdev: radeon_device pointer
>> + * @ring: radeon ring buffer object
>> + * @semaphore: radeon semaphore object
>> + * @emit_wait: Is this a sempahore wait?
>> + *
>> + * Emits a semaphore signal/wait packet to the CP ring and prevents the PFP
>> + * from running ahead of semaphore waits.
>> + */
>>   bool cik_semaphore_ring_emit(struct radeon_device *rdev,
>>                               struct radeon_ring *ring,
>>                               struct radeon_semaphore *semaphore,
>> @@ -3932,6 +3943,12 @@ bool cik_semaphore_ring_emit(struct radeon_device *rdev,
>>          radeon_ring_write(ring, lower_32_bits(addr));
>>          radeon_ring_write(ring, (upper_32_bits(addr) & 0xffff) | sel);
>>
>> +       if (emit_wait && ring->idx == RADEON_RING_TYPE_GFX_INDEX) {
>> +               /* Prevent the PFP from running ahead of the semaphore wait */
>> +               radeon_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
>> +               radeon_ring_write(ring, 0x0);
>> +       }
>> +
>>          return true;
>>   }
>>
>> diff --git a/drivers/gpu/drm/radeon/r600.c b/drivers/gpu/drm/radeon/r600.c
>> index c70a504..d659fa5 100644
>> --- a/drivers/gpu/drm/radeon/r600.c
>> +++ b/drivers/gpu/drm/radeon/r600.c
>> @@ -2753,6 +2753,17 @@ void r600_fence_ring_emit(struct radeon_device *rdev,
>>          }
>>   }
>>
>> +/**
>> + * r600_semaphore_ring_emit - emit a semaphore on the CP ring
>> + *
>> + * @rdev: radeon_device pointer
>> + * @ring: radeon ring buffer object
>> + * @semaphore: radeon semaphore object
>> + * @emit_wait: Is this a sempahore wait?
>> + *
>> + * Emits a semaphore signal/wait packet to the CP ring and prevents the PFP
>> + * from running ahead of semaphore waits.
>> + */
>>   bool r600_semaphore_ring_emit(struct radeon_device *rdev,
>>                                struct radeon_ring *ring,
>>                                struct radeon_semaphore *semaphore,
>> @@ -2768,6 +2779,12 @@ bool r600_semaphore_ring_emit(struct radeon_device *rdev,
>>          radeon_ring_write(ring, lower_32_bits(addr));
>>          radeon_ring_write(ring, (upper_32_bits(addr) & 0xff) | sel);
>>
>> +       if (emit_wait) {
>> +               /* Prevent the PFP from running ahead of the semaphore wait */
>> +               radeon_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
>> +               radeon_ring_write(ring, 0x0);
>> +       }
>> +
>>          return true;
>>   }
>>
>> diff --git a/drivers/gpu/drm/radeon/r600d.h b/drivers/gpu/drm/radeon/r600d.h
>> index f94e7a9..1272c0d 100644
>> --- a/drivers/gpu/drm/radeon/r600d.h
>> +++ b/drivers/gpu/drm/radeon/r600d.h
>> @@ -1597,6 +1597,7 @@
>>                   */
>>   #              define PACKET3_CP_DMA_CMD_SAIC      (1 << 28)
>>   #              define PACKET3_CP_DMA_CMD_DAIC      (1 << 29)
>> +#define PACKET3_PFP_SYNC_ME                            0x42
>>   #define        PACKET3_SURFACE_SYNC                            0x43
>>   #              define PACKET3_CB0_DEST_BASE_ENA    (1 << 6)
>>   #              define PACKET3_FULL_CACHE_ENA       (1 << 20) /* r7xx+ only */
>> diff --git a/drivers/gpu/drm/radeon/radeon_vm.c b/drivers/gpu/drm/radeon/radeon_vm.c
>> index 058f200..9c8358f 100644
>> --- a/drivers/gpu/drm/radeon/radeon_vm.c
>> +++ b/drivers/gpu/drm/radeon/radeon_vm.c
>> @@ -238,9 +238,7 @@ 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! */
>> -       /* 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) {
>> +       if (!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);
>> --
>> 1.9.1
>>