[PATCH] drm/radeon/uvd: revert lower msg&fb buffer requirements on UVD3

Christian König deathsimple at vodafone.de
Wed Oct 16 09:15:59 CEST 2013


Am 16.10.2013 00:05, schrieb Grigori Goronzy:
> On 15.10.2013 20:12, Christian König wrote:
>> From: Christian König <christian.koenig at amd.com>
>>
>> This only seem to work for H.264 but not for VC-1 streams.
>>
>> Need to investigate further why exactly.
>>
>
> Thanks for the quick investigation. This is really strange. The 
> corruption looks very similar to what I saw with VC-1 simple/main 
> earlier, before it was disabled. Could there be some kind of connection?
>

Codecs usually work from top left to bottom right. What happens here is 
that the decoding process initially works fine and then hits a point 
where it needs a parameter from the message buffer. This parameter is 
missing/incorrect because it uses the wrong segment identifier for the 
memory access.

So decoding fails and we only end up with a few correct pixels at the 
top and only garbage in the rest of the image. The same thing currently 
happens with MPEG1 streams as well.

It's not really related apart from the fact that all failed decoding 
processes look more or less the same.

Christian.

> Grigori
>
>> This reverts commit 4b40e5921230beb1951f04d2b1b92c4c88fbad43.
>>
>> Signed-off-by: Christian König <christian.koenig at amd.com>
>> ---
>>   drivers/gpu/drm/radeon/radeon_cs.c  | 3 +--
>>   drivers/gpu/drm/radeon/radeon_uvd.c | 3 ++-
>>   drivers/gpu/drm/radeon/uvd_v1_0.c   | 4 ++--
>>   3 files changed, 5 insertions(+), 5 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/radeon/radeon_cs.c 
>> b/drivers/gpu/drm/radeon/radeon_cs.c
>> index 66c2228..80285e3 100644
>> --- a/drivers/gpu/drm/radeon/radeon_cs.c
>> +++ b/drivers/gpu/drm/radeon/radeon_cs.c
>> @@ -85,9 +85,8 @@ static int radeon_cs_parser_relocs(struct 
>> radeon_cs_parser *p)
>>              VRAM, also but everything into VRAM on AGP cards to avoid
>>              image corruptions */
>>           if (p->ring == R600_RING_TYPE_UVD_INDEX &&
>> -            p->rdev->family < CHIP_PALM &&
>>               (i == 0 || drm_pci_device_is_agp(p->rdev->ddev))) {
>> -
>> +            /* TODO: is this still needed for NI+ ? */
>>               p->relocs[i].lobj.domain =
>>                   RADEON_GEM_DOMAIN_VRAM;
>>
>> diff --git a/drivers/gpu/drm/radeon/radeon_uvd.c 
>> b/drivers/gpu/drm/radeon/radeon_uvd.c
>> index 4f2e73f..308eff5 100644
>> --- a/drivers/gpu/drm/radeon/radeon_uvd.c
>> +++ b/drivers/gpu/drm/radeon/radeon_uvd.c
>> @@ -476,7 +476,8 @@ static int radeon_uvd_cs_reloc(struct 
>> radeon_cs_parser *p,
>>           return -EINVAL;
>>       }
>>
>> -    if (p->rdev->family < CHIP_PALM && (cmd == 0 || cmd == 0x3) &&
>> +    /* TODO: is this still necessary on NI+ ? */
>> +    if ((cmd == 0 || cmd == 0x3) &&
>>           (start >> 28) != (p->rdev->uvd.gpu_addr >> 28)) {
>>           DRM_ERROR("msg/fb buffer %LX-%LX out of 256MB segment!\n",
>>                 start, end);
>> diff --git a/drivers/gpu/drm/radeon/uvd_v1_0.c 
>> b/drivers/gpu/drm/radeon/uvd_v1_0.c
>> index 3100fa9..7266805 100644
>> --- a/drivers/gpu/drm/radeon/uvd_v1_0.c
>> +++ b/drivers/gpu/drm/radeon/uvd_v1_0.c
>> @@ -212,8 +212,8 @@ int uvd_v1_0_start(struct radeon_device *rdev)
>>       /* enable VCPU clock */
>>       WREG32(UVD_VCPU_CNTL,  1 << 9);
>>
>> -    /* enable UMC and NC0 */
>> -    WREG32_P(UVD_LMI_CTRL2, 1 << 13, ~((1 << 8) | (1 << 13)));
>> +    /* enable UMC */
>> +    WREG32_P(UVD_LMI_CTRL2, 0, ~(1 << 8));
>>
>>       /* boot up the VCPU */
>>       WREG32(UVD_SOFT_RESET, 0);
>>
>



More information about the dri-devel mailing list