[PATCH V2] drm/amdgpu: Fix SDMA UTC_L1 handling during start/stop sequences

Lazar, Lijo lijo.lazar at amd.com
Mon Jun 16 15:04:51 UTC 2025



On 6/16/2025 8:28 PM, Alex Deucher wrote:
> On Mon, Jun 16, 2025 at 10:47 AM Jesse Zhang <jesse.zhang at amd.com> wrote:
>>
>> This commit makes two key fixes to SDMA v4.4.2 handling:
>>
>> 1. disable UTC_L1 in sdma_cntl register when stopping SDMA engines
>>    by reading the current value before modifying UTC_L1_ENABLE bit.
>>
>> 2. Ensure UTC_L1_ENABLE is consistently managed by:
>>    - Adding the missing register write when enabling UTC_L1 during start
>>    - Keeping UTC_L1 enabled by default as per hardware requirements
>>
>> Fixes the commit 7f03b1d14d51 ("drm/amdgpu:
>> Remove sdma halt/unhalt during frontdoor load") by maintaining proper
>> SDMA controller state during initialization sequences.
> 
> Proper fixes tag formatting:
> 
> Fixes: 7f03b1d14d51 ("drm/amdgpu: Remove sdma halt/unhalt during
> frontdoor load")
> 

The default reset value of that field is 1. Keeping SDMA unhalted but
rings disabled shouldn't ideally be a problem - should be similar to the
case when it's loaded also.

Thanks,
Lijo

>>
>> v2: Correct SDMA_CNTL setting (Philip)
>>
>> Suggested-by: Jonathan Kim <jonathan.kim at amd.com>
>> Signed-off-by: Jesse Zhang <Jesse.Zhang at amd.com>
> 
> Acked-by: Alex Deucher <alexander.deucher at amd.com>
> 
>> ---
>>  drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c | 6 +++++-
>>  1 file changed, 5 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c b/drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c
>> index e6af50c947a1..52157acab55a 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c
>> @@ -492,7 +492,7 @@ static void sdma_v4_4_2_inst_gfx_stop(struct amdgpu_device *adev,
>>  {
>>         struct amdgpu_ring *sdma[AMDGPU_MAX_SDMA_INSTANCES];
>>         u32 doorbell_offset, doorbell;
>> -       u32 rb_cntl, ib_cntl;
>> +       u32 rb_cntl, ib_cntl, sdma_cntl;
>>         int i;
>>
>>         for_each_inst(i, inst_mask) {
>> @@ -504,6 +504,9 @@ static void sdma_v4_4_2_inst_gfx_stop(struct amdgpu_device *adev,
>>                 ib_cntl = RREG32_SDMA(i, regSDMA_GFX_IB_CNTL);
>>                 ib_cntl = REG_SET_FIELD(ib_cntl, SDMA_GFX_IB_CNTL, IB_ENABLE, 0);
>>                 WREG32_SDMA(i, regSDMA_GFX_IB_CNTL, ib_cntl);
>> +               sdma_cntl = RREG32_SDMA(i, SDMA_CNTL);
>> +               sdma_cntl = REG_SET_FIELD(sdma_cntl, SDMA_CNTL, UTC_L1_ENABLE, 0);
>> +               WREG32_SDMA(i, regSDMA_CNTL, sdma_cntl);
>>
>>                 if (sdma[i]->use_doorbell) {
>>                         doorbell = RREG32_SDMA(i, regSDMA_GFX_DOORBELL);
>> @@ -997,6 +1000,7 @@ static int sdma_v4_4_2_inst_start(struct amdgpu_device *adev,
>>                 /* set utc l1 enable flag always to 1 */
>>                 temp = RREG32_SDMA(i, regSDMA_CNTL);
>>                 temp = REG_SET_FIELD(temp, SDMA_CNTL, UTC_L1_ENABLE, 1);
>> +               WREG32_SDMA(i, regSDMA_CNTL, temp);
>>
>>                 if (amdgpu_ip_version(adev, SDMA0_HWIP, 0) < IP_VERSION(4, 4, 5)) {
>>                         /* enable context empty interrupt during initialization */
>> --
>> 2.34.1
>>



More information about the amd-gfx mailing list