On 24.09.25 01:00, [email protected] wrote:
> On Tue, 2025-09-23 at 15:10 +0200, Christian König wrote:
>> The Constant Engine found on gfx6-gfx10 HW has been a notorious
>> source of
>> problems.
>>
>> RADV never used it in the first place, radeonsi only used it for a
>> few
>> releases around 2017 for gfx6-gfx9 before dropping support for it as
>> well.
>>
>> While investigating another problem I just recently found that
>> submitting
>> to the CE seems to be completely broken on gfx9 for quite a while.
>>
>> Since nobody complained about that problem it most likely means that
>> nobody is using any of the affected radeonsi versions on current
>> Linux
>> kernels any more.
>>
>> So to potentially phase out the support for the CE and eliminate
>> another
>> source of problems block submitting CE IBs unless it is enabled again
>> using a debug flag.
>>
>> Signed-off-by: Christian König <[email protected]>
>
> Acked-by: Timur Kristóf <[email protected]>
>
> Hi Christian,
>
> Would you be open to receiving a patch to stop emitting the CE related
> workarounds when the CE is not enabled?
>
> Alternatively, could we stop emitting them altogether now that the CE
> is disabled by default?
Not yet. I want to push that upstream, wait for quite a while and when nobody
complains just completely remove the CE support including all the extra
overhead we currently do in the submission path for it.
> Also, should the new debug flag be documented?
Where should we put that? I already noted how to enable it again in the
ratelimited error message printed when you try to use the CE.
Regards,
Christian.
>
> Thanks & best regards,
> Timur
>
>
>> ---
>> drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 +
>> drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 6 ++++++
>> drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 8 +++++++-
>> 3 files changed, 14 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> index 2a0df4cabb99..6f5b4a0e0a34 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> @@ -1290,6 +1290,7 @@ struct amdgpu_device {
>> bool
>> debug_disable_gpu_ring_reset;
>> bool debug_vm_userptr;
>> bool debug_disable_ce_logs;
>> + bool debug_enable_ce_cs;
>>
>> /* Protection for the following isolation structure */
>> struct mutex enforce_isolation_mutex;
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>> index 744e6ff69814..322890e2c899 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
>> @@ -364,6 +364,12 @@ static int amdgpu_cs_p2_ib(struct
>> amdgpu_cs_parser *p,
>> if (p->uf_bo && ring->funcs->no_user_fence)
>> return -EINVAL;
>>
>> + if (!p->adev->debug_enable_ce_cs &&
>> + chunk_ib->flags & AMDGPU_IB_FLAG_CE) {
>> + dev_err_ratelimited(p->adev->dev, "CE CS is blocked,
>> use debug=0x400 to override\n");
>> + return -EINVAL;
>> + }
>> +
>> if (chunk_ib->ip_type == AMDGPU_HW_IP_GFX &&
>> chunk_ib->flags & AMDGPU_IB_FLAG_PREEMPT) {
>> if (chunk_ib->flags & AMDGPU_IB_FLAG_CE)
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>> index ece251cbe8c3..3b3fc734c0f8 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>> @@ -144,7 +144,8 @@ enum AMDGPU_DEBUG_MASK {
>> AMDGPU_DEBUG_DISABLE_GPU_RING_RESET = BIT(6),
>> AMDGPU_DEBUG_SMU_POOL = BIT(7),
>> AMDGPU_DEBUG_VM_USERPTR = BIT(8),
>> - AMDGPU_DEBUG_DISABLE_RAS_CE_LOG = BIT(9)
>> + AMDGPU_DEBUG_DISABLE_RAS_CE_LOG = BIT(9),
>> + AMDGPU_DEBUG_ENABLE_CE_CS = BIT(10)
>> };
>>
>> unsigned int amdgpu_vram_limit = UINT_MAX;
>> @@ -2289,6 +2290,11 @@ static void amdgpu_init_debug_options(struct
>> amdgpu_device *adev)
>> pr_info("debug: disable kernel logs of correctable
>> errors\n");
>> adev->debug_disable_ce_logs = true;
>> }
>> +
>> + if (amdgpu_debug_mask & AMDGPU_DEBUG_ENABLE_CE_CS) {
>> + pr_info("debug: allowing command submission to CE
>> engine\n");
>> + adev->debug_enable_ce_cs = true;
>> + }
>> }
>>
>> static unsigned long amdgpu_fix_asic_type(struct pci_dev *pdev,
>> unsigned long flags)
