drm/amdgpu: make compute timeouts consistent
Subsystems (2)
Alex Deucher
maintainerPATCH v2 1/1 drm/amdgpu: make compute timeouts consistent
For kernel compute queues, align the timeout with other kernel queues (10 sec). This had previously been set higher for OpenCL when it used kernel queues, but now OpenCL uses KFD user queues which don't have a timeout limitation. This also aligns with SR-IOV which already used a shorter timeout.
Additionally the longer timeout negatively impacts the user experience with kernel queues for interactive applications.
---
V2: fix documentation as well
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 12 ++---------- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 10 +++++----- 2 files changed, 7 insertions(+), 15 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +2 -10
@@ -4171,18 +4171,10 @@ static int amdgpu_device_get_job_timeout_settings(struct amdgpu_device *adev)int ret =0 ;/* -* By default timeout for non compute jobs is 10000 -* and 60000 for compute jobs. -* In SR-IOV or passthrough mode, timeout for compute -* jobs are 60000 by default. +* By default timeout for jobs is 10 sec */ - adev->gfx_timeout =msecs_to_jiffies (10000 ); + adev->compute_timeout = adev->gfx_timeout =msecs_to_jiffies (10000 ); adev->sdma_timeout = adev->video_timeout = adev->gfx_timeout; -if (amdgpu_sriov_vf (adev)) - adev->compute_timeout =amdgpu_sriov_is_pp_one_vf (adev) ? -msecs_to_jiffies (60000 ) :msecs_to_jiffies (10000 ); -else - adev->compute_timeout =msecs_to_jiffies (60000 );if (strnlen (input,AMDGPU_MAX_TIMEOUT_PARAM_LENGTH )) { while ((timeout_setting = strsep(&input, ",")) &&
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +5 -5
@@ -362,12 +362,12 @@ module_param_named(svm_default_granularity, amdgpu_svm_default_granularity, uint* The second one is for Compute. The third and fourth ones are * for SDMA and Video. * -* By default(with no lockup_timeout settings), the timeout for all non-compute(GFX, SDMA and Video) -* jobs is 10000. The timeout for compute is 60000. +* By default(with no lockup_timeout settings), the timeout for all jobs is 10000. */ -MODULE_PARM_DESC (lockup_timeout,"GPU lockup timeout in ms (default: for bare metal 10000 for non-compute jobs and 60000 for compute jobs; " -"for passthrough or sriov, 10000 for all jobs. 0: keep default value. negative: infinity timeout), format: for bare metal [Non-Compute] or [GFX,Compute,SDMA,Video]; " -"for passthrough or sriov [all jobs] or [GFX,Compute,SDMA,Video]." ); +MODULE_PARM_DESC (lockup_timeout, +"GPU lockup timeout in ms (default: 10000 for all jobs. " +"0: keep default value. negative: infinity timeout), format: for bare metal [Non-Compute] or [GFX,Compute,SDMA,Video]; " +"for passthrough or sriov [all jobs] or [GFX,Compute,SDMA,Video]." );module_param_string (lockup_timeout, amdgpu_lockup_timeout,sizeof (amdgpu_lockup_timeout),0444 );/**
Signature · 1 line
2.49.0
Ping?
Quoted 71 lines · Alex Deucher
For kernel compute queues, align the timeout with other kernel queues (10 sec). This had previously been set higher for OpenCL when it used kernel queues, but now OpenCL uses KFD user queues which don't have a timeout limitation. This also aligns with SR-IOV which already used a shorter timeout.
Additionally the longer timeout negatively impacts the user experience with kernel queues for interactive applications.
Signed-off-by Alex Deucher <[email protected]>---
V2: fix documentation as well
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 12 ++---------- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 10 +++++----- 2 files changed, 7 insertions(+), 15 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +2
-10@@ -4171,18 +4171,10 @@ static int amdgpu_device_get_job_timeout_settings(struct amdgpu_device *adev)int ret =0 ;/* -* By default timeout for non compute jobs is 10000 -* and 60000 for compute jobs. -* In SR-IOV or passthrough mode, timeout for compute -* jobs are 60000 by default. +* By default timeout for jobs is 10 sec */ - adev->gfx_timeout =msecs_to_jiffies (10000 ); + adev->compute_timeout = adev->gfx_timeout =msecs_to_jiffies (10000 ); adev->sdma_timeout = adev->video_timeout = adev->gfx_timeout; -if (amdgpu_sriov_vf (adev)) - adev->compute_timeout =amdgpu_sriov_is_pp_one_vf (adev) ? -msecs_to_jiffies (60000 ) :msecs_to_jiffies (10000 ); -else - adev->compute_timeout =msecs_to_jiffies (60000 );if (strnlen (input,AMDGPU_MAX_TIMEOUT_PARAM_LENGTH )) { while ((timeout_setting = strsep(&input, ",")) &&drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +5
-6@@ -362,12 +362,12 @@ module_param_named(svm_default_granularity, amdgpu_svm_default_granularity, uint* The second one is for Compute. The third and fourth ones are * for SDMA and Video. * -* By default(with no lockup_timeout settings), the timeout for all non-compute(GFX, SDMA and Video) -* jobs is 10000. The timeout for compute is 60000. +* By default(with no lockup_timeout settings), the timeout for all jobs is 10000. */ -MODULE_PARM_DESC (lockup_timeout,"GPU lockup timeout in ms (default: for bare metal 10000 for non-compute jobs and 60000 for compute jobs; " -"for passthrough or sriov, 10000 for all jobs. 0: keep default value. negative: infinity timeout), format: for bare metal [Non-Compute] or [GFX,Compute,SDMA,Video]; " -"for passthrough or sriov [all jobs] or [GFX,Compute,SDMA,Video]." ); +MODULE_PARM_DESC (lockup_timeout, +"GPU lockup timeout in ms (default: 10000 for all jobs. " +"0: keep default value. negative: infinity timeout), format: for bare metal [Non-Compute] or [GFX,Compute,SDMA,Video]; " +"for passthrough or sriov [all jobs] or [GFX,Compute,SDMA,Video]." );module_param_string (lockup_timeout, amdgpu_lockup_timeout,sizeof (amdgpu_lockup_timeout),0444 );/** -- 2.49.0
Russell, Kent
Reviewed-by[Public]
Quoted 93 lines
-----Original Message-----
From: amd-gfx <[email protected]> On Behalf Of Alex Deucher
Sent: Tuesday, July 15, 2025 9:45 AM
To: Deucher, Alexander <[email protected]>
Cc: [email protected]
Subject: Re: [PATCH V2] drm/amdgpu: make compute timeouts consistentPing?
Quoted 15 lines · Alex Deucher
For kernel compute queues, align the timeout with other kernel queues (10 sec). This had previously been set higher for OpenCL when it used kernel queues, but now OpenCL uses KFD user queues which don't have a timeout limitation. This also aligns with SR-IOV which already used a shorter timeout.
Additionally the longer timeout negatively impacts the user experience with kernel queues for interactive applications.
Signed-off-by Alex Deucher <[email protected]>---
Review
Quoted 7 lines
V2: fix documentation as well
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 12 ++---------- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 10 +++++----- 2 files changed, 7 insertions(+), 15 deletions(-)(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.cdrivers/gpu/drm/amd/amdgpu/amdgpu_device.c static int
Quoted 4 lines
index
c8a6b3689deae..58a0ee99eb2f8100644drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +0
-0@@ -4171,18 +4171,10 @@ static int
amdgpu_device_get_job_timeout_settings(struct amdgpu_device *adev)Review
Quoted 21 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +2
-10int ret =0 ;/* -* By default timeout for non compute jobs is 10000 -* and 60000 for compute jobs. -* In SR-IOV or passthrough mode, timeout for compute -* jobs are 60000 by default. +* By default timeout for jobs is 10 sec */ - adev->gfx_timeout =msecs_to_jiffies (10000 ); + adev->compute_timeout = adev->gfx_timeout =msecs_to_jiffies (10000 ); adev->sdma_timeout = adev->video_timeout = adev->gfx_timeout; -if (amdgpu_sriov_vf (adev)) - adev->compute_timeout =amdgpu_sriov_is_pp_one_vf (adev) ? -msecs_to_jiffies (60000 ) :msecs_to_jiffies (10000 ); -else - adev->compute_timeout =msecs_to_jiffies (60000 );if (strnlen (input,AMDGPU_MAX_TIMEOUT_PARAM_LENGTH )) { while ((timeout_setting = strsep(&input, ",")) &&(quoted diff) +0
-0
b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.cdrivers/gpu/drm/amd/amdgpu/amdgpu_drv.c module_param_named(svm_default_granularity,
Quoted 4 lines
index
7e3fa76227033..7bc326d0b4000100644drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +0
-0@@ -362,12 +362,12 @@ module_param_named(svm_default_granularity,
amdgpu_svm_default_granularity, uintdrivers/gpu/drm/amd/amdgpu/amdgpu_drv.c module_param_named(svm_default_granularity, amdgpu_svm_default_granularity, uint
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +0
-1* The second one is for Compute. The third and fourth ones are * for SDMA and Video. * -* By default(with no lockup_timeout settings), the timeout for all non-
compute(GFX, SDMA and Video)drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c module_param_named(svm_default_granularity, amdgpu_svm_default_granularity, uint
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +1
-2-* jobs is 10000. The timeout for compute is 60000. +* By default(with no lockup_timeout settings), the timeout for all jobs is 10000. */ -MODULE_PARM_DESC(lockup_timeout, "GPU lockup timeout in ms (default: forbare metal 10000 for non-compute jobs and 60000 for compute jobs; "
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c module_param_named(svm_default_granularity, · line 369
Quoted 1 line
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +0
-1- "for passthrough or sriov, 10000 for all jobs. 0: keep default value.negative: infinity timeout), format: for bare metal [Non-Compute] or [GFX,Compute,SDMA,Video]; "
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c module_param_named(svm_default_granularity, amdgpu_svm_default_granularity, uint
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +3
-1- "for passthrough or sriov [all jobs] or [GFX,Compute,SDMA,Video]."); +MODULE_PARM_DESC (lockup_timeout, +"GPU lockup timeout in ms (default: 10000 for all jobs. " + "0: keep default value. negative: infinity timeout), format: for bare metal[Non-Compute] or [GFX,Compute,SDMA,Video]; "
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c module_param_named(svm_default_granularity, · line 371
Quoted 2 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +1
-0+ "for passthrough or sriov [all jobs] or [GFX,Compute,SDMA,Video]."); module_param_string(lockup_timeout, amdgpu_lockup_timeout,
sizeof(amdgpu_lockup_timeout), 0444);Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +0
-1/** -- 2.49.0