Hi, On 2/18/26 14:44, Mike Lothian wrote:
This fixes a timeout whilst running GravityMark v1.89 Fullscreen VK RT
I'd argue this doesn't "fix" anything, it papers over the app doing a really long-running submission.
Submissions that run this extremely long are obviously prone to timing out - that's a feature, not a bug. It's likely that this adjustment fixes things for your machine only, and slower GPUs will still hit a timeout. IMHO userspace (that is, Vulkan applications) should be changed to not submit this much work at once, instead.
I'd also argue that increasing the timeout again hurts responsiveness in GPU hang scenarios - you kind of want to get the user back to a functioning session as quickly as possible. The current timeout of 2 seconds has some prior art to it (Windows's TDR timeout is exactly the same), so I don't see much justification in changing anything here.
Thanks, Natalie
Fixes: 1bea57ea7544 ("drm/amdgpu: reduce queue timeout to 2 seconds v2") Signed-off-by: Mike Lothian <[email protected]> --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index c7f44422939f..00f495b8c4f0 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -4204,9 +4204,9 @@ static int amdgpu_device_get_job_timeout_settings(struct amdgpu_device *adev) long timeout; int ret = 0;- /* By default timeout for all queues is 2 sec */+ /* By default timeout for all queues is 3 sec */ adev->gfx_timeout = adev->compute_timeout = adev->sdma_timeout = - adev->video_timeout = msecs_to_jiffies(2000); + adev->video_timeout = msecs_to_jiffies(3000);if (!strnlen(input, AMDGPU_MAX_TIMEOUT_PARAM_LENGTH))return 0;
