Re: [PATCH 4/4] drm/i915/display: use fixed 10 second timeout for fence wait

Hello,

On 8/19/26 16:05, Jani Nikula wrote:
> i915 has a configurable fence wait timeout, defaulting to 10 seconds,
> and xe uses MAX_SCHEDULE_TIMEOUT i.e. it has no timeout.
>
> Use a fixed 10 second timeout for both drivers instead of having them do
> different things. Ditch the now unused compat header from xe.
>
> i915 actually needs the timeout to deal with CI reset tests, but there
> shouldn't be any harm in having a 10 second timeout in xe either.
>
> Cc: Ville Syrjälä <[email protected]>
> Cc: Maarten Lankhorst <[email protected]>
> Signed-off-by: Jani Nikula <[email protected]>
>
> ---
>
> There have been a number of stabs at this in the past (see links below),
> and all of them have been rejected with one reason or another.
>
> My primary goal here is just to remove the annoying i915_fence_timeout()
> usage and i915 vs. xe difference. Further cleanup can come later (or
> never, as it usually is).
>
> https://lore.kernel.org/r/[email protected]
> https://lore.kernel.org/r/[email protected]
> https://lore.kernel.org/r/8cde57236982fc3ca1a5608332943d3b9274a45a.1775661609.git.jani.nikula@intel.com
> https://lore.kernel.org/r/[email protected]
> ---
>  drivers/gpu/drm/i915/display/intel_display.c     |  4 +---
>  .../gpu/drm/xe/compat-i915-headers/i915_config.h | 16 ----------------
>  2 files changed, 1 insertion(+), 19 deletions(-)
>  delete mode 100644 drivers/gpu/drm/xe/compat-i915-headers/i915_config.h
>
> diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
> index aa5d41dac8a9..c3a11e9e69e2 100644
> --- a/drivers/gpu/drm/i915/display/intel_display.c
> +++ b/drivers/gpu/drm/i915/display/intel_display.c
> @@ -50,7 +50,6 @@
>  #include "g4x_dp.h"
>  #include "g4x_hdmi.h"
>  #include "hsw_ips.h"
> -#include "i915_config.h"
>  #include "i9xx_plane.h"
>  #include "i9xx_plane_regs.h"
>  #include "i9xx_wm.h"
> @@ -7284,8 +7283,7 @@ static void intel_atomic_commit_fence_wait(struct intel_atomic_state *state)
>  		if (!new_plane_state->fence)
>  			continue;
>  
> -		ret = dma_fence_wait_timeout(new_plane_state->fence, false,
> -					     i915_fence_timeout());
> +		ret = dma_fence_wait_timeout(new_plane_state->fence, false, 10 * HZ);
>  		if (!ret)
>  			ret = -ETIMEDOUT;
>  		if (ret < 0) {
>
Not waiting on fences until they're signaled is an existing issue in i915, and was added as a workaround for a deadlock when GPU reset has to reset display as well on old platforms. This requires a modeset to disable everything, which may hang since it could wait on a fence that may only be signaled after gpu reset completes.

Can you test if gpu_reset_clobbers_display and force_reset_modeset_test are set, and if not set the timeout to infinity on i915 as well?

Kind regards,
~Maarten Lankhorst