[Intel-gfx] [PATCH 2/8] drm/i915: Don't disable interrupts on PREEMPT_RT during atomic updates

[Ville Syrjälä]

On Tue, Oct 05, 2021 at 05:00:40PM +0200, Sebastian Andrzej Siewior wrote:
> From: Mike Galbraith <umgwanakikbuti at gmail.com>
> 
> Commit
>    8d7849db3eab7 ("drm/i915: Make sprite updates atomic")
> 
> started disabling interrupts across atomic updates. This breaks on PREEMPT_RT
> because within this section the code attempt to acquire spinlock_t locks which
> are sleeping locks on PREEMPT_RT.
> 
> According to the comment the interrupts are disabled to avoid random delays and
> not required for protection or synchronisation.
> If this needs to happen with disabled interrupts on PREEMPT_RT, and the
> whole section is restricted to register access then all sleeping locks
> need to be acquired before interrupts are disabled and some function
> maybe moved after enabling interrupts again.
> This includes:
> - prepare_to_wait() + finish_wait() due its wake queue.
> - drm_crtc_vblank_put() -> vblank_disable_fn() drm_device::vbl_lock.
> - skl_pfit_enable(), intel_update_plane(), vlv_atomic_update_fifo() and
>   maybe others due to intel_uncore::lock
> - drm_crtc_arm_vblank_event() due to drm_device::event_lock and
>   drm_device::vblank_time_lock.
> 
> Don't disable interrupts on PREEMPT_RT during atomic updates.
> 
> [bigeasy: drop local locks, commit message]

I have my doubts about meething the deadlines here.
CONFIG_DRM_I915_DEBUG_VBLANK_EVADE should scream a bunch if it
looks like we're missing it.

That said, we already miss the deadline sometimes, esp. with
lockdep and whatnot enabled which makes the locking very expensive.

Also some ideas how to reduce the overhead:
- Try to make the mmio accesses lockless as much s possible
- Reduce the amount of work we do in the critical section

Anyways, RT hasn't really been on anyone's radar so no one
has yet spent significant amount of brain cells on this.

> 
> Signed-off-by: Mike Galbraith <umgwanakikbuti at gmail.com>
> Signed-off-by: Sebastian Andrzej Siewior <bigeasy at linutronix.de>
> ---
>  drivers/gpu/drm/i915/display/intel_crtc.c | 15 ++++++++++-----
>  1 file changed, 10 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/display/intel_crtc.c b/drivers/gpu/drm/i915/display/intel_crtc.c
> index 254e67141a776..7a39029b083f4 100644
> --- a/drivers/gpu/drm/i915/display/intel_crtc.c
> +++ b/drivers/gpu/drm/i915/display/intel_crtc.c
> @@ -425,7 +425,8 @@ void intel_pipe_update_start(const struct intel_crtc_state *new_crtc_state)
>  	 */
>  	intel_psr_wait_for_idle(new_crtc_state);
>  
> -	local_irq_disable();
> +	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
> +		local_irq_disable();
>  
>  	crtc->debug.min_vbl = min;
>  	crtc->debug.max_vbl = max;
> @@ -450,11 +451,13 @@ void intel_pipe_update_start(const struct intel_crtc_state *new_crtc_state)
>  			break;
>  		}
>  
> -		local_irq_enable();
> +		if (!IS_ENABLED(CONFIG_PREEMPT_RT))
> +			local_irq_enable();
>  
>  		timeout = schedule_timeout(timeout);
>  
> -		local_irq_disable();
> +		if (!IS_ENABLED(CONFIG_PREEMPT_RT))
> +			local_irq_disable();
>  	}
>  
>  	finish_wait(wq, &wait);
> @@ -487,7 +490,8 @@ void intel_pipe_update_start(const struct intel_crtc_state *new_crtc_state)
>  	return;
>  
>  irq_disable:
> -	local_irq_disable();
> +	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
> +		local_irq_disable();
>  }
>  
>  #if IS_ENABLED(CONFIG_DRM_I915_DEBUG_VBLANK_EVADE)
> @@ -566,7 +570,8 @@ void intel_pipe_update_end(struct intel_crtc_state *new_crtc_state)
>  		new_crtc_state->uapi.event = NULL;
>  	}
>  
> -	local_irq_enable();
> +	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
> +		local_irq_enable();
>  
>  	/* Send VRR Push to terminate Vblank */
>  	intel_vrr_send_push(new_crtc_state);
> -- 
> 2.33.0

-- 
Ville Syrjälä
Intel