[PATCH] drm/omap: fix incorrect lock state

[Laurent Pinchart]

Hi Tomi,

Thank you for the patch.

On Wed, Aug 19, 2020 at 01:30:21PM +0300, Tomi Valkeinen wrote:
> After commit 92cc68e35863c1c61c449efa2b2daef6e9926048 ("drm/vblank: Use
> spin_(un)lock_irq() in drm_crtc_vblank_on()") omapdrm locking is broken:
> 
> WARNING: inconsistent lock state
> 5.8.0-rc2-00483-g92cc68e35863 #13 Tainted: G        W
> --------------------------------
> inconsistent {HARDIRQ-ON-W} -> {IN-HARDIRQ-W} usage.
> swapper/0/0 [HC1[1]:SC0[0]:HE0:SE1] takes:
> ea98222c (&dev->event_lock#2){?.+.}-{2:2}, at: drm_handle_vblank+0x4c/0x520 [drm]
> {HARDIRQ-ON-W} state was registered at:
>   trace_hardirqs_on+0x9c/0x1ec
>   _raw_spin_unlock_irq+0x20/0x58
>   omap_crtc_atomic_enable+0x54/0xa0 [omapdrm]
>   drm_atomic_helper_commit_modeset_enables+0x218/0x270 [drm_kms_helper]
>   omap_atomic_commit_tail+0x48/0xc4 [omapdrm]
>   commit_tail+0x9c/0x190 [drm_kms_helper]
>   drm_atomic_helper_commit+0x154/0x188 [drm_kms_helper]
>   drm_client_modeset_commit_atomic+0x228/0x268 [drm]
>   drm_client_modeset_commit_locked+0x60/0x1d0 [drm]
>   drm_client_modeset_commit+0x24/0x40 [drm]
>   drm_fb_helper_restore_fbdev_mode_unlocked+0x54/0xa8 [drm_kms_helper]
>   drm_fb_helper_set_par+0x2c/0x5c [drm_kms_helper]
>   drm_fb_helper_hotplug_event.part.0+0xa0/0xbc [drm_kms_helper]
>   drm_kms_helper_hotplug_event+0x24/0x30 [drm_kms_helper]
>   output_poll_execute+0x1a8/0x1c0 [drm_kms_helper]
>   process_one_work+0x268/0x800
>   worker_thread+0x30/0x4e0
>   kthread+0x164/0x190
>   ret_from_fork+0x14/0x20
> 
> The reason for this is that omapdrm calls drm_crtc_vblank_on() while
> holding event_lock taken with spin_lock_irq().
> 
> It is not clear why drm_crtc_vblank_on() and drm_crtc_vblank_get() are
> called while holding event_lock. I don't see any problem with moving
> those calls outside the lock, which is what this patch does.
> 
> Signed-off-by: Tomi Valkeinen <tomi.valkeinen at ti.com>

Reviewed-by: Laurent Pinchart <laurent.pinchart at ideasonboard.com>

I've had a quick look at other callers of drm_crtc_vblank_on() and none
of them call it with a spinlock held in the same function. I have
however only checked locally, not up the call stack.

> ---
>  drivers/gpu/drm/omapdrm/omap_crtc.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/omapdrm/omap_crtc.c b/drivers/gpu/drm/omapdrm/omap_crtc.c
> index 6d40914675da..328a4a74f534 100644
> --- a/drivers/gpu/drm/omapdrm/omap_crtc.c
> +++ b/drivers/gpu/drm/omapdrm/omap_crtc.c
> @@ -451,11 +451,12 @@ static void omap_crtc_atomic_enable(struct drm_crtc *crtc,
>  	if (omap_state->manually_updated)
>  		return;
>  
> -	spin_lock_irq(&crtc->dev->event_lock);
>  	drm_crtc_vblank_on(crtc);
> +
>  	ret = drm_crtc_vblank_get(crtc);
>  	WARN_ON(ret != 0);
>  
> +	spin_lock_irq(&crtc->dev->event_lock);
>  	omap_crtc_arm_event(crtc);
>  	spin_unlock_irq(&crtc->dev->event_lock);
>  }

-- 
Regards,

Laurent Pinchart