[PATCH] drm/drm_vblank: use drm_warn_once() to warn undefined mode timing

[Shawn Guo]

On Fri, Oct 16, 2020 at 09:58:46AM +0200, Daniel Vetter wrote:
> On Fri, Oct 16, 2020 at 9:13 AM Shawn Guo <shawn.guo at linaro.org> wrote:
> >
> > Commit 5caa0feafcc6 ("drm/vblank: Lock down vblank->hwmode more") added
> > WARN_ON_ONCE() for atomic drivers to warn the case that vsync is enabled
> > before a mode has been set on CRTC.  This happens sometimes during the
> > initial mode setting of a CRTC.  It also happens on Android running HWC2
> > backed with drm_hwcomposer, where HWC2::SetVsyncEnabled could be called
> > before the atomic mode setting on CRTC happens.
> >
> > In this case, there is nothing really bad to happen as kernel function
> > returns as no-op.  So using WARN() version might be overkilled,
> > considering some user space crash reporting services may treat kernel
> > WARNINGS as crashes.  Let's drop WARN_ON_ONCE() and change drm_dbg_core()
> > to drm_warn_once() for warning undefined mode timing.
> 
> This indicates a bug in your driver. Please fix it there, not by
> shutting up the core code complaining about that. Either you're
> getting vblank timestamps when the vblank isn't set up yet
> (drm_crtc_vblank_on/off) or there's some other race going on in your
> driver code resulting in this.

Thanks for the comment, Daniel.

I'm hitting this warning on an Android running drm_hwcomposer.  I'm
indeed getting vblank timestamps request before drm_crtc_vblank_on() is
called.  I'm not sure this is a bug or race condition in the driver
code, as both vblank timestamps and on/off requests are coming from user
space ioctl for my case.  @Sean, that means the problem is in Android
drm_hwcomposer code?

Shawn