[PATCH] drm/vkms: Don't warn hrtimer_forward_now failure.

Tetsuo Handa penguin-kernel at i-love.sakura.ne.jp
Tue May 26 04:18:45 UTC 2020

On 2020/05/26 2:00, Daniel Vetter wrote:
> Forgot to add: I did run this quickly with vkms as secondary output.
> Good fireworks show, but there's an entire army of additional splats
> after the first one. The WARN_ON you're removing is only the
> messenger, the real bug is probably one of the later backtraces. Or at
> least points more clearly at the real bug.
> -Daniel
>> Debug printk() patch shows that hrtimer_forward_now() can return larger than 1.
>> What is the reason you are expecting hrtimer_forward_now() to always return 1 ?
>>> If the vkms driver is loaded, and there's nothing else going on, then what
>>> I expect to happen is that this virtual hw is entirely off. And in that
>>> case, the vblank ioctl should be rejected outright. So there's definitely
>>> something fishy going on to begin with.
>>> If otoh the virtual hw is somehow on (maybe fbcon gets loaded, no idea),
>>> then the vblank wait shouldn't just blow up like this.

I see. Another debug printk() patch ( https://syzkaller.appspot.com/text?tag=Patch&x=135f254a100000 ) found that

  [   86.613767][ T7837] [drm:drm_stub_open] 
  [   86.621592][ T7837] [drm:drm_open] pid = 7837, minor = 1
  [   86.651632][ T7837] [drm:drm_ioctl] pid=7837, dev=0xe201, auth=1, DRM_IOCTL_WAIT_VBLANK
  [   86.663391][ T7837] [drm:drm_calc_timestamping_constants] *ERROR* crtc 32: Can't calculate constants, dotclock = 0!
  [   86.679151][ T7837] [drm:drm_calc_timestamping_constants] crtc 32: hwmode: htotal 0, vtotal 0, vdisplay 0
  [   86.688865][ T7837] [drm:drm_calc_timestamping_constants] crtc 32: clock 0 kHz framedur 0 linedur 0
  [   86.688899][ T7837] Starting out->vblank_hrtimer : out->period_ns=0 vblank->framedur_ns=0
  [   86.689348][ T7837] [drm:drm_vblank_enable] enabling vblank on crtc 0, ret: 0
  [   86.706883][ T7837] [drm:drm_update_vblank_count] updating vblank count on crtc 0: current=0, diff=0, hw=0 hw_last=0
  [   86.706962][    C1] now=86714983667 interval=0 ret_overrun=17693375

hrtimer_start(&out->vblank_hrtimer, 0, HRTIMER_MODE_REL) from vkms_enable_vblank() is
causing immediate vkms_vblank_simulate() callback. The reason of vblank->framedur_ns == 0
is that drm_calc_timestamping_constants() fails to calculate framedur_ns (which is 0 by default)
due to mode->crtc_clock <= 0. Thus, somehow initializing mode->crtc_clock > 0 might be able
to solve this problem.

By the way, if you have an environment that can reproduce this problem, you can
test with debug printk() patch above applied.

More information about the dri-devel mailing list