[PATCH] drm/vkms: Add support for vkms work without vblank

[Daniel Vetter]

On Tue, Aug 21, 2018 at 4:52 PM, Rodrigo Siqueira
<rodrigosiqueiramelo at gmail.com> wrote:
> On 08/21, Daniel Vetter wrote:
>> On Mon, Aug 20, 2018 at 08:55:44PM -0300, Rodrigo Siqueira wrote:
>> > Currently, vkms needs VBlank to work well. This patch adds another
>> > operation model that make vkms works without VBlank support.
>> >
>> > Signed-off-by: Rodrigo Siqueira <rodrigosiqueiramelo at gmail.com>
>> > ---
>> >  drivers/gpu/drm/vkms/vkms_crtc.c | 10 ++++++++++
>> >  drivers/gpu/drm/vkms/vkms_drv.c  | 12 ++++++++++--
>> >  drivers/gpu/drm/vkms/vkms_drv.h  |  1 +
>> >  3 files changed, 21 insertions(+), 2 deletions(-)
>> >
>> > diff --git a/drivers/gpu/drm/vkms/vkms_crtc.c b/drivers/gpu/drm/vkms/vkms_crtc.c
>> > index bfe6e0312cc4..001fa5c1d326 100644
>> > --- a/drivers/gpu/drm/vkms/vkms_crtc.c
>> > +++ b/drivers/gpu/drm/vkms/vkms_crtc.c
>> > @@ -145,12 +145,22 @@ static const struct drm_crtc_funcs vkms_crtc_funcs = {
>> >  static void vkms_crtc_atomic_enable(struct drm_crtc *crtc,
>> >                                 struct drm_crtc_state *old_state)
>> >  {
>> > +   struct vkms_output *vkms_out = drm_crtc_to_vkms_output(crtc);
>> > +
>> > +   if (vkms_out->disable_vblank)
>> > +           return;
>> > +
>> >     drm_crtc_vblank_on(crtc);
>> >  }
>> >
>> >  static void vkms_crtc_atomic_disable(struct drm_crtc *crtc,
>> >                                  struct drm_crtc_state *old_state)
>> >  {
>> > +   struct vkms_output *vkms_out = drm_crtc_to_vkms_output(crtc);
>> > +
>> > +   if (vkms_out->disable_vblank)
>> > +           return;
>> > +
>> >     drm_crtc_vblank_off(crtc);
>> >  }
>> >
>> > diff --git a/drivers/gpu/drm/vkms/vkms_drv.c b/drivers/gpu/drm/vkms/vkms_drv.c
>> > index 5d78bd97e69c..f30c4113d4c4 100644
>> > --- a/drivers/gpu/drm/vkms/vkms_drv.c
>> > +++ b/drivers/gpu/drm/vkms/vkms_drv.c
>> > @@ -21,6 +21,10 @@
>> >
>> >  static struct vkms_device *vkms_device;
>> >
>> > +static bool disable_vblank;
>> > +module_param_named(disablevblank, disable_vblank, bool, 0444);
>> > +MODULE_PARM_DESC(disablevblank, "Disable vblank interrupt");
>>
>> I think the info output is much clearer, I'd rename this to virtual_hw,
>> with 0 as default, and a help text along of "Enable virtual HW mode
>> (disables vblanks and immediately completes flips)."
>
> You right, virtual_hw is much better :)
> I will fix it.
>
>> > +
>> >  static const struct file_operations vkms_driver_fops = {
>> >     .owner          = THIS_MODULE,
>> >     .open           = drm_open,
>> > @@ -105,9 +109,13 @@ static int __init vkms_init(void)
>> >             goto out_fini;
>> >     }
>> >
>> > -   vkms_device->drm.irq_enabled = true;
>> > +   vkms_device->output.disable_vblank = disable_vblank;
>> > +   vkms_device->drm.irq_enabled = !disable_vblank ? true : false;
>> > +
>> > +   if (disable_vblank)
>> > +           DRM_INFO("Virtual Hardware mode enabled");
>> >
>> > -   ret = drm_vblank_init(&vkms_device->drm, 1);
>> > +   ret = (disable_vblank) ? 0 : drm_vblank_init(&vkms_device->drm, 1);
>> >     if (ret) {
>> >             DRM_ERROR("Failed to vblank\n");
>> >             goto out_fini;
>> > diff --git a/drivers/gpu/drm/vkms/vkms_drv.h b/drivers/gpu/drm/vkms/vkms_drv.h
>> > index f156c930366a..9ee862e6f7e5 100644
>> > --- a/drivers/gpu/drm/vkms/vkms_drv.h
>> > +++ b/drivers/gpu/drm/vkms/vkms_drv.h
>> > @@ -52,6 +52,7 @@ struct vkms_output {
>> >     struct drm_encoder encoder;
>> >     struct drm_connector connector;
>> >     struct hrtimer vblank_hrtimer;
>> > +   bool disable_vblank;
>> >     ktime_t period_ns;
>> >     struct drm_pending_vblank_event *event;
>> >     bool crc_enabled;
>>
>> The vblank handling parts look correct to me.
>>
>> Does kms_flip work with this? Virtual hw immediately completes the
>> drm_crtc_state->event in the atomic_update callback, that part seems to be
>> missing from your patch.
>
> I tested with patched version of kms_flip to skip vblank. Here is the patch:
>
> https://patchwork.kernel.org/patch/10570861/
>
> About the drm_crtc_state->event, afaiu, the code in the function
> vkms_crtc_atomic_flush immediately send the vblank event to the user
> space via drm_crtc_send_vblank_event which make things work in the user
> space. Make sense? Or I missed something?

Ah, the magic of the vblank_get() == 0 check. A bit tricky, but yes
this indeed works. I think it'd be good to explain this in the commit
message.

>> Also I like the idea of switching away from the generic -EINVAL to -ENOTTY
>> for when vblanks aren't supported. Do you have that somewhere (as a
>> separate patch).
>
> I already started to investigate if the change between -EINVAL to
> -ENOTTY will break an existing kms userspace compositors. I already
> checked:
>
> * kwin
> * sway
> * wayland-core
> * weston
>
> I still have to check the X drivers, chromium, and drm_hwc.

Awesome, keep up the great work.

Cheers, Daniel

>
> Thanks for the feedback.
>
>> -Daniel
>> --
>> Daniel Vetter
>> Software Engineer, Intel Corporation
>> http://blog.ffwll.ch
>
> --
> Rodrigo Siqueira
> http://siqueira.tech
> Graduate Student
> Department of Computer Science
> University of São Paulo



-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch