Re: [PATCH] drm/crtc: Add a helper func to get a registered crtc from its index

[Thierry Reding]

On Tue, Aug 25, 2015 at 11:36:18AM +0200, Daniel Vetter wrote:
> On Tue, Aug 25, 2015 at 11:13:51AM +0800, Xinliang Liu wrote:
> > This patch add a helper func to get a registered crtc from its index.
> > In some case, where we know the crtc's index and we want to know the
> > crtc too.
> > 
> > For example, the enable_vblank func of struct drm_driver:
> > In the implementation of this func, we know the index of the crtc but
> > we want to know the crtc. This helper func can get the crtc easily.
> > A sample impelmentation of enable_vblank is as shown as bellow:
> > 
> > int hisi_drm_crtc_enable_vblank(struct drm_device *dev, int c)
> > {
> >     struct drm_crtc *crtc = drm_get_crtc_from_index(dev, c);
> >     struct hisi_crtc *hcrtc = to_hisi_crtc(crtc);
> >     struct hisi_crtc_ops *ops = hcrtc->ops;
> >     int ret = 0;
> > 
> >     if (ops->enable_vblank)
> >             ret = ops->enable_vblank(hcrtc);
> > 
> >     return ret;
> > }
> > 
> > Signed-off-by: Xinliang Liu <xinliang....@linaro.org>
> 
> Yeah unfortunately drm_irq.c is still stick in the old pre-KMS days. I
> think we should go a bit further here though to allow new drivers to be
> completely free of int pipe:

Of course you meant to say /unsigned/ int pipe =)

> - add a new array pointer dev->mode_conifg.crtc_arr, which is
>   (re-)allocated in drm_crtc_init_with_planes. Then a pipe->crtc lookup
>   will be just
> 
>       crtc = dev->mode_config.crtc_arr[pipe];
> 
> - add new hooks for vblank handling int drm_crtc_helper_funcs for
>   enable_vblanke, disable_vblank, get_vblank_timestamp and get_scanoutpos.
>   Ofc also anotate the docs for the existing hooks and make it clear new
>   drivers should use the new ones. Ofc these new hooks should directly
>   take a struct drm_crtc * instead of inte pipe.

I have a couple patches to address this partially, which came about as a
result of the int crtc/crtc_id/pipe/whatever -> unsigned int pipe
conversion work that I've been doing.

> - change the code in drm_irq.c to wrap all callbacks and first check
>   whether the new ones are there and only if that's not the case call the
>   old ones.
> 
> With these changes drivers can be completely free of int pipe and use
> struct drm_crtc exclusivly I think, and the mess would be fully restricted
> to drm_irq.c.

I like the idea of moving the callbacks to drm_crtc_helper_funcs. That
allows us to introduce this step by step, without a flag date when every
driver needs to switch the drm_driver functions over.

Thierry

signature.asc
Description: PGP signature