Re: [PATCH RFC v3 2/6] drm/sprd: add Unisoc's drm kms master

[tang pengchuan <kevin3.tang@xxxxxxxxx>]

On Tue, Feb 25, 2020 at 3:38 PM Thomas Zimmermann <tzimmermann@xxxxxxx> wrote:

Hi

Am 23.02.20 um 05:26 schrieb tang pengchuan:
>
>
> On Sun, Feb 23, 2020 at 5:27 AM Sam Ravnborg <sam@xxxxxxxxxxxx
> <mailto:sam@xxxxxxxxxxxx>> wrote:
>
>     Hi Kevin/tang.
>
>     Thanks for the quick and detailed feedback.
>     Your questions are addressed below.
>
>             Sam
>
>
>     > > > +static int sprd_drm_bind(struct device *dev)
>     > > > +{
>     > > > +     struct drm_device *drm;
>     > > > +     struct sprd_drm *sprd;
>     > > > +     int err;
>     > > > +
>     > > > +     drm = drm_dev_alloc(&sprd_drm_drv, dev);
>     > > > +     if (IS_ERR(drm))
>     > > > +             return PTR_ERR(drm);
>     > > You should embed drm_device in struct sprd_drm.
>     > > See example code in drm/drm_drv.c
>     > >
>     > > This is what modern drm drivers do.
>     > >
>     > > I *think* you can drop the component framework if you do this.
>     > >
>     > I have read it(drm/drm_drv.c) carefully, if drop the component
>     framework,
>     > the whole our drm driver maybe need to redesign, so i still want
>     to keep
>     > current design.
>     OK, fine.
>
>     > > > +     sprd_drm_mode_config_init(drm);
>     > > > +
>     > > > +     /* bind and init sub drivers */
>     > > > +     err = component_bind_all(drm->dev, drm);
>     > > > +     if (err) {
>     > > > +             DRM_ERROR("failed to bind all component.\n");
>     > > > +             goto err_dc_cleanup;
>     > > > +     }
>     > > When you have a drm_device - which you do here.
>     > > Then please use drm_err() and friends for error messages.
>     > > Please verify all uses of DRM_XXX
>     > >
>     >   modern drm drivers need drm_xxx to replace DRM_XXX?
>     Yes, use of DRM_XXX is deprecated - when you have a drm_device.
>
>     > >
>     > > > +     /* with irq_enabled = true, we can use the vblank
>     feature. */
>     > > > +     drm->irq_enabled = true;
>     > > I cannot see any irq being installed. This looks wrong.
>     > >
>     > Our display controller isr is been register on crtc
>     driver(sprd_dpu.c), not
>     > here.
>
>     I think you just need to move this to next patch and then it is fine.
>
> Here is the advice given by Thomas Zimmermann, similar to the advice you
> gave.
> I have given thomas feedback about my questions, maybe thomas didn't see
> my email, so there is no reply.

I have been busy last week. Sorry for not getting back to you.

>
> But I've always been confused, because irq is initialized in drm driver
> for other guys, why not for me?

Do you have an example?

Hi Thomas,

The the irq is initialized in the sub-device code, but the device state is set on kms driver.

E.g

Here is the device state set on kms driver:

 static int mtk_drm_kms_init(struct drm_device *drm)
......

 	/*
 	 * We don't use the drm_irq_install() helpers provided by the DRM
 	 * core, so we need to set this manually in order to allow the
 	 * DRM_IOCTL_WAIT_VBLANK to operate correctly.
 	 */
 	drm->irq_enabled = true;

Here is irq install on subdev:
 static int mtk_disp_rdma_probe(struct platform_device *pdev)
......
 	ret = devm_request_irq(dev, irq, mtk_disp_rdma_irq_handler,
 			       IRQF_TRIGGER_NONE, dev_name(dev), priv);

I'm not sure if my understanding is wrong...

Best regards
Thomas

> Can you help to tell the reason in detail, looking forward to your answers.
>
> Thomas's suggestion：
> -------------------------------------------------------------------------------------------
>
> This line indicates the problem's design. The irq is initialized in the
> sub-device code, but the device state is set here. Instead both should
> be set in the same place.
>
>> +
>> +     /* reset all the states of crtc/plane/encoder/connector */
>> +     drm_mode_config_reset(drm);
>> +
>> +     /* init kms poll for handling hpd */
>> +     drm_kms_helper_poll_init(drm);
>
> Most of this function's code should be moved into the sub-device bind
> function.
>
> Here, maybe do:
>
>  * allocate device structures
>  * call component_bind_all()
>  * on success, register device
>
> The sub-device function should then do
>
>  * init modesetting, crtc, planes, etc.
>  * do drm_mode_config_reset()
>  * init vblanking
>  * init the irq
>  * do drm_kms_helper_poll_init()
>
> roughtly in that order. It makes sense to call drm_vblank_init() after
> drm_mode_config_reset(), as vblanking uses some of the mode-config fields. 
> ------------------------------------------------------------------------------------------------------
>
>
>             Sam
>
>
> _______________________________________________
> dri-devel mailing list
> dri-devel@xxxxxxxxxxxxxxxxxxxxx
> https://lists.freedesktop.org/mailman/listinfo/dri-devel
>

--
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer

_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/dri-devel