Re: [PATCH v3 drm-dp 7/8] drm/hisilicon/hibmc: Enable this hot plug detect of irq feature

[Yongbang Shi <shiyongbang@huawei.com>]

> On Sat, Mar 01, 2025 at 04:45:40PM +0800, Yongbang Shi wrote:
>>> On Thu, Feb 27, 2025 at 09:46:10PM +0800, Yongbang Shi wrote:
>>>>> On Tue, Feb 25, 2025 at 09:57:17PM +0800, Yongbang Shi wrote:
>>>>>>> On Mon, 24 Feb 2025 at 16:03, Yongbang Shi <shiyongbang@huawei.com> wrote:
>>>>>>>>> On Sat, Feb 22, 2025 at 06:35:48PM +0800, Yongbang Shi wrote:
>>>>>>>>>>>> +static int hibmc_dp_hpd_event(struct drm_client_dev *client)
>>>>>>>>>>>> +{
>>>>>>>>>>>> +  struct hibmc_dp *dp = container_of(client, struct hibmc_dp, client);
>>>>>>>>>>>> +  struct hibmc_drm_private *priv = to_hibmc_drm_private(dp->drm_dev);
>>>>>>>>>>>> +  struct drm_display_mode *mode = &priv->crtc.state->adjusted_mode;
>>>>>>>>>>>> +  int ret;
>>>>>>>>>>>> +
>>>>>>>>>>>> +  if (dp->hpd_status) {
>>>>>>>>>>>> +          hibmc_dp_hpd_cfg(&priv->dp);
>>>>>>>>>>>> +          ret = hibmc_dp_prepare(dp, mode);
>>>>>>>>>>>> +          if (ret)
>>>>>>>>>>>> +                  return ret;
>>>>>>>>>>>> +
>>>>>>>>>>>> +          hibmc_dp_display_en(dp, true);
>>>>>>>>>>>> +  } else {
>>>>>>>>>>>> +          hibmc_dp_display_en(dp, false);
>>>>>>>>>>>> +          hibmc_dp_reset_link(&priv->dp);
>>>>>>>>>>>> +  }
>>>>>>>>>>> If I understand this correctly, you are using a separate drm_client to
>>>>>>>>>>> enable and disable the link & display. Why is it necessary? Existing
>>>>>>>>>>> drm_clients and userspace compositors use drm framework, they should be
>>>>>>>>>>> able to turn the display on and off as required.
>>>>>>>>>>>
>>>>>>>>>> Thanks for your asking, there are cfg/reset process when the connector 's pluging in/out.
>>>>>>>>>> We want to cfg DP registers again when the connector changes. Not only dp link training, but also cfg
>>>>>>>>>> the different video modes into DP registers.
>>>>>>>>> Why? The link training and mode programming should happen during
>>>>>>>>> pre_enable / enable stage (legacy or atomic).
>>>>>>>> Hi Dmitry,
>>>>>>>>
>>>>>>>> Right, that's what I'm curious about. It won't call encoder enble/disable functions when I triggered HPD.
>>>>>>>> And I'm sure the drm_connector_helper_hpd_irq_event() is called. So I add a drm_client for it.I
>>>>>>> It should be userspace, who triggers the enable/disable (or it should
>>>>>>> be the in-kernel fbdev / fbcon, which interface through the generic
>>>>>>> drm_fbdev client).
>>>>>> Right, I knew it. When I insmode my driver firstly (or restart display service), it will call disable, modeset and enable,
>>>>>> by user, but it won't call when HPD triggered .
>>>>> - Is HPD even properly delivered to userspace? What kind of compsitor
>>>>>      are you using? Is .detect working properly and reporting a correct
>>>>>      plug-in state?
>>>> Thanks for your answering. I'm not very good understanding about userspace in framework. In my opinion, when I call
>>>> this drm_connector_helper_hpd_irq_event(), the HPD will deliver to userspace.
>>>> I use Xorg, and the display service is GDM.
>>>> The .detect is called and the getting modes info is correct.
>>>> I find that it would only trigger(disable, modeset and enable), when I changed resolutions, restart display service and insmod driver.
>>> You can go to the display settings in GDM. It would be interesting to
>>> observe if it notes the second monitor or not. Last, but not least, you
>>> can use a simple tool like 'xrandr' under your XOrg session to set the
>>> display resolution.
>> Thank you for your advice!
>> Right, there are DP and VGA two monitors. I tried to totally remove the vga connector in driver, the problem is gone.
>> So do I need to clear the vga connector, if dp is plugged in?
> Unless your hardware can not manage two outputs at the same time, no,
> you don't have to. Just check how it behaves on x86 systems. Ideally
> your driver should have the same behaviour.

Our hardware cannot support two outputs with different timing, so I used the one crtc and one plane that DP and VGA share. And just add a new DP connector
with a encoder, just like the previous VGA's code logic. But the HPD problem makes me feel confused, should I change the framwork structure to slove this problem?

And also, I will check whether this driver works good on the x86 server. Right now, I'm testing on arm64 server.

>> And also, I used xrandr to set modes after 'startx'. Changing resolutions works,
>> but there are errs when set some low resolutions.
> That's a separate topic, most likely related to timing or to some other
> issues. You can fix that separately (but please do, switching modes
> should work).

Okay!