Re: [PATCH v3 04/27] drm: Don't test for IRQ support in VBLANK ioctls

From: Thomas Zimmermann <tzimmermann@suse.de>
To: Jani Nikula <jani.nikula@linux.intel.com>,
	Thierry Reding <thierry.reding@gmail.com>
Cc: hamohammed.sa@gmail.com, emma@anholt.net, airlied@linux.ie,
	nouveau@lists.freedesktop.org, rodrigo.vivi@intel.com,
	liviu.dudau@arm.com, alexandre.torgue@foss.st.com,
	dri-devel@lists.freedesktop.org, michal.simek@xilinx.com,
	melissa.srw@gmail.com, linux-tegra@vger.kernel.org,
	laurent.pinchart@ideasonboard.com, benjamin.gaignard@linaro.org,
	linux@armlinux.org.uk, mihail.atanassov@arm.com,
	linux-stm32@st-md-mailman.stormreply.com,
	linux-samsung-soc@vger.kernel.org, jy0922.shim@samsung.com,
	krzysztof.kozlowski@canonical.com,
	linux-rockchip@lists.infradead.org,
	linux-mediatek@lists.infradead.org, wens@csie.org,
	jernej.skrabec@gmail.com, jonathanh@nvidia.com,
	xinliang.liu@linaro.org, kong.kongxinwei@hisilicon.com,
	james.qian.wang@arm.com, linux-imx@nxp.com,
	Daniel Vetter <daniel.vetter@ffwll.ch>,
	linux-graphics-maintainer@vmware.com,
	intel-gfx@lists.freedesktop.org, bskeggs@redhat.com,
	chunkuang.hu@kernel.org, puck.chen@hisilicon.com,
	s.hauer@pengutronix.de, rodrigosiqueiramelo@gmail.com,
	laurentiu.palcu@oss.nxp.com, matthias.bgg@gmail.com,
	kernel@pengutronix.de, linux-arm-kernel@lists.infradead.org,
	mcoquelin.stm32@gmail.com, amd-gfx@lists.freedesktop.org,
	hyun.kwon@xilinx.com, tomba@kernel.org, jyri.sarha@iki.fi,
	yannick.fertre@foss.st.com, Xinhui.Pan@amd.com,
	sw0312.kim@samsung.com, hjc@rock-chips.com,
	christian.koenig@amd.com, linux-sunxi@lists.linux.dev,
	kyungmin.park@samsung.com,
	kieran.bingham+renesas@ideasonboard.com,
	philippe.cornu@foss.st.com, alexander.deucher@amd.com,
	tiantao6@hisilicon.com, shawnguo@kernel.org
Subject: Re: [PATCH v3 04/27] drm: Don't test for IRQ support in VBLANK ioctls
Date: Thu, 24 Jun 2021 14:57:44 +0200	[thread overview]
Message-ID: <0b0b7cda-9808-f0ed-4122-2d8eeaa8f7c3@suse.de> (raw)
In-Reply-To: <87zgvfsalz.fsf@intel.com>

[-- Attachment #1.1: Type: text/plain, Size: 3127 bytes --]

Hi

Am 24.06.21 um 14:36 schrieb Jani Nikula:
> On Thu, 24 Jun 2021, Thierry Reding <thierry.reding@gmail.com> wrote:
>> On Thu, Jun 24, 2021 at 11:07:57AM +0200, Thomas Zimmermann wrote:
>>> Hi
>>>
>>> Am 24.06.21 um 10:51 schrieb Jani Nikula:
>>>> On Thu, 24 Jun 2021, Thomas Zimmermann <tzimmermann@suse.de> wrote:
>>>>> Hi
>>>>>
>>>>> Am 24.06.21 um 10:06 schrieb Jani Nikula:
>>>>>> On Thu, 24 Jun 2021, Thomas Zimmermann <tzimmermann@suse.de> wrote:
>>>>>>> diff --git a/drivers/gpu/drm/drm_vblank.c b/drivers/gpu/drm/drm_vblank.c
>>>>>>> index 3417e1ac7918..10fe16bafcb6 100644
>>>>>>> --- a/drivers/gpu/drm/drm_vblank.c
>>>>>>> +++ b/drivers/gpu/drm/drm_vblank.c
>>>>>>> @@ -1748,8 +1748,16 @@ int drm_wait_vblank_ioctl(struct drm_device *dev, void *data,
>>>>>>>     	unsigned int pipe_index;
>>>>>>>     	unsigned int flags, pipe, high_pipe;
>>>>>>> -	if (!dev->irq_enabled)
>>>>>>> -		return -EOPNOTSUPP;
>>>>>>> +#if defined(CONFIG_DRM_LEGACY)
>>>>>>> +	if  (unlikely(drm_core_check_feature(dev, DRIVER_LEGACY))) {
>>>>>>> +		if (!dev->irq_enabled)
>>>>>>> +			return -EOPNOTSUPP;
>>>>>>> +	} else /* if DRIVER_MODESET */
>>>>>>> +#endif
>>>>>>> +	{
>>>>>>> +		if (!drm_dev_has_vblank(dev))
>>>>>>> +			return -EOPNOTSUPP;
>>>>>>> +	}
>>>>>>
>>>>>> Sheesh I hate this kind of inline #ifdefs.
>>>>>>
>>>>>> Two alternate suggestions that I believe should be as just efficient:
>>>>>
>>>>> Or how about:
>>>>>
>>>>> static bool drm_wait_vblank_supported(struct drm_device *dev)
>>>>>
>>>>> {
>>>>>
>>>>> if defined(CONFIG_DRM_LEGACY)
>>>>> 	if  (unlikely(drm_core_check_feature(dev, DRIVER_LEGACY)))
>>>>>
>>>>> 		return dev->irq_enabled;
>>>>>
>>>>> #endif
>>>>> 	return drm_dev_has_vblank(dev);
>>>>>
>>>>> }
>>>>>
>>>>>
>>>>> ?
>>>>>
>>>>> It's inline, but still readable.
>>>>
>>>> It's definitely better than the original, but it's unclear to me why
>>>> you'd prefer this over option 2) below. I guess the only reason I can
>>>> think of is emphasizing the conditional compilation. However,
>>>> IS_ENABLED() is widely used in this manner specifically to avoid inline
>>>> #if, and the compiler optimizes it away.
>>>
>>> It's simply more readable to me as the condition is simpler. But option 2 is
>>> also ok.
>>
>> Perhaps do something like this, then:
>>
>> 	if (IS_ENABLED(CONFIG_DRM_LEGACY)) {
>> 		if (unlikely(drm_core_check_feature(dev, DRIVER_LEGACY)))
>> 			return dev->irq_enabled;
>> 	}
>>
>> 	return drm_dev_has_vblank(dev);
>>
>> That's about just as readable as the variant involving the preprocessor
>> but has all the benefits of not using the preprocessor.
> 
> Looks like a winner to me. :)

That's the most readable.

But I just remembered that irq_enabled will likely become legacy-only in 
the device structure. We'll need an ifdef variant then. :/

Best regards
Thomas

> 
> BR,
> Jani.
> 
> 

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

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 840 bytes --]