[PATCH] media: rkisp1: Depend on PM

[PATCH] media: rkisp1: Depend on PM

[Igor Velkov via B4 Relay]

From: Igor Velkov <iav@iav.lv>

rkisp1_runtime_resume() enables the clocks and lets the interrupt
handlers run. Without CONFIG_PM it is never called, so rkisp1_probe()
reads RKISP1_CIF_VI_ID with the clocks disabled and every interrupt
handler returns IRQ_NONE.

ARCH_ROCKCHIP and arm64 ARCH_MXC select PM, and no arm32 i.MX has this
ISP, so only COMPILE_TEST builds lose the driver. rkcif already depends
on PM.

Signed-off-by: Igor Velkov <iav@iav.lv>
---
Split out of the rkvdec CONFIG_PM discussion, as Nicolas asked [1].
Without PM, olddefconfig on arm64 COMPILE_TEST now drops the driver;
with PM it builds as before (W=1).

[1] https://lore.kernel.org/r/8aafdbbf04a0d939c6b4d7d5577d97bb58d1c439.camel@collabora.com
---
 drivers/media/platform/rockchip/rkisp1/Kconfig | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/media/platform/rockchip/rkisp1/Kconfig b/drivers/media/platform/rockchip/rkisp1/Kconfig
index f53eb1f3f3e7..c825fbacee96 100644
--- a/drivers/media/platform/rockchip/rkisp1/Kconfig
+++ b/drivers/media/platform/rockchip/rkisp1/Kconfig
@@ -4,6 +4,7 @@ config VIDEO_ROCKCHIP_ISP1
 	depends on V4L_PLATFORM_DRIVERS
 	depends on VIDEO_DEV && OF
 	depends on ARCH_ROCKCHIP || ARCH_MXC || COMPILE_TEST
+	depends on PM
 	select MEDIA_CONTROLLER
 	select VIDEO_V4L2_SUBDEV_API
 	select VIDEOBUF2_DMA_CONTIG

---
base-commit: 27285b099fbc5cb2ae3aef20f3bbd7b3c88b3005
change-id: 20261102-rkisp1-depends-on-pm-219d94239b56

Best regards,
-- 
Igor Velkov <iav@iav.lv>

Re: [PATCH] media: rkisp1: Depend on PM

[Laurent Pinchart]

On Mon, Nov 02, 2026 at 05:49:46PM +0000, Igor Velkov via B4 Relay wrote:
> From: Igor Velkov <iav@iav.lv>
> 
> rkisp1_runtime_resume() enables the clocks and lets the interrupt
> handlers run. Without CONFIG_PM it is never called, so rkisp1_probe()
> reads RKISP1_CIF_VI_ID with the clocks disabled and every interrupt
> handler returns IRQ_NONE.
> 
> ARCH_ROCKCHIP and arm64 ARCH_MXC select PM,

Why do we need this patch then ?

> and no arm32 i.MX has this
> ISP, so only COMPILE_TEST builds lose the driver. rkcif already depends
> on PM.
> 
> Signed-off-by: Igor Velkov <iav@iav.lv>
> ---
> Split out of the rkvdec CONFIG_PM discussion, as Nicolas asked [1].
> Without PM, olddefconfig on arm64 COMPILE_TEST now drops the driver;
> with PM it builds as before (W=1).
> 
> [1] https://lore.kernel.org/r/8aafdbbf04a0d939c6b4d7d5577d97bb58d1c439.camel@collabora.com
> ---
>  drivers/media/platform/rockchip/rkisp1/Kconfig | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/media/platform/rockchip/rkisp1/Kconfig b/drivers/media/platform/rockchip/rkisp1/Kconfig
> index f53eb1f3f3e7..c825fbacee96 100644
> --- a/drivers/media/platform/rockchip/rkisp1/Kconfig
> +++ b/drivers/media/platform/rockchip/rkisp1/Kconfig
> @@ -4,6 +4,7 @@ config VIDEO_ROCKCHIP_ISP1
>  	depends on V4L_PLATFORM_DRIVERS
>  	depends on VIDEO_DEV && OF
>  	depends on ARCH_ROCKCHIP || ARCH_MXC || COMPILE_TEST
> +	depends on PM
>  	select MEDIA_CONTROLLER
>  	select VIDEO_V4L2_SUBDEV_API
>  	select VIDEOBUF2_DMA_CONTIG
> 
> ---
> base-commit: 27285b099fbc5cb2ae3aef20f3bbd7b3c88b3005
> change-id: 20261102-rkisp1-depends-on-pm-219d94239b56

-- 
Regards,

Laurent Pinchart

Re: [PATCH] media: rkisp1: Depend on PM

[Igor Velkov]

On Mon, Nov 02, 2026, Laurent Pinchart wrote:
> > ARCH_ROCKCHIP and arm64 ARCH_MXC select PM,
>
> Why do we need this patch then ?

Only COMPILE_TEST builds reach PM=n, and there the driver still binds and
probe reads RKISP1_CIF_VI_ID with the clocks off. Nicolas asked to close
the PM=n case per driver [1]; rkcif already depends on PM. If you'd rather
leave COMPILE_TEST-only configs alone, I'll drop it.

[1] https://lore.kernel.org/r/8aafdbbf04a0d939c6b4d7d5577d97bb58d1c439.camel@collabora.com

Re: [PATCH] media: rkisp1: Depend on PM

[Laurent Pinchart]

On Tue, Nov 03, 2026 at 02:55:00AM +0200, Igor Velkov wrote:
> On Mon, Nov 02, 2026, Laurent Pinchart wrote:
> > > ARCH_ROCKCHIP and arm64 ARCH_MXC select PM,
> >
> > Why do we need this patch then ?
> 
> Only COMPILE_TEST builds reach PM=n, and there the driver still binds and
> probe reads RKISP1_CIF_VI_ID with the clocks off.

What does the driver bind to when running on a system that has PM=n ? On
which platform can that happen ?

> Nicolas asked to close
> the PM=n case per driver [1]; rkcif already depends on PM. If you'd rather
> leave COMPILE_TEST-only configs alone, I'll drop it.
> 
> [1] https://lore.kernel.org/r/8aafdbbf04a0d939c6b4d7d5577d97bb58d1c439.camel@collabora.com

-- 
Regards,

Laurent Pinchart

Re: [PATCH] media: rkisp1: Depend on PM

[Igor Velkov]

On Tue, Nov 03, 2026, Laurent Pinchart wrote:
> What does the driver bind to when running on a system that has PM=n ? On
> which platform can that happen ?

None that I can name: the nodes exist only on Rockchip and i.MX8 boards,
and those select PM. The patch came out of the rkvdec thread, where Nicolas
pointed out that with PM=n the first MMIO access can hang and asked to
check similar drivers [1]; rkisp1 and rga had the same pattern. If that
case is not worth guarding, I'll drop it.

[1] https://lore.kernel.org/r/8aafdbbf04a0d939c6b4d7d5577d97bb58d1c439.camel@collabora.com

Re: [PATCH] media: rkisp1: Depend on PM

[Laurent Pinchart]

On Tue, Nov 03, 2026 at 01:25:45AM +0200, Igor Velkov wrote:
> On Tue, Nov 03, 2026, Laurent Pinchart wrote:
> > What does the driver bind to when running on a system that has PM=n ? On
> > which platform can that happen ?
> 
> None that I can name: the nodes exist only on Rockchip and i.MX8 boards,
> and those select PM. The patch came out of the rkvdec thread, where Nicolas
> pointed out that with PM=n the first MMIO access can hang and asked to
> check similar drivers [1]; rkisp1 and rga had the same pattern. If that
> case is not worth guarding, I'll drop it.

Given that it can't happen on any platform, I don't think the patch is
needed.

> [1] https://lore.kernel.org/r/8aafdbbf04a0d939c6b4d7d5577d97bb58d1c439.camel@collabora.com

-- 
Regards,

Laurent Pinchart

Re: [PATCH] media: rkisp1: Depend on PM

[Nicolas Dufresne]

[-- Attachment #1: Type: text/plain, Size: 833 bytes --]

Hi,

Le mardi 03 novembre 2026 à 02:55 +0200, Igor Velkov a écrit :
> On Mon, Nov 02, 2026, Laurent Pinchart wrote:
> > > ARCH_ROCKCHIP and arm64 ARCH_MXC select PM,
> > 
> > Why do we need this patch then ?
> 
> Only COMPILE_TEST builds reach PM=n, and there the driver still binds and
> probe reads RKISP1_CIF_VI_ID with the clocks off. Nicolas asked to close
> the PM=n case per driver [1]; rkcif already depends on PM. If you'd rather
> leave COMPILE_TEST-only configs alone, I'll drop it.
> 
> [1] https://lore.kernel.org/r/8aafdbbf04a0d939c6b4d7d5577d97bb58d1c439.camel@collabora.com


The rkvdec coded was using macros capable of supporting both, so it was logical
to make that functional, though if non of the Rockchip ARCH (32bit and 64bit)
can be built without PM, then indeed we can drop.

Nicolas

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 228 bytes --]

Re: [PATCH] media: rkisp1: Depend on PM

[Igor Velkov]

On Mon, Nov 02, 2026, Nicolas Dufresne wrote:
> though if non of the Rockchip ARCH (32bit and 64bit) can be built
> without PM, then indeed we can drop.

ARCH_ROCKCHIP selects PM on both arm and arm64, so I'm dropping this one
and the same patch for rga.