Re: [PATCH] clk: rockchip: rk3568: Add PCIe pipe clock gates

Re: [PATCH] clk: rockchip: rk3568: Add PCIe pipe clock gates

[Heiko Stuebner]

Hi Shawn,

Am Donnerstag, 5. März 2026, 09:06:51 Mitteleuropäische Normalzeit schrieb Shawn Lin:
> The PCIe pipe clocks are currently left as orphan clocks and remain
> enabled indefinitely, which is suboptimal. Add the missing clock gates
> so the PCIe driver can explicitly manage them when not in use.
> 
> Signed-off-by: Shawn Lin <shawn.lin@rock-chips.com>
> ---
> Hi Heiko,
> 
> This patch should be considered only after David's patch[1] is merged,
> otherwise it breaks PCIe on RK3566/RK3568 due to clk_disable_unused().

This would also break compatiblity with old devicetrees.

The devicetree used by the kernel can also be embedded for example
in an already installed u-boot and handed to the kernel from there.
You cannot expect board devicetrees and the kernel image to be updated
in lockstep.

We also sort of guarantee this backwards compatiblity - old DTB with
newer kernel.

The solution is to add a CLK_IGNORE_UNUSED flag (but not CLK_IS_CRITICAL)
to the newly added gate.

This lets the clock stay and keeps the original behavious on old DTBs,
but also allows the PCIe driver in newer kernels to turn off the clock
if it wants to.

Thanks
Heiko

Re: [PATCH] clk: rockchip: rk3568: Add PCIe pipe clock gates

[Shawn Lin]

Hi Heiko,

在 2026/03/06 星期五 18:52, Heiko Stuebner 写道:
> Hi Shawn,
> 
> Am Donnerstag, 5. März 2026, 09:06:51 Mitteleuropäische Normalzeit schrieb Shawn Lin:
>> The PCIe pipe clocks are currently left as orphan clocks and remain
>> enabled indefinitely, which is suboptimal. Add the missing clock gates
>> so the PCIe driver can explicitly manage them when not in use.
>>
>> Signed-off-by: Shawn Lin <shawn.lin@rock-chips.com>
>> ---
>> Hi Heiko,
>>
>> This patch should be considered only after David's patch[1] is merged,
>> otherwise it breaks PCIe on RK3566/RK3568 due to clk_disable_unused().
> 
> This would also break compatiblity with old devicetrees.
> 
> The devicetree used by the kernel can also be embedded for example
> in an already installed u-boot and handed to the kernel from there.
> You cannot expect board devicetrees and the kernel image to be updated
> in lockstep.

Indeed.

> 
> We also sort of guarantee this backwards compatiblity - old DTB with
> newer kernel.
> 
> The solution is to add a CLK_IGNORE_UNUSED flag (but not CLK_IS_CRITICAL)
> to the newly added gate.
> 

Thanks, will improve it.

> This lets the clock stay and keeps the original behavious on old DTBs,
> but also allows the PCIe driver in newer kernels to turn off the clock
> if it wants to.
> 
> Thanks
> Heiko
> 
> 
> 
>