Re: [PATCH v8 2/3] arm64: dts: qcom: kodiak: enable inline crypto engine for SDHC

On 6/29/26 2:32 PM, Kuldeep Singh wrote:
> On 29-06-2026 17:24, Konrad Dybcio wrote:
>> On 6/29/26 9:35 AM, Kuldeep Singh wrote:
>>> From: Neeraj Soni <[email protected]>
>>>
>>> Add the dedicated ICE node for kodiak and reference it from the SDHC
>>> controller via qcom,ice.
>>>
>>> Keep the ICE node disabled by default in kodiak.dtsi and enable it in
>>> board DTS files where the corresponding SDHC node is enabled.
>>
>> Is there a reason to disable it in the first place?
>>
>> Keeping it enabled it would ensure that the core is parked
>> if it was left on by the bootloader
> Yes, observed an issue on kodiak Soc which supports both ufs-ice and
> sdhc-ice.
> 
> CRYPTO_INLINE_ENGINE flag can be set to m via 2 configs MMC_CRYPTO and
> SCSI_UFS_CRYPTO. In default defconfig, SCSI_UFS_CRYPTO is enabled
> default and hence, CRYPTO_INLINE_ENGINE is enabled irrespective of
> MMC_CRYPTO state.
> https://elixir.bootlin.com/linux/v7.1.2/source/arch/arm64/configs/defconfig#L1325
> https://elixir.bootlin.com/linux/v7.1.2/source/drivers/mmc/host/Kconfig#L614
> https://elixir.bootlin.com/linux/v7.1.2/source/drivers/ufs/host/Kconfig#L62
> 
> Kodiak Soc supporting both ufs/sdhc results in initiating both ice
> instances which in turn votes the clock in driver. In general, ice clock
> suspend/resume sequence is later controller by ufs/sdhc controller once
> the corresponding ice instance is initialised.
> 
> Since, qcs6490-rb3gen2(having only ufs storage) doesn't have emmc
> storage but sdhc-ice DT node still probes the driver, later sdhci-msm
> never takes ice instance(as sdhc_1 is disabled) but ice-clocks are
> remained voted on.
> 
> This results in an issue and hence, trying to fix it by enabling ice
> only when corresponding storage controller is enabled be it ufs or sdhc.
> ufs/sdhc are default disabled and enabled per DTS, hence following same
> convention with ice DT nodes too.
> 
> I am planning to update other DTs too with same change.
> Suggestions are welcome!

qcom_ice_probe()
  -> qcom_ice_create()
    -> devm_clk_get_optional_enabled()

If we remove the _enabled suffix and put a ice_resume() in ice_get(),
I believe this is no longer an issue

Konrad