Re: [PATCH v6 07/11] arm64: dts: qcom: sm6115: Add Crypto Engine support

From: Konrad Dybcio
Date: Fri May 19 2023 - 06:51:49 EST




On 19.05.2023 12:49, Bhupesh Sharma wrote:
> On Fri, 19 May 2023 at 16:12, Konrad Dybcio <konrad.dybcio@xxxxxxxxxx> wrote:
>>
>> On 19.05.2023 12:22, Bhupesh Sharma wrote:
>>> Hi Stephan,
>>>
>>> On Fri, 19 May 2023 at 15:40, Stephan Gerhold <stephan@xxxxxxxxxxx> wrote:
>>>>
>>>> Hi Bhupesh,
>>>>
>>>> Not sure if this is the latest version of this series since it's pretty
>>>> old but I didn't find a new one. Just came here because you mentioned
>>>> RB1/RB2 [1] in my bam_dma patch and they don't have any BAM defined
>>>> upstream yet.
>>>>
>>>> [1]: https://lore.kernel.org/linux-arm-msm/CAH=2Ntw0BZH=RGp14mYLhX7D6jV5O5eDKRQbby=uCy85xMDU_g@xxxxxxxxxxxxxx/
>>>>
>>>> On Wed, Apr 05, 2023 at 12:58:32PM +0530, Bhupesh Sharma wrote:
>>>>> Add crypto engine (CE) and CE BAM related nodes and definitions to
>>>>> 'sm6115.dtsi'.
>>>>>
>>>>> Signed-off-by: Bhupesh Sharma <bhupesh.sharma@xxxxxxxxxx>
>>>>> ---
>>>>> arch/arm64/boot/dts/qcom/sm6115.dtsi | 22 ++++++++++++++++++++++
>>>>> 1 file changed, 22 insertions(+)
>>>>>
>>>>> diff --git a/arch/arm64/boot/dts/qcom/sm6115.dtsi b/arch/arm64/boot/dts/qcom/sm6115.dtsi
>>>>> index 2a51c938bbcb..ebac026b4cc7 100644
>>>>> --- a/arch/arm64/boot/dts/qcom/sm6115.dtsi
>>>>> +++ b/arch/arm64/boot/dts/qcom/sm6115.dtsi
>>>>> @@ -650,6 +650,28 @@ usb_hsphy: phy@1613000 {
>>>>> status = "disabled";
>>>>> };
>>>>>
>>>>> + cryptobam: dma-controller@1b04000 {
>>>>> + compatible = "qcom,bam-v1.7.4", "qcom,bam-v1.7.0";
>>>>> + reg = <0x0 0x01b04000 0x0 0x24000>;
>>>>> + interrupts = <GIC_SPI 247 IRQ_TYPE_LEVEL_HIGH>;
>>>>> + #dma-cells = <1>;
>>>>> + qcom,ee = <0>;
>>>>> + qcom,controlled-remotely;
>>>>> + num-channels = <8>;
>>>>> + qcom,num-ees = <2>;
>>>>> + iommus = <&apps_smmu 0x94 0x11>,
>>>>> + <&apps_smmu 0x96 0x11>;
>>>>> + };
>>>>> +
>>>>> + crypto: crypto@1b3a000 {
>>>>> + compatible = "qcom,sm6115-qce", "qcom,sm8150-qce", "qcom,qce";
>>>>> + reg = <0x0 0x01b3a000 0x0 0x6000>;
>>>>> + dmas = <&cryptobam 6>, <&cryptobam 7>;
>>>>> + dma-names = "rx", "tx";
>>>>> + iommus = <&apps_smmu 0x94 0x11>,
>>>>> + <&apps_smmu 0x96 0x11>;
>>>>
>>>> Shouldn't you have clocks = <&rpmcc RPM_SMD_CE1_CLK> here to make sure
>>>> the clock for the crypto engine is on? Your binding patch (PATCH 06/11)
>>>> says "Crypto Engine block on Qualcomm SoCs SM6115 and QCM2290 do not
>>>> require clocks strictly" but doesn't say why.
>>>>
>>>> Make sure you don't rely on having rpmcc keep unused clocks on
>>>> permanently. This is the case at the moment, but we would like to change
>>>> this [2]. Adding new users that rely on this broken behavior would just
>>>> make this effort even more complicated.
>>>>
>>>> If you also add the clock to the cryptobam then you should be able to
>>>> see the advantage of my bam_dma patch [3]. It allows you to drop
>>>> "num-channels" and "qcom,num-ees" from the cryptobam in your changes
>>>> above because it can then be read directly from the BAM registers.
>>>
>>> Thanks for pointing this out. Actually that's why I was using your
>>> patch while testing with RB1/RB2 :)
>>>
>>> Yes, so the background is that I am preparing a new version of this
>>> crypto enablement patchset.
>>> Also your assumption about the clocks being turned on by the firmware
>>> is true for RB1/RB2 devices, so enabling them via Linux is optional as
>>> per Qualcomm enggs.
>> This is not necessarily true. Currently it's kept always-on on
>> by clk_smd_rpm_handoff, but that's a hack from 10 years ago when smd
>> was still new.
>>
>>>
>>> So, I am testing the new patchset right now with 'clock' entries
>>> provided in the .dtsi and see if that causes any issue / improvement
>>> (etc.)
>> It won't change since it's on anyway, but that won't be a given for long.
>
> Right, so that's what I observe: RPM_SMD_CE1_CLK is always on by the
> time crypto _probe gets called.
> So, IMO let's not mix this patchset with the other fix which probably
> will fix the 10-year old clk_smd_rpm handoff keeping
> these clocks on.
>
> Probably that should be a separate changeset - requiring very thorough
> checks to make sure that we don't break
> working platforms.
It's not about mixing patchsets, the nodes should reflect all the clock/
power-domain/regulator/pinctrl/etc. dependencies from their introduction.
Remember, dt describes the hardware, not the software or firmware.

That - among other things - ensures backwards compatibility can be
preserved.
>
> Thanks.