Re: [PATCH v1] clk: qcom: clk-rpmh: Add IPA clock support

From: David Dai <daidavid1@codeaurora.org>
To: Stephen Boyd <sboyd@kernel.org>,
	linux-arm-msm@vger.kernel.org, linux-clk@vger.kernel.org,
	linux-kernel@vger.kernel.org
Cc: georgi.djakov@linaro.org, bjorn.andersson@linaro.org,
	evgreen@google.com, tdas@codeaurora.org, elder@linaro.org
Subject: Re: [PATCH v1] clk: qcom: clk-rpmh: Add IPA clock support
Date: Wed, 16 Jan 2019 16:54:39 -0800	[thread overview]
Message-ID: <7db43e9e-0f5c-9c26-6f26-c601ec69f964@codeaurora.org> (raw)
In-Reply-To: <154748442276.169631.994019688641603420@swboyd.mtv.corp.google.com>

On 1/14/2019 8:47 AM, Stephen Boyd wrote:
> Quoting David Dai (2019-01-11 16:56:14)
>> On 1/9/2019 11:28 AM, Stephen Boyd wrote:
>>> Quoting David Dai (2018-12-13 18:35:04)
>>>> +
>>>> +#define BCM_TCS_CMD(valid, vote)                               \
>>>> +       (BCM_TCS_CMD_COMMIT_MASK |                              \
>>>> +       ((valid) << BCM_TCS_CMD_VALID_SHIFT) |                  \
>>>> +       ((cpu_to_le32(vote) &                                   \
>>> Why?
>> Sorry, could you clarify this question? If you're referring to the
>> cpu_to_le32, shouldn't we be explicit about converting endianness even
>> if it might be redundant for this particular qcom platform?
> Is only the vote part of the message in little endian format and the
> rest is native CPU endianess? It's very odd to see that jammed in the
> middle of a bit packing statement like that. Typically the whole u32
> would be in one or the other endianness. Also, sparse right complains
> about this macro and it's usage, so something is wrong.
Point taken, I'll leave it out of the macro for now.
> I think one other problem is that rpmh API doesn't really talk about
> endianness here and that's busted. So if you want to fix endianness
> issues that needs to be fixed first.

Since a similar macro is being used as part of the interconnect provider 
driver, I was thinking a better place for this macro might be in the 
tcs.h as part of the rpmh driver? I could submit a different patch for 
rpmh that mentions endianness along with this change.

>>>> @@ -29,6 +54,7 @@
>>>>     * @aggr_state:                rpmh clock aggregated state
>>>>     * @last_sent_aggr_state: rpmh clock last aggr state sent to RPMh
>>>>     * @valid_state_mask:  mask to determine the state of the rpmh clock
>>>> + * @aux_data:          data specific to the bcm rpmh resource
>>> But there isn't aux_data. Is this supposed to be unit? And what is unit
>>> going to be? 1000?
>> Supposed to be unit, the value is on the order of Khz.
> Ok, hopefully the kernel-doc comes out easy to understand.
>
>>>> @@ -217,6 +340,7 @@ static unsigned long clk_rpmh_recalc_rate(struct clk_hw *hw,
>>>>    DEFINE_CLK_RPMH_VRM(sdm845, rf_clk1, rf_clk1_ao, "rfclka1", 1);
>>>>    DEFINE_CLK_RPMH_VRM(sdm845, rf_clk2, rf_clk2_ao, "rfclka2", 1);
>>>>    DEFINE_CLK_RPMH_VRM(sdm845, rf_clk3, rf_clk3_ao, "rfclka3", 1);
>>>> +DEFINE_CLK_RPMH_BCM(sdm845, ipa, "IP0");
>>> It's really IP0?! What was wrong with IPA?
>> Yeah... convention seems to be 2 letters and then a number for most BCM
>> resources.
> OK.
>
-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project