Received-SPF: pass (google.com: domain of linux-kernel+bounces-91008-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45d1:ec00::1 as permitted sender) client-ip=2604:1380:45d1:ec00::1;
Message-ID: <98e154f7-a01d-43d0-bd0b-70122ad880c6@linaro.org>
Date: Mon, 4 Mar 2024 18:36:40 +0100
Precedence: bulk
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH] interconnect: qcom: x1e80100: Remove inexistent ACV_PERF
 BCM
To: Mike Tipton <quic_mdtipton@quicinc.com>
Cc: Bjorn Andersson <andersson@kernel.org>, Georgi Djakov
 <djakov@kernel.org>, Abel Vesa <abel.vesa@linaro.org>,
 Sibi Sankar <quic_sibis@quicinc.com>,
 Rajendra Nayak <quic_rjendra@quicinc.com>,
 Marijn Suijten <marijn.suijten@somainline.org>,
 linux-arm-msm@vger.kernel.org, linux-pm@vger.kernel.org,
 linux-kernel@vger.kernel.org
References: <20240302-topic-faux_bcm_x1e-v1-1-c40fab7c4bc5@linaro.org>
 <20240304164009.GB25492@hu-mdtipton-lv.qualcomm.com>
Content-Language: en-US
From: Konrad Dybcio <konrad.dybcio@linaro.org>
In-Reply-To: <20240304164009.GB25492@hu-mdtipton-lv.qualcomm.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit


On 3/4/24 17:40, Mike Tipton wrote:
> On Sat, Mar 02, 2024 at 03:22:49AM +0100, Konrad Dybcio wrote:
>> Booting the kernel on X1E results in a message like:
>>
>> [    2.561524] qnoc-x1e80100 interconnect-0: ACV_PERF could not find RPMh address
>>
>> And indeed, taking a look at cmd-db, no such BCM exists. Remove it.
>>
>> Fixes: 9f196772841e ("interconnect: qcom: Add X1E80100 interconnect provider driver")
>> Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
> 
> Reviewed-by: Mike Tipton <quic_mdtipton@quicinc.com>
> 
> For some background, ACV "perf mode" does exist, but not as a separate
> BCM. It's controlled by a separate bit in the ACV mask. By default, the
> ACV node only sets the bit indicating the HLOS voter. It doesn't assert
> the perf mode bit.
> 
> Enabling perf mode toggles different trade-offs within the DDR subsystem
> for slightly improved performance at the expense of slightly higher
> power. There are limited use cases of this downstream, where we expose
> control over this bit to clients through icc_set_tag(). It primarily
> improves certain latency sensitive benchmarks, AFAIK. I don't think it
> has much impact on real world use cases.
> 
> This is true for many other targets as well, not just x1e80100.
> 
> Voting for perf-mode is entirely optional and in most cases also
> entirely unnecessary. So, removing this broken way to control it without
> adding the proper control is totally fine.
> 
> I have a local series to add the proper support, but haven't posted it
> yet. There aren't any users for it upstream yet, nor am I aware of any
> near term plans to add them. So, it would be unused for a little while,
> at least. That said, anybody could use it to set that tag on their BW
> votes on the off-chance it improves performance and they don't care
> about the power trade-offs.
> 
> I could post the series soon if there's interest.

I think adding a sysfs entry for toggling this could be very interesting.

Userspace could toggle this based on "power profile"-style settings.

Konrad