[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CALHNRZ-660RVcYLo9Pxxj9gz1s0x4nLYOSFwbtiEwSU1qbvA5Q@mail.gmail.com>
Date: Tue, 2 Sep 2025 19:55:36 -0500
From: Aaron Kling <webgeek1234@...il.com>
To: Mikko Perttunen <mperttunen@...dia.com>
Cc: Krzysztof Kozlowski <krzk@...nel.org>, Rob Herring <robh@...nel.org>, Conor Dooley <conor+dt@...nel.org>,
Thierry Reding <thierry.reding@...il.com>, Jonathan Hunter <jonathanh@...dia.com>,
Krzysztof Kozlowski <krzk+dt@...nel.org>, MyungJoo Ham <myungjoo.ham@...sung.com>,
Kyungmin Park <kyungmin.park@...sung.com>, Chanwoo Choi <cw00.choi@...sung.com>,
Dmitry Osipenko <digetx@...il.com>, linux-kernel@...r.kernel.org, devicetree@...r.kernel.org,
linux-tegra@...r.kernel.org, linux-pm@...r.kernel.org
Subject: Re: [PATCH RFC 0/7] Support Tegra210 actmon for dynamic EMC scaling
On Tue, Sep 2, 2025 at 6:51 PM Mikko Perttunen <mperttunen@...dia.com> wrote:
>
> On Friday, August 29, 2025 1:01 PM Aaron Kling via B4 Relay wrote:
> > This series adds interconnect support to tegra210 MC and EMC, then
> > enables actmon. This enables dynamic emc scaling.
> >
> > This series is marked RFC for two reasons:
> >
> > 1) Calculating rate from bandwidth usage results in double the expected
> > rate. I thought this might be due to the ram being 64-bit, but the
> > related CFG5 register reports 32-bit on both p2371-2180 and
> > p3450-0000. I'm using the calculation used for Tegra124 and haven't
> > seen seen anything obviously different between the ram handling on
> > these archs to cause a different result. I have considered that the
> > number of channels might affect the reporting, and factoring in that
> > variable does result in the correct rate, but I don't want to assume
> > that's correct without confirmation.
>
> My thinking is also that this is due to the channels. L4T says
>
> /*
> * Tegra11 has dual 32-bit memory channels, while
> * Tegra12 has single 64-bit memory channel. Tegra21
> * has either dual 32 bit channels (LP4) or a single
> * 64 bit channel (LP3).
> *
> * MC effectively operates as 64-bit bus.
> */
>
> next to calculating bw_to_freq, and proceeds to use the same divisor for T114 to T210. Regarding the CFG5_DRAM_WIDTH field, I'm guessing it gives the width for one channel, but I'm not sure how it would function for other memory types -- I'm not sure if any Tegra210 devices using memory other than LPDDR4 were ever released.
Mmm. "MC effectively operates as 64-bit bus." So I could just hardcode
64-bit dram width and skip reading the CFG5_DRAM_WIDTH field
altogether. And regardless of the layout, if I'm reading that
correctly, the calculation would remain correct. That should get the
numbers I'm expecting on the devkits, but I will verify again before
uploading a new revision.
Aaron
Powered by blists - more mailing lists