On Tue, Dec 23, 2025 at 2:01 PM Alexandre Demers <[email protected]> wrote: > > Hi again Alex, > > First, a follow up on VDDCI. From some sources I've found, I > understand that it would be the VDD at the interface. Which interface > is unclear, but one source was referring to the voltage at the PCIe > interface. Would that make sense?
I'm pretty sure it's the memory interface. Power team said same thing from what they can remember. > > Next, I'd like your thoughts and inputs on a project I'd like to work > on (in fact, I've been working on it in the last few weeks already). > I'd like to bring BACO on par with CI. I know there is a minimal > implementation for reset purposes already present. I think it is There is no BACO support in the driver for SI yet. The reset support for SI doesn't use BACO. > linked to the legacy dpm. And then, I'd like to go as far as possible > in moving from the legacy DPM code to use the same (or similar) code > path used by CI+. Do you think this could work? You could port the existing SI dpm code to the powerplay code base that is used by CI, but you can't reuse the CI code directly. It would need its own BACO sequence. I should be able to dig that up for you if you get to that point. > > Also, was SMU6 significantly different from SMU7? Looking at how the > driver sends messages to SMU7 and 8, it looks like an iterative > process where new messages/features were added to the previous SMU > version. Can we expect SMU6 to work mostly like the two subsequent > versions with a subset of their supported messages/features? Sort of. We've been iterating on SMU firmware interfaces since R7xx. There are a number of differences. We supported both SI and CI in radeon with the same framework. When we started amdgpu, we inherited the radeon interface there too, but later ported CI to the powerplay interface. So it's doable. You'll have to add smu6 equivalents of smu7_hwmgr.c and smu7_smumgr.c, add powerplay table parsing, etc. You can reference the existing code bases in amdgpu and radeon. > > Finally, was SMU6 able to use any of the power tune features? CI is > able to receive a request to use one of a few power profiles. Was this > only available from CI or was it also present under SI? The powertune stuff is already there. For the profiles, do you mean the PP_SMC_POWER_PROFILE_* stuff? That was only available on SMU7. > > Thank you very much for your time and knowledge and have a great > Holidays season. Happy New Year! Alex > > On Mon, Dec 8, 2025 at 9:32 AM Alex Deucher <[email protected]> wrote: > > > > On Sat, Dec 6, 2025 at 11:30 AM Alexandre Demers > > <[email protected]> wrote: > > > > > > OK, I was hoping for a similar SMU7 implementation never completed. > > > > > > I see both VDDC and VDDCI (for Evergreen+) values available. When > > > looking at amdgpu_pm_info, VDDC would be reported as VDDGFX I suppose > > > (and inX_label under /sys/class/drm/cardX/device/hwmon/hwmonY/), isn't > > > it? > > > > Yes. > > > > > > > > What would VDDCI correspond to? I mean, how should it be displayed > > > (maybe it has its specific inX_label)? > > > > I don't recall off hand what voltage domain VDDCI was off hand. I > > think it may have been the memory interface. I'll see if I can dig it > > up. > > > > Alex > > > > > > > > Cheers, > > > Alexandre > > > > > > On Mon, Dec 1, 2025 at 10:26 AM Alex Deucher <[email protected]> > > > wrote: > > > > > > > > On Sat, Nov 29, 2025 at 8:44 PM Alexandre Demers > > > > <[email protected]> wrote: > > > > > > > > > > Hi, > > > > > > > > > > I was wondering if SI GPUs support reporting power and voltage values > > > > > at the hardware level? I read somewhere that it was supported from > > > > > GCN1.2 (Tonga and over), but I haven't found anything on prior > > > > > generations. Going through the register names, I wasn't able to > > > > > identify any who may correspond to power or voltage level. > > > > > > > > > > So I'm asking, just in case it is supported and I missed the > > > > > information while searching for it. > > > > > > > > Correct. There is no interface to query the power. For voltages you > > > > can look at mmTARGET_AND_CURRENT_PROFILE_INDEX to find out the current > > > > DPM level and then convert that into struct si_ps to look up the > > > > voltages, similar to what is done for sclk and mclk. See > > > > si_dpm_read_sensor(). > > > > > > > > Alex
