Re: [RFT PATCH 0/4] hwmon: k10temp driver improvements

[Ondrej Äerman]

DÅa 17. 1. 2020 o 19:46 Guenter Roeck napÃsal(a):
> On Fri, Jan 17, 2020 at 10:46:25AM +0100, Ondrej Äerman wrote:
> > DÅa 16. 1. 2020 o 15:17 Guenter Roeck napÃsal(a):
> > > This patch series implements various improvements for the k10temp driver.
> > >
> > > Patch 1/4 introduces the use of bit operations.
> > >
> > > Patch 2/4 converts the driver to use the devm_hwmon_device_register_with_info
> > > API. This not only simplifies the code and reduces its size, it also
> > > makes the code easier to maintain and enhance.
> > >
> > > Patch 3/4 adds support for reporting Core Complex Die (CCD) temperatures
> > > on Ryzen 3 (Zen2) CPUs.
> > >
> > > Patch 4/4 adds support for reporting core and SoC current and voltage
> > > information on Ryzen CPUs.
> > >
> > > With all patches in place, output on Ryzen 3900 CPUs looks as follows
> > > (with the system under load).
> > >
> > > k10temp-pci-00c3
> > > Adapter: PCI adapter
> > > Vcore:        +1.36 V
> > > Vsoc:         +1.18 V
> > > Tdie:         +86.8ÂC  (high = +70.0ÂC)
> > > Tctl:         +86.8ÂC
> > > Tccd1:        +80.0ÂC
> > > Tccd2:        +81.8ÂC
> > > Icore:       +44.14 A
> > > Isoc:        +13.83 A
> > >
> > > The patch series has only been tested with Ryzen 3900 CPUs. Further test
> > > coverage will be necessary before the changes can be applied to the Linux
> > > kernel.
> > Hello everyone, I am the author of https://github.com/ocerman/zenpower/ .
> >
> > It is nice to see this merged.
> >
> > I just want to warn you that there have been reported issues with
> > Threadripper CPUs to zenpower issue tracker. Also I think that no-one tested
> > EPYC CPUs.
> >
> > Most of the stuff I was able to figure out by trial-and-error approach and
> > unfortunately because I do not own any Threadripper CPU I was not able to
> > test and fix reported problems.
> Thanks a lot for the note. The key problem seems to be that Threadripper
> doesn't report SoC current and voltage. Is that correct ? If so, that
> should be easy to solve.

Hello,

I thought that initially, but I was wrong. It seems like that these 
multi-node CPUs are reporting SOC and Core voltage/current data at 
particular node. Look at this HWiNFO64 screenshot of 2990WX for 
reference: https://i.imgur.com/yM9X5nd.jpg . They also may be using 
different addresses and/or factors.

> On a side note, drivers/gpu/drm/amd/include/asic_reg/thm/thm_10_0_offset.h
> suggests that two more temperature sensors might be available at 0x0005995C
> and 0x00059960 (DIE3_TEMP and SW_TEMP). Have you ever tried that ?
>
> Thanks,
> Guenter

I was aware of 0005995c and I thought that it could be Tdie3 (that's why 
I have included it in debug output, someone already shared that 3960X is 
reporting data on that address). I think this one can be safely included.

I was not aware of the other address, I will try it.

Ondrej.