On Mon, Mar 20, 2023 at 02:45:15PM +0300, Paul Fertser wrote:

Hello,

We are seeing wrong DTS temperatures on at least "Intel(R) Xeon(R)
Bronze 3204 CPU @ 1.90GHz" and most probably other Skylake Xeon CPUs
are also affected, see inline.

Thanks for the report! I guess we need a fix for this indeed.

On Tue, Feb 08, 2022 at 04:36:36PM +0100, Iwona Winiarska wrote:

quoted

Add peci-cputemp driver for Digital Thermal Sensor (DTS) thermal
readings of the processor package and processor cores that are
accessible via the PECI interface.

...

quoted

+static const struct cpu_info cpu_hsx = {
+	.reg		= &resolved_cores_reg_hsx,
+	.min_peci_revision = 0x33,
+	.thermal_margin_to_millidegree = &dts_eight_dot_eight_to_millidegree,
+};
+
+static const struct cpu_info cpu_icx = {
+	.reg		= &resolved_cores_reg_icx,
+	.min_peci_revision = 0x40,
+	.thermal_margin_to_millidegree = &dts_ten_dot_six_to_millidegree,
+};

...

quoted

+	{
+		.name = "peci_cpu.cputemp.skx",
+		.driver_data = (kernel_ulong_t)&cpu_hsx,
+	},

With this configuration we get this data:

/sys/bus/peci/devices/0-30/peci_cpu.cputemp.skx.48/hwmon/hwmon15# grep . temp[123]_{label,input}
temp1_label:Die
temp2_label:DTS
temp3_label:Tcontrol
temp1_input:30938
temp2_input:67735
temp3_input:80000

On the host system "sensors" report

Package id 0:  +31.C (high = +80.C, crit = +90.C)

So I conclude Die temperature as retrieved over PECI is correct while
DTS is mis-calculated. The old downstream code in OpenBMC was using
ten_dot_six_to_millidegree() function for conversion, and that was
providing expected results. And indeed if we reverse the calculation
here we get 80000 - ((80000-67735) * 256 / 64) = 30940 which matches
expectations.

-- 
With Best Regards,
Andy Shevchenko



_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel