linux-kernel - Re: [Regression] 6.11.0-rc1: AMD CPU boot with error when CPPC feature disabled by BIOS

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-Id: <df99663b-23bd-43ac-b277-cd6700f16809@app.fastmail.com>
Date: Thu, 26 Sep 2024 13:56:21 -0700
From: "Luna Nova" <lists@...nova.dev>
To: "Gautham R.Shenoy" <gautham.shenoy@....com>
Cc: perry.yuan@....com, Xiaojian.Du@....com, linux-kernel@...r.kernel.org,
 linux-pm@...r.kernel.org, "David Wang" <00107082@....com>
Subject: Re: [Regression] 6.11.0-rc1: AMD CPU boot with error when CPPC feature
 disabled by BIOS

Hi Gautham,

I'm seeing the same message on a server board with an EPYC Rome 7K62 CPU.
CPPC is set to enabled in the UEFI firmware settings.

Kernel: 6.11.0 (6.11.0 #1-NixOS SMP PREEMPT_DYNAMIC Sun Sep 15 14:57:56 UTC 2024 x86_64 GNU/Linux)
Board: Gigabyte MZ22-G20-00 Rev 1.0 (in a G292-Z20 Rev 100)
UEFI Firwmare: R23_F01 (2021-09-06, latest available version at time of this message)
AGESA PI Version 1.0.0.C.

CONFIG_ACPI_CPPC_LIB=y
CONFIG_X86_AMD_PSTATE=y
CONFIG_X86_AMD_PSTATE_DEFAULT_MODE=3
CONFIG_X86_AMD_PSTATE_UT=m

$ cat /proc/cmdline
initrd=\EFI\nixos\z16gakzlwypxbjzm5y93x10cjmxjvial-initrd-linux-6.11-initrd.efi init=/nix/store/cqhw9x7w7dc3avwri4i2lk0mgc31arll-nixos-system-tsukiakari-nixos-24.11/init sysrq_always_enabled fsck.mode=force loglevel=4 audit=0 amd_pstate=guided amd_pstate.shared_mem=1 amdgpu.lockup_timeout=10000,10000,10000,10000
$ sudo dmesg | grep pstate
amd_pstate: min_freq(0) or max_freq(0) or nominal_freq(0) value is incorrect
(Repeats for each core)
amd_pstate: failed to register with return -19
stage-1-init: [Thu Sep 26 20:04:53 UTC 2024] loading module amd_pstate_ut...
amd_pstate_ut: 1    amd_pstate_ut_acpi_cpc_valid  success!
amd_pstate_ut: 2    amd_pstate_ut_check_enabled   success!
amd_pstate_ut: 3    amd_pstate_ut_check_perf      success!
amd_pstate_ut: 4    amd_pstate_ut_check_freq      success!

It seems odd that amd_pstate fails to load but amd_pstate_ut reports success for all checks.

> it appears that the CPPC version on your platform is v2 which does not
> advertise the nominal_freq and the lowest_freq. In the absence of these,
> it is not possible for the amd-pstate driver to infer the
> min/max_freq. Which is why the driver bails at this later stage.

> The way around it is to add a quirk for your BIOS as done in this commit
> from Perry:
> eb8b6c368202 ("cpufreq: amd-pstate: Add quirk for the pstate CPPC capabilities missing")

Perry's patch you referenced as an example above targets the same 7K62 CPU but requires one specific BIOS version.
Should I submit a patch adding the version on this system to that quirk?

I'm confused by the quirk code: it's called "AMD EPYC 7K62" but it matches by BIOS revision and doesn't check the CPU model.
An earlier version of the quirk included `boot_cpu_data.x86 == 0x17 && boot_cpu_data.x86_model == 0x31` to check the model; it now uses the nominal frequencies for a 7K62 regardless of the CPU model if the BIOS revision matches.

Best,
Luna