Experimental Loongson3 CPUFreq ABIv1 driver

[ziyao233]

Topic Description

• linux-kernel: Support cpufreq on Loongson 3 with older SMC firmware
  Introduces the experimental downstream patch for CPUFreq support on
  Loongson 3 machines with older SMC firmware (CMD_GET_VERSION returns 0
  instead of a value greater than 0).
  Signed-off-by: Yao Zi [email protected]

Package(s) Affected

• linux-kernel-${__VER}: 6.14.9-3

Security Update?

No

Build Order

#buildit linux-kernel

Test Build(s) Done

Primary Architectures

• LoongArch 64-bit loongarch64

[xry111]

I really don't like the terminology "ABIv1."

Those pairs are in fact, completely different and orthogonal concepts:

• Downstream "Linux 4.19.0" UAPI vs. Linux 5.19.0 UAPI
• Stack-based relocs vs. relocs directly operating on the immediate slot
• BPI vs. UEFI
• SMC shipped with current HW vs. SMC implementing the interface expected by the upstream kernel
• Maybe others...

Mixing up those concepts into "ABIv1" vs "ABIv2" has already caused many misunderstandings.

[ziyao233]

On Fri, Jun 06, 2025 at 03:55:20AM -0700, Xi Ruoyao wrote: xry111 left a comment (AOSC-Dev/aosc-os-abbs#11270) I really don't like the terminology "ABIv1." Those pairs are in fact, completely different and orthogonal concepts: - Downstream "Linux 4.19.0" UAPI vs. Linux 5.19.0 UAPI - Stack-based relocs vs. relocs directly operating on the immediate slot - BPI vs. UEFI - SMC shipped with current HW vs. SMC implementing the interface expected by the upstream kernel - Maybe others... Mixing up those concepts into "ABIv1" vs "ABIv2" *has already* caused many misunderstandings.

Yeah, I've considered the risk of mixing things up as well, but didn't come up with a reasonable name. So I just stopped waiting and made up a quick patch with a draft commit message for testing. As suggested by @xen0n, maybe SMCv0 is a better name, since the easist way to distinguish these two versions of firmware is by return value of CMD_GET_VERSION, where the older returns 0.

…

-- Reply to this email directly or view it on GitHub: #11270 (comment) You are receiving this because you authored the thread. Message ID: ***@***.***>

[ziyao233]

@KatyushaScarlet reported that the CPUFreq driver causes frequently amdgpu resets on TongFang L820-T2 when using amdgpu instead of radeon driver. I'll dig it further later.

[KatyushaScarlet]

@KatyushaScarlet reported that the CPUFreq driver causes frequently amdgpu resets on TongFang L820-T2 when using amdgpu instead of radeon driver. I'll dig it further later.

1. Log 1: After unlocking the screen in KDE6 and clicking the tray icon, the desktop froze, blinked, and reported a GPU reset.
   dmesg-l860-t2-amdgpu-reset-6.14.9-log1.log

2. Log 2: After reproducing the same steps, the GPU reset occurred. I then removed the loongson3_cpufreq driver using rmmod, and the GPU stopped resetting. However, when I reinserted the driver using modprobe, the GPU reset again.
   dmesg-l860-t2-amdgpu-reset-6.14.9-log2.log