feat: dj with the ryzen_smu kernel module on a few systems and I have a patch that improves monitoring support for newer/less-covered PM tables.

[inode64]

Tested systems:

• Dragon Range / Raphael-compatible system
  • PM table: 0x540104
  • SMU FW: 4.84.4.0
• AMD EPYC 4344P
  • Family/model: Zen 4, model 97
  • PM table: 0x540104
  • SMU FW: 84.79.225
• AMD Ryzen 5 3400G / Picasso
  • PM table: 0x1e0004
  • SMU FW: 4.30.92.0

Main changes:

• Added support for reading pm_table_version and pm_table_size from the ryzen_smu sysfs interface.
• Added a fallback path for PM table version/address handling when using the kernel module on families where the usual SMU table messages are not available.
• Added SMU firmware version reporting using SMU message 0x2 with arg0 = 1.
• Fixed PM table initialisation for 0x540104: offset 0x0 is legitimately zero on this table, so the empty-table check now uses offset 0x8.
• Added monitored values for 0x540104, including:
  • fast PPT limit/value
  • socket power
  • SoC voltage
  • FCLK
  • MEM clock
  • Tctl limit/value
  • VDD TDC value
  • per-core power, voltage, temperature, requested clock, effective clock, C0, CC1 and C6 residency
• Added conservative VDD TDC/EDC limit detection for 0x540104:
  • TDC LIMIT VDD from 0x20 when the value is plausible
  • EDC LIMIT VDD from 0xF4 when the value is plausible
  • sentinel/unlimited values like 1000A are hidden instead of displayed as real limits
• Improved Tctl reporting for 0x540104:
  • Dragon Range/Raphael systems track Tctl at 0x2C
  • EPYC 4004 tracks the real loaded Tctl at 0xF8
  • the code now uses the higher plausible value, matching k10temp on the EPYC system
• Added Picasso 0x1e0004 monitoring improvements:
  • fixed Tctl value using the hotter core from the later temperature block
  • added per-core power, voltage, temperature and clock
  • added GFX clock
  • added MEM clock
• Made the --info output cleaner:
  • extended values are only printed when the getter returns a valid value
  • per-core columns are dynamic, so unsupported columns are not shown as nan
• Added portable x86_64 build options:
  • ENABLE_IPO=OFF
  • STRIP_X86_ISA_NOTE=ON
  • this avoids binaries being rejected on older x86_64 CPUs because of an overly strict .note.gnu.property ISA marker

Validation:

• The portable binary was tested remotely on:
  • Ryzen 5 3400G / Picasso
  • EPYC 4344P / PM table 0x540104
• On the EPYC 4344P, loaded k10temp reported around 82.8-83.1 C, and RyzenAdj now reports THM VALUE CORE 83.062, matching the kernel sensor.
• On Picasso, the new per-core table and corrected Tctl were validated against sensors, cpufreq and PM table dumps.

Some values are intentionally still hidden as nan because I could not validate stable offsets yet, especially STAPM/slow PPT on 0x540104 and some SoC/EDC value fields.