[Bug 221796] New: intel_pmc_core: Arrow Lake-HX (Core Ultra 9 275HX) uses ARL-H DMU GUID and fails to expose Die C6 telemetry / existing

[Bug 221796] New: intel_pmc_core: Arrow Lake-HX (Core Ultra 9 275HX) uses ARL-H DMU GUID and fails to expose Die C6 telemetry / existing

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=221796

            Bug ID: 221796
           Summary: intel_pmc_core: Arrow Lake-HX (Core Ultra 9 275HX)
                    uses ARL-H DMU GUID and fails to expose Die C6
                    telemetry / existing
           Product: Drivers
           Version: 2.5
          Hardware: Intel
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P3
         Component: Platform_x86
          Assignee: drivers_platform_x86@kernel-bugs.osdl.org
          Reporter: marek.suranic@gmail.com
        Regression: No

HARDWARE / SYSTEM
=================

Laptop: MSI Vector 16 HX AI A2XW
CPU: Intel Core Ultra 9 275HX
CPU family: 6
CPU model: 198 (0xc6)
CPU stepping: 2
CPU topology: 24 cores, 1 thread per core
Operating system: Fedora Linux 44
Kernel: 7.1.4-204.fc44.x86_64
Architecture: x86_64
Preemption model: PREEMPT_DYNAMIC


PROBLEM
=======

With the unmodified intel_pmc_core module, the kernel reports:

    intel_pmc_core INT33A1:00: pmc_core: couldn't get DMU telem endpoint -19
    intel_pmc_core INT33A1:00: initialized

The driver otherwise initializes, but the following debugfs entry is
missing:

    /sys/kernel/debug/pmc_core/die_c6_us_show


RELEVANT INTEL PMT ENDPOINT
===========================

The platform exposes the following telemetry endpoint:

    /sys/class/intel_pmt/telem4/guid:   0x1a06a101
    /sys/class/intel_pmt/telem4/size:   136
    /sys/class/intel_pmt/telem4/offset: 0

The available GUID 0x1a06a101 corresponds to
ARL_H_PMT_DMU_GUID.

The Core Ultra 9 275HX has CPU model 0xc6 and is selected through the
generic Arrow Lake configuration. The generic ARL_PMT_DMU_GUIDS list
only checks ARL_PMT_DMU_GUID (0x1a06a102), so the available
0x1a06a101 endpoint is not found.


TESTED PATCH
============

    -static u32 ARL_PMT_DMU_GUIDS[] =
    -    {ARL_PMT_DMU_GUID, 0x0};
    +static u32 ARL_PMT_DMU_GUIDS[] =
    +    {ARL_PMT_DMU_GUID, ARL_H_PMT_DMU_GUID, 0x0};


RESULT
======

After rebuilding and signing intel_pmc_core against kernel
7.1.4-204.fc44.x86_64:

- The "couldn't get DMU telem endpoint -19" message disappeared.
- intel_pmc_core initialized successfully.
- /sys/kernel/debug/pmc_core/die_c6_us_show became available.
- The Die C6 telemetry value became readable.

Command and output:

    $ sudo cat /sys/kernel/debug/pmc_core/die_c6_us_show
    31917705

Relevant kernel log after applying the patch:

    intel_pmc_core: loading out-of-tree module taints kernel.
    intel_pmc_core INT33A1:00: initialized

The out-of-tree taint is expected because the patched module was built
and installed separately for testing.


CONCLUSION
==========

This suggests that Arrow Lake-HX model 0xc6 systems may need
ARL_H_PMT_DMU_GUID as a fallback in ARL_PMT_DMU_GUIDS.

-- 
You may reply to this email to add a comment.

You are receiving this mail because:
You are watching the assignee of the bug.