Re: [PATCH v6 17/18] drm/xe/vsec: Update PMT internal access

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] Uninitialized MMIO offset used in PMT telemetry reads during hardware monitor registration, causing invalid MMIO accesses.
--

commit d59cca0476ebbe3cb43377a85bcbb964d1b8401f
Author: Michael J. Ruhl <michael.j.ruhl@intel.com>

drm/xe/vsec: Update PMT internal access

Current PMT internal access assumes BMG devices. Use the new PMT GUID API
to read the device specific guid, and cleanup newly unused mmio parameter.

[ ... ]

> diff --git a/drivers/gpu/drm/xe/xe_hwmon.c b/drivers/gpu/drm/xe/xe_hwmon.c
> index b9740224e07ff..c44a7d24816d1 100644
> --- a/drivers/gpu/drm/xe/xe_hwmon.c
> +++ b/drivers/gpu/drm/xe/xe_hwmon.c
> @@ -517,9 +517,15 @@ xe_hwmon_energy_get(struct xe_hwmon *hwmon, int channel, long *energy)
>  
>  	if (hwmon->xe->info.platform == XE_BATTLEMAGE) {
>  		u64 pmt_val;
> +		u32 guid;
>  
> -		ret = xe_pmt_telem_read(hwmon->xe->drm.dev,
> -					xe_mmio_read32(mmio, BMG_PUNIT_TELEMETRY_GUID),
> +		ret = xe_vsec_get_guid(hwmon->xe, &guid);
> +		if (ret) {
> +			drm_warn(&hwmon->xe->drm, "PMT device is not available: %d\n", ret);
> +			*energy = 0;
> +			return;
> +		}
> +		ret = xe_pmt_telem_read(hwmon->xe->drm.dev, guid,

[Severity: High]
This is a pre-existing issue, but does this telemetry read target an invalid
MMIO address during hardware monitor registration?

xe_device_probe() calls xe_hwmon_register() before xe_vsec_init().
xe_hwmon_register() internally calls xe_hwmon_energy_get() during
preregistration, triggering this xe_pmt_telem_read().

Because xe->pmt.base_offset is still uninitialized (0) at this point, the
read in xe_pmt_telem_read():

drivers/gpu/drm/xe/xe_vsec.c:xe_pmt_telem_read() {
    void __iomem *telem_addr = xe->mmio.regs + xe->pmt.base_offset;
    ...
    telem_addr += offset + user_offset;
    ...
    memcpy_fromio(data, telem_addr, count);
}

seems to evaluate to an incorrect platform-specific telemetry offset (e.g.,
xe->mmio.regs + 0 + 0x0230) instead of the correct base. Reading unmapped
PCI address space can cause a PCIe Unsupported Request and Machine Check
Exception during boot on BMG platforms.

>  					&pmt_val, BMG_ENERGY_STATUS_PMT_OFFSET,	sizeof(pmt_val));

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260904182451.1164868-20-michael.j.ruhl@intel.com?part=17