Re: [PATCH v11 00/23] Allow AET to use PMT as loadable module

[Luck, Tony]

On Mon, Aug 31, 2026 at 10:43:58AM -0700, Tony Luck wrote:
> Requiring INTEL_PMT_TELEMETRY=y to enable AET is a functional workaround
> to enable enumeration of Application Energy Telemetry (AET) events, but
> unacceptable to many users. It results in increased configuration complexity,
> increased kernel memory footprint and inability to patch problems by unloading
> a module and loading an updated version.
> 
> Add a registration function to the AET code that can be used by
> INTEL_PMT_TELEMETRY to provide the enumeration functions.
> 
> INTEL_PMT_TELEMETRY can be loaded/unloaded independently of
> resctrl file system mount/unmount. Perform enumeration on
> every mount and cleanup on every unmount.

Summary of Sashiko findings:
https://sashiko.dev/#/patchset/20260831174421.13921-1-tony.luck%40intel.com

 [PATCH v11 09/23] fs/resctrl: Remove redundant calls to resctrl_mon_capable()

Pre-existing issue that systems supporting only monitoring will not
report any tasks in the default group. Issue is real. Sashiko suggests
removing the check for resctrl_alloc_capable(). I should have done this
anyway for the same optimization reasons that I removed resctrl_mon_capable()
from is_rmid_match().


 [PATCH v11 16/23] platform/x86/intel/pmt: Register enumeration functions with resctrl

Pre-existing potential race. Sashiko suggests calling intel_pmt_get_features()
under ep_lock.


 [PATCH v11 18/23] fs/resctrl: Call arch code for every mount

Two broken bits, both with easy (I hope) fixes:

1) rdt_get_tree() should "return ret;" not "return 0;" 
2) resctrl_arch_unmount() called in error path before cpus_read_unlock()
   could deadlock. Move the resctrl_arch_unmount() after the unlock.


 [PATCH v11 20/23] platform/x86/intel/pmt: Inform resctrl when MMIO maps are being removed

Sashiko is concerned about non-PCI telemetry end points. Not a real
issue. Endpoints are enumerated using PCIe DVSEC - so they have to be
PCIe devices.

Pollution of kernel log with dev_warn(). Ilpo also flagged this. It's a
"can't happen" test. I will downgrade severity to dev_dbg()

Pre-existing issue in telemetry driver that other places need to handle
unbind/remove of devices and unmap of MMIO.

 [PATCH v11 23/23] x86/resctrl: Document telemetry mount timing caveat

I'm still struggling to come up with a systemd/udev way to delay
mounting the resctrl file system until after telemetry enumeration
is complete. Sashiko says there are gaps in the example I provide.
The whole namespace issue is a mystery to me :-(

-Tony