Re: [PATCH] arm64: mpam: Document when to set arm64.nompam

Hi Ben,

On Thu, 3 Sept 2026 at 10:55, Ben Horgan [off-list ref] wrote:
...

quoted

+arm64.nompam
+============

This assumes that the reader already knows or can guess that arm64.nompam is a command line
parameter. Perhaps the section name can be 'Command line parameters' with a single 'arm64.nompam'
subsection underneath or all in the section name.

Will do: a "Command line parameters" section with arm64.nompam as its
subsection.

quoted

+Firmware must enable MPAM at the highest implemented exception level and
+leave the MPAM system registers accessible to the kernel, or trap the
+accesses and emulate MPAM as disabled. Where it does neither, the CPUs
+still advertise MPAM in the ID registers, the first kernel access to an
+MPAM register traps to EL3, and the boot fails. ``arm64.nompam`` exists
+for that firmware: it makes the kernel treat the CPUs as not implementing
+MPAM, so no MPAM system register is accessed. Set it only on a machine
+that does not boot without it.
+
+It is not a way to turn MPAM off. On a system whose firmware has enabled
+MPAM at EL3, the option leaves the trap controls in MPAM2_EL2 and> +MPAMHCR_EL2 unwritten, and their reset values are UNKNOWN.

Enabling MPAM at EL3 is separate from disabling traps. Setting MPAM3_EL3.MPAMEN to 1 means that the
values in MPAMn_ELx are used to determine the PARTID and PMG for CPU requests. At this point EL3 can
configure the PARTID and PMG for the different exception levels and MPAM can be used in the system
under firmware control. Setting MPAM3_EL3.TRAPLOWER to 0 is what enables EL2 to access the MPAM
registers without trapping and configure the traps for MPAM register accesses at EL1.

Thanks, I'll separate the two: TRAPLOWER being clear is what leaves
the EL2 trap controls unwritten, and MPAMEN is what makes a guest's
choice of PARTID and PMG take effect.

quoted

KVM still
+hides MPAM from guests but no longer enables the traps that stop a guest
+from using it, so a guest may be able to choose its own PARTID and PMG.

Shouldn't this also mention that the MPAM_*_EL1 are not save restored and so if the traps aren't
enabled provide a side channel between guests.

Yes, I'll add that KVM does not save or restore MPAM0_EL1 and
MPAM1_EL1, so what one guest writes is still there for the next guest
on that CPU.

I'll send a v2 with these once others have had a chance to comment as well.

Cheers,
/fuad

Thanks,

Ben

quoted

 Reporting Bugs
 ==============
 If you are not seeing the counters or controls you expect please share the