Re: [PATCH v3 1/2] KVM: SVM: Increase X2AVIC limit to 4096 vcpus

[Sean Christopherson]

On Fri, Jul 18, 2025, Naveen N Rao wrote:
> On Mon, Jun 23, 2025 at 04:17:13PM -0700, Sean Christopherson wrote:
> > On Thu, Feb 20, 2025, Naveen N Rao (AMD) wrote:
> > > +		if (x2avic_4k_vcpu_supported) {
> > > +			x2avic_max_physical_id = X2AVIC_MAX_PHYSICAL_ID_4K;
> > > +			avic_physical_max_index_mask = AVIC_PHYSICAL_MAX_INDEX_4K_MASK;
> > > +		} else {
> > > +			x2avic_max_physical_id = X2AVIC_MAX_PHYSICAL_ID;
> > > +			avic_physical_max_index_mask = AVIC_PHYSICAL_MAX_INDEX_MASK;
> > > +		}
> > > +
> > > +		pr_info("x2AVIC enabled%s\n",
> > > +			x2avic_4k_vcpu_supported ? " (w/ 4K-vcpu)" : "");
> > 
> > Maybe print the max number of vCPUs that are supported?  That way there is clear
> > signal when 4k *isn't* supported (and communicating the max number of vCPUs in
> > the !4k case would be helpful too).
> 
> I'm tempted to go the opposite way and not print that 4k vCPUs are 
> supported by x2AVIC. As it is, there are many reasons AVIC may be 
> inhibited and lack of 4k vCPU support is just one other reason, but only
> for large VMs.

This isn't just about AVIC being inhibited though, it's about communicating
hardware support to the admin/user.  While I usually advocate *against* using
printk to log information, I find SVM's pr_info()s about what is/isn't enabled
during module load to be extremely useful, e.g. as sanity checks.  I (re)load
kvm-amd.ko on various hardware configurations on a regular basis, and more than
once the prints have helped me "remember" which platforms do/don't have SEV-ES,
AVIC, etc, and/or detect that I loaded kvm-amd.ko with the wrong overrides.

> Most users shouldn't have to care: where possible, AVIC will be enabled 
> by default (once that patch series lands). Users who truly care about 
> AVIC will anyway need to confirm AVIC isn't inhibited since looking at 
> the kernel log won't be sufficient. Those users can very well use cpuid 
> to figure out if 4k vCPU support is present.

If there wasn't already an "x2AVIC enabled" print, I would probably lean toward
doing nothing.  But since pr_info("x2AVIC enabled\n") already exists, and has
plently of free space for adding extra information, there's basically zero downside
to printing out the number of supported CPUs.  And it's not just a binary yes/no,
e.g. I would wager most people couldn't state the number of vCPUs supported by
the "old" x2AVIC.