Re: [kvm-unit-tests PATCH] arm: Remove MAX_SMP probe loop

[Andrew Jones <andrew.jones@linux.dev>]

On Thu, Jan 05, 2023 at 11:09:58PM +0000, Colton Lewis wrote:
> Andrew Jones <andrew.jones@linux.dev> writes:
> > On Tue, Dec 20, 2022 at 04:32:00PM +0000, Colton Lewis wrote:
> > > Alexandru Elisei <alexandru.elisei@arm.com> writes:
> > Ah, I think I understand now. Were you running 32-bit arm tests? If so,
> > it'd be good to point that out explicitly in the commit message (the
> > 'arm:' prefix in the summary is ambiguous).
> 
> No, this was happening on arm64. Since it had been a while since I noted
> this issue, I reviewed it and realized the issue was only happening
> using -accel tcg. That was automatically being used on my problem test
> machine without me noticing. That's where the limit of 8 seems to be
> coming from and why the loop is triggered.
> 
> qemu-system-aarch64: Number of SMP CPUs requested (152) exceeds max CPUs
> supported by machine 'mach-virt' (8)
> 
> Since this case doesn't directly involve KVM, I doubt anyone cares about
> a fix.
> 
> > Assuming the loop body was running because it needed to reduce MAX_SMP to
> > 8 or lower for 32-bit arm tests, then we should be replacing the loop with
> > something that caps MAX_SMP at 8 for 32-bit arm tests instead.
> 
> We could cap at 8 for ACCEL=tcg. Even if no one cares, I'm tempted to do
> it so no one hits the same little landmine as me in the future.

TCG supports up to 255 CPUs. The only reason it'd have a max of 8 is if
you were configuring a GICv2 instead of a GICv3. Using gic-version=3 or
gic-version=max should allow the 152 CPUs to work. Actually, I should
have asked about your gic version instead of whether or not the VM was
AArch32 in the first place. I was incorrectly associating the gicv2
limits with arm32 since my memories of these things have started to
blur together...

Thanks,
drew