On Fri, Jun 24, 2016 at 05:41:55PM +0100, Peter Maydell wrote: > On 24 June 2016 at 17:15, Andrew Jones <drjo...@redhat.com> wrote: > > On Fri, Jun 24, 2016 at 06:03:21PM +0200, Andrew Jones wrote: > >> So we can either > >> a) play it safe and always use clusters of 4 for ARM guests, and > >> KVM will get "fixed" when we start managing the guest's MPIDR > >> from userspace, or > >> b) use 8 here, as TCG always has, and KVM does for AArch32 guests. > >> This might be less safe, but also improves SGI efficiency. > > > > Actually AArch32 guests would even use all 16 tlist bits on gicv3, if > > there was a KVM host available to try it. So the (b) option shouldn't > > be "use 8" it should be "don't treat 32-bit guests differently" > > KVM AArch32 is 4 CPUs per cluster: > http://lxr.free-electrons.com/source/arch/arm/kvm/coproc.c#L109
Hmm... yes, it should use coproc.c, but here's what I get when I test qemu-system-aarch64 \ -machine virt,gic-version=2,accel=kvm \ -cpu host,aarch64=off \ -device virtio-serial-device \ -device virtconsole,chardev=ctd \ -chardev testdev,id=ctd \ -display none -serial stdio \ -kernel arm/selftest.flat \ -append smp -smp 8 PSCI version 0.2 PASS: selftest: smp: PSCI version PASS: selftest: smp: CPU( 1) mpidr=80000001 PASS: selftest: smp: CPU( 2) mpidr=80000002 PASS: selftest: smp: CPU( 3) mpidr=80000003 PASS: selftest: smp: CPU( 4) mpidr=80000004 PASS: selftest: smp: CPU( 5) mpidr=80000005 PASS: selftest: smp: CPU( 6) mpidr=80000006 PASS: selftest: smp: CPU( 7) mpidr=80000007 PASS: selftest: smp: CPU( 0) mpidr=80000000 SUMMARY: 9 tests (arm/selftest.flat built from https://github.com/rhdrjones/kvm-unit-tests/tree/arm/gic) Other configurations give the expected mpidrs. I'll look into it more next week. Thanks, drew
