On Wed, Mar 07, 2018 at 09:29:10PM +0100, Nikola Ciprich wrote:

Hi,

...

...

...

I'd like to report that when upgrading our cluster from 4.14.18 to 4.14.24-rc1 (with live guests migration), almost none of guests survived..

What's your hardware setup, intel with IBPB enabled microcode?

Intel(R) Xeon(R) CPU E5-2650 0 @ 2.00GHz

therefore I suppose no IBPB (at least meltdown checker reports so)

...

Does guests hang right after live migration?

yes, just tried it.

...

Are you able to reproduce the problem, does it work with latest upstream?

yup, so I'm able to reproduce quickly. I'll revert the cluster to 4.14.18 now, but setup test system just afterwards, so and test the patch you've proposed.

...

Not sure it helps, but following patch is missing in 4.14.24

commit 37b95951c58fdf08dc10afa9d02066ed9f176fb5 upstream.

kvm_valid_sregs() should use X86_CR0_PG and X86_CR4_PAE to check bit status rather than X86_CR0_PG_BIT and X86_CR4_PAE_BIT. This patch is to fix it.

Fixes: f29810335965a(KVM/x86: Check input paging mode when cs.l is set) Reported-by: Jeremi Piotrowski jeremi.piotrowski@gmail.com Cc: Paolo Bonzini pbonzini@redhat.com Cc: Radim Krčmář rkrcmar@redhat.com Signed-off-by: Tianyu Lan Tianyu.Lan@microsoft.com Signed-off-by: Radim Krčmář rkrcmar@redhat.com

I'll test and report.

so indeed, this one on top of 4.14.24-rc1 fixes the migration for me. Greg, could you queue this one up please?

As was already pointed out, this is already queued up to be in the next release.

thanks,

greg k-h