KVM: x86: lower default for halt_poll_ns
authorPaolo Bonzini <pbonzini@redhat.com>
Tue, 18 Apr 2017 10:41:18 +0000 (12:41 +0200)
committerRadim Krčmář <rkrcmar@redhat.com>
Tue, 16 May 2017 19:15:50 +0000 (21:15 +0200)
commitb401ee0b85a53e89739ff68a5b1a0667d664afc9
tree68a3164a492eb30017c61a727a2c66812d9ed576
parent0780516a18f87e881e42ed815f189279b0a1743c
KVM: x86: lower default for halt_poll_ns

In some fio benchmarks, halt_poll_ns=400000 caused CPU utilization to
increase heavily even in cases where the performance improvement was
small.  In particular, bandwidth divided by CPU usage was as much as
60% lower.

To some extent this is the expected effect of the patch, and the
additional CPU utilization is only visible when running the
benchmarks.  However, halving the threshold also halves the extra
CPU utilization (from +30-130% to +20-70%) and has no negative
effect on performance.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
arch/x86/include/asm/kvm_host.h