KVM: VMX: Ensure vcpu time stamp counter is monotonous
authorAvi Kivity <avi@qumranet.com>
Wed, 13 Jun 2007 16:55:28 +0000 (19:55 +0300)
committerAvi Kivity <avi@qumranet.com>
Mon, 16 Jul 2007 09:05:47 +0000 (12:05 +0300)
commit7700270ee3c1324c18f5b7c36ee5ba1a4165919a
treeb5b1f47ca6ab565bcfb7974ea614da6ddebce0a4
parent94cea1bb9d050c3200b36420cc03ba744dfd4338
KVM: VMX: Ensure vcpu time stamp counter is monotonous

If the time stamp counter goes backwards, a guest delay loop can become
infinite.  This can happen if a vcpu is migrated to another cpu, where
the counter has a lower value than the first cpu.

Since we're doing an IPI to the first cpu anyway, we can use that to pick
up the old tsc, and use that to calculate the adjustment we need to make
to the tsc offset.

Signed-off-by: Avi Kivity <avi@qumranet.com>
drivers/kvm/vmx.c