hypervisor: kvm: Ignore -EINVAL from KVM_KVMCLOCK_CTRL ioctl()

If the guest hasn't initialised a PV clock then the KVM_KVMCLOCK_CTRL
ioctl will return -EINVAL. Therefore if running in the firmware or an OS
that doesn't use the PV clock then we should ignore that error

Tested by migrating a VM that has not yet booted into the Linux kernel
(just in firmware) by specifying no disk image:

e.g. target/debug/cloud-hypervisor --kernel ~/workloads/hypervisor-fw --api-socket /tmp/api --serial tty --console off

Fixes: #3586

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
This commit is contained in:
Rob Bradford 2022-01-18 17:08:42 +00:00 committed by Sebastien Boeuf
parent b7b3b45364
commit 658658e76c

View File

@ -1051,9 +1051,16 @@ impl cpu::Vcpu for KvmVcpu {
/// potential soft lockups when being resumed.
///
fn notify_guest_clock_paused(&self) -> cpu::Result<()> {
self.fd
.kvmclock_ctrl()
.map_err(|e| cpu::HypervisorCpuError::NotifyGuestClockPaused(e.into()))
if let Err(e) = self.fd.kvmclock_ctrl() {
// Linux kernel returns -EINVAL if the PV clock isn't yet initialised
// which could be because we're still in firmware or the guest doesn't
// use KVM clock.
if e.errno() != libc::EINVAL {
return Err(cpu::HypervisorCpuError::NotifyGuestClockPaused(e.into()));
}
}
Ok(())
}
#[cfg(any(target_arch = "arm", target_arch = "aarch64"))]
fn vcpu_init(&self, kvi: &VcpuInit) -> cpu::Result<()> {