On Tue, 2021-03-30 at 12:59 -0400, Paolo Bonzini wrote:
> @@ -2686,13 +2688,13 @@ static int kvm_guest_time_update(struct
> kvm_vcpu *v)
>          * If the host uses TSC clock, then passthrough TSC as stable
>          * to the guest.
>          */
> -       spin_lock(&ka->pvclock_gtod_sync_lock);
> +       spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
>         use_master_clock = ka->use_master_clock;
>         if (use_master_clock) {
>                 host_tsc = ka->master_cycle_now;
>                 kernel_ns = ka->master_kernel_ns;
>         }
> -       spin_unlock(&ka->pvclock_gtod_sync_lock);
> +       spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
> 
>         /* Keep irq disabled to prevent changes to the clock */
>         local_irq_save(flags);

That seems a little gratuitous at the end; restoring the flags as part
of the spin_unlock_irqrestore() and then immediately calling
local_irq_save().

Is something going to complain if we just use spin_unlock() there and
then later restore the flags with the existing local_irq_restore()?

Or should we move the local_irq_save() up above the existing
spin_lock() and leave the spin lock/unlock as they are?