On Wed, Nov 23, 2022, David Woodhouse wrote: > On Wed, 2022-11-23 at 19:32 +0000, Sean Christopherson wrote: > > Right. Might be worth adding a comment at some point to call out that disabling > > IRQs may not be strictly required for all users, but it's done for simplicity. > > Ah, if/when we add kvm_gpc_lock(), that would be the perfect place to document > > the behavior. > > Yeah. Or perhaps the kvm_gpc_lock() should go with that 'not required > for all users, but done for simplicity' angle too, and always disable > IRQs? I was thinking the latter (always disable IRQs in kvm_gpc_lock()). Sorry I didn't make that clear. I completely agree that fewer conditionals in this code is better, I was mostly trying to figure out if there is some edge case I was missing.