On Tue, Feb 20, 2024 at 05:53:55PM +0000, Marc Zyngier wrote: > On Tue, 20 Feb 2024 17:43:03 +0000, Oliver Upton <oliver.upton@xxxxxxxxx> wrote: > > I think we can get into here both from contexts w/ interrupts disabled > > or enabled. irqfd_wakeup() expects to be called w/ interrupts disabled. > > > > All the more reason to use irqsave() / irqrestore() flavors of all of > > this, and a reminder to go check all callsites that implicitly take the > > xa_lock. > > Sounds good. Maybe you can also update the locking order > "documentation" to include the xa_lock? I expect that it will > ultimately replace lpi_list_lock. Yep, I got to the point of deleting the lpi_list_lock on the full series, which is where I update the documentation. I really didn't want people to know I'm adding yet another layer of locking in the interim... Anyways, I think there's sufficient feedback to justify a respin. I'll make sure the documentation is updated w/ the xa_lock for the stuff I'm trying to land in 6.9. -- Thanks, Oliver
