On Thu, 2022-11-24 at 01:31 +0100, Paolo Bonzini wrote: > I have picked them into both kvm/master and kvm/queue. Thanks. > The gpc series probably will be left for 6.3. I had already removed > Sean's bits for the gpc and will rebase on top of your runstate > compatibility fixes, which I'm cherry-picking into kvm/queue. > > But wow, is that runstate compatibility patch ugly. Is it really > worth it having the two separate update paths, one which is ugly > because of BUILD_BUG_ON assertions and one which is ugly because of > the two-page stuff? The BUILD_BUG_ON() assertions could move. I quite liked having them there as documentation for compat handling, but I'm happy to move them. There's plenty of *actual* comments on the compat handling too ;) I do think it's worth the separate paths because the *common* case doesn't have to bounce it through the kernel stack at all, and it's a relatively fast path because it happens on each schedule in/out. The fast path isn't the part I hate. Removing *that* doesn't really make things much nicer. If you want to move the BUILD_BUG_ONs into the slow path and then we can concentrate our hate on that part and leave the fast path even nicer, that's fine though ;)
Attachment:
smime.p7s
Description: S/MIME cryptographic signature