Messages in this thread | | | Date | Fri, 20 Feb 2009 13:59:19 +0100 | From | Ingo Molnar <> | Subject | Re: [PATCH] x86: enable x2apic early at the first point |
| |
* Gleb Natapov <gleb@redhat.com> wrote:
> On Fri, Feb 20, 2009 at 12:06:51PM +0100, Ingo Molnar wrote: > > > > > If you'll revert my patch it will not be possible to use > > > > > x2apic in KVM (at least without KVM implementing interrupt > > > > > remapping which is unneeded otherwise) and x2apic interface is > > > > > much better for vitalization. Instead of reverting the patch > > > > > it will be better to add check if x2apic can be used without > > > > > intr-remmaping (all CPUs belong to cluster 0) or allow > > > > > enabling of x2apic without IR if running as a guest. > > > > > > > > yep, that would be required - because your patch can break real > > > > systems right now. Mind sending me a fix for it? > > > > > > > > I've applied Yinghai's fix as well, so please base it on latest > > > > tip:master. > > > > > > > OK. Will send next week. > > > > ok, that's too long of a breakage window - i'll revert it then, > > please send a new version once you have it. > > > > What approach you actually prefer? Enable x2apic without > intr-remapping only when running in KVM, or even if running on > real HW but all CPUs are in cluster 0? If former then below is > updated patch (tested only in KVM) if later then it's > definitely for next week :)
The latter would be the ideal solution - i.e. next week :)
x2apic is a rare feature, and the more exposure we give it the more tested it becomes. It's also obviously good for general code structure if core CPU features are separated from chipset/iommu features.
Plus, x2apic accesses could actually be faster than UC accesses to the lapic, so whenever we can we should enter x2apic mode - even if intr-remap is not turned on (because not needed).
Ingo
| |