On Thu, Jul 26, 2012 at 11:21 PM, Peter Maydell <peter.maydell@xxxxxxxxxx> wrote: > On 26 July 2012 22:15, Christoffer Dall <c.dall@xxxxxxxxxxxxxxxxxxxxxx> wrote: >> On Tue, Jul 3, 2012 at 3:51 PM, Peter Maydell <peter.maydell@xxxxxxxxxx> wrote: >>> We've gone back and forth on this one, I think. For an A15 guest >>> we want to say that userspace GIC isn't permitted (ie you must use >>> the in-kernel-irqchip). For an A9 guest you would have to use the >>> userspace GIC, but on the other hand we don't currently actually >>> support an A9 guest... It seems a little unfortunate to remove >>> code that is actually working but I could certainly see the >>> cleanliness argument for dropping it and letting anybody who >>> wanted an A9 guest put it back later. >>> >> I prefer keeping the user space code as the virtualization extensions >> do not mandate VGIC support, so one could imagine hypothetical future >> cores without VGIC support. > > I think you can file that one in the "theoretically possible > but vanishingly unlikely" bucket. Having said that... > vanishingly is a convincing word. >> Also, we don't really have any profiling >> information about the VGIC implementation yet, so the user space >> support version may be useful for some performance poking, maybe. > > ...if you want to retain the kernel and ABI end of it, the qemu > end is pretty clean (or will be once I have managed to get the > required cleanup of kvm_irqchip_in_kernel() through code review) > so I'm happy to keep it at that end. > I'm not married to keeping it around, but I would like trying out both versions on some stable platform and be sure the vgic code is stable before nuking it. If we want to drop it before the final merge it shouldn't be a big deal. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
