On Wed, May 25, 2016 at 03:26:32PM +0100, Marc Zyngier wrote: > The dust has not yet settled on the new vgic, but here's the first > batch of fixes anyway! > > The first two patches are courtesy of Christoffer, who noticed that we > fail to clear LRs that generate a maintenance interrupt, leading to > potential screaming interrupts. This bug already exists in the current > mainline (cc stable?) > > The next two address a bug where we fail to properly resample the line > level on exit, which could result in spurious interrupts being > injected. This is specific to the new vgic implementation. > > The following two patches tighten our GICv3 emulation by preventing > the guest from changing the SRE setting. This bug already exists in > mainline, though it is hardly critical. > > The last patch is actually a performance optimization: if the guest is > using GICv3, we can drop a number of barriers (since we don't need to > change SRE, and there is no memory-mapped view to synchronize > with). This results in a world switch that is 2.5% faster on my LS2085 > box when running GICv3 guests. I suspect that the bigger the box, the > bigger this impact this change will have (system-wide DSBs don't really > come cheap). > > These patches have been tested on top of the kvmarm/next branch. > Thanks, applied to queue (with cc stable for v4.6+ as I only think this applies post our optimizations that went in for v4.6) -Christoffer -- 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
