On Tue, Aug 14, 2012 at 10:49 PM, Rusty Russell <rusty@xxxxxxxxxxxxxxx> wrote: > On Mon, 13 Aug 2012 23:15:07 -0400, Christoffer Dall <c.dall@xxxxxxxxxxxxxxxxxxxxxx> wrote: >> On Sun, Aug 5, 2012 at 8:41 PM, Rusty Russell <rusty@xxxxxxxxxxxxxxx> wrote: >> > + /* See ARM ARM B1.14.1: "Hyp traps on instructions >> > + * that fail their condition code check" */ >> >> not sure if checkpatch.pl complains, but this is not kernel style >> commenting, strictly speaking, is it? > > Yes, I couldn't get it under a line, so it needs wings. I think it's > stupid, but it is the Kernel Way. > >> > + if (!kvm_condition_valid(vcpu)) >> > + return 0; >> >> eh, this just exits to qemu without any further notice or explanation >> and the guest will re-execute the instruction and the hardware will >> trap again. Should we not fast forward past the instruction and adjust >> the ITSTATE accordingly? >> >> and then this would be return 1 now. >> >> if you confirm you agree, I will simply make the adjustment and merge. > > Yes, you're right. This is the problem with untestable code :( > unfortunately it is complicated by the fact that we need to know how long the instruction is to be able to skip it, which would mean that we would have to do the VA to PA on every HVC exception and copy the instruction from the guest check the length - unless we assume that HSR.ISV == 1 whenever this happens...? -Christoffer _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm
