On Tue, Oct 16, 2012 at 11:04 AM, Alexander Graf <agraf@xxxxxxx> wrote: > On 10/16/2012 05:00 PM, Marc Zyngier wrote: >> >> On 16/10/12 14:43, Christoffer Dall wrote: >>> >>> On Mon, Oct 15, 2012 at 11:19 AM, Marc Zyngier<marc.zyngier@xxxxxxx> >>> wrote: >>>> >>>> Now that our world-switch is mostly layout independant, it becomes >>>> relatively easy to move the usr_regs array to struct pt_regs. >>>> >>>> This gives us a common abstraction with the rest of the kernel, and >>>> makes it similar to what is being done on the arm64 side. >>>> >>> Do you have any code that does this which you're referring to that the >>> rest of us can see as to evaluate this argument? >> >> The arm64 bit is not public yet. It will as soon as I have the green >> light for it. >> >> When it comes to the use of pt_regs, grep-ing around in arch/arm will >> give you a pretty compelling view of its usage, including kgdb which has >> requirements that are similar to KVM's. >> >> It is a matter of uniformity in the arch/arm tree, the same way we'd >> like to see as few as possible implementations of the opcode stuff. We >> had way too many cases of the NIH syndrome in the past, and it would be >> nice not to repeat the mistake. > > > Yeah, we had a patch doing pt_regs on PPC as well a while back. The only > reason we did not go for it is because we need to heavily PV'nize some > registers (shared struct between guest and host) to be able to achieve good > speeds with trap-and-emulate. > > Since you guys don't have that problem, I don't see any reason not to use > pt_regs. > ok, fair enough. Maybe we can put something funny in orig_r0's place some day. _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm
