On Fri, Oct 13, 2017 at 10:15:09AM +0100, Alex Bennée wrote:
>
> Christoffer Dall <cdall@xxxxxxxxxx> writes:
>
> > On Fri, Oct 06, 2017 at 12:39:20PM +0100, Alex Bennée wrote:
> >> If we are using guest debug to single-step the guest we need to ensure
> >> we exit after emulating the instruction. This only affects
> >> instructions completely emulated by the kernel. For userspace emulated
> >> instructions we need to exit and return to complete the emulation.
> >>
> >> We fake debug.arch.hsr to contain ESR_ELx_EC_SOFTSTP_LOW so QEMU knows
> >> it was a single-step event (and without altering the userspace ABI).
> >>
> >> Signed-off-by: Alex Bennée <alex.bennee@xxxxxxxxxx>
> >> ---
> >> arch/arm64/kvm/handle_exit.c | 48 +++++++++++++++++++++++++++++++-------------
> >> 1 file changed, 34 insertions(+), 14 deletions(-)
> >>
> >> diff --git a/arch/arm64/kvm/handle_exit.c b/arch/arm64/kvm/handle_exit.c
> >> index 7debb74843a0..c918d291cb58 100644
> >> --- a/arch/arm64/kvm/handle_exit.c
> >> +++ b/arch/arm64/kvm/handle_exit.c
> >> @@ -178,6 +178,39 @@ static exit_handle_fn kvm_get_exit_handler(struct kvm_vcpu *vcpu)
> >> return arm_exit_handlers[hsr_ec];
> >> }
> >>
> >> +/*
> >> + * When handling traps we need to ensure exit the guest if we
> >> + * completely emulated the instruction while single-stepping. Stuff to
> >> + * be emulated in userspace needs to complete that first.
> >> + */
> >
> > I really don't understand the first sentence here. We are already out
> > of the guest, so do you mean a return to userspace?
> > I think the second sentence could be more clear as well. Is 'stuff' not
> > actually 'MMIO emulation' or 'emulation' more broadly?
>
> Your right - it's sloppily worded how about:
>
> /*
> * We may be single-stepping an emulated instruction. If the emulation
> * has been completed in-kernel we can return to userspace with a
> * KVM_EXIT_DEBUG, otherwise the userspace needs to complete it's
s/it's/its/
> * emulation first.
> */
Otherwise looks much better, thanks.
>
> For x86 there is also IO emulation but in principle anything that might
> be passed off to userspace to be completed should be done first.
>
> >
> >> +
> >> +static int handle_trap_exceptions(struct kvm_vcpu *vcpu, struct kvm_run *run)
> >> +{
> >> + int handled;
> >> +
> >> + /*
> >> + * See ARM ARM B1.14.1: "Hyp traps on instructions
> >> + * that fail their condition code check"
> >> + */
> >> + if (!kvm_condition_valid(vcpu)) {
> >> + kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
> >> + handled = 1;
> >> + } else {
> >> + exit_handle_fn exit_handler;
> >> +
> >> + exit_handler = kvm_get_exit_handler(vcpu);
> >> + handled = exit_handler(vcpu, run);
> >> + }
> >> +
> >> + if (handled && (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)) {
> >
> > Don't you want if (handled == 1) or if (handled > 0) ?
> >
> > If there was an error I think we want to just return that to userspace
> > and not override it and present single-stepping.
>
> Yes, I'll fix it.
>
Thanks,
-Christoffer
