On Fri, 21 Jan 2022 13:08:41 -0300
Daniel Henrique Barboza <[email protected]> wrote:
> cpu_interrupt_exittb() was introduced by commit 044897ef4a22
> ("target/ppc: Fix system lockups caused by interrupt_request state
> corruption") as a way to wrap cpu_interrupt() helper in BQL.
>
> After that, commit 6d38666a8931 ("ppc: Ignore the CPU_INTERRUPT_EXITTB
> interrupt with KVM") added a condition to skip this interrupt if we're
> running with KVM.
>
> Problem is that the change made by the above commit, testing for
> !kvm_enabled() at the start of cpu_interrupt_exittb():
>
> static inline void cpu_interrupt_exittb(CPUState *cs)
> {
> if (!kvm_enabled()) {
> return;
> }
> (... do cpu_interrupt(cs, CPU_INTERRUPT_EXITTB) ...)
>
> is doing the opposite of what it intended to do. This will return
> immediately if not kvm_enabled(), i.e. it's a emulated CPU, and if
> kvm_enabled() it will proceed to fire CPU_INTERRUPT_EXITTB.
>
Ohh boy... this is embarrassing :-\
> Fix the 'skip KVM' condition so the function is a no-op when
> kvm_enabled().
>
> CC: Greg Kurz <[email protected]>
> Resolves: https://gitlab.com/qemu-project/qemu/-/issues/809
> Fixes: 6d38666a8931 ("ppc: Ignore the CPU_INTERRUPT_EXITTB interrupt with
> KVM")
> Signed-off-by: Daniel Henrique Barboza <[email protected]>
> ---
Reviewed-by: Greg Kurz <[email protected]>
> target/ppc/helper_regs.c | 6 +++++-
> 1 file changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/target/ppc/helper_regs.c b/target/ppc/helper_regs.c
> index 8671b7bb69..7dca585ddd 100644
> --- a/target/ppc/helper_regs.c
> +++ b/target/ppc/helper_regs.c
> @@ -201,7 +201,11 @@ void cpu_get_tb_cpu_state(CPUPPCState *env, target_ulong
> *pc,
>
> void cpu_interrupt_exittb(CPUState *cs)
> {
> - if (!kvm_enabled()) {
> + /*
> + * We don't need to worry about translation blocks
> + * when running with KVM.
> + */
> + if (kvm_enabled()) {
> return;
> }
>
