On 2011-10-14 11:03, Lai Jiangshan wrote: > Currently, NMI interrupt is blindly sent to all the vCPUs when NMI > button event happens. This doesn't properly emulate real hardware on > which NMI button event triggers LINT1. Because of this, NMI is sent to > the processor even when LINT1 is masked in LVT. For example, this > causes the problem that kdump initiated by NMI sometimes doesn't work > on KVM, because kdump assumes NMI is masked on CPUs other than CPU0. > > With this patch, inject-nmi request is handled as follows. > > - When in-kernel irqchip is enabled and KVM_SET_LINT1 is enabled, > inject LINT1 instead of NMI interrupt. > > - otherwise when in-kernel irqchip is enabled, get the in-kernel > LAPIC states and test the APIC_LVT_MASKED, if LINT1 is unmasked, > and then delivering the NMI directly. > > - otherwise, userland lapic emulates NMI button and inject NMI > if it is unmasked. > > Signed-off-by: Lai Jiangshan <[email protected]> > Reported-by: Kenji Kaneshige <[email protected]> > --- > hw/apic.c | 72 > +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ > hw/apic.h | 1 + > monitor.c | 6 ++++- > 3 files changed, 78 insertions(+), 1 deletions(-) > > diff --git a/hw/apic.c b/hw/apic.c > index 69d6ac5..91b82d0 100644 > --- a/hw/apic.c > +++ b/hw/apic.c > @@ -205,6 +205,78 @@ void apic_deliver_pic_intr(DeviceState *d, int level) > } > } > > +#ifdef KVM_CAP_IRQCHIP
Please read all my comments. That unfortunately also applies to the rest
of the patch.
> +static inline uint32_t kapic_reg(struct kvm_lapic_state *kapic, int reg_id);
> +
> +static void kvm_irqchip_deliver_nmi(void *p)
> +{
> + APICState *s = p;
> + struct kvm_lapic_state klapic;
> + uint32_t lvt;
> +
> + kvm_get_lapic(s->cpu_env, &klapic);
> + lvt = kapic_reg(&klapic, 0x32 + APIC_LVT_LINT1);
> +
> + if (lvt & APIC_LVT_MASKED) {
> + return;
> + }
> +
> + if (((lvt >> 8) & 7) != APIC_DM_NMI) {
> + return;
> + }
> +
> + kvm_vcpu_ioctl(s->cpu_env, KVM_NMI);
> +}
> +
> +static void __apic_deliver_nmi(APICState *s)
> +{
> + if (kvm_irqchip_in_kernel()) {
> + run_on_cpu(s->cpu_env, kvm_irqchip_deliver_nmi, s);
> + } else {
> + apic_local_deliver(s, APIC_LVT_LINT1);
> + }
> +}
> +#else
> +static void __apic_deliver_nmi(APICState *s)
> +{
> + apic_local_deliver(s, APIC_LVT_LINT1);
> +}
> +#endif
> +
> +enum {
> + KVM_SET_LINT1_UNKNOWN,
> + KVM_SET_LINT1_ENABLED,
> + KVM_SET_LINT1_DISABLED,
> +};
> +
> +static void kvm_set_lint1(void *p)
> +{
> + CPUState *env = p;
> +
> + kvm_vcpu_ioctl(env, KVM_SET_LINT1);
> +}
> +
> +void apic_deliver_nmi(DeviceState *d)
> +{
> + APICState *s = DO_UPCAST(APICState, busdev.qdev, d);
> + static int kernel_lint1 = KVM_SET_LINT1_UNKNOWN;
> +
> + if (kernel_lint1 == KVM_SET_LINT1_UNKNOWN) {
> + if (kvm_enabled() && kvm_irqchip_in_kernel() &&
> + kvm_check_extension(kvm_state, KVM_CAP_SET_LINT1)) {
That CAP test belongs where the injection shall happen. Here you decide
about user space vs. kernel space APIC model.
Let's try it together:
if kvm_enabled && kvm_irqchip_in_kernel
run_on_cpu(kvm_apic_deliver_nmi)
else
apic_local_deliver(APIC_LVT_LINT1)
with kvm_acpi_deliver_nmi like this:
if !check_extention(CAP_SET_LINT1)
get_kernel_apic_state
if !nmi_acceptable
return
kvm_vcpu_ioctl(KVM_NMI)
Please don't trust me blindly and re-check, but this is how the scenario
looks like to me.
Thanks for your patience,
Jan
signature.asc
Description: OpenPGP digital signature
