| Message ID | 1241579779-17974-1-git-send-email-glommer@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Tue, May 05, 2009 at 11:16:19PM -0400, Glauber Costa wrote: > When we create a new vcpu, we need to make sure that > all of the state it is going to use (apic state, for example) > already exists. We can do it nicely by making sure kvm_init_vcpu > is executed after everything else in cpu creation. > > After that, the first call to KVM_SET_LAPIC ioctl will not find an > existant vcpu. So we introduce a function that tell us that the vcpu > is already initialized, and is it safe to call the ioctl. Otherwise, > just don't botter. Why did you decide to drop the additional wait vcpu->inited thing you had in the previous patch? I think its nice to make the synchronization explicit. Isnt your current solution somewhat trickier? And if you disagree with me (which you should avoid for safety reasons), you need to regenerate the patch since it'll reject against qemu-kvm.git as of today. > We then force the execution of the KVM_SET_LAPIC from within the new > vcpu thread. > > Signed-off-by: Glauber Costa <glommer@redhat.com> > --- > hw/apic.c | 21 +++++++++++---------- > hw/pc.c | 2 ++ > qemu-kvm.c | 10 ++++++++++ > qemu-kvm.h | 4 ++++ > target-i386/helper.c | 2 -- > 5 files changed, 27 insertions(+), 12 deletions(-) > -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed, May 06, 2009 at 10:53:06AM -0300, Marcelo Tosatti wrote: > On Tue, May 05, 2009 at 11:16:19PM -0400, Glauber Costa wrote: > > When we create a new vcpu, we need to make sure that > > all of the state it is going to use (apic state, for example) > > already exists. We can do it nicely by making sure kvm_init_vcpu > > is executed after everything else in cpu creation. > > > > After that, the first call to KVM_SET_LAPIC ioctl will not find an > > existant vcpu. So we introduce a function that tell us that the vcpu > > is already initialized, and is it safe to call the ioctl. Otherwise, > > just don't botter. > > Why did you decide to drop the additional wait vcpu->inited thing you > had in the previous patch? I think its nice to make the synchronization > explicit. Explicit is good. Not needed is even better. If we make sure to call kvm_vcpu_init() after everything is already initialized, we avoid the need to have any kind of sync at all, since it serializes naturally. > > Isnt your current solution somewhat trickier? I believe it is simpler. > > And if you disagree with me (which you should avoid for safety reasons), > you need to regenerate the patch since it'll reject against qemu-kvm.git > as of today. I can do that happily (both disagreeing with you, and regenerating the patch) -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/hw/apic.c b/hw/apic.c index 8c059f6..b8112f1 100644 --- a/hw/apic.c +++ b/hw/apic.c @@ -891,6 +891,15 @@ static void kvm_kernel_lapic_load_from_user(APICState *s) #endif +void qemu_kvm_load_lapic(CPUState *env) +{ +#ifdef KVM_CAP_IRQCHIP + if (kvm_enabled() && kvm_vcpu_inited(env) && qemu_kvm_irqchip_in_kernel()) { + kvm_kernel_lapic_load_from_user(env->apic_state); + } +#endif +} + static void apic_save(QEMUFile *f, void *opaque) { APICState *s = opaque; @@ -965,11 +974,7 @@ static int apic_load(QEMUFile *f, void *opaque, int version_id) if (version_id >= 2) qemu_get_timer(f, s->timer); -#ifdef KVM_CAP_IRQCHIP - if (kvm_enabled() && qemu_kvm_irqchip_in_kernel()) { - kvm_kernel_lapic_load_from_user(s); - } -#endif + qemu_kvm_load_lapic(s->cpu_env); return 0; } @@ -991,11 +996,7 @@ static void apic_reset(void *opaque) */ s->lvt[APIC_LVT_LINT0] = 0x700; } -#ifdef KVM_CAP_IRQCHIP - if (kvm_enabled() && qemu_kvm_irqchip_in_kernel()) { - kvm_kernel_lapic_load_from_user(s); - } -#endif + qemu_kvm_load_lapic(s->cpu_env); } static CPUReadMemoryFunc *apic_mem_read[3] = { diff --git a/hw/pc.c b/hw/pc.c index db34f53..afa9eac 100644 --- a/hw/pc.c +++ b/hw/pc.c @@ -853,6 +853,8 @@ CPUState *pc_new_cpu(int cpu, const char *cpu_model, int pci_enabled) if (pci_enabled) { apic_init(env); } + if (kvm_enabled()) + kvm_init_vcpu(env); return env; } diff --git a/qemu-kvm.c b/qemu-kvm.c index 68a9218..2e94693 100644 --- a/qemu-kvm.c +++ b/qemu-kvm.c @@ -427,6 +427,9 @@ static void *ap_main_loop(void *_env) kvm_create_vcpu(kvm_context, env->cpu_index); kvm_qemu_init_env(env); + /* APIC state creation takes place before we get here. So despite the fact that + * apic_reset() (called by apic_init) will also load the apic state, we have to redo it here + */ #ifdef USE_KVM_DEVICE_ASSIGNMENT /* do ioperm for io ports of assigned devices */ LIST_FOREACH(data, &ioperm_head, entries) @@ -438,6 +441,8 @@ static void *ap_main_loop(void *_env) current_env->kvm_cpu_state.created = 1; pthread_cond_signal(&qemu_vcpu_cond); + qemu_kvm_load_lapic(env); + /* and wait for machine initialization */ while (!qemu_system_ready) qemu_cond_wait(&qemu_system_cond); @@ -455,6 +460,11 @@ void kvm_init_vcpu(CPUState *env) qemu_cond_wait(&qemu_vcpu_cond); } +int kvm_vcpu_inited(CPUState *env) +{ + return env->kvm_cpu_state.created; +} + int kvm_init_ap(void) { #ifdef TARGET_I386 diff --git a/qemu-kvm.h b/qemu-kvm.h index ca59af8..a307976 100644 --- a/qemu-kvm.h +++ b/qemu-kvm.h @@ -16,6 +16,7 @@ int kvm_main_loop(void); int kvm_qemu_init(void); int kvm_qemu_create_context(void); int kvm_init_ap(void); +int kvm_vcpu_inited(CPUState *env); void kvm_qemu_destroy(void); void kvm_load_registers(CPUState *env); void kvm_save_registers(CPUState *env); @@ -31,6 +32,9 @@ int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap); int kvm_qemu_init_env(CPUState *env); int kvm_qemu_check_extension(int ext); void kvm_apic_init(CPUState *env); +/* called from vcpu initialization */ +void qemu_kvm_load_lapic(CPUState *env); + int kvm_set_irq(int irq, int level, int *status); int kvm_physical_memory_set_dirty_tracking(int enable); diff --git a/target-i386/helper.c b/target-i386/helper.c index 7440983..a273cac 100644 --- a/target-i386/helper.c +++ b/target-i386/helper.c @@ -1695,7 +1695,5 @@ CPUX86State *cpu_x86_init(const char *cpu_model) #ifdef CONFIG_KQEMU kqemu_init(env); #endif - if (kvm_enabled()) - kvm_init_vcpu(env); return env; }
When we create a new vcpu, we need to make sure that all of the state it is going to use (apic state, for example) already exists. We can do it nicely by making sure kvm_init_vcpu is executed after everything else in cpu creation. After that, the first call to KVM_SET_LAPIC ioctl will not find an existant vcpu. So we introduce a function that tell us that the vcpu is already initialized, and is it safe to call the ioctl. Otherwise, just don't botter. We then force the execution of the KVM_SET_LAPIC from within the new vcpu thread. Signed-off-by: Glauber Costa <glommer@redhat.com> --- hw/apic.c | 21 +++++++++++---------- hw/pc.c | 2 ++ qemu-kvm.c | 10 ++++++++++ qemu-kvm.h | 4 ++++ target-i386/helper.c | 2 -- 5 files changed, 27 insertions(+), 12 deletions(-)