Skip to contentSkip to conversation

kvmclock: Add comment explaining why we need cpu_clean_all_dirty()

PATCH In review Eduardo Habkost 2 participants 2 messages Nov 3, 2014 – Nov 5, 2014

lore
Share & tools
lore

Eduardo Habkost

author
#0 lore raw

PATCH 1/1 kvmclock: Add comment explaining why we need cpu_clean_all_dirty()

Try to explain why commit 317b0a6d8ba44e9bf8f9c3dbd776c4536843d82c needed a cpu_clean_all_dirty() call just after calling cpu_synchronize_all_states().

Signed-off-by Eduardo Habkost <[email protected]>
Cc Andrey Korolyov <[email protected]>
Cc Marcin Gibuła <[email protected]>
Cc Marcelo Tosatti <[email protected]>
Cc Paolo Bonzini <[email protected]>
---
 hw/i386/kvm/clock.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)
hw/i386/kvm/clock.c +14 -0
index 1ac60d6..58be2bd 100644
@@ -127,7 +127,21 @@ static void kvmclock_vm_state_change(void *opaque, int running,
         }
 
         cpu_synchronize_all_states();
+        /* In theory, the cpu_synchronize_all_states() call above wouldn't
+         * affect the rest of the code, as the VCPU state inside CPUState
+         * is supposed to always match the VCPU state on the kernel side.
+         *
+         * In practice, calling cpu_synchronize_state() too soon will load the
+         * kernel-side APIC state into X86CPU.apic_state too early, APIC state
+         * won't be reloaded later because CPUState.vcpu_dirty==true, and
+         * outdated APIC state may be migrated to another host.
+         *
+         * The real fix would be to make sure outdated APIC state is read
+         * from the kernel again when necessary. While this is not fixed, we
+         * need the cpu_clean_all_dirty() call below.
+         */
         cpu_clean_all_dirty();
+
         ret = kvm_vm_ioctl(kvm_state, KVM_GET_CLOCK, &data);
         if (ret < 0) {
             fprintf(stderr, "KVM_GET_CLOCK failed: %s\n", strerror(ret));
Signature · 1 line
1.9.3
#1 lore raw
Quoted 30 lines · Eduardo Habkost

Try to explain why commit 317b0a6d8ba44e9bf8f9c3dbd776c4536843d82c needed a cpu_clean_all_dirty() call just after calling cpu_synchronize_all_states().

Signed-off-by Eduardo Habkost <[email protected]>
Cc Andrey Korolyov <[email protected]>
Cc Marcin Gibuła <[email protected]>
Cc Marcelo Tosatti <[email protected]>
Cc Paolo Bonzini <[email protected]>
---
 hw/i386/kvm/clock.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)
hw/i386/kvm/clock.c +8 -0
index 1ac60d6..58be2bd 100644
@@ -127,7 +127,21 @@ static void kvmclock_vm_state_change(void *opaque, int running,
         }
 
         cpu_synchronize_all_states();
+        /* In theory, the cpu_synchronize_all_states() call above wouldn't
+         * affect the rest of the code, as the VCPU state inside CPUState
+         * is supposed to always match the VCPU state on the kernel side.
+         *
+         * In practice, calling cpu_synchronize_state() too soon will load the
+         * kernel-side APIC state into X86CPU.apic_state too early, APIC state
+         * won't be reloaded later because CPUState.vcpu_dirty==true, and
+         * outdated APIC state may be migrated to another host.

hw/i386/kvm/clock.c static void kvmclock_vm_state_change(void *opaque, int running, · line 147

Quoted 10 lines
hw/i386/kvm/clock.c +6 -0
+         *
+         * The real fix would be to make sure outdated APIC state is read
+         * from the kernel again when necessary. While this is not fixed, we
+         * need the cpu_clean_all_dirty() call below.
+         */
         cpu_clean_all_dirty();
+
         ret = kvm_vm_ioctl(kvm_state, KVM_GET_CLOCK, &data);
         if (ret < 0) {
             fprintf(stderr, "KVM_GET_CLOCK failed: %s\n", strerror(ret));

Thanks, applying to uq/master.

Paolo