kvmclock: add a new function to update env->tsc.WW-2015-52

(Back port from upstream, commit id: 0fd7e098db30e302d27920487f0afec)

The commit 317b0a6d8 fixed an issue which caused by the outdated
env->tsc value, but the fix lead to 'cpu_synchronize_all_states()'
called twice during live migration. The 'cpu_synchronize_all_states()'
takes about 130us for a VM which has 4 vcpus, it's a bit expensive.

Synchronize the whole CPU context just for updating env->tsc is too
wasting, this patch use a new function to update the env->tsc.
Comparing to 'cpu_synchronize_all_states()', it only takes about 20us.

Change-Id: I7dc8371aa17b005d073249c3732d9a7424f20a25
Signed-off-by: Liang Li <liang.z.li@intel.com>
Message-Id: <1446695464-27116-2-git-send-email-liang.z.li@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

3 files changed, 48 insertions, 16 deletions

diff --git a/qemu/hw/i386/kvm/clock.c b/qemu/hw/i386/kvm/clock.c
index efdf16584..0593a3f1f 100644
--- a/qemu/hw/i386/kvm/clock.c
+++ b/qemu/hw/i386/kvm/clock.c
@@ -17,7 +17,7 @@
 #include "qemu/host-utils.h"
 #include "sysemu/sysemu.h"
 #include "sysemu/kvm.h"
-#include "sysemu/cpus.h"
+#include "kvm_i386.h"
 #include "hw/sysbus.h"
 #include "hw/kvm/clock.h"
 
@@ -125,21 +125,7 @@ static void kvmclock_vm_state_change(void *opaque, int running,
             return;
         }
 
-        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-         * outdated APIC state may be migrated to another host.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-         * The real fix would be to make sure outdated APIC state is read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();
+        kvm_synchronize_all_tsc();
 
         ret = kvm_vm_ioctl(kvm_state, KVM_GET_CLOCK, &data);
         if (ret < 0) {
diff --git a/qemu/target-i386/kvm.c b/qemu/target-i386/kvm.c
index 066d03d99..721c580ed 100644
--- a/qemu/target-i386/kvm.c
+++ b/qemu/target-i386/kvm.c
@@ -96,6 +96,51 @@ bool kvm_allows_irq0_override(void)
     return !kvm_irqchip_in_kernel() || kvm_has_gsi_routing();
 }
 
+static int kvm_get_tsc(CPUState *cs)
+{
+    X86CPU *cpu = X86_CPU(cs);
+    CPUX86State *env = &cpu->env;
+    struct {
+        struct kvm_msrs info;
+        struct kvm_msr_entry entries[1];
+    } msr_data;
+    int ret;
+
+    if (env->tsc_valid) {
+        return 0;
+    }
+
+    msr_data.info.nmsrs = 1;
+    msr_data.entries[0].index = MSR_IA32_TSC;
+    env->tsc_valid = !runstate_is_running();
+
+    ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MSRS, &msr_data);
+    if (ret < 0) {
+        return ret;
+    }
+
+    env->tsc = msr_data.entries[0].data;
+    return 0;
+}
+
+static inline void do_kvm_synchronize_tsc(void *arg)
+{
+    CPUState *cpu = arg;
+
+    kvm_get_tsc(cpu);
+}
+
+void kvm_synchronize_all_tsc(void)
+{
+    CPUState *cpu;
+
+    if (kvm_enabled()) {
+        CPU_FOREACH(cpu) {
+            run_on_cpu(cpu, do_kvm_synchronize_tsc, cpu);
+        }
+    }
+}
+
 static struct kvm_cpuid2 *try_get_cpuid(KVMState *s, int max)
 {
     struct kvm_cpuid2 *cpuid;
diff --git a/qemu/target-i386/kvm_i386.h b/qemu/target-i386/kvm_i386.h
index e557e94f4..c1b312ba2 100644
--- a/qemu/target-i386/kvm_i386.h
+++ b/qemu/target-i386/kvm_i386.h
@@ -15,6 +15,7 @@
 
 bool kvm_allows_irq0_override(void);
 bool kvm_has_smm(void);
+void kvm_synchronize_all_tsc(void);
 void kvm_arch_reset_vcpu(X86CPU *cs);
 void kvm_arch_do_init_vcpu(X86CPU *cs);