KVM: x86 shared msr infrastructure

The various syscall-related MSRs are fairly expensive to switch.  Currently
we switch them on every vcpu preemption, which is far too often:

- if we're switching to a kernel thread (idle task, threaded interrupt,
  kernel-mode virtio server (vhost-net), for example) and back, then
  there's no need to switch those MSRs since kernel threasd won't
  be exiting to userspace.

- if we're switching to another guest running an identical OS, most likely
  those MSRs will have the same value, so there's little point in reloading
  them.

- if we're running the same OS on the guest and host, the MSRs will have
  identical values and reloading is unnecessary.

This patch uses the new user return notifiers to implement last-minute
switching, and checks the msr values to avoid unnecessary reloading.

Signed-off-by: Avi Kivity <avi@redhat.com>

3 files changed, 85 insertions, 0 deletions

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 0558ff8c32a..26a74b7bb6b 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -809,4 +809,7 @@ int kvm_cpu_has_interrupt(struct kvm_vcpu *vcpu);
 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu);
 int kvm_cpu_get_interrupt(struct kvm_vcpu *v);
 
+void kvm_define_shared_msr(unsigned index, u32 msr);
+void kvm_set_shared_msr(unsigned index, u64 val);
+
 #endif /* _ASM_X86_KVM_HOST_H */
diff --git a/arch/x86/kvm/Kconfig b/arch/x86/kvm/Kconfig
index b84e571f417..4cd49833246 100644
--- a/arch/x86/kvm/Kconfig
+++ b/arch/x86/kvm/Kconfig
@@ -28,6 +28,7 @@ config KVM
         select HAVE_KVM_IRQCHIP
         select HAVE_KVM_EVENTFD
         select KVM_APIC_ARCHITECTURE
+        select USER_RETURN_NOTIFIER
         ---help---
           Support hosting fully virtualized guest machines using hardware
           virtualization extensions.  You will need a fairly recent
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index e16cdc9ec0c..58c5cddf363 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -37,6 +37,7 @@
 #include <linux/iommu.h>
 #include <linux/intel-iommu.h>
 #include <linux/cpufreq.h>
+#include <linux/user-return-notifier.h>
 #include <trace/events/kvm.h>
 #undef TRACE_INCLUDE_FILE
 #define CREATE_TRACE_POINTS
@@ -87,6 +88,25 @@ EXPORT_SYMBOL_GPL(kvm_x86_ops);
 int ignore_msrs = 0;
 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
 
+#define KVM_NR_SHARED_MSRS 16
+
+struct kvm_shared_msrs_global {
+        int nr;
+        struct kvm_shared_msr {
+                u32 msr;
+                u64 value;
+        } msrs[KVM_NR_SHARED_MSRS];
+};
+
+struct kvm_shared_msrs {
+        struct user_return_notifier urn;
+        bool registered;
+        u64 current_value[KVM_NR_SHARED_MSRS];
+};
+
+static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
+static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
+
 struct kvm_stats_debugfs_item debugfs_entries[] = {
         { "pf_fixed", VCPU_STAT(pf_fixed) },
         { "pf_guest", VCPU_STAT(pf_guest) },
@@ -123,6 +143,64 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
         { NULL }
 };
 
+static void kvm_on_user_return(struct user_return_notifier *urn)
+{
+        unsigned slot;
+        struct kvm_shared_msr *global;
+        struct kvm_shared_msrs *locals
+                = container_of(urn, struct kvm_shared_msrs, urn);
+
+        for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
+                global = &shared_msrs_global.msrs[slot];
+                if (global->value != locals->current_value[slot]) {
+                        wrmsrl(global->msr, global->value);
+                        locals->current_value[slot] = global->value;
+                }
+        }
+        locals->registered = false;
+        user_return_notifier_unregister(urn);
+}
+
+void kvm_define_shared_msr(unsigned slot, u32 msr)
+{
+        int cpu;
+        u64 value;
+
+        if (slot >= shared_msrs_global.nr)
+                shared_msrs_global.nr = slot + 1;
+        shared_msrs_global.msrs[slot].msr = msr;
+        rdmsrl_safe(msr, &value);
+        shared_msrs_global.msrs[slot].value = value;
+        for_each_online_cpu(cpu)
+                per_cpu(shared_msrs, cpu).current_value[slot] = value;
+}
+EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
+
+static void kvm_shared_msr_cpu_online(void)
+{
+        unsigned i;
+        struct kvm_shared_msrs *locals = &__get_cpu_var(shared_msrs);
+
+        for (i = 0; i < shared_msrs_global.nr; ++i)
+                locals->current_value[i] = shared_msrs_global.msrs[i].value;
+}
+
+void kvm_set_shared_msr(unsigned slot, u64 value)
+{
+        struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
+
+        if (value == smsr->current_value[slot])
+                return;
+        smsr->current_value[slot] = value;
+        wrmsrl(shared_msrs_global.msrs[slot].msr, value);
+        if (!smsr->registered) {
+                smsr->urn.on_user_return = kvm_on_user_return;
+                user_return_notifier_register(&smsr->urn);
+                smsr->registered = true;
+        }
+}
+EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
+
 unsigned long segment_base(u16 selector)
 {
         struct descriptor_table gdt;
@@ -4815,6 +4893,9 @@ int kvm_arch_hardware_enable(void *garbage)
                 int cpu = raw_smp_processor_id();
                 per_cpu(cpu_tsc_khz, cpu) = 0;
         }
+
+        kvm_shared_msr_cpu_online();
+
         return kvm_x86_ops->hardware_enable(garbage);
 }