KVM: use KVM_CAP_IMMEDIATE_EXIT
The purpose of the KVM_SET_SIGNAL_MASK API is to let userspace "kick"
a VCPU out of KVM_RUN through a POSIX signal. A signal is attached
to a dummy signal handler; by blocking the signal outside KVM_RUN and
unblocking it inside, this possible race is closed:
VCPU thread service thread
--------------------------------------------------------------
check flag
set flag
raise signal
(signal handler does nothing)
KVM_RUN
However, one issue with KVM_SET_SIGNAL_MASK is that it has to take
tsk->sighand->siglock on every KVM_RUN. This lock is often on a
remote NUMA node, because it is on the node of a thread's creator.
Taking this lock can be very expensive if there are many userspace
exits (as is the case for SMP Windows VMs without Hyper-V reference
time counter).
KVM_CAP_IMMEDIATE_EXIT provides an alternative, where the flag is
placed directly in kvm_run so that KVM can see it:
VCPU thread service thread
--------------------------------------------------------------
raise signal
signal handler
set run->immediate_exit
KVM_RUN
check run->immediate_exit
The previous patches changed QEMU so that the only blocked signal is
SIG_IPI, so we can now stop using KVM_SET_SIGNAL_MASK and sigtimedwait
if KVM_CAP_IMMEDIATE_EXIT is available.
On a 14-VCPU guest, an "inl" operation goes down from 30k to 6k on
an unlocked (no BQL) MemoryRegion, or from 30k to 15k if the BQL
is involved.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
diff --git a/kvm-all.c b/kvm-all.c
index 434c720..9040bd5 100644
--- a/kvm-all.c
+++ b/kvm-all.c
@@ -120,6 +120,7 @@
bool kvm_direct_msi_allowed;
bool kvm_ioeventfd_any_length_allowed;
bool kvm_msi_use_devid;
+static bool kvm_immediate_exit;
static const KVMCapabilityInfo kvm_required_capabilites[] = {
KVM_CAP_INFO(USER_MEMORY),
@@ -1619,6 +1620,7 @@
goto err;
}
+ kvm_immediate_exit = kvm_check_extension(s, KVM_CAP_IMMEDIATE_EXIT);
s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
/* If unspecified, use the default value */
@@ -1899,6 +1901,20 @@
static __thread bool have_sigbus_pending;
#endif
+static void kvm_cpu_kick(CPUState *cpu)
+{
+ atomic_set(&cpu->kvm_run->immediate_exit, 1);
+}
+
+static void kvm_cpu_kick_self(void)
+{
+ if (kvm_immediate_exit) {
+ kvm_cpu_kick(current_cpu);
+ } else {
+ qemu_cpu_kick_self();
+ }
+}
+
static void kvm_eat_signals(CPUState *cpu)
{
struct timespec ts = { 0, 0 };
@@ -1907,6 +1923,15 @@
sigset_t chkset;
int r;
+ if (kvm_immediate_exit) {
+ atomic_set(&cpu->kvm_run->immediate_exit, 0);
+ /* Write kvm_run->immediate_exit before the cpu->exit_request
+ * write in kvm_cpu_exec.
+ */
+ smp_wmb();
+ return;
+ }
+
sigemptyset(&waitset);
sigaddset(&waitset, SIG_IPI);
@@ -1955,9 +1980,14 @@
* instruction emulation. This self-signal will ensure that we
* leave ASAP again.
*/
- qemu_cpu_kick_self();
+ kvm_cpu_kick_self();
}
+ /* Read cpu->exit_request before KVM_RUN reads run->immediate_exit.
+ * Matching barrier in kvm_eat_signals.
+ */
+ smp_rmb();
+
run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);
attrs = kvm_arch_post_run(cpu, run);
@@ -2431,8 +2461,12 @@
return r;
}
-static void dummy_signal(int sig)
+static void kvm_ipi_signal(int sig)
{
+ if (current_cpu) {
+ assert(kvm_immediate_exit);
+ kvm_cpu_kick(current_cpu);
+ }
}
void kvm_init_cpu_signals(CPUState *cpu)
@@ -2442,7 +2476,7 @@
struct sigaction sigact;
memset(&sigact, 0, sizeof(sigact));
- sigact.sa_handler = dummy_signal;
+ sigact.sa_handler = kvm_ipi_signal;
sigaction(SIG_IPI, &sigact, NULL);
pthread_sigmask(SIG_BLOCK, NULL, &set);
@@ -2451,7 +2485,11 @@
pthread_sigmask(SIG_SETMASK, &set, NULL);
#endif
sigdelset(&set, SIG_IPI);
- r = kvm_set_signal_mask(cpu, &set);
+ if (kvm_immediate_exit) {
+ r = pthread_sigmask(SIG_SETMASK, &set, NULL);
+ } else {
+ r = kvm_set_signal_mask(cpu, &set);
+ }
if (r) {
fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r));
exit(1);