[PATCH 16/23] kvm: Fix race between timer signals and vcpu entry under !IOTHREAD

From: Marcelo Tosatti <mtosatti@redhat.com>
To: Anthony Liguori <aliguori@us.ibm.com>
Cc: qemu-devel@nongnu.org, kvm@vger.kernel.org,
	Jan Kiszka <jan.kiszka@siemens.com>,
	Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>,
	Marcelo Tosatti <mtosatti@redhat.com>
Subject: [PATCH 16/23] kvm: Fix race between timer signals and vcpu entry under !IOTHREAD
Date: Fri,  4 Feb 2011 13:47:19 -0200	[thread overview]
Message-ID: <4a3cf4661e663df87b315d7556b047b0d8b2aa58.1296834446.git.mtosatti@redhat.com> (raw)
In-Reply-To: <cover.1296834446.git.mtosatti@redhat.com>

From: Jan Kiszka <jan.kiszka@siemens.com>

Found by Stefan Hajnoczi: There is a race in kvm_cpu_exec between
checking for exit_request on vcpu entry and timer signals arriving
before KVM starts to catch them. Plug it by blocking both timer related
signals also on !CONFIG_IOTHREAD and process those via signalfd.

As this fix depends on real signalfd support (otherwise the timer
signals only kick the compat helper thread, and the main thread hangs),
we need to detect the invalid constellation and abort configure.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
CC: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
---
 configure |    6 ++++++
 cpus.c    |   31 ++++++++++++++++++++++++++++++-
 2 files changed, 36 insertions(+), 1 deletions(-)

diff --git a/configure b/configure
index 598e8e1..a3f5345 100755
--- a/configure
+++ b/configure
@@ -2057,6 +2057,12 @@ EOF
 
 if compile_prog "" "" ; then
   signalfd=yes
+elif test "$kvm" = "yes" -a "$io_thread" != "yes"; then
+  echo
+  echo "ERROR: Host kernel lacks signalfd() support,"
+  echo "but KVM depends on it when the IO thread is disabled."
+  echo
+  exit 1
 fi
 
 # check if eventfd is supported
diff --git a/cpus.c b/cpus.c
index 359361f..18caf47 100644
--- a/cpus.c
+++ b/cpus.c
@@ -327,6 +327,12 @@ static void qemu_kvm_eat_signals(CPUState *env)
             exit(1);
         }
     } while (sigismember(&chkset, SIG_IPI) || sigismember(&chkset, SIGBUS));
+
+#ifndef CONFIG_IOTHREAD
+    if (sigismember(&chkset, SIGIO) || sigismember(&chkset, SIGALRM)) {
+        qemu_notify_event();
+    }
+#endif
 }
 
 #else /* _WIN32 */
@@ -376,11 +382,15 @@ static void qemu_kvm_init_cpu_signals(CPUState *env)
 
     sigemptyset(&set);
     sigaddset(&set, SIG_IPI);
+    sigaddset(&set, SIGIO);
+    sigaddset(&set, SIGALRM);
     pthread_sigmask(SIG_BLOCK, &set, NULL);
 
     pthread_sigmask(SIG_BLOCK, NULL, &set);
     sigdelset(&set, SIG_IPI);
     sigdelset(&set, SIGBUS);
+    sigdelset(&set, SIGIO);
+    sigdelset(&set, SIGALRM);
     r = kvm_set_signal_mask(env, &set);
     if (r) {
         fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r));
@@ -389,13 +399,32 @@ static void qemu_kvm_init_cpu_signals(CPUState *env)
 #endif
 }
 
+#ifndef _WIN32
+static sigset_t block_synchronous_signals(void)
+{
+    sigset_t set;
+
+    sigemptyset(&set);
+    if (kvm_enabled()) {
+        /*
+         * We need to process timer signals synchronously to avoid a race
+         * between exit_request check and KVM vcpu entry.
+         */
+        sigaddset(&set, SIGIO);
+        sigaddset(&set, SIGALRM);
+    }
+
+    return set;
+}
+#endif
+
 int qemu_init_main_loop(void)
 {
 #ifndef _WIN32
     sigset_t blocked_signals;
     int ret;
 
-    sigemptyset(&blocked_signals);
+    blocked_signals = block_synchronous_signals();
 
     ret = qemu_signalfd_init(blocked_signals);
     if (ret) {
-- 
1.7.2.3

