cpus-common: move exclusive work infrastructure from linux-user

This will serve as the base for async_safe_run_on_cpu.  Because
start_exclusive uses CPU_FOREACH, merge exclusive_lock with
qemu_cpu_list_lock: together with a call to exclusive_idle (via
cpu_exec_start/end) in cpu_list_add, this protects exclusive work
against concurrent CPU addition and removal.

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

5 files changed, 127 insertions, 105 deletions

diff --git a/bsd-user/main.c b/bsd-user/main.c
index 6dfa912..35125b7 100644
--- a/bsd-user/main.c
+++ b/bsd-user/main.c
@@ -67,23 +67,6 @@ int cpu_get_pic_interrupt(CPUX86State *env)
 }
 #endif
 
-/* These are no-ops because we are not threadsafe.  */
-static inline void cpu_exec_start(CPUState *cpu)
-{
-}
-
-static inline void cpu_exec_end(CPUState *cpu)
-{
-}
-
-static inline void start_exclusive(void)
-{
-}
-
-static inline void end_exclusive(void)
-{
-}
-
 void fork_start(void)
 {
 }
diff --git a/cpus-common.c b/cpus-common.c
index d6cd426..7d935fd 100644
--- a/cpus-common.c
+++ b/cpus-common.c
@@ -23,11 +23,21 @@
 #include "sysemu/cpus.h"
 
 static QemuMutex qemu_cpu_list_lock;
+static QemuCond exclusive_cond;
+static QemuCond exclusive_resume;
 static QemuCond qemu_work_cond;
 
+static int pending_cpus;
+
 void qemu_init_cpu_list(void)
 {
+    /* This is needed because qemu_init_cpu_list is also called by the
+     * child process in a fork.  */
+    pending_cpus = 0;
+
     qemu_mutex_init(&qemu_cpu_list_lock);
+    qemu_cond_init(&exclusive_cond);
+    qemu_cond_init(&exclusive_resume);
     qemu_cond_init(&qemu_work_cond);
 }
 
@@ -55,6 +65,12 @@ static int cpu_get_free_index(void)
     return cpu_index;
 }
 
+static void finish_safe_work(CPUState *cpu)
+{
+    cpu_exec_start(cpu);
+    cpu_exec_end(cpu);
+}
+
 void cpu_list_add(CPUState *cpu)
 {
     qemu_mutex_lock(&qemu_cpu_list_lock);
@@ -66,6 +82,8 @@ void cpu_list_add(CPUState *cpu)
     }
     QTAILQ_INSERT_TAIL(&cpus, cpu, node);
     qemu_mutex_unlock(&qemu_cpu_list_lock);
+
+    finish_safe_work(cpu);
 }
 
 void cpu_list_remove(CPUState *cpu)
@@ -148,6 +166,70 @@ void async_run_on_cpu(CPUState *cpu, run_on_cpu_func func, void *data)
     queue_work_on_cpu(cpu, wi);
 }
 
+/* Wait for pending exclusive operations to complete.  The CPU list lock
+   must be held.  */
+static inline void exclusive_idle(void)
+{
+    while (pending_cpus) {
+        qemu_cond_wait(&exclusive_resume, &qemu_cpu_list_lock);
+    }
+}
+
+/* Start an exclusive operation.
+   Must only be called from outside cpu_exec, takes
+   qemu_cpu_list_lock.   */
+void start_exclusive(void)
+{
+    CPUState *other_cpu;
+
+    qemu_mutex_lock(&qemu_cpu_list_lock);
+    exclusive_idle();
+
+    /* Make all other cpus stop executing.  */
+    pending_cpus = 1;
+    CPU_FOREACH(other_cpu) {
+        if (other_cpu->running) {
+            pending_cpus++;
+            qemu_cpu_kick(other_cpu);
+        }
+    }
+    while (pending_cpus > 1) {
+        qemu_cond_wait(&exclusive_cond, &qemu_cpu_list_lock);
+    }
+}
+
+/* Finish an exclusive operation.  Releases qemu_cpu_list_lock.  */
+void end_exclusive(void)
+{
+    pending_cpus = 0;
+    qemu_cond_broadcast(&exclusive_resume);
+    qemu_mutex_unlock(&qemu_cpu_list_lock);
+}
+
+/* Wait for exclusive ops to finish, and begin cpu execution.  */
+void cpu_exec_start(CPUState *cpu)
+{
+    qemu_mutex_lock(&qemu_cpu_list_lock);
+    exclusive_idle();
+    cpu->running = true;
+    qemu_mutex_unlock(&qemu_cpu_list_lock);
+}
+
+/* Mark cpu as not executing, and release pending exclusive ops.  */
+void cpu_exec_end(CPUState *cpu)
+{
+    qemu_mutex_lock(&qemu_cpu_list_lock);
+    cpu->running = false;
+    if (pending_cpus > 1) {
+        pending_cpus--;
+        if (pending_cpus == 1) {
+            qemu_cond_signal(&exclusive_cond);
+        }
+    }
+    exclusive_idle();
+    qemu_mutex_unlock(&qemu_cpu_list_lock);
+}
+
 void process_queued_cpu_work(CPUState *cpu)
 {
     struct qemu_work_item *wi;
diff --git a/cpus.c b/cpus.c
index c3afd18..fbd70f5 100644
--- a/cpus.c
+++ b/cpus.c
@@ -1457,7 +1457,9 @@ static int tcg_cpu_exec(CPUState *cpu)
         cpu->icount_decr.u16.low = decr;
         cpu->icount_extra = count;
     }
+    cpu_exec_start(cpu);
     ret = cpu_exec(cpu);
+    cpu_exec_end(cpu);
 #ifdef CONFIG_PROFILER
     tcg_time += profile_getclock() - ti;
 #endif
diff --git a/include/qom/cpu.h b/include/qom/cpu.h
index c04e510..f872614 100644
--- a/include/qom/cpu.h
+++ b/include/qom/cpu.h
@@ -242,7 +242,8 @@ struct qemu_work_item;
  * @nr_threads: Number of threads within this CPU.
  * @numa_node: NUMA node this CPU is belonging to.
  * @host_tid: Host thread ID.
- * @running: #true if CPU is currently running (usermode).
+ * @running: #true if CPU is currently running;
+ * valid under cpu_list_lock.
  * @created: Indicates whether the CPU thread has been successfully created.
  * @interrupt_request: Indicates a pending interrupt request.
  * @halted: Nonzero if the CPU is in suspended state.
@@ -819,6 +820,47 @@ void cpu_remove_sync(CPUState *cpu);
 void process_queued_cpu_work(CPUState *cpu);
 
 /**
+ * cpu_exec_start:
+ * @cpu: The CPU for the current thread.
+ *
+ * Record that a CPU has started execution and can be interrupted with
+ * cpu_exit.
+ */
+void cpu_exec_start(CPUState *cpu);
+
+/**
+ * cpu_exec_end:
+ * @cpu: The CPU for the current thread.
+ *
+ * Record that a CPU has stopped execution and exclusive sections
+ * can be executed without interrupting it.
+ */
+void cpu_exec_end(CPUState *cpu);
+
+/**
+ * start_exclusive:
+ *
+ * Wait for a concurrent exclusive section to end, and then start
+ * a section of work that is run while other CPUs are not running
+ * between cpu_exec_start and cpu_exec_end.  CPUs that are running
+ * cpu_exec are exited immediately.  CPUs that call cpu_exec_start
+ * during the exclusive section go to sleep until this CPU calls
+ * end_exclusive.
+ *
+ * Returns with the CPU list lock taken (which nests outside all
+ * other locks except the BQL).
+ */
+void start_exclusive(void);
+
+/**
+ * end_exclusive:
+ *
+ * Concludes an exclusive execution section started by start_exclusive.
+ * Releases the CPU list lock.
+ */
+void end_exclusive(void);
+
+/**
  * qemu_init_vcpu:
  * @cpu: The vCPU to initialize.
  *
diff --git a/linux-user/main.c b/linux-user/main.c
index e3eca40..c8f8573 100644
--- a/linux-user/main.c
+++ b/linux-user/main.c
@@ -107,28 +107,11 @@ int cpu_get_pic_interrupt(CPUX86State *env)
 /***********************************************************/
 /* Helper routines for implementing atomic operations.  */
 
-/* To implement exclusive operations we force all cpus to syncronise.
-   We don't require a full sync, only that no cpus are executing guest code.
-   The alternative is to map target atomic ops onto host equivalents,
-   which requires quite a lot of per host/target work.  */
-static QemuMutex exclusive_lock;
-static QemuCond exclusive_cond;
-static QemuCond exclusive_resume;
-static int pending_cpus;
-
-void qemu_init_cpu_loop(void)
-{
-    qemu_mutex_init(&exclusive_lock);
-    qemu_cond_init(&exclusive_cond);
-    qemu_cond_init(&exclusive_resume);
-}
-
 /* Make sure everything is in a consistent state for calling fork().  */
 void fork_start(void)
 {
     cpu_list_lock();
     qemu_mutex_lock(&tcg_ctx.tb_ctx.tb_lock);
-    qemu_mutex_lock(&exclusive_lock);
     mmap_fork_start();
 }
 
@@ -144,84 +127,15 @@ void fork_end(int child)
                 QTAILQ_REMOVE(&cpus, cpu, node);
             }
         }
-        pending_cpus = 0;
-        qemu_mutex_init(&exclusive_lock);
-        qemu_cond_init(&exclusive_cond);
-        qemu_cond_init(&exclusive_resume);
         qemu_mutex_init(&tcg_ctx.tb_ctx.tb_lock);
         qemu_init_cpu_list();
         gdbserver_fork(thread_cpu);
     } else {
-        qemu_mutex_unlock(&exclusive_lock);
         qemu_mutex_unlock(&tcg_ctx.tb_ctx.tb_lock);
         cpu_list_unlock();
     }
 }
 
-/* Wait for pending exclusive operations to complete.  The exclusive lock
-   must be held.  */
-static inline void exclusive_idle(void)
-{
-    while (pending_cpus) {
-        qemu_cond_wait(&exclusive_resume, &exclusive_lock);
-    }
-}
-
-/* Start an exclusive operation.
-   Must only be called from outside cpu_exec.   */
-static inline void start_exclusive(void)
-{
-    CPUState *other_cpu;
-
-    qemu_mutex_lock(&exclusive_lock);
-    exclusive_idle();
-
-    pending_cpus = 1;
-    /* Make all other cpus stop executing.  */
-    CPU_FOREACH(other_cpu) {
-        if (other_cpu->running) {
-            pending_cpus++;
-            cpu_exit(other_cpu);
-        }
-    }
-    while (pending_cpus > 1) {
-        qemu_cond_wait(&exclusive_cond, &exclusive_lock);
-    }
-}
-
-/* Finish an exclusive operation.  */
-static inline void __attribute__((unused)) end_exclusive(void)
-{
-    pending_cpus = 0;
-    qemu_cond_broadcast(&exclusive_resume);
-    qemu_mutex_unlock(&exclusive_lock);
-}
-
-/* Wait for exclusive ops to finish, and begin cpu execution.  */
-static inline void cpu_exec_start(CPUState *cpu)
-{
-    qemu_mutex_lock(&exclusive_lock);
-    exclusive_idle();
-    cpu->running = true;
-    qemu_mutex_unlock(&exclusive_lock);
-}
-
-/* Mark cpu as not executing, and release pending exclusive ops.  */
-static inline void cpu_exec_end(CPUState *cpu)
-{
-    qemu_mutex_lock(&exclusive_lock);
-    cpu->running = false;
-    if (pending_cpus > 1) {
-        pending_cpus--;
-        if (pending_cpus == 1) {
-            qemu_cond_signal(&exclusive_cond);
-        }
-    }
-    exclusive_idle();
-    qemu_mutex_unlock(&exclusive_lock);
-}
-
-
 #ifdef TARGET_I386
 /***********************************************************/
 /* CPUX86 core interface */
@@ -4245,7 +4159,6 @@ int main(int argc, char **argv, char **envp)
     int execfd;
 
     qemu_init_cpu_list();
-    qemu_init_cpu_loop();
     module_call_init(MODULE_INIT_QOM);
 
     if ((envlist = envlist_create()) == NULL) {