qom: convert the CPU list to RCU

Iterating over the list without using atomics is undefined behaviour,
since the list can be modified concurrently by other threads (e.g.
every time a new thread is created in user-mode).

Fix it by implementing the CPU list as an RCU QTAILQ. This requires
a little bit of extra work to traverse list in reverse order (see
previous patch), but other than that the conversion is trivial.

Signed-off-by: Emilio G. Cota <cota@braap.org>
Message-Id: <20180819091335.22863-12-cota@braap.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

cpus-common.c

@@ -84,7 +84,7 @@ void cpu_list_add(CPUState *cpu)
     } else {
         assert(!cpu_index_auto_assigned);
     }
-    QTAILQ_INSERT_TAIL(&cpus, cpu, node);
+    QTAILQ_INSERT_TAIL_RCU(&cpus, cpu, node);
     qemu_mutex_unlock(&qemu_cpu_list_lock);
 
     finish_safe_work(cpu);
@@ -101,7 +101,7 @@ void cpu_list_remove(CPUState *cpu)
 
     assert(!(cpu_index_auto_assigned && cpu != QTAILQ_LAST(&cpus, CPUTailQ)));
 
-    QTAILQ_REMOVE(&cpus, cpu, node);
+    QTAILQ_REMOVE_RCU(&cpus, cpu, node);
     cpu->cpu_index = UNASSIGNED_CPU_INDEX;
     qemu_mutex_unlock(&qemu_cpu_list_lock);
 }

cpus.c

@@ -1491,7 +1491,7 @@ static void *qemu_tcg_rr_cpu_thread_fn(void *arg)
             atomic_mb_set(&cpu->exit_request, 0);
         }
 
-        qemu_tcg_rr_wait_io_event(cpu ? cpu : QTAILQ_FIRST(&cpus));
+        qemu_tcg_rr_wait_io_event(cpu ? cpu : first_cpu);
         deal_with_unplugged_cpus();
     }
 

include/qom/cpu.h

@@ -26,6 +26,7 @@
 #include "exec/memattrs.h"
 #include "qapi/qapi-types-run-state.h"
 #include "qemu/bitmap.h"
+#include "qemu/rcu_queue.h"
 #include "qemu/queue.h"
 #include "qemu/thread.h"
 
@@ -442,13 +443,11 @@ struct CPUState {
 
 QTAILQ_HEAD(CPUTailQ, CPUState);
 extern struct CPUTailQ cpus;
-#define CPU_NEXT(cpu) QTAILQ_NEXT(cpu, node)
-#define CPU_FOREACH(cpu) QTAILQ_FOREACH(cpu, &cpus, node)
+#define first_cpu        QTAILQ_FIRST_RCU(&cpus)
+#define CPU_NEXT(cpu)    QTAILQ_NEXT_RCU(cpu, node)
+#define CPU_FOREACH(cpu) QTAILQ_FOREACH_RCU(cpu, &cpus, node)
 #define CPU_FOREACH_SAFE(cpu, next_cpu) \
-    QTAILQ_FOREACH_SAFE(cpu, &cpus, node, next_cpu)
-#define CPU_FOREACH_REVERSE(cpu) \
-    QTAILQ_FOREACH_REVERSE(cpu, &cpus, CPUTailQ, node)
-#define first_cpu QTAILQ_FIRST(&cpus)
+    QTAILQ_FOREACH_SAFE_RCU(cpu, &cpus, node, next_cpu)
 
 extern __thread CPUState *current_cpu;
 

linux-user/main.c

@@ -126,7 +126,7 @@ void fork_end(int child)
            Discard information about the parent threads.  */
         CPU_FOREACH_SAFE(cpu, next_cpu) {
             if (cpu != thread_cpu) {
-                QTAILQ_REMOVE(&cpus, cpu, node);
+                QTAILQ_REMOVE_RCU(&cpus, cpu, node);
             }
         }
         qemu_init_cpu_list();

linux-user/syscall.c

@@ -8157,7 +8157,7 @@ abi_long do_syscall(void *cpu_env, int num, abi_long arg1,
             TaskState *ts;
 
             /* Remove the CPU from the list.  */
-            QTAILQ_REMOVE(&cpus, cpu, node);
+            QTAILQ_REMOVE_RCU(&cpus, cpu, node);
 
             cpu_list_unlock();
 

target/s390x/cpu_models.c

@@ -1096,7 +1096,7 @@ void s390_set_qemu_cpu_model(uint16_t type, uint8_t gen, uint8_t ec_ga,
     const S390CPUDef *def = s390_find_cpu_def(type, gen, ec_ga, NULL);
 
     g_assert(def);
-    g_assert(QTAILQ_EMPTY(&cpus));
+    g_assert(QTAILQ_EMPTY_RCU(&cpus));
 
     /* TCG emulates some features that can usually not be enabled with
      * the emulated machine generation. Make sure they can be enabled