[4/4] plugins: make qemu_plugin_user_exit's locking order consistent with fork_start's

Commit Message

Emilio Cota Jan. 8, 2023, 4:39 p.m. UTC

To fix potential deadlocks as reported by tsan.

Signed-off-by: Emilio Cota <[e-mail]>
---
 plugins/core.c | 16 +++++++++++-----
 1 file changed, 11 insertions(+), 5 deletions(-)

Comments

Richard Henderson Jan. 8, 2023, 7:31 p.m. UTC | #1

On 1/8/23 08:39, Emilio Cota wrote:
> To fix potential deadlocks as reported by tsan.
> 
> Signed-off-by: Emilio Cota<[e-mail]>
> ---
>   plugins/core.c | 16 +++++++++++-----
>   1 file changed, 11 insertions(+), 5 deletions(-)

Reviewed-by: Richard Henderson <[e-mail]>

r~

Patch

diff --git a/plugins/core.c b/plugins/core.c
index ccb770a485..728bacef95 100644
--- a/plugins/core.c
+++ b/plugins/core.c
@@ -500,10 +500,17 @@  void qemu_plugin_user_exit(void)
     enum qemu_plugin_event ev;
     CPUState *cpu;
 
-    QEMU_LOCK_GUARD(&plugin.lock);
-
+    /*
+     * Locking order: we must acquire locks in an order that is consistent
+     * with the one in fork_start(). That is:
+     * - start_exclusive(), which acquires qemu_cpu_list_lock,
+     *   must be called before acquiring plugin.lock.
+     * - tb_flush(), which acquires mmap_lock(), must be called
+     *   while plugin.lock is not held.
+     */
     start_exclusive();
 
+    qemu_rec_mutex_lock(&plugin.lock);
     /* un-register all callbacks except the final AT_EXIT one */
     for (ev = 0; ev < QEMU_PLUGIN_EV_MAX; ev++) {
         if (ev != QEMU_PLUGIN_EV_ATEXIT) {
@@ -513,13 +520,12 @@  void qemu_plugin_user_exit(void)
             }
         }
     }
-
-    tb_flush(current_cpu);
-
     CPU_FOREACH(cpu) {
         qemu_plugin_disable_mem_helpers(cpu);
     }
+    qemu_rec_mutex_unlock(&plugin.lock);
 
+    tb_flush(current_cpu);
     end_exclusive();
 
     /* now it's safe to handle the exit case */