diff --git a/src/debug.c b/src/debug.c
index 30213c9b9..a710f36b0 100644
--- a/src/debug.c
+++ b/src/debug.c
@@ -70,7 +70,6 @@
  * non-null. It's only used by debuggers for core dump analysis.
  */
 struct buffer *thread_dump_buffer = NULL;
-unsigned int panic_started = 0;
 unsigned int debug_commands_issued = 0;
 
 /* dumps a backtrace of the current thread that is appended to buffer <buf>.
@@ -397,10 +396,19 @@ static int debug_parse_cli_show_libs(char **args, char *payload, struct appctx *
 }
 #endif
 
-/* dumps a state of all threads into the trash and on fd #2, then aborts. */
+/* Dumps a state of all threads into the trash and on fd #2, then aborts.
+ * A copy will be put into a trash chunk that's assigned to thread_dump_buffer
+ * so that the debugger can easily find it. This buffer might be truncated if
+ * too many threads are being dumped, but at least we'll dump them all on stderr.
+ * If thread_dump_buffer is set, it means that a panic has already begun.
+ */
 void ha_panic()
 {
-	if (HA_ATOMIC_FETCH_ADD(&panic_started, 1) != 0) {
+	struct buffer *old;
+	unsigned int thr;
+
+	old = NULL;
+	if (!HA_ATOMIC_CAS(&thread_dump_buffer, &old, get_trash_chunk())) {
 		/* a panic dump is already in progress, let's not disturb it,
 		 * we'll be called via signal DEBUGSIG. By returning we may be
 		 * able to leave a current signal handler (e.g. WDT) so that
@@ -408,11 +416,17 @@ void ha_panic()
 		 */
 		return;
 	}
-	thread_dump_buffer = &trash;
+
 	chunk_reset(&trash);
 	chunk_appendf(&trash, "Thread %u is about to kill the process.\n", tid + 1);
-	ha_thread_dump_all_to_trash();
-	DISGUISE(write(2, trash.area, trash.data));
+
+	for (thr = 0; thr < global.nbthread; thr++) {
+		ha_thread_dump(&trash, thr);
+		DISGUISE(write(2, trash.area, trash.data));
+		b_force_xfer(thread_dump_buffer, &trash, b_room(thread_dump_buffer));
+		chunk_reset(&trash);
+	}
+
 	for (;;)
 		abort();
 }