proc-tid0 uses SIGALRM to stop the test after one second, but the
signal handler calls exit(), which is not async-signal-safe.

Set a sig_atomic_t flag from the handler instead and let the main loop
terminate normally. This also preserves the existing atexit cleanup
that kills the worker child.

Fixes: 0658a0961b0a ("procfs: do not list TID 0 in /proc/<pid>/task")
Signed-off-by: Paul Dolan <paul.dolan.dev@gmail.com>
---
 tools/testing/selftests/proc/proc-tid0.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/tools/testing/selftests/proc/proc-tid0.c b/tools/testing/selftests/proc/proc-tid0.c
index 58c1d7c90a8e..79d3704b2b33 100644
--- a/tools/testing/selftests/proc/proc-tid0.c
+++ b/tools/testing/selftests/proc/proc-tid0.c
@@ -25,6 +25,8 @@
 
 static pid_t pid = -1;
 
+static volatile sig_atomic_t alarmed;
+
 static void atexit_hook(void)
 {
 	if (pid > 0) {
@@ -39,7 +41,7 @@ static void *f(void *_)
 
 static void sigalrm(int _)
 {
-	exit(0);
+	alarmed = 1;
 }
 
 int main(void)
@@ -62,7 +64,7 @@ int main(void)
 		signal(SIGALRM, sigalrm);
 		alarm(1);
 
-		while (1) {
+		while (!alarmed) {
 			DIR *d = opendir(buf);
 			struct dirent *de;
 			while ((de = readdir(d))) {
-- 
2.55.0

On Fri, Sep 04, 2026 at 01:37:52PM +0100, Paul Dolan wrote:
> proc-tid0 uses SIGALRM to stop the test after one second, but the
> signal handler calls exit(), which is not async-signal-safe.
> 
> Set a sig_atomic_t flag from the handler instead and let the main loop
> terminate normally. This also preserves the existing atexit cleanup
> that kills the worker child.
> 
> Fixes: 0658a0961b0a ("procfs: do not list TID 0 in /proc/<pid>/task")
> Signed-off-by: Paul Dolan <paul.dolan.dev@gmail.com>

Yeah, yeah.

Reviewed-off-by: Alexey Dobriyan <adobriyan@gmail.com>

>  static void sigalrm(int _)
>  {
> -	exit(0);
> +	alarmed = 1;
>  }