rt_sigtimedwait01 failed in armv7l

LTP New Cui Bixuan 2 participants 5 messages Dec 13, 2016 – Jan 16, 2017

Share & tools

Cui Bixuan

author

2016-12-13 09:09 UTC #0 lore raw

Hi, These is a fail about rt_sigtimedwait01(armv7l GNU/Linux and linux 3.10):

root@arma9el:/tmp# ./rt_sigtimedwait01
rt_sigtimedwait01    0  TINFO  :  0x9fd0, 10
rt_sigtimedwait01    1  TPASS  :  Test passed
rt_sigtimedwait01    0  TINFO  :  0x9fd0, 10
rt_sigtimedwait01    2  TPASS  :  Test passed
...
rt_sigtimedwait01    0  TINFO  :  0x9fd0, 12
rt_sigtimedwait01   11  TPASS  :  Test passed
rt_sigtimedwait01    0  TINFO  :  0x9fd0, 15
rt_sigtimedwait01   12  TPASS  :  Test passed
rt_sigtimedwait01   12  TPASS  :  sigwaitinfo restored the original mask
rt_sigtimedwait01    0  TINFO  :  0x9fd0, -1

rt_sigtimedwait01 13 TFAIL : /home/cuibixuan/ltp/testcases/kernel/syscalls/rt_sigtimedwait/../sigwaitinfo/sigwaitinfo01.c:56:
test_masked_matching_rt (/home/cuibixuan/ltp/testcases/kernel/syscalls/rt_sigtimedwait/../sigwaitinfo/sigwaitinfo01.c: 274): Unexpected return value; expected 35, got 36
rt_sigtimedwait01 13 TFAIL : /home/cuibixuan/ltp/testcases/kernel/syscalls/rt_sigtimedwait/../sigwaitinfo/sigwaitinfo01.c:56:
test_masked_matching_rt (/home/cuibixuan/ltp/testcases/kernel/syscalls/rt_sigtimedwait/../sigwaitinfo/sigwaitinfo01.c: 282): Unexpected return value; expected 36, got 35
rt_sigtimedwait01 13 TPASS : sigwaitinfo restored the original mask

I add 'sleep(1)' into test_masked_matching_rt() of ltp/testcases/kernel/syscalls/sigwaitinfo/sigwaitinfo01.c:268

void test_masked_matching_rt(swi_func sigwaitinfo, int signo)
...
/* Run a child that will wake us up */

        child[0] = create_sig_proc(0, signo, 1);
+       sleep(1);
        child[1] = create_sig_proc(0, signo + 1, 1);

Then the case pass:

rt_sigtimedwait01   13  TPASS  :  Test passed
rt_sigtimedwait01   13  TPASS  :  Test passed
rt_sigtimedwait01   13  TPASS  :  sigwaitinfo restored the original mask

Or run it by 'chrt 10 ./rt_sigtimedwait01' then pass.

Why? Can someone give me a hand.

Thanks,
Cui Bixuan

Cyril Hrubis

2016-12-13 10:43 UTC #1 lore raw

Hi!

Quoted 30 lines

Hi, These is a fail about rt_sigtimedwait01(armv7l GNU/Linux and linux 3.10):
root@arma9el:/tmp# ./rt_sigtimedwait01
rt_sigtimedwait01    0  TINFO  :  0x9fd0, 10
rt_sigtimedwait01    1  TPASS  :  Test passed
rt_sigtimedwait01    0  TINFO  :  0x9fd0, 10
rt_sigtimedwait01    2  TPASS  :  Test passed
...
rt_sigtimedwait01    0  TINFO  :  0x9fd0, 12
rt_sigtimedwait01   11  TPASS  :  Test passed
rt_sigtimedwait01    0  TINFO  :  0x9fd0, 15
rt_sigtimedwait01   12  TPASS  :  Test passed
rt_sigtimedwait01   12  TPASS  :  sigwaitinfo restored the original mask
rt_sigtimedwait01    0  TINFO  :  0x9fd0, -1
rt_sigtimedwait01 13 TFAIL : /home/cuibixuan/ltp/testcases/kernel/syscalls/rt_sigtimedwait/../sigwaitinfo/sigwaitinfo01.c:56:
test_masked_matching_rt (/home/cuibixuan/ltp/testcases/kernel/syscalls/rt_sigtimedwait/../sigwaitinfo/sigwaitinfo01.c: 274): Unexpected return value; expected 35, got 36
rt_sigtimedwait01 13 TFAIL : /home/cuibixuan/ltp/testcases/kernel/syscalls/rt_sigtimedwait/../sigwaitinfo/sigwaitinfo01.c:56:
test_masked_matching_rt (/home/cuibixuan/ltp/testcases/kernel/syscalls/rt_sigtimedwait/../sigwaitinfo/sigwaitinfo01.c: 282): Unexpected return value; expected 36, got 35
rt_sigtimedwait01 13 TPASS : sigwaitinfo restored the original mask
I add 'sleep(1)' into test_masked_matching_rt() of ltp/testcases/kernel/syscalls/sigwaitinfo/sigwaitinfo01.c:268
void test_masked_matching_rt(swi_func sigwaitinfo, int signo)
...
/* Run a child that will wake us up */
        child[0] = create_sig_proc(0, signo, 1);
+       sleep(1);
        child[1] = create_sig_proc(0, signo + 1, 1);

Quoted 8 lines

Then the case pass:
rt_sigtimedwait01   13  TPASS  :  Test passed
rt_sigtimedwait01   13  TPASS  :  Test passed
rt_sigtimedwait01   13  TPASS  :  sigwaitinfo restored the original mask
Or run it by 'chrt 10 ./rt_sigtimedwait01' then pass.
Why? Can someone give me a hand.

I guess that the problem here is that the order in which the test children and parent are executed is undefined.

The signals should be delivered starting from lowest numbered but that applies only when both signals are pending at the point we call sigwaitinfo(). If, by a chance, the first child haven't had chance to finish until we get to the sigwaitinfo() call we will get the signo+1 signal first, since that is the only one pending at that point.

So the fix here would be making sure that both children are finished before we call the sigwaitinfo(). Quite possibly doing waitpid() for both of them will suffice.

Signature · 2 lines

Cyril Hrubis
[email protected]

Cui Bixuan

author

2016-12-15 02:37 UTC #2 lore raw

Quoted 9 lines · Cyril Hrubis

I guess that the problem here is that the order in which the test children and parent are executed is undefined.
The signals should be delivered starting from lowest numbered but that applies only when both signals are pending at the point we call sigwaitinfo(). If, by a chance, the first child haven't had chance to finish until we get to the sigwaitinfo() call we will get the signo+1 signal first, since that is the only one pending at that point.

Quoted 3 lines

So the fix here would be making sure that both children are finished before we call the sigwaitinfo(). Quite possibly doing waitpid() for both of them will suffice.

Thank you. I will try to fix it.

Thanks,
Cui Bixuan

Cui Bixuan

author

2016-12-28 12:14 UTC #3 lore raw

PATCH 1/1 rt_sigtimedwait01: fix probability of failure

If child[0] of test_masked_matching_rt() haven't had chance to finish until we get to the sigwaitinfo() call, the case will fail.

So we should add waitpid() to make sure that both children are finished.

Signed-off-by Cui Bixuan <[email protected]>

---
 testcases/kernel/syscalls/utils/common_j_h.c | 3 +++
 1 file changed, 3 insertions(+)

testcases/kernel/syscalls/utils/common_j_h.c +3 -0

index 43165ca..784ca04 100644
@@ -119,6 +119,7 @@ static void sigterm_handler(int sig)
 pid_t create_sig_proc(unsigned long usec, int sig, unsigned count)
 {
 	pid_t pid, cpid;
+	int status;
 
 	pid = getpid();
 	WITH_SIGNALS_BLOCKED(if ((cpid = fork()) == 0) {

@@ -138,6 +139,8 @@ pid_t create_sig_proc(unsigned long usec, int sig, unsigned count)
 		EPRINTF("fork failed.\n");
 		return cpid;
 	default:
+		/* wait cpid finish */
+		waitpid(cpid, &status, 0);
 		return cpid;
 	}
 }

Signature · 1 line

1.8.3.4

Cyril Hrubis

2017-01-16 13:33 UTC #4 lore raw

PATCH 1/1 rt_sigtimedwait01: fix probability of failure

Hi!

testcases/kernel/syscalls/utils/common_j_h.c pid_t create_sig_proc(unsigned long usec, int sig, unsigned count) · line 142

Quoted 7 lines

testcases/kernel/syscalls/utils/common_j_h.c +1 -0

 	pid = getpid();
 	WITH_SIGNALS_BLOCKED(if ((cpid = fork()) == 0) {

@@ -138,6 +139,8 @@ pid_t create_sig_proc(unsigned long usec, int sig, unsigned count)
 		EPRINTF("fork failed.\n");
 		return cpid;
 	default:
+		/* wait cpid finish */

The comment here is redundant and useless.

Quoted 8 lines

testcases/kernel/syscalls/utils/common_j_h.c +1 -0
+		waitpid(cpid, &status, 0);
 		return cpid;
 	}
 }
Signature · 1 line
1.8.3.4

Signature · 2 lines

Cyril Hrubis
[email protected]