selftests: coredump: Some bug fixes
cover selftests: coredump: Some bug fixes
Hi,
While trying the coredump test on qemu-system-riscv64, I observed test failures for various reasons.
This series makes the test works on qemu-system-riscv64.
Best regards,
Nam
Nam Cao (3):
selftests: coredump: Properly initialize pointer
selftests: coredump: Use waitpid() instead of busy-wait
selftests: coredump: Raise timeout to 2 minutes
tools/testing/selftests/coredump/stackdump | 6 +----- .../testing/selftests/coredump/stackdump_test.c | 17 +++++++++-------- 2 files changed, 10 insertions(+), 13 deletions(-)
Signature · 1 line
2.39.5
PATCH 1/3 selftests: coredump: Properly initialize pointer
The buffer pointer "line" is not initialized. This pointer is passed to getline().
It can still work if the stack is zero-initialized, because getline() can work with a NULL pointer as buffer.
But this is obviously broken. This bug shows up while running the test on a riscv64 machine.
Fix it by properly initializing the pointer.
--- tools/testing/selftests/coredump/stackdump_test.c | 2 ++ 1 file changed, 2 insertions(+)
tools/testing/selftests/coredump/stackdump_test.c +2 -0
@@ -100,6 +100,8 @@ TEST_F(coredump, stackdump)FILE *file;pid_t pid; + line =NULL ; +/* * Step 1: Setup core_pattern so that the stackdump script is executed when the child * process crashes
Signature · 1 line
2.39.5
PATCH 2/3 selftests: coredump: Use waitpid() instead of busy-wait
In reply to Nam CaoThe test waits for coredump to finish by busy-waiting for the stackdump_values file to be created. The maximum wait time is 10 seconds.
This doesn't work for slow machine (qemu-system-riscv64), because coredump takes longer.
Switch to use waitpid().
With this, the stack_values file doesn't need to be atomically written by coredump anymore, therefore simplify the stackdump script.
--- tools/testing/selftests/coredump/stackdump | 6 +----- tools/testing/selftests/coredump/stackdump_test.c | 13 ++++++------- 2 files changed, 7 insertions(+), 12 deletions(-)
tools/testing/selftests/coredump/stackdump +1 -5
@@ -4,11 +4,7 @@ CRASH_PROGRAM_ID=$1 STACKDUMP_FILE=$2 -TMP=$(mktemp) - for t in /proc/$CRASH_PROGRAM_ID/task/*; do tid=$(basename $t) - cat /proc/$tid/stat | awk '{print $29}' >> $TMP + cat /proc/$tid/stat | awk '{print $29}' >> $STACKDUMP_FILE done - -mv $TMP $STACKDUMP_FILE
tools/testing/selftests/coredump/stackdump_test.c +6 -7
@@ -96,7 +96,7 @@ TEST_F(coredump, stackdump)char *test_dir, *line;size_t line_length;char buf[PATH_MAX ]; -int ret, i; +int ret, i, status;FILE *file;pid_t pid;
@@ -131,12 +131,11 @@ TEST_F(coredump, stackdump)/* * Step 3: Wait for the stackdump script to write the stack pointers to the stackdump file */ -for (i =0 ; i <10 ; ++i) { - file =fopen (STACKDUMP_FILE ,"r" ); -if (file) -break ; -sleep (1 ); - } +waitpid (pid, &status,0 ); +ASSERT_TRUE (WIFSIGNALED (status)); +ASSERT_TRUE (WCOREDUMP (status)); + + file =fopen (STACKDUMP_FILE ,"r" );ASSERT_NE (file,NULL );/* Step 4: Make sure all stack pointer values are non-zero */
Signature · 1 line
2.39.5
PATCH 3/3 selftests: coredump: Raise timeout to 2 minutes
In reply to Nam CaoThe test's runtime (nearly 20s) is dangerously close to the limit (30s) on qemu-system-riscv64:
$ time ./stackdump_test > /dev/null real 0m19.210s user 0m0.077s sys 0m0.359s
There could be machines slower than qemu-system-riscv64. Therefore raise the test timeout to 2 minutes to be safe.
--- tools/testing/selftests/coredump/stackdump_test.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
tools/testing/selftests/coredump/stackdump_test.c +1 -1
@@ -89,7 +89,7 @@ FIXTURE_TEARDOWN(coredump)fprintf (stderr,"Failed to cleanup stackdump test: %s\n" , reason); } -TEST_F(coredump, stackdump) +TEST_F_TIMEOUT(coredump, stackdump, 120) {struct sigaction action = {};unsigned long long stack;
Signature · 1 line
2.39.5
Re: [PATCH 1/3] selftests: coredump: Properly initialize pointer
In reply to Nam CaoQuoted 2 lines · Nam Cao
The buffer pointer "line" is not initialized. This pointer is passed to
getline().
Ouch.
Quoted 19 lines
It can still work if the stack is zero-initialized, because
getline()can work with a NULL pointer as buffer.But this is obviously broken. This bug shows up while running the test on a riscv64 machine.
Fix it by properly initializing the pointer.
Fixes 15858da53542 ("selftests: coredump: Add stackdump test")Signed-off-by Nam Cao <[email protected]>--- tools/testing/selftests/coredump/stackdump_test.c | 2 ++ 1 file changed, 2 insertions(+)tools/testing/selftests/coredump/stackdump_test.c +0
-0
tools/testing/selftests/coredump/stackdump_test.c TEST_F(coredump, stackdump) · line 103
Quoted 4 lines
tools/testing/selftests/coredump/stackdump_test.c +1
-0@@ -100,6 +100,8 @@ TEST_F(coredump, stackdump)FILE *file;pid_t pid; + line =NULL ;
The syntax of getline(3) is quite interesting, since it allocates/reallocates/uses the lineptr as needed and possibly requires the application to free the data. I recommend moving the initialization down to the getline() call and also add the corresponding free().
Something like this:
tools/testing/selftests/coredump/stackdump_test.c +2 -0
@@ -138,10 +138,12 @@ TEST_F(coredump, stackdump)ASSERT_NE (file,NULL );/* Step 4: Make sure all stack pointer values are non-zero */ + line =NULL ;for (i =0 ;-1 !=getline (&line, &line_length, file); ++i) { stack =strtoull (line,NULL ,10 );ASSERT_NE (stack,0 ); } +free (line);ASSERT_EQ (i,1 +NUM_THREAD_SPAWN );
Because of how getline() works, technically your patch is good enough. But we should probably excercise more precision in the use of getline() so as to set a good example.
John Ogness
Re: [PATCH 2/3] selftests: coredump: Use waitpid() instead of busy-wait
In reply to Nam CaoQuoted 7 lines · Nam Cao
The test waits for coredump to finish by busy-waiting for the
stackdump_valuesfile to be created. The maximum wait time is 10 seconds.This doesn't work for slow machine (qemu-system-riscv64), because coredump takes longer.
Switch to use
waitpid().
Note that you are now assuming that returning from waitpid() means that:
1. the coredumping has completed
and
2. the STACKDUMP_FILE with all its contents are visible to the parent process
Quoted 17 lines
With this, the
stack_valuesfile doesn't need to be atomically written by coredump anymore, therefore simplify the stackdump script.Fixes 15858da53542 ("selftests: coredump: Add stackdump test")Signed-off-by Nam Cao <[email protected]>--- tools/testing/selftests/coredump/stackdump | 6 +----- tools/testing/selftests/coredump/stackdump_test.c | 13 ++++++------- 2 files changed, 7 insertions(+), 12 deletions(-)tools/testing/selftests/coredump/stackdump +0
-0@@ -4,11 +4,7 @@ CRASH_PROGRAM_ID=$1 STACKDUMP_FILE=$2
tools/testing/selftests/coredump/stackdump line 14
Quoted 10 lines
tools/testing/selftests/coredump/stackdump +1
-5-TMP=$(mktemp) - for t in /proc/$CRASH_PROGRAM_ID/task/*; do tid=$(basename $t) - cat /proc/$tid/stat | awk '{print $29}' >> $TMP + cat /proc/$tid/stat | awk '{print $29}' >> $STACKDUMP_FILE done - -mv $TMP $STACKDUMP_FILE
I would leave this as it was. Then the availability of STACKDUMP_FILE means the full contents are available.
Quoted 16 lines
tools/testing/selftests/coredump/stackdump_test.c +1
-1@@ -96,7 +96,7 @@ TEST_F(coredump, stackdump)char *test_dir, *line;size_t line_length;char buf[PATH_MAX ]; -int ret, i; +int ret, i, status;FILE *file;pid_t pid;@@ -131,12 +131,11 @@ TEST_F(coredump, stackdump)/* * Step 3: Wait for the stackdump script to write the stack pointers to the stackdump file
tools/testing/selftests/coredump/stackdump_test.c TEST_F(coredump, stackdump) · line 136
Quoted 10 lines
tools/testing/selftests/coredump/stackdump_test.c +3
-6*/ -for (i =0 ; i <10 ; ++i) { - file =fopen (STACKDUMP_FILE ,"r" ); -if (file) -break ; -sleep (1 ); - } +waitpid (pid, &status,0 ); +ASSERT_TRUE (WIFSIGNALED (status)); +ASSERT_TRUE (WCOREDUMP (status));
Why not just put these 3 lines above the for-loop? So you would wait for the process to end, then go into the 20-second timeout loop waiting for STACKDUMP_FILE to show up.
John Ogness