Re: [PATCH v2 6/8] selftests/mm: Make migration test robust to failure

[David Hildenbrand <david@redhat.com>]

On 18.07.23 14:42, Ryan Roberts wrote:
> On 18/07/2023 12:24, David Hildenbrand wrote:
>> On 18.07.23 13:23, David Hildenbrand wrote:
>>> On 18.07.23 12:49, Ryan Roberts wrote:
>>>> On 17/07/2023 18:40, David Hildenbrand wrote:
>>>>> On 17.07.23 12:31, Ryan Roberts wrote:
>>>>>> The `migration` test currently has a number of robustness problems that
>>>>>> cause it to hang and leak resources.
>>>>>>
>>>>>> Timeout: There are 3 tests, which each previously ran for 60 seconds.
>>>>>> However, the timeout in mm/settings for a single test binary was set to
>>>>>> 45 seconds. So when run using run_kselftest.sh, the top level timeout
>>>>>> would trigger before the test binary was finished. Solve this by meeting
>>>>>> in the middle; each of the 3 tests now runs for 20 seconds (for a total
>>>>>> of 60), and the top level timeout is set to 90 seconds.
>>>>>>
>>>>>> Leaking child processes: the `shared_anon` test fork()s some children
>>>>>> but then an ASSERT() fires before the test kills those children. The
>>>>>> assert causes immediate exit of the parent and leaking of the children.
>>>>>> Furthermore, if run using the run_kselftest.sh wrapper, the wrapper
>>>>>> would get stuck waiting for those children to exit, which never happens.
>>>>>> Solve this by deferring any asserts until after the children are killed.
>>>>>> The same pattern is used for the threaded tests for uniformity.
>>>>>>
>>>>>> With these changes, the test binary now runs to completion on arm64,
>>>>>> with 2 tests passing and the `shared_anon` test failing.
>>>>>>
>>>>>> Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
>>>>>> ---
>>>>>>      tools/testing/selftests/mm/migration.c | 14 ++++++++++----
>>>>>>      tools/testing/selftests/mm/settings    |  2 +-
>>>>>>      2 files changed, 11 insertions(+), 5 deletions(-)
>>>>>>
>>>>>> diff --git a/tools/testing/selftests/mm/migration.c
>>>>>> b/tools/testing/selftests/mm/migration.c
>>>>>> index 379581567f27..189d7d9070e8 100644
>>>>>> --- a/tools/testing/selftests/mm/migration.c
>>>>>> +++ b/tools/testing/selftests/mm/migration.c
>>>>>> @@ -15,7 +15,7 @@
>>>>>>      #include <time.h>
>>>>>>        #define TWOMEG (2<<20)
>>>>>> -#define RUNTIME (60)
>>>>>> +#define RUNTIME (20)
>>>>>>        #define ALIGN(x, a) (((x) + (a - 1)) & (~((a) - 1)))
>>>>>>      @@ -118,6 +118,7 @@ TEST_F_TIMEOUT(migration, private_anon, 2*RUNTIME)
>>>>>>      {
>>>>>>          uint64_t *ptr;
>>>>>>          int i;
>>>>>> +    int ret;
>>>>>>            if (self->nthreads < 2 || self->n1 < 0 || self->n2 < 0)
>>>>>>              SKIP(return, "Not enough threads or NUMA nodes available");
>>>>>> @@ -131,9 +132,10 @@ TEST_F_TIMEOUT(migration, private_anon, 2*RUNTIME)
>>>>>>              if (pthread_create(&self->threads[i], NULL, access_mem, ptr))
>>>>>>                  perror("Couldn't create thread");
>>>>>>      -    ASSERT_EQ(migrate(ptr, self->n1, self->n2), 0);
>>>>>> +    ret = migrate(ptr, self->n1, self->n2);
>>>>>>          for (i = 0; i < self->nthreads - 1; i++)
>>>>>>              ASSERT_EQ(pthread_cancel(self->threads[i]), 0);
>>>>>> +    ASSERT_EQ(ret, 0);
>>>>>
>>>>> Why is that required? This does not involve fork.
>>>>
>>>> It's not required. I was just trying to keep everything aligned to the same
>>>> pattern.
>>>>
>>>>>
>>>>>>      }
>>>>>>        /*
>>>>>> @@ -144,6 +146,7 @@ TEST_F_TIMEOUT(migration, shared_anon, 2*RUNTIME)
>>>>>>          pid_t pid;
>>>>>>          uint64_t *ptr;
>>>>>>          int i;
>>>>>> +    int ret;
>>>>>>            if (self->nthreads < 2 || self->n1 < 0 || self->n2 < 0)
>>>>>>              SKIP(return, "Not enough threads or NUMA nodes available");
>>>>>> @@ -161,9 +164,10 @@ TEST_F_TIMEOUT(migration, shared_anon, 2*RUNTIME)
>>>>>>                  self->pids[i] = pid;
>>>>>>          }
>>>>>>      -    ASSERT_EQ(migrate(ptr, self->n1, self->n2), 0);
>>>>>> +    ret = migrate(ptr, self->n1, self->n2);
>>>>>>          for (i = 0; i < self->nthreads - 1; i++)
>>>>>>              ASSERT_EQ(kill(self->pids[i], SIGTERM), 0);
>>>>>> +    ASSERT_EQ(ret, 0);
>>>>>
>>>>>
>>>>> Might be cleaner to also:
>>>>
>>>> Or instead of? I agree this is neater, so will undo the moving of the ASSERT()
>>>> and rely on this prctl.
>>>
>>> I was thinking about possible races when our parent process already
>>> quits before our child managed to set the prctl. prctl() won't do
>>> anything in that case, hmmmm.
>>>
>>> But similarly, existing code might already trigger the migrate() + kill
>>> before the child processes even started to access_mem().
>>>
>>> Racy :)
>>>
>>
>> Maybe what would work, is checking after the prctl() in the child if the parent
>> is already gone.
> 
> 
> Like this?
> 
> 	if (!pid) {
> 		prctl(PR_SET_PDEATHSIG, SIGHUP);
> 		/* Parent may have died before prctl so check now. */
> 		if (getppid() == 1)
> 			kill(getpid(), SIGHUP);
> 		access_mem(ptr);
> 	}

Staring at forget_original_parent(), that order should work.

I do wonder if there is a nicer way to handle that, but maybe that 
already is the "nice" way.

-- 
Cheers,

David / dhildenb