[Xenomai-core] Posix mutex unit test fails on git head

[Xenomai-core] Posix mutex unit test fails on git head

[Jan Kiszka]

Gilles,

can you (or someone else) confirm this:

# mutex-torture-posix
simple_wait
simple mutex_lock 1: 1 (Operation not permitted)

I'm on a x86_64 target (2.6.32.8 with latest I-pipe), Xenomai is
unpatched git head.

[ But I'm currently facing general stability issues with 2.6.32 again.
And sigtest is failing loudly here, typically hard lockups. Ugh. ]

Jan

-- 
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux

Re: [Xenomai-core] Posix mutex unit test fails on git head

[Jan Kiszka]

Jan Kiszka wrote:
> Gilles,
> 
> can you (or someone else) confirm this:
> 
> # mutex-torture-posix
> simple_wait
> simple mutex_lock 1: 1 (Operation not permitted)
> 
> I'm on a x86_64 target (2.6.32.8 with latest I-pipe), Xenomai is
> unpatched git head.
> 

Looks like a bug in the test case: We do not wrap sched_setscheduler
which it uses, and my config does not auto-shadow main. Fix will follow.

Jan

-- 
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux

Re: [Xenomai-core] Posix mutex unit test fails on git head

[Gilles Chanteperdrix]

Jan Kiszka wrote:
> Jan Kiszka wrote:
>> Gilles,
>>
>> can you (or someone else) confirm this:
>>
>> # mutex-torture-posix
>> simple_wait
>> simple mutex_lock 1: 1 (Operation not permitted)
>>
>> I'm on a x86_64 target (2.6.32.8 with latest I-pipe), Xenomai is
>> unpatched git head.
>>
> 
> Looks like a bug in the test case: We do not wrap sched_setscheduler
> which it uses, and my config does not auto-shadow main. Fix will follow.

The test works here, but fails in cond_signaler, mutex_lock 2, just
checked out v2.5.1 to see if we already had the bug.

> 
> Jan
> 


-- 
					    Gilles.

Re: [Xenomai-core] Posix mutex unit test fails on git head

[Jan Kiszka]

Gilles Chanteperdrix wrote:
> Jan Kiszka wrote:
>> Jan Kiszka wrote:
>>> Gilles,
>>>
>>> can you (or someone else) confirm this:
>>>
>>> # mutex-torture-posix
>>> simple_wait
>>> simple mutex_lock 1: 1 (Operation not permitted)
>>>
>>> I'm on a x86_64 target (2.6.32.8 with latest I-pipe), Xenomai is
>>> unpatched git head.
>>>
>> Looks like a bug in the test case: We do not wrap sched_setscheduler
>> which it uses, and my config does not auto-shadow main. Fix will follow.
> 
> The test works here, but fails in cond_signaler, mutex_lock 2, just
> checked out v2.5.1 to see if we already had the bug.
> 

Weird. Works for me with this fix:

diff --git a/src/testsuite/unit/mutex-torture.c b/src/testsuite/unit/mutex-torture.c
index 70bda14..ebdd4f2 100644
--- a/src/testsuite/unit/mutex-torture.c
+++ b/src/testsuite/unit/mutex-torture.c
@@ -657,7 +657,7 @@ int main(void)
 	/* Set scheduling parameters for the current process */
 #ifdef XENO_POSIX
 	sparam.sched_priority = 2;
-	sched_setscheduler(0, SCHED_FIFO, &sparam);
+	pthread_setschedparam(pthread_self(), SCHED_FIFO, &sparam);
 #else /* __NATIVE_SKIN__ */
 	rt_task_shadow(&main_tid, "main_task", 2, 0);
 #endif /* __NATIVE_SKIN__ */


# mutex-torture-posix
simple_wait
recursive_wait
errorcheck_wait
mode_switch
pi_wait
lock_stealing
lock_stealing mutex_trylock: not supported
simple_condwait
recursive_condwait
cond_signaler, mutex_lock 2 waited 0.854 us


Jan

-- 
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux

Re: [Xenomai-core] Posix mutex unit test fails on git head

[Gilles Chanteperdrix]

Jan Kiszka wrote:
> Gilles Chanteperdrix wrote:
>> Jan Kiszka wrote:
>>> Jan Kiszka wrote:
>>>> Gilles,
>>>>
>>>> can you (or someone else) confirm this:
>>>>
>>>> # mutex-torture-posix
>>>> simple_wait
>>>> simple mutex_lock 1: 1 (Operation not permitted)
>>>>
>>>> I'm on a x86_64 target (2.6.32.8 with latest I-pipe), Xenomai is
>>>> unpatched git head.
>>>>
>>> Looks like a bug in the test case: We do not wrap sched_setscheduler
>>> which it uses, and my config does not auto-shadow main. Fix will follow.
>> The test works here, but fails in cond_signaler, mutex_lock 2, just
>> checked out v2.5.1 to see if we already had the bug.
>>
> 
> Weird. Works for me with this fix:
> 
> diff --git a/src/testsuite/unit/mutex-torture.c b/src/testsuite/unit/mutex-torture.c
> index 70bda14..ebdd4f2 100644
> --- a/src/testsuite/unit/mutex-torture.c
> +++ b/src/testsuite/unit/mutex-torture.c
> @@ -657,7 +657,7 @@ int main(void)
>  	/* Set scheduling parameters for the current process */
>  #ifdef XENO_POSIX
>  	sparam.sched_priority = 2;
> -	sched_setscheduler(0, SCHED_FIFO, &sparam);
> +	pthread_setschedparam(pthread_self(), SCHED_FIFO, &sparam);
>  #else /* __NATIVE_SKIN__ */
>  	rt_task_shadow(&main_tid, "main_task", 2, 0);
>  #endif /* __NATIVE_SKIN__ */
> 
> 
> # mutex-torture-posix
> simple_wait
> recursive_wait
> errorcheck_wait
> mode_switch
> pi_wait
> lock_stealing
> lock_stealing mutex_trylock: not supported
> simple_condwait
> recursive_condwait
> cond_signaler, mutex_lock 2 waited 0.854 us

No, that is a bug. We should have waited 10ms, and we wait less than a us.

-- 
					    Gilles.

Re: [Xenomai-core] Posix mutex unit test fails on git head

[Jan Kiszka]

Gilles Chanteperdrix wrote:
> Jan Kiszka wrote:
>> Gilles Chanteperdrix wrote:
>>> Jan Kiszka wrote:
>>>> Jan Kiszka wrote:
>>>>> Gilles,
>>>>>
>>>>> can you (or someone else) confirm this:
>>>>>
>>>>> # mutex-torture-posix
>>>>> simple_wait
>>>>> simple mutex_lock 1: 1 (Operation not permitted)
>>>>>
>>>>> I'm on a x86_64 target (2.6.32.8 with latest I-pipe), Xenomai is
>>>>> unpatched git head.
>>>>>
>>>> Looks like a bug in the test case: We do not wrap sched_setscheduler
>>>> which it uses, and my config does not auto-shadow main. Fix will follow.
>>> The test works here, but fails in cond_signaler, mutex_lock 2, just
>>> checked out v2.5.1 to see if we already had the bug.
>>>
>> Weird. Works for me with this fix:
>>
>> diff --git a/src/testsuite/unit/mutex-torture.c b/src/testsuite/unit/mutex-torture.c
>> index 70bda14..ebdd4f2 100644
>> --- a/src/testsuite/unit/mutex-torture.c
>> +++ b/src/testsuite/unit/mutex-torture.c
>> @@ -657,7 +657,7 @@ int main(void)
>>  	/* Set scheduling parameters for the current process */
>>  #ifdef XENO_POSIX
>>  	sparam.sched_priority = 2;
>> -	sched_setscheduler(0, SCHED_FIFO, &sparam);
>> +	pthread_setschedparam(pthread_self(), SCHED_FIFO, &sparam);
>>  #else /* __NATIVE_SKIN__ */
>>  	rt_task_shadow(&main_tid, "main_task", 2, 0);
>>  #endif /* __NATIVE_SKIN__ */
>>
>>
>> # mutex-torture-posix
>> simple_wait
>> recursive_wait
>> errorcheck_wait
>> mode_switch
>> pi_wait
>> lock_stealing
>> lock_stealing mutex_trylock: not supported
>> simple_condwait
>> recursive_condwait
>> cond_signaler, mutex_lock 2 waited 0.854 us
> 
> No, that is a bug. We should have waited 10ms, and we wait less than a us.
> 

Ah - well, we do not properly return ETIMEDOUT anymore, do we?

Jan

-- 
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux

Re: [Xenomai-core] Posix mutex unit test fails on git head

[Gilles Chanteperdrix]

Jan Kiszka wrote:
> Gilles Chanteperdrix wrote:
>> Jan Kiszka wrote:
>>> Gilles Chanteperdrix wrote:
>>>> Jan Kiszka wrote:
>>>>> Jan Kiszka wrote:
>>>>>> Gilles,
>>>>>>
>>>>>> can you (or someone else) confirm this:
>>>>>>
>>>>>> # mutex-torture-posix
>>>>>> simple_wait
>>>>>> simple mutex_lock 1: 1 (Operation not permitted)
>>>>>>
>>>>>> I'm on a x86_64 target (2.6.32.8 with latest I-pipe), Xenomai is
>>>>>> unpatched git head.
>>>>>>
>>>>> Looks like a bug in the test case: We do not wrap sched_setscheduler
>>>>> which it uses, and my config does not auto-shadow main. Fix will follow.
>>>> The test works here, but fails in cond_signaler, mutex_lock 2, just
>>>> checked out v2.5.1 to see if we already had the bug.
>>>>
>>> Weird. Works for me with this fix:
>>>
>>> diff --git a/src/testsuite/unit/mutex-torture.c b/src/testsuite/unit/mutex-torture.c
>>> index 70bda14..ebdd4f2 100644
>>> --- a/src/testsuite/unit/mutex-torture.c
>>> +++ b/src/testsuite/unit/mutex-torture.c
>>> @@ -657,7 +657,7 @@ int main(void)
>>>  	/* Set scheduling parameters for the current process */
>>>  #ifdef XENO_POSIX
>>>  	sparam.sched_priority = 2;
>>> -	sched_setscheduler(0, SCHED_FIFO, &sparam);
>>> +	pthread_setschedparam(pthread_self(), SCHED_FIFO, &sparam);
>>>  #else /* __NATIVE_SKIN__ */
>>>  	rt_task_shadow(&main_tid, "main_task", 2, 0);
>>>  #endif /* __NATIVE_SKIN__ */
>>>
>>>
>>> # mutex-torture-posix
>>> simple_wait
>>> recursive_wait
>>> errorcheck_wait
>>> mode_switch
>>> pi_wait
>>> lock_stealing
>>> lock_stealing mutex_trylock: not supported
>>> simple_condwait
>>> recursive_condwait
>>> cond_signaler, mutex_lock 2 waited 0.854 us
>> No, that is a bug. We should have waited 10ms, and we wait less than a us.
>>
> 
> Ah - well, we do not properly return ETIMEDOUT anymore, do we?

No, it is a recursive mutex issue. We are able to acquire a mutex,
whereas we should not, someone else has it. What I do not understand, is
that I am sure I had tested this before the 2.5.0 release, and that even
with the 2.5.0 the test fails.

> 
> Jan
> 


-- 
					    Gilles.

Re: [Xenomai-core] Posix mutex unit test fails on git head

[Gilles Chanteperdrix]

Gilles Chanteperdrix wrote:
> Jan Kiszka wrote:
>> Gilles Chanteperdrix wrote:
>>> Jan Kiszka wrote:
>>>> Jan Kiszka wrote:
>>>>> Gilles,
>>>>>
>>>>> can you (or someone else) confirm this:
>>>>>
>>>>> # mutex-torture-posix
>>>>> simple_wait
>>>>> simple mutex_lock 1: 1 (Operation not permitted)
>>>>>
>>>>> I'm on a x86_64 target (2.6.32.8 with latest I-pipe), Xenomai is
>>>>> unpatched git head.
>>>>>
>>>> Looks like a bug in the test case: We do not wrap sched_setscheduler
>>>> which it uses, and my config does not auto-shadow main. Fix will follow.
>>> The test works here, but fails in cond_signaler, mutex_lock 2, just
>>> checked out v2.5.1 to see if we already had the bug.
>>>
>> Weird. Works for me with this fix:
>>
>> diff --git a/src/testsuite/unit/mutex-torture.c b/src/testsuite/unit/mutex-torture.c
>> index 70bda14..ebdd4f2 100644
>> --- a/src/testsuite/unit/mutex-torture.c
>> +++ b/src/testsuite/unit/mutex-torture.c
>> @@ -657,7 +657,7 @@ int main(void)
>>  	/* Set scheduling parameters for the current process */
>>  #ifdef XENO_POSIX
>>  	sparam.sched_priority = 2;
>> -	sched_setscheduler(0, SCHED_FIFO, &sparam);
>> +	pthread_setschedparam(pthread_self(), SCHED_FIFO, &sparam);
>>  #else /* __NATIVE_SKIN__ */
>>  	rt_task_shadow(&main_tid, "main_task", 2, 0);
>>  #endif /* __NATIVE_SKIN__ */
>>
>>
>> # mutex-torture-posix
>> simple_wait
>> recursive_wait
>> errorcheck_wait
>> mode_switch
>> pi_wait
>> lock_stealing
>> lock_stealing mutex_trylock: not supported
>> simple_condwait
>> recursive_condwait
>> cond_signaler, mutex_lock 2 waited 0.854 us
> 
> No, that is a bug. We should have waited 10ms, and we wait less than a us.
> 

Ok. Should be fixed by commit af93ec87f975b387243127090b578d57922b38dc.
The bug was on the same goddamn pass as the one we are trying to fix to
return proper return values to user-space.

-- 
					    Gilles.

Re: [Xenomai-core] Posix mutex unit test fails on git head

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 2186 bytes --]

Gilles Chanteperdrix wrote:
> Gilles Chanteperdrix wrote:
>> Jan Kiszka wrote:
>>> Gilles Chanteperdrix wrote:
>>>> Jan Kiszka wrote:
>>>>> Jan Kiszka wrote:
>>>>>> Gilles,
>>>>>>
>>>>>> can you (or someone else) confirm this:
>>>>>>
>>>>>> # mutex-torture-posix
>>>>>> simple_wait
>>>>>> simple mutex_lock 1: 1 (Operation not permitted)
>>>>>>
>>>>>> I'm on a x86_64 target (2.6.32.8 with latest I-pipe), Xenomai is
>>>>>> unpatched git head.
>>>>>>
>>>>> Looks like a bug in the test case: We do not wrap sched_setscheduler
>>>>> which it uses, and my config does not auto-shadow main. Fix will follow.
>>>> The test works here, but fails in cond_signaler, mutex_lock 2, just
>>>> checked out v2.5.1 to see if we already had the bug.
>>>>
>>> Weird. Works for me with this fix:
>>>
>>> diff --git a/src/testsuite/unit/mutex-torture.c b/src/testsuite/unit/mutex-torture.c
>>> index 70bda14..ebdd4f2 100644
>>> --- a/src/testsuite/unit/mutex-torture.c
>>> +++ b/src/testsuite/unit/mutex-torture.c
>>> @@ -657,7 +657,7 @@ int main(void)
>>>  	/* Set scheduling parameters for the current process */
>>>  #ifdef XENO_POSIX
>>>  	sparam.sched_priority = 2;
>>> -	sched_setscheduler(0, SCHED_FIFO, &sparam);
>>> +	pthread_setschedparam(pthread_self(), SCHED_FIFO, &sparam);
>>>  #else /* __NATIVE_SKIN__ */
>>>  	rt_task_shadow(&main_tid, "main_task", 2, 0);
>>>  #endif /* __NATIVE_SKIN__ */
>>>
>>>
>>> # mutex-torture-posix
>>> simple_wait
>>> recursive_wait
>>> errorcheck_wait
>>> mode_switch
>>> pi_wait
>>> lock_stealing
>>> lock_stealing mutex_trylock: not supported
>>> simple_condwait
>>> recursive_condwait
>>> cond_signaler, mutex_lock 2 waited 0.854 us
>> No, that is a bug. We should have waited 10ms, and we wait less than a us.
>>
> 
> Ok. Should be fixed by commit af93ec87f975b387243127090b578d57922b38dc.
> The bug was on the same goddamn pass as the one we are trying to fix to
> return proper return values to user-space.
> 

Yes, looks good!

I've hacked up some additional (basic) test cases for times mutex and
cond waiting, and now I'm looking into the posix side of that infamous
workaround.

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 257 bytes --]