rtems 4.6.0pre5: memory leak after deleting a task used printf

[Till Straumann]

Joel Sherrill  wrote:
> Valery Pykhtin wrote:
> 
>> Hi all,
>>
>> It seems there is a memory leak when deleting a task used printf.
>>
>> I created the test:
>>
>> 1. Calculating the amount of free/used blocks in c heap using 
>> _Heap_Get_information.
>> 2. Starting new task
>> 3. Task does printf and selfdeletes
>> 4. Calculating the amount of free/used blocks in c heap using 
>> _Heap_Get_information.
>> 5. Comparing values before starting a task with values after.
>> 6. Goto point 1
> 
> 
> If newlib is allocating the memory implicitly, then the
> thread-exit delete extension in cpukit/libcsupport/src/newlibc.c
> should be doing or calling something that frees it.  Look at
> that code and see if it is missing something.
> 
> This wouldn't be the first time that newlib allocated memory
> that it's clean up routines did not account for.

Well, we don't call their cleanup routines in the first place
__wrapup_reent() / __reclaim_reent() are commented out...

I did a similar test under rtems-4.6 (newlib 1.11.0) and
I observe a memory leak if a task uses more than 3 file
descriptors since excess '__sglue' memory (allocated in
__sfmoreglue) is not released.

Seems that our newlibc support routines are quite dependent
on newlib internals and have probably not been revised
thoroughly in a while...

-- Till

> 
> 
> 
>> After each iteration the number of used blocks is continously 
>> increasing. To catch the bandit I created memory allocation tracer 
>> wich logs every _Heap_Allocate and _Heap_Free and (log include stack 
>> traces for each allocation) and marks those blocks that were not freed.
>>
>> The stack trace of the bandit:
>>
>> #0 __wrap__Heap_Allocate
>> #1 _Region_Allocate_segment
>> #2 rtems_region_get_segment
>> #3 malloc
>> #4 _malloc_r
>> #5 __smakebuf  at newlib/libc/stdio/makebuf.c:81
>> #6 __swsetup at newlib/libc/stdio/wsetup.c:63
>> #7 _vfprintf_r at newlib/libc/stdio/vfprintf.c:519
>> #8 vfprintf at newlib/libc/stdio/vfprintf.c:367
>> #9 printf at newlib/libc/stdio/printf.c:62
>> #10 TestTask
>>
>> where TestTask is my test task, __wrap__Heap_Allocate  - 
>> _Heap_Allocate wrapper.
>>
>> I can send the test I've used, it contains the memory allocation 
>> tracer and completly ready to run under debugger. My BSP is pc586. 
>> Note that the test uses builtin GCC functions __builtin_frame_address 
>> and __builtin_return_address.
>>
>> Tommorow I'm going to repeat the test with RTEMS 4.6.2
>>
>> Valery
>>
> 
>