Malloc tests

[Joel Sherrill]

On Fri, Jan 7, 2022 at 8:25 PM zack leung <zakthertemsdev at gmail.com> wrote:
>
> I think that the malloc tests is calculated differently than  alloc_size+ allocsize mod it looks like this
>  *alloc_size = (uintptr_t) next_block + HEAP_ALLOC_BONUS - alloc_begin;
> when i run the comparison  i get 8 (with heap alignment) and the function gives me 12.
>  is the heap alloc bonus part of the current block?

The allocation extra is intended to ensure the block is in even
"pages". The size
of a page is specified when the heap instance is created but it is
CPU_HEAP_ALIGMENT for this specific heap.

If CPU_HEAP_ALIGMENT is 16 and you malloc(13), then the block
returned to you is really 16 bytes in size and you should only expect 13 to
be there.

The "bonus" will be between 0 and CPU_HEAP_ALIGNMENT - 1.

But the allocated size adjusted up to the next even multiple of
CPU_HEAP_ALIGNMENT should be the usable size.

--joel

>
> On Thu, 6 Jan 2022 at 21:28, Joel Sherrill <joel at rtems.org> wrote:
>>
>> On Thu, Jan 6, 2022 at 2:55 PM Gedare Bloom <gedare at rtems.org> wrote:
>> >
>> > On Tue, Jan 4, 2022 at 6:10 PM zack leung <zakthertemsdev at gmail.com> wrote:
>> > >
>> > > Helllo  ,
>> > > I'm working on a patch for malloc_get_usable size right now so far i have
>> > > this test case for malloc, I just make sure that the value is null and i
>> > > just malloc an int and then  i make a call to the function malloc_usable
>> > > size and then i compare it like this.
>> > >
>> > > static void test_malloc_usablesize( void ){
>> > >    int * a = malloc(sizeof(int));
>> > >    rtems_test_assert((int) malloc_usable_size(a) == 12);
>> > >    free (a);
>> > >
>> > >    int * b = NULL;
>> > >    rtems_test_assert( 0 == malloc_usable_size(b));
>> > >    free(b);
>> > >
>> > >
>> > > }
>> > > Is there a good amount of storage to allocate? Also I heard someone made a
>> >
>> > I think that this test case is quite brittle. The "usable size" will
>> > depend on the implementation details of malloc, and the conditions of
>> > the heap when the allocation request gets made. I don't have better
>> > ideas however for how to test the correctness of the usable_size
>> > function. Maybe there are certain sequences of malloc/free calls that
>> > can be relied upon to give you deterministic sizes of allocations?
>>
>> My suggestion in private chat was that you can depend on the
>> malloc heap being initialized with CPU_HEAP_ALIGNMENT. That's
>> what that macro is specifically for. It is used in
>> include/rtems/score/wkspaceinit*.h
>> and has been since the dawn of time.
>>
>> Then the size for a valid pointer from malloc() should be between
>> the allocated size and the next number on a CPU_HEAP_ALIGNMENT
>> boundary. I think the exact number is actually something like this:
>>
>> expected = alloc_size;
>> mod = alloc_size % CPU_HEAP_ALIGMENT;
>> expected += mod;
>>
>> Adding a helper function in the test to compute the expected
>> size allocated and validate the return may be wise if multiple
>> size allocations are going to be tested.
>>
>>
>> > > formatter for rtems source files.  Can someone tell me how to use it and if
>> > > it is on the main branch?
>> >
>> > This was part of a gsoc project last year. We haven't quite made the
>> > switch over to it and the associated coding style I think, but you can
>> > find the code via
>> > https://devel.rtems.org/wiki/GSoC/2021#StudentsSummerofCodeTrackingTable
>> > the student was Meh Mbeh Ida Delphine
>> >
>> >
>> > > _______________________________________________
>> > > devel mailing list
>> > > devel at rtems.org
>> > > http://lists.rtems.org/mailman/listinfo/devel
>> > _______________________________________________
>> > devel mailing list
>> > devel at rtems.org
>> > http://lists.rtems.org/mailman/listinfo/devel