FreeRTOS: Memory Management

Kernel
1

paulromero wrote on Sunday, May 22, 2016:

Dear Forum:

Is is possible to configure FreeRTOS such that the heap. implementation
of prvPortMalloc() etc. can be used on separate heap and ram segments
the concurrently ?

More clearly, I would like to use the same memory
management algrorithm for two completey separate areas of memory.
One is a large ram segment and the other is the default heap segment.

I am almost certan the names of the allocation and free routines
must have distinct names and something needs to be done to point
the at the right segment during initialization.

Best Regards,

Paul R.

2

richard_damon wrote on Sunday, May 22, 2016:

You probably want to make a copy of the heap#.c file and give the functions a new name to make the version to handle the separate area. You can then add whatever is needed to force the heap array into the right memory, this is very system dependent.

3

heinbali01 wrote on Monday, May 23, 2016:

Do you want two different malloc routines, e.g.:

    /* Malloc internal RAM. */
    void *pvPortMalloc();
    void vPortFree();

    /* Malloc external RAM. */
    void *pvPortMalloc_2();
    void vPortFree_2();

Or do you want a single malloc/free couple that manages both areas of memory?
In the last case you can use heap_5.c

4

rtel wrote on Monday, May 23, 2016:

Further to Hein’s post, here is a link with a little more information on
heap_5: http://www.freertos.org/a00111.html#heap_5

Regards.

5

paulromero wrote on Wednesday, May 25, 2016:

Hi Richard:

I would like to use two different malloc() routines
with independent memory management.

I designed a best fit memory management algorithm
which has been extensively tested and and moderately
analyzed. It is pretty much system independent. All
you need to do is point at the chunk of memory you
want to use. However, it is not reentrant and the
code could use some refinement.

The code that implements the algorithm and the proof of
it are attached to this note for your perusal. Can
you think of a way to make it safe for use in a
multitasking environment without using mutexes ?
I assume no one is dumb enough to try to use in
an interrupt handler or critical section.

Best Regards,

Paul R.

6

richard_damon wrote on Wednesday, May 25, 2016:

The simplest way to get mutual exclusion is a mutex, that is the basic purpose of them (which is what gives them their name). The alternative would be to do something to disable task switching for the duration, either with a critical section (which may cause unexceptable delays to interrupts) or to disable the scheduler, which will perhaps hold off a high priority task needlessly.

What is your problem with using a mutex?

7

paulromero wrote on Wednesday, May 25, 2016:

Hi Richard:

I have no problem, and actually like, using mutexes. However, some folks
complain that they degrade performance.

Best Regards,

Paul R.

8

rtel wrote on Wednesday, May 25, 2016:

On the assumption the allocation is going to be fast, and taking your
point that nobody should be doing this in an interrupt (although you
would be surprised what people do ;o) I would second Richard D’s
suggestion of suspending the scheduler.

vTaskSuspendAll();
// Do you thing here
xTaskResumeAll();

It might be that you do not need to do this around the entire allocation
function, but just at select points within the algorithm where
simultaneous access would be an issue.