Dumb question. What is the best way to get the free memory available on the maple?
In avr-gcc you would do this.
extern int __bss_end;
extern void *__brkval;
int getFreeMemory()
{
int free_memory;
if((int)__brkval == 0)
free_memory = ((int)&free_memory) - ((int)&__bss_end);
else
free_memory = ((int)&free_memory) - ((int)__brkval);
return free_memory;
}
shark2600
Member
There is a FreeRam function in dinau's SdFatMaple-vld lib. It is under condition #if 0, I couldn't get it compiled, one symbol is missing. Hope this can help you.
https://bitbucket.org/dinau/sdfatmaple-vld/src/a386d1505010/SdFatUtil.h
/** Return the number of bytes currently free in RAM. */
#if 0
static int FreeRam(void) {
extern int __bss_end;
extern int* __brkval;
int free_memory;
if (reinterpret_cast<int>(__brkval) == 0) {
// if no heap use from end of bss section
free_memory = reinterpret_cast<int>(&free_memory)
- reinterpret_cast<int>(&__bss_end);
} else {
// use from top of stack to heap
free_memory = reinterpret_cast<int>(&free_memory)
- reinterpret_cast<int>(__brkval);
}
return free_memory;
}
#endif
The reason that doesnt compile is because its basically the same avr-gcc code as above.
I am playing around with something along these lines.
extern caddr_t _sbrk(int incr);
#ifndef CONFIG_HEAP_END
extern char _lm_heap_end;
#define CONFIG_HEAP_END ((caddr_t)&_lm_heap_end)
#endif
int getFreeMemory()
{
return CONFIG_HEAP_END - _sbrk(0);
}
If anyone has any better ideas that would be swell.
mbolivar
LeafLabs
libmaple's linker script will always define _lm_heap_end; however, on most boards, its value is the initial stack pointer (so the heap and stack overlap).
Probably the right thing to do is put it somewhere below the stack, so there's a minimum guaranteed stack size. Then, as long as you don't use too much stack, you could use _lm_heap_end safely to build a getFreeMemory() type function.
Would you care to prepare a patch? See support/ld/common.inc:
/*
* Heap: Linker scripts may choose a custom heap by overriding
* _lm_heap_start and _lm_heap_end. Otherwise, the heap is in
* internal SRAM, beginning after .bss, and growing towards
* the stack.
*
* I'm shoving these here naively; there's probably a cleaner way
* to go about this. [mbolivar]
*/
_lm_heap_start = DEFINED(_lm_heap_start) ? _lm_heap_start : _end;
_lm_heap_end = DEFINED(_lm_heap_end) ? _lm_heap_end : __msp_init;
mbolivar
LeafLabs
I was trying to debug some FreeRTOS code which allocates from the heap outside of this mechanism anyways.
Curious -- what do you mean? If the FreeRTOS code uses malloc(), then it should be using newlib's implementation, which uses our _sbrk() to request more space. Our _sbrk() respects _lm_heap_end, so you should be OK, unless I've misunderstood.
Actually the freertos that you are distributing uses heap_2 which does not use newlib, and i couldn't get heap_3 which would wrap newlib's implementation to work (which sucks because I am very fond of stdio and most of the ansi stuff). I will try to post some blog stuff about my experiences with "FreeRtos vs libmaple vs Newlib". As the toolchain moves forward it should get less painful one would hope. http://michaldemin.wordpress.com/2010/03/09/freertos-newlib-on-cortex-m3/ does a pretty good job of describing some of the issue. Although things that depend on how you compile newlib really don't generalize or integrate well.
chipaudette
Member
Has there been any progress on this? I used the information above to get a value returned by _sbrk, but I'm not sure if it is meaningful. Here's what I did to get it to return something at all...
#include "syscalls.c"
extern caddr_t _sbrk(int incr);
#ifndef CONFIG_HEAP_END
extern char _lm_heap_end;
#define CONFIG_HEAP_END ((caddr_t)&_lm_heap_end)
#endif
int getFreeMemory()
{
return CONFIG_HEAP_END - _sbrk(0);
}I did not edit my common.inc. It is stock from the 0.0.12 distrubtion. For my current program, calling SerialUSB.println(getFreeMemory()); returns a value of 14224. Is this really the number of bytes that I have free? Or, do I need to do the mod to the common.inc?
Thanks,
Chip
As I was revisiting sys calls for other things. I was able to resolve this in a similar way
#ifndef CONFIG_HEAP_START
extern char _lm_heap_start;
#define CONFIG_HEAP_START ((caddr_t)&_lm_heap_start)
#endif
#ifndef CONFIG_HEAP_END
extern char _lm_heap_end;
#define CONFIG_HEAP_END ((caddr_t)&_lm_heap_end)
#endif
unsigned long int getFreeMemory()
{ static long int *testMalloc;
unsigned long int freeMemory;
testMalloc = (long int *) malloc(10);
if (testMalloc==0) return 0;
freeMemory = (unsigned long int)CONFIG_HEAP_END - (unsigned long int)(testMalloc);
free(testMalloc);
return freeMemory;
}I am getting a similar number ~13K which makes sense given that there is 20k to work with.
This also flys in the face of the documentation that claims that you can't malloc.
Also I couldn't get anything meaningful out of _sbrk by itself my current solution assumes that newlib is pretty dumb about its actual allocation so it remains to me unsolved. So I am perplexed at best.....
unsigned long int getFreeMemory() // none of this works.
{ static long int *testMalloc;
unsigned long int freeMemory;
testMalloc = (long int *) malloc(10);
if (testMalloc==0) return 0;
// freeMemory = (unsigned long int)CONFIG_HEAP_END - (unsigned long int)(testMalloc);
freeMemory = (unsigned long int)(&freeMemory) - (unsigned long int)(testMalloc);
// freeMemory = (unsigned long int)CONFIG_HEAP_END - (unsigned long int) _sbrk(0);
// freeMemory = (long unsigned int) (&freeMemory) - (unsigned long int) _sbrk(0);
free(testMalloc);
return freeMemory;
}
You must log in to post.