Important changes to forums and questions
All forums and questions are now archived. To start a new conversation or read the latest updates go to forums.mbed.com.
How to check for RAM available?
Vincent (pan-) Coubard.
43
replies
Has we finally sorted out this problem? I tested John's code today with the latest mbed compiler. It is still crash without return to NULL or causing hard_fault.
John Kemp wrote:
On a related note, if you add up the allocations and the statically used RAM, it crashes out once it hits about 22kB used.
I've recently run into a memory issue and while researching ideas I came across this topic. I compiled John's program to see if it can shed light on my crashes when dealing with strings as I'm crashing with string concatenations (assuming a bad malloc()).
John Kemp wrote:
On a related note, if you add up the allocations and the statically used RAM, it crashes out once it hits about 22kB used.
I used your program and modified a few parameters to see if only 22kB is used is the correct assumption. I tried a few different variable combinations and came up with this table.
*ptr int size char size index memory used 4 0 1 3432 17160 Busout removed *27456 corrected for 8 byte structure 4 0 1 3404 17020 *27232 4 0 0 3404 13616 *27232 4 4 0 3404 27232 4 4 14 1135 24970 4 4 28 687 24732 Busout removed 4 4 28 682 24552 4 4 40 568 27264 4 4 80 309 27192 4 4 120 212 27136 4 4 180 141 26508 4 4 190 135 26730 4 4 200 130 27040 Busout removed 4 4 200 129 26832 4 4 208 124 26784 4 4 210 120 26160 4 4 220 116 26448 4 4 260 99 26532 4 4 300 86 26488 4 4 400 65 26520
From what the table shows the more the while-loop runs the faster the heap and stack crash together.
Memory used = (*ptr+int+char)*index
It also makes since that the hard fault isn't triggers as from what I have read those are disabled with the online compiler. So I removed it from the code.
The final modified code is:
#include "mbed.h"
Serial pc(USBTX, USBRX);
void fail()
{
pc.printf("Caught the failure\r\n");
}
class Test
{
public:
int something;
char woot[200];
};
int main() {
pc.baud(921600);
Test *dummy;
int index = 0;
while(1) {
dummy = new Test;
if (dummy == NULL) fail();
pc.printf("%u\r\n", index++);
}
}
- Edit - Sort of off topic... but I did some testing and It seems the smallest structure is 8 bytes. If this is used an anomaly can be seen where 8 extra bytes go missing.
Empty Structure is 8 (ptr and NULL, 3 byte pad) Index Address malloc too big delta index 494 268440464 16 494 1006 268444568 16 512 1518 268448672 16 512 2030 268452776 16 512 2542 268456880 16 512 2991 268460480 16 449 3215 268462280 16 224 3327 268463184 16 112 3382 268463632 16 55 3409 268463856 16 27 3422 268463968 16 13 3428 268464024 16 6 3431 268464056 16 3
John Kemp wrote:
I've published a program here: http://mbed.org/users/Kemp/programs/memory_test/
Unless I messed up when typing it out, this demonstrates the crash on memory allocation (without returning NULL or causing a hard fault).
The test program does what is expected. operator new will never return NULL except when it is called with std::no_throw. So, in this case, when there is no memory left, new will call std::abort which will call HardFaulthandler.
So, if you want that new return NULL when there is no memory left, use the nothrow version of the operator.
nothrow usage
#include <new> // will return NULL if there is no memory left instead of calling std::abort Foo* foo = new(std::nothrow) Foo;
I've published a program here: http://mbed.org/users/Kemp/programs/memory_test/
Unless I messed up when typing it out, this demonstrates the crash on memory allocation (without returning NULL or causing a hard fault).