On Sünndag 20 Februar 2005 21:47, Matt Mackall wrote: > I've been sitting on this for over a year now, kicking it out in the > hopes that someone finds it useful. kernel.org was down when I was > tidying this up so it's against 2.6.10 which is what I had handy. > > /proc/kmalloc allocation tracing
Nice. I have done something similar for the buddy allocator but never
got around to sending it.
> This quick hack adds accounting for kmalloc/kfree callers. This can
> aid in tracking down memory leaks and large dynamic memory users. The
> stock version use ~280k of memory for hash tables and can track 32k
> active allocations.
>
> Here's some sample output from my laptop:
>
> total bytes allocated: 47118848
> slack bytes allocated: 8717262
> net bytes allocated: 2825920
> number of allocs: 132796
> number of frees: 122629
> number of callers: 325
> lost callers: 0
> lost allocs: 0
> unknown frees: 0
>
> total slack net alloc/free caller
> 24576 0 0 3/3 copy_thread+0x1ad
The format is not really easy to parse, it probably makes sense to
split the two parts into separate files. I also think that debugfs
would be a more appropriate place to put this in than procfs.
> +void __kmalloc_account(const void *caller, const void *addr, int size, int
> req)
> +{
> + int i, hasha, hashc;
> + unsigned long flags;
> +
> + spin_lock_irqsave(&kma_lock, flags);
> + if(req >= 0) /* kmalloc */
> + {
> + /* find callers slot */
> + hashc = kma_hash(caller, MAX_CALLER_TABLE);
> + for (i = 0; i < MAX_CALLER_TABLE; i++) {
> + if (!kma_caller[hashc].caller ||
> + kma_caller[hashc].caller == caller)
> + break;
> + hashc = (hashc + 1) % MAX_CALLER_TABLE;
> + }
The housekeeping that is needed for the hash implementation is rather
complicated. The code that I wrote did a static allocation from inside
a macro, like
#define kmalloc(_size, _gfp) \
({ \
static struct kma_caller _caller \
__attribute__((section(".kmalloc.data"))) = { \
.func = __FUNCTION__, \
.line = __LINE__, \
}; \
_caller.count++; \
_caller.size += (_size); \
__kmalloc((_size), (_gfp)); \
})
Then I could simply print out all allocations by walking through the
special linker section. OTOH, your implementation has the advantage
that it can directly match kmalloc/kfree pairs and that it does not
rely on special linker magic.
Arnd <><
pgp01ROXTHZ22.pgp
Description: signature
