Re: [PATCH 01/15] kmemleak: Add the base support

[Catalin Marinas]

On Mon, 2008-12-01 at 08:49 +0200, Pekka Enberg wrote:
> On Mon, Dec 1, 2008 at 7:39 AM, Yasunori Goto <y-goto@xxxxxxxxxxxxxx> wrote:
> >> +/*
> >> + * Insert a pointer into the pointer hash table.
> >> + */
> >> +static inline void create_object(unsigned long ptr, size_t size, int ref_count)
> >> +{
> >> +     unsigned long flags;
> >> +     struct memleak_object *object;
> >> +     struct prio_tree_node *node;
> >> +     struct stack_trace trace;
> >> +
> >> +     object = kmem_cache_alloc(object_cache, GFP_ATOMIC);
> >> +     if (!object)
> >> +             panic("kmemleak: cannot allocate a memleak_object structure\n");
> >
> > IIRC, GFP_ATOMIC allocation sometimes fails. (ex. when page cache occupies all
> > area). It seems to be easy to panic. Is it intended?
> 
> Yup, GFP_ATOMIC can fail as can any memory allocation on out-of-memory
> conditions unless you specify GFP_NOFAIL which will either succeed or
> lock up the box. I think you can just WARN_ON() here? However, it's
> probably safer to pass gfp flags from the callers here; otherwise we
> end up doing tons of GFP_ATOMIC allocations which is not healthy in
> general.

Yes, good idea, GFP_ATOMIC isn't always needed.

> Also, I see some other BUG_ON() calls in the code which probably
> should be converted to WARN_ON() as well.

This was also raised in the past by Ingo and it's on my to-do list to
fix. Basically, if an allocation for internal kmemleak structures fails,
kmemleak should print a warning and disable the tracing since further
leak reports are irrelevant. Once I add the run-time kmemleak
configuration/disabling, I'll change the BUG_ON and panic calls to
something less severe.

-- 
Catalin

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/