On Fri, 3 Apr 2015, Drokin, Oleg wrote:
On Apr 2, 2015, at 6:18 AM, Julia Lawall wrote:
>> Julia, I wonder if you happen to have a bunch of other patches to get rid of the
rest of OBD_ALLOC and OBD_FREE stuff by any chance?
> I can generate them again, but I wasn't clear on what was wanted. I would
> really prefer something where it is explicit at the call site that an
> assignment is taking place. If we can have x = obd_alloc(...) and
> obd_free(x,...) (I don't have time to look up the exact arguments at the
> moment), then I can take care of that). I still think it is too bad that
> this code won't benefit from rules written for more generic memory
> allocation functions, but if the extra debugging facility provided by
> these functions is useful, then I guess it is reasonable to keep it.
Like I mentioned sometime last year - it's now pretty easy to replace the memleak
detection with other in-kernel mechanisms some of which are in fact even better
than what we have. And considering our mechanisms are totally broken now by the mixup of
wrapped vs nonwrapped allocation/freeing - there's no point in holding to it
remaining at all.
The only last bit of useful functionality left, I imagine, is the ability to redirect
to regular kmalloc or to vmalloc based on the allocation size (there's kvfree already
freeing part of it).
Other than that the wrappers could go away at any time now, I think.