Re: [PATCH] mm/sl[aou]b: make kfree() aware of error pointers

[Valdis . Kletnieks]

On Tue, 09 Sep 2014 16:21:14 -0700, Andrew Morton said:
> On Tue, 9 Sep 2014 23:25:28 +0200 (CEST) Jiri Kosina <jkosina@xxxxxxx> wrote:
> kfree() is quite a hot path to which this will add overhead.  And we
> have (as far as we know) no code which will actually use this at
> present.

We already do a check for ZERO_SIZE_PTR, and given that dereferencing *that* is
instant death for the kernel, and we see it very rarely, I'm going to guess
that IS_ERR(ptr) *has* to be true more often than ZERO_SIZE_PTR, and thus even
more advantageous to short-circuit.

I guess it depends on a few things:

1) How many instances of 'if (!IS_ERR(foo)) kfree(foo);' are in the tree, and
what percent of kfree() calls executed have the guard on them

2) How many of the hot calls can/will get the guard removed.

3) How many cycles, if any, this adds to the path (a non-trivial question on
superscalar architectures), compared with doing a test before calling kfree()

I unfortunately have no earthly clue what the values of any of those
three quantities are....

Attachment: pgpndOyQxGwfi.pgp
Description: PGP signature