Messages in this thread | | | Subject | Re: [PATCH] Export symbol ksize() | From | Matt Mackall <> | Date | Sun, 15 Feb 2009 19:28:46 -0600 |
| |
On Mon, 2009-02-16 at 09:21 +0800, Herbert Xu wrote: > On Sun, Feb 15, 2009 at 05:00:52PM -0800, Andrew Morton wrote: > > > > But kmem_cache_size() would tell you how much extra secret memory there > > is available after the object? > > > > How that gets along with redzoning is a bit of a mystery though. > > > > The whole concept is quite hacky and nasty, isn't it?. Does > > networking/crypto actually show any gain from pulling this stunt? > > I see no point in calling ksize on memory that's not kmalloced. > So no there is nothing to be gained from having kmem_cache_ksize. > > However, for kmalloced memory we're wasting hundreds of bytes > for the standard 1500 byte allocation without ksize which means > that we're doing reallocations (and sometimes copying) when it > isn't necessary.
Yeah. That sucks. We should probably stick in an skb-friendly slab size and see what happens on network benchmarks.
-- http://selenic.com : development and support for Mercurial and Linux
| |