Re: [RFC][PATCH] Check write to slab memory which freed already using mudflap

[Nick Piggin <npiggin@suse.de>]

On Fri, Jul 10, 2009 at 01:10:11PM +0300, Pekka Enberg wrote:
> Hi David,
> 
> On Fri, Jul 10, 2009 at 1:03 PM, David Rientjes<rientjes@google.com> wrote:
> > It's my opinion that slab is on its way out when there's no benchmark that
> > shows it is superior by any significant amount.  If that happens (and if
> > its successor is slub, slqb, or a yet to be implemented allocator), we can
> > probably start a discussion on what's in and what's out at that time.
> 
> Performance matters a lot, but it's certainly not the only priority
> here. Look at slab, it's bootstrap code is a train-wreck which bit us
> in the ass when we did the earlyslab thing and the NUMA support is
> pretty horrible. The code base hasn't received much attention for the
> past few years so unless someone steps up to clean it all, it's on
> it's way out, like it or not.
> 
> So if you care about performance and have benchmarks that are _known_
> to regress, you might want to focus your efforts on SLUB and/or SLQB
> because that's where the action happens these days.

Well OTOH performance is something that has some pretty fixed limits
by the design. If SLQB has any performance problems I can't fix, I
do intend to try to take the SLAB base allocator design and implement
it with more modern SL[UQ]B coding style and bootstrap code.

The reason I made some changes with SLQB is because I was trying to
remove nr_cpus*nr_nodes*lots allocations for the alien caches, and
I think the linked list style of queueing allows you to be more flexible
in queue sizes, and more cache efficient (especially when moving them
around).

We'll see...