Messages in this thread |  | | | Date | Mon, 26 Jan 2009 12:17:28 -0500 (EST) | | From | Christoph Lameter <> | | Subject | Re: [PATCH] SLUB: revert direct page allocator pass through |
| |
On Sat, 24 Jan 2009, Nick Piggin wrote:
> > Or if you prefer order-0. Do a single linked list like SLQB does. > > The fundamental issues I guess are that slab pages are kernel mapped, and > within a given slab, the zone and movability are irrelevant.
Right. There are multiple subsystems that all have the same requirements.
> Other ones which could be changed but could introduce regressions are > watermarks, buddy merging, and struct page error checking and setup.
Isnt it possible to defer that (queuing them (sigh)). A bitmap could be used to avoid queuing and may even allow fully concurrent allocations without locks. Use a counter to check watermarks once in a while.
Struct page error checking could only be enabled for debugging purposes.
> I brought all this up when it was discussed. Did you find any ways to > improve anything?
I played around with some bitmap concepts having basically a bit vector per MAX_ORDER page (plus giving a preferred order to each MAX_ORDER chunk in order to avoid fragmenation) but I had to abandon that at the time due to time constraints.
The bitmap / counter approach could get a way with testing/incrementing a percpu counter and flipping a bit. One of the reasons that I am quite pushy on the per cpu atomic ops vs. interrupts is that these would allow a page allocator fastpath w/o disabling interrupts.
> (I did make that patch to enable refcounting to be avoided FWIW, which > avoids a couple of atomic operations, but I don't think it brought > performance up too much, but I still intend to dust it off at some > point).
Well if the page stays with a refcount of one then we do not need to check the refcount at all but just push it in an out of some queue / bitmap or something.
|  |