Messages in this thread | | | Date | Fri, 29 Feb 2008 11:57:16 -0800 (PST) | From | Christoph Lameter <> | Subject | Re: [PATCH] Implement slub fastpath in terms of freebase and freeoffset |
| |
On Fri, 29 Feb 2008, Mathieu Desnoyers wrote:
> > Then we do not need the page->end field anymore right? I will try > > to rediff your patch against current slab-mm and see how we can proceed > > from there. > > I guess you could replace page->end with the (expensive) > page_address(page) + 1 in new_slab. > > You would also have to change slab_address() to use > page_address(page). Since this is used by check_valid_pointer, I am not > sure it's a good idea to use this slow operation there. > > Or maybe you have a different alternative in mind ?
page->end was introduced to have a unique pointer for each slab. The counter that you introduces can take over that role. So we could use a NULL pointer again as the end of the objects? Or some constant?
I am going to revert the change that introduced page->end for 2.6.25 since it seems that the new fastpath does not need this. It was specific to the old fastpath.
| |