Re: 'Change the vm_map lookup algorithm'

看板DFBSD_kernel作者時間15年前 (2010/10/04 07:32), 編輯推噓0(000)
留言0則, 0人參與, 最新討論串3/3 (看更多)
#1 on my list of things that can make the system faster now would be how to MP-lock the VM system. The problem the VM system has is that there is a billion different ways to drill down into it. Via the filesystem buffer cache, via VM objects, via the pmap (pmap->pte->vm_page), via pv_entry's in the pmap, via VM pages, via the vm_map... those are the main ones. So what we need to do is to come up with a way to MP lock the VM system that is more fine-grained than our current vm_token but which allows the above various methods to be implemented without having to acquire tons of different locks on the way down. This will bear a strong relationship to whatever indexing mechanic is chosen for vm_map's, and perhaps also vm_object's which currently use a RB tree as well (vm_object->rb_memq and vm_object->swblock_root). -Matt Matthew Dillon <dillon@backplane.com>
文章代碼(AID): #1CgH9s_q (DFBSD_kernel)
文章代碼(AID): #1CgH9s_q (DFBSD_kernel)