| Date | Thu, 11 Jun 2015 09:46:46 -0700 | From | "H. Peter Anvin" <> | Subject | Re: [RFC PATCH 00/12] x86/mm: Implement lockless pgd_alloc()/pgd_free() |
| |
On 06/11/2015 07:07 AM, Ingo Molnar wrote: > Waiman Long reported 'pgd_lock' contention on high CPU count systems and proposed > moving pgd_lock on a separate cacheline to eliminate false sharing and to reduce > some of the lock bouncing overhead. > > I think we can do much better: this series eliminates the pgd_list and makes > pgd_alloc()/pgd_free() lockless.
This ties into a slightly different issue, which is how to deal *properly* with "special" PGDs like the 1:1 trampoline and the UEFI page tables. These, too, should be able to incorporate, possibly more than once, page tables further down.
-hpa
|