On Wed, 2 Jan 2013, Christoph Lameter wrote: > On Wed, 2 Jan 2013, Hugh Dickins wrote: > > > Recent NUMA enhancements are not to blame: this dates back to 2.6.35, > > when commit e17f74af351c "mempolicy: don't call mpol_set_nodemask() > > when no_context" skipped mpol_parse_str()'s call to mpol_set_nodemask(), > > which used to initialize v.preferred_node, or set MPOL_F_LOCAL in flags. > > With slab poisoning, you can then rely on mpol_to_str() to set the bit > > for node 0x6b6b, probably in the next page above the caller's stack. > > Ugly. But 2.6.35 means that the patch was not included in several > enterprise linux releases. Thanks, that's some relief. I forgot to mention that a good test for whether your particular kernel (with who knows what additional patches applied) is affected, is to mount -o remount,mpol=local /dev/shm # which should be a tmpfs grep /dev/shm /proc/mounts If that says "mpol=prefer" then you're affected and need the fix; if it says "mpol=local" (like 2.6.34 or after this fix) then you're safe. (Conversely, setting "mpol=prefer" shows up as "mpol=local" after the, fix, since that's what prefer without a node specification amounts to.) > > > I don't understand why MPOL_LOCAL is described as a pseudo-policy: > > it's a reasonable policy which suffers from a confusing implementation > > in terms of MPOL_PREFERRED with MPOL_F_LOCAL. I believe this would be > > much more robust if MPOL_LOCAL were recognized in switch statements > > throughout, MPOL_F_LOCAL deleted, and MPOL_PREFERRED use the (possibly > > empty) nodes mask like everyone else, instead of its preferred_node > > variant (I presume an optimization from the days before MPOL_LOCAL). > > But that would take me too long to get right and fully tested. > > The current approaches to implementing NUMA scheduling are making > MPOL_LOCAL an explicit policy. See > https://patchwork.kernel.org/patch/1703641/. It's a good step in the right direction. > > Does that address the concerns? It makes no difference to this bug, and does not go far enough to remove all the MPOL_F_LOCAL MPOL_PREFERRED MPOL_LOCAL twistiness. Hugh -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>