Messages in this thread | | | From | Mel Gorman <> | Subject | [PATCH 0/4] NUMA balancing related fixlets | Date | Thu, 2 Oct 2014 14:29:14 +0100 |
| |
The following is in reaction to the thread "pipe/page fault oddness". The first patch cleans up a lot of mess related to ARCH_USES_NUMA_PROT_NONE. It's already in Andrew's tree but included here for context.
The second patch removes a race between migration and mprotect, particularly from compaction context.
The third patch should have no real effect but it's an oversight that should be corrected in case MPOL_MF_LAZY is exposed to userspace.
The fourth patch addresses a problem whereby a split_huge_page can leave a pte_numa PTE in the wrong location and is one potential source of the bug Dave observed.
Note that none of this prevents a pte_protnone approach being applied on top or replacing patch 4 with it. They should not be in conflict but if patch 4 in particular addresses Dave's problem then it may be considerably lower risk for 3.17.
arch/powerpc/include/asm/pgtable.h | 57 ++++++++--------------------------- arch/powerpc/include/asm/pte-common.h | 5 +++ arch/x86/Kconfig | 1 - arch/x86/include/asm/pgtable_types.h | 14 +++++++++ include/asm-generic/pgtable.h | 27 ++++++----------- init/Kconfig | 11 ------- mm/huge_memory.c | 2 -- mm/mempolicy.c | 4 ++- mm/migrate.c | 4 ++- 9 files changed, 46 insertions(+), 79 deletions(-)
-- 1.8.4.5
| |