Re: [RFC PATCH 0/3] Add support for compound page migration in mm_iommu_get

On 09/03/2018 10:07 PM, Aneesh Kumar K.V wrote:

This patch series add support for migrating compound pages if we find them in the
CMA area before taking long term page reference for VFIO.

We now call lru_add_drain_all instead of lru_add_drain() which means we 
now have higher chances of isolate_lru_page succeeding. The patch also 
migrate all the pages in one call, instead of one page at a time.

Testing:
* TODO: test with hugetlb backed guest ram.
* Testing done with a code change as below

-		if (is_migrate_cma_page(pages[i]) && migrate_allow) {
+		if (migrate_allow) {

...
+	migrate_allow = false;


Aneesh Kumar K.V (3):
   mm: Export alloc_migrate_huge_page
   powerpc/mm/iommu: Allow large IOMMU page size only for hugetlb backing
   powerpc/mm/iommu: Allow migration of cma allocated pages during
     mm_iommu_get

  arch/powerpc/mm/mmu_context_iommu.c | 209 +++++++++++++++++-----------
  include/linux/hugetlb.h             |   2 +
  mm/hugetlb.c                        |   4 +-
  3 files changed, 128 insertions(+), 87 deletions(-)