From: Ralph Campbell rcampbell@nvidia.com

Private ZONE_DEVICE pages use a special pte entry and thus are not present. Properly handle this case in map_pte(), it is already handled in check_pte(), the map_pte() part was lost in some rebase most probably.

Without this patch the slow migration path can not migrate back to any private ZONE_DEVICE memory to regular memory. This was found after stress testing migration back to system memory. This ultimatly can lead to the CPU constantly page fault looping on the special swap entry.

Changes since v2: - add comments explaining what is going on Changes since v1: - properly lock pte directory in map_pte()

Signed-off-by: Ralph Campbell rcampbell@nvidia.com Signed-off-by: Jérôme Glisse jglisse@redhat.com Reviewed-by: Balbir Singh bsingharora@gmail.com Cc: Andrew Morton akpm@linux-foundation.org Cc: Kirill A. Shutemov kirill.shutemov@linux.intel.com Cc: stable@vger.kernel.org --- mm/page_vma_mapped.c | 24 +++++++++++++++++++++++- 1 file changed, 23 insertions(+), 1 deletion(-)

diff --git a/mm/page_vma_mapped.c b/mm/page_vma_mapped.c index ae3c2a35d61b..11df03e71288 100644 --- a/mm/page_vma_mapped.c +++ b/mm/page_vma_mapped.c @@ -21,7 +21,29 @@ static bool map_pte(struct page_vma_mapped_walk *pvmw) if (!is_swap_pte(*pvmw->pte)) return false; } else { - if (!pte_present(*pvmw->pte)) + /* + * We get here when we are trying to unmap a private + * device page from the process address space. Such + * page is not CPU accessible and thus is mapped as + * a special swap entry, nonetheless it still does + * count as a valid regular mapping for the page (and + * is accounted as such in page maps count). + * + * So handle this special case as if it was a normal + * page mapping ie lock CPU page table and returns + * true. + * + * For more details on device private memory see HMM + * (include/linux/hmm.h or mm/hmm.c). + */ + if (is_swap_pte(*pvmw->pte)) { + swp_entry_t entry; + + /* Handle un-addressable ZONE_DEVICE memory */ + entry = pte_to_swp_entry(*pvmw->pte); + if (!is_device_private_entry(entry)) + return false; + } else if (!pte_present(*pvmw->pte)) return false; } }