mm/hmm: add missing call to hmm_pte_need_fault in HMM_PFN_SPECIAL handling
authorJason Gunthorpe <jgg@mellanox.com>
Thu, 5 Mar 2020 18:27:20 +0000 (14:27 -0400)
committerJason Gunthorpe <jgg@mellanox.com>
Thu, 26 Mar 2020 17:33:37 +0000 (14:33 -0300)
Currently if a special PTE is encountered hmm_range_fault() immediately
returns EFAULT and sets the HMM_PFN_SPECIAL error output (which nothing
uses).

EFAULT should only be returned after testing with hmm_pte_need_fault().

Also pte_devmap() and pte_special() are exclusive, and there is no need to
check IS_ENABLED, pte_special() is stubbed out to return false on
unsupported architectures.

Fixes: 992de9a8b751 ("mm/hmm: allow to mirror vma of a file on a DAX backed filesystem")
Reviewed-by: Ralph Campbell <rcampbell@nvidia.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
mm/hmm.c

index 3a03fcf..9c82ea9 100644 (file)
--- a/mm/hmm.c
+++ b/mm/hmm.c
@@ -339,16 +339,21 @@ static int hmm_vma_handle_pte(struct mm_walk *walk, unsigned long addr,
                        pte_unmap(ptep);
                        return -EBUSY;
                }
-       } else if (IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL) && pte_special(pte)) {
-               if (!is_zero_pfn(pte_pfn(pte))) {
+       }
+
+       /*
+        * Since each architecture defines a struct page for the zero page, just
+        * fall through and treat it like a normal page.
+        */
+       if (pte_special(pte) && !is_zero_pfn(pte_pfn(pte))) {
+               hmm_pte_need_fault(hmm_vma_walk, orig_pfn, 0, &fault,
+                                  &write_fault);
+               if (fault || write_fault) {
                        pte_unmap(ptep);
-                       *pfn = range->values[HMM_PFN_SPECIAL];
                        return -EFAULT;
                }
-               /*
-                * Since each architecture defines a struct page for the zero
-                * page, just fall through and treat it like a normal page.
-                */
+               *pfn = range->values[HMM_PFN_SPECIAL];
+               return 0;
        }
 
        *pfn = hmm_device_entry_from_pfn(range, pte_pfn(pte)) | cpu_flags;