mm: fix pfn_t to page conversion in vm_insert_mixed
authorDan Williams <dan.j.williams@intel.com>
Tue, 26 Jan 2016 17:48:05 +0000 (09:48 -0800)
committerDan Williams <dan.j.williams@intel.com>
Sun, 31 Jan 2016 17:07:15 +0000 (09:07 -0800)
pfn_t_to_page() honors the flags in the pfn_t value to determine if a
pfn is backed by a page.  However, vm_insert_mixed() was originally
written to use pfn_valid() to make this determination.  To restore the
old/correct behavior, ignore the pfn_t flags in the !pfn_t_devmap() case
and fallback to trusting pfn_valid().

Fixes: 01c8f1c44b83 ("mm, dax, gpu: convert vm_insert_mixed to pfn_t")
Cc: Dave Hansen <dave@sr71.net>
Cc: David Airlie <airlied@linux.ie>
Reported-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Tested-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
mm/memory.c

index 30991f83d0bf54dc537f1927a10883aa5787dd97..93ce37989471174439d8973f19b12e776e218cc3 100644 (file)
@@ -1591,10 +1591,15 @@ int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
         * than insert_pfn).  If a zero_pfn were inserted into a VM_MIXEDMAP
         * without pte special, it would there be refcounted as a normal page.
         */
-       if (!HAVE_PTE_SPECIAL && pfn_t_valid(pfn)) {
+       if (!HAVE_PTE_SPECIAL && !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) {
                struct page *page;
 
-               page = pfn_t_to_page(pfn);
+               /*
+                * At this point we are committed to insert_page()
+                * regardless of whether the caller specified flags that
+                * result in pfn_t_has_page() == false.
+                */
+               page = pfn_to_page(pfn_t_to_pfn(pfn));
                return insert_page(vma, addr, page, vma->vm_page_prot);
        }
        return insert_pfn(vma, addr, pfn, vma->vm_page_prot);