x86/mm/pageattr: Add a PUD error unwinding path
authorBorislav Petkov <bp@suse.de>
Thu, 31 Oct 2013 16:25:05 +0000 (17:25 +0100)
committerMatt Fleming <matt.fleming@intel.com>
Sat, 2 Nov 2013 11:09:28 +0000 (11:09 +0000)
In case we encounter an error during the mapping of a region, we want to
unwind what we've established so far exactly the way we did the mapping.
This is the PUD part kept deliberately small for easier review.

Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
arch/x86/mm/pageattr.c

index 2a1308a..1cbdbbc 100644 (file)
@@ -666,6 +666,51 @@ static int split_large_page(pte_t *kpte, unsigned long address)
        return 0;
 }
 
+#define unmap_pmd_range(pud, start, pre_end)           do {} while (0)
+
+static void unmap_pud_range(pgd_t *pgd, unsigned long start, unsigned long end)
+{
+       pud_t *pud = pud_offset(pgd, start);
+
+       /*
+        * Not on a GB page boundary?
+        */
+       if (start & (PUD_SIZE - 1)) {
+               unsigned long next_page = (start + PUD_SIZE) & PUD_MASK;
+               unsigned long pre_end   = min_t(unsigned long, end, next_page);
+
+               unmap_pmd_range(pud, start, pre_end);
+
+               start = pre_end;
+               pud++;
+       }
+
+       /*
+        * Try to unmap in 1G chunks?
+        */
+       while (end - start >= PUD_SIZE) {
+
+               if (pud_large(*pud))
+                       pud_clear(pud);
+               else
+                       unmap_pmd_range(pud, start, start + PUD_SIZE);
+
+               start += PUD_SIZE;
+               pud++;
+       }
+
+       /*
+        * 2M leftovers?
+        */
+       if (start < end)
+               unmap_pmd_range(pud, start, end);
+
+       /*
+        * No need to try to free the PUD page because we'll free it in
+        * populate_pgd's error path
+        */
+}
+
 static int alloc_pte_page(pmd_t *pmd)
 {
        pte_t *pte = (pte_t *)get_zeroed_page(GFP_KERNEL | __GFP_NOTRACK);
@@ -883,9 +928,20 @@ static int populate_pgd(struct cpa_data *cpa, unsigned long addr)
        pgprot_val(pgprot) |=  pgprot_val(cpa->mask_set);
 
        ret = populate_pud(cpa, addr, pgd_entry, pgprot);
-       if (ret < 0)
-               return ret;
+       if (ret < 0) {
+               unmap_pud_range(pgd_entry, addr,
+                               addr + (cpa->numpages << PAGE_SHIFT));
 
+               if (allocd_pgd) {
+                       /*
+                        * If I allocated this PUD page, I can just as well
+                        * free it in this error path.
+                        */
+                       pgd_clear(pgd_entry);
+                       free_page((unsigned long)pud);
+               }
+               return ret;
+       }
        cpa->numpages = ret;
        return 0;
 }