#ifdef very expensive debug check in page fault path

This patch puts #ifdef CONFIG_DEBUG_VM around a check in vm_normal_page
that verifies that a pfn is valid.  This patch increases performance of the
page fault microbenchmark in lmbench by 13% and overall dbench performance
by 7% on s390x.  pfn_valid() is an expensive operation on s390 that needs a
high double digit amount of CPU cycles.  Nick Piggin suggested that
pfn_valid() involves an array lookup on systems with sparsemem, and
therefore is an expensive operation there too.

The check looks like a clear debug thing to me, it should never trigger on
regular kernels.  And if a pte is created for an invalid pfn, we'll find
out once the memory gets accessed later on anyway.  Please consider
inclusion of this patch into mm.

Signed-off-by: Carsten Otte <cotte@de.ibm.com>
Acked-by: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

diff --git a/mm/memory.c b/mm/memory.c
index 4bf0b6d0eb2a675bbd1083099448d54979dff304..6dd1cd88bfb67411767b28e0cdc11a35b52fc92a 100644 (file)
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -392,6 +392,7 @@ struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_
                        return NULL;
        }
 
+#ifdef CONFIG_DEBUG_VM
        /*
         * Add some anal sanity checks for now. Eventually,
         * we should just do "return pfn_to_page(pfn)", but
@@ -402,6 +403,7 @@ struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_
                print_bad_pte(vma, pte, addr);
                return NULL;
        }
+#endif
 
        /*
         * NOTE! We still have PageReserved() pages in the page