x86/mm/cpa: Optimize same protection check
authorThomas Gleixner <tglx@linutronix.de>
Mon, 17 Sep 2018 14:29:15 +0000 (16:29 +0200)
committerThomas Gleixner <tglx@linutronix.de>
Thu, 27 Sep 2018 18:39:39 +0000 (20:39 +0200)
When the existing mapping is correct and the new requested page protections
are the same as the existing ones, then further checks can be omitted and the
large page can be preserved. The slow path 4k wise check will not come up with
a different result.

Before:

 1G pages checked:                    2
 1G pages sameprot:                   0
 1G pages preserved:                  0
 2M pages checked:                  540
 2M pages sameprot:                 466
 2M pages preserved:                 47
 4K pages checked:               800709
 4K pages set-checked:             7668

After:

 1G pages checked:                    2
 1G pages sameprot:                   0
 1G pages preserved:                  0
 2M pages checked:                  538
 2M pages sameprot:                 466
 2M pages preserved:                 47
 4K pages checked:               560642
 4K pages set-checked:             7668

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Dave Hansen <dave.hansen@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Bin Yang <bin.yang@intel.com>
Cc: Mark Gross <mark.gross@intel.com>
Link: https://lkml.kernel.org/r/20180917143546.424477581@linutronix.de
arch/x86/mm/pageattr.c

index 19781b0..5160334 100644 (file)
@@ -840,6 +840,20 @@ static int __should_split_large_page(pte_t *kpte, unsigned long address,
        }
 
        /*
+        * Optimization: If the requested pgprot is the same as the current
+        * pgprot, then the large page can be preserved and no updates are
+        * required independent of alignment and length of the requested
+        * range. The above already established that the current pgprot is
+        * correct, which in consequence makes the requested pgprot correct
+        * as well if it is the same. The static protection scan below will
+        * not come to a different conclusion.
+        */
+       if (pgprot_val(req_prot) == pgprot_val(old_prot)) {
+               cpa_inc_lp_sameprot(level);
+               return 0;
+       }
+
+       /*
         * Make sure that the requested pgprot does not violate the static
         * protections. Check the full large page whether one of the pages
         * in it results in a different pgprot than the first one of the