OSDN Git Service

x86/mm: Disable 1GB direct mappings when disabling 2MB mappings
authorVlastimil Babka <vbabka@suse.cz>
Mon, 12 Jun 2017 07:21:30 +0000 (09:21 +0200)
committerIngo Molnar <mingo@kernel.org>
Tue, 13 Jun 2017 06:33:00 +0000 (08:33 +0200)
The kmemleak and debug_pagealloc features both disable using huge pages for
direct mappings so they can do cpa() on page level granularity in any context.

However they only do that for 2MB pages, which means 1GB pages can still be
used if the CPU supports it, unless disabled by a boot param, which is
non-obvious. Disable also 1GB pages when disabling 2MB pages.

Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vegard Nossum <vegardno@ifi.uio.no>
Cc: linux-mm@kvack.org
Link: http://lkml.kernel.org/r/2be70c78-6130-855d-3dfa-d87bd1dd4fda@suse.cz
Signed-off-by: Ingo Molnar <mingo@kernel.org>
arch/x86/mm/init.c

index cbc87ea..9b3f9fa 100644 (file)
@@ -161,16 +161,16 @@ static int page_size_mask;
 
 static void __init probe_page_size_mask(void)
 {
-#if !defined(CONFIG_KMEMCHECK)
        /*
         * For CONFIG_KMEMCHECK or pagealloc debugging, identity mapping will
         * use small pages.
         * This will simplify cpa(), which otherwise needs to support splitting
         * large pages into small in interrupt context, etc.
         */
-       if (boot_cpu_has(X86_FEATURE_PSE) && !debug_pagealloc_enabled())
+       if (boot_cpu_has(X86_FEATURE_PSE) && !debug_pagealloc_enabled() && !IS_ENABLED(CONFIG_KMEMCHECK))
                page_size_mask |= 1 << PG_LEVEL_2M;
-#endif
+       else
+               direct_gbpages = 0;
 
        /* Enable PSE if available */
        if (boot_cpu_has(X86_FEATURE_PSE))