On 11/16/2015 02:16 PM, Suzuki K. Poulose wrote:

On 13/10/15 09:34, Catalin Marinas wrote:

quoted

On Mon, Oct 12, 2015 at 06:52:56PM +0300, Andrey Ryabinin wrote:

quoted

Andrey Ryabinin (3):
   arm64: move PGD_SIZE definition to pgalloc.h
   arm64: add KASAN support
   Documentation/features/KASAN: arm64 supports KASAN now

Linus Walleij (1):
   ARM64: kasan: print memory assignment

Patches queued for 4.4. Thanks.

Hi,

I get the following failure with KASAN + 16K_PAGES + 48BIT_VA, with 4.4-rc1:


arch/arm64/mm/kasan_init.c: In function ?kasan_early_init?:
include/linux/compiler.h:484:38: error: call to ?__compiletime_assert_95? declared with attribute error: BUILD_BUG_ON failed: !IS_ALIGNED(KASAN_SHADOW_END, PGDIR_SIZE)
  _compiletime_assert(condition, msg, __compiletime_assert_, __LINE__)
                                      ^
include/linux/compiler.h:467:4: note: in definition of macro ?__compiletime_assert?
    prefix ## suffix();    \
    ^
include/linux/compiler.h:484:2: note: in expansion of macro ?_compiletime_assert?
  _compiletime_assert(condition, msg, __compiletime_assert_, __LINE__)
  ^
include/linux/bug.h:50:37: note: in expansion of macro ?compiletime_assert?
 #define BUILD_BUG_ON_MSG(cond, msg) compiletime_assert(!(cond), msg)
                                     ^
include/linux/bug.h:74:2: note: in expansion of macro ?BUILD_BUG_ON_MSG?
  BUILD_BUG_ON_MSG(condition, "BUILD_BUG_ON failed: " #condition)
  ^
arch/arm64/mm/kasan_init.c:95:2: note: in expansion of macro ?BUILD_BUG_ON?
  BUILD_BUG_ON(!IS_ALIGNED(KASAN_SHADOW_END, PGDIR_SIZE));


---

The problem is that the PGDIR_SIZE is (1UL << 47) with 16K+48bit, which makes
the KASAN_SHADOW_END unaligned(which is aligned to (1UL << (48 - 3)) ). Is the
alignment really needed ? Thoughts on how best we could fix this ?

Yes, it's really needed, because some code relies on this (e.g.  clear_pgs() and kasan_init()).
But it should be possible to get rid of this requirement.

At first we need to rework clear_pgs().
The purpose of clear_pgs() is to remove kasan shadow from swapper_pg_dir.
So clear_pgs() should clear the top most kasan_zero_* entries from page tables.
Previously it was enough to clear PGDs, in case of 16K_PAGES + 48BIT_VA we probably need to clear PMDs


We also have to change following part of kasan_init()
...
	/*
	 * We are going to perform proper setup of shadow memory.
	 * At first we should unmap early shadow (clear_pgds() call bellow).
	 * However, instrumented code couldn't execute without shadow memory.
	 * tmp_pg_dir used to keep early shadow mapped until full shadow
	 * setup will be finished.
	 */
	memcpy(tmp_pg_dir, swapper_pg_dir, sizeof(tmp_pg_dir));


Besides tmp_pg_dir we will need one more temporary page table to store those entries
which later will be removed from swapper_pg_dir by clear_pgds().

Cheers
Suzuki