sparc32: fix broken set_pte()
authorKirill A. Shutemov <kirill.shutemov@linux.intel.com>
Wed, 11 Feb 2015 23:25:35 +0000 (15:25 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 12 Feb 2015 01:06:01 +0000 (17:06 -0800)
commit4ecf886045152d2ddf98ae74e39f789868ac1f98
tree6cde1110b8bad8824ffc882640e89a1c4af3d68c
parent9fbc1f635fd0bd28cb32550211bf095753ac637a
sparc32: fix broken set_pte()

32-bit sparc uses swap instruction to implement set_pte().  It called
using GCC inline assembler.  But it misses the "memory" clobber to
indicate that pte value will be updated in memory.

As result GCC doesn't know that it cannot postpone pte pointer dereference
which occurs before set_pte() to post-set_pte() time.

It leads to real-world bugs -- [1]. In this situation we have code:

ptent = ptep_modify_prot_start(mm, addr, pte);
ptent = pte_modify(ptent, newprot);
...
ptep_modify_prot_commit(mm, addr, pte, ptent);

ptep_modify_prot_start() in sparc case is just 'pte' dereference plus
pte_clear().  pte_clear() calls broken set_pte().  GCC thinks it's valid
to dereference 'pte' again on pte_modify() and gets cleared pte.
ptep_modify_prot_commit() puts 'pteent' with pfn==0 back to page table,
which eventually leads to the crash.

[1] http://lkml.kernel.org/r/54C06B19.8060305@roeck-us.net

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reported-by: Guenter Roeck <linux@roeck-us.net>
Tested-by: Guenter Roeck <linux@roeck-us.net>
Cc: Paul Moore <pmoore@redhat.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: David Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
arch/sparc/include/asm/pgtable_32.h