x86: avoid theoretical vmalloc fault loop
authorAndi Kleen <andi@firstfloor.org>
Fri, 9 Jan 2009 20:17:43 +0000 (12:17 -0800)
committerIngo Molnar <mingo@elte.hu>
Mon, 12 Jan 2009 18:24:21 +0000 (19:24 +0100)
commitf313e12308f7c5ea645f18e759d104d088b18615
tree4043dfa82448c1ef1bba81e58eb1fe162e82a69c
parentf45ac22ae2b8fc5b4c32d9b8d17ea419a8701d89
x86: avoid theoretical vmalloc fault loop

Ajith Kumar noticed:

 I was going through the vmalloc fault handling for x86_64 and am unclear
 about the following lines in the vmalloc_fault() function.

 pgd = pgd_offset(current->mm ?: &init_mm, address);
 pgd_ref = pgd_offset_k(address);

 Here the intention is to get the pgd corresponding to the current process
 and sync it up with the pgd in init_mm(obtained from pgd_offset_k).
 However, for kernel threads current->mm is NULL and hence pgd =
 pgd_offset(init_mm, address) = pgd_ref which means the fault handler
 returns without setting the pgd entry in the MM structure in the context
 of which the kernel thread has faulted.  This could lead to never-ending
 faults and busy looping of kernel threads like pdflush.  So, shouldn't the
 pgd = pgd_offset(current->mm ?: &init_mm, address); be pgd =
 pgd_offset(current->active_mm ?: &init_mm, address);

We can use active_mm unconditionally because it should be always set.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
arch/x86/mm/fault.c