Re: pagefault_disable (was Re: [patch 6/6] mm: fix pagecache write deadlocks)

[Nick Piggin <nickpiggin@yahoo.com.au>]

Peter Zijlstra wrote:
> On Tue, 2006-10-17 at 01:24 +1000, Nick Piggin wrote:

>>Also, the rest of the kernel tree (mainly uaccess and futexes) should be
>>converted ;)
> 
> 
> Yeah, lotsa places to touch.
> 
> 
>>>Index: linux-2.6/mm/filemap.h
>>>===================================================================
>>>--- linux-2.6.orig/mm/filemap.h	2006-10-14 20:20:20.000000000 +0200
>>>+++ linux-2.6/mm/filemap.h	2006-10-15 17:17:45.000000000 +0200
>>>@@ -21,6 +21,22 @@ __filemap_copy_from_user_iovec_inatomic(
>>> 					size_t bytes);
>>> 
>>> /*
>>>+ * By increasing the preempt_count we make sure the arch preempt
>>>+ * handler bails out early, before taking any locks, so that the copy
>>>+ * operation gets terminated early.
>>>+ */
>>>+pagefault_static inline void disable(void)
>>>+{
>>>+	inc_preempt_count();
> 
> 
> I think we also need a barrier(); here. We need to make sure the preempt
> count is written to memory before we hit the fault handler.

It will come from this thread, but I guess the fault is not an event the
compiler can forsee, so indeed it might optimise this into the wrong place.
Perhaps not with any copy*user implementation we have, but at least in
theory...

>>>+pagefault_static inline void enable(void)
>>>+{
>>>+	dec_preempt_count();
>>>+	preempt_check_resched();
>>>+}

You'll want barriers before and after the dec_preempt_count, for similar
reasons.

-- 
SUSE Labs, Novell Inc.
Send instant messages to your online friends http://au.messenger.yahoo.com