radix-tree: fix radix_tree_replace_slot

I was waiting to hear from Nick on this, but I understand that he has severely injured one hand, restricting his keyboard access for a while.

Christoph has been too swamped to take a look, either. So, having
tested it myself, I'll send it on.

Paul:

could you take a look at this one. Also, I'll be sending another rcu-radix-tree "cleanup" patch that I'd like your opinion on.

Lee

Fix radix tree direct slot replacement - 2.6.18-rc4-mm2

radix_tree_replace_slot() was assigning to local variable 'slot' instead of to where pslot pointed. Changed to directly replace location pointed to by argument pslot.

Added comments specifying required locking.

Note that we do not need to rcu_dereference() the slot to obtain the direct pointer flag, as we hold the tree write locked.

Fixes the migration corruption that we were seeing since the rcu-radix-tree patches went in. With this patch, we can back out page-migration-replace-radix_tree_lookup_slot-with-radix_tree_lockup.patch to use the more efficient direct access to radix tree slot.

 include/linux/radix-tree.h |    5 ++++-
 1 files changed, 4 insertions(+), 1 deletion(-)

Index: linux-2.6.18-rc4-mm2/include/linux/radix-tree.h
===================================================================
@@ -133,12 +133,15 @@ static inline void *radix_tree_deref_slo
  * radix_tree_replace_slot	- replace item in a slot
  * @pslot:	pointer to slot, returned by radix_tree_lookup_slot
  * @item:	new item to store in the slot.
+ *
+ * For use with radix_tree_lookup_slot().  Caller must hold tree write locked
+ * across slot lookup and replacement.
  */
 static inline void radix_tree_replace_slot(void *pslot, void *item)
 {
 	void *slot = *(void **)pslot;
 	BUG_ON(radix_tree_is_direct_ptr(item));
-	rcu_assign_pointer(slot,
+	rcu_assign_pointer(*(void **)pslot,
 		(void *)((unsigned long)item |
 			((unsigned long)slot & RADIX_TREE_DIRECT_PTR)));
 }

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Would it not be better to change the calling conventions of radix_tree_replace_slot? It should get passsed a void **pslot right?

static inline void radix_tree_replace_slot(void **pslot, void *item)

{
	BUG_ON(radix_tree_is_direct_ptr(item));
	rcu_assign_pointer(*pslot,
		(void *)((unsigned long)item |
			((unsigned long)*pslot & RADIX_TREE_DIRECT_PTR)));
}

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

I did consider that, and I looked at where the value of pslot came from [radix_tree_lookup_slot()] and all of the casts and other uses of that value. I decided to limit the change to the one line that I submitted. I don't actually recall my reasoning now. Probably to maintain the symmetry of _deref_slot() and _replace_slot().

Your suggestion certainly looks cleaner. Could/should also change _deref_slot()'s arg to 'void **pslot' and save a cast?

Lee

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Yeah sorry Lee, ping times have gone up and throughput down.

This does look like the right fix to me. Thanks very much for tracking it down. Both these patches are

BTW. I'll have to do a little rework on the radix tree to get the lockless pagecache patches working properly after the direct-data patches (based on review feedback)... that will take me a while :(

In the meantime, RCU radix tree isn't getting any lock-free testing because everything still takes tree_lock... I wonder if Andrew would be brave enough to lift the tree_lock from mm/readahead.c ?

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

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Color me blind -- I certainly missed this one earlier.

Yes, the original is certainly not very functional...

Thanx, Paul