[ofa-general] Re: [PATCH 08 of 11] anon-vma-rwsem

[Andrea Arcangeli]

On Thu, May 08, 2008 at 08:30:20AM +0300, Pekka Enberg wrote:
> On Thu, May 8, 2008 at 8:27 AM, Pekka Enberg <penberg at cs.helsinki.fi> wrote:
> > You might want to read carefully what Linus wrote:
> >
> >  > The one that already has a 4 byte padding thing on x86-64 just after the
> >  > spinlock? And that on 32-bit x86 (with less than 256 CPU's) would have two
> >  > bytes of padding if we didn't just make the spinlock type unconditionally
> >  > 32 bits rather than the 16 bits we actually _use_?
> >
> >  So you need to add the flag _after_ ->lock and _before_ ->head....
> 
> Oh should have taken my morning coffee first, before ->lock works
> obviously as well.

Sorry, Linus's right: I didn't realize the "after the spinlock" was
literally after the spinlock, I didn't see the 4 byte padding when I
read the code and put the flag:1 in. If put between ->lock and ->head
it doesn't take more memory on x86-64 as described literlly. So the
next would be to find another place like that in the address
space. Perhaps after the private_lock using the same trick or perhaps
the slab alignment won't actually alter the number of slabs per page
regardless.

I leave that to Christoph, he's surely better than me at doing this, I
give it up entirely and I consider my attempt to merge a total failure
and I strongly regret it.

On a side note the anon_vma will change to this when XPMEM support is
compiled in:

 struct anon_vma {
-	spinlock_t lock;	/* Serialize access to vma list */
+	atomic_t refcount;	/* vmas on the list */
+	struct rw_semaphore sem;/* Serialize access to vma list */
 	struct list_head head;	   /* List of private "related" vmas
	*/
 };

not sure if it'll grow in size or not after that but let's say it's
not a big deal.