Re: [PATCH 1/2] xfs: fix xfs_mark_inode_dirty during umount

[Christoph Hellwig <hch@xxxxxxxxxxxxx>]

On Tue, Aug 30, 2011 at 05:20:13PM +1000, Dave Chinner wrote:
> Now that may have been true on Irix/MIPS which had strong memory
> ordering so only compiler barriers were necessary.
> 
> However, normally when we talk about ordered memory semantics in
> Linux, we cannot assume strong ordering - if we have ordering
> requirements, we have to guarantee ordering by explicit use of
> memory barriers, right?

Probably.  But I'm not worried about that so much, it's just timestamps
we're talking about as the size already has the ilock unlock as full
barrier, and we're going to kill this code soon anyway.