Re: [PATCH v4 19/19] fs: handle inode->i_version more efficiently

[Jeff Layton]

On Mon, 2018-01-08 at 05:30 -0800, Matthew Wilcox wrote:
> On Fri, Dec 22, 2017 at 07:05:56AM -0500, Jeff Layton wrote:
> > +	cur = inode_peek_iversion_raw(inode);
> > +	for (;;) {
> > +		/* If flag is clear then we needn't do anything */
> > +		if (!force && !(cur & I_VERSION_QUERIED))
> > +			return false;
> > +		/* Since lowest bit is flag, add 2 to avoid it */
> > +		new = (cur & ~I_VERSION_QUERIED) + I_VERSION_INCREMENT;
> 
> Isn't this an extraordinarily complicated way of spelling:
> 
> 		new = cur + 1;
> 
> We know 'cur' has I_VERSION_QUERIED set, so clearing that bit and adding
> two is going to be the same as adding 1 ... right?
> 

It would be, but if "force" is true, then I_VERSION_QUERIED may not be
set.
-- 
Jeff Layton <jlayton@xxxxxxxxxx>