Re: [PATCH 1/2] mm: Make follow_pte_pmd an inline

[Matthew Wilcox]

On Thu, Dec 21, 2017 at 02:29:43PM -0700, Ross Zwisler wrote:
> On Tue, Dec 19, 2017 at 08:58:22AM -0800, Matthew Wilcox wrote:
> > From: Matthew Wilcox <mawilcox@xxxxxxxxxxxxx>
> > 
> > The one user of follow_pte_pmd (dax) emits a sparse warning because
> > it doesn't know that follow_pte_pmd conditionally returns with the
> > pte/pmd locked.  The required annotation is already there; it's just
> > in the wrong file.
> 
> Can you help me find the required annotation that is already there but in the
> wrong file?

You cut it out ... that was the entire contents of the patch!
The cond_lock annotation is correct, but sparse doesn't look across
compilation units, so it can't see the one that's in mm/memory.c when
it's compiling fs/dax.c.  That's why it needs to be in a header file.

> This does seem to quiet a lockep warning in fs/dax.c, but I think we still
> have a related one in mm/memory.c:
> 
> mm/memory.c:4204:5: warning: context imbalance in '__follow_pte_pmd' - different lock contexts for basic block
> 
> Should we deal with this one as well?

I'm not sure how to deal with that one, to be honest.