Re: [PATCH 07/12] mempolicy: mpol_shared_policy_init() without pseudo-vma

[Hugh Dickins]

On Mon, 25 Sep 2023, Matthew Wilcox wrote:
> On Mon, Sep 25, 2023 at 01:29:28AM -0700, Hugh Dickins wrote:
> > +		/* alloc node covering entire file; adds ref to new */
> 
> This comment is confusing.  sp_alloc initialises the refcount of 'n' to 1.
> Which is the same memory referred to by the name 'new' in __mpol_dup(),
> but in this function, the name "new" refers to the mempolicy called
> "old" in __mpol_dup().

No promises, but I'll see if I can make it look better in v2.

> 
> > +		n = sp_alloc(0, MAX_LFS_FILESIZE >> PAGE_SHIFT, new);
> > +		if (n)
> > +			sp_insert(sp, n);
> >  put_new:
> >  		mpol_put(new);			/* drop initial ref */
> >  free_scratch:
> 
> This is all a bit inefficient, really.  We call mpol_new() to get a
> new mpol, then we set it up, then we dup it, then we free it.  It'd
> be nice if we could donate it instead of copying it.  Maybe you'll
> do something like that later.

"later" is probably the operative word.  I do have an unincluded 2017
patch where I had that same realization, and wrote "I suspect that this
series of commits may be adding to an absurdity of over-mpol_dup()ing:
but that's for some other future cleanup, right now I'm just happy not
to be corrupting or leaking mpols."

Hugh