Re: [PATCH] mtd: partitions: fix of_node_get/put balance in parser

[Miquel Raynal <miquel.raynal@bootlin.com>]

Hi Boris,

Boris Brezillon <boris.brezillon@bootlin.com> wrote on Mon, 10 Sep 2018
14:53:12 +0200:

> Hi Miquel,
> 
> On Fri, 7 Sep 2018 16:38:24 +0200
> Miquel Raynal <miquel.raynal@bootlin.com> wrote:
> 
> > Hello,
> > 
> > I forgot to add Rafal which I know worked a lot on the parsers.
> > 
> > Miquel Raynal <miquel.raynal@bootlin.com> wrote on Fri,  7 Sep 2018
> > 16:35:54 +0200:
> >   
> > > While at first mtd_part_of_parse() would just call
> > > of_get_chil_by_name(), it has been edited to first try to get the OF
> > > node thanks to mtd_get_of_node() and fallback on
> > > of_get_child_by_name().
> > > 
> > > A of_node_put() was a bit below in the code, to balance the
> > > of_get_child_by_name(). However, despite its name, mtd_get_of_node()
> > > does not take a reference on the OF node.  
> 
> That's probably something we should patch at some point, but that
> implies patching all mtd_get_of_node() users at the same time, so let's
> keep that for later.
> 
> BTW, if mtd_get_of_node() was actually retaining a reference, you
> would miss an of_node_put() in the !mtd_is_partition(master) case.

I think there is a misunderstanding here: mtd_get_of_node() is not
retaining a reference, and I do not think it should! It is by design a
helper to shortcut from the MTD device to the related FW node. Maybe
calling it differently than "get" would be definitely less prone to
errors. Maybe mtd_to_of_node() would be better?

> 
> > > It is a simple helper hiding
> > > some pointer logic to retrieve the OF node related to an MTD
> > > device. People often used it this way:
> > > 
> > >     of_node_put(mtd_get_of_node(<mtd>)).  
> 
> I don't get your point. Are you saying other places in the code are
> doing the wrong thing? Should we fix them too?

No, other places are doing the right thing. I think if the helper was
named "mtd_to_of_node()" that would be much clearer for everyone and
of_node_get(mtd_to_of_node(mtd)) would be the way to retain a reference
on the OF node.

I don't think creating a helper for that would be better because I
really prefer seeing the of_node_get() in the code, meaning an
of_node_put() will be needed at some point.

> 
> > > 
> > > The direct effect of such unbalanced reference counting is visible by
> > > rmmod'ing any module that would have added MTD partitions:
> > > 
> > >     OF: ERROR: Bad of_node_put() on <of_path_to_partition>
> > > 
> > > As it seems normal to get a reference on the OF node during the
> > > of_property_for_each_string() that follows, add a call to
> > > of_node_get() when relevant.
> > > 
> > > Fixes: 76a832254ab0 ("mtd: partitions: use DT info for parsing partitions with "compatible" prop")
> > > Cc: stable@vger.kernel.org  
> 
> You can drop the above line since this patch is fixing a bug introduced
> in 4.19-rc1.

Right!

> 
> > > Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
> > > ---
> > >  drivers/mtd/mtdpart.c | 5 ++++-
> > >  1 file changed, 4 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/drivers/mtd/mtdpart.c b/drivers/mtd/mtdpart.c
> > > index 52e2cb35fc79..99c460facd5e 100644
> > > --- a/drivers/mtd/mtdpart.c
> > > +++ b/drivers/mtd/mtdpart.c
> > > @@ -873,8 +873,11 @@ static int mtd_part_of_parse(struct mtd_info *master,
> > >  	int ret, err = 0;
> > >  
> > >  	np = mtd_get_of_node(master);
> > > -	if (!mtd_is_partition(master))
> > > +	if (mtd_is_partition(master))
> > > +		of_node_get(np);
> > > +	else
> > >  		np = of_get_child_by_name(np, "partitions");
> > > +
> > >  	of_property_for_each_string(np, "compatible", prop, compat) {
> > >  		parser = mtd_part_get_compatible_parser(compat);
> > >  		if (!parser)    
> 
> The patch itself looks correct, but I'd like some clarification about
> the commit message before applying it.
> 
> Thanks,
> 
> Boris
> 

Thanks,
Miquèl