Re: [PATCH] arm/xen: Fix refcount leak in xen_dt_guest_init

[Miaoqian Lin <linmq006@xxxxxxxxx>]

Hi,

On Fri, Mar 11, 2022 at 06:01:11PM -0800, Stefano Stabellini wrote:
> On Wed, 9 Mar 2022, Miaoqian Lin wrote:
> > The of_find_compatible_node() function returns a node pointer with
> > refcount incremented, We should use of_node_put() on it when done
> > Add the missing of_node_put() to release the refcount.
> > 
> > Fixes: 9b08aaa3199a ("ARM: XEN: Move xen_early_init() before efi_init()")
> > Signed-off-by: Miaoqian Lin <linmq006@xxxxxxxxx>
> 
> Thanks for the patch!
> 
> 
> > ---
> >  arch/arm/xen/enlighten.c | 1 +
> >  1 file changed, 1 insertion(+)
> > 
> > diff --git a/arch/arm/xen/enlighten.c b/arch/arm/xen/enlighten.c
> > index ec5b082f3de6..262f45f686b6 100644
> > --- a/arch/arm/xen/enlighten.c
> > +++ b/arch/arm/xen/enlighten.c
> > @@ -424,6 +424,7 @@ static void __init xen_dt_guest_init(void)
> >  
> >     if (of_address_to_resource(xen_node, GRANT_TABLE_INDEX, &res)) {
> >             pr_err("Xen grant table region is not found\n");
> > +           of_node_put(xen_node);
> >             return;
> >     }
> 
> This is adding a call to of_node_put on the error path. Shouldn't it
> be called also in the non-error path?

You're right. It should be called also in the non-error path.
I made a mistake.

> Also, there is another instance of of_address_to_resource being called
> in this file (in arch_xen_unpopulated_init), does it make sense to call
> of_node_put there too?

I think so, becase device node pointer np is a local variable.
So the reference it taken should be released in the scope.

I look into the whole codebase for this kind of usage pattern
($ret=of_find_compatible_node();of_address_to_resource($ret,_,_), 
$ret is a local variable), Most of them call of_node_put() when done. 
And document of of_find_compatible_node() also mentions
> Return: A node pointer with refcount incremented, use
> of_node_put() on it when done.

But I am not sure, Since I am unfamiliar with other code logic.
It better if the developers could double check. I found some 
similar cases in arch/arm.