Re: [PATCH] PCI: kirin: Fix of_node_put() issue in pcie-kirin

[Rob Herring]

On Wed, Nov 3, 2021 at 9:31 AM Bjorn Helgaas <helgaas@xxxxxxxxxx> wrote:
>
> [+to Mauro, author of code being changed,
> Rob for "of_pci_get_devfn()" naming question]
>
> On Wed, Nov 03, 2021 at 02:25:18AM -0400, Wan Jiabing wrote:
> > Fix following coccicheck warning:
> > ./drivers/pci/controller/dwc/pcie-kirin.c:414:2-34: WARNING: Function
> > for_each_available_child_of_node should have of_node_put() before return.
> >
> > Early exits from for_each_available_child_of_node should decrement the
> > node reference counter. Replace return by goto here.
> >
> > Signed-off-by: Wan Jiabing <wanjiabing@xxxxxxxx>
> > ---
> >  drivers/pci/controller/dwc/pcie-kirin.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/pci/controller/dwc/pcie-kirin.c b/drivers/pci/controller/dwc/pcie-kirin.c
> > index 06017e826832..23a2c076ce53 100644
> > --- a/drivers/pci/controller/dwc/pcie-kirin.c
> > +++ b/drivers/pci/controller/dwc/pcie-kirin.c
> > @@ -422,7 +422,8 @@ static int kirin_pcie_parse_port(struct kirin_pcie *pcie,
> >                       pcie->num_slots++;
> >                       if (pcie->num_slots > MAX_PCI_SLOTS) {
> >                               dev_err(dev, "Too many PCI slots!\n");
> > -                             return -EINVAL;
> > +                             ret = -EINVAL;
> > +                             goto put_node;
> >                       }
> >
> >                       ret = of_pci_get_devfn(child);
>
> This is a change to the code added here:
>   https://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci.git/commit/?id=31bd24f0cfe0
>
> This fix looks right to me; all the other early exits from the inner
> loop drop the "child" reference.
>
> But this is a nested loop and the *outer* loop also increments
> refcounts, and I don't see that outer loop reference on "parent" being
> dropped at all:
>
>     for_each_available_child_of_node(node, parent) {
>       for_each_available_child_of_node(parent, child) {
>         ...
>         if (error)
>           goto put_node;
>       }
>     }
>
>   put_node:
>     of_node_put(child);

Indeed. There should be a put on the parent.

This whole function is less than ideal as it assumes the board has 2
levels of PCI nodes. This might be another (like brcmstb thread) place
to use .add_bus(). But that's a problem for another day.

> The "of_pci_get_devfn()" immediately after is unrelated, but possibly
> a confusing name.  "Get" often suggests a reference count being
> increased, but that's not the case with of_pci_get_devfn().

I guess you have to know that only nodes are refcounted and property
functions never are.

There's also of_pci_get_max_link_speed() and of_get_pci_domain_nr().

Rob