Re: [PATCH] mtd: ck804xrom: fix missing pci device put in error paths

[James Bond]

Hi Miquèl,
Thanks for your feedback.
I have just rechecked this function and find that "pdev" currently is
already put
inside ck804xrom_cleanup, so my previous patch is meaningless...
The current calling order is like:

window->pdev = pci_dev_get(pdev);
...
    ck804xrom_cleanup(window)
             ->
                  pci_dev_put(window->pdev);
And sorry for bothering you.
Thanks,
James

On Thu, Aug 27, 2020 at 7:46 AM Miquel Raynal <miquel.raynal@xxxxxxxxxxx> wrote:
>
> Hi James,
>
> James Bond <jameslouisebond@xxxxxxxxx> wrote on Fri, 21 Aug 2020
> 02:05:36 -0500:
>
> > pci_dev_get increases the refcount of "pdev".
> > In the error paths, pci_dev_put should be called
> > to handle the "pdev" and decrease the corresponding refcount.
> >
> > Fixes: 90afffc8bd79 ("[MTD] [MAPS] Support for BIOS flash chips on the nvidia ck804 southbridge")
> > Signed-off-by: James Bond <jameslouisebond@xxxxxxxxx>
> > ---
> >  drivers/mtd/maps/ck804xrom.c | 3 +++
> >  1 file changed, 3 insertions(+)
> >
> > diff --git a/drivers/mtd/maps/ck804xrom.c b/drivers/mtd/maps/ck804xrom.c
> > index 460494212f6a..16af8b5ee653 100644
> > --- a/drivers/mtd/maps/ck804xrom.c
> > +++ b/drivers/mtd/maps/ck804xrom.c
> > @@ -195,6 +195,7 @@ static int __init ck804xrom_init_one(struct pci_dev *pdev,
> >       if (!window->virt) {
> >               printk(KERN_ERR MOD_NAME ": ioremap(%08lx, %08lx) failed\n",
> >                       window->phys, window->size);
> > +             pci_dev_put(pdev);
> >               goto out;
> >       }
> >
> > @@ -222,6 +223,7 @@ static int __init ck804xrom_init_one(struct pci_dev *pdev,
> >
> >               if (!map) {
> >                       printk(KERN_ERR MOD_NAME ": kmalloc failed");
> > +                     pci_dev_put(pdev);
> >                       goto out;
> >               }
> >               memset(map, 0, sizeof(*map));
> > @@ -295,6 +297,7 @@ static int __init ck804xrom_init_one(struct pci_dev *pdev,
> >               if (mtd_device_register(map->mtd, NULL, 0)) {
> >                       map_destroy(map->mtd);
> >                       map->mtd = NULL;
> > +                     pci_dev_put(pdev);
> >                       goto out;
> >               }
> >
>
> I suppose in these three cases, the window->maps list will be empty and
> you will end up returning -ENODEV and the bottom of the function? If
> yes, it woudl probably be better to move these pci_dev_put() calls to
> this location.
>
> Otherwise, it might bit interesting to clean up a little bit the error
> path and perhaps have a distinct success vs. failure path.
>
>
> Thanks,
> Miquèl