Re: [PATCH 1/3] msi: add forgotten pci_dev_put(pdev) to populate_msi_sysfs()

[Veaceslav Falico <vfalico@redhat.com>]

On Wed, Sep 25, 2013 at 05:35:54PM -0600, Bjorn Helgaas wrote:
>On Wed, Sep 25, 2013 at 5:23 PM, Neil Horman <nhorman@tuxdriver.com> wrote:
>> On Wed, Sep 25, 2013 at 03:08:05PM -0600, Bjorn Helgaas wrote:
>>> [+cc Neil (he added this code in da8d1c8ba4), Greg]
>>>
>>> On Mon, Sep 16, 2013 at 7:47 PM, Veaceslav Falico <vfalico@redhat.com> wrote:
>>> > Before trying to kobject_init_and_add(), we add a reference to pdev via
>>> > pci_dev_get(pdev). However, if it fails to init and/or add the kobject, we
>>> > don't return it back - even on out_unroll.
>>> >
>>> > Fix this by adding pci_dev_put(pdev) before going to unrolling section.
>>> >
>>> > CC: Bjorn Helgaas <bhelgaas@google.com>
>>> > CC: linux-pci@vger.kernel.org
>>> > CC: linux-kernel@vger.kernel.org
>>> > Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
>>> > ---
>>> >  drivers/pci/msi.c | 4 +++-
>>> >  1 file changed, 3 insertions(+), 1 deletion(-)
>>> >
>>> > diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c
>>> > index d5f90d6..14bf578 100644
>>> > --- a/drivers/pci/msi.c
>>> > +++ b/drivers/pci/msi.c
>>> > @@ -534,8 +534,10 @@ static int populate_msi_sysfs(struct pci_dev *pdev)
>>> >                 pci_dev_get(pdev);
>>> >                 ret = kobject_init_and_add(kobj, &msi_irq_ktype, NULL,
>>> >                                      "%u", entry->irq);
>>> > -               if (ret)
>>> > +               if (ret) {
>>> > +                       pci_dev_put(pdev);
>>> >                         goto out_unroll;
>>> > +               }
>>> >
>>> >                 count++;
>>> >         }
>>>
>>> I don't understand why this code does the pci_dev_get() in the first
>>> place.  The pdev->msi_list of msi_desc structs is private to the
>>> pci_dev, and even without bumping the refcount, there should be no way
>>> for the pci_dev to be freed before the msi_desc.
>>>
>> Its been a few years now, but IIRC I did the pci_dev_get/put here to ensure that
>> people didn't try to remove the device prior to freeing all their interrupts
>> (i.e I didn't want a broken driver to go through its remove routine without
>> freeing all its irqs).  That might have been the wrong thing to do, but thats
>> what bubbles to the front of my head when looking at this.
>
>That sounds plausible, but I think I'd rather deal with that by having
>the PCI core remove logic free all the interrupts.  I *think* that's
>already in place, i.e., pci_free_resources() calls
>msi_remove_pci_irq_vectors().  So I propose that we remove the
>pci_dev_get()/put() unless we come up with a more compelling reason
>for it.
>
>>> I also don't understand this nearby code (the same pattern appears in
>>> free_msi_irqs()):
>>>
>>>     out_unroll:
>>>         list_for_each_entry(entry, &pdev->msi_list, list) {
>>>                 if (!count)
>>>                         break;
>>>                 kobject_del(&entry->kobj);
>>>                 kobject_put(&entry->kobj);
>>>                 count--;
>>>         }
>>>
>>> Why do we call kobject_del() here?  The kobject_put() will call
>>> kobject_del() anyway, so it looks redundant.
>>> Documentation/kobject.txt says kobject_del() must be called explicitly
>>> to break a circular reference, but I don't think we have that here.
>>>
>>  I think thats exactly why I did it, because of the documentation.  I agree
>> however, it does look redundant.  Harmless, but redundant.
>
>OK, thanks.  I think we should remove it on the grounds that it's not
>needed and removing it will make this code look more similar to other
>callers of kobject_init_and_add(), which means bugs will have fewer
>places to hide.
Hi,

Sounds great, I'll add this as new patches and send this in a separate
patchset with the unregister msi_kset in free_msi_irqs patch, so that it
won't depend on the fix (both removing kobject_del() and
pci_dev_get/put()).

Thank you!

>
>Thanks, Neil!
>
>Bjorn