Re: kernel panic becase pci register more than once

[Christoph Hellwig <hch@xxxxxx>]

On Mon, Sep 17, 2018 at 02:15:03PM -0500, Bjorn Helgaas wrote:
> A WARN_ON() shouldn't cause the kernel to panic.  Do you mean that it
> causes a backtrace?  I think that's the intended behavior, because
> calling this twice would be a driver bug, and we want to find and fix
> those bugs.
> 
> > diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c
> > index f2ef896..324840f 100644
> > --- a/drivers/pci/msi.c
> > +++ b/drivers/pci/msi.c
> > @@ -958,7 +958,9 @@ static int __pci_enable_msix(struct pci_dev *dev,
> > struct msix_entry *entries,
> >                         }
> >                 }
> >         }
> > -       WARN_ON(!!dev->msix_enabled);
> > +
> > +       if (dev->msix_enabled)
> > +               return -EINVAL;

This is a grave driver bug, so we should keep a trace.  Then again
handling even grave driver bugs more gracefull is a good idea, so I'd
vote for:

	if (WARN_ON_ONCE(dev->msix_enabled))
		return -EINVAL;