My static checker detected that free_irq() is called even after
request_irq() failed in ddb_probe(). In this case, the kernel may try to
free dev->pdev->irq although the IRQ is not assigned. This event rarely
occurs, but always introduces a warning if it happens.

"goto fail1" always results in disabling enabled MSI and releasing a
requested IRQ. It seems like the former handling is necessary. So I added
a conditional branch before the free_irq() (stat == 0 means request_irq()
succeeds).
Signed-off-by: Takeshi Yoshimura <yos@sslab.ics.keio.ac.jp>
Signed-off-by: Mauro Carvalho Chehab <mchehab@osg.samsung.com>