Re: [PATCH v9 tty-next 2/4] serial: 8250_pci1xxxx: Add driver for quad-uart support

[Andy Shevchenko]

On Fri, Dec 16, 2022 at 09:30:42AM +0000, Tharunkumar.Pasumarthi@xxxxxxxxxxxxx wrote:
> > From: Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>
> > Sent: Friday, December 16, 2022 1:56 PM

...

> > > > > +             pci_iounmap(pdev, priv->membase);
> > > >
> > > > Here is inconsistency on how you interpret pci_*() calls when
> > > > pcim_enable_device() has been used. I.e. for IRQ you don't
> > > > deallocate resources explicitly (yes, it's done automatically
> > > > anyway), but you explicitly call pci_iounmap(). Choose a single approach
> > for all of them.
> > >
> > > AFAIK call to pci_iounmap cannot be avoided since pci_ioremap_bar is not
> > 'managed' API.
> > > You suggest calling pci_free_irq_vectors (even though it is not mandatory)?
> > 
> > Why is it not mandatory?
> 
> Following is the reason why I felt calling pci_free_irq_vectors is not
> mandatory.  Correct me if my understanding is wrong.
> 
> Following is the Callback Sequence (from 6.1 kernel) that gets executed upon
> calling pci_alloc_irq_vectors:
> pci_alloc_irq_vectors =>
>  pci_alloc_irq_vectors_affinity =>
> __pci_enable_msi_range =>
>  pci_setup_msi_context =>
>  pcim_setup_msi_release =>
> 
> devm_add_action(&dev->dev, pcim_msi_release, dev);
> 
> Inside pcim_msi_release: (called since pcim_enable_device is used)
> pci_free_irq_vectors(dev);
> 
> From this sequence, it seemed like, kernel takes care of freeing irq and
> calling pci_free_irq_vectors is not required.

Technically you are correct, but... read this discussion, and pay attention to
what Christoph said there:

https://lore.kernel.org/linux-serial/0250c46e-da6c-71f3-50ae-b7c17fd0bd2c@xxxxxxxxxxx/

-- 
With Best Regards,
Andy Shevchenko