On Thu, Oct 10, 2024 at 10:09:12AM +0200, Philipp Stanner wrote:

On Wed, 2024-10-09 at 20:32 +0200, Heiner Kallweit wrote:

quoted

On 09.10.2024 10:35, Philipp Stanner wrote:

...

quoted

quoted

To do so, a pci_intx() version that is always-managed, and one that
is
never-managed are provided. Then, all pci_intx() users are ported
to the
version they need. Afterwards, pci_intx() can be cleaned up and the
users of the never-managed version be ported back to pci_intx().

This way we'd get this PCI API consistent again.

AFAICS pci_intx() is used only by drivers which haven't been
converted
to the pci_alloc_irq_vectors() API yet. Wouldn't it be better to do
this
instead of trying to improve pci_intx()?

My first impression was the same...

This would be the créme-de-la-créme-solution, yes.

But such a portation would require more detailed knowledge of the old
drivers.

In this discussion, Alex points out that at least in some drivers, you
can't replace pci_intx() without further ado:
https://lore.kernel.org/all/20240904151020.486f599e.alex.williamson@redhat.com/ (local)

What we could do is mark pci_intx() and pcim_intx() as deprecated and
point everyone to pci_alloc_irq_vectors(). Then someone can look into
porting the old drivers at some point in the future.

...but here I got the point by Philipp.

-- 
With Best Regards,
Andy Shevchenko