Re: [PATCH V3] Delay allocation of PCI device IRQs from boot time until bus scan time to fix PCI hotplugging

[Bjorn Helgaas <bhelgaas@google.com>]

On Thu, Oct 02, 2014 at 05:07:28AM +0100, matt@masarand.com wrote:
> Sorry, the previous submission was damaged by my editor which purged the sign off lines from the patches,
> this is a fixed version.
> 
> pci_fixup_irqs is the current method used in most arches to assign IRQs to
> PCI devices. This has a number of flaws including it requiring an extra walk
> of the PCI bus and most importantly not running for devices which are added
> after boot time, this includes hot-added devices on boards that support this.
> 
> This patch set modifies the IRQ mapping and swizzling infrastructure
> such that functions are registered by the boot code (instead of being
> run directly by the boot code) and then run later in the enable_device
> path such that it will apply to all devices and not only those inserted
> at boot time.
> 
> This is cleaner as it unifies how each architecture allocates PCI IRQs as much
> as possible (some arches were particularly resistant to these changes so some
> work-arounds have been used or the change disabled completely).
> 
> The caveat here is that I have been forced to modify some architecture specific
> code for various architectures which I cannot test due to not having such boards
> available. The code seems correct and the changes in most cases are small and
> trivial however I have not tested all the patches for rare arches.
> 
> NEW: Since the last version all changes and suggested bug fixes have been implemented
> the patch-set has also been rebased onto the latest kernel version to ease merging.

Hi Matthew,

I'm not sure where we are with this.  This has a lot of good fixes that I'd
like to merge, but I think there were some comments that needed to be
addressed.  But I might have missed something, so I'm just checking to see
if I accidentally dropped the ball.

Bjorn