[dpdk-dev] [PATCH 00/10] igb_uio patches

[Carew, Alan]

-----Original Message-----
From: Stephen Hemminger [mailto:stephen at networkplumber.org] 
Sent: Tuesday, June 10, 2014 6:41 PM
To: Carew, Alan
Cc: dev at dpkd.org
Subject: Re: [PATCH 00/10] igb_uio patches

On Tue, 10 Jun 2014 15:33:31 +0000
"Carew, Alan" <alan.carew at intel.com> wrote:

> Good catches and fixes in this set, would it be possible to refactor the "enum igbuio_intr_mode" to a common header files as with http://dpdk.org/ml/archives/dev/2014-June/003127.html

Actually, thought that was ugly. The userspace shouldn't have to see all these
definitions.

Not a big fan of doing this and it would go against goal of making a version
that could be merged upstream.

Stephen,

One alternative for us is then to take Neil Horman's idea of scanning /sys/bus/pci/devices/<D:B:D.F>/msi_irqs/
This gives us at least (MSI | MSI-X) | INT-X, however the Virtio specification mentions MSI-X explicitly for determining the correct offset and might not address the underlying problem my patch set was addressing, I have not tried Virtio with MSI.

A question then, how does uio_pci_generic handle interrupt modes in terms of userspace interface when determining the configured interrupt mode?

In the short term while we still work with igb_uio we may have to rely on SYSFS, but there is nothing stopping us from having device specific code paths that does not exclude us from ignoring igb_uio. The shared headers also gives us the benefit of having an internal interface(rte_pci_device: rte_intr_mode ) already in place when  swapping out kernel drivers, Anatoly's VFIO implementation already uses such an interface.

Thanks,
Alan