Access to the PCI config space

Access to the PCI config space

[Patrick Mahan]

Has anyone contemplated providing access to the PCI config space
(pci_read_config_[byte|word|dword]) for DPDK?  It seems simple to
perform, some preliminary testing by me seems to indicate opening
the /sys/bus/pci/devices/<domain:bus:loc.func>/config file, seeking
to the correct offset and reading off the value.

It seems to be correctly returning expected values and sysfs-pci.txt
seems to indicate that this is RW so you can set values as well.
It would seem a natural to make the PCI support code in DPDK more
fully capable of porting PCI drivers.

I realize that somethings should be avoid (ie. causing interrupts),
but I know that my device needs to modify bits in both the DEVCTL
field of the PCI config and a BAR0 register to ensure they have
the same value.

Thoughts?

Thanks,

Patrick

Re: Access to the PCI config space

[Stephen Hemminger]

On Wed, 21 Aug 2013 18:13:21 -0700
Patrick Mahan <mahan-5dHXHCkEAVbYtjvyW6yDsg@public.gmane.org> wrote:

> Has anyone contemplated providing access to the PCI config space
> (pci_read_config_[byte|word|dword]) for DPDK?  It seems simple to
> perform, some preliminary testing by me seems to indicate opening
> the /sys/bus/pci/devices/<domain:bus:loc.func>/config file, seeking
> to the correct offset and reading off the value.
> 
> It seems to be correctly returning expected values and sysfs-pci.txt
> seems to indicate that this is RW so you can set values as well.
> It would seem a natural to make the PCI support code in DPDK more
> fully capable of porting PCI drivers.
> 
> I realize that somethings should be avoid (ie. causing interrupts),
> but I know that my device needs to modify bits in both the DEVCTL
> field of the PCI config and a BAR0 register to ensure they have
> the same value.
> 
> Thoughts?
> 
> Thanks,
> 
> Patrick

I played with that and it worked, but proved be unnecessary for the
hardware we support. If your device needs it, then sure it makes sense
to be supported genericly.