VirtIO based network driver works well after i8259 masks maintenance correction

Pavel Pisa ppisa4lists at pikron.com
Sun Sep 25 01:14:23 UTC 2016


Hello Jin-Hyun Kim, Hyun-Wook Jin  and others,

I have returned back to VirtIO based network driver
after I have make PCI based E100 card work by woraround
which re-enables IRQ in driver worker daemon.

In meanwhile, I have included alternative networking setup
with E100 to RTEMS QEMU Wiki page

https://devel.rtems.org/wiki/Developer/Simulators/QEMU#Inteli82557be100PCICardEmulation

I have cleaned a VirtIO driver a little but including
IRQ workaround would be really complicated in this case.
Interrupt source signalling is cleaned by inport_byte
from corresponding VIRTIO_PCI_ISR. But if interrupt
is disabled on controller level by i8259 then reenable
would to be moved to rtems_bsdnet worker thread.
But that would violate all ideas about separation
of layers and encapsulation. Not by one, but over three
levels. So I have corrected (at least according my feeling)
i8259 IRQ processing.

The VirtIO networking seems to work great with this fix.
There is updated VirtIO support on branch "devel-virtio"
of my GitHub repository

  https://github.com/ppisa/rtems/tree/devel-virtio

I would like express big thanks to authors that they
have contributed port.

The last time there has not been reached final decision
if the code should be included in RTEMS or if there is
plug on old networking stack.

I think that it works well and can be used before
there is other alternative location for the code.

As for the code, I would be happy if it can be moved
out of i386 BSP directory or at least that part which
is independent if there is some i386 specific code.
But it uses generic IRQ now, generic PCI so it
should be usable with all other architectures.

There is even another reason why to included virtio
support. It can be used (if extended) even for providing
disk device to the virtualized environment and this
part probably stays in core RTEMS codebase even if
libbsd is used for networking.

Best wishes,

               Pavel




More information about the devel mailing list