Re: [Qemu-devel] [RFC PATCH 0/9] Configurable PCI device addresses

[Anthony Liguori <anthony@codemonkey.ws>]

Markus Armbruster wrote:
> This patch series makes PCI device addresses configurable for a number
> of devices.  For now, I just covered the ones I consider important
> and/or where the change is somewhat invasive.
>
> Why is a configurable PCI device address useful?  Currently, QEMU
> assigns a few device addresses statically, and the rest dynamically on
> a first come, first serve basis[*].  If you add or remove a device,
> all devices registering later change address.  I'm told a certain
> operating system that can't be named in polite company can get rather
> upset about that.
>
> Making the device address configurable looks like the simplest
> solution to this problem.  Only users who really mind changing
> addresses have to deal with the new configuration parameter.
>   

I think we're starting to get to the tipping point where we really just 
need to introduce a machine config file.

Paul, you mentioned you were working on something before?  Care to 
update us on the status?

I know there was a fair bit of support before for a FDT based machine 
config.

Regards,

Anthony Liguori

> I started with Izik's proposal to "allow setting static slot values
> for pci devices from the command line", see
>
>     http://lists.gnu.org/archive/html/qemu-devel/2008-01/msg00601.html
>     http://lists.gnu.org/archive/html/qemu-devel/2007-11/msg00620.html
>
> but ended up doing it quite differently.
>
> I'm just soliciting comments.  I do *not* ask for a commit of anything
> right now.
>
> The patch series consists of these parts:
>
> [1/9] PCI device registration helpers
> [2/9] Clean up handling of name=value,... part of -vga option argument
> [3/9] Support pci=... in option argument of -vga
> [4/9] Convert virtio_init_pci() to pci_register_device_2()
> [5/9] Support pci=... in option argument of -net nic
> [6/9] Make drives_opt[] accessible from device initialization
> [7/9] Support pci=... in option argument of -drive if=virtio
> [8/9] New option -audio as a more flexible alternative to -soundhw
> [9/9] Support pci=... in option argument of -audio
>
> Overall diffstat:
>
>  hw/ac97.c           |   11 +---
>  hw/audiodev.h       |    4 +-
>  hw/cirrus_vga.c     |    6 +-
>  hw/e1000.c          |    6 +-
>  hw/eepro100.c       |   19 ++++----
>  hw/es1370.c         |   11 +---
>  hw/mips_malta.c     |   17 +++----
>  hw/ne2000.c         |    7 +--
>  hw/pc.c             |   23 ++++------
>  hw/pc.h             |   11 +---
>  hw/pci.c            |  109 ++++++++++++++++++++++++++++++++++++++++++++--
>  hw/pci.h            |   25 ++++++----
>  hw/pcnet.c          |    6 +-
>  hw/ppc440_bamboo.c  |    5 +-
>  hw/ppc_chrp.c       |    4 +-
>  hw/ppc_oldworld.c   |    4 +-
>  hw/ppc_prep.c       |    4 +-
>  hw/r2d.c            |    8 ++--
>  hw/realview.c       |    2 +-
>  hw/rtl8139.c        |    7 +--
>  hw/sun4u.c          |    4 +-
>  hw/versatilepb.c    |    2 +-
>  hw/vga.c            |   40 ++++++++--------
>  hw/virtio-balloon.c |    2 +-
>  hw/virtio-blk.c     |    4 +-
>  hw/virtio-blk.h     |    2 +-
>  hw/virtio-console.c |    2 +-
>  hw/virtio-net.c     |    5 +-
>  hw/virtio-net.h     |    2 +-
>  hw/virtio.c         |    5 +-
>  hw/virtio.h         |    2 +-
>  hw/vmware_vga.c     |    7 ++-
>  net.c               |    1 +
>  net.h               |    1 +
>  sysemu.h            |   13 +++++-
>  vl.c                |  120 ++++++++++++++++++++++++++++++++-------------------
>  36 files changed, 315 insertions(+), 186 deletions(-)
>
>
> [*] To be precise: it assigns the dev.func part dynamically.  The
> domain:bus: part is still statically assigned.
>
>
>