Re: [Qemu-devel] [PATCH v8 00/16] spapr: add support for pci hotplug

[Michael Roth <mdroth@linux.vnet.ibm.com>]

Quoting Michael Roth (2015-04-30 16:04:49)
> Quoting David Gibson (2015-04-29 23:13:04)
> > On Wed, Apr 22, 2015 at 01:28:04AM -0500, Michael Roth wrote:
> > > These patches are based on spapr-next, and can also be obtained
> > > from:
> > 
> > Thanks Michael.
> > 
> > I've made a few small conflict fixes and merged this into spapr-next.
> 
> Great, thanks!
> 
> > 
> > Could you please:
> > 
> >   a) Pull the new spapr-next and test that your code still works
> 
> I've run it through all my basic testing and things look good. I did
> need one change though regarding the 2.4 machine type intro though:
> 
> diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
> index a8116d0..8fbcaf5 100644
> --- a/hw/ppc/spapr.c
> +++ b/hw/ppc/spapr.c
> @@ -1919,6 +1919,7 @@ static void spapr_machine_register_types(void)
>      type_register_static(&spapr_machine_2_1_info);
>      type_register_static(&spapr_machine_2_2_info);
>      type_register_static(&spapr_machine_2_3_info);
> +    type_register_static(&spapr_machine_2_4_info);
>  }
>  
>  type_init(spapr_machine_register_types)
> 
> Just needs to get squashed into "pseries: Add pseries-2.4 machine type"
> I'd imagine.
> 
> > 
> >   b) Send me some minimal instructions for doing a basic test of PCI
> >      hotplug.  I have some spapr cleanups in mind, and I'd like to be
> >      able to check for myself that I at least haven't broken things
> >      totally.
> 
> For all the following tests, the following guest distros (assuming they're
> updated) should support the hotplug event and will do the whole device
> bring-up/tear-down in response to hotplug/unplug: Fedora 21+, RHEL 6.6+, 
> rhel 7.1+, sles12+, ubuntu 14.04+.
> 
> For quick testing just booting to SLOF you can do a reboot after
> device_add/device_del to get the device picked up by SLOF PCI enumeration,
> or finalized by QEMU's reset hooks, respectively. This won't exercise any
> RTAS calls or actual device functionality, but should touch most of the
> hotplug-related code.
> 
> = BASIC TEST
> 
>  # start guest with an extra PHB to cover hotplug across multiple PHBs
> 
>  qemu ... -device spapr-pci-host-bridge,index=1,phb1
> 
>  # from QEMU HMP monitor
> 
>  netdev_add user,id=hpnet0.0
>  device_add virtio-net-pci,id=hp0.0,netdev=hpnet0.0 #PHB 0
>  netdev_add user,id=hpnet1.0
>  device_add virtio-net-pci,id=hp1.0,netdev=hpnet1.0,bus=phb1.0 #PHB 1
> 
>  <if running supported distro, verify devices show up in lspci, and
>   that interfaces can be brought online and lease IPs>
>  <if just booting to SLOF, verify devices are preset via 'info pci'
>   monitor command>
> 
>  device_del hp0.0
>  netdev_del hpnet0.0
>  device_del hp1.0
>  netdev_del hpnet1.0
> 
>  <if running supported distro, verify devices are removed from lspci>
>  <if just booting to SLOF, issue system_reset and verify devices
>   are removed from output of 'info pci' monitor command>
> 
> = MULTIFUNCTION HOTPLUG TEST
> 
>  #start guest as above
> 
>  netdev_add user,id=hpnet1.0.0
>  device_add virtio-net-pci,id=hp1.0.0,netdev=hpnet1.0.0,bus=phb1.0,\
>     addr=0x0.0,multifunction=on
>  netdev_add user,id=hpnet1.0.1
>  device_add virtio-net-pci,id=hp1.0.1,netdev=hpnet1.0.1,bus=phb1.0,\
>     addr=0x0.1,multifunction=on
> 
>  <verify each function as above, and that they're functions of the same
>   device>
> 
>  device_del hp1.0.0
>  netdev_del hpnet1.0.0
>  device_del hp1.0.1
>  netdev_del hpnet1.0.1

Actually this should probably be:

  device_del hp1.0.1
  netdev_del hpnet1.0.1
  device_del hp1.0.0
  netdev_del hpnet1.0.0

From what I recall multifunction devices are expected to have a 0
function present when attached, and behavior is undefined otherwise.

> 
>  <verify each function as above>
> 
> > 
> > -- 
> > David Gibson                    | I'll have my music baroque, and my code
> > david AT gibson.dropbear.id.au  | minimalist, thank you.  NOT _the_ _other_
> >                                 | _way_ _around_!
> > http://www.ozlabs.org/~dgibson