Re: make test failure for iotest 267 on qemu-system-sparc64

[Mark Cave-Ayland]

On 22/06/2020 10:25, Thomas Huth wrote:

>> After a recent rebase I've started seeing iotest 267 after running "make 
>> check" on
>> qemu-system-sparc64. The diff output looks similar to this:
>>
>> --- /home/build/src/qemu/git/qemu/tests/qemu-iotests/267.out    2020-01-22
>> 17:57:54.246650995 +0000
>> +++ /home/build/src/qemu/git/qemu/tests/qemu-iotests/267.out.bad        
>> 2020-03-07
>> 09:25:36.044451658 +0000
>> @@ -41,13 +41,7 @@
>>  Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=134217728
>>  Testing: -drive driver=IMGFMT,file=TEST_DIR/t.IMGFMT,if=none -device
>> virtio-blk,drive=none0
>>  QEMU X.Y.Z monitor - type 'help' for more information
>> -(qemu) savevm snap0
>> -(qemu) info snapshots
>> -List of snapshots present on all disks:
>> -ID        TAG                 VM SIZE                DATE       VM CLOCK
>> ---        snap0                  SIZE yyyy-mm-dd hh:mm:ss   00:00:00.000
>> -(qemu) loadvm snap0
>> -(qemu) quit
>> +(qemu) QEMU_PROG: -device virtio-blk,drive=none0: PCI: no slot/function 
>> available
>> for virtio-blk-pci, all in use or reserved
>>
>> The error message here is because the sun4u machine PCI root (default) bus 
>> contains
>> only 2 PCI bridges and cannot have devices plugged into it directly. An 
>> example of
>> how to use virtio-blk-pci on qemu-system-sparc64 is shown at
>> https://wiki.qemu.org/Documentation/Platforms/SPARC#All_PCI_devices_are_attached_behind_one_of_the_simba_PCI_bridges.
>>
>> It seems that all that is missing is a way to specify the bus= parameter for 
>> the
>> virtio-blk-pci device for this test to work. Can anyone suggest a suitable 
>> way to do
>> this?
> 
>  Hi!
> 
> Sorry for the late reply... but better late than never:

Heh, that's okay :)

> I think you likely don't want to change each and every current and
> future iotest to contain a "bus=..." parameter - that's not reasonable.
> So I see two possibilities to get "make check" working here again:
> 
> 1) From a user point of view, it would be great if PCI devices could be
> used on the sun4u machine also without specifying the "bus=..."
> property. Is there a way to convince QEMU to plug PCI devices in a
> different bus with free slots in case the root bus is already full? ...
> I had a quick look at the sources, but unfortunately I failed to spot
> the code that decides which PCI bus should be used in case no "bus=..."
> property has been specified, so no clue whether this is feasible or not.

It has been a while since I looked at this, but from memory the problem is that 
the
default bus value is 0 which matches the first (root) PCI bus. Which 
unfortunately
can't have anything plugged into it for sun4u.

Perhaps the QEMU machine needs a way to be able to override the id of the 
default PCI
bus which is used for -device?

> 2) Simply change the _require_devices function in
> tests/qemu-iotests/common.rc to filter out the virtio and pci devices on
> sparc64. Something like this should do the job, I hope:
> 
> diff a/tests/qemu-iotests/common.rc b/tests/qemu-iotests/common.rc
> --- a/tests/qemu-iotests/common.rc
> +++ b/tests/qemu-iotests/common.rc
> @@ -788,6 +788,11 @@ _require_devices()
>  {
>      available=$($QEMU -M none -device help | \
>                  grep ^name | sed -e 's/^name "//' -e 's/".*$//')
> +    case "$QEMU_PROG" in
> +        *qemu-system-sparc64)
> +            available=$(grep -v -i -E 'pci|virtio' <<< $available)
> +            ;;
> +    esac
>      for device
>      do
>          if ! echo "$available" | grep -q "$device" ; then

I'd prefer a solution that keeps the virtio tests if possible, since 5.0 is 
actually
broken for booting from a virtio-blk-pci device - there were a couple of changes
leading up to release that broke OpenBIOS (see
https://mail.coreboot.org/hyperkitty/list/openbios@openbios.org/2020/5/) that 
managed
to slip by.


ATB,

Mark.