Re: [kvm-unit-tests PATCH 1/2] Add the possibility to do simple migration tests

[Thomas Huth <thuth@redhat.com>]

On 11.03.2017 02:42, David Matlack wrote:
> On Thu, Mar 9, 2017 at 9:27 AM, Thomas Huth <thuth@redhat.com> wrote:
>> To be able to do simple migration tests with kvm-unit-tests, too,
>> add a helper script that does all the necessary work: Start two
>> instances of QEMU (source and destination) with QMP sockets for
>> sending commands to them, then trigger the migration from one
>> instance to the other and finally signal the end of the migration
>> to the guest by injecting an NMI.
>> This helper script is now used automatically for powerpc tests
>> if the test is put into the "migration" group in the unittests.cfg
>> file.
>>
>> Signed-off-by: Thomas Huth <thuth@redhat.com>
>> ---
>>  powerpc/run                      |  5 ++++
>>  scripts/qemu-migration-helper.sh | 51 ++++++++++++++++++++++++++++++++++++++++
>>  scripts/runtime.bash             |  3 +++
>>  3 files changed, 59 insertions(+)
>>  create mode 100755 scripts/qemu-migration-helper.sh
>>
>> diff --git a/powerpc/run b/powerpc/run
>> index 6269abb..126fed5 100755
>> --- a/powerpc/run
>> +++ b/powerpc/run
>> @@ -41,6 +41,11 @@ if ! $qemu -machine '?' 2>&1 | grep 'pseries' > /dev/null; then
>>         exit 2
>>  fi
>>
>> +if [ "$MIGRATION" = "1" ]; then
>> +       export QEMU_BIN="$qemu"
>> +       qemu=`dirname $0`/../scripts/qemu-migration-helper.sh
>> +fi
>> +
>>  M='-machine pseries'
>>  M+=",accel=$ACCEL"
>>  command="$qemu -nodefaults $M -bios $FIRMWARE"
>> diff --git a/scripts/qemu-migration-helper.sh b/scripts/qemu-migration-helper.sh
>> new file mode 100755
>> index 0000000..0cb7e4a
>> --- /dev/null
>> +++ b/scripts/qemu-migration-helper.sh
>> @@ -0,0 +1,51 @@
>> +#!/bin/sh
>> +
>> +# This script runs two instances of QEMU and then migrates the guest from one
>> +# instance to the other. The end of the migration is signalled to the guest by
>> +# injecting an NMI.
>> +
>> +if [ "x$QEMU_BIN" = "x" ]; then
>> +    echo "QEMU_BIN must be set to the QEMU binary"
>> +    exit 1
>> +fi
>> +
>> +if ! command -v nc >/dev/null 2>&1; then
>> +    echo "$0 needs nc (netcat)"
>> +    exit 1
>> +fi
>> +
>> +qmp_cmd()
>> +{
>> +    echo '{ "execute": "qmp_capabilities" }{ "execute":' "$2" '}' | nc -U $1
>> +}
>> +
>> +tempname=`mktemp`
>> +qmp1=${tempname}.qmp1
>> +qmp2=${tempname}.qmp2
>> +qmpout1=${tempname}.qmpout1
>> +qmpout2=${tempname}.qmpout2
>> +stdout1=${tempname}.stdout1
>> +stdout2=${tempname}.stdout2
>> +migsock=${tempname}.mig
>> +
>> +$QEMU_BIN $* -chardev socket,id=mon1,path=${qmp1},server,nowait \
>> +       -mon chardev=mon1,mode=control > ${stdout1} &
>> +
>> +$QEMU_BIN $* -chardev socket,id=mon2,path=${qmp2},server,nowait \
>> +       -mon chardev=mon2,mode=control -incoming unix:${migsock} > ${stdout2} &
>> +
>> +sleep 1
> 
> Suggest monitoring the QEMU serial port for a specific token to know
> when to initiate the migration. e.g. the guest can print
> "MIGRATE_ME_NOW" when it's ready. This would avoid initiating a
> migration too early.

Yes, that's a good idea, I'll add a check for something like this.

>> +
>> +qmp_cmd ${qmp1} '"migrate", "arguments": { "uri": "unix:'${migsock}'" }' > ${qmpout1}
>> +while qmp_cmd ${qmp1} '"query-migrate"' | grep -q '"active"' ; do
>> +    sleep 1
>> +done
>> +qmp_cmd ${qmp1} '"quit"'> ${qmpout1} 2>/dev/null
>> +
>> +qmp_cmd ${qmp2} '"inject-nmi"'> ${qmpout2}
> 
> Does "inject-nmi" work on other architectures?

It should work at least on x86. Not sure about ARM though.

> An alternative approach is to add a new Qemu testdev register that
> counts the number of times a VM has been migrated. The guest can
> monitor this to know when the migration is complete.

AFAIK the testdev only works on x86, so that's not an option for powerpc
here.

 Thomas