Re: [kvm-unit-tests PATCH v3] runtime: Allow to specify properties for accelerator

[Gavin Shan <gshan@redhat.com>]

Hi Drew,

On 6/19/23 18:45, Andrew Jones wrote:
> On Thu, Jun 15, 2023 at 04:21:48PM +1000, Gavin Shan wrote:
>> There are extra properties for accelerators to enable the specific
>> features. For example, the dirty ring for KVM accelerator can be
>> enabled by "-accel kvm,dirty-ring-size=65536". Unfortuntely, the
>> extra properties for the accelerators aren't supported. It makes
>> it's impossible to test the combination of KVM and dirty ring
>> as the following error message indicates.
>>
>>    # cd /home/gavin/sandbox/kvm-unit-tests/tests
>>    # QEMU=/home/gavin/sandbox/qemu.main/build/qemu-system-aarch64 \
>>      ACCEL=kvm,dirty-ring-size=65536 ./its-migration
>>       :
>>    BUILD_HEAD=2fffb37e
>>    timeout -k 1s --foreground 90s /home/gavin/sandbox/qemu.main/build/qemu-system-aarch64 \
>>    -nodefaults -machine virt -accel kvm,dirty-ring-size=65536 -cpu cortex-a57             \
>>    -device virtio-serial-device -device virtconsole,chardev=ctd -chardev testdev,id=ctd   \
>>    -device pci-testdev -display none -serial stdio -kernel _NO_FILE_4Uhere_ -smp 160      \
>>    -machine gic-version=3 -append its-pending-migration # -initrd /tmp/tmp.gfDLa1EtWk
>>    qemu-system-aarch64: kvm_init_vcpu: kvm_arch_init_vcpu failed (0): Invalid argument
>>
>> Allow to specify extra properties for accelerators. With this, the
>> "its-migration" can be tested for the combination of KVM and dirty
>> ring.
>>
>> Signed-off-by: Gavin Shan <gshan@redhat.com>
>> ---
>> v3: Split $ACCEL to $ACCEL and $ACCEL_PROPS in get_qemu_accelerator()
>>      and don't print them as output, suggested by Drew.
>> ---
>>   arm/run               | 12 ++++--------
>>   powerpc/run           |  5 ++---
>>   s390x/run             |  5 ++---
>>   scripts/arch-run.bash | 21 +++++++++++++--------
>>   x86/run               |  5 ++---
>>   5 files changed, 23 insertions(+), 25 deletions(-)
>>
>> diff --git a/arm/run b/arm/run
>> index c6f25b8..d9ebe59 100755
>> --- a/arm/run
>> +++ b/arm/run
>> @@ -10,10 +10,8 @@ if [ -z "$KUT_STANDALONE" ]; then
>>   fi
>>   processor="$PROCESSOR"
>>   
>> -accel=$(get_qemu_accelerator) ||
>> -	exit $?
>> -
>> -if [ "$accel" = "kvm" ]; then
>> +get_qemu_accelerator || exit $?
>> +if [ "$ACCEL" = "kvm" ]; then
>>   	QEMU_ARCH=$HOST
>>   fi
>>   
>> @@ -23,11 +21,9 @@ qemu=$(search_qemu_binary) ||
>>   if [ "$QEMU" ] && [ -z "$ACCEL" ] &&
>>      [ "$HOST" = "aarch64" ] && [ "$ARCH" = "arm" ] &&
>>      [ "$(basename $QEMU)" = "qemu-system-arm" ]; then
>> -	accel=tcg
>> +	ACCEL="tcg"
>>   fi
>>   
> 
> As I pointed out in the v2 review we can't just s/accel/ACCEL/ without
> other changes. Now ACCEL will also be set when the above condition
> is checked, making it useless. Please ensure the test case that commit
> c7d6c7f00e7c ("arm/run: Use TCG with qemu-system-arm on arm64 systems")
> fixed still works with your patch.
> 

Sorry that I missed your comments for v2. In order to make the test case
in c7d6c7f00e7c working, we just need to call set_qemu_accelerator() after
the chunk of code, like below. When $ACCEL is set to "tcg" by the conditional
code, it won't be changed in the following set_qemu_accelerator().

Could you Please confirm if it looks good to you so that I can integrate
the changes to v4 and post it.

arm/run
--------

processor="$PROCESSOR"

if [ "$QEMU" ] && [ -z "$ACCEL" ] &&
    [ "$HOST" = "aarch64" ] && [ "$ARCH" = "arm" ] &&
    [ "$(basename $QEMU)" = "qemu-system-arm" ]; then
         ACCEL="tcg"
fi

set_qemu_accelerator || exit $?
if [ "$ACCEL" = "kvm" ]; then
         QEMU_ARCH=$HOST
fi


Thanks,
Gavin