Re: [kvm-unit-tests RFC PATCH 4/5] scripts: Generate kvmtool standalone tests

[Alexandru Elisei <alexandru.elisei@arm.com>]

Hi Drew,

On 9/8/21 5:07 PM, Andrew Jones wrote:
> On Wed, Sep 08, 2021 at 04:37:39PM +0100, Alexandru Elisei wrote:
>> Hi Drew,
>>
>> On 9/7/21 11:21 AM, Andrew Jones wrote:
>>> On Fri, Jul 02, 2021 at 05:31:21PM +0100, Alexandru Elisei wrote:
>>>> Add support for the standalone target when running kvm-unit-tests under
>>>> kvmtool.
>>>>
>>>> Example command line invocation:
>>>>
>>>> $ ./configure --target=kvmtool
>>>> $ make clean && make standalone
>>>>
>>>> Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com>
>>>> ---
>>>>  scripts/mkstandalone.sh | 14 +++++++-------
>>>>  1 file changed, 7 insertions(+), 7 deletions(-)
>>>>
>>>> diff --git a/scripts/mkstandalone.sh b/scripts/mkstandalone.sh
>>>> index 16f461c06842..d84bdb7e278c 100755
>>>> --- a/scripts/mkstandalone.sh
>>>> +++ b/scripts/mkstandalone.sh
>>>> @@ -44,6 +44,10 @@ generate_test ()
>>>>  	config_export ARCH_NAME
>>>>  	config_export PROCESSOR
>>>>  
>>>> +	if [ "$ARCH" = "arm64" ] || [ "$ARCH" = "arm" ]; then
>>>> +		config_export TARGET
>>>> +	fi
>>> Should export unconditionally, since we'll want TARGET set
>>> unconditionally.
>> Yes, will do.
>>
>>>> +
>>>>  	echo "echo BUILD_HEAD=$(cat build-head)"
>>>>  
>>>>  	if [ ! -f $kernel ]; then
>>>> @@ -59,7 +63,7 @@ generate_test ()
>>>>  		echo 'export FIRMWARE'
>>>>  	fi
>>>>  
>>>> -	if [ "$ENVIRON_DEFAULT" = "yes" ] && [ "$ERRATATXT" ]; then
>>>> +	if [ "$TARGET" != "kvmtool" ] && [ "$ENVIRON_DEFAULT" = "yes" ] && [ "$ERRATATXT" ]; then
>>> I think it would be better to ensure that ENVIRON_DEFAULT is "no" for
>>> TARGET=kvmtool in configure.
>> From looking at the code, it is my understanding that with ENVIRON_DEFAULT=yes, an
>> initrd file is generated with the contents of erratatxt and other information, in
>> a key=value pair format. This initrd is then passed on to the test (please correct
>> me if I'm wrong). With ENVIRON_DEFAULT=no (set via ./configure
>> --disable-default-environ), this initrd is not generated.
>>
>> kvmtool doesn't have support for passing an initrd when loading firmware, so yes,
>> I believe the default should be no.
>>
>> However, I have two questions:
>>
>> 1. What happens when the user specifically enables the default environ via
>> ./configure --enable-default-environ --target=kvmtool? In my opinion, that should
>> be an error because the user wants something that is not possible with kvmtool
>> (loading an image with --firmware in kvmtool means that the initrd image it not
>> loaded into the guest memory and no node is generated for it in the dtb), but I
>> would like to hear your thoughts about it.
> As part of the forcing ENVIRON_DEFAULT to "no" for kvmtool in configure an
> error should be generated if a user tries to explicitly enable it.
>
>> 2. If the default environment is disabled, is it still possible for an user to
>> pass an initrd via other means? I couldn't find where that is implemented, so I'm
>> guessing it's not possible.
> Yes, a user could have a KVM_UNIT_TESTS_ENV environment variable set when
> they launch the tests. If that variable points to a file then it will get
> passed as an initrd. I guess you should also report a warning in arm/run
> if KVM_UNIT_TESTS_ENV is set which states that the environment file will
> be ignored when running with kvmtool.

Thank you for explaining it, I had looked at
scripts/arch-run.bash::initrd_create(), but it didn't click that setting the
KVM_UNIT_TESTS_ENV environment variable is enough to generate and use the initrd.

After looking at the code some more, in the logs the -initrd argument is shown as
a comment, instead of an actual argument that is passed to qemu:

timeout -k 1s --foreground 90s /usr/bin/qemu-system-aarch64 -nodefaults -machine
virt,gic-version=host,accel=kvm -cpu host -device virtio-serial-device -device
virtconsole,chardev=ctd -chardev testdev,id=ctd -device pci-testdev -display none
-serial stdio -kernel arm/cache.flat -smp 1 # -initrd /tmp/tmp.rUIZ3h9KLJ
QEMU_ACCEL = kvm
INFO: IDC-DIC: dcache clean to PoU required
INFO: IDC-DIC: icache invalidation to PoU required
PASS: IDC-DIC: code generation
SUMMARY: 1 tests

This is done intentionally in scripts/arch-run.bash::run_qemu(). I don't
understand the reason for that. When I first looked at the logs, I was sure that
no initrd is passed to the test. I had to go dig through the scripts to figure out
that the "#" sign (which marks the beginning of a comment) is not present in the
qemu invocation.

Thanks,

Alex

>
> There aren't currently any other ways to invoke the addition of the
> -initrd command line option, because so far we only support passing a
> single file to test (the environment "file"). If we ever want to pass
> more files, then we'd need to create a simple file system on the initrd
> and make it possible to add -initrd even when no environment is desired.
> But, that may never happen.
>
> Thanks,
> drew
>
>> Thanks,
>>
>> Alex
>>
>>>
>>>>  		temp_file ERRATATXT "$ERRATATXT"
>>>>  		echo 'export ERRATATXT'
>>>>  	fi
>>>> @@ -95,12 +99,8 @@ function mkstandalone()
>>>>  	echo Written $standalone.
>>>>  }
>>>>  
>>>> -if [ "$TARGET" = "kvmtool" ]; then
>>>> -	echo "Standalone tests not supported with kvmtool"
>>>> -	exit 2
>>>> -fi
>>>> -
>>>> -if [ "$ENVIRON_DEFAULT" = "yes" ] && [ "$ERRATATXT" ] && [ ! -f "$ERRATATXT" ]; then
>>>> +if [ "$TARGET" != "kvmtool" ] && [ "$ENVIRON_DEFAULT" = "yes" ] && \
>>>> +		[ "$ERRATATXT" ] && [ ! -f "$ERRATATXT" ]; then
>>>>  	echo "$ERRATATXT not found. (ERRATATXT=$ERRATATXT)" >&2
>>>>  	exit 2
>>>>  fi
>>>> -- 
>>>> 2.32.0
>>>>
>>> Thanks,
>>> drew 
>>>