On Tue, Jan 21, 2025 at 03:54:17PM +0000, Alexandru Elisei wrote:

Hi Drew,

On Tue, Jan 21, 2025 at 03:48:55PM +0100, Andrew Jones wrote:

quoted

On Mon, Jan 20, 2025 at 04:43:01PM +0000, Alexandru Elisei wrote:

quoted

Arm and arm64 support running the tests under kvmtool. Unsurprisingly,
kvmtool and qemu have a different command line syntax for configuring and
running a virtual machine.

On top of that, when kvm-unit-tests has been configured to run under
kvmtool (via ./configure --target=kvmtool), the early UART address changes,
and if then the tests are run with qemu, this warning is displayed:

WARNING: early print support may not work. Found uart at 0x9000000, but early base is 0x1000000.

At the moment, the only way to run a test under kvmtool is manually, as no
script has any knowledge of how to invoke kvmtool. Also, unless one looks
at the logs, it's not obvious that the test runner is using qemu to run the
tests, and not kvmtool.

To avoid any confusion for unsuspecting users, refuse to run a test via the
testing scripts when kvm-unit-tests has been configured for kvmtool.

There are four different ways to run a test using the test infrastructure:
with run_tests.sh, by invoking arm/run or arm/efi/run with the correct
parameters (only the arm directory is mentioned here because the tests can
be configured for kvmtool only on arm and arm64), and by creating
standalone tests. Add a check in each of these locations for the supported
virtual machine manager.

Signed-off-by: Alexandru Elisei <redacted>
---
 arm/efi/run             | 8 ++++++++
 arm/run                 | 9 +++++++++
 run_tests.sh            | 8 ++++++++
 scripts/mkstandalone.sh | 8 ++++++++
 4 files changed, 33 insertions(+)

diff --git a/arm/efi/run b/arm/efi/run
index 8f41fc02df31..916f4c4deef6 100755
--- a/arm/efi/run
+++ b/arm/efi/run

@@ -12,6 +12,14 @@ fi
 source config.mak
 source scripts/arch-run.bash
 
+case "$TARGET" in
+qemu)
+    ;;
+*)
+    echo "$0 does not support '$TARGET'"
+    exit 2
+esac
+
 if [ -f /usr/share/qemu-efi-aarch64/QEMU_EFI.fd ]; then
 	DEFAULT_UEFI=/usr/share/qemu-efi-aarch64/QEMU_EFI.fd
 elif [ -f /usr/share/edk2/aarch64/QEMU_EFI.silent.fd ]; then

diff --git a/arm/run b/arm/run
index efdd44ce86a7..6db32cf09c88 100755
--- a/arm/run
+++ b/arm/run

@@ -8,6 +8,15 @@ if [ -z "$KUT_STANDALONE" ]; then
 	source config.mak
 	source scripts/arch-run.bash
 fi
+
+case "$TARGET" in
+qemu)
+    ;;
+*)
+   echo "'$TARGET' not supported"
+   exit 3

I think we want exit code 2 here.

Exit code 2 is already in use in arm/run. Now that I'm looking more closely
at it, exit code 2 is already in use in run_tests.sh, same for
mkstandalone.sh and arm/efi/run.

How about using 3 everywhere as the exit code?

In kvm-unit-tests, exit code 2 is what we use for "most likely a run
script failed" (see the comment above run_qemu() in
scripts/arch-run.bash). We don't try to create a new error code for each
type of error, but we do have the error message as well. So if there's a
higher level runner, which runs this runner, it only needs to learn that
2 is likely a script failure and that an error message will hopefully
point the way to the problem.

Thanks,
drew