On 6/25/21 11:17 AM, Igor Mammedov wrote:
> Currently it is not possible to create tests that have KVM as a hard
> requirement on a host that doesn't support KVM for tested target
> binary (modulo going through the trouble of compiling out
> the offending test case).
>
> Following scenario makes test fail when it's run on non x86 host:
> qemu-system-x86_64 -enable-kvm -M q35,kernel-irqchip=on -smp 1,maxcpus=288
>
> This patch introduces qtest_has_accel() to let users check if accel is
> available in advance and avoid executing non run-able test-cases.
>
> It implements detection of TCG and KVM only, the rest could be
> added later on, when we actually start testing them in qtest.
>
> Signed-off-by: Igor Mammedov <imamm...@redhat.com>
> ---
> PS:
> There is an alternative 'runtime' approach on list:
> 'query-accels' series.
> https://patchwork.kernel.org/project/qemu-devel/patch/20210503211020.894589-4-phi...@redhat.com/
>
> on upside this impl. much cheaper to execute than the 'query-accels'
> as it doesn't need to run QEMU for probing.
>
> v4:
> - s/qtest_has_kvm/qtest_has_accel/
> - add TCG detection (based on --disable-tcg)
> v3:
> - make it work with --disable-kvm
> (Claudio Fontana <cfont...@suse.de>)
> v2:
> - fix access() check.
> s/access()/!access()/
> - format C array items at meson.build time
>
> CC: th...@redhat.com
> CC: lviv...@redhat.com
> CC: f4...@amsat.org
> ---
> tests/qtest/libqos/libqtest.h | 8 ++++++++
> meson.build | 6 ++++++
> tests/qtest/libqtest.c | 27 +++++++++++++++++++++++++++
> 3 files changed, 41 insertions(+)
>
> diff --git a/tests/qtest/libqos/libqtest.h b/tests/qtest/libqos/libqtest.h
> index a68dcd79d4..59e9271195 100644
> --- a/tests/qtest/libqos/libqtest.h
> +++ b/tests/qtest/libqos/libqtest.h
> @@ -588,6 +588,14 @@ bool qtest_big_endian(QTestState *s);
> */
> const char *qtest_get_arch(void);
>
> +/**
> + * qtest_has_accel:
> + * @accel_name: Accelerator name to check for.
> + *
> + * Returns: true if the accelerator is built in.
> + */
> +bool qtest_has_accel(const char *accel_name);
> +
> /**
> * qtest_add_func:
> * @str: Test case path.
> diff --git a/meson.build b/meson.build
> index d8a92666fb..7eeb201179 100644
> --- a/meson.build
> +++ b/meson.build
> @@ -76,6 +76,12 @@ else
> kvm_targets = []
> endif
>
> +kvm_targets_c = ''
> +if not get_option('kvm').disabled() and targetos == 'linux'
> + kvm_targets_c = '"' + '" ,"'.join(kvm_targets) + '"'
> +endif
> +config_host_data.set('CONFIG_KVM_TARGETS', kvm_targets_c)
> +
> accelerator_targets = { 'CONFIG_KVM': kvm_targets }
> if cpu in ['x86', 'x86_64', 'arm', 'aarch64']
> # i368 emulator provides xenpv machine type for multiple architectures
> diff --git a/tests/qtest/libqtest.c b/tests/qtest/libqtest.c
> index 825b13a44c..643769e416 100644
> --- a/tests/qtest/libqtest.c
> +++ b/tests/qtest/libqtest.c
> @@ -920,6 +920,33 @@ const char *qtest_get_arch(void)
> return end + 1;
> }
>
> +bool qtest_has_accel(const char *accel_name)
> +{
> + if (g_str_equal(accel_name, "tcg")) {
> +#if defined(CONFIG_TCG)
The point of the "alternative 'runtime' approach" was to remove
access to accel-specific definitions in tests/ and you are
adding yet another one :( Maybe now it is acceptable, I am not
sure since I haven't followed the latest accelerator modularization
effort series. So I'll defer that to the active developers doing
it.
> + return true;
> +#else
> + return false;
> +#endif
> + } else if (g_str_equal(accel_name, "kvm")) {
> + int i;
> + const char *arch = qtest_get_arch();
> + const char *targets[] = { CONFIG_KVM_TARGETS };
> +
> + for (i = 0; i < ARRAY_SIZE(targets); i++) {
> + if (!strncmp(targets[i], arch, strlen(arch))) {
> + if (!access("/dev/kvm", R_OK | W_OK)) {
> + return true;
> + }
> + }
> + }
> + } else {
> + /* not implemented */
> + g_assert_not_reached();
> + }
> + return false;
> +}
> +
> bool qtest_get_irq(QTestState *s, int num)
> {
> /* dummy operation in order to make sure irq is up to date */
>
