[PATCH v3] tests/intel/xe_fault_injection: Inject errors during exec queue create IOCTL
Gurram, Pravalika
pravalika.gurram at intel.com
Thu Feb 27 05:43:33 UTC 2025
> -----Original Message-----
> From: Dandamudi, Priyanka <priyanka.dandamudi at intel.com>
> Sent: Thursday, February 27, 2025 10:59 AM
> To: Gurram, Pravalika <pravalika.gurram at intel.com>; igt-
> dev at lists.freedesktop.org; Dugast, Francois <francois.dugast at intel.com>
> Cc: Gurram, Pravalika <pravalika.gurram at intel.com>
> Subject: RE: [PATCH v3] tests/intel/xe_fault_injection: Inject errors during
> exec queue create IOCTL
>
>
>
> > -----Original Message-----
> > From: igt-dev <igt-dev-bounces at lists.freedesktop.org> On Behalf Of
> > Pravalika Gurram
> > Sent: 25 February 2025 01:04 PM
> > To: igt-dev at lists.freedesktop.org; Dugast, Francois
> > <francois.dugast at intel.com>
> > Cc: Gurram, Pravalika <pravalika.gurram at intel.com>
> > Subject: [PATCH v3] tests/intel/xe_fault_injection: Inject errors
> > during exec queue create IOCTL
> >
> > Use the fault injection infrastructure to make targeted internal KMD
> > functions fail when executing xe_exec_queue_create_ioctl() so that
> > more code paths are tested, such as error handling and unwinding.
> > include xe_exec_queue_create_bind only for VM_BIND engine class.
> > xe_exec_queue_destroy_ioctl() and
> > xe_exec_queue_get_property_ioctl() are not considered as there is no
> > unwinding code to test with fault injection.
> >
> > Signed-off-by: Pravalika Gurram <pravalika.gurram at intel.com>
> > ---
> > tests/intel/xe_fault_injection.c | 50
> > ++++++++++++++++++++++++++++++++
> > 1 file changed, 50 insertions(+)
> >
> > diff --git a/tests/intel/xe_fault_injection.c
> > b/tests/intel/xe_fault_injection.c
> > index 7ae941367..f25ad9b27 100644
> > --- a/tests/intel/xe_fault_injection.c
> > +++ b/tests/intel/xe_fault_injection.c
> > @@ -185,6 +185,37 @@ inject_fault_probe(int fd, char pci_slot[], const
> > char
> > function_name[])
> > injection_list_do(INJECTION_LIST_REMOVE, function_name); }
> >
> > +/**
> > + * SUBTEST: exec-queue-create-fail-%s
> > + * Description: inject an error in function %arg[1] used in exec
> > +queue create IOCTL to make it fail
> > + * Functionality: fault
> > + *
> > + * arg[1]:
> > + * @xe_exec_queue_create: xe_exec_queue_create
> > + * @xe_hw_engine_group_add_exec_queue:
> > xe_hw_engine_group_add_exec_queue
> > + * @xe_vm_add_compute_exec_queue:
> > xe_vm_add_compute_exec_queue
> > + * @xe_exec_queue_create_bind: xe_exec_queue_create_bind
> > + */
> > +static void
> > +exec_queue_create_fail(int fd, struct drm_xe_engine_class_instance
> > *instance,
> > + const char function_name[], unsigned int flags) {
> > + uint32_t exec_queue_id;
> > + uint32_t vm = xe_vm_create(fd, flags, 0);
> > + /* sanity check */
> > + igt_assert_eq(__xe_exec_queue_create(fd, vm, 1, 1, instance, 0,
> > &exec_queue_id), 0);
> > + xe_exec_queue_destroy(fd, exec_queue_id);
> > +
> > + ignore_faults_in_dmesg(function_name);
> > + injection_list_do(INJECTION_LIST_ADD, function_name);
> > + set_retval(function_name, INJECT_ERRNO);
> > + igt_assert(__xe_exec_queue_create(fd, vm, 1, 1, instance, 0,
> > &exec_queue_id) != 0);
> > + injection_list_do(INJECTION_LIST_REMOVE, function_name);
> > +
> > + igt_assert_eq(__xe_exec_queue_create(fd, vm, 1, 1, instance, 0,
> > &exec_queue_id), 0);
> > + xe_exec_queue_destroy(fd, exec_queue_id); }
> > +
> > static int
> > simple_vm_create(int fd, unsigned int flags) { @@ -283,6 +314,7 @@
> > vm_bind_fail(int fd, const char function_name[]) igt_main {
> > int fd;
> > + struct drm_xe_engine_class_instance *hwe;
> > char pci_slot[NAME_MAX];
> > const struct section {
> > const char *name;
> > @@ -318,6 +350,14 @@ igt_main
> > { }
> > };
> >
> > + const struct section exec_queue_create_fail_functions[] = {
> > + { "xe_exec_queue_create", 0 },
> > + { "xe_hw_engine_group_add_exec_queue", 0 },
> > + { "xe_vm_add_compute_exec_queue",
> > DRM_XE_VM_CREATE_FLAG_LR_MODE },
> > + { "xe_exec_queue_create_bind", 0 },
> > + { }
> > + };
> > +
> > igt_fixture {
> > igt_require(fail_function_injection_enabled());
> > fd = drm_open_driver(DRIVER_XE);
> > @@ -333,6 +373,16 @@ igt_main
> > igt_subtest_f("vm-bind-fail-%s", s->name)
> > vm_bind_fail(fd, s->name);
> >
> > + for (const struct section *s = exec_queue_create_fail_functions; s-
> > >name; s++)
> > + igt_subtest_f("exec-queue-create-fail-%s", s->name)
> > + xe_for_each_engine(fd, hwe) {
> > + if (hwe->engine_class ==
> > DRM_XE_ENGINE_CLASS_VM_BIND &&
> > + strcmp(s->name,
> > "xe_exec_queue_create_bind"))
> > + exec_queue_create_fail(fd, hwe,
> > s[3].name, s[3].flags);
> > + else if (strcmp(s->name,
> > "xe_exec_queue_create_bind"))
> > + exec_queue_create_fail(fd, hwe, s-
> > >name, s->flags);
> In either if-else case you are trying to validate only for
> xe_exec_queue_create_bind out of all the list, what about the others
> {"xe_exec_queue_create", 0 },
> + { "xe_hw_engine_group_add_exec_queue", 0 },
> + { "xe_vm_add_compute_exec_queue",
> DRM_XE_VM_CREATE_FLAG_LR_MODE },
> these are not considered at all. Can you check this once.
xe_exec_queue_create_bind is needed only for VM_BIND instance in kernel this is covered in if and else in kernel
The remaining functions are needed for other engine instances. Hence excluding the VM_BIND instance for other functions
> > +
> > +
> > igt_fixture {
> > xe_sysfs_driver_do(fd, pci_slot,
> XE_SYSFS_DRIVER_UNBIND);
> > }
> > --
> > 2.34.1
More information about the igt-dev
mailing list