[PATCH v2 i-g-t] tests/intel/xe_fault_injection: Inject errors during observation IOCTL

[Pottumuttu, Sai Teja]

On 13-03-2025 15:10, nakshtra.goyal at intel.com wrote:
> From: Nakshtra Goyal <nakshtra.goyal at intel.com>
> 
> Use the fault injection infrastructure to make targeted internal KMD
> functions fail when executing xe_observation_ioctl()
> so that more code paths are tested, such as error handling and unwinding.
> 
> v1: Adding remove config at end of test , Adding igt_sysfs into igt_assert
> so test fails if sysfs read fails (Sai Teja)
> 
> Signed-off-by: Nakshtra Goyal <nakshtra.goyal at intel.com>
> ---
>   tests/intel/xe_fault_injection.c | 56 +++++++++++++++++++++++++++++++-
>   tests/meson.build                |  1 +
>   2 files changed, 56 insertions(+), 1 deletion(-)
> 
> diff --git a/tests/intel/xe_fault_injection.c b/tests/intel/xe_fault_injection.c
> index 5409d19a6..1f150fbb0 100644
> --- a/tests/intel/xe_fault_injection.c
> +++ b/tests/intel/xe_fault_injection.c
> @@ -21,6 +21,7 @@
>   #include "lib/intel_pat.h"
>   #include "xe/xe_ioctl.h"
>   #include "xe/xe_query.h"
> +#include "xe/xe_oa.h"
>   
>   #define INJECT_ERRNO	-ENOMEM
>   #define BO_ADDR		0x1a0000
> @@ -314,10 +315,51 @@ vm_bind_fail(int fd, const char function_name[])
>   	igt_assert_eq(simple_vm_bind(fd, vm), 0);
>   }
>   
> +/**
> + * SUBTEST: oa-add-config-fail-%s
> + * Description: inject an error in function %arg[1] used in oa add config IOCTL to make it fail
> + * Functionality: fault
> + *
> + * arg[1]:
> + * @xe_oa_alloc_regs:		xe_oa_alloc_regs
> + */
> +static void
> +oa_add_config_fail(int fd, int sysfs, int devid, const char function_name[])
> +{
> +	char path[512];
> +	uint64_t config_id;
> +#define SAMPLE_MUX_REG (intel_graphics_ver(devid) >= IP_VER(20, 0) ?	\
> +			0x13000 /* PES* */ : 0x9888 /* NOA_WRITE */)
> +
> +	uint32_t mux_regs[] = { SAMPLE_MUX_REG, 0x0 };
> +	struct drm_xe_oa_config config;
> +	const char *uuid = "01234567-0123-0123-0123-0123456789ab";

Looking at other oa tests, I think its a good practice to make sure that 
we don't have a config with that uuid already and if its there, we will 
have to remove it before proceeding.

> +
> +	memset(&config, 0, sizeof(config));
> +	memcpy(config.uuid, uuid, sizeof(config.uuid));
> +	config.n_regs = 1;
> +	config.regs_ptr = to_user_pointer(mux_regs);
> +	snprintf(path, sizeof(path), "metrics/%s/id", uuid);
> +
> +	igt_assert_lt(0, intel_xe_perf_ioctl(fd, DRM_XE_OBSERVATION_OP_ADD_CONFIG, &config));
> +	igt_assert(igt_sysfs_scanf(sysfs, path, "%"PRIu64, &config_id) == 1);

Looks like instead of reading sysfs we can do something like this as 
well here,

int ret;
ret = intel_xe_perf_ioctl(fd, DRM_XE_OBSERVATION_OP_ADD_CONFIG, &config);
igt_assert_lt(0, ret);
igt_assert_eq(intel_xe_perf_ioctl(fd, 
DRM_XE_OBSERVATION_OP_REMOVE_CONFIG, &ret), 0);

Do you think sysfs read gives any advantage here?

> +	igt_assert_eq(intel_xe_perf_ioctl(fd, DRM_XE_OBSERVATION_OP_REMOVE_CONFIG, &config_id), 0);
> +
> +	ignore_faults_in_dmesg(function_name);
> +	injection_list_do(INJECTION_LIST_ADD, function_name);
> +	set_retval(function_name, INJECT_ERRNO);
> +	igt_assert_lt(intel_xe_perf_ioctl(fd, DRM_XE_OBSERVATION_OP_ADD_CONFIG, &config), 0);

Maybe it would be better to do assert_eq to INJECT_ERRNO instead of just 
saying that return value should be less than 0?

> +	injection_list_do(INJECTION_LIST_REMOVE, function_name);
> +
> +	igt_assert_lt(0, intel_xe_perf_ioctl(fd, DRM_XE_OBSERVATION_OP_ADD_CONFIG, &config));
> +	igt_assert_eq(intel_xe_perf_ioctl(fd, DRM_XE_OBSERVATION_OP_REMOVE_CONFIG, &config_id), 0);

The config_id here can be different I think. We need to get the new 
config_id.

Sorry for not observing these while reviewing v1.

Thanks,
Sai Teja

> +}
> +
>   igt_main
>   {
> -	int fd;
> +	int fd, sysfs;
>   	struct drm_xe_engine_class_instance *hwe;
> +	static uint32_t devid;
>   	char pci_slot[NAME_MAX];
>   	const struct section {
>   		const char *name;
> @@ -366,9 +408,16 @@ igt_main
>   		{ }
>   	};
>   
> +	const struct section oa_add_config_fail_functions[] = {
> +		{ "xe_oa_alloc_regs"},
> +		{ }
> +	};
> +
>   	igt_fixture {
>   		igt_require(fail_function_injection_enabled());
>   		fd = drm_open_driver(DRIVER_XE);
> +		devid = intel_get_drm_devid(fd);
> +		sysfs = igt_sysfs_open(fd);
>   		igt_device_get_pci_slot_name(fd, pci_slot);
>   		setup_injection_fault();
>   	}
> @@ -393,6 +442,10 @@ igt_main
>   				if (hwe->engine_class == DRM_XE_ENGINE_CLASS_VM_BIND)
>   					exec_queue_create_fail(fd, hwe, s->name, s->flags);
>   
> +	for (const struct section *s = oa_add_config_fail_functions; s->name; s++)
> +		igt_subtest_f("oa-add-config-fail-%s", s->name)
> +			oa_add_config_fail(fd, sysfs, devid, s->name);
> +
>   	igt_fixture {
>   		xe_sysfs_driver_do(fd, pci_slot, XE_SYSFS_DRIVER_UNBIND);
>   	}
> @@ -402,6 +455,7 @@ igt_main
>   			inject_fault_probe(fd, pci_slot, s->name);
>   
>   	igt_fixture {
> +		close(sysfs);
>   		drm_close_driver(fd);
>   		xe_sysfs_driver_do(fd, pci_slot, XE_SYSFS_DRIVER_BIND);
>   	}
> diff --git a/tests/meson.build b/tests/meson.build
> index 2f5406523..c4a9093c4 100644
> --- a/tests/meson.build
> +++ b/tests/meson.build
> @@ -401,6 +401,7 @@ extra_dependencies = {
>   	'perf_pmu':  [ lib_igt_perf ],
>   	'sw_sync': [ libatomic ],
>   	'xe_oa': [ lib_igt_xe_oa ],
> +	'xe_fault_injection': [ lib_igt_xe_oa ],
>   }
>   
>   test_executables = []