[Intel-xe] [PATCH V2 i-g-t] tests/xe_drm_fdinfo: Add tests to read and verify fdinfo

Kumar, Janga Rahul janga.rahul.kumar at intel.com
Wed Sep 13 08:31:52 UTC 2023



> -----Original Message-----
> From: Upadhyay, Tejas <tejas.upadhyay at intel.com>
> Sent: Friday, September 8, 2023 1:45 PM
> To: igt-dev at lists.freedesktop.org
> Cc: intel-xe at lists.freedesktop.org; Aravind Iddamsetty
> <aravind.iddamsetty at linux.intel.com>; Ghimiray, Himal Prasad
> <himal.prasad.ghimiray at intel.com>; Dandamudi, Priyanka
> <priyanka.dandamudi at intel.com>; Kumar, Janga Rahul
> <janga.rahul.kumar at intel.com>; Kamil Konieczny
> <kamil.konieczny at linux.intel.com>; Upadhyay, Tejas
> <tejas.upadhyay at intel.com>
> Subject: [PATCH V2 i-g-t] tests/xe_drm_fdinfo: Add tests to read and verify
> fdinfo
> 
> Add tests to verify fdinfo published by xe KMD with following subtests :
> @basic
> @drm-total-residents
> @drm-shared
> @drm-active

It would be better if you can add sample output of fdinfo here or in the test.
> 
> TODO: drm-purgeable test is not possbile as we consider objects in system
> memory to be purgeable but current xe KMD does not allow BOs to be
> created in system memory (instead KMD creates it in XE_PL_TT which is
> VRAM backed system memory).
> 
> V2:
>   - Replace printf with igt_info - Kamil
>   - run checkpatch - Kamil
>   - use igt_assert_f - Kamil
> 
> Signed-off-by: Tejas Upadhyay <tejas.upadhyay at intel.com>
> ---
>  tests/intel/xe_drm_fdinfo.c | 266
> ++++++++++++++++++++++++++++++++++++
>  tests/meson.build           |   1 +
>  2 files changed, 267 insertions(+)
>  create mode 100644 tests/intel/xe_drm_fdinfo.c
> 
> diff --git a/tests/intel/xe_drm_fdinfo.c b/tests/intel/xe_drm_fdinfo.c new
> file mode 100644 index 000000000..8aa753d95
> --- /dev/null
> +++ b/tests/intel/xe_drm_fdinfo.c
> @@ -0,0 +1,266 @@
> +// SPDX-License-Identifier: MIT
> +/*
> + * Copyright © 2023 Intel Corporation
> + */
> +
> +#include "igt.h"
> +#include "igt_core.h"
> +#include "igt_device.h"
> +#include "igt_drm_fdinfo.h"
> +#include "lib/igt_syncobj.h"
> +#include "xe_drm.h"
> +#include "xe/xe_ioctl.h"
> +#include "xe/xe_query.h"
> +#include "xe/xe_spin.h"
> +/**
> + * TEST: xe drm fdinfo
> + * Description: Read and verify drm client memory consumption using
> +fdinfo
> + * Feature: SMI, core
> + * Category: Software building block
> + * Sub-category: driver
> + * Functionality: Per client memory statistics
> + * Run type: FULL
> + * Test category: SysMan
> + *
> + * SUBTEST: basic
> + * Description: Check if basic fdinfo content is present
> + *
> + * SUBTEST: drm-total-resident
> + * Description: Create and compare total and resident memory
> +consumption by client
> + *
> + * SUBTEST: drm-shared
> + * Description: Create and compare shared memory consumption by client
> + *
> + * SUBTEST: drm-active
> + * Description: Create and compare active memory consumption by client
> +*/
> +
> +IGT_TEST_DESCRIPTION("Read and verify drm client memory consumption
> +using fdinfo");
> +
> +/* Subtests */
> +static void test_active(int fd, struct drm_xe_engine_class_instance
> +*eci) {
> +	struct drm_client_fdinfo info = { };
> +	uint32_t vm;
> +	uint64_t addr = 0x1a0000;
> +	struct drm_xe_sync sync[2] = {
> +		{ .flags = DRM_XE_SYNC_SYNCOBJ | DRM_XE_SYNC_SIGNAL,
> },
> +		{ .flags = DRM_XE_SYNC_SYNCOBJ | DRM_XE_SYNC_SIGNAL,
> },
> +	};
> +	struct drm_xe_exec exec = {
> +		.num_batch_buffer = 1,
> +		.num_syncs = 2,
> +		.syncs = to_user_pointer(sync),
> +	};
> +#define N_EXEC_QUEUES   2
> +	uint32_t exec_queues[N_EXEC_QUEUES];
> +	uint32_t bind_exec_queues[N_EXEC_QUEUES];
> +	uint32_t syncobjs[N_EXEC_QUEUES + 1];
> +	size_t bo_size;
> +	uint32_t bo = 0;
> +	int region = 0x2; /* VRAM 0 */
> +	struct {
> +		struct xe_spin spin;
> +		uint32_t batch[16];
> +		uint64_t pad;
> +		uint32_t data;
> +	} *data;
> +	int i, b;
> +
> +	vm = xe_vm_create(fd, DRM_XE_VM_CREATE_ASYNC_BIND_OPS, 0);
> +	bo_size = sizeof(*data) * N_EXEC_QUEUES;
> +	bo_size = ALIGN(bo_size + xe_cs_prefetch_size(fd),
> +			xe_get_default_alignment(fd));
> +	bo = xe_bo_create_flags(fd, vm, bo_size,
> +			region);
> +	data = xe_bo_map(fd, bo, bo_size);
> +
> +	for (i = 0; i < N_EXEC_QUEUES; i++) {
> +		exec_queues[i] = xe_exec_queue_create(fd, vm, eci, 0);
> +		bind_exec_queues[i] = xe_bind_exec_queue_create(fd, vm,
> 0);
> +		syncobjs[i] = syncobj_create(fd, 0);
> +	}
> +	syncobjs[N_EXEC_QUEUES] = syncobj_create(fd, 0);
> +
> +	sync[0].handle = syncobj_create(fd, 0);
> +	xe_vm_bind_async(fd, vm, bind_exec_queues[0], bo, 0, addr,
> bo_size,
> +			sync, 1);
> +
> +	for (i = 0; i < N_EXEC_QUEUES; i++) {
> +		uint64_t spin_offset = (char *)&data[i].spin - (char *)data;
> +		uint64_t spin_addr = addr + spin_offset;
> +		int e = i;
> +
> +		if (i == 0) {
> +			xe_spin_init(&data[i].spin, spin_addr, true);
> +			exec.exec_queue_id = exec_queues[e];
> +			exec.address = spin_addr;
> +			sync[0].flags &= ~DRM_XE_SYNC_SIGNAL;
> +			sync[1].flags |= DRM_XE_SYNC_SIGNAL;
> +			sync[1].handle = syncobjs[e];
> +			xe_exec(fd, &exec);
> +			xe_spin_wait_started(&data[i].spin);
> +
> +			addr += bo_size;
> +			sync[1].flags &= ~DRM_XE_SYNC_SIGNAL;
> +			sync[1].handle = syncobjs[e];
> +			xe_vm_bind_async(fd, vm, bind_exec_queues[e], bo,
> 0, addr,
> +					bo_size, sync + 1, 1);
> +			addr += bo_size;
> +		} else {
> +			sync[0].flags |= DRM_XE_SYNC_SIGNAL;
> +			xe_vm_bind_async(fd, vm, bind_exec_queues[e], bo,
> 0, addr,
> +					bo_size, sync, 1);
> +		}
> +
> +		b = igt_parse_drm_fdinfo(fd, &info, NULL, 0, NULL, 0);
> +		igt_assert_f(b != 0, "failed with err:%d\n", errno);
> +
> +		/* Client memory consumption includes public objects
> +		 * as well as internal objects hence if bo is active on
> +		 * N_EXEC_QUEUES consumption should be
> +		 * >= bo_size + bo_size * N_EXEC_QUEUES
> +		 */
> +		igt_info("info.total:%ld active:%ld bo_size:%ld\n",
> +				info.region_mem[region].total,
> +				info.region_mem[region].active, bo_size);
> +		igt_assert(info.region_mem[region].active >=
> +				bo_size + bo_size * N_EXEC_QUEUES);
> +	}
> +	xe_spin_end(&data[0].spin);
> +
> +	syncobj_destroy(fd, sync[0].handle);
> +	sync[0].handle = syncobj_create(fd, 0);
> +	sync[0].flags |= DRM_XE_SYNC_SIGNAL;
> +	xe_vm_unbind_all_async(fd, vm, 0, bo, sync, 1);
> +	igt_assert(syncobj_wait(fd, &sync[0].handle, 1, INT64_MAX, 0,
> NULL));
> +
> +	syncobj_destroy(fd, sync[0].handle);
> +	for (i = 0; i < N_EXEC_QUEUES; i++) {
> +		syncobj_destroy(fd, syncobjs[i]);
> +		xe_exec_queue_destroy(fd, exec_queues[i]);
> +		xe_exec_queue_destroy(fd, bind_exec_queues[i]);
> +	}
> +
> +	munmap(data, bo_size);
> +	gem_close(fd, bo);
> +	xe_vm_destroy(fd, vm);
> +}
> +
> +static void test_shared(int xe)
> +{
> +	struct drm_client_fdinfo info = { };
> +	struct drm_gem_flink flink;
> +	struct drm_gem_open open_struct;
> +	uint32_t bo;
> +	int region = 0x1;
> +	int ret;
> +
> +	bo = xe_bo_create_flags(xe, 0, 4096, region);
> +
> +	flink.handle = bo;
> +	ret = igt_ioctl(xe, DRM_IOCTL_GEM_FLINK, &flink);
> +	igt_assert_eq(ret, 0);
> +
> +	open_struct.name = flink.name;
> +	ret = igt_ioctl(xe, DRM_IOCTL_GEM_OPEN, &open_struct);
> +	igt_assert_eq(ret, 0);
> +	igt_assert(open_struct.handle != 0);
> +
> +	ret = igt_parse_drm_fdinfo(xe, &info, NULL, 0, NULL, 0);
> +	igt_assert_f(ret != 0, "failed with err:%d\n", errno);
> +
> +	igt_info("info.total:%ld shared:%ld\n",
> +			info.region_mem[region].total,
> +			info.region_mem[region].shared);
> +	igt_assert_eq(info.region_mem[region].shared,
> +			2 * 4096);
	Use expected_val kind of variables to check for values avoid math operations in asserts to improve readability.
	Pls update this through out the test.
> +
> +	gem_close(xe, bo);
> +}
> +
> +static void test_total_resident(int xe) {
> +	struct drm_xe_query_mem_region *memregion;
> +	uint64_t memreg = all_memory_regions(xe), region;
> +	struct drm_client_fdinfo info = { };
> +	uint32_t vm;
> +	uint32_t handle;
> +	uint64_t addr = 0x1a0000;
> +	int ret;
> +
> +	vm = xe_vm_create(xe, DRM_XE_VM_CREATE_SCRATCH_PAGE, 0);
> +
> +	xe_for_each_mem_region(xe, memreg, region) {
> +		memregion = xe_mem_region(xe, region);
> +
> +		handle = xe_bo_create_flags(xe, vm, memregion-
> >min_page_size, region);
> +		xe_vm_bind_sync(xe, vm, handle, 0, addr, memregion-
> >min_page_size);
> +
> +		ret = igt_parse_drm_fdinfo(xe, &info, NULL, 0, NULL, 0);
> +		igt_assert_f(ret != 0, "failed with err:%d\n", errno);
> +		/* currently xe KMD maps memory class system region to
> +		 * XE_PL_TT thus we need memregion->instance + 1
		Instead of adding 1(constant), It would be neat if you can define some macro and use it 
> +		 */
> +		igt_info("info.total:%ld resident:%ld for instance: %d
> bo_size:%d\n",
> +				info.region_mem[memregion->instance +
> 1].total,
> +				info.region_mem[memregion->instance +
> 1].resident,
> +				memregion->instance, memregion-
> >min_page_size);
> +		/* Client memory consumption includes public objects
> +		 * as well as internal objects hence it should be
> +		 * >= memregion->min_page_size
> +		 */
> +		igt_assert(info.region_mem[memregion->instance + 1].total
> >=
> +				memregion->min_page_size);
> +		igt_assert(info.region_mem[memregion->instance +
> 1].resident >=
> +				memregion->min_page_size);
> +		xe_vm_unbind_sync(xe, vm, 0, addr, memregion-
> >min_page_size);
> +		gem_close(xe, handle);
> +	}
> +
> +	xe_vm_destroy(xe, vm);
> +}
> +
> +static void basic(int xe)
> +{
> +	struct drm_client_fdinfo info = { };
> +	unsigned int ret;
> +
> +	ret = igt_parse_drm_fdinfo(xe, &info, NULL, 0, NULL, 0);
> +	igt_assert_f(ret != 0, "failed with err:%d\n", errno);
> +
> +	igt_assert(!strcmp(info.driver, "xe")); }
	This we could have checked inside fixture itself as if info is not of xe driver, we can skip the tests?

Thanks,
Rahul
> +
> +igt_main
> +{
> +	int xe;
> +
> +	igt_fixture {
> +		struct drm_client_fdinfo info = { };
> +
> +		xe = drm_open_driver(DRIVER_XE);
> +		igt_require_xe(xe);
> +		igt_require(igt_parse_drm_fdinfo(xe, &info, NULL, 0, NULL,
> 0));
> +	}
> +
> +	igt_describe("Check if basic fdinfo content is present");
	We are just checking the info.driver only in basic, can we improve by checking any other fields to be non-zero/valid to improve the test.
> +	igt_subtest("basic")
> +		basic(xe);
> +
> +	igt_describe("Create and compare total and resident memory
> consumption by client");
> +	igt_subtest("drm-total-resident")
> +		test_total_resident(xe);
> +
> +	igt_describe("Create and compare shared memory consumption by
> client");
> +	igt_subtest("drm-shared")
> +		test_shared(xe);
> +
> +	igt_describe("Create and compare active memory consumption by
> client");
> +	igt_subtest("drm-active")
> +		test_active(xe, xe_hw_engine(xe, 0));
> +
> +	igt_fixture {
> +		drm_close_driver(xe);
> +	}
> +}
> diff --git a/tests/meson.build b/tests/meson.build index
> aa8e3434c..f5f495050 100644
> --- a/tests/meson.build
> +++ b/tests/meson.build
> @@ -275,6 +275,7 @@ intel_xe_progs = [
>  	'xe_compute',
>  	'xe_dma_buf_sync',
>  	'xe_debugfs',
> +	'xe_drm_fdinfo',
>  	'xe_evict',
>  	'xe_exec_balancer',
>  	'xe_exec_basic',
> --
> 2.25.1



More information about the Intel-xe mailing list