Re: [i-g-t] tests/xe_vm: Report OOM for vm_bind ioctl under memory pressure

["Thomas Hellström" <thomas.hellstrom@linux.intel.com>]

On Wed, 2025-07-16 at 17:22 +0530, priyanka.dandamudi@intel.com wrote:
> From: Priyanka Dandamudi <priyanka.dandamudi@intel.com>
> 
> Add a test which create buffer objects on an LR vm and vm_binding
> buffer objects in a loop until it reaches OOM.
> This is to check that buffer objects on a single vm doesnot get
> evicted
> and instead report with ENOMEM in non fault mode.
> 
> Signed-off-by: Priyanka Dandamudi <priyanka.dandamudi@intel.com>
> Acked-by: Sai Gowtham Ch <sai.gowtham.ch@intel.com>
> ---
>  tests/intel/xe_vm.c | 65
> +++++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 65 insertions(+)
> 
> diff --git a/tests/intel/xe_vm.c b/tests/intel/xe_vm.c
> index c1abb08bb..e808a0b4e 100644
> --- a/tests/intel/xe_vm.c
> +++ b/tests/intel/xe_vm.c
> @@ -2368,6 +2368,65 @@ static void invalid_vm_id(int fd)
>  	do_ioctl_err(fd, DRM_IOCTL_XE_VM_DESTROY, &destroy, ENOENT);
>  }
>  
> +/**
> + * SUBTEST: out-of-memory
> + * Description: Test if vm_bind ioctl results in oom
> + * when creating and vm_binding buffer objects on an LR vm beyond
> available visible vram size.
> + * Functionality: oom
> + * Test category: functionality test
> + */
> +static void test_oom(int fd)
> +{
> +#define USER_FENCE_VALUE 0xdeadbeefdeadbeefull
> +#define BO_SIZE xe_bb_size(fd, SZ_512M)
> +#define MAX_BUFS (int)(xe_visible_vram_size(fd, 0) / BO_SIZE)
> +	uint64_t addr = 0x1a0000;
> +	uint64_t vm_sync;
> +	uint32_t bo[MAX_BUFS + 1];
> +	uint32_t *data[MAX_BUFS + 1];
> +	uint32_t vm;
> +	struct drm_xe_sync sync[1] = {
> +		{ .type = DRM_XE_SYNC_TYPE_USER_FENCE, .flags =
> DRM_XE_SYNC_FLAG_SIGNAL,
> +		  .timeline_value = USER_FENCE_VALUE },
> +	};
> +	size_t bo_size = BO_SIZE;
> +	int total_bufs = MAX_BUFS;
> +
> +	vm = xe_vm_create(fd, DRM_XE_VM_CREATE_FLAG_LR_MODE, 0);
> +	for (int iter = 0; iter <= total_bufs; iter++) {
> +		bo[iter] = xe_bo_create(fd, 0, bo_size,
> +					vram_if_possible(fd, 0),
> +					DRM_XE_GEM_CREATE_FLAG_DEFER
> _BACKING |
> +					DRM_XE_GEM_CREATE_FLAG_NEEDS
> _VISIBLE_VRAM);
> +
> +		sync[0].addr = to_user_pointer(&vm_sync);
> +		if (iter < total_bufs)
> +			xe_vm_bind_async(fd, vm, 0, bo[iter], 0,
> +					 addr + bo_size * iter,
> bo_size, sync, 1);

This will error if VRAM was smaller than estimated, right? I think that
may happen if previous tests in CI have leaked VRAM memory and IMO the
test should still pass in that case, perhaps with a warning or debug
message as long as we see an OOM error.


> +		else {
> +			igt_assert_eq(__xe_vm_bind(fd, vm, 0,
> bo[iter], 0,
> +					           addr + bo_size *
> iter, bo_size,
> +						  
> DRM_XE_VM_BIND_OP_MAP, 0, sync, 1, 0,
> +						  
> DEFAULT_PAT_INDEX, 0), -ENOMEM);

XeKMD may return -ENOMEM or -ENOSPC on these errors, so we should also
check for -ENOSPC.


Otherwise LGTM.
/Thomas


> +			break;
> +
> +		}
> +		xe_wait_ufence(fd, &vm_sync, USER_FENCE_VALUE, 0,
> NSEC_PER_SEC);
> +		vm_sync = 0;
> +		data[iter] = xe_bo_map(fd, bo[iter], bo_size);
> +		memset(data[iter], 0, bo_size);
> +	}
> +
> +	for (int iter = 0; iter < total_bufs; iter++) {
> +		sync[0].addr = to_user_pointer(&vm_sync);
> +		xe_vm_unbind_async(fd, vm, 0, 0, addr + bo_size *
> iter, bo_size,
> +				   sync, 1);
> +		xe_wait_ufence(fd, &vm_sync, USER_FENCE_VALUE, 0,
> NSEC_PER_SEC);
> +		munmap(data[iter], bo_size);
> +		gem_close(fd, bo[iter]);
> +	}
> +}
> +
>  igt_main
>  {
>  	struct drm_xe_engine_class_instance *hwe, *hwe_non_copy =
> NULL;
> @@ -2759,6 +2818,12 @@ igt_main
>  	igt_subtest("invalid-vm-id")
>  		invalid_vm_id(fd);
>  
> +	igt_subtest("out-of-memory") {
> +		igt_require(xe_has_vram(fd));
> +		igt_assert(xe_visible_vram_size(fd, 0));
> +		test_oom(fd);
> +	}
> +
>  	igt_fixture
>  		drm_close_driver(fd);
>  }