[PATCH 1/2] tests/intel/xe_vm: Add bind array conflict tests
Matthew Brost
matthew.brost at intel.com
Fri Jul 19 19:27:45 UTC 2024
Add bind array conflict tests which modify the same address space within
1 bind array. Not a use case but stresses the driver logic. Also add an
error injection section which tests error of bind IOCTL.
v2:
- Use do_ioctl_err (Matthew Auld)
- Move xe_vm_bind_array_err to xe_vm.c
Signed-off-by: Matthew Brost <matthew.brost at intel.com>
---
tests/intel/xe_vm.c | 281 ++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 281 insertions(+)
diff --git a/tests/intel/xe_vm.c b/tests/intel/xe_vm.c
index d507e7cb16..e9711a60a5 100644
--- a/tests/intel/xe_vm.c
+++ b/tests/intel/xe_vm.c
@@ -850,6 +850,24 @@ test_bind_execqueues_independent(int fd, struct drm_xe_engine_class_instance *ec
xe_vm_destroy(fd, vm);
}
+static void xe_vm_bind_array_err(int fd, uint32_t vm, uint32_t exec_queue,
+ struct drm_xe_vm_bind_op *bind_ops,
+ uint32_t num_bind, struct drm_xe_sync *sync,
+ uint32_t num_syncs, int err)
+{
+ struct drm_xe_vm_bind bind = {
+ .vm_id = vm,
+ .num_binds = num_bind,
+ .vector_of_binds = (uintptr_t)bind_ops,
+ .num_syncs = num_syncs,
+ .syncs = (uintptr_t)sync,
+ .exec_queue_id = exec_queue,
+ };
+
+ igt_assert(num_bind > 1);
+ do_ioctl_err(fd, DRM_IOCTL_XE_VM_BIND, &bind, err);
+}
+
#define BIND_ARRAY_BIND_EXEC_QUEUE_FLAG (0x1 << 0)
@@ -992,6 +1010,257 @@ test_bind_array(int fd, struct drm_xe_engine_class_instance *eci, int n_execs,
xe_vm_destroy(fd, vm);
}
+/**
+ * SUBTEST: bind-array-conflict
+ * Description: Test bind array with conflicting address
+ * Functionality: bind exec_queues and page table updates
+ * Test category: functionality test
+ *
+ * SUBTEST: bind-no-array-conflict
+ * Description: Test binding with conflicting address
+ * Functionality: bind and page table updates
+ * Test category: functionality test
+ *
+ * SUBTEST: bind-array-conflict-error-inject
+ * Description: Test bind array with conflicting address plus error injection
+ * Functionality: bind exec_queues and page table updates error paths
+ * Test category: functionality test
+ */
+static void
+test_bind_array_conflict(int fd, struct drm_xe_engine_class_instance *eci,
+ bool no_array, bool error_inject)
+{
+ uint32_t vm;
+ uint64_t addr = 0x1a00000;
+ struct drm_xe_sync sync[2] = {
+ { .type = DRM_XE_SYNC_TYPE_SYNCOBJ, .flags = DRM_XE_SYNC_FLAG_SIGNAL, },
+ { .type = DRM_XE_SYNC_TYPE_SYNCOBJ, .flags = DRM_XE_SYNC_FLAG_SIGNAL, },
+ };
+ struct drm_xe_exec exec = {
+ .num_batch_buffer = 1,
+ .syncs = to_user_pointer(sync),
+ };
+ uint32_t exec_queue;
+#define BIND_ARRAY_CONFLICT_NUM_BINDS 4
+ struct drm_xe_vm_bind_op bind_ops[BIND_ARRAY_CONFLICT_NUM_BINDS] = { };
+#define ONE_MB 0x100000
+ size_t bo_size = 8 * ONE_MB;
+ uint32_t bo = 0, bo2 = 0;
+ void *map, *map2 = NULL;
+ struct {
+ uint32_t batch[16];
+ uint64_t pad;
+ uint32_t data;
+ } *data = NULL;
+ const struct binds {
+ uint64_t size;
+ uint64_t offset;
+ uint32_t op;
+ } bind_args[] = {
+ { ONE_MB, 0, DRM_XE_VM_BIND_OP_MAP },
+ { 2 * ONE_MB, ONE_MB, DRM_XE_VM_BIND_OP_MAP },
+ { 3 * ONE_MB, 3 * ONE_MB, DRM_XE_VM_BIND_OP_MAP },
+ { 4 * ONE_MB, ONE_MB, DRM_XE_VM_BIND_OP_UNMAP },
+ };
+ const struct execs {
+ uint64_t offset;
+ } exec_args[] = {
+ { 0 },
+ { ONE_MB / 2 },
+ { ONE_MB / 4 },
+ { 5 * ONE_MB },
+ };
+ int i, b, n_execs = 4;
+
+ vm = xe_vm_create(fd, 0, 0);
+
+ bo = xe_bo_create(fd, vm, bo_size,
+ vram_if_possible(fd, eci->gt_id),
+ DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM);
+ map = xe_bo_map(fd, bo, bo_size);
+
+ exec_queue = xe_exec_queue_create(fd, vm, eci, 0);
+
+ /* Map some memory that will be over written */
+ if (error_inject) {
+ bo2 = xe_bo_create(fd, vm, bo_size,
+ vram_if_possible(fd, eci->gt_id),
+ DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM);
+ map2 = xe_bo_map(fd, bo2, bo_size);
+
+ i = 0;
+ sync[0].handle = syncobj_create(fd, 0);
+ xe_vm_bind_async(fd, vm, 0, bo2, bind_args[i].offset,
+ addr + bind_args[i].offset, bind_args[i].size,
+ sync, 1);
+ {
+ uint64_t batch_offset = (char *)&data[i].batch - (char *)data;
+ uint64_t batch_addr = addr + exec_args[i].offset + batch_offset;
+ uint64_t sdi_offset = (char *)&data[i].data - (char *)data;
+ uint64_t sdi_addr = addr + exec_args[i].offset + sdi_offset;
+ data = map2 + exec_args[i].offset;
+
+ b = 0;
+ data[i].batch[b++] = MI_STORE_DWORD_IMM_GEN4;
+ data[i].batch[b++] = sdi_addr;
+ data[i].batch[b++] = sdi_addr >> 32;
+ data[i].batch[b++] = 0xc0ffee;
+ data[i].batch[b++] = MI_BATCH_BUFFER_END;
+ igt_assert(b <= ARRAY_SIZE(data[i].batch));
+
+ sync[0].flags &= ~DRM_XE_SYNC_FLAG_SIGNAL;
+ sync[1].flags |= DRM_XE_SYNC_FLAG_SIGNAL;
+ sync[1].handle = syncobj_create(fd, 0);
+ exec.num_syncs = 2;
+
+ exec.exec_queue_id = exec_queue;
+ exec.address = batch_addr;
+ xe_exec(fd, &exec);
+ }
+
+ igt_assert(syncobj_wait(fd, &sync[0].handle, 1, INT64_MAX, 0, NULL));
+ igt_assert(syncobj_wait(fd, &sync[1].handle, 1, INT64_MAX, 0, NULL));
+
+ syncobj_reset(fd, &sync[0].handle, 1);
+ syncobj_destroy(fd, sync[1].handle);
+
+ data = map2 + exec_args[i].offset;
+ igt_assert_eq(data[i].data, 0xc0ffee);
+ }
+
+ if (no_array) {
+ sync[0].handle = syncobj_create(fd, 0);
+ for (i = 0; i < BIND_ARRAY_CONFLICT_NUM_BINDS; ++i) {
+ if (bind_args[i].op == DRM_XE_VM_BIND_OP_MAP)
+ xe_vm_bind_async(fd, vm, 0, bo,
+ bind_args[i].offset,
+ addr + bind_args[i].offset,
+ bind_args[i].size,
+ sync, 1);
+ else
+ xe_vm_unbind_async(fd, vm, 0,
+ bind_args[i].offset,
+ addr + bind_args[i].offset,
+ bind_args[i].size,
+ sync, 1);
+ }
+ } else {
+ for (i = 0; i < BIND_ARRAY_CONFLICT_NUM_BINDS; ++i) {
+ bind_ops[i].obj = bind_args[i].op == DRM_XE_VM_BIND_OP_MAP ?
+ bo : 0;
+ bind_ops[i].obj_offset = bind_args[i].offset;
+ bind_ops[i].range = bind_args[i].size;
+ bind_ops[i].addr = addr + bind_args[i].offset;
+ bind_ops[i].op = bind_args[i].op;
+ bind_ops[i].flags = 0;
+ bind_ops[i].prefetch_mem_region_instance = 0;
+ bind_ops[i].pat_index = intel_get_pat_idx_wb(fd);
+ bind_ops[i].reserved[0] = 0;
+ bind_ops[i].reserved[1] = 0;
+ }
+
+ if (error_inject) {
+ sync[0].flags |= DRM_XE_SYNC_FLAG_SIGNAL;
+ bind_ops[BIND_ARRAY_CONFLICT_NUM_BINDS - 1].flags |=
+ 0x1 << 31;
+ xe_vm_bind_array_err(fd, vm, 0, bind_ops,
+ BIND_ARRAY_CONFLICT_NUM_BINDS,
+ sync, 1, -ENOSPC);
+ bind_ops[BIND_ARRAY_CONFLICT_NUM_BINDS - 1].flags &=
+ ~(0x1 << 31);
+
+ /* Verify existing mappings still works */
+ i = 1;
+ {
+ uint64_t batch_offset = (char *)&data[i].batch - (char *)data;
+ uint64_t batch_addr = addr + exec_args[i].offset + batch_offset;
+ uint64_t sdi_offset = (char *)&data[i].data - (char *)data;
+ uint64_t sdi_addr = addr + exec_args[i].offset + sdi_offset;
+ data = map2 + exec_args[i].offset;
+
+ b = 0;
+ data[i].batch[b++] = MI_STORE_DWORD_IMM_GEN4;
+ data[i].batch[b++] = sdi_addr;
+ data[i].batch[b++] = sdi_addr >> 32;
+ data[i].batch[b++] = 0xc0ffee;
+ data[i].batch[b++] = MI_BATCH_BUFFER_END;
+ igt_assert(b <= ARRAY_SIZE(data[i].batch));
+
+ sync[1].flags |= DRM_XE_SYNC_FLAG_SIGNAL;
+ sync[1].handle = syncobj_create(fd, 0);
+ exec.num_syncs = 2;
+
+ exec.exec_queue_id = exec_queue;
+ exec.address = batch_addr;
+ xe_exec(fd, &exec);
+ }
+
+ igt_assert(syncobj_wait(fd, &sync[0].handle, 1, INT64_MAX, 0, NULL));
+ igt_assert(syncobj_wait(fd, &sync[1].handle, 1, INT64_MAX, 0, NULL));
+
+ syncobj_destroy(fd, sync[0].handle);
+ syncobj_destroy(fd, sync[1].handle);
+
+ data = map2 + exec_args[i].offset;
+ igt_assert_eq(data[i].data, 0xc0ffee);
+ sync[0].flags |= DRM_XE_SYNC_FLAG_SIGNAL;
+ }
+ sync[0].handle = syncobj_create(fd, 0);
+ xe_vm_bind_array(fd, vm, 0, bind_ops,
+ BIND_ARRAY_CONFLICT_NUM_BINDS, sync, 1);
+ }
+
+ for (i = 0; i < n_execs; i++) {
+ uint64_t batch_offset = (char *)&data[i].batch - (char *)data;
+ uint64_t batch_addr = addr + exec_args[i].offset + batch_offset;
+ uint64_t sdi_offset = (char *)&data[i].data - (char *)data;
+ uint64_t sdi_addr = addr + exec_args[i].offset + sdi_offset;
+ data = map + exec_args[i].offset;
+
+ b = 0;
+ data[i].batch[b++] = MI_STORE_DWORD_IMM_GEN4;
+ data[i].batch[b++] = sdi_addr;
+ data[i].batch[b++] = sdi_addr >> 32;
+ data[i].batch[b++] = 0xc0ffee;
+ data[i].batch[b++] = MI_BATCH_BUFFER_END;
+ igt_assert(b <= ARRAY_SIZE(data[i].batch));
+
+ sync[0].flags &= ~DRM_XE_SYNC_FLAG_SIGNAL;
+ sync[1].flags |= DRM_XE_SYNC_FLAG_SIGNAL;
+ if (i == n_execs - 1) {
+ sync[1].handle = syncobj_create(fd, 0);
+ exec.num_syncs = 2;
+ } else {
+ exec.num_syncs = 1;
+ }
+
+ exec.exec_queue_id = exec_queue;
+ exec.address = batch_addr;
+ xe_exec(fd, &exec);
+ }
+
+ igt_assert(syncobj_wait(fd, &sync[0].handle, 1, INT64_MAX, 0, NULL));
+ igt_assert(syncobj_wait(fd, &sync[1].handle, 1, INT64_MAX, 0, NULL));
+
+ for (i = 0; i < n_execs; i++) {
+ data = map + exec_args[i].offset;
+ igt_assert_eq(data[i].data, 0xc0ffee);
+ }
+
+ syncobj_destroy(fd, sync[0].handle);
+ syncobj_destroy(fd, sync[1].handle);
+ xe_exec_queue_destroy(fd, exec_queue);
+
+ munmap(map, bo_size);
+ if (map2)
+ munmap(map, bo_size);
+ gem_close(fd, bo);
+ if (bo2)
+ gem_close(fd, bo2);
+ xe_vm_destroy(fd, vm);
+}
+
+
#define LARGE_BIND_FLAG_MISALIGNED (0x1 << 0)
#define LARGE_BIND_FLAG_SPLIT (0x1 << 1)
#define LARGE_BIND_FLAG_USERPTR (0x1 << 2)
@@ -2144,6 +2413,18 @@ igt_main
test_bind_array(fd, hwe, 16,
BIND_ARRAY_BIND_EXEC_QUEUE_FLAG);
+ igt_subtest("bind-array-conflict")
+ xe_for_each_engine(fd, hwe)
+ test_bind_array_conflict(fd, hwe, false, false);
+
+ igt_subtest("bind-no-array-conflict")
+ xe_for_each_engine(fd, hwe)
+ test_bind_array_conflict(fd, hwe, true, false);
+
+ igt_subtest("bind-array-conflict-error-inject")
+ xe_for_each_engine(fd, hwe)
+ test_bind_array_conflict(fd, hwe, false, true);
+
for (bind_size = 0x1ull << 21; bind_size <= 0x1ull << 31;
bind_size = bind_size << 1) {
igt_subtest_f("large-binds-%lld",
--
2.34.1
More information about the igt-dev
mailing list