Hi Andi,
For next time, I find it difficult to follow all these variables,
it's easier to read
man = ttm_manager_type(priv->ttm_dev, TTM_PL_SYSTEM);
than
mem_type = TTM_PL_SYSTEM;
...
...
...
man = ttm_manager_type(priv->ttm_dev, mem_type);
+ bo = ttm_bo_kunit_init(test, test->priv, size);
+ bo->type = bo_type;
same here... the bo_type variable is not giving any value.
bo->type = ttm_bo_type_device;
is way more readable. You keep doing this all the way and I need
to check everytime what's the value in the declaration :-)
The idea was that I'd provide these as parameters, but I limited the
scope of my tests and stick to set values. Also, in some cases, I
defined them because I keep checking for them in assertions, and didn't
want to repeat myself.
I'm not going to comment on this anymore.
+ if (params->with_ttm) {
+ old_tt = priv->ttm_dev->funcs->ttm_tt_create(bo, 0);
+ ttm_pool_alloc(&priv->ttm_dev->pool, old_tt, &ctx);
+ bo->ttm = old_tt;
+ }
+
+ err = ttm_resource_alloc(bo, place, &bo->resource);
+ KUNIT_EXPECT_EQ(test, err, 0);
+ KUNIT_ASSERT_EQ(test, man->usage, size);
+
+ placement = kunit_kzalloc(test, sizeof(*placement), GFP_KERNEL);
+ KUNIT_ASSERT_NOT_NULL(test, placement);
+
+ ttm_bo_reserve(bo, false, false, NULL);
+ err = ttm_bo_validate(bo, placement, &ctx);
+ dma_resv_unlock(bo->base.resv);
why don't you use here ttm_bo_unreserve()?
That's a good question! I think that unreserve would work here as well.
(...)
+static void ttm_bo_validate_move_fence_signaled(struct kunit *test)
+{
+ struct ttm_test_devices *priv = test->priv;
+ struct ttm_buffer_object *bo;
+ struct ttm_place *place;
+ struct ttm_placement *placement;
+ struct ttm_resource_manager *man;
+ enum ttm_bo_type bo_type = ttm_bo_type_device;
+ uint32_t mem_type = TTM_PL_SYSTEM;
+ struct ttm_operation_ctx ctx = { };
+ uint32_t size = ALIGN(BO_SIZE, PAGE_SIZE);
+ int err;
+
+ man = ttm_manager_type(priv->ttm_dev, mem_type);
+ spin_lock_init(&fence_lock);
where are we using the fence_lock here?
Argh, it's a copy-paste mistake, sorry. We don't need it, as we use a
stub fence in man->move. I will delete that.
+ man->move = dma_fence_get_stub();
+
+ bo = ttm_bo_kunit_init(test, test->priv, size);
+ bo->type = bo_type;
+ place = ttm_place_kunit_init(test, mem_type, 0);
+ placement = ttm_placement_kunit_init(test, place, 1, NULL, 0);
+
+ ttm_bo_reserve(bo, false, false, NULL);
+ err = ttm_bo_validate(bo, placement, &ctx);
+ dma_resv_unlock(bo->base.resv);
+
+ KUNIT_EXPECT_EQ(test, err, 0);
+ KUNIT_EXPECT_EQ(test, bo->resource->mem_type, mem_type);
+ KUNIT_EXPECT_EQ(test, ctx.bytes_moved, size);
Do we want to check also bo->ttm here?
Hmm, I think that I covered that case with "normal" testing of
ttm_bo_validate, and kept this one minimal. Here, I'm just making sure
that a signaled move fence doesn't get into a way of BO validation.
+ ttm_bo_put(bo);
+ dma_fence_put(man->move);
+}
+
+static const struct ttm_bo_validate_test_case ttm_bo_validate_wait_cases[] = {
+ {
+ .description = "Waits for GPU",
+ .no_gpu_wait = false,
+ },
+ {
+ .description = "Tries to lock straight away",
+ .no_gpu_wait = true,
+ },
+};
+
+KUNIT_ARRAY_PARAM(ttm_bo_validate_wait, ttm_bo_validate_wait_cases,
+ ttm_bo_validate_case_desc);
+
+static int threaded_fence_signal(void *arg)
+{
+ struct dma_fence *fence = arg;
+ int err;
+
+ msleep(20);
+ err = dma_fence_signal(fence);
+
+ return err;
if you do here "return dma_fence_signal(fence);" you don't need
for err.
That is true!
Not a binding review, though, your choice.
You spotted a couple of things I can improve, so I plan to include them
in the next version, after getting more review comments.
All the best,
Karolina
