Re: [PATCH 07/13] drm/vkms: Allow to configure multiple planes

["José Expósito" <jose.exposito89@gmail.com>]

On Thu, Jan 30, 2025 at 02:48:19PM +0100, Louis Chauvet wrote:
> On 29/01/25 - 12:00, José Expósito wrote:
> > Add a list of planes to vkms_config and create as many planes as
> > configured during output initialization.
> > 
> > For backwards compatibility, add one primary plane and, if configured,
> > one cursor plane and NUM_OVERLAY_PLANES planes to the default
> > configuration.
> > 
> > Signed-off-by: Louis Chauvet <louis.chauvet@bootlin.com>
> > Signed-off-by: José Expósito <jose.exposito89@gmail.com>
> 
> Co-developped-by: Louis Chauvet <louis.chauvet@bootlin.com>
> Signed-off-by: Louis Chauvet <louis.chauvet@bootlin.com>
> Signed-off-by: José Expósito <jose.exposito89@gmail.com>
> 
> [...]
> 
> > diff --git a/drivers/gpu/drm/vkms/tests/vkms_config_test.c b/drivers/gpu/drm/vkms/tests/vkms_config_test.c
> 
> [...]
> 
> > +static void vkms_config_test_get_planes(struct kunit *test)
> > +{
> > +	struct vkms_config *config;
> > +	struct vkms_config_plane *plane_cfg1, *plane_cfg2;
> > +	struct vkms_config_plane **array;
> > +	size_t length;
> > +
> > +	config = vkms_config_create("test");
> > +	KUNIT_ASSERT_NOT_ERR_OR_NULL(test, config);
> > +
> > +	array = vkms_config_get_planes(config, &length);
> > +	KUNIT_ASSERT_EQ(test, length, 0);
> > +	KUNIT_ASSERT_NULL(test, array);
> > +
> > +	plane_cfg1 = vkms_config_add_plane(config);
> > +	array = vkms_config_get_planes(config, &length);
> > +	KUNIT_ASSERT_EQ(test, length, 1);
> > +	KUNIT_ASSERT_PTR_EQ(test, array[0], plane_cfg1);
> > +	kfree(array);
> > +
> > +	plane_cfg2 = vkms_config_add_plane(config);
> > +	array = vkms_config_get_planes(config, &length);
> > +	KUNIT_ASSERT_EQ(test, length, 2);
> > +	KUNIT_ASSERT_PTR_EQ(test, array[0], plane_cfg1);
> > +	KUNIT_ASSERT_PTR_EQ(test, array[1], plane_cfg2);
> > +	kfree(array);
> > +
> > +	vkms_config_destroy_plane(plane_cfg1);
> > +	array = vkms_config_get_planes(config, &length);
> > +	KUNIT_ASSERT_EQ(test, length, 1);
> > +	KUNIT_ASSERT_PTR_EQ(test, array[0], plane_cfg2);
> > +	kfree(array);
> > +
> > +	vkms_config_destroy(config);
> > +}
> 
> In this test I have the feeling that vkms_config_get_planes always returns 
> a predictable order. It is maybe trivial here, but I would prefer to shows 
> that the order is not stable, for example:
> 
> 	bool plane_cfg1_found = false;
> 	bool plane_cfg2_found = false;
> 
> 	vkms_config_for_each_plane(config, plane_cfg) {
> 		if (plane_cfg == plane_cfg1)
> 			plane_cfg1_found = true;
> 		else if (plane_cfg == plane_cfg2)
> 			plane_cfg2_found = true;
> 		else
> 			KUNIT_FAILS("Unexpected plane");
> 	}
> 
> 	KUNIT_ASSERT(test, plane_cfg1_found);
> 	KUNIT_ASSERT(test, plane_cfg2_found);
> 
> [...]
> 
> > +static void vkms_config_test_valid_plane_number(struct kunit *test)
> > +{
> > +	struct vkms_config *config;
> > +	struct vkms_config_plane *plane_cfg;
> > +	int n;
> > +
> > +	config = vkms_config_default_create(false, false, false);
> > +	KUNIT_ASSERT_NOT_ERR_OR_NULL(test, config);
> > +
> > +	/* Invalid: No planes */
> > +	plane_cfg = list_first_entry(&config->planes, typeof(*plane_cfg), link);
> > +	vkms_config_destroy_plane(plane_cfg);
> > +	KUNIT_EXPECT_FALSE(test, vkms_config_is_valid(config));
> > +
> > +	/* Invalid: Too many planes */
> > +	for (n = 0; n <= 32; n++)
> > +		vkms_config_add_plane(config);
> > +
> > +	KUNIT_EXPECT_FALSE(test, vkms_config_is_valid(config));
> > +
> > +	vkms_config_destroy(config);
> > +}
> 
> For this function, the naming is a bit strange, it says 
> "valid_plane_number", but you test only invalid plane number.

The reason for this naming is that it tests the valid_plane_number()
function called by vkms_config_is_valid(). The applies for the other
valid_* tests.

However, I don't mind changing its name to so it reflects the test
rather than the tested function.

Changed in v2.

> 
> Can you rename it to vkms_config_test_invalid_plane_number?
> 
> [...]
> 
> > diff --git a/drivers/gpu/drm/vkms/vkms_config.c b/drivers/gpu/drm/vkms/vkms_config.c
> 
> [...]
> 
> > +struct vkms_config_plane **vkms_config_get_planes(const struct vkms_config *config,
> > +						  size_t *out_length)
> > +{
> > +	struct vkms_config_plane **array;
> > +	struct vkms_config_plane *plane_cfg;
> > +	size_t length;
> > +	int n = 0;
> > +
> > +	length = list_count_nodes((struct list_head *)&config->planes);
> > +	if (length == 0) {
> > +		*out_length = length;
> > +		return NULL;
> > +	}
> > +
> > +	array = kmalloc_array(length, sizeof(*array), GFP_KERNEL);
> > +	if (!array)
> > +		return ERR_PTR(-ENOMEM);
> > +
> > +	list_for_each_entry(plane_cfg, &config->planes, link) {
> > +		array[n] = plane_cfg;
> > +		n++;
> > +	}
> > +
> > +	*out_length = length;
> > +	return array;
> > +}
> 
> To join the comment on the test, I am not a big fan of creating a new list 
> to return to the caller, for three reasons:
> - the caller needs to manage an other pointer;
> - the caller needs to understand that the content of the array is only 
>   valid if: the config is not freed, nobody else removed anything from the 
>   planes;
> - the caller may think this list always have the same order if he looks at 
>   the tests.
> 
> I would prefer a simple macro to do an iteration over the config->planes 
> list: (I did not test this macro, but you have this idea)
> 
> 	#define vkms_config_iter_plane(config, plane_cfg) \
> 		list_for_each_entry((plane_cfg), &(config).planes, link)
> 
> This way:
> - no new pointer to manage;
> - if one day we have concurency issue, we just have to protect config, not 
>   config+all the planes;
> - there is no expected order.
> 
> [...]
> 
> >  bool vkms_config_is_valid(struct vkms_config *config)
> >  {
> > +	if (!valid_plane_number(config))
> > +		return false;
> > +
> > +	if (!valid_plane_type(config))
> > +		return false;
> > +
> >  	return true;
> >  }
> 
> I really like the idea to split the validation function, way simpler!
> 
> [...]
> 
> > +void vkms_config_destroy_plane(struct vkms_config_plane *plane_cfg)
> > +{
> > +	list_del(&plane_cfg->link);
> > +	kfree(plane_cfg);
> > +}
> 
> I would prefer a "standard" function pair, i.e.: add/remove or 
> create/destroy, not add/destroy.
> 
> For me it should be create/destroy, you create the plane by using a 
> config, so it is clear it will be attached to it.
> 
> If you choose add/remove, you should explains in the documentation that 
> remove is also doing kfree.
> 
> [...]
> 
> > diff --git a/drivers/gpu/drm/vkms/vkms_output.c b/drivers/gpu/drm/vkms/vkms_output.c
> 
> [...]
> 
> > @@ -11,61 +11,63 @@ int vkms_output_init(struct vkms_device *vkmsdev)
> >  	struct vkms_connector *connector;
> >  	struct drm_encoder *encoder;
> >  	struct vkms_output *output;
> > -	struct vkms_plane *primary, *overlay, *cursor = NULL;
> > -	int ret;
> > +	struct vkms_plane *primary = NULL, *cursor = NULL;
> > +	struct vkms_config_plane **plane_cfgs = NULL;
> > +	size_t n_planes;
> > +	int ret = 0;
> >  	int writeback;
> >  	unsigned int n;
> 
> I think it could be interesting to have a vkms_config_is_valid call here. 
> It will avoid raising DRM errors or create unexpected devices.
> 
> It will also garantee in a later patch that 
> vkms_config_crtc_get_primary_plane is a valid pointer.
> 
> > -	/*
> > -	 * Initialize used plane. One primary plane is required to perform the composition.
> > -	 *
> > -	 * The overlay and cursor planes are not mandatory, but can be used to perform complex
> > -	 * composition.
> > -	 */
> > -	primary = vkms_plane_init(vkmsdev, DRM_PLANE_TYPE_PRIMARY);
> > -	if (IS_ERR(primary))
> > -		return PTR_ERR(primary);
> > +	plane_cfgs = vkms_config_get_planes(vkmsdev->config, &n_planes);
> > +	if (IS_ERR(plane_cfgs))
> > +		return PTR_ERR(plane_cfgs);
> 
> If you agree on the iterator implementation, this code could be simplified 
> a lot.
> 
> > -	if (vkmsdev->config->cursor) {
> > -		cursor = vkms_plane_init(vkmsdev, DRM_PLANE_TYPE_CURSOR);
> > -		if (IS_ERR(cursor))
> > -			return PTR_ERR(cursor);
> > +	for (n = 0; n < n_planes; n++) {
> > +		struct vkms_config_plane *plane_cfg;
> > +		enum drm_plane_type type;
> > +
> > +		plane_cfg = plane_cfgs[n];
> > +		type = vkms_config_plane_get_type(plane_cfg);
> > +
> > +		plane_cfg->plane = vkms_plane_init(vkmsdev, type);
> 
> Can we pass plane_cfg in vkms_plane_init? This way we don't have to 
> touch vkms_output_init when adding new vkms_config_plane members.

While it'll be required once we allow to configure more parameters, I don't
think we need it right now. To keep things as simple as possible, I'd prefer to
delay it until required.

Thanks,
Jose

> > +		if (IS_ERR(plane_cfg->plane)) {
> > +			DRM_DEV_ERROR(dev->dev, "Failed to init vkms plane\n");
> > +			ret = PTR_ERR(plane_cfg->plane);
> > +			goto err_free;
> > +		}
> > +
> > +		if (type == DRM_PLANE_TYPE_PRIMARY)
> > +			primary = plane_cfg->plane;
> > +		else if (type == DRM_PLANE_TYPE_CURSOR)
> > +			cursor = plane_cfg->plane;
> >  	}
> 
> [...]