On 08/08/2023 07:56, Tomi Valkeinen wrote:
> On 02/08/2023 16:54, Hans Verkuil wrote:
>> On 20/07/2023 09:50, Tomi Valkeinen wrote:
>>> Add a simple test for VIDIOC_SUBDEV_G_ROUTING/VIDIOC_SUBDEV_S_ROUTING.
>>>
>>> We can't (at least at the moment) really know here what kind of routings
>>> the driver would accept, but we can test a VIDIOC_SUBDEV_G_ROUTING call,
>>> followed by a VIDIOC_SUBDEV_S_ROUTING call with the routing we
>>> received.
>>>
>>> Additionally, we can check that the returned pads and flags look fine,
>>> and also that setting obviously invalid routing will fail.
>>>
>>> Signed-off-by: Tomi Valkeinen <tomi.valkeinen@xxxxxxxxxxxxxxxx>
>>> ---
>>> utils/v4l2-compliance/v4l2-compliance.cpp | 12 ++++
>>> utils/v4l2-compliance/v4l2-compliance.h | 1 +
>>> utils/v4l2-compliance/v4l2-test-subdevs.cpp | 74 +++++++++++++++++++++
>>> 3 files changed, 87 insertions(+)
>>>
>>> diff --git a/utils/v4l2-compliance/v4l2-compliance.cpp b/utils/v4l2-compliance/v4l2-compliance.cpp
>>> index e8359b2f..4b232314 100644
>>> --- a/utils/v4l2-compliance/v4l2-compliance.cpp
>>> +++ b/utils/v4l2-compliance/v4l2-compliance.cpp
>>> @@ -1249,6 +1249,18 @@ void testNode(struct node &node, struct node &node_m2m_cap, struct node &expbuf_
>>> node.is_passthrough_subdev = has_source && has_sink;
>>> if (has_routes) {
>>> + printf("Sub-Device routing ioctls:\n");
>>> +
>>> + for (unsigned which = V4L2_SUBDEV_FORMAT_TRY;
>>> + which <= V4L2_SUBDEV_FORMAT_ACTIVE; which++) {
>>> +
>>> + printf("\ttest %s VIDIOC_SUBDEV_G_ROUTING/VIDIOC_SUBDEV_S_ROUTING: %s\n",
>>> + which ? "Active" : "Try",
>>> + ok(testSubDevRouting(&node, which)));
>>> + }
>>> +
>>> + printf("\n");
>>> +
>>> for (unsigned which = V4L2_SUBDEV_FORMAT_TRY;
>>> which <= V4L2_SUBDEV_FORMAT_ACTIVE; which++) {
>>> diff --git a/utils/v4l2-compliance/v4l2-compliance.h b/utils/v4l2-compliance/v4l2-compliance.h
>>> index 0cd43980..35b2274b 100644
>>> --- a/utils/v4l2-compliance/v4l2-compliance.h
>>> +++ b/utils/v4l2-compliance/v4l2-compliance.h
>>> @@ -375,6 +375,7 @@ int testSubDevEnum(struct node *node, unsigned which, unsigned pad, unsigned str
>>> int testSubDevFormat(struct node *node, unsigned which, unsigned pad, unsigned stream);
>>> int testSubDevSelection(struct node *node, unsigned which, unsigned pad, unsigned stream);
>>> int testSubDevFrameInterval(struct node *node, unsigned pad, unsigned stream);
>>> +int testSubDevRouting(struct node *node, unsigned which);
>>> // Buffer ioctl tests
>>> int testReqBufs(struct node *node);
>>> diff --git a/utils/v4l2-compliance/v4l2-test-subdevs.cpp b/utils/v4l2-compliance/v4l2-test-subdevs.cpp
>>> index 5545b0dc..e59d67f7 100644
>>> --- a/utils/v4l2-compliance/v4l2-test-subdevs.cpp
>>> +++ b/utils/v4l2-compliance/v4l2-test-subdevs.cpp
>>> @@ -551,3 +551,77 @@ int testSubDevSelection(struct node *node, unsigned which, unsigned pad, unsigne
>>> return have_sel ? 0 : ENOTTY;
>>> }
>>> +
>>> +int testSubDevRouting(struct node *node, unsigned which)
>>> +{
>>> + const uint32_t all_route_flags_mask = V4L2_SUBDEV_ROUTE_FL_ACTIVE;
>>> + struct v4l2_subdev_routing routing = {};
>>> + struct v4l2_subdev_route routes[NUM_ROUTES_MAX] = {};
>>> + unsigned int i;
>>> + int ret;
>>> +
>>> + routing.which = which;
>>> + routing.routes = (__u64)&routes;
>>> + routing.num_routes = 0;
>>> + memset(routing.reserved, 0xff, sizeof(routing.reserved));
>>> +
>>> + /*
>>> + * First test that G_ROUTING either returns success, or ENOSPC and
>>> + * updates num_routes.
>>> + */
>>> +
>>> + ret = doioctl(node, VIDIOC_SUBDEV_G_ROUTING, &routing);
>>> + fail_on_test(ret && ret != ENOSPC);
>>> + fail_on_test(ret == ENOSPC && routing.num_routes == 0);
>>> + fail_on_test(check_0(routing.reserved, sizeof(routing.reserved)));
>>> +
>>> + if (routing.num_routes)
>>> + return 0;
>>
>> Shouldn't this be 'if (!routing.num_routes)'?
>
> Yes...
>
>>> +
>>> + /* Then get the actual routes */
>>> +
>>> + routing.num_routes = NUM_ROUTES_MAX;
>>> + fail_on_test(doioctl(node, VIDIOC_SUBDEV_G_ROUTING, &routing));
>>
>> I assume that num_routes is always updated to the actual number of routes, right?
>
> If VIDIOC_SUBDEV_G_ROUTING succeeds, yes, num_routes is updated.
>
>> That's not actually explained clearly in the spec. It says that if the app provided
>> num_routes is too small, then it is updated, but it says nothing about what happens
>> if the app provided value is too large.
>
> Ok. I need to update the doc.
>
>> Assuming I am right, then I would rewrite this code as follows:
>>
>> __u32 num_routes = routing.num_routes;
>> routing.num_routes = num_routes + 1;
>> fail_on_test(doioctl(node, VIDIOC_SUBDEV_G_ROUTING, &routing));
>> fail_on_test(routing.num_routes != num_routes);
>
> Yes, I think this looks fine.
>
> Btw, you use __u32 above. Is there any style guide for these? I used uint32_t in this function for another variable, and I'd use it here too.
It is derived from a kernel structure, and the kernel API uses __u32. So I prefer to
use __u32 for such things. I personally think that is good practice since it helps
indicate that it is a kernel-API-related variable.
I also like it because it is shorter than uint32_t :-)
Regards,
Hans
>
>>> +
>>> + /* Check the validity of route pads and flags */
>>> +
>>> + if (node->pads) {
>>> + for (i = 0; i < routing.num_routes; ++i) {
>>> + const struct v4l2_subdev_route *route = &routes[i];
>>> + const struct media_pad_desc *sink;
>>> + const struct media_pad_desc *source;
>>> +
>>> + fail_on_test(route->sink_pad >= node->entity.pads);
>>> + fail_on_test(route->source_pad >= node->entity.pads);
>>> +
>>> + sink = &node->pads[route->sink_pad];
>>> + source = &node->pads[route->source_pad];
>>> +
>>> + fail_on_test(!(sink->flags & MEDIA_PAD_FL_SINK));
>>> + fail_on_test(!(source->flags & MEDIA_PAD_FL_SOURCE));
>>> + fail_on_test(route->flags & ~all_route_flags_mask);
>>> + }
>>> + }
>>> +
>>> + /* Set the same routes back, which should always succeed. */
>>> +
>>> + memset(routing.reserved, 0xff, sizeof(routing.reserved));
>>> + fail_on_test(doioctl(node, VIDIOC_SUBDEV_S_ROUTING, &routing));
>>> + fail_on_test(check_0(routing.reserved, sizeof(routing.reserved)));
>>> +
>>> + /* Test setting invalid pads */
>>> +
>>> + if (node->pads) {
>>> + for (i = 0; i < routing.num_routes; ++i) {
>>> + struct v4l2_subdev_route *route = &routes[i];
>>> +
>>> + route->sink_pad = node->entity.pads + 1;
>>> + }
>>> +
>>> + memset(routing.reserved, 0xff, sizeof(routing.reserved));
>>> + fail_on_test(doioctl(node, VIDIOC_SUBDEV_S_ROUTING, &routing) != EINVAL);
>>> + fail_on_test(check_0(routing.reserved, sizeof(routing.reserved)));
>
> After fixing the num_routes check, I noticed that this one is broken too. If S_ROUTING fails early enough, the reserved field won't get cleared, so we can't have this check here.
>
>>> + }
>>> +
>>> + return 0;
>>> +}
>>
>> Regards,
>>
>> Hans
>
