Nvv4l2decoder blocks indefinitely

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson / GPU) GPU T4, Titan V
• DeepStream Version 6.2, 6.1.1
• JetPack Version (valid for Jetson only)
• TensorRT Version 8.5.2
• NVIDIA GPU Driver Version (valid for GPU only)
525.85.12
• Issue Type( questions, new requirements, bugs)

Some possibly defective video files (observed for h264, vp9) cause nvv4l2decoder to block indefinitely.

Reproduce like this:

gst-launch-1.0 uridecodebin uri=https://github.com/webmproject/libwebm/blob/libwebm-1.0.0.29/testing/testdata/accurate_cluster_duration_last_frame.webm?raw=true ! fakesink

This blocks indefinitely, and when pressing Ctrl+C, one sees that setting the pipeline to NULL also blocks. This is a huge problem in our application, as it can effectively block the application.

One can reduce the problem to

gst-launch-1.0 filesrc location=testdata/accurate_cluster_duration_last_frame.webm ! typefind ! matroskademux ! multiqueue ! nvv4l2decoder ! fakesink

Replacing nvv4l2decoder by vp9dec gives an error, but at least does not block.

Attaching gdb to the blocked process shows the following:

(gdb) thread apply all bt

Thread 8 (Thread 0x7f1cb37fe700 (LWP 5248)):
#0  futex_wait_cancelable (private=<optimized out>, expected=0, futex_word=0x7f1c90004810) at ../sysdeps/nptl/futex-internal.h:183
#1  __pthread_cond_wait_common (abstime=0x0, clockid=0, mutex=0x7f1c900047c0, cond=0x7f1c900047e8) at pthread_cond_wait.c:508
#2  __pthread_cond_wait (cond=0x7f1c900047e8, mutex=0x7f1c900047c0) at pthread_cond_wait.c:647
#3  0x00007f1cc632e093 in NvOsSemaphoreWait(NvOsSemaphoreRec*) () at ///opt/nvidia/deepstream/deepstream-6.2/lib/libcuvidv4l2.so
#4  0x00007f1cc633d057 in cuvidv4l2_dec_thread_func(void*) () at ///opt/nvidia/deepstream/deepstream-6.2/lib/libcuvidv4l2.so
#5  0x00007f1cc632deb0 in  () at ///opt/nvidia/deepstream/deepstream-6.2/lib/libcuvidv4l2.so
#6  0x00007f1cc80fc609 in start_thread (arg=<optimized out>) at pthread_create.c:477
#7  0x00007f1cc8021133 in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95

It appears nvv4l2decoder is deadlocked.

The libwebm repository contains more files which cause this behavior (GitHub - webmproject/libwebm: Mirror only. Please do not send pull requests., testing/testdata)

3

We’ll check it. But there may be something special about these videos. It cannot be played by other player too(VLC, FFMPEG, Google Chrome). But we can play the bbb_480p_vp9_opus_1second.webm video. So you can also ask the author of the libwebm repository how to decode these videos correctly.

4

Thanks. The problem is not so much that the videos cannot be decoded. But the nvv4l2decoder cannot switch to state NULL and blocks. This should never happen even on invalid data.

6

I can repro the problems with the video you attached. We’ll check and fix it in the future version.

7

Hi @crops.limits-09 , we have fixed this issue in the latest version. You can try to upgrade DeepStream Version and check that. Thanks