Corr_err: discarding frame, flags: 0, err_data: 0 while streaming 4K@60 on Jetson Orin NX (L4T 36.5)

Hi,

I am using Jetson Orin NX with JetPack/L4T 36.5 and streaming a 3840×2160 @ 60 FPS UYVY video through a CSI-2 bridge.

The video stream is stable, and I can receive/display clear frames. However, the kernel continuously reports the following message during streaming:

tegra-camrtc-capture-vi tegra-capture-vi: corr_err: discarding frame 0, flags: 0, err_data: 0

The trace_pipe output shows continuous SOF/EOF events, and there are no CHANSEL_SHORT_FRAME, CHANSEL_LONG_FRAME, or CHANSEL_NOMATCH errors like below.

dmesg.txt (2.6 KB)

trace.txt (10.9 KB)

Could you please help us understand:

  1. What typically causes this corr_err: discarding frame message when flags and err_data are both 0?
  2. Is this related to VI timing or device tree configuration (such as pix_clk_hz, line_length, or frame rate)?
  3. Is there a recommended way to eliminate this message if the video stream itself is functioning correctly?

I have attached the kernel dmesg log and the trace_pipe output for reference.

Any guidance would be greatly appreciated.

Thank you.

hello kamalesh1,

>> Q1
discarding frame emitted when descr->status.status != CAPTURE_STATUS_SUCCESS but the channel is not marked unrecoverably as fatal error.
the driver calls that a correctable error, marks frame_err, and releases the buffer as an error rather than returning it as a good frame.

>> Q2
>> Q3
per your logs, it seems discarding frame has reported occasionally. it might due to consumer is too slow, you may give it a try to lower the frame-rate for testing.

Hello Chang,

Thank you for the clarification regarding CAPTURE_STATUS_SUCCESS and the correctable error handling.

I performed an additional test by reducing the stream to 3840×2160 @ 30 FPS, and in that configuration I do not observe the “corr_err: discarding frame” messages. The issue appears only when streaming at 3840×2160 @ 60 FPS.

Given that 4K30 works cleanly while 4K60 occasionally reports discarded frames, could this indicate that the VI/capture pipeline is unable to process the incoming frames at 4K60 in real time (for example due to bandwidth, scheduling, or consumer latency limitations)?

For reference:

  • Platform: Jetson Orin NX
  • L4T version: 36.5
  • Input format: UYVY 4:2:2
  • Resolution: 3840×2160
  • 4K30: No discarded-frame messages observed
  • 4K60: Occasional “corr_err: discarding frame” messages observed

Also, are there any recommended VI/CSI/EMC clock settings, performance modes (for example jetson_clocks), or device-tree bandwidth parameters that should be tuned specifically for sustained 4K60 UYVY capture on Orin NX?

Any guidance on whether this behavior is expected at 4K60, or how to eliminate the discarded-frame events, would be greatly appreciated.

Thank you.

Regards,
Kamal

hello kamalesh1,

we may check whether 60-fps failing because the consumer is too slow, or because the 60-fps CSI/VI/bandwidth configuration has insufficient margin.

please have another test to capture-only, without display/encode/file-write.
let’s check the discarding frame is caused by the consumer pipeline or by the CSI/VI input path.
you may see-also developer guide,
for instance, Applications Using V4L2 IOCTL Directly.
$ v4l2-ctl -d /dev/video0 --set-fmt-video=width=1920,height=1080,pixelformat=UYVY --set-ctrl bypass_mode=0 --stream-mmap --stream-count=300
or.. Applications Using GStreamer with V4L2 Source Plugin.
$ gst-launch-1.0 v4l2src device=/dev/video1 num-buffers=300 ! 'video/x-raw, width=1920, height=1080, framerate=30/1, format=YUY2' ! fpsdisplaysink text-overlay=0 name=sink_0 video-sink=fakesink sync=0 -v

furthermore,
you should configure the system for testing in max performance mode.
$ sudo nvpmodel -q
$ sudo nvpmodel -m <highest supported mode for the module>
$ sudo /usr/bin/jetson_clocks

Hi,

Thank you for your suggestions.

I have already performed the recommended tests with the system configured for maximum performance by running:

Since my source outputs only 3840×2160 @ 60 fps, I modified the test commands accordingly. Specifically:

  • I used 3840×2160 @ 60 fps instead of 1920×1080 @ 60 fps.
  • In the GStreamer pipeline, I used UYVY instead of YUY2, as the camera outputs UYVY format.

Even with these changes and the system running in maximum performance mode, I still observe the following kernel message:

tegra-camrtc-capture-vi tegra-capture-vi: corr_err: discarding frame 0, flags: 0, err_data: 0

One observation is that when I configure the source to output 3840×2160 @ 30 fps, I do not see any discarding frame messages. The issue only occurs at 3840×2160 @ 60 fps.

Could you please clarify whether this indicates that the VI/CSI pipeline is unable to process the incoming 4K60 stream in time, or whether there is a bandwidth/timing configuration issue in the device tree or camera driver?

Any guidance on what additional parameters or configurations should be checked for stable 4K60 capture would be greatly appreciated.

Thank you.

Best regards,

Kamal

hello kamalesh1,

just for confirmation, did you also update the pixel clock rate, pix_clk_hz?
in general, 4K60 has 2x pixel clock rate with 4K30.

Hi Chang,

Yes, I have given the pixel clock to 594 MHZ only.

Thanks.

hello kamalesh1,

as you can see… Sensor Pixel Clock, pixel_clk_hz = sensor output size × frame rate
you should have different pixel clock rate for 30/60-fps sensor modes.

Hi,

Yes, I have configured different pixel_clk_hz values for the two sensor modes as recommended.

  • 3840 × 2160 @ 30 fps: pixel_clk_hz = 248000000
  • 3840 × 2160 @ 60 fps: pixel_clk_hz = 497000000

I have tested with these corresponding pixel clock values, but I still observe the same corr_err: discarding frame messages at 4K@60 fps. There was no noticeable improvement.

Thank you.

hello kamalesh1,

please verify the actual CSI output from the source side:

  1. Confirm the 4K60 sensor/bridge register table: PLL, HTS/VTS, frame length, line length, CSI lane count, and MIPI lane rate.
  2. Confirm the bridge is really outputting 3840x2160 UYVY at 60 fps with the expected blanking.
  3. Check whether DPHY deskew is enabled/sent at 4K60 if per-lane rate is above 1.5 Gbps.
  4. Check bridge/sensor status registers for CSI FIFO underrun/overrun, CRC/ECC, packet, or link errors.
  5. Try lower step rates such as 4K45/50/55. If errors begin at a threshold, that points to source-side lane-rate or timing margin.
  6. If possible, validate the MIPI output with a CSI analyzer or compare against another known-good receiver.

in the meanwhile,
you may checking VI tracing logs, also collect trace with rtcpu_nvcsi_intr lines for any suspect errors.
if PHY interrupts appear only at 4K60, that would support a sensor/bridge/link-side issue.

Hi Chang,

Thanks for the detailed checklist. Here’s where we stand on each point:

  1. PLL / HTS-VTS / lane config / lane rate — confirmed from bridge debug log:

    • CSI lanes: 4 (ucTxLaneNum:0x04)
    • Per-lane MIPI rate: ulPhyClk = 2495888 kbps ≈ 2.5 Gbps/lane
    • Htotal:4400, Hactive:3840, Hfp:176, Hbp:296, Hsw:88
    • Vtotal:2250, Vactive:2160, Vfp:8, Vbp:72, Vsw:10
    • Format: YUV422 8-bit (UYVY)
    • Frame rate check: pixel clock = Htotal x Vtotal x fps = 4400 x 2250 x 60 = 594,000,000 Hz. ulHalfPixClk:296986 x 2 ≈ 594 MHz, confirming true 60fps at this timing.
      Thanks.
  2. Bridge output confirmed as 3840x2160 UYVY @ 60fps with the blanking above — matches spec.

  3. Whenever the per-lane rate is pushed above 1.5 Gbps, we get the same error, regardless of exact resolution/rate.

  4. Bridge/sensor error/status registers — checked sensor registers, no errors reported; MIPI signals verified clean with a probe.

  5. Step rate test — done (see point 3 above). Consistent failure above 1.5Gbps/lane, clean below it.

    After receiveing the deskew pattern provides clear video output right?

    Thanks.

    Regards,
    Kamal

thanks for sharing detailed information.

since the issue appears whenever the MIPI DPHY lane rate is above 1.5 Gbps/lane, this now strongly points to DPHY deskew calibration rather than a general VI throughput issue.

according to developer guide, SerDes Pixel Clock.
for DPHY operation data rate above 1.5 Gbps, it must send the required deskew calibration sequence. the deskew pattern is not video data itself; it is used by the receiver to calibrate lane skew before normal image data is captured.

the camera firmware will continue to wait for deskew signal from the sensor side. it’ll enable pixel parser when deskew calibration has completed, after successful deskew calibration, normal video frames should be received correctly.

hence..
please give it a try for sending a reset signal to the bridge device when s_stream() has called.
for instance, a bridge reset can clear stale CSI TX state, FIFO state, PLL/PHY state, and force the bridge to re-lock and resend the DPHY init/deskew sequence before normal video packets.