# [Bug Report] TC358743 (HDMI-to-CSI2) V4L2 STREAMON crashes kernel via tegra_camera_platform NULL pointer — same DCE RPC failure as thread 377392

**Cross-referencing:** [nvarguscamerasrc fails with NvRmGpuLibOpen error=4, DCE RPC control call failures at boot (thread 377392)]( Jetson Orin Nano Devkit / JetPack 7.2 (R39.2.0) — nvarguscamerasrc fails with NvRmGpuLibOpen error=4, DCE RPC control call failures at boot )

## Environment

- Hardware: Custom Jetson Orin NX carrier board

- HDMI capture module: Geekworm C790 (TC358743 HDMI-to-CSI-2 bridge)

- L4T: R39.2.1

- JetPack: 7.2

- Kernel: 6.8.12-1021-tegra

- Base DTB: `tegra234-p3768-0000+p3767-0003-nv.dtb`

- Overlay in use: `tegra234-p3767-camera-p3768-tc358743-A.dtbo` (NVIDIA’s own official overlay for this chip/carrier)

## Symptom

`VIDIOC_STREAMON` on `/dev/video0` (TC358743 → NVCSI → VI capture chain) causes a kernel NULL pointer dereference and full system crash/reboot — not a userspace error, a hard kernel oops that takes down the whole SoC.

Reproducer:

```

v4l2-ctl -d /dev/v4l-subdev0 --set-dv-bt-timings query

v4l2-ctl -d /dev/v4l-subdev0 --set-subdev-fmt pad=0,code=0x200f,width=1920,height=1080

v4l2-ctl -d /dev/video0 --set-fmt-video=width=1920,height=1080,pixelformat=UYVY

yavta -f UYVY -s 1920x1080 -n 2 --capture=1 -F /dev/video0

```

All format/timings negotiation succeeds correctly (subdev reports `UYVY8_1X16/1920x1080`, `dv.current` matches `dv.detect` at 1920x1080p60, video node reports correct `Size Image: 4147200`, buffers allocate and map correctly). The crash occurs specifically when STREAMON attempts to start the actual capture.

## Kernel oops

```

Unable to handle kernel NULL pointer dereference at virtual address 0000000000000078

Internal error: Oops: 0000000096000004 [#1] PREEMPT SMP

CPU: 2 PID: 9250 Comm: yavta Tainted: G W OE 6.8.12-1021-tegra #1

Hardware name: NVIDIA NVIDIA Jetson Orin Nano Developer Kit/Jetson, BIOS 39.2.1-gcid-46758480 08/07/2026

pc : tegra_camera_emc_clk_enable+0x18/0x134 [tegra_camera_platform]

lr : convert_table_to_blob+0x7e0/0x1e90 [tegra_camera]

Call trace:

tegra_camera_emc_clk_enable+0x18/0x134 [tegra_camera_platform]

convert_table_to_blob+0x7e0/0x1e90 [tegra_camera]

dequeue_dequeue_buffer+0xa4/0x154 [tegra_camera]

vb2_core_queue_release+0xcc/0x1f0 [videobuf2_common]

vb2_core_streamon+0x100/0x1d4 [videobuf2_common]

vb2_ioctl_streamon+0x4c/0x98 [videobuf2_v4l2]

video_usercopy+0x2d0/0x814 [videodev]

video_ioctl2+0x18/0x4c [videodev]

__arm64_sys_ioctl+0xb4/0xf4

```

## Boot-time precursor — matches thread 377392 exactly

On every boot, well before any camera driver activity, we see the same DCE RPC failure signature reported in thread 377392:

```

[ 12.038573] dce: dce_ipc_channel_init_unlocked:248 Invalid Channel State [0x0] for ch_type [2]

[ 12.117500] dce: dce_admin_setup_clients_ipc:978 Get queue info failed for [2]

[ 15.219995] NVRM: GPU0 rpcRmApiControl_dce: NVRM_RPC_DCE: Failed RM ctrl call cmd:0x731341 result 0xffff: Failure: Generic Error [NV_ERR_GENERIC]

```

This repeats dozens of times across multiple boot-time timestamps.

## What we’ve ruled out

- **Device tree config** — checked base DTB and the official NVIDIA TC358743 overlay; neither contains a `tegra-camera-platform` node, so this doesn’t appear to be a missing/incorrect DT entry on this kernel generation.

- **CSI bandwidth** — separately confirmed and fixed. RGB888 (24bpp) exceeded our board’s configured CSI link bandwidth and caused a *different*, earlier hard hang; switching to UYVY (16bpp) resolved that but did not prevent this STREAMON crash.

- **The TC358743 driver itself** — the crash trace never touches the `tc358743` module; it’s entirely inside `tegra_camera_platform`/`tegra_camera`.

## Why we think this is the same bug as 377392

Same DCE RPC failure code (`cmd:0x731341 result 0xffff`) at boot, same L4T/kernel version, but manifesting downstream in a different subsystem — a hard V4L2/kernel crash via `tegra_camera_platform` on our end, vs. a userspace Argus/`NvRmGpuLibOpen` segfault in that thread. This suggests one root-cause DCE initialization failure with multiple possible downstream symptoms depending on which camera/GPU code path gets exercised.

## Questions for NVIDIA

1. Is this DCE RPC failure (`cmd:0x731341`) a known issue in R39.2.x, and is a fix targeted for a specific point release?

2. Is the `tegra_camera_platform` NULL pointer at `tegra_camera_emc_clk_enable` a known consequence of the DCE init failure, or a separate bug worth tracking independently?

3. Is there a runtime workaround (e.g., forcing a DCE re-init, disabling EMC scaling for the camera path) that avoids the crash without a full kernel patch?

Happy to provide full `dmesg`, `journalctl -k`, device tree dumps, or test additional reproduction steps.

Hello @karhowe.lee!

Based on the title and content of your topic, it looks like it may receive better visibility and feedback in a different category. We took the liberty of moving it for you.

If this was an incorrect assessment, please send me a direct message.

Disclaimer: this moderation suggestion and message were generated with AI assistance.

hello karhowe.lee

may I double check your release version, are you working with native JetPack 7.2.1 public release version?
besides.. please setup serial console for sharing complete kernel logs for reference. you may upload it as a single text file.

Update: I’ve reproduced this exact crash on a second, completely different board and L4T generation, which rules out anything carrier-specific.

New environment

  • Hardware: Stock NVIDIA Jetson Orin Nano Developer Kit (unmodified dev kit carrier p3768-0000 + module p3767-0003)
  • HDMI capture module: Same Geekworm C790 (TC358743)
  • L4T: R36.5.2
  • JetPack: 6.x
  • Kernel: 5.15.199-tegra
  • Overlay: tegra234-p3767-camera-p3768-tc358743-A.dtbo (NVIDIA’s official overlay, applied via jetson-io.py)

Reproducer

Same as before - EDID programmed, DV timings detected correctly (1920x1080p60, 148.5MHz), format set to UYVY, full media-ctl pipeline shows all links ENABLED. Crash occurs on STREAMON.

New crash trace (captured via pstore/ramoops after hard reboot)

Unable to handle kernel NULL pointer dereference at virtual address 0000000000000078
Internal error: Oops: 0000000096000004 [#1] PREEMPT SMP
CPU: 2 PID: 5238 Comm: v4l2-ctl Tainted: G W OE 5.15.199-tegra #1
Hardware name: NVIDIA NVIDIA Jetson Orin Nano Developer Kit/Jetson, BIOS 36.5.0-gcid-43688277 01/16/2026
pc : tegra_camera_emc_clk_enable+0x24/0x160 [tegra_camera_platform]
lr : tegra_camera_emc_clk_enable+0x1c/0x160 [tegra_camera_platform]
Call trace:
 tegra_camera_emc_clk_enable+0x24/0x160 [tegra_camera_platform]
 convert_table_to_blob+0x978/0x1e50 [tegra_camera]
 dequeue_dequeue_buffer+0xc0/0x200 [tegra_camera]
 vb2_core_queue_release+0xe4/0x1f0 [videobuf2_common]
 vb2_core_streamon+0xac/0x470 [videobuf2_common]
 vb2_ioctl_streamon+0x74/0xd0 [videobuf2_v4l2]
 video_usercopy+0x280/0x680 [videodev]
 video_ioctl2+0x44/0x90 [videodev]
 __arm64_sys_ioctl+0xbc/0x100
---[ end trace ]---

Identical function (tegra_camera_emc_clk_enable) and nearly identical call chain to the original report, on a different kernel (5.15.199 vs 6.8.12), different L4T major version (R36 vs R39), and stock hardware instead of a custom carrier. This strongly suggests the bug is in shared tegra_camera_platform/tegra_camera code that’s persisted across at least two L4T release generations, not something introduced recently or tied to non-standard hardware.

Workaround attempted (did not help)

Tried forcing static max clocks (jetson_clocks) plus locking EMC rate via /sys/kernel/debug/bpmp/debug/clk/emc/mrq_rate_locked=1 before STREAMON. Same crash, same signature - so this isn’t a dynamic-scaling-timing race that clock pinning can avoid; looks like a genuine NULL pointer bug in the EMC clock enable path itself.

Happy to provide the full pstore dump, extlinux.conf, or device tree blob dump if useful. Given this now reproduces on stock reference hardware with the officially-provided overlay, it should be straightforward for NVIDIA to reproduce in-house with just a C790 module and a dev kit.

hello karhowe.lee,

it’s reproduced with HDMI capture module, there’s no tegra-camera-platform node.
according to the logs, it should due to tegra_camera_emc_clk_enable() dereferences tegra_camera_misc.parent while it is NULL.
for instance,
$public_sources/kernel_src/nvidia-oot/drivers/video/tegra/camera/tegra_camera_platform.c

int tegra_camera_emc_clk_enable(void)
{
        info = dev_get_drvdata(tegra_camera_misc.parent); <-- it is already NULL

let’s give it a try with below kernel hardening, you should apply the guard to both tegra_camera_emc_clk_enable() and tegra_camera_emc_clk_disable() for quick testing.
for instance,

int tegra_camera_emc_clk_enable(void)
{

+       if (!tegra_camera_misc.parent)
+           return -ENODEV;

        info = dev_get_drvdata(tegra_camera_misc.parent);
        if (!info)
                return -EINVAL;

if that address the STREAMON kernel panic,
please try adding below into DTBO, which the tegra_camera_platform driver only probes against this compatible string.
for instance,

        __overlay__ {
+                tegra_camera_platform: tegra-camera-platform {
+                        compatible = "nvidia,tegra-camera-platform";
+                        status = "okay";