Tracked on RSDSO-21191 Support Nvidia JP 5.1.x/RSDSO-21146 Support Nvidia JP6.2.2

[Kontra2B]

Add support for Jetpack versions:
5.1.6
6.2.2

[Kontra2B]

The builds fail due to the issue addressed in #401

[Copilot]

Pull request overview

Adds support for NVIDIA JetPack 5.1.6 and 6.2.2 across the workspace scripts and kernel patchsets, with CI workflows to validate builds.

Changes:

• Extend setup/sync scripts to recognize JP 5.1.6 and 6.2.2 (and prefer JP_INPUT_VERSION sources files).
• Add JP-version mapping “pointer” files and introduce new kernel patch directories for 5.1.6 / 6.2.2.
• Add GitHub Actions workflows to build JP 5.1.6 and JP 6.2.2 in CI.

Reviewed changes

Copilot reviewed 17 out of 18 changed files in this pull request and generated 10 comments.

Show a summary per file

File	Description
scripts/sync-sources.sh	Prefer sources_${JP_INPUT_VERSION} over sources_${JETPACK_VERSION} when present
scripts/setup-common	Add JP 6.2.2 / 5.1.6 support and adjust -h behavior
nvidia-oot/6.2.2	Map JP 6.2.2 OOT patchset to 6.2.1 baseline
kernel/kernel-5.10/5.1.6	Map JP 5.1.6 kernel baseline to 5.1.2
kernel/nvidia/5.1.6/0001-Porting-driver-patches-to-jetpack-5.1.2.patch	Kernel driver patchset (ported) used for JP 5.1.6
kernel/nvidia/5.1.6/0002-nvidia-vi5-prevent-high-cpu-usage.patch	VI5 CPU usage fix included for JP 5.1.6
kernel/nvidia/5.1.6/0003-vi-channel-verify-s-g-callbacks.patch	Reuse 5.1.2 patch for JP 5.1.6
kernel/nvidia/5.1.6/0004-Adding-stream-restart-upon-capture-timeout.patch	Add VI stream restart-on-timeout logic for JP 5.1.6
kernel/nvidia/5.1.6/0005-Add-GPIO-tunneling-for-the-external-sync-support.patch	Reuse 5.1.2 patch for JP 5.1.6
kernel/nvidia/5.1.6/0006-Adding-max96712-support-for-D4xx.patch	Reuse 5.1.2 patch for JP 5.1.6
kernel/kernel-jammy-src/6.2.2/0001-kernel-enable-HID_SENSOR-without-USB-impact.patch	Reuse 6.2.1 patch for JP 6.2.2
kernel/kernel-jammy-src/6.2.2/0002-realsense-camera-formats-jammy-master.patch	Reuse 6.2.1 patch for JP 6.2.2
kernel/kernel-jammy-src/6.2.2/0003-realsense-metadata-jammy-master.patch	Reuse 6.2.1 patch for JP 6.2.2
kernel/kernel-jammy-src/6.2.2/0004-realsense-powerlinefrequency-control-fix-jammy.patch	Reuse 6.2.1 patch for JP 6.2.2
kernel/kernel-jammy-src/6.2.2/0005-support-for-dynamic-change-of-i2c-bus-clk-rate.patch	Add dynamic I2C bus clock rate control for JP 6.2.2
.github/workflows/build-jp5.1.6.yml	CI workflow for building JP 5.1.6
.github/workflows/build-jp6.2.2.yml	CI workflow for building JP 6.2.2

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 18 out of 19 changed files in this pull request and generated 11 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 18 out of 19 changed files in this pull request and generated 13 comments.

Comments suppressed due to low confidence (4)

kernel/nvidia/5.1.6/0001-Porting-driver-patches-to-jetpack-5.1.2.patch:1

• Using bitwise-OR to accumulate error codes can corrupt the original errno (e.g., combining negative values) and makes it hard to return the first/most relevant failure. Prefer returning immediately on first error (or preserving the first negative errno) rather than err |= ....

From 1b06cf3323a3d7d1437071dc3789e9487b9b5733 Mon Sep 17 00:00:00 2001

kernel/nvidia/5.1.6/0001-Porting-driver-patches-to-jetpack-5.1.2.patch:1

• head/tail/num_buffers are reset outside the spinlock even though they are part of the same protected ring-buffer state used elsewhere. This introduces a race with queueing/dequeueing metadata buffers. Reset these fields under the same spinlock (and remove the extra semicolon on tail = 0;;).

From 1b06cf3323a3d7d1437071dc3789e9487b9b5733 Mon Sep 17 00:00:00 2001

kernel/kernel-jammy-src/6.2.2/0005-support-for-dynamic-change-of-i2c-bus-clk-rate.patch:1

• bus_clk_rate is stored as unsigned long in struct i2c_adapter, but sysfs show uses %ld and store parses into an int (memparse returns an unsigned long), which can truncate/overflow and mis-display values. Also, the sysfs store updates adap->bus_clk_rate without taking the i2c bus lock (unlike i2c_set_adapter_bus_clk_rate), which can race with active transfers. Use sysfs_emit()/%lu, parse into unsigned long, validate range, and update under i2c_lock_bus (or call the new helper) to make the change thread-safe.

From ce6faaedef68204e8b132bf44a0e4d2432dfb7ad Mon Sep 17 00:00:00 2001

kernel/kernel-jammy-src/6.2.2/0005-support-for-dynamic-change-of-i2c-bus-clk-rate.patch:1

• bus_clk_rate is stored as unsigned long in struct i2c_adapter, but sysfs show uses %ld and store parses into an int (memparse returns an unsigned long), which can truncate/overflow and mis-display values. Also, the sysfs store updates adap->bus_clk_rate without taking the i2c bus lock (unlike i2c_set_adapter_bus_clk_rate), which can race with active transfers. Use sysfs_emit()/%lu, parse into unsigned long, validate range, and update under i2c_lock_bus (or call the new helper) to make the change thread-safe.

From ce6faaedef68204e8b132bf44a0e4d2432dfb7ad Mon Sep 17 00:00:00 2001

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 19 out of 20 changed files in this pull request and generated 4 comments.

Comments suppressed due to low confidence (7)

kernel/nvidia/5.1.6/0001-Porting-driver-patches-to-jetpack-5.1.2.patch:1

• vb2_buffer_done() is invoked while holding chan->embedded.spin_lock. Buffer completion can trigger callbacks/locking that is not safe under a spinlock in many vb2 paths. Move the vb2_buffer_done() call outside the spinlocked section (e.g., decide under lock, then complete after unlock).

From 1b06cf3323a3d7d1437071dc3789e9487b9b5733 Mon Sep 17 00:00:00 2001

kernel/nvidia/5.1.6/0001-Porting-driver-patches-to-jetpack-5.1.2.patch:1

• vb2_buffer_done() is called under spin_lock here as well, which has the same locking-safety concern as in tegra_metadata_buffer_queue. Consider collecting the buffers to complete into a temporary list/array while holding the lock, then releasing the spinlock and completing them afterwards.

From 1b06cf3323a3d7d1437071dc3789e9487b9b5733 Mon Sep 17 00:00:00 2001

kernel/nvidia/5.1.6/0001-Porting-driver-patches-to-jetpack-5.1.2.patch:1

• Both S_FMT and TRY_FMT return success without validating format or normalizing fields (e.g., dataformat and buffersize). This can cause user space to believe an unsupported meta format/buffersize is accepted, while the queue later rejects it (or worse, accepts inconsistent state). At minimum, enforce V4L2_META_FMT_D4XX and a fixed/validated buffersize here and return -EINVAL for unsupported requests.

From 1b06cf3323a3d7d1437071dc3789e9487b9b5733 Mon Sep 17 00:00:00 2001

kernel/nvidia/5.1.6/0001-Porting-driver-patches-to-jetpack-5.1.2.patch:1

• interval is stack-allocated and passed to g_frame_interval without initializing required fields (notably interval.pad, which you do set in the corresponding s_frame_interval path). This can lead to undefined behavior or EINVAL depending on the subdev implementation. Initialize interval (e.g., zero it and set .pad consistently) and consider mapping -ENOIOCTLCMD to -ENOTTY as is done elsewhere in this file.

From 1b06cf3323a3d7d1437071dc3789e9487b9b5733 Mon Sep 17 00:00:00 2001

kernel/nvidia/5.1.6/0001-Porting-driver-patches-to-jetpack-5.1.2.patch:1

• The code copies 255 bytes into the metadata buffer but sets the vb2 plane payload to 68 bytes. This inconsistency will cause user space to only see 68 bytes as valid (or suggests the copy size is wrong). Align the payload size with the actual produced metadata size (either copy 68, or set payload to 255, or compute it dynamically).

From 1b06cf3323a3d7d1437071dc3789e9487b9b5733 Mon Sep 17 00:00:00 2001

kernel/nvidia/5.1.6/0001-Porting-driver-patches-to-jetpack-5.1.2.patch:1

• Returning -1 is non-standard for kernel code and loses errno semantics for callers/diagnostics. Use an appropriate negative errno (e.g., -ENODEV when the CSI subdev can’t be found) so error handling remains consistent.

From 1b06cf3323a3d7d1437071dc3789e9487b9b5733 Mon Sep 17 00:00:00 2001

kernel/nvidia/5.1.6/0001-Porting-driver-patches-to-jetpack-5.1.2.patch:1

• Changing TEGRA_STRIDE_ALIGNMENT from 1 to 64 is a global behavior change that will affect buffer layout/stride for all formats and all users of this header, not just D4xx-specific paths. If this is only required for certain modes/devices, consider making the alignment conditional (per-format/per-device) or document why the global change is safe; otherwise it can cause unexpected size/stride differences in existing deployments.

From 1b06cf3323a3d7d1437071dc3789e9487b9b5733 Mon Sep 17 00:00:00 2001

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[ejgoldik]

Looks like debug prints left by mistake

[Kontra2B]

Removed

[ejgoldik]

Looks like debug prints left by mistake

[Kontra2B]

Removed

[ejgoldik]

Just remove the debug prints in build_all (Or make them official-looking if you think they should remain)