Device

SoC BSP Kernel Drift in November 2026: Linux 5.15 EOL Looms as ST, Qualcomm, and TI Split Between 6.18 and 7.3

Unpinned Yocto BSPs for ST STM32MP1/MP2 are silently shifting to Linux 6.18 while Linux 5.15 hits projected EOL in December 2026. Here is the verified vendor branch matrix and release-engineering playbook.

By 0xScorPionSon

9 min read

SoC BSP Kernel Drift in November 2026: Linux 5.15 EOL Looms as ST, Qualcomm, and TI Split Between 6.18 and 7.3

Embedded Linux SoC BSP Kernel Drift November 2026 Infographic

Executive Summary

Between October 14 and November 13, 2026, multiple System-on-Chip (SoC) Board Support Package (BSP) repositories quietly shifted kernel branches—creating subtle build-reproducibility traps for Yocto manifests and urgent security lifecycle deadlines for embedded hardware teams. According to the TECH VEDA Vendor BSP Status Report (published November 13, 2026) and cross-checked against The Linux Kernel Archives (kernel.org) and STMicroelectronics' meta-st-stm32mp repository, three major dynamics define embedded Linux device engineering this month:

  1. Unpinned Yocto Branch Drift on ST STM32MP1 and STM32MP2: The unversioned wrynose branch of meta-st-stm32mp now builds Linux 6.18 (v6.18-stm32mp-r1, Makefile 6.18.37), whereas the versioned release branch wrynose_v26.02.18 still builds Linux 6.6 (v6.6-stm32mp). Teams pinning wrynose by branch name without a commit hash or release tag risk silently jumping kernel major series across CI builds.
  2. Linux 6.18 LTS vs. 7.3 Branches on Qualcomm and TI: Qualcomm (qcom-7.3, head November 11) and Texas Instruments (ti-linux-7.3.y, head November 10) have both published Linux 7.3 trees, but kernel.org does not list 7.3 as a longterm maintenance series. Production baselines on Qualcomm QCS6490 and TI AM335x/AM62A/AM62x/J721E should remain pinned to qcom-6.18.y and ti-linux-6.18.y (projected EOL December 2028).
  3. Imminent Linux 5.15 & 5.10 EOL (December 2026): With fewer than 60 days remaining before kernel.org's projected December 2026 end of life for Linux 5.15 and 5.10, SoC BSPs still anchored on 5.15—such as the Allwinner T507 and T527—must migrate to a supported LTS branch, move to mainline, or assume self-managed CVE backporting.

This briefing maps the verified November 2026 BSP branch movements, explains the Yocto/kernel lifecycle architecture behind these splits, and provides a read-only Python & Git audit script plus a hardening playbook for embedded device teams.


Technical Mechanics and Architecture

How Vendor BSP Branching Diverges from kernel.org LTS

Embedded Linux firmware rarely tracks mainline kernel.org directly. Instead, silicon vendors fork an upstream longterm kernel series, layer out-of-tree SoC device trees, GPU/NPU drivers, and power-management patches on top, and expose them through Yocto/OpenEmbedded BSP layers (such as meta-st-stm32mp). When a vendor introduces a new kernel series or Civil Infrastructure Platform (CIP) maintenance branch on an unversioned layer head, downstream product manifests that track branch names instead of immutable SHAs or versioned tags inherit breaking kernel-ABI and device-tree shifts:

+-----------------------------------------------------------------------------+
|       EMBEDDED SOC BSP BRANCHING & YOCTO PINNING ARCHITECTURE (NOV 2026)    |
+-----------------------------------------------------------------------------+
|                                                                             |
|  [ UPSTREAM KERNEL.ORG LONGTERM SERIES ]                                    |
|   |-- 5.10 / 5.15 -> Projected EOL: Dec 2026  (e.g., Allwinner T507/T527)   |
|   |-- 6.1  / 6.6  -> Projected EOL: Dec 2027  (e.g., Rockchip 6.1, ST 6.6)  |
|   |-- 6.12 / 6.18 -> Projected EOL: Dec 2028  (e.g., Renesas 6.12, Qcom/TI) |
|   +-- 7.3         -> Non-LTS (No longterm entry on kernel.org as of Nov 13) |
|             |                                                               |
|             v                                                               |
|  [ VENDOR KERNEL & YOCTO LAYER REPOSITORIES ]                               |
|   |-- ST meta-st-stm32mp:                                                   |
|   |    |-- wrynose (unversioned)    -> KERNEL: v6.18-stm32mp-r1             |
|   |    +-- wrynose_v26.02.18 (tag)  -> KERNEL: v6.6-stm32mp                 |
|   +-- Renesas RZ/G2L:                                                       |
|        +-- rz-6.12-cip14 (active Nov 11) vs rz-6.12-cip15 (tracker target)  |
|             |                                                               |
|             v                                                               |
|  [ DEVICE CI / PRODUCTION MANIFEST ]                                        |
|   |-- Unpinned branch ("wrynose")   -> Silent 6.6 -> 6.18 jump + DT breakage|
|   +-- Pinned tag / commit SHA       -> Reproducible builds + controlled port|
+-----------------------------------------------------------------------------+

Verified November 2026 SoC BSP Status Matrix

Data cross-checked on November 13–14, 2026 across TECH VEDA's BSP Tracker, kernel.org Releases, and STMicroelectronics GitHub:

SoC / Platform Active Production / Tracked Branch Newer / Alternate Branch Observed Upstream kernel.org Projected EOL Engineering Action Required
ST STM32MP1 / STM32MP2 v6.6-stm32mp (wrynose_v26.02.18) v6.18-stm32mp (Makefile 6.18.37; unversioned wrynose layer) 6.6: Dec 2027; 6.18: Dec 2028 Pin wrynose_v26.02.18 or exact SHA; evaluate 6.18 on dev boards before Dec 2027
Renesas RZ/G2L rz-6.12-cip15 (head Oct 30, 2026) rz-6.12-cip14 (still received commit Nov 11, 2026) 6.12: Dec 2028 Audit whether recipe pins cip14 or cip15; diff patches before next maintenance release
Qualcomm QCS6490 qcom-6.18.y (head Nov 12, 2026) qcom-7.3 (head Nov 11, 2026; non-LTS) 6.18: Dec 2028 Stay on qcom-6.18.y; do not rebase shipping products to 7.3 unless promoted to LTS
TI AM335x / AM62A / AM62x / J721E ti-linux-6.18.y (head Nov 5, 2026) ti-linux-7.3.y (head Nov 10, 2026; non-LTS) 6.18: Dec 2028 Stay on ti-linux-6.18.y; monitor TI mirror vs. primary git host
AMD Zynq UltraScale+ MPSoC xlnx_rebase_v6.18_LTS (head Nov 4, 2026) — 6.18: Dec 2028 Routine LTS tracking; pull latest 6.18 security backports
Microchip SAMA7G5 linux-6.18-mchp (head Nov 6, 2026) — 6.18: Dec 2028 Routine LTS tracking; verify board defconfig against Nov 6 head
NXP i.MX6ULL / i.MX8M Plus / i.MX93 lf-6.18.y (head Sep 8, 2026) — 6.18: Dec 2028 Unchanged since Sep 8; monitor for Q4 CVE backports
Rockchip RK3588 / RK3588S develop-6.1 (head Jul 29, 2026) — 6.1: Dec 2027 No branch move; plan 2027 migration path ahead of Dec 2027 EOL
Allwinner T507 / T527 Linux 5.15 — Dec 2026 (<60 days) Urgent: select newer vendor kernel, mainline migration, or internal CVE backporting

Defensive Build-Engineering Code: Detecting Unpinned Yocto BSPs and Looming Kernel EOLs

To prevent CI pipelines from silently pulling an unpinned branch like ST's wrynose (or shipping a kernel approaching kernel.org EOL), embedded teams can run a read-only manifest audit script in CI. The following Python script parses a repo/KAS-style project configuration or uname -r inventory and flags unpinned branch references alongside kernel.org projected EOL dates:

import re
from dataclasses import dataclass

# Verified kernel.org longterm projected EOLs as of November 13, 2026
KERNEL_ORG_LTS_EOL = {
    "6.18": "2028-12",
    "6.12": "2028-12",
    "6.6": "2027-12",
    "6.1": "2027-12",
    "5.15": "2026-12",
    "5.10": "2026-12",
}

SHA1_RE = re.compile(r"^[0-9a-f]{40}$", re.IGNORECASE)

@dataclass
class BspSourceEntry:
    name: str
    branch_or_tag: str
    revision: str
    kernel_series: str

def audit_bsp_manifest(entries: list[BspSourceEntry]) -> list[str]:
    """
    Flags unpinned Yocto/BSP layer references, non-LTS kernel series,
    and kernels approaching kernel.org projected EOL (Dec 2026).
    """
    warnings: list[str] = []

    for entry in entries:
        is_sha_pinned = bool(SHA1_RE.match(entry.revision.strip()))
        if not is_sha_pinned:
            warnings.append(
                f"[UNPINNED] {entry.name} tracks '{entry.branch_or_tag}' "
                f"(rev='{entry.revision}') without an immutable 40-char commit SHA."
            )

        eol = KERNEL_ORG_LTS_EOL.get(entry.kernel_series)
        if eol is None:
            warnings.append(
                f"[NON-LTS] {entry.name} targets kernel {entry.kernel_series}, "
                "which is NOT listed on kernel.org's longterm table."
            )
        elif eol == "2026-12":
            warnings.append(
                f"[IMMINENT EOL] {entry.name} runs Linux {entry.kernel_series} "
                f"(projected EOL {eol}); plan migration or self-managed backports."
            )

    return warnings

Evidence: Upstream kernel.org Lifecycle and Verification Boundaries

Cross-checking the November 13, 2026 TECH VEDA report against primary sources confirms both the exact kernel timelines and several important verification boundaries:

  • Primary kernel.org LTS Confirmation: kernel.org's Active Kernel Releases table confirms six active longterm maintenance series: 6.18 (released 2025-11-30, EOL Dec, 2028), 6.12 (released 2024-11-17, EOL Dec, 2028), 6.6 (released 2023-10-29, EOL Dec, 2027), 6.1 (released 2022-12-11, EOL Dec, 2027), 5.15 (released 2021-10-31, EOL Dec, 2026), and 5.10 (released 2020-12-13, EOL Dec, 2026). Kernel 7.3 is absent from the longterm table.
  • Primary STMicroelectronics Layer Confirmation: Inspecting STMicroelectronics/meta-st-stm32mp on GitHub confirms that the default wrynose branch actively carries KERNEL: v6.18-stm32mp-r1, U-BOOT: v2026.04-stm32mp-r1, OPTEE: 4.8.0-stm32mp-r1, A/B firmware WIC examples, and SBOM fragment tooling—marking a full platform generation jump from the 6.6 recipe in wrynose_v26.02.18.
  • Renesas cip14 vs. cip15 Nuance: Although the tracker moved RZ/G2L from rz-6.12-cip14 (last tracked commit May 25, 2026) to rz-6.12-cip15 (head October 30, 2026), a check on November 13 revealed a fresh commit on cip14 dated November 11, 2026, proving cip14 is not frozen.
  • Confidence & Benchmark Caveat: Confidence Level: High on kernel.org EOL projections, ST's wrynose 6.18 shift, and the presence of Qualcomm/TI 7.3 branches. Verification & Benchmark Caveats: Per TECH VEDA's disclosure, TI branches were read from TI's public GitHub mirror rather than its primary git host, and Allwinner/Amlogic/NVIDIA rows were not re-verified against vendor git remotes in the November 13 pass. Additionally, no cross-SoC runtime performance or boot-time benchmarks comparing 6.6, 6.12, 6.18, and 7.3 were published in these advisories; teams should benchmark target peripherals locally before promoting 6.18 or 7.3 to production.

Hardening & Release-Engineering Playbook

Step 1: Audit All Yocto & Buildroot Manifests for Unpinned Branches

Inspect every KAS, repo XML, or bblayers.conf source entry for bare branch names like wrynose, scarthgap, or main without a commit / SRCREV pin. Use git ls-remote to verify where upstream heads point before running a clean CI build:

git ls-remote --heads https://github.com/STMicroelectronics/meta-st-stm32mp.git 'wrynose*'
git ls-remote --heads https://github.com/STMicroelectronics/linux.git 'v6.*-stm32mp'

Step 2: Lock Production Shipping Lines to 6.18.y / 6.12 LTS (Avoid 7.3 for Production)

If your team builds on Qualcomm QCS6490 or TI AM335x/AM62A/AM62x/J721E, keep production SRCREV pins on qcom-6.18.y or ti-linux-6.18.y (projected EOL December 2028). Treat qcom-7.3 and ti-linux-7.3.y strictly as evaluation branches until kernel.org designates a 7.x longterm series.

Step 3: Verify Running Target Kernels in the Field

On deployed hardware, confirm the exact kernel release and vendor suffix rather than relying on build-server assumptions:

uname -r

Step 4: Execute a Linux 5.15 / 5.10 Exit Plan Before December 2026

For any shipping device on Linux 5.15 or 5.10 (such as Allwinner T507/T527 platforms), decide immediately among three paths: (a) upgrade to a vendor 6.6/6.12/6.18 BSP, (b) evaluate mainline Linux support for your board's peripherals, or (c) establish an internal CVE triage and backporting workflow for out-of-support kernels.

Step 5: Rollback & Staged Evaluation Notes

Never migrate a shipping 6.6 product (such as STM32MP1/MP2) directly onto an unversioned 6.18 layer branch in production. Stand up a dedicated evaluation build on development hardware to catalog required Device Tree (DT) and out-of-tree driver API changes while keeping production CI pinned to the tagged wrynose_v26.02.18 release so you can roll back cleanly via a single manifest tag revert.


Frequently Asked Questions

1. Should embedded teams upgrade Qualcomm or TI boards to the new Linux 7.3 BSP branches?

No, not for shipping production devices. As of November 13, 2026, kernel.org does not list 7.3 on its longterm maintenance table; production baselines should remain on qcom-6.18.y and ti-linux-6.18.y, which have a projected EOL of December 2028.

2. Why did my ST STM32MP1/MP2 Yocto build suddenly switch from Linux 6.6 to 6.18?

If your manifest pins the unversioned wrynose branch of STMicroelectronics/meta-st-stm32mp, ST updated wrynose to carry v6.18-stm32mp-r1 recipes while the older versioned release wrynose_v26.02.18 continues to build Linux 6.6.

3. Does Renesas RZ/G2L moving from rz-6.12-cip14 to rz-6.12-cip15 change its EOL date?

No. Both cip14 and cip15 are based on Linux 6.12, for which kernel.org projects an end-of-life date of December 2028. However, because cip14 still received a commit on November 11, 2026, teams should verify which CIP branch their recipe pins.

4. What happens to devices running Linux 5.15 after December 2026?

kernel.org projects the end of upstream longterm maintenance for Linux 5.15 (and 5.10) in December 2026. After EOL, upstream stable maintainers stop backporting security and bug fixes, leaving product teams to upgrade kernels or maintain patches themselves.

5. Is a kernel.org "Projected EOL" date a hard cutoff?

No. As kernel.org notes, projected EOL dates are not set in stone—new longterm kernels typically start with a 2-year projection that can be extended based on industry demand—but engineering teams should treat projected EOLs as hard planning deadlines for compliance and CVE response.


Sources

More on Device

Tell us what is not working

A thirty-minute call, no deck and no obligation. You will leave with a straight answer about whether this is worth building - including if that answer is no.

Start a project

Direct line

hello@codebumble.net
Dhaka, Bangladesh  :   local timeCalls in English or Bangla