Skip to main content

TECH VEDA

Embedded Linux on Edge-AI 23rd Sept 2026 enrollingLinux kernel & Device drivers starts on 24th Oct 2026 enrollingCorporate on-site training - Submit proposal Pick your modulesSharpen your kernel skills: deep dives, drivers, Yocto, CVEs, careers — updated daily. Read the blog →Embedded Linux fast track starts 23rd sept 2026 enrollingEmbedded Linux Mastery track starts 23rd sept 2026 enrollingLinux systems engineering starts 23rd sept 2026 enrolling

BSP Tracker  /  BSP Status / STMicroelectronics

ST STM32MP2 (STM32MP257)

Mostly fine for CPU, peripherals and video decode. Not if the NPU is why you chose the part: the Neural-ART accelerator has no mainline driver, so an AI product is locked to OpenSTLinux and X-LINUX-AI. Note the SoC landed upstream (v6.5) BEFORE the vendor branch it ships on (6.6).

BSP branch v6.6-stm32mpLinux 6.6Mainline since v6.5
See all 17 →
Security cliff
Upstream fixes for Linux 6.6 stop in December 2027 — 16 months from now.
The vendor BSP tracks Linux 6.6 on branch v6.6-stm32mp. After that date no fixes arrive from upstream, and maintaining this kernel becomes your job.

Mainline instead of the vendor BSP

Can you ship mainline instead, and what exactly do you lose if you do?

Mainline, with named losses

BlockIPUpstreamDetail
VPU decodeHantro (VeriSilicon)upstreamCONFIG_VIDEO_HANTRO_STM32MP25 + stm32mp25_vpu_hw.c — decode driver is in mainline.
VPU encodeHantro H1 (VeriSilicon)upstreamstm32mp25_venc_variant in drivers/media/platform/verisilicon registers HANTRO_JPEG_ENCODER only (JPEG / M-JPEG). There is no H.264 or HEVC video encode upstream, so a mainline video-recording product is not covered.
NPUST Neural-ART (1.35 TOPS)missingNo ST driver in drivers/accel. AI is X-LINUX-AI toolchain only.
GPUVeriSilicon 3Dunverified
ISPDCMIPPunverified
Mainline since v6.5, verified by bracketing the device tree against the kernel tags: proof 1 · proof 2

Vendor BSP

Repository. https://github.com/STMicroelectronics/linux

Newest branch that actually exists. v6.6-stm32mp, tracking Linux 6.6. Last commit 2026-06-01.

Cadence. There is NO separate STM32MP2 branch and NO 6.12 branch. v6.6-stm32mp is the newest branch that exists, full stop.

Silicon longevity

The vendor publishes no commitment. Same as STM32MP1: no per-part table. A '10 years / 2029' figure surfaced during research but was not supported by any text on ST's page, and was discarded.

Vendor source

Yocto

Layer. meta-st-stm32mp. A branch for the current LTS exists: wrynose_v26.02.18.

master is a 2019 stub. Never build from it.

Hardware you can buy

Evaluate on. STM32MP257F-DK, STM32MP257F-EV1

Production modules.

What to do about it

ST STM32MP2 (STM32MP257): upstream fixes stop December 2027. That leaves 16 months. After that date the kernel this BSP is based on receives no further security fixes from upstream, and maintaining it becomes your team’s responsibility.

There are three options:

  1. Backport the fixes yourself. Your team maintains a kernel that nobody else maintains: tracking each CVE, applying the patch to a tree that has diverged from upstream, and testing the result, for as long as the product ships.
  2. Move to the vendor’s newer BSP. This gives you a few more years of upstream fixes, and it costs a second bring-up: a new kernel, your patches forward-ported, drivers re-validated, and the product re-tested.
  3. Move to mainline. Mainline supports part of this SoC. The blocks marked above are the ones a migration has to solve or replace. It is the only option that does not have to be repeated when the next vendor kernel reaches end of life.

All three are engineering work, and all three are work we do: reading vendor kernel trees, checking what mainline actually supports, and moving products from one to the other. This page is built from the same work.

Embedded Linux engineering →Talk to us about your BSP

Get the updates

An email when a kernel series reaches end of life, a board row changes, or a layer picks up a release branch. Nothing else.

RSS