Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
29 commits
Select commit Hold shift + click to select a range
347fdfc
axera: add br-ext-chip-axera for AX620E and AX620U
dimerr Oct 24, 2026
bab8e68
axera: wire the real kernel configs the board defconfigs point at
dimerr Oct 24, 2026
d0e158f
ax620u_lite: build fixes found by the first full buildroot build
openipc-ai Nov 18, 2026
a76dded
axera-osdrv-ax620u: load_axera as the stock firmware loads them
openipc-ai Nov 18, 2026
ec00df4
ax620u: a kernel config that can run the image
openipc-ai Nov 18, 2026
d0b582c
ax620u: the Axera SDK kernel and OpenIPC's NAND layout
openipc-ai Nov 18, 2026
e6ac7a1
axera-osdrv-ax620u: load every module, ax_audio and ax_pcal6416a too
openipc-ai Nov 18, 2026
dbf1ebf
ipctool: an Axera HAL, so S70vendor runs load_axera
openipc-ai Nov 18, 2026
e50f713
ax620e_lite: make the build finish
dimerr Nov 18, 2026
9f7431e
ax620e: a camera kernel config instead of arm64 defconfig
dimerr Nov 18, 2026
5d043bb
axera-osdrv-ax620e: load_axera in depends= order, with a CMM pool
dimerr Nov 18, 2026
4661d42
ax620e: Xiongmai IPC devicetree and a NOR image its U-Boot boots
dimerr Nov 18, 2026
26c0897
ax620e: README: what the stock Xiongmai U-Boot needs, known blockers
dimerr Nov 18, 2026
13f5973
ax620e: the camera kernel from the axera-ax620e SDK port, its in-tree DT
openipc-ai Nov 18, 2026
ac78b70
ipctool: Axera HAL, AX620E and AX620U
openipc-ai Nov 18, 2026
dba3f44
ax620e: the standard OpenIPC NOR layout, with the OpenIPC U-Boot
openipc-ai Nov 18, 2026
a188e58
ipctool: one Axera HAL patch for both chips
openipc-ai Nov 18, 2026
d0c5c40
Merge remote-tracking branch 'upstream/master' into axera-combined
openipc-ai Nov 18, 2026
4013ffc
ax620e: the arm64 uImage without the split layout
openipc-ai Nov 18, 2026
fafd9d4
ci-matrix: account for the Axera boards
openipc-ai Nov 18, 2026
80420f3
axera-osdrv: load_axera with modprobe, provenance in the READMEs
openipc-ai Nov 18, 2026
f7bc3d0
ipctool: Axera HAL patch from ipctool#290 at e6386eb
openipc-ai Nov 18, 2026
837f87c
axera: README: what was tested, in QEMU, and what was not
openipc-ai Nov 18, 2026
e9f1272
ci: build ax620e_lite and ax620u_lite
openipc-ai Nov 18, 2026
aedb3ee
axera: sensor tuning from the stock firmwares
openipc-ai Nov 19, 2026
624139c
Merge remote-tracking branch 'origin/master' into axera
openipc-ai Nov 20, 2026
cf8baba
ci-matrix: ax620e_lite and ax620u_lite build weekly, not every night
openipc-ai Nov 20, 2026
0bebcd8
ci-matrix: a widening pull request still builds the weekly boards it …
openipc-ai Nov 20, 2026
befce01
ipctool: drop the Axera HAL patch, upstream since ipctool#290 (b1f0324)
openipc-ai Nov 20, 2026
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 14 additions & 5 deletions .github/scripts/ci-matrix.py
Original file line number Diff line number Diff line change
Expand Up @@ -176,9 +176,8 @@
# Not firmware images at all: no BR2_OPENIPC_VARIANT, so nothing the repack
# step or the manifest knows how to name.
"hi3519dv500_toolchain": "toolchain config, produces no firmware image",
# Axera has no image yet. These two configs publish its toolchains so
# majestic can be ported first. The boards follow once a camera has run
# them (#2670).
# Axera's toolchain configs publish toolchain.axera-ax620{e,u} (#2852) for
# the ax620e_lite and ax620u_lite boards, which are in WEEKLY_BOARDS.
"ax620e_toolchain": "toolchain config, produces no firmware image",
"ax620u_toolchain": "toolchain config, produces no firmware image",
# T41 publishes its toolchain ahead of its boards (#2877), whose build
Expand Down Expand Up @@ -284,6 +283,8 @@
# They are part of the tree like any other built board, so --self-test holds
# them to the same contract, but they cost more than they are touched: the v83x
# pair compile their own gcc and musl before the firmware, and change rarely.
# The Axera pair are new and not yet run on a camera, so a weekly image is
# enough until they are (#2670).
#
# A pull request builds them only for a change nothing else covers: their own
# defconfigs, their board directory, or a package no nightly board builds. A
Expand All @@ -295,7 +296,7 @@
# six days their last images stay where flashers and sysupgrade find them. A
# nightly that skips because HEAD has not moved skips them too; they wait for the
# next WEEKLY_DAY with a new commit.
WEEKLY_BOARDS = ["v83x_lite", "v83x_ultimate"]
WEEKLY_BOARDS = ["v83x_lite", "v83x_ultimate", "ax620e_lite", "ax620u_lite"]
WEEKLY_DAY = 6 # datetime.weekday(): Sunday, UTC

# Board directories with no board in ALL_BOARDS. Their defconfigs exist but are
Expand Down Expand Up @@ -646,8 +647,12 @@ def classify(tree, changed, labels=(), event="pull_request", draft=False,
return _decision(list(tree.built), True, reason=f"{FULL_LABEL} label set")

# On a pull request the full matrix is the nightly one: a change that
# widens leaves the weekly boards to their weekly run.
# widens leaves the weekly boards to their weekly run. A weekly board this
# pull request adds to the matrix is the exception: nothing has built it
# yet, so it is built here rather than first on a Sunday.
full = list(tree.nightly)
if base_boards is not None and f".github/scripts/{SELECTOR}" in changed:
full += [b for b in tree.weekly if b not in base_boards]

if draft:
return _decision([], False, reason="draft pull request")
Expand Down Expand Up @@ -1178,6 +1183,10 @@ def self_test():
if len(classify(tree, selector + ["general/overlay/etc/passwd"],
base_boards=set(tree.built))["rows"]) != full:
problems.append("selector edit plus an overlay file must still widen")
added = set(tree.built) - {WEEKLY_BOARDS[0]}
if set(classify(tree, selector + ["general/overlay/etc/passwd"],
base_boards=added)["rows"]) != set(tree.nightly) | {WEEKLY_BOARDS[0]}:
problems.append("a widening PR that adds a weekly board must build it too")
# 5c. A change to the package index made only of source lines builds the
# boards enabling those packages plus the smoke set; anything else in
# it, or a diff we could not read, widens.
Expand Down
64 changes: 64 additions & 0 deletions br-ext-chip-axera/board/ax620e/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,64 @@
# AX620E / AX630C

Kernel: OpenIPC/linux `axera-ax620e`, Linux 5.15.73 -- the version the
closed Axera media modules of a real Xiongmai IVG-A6 (AX620E) camera were
built against (their `vermagic=`). The branch carries the Axera SDK
drivers ported to 5.15 that the camera and those modules need: clocks,
pinctrl, GPIO, resets, the vendor SPI NOR controller driver + AX_DMAPER,
EQOS Ethernet + internal EPHY, UART, watchdog (also the reboot path), the
built-in OSAL (`AX_OSAL_*`) with `/proc/ax_proc`, hwinfo, os_mem,
ax_hrtimer, and the board devicetree `axera/ax620e-xm-ipc.dts` (the stock
camera's own DT turned back into source).

`ax620e.lite.config` is the branch's `ax620e_xm_ipc_defconfig`. Where the
vendor modules depend on it, it matches the vendor kernel: SMP, PREEMPT,
HZ=100, module unload, no modversions, KPROBES (sizes `struct module`
like the vendor `.ko`). Release must be exactly `5.15.73`.

To build against a local kernel tree instead of the branch tarball:

make BOARD=ax620e_lite LINUX_OVERRIDE_SRCDIR=/path/to/linux

## Flash layout and bootloader

The standard OpenIPC NOR layout (lite): boot 256K, env 64K @0x40000,
kernel 2M @0x50000 (`uImage`, LZMA multi-file with the DTB), rootfs 5M
@0x250000, rootfs_data after it; the NOR is `spi_nor.0` (the devicetree
label the vendor kernel and its `mtdparts=` use).

The boot partition holds the vendor SPL (64K @0, kept: the boot ROM loads
and checks it, it brings up DDR and loads U-Boot's image from 0x10000) and
the OpenIPC U-Boot built from the Axera SDK's U-Boot 2020.04
(`openipc_ax620e_nor_defconfig`, ax_gzip payload behind the SDK image
header, signed with the SDK's test key -- the key the stock image carries).
It starts with OpenIPC's default environment (`bootcmd=run bootflash`,
`osmem`, `extras`, `uk`/`ur`/`fwupd`), saves it on first start, and boots
the kernel slot. With the stock Xiongmai U-Boot instead, the firmware has to
sit at 0x60000 and the env needs `xmuart=0` and the OpenIPC bootcmd.

At boot `S70vendor` runs `load_axera` (ipcinfo -v says `axera`): the 21
vendor modules load in the stock loader's order, `ax_cmm` gets the DRAM
above `mem=` as its pool.

## Tested

In QEMU only (OpenIPC/qemu-bench `-M ax620e`, `-global
axera-spi-nor.env-from-flash=on`), from a whole-flash image of the vendor
SPL, the OpenIPC U-Boot, this build's `uImage` at 0x50000 and
`rootfs.squashfs` at 0x250000, env blank:

- U-Boot saves its default env and boots the kernel slot; kernel
`5.15.73`, `mtdparts=spi_nor.0:256k(boot),64k(env),2048k(kernel),5120k(rootfs),-(rootfs_data)`;
- squashfs root, `/dev/mtdblock4` (rootfs_data) mounted as jffs2 on
`/overlay`, overlayfs on `/`;
- first login sets the root password; `ipcinfo -v/-c/-f` print
`axera`/`ax620e`/`ax620e`; `S70vendor` runs `load_axera`: all 21
modules in `modules.dep` load (`lsmod`), no failure logged, no oops;
- a file written to `/etc` and the password survive a reboot.

Real hardware has not run this image yet: clocks, EPHY, watchdog, SPI
dual reads and the vendor SPL's secure-boot check are unverified there.

Known gaps: no majestic build for Axera (BR2_PACKAGE_MAJESTIC is off);
`ax_vdec` fails its probe in QEMU (no decoder model) exactly as on the
stock firmware there.
Loading
Loading