Kernel tree exists only as an uncommitted working tree on ouranos #4

New issue

Closed

opened 2026-08-27 23:22:23 +00:00 by tiagoagueda · 1 comment

tiagoagueda commented

2026-08-27 23:22:23 +00:00

Owner

uname -r is 7.2.0-g45c13f3f9e3b-dirty, built from ~/a80/linux on ouranos, which has
no commits. Modified there:

arch/arm/mach-sunxi/mc_smp.c — A15 v-f tables, rail enable, domain isolation, NEON reset
ordering, and the cpu_table fix (an upstream bug)
drivers/soc/sunxi/sunxi_arisc.c + include/linux/soc/sunxi/arisc.h — ~330 lines that
exist nowhere else; the AR100 beacon handshake was expensive to work out and is undocumented
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c — the MDIO settle poll
arch/arm/boot/dts/allwinner/sun9i-a80.dtsi and the board DTS
drivers/phy/allwinner/phy-sun50i-usb3.c, drivers/thermal/sun8i_thermal.c, clk and cpufreq
glue

If that host's disk dies, this board becomes unreproducible and every fix — DRAM sizing, USB3,
the real pin map, the MDIO settle — is gone. This is the only risk in the tracker that is
genuinely irreversible.

Done when

the tree is committed to a branch on ouranos
a patch series is exported into this repo under patches/
the board can be rebuilt from git alone, with ouranos treated as disposable

`uname -r` is `7.2.0-g45c13f3f9e3b-dirty`, built from `~/a80/linux` on ouranos, which has **no commits**. Modified there: - `arch/arm/mach-sunxi/mc_smp.c` — A15 v-f tables, rail enable, domain isolation, NEON reset ordering, and the `cpu_table` fix (an upstream bug) - `drivers/soc/sunxi/sunxi_arisc.c` + `include/linux/soc/sunxi/arisc.h` — **~330 lines that exist nowhere else**; the AR100 beacon handshake was expensive to work out and is undocumented - `drivers/net/ethernet/stmicro/stmmac/stmmac_main.c` — the MDIO settle poll - `arch/arm/boot/dts/allwinner/sun9i-a80.dtsi` and the board DTS - `drivers/phy/allwinner/phy-sun50i-usb3.c`, `drivers/thermal/sun8i_thermal.c`, clk and cpufreq glue If that host's disk dies, this board becomes unreproducible and every fix — DRAM sizing, USB3, the real pin map, the MDIO settle — is gone. **This is the only risk in the tracker that is genuinely irreversible.** **Done when** - [ ] the tree is committed to a branch on ouranos - [ ] a patch series is exported into this repo under `patches/` - [ ] the board can be rebuilt from git alone, with ouranos treated as disposable

tiagoagueda added the

infra

P1-critical

labels

2026-08-27 23:22:23 +00:00

tiagoagueda referenced this issue

2026-08-27 23:36:38 +00:00

Upstream: mc_smp cpu_table fix (sunxi) #27

tiagoagueda referenced this issue

2026-08-27 23:36:39 +00:00

Upstream: RFC the AR100 driver before writing v1 #31

tiagoagueda commented

2026-08-27 23:44:47 +00:00

Author

Owner

Done 2026-08-28. Both working trees are committed on draco-aw80 branches on ouranos, and exported into patches/ as ten series split along maintainer boundaries.

The AR100 driver was the irreplaceable part. Verified after export by reconstructing the file from the patch and comparing against the original:

de76fc4a2bafa1de54d03f73cecb0d1a   396 lines   identical

Bases: linux 45c13f3f9e3b, u-boot 964ad5b5c91b (v2026.10-rc3). patches/configs/ holds savedefconfig output for both, so a build is reproducible without the original .config.

Not fully closed by this, deliberately: the trees on ouranos are committed but still live only on that host. The patches here are a faithful export, but an export is not the tree with its history. Pushing those branches to a remote of their own would close the loop.

Nothing has been submitted upstream.

Done 2026-08-28. Both working trees are committed on `draco-aw80` branches on ouranos, and exported into `patches/` as ten series split along maintainer boundaries. The AR100 driver was the irreplaceable part. Verified after export by reconstructing the file from the patch and comparing against the original: ``` de76fc4a2bafa1de54d03f73cecb0d1a 396 lines identical ``` Bases: linux `45c13f3f9e3b`, u-boot `964ad5b5c91b` (v2026.10-rc3). `patches/configs/` holds savedefconfig output for both, so a build is reproducible without the original `.config`. **Not fully closed by this, deliberately:** the trees on ouranos are committed but still live only on that host. The patches here are a faithful export, but an export is not the tree with its history. Pushing those branches to a remote of their own would close the loop. Nothing has been submitted upstream.

tiagoagueda closed this issue

2026-08-27 23:44:47 +00:00