Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
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
1 change: 1 addition & 0 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -81,6 +81,7 @@ One binary to rule them all. The 1bit engine is model agnostic: any Hugging Face
| 2026-10-30 | The HRX pin carries the DeepSeek-V4.1 port; ONNX Runtime runs in-process with our own execution provider on the NPU; the NPU and the GPU are measured running at the same time ([#672](https://github.com/1bit-MONSTER/engine/pull/672), [#647](https://github.com/1bit-MONSTER/engine/pull/647), [docs/hybrid.md](docs/hybrid.md)) |
| 2026-11-01 | **v2026.44 is a draft**: candidate 19 is gated in software (1959/1959 kernel ops, NPU 82.03 tok/s, `VMTEST` pass through install and installed-disk boot, `VMTEST-RELEASE` 10 pass / 9 skip / 0 fail) but has **not** been booted or installed on hardware; candidate 20 takes the full gate, hardware boot and install included, before it is published ([docs/releases.md](docs/releases.md)) |
| 2026-11-02 | **1bit OS becomes a Puppy Linux appliance**: `ResolutePup64` (woof-CE, Ubuntu 26.04 base) with our MOK-signed kernel, Lemonade and the engine embedded, headless by default with a `desktop` boot entry, `apt` and a persistent `1BIT-DATA`; it boots on the board with Secure Boot on ([docs/os-v3.md](docs/os-v3.md), [os/README.md](os/README.md)) |
| 2026-11-14 | The Bosgame BeyondMax's rear USB-A and Type-C ports work under Linux: a plug test on every external port passed once the keep-awake rule holds the xHCI controllers awake (USB4 devices still untested). The MTP drafter no longer crashes on a model's first request, and the HIP kernel and HRX pins moved forward through the hardware gate ([docs/axb35-usb-ports.md](docs/axb35-usb-ports.md), [#2431](https://github.com/1bit-MONSTER/engine/pull/2431), [#2453](https://github.com/1bit-MONSTER/engine/pull/2453), [#2461](https://github.com/1bit-MONSTER/engine/pull/2461)) |

The release goes out when every item on the release checklist is done ([docs/releases.md](docs/releases.md)); the weekly Lemonade pin moves with Lemonade's releases.

Expand Down
36 changes: 35 additions & 1 deletion docs/axb35-usb-ports.md
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,33 @@ limitations under the License.
-->
# AXB35-02 (Bosgame BeyondMax): the rear USB-A and Type-C ports

**Status: fix merged, mechanism verified on the board, functional plug test still outstanding.** engine#823.
**Status: verified on hardware 2026-11-14 (BIOS 1.10): every external port enumerates with the keep-awake rule; USB4 device untested.** engine#823.

| physical port | xHCI controller | measured with the controllers kept in D0 |
|---|---|---|
| front USB-A | `c7:00.0` (`1588`) | worked before the rule as well (the one controller that stays in D0) |
| front Type-C | `c7:00.3` (`1589`) | SuperSpeed 5 Gbps, USB storage |
| rear Type-C | `c7:00.4` (`158b`) | SuperSpeed 5 Gbps, USB storage |
| rear USB-A | `c7:00.0` (`1588`), port 4 | high-speed with a USB 2 stick; a USB 3 device on this port is untested |
| rear USB-A next to the power connector | `c5:00.4` (`1587`) | SuperSpeed 5 Gbps, USB storage |

How it was measured: on the board's Ubuntu host (kernel `7.4.0-rc1-1bit`), first with `power/control=on` written by hand on the
four xHCI functions, then with `91-1bit-xhci-keepawake.rules` installed in `/etc/udev/rules.d`. Both produce the same state, and
the rule's device list matches these four controllers exactly. Before the change, the three silent controllers had taken no
interrupts since boot, and every port read `RxDetect` with no connect. So the D3cold autosuspend is the whole cause for USB 2 and
USB 3 devices.

Not covered yet:

- a USB4/Thunderbolt device on a Type-C port (both USB4 routers still report no link);
- a USB 3 device on the `c7:00.0` rear USB-A port;
- a 1bit OS boot with the rule loaded from the image.

One enclosure (a Realtek RTL9210, the device from the original report) enumerates at high-speed (USB 2) on **every** port, including
ports where another drive reaches 5 Gbps. That limit belongs to the enclosure or its cable, not to the board.

The short, copy-pasteable version (symptom, rule, check) is the blog post
[Bosgame M5 USB-C not working on Linux: the fix](https://1bit.gg/blog-bosgame-m5-usb-c-linux-fix.html).

On this board, stock boots leave three of the four xHCI functions in `D3cold`, so the rear USB-A ports and both Type-C ports never
enumerate; only the controller that carries the front USB-A (and Bluetooth, and the JetKVM) stays in `D0`. Windows drives every
Expand Down Expand Up @@ -83,6 +109,10 @@ Measured on the board 2026-11-10, BIOS 1.10, a dev boot with the rule not instal
0000:c7:00.6 0x158e thunderbolt D3cold <- the rule does NOT match this
```

*Corrected 2026-11-14 by the plug test above:* USB 2 and USB 3 devices on **both** Type-C ports run through xHCI functions the
rule covers (`c7:00.3`, `c7:00.4`); only a USB4/Thunderbolt tunnel uses the routers. The reading below was written before the plug,
and is kept as written.

The **rear USB-A** ports hang off the four xHCI functions the rule matches. The **Type-C** ports hang off the two USB4 host routers
(`0x158d`/`0x158e`), which the rule's device list does not contain, and which are in `D3cold` with the `thunderbolt` driver bound and
four live IRQs. So:
Expand Down Expand Up @@ -117,6 +147,10 @@ USB4 side is alive at PCI level and the missing thing is below it.

## What is still owed

*Updated 2026-11-14:* the plug test on a host with the rule passed on every external port (table at the top). What remains is
the same plug **on a 1bit-OS boot** with the rule from the image, a USB 3 device on `c7:00.0`, and a USB4/Thunderbolt device.
The paragraph below is the earlier plan, kept as written.

One device into a rear USB-A or a Type-C port **on a 1bit-OS boot**, checked with `--require-awake` first. If a port is then still
dead the next diagnostic is the one no one has run: compare `_DSM` on the xHCI/USB4 ACPI nodes against what Windows sends. Until
that plug, the honest row is *"fix merged and its mechanism verified on the board; functional confirmation pending"* — which is what
Expand Down
25 changes: 13 additions & 12 deletions docs/install.md
Original file line number Diff line number Diff line change
Expand Up @@ -87,8 +87,8 @@ appliance is blank ([os-v3.md](os-v3.md)).

## 1. Run from the stick (no install)

1. Put the stick in the machine. On the Bosgame BeyondMax use the **front USB-A** port
([7](#7-the-bosgame-beyondmax-has-one-usable-usb-port)).
1. Put the stick in the machine. On the Bosgame BeyondMax, boot from the **front USB-A** port
([7](#7-the-bosgame-beyondmax-usb-ports)).
2. Boot it. The default entry, `1bit OS (headless console, 7.4.0-rc1-1bit)`, comes up headless:
a console with no X, whose first screen prints the box's address and how to connect to it (web UI,
API, `ssh`), above a menu whose item 1 turns the Lemonade API key on (the shipped default is no key)
Expand Down Expand Up @@ -266,18 +266,19 @@ mokutil --sb-state # SecureBoot enabled
mokutil --list-enrolled | grep -i "1bit\|strixhalo"
```

## 7. The Bosgame BeyondMax has one usable USB port
## 7. The Bosgame BeyondMax USB ports

On the Bosgame BeyondMax (`AXB35-02`) **only the front USB-A port works under Linux**: a drive plugged into
the front or rear Type-C, or a rear USB-A port, produces no kernel event at all — the xHCI port registers
stay in RxDetect, and three of the four controllers have taken no interrupt. The same ports work in Windows
on the same box, so it is a firmware/EC gap, not a driver in this image
([#823](https://github.com/1bit-MONSTER/engine/issues/823)). Boot from the front USB-A port.
On the Bosgame BeyondMax (`AXB35-02`), three of the four xHCI controllers autosuspend to `D3cold` under Linux. A drive plugged
into the front or rear Type-C port, or into a rear USB-A port, then produces no kernel event at all, while the same ports work in
Windows ([#823](https://github.com/1bit-MONSTER/engine/issues/823)). The image ships a udev rule,
`91-1bit-xhci-keepawake.rules`, that keeps the four controllers awake. With them kept awake, **every external port enumerated** in
a plug test on hardware on 2026-11-14 (BIOS 1.10, on a host with the rule installed).

That is measured on **BIOS 1.09 and, since the owner's reflash, on 1.10** (11/04/2026): the newer firmware
does not change the picture — three of the four controllers still autosuspend to `D3cold` — so the front
port is still the one to use. The capture, the per-controller states and the keepawake rule that follows
from them are in [axb35-usb-ports.md](axb35-usb-ports.md).
Boot the stick from the **front USB-A** port all the same. It is the one port that works before the rule applies, and the same
plug test has not yet been run on a 1bit OS boot, where the rule comes from the image. USB4/Thunderbolt devices are untested. The
per-port results, the controller states and a read-only check (`tools/xhci_keepawake_check.sh`) are in
[axb35-usb-ports.md](axb35-usb-ports.md). The same fix for another Linux install on this board is in
[Bosgame M5 USB-C not working on Linux: the fix](https://1bit.gg/blog-bosgame-m5-usb-c-linux-fix.html).

## What was verified for this page

Expand Down
5 changes: 3 additions & 2 deletions docs/kernel.md
Original file line number Diff line number Diff line change
Expand Up @@ -52,8 +52,9 @@ list, with the facts that go stale marked as facts and the commands that do not.
rebuild on any machine that installs them — `scripts/mk-kernel-packages.sh` builds the set for deb,
rpm and Arch, and [install-any-distro.md](install-any-distro.md), "Kernel and NPU driver packages",
is the install. The board's other firmware-side quirk is not a module at all: three of its four xHCI
controllers park in `D3cold`, so the rear USB-A and both Type-C ports never enumerate, and the fix
that ships for it is a udev rule — written up, with the ACPI evidence and the one-command check, in
controllers park in `D3cold`, so the rear USB-A and both Type-C ports never enumerate. The fix that
ships for it is a udev rule that keeps them awake; with it, every external port enumerated in a plug
test on hardware (2026-11-14). It is written up, with the per-port results and the one-command check, in
[axb35-usb-ports.md](axb35-usb-ports.md).

## 7.4 on the box: what changed, and what to watch
Expand Down
3 changes: 2 additions & 1 deletion docs/registry.md
Original file line number Diff line number Diff line change
Expand Up @@ -44,7 +44,8 @@ them `zamba`, `zamba2`, `blackmamba`; `qwen4exp` is back since pin `e57beb97`) a
The `zinc` column (ZINC's `parseArchitecture`; 63 HF architectures on 2026-10-03) left with ZINC in
the core strip (2026-10-04).

On 2026-10-04 (core strip): 290 HF architectures mapped, hrx 290, npu 7. The file
At the current pins (llama.cpp `8c12dfa6`, 2026-11-15): 350 HF architectures mapped, hrx 350, npu 9, and 171 GGUF
architectures in `gguf_architectures.hrx`. On 2026-10-04 (core strip): 290 HF architectures mapped, hrx 290, npu 7. The file
also lists `gguf_architectures.hrx`, every GGUF architecture the pinned fork's runtime builds (141
then): CMakeLists.txt compiles that list into `1bit serve`, which refuses any other architecture on
hrx and cpu with a pointer to Lemonade's llamacpp backend. Mapped means a backend's code accepts the architecture; the census below reports how many were checked. `registry/architectures.json` always holds the current counts and the pins they come from: every pin bump regenerates it (`scripts/registry-regen.sh`, run by the bump workflows), and CI's `registry_pins` test fails a pin that moved without it.
Expand Down
2 changes: 1 addition & 1 deletion docs/releases.md
Original file line number Diff line number Diff line change
Expand Up @@ -172,7 +172,7 @@ What ran on **c19** (Strix Halo and QEMU, 1 November) — every gate except the
| `sha256sum -c SHA256SUMS`, `sbverify`, module/kernel identity | 3/3 OK, signature OK, one module tree for the shipped kernel |
| stick write | device readback `3c3c08c2…` **= the image**, `prepare-stick` 11/11 and `--check` exit 0, `/data` grown to the whole device |
| no DwarfStar | **0** `ds4/` files and **0** `hipblaslt` references; the image's catalogue has no `ds4` recipe |
| AXB35-02 rear USB-A and Type-C ports ([#823](https://github.com/1bit-MONSTER/engine/issues/823)) | on this board three of the four xHCI functions autosuspend to `D3cold`, so only the controller behind the front USB-A enumerates; the image ships `91-1bit-xhci-keepawake.rules` to keep the four awake, and **a device on a rear port has not yet been confirmed to enumerate on a 1bit-OS boot**. Check with `tools/xhci_keepawake_check.sh --require-awake` (read-only; verdict `all awake` is the only one a plug test is valid under), mechanism and per-boot captures in `docs/axb35-usb-ports.md` |
| AXB35-02 rear USB-A and Type-C ports ([#823](https://github.com/1bit-MONSTER/engine/issues/823)) | on this board three of the four xHCI functions autosuspend to `D3cold`, so only the controller behind the front USB-A enumerates; the image ships `91-1bit-xhci-keepawake.rules` to keep the four awake, and with the four kept awake **every external port enumerated in a plug test on hardware on 2026-11-14** (BIOS 1.10, on a host with the rule installed). **The same test on a 1bit-OS boot, and a USB4/Thunderbolt device, are still to be done**. Check with `tools/xhci_keepawake_check.sh --require-awake` (read-only; verdict `all awake` is the only one a plug test is valid under), mechanism and per-boot captures in `docs/axb35-usb-ports.md` |

**What did not run on c19: the hardware boot and the hardware install.** The only hardware evidence in this document is
candidate **c18**'s — a different, earlier candidate (`rc-2026-10-31c`): live USB boot with Secure Boot on,
Expand Down
Loading