Replies: 2 comments
|
Canaan Avalon Nano 3s Mujina src code repo: https://github.com/aadhi1014/nano-mujina |
0 replies
|
Summary and transcript are now in the top post. |
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Summary generated by AI from the meeting recording.
Post-Meeting Summary and Transcript
Attendees: Ryan (host), Aadhi, Brett, Dylan, Econoalchemist, PizzAndy, Schnitzel, Susan. Referenced but absent: Jayr, Ronald, Skot, Tyler.
Summary
RY3T Nova: surplus solar to hashrate (01:18)
Schnitzel presented Nova, RY3T's product built on Mujina and the subject of 256F's X post earlier that day. Nova packages an S19j Pro with a quieter radial fan, and its Rust binary runs directly on the stock control board, with no added Raspberry Pi. It reads excess solar production from the home's inverter or smart meter over Modbus or MQTT and follows it within three to five seconds. Units stack and discover each other, and the mined bitcoin goes to the homeowner. Nova answers the collapse of European feed-in tariffs. Swiss injection rates fell from about 30 cents to 6, and large systems now pay real-time prices, sometimes negative, to inject. Most of Nova should become open source; partner licensing is still being worked through, and Mujina's license requires opening only changes to Mujina itself, not the whole product.
Dependency supply chain hardening (10:19)
Ryan walked through the merged supply chain PRs (#91 and #97), prompted by the Coldcard discussion at the last call (#87). Cargo.lock now pins every crate, tools and GitHub Actions and container images are pinned, and updates pass through cargo-cooldown, which refuses crate releases younger than seven days. A nightly GitHub Actions job audits dependencies against the advisory database and files issues, with a fast-forward path to update a single package immediately when an advisory demands it. Schnitzel suggested seven days may be too short (his day job settled on two weeks), but seven is already among the longest cooldowns Ryan found in the wild, so it stands.
Red team findings and a security policy (15:24)
Schnitzel's "256 red team", Kimi K3 in a pen-test harness under opencode, probed Mujina by static analysis and by fuzzing the CPU miner with a malicious pool sending non-standard Stratum V1 traffic. Since Stratum V1 is unencrypted, the same payloads could come from a man in the middle, so a misbehaving upstream could take down a mining site. Ryan fixed one category in #89; three similar reports remain open, none urgent (#90, #98, and #99). Because the reports arrived as patch PRs and some fixes belong elsewhere, Ryan will convert some to issues and add a SECURITY.md describing how to report, which reporting agents look for. Anyone is welcome to pick up an open report; comment with the intended fix first, since several need design decisions.
Open PRs: config, SV2, BZM2 (22:19)
Jayr's first configuration implementation PR is in (#88), and his SV2 series (#65) is about halfway through review. Ronald's BZM2 series (#68 through #71) changes code beyond the BZM2 chip itself, so Ryan will reach out about which parts are meant for mainline.
Porting Mujina to the Avalon Nano 3S (24:25)
Aadhi told the story behind his announcement (X thread). Canaan open sourced their cgminer for the Nano 3S, which let him build an image with SSH enabled. When poking the stock RTOS from software stalled, he put logic analyzers on the connectors and reverse engineered the ASIC protocol by capture and replay; a missed power enable cost two weeks. He rebuilt the RTOS side from scratch on Canaan's Kendryte SDK for the K230D (big core runs the RTOS, little core runs Linux), wrote his own IPC, and connected Mujina to it on the Linux side, replacing the stock miner binaries. A stubborn hashrate shortfall turned out to be thermal. Each chip's SHA engines choose among four PLLs by temperature and error rate, so running the fans at maximum kept the chips too cool to pick the faster PLLs. The RTOS code will live in its own repo, nano-mujina, with only the IPC glue in mainline. The Mini and Cube share the processor and should port easily; the A15 series (K210) will be harder. He also applied for a job at Canaan.
S19 interest and official installers (38:21)
A new reply on the forum's S19 hacking thread asked for an easy way in. Ryan's current milestone is the answer. He is merging S19 and S21 support from Schnitzel's and Skot's forks and his own local work so the mainline driver covers all the Bitmain chips at once, and will then wrap official installers around it. Until that exists, S19s are not for the average user. The website now carries a hardware compatibility matrix. Tyler's X Space with Canaan, much of it on hashrate heating, is worth a listen.
Buzz as the future chat home (41:06)
256F is trying Buzz internally as part of consolidating its infrastructure; Schnitzel set it up mostly for Tyler to evaluate, and it is not public yet. If adopted it would replace Telegram rather than add another channel. Dylan wrote one of the Buzz mobile apps, still buggy but workable, which matters because phone notifications were Ryan's main concern about committing to it.
Follow-ups and wrap (43:51)
Two items from last call remain open: writing up the user promises (what the project owes users on safety and security even without holding keys) and making the bm13xx chip reference reachable from the website. A quiet open floor ended a short call at about 47 minutes.
Action items
Transcript
Full transcript on the wiki: https://github.com/256foundation/mujina/wiki/dev-call-20260817
Agenda (as planned)
Agenda
Meeting: Monday 2026-08-17 at 1700 UTC (noon US Central), https://meet.jit.si/mujina-dev-call
Note: These meetings are recorded and transcribed by AI.
Purpose: keep everyone aware of what's happening, roughly following the
development workflow. Not a replacement for the other communication
channels.
Announcements & host topics
the first commercial mining product built on Mujina. Nova turns
surplus solar into hashrate, tracks PV curves in seconds, and
mixes hashboards on stock control boards.
PRs
build: harden the dependency supply chain #91 and
chore(deps): clear the advisories and update the lock #97
container base images are pinned, and
cargo-cooldown
holds fresh crate releases back before they can enter the
tree. The four dependency advisory issues the audit raised
are closed.
the Stratum v1 client against a misbehaving pool.
its consumers.
silently.
want them as issues first? The repo has no SECURITY.md yet;
one could set that expectation.
design (MIP-0001: Configuration and API mujina-mips#1):
a config tree with sources. Bears on the pool configuration
issue, feat: pool configuration via MIP-001 configuration and API #83.
license allowlist (a transitive dependency under the ISC
license); being sorted in the PR thread.
through feat(api): BZM2 diagnostic endpoints and hardware documentation #71
from the last call).
Issues
Discussions
that his Avalon Nano 3S is hashing under Mujina: he reverse
engineered the miner, built the port, and open sourced it. He
plans a
Good Vibes
discussion with the details. In a
follow-up,
the Nano 3S runs DOOM.
Forum & Chat
answer: Schnitzel, Skot, and AgentP each have S19 forks, but
installing still means SSH sideloading; no easy-load image
yet.
tested on his Bitaxes
(message).
control board while building a loader for flashing
Bitaxe Raw style
firmware; recovery over direct USB in progress
(message).
live on X with Canaan in America
and took questions
(message).
channels, and a Buzz relay experiment may eventually offer
public channels again
(message).
Follow-up
Still open from the last call's action items:
(call section).
bm13xx chip reference
from mujina.org
(call section).
Other
Open floor
All reactions