R

Lost catgirl. Full-stack hacker.

Formerly infosec, professionally. Recovering from deep burnout.

Sometimes refers to ourselves in plural form.

🏳️‍⚧️🏳️‍🌈

Chuuni rights!

♫ Wake up, leave your hesitation ♫

ABC (American-born Chinese) but not in America

Pronouns: she/they//佢//她
Accept-Language: en, yue, zh, de;q=0.5
A/S/L: NaN, maybe (ask nicely), London
Website: https://arcanenibble.com
Email: [email protected]

Public Key

npub1xfvrmeg904uatw05gsrwwnl8j7680j2vg5cdynn9fwskhx6adprqwzx63w

NIP-05 Address

[email protected]

Profile Code

nprofile1qqsrykpau5zh67w4h86ygph8flne0drhe9xy2vxjfej5hgttndwks3sprpmhxue69uhhyetvv9ujumt0d4hhxarj9ecxjmntqyg8wumn8ghj7mn0wd68ytnhd9hx2r6y6gh

Publishing to

relay.momostr.pink

nostr.wine

nos.lol

Author Public Key

npub1xfvrmeg904uatw05gsrwwnl8j7680j2vg5cdynn9fwskhx6adprqwzx63w

Show more details

Last Notes

2026-04-20 00:34:06 UTC

🌟 Project Release Announcement Time! 🎉
I wrote a WebUSB extension for Firefox.
This lets web pages access USB devices (with your permission). This has long been supported by Chrome, but Mozilla has not wanted it, at least by default. Fortunately, because add-ons are a thing (for now. make sure to keep fighting for this!), it's possible to change that!
Common reasons for wanting this include programming microcontrollers, 3d-printers, smartphone bootloaders, and similar "physical computing" projects.
This works on Windows, Linux, and macOS
You can get the source code here: https://github.com/ArcaneNibble/awawausb
In order to make this work, you will need to install a small program on your computer. This is explained in the README. You can then install the .xpi file (on the GitHub Releases page) into your browser (which was _just_ auto-approved while I was typing this announcement up).
Please boost (if you want), test, report bugs, etc. etc. (although do keep in mind that this code is entirely written by a single catgirl)
i think this is the perfect type of project to drop late on a sunday / very early on a monday?
https://mastodon-data.content.as207960.net/media_attachments/files/116/434/172/014/901/094/original/8df883ef611439fe.mp4

2026-03-15 04:08:51 UTC

So, I recently saw some quiet discussion about a paper where researchers reverse-engineered and disclosed some attacks against PhotoDNA, the very-super-duper-secret algorithm used by tech megacorps to scan for illegal images.
They didn't make any code public, and so... I did: https://github.com/ArcaneNibble/open-alleged-photodna
A _complete_ reverse-engineering and commented Python reimplementation of the algorithm from publicly-leaked binaries.
This means that studying the algorithm and any potential flaws is now much more accessible.
This took only about two days (once I knew that there even _was_ a leaked binary to compare against), which just goes to again show that security through obscurity never works.
🔁 encouraged

2026-03-01 03:27:08 UTC

I just documented the protocol that udev uses to broadcast messages to clients: https://arcanenibble.github.io/hardware-hotplug-events-on-linux-the-gory-details.html
Comes with bonus free sample code <strike>on the included CD-ROM</strike>!
This is how programs on Linux listen for events such as "you plugged in a USB device" in order to.... do something useful with that information (such as opening the files on it). Normally, this "just works", but the low-level details of _how_ it works doesn't seem to be that well known.
boosts ok!

2026-01-28 21:16:12 UTC

- reply

it's been pointed out already that ai art ~= clip art, except worse

2026-01-27 18:02:53 UTC

- reply

ehhh just put some brown tape on the blue wire, it'll be fine
(is this allowed by EU-style wiring conventions?)

2025-09-26 23:29:11 UTC

question for the USB experts (boosts okay)
do the USB specification and/or operating system APIs in practice define any sequencing requirements between requests on the control endpoint vs requests on a bulk endpoint?
specifically, i am thinking of the requirement "For endpoints using data toggle, regardless of whether an endpoint has the Halt feature set, a ClearFeature(ENDPOINT_HALT) request always results in the data toggle being reinitialized to DATA0."
if i have a BULK IN endpoint which is *not* halted, and a program running on the host platform tries to simultaneously read from the BULK IN endpoint as well as issuing a ClearFeature(ENDPOINT_HALT), can these requests race each other? what are the permissible outcomes of this situation?

2025-08-23 02:11:25 UTC

- reply

unfortunately, issues of other tracking sources, leakage through keyboards/IMEs, etc. are "out of scope", and individual apps can't really do anything about it thanks to how mobile phone operating systems are built (i.e. part of the tradeoff of their app-based security model)
imo signal's desktop app is... acceptable. a lot of the mess here is because GUI toolkits are *also* themselves a mess (electron didn't get popular for no reason, have you _tried_ developing a native app for multiple desktop platforms?)
signal is definitely a bit hostile towards outsiders, but once again they have justifications for it (it is easier to evolve, including fixing security issues in, an ecosystem that they have control over)
i hate this situation too

2025-08-23 01:34:21 UTC

- reply

serious answer: because mobile phones actually do have (in certain specific scenarios) "better" security than desktops, and secondly as a calculated strategic/political move to maximize reach/impact
mobile phones tend to have actual security boundaries between apps, whereas desktops traditionally do not and only have security boundaries between users (which isn't very useful on single-user *personal* computers, as this security model predates them)
like it or not, the balance between desktop/mobile traffic on the internet has completely flipped over towards mobile about a decade ago, so "most users" really are accessing the internet via mobile devices. part of making "privacy" software actually useful is that "normal" people also need to be using it (you don't want to become "the app only used by drug dealers and terrorists")
it's a shitty situation without any easy ways out

2025-08-16 19:20:12 UTC

- reply

i think i'm missing where exactly you're coming from, but i've generally found, at least with the work i've previously done or been drawn towards, that "comments" are typically the least powerful/useful way to convey thinking
i've generally gotten much more value out of architecture diagrams, project-planning-related documents, or even academia-style algorithm explanations

2024-12-22 23:58:31 UTC

- reply

first image is a SH-2 memory map, second image is a SH-4 memory map
link dump:
https://segaretro.org/images/2/2c/SH7604_Hardware_Manual.pdf
https://retrocdn.net/images/8/88/Hitachi_SH7095_Hardware_User_Manual.pdf
https://www.renesas.com/en/document/mah/sh7750-sh7750s-sh7750r-group-users-manual-hardware?r=1055161
https://mastodon-data.content.as207960.net/media_attachments/files/113/699/180/058/697/876/original/336d9dc1d619e707.png
https://mastodon-data.content.as207960.net/media_attachments/files/113/699/181/342/217/916/original/5cdfb3b0520eb530.png

2024-12-22 22:15:46 UTC

- reply

mostly that they tended to live at the upper end of the address space, and a UART ("SCI") and timer of some kind was pretty common on many parts
the rest of the address space tended to be decoded into separate external chip select lines, some of which could interface things like sdram. on some parts (sega-consoles-era) different physical addresses also controlled things like cacheability

2024-12-06 15:12:33 UTC

- reply

https://mastodon.social/@whitequark/113522701844163803

2024-11-21 20:07:37 UTC

- reply

can you maintain "feminine" physical traits (i.e. fat distribution, skin texture) with that much T, or do you not particularly care?

2024-11-11 12:49:45 UTC

- reply

oh wait, is this specifically for the cockpit and not the rest of the aircraft?

2024-11-11 12:48:56 UTC

- reply

seems reasonable to make sure someone doesn't without thinking grab an alcoholic beverage from the nearest catering trolley

2024-11-09 17:41:39 UTC

- reply

in the end, the decisions are up to you. we are currently a funemployed autistic burnout stuck in administrative limbo, you should weigh our advice accordingly. we're just trying to make our own attempt to convey some of what we've been forced to realize the hard way over time. it may not apply to you! 🤷‍♀️

2024-11-09 17:36:44 UTC

- reply

again, it's not particularly about any specific platforms, it's about _vibes_
for example, you could easily stash a small collection of black friday bestbuy craptops to test on for not very much money
the specific object-level details of _how_ you bring "magic" or "vision" or however you describe it down into something like a "test matrix" is of course up to interpretation

2024-11-09 17:26:20 UTC

- reply

it seems to be more focusing on the general lack of emphasis placed on testing against "baseline" hardware, rather than any specific bug

2024-11-09 11:48:02 UTC

- reply

oh dang, that's way more streamlined compared to when channers were doing this over a decade ago to manipulate a poll for the lulz

2024-11-06 03:09:03 UTC

- reply

... okay, the nRF52 SoftDevice stack just natively supports at least 4 concurrent connections, so that should "just work" if you build a device around that chip

2024-11-06 03:04:43 UTC

- reply

can you interleave two ble frequency hop patterns if you have sufficient low-level control of the stack and enough patience to debug it?

2024-10-30 11:35:03 UTC

- reply

something potentially for your toolkit?

2024-10-30 11:34:43 UTC

oh btw jsyk python supports being built for WASM WASI, including linking with Rust PyO3
it's just not properly documented and involves a bunch of build system fuckery
https://mastodon-data.content.as207960.net/media_attachments/files/113/396/154/178/862/643/original/74ecaf1f6c6ac41d.png
https://mastodon-data.content.as207960.net/media_attachments/files/113/396/154/387/528/914/original/965ce651a71422cc.png

2024-10-23 14:37:37 UTC

- reply

fwiw there are already existing VST wrappers for Linux, such as LinVst
what is it you need that isn't already addressed by that?

2024-10-23 14:00:57 UTC

- reply

if you do not care about doing CPU emulation, you can "just" load/run 32-bit x86 code in a 64-bit x86_64 process by changing the cs register to 0x23 (and 0x33 to get back). silly infosec people have named this trick "Heaven's Gate". you might also want personality(2) to set ADDR_LIMIT_3GB
glhf with the magic around setting up the FS/GS segments and vDSO if you actually try to do this
WINE's "new WoW64" might already do something to this effect?

2024-10-17 16:22:22 UTC

- reply

this is just a static-PIE executable, no?

2024-09-24 19:08:26 UTC

- reply

Rust has this type of hacks too!
https://github.com/rust-lang/rust/blob/c682aa162b0d41e21cc6748f4fecfe01efb69d1f/compiler/rustc_hir_analysis/src/check/wfcheck.rs#L141-L167
https://mastodon-data.content.as207960.net/media_attachments/files/113/194/098/773/495/214/original/3128b601d74141f5.png

2024-08-19 11:17:53 UTC

- reply

tangentially related, interesting things happen if you're Nintendo and want hard hold plating everywhere for some reason (durability/reliability? solderability?)
https://gbhwdb.gekkio.fi/static/AGB-BPEJ-0/fexcollects-1_04_without_battery.jpg
see the drill holes near "U3 RTC", the ROM, and the "AGB" text, which are separating traces that were being used to short pads together for the plating process

2024-08-11 01:49:41 UTC

- reply

oh noes, but fyi we contributed black magic debug support for it a while back if you want to try that (not in the normal build because of code size limits though)

2024-08-09 08:45:22 UTC

- reply

all of the horrible formats other than FST (i.e. LXT/LXT2/VZT) are no longer supported and have been removed in their working master branch (after I filed an issue reporting multiple memory corruptions across all of them)