mjg59

9,131 karmajoined vor 15 Jahren

Linux kernel and security developer at Nvidia, experienced in power management, mobile hardware and firmware interaction. I'm at https://nondeterministic.computer/@mjg59/ on the Fediverse and https://bsky.app/profile/mjg59.eicar-test-file.zip on Bluesky, and blog at mjg59.dreamwidth.org . I used to work on fruitflies. This is more relaxing.

[ my public key: https://keybase.io/mjg59; my proof: https://keybase.io/mjg59/sigs/1nzROHRbn3xz5zA0YGMRsbiPahYUreIHH3JNmK3a1oE ]

comments

vor 8 Stundendiscuss

My understanding is that it's not so much that the novelisation is of the director's cut, but that the people writing them are typically working from (at best) the shooting script (and at worst, an earlier version of the script). The book needs to come out at around the same time as the movie, and there simply isn't enough time to re-edit it and print it after the final editing decisions are made. You'll frequently find sections in the book version that only end up on-screen as a director's cut or in deleted scenes, and sometimes just never at all.

vor 3 Tagendiscuss

You say that, and also remote attestation is how Signal knows it's talking to a legitimate SGX enclave running the expected payload

vor 4 Tagendiscuss

Isn't that lower than a purely resistive heater?

vor 5 Tagendiscuss

As someone with experience (albeit almost 20 years out of date) experience of wet lab DNA collection and sequencing - this stuff is hard, you will fail a lot, and you will fail a lot more if you don't have an extremely clean environment to do this in. And once you have data, you should be asking yourself how accurate it is given the environment you collected it in, you should be looking at correlated sequence errors that are not taken into account[1].

But also: genetic counselling is a real thing that real people study. Please don't ask an LLM questions about what your genes are going to do to you without having access to someone who has the ability to contexualise the data and put you in touch with relevant experts. I have a PhD in this and I would not trust myself to be able to interpret data about myself in a detached and rational way.

(And: why is the link to Molecular Biology of the Cell to the 6th edition, when the 7th came out 4 years ago? Random fact: the first three editions were co-authored by my supervisor during my first PhD attempt, who went on to demonstrate that Roger Penrose's ideas about the importance of microtubules in chemotaxis in E. coli were absolute bullshit. Great guy)

[1] I spent a while analysing very early (by commercial standards) Illumina data in 2007, and being able to align stuff to reference genomes made it possible to identify certain biases. Nanopore technology is likely to have more of those, and if you don't have the ability to take those into account you may have a very bad time

vor 5 Tagendiscuss

We're talking about the equivalent of booting with init=/bin/sh, I don't remember kernel init on x86 ever being that slow

vor 8 Tagendiscuss

ARM isn't consistent in the way that the x86 platform is. Each vendor's SoC usually requires specific support.

vor 10 Tagendiscuss

"Quality" is maybe overloaded. Upstream requires the code to meet their sense of taste, and some of that is about quality, and some of that is about undocumented design concepts. It's not hard to meet the quality bar. Meeting the design requirements is extremely hard.

vor 10 Tagendiscuss

Linux tries to avoid special cases. That means that when someone shows up with a new driver that's either not something that fits into an existing category, or which sort of (but doesn't entirely) overlap with an existing driver, there's an extended set of design discussions about how to make this new driver fit into existing infrastructure in a way that's consistent with what's there and which also allows new things to exist.

That sounds great from a design perspective, but it can also lead to cases where people are attempting to design for utter unknowns - potential futures that may or may not exist, theoretical understandings of how hardware works, that kind of thing. It frequently prevents new drivers being merged without significant modification, and sometimes it results in a need to entirely rearchitect the relevant part of the kernel before the driver can even be considered (and also now you need to split that driver into three parts). Upstreaming is hard.

vor 10 Tagendiscuss

If you've ever been developing software for a Mac for any length of time and searched for a weird error, there's an extremely high probability you've ended up on a post written by Quinn that gives you a pointer to how to fix it.

vor 10 Tagendiscuss

My coworkers like to complain that searching for anything they're working on leads them to either old blog posts written by me, or (if they're currently working on MacOS issues) posts by Quinn. It's funny because it's entirely my experience as well. Apple's attitude towards secrecy means that a huge amount of knowledge is simply never shared, and we're left with Quinn as an incredibly rare portal of knowledge between the inside of Apple and the rest of the world. Quinn, you've apparently seen some shit. Thank you for sharing it with us. I've worked with at least three teams who could never have deployed what we did without you.

vor 12 Tagendiscuss

There's US1549 and FR4102 - are there any others? But yes, the assumption about failures being independent is obviously violated by large flocks of birds, although this has also resulted in the loss of 4-engined aircraft.

vor 13 Tagendiscuss

> or indeed any I/O ports.

PCI has the concept of I/O ports even if the CPU doesn't. The details differ depending on whether it's PCI or PCIe, but the net effect is that whatever's responsible for putting the access on the bus exposes an MMIO window that corresponds to the the port addresses available, and generates port I/O cycles in response to accesses to that window.

vor 13 Tagendiscuss

All else being equal, potentially - although as I mentioned there have been cases where one engine falling off a 4 engine aircraft hit another in the process. But ETOPS certification is based on it being demonstrated that engines are sufficiently reliable that the probability of an independent failure is incredibly unlikely, and also requires that operators have a stricter maintenance process. The only dual engine failures on modern two-engine aircraft I can think of off-hand have been fuel exhaustion (either actually being out of fuel, or ice blocking fuel filters in the case of BA38), and would have affected 4-engine aircraft just as badly.

vor 13 Tagendiscuss

More people have died due to one engine falling off a 747 and knocking off the other engine on the same wing than have died due to dual engine failure on an ETOPS certified aircraft

vor 18 Tagendiscuss

Shim, the first stage bootloader on Linux, is designed to be updated infrequently. Distributions embed their own signing certificate in it and have that binary signed by Microsoft. The actual bootloader (typically either grub or systemd-boot) is then signed with the distribution certificate, as is the kernel. Distributions get to set their own policy around how long that certificate lasts for, it's entirely unrelated to the Microsoft certificate expiry.

vor 19 Tagendiscuss

It's actually largely because I demanded it, our customers weren't paying attention at all - Fedora was going to be hit much worse than RHEL

vor 19 Tagendiscuss

Nothing stops vendors from shipping additional keys, and several do - I've seen multiple laptops that included an Ubuntu signing cert (contrary to Canonical's recommendation)

vor 19 Tagendiscuss

The short answer is simply that nobody credible has offered to run such a service. The Linux Foundation investigated it and concluded that it was impractical to do so. Since secure boot rolled out we've seen a couple of pieces of malware that have explicitly attempted to bypass it (largely through vulnerabilities in Microsoft's bootloaders, ironically) which strongly implies that it's an obstacle to their goals rather than mere security theater.

vor 29 Tagendiscuss

The majority of code in ffmpeg today isn't written by Fabrice, but also there's multiple axes that people view programming ability on. Some people can write software that will do things you couldn't imagine given the constraints. Some people can write software that is resilient against all malformed input. Sometimes these people are the same people, but frequently they're not.

vor 30 Tagendiscuss

> ARM SystemReady devices (which is a requirement for Thunderbolt 4+ on ARM, so Macs are included)

Either this is untrue or misinterpreted - the SystemReady DeviceTree band (the only one Macs could possibly fit into, given they don't implement ACPI) still requires that devices implement EBBR, which requires that devices implement UEFI. Macs don't, and so are very much not SystemReady compliant.