gym week #3

The Kernel Report - Jonathan Corbet (@corbet), @LWN

The recording of this recent talk is now available on the #ossna2024 schedule page: https://ossna2024.sched.com/event/1aBNs/the-kernel-report-jonathan-corbet-lwnnet

Slides can be found here: https://static.lwn.net/talks/2024/kr-ossna.pdf

Direct link to the recording: https://www.youtube.com/watch?v=DAqjl_x4hZc

#Linux #kernel #LinuxKernel

Browsing without account:

• #Instagram: https://imginn.com/
• #Twitter / #X: https://nitter.poast.org/

@lupyuen You can do memory protection also with RISC-V without MMU by configuring so called Physical Memory Protection (PMP) registers in M-mode and running applications in U-mode. So it is memory protection based on physical memory partitioning instead of virtual memory. RISC-V would be a good platform for product low-cost but still quite secure products.

EDIT: Oops, meant to write U-mode, S-mode is one that exists when there is an MMU :-)

@lupyuen I have one SBC for RISC-V (VisionFive2) but would be nice to get one MMUless SBC too (i.e. without S-mode). I.e. kind of RISC-V "arduino".

@liskin @ljs @vbabka do not want to sound like boomer but people should neither "fan" linux nor rust or anything like that. inspiration should come from hardware, domain and stuff like that. they are translatable from kernel to kernel and from language to language, and that is what matter in the end.

@edwintorok @sdbbp right thanks for to great answers! i mean mostly hlt and similar are used to conserve energy (e.g. in PM an similar), and e.g. locks just reschedule so i could imagine that in e.g. linux nop should not cause any actual collateral damage.

i'm just used to specs where opcode does one and only one thing :-)

I wish there was USB-stick that would send one single key press when the button is pressed. That's all it would do.

@liskin @ljs @vbabka doing symbian was not all that bad.

i was not in the UI layer. i was in the team doing audio subsystem, which contained resource manager, DSP codecs, audio policy etc. e.g. setting up the correct codecs, managing memory for audio, opening correct streams depending on phone state (like incoming call when you have video playback ongoing).

i also learned basics of ARM CPU's at the time, which is obviously still useful knowledge, JTAG debugging with Lauterbach TRACE32 and stuff like that. so even tho symbian is a relic, all the skills are still useful and can be applied :-) that way i got into middleware and operating systems.

when i started in 2003 symbian phones were larger and had two ARM CPU's separated by a physical bus: the first was for the phone OS and second one was for the application OS (symbian). when series 60 3.0 came along around 2004 the hardware architecture moved into one CPU model and to a real-time kernel called EKA2, which hosted the whole symbian in one of its threads.

One thing that was tedious was the edit-compile-run cycle. If I had to fix a bug, it took about 45-60 minutes to run it on a device, as both compilation and flashing took a while.

Doing e.g. Linux kernel is not that hard if you've worked in any possible operating systems before because in the end it is all about understanding hardware and the domain where you are working and the source code's layout is just like a different legislation and policies in a foreign country.

OK, so WFI on RISC-V is not detemistic! It can be same as HLT (x86) or WAIT (ARM) but could also be nop 🤷 Beyond me tbh. #riscv

@ljs if nokia clash had happened maybe few months later i might be living in canada and maybe still working for blackberry, and developing webkit ;-) i'm a kernel dev by accident (sort of).

@ljs it is also nice to get chance to work a bit entities like ethereum foundation (and earlier enarx) because it is at the same time chance to see how all this looks from a perspective of a users and not just e.g. kselftest :-)

@ljs yeah i actually do enjoy now a lot of working in tee and confidential computing area. at some point i thought i was bored but now i'm feeling confident that it is good place to stay. also it helps that distributions and cloud service provides have seemed to gained new interest to these technologies and they are getting enabled (like TPM2 across the board). So yeah probably also plan to stick on making sure that we can identify software that runs in host and also in the cloud (not just server where the software runs) :-) i don't think this is coming less important at least...

@ljs I'm not young, I'm middle-aged and have also done bunch of non-kernel things :-) 43 now and been in industry since 2003 starting from Symbian OS and doing even one full 3D engine and bunch of things. In 2010-11 I did for fun initial bits of WebGL for QtWebKit and then Research in Motion contacted me and wanted me to come to visit in Canada. But then Nokia clash happened and I ended up to Intel and have been working on kernel since (there I implemented TPM2 and SGX support and some other stuff). After Intel I went to a startup called Profian where I implemented SGX backend and some other bits for https://enarx.dev/.

I took the researcher job and rejected factors better paying industry job because my juices were gone after startup madness. Also there was chance to learn RISC-V (see e.g. https://www.sochub.fi/).

So somehow I've converged into the "hardware TEE jack-of-all-trades-but-master-of-none" type of person in Linux :-) Do not mind and I always find some new endeavors like connecting with ethereum foundation (they paid me a trip to Tallinn where I spent weekend and brainstormed ideas for better Linux support).

So yeah in my work life the only stable thing has been chaos and uncertainty, and I'm used to it, so I never really feel that much pressure or stress even if the platform is not that stable :-)

edit: s/ethernet/ethereum/ ;-)

@ljs agreed, this is true and more or less what i've been doing :-) i rejected industry job to be able to study new things like riscv-architecture. and also now i'm doing some collaboration with ethereum foundation as it seems interesting and fun (do not own any of that currency btw but tech is fun in general). so i'm not actually worried that this would not lead to something but still doing a bit of seek to see what there might be for me.

@ljs i'm in half-hobbyist mode. i'm working on riscv soc's and doing maintainer stuff on side but yeah getting full-time to kernel would be great :-)

@ljs Congratulations! still working on mine, some ideas are emerging but nothing locked in yet. My contract with the university ends in the end of September so there is luckily also some time to consider :-) Took a 1.5 years from the industry and now looking for coming back.

Everyone should immediately stop contributing to the stack overflow and its network. The human touch is what made it unique. Delete your profile from SO AND all your answers. Freeloaders are making money out of human contributions.

oct-git and its sibling project https://crates.io/crates/openpgp-card-ssh-agent are concrete options for OpenPGP card users to explore OpenPGP use without GnuPG, today.

In addition, the #rsop "Stateless OpenPGP" tool (https://crates.io/crates/rsop) also supports using OpenPGP card devices (see https://datatracker.ietf.org/doc/draft-dkg-openpgp-stateless-cli/ for more).

@cake you can only understand something if you are involved with it. no need to be judgemental.