Linus Torvalds releases Linux 5.19 – using Asahi on an Arm-powered Mac Linus Torvalds has released version 5.19 of the project, and hailed Apple's homebrew silicon – and the Asahi Linux distribution that runs on it – for making Arm-powered computers useful for developers. In his announcement of the release, Torvalds called out work to support for the made-in-China Loongarch RISC architecture as …
I understood it as saying that now with Apple silicon you can have an actual Aarm64 arch to develop on. (I mean nobody would develop on a phone with an Aarm64 SoC would they?)
This isn't a bad thing as the alternative is something that has it's origins in Intel, a company that I think can go and fuck itself many times over. So can Nvidia. And you too Broadcom.
> I used to like linux but he has just given apple a free plug. So fuck linus
What was that story about a well meaning disruptor, seen as a messiah, finding himself despairing at the zealots that his popular movement enabled? ... Dune maybe, or Life of Brian... possibly Apocalypse Now. Or Lawrence of Arabia... Ah okay, it's an archetype. A cautionary tale, part of our societal immune system.
Tweet by Hector Martin from Asahi:
"Looks like Apple changed the requirements for Mach-O kernel files in 12.1, breaking our existing installation process... and they *also* added a raw image mode that will never break again and doesn't require Mach-Os.
And people said they wouldn't help. This is intended for us."
Hi, Simon Sharwood, APAC Editor,
Some of a different mindset and understanding would posit the conclusion reached in the last paragraph is diametrically opposed and at odds with the two preceding paragraphs and Torvald's post footnote ...... which much more likely suggests and indicates that subsequent releases are going to make a real difference.
Bravo, Linus ...... bogged down systems are long overdue a program refresh/projects overhaul/radical rehash/kick up the arse to reroot and reboot them onto a much more enlightened path than the pedestrian trudge through the swamps of the present that the past has provided for current memories to process and try to make greater sense of with clearer mass media presentations of advanced intelligent visions.
And who on Earth would deny that such is definitely needed for an absolutely fabulous change/fantastic progress/quantum communications leap breakthrough?
And to any and all who see everything strange and new as a Persistent ACTive Cyber Threat rather than being able to enable deployment of enjoyment and EMPowerment of ITs Treats in and/or from Vast Almighty Fields of AI Endeavour and Remote Autonomous Provision, there's not much to enjoy in the current default alternative always being offered/forwarded to media for mass public maniacal manipulation of future non-captive perception/enlightening understanding with the following popular ruse which introduces the proxy variable very much to the fore of established thinking and vested interest think tank bodies/exhausted vessels of once-upon-a-past-time novel influence.
There's nothing to be gained in continuing to flog that quite obviously stone dead horse, Trojan and otherwise.
This post has been deleted by its author
This post has been deleted by its author
All releases from Mac OS X 10.5 Leopard[10] and after are UNIX 03 certified,[11] with an exception for OS X 10.7 Lion.
Nobody has had any reason to Unix certify Linux and thus Linux is a Unix like OS.
And who the hell cares.
Apparently Linux is the more powerful OS as Linux is the leading operating system on servers (over 96.4% of the top 1 million web servers' operating systems are Linux),[28] leads other big iron systems such as mainframe computers, and is the only OS used on TOP500 supercomputers (since November 2017, having gradually eliminated all competitors).
https://en.wikipedia.org/wiki/Linux
https://en.wikipedia.org/wiki/MacOS
https://en.wikipedia.org/wiki/Unix
I follow IBM's advice from the 90's.
You need 2+ vendors in order to support a platform. So Apple and their non-standard aarch64 extensions is pretty much out. Apple might be a fun distraction for developers but is honestly a dead end in the long run. The Apple macppc arch was also a fun distraction whilst it lasted.
Erm... Apple is hardly non-standard. It was x86 with an AMD GPU before. ARM64 now. Most of the other chips in an iPhone/MacBook are pretty normal as well.
What isn't normal? Knowing that the build quality is going to be very high. (In before people complaining about the butterfly keyboard, despite that I have never met anyone who actually experienced the problem in real life)
> Apple is hardly non-standard
Apple's recent chip is a non-standard aarch64.
Apple's x86 days used a non-standard UEFI
Apple's PPC days used a non-standard POWER based ISA
Apple's future RISC-V days *will* be non-standard.
Build quality is... glue... But what good is build quality anyway when it is laying in the landfill because it is non-standard and the sole hardware vendor has deprecated it?
Lmao, RISC-V. They've been trying to get it going since 2010 with all the might of universities and some of the biggest silicon companies (and Intel) and even now it's barely competitive with a PowerPC 970. RISC-V is more of a warm and fuzzy thought experiment (whose governing body recommended people check out a literal scam) than an ISA.
ARM itself already dropped most of the standards, which is why the teams getting Linux to run on it have to hack around with their firmware. To get Linux booting on an Intel-based Mac, you put in your disk and select it in the boot menu. One standard down. In addition, it's not just ARM64. It's ARM64 with Apple's extensions added, and they haven't documented everything they did. You can just ignore their extensions and compile for straight ARM64, at least for now, but it probably won't be as efficient. They don't follow standards when it suits them because they don't care about the limitations of not doing so. To be honest, I think a lot of their users don't care either, so maybe it's logical if annoying.
As for "very high" build quality, that's being charitable. They have acceptable build quality most of the time. Some models from the competition are clearly worse, but Apple's not perfect and the other manufacturers do it too. Apple has design defects, they have unrepairability issues, they have had model-specific problems and persistent ones. If you compare it to the cheapest Chromebook made by another manufacturer, Apple's looks better. If you compare it to a laptop of similar cost from another manufacturer, it's a toss-up who will end up with the better quality, design, or cost-efficiency.
Apple build quality is crap. I'm forced to use a MacBook through work, not choice. First one, a 2015 example, had huge problems with overheating. The engineer who came to fix it the first time managed to leave one of the batteries floating loose in the case. After a few more overheating issues that I learn't to fix myself by disassembly and clearing out, or buying a new, fan it gave up entirely. Oddly enough, I installed Linux on it and now it works fine, whatever chip that failed and caused MacOS to barf is unused by Linux. The 2020 MacBook Pro replacement is OK for heat problems, but the keyboard issues have had it back in the repair shop once already. Meanwhile, that Lenovo P50 keeps on chugging. Heavy, inefficient, but Ubuntu 20.04 just works.
It's humbling that an ARM64 RaspberryPi is more than sufficient to cope with my pathetic attempts at coding. But then I was overwhelmed when I got my first Z-80 and discovered Assembler was easier than machine code making me a bit of a wimp at the time.
But good that the big boys & girls may be taking ARM64 more seriously. I look forward to the trickle down effect of not seeing so many IDE related infrastructure restricted to x64 architectures.
Don't know why this is downvoted. My first coding on a microprocessor was 6502 in a PET. No assembler, hand written hex machine code - just like real men (I was 14). I wouldn't offer any flavour of RaspberryPi as a good primer for understanding low level coding - too isolated from the hardware. A GUI, and interpreted scripting languages are all well and good, but I would push a fledgling programmer towards the hardware fun/hell that is Arduino.
Upvoted, but...
being isolated from the hardware is a reasonable place to be. It comes down to perspective / interest.
Some would feel that working at machine instruction / register / IO bus / memory read-write is too isolated...
Some would think that working on microcode / macrocode is too isolated...
Some would think that working at the logic gate level is too isolated.
I heard (~1990s) Gene Amdahl, then in his 70s, talk about the next 30-40 years of hardware development, clearly /very/ knowledgeable about when various stages of 'n nanometer' manufacturing would hit barriers, and why, when quantum effects would limit how small (and hence fast) individual components could get... the role and limitations of increasing parallelisation...
There are nuclear physicists who think that is too isolated from fundamental particles...
And theoretical physicists / applied mathematicians who think that 3 (+ time) dimensions are too isolated from the equations of the universe...
And, quite possibly, Buddhists / astrophysicists who...
- https://www.goodreads.com/book/show/222652.The_Quantum_and_the_Lotus
A RasPi is a good primer / entry point for some people. For others anything lower-level than Lisp is of limited interest. There is no best entry point for all people.
The issue here is that the Apple M1 and M2 hardware, while built around one of the ARM64 ISA versions, are not really ARM processors.
Apple, as an Architecture License holder, takes ARM designs, and builds their own processor around them. This means that they can add whole groups of instructions and other features that they then use in their products. It's going to be an Apple EEE policy.
I suspect that the stellar performance of the M1 and M2 processors is as a result of some of these changes. This means that building Linux for ARM and M* processors, you have a choice to either embrace the changes and get good performance at the expense of effictively having a single supplier lock-in, or keep to the core standard, and potentially lose out and only get mediocre performance.
The ARM environment is fragmenting too much. I think that RiscV is likely to look increasingly more attractive as an alternative architecture for many people.
Linux is built for the Armv8-A/AArch64 ("arm64" in Linux parlance) architecture. Do you have any actual counter examples of "mediocre performance" to prove your "suspicions" of this not getting good performance?
Do you have any actual examples of "The ARM environment is fragmenting too much" and e.g. comparisons with other architectures?
I don't have any examples, the architecture is just too new. But you've spotted the problem, just not thought it through to it's logical conclusion.
M1 and M2 and the variants of Arm64+, which include Apple specific instructions, so what I said was stellar performance has been measured for MacOS, and as Apple own the hardware and OS, it would make no sense for them not to take advantage of all of the hardware tweaks they've added on top of Arm64. This also has the effect of preventing other people building Arm systems that run MacOS without access to M1 or M2 processors.
I would expect that the current crop of Linux builds are for standard Arm64 with the minimum of changes (it's still early days). If there have been no modifications to the compilers to generate the extra instructions that Apple have added, if there were meaningful comparative benchmarks for Ashahi and MacOS, I would expect MacOS to win hands down. The question would be by how much.
Until Apple open up the additions, and the compiler writers and distro owners incorporate the changes into the source tree, I would expect this to remain to be the case. And I believe that Apple are not that likely to put significant effort into a Linux port to their silicon.
I have experience here. I work with IBM Power, and have worked on HPC sites, and while you may still be able to run code generated for 604 and maybe even earlier processors on current Power processors, the same binaries will be much slower than the same source compiled specifically for, say, Power10 with all the bells and whistles turned on on the compiler to take advantage of the ISA changes that have occurred in the last 25 years.
Going back to Apple, until they open up the architectural changes to the community, any Linux generated against the ordinary Arm64 model will miss out on speed and power tricks, and will thus perform worse.
When it comes to fragmented Arm infrastructure, look at the number of SoCs out there, and work out how many of them have hardware specific to that SoC. I'm pretty certain it's not just me saying this.
Of course, writing to the common subset should work fine, but you will never get the performance that you would if you used all of the features.
Phoronix ran their benchmark suite on macOS and the first very crude version of Asahi. Go to Phoronix and search for Asahi benchmark. It's completely unfair benchmarking a first alpha, since it's not optimized at all for the CPU power states/core layout/SSD/etc. but in fact in many tests they were quite close and in a number of them Asahi had an upper hand.
So I hadn't heard of these additions of which y'all speak before, though I do know Apple have implemented many of the optional ARM extensions with ARM v8.5. In any case Asahi performs very well indeed and there was not a decisive slam-dunk for macOS - just a few outliers, which is to be expected at this point. Meaning the M1 is a very performant general purpose processor for any OS, even Windows ARM runs well in a VM.
Where macOS might have a specific advantage, not shown in the phoronix tests, are things like the machine learning module and media enc/dec units. These would probably need new drivers implemented in Linux.
Until Apple open up the additions, and the compiler writers and distro owners incorporate the changes into the source tree, I would expect this to remain to be the case.
The flaw in your statement here is that the assertion that the OS would not take advantage of the hardware, in reality i would suggest that the hardware has been designed to maximise the performance of the OS which is where the benefit comes from owning both and not something Intel or AMD would do for any OS even Windows.
With Apples M chips there will always be performance gains to be had but they may need some fundamental changes to the kernel to enable maximum use of the the specific functionality which can cause bloat or impact other hardware implementations and may not even be of value or use to most users, a bit like Intel's AVX-512 extensions that are great in very limited scenarios so is it worth making the changes to use them?
There is the possibility that the experience of building for a variety of ARM-like chips will be useful down the road. After all, there is nothing to stop RiscV from becoming fragmented as people customise it to their needs.
It might be that some elements of Apple's CPU designs are later adopted by mainstream ARM, and thus lessons learnt from optimising Linux for M* can be reused in future.
"It might be that some elements of Apple's CPU designs are later adopted by mainstream ARM"
They already have been, they were part of the ISA for a while but ARM had not provided a reference implementation so nobody had included them yet. Apple were just the first the first to implement these new features, and there is now a reference implementation for them in the new Cortex designs so we should see them come to other vendors CPUs soon.
Vendors who make their own custom cores like Apple and NVIDIA will typically ship new features faster then vendors who re-badge the stock Cortex designs. If you are going to the expense of designing your own custom core then you are doing it because you want higher performance then the reference designs offer, so of course you are going to design your core to the latest ISA spec with all the optional bells and whistles.
Also RISC-V has the potential to become a complete cluster with 3rd party extensions, and with the decision to allow variable length instructions.
Apple, as an Architecture License holder, takes ARM designs, and builds their own processor around them.
But in order to call them aarch64 (which they do) they also have to comply with the ARM validation tests.
I suspect that the stellar performance of the M1 and M2 processors is as a result of some of these changes.
Do you have any evidence of this? What changes specifically are you referring to?
Could it just not be that Apple have designed a good desktop CPU, whereas most other ARMs that people interact with are designed with thought to power consumption first and performance second?
I'm attempting some aarch64 coding on the M1 at the moment, nothing more than baby steps, but I'm also not looking at anything other than the ARM ISA documentation and it's all working just fine so far.
Apple do have some other custom bits on the chip to do various thing, but this isn't any different from any other SoC vendor adding their additions.
Whilst Apple have made absolutly no effort to support Linux, or help out, I'm not convinced there's any 'secret magic' in the CPU core, other than it being a well designed high performance implementation of the ARM 64 bit architecture.
And what of the notion that Apple are decided to create and administer to their own Multiverse for others to carbon copy and enjoy via processes which require leading instruction sets from programs and projects foundationally built around their virtual machinery and widespread broad band of fanatical acolytes?
Or do you expect Apple to not be so interested in such a cosmic capture as would render to them, as a postmodern Caesar, a more universal command and remote absolute control than has ever before been delivered?
Bet against that whenever such competition or opposition is richer than Croesus and you need to prepare to be content to lose everything and your shirt.
I remember many years ago, when apparently Linus had a go at Apple hardware, a very loud and highly optimistic header that went something like this - "Linus has switched from Linux to Apple" or "Linus gave up on Linux for Apple".
Lots of Windows hardware has found a new expanded life running Linux. I haven't used any Apple hardware since Apple II+ but I assume it's not more rubbish than Windows laptops so if it will be possible (and not too complicated) to run Linux on some Apple hardware then why not.
There are some opinions here about fragmentation, ARM and RiscV, but how do you develop anything better if you are not allowed to change anything on something already available.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
