Initial Apple M1 SoC Support Aims For Linux 5.13 Kernel

[sverris]

Why should someone do this work?
Is it so much fun?

It's good to have when this hardware becomes older. Because Apple's support will run out - and then it is great to have a decent, safe, fast and long-term supported OS...

[morydris]

Actually. The ARM ISA is about as old as x86.
The difference is a couple of years.

You are talking about the original Arm ISA that evolved over many generation. M1 doesn't support that.

The ISA of the M1 is AArch64 which was a ground up redesign and different in many important ways.
It was "Announced in October 2011" so dramatically younger than x86. You might argue that we should
compare to AMD64 which was "released in 2000", but unlike AArch64, AMD64 is a clear extension of the 32-bit
predecessor and shares nearly all of its deficiencies.

AArch64 is a very well designed ISA and obviously enables high-performance implementations. It's not
the only well designed ISA though.

[morydris]

who cares?

I do, and so much that I'm significantly sponsoring the work.

Why do I care? Because the M1 is simply the fastest processor in existence on my workload (excluding heroic LN2 overclocking). This comes about by being exceptionally strong on single threaded integer workloads, beating _EVERYONE_ on a per-watt basis. I couldn't case less about graphics though. I just need PCIe to work, to enable networking and storage. Thunderbolt would be welcome too. Once that's in place, I'll use it as yet another remotely accessible compute server.

[cynical]

My laptop battery died recently, and rather than replace it, (my laptop is six years old now and really showing its age) I decided to shop around. I have to say, the lack of ARM laptops is seriously disappointing given the promising battery life improvements.

I’m also no longer interested in waiting an extra year for my hardware to be fully supported, so I went with the Dell XPS 13 9310 (the developer edition with Ubuntu pre-installed) Hopefully Tigerlake is a decent platform. Broadwell was pretty good to me.

Figured I would support a company that makes an effort to support my favorite OS.

That being said, I can’t agree with this effort. As cool as it is from a tech standpoint to have a fast ARM SoC running Linux, Apple is a terrible company. Shoveling money towards them is rewarding them for their user hostile practices.

I hope RISC-V takes off in the consumer space so we have more choices. I honestly thought ARM would have done better by now after all that talk about Qualcomm’s desktop oriented Snapdragon chips.

[smitty3268]

Why do I care? Because the M1 is simply the fastest processor in existence on my workload (excluding heroic LN2 overclocking). This comes about by being exceptionally strong on single threaded integer workloads

What is this workload, out of curiosity?

[AdrianBc]

Actually. The ARM ISA is about as old as x86.
The difference is a couple of years.

No, the difference is about 15 years.

ARM was introduced in 1985.

The x86 ISA started in 1970 as the Datapoint 2200 of Computer Terminal Corporation.

The 8-bit ISA of Datapoint 2200 was implemented in PMOS as the Intel 8008 in 1972.

Then the 8-bit ISA was extended in 1974 in the Intel 8080, then in 1976 in the Zilog Z-80.

Then the 8-bit ISA was extended to 16-bit in 1978, in the Intel 8086.
The 16-bit ISA was extended again in 1980, with Intel 8087, then in 1982, in Intel 80186 and Intel 80286.

The 16-bit ISA was extended to 32-bit in 1985, in Intel 80386.


So what is true is that the 32-bit variant of the Intel/AMD ISA has the same age as the ARM ISA (both being introduced in 1985).

However, the ARM ISA was a new ISA, unencumbered by the past, even if it has taken various ideas from IBM 801, MOS Technology 6502 and others.

On the other hand, the 80386 ISA was the result of continuous additions made every few years to an ISA introduced in 1970.

Intel 80386, like the latest AMD Zen 3 and Intel Tiger Lake, continued to be able to execute most of the instructions of the Datapoint 2200 from 1970 (albeit with a different binary encoding, because 8086 has changed in 1978 the binary encoding of the instructions).

[Alexmitter]

Actually. The ARM ISA is about as old as x86.
The difference is a couple of years.

And not just that, the x86 ISA is not even handled directly anymore, its commands get split up in smaller tasks and executed by RISC type cores in all x86 CPUs since the Pentium Pro.

Both x86 and the ARM ISA suck in that you can only work with them if a certain company is OK with that. And that company is Intel on one side and soon Nvidia on the other side.

[Alexmitter]

...say Hello! to Pine64's PineBook Pro, and its mainline kernel (and u-boot) support.

Have that thing too, love it. It is not a 15W SoC like the M1, but its properly mainline, made by a company and chip producer that does not oppose free software, and it has a proper keyboard unlike that macbook.

For around 200 bucks, its the perfect start into the RISC notebook world.

[ldesnogu]

Actually. The ARM ISA is about as old as x86.
The difference is a couple of years.

From a silicon point of view first 8086 was released in 1978 while the first ARM was released in 1985. In both cases design started about 2 years before.

If we talk about later variants of the ISA, AArch64 vs x86-64 then we are talking 2011 vs 1999.

EDIT: oops AdrianBc and morydris already pointed this out

[Boland]

It may be optimized for Metal since it is the macOS main graphics API. But it definitely runs OpenGL since macOS still provides OpenGL API on the current M1 machines. It is deprecated though, so no one should expect a stellar implementation

OpenGL runs on top of Metal with the M1 (and all recent iOS devices).