ARM to charge end users now instead of just wafer charge

[snarkysparky]

So I read that ARM will move to only license its cores to companies that have restrictive agreements with ARM.  So that ARM can charge a percentage of the end uses device cost.

Sooo does this mean DigiKey and Mouser will not be able to sell ARM chips cause they can't control end use?

Makes no sense to me.

https://arstechnica.com/gadgets/2023/03/risc-y-business-arm-wants-to-charge-dramatically-more-for-chip-licenses/?comments=1&comments-page=1

[ataradov]

They might only care about high-end MPU market where there are a very limited number of final consumers. I doubt that would apply to MCUs.

And I'm not sure if the interpretation of the news is fully correct and I don't think anyone has seen the new license agreement.

It sounds like a way to go out with the bang. Everyone would just move to RISC-V over time. People already started to move and this just creates a lot of pressure.

[JPortici]

 
It's like they want us to not use their cores.
How long until RISC-V has an alternative to Trust-Zone so we don't have any more reason to use ARM anymore?

[Marco]

Together with the Qualcomm/Nuvia lawsuit ARM is going to kill ARM competitiveness for anyone but Apple. Apple probably has a grand fathered contract running into the next century.

GG.

[ataradov]

Plot twist - Apple paid ARM to kill their product for anyone but Apple.

[AVI-crak]

ARM licenses are very aggressive and varied. There is a variant of a simple user, with the right to manufacture chips. Advanced developer option - with the right to make changes. And the option of a super advanced developer - giving the right to create a new version of the chip, with a new unique command system.
Aggressiveness lies in the fact that ARM requires absolutely all the developments, with the aim of further resale. The author of developments receives a unique license, with a significantly reduced price tag.
All independent development of ARM ended in the era of 16-bit chips, for which she nevertheless managed to sell licenses. And now he owns an impressive fleet of used cars that have long gone out of fashion, but still sell confidently.

So, Apple just bought everything it could. Apple used all the best parts of the projects, added a little of its own, and got a great version of the M1. But the ARM license continues to work!!! ARM wants to get "my charm", very very much. Apple knows how to feed breakfasts, and knows how to bargain, but it cannot share - a taboo at the genetic level.
This is a very slow war. War of two armies of lawyers. Do not count on a quick ending, this is not an action movie for you.

[snarkysparky]

will we still be able to buy stm32

[MarkS]

will we still be able to buy stm32

Why? You can't even find them now. 

[SiliconWizard]

Plot twist - Apple paid ARM to kill their product for anyone but Apple.

Funny thing is that they are slowly killing the markets their were predominant in while betting everything on a market in which they were essentially non-existant (desktop computing and related).
Not only is that a good way of infuriating your existing customer base, but also a risky way of putting all your eggs in a new basket.
But what do we know.

[brucehoult]

How long until RISC-V has an alternative to Trust-Zone so we don't have any more reason to use ARM anymore?

As far as I am aware, Trust Zone doesn't do anything more than RISC-V PMP (Physical Memory Protection), part of the original RISC-V specification ratified in July 2019.

The original FE310-G000 -- the very first commercially available RISC-V MCU in the HiFive1 in December 2016 -- didn't have PMP (or User mode) but the FE310-G002 in the HiFive1 rev B in March 2019 does.

If a RISC-V core implements U mode then it will almost certainly have PMP too.

[PCB.Wiz]

So I read that ARM will move to only license its cores to companies that have restrictive agreements with ARM.  So that ARM can charge a percentage of the end uses device cost.

Sooo does this mean DigiKey and Mouser will not be able to sell ARM chips cause they can't control end use?

Makes no sense to me.

https://arstechnica.com/gadgets/2023/03/risc-y-business-arm-wants-to-charge-dramatically-more-for-chip-licenses/?comments=1&comments-page=1

Interesting move.
later adding that Arm has been "frustrated by customers’ reluctance to accept the new arrangement."
Entirely predictable, and yet and ARM were surprised / frustrated

The admin of this has to be much more complicated, and with many other suppliers offering existing silicon under existing business deals, trying to change horses like this seems risky / doomed.
How do ARM even hope to police this ? 

[ataradov]

At least on MCUs TrustZone has a lot more than just memory protection. It defines secure and non-secure code sections, which are different from privileges (which also present in both secure and non-secure worlds).

And it also defines a ways those sections can interact. For example, when jumping into secure section from a non-secure section, the first instruction must be a special gateway instruction, otherwise an exception will happen. This prevents jumps into arbitrary code, only allowing jumping into true entry points. This also stops a lot (but not all) of glitching attacks, you can't simply glitch your PC into an arbitrary secure code if the processor is not already in the secure mode.

I'm not sure if this is useful for a lot of application or a real show stopper for adoption of any other ISA, most likely not. But it is there, and some people want it.

[brucehoult]

And it also defines a ways those sections can interact. For example, when jumping into secure section from a non-secure section, the first instruction must be a special gateway instruction, otherwise an exception will happen. This prevents jumps into arbitrary code, only allowing jumping into true entry points.

Aren't those pretty much orthogonal features? IIRC FEAT_BTI was introduced in ARMv8.5-A.

Everyone is getting that stuff. My x86 Linux box makes code with ENDBR64 in it. In RISC-V the "Zisslpcfi" extension (landing pads and shadow stack) is in development with ratification planned for Q3 this year.

https://github.com/riscv/riscv-cfi/blob/main/cfi_forward.adoc

Upstreaming into the Linux kernel has started:

https://lore.kernel.org/lkml/20230213045351.3945824-1-debug@rivosinc.com/

[ataradov]

Aren't those pretty much orthogonal features?

Which ones? Memory protection and secure/non-secure execution modes? They are orthogonal. TrustZone is execution modes. MPU provides memory protection. Both secure and non-secure modes can have different memory mappings.

FEAT_BTI is a general feature for any code and identifies all branch targets. SG (secure gateway) instruction is specific for switching between the modes. No other branches within the secure or non-secure mode need to be identified.

ARM is also confusing because TrustZone for Cortex-A and for Cortex-M mostly have nothing in common, except for the name. They do different things and have different functions.

[JPortici]

How long until RISC-V has an alternative to Trust-Zone so we don't have any more reason to use ARM anymore?

As far as I am aware, Trust Zone doesn't do anything more than RISC-V PMP (Physical Memory Protection), part of the original RISC-V specification ratified in July 2019.

The original FE310-G000 -- the very first commercially available RISC-V MCU in the HiFive1 in December 2016 -- didn't have PMP (or User mode) but the FE310-G002 in the HiFive1 rev B in March 2019 does.

If a RISC-V core implements U mode then it will almost certainly have PMP too.

TrustZone also ticks a very big box in security-related "IOT" projects, that is a feature that carries some weight

EDIT: I was referring to microcontroller cores. There must be a reason why basically every new ARM core has been a cortex Mx3

[ataradov]

There must be a reason why basically every new ARM core has been a cortex Mx3

I'd say the reason is better instruction set and just the core being newer. Many of those implementations either don't have TZ enabled or offer alternative device without TZ. So, there is enough of a market pull to bother making two different versions.

[NorthGuy]

All major OSes (Windows, Linux, Mac) can run on ARM. Microsoft even sell a special PC specifically designed to help people in porting their creatures to ARM. They call it "Development Kit" (or similar).

RISC-V is seriously behind. Do you think RISC-V can beat ARM in this? How long until we have Windows for RISC-V, or Ubuntu mainstream distribution for RISC-V?

[zilp]

There is already an inofficial Debian port for riscv64 that boots and runs a lot of software, so I'd guess it's probably mostly a matter of hardware availability whether a more end-user oriented distribution picks that up and how fast remaining problems are ironed out.

[brucehoult]

RISC-V is seriously behind.

Of course it is. RISC-V is very very new. The basic RISC-V instruction set RV{32,64}GC was not definitively frozen until July 2019, less than four years ago. Many of the features necessary to move from microcontrollers up to the desktop and servers -- vectors, bit-oriented things such as clz and popcount and byte-reversal, hypervisor, cache preload and flush -- were added only in November 2021, 16 months ago.

It takes several years to design a modern CPU core, and then once the core is designed it takes four or five more years to design SoCs around it, make test versions, get into mass production, and get on to consumer-priced boards.

That is the same for both ARM and RISC-V as well as others.

Do you think RISC-V can beat ARM in this?

Given the current rate of progress: absolutely. Multiple credible RISC-V companies have cores on the drawing board or even already announced (e.g. Ventana, MIPS) similar in capability to Apple M1 from just two years ago, and ahead of anything ARM has announced.

Party between ARM-licensed cores and RISC-V in actual shipping hardware you can buy will happen in around 2026. Apple will probably still be ahead, as they are far ahead of ARM now.

Right now there exists a RISC-V SoC and EVB from SOPHGO, the SG2042, with sixty four OoO cores similar to ARM A72 running at 2.0 GHz. I've been promised ssh access to one shortly ... I hope this week.

How long until we have Windows for RISC-V, or Ubuntu mainstream distribution for RISC-V?

The former you need to ask Microsoft and no one else.

Ubuntu had official images for the HiFive Unmatched and HiFive Unleashed in June 2021. They currently have official 22.04LTS and 22.10 images for the StarFive VisionFive 2, the Sipeed LicheeRV (with dock), and the PolarFire SoC FPGA.

Most new RISC-V boards have the ability to put the boot and Machine mode code that is dependent on the CPU core, the SoC, or the board in SPI flash and then all run an identical OS image with Linux kernel and root filesystem. The work to make that smooth should hopefully be completed before 24.04LTS.

[SiliconWizard]

Point is, ARM is certainly much more advanced for "desktop" computing than RISC-V currently is.
Apple has initiated a big market for that (which ARM didn't have before) and they may want to expand on this - things are pretty much almost ready. Even for Windows.
Their strategy may well be to progressively leave the MCU market to RISC-V MCUs (which is already threatening in China anyway) while focusing on higher-performance CPUs (mobile phones/tablets and laptops). That may not be a stupid move if that's the idea.

[brucehoult]

There is already an inofficial Debian port for riscv64 that boots and runs a lot of software

There has been for many years. And Fedora.

Debian doesn't provide ready-made images. You need to use debootstrap.

[brucehoult]

Point is, ARM is certainly much more advanced for "desktop" computing than RISC-V currently is.

CURRRENTLY, yes, of course.

But not for much longer. Less than the time it takes to design a chip and get it into production.

In fact the Intel "Horse Creek" SoC due out on the HiFive Pro in late summer this year should be in the same ballpark with the current best ARM SBCs with the RK3588.

Apple has initiated a big market for that (which ARM didn't have before)

... which ARM doesn't have NOW.

Apple's computers run the ARM ISA very quickly, but Apple isn't selling the chips to anyone else and you're not going to see a lot of Linux or Windows or Android running on that Apple hardware.

Yes, I know about Asahi Linux. I've been a patron of that project for more than two years. I repeat: you're not going to see a lot of machines running it.

And Apple says Microsoft is welcome to port Windows to their hardware. I can't see that happening.

[westfw]

Apple paid ARM to kill their product for anyone but Apple.

Huh.  Apple effectively killed a high-end MIPS chip we were interested in, once upon a time.


I don't understand how (or IF) the supposed new licensing scheme would/could apply to chip manufacturers.  For them, the chip IS the end product.  The various manufactures licensing ARM architectures to implement proprietary chips for cellphones and such could be subject to such a license.


If it made sense at all - ARM must be very confident that the IP that they are licensing is far beyond what any of the RISC-V vendors are offering (and they might be right.  For a while.)  Remember that the only "open" part of RISC-V is the Instruction set architecture.

[brucehoult]

Remember that the only "open" part of RISC-V is the Instruction set architecture.

Which is the most useful thing, because it means that if some leading vendor of RISC-V IP starts to fall behind in performance or raises prices to stupid levels then any other suitably skilled company can swoop in and replace them, allowing everyone to keep using their existing software and tooling.

Unlike ARM's customers, who if they are unhappy have to change ISAs.

PLUS, while no one is required to, in practice there are a number of good quality RISC-V cores that have been open-sourced, license free, ranging from the original Berkeley Rocket (5 stage in order), to Western Digital SWeRV (dual issue), to THead C910 (A72-like OoO, plus other lesser cores), to Paul Campbell's WIP (but working) VROOM (8 wide OoO, aiming for Apple M1 territory).

That THead C910 is in two SoCs that are on prototype boards right now and hitting the market soon: the TH1520 with quad cores at up to 2.5 GHz, and the SG2042 with sixty four cores at 2.0 GHz. The TH1520 will be on a $99 board in April or May, and the same manufacturer has been teasing a proposal for a SG2042 board for $999.

[AndyC_772]

How do ARM even hope to police this ?

I suspect it'll be quite simple.

Small devices - Cortex-M microcontrollers, for example, will be unaffected. These devices are all about ubiquity; they're used by so many different end customers, in such enormous numbers of discrete designs, that the logistics of trying to register each and every one and track the product cost and license fee due in each case, would be a bureaucratic nightmare. They're also able to be designed out of embedded applications without anyone outside the engineering team really noticing.

Bigger devices - the the Cortex-A family, with multiple cores running at GHz speeds - would be the ones to suffer. Designing them out is hard when you're making a phone that's expected to run the enormous existing library of apps. ARM just restricts supply of those chips to the phone manufacturers and charges them a fee per SKU.