x86-S: Intel wants to drop legacy compatibility from processors

The long-awaited clean-up of the decades-old x86 architecture finally approaching?

The scenario of Intel dropping backwards compatibility and creating a new version of x86 processors free of various “legacy burdens” has been discussed for a long time. The hype around ARM processors and the perceived or real advantages they gain by not carrying similar “baggage” highlighted this topic again. Intel has unveiled a proposal for a simplified pure 64-bit x86-S architecture that could bring about such a revolution, now.

Intel has now released documents proposing an architecture called x86-S, which is a planned future revision of the x86 instruction set and processor platform, currently still in the draft phase. So it isn’t necessarily anything definitive for now. The S designation should mean “simplified”. Thus, it is not an entirely new instruction set, but it adds some changes to the existing x86-64 that will break decades of compatibility.

The x86-S architecture should be 64-bit only, or at least partially so. Today’s x86 processors are 64-bit, but they retain full compatibility with not only 32-bit but also 16-bit mode. They boot (reset) in 16-bit mode and only then does the firmware switch to 64-bit, during firmware initialization. This would be removed by the x86-S architecture, the processors would always natively work only in 64-bit mode.

It would not, however, mean a complete elimination of 32-bit compatibility, as ARM processor manufacturers have done. Chips based on the x86-S architecture would no longer be able to run a native 32-bit operating system, but support for running 32-bit software in user space mode on top of a 64-bit operating system would remain. The gist is that the Ring 0 privilege level would no longer be available for 32-bit mode.

Simplified CPU booting directly in 64-bit mode (source: Intel)

However, compatibility with legacy 32-bit OSes (such as Windows XP or even older systems) should be possible to attain through virtualization. So a 32-bit OS would not work directly on hardware, but it should be able to run with a suitable hypervisor that handles the incompatibility of x86-S with the original x86 platform.

However, from what Intel says, 64-bit operating systems will have to be modified as well, so new versions of Windows and Linux will probably be needed for the x86-S. Virtualization should help in this case too and allow usage of older 64-bit OSes. Conversely, the ability to run any 16-bit code (not just kernel but user space too) natively on the baremetal hardware will be completely lost. Such applications will therefore have to be emulated on x86-S.

These changes will help to simplify the processor design a bit and save some transistors and perhaps even some power consumption, too. It probably won’t bring a huge amount of savings, but it will simplify the validation (verifying that the CPU is working correctly), which is a time consuming and very hard part of processor development. It will also eliminate some of the bugs that can be introduced into these more obscure parts of CPUs. Intel notes that these legacy modes and functionality tend to be used very little in practice outside of that short usage during boot time.

Along with 16-bit and 32-bit mode, Intel proposes to drop several other features. For example, Ring 1 and Ring 2 privilege levels (which are mostly unused by modern software) and access to I/O ports from user space (Ring 3) are to be removes. The latter would mean that some software would have to be rewritten to use MMIO. There are also proposed changes to interrupt handling. With the changes proposed for he x86-S architecture, it would also be possible to switch between a four-level and five-level page tables without the need to switch CPU to unpaged mode (which is now a required intermediate step).

Proposed changes in the x86-S architecture (source: Intel)

When various analysts or commenters considered the eventual redesign of the x86 processor platform in the past, it was often assumed that Intel could remove support for, for example, MMX and x87 instructions (and their registers), which would also simplify the processor design. But this would stop a large number of compiled applications from working at the user space level, not just at the OS (kernel) level. The x86-S design does not seem to include anything like this so far. It looks like pure userspace software might not be affected much, and thus there might not be that many broken legacy programs.

Historic x86 processors from AMD and Intel (photo: Cnews)

The proposed changes look quite a bit less radical than what ARM has done with the ARMv8 and ARMv9 architectures, which also started out by only allowing 32-bit software to run only in user space (while no longer supporting 32-bit kernel/OS) followed by abandoning any 32-bit backwards compatibility entirely. However, ARM’s 64-bit instruction set is a new (from the grounds up designed) ISA completely separated from the 32-bit one. The x86 architecture (courtesy of AMD) has implemented the 64-bit ISA as an extension of the 32-bit mode on the other hand, so cutting deeper into x86 legacy would create far-reaching incompatibilities with existing 64-bit software.

In any case, this proposal is probably still open for discussion and Intel wants to get some feedback and suggestions from users, among other things. Thus, the final version of the “legacy-free” x86 architecture may end up differing in various ways as well.

Sources: Tom’s Hardware, Intel (1, 2)

English translation and edit by Jozef Dudáš


  •  
  •  
  •  
Flattr this!

Zen 6 finally brings more advanced 2.5D chiplets, has RDNA 5 GPU

So far, little information has emerged about what to expect from AMD Ryzen processors with the Zen 6 architecture. The core itself should be an evolution of Zen 5, given AMD’s model resembling “tick-tock”, where even-numbered cores are less aggressive enhancements of tech introduced in odd-numbered ground-up redesigns. But it looks like everything else in the SoC could be a radical shift from the aging concept of Ryzen processor uncores. Read more “Zen 6 finally brings more advanced 2.5D chiplets, has RDNA 5 GPU” »

  •  
  •  
  •  

AMD confirms Zen 5 details: 6 ALUs, full-performance AVX-512

Zen 5-based AMD processors will launch this year (perhaps in Q3). Unlike Zen 4 which was just Zen 3 refined, Zen 5 will be another big upgrade, and according to various hints including statements by the architect Mike Clark, it could be AMD’s most interesting core since the first Zen. Interestingly, up until now, the only information about it has come from a single YouTuber source. But it has just been officially confirmed directly by AMD. Read more “AMD confirms Zen 5 details: 6 ALUs, full-performance AVX-512” »

  •  
  •  
  •  

Arrow Lake CPU spotted. Missing HT and AVX-512 confirmed

Couple days ago, documents leaked on Arrow Lake-S processors and Generation 800 chipsets – Intel’s next-gen desktop platform with the new LGA 1851 socket, due out in the second half of the year. Besides the things we already analysed, those documents also indicate that Arrow Lake P-cores have only one thread. This has now been confirmed by a log from testing a sample of this CPU. After 22 years, Intel processors are dropping HT. Read more “Arrow Lake CPU spotted. Missing HT and AVX-512 confirmed” »

  •  
  •  
  •  

Comments (2) Add comment

  1. I remember reading about the possibility of removal of some specialized blocks and translating these instructions to AVX to save some space, though it’s possible that either the translator would be big enough to not bring any savings or the savings are not (yet) worth the effort

    what I’d like to know is: will x86-S software run on x88-64 hardware? looks like the OS would likely need some tweaks in the booting process at the very least, but userspace should be fine?

    1. “will x86-S software run on x88-64 hardware? looks like the OS would likely need some tweaks in the booting process at the very least, but userspace should be fine?”

      It does look likely to me and it would make sense. To be confirmed of course, but probably good chances.

Leave a Reply

Your email address will not be published. Required fields are marked *