Intel MPX Support Will Be Removed From Linux - Memory Protection Extensions Appear Dead

[phoronix]

Phoronix: Intel MPX Support Will Be Removed From Linux - Memory Protection Extensions Appear Dead

Back in April was a discussion about dropping MPX support from the Linux kernel but no action taken. Now though an Intel developer is preparing to see this Memory Protection Extensions functionality removed from the mainline Linux kernel...

http://www.phoronix.com/scan.php?page=news_item&px=Intel-MPX-Kernel-Removal-Patch

[cl333r]

This is effectively useless and unused functionality in CPU? Anything else unused in CPUs?

[wizard69]

This is effectively useless and unused functionality in CPU? Anything else unused in CPUs?

I would imagine that once in 64 bit mode many addressing modes go unused. That is one advantage new architectures have, little if any hardware dedicated to legacy support.

I could see ARM dropping all 32 but support from its processors soon. This would give us an AArch 64 clean processor. I wouldn't be surprised if Apple is first due to how hard they have pushed developers to 64 bit. When it happens ARMs cores will get faster and lower powered. Being 64 bit only means an entire instruction set support gets deleted.

[numacross]

This is effectively useless and unused functionality in CPU? Anything else unused in CPUs?

It's present only in Skylake and newer together with Goldmont on the Atom side. The Wikipedia article summarizes a study of MPX that made it not so desirable together with problematic support from Intel's side in GCC.

There are many unused parts in CPUs, not only in x86. Another example of potentially dubious x86 extension is Intel SGX which PoCs of attacks presented.

[numacross]

I would imagine that once in 64 bit mode many addressing modes go unused. That is one advantage new architectures have, little if any hardware dedicated to legacy support.

I could see ARM dropping all 32 but support from its processors soon. This would give us an AArch 64 clean processor. I wouldn't be surprised if Apple is first due to how hard they have pushed developers to 64 bit. When it happens ARMs cores will get faster and lower powered. Being 64 bit only means an entire instruction set support gets deleted.

I'm not sure if legacy support in terms of x86 on x64-capable CPUs takes any considerable amount of hardware. The decoders are really small in comparison to the execution units, the registers and the giant caches. x86 processors have been microcoded for a long time, which means they are actually not implementing x86 in hardware and both 32- and 64-bit modes can reuse parts of the processor.

On the ARM side it's the same for ARMv8 which supports 3 instruction sets: A32, T32 and A64. For the sense of scale let's take a Samsung Mongoose M1 core from 2017 on 14nm. An annotated schematic of one core shows that the instruction decoding is not very big. When you take into consideration the entire processor it gets even smaller

[jpg44]

The legacy support in x86 is a miniscule amount of hardware, a lot of it is probably simply reused in higher address modes, you start in 16 bit mode, your really using 16 out of the 64 lines, then go to 64 bit, and those 16 bit address lines are still use with additional ones coming online. Some Intel engineer said its a miniscule amount of circuitry to support legacy and really has no significant effect on the CPUs performance.

[carewolf]

This is effectively useless and unused functionality in CPU? Anything else unused in CPUs?

16bit mode comes to mind. I wonder if the processors also still has the 286-era memory protection logic (segments instead of pages). They probably do to support OS/2, and other ancient stuff still used in some niches.

[s_j_newbury]

The legacy support in x86 is a miniscule amount of hardware, a lot of it is probably simply reused in higher address modes, you start in 16 bit mode, your really using 16 out of the 64 lines, then go to 64 bit, and those 16 bit address lines are still use with additional ones coming online. Some Intel engineer said its a miniscule amount of circuitry to support legacy and really has no significant effect on the CPUs performance.

But does have a significant effect on the design!

[jabl]

I would imagine that once in 64 bit mode many addressing modes go unused. That is one advantage new architectures have, little if any hardware dedicated to legacy support.

I could see ARM dropping all 32 but support from its processors soon. This would give us an AArch 64 clean processor. I wouldn't be surprised if Apple is first due to how hard they have pushed developers to 64 bit. When it happens ARMs cores will get faster and lower powered. Being 64 bit only means an entire instruction set support gets deleted.

Falkor is aarch64 only, FWIW.

[Weasel]

I would imagine that once in 64 bit mode many addressing modes go unused. That is one advantage new architectures have, little if any hardware dedicated to legacy support.

I could see ARM dropping all 32 but support from its processors soon. This would give us an AArch 64 clean processor. I wouldn't be surprised if Apple is first due to how hard they have pushed developers to 64 bit. When it happens ARMs cores will get faster and lower powered. Being 64 bit only means an entire instruction set support gets deleted.

Imagine is a good word you used, since it's pure fantasy, to put it nicely. But you'll keep re-iterating the same nonsense over and over no matter how many times you're proven wrong, so it's pointless anyway.

Ignorance is not bliss though.