ARMv8.5-A Support Being Prepped To Battle Spectre-Style Vulnerabilities

[phoronix]

Phoronix: ARMv8.5-A Support Being Prepped To Battle Spectre-Style Vulnerabilities

Earlier this month Arm began publishing details of the ARMv8.5-A instruction set update, which is expected to be officially documented and released by the end of Q1'2019, while the LLVM compiler stack has already received initial support for the interesting additions...

http://www.phoronix.com/scan.php?page=news_item&px=LLVM-ARMv8.5-A-Enablement

[mhartzel]

I was hoping designers would be getting rid of hardware features that are inherently unsafe. This seems not to be the case here. It may be a long wait until we have hardware not vulnerable to these kinds of exploits anymore.......

[andrebrait]

I was hoping designers would be getting rid of hardware features that are inherently unsafe. This seems not to be the case here. It may be a long wait until we have hardware not vulnerable to these kinds of exploits anymore.......

I don't think they're willing to sacrifice so much performance yet.

[GunpowaderGuy]

mhartzel yeah, do we really need these kinds of extensions ? aren´t they ( intel is doing something similar ) just being developed because some microarchitectures could not be made safe on time the proper way ? ( suposedly zen will be made inmune on its next iteration , risc's 5 boom has always been )

[hajj_3]

I was hoping designers would be getting rid of hardware features that are inherently unsafe. This seems not to be the case here. It may be a long wait until we have hardware not vulnerable to these kinds of exploits anymore.......

RISC-V Isa isn't vulnerable. Lets hope that some companies make some nice risc-v cores and that they become popular.

[GunpowaderGuy]

hajj_3 is not isas that are vulnerable but microarchitectures that implement them

[name99]

RISC-V Isa isn't vulnerable. Lets hope that some companies make some nice risc-v cores and that they become popular.

BS. ISA is not relevant, the degree of speculation is what's relevant. RISC-V isn't vulnerable because even their most aggressive implementations, something like BOOM, just aren't very aggressive.

More interesting is the question of why bother with this rather than hardening the CPU by doing things like attaching ASIDs to branch predictor entries, and quarantining cache lines loaded speculatively until the load completes? My GUESS is that they expect ARM cores to do that (internally, programmer invisible) but you still have the problem of the kernel as a monolithic block, and these instructions are supposed to be used at various sensitive points in the kernel, But I'm just guessing here.

Finally "but it's going to be well into 2019 at the earliest before actually seeing any ARMv8.5-A hardware."
Let's be honest here, that hardware will NOT arrive in 2019. The first implementations will surely be, like with ARMv8.3A, Apple implementations, and, assuming the current Apple pace is maintained (which seems reasonable) they will arrive in September 2020 with the A14.

[carewolf]

I was hoping designers would be getting rid of hardware features that are inherently unsafe. This seems not to be the case here. It may be a long wait until we have hardware not vulnerable to these kinds of exploits anymore.......

The problem is that then we would have to bring back prefetch. Speculative execution with speculative loads is what made prefetch instructions a thing of the past that can now only slow a program down.

[misGnomer]

Phoronix: ARMv8.5-A Support Being Prepped To Battle Spectre-Style Vulnerabilities

it's going to be well into 2019 at the earliest before actually seeing any ARMv8.5-A hardware.

Unless things change we'll have more or less fully open-source platforms running on the fixed 2019 ARM hardware around 2022, but all that support will be based on 2019 kernel versions - or even older ones - until the end. RISC-V even at its infantile stage looks promising unless some openminded ARM vendor takes a different approach to incorporating suffocating proprietary components. Keep those at arms length!

[Weasel]

I was hoping designers would be getting rid of hardware features that are inherently unsafe. This seems not to be the case here. It may be a long wait until we have hardware not vulnerable to these kinds of exploits anymore.......

• Get a hammer.
• Smash your router or modem or wireless antenna/chip or whatever.
• Profit

You just got rid of inherently unsafe hardware. Congratulations!