Discussion in 'Article Discussion' started by bit-tech, 25 May 2018.
Internal Server Error
Ooh, that's exciting. @MLyons: SUMMAT'S BROK.
It's working now.
Was related to a restart for GDPR related stuff. All is well again.
So maybe I'm reading this wrong, but if Intel's current Virtual Fences fix doesn't work but isn't implemented in any silicon yet, doesn't that give Intel the opportunity to fix the issue before it's baked into any products? Or is Virtual Fences already being added into next gen as-yet-unreleased-but-too-late-to-change silicon?
Something is still broken. Look at the lower-left corner of this image for the address the @MLyons link in your post points to. It doesn't do it in the forum though, only in the article page.
Very odd. When you hover over it in the forums it works exactly as you would expect. I'm a little stumped on that. Make a thread in feedback to prevent cluttering the article up and I'll look into this more.
The latter: if they're talking about putting it in chips due later this year, it's way too late to be making changes like that. Best case, you're looking at the generation after - or even the one after that. That's even assuming Intel bothers at all - I mean, even its workaround is going to be optional.
Given the relatively short time since the attack angle has been discovered variant 5 - 17 (or whatever they end up being called) are inevitable, so regardless of what happens with variant 4 they will have to keep working on securing speculative execution for years to come.
I'd expect EVERYONE's near-future silicon is vulnerable to SPECTRE class attacks. AMd, IBM, ARM, anyone who utilises Speculative Execution is going to be vulnerable for a few years as the 'easy' applications of SPECTRE to attacks are found, and methods to retain the performance of speculative execution with leakage are found.
This won't be an easy or quick fix for anyone.
Well this is all very heartening
Article updated with confirmation that the in-silicon fix does not extend to Variant 4.
Intel has confirmed that its in-silicon hardware protection does not extend to Variant 4, and that it will be relying on the microcode mitigation - which, it must be remembered, is disabled by default - on both current-gen and next-gen processors. 'As we shared in our announcement on March 15, those design changes provide protection against Variant 2 and 3,' an Intel spokesperson tells us. 'For Variant 4 – in addition to the browser-based mitigations that are already available -- we’ve added functionality into our microcode called the Speculative Store Bypass Disable (SSBD) bit. This functionality will continue to be utilised on future hardware platforms ensuring customers can stay protected.'
Welp, we had a good run.
Time to head into the pile of obsolete technology and get the old Sparc workstations, 486 desktops, and Raspberry Pis out. In-order execution is the new word of the day.
Have AMD announced an ETA for their hardware fixes (though presumably Zen 2 next year would include some mitigation)? We know that SPECTRE Variant 2 has a microcode fix - if any motherboard manufacturers actually roll it out, that is - and are not producing a microcode fix for Variant 4 (and, like Intel, have recommended Memory Disambiguation remain enabled).
1. I'm still annoyed that it will cost the end user current, non-discounted prices to upgrade and remove a fundamental flaw, or have ten percent performance taken away
2. I STILL have no idea whether any fixes have been applied to my system as I can't find a single comprehensive list of required patches, dependent on components used
I think I'd be inclined to just push an overclock further and incur no cash cost, if only I could answer 2 above.
It's not a list but doesn't InSpectre by GRC tell you if you're all patched up.
Separate names with a comma.