Intel Skylake/Kaby Lake processors: Broken Hyper-Threading?

[Phynaz]

By your definition, every chip ever made by any mfg has major bugs, and is fundamentally broken, and all of these major bugs should be written about extensively.
That's going to take a lot of time and a lot of writing.

I strongly suggest he doesn't look at the Technical Service Bulletins for his vehicle. He may never drive again .

 

[richaron]

It has had as much exposure as any other bug. It was exposed they exact same way AMD and Intel have been disclosing bugs for decades. Just because you haven't heard of them in the sky is falling posts doesn't mean they don't occur. Have you read any of the processor revision documentation that either LTC8K6 or I have linked to?

Here's one for AMD:
http://support.amd.com/TechDocs/55370_Rev_Guide_For_Family_15h_Models_70h-7Fh_Processors.pdf

Lol do I sound like I'm panicking? Do I sound like I think the sky is falling down? I'd suggest this is a good opportunity to recalibrate your opinion of people who are panicking or not. I'm not panicking here buddy.

But I do see a major bug in a major CPU. And I count it as a major bug because the system fails running tasks it's supposed to be able to handle; quite simple really. So I think this bug deserves as much attention as similar bugs in the past.

And again I would like to thank the open source ethos for bringing this bug to the light of day. Like most of us I only heard about this bug recently, personally I heard it first from Arstechnica, and like most of the recent announcements this was a result of the Debian Linux announcements.

 

[tamz_msc]

As far as Ryzen segfaulting in gcc issue is concerned, it was originally reported in the Gentoo forums and till now it isn't clear if it is a CPU bug, or if it is, it's unclear as to why it happens. People over there tried different things, some of which worked, while others didn't, so it's better to not jump to conclusions.

 

[JoeRambo]

I'm not doing that. I'm just saying people downplaying such a bug are clueless. No matter what CPU has it, AMD or Intel.

Haha, there was some other dude in this thread who does not bother to even update BIOS to "fix" this. I am with him, running 7700K and not caring at all. Actually i believe eventually microcode will hit Windows update, if that did not happen before.
And being actual programmer with a lot of assembly/disassembly experience, I can evaluate the risks rather well. Also the fact that it was discovered so late and in very specific way makes me rather confident.

 

[mv2devnull]

JoeRambo, what's worse, a crash or production of bad data without a crash?

That is a good question.

This "condition" is triggered only when both logical cores are in use. HPC (high-performance computing) and virtualization both tend to use "more cores", if possible. Whether either has any vulnerable code is an another matter.

The fact is that those use cases tend to run hours, days, weeks. Silent corruption is evil, if you cannot detect it from the results. A lot of future work might be based on those results. A crash is more honest, but you still have to commit resources (time and money) to figure out why there was a crash, to prevent it repeating, and to recompute.

Awareness is important. That leads to informed risk analysis and decisions.

 

[richaron]

I strongly suggest he doesn't look at the Technical Service Bulletins for his vehicle. He may never drive again .

I sounds like people in this thread are suggesting everyone who thinks this is news are simply uninformed? Or that they are somehow overreacting to "business as usual"?

I don't think this Bug would be getting from page news on numerous tech sites if that was the case...

 

[Phynaz]

I sounds like people in this thread are suggesting everyone who thinks this is news are simply uninformed? Or that they are somehow overreacting to "business as usual"?

Being uninformed leads to overreacting to business as usual.

 

[VirtualLarry]

Let me guess, Phynaz, you thought that the "FDIV bug" was also unimportant... I mean, who else but a few niche "spreadsheet jockeys" would ever be affected...?

My position on this is, it's a serious bug, and needs to be fixed, and looks like it already has been (microcode update?).

My opinion is that functionality bugs are more important to fix than performance-related "bugs".

So, yes, I consider this a more serious bug than the original Phenom TLB bug.

 

[Atari2600]

JoeRambo, what's worse, a crash or production of bad data without a crash?

Bad data.

A machine crash (due to this) isn't going to kill anyone. Bad data might.

 

[JoeRambo]

From newest architecture reference manual

Notice that, unlike Skylake microarchitecture for client, the Skylake Server microarchitecture has its
front end loop stream detector (LSD) disabled.

One has to wonder if this is connected to this bug, or Intel disabled LSD and gave up power efficiency (and performance both directly and indirectly) for some other reasons.

EDIT: also wording is important here, "disable" is not same as "not present"

 

[gx_saurav]

Gigabyte released BIOS update F24b to fix this bug two days ago for my B150M-D3H motherboard.

I suggest you guys check updates for your motherboard's BIOS too.

 

[Nothingness]

History of the bug as found by people using OCaml: http://gallium.inria.fr/blog/intel-skylake-bug

The code sequence that lead to the issue doesn't look that crazy and is indeed generated by gcc:

movq    (%rdi), %rax
[...]
andb    $252, %ah
movq    %rax, (%rdi)

gcc has reduced code size by changing andq $0xFFFFFFFFFFFFFCFF, %rax to andb $0xFC, %ah.