Intel processor bug leads to Windows and Linux kernel updates and possible performance hits

[Bit_Guardian]

9 minutes ago, Hairy Weasel said:

I agree, but this one stands out because the solution appears to be very costly in terms of performance. The 30% mentioned is probably a worst case scenario, but even 10% is a lot.

See my post just above this reply. This should be microcode fixable for sure in the medium to long term.

[GoodBytes]

This news entry does not meet the posting guidelines, and the forum quotes aren't being used. Therefore, until it is fixed by @Hairy Weasel, the thread has bee moved out of the Tech News Section and moved to the CPU's, Motherboard and Memory forum section.

 

 

[TheGlenlivet]

2 minutes ago, Bit_Guardian said:

I think Lendacky doth protest too much, as everything Excavator and below is still vulnerable to the APIC shift bug that used to affect Westmere on down (Intel patched, and AMD has yet to).

Yeah, but a 6700K is (maybe was) a viable option for a system that could play top tier games etc.  Excavator hasn't been recommended for a new build for a while.  I would say this is still a way bigger hit to intel, providing everything we are seeing is true.

[Bit_Guardian]

22 minutes ago, TheGlenlivet said:

Yeah, but a 6700K is (maybe was) a viable option for a system that could play top tier games etc.  Excavator hasn't been recommended for a new build for a while.  I would say this is still a way bigger hit to intel, providing everything we are seeing is true.

Thing is, he worded that response very deceptively. There are plenty of privileged memory reference bugs in AMD's architectures that are unpatched. He just specifically skirted looking bad by saying "we're already safe against what could be patched by (insert the acronyms here because I suck at them)".

[leadeater]

1 hour ago, Bit_Guardian said:

Everything Gen 6 on down actually.

Sounds more like all of them to me, newer ones are just less impacted but still affected.

 

Quote

It is understood the bug is present in modern Intel processors produced in the past decade. 

 

Quote

More recent Intel chips have features – specifically, PCID – to reduce the performance hit.

I can't seem to find any decent info on what generation PCID was added to Intel CPU architecture, hope someone knows because I really want to find out.

 

Quote

KPTI fixes these leaks by separating user space and kernel space page tables entirely. On recent x86 processors, a TLB flush can be avoided using the process context identifiers (PCID) feature, but even then it comes at a significant performance cost particularly in syscall-heavy and interrupt-heavy workloads. The overhead was measured to be 0.28% according to KAISER's original authors,[2] but roughly 5% for most workloads by a Linux developer.[1]

https://en.wikipedia.org/wiki/Kernel_page-table_isolation

 

Up to 30% seems far too extreme, you'd have to be in some very special select group to get impacted by that much. What I want to know is how badly this will effect virtual hosts and VM performance, because if it's 5%-10% then holy crap not good.

 

45 minutes ago, Bit_Guardian said:

See my post just above this reply. This should be microcode fixable for sure in the medium to long term.

Article says it cannot be fixed by microcode update, that would have been the first option explored too by the sounds of how annoying this was and the performance impact of the proposed fix.

[Noirgheos]

The real question is, how will this affect gaming and other tasks like rendering? 

[leadeater]

Quote

With the advent of virtualization for server consolidation, a lot of effort has gone into making the x86 architecture easier to virtualize and to ensure better performance of virtual machines on x86 hardware.[19][20]

 

Normally, entries in the x86 TLBs are not associated with a particular address space; they implicitly refer to the current address space. Hence, every time there is a change in address space, such as a context switch, the entire TLB has to be flushed. Maintaining a tag that associates each TLB entry with an address space in software and comparing this tag during TLB lookup and TLB flush is very expensive, especially since the x86 TLB is designed to operate with very low latency and completely in hardware. In 2008, both Intel (Nehalem)[21] and AMD (SVM)[22] have introduced tags as part of the TLB entry and dedicated hardware that checks the tag during lookup. Even though these are not fully exploited, it is envisioned that in the future, these tags will identify the address space to which every TLB entry belongs. Thus a context switch will not result in the flushing of the TLB – but just changing the tag of the current address space to the tag of the address space of the new task.

https://en.wikipedia.org/wiki/Translation_lookaside_buffer

 

Sounds like this issue stems from all the virtualization extensions added to CPUs to increase VM performance.

[leadeater]

Sounds like RIP any 32bit processes.

Quote

So what is PCID?  see text below from Intel SDM.
Process-context identifiers (PCIDs) are a facility by which a logical processor may cache information for multiple linear-address spaces. The processor may retain cached information when software switches to a different linear address space with a different PCID.
When a logical processor creates entries in the TLBs and paging-structure caches, it associates those entries with the current PCID. When using entries in the TLBs and paging-structure caches to translate a linear address, a logical processor uses only those entries associated with the current PCID.

However, the PCID feature is only available on x64 mode (Intel IA-32e mode), which means only 64bit OS can use it. The PCID is a 12-bit value stored in CR3 register for each address space, see below from SDM manual. 

http://hypervsir.blogspot.co.nz/2014/11/improve-performance-for-separating.html

 

[Bananasplit_00]

1 hour ago, Bit_Guardian said:

Everything Gen 6 on down actually.

So Kaby and Coffee are safe? Didn't say so in the article? 

 

1 hour ago, lots of unexplainable lag said:

Add a 60%+ performance hit for Skylake to that.

 

Would also like to point out AMD CPUs aren't hit with this bug at all, as stated here by AMD engineer Tom Lendacky: https://lkml.org/lkml/2017/12/27/2

 

Holy hell that's BRUTAL... RIP Skylake

[Bit_Guardian]

59 minutes ago, leadeater said:

Sounds more like all of them to me, newer ones are just less impacted but still affected.

 

 

I can't seem to find any decent info on what generation PCID was added to Intel CPU architecture, hope someone knows because I really want to find out.

 

https://en.wikipedia.org/wiki/Kernel_page-table_isolation

 

Up to 30% seems far too extreme, you'd have to be in some very special select group to get impacted by that much. What I want to know is how badly this will effect virtual hosts and VM performance, because if it's 5%-10% then holy crap not good.

 

Article says it cannot be fixed by microcode update, that would have been the first option explored too by the sounds of how annoying this was and the performance impact of the proposed fix.

I have yet to see evidence to back up that it can't be, and I have previous similar cases where it was fixable via microcode. And options can be explored in parallel. It might be much faster to deliver this interim solution, and doing both would be best practice in such a case.

[leadeater]

1 minute ago, Bit_Guardian said:

I have yet to see evidence to back up that it can't be, and I have previous similar cases where it was fixable via microcode. And options can be explored in parallel. It might be much faster to deliver this interim solution, and doing both would be best practice in such a case.

Comments like that tend to come from people in the know about the issue, past bugs have said in the articles that microcode fixes are possible and they have come. This one has said it can't, I tend to believe that. More than one report is also saying it can't, likely referencing the same source.

 

For the most part people aren't going to notice any performance decrease, the people actually working on the issue and testing the kernel patches have said single digit regression with 5% being a rough guide. It can be lower than that and I suspect gaming is one that will be less effected.

 

It's also likely benchmark testing applications are not going to show the performance decrease either given how they work and their purpose.

[Noirgheos]

3 minutes ago, leadeater said:

Comments like that tend to come from people in the know about the issue, past bugs have said in the articles that microcode fixes are possible and they have come. This one has said it can't, I tend to believe that. More than one report is also saying it can't, likely referencing the same source.

 

For the most part people aren't going to notice any performance decrease, the people actually working on the issue and testing the kernel patches have said single digit regression with 5% being a rough guide. It can be lower than that and I suspect gaming is one that will be less effected.

 

It's also likely benchmark testing applications are not going to show the performance decrease either given how they work and their purpose.

I wonder if CPUs will be re-benchmarked by all the popular sites after the fix rolls out...

 

To that end, why not have something like a whitelist for applications that need the patch?

[BenasBr]

Does this affect kaby lake and coffee lake?

[Emberstone]

If this affects performance, there better be a way to opt out of it. I couldn't give two shits about a security bug that likely won't affect the vast majority of users.

[leadeater]

27 minutes ago, Noirgheos said:

To that end, why not have something like a whitelist for applications that need the patch?

It's an OS kernel patch which by it's nature will effect every application.

[Noirgheos]

Just now, leadeater said:

It's an OS kernel patch which by it's nature will effect every application.

Well, they better at least allow us to opt out. Losing any kind of performance after basing the purchase on reviews that are not proper anymore seems wrong. May also mislead future buyers.

[Bit_Guardian]

1 hour ago, Noirgheos said:

Well, they better at least allow us to opt out. Losing any kind of performance after basing the purchase on reviews that are not proper anymore seems wrong. May also mislead future buyers.

Won't happen unless you opt out of OS updates entirely.

[Noirgheos]

Just now, Bit_Guardian said:

Won't happen unless you opt out of OS updates entirely.

I do have multiple older images of Windows 10 on USBs. Guess I'll use them from now on if the hit is too great, and until I upgrade to a CPU that isn't affected.

[itrr14]

1 minute ago, Noirgheos said:

I do have multiple older images of Windows 10 on USBs. Guess I'll use them from now on if the hit is too great, and until I upgrade to a CPU that isn't affected.

Just remember that it would mainly affect tasks that are affected by this bug in general. So unless you do a ton of VM work, you might not even notice the 5% they are talking about.

[Noirgheos]

1 minute ago, itrr14 said:

Just remember that it would mainly affect tasks that are affected by this bug in general. So unless you do a ton of VM work, you might not even notice the 5% they are talking about.

https://www.phoronix.com/scan.php?page=article&item=linux-415-x86pti&num=2

 

Seems to be close to 0% for rendering tasks, and that likely extends to gaming. When actual game benchmarks do release, I'll decide then. For now, it's a good thing I have those images.

[itrr14]

3 minutes ago, Noirgheos said:

https://www.phoronix.com/scan.php?page=article&item=linux-415-x86pti&num=2

 

Seems to be close to 0% for rendering tasks, and that likely extends to gaming. When actual game benchmarks do release, I'll decide then. For now, it's a good thing I have those images.

It more seems to be anything that has a large amount of system calls is affected more.

 

And fair enough, smart of you to keep images.

[Noirgheos]

Just now, itrr14 said:

It more seems to be anything that has a large amount of system calls is affected more.

The only program that I use and know for a fact is system call heavy, is Visual Studio. I don't mind waiting an extra second or two to compile code, but it does add up over time...

 

If they fix it for Ice Lake, I may jump ship, since it seems to be compatible with Z370 according to all reports. If not, Ryzen+ it is.

[DarkSmith2]

14 minutes ago, Noirgheos said:

I do have multiple older images of Windows 10 on USBs. Guess I'll use them from now on if the hit is too great, and until I upgrade to a CPU that isn't affected.

win10 after version 1511 is crippling your performance so hard for gaming anyway. Just look how much services they decided to implement that hard that you cant even disable them anymore. This is what iam running tweaked:

You are not going to come even close with version 1709 .. gave me up to 40% more FPS in some titles, pubg went up from 70FPS to 110FPS, after game patches im sitting now on ~140fps with a 2600k and a GTX970, CSGO went up 50FPS aswell.

 

Also noticed horrible stutters before while gaming that are completly gone.

[itrr14]

1 minute ago, Noirgheos said:

The only program that I use and know for a fact is system call heavy, is Visual Studio. I don't mind waiting an extra second or two to compile code, but it does add up over time...

 

If they fix it for Ice Lake, I may jump ship, since it seems to be compatible with Z370 according to all reports. If not, Ryzen+ it is.

Well if it affects Visual Studio, I'm going to be screwed. Just starting to get into even larger projects in school, and it may not take too long right now, but if Skylake is affected up to 60% as that one report is saying, I'm going to need something faster.

[Noirgheos]

3 minutes ago, DarkSmith2 said:

win10 after version 1511 is crippling your performance so hard for gaming anyway. Just look how much services they decided to implement that hard that you cant even disable them anymore. This is what iam running tweaked:

You are not going to come even close with version 1709 .. gave me up to 40% more FPS in some titles, pubg went up from 70FPS to 110FPS, after game patches im sitting now on ~140fps with a 2600k and a GTX970, CSGO went up 50FPS aswell.

 

Also noticed horrible stutters before while gaming that are completly gone.

 

Think I'm good.