It would seem some redesign / patches needed on windows and linux to prevent some mixup of kernel and user space datas:
https://www.theregister.co.uk/2018/01/02/intel_cpu_design_flaw/

Potentially can affect on performance.

They say AMD CPUs are not affected by the bug... but not sure if the fix will cut out also 30% of AMD performance.
(also ARM might be affected by similar flaw ...)

From internet forums...
"First, the good news: gaming performance appear unaffected by this. This does make sense, since in gaming applications the CPU is going to spend most of its time executing instructions and simulating stuff like physics and not so much on switching system calls. Raw math-heavy tasks like encoding video also appear unaffected as well.

The bad news is that benchmarks that simulate file transfer and SQL operations are heavily affected. These operations require a lot of system calls and context switching, and require spending a lot of CPU cycles to refresh the kernel memory. So if you move lots of files frequently, or if you're running a server or a data centre, prepare to have your day ruined."

Last edited by KimmoK on 03-Jan-2018 at 12:30 PM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@KimmoK

Ya pay peanuts, you get monkeys.

Amiga is about quality not quantity, therefore we've avoided issues.
Happy to look on and chuckle tho.
Sent from my android phone ;)

_________________
Never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway.

I serve King Jesus.
What/who do you serve?

@KimmoK

The hardware bug is not present on AMD CPU's. AMD is working on a kernel patch to exclude them from the Intel patch.

@WolfpackN64

"AMD is working on a kernel patch to exclude them from the Intel patch."

That would be nice, so only 80% of the IT world slows down... 8-o


...
It's interesting that ARM might have similar kind of patch coming to differently isolate kernel and user spaces.
(I wonder how Telecom companies handle it, do they delay the move from PPC to ARM? Do they continue without protected memory spaces...)

I imagine the non-memory protected OSs are immune to this bug. (But it's pretty weak comfort as there is total lack of protection.)
But this might affect slightly the selection of host OS + CPU on emulated Amiga systems.

Also, I imagine true microkernel OSs are not (much) affected by this bug (might not affect MorphOS4x64 & QNX etc.).

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@KimmoK

Microkernel OS's will probably be less affected.
Of course, in Monolithic kernel OS', you want memory separation between the kernel and userspace, but not too strict or you'll suffer a performance penalty. The hardware bug made this quite necessary. This will mean Intel's IPC advantage over AMD's Ryzen just shrunk without AMD even having to do anything. this will hurt Intel in the datacenter big time.

@Barana

Quote:

Barana wrote: @KimmoK Ya pay peanuts, you get monkeys. Amiga is about quality not quantity, therefore we've avoided issues. Happy to look on and chuckle tho. Sent from my android phone ;)

With "quality" I guess you mean providing no protection at all?

@WolfpackN64

Quote:

Microkernel OS's will probably be less affected.

So QNX, AROS, and maybe Haiku will be OK, everyone else not? (I understand the major OS's to be monolithic.)

_________________
I've decided to follow an awful lot of people I respect and leave AmigaWorld. If for some reason you want to talk to me, it shouldn't take much effort to find me.

@simplex

Haiku, Windows and macOS are hybrid kernel systems. Technically, all three are affected. But Haiku has the safety through obscurity advantage. They will have more time to work this out, while macOS, Windows and Linux will need to fix this ASAP.

@Barana
Quote:

Amiga is about quality not quantity, therefore we've avoided issues.

The bug makes it possible under certain circumstances for programs running in user mode to access memory in kernel space. The thing is, AmigaOS doesn't distinguish between kernel and userspace at all and doesn't have any concept of protection rings. So essentially all memory is kernel memory if you like. It's like avoiding break ins by leaving the front door unlocked.

Last edited by AdvancedFollower on 03-Jan-2018 at 05:57 PM.

I never liked Intel/x86 anyway.

@megol

No, by quality I meant $ppc as opposed to cheap commodity intel/arm.

_________________
Never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway.

I serve King Jesus.
What/who do you serve?

@Thread

It would seem AMD and Arm are affected as well.

OSNews


So pretty much every modern phone and computer out there.

Last edited by Arnie on 04-Jan-2018 at 07:33 AM.

@Arnie

Meltdown seem to affect mainly intel. (it's SW patch/fix/workaround affect performance)

Spectre can affect also other CPUs, harder to fix, more difficult to exploit, etc...

https://spectreattack.com/#faq-systems-meltdown

https://www.theregister.co.uk/2018/01/04/intels_spin_the_registers_annotations/

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@Arnie

Are powerpc affected? If not then the powerpc choice was not as bad after all

_________________
No PowerPC, No Fun
Make Amiga Great Again

@Arnie

Nevertheless, the fix is disabled on AMD CPU's in Linux. The fix with the exeption was pulled by Linus just yesterday.

@simplex

Quote:

simplex wrote: @WolfpackN64 Quote: Microkernel OS's will probably be less affected. So QNX, AROS, and maybe Haiku will be OK, everyone else not? (I understand the major OS's to be monolithic.)

A proper microkernel makes things harder but not impossible.

Note that there are two problems here:
One that allows unprivileged programs to do real damage - this is an Intel bug.
One that uses shared resources to get privileged information - this is a general problem (not a bug as such).

If one protected process is running on the same processor as another process the attacker can prime the branch predictor and then use whatever mechanism the OS use to request the protected one to do something.
The protected process then will have a branch misprediction causing it to expose some information otherwise not visible to the attacker by reading something into the cache.
The protected process returns/replies via the OS mechanism.
The attacker then times accessing the cache to determine what that information was.

Shared resources have always been a problem and here two processes sharing cache and branch predictors state can expose information via mechanisms not protected against.

@Barana

So you are saying that context switches flush the branch predictor and/or caches on PPC?

I don't think so.

@Fransexy
Quote:

Are powerpc affected?

I'm unable to find a definitive answer to that, most reports say Intel, AMD and Arm there have been some that just say plus some other vendors. I did see somewhere Sun Sparc and an IBM processor are immune as they operate in a different manor.

@Arnie

It seems to be a problem for CPUs with out-of-order instruction execution, or at least a subset of such CPUs. The fact that it affects both x86 and ARM would seem to indicate that this is an unforeseen use of out-of-order execution, and not implementation "bugs" per se.

Update:

Quote:

...Intel and AMD x86_64, IBM Power, IBM zSeries and 64-bit ARM.

https://www.suse.com/support/kb/doc/?id=7022512

Last edited by bison on 04-Jan-2018 at 09:16 PM.

_________________
"Unix is supposed to fix that." -- Jay Miner

@simplex

Quote:

simplex wrote: @WolfpackN64 Quote: Microkernel OS's will probably be less affected. So QNX, AROS, and maybe Haiku will be OK, everyone else not? (I understand the major OS's to be monolithic.)

No, not exactly.

Some OSes with micro kernels have a the kernel and user memory on different pages, this will help.

On AROS (and every other Amigoid) it doesn't matter if your CPU is affected or not, because they don't have any security. A third party program can scan the whole memory, collect password and even modify system settings or execute other code.

Last edited by OneTimer1 on 04-Jan-2018 at 10:25 PM.