Skip to contentSkip to conversation

Bugs aren't features: X86_FEATURE_FXSAVE_LEAK

DISCUSSION H. Peter Anvin 4 participants 6 messages May 5, 2006 – May 9, 2006

lore
Share & tools
lore
#0 lore raw

The recent fix for the AMD FXSAVE information leak had a problematic side effect. It introduced an entry in the x86 features vector which is a bug, not a feature.

The problem with this is that the features vector is designed so that it can be ANDed between CPUs to find out the common feature set. However, bugs aren't features, and bugs should be ORd, not ANDed. In that sense, the *absence* of a bug is a feature.

There are two possible ways of dealing with this:

a) put bugs in the features vector, but sense-inversed, i.e. 1 = bug absent; 0 = bug present.

b) add a separate bugs vector.

When I originally wrote the code I always meant to do (b), but never got around to it. It's clear, though, that whatever way we go about this, it should also incorporate the fdiv, hlt, f00f, and coma bugs.

-hpa

#1 lore raw
Quoted 3 lines · H. Peter Anvin

The recent fix for the AMD FXSAVE information leak had a problematic side effect. It introduced an entry in the x86 features vector which is a bug, not a feature.

It's a non issue because it affects all AMD CPUs (except K5/K6). You'll never find a system where only some CPUs have this problem.

-Andi

#2 lore raw
Quoted 9 lines · Andi Kleen
Quoted 3 lines · H. Peter Anvin

The recent fix for the AMD FXSAVE information leak had a problematic side effect. It introduced an entry in the x86 features vector which is a bug, not a feature.

It's a non issue because it affects all AMD CPUs (except K5/K6). You'll never find a system where only some CPUs have this problem.

It's still wrong architecturally, and we should have a sane way to deal with this as well as other bugs. This isn't the only bug -- we're getting a decent-size collection of them already -- and not all of them is going to have that particular property.

-hpa

#3 lore raw
Quoted 7 lines · H. Peter Anvin
Quoted 6 lines · Andi Kleen
Quoted 3 lines · H. Peter Anvin

The recent fix for the AMD FXSAVE information leak had a problematic side effect. It introduced an entry in the x86 features vector which is a bug, not a feature.

Quoted 7 lines
Quoted 3 lines

It's a non issue because it affects all AMD CPUs (except K5/K6). You'll never find a system where only some CPUs have this problem.

It's still wrong architecturally, and we should have a sane way to deal with this as well as other bugs. This isn't the only bug -- we're getting a decent-size collection of them already -- and not all of them is going to have that particular property.

I'm not very worried because mixing steppings is generally unsupported anyways and in Linux also doesn't work well with it.

(e.g. my old rig with one CPU FXSAVE capable, one CPU not never worked without kernel hacks)

It's probably not worth the effort anyways. The people who build such strange systems can tweak their kernels for it too.

-Andi

#4 lore raw
In reply to Andi Kleen
Quoted 8 lines · Andi Kleen
Quoted 3 lines · H. Peter Anvin

The recent fix for the AMD FXSAVE information leak had a problematic side effect. It introduced an entry in the x86 features vector which is a bug, not a feature.

It's a non issue because it affects all AMD CPUs (except K5/K6). You'll never find a system where only some CPUs have this problem.

I have a dual CPU system (Tyan Tomcat 1564D) where only one CPU is reported to have the F00F bug (iirc). So yes, there can be SMP-systems where the CPU's have different bugs.

/Christer

Signature · 4 lines
"Just how much can I get away with and still go to heaven?"

Freelance consultant specializing in device driver programming for Linux 
Christer Weinigel <[email protected]>  http://www.weinigel.se
#5 lore raw
In reply to Andi Kleen
Quoted 9 lines · Andi Kleen
Quoted 3 lines · H. Peter Anvin

The recent fix for the AMD FXSAVE information leak had a problematic side effect. It introduced an entry in the x86 features vector which is a bug, not a feature.

It's a non issue because it affects all AMD CPUs (except K5/K6). You'll never find a system where only some CPUs have this problem.

The initial argument is dead on, bugs should be either presented as a "fixed" feature, or there could (should?) be a bugs vector which could be or'd. It's at least possible that AMD might fix this some time.

Signature · 3 lines
   -bill davidsen ([email protected])
"The secret to procrastination is to put things off until the
 last possible moment - but no longer"  -me