HyperThreading begining of the end!

  • Thread starter Thread starter Rthoreau
  • Start date Start date
R

Rthoreau

As most people know by now Hyperthreading, has a security flaw found by
researchers working with the BSD projects.

http://www.daemonology.net/hyperthreading-considered-harmful/

Another good article can be found here:

http://www.eweek.com/article2/0,1759,1815954,00.asp

After reading the eweek article, it seems that a few people are miffed
at this discovery, one even suggests it might be easier to just take
the machine. Another one suggest that it would best be solved in a
software update, and for now to just turn off HyperThreading.

We know that Intel is moving away from Netburst, and that for the most
part HyperThreading will not be a factor in dual core chips. It only
seems to affect the current lines based on Netburst designs.

I think this will hurry up the demise of HyperThreading, as I really
don't see a big push for software makers like MS to push out an update.
I could see Intel pushing dual core as on option, but wait they have
not released dual core for servers. Since servers will be the most
likely to be affected, I see this as a another AMD win, reason for
switching to dual core AMD chips.

Rthoreau
 
Using a finger dipped in purple ink said:
I think this will hurry up the demise of HyperThreading,

What would POSSIBLY lead you to THAT conclusion??

This apparently only affects multi-user Linux systems.

Do you REALLY think Intel will STOP using Hyper-Threading because of
a Linux problem, and a SMALL one at that??
 
As most people know by now Hyperthreading, has a security flaw found by
researchers working with the BSD projects.

It's not really so much a security flaw in Hyperthreading itself as it
is a security flaw in the operating systems that can be exploited
through Hyperthreading. It's also a flaw that should be WELL down the
list of security concerns for most people, though obviously no
security concerns can be completely ignored.
http://www.daemonology.net/hyperthreading-considered-harmful/

Another good article can be found here:

http://www.eweek.com/article2/0,1759,1815954,00.asp

After reading the eweek article, it seems that a few people are miffed
at this discovery, one even suggests it might be easier to just take
the machine.

And he's quite right! Exploiting this issue to gain any meaningful
information would be excruciatingly difficult. Actually picking up
the machine and marching out the door with it is probably the easiest
way to get this info, though it tends to be a bit... um... obvious. A
much more important concern is for things like buffer overflows and
more standard methods of gaining increased permissions on a system (as
is noted in these articles, this is not a remote exploit but one that
requires a user already have at least limited-permission access to a
system).
Another one suggest that it would best be solved in a
software update, and for now to just turn off HyperThreading.

It can and is being solved with software updates. As a note, it's
theoretically possible that this sort of exploit could be a factor in
some other multithreaded architectures, eg. IBM's Power5. It's not
entirely unlike some other exploits that could theoretically exist for
more standard multiprocessor setups as well as dual-core chip. The
only real difference here is that it
We know that Intel is moving away from Netburst, and that for the most
part HyperThreading will not be a factor in dual core chips. It only
seems to affect the current lines based on Netburst designs.

I think this will hurry up the demise of HyperThreading, as I really
don't see a big push for software makers like MS to push out an update.

I'd strongly disagree here, this is mostly a non-issue. As mentioned
above, actually gaining any useful information from this would be MUCH
more complicated than with any of the multitude of existing security
flaws, particularly considering it requires that the user already have
access to the system.

As for software updates, at least one BSD has already updated their
code and others will work on it. Since this is such a low-priority
threat it might take some time for others to release updates, but they
will happen eventually.

In any case, the real issue here is more just that this is a new style
of exploit we don't see much of, that is why it's getting some press.
There are thousands upon thousands of other known exploits that would
be much easier for a malicious user to implement. But buffer overflow
exploits are boring these days, we've all seen too many of them to
really care much about.
 
It's not really so much a security flaw in Hyperthreading itself as it
is a security flaw in the operating systems that can be exploited
through Hyperthreading. It's also a flaw that should be WELL down the
list of security concerns for most people, though obviously no
security concerns can be completely ignored.

"WELL down the list of security concerns for most people" would include
anyone using _any_ sort of Windows.
 
Never said:
What would POSSIBLY lead you to THAT conclusion??

This apparently only affects multi-user Linux systems.

Do you REALLY think Intel will STOP using Hyper-Threading because of
a Linux problem, and a SMALL one at that??

What lead you to believe this only affects multiuser linux systems? It
affects any OS that can turn the Hyperthreading on. It was just
discovered by somebody who was working on a multiuser BSD system.

Yousuf Khan
 
I agree, Intel is already phasing out HyperThreading, and Netburst even
if they fix the problem, HyperThreading will have a stigma for those
who do not know any better.

What gets me is those who run Xeon servers, who might need to shut off
HyperThreading, until a fix can be made. This could hurt them if they
have any major applications that really use the HyperThreading ability.

As we all know not that many programs are designed with just
HyperThreading in mind. Usually if it works good in HyperThreading it
works good with general SMP, or dual core.

Lets face it a lot of people using Xeons, are running some kind of
Windows software, just look at how long it took SP2 to be adopted by
the majority of users. The system admins might turn off HyperThreading,
but do you really expect them to patch the system, when they did not do
it for SP2?

I think this will hurt sales of any Netburst based system, ie servers,
especially when you consider the alternatives. All this will do is
make the marketing teams work overtime, or push Intel even harder to
release new products based on the newer designs. Any Netburst solution
will suffer from this problem, and if the solution is in software it
could make a performance hit.

So I still stick by my assertion, that this will hurry up the ending of
HyperThreading.

Rthoreau
 
Rthoreau said:
I agree, Intel is already phasing out HyperThreading, and Netburst even
if they fix the problem, HyperThreading will have a stigma for those
who do not know any better.

What gets me is those who run Xeon servers, who might need to shut off
HyperThreading, until a fix can be made. This could hurt them if they
have any major applications that really use the HyperThreading ability.

I don't think there's any applications at all that specifically make use
of HT. There are multithreaded apps in general and that's about it.
Multithreading will continue to work whether HT is turned on or not,
even in single processor non-HT environments; that's what an OS's task
scheduler is all about. No one is going to miss it.

It was only really useful to people who found the added responsiveness
of foreground apps when heavy background apps were running; but that
wasn't a multithreaded app, that was multiple single-threaded apps. In a
server environment, you don't have to worry about foreground vs.
background apps, they are all background apps.
Lets face it a lot of people using Xeons, are running some kind of
Windows software, just look at how long it took SP2 to be adopted by
the majority of users. The system admins might turn off HyperThreading,
but do you really expect them to patch the system, when they did not do
it for SP2?

I don't think anybody is going to be too worried about this security
threat whether they patch it or not. The app has to be placed into the
system by somebody with administrator privileges, at least initially. It
can run non-privileged, but it has to be installed by privilege.
Secondly even if it's running, chances of it finding anything useful are
small. There's going to be a lot of random data it's going to have to
sort through, it's a needle in a haystack.

As for how long it took SP2 to be installed, it's not surprising.
Microsoft has been known to break things with its SP's. People are right
to wait it out.
I think this will hurt sales of any Netburst based system, ie servers,
especially when you consider the alternatives. All this will do is
make the marketing teams work overtime, or push Intel even harder to
release new products based on the newer designs. Any Netburst solution
will suffer from this problem, and if the solution is in software it
could make a performance hit.

So I still stick by my assertion, that this will hurry up the ending of
HyperThreading.

It won't hurt the sales of Netburst systems. What alternatives are
there? Most of the manufacturers aren't allowed to sell AMD based
systems due to their monopolistic contracts with Intel, so just like
with Microsoft software, they'll mumble and grumble, and continue to
purchase them.

Yousuf Khan
 
Using a finger dipped in purple ink said:
I think this will hurt sales of any Netburst based system,

Just WHY would a MINOR software problem (easily fixed)
that exploits Hyper-Threading, have any effect on NetBurst??
 
I agree, Intel is already phasing out HyperThreading, and Netburst even
if they fix the problem, HyperThreading will have a stigma for those
who do not know any better.

Latest I heard was that Intel planned on continuing to use
multithreading with their next generation of processor core. As for
phasing out Netburst, there's nothing too unusual about that, after
all it has already been 4 and a half years since it was introduced.
Most architectures only have a 4-6 year lifespan, and given that the
replacement to Netburst is still expected to be more than a year away,
this architecture will actually be one of the longest-lived of any
that Intel has made.
What gets me is those who run Xeon servers, who might need to shut off
HyperThreading, until a fix can be made. This could hurt them if they
have any major applications that really use the HyperThreading ability.

As we all know not that many programs are designed with just
HyperThreading in mind. Usually if it works good in HyperThreading it
works good with general SMP, or dual core.

True enough, though along the same lines, most applications won't be
hurt too badly if they disable hyperthreading.
Lets face it a lot of people using Xeons, are running some kind of
Windows software, just look at how long it took SP2 to be adopted by
the majority of users. The system admins might turn off HyperThreading,
but do you really expect them to patch the system, when they did not do
it for SP2?

Err, if people are running a server on WinXP then some extremely
obscure local-access-only raised permissions hole involving
hyperthreading is probably not on their radar! At the very least I
would expect people running Windows server to be using either Win2K
Server of Win2K3 Server.

As for patching vs. not patching, that's another issue altogether and
is unlikely to be affected by this issue one way or the other.
I think this will hurt sales of any Netburst based system, ie servers,
especially when you consider the alternatives.

Uhh, in my experience dealing with customers, 95% of those who will
purchase servers will NEVER even notice this one and when they do, it
will be *WELL* down on their list of priorities.. Somewhere between
which slot in the rack the server should reside and what colour the
front bezel should be.

Really, this issue is so obscure that it's unlikely to have any effect
at all beyond a couple of days worth of press. This bug has already
had it's 15 minutes of fame and will fade away rather quickly.
Patches will comes and most people will apply said patches as part of
their regular patching schedule. Believe it or not, life will
continue.
 
Back
Top