"w_tom" <w_tom1@usa.net> writes:

> Todd describes a condition that causes a crash; not hardware
> damage.
> He also describes a manufacturer's design defect - not a conditon made
> possible in the field. Motherboards, et al have been made, upgraded,
> and enhanced from previous experiences so many times. Design is mostly
> provided to motherboard manufacturers from Intel, AMD, and other high
> tech companies.

Among those, one I worked for and hold a circuit design patent with.

> Design is repeated so often as to be well established.

This couldn't be more false in the microprocessor realm where the
state of the art requires reinvention of architecture implementations
to achieve speed gains. Any student of an introductory computer
architecture course could tell you that.

> What sells as cheap would cost massively if done from scratch. No,
> those cheap motherboards have massive design and experience within
> them. For example, today's Intel CPUs cannot overheat destructively
> because of designs well proven in the 1980s.
>
> Even circuits inside ICs are designed so that unintended logic states
> cannot cause damage.

Not universally true. To achieve the speeds of modern
microprocessors, dynamic logic is used in places, and, it is not at
all inconceivable that software on an appropriately privileged
execution layer could write to a microcode that could lead to an
unanticipated logic state that could precipitate a hardware failure.

It's necessarily a goddamn hard thing to try to exploit given the
granularity with which one would need to know the microarchitecture of
the CPU in question, though.

> But then also logic inside a microsequencer is not accessible to a
> virus.

Again, this is typically the case with a given OS/motherboard/CPU
combination, but there's no way to guarantee that this is universally
true.

> designs. Take some CMOS logic gates. Ties all those outputs together.
> But the logic gate outputs at various and contradictory logic states.
> Are any logic gates damage? Of course not.

In a massively integrated design like a CPU, if you get enough of
those transistors drawing crowbar current from the supply rails, you
are going to overheat the device, and cause transitor damage. You
don't think every single logic gate on a cpu with millions of
transistors is designed to have individual short circuit protection do
you?

> implemented in the design makes hardware damage unlikely and totally
> unacceptable.

Totally unacceptable to a consumer if it happened, sure. But do you
really think that with all the design points you need to hit in a
modern CPU design that all possibilities of malicious microcode are
taken into account in every design? In the rush to get things out
the door and to market, mistakes do get made. We just haven't seen a
high profile exploit of such an issue yet, but we can't say that it
can't possibly happen.


--
Todd H.
http://www.toddh.net/