Re: Pentium DEATH in user-mode

Alain Knaff (knaff@tux.org)
Sat, 8 Nov 1997 16:55:26 +0100 (MET)

In linux, you wrote:
>
>
>On Fri, 7 Nov 1997, Richard B. Johnson wrote:
>
>> On Fri, 7 Nov 1997, Robert R. Collins wrote:
>>
>> > Richard B. Johnson, you wrote:
>> > >
>> > >The following user-mode program will stop the internal clock
>> > >of a Pentium.
>> >
>> > Actually, it doesn't stop the internal clock. It sends out two
>> > (back-to-back) locked memory read cycles, without any intervening locked
>> > memory write cycles. This (most likely) causes the state machines of the
>> > chipsets to go bonkers. The net effect is a crashed system.
>> >
>> > >
>> > >char x[5]={0xf0,0x0f,0xc7,0xc8,0x00};
>
> Is is possible to make a kernel patch or something else to keep an
>eye out for these byte codes? Without some sort of protection this could
>case some major problems with pentium based linux boxes.
> Cheers Adam

Not easily. After all, these instructions are not privileged at all,
so they don't trigger any trap which the kernel could disable. The
only way to stop this from kernel mode would be to scan all executable
page for this sequence as soon as it is brought into memory. And this
would not work for pages which are both writable and executable, as
the program could change the contents after the mapping is done.

So, no, either upgrade, or watch closely who you grant an account on
your Linux box...

Regards,

Alain