On Thu, 30 Jan 1997, Praveen Dwivedi wrote:
> >
> > > Recently I saw this behaviour on linux 2.0 running on i486.
> > > Instruction invd ( supposed to invalidate cache) can lock the m/c
> > > completely.
> >
> Alan Cox writes -
>
> > If you can do this from userspace can you try the same with a different
> > OS on the same hardware. It might he hardware, it might be trap handling
> > it might be a buggy chip or a general flaw in a stepping
> >
> > I wouldnt rule out hardware. The 386 can be crashed with an instruction
> > from userspace
> >
> > Alan
> >
>
> This seems to be the conclusion I would infer too. This is the output
> of cat /proc/cpuinfo:
> processor : 0
> cpu : 486
> model : DX/4-WB
> vendor_id : AuthenticAMD
> stepping : 4
> fdiv_bug : no
> hlt_bug : no
> fpu : yes
> fpu_exception : yes
> cpuid : yes
> wp : yes
> flags : fpu
> bogomips : 49.86
>
> I tried the same program (containing INVD instruction) on a
> 486 from Intel, nothing spectular happened. Though I did not get
> fault there too, at least no crash. As far as my knowledge goes
> INVD should be a privileged instruction on both 486 and Pentium.
> AMD can produce such CPUs where a userspace program having invd instruction
> can crash the machine. The poor OS can't help if Hardware is buggy.
>
> Cheers,
> Praveen
>
>
> --
>
> ----------------------------------------------------------------------
> Praveen Dwivedi pkd@sequent.com
> --
>