Re: [PATCH 2/9] x86 idle: remove NOP cpuinfo_x86.hlt_works_ok flag

[Alan Cox]

On Thu, 31 Mar 2011 07:19:50 -0700
"H. Peter Anvin" <hpa@xxxxxxxxx> wrote:

> On 03/31/2011 02:35 AM, Ingo Molnar wrote:
> > 
> > * Alan Cox <alan@xxxxxxxxxxxxxxxxxxx> wrote:
> > 
> >>> Btw., we used to auto-detect broken HLT systems IIRC - but that got lost 
> >>> already. We should at least honor the boot parameter.
> >>
> >> I don't believe we ever auto detected them or found a way to do so. That
> >> was why the HLT message was printed before hlt was executed.
> > 
> > Yeah - the CPU hang was unrecoverably deep so no auto-detection was possible.
> > 
> > That's seriously ancient stuff - still, keeping the boot option around (<10 
> > lines of code) does not hurt anyone.
> > 
> 
> What it was was bad power supplies or low-capacitance, high-inductance
> power distribution that happened to work with MS-DOS which always burned
> the CPU at 100% and therefore left the power draw relatively consistent
> current.  A proper OS putting the CPU in HLT produced a lot more high
> frequency noise on the power busses, with disastrous results without
> proper bypass.

And also chipset errata in some cases - eg some revisions of the CS5510
hung the box solid if a CPU hlt occurred during an IDE transfer. I don't
think any CS5510s are still around although I've had mail from someone
with a CS5520 in use not that long ago so who knows!

The joy of ancient history.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/