Re: [PATCH] perf_counter: extensible perf_counter_attr

[Corey Ashford <cjashfor@linux.vnet.ibm.com>]

Paul Mackerras wrote:
> Corey Ashford writes:
> 
>> Ok, some disclosure here: we have not yet supported either of these features in 
>> the Power PMU's in any open source code (and perhaps the proprietary code too, 
>> but I don't know about that).  Since these features are described in an IBM 
>> proprietary document, I can't describe how they work here, except to say that 
>> they are present in the chip.
> 
> Actually, the public PPC970FX user manual has a chapter on the
> performance monitor unit which describes both thresholding and the
> instruction matching CAM, among other things.

Ah, good point!

Anyone interested can find the 970FX (G5) manual here: www.cs.lth.se/EDA116/G5.pdf
It's supposed to be on the IBM web site too, but for some reason, the link is 
broken (reported!).

> 
> I think we can support thresholding using higher-order bits of the
> event code when the low bits == PM_THRESH_TIMEO.  We can only have one
> PM_THRESH_TIMEO event on the PMU at any given time, so there can't be
> any conflict over the thresholder settings.
> 
> The IMC is more problematic, but we can't do much with it on POWER5
> and later processors anyway, due to various things being accessible
> only in hypervisor mode, so I have been ignoring it. :)
> 
> Paul.

Ug.  Well, there's always bare-metal Linux :)

- Corey