On 30/05/08 12:24 -0400, Lennart Sorensen wrote:
quoted text
> On Fri, May 30, 2008 at 10:05:05AM -0600, Jordan Crouse wrote:
> > I think the hamster dropping code is queued for 2.6.27, so at least we're
> > up to mammals.
> > 
> > The story here is that in an unfortunate instance of bad planning the
> > MFGPT timers can only be configured once, so a module can't allocate a
> > timer at init and release it when it is done.  The original object of this
> > code was to try give the timer back to a module if it happened to go away
> > and come back, but that is clearly a more complex process then just simply
> > storing the module name, and this code fell into bitrot.
> 
> It is a bit unfortunate that someone decided to design 'configure once'
> hardware.  What were they thinking?

Agreed, very sub-optimal.   For all those prospective silicon vendors out
there, this is a good lesson.  Always let your software people review
the specification before you freeze the RTL - it will save you grief in
the long run. </sermon>

quoted text
> I run a watchdog using the mfgpt, and I simply tore out the code that
> prevents reuse of the timers, and I decided which timers I am going to
> use for which purpose and never reuse them for anything else, so the
> parts of the configuration in the hardware that is fixed isn't an issue
> then.  I start the watchdog from grub, so I had to override the check
> when the kernel takes over the timer watchdog management after all.

I expect that this will be the primary usage model - every platform needs to
have a gentleman's agreement for how the timers are allocated and used.

quoted text
> > So its not so much that we need to drop module support, rather it needs to
> > be understood that if you remove and insert your module on a regular basis
> > you will run out of timers, and deprive others of the timers too. I think
> > that is a reasonable restriction to impose, given the limited usefulness 
> > of these timers for general purpose use.
> 
> They are useful timers, but yes perhaps not for general purpose.  As a
> source of interrupts at certain intervals or as a watchdog they are not
> too bad.

For the longest time, I thought they were only good as a watchdog or
driving an external output.  I had to eat some crow when we figured out
that we could use them as the tick source for tickless, so I no longer say
that they won't be good for anything else, but their usefulness is
certainly limited.

-- 
Jordan Crouse
Systems Software Development Engineer 
Advanced Micro Devices, Inc.

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