RE: [ECOS] Clock/interrupt latency

[Gary Thomas <gary at mlbassoc dot com>]

On Thu, 2003-08-28 at 03:55, Daniel Lidsten wrote:
> > > Hi,
> > > 
> > > I have a question regarding the clock/interrupt latency (on a 
> > > MPC823e). When i run the tm_basic test then i get a very high 
> > > variation in the measured times as you can see below.
> > > 
> > > If i look in the realtime characterization section in the 
> > user manual 
> > > then i find a variation of 3 to 5 times between maximun and 
> > average to 
> > > be normal, not as much as my measure.
> > > 
> > > ---- tm_basic diag------
> > >     4.75    3.84  158.72    0.00            Clock/interrupt latency
> > >     5.32    3.84   12.16    0.00            Clock DSR latency
> > > ---- tm_basic ------
> > > 
> > > I tried to define diag_printf as printf and then i got the below 
> > > measurements. How come that diag_printf has that effect on 
> > my measure?
> > > 
> > > ---- tm_basic printf ------
> > >    4.75    3.84    8.96    0.00            Clock/interrupt latency
> > >    5.20    3.52   11.84    0.00            Clock DSR latency
> > > ---- tm_basic ------
> > 
> > diag_printf uses a very simple serial driver which will work 
> > even under the worst conditions of a very limited system 
> > which is currently crashing etc.... To achieve this its very 
> > simple, polled and with interrupts disabled. Having 
> > interrupts disabled will thus affect interrupt latency. It 
> > looks like tm_basic is for some reason printing while making 
> > measurements.
> > 
> > How are you running tm_basic? With gdb? dbg over ethernet? Or 
> > something simple like using redboot to load it into RAM and then go?
> > 
> >       Andrew
> 
> I know that diag_printf is a simple driver but since it is used as
> default in tm_basic then i assume that no printing is done during the
> measurements. As i understand the MPC then the decrementer isnt
> maskable, which means that it cannot be put on hold (pending) so why
> does one interrupt (max time) takes that long time. Even if diag_printf
> disables all interrupt then i think that we will serve the decrementer
> interrupt, am i wrong?
> 

Sorry, yes.  On the PowerPC, decrementer interrupts can only happen
when interrupts (there is only one "master enable") are masked.

'tm_basic' is designed to take this into account, by taking time
measurements only when printing is not taking place.

> I am running eCos from RAM using an ICE debugger through a BDM
> interface.
> 
> Regards, Daniel
-- 
Gary Thomas <gary@mlbassoc.com>
MLB Associates


-- 
Before posting, please read the FAQ: http://sources.redhat.com/fom/ecos
and search the list archive: http://sources.redhat.com/ml/ecos-discuss