Re: PREEMPT_RT vs ADEOS: the numbers, part 1

[Ingo Molnar <mingo@elte.hu>]

* Karim Yaghmour <karim@opersys.com> wrote:

> Ingo Molnar wrote:
> > how were interrupt response times measured, precisely? What did the 
> > target (measured) system have to do to respond to an interrupt? Did you 
> > use the RTC to measure IRQ latencies?
> 
> The logger used two TSC values. One prior to shooting the interrupt to 
> the target, and one when receiving the response. Responding to an 
> interrupt meant that a driver was hooked to the target's parallel port 
> interrupt and simply acted by toggling an output pin on the parallel 
> port, which in turn was hooked onto the logger's parallel port in a 
> similar fashion. [...]

FYI, there's a new feature in the -V0.7.48-25 (and later) -RT patches 
that implements this: CONFIG_LPPTEST. It is a simple standalone driver 
and userspace utility from Thomas Gleixner that can be used to measure 
the IRQ-latency of the system over a null-modem-parallel-cable.

to use it, enable CONFIG_LPPEST in the .config [disable CONFIG_PARPORT 
first], boot the kernel on both the target and the host systems, and 
then run the scripts/testlpp utility on the host system which will 
measure latencies and will do a maximum-search.

(the driver assumes normal LPT1 PC layout - 0x378/IRQ7)

	Ingo