Interrupt Latency

Interrupt Latency

[Jon Baker]

I am curious about interrupt latency.

I have an Embedded Planet RPX Classic CLLF_BW31 MPC860 running at 48Mhz with
non-realtime Hard Hat 1.2 with the 2.2.14 kernel.

In general what kind of interrupt latency can I expect?  I am seeing a pretty
consistent 10 us min IRQ2 latency but occassionally see up to 70 us max.  I am
not using IRQ0 or IRQ1 (therefore IRQ2 is highest priority) and we monitored
this with my application code doing next to nothing.  Basically I am only
running ISR2.  First, I was surprised the 10 us min latency, I thought I might
occassionally see a much quicker response.  Second, I was surprised to see the
occassional slow response of 70 us max.  It would be nice if we could get the
max under 50 us.  Or do you have to go to a real-time kernel?  I am looking into
my code to see that I am not masking the interrupt for too long (or even at all)
and checking the driver code efficiency.  Could the non-realtime linux kernel
mask the interrupts for that long?  I am going to look into what all linux is
doing, I am thinking it is not doing much but have not verified this.  I tried
to find some general interrupt latency numbers on the web through Monta Vista or
others but did not find anything relevant.

Jon Baker

===================================
Jon Baker
Software Engineer
Efficient Channel Coding, Inc.
216-635-1610
www.eccincorp.com
===================================


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Interrupt Latency

[Dan Malek]

Jon Baker wrote:


> ..... It would be nice if we could
> get the
> max under 50 us.  Or do you have to go to a real-time kernel?

Real-time != Real fast.......

Real-time indicates a predictable maximum/minimum, whatever is
appropriate for a given resource to be managed.  Since you asked
a question that implied a maximum interrupt latency requirement,
you are designing a real-time system.  Linux is not a real time
kernel, you will need to use something like RTLinux to meet such
requirements.


	-- Dan


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Interrupt Latency

[Wolfgang Denk]

Dear Jon,

in message <3D0797F8.4030100@eccincorp.com> you wrote:
>
> I have an Embedded Planet RPX Classic CLLF_BW31 MPC860 running at 48Mhz with
> non-realtime Hard Hat 1.2 with the 2.2.14 kernel.
>
> In general what kind of interrupt latency can I expect?  I am seeing a pretty
> consistent 10 us min IRQ2 latency but occassionally see up to 70 us max.  I am

This is about what you can get on such a CPU.

Even with RTAI, running the latency calibration test,  you  will  see
average latencies af some 14 us.

> running ISR2.  First, I was surprised the 10 us min latency, I thought I might
> occassionally see a much quicker response.  Second, I was surprised to see the

What do you expect? This is just a 50 MHz system  with  tiny  caches.
This is not the fast number-cruncher you seem to expect...

> occassional slow response of 70 us max.  It would be nice if we could get the
> max under 50 us.  Or do you have to go to a real-time kernel?  I am looking into

Yes, if you need guaranteed response times you have to use a hard  RT
capable system; my recommendation is RTAI.

> and checking the driver code efficiency.  Could the non-realtime linux kernel
> mask the interrupts for that long?  I am going to look into what all linux is

Sure it can.

Latencies scale more or less with the clock frequency; CPU clock, bus
clock and cache size have visible impact. Here are  some  results  we
got when running the RTAI latency test on a couple of systems:

IBM PowerPC 405GP Rev. D at 198 MHz 16 kB I-Cache 8 kB D-Cache:

	Interrupt latency approx.  5.9 us

MPC855 at 80 MHz / 40 MHz bus clock  4 kB I-Cache 4 kB D-Cache:

	Interrupt latency approx. 22.0 us

MPC860DP at 50 MHz / 50 MHz bus clock  16 kB I-Cache 8 kB D-Cache:

	Interrupt latency approx. 16.0 us

MPC8240 at 247.500 MHz 16 kB I-Cache 16 kB D-Cache:

	Interrupt latency approx.  4.5 us

MPC8260 at 199.999 MHz

	Interrupt latency approx.  4.0 us

PowerPC 750 @ 300 Mhz:

	Interrupt latency approx.  3.0 us

PowerPC 750 at 450 MHz 1 MB L2-Cache

	Interrupt latency approx.  1.5 us


Wolfgang Denk

--
Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
Phone: (+49)-8142-4596-87  Fax: (+49)-8142-4596-88  Email: wd@denx.de
"I've finally learned what `upward compatible' means. It means we get
to keep all our old mistakes." - Dennie van Tassel

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/