Q: IXP425 CPU (over) clock ?

Do you have a question? Post it now! No Registration Necessary

Translate This Thread From English to

Threaded View
Hello,

I am fiddling around with an NSLU2 from Linksys, which has a IXP425
operating at 266 MHz, if you remove a distinct resistor.

Now I tried to overclock the CPU by replacing the 33 MHz OSC with a 36
MHz Oscillator. But the reported CPU frequency is still 266 MHz.

Does the CPU incorporate something like a clock speed limit?

Does anybody know?

I tried to get info by downloading several in depth whitepaper from
Intel download website, but although they were in depth, they were not
depp enough.

In good hope for answers

Heinz

Re: Q: IXP425 CPU (over) clock ?
Quoted text here. Click to load it

How do you think the system should detect the changed speed ? If the
single OSC is changed the system time will be faster together with the
CPU and thus the speed test results in the same value, even though the
working speed is faster. _Hopefully_ the Ethernet, serial port and USB
clocks are not derived from that OSC, otherwise you will see
communication problems.

-Michael

Re: Q: IXP425 CPU (over) clock ?
Quoted text here. Click to load it

Hello,

I do not have any problems with ethernet or USB.

The bogo mips with teh original settings is somewaht around 133, because
a set of resistors forces the CPU to slow down to 1/2 of the nominal CPU
speed.

If you remove one resistor, you "de-inderclock" to the CPU frequency of
266 MHz.

I expected the CPU speed  (bogo mips) to show more than 266 after the
OSC replacement.

I startet with just 3 MHz more, because I wanted to try it stepwise from
slower to faster.

As far as I know, the 266 MHz is derived from the OCS-freq. multiplied
with a certain factor. The factor will be 266MHz / 33 MHz = 8,060606...

I expected the CPU speed would be about 36MHz * 8,060606... = 290 MHz
after the replacement. Furthermore, I expected that bogo mips will
calculate the value, not just printing a preset value.

At the moment, I wonder if the CPU is limiting the incoming osc
frequency by itself. But I cannot find any clues, that this might be the
reality.

Bye,
Heinz

Re: Q: IXP425 CPU (over) clock ?
Please re-read my post.

A frequency, bogo-mips or whatever speed measurement needs to be made
against a known clock. So the de-underclocking should show with the
bogo-mips, as the timer used supposedly is running at the same speed as
before. Unless there is a second OSC, the timer runs faster, too, if you
change the OSC speed outside of the chip. So the measured CPU speed
value will be the same even if the CPU runs faster.

-Michael

Re: Q: IXP425 CPU (over) clock ?
Quoted text here. Click to load it

... but does the CPU "know" about the external oscillators frequency?

I cannot imagine, that the CPU knows that there has to be a 33MHz osc
somewhere on the PCB and that by definiton the CPU speed has to be 266 MHz.

As far as I understood the ixp425 manual, intel is manufacturing the
CPUs in three different speeds.

That brought me to the conclusion, that the cpu might be able to somehow
reduce the clock that is entered into the clock in pin. But that is very
vague.

Bye,
heinz


Re: Q: IXP425 CPU (over) clock ?
Quoted text here. Click to load it

How should it know ? The OSC does not have a communication channel to
tell the CPU at what speed it is supposed to run.

Quoted text here. Click to load it

You need to use software to set the internal clock speed generation
(derived from the osc input) for the CPU and the hardware timers and the
software timers of the OS according to the osc used, if you want that
e.g. the jiffie ticker works as expected.

Quoted text here. Click to load it

That is just the max speed the chip is guaranteed to work with. It can
work slower if using another osc and/or setting other parameters by
software/hardware configuration. Now it will consume less power.

Quoted text here. Click to load it

It can't. but it can set parameters for it's internal clock generating
hardware and modify it's running speed by that.

-Michael

Re: Q: IXP425 CPU (over) clock ?
Quoted text here. Click to load it

Hello,

I would say that the system is measuring against the RealTimeClock as an
  independant source.

BTW, how do you think the system is detecting speed? In my opinion, it
has to count cycles and compare it with a reference souce, e.g. the RTC.

Am I wrong?

Heinz

Re: Q: IXP425 CPU (over) clock ?
Quoted text here. Click to load it

Is there a separate crystal for the real time clock on the PCB ? If yes
this is possible. You would need to look at the Kernel source to be sure.

Quoted text here. Click to load it

Of course you are right, but I suppose that there is a software module
that calculated the speed and another module that creates some kind of
ticker that is used by the first one. Both might depend on the board you
are using by means of Kernel configuration and/or special source code.

-Michael

Re: Q: IXP425 CPU (over) clock ?

Quoted text here. Click to load it

The RTC is usually only used to set a counter at system start. The
counter is then run by a timer which, on a typical ARM ASIC is derived
from the system clock.

Kind regards,

Iwo


Re: Q: IXP425 CPU (over) clock ?
Heinz Rohde a écrit :
Quoted text here. Click to load it

No you are right, but what if your RTC is using the same clock
source as your chip?

The IXP425 has internal timers, so if these are used to serve
as a reference for BogoMIPS for instance, they will be wrong
since their speed is derived from the external OSC.  Cf section
2.1.14 of http://download.intel.com/design/network/datashts/25247905.pdf
which says:

   The IXP42X product line and IXC1100 control plane processors
   consists of four internal timers operating at 66.66 MHz
   (which is 2 * OSC_IN input pin.) to allow task scheduling and
   prevent software lock-ups. The device has four 32-bit counters:
    Watch-Dog Timer  Timestamp Timer  Two general-purpose timers

If you want to be sure of what timer is used, take a look at the
Linux kernel source.


            Laurent

Site Timeline