Thomas De Schampheleire wrote:
Hi,
This message was originally posted on the osol-code mailing list, but
someone suggested this is more appropriate for the performance community.
I am using a full-system simulator (Simics) on a simulated OpenSolaris
installation for a research project. The aim of the project is to
implement adaptive processor shutdown for energy savings.
Neat.
We would like to measure the dependence of the interrupt latency
(interrupt response time) on the power state of a processor.
Ok.... how does your simulator model power states? Not many SPARC
CPUs support multiple running power states...
By adding a special ("magic") assembler instruction to the kernel code,
we can halt the simulation and consequently do memory reads/writes or
execute code in the simulator. I was thinking of adding such a magic
instruction at the beginning of the general interrupt code, and just
before the execution of the actual interrupt handler (for that specific
interrupt). I could then find a way of reading out current system time
(I believe this is possible from the simulator, otherwise I could maybe
read out system time directly from memory?).
Is this a correct approach?
Seems plausible...
I would thus add these magic instructions at line 63 (entry) and right
before lines 558 and 1181 (before the actual handler is called)
referring to
http://src.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/uts/sun4/ml/interrupt.s
<http://src.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/uts/sun4/ml/interrupt.s>.
Is this correct?
In interrupt.s, I see that some kind of timestamp is saved before the
interrupt is called. What purpose does it serve, and where is it stored?
This is for interrupt accounting... look at thread structure for
t_intr_start field.
And where can I find a list and explanation of the different assembler
instructions that are used? This would be most useful to me!
I'd take a look at a copy of the SPARC architecture manual...
http://developers.sun.com/solaris/articles/sparcv9.pdf
- Bart
Thank you, Thomas
------------------------------------------------------------------------
_______________________________________________
perf-discuss mailing list
perf-discuss@opensolaris.org
--
Bart Smaalders Solaris Kernel Performance
[EMAIL PROTECTED] http://blogs.sun.com/barts
_______________________________________________
perf-discuss mailing list
perf-discuss@opensolaris.org
