RFC: performance monitor/oprofile support for e500

Kumar Gala kumar.gala at freescale.com
Mon Nov 1 16:43:52 EST 2004


Andy,

Some feedback:

* Any reason to get ride of the 604 code now,w/o replacing it with 
something?
* Why have you introduce two CPU features (CPU_FTR_CAN_USE_PMON_INTR & 
CPU_FTR_FSL_BOOKE_PMON) you dont use ?
* In the PerformanceMonitor can't you use regs->nip for the same 
purpose of regs->sia?
* Does arch/ppc/kernel/perfmon.c really need all the headers you are 
including?
* Does perfmon.c make more sense as perfmon_fsl_booke.c
* op_ppc32_setup, I'd prefer the saving & restoring of the perf_irq 
like arch/ppc64/oprofile/common.c does
* a number of issues related to SMP
* any reason get_kernel is a function?
* Do ctr_read & ctr_write really belong in op_impl.h, especially since 
different implementations will be needed for classic ppc.


I sent a note to Paul M. about putting num_counters into cpu_specs 
table.

- kumar

On Oct 29, 2004, at 8:16 PM, Fleming Andy-afleming wrote:

> This patch:
>  * Adds support for using the performance monitor counters and the
> associated interrupt, but only enables it currently for e500
>  * Removes the old 604 performance monitor code, suggested by various
> people on IRC
> * Adds oprofile support for the e500, with potential for expansion to
> support other ppc32 processors
>
> Andy Fleming
>  NCSG Open Source Team
>  Freescale Semiconductor, Inc
> <oprofile-kernel-patch><ATT2027925.txt>




More information about the Linuxppc-embedded mailing list