Hi Stefane,
Thanks for the comments. Some notes, why can't I hardcode PMC values
for a platform which I write all the register definitions for? I do this
on MIPS also...For performance sake, I really want these functions to be
as fast as possible. For the PPC32 and PPC64 series, all can read these
PMC's as written in the header file. It could be a 'limitation' of this
particular architecture that the mapping is fixed...in fact, that might
be a benefit! I'm planning on putting the SDAR and SIAR (the address
sample registers) after PMC8.
I booted the kernel, but I think it choked...have to wait for the old
man to send me the boot log.
Phil
Stephane Eranian wrote:
Phil,
On Tue, Nov 07, 2006 at 06:13:08PM +0100, Philip J. Mucci wrote:
Hi Stefane,
I've been playing around on the PPC32 box I have, in preparation for
looking at the PPC64 code. (I have a box, just not root access to the
latter.)
I've reworked a lot of the code to resemble the MIPS code in many
ways...especially since the MIPS code is so much simpler to understand.
I've also removed the TB register from the PMD defs for now and included
skeletons for PPC970, Cell and Power4.
I looked at some of the files and in particular asm-powerpc/perfmon.h
and I have a few remarks.
You cannot hardcode the mapping from PMCx to SPRN_* in pfm_arch_write_pm*().
The mapping is driven by the mapping table. As such the SPRN_* values to use
must come from the table and not be in the switch() statement. This is how we
do this for x86 for instance.
Some of the complexity of x86 mapping comes from the P4 and hyperthreading
support.
But I have now duplicated the HW address of each register into the generic
mapping
table pfm_pmu_conf->pmc_desc[].hw_addr. You could use this instead if you
ppopulate your mapping table correctly.
Anyways, attached are files for PPC32 and PPC64 support that get us a
little closer to working support. My bet is that with proper definitions
of the registers in the PPC32 file, we'll be very close to working PPC32
support (and as such, other PPC support). I believe I sent you the name
of the machine if you want to have access...you can find my patched and
built kernel in /usr/src there.
If I get time, I'm going to boot the kernel I built using the following
and see if I can get it to do something.
You still need some minimal libpfm support to get examples/self working...
anyway, this is good progress...
Thanks Phil!
_______________________________________________
perfmon mailing list
[email protected]
http://www.hpl.hp.com/hosted/linux/mail-archives/perfmon/
