Re: [CFT]hwpmc update for sandybridge-e

On Wed, Oct 17, 2012 at 11:41 AM, Jim Harris <jim.harris_at_gmail.com> wrote:

> On Mon, Oct 15, 2012 at 5:49 PM, hiren panchasara
> <hiren.panchasara_at_gmail.com> wrote:
> <snip>
> >
> > Thanks Fabien for getting back quickly. As I am not clear about uncore
> > support, I've removed it entirely as you suggested. I've tried to
> > incorporate other comments too.
> >
> > New set of diffs,  pmccontrol and pmctest.py results are here:
> >
> > http://www.strugglingcoder.info/patches/hwpmc_sbx_2.txt
> > http://www.strugglingcoder.info/patches/pmccontrol_2.txt
> > http://www.strugglingcoder.info/patches/pmctestpy_2.txt
> >
> > I now have following event counters failing from pmctest.py script:
> >
> > CYCLE_ACTIVITY.CYCLES_L2_PENDING
> > BACLEARS.ANY
> > L2_TRANS.DEMAND_DATA_RD
> > L2_TRANS.RF0
> > SQ_MISC.SPLIT_LOCK
>
> Hi Hiren,
>
> With your patch, I see different event counters failing on my SNB Xeon
> system:
>
> CYCLE_ACTIVITY.CYCLES_L2_PENDING - this is due to missing IAP_F_FM
> which Fabien already pointed out.  There are 5 total that need this
> added - this one is the only one that fails from the pmctest.py script
> because it's the only one with bit 0 set (IAP_M_CORE) in its flags
> which causes a hard pmc_allocation failure.
> MEM_LOAD_UOPS_LLC_MISS_RETIRED.LOCAL_DRAM - this is due to adding this
> event to the SNB Xeon event list in the header file, but IAP_F_SBX is
> missing from the associated IAPDESCR for D3H_01H  in hwpmc_core.c.
>

Thanks for the pointers Jim!

It seems that everything works now with:

http://www.strugglingcoder.info/patches/hwpmc_sbx_3.txt
http://www.strugglingcoder.info/patches/pmccontrol_3.txt
http://www.strugglingcoder.info/patches/pmctestpy_3.txt

As it can be seen, I do not see any errors with the test script.
I've also tried to take care of the sorting issues Fabien mentioned.

Can you please give this a whirl?

>
> It's not clear to me why you are seeing failures on those other
> events.  Maybe make sure libpmc/pmcstat/pmccontrol/hwpmc are all in
> sync?
>

I have a few questions regarding this. For my testing I am using hwpmc as a
module.
Problem is that any change to pmc_events.h is not reflected when I do:
1) rebuild /sys/modules/hwpmc - unload/load new module
2) rebuild /lib/libpmc/
3) rebuild /usr.sbin/pmccontrol and pmcstat

I have to do buildworld for changes in pmc_events.h to take effect. (which
kind of explains my delayed responses :D )
I know I am missing something here and there should be a better way to do
this. Any other changes (other than pmc_events.h)
can be tested with above 3 steps.


> Thanks for working on this!
>
Entirely my pleasure :-)

Thanks,
Hiren