On Fri, Nov 16, 2012 at 11:35:37AM -0800, Sukadev Bhattiprolu wrote:

Jiri Olsa [jolsa@redhat.com] wrote:
| > 
| > Can we remove the assumption that the event id is a generic event that
| > has PERF_COUNT_HW_ prefix and also let the architectures pass in a "show"
| > function ? This would allow architectures to display any arch specific
| > events that don't yet have a generic counterpart.
| > 
| > IOW, can we do something like this (untested) and make PERF_EVENT_ATTR global:
| 
| hm, then you probably can use following:
| 
| http://www.spinics.net/lists/kernel/msg1434233.html

For now, power events can simply be u64 - so am hoping to have something
like this and replace the raw codes:

#define PM_CYC			0x1e
#define PM_GCT_NOSLOT_CYC	0x100f8

EVENT_ATTR_STR() is interesting, but would require new/extra macros like

#define PM_CYC_STR		"0x1e"
#define PM_GCT_NOSLOT_CYC_STR	"0x100f8"

I went down the path we discussed earlier - to have x86 and Power share
the EVENT_ATTR() macros. This is making the x86 code less concise

	EVENT_ATTR(cpu-cycles,	CPU_CYCLES)

becomes

	EVENT_ATTR(cpu-cycles,	PERF_COUNT_HW_CPU_CYCLES, events_sysfs_show)

with each EVENT_ATTR() explcitly adding events_sysfs_show or needing another
wrapper around the wrappers.

Still tweaking it, but would it be flexible to make 'struct pmu_events_attr'
common and let architectures define the EVENT_ATTR() as needed ? Would
duplicate some code but hopefully not a lot.

sounds good, let's see ;)

jirka