On 07/05/2012 11:47 AM, Will Deacon wrote:
Hi Maynard,

On Thu, Jul 05, 2012 at 05:32:02PM +0100, Maynard Johnson wrote:
*Will*, it would be great if you could test out Gergely's + my patch in you cross-compile
environment, too.
It's on the list, I promise! Actually, I had some reservations about using a
header file to determine the availability of perf events on the target. On
ARM at least, there are ARM-specific kernel config options which you need
selected to enable hardware perf events and on top of that the platform may
have to do some specific initialisation to enable interrupts. 
This type of thing sounds like it might be best done as a runtime check.  See the call to _check_perf_events_cap() in operf.cpp:main.  If that capability check is not sufficient for what you're talking about, perhaps it can be extended.

That's why I
preferred the --with-perf-events option, since it makes the choice explicit.

That said, I'll try and make some time to test both the approaches in my