On Wed, 24 Jul 2013, Andi Kleen wrote: > On Tue, Jul 23, 2013 at 05:27:43PM -0400, Vince Weaver wrote: > > > So format/precise would be > > flags:15-16 > > on little endian. Looks good? I'd prefer if it were precise_ip:0-1 as the code you use to set the value looks like attr->precise_ip=2; I don't think you can even set bits 15-16 of flags directly, can you? Althought having to have a parser being able to parse the names of all 50+ perf_event->attr arguments, then parse hex/decimal strings from another file in sysfs, then trying to map that to binary fields in a complicated structure that is then passed to a syscall gets a bit stupid at some point; this is one of the silliest interfaces I've had to deal with and I've done low-level X11 programming before. Ideally we could do something sane, such as maybe having the event descriptions in a library in userspace where it belongs but it's a bit late to close the barn door now. > > I'm a bit grumpy about this because I just finished fixing the fallout > > from your last time breaking the ABI a few weeks ago when your broken code > > started dumping non-hex fields into the sysfs event strings. I've learned > > Not sure what you're talking about? I'm pretty sure it was one of your patches that added the new memory stuff which had "ldlat=1" in one of the events, and that was not a hex value prefixed by 0x as specified in the ABI documentation and thus broke various parsers. By the time I caught this the perf_event maintainers declared it was too late to do anything about it. Vince -- To unsubscribe from this list: send the line "unsubscribe trinity" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
