On Wed, 8 Jun 2016, tip-bot for Dave Hansen wrote:
> Commit-ID: ef5f9f47d4ec4cf42bac48c7c4dafacc1b9f0630
> Gitweb: http://git.kernel.org/tip/ef5f9f47d4ec4cf42bac48c7c4dafacc1b9f0630
> Author: Dave Hansen <dave.han...@linux.intel.com>
> AuthorDate: Thu, 2 Jun 2016 17:19:29 -0700
> Committer: Ingo Molnar <mi...@kernel.org>
> CommitDate: Wed, 8 Jun 2016 12:05:58 +0200
>
> perf/x86/intel: Use Intel family macros for core perf events
>
> Use the new model number macros instead of spelling things out
> in the comments.
...
> #include "../perf_event.h"
> @@ -3319,11 +3320,11 @@ static int intel_snb_pebs_broken(int cpu)
> u32 rev = UINT_MAX; /* default to broken for unknown models */
>
> switch (cpu_data(cpu).x86_model) {
> - case 42: /* SNB */
> + case INTEL_FAM6_SANDYBRIDGE:
I personally find this a step backwards.
Currently it's bad enough when someone reports a problem with PAPI/perf
"my Xeon processor isn't detected properly"
eventually you can get them to send /proc/cpuinfo so you can try to find
out what processor they really have, and in that case you can get
the family/model numbers in plain decimal, and you can easily look up
if things are supported in the various files.
Now I then need to go look in the source, then find that it says
INTEL_FAM6_SANDYBRIDGE and then need to take the extra step to find
whatever header file defines these values.
Anway I doubt this will change anyone's mind but just wanted to register
my complaint.
Vince
and don't get me started about trying to somehow match a model number to a
code name using the AMD or intel documentation, without resorting to
wikipedia or random tech sites. I'm looking at you AMD fam15h model 60h.
Also "intel 6th generation i7"
