[RFC] perf: perf_event_attr anon unions and static initializer issue

[RFC] perf: perf_event_attr anon unions and static initializer issue

[Stephane Eranian]

Hi,

I am writing a little test program for the perf_event API which is using
hardcoded events. Some of those events (SNBEP uncore events) require
a value for config1. I was naively assuming, one could simply do:

struct perf_event_attr attr = { .config = 0x1234, .config1 = 0x456 };

However this does not work with gcc version < 4.6 apparently. Don't
know how this even works with other compilers. I looked online and
I could not really find a good way of solving this that would be portable
across compiler versions.

If we name those unions, then this will break lots of existing programs.
We'd have to create a "named" version of perf_event_attr and then
memcpy() at runtime. Or simply initialize the .config1 and such fields at
runtime from a separate table.

Does anyone have a better solution to propose?

Re: [RFC] perf: perf_event_attr anon unions and static initializer issue

[Peter Zijlstra]

On Fri, 2012-10-05 at 12:36 +0200, Stephane Eranian wrote:

> struct perf_event_attr attr = { .config = 0x1234, .config1 = 0x456 };

> Does anyone have a better solution to propose?


  struct perf_event_attr attr = { 
	.config = 0x1234, 
	{ .config1 = 0x5678 },
  };

sometimes works, but apparently not in this case.. its a bit unfortunate
indeed. EDG based compilers and the latest C std use your version --
hence also 4.6 supporting it.

Yeah, I'm afraid we're stuck with this until a future where modern C
isn't modern anymore.

Re: [RFC] perf: perf_event_attr anon unions and static initializer issue

[Stephane Eranian]

On Fri, Oct 5, 2012 at 1:01 PM, Peter Zijlstra <peterz@infradead.org> wrote:
> On Fri, 2012-10-05 at 12:36 +0200, Stephane Eranian wrote:
>
>> struct perf_event_attr attr = { .config = 0x1234, .config1 = 0x456 };
>
>> Does anyone have a better solution to propose?
>
>
>   struct perf_event_attr attr = {
>         .config = 0x1234,
>         { .config1 = 0x5678 },
>   };
>
> sometimes works, but apparently not in this case.. its a bit unfortunate
> indeed. EDG based compilers and the latest C std use your version --
> hence also 4.6 supporting it.
>
> Yeah, I'm afraid we're stuck with this until a future where modern C
> isn't modern anymore.

Yeah, unfortunately.
I ended up creating a separate table for config1 and initializing attr.config1
at runtime. That way it works regardless of the compiler....

Re: [RFC] perf: perf_event_attr anon unions and static initializer issue

[Borislav Petkov]

On Fri, Oct 05, 2012 at 01:49:44PM +0200, Stephane Eranian wrote:
> Yeah, unfortunately. I ended up creating a separate table for config1
> and initializing attr.config1 at runtime. That way it works regardless
> of the compiler....

Did the same thing in the persistent events patches recently:

http://marc.info/?l=linux-kernel&m=134513917126309

-- 
Regards/Gruss,
Boris.