Re: [PATCH] perf_counter: extensible perf_counter_attr

From: Paul Mackerras
Date: Tue Jun 09 2009 - 05:58:24 EST

Ingo Molnar writes:

> * Paul Mackerras <paulus@xxxxxxxxx> wrote:

> > Since this is a user-visible structure, you've just introduced an
> > ABI difference between 32-bit and 64-bit processes, which we've
> > managed to avoid so far. [...]
> This is a good point too - handling pointers in ABIs is possible but
> should be avoided as much as possible: it creates the need to
> introduce a compat_sys_perf_counter_open() and doubles the syscall
> table complexity.
> Lets do the s/__reserved_1/attr_size ABI i suggested (i outlined
> various properties of it in the previous mail). Agreed?

Yep, sounds good.

We also need a way to allow access to random machine-specific features
that might not be supported in a generic way, such as the instruction
matching CAM on POWER4/PPC970, or (apparently) PEBS. That's what I
had intended exclusive groups + the extra_config_len field to be used
for, but Peter removed extra_config_len...

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at