Re: [RFC PATCH 4/9] mce: Add persistent events

[Ingo Molnar <mingo@elte.hu>]

* Steven Rostedt <rostedt@goodmis.org> wrote:

> On Fri, 2010-08-06 at 16:57 +0200, Peter Zijlstra wrote:
> > On Fri, 2010-08-06 at 16:25 +0200, Borislav Petkov wrote:
> > > +static const struct file_operations perf_mce_fops = {
> > > +       .llseek         = no_llseek,
> > > +       .open           = mce_perf_open,
> > > +       .poll           = perf_poll,
> > > +       .unlocked_ioctl = perf_ioctl,
> > > +       .compat_ioctl   = perf_ioctl,
> > > +       .mmap           = perf_mmap,
> > > +       .fasync         = perf_fasync,
> > > +       .release        = perf_release,
> > > +}; 
> > 
> > I'd rather see this part of the persistent bits live in
> > kernel/perf_event.c, that way you don't need the previous patch either.
> > 
> 
> This is part of what I hate about the perf design. The fact that 
> everything needs to be very coupled. I would like the infrastructure 
> to be more flexible.

Well, Peter's comment was mostly about not making this arch specific but 
core kernel - the new bits can live in kernel/perf_event_persistent.c 
just fine.

There is no 'flexibility' in each arch doing the same thing with small 
variations.

More generally i agree with you that better modularization helps - a 
nice improvement in this area would be to do a kernel/perf_event/ 
splitup: i.e. split kernel/perf_event.c (which is getting a tad big) 
into:

  kernel/events/output.c
  kernel/events/trace.c
  kernel/events/core.c
  kernel/events/syscall.c
  kernel/events/persistent.c

Maybe even move hw_breakpoint.c there, etc. There's already various 
wrappers in perf_event.c that could be split out of the core code to 
increase modularity. Later on we could have a sysfs.c, etc.

Any taker for such a massive restructuring? :)

Thanks,

	Ingo