On Mon, 17 May 2010, Peter Zijlstra wrote:
> On Fri, 2010-05-07 at 10:05 +0100, Eric B Munson wrote:
> > This patch splits tracking of executable and non-executable mmaps.
> > Executable mmaps are tracked normally and non-executable are
> > tracked when --data is used.
> >
> > Signed-off-by: Anton Blanchard <anton@xxxxxxxxx>
> >
> > Updated code for stable perf ABI
> > Signed-off-by: Eric B Munson <ebmunson@xxxxxxxxxx>
>
> > +++ b/include/linux/perf_event.h
> > @@ -197,6 +197,7 @@ struct perf_event_attr {
> > exclude_hv : 1, /* ditto hypervisor */
> > exclude_idle : 1, /* don't count when idle */
> > mmap : 1, /* include mmap data */
> > + mmap_exec : 1, /* include exec mmap data*/
> > comm : 1, /* include comm data */
> > freq : 1, /* use freq, not period */
> > inherit_stat : 1, /* per task counts */
>
> You cannot add a field in the middle, that breaks ABI.
>
> > -static inline void perf_event_mmap(struct vm_area_struct *vma)
> > -{
> > - if (vma->vm_flags & VM_EXEC)
> > - __perf_event_mmap(vma);
> > -}
>
> Also, the current behaviour of perf_event_attr::mmap() is to trace
> VM_EXEC maps only, apps relying on that will be broken after this patch
> because they'd have to set mmap_exec.
>
> If you want to do this, you'll have to add mmap_data (to the tail of the
> bitfield) and have that add !VM_EXEC mmap() tracing.
>
Thanks, I will get right on the changes.
--
Eric B Munson
IBM Linux Technology Center
ebmunson@xxxxxxxxxx
Attachment:
signature.asc
Description: Digital signature