Re: [PATCH] powerpc: Keep track of emulated instructions

[Michael Neuling <mikey@neuling.org>]

In message <alpine.LRH.2.00.0904070926001.17130@vixen.sonytel.be> you wrote:
> On Tue, 7 Apr 2009, Michael Neuling wrote:
> > In message <alpine.LRH.2.00.0904061430090.11076@vixen.sonytel.be> you wrote
:
> > > On Fri, 3 Apr 2009, Kumar Gala wrote:
> > > > On Apr 3, 2009, at 7:08 AM, Geert Uytterhoeven wrote:
> > > > >Finally (after ca. 1.5 years), he're an updated version of my patch to
 k=
> > > eep
> > > > >track of emulated instructions.  In the light of Kumar's `Emulate enou
gh=
> > >  of
> > > > >SPE
> > > > >instructions to make gcc happy' patch, he probably also wants to keep 
tr=
> > > ack
> > > > >of
> > > > >the actual runtime overhead.
> > > > >
> > > > >Changes since last version:
> > > > > - arch/powerpc/kernel/sysfs.c is now compiled on ppc32, so we can pro
vi=
> > > de
> > > > >   counters in sysfs on ppc32, too,
> > > > > - WARN_EMULATED() is a no-op if CONFIG_SYSCTL is disabled,
> > > > > - Add warnings for altivec,
> > > > > - Add warnings for recently introduced emulation of vsx and isel
> > > > >   instructions.
> > > > =
> > > 
> > > > pretty cool.  Do we think counters should be sysfs or debugfs?
> > > 
> > > What do you prefer?
> > > 
> > > On MIPS, unaligned exception handling control is in debugfs.
> > 
> > Would per process counters be too hard?  Stick them in the thread_struct
> > and export them via /proc/<pid>/emulated. 
> 
> But they go away as soon as the process exits, right?

True.  

taskstats would be better then but you'd have to start touching generic
code for that.

Mikey