Re: Call stack profiling

[David S. <ds...@al...>]

John Levon wrote:
> 
> On Fri, Sep 13, 2002 at 08:29:38PM +0200, David Smith wrote:
> 
> > All that is needed is a patch to the nmi handler to make sure the page
> > fault handler knows we're in an interrupt (see below).
> >
> > The patch I'm currently using has get_user() instead of locking, and
> > so far there have been no problems (that is still all with RTC on
> > a single pentium classic, though...).
> 
> If you can send me a patch that should work using NMI, I'll see to give
> it a go on my two-way.
>

This is the patch with get_user() instead of locking. (It also has
some code I was playing with to check that the return address from
syscall_regs() is actually "int 0x80", but that shouldn't hurt :)

The changes to op_cpu_type.c anc op_events_desc.c are just to 
be able to see the extra counter with RTC.

> My 2.5 stuff might actually allow true call stack profiling. e.g. set
> depth 4, then simply dump 4 EIPs in the sample buffer. This would
> increase throughput by a factor of 4, but would allow true call stack
> profiling. I have no idea whether this would be too slow, but it's
> something interesting to consider.

That would be great!

/David