Re: [PATCH 5/6] perf_counter: add more context information

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Thu, 2009-04-02 at 20:34 +0200, Ingo Molnar wrote:
> * Peter Zijlstra <a.p.zijlstra@chello.nl> wrote:
> 
> > On Thu, 2009-04-02 at 20:18 +0200, Ingo Molnar wrote:
> > > * Peter Zijlstra <a.p.zijlstra@chello.nl> wrote:
> > > 
> > > > On Thu, 2009-04-02 at 13:36 +0200, Ingo Molnar wrote:
> > > > 
> > > > > > -#define MAX_STACK_DEPTH		255
> > > > > > +#define MAX_STACK_DEPTH		254
> > > > > >  
> > > > > >  struct perf_callchain_entry {
> > > > > > -	u64	nr;
> > > > > > +	u32	nr, hv, kernel, user;
> > > > > >  	u64	ip[MAX_STACK_DEPTH];
> > > > > >  };
> > > > 
> > > > Oh, and Paul suggested using u16s right after I send it out. So 
> > > > I'll either send an update or send a incremental in case you 
> > > > already applied it.
> > > 
> > > yes, that's probably a good idea. Although u8 might be even better - 
> > > do we ever want to do more than 256 deep stack vectors? Even those 
> > > would take quite some time to construct and pass down.
> > 
> > We'd have to pad it with 4 more bytes to remain u64 aligned,
> 
> ok, indeed.
> 
> > [...] also, why restrict ourselves. That MAX_STACK_DEPTH limit is 
> > trivially fixable if indeed someone finds its insufficient.
> 
> well .. think about it: walking more than 256 stack frames for every 
> IRQ event? Getting backtraces like:
> 
>  <func_0+0x123>
...
>  <func_269+0x123>
> 
> does that make much sense _per event_? How do you visualize it?

You can use it to calculate aggregate times. Eg. attribute the time
spend in func_0 to func_1 to func_2 etc. And use a tree view based on
these call-chains, allowing you to drill-down -- which is basically what
the sysprof GUI does.