Re: [RFC PATCH] LTTng instrumentation mm (updated)

[Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>]

* Dave Hansen (haveblue@us.ibm.com) wrote:
> On Fri, 2007-11-30 at 12:05 -0500, Mathieu Desnoyers wrote:
> > 
> > 
> > Given a trace including :
> > - Swapfiles initially used
> > - multiple swapon/swapoff
> > - swap in/out events
> > 
> > We would like to be able to tell which swap file the information has
> > been written to/read from at any given time during the trace.
> 
> Oh, tracing is expected to be on at all times?  I figured someone would
> encounter a problem, then turn it on to dig down a little deeper, then
> turn it off.
> 

Yep, it can be expected to be on at all times, especially on production
systems using "flight recorder" tracing to record information in a
circular buffer, then dumping the buffers when some triggers (error
conditions) happens.

> As for why I care what is in /proc/swaps.  Take a look at this:
> 
> struct swap_info_struct *
> get_swap_info_struct(unsigned type)
> {
>         return &swap_info[type];
> }
> 
> Then, look at the proc functions: 
> 
> static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
> {
>         struct swap_info_struct *ptr;
>         struct swap_info_struct *endptr = swap_info + nr_swapfiles;
> 
>         if (v == SEQ_START_TOKEN)
>                 ptr = swap_info;
> ...
> 
> I guess if that swap_info[] has any holes, we can't relate indexes in
> there right back to /proc/swaps, but maybe we should add some
> information so that we _can_.
> 

The if (!(ptr->flags & SWP_USED) test in swap_next seems to skip the
unused swap_info entries.

Why should we care about get_swap_info_struct always returning a "used"
swap info struct ?

> -- Dave
> 

-- 
Mathieu Desnoyers
Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68