Re: [RFC][PATCH 0/2] Syscalls tracing

[Frederic Weisbecker <fweisbec@gmail.com>]

On Sun, Mar 08, 2009 at 12:24:41PM +0100, Frederic Weisbecker wrote:
> On Sat, Mar 07, 2009 at 05:02:47PM +0100, Frederic Weisbecker wrote:
> > On Sat, Mar 07, 2009 at 01:15:18PM +0100, Peter Zijlstra wrote:
> > > On Sat, 2009-03-07 at 05:52 +0100, Frederic Weisbecker wrote:
> > > > Here is a first attempt, quick one-shot, to provide a syscall tracing
> > > > infrastructure on ftrace.
> > > > 
> > > > The RFC prefix is here to reflect its ugliness on various parts.
> > > > The compromise between tracing reliabilty and speed is hard to balance.
> > > > For example I guess the basic and horrid string mask should be dropped in favour
> > > > of something else, which takes care of the volatile strings from the userspace.
> > > > 
> > > > But I hope a lot of ideas to make it better will come along this discussion.
> > > 
> > > Can't you abuse the SYSCALL_DEFINE macros? This current approach looks
> > > like it will replicate the syscall table.
> > > 
> > 
> > 
> > Ah, I did not even think about it.
> > I will be able to get the number of parameters. Sounds good. But I will
> > still need a way to store their format somewhere.
> >
> 
> Ok, we can iterate through sections datas for each one and then generate the format string
> depending of the types of the parameters. We can even to it once at boot time.
> The last thing is the need to match the exact syscall entry from this section when we enter
> a syscall. Don't know yet how I will do that but I will think about it.
> 

Ah and the string mask is actually useful to determine if we one parameter is a string.
If so we can decide to save it on tracing time or not.