Re: [RFC][PATCH 2/2] tracing/x86: basic implementation of syscall tracing for x86-64

[Frederic Weisbecker <fweisbec@gmail.com>]

On Sun, Mar 15, 2009 at 05:44:05AM +0100, Ingo Molnar wrote:
> 
> * Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca> wrote:
> 
> > * Ingo Molnar (mingo@elte.hu) wrote:
> > > 
> > > * Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca> wrote:
> > > 
> > > > * Frederic Weisbecker (fweisbec@gmail.com) wrote:
> > > >
> > > > > Provide the ptrace hooks and arch specific syscall numbers 
> > > > > to ftrace arch indepedant syscall numbers. For now it only 
> > > > > supports 4 syscalls to provide an example.
> > > > > 
> > > > 
> > > > Hi Frederic,
> > > > 
> > > > I already have the equivalent TIF_KERNEL_TRACE flag in my 
> > > > LTTng tree added to every Linux architecture. You might want 
> > > > to re-use this work rather than re-doing this. I don't mind 
> > > > changing the flag name.
> > > 
> > > Yeah. Note that the TIF bits are just one part - there are other 
> > > bits needed for HAVE_FTRACE_SYSCALLS arch support.
> > > 
> > > Also, i'd eventually expect the TIF bits to be converted to a 
> > > tracehook callback, not spread it to other architectures.
> > > 
> > 
> > The nice part about the TIF bit is that it permits adding this 
> > syscall tracing feature without increasing the size of the 
> > thread_info struct nor adding any extra tests in entry.S. Does 
> > the tracehook callback have these features ?
> 
> yes. Tracehook just factors out common interfacing points - with 
> one specific implementation for now: ptrace. For syscall tracing 
> that means it wraps TIF_SYSCALL_TRACE in essence.
> 
> 	Ingo


Since I didn't want to experience conflicts against ptrace, I haven't
played with the tracehooks.

But anyway, I wrote this low level part of syscall tracing having in mind
the fact that utrace does this work very much better. That's why I wrote it
to be very basic and simple enough to do the job, but waiting for a better
code that is already written somewhere else.

So now there is some code which does that on both utrace and Lttng.

I would be pleased to see these patches which handle these flags properly
on LKML :-)

Frederic.