Re: [PATCH v4 0/4] tracing: tweaks for generic syscall events

[Frederic Weisbecker]

On Tue, Aug 25, 2009 at 10:41:22AM -0400, Jason Baron wrote:
> right, the compat layer as well as the core kernel need additional
> DEFINE_SYSCALL() macros to catch all the the syscalls. However, I think
> the compat layer is a bit more involved in that it often makes use of
> the core kernel syscalls, but the mapping between syscall number is
> different. So, we need another array, or to augment the existing one, to
> cover the compat syscalls. We also need to detect 32-bit processes in
> the syscall entry path to determine which array to use, and we need to
> grab the arguments differently. So there is a bunch of work here.


Hmm, indeed we may need a parallel compat_syscall_table for the metadata
and other kind of special treatements.
I wonder how ptrace sort it out in syscall_trace_enter() to guess the
origin of the syscall (ia32 or normal), since the table is not mapped
with the same numbers.

 
> Also, we have the question of whether we need separate entries in the
> events/syscalls directly for 32-bit process syscalls that call the same
> 64-bit syscall interfaces. Should they be parsed as 64-bit argument
> values event thought they are 32-bit? To reduce complexity, I would say
> the 32-bit syscall entries should be the same as the 64-bit ones. That
> said there will be a bunch of new "compat_sys*" etc. entries.
> 


We could probably re-route the compat syscall tracing to their
homologuous 64 bits tracepoints, but one may want to only trace the
compat_syscalls, use filters only on them, activate only some of them,
etc...

To do such routing, we could have a simple table that resolves
a compat syscall number to its real 64 bits syscall number and we could
then pass this number plus a flag that set its compat state in the ring
buffer entry.
That would avoid the need of using the DEFINE_SYSCALLx() in fs/compat.c

But such ghost tracepoints would also complexify too much the filter
processing, the individual tracepoints toggling, etc...
So I think having real compat tracepoints would actually be more simple.

We could still use shortcuts in userspace if we want to enable
sys_enter_open and sys_enter_compat_open at the same time.

Hmm?

> > Is someone willing to cover them?
> > 
> 
> I can take a stab at it.


Thanks a lot!


> thanks,
> 
> -Jason

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/