This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [PATCH 1/4] catch syscall -- try 4 -- Architecture-independent part
- From: Sérgio Durigan Júnior <sergiodj at linux dot vnet dot ibm dot com>
- To: Daniel Jacobowitz <drow at false dot org>
- Cc: Pedro Alves <pedro at codesourcery dot com>, gdb-patches at sourceware dot org
- Date: Sun, 08 Mar 2009 16:15:58 -0300
- Subject: Re: [PATCH 1/4] catch syscall -- try 4 -- Architecture-independent part
- References: <1232929831.26873.22.camel@miki> <200901260053.06295.pedro@codesourcery.com> <1232945747.26873.27.camel@miki> <1232989355.26873.39.camel@miki> <20090201193306.GJ4597@caradoc.them.org> <1235561189.14363.20.camel@miki> <20090227221133.GA12904@caradoc.them.org>
Hi Daniel,
Thanks for your comments.
On Fri, 2009-02-27 at 17:11 -0500, Daniel Jacobowitz wrote:
> > > > +/* Implement the "print_one" breakpoint_ops method for syscall
> > > > + catchpoints. */
> > > > +
> > > > +static void
> > > > +print_one_catch_syscall (struct breakpoint *b, CORE_ADDR *last_addr)
> > > > +{
> > >
> > > Have you tried hitting a syscall catchpoint in MI mode, and is the
> > > output anything useful?
> >
> > No, unfortunately I haven't. Actually, I must first learn how to use the
> > MI interface, but that should not be hard :-).
>
> I'd suggest doing that as part of this submission so that we know
> you're on the right track. It isn't too hard; you can start by
> looking at the test logs from gdb.mi tests, if that helps.
All right, I'll do that today.
> > > > +# Fills the struct syscall (passed as argument) with the corresponding
> > > > +# system call represented by syscall_number.
> > > > +M:void:get_syscall_by_number:int syscall_number, struct syscall *s:syscall_number, s
> > > > +
> > > > +# Fills the struct syscall (passed as argument) with the corresponding
> > > > +# system call represented by syscall_name.
> > > > +M:void:get_syscall_by_name:const char *syscall_name, struct syscall *s:syscall_name, s
> > > > +
> > > > +# Returns the array containing the syscall names for the architecture.
> > > > +M:const char **:get_syscall_names:void:
> > >
> > > If every target is going to use XML for this, these three do not need
> > > to be gdbarch methods and the support code can move from linux-tdep.c
> > > to xml-syscall.c.
> >
> > As far as I understood (from our discussion a few months ago), not every
> > target is supposed to use the XML for syscalls. That's specially true
> > for embedded systems and/or architectures for which the XML file is
> > missing (for some obscure reason, don't know). That's why I thought it
> > would be better not to generalize.
>
> I don't think this is a big deal. If it is, we can handle it the same
> way as for target-descriptions: pre-compile them into GDB.
So I won't modify anything, ok?
> > > > + if (target_passed_by_entrypoint () > 0
> > > > + && catch_syscall_enabled () > 0)
> > > > + request = PT_SYSCALL;
> > > > + else
> > > > + request = PT_CONTINUE;
> > >
> > > Why is target_passed_by_entrypoint still necessary? If we understand
> > > why, I think there'll be some other more appropriate flag. Is it to
> > > avoid using PTRACE_SYSCALL when the shell is running, before the
> > > application starts?
> >
> > It's been a long time since I added this check, but as far as I
> > remember, that's exactly the reason. I tried to remove this, and the
> > testcase simply freezes. Do you have another idea? :-)
>
> Not sure that the flag exists any more, but you're trying to avoid it
> when called by startup_inferior. I suppose you could use the
> inferior_created observer (not new_inferior! The distinction is not
> too clear in the manual but that one is too early). The problem is,
> again, that this flag needs to be per-inferior.
>
> Pedro, any thoughts?
What do you mean by "Not sure that the flag exists any more"? Also, I'm
waiting for Pedro's reply.
> > > > diff --git a/gdb/linux-nat.c b/gdb/linux-nat.c
> > > > index 9a7e39c..1d0f66f 100644
> > > > --- a/gdb/linux-nat.c
> > > > +++ b/gdb/linux-nat.c
> > > > @@ -676,6 +676,7 @@ linux_child_post_attach (int pid)
> > > > {
> > > > linux_enable_event_reporting (pid_to_ptid (pid));
> > > > check_for_thread_db ();
> > > > + linux_enable_tracesysgood (pid_to_ptid (pid));
> > > > }
> > > >
> > > > static void
> > > > @@ -683,6 +684,7 @@ linux_child_post_startup_inferior (ptid_t ptid)
> > > > {
> > > > linux_enable_event_reporting (ptid);
> > > > check_for_thread_db ();
> > > > + linux_enable_tracesysgood (ptid);
> > > > }
> > > >
> > > > static int
> > > > @@ -4160,6 +4162,7 @@ linux_target_install_ops (struct target_ops *t)
> > > > t->to_follow_fork = linux_child_follow_fork;
> > > > t->to_find_memory_regions = linux_nat_find_memory_regions;
> > > > t->to_make_corefile_notes = linux_nat_make_corefile_notes;
> > > > + t->to_passed_by_entrypoint = linux_passed_by_entrypoint;
> > > >
> > > > super_xfer_partial = t->to_xfer_partial;
> > > > t->to_xfer_partial = linux_xfer_partial;
> > >
> > > These bits must be for another patch in the series :-)
> >
> > I'm sorry, I didn't understand what you meant by that :-(. These
> > modifications are all architecture-independent, so this is the right
> > place for them right?
>
> No - since they're specific to Linux. Also, I don't think they'll
> compile at this point, you haven't added the function yet.
Well, I know it's a shame, but the patches for catch syscall don't
compile alone at all. You must have all four patches in order to get
things working, and I haven't even tested if the patches can be compiled
independently. However, you're right regarding this piece of code: it
should be placed in other patch.
Regards,
--
Sérgio Durigan Júnior
Linux on Power Toolchain - Software Engineer
Linux Technology Center - LTC
IBM Brazil