This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [RFA] cleanup of syscall consts in process record
> Date: Tue, 08 Sep 2009 12:06:27 -0700
> From: Michael Snyder <msnyder@vmware.com>
>
> Joel Brobecker wrote:
> >> 2009-09-08 Michael Snyder <msnyder@vmware.com>
> >>
> >> * amd64-linux-tdep.h (enum amd64_syscall): New enum consts,
> >> to replace literal consts used in amd64-linux-tdep.c
> >> * linux-record.h (enum gdb_syscall): New enum consts, to replace
> >> literal consts used in amd64-linux-tdep.c and linux-record.c.
> >> * amd64-linux-tdep.c (amd64_canonicalize_syscall): New function,
> >> translate from native amd64 Linux syscall id to internal gdb id.
> >> (amd64_linux_syscall_record): Switch statement abstracted out
> >> and replaced with a call to amd64_canonicalize_syscall.
> >> * linux-record.c (record_linux_system_call): Replace literal
> >> consts with enum consts.
> >> * i386-linux-tdep.c (i386_canonicalize_syscall): New function,
> >> trivially translate from native i386 Linux syscalls to gdb syscalls.
> >> (i386_linux_intx80_sysenter_record):
> >
> > Nice! I really like this version much better. The approach you took
> > with i386 made me wonder whether we really need the amd64_syscall enum
> > at all - we could have used a plain int as the argument to
> > amd64_canonicalize_syscall, and use plain numbers there, rather than
> > having an enum that's only used there. I don't mind, though, so
> > don't worry about it unless you agree as well.
>
> Nah, the idea was to get rid of magic numbers,
> plus it makes the code more readable. Self-documenting.
>
>
> > Note that this should also fix the issue that Hui reported about
> > building on cygwin with --enable-64-bit-bfd. So I'll remove Hui's
> > patch from my list.
>
> Yes, I broke down and included Hui's patch in this one.
>
> > Just one comment:
> >
> >> +static enum gdb_syscall
> >> +i386_canonicalize_syscall (int syscall)
> >> +{
> >> + enum { i386_syscall_max = 499 };
> >> +
> >> + if (syscall <= i386_syscall_max)
> >> + return syscall;
> >
> > I thought that we should incorporate Mark's suggestion of checking
> > syscall against negative values. But I now realize that if syscall
> > is negative, we'll return a value that's equivalent to returning -1.
> > And the check against negative values in i386_linux_intx80_sysenter_record
> > should then catch it.
>
> Correct.
>
> So this is approved, then?
Fine with me.