Re: Three weeks to branching (gdb 7.5 release)

["H.J. Lu" <hjl.tools@gmail.com>]

On Sun, May 20, 2012 at 1:43 PM, Mark Kettenis <mark.kettenis@xs4all.nl> wrote:
>> Date: Sun, 20 May 2012 08:40:26 -0700
>> From: "H.J. Lu" <hjl.tools@gmail.com>
>>
>> On Fri, May 11, 2012 at 11:43 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
>> > On Fri, May 11, 2012 at 11:17 AM, Joel Brobecker <brobecker@adacore.com> wrote:
>> >> Hello,
>> >>
>> >> Just a quick heads up: The current tentative date for branching
>> >> GDB (7.5 release) is Mon Jun 4th, which is a little over three weeks
>> >> away.
>> >>
>> >> I've created a wiki page for known issues that need to be fixed
>> >> before then:
>> >>
>> >>    http://sourceware.org/gdb/wiki/GDB_7.5_Release
>> >>
>> >> When you add an issue, please make sure you add a name so we know
>> >> who is coordinating the effort.  If you don't know who can work
>> >> on it, please just post the issue here, and we'll try to find some
>> >> help.
>> >>
>> >> I only know of one issue, which is a noticeable performance degradation
>> >> that was reported a while ago:
>> >>
>> >
>> > I'd like to merge x32 into GDB 7.5.  My x32 change is on hjl/x32/master
>> > branch at
>> >
>> > http://sourceware.org/git/?p=gdb.git;a=summary
>> >
>> > The current diff only has 864 lines.  One patch:
>> >
>> > http://sourceware.org/ml/gdb-patches/2012-05/msg00097.html
>> >
>> > isn't reviewed yet.  I will open a meta bug for x32 integration.
>> >
>>
>> I opened:
>>
>> http://sourceware.org/bugzilla/show_bug.cgi?id=14099
>>
>> Thanks for help from everyone.  The full GDBserver x32 support
>> as well as partial GDB x32 support have been checked in.  The
>> remaining patches are:
>>
>> http://sourceware.org/ml/gdb-patches/2012-04/msg00195.html
>> http://sourceware.org/ml/gdb-patches/2012-04/msg00191.html
>> http://sourceware.org/ml/gdb-patches/2012-05/msg00744.html
>> http://sourceware.org/ml/gdb-patches/2012-05/msg00531.html
>> http://sourceware.org/ml/gdb-patches/2012-05/msg00533.html
>> http://sourceware.org/ml/gdb-patches/2012-05/msg00489.html
>> http://sourceware.org/ml/gdb-patches/2012-05/msg00438.html
>> http://sourceware.org/ml/gdb-patches/2012-05/msg00097.html
>>
>> I would appreciate help to get them reviewed and approved.
>
> As I wrote before, I don't think adding lots if if-statements is the
> proper way to add a new ABI to GDB.  The proper way is to do it like
> the diff below.  In that diff, I'm not entirely confident that calling
> amd64_linux_init_abi() from amd64_x32_linux_init_abi() makes all that
> much sense.  For example the amd64_linux_record_tdep stuff probably
> isn't right for the x32 ABI.  But at least this will give us a
> starting point where we won't end up adding
>
>  if (gdbarch_ptr_bit (gdbarch) == 32)
>    {
>      ...
>    }

Please take a look at

http://sourceware.org/ml/gdb-patches/2012-04/msg00195.html

It doesn't add any  (gdbarch_ptr_bit (gdbarch) == 32).  It just changes
it to bits_per_word.  I add one  "gdbarch_ptr_bit (gdbarch) == 32" in
amd64_linux_sigtramp_start and I will remove them from

http://sourceware.org/ml/gdb-patches/2012-05/msg00744.html

>  else
>    {
>    }
>
> blocks all over the place.
>
>
>
> Index: amd64-linux-tdep.c
> ===================================================================
> RCS file: /cvs/src/src/gdb/amd64-linux-tdep.c,v
> retrieving revision 1.50
> diff -u -p -r1.50 amd64-linux-tdep.c
> --- amd64-linux-tdep.c  12 May 2012 08:54:03 -0000      1.50
> +++ amd64-linux-tdep.c  20 May 2012 20:31:53 -0000
> @@ -1543,6 +1543,24 @@ amd64_linux_init_abi (struct gdbarch_inf
>
>   tdep->i386_syscall_record = amd64_linux_syscall_record;
>  }
> +
> +static void
> +amd64_x32_linux_init_abi(struct gdbarch_info info, struct gdbarch *gdbarch)
> +{
> +  struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
> +  const struct target_desc *tdesc = info.target_desc;
> +
> +  amd64_linux_init_abi (info, gdbarch);
> +  amd64_x32_init_abi (info, gdbarch);
> +
> +  if (! tdesc_has_registers (tdesc))
> +    tdesc = tdesc_amd64_linux;

I assume you meant tdesc_x32_linux here.  The problem is
when we reach here, if (! tdesc_has_registers (tdesc)) will always
be false since tdep->tdesc has been set by amd64_linux_init_abi.

> +  tdep->tdesc = tdesc;
> +
> +  /* GNU/Linux uses SVR4-style shared libraries.  */
> +  set_solib_svr4_fetch_link_map_offsets
> +    (gdbarch, svr4_ilp32_fetch_link_map_offsets);
> +}
>
>
>  /* Provide a prototype to silence -Wmissing-prototypes.  */
> @@ -1553,6 +1571,8 @@ _initialize_amd64_linux_tdep (void)
>  {
>   gdbarch_register_osabi (bfd_arch_i386, bfd_mach_x86_64,
>                          GDB_OSABI_LINUX, amd64_linux_init_abi);
> +  gdbarch_register_osabi (bfd_arch_i386, bfd_mach_x64_32,
> +                         GDB_OSABI_LINUX, amd64_x32_linux_init_abi);
>
>   /* Initialize the Linux target description.  */
>   initialize_tdesc_amd64_linux ();
> Index: amd64-tdep.c
> ===================================================================
> RCS file: /cvs/src/src/gdb/amd64-tdep.c,v
> retrieving revision 1.104
> diff -u -p -r1.104 amd64-tdep.c
> --- amd64-tdep.c        14 May 2012 18:56:40 -0000      1.104
> +++ amd64-tdep.c        20 May 2012 20:31:54 -0000
> @@ -258,7 +258,8 @@ static const char *amd64_word_names[] =
>  static const char *amd64_dword_names[] =
>  {
>   "eax", "ebx", "ecx", "edx", "esi", "edi", "ebp", "esp",
> -  "r8d", "r9d", "r10d", "r11d", "r12d", "r13d", "r14d", "r15d"
> +  "r8d", "r9d", "r10d", "r11d", "r12d", "r13d", "r14d", "r15d",
> +  "eip"
>  };
>
>  /* Return the name of register REGNUM.  */
> @@ -2729,6 +2730,43 @@ amd64_init_abi (struct gdbarch_info info
>   set_gdbarch_stap_parse_special_token (gdbarch,
>                                        i386_stap_parse_special_token);
>  }
> +
> +
> +static struct type *
> +amd64_x32_pseudo_register_type (struct gdbarch *gdbarch, int regnum)
> +{
> +  struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
> +
> +  switch (regnum - tdep->eax_regnum)
> +    {
> +    case AMD64_RBP_REGNUM:     /* %ebp */
> +    case AMD64_RSP_REGNUM:     /* %esp */
> +      return builtin_type (gdbarch)->builtin_data_ptr;
> +    case AMD64_RIP_REGNUM:     /* %eip */
> +      return builtin_type (gdbarch)->builtin_func_ptr;
> +    }
> +
> +  return i386_pseudo_register_type (gdbarch, regnum);
> +}
> +
> +void
> +amd64_x32_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch)
> +{
> +  struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
> +  const struct target_desc *tdesc = info.target_desc;
> +
> +  amd64_init_abi (info, gdbarch);
> +
> +  if (! tdesc_has_registers (tdesc))
> +    tdesc = tdesc_x32;

Again, " if (! tdesc_has_registers (tdesc))" will always false
since tdep->tdesc has been set in amd64_init_abi.  How
do we solve it?  My suggestion is to add a new function
which is similar to amd64_init_abi, but takes a new argument,
const struct target_desc *, as the default tdesc. Both
amd64_init_abi and amd64_x32_init_abi will call this
function.  Will it work for you?

Thanks.


-- 
H.J.