Use elfcore_write_linux_prstatus in gdb

Use elfcore_write_linux_prstatus in gdb

[Alan Modra]

The gdb side.  Nothing fancy here, and no support for targets that
need gdbarch_elfcore_write_linux_prstatus, mips n32, sparc64 if
writing any of pr_{u,s,cu,cs}time, and probably x86-64 x32.

diff --git a/gdb/gdbarch.sh b/gdb/gdbarch.sh
index 4ac6b90..981f8db 100755
--- a/gdb/gdbarch.sh
+++ b/gdb/gdbarch.sh
@@ -704,6 +704,9 @@ M:char *:make_corefile_notes:bfd *obfd, int *note_size:obfd, note_size
 # default.
 F:char *:elfcore_write_linux_prpsinfo:bfd *obfd, char *note_data, int *note_size, const struct elf_internal_linux_prpsinfo *info:obfd, note_data, note_size, info
 
+# prstatus, as for prpsinfo
+F:char *:elfcore_write_linux_prstatus:bfd *obfd, char *note_data, int *note_size, const struct elf_internal_linux_prstatus *info:obfd, note_data, note_size, info
+
 # Find core file memory regions
 M:int:find_memory_regions:find_memory_region_ftype func, void *data:func, data
 
@@ -1248,6 +1251,7 @@ struct stap_parse_info;
 struct parser_state;
 struct ravenscar_arch_ops;
 struct elf_internal_linux_prpsinfo;
+struct elf_internal_linux_prstatus;
 struct mem_range;
 struct syscalls_info;
 struct thread_info;
diff --git a/gdb/linux-tdep.c b/gdb/linux-tdep.c
index 29f5f3c..03a273e 100644
--- a/gdb/linux-tdep.c
+++ b/gdb/linux-tdep.c
@@ -1564,9 +1564,31 @@ linux_collect_regset_section_cb (const char *sect_name, int size,
 
   /* PRSTATUS still needs to be treated specially.  */
   if (strcmp (sect_name, ".reg") == 0)
-    data->note_data = (char *) elfcore_write_prstatus
-      (data->obfd, data->note_data, data->note_size, data->lwp,
-       gdb_signal_to_host (data->stop_signal), buf);
+    {
+      struct elf_internal_linux_prstatus prstatus;
+
+      memset (&prstatus, 0, sizeof (prstatus));
+      prstatus.pr_reg = buf;
+      prstatus.pr_reg_size = size;
+      prstatus.pr_pid = data->lwp;
+      prstatus.pr_cursig = gdb_signal_to_host (data->stop_signal);
+      if (gdbarch_elfcore_write_linux_prstatus_p (data->gdbarch))
+	data->note_data = gdbarch_elfcore_write_linux_prstatus (data->gdbarch,
+								data->obfd,
+								data->note_data,
+								data->note_size,
+								&prstatus);
+      else if (gdbarch_ptr_bit (data->gdbarch) == 64)
+	data->note_data = elfcore_write_linux_prstatus64 (data->obfd,
+							  data->note_data,
+							  data->note_size,
+							  &prstatus);
+      else
+	data->note_data = elfcore_write_linux_prstatus32 (data->obfd,
+							  data->note_data,
+							  data->note_size,
+							  &prstatus);
+    }
   else
     data->note_data = (char *) elfcore_write_register_note
       (data->obfd, data->note_data, data->note_size,


-- 
Alan Modra
Australia Development Lab, IBM

Re: Use elfcore_write_linux_prstatus in gdb

[Pedro Alves]

On 01/21/2016 09:28 AM, Alan Modra wrote:
> The gdb side.  Nothing fancy here, and no support for targets that
> need gdbarch_elfcore_write_linux_prstatus, mips n32, sparc64 if
> writing any of pr_{u,s,cu,cs}time, and probably x86-64 x32.

Thanks a lot for doing all this!  I don't have spare cycles either to
drive this all the way to completion for all archs, I think we can still
move forward with it, by adding a stop-gap implementation those
archs, which just calls elfcore_write_prstatus.  I'll send it along
in a bit.

Thanks,
Pedro Alves

Re: Use elfcore_write_linux_prstatus in gdb

[Pedro Alves]

On 01/21/2016 12:25 PM, Pedro Alves wrote:
> On 01/21/2016 09:28 AM, Alan Modra wrote:
>> The gdb side.  Nothing fancy here, and no support for targets that
>> need gdbarch_elfcore_write_linux_prstatus, mips n32, sparc64 if
>> writing any of pr_{u,s,cu,cs}time, and probably x86-64 x32.
> 
> Thanks a lot for doing all this!  I don't have spare cycles either to
> drive this all the way to completion for all archs, I think we can still
> move forward with it, by adding a stop-gap implementation those
> archs, which just calls elfcore_write_prstatus.  I'll send it along
> in a bit.

Like this.  I've pushed this to the users/palves/core-prstatus-rework
branch.  I've tested on x86_64 with:

 make check RUNTESTFLAGS="--target_board unix/-m32" TESTS="*/*core*.exp"
 make check RUNTESTFLAGS="--target_board unix/-m64" TESTS="*/*core*.exp"

and it seems to work fine.  Also did the same after installing the
stop-gap hook on x86-64, to make sure that works.

From 341d977d0d79d2b07e5a4d41e5be2f6977033fb3 Mon Sep 17 00:00:00 2001
From: Alan Modra <amodra@gmail.com>
Date: Thu, 21 Jan 2016 19:58:55 +1030
Subject: [PATCH] Use elfcore_write_linux_prstatus in gdb

The gdb side.  Nothing fancy here, though it'd be easier to fill in
more fields now.  Targets that are known to need a custom
gdbarch_elfcore_write_linux_prstatus but don't have one yet can
install a stop-gap hook implementation that still calls into
elfcore_write_prstatus.  It is currently installed for MIPS n32 and
Sparc64 (though we're not writting any of pr_{u,s,cu,cs}time yet).
Might be needed for x86-64 x32 too.

Signed-off-by: Pedro Alves <palves@redhat.com>
---
 gdb/gdbarch.c            | 32 ++++++++++++++++++++++++++++++++
 gdb/gdbarch.h            | 15 +++++++++++----
 gdb/gdbarch.sh           | 10 ++++++----
 gdb/linux-tdep.c         | 43 ++++++++++++++++++++++++++++++++++++++++---
 gdb/linux-tdep.h         |  8 ++++++++
 gdb/mips-linux-tdep.c    |  5 +++++
 gdb/sparc64-linux-tdep.c |  5 +++++
 7 files changed, 107 insertions(+), 11 deletions(-)

diff --git a/gdb/gdbarch.c b/gdb/gdbarch.c
index 4143744..fe0acbc 100644
--- a/gdb/gdbarch.c
+++ b/gdb/gdbarch.c
@@ -265,6 +265,7 @@ struct gdbarch
   gdbarch_iterate_over_regset_sections_ftype *iterate_over_regset_sections;
   gdbarch_make_corefile_notes_ftype *make_corefile_notes;
   gdbarch_elfcore_write_linux_prpsinfo_ftype *elfcore_write_linux_prpsinfo;
+  gdbarch_elfcore_write_linux_prstatus_ftype *elfcore_write_linux_prstatus;
   gdbarch_find_memory_regions_ftype *find_memory_regions;
   gdbarch_core_xfer_shared_libraries_ftype *core_xfer_shared_libraries;
   gdbarch_core_xfer_shared_libraries_aix_ftype *core_xfer_shared_libraries_aix;
@@ -609,6 +610,7 @@ verify_gdbarch (struct gdbarch *gdbarch)
   /* Skip verify of iterate_over_regset_sections, has predicate.  */
   /* Skip verify of make_corefile_notes, has predicate.  */
   /* Skip verify of elfcore_write_linux_prpsinfo, has predicate.  */
+  /* Skip verify of elfcore_write_linux_prstatus, has predicate.  */
   /* Skip verify of find_memory_regions, has predicate.  */
   /* Skip verify of core_xfer_shared_libraries, has predicate.  */
   /* Skip verify of core_xfer_shared_libraries_aix, has predicate.  */
@@ -931,6 +933,12 @@ gdbarch_dump (struct gdbarch *gdbarch, struct ui_file *file)
                       "gdbarch_dump: elfcore_write_linux_prpsinfo = <%s>\n",
                       host_address_to_string (gdbarch->elfcore_write_linux_prpsinfo));
   fprintf_unfiltered (file,
+                      "gdbarch_dump: gdbarch_elfcore_write_linux_prstatus_p() = %d\n",
+                      gdbarch_elfcore_write_linux_prstatus_p (gdbarch));
+  fprintf_unfiltered (file,
+                      "gdbarch_dump: elfcore_write_linux_prstatus = <%s>\n",
+                      host_address_to_string (gdbarch->elfcore_write_linux_prstatus));
+  fprintf_unfiltered (file,
                       "gdbarch_dump: fast_tracepoint_valid_at = <%s>\n",
                       host_address_to_string (gdbarch->fast_tracepoint_valid_at));
   fprintf_unfiltered (file,
@@ -3466,6 +3474,30 @@ set_gdbarch_elfcore_write_linux_prpsinfo (struct gdbarch *gdbarch,
 }
 
 int
+gdbarch_elfcore_write_linux_prstatus_p (struct gdbarch *gdbarch)
+{
+  gdb_assert (gdbarch != NULL);
+  return gdbarch->elfcore_write_linux_prstatus != NULL;
+}
+
+char *
+gdbarch_elfcore_write_linux_prstatus (struct gdbarch *gdbarch, bfd *obfd, char *note_data, int *note_size, const struct elf_internal_linux_prstatus *info)
+{
+  gdb_assert (gdbarch != NULL);
+  gdb_assert (gdbarch->elfcore_write_linux_prstatus != NULL);
+  if (gdbarch_debug >= 2)
+    fprintf_unfiltered (gdb_stdlog, "gdbarch_elfcore_write_linux_prstatus called\n");
+  return gdbarch->elfcore_write_linux_prstatus (obfd, note_data, note_size, info);
+}
+
+void
+set_gdbarch_elfcore_write_linux_prstatus (struct gdbarch *gdbarch,
+                                          gdbarch_elfcore_write_linux_prstatus_ftype elfcore_write_linux_prstatus)
+{
+  gdbarch->elfcore_write_linux_prstatus = elfcore_write_linux_prstatus;
+}
+
+int
 gdbarch_find_memory_regions_p (struct gdbarch *gdbarch)
 {
   gdb_assert (gdbarch != NULL);
diff --git a/gdb/gdbarch.h b/gdb/gdbarch.h
index 3fadcd1..5e00b32 100644
--- a/gdb/gdbarch.h
+++ b/gdb/gdbarch.h
@@ -61,6 +61,7 @@ struct stap_parse_info;
 struct parser_state;
 struct ravenscar_arch_ops;
 struct elf_internal_linux_prpsinfo;
+struct elf_internal_linux_prstatus;
 struct mem_range;
 struct syscalls_info;
 struct thread_info;
@@ -818,10 +819,10 @@ typedef char * (gdbarch_make_corefile_notes_ftype) (struct gdbarch *gdbarch, bfd
 extern char * gdbarch_make_corefile_notes (struct gdbarch *gdbarch, bfd *obfd, int *note_size);
 extern void set_gdbarch_make_corefile_notes (struct gdbarch *gdbarch, gdbarch_make_corefile_notes_ftype *make_corefile_notes);
 
-/* The elfcore writer hook to use to write Linux prpsinfo notes to core
-   files.  Most Linux architectures use the same prpsinfo32 or
-   prpsinfo64 layouts, and so won't need to provide this hook, as we
-   call the Linux generic routines in bfd to write prpsinfo notes by
+/* The elfcore writer hooks to use to write Linux notes to core files.
+   Most Linux architectures use the same prpsinfo{32,64} and
+   prstatus{32,64} layouts, and so won't need to provide these hooks,
+   as we call the Linux generic routines in bfd to write notes by
    default. */
 
 extern int gdbarch_elfcore_write_linux_prpsinfo_p (struct gdbarch *gdbarch);
@@ -830,6 +831,12 @@ typedef char * (gdbarch_elfcore_write_linux_prpsinfo_ftype) (bfd *obfd, char *no
 extern char * gdbarch_elfcore_write_linux_prpsinfo (struct gdbarch *gdbarch, bfd *obfd, char *note_data, int *note_size, const struct elf_internal_linux_prpsinfo *info);
 extern void set_gdbarch_elfcore_write_linux_prpsinfo (struct gdbarch *gdbarch, gdbarch_elfcore_write_linux_prpsinfo_ftype *elfcore_write_linux_prpsinfo);
 
+extern int gdbarch_elfcore_write_linux_prstatus_p (struct gdbarch *gdbarch);
+
+typedef char * (gdbarch_elfcore_write_linux_prstatus_ftype) (bfd *obfd, char *note_data, int *note_size, const struct elf_internal_linux_prstatus *info);
+extern char * gdbarch_elfcore_write_linux_prstatus (struct gdbarch *gdbarch, bfd *obfd, char *note_data, int *note_size, const struct elf_internal_linux_prstatus *info);
+extern void set_gdbarch_elfcore_write_linux_prstatus (struct gdbarch *gdbarch, gdbarch_elfcore_write_linux_prstatus_ftype *elfcore_write_linux_prstatus);
+
 /* Find core file memory regions */
 
 extern int gdbarch_find_memory_regions_p (struct gdbarch *gdbarch);
diff --git a/gdb/gdbarch.sh b/gdb/gdbarch.sh
index 4ac6b90..71e3e8c 100755
--- a/gdb/gdbarch.sh
+++ b/gdb/gdbarch.sh
@@ -697,12 +697,13 @@ M:void:iterate_over_regset_sections:iterate_over_regset_sections_cb *cb, void *c
 # Create core file notes
 M:char *:make_corefile_notes:bfd *obfd, int *note_size:obfd, note_size
 
-# The elfcore writer hook to use to write Linux prpsinfo notes to core
-# files.  Most Linux architectures use the same prpsinfo32 or
-# prpsinfo64 layouts, and so won't need to provide this hook, as we
-# call the Linux generic routines in bfd to write prpsinfo notes by
+# The elfcore writer hooks to use to write Linux notes to core files.
+# Most Linux architectures use the same prpsinfo{32,64} and
+# prstatus{32,64} layouts, and so won't need to provide these hooks,
+# as we call the Linux generic routines in bfd to write notes by
 # default.
 F:char *:elfcore_write_linux_prpsinfo:bfd *obfd, char *note_data, int *note_size, const struct elf_internal_linux_prpsinfo *info:obfd, note_data, note_size, info
+F:char *:elfcore_write_linux_prstatus:bfd *obfd, char *note_data, int *note_size, const struct elf_internal_linux_prstatus *info:obfd, note_data, note_size, info
 
 # Find core file memory regions
 M:int:find_memory_regions:find_memory_region_ftype func, void *data:func, data
@@ -1248,6 +1249,7 @@ struct stap_parse_info;
 struct parser_state;
 struct ravenscar_arch_ops;
 struct elf_internal_linux_prpsinfo;
+struct elf_internal_linux_prstatus;
 struct mem_range;
 struct syscalls_info;
 struct thread_info;
diff --git a/gdb/linux-tdep.c b/gdb/linux-tdep.c
index 29f5f3c..d432c47 100644
--- a/gdb/linux-tdep.c
+++ b/gdb/linux-tdep.c
@@ -1526,6 +1526,21 @@ linux_make_mappings_corefile_notes (struct gdbarch *gdbarch, bfd *obfd,
   return note_data;
 }
 
+/* See linux-tdep.h.
+
+   FIXME: This should be reimplemented on the BFD side along the lines
+   of elfcore_write_linux_prstatus{32,64}, instead of relying on
+   elfcore_write_prstatus.  */
+
+char *
+gdb_deprecated_elfcore_write_linux_prstatus
+  (bfd *obfd, char *note_data, int *note_size,
+   const struct elf_internal_linux_prstatus *info)
+{
+  return elfcore_write_prstatus (obfd, note_data, note_size, info->pr_pid,
+				 info->pr_cursig, info->pr_reg);
+}
+
 /* Structure for passing information from
    linux_collect_thread_registers via an iterator to
    linux_collect_regset_section_cb. */
@@ -1564,9 +1579,31 @@ linux_collect_regset_section_cb (const char *sect_name, int size,
 
   /* PRSTATUS still needs to be treated specially.  */
   if (strcmp (sect_name, ".reg") == 0)
-    data->note_data = (char *) elfcore_write_prstatus
-      (data->obfd, data->note_data, data->note_size, data->lwp,
-       gdb_signal_to_host (data->stop_signal), buf);
+    {
+      struct elf_internal_linux_prstatus prstatus;
+
+      memset (&prstatus, 0, sizeof (prstatus));
+      prstatus.pr_reg = buf;
+      prstatus.pr_reg_size = size;
+      prstatus.pr_pid = data->lwp;
+      prstatus.pr_cursig = gdb_signal_to_host (data->stop_signal);
+      if (gdbarch_elfcore_write_linux_prstatus_p (data->gdbarch))
+	data->note_data = gdbarch_elfcore_write_linux_prstatus (data->gdbarch,
+								data->obfd,
+								data->note_data,
+								data->note_size,
+								&prstatus);
+      else if (gdbarch_ptr_bit (data->gdbarch) == 64)
+	data->note_data = elfcore_write_linux_prstatus64 (data->obfd,
+							  data->note_data,
+							  data->note_size,
+							  &prstatus);
+      else
+	data->note_data = elfcore_write_linux_prstatus32 (data->obfd,
+							  data->note_data,
+							  data->note_size,
+							  &prstatus);
+    }
   else
     data->note_data = (char *) elfcore_write_register_note
       (data->obfd, data->note_data, data->note_size,
diff --git a/gdb/linux-tdep.h b/gdb/linux-tdep.h
index 328c7f0..ae50b14 100644
--- a/gdb/linux-tdep.h
+++ b/gdb/linux-tdep.h
@@ -44,4 +44,12 @@ extern void linux_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch);
 
 extern int linux_is_uclinux (void);
 
+/* Stop-gap gdbarch_elfcore_write_linux_prstatus implementation for
+   architectures whose prstatus{32,64} layout differs from the
+   standard one, and for which BFD doesn't provide a custom writer
+   yet.  */
+extern char *gdb_deprecated_elfcore_write_linux_prstatus
+  (bfd *obfd, char *note_data, int *note_size,
+   const struct elf_internal_linux_prstatus *info);
+
 #endif /* linux-tdep.h */
diff --git a/gdb/mips-linux-tdep.c b/gdb/mips-linux-tdep.c
index 8dc0566..b2fcf23 100644
--- a/gdb/mips-linux-tdep.c
+++ b/gdb/mips-linux-tdep.c
@@ -1683,6 +1683,11 @@ mips_linux_init_abi (struct gdbarch_info info,
 				      &micromips_linux_n32_rt_sigframe);
 	tramp_frame_prepend_unwinder (gdbarch, &mips_linux_n32_rt_sigframe);
 	set_xml_syscall_file_name (gdbarch, "syscalls/mips-n32-linux.xml");
+
+	/* N32 uses a different prstatus compared to most other Linux
+	   archs.  */
+	set_gdbarch_elfcore_write_linux_prstatus
+	  (gdbarch, gdb_deprecated_elfcore_write_linux_prstatus);
 	break;
       case MIPS_ABI_N64:
 	set_gdbarch_get_longjmp_target (gdbarch,
diff --git a/gdb/sparc64-linux-tdep.c b/gdb/sparc64-linux-tdep.c
index 1336d73..3a50900 100644
--- a/gdb/sparc64-linux-tdep.c
+++ b/gdb/sparc64-linux-tdep.c
@@ -307,6 +307,11 @@ sparc64_linux_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch)
   tdep->fpregset = &sparc64_linux_fpregset;
   tdep->sizeof_fpregset = 280;
 
+  /* Sparc64 uses a different prstatus compared to most other Linux
+     archs.  */
+  set_gdbarch_elfcore_write_linux_prstatus
+    (gdbarch, gdb_deprecated_elfcore_write_linux_prstatus);
+
   tramp_frame_prepend_unwinder (gdbarch, &sparc64_linux_rt_sigframe);
 
   /* Hook in the DWARF CFI frame unwinder.  */
-- 
1.9.3

Re: Use elfcore_write_linux_prstatus in gdb

[H.J. Lu]

On Thu, Jan 21, 2016 at 5:11 AM, Pedro Alves <palves@redhat.com> wrote:
> On 01/21/2016 12:25 PM, Pedro Alves wrote:
>> On 01/21/2016 09:28 AM, Alan Modra wrote:
>>> The gdb side.  Nothing fancy here, and no support for targets that
>>> need gdbarch_elfcore_write_linux_prstatus, mips n32, sparc64 if
>>> writing any of pr_{u,s,cu,cs}time, and probably x86-64 x32.
>>
>> Thanks a lot for doing all this!  I don't have spare cycles either to
>> drive this all the way to completion for all archs, I think we can still
>> move forward with it, by adding a stop-gap implementation those
>> archs, which just calls elfcore_write_prstatus.  I'll send it along
>> in a bit.
>
> Like this.  I've pushed this to the users/palves/core-prstatus-rework
> branch.  I've tested on x86_64 with:
>
>  make check RUNTESTFLAGS="--target_board unix/-m32" TESTS="*/*core*.exp"
>  make check RUNTESTFLAGS="--target_board unix/-m64" TESTS="*/*core*.exp"

It failed

 make check RUNTESTFLAGS="--target_board unix/-mx32" TESTS="*/*core*.exp"

Please take a look at bfd/hosts/x86-64linux.h to see how to dump core
x86-64.

Thanks.


H.J.

Re: Use elfcore_write_linux_prstatus in gdb

[Pedro Alves]

On 01/21/2016 01:24 PM, H.J. Lu wrote:
> On Thu, Jan 21, 2016 at 5:11 AM, Pedro Alves <palves@redhat.com> wrote:
>> On 01/21/2016 12:25 PM, Pedro Alves wrote:
>>> On 01/21/2016 09:28 AM, Alan Modra wrote:
>>>> The gdb side.  Nothing fancy here, and no support for targets that
>>>> need gdbarch_elfcore_write_linux_prstatus, mips n32, sparc64 if
>>>> writing any of pr_{u,s,cu,cs}time, and probably x86-64 x32.
>>>
>>> Thanks a lot for doing all this!  I don't have spare cycles either to
>>> drive this all the way to completion for all archs, I think we can still
>>> move forward with it, by adding a stop-gap implementation those
>>> archs, which just calls elfcore_write_prstatus.  I'll send it along
>>> in a bit.
>>
>> Like this.  I've pushed this to the users/palves/core-prstatus-rework
>> branch.  I've tested on x86_64 with:
>>
>>  make check RUNTESTFLAGS="--target_board unix/-m32" TESTS="*/*core*.exp"
>>  make check RUNTESTFLAGS="--target_board unix/-m64" TESTS="*/*core*.exp"
> 
> It failed
> 
>  make check RUNTESTFLAGS="--target_board unix/-mx32" TESTS="*/*core*.exp"
> 
> Please take a look at bfd/hosts/x86-64linux.h to see how to dump core
> x86-64.

I don't have x32 setup on my machine.  AFAICS, elf_prstatusx32 is exactly
like elf_prstatus32, except it takes a 64-bit regset.
Given gdbarch_ptr_bit==32 on x32, and:

      else if (gdbarch_ptr_bit (data->gdbarch) == 64)
	data->note_data = elfcore_write_linux_prstatus64 (data->obfd,
							  data->note_data,
							  data->note_size,
							  &prstatus);
      else
	data->note_data = elfcore_write_linux_prstatus32 (data->obfd,
							  data->note_data,
							  data->note_size,
							  &prstatus);

elfcore_write_linux_prstatus32 should be reached.  Other than that,
I have no idea (other than installing the stop gap on x32 too).

Thanks,
Pedro Alves

Re: Use elfcore_write_linux_prstatus in gdb

[H.J. Lu]

On Thu, Jan 21, 2016 at 6:05 AM, Pedro Alves <palves@redhat.com> wrote:
> On 01/21/2016 01:24 PM, H.J. Lu wrote:
>> On Thu, Jan 21, 2016 at 5:11 AM, Pedro Alves <palves@redhat.com> wrote:
>>> On 01/21/2016 12:25 PM, Pedro Alves wrote:
>>>> On 01/21/2016 09:28 AM, Alan Modra wrote:
>>>>> The gdb side.  Nothing fancy here, and no support for targets that
>>>>> need gdbarch_elfcore_write_linux_prstatus, mips n32, sparc64 if
>>>>> writing any of pr_{u,s,cu,cs}time, and probably x86-64 x32.
>>>>
>>>> Thanks a lot for doing all this!  I don't have spare cycles either to
>>>> drive this all the way to completion for all archs, I think we can still
>>>> move forward with it, by adding a stop-gap implementation those
>>>> archs, which just calls elfcore_write_prstatus.  I'll send it along
>>>> in a bit.
>>>
>>> Like this.  I've pushed this to the users/palves/core-prstatus-rework
>>> branch.  I've tested on x86_64 with:
>>>
>>>  make check RUNTESTFLAGS="--target_board unix/-m32" TESTS="*/*core*.exp"
>>>  make check RUNTESTFLAGS="--target_board unix/-m64" TESTS="*/*core*.exp"
>>
>> It failed
>>
>>  make check RUNTESTFLAGS="--target_board unix/-mx32" TESTS="*/*core*.exp"
>>
>> Please take a look at bfd/hosts/x86-64linux.h to see how to dump core
>> x86-64.
>
> I don't have x32 setup on my machine.  AFAICS, elf_prstatusx32 is exactly
> like elf_prstatus32, except it takes a 64-bit regset.
> Given gdbarch_ptr_bit==32 on x32, and:
>
>       else if (gdbarch_ptr_bit (data->gdbarch) == 64)
>         data->note_data = elfcore_write_linux_prstatus64 (data->obfd,
>                                                           data->note_data,
>                                                           data->note_size,
>                                                           &prstatus);
>       else
>         data->note_data = elfcore_write_linux_prstatus32 (data->obfd,
>                                                           data->note_data,
>                                                           data->note_size,
>                                                           &prstatus);
>
> elfcore_write_linux_prstatus32 should be reached.  Other than that,
> I have no idea (other than installing the stop gap on x32 too).
>

x32 has different core types (core layout) from m32 and m64.
See bfd/hosts/x86-64linux.h for details.  You need
elfcore_write_linux_prstatusx32 or change
elfcore_write_linux_prstatus32 to suport x32.


-- 
H.J.

Re: Use elfcore_write_linux_prstatus in gdb

[Pedro Alves]

On 01/21/2016 02:17 PM, H.J. Lu wrote:
> On Thu, Jan 21, 2016 at 6:05 AM, Pedro Alves <palves@redhat.com> wrote:
>> On 01/21/2016 01:24 PM, H.J. Lu wrote:

>>> Please take a look at bfd/hosts/x86-64linux.h to see how to dump core
>>> x86-64.
>>
>> I don't have x32 setup on my machine.  AFAICS, elf_prstatusx32 is exactly
>> like elf_prstatus32, except it takes a 64-bit regset.
...

>>
>> elfcore_write_linux_prstatus32 should be reached.  Other than that,
>> I have no idea (other than installing the stop gap on x32 too).
>>
> 
> x32 has different core types (core layout) from m32 and m64.
> See bfd/hosts/x86-64linux.h for details.  

As I said, "AFAICS, elf_prstatusx32 is exactly like elf_prstatus32,
except it takes a 64-bit regset."  It's probably an issue
of alignment/padding?

> You need
> elfcore_write_linux_prstatusx32 or change
> elfcore_write_linux_prstatus32 to suport x32.

I really can't be working further on this myself right now, sorry.

I've updated the branch to install the stop-gap on x32 as well, which
really ends up calling the same function we call today.
Please give it a try.

Thanks,
Pedro Alves

Re: Use elfcore_write_linux_prstatus in gdb

[H.J. Lu]

On Thu, Jan 21, 2016 at 6:26 AM, Pedro Alves <palves@redhat.com> wrote:
> On 01/21/2016 02:17 PM, H.J. Lu wrote:
>> On Thu, Jan 21, 2016 at 6:05 AM, Pedro Alves <palves@redhat.com> wrote:
>>> On 01/21/2016 01:24 PM, H.J. Lu wrote:
>
>>>> Please take a look at bfd/hosts/x86-64linux.h to see how to dump core
>>>> x86-64.
>>>
>>> I don't have x32 setup on my machine.  AFAICS, elf_prstatusx32 is exactly
>>> like elf_prstatus32, except it takes a 64-bit regset.
> ...
>
>>>
>>> elfcore_write_linux_prstatus32 should be reached.  Other than that,
>>> I have no idea (other than installing the stop gap on x32 too).
>>>
>>
>> x32 has different core types (core layout) from m32 and m64.
>> See bfd/hosts/x86-64linux.h for details.
>
> As I said, "AFAICS, elf_prstatusx32 is exactly like elf_prstatus32,
> except it takes a 64-bit regset."  It's probably an issue
> of alignment/padding?
>
>> You need
>> elfcore_write_linux_prstatusx32 or change
>> elfcore_write_linux_prstatus32 to suport x32.
>
> I really can't be working further on this myself right now, sorry.
>
> I've updated the branch to install the stop-gap on x32 as well, which
> really ends up calling the same function we call today.
> Please give it a try.
>

I will take a look.

-- 
H.J.

Re: Use elfcore_write_linux_prstatus in gdb

[Alan Modra]

On Thu, Jan 21, 2016 at 02:26:11PM +0000, Pedro Alves wrote:
> As I said, "AFAICS, elf_prstatusx32 is exactly like elf_prstatus32,
> except it takes a 64-bit regset."  It's probably an issue
> of alignment/padding?

Yes, I guess that the presence of elf_gregset64_t for x32 changes the
alignment of the external struct, inserting 4 bytes of padding at the
end.

-- 
Alan Modra
Australia Development Lab, IBM

Re: Use elfcore_write_linux_prstatus in gdb

[H.J. Lu]

On Thu, Jan 21, 2016 at 6:26 AM, Pedro Alves <palves@redhat.com> wrote:
> On 01/21/2016 02:17 PM, H.J. Lu wrote:
>> On Thu, Jan 21, 2016 at 6:05 AM, Pedro Alves <palves@redhat.com> wrote:
>>> On 01/21/2016 01:24 PM, H.J. Lu wrote:
>
>>>> Please take a look at bfd/hosts/x86-64linux.h to see how to dump core
>>>> x86-64.
>>>
>>> I don't have x32 setup on my machine.  AFAICS, elf_prstatusx32 is exactly
>>> like elf_prstatus32, except it takes a 64-bit regset.
> ...
>
>>>
>>> elfcore_write_linux_prstatus32 should be reached.  Other than that,
>>> I have no idea (other than installing the stop gap on x32 too).
>>>
>>
>> x32 has different core types (core layout) from m32 and m64.
>> See bfd/hosts/x86-64linux.h for details.
>
> As I said, "AFAICS, elf_prstatusx32 is exactly like elf_prstatus32,
> except it takes a 64-bit regset."  It's probably an issue
> of alignment/padding?
>
>> You need
>> elfcore_write_linux_prstatusx32 or change
>> elfcore_write_linux_prstatus32 to suport x32.
>
> I really can't be working further on this myself right now, sorry.
>
> I've updated the branch to install the stop-gap on x32 as well, which
> really ends up calling the same function we call today.
> Please give it a try.
>

It works.

Thanks.

-- 
H.J.

Re: Use elfcore_write_linux_prstatus in gdb

[Pedro Alves]

On 01/21/2016 03:59 PM, H.J. Lu wrote:
> On Thu, Jan 21, 2016 at 6:26 AM, Pedro Alves <palves@redhat.com> wrote:

>> I've updated the branch to install the stop-gap on x32 as well, which
>> really ends up calling the same function we call today.
>> Please give it a try.
>>
> 
> It works.
> 
> Thanks.
> 

Thanks.  As this risks breaking core dumping for many architectures,
I think we should hold on moving forward with it until after
gdb 7.11 branches.

Thanks,
Pedro Alves