Mirror of the gdb-patches mailing list
 help / color / mirror / Atom feed
From: Aleksandar Ristovski <aristovski@qnx.com>
To: gdb-patches@sources.redhat.com
Subject: [patch] use .gnu.attributes to detect e500 machine type
Date: Wed, 08 Apr 2009 14:48:00 -0000	[thread overview]
Message-ID: <gridf7$uu0$1@ger.gmane.org> (raw)

[-- Attachment #1: Type: text/plain, Size: 2287 bytes --]

Hello,


The patch introduces determining e500 cpu from 
.gnu.attributes; if gnu attributes section is not present, 
fall-back to the old way of checking the presence of 
.PPC.EMB.apuinfo section.

It also introduces a function for "show powerpc vector-abi" 
to print both ABI being used and if different, global 
setting (output can be seen below). This could be a separate 
patch - let me know if you want it separately.

Background of the patch:

Currently, gdb relies on presence .PPC.EMB.apuinfo section. 
This is not completely correct since apuinfo section can 
contain information about any APU instructions present, and 
not only SPE. So we could have a binary with 
.PPC.EMB.apuinfo section and without SPE APU. Further, 
.PPC.EMB.apuinfo may not be present at all even if -me500 is 
specified. This is due to gas emitting this section only if 
it encounters SPE APU instructions in the assembly.

.gnu.attributes section records user setting, whether there 
were SPE instructions generated or not. We use gnu object 
attributes in rs6000 already (for determining vector-abi and 
float-point abi).

The only problem with .gnu.attribute is that if binary is 
compiled with pre-4.3 gcc, .gnu.attributes section does not 
get generated at all. (.gnu_attribute was added in 4.3). 
This is why we fall-back to .PPC.EMB.apuinfo presence.

Results of the patch:

After starting gdb, no executable:
(gdb) show powerpc vector-abi
The vector ABI in use is "generic". Default setting is "auto".
(gdb) file /mnt/win/ppcspetest_g
Reading symbols from /mnt/win/ppcspetest_g...done.
(gdb) show powerpc vector-abi
The vector ABI in use is "spe". Default setting is "auto".

** If user overrides selected vector-abi, this is the display:
(gdb) set powerpc vector-abi generic
(gdb) show powerpc vector-abi
The vector ABI is "generic"


"show powerpc vector-abi" will now display both selected 
vector-abi and user setting ("Default setting is ..."). Only 
if the two concide, it will only say "Target vector ABI is 
...").


Thanks,

Aleksandar Ristovski
QNX Software Systems


ChangeLog
* rs6000-tdep.c (rs6000_gdbarch_init): Use .gnu.attributes 
to see
if this is a e500 binary (with SPE APU).
(powerpc_show_vector_abi): New function.
(_initialize_rs6000_tdep): Use powerpc_show_vector_abi.



[-- Attachment #2: rs6000-tdep-mach_type_from_gnu_obj_attr-20090408-1.diff --]
[-- Type: text/x-patch, Size: 3074 bytes --]

Index: gdb/rs6000-tdep.c
===================================================================
RCS file: /cvs/src/src/gdb/rs6000-tdep.c,v
retrieving revision 1.328
diff -u -p -r1.328 rs6000-tdep.c
--- gdb/rs6000-tdep.c	22 Feb 2009 01:02:19 -0000	1.328
+++ gdb/rs6000-tdep.c	8 Apr 2009 13:13:12 -0000
@@ -3385,11 +3385,34 @@ rs6000_gdbarch_init (struct gdbarch_info
 
   if (info.abfd)
     {
-      sect = bfd_get_section_by_name (info.abfd, ".PPC.EMB.apuinfo");
-      if (sect)
+      /* We want to check .gnu.attributes first as it records user's 
+	 setting.  In particular, we want to know if this binary is meant
+	 to be executed on a popwer pc with  SPE APU.  To do that,
+	 we look for tag 8. Failing that, resort to detecting apuinfo 
+	 section.  See 'as.pdf', section 'Object Attributes'.   */
+      int tag8_value; /* Tag_GNU_Power_ABI_Vector */
+      unsigned long m = 0;
+
+      tag8_value = bfd_elf_get_obj_attr_int (info.abfd, OBJ_ATTR_GNU,
+					     Tag_GNU_Power_ABI_Vector);
+
+      /* The vector ABI used by this object file. The value will be:
+	 - 0 for files not affected by the vector ABI.
+	 - 1 for files using general purpose registers to pass vectors.
+	 - 2 for files using AltiVec registers to pass vectors.
+	 - 3 for files using SPE registers to pass vectors.  */
+      if (tag8_value == 3)
+	m = bfd_mach_ppc_e500;
+      else
+	{
+	  sect = bfd_get_section_by_name (info.abfd, ".PPC.EMB.apuinfo");
+	  if (sect)
+	    m = bfd_mach_ppc_e500;
+	}
+      if (m != 0)
 	{
+	  mach = m;
 	  arch = info.bfd_arch_info->arch;
-	  mach = bfd_mach_ppc_e500;
 	  bfd_default_set_arch_mach (&abfd, arch, mach);
 	  info.bfd_arch_info = bfd_get_arch_info (&abfd);
 	}
@@ -4003,6 +4026,36 @@ powerpc_set_vector_abi (char *args, int 
     internal_error (__FILE__, __LINE__, "could not update architecture");
 }
 
+static void
+powerpc_show_vector_abi (struct ui_file *file,
+			 int from_tty,
+			 struct cmd_list_element *c,
+			 const char *value)
+{
+  struct gdbarch *gdbarch = current_gdbarch;
+  enum powerpc_vector_abi tdep_vector_abi;
+  int global_msg_only = 0;
+
+  if (gdbarch == NULL)
+    global_msg_only = 1;
+  else 
+    {
+      tdep_vector_abi = gdbarch_tdep (gdbarch)->vector_abi;
+      if (strcmp (powerpc_vector_abi_string,
+		  powerpc_vector_strings[tdep_vector_abi]) == 0)
+	global_msg_only = 1;
+    }
+
+  if (global_msg_only)
+    printf_unfiltered ("The vector ABI is \"%s\"\n",
+		       powerpc_vector_abi_string);
+  else
+    printf_unfiltered ("\
+The vector ABI in use is \"%s\". Default setting is \"%s\".\n",
+		       powerpc_vector_strings[tdep_vector_abi],
+		       powerpc_vector_abi_string); 
+}
+
 /* Initialization code.  */
 
 extern initialize_file_ftype _initialize_rs6000_tdep; /* -Wmissing-prototypes */
@@ -4056,6 +4109,6 @@ _initialize_rs6000_tdep (void)
 			&powerpc_vector_abi_string,
 			_("Set the vector ABI."),
 			_("Show the vector ABI."),
-			NULL, powerpc_set_vector_abi, NULL,
+			NULL, powerpc_set_vector_abi, powerpc_show_vector_abi,
 			&setpowerpccmdlist, &showpowerpccmdlist);
 }

             reply	other threads:[~2009-04-08 14:48 UTC|newest]

Thread overview: 8+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2009-04-08 14:48 Aleksandar Ristovski [this message]
2009-04-14 19:48 ` Aleksandar Ristovski
2009-04-28 21:31 ` Joel Brobecker
2009-05-22 19:21   ` Aleksandar Ristovski
2009-07-06 16:34     ` Aleksandar Ristovski
2009-07-06 17:48       ` Daniel Jacobowitz
2009-07-06 21:26         ` Aleksandar Ristovski
2009-07-21 15:41           ` Aleksandar Ristovski

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to='gridf7$uu0$1@ger.gmane.org' \
    --to=aristovski@qnx.com \
    --cc=gdb-patches@sources.redhat.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox