Re: [RFC] Add support for PPC Altivec registers in gcore

[Carlos Eduardo Seo <cseo@linux.vnet.ibm.com>]

[-- Attachment #1: Type: text/plain, Size: 1061 bytes --]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1



Ulrich Weigand wrote:
>
> Completely removing all the fallbacks seems quite a bit of work;
> if you want to work on it that would certainly be welcome.
>
> However, I think it would be OK to start with having the gdbarch
> variable only contain the *extra* register sets, beyond the core
> and floating-point sets.
>
Ok. So, this would require removing the FILL_FPXREGSET ifdefs and all
the fill_fpxregset fallbacks. I think this is OK, but won't it break
anything in x86?
>
> If we do not have a collect_regset, we certainly do not want to
> write an note section containing uninitialized data.  The
> elfcore_write_ppc_vmx call needs to be inside the "if" block ...
Corrected.

Thanks,

- --
Carlos Eduardo Seo
Software Engineer
IBM Linux Technology Center
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.7 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFHJ54hqvq7Aov/qQARArXXAJ4z8a5f0YMzM2QxKX2QkuiJLTcGhgCbBsnS
NrH8jrngJzujytVbM/7wlXI=
=kR1E
-----END PGP SIGNATURE-----


[-- Attachment #2: gcore-altivec.diff --]
[-- Type: text/x-patch, Size: 4467 bytes --]

2007-10-30  Carlos Eduardo Seo  <cseo@linux.vnet.ibm.com>

	* linux-nat.c (linux_nat_do_thread_registers): Added
	support for PPC Altivec registers.
	* ppc-linux-nat.c: Removed definitions of SIZEOF_VRREGS
	and gdb_vrregset_t.
	* gregset.h: Added definitions of SIZEOF_VRREGS and
	gdb_vrregset_t.

Index: src/gdb/linux-nat.c
===================================================================
--- src.orig/gdb/linux-nat.c
+++ src/gdb/linux-nat.c
@@ -2616,6 +2616,7 @@ linux_nat_do_thread_registers (bfd *obfd
 {
   gdb_gregset_t gregs;
   gdb_fpregset_t fpregs;
+  gdb_vrregset_t vrregs;
 #ifdef FILL_FPXREGSET
   gdb_fpxregset_t fpxregs;
 #endif
@@ -2676,6 +2677,19 @@ linux_nat_do_thread_registers (bfd *obfd
 					       note_size,
 					       &fpxregs, sizeof (fpxregs));
 #endif
+
+  if (core_regset_p
+      && (regset = gdbarch_regset_from_core_section (gdbarch, ".reg-ppc-vmx",
+						     sizeof (vrregs))) != NULL
+      && regset->collect_regset != NULL)
+    {
+      regset->collect_regset (regset, regcache, -1,
+			      &vrregs, sizeof (vrregs));
+      note_data = (char *) elfcore_write_ppc_vmx (obfd,
+						  note_data,
+						  note_size,
+						  &vrregs, sizeof (vrregs));
+    }
   return note_data;
 }
 
Index: src/gdb/ppc-linux-nat.c
===================================================================
--- src.orig/gdb/ppc-linux-nat.c
+++ src/gdb/ppc-linux-nat.c
@@ -79,38 +79,6 @@
 #define PTRACE_GETSIGINFO    0x4202
 #endif
 
-/* This oddity is because the Linux kernel defines elf_vrregset_t as
-   an array of 33 16 bytes long elements.  I.e. it leaves out vrsave.
-   However the PTRACE_GETVRREGS and PTRACE_SETVRREGS requests return
-   the vrsave as an extra 4 bytes at the end.  I opted for creating a
-   flat array of chars, so that it is easier to manipulate for gdb.
-
-   There are 32 vector registers 16 bytes longs, plus a VSCR register
-   which is only 4 bytes long, but is fetched as a 16 bytes
-   quantity. Up to here we have the elf_vrregset_t structure.
-   Appended to this there is space for the VRSAVE register: 4 bytes.
-   Even though this vrsave register is not included in the regset
-   typedef, it is handled by the ptrace requests.
-
-   Note that GNU/Linux doesn't support little endian PPC hardware,
-   therefore the offset at which the real value of the VSCR register
-   is located will be always 12 bytes.
-
-   The layout is like this (where x is the actual value of the vscr reg): */
-
-/* *INDENT-OFF* */
-/*
-   |.|.|.|.|.....|.|.|.|.||.|.|.|x||.|
-   <------->     <-------><-------><->
-     VR0           VR31     VSCR    VRSAVE
-*/
-/* *INDENT-ON* */
-
-#define SIZEOF_VRREGS 33*16+4
-
-typedef char gdb_vrregset_t[SIZEOF_VRREGS];
-
-
 /* On PPC processors that support the the Signal Processing Extension
    (SPE) APU, the general-purpose registers are 64 bits long.
    However, the ordinary Linux kernel PTRACE_PEEKUSER / PTRACE_POKEUSER
Index: src/gdb/gregset.h
===================================================================
--- src.orig/gdb/gregset.h
+++ src/gdb/gregset.h
@@ -72,4 +72,37 @@ extern void fill_fpxregset (const struct
 			    gdb_fpxregset_t *fpxregs, int regno);
 #endif
 
+/* For PPC Altivec support  */
+
+/* This oddity is because the Linux kernel defines elf_vrregset_t as
+   an array of 33 16 bytes long elements.  I.e. it leaves out vrsave.
+   However the PTRACE_GETVRREGS and PTRACE_SETVRREGS requests return
+   the vrsave as an extra 4 bytes at the end.  I opted for creating a
+   flat array of chars, so that it is easier to manipulate for gdb.
+
+   There are 32 vector registers 16 bytes longs, plus a VSCR register
+   which is only 4 bytes long, but is fetched as a 16 bytes
+   quantity. Up to here we have the elf_vrregset_t structure.
+   Appended to this there is space for the VRSAVE register: 4 bytes.
+   Even though this vrsave register is not included in the regset
+   typedef, it is handled by the ptrace requests.
+
+   Note that GNU/Linux doesn't support little endian PPC hardware,
+   therefore the offset at which the real value of the VSCR register
+   is located will be always 12 bytes.
+
+   The layout is like this (where x is the actual value of the vscr reg): */
+
+/* *INDENT-OFF* */
+/*
+   |.|.|.|.|.....|.|.|.|.||.|.|.|x||.|
+   <------->     <-------><-------><->
+     VR0           VR31     VSCR    VRSAVE
+*/
+/* *INDENT-ON* */
+
+#define SIZEOF_VRREGS 33*16+4
+
+typedef char gdb_vrregset_t[SIZEOF_VRREGS];
+
 #endif

[-- Attachment #3: gcore-altivec.diff.sig --]
[-- Type: application/octet-stream, Size: 65 bytes --]