This is the mail archive of the gdb-patches@sourceware.org mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [RFA][3/5] New port: Cell BE SPU (the port itself)


Mark Kettenis wrote:

> That's not really what I'm talking about.  I'm talking about having a single
> gdb binary that is able to execute both native powerpc code and spu code
> (but not at the same time).  As far as I can tell, it should be possible
> to do that by implementing to_can_run() in both ppc-linux-nat.c and
> spu-linux-nat.c, such that only one returns true at any time.

I've tried but failed to make that work.  The problem is that both
ppc-linux-nat.c and spu-linux-nat.c are based on inf-ptrace.c, and
it is currently impossible to register two targets resulting from
inf_ptrace_create at the same time, due to the ptrace_ops_hack.

Appended below is the code I'm currently using.  It requires building
a native powerpc64 target with --enable-targets=spu (which is necessary
anyway to get spu BFD support, and now also triggers inclusion of
both spu-linux-nat.c on the host side and spu-tdep.c on the target
side).

Even if this would work somehow, some other features of my original
patch appear difficult to implement.  For example, how would a gdbserver
server setup look like?  Also, with my approach I can attach to a thread
without having to specify an exec file, and GDB will retrieve a possibly
embedded SPU exec file from the inferior -- this also looks difficult.

More generally, your approach of going towards a single debugger for
Cell is of course something I would like to see work in the end -- but
that is something that requires difficult problems to solve first.  
I had hoped to start out with a strictly single-target SPU standalone
port first, that would be easily acceptable by avoiding those difficult
issues, and *then* in a second step address the combined Cell debugger.

I would be happy to try out whatever suggestions you might have to
achieve that goal ...

> >  I have a set of patches that does appear to work so far; it is based
> >  primarily on switching current_gdbarch on thread switch.  However,
> >  there's still some work to be done before this is in a shape suitable
> >  for mainline inclusion.
> 
> Andrew Cagney has talked a fair bit about this sort of things in the past.
> His idea was that each frame would have a gdbarch.  But a gdbarch per
> thread probably makes more sense.

In the end having a gdbarch per frame would be preferrable, so that
you can model things like: PPU code calls spu_run, which starts SPU
code execution, then a signal arrives and is handled on the PPU side.
However, even my experimental multi-arch debugger currently does not
support this.

Bye,
Ulrich


diff -urN gdb-orig/gdb/config/powerpc/linux-cell.mt gdb-head/gdb/config/powerpc/linux-cell.mt
--- gdb-orig/gdb/config/powerpc/linux-cell.mt	1970-01-01 01:00:00.000000000 +0100
+++ gdb-head/gdb/config/powerpc/linux-cell.mt	2006-11-16 04:24:03.000000000 +0100
@@ -0,0 +1,5 @@
+# Target: Linux on Cell BE (PowerPC + SPU)
+TDEPFILES= rs6000-tdep.o ppc-linux-tdep.o ppc-sysv-tdep.o spu-tdep.o \
+	solib.o solib-svr4.o solib-legacy.o corelow.o symfile-mem.o
+DEPRECATED_TM_FILE= tm-linux.h
+
diff -urN gdb-orig/gdb/config/powerpc/ppc64-linux-cell.mh gdb-head/gdb/config/powerpc/ppc64-linux-cell.mh
--- gdb-orig/gdb/config/powerpc/ppc64-linux-cell.mh	1970-01-01 01:00:00.000000000 +0100
+++ gdb-head/gdb/config/powerpc/ppc64-linux-cell.mh	2006-11-16 04:25:21.000000000 +0100
@@ -0,0 +1,18 @@
+# Host: Cell BE (PowerPC64 + SPU), running Linux
+
+XM_CLIBS=
+
+NAT_FILE= nm-ppc64-linux.h
+NATDEPFILES= inf-ptrace.o fork-child.o \
+	ppc-linux-nat.o proc-service.o linux-thread-db.o \
+	gcore.o linux-nat.o spu-linux-nat.o linux-fork.o
+
+# The PowerPC has severe limitations on TOC size, and uses them even
+# for non-PIC code.  GDB overflows those tables when compiling with
+# -mfull-toc (the default), so we need to ask GCC to use as few TOC
+# entries as possible.
+MH_CFLAGS= -mminimal-toc
+
+# The dynamically loaded libthread_db needs access to symbols in the
+# gdb executable.
+LOADLIBES= -ldl -rdynamic
diff -urN gdb-orig/gdb/configure.host gdb-head/gdb/configure.host
--- gdb-orig/gdb/configure.host	2006-02-10 22:53:51.000000000 +0100
+++ gdb-head/gdb/configure.host	2006-11-16 04:21:41.000000000 +0100
@@ -113,7 +113,18 @@
 			gdb_host=nbsd ;;
 powerpc-*-openbsd*)	gdb_host=obsd ;;
 
-powerpc64-*-linux*)     gdb_host=ppc64-linux ;;
+powerpc64-*-linux*)	gdb_host=ppc64-linux
+			# If we were built with --enable-targets=spu,
+			# enable combined Cell BE debugging.
+			if test -n "$enable_targets" ; then
+			  for targ in `echo $enable_targets | sed 's/,/ /g'`
+			    do
+			      case "$targ" in
+				all | spu*) gdb_host=ppc64-linux-cell ;;
+			      esac
+			    done
+			fi
+			;;
 
 rs6000-*-lynxos*)	gdb_host=rs6000lynx ;;
 rs6000-*-aix4*)		gdb_host=aix4 ;;
diff -urN gdb-orig/gdb/configure.tgt gdb-head/gdb/configure.tgt
--- gdb-orig/gdb/configure.tgt	2006-11-10 02:28:35.000000000 +0100
+++ gdb-head/gdb/configure.tgt	2006-11-16 04:19:50.000000000 +0100
@@ -161,6 +161,16 @@
 			build_gdbserver=yes
 			;;
 powerpc64-*-linux*)	gdb_target=linux
+			# If we were built with --enable-targets=spu,
+			# enable combined Cell BE debugging.
+			if test -n "$enable_targets" ; then
+			  for targ in `echo $enable_targets | sed 's/,/ /g'`
+			    do
+			      case "$targ" in
+				all | spu*) gdb_target=linux-cell ;;
+			      esac
+			    done
+			fi
 			build_gdbserver=yes
 			;;
 powerpc*-*-*)		if test -f ../sim/ppc/Makefile; then
diff -urN gdb-orig/gdb/Makefile.in gdb-head/gdb/Makefile.in
--- gdb-orig/gdb/Makefile.in	2006-11-10 02:28:35.000000000 +0100
+++ gdb-head/gdb/Makefile.in	2006-11-16 04:16:15.000000000 +0100
@@ -799,6 +799,7 @@
 sparc64_tdep_h = sparc64-tdep.h $(sparc_tdep_h)
 sparc_nat_h = sparc-nat.h
 sparc_tdep_h = sparc-tdep.h
+spu_tdep_h = spu-tdep.h
 srec_h = srec.h
 stabsread_h = stabsread.h
 stack_h = stack.h
@@ -1496,6 +1497,7 @@
 	sparc64-tdep.c sparc64fbsd-nat.c sparc64fbsd-tdep.c \
 	sparc64nbsd-nat.c sparc64nbsd-tdep.c sparc64obsd-tdep.c \
 	sparcnbsd-nat.c sparcnbsd-tdep.c sparcobsd-tdep.c \
+	spu-linux-nat.c spu-tdep.c \
 	v850-tdep.c \
 	vax-nat.c vax-tdep.c vaxbsd-nat.c vaxnbsd-tdep.c \
 	win32-nat.c \
@@ -2718,6 +2720,15 @@
 	$(frame_unwind_h) $(gdbcore_h) $(gdbtypes_h) $(inferior_h) \
 	$(symtab_h) $(objfiles_h) $(osabi_h) $(regcache_h) $(target_h) \
 	$(value_h) $(gdb_assert_h) $(gdb_string_h) $(sparc_tdep_h)
+spu-linux-nat.o: spu-linux-nat.c $(defs_h) $(gdbcore_h) $(gdb_string_h) \
+	$(target_h) $(inferior_h) $(inf_ptrace_h) $(regcache_h) $(symfile_h) \
+	$(gdb_wait_h) $(spu_tdep_h)
+spu-tdep.o: spu-tdep.c $(defs_h) $(arch_utils_h) $(gdbtypes_h) $(gdbcmd_h) \
+	$(gdbcore_h) $(gdb_string_h) $(gdb_assert_h) $(frame_h) \
+	$(frame_unwind_h) $(frame_base_h) $(trad_frame_h) $(symtab_h) \
+	$(symfile_h) $(value_h) $(inferior_h) $(dis_asm_h) $(objfiles_h) \
+	$(language_h) $(regcache_h) $(reggroups_h) $(floatformat_h) \
+	$(spu_tdep_h)
 stabsread.o: stabsread.c $(defs_h) $(gdb_string_h) $(bfd_h) $(gdb_obstack_h) \
 	$(symtab_h) $(gdbtypes_h) $(expression_h) $(symfile_h) $(objfiles_h) \
 	$(aout_stab_gnu_h) $(libaout_h) $(aout_aout64_h) $(gdb_stabs_h) \
diff -urN gdb-orig/gdb/ppc-linux-nat.c gdb-head/gdb/ppc-linux-nat.c
--- gdb-orig/gdb/ppc-linux-nat.c	2006-05-05 22:45:03.000000000 +0200
+++ gdb-head/gdb/ppc-linux-nat.c	2006-11-16 04:29:44.000000000 +0100
@@ -1002,6 +1002,14 @@
     }
 }
 
+/* Override the to_can_run routine.  */
+static int
+ppc_linux_can_run (void)
+{
+  return !exec_bfd || bfd_get_arch (exec_bfd) != bfd_arch_spu;
+}
+
+
 void _initialize_ppc_linux_nat (void);
 
 void
@@ -1012,6 +1020,9 @@
   /* Fill in the generic GNU/Linux methods.  */
   t = linux_target ();
 
+  /* Override to_can_run routine.  */
+  t->to_can_run = ppc_linux_can_run;
+
   /* Add our register access methods.  */
   t->to_fetch_registers = ppc_linux_fetch_inferior_registers;
   t->to_store_registers = ppc_linux_store_inferior_registers;
diff -urN gdb-orig/gdb/spu-linux-nat.c gdb-head/gdb/spu-linux-nat.c
--- gdb-orig/gdb/spu-linux-nat.c	1970-01-01 01:00:00.000000000 +0100
+++ gdb-head/gdb/spu-linux-nat.c	2006-11-16 04:28:43.000000000 +0100
@@ -0,0 +1,565 @@
+/* SPU native-dependent code for GDB, the GNU debugger.
+   Copyright (C) 2006 Free Software Foundation, Inc.
+
+   Contributed by Ulrich Weigand <uweigand@de.ibm.com>.
+
+   This file is part of GDB.
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 2 of the License, or
+   (at your option) any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program; if not, write to the Free Software
+   Foundation, Inc., 51 Franklin Street, Fifth Floor,
+   Boston, MA 02110-1301, USA.  */
+
+#include "defs.h"
+#include "gdbcore.h"
+#include "gdb_string.h"
+#include "target.h"
+#include "inferior.h"
+#include "inf-ptrace.h"
+#include "regcache.h"
+#include "symfile.h"
+#include "gdb_wait.h"
+
+#include <sys/ptrace.h>
+#include <asm/ptrace.h>
+#include <sys/types.h>
+#include <sys/param.h>
+
+#include "spu-tdep.h"
+
+/* PPU side system calls.  */
+#define INSTR_SC	0x44000002
+#define NR_spu_run	0x0116
+
+
+/* Fetch PPU register REGNO.  */
+static CORE_ADDR
+fetch_ppc_register (int regno)
+{
+  PTRACE_TYPE_RET res;
+
+  int tid = TIDGET (inferior_ptid);
+  if (tid == 0)
+    tid = PIDGET (inferior_ptid);
+
+#ifndef __powerpc64__
+  /* If running as a 32-bit process on a 64-bit system, we attempt
+     to get the full 64-bit register content of the target process.
+     If the PPC special ptrace call fails, we're on a 32-bit system;
+     just fall through to the regular ptrace call in that case.  */
+  {
+    gdb_byte buf[8];
+
+    errno = 0;
+    ptrace (PPC_PTRACE_PEEKUSR_3264, tid,
+	    (PTRACE_TYPE_ARG3) (regno * 8), buf);
+    if (errno == 0)
+      ptrace (PPC_PTRACE_PEEKUSR_3264, tid,
+	      (PTRACE_TYPE_ARG3) (regno * 8 + 4), buf + 4);
+    if (errno == 0)
+      return (CORE_ADDR) *(unsigned long long *)buf;
+  }
+#endif
+
+  errno = 0;
+  res = ptrace (PT_READ_U, tid,
+	 	(PTRACE_TYPE_ARG3) (regno * sizeof (PTRACE_TYPE_RET)), 0);
+  if (errno != 0)
+    {
+      char mess[128];
+      xsnprintf (mess, sizeof mess, "reading PPC register #%d", regno);
+      perror_with_name (_(mess));
+    }
+
+  return (CORE_ADDR) (unsigned long) res;
+}
+
+/* Fetch WORD from PPU memory at (aligned) MEMADDR in thread TID.  */
+static int
+fetch_ppc_memory_1 (int tid, CORE_ADDR memaddr, PTRACE_TYPE_RET *word)
+{
+  errno = 0;
+
+#ifndef __powerpc64__
+  if (memaddr >> 32)
+    {
+      unsigned long long addr_8 = (unsigned long long) memaddr;
+      ptrace (PPC_PTRACE_PEEKTEXT_3264, tid, (PTRACE_TYPE_ARG3) &addr_8, word);
+    }
+  else
+#endif
+    *word = ptrace (PT_READ_I, tid, (PTRACE_TYPE_ARG3) (size_t) memaddr, 0);
+
+  return errno;
+}
+
+/* Store WORD into PPU memory at (aligned) MEMADDR in thread TID.  */
+static int
+store_ppc_memory_1 (int tid, CORE_ADDR memaddr, PTRACE_TYPE_RET word)
+{
+  errno = 0;
+
+#ifndef __powerpc64__
+  if (memaddr >> 32)
+    {
+      unsigned long long addr_8 = (unsigned long long) memaddr;
+      ptrace (PPC_PTRACE_POKEDATA_3264, tid, (PTRACE_TYPE_ARG3) &addr_8, word);
+    }
+  else
+#endif
+    ptrace (PT_WRITE_D, tid, (PTRACE_TYPE_ARG3) (size_t) memaddr, word);
+
+  return errno;
+}
+
+/* Fetch LEN bytes of PPU memory at MEMADDR to MYADDR.  */
+static int
+fetch_ppc_memory (CORE_ADDR memaddr, gdb_byte *myaddr, int len)
+{
+  int i, ret;
+
+  CORE_ADDR addr = memaddr & -(CORE_ADDR) sizeof (PTRACE_TYPE_RET);
+  int count = ((((memaddr + len) - addr) + sizeof (PTRACE_TYPE_RET) - 1)
+	       / sizeof (PTRACE_TYPE_RET));
+  PTRACE_TYPE_RET *buffer;
+
+  int tid = TIDGET (inferior_ptid);
+  if (tid == 0)
+    tid = PIDGET (inferior_ptid);
+
+  buffer = (PTRACE_TYPE_RET *) alloca (count * sizeof (PTRACE_TYPE_RET));
+  for (i = 0; i < count; i++, addr += sizeof (PTRACE_TYPE_RET))
+    if ((ret = fetch_ppc_memory_1 (tid, addr, &buffer[i])) != 0)
+      return ret;
+
+  memcpy (myaddr,
+	  (char *) buffer + (memaddr & (sizeof (PTRACE_TYPE_RET) - 1)),
+	  len);
+
+  return 0;
+}
+
+/* Store LEN bytes from MYADDR to PPU memory at MEMADDR.  */
+static int
+store_ppc_memory (CORE_ADDR memaddr, const gdb_byte *myaddr, int len)
+{
+  int i, ret;
+
+  CORE_ADDR addr = memaddr & -(CORE_ADDR) sizeof (PTRACE_TYPE_RET);
+  int count = ((((memaddr + len) - addr) + sizeof (PTRACE_TYPE_RET) - 1)
+	       / sizeof (PTRACE_TYPE_RET));
+  PTRACE_TYPE_RET *buffer;
+
+  int tid = TIDGET (inferior_ptid);
+  if (tid == 0)
+    tid = PIDGET (inferior_ptid);
+
+  buffer = (PTRACE_TYPE_RET *) alloca (count * sizeof (PTRACE_TYPE_RET));
+
+  if (addr != memaddr || len < (int) sizeof (PTRACE_TYPE_RET))
+    if ((ret = fetch_ppc_memory_1 (tid, addr, &buffer[0])) != 0)
+      return ret;
+
+  if (count > 1)
+    if ((ret = fetch_ppc_memory_1 (tid, addr + (count - 1)
+					       * sizeof (PTRACE_TYPE_RET),
+				   &buffer[count - 1])) != 0)
+      return ret;
+
+  memcpy ((char *) buffer + (memaddr & (sizeof (PTRACE_TYPE_RET) - 1)),
+          myaddr, len);
+
+  for (i = 0; i < count; i++, addr += sizeof (PTRACE_TYPE_RET))
+    if ((ret = store_ppc_memory_1 (tid, addr, buffer[i])) != 0)
+      return ret;
+
+  return 0;
+}
+
+
+/* If the PPU thread is currently stopped on a spu_run system call,
+   return to FD and ADDR the file handle and NPC parameter address
+   used with the system call.  Return non-zero if successful.  */
+static int 
+parse_spufs_run (int *fd, CORE_ADDR *addr)
+{
+  gdb_byte buf[4];
+  CORE_ADDR pc = fetch_ppc_register (32);  /* nip */
+
+  /* Fetch instruction preceding current NIP.  */
+  if (fetch_ppc_memory (pc-4, buf, 4) != 0)
+    return 0;
+  /* It should be a "sc" instruction.  */
+  if (extract_unsigned_integer (buf, 4) != INSTR_SC)
+    return 0;
+  /* System call number should be NR_spu_run.  */
+  if (fetch_ppc_register (0) != NR_spu_run)
+    return 0;
+
+  /* Register 3 contains fd, register 4 the NPC param pointer.  */
+  *fd = fetch_ppc_register (34);  /* orig_gpr3 */
+  *addr = fetch_ppc_register (4);
+  return 1;
+}
+
+
+/* Copy LEN bytes at OFFSET in spufs file ANNEX into/from READBUF or WRITEBUF,
+   using the /proc file system.  */
+static LONGEST
+spu_proc_xfer_spu (const char *annex, gdb_byte *readbuf,
+		   const gdb_byte *writebuf,
+		   ULONGEST offset, LONGEST len)
+{
+  char buf[128];
+  int fd = 0;
+  int ret = -1;
+  int pid = PIDGET (inferior_ptid);
+
+  if (!annex)
+    return 0;
+
+  xsnprintf (buf, sizeof buf, "/proc/%d/fd/%s", pid, annex);
+  fd = open (buf, writebuf? O_WRONLY : O_RDONLY);
+  if (fd <= 0)
+    return -1;
+
+  if (offset != 0
+      && lseek (fd, (off_t) offset, SEEK_SET) != (off_t) offset)
+    {
+      close (fd);
+      return -1;
+    }
+
+  if (writebuf)
+    ret = write (fd, writebuf, (size_t) len);
+  else if (readbuf)
+    ret = read (fd, readbuf, (size_t) len);
+
+  close (fd);
+  return ret;
+}
+
+
+/* Inferior memory should contain an SPE executable image at location ADDR.
+   Allocate a BFD representing that executable.  Return NULL on error.  */
+
+static void *
+spu_bfd_iovec_open (struct bfd *nbfd, void *open_closure)
+{
+  return open_closure;
+}
+
+static int
+spu_bfd_iovec_close (struct bfd *nbfd, void *stream)
+{
+  xfree (stream);
+  return 1;
+}
+
+static file_ptr
+spu_bfd_iovec_pread (struct bfd *abfd, void *stream, void *buf,
+	             file_ptr nbytes, file_ptr offset)
+{
+  CORE_ADDR addr = *(CORE_ADDR *)stream;
+
+  if (fetch_ppc_memory (addr + offset, buf, nbytes) != 0)
+    {
+      bfd_set_error (bfd_error_invalid_operation);
+      return -1;
+    }
+
+  return nbytes;
+}
+
+static bfd *
+spu_bfd_open (CORE_ADDR addr)
+{
+  struct bfd *nbfd;
+
+  CORE_ADDR *open_closure = xmalloc (sizeof (CORE_ADDR));
+  *open_closure = addr;
+
+  nbfd = bfd_openr_iovec (xstrdup ("<in-memory>"), "elf32-spu",
+			  spu_bfd_iovec_open, open_closure,
+			  spu_bfd_iovec_pread, spu_bfd_iovec_close);
+  if (!nbfd)
+    return NULL;
+
+  if (!bfd_check_format (nbfd, bfd_object))
+    {
+      bfd_close (nbfd);
+      return NULL;
+    }
+
+  return nbfd;
+}
+
+/* INFERIOR_FD is a file handle passed by the inferior to the
+   spu_run system call.  Assuming the SPE context was allocated
+   by the libspe library, try to retrieve the main SPE executable
+   file from its copy within the target process.  */
+static void
+spu_symbol_file_add_from_memory (int inferior_fd)
+{
+  CORE_ADDR addr;
+  struct bfd *nbfd;
+
+  char id[128];
+  char annex[32];
+  int len;
+
+  /* Read object ID.  */
+  xsnprintf (annex, sizeof annex, "%d/object-id", inferior_fd);
+  len = spu_proc_xfer_spu (annex, id, NULL, 0, sizeof id);
+  if (len <= 0 || len >= sizeof id)
+    return;
+  id[len] = 0;
+  if (sscanf (id, "0x%llx", &addr) != 1)
+    return;
+
+  /* Open BFD representing SPE executable and read its symbols.  */
+  nbfd = spu_bfd_open (addr);
+  if (nbfd)
+    symbol_file_add_from_bfd (nbfd, 0, NULL, 1, 0);
+}
+
+
+/* Override the post_startup_inferior routine to continue running
+   the inferior until the first spu_run system call.  */
+static void
+spu_child_post_startup_inferior (ptid_t ptid)
+{
+  int fd;
+  CORE_ADDR addr;
+
+  int tid = TIDGET (ptid);
+  if (tid == 0)
+    tid = PIDGET (ptid);
+  
+  while (!parse_spufs_run (&fd, &addr))
+    {
+      ptrace (PT_SYSCALL, tid, (PTRACE_TYPE_ARG3) 0, 0);
+      waitpid (tid, NULL, __WALL | __WNOTHREAD);
+    }
+}
+
+/* Override the post_attach routine to try load the SPE executable
+   file image from its copy inside the target process.  */
+static void
+spu_child_post_attach (int pid)
+{
+  int fd;
+  CORE_ADDR addr;
+
+  /* Like child_post_startup_inferior, if we happened to attach to
+     the inferior while it wasn't currently in spu_run, continue 
+     running it until we get back there.  */
+  while (!parse_spufs_run (&fd, &addr))
+    {
+      ptrace (PT_SYSCALL, pid, (PTRACE_TYPE_ARG3) 0, 0);
+      waitpid (pid, NULL, __WALL | __WNOTHREAD);
+    }
+
+  /* If the user has not provided an executable file, try to extract
+     the image from inside the target process.  */
+  if (!get_exec_file (0))
+    spu_symbol_file_add_from_memory (fd);
+}
+
+/* Wait for child PTID to do something.  Return id of the child,
+   minus_one_ptid in case of error; store status into *OURSTATUS.  */
+static ptid_t
+spu_child_wait (ptid_t ptid, struct target_waitstatus *ourstatus)
+{
+  int save_errno;
+  int status;
+  pid_t pid;
+
+  do
+    {
+      set_sigint_trap ();	/* Causes SIGINT to be passed on to the
+				   attached process.  */
+      set_sigio_trap ();
+
+      pid = waitpid (PIDGET (ptid), &status, 0);
+      if (pid == -1 && errno == ECHILD)
+	/* Try again with __WCLONE to check cloned processes.  */
+	pid = waitpid (PIDGET (ptid), &status, __WCLONE);
+
+      save_errno = errno;
+
+      /* Make sure we don't report an event for the exit of the
+         original program, if we've detached from it.  */
+      if (pid != -1 && !WIFSTOPPED (status) && pid != PIDGET (inferior_ptid))
+	{
+	  pid = -1;
+	  save_errno = EINTR;
+	}
+
+      clear_sigio_trap ();
+      clear_sigint_trap ();
+    }
+  while (pid == -1 && save_errno == EINTR);
+
+  if (pid == -1)
+    {
+      warning ("Child process unexpectedly missing: %s",
+	       safe_strerror (save_errno));
+
+      /* Claim it exited with unknown signal.  */
+      ourstatus->kind = TARGET_WAITKIND_SIGNALLED;
+      ourstatus->value.sig = TARGET_SIGNAL_UNKNOWN;
+      return minus_one_ptid;
+    }
+
+  store_waitstatus (ourstatus, status);
+  return pid_to_ptid (pid);
+}
+
+/* Override the fetch_inferior_register routine.  */
+static void
+spu_fetch_inferior_registers (int regno)
+{
+  int fd;
+  CORE_ADDR addr;
+
+  /* We must be stopped on a spu_run system call.  */
+  if (!parse_spufs_run (&fd, &addr))
+    return;
+
+  /* The ID register holds the spufs file handle.  */
+  if (regno == -1 || regno == SPU_ID_REGNUM)
+    {
+      char buf[4];
+      store_unsigned_integer (buf, 4, fd);
+      regcache_raw_supply (current_regcache, SPU_ID_REGNUM, buf);
+    }
+
+  /* The NPC register is found at ADDR.  */
+  if (regno == -1 || regno == SPU_PC_REGNUM)
+    {
+      gdb_byte buf[4];
+      if (fetch_ppc_memory (addr, buf, 4) == 0)
+	regcache_raw_supply (current_regcache, SPU_PC_REGNUM, buf);
+    }
+
+  /* The GPRs are found in the "regs" spufs file.  */
+  if (regno == -1 || (regno >= 0 && regno < SPU_NUM_GPRS))
+    {
+      gdb_byte buf[16 * SPU_NUM_GPRS];
+      char annex[32];
+      int i;
+
+      xsnprintf (annex, sizeof annex, "%d/regs", fd);
+      if (spu_proc_xfer_spu (annex, buf, NULL, 0, sizeof buf) == sizeof buf)
+	for (i = 0; i < SPU_NUM_GPRS; i++)
+	  regcache_raw_supply (current_regcache, i, buf + i*16);
+    }
+}
+
+/* Override the store_inferior_register routine.  */
+static void
+spu_store_inferior_registers (int regno)
+{
+  int fd;
+  CORE_ADDR addr;
+
+  /* We must be stopped on a spu_run system call.  */
+  if (!parse_spufs_run (&fd, &addr))
+    return;
+
+  /* The NPC register is found at ADDR.  */
+  if (regno == -1 || regno == SPU_PC_REGNUM)
+    {
+      gdb_byte buf[4];
+      regcache_raw_collect (current_regcache, SPU_PC_REGNUM, buf);
+      store_ppc_memory (addr, buf, 4);
+    }
+
+  /* The GPRs are found in the "regs" spufs file.  */
+  if (regno == -1 || (regno >= 0 && regno < SPU_NUM_GPRS))
+    {
+      gdb_byte buf[16 * SPU_NUM_GPRS];
+      char annex[32];
+      int i;
+
+      for (i = 0; i < SPU_NUM_GPRS; i++)
+	regcache_raw_collect (current_regcache, i, buf + i*16);
+
+      xsnprintf (annex, sizeof annex, "%d/regs", fd);
+      spu_proc_xfer_spu (annex, NULL, buf, 0, sizeof buf);
+    }
+}
+
+/* Override the to_xfer_partial routine.  */
+static LONGEST 
+spu_xfer_partial (struct target_ops *ops,
+		  enum target_object object, const char *annex,
+		  gdb_byte *readbuf, const gdb_byte *writebuf,
+		  ULONGEST offset, LONGEST len)
+{
+  if (object == TARGET_OBJECT_MEMORY)
+    {
+      int fd;
+      CORE_ADDR addr;
+      char mem_annex[32];
+
+      /* We must be stopped on a spu_run system call.  */
+      if (!parse_spufs_run (&fd, &addr))
+	return 0;
+
+      /* Use the "mem" spufs file to access SPU local store.  */
+      xsnprintf (mem_annex, sizeof mem_annex, "%d/mem", fd);
+      return spu_proc_xfer_spu (mem_annex, readbuf, writebuf, offset, len);
+    }
+
+  return 0;
+}
+
+/* Override the to_can_use_hw_breakpoint routine.  */
+static int
+spu_can_use_hw_breakpoint (int type, int cnt, int othertype)
+{
+  return 0;
+}
+
+/* Override the to_can_run routine.  */
+static int
+spu_can_run (void)
+{
+  return exec_bfd && bfd_get_arch (exec_bfd) == bfd_arch_spu;
+}
+
+
+/* Initialize SPU native target.  */
+void 
+_initialize_spu_nat (void)
+{
+  /* Generic ptrace methods.  */
+  struct target_ops *t;
+  t = inf_ptrace_target ();
+
+  /* Add SPU methods.  */
+  t->to_can_run = spu_can_run;
+  t->to_post_attach = spu_child_post_attach;  
+  t->to_post_startup_inferior = spu_child_post_startup_inferior;
+  t->to_wait = spu_child_wait;
+  t->to_fetch_registers = spu_fetch_inferior_registers;
+  t->to_store_registers = spu_store_inferior_registers;
+  t->to_xfer_partial = spu_xfer_partial;
+  t->to_can_use_hw_breakpoint = spu_can_use_hw_breakpoint;
+
+  /* Register SPU target.  */
+  add_target (t);
+}
+
diff -urN gdb-orig/gdb/spu-tdep.c gdb-head/gdb/spu-tdep.c
--- gdb-orig/gdb/spu-tdep.c	1970-01-01 01:00:00.000000000 +0100
+++ gdb-head/gdb/spu-tdep.c	2006-11-16 04:16:15.000000000 +0100
@@ -0,0 +1,1099 @@
+/* SPU target-dependent code for GDB, the GNU debugger.
+   Copyright (C) 2006 Free Software Foundation, Inc.
+
+   Contributed by Ulrich Weigand <uweigand@de.ibm.com>.
+   Based on a port by Sid Manning <sid@us.ibm.com>.
+
+   This file is part of GDB.
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 2 of the License, or
+   (at your option) any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program; if not, write to the Free Software
+   Foundation, Inc., 51 Franklin Street, Fifth Floor,
+   Boston, MA 02110-1301, USA.  */
+
+#include "defs.h"
+#include "arch-utils.h"
+#include "gdbtypes.h"
+#include "gdbcmd.h"
+#include "gdbcore.h"
+#include "gdb_string.h"
+#include "gdb_assert.h"
+#include "frame.h"
+#include "frame-unwind.h"
+#include "frame-base.h"
+#include "trad-frame.h"
+#include "symtab.h"
+#include "symfile.h"
+#include "value.h"
+#include "inferior.h"
+#include "dis-asm.h"
+#include "objfiles.h"
+#include "language.h"
+#include "regcache.h"
+#include "reggroups.h"
+#include "floatformat.h"
+
+#include "spu-tdep.h"
+
+
+/* Registers.  */
+
+static const char *
+spu_register_name (int reg_nr)
+{
+  static char *register_names[] = 
+    {
+      "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7",
+      "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15",
+      "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23",
+      "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31",
+      "r32", "r33", "r34", "r35", "r36", "r37", "r38", "r39",
+      "r40", "r41", "r42", "r43", "r44", "r45", "r46", "r47",
+      "r48", "r49", "r50", "r51", "r52", "r53", "r54", "r55",
+      "r56", "r57", "r58", "r59", "r60", "r61", "r62", "r63",
+      "r64", "r65", "r66", "r67", "r68", "r69", "r70", "r71",
+      "r72", "r73", "r74", "r75", "r76", "r77", "r78", "r79",
+      "r80", "r81", "r82", "r83", "r84", "r85", "r86", "r87",
+      "r88", "r89", "r90", "r91", "r92", "r93", "r94", "r95",
+      "r96", "r97", "r98", "r99", "r100", "r101", "r102", "r103",
+      "r104", "r105", "r106", "r107", "r108", "r109", "r110", "r111",
+      "r112", "r113", "r114", "r115", "r116", "r117", "r118", "r119",
+      "r120", "r121", "r122", "r123", "r124", "r125", "r126", "r127",
+      "id", "pc", "sp"
+    };
+
+  if (reg_nr < 0)
+    return NULL;
+  if (reg_nr >= sizeof register_names / sizeof *register_names)
+    return NULL;
+
+  return register_names[reg_nr];
+}
+
+static struct type *
+spu_register_type (struct gdbarch *gdbarch, int reg_nr)
+{
+  if (reg_nr < SPU_NUM_GPRS)
+    return builtin_type_vec128;
+
+  switch (reg_nr)
+    {
+    case SPU_ID_REGNUM:
+      return builtin_type_uint32;
+
+    case SPU_PC_REGNUM:
+      return builtin_type_void_func_ptr;
+
+    case SPU_SP_REGNUM:
+      return builtin_type_void_data_ptr;
+
+    default:
+      internal_error (__FILE__, __LINE__, "invalid regnum");
+    }
+}
+
+/* Pseudo registers for preferred slots - stack pointer.  */
+
+static void
+spu_pseudo_register_read (struct gdbarch *gdbarch, struct regcache *regcache,
+                          int regnum, gdb_byte *buf)
+{
+  gdb_byte reg[16];
+
+  switch (regnum)
+    {
+    case SPU_SP_REGNUM:
+      regcache_raw_read (regcache, SPU_RAW_SP_REGNUM, reg);
+      memcpy (buf, reg, 4);
+      break;
+
+    default:
+      internal_error (__FILE__, __LINE__, _("invalid regnum"));
+    }
+}
+
+static void
+spu_pseudo_register_write (struct gdbarch *gdbarch, struct regcache *regcache,
+                           int regnum, const gdb_byte *buf)
+{
+  gdb_byte reg[16];
+
+  switch (regnum)
+    {
+    case SPU_SP_REGNUM:
+      regcache_raw_read (regcache, SPU_RAW_SP_REGNUM, reg);
+      memcpy (reg, buf, 4);
+      regcache_raw_write (regcache, SPU_RAW_SP_REGNUM, reg);
+      break;
+
+    default:
+      internal_error (__FILE__, __LINE__, _("invalid regnum"));
+    }
+}
+
+/* Value conversion -- access scalar values at the preferred slot.  */
+
+static int
+spu_convert_register_p (int regno, struct type *type)
+{
+  return regno < SPU_NUM_GPRS && TYPE_LENGTH (type) < 16;
+}
+
+static void
+spu_register_to_value (struct frame_info *frame, int regnum,
+                       struct type *valtype, gdb_byte *out)
+{
+  gdb_byte in[16];
+  int len = TYPE_LENGTH (valtype);
+  int preferred_slot = len < 4 ? 4 - len : 0;
+  gdb_assert (len < 16);
+
+  get_frame_register (frame, regnum, in);
+  memcpy (out, in + preferred_slot, len);
+}
+
+static void
+spu_value_to_register (struct frame_info *frame, int regnum,
+                       struct type *valtype, const gdb_byte *in)
+{
+  gdb_byte out[16];
+  int len = TYPE_LENGTH (valtype);
+  int preferred_slot = len < 4 ? 4 - len : 0;
+  gdb_assert (len < 16);
+
+  memset (out, 0, 16);
+  memcpy (out + preferred_slot, in, len);
+  put_frame_register (frame, regnum, out);
+}
+
+/* Register groups.  */
+
+static int
+spu_register_reggroup_p (struct gdbarch *gdbarch, int regnum,
+			 struct reggroup *group)
+{
+  /* Registers displayed via 'info regs'.  */
+  if (group == general_reggroup)
+    return 1;
+
+  /* Registers displayed via 'info float'.  */
+  if (group == float_reggroup)
+    return 0;
+
+  /* Registers that need to be saved/restored in order to
+     push or pop frames.  */
+  if (group == save_reggroup || group == restore_reggroup)
+    return 1;
+
+  return default_register_reggroup_p (gdbarch, regnum, group);
+}
+
+
+/* Decoding SPU instructions.  */
+
+enum
+  {
+    op_lqd   = 0x34,
+    op_lqx   = 0x3c4,
+    op_lqa   = 0x61,
+    op_lqr   = 0x67,
+    op_stqd  = 0x24,
+    op_stqx  = 0x144,
+    op_stqa  = 0x41,
+    op_stqr  = 0x47,
+
+    op_il    = 0x081,
+    op_ila   = 0x21,
+    op_a     = 0x0c0,
+    op_ai    = 0x1c,
+
+    op_selb  = 0x4,
+
+    op_br    = 0x64,
+    op_bra   = 0x60,
+    op_brsl  = 0x66,
+    op_brasl = 0x62,
+    op_brnz  = 0x42,
+    op_brz   = 0x40,
+    op_brhnz = 0x46,
+    op_brhz  = 0x44,
+    op_bi    = 0x1a8,
+    op_bisl  = 0x1a9,
+    op_biz   = 0x128,
+    op_binz  = 0x129,
+    op_bihz  = 0x12a,
+    op_bihnz = 0x12b,
+  };
+
+static int
+is_rr (unsigned int insn, int op, int *rt, int *ra, int *rb)
+{
+  if ((insn >> 21) == op)
+    {
+      *rt = insn & 127;
+      *ra = (insn >> 7) & 127;
+      *rb = (insn >> 14) & 127;
+      return 1;
+    }
+
+  return 0;
+}
+
+static int
+is_rrr (unsigned int insn, int op, int *rt, int *ra, int *rb, int *rc)
+{
+  if ((insn >> 28) == op)
+    {
+      *rt = (insn >> 21) & 127;
+      *ra = (insn >> 7) & 127;
+      *rb = (insn >> 14) & 127;
+      *rc = insn & 127;
+      return 1;
+    }
+
+  return 0;
+}
+
+static int
+is_ri7 (unsigned int insn, int op, int *rt, int *ra, int *i7)
+{
+  if ((insn >> 21) == op)
+    {
+      *rt = insn & 127;
+      *ra = (insn >> 7) & 127;
+      *i7 = (((insn >> 14) & 127) ^ 0x40) - 0x40;
+      return 1;
+    }
+
+  return 0;
+}
+
+static int
+is_ri10 (unsigned int insn, int op, int *rt, int *ra, int *i10)
+{
+  if ((insn >> 24) == op)
+    {
+      *rt = insn & 127;
+      *ra = (insn >> 7) & 127;
+      *i10 = (((insn >> 14) & 0x3ff) ^ 0x200) - 0x200;
+      return 1;
+    }
+
+  return 0;
+}
+
+static int
+is_ri16 (unsigned int insn, int op, int *rt, int *i16)
+{
+  if ((insn >> 23) == op)
+    {
+      *rt = insn & 127;
+      *i16 = (((insn >> 7) & 0xffff) ^ 0x8000) - 0x8000;
+      return 1;
+    }
+
+  return 0;
+}
+
+static int
+is_ri18 (unsigned int insn, int op, int *rt, int *i18)
+{
+  if ((insn >> 25) == op)
+    {
+      *rt = insn & 127;
+      *i18 = (((insn >> 7) & 0x3ffff) ^ 0x20000) - 0x20000;
+      return 1;
+    }
+
+  return 0;
+}
+
+static int
+is_branch (unsigned int insn, int *offset, int *reg)
+{
+  int rt, i7, i16;
+
+  if (is_ri16 (insn, op_br, &rt, &i16)
+      || is_ri16 (insn, op_brsl, &rt, &i16)
+      || is_ri16 (insn, op_brnz, &rt, &i16)
+      || is_ri16 (insn, op_brz, &rt, &i16)
+      || is_ri16 (insn, op_brhnz, &rt, &i16)
+      || is_ri16 (insn, op_brhz, &rt, &i16))
+    {
+      *reg = SPU_PC_REGNUM;
+      *offset = i16 << 2;
+      return 1;
+    }
+
+  if (is_ri16 (insn, op_bra, &rt, &i16)
+      || is_ri16 (insn, op_brasl, &rt, &i16))
+    {
+      *reg = -1;
+      *offset = i16 << 2;
+      return 1;
+    }
+
+  if (is_ri7 (insn, op_bi, &rt, reg, &i7)
+      || is_ri7 (insn, op_bisl, &rt, reg, &i7)
+      || is_ri7 (insn, op_biz, &rt, reg, &i7)
+      || is_ri7 (insn, op_binz, &rt, reg, &i7)
+      || is_ri7 (insn, op_bihz, &rt, reg, &i7)
+      || is_ri7 (insn, op_bihnz, &rt, reg, &i7))
+    {
+      *offset = 0;
+      return 1;
+    }
+
+  return 0;
+}
+
+
+/* Prolog parsing.  */
+
+struct spu_prologue_data
+  {
+    /* Stack frame size.  -1 if analysis was unsuccessful.  */
+    int size;
+
+    /* How to find the CFA.  The CFA is equal to SP at function entry.  */
+    int cfa_reg;
+    int cfa_offset;
+
+    /* Offset relative to CFA where a register is saved.  -1 if invalid.  */
+    int reg_offset[SPU_NUM_GPRS];
+  };
+
+static CORE_ADDR
+spu_analyze_prologue (CORE_ADDR start_pc, CORE_ADDR end_pc,
+                      struct spu_prologue_data *data)
+{
+  int found_sp = 0;
+  int found_fp = 0;
+  int found_lr = 0;
+  int reg_immed[SPU_NUM_GPRS];
+  gdb_byte buf[16];
+  CORE_ADDR prolog_pc = start_pc;
+  CORE_ADDR pc;
+  int i;
+
+
+  /* Initialize DATA to default values.  */
+  data->size = -1;
+
+  data->cfa_reg = SPU_RAW_SP_REGNUM;
+  data->cfa_offset = 0;
+
+  for (i = 0; i < SPU_NUM_GPRS; i++)
+    data->reg_offset[i] = -1;
+
+  /* Set up REG_IMMED array.  This is non-zero for a register if we know its
+     preferred slot currently holds this immediate value.  */
+  for (i = 0; i < SPU_NUM_GPRS; i++)
+      reg_immed[i] = 0;
+
+  /* Scan instructions until the first branch.
+
+     The following instructions are important prolog components:
+
+	- The first instruction to set up the stack pointer.
+	- The first instruction to set up the frame pointer.
+	- The first instruction to save the link register.
+
+     We return the instruction after the latest of these three,
+     or the incoming PC if none is found.  The first instruction
+     to set up the stack pointer also defines the frame size.
+
+     Note that instructions saving incoming arguments to their stack
+     slots are not counted as important, because they are hard to
+     identify with certainty.  This should not matter much, because
+     arguments are relevant only in code compiled with debug data,
+     and in such code the GDB core will advance until the first source
+     line anyway, using SAL data.
+
+     For purposes of stack unwinding, we analyze the following types
+     of instructions in addition:
+
+      - Any instruction adding to the current frame pointer.
+      - Any instruction loading an immediate constant into a register.
+      - Any instruction storing a register onto the stack.
+
+     These are used to compute the CFA and REG_OFFSET output.  */
+
+  for (pc = start_pc; pc < end_pc; pc += 4)
+    {
+      unsigned int insn;
+      int rt, ra, rb, rc, immed;
+
+      if (target_read_memory (pc, buf, 4))
+	break;
+      insn = extract_unsigned_integer (buf, 4);
+
+      /* AI is the typical instruction to set up a stack frame.
+         It is also used to initialize the frame pointer.  */
+      if (is_ri10 (insn, op_ai, &rt, &ra, &immed))
+	{
+	  if (rt == data->cfa_reg && ra == data->cfa_reg)
+	    data->cfa_offset -= immed;
+
+	  if (rt == SPU_RAW_SP_REGNUM && ra == SPU_RAW_SP_REGNUM
+	      && !found_sp)
+	    {
+	      found_sp = 1;
+	      prolog_pc = pc + 4;
+
+	      data->size = -immed;
+	    }
+	  else if (rt == SPU_FP_REGNUM && ra == SPU_RAW_SP_REGNUM
+		   && !found_fp)
+	    {
+	      found_fp = 1;
+	      prolog_pc = pc + 4;
+
+	      data->cfa_reg = SPU_FP_REGNUM;
+	      data->cfa_offset -= immed;
+	    }
+	}
+
+      /* A is used to set up stack frames of size >= 512 bytes.
+         If we have tracked the contents of the addend register,
+         we can handle this as well.  */
+      else if (is_rr (insn, op_a, &rt, &ra, &rb))
+	{
+	  if (rt == data->cfa_reg && ra == data->cfa_reg)
+	    {
+	      if (reg_immed[rb] != 0)
+		data->cfa_offset -= reg_immed[rb];
+	      else
+		data->cfa_reg = -1;  /* We don't know the CFA any more.  */
+	    }
+
+	  if (rt == SPU_RAW_SP_REGNUM && ra == SPU_RAW_SP_REGNUM
+	      && !found_sp)
+	    {
+	      found_sp = 1;
+	      prolog_pc = pc + 4;
+
+	      if (reg_immed[rb] != 0)
+		data->size = -reg_immed[rb];
+	    }
+	}
+
+      /* We need to track IL and ILA used to load immediate constants
+         in case they are later used as input to an A instruction.  */
+      else if (is_ri16 (insn, op_il, &rt, &immed))
+	{
+	  reg_immed[rt] = immed;
+	}
+
+      else if (is_ri18 (insn, op_ila, &rt, &immed))
+	{
+	  reg_immed[rt] = immed & 0x3ffff;
+	}
+
+      /* STQD is used to save registers to the stack.  */
+      else if (is_ri10 (insn, op_stqd, &rt, &ra, &immed))
+	{
+	  if (ra == data->cfa_reg)
+	    data->reg_offset[rt] = data->cfa_offset - (immed << 4);
+
+	  if (ra == data->cfa_reg && rt == SPU_LR_REGNUM
+              && !found_lr)
+	    {
+	      found_lr = 1;
+	      prolog_pc = pc + 4;
+	    }
+	}
+
+      /* _start uses SELB to set up the stack pointer.  */
+      else if (is_rrr (insn, op_selb, &rt, &ra, &rb, &rc))
+	{
+	  if (rt == SPU_RAW_SP_REGNUM && !found_sp)
+	    found_sp = 1;
+	}
+
+      /* We terminate if we find a branch.  */
+      else if (is_branch (insn, &immed, &ra))
+	break;
+    }
+
+
+  /* If we successfully parsed until here, and didn't find any instruction
+     modifying SP, we assume we have a frameless function.  */
+  if (!found_sp)
+    data->size = 0;
+
+  /* Return cooked instead of raw SP.  */
+  if (data->cfa_reg == SPU_RAW_SP_REGNUM)
+    data->cfa_reg = SPU_SP_REGNUM;
+
+  return prolog_pc;
+}
+
+/* Return the first instruction after the prologue starting at PC.  */
+static CORE_ADDR
+spu_skip_prologue (CORE_ADDR pc)
+{
+  struct spu_prologue_data data;
+  return spu_analyze_prologue (pc, (CORE_ADDR)-1, &data);
+}
+
+/* Return the frame pointer in use at address PC.  */
+static void
+spu_virtual_frame_pointer (CORE_ADDR pc, int *reg, LONGEST *offset)
+{
+  struct spu_prologue_data data;
+  spu_analyze_prologue (pc, (CORE_ADDR)-1, &data);
+
+  if (data.size != -1 && data.cfa_reg != -1)
+    {
+      /* The 'frame pointer' address is CFA minus frame size.  */
+      *reg = data.cfa_reg;
+      *offset = data.cfa_offset - data.size;
+    }
+  else
+    {
+      /* ??? We don't really know ... */
+      *reg = SPU_SP_REGNUM;
+      *offset = 0;
+    }
+}
+
+/* Normal stack frames.  */
+
+struct spu_unwind_cache
+{
+  CORE_ADDR func;
+  CORE_ADDR frame_base;
+  CORE_ADDR local_base;
+
+  struct trad_frame_saved_reg *saved_regs;
+};
+
+static struct spu_unwind_cache *
+spu_frame_unwind_cache (struct frame_info *next_frame,
+			void **this_prologue_cache)
+{
+  struct spu_unwind_cache *info;
+  struct spu_prologue_data data;
+
+  if (*this_prologue_cache)
+    return *this_prologue_cache;
+
+  info = FRAME_OBSTACK_ZALLOC (struct spu_unwind_cache);
+  *this_prologue_cache = info;
+  info->saved_regs = trad_frame_alloc_saved_regs (next_frame);
+  info->frame_base = 0;
+  info->local_base = 0;
+
+  /* Find the start of the current function, and analyze its prologue.  */
+  info->func = frame_func_unwind (next_frame);
+  if (info->func == 0)
+    {
+      /* Fall back to using the current PC as frame ID.  */
+      info->func = frame_pc_unwind (next_frame);
+      data.size = -1;
+    }
+  else
+    spu_analyze_prologue (info->func, frame_pc_unwind (next_frame), &data);
+
+
+  /* If successful, use prologue analysis data.  */
+  if (data.size != -1 && data.cfa_reg != -1)
+    {
+      CORE_ADDR cfa;
+      int i;
+      gdb_byte buf[16];
+
+      /* Determine CFA via unwound CFA_REG plus CFA_OFFSET.  */
+      frame_unwind_register (next_frame, data.cfa_reg, buf);
+      cfa = extract_unsigned_integer (buf, 4) + data.cfa_offset;
+
+      /* Call-saved register slots.  */
+      for (i = 0; i < SPU_NUM_GPRS; i++)
+	if (i == SPU_LR_REGNUM
+	    || (i >= SPU_SAVED1_REGNUM && i <= SPU_SAVEDN_REGNUM))
+	  if (data.reg_offset[i] != -1)
+	    info->saved_regs[i].addr = cfa - data.reg_offset[i];
+
+      /* The previous PC comes from the link register.  */
+      if (trad_frame_addr_p (info->saved_regs, SPU_LR_REGNUM))
+	info->saved_regs[SPU_PC_REGNUM] = info->saved_regs[SPU_LR_REGNUM];
+      else
+	info->saved_regs[SPU_PC_REGNUM].realreg = SPU_LR_REGNUM;
+
+      /* The previous SP is equal to the CFA.  */
+      trad_frame_set_value (info->saved_regs, SPU_SP_REGNUM, cfa);
+
+      /* Frame bases.  */
+      info->frame_base = cfa;
+      info->local_base = cfa - data.size;
+    }
+
+  /* Otherwise, fall back to reading the backchain link.  */
+  else
+    {
+      CORE_ADDR reg, backchain;
+
+      /* Get the backchain.  */
+      reg = frame_unwind_register_unsigned (next_frame, SPU_SP_REGNUM);
+      backchain = read_memory_unsigned_integer (reg, 4);
+
+      /* A zero backchain terminates the frame chain.  Also, sanity
+         check against the local store size limit.  */
+      if (backchain != 0 && backchain < SPU_LS_SIZE)
+	{
+	  /* Assume the link register is saved into its slot.  */
+	  if (backchain + 16 < SPU_LS_SIZE)
+	    info->saved_regs[SPU_LR_REGNUM].addr = backchain + 16;
+
+	  /* This will also be the previous PC.  */
+	  if (trad_frame_addr_p (info->saved_regs, SPU_LR_REGNUM))
+	    info->saved_regs[SPU_PC_REGNUM] = info->saved_regs[SPU_LR_REGNUM];
+	  else
+	    info->saved_regs[SPU_PC_REGNUM].realreg = SPU_LR_REGNUM;
+
+	  /* The previous SP will equal the backchain value.  */
+	  trad_frame_set_value (info->saved_regs, SPU_SP_REGNUM, backchain);
+
+          /* Frame bases.  */
+	  info->frame_base = backchain;
+	  info->local_base = reg;
+	}
+    }
+ 
+  return info;
+}
+
+static void
+spu_frame_this_id (struct frame_info *next_frame,
+		   void **this_prologue_cache, struct frame_id *this_id)
+{
+  struct spu_unwind_cache *info =
+    spu_frame_unwind_cache (next_frame, this_prologue_cache);
+
+  if (info->frame_base == 0)
+    return;
+
+  *this_id = frame_id_build (info->frame_base, info->func);
+}
+
+static void
+spu_frame_prev_register (struct frame_info *next_frame,
+			 void **this_prologue_cache,
+			 int regnum, int *optimizedp,
+			 enum lval_type *lvalp, CORE_ADDR * addrp,
+			 int *realnump, gdb_byte *bufferp)
+{
+  struct spu_unwind_cache *info
+    = spu_frame_unwind_cache (next_frame, this_prologue_cache);
+
+  /* Special-case the stack pointer.  */
+  if (regnum == SPU_RAW_SP_REGNUM)
+    regnum = SPU_SP_REGNUM;
+
+  trad_frame_get_prev_register (next_frame, info->saved_regs, regnum,
+				optimizedp, lvalp, addrp, realnump, bufferp);
+}
+
+static const struct frame_unwind spu_frame_unwind = {
+  NORMAL_FRAME,
+  spu_frame_this_id,
+  spu_frame_prev_register
+};
+
+const struct frame_unwind *
+spu_frame_sniffer (struct frame_info *next_frame)
+{
+  return &spu_frame_unwind;
+}
+
+static CORE_ADDR
+spu_frame_base_address (struct frame_info *next_frame, void **this_cache)
+{
+  struct spu_unwind_cache *info
+    = spu_frame_unwind_cache (next_frame, this_cache);
+  return info->local_base;
+}
+
+static const struct frame_base spu_frame_base = {
+  &spu_frame_unwind,
+  spu_frame_base_address,
+  spu_frame_base_address,
+  spu_frame_base_address
+};
+
+static CORE_ADDR
+spu_unwind_pc (struct gdbarch *gdbarch, struct frame_info *next_frame)
+{
+  return frame_unwind_register_unsigned (next_frame, SPU_PC_REGNUM);
+}
+
+static CORE_ADDR
+spu_unwind_sp (struct gdbarch *gdbarch, struct frame_info *next_frame)
+{
+  return frame_unwind_register_unsigned (next_frame, SPU_SP_REGNUM);
+}
+
+
+/* Function calling convention.  */
+
+static int
+spu_scalar_value_p (struct type *type)
+{
+  switch (TYPE_CODE (type))
+    {
+    case TYPE_CODE_INT:
+    case TYPE_CODE_ENUM:
+    case TYPE_CODE_RANGE:
+    case TYPE_CODE_CHAR:
+    case TYPE_CODE_BOOL:
+    case TYPE_CODE_PTR:
+    case TYPE_CODE_REF:
+      return TYPE_LENGTH (type) <= 16;
+
+    default:
+      return 0;
+    }
+}
+
+static void
+spu_value_to_regcache (struct regcache *regcache, int regnum,
+		       struct type *type, const gdb_byte *in)
+{
+  int len = TYPE_LENGTH (type);
+
+  if (spu_scalar_value_p (type))
+    {
+      int preferred_slot = len < 4 ? 4 - len : 0;
+      regcache_cooked_write_part (regcache, regnum, preferred_slot, len, in);
+    }
+  else
+    {
+      while (len >= 16)
+	{
+	  regcache_cooked_write (regcache, regnum++, in);
+	  in += 16;
+	  len -= 16;
+	}
+
+      if (len > 0)
+	regcache_cooked_write_part (regcache, regnum, 0, len, in);
+    }
+}
+
+static void
+spu_regcache_to_value (struct regcache *regcache, int regnum,
+		       struct type *type, gdb_byte *out)
+{
+  int len = TYPE_LENGTH (type);
+
+  if (spu_scalar_value_p (type))
+    {
+      int preferred_slot = len < 4 ? 4 - len : 0;
+      regcache_cooked_read_part (regcache, regnum, preferred_slot, len, out);
+    }
+  else
+    {
+      while (len >= 16)
+	{
+	  regcache_cooked_read (regcache, regnum++, out);
+	  out += 16;
+	  len -= 16;
+	}
+
+      if (len > 0)
+	regcache_cooked_read_part (regcache, regnum, 0, len, out);
+    }
+}
+
+static CORE_ADDR
+spu_push_dummy_call (struct gdbarch *gdbarch, struct value *function,
+		     struct regcache *regcache, CORE_ADDR bp_addr,
+		     int nargs, struct value **args, CORE_ADDR sp,
+		     int struct_return, CORE_ADDR struct_addr)
+{
+  int i;
+  int regnum = SPU_ARG1_REGNUM;
+  int stack_arg = -1;
+  gdb_byte buf[16];
+
+  /* Set the return address.  */
+  memset (buf, 0, sizeof buf);
+  store_unsigned_integer (buf, 4, bp_addr);
+  regcache_cooked_write (regcache, SPU_LR_REGNUM, buf);
+
+  /* If STRUCT_RETURN is true, then the struct return address (in
+     STRUCT_ADDR) will consume the first argument-passing register.
+     Both adjust the register count and store that value.  */
+  if (struct_return)
+    {
+      memset (buf, 0, sizeof buf);
+      store_unsigned_integer (buf, 4, struct_addr);
+      regcache_cooked_write (regcache, regnum++, buf);
+    }
+
+  /* Fill in argument registers.  */
+  for (i = 0; i < nargs; i++)
+    {
+      struct value *arg = args[i];
+      struct type *type = check_typedef (value_type (arg));
+      const gdb_byte *contents = value_contents (arg);
+      int len = TYPE_LENGTH (type);
+      int n_regs = align_up (len, 16) / 16;
+
+      /* If the argument doesn't wholly fit into registers, it and
+	 all subsequent arguments go to the stack.  */
+      if (regnum + n_regs - 1 > SPU_ARGN_REGNUM)
+	{
+	  stack_arg = i;
+	  break;
+	}
+
+      spu_value_to_regcache (regcache, regnum, type, contents);
+      regnum += n_regs;
+    }
+
+  /* Overflow arguments go to the stack.  */
+  if (stack_arg != -1)
+    {
+      CORE_ADDR ap;
+
+      /* Allocate all required stack size.  */
+      for (i = stack_arg; i < nargs; i++)
+	{
+	  struct type *type = check_typedef (value_type (args[i]));
+	  sp -= align_up (TYPE_LENGTH (type), 16);
+	}
+
+      /* Fill in stack arguments.  */
+      ap = sp;
+      for (i = stack_arg; i < nargs; i++)
+	{
+	  struct value *arg = args[i];
+	  struct type *type = check_typedef (value_type (arg));
+	  int len = TYPE_LENGTH (type);
+	  int preferred_slot;
+	  
+	  if (spu_scalar_value_p (type))
+	    preferred_slot = len < 4 ? 4 - len : 0;
+	  else
+	    preferred_slot = 0;
+
+	  target_write_memory (ap + preferred_slot, value_contents (arg), len);
+	  ap += align_up (TYPE_LENGTH (type), 16);
+	}
+    }
+
+  /* Allocate stack frame header.  */
+  sp -= 32;
+
+  /* Finally, update the SP register.  */
+  regcache_cooked_write_unsigned (regcache, SPU_SP_REGNUM, sp);
+
+  return sp;
+}
+
+static struct frame_id
+spu_unwind_dummy_id (struct gdbarch *gdbarch, struct frame_info *next_frame)
+{
+  return frame_id_build (spu_unwind_sp (gdbarch, next_frame),
+			 spu_unwind_pc (gdbarch, next_frame));
+}
+
+/* Function return value access.  */
+
+static enum return_value_convention
+spu_return_value (struct gdbarch *gdbarch, struct type *type,
+                  struct regcache *regcache, gdb_byte *out, const gdb_byte *in)
+{
+  enum return_value_convention rvc;
+
+  if (TYPE_LENGTH (type) <= (SPU_ARGN_REGNUM - SPU_ARG1_REGNUM + 1) * 16)
+    rvc = RETURN_VALUE_REGISTER_CONVENTION;
+  else
+    rvc = RETURN_VALUE_STRUCT_CONVENTION;
+
+  if (in)
+    {
+      switch (rvc)
+	{
+	case RETURN_VALUE_REGISTER_CONVENTION:
+	  spu_value_to_regcache (regcache, SPU_ARG1_REGNUM, type, in);
+	  break;
+
+	case RETURN_VALUE_STRUCT_CONVENTION:
+	  error ("Cannot set function return value.");
+	  break;
+	}
+    }
+  else if (out)
+    {
+      switch (rvc)
+	{
+	case RETURN_VALUE_REGISTER_CONVENTION:
+	  spu_regcache_to_value (regcache, SPU_ARG1_REGNUM, type, out);
+	  break;
+
+	case RETURN_VALUE_STRUCT_CONVENTION:
+	  error ("Function return value unknown.");
+	  break;
+	}
+    }
+
+  return rvc;
+}
+
+
+/* Breakpoints.  */
+
+static const gdb_byte *
+spu_breakpoint_from_pc (CORE_ADDR * pcptr, int *lenptr)
+{
+  static const gdb_byte breakpoint[] = { 0x00, 0x00, 0x3f, 0xff };
+
+  *lenptr = sizeof breakpoint;
+  return breakpoint;
+}
+
+
+/* Software single-stepping support.  */
+
+void
+spu_software_single_step (enum target_signal signal, int insert_breakpoints_p)
+{
+  if (insert_breakpoints_p)
+    {
+      CORE_ADDR pc, next_pc;
+      unsigned int insn;
+      int offset, reg;
+      gdb_byte buf[4];
+
+      regcache_cooked_read (current_regcache, SPU_PC_REGNUM, buf);
+      pc = extract_unsigned_integer (buf, 4);
+
+      if (target_read_memory (pc, buf, 4))
+	return;
+      insn = extract_unsigned_integer (buf, 4);
+
+       /* Next sequential instruction is at PC + 4, except if the current
+	  instruction is a PPE-assisted call, in which case it is at PC + 8.
+	  Wrap around LS limit to be on the safe side.  */
+      if ((insn & 0xffffff00) == 0x00002100)
+	next_pc = (pc + 8) & (SPU_LS_SIZE - 1) & -4;
+      else
+	next_pc = (pc + 4) & (SPU_LS_SIZE - 1) & -4;
+
+      insert_single_step_breakpoint (next_pc);
+
+      if (is_branch (insn, &offset, &reg))
+	{
+	  CORE_ADDR target = offset;
+
+	  if (reg == SPU_PC_REGNUM)
+	    target += pc;
+	  else if (reg != -1)
+	    {
+	      regcache_cooked_read_part (current_regcache, reg, 0, 4, buf);
+	      target += extract_unsigned_integer (buf, 4);
+	    }
+
+	  target = target & (SPU_LS_SIZE - 1) & -4;
+	  if (target != next_pc)
+	    insert_single_step_breakpoint (target);
+	}
+    }
+  else
+    remove_single_step_breakpoints ();
+}
+
+
+/* Set up gdbarch struct.  */
+
+static struct gdbarch *
+spu_gdbarch_init (struct gdbarch_info info, struct gdbarch_list *arches)
+{
+  struct gdbarch *gdbarch;
+
+  /* Find a candidate among the list of pre-declared architectures.  */
+  arches = gdbarch_list_lookup_by_info (arches, &info);
+  if (arches != NULL)
+    return arches->gdbarch;
+
+  /* Is is for us?  */
+  if (info.bfd_arch_info->mach != bfd_mach_spu)
+    return NULL;
+
+  /* Yes, create a new architecture.  */
+  gdbarch = gdbarch_alloc (&info, NULL);
+
+  /* Disassembler.  */
+  set_gdbarch_print_insn (gdbarch, print_insn_spu);
+
+  /* Registers.  */
+  set_gdbarch_num_regs (gdbarch, SPU_NUM_REGS);
+  set_gdbarch_num_pseudo_regs (gdbarch, SPU_NUM_PSEUDO_REGS);
+  set_gdbarch_sp_regnum (gdbarch, SPU_SP_REGNUM);
+  set_gdbarch_pc_regnum (gdbarch, SPU_PC_REGNUM);
+  set_gdbarch_register_name (gdbarch, spu_register_name);
+  set_gdbarch_register_type (gdbarch, spu_register_type);
+  set_gdbarch_pseudo_register_read (gdbarch, spu_pseudo_register_read);
+  set_gdbarch_pseudo_register_write (gdbarch, spu_pseudo_register_write);
+  set_gdbarch_convert_register_p (gdbarch, spu_convert_register_p);
+  set_gdbarch_register_to_value (gdbarch, spu_register_to_value);
+  set_gdbarch_value_to_register (gdbarch, spu_value_to_register);
+  set_gdbarch_register_reggroup_p (gdbarch, spu_register_reggroup_p);
+
+  /* Data types.  */
+  set_gdbarch_char_signed (gdbarch, 0);
+  set_gdbarch_ptr_bit (gdbarch, 32);
+  set_gdbarch_addr_bit (gdbarch, 32);
+  set_gdbarch_short_bit (gdbarch, 16);
+  set_gdbarch_int_bit (gdbarch, 32);
+  set_gdbarch_long_bit (gdbarch, 32);
+  set_gdbarch_long_long_bit (gdbarch, 64);
+  set_gdbarch_float_bit (gdbarch, 32);
+  set_gdbarch_double_bit (gdbarch, 64);
+  set_gdbarch_long_double_bit (gdbarch, 64);
+  set_gdbarch_float_format (gdbarch, &floatformat_ieee_single_big);
+  set_gdbarch_double_format (gdbarch, &floatformat_ieee_double_big);
+  set_gdbarch_long_double_format (gdbarch, &floatformat_ieee_double_big);
+
+  /* Inferior function calls.  */
+  set_gdbarch_push_dummy_call (gdbarch, spu_push_dummy_call);
+  set_gdbarch_unwind_dummy_id (gdbarch, spu_unwind_dummy_id);
+  set_gdbarch_return_value (gdbarch, spu_return_value);
+
+  /* Frame handling.  */
+  set_gdbarch_inner_than (gdbarch, core_addr_lessthan);
+  frame_unwind_append_sniffer (gdbarch, spu_frame_sniffer);
+  frame_base_set_default (gdbarch, &spu_frame_base);
+  set_gdbarch_unwind_pc (gdbarch, spu_unwind_pc);
+  set_gdbarch_unwind_sp (gdbarch, spu_unwind_sp);
+  set_gdbarch_virtual_frame_pointer (gdbarch, spu_virtual_frame_pointer);
+  set_gdbarch_frame_args_skip (gdbarch, 0);
+  set_gdbarch_skip_prologue (gdbarch, spu_skip_prologue);
+
+  /* Breakpoints.  */
+  set_gdbarch_decr_pc_after_break (gdbarch, 4);
+  set_gdbarch_breakpoint_from_pc (gdbarch, spu_breakpoint_from_pc);
+  set_gdbarch_cannot_step_breakpoint (gdbarch, 1);
+  set_gdbarch_software_single_step (gdbarch, spu_software_single_step);
+
+  return gdbarch;
+}
+
+void
+_initialize_spu_tdep (void)
+{
+  register_gdbarch_init (bfd_arch_spu, spu_gdbarch_init);
+}
diff -urN gdb-orig/gdb/spu-tdep.h gdb-head/gdb/spu-tdep.h
--- gdb-orig/gdb/spu-tdep.h	1970-01-01 01:00:00.000000000 +0100
+++ gdb-head/gdb/spu-tdep.h	2006-11-16 04:16:15.000000000 +0100
@@ -0,0 +1,50 @@
+/* SPU target-dependent code for GDB, the GNU debugger.
+   Copyright (C) 2006 Free Software Foundation, Inc.
+
+   This file is part of GDB.
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 2 of the License, or
+   (at your option) any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program; if not, write to the Free Software
+   Foundation, Inc., 51 Franklin Street, Fifth Floor,
+   Boston, MA 02110-1301, USA.  */
+
+#ifndef SPU_TDEP_H
+#define SPU_TDEP_H
+
+/* Number of registers.  */
+#define SPU_NUM_REGS         130
+#define SPU_NUM_PSEUDO_REGS  1
+#define SPU_NUM_GPRS	     128
+
+/* Register numbers of various important registers.  */
+enum spu_regnum
+{
+  /* SPU calling convention.  */
+  SPU_LR_REGNUM = 0,		/* Link register.  */
+  SPU_RAW_SP_REGNUM = 1,	/* Stack pointer (full register).  */
+  SPU_ARG1_REGNUM = 3,		/* First argument register.  */
+  SPU_ARGN_REGNUM = 74,		/* Last argument register.  */
+  SPU_SAVED1_REGNUM = 80,	/* First call-saved register.  */
+  SPU_SAVEDN_REGNUM = 127,	/* Last call-saved register.  */
+  SPU_FP_REGNUM = 127,		/* Frame pointer.  */
+
+  /* Special registers.  */
+  SPU_ID_REGNUM = 128,		/* SPU ID register.  */
+  SPU_PC_REGNUM = 129,		/* Next program counter.  */
+  SPU_SP_REGNUM = 130		/* Stack pointer (preferred slot).  */
+};
+
+/* Local store.  */
+#define SPU_LS_SIZE          0x40000
+
+#endif


-- 
  Dr. Ulrich Weigand
  GNU Toolchain for Linux on System z and Cell BE
  Ulrich.Weigand@de.ibm.com


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]