[patch] Fix i386 memory-by-register access on amd64

[Jan Kratochvil]

On Tue, 07 Jul 2009 18:24:06 +0200, Ulrich Weigand wrote:
> target_thread_architecture is wrong for this purpose;
+
> This should be done inside the TARGET_OBJECT_MEMORY case;

Fixed in the patch.


> (The assert seems superfluous to me; "offset" is a local variable to this
> function, so we should know its type already.  Other code in this function
> would already fail if offset were of any other type.)

Originally I thought OFFSET should be later changed to CORE_ADDR.  Now I see
for other `enum target_object' larger OFFSET may make sense.


On Tue, 07 Jul 2009 20:43:18 +0200, Mark Kettenis wrote:
> > Date: Tue, 7 Jul 2009 20:22:04 +0200
> > From: Jan Kratochvil <jan.kratochvil@redhat.com>
> > 
> > On Tue, 07 Jul 2009 19:59:43 +0200, Mark Kettenis wrote:
> > > But then I don't understand Jan's diff at all.  Linux has its own
> > > implementation for TARGET_OBJECT_MEMORY in linux-nat.c.  Why isn't
> > > that one used?
> > 
> > Expecting the same problem must affect even non-Linux ptrace usage.  I will
> > move it to linux-nat.c if you think so.
> 
> I suppose I should have been more specific.  You are trying to fix a
> bug that you see on Linux isn't it?

Yes.

> And this fix should only be relevant for TARGET_OBJECT_MEMORY isn't
> it?

Yes.  I was not sure before and I though it would not hurt other
`enum target_object's but - you are right, yes.


> But Linux has its own implementation for doing TARGET_OBJECT_MEMORY
> xfers in linux_proc_xfer_partial.

I did not much notice linux_proc_xfer_partial role in this bug, thanks.

* I was not able to reproduce the problem for linux_proc_xfer_partial without
  removing there:
    /* Don't bother for one word.  */
    if (len < 3 * sizeof (long))
      return 0;
  because:
   * x/gx is the largest size - 8 bytes - which is still smaller.
   * x/NUMx generates NUM small transfers (not one NUM*8 transfer).
   * These already truncate the address to gdbarch_addr_bit in the caller:
     * dump memory
     * print *(struct large *) $ebx

* inf_ptrace_xfer_partial gets silently used as a fallback when
  linux_proc_xfer_partial fails.

  * linux_proc_xfer_partial has been broken so far even on native 32bit GDB
    debugging 32bit inferior with GDB built using --enable-64-bit-bfd (and
    thus having 64-bit CORE_ADDR).  Due to the silent fallback just nobody has
    noticed it.

Therefore fixed linux_xfer_partial.  Therefore it no longer needs the fix to
be present also in inf_ptrace_xfer_partial.  I do not know how non-Linux OSes
handle the debugging of 32bit inferior on 64bit GDB so I have no opinion
whether the inf_ptrace_xfer_partial patch makes sense, i can drop it.


> So I don't understand how this change fixes anything on Linux.

With my former patch linux_proc_xfer_partial failed and the fix in the ptrace
backend did handle the transfer.


Thanks,
Jan


2009-07-08  Jan Kratochvil  <jan.kratochvil@redhat.com>

	Fix memory access from signed 32bit inferior registers on 64bit GDB.
	* inf-ptrace.c (inf_ptrace_xfer_partial <TARGET_OBJECT_MEMORY>): New
	variable addr_bit.  Mask OFFSET by the ADDR_BIT width.
	* linux-nat.c (linux_xfer_partial <TARGET_OBJECT_MEMORY>): Likewise.

2009-07-08  Jan Kratochvil  <jan.kratochvil@redhat.com>

	* gdb.arch/amd64-i386-address.exp, gdb.arch/amd64-i386-address.S: New.

--- a/gdb/inf-ptrace.c
+++ b/gdb/inf-ptrace.c
@@ -457,6 +457,19 @@ inf_ptrace_xfer_partial (struct target_ops *ops, enum target_object object,
   switch (object)
     {
     case TARGET_OBJECT_MEMORY:
+      {
+	int addr_bit = gdbarch_addr_bit (target_gdbarch);
+
+	/* GDB calculates all the addresses in possibly larget width of the
+	   address.  Address width needs to be masked before its final use
+	   - either by linux_proc_xfer_partial or inf_ptrace_xfer_partial.
+
+	   Compare ADDR_BIT first to avoid a compiler warning on shift
+	   overflow.  */
+
+	if (addr_bit < (sizeof (ULONGEST) * HOST_CHAR_BIT))
+	  offset &= ((ULONGEST) 1 << addr_bit) - 1;
+      }
 #ifdef PT_IO
       /* OpenBSD 3.1, NetBSD 1.6 and FreeBSD 5.0 have a new PT_IO
 	 request that promises to be much more efficient in reading
--- a/gdb/linux-nat.c
+++ b/gdb/linux-nat.c
@@ -4260,6 +4260,20 @@ linux_xfer_partial (struct target_ops *ops, enum target_object object,
     return linux_nat_xfer_osdata (ops, object, annex, readbuf, writebuf,
                                offset, len);
 
+  /* GDB calculates all the addresses in possibly larget width of the address.
+     Address width needs to be masked before its final use - either by
+     linux_proc_xfer_partial or inf_ptrace_xfer_partial.
+
+     Compare ADDR_BIT first to avoid a compiler warning on shift overflow.  */
+
+  if (object == TARGET_OBJECT_MEMORY)
+    {
+      int addr_bit = gdbarch_addr_bit (target_gdbarch);
+
+      if (addr_bit < (sizeof (ULONGEST) * HOST_CHAR_BIT))
+	offset &= ((ULONGEST) 1 << addr_bit) - 1;
+    }
+
   xfer = linux_proc_xfer_partial (ops, object, annex, readbuf, writebuf,
 				  offset, len);
   if (xfer != 0)
--- /dev/null
+++ b/gdb/testsuite/gdb.arch/amd64-i386-address.S
@@ -0,0 +1,24 @@
+/* Copyright 2009 Free Software Foundation, Inc.
+
+   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 3 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, see <http://www.gnu.org/licenses/>.
+
+   This file is part of the gdb testsuite.  */
+
+_start:	.globl	_start
+	movl	$0xdeadf00d, %eax
+	pushl	%eax
+	movl	%esp, %ebx
+	int3
+	nop
+	nop
--- /dev/null
+++ b/gdb/testsuite/gdb.arch/amd64-i386-address.exp
@@ -0,0 +1,43 @@
+# Copyright 2009 Free Software Foundation, Inc.
+
+# 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 3 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, see <http://www.gnu.org/licenses/>.
+
+# This file is part of the gdb testsuite.
+
+# Test UNsigned extension of the 32-bit inferior address on a 64-bit host.
+# On native 32-bit host the test always PASSed.
+
+if {![istarget "x86_64-*-*"] && ![istarget "i?86-*-*"]} then {
+    verbose "Skipping amd64->i386 adress test."
+    return
+}
+
+if [prepare_for_testing amd64-i386-address.exp amd64-i386-address amd64-i386-address.S [list debug "additional_flags=-m32 -nostdlib"]] {
+    return -1
+}
+
+gdb_run_cmd
+
+set test "trap stop"
+gdb_test_multiple "" $test {
+    -re "Program received signal SIGTRAP,.*_start .*$gdb_prompt $" {
+	pass $test
+    }
+}
+
+gdb_test "x/wx \$esp" "0x\[0-9a-f\]*:\t0xdeadf00d"
+
+# Failure case would be:
+# 	0xff8d7f00:     Cannot access memory at address 0xff8d7f00
+gdb_test "x/wx \$ebx" "0x\[0-9a-f\]*:\t0xdeadf00d"