PR gdb/15871: Unavailable entry value is not shown correctly

In entry-values.exp, we have a test where the entry value of 'j' is
unavailable, so it is expected that printing j@entry yields
"<unavailable>".  However, the actual output is:

 (gdb) frame
 #0  0x0000000000400540 in foo (i=0, i@entry=2, j=2, j@entry=<error reading variable: Cannot access memory at address 0x6009e8>)

The error is thrown here:

#0  throw_it (reason=RETURN_ERROR, error=MEMORY_ERROR, fmt=0x8cd550 "Cannot access memory at address %s", ap=0x7fffffffc8e8) at ../../src/gdb/exceptions.c:373
#1  0x00000000005e2f9c in throw_error (error=MEMORY_ERROR, fmt=0x8cd550 "Cannot access memory at address %s") at ../../src/gdb/exceptions.c:422
#2  0x0000000000673a5f in memory_error (status=5, memaddr=6293992) at ../../src/gdb/corefile.c:204
#3  0x0000000000673aea in read_memory (memaddr=6293992, myaddr=0x7fffffffca60 "\200\316\377\377\377\177", len=4) at ../../src/gdb/corefile.c:223
#4  0x00000000006784d1 in dwarf_expr_read_mem (baton=0x7fffffffcd50, buf=0x7fffffffca60 "\200\316\377\377\377\177", addr=6293992, len=4) at ../../src/gdb/dwarf2loc.c:334
#5  0x000000000067645e in execute_stack_op (ctx=0x1409480, op_ptr=0x7fffffffce87 "\237<\005@", op_end=0x7fffffffce88 "<\005@") at ../../src/gdb/dwarf2expr.c:1045
#6  0x0000000000674e29 in dwarf_expr_eval (ctx=0x1409480, addr=0x7fffffffce80 "\003\350\t`", len=8) at ../../src/gdb/dwarf2expr.c:364
#7  0x000000000067c5b2 in dwarf2_evaluate_loc_desc_full (type=0x10876d0, frame=0xd8ecc0, data=0x7fffffffce80 "\003\350\t`", size=8, per_cu=0xf24c40, byte_offset=0)
    at ../../src/gdb/dwarf2loc.c:2236
#8  0x000000000067cc65 in dwarf2_evaluate_loc_desc (type=0x10876d0, frame=0xd8ecc0, data=0x7fffffffce80 "\003\350\t`", size=8, per_cu=0xf24c40)
    at ../../src/gdb/dwarf2loc.c:2407
#9  0x000000000067a5d4 in dwarf_entry_parameter_to_value (parameter=0x13a7960, deref_size=18446744073709551615, type=0x10876d0, caller_frame=0xd8ecc0, per_cu=0xf24c40)
    at ../../src/gdb/dwarf2loc.c:1160
#10 0x000000000067a962 in value_of_dwarf_reg_entry (type=0x10876d0, frame=0xd8de70, kind=CALL_SITE_PARAMETER_DWARF_REG, kind_u=...) at ../../src/gdb/dwarf2loc.c:1310
#11 0x000000000067aaca in value_of_dwarf_block_entry (type=0x10876d0, frame=0xd8de70, block=0xf1c2d4 "Q", block_len=1) at ../../src/gdb/dwarf2loc.c:1363
#12 0x000000000067e7c9 in locexpr_read_variable_at_entry (symbol=0x13a7540, frame=0xd8de70) at ../../src/gdb/dwarf2loc.c:3326
#13 0x00000000005daab6 in read_frame_arg (sym=0x13a7540, frame=0xd8de70, argp=0x7fffffffd0e0, entryargp=0x7fffffffd100) at ../../src/gdb/stack.c:362
#14 0x00000000005db384 in print_frame_args (func=0x13a7470, frame=0xd8de70, num=-1, stream=0xea3890) at ../../src/gdb/stack.c:669
#15 0x00000000005dc338 in print_frame (frame=0xd8de70, print_level=1, print_what=SRC_AND_LOC, print_args=1, sal=...) at ../../src/gdb/stack.c:1199
#16 0x00000000005db8ee in print_frame_info (frame=0xd8de70, print_level=1, print_what=SRC_AND_LOC, print_args=1) at ../../src/gdb/stack.c:851
#17 0x00000000005da2bb in print_stack_frame (frame=0xd8de70, print_level=1, print_what=SRC_AND_LOC) at ../../src/gdb/stack.c:169
#18 0x00000000005de236 in frame_command (level_exp=0x0, from_tty=1) at ../../src/gdb/stack.c:2265

dwarf2_evaluate_loc_desc_full (frame #7) knows to handle
NOT_AVAILABLE_ERROR errors, but read_memory always throws
a generic error.

Presently, only the value machinery knows to handle unavailable
memory.  We need to push the awareness down to the target_xfer layer,
making it return a finer grained error indication.  We can only return
a generic -1 nowadays, which leaves the upper layers with no clue on
why the xfer failed.  Use target_xfer_partial directly, rather than
propagating the error through target_read_memory so as to get a better
address to display in the error message.

(target_read_memory & friends build on top of target_read (thus the
target_xfer machinery), but turn all errors to EIO, an errno value.  I
think this is a mistake, and we'd better convert all these to return a
target_xfer_error too, but that can be done separately.  I looked
around a bit over memory_error calls, and the need to handle random
errno values, other than the EIOs gdb itself hardcodes, probably comes
(only) from deprecated_xfer_memory, which uses errno for error
indication, but I didn't look exhaustively.  We should really get rid
of deprecated_xfer_memory and of passing down errno values as error
indication in target_read & friends methods).

Tested on x86_64 Fedora 17, native and gdbserver.  Fixes the test in
the PR, which will be added to the testsuite later.

gdb/
2013-08-22  Pedro Alves  <palves@redhat.com>

	PR gdb/15871
	* corefile.c (target_xfer_memory_error): New function.
	(memory_error): Defer EIO to target_memory_error.
	(read_memory): Use target_xfer_partial, and handle finer-grained
	target xfer errors.
	* target.c (target_xfer_error_to_string): New function.
	(memory_xfer_partial_1): If memory is known to be
	unavailable, return TARGET_XFER_E_UNAVAILABLE instead of -1.
	(target_xfer_partial): Make extern.
	* target.h (enum target_xfer_error): New enum.
	(target_xfer_error_to_string): Declare function.
	(target_xfer_partial): Declare function.
	(struct target_ops) <xfer_partial>: Adjust describing comment.

gdb/target.c

@@ -81,12 +81,6 @@ static LONGEST current_xfer_partial (struct target_ops *ops,
 				     const gdb_byte *writebuf,
 				     ULONGEST offset, LONGEST len);
 
-static LONGEST target_xfer_partial (struct target_ops *ops,
-				    enum target_object object,
-				    const char *annex,
-				    void *readbuf, const void *writebuf,
-				    ULONGEST offset, LONGEST len);
-
 static struct gdbarch *default_thread_architecture (struct target_ops *ops,
 						    ptid_t ptid);
 
@@ -1238,6 +1232,21 @@ target_translate_tls_address (struct objfile *objfile, CORE_ADDR offset)
   return addr;
 }
 
+const char *
+target_xfer_error_to_string (enum target_xfer_error err)
+{
+#define CASE(X) case X: return #X
+  switch (err)
+    {
+      CASE(TARGET_XFER_E_IO);
+      CASE(TARGET_XFER_E_UNAVAILABLE);
+    default:
+      return "<unknown>";
+    }
+#undef CASE
+};
+
+
 #undef	MIN
 #define MIN(A, B) (((A) <= (B)) ? (A) : (B))
 
@@ -1523,7 +1532,7 @@ memory_xfer_partial_1 (struct target_ops *ops, enum target_object object,
 
 	      /* No use trying further, we know some memory starting
 		 at MEMADDR isn't available.  */
-	      return -1;
+	      return TARGET_XFER_E_UNAVAILABLE;
 	    }
 
 	  /* Don't try to read more than how much is available, in
@@ -1700,7 +1709,7 @@ make_show_memory_breakpoints_cleanup (int show)
 
 /* For docs see target.h, to_xfer_partial.  */
 
-static LONGEST
+LONGEST
 target_xfer_partial (struct target_ops *ops,
 		     enum target_object object, const char *annex,
 		     void *readbuf, const void *writebuf,