Explicitly mark vtables as code space

Ports for Hardvard architectures will typically have in their
pointer_to_address hook a check for TYPE_CODE_SPACE in their
"pointer_to_address" gdbarch method.  E.g., rl78's:

  /* Is it a code address?  */
  if (TYPE_CODE (TYPE_TARGET_TYPE (type)) == TYPE_CODE_FUNC
      || TYPE_CODE (TYPE_TARGET_TYPE (type)) == TYPE_CODE_METHOD
      || TYPE_CODE_SPACE (TYPE_TARGET_TYPE (type))
      || TYPE_LENGTH (type) == 4)
    return rl78_make_instruction_address (addr);
  else
    return rl78_make_data_address (addr);

The avr port is similar.

The vtable type is a struct type gdb itself bakes.  Being neither a
function, nor a method, and absent explicit flagging as residing in
code space, ends up being considered data.

This patch marks the type as code when it is created, on the theory
that this is needed for all Hardvard architectures.  I believe this
should make no difference on archs with flat address space.  Testing
on x86-64 GNU/Linux shows no changes.

gdb/
2014-02-12  Pedro Alves  <palves@redhat.com>
	    Kevin Buettner <kevinb@redhat.com>

	* gnu-v3-abi.c (build_gdb_vtable_type): Return a type marked with
	TYPE_INSTANCE_FLAG_CODE_SPACE.

Kevin Buettner, at
<https://sourceware.org/ml/gdb-patches/2014-02/msg00338.html>, writes,
re. rl78:

This patch, for rl78 using the default multilib, fixes 5 failures in
gdb.cp/casts.exp, 2 failures in gdb.cp/class2.exp, 115 failures in
gdb.mi/mi-var-rtti.exp, and 2 failures in gdb.python/py-value.exp.

It introduces 9 failures (regressions) in gdb.mi/mi-var-rtti.exp.

One of the regressions is:

 FAIL: gdb.mi/mi-var-rtti.exp: list children of s.public in type_update_when_use_rtti

The relevant lines from the log file from a pre-patch test run are as
follows:

 -var-list-children  S.public
 ^done,numchild="1",children=[child={name="S.public.ptr",exp="ptr",numchild="1",type="Base *",thread-id="1"}],has_more="0"
 (gdb)
 PASS: gdb.mi/mi-var-rtti.exp: list children of s.public in type_update_when_use_rtti
 Expecting: \^done,numchild=".*",children=\[child={name="S.public.ptr.public",exp="public",numchild="1"(,thread-id="[0-9]+")?}.*\],has_more="0"
 Expecting: ^(-var-list-children  S\.public\.ptr  [
 ]+)?(\^done,numchild=".*",children=\[child={name="S.public.ptr.public",exp="public",numchild="1"(,thread-id="[0-9]+")?}.*\],has_more="0"[
 ]+[(]gdb[)]
 [ ]*)

The same set of lines for the failing (post-patch) run are instead:

 -var-list-children  S.public
 &"warning: can't find linker symbol for virtual table for `Base' value\n"
 &"warning:   found `typeinfo for __cxxabiv1::__vmi_class_type_info' instead\n"
 &"warning: can't find linker symbol for virtual table for `Base' value\n"
 &"warning:   found `typeinfo for __cxxabiv1::__vmi_class_type_info' instead\n"
 &"warning: can't find linker symbol for virtual table for `Base' value\n"
 &"warning:   found `typeinfo for __cxxabiv1::__vmi_class_type_info' instead\n"
 ^done,numchild="1",children=[child={name="S.public.ptr",exp="ptr",numchild="1",type="Base *",thread-id="1"}],has_more="0"
 (gdb)
 FAIL: gdb.mi/mi-var-rtti.exp: list children of s.public in type_update_when_use_rtti
 Expecting: \^done,numchild=".*",children=\[child={name="S.public.ptr.public",exp="public",numchild="1"(,thread-id="[0-9]+")?}.*\],has_more="0"
 Expecting: ^(-var-list-children  S\.public\.ptr  [
 ]+)?(\^done,numchild=".*",children=\[child={name="S.public.ptr.public",exp="public",numchild="1"(,thread-id="[0-9]+")?}.*\],has_more="0"[
 ]+[(]gdb[)]
 [ ]*)

Note that the difference between these are the warnings regarding
linker symbols.  Aside from the warnings, the result is the same.
I.e.  gdb produces the correct answer despite the warnings.  The
reason for the other 8 failures is the same: the test harness is not
expecting these warnings.

I spent some time (a while ago) looking at the reason for these
warnings.  As I recall, we are now getting further along in the type
resolution process than we were without my patch.  I.e.  for the
example above, the code in question never got to the point where it
was looking for the specified linker symbol.  So it seems to me that,
at worst, my patch exposes some other problem, but is not directly the
cause of the problem.

diff --git a/gdb/ChangeLog b/gdb/ChangeLog
index 5e9e9b8c6f4845c72701d2cef8a13a003aa9f4eb..41397b21e63095ae708e064cfa8c43aa69a94f14 100644 (file)
--- a/gdb/ChangeLog
+++ b/gdb/ChangeLog
@@ -1,3 +1,9 @@
+2014-02-12  Pedro Alves  <palves@redhat.com>
+           Kevin Buettner <kevinb@redhat.com>
+
+       * gnu-v3-abi.c (build_gdb_vtable_type): Return a type marked with
+       TYPE_INSTANCE_FLAG_CODE_SPACE.
+
 2014-02-12  Pedro Alves  <palves@redhat.com>
 
        * h8300-tdep.c (pseudo_from_raw_register)

diff --git a/gdb/gnu-v3-abi.c b/gdb/gnu-v3-abi.c
index a08dc368b4fd19ffaaab3551f190f19566ab3a6f..ceccbe5288663514810ff57f752472aaeeb8eb57 100644 (file)
--- a/gdb/gnu-v3-abi.c
+++ b/gdb/gnu-v3-abi.c
@@ -171,7 +171,7 @@ build_gdb_vtable_type (struct gdbarch *arch)
   TYPE_TAG_NAME (t) = "gdb_gnu_v3_abi_vtable";
   INIT_CPLUS_SPECIFIC (t);
 
-  return t;
+  return make_type_with_address_space (t, TYPE_INSTANCE_FLAG_CODE_SPACE);
 }