diff mbox

s390: Fix infcalls passing a single-field struct with static members

Message ID m3tv8f5oc1.fsf@oc0404454431.ibm.com
State New, archived
Headers

Commit Message

Andreas Arnez Oct. 11, 2019, 2:49 p.m. UTC 
  The infcall-nested-structs test case yields 36 FAILs on s390x because GCC
and GDB disagree on how to pass a C++ struct like this as an argument to a
function:

  struct s { float x; static float y; };

For the purpose of argument passing, GCC ignores static fields, while GDB
does not.  Thus GCC passes the argument in a floating-point register and
GDB passes it via memory.

Fix this by explicitly ignoring static fields when detecting single-field
structs.

gdb/ChangeLog:

	* s390-tdep.c (s390_effective_inner_type): Ignore static fields
	when unwrapping single-field structs.
---
 gdb/s390-tdep.c | 21 ++++++++++++++++++---
 1 file changed, 18 insertions(+), 3 deletions(-)

Comments

Tom Tromey Oct. 11, 2019, 5:46 p.m. UTC | #1 
>>>>> "Andreas" == Andreas Arnez <arnez@linux.ibm.com> writes:

Andreas> The infcall-nested-structs test case yields 36 FAILs on s390x because GCC
Andreas> and GDB disagree on how to pass a C++ struct like this as an argument to a
Andreas> function:

Andreas>   struct s { float x; static float y; };

Andreas> For the purpose of argument passing, GCC ignores static fields, while GDB
Andreas> does not.  Thus GCC passes the argument in a floating-point register and
Andreas> GDB passes it via memory.

Andreas> Fix this by explicitly ignoring static fields when detecting single-field
Andreas> structs.

FWIW this looked reasonable to me.

Tom
Andreas Arnez Oct. 15, 2019, 12:23 p.m. UTC | #2 
On Fri, Oct 11 2019, Tom Tromey wrote:

>>>>>> "Andreas" == Andreas Arnez <arnez@linux.ibm.com> writes:
>
> Andreas> The infcall-nested-structs test case yields 36 FAILs on s390x because GCC
> Andreas> and GDB disagree on how to pass a C++ struct like this as an argument to a
> Andreas> function:
>
> Andreas>   struct s { float x; static float y; };
>
> Andreas> For the purpose of argument passing, GCC ignores static fields, while GDB
> Andreas> does not.  Thus GCC passes the argument in a floating-point register and
> Andreas> GDB passes it via memory.
>
> Andreas> Fix this by explicitly ignoring static fields when detecting single-field
> Andreas> structs.
>
> FWIW this looked reasonable to me.

Thanks for having a look at this!

Pushed as ba18312d8f2cecf88b6edcbc277f433a62131e58.

--
Andreas
diff mbox

Patch

diff --git a/gdb/s390-tdep.c b/gdb/s390-tdep.c
index e7f1215e1a..6bd0528cf4 100644
--- a/gdb/s390-tdep.c
+++ b/gdb/s390-tdep.c
@@ -1637,11 +1637,26 @@  s390_address_class_name_to_type_flags (struct gdbarch *gdbarch,
 static struct type *
 s390_effective_inner_type (struct type *type, unsigned int min_size)
 {
-  while (TYPE_CODE (type) == TYPE_CODE_STRUCT
-	 && TYPE_NFIELDS (type) == 1)
+  while (TYPE_CODE (type) == TYPE_CODE_STRUCT)
     {
-      struct type *inner = check_typedef (TYPE_FIELD_TYPE (type, 0));
+      struct type *inner = NULL;
 
+      /* Find a non-static field, if any.  Unless there's exactly one,
+	 abort the unwrapping.  */
+      for (int i = 0; i < TYPE_NFIELDS (type); i++)
+	{
+	  struct field f = TYPE_FIELD (type, i);
+
+	  if (field_is_static (&f))
+	    continue;
+	  if (inner != NULL)
+	    return type;
+	  inner = FIELD_TYPE (f);
+	}
+
+      if (inner == NULL)
+	break;
+      inner = check_typedef (inner);
       if (TYPE_LENGTH (inner) < min_size)
 	break;
       type = inner;