[RFC] Sanitize i386 targets

[Mark Kettenis <kettenis at wins dot uva dot nl>]

In a hope to multi-arch the i386 targets, i'm trying to cleanup the
target dependent stuff in the config/i386 directory.  In doing so, I
noticed that quite a few targets make use of the tm-i386v.h file.
This file is supposed to give target dependent definitions for System
V-derived systems (presumably release III, a.k.a. SVR3), but quite a
few targets including this file bear no resemblence to UNIX at all.
Most of these are the "embedded" targets, i.e. i386-elf, i386-aout,
i386-coff, vxWorks.  Another example is Cygwin.  It would make more
sense to me if these targets used the "tm-i386.h" header directly.

Now what exactly does "tm-i386v.h" provide that "tm-i386.h" doesn't?
Hardly anything.  In fact it overrides some stuff from "tm-i386.h"
with code that provides less functionality:

1. EXTRACT_RETURN_VALUE and STORE_RETURN_VALUE:

   These are implemented as a simple memcpy.  I suspect that even SVR3
   used the same conventions for return values as the current
   SVR4/i386 ABI (except perhaps for small structs).  Basically, this
   means that only simple scalar return values (integers up to 32 bits
   and pointers) are supported.  No 64-bit integers and probably no
   floating point return values.  It's also very likely that returning
   small structs won't work right.

   I'm fairly sure that the implementation provided by "tm-i386.h"
   does the right thing for all supported i?86 targets (except the
   symmetry, which provides its own version of EXTRACT_RETURN_VALUE
   anyway).  Therefore I propose to zap these macros from
   "tm-i386v.h".

2. FRAMELESS_FUNCTION_INVOCATION, FRAME_SAVED_PC, FRAME_NUM_ARGS:

   These are almost identical to the ones provided by "tm-i386.h",
   except that the logic to handle signal handler frames is absent.
   Previously the version of FRAME_SAVED_PC in "tm-i386.h" caused a
   linker failure for targets that didn't implement
   sigtramp_saved_pc.  That's probably the reason why these targets
   used "tm-i386v.h" in the first place.  However I recently checked
   in some changes to overcome this problem.  The additional signal
   handler frame logic shouldn't do any harm since none of these
   targets define IN_SIGTRAMP and therefore never flag any frames as
   corresponding to a signal handler.  So these can be zapped too.

In the end zapping these macros leaves only
START_INFERIOR_TRAPS_EXPECTED.  Now that's a bit of SVR3-specific
information.  But it's irrelevant to non-SVR3 targets, so I think
those can simply use "tm-i386.h".  Building a i386-elf or i386-cygwin
cross debugger with such a change does indeed succeed.  Unfortunately
I cannot really check if such a cross debugger does indeed work.  So I
would appreciate it if someone who can test such a compiler could do
so.  Try to get GDB to display the return value of test_return() in
the following program:

  #include <limits.h>
  #include <stdio.h>
  #include <stdlib.h>

  long long
  test_return (void)
  {
    return 2 * LONG_MAX;
  }

  int
  main (void)
  {
    printf ("test_return returned %Ld\n", test_return ());

    return EXIT_SUCCESS;
  }

A preliminary patch is attached.

Comments are also very welcome.

Thanks,

Mark


Index: config/i386/tm-i386v.h
===================================================================
RCS file: /cvs/src/src/gdb/config/i386/tm-i386v.h,v
retrieving revision 1.6
diff -u -p -r1.6 tm-i386v.h
--- config/i386/tm-i386v.h 2001/05/08 12:00:49 1.6
+++ config/i386/tm-i386v.h 2001/05/10 18:16:30
@@ -24,7 +24,6 @@
 
 /* First pick up the generic *86 target file. */
 
-#include "regcache.h"
 #include "i386/tm-i386.h"
 
 /* Number of traps that happen between exec'ing the shell to run an
@@ -33,52 +32,5 @@
 
 #undef  START_INFERIOR_TRAPS_EXPECTED
 #define START_INFERIOR_TRAPS_EXPECTED 4
-
-/* Extract from an array REGBUF containing the (raw) register state
-   a function return value of type TYPE, and copy that, in virtual format,
-   into VALBUF.  */
-
-#undef  EXTRACT_RETURN_VALUE
-#define EXTRACT_RETURN_VALUE(TYPE,REGBUF,VALBUF) \
-  memcpy ((VALBUF), (REGBUF), TYPE_LENGTH (TYPE))
-
-/* Write into appropriate registers a function return value
-   of type TYPE, given in virtual format.  */
-
-#undef  STORE_RETURN_VALUE
-#define STORE_RETURN_VALUE(TYPE,VALBUF) \
-  write_register_bytes (0, VALBUF, TYPE_LENGTH (TYPE))
-
-
-/* Describe the pointer in each stack frame to the previous stack frame
-   (its caller).  */
-
-/* FRAME_CHAIN takes a frame's nominal address
-   and produces the frame's chain-pointer. */
-
-#undef  FRAME_CHAIN
-#define FRAME_CHAIN(thisframe) \
-  (!inside_entry_file ((thisframe)->pc) ? \
-   read_memory_integer ((thisframe)->frame, 4) :\
-   0)
-
-/* Define other aspects of the stack frame.  */
-
-/* A macro that tells us whether the function invocation represented
-   by FI does not have a frame on the stack associated with it.  If it
-   does not, FRAMELESS is set to 1, else 0.  */
-
-#undef  FRAMELESS_FUNCTION_INVOCATION
-#define FRAMELESS_FUNCTION_INVOCATION(FI) \
-  (frameless_look_for_prologue (FI))
-
-#undef  FRAME_SAVED_PC
-#define FRAME_SAVED_PC(FRAME) (read_memory_integer ((FRAME)->frame + 4, 4))
-
-/* Return number of args passed to a frame.
-   Can return -1, meaning no way to tell.  */
-
-#undef  FRAME_NUM_ARGS
-#define FRAME_NUM_ARGS(fi) (-1)
 
 #endif /* ifndef TM_I386V_H */