This is the mail archive of the
gdb-patches@sources.redhat.com
mailing list for the GDB project.
[patch/hppa] new in_epilogue_p method
- From: Randolph Chung <randolph at tausq dot org>
- To: gdb-patches at sources dot redhat dot com
- Date: Fri, 3 Dec 2004 16:54:52 -0800
- Subject: [patch/hppa] new in_epilogue_p method
- Reply-to: Randolph Chung <randolph at tausq dot org>
This patch teaches gdb about the hppa epilogue sequences, and fixes a
bug in software watchpoint handling. It's a bit ugly since there's no
clearly defined way to identify the epilogue... possibly we can do
better than this for dwarf2-enabled targets when we support dwarf2
unwinding ...
i'll give this a few days for comments
btw, now that "read_memory_nobpt" is deprecated, how are we supposed to
get that functionality when there is no "frame"? Do we just call
safe_frame_unwind_memeory with get_current_frame(), noting that the
frame argument to safe_frame_unwind_memory is not actually used?
randolph
2004-12-03 Randolph Chung <tausq@debian.org>
* hppa-tdep.c (hppa_in_function_epilogue_p): New function.
(hppa_gdbarch_init): Set in_epilogue_p method.
Index: hppa-tdep.c
===================================================================
RCS file: /cvs/src/src/gdb/hppa-tdep.c,v
retrieving revision 1.184
diff -u -p -r1.184 hppa-tdep.c
--- hppa-tdep.c 3 Dec 2004 23:59:52 -0000 1.184
+++ hppa-tdep.c 4 Dec 2004 00:46:18 -0000
@@ -621,6 +617,44 @@ find_unwind_entry (CORE_ADDR pc)
return NULL;
}
+/* The epilogue is defined here as the area either on the `bv' instruction
+ itself or an instruction which destroys the function's stack frame.
+
+ We do not assume that the epilogue is at the end of a function as we can
+ also have return sequences in the middle of a function. */
+static int
+hppa_in_function_epilogue_p (struct gdbarch *gdbarch, CORE_ADDR pc)
+{
+ unsigned long status;
+ unsigned int inst;
+ char buf[4];
+ int off;
+
+ status = deprecated_read_memory_nobpt (pc, buf, 4);
+ if (status != 0)
+ return 0;
+
+ inst = extract_unsigned_integer (buf, 4);
+
+ /* The most common way to perform a stack adjustment ldo X(sp),sp
+ We are destroying a stack frame if the offset is negative. */
+ if ((inst & 0xffffc000) == 0x37de0000
+ && hppa_extract_14 (inst) < 0)
+ return 1;
+
+ /* ldw,mb D(sp),X or ldd,mb D(sp),X */
+ if (((inst & 0x0fc010e0) == 0x0fc010e0
+ || (inst & 0x0fc010e0) == 0x0fc010e0)
+ && hppa_extract_14 (inst) < 0)
+ return 1;
+
+ /* bv %r0(%rp) or bv,n %r0(%rp) */
+ if (inst == 0xe840c000 || inst == 0xe840c002)
+ return 1;
+
+ return 0;
+}
+
static const unsigned char *
hppa_breakpoint_from_pc (CORE_ADDR *pc, int *len)
{
@@ -2529,6 +2535,8 @@ hppa_gdbarch_init (struct gdbarch_info i
/* The following gdbarch vector elements do not depend on the address
size, or in any other gdbarch element previously set. */
set_gdbarch_skip_prologue (gdbarch, hppa_skip_prologue);
+ set_gdbarch_in_function_epilogue_p (gdbarch,
+ hppa_in_function_epilogue_p);
set_gdbarch_inner_than (gdbarch, core_addr_greaterthan);
set_gdbarch_sp_regnum (gdbarch, HPPA_SP_REGNUM);
set_gdbarch_fp0_regnum (gdbarch, HPPA_FP0_REGNUM);
--
Randolph Chung
Debian GNU/Linux Developer, hppa/ia64 ports
http://www.tausq.org/