This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
[commit] mn10300 prologue analyzer updates
- From: Kevin Buettner <kevinb at redhat dot com>
- To: gdb-patches at sourceware dot org
- Date: Mon, 28 Jan 2008 17:19:15 -0700
- Subject: [commit] mn10300 prologue analyzer updates
I've just committed that patch below. It updates the mn10300 prologue
analyzer.
* mn10300-tdep.c (mn10300_analyze_prologue): Check for an
instruction pattern that appears frequently in position
independent code. Fix bug in code which looks for "fmov" and
backtracks if no "fmov" is found.
Index: mn10300-tdep.c
===================================================================
RCS file: /cvs/src/src/gdb/mn10300-tdep.c,v
retrieving revision 1.152
diff -u -p -r1.152 mn10300-tdep.c
--- mn10300-tdep.c 16 Jan 2008 04:56:34 -0000 1.152
+++ mn10300-tdep.c 29 Jan 2008 00:01:23 -0000
@@ -608,6 +608,22 @@ mn10300_analyze_prologue (struct gdbarch
goto finish_prologue;
}
+ /* Check for "mov pc, a2", an instruction found in optimized, position
+ independent code. Skip it if found. */
+ if (buf[0] == 0xf0 && buf[1] == 0x2e)
+ {
+ addr += 2;
+
+ /* Quit now if we're beyond the stop point. */
+ if (addr >= stop)
+ goto finish_prologue;
+
+ /* Get the next two bytes so the prologue scan can continue. */
+ status = read_memory_nobpt (addr, buf, 2);
+ if (status != 0)
+ goto finish_prologue;
+ }
+
if (AM33_MODE (gdbarch) == 2)
{
/* Determine if any floating point registers are to be saved.
@@ -635,6 +651,7 @@ mn10300_analyze_prologue (struct gdbarch
was actually encountered. As a consequence, ``addr'' would
sometimes be advanced even when no fmov instructions were found. */
CORE_ADDR restore_addr = addr;
+ int fmov_found = 0;
/* First, look for add -SIZE,sp (i.e. add imm8,sp (0xf8feXX)
or add imm16,sp (0xfafeXXXX)
@@ -715,12 +732,9 @@ mn10300_analyze_prologue (struct gdbarch
break;
/* An fmov instruction has just been seen. We can
- now really commit to the pattern match. Set the
- address to restore at the end of this speculative
- bit of code to the actually address that we've
- been incrementing (or not) throughout the
- speculation. */
- restore_addr = addr;
+ now really commit to the pattern match. */
+
+ fmov_found = 1;
/* Get the floating point register number from the
2nd and 3rd bytes of the "fmov" instruction:
@@ -739,29 +753,18 @@ mn10300_analyze_prologue (struct gdbarch
imm_size = (buf[0] == 0xf9) ? 3 : 4;
}
}
- else
- {
- /* No "fmov" was found. Reread the two bytes at the original
- "addr" to reset the state. */
- addr = restore_addr;
- if (!safe_frame_unwind_memory (fi, addr, buf, 2))
- goto finish_prologue;
- }
}
- /* else the prologue consists entirely of an "add -SIZE,sp"
- instruction. Handle this below. */
}
- /* else no "add -SIZE,sp" was found indicating no floating point
- registers are saved in this prologue. */
-
- /* In the pattern match code contained within this block, `restore_addr'
- is set to the starting address at the very beginning and then
- iteratively to the next address to start scanning at once the
- pattern match has succeeded. Thus `restore_addr' will contain
- the address to rewind to if the pattern match failed. If the
- match succeeded, `restore_addr' and `addr' will already have the
- same value. */
- addr = restore_addr;
+ /* If no fmov instructions were found by the above sequence, reset
+ the state and pretend that the above bit of code never happened. */
+ if (!fmov_found)
+ {
+ addr = restore_addr;
+ status = read_memory_nobpt (addr, buf, 2);
+ if (status != 0)
+ goto finish_prologue;
+ stack_extra_size = 0;
+ }
}
/* Now see if we set up a frame pointer via "mov sp,a3" */