This is the mail archive of the
gdb@sourceware.org
mailing list for the GDB project.
PowerPC prologue analysis
- From: Aleksandar Ristovski <aristovski at qnx dot com>
- To: gdb at sources dot redhat dot com
- Date: Mon, 28 Jul 2008 16:09:54 -0400
- Subject: PowerPC prologue analysis
Hello,
In the code, in rs6000-tdep.c around line 3334, there is a comment stating:
/* if != -1, fdata.saved_gpr is the smallest number of saved_gpr.
All gpr's from saved_gpr to gpr31 are saved. */
I am, however, witnessing a function that appears to be saving r30, but not r31 (see the disassembly below). This, in turn, causes gdb to unwind r31 from a 'saved' area even though the area does not exist.
I am not very familiar with PowerPC ABI, but from what I gather reading the "function call" section, but can not see where is it stated that if r30 is saved, then r31 must be saved too? But again, I haven't studied the ABI very thoroughly and might be missing that line.
Just for the reference, here is the disassembly of the function:
(gdb) disassemble foo
Dump of assembler code for function foo
0xfe346aa0 <foo+0>: stwu r1,-16(r1)
0xfe346aa4 <foo+4>: mflr r0
0xfe346aa8 <foo+8>: bl 0xfe37ca18
0xfe346aac <foo+12>: mr r4,r3
0xfe346ab0 <foo+16>: stw r30,8(r1)
0xfe346ab4 <foo+20>: mflr r30
0xfe346ab8 <foo+24>: li r5,0
0xfe346abc <foo+28>: li r6,0
0xfe346ac0 <foo+32>: stw r0,20(r1)
0xfe346ac4 <foo+36>: lwz r3,-176(r30)
0xfe346ac8 <foo+40>: bl 0xfe37d738
0xfe346acc <foo+44>: lwz r0,20(r1)
0xfe346ad0 <foo+48>: lwz r30,8(r1)
0xfe346ad4 <foo+52>: addi r1,r1,16
0xfe346ad8 <foo+56>: mtlr r0
0xfe346adc <foo+60>: blr
End of assembler dump.
Thanks,
Aleksandar Ristovski
QNX Software Systems