* [PATCH] parisc: add CALLER_ADDR{0-6} macros
@ 2009-10-25 21:48 Helge Deller
2009-10-27 4:49 ` Randolph Chung
0 siblings, 1 reply; 3+ messages in thread
From: Helge Deller @ 2009-10-25 21:48 UTC (permalink / raw)
To: linux-parisc, Kyle McMartin, Randolph Chung
Signed-off-by: Helge Deller <deller@gmx.de>
diff --git a/arch/parisc/include/asm/ftrace.h b/arch/parisc/include/asm/ftrace.h
index 2fa05dd..72c0faf 100644
--- a/arch/parisc/include/asm/ftrace.h
+++ b/arch/parisc/include/asm/ftrace.h
@@ -20,6 +20,20 @@ struct ftrace_ret_stack {
* Defined in entry.S
*/
extern void return_to_handler(void);
+
+
+extern unsigned long return_address(unsigned int);
+
+#define HAVE_ARCH_CALLER_ADDR
+
+#define CALLER_ADDR0 ((unsigned long)__builtin_return_address(0))
+#define CALLER_ADDR1 return_address(1)
+#define CALLER_ADDR2 return_address(2)
+#define CALLER_ADDR3 return_address(3)
+#define CALLER_ADDR4 return_address(4)
+#define CALLER_ADDR5 return_address(5)
+#define CALLER_ADDR6 return_address(6)
+
#endif /* __ASSEMBLY__ */
#endif /* _ASM_PARISC_FTRACE_H */
diff --git a/arch/parisc/kernel/unwind.c b/arch/parisc/kernel/unwind.c
index 69dad5a..6c631bf 100644
--- a/arch/parisc/kernel/unwind.c
+++ b/arch/parisc/kernel/unwind.c
@@ -417,3 +429,31 @@ int unwind_to_user(struct unwind_frame_info *info)
return ret;
}
+
+unsigned long return_address(unsigned int level)
+{
+ struct unwind_frame_info info;
+ struct pt_regs r;
+ unsigned long sp;
+
+ /* initialize unwind info */
+ asm volatile ("copy %%r30, %0" : "=r"(sp));
+ memset(&r, 0, sizeof(struct pt_regs));
+ r.iaoq[0] = (unsigned long) current_text_addr();
+ r.gr[2] = (unsigned long) __builtin_return_address(0);
+ r.gr[30] = sp;
+ unwind_frame_init(&info, current, &r);
+
+ /* unwind stack */
+ ++level;
+ do {
+ if (unwind_once(&info) < 0 || info.ip == 0)
+ return 0;
+ if (!__kernel_text_address(info.ip)) {
+ return 0;
+ }
+ } while (info.ip && level--);
+
+ return info.ip;
+}
+
^ permalink raw reply related [flat|nested] 3+ messages in thread
* Re: [PATCH] parisc: add CALLER_ADDR{0-6} macros
2009-10-25 21:48 [PATCH] parisc: add CALLER_ADDR{0-6} macros Helge Deller
@ 2009-10-27 4:49 ` Randolph Chung
2009-10-27 13:41 ` Helge Deller
0 siblings, 1 reply; 3+ messages in thread
From: Randolph Chung @ 2009-10-27 4:49 UTC (permalink / raw)
To: Helge Deller; +Cc: linux-parisc, Kyle McMartin
Helge,
> +unsigned long return_address(unsigned int level)
> +{
> + struct unwind_frame_info info;
> + struct pt_regs r;
> + unsigned long sp;
> +
> + /* initialize unwind info */
> + asm volatile ("copy %%r30, %0" : "=r"(sp));
> + memset(&r, 0, sizeof(struct pt_regs));
> + r.iaoq[0] = (unsigned long) current_text_addr();
> + r.gr[2] = (unsigned long) __builtin_return_address(0);
> + r.gr[30] = sp;
> + unwind_frame_init(&info, current, &r);
> +
> + /* unwind stack */
> + ++level;
> + do {
> + if (unwind_once(&info) < 0 || info.ip == 0)
> + return 0;
> + if (!__kernel_text_address(info.ip)) {
> + return 0;
> + }
> + } while (info.ip && level--);
> +
> + return info.ip;
> +}
> +
Can you show an objdump of this function once it is compiled? I suspect
the stack pointer initialization here is not reliable.
Ideally unwind_frame_init is called with the frame address in gr[30].
With a big struct like pt_regs on the stack, the sp initialization might
be quite far from the actual frame address.
The unwind_once() code uses like heuristics to try to recover from
inaccurate stack pointers (by aligning and stepping the frame 64 bytes
at a time) but that is really a brute force guess.
I realize I used a similar construct in traps.c, but even there I think
it doesn't work reliably.
Maybe somebody else on the list (Dave? :) can suggest a better way to do
this.
randolph
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] parisc: add CALLER_ADDR{0-6} macros
2009-10-27 4:49 ` Randolph Chung
@ 2009-10-27 13:41 ` Helge Deller
0 siblings, 0 replies; 3+ messages in thread
From: Helge Deller @ 2009-10-27 13:41 UTC (permalink / raw)
To: Randolph Chung, linux-parisc, Kyle McMartin
* Randolph Chung <randolph@tausq.org>:
>> +unsigned long return_address(unsigned int level)
>> +{
>> + struct unwind_frame_info info;
>> + struct pt_regs r;
>> + unsigned long sp;
>> +
>> + /* initialize unwind info */
>> + asm volatile ("copy %%r30, %0" : "=r"(sp));
>> + memset(&r, 0, sizeof(struct pt_regs));
>> + r.iaoq[0] = (unsigned long) current_text_addr();
>> + r.gr[2] = (unsigned long) __builtin_return_address(0);
>> + r.gr[30] = sp;
>> + unwind_frame_init(&info, current, &r);
>> +
>> + /* unwind stack */
>> + ++level;
>> + do {
>> + if (unwind_once(&info) < 0 || info.ip == 0)
>> + return 0;
>> + if (!__kernel_text_address(info.ip)) {
>> + return 0;
>> + }
>> + } while (info.ip && level--);
>> +
>> + return info.ip;
>> +}
>> +
>
> Can you show an objdump of this function once it is compiled? I suspect
> the stack pointer initialization here is not reliable.
> Ideally unwind_frame_init is called with the frame address in gr[30].
> With a big struct like pt_regs on the stack, the sp initialization might
> be quite far from the actual frame address.
>
> The unwind_once() code uses like heuristics to try to recover from
> inaccurate stack pointers (by aligning and stepping the frame 64 bytes
> at a time) but that is really a brute force guess.
>
> I realize I used a similar construct in traps.c, but even there I think
> it doesn't work reliably.
Hi Randolph,
I agree, that those kind of implementations may not always be reliable.
Nevertheless, I started off with a copy of your code, but it didn't
worked at all. I assume due to some compiler-trickery to include the
static local functions in other functions it does work in traps.c
nevertheless.
I did sucessfully tested the patch I submitted here, both in 32- and 64-bit.
Below is the disassembly of the code which you asked for...
Helge
Disassembly of section .text.return_address:
00000000 <return_address>:
0: 08 03 02 41 copy r3,r1
4: 6b c2 3f d9 stw rp,-14(sp)
8: 08 1e 02 43 copy sp,r3
c: 6f c1 04 80 stw,ma r1,240(sp)
10: 68 67 04 00 stw r7,200(r3)
14: 08 02 02 47 copy rp,r7
18: 68 66 04 08 stw r6,204(r3)
1c: 08 1a 02 46 copy r26,r6
20: 68 65 04 10 stw r5,208(r3)
24: 68 64 04 18 stw r4,20c(r3)
28: 08 1e 02 44 copy sp,r4
2c: 34 65 00 50 ldo 28(r3),r5
30: 34 19 00 00 ldi 0,r25
34: 08 05 02 5a copy r5,r26
38: e8 40 00 00 b,l 40 <return_address+0x40>,rp
3c: 34 18 03 b0 ldi 1d8,r24
40: 68 67 00 60 stw r7,30(r3)
44: eb 80 40 00 blr r0,ret0
48: 08 00 02 40 nop
4c: 68 64 01 40 stw r4,a0(r3)
50: 68 7c 03 a0 stw ret0,1d0(r3)
54: 03 c0 08 bc mfctl tr6,ret0
58: 34 64 00 10 ldo 8(r3),r4
5c: 0f 80 10 99 ldw 0(ret0),r25
60: 08 04 02 5a copy r4,r26
64: e8 40 00 00 b,l 6c <return_address+0x6c>,rp
68: 08 05 02 58 copy r5,r24
6c: 08 04 02 5a copy r4,r26
70: e8 40 00 00 b,l 78 <return_address+0x78>,rp
74: 34 c6 00 02 ldo 1(r6),r6
78: 8f 80 60 80 cmpib,> 0,ret0,c0 <return_address+0xc0>
7c: 34 1c 00 00 ldi 0,ret0
80: 48 7c 00 20 ldw 10(r3),ret0
84: 87 80 20 68 cmpib,= 0,ret0,c0 <return_address+0xc0>
88: 08 1c 02 5a copy ret0,r26
8c: e8 40 00 00 b,l 94 <return_address+0x94>,rp
90: 08 00 02 40 nop
94: 87 80 20 40 cmpib,= 0,ret0,bc <return_address+0xbc>
98: 48 7c 00 20 ldw 10(r3),ret0
9c: 87 80 20 40 cmpib,= 0,ret0,c4 <return_address+0xc4>
a0: 48 62 3f d9 ldw -14(r3),rp
a4: 84 c0 20 30 cmpib,= 0,r6,c4 <return_address+0xc4>
a8: 08 04 02 5a copy r4,r26
ac: e8 40 00 00 b,l b4 <return_address+0xb4>,rp
b0: 34 c6 3f ff ldo -1(r6),r6
b4: 87 80 7f 95 cmpib,<= 0,ret0,84 <return_address+0x84>
b8: 48 7c 00 20 ldw 10(r3),ret0
bc: 34 1c 00 00 ldi 0,ret0
c0: 48 62 3f d9 ldw -14(r3),rp
c4: 48 67 04 00 ldw 200(r3),r7
c8: 48 66 04 08 ldw 204(r3),r6
cc: 48 65 04 10 ldw 208(r3),r5
d0: 48 64 04 18 ldw 20c(r3),r4
d4: 34 7e 00 80 ldo 40(r3),sp
d8: e8 40 c0 00 bv r0(rp)
dc: 4f c3 3f 81 ldw,mb -40(sp),r3
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2009-10-27 13:41 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2009-10-25 21:48 [PATCH] parisc: add CALLER_ADDR{0-6} macros Helge Deller
2009-10-27 4:49 ` Randolph Chung
2009-10-27 13:41 ` Helge Deller
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).