Re: [PATCH] arm64: perf: Fix callchain parse error with kernel tracepoint events

From: Hou Pengyang
Date: Wed Apr 29 2015 - 21:50:31 EST

On 2015/4/29 18:12, Will Deacon wrote:

On Tue, Apr 28, 2015 at 02:20:48PM +0100, Hou Pengyang wrote:
For ARM64, when tracing with tracepoint events, the IP and cpsr are set
to 0, preventing the perf code parsing the callchain and resolving the
symbols correctly.

./perf record -e sched:sched_switch -g --call-graph dwarf ls
[ perf record: Captured and wrote 0.146 MB ]
./perf report -f
Samples: 194 of event 'sched:sched_switch', Event count (approx.): 194
Children Self Command Shared Object Symbol
100.00% 100.00% ls [unknown] [.] 0000000000000000

The fix is to implement perf_arch_fetch_caller_regs for ARM64, which fills
several necessary registers used for callchain unwinding, including pc,sp,
fp and psr .

With this patch, callchain can be parsed correctly as follows:

+ 2.63% 0.00% ls [kernel.kallsyms] [k] vfs_symlink
+ 2.63% 0.00% ls [kernel.kallsyms] [k] follow_down
+ 2.63% 0.00% ls [kernel.kallsyms] [k] pfkey_get
+ 2.63% 0.00% ls [kernel.kallsyms] [k] do_execveat_common.isra.33
- 2.63% 0.00% ls [kernel.kallsyms] [k] pfkey_send_policy_notify

For tracepoint event, stack parsing also doesn't work well for ARM. Jean Pihet
comed up a patch:

Any chance you could revive that series too, please? I'd like to update both
arm and arm64 together, since we're currently working at merging the two
perf backends and introducing discrepencies is going to delay that even

Signed-off-by: Hou Pengyang <houpengyang@xxxxxxxxxx>
arch/arm64/include/asm/perf_event.h | 16 ++++++++++++++++
1 file changed, 16 insertions(+)

diff --git a/arch/arm64/include/asm/perf_event.h b/arch/arm64/include/asm/perf_event.h
index d26d1d5..16a074f 100644
--- a/arch/arm64/include/asm/perf_event.h
+++ b/arch/arm64/include/asm/perf_event.h
@@ -24,4 +24,20 @@ extern unsigned long perf_misc_flags(struct pt_regs *regs);
#define perf_misc_flags(regs) perf_misc_flags(regs)

+#define perf_arch_fetch_caller_regs(regs, __ip) { \
+ unsigned long sp; \
+ __asm__ ("mov %[sp], sp\n" : [sp] "=r" (sp)); \
+ (regs)->pc = (__ip); \
+ __asm__ ( \
+ "str %[sp], %[_arm64_sp] \n\t" \
+ "str x29, %[_arm64_fp] \n\t" \
+ "mrs %[_arm64_cpsr], spsr_el1 \n\t" \
+ : [_arm64_sp] "=m" (regs->sp), \
+ [_arm64_fp] "=m" (regs->regs[29]), \
+ [_arm64_cpsr] "=r" (regs->pstate) \

Does this really all need to be in assembly code? Ideally we'd use something
like __builtin_stack_pointer and __builtin_frame_pointer. That just leaves
the CPSR, but given that it's (a) only used for user_mode_regs tests and (b)
this macro is only used by ftrace, then we just set it to a static value
indicating that we're at EL1.

So I *think* we should be able to write this as three lines of C.

Hi, will, as you said, we can get fp by __builtin_frame_address() and pstate by setting it to a static value. However, for sp, there isn't a gcc builtin fuction like __builtin_stack_pointer, so assembly code is needed. What's more, if CONFIG_FRAME_POINTER is close, can fp be got by __builtin_frame_address()?



To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at