arm64: perf: Fix callchain parse error with kernel tracepoint events
authorHou Pengyang <houpengyang@huawei.com>
Sun, 10 May 2015 11:07:40 +0000 (11:07 +0000)
committerCatalin Marinas <catalin.marinas@arm.com>
Tue, 19 May 2015 16:49:02 +0000 (17:49 +0100)
For ARM64, when tracing with tracepoint events, the IP and pstate 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.data ]
 ./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 spsr .

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
     pfkey_send_policy_notify
     pfkey_get
     v9fs_vfs_rename
     page_follow_link_light
     link_path_walk
     el0_svc_naked
    .......

Signed-off-by: Hou Pengyang <houpengyang@huawei.com>
Acked-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
arch/arm64/include/asm/perf_event.h

index d26d1d53c0d706bbe14f8000558cc04bac2f2ad3..6471773db6fd8c067833ea92f647cca865e6fb4d 100644 (file)
@@ -24,4 +24,11 @@ extern unsigned long perf_misc_flags(struct pt_regs *regs);
 #define perf_misc_flags(regs)  perf_misc_flags(regs)
 #endif
 
+#define perf_arch_fetch_caller_regs(regs, __ip) { \
+       (regs)->pc = (__ip);    \
+       (regs)->regs[29] = (unsigned long) __builtin_frame_address(0); \
+       (regs)->sp = current_stack_pointer; \
+       (regs)->pstate = PSR_MODE_EL1h; \
+}
+
 #endif