LoongArch: Fix callchain parse error with kernel tracepoint events again
authorHuacai Chen <chenhuacai@loongson.cn>
Tue, 14 May 2024 04:24:18 +0000 (12:24 +0800)
committerHuacai Chen <chenhuacai@loongson.cn>
Tue, 14 May 2024 04:24:18 +0000 (12:24 +0800)
With commit d3119bc985fb645 ("LoongArch: Fix callchain parse error with
kernel tracepoint events"), perf can parse kernel callchain, but not
complete and sometimes maybe error. The reason is LoongArch's unwinders
(guess, prologue and orc) don't really need fp (i.e., regs[22]), and
they use sp (i.e., regs[3]) as the frame address rather than the current
stack pointer.

Fix that by removing the assignment of regs[22], and instead assign the
__builtin_frame_address(0) to regs[3].

Without fix:

  Children      Self  Command        Shared Object      Symbol
  ........  ........  .............  .................  ................
  33.91%    33.91%    swapper        [kernel.vmlinux]   [k] __schedule
            |
            |--33.04%--__schedule
            |
             --0.87%--__arch_cpu_idle
                       __schedule

With this fix:

  Children      Self  Command        Shared Object      Symbol
  ........  ........  .............  .................  ................
  31.16%    31.16%    swapper        [kernel.vmlinux]   [k] __schedule
            |
            |--20.63%--smpboot_entry
            |          cpu_startup_entry
            |          schedule_idle
            |          __schedule
            |
             --10.53%--start_kernel
                       cpu_startup_entry
                       schedule_idle
                       __schedule

Fixes: d3119bc985fb645 ("LoongArch: Fix callchain parse error with kernel tracepoint events")
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
arch/loongarch/include/asm/perf_event.h

index 52b638059e40b31645a62243e467c09e7d7ce0cf..f948a0676daf80382f08cfbfe7342e7076b81c5a 100644 (file)
@@ -13,8 +13,7 @@
 
 #define perf_arch_fetch_caller_regs(regs, __ip) { \
        (regs)->csr_era = (__ip); \
-       (regs)->regs[3] = current_stack_pointer; \
-       (regs)->regs[22] = (unsigned long) __builtin_frame_address(0); \
+       (regs)->regs[3] = (unsigned long) __builtin_frame_address(0); \
 }
 
 #endif /* __LOONGARCH_PERF_EVENT_H__ */