Re: [RFC PATCH] powerpc/ftrace: Refactoring and support for -fpatchable-function-entry

Le 23/05/2023 à 11:31, Naveen N Rao a écrit :

Christophe Leroy wrote:

quoted

That's better, but still more time than original implementation:

+20% to activate function tracer (was +40% with your RFC)
+21% to activate nop tracer (was +24% with your RFC)

perf record (without strict kernel rwx) :

     17.75%  echo     [kernel.kallsyms]   [k] ftrace_check_record
      9.76%  echo     [kernel.kallsyms]   [k] ftrace_replace_code
      6.53%  echo     [kernel.kallsyms]   [k] patch_instruction
      5.21%  echo     [kernel.kallsyms]   [k] __ftrace_hash_rec_update
      4.26%  echo     [kernel.kallsyms]   [k] ftrace_get_addr_curr
      4.18%  echo     [kernel.kallsyms]   [k] ftrace_get_call_inst.isra.0
      3.45%  echo     [kernel.kallsyms]   [k] ftrace_get_addr_new
      3.08%  echo     [kernel.kallsyms]   [k] function_trace_call
      2.20%  echo     [kernel.kallsyms]   [k] 
__rb_reserve_next.constprop.0
      2.05%  echo     [kernel.kallsyms]   [k] copy_page
      1.91%  echo     [kernel.kallsyms]   [k] 
ftrace_create_branch_inst.constprop.0
      1.83%  echo     [kernel.kallsyms]   [k] ftrace_rec_iter_next
      1.83%  echo     [kernel.kallsyms]   [k] rb_commit
      1.69%  echo     [kernel.kallsyms]   [k] ring_buffer_lock_reserve
      1.54%  echo     [kernel.kallsyms]   [k] trace_function
      1.39%  echo     [kernel.kallsyms]   [k] 
__call_rcu_common.constprop.0
      1.25%  echo     ld-2.23.so          [.] do_lookup_x
      1.17%  echo     [kernel.kallsyms]   [k] ftrace_rec_iter_record
      1.03%  echo     [kernel.kallsyms]   [k] unmap_page_range
      0.95%  echo     [kernel.kallsyms]   [k] flush_dcache_icache_page
      0.95%  echo     [kernel.kallsyms]   [k] ftrace_lookup_ip

Ok, I simplified this further, and this is as close to the previous fast 
path as we can get (applies atop the original RFC). The only difference 
left is the ftrace_rec iterator.

That's not better, that's even slightly worse (less than 1%).

I will try to investigate why.

Christophe