Re: [RFC] ftrace: Add support to keep some functions out of ftrace

[Jiri Olsa <olsajiri@gmail.com>]

On Mon, Aug 15, 2022 at 05:41:27PM +0200, Peter Zijlstra wrote:
> On Mon, Aug 15, 2022 at 05:28:02PM +0200, Peter Zijlstra wrote:
> > On Mon, Aug 15, 2022 at 08:17:42AM -0700, Alexei Starovoitov wrote:
> > > It's hiding a fake function from ftrace, since it's not a function
> > > and ftrace infra shouldn't show it tracing logs.
> > > In other words it's a _notrace_ function with nop5.
> > 
> > Then make it a notrace function with a nop5 in it. That isn't hard.
> > 
> > The whole problem is that it isn't a notrace function and you're abusing
> > a __fentry__ site.
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/peterz/queue.git/commit/?h=x86/fineibt&id=8d075bdf11193f1d276bf19fa56b4b8dfe24df9e
> 
> foo.c:
> 
> __attribute__((__no_instrument_function__))
> __attribute__((patchable_function_entry(5)))
> void my_func(void)
> {
> }
> 
> void my_foo(void)
> {
> }
> 
> gcc -c foo.c -pg -mfentry -mcmodel=kernel -fno-PIE -O2
> 
> foo.o:     file format elf64-x86-64
> 
> 
> Disassembly of section .text:
> 
> 0000000000000000 <my_func>:
>    0:   f3 0f 1e fa             endbr64 
>    4:   90                      nop
>    5:   90                      nop
>    6:   90                      nop
>    7:   90                      nop
>    8:   90                      nop
>    9:   c3                      ret    
>    a:   66 0f 1f 44 00 00       nopw   0x0(%rax,%rax,1)
> 
> 0000000000000010 <my_foo>:
>   10:   f3 0f 1e fa             endbr64 
>   14:   e8 00 00 00 00          call   19 <my_foo+0x9>  15: R_X86_64_PLT32      __fentry__-0x4
>   19:   c3                      ret    
> 

ok, so the problem with __attribute__((patchable_function_entry(5))) is that
it puts function address into __patchable_function_entries section, which is
one of ftrace locations source:

  #define MCOUNT_REC()    . = ALIGN(8);     \
    __start_mcount_loc = .;                 \
    KEEP(*(__mcount_loc))                   \
    KEEP(*(__patchable_function_entries))   \
    __stop_mcount_loc = .;                  \
   ...


it looks like __patchable_function_entries is used for other than x86 archs,
so we perhaps we could have x86 specific MCOUNT_REC macro just with
__mcount_loc section?

jirka