Le 03/03/2022 à 17:59, Naveen N. Rao a écrit :

Christophe Leroy wrote:

quoted

ftrace_caller() and frace_regs_caller() have now a lot in common.

Refactor them using GAS macros.

Thanks for doing this - this looks much better.

quoted

Signed-off-by: Christophe Leroy <redacted>
---
 arch/powerpc/kernel/trace/ftrace_mprofile.S | 147 ++++++--------------
 1 file changed, 45 insertions(+), 102 deletions(-)

diff --git a/arch/powerpc/kernel/trace/ftrace_mprofile.S 

b/arch/powerpc/kernel/trace/ftrace_mprofile.S
index 76dab07fd8fd..630b2de9957b 100644

--- a/arch/powerpc/kernel/trace/ftrace_mprofile.S
+++ b/arch/powerpc/kernel/trace/ftrace_mprofile.S

@@ -32,7 +32,7 @@

  * Our job is to save the register state into a struct pt_regs (on 
the stack)
  * and then arrange for the ftrace function to be called.
  */
-_GLOBAL(ftrace_regs_caller)
+.macro    ftrace_regs_entry allregs
     /* Save the original return address in A's stack frame */
 #ifdef CONFIG_MPROFILE_KERNEL
     PPC_STL    r0,LRSAVE(r1)

@@ -43,7 +43,7 @@ _GLOBAL(ftrace_regs_caller)

     /* Save all gprs to pt_regs */
     SAVE_GPR(0, r1)
-    SAVE_GPRS(2, 11, r1)
+    SAVE_GPRS(3, 10, r1)

 #ifdef CONFIG_PPC64
     /* Ok to continue? */

@@ -52,17 +52,29 @@ _GLOBAL(ftrace_regs_caller)

     beq    ftrace_no_trace
 #endif

The ability to disable ftrace in certain code paths through 
paca_struct->ftrace_enabled will also be relevant on ppc32 - it will be 
nice if it can be introduced there.

Ah ? I understood from commit ea678ac627e0 ("powerpc64/ftrace: Add a 
field in paca to disable ftrace in unsafe code paths") that it was for 
when it runs in real mode. PPC32 doesn't run any C code in real mode.

Are there any other situations that real_mode where we'd like to disable 
it ? If so we could use the thread_struct as we don't have paca on PPC32.


Christophe