[PATCH] powerpc: inline current_stack_pointer()

[PATCH] powerpc: inline current_stack_pointer()

[Christophe Leroy]

current_stack_pointeur() is a single instruction function. it
It is not worth breaking the execution flow with a bl/blr for a
single instruction

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
 arch/powerpc/include/asm/reg.h  | 7 ++++++-
 arch/powerpc/kernel/misc.S      | 4 ----
 arch/powerpc/kernel/ppc_ksyms.c | 2 --
 3 files changed, 6 insertions(+), 7 deletions(-)

diff --git a/arch/powerpc/include/asm/reg.h b/arch/powerpc/include/asm/reg.h
index c1e82e9..7ce6777 100644
--- a/arch/powerpc/include/asm/reg.h
+++ b/arch/powerpc/include/asm/reg.h
@@ -1301,7 +1301,12 @@ static inline unsigned long mfvtb (void)
 
 #define proc_trap()	asm volatile("trap")
 
-extern unsigned long current_stack_pointer(void);
+static inline unsigned long current_stack_pointer(void)
+{
+	register unsigned long *ptr asm("r1");
+
+	return *ptr;
+}
 
 extern unsigned long scom970_read(unsigned int address);
 extern void scom970_write(unsigned int address, unsigned long value);
diff --git a/arch/powerpc/kernel/misc.S b/arch/powerpc/kernel/misc.S
index 0d43219..7ce26d4 100644
--- a/arch/powerpc/kernel/misc.S
+++ b/arch/powerpc/kernel/misc.S
@@ -114,7 +114,3 @@ _GLOBAL(longjmp)
 	mtlr	r0
 	mr	r3,r4
 	blr
-
-_GLOBAL(current_stack_pointer)
-	PPC_LL	r3,0(r1)
-	blr
diff --git a/arch/powerpc/kernel/ppc_ksyms.c b/arch/powerpc/kernel/ppc_ksyms.c
index 9f01e28..eb5c5dc 100644
--- a/arch/powerpc/kernel/ppc_ksyms.c
+++ b/arch/powerpc/kernel/ppc_ksyms.c
@@ -33,5 +33,3 @@ EXPORT_SYMBOL(store_vr_state);
 #ifdef CONFIG_EPAPR_PARAVIRT
 EXPORT_SYMBOL(epapr_hypercall_start);
 #endif
-
-EXPORT_SYMBOL(current_stack_pointer);
-- 
2.1.0

Re: [PATCH] powerpc: inline current_stack_pointer()

[Gabriel Paubert]

On Mon, May 23, 2016 at 10:46:02AM +0200, Christophe Leroy wrote:
> current_stack_pointeur() is a single instruction function. it
> It is not worth breaking the execution flow with a bl/blr for a
> single instruction

Are you sure that the result is always the same? 

Calling an external function prevents the compiler from considering the
caller of of current_stack_pointer a leaf function, which certainly 
does not help the compiler, but in a leaf function the compiler is free 
not to establish a new frame.

If the compiler decides to establishes a new frame (typically with 
"stwu r1,-frame_size(r1)"), *r1 is the previous stack pointer, or
the caller's stack pointer, or the current function frame pointer if
I remember correctly the ABI definitions. 

However, if the compiler decides that it can get away without a frame
for the function, *r1 is the stack pointer of the caller's caller.

Depending on the application, this may or may not be important.

By the way, isn't there a GCC builtin which can perform this task,
for example builtin_frame_address()?

    Gabriel
> 
> Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
> ---
>  arch/powerpc/include/asm/reg.h  | 7 ++++++-
>  arch/powerpc/kernel/misc.S      | 4 ----
>  arch/powerpc/kernel/ppc_ksyms.c | 2 --
>  3 files changed, 6 insertions(+), 7 deletions(-)
> 
> diff --git a/arch/powerpc/include/asm/reg.h b/arch/powerpc/include/asm/reg.h
> index c1e82e9..7ce6777 100644
> --- a/arch/powerpc/include/asm/reg.h
> +++ b/arch/powerpc/include/asm/reg.h
> @@ -1301,7 +1301,12 @@ static inline unsigned long mfvtb (void)
>  
>  #define proc_trap()	asm volatile("trap")
>  
> -extern unsigned long current_stack_pointer(void);
> +static inline unsigned long current_stack_pointer(void)
> +{
> +	register unsigned long *ptr asm("r1");
> +
> +	return *ptr;
> +}
>  
>  extern unsigned long scom970_read(unsigned int address);
>  extern void scom970_write(unsigned int address, unsigned long value);
> diff --git a/arch/powerpc/kernel/misc.S b/arch/powerpc/kernel/misc.S
> index 0d43219..7ce26d4 100644
> --- a/arch/powerpc/kernel/misc.S
> +++ b/arch/powerpc/kernel/misc.S
> @@ -114,7 +114,3 @@ _GLOBAL(longjmp)
>  	mtlr	r0
>  	mr	r3,r4
>  	blr
> -
> -_GLOBAL(current_stack_pointer)
> -	PPC_LL	r3,0(r1)
> -	blr
> diff --git a/arch/powerpc/kernel/ppc_ksyms.c b/arch/powerpc/kernel/ppc_ksyms.c
> index 9f01e28..eb5c5dc 100644
> --- a/arch/powerpc/kernel/ppc_ksyms.c
> +++ b/arch/powerpc/kernel/ppc_ksyms.c
> @@ -33,5 +33,3 @@ EXPORT_SYMBOL(store_vr_state);
>  #ifdef CONFIG_EPAPR_PARAVIRT
>  EXPORT_SYMBOL(epapr_hypercall_start);
>  #endif
> -
> -EXPORT_SYMBOL(current_stack_pointer);
> -- 
> 2.1.0
> _______________________________________________
> Linuxppc-dev mailing list
> Linuxppc-dev@lists.ozlabs.org
> https://lists.ozlabs.org/listinfo/linuxppc-dev

Re: [PATCH] powerpc: inline current_stack_pointer()

[Paul Mackerras]

On Mon, May 23, 2016 at 07:17:38PM +0200, Gabriel Paubert wrote:
> On Mon, May 23, 2016 at 10:46:02AM +0200, Christophe Leroy wrote:
> > current_stack_pointeur() is a single instruction function. it
> > It is not worth breaking the execution flow with a bl/blr for a
> > single instruction
> 
> Are you sure that the result is always the same? 
> 
> Calling an external function prevents the compiler from considering the
> caller of of current_stack_pointer a leaf function, which certainly 
> does not help the compiler, but in a leaf function the compiler is free 
> not to establish a new frame.
> 
> If the compiler decides to establishes a new frame (typically with 
> "stwu r1,-frame_size(r1)"), *r1 is the previous stack pointer, or
> the caller's stack pointer, or the current function frame pointer if
> I remember correctly the ABI definitions. 
> 
> However, if the compiler decides that it can get away without a frame
> for the function, *r1 is the stack pointer of the caller's caller.
> 
> Depending on the application, this may or may not be important.

Right.  I think I wrote the original current_stack_pointer()
implementation, and that I deliberately didn't make it an inline
so that the caller would have to establish its own stack frame,
and thus its stack pointer value would be a well-defined thing.

Paul.

Re: [PATCH] powerpc: inline current_stack_pointer()

[Segher Boessenkool]

On Mon, May 23, 2016 at 10:46:02AM +0200, Christophe Leroy wrote:
> +static inline unsigned long current_stack_pointer(void)
> +{
> +	register unsigned long *ptr asm("r1");
> +
> +	return *ptr;
> +}

Register asm is only guaranteed to work as input to inline asm.  NAK.


Segher

Re: [PATCH] powerpc: inline current_stack_pointer()

[Christophe Leroy]

Le 23/05/2016 à 22:22, Segher Boessenkool a écrit :
> On Mon, May 23, 2016 at 10:46:02AM +0200, Christophe Leroy wrote:
>> +static inline unsigned long current_stack_pointer(void)
>> +{
>> +	register unsigned long *ptr asm("r1");
>> +
>> +	return *ptr;
>> +}
> Register asm is only guaranteed to work as input to inline asm.  NAK.
>
Does it mean that the following declaration in 
arch/powerpc/include/asm/paca.h is wrong too ?

register struct paca_struct *local_paca asm("r13");

Christophe

Re: [PATCH] powerpc: inline current_stack_pointer()

[Segher Boessenkool]

On Tue, May 24, 2016 at 07:39:59AM +0200, Christophe Leroy wrote:
> >>+static inline unsigned long current_stack_pointer(void)
> >>+{
> >>+	register unsigned long *ptr asm("r1");
> >>+
> >>+	return *ptr;
> >>+}
> >Register asm is only guaranteed to work as input to inline asm.  NAK.
> >
> Does it mean that the following declaration in 
> arch/powerpc/include/asm/paca.h is wrong too ?
> 
> register struct paca_struct *local_paca asm("r13");

That one is fine, because it is a global var.

https://gcc.gnu.org/onlinedocs/gcc/Explicit-Register-Variables.html


Segher

Re: [PATCH] powerpc: inline current_stack_pointer()

[Anton Blanchard]

Hi,

> current_stack_pointeur() is a single instruction function. it
> It is not worth breaking the execution flow with a bl/blr for a
> single instruction

Check out bfe9a2cfe91a ("powerpc: Reimplement __get_SP() as a function
not a define") to see why we made it a function.

Anton