[PATCH 0/2] ARM: Avoid to clear bss multiple times

[PATCH 0/2] ARM: Avoid to clear bss multiple times

[Sascha Hauer]

Currently we clear the bss everytime we call setup_c(). We can do better
and clear it only once which makes for example alloc_pte() in the PBL
MMU setup preserve its state and we can call mmu_early_enable() multiple
times without harm.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
---
Sascha Hauer (2):
      ARM: setupc_32: remove relocation code from setup_c
      ARM: setup_c: avoid clearing BSS twice

 arch/arm/cpu/common.c    |  2 ++
 arch/arm/cpu/setupc_32.S | 35 +++++++++++++----------------------
 arch/arm/cpu/setupc_64.S | 10 ++++++++--
 3 files changed, 23 insertions(+), 24 deletions(-)
---
base-commit: bd5518cd7e34861706f87c8ca67b640d6257d5b8
change-id: 20260226-arm-setup-c-30f0dccdde71

Best regards,
-- 
Sascha Hauer <s.hauer@pengutronix.de>

[PATCH 1/2] ARM: setupc_32: remove relocation code from setup_c

[Sascha Hauer]

All callers of setup_c() already call relocate_to_current_adr() or
relocate_to_adr() before setup_c(), so the relocation (memcpy) code
in setup_c is dead and also the sync_caches_for_execution() is unnecessary
Remove it and reduce setup_c to just clearing BSS.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
---
 arch/arm/cpu/setupc_32.S | 27 +++++----------------------
 1 file changed, 5 insertions(+), 22 deletions(-)

diff --git a/arch/arm/cpu/setupc_32.S b/arch/arm/cpu/setupc_32.S
index 0134637f62..c2c3f97528 100644
--- a/arch/arm/cpu/setupc_32.S
+++ b/arch/arm/cpu/setupc_32.S
@@ -7,33 +7,16 @@
 .section .text.setupc
 
 /*
- * setup_c: copy binary to link address, clear bss and
- * continue executing at new address.
- *
- * This function does not return to the address it is
- * called from, but to the same location in the copied
- * binary.
+ * setup_c: clear bss
  */
 ENTRY(setup_c)
-	push	{r4, r5}
-	mov	r5, lr
-	bl	get_runtime_offset
-	subs	r4, r0, #0
-	beq	1f			/* skip memcpy if already at correct address */
-	ldr	r0,=_text
-	ldr	r2,=__bss_start
-	sub	r2, r2, r0
-	add	r1, r0, r4
-	bl	__memcpy			/* memcpy(_text, _text + offset, __bss_start - _text) */
-1:	ldr	r0, =__bss_start
+	mov	r4, lr
+	ldr	r0, =__bss_start
 	mov	r1, #0
 	ldr	r2, =__bss_stop
 	sub	r2, r2, r0
-	bl	__memset			/* clear bss */
-	bl	sync_caches_for_execution
-	sub	lr, r5, r4		/* adjust return address to new location */
-	pop	{r4, r5}
-	ret	lr
+	bl	__memset		/* clear bss */
+	ret	r4
 ENDPROC(setup_c)
 
 /*

-- 
2.47.3

[PATCH 2/2] ARM: setup_c: avoid clearing BSS twice

[Sascha Hauer]

Add a bss_cleared variable that is set after the first call to
setup_c(). On subsequent calls, the BSS clear is skipped to avoid
wiping already-initialized data.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
---
 arch/arm/cpu/common.c    |  2 ++
 arch/arm/cpu/setupc_32.S | 10 +++++++++-
 arch/arm/cpu/setupc_64.S | 10 ++++++++--
 3 files changed, 19 insertions(+), 3 deletions(-)

diff --git a/arch/arm/cpu/common.c b/arch/arm/cpu/common.c
index e86c89f808..40742894c2 100644
--- a/arch/arm/cpu/common.c
+++ b/arch/arm/cpu/common.c
@@ -77,6 +77,8 @@ int __pure cpu_architecture(void)
 }
 #endif
 
+int bss_cleared;
+
 extern int __boot_cpu_mode;
 
 int boot_cpu_mode(void)
diff --git a/arch/arm/cpu/setupc_32.S b/arch/arm/cpu/setupc_32.S
index c2c3f97528..d923a8f621 100644
--- a/arch/arm/cpu/setupc_32.S
+++ b/arch/arm/cpu/setupc_32.S
@@ -7,16 +7,24 @@
 .section .text.setupc
 
 /*
- * setup_c: clear bss
+ * setup_c: clear bss if not yet done
  */
 ENTRY(setup_c)
+	ldr	r0, =bss_cleared
+	ldr	r1, [r0]
+	cmp	r1, #0
+	bne	1f			/* skip if already done */
 	mov	r4, lr
 	ldr	r0, =__bss_start
 	mov	r1, #0
 	ldr	r2, =__bss_stop
 	sub	r2, r2, r0
 	bl	__memset		/* clear bss */
+	ldr	r0, =bss_cleared
+	mov	r1, #1
+	str	r1, [r0]		/* mark bss cleared */
 	ret	r4
+1:	ret	lr
 ENDPROC(setup_c)
 
 /*
diff --git a/arch/arm/cpu/setupc_64.S b/arch/arm/cpu/setupc_64.S
index fd95187a04..8262deb512 100644
--- a/arch/arm/cpu/setupc_64.S
+++ b/arch/arm/cpu/setupc_64.S
@@ -7,17 +7,23 @@
 .section .text.setupc
 
 /*
- * setup_c: clear bss
+ * setup_c: clear bss if not yet done
  */
 ENTRY(setup_c)
+	adr_l	x0, bss_cleared
+	ldr	w1, [x0]
+	cbnz	w1, 1f			/* skip if already done */
 	mov	x15, x30
 	adr_l	x0, __bss_start
 	mov	x1, #0
 	adr_l	x2, __bss_stop
 	sub	x2, x2, x0
 	bl	__memset		/* clear bss */
+	adr_l	x0, bss_cleared
+	mov	w1, #1
+	str	w1, [x0]		/* mark bss cleared */
 	mov	x30, x15
-	ret
+1:	ret
 ENDPROC(setup_c)
 
 /*

-- 
2.47.3

Re: [PATCH 2/2] ARM: setup_c: avoid clearing BSS twice

[Ahmad Fatoum]

Hi,

On 2/26/26 13:23, Sascha Hauer wrote:
> Add a bss_cleared variable that is set after the first call to
> setup_c(). On subsequent calls, the BSS clear is skipped to avoid
> wiping already-initialized data.
> 
> Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
> Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
> ---
>  arch/arm/cpu/common.c    |  2 ++
>  arch/arm/cpu/setupc_32.S | 10 +++++++++-
>  arch/arm/cpu/setupc_64.S | 10 ++++++++--
>  3 files changed, 19 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/arm/cpu/common.c b/arch/arm/cpu/common.c
> index e86c89f808..40742894c2 100644
> --- a/arch/arm/cpu/common.c
> +++ b/arch/arm/cpu/common.c
> @@ -77,6 +77,8 @@ int __pure cpu_architecture(void)
>  }
>  #endif
>  
> +int bss_cleared;

This will go into BSS, so you implicitly assume BSS to be already
zero as this is accessed first before BSS initialization.

I suggest, along with moving it into the .data section, that you
also move it into the assembly file, so the consumer is directly
next to it:

       .pushsection .data
       .align 2
setup_c_done:
       .word 0
       .popsection

> +
>  extern int __boot_cpu_mode;
>  
>  int boot_cpu_mode(void)
> diff --git a/arch/arm/cpu/setupc_32.S b/arch/arm/cpu/setupc_32.S
> index c2c3f97528..d923a8f621 100644
> --- a/arch/arm/cpu/setupc_32.S
> +++ b/arch/arm/cpu/setupc_32.S
> @@ -7,16 +7,24 @@
>  .section .text.setupc
>  
>  /*
> - * setup_c: clear bss
> + * setup_c: clear bss if not yet done
>   */
>  ENTRY(setup_c)
> +	ldr	r0, =bss_cleared
> +	ldr	r1, [r0]
> +	cmp	r1, #0
> +	bne	1f			/* skip if already done */
>  	mov	r4, lr

Move this instruction to the start, so ...

>  	ldr	r0, =__bss_start
>  	mov	r1, #0
>  	ldr	r2, =__bss_stop
>  	sub	r2, r2, r0
>  	bl	__memset		/* clear bss */
> +	ldr	r0, =bss_cleared
> +	mov	r1, #1
> +	str	r1, [r0]		/* mark bss cleared */
>  	ret	r4
> +1:	ret	lr

This instruction can be dropped and 1: just points at the existing
ret r4?

>  ENDPROC(setup_c)
>  
>  /*
> diff --git a/arch/arm/cpu/setupc_64.S b/arch/arm/cpu/setupc_64.S
> index fd95187a04..8262deb512 100644
> --- a/arch/arm/cpu/setupc_64.S
> +++ b/arch/arm/cpu/setupc_64.S
> @@ -7,17 +7,23 @@
>  .section .text.setupc
>  
>  /*
> - * setup_c: clear bss
> + * setup_c: clear bss if not yet done
>   */
>  ENTRY(setup_c)
> +	adr_l	x0, bss_cleared
> +	ldr	w1, [x0]
> +	cbnz	w1, 1f			/* skip if already done */
>  	mov	x15, x30
>  	adr_l	x0, __bss_start
>  	mov	x1, #0
>  	adr_l	x2, __bss_stop
>  	sub	x2, x2, x0
>  	bl	__memset		/* clear bss */
> +	adr_l	x0, bss_cleared
> +	mov	w1, #1
> +	str	w1, [x0]		/* mark bss cleared */
>  	mov	x30, x15
> -	ret
> +1:	ret
>  ENDPROC(setup_c)
>  
>  /*
> 


-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |