[PATCH v3 1/2] powerpc/ftrace: support CONFIG_FUNCTION_GRAPH_RETVAL

[PATCH v3 1/2] powerpc/ftrace: support CONFIG_FUNCTION_GRAPH_RETVAL

[Aditya Bodkhe]

commit a1be9ccc57f0 ("function_graph: Support recording and printing the
return value of function") introduced support for function graph return
value tracing.

Additionally, commit a3ed4157b7d8 ("fgraph: Replace fgraph_ret_regs with
ftrace_regs") further refactored and optimized the implementation,
making `struct fgraph_ret_regs` unnecessary.

This patch enables the above modifications for powerpc all, ensuring that
function graph return value tracing is available on this architecture.

In this patch we have redefined two functions:
- 'ftrace_regs_get_return_value()' - the existing implementation on
ppc returns -ve of return value based on some conditions not
relevant to our patch.
- 'ftrace_regs_get_frame_pointer()' - always returns 0 in current code .

We also allocate stack space to equivalent of 'SWITCH_FRAME_SIZE',
allowing us to directly use predefined offsets like 'GPR3' and 'GPR4'
this keeps code clean and consistent with already defined offsets .

After this patch, v6.14+ kernel can also be built with FPROBE on powerpc
but there are a few other build and runtime dependencies for FPROBE to
work properly. The next patch addresses them.

Tested-by: Venkat Rao Bagalkote <venkat88@linux.ibm.com>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Aditya Bodkhe <adityab1@linux.ibm.com>
---

Changelog:
v2 -> v3:
- Rebase on Linux 6.17-rc4 

v2:https://lore.kernel.org/all/20250722085648.1640-1-adityab1@linux.ibm.com/ 

v1 -> v2:
- Added explanation for redefining ftrace_regs_get_return_value()
and ftrace_regs_get_frame_pointer()
- Explained why stack space equivalent to 'SWITCH_FRAME_SIZE' is allocated.

v1:https://lore.kernel.org/all/20250528134820.74121-1-adityab1@linux.ibm.com/ 

---
 arch/powerpc/Kconfig                     |  1 +
 arch/powerpc/include/asm/ftrace.h        | 15 +++++++++
 arch/powerpc/kernel/trace/ftrace_entry.S | 42 ++++++++++++++----------
 3 files changed, 41 insertions(+), 17 deletions(-)

diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
index 93402a1d9c9f..451d0e140051 100644
--- a/arch/powerpc/Kconfig
+++ b/arch/powerpc/Kconfig
@@ -246,6 +246,7 @@ config PPC
 	select HAVE_FUNCTION_ARG_ACCESS_API
 	select HAVE_FUNCTION_DESCRIPTORS	if PPC64_ELF_ABI_V1
 	select HAVE_FUNCTION_ERROR_INJECTION
+	select HAVE_FUNCTION_GRAPH_FREGS
 	select HAVE_FUNCTION_GRAPH_TRACER
 	select HAVE_FUNCTION_TRACER		if !COMPILE_TEST && (PPC64 || (PPC32 && CC_IS_GCC))
 	select HAVE_GCC_PLUGINS			if GCC_VERSION >= 50200   # plugin support on gcc <= 5.1 is buggy on PPC
diff --git a/arch/powerpc/include/asm/ftrace.h b/arch/powerpc/include/asm/ftrace.h
index 82da7c7a1d12..6ffc9c9cf4e3 100644
--- a/arch/powerpc/include/asm/ftrace.h
+++ b/arch/powerpc/include/asm/ftrace.h
@@ -50,6 +50,21 @@ static __always_inline struct pt_regs *arch_ftrace_get_regs(struct ftrace_regs *
 		asm volatile("mfmsr %0" : "=r" ((_regs)->msr));	\
 	} while (0)
 
+#undef ftrace_regs_get_return_value
+static __always_inline unsigned long
+ftrace_regs_get_return_value(const struct ftrace_regs *fregs)
+{
+	return arch_ftrace_regs(fregs)->regs.gpr[3];
+}
+#define ftrace_regs_get_return_value ftrace_regs_get_return_value
+
+#undef ftrace_regs_get_frame_pointer
+static __always_inline unsigned long
+ftrace_regs_get_frame_pointer(const struct ftrace_regs *fregs)
+{
+	return arch_ftrace_regs(fregs)->regs.gpr[1];
+}
+
 static __always_inline void
 ftrace_regs_set_instruction_pointer(struct ftrace_regs *fregs,
 				    unsigned long ip)
diff --git a/arch/powerpc/kernel/trace/ftrace_entry.S b/arch/powerpc/kernel/trace/ftrace_entry.S
index 3565c67fc638..6599fe3c6234 100644
--- a/arch/powerpc/kernel/trace/ftrace_entry.S
+++ b/arch/powerpc/kernel/trace/ftrace_entry.S
@@ -409,23 +409,31 @@ EXPORT_SYMBOL(_mcount)
 _GLOBAL(return_to_handler)
 	/* need to save return values */
 #ifdef CONFIG_PPC64
-	std	r4,  -32(r1)
-	std	r3,  -24(r1)
+	stdu	r1, -SWITCH_FRAME_SIZE(r1)
+	std	r4, GPR4(r1)
+	std	r3, GPR3(r1)
+	/* Save previous stack pointer (r1) */
+	addi	r3, r1, SWITCH_FRAME_SIZE
+	std	r3, GPR1(r1)
 	/* save TOC */
-	std	r2,  -16(r1)
-	std	r31, -8(r1)
+	std	r2,  24(r1)
+	std	r31, 32(r1)
 	mr	r31, r1
-	stdu	r1, -112(r1)
-
+	/* pass ftrace_regs/pt_regs to ftrace_return_to_handler */
+	addi	r3, r1, STACK_INT_FRAME_REGS
 	/*
 	 * We might be called from a module.
 	 * Switch to our TOC to run inside the core kernel.
 	 */
 	LOAD_PACA_TOC()
 #else
-	stwu	r1, -16(r1)
-	stw	r3, 8(r1)
-	stw	r4, 12(r1)
+	stwu	r1, -SWITCH_FRAME_SIZE(r1)
+	stw	r4, GPR4(r1)
+	stw	r3, GPR3(r1)
+	addi	r3, r1, SWITCH_FRAME_SIZE
+	stw	r3, GPR1(r1)
+	/* pass ftrace_regs/pt_regs to ftrace_return_to_handler */
+	addi	r3, r1, STACK_INT_FRAME_REGS
 #endif
 
 	bl	ftrace_return_to_handler
@@ -435,15 +443,15 @@ _GLOBAL(return_to_handler)
 	mtlr	r3
 
 #ifdef CONFIG_PPC64
-	ld	r1, 0(r1)
-	ld	r4,  -32(r1)
-	ld	r3,  -24(r1)
-	ld	r2,  -16(r1)
-	ld	r31, -8(r1)
+	ld	r4,  GPR4(r1)
+	ld	r3,  GPR3(r1)
+	ld	r2,  24(r1)
+	ld	r31, 32(r1)
+	ld	r1,  0(r1)
 #else
-	lwz	r3, 8(r1)
-	lwz	r4, 12(r1)
-	addi	r1, r1, 16
+	lwz	r3, GPR3(r1)
+	lwz	r4, GPR4(r1)
+	addi	r1, r1, SWITCH_FRAME_SIZE
 #endif
 
 	/* Jump back to real return address */
-- 
2.51.0

[PATCH v3 2/2] powerpc/fprobe: fix updated fprobe for function-graph tracer

[Aditya Bodkhe]

From: Hari Bathini <hbathini@linux.ibm.com>

Since commit 4346ba160409 ("fprobe: Rewrite fprobe on function-graph
tracer"), FPROBE depends on HAVE_FUNCTION_GRAPH_FREGS. With previous
patch adding HAVE_FUNCTION_GRAPH_FREGS for powerpc, FPROBE can be
enabled on powerpc. But with the commit b5fa903b7f7c ("fprobe: Add
fprobe_header encoding feature"), asm/fprobe.h header is needed to
define arch dependent encode/decode macros. The fprobe header MSB
pattern on powerpc is not 0xf. So, define FPROBE_HEADER_MSB_PATTERN
expected on powerpc.

Also, commit 762abbc0d09f ("fprobe: Use ftrace_regs in fprobe exit
handler") introduced HAVE_FTRACE_REGS_HAVING_PT_REGS for archs that
have pt_regs in ftrace_regs. Advertise that on powerpc to reuse
common definitions like ftrace_partial_regs().

Acked-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Signed-off-by: Hari Bathini <hbathini@linux.ibm.com>
Signed-off-by: Aditya Bodkhe <aditya.b1@linux.ibm.com>
---
 arch/powerpc/Kconfig              |  1 +
 arch/powerpc/include/asm/fprobe.h | 12 ++++++++++++
 2 files changed, 13 insertions(+)
 create mode 100644 arch/powerpc/include/asm/fprobe.h

diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
index 451d0e140051..35d32cc43255 100644
--- a/arch/powerpc/Kconfig
+++ b/arch/powerpc/Kconfig
@@ -243,6 +243,7 @@ config PPC
 	select HAVE_EFFICIENT_UNALIGNED_ACCESS
 	select HAVE_GUP_FAST
 	select HAVE_FTRACE_GRAPH_FUNC
+	select HAVE_FTRACE_REGS_HAVING_PT_REGS
 	select HAVE_FUNCTION_ARG_ACCESS_API
 	select HAVE_FUNCTION_DESCRIPTORS	if PPC64_ELF_ABI_V1
 	select HAVE_FUNCTION_ERROR_INJECTION
diff --git a/arch/powerpc/include/asm/fprobe.h b/arch/powerpc/include/asm/fprobe.h
new file mode 100644
index 000000000000..d64bc28fb3d3
--- /dev/null
+++ b/arch/powerpc/include/asm/fprobe.h
@@ -0,0 +1,12 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _ASM_PPC_FPROBE_H
+#define _ASM_PPC_FPROBE_H
+
+#include <asm-generic/fprobe.h>
+
+#ifdef CONFIG_64BIT
+#undef FPROBE_HEADER_MSB_PATTERN
+#define FPROBE_HEADER_MSB_PATTERN	(PAGE_OFFSET & ~FPROBE_HEADER_MSB_MASK)
+#endif
+
+#endif /* _ASM_PPC_FPROBE_H */
-- 
2.51.0

Re: [PATCH v3 1/2] powerpc/ftrace: support CONFIG_FUNCTION_GRAPH_RETVAL

[Hari Bathini]

On 09/09/25 10:26 am, Aditya Bodkhe wrote:
> commit a1be9ccc57f0 ("function_graph: Support recording and printing the
> return value of function") introduced support for function graph return
> value tracing.
> 
> Additionally, commit a3ed4157b7d8 ("fgraph: Replace fgraph_ret_regs with
> ftrace_regs") further refactored and optimized the implementation,
> making `struct fgraph_ret_regs` unnecessary.
> 
> This patch enables the above modifications for powerpc all, ensuring that
> function graph return value tracing is available on this architecture.
> 
> In this patch we have redefined two functions:
> - 'ftrace_regs_get_return_value()' - the existing implementation on
> ppc returns -ve of return value based on some conditions not
> relevant to our patch.
> - 'ftrace_regs_get_frame_pointer()' - always returns 0 in current code .
> 
> We also allocate stack space to equivalent of 'SWITCH_FRAME_SIZE',
> allowing us to directly use predefined offsets like 'GPR3' and 'GPR4'
> this keeps code clean and consistent with already defined offsets .
> 
> After this patch, v6.14+ kernel can also be built with FPROBE on powerpc
> but there are a few other build and runtime dependencies for FPROBE to
> work properly. The next patch addresses them.
> 

Looks good to me.

Reviewed-by: Hari Bathini <hbathini@linux.ibm.com>

> Tested-by: Venkat Rao Bagalkote <venkat88@linux.ibm.com>
> Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
> Signed-off-by: Aditya Bodkhe <adityab1@linux.ibm.com>
> ---
> 
> Changelog:
> v2 -> v3:
> - Rebase on Linux 6.17-rc4
> 
> v2:https://lore.kernel.org/all/20250722085648.1640-1-adityab1@linux.ibm.com/
> 
> v1 -> v2:
> - Added explanation for redefining ftrace_regs_get_return_value()
> and ftrace_regs_get_frame_pointer()
> - Explained why stack space equivalent to 'SWITCH_FRAME_SIZE' is allocated.
> 
> v1:https://lore.kernel.org/all/20250528134820.74121-1-adityab1@linux.ibm.com/
> 
> ---
>   arch/powerpc/Kconfig                     |  1 +
>   arch/powerpc/include/asm/ftrace.h        | 15 +++++++++
>   arch/powerpc/kernel/trace/ftrace_entry.S | 42 ++++++++++++++----------
>   3 files changed, 41 insertions(+), 17 deletions(-)
> 
> diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
> index 93402a1d9c9f..451d0e140051 100644
> --- a/arch/powerpc/Kconfig
> +++ b/arch/powerpc/Kconfig
> @@ -246,6 +246,7 @@ config PPC
>   	select HAVE_FUNCTION_ARG_ACCESS_API
>   	select HAVE_FUNCTION_DESCRIPTORS	if PPC64_ELF_ABI_V1
>   	select HAVE_FUNCTION_ERROR_INJECTION
> +	select HAVE_FUNCTION_GRAPH_FREGS
>   	select HAVE_FUNCTION_GRAPH_TRACER
>   	select HAVE_FUNCTION_TRACER		if !COMPILE_TEST && (PPC64 || (PPC32 && CC_IS_GCC))
>   	select HAVE_GCC_PLUGINS			if GCC_VERSION >= 50200   # plugin support on gcc <= 5.1 is buggy on PPC
> diff --git a/arch/powerpc/include/asm/ftrace.h b/arch/powerpc/include/asm/ftrace.h
> index 82da7c7a1d12..6ffc9c9cf4e3 100644
> --- a/arch/powerpc/include/asm/ftrace.h
> +++ b/arch/powerpc/include/asm/ftrace.h
> @@ -50,6 +50,21 @@ static __always_inline struct pt_regs *arch_ftrace_get_regs(struct ftrace_regs *
>   		asm volatile("mfmsr %0" : "=r" ((_regs)->msr));	\
>   	} while (0)
>   
> +#undef ftrace_regs_get_return_value
> +static __always_inline unsigned long
> +ftrace_regs_get_return_value(const struct ftrace_regs *fregs)
> +{
> +	return arch_ftrace_regs(fregs)->regs.gpr[3];
> +}
> +#define ftrace_regs_get_return_value ftrace_regs_get_return_value
> +
> +#undef ftrace_regs_get_frame_pointer
> +static __always_inline unsigned long
> +ftrace_regs_get_frame_pointer(const struct ftrace_regs *fregs)
> +{
> +	return arch_ftrace_regs(fregs)->regs.gpr[1];
> +}
> +
>   static __always_inline void
>   ftrace_regs_set_instruction_pointer(struct ftrace_regs *fregs,
>   				    unsigned long ip)
> diff --git a/arch/powerpc/kernel/trace/ftrace_entry.S b/arch/powerpc/kernel/trace/ftrace_entry.S
> index 3565c67fc638..6599fe3c6234 100644
> --- a/arch/powerpc/kernel/trace/ftrace_entry.S
> +++ b/arch/powerpc/kernel/trace/ftrace_entry.S
> @@ -409,23 +409,31 @@ EXPORT_SYMBOL(_mcount)
>   _GLOBAL(return_to_handler)
>   	/* need to save return values */
>   #ifdef CONFIG_PPC64
> -	std	r4,  -32(r1)
> -	std	r3,  -24(r1)
> +	stdu	r1, -SWITCH_FRAME_SIZE(r1)
> +	std	r4, GPR4(r1)
> +	std	r3, GPR3(r1)
> +	/* Save previous stack pointer (r1) */
> +	addi	r3, r1, SWITCH_FRAME_SIZE
> +	std	r3, GPR1(r1)
>   	/* save TOC */
> -	std	r2,  -16(r1)
> -	std	r31, -8(r1)
> +	std	r2,  24(r1)
> +	std	r31, 32(r1)
>   	mr	r31, r1
> -	stdu	r1, -112(r1)
> -
> +	/* pass ftrace_regs/pt_regs to ftrace_return_to_handler */
> +	addi	r3, r1, STACK_INT_FRAME_REGS
>   	/*
>   	 * We might be called from a module.
>   	 * Switch to our TOC to run inside the core kernel.
>   	 */
>   	LOAD_PACA_TOC()
>   #else
> -	stwu	r1, -16(r1)
> -	stw	r3, 8(r1)
> -	stw	r4, 12(r1)
> +	stwu	r1, -SWITCH_FRAME_SIZE(r1)
> +	stw	r4, GPR4(r1)
> +	stw	r3, GPR3(r1)
> +	addi	r3, r1, SWITCH_FRAME_SIZE
> +	stw	r3, GPR1(r1)
> +	/* pass ftrace_regs/pt_regs to ftrace_return_to_handler */
> +	addi	r3, r1, STACK_INT_FRAME_REGS
>   #endif
>   
>   	bl	ftrace_return_to_handler
> @@ -435,15 +443,15 @@ _GLOBAL(return_to_handler)
>   	mtlr	r3
>   
>   #ifdef CONFIG_PPC64
> -	ld	r1, 0(r1)
> -	ld	r4,  -32(r1)
> -	ld	r3,  -24(r1)
> -	ld	r2,  -16(r1)
> -	ld	r31, -8(r1)
> +	ld	r4,  GPR4(r1)
> +	ld	r3,  GPR3(r1)
> +	ld	r2,  24(r1)
> +	ld	r31, 32(r1)
> +	ld	r1,  0(r1)
>   #else
> -	lwz	r3, 8(r1)
> -	lwz	r4, 12(r1)
> -	addi	r1, r1, 16
> +	lwz	r3, GPR3(r1)
> +	lwz	r4, GPR4(r1)
> +	addi	r1, r1, SWITCH_FRAME_SIZE
>   #endif
>   
>   	/* Jump back to real return address */