Skip to contentSkip to conversation

riscv: ftrace: Fix the logic issue in DYNAMIC_FTRACE selection

PATCH In review ChenMiao 3 participants 7 messages Jul 3, 2025 – Jul 4, 2025 1 message not in the archive RISC-V

lore
Share & tools
lore

ChenMiao

author
#0 lore raw

PATCH 1/1 riscv: ftrace: Fix the logic issue in DYNAMIC_FTRACE selection

From: chenmiao <[email protected]>

When I was reading the source code of ftrace, I learned that ftrace has two types: static and dynamic. Initially, I planned to prioritize reading the static source code, so I disabled the enable dynamic option in RISCV.

[*]   Kernel Function Tracer
[ ]   Kernel Function Graph Tracer
[ ]   enable/disable function tracing dynamically (NEW)

However, when I tried to compile it, the build failed.

./include/linux/ftrace.h:190:16: error: implicit declaration of function ‘arch_ftrace_get_regs’; did you mean ‘arch_ftrace_regs’? [-Wimplicit-function-declaration]

  190 |         return arch_ftrace_get_regs(fregs);
      |                ^~~~~~~~~~~~~~~~~~~~
      |                arch_ftrace_regs

After comparing it with the ARM64 architecture, I found that ARM64 automatically enables DYNAMIC_FTRACE by default once FUNCTION_TRACER is turned on, and this cannot be set to "no". Therefore, I believe the optional DYNAMIC_FTRACE setting in RISC-V has a logic flaw—if FUNCTION_TRACER is enabled, DYNAMIC_FTRACE should also be enabled, and vice versa. Moreover, it's clear that RISC-V lacks the necessary support to successfully compile the kernel when DYNAMIC_FTRACE is disabled.

[*]   Kernel Function Tracer
[ ]   Kernel Function Graph Tracer
-*-   enable/disable function tracing dynamically
---
 arch/riscv/Kconfig | 1 +
 1 file changed, 1 insertion(+)
arch/riscv/Kconfig +1 -0
index 36061f473..f7fc8b460 100644
@@ -97,6 +97,7 @@ config RISCV
 	select CLONE_BACKWARDS
 	select COMMON_CLK
 	select CPU_PM if CPU_IDLE || HIBERNATION || SUSPEND
+	select DYNAMIC_FTRACE if FUNCTION_TRACER
 	select EDAC_SUPPORT
 	select FRAME_POINTER if PERF_EVENTS || (FUNCTION_TRACER && !DYNAMIC_FTRACE)
 	select FTRACE_MCOUNT_USE_PATCHABLE_FUNCTION_ENTRY if DYNAMIC_FTRACE
Signature · 1 line
2.45.2
#1 lore raw

Hi ChenMiao,

Quoted 20 lines · ChenMiao

From: chenmiao <[email protected]>

When I was reading the source code of ftrace, I learned that ftrace has two types: static and dynamic. Initially, I planned to prioritize reading the static source code, so I disabled the enable dynamic option in RISCV.

[*]   Kernel Function Tracer
[ ]   Kernel Function Graph Tracer
[ ]   enable/disable function tracing dynamically (NEW)

However, when I tried to compile it, the build failed.

./include/linux/ftrace.h:190:16: error: implicit declaration of function ‘arch_ftrace_get_regs’; did you mean ‘arch_ftrace_regs’? [-Wimplicit-function-declaration]

   190 |         return arch_ftrace_get_regs(fregs);
       |                ^~~~~~~~~~~~~~~~~~~~
       |                arch_ftrace_regs
Quoted 8 lines

After comparing it with the ARM64 architecture, I found that ARM64 automatically enables DYNAMIC_FTRACE by default once FUNCTION_TRACER is turned on, and this cannot be set to "no". Therefore, I believe the optional DYNAMIC_FTRACE setting in RISC-V has a logic flaw—if FUNCTION_TRACER is enabled, DYNAMIC_FTRACE should also be enabled, and vice versa. Moreover, it's clear that RISC-V lacks the necessary support to successfully compile the kernel when DYNAMIC_FTRACE is disabled.

We could support static ftrace, but I don't think we should, so I agree with this patch. In fact I had just prepared a patch for this here https://github.com/linux-riscv/linux/pull/556/commits/0481092a5bec3818658981c11f629e06e66382b3 which is a bit more complete since I have removed some dead code.

Let's see what other people think about supporting static ftrace, I have added Steven in cc if he has an opinion.

Thanks,

Alex

Quoted 22 lines
[*]   Kernel Function Tracer
[ ]   Kernel Function Graph Tracer
-*-   enable/disable function tracing dynamically
---
  arch/riscv/Kconfig | 1 +
  1 file changed, 1 insertion(+)
arch/riscv/Kconfig +1 -0
index 36061f473..f7fc8b460 100644
@@ -97,6 +97,7 @@ config RISCV
  	select CLONE_BACKWARDS
  	select COMMON_CLK
  	select CPU_PM if CPU_IDLE || HIBERNATION || SUSPEND
+	select DYNAMIC_FTRACE if FUNCTION_TRACER
  	select EDAC_SUPPORT
  	select FRAME_POINTER if PERF_EVENTS || (FUNCTION_TRACER && !DYNAMIC_FTRACE)
  	select FTRACE_MCOUNT_USE_PATCHABLE_FUNCTION_ENTRY if DYNAMIC_FTRACE
#2 lore raw
Quoted 8 lines · On Thu, 3 Jul 2025 15:00:02 +0200 Alexandre Ghiti <[email protected]> wrote:

We could support static ftrace, but I don't think we should, so I agree with this patch. In fact I had just prepared a patch for this here
https://github.com/linux-riscv/linux/pull/556/commits/0481092a5bec3818658981c11f629e06e66382b3
which is a bit more complete since I have removed some dead code.

Let's see what other people think about supporting static ftrace, I have added Steven in cc if he has an opinion.

Yes, please only support the dynamic ftrace. The static is there only to help archs to get ftrace up and running. Once dynamic is supported, static should not be used.

Hmm, maybe I should just remove the prompt for DYNAMIC_FTRACE.

That is, once it is supported by an architecture, it should be the only thing used.

-- Steve

#3 lore raw

Hi Steve,

Quoted 15 lines · Steven Rostedt
Quoted 7 lines · On Thu, 3 Jul 2025 15:00:02 +0200 Alexandre Ghiti <[email protected]> wrote:

We could support static ftrace, but I don't think we should, so I agree with this patch. In fact I had just prepared a patch for this here
https://github.com/linux-riscv/linux/pull/556/commits/0481092a5bec3818658981c11f629e06e66382b3
which is a bit more complete since I have removed some dead code.

Let's see what other people think about supporting static ftrace, I have added Steven in cc if he has an opinion.

Yes, please only support the dynamic ftrace. The static is there only to help archs to get ftrace up and running. Once dynamic is supported, static should not be used.

Quoted 6 lines

Hmm, maybe I should just remove the prompt for DYNAMIC_FTRACE.

That is, once it is supported by an architecture, it should be the only thing used.

-- Steve

Thanks for your input.

@ChenMiao: can you come up with a v2 that, in addition, deletes the dead code and with a commit log that explains what Steven said? If not possible for you, let me know and I'll do it.

Thanks,

Alex

A message that is not in the archive (sent off-list or not yet received): CAKxVwgeDcqo83ZV+xBcHwNuMk6yeU+yp7RYo22OARAVOBgrsJQ@mail.gmail.com

#5 lore raw

Hi Chen Miao,

Quoted 3 lines · Miao Chen

Hello Alex,

Should I wait for Steven to finish his work before starting mine? He seems to be refactoring a lot of CONFIG logic, which might cause conflicts.

I would say no, don't wait for Steven since his work won't land in 6.16, we need a fix to prevent build failures for this release.

Thanks,

Alex

Quoted 50 lines

Thanks,

Chen Miao

Alexandre Ghiti <[email protected]> 于2025年7月4日周五 00:06写道:

Hi Steve,

On 7/3/25 17:40, Steven Rostedt wrote:
> On Thu, 3 Jul 2025 15:00:02 +0200
> Alexandre Ghiti <[email protected]> wrote:
>
>
>> We could support static ftrace, but I don't think we should, so I agree
>> with this patch. In fact I had just prepared a patch for this here >>
https://github.com/linux-riscv/linux/pull/556/commits/0481092a5bec3818658981c11f629e06e66382b3
>> which is a bit more complete since I have removed some dead code. >>
>> Let's see what other people think about supporting static ftrace, I have
>> added Steven in cc if he has an opinion.
> Yes, please only support the dynamic ftrace. The static is there only
> to help archs to get ftrace up and running. Once dynamic is supported,
> static should not be used.
>
> Hmm, maybe I should just remove the prompt for DYNAMIC_FTRACE. >
> That is, once it is supported by an architecture, it should be the only
> thing used.
>
> -- Steve

Thanks for your input.

@ChenMiao: can you come up with a v2 that, in addition, deletes the dead
code and with a commit log that explains what Steven said? If not possible for you, let me know and I'll do it.

Thanks,

Alex

#6 lore raw
Quoted 10 lines · Alexandre Ghiti

Hi Chen Miao,

Quoted 3 lines · Miao Chen

Hello Alex,

Should I wait for Steven to finish his work before starting mine? He seems to be refactoring a lot of CONFIG logic, which might cause conflicts.

I would say no, don't wait for Steven since his work won't land in 6.16, we need a fix to prevent build failures for this release.

Agreed,

-- Steve

Quoted 2 lines
Quoted 2 lines
author
#7 lore raw
In reply to Alexandre Ghiti

Hi Alex,

Ok, I'll start my work right now.

Thanks,

Chen Miao

Quoted 69 lines · Alexandre Ghiti <[email protected]> 于2025年7月4日周五 20:28写道:

Hi Chen Miao,

Quoted 5 lines · Miao Chen

Hello Alex,

Should I wait for Steven to finish his work before starting mine? He seems to be refactoring a lot of CONFIG logic, which might cause conflicts.

I would say no, don't wait for Steven since his work won't land in 6.16, we need a fix to prevent build failures for this release.

Thanks,

Alex

Quoted 50 lines

Thanks,

Chen Miao

Alexandre Ghiti <[email protected]> 于2025年7月4日周五 00:06写道:

Hi Steve,

On 7/3/25 17:40, Steven Rostedt wrote:
> On Thu, 3 Jul 2025 15:00:02 +0200
> Alexandre Ghiti <[email protected]> wrote:
>
>
>> We could support static ftrace, but I don't think we should, so I agree
>> with this patch. In fact I had just prepared a patch for this here >>
https://github.com/linux-riscv/linux/pull/556/commits/0481092a5bec3818658981c11f629e06e66382b3
>> which is a bit more complete since I have removed some dead code. >>
>> Let's see what other people think about supporting static ftrace, I have
>> added Steven in cc if he has an opinion.
> Yes, please only support the dynamic ftrace. The static is there only
> to help archs to get ftrace up and running. Once dynamic is supported,
> static should not be used.
>
> Hmm, maybe I should just remove the prompt for DYNAMIC_FTRACE. >
> That is, once it is supported by an architecture, it should be the only
> thing used.
>
> -- Steve

Thanks for your input.

@ChenMiao: can you come up with a v2 that, in addition, deletes the dead
code and with a commit log that explains what Steven said? If not possible for you, let me know and I'll do it.

Thanks,

Alex