Skip to contentSkip to conversation

kernel: stop_machine: report (un)dead code (and feedback request)

DISCUSSION Valentin Rothberg 3 participants 3 messages Nov 26, 2015 – Nov 27, 2015

lore
Share & tools
lore
#0 lore raw

Hi Chris,

your commit 4c477de14237 ("kernel: remove stop_machine() Kconfig dependency") has shown up in today's linux-next tree (20151126). The commit changes the #ifdef condition in kernel/stop_machine.c from/to:

    -#if defined(CONFIG_STOP_MACHINE) && defined(CONFIG_SMP)
    +#if defined(CONFIG_SMP) || defined(CONFIG_HOTPLUG_CPU)

Although this change fixes certain configs on X86, the condition now is a tautology since CONFIG_SMP is already required to compile the file:

    kernel/Makefile:65:obj-$(CONFIG_SMP) += stop_machine.o

AFAIU, we can safely remove this #ifdef?

I detected this issue with undertaker-checkpatch [1]. We're a research group at the University of Erlangen-Nuremberg, running daily checks on linux-next to report Kconfig related issues, such as dead and undead code, or references on undefined symbols.

One of the students, Patrick, who writes his Bachelor's thesis in our group implemented a web-framework to visualize kernel code w.r.t. CPP #ifdef statements and the bugs our tools [1] detect. You can find the visualization of the mentioned issue at [2]. The description of the tautology in the #ifdef condition can be found by clicking on the 'details' menu of the block.

We'll be happy to receive any kind of feedback of the recipients of this mail. Do you like/dislike the visualization? Does it help you to understand the described issue? etc. The main purpose of the framework is to aid when sending reports. It's much easier and nicer to point to a website visualizing a problem than just describing it textually.

Kind regards,
Valentin

[1] https://undertaker.cs.fau.de
[2] http://cados.cs.fau.de:4567/4c477de14237193c99329ca82965c1211b4ee81e/kernel/stop_machine.c

#1 lore raw
Quoted 10 lines · Valentin Rothberg

Hi Chris,

your commit 4c477de14237 ("kernel: remove stop_machine() Kconfig dependency") has shown up in today's linux-next tree (20151126). The commit changes the #ifdef condition in kernel/stop_machine.c from/to:

    -#if defined(CONFIG_STOP_MACHINE) && defined(CONFIG_SMP)
    +#if defined(CONFIG_SMP) || defined(CONFIG_HOTPLUG_CPU)
Quoted 6 lines

Although this change fixes certain configs on X86, the condition now is a tautology since CONFIG_SMP is already required to compile the file:

    kernel/Makefile:65:obj-$(CONFIG_SMP) += stop_machine.o

AFAIU, we can safely remove this #ifdef?

That seems logical. The argument in favour of it would be to keep the ifdeffery around the function defintion the same as the function declaration in stop_machine.h.
-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#2 lore raw
Quoted 14 lines · Chris Wilson
Quoted 13 lines · Valentin Rothberg

Hi Chris,

your commit 4c477de14237 ("kernel: remove stop_machine() Kconfig dependency") has shown up in today's linux-next tree (20151126). The commit changes the #ifdef condition in kernel/stop_machine.c from/to:

    -#if defined(CONFIG_STOP_MACHINE) && defined(CONFIG_SMP)
    +#if defined(CONFIG_SMP) || defined(CONFIG_HOTPLUG_CPU)

Although this change fixes certain configs on X86, the condition now is a tautology since CONFIG_SMP is already required to compile the file:

Quoted 7 lines
Quoted 3 lines
    kernel/Makefile:65:obj-$(CONFIG_SMP) += stop_machine.o

AFAIU, we can safely remove this #ifdef?

That seems logical. The argument in favour of it would be to keep the ifdeffery around the function defintion the same as the function declaration in stop_machine.h.

But this would introduce a bit of fragility: we could re-introduce the same regression that the commit fixes, if we ever changed the SMP dependency for stop_machine.c.

Thanks,

Ingo