Skip to contentSkip to conversation

Documentation: warn users not to use select on choice options in Kconfig

PATCH In review Julian Braha 5 participants 10 messages Jul 15 – Aug 4

Subsystems (3)
lore
Share & tools
lore

Julian Braha

author reviewer
#0 lore raw

PATCH 1/1 Documentation: warn users not to use select on choice options in Kconfig

Hengelein's master thesis about Kconfig from 2015 wrote that 'select' does not work on member options inside of a 'choice'.

So I've re-implemented his check as part of kconfirm:
https://github.com/julianbraha/kconfirm

and discovered several instances of these dead selects in use, and fixing them is not always trivial. For example, see:
https://lore.kernel.org/all/[email protected]/

Let's add a warning to the documentation (a place where developers are more likely to see it than a master's thesis...)

---
 Documentation/kbuild/kconfig-language.rst | 5 +++++
 1 file changed, 5 insertions(+)
Documentation/kbuild/kconfig-language.rst +5 -0
index d9338407c1c6..ab2822d7ca1c 100644
@@ -164,6 +164,11 @@ applicable everywhere (see syntax).
 	That will limit the usefulness but on the other hand avoid
 	the illegal configurations all over.
 
+	select should also not be used on the member options of a choice.
+	Selecting a choice option will do nothing (dead code). Instead,
+	consider making these a dependency, or using the defaults of the
+	choice to enable the member option when it's needed.
+
 	If "select" <symbol> is followed by "if" <expr>, <symbol> will be
 	selected by the logical AND of the value of the current menu symbol
 	and <expr>. This means, the lower limit can be downgraded due to the
Signature · 1 line
2.54.0
Reviewed-by on v1 1/1
#1 lore raw
Quoted 6 lines · Julian Braha

Hengelein's master thesis about Kconfig from 2015 wrote that 'select' does not work on member options inside of a 'choice'.

So I've re-implemented his check as part of kconfirm:
https://github.com/julianbraha/kconfirm

Quoted 8 lines

and discovered several instances of these dead selects in use, and fixing them is not always trivial. For example, see:
https://lore.kernel.org/all/[email protected]/

Let's add a warning to the documentation (a place where developers are more likely to see it than a master's thesis...)

Yeah, it'd be good to add a checkpatch or even build-time check too.

Signature · 1 line
Kees Cook
author reviewer
#2 lore raw

Hi Kees,

Quoted 1 line · Kees Cook

Yeah, it'd be good to add a checkpatch or even build-time check too.

I'll look into adding a check for this into the kconfig interpreter.

- Julian Braha

#3 lore raw
Quoted 6 lines · Julian Braha

Hi Kees,

Quoted 1 line · Kees Cook

Yeah, it'd be good to add a checkpatch or even build-time check too.

I'll look into adding a check for this into the kconfig interpreter.

This. I wish there was a kconfig lint option that would whine about all the things that kconfig-language.rst explicitly tells you *not* to use select for. (So I could perhaps eventually stop replying to people with that stuff.)

BR,
Jani.

Signature · 1 line
Jani Nikula, Intel
maintainer
#4 lore raw
In reply to Julian Braha
Quoted 1 line · Julian Braha

Documentation: warn users not to use select on choice options in Kconfig

Applied to kbuild/linux.git (kbuild-next-unstable), thanks!

[1/1] Documentation: warn users not to use select on choice options in Kconfig

      https://git.kernel.org/kbuild/c/d10cfe72

Please look out for regression or issue reports or other follow up comments, as they may result in the patch/series getting dropped, reverted or modified (e.g. trailers).

Patches applied to the kbuild-next-unstable branch are accepted pending wider testing in linux-next and any post-commit review; they will generally be moved to the kbuild-next branch in about a week if no issues are found.

Best regards,

Signature · 1 line
Nicolas
maintainer
#5 lore raw
In reply to Julian Braha
Quoted 2 lines · Julian Braha

Hengelein's master thesis about Kconfig from 2015 wrote that 'select' does not work on member options inside of a 'choice'.

is this true also for 'imply'? Adding the check to kconfig seems to be pretty simple:

scripts/kconfig/menu.c +5 -0
index b2d8d4e11e07..149d3ee59b58 100644
@@ -287,6 +287,11 @@ static void sym_check_prop(struct symbol *sym)
 				    "'%s' has wrong type. '%s' only "
 				    "accept arguments of bool and "
 				    "tristate type", sym2->name, use);
+			if (sym_is_choice_value(sym2))
+				prop_warn(prop,
+				    "config symbol '%s' uses %s for '%s', "
+				    "but '%s' is a choice value", sym->name,
+				    use, sym2->name, sym2->name);
 			break;
 		case P_RANGE:
 			if (sym->type != S_INT && sym->type != S_HEX)

which currently reveals

kernel/time/Kconfig:136:warning: config symbol 'NO_HZ_FULL' uses select for 'VIRT_CPU_ACCOUNTING_GEN', but 'VIRT_CPU_ACCOUNTING_GEN' is a choice value drivers/mtd/maps/Kconfig:104:warning: config symbol 'MTD_PHYSMAP_IXP4XX' uses select for 'MTD_CFI_BE_BYTE_SWAP', but 'MTD_CFI_BE_BYTE_SWAP' is a choice value drivers/tee/qcomtee/Kconfig:8:warning: config symbol 'QCOMTEE' uses select for 'QCOM_TZMEM_MODE_SHMBRIDGE', but 'QCOM_TZMEM_MODE_SHMBRIDGE' is a choice value

on linux-next, with the NO_HZ_FULL also being addressed by Julian [1].

Julian, please call out if you want help for pushing these.

I am considering to prepare a patch and add it for kbuild-for-next (but not to kbuild-next) to raise attention.

Kind regards,
Nicolas

[1]: https://lore.kernel.org/r/[email protected]

author reviewer
#6 lore raw

Hi Nicolas,

Quoted 5 lines · Nicolas Schier
Quoted 2 lines · Julian Braha

Hengelein's master thesis about Kconfig from 2015 wrote that 'select' does not work on member options inside of a 'choice'.

is this true also for 'imply'?

Yes, the same problem goes for imply. But I didn't find any of these 'imply-choice' instances in use. Of course, if we're going to warn / error for select, then it makes sense to do the same for imply.

Quoted 22 lines

Adding the check to kconfig seems to be
pretty simple:

scripts/kconfig/menu.c +5 -0
index b2d8d4e11e07..149d3ee59b58 100644
@@ -287,6 +287,11 @@ static void sym_check_prop(struct symbol *sym)
 				    "'%s' has wrong type. '%s' only "
 				    "accept arguments of bool and "
 				    "tristate type", sym2->name, use);
+			if (sym_is_choice_value(sym2))
+				prop_warn(prop,
+				    "config symbol '%s' uses %s for '%s', "
+				    "but '%s' is a choice value", sym->name,
+				    use, sym2->name, sym2->name);
 			break;
 		case P_RANGE:
 			if (sym->type != S_INT && sym->type != S_HEX)

which currently reveals

Quoted 7 lines

kernel/time/Kconfig:136:warning: config symbol 'NO_HZ_FULL' uses select for 'VIRT_CPU_ACCOUNTING_GEN', but 'VIRT_CPU_ACCOUNTING_GEN' is a choice value drivers/mtd/maps/Kconfig:104:warning: config symbol 'MTD_PHYSMAP_IXP4XX' uses select for 'MTD_CFI_BE_BYTE_SWAP', but 'MTD_CFI_BE_BYTE_SWAP' is a choice value drivers/tee/qcomtee/Kconfig:8:warning: config symbol 'QCOMTEE' uses select for 'QCOM_TZMEM_MODE_SHMBRIDGE', but 'QCOM_TZMEM_MODE_SHMBRIDGE' is a choice value

on linux-next, with the NO_HZ_FULL also being addressed by Julian [1].

Julian, please call out if you want help for pushing these.

Thanks for offering :)

I actually detected 5 of these across all architectures, and already submitted patches for all of them (thank you, Arnd, for all of your help):
1.
https://lore.kernel.org/all/[email protected]/
2.
https://lore.kernel.org/all/[email protected]/
3.
https://lore.kernel.org/all/[email protected]/
4.
https://lore.kernel.org/all/[email protected]/

With #5 already in linux-next:
5.
https://lore.kernel.org/all/[email protected]/

Quoted 2 lines

I am considering to prepare a patch and add it for kbuild-for-next (but not to kbuild-next) to raise attention.

I think the Kconfig interpreter should completely error out for these, but of course, we need to wait all of these patches to be applied first.

- Julian Braha

#7 lore raw
Quoted 20 lines · Julian Braha
Quoted 19 lines · Nicolas Schier

Adding the check to kconfig seems to be
pretty simple:

scripts/kconfig/menu.c +5 -0
index b2d8d4e11e07..149d3ee59b58 100644
@@ -287,6 +287,11 @@ static void sym_check_prop(struct symbol *sym)
 				    "'%s' has wrong type. '%s' only "
 				    "accept arguments of bool and "
 				    "tristate type", sym2->name, use);
+			if (sym_is_choice_value(sym2))
+				prop_warn(prop,
+				    "config symbol '%s' uses %s for '%s', "
+				    "but '%s' is a choice value", sym->name,
+				    use, sym2->name, sym2->name);
 			break;
 		case P_RANGE:
 			if (sym->type != S_INT && sym->type != S_HEX)

...

Quoted 16 lines

I actually detected 5 of these across all architectures, and already submitted patches for all of them (thank you, Arnd, for all of your help):
1.
https://lore.kernel.org/all/[email protected]/
2.
https://lore.kernel.org/all/[email protected]/
3.
https://lore.kernel.org/all/[email protected]/
4.
https://lore.kernel.org/all/[email protected]/

With #5 already in linux-next:
5.
https://lore.kernel.org/all/[email protected]/

Quoted 5 lines
Quoted 2 lines

I am considering to prepare a patch and add it for kbuild-for-next (but not to kbuild-next) to raise attention.

I think the Kconfig interpreter should completely error out for these, but of course, we need to wait all of these patches to be applied first.

It looks like number 4 has been picked up as well. Given that we only need three more patches to be accepted to turn this into a hard error, I think it would be better to hold off on taking the documentation patch and just push for the hard error next cycle. If these patches are not applied to other trees before 7.3-rc1, we can ask to take them via Kbuild with Acks and/or stable branches for 7.4.

Signature · 2 lines
Cheers,
Nathan
author reviewer
#8 lore raw
Quoted 6 lines · Nathan Chancellor

It looks like number 4 has been picked up as well. Given that we only need three more patches to be accepted to turn this into a hard error, I think it would be better to hold off on taking the documentation patch and just push for the hard error next cycle. If these patches are not applied to other trees before 7.3-rc1, we can ask to take them via Kbuild with Acks and/or stable branches for 7.4.

All good on my end, we can drop this patch.

- Julian Braha

maintainer
#9 lore raw
Quoted 7 lines · Julian Braha
Quoted 6 lines · Nathan Chancellor

It looks like number 4 has been picked up as well. Given that we only need three more patches to be accepted to turn this into a hard error, I think it would be better to hold off on taking the documentation patch and just push for the hard error next cycle. If these patches are not applied to other trees before 7.3-rc1, we can ask to take them via Kbuild with Acks and/or stable branches for 7.4.

sounds good. I'll prepare a patch for the hard error.

Quoted 1 line

All good on my end, we can drop this patch.

thanks, removed commit 63dad1769941 ("Documentation: warn users not to use select on choice options in Kconfig") from kbuild-next-unstable.

Quoted 2 lines

- Julian Braha

Signature · 1 line
Nicolas