[PATCH] kconfig: warn on dead default

[PATCH] kconfig: warn on dead default

[Julian Braha]

The dead default check was originally introduced with kconfirm:
https://lore.kernel.org/all/6ec4df6d-1445-48ca-8f54-1d1a83c4716d@gmail.com/

While I'm still working on that tool, it's not yet ready for inclusion
into the tree. I am currently waiting for common distro packagers to
package the parsing library before submitting the next RFC iteration.

However, the dead default check is more impactful than the other checks:
all 4 dead defaults that were detected should not have been dead and could
cause misconfiguration bugs. But fortunately, these were just for kunit
tests. The 3 patches to fix them have all since been merged:
commit aef656a0e6c0 ("powerpc: fix dead default for GUEST_STATE_BUFFER_TEST")
commit 30cc5e2ad826 ("s390/Kconfig: Cleanup defaults for selftests")
commit df75430515c3 ("drm: fix dead default for DRM_TTM_KUNIT_TEST")

We can actually check for dead defaults while evaluating Kconfig, which
should be even more effective at preventing future instances than keeping
it in a static checker.

Note that this patch will only trigger a warning when the default values
are different, in other words, pure duplicate defaults won't cause a
warning, as they are simply redundant.

Signed-off-by: Julian Braha <julianbraha@gmail.com>
---
 scripts/kconfig/menu.c                        | 22 +++++++++-
 .../kconfig/tests/warn_dead_default/Kconfig   | 40 +++++++++++++++++++
 .../tests/warn_dead_default/__init__.py       |  8 ++++
 .../tests/warn_dead_default/expected_stderr   |  4 ++
 4 files changed, 73 insertions(+), 1 deletion(-)
 create mode 100644 scripts/kconfig/tests/warn_dead_default/Kconfig
 create mode 100644 scripts/kconfig/tests/warn_dead_default/__init__.py
 create mode 100644 scripts/kconfig/tests/warn_dead_default/expected_stderr

diff --git a/scripts/kconfig/menu.c b/scripts/kconfig/menu.c
index b2d8d4e11e07..8c280292f9cd 100644
--- a/scripts/kconfig/menu.c
+++ b/scripts/kconfig/menu.c
@@ -242,13 +242,33 @@ static int menu_validate_number(struct symbol *sym, struct symbol *sym2)
 
 static void sym_check_prop(struct symbol *sym)
 {
-	struct property *prop;
+	struct property *prev, *prop;
 	struct symbol *sym2;
 	char *use;
 
 	for (prop = sym->prop; prop; prop = prop->next) {
 		switch (prop->type) {
 		case P_DEFAULT:
+			for_all_defaults(sym, prev) {
+				if (prev == prop)
+					break;
+				if (expr_is_yes(prev->visible.expr)) {
+					if (!expr_eq(prev->expr, prop->expr))
+						prop_warn(prop,
+							"default for '%s' is unreachable: earlier default at %s:%d is unconditional",
+							sym->name ? sym->name : "<choice>",
+							prev->filename, prev->lineno);
+					break;
+				}
+				if (expr_eq(prev->visible.expr, prop->visible.expr)) {
+					if (!expr_eq(prev->expr, prop->expr))
+						prop_warn(prop,
+							"default for '%s' has the same condition as the earlier default at %s:%d",
+							sym->name ? sym->name : "<choice>",
+							prev->filename, prev->lineno);
+					break;
+				}
+			}
 			if ((sym->type == S_STRING || sym->type == S_INT || sym->type == S_HEX) &&
 			    prop->expr->type != E_SYMBOL)
 				prop_warn(prop,
diff --git a/scripts/kconfig/tests/warn_dead_default/Kconfig b/scripts/kconfig/tests/warn_dead_default/Kconfig
new file mode 100644
index 000000000000..adf421d73dbd
--- /dev/null
+++ b/scripts/kconfig/tests/warn_dead_default/Kconfig
@@ -0,0 +1,40 @@
+# SPDX-License-Identifier: GPL-2.0
+
+config A
+	bool
+
+config B
+	bool
+
+config UNCONDITIONAL
+	int
+	default 1
+	default 2
+
+config CONDITIONAL
+	int
+	default 1 if A
+	default 2 if A
+	default 3 if B
+
+config CONDITIONAL_COMMUTATIVE
+	int
+	default 1 if A && B
+	default 2 if B && A
+
+config CONTROL
+	int
+	default 1 if A
+	default 2 if B
+	default 3
+
+choice
+	prompt "test choice"
+	default C
+	default D
+
+	config C
+		bool "C"
+	config D
+		bool "D"
+endchoice
diff --git a/scripts/kconfig/tests/warn_dead_default/__init__.py b/scripts/kconfig/tests/warn_dead_default/__init__.py
new file mode 100644
index 000000000000..911b30ce19fe
--- /dev/null
+++ b/scripts/kconfig/tests/warn_dead_default/__init__.py
@@ -0,0 +1,8 @@
+# SPDX-License-Identifier: GPL-2.0
+"""
+Test detection of dead defaults (different defaults that can never be active).
+"""
+
+def test(conf):
+    assert conf.olddefconfig() == 0
+    assert conf.stderr_contains('expected_stderr')
diff --git a/scripts/kconfig/tests/warn_dead_default/expected_stderr b/scripts/kconfig/tests/warn_dead_default/expected_stderr
new file mode 100644
index 000000000000..baa20bf33910
--- /dev/null
+++ b/scripts/kconfig/tests/warn_dead_default/expected_stderr
@@ -0,0 +1,4 @@
+Kconfig:12:warning: default for 'UNCONDITIONAL' is unreachable: earlier default at Kconfig:11 is unconditional
+Kconfig:17:warning: default for 'CONDITIONAL' has the same condition as the earlier default at Kconfig:16
+Kconfig:23:warning: default for 'CONDITIONAL_COMMUTATIVE' has the same condition as the earlier default at Kconfig:22
+Kconfig:34:warning: default for '<choice>' is unreachable: earlier default at Kconfig:33 is unconditional
-- 
2.53.0

Re: [PATCH] kconfig: warn on dead default

[Sam Ravnborg]

Hi Julian.

On Sat, Jun 06, 2026 at 03:00:08PM +0100, Julian Braha wrote:
> The dead default check was originally introduced with kconfirm:
> https://lore.kernel.org/all/6ec4df6d-1445-48ca-8f54-1d1a83c4716d@gmail.com/
> 
> While I'm still working on that tool, it's not yet ready for inclusion
> into the tree. I am currently waiting for common distro packagers to
> package the parsing library before submitting the next RFC iteration.
> 
> However, the dead default check is more impactful than the other checks:
> all 4 dead defaults that were detected should not have been dead and could
> cause misconfiguration bugs. But fortunately, these were just for kunit
> tests. The 3 patches to fix them have all since been merged:
> commit aef656a0e6c0 ("powerpc: fix dead default for GUEST_STATE_BUFFER_TEST")
> commit 30cc5e2ad826 ("s390/Kconfig: Cleanup defaults for selftests")
> commit df75430515c3 ("drm: fix dead default for DRM_TTM_KUNIT_TEST")
> 
> We can actually check for dead defaults while evaluating Kconfig, which
> should be even more effective at preventing future instances than keeping
> it in a static checker.
> 
> Note that this patch will only trigger a warning when the default values
> are different, in other words, pure duplicate defaults won't cause a
> warning, as they are simply redundant.

A drive-by comment below.

> 
> Signed-off-by: Julian Braha <julianbraha@gmail.com>
> ---
>  scripts/kconfig/menu.c                        | 22 +++++++++-
>  .../kconfig/tests/warn_dead_default/Kconfig   | 40 +++++++++++++++++++
>  .../tests/warn_dead_default/__init__.py       |  8 ++++
>  .../tests/warn_dead_default/expected_stderr   |  4 ++
>  4 files changed, 73 insertions(+), 1 deletion(-)
>  create mode 100644 scripts/kconfig/tests/warn_dead_default/Kconfig
>  create mode 100644 scripts/kconfig/tests/warn_dead_default/__init__.py
>  create mode 100644 scripts/kconfig/tests/warn_dead_default/expected_stderr
> 
> diff --git a/scripts/kconfig/menu.c b/scripts/kconfig/menu.c
> index b2d8d4e11e07..8c280292f9cd 100644
> --- a/scripts/kconfig/menu.c
> +++ b/scripts/kconfig/menu.c
> @@ -242,13 +242,33 @@ static int menu_validate_number(struct symbol *sym, struct symbol *sym2)
>  
>  static void sym_check_prop(struct symbol *sym)
>  {
> -	struct property *prop;
> +	struct property *prev, *prop;
>  	struct symbol *sym2;
>  	char *use;
>  
>  	for (prop = sym->prop; prop; prop = prop->next) {
This part iterates through all properties of sym.
>  		switch (prop->type) {
>  		case P_DEFAULT:
> +			for_all_defaults(sym, prev) {
And this part iterates through all properties of sym and will match only
P_DEFAULT.

I looks redundant with the two loops were we for the latter knows it is
a P_DEFAULT.

> +				if (prev == prop)
> +					break;
> +				if (expr_is_yes(prev->visible.expr)) {
> +					if (!expr_eq(prev->expr, prop->expr))
> +						prop_warn(prop,
> +							"default for '%s' is unreachable: earlier default at %s:%d is unconditional",
> +							sym->name ? sym->name : "<choice>",
> +							prev->filename, prev->lineno);
> +					break;
> +				}
> +				if (expr_eq(prev->visible.expr, prop->visible.expr)) {
> +					if (!expr_eq(prev->expr, prop->expr))
> +						prop_warn(prop,
> +							"default for '%s' has the same condition as the earlier default at %s:%d",
> +							sym->name ? sym->name : "<choice>",
> +							prev->filename, prev->lineno);
> +					break;
> +				}
> +			}
>  			if ((sym->type == S_STRING || sym->type == S_INT || sym->type == S_HEX) &&
>  			    prop->expr->type != E_SYMBOL)
>  				prop_warn(prop,

	Sam

Re: [PATCH] kconfig: warn on dead default

[Julian Braha]

On 6/6/26 15:23, Sam Ravnborg wrote:
> I looks redundant with the two loops were we for the latter knows it is
> a P_DEFAULT.

Hi Sam,

It may look redundant, but we do need to check each default we encounter
in the
```
for (prop = sym->prop; prop; prop = prop->next) {
```
loop against the previous defaults to know if we've encountered a
duplicate default condition.

I guess the other approach would be to copy each default as we encounter
them into a hashtable and check against that, instead? Let me know if
you have another, preferred approach.

- Julian Braha

Re: [PATCH] kconfig: warn on dead default

[Sam Ravnborg]

Hi Julian.

On Sat, Jun 06, 2026 at 04:01:34PM +0100, Julian Braha wrote:
> On 6/6/26 15:23, Sam Ravnborg wrote:
> > I looks redundant with the two loops were we for the latter knows it is
> > a P_DEFAULT.
> 
> Hi Sam,
> 
> It may look redundant, but we do need to check each default we encounter
> in the
> ```
> for (prop = sym->prop; prop; prop = prop->next) {
> ```
> loop against the previous defaults to know if we've encountered a
> duplicate default condition.
> 
> I guess the other approach would be to copy each default as we encounter
> them into a hashtable and check against that, instead? Let me know if
> you have another, preferred approach.

It was just a drive-by comment, keep it as simple as possible.

	Sam

Re: [PATCH] kconfig: warn on dead default

[Nathan Chancellor]

Hi Julian,

On Sat, Jun 06, 2026 at 03:00:08PM +0100, Julian Braha wrote:
> The dead default check was originally introduced with kconfirm:
> https://lore.kernel.org/all/6ec4df6d-1445-48ca-8f54-1d1a83c4716d@gmail.com/
> 
> While I'm still working on that tool, it's not yet ready for inclusion
> into the tree. I am currently waiting for common distro packagers to
> package the parsing library before submitting the next RFC iteration.
> 
> However, the dead default check is more impactful than the other checks:
> all 4 dead defaults that were detected should not have been dead and could
> cause misconfiguration bugs. But fortunately, these were just for kunit
> tests. The 3 patches to fix them have all since been merged:
> commit aef656a0e6c0 ("powerpc: fix dead default for GUEST_STATE_BUFFER_TEST")
> commit 30cc5e2ad826 ("s390/Kconfig: Cleanup defaults for selftests")
> commit df75430515c3 ("drm: fix dead default for DRM_TTM_KUNIT_TEST")
> 
> We can actually check for dead defaults while evaluating Kconfig, which
> should be even more effective at preventing future instances than keeping
> it in a static checker.
> 
> Note that this patch will only trigger a warning when the default values
> are different, in other words, pure duplicate defaults won't cause a
> warning, as they are simply redundant.
> 
> Signed-off-by: Julian Braha <julianbraha@gmail.com>

Nicolas pushed this into -next, where it fires several times for me. I
have not looked at these closely yet, I just wanted to get the report
out so that more eyes could be on it.

ARCH=powerpc allmodconfig:

  arch/powerpc/platforms/Kconfig.cputype:269:warning: default for 'TUNE_CPU' has the same condition as the earlier default at arch/powerpc/platforms/Kconfig.cputype:268
  arch/powerpc/platforms/Kconfig.cputype:270:warning: default for 'TUNE_CPU' has the same condition as the earlier default at arch/powerpc/platforms/Kconfig.cputype:268
  lib/Kconfig.debug:1083:warning: default for 'PANIC_TIMEOUT' is unreachable: earlier default at arch/powerpc/Kconfig:350 is unconditional
  arch/Kconfig:1194:warning: default for 'PGTABLE_LEVELS' is unreachable: earlier default at arch/powerpc/Kconfig:482 is unconditional

ARCH=s390 allmodconfig:

  arch/Kconfig:1194:warning: default for 'PGTABLE_LEVELS' is unreachable: earlier default at arch/s390/Kconfig:295 is unconditional
  lib/Kconfig:372:warning: default for 'HAS_IOMEM' is unreachable: earlier default at arch/s390/Kconfig:760 is unconditional

When building with clang-23, which supports '-fms-anonymous-struct':

  init/Kconfig:978:warning: default for 'CC_MS_EXTENSIONS' is unreachable: earlier default at init/Kconfig:977 is unconditional

I think the init/Kconfig and arch/powerpc/platforms/Kconfig.cputype ones
are intentional due to how cc-option works but I know those might be
hard to silence because they are already evaluated by this stage if I
understand correctly? Not sure about the other ones (apologies if they
are already known or have fixes).

-- 
Cheers,
Nathan

Re: [PATCH] kconfig: warn on dead default

[Nicolas Schier]

On Mon, Jul 06, 2026 at 10:31:43PM -0700, Nathan Chancellor wrote:
> Hi Julian,
> 
> On Sat, Jun 06, 2026 at 03:00:08PM +0100, Julian Braha wrote:
> > The dead default check was originally introduced with kconfirm:
> > https://lore.kernel.org/all/6ec4df6d-1445-48ca-8f54-1d1a83c4716d@gmail.com/
> > 
> > While I'm still working on that tool, it's not yet ready for inclusion
> > into the tree. I am currently waiting for common distro packagers to
> > package the parsing library before submitting the next RFC iteration.
> > 
> > However, the dead default check is more impactful than the other checks:
> > all 4 dead defaults that were detected should not have been dead and could
> > cause misconfiguration bugs. But fortunately, these were just for kunit
> > tests. The 3 patches to fix them have all since been merged:
> > commit aef656a0e6c0 ("powerpc: fix dead default for GUEST_STATE_BUFFER_TEST")
> > commit 30cc5e2ad826 ("s390/Kconfig: Cleanup defaults for selftests")
> > commit df75430515c3 ("drm: fix dead default for DRM_TTM_KUNIT_TEST")
> > 
> > We can actually check for dead defaults while evaluating Kconfig, which
> > should be even more effective at preventing future instances than keeping
> > it in a static checker.
> > 
> > Note that this patch will only trigger a warning when the default values
> > are different, in other words, pure duplicate defaults won't cause a
> > warning, as they are simply redundant.
> > 
> > Signed-off-by: Julian Braha <julianbraha@gmail.com>
> 
> Nicolas pushed this into -next, where it fires several times for me. I
> have not looked at these closely yet, I just wanted to get the report
> out so that more eyes could be on it.
> 
> ARCH=powerpc allmodconfig:
> 
>   arch/powerpc/platforms/Kconfig.cputype:269:warning: default for 'TUNE_CPU' has the same condition as the earlier default at arch/powerpc/platforms/Kconfig.cputype:268
>   arch/powerpc/platforms/Kconfig.cputype:270:warning: default for 'TUNE_CPU' has the same condition as the earlier default at arch/powerpc/platforms/Kconfig.cputype:268
>   lib/Kconfig.debug:1083:warning: default for 'PANIC_TIMEOUT' is unreachable: earlier default at arch/powerpc/Kconfig:350 is unconditional
>   arch/Kconfig:1194:warning: default for 'PGTABLE_LEVELS' is unreachable: earlier default at arch/powerpc/Kconfig:482 is unconditional
> 
> ARCH=s390 allmodconfig:
> 
>   arch/Kconfig:1194:warning: default for 'PGTABLE_LEVELS' is unreachable: earlier default at arch/s390/Kconfig:295 is unconditional
>   lib/Kconfig:372:warning: default for 'HAS_IOMEM' is unreachable: earlier default at arch/s390/Kconfig:760 is unconditional
> 
> When building with clang-23, which supports '-fms-anonymous-struct':
> 
>   init/Kconfig:978:warning: default for 'CC_MS_EXTENSIONS' is unreachable: earlier default at init/Kconfig:977 is unconditional
> 
> I think the init/Kconfig and arch/powerpc/platforms/Kconfig.cputype ones
> are intentional due to how cc-option works but I know those might be
> hard to silence because they are already evaluated by this stage if I
> understand correctly? Not sure about the other ones (apologies if they
> are already known or have fixes).

thanks for the report.  I'll remove the patch from kbuild-for-next
again as there is not obvious immediate solution visible to me right
now.

-- 
Nicolas

Re: [PATCH] kconfig: warn on dead default

[Julian Braha]

Hi Nathan,

On 7/7/26 06:31, Nathan Chancellor wrote:
> Nicolas pushed this into -next, where it fires several times for me. I
> have not looked at these closely yet, I just wanted to get the report
> out so that more eyes could be on it.

Thank you for sharing, I've taken a close look.

> ARCH=powerpc allmodconfig:
> 
>   arch/powerpc/platforms/Kconfig.cputype:269:warning: default for 'TUNE_CPU' has the same condition as the earlier default at arch/powerpc/platforms/Kconfig.cputype:268
>   arch/powerpc/platforms/Kconfig.cputype:270:warning: default for 'TUNE_CPU' has the same condition as the earlier default at arch/powerpc/platforms/Kconfig.cputype:268
>   lib/Kconfig.debug:1083:warning: default for 'PANIC_TIMEOUT' is unreachable: earlier default at arch/powerpc/Kconfig:350 is unconditional
>   arch/Kconfig:1194:warning: default for 'PGTABLE_LEVELS' is unreachable: earlier default at arch/powerpc/Kconfig:482 is unconditional
> 
> ARCH=s390 allmodconfig:
> 
>   arch/Kconfig:1194:warning: default for 'PGTABLE_LEVELS' is unreachable: earlier default at arch/s390/Kconfig:295 is unconditional
>   lib/Kconfig:372:warning: default for 'HAS_IOMEM' is unreachable: earlier default at arch/s390/Kconfig:760 is unconditional

It seems to me that all of these instances are probably intended by the
architectures to override the global default.

Personally, I find this redefine-and-override style confusing as it
requires the reader to think about the order of imports in kconfig. I'll
throw them in my kconfig cleanup pile, and see what the relevant
maintainers think of my version.

> When building with clang-23, which supports '-fms-anonymous-struct':
> 
>   init/Kconfig:978:warning: default for 'CC_MS_EXTENSIONS' is unreachable: earlier default at init/Kconfig:977 is unconditional
> 
> I think the init/Kconfig and arch/powerpc/platforms/Kconfig.cputype ones
> are intentional due to how cc-option works but I know those might be
> hard to silence because they are already evaluated by this stage if I
> understand correctly? Not sure about the other ones (apologies if they
> are already known or have fixes).

Yeah, let's leave these alone.

Best not to merge this. Thank you all for taking a look.

- Julian Braha