Build error in next-20260223 gcc-plugins/latent_entropy_plugin.so

Build error in next-20260223 gcc-plugins/latent_entropy_plugin.so

[Valdis Klētnieks]

Building next-20260223 dies an early death:

  HOSTLD  arch/x86/tools/relocs
  HOSTCC  arch/x86/tools/vdso2c
  HOSTCXX scripts/gcc-plugins/latent_entropy_plugin.so
In file included from scripts/gcc-plugins/latent_entropy_plugin.c:78:
scripts/gcc-plugins/gcc-common.h: In function 'void debug_tree(const_tree)':
scripts/gcc-plugins/gcc-common.h:417:20: error: 'CONST_CAST_TREE' was not declared in this scope; did you mean 'CONST_CAST_EXPR'?
  417 |         debug_tree(CONST_CAST_TREE(t));
      |                    ^~~~~~~~~~~~~~~
      |                    CONST_CAST_EXPR
scripts/gcc-plugins/gcc-common.h: In function 'void debug_gimple_stmt(const_gimple_ptr)':
scripts/gcc-plugins/gcc-common.h:312:47: error: expected primary-expression before ',' token
  312 | #define CONST_CAST_GIMPLE(X) CONST_CAST(gimple, (X))
      |                                               ^
scripts/gcc-plugins/gcc-common.h:422:27: note: in expansion of macro 'CONST_CAST_GIMPLE'
  422 |         debug_gimple_stmt(CONST_CAST_GIMPLE(s));
      |                           ^~~~~~~~~~~~~~~~~
scripts/gcc-plugins/gcc-common.h:312:30: error: 'CONST_CAST' was not declared in this scope; did you mean 'CONST_INT'?
  312 | #define CONST_CAST_GIMPLE(X) CONST_CAST(gimple, (X))
      |                              ^~~~~~~~~~
scripts/gcc-plugins/gcc-common.h:422:27: note: in expansion of macro 'CONST_CAST_GIMPLE'
  422 |         debug_gimple_stmt(CONST_CAST_GIMPLE(s));
      |                           ^~~~~~~~~~~~~~~~~
make[3]: *** [scripts/gcc-plugins/Makefile:54: scripts/gcc-plugins/latent_entropy_plugin.so] Error 1
make[2]: *** [scripts/Makefile.build:546: scripts/gcc-plugins] Error 2
make[1]: *** [/usr/src/linux-next/Makefile:1316: scripts] Error 2
make: *** [Makefile:248: __sub-make] Error 2

'git blame' isn't showing any recent changes..

next-20260109 built OK.   Any ideas before I start bisecting in more detail?

\0\0\0\0\0\0\0\0

Re: Build error in next-20260223 gcc-plugins/latent_entropy_plugin.so

[Kees Cook]

On Wed, Feb 25, 2026 at 12:20:01AM -0500, Valdis Klētnieks wrote:
> Building next-20260223 dies an early death:
> 
>   HOSTLD  arch/x86/tools/relocs
>   HOSTCC  arch/x86/tools/vdso2c
>   HOSTCXX scripts/gcc-plugins/latent_entropy_plugin.so
> In file included from scripts/gcc-plugins/latent_entropy_plugin.c:78:
> scripts/gcc-plugins/gcc-common.h: In function 'void debug_tree(const_tree)':
> scripts/gcc-plugins/gcc-common.h:417:20: error: 'CONST_CAST_TREE' was not declared in this scope; did you mean 'CONST_CAST_EXPR'?
>   417 |         debug_tree(CONST_CAST_TREE(t));
>       |                    ^~~~~~~~~~~~~~~
>       |                    CONST_CAST_EXPR
> scripts/gcc-plugins/gcc-common.h: In function 'void debug_gimple_stmt(const_gimple_ptr)':
> scripts/gcc-plugins/gcc-common.h:312:47: error: expected primary-expression before ',' token
>   312 | #define CONST_CAST_GIMPLE(X) CONST_CAST(gimple, (X))
>       |                                               ^
> scripts/gcc-plugins/gcc-common.h:422:27: note: in expansion of macro 'CONST_CAST_GIMPLE'
>   422 |         debug_gimple_stmt(CONST_CAST_GIMPLE(s));
>       |                           ^~~~~~~~~~~~~~~~~
> scripts/gcc-plugins/gcc-common.h:312:30: error: 'CONST_CAST' was not declared in this scope; did you mean 'CONST_INT'?
>   312 | #define CONST_CAST_GIMPLE(X) CONST_CAST(gimple, (X))
>       |                              ^~~~~~~~~~
> scripts/gcc-plugins/gcc-common.h:422:27: note: in expansion of macro 'CONST_CAST_GIMPLE'
>   422 |         debug_gimple_stmt(CONST_CAST_GIMPLE(s));
>       |                           ^~~~~~~~~~~~~~~~~
> make[3]: *** [scripts/gcc-plugins/Makefile:54: scripts/gcc-plugins/latent_entropy_plugin.so] Error 1
> make[2]: *** [scripts/Makefile.build:546: scripts/gcc-plugins] Error 2
> make[1]: *** [/usr/src/linux-next/Makefile:1316: scripts] Error 2
> make: *** [Makefile:248: __sub-make] Error 2
> 
> 'git blame' isn't showing any recent changes..
> 
> next-20260109 built OK.   Any ideas before I start bisecting in more detail?
> 
> ????????

Which GCC version?

-- 
Kees Cook

Re: Build error in next-20260223 gcc-plugins/latent_entropy_plugin.so

[Valdis Klētnieks]

On Tue, 24 Feb 2026 23:33:32 -0800, Kees Cook said:
> 
> Which GCC version?

Oh nuts.  I thought I included that.. Dug into it more, looks like a GCC
bug rather than a kernel bug.

From the gcc RPM changelog:

This one works.

* Tue Jan 27 2026 Jakub Jelinek <jakub@redhat.com> 16.0.1-0.5
- update from trunk

This doesn't:

* Mon Feb 09 2026 Jakub Jelinek <jakub@redhat.com> 16.0.1-0.7
- update from trunk


\0\0\0\0\0\0\0\0

Re: Build error in next-20260223 gcc-plugins/latent_entropy_plugin.so

[Ingo Saitz]

I get the same error with

    gcc (Debian 16-20260308-1) 16.0.1 20260308 (experimental) [trunk r16-7949-g6c5de6335f5]

It seems gcc removed the CONST_CAST macros in gcc-16. CONST_CAST_TREE
got removed in commit c3d96ff, CONST_CAST in commit 458c792.

    https://github.com/gcc-mirror/gcc/commit/c3d96ff9e916c02584aa081f03ab999292efbb50
    https://github.com/gcc-mirror/gcc/commit/458c7926d48959abcb2c1adaa22458e27459a551

So this does not seem to be a gcc bug, but the CONST_CAST macros being
gone for good.

I attached a quick patch to get the kernel build with the current gcc-16
development version, but it still needs to be guarded against different
gcc versions.

    Ingo
-- 
const_cast<long double>(Λ)

Re: Build error in next-20260223 gcc-plugins/latent_entropy_plugin.so

[Ingo Saitz]

[-- Attachment #1: Type: text/plain, Size: 140 bytes --]

On Sat, Mar 14, 2026 at 02:50:46PM +0100, Ingo Saitz wrote:
> I attached [...]

Grml, here you go

    Ingo
-- 
const_cast<long double>(Λ)

[-- Attachment #2: 0001-Replace-CONST_CAST-with-const_cast.patch --]
[-- Type: text/x-diff, Size: 1671 bytes --]

From 552485b879d19780afd3614d69508c3ea7cb242f Mon Sep 17 00:00:00 2001
From: Ingo Saitz <ingo@hannover.ccc.de>
Date: Sat, 14 Mar 2026 14:24:56 +0100
Subject: [PATCH] Replace CONST_CAST with const_cast<>.

For gcc-16, this was removed in gcc trunk

see commits
  c3d96ff9e916c02584aa081f03ab999292efbb50
  458c7926d48959abcb2c1adaa22458e27459a551
---
 scripts/gcc-plugins/gcc-common.h | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/scripts/gcc-plugins/gcc-common.h b/scripts/gcc-plugins/gcc-common.h
index 8f1b3500f8e2..7b3517b31e0a 100644
--- a/scripts/gcc-plugins/gcc-common.h
+++ b/scripts/gcc-plugins/gcc-common.h
@@ -309,7 +309,7 @@ typedef const gimple *const_gimple_ptr;
 #define gimple gimple_ptr
 #define const_gimple const_gimple_ptr
 #undef CONST_CAST_GIMPLE
-#define CONST_CAST_GIMPLE(X) CONST_CAST(gimple, (X))
+#define CONST_CAST_GIMPLE(X) const_cast<gimple>((X))
 
 /* gimple related */
 static inline gimple gimple_build_assign_with_ops(enum tree_code subcode, tree lhs, tree op1, tree op2 MEM_STAT_DECL)
@@ -414,7 +414,7 @@ static inline void ipa_remove_stmt_references(symtab_node *referring_node, gimpl
 #ifdef __cplusplus
 static inline void debug_tree(const_tree t)
 {
-	debug_tree(CONST_CAST_TREE(t));
+	debug_tree(const_cast<tree>(t));
 }
 
 static inline void debug_gimple_stmt(const_gimple s)
@@ -422,7 +422,7 @@ static inline void debug_gimple_stmt(const_gimple s)
 	debug_gimple_stmt(CONST_CAST_GIMPLE(s));
 }
 #else
-#define debug_tree(t) debug_tree(CONST_CAST_TREE(t))
+#define debug_tree(t) debug_tree(const_cast<tree>(t))
 #define debug_gimple_stmt(s) debug_gimple_stmt(CONST_CAST_GIMPLE(s))
 #endif
 
-- 
2.53.0

Re: Build error in next-20260223 gcc-plugins/latent_entropy_plugin.so

[Kees Cook]

On Sat, Mar 14, 2026 at 03:44:08PM +0100, Ingo Saitz wrote:
> On Sat, Mar 14, 2026 at 02:50:46PM +0100, Ingo Saitz wrote:
> > I attached [...]
> 
> Grml, here you go
> 
>     Ingo
> -- 
> const_cast<long double>(Λ)

> From 552485b879d19780afd3614d69508c3ea7cb242f Mon Sep 17 00:00:00 2001
> From: Ingo Saitz <ingo@hannover.ccc.de>
> Date: Sat, 14 Mar 2026 14:24:56 +0100
> Subject: [PATCH] Replace CONST_CAST with const_cast<>.
> 
> For gcc-16, this was removed in gcc trunk
> 
> see commits
>   c3d96ff9e916c02584aa081f03ab999292efbb50
>   458c7926d48959abcb2c1adaa22458e27459a551
> ---
>  scripts/gcc-plugins/gcc-common.h | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/scripts/gcc-plugins/gcc-common.h b/scripts/gcc-plugins/gcc-common.h
> index 8f1b3500f8e2..7b3517b31e0a 100644
> --- a/scripts/gcc-plugins/gcc-common.h
> +++ b/scripts/gcc-plugins/gcc-common.h
> @@ -309,7 +309,7 @@ typedef const gimple *const_gimple_ptr;
>  #define gimple gimple_ptr
>  #define const_gimple const_gimple_ptr
>  #undef CONST_CAST_GIMPLE
> -#define CONST_CAST_GIMPLE(X) CONST_CAST(gimple, (X))
> +#define CONST_CAST_GIMPLE(X) const_cast<gimple>((X))
>  
>  /* gimple related */
>  static inline gimple gimple_build_assign_with_ops(enum tree_code subcode, tree lhs, tree op1, tree op2 MEM_STAT_DECL)
> @@ -414,7 +414,7 @@ static inline void ipa_remove_stmt_references(symtab_node *referring_node, gimpl
>  #ifdef __cplusplus
>  static inline void debug_tree(const_tree t)
>  {
> -	debug_tree(CONST_CAST_TREE(t));
> +	debug_tree(const_cast<tree>(t));
>  }
>  
>  static inline void debug_gimple_stmt(const_gimple s)
> @@ -422,7 +422,7 @@ static inline void debug_gimple_stmt(const_gimple s)
>  	debug_gimple_stmt(CONST_CAST_GIMPLE(s));
>  }
>  #else
> -#define debug_tree(t) debug_tree(CONST_CAST_TREE(t))
> +#define debug_tree(t) debug_tree(const_cast<tree>(t))
>  #define debug_gimple_stmt(s) debug_gimple_stmt(CONST_CAST_GIMPLE(s))
>  #endif

Thanks for this!

Should CONST_CAST_TREE be explicitly defined instead, to match
CONST_CAST_GIMPLE in the this header file? It seems weird to have one of
two done this way: we should either do both or neither (CONST_CAST_GIMPLE
is only used in the header file too).

Is const_cast<gimple> and const_cast<tree> valid back through GCC 8
(earliest Linux-supported GCC)?

-Kees

-- 
Kees Cook

Re: Build error in next-20260223 gcc-plugins/latent_entropy_plugin.so

[Ingo Saitz]

On Fri, Mar 20, 2026 at 12:23:10PM -0700, Kees Cook wrote:
> Thanks for this!
> 
> Should CONST_CAST_TREE be explicitly defined instead, to match
> CONST_CAST_GIMPLE in the this header file? It seems weird to have one of
> two done this way: we should either do both or neither (CONST_CAST_GIMPLE
> is only used in the header file too).

My first idea was to simplay guard this depending on the gcc version,
since the CONST_CAST_TREE type changed from (union tree_node *) to tree.
But the typedef union tree_node *tree (in gcc source: gcc/coretypes.h)
is already there in gcc-8, and const_cast is an old c++ feature.

> Is const_cast<gimple> and const_cast<tree> valid back through GCC 8
> (earliest Linux-supported GCC)?

I successfully compiled and booted linux 6.19.8 with my patch applied
with gcc-8 (Debian 8.3.0-7) 8.3.0, so it think this patch does work with
all supported gcc versions and a guard is not needed indeed.

I tried to minimize the changes, since I don't really know how gcc
plugins actually work. But changing CONST_CAST_GIMPLE to
const_cast<gimple>, too, should not fail since that's now a simple text
replacement.

    Ingo
-- 
const_cast<long double>(Λ)