[PATCH 1/1] mm,swapops: Fix compile error on set_pmd_migration_entry.

[PATCH 1/1] mm,swapops: Fix compile error on set_pmd_migration_entry.

[Wale Zhang]

../include/linux/swapops.h: In function ‘set_pmd_migration_entry’:
../include/linux/swapops.h:346:1: error: no return statement in function returning non-void [-Werror=return-type]
  346 | }
      | ^
cc1: some warnings being treated as errors

Signed-off-by: Wale Zhang <wale.zhang.ftd@gmail.com>
---
 include/linux/swapops.h | 1 +
 1 file changed, 1 insertion(+)

diff --git a/include/linux/swapops.h b/include/linux/swapops.h
index 8cfc966eae48..84eb36d97212 100644
--- a/include/linux/swapops.h
+++ b/include/linux/swapops.h
@@ -343,6 +343,7 @@ static inline int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
 		struct page *page)
 {
 	BUILD_BUG();
+	return 0;
 }
 
 static inline void remove_migration_pmd(struct page_vma_mapped_walk *pvmw,
-- 
2.43.0

Re: [PATCH 1/1] mm,swapops: Fix compile error on set_pmd_migration_entry.

[Barry Song]

On Thu, Dec 25, 2025 at 10:41 PM Wale Zhang <wale.zhang.ftd@gmail.com> wrote:
>
> ../include/linux/swapops.h: In function ‘set_pmd_migration_entry’:
> ../include/linux/swapops.h:346:1: error: no return statement in function returning non-void [-Werror=return-type]
>   346 | }
>       | ^
> cc1: some warnings being treated as errors
>
> Signed-off-by: Wale Zhang <wale.zhang.ftd@gmail.com>

Sorry. NAK.

This looks odd. We already have a BUILD_BUG(), so no fix is needed.

we should already have:
././include/linux/compiler_types.h:631:45: error: call to
‘__compiletime_assert_907’ declared with attribute error: BUILD_BUG
failed
  631 |         _compiletime_assert(condition, msg,
__compiletime_assert_, __COUNTER__)
      |                                             ^
././include/linux/compiler_types.h:612:25: note: in definition of
macro ‘__compiletime_assert’
  612 |                         prefix ## suffix();
         \
      |                         ^~~~~~

Do you actually see this being built? If so, please help identify why it
gets built.

> ---
>  include/linux/swapops.h | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/include/linux/swapops.h b/include/linux/swapops.h
> index 8cfc966eae48..84eb36d97212 100644
> --- a/include/linux/swapops.h
> +++ b/include/linux/swapops.h
> @@ -343,6 +343,7 @@ static inline int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
>                 struct page *page)
>  {
>         BUILD_BUG();
> +       return 0;
>  }
>
>  static inline void remove_migration_pmd(struct page_vma_mapped_walk *pvmw,
> --
> 2.43.0
>

Thanks
Barry

Re: [PATCH 1/1] mm,swapops: Fix compile error on set_pmd_migration_entry.

[wale zhang]

On Thu, Dec 25, 2025 at 6:22 PM Barry Song <21cnbao@gmail.com> wrote:
>
> On Thu, Dec 25, 2025 at 10:41 PM Wale Zhang <wale.zhang.ftd@gmail.com> wrote:
> >
> > ../include/linux/swapops.h: In function ‘set_pmd_migration_entry’:
> > ../include/linux/swapops.h:346:1: error: no return statement in function returning non-void [-Werror=return-type]
> >   346 | }
> >       | ^
> > cc1: some warnings being treated as errors
> >
> > Signed-off-by: Wale Zhang <wale.zhang.ftd@gmail.com>
>
> Sorry. NAK.
>
> This looks odd. We already have a BUILD_BUG(), so no fix is needed.
>
> we should already have:
> ././include/linux/compiler_types.h:631:45: error: call to
> ‘__compiletime_assert_907’ declared with attribute error: BUILD_BUG
> failed
>   631 |         _compiletime_assert(condition, msg,
> __compiletime_assert_, __COUNTER__)
>       |                                             ^
> ././include/linux/compiler_types.h:612:25: note: in definition of
> macro ‘__compiletime_assert’
>   612 |                         prefix ## suffix();
>          \
>       |                         ^~~~~~
>
> Do you actually see this being built? If so, please help identify why it
> gets built.

Hello Barry,

I compile the 6.19-rc2 for arm64 with -O0, this kind of error log will appear.
In file included from /home/wale/repo/linux/master/include/linux/leafops.h:11,
                 from
/home/wale/repo/linux/master/include/linux/userfaultfd_k.h:19,
                 from /home/wale/repo/linux/master/include/linux/mm_inline.h:10,
                 from /home/wale/repo/linux/master/mm/internal.h:13,
                 from /home/wale/repo/linux/master/mm/fadvise.c:24:
/home/wale/repo/linux/master/include/linux/swapops.h: In function
‘set_pmd_migration_entry’:
/home/wale/repo/linux/master/include/linux/swapops.h:346:1: error: no
return statement in function returning non-void [-Werror=return-type]
  346 | }
      | ^
cc1: some warnings being treated as errors
make[4]: *** [/home/wale/repo/linux/master/scripts/Makefile.build:287:
mm/fadvise.o] Error 1
make[4]: *** Waiting for unfinished jobs....

There is an example on my
github(https://github.com/wale-ftd/linux/commits/linux-6.12/) that
used -O0 to compile the 6.12 kernel. If you're interested, you can
check them out.

>
> > ---
> >  include/linux/swapops.h | 1 +
> >  1 file changed, 1 insertion(+)
> >
> > diff --git a/include/linux/swapops.h b/include/linux/swapops.h
> > index 8cfc966eae48..84eb36d97212 100644
> > --- a/include/linux/swapops.h
> > +++ b/include/linux/swapops.h
> > @@ -343,6 +343,7 @@ static inline int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
> >                 struct page *page)
> >  {
> >         BUILD_BUG();
> > +       return 0;
> >  }
> >
> >  static inline void remove_migration_pmd(struct page_vma_mapped_walk *pvmw,
> > --
> > 2.43.0
> >
>
> Thanks
> Barry

--
Thanks
Wale

Re: [PATCH 1/1] mm,swapops: Fix compile error on set_pmd_migration_entry.

[Barry Song]

On Sat, Dec 27, 2025 at 12:46 AM wale zhang <wale.zhang.ftd@gmail.com> wrote:
>
> On Thu, Dec 25, 2025 at 6:22 PM Barry Song <21cnbao@gmail.com> wrote:
> >
> > On Thu, Dec 25, 2025 at 10:41 PM Wale Zhang <wale.zhang.ftd@gmail.com> wrote:
> > >
> > > ../include/linux/swapops.h: In function ‘set_pmd_migration_entry’:
> > > ../include/linux/swapops.h:346:1: error: no return statement in function returning non-void [-Werror=return-type]
> > >   346 | }
> > >       | ^
> > > cc1: some warnings being treated as errors
> > >
> > > Signed-off-by: Wale Zhang <wale.zhang.ftd@gmail.com>
> >
> > Sorry. NAK.
> >
> > This looks odd. We already have a BUILD_BUG(), so no fix is needed.
> >
> > we should already have:
> > ././include/linux/compiler_types.h:631:45: error: call to
> > ‘__compiletime_assert_907’ declared with attribute error: BUILD_BUG
> > failed
> >   631 |         _compiletime_assert(condition, msg,
> > __compiletime_assert_, __COUNTER__)
> >       |                                             ^
> > ././include/linux/compiler_types.h:612:25: note: in definition of
> > macro ‘__compiletime_assert’
> >   612 |                         prefix ## suffix();
> >          \
> >       |                         ^~~~~~
> >
> > Do you actually see this being built? If so, please help identify why it
> > gets built.
>
> Hello Barry,
>
> I compile the 6.19-rc2 for arm64 with -O0, this kind of error log will appear.
> In file included from /home/wale/repo/linux/master/include/linux/leafops.h:11,
>                  from
> /home/wale/repo/linux/master/include/linux/userfaultfd_k.h:19,
>                  from /home/wale/repo/linux/master/include/linux/mm_inline.h:10,
>                  from /home/wale/repo/linux/master/mm/internal.h:13,
>                  from /home/wale/repo/linux/master/mm/fadvise.c:24:
> /home/wale/repo/linux/master/include/linux/swapops.h: In function
> ‘set_pmd_migration_entry’:
> /home/wale/repo/linux/master/include/linux/swapops.h:346:1: error: no
> return statement in function returning non-void [-Werror=return-type]
>   346 | }
>       | ^
> cc1: some warnings being treated as errors
> make[4]: *** [/home/wale/repo/linux/master/scripts/Makefile.build:287:
> mm/fadvise.o] Error 1
> make[4]: *** Waiting for unfinished jobs....
>
> There is an example on my
> github(https://github.com/wale-ftd/linux/commits/linux-6.12/) that
> used -O0 to compile the 6.12 kernel. If you're interested, you can
> check them out.

This is not my question. The BUILD_BUG here means this code
should never be built. If it is built, then that indicates a bug.
So my question is: under what kernel configuration would this
code actually get built?

Thanks
Barry

Re: [PATCH 1/1] mm,swapops: Fix compile error on set_pmd_migration_entry.

[wale zhang]

On Sat, Dec 27, 2025 at 2:43 AM Barry Song <21cnbao@gmail.com> wrote:
>
> On Sat, Dec 27, 2025 at 12:46 AM wale zhang <wale.zhang.ftd@gmail.com> wrote:
> >
> > On Thu, Dec 25, 2025 at 6:22 PM Barry Song <21cnbao@gmail.com> wrote:
> > >
> > > On Thu, Dec 25, 2025 at 10:41 PM Wale Zhang <wale.zhang.ftd@gmail.com> wrote:
> > > >
> > > > ../include/linux/swapops.h: In function ‘set_pmd_migration_entry’:
> > > > ../include/linux/swapops.h:346:1: error: no return statement in function returning non-void [-Werror=return-type]
> > > >   346 | }
> > > >       | ^
> > > > cc1: some warnings being treated as errors
> > > >
> > > > Signed-off-by: Wale Zhang <wale.zhang.ftd@gmail.com>
> > >
> > > Sorry. NAK.
> > >
> > > This looks odd. We already have a BUILD_BUG(), so no fix is needed.
> > >
> > > we should already have:
> > > ././include/linux/compiler_types.h:631:45: error: call to
> > > ‘__compiletime_assert_907’ declared with attribute error: BUILD_BUG
> > > failed
> > >   631 |         _compiletime_assert(condition, msg,
> > > __compiletime_assert_, __COUNTER__)
> > >       |                                             ^
> > > ././include/linux/compiler_types.h:612:25: note: in definition of
> > > macro ‘__compiletime_assert’
> > >   612 |                         prefix ## suffix();
> > >          \
> > >       |                         ^~~~~~
> > >
> > > Do you actually see this being built? If so, please help identify why it
> > > gets built.
> >
> > Hello Barry,
> >
> > I compile the 6.19-rc2 for arm64 with -O0, this kind of error log will appear.
> > In file included from /home/wale/repo/linux/master/include/linux/leafops.h:11,
> >                  from
> > /home/wale/repo/linux/master/include/linux/userfaultfd_k.h:19,
> >                  from /home/wale/repo/linux/master/include/linux/mm_inline.h:10,
> >                  from /home/wale/repo/linux/master/mm/internal.h:13,
> >                  from /home/wale/repo/linux/master/mm/fadvise.c:24:
> > /home/wale/repo/linux/master/include/linux/swapops.h: In function
> > ‘set_pmd_migration_entry’:
> > /home/wale/repo/linux/master/include/linux/swapops.h:346:1: error: no
> > return statement in function returning non-void [-Werror=return-type]
> >   346 | }
> >       | ^
> > cc1: some warnings being treated as errors
> > make[4]: *** [/home/wale/repo/linux/master/scripts/Makefile.build:287:
> > mm/fadvise.o] Error 1
> > make[4]: *** Waiting for unfinished jobs....
> >
> > There is an example on my
> > github(https://github.com/wale-ftd/linux/commits/linux-6.12/) that
> > used -O0 to compile the 6.12 kernel. If you're interested, you can
> > check them out.
>
> This is not my question. The BUILD_BUG here means this code
> should never be built. If it is built, then that indicates a bug.
> So my question is: under what kernel configuration would this
> code actually get built?

Hello Barry,

I'm very sorry for misunderstanding your meaning.

For -O0, __OPTIMIZE__ isn't defined, so line 615 is hit.

595 #ifdef __OPTIMIZE__
602 # define __compiletime_assert(condition, msg, prefix, suffix)           \
612                         prefix ## suffix();
614 #else
615 # define __compiletime_assert(condition, msg, prefix, suffix)
((void)(condition))
616 #endif

Finally, set_pmd_migration_entry is defined as follow:
static inline int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
                struct page *page)
{
-        BUILD_BUG();
+        ((void)(0));
}

> Thanks
> Barry

--
Thanks
Wale

Re: [PATCH 1/1] mm,swapops: Fix compile error on set_pmd_migration_entry.

[Barry Song]

On Sat, Dec 27, 2025 at 8:51 PM wale zhang <wale.zhang.ftd@gmail.com> wrote:
>
> On Sat, Dec 27, 2025 at 2:43 AM Barry Song <21cnbao@gmail.com> wrote:
> >
> > On Sat, Dec 27, 2025 at 12:46 AM wale zhang <wale.zhang.ftd@gmail.com> wrote:
> > >
> > > On Thu, Dec 25, 2025 at 6:22 PM Barry Song <21cnbao@gmail.com> wrote:
> > > >
> > > > On Thu, Dec 25, 2025 at 10:41 PM Wale Zhang <wale.zhang.ftd@gmail.com> wrote:
> > > > >
> > > > > ../include/linux/swapops.h: In function ‘set_pmd_migration_entry’:
> > > > > ../include/linux/swapops.h:346:1: error: no return statement in function returning non-void [-Werror=return-type]
> > > > >   346 | }
> > > > >       | ^
> > > > > cc1: some warnings being treated as errors
> > > > >
> > > > > Signed-off-by: Wale Zhang <wale.zhang.ftd@gmail.com>
> > > >
> > > > Sorry. NAK.
> > > >
> > > > This looks odd. We already have a BUILD_BUG(), so no fix is needed.
> > > >
> > > > we should already have:
> > > > ././include/linux/compiler_types.h:631:45: error: call to
> > > > ‘__compiletime_assert_907’ declared with attribute error: BUILD_BUG
> > > > failed
> > > >   631 |         _compiletime_assert(condition, msg,
> > > > __compiletime_assert_, __COUNTER__)
> > > >       |                                             ^
> > > > ././include/linux/compiler_types.h:612:25: note: in definition of
> > > > macro ‘__compiletime_assert’
> > > >   612 |                         prefix ## suffix();
> > > >          \
> > > >       |                         ^~~~~~
> > > >
> > > > Do you actually see this being built? If so, please help identify why it
> > > > gets built.
> > >
> > > Hello Barry,
> > >
> > > I compile the 6.19-rc2 for arm64 with -O0, this kind of error log will appear.
> > > In file included from /home/wale/repo/linux/master/include/linux/leafops.h:11,
> > >                  from
> > > /home/wale/repo/linux/master/include/linux/userfaultfd_k.h:19,
> > >                  from /home/wale/repo/linux/master/include/linux/mm_inline.h:10,
> > >                  from /home/wale/repo/linux/master/mm/internal.h:13,
> > >                  from /home/wale/repo/linux/master/mm/fadvise.c:24:
> > > /home/wale/repo/linux/master/include/linux/swapops.h: In function
> > > ‘set_pmd_migration_entry’:
> > > /home/wale/repo/linux/master/include/linux/swapops.h:346:1: error: no
> > > return statement in function returning non-void [-Werror=return-type]
> > >   346 | }
> > >       | ^
> > > cc1: some warnings being treated as errors
> > > make[4]: *** [/home/wale/repo/linux/master/scripts/Makefile.build:287:
> > > mm/fadvise.o] Error 1
> > > make[4]: *** Waiting for unfinished jobs....
> > >
> > > There is an example on my
> > > github(https://github.com/wale-ftd/linux/commits/linux-6.12/) that
> > > used -O0 to compile the 6.12 kernel. If you're interested, you can
> > > check them out.
> >
> > This is not my question. The BUILD_BUG here means this code
> > should never be built. If it is built, then that indicates a bug.
> > So my question is: under what kernel configuration would this
> > code actually get built?
>
> Hello Barry,
>
> I'm very sorry for misunderstanding your meaning.

No. You still don’t understand my point. The BUILD_BUG means this
code will never have a chance to be compiled under any possible
kernel configuration. That makes it dead code. If there is any
kernel .config in which it can be built, then we should fix it.

This set_pmd_migration_entry() has no chance of ever being built.

let me give you an example:
#if 0
aaa;
#endif

main()
{

}

Imagine this set_pmd_migration_entry() as aaa.

>
> For -O0, __OPTIMIZE__ isn't defined, so line 615 is hit.
>
> 595 #ifdef __OPTIMIZE__
> 602 # define __compiletime_assert(condition, msg, prefix, suffix)           \
> 612                         prefix ## suffix();
> 614 #else
> 615 # define __compiletime_assert(condition, msg, prefix, suffix)
> ((void)(condition))
> 616 #endif
>
> Finally, set_pmd_migration_entry is defined as follow:
> static inline int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
>                 struct page *page)
> {
> -        BUILD_BUG();
> +        ((void)(0));
> }
>

Thanks
Barry

Re: [PATCH 1/1] mm,swapops: Fix compile error on set_pmd_migration_entry.

[wale zhang]

On Sat, Dec 27, 2025 at 4:16 PM Barry Song <21cnbao@gmail.com> wrote:
>
> On Sat, Dec 27, 2025 at 8:51 PM wale zhang <wale.zhang.ftd@gmail.com> wrote:
> >
> > On Sat, Dec 27, 2025 at 2:43 AM Barry Song <21cnbao@gmail.com> wrote:
> > >
> > > On Sat, Dec 27, 2025 at 12:46 AM wale zhang <wale.zhang.ftd@gmail.com> wrote:
> > > >
> > > > On Thu, Dec 25, 2025 at 6:22 PM Barry Song <21cnbao@gmail.com> wrote:
> > > > >
> > > > > On Thu, Dec 25, 2025 at 10:41 PM Wale Zhang <wale.zhang.ftd@gmail.com> wrote:
> > > > > >
> > > > > > ../include/linux/swapops.h: In function ‘set_pmd_migration_entry’:
> > > > > > ../include/linux/swapops.h:346:1: error: no return statement in function returning non-void [-Werror=return-type]
> > > > > >   346 | }
> > > > > >       | ^
> > > > > > cc1: some warnings being treated as errors
> > > > > >
> > > > > > Signed-off-by: Wale Zhang <wale.zhang.ftd@gmail.com>
> > > > >
> > > > > Sorry. NAK.
> > > > >
> > > > > This looks odd. We already have a BUILD_BUG(), so no fix is needed.
> > > > >
> > > > > we should already have:
> > > > > ././include/linux/compiler_types.h:631:45: error: call to
> > > > > ‘__compiletime_assert_907’ declared with attribute error: BUILD_BUG
> > > > > failed
> > > > >   631 |         _compiletime_assert(condition, msg,
> > > > > __compiletime_assert_, __COUNTER__)
> > > > >       |                                             ^
> > > > > ././include/linux/compiler_types.h:612:25: note: in definition of
> > > > > macro ‘__compiletime_assert’
> > > > >   612 |                         prefix ## suffix();
> > > > >          \
> > > > >       |                         ^~~~~~
> > > > >
> > > > > Do you actually see this being built? If so, please help identify why it
> > > > > gets built.
> > > >
> > > > Hello Barry,
> > > >
> > > > I compile the 6.19-rc2 for arm64 with -O0, this kind of error log will appear.
> > > > In file included from /home/wale/repo/linux/master/include/linux/leafops.h:11,
> > > >                  from
> > > > /home/wale/repo/linux/master/include/linux/userfaultfd_k.h:19,
> > > >                  from /home/wale/repo/linux/master/include/linux/mm_inline.h:10,
> > > >                  from /home/wale/repo/linux/master/mm/internal.h:13,
> > > >                  from /home/wale/repo/linux/master/mm/fadvise.c:24:
> > > > /home/wale/repo/linux/master/include/linux/swapops.h: In function
> > > > ‘set_pmd_migration_entry’:
> > > > /home/wale/repo/linux/master/include/linux/swapops.h:346:1: error: no
> > > > return statement in function returning non-void [-Werror=return-type]
> > > >   346 | }
> > > >       | ^
> > > > cc1: some warnings being treated as errors
> > > > make[4]: *** [/home/wale/repo/linux/master/scripts/Makefile.build:287:
> > > > mm/fadvise.o] Error 1
> > > > make[4]: *** Waiting for unfinished jobs....
> > > >
> > > > There is an example on my
> > > > github(https://github.com/wale-ftd/linux/commits/linux-6.12/) that
> > > > used -O0 to compile the 6.12 kernel. If you're interested, you can
> > > > check them out.
> > >
> > > This is not my question. The BUILD_BUG here means this code
> > > should never be built. If it is built, then that indicates a bug.
> > > So my question is: under what kernel configuration would this
> > > code actually get built?
> >
> > Hello Barry,
> >
> > I'm very sorry for misunderstanding your meaning.
>
> No. You still don’t understand my point. The BUILD_BUG means this
> code will never have a chance to be compiled under any possible
> kernel configuration. That makes it dead code. If there is any
> kernel .config in which it can be built, then we should fix it.
>
> This set_pmd_migration_entry() has no chance of ever being built.

Hello Barry,

set_pmd_migration_entry() is defined in include/linux/swapops.h.
For not defining CONFIG_ARCH_ENABLE_THP_MIGRATION, line 342~346 will
be built as long as *.c file includes swapops.h.
There are many .c files including the header file directly or
indirectly, such as fadvise.c, shm.c ...

include/linux/swapops.h
324 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
325 extern int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
326                 struct page *page);
341 #else  /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
342 static inline int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
343                 struct page *page)
344 {
345         BUILD_BUG();
346 }

>
> let me give you an example:
> #if 0
> aaa;
> #endif
>
> main()
> {
>
> }
>
> Imagine this set_pmd_migration_entry() as aaa.
>
> >
> > For -O0, __OPTIMIZE__ isn't defined, so line 615 is hit.
> >
> > 595 #ifdef __OPTIMIZE__
> > 602 # define __compiletime_assert(condition, msg, prefix, suffix)           \
> > 612                         prefix ## suffix();
> > 614 #else
> > 615 # define __compiletime_assert(condition, msg, prefix, suffix)
> > ((void)(condition))
> > 616 #endif
> >
> > Finally, set_pmd_migration_entry is defined as follow:
> > static inline int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
> >                 struct page *page)
> > {
> > -        BUILD_BUG();
> > +        ((void)(0));
> > }
> >
>
> Thanks
> Barry

--
Thanks
Wale

On Sat, Dec 27, 2025 at 4:16 PM Barry Song <21cnbao@gmail.com> wrote:
>
> On Sat, Dec 27, 2025 at 8:51 PM wale zhang <wale.zhang.ftd@gmail.com> wrote:
> >
> > On Sat, Dec 27, 2025 at 2:43 AM Barry Song <21cnbao@gmail.com> wrote:
> > >
> > > On Sat, Dec 27, 2025 at 12:46 AM wale zhang <wale.zhang.ftd@gmail.com> wrote:
> > > >
> > > > On Thu, Dec 25, 2025 at 6:22 PM Barry Song <21cnbao@gmail.com> wrote:
> > > > >
> > > > > On Thu, Dec 25, 2025 at 10:41 PM Wale Zhang <wale.zhang.ftd@gmail.com> wrote:
> > > > > >
> > > > > > ../include/linux/swapops.h: In function ‘set_pmd_migration_entry’:
> > > > > > ../include/linux/swapops.h:346:1: error: no return statement in function returning non-void [-Werror=return-type]
> > > > > >   346 | }
> > > > > >       | ^
> > > > > > cc1: some warnings being treated as errors
> > > > > >
> > > > > > Signed-off-by: Wale Zhang <wale.zhang.ftd@gmail.com>
> > > > >
> > > > > Sorry. NAK.
> > > > >
> > > > > This looks odd. We already have a BUILD_BUG(), so no fix is needed.
> > > > >
> > > > > we should already have:
> > > > > ././include/linux/compiler_types.h:631:45: error: call to
> > > > > ‘__compiletime_assert_907’ declared with attribute error: BUILD_BUG
> > > > > failed
> > > > >   631 |         _compiletime_assert(condition, msg,
> > > > > __compiletime_assert_, __COUNTER__)
> > > > >       |                                             ^
> > > > > ././include/linux/compiler_types.h:612:25: note: in definition of
> > > > > macro ‘__compiletime_assert’
> > > > >   612 |                         prefix ## suffix();
> > > > >          \
> > > > >       |                         ^~~~~~
> > > > >
> > > > > Do you actually see this being built? If so, please help identify why it
> > > > > gets built.
> > > >
> > > > Hello Barry,
> > > >
> > > > I compile the 6.19-rc2 for arm64 with -O0, this kind of error log will appear.
> > > > In file included from /home/wale/repo/linux/master/include/linux/leafops.h:11,
> > > >                  from
> > > > /home/wale/repo/linux/master/include/linux/userfaultfd_k.h:19,
> > > >                  from /home/wale/repo/linux/master/include/linux/mm_inline.h:10,
> > > >                  from /home/wale/repo/linux/master/mm/internal.h:13,
> > > >                  from /home/wale/repo/linux/master/mm/fadvise.c:24:
> > > > /home/wale/repo/linux/master/include/linux/swapops.h: In function
> > > > ‘set_pmd_migration_entry’:
> > > > /home/wale/repo/linux/master/include/linux/swapops.h:346:1: error: no
> > > > return statement in function returning non-void [-Werror=return-type]
> > > >   346 | }
> > > >       | ^
> > > > cc1: some warnings being treated as errors
> > > > make[4]: *** [/home/wale/repo/linux/master/scripts/Makefile.build:287:
> > > > mm/fadvise.o] Error 1
> > > > make[4]: *** Waiting for unfinished jobs....
> > > >
> > > > There is an example on my
> > > > github(https://github.com/wale-ftd/linux/commits/linux-6.12/) that
> > > > used -O0 to compile the 6.12 kernel. If you're interested, you can
> > > > check them out.
> > >
> > > This is not my question. The BUILD_BUG here means this code
> > > should never be built. If it is built, then that indicates a bug.
> > > So my question is: under what kernel configuration would this
> > > code actually get built?
> >
> > Hello Barry,
> >
> > I'm very sorry for misunderstanding your meaning.
>
> No. You still don’t understand my point. The BUILD_BUG means this
> code will never have a chance to be compiled under any possible
> kernel configuration. That makes it dead code. If there is any
> kernel .config in which it can be built, then we should fix it.
>
> This set_pmd_migration_entry() has no chance of ever being built.
>
> let me give you an example:
> #if 0
> aaa;
> #endif
>
> main()
> {
>
> }
>
> Imagine this set_pmd_migration_entry() as aaa.
>
> >
> > For -O0, __OPTIMIZE__ isn't defined, so line 615 is hit.
> >
> > 595 #ifdef __OPTIMIZE__
> > 602 # define __compiletime_assert(condition, msg, prefix, suffix)           \
> > 612                         prefix ## suffix();
> > 614 #else
> > 615 # define __compiletime_assert(condition, msg, prefix, suffix)
> > ((void)(condition))
> > 616 #endif
> >
> > Finally, set_pmd_migration_entry is defined as follow:
> > static inline int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
> >                 struct page *page)
> > {
> > -        BUILD_BUG();
> > +        ((void)(0));
> > }
> >
>
> Thanks
> Barry

Re: [PATCH 1/1] mm,swapops: Fix compile error on set_pmd_migration_entry.

[Barry Song]

On Sat, Dec 27, 2025 at 5:58 PM wale zhang <wale.zhang.ftd@gmail.com> wrote:
>
> On Sat, Dec 27, 2025 at 4:16 PM Barry Song <21cnbao@gmail.com> wrote:
> >
> > On Sat, Dec 27, 2025 at 8:51 PM wale zhang <wale.zhang.ftd@gmail.com> wrote:
> > >
> > > On Sat, Dec 27, 2025 at 2:43 AM Barry Song <21cnbao@gmail.com> wrote:
> > > >
> > > > On Sat, Dec 27, 2025 at 12:46 AM wale zhang <wale.zhang.ftd@gmail.com> wrote:
> > > > >
> > > > > On Thu, Dec 25, 2025 at 6:22 PM Barry Song <21cnbao@gmail.com> wrote:
> > > > > >
> > > > > > On Thu, Dec 25, 2025 at 10:41 PM Wale Zhang <wale.zhang.ftd@gmail.com> wrote:
> > > > > > >
> > > > > > > ../include/linux/swapops.h: In function ‘set_pmd_migration_entry’:
> > > > > > > ../include/linux/swapops.h:346:1: error: no return statement in function returning non-void [-Werror=return-type]
> > > > > > >   346 | }
> > > > > > >       | ^
> > > > > > > cc1: some warnings being treated as errors
> > > > > > >
> > > > > > > Signed-off-by: Wale Zhang <wale.zhang.ftd@gmail.com>
> > > > > >
> > > > > > Sorry. NAK.
> > > > > >
> > > > > > This looks odd. We already have a BUILD_BUG(), so no fix is needed.
> > > > > >
> > > > > > we should already have:
> > > > > > ././include/linux/compiler_types.h:631:45: error: call to
> > > > > > ‘__compiletime_assert_907’ declared with attribute error: BUILD_BUG
> > > > > > failed
> > > > > >   631 |         _compiletime_assert(condition, msg,
> > > > > > __compiletime_assert_, __COUNTER__)
> > > > > >       |                                             ^
> > > > > > ././include/linux/compiler_types.h:612:25: note: in definition of
> > > > > > macro ‘__compiletime_assert’
> > > > > >   612 |                         prefix ## suffix();
> > > > > >          \
> > > > > >       |                         ^~~~~~
> > > > > >
> > > > > > Do you actually see this being built? If so, please help identify why it
> > > > > > gets built.
> > > > >
> > > > > Hello Barry,
> > > > >
> > > > > I compile the 6.19-rc2 for arm64 with -O0, this kind of error log will appear.
> > > > > In file included from /home/wale/repo/linux/master/include/linux/leafops.h:11,
> > > > >                  from
> > > > > /home/wale/repo/linux/master/include/linux/userfaultfd_k.h:19,
> > > > >                  from /home/wale/repo/linux/master/include/linux/mm_inline.h:10,
> > > > >                  from /home/wale/repo/linux/master/mm/internal.h:13,
> > > > >                  from /home/wale/repo/linux/master/mm/fadvise.c:24:
> > > > > /home/wale/repo/linux/master/include/linux/swapops.h: In function
> > > > > ‘set_pmd_migration_entry’:
> > > > > /home/wale/repo/linux/master/include/linux/swapops.h:346:1: error: no
> > > > > return statement in function returning non-void [-Werror=return-type]
> > > > >   346 | }
> > > > >       | ^
> > > > > cc1: some warnings being treated as errors
> > > > > make[4]: *** [/home/wale/repo/linux/master/scripts/Makefile.build:287:
> > > > > mm/fadvise.o] Error 1
> > > > > make[4]: *** Waiting for unfinished jobs....
> > > > >
> > > > > There is an example on my
> > > > > github(https://github.com/wale-ftd/linux/commits/linux-6.12/) that
> > > > > used -O0 to compile the 6.12 kernel. If you're interested, you can
> > > > > check them out.
> > > >
> > > > This is not my question. The BUILD_BUG here means this code
> > > > should never be built. If it is built, then that indicates a bug.
> > > > So my question is: under what kernel configuration would this
> > > > code actually get built?
> > >
> > > Hello Barry,
> > >
> > > I'm very sorry for misunderstanding your meaning.
> >
> > No. You still don’t understand my point. The BUILD_BUG means this
> > code will never have a chance to be compiled under any possible
> > kernel configuration. That makes it dead code. If there is any
> > kernel .config in which it can be built, then we should fix it.
> >
> > This set_pmd_migration_entry() has no chance of ever being built.
>
> Hello Barry,
>
> set_pmd_migration_entry() is defined in include/linux/swapops.h.
> For not defining CONFIG_ARCH_ENABLE_THP_MIGRATION, line 342~346 will
> be built as long as *.c file includes swapops.h.
> There are many .c files including the header file directly or
> indirectly, such as fadvise.c, shm.c ...

As far as I understand, set_pmd_migration_entry is never
called unless CONFIG_ARCH_ENABLE_THP_MIGRATION is enabled.
If that is correct, no fix is required.

Re: [PATCH 1/1] mm,swapops: Fix compile error on set_pmd_migration_entry.

[Andrew Morton]

On Sat, 27 Dec 2025 17:58:40 +0800 wale zhang <wale.zhang.ftd@gmail.com> wrote:

> > No. You still don’t understand my point. The BUILD_BUG means this
> > code will never have a chance to be compiled under any possible
> > kernel configuration. That makes it dead code. If there is any
> > kernel .config in which it can be built, then we should fix it.
> >
> > This set_pmd_migration_entry() has no chance of ever being built.
> 
> Hello Barry,
> 

Please share the .config which was used to trigger this build error.

(It is strange to have this should-never-be-compiled code.  Why not just
delete it and let the compiler/linker report the error.)

Re: [PATCH 1/1] mm,swapops: Fix compile error on set_pmd_migration_entry.

[Barry Song]

On Mon, Dec 29, 2025 at 10:37 AM Andrew Morton <akpm@linux-foundation.org> wrote:
>
> On Sat, 27 Dec 2025 17:58:40 +0800 wale zhang <wale.zhang.ftd@gmail.com> wrote:
>
> > > No. You still don’t understand my point. The BUILD_BUG means this
> > > code will never have a chance to be compiled under any possible
> > > kernel configuration. That makes it dead code. If there is any
> > > kernel .config in which it can be built, then we should fix it.
> > >
> > > This set_pmd_migration_entry() has no chance of ever being built.
> >
> > Hello Barry,
> >
>
> Please share the .config which was used to trigger this build error.
>
> (It is strange to have this should-never-be-compiled code.  Why not just
> delete it and let the compiler/linker report the error.)

Right. We should entirely drop it in mainline. If Wale does want an
O0 build, then the only call to set_pmd_migration_entry() that is not
under

#ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION

is migrate_vma_collect_huge_pmd() in mm/migrate_device.c.

Then what we really need is as follows:

diff --git a/include/linux/swapops.h b/include/linux/swapops.h
index 8cfc966eae48..fdd4375d5d90 100644
--- a/include/linux/swapops.h
+++ b/include/linux/swapops.h
@@ -339,17 +339,6 @@ static inline pmd_t swp_entry_to_pmd(swp_entry_t entry)
 }
 
 #else  /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
-static inline int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
-		struct page *page)
-{
-	BUILD_BUG();
-}
-
-static inline void remove_migration_pmd(struct page_vma_mapped_walk *pvmw,
-		struct page *new)
-{
-	BUILD_BUG();
-}
 
 static inline void pmd_migration_entry_wait(struct mm_struct *m, pmd_t *p) { }
 
diff --git a/mm/migrate_device.c b/mm/migrate_device.c
index 23379663b1e1..74d23c91626a 100644
--- a/mm/migrate_device.c
+++ b/mm/migrate_device.c
@@ -195,8 +195,9 @@ static int migrate_vma_collect_huge_pmd(pmd_t *pmdp, unsigned long start,
 		return migrate_vma_collect_skip(start, end, walk);
 	}
 
-	if (thp_migration_supported() &&
-		(migrate->flags & MIGRATE_VMA_SELECT_COMPOUND) &&
+
+#ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
+	if ((migrate->flags & MIGRATE_VMA_SELECT_COMPOUND) &&
 		(IS_ALIGNED(start, HPAGE_PMD_SIZE) &&
 		 IS_ALIGNED(end, HPAGE_PMD_SIZE))) {
 
@@ -226,6 +227,7 @@ static int migrate_vma_collect_huge_pmd(pmd_t *pmdp, unsigned long start,
 		spin_unlock(ptl);
 		return 0;
 	}
+#endif
 
 fallback:
 	spin_unlock(ptl);

Thanks
Barry

Re: [PATCH 1/1] mm,swapops: Fix compile error on set_pmd_migration_entry.

[wale zhang]

On Mon, Dec 29, 2025 at 6:07 AM Barry Song <21cnbao@gmail.com> wrote:
>
> On Mon, Dec 29, 2025 at 10:37 AM Andrew Morton <akpm@linux-foundation.org> wrote:
> >
> > On Sat, 27 Dec 2025 17:58:40 +0800 wale zhang <wale.zhang.ftd@gmail.com> wrote:
> >
> > > > No. You still don’t understand my point. The BUILD_BUG means this
> > > > code will never have a chance to be compiled under any possible
> > > > kernel configuration. That makes it dead code. If there is any
> > > > kernel .config in which it can be built, then we should fix it.
> > > >
> > > > This set_pmd_migration_entry() has no chance of ever being built.
> > >
> > > Hello Barry,
> > >
> >
> > Please share the .config which was used to trigger this build error.
> >
> > (It is strange to have this should-never-be-compiled code.  Why not just
> > delete it and let the compiler/linker report the error.)
>
> Right. We should entirely drop it in mainline. If Wale does want an
> O0 build, then the only call to set_pmd_migration_entry() that is not
> under
>
> #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
>
> is migrate_vma_collect_huge_pmd() in mm/migrate_device.c.
>
> Then what we really need is as follows:
>
> diff --git a/include/linux/swapops.h b/include/linux/swapops.h
> index 8cfc966eae48..fdd4375d5d90 100644
> --- a/include/linux/swapops.h
> +++ b/include/linux/swapops.h
> @@ -339,17 +339,6 @@ static inline pmd_t swp_entry_to_pmd(swp_entry_t entry)
>  }
>
>  #else  /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
> -static inline int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
> -               struct page *page)
> -{
> -       BUILD_BUG();
> -}
> -
> -static inline void remove_migration_pmd(struct page_vma_mapped_walk *pvmw,
> -               struct page *new)
> -{
> -       BUILD_BUG();
> -}
>
>  static inline void pmd_migration_entry_wait(struct mm_struct *m, pmd_t *p) { }
>
> diff --git a/mm/migrate_device.c b/mm/migrate_device.c
> index 23379663b1e1..74d23c91626a 100644
> --- a/mm/migrate_device.c
> +++ b/mm/migrate_device.c
> @@ -195,8 +195,9 @@ static int migrate_vma_collect_huge_pmd(pmd_t *pmdp, unsigned long start,
>                 return migrate_vma_collect_skip(start, end, walk);
>         }
>
> -       if (thp_migration_supported() &&
> -               (migrate->flags & MIGRATE_VMA_SELECT_COMPOUND) &&
> +
> +#ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
> +       if ((migrate->flags & MIGRATE_VMA_SELECT_COMPOUND) &&
>                 (IS_ALIGNED(start, HPAGE_PMD_SIZE) &&
>                  IS_ALIGNED(end, HPAGE_PMD_SIZE))) {
>
> @@ -226,6 +227,7 @@ static int migrate_vma_collect_huge_pmd(pmd_t *pmdp, unsigned long start,
>                 spin_unlock(ptl);
>                 return 0;
>         }
> +#endif
>
>  fallback:
>         spin_unlock(ptl);
>
> Thanks
> Barry
>

OK. I will prepare the next patch to delete this unnecessary codes.
Thank you for your guidance.

--
Thanks
Wale

Re: [PATCH 1/1] mm,swapops: Fix compile error on set_pmd_migration_entry.

[Matthew Wilcox]

On Fri, Dec 26, 2025 at 07:45:48PM +0800, wale zhang wrote:
> I compile the 6.19-rc2 for arm64 with -O0, this kind of error log will appear.

I don't think we support compilation with -O0.  Why did you think this
would work, and why did you think this was a good idea?

Re: [PATCH 1/1] mm,swapops: Fix compile error on set_pmd_migration_entry.

[wale zhang]

On Tue, Dec 30, 2025 at 1:00 PM Matthew Wilcox <willy@infradead.org> wrote:
>
> On Fri, Dec 26, 2025 at 07:45:48PM +0800, wale zhang wrote:
> > I compile the 6.19-rc2 for arm64 with -O0, this kind of error log will appear.
>
> I don't think we support compilation with -O0.  Why did you think this
> would work, and why did you think this was a good idea?

Hello Matthew,

Compilation with -O0 is mainly for beginners to learn and debug.
I don't want to encounter the same problems every time I pull the
latest code. Perhaps other people will also encounter the same
problem.

And I see that include/linux/pgtable.h has the same usage.
 320 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
 321                                         unsigned long address, pmd_t *pmdp,
 322                                         pmd_t entry, int dirty)
 323 {
 324         BUILD_BUG();
 325         return 0;
 326 }

Thanks
Wale