[PATCH v2 0/3] mm: Restrict the static definition of the per-CPU variable _shared_alloc_tag to s390 and alpha architectures only

[PATCH v2 0/3] mm: Restrict the static definition of the per-CPU variable _shared_alloc_tag to s390 and alpha architectures only

[Hao Ge]

From: Hao Ge <gehao@kylinos.cn>

Recently discovered this entry while checking kallsyms on ARM64:
ffff800083e509c0 D _shared_alloc_tag

If ARCH_NEEDS_WEAK_PER_CPU is not defined((it is only defined for
s390 and alpha architectures),there's no need to statically define
the percpu variable _shared_alloc_tag. As the number of CPUs
increases,the wasted memory will grow correspondingly.

Therefore,we need to implement isolation for this purpose.

However,currently ARCH_NEEDS_WEAK_PER_CPU is a #define and
is enclosed within the #if defined(MODULE) conditional block.

When building the core kernel code for s390 or alpha architectures,
ARCH_NEEDS_WEAK_PER_CPU remains undefined (as it is gated
by #if defined(MODULE)). However,when building modules for these
architectures,the macro is explicitly defined.

Therefore,we need to make ARCH_NEEDS_WEAK_PER_CPU a Kconfig option.
And replace all instances of ARCH_NEEDS_WEAK_PER_CPU in the kernel
code with MODULE_NEEDS_WEAK_PER_CPU,MODULE_NEEDS_WEAK_PER_CPU might
be a more accurate description,because it was only needed for modules.
Then,when defining the percpu variable _shared_alloc_tag,wrap it with the
CONFIG_ARCH_NEEDS_WEAK_PER_CPU condition.

The following version can be regarded as the most original version:
https://lore.kernel.org/all/20250529073537.563107-1-hao.ge@linux.dev/
But unfortunately,it caused build errors on s390.
Based on Suren's guidance and suggestions,
I've refined it into this patch series.
Many thanks to Suren for his patient instruction.

Verify:
     1. On Arm64:
        nm vmlinux | grep "_shared_alloc_tag",no output is returned.
     2. On S390:
        Compile tested.
        nm vmlinux | grep "_shared_alloc_tag"
        00000000015605b4 r __crc__shared_alloc_tag
        0000000001585fef r __kstrtab__shared_alloc_tag
        0000000001586897 r __kstrtabns__shared_alloc_tag
        00000000014f6548 r __ksymtab__shared_alloc_tag
        0000000001a8fa28 D _shared_alloc_tag
        nm net/ceph/libceph.ko | grep "_shared"
        U _shared_alloc_tag
     3. On alpha
        Compile tested.
        nm vmlinux | grep "_shared_alloc_tag"
        fffffc0000b080fa r __kstrtab__shared_alloc_tag
        fffffc0000b07ee7 r __kstrtabns__shared_alloc_tag
        fffffc0000adee98 r __ksymtab__shared_alloc_tag
        fffffc0000b83d38 D _shared_alloc_tag
        nm crypto/cryptomgr.ko | grep "_share"
        U _shared_alloc_tag

v2:
    Heiko pointed out that when defining MODULE_NEEDS_WEAK_PER_CPU,
    the CONFIG_ARCH_NEEDS_WEAK_PER_CPU condition in the v1 version
    should be removed,as it is always true for s390 and alpha
    architectures.And He also pointed out that patches 2-4 need to
    be merged into one patch. Modify the code according to the suggestions
    and update the corresponding commit message.

Hao Ge (3):
  mm/Kconfig: add ARCH_NEEDS_WEAK_PER_CPU Option and enable it for
    s390/alpha
  mm: replace ARCH_NEEDS_WEAK_PER_CPU with MODULE_NEEDS_WEAK_PER_CPU
  mm/alloc_tag: add the CONFIG_ARCH_NEEDS_WEAK_PER_CPU macro when
    statically defining the percpu variable _shared_alloc_tag

 arch/alpha/Kconfig              | 1 +
 arch/alpha/include/asm/percpu.h | 2 +-
 arch/s390/Kconfig               | 1 +
 arch/s390/include/asm/percpu.h  | 2 +-
 include/linux/alloc_tag.h       | 6 +++---
 include/linux/percpu-defs.h     | 4 ++--
 lib/alloc_tag.c                 | 2 ++
 mm/Kconfig                      | 4 ++++
 8 files changed, 15 insertions(+), 7 deletions(-)

-- 
2.25.1

[PATCH v2 1/3] mm/Kconfig: add ARCH_NEEDS_WEAK_PER_CPU Option and enable it for s390/alpha

[Hao Ge]

From: Hao Ge <gehao@kylinos.cn>

Recently discovered this entry while checking kallsyms on ARM64:
ffff800083e509c0 D _shared_alloc_tag

If ARCH_NEEDS_WEAK_PER_CPU is not defined((it is only defined for
s390 and alpha architectures),there's no need to statically define
the percpu variable _shared_alloc_tag. As the number of CPUs
increases,the wasted memory will grow correspondingly.

Therefore,we need to implement isolation for this purpose.

However,currently ARCH_NEEDS_WEAK_PER_CPU is a #define and
is enclosed within the #if defined(MODULE) conditional block.

When building the core kernel code for s390 or alpha architectures,
ARCH_NEEDS_WEAK_PER_CPU remains undefined (as it is gated
by #if defined(MODULE)).However,when building modules for these
architectures,the macro is explicitly defined.

Therefore,we need to make ARCH_NEEDS_WEAK_PER_CPU a Kconfig option.
And replace all instances of ARCH_NEEDS_WEAK_PER_CPU in the kernel
code with MODULE_NEEDS_WEAK_PER_CPU,MODULE_NEEDS_WEAK_PER_CPU might
be a more accurate description.
Then,when defining the percpu variable _shared_alloc_tag,wrap it
with the CONFIG_ARCH_NEEDS_WEAK_PER_CPU condition.

Therefore,we should add the ARCH_NEEDS_WEAK_PER_CPU option
to the mm Kconfig file and enable it for the s390 and alpha
architectures.

Suggested-by: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Hao Ge <gehao@kylinos.cn>
---
 arch/alpha/Kconfig | 1 +
 arch/s390/Kconfig  | 1 +
 mm/Kconfig         | 4 ++++
 3 files changed, 6 insertions(+)

diff --git a/arch/alpha/Kconfig b/arch/alpha/Kconfig
index 109a4cddcd13..ce9bc842e384 100644
--- a/arch/alpha/Kconfig
+++ b/arch/alpha/Kconfig
@@ -7,6 +7,7 @@ config ALPHA
 	select ARCH_HAS_DMA_OPS if PCI
 	select ARCH_MIGHT_HAVE_PC_PARPORT
 	select ARCH_MIGHT_HAVE_PC_SERIO
+	select ARCH_NEEDS_WEAK_PER_CPU
 	select ARCH_NO_PREEMPT
 	select ARCH_NO_SG_CHAIN
 	select ARCH_USE_CMPXCHG_LOCKREF
diff --git a/arch/s390/Kconfig b/arch/s390/Kconfig
index 0c16dc443e2f..716031d22346 100644
--- a/arch/s390/Kconfig
+++ b/arch/s390/Kconfig
@@ -132,6 +132,7 @@ config S390
 	select ARCH_INLINE_WRITE_UNLOCK_IRQ
 	select ARCH_INLINE_WRITE_UNLOCK_IRQRESTORE
 	select ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE
+	select ARCH_NEEDS_WEAK_PER_CPU
 	select ARCH_STACKWALK
 	select ARCH_SUPPORTS_ATOMIC_RMW
 	select ARCH_SUPPORTS_DEBUG_PAGEALLOC
diff --git a/mm/Kconfig b/mm/Kconfig
index e113f713b493..2f55cc95cfcb 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -929,6 +929,10 @@ config ARCH_SUPPORTS_PUD_PFNMAP
 	def_bool y
 	depends on ARCH_SUPPORTS_HUGE_PFNMAP && HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
 
+# s390 and alpha should be enabled,see comments for DECLARE_PER_CPU_SECTION
+config ARCH_NEEDS_WEAK_PER_CPU
+       bool
+
 #
 # UP and nommu archs use km based percpu allocator
 #
-- 
2.25.1

[PATCH v2 2/3] mm: replace ARCH_NEEDS_WEAK_PER_CPU with MODULE_NEEDS_WEAK_PER_CPU

[Hao Ge]

From: Hao Ge <gehao@kylinos.cn>

As stated in the first patch of this patch series,
we make ARCH_NEEDS_WEAK_PER_CPU a Kconfig option.

Previously,ARCH_NEEDS_WEAK_PER_CPU was enclosed within
the #if defined(MODULE) conditional block,and it was only
needed for modules.

So,replace all instances of ARCH_NEEDS_WEAK_PER_CPU
in the kernel code with MODULE_NEEDS_WEAK_PER_CPU,

Suggested-by: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Hao Ge <gehao@kylinos.cn>
---
 arch/alpha/include/asm/percpu.h | 2 +-
 arch/s390/include/asm/percpu.h  | 2 +-
 include/linux/alloc_tag.h       | 6 +++---
 include/linux/percpu-defs.h     | 4 ++--
 4 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/arch/alpha/include/asm/percpu.h b/arch/alpha/include/asm/percpu.h
index 6923249f2d49..b164d3720e9e 100644
--- a/arch/alpha/include/asm/percpu.h
+++ b/arch/alpha/include/asm/percpu.h
@@ -11,7 +11,7 @@
  * Always use weak definitions for percpu variables in modules.
  */
 #if defined(MODULE) && defined(CONFIG_SMP)
-#define ARCH_NEEDS_WEAK_PER_CPU
+#define MODULE_NEEDS_WEAK_PER_CPU
 #endif
 
 #include <asm-generic/percpu.h>
diff --git a/arch/s390/include/asm/percpu.h b/arch/s390/include/asm/percpu.h
index 84f6b8357b45..0afc7ce9c26f 100644
--- a/arch/s390/include/asm/percpu.h
+++ b/arch/s390/include/asm/percpu.h
@@ -18,7 +18,7 @@
  * generate external references.
  */
 #if defined(MODULE)
-#define ARCH_NEEDS_WEAK_PER_CPU
+#define MODULE_NEEDS_WEAK_PER_CPU
 #endif
 
 /*
diff --git a/include/linux/alloc_tag.h b/include/linux/alloc_tag.h
index 8f7931eb7d16..f349cca0ebed 100644
--- a/include/linux/alloc_tag.h
+++ b/include/linux/alloc_tag.h
@@ -88,7 +88,7 @@ static inline struct alloc_tag *ct_to_alloc_tag(struct codetag *ct)
 	return container_of(ct, struct alloc_tag, ct);
 }
 
-#ifdef ARCH_NEEDS_WEAK_PER_CPU
+#ifdef MODULE_NEEDS_WEAK_PER_CPU
 /*
  * When percpu variables are required to be defined as weak, static percpu
  * variables can't be used inside a function (see comments for DECLARE_PER_CPU_SECTION).
@@ -102,7 +102,7 @@ DECLARE_PER_CPU(struct alloc_tag_counters, _shared_alloc_tag);
 		.ct = CODE_TAG_INIT,						\
 		.counters = &_shared_alloc_tag };
 
-#else /* ARCH_NEEDS_WEAK_PER_CPU */
+#else /* MODULE_NEEDS_WEAK_PER_CPU */
 
 #ifdef MODULE
 
@@ -123,7 +123,7 @@ DECLARE_PER_CPU(struct alloc_tag_counters, _shared_alloc_tag);
 
 #endif /* MODULE */
 
-#endif /* ARCH_NEEDS_WEAK_PER_CPU */
+#endif /* MODULE_NEEDS_WEAK_PER_CPU */
 
 DECLARE_STATIC_KEY_MAYBE(CONFIG_MEM_ALLOC_PROFILING_ENABLED_BY_DEFAULT,
 			mem_alloc_profiling_key);
diff --git a/include/linux/percpu-defs.h b/include/linux/percpu-defs.h
index 0aeb0e276a3e..b4ecfc3a7b2b 100644
--- a/include/linux/percpu-defs.h
+++ b/include/linux/percpu-defs.h
@@ -64,13 +64,13 @@
  * 2. Static percpu variables cannot be defined inside a function.
  *
  * Archs which need weak percpu definitions should define
- * ARCH_NEEDS_WEAK_PER_CPU in asm/percpu.h when necessary.
+ * MODULE_NEEDS_WEAK_PER_CPU in asm/percpu.h when necessary.
  *
  * To ensure that the generic code observes the above two
  * restrictions, if CONFIG_DEBUG_FORCE_WEAK_PER_CPU is set weak
  * definition is used for all cases.
  */
-#if defined(ARCH_NEEDS_WEAK_PER_CPU) || defined(CONFIG_DEBUG_FORCE_WEAK_PER_CPU)
+#if defined(MODULE_NEEDS_WEAK_PER_CPU) || defined(CONFIG_DEBUG_FORCE_WEAK_PER_CPU)
 /*
  * __pcpu_scope_* dummy variable is used to enforce scope.  It
  * receives the static modifier when it's used in front of
-- 
2.25.1

[PATCH v2 3/3] mm/alloc_tag: add the CONFIG_ARCH_NEEDS_WEAK_PER_CPU macro when statically defining the percpu variable _shared_alloc_tag

[Hao Ge]

From: Hao Ge <gehao@kylinos.cn>

Recently discovered this entry while checking kallsyms on ARM64:
ffff800083e509c0 D _shared_alloc_tag

If CONFIG_ARCH_NEEDS_WEAK_PER_CPU is not defined(it is only defined for
s390 and alpha architectures),there's no need to statically define
the percpu variable _shared_alloc_tag. As the number of CPUs
increases,the wasted memory will grow correspondingly.

Enclose the definition of _shared_alloc_tag within the
CONFIG_ARCH_NEEDS_WEAK_PER_CPU condition.

Suggested-by: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Hao Ge <gehao@kylinos.cn>
---
 lib/alloc_tag.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/lib/alloc_tag.c b/lib/alloc_tag.c
index c7f602fa7b23..14fd66f26e42 100644
--- a/lib/alloc_tag.c
+++ b/lib/alloc_tag.c
@@ -24,8 +24,10 @@ static bool mem_profiling_support;
 
 static struct codetag_type *alloc_tag_cttype;
 
+#ifdef CONFIG_ARCH_NEEDS_WEAK_PER_CPU
 DEFINE_PER_CPU(struct alloc_tag_counters, _shared_alloc_tag);
 EXPORT_SYMBOL(_shared_alloc_tag);
+#endif
 
 DEFINE_STATIC_KEY_MAYBE(CONFIG_MEM_ALLOC_PROFILING_ENABLED_BY_DEFAULT,
 			mem_alloc_profiling_key);
-- 
2.25.1

Re: [PATCH v2 0/3] mm: Restrict the static definition of the per-CPU variable _shared_alloc_tag to s390 and alpha architectures only

[Suren Baghdasaryan]

On Thu, Jun 12, 2025 at 8:06 PM Hao Ge <hao.ge@linux.dev> wrote:
>
> From: Hao Ge <gehao@kylinos.cn>
>
> Recently discovered this entry while checking kallsyms on ARM64:
> ffff800083e509c0 D _shared_alloc_tag
>
> If ARCH_NEEDS_WEAK_PER_CPU is not defined((it is only defined for
> s390 and alpha architectures),there's no need to statically define
> the percpu variable _shared_alloc_tag. As the number of CPUs
> increases,the wasted memory will grow correspondingly.
>
> Therefore,we need to implement isolation for this purpose.
>
> However,currently ARCH_NEEDS_WEAK_PER_CPU is a #define and
> is enclosed within the #if defined(MODULE) conditional block.
>
> When building the core kernel code for s390 or alpha architectures,
> ARCH_NEEDS_WEAK_PER_CPU remains undefined (as it is gated
> by #if defined(MODULE)). However,when building modules for these
> architectures,the macro is explicitly defined.
>
> Therefore,we need to make ARCH_NEEDS_WEAK_PER_CPU a Kconfig option.
> And replace all instances of ARCH_NEEDS_WEAK_PER_CPU in the kernel
> code with MODULE_NEEDS_WEAK_PER_CPU,MODULE_NEEDS_WEAK_PER_CPU might
> be a more accurate description,because it was only needed for modules.
> Then,when defining the percpu variable _shared_alloc_tag,wrap it with the
> CONFIG_ARCH_NEEDS_WEAK_PER_CPU condition.
>
> The following version can be regarded as the most original version:
> https://lore.kernel.org/all/20250529073537.563107-1-hao.ge@linux.dev/
> But unfortunately,it caused build errors on s390.
> Based on Suren's guidance and suggestions,
> I've refined it into this patch series.
> Many thanks to Suren for his patient instruction.

I think the first two patches in your patchset should be merged together.

>
> Verify:
>      1. On Arm64:
>         nm vmlinux | grep "_shared_alloc_tag",no output is returned.
>      2. On S390:
>         Compile tested.
>         nm vmlinux | grep "_shared_alloc_tag"
>         00000000015605b4 r __crc__shared_alloc_tag
>         0000000001585fef r __kstrtab__shared_alloc_tag
>         0000000001586897 r __kstrtabns__shared_alloc_tag
>         00000000014f6548 r __ksymtab__shared_alloc_tag
>         0000000001a8fa28 D _shared_alloc_tag
>         nm net/ceph/libceph.ko | grep "_shared"
>         U _shared_alloc_tag
>      3. On alpha
>         Compile tested.
>         nm vmlinux | grep "_shared_alloc_tag"
>         fffffc0000b080fa r __kstrtab__shared_alloc_tag
>         fffffc0000b07ee7 r __kstrtabns__shared_alloc_tag
>         fffffc0000adee98 r __ksymtab__shared_alloc_tag
>         fffffc0000b83d38 D _shared_alloc_tag
>         nm crypto/cryptomgr.ko | grep "_share"
>         U _shared_alloc_tag
>
> v2:
>     Heiko pointed out that when defining MODULE_NEEDS_WEAK_PER_CPU,
>     the CONFIG_ARCH_NEEDS_WEAK_PER_CPU condition in the v1 version
>     should be removed,as it is always true for s390 and alpha
>     architectures.And He also pointed out that patches 2-4 need to
>     be merged into one patch. Modify the code according to the suggestions
>     and update the corresponding commit message.
>
> Hao Ge (3):
>   mm/Kconfig: add ARCH_NEEDS_WEAK_PER_CPU Option and enable it for
>     s390/alpha
>   mm: replace ARCH_NEEDS_WEAK_PER_CPU with MODULE_NEEDS_WEAK_PER_CPU
>   mm/alloc_tag: add the CONFIG_ARCH_NEEDS_WEAK_PER_CPU macro when
>     statically defining the percpu variable _shared_alloc_tag
>
>  arch/alpha/Kconfig              | 1 +
>  arch/alpha/include/asm/percpu.h | 2 +-
>  arch/s390/Kconfig               | 1 +
>  arch/s390/include/asm/percpu.h  | 2 +-
>  include/linux/alloc_tag.h       | 6 +++---
>  include/linux/percpu-defs.h     | 4 ++--
>  lib/alloc_tag.c                 | 2 ++
>  mm/Kconfig                      | 4 ++++
>  8 files changed, 15 insertions(+), 7 deletions(-)
>
> --
> 2.25.1
>

Re: [PATCH v2 3/3] mm/alloc_tag: add the CONFIG_ARCH_NEEDS_WEAK_PER_CPU macro when statically defining the percpu variable _shared_alloc_tag

[Suren Baghdasaryan]

On Thu, Jun 12, 2025 at 8:06 PM Hao Ge <hao.ge@linux.dev> wrote:
>
> From: Hao Ge <gehao@kylinos.cn>
>
> Recently discovered this entry while checking kallsyms on ARM64:
> ffff800083e509c0 D _shared_alloc_tag
>
> If CONFIG_ARCH_NEEDS_WEAK_PER_CPU is not defined(it is only defined for
> s390 and alpha architectures),there's no need to statically define
> the percpu variable _shared_alloc_tag. As the number of CPUs
> increases,the wasted memory will grow correspondingly.

I don't think this patch needs all this background. Just say that
_shared_alloc_tag is not used when CONFIG_ARCH_NEEDS_WEAK_PER_CPU=n,
so you want to compile it out.

>
> Enclose the definition of _shared_alloc_tag within the
> CONFIG_ARCH_NEEDS_WEAK_PER_CPU condition.
>
> Suggested-by: Suren Baghdasaryan <surenb@google.com>
> Signed-off-by: Hao Ge <gehao@kylinos.cn>
> ---
>  lib/alloc_tag.c | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/lib/alloc_tag.c b/lib/alloc_tag.c
> index c7f602fa7b23..14fd66f26e42 100644
> --- a/lib/alloc_tag.c
> +++ b/lib/alloc_tag.c
> @@ -24,8 +24,10 @@ static bool mem_profiling_support;
>
>  static struct codetag_type *alloc_tag_cttype;
>
> +#ifdef CONFIG_ARCH_NEEDS_WEAK_PER_CPU
>  DEFINE_PER_CPU(struct alloc_tag_counters, _shared_alloc_tag);
>  EXPORT_SYMBOL(_shared_alloc_tag);
> +#endif
>
>  DEFINE_STATIC_KEY_MAYBE(CONFIG_MEM_ALLOC_PROFILING_ENABLED_BY_DEFAULT,
>                         mem_alloc_profiling_key);
> --
> 2.25.1
>