Skip to contentSkip to conversation

enable PT_RECLAIM on all 64-bit architectures

PATCH Superseded Qi Zheng 4 participants 23 messages Dec 17, 2025 – Jan 27

Subsystems (10)
Memory management x86 MIPS MM core +6 more subsystems

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Qi Zheng

author
#0 lore raw

v3 cover enable PT_RECLAIM on all 64-bit architectures

From: Qi Zheng <[email protected]>

Changes in v3:
- modify the commit message in [PATCH v3 1/7] (suggested by David Hildenbrand)
- make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE instead of 64BIT
- collect Acked-by
- rebase onto the next-20251217

Changelog in v2:
- fix compilation errors (reported by Magnus Lindholm and kernel test robot)
- adjust some code style (suggested by Huacai Chen)
- make PT_RECLAIM user-unselectable (suggested by David Hildenbrand)
- rebase onto the next-20251119

Hi all,

This series aims to enable PT_RECLAIM on all 64-bit architectures.

On a 64-bit system, madvise(MADV_DONTNEED) may cause a large number of empty PTE page table pages (such as 100GB+). To resolve this problem, we need to enable PT_RECLAIM, which depends on MMU_GATHER_RCU_TABLE_FREE.

Therefore, this series first enables MMU_GATHER_RCU_TABLE_FREE on all 64-bit architectures, and finally makes PT_RECLAIM depend on MMU_GATHER_RCU_TABLE_FREE. This way, PT_RECLAIM can be enabled by default on all 64-bit architectures.

Of course, this will also be enabled on some 32-bit architectures that already support MMU_GATHER_RCU_TABLE_FREE. That's fine, PT_RECLAIM works well on all 32-bit architectures as well. Although the benefit isn't significant, there's still memory that can be reclaimed. Perhaps PT_RECLAIM can be enabled on all 32-bit architectures in the future.

Comments and suggestions are welcome!

Thanks,
Qi

Qi Zheng (7):
mm: change mm/pt_reclaim.c to use asm/tlb.h instead of
asm-generic/tlb.h
alpha: mm: enable MMU_GATHER_RCU_TABLE_FREE
LoongArch: mm: enable MMU_GATHER_RCU_TABLE_FREE
mips: mm: enable MMU_GATHER_RCU_TABLE_FREE
parisc: mm: enable MMU_GATHER_RCU_TABLE_FREE
um: mm: enable MMU_GATHER_RCU_TABLE_FREE
mm: make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE

 arch/alpha/Kconfig                   | 1 +
 arch/alpha/include/asm/tlb.h         | 6 +++---
 arch/loongarch/Kconfig               | 1 +
 arch/loongarch/include/asm/pgalloc.h | 7 +++----
 arch/mips/Kconfig                    | 1 +
 arch/mips/include/asm/pgalloc.h      | 7 +++----
 arch/parisc/Kconfig                  | 1 +
 arch/parisc/include/asm/tlb.h        | 4 ++--
 arch/um/Kconfig                      | 1 +
 arch/x86/Kconfig                     | 1 -
 mm/Kconfig                           | 9 ++-------
 mm/pt_reclaim.c                      | 2 +-
 12 files changed, 19 insertions(+), 22 deletions(-)
Signature · 1 line
2.20.1
author
#1 lore raw

PATCH v3 1/7 mm: change mm/pt_reclaim.c to use asm/tlb.h instead of asm-generic/tlb.h

From: Qi Zheng <[email protected]>

Generally, the asm/tlb.h will include asm-generic/tlb.h, so change mm/pt_reclaim.c to use asm/tlb.h instead of asm-generic/tlb.h. This is a preparation for enabling CONFIG_PT_RECLAIM on other architectures, such as alpha.

---
 mm/pt_reclaim.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/pt_reclaim.c +1 -1
index 0d9cfbf4fe5d8..46771cfff8239 100644
@@ -2,7 +2,7 @@
 #include <linux/hugetlb.h>
 #include <linux/pgalloc.h>
 
-#include <asm-generic/tlb.h>
+#include <asm/tlb.h>
 
 #include "internal.h"
 
Signature · 1 line
2.20.1
author
#2 lore raw

PATCH v3 2/7 alpha: mm: enable MMU_GATHER_RCU_TABLE_FREE

In reply to Qi Zheng

From: Qi Zheng <[email protected]>

On a 64-bit system, madvise(MADV_DONTNEED) may cause a large number of empty PTE page table pages (such as 100GB+). To resolve this problem, first enable MMU_GATHER_RCU_TABLE_FREE to prepare for enabling the PT_RECLAIM feature, which resolves this problem.

Cc Richard Henderson <[email protected]>
Cc Matt Turner <[email protected]>
---
 arch/alpha/Kconfig           | 1 +
 arch/alpha/include/asm/tlb.h | 6 +++---
 2 files changed, 4 insertions(+), 3 deletions(-)
arch/alpha/Kconfig +1 -0
index 80367f2cf821c..6c7dbf0adad62 100644
@@ -38,6 +38,7 @@ config ALPHA
 	select OLD_SIGSUSPEND
 	select CPU_NO_EFFICIENT_FFS if !ALPHA_EV67
 	select MMU_GATHER_NO_RANGE
+	select MMU_GATHER_RCU_TABLE_FREE
 	select SPARSEMEM_EXTREME if SPARSEMEM
 	select ZONE_DMA
 	help
arch/alpha/include/asm/tlb.h +3 -3
index 4f79e331af5ea..ad586b898fd6b 100644
@@ -4,7 +4,7 @@
 
 #include <asm-generic/tlb.h>
 
-#define __pte_free_tlb(tlb, pte, address)		pte_free((tlb)->mm, pte)
-#define __pmd_free_tlb(tlb, pmd, address)		pmd_free((tlb)->mm, pmd)
- 
+#define __pte_free_tlb(tlb, pte, address)	tlb_remove_ptdesc((tlb), page_ptdesc(pte))
+#define __pmd_free_tlb(tlb, pmd, address)	tlb_remove_ptdesc((tlb), virt_to_ptdesc(pmd))
+
 #endif
Signature · 1 line
2.20.1
author
#3 lore raw

PATCH v3 3/7 LoongArch: mm: enable MMU_GATHER_RCU_TABLE_FREE

In reply to Qi Zheng

From: Qi Zheng <[email protected]>

On a 64-bit system, madvise(MADV_DONTNEED) may cause a large number of empty PTE page table pages (such as 100GB+). To resolve this problem, first enable MMU_GATHER_RCU_TABLE_FREE to prepare for enabling the PT_RECLAIM feature, which resolves this problem.

Cc Huacai Chen <[email protected]>
Cc WANG Xuerui <[email protected]>
---
 arch/loongarch/Kconfig               | 1 +
 arch/loongarch/include/asm/pgalloc.h | 7 +++----
 2 files changed, 4 insertions(+), 4 deletions(-)
arch/loongarch/Kconfig +1 -0
index 730f342145197..43d5b863e1fb2 100644
@@ -187,6 +187,7 @@ config LOONGARCH
 	select IRQ_LOONGARCH_CPU
 	select LOCK_MM_AND_FIND_VMA
 	select MMU_GATHER_MERGE_VMAS if MMU
+	select MMU_GATHER_RCU_TABLE_FREE
 	select MODULES_USE_ELF_RELA if MODULES
 	select NEED_PER_CPU_EMBED_FIRST_CHUNK
 	select NEED_PER_CPU_PAGE_FIRST_CHUNK
arch/loongarch/include/asm/pgalloc.h +3 -4
index 08dcc698ec184..248f62d0b590e 100644
@@ -55,8 +55,7 @@ static inline pte_t *pte_alloc_one_kernel(struct mm_struct *mm)
 	return pte;
 }
 
-#define __pte_free_tlb(tlb, pte, address)	\
-	tlb_remove_ptdesc((tlb), page_ptdesc(pte))
+#define __pte_free_tlb(tlb, pte, address)	tlb_remove_ptdesc((tlb), page_ptdesc(pte))
 
 #ifndef __PAGETABLE_PMD_FOLDED
 
@@ -79,7 +78,7 @@ static inline pmd_t *pmd_alloc_one(struct mm_struct *mm, unsigned long address)
 	return pmd;
 }
 
-#define __pmd_free_tlb(tlb, x, addr)	pmd_free((tlb)->mm, x)
+#define __pmd_free_tlb(tlb, x, addr)	tlb_remove_ptdesc((tlb), virt_to_ptdesc(x))
 
 #endif
 
@@ -99,7 +98,7 @@ static inline pud_t *pud_alloc_one(struct mm_struct *mm, unsigned long address)
 	return pud;
 }
 
-#define __pud_free_tlb(tlb, x, addr)	pud_free((tlb)->mm, x)
+#define __pud_free_tlb(tlb, x, addr)	tlb_remove_ptdesc((tlb), virt_to_ptdesc(x))
 
 #endif /* __PAGETABLE_PUD_FOLDED */
 
Signature · 1 line
2.20.1
author
#4 lore raw

PATCH v3 4/7 mips: mm: enable MMU_GATHER_RCU_TABLE_FREE

In reply to Qi Zheng

From: Qi Zheng <[email protected]>

On a 64-bit system, madvise(MADV_DONTNEED) may cause a large number of empty PTE page table pages (such as 100GB+). To resolve this problem, first enable MMU_GATHER_RCU_TABLE_FREE to prepare for enabling the PT_RECLAIM feature, which resolves this problem.

Cc Thomas Bogendoerfer <[email protected]>
---
 arch/mips/Kconfig               | 1 +
 arch/mips/include/asm/pgalloc.h | 7 +++----
 2 files changed, 4 insertions(+), 4 deletions(-)
arch/mips/Kconfig +1 -0
index b88b97139fa8e..c0c94e26ce396 100644
@@ -99,6 +99,7 @@ config MIPS
 	select IRQ_FORCED_THREADING
 	select ISA if EISA
 	select LOCK_MM_AND_FIND_VMA
+	select MMU_GATHER_RCU_TABLE_FREE
 	select MODULES_USE_ELF_REL if MODULES
 	select MODULES_USE_ELF_RELA if MODULES && 64BIT
 	select PERF_USE_VMALLOC
arch/mips/include/asm/pgalloc.h +3 -4
index 7a04381efa0b5..895bf79e76762 100644
@@ -48,8 +48,7 @@ static inline void pud_populate(struct mm_struct *mm, pud_t *pud, pmd_t *pmd)
 extern void pgd_init(void *addr);
 extern pgd_t *pgd_alloc(struct mm_struct *mm);
 
-#define __pte_free_tlb(tlb, pte, address)	\
-	tlb_remove_ptdesc((tlb), page_ptdesc(pte))
+#define __pte_free_tlb(tlb, pte, address)	tlb_remove_ptdesc((tlb), page_ptdesc(pte))
 
 #ifndef __PAGETABLE_PMD_FOLDED
 
@@ -72,7 +71,7 @@ static inline pmd_t *pmd_alloc_one(struct mm_struct *mm, unsigned long address)
 	return pmd;
 }
 
-#define __pmd_free_tlb(tlb, x, addr)	pmd_free((tlb)->mm, x)
+#define __pmd_free_tlb(tlb, x, addr)	tlb_remove_ptdesc((tlb), virt_to_ptdesc(x))
 
 #endif
 
@@ -97,7 +96,7 @@ static inline void p4d_populate(struct mm_struct *mm, p4d_t *p4d, pud_t *pud)
 	set_p4d(p4d, __p4d((unsigned long)pud));
 }
 
-#define __pud_free_tlb(tlb, x, addr)	pud_free((tlb)->mm, x)
+#define __pud_free_tlb(tlb, x, addr)	tlb_remove_ptdesc((tlb), virt_to_ptdesc(x))
 
 #endif /* __PAGETABLE_PUD_FOLDED */
 
Signature · 1 line
2.20.1
author
#5 lore raw

PATCH v3 5/7 parisc: mm: enable MMU_GATHER_RCU_TABLE_FREE

In reply to Qi Zheng

From: Qi Zheng <[email protected]>

On a 64-bit system, madvise(MADV_DONTNEED) may cause a large number of empty PTE page table pages (such as 100GB+). To resolve this problem, first enable MMU_GATHER_RCU_TABLE_FREE to prepare for enabling the PT_RECLAIM feature, which resolves this problem.

Cc "James E.J. Bottomley" <[email protected]>
Cc Helge Deller <[email protected]>
---
 arch/parisc/Kconfig           | 1 +
 arch/parisc/include/asm/tlb.h | 4 ++--
 2 files changed, 3 insertions(+), 2 deletions(-)
arch/parisc/Kconfig +1 -0
index 47fd9662d8005..62d5a89d5c7bc 100644
@@ -79,6 +79,7 @@ config PARISC
 	select GENERIC_CLOCKEVENTS
 	select CPU_NO_EFFICIENT_FFS
 	select THREAD_INFO_IN_TASK
+	select MMU_GATHER_RCU_TABLE_FREE
 	select NEED_DMA_MAP_STATE
 	select NEED_SG_DMA_LENGTH
 	select HAVE_ARCH_KGDB
arch/parisc/include/asm/tlb.h +2 -2
index 44235f367674d..4501fee0a8fa4 100644
@@ -5,8 +5,8 @@
 #include <asm-generic/tlb.h>
 
 #if CONFIG_PGTABLE_LEVELS == 3
-#define __pmd_free_tlb(tlb, pmd, addr)	pmd_free((tlb)->mm, pmd)
+#define __pmd_free_tlb(tlb, pmd, addr)	tlb_remove_ptdesc((tlb), virt_to_ptdesc(pmd))
 #endif
-#define __pte_free_tlb(tlb, pte, addr)	pte_free((tlb)->mm, pte)
+#define __pte_free_tlb(tlb, pte, addr)	tlb_remove_ptdesc((tlb), page_ptdesc(pte))
 
 #endif
Signature · 1 line
2.20.1
author
#6 lore raw

PATCH v3 6/7 um: mm: enable MMU_GATHER_RCU_TABLE_FREE

In reply to Qi Zheng

From: Qi Zheng <[email protected]>

On a 64-bit system, madvise(MADV_DONTNEED) may cause a large number of empty PTE page table pages (such as 100GB+). To resolve this problem, first enable MMU_GATHER_RCU_TABLE_FREE to prepare for enabling the PT_RECLAIM feature, which resolves this problem.

Cc Richard Weinberger <[email protected]>
Cc Anton Ivanov <[email protected]>
Cc Johannes Berg <[email protected]>
---
 arch/um/Kconfig | 1 +
 1 file changed, 1 insertion(+)
arch/um/Kconfig +1 -0
index 8415d39b0d430..098cda44db225 100644
@@ -42,6 +42,7 @@ config UML
 	select HAVE_SYSCALL_TRACEPOINTS
 	select THREAD_INFO_IN_TASK
 	select SPARSE_IRQ
+	select MMU_GATHER_RCU_TABLE_FREE
 
 config MMU
 	bool
Signature · 1 line
2.20.1
author
#7 lore raw

PATCH v3 7/7 mm: make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE

In reply to Qi Zheng

From: Qi Zheng <[email protected]>

The PT_RECLAIM can work on all architectures that support MMU_GATHER_RCU_TABLE_FREE, so make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE.

BTW, change PT_RECLAIM to be enabled by default, since nobody should want to turn it off.

---
 arch/x86/Kconfig | 1 -
 mm/Kconfig       | 9 ++-------
 2 files changed, 2 insertions(+), 8 deletions(-)
arch/x86/Kconfig +0 -1
index 80527299f859a..0d22da56a71b0 100644
@@ -331,7 +331,6 @@ config X86
 	select FUNCTION_ALIGNMENT_4B
 	imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI
 	select HAVE_DYNAMIC_FTRACE_NO_PATCHABLE
-	select ARCH_SUPPORTS_PT_RECLAIM		if X86_64
 	select ARCH_SUPPORTS_SCHED_SMT		if SMP
 	select SCHED_SMT			if SMP
 	select ARCH_SUPPORTS_SCHED_CLUSTER	if SMP
mm/Kconfig +2 -7
index bd0ea5454af82..fc00b429b7129 100644
@@ -1447,14 +1447,9 @@ config ARCH_HAS_USER_SHADOW_STACK
 	  The architecture has hardware support for userspace shadow call
           stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss).
 
-config ARCH_SUPPORTS_PT_RECLAIM
-	def_bool n
-
 config PT_RECLAIM
-	bool "reclaim empty user page table pages"
-	default y
-	depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP
-	select MMU_GATHER_RCU_TABLE_FREE
+	def_bool y
+	depends on MMU_GATHER_RCU_TABLE_FREE
 	help
 	  Try to reclaim empty user page table pages in paths other than munmap
 	  and exit_mmap path.
Signature · 1 line
2.20.1
#8 lore raw

Re: [PATCH v3 7/7] mm: make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE

Quoted 45 lines · Qi Zheng

From: Qi Zheng <[email protected]>

The PT_RECLAIM can work on all architectures that support MMU_GATHER_RCU_TABLE_FREE, so make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE.

BTW, change PT_RECLAIM to be enabled by default, since nobody should want to turn it off.

---
arch/x86/Kconfig | 1 -
mm/Kconfig | 9 ++-------

2 files changed, 2 insertions(+), 8 deletions(-)
arch/x86/Kconfig +0 -1
index 80527299f859a..0d22da56a71b0 100644
@@ -331,7 +331,6 @@ config X86
 	select FUNCTION_ALIGNMENT_4B
 	imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI
 	select HAVE_DYNAMIC_FTRACE_NO_PATCHABLE
-	select ARCH_SUPPORTS_PT_RECLAIM		if X86_64
 	select ARCH_SUPPORTS_SCHED_SMT		if SMP
 	select SCHED_SMT			if SMP
 	select ARCH_SUPPORTS_SCHED_CLUSTER	if SMP
mm/Kconfig +0 -3
index bd0ea5454af82..fc00b429b7129 100644
@@ -1447,14 +1447,9 @@ config ARCH_HAS_USER_SHADOW_STACK
 	  The architecture has hardware support for userspace shadow call
          stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss).
 
-config ARCH_SUPPORTS_PT_RECLAIM
-	def_bool n
-

config PT_RECLAIM

mm/Kconfig +2 -4
-	bool "reclaim empty user page table pages"
-	default y
-	depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP
-	select MMU_GATHER_RCU_TABLE_FREE
+	def_bool y
+	depends on MMU_GATHER_RCU_TABLE_FREE
Quoted 3 lines

help
Try to reclaim empty user page table pages in paths other than munmap and exit_mmap path.

Hi, Qi

I am new to PT_RECLAIM, when reading related code I got one question.

Before this patch, we could have this config combination:

CONFIG_MMU_GATHER_RCU_TABLE_FREE & !CONFIG_PT_RECLAIM

This means tlb_remove_table_free() is rcu version while tlb_remove_table_one() is semi rcu version.

I am curious could we use rcu version tlb_remove_table_one() for this case? Use rcu version tlb_remove_table_one() if CONFIG_MMU_GATHER_RCU_TABLE_FREE. Is there some limitation here?

Thanks in advance for your explanation.

Signature · 2 lines
Wei Yang
Help you, Help me
author
#9 lore raw

Re: [PATCH v3 7/7] mm: make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE

Quoted 56 lines · Wei Yang
Quoted 45 lines · Qi Zheng

From: Qi Zheng <[email protected]>

The PT_RECLAIM can work on all architectures that support MMU_GATHER_RCU_TABLE_FREE, so make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE.

BTW, change PT_RECLAIM to be enabled by default, since nobody should want to turn it off.

---
arch/x86/Kconfig | 1 -
mm/Kconfig | 9 ++-------

2 files changed, 2 insertions(+), 8 deletions(-)
arch/x86/Kconfig +0 -1
index 80527299f859a..0d22da56a71b0 100644
@@ -331,7 +331,6 @@ config X86
 	select FUNCTION_ALIGNMENT_4B
 	imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI
 	select HAVE_DYNAMIC_FTRACE_NO_PATCHABLE
-	select ARCH_SUPPORTS_PT_RECLAIM		if X86_64
 	select ARCH_SUPPORTS_SCHED_SMT		if SMP
 	select SCHED_SMT			if SMP
 	select ARCH_SUPPORTS_SCHED_CLUSTER	if SMP
mm/Kconfig +0 -3
index bd0ea5454af82..fc00b429b7129 100644
@@ -1447,14 +1447,9 @@ config ARCH_HAS_USER_SHADOW_STACK
 	  The architecture has hardware support for userspace shadow call
           stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss).
 
-config ARCH_SUPPORTS_PT_RECLAIM
-	def_bool n
-

config PT_RECLAIM

mm/Kconfig +2 -4
-	bool "reclaim empty user page table pages"
-	default y
-	depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP
-	select MMU_GATHER_RCU_TABLE_FREE
+	def_bool y
+	depends on MMU_GATHER_RCU_TABLE_FREE
Quoted 3 lines

help
Try to reclaim empty user page table pages in paths other than munmap and exit_mmap path.

Hi, Qi

I am new to PT_RECLAIM, when reading related code I got one question.

Before this patch, we could have this config combination:

Quoted 8 lines

CONFIG_MMU_GATHER_RCU_TABLE_FREE & !CONFIG_PT_RECLAIM

This means tlb_remove_table_free() is rcu version while tlb_remove_table_one() is semi rcu version.

I am curious could we use rcu version tlb_remove_table_one() for this case? Use rcu version tlb_remove_table_one() if CONFIG_MMU_GATHER_RCU_TABLE_FREE. Is there some limitation here?

I think there's no problem. The rcu version can also ensure that the fast GUP works well.

Quoted 4 lines

Thanks in advance for your explanation.

#10 lore raw

Re: [PATCH v3 7/7] mm: make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE

Quoted 64 lines · Qi Zheng
Quoted 61 lines · Wei Yang
Quoted 45 lines · Qi Zheng

From: Qi Zheng <[email protected]>

The PT_RECLAIM can work on all architectures that support MMU_GATHER_RCU_TABLE_FREE, so make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE.

BTW, change PT_RECLAIM to be enabled by default, since nobody should want to turn it off.

---
arch/x86/Kconfig | 1 -
mm/Kconfig | 9 ++-------

2 files changed, 2 insertions(+), 8 deletions(-)
arch/x86/Kconfig +0 -1
index 80527299f859a..0d22da56a71b0 100644
@@ -331,7 +331,6 @@ config X86
 	select FUNCTION_ALIGNMENT_4B
 	imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI
 	select HAVE_DYNAMIC_FTRACE_NO_PATCHABLE
-	select ARCH_SUPPORTS_PT_RECLAIM		if X86_64
 	select ARCH_SUPPORTS_SCHED_SMT		if SMP
 	select SCHED_SMT			if SMP
 	select ARCH_SUPPORTS_SCHED_CLUSTER	if SMP
mm/Kconfig +0 -3
index bd0ea5454af82..fc00b429b7129 100644
@@ -1447,14 +1447,9 @@ config ARCH_HAS_USER_SHADOW_STACK
 	  The architecture has hardware support for userspace shadow call
           stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss).
 
-config ARCH_SUPPORTS_PT_RECLAIM
-	def_bool n
-

config PT_RECLAIM

mm/Kconfig +2 -4
-	bool "reclaim empty user page table pages"
-	default y
-	depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP
-	select MMU_GATHER_RCU_TABLE_FREE
+	def_bool y
+	depends on MMU_GATHER_RCU_TABLE_FREE
Quoted 3 lines

help
Try to reclaim empty user page table pages in paths other than munmap and exit_mmap path.

Hi, Qi

I am new to PT_RECLAIM, when reading related code I got one question.

Before this patch, we could have this config combination:

CONFIG_MMU_GATHER_RCU_TABLE_FREE & !CONFIG_PT_RECLAIM

This means tlb_remove_table_free() is rcu version while tlb_remove_table_one() is semi rcu version.

Quoted 7 lines
Quoted 3 lines

I am curious could we use rcu version tlb_remove_table_one() for this case? Use rcu version tlb_remove_table_one() if CONFIG_MMU_GATHER_RCU_TABLE_FREE. Is there some limitation here?

I think there's no problem. The rcu version can also ensure that the fast GUP works well.

Thanks for your quick response :-)

And Happy New Year

So my little suggestion is move the definition of __tlb_remove_table_one() under CONFIG_MMU_GATHER_RCU_TABLE_FREE. Do you thinks this would be more clear?

Quoted 4 lines
Quoted 4 lines

Thanks in advance for your explanation.

Signature · 2 lines
Wei Yang
Help you, Help me
#11 lore raw

Re: [PATCH v3 7/7] mm: make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE

Quoted 73 lines · Wei Yang
Quoted 64 lines · Qi Zheng
Quoted 61 lines · Wei Yang
Quoted 45 lines · Qi Zheng

From: Qi Zheng <[email protected]>

The PT_RECLAIM can work on all architectures that support MMU_GATHER_RCU_TABLE_FREE, so make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE.

BTW, change PT_RECLAIM to be enabled by default, since nobody should want to turn it off.

---
arch/x86/Kconfig | 1 -
mm/Kconfig | 9 ++-------

2 files changed, 2 insertions(+), 8 deletions(-)
arch/x86/Kconfig +0 -1
index 80527299f859a..0d22da56a71b0 100644
@@ -331,7 +331,6 @@ config X86
 	select FUNCTION_ALIGNMENT_4B
 	imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI
 	select HAVE_DYNAMIC_FTRACE_NO_PATCHABLE
-	select ARCH_SUPPORTS_PT_RECLAIM		if X86_64
 	select ARCH_SUPPORTS_SCHED_SMT		if SMP
 	select SCHED_SMT			if SMP
 	select ARCH_SUPPORTS_SCHED_CLUSTER	if SMP
mm/Kconfig +0 -3
index bd0ea5454af82..fc00b429b7129 100644
@@ -1447,14 +1447,9 @@ config ARCH_HAS_USER_SHADOW_STACK
 	  The architecture has hardware support for userspace shadow call
            stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss).
 
-config ARCH_SUPPORTS_PT_RECLAIM
-	def_bool n
-

config PT_RECLAIM

mm/Kconfig +2 -4
-	bool "reclaim empty user page table pages"
-	default y
-	depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP
-	select MMU_GATHER_RCU_TABLE_FREE
+	def_bool y
+	depends on MMU_GATHER_RCU_TABLE_FREE
Quoted 3 lines

help
Try to reclaim empty user page table pages in paths other than munmap and exit_mmap path.

Hi, Qi

I am new to PT_RECLAIM, when reading related code I got one question.

Before this patch, we could have this config combination:

CONFIG_MMU_GATHER_RCU_TABLE_FREE & !CONFIG_PT_RECLAIM

This means tlb_remove_table_free() is rcu version while tlb_remove_table_one() is semi rcu version.

Quoted 7 lines
Quoted 3 lines

I am curious could we use rcu version tlb_remove_table_one() for this case? Use rcu version tlb_remove_table_one() if CONFIG_MMU_GATHER_RCU_TABLE_FREE. Is there some limitation here?

I think there's no problem. The rcu version can also ensure that the fast GUP works well.

Quoted 7 lines

Thanks for your quick response :-)

And Happy New Year

So my little suggestion is move the definition of __tlb_remove_table_one() under CONFIG_MMU_GATHER_RCU_TABLE_FREE. Do you thinks this would be more clear?

Do you mean

mm/mmu_gather.c +1 -1
index 2faa23d7f8d42..6aeba4bae68d2 100644
@@ -319,7 +319,7 @@ static inline void tlb_table_invalidate(struct mmu_gather *tlb)
        }
 }
 
-#ifdef CONFIG_PT_RECLAIM
+#ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE
 static inline void __tlb_remove_table_one_rcu(struct rcu_head *head)
 {
        struct ptdesc *ptdesc;

?

Signature · 3 lines
Cheers

David
#12 lore raw

Re: [PATCH v3 7/7] mm: make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE

Quoted 100 lines · David Hildenbrand (Red Hat)
Quoted 73 lines · Wei Yang
Quoted 64 lines · Qi Zheng
Quoted 61 lines · Wei Yang
Quoted 45 lines · Qi Zheng

From: Qi Zheng <[email protected]>

The PT_RECLAIM can work on all architectures that support MMU_GATHER_RCU_TABLE_FREE, so make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE.

BTW, change PT_RECLAIM to be enabled by default, since nobody should want to turn it off.

---
arch/x86/Kconfig | 1 -
mm/Kconfig | 9 ++-------

2 files changed, 2 insertions(+), 8 deletions(-)
arch/x86/Kconfig +0 -1
index 80527299f859a..0d22da56a71b0 100644
@@ -331,7 +331,6 @@ config X86
 	select FUNCTION_ALIGNMENT_4B
 	imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI
 	select HAVE_DYNAMIC_FTRACE_NO_PATCHABLE
-	select ARCH_SUPPORTS_PT_RECLAIM		if X86_64
 	select ARCH_SUPPORTS_SCHED_SMT		if SMP
 	select SCHED_SMT			if SMP
 	select ARCH_SUPPORTS_SCHED_CLUSTER	if SMP
mm/Kconfig +0 -3
index bd0ea5454af82..fc00b429b7129 100644
@@ -1447,14 +1447,9 @@ config ARCH_HAS_USER_SHADOW_STACK
 	  The architecture has hardware support for userspace shadow call
            stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss).
 
-config ARCH_SUPPORTS_PT_RECLAIM
-	def_bool n
-

config PT_RECLAIM

mm/Kconfig +2 -4
-	bool "reclaim empty user page table pages"
-	default y
-	depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP
-	select MMU_GATHER_RCU_TABLE_FREE
+	def_bool y
+	depends on MMU_GATHER_RCU_TABLE_FREE
Quoted 3 lines

help
Try to reclaim empty user page table pages in paths other than munmap and exit_mmap path.

Hi, Qi

I am new to PT_RECLAIM, when reading related code I got one question.

Before this patch, we could have this config combination:

CONFIG_MMU_GATHER_RCU_TABLE_FREE & !CONFIG_PT_RECLAIM

This means tlb_remove_table_free() is rcu version while tlb_remove_table_one() is semi rcu version.

Quoted 7 lines
Quoted 3 lines

I am curious could we use rcu version tlb_remove_table_one() for this case? Use rcu version tlb_remove_table_one() if CONFIG_MMU_GATHER_RCU_TABLE_FREE. Is there some limitation here?

I think there's no problem. The rcu version can also ensure that the fast GUP works well.

Quoted 7 lines

Thanks for your quick response :-)

And Happy New Year

So my little suggestion is move the definition of __tlb_remove_table_one() under CONFIG_MMU_GATHER_RCU_TABLE_FREE. Do you thinks this would be more clear?

Do you mean

mm/mmu_gather.c +1 -1
index 2faa23d7f8d42..6aeba4bae68d2 100644
@@ -319,7 +319,7 @@ static inline void tlb_table_invalidate(struct mmu_gather
 *tlb)
       }
 }
 
-#ifdef CONFIG_PT_RECLAIM
+#ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE
 static inline void __tlb_remove_table_one_rcu(struct rcu_head *head)
 {
       struct ptdesc *ptdesc;
Quoted 1 line

?

Sorry for the late reply.

Yes, and maybe we can move the definition to the
#ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE code block above, then to be next to tlb_remove_table_free().

So that we always have rcu version when CONFIG_MMU_GATHER_RCU_TABLE_FREE.

Quoted 5 lines
Signature · 4 lines
-- 
Cheers

David
Signature · 2 lines
Wei Yang
Help you, Help me
author
#13 lore raw

Re: [PATCH v3 7/7] mm: make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE

Quoted 105 lines · Wei Yang
Quoted 100 lines · David Hildenbrand (Red Hat)
Quoted 73 lines · Wei Yang
Quoted 64 lines · Qi Zheng
Quoted 61 lines · Wei Yang
Quoted 45 lines · Qi Zheng

From: Qi Zheng <[email protected]>

The PT_RECLAIM can work on all architectures that support MMU_GATHER_RCU_TABLE_FREE, so make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE.

BTW, change PT_RECLAIM to be enabled by default, since nobody should want to turn it off.

---
arch/x86/Kconfig | 1 -
mm/Kconfig | 9 ++-------

2 files changed, 2 insertions(+), 8 deletions(-)
arch/x86/Kconfig +0 -1
index 80527299f859a..0d22da56a71b0 100644
@@ -331,7 +331,6 @@ config X86
 	select FUNCTION_ALIGNMENT_4B
 	imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI
 	select HAVE_DYNAMIC_FTRACE_NO_PATCHABLE
-	select ARCH_SUPPORTS_PT_RECLAIM		if X86_64
 	select ARCH_SUPPORTS_SCHED_SMT		if SMP
 	select SCHED_SMT			if SMP
 	select ARCH_SUPPORTS_SCHED_CLUSTER	if SMP
mm/Kconfig +0 -3
index bd0ea5454af82..fc00b429b7129 100644
@@ -1447,14 +1447,9 @@ config ARCH_HAS_USER_SHADOW_STACK
 	  The architecture has hardware support for userspace shadow call
             stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss).
 
-config ARCH_SUPPORTS_PT_RECLAIM
-	def_bool n
-

config PT_RECLAIM

mm/Kconfig +2 -4
-	bool "reclaim empty user page table pages"
-	default y
-	depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP
-	select MMU_GATHER_RCU_TABLE_FREE
+	def_bool y
+	depends on MMU_GATHER_RCU_TABLE_FREE
Quoted 3 lines

help
Try to reclaim empty user page table pages in paths other than munmap and exit_mmap path.

Hi, Qi

I am new to PT_RECLAIM, when reading related code I got one question.

Before this patch, we could have this config combination:

CONFIG_MMU_GATHER_RCU_TABLE_FREE & !CONFIG_PT_RECLAIM

This means tlb_remove_table_free() is rcu version while tlb_remove_table_one() is semi rcu version.

Quoted 7 lines
Quoted 3 lines

I am curious could we use rcu version tlb_remove_table_one() for this case? Use rcu version tlb_remove_table_one() if CONFIG_MMU_GATHER_RCU_TABLE_FREE. Is there some limitation here?

I think there's no problem. The rcu version can also ensure that the fast GUP works well.

Quoted 7 lines

Thanks for your quick response :-)

And Happy New Year

So my little suggestion is move the definition of __tlb_remove_table_one() under CONFIG_MMU_GATHER_RCU_TABLE_FREE. Do you thinks this would be more clear?

Do you mean

mm/mmu_gather.c +1 -1
index 2faa23d7f8d42..6aeba4bae68d2 100644
@@ -319,7 +319,7 @@ static inline void tlb_table_invalidate(struct mmu_gather
 *tlb)
        }
 }
 
-#ifdef CONFIG_PT_RECLAIM
+#ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE
 static inline void __tlb_remove_table_one_rcu(struct rcu_head *head)
 {
        struct ptdesc *ptdesc;
Quoted 1 line

?

Sorry for the late reply.

Quoted 5 lines

Yes, and maybe we can move the definition to the
#ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE code block above, then to be next to tlb_remove_table_free().

So that we always have rcu version when CONFIG_MMU_GATHER_RCU_TABLE_FREE.

LGTM, could you help submit an official patch?

Thanks,
Qi

Quoted 7 lines
Quoted 6 lines
Signature · 4 lines
-- 
Cheers

David
#14 lore raw

Re: [PATCH v3 7/7] mm: make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE

Quoted 108 lines · Qi Zheng
Quoted 105 lines · Wei Yang
Quoted 100 lines · David Hildenbrand (Red Hat)
Quoted 73 lines · Wei Yang
Quoted 64 lines · Qi Zheng
Quoted 61 lines · Wei Yang
Quoted 45 lines · Qi Zheng

From: Qi Zheng <[email protected]>

The PT_RECLAIM can work on all architectures that support MMU_GATHER_RCU_TABLE_FREE, so make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE.

BTW, change PT_RECLAIM to be enabled by default, since nobody should want to turn it off.

---
arch/x86/Kconfig | 1 -
mm/Kconfig | 9 ++-------

2 files changed, 2 insertions(+), 8 deletions(-)
arch/x86/Kconfig +0 -1
index 80527299f859a..0d22da56a71b0 100644
@@ -331,7 +331,6 @@ config X86
 	select FUNCTION_ALIGNMENT_4B
 	imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI
 	select HAVE_DYNAMIC_FTRACE_NO_PATCHABLE
-	select ARCH_SUPPORTS_PT_RECLAIM		if X86_64
 	select ARCH_SUPPORTS_SCHED_SMT		if SMP
 	select SCHED_SMT			if SMP
 	select ARCH_SUPPORTS_SCHED_CLUSTER	if SMP
mm/Kconfig +0 -3
index bd0ea5454af82..fc00b429b7129 100644
@@ -1447,14 +1447,9 @@ config ARCH_HAS_USER_SHADOW_STACK
 	  The architecture has hardware support for userspace shadow call
             stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss).
 
-config ARCH_SUPPORTS_PT_RECLAIM
-	def_bool n
-

config PT_RECLAIM

mm/Kconfig +2 -4
-	bool "reclaim empty user page table pages"
-	default y
-	depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP
-	select MMU_GATHER_RCU_TABLE_FREE
+	def_bool y
+	depends on MMU_GATHER_RCU_TABLE_FREE
Quoted 3 lines

help
Try to reclaim empty user page table pages in paths other than munmap and exit_mmap path.

Hi, Qi

I am new to PT_RECLAIM, when reading related code I got one question.

Before this patch, we could have this config combination:

CONFIG_MMU_GATHER_RCU_TABLE_FREE & !CONFIG_PT_RECLAIM

This means tlb_remove_table_free() is rcu version while tlb_remove_table_one() is semi rcu version.

Quoted 7 lines
Quoted 3 lines

I am curious could we use rcu version tlb_remove_table_one() for this case? Use rcu version tlb_remove_table_one() if CONFIG_MMU_GATHER_RCU_TABLE_FREE. Is there some limitation here?

I think there's no problem. The rcu version can also ensure that the fast GUP works well.

Quoted 7 lines

Thanks for your quick response :-)

And Happy New Year

So my little suggestion is move the definition of __tlb_remove_table_one() under CONFIG_MMU_GATHER_RCU_TABLE_FREE. Do you thinks this would be more clear?

Do you mean

mm/mmu_gather.c +1 -1
index 2faa23d7f8d42..6aeba4bae68d2 100644
@@ -319,7 +319,7 @@ static inline void tlb_table_invalidate(struct mmu_gather
 *tlb)
        }
 }
 
-#ifdef CONFIG_PT_RECLAIM
+#ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE
 static inline void __tlb_remove_table_one_rcu(struct rcu_head *head)
 {
        struct ptdesc *ptdesc;
Quoted 1 line

?

Sorry for the late reply.

Quoted 8 lines
Quoted 5 lines

Yes, and maybe we can move the definition to the
#ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE code block above, then to be next to tlb_remove_table_free().

So that we always have rcu version when CONFIG_MMU_GATHER_RCU_TABLE_FREE.

LGTM, could you help submit an official patch?

Sure.

Since this is trivial cleanup, I will post it till next merge window.

Quoted 10 lines

Thanks,
Qi

Quoted 7 lines
Quoted 6 lines
Signature · 4 lines
-- 
Cheers

David
Signature · 2 lines
Wei Yang
Help you, Help me
#15 lore raw

Re: [PATCH v3 7/7] mm: make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE

In reply to Qi Zheng
Quoted 38 lines · Qi Zheng

From: Qi Zheng <[email protected]>

The PT_RECLAIM can work on all architectures that support MMU_GATHER_RCU_TABLE_FREE, so make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE.

BTW, change PT_RECLAIM to be enabled by default, since nobody should want to turn it off.

---
  arch/x86/Kconfig | 1 -
  mm/Kconfig       | 9 ++-------
  2 files changed, 2 insertions(+), 8 deletions(-)
arch/x86/Kconfig +0 -1
index 80527299f859a..0d22da56a71b0 100644
@@ -331,7 +331,6 @@ config X86
  	select FUNCTION_ALIGNMENT_4B
  	imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI
  	select HAVE_DYNAMIC_FTRACE_NO_PATCHABLE
-	select ARCH_SUPPORTS_PT_RECLAIM		if X86_64
  	select ARCH_SUPPORTS_SCHED_SMT		if SMP
  	select SCHED_SMT			if SMP
  	select ARCH_SUPPORTS_SCHED_CLUSTER	if SMP
mm/Kconfig +0 -3
index bd0ea5454af82..fc00b429b7129 100644
@@ -1447,14 +1447,9 @@ config ARCH_HAS_USER_SHADOW_STACK
  	  The architecture has hardware support for userspace shadow call
            stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss).
 
-config ARCH_SUPPORTS_PT_RECLAIM
-	def_bool n
-

mm/Kconfig config ARCH_HAS_USER_SHADOW_STACK · line 1455

Quoted 10 lines
mm/Kconfig +2 -4
  config PT_RECLAIM
-	bool "reclaim empty user page table pages"
-	default y
-	depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP
-	select MMU_GATHER_RCU_TABLE_FREE
+	def_bool y
+	depends on MMU_GATHER_RCU_TABLE_FREE
  	help
  	  Try to reclaim empty user page table pages in paths other than munmap
  	  and exit_mmap path.

This patch seems to make s390x compilations sometimes unhappy:

Unverified Warning (likely false positive, kindly check if interested):

mm/memory.c:1911 zap_pte_range() error: uninitialized symbol 'pmdval'.

Warning ids grouped by kconfigs:

recent_errors
`-- s390-randconfig-r072-20260117
`-- mm-memory.c-zap_pte_range()-error:uninitialized-symbol-pmdval-.

I assume the compiler is not able to figure out that only when try_get_and_clear_pmd() returns false that pmdval could be uninitialized.

Maybe it has to do with LTO?

After all, that function resides in a different compilation unit.

Which makes me wonder whether we want to just move try_get_and_clear_pmd() and reclaim_pt_is_enabled() to internal.h or even just memory.c?

But then, maybe we could remove pt_reclaim.c completely and just have try_to_free_pte() in memory.c as well?

I would just do the following cleanup:

From cfe97092f71fcc88f729f07ee0bc6816e3e398f0 Mon Sep 17 00:00:00 2001
From: "David Hildenbrand (Red Hat)" <[email protected]>
Date: Sun, 18 Jan 2026 12:20:55 +0100
Subject: [PATCH] mm: move pte table reclaim code to memory.c

Let's move the code and clean it up a bit along the way.

---
 MAINTAINERS     |  1 -
 mm/internal.h   | 18 -------------
 mm/memory.c     | 70 ++++++++++++++++++++++++++++++++++++++++++-----
 mm/pt_reclaim.c | 72 -------------------------------------------------
 4 files changed, 64 insertions(+), 97 deletions(-)
 delete mode 100644 mm/pt_reclaim.c
MAINTAINERS +0 -1
index 11720728d92f2..28e8e28bca3e5 100644
@@ -16692,7 +16692,6 @@ R:	Shakeel Butt <[email protected]>
 R:	Lorenzo Stoakes <[email protected]>
 L:	[email protected]
 S:	Maintained
-F:	mm/pt_reclaim.c
 F:	mm/vmscan.c
 F:	mm/workingset.c
 
mm/internal.h +0 -18
index 9508dbaf47cd4..ef71a1d9991f2 100644
@@ -1745,24 +1745,6 @@ int walk_page_range_debug(struct mm_struct *mm, unsigned long start,
 			  unsigned long end, const struct mm_walk_ops *ops,
 			  pgd_t *pgd, void *private);
 
-/* pt_reclaim.c */
-bool try_get_and_clear_pmd(struct mm_struct *mm, pmd_t *pmd, pmd_t *pmdval);
-void free_pte(struct mm_struct *mm, unsigned long addr, struct mmu_gather *tlb,
-	      pmd_t pmdval);
-void try_to_free_pte(struct mm_struct *mm, pmd_t *pmd, unsigned long addr,
-		     struct mmu_gather *tlb);
-
-#ifdef CONFIG_PT_RECLAIM
-bool reclaim_pt_is_enabled(unsigned long start, unsigned long end,
-			   struct zap_details *details);
-#else
-static inline bool reclaim_pt_is_enabled(unsigned long start, unsigned long end,
-					 struct zap_details *details)
-{
-	return false;
-}
-#endif /* CONFIG_PT_RECLAIM */
-
 void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm);
 int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm);
 
mm/memory.c +64 -6

99 lines not on this page · raw message

mm/pt_reclaim.c deleted +0 -72

72 lines not on this page · raw message

Signature · 9 lines
2.52.0


Completely untested, of course.

-- 
Cheers

David
author
#16 lore raw

Re: [PATCH v3 7/7] mm: make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE

Quoted 51 lines · David Hildenbrand (Red Hat)
Quoted 39 lines · Qi Zheng

From: Qi Zheng <[email protected]>

The PT_RECLAIM can work on all architectures that support MMU_GATHER_RCU_TABLE_FREE, so make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE.

BTW, change PT_RECLAIM to be enabled by default, since nobody should want to turn it off.

---
  arch/x86/Kconfig | 1 -
  mm/Kconfig       | 9 ++-------

  2 files changed, 2 insertions(+), 8 deletions(-)
arch/x86/Kconfig +0 -0
index 80527299f859a..0d22da56a71b0 100644
@@ -331,7 +331,6 @@ config X86

      select FUNCTION_ALIGNMENT_4B
      imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI select HAVE_DYNAMIC_FTRACE_NO_PATCHABLE -    select ARCH_SUPPORTS_PT_RECLAIM        if X86_64 select ARCH_SUPPORTS_SCHED_SMT        if SMP select SCHED_SMT            if SMP select ARCH_SUPPORTS_SCHED_CLUSTER    if SMP

mm/Kconfig +0 -0
index bd0ea5454af82..fc00b429b7129 100644
@@ -1447,14 +1447,9 @@ config ARCH_HAS_USER_SHADOW_STACK

        The architecture has hardware support for userspace shadow call stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss).

mm/Kconfig +0 -3
-config ARCH_SUPPORTS_PT_RECLAIM
-    def_bool n
-

  config PT_RECLAIM
-    bool "reclaim empty user page table pages"

Quoted 8 lines

-    default y
-    depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP
-    select MMU_GATHER_RCU_TABLE_FREE
+    def_bool y
+    depends on MMU_GATHER_RCU_TABLE_FREE
      help
        Try to reclaim empty user page table pages in paths other than munmap and exit_mmap path.

This patch seems to make s390x compilations sometimes unhappy:

Quoted 1 line

Unverified Warning (likely false positive, kindly check if interested):

I believe it is a false positive.

Quoted 33 lines

    mm/memory.c:1911 zap_pte_range() error: uninitialized symbol 'pmdval'.

Warning ids grouped by kconfigs:

recent_errors
`-- s390-randconfig-r072-20260117
    `-- mm-memory.c-zap_pte_range()-error:uninitialized-symbol-pmdval-.

I assume the compiler is not able to figure out that only when try_get_and_clear_pmd() returns false that pmdval could be uninitialized.

Maybe it has to do with LTO?

After all, that function resides in a different compilation unit.

Which makes me wonder whether we want to just move try_get_and_clear_pmd() and reclaim_pt_is_enabled() to internal.h or even just memory.c?

But then, maybe we could remove pt_reclaim.c completely and just have try_to_free_pte() in memory.c as well?

I would just do the following cleanup:

From cfe97092f71fcc88f729f07ee0bc6816e3e398f0 Mon Sep 17 00:00:00 2001
From: "David Hildenbrand (Red Hat)" <[email protected]>
Date: Sun, 18 Jan 2026 12:20:55 +0100
Subject: [PATCH] mm: move pte table reclaim code to memory.c

Let's move the code and clean it up a bit along the way.

Quoted 8 lines
---
  MAINTAINERS     |  1 -
  mm/internal.h   | 18 -------------
  mm/memory.c     | 70 ++++++++++++++++++++++++++++++++++++++++++-----
  mm/pt_reclaim.c | 72 -------------------------------------------------
  4 files changed, 64 insertions(+), 97 deletions(-)

 delete mode 100644 mm/pt_reclaim.c

Make sense, and LGTM. The reason it was placed in mm/pt_reclaim.c before was because there would be other paths calling these functions in the future. However, it can be separated out or put into a header file when there are actually such callers.

would you be willing to send out an official patch?

Thanks,
Qi

Quoted 240 lines
MAINTAINERS +0 -1
index 11720728d92f2..28e8e28bca3e5 100644
@@ -16692,7 +16692,6 @@ R:    Shakeel Butt <[email protected]>
  R:    Lorenzo Stoakes <[email protected]>
  L:    [email protected]
  S:    Maintained
-F:    mm/pt_reclaim.c
  F:    mm/vmscan.c
  F:    mm/workingset.c
 
mm/internal.h +0 -12
index 9508dbaf47cd4..ef71a1d9991f2 100644
@@ -1745,24 +1745,6 @@ int walk_page_range_debug(struct mm_struct *mm, 
 unsigned long start,
                unsigned long end, const struct mm_walk_ops *ops,
                pgd_t *pgd, void *private);
 
-/* pt_reclaim.c */
-bool try_get_and_clear_pmd(struct mm_struct *mm, pmd_t *pmd, pmd_t 
 *pmdval);
-void free_pte(struct mm_struct *mm, unsigned long addr, struct 
 mmu_gather *tlb,
-          pmd_t pmdval);
-void try_to_free_pte(struct mm_struct *mm, pmd_t *pmd, unsigned long addr,
-             struct mmu_gather *tlb);
-
-#ifdef CONFIG_PT_RECLAIM
-bool reclaim_pt_is_enabled(unsigned long start, unsigned long end,
-               struct zap_details *details);
-#else
-static inline bool reclaim_pt_is_enabled(unsigned long start, unsigned 

long end,
-                     struct zap_details *details)

mm/internal.h +0 -5
-{
-    return false;
-}
-#endif /* CONFIG_PT_RECLAIM */
-
  void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm);
  int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm);
mm/memory.c +57 -0
index f2e9e05388743..a09226761a07f 100644
@@ -1824,11 +1824,68 @@ static inline int do_zap_pte_range(struct 
 mmu_gather *tlb,
      return nr;
  }
 
+static bool pte_table_reclaim_enabled(unsigned long start, unsigned 
 long end,
+        struct zap_details *details)
+{
+    if (!IS_ENABLED(CONFIG_PT_RECLAIM))
+        return false;
+    return details && details->reclaim_pt && (end - start >= PMD_SIZE);
+}
+
+static bool zap_empty_pte_table(struct mm_struct *mm, pmd_t *pmd, pmd_t 
 *pmdval)
+{
+    spinlock_t *pml = pmd_lockptr(mm, pmd);
+
+    if (!spin_trylock(pml))
+        return false;
+
+    *pmdval = pmdp_get_lockless(pmd);
+    pmd_clear(pmd);
+    spin_unlock(pml);
+
+    return true;
+}
+
+static bool zap_pte_table_if_empty(struct mm_struct *mm, pmd_t *pmd,
+        unsigned long addr, pmd_t *pmdval)
+{
+    spinlock_t *pml, *ptl = NULL;
+    pte_t *start_pte, *pte;
+    int i;
+
+    pml = pmd_lock(mm, pmd);
+    start_pte = pte_offset_map_rw_nolock(mm, pmd, addr, pmdval, &ptl);
+    if (!start_pte)
+        goto out_ptl;
+    if (ptl != pml)
+        spin_lock_nested(ptl, SINGLE_DEPTH_NESTING);
+
+    for (i = 0, pte = start_pte; i < PTRS_PER_PTE; i++, pte++) {
+        if (!pte_none(ptep_get(pte)))
+            goto out_ptl;
+    }
+    pte_unmap(start_pte);
+
+    pmd_clear(pmd);
+
+    if (ptl != pml)
+        spin_unlock(ptl);
+    spin_unlock(pml);
+    return true;
+out_ptl:
+    if (start_pte)
+        pte_unmap_unlock(start_pte, ptl);
+    if (ptl != pml)
+        spin_unlock(pml);
+    return false;
+}
+
  static unsigned long zap_pte_range(struct mmu_gather *tlb,
                  struct vm_area_struct *vma, pmd_t *pmd,
                  unsigned long addr, unsigned long end,
                  struct zap_details *details)
  {
+    bool can_reclaim_pt = pte_table_reclaim_enabled(addr, end, details);

     bool force_flush = false, force_break = false; struct mm_struct *mm = tlb->mm;
     int rss[NR_MM_COUNTERS];

mm/memory.c +1 -2
@@ -1837,7 +1894,6 @@ static unsigned long zap_pte_range(struct 
 mmu_gather *tlb,
      pte_t *pte;
      pmd_t pmdval;
      unsigned long start = addr;
-    bool can_reclaim_pt = reclaim_pt_is_enabled(start, end, details);
      bool direct_reclaim = true;
      int nr;
@@ -1878,7 +1934,7 @@ static unsigned long zap_pte_range(struct 
 mmu_gather *tlb,
       * from being repopulated by another thread.
       */
      if (can_reclaim_pt && direct_reclaim && addr == end)
-        direct_reclaim = try_get_and_clear_pmd(mm, pmd, &pmdval);
+        direct_reclaim = zap_empty_pte_table(mm, pmd, &pmdval);
 
      add_mm_rss_vec(mm, rss);

     lazy_mmu_mode_disable();

mm/memory.c +6 -4
@@ -1907,10 +1963,12 @@ static unsigned long zap_pte_range(struct 
 mmu_gather *tlb,
      }
 
      if (can_reclaim_pt) {
-        if (direct_reclaim)
-            free_pte(mm, start, tlb, pmdval);
-        else
-            try_to_free_pte(mm, pmd, start, tlb);
+        if (!direct_reclaim)
+            direct_reclaim = zap_pte_table_if_empty(mm, pmd, start, 
 &pmdval);
+        if (direct_reclaim) {
+            pte_free_tlb(tlb, pmd_pgtable(pmdval), addr);
+            mm_dec_nr_ptes(mm);
+        }
      }

     return addr;

mm/pt_reclaim.c deleted +0 -15
deleted file mode 100644
index 46771cfff8239..0000000000000
@@ -1,72 +0,0 @@
-// SPDX-License-Identifier: GPL-2.0
-#include <linux/hugetlb.h>
-#include <linux/pgalloc.h>
-
-#include <asm/tlb.h>
-
-#include "internal.h"
-
-bool reclaim_pt_is_enabled(unsigned long start, unsigned long end,
-               struct zap_details *details)
-{
-    return details && details->reclaim_pt && (end - start >= PMD_SIZE);
-}
-
-bool try_get_and_clear_pmd(struct mm_struct *mm, pmd_t *pmd, pmd_t 

*pmdval)

mm/pt_reclaim.c deleted +0 -14
-{
-    spinlock_t *pml = pmd_lockptr(mm, pmd);
-
-    if (!spin_trylock(pml))
-        return false;
-
-    *pmdval = pmdp_get_lockless(pmd);
-    pmd_clear(pmd);
-    spin_unlock(pml);
-
-    return true;
-}
-
-void free_pte(struct mm_struct *mm, unsigned long addr, struct 

mmu_gather *tlb,
-          pmd_t pmdval)

mm/pt_reclaim.c deleted +0 -42
-{
-    pte_free_tlb(tlb, pmd_pgtable(pmdval), addr);
-    mm_dec_nr_ptes(mm);
-}
-
-void try_to_free_pte(struct mm_struct *mm, pmd_t *pmd, unsigned long addr,
-             struct mmu_gather *tlb)
-{
-    pmd_t pmdval;
-    spinlock_t *pml, *ptl = NULL;
-    pte_t *start_pte, *pte;
-    int i;
-
-    pml = pmd_lock(mm, pmd);
-    start_pte = pte_offset_map_rw_nolock(mm, pmd, addr, &pmdval, &ptl);
-    if (!start_pte)
-        goto out_ptl;
-    if (ptl != pml)
-        spin_lock_nested(ptl, SINGLE_DEPTH_NESTING);
-
-    /* Check if it is empty PTE page */
-    for (i = 0, pte = start_pte; i < PTRS_PER_PTE; i++, pte++) {
-        if (!pte_none(ptep_get(pte)))
-            goto out_ptl;
-    }
-    pte_unmap(start_pte);
-
-    pmd_clear(pmd);
-
-    if (ptl != pml)
-        spin_unlock(ptl);
-    spin_unlock(pml);
-
-    free_pte(mm, addr, tlb, pmdval);
-
-    return;
-out_ptl:
-    if (start_pte)
-        pte_unmap_unlock(start_pte, ptl);
-    if (ptl != pml)
-        spin_unlock(pml);
-}
#17 lore raw

Re: [PATCH v3 7/7] mm: make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE

Quoted 101 lines · Qi Zheng
Quoted 52 lines · David Hildenbrand (Red Hat)
Quoted 46 lines · Qi Zheng

From: Qi Zheng <[email protected]>

The PT_RECLAIM can work on all architectures that support MMU_GATHER_RCU_TABLE_FREE, so make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE.

BTW, change PT_RECLAIM to be enabled by default, since nobody should want to turn it off.

---
   arch/x86/Kconfig | 1 -
   mm/Kconfig       | 9 ++-------
   2 files changed, 2 insertions(+), 8 deletions(-)
arch/x86/Kconfig +0 -1
index 80527299f859a..0d22da56a71b0 100644
@@ -331,7 +331,6 @@ config X86
       select FUNCTION_ALIGNMENT_4B
       imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI
       select HAVE_DYNAMIC_FTRACE_NO_PATCHABLE
-    select ARCH_SUPPORTS_PT_RECLAIM        if X86_64
       select ARCH_SUPPORTS_SCHED_SMT        if SMP
       select SCHED_SMT            if SMP
       select ARCH_SUPPORTS_SCHED_CLUSTER    if SMP
mm/Kconfig +2 -7
index bd0ea5454af82..fc00b429b7129 100644
@@ -1447,14 +1447,9 @@ config ARCH_HAS_USER_SHADOW_STACK
         The architecture has hardware support for userspace shadow call
             stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss).
-config ARCH_SUPPORTS_PT_RECLAIM
-    def_bool n
-
   config PT_RECLAIM
-    bool "reclaim empty user page table pages"
-    default y
-    depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP
-    select MMU_GATHER_RCU_TABLE_FREE
+    def_bool y
+    depends on MMU_GATHER_RCU_TABLE_FREE
       help
         Try to reclaim empty user page table pages in paths other than
Quoted 2 lines

munmap
        and exit_mmap path.

This patch seems to make s390x compilations sometimes unhappy:

Quoted 1 line

Unverified Warning (likely false positive, kindly check if interested):

I believe it is a false positive.

Quoted 33 lines

    mm/memory.c:1911 zap_pte_range() error: uninitialized symbol 'pmdval'.

Warning ids grouped by kconfigs:

recent_errors
`-- s390-randconfig-r072-20260117
    `-- mm-memory.c-zap_pte_range()-error:uninitialized-symbol-pmdval-.

I assume the compiler is not able to figure out that only when try_get_and_clear_pmd() returns false that pmdval could be uninitialized.

Maybe it has to do with LTO?

After all, that function resides in a different compilation unit.

Which makes me wonder whether we want to just move try_get_and_clear_pmd() and reclaim_pt_is_enabled() to internal.h or even just memory.c?

But then, maybe we could remove pt_reclaim.c completely and just have try_to_free_pte() in memory.c as well?

I would just do the following cleanup:

From cfe97092f71fcc88f729f07ee0bc6816e3e398f0 Mon Sep 17 00:00:00 2001
From: "David Hildenbrand (Red Hat)" <[email protected]>
Date: Sun, 18 Jan 2026 12:20:55 +0100
Subject: [PATCH] mm: move pte table reclaim code to memory.c

Let's move the code and clean it up a bit along the way.

Quoted 8 lines
---
   MAINTAINERS     |  1 -
   mm/internal.h   | 18 -------------
   mm/memory.c     | 70 ++++++++++++++++++++++++++++++++++++++++++-----
   mm/pt_reclaim.c | 72 -------------------------------------------------
   4 files changed, 64 insertions(+), 97 deletions(-)

 delete mode 100644 mm/pt_reclaim.c

Quoted 4 lines

Make sense, and LGTM. The reason it was placed in mm/pt_reclaim.c before was because there would be other paths calling these functions in the future. However, it can be separated out or put into a header file when there are actually such callers.

Most relevant zapping better happens in memory.c :)

There is, of course, zapping due to RMAP unmap, but that mostly targets individual PTEs, and not a complete pte table.

Likely, if ever required, we should expose a proper zapping interface from memory.c to other users, assuming the existing one is not suitable.

Quoted 2 lines

would you be willing to send out an official patch?

Yes, I can send one out, thanks.

Signature · 3 lines
Cheers

David
maintainer Acked-by on v3 7/7
#18 lore raw

Re: [PATCH v3 7/7] mm: make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE

In reply to Qi Zheng
Quoted 8 lines · Qi Zheng

From: Qi Zheng <[email protected]>

The PT_RECLAIM can work on all architectures that support MMU_GATHER_RCU_TABLE_FREE, so make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE.

BTW, change PT_RECLAIM to be enabled by default, since nobody should want to turn it off.

Right, and if there is ever a need to, I wonder whether that should be a boottime/runtime toggle instead.

So far we haven't heard of any relevant runtime overheads that causes problems.

Quoted 30 lines
---
  arch/x86/Kconfig | 1 -
  mm/Kconfig       | 9 ++-------
  2 files changed, 2 insertions(+), 8 deletions(-)
arch/x86/Kconfig +0 -1
index 80527299f859a..0d22da56a71b0 100644
@@ -331,7 +331,6 @@ config X86
  	select FUNCTION_ALIGNMENT_4B
  	imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI
  	select HAVE_DYNAMIC_FTRACE_NO_PATCHABLE
-	select ARCH_SUPPORTS_PT_RECLAIM		if X86_64
  	select ARCH_SUPPORTS_SCHED_SMT		if SMP
  	select SCHED_SMT			if SMP
  	select ARCH_SUPPORTS_SCHED_CLUSTER	if SMP
mm/Kconfig +0 -3
index bd0ea5454af82..fc00b429b7129 100644
@@ -1447,14 +1447,9 @@ config ARCH_HAS_USER_SHADOW_STACK
  	  The architecture has hardware support for userspace shadow call
            stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss).
 
-config ARCH_SUPPORTS_PT_RECLAIM
-	def_bool n
-

mm/Kconfig config ARCH_HAS_USER_SHADOW_STACK · line 1455

Quoted 10 lines
mm/Kconfig +2 -4
  config PT_RECLAIM
-	bool "reclaim empty user page table pages"
-	default y
-	depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP
-	select MMU_GATHER_RCU_TABLE_FREE
+	def_bool y
+	depends on MMU_GATHER_RCU_TABLE_FREE
  	help
  	  Try to reclaim empty user page table pages in paths other than munmap
  	  and exit_mmap path.

Nothing jumped at me. Hopefully we're not missing something important :)

Signature · 3 lines
Cheers

David
#19 lore raw

Re: [PATCH v3 7/7] mm: make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE

In reply to Qi Zheng
Quoted 38 lines · Qi Zheng

From: Qi Zheng <[email protected]>

The PT_RECLAIM can work on all architectures that support MMU_GATHER_RCU_TABLE_FREE, so make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE.

BTW, change PT_RECLAIM to be enabled by default, since nobody should want to turn it off.

---
 arch/x86/Kconfig | 1 -
 mm/Kconfig       | 9 ++-------
 2 files changed, 2 insertions(+), 8 deletions(-)
arch/x86/Kconfig +0 -1
index 80527299f859a..0d22da56a71b0 100644
@@ -331,7 +331,6 @@ config X86
 	select FUNCTION_ALIGNMENT_4B
 	imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI
 	select HAVE_DYNAMIC_FTRACE_NO_PATCHABLE
-	select ARCH_SUPPORTS_PT_RECLAIM		if X86_64
 	select ARCH_SUPPORTS_SCHED_SMT		if SMP
 	select SCHED_SMT			if SMP
 	select ARCH_SUPPORTS_SCHED_CLUSTER	if SMP
mm/Kconfig +0 -3
index bd0ea5454af82..fc00b429b7129 100644
@@ -1447,14 +1447,9 @@ config ARCH_HAS_USER_SHADOW_STACK
 	  The architecture has hardware support for userspace shadow call
           stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss).
 
-config ARCH_SUPPORTS_PT_RECLAIM
-	def_bool n
-

mm/Kconfig config ARCH_HAS_USER_SHADOW_STACK · line 1455

Quoted 10 lines
mm/Kconfig +2 -4
 config PT_RECLAIM
-	bool "reclaim empty user page table pages"
-	default y
-	depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP
-	select MMU_GATHER_RCU_TABLE_FREE
+	def_bool y
+	depends on MMU_GATHER_RCU_TABLE_FREE
 	help
 	  Try to reclaim empty user page table pages in paths other than munmap
 	  and exit_mmap path.

Hi,

This patch unfortunately results in a WARN_ON_ONCE and unaligned accesses on sparc64:

$ stress-ng --mmaphuge 20 -t 60
stress-ng: info:  [559] setting to a 1 min run per stressor
stress-ng: info:  [559] dispatching hogs: 20 mmaphuge
[  560.592569] ------------[ cut here ]------------
[  560.592663] WARNING: kernel/rcu/tree.c:3098 at __call_rcu_common.constprop.0+0x200/0x760, CPU#4: stress-ng-mmaph/568
[  560.592777] CPU: 4 UID: 1000 PID: 568 Comm: stress-ng-mmaph Not tainted 6.19.0-rc5-00127-g62fc9f6ccb97 #8 VOLUNTARY
[  560.592805] Call Trace:
[  560.592812] [<00000000004368b8>] dump_stack+0x8/0x60
[  560.592844] [<0000000000482a60>] __warn+0xe0/0x140
[  560.592878] [<0000000000482b64>] warn_slowpath_fmt+0xa4/0x120
[  560.592901] [<0000000000526a40>] __call_rcu_common.constprop.0+0x200/0x760
[  560.592931] [<0000000000526fd0>] call_rcu+0x10/0x20
[  560.592954] [<0000000000730838>] tlb_remove_table+0x98/0xc0
[  560.592986] [<000000000071bec4>] free_pgd_range+0x224/0x4c0
[  560.593021] [<000000000071c35c>] free_pgtables+0x1fc/0x240
[  560.593042] [<000000000074a6f0>] vms_clear_ptes+0x110/0x140
[  560.593068] [<000000000074c3dc>] vms_complete_munmap_vmas+0x5c/0x280
[  560.593094] [<000000000074de5c>] do_vmi_align_munmap+0x1dc/0x260
[  560.593117] [<000000000074df80>] do_vmi_munmap+0xa0/0x140
[  560.593142] [<000000000074fb2c>] __vm_munmap+0x8c/0x160
[  560.593168] [<000000000072cfd4>] vm_munmap+0x14/0x40
[  560.593190] [<00000000004402a8>] sys_64_munmap+0x88/0xa0
[  560.593221] [<0000000000406274>] linux_sparc_syscall+0x34/0x44
[  560.593274] ---[ end trace 0000000000000000 ]---
[  560.593960] log_unaligned: 209 callbacks suppressed
[  560.593979] Kernel unaligned access at TPC[526a4c] __call_rcu_common.constprop.0+0x20c/0x760
[  560.594121] Kernel unaligned access at TPC[526864] __call_rcu_common.constprop.0+0x24/0x760
[  560.594198] Kernel unaligned access at TPC[52b3c4] rcu_segcblist_enqueue+0x24/0x40
[  560.594275] Kernel unaligned access at TPC[526860] __call_rcu_common.constprop.0+0x20/0x760
[  560.594360] Kernel unaligned access at TPC[526864] __call_rcu_common.constprop.0+0x24/0x760
[  567.054127] log_unaligned: 1105 callbacks suppressed
[  567.054167] Kernel unaligned access at TPC[526860] __call_rcu_common.constprop.0+0x20/0x760
[  567.054331] Kernel unaligned access at TPC[526864] __call_rcu_common.constprop.0+0x24/0x760
[  567.054410] Kernel unaligned access at TPC[52b3c4] rcu_segcblist_enqueue+0x24/0x40
...

I bisected to this one on mm-unstable from approximately 2026-01-12.

The warning is from

	/* Misaligned rcu_head! */
	WARN_ON_ONCE((unsigned long)head & (sizeof(void *) - 1));

in __call_rcu_common() and the unaligned accesses follows from there.

Regards,
Andreas

author
#20 lore raw

Re: [PATCH v3 7/7] mm: make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE

Quoted 81 lines · Andreas Larsson
Quoted 38 lines · Qi Zheng

From: Qi Zheng <[email protected]>

The PT_RECLAIM can work on all architectures that support MMU_GATHER_RCU_TABLE_FREE, so make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE.

BTW, change PT_RECLAIM to be enabled by default, since nobody should want to turn it off.

---
  arch/x86/Kconfig | 1 -
  mm/Kconfig       | 9 ++-------
  2 files changed, 2 insertions(+), 8 deletions(-)
arch/x86/Kconfig +0 -1
index 80527299f859a..0d22da56a71b0 100644
@@ -331,7 +331,6 @@ config X86
  	select FUNCTION_ALIGNMENT_4B
  	imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI
  	select HAVE_DYNAMIC_FTRACE_NO_PATCHABLE
-	select ARCH_SUPPORTS_PT_RECLAIM		if X86_64
  	select ARCH_SUPPORTS_SCHED_SMT		if SMP
  	select SCHED_SMT			if SMP
  	select ARCH_SUPPORTS_SCHED_CLUSTER	if SMP
mm/Kconfig +0 -3
index bd0ea5454af82..fc00b429b7129 100644
@@ -1447,14 +1447,9 @@ config ARCH_HAS_USER_SHADOW_STACK
  	  The architecture has hardware support for userspace shadow call
            stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss).
 
-config ARCH_SUPPORTS_PT_RECLAIM
-	def_bool n
-

mm/Kconfig config ARCH_HAS_USER_SHADOW_STACK · line 1455

Quoted 10 lines
mm/Kconfig +2 -4
  config PT_RECLAIM
-	bool "reclaim empty user page table pages"
-	default y
-	depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP
-	select MMU_GATHER_RCU_TABLE_FREE
+	def_bool y
+	depends on MMU_GATHER_RCU_TABLE_FREE
  	help
  	  Try to reclaim empty user page table pages in paths other than munmap
  	  and exit_mmap path.

Hi,

This patch unfortunately results in a WARN_ON_ONCE and unaligned accesses on sparc64:

$ stress-ng --mmaphuge 20 -t 60
stress-ng: info:  [559] setting to a 1 min run per stressor
stress-ng: info:  [559] dispatching hogs: 20 mmaphuge
[  560.592569] ------------[ cut here ]------------
[  560.592663] WARNING: kernel/rcu/tree.c:3098 at __call_rcu_common.constprop.0+0x200/0x760, CPU#4: stress-ng-mmaph/568
[  560.592777] CPU: 4 UID: 1000 PID: 568 Comm: stress-ng-mmaph Not tainted 6.19.0-rc5-00127-g62fc9f6ccb97 #8 VOLUNTARY
[  560.592805] Call Trace:
[  560.592812] [<00000000004368b8>] dump_stack+0x8/0x60
[  560.592844] [<0000000000482a60>] __warn+0xe0/0x140
[  560.592878] [<0000000000482b64>] warn_slowpath_fmt+0xa4/0x120
[  560.592901] [<0000000000526a40>] __call_rcu_common.constprop.0+0x200/0x760
[  560.592931] [<0000000000526fd0>] call_rcu+0x10/0x20
[  560.592954] [<0000000000730838>] tlb_remove_table+0x98/0xc0
[  560.592986] [<000000000071bec4>] free_pgd_range+0x224/0x4c0
[  560.593021] [<000000000071c35c>] free_pgtables+0x1fc/0x240
[  560.593042] [<000000000074a6f0>] vms_clear_ptes+0x110/0x140
[  560.593068] [<000000000074c3dc>] vms_complete_munmap_vmas+0x5c/0x280
[  560.593094] [<000000000074de5c>] do_vmi_align_munmap+0x1dc/0x260
[  560.593117] [<000000000074df80>] do_vmi_munmap+0xa0/0x140
[  560.593142] [<000000000074fb2c>] __vm_munmap+0x8c/0x160
[  560.593168] [<000000000072cfd4>] vm_munmap+0x14/0x40
[  560.593190] [<00000000004402a8>] sys_64_munmap+0x88/0xa0
[  560.593221] [<0000000000406274>] linux_sparc_syscall+0x34/0x44
[  560.593274] ---[ end trace 0000000000000000 ]---
[  560.593960] log_unaligned: 209 callbacks suppressed
[  560.593979] Kernel unaligned access at TPC[526a4c] __call_rcu_common.constprop.0+0x20c/0x760
Quoted 8 lines
[  560.594121] Kernel unaligned access at TPC[526864] __call_rcu_common.constprop.0+0x24/0x760
[  560.594198] Kernel unaligned access at TPC[52b3c4] rcu_segcblist_enqueue+0x24/0x40
[  560.594275] Kernel unaligned access at TPC[526860] __call_rcu_common.constprop.0+0x20/0x760
[  560.594360] Kernel unaligned access at TPC[526864] __call_rcu_common.constprop.0+0x24/0x760
[  567.054127] log_unaligned: 1105 callbacks suppressed
[  567.054167] Kernel unaligned access at TPC[526860] __call_rcu_common.constprop.0+0x20/0x760
[  567.054331] Kernel unaligned access at TPC[526864] __call_rcu_common.constprop.0+0x24/0x760
[  567.054410] Kernel unaligned access at TPC[52b3c4] rcu_segcblist_enqueue+0x24/0x40

Thanks for your report!

On sparc64, pmd and pud levels are not of struct page:

__pmd_free_tlb/__pud_free_tlb
--> pgtable_free_tlb(tlb, pud/pmd, false). <=== is_page == false

    --> tlb_remove_table

So in __tlb_remove_table_one(), the table cannot be treated as ptdesc because it does not have an pt_rcu_head member.

Hi David, it seems we still need to keep ARCH_SUPPORTS_PT_RECLAIM?

Thanks,
Qi

Quoted 14 lines

...

I bisected to this one on mm-unstable from approximately 2026-01-12.

The warning is from

	/* Misaligned rcu_head! */
	WARN_ON_ONCE((unsigned long)head & (sizeof(void *) - 1));

in __call_rcu_common() and the unaligned accesses follows from there.

Regards,
Andreas

#21 lore raw

Re: [PATCH v3 7/7] mm: make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE

Quoted 93 lines · Qi Zheng
Quoted 81 lines · Andreas Larsson
Quoted 38 lines · Qi Zheng

From: Qi Zheng <[email protected]>

The PT_RECLAIM can work on all architectures that support MMU_GATHER_RCU_TABLE_FREE, so make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE.

BTW, change PT_RECLAIM to be enabled by default, since nobody should want to turn it off.

---
   arch/x86/Kconfig | 1 -
   mm/Kconfig       | 9 ++-------
   2 files changed, 2 insertions(+), 8 deletions(-)
arch/x86/Kconfig +0 -1
index 80527299f859a..0d22da56a71b0 100644
@@ -331,7 +331,6 @@ config X86
   	select FUNCTION_ALIGNMENT_4B
   	imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI
   	select HAVE_DYNAMIC_FTRACE_NO_PATCHABLE
-	select ARCH_SUPPORTS_PT_RECLAIM		if X86_64
   	select ARCH_SUPPORTS_SCHED_SMT		if SMP
   	select SCHED_SMT			if SMP
   	select ARCH_SUPPORTS_SCHED_CLUSTER	if SMP
mm/Kconfig +0 -3
index bd0ea5454af82..fc00b429b7129 100644
@@ -1447,14 +1447,9 @@ config ARCH_HAS_USER_SHADOW_STACK
   	  The architecture has hardware support for userspace shadow call
             stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss).
 
-config ARCH_SUPPORTS_PT_RECLAIM
-	def_bool n
-

mm/Kconfig config ARCH_HAS_USER_SHADOW_STACK · line 1455

Quoted 10 lines
mm/Kconfig +2 -4
   config PT_RECLAIM
-	bool "reclaim empty user page table pages"
-	default y
-	depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP
-	select MMU_GATHER_RCU_TABLE_FREE
+	def_bool y
+	depends on MMU_GATHER_RCU_TABLE_FREE
   	help
   	  Try to reclaim empty user page table pages in paths other than munmap
   	  and exit_mmap path.

Hi,

This patch unfortunately results in a WARN_ON_ONCE and unaligned accesses on sparc64:

$ stress-ng --mmaphuge 20 -t 60
stress-ng: info:  [559] setting to a 1 min run per stressor
stress-ng: info:  [559] dispatching hogs: 20 mmaphuge
[  560.592569] ------------[ cut here ]------------
[  560.592663] WARNING: kernel/rcu/tree.c:3098 at __call_rcu_common.constprop.0+0x200/0x760, CPU#4: stress-ng-mmaph/568
[  560.592777] CPU: 4 UID: 1000 PID: 568 Comm: stress-ng-mmaph Not tainted 6.19.0-rc5-00127-g62fc9f6ccb97 #8 VOLUNTARY
[  560.592805] Call Trace:
[  560.592812] [<00000000004368b8>] dump_stack+0x8/0x60
[  560.592844] [<0000000000482a60>] __warn+0xe0/0x140
[  560.592878] [<0000000000482b64>] warn_slowpath_fmt+0xa4/0x120
[  560.592901] [<0000000000526a40>] __call_rcu_common.constprop.0+0x200/0x760
[  560.592931] [<0000000000526fd0>] call_rcu+0x10/0x20
[  560.592954] [<0000000000730838>] tlb_remove_table+0x98/0xc0
[  560.592986] [<000000000071bec4>] free_pgd_range+0x224/0x4c0
[  560.593021] [<000000000071c35c>] free_pgtables+0x1fc/0x240
[  560.593042] [<000000000074a6f0>] vms_clear_ptes+0x110/0x140
[  560.593068] [<000000000074c3dc>] vms_complete_munmap_vmas+0x5c/0x280
[  560.593094] [<000000000074de5c>] do_vmi_align_munmap+0x1dc/0x260
[  560.593117] [<000000000074df80>] do_vmi_munmap+0xa0/0x140
[  560.593142] [<000000000074fb2c>] __vm_munmap+0x8c/0x160
[  560.593168] [<000000000072cfd4>] vm_munmap+0x14/0x40
[  560.593190] [<00000000004402a8>] sys_64_munmap+0x88/0xa0
[  560.593221] [<0000000000406274>] linux_sparc_syscall+0x34/0x44
[  560.593274] ---[ end trace 0000000000000000 ]---
[  560.593960] log_unaligned: 209 callbacks suppressed
[  560.593979] Kernel unaligned access at TPC[526a4c] __call_rcu_common.constprop.0+0x20c/0x760
Quoted 8 lines
[  560.594121] Kernel unaligned access at TPC[526864] __call_rcu_common.constprop.0+0x24/0x760
[  560.594198] Kernel unaligned access at TPC[52b3c4] rcu_segcblist_enqueue+0x24/0x40
[  560.594275] Kernel unaligned access at TPC[526860] __call_rcu_common.constprop.0+0x20/0x760
[  560.594360] Kernel unaligned access at TPC[526864] __call_rcu_common.constprop.0+0x24/0x760
[  567.054127] log_unaligned: 1105 callbacks suppressed
[  567.054167] Kernel unaligned access at TPC[526860] __call_rcu_common.constprop.0+0x20/0x760
[  567.054331] Kernel unaligned access at TPC[526864] __call_rcu_common.constprop.0+0x24/0x760
[  567.054410] Kernel unaligned access at TPC[52b3c4] rcu_segcblist_enqueue+0x24/0x40
Quoted 3 lines

Thanks for your report!

On sparc64, pmd and pud levels are not of struct page:

Can you elaborate, I don't understand what you mean :)

Is it also a problem on architectures like s390x and ppc, where we squeeze multiple page tables into a physical pages?

Quoted 9 lines

__pmd_free_tlb/__pud_free_tlb
--> pgtable_free_tlb(tlb, pud/pmd, false). <=== is_page == false

      --> tlb_remove_table

So in __tlb_remove_table_one(), the table cannot be treated as ptdesc because it does not have an pt_rcu_head member.

Hi David, it seems we still need to keep ARCH_SUPPORTS_PT_RECLAIM?

Or we invert it and only disable it for the known-problematic architectures?

Signature · 3 lines
Cheers

David
author
#22 lore raw

Re: [PATCH v3 7/7] mm: make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE

Quoted 93 lines · David Hildenbrand (Red Hat)
Quoted 92 lines · Qi Zheng
Quoted 88 lines · Andreas Larsson
Quoted 39 lines · Qi Zheng

From: Qi Zheng <[email protected]>

The PT_RECLAIM can work on all architectures that support MMU_GATHER_RCU_TABLE_FREE, so make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE.

BTW, change PT_RECLAIM to be enabled by default, since nobody should want to turn it off.

---
   arch/x86/Kconfig | 1 -
   mm/Kconfig       | 9 ++-------

   2 files changed, 2 insertions(+), 8 deletions(-)
arch/x86/Kconfig +0 -0
index 80527299f859a..0d22da56a71b0 100644
@@ -331,7 +331,6 @@ config X86

       select FUNCTION_ALIGNMENT_4B
       imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI select HAVE_DYNAMIC_FTRACE_NO_PATCHABLE -    select ARCH_SUPPORTS_PT_RECLAIM        if X86_64 select ARCH_SUPPORTS_SCHED_SMT        if SMP select SCHED_SMT            if SMP select ARCH_SUPPORTS_SCHED_CLUSTER    if SMP

mm/Kconfig +0 -0
index bd0ea5454af82..fc00b429b7129 100644
@@ -1447,14 +1447,9 @@ config ARCH_HAS_USER_SHADOW_STACK

         The architecture has hardware support for userspace shadow call stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss).

mm/Kconfig +0 -3
-config ARCH_SUPPORTS_PT_RECLAIM
-    def_bool n
-

   config PT_RECLAIM
-    bool "reclaim empty user page table pages"

Quoted 8 lines

-    default y
-    depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP
-    select MMU_GATHER_RCU_TABLE_FREE
+    def_bool y
+    depends on MMU_GATHER_RCU_TABLE_FREE
       help
         Try to reclaim empty user page table pages in paths other than munmap and exit_mmap path.

Hi,

This patch unfortunately results in a WARN_ON_ONCE and unaligned accesses on sparc64:

$ stress-ng --mmaphuge 20 -t 60
stress-ng: info:  [559] setting to a 1 min run per stressor
stress-ng: info:  [559] dispatching hogs: 20 mmaphuge
[  560.592569] ------------[ cut here ]------------
[  560.592663] WARNING: kernel/rcu/tree.c:3098 at __call_rcu_common.constprop.0+0x200/0x760, CPU#4: stress-ng-mmaph/568
[  560.592777] CPU: 4 UID: 1000 PID: 568 Comm: stress-ng-mmaph Not tainted 6.19.0-rc5-00127-g62fc9f6ccb97 #8 VOLUNTARY
[  560.592805] Call Trace:
[  560.592812] [<00000000004368b8>] dump_stack+0x8/0x60
[  560.592844] [<0000000000482a60>] __warn+0xe0/0x140
[  560.592878] [<0000000000482b64>] warn_slowpath_fmt+0xa4/0x120
[  560.592901] [<0000000000526a40>] __call_rcu_common.constprop.0+0x200/0x760
[  560.592931] [<0000000000526fd0>] call_rcu+0x10/0x20
[  560.592954] [<0000000000730838>] tlb_remove_table+0x98/0xc0
[  560.592986] [<000000000071bec4>] free_pgd_range+0x224/0x4c0
[  560.593021] [<000000000071c35c>] free_pgtables+0x1fc/0x240
[  560.593042] [<000000000074a6f0>] vms_clear_ptes+0x110/0x140
[  560.593068] [<000000000074c3dc>] vms_complete_munmap_vmas+0x5c/0x280
[  560.593094] [<000000000074de5c>] do_vmi_align_munmap+0x1dc/0x260
[  560.593117] [<000000000074df80>] do_vmi_munmap+0xa0/0x140
[  560.593142] [<000000000074fb2c>] __vm_munmap+0x8c/0x160
[  560.593168] [<000000000072cfd4>] vm_munmap+0x14/0x40
[  560.593190] [<00000000004402a8>] sys_64_munmap+0x88/0xa0
[  560.593221] [<0000000000406274>] linux_sparc_syscall+0x34/0x44
[  560.593274] ---[ end trace 0000000000000000 ]---
[  560.593960] log_unaligned: 209 callbacks suppressed
[  560.593979] Kernel unaligned access at TPC[526a4c] __call_rcu_common.constprop.0+0x20c/0x760
[  560.594121] Kernel unaligned access at TPC[526864] __call_rcu_common.constprop.0+0x24/0x760
[  560.594198] Kernel unaligned access at TPC[52b3c4] rcu_segcblist_enqueue+0x24/0x40
[  560.594275] Kernel unaligned access at TPC[526860] __call_rcu_common.constprop.0+0x20/0x760
[  560.594360] Kernel unaligned access at TPC[526864] __call_rcu_common.constprop.0+0x24/0x760
[  567.054127] log_unaligned: 1105 callbacks suppressed
[  567.054167] Kernel unaligned access at TPC[526860] __call_rcu_common.constprop.0+0x20/0x760
[  567.054331] Kernel unaligned access at TPC[526864] __call_rcu_common.constprop.0+0x24/0x760
[  567.054410] Kernel unaligned access at TPC[52b3c4] rcu_segcblist_enqueue+0x24/0x40
Quoted 5 lines
Quoted 3 lines

Thanks for your report!

On sparc64, pmd and pud levels are not of struct page:

Can you elaborate, I don't understand what you mean :)

On sparc64:

static inline void pgtable_free_tlb(struct mmu_gather *tlb, void *table, bool is_page)

{
	unsigned long pgf = (unsigned long)table;
	if (is_page)
		pgf |= 0x1UL;
	tlb_remove_table(tlb, (void *)pgf);
}

static inline void __tlb_remove_table(void *_table)

{
	void *table = (void *)((unsigned long)_table & ~0x1UL);
	bool is_page = false;

	if ((unsigned long)_table & 0x1UL)
		is_page = true;
	pgtable_free(table, is_page);
}

void pgtable_free(void *table, bool is_page)

{
	if (is_page)
		__pte_free(table);
	else
		kmem_cache_free(pgtable_cache, table);
}

For pmd and pud levels, is_page is false, so we can not do the following in __tlb_remove_table_one().

```
	ptdesc = table;
	call_rcu(&ptdesc->pt_rcu_head, __tlb_remove_table_one_rcu);
```
Quoted 2 lines

Is it also a problem on architectures like s390x and ppc, where we squeeze multiple page tables into a physical pages?

For ppc, it's the same as for sparc64.

For s390x, it supports MMU_GATHER_RCU_TABLE_FREE and define its own pxx_free_tlb(), but these all call tlb_remove_ptdesc(), so there is no problem.

Quoted 6 lines
Quoted 6 lines

__pmd_free_tlb/__pud_free_tlb
--> pgtable_free_tlb(tlb, pud/pmd, false). <=== is_page == false --> tlb_remove_table

Quoted 6 lines
Quoted 4 lines

So in __tlb_remove_table_one(), the table cannot be treated as ptdesc because it does not have an pt_rcu_head member.

Hi David, it seems we still need to keep ARCH_SUPPORTS_PT_RECLAIM?

Or we invert it and only disable it for the known-problematic architectures?

Yes, the problem lies with those architectures that support MMU_GATHER_RCU_TABLE_FREE and define their own _tlb_remove_table().

So my plan is as follows:

1. convert __HAVE_ARCH_TLB_REMOVE_TABLE to CONFIG_HAVE_ARCH_TLB_REMOVE_TABLE config
2. make PT_RECLAIM depends on MMU_GATHER_RCU_TABLE_FREE && !HAVE_ARCH_TLB_REMOVE_TABLE

I'll send v4 soon.

Thanks,
Qi

Quoted 1 line