[PATCH 0/7] enable PT_RECLAIM on all 64-bit architectures

[PATCH 0/7] enable PT_RECLAIM on all 64-bit architectures

[Qi Zheng]

From: Qi Zheng <zhengqi.arch@bytedance.com>

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
&& 64BIT. This way, PT_RECLAIM can be enabled by default on all 64-bit
architectures.

Comments and suggestions are welcome!

Thanks,
Qi

Qi Zheng (7):
  alpha: mm: enable MMU_GATHER_RCU_TABLE_FREE
  arc: 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 depend on MMU_GATHER_RCU_TABLE_FREE && 64BIT

 arch/alpha/Kconfig                   | 1 +
 arch/alpha/include/asm/tlb.h         | 8 +++++---
 arch/arc/Kconfig                     | 1 +
 arch/arc/include/asm/pgalloc.h       | 9 ++++++---
 arch/loongarch/Kconfig               | 1 +
 arch/loongarch/include/asm/pgalloc.h | 6 ++++--
 arch/mips/Kconfig                    | 1 +
 arch/mips/include/asm/pgalloc.h      | 6 ++++--
 arch/parisc/Kconfig                  | 1 +
 arch/parisc/include/asm/tlb.h        | 6 ++++--
 arch/um/Kconfig                      | 1 +
 arch/x86/Kconfig                     | 1 -
 mm/Kconfig                           | 6 +-----
 13 files changed, 30 insertions(+), 18 deletions(-)

-- 
2.20.1

[PATCH 1/7] alpha: mm: enable MMU_GATHER_RCU_TABLE_FREE

[Qi Zheng]

From: Qi Zheng <zhengqi.arch@bytedance.com>

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.

Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Richard Henderson <richard.henderson@linaro.org>
Cc: Matt Turner <mattst88@gmail.com>
---
 arch/alpha/Kconfig           | 1 +
 arch/alpha/include/asm/tlb.h | 8 +++++---
 2 files changed, 6 insertions(+), 3 deletions(-)

diff --git a/arch/alpha/Kconfig b/arch/alpha/Kconfig
index 80367f2cf821c..681ed894d9e72 100644
--- a/arch/alpha/Kconfig
+++ b/arch/alpha/Kconfig
@@ -40,6 +40,7 @@ config ALPHA
 	select MMU_GATHER_NO_RANGE
 	select SPARSEMEM_EXTREME if SPARSEMEM
 	select ZONE_DMA
+	select MMU_GATHER_RCU_TABLE_FREE
 	help
 	  The Alpha is a 64-bit general-purpose processor designed and
 	  marketed by the Digital Equipment Corporation of blessed memory,
diff --git a/arch/alpha/include/asm/tlb.h b/arch/alpha/include/asm/tlb.h
index 4f79e331af5ea..4fe5a901720f0 100644
--- a/arch/alpha/include/asm/tlb.h
+++ b/arch/alpha/include/asm/tlb.h
@@ -4,7 +4,9 @@
 
 #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
-- 
2.20.1

Re: [PATCH 1/7] alpha: mm: enable MMU_GATHER_RCU_TABLE_FREE

[Magnus Lindholm]

Hi,

I applied your patches to a fresh pull of torvalds/linux.git repo but was unable
to build the kernel (on Alpha) with this patch applied.

I made the following changes in order to get it to build on Alpha:

diff --git a/mm/pt_reclaim.c b/mm/pt_reclaim.c
index 7e9455a18aae..6761b0c282bf 100644
--- a/mm/pt_reclaim.c
+++ b/mm/pt_reclaim.c
@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: GPL-2.0
 #include <linux/hugetlb.h>
-#include <asm-generic/tlb.h>
 #include <asm/pgalloc.h>
+#include <asm/tlb.h>

 #include "internal.h"


/Magnus

Re: [PATCH 1/7] alpha: mm: enable MMU_GATHER_RCU_TABLE_FREE

[Qi Zheng]

Hi Magnus,

On 11/15/25 3:13 AM, Magnus Lindholm wrote:
> Hi,
> 
> I applied your patches to a fresh pull of torvalds/linux.git repo but was unable
> to build the kernel (on Alpha) with this patch applied.
> 
> I made the following changes in order to get it to build on Alpha:

Thanks! Will fix it in the next version.

> 
> diff --git a/mm/pt_reclaim.c b/mm/pt_reclaim.c
> index 7e9455a18aae..6761b0c282bf 100644
> --- a/mm/pt_reclaim.c
> +++ b/mm/pt_reclaim.c
> @@ -1,7 +1,7 @@
>   // SPDX-License-Identifier: GPL-2.0
>   #include <linux/hugetlb.h>
> -#include <asm-generic/tlb.h>
>   #include <asm/pgalloc.h>
> +#include <asm/tlb.h>
> 
>   #include "internal.h"
> 
> 
> /Magnus

[PATCH 2/7] arc: mm: enable MMU_GATHER_RCU_TABLE_FREE

[Qi Zheng]

From: Qi Zheng <zhengqi.arch@bytedance.com>

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.

Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Vineet Gupta <vgupta@kernel.org>
---
 arch/arc/Kconfig               | 1 +
 arch/arc/include/asm/pgalloc.h | 9 ++++++---
 2 files changed, 7 insertions(+), 3 deletions(-)

diff --git a/arch/arc/Kconfig b/arch/arc/Kconfig
index f27e6b90428e4..47db93952386d 100644
--- a/arch/arc/Kconfig
+++ b/arch/arc/Kconfig
@@ -54,6 +54,7 @@ config ARC
 	select HAVE_ARCH_JUMP_LABEL if ISA_ARCV2 && !CPU_ENDIAN_BE32
 	select TRACE_IRQFLAGS_SUPPORT
 	select HAVE_EBPF_JIT if ISA_ARCV2
+	select MMU_GATHER_RCU_TABLE_FREE
 
 config LOCKDEP_SUPPORT
 	def_bool y
diff --git a/arch/arc/include/asm/pgalloc.h b/arch/arc/include/asm/pgalloc.h
index dfae070fe8d55..b1c6619435613 100644
--- a/arch/arc/include/asm/pgalloc.h
+++ b/arch/arc/include/asm/pgalloc.h
@@ -72,7 +72,8 @@ static inline void p4d_populate(struct mm_struct *mm, p4d_t *p4dp, pud_t *pudp)
 	set_p4d(p4dp, __p4d((unsigned long)pudp));
 }
 
-#define __pud_free_tlb(tlb, pmd, addr)  pud_free((tlb)->mm, pmd)
+#define __pud_free_tlb(tlb, pud, addr)	\
+	tlb_remove_ptdesc((tlb), virt_to_ptdesc(pud))
 
 #endif
 
@@ -83,10 +84,12 @@ static inline void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmdp)
 	set_pud(pudp, __pud((unsigned long)pmdp));
 }
 
-#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 /* _ASM_ARC_PGALLOC_H */
-- 
2.20.1

Re: [PATCH 2/7] arc: mm: enable MMU_GATHER_RCU_TABLE_FREE

[Qi Zheng]

On 11/14/25 7:11 PM, Qi Zheng wrote:
> From: Qi Zheng <zhengqi.arch@bytedance.com>
> 
> 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.
> 
> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
> Cc: Vineet Gupta <vgupta@kernel.org>
> ---
>   arch/arc/Kconfig               | 1 +
>   arch/arc/include/asm/pgalloc.h | 9 ++++++---
>   2 files changed, 7 insertions(+), 3 deletions(-)

Strangely, it seems that only ARC does not define CONFIG_64BIT?

Does the ARC architecture support 64-bit? Did I miss something?

> 
> diff --git a/arch/arc/Kconfig b/arch/arc/Kconfig
> index f27e6b90428e4..47db93952386d 100644
> --- a/arch/arc/Kconfig
> +++ b/arch/arc/Kconfig
> @@ -54,6 +54,7 @@ config ARC
>   	select HAVE_ARCH_JUMP_LABEL if ISA_ARCV2 && !CPU_ENDIAN_BE32
>   	select TRACE_IRQFLAGS_SUPPORT
>   	select HAVE_EBPF_JIT if ISA_ARCV2
> +	select MMU_GATHER_RCU_TABLE_FREE
>   
>   config LOCKDEP_SUPPORT
>   	def_bool y
> diff --git a/arch/arc/include/asm/pgalloc.h b/arch/arc/include/asm/pgalloc.h
> index dfae070fe8d55..b1c6619435613 100644
> --- a/arch/arc/include/asm/pgalloc.h
> +++ b/arch/arc/include/asm/pgalloc.h
> @@ -72,7 +72,8 @@ static inline void p4d_populate(struct mm_struct *mm, p4d_t *p4dp, pud_t *pudp)
>   	set_p4d(p4dp, __p4d((unsigned long)pudp));
>   }
>   
> -#define __pud_free_tlb(tlb, pmd, addr)  pud_free((tlb)->mm, pmd)
> +#define __pud_free_tlb(tlb, pud, addr)	\
> +	tlb_remove_ptdesc((tlb), virt_to_ptdesc(pud))
>   
>   #endif
>   
> @@ -83,10 +84,12 @@ static inline void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmdp)
>   	set_pud(pudp, __pud((unsigned long)pmdp));
>   }
>   
> -#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 /* _ASM_ARC_PGALLOC_H */

Re: [PATCH 2/7] arc: mm: enable MMU_GATHER_RCU_TABLE_FREE

[Vineet Gupta]

On 11/14/25 03:20, Qi Zheng wrote:
> Strangely, it seems that only ARC does not define CONFIG_64BIT?
>
> Does the ARC architecture support 64-bit? Did I miss something?

ARC is 32-bit only !

-Vineet

Re: [PATCH 2/7] arc: mm: enable MMU_GATHER_RCU_TABLE_FREE

[Qi Zheng]

On 11/15/25 7:10 AM, Vineet Gupta wrote:
> On 11/14/25 03:20, Qi Zheng wrote:
>> Strangely, it seems that only ARC does not define CONFIG_64BIT?
>>
>> Does the ARC architecture support 64-bit? Did I miss something?
> 
> ARC is 32-bit only !

Got it! Will drop this patch in the next version.

Thanks!

> 
> -Vineet

[PATCH 3/7] loongarch: mm: enable MMU_GATHER_RCU_TABLE_FREE

[Qi Zheng]

From: Qi Zheng <zhengqi.arch@bytedance.com>

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.

Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Huacai Chen <chenhuacai@kernel.org>
Cc: WANG Xuerui <kernel@xen0n.name>
---
 arch/loongarch/Kconfig               | 1 +
 arch/loongarch/include/asm/pgalloc.h | 6 ++++--
 2 files changed, 5 insertions(+), 2 deletions(-)

diff --git a/arch/loongarch/Kconfig b/arch/loongarch/Kconfig
index 5b1116733d881..3bf2f2a9cd647 100644
--- a/arch/loongarch/Kconfig
+++ b/arch/loongarch/Kconfig
@@ -210,6 +210,7 @@ config LOONGARCH
 	select USER_STACKTRACE_SUPPORT
 	select VDSO_GETRANDOM
 	select ZONE_DMA32
+	select MMU_GATHER_RCU_TABLE_FREE
 
 config 32BIT
 	bool
diff --git a/arch/loongarch/include/asm/pgalloc.h b/arch/loongarch/include/asm/pgalloc.h
index 1c63a9d9a6d35..0539d04bf1525 100644
--- a/arch/loongarch/include/asm/pgalloc.h
+++ b/arch/loongarch/include/asm/pgalloc.h
@@ -79,7 +79,8 @@ 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 +100,8 @@ 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 */
 
-- 
2.20.1

Re: [PATCH 3/7] loongarch: mm: enable MMU_GATHER_RCU_TABLE_FREE

[Huacai Chen]

Hi, Qi Zheng,

We usually use LoongArch rather than loongarch, but if you want to
keep consistency for all patches, just do it.

On Fri, Nov 14, 2025 at 7:13 PM Qi Zheng <qi.zheng@linux.dev> wrote:
>
> From: Qi Zheng <zhengqi.arch@bytedance.com>
>
> 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.
>
> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
> Cc: Huacai Chen <chenhuacai@kernel.org>
> Cc: WANG Xuerui <kernel@xen0n.name>
> ---
>  arch/loongarch/Kconfig               | 1 +
>  arch/loongarch/include/asm/pgalloc.h | 6 ++++--
>  2 files changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/arch/loongarch/Kconfig b/arch/loongarch/Kconfig
> index 5b1116733d881..3bf2f2a9cd647 100644
> --- a/arch/loongarch/Kconfig
> +++ b/arch/loongarch/Kconfig
> @@ -210,6 +210,7 @@ config LOONGARCH
>         select USER_STACKTRACE_SUPPORT
>         select VDSO_GETRANDOM
>         select ZONE_DMA32
> +       select MMU_GATHER_RCU_TABLE_FREE
Please use alpha-betical order.

>
>  config 32BIT
>         bool
> diff --git a/arch/loongarch/include/asm/pgalloc.h b/arch/loongarch/include/asm/pgalloc.h
> index 1c63a9d9a6d35..0539d04bf1525 100644
> --- a/arch/loongarch/include/asm/pgalloc.h
> +++ b/arch/loongarch/include/asm/pgalloc.h
> @@ -79,7 +79,8 @@ 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))
I think we can define it in one line.

>
>  #endif
>
> @@ -99,7 +100,8 @@ 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))
The same.

Other patches have the same problem.

Huacai

>
>  #endif /* __PAGETABLE_PUD_FOLDED */
>
> --
> 2.20.1
>

Re: [PATCH 3/7] loongarch: mm: enable MMU_GATHER_RCU_TABLE_FREE

[Qi Zheng]

Hi Huacai,

On 11/14/25 10:17 PM, Huacai Chen wrote:
> Hi, Qi Zheng,
> 
> We usually use LoongArch rather than loongarch, but if you want to
> keep consistency for all patches, just do it.

OK, will change to use LoongArch.

> 
> On Fri, Nov 14, 2025 at 7:13 PM Qi Zheng <qi.zheng@linux.dev> wrote:
>>
>> From: Qi Zheng <zhengqi.arch@bytedance.com>
>>
>> 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.
>>
>> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
>> Cc: Huacai Chen <chenhuacai@kernel.org>
>> Cc: WANG Xuerui <kernel@xen0n.name>
>> ---
>>   arch/loongarch/Kconfig               | 1 +
>>   arch/loongarch/include/asm/pgalloc.h | 6 ++++--
>>   2 files changed, 5 insertions(+), 2 deletions(-)
>>
>> diff --git a/arch/loongarch/Kconfig b/arch/loongarch/Kconfig
>> index 5b1116733d881..3bf2f2a9cd647 100644
>> --- a/arch/loongarch/Kconfig
>> +++ b/arch/loongarch/Kconfig
>> @@ -210,6 +210,7 @@ config LOONGARCH
>>          select USER_STACKTRACE_SUPPORT
>>          select VDSO_GETRANDOM
>>          select ZONE_DMA32
>> +       select MMU_GATHER_RCU_TABLE_FREE
> Please use alpha-betical order.

OK, will do.

> 
>>
>>   config 32BIT
>>          bool
>> diff --git a/arch/loongarch/include/asm/pgalloc.h b/arch/loongarch/include/asm/pgalloc.h
>> index 1c63a9d9a6d35..0539d04bf1525 100644
>> --- a/arch/loongarch/include/asm/pgalloc.h
>> +++ b/arch/loongarch/include/asm/pgalloc.h
>> @@ -79,7 +79,8 @@ 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))
> I think we can define it in one line.

will do.

> 
>>
>>   #endif
>>
>> @@ -99,7 +100,8 @@ 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))
> The same.
> 
> Other patches have the same problem.

Got it, will convert them all to the one-line type.

Thanks,
Qi

> 
> Huacai
> 
>>
>>   #endif /* __PAGETABLE_PUD_FOLDED */
>>
>> --
>> 2.20.1
>>

Re: [PATCH 3/7] loongarch: mm: enable MMU_GATHER_RCU_TABLE_FREE

[Qi Zheng]

Hi Huacai,

On 11/14/25 10:17 PM, Huacai Chen wrote:
> Hi, Qi Zheng,

[...]

>>
>> -#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))
> I think we can define it in one line.

Do we need to change __pte_free_tlb() to a single-line format
as well?

Thanks,
Qi


>>

Re: [PATCH 3/7] loongarch: mm: enable MMU_GATHER_RCU_TABLE_FREE

[Huacai Chen]

On Mon, Nov 17, 2025 at 2:42 PM Qi Zheng <qi.zheng@linux.dev> wrote:
>
> Hi Huacai,
>
> On 11/14/25 10:17 PM, Huacai Chen wrote:
> > Hi, Qi Zheng,
>
> [...]
>
> >>
> >> -#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))
> > I think we can define it in one line.
>
> Do we need to change __pte_free_tlb() to a single-line format
> as well?
Yes, there is no 80 columns limit now.

Huacai

>
> Thanks,
> Qi
>
>
> >>
>
>

[PATCH 4/7] mips: mm: enable MMU_GATHER_RCU_TABLE_FREE

[Qi Zheng]

From: Qi Zheng <zhengqi.arch@bytedance.com>

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.

Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
---
 arch/mips/Kconfig               | 1 +
 arch/mips/include/asm/pgalloc.h | 6 ++++--
 2 files changed, 5 insertions(+), 2 deletions(-)

diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
index e8683f58fd3e2..0ee8820a354c4 100644
--- a/arch/mips/Kconfig
+++ b/arch/mips/Kconfig
@@ -108,6 +108,7 @@ config MIPS
 	select TRACE_IRQFLAGS_SUPPORT
 	select ARCH_HAS_ELFCORE_COMPAT
 	select HAVE_ARCH_KCSAN if 64BIT
+	select MMU_GATHER_RCU_TABLE_FREE
 
 config MIPS_FIXUP_BIGPHYS_ADDR
 	bool
diff --git a/arch/mips/include/asm/pgalloc.h b/arch/mips/include/asm/pgalloc.h
index 942af87f1cddb..c00f445045f43 100644
--- a/arch/mips/include/asm/pgalloc.h
+++ b/arch/mips/include/asm/pgalloc.h
@@ -72,7 +72,8 @@ 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
 
@@ -98,7 +99,8 @@ 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 */
 
-- 
2.20.1

[PATCH 5/7] parisc: mm: enable MMU_GATHER_RCU_TABLE_FREE

[Qi Zheng]

From: Qi Zheng <zhengqi.arch@bytedance.com>

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.

Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Helge Deller <deller@gmx.de>
---
 arch/parisc/Kconfig           | 1 +
 arch/parisc/include/asm/tlb.h | 6 ++++--
 2 files changed, 5 insertions(+), 2 deletions(-)

diff --git a/arch/parisc/Kconfig b/arch/parisc/Kconfig
index 47fd9662d8005..946cbe21a4118 100644
--- a/arch/parisc/Kconfig
+++ b/arch/parisc/Kconfig
@@ -92,6 +92,7 @@ config PARISC
 	select TRACE_IRQFLAGS_SUPPORT
 	select HAVE_FUNCTION_DESCRIPTORS if 64BIT
 	select PCI_MSI_ARCH_FALLBACKS if PCI_MSI
+	select MMU_GATHER_RCU_TABLE_FREE
 
 	help
 	  The PA-RISC microprocessor is designed by Hewlett-Packard and used
diff --git a/arch/parisc/include/asm/tlb.h b/arch/parisc/include/asm/tlb.h
index 44235f367674d..ab7d4113df61a 100644
--- a/arch/parisc/include/asm/tlb.h
+++ b/arch/parisc/include/asm/tlb.h
@@ -5,8 +5,10 @@
 #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
-- 
2.20.1

[PATCH 6/7] um: mm: enable MMU_GATHER_RCU_TABLE_FREE

[Qi Zheng]

From: Qi Zheng <zhengqi.arch@bytedance.com>

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.

Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Richard Weinberger <richard@nod.at>
Cc: Anton Ivanov <anton.ivanov@cambridgegreys.com>
Cc: Johannes Berg <johannes@sipsolutions.net>
---
 arch/um/Kconfig | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/um/Kconfig b/arch/um/Kconfig
index 097c6a6265ef3..47a41bc77bb24 100644
--- a/arch/um/Kconfig
+++ b/arch/um/Kconfig
@@ -41,6 +41,7 @@ config UML
 	select HAVE_SYSCALL_TRACEPOINTS
 	select THREAD_INFO_IN_TASK
 	select SPARSE_IRQ
+	select MMU_GATHER_RCU_TABLE_FREE
 
 config MMU
 	bool
-- 
2.20.1

[PATCH 7/7] mm: make PT_RECLAIM depend on MMU_GATHER_RCU_TABLE_FREE && 64BIT

[Qi Zheng]

From: Qi Zheng <zhengqi.arch@bytedance.com>

Make PT_RECLAIM depend on MMU_GATHER_RCU_TABLE_FREE so that PT_RECLAIM can
be enabled by default on all architectures that support
MMU_GATHER_RCU_TABLE_FREE.

Considering that a large number of PTE page table pages (such as 100GB+)
can only be caused on a 64-bit system, let PT_RECLAIM also depend on
64BIT.

Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
---
 arch/x86/Kconfig | 1 -
 mm/Kconfig       | 6 +-----
 2 files changed, 1 insertion(+), 6 deletions(-)

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index eac2e86056902..96bff81fd4787 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -330,7 +330,6 @@ 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
diff --git a/mm/Kconfig b/mm/Kconfig
index a5a90b169435d..e795fbd69e50c 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -1440,14 +1440,10 @@ 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
+	depends on MMU_GATHER_RCU_TABLE_FREE && MMU && SMP && 64BIT
 	help
 	  Try to reclaim empty user page table pages in paths other than munmap
 	  and exit_mmap path.
-- 
2.20.1

Re: [PATCH 7/7] mm: make PT_RECLAIM depend on MMU_GATHER_RCU_TABLE_FREE && 64BIT

[kernel test robot]

Hi Qi,

kernel test robot noticed the following build errors:

[auto build test ERROR on deller-parisc/for-next]
[also build test ERROR on uml/next tip/x86/core akpm-mm/mm-everything linus/master v6.18-rc5 next-20251114]
[cannot apply to uml/fixes vgupta-arc/for-next vgupta-arc/for-curr]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Qi-Zheng/alpha-mm-enable-MMU_GATHER_RCU_TABLE_FREE/20251114-191543
base:   https://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux.git for-next
patch link:    https://lore.kernel.org/r/0a4d1e6f0bf299cafd1fc624f965bd1ca542cea8.1763117269.git.zhengqi.arch%40bytedance.com
patch subject: [PATCH 7/7] mm: make PT_RECLAIM depend on MMU_GATHER_RCU_TABLE_FREE && 64BIT
config: arm64-randconfig-004-20251115 (https://download.01.org/0day-ci/archive/20251115/202511150845.XqOxPJxe-lkp@intel.com/config)
compiler: aarch64-linux-gcc (GCC) 8.5.0
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20251115/202511150845.XqOxPJxe-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202511150845.XqOxPJxe-lkp@intel.com/

All errors (new ones prefixed by >>):

   In file included from mm/pt_reclaim.c:3:
   mm/pt_reclaim.c: In function 'free_pte':
>> include/asm-generic/tlb.h:731:3: error: implicit declaration of function '__pte_free_tlb'; did you mean 'pte_free_tlb'? [-Werror=implicit-function-declaration]
      __pte_free_tlb(tlb, ptep, address);  \
      ^~~~~~~~~~~~~~
   mm/pt_reclaim.c:31:2: note: in expansion of macro 'pte_free_tlb'
     pte_free_tlb(tlb, pmd_pgtable(pmdval), addr);
     ^~~~~~~~~~~~
   cc1: some warnings being treated as errors


vim +731 include/asm-generic/tlb.h

a00cc7d9dd93d6 Matthew Wilcox         2017-02-24  701  
a00cc7d9dd93d6 Matthew Wilcox         2017-02-24  702  #define tlb_remove_pud_tlb_entry(tlb, pudp, address)			\
a00cc7d9dd93d6 Matthew Wilcox         2017-02-24  703  	do {								\
2631ed00b04988 Peter Zijlstra (Intel  2020-06-25  704) 		tlb_flush_pud_range(tlb, address, HPAGE_PUD_SIZE);	\
a00cc7d9dd93d6 Matthew Wilcox         2017-02-24  705  		__tlb_remove_pud_tlb_entry(tlb, pudp, address);		\
a00cc7d9dd93d6 Matthew Wilcox         2017-02-24  706  	} while (0)
a00cc7d9dd93d6 Matthew Wilcox         2017-02-24  707  
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  708  /*
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  709   * For things like page tables caches (ie caching addresses "inside" the
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  710   * page tables, like x86 does), for legacy reasons, flushing an
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  711   * individual page had better flush the page table caches behind it. This
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  712   * is definitely how x86 works, for example. And if you have an
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  713   * architected non-legacy page table cache (which I'm not aware of
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  714   * anybody actually doing), you're going to have some architecturally
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  715   * explicit flushing for that, likely *separate* from a regular TLB entry
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  716   * flush, and thus you'd need more than just some range expansion..
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  717   *
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  718   * So if we ever find an architecture
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  719   * that would want something that odd, I think it is up to that
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  720   * architecture to do its own odd thing, not cause pain for others
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  721   * http://lkml.kernel.org/r/CA+55aFzBggoXtNXQeng5d_mRoDnaMBE5Y+URs+PHR67nUpMtaw@mail.gmail.com
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  722   *
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  723   * For now w.r.t page table cache, mark the range_size as PAGE_SIZE
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  724   */
b5bc66b7131087 Aneesh Kumar K.V       2016-12-12  725  
a90744bac57c3c Nicholas Piggin        2018-07-13  726  #ifndef pte_free_tlb
9e1b32caa525cb Benjamin Herrenschmidt 2009-07-22  727  #define pte_free_tlb(tlb, ptep, address)			\
^1da177e4c3f41 Linus Torvalds         2005-04-16  728  	do {							\
2631ed00b04988 Peter Zijlstra (Intel  2020-06-25  729) 		tlb_flush_pmd_range(tlb, address, PAGE_SIZE);	\
22a61c3c4f1379 Peter Zijlstra         2018-08-23  730  		tlb->freed_tables = 1;				\
9e1b32caa525cb Benjamin Herrenschmidt 2009-07-22 @731  		__pte_free_tlb(tlb, ptep, address);		\
^1da177e4c3f41 Linus Torvalds         2005-04-16  732  	} while (0)
a90744bac57c3c Nicholas Piggin        2018-07-13  733  #endif
^1da177e4c3f41 Linus Torvalds         2005-04-16  734  

-- 
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [PATCH 7/7] mm: make PT_RECLAIM depend on MMU_GATHER_RCU_TABLE_FREE && 64BIT

[kernel test robot]

Hi Qi,

kernel test robot noticed the following build errors:

[auto build test ERROR on deller-parisc/for-next]
[also build test ERROR on uml/next tip/x86/core akpm-mm/mm-everything linus/master v6.18-rc5 next-20251114]
[cannot apply to uml/fixes vgupta-arc/for-next vgupta-arc/for-curr]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Qi-Zheng/alpha-mm-enable-MMU_GATHER_RCU_TABLE_FREE/20251114-191543
base:   https://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux.git for-next
patch link:    https://lore.kernel.org/r/0a4d1e6f0bf299cafd1fc624f965bd1ca542cea8.1763117269.git.zhengqi.arch%40bytedance.com
patch subject: [PATCH 7/7] mm: make PT_RECLAIM depend on MMU_GATHER_RCU_TABLE_FREE && 64BIT
config: arm64-randconfig-002-20251115 (https://download.01.org/0day-ci/archive/20251115/202511150832.iAyO0SAW-lkp@intel.com/config)
compiler: clang version 18.1.8 (https://github.com/llvm/llvm-project 3b5b5c1ec4a3095ab096dd780e84d7ab81f3d7ff)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20251115/202511150832.iAyO0SAW-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202511150832.iAyO0SAW-lkp@intel.com/

All errors (new ones prefixed by >>):

>> mm/pt_reclaim.c:31:2: error: call to undeclared function '__pte_free_tlb'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
      31 |         pte_free_tlb(tlb, pmd_pgtable(pmdval), addr);
         |         ^
   include/asm-generic/tlb.h:731:3: note: expanded from macro 'pte_free_tlb'
     731 |                 __pte_free_tlb(tlb, ptep, address);             \
         |                 ^
   1 error generated.


vim +/__pte_free_tlb +31 mm/pt_reclaim.c

6375e95f381e3d Qi Zheng 2024-12-04  27  
6375e95f381e3d Qi Zheng 2024-12-04  28  void free_pte(struct mm_struct *mm, unsigned long addr, struct mmu_gather *tlb,
6375e95f381e3d Qi Zheng 2024-12-04  29  	      pmd_t pmdval)
6375e95f381e3d Qi Zheng 2024-12-04  30  {
6375e95f381e3d Qi Zheng 2024-12-04 @31  	pte_free_tlb(tlb, pmd_pgtable(pmdval), addr);
6375e95f381e3d Qi Zheng 2024-12-04  32  	mm_dec_nr_ptes(mm);
6375e95f381e3d Qi Zheng 2024-12-04  33  }
6375e95f381e3d Qi Zheng 2024-12-04  34  

-- 
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [PATCH 7/7] mm: make PT_RECLAIM depend on MMU_GATHER_RCU_TABLE_FREE && 64BIT

[David Hildenbrand (Red Hat)]

On 14.11.25 12:11, Qi Zheng wrote:
> From: Qi Zheng <zhengqi.arch@bytedance.com>

Subject: s/&&/&/

> 
> Make PT_RECLAIM depend on MMU_GATHER_RCU_TABLE_FREE so that PT_RECLAIM can
> be enabled by default on all architectures that support
> MMU_GATHER_RCU_TABLE_FREE.
> 
> Considering that a large number of PTE page table pages (such as 100GB+)
> can only be caused on a 64-bit system, let PT_RECLAIM also depend on
> 64BIT.
> 
> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
> ---
>   arch/x86/Kconfig | 1 -
>   mm/Kconfig       | 6 +-----
>   2 files changed, 1 insertion(+), 6 deletions(-)
> 
> diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
> index eac2e86056902..96bff81fd4787 100644
> --- a/arch/x86/Kconfig
> +++ b/arch/x86/Kconfig
> @@ -330,7 +330,6 @@ 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
> diff --git a/mm/Kconfig b/mm/Kconfig
> index a5a90b169435d..e795fbd69e50c 100644
> --- a/mm/Kconfig
> +++ b/mm/Kconfig
> @@ -1440,14 +1440,10 @@ 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
> +	depends on MMU_GATHER_RCU_TABLE_FREE && MMU && SMP && 64BIT

Who would we have MMU_GATHER_RCU_TABLE_FREE without MMU? (can we drop 
the MMU part)

Why do we care about SMP in the first place? (can we frop SMP)

But I also wonder why we need "MMU_GATHER_RCU_TABLE_FREE && 64BIT":

Would it be harmful on 32bit (sure, we might not reclaim as much, but 
still there is memory to be reclaimed?)?

If all 64BIT support MMU_GATHER_RCU_TABLE_FREE (as you previously 
state), why can't we only check for 64BIT?

-- 
Cheers

David

Re: [PATCH 0/7] enable PT_RECLAIM on all 64-bit architectures

[David Hildenbrand (Red Hat)]

On 14.11.25 12:11, Qi Zheng wrote:
> From: Qi Zheng <zhengqi.arch@bytedance.com>
> 
> 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.
> 

Makes sense!

> 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
> && 64BIT. This way, PT_RECLAIM can be enabled by default on all 64-bit
> architectures.

Could we then even go ahead and stop making PT_RECLAIM user-selectable?

-- 
Cheers

David