support "THPeligible" semantics for mTHP with anonymous shmem

PATCH Superseded Bang Li 5 participants 28 messages Jun 28, 2024 – Jul 9, 2024

Subsystems (5)

Memory management Proc filesystem Memory mapping THP +1 more subsystems

lore

Share & tools

lore

Bang Li

author

2024-06-28 10:49 UTC #0 lore raw

PATCH 1/1 support "THPeligible" semantics for mTHP with anonymous shmem

After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.

Signed-off-by Bang Li <[email protected]>

---
 fs/proc/task_mmu.c      | 10 +++++++---
 include/linux/huge_mm.h | 11 +++++++++++
 mm/shmem.c              |  9 +--------
 3 files changed, 19 insertions(+), 11 deletions(-)

fs/proc/task_mmu.c +7 -3

index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
 {
 	struct vm_area_struct *vma = v;
 	struct mem_size_stats mss = {};
+	bool thp_eligible;
 
 	smap_gather_stats(vma, &mss, 0);

@@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void *v)
 
 	__show_smap(m, &mss, false);
 
-	seq_printf(m, "THPeligible:    %8u\n",
-		   !!thp_vma_allowable_orders(vma, vma->vm_flags,
-			   TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
+	thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
+						TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL);
+	if (vma_is_anon_shmem(vma))
+		thp_eligible = !!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+							vma, vma->vm_pgoff, thp_eligible);
+	seq_printf(m, "THPeligible:    %8u\n", thp_eligible);
 
 	if (arch_pkeys_enabled())
 		seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));

include/linux/huge_mm.h +11 -0

index 212cca384d7e..f87136f38aa1 100644
@@ -267,6 +267,10 @@ unsigned long thp_vma_allowable_orders(struct vm_area_struct *vma,
 	return __thp_vma_allowable_orders(vma, vm_flags, tva_flags, orders);
 }
 
+unsigned long shmem_allowable_huge_orders(struct inode *inode,
+				struct vm_area_struct *vma, pgoff_t index,
+				bool global_huge);
+
 struct thpsize {
 	struct kobject kobj;
 	struct list_head node;

@@ -460,6 +464,13 @@ static inline unsigned long thp_vma_allowable_orders(struct vm_area_struct *vma,
 	return 0;
 }
 
+static inline unsigned long shmem_allowable_huge_orders(struct inode *inode,
+				struct vm_area_struct *vma, pgoff_t index,
+				bool global_huge)
+{
+	return 0;
+}
+
 #define transparent_hugepage_flags 0UL
 
 #define thp_get_unmapped_area	NULL

mm/shmem.c +1 -8

index d495c0701a83..aa85df9c662a 100644
@@ -1622,7 +1622,7 @@ static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
 }
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-static unsigned long shmem_allowable_huge_orders(struct inode *inode,
+unsigned long shmem_allowable_huge_orders(struct inode *inode,
 				struct vm_area_struct *vma, pgoff_t index,
 				bool global_huge)
 {

@@ -1707,13 +1707,6 @@ static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault
 	return orders;
 }
 #else
-static unsigned long shmem_allowable_huge_orders(struct inode *inode,
-				struct vm_area_struct *vma, pgoff_t index,
-				bool global_huge)
-{
-	return 0;
-}
-
 static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault *vmf,
 					   struct address_space *mapping, pgoff_t index,
 					   unsigned long orders)

Signature · 1 line

2.19.1.6.gb485710b

Baolin Wang

reviewer

2024-07-01 06:47 UTC #1 lore raw

CC Barry.

Quoted 8 lines · Bang Li

After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.

I did not see a consensus that "THP*" related statistics should contain mTHP in previous discussion [1].

In addition, if we all agree that "THPeligible" should include mTHP statistics, you should update the corresponding documentation to keep consistency.

[1] https://lore.kernel.org/linux-mm/[email protected]/T/#md7a77056110cebcc2a9b3cd7e4a8d682667f6ba5

Quoted 91 lines

Signed-off-by Bang Li <[email protected]>

---
  fs/proc/task_mmu.c      | 10 +++++++---
  include/linux/huge_mm.h | 11 +++++++++++
  mm/shmem.c              |  9 +--------
  3 files changed, 19 insertions(+), 11 deletions(-)

fs/proc/task_mmu.c +7 -3

index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
  {
  	struct vm_area_struct *vma = v;
  	struct mem_size_stats mss = {};
+	bool thp_eligible;
 
  	smap_gather_stats(vma, &mss, 0);

@@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void *v)
 
  	__show_smap(m, &mss, false);
 
-	seq_printf(m, "THPeligible:    %8u\n",
-		   !!thp_vma_allowable_orders(vma, vma->vm_flags,
-			   TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
+	thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
+						TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL);
+	if (vma_is_anon_shmem(vma))
+		thp_eligible = !!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+							vma, vma->vm_pgoff, thp_eligible);
+	seq_printf(m, "THPeligible:    %8u\n", thp_eligible);
 
  	if (arch_pkeys_enabled())
  		seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));

include/linux/huge_mm.h +11 -0

index 212cca384d7e..f87136f38aa1 100644
@@ -267,6 +267,10 @@ unsigned long thp_vma_allowable_orders(struct vm_area_struct *vma,
  	return __thp_vma_allowable_orders(vma, vm_flags, tva_flags, orders);
  }
 
+unsigned long shmem_allowable_huge_orders(struct inode *inode,
+				struct vm_area_struct *vma, pgoff_t index,
+				bool global_huge);
+
  struct thpsize {
  	struct kobject kobj;
  	struct list_head node;

@@ -460,6 +464,13 @@ static inline unsigned long thp_vma_allowable_orders(struct vm_area_struct *vma,
  	return 0;
  }
 
+static inline unsigned long shmem_allowable_huge_orders(struct inode *inode,
+				struct vm_area_struct *vma, pgoff_t index,
+				bool global_huge)
+{
+	return 0;
+}
+
  #define transparent_hugepage_flags 0UL
 
  #define thp_get_unmapped_area	NULL

mm/shmem.c +1 -8

index d495c0701a83..aa85df9c662a 100644
@@ -1622,7 +1622,7 @@ static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
  }
 
  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-static unsigned long shmem_allowable_huge_orders(struct inode *inode,
+unsigned long shmem_allowable_huge_orders(struct inode *inode,
  				struct vm_area_struct *vma, pgoff_t index,
  				bool global_huge)
  {

@@ -1707,13 +1707,6 @@ static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault
  	return orders;
  }
  #else
-static unsigned long shmem_allowable_huge_orders(struct inode *inode,
-				struct vm_area_struct *vma, pgoff_t index,
-				bool global_huge)
-{
-	return 0;
-}
-
  static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault *vmf,
  					   struct address_space *mapping, pgoff_t index,
  					   unsigned long orders)

David Hildenbrand

2024-07-01 06:55 UTC #2 lore raw

Quoted 15 lines · Baolin Wang

CC Barry.
Quoted 8 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
I did not see a consensus that "THP*" related statistics should contain mTHP in previous discussion [1].

Quoted 7 lines

In addition, if we all agree that "THPeligible" should include mTHP statistics, you should update the corresponding documentation to keep consistency.
[1]
https://lore.kernel.org/linux-mm/[email protected]/T/#md7a77056110cebcc2a9b3cd7e4a8d682667f6ba5

Fortunately, documentation (Documentation/filesystems/proc.rst) says:

"THPeligible" indicates whether the mapping is eligible for allocating naturally aligned THP pages of any currently enabled size. 1 if true, 0 otherwise."

So that documentation is already pretty clear (we just have to make sure the other ones are properly documented, for example as raised in reply to [1]).

Signature · 3 lines

Cheers,

David / dhildenb

Baolin Wang

reviewer

2024-07-01 07:18 UTC #3 lore raw

Quoted 24 lines · David Hildenbrand

Quoted 15 lines · Baolin Wang
CC Barry.
Quoted 8 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
I did not see a consensus that "THP*" related statistics should contain mTHP in previous discussion [1].
Quoted 7 lines
In addition, if we all agree that "THPeligible" should include mTHP statistics, you should update the corresponding documentation to keep consistency.
[1]
https://lore.kernel.org/linux-mm/[email protected]/T/#md7a77056110cebcc2a9b3cd7e4a8d682667f6ba5

Quoted 5 lines

Fortunately, documentation (Documentation/filesystems/proc.rst) says:
"THPeligible" indicates whether the mapping is eligible for allocating naturally aligned THP pages of any currently enabled size. 1 if true, 0 otherwise."
So that documentation is already pretty clear (we just have to make sure the other ones are properly documented, for example as raised in reply to [1]).

OK, great. Thanks.

Bang Li

author

2024-07-01 08:23 UTC #4 lore raw

In reply to David Hildenbrand

Hi David,

Quoted 24 lines · David Hildenbrand

Quoted 15 lines · Baolin Wang
CC Barry.
Quoted 8 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
I did not see a consensus that "THP*" related statistics should contain mTHP in previous discussion [1].
Quoted 7 lines
In addition, if we all agree that "THPeligible" should include mTHP statistics, you should update the corresponding documentation to keep consistency.
[1]
https://lore.kernel.org/linux-mm/[email protected]/T/#md7a77056110cebcc2a9b3cd7e4a8d682667f6ba5

Quoted 6 lines

Fortunately, documentation (Documentation/filesystems/proc.rst) says:
"THPeligible" indicates whether the mapping is eligible for allocating naturally aligned THP pages of any currently enabled size. 1 if true, 0 otherwise."
So that documentation is already pretty clear (we just have to make sure the other ones are properly documented, for example as raised in reply to [1]).

Thanks for the clarification.

Thanks,
Bang

David Hildenbrand

2024-07-01 06:57 UTC #5 lore raw

In reply to Bang Li

Quoted 30 lines · Bang Li

After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
  fs/proc/task_mmu.c      | 10 +++++++---
  include/linux/huge_mm.h | 11 +++++++++++
  mm/shmem.c              |  9 +--------
  3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +1 -0
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
  {
  	struct vm_area_struct *vma = v;
  	struct mem_size_stats mss = {};
+	bool thp_eligible;
 
  	smap_gather_stats(vma, &mss, 0);
 
@@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void *v)
 

fs/proc/task_mmu.c static int show_smap(struct seq_file *m, void *v) · line 890

Quoted 10 lines

fs/proc/task_mmu.c +5 -3

  	__show_smap(m, &mss, false);
 
-	seq_printf(m, "THPeligible:    %8u\n",
-		   !!thp_vma_allowable_orders(vma, vma->vm_flags,
-			   TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
+	thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
+						TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL);
+	if (vma_is_anon_shmem(vma))
+		thp_eligible = !!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+							vma, vma->vm_pgoff, thp_eligible);

I would have thought the correct fix is to return the correct result from thp_vma_allowable_orders().

Signature · 3 lines

Cheers,

David / dhildenb

Bang Li

author

2024-07-01 08:24 UTC #6 lore raw

Hi David,

Thanks for you review!

Quoted 32 lines · David Hildenbrand

Quoted 31 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
fs/proc/task_mmu.c | 10 +++++++---
include/linux/huge_mm.h | 11 +++++++++++
mm/shmem.c | 9 +--------
  3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +1 -3
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
   {
       struct vm_area_struct *vma = v;
       struct mem_size_stats mss = {};
+    bool thp_eligible;
       smap_gather_stats(vma, &mss, 0);
@@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void *v)
       __show_smap(m, &mss, false);
-    seq_printf(m, "THPeligible:    %8u\n",
-           !!thp_vma_allowable_orders(vma, vma->vm_flags,
-               TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));

Quoted 9 lines

fs/proc/task_mmu.c static int show_smap(struct seq_file *m, void *v) · line 890
Quoted 6 lines
fs/proc/task_mmu.c +5 -0
+    thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
+                        TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL);
+    if (vma_is_anon_shmem(vma))
+        thp_eligible = 
 !!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+                            vma, vma->vm_pgoff, thp_eligible);
I would have thought the correct fix is to return the correct result from thp_vma_allowable_orders().

Agreed. I'll try to reimplement this in thp_vma_allowable_orders().

Thanks again for the review!
Bang

Ryan Roberts

reviewer

2024-07-01 07:55 UTC #7 lore raw

In reply to Bang Li

Quoted 30 lines · Bang Li

After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
 fs/proc/task_mmu.c      | 10 +++++++---
 include/linux/huge_mm.h | 11 +++++++++++
 mm/shmem.c              |  9 +--------
 3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +1 -0
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
 {
 	struct vm_area_struct *vma = v;
 	struct mem_size_stats mss = {};
+	bool thp_eligible;
 
 	smap_gather_stats(vma, &mss, 0);
 
@@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void *v)
 

fs/proc/task_mmu.c static int show_smap(struct seq_file *m, void *v) · line 890

Quoted 10 lines

fs/proc/task_mmu.c +5 -3

 	__show_smap(m, &mss, false);
 
-	seq_printf(m, "THPeligible:    %8u\n",
-		   !!thp_vma_allowable_orders(vma, vma->vm_flags,
-			   TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
+	thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
+						TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL);
+	if (vma_is_anon_shmem(vma))
+		thp_eligible = !!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+							vma, vma->vm_pgoff, thp_eligible);

Afraid I haven't been following the shmem mTHP support work as much as I would have liked, but is there a reason why we need a separate function for shmem? Couldn't (shouldn't) thp_vma_allowable_orders() be taught to handle shmem too?

Quoted 60 lines

fs/proc/task_mmu.c +1 -0

+	seq_printf(m, "THPeligible:    %8u\n", thp_eligible);
 
 	if (arch_pkeys_enabled())
 		seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));

include/linux/huge_mm.h +11 -0

index 212cca384d7e..f87136f38aa1 100644
@@ -267,6 +267,10 @@ unsigned long thp_vma_allowable_orders(struct vm_area_struct *vma,
 	return __thp_vma_allowable_orders(vma, vm_flags, tva_flags, orders);
 }
 
+unsigned long shmem_allowable_huge_orders(struct inode *inode,
+				struct vm_area_struct *vma, pgoff_t index,
+				bool global_huge);
+
 struct thpsize {
 	struct kobject kobj;
 	struct list_head node;

@@ -460,6 +464,13 @@ static inline unsigned long thp_vma_allowable_orders(struct vm_area_struct *vma,
 	return 0;
 }
 
+static inline unsigned long shmem_allowable_huge_orders(struct inode *inode,
+				struct vm_area_struct *vma, pgoff_t index,
+				bool global_huge)
+{
+	return 0;
+}
+
 #define transparent_hugepage_flags 0UL
 
 #define thp_get_unmapped_area	NULL

mm/shmem.c +1 -8

index d495c0701a83..aa85df9c662a 100644
@@ -1622,7 +1622,7 @@ static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
 }
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-static unsigned long shmem_allowable_huge_orders(struct inode *inode,
+unsigned long shmem_allowable_huge_orders(struct inode *inode,
 				struct vm_area_struct *vma, pgoff_t index,
 				bool global_huge)
 {

@@ -1707,13 +1707,6 @@ static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault
 	return orders;
 }
 #else
-static unsigned long shmem_allowable_huge_orders(struct inode *inode,
-				struct vm_area_struct *vma, pgoff_t index,
-				bool global_huge)
-{
-	return 0;
-}
-
 static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault *vmf,
 					   struct address_space *mapping, pgoff_t index,
 					   unsigned long orders)

Baolin Wang

reviewer

2024-07-01 08:33 UTC #8 lore raw

Quoted 33 lines · Ryan Roberts

Quoted 32 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
  fs/proc/task_mmu.c      | 10 +++++++---
  include/linux/huge_mm.h | 11 +++++++++++
  mm/shmem.c              |  9 +--------
  3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +1 -0
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
  {
  	struct vm_area_struct *vma = v;
  	struct mem_size_stats mss = {};
+	bool thp_eligible;
 
  	smap_gather_stats(vma, &mss, 0);
 
@@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void *v)
 
  	__show_smap(m, &mss, false);
 

Quoted 11 lines

fs/proc/task_mmu.c static int show_smap(struct seq_file *m, void *v) · line 890
Quoted 8 lines
fs/proc/task_mmu.c +5 -3
-	seq_printf(m, "THPeligible:    %8u\n",
-		   !!thp_vma_allowable_orders(vma, vma->vm_flags,
-			   TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
+	thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
+						TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL);
+	if (vma_is_anon_shmem(vma))
+		thp_eligible = !!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+							vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I would have liked, but is there a reason why we need a separate function for shmem?

Since shmem_allowable_huge_orders() only uses shmem specific logic to determine if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.

IIUC, I agree with David's suggestion that the shmem_allowable_huge_orders() helper function could be used in thp_vma_allowable_orders() to support shmem mTHP. Something like:

mm/huge_memory.c +7 -4

index c7ce28f6b7f3..9677fe6cf478 100644
@@ -151,10 +151,13 @@ unsigned long __thp_vma_allowable_orders(struct vm_area_struct *vma,
         * Must be done before hugepage flags check since shmem has its
         * own flags.
         */
-       if (!in_pf && shmem_file(vma->vm_file))
-               return shmem_is_huge(file_inode(vma->vm_file), vma->vm_pgoff,
-                                    !enforce_sysfs, vma->vm_mm, vm_flags)
-                       ? orders : 0;
+       if (!in_pf && shmem_file(vma->vm_file)) {
+               bool global_huge = shmem_is_huge(file_inode(vma->vm_file), vma->vm_pgoff,
+                                    !enforce_sysfs, vma->vm_mm, vm_flags);
+
+               return shmem_allowable_huge_orders(file_inode(vma->vm_file),
+                                       vma, vma->vm_pgoff, global_huge);
+       }
 
        if (!vma_is_anonymous(vma)) {
                /*

Quoted 62 lines

Couldn't (shouldn't) thp_vma_allowable_orders() be taught to handle shmem too?

Quoted 60 lines

fs/proc/task_mmu.c +1 -0

+	seq_printf(m, "THPeligible:    %8u\n", thp_eligible);
 
  	if (arch_pkeys_enabled())
  		seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));

include/linux/huge_mm.h +11 -0

index 212cca384d7e..f87136f38aa1 100644
@@ -267,6 +267,10 @@ unsigned long thp_vma_allowable_orders(struct vm_area_struct *vma,
  	return __thp_vma_allowable_orders(vma, vm_flags, tva_flags, orders);
  }
 
+unsigned long shmem_allowable_huge_orders(struct inode *inode,
+				struct vm_area_struct *vma, pgoff_t index,
+				bool global_huge);
+
  struct thpsize {
  	struct kobject kobj;
  	struct list_head node;

@@ -460,6 +464,13 @@ static inline unsigned long thp_vma_allowable_orders(struct vm_area_struct *vma,
  	return 0;
  }
 
+static inline unsigned long shmem_allowable_huge_orders(struct inode *inode,
+				struct vm_area_struct *vma, pgoff_t index,
+				bool global_huge)
+{
+	return 0;
+}
+
  #define transparent_hugepage_flags 0UL
 
  #define thp_get_unmapped_area	NULL

mm/shmem.c +1 -8

index d495c0701a83..aa85df9c662a 100644
@@ -1622,7 +1622,7 @@ static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
  }
 
  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-static unsigned long shmem_allowable_huge_orders(struct inode *inode,
+unsigned long shmem_allowable_huge_orders(struct inode *inode,
  				struct vm_area_struct *vma, pgoff_t index,
  				bool global_huge)
  {

@@ -1707,13 +1707,6 @@ static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault
  	return orders;
  }
  #else
-static unsigned long shmem_allowable_huge_orders(struct inode *inode,
-				struct vm_area_struct *vma, pgoff_t index,
-				bool global_huge)
-{
-	return 0;
-}
-
  static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault *vmf,
  					   struct address_space *mapping, pgoff_t index,
  					   unsigned long orders)

Ryan Roberts

reviewer

2024-07-01 08:40 UTC #9 lore raw

Quoted 44 lines · Baolin Wang

Quoted 32 lines · Ryan Roberts
Quoted 31 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
fs/proc/task_mmu.c | 10 +++++++---
include/linux/huge_mm.h | 11 +++++++++++
mm/shmem.c | 9 +--------
  3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +1 -0
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
   {
       struct vm_area_struct *vma = v;
       struct mem_size_stats mss = {};
+    bool thp_eligible;
         smap_gather_stats(vma, &mss, 0);
   @@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void *v)
         __show_smap(m, &mss, false);
- seq_printf(m, "THPeligible: %8u\n",
- !!thp_vma_allowable_orders(vma, vma->vm_flags, - TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
Quoted 8 lines
Quoted 5 lines
+ thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags, + TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL); + if (vma_is_anon_shmem(vma))
+ thp_eligible = !!shmem_allowable_huge_orders(file_inode(vma->vm_file), + vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I would have liked, but is there a reason why we need a separate function for shmem?

Quoted 5 lines

Since shmem_allowable_huge_orders() only uses shmem specific logic to determine if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.

My point was really that a single interface (thp_vma_allowable_orders) should be used to get this information. I have no strong opinon on how the implementation of that interface looks. What you suggest below seems perfectly reasonable to me.

Quoted 4 lines

IIUC, I agree with David's suggestion that the shmem_allowable_huge_orders() helper function could be used in thp_vma_allowable_orders() to support shmem mTHP. Something like:

I hadn't seen David's suggestion until after I sent my mail. But I think we are both suggesting the same thing.

Quoted 88 lines

mm/huge_memory.c +7 -4

index c7ce28f6b7f3..9677fe6cf478 100644
@@ -151,10 +151,13 @@ unsigned long __thp_vma_allowable_orders(struct
 vm_area_struct *vma,
          * Must be done before hugepage flags check since shmem has its
          * own flags.
          */
-       if (!in_pf && shmem_file(vma->vm_file))
-               return shmem_is_huge(file_inode(vma->vm_file), vma->vm_pgoff,
-                                    !enforce_sysfs, vma->vm_mm, vm_flags)
-                       ? orders : 0;
+       if (!in_pf && shmem_file(vma->vm_file)) {
+               bool global_huge = shmem_is_huge(file_inode(vma->vm_file),
 vma->vm_pgoff,
+                                    !enforce_sysfs, vma->vm_mm, vm_flags);
+
+               return shmem_allowable_huge_orders(file_inode(vma->vm_file),
+                                       vma, vma->vm_pgoff, global_huge);
+       }

if (!vma_is_anonymous(vma)) {
/*

Quoted 62 lines

Couldn't (shouldn't) thp_vma_allowable_orders() be taught to handle shmem too?

Quoted 60 lines

+ seq_printf(m, "THPeligible: %8u\n", thp_eligible); if (arch_pkeys_enabled())
seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma));

include/linux/huge_mm.h +3 -0

index 212cca384d7e..f87136f38aa1 100644
@@ -267,6 +267,10 @@ unsigned long thp_vma_allowable_orders(struct
 vm_area_struct *vma,
       return __thp_vma_allowable_orders(vma, vm_flags, tva_flags, orders);
   }
   +unsigned long shmem_allowable_huge_orders(struct inode *inode,
+                struct vm_area_struct *vma, pgoff_t index,
+                bool global_huge);
+
   struct thpsize {
       struct kobject kobj;

struct list_head node;

include/linux/huge_mm.h +6 -0

@@ -460,6 +464,13 @@ static inline unsigned long
 thp_vma_allowable_orders(struct vm_area_struct *vma,
       return 0;
   }
   +static inline unsigned long shmem_allowable_huge_orders(struct inode *inode,
+                struct vm_area_struct *vma, pgoff_t index,
+                bool global_huge)
+{
+    return 0;
+}
+
   #define transparent_hugepage_flags 0UL
     #define thp_get_unmapped_area    NULL

mm/shmem.c +1 -1

index d495c0701a83..aa85df9c662a 100644
@@ -1622,7 +1622,7 @@ static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t
 limit_gfp)
   }
     #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-static unsigned long shmem_allowable_huge_orders(struct inode *inode,
+unsigned long shmem_allowable_huge_orders(struct inode *inode,
                   struct vm_area_struct *vma, pgoff_t index,
                   bool global_huge)
   {

@@ -1707,13 +1707,6 @@ static unsigned long shmem_suitable_orders(struct

inode *inode, struct vm_fault
return orders;
}
#else

mm/shmem.c +0 -7

-static unsigned long shmem_allowable_huge_orders(struct inode *inode,
-                struct vm_area_struct *vma, pgoff_t index,
-                bool global_huge)
-{
-    return 0;
-}
-

static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault *vmf,
struct address_space *mapping, pgoff_t index, unsigned long orders)

Baolin Wang

reviewer

2024-07-01 08:46 UTC #10 lore raw

Quoted 56 lines · Ryan Roberts

Quoted 44 lines · Baolin Wang
Quoted 32 lines · Ryan Roberts
Quoted 31 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
   fs/proc/task_mmu.c      | 10 +++++++---
   include/linux/huge_mm.h | 11 +++++++++++
   mm/shmem.c              |  9 +--------
   3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +1 -0
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
   {
       struct vm_area_struct *vma = v;
       struct mem_size_stats mss = {};
+    bool thp_eligible;
         smap_gather_stats(vma, &mss, 0);
   @@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void *v)
         __show_smap(m, &mss, false);
- seq_printf(m, "THPeligible: %8u\n",
- !!thp_vma_allowable_orders(vma, vma->vm_flags, - TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
Quoted 8 lines
Quoted 5 lines
+ thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags, + TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL); + if (vma_is_anon_shmem(vma))
+ thp_eligible = !!shmem_allowable_huge_orders(file_inode(vma->vm_file), + vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I would have liked, but is there a reason why we need a separate function for shmem?
Quoted 5 lines
Since shmem_allowable_huge_orders() only uses shmem specific logic to determine if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.
My point was really that a single interface (thp_vma_allowable_orders) should be used to get this information. I have no strong opinon on how the implementation of that interface looks. What you suggest below seems perfectly reasonable to me.
Quoted 1 line

Quoted 5 lines

Quoted 3 lines
IIUC, I agree with David's suggestion that the shmem_allowable_huge_orders() helper function could be used in thp_vma_allowable_orders() to support shmem mTHP. Something like:
I hadn't seen David's suggestion until after I sent my mail. But I think we are > both suggesting the same thing.

OK. Great :)

David Hildenbrand

2024-07-01 08:48 UTC #11 lore raw

In reply to Ryan Roberts

Quoted 51 lines · Ryan Roberts

Quoted 44 lines · Baolin Wang
Quoted 32 lines · Ryan Roberts
Quoted 31 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
   fs/proc/task_mmu.c      | 10 +++++++---
   include/linux/huge_mm.h | 11 +++++++++++
   mm/shmem.c              |  9 +--------
   3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +1 -0
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
   {
       struct vm_area_struct *vma = v;
       struct mem_size_stats mss = {};
+    bool thp_eligible;
         smap_gather_stats(vma, &mss, 0);
   @@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void *v)
         __show_smap(m, &mss, false);
- seq_printf(m, "THPeligible: %8u\n",
- !!thp_vma_allowable_orders(vma, vma->vm_flags, - TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
Quoted 8 lines
Quoted 5 lines
+ thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags, + TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL); + if (vma_is_anon_shmem(vma))
+ thp_eligible = !!shmem_allowable_huge_orders(file_inode(vma->vm_file), + vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I would have liked, but is there a reason why we need a separate function for shmem?
Quoted 5 lines
Since shmem_allowable_huge_orders() only uses shmem specific logic to determine if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.

Quoted 3 lines

My point was really that a single interface (thp_vma_allowable_orders) should be used to get this information. I have no strong opinon on how the implementation of that interface looks. What you suggest below seems perfectly reasonable to me.

Right. thp_vma_allowable_orders() might require some care as discussed in other context (cleanly separate dax and shmem handling/orders). But that would be follow-up cleanups.

Signature · 3 lines

Cheers,

David / dhildenb

Ryan Roberts

reviewer

2024-07-01 08:51 UTC #12 lore raw

Quoted 53 lines · David Hildenbrand

Quoted 52 lines · Ryan Roberts
Quoted 50 lines · Baolin Wang
Quoted 33 lines · Ryan Roberts
Quoted 32 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
   fs/proc/task_mmu.c      | 10 +++++++---
   include/linux/huge_mm.h | 11 +++++++++++
   mm/shmem.c              | 9 +--------
   3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +1 -0
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
    {
        struct vm_area_struct *vma = v;
        struct mem_size_stats mss = {};
+    bool thp_eligible;
          smap_gather_stats(vma, &mss, 0);
    @@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void *v)
          __show_smap(m, &mss, false);
- seq_printf(m, "THPeligible: %8u\n",
- !!thp_vma_allowable_orders(vma, vma->vm_flags, - TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL)); + thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
Quoted 8 lines
Quoted 5 lines
+                        TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL); +    if (vma_is_anon_shmem(vma))
+        thp_eligible = !!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+                            vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I would have liked, but is there a reason why we need a separate function for shmem?
Since shmem_allowable_huge_orders() only uses shmem specific logic to determine if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.

Quoted 7 lines

Quoted 3 lines
My point was really that a single interface (thp_vma_allowable_orders) should be used to get this information. I have no strong opinon on how the implementation of that interface looks. What you suggest below seems perfectly reasonable to me.
Right. thp_vma_allowable_orders() might require some care as discussed in other context (cleanly separate dax and shmem handling/orders). But that would be follow-up cleanups.

Are you planning to do that, or do you want me to send a patch?

David Hildenbrand

2024-07-01 08:57 UTC #13 lore raw

Quoted 58 lines · Ryan Roberts

Quoted 57 lines · David Hildenbrand
Quoted 55 lines · Ryan Roberts
Quoted 45 lines · Baolin Wang
Quoted 33 lines · Ryan Roberts
Quoted 32 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
    fs/proc/task_mmu.c      | 10 +++++++---
    include/linux/huge_mm.h | 11 +++++++++++
    mm/shmem.c              |  9 +--------
    3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +1 -0
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
    {
        struct vm_area_struct *vma = v;
        struct mem_size_stats mss = {};
+    bool thp_eligible;
          smap_gather_stats(vma, &mss, 0);
    @@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void *v)
          __show_smap(m, &mss, false);
- seq_printf(m, "THPeligible: %8u\n",
- !!thp_vma_allowable_orders(vma, vma->vm_flags, - TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL)); + thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
Quoted 8 lines
Quoted 5 lines
+                        TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL); +    if (vma_is_anon_shmem(vma))
+        thp_eligible = !!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+                            vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I would have liked, but is there a reason why we need a separate function for shmem?
Quoted 5 lines
Since shmem_allowable_huge_orders() only uses shmem specific logic to determine if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.
My point was really that a single interface (thp_vma_allowable_orders) should be used to get this information. I have no strong opinon on how the implementation of that interface looks. What you suggest below seems perfectly reasonable to me.

Quoted 5 lines

Quoted 3 lines
Right. thp_vma_allowable_orders() might require some care as discussed in other context (cleanly separate dax and shmem handling/orders). But that would be follow-up cleanups.
Are you planning to do that, or do you want me to send a patch?

I'm planning on looking into some details, especially the interaction with large folios in the pagecache. I'll let you know once I have a better idea what actually should be done :)

Signature · 3 lines

Cheers,

David / dhildenb

Ryan Roberts

reviewer

2024-07-01 09:14 UTC #14 lore raw

Quoted 67 lines · David Hildenbrand

Quoted 66 lines · Ryan Roberts
Quoted 64 lines · David Hildenbrand
Quoted 54 lines · Ryan Roberts
Quoted 46 lines · Baolin Wang
Quoted 34 lines · Ryan Roberts
Quoted 33 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
    fs/proc/task_mmu.c      | 10 +++++++---
    include/linux/huge_mm.h | 11 +++++++++++
    mm/shmem.c              | 9 +--------
    3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +1 -0
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
     {
         struct vm_area_struct *vma = v;
         struct mem_size_stats mss = {};
+    bool thp_eligible;
           smap_gather_stats(vma, &mss, 0);
     @@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void *v)
           __show_smap(m, &mss, false);
    -    seq_printf(m, "THPeligible:    %8u\n",
-           !!thp_vma_allowable_orders(vma, vma->vm_flags,
-               TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL)); +    thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags, +                        TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL);
Quoted 8 lines
Quoted 4 lines
+ if (vma_is_anon_shmem(vma))
+ thp_eligible = !!shmem_allowable_huge_orders(file_inode(vma->vm_file), + vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I would
have liked, but is there a reason why we need a separate function for shmem?
Quoted 6 lines
Since shmem_allowable_huge_orders() only uses shmem specific logic to determine
if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.
Quoted 5 lines
My point was really that a single interface (thp_vma_allowable_orders) should be
used to get this information. I have no strong opinon on how the implementation of that interface looks. What you suggest below seems perfectly reasonable to me.
Right. thp_vma_allowable_orders() might require some care as discussed in other context (cleanly separate dax and shmem handling/orders). But that would be follow-up cleanups.

Quoted 5 lines

Quoted 1 line
Are you planning to do that, or do you want me to send a patch?
I'm planning on looking into some details, especially the interaction with large folios in the pagecache. I'll let you know once I have a better idea what actually should be done :)

OK great - I'll scrub it from my todo list... really getting things done today :)

David Hildenbrand

2024-07-01 09:17 UTC #15 lore raw

Quoted 68 lines · Ryan Roberts

Quoted 67 lines · David Hildenbrand
Quoted 66 lines · Ryan Roberts
Quoted 64 lines · David Hildenbrand
Quoted 54 lines · Ryan Roberts
Quoted 46 lines · Baolin Wang
Quoted 34 lines · Ryan Roberts
Quoted 33 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
     fs/proc/task_mmu.c      | 10 +++++++---
     include/linux/huge_mm.h | 11 +++++++++++
     mm/shmem.c              |  9 +--------
     3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +1 -0
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
     {
         struct vm_area_struct *vma = v;
         struct mem_size_stats mss = {};
+    bool thp_eligible;
           smap_gather_stats(vma, &mss, 0);
     @@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void *v)
           __show_smap(m, &mss, false);
    -    seq_printf(m, "THPeligible:    %8u\n",
-           !!thp_vma_allowable_orders(vma, vma->vm_flags,
-               TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL)); +    thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags, +                        TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL);
Quoted 8 lines
Quoted 4 lines
+ if (vma_is_anon_shmem(vma))
+ thp_eligible = !!shmem_allowable_huge_orders(file_inode(vma->vm_file), + vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I would
have liked, but is there a reason why we need a separate function for shmem?
Quoted 6 lines
Since shmem_allowable_huge_orders() only uses shmem specific logic to determine
if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.
Quoted 5 lines
My point was really that a single interface (thp_vma_allowable_orders) should be
used to get this information. I have no strong opinon on how the implementation of that interface looks. What you suggest below seems perfectly reasonable to me.
Right. thp_vma_allowable_orders() might require some care as discussed in other context (cleanly separate dax and shmem handling/orders). But that would be follow-up cleanups.

Quoted 7 lines

Quoted 5 lines
Quoted 1 line
Are you planning to do that, or do you want me to send a patch?
I'm planning on looking into some details, especially the interaction with large folios in the pagecache. I'll let you know once I have a better idea what actually should be done :)
OK great - I'll scrub it from my todo list... really getting things done today :)

Resolved the khugepaged thiny already? :P

[khugepaged not active when only enabling the sub-size via the 2M folder IIRC]

Signature · 3 lines

Cheers,

David / dhildenb

Ryan Roberts

reviewer

2024-07-01 10:16 UTC #16 lore raw

Quoted 80 lines · David Hildenbrand

Quoted 79 lines · Ryan Roberts
Quoted 77 lines · David Hildenbrand
Quoted 66 lines · Ryan Roberts
Quoted 65 lines · David Hildenbrand
Quoted 63 lines · Ryan Roberts
Quoted 48 lines · Baolin Wang
Quoted 36 lines · Ryan Roberts
Quoted 35 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
     fs/proc/task_mmu.c      | 10 +++++++---
     include/linux/huge_mm.h | 11 +++++++++++
     mm/shmem.c              | 9 +--------
     3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +1 -0
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
      {
          struct vm_area_struct *vma = v;
          struct mem_size_stats mss = {};
+    bool thp_eligible;
            smap_gather_stats(vma, &mss, 0);
     @@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void *v)
           __show_smap(m, &mss, false);
     -    seq_printf(m, "THPeligible:    %8u\n",
-           !!thp_vma_allowable_orders(vma, vma->vm_flags,
-               TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL)); +    thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags, +                        TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL); +    if (vma_is_anon_shmem(vma))
Quoted 8 lines
Quoted 3 lines
+ thp_eligible = !!shmem_allowable_huge_orders(file_inode(vma->vm_file), + vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I would
have liked, but is there a reason why we need a separate function for shmem?
Quoted 7 lines
Since shmem_allowable_huge_orders() only uses shmem specific logic to determine
if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making
it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.
My point was really that a single interface (thp_vma_allowable_orders) should be
used to get this information. I have no strong opinon on how the implementation
of that interface looks. What you suggest below seems perfectly reasonable to me.
Quoted 6 lines
Quoted 4 lines
Right. thp_vma_allowable_orders() might require some care as discussed in other
context (cleanly separate dax and shmem handling/orders). But that would be follow-up cleanups.
Are you planning to do that, or do you want me to send a patch?
I'm planning on looking into some details, especially the interaction with large folios in the pagecache. I'll let you know once I have a better idea what actually should be done :)

Quoted 5 lines

Quoted 1 line
OK great - I'll scrub it from my todo list... really getting things done today :)
Resolved the khugepaged thiny already? :P
[khugepaged not active when only enabling the sub-size via the 2M folder IIRC]

Hmm... baby brain?

Sorry about that. I've been a bit useless lately. For some reason it wasn't on my list, but its there now. Will prioritise it, because I agree it's not good.

David Hildenbrand

2024-07-01 10:22 UTC #17 lore raw

Quoted 81 lines · Ryan Roberts

Quoted 80 lines · David Hildenbrand
Quoted 79 lines · Ryan Roberts
Quoted 77 lines · David Hildenbrand
Quoted 66 lines · Ryan Roberts
Quoted 65 lines · David Hildenbrand
Quoted 63 lines · Ryan Roberts
Quoted 48 lines · Baolin Wang
Quoted 36 lines · Ryan Roberts
Quoted 35 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
      fs/proc/task_mmu.c      | 10 +++++++---
      include/linux/huge_mm.h | 11 +++++++++++
      mm/shmem.c              |  9 +--------
      3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +1 -0
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
      {
          struct vm_area_struct *vma = v;
          struct mem_size_stats mss = {};
+    bool thp_eligible;
            smap_gather_stats(vma, &mss, 0);
      @@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void
 *v)
           __show_smap(m, &mss, false);
     -    seq_printf(m, "THPeligible:    %8u\n",
-           !!thp_vma_allowable_orders(vma, vma->vm_flags,
-               TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL)); +    thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags, +                        TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL); +    if (vma_is_anon_shmem(vma))
Quoted 8 lines
Quoted 3 lines
+ thp_eligible = !!shmem_allowable_huge_orders(file_inode(vma->vm_file), + vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I would
have liked, but is there a reason why we need a separate function for shmem?
Quoted 7 lines
Since shmem_allowable_huge_orders() only uses shmem specific logic to determine
if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making
it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.
My point was really that a single interface (thp_vma_allowable_orders) should be
used to get this information. I have no strong opinon on how the implementation
of that interface looks. What you suggest below seems perfectly reasonable to me.
Quoted 6 lines
Quoted 4 lines
Right. thp_vma_allowable_orders() might require some care as discussed in other
context (cleanly separate dax and shmem handling/orders). But that would be follow-up cleanups.
Are you planning to do that, or do you want me to send a patch?
I'm planning on looking into some details, especially the interaction with large folios in the pagecache. I'll let you know once I have a better idea what actually should be done :)

Quoted 7 lines

Quoted 5 lines
Quoted 1 line
OK great - I'll scrub it from my todo list... really getting things done today :)
Resolved the khugepaged thiny already? :P
[khugepaged not active when only enabling the sub-size via the 2M folder IIRC]
Hmm... baby brain?

I think I only mentioned it in a private mail at some point.

Quoted 3 lines

Sorry about that. I've been a bit useless lately. For some reason it wasn't on my list, but its there now. Will prioritise it, because I agree it's not good.

IIRC, if you do

echo never > /sys/kernel/mm/transparent_hugepage/enabled
echo always > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled

khugepaged will not get activated.

Signature · 3 lines

Cheers,

David / dhildenb

Yang Shi

2024-07-01 18:20 UTC #18 lore raw

Quoted 100 lines · David Hildenbrand

Quoted 81 lines · Ryan Roberts
Quoted 80 lines · David Hildenbrand
Quoted 79 lines · Ryan Roberts
Quoted 77 lines · David Hildenbrand
Quoted 66 lines · Ryan Roberts
Quoted 65 lines · David Hildenbrand
Quoted 63 lines · Ryan Roberts
Quoted 48 lines · Baolin Wang
Quoted 36 lines · Ryan Roberts
Quoted 35 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
      fs/proc/task_mmu.c      | 10 +++++++---
      include/linux/huge_mm.h | 11 +++++++++++
      mm/shmem.c              |  9 +--------
      3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +1 -0
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
      {
          struct vm_area_struct *vma = v;
          struct mem_size_stats mss = {};
+    bool thp_eligible;
            smap_gather_stats(vma, &mss, 0);
      @@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void
 *v)
__show_smap(m, &mss, false);
- seq_printf(m, "THPeligible: %8u\n",
fs/proc/task_mmu.c +3 -2
-           !!thp_vma_allowable_orders(vma, vma->vm_flags,
-               TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
+    thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
+                        TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL);
+    if (vma_is_anon_shmem(vma))
Quoted 8 lines
Quoted 3 lines
+ thp_eligible =
!!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+ vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I would
have liked, but is there a reason why we need a separate function for shmem?
Quoted 7 lines
Since shmem_allowable_huge_orders() only uses shmem specific logic to determine
if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making
it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.
My point was really that a single interface (thp_vma_allowable_orders) should be
used to get this information. I have no strong opinon on how the implementation
of that interface looks. What you suggest below seems perfectly reasonable to me.
Quoted 6 lines
Quoted 4 lines
Right. thp_vma_allowable_orders() might require some care as discussed in other
context (cleanly separate dax and shmem handling/orders). But that would be follow-up cleanups.
Are you planning to do that, or do you want me to send a patch?
I'm planning on looking into some details, especially the interaction with large folios in the pagecache. I'll let you know once I have a better idea what actually should be done :)
Quoted 7 lines
Quoted 5 lines
Quoted 1 line
OK great - I'll scrub it from my todo list... really getting things done today :)
Resolved the khugepaged thiny already? :P
[khugepaged not active when only enabling the sub-size via the 2M folder IIRC]
Hmm... baby brain?
:)
I think I only mentioned it in a private mail at some point.
Quoted 3 lines
Sorry about that. I've been a bit useless lately. For some reason it wasn't on my list, but its there now. Will prioritise it, because I agree it's not good.

Quoted 6 lines

IIRC, if you do
echo never > /sys/kernel/mm/transparent_hugepage/enabled
echo always > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled
khugepaged will not get activated.

khugepaged is controlled by the top level knob. But the above setting sounds confusing, can we disable the top level knob, but enable it on a per-order basis? TBH, it sounds weird and doesn't make too much sense to me.

Quoted 7 lines

--
Cheers,
David / dhildenb

Ryan Roberts

reviewer

2024-07-02 08:24 UTC #19 lore raw

Quoted 101 lines · Yang Shi

Quoted 100 lines · David Hildenbrand
Quoted 81 lines · Ryan Roberts
Quoted 80 lines · David Hildenbrand
Quoted 79 lines · Ryan Roberts
Quoted 77 lines · David Hildenbrand
Quoted 66 lines · Ryan Roberts
Quoted 65 lines · David Hildenbrand
Quoted 63 lines · Ryan Roberts
Quoted 48 lines · Baolin Wang
Quoted 36 lines · Ryan Roberts
Quoted 35 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
      fs/proc/task_mmu.c      | 10 +++++++---
      include/linux/huge_mm.h | 11 +++++++++++
      mm/shmem.c              |  9 +--------
      3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +1 -0
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
      {
          struct vm_area_struct *vma = v;
          struct mem_size_stats mss = {};
+    bool thp_eligible;
            smap_gather_stats(vma, &mss, 0);
      @@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void
 *v)
__show_smap(m, &mss, false);
- seq_printf(m, "THPeligible: %8u\n",
fs/proc/task_mmu.c +3 -2
-           !!thp_vma_allowable_orders(vma, vma->vm_flags,
-               TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
+    thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
+                        TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL);
+    if (vma_is_anon_shmem(vma))
Quoted 8 lines
Quoted 3 lines
+ thp_eligible =
!!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+ vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I would
have liked, but is there a reason why we need a separate function for shmem?
Quoted 7 lines
Since shmem_allowable_huge_orders() only uses shmem specific logic to determine
if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making
it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.
My point was really that a single interface (thp_vma_allowable_orders) should be
used to get this information. I have no strong opinon on how the implementation
of that interface looks. What you suggest below seems perfectly reasonable to me.
Quoted 6 lines
Quoted 4 lines
Right. thp_vma_allowable_orders() might require some care as discussed in other
context (cleanly separate dax and shmem handling/orders). But that would be follow-up cleanups.
Are you planning to do that, or do you want me to send a patch?
I'm planning on looking into some details, especially the interaction with large folios in the pagecache. I'll let you know once I have a better idea what actually should be done :)
Quoted 7 lines
Quoted 5 lines
Quoted 1 line
OK great - I'll scrub it from my todo list... really getting things done today :)
Resolved the khugepaged thiny already? :P
[khugepaged not active when only enabling the sub-size via the 2M folder IIRC]
Hmm... baby brain?
:)
I think I only mentioned it in a private mail at some point.
Quoted 3 lines
Sorry about that. I've been a bit useless lately. For some reason it wasn't on my list, but its there now. Will prioritise it, because I agree it's not good.

Quoted 8 lines

Quoted 6 lines
IIRC, if you do
echo never > /sys/kernel/mm/transparent_hugepage/enabled
echo always > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled
khugepaged will not get activated.
khugepaged is controlled by the top level knob.

What do you mean by "top level knob"? I assume
/sys/kernel/mm/transparent_hugepage/enabled ?

If so, that's not really a thing in its own right; its just the legacy PMD-size THP control, and we only take any notice of it if a per-size control is set to "inherit". So if we have:

# echo always > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled

Then by design, /sys/kernel/mm/transparent_hugepage/enabled should be ignored.

Quoted 4 lines

But the above setting
sounds confusing, can we disable the top level knob, but enable it on a per-order basis? TBH, it sounds weird and doesn't make too much sense to me.

Well that's the design and that's how its documented. It's done this way for back-compat. All controls are now per-size. But at boot, we default all per-size controls to "never" except for the PMD-sized control, which is defaulted to "inherit". That way, an unenlightened user-space can still control PMD-sized THP via the legacy (top-level) control. But enlightened apps can directly control per-size.

I'm not sure how your way would work, because you would have 2 controls competing to do the same thing?

Quoted 8 lines

Quoted 8 lines
--
Cheers,
David / dhildenb

David Hildenbrand

2024-07-02 08:28 UTC #20 lore raw

Quoted 109 lines · Ryan Roberts

Quoted 108 lines · Yang Shi
Quoted 106 lines · David Hildenbrand
Quoted 81 lines · Ryan Roberts
Quoted 80 lines · David Hildenbrand
Quoted 79 lines · Ryan Roberts
Quoted 77 lines · David Hildenbrand
Quoted 66 lines · Ryan Roberts
Quoted 65 lines · David Hildenbrand
Quoted 63 lines · Ryan Roberts
Quoted 48 lines · Baolin Wang
Quoted 36 lines · Ryan Roberts
Quoted 35 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for
anonymous shmem"), we can configure different policies through
the multi-size THP sysfs interface for anonymous shmem. But
currently "THPeligible" indicates only whether the mapping is
eligible for allocating THP-pages as well as the THP is PMD
mappable or not for anonymous shmem, we need to support semantics
for mTHP with anonymous shmem similar to those for mTHP with
anonymous memory.
Signed-off-by: Bang Li <[email protected]>
---
fs/proc/task_mmu.c | 10 +++++++---
include/linux/huge_mm.h | 11 +++++++++++
mm/shmem.c | 9 +--------
3 files changed, 19 insertions(+), 11 deletions(-)
diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
index 93fb2c61b154..09b5db356886 100644
--- a/fs/proc/task_mmu.c
+++ b/fs/proc/task_mmu.c
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
{
struct vm_area_struct *vma = v;
struct mem_size_stats mss = {};
+ bool thp_eligible;
smap_gather_stats(vma, &mss, 0);
@@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void
*v)
__show_smap(m, &mss, false);
- seq_printf(m, "THPeligible: %8u\n",
- !!thp_vma_allowable_orders(vma, vma->vm_flags,
- TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
+ thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
+ TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL);
+ if (vma_is_anon_shmem(vma))
Quoted 8 lines
Quoted 3 lines
+ thp_eligible =
!!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+ vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I would
have liked, but is there a reason why we need a separate function for shmem?
Quoted 7 lines
Since shmem_allowable_huge_orders() only uses shmem specific logic to determine
if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making
it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.
My point was really that a single interface (thp_vma_allowable_orders) should be
used to get this information. I have no strong opinon on how the implementation
of that interface looks. What you suggest below seems perfectly reasonable to me.
Quoted 6 lines
Quoted 4 lines
Right. thp_vma_allowable_orders() might require some care as discussed in other
context (cleanly separate dax and shmem handling/orders). But that would be follow-up cleanups.
Are you planning to do that, or do you want me to send a patch?
I'm planning on looking into some details, especially the interaction with large folios in the pagecache. I'll let you know once I have a better idea what actually should be done :)
Quoted 7 lines
Quoted 5 lines
Quoted 1 line
OK great - I'll scrub it from my todo list... really getting things done today :)
Resolved the khugepaged thiny already? :P
[khugepaged not active when only enabling the sub-size via the 2M folder IIRC]
Hmm... baby brain?
:)
I think I only mentioned it in a private mail at some point.
Quoted 3 lines
Sorry about that. I've been a bit useless lately. For some reason it wasn't on my list, but its there now. Will prioritise it, because I agree it's not good.
IIRC, if you do
echo never > /sys/kernel/mm/transparent_hugepage/enabled
echo always > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled
khugepaged will not get activated.

Quoted 8 lines

Quoted 1 line
khugepaged is controlled by the top level knob.
What do you mean by "top level knob"? I assume
/sys/kernel/mm/transparent_hugepage/enabled ?
If so, that's not really a thing in its own right; its just the legacy PMD-size THP control, and we only take any notice of it if a per-size control is set to "inherit". So if we have:

In a simpler world, where "enabled" would have been a boolean (true / false), we could have made it a universal killswitch that is AND'ed with the other ones.

Unfortunately, we don't live in such a simple world.

Signature · 3 lines

Cheers,

David / dhildenb

Yang Shi

2024-07-03 16:08 UTC #21 lore raw

In reply to Ryan Roberts

Quoted 108 lines · Ryan Roberts

Quoted 106 lines · Yang Shi
Quoted 105 lines · David Hildenbrand
Quoted 81 lines · Ryan Roberts
Quoted 80 lines · David Hildenbrand
Quoted 79 lines · Ryan Roberts
Quoted 77 lines · David Hildenbrand
Quoted 66 lines · Ryan Roberts
Quoted 65 lines · David Hildenbrand
Quoted 63 lines · Ryan Roberts
Quoted 48 lines · Baolin Wang
Quoted 36 lines · Ryan Roberts
Quoted 35 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for
anonymous shmem"), we can configure different policies through
the multi-size THP sysfs interface for anonymous shmem. But
currently "THPeligible" indicates only whether the mapping is
eligible for allocating THP-pages as well as the THP is PMD
mappable or not for anonymous shmem, we need to support semantics
for mTHP with anonymous shmem similar to those for mTHP with
anonymous memory.
Signed-off-by: Bang Li <[email protected]>
---
fs/proc/task_mmu.c | 10 +++++++---
include/linux/huge_mm.h | 11 +++++++++++
mm/shmem.c | 9 +--------
3 files changed, 19 insertions(+), 11 deletions(-)
diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
index 93fb2c61b154..09b5db356886 100644
--- a/fs/proc/task_mmu.c
+++ b/fs/proc/task_mmu.c
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
{
struct vm_area_struct *vma = v;
struct mem_size_stats mss = {};
+ bool thp_eligible;
smap_gather_stats(vma, &mss, 0);
@@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void
*v)
__show_smap(m, &mss, false);
- seq_printf(m, "THPeligible: %8u\n",
- !!thp_vma_allowable_orders(vma, vma->vm_flags,
- TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
+ thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
+ TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL);
+ if (vma_is_anon_shmem(vma))
Quoted 8 lines
Quoted 3 lines
+ thp_eligible =
!!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+ vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I would
have liked, but is there a reason why we need a separate function for shmem?
Quoted 7 lines
Since shmem_allowable_huge_orders() only uses shmem specific logic to determine
if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making
it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.
My point was really that a single interface (thp_vma_allowable_orders) should be
used to get this information. I have no strong opinon on how the implementation
of that interface looks. What you suggest below seems perfectly reasonable to me.
Quoted 6 lines
Quoted 4 lines
Right. thp_vma_allowable_orders() might require some care as discussed in other
context (cleanly separate dax and shmem handling/orders). But that would be follow-up cleanups.
Are you planning to do that, or do you want me to send a patch?
I'm planning on looking into some details, especially the interaction with large folios in the pagecache. I'll let you know once I have a better idea what actually should be done :)
Quoted 7 lines
Quoted 5 lines
Quoted 1 line
OK great - I'll scrub it from my todo list... really getting things done today :)
Resolved the khugepaged thiny already? :P
[khugepaged not active when only enabling the sub-size via the 2M folder IIRC]
Hmm... baby brain?
:)
I think I only mentioned it in a private mail at some point.
Quoted 3 lines
Sorry about that. I've been a bit useless lately. For some reason it wasn't on my list, but its there now. Will prioritise it, because I agree it's not good.
IIRC, if you do
echo never > /sys/kernel/mm/transparent_hugepage/enabled
echo always > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled

Quoted 6 lines

Quoted 3 lines
Quoted 1 line
khugepaged will not get activated.
khugepaged is controlled by the top level knob.
What do you mean by "top level knob"? I assume
/sys/kernel/mm/transparent_hugepage/enabled ?

Yes.

Quoted 14 lines

If so, that's not really a thing in its own right; its just the legacy PMD-size THP control, and we only take any notice of it if a per-size control is set to "inherit". So if we have:
# echo always > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled
Then by design, /sys/kernel/mm/transparent_hugepage/enabled should be ignored.
Quoted 4 lines
But the above setting
sounds confusing, can we disable the top level knob, but enable it on a per-order basis? TBH, it sounds weird and doesn't make too much sense to me.

Quoted 6 lines

Well that's the design and that's how its documented. It's done this way for back-compat. All controls are now per-size. But at boot, we default all per-size controls to "never" except for the PMD-sized control, which is defaulted to "inherit". That way, an unenlightened user-space can still control PMD-sized THP via the legacy (top-level) control. But enlightened apps can directly control per-size.

OK, good to know.

Quoted 3 lines

I'm not sure how your way would work, because you would have 2 controls competing to do the same thing?

I don't see how they compete if they are 2-level knobs. And I failed to see how it achieved back-compat. For example, memcached reads /sys/kernel/mm/transparent_hugepage/enabled to determine whether it should manage memory in huge page (2M) granularity. If the setting is set to :

# echo never > /sys/kernel/mm/transparent_hugepage/enabled
# echo always > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled

memcached will manage memory in 4K granularity, but 2M THP is actually enabled unless memcached checks the per-order knobs.

If we use 2-level mode, memcached doesn't need check per-order setting at all in order to know whether THP is enabled or not. And it actually doesn't care about what orders are enabled, it assumes THP size is 2M (or PMD size). Even though 2M is not enabled but lower orders are enabled, memcached still can fully utilize the mTHP since the memory chunk managed by memcached is still 2M aligned in this setting. So unenlightened applications still can work well. Jemalloc should do the similar thing if I remember correctly.

Quoted 10 lines

Quoted 9 lines
Quoted 9 lines
--
Cheers,
David / dhildenb

David Hildenbrand

2024-07-03 16:19 UTC #22 lore raw

Quoted 151 lines · Yang Shi

Quoted 108 lines · Ryan Roberts
Quoted 106 lines · Yang Shi
Quoted 105 lines · David Hildenbrand
Quoted 81 lines · Ryan Roberts
Quoted 80 lines · David Hildenbrand
Quoted 79 lines · Ryan Roberts
Quoted 77 lines · David Hildenbrand
Quoted 66 lines · Ryan Roberts
Quoted 65 lines · David Hildenbrand
Quoted 63 lines · Ryan Roberts
Quoted 48 lines · Baolin Wang
Quoted 44 lines · Ryan Roberts
On 28/06/2024 11:49, Bang Li wrote:
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for
anonymous shmem"), we can configure different policies through
the multi-size THP sysfs interface for anonymous shmem. But
currently "THPeligible" indicates only whether the mapping is
eligible for allocating THP-pages as well as the THP is PMD
mappable or not for anonymous shmem, we need to support semantics
for mTHP with anonymous shmem similar to those for mTHP with
anonymous memory.
Signed-off-by: Bang Li <[email protected]>
---
fs/proc/task_mmu.c | 10 +++++++---
include/linux/huge_mm.h | 11 +++++++++++
mm/shmem.c | 9 +--------
3 files changed, 19 insertions(+), 11 deletions(-)
diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
index 93fb2c61b154..09b5db356886 100644
--- a/fs/proc/task_mmu.c
+++ b/fs/proc/task_mmu.c
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
{
struct vm_area_struct *vma = v;
struct mem_size_stats mss = {};
+ bool thp_eligible;
smap_gather_stats(vma, &mss, 0);
@@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void
*v)
__show_smap(m, &mss, false);
- seq_printf(m, "THPeligible: %8u\n",
- !!thp_vma_allowable_orders(vma, vma->vm_flags,
- TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
+ thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
+ TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL);
+ if (vma_is_anon_shmem(vma))
+ thp_eligible =
!!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+ vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I
would
have liked, but is there a reason why we need a separate function for
shmem?
Quoted 7 lines
Since shmem_allowable_huge_orders() only uses shmem specific logic to determine
if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making
it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.
My point was really that a single interface (thp_vma_allowable_orders) should be
used to get this information. I have no strong opinon on how the implementation
of that interface looks. What you suggest below seems perfectly reasonable to me.
Quoted 6 lines
Quoted 4 lines
Right. thp_vma_allowable_orders() might require some care as discussed in other
context (cleanly separate dax and shmem handling/orders). But that would be follow-up cleanups.
Are you planning to do that, or do you want me to send a patch?
I'm planning on looking into some details, especially the interaction with large folios in the pagecache. I'll let you know once I have a better idea what actually should be done :)
Quoted 7 lines
Quoted 5 lines
Quoted 1 line
OK great - I'll scrub it from my todo list... really getting things done today :)
Resolved the khugepaged thiny already? :P
[khugepaged not active when only enabling the sub-size via the 2M folder IIRC]
Hmm... baby brain?
:)
I think I only mentioned it in a private mail at some point.
Quoted 3 lines
Sorry about that. I've been a bit useless lately. For some reason it wasn't on my list, but its there now. Will prioritise it, because I agree it's not good.
IIRC, if you do
echo never > /sys/kernel/mm/transparent_hugepage/enabled
echo always > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled
Quoted 6 lines
Quoted 3 lines
Quoted 1 line
khugepaged will not get activated.
khugepaged is controlled by the top level knob.
What do you mean by "top level knob"? I assume
/sys/kernel/mm/transparent_hugepage/enabled ?
Yes.
Quoted 14 lines
If so, that's not really a thing in its own right; its just the legacy PMD-size THP control, and we only take any notice of it if a per-size control is set to "inherit". So if we have:
# echo always > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled
Then by design, /sys/kernel/mm/transparent_hugepage/enabled should be ignored.
Quoted 4 lines
But the above setting
sounds confusing, can we disable the top level knob, but enable it on a per-order basis? TBH, it sounds weird and doesn't make too much sense to me.
Quoted 6 lines
Well that's the design and that's how its documented. It's done this way for back-compat. All controls are now per-size. But at boot, we default all per-size controls to "never" except for the PMD-sized control, which is defaulted to "inherit". That way, an unenlightened user-space can still control PMD-sized THP via the legacy (top-level) control. But enlightened apps can directly control per-size.
OK, good to know.
Quoted 3 lines
I'm not sure how your way would work, because you would have 2 controls competing to do the same thing?
I don't see how they compete if they are 2-level knobs. And I failed to see how it achieved back-compat. For example, memcached reads /sys/kernel/mm/transparent_hugepage/enabled to determine whether it should manage memory in huge page (2M) granularity. If the setting is set to :

Quoted 5 lines

# echo never > /sys/kernel/mm/transparent_hugepage/enabled
# echo always > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled
memcached will manage memory in 4K granularity, but 2M THP is actually enabled unless memcached checks the per-order knobs.

And you can still do it the old way and keep it all working with existing software (compat mode as default).

It's just another option and some software might need updates to benefit from it (just like if you would enable other folio sizes).

You can happily do

echo always > /sys/kernel/mm/transparent_hugepage/enabled
echo inherit > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled

It's an admin choice.

Signature · 3 lines

Cheers,

David / dhildenb

Ryan Roberts

reviewer

2024-07-04 09:43 UTC #23 lore raw

In reply to Yang Shi

Quoted 139 lines · Yang Shi

Quoted 108 lines · Ryan Roberts
Quoted 106 lines · Yang Shi
Quoted 105 lines · David Hildenbrand
Quoted 81 lines · Ryan Roberts
Quoted 80 lines · David Hildenbrand
Quoted 79 lines · Ryan Roberts
Quoted 77 lines · David Hildenbrand
Quoted 66 lines · Ryan Roberts
Quoted 65 lines · David Hildenbrand
Quoted 63 lines · Ryan Roberts
Quoted 48 lines · Baolin Wang
Quoted 44 lines · Ryan Roberts
On 28/06/2024 11:49, Bang Li wrote:
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for
anonymous shmem"), we can configure different policies through
the multi-size THP sysfs interface for anonymous shmem. But
currently "THPeligible" indicates only whether the mapping is
eligible for allocating THP-pages as well as the THP is PMD
mappable or not for anonymous shmem, we need to support semantics
for mTHP with anonymous shmem similar to those for mTHP with
anonymous memory.
Signed-off-by: Bang Li <[email protected]>
---
fs/proc/task_mmu.c | 10 +++++++---
include/linux/huge_mm.h | 11 +++++++++++
mm/shmem.c | 9 +--------
3 files changed, 19 insertions(+), 11 deletions(-)
diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
index 93fb2c61b154..09b5db356886 100644
--- a/fs/proc/task_mmu.c
+++ b/fs/proc/task_mmu.c
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
{
struct vm_area_struct *vma = v;
struct mem_size_stats mss = {};
+ bool thp_eligible;
smap_gather_stats(vma, &mss, 0);
@@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void
*v)
__show_smap(m, &mss, false);
- seq_printf(m, "THPeligible: %8u\n",
- !!thp_vma_allowable_orders(vma, vma->vm_flags,
- TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
+ thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
+ TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL);
+ if (vma_is_anon_shmem(vma))
+ thp_eligible =
!!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+ vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I
would
have liked, but is there a reason why we need a separate function for
shmem?
Quoted 7 lines
Since shmem_allowable_huge_orders() only uses shmem specific logic to determine
if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making
it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.
My point was really that a single interface (thp_vma_allowable_orders) should be
used to get this information. I have no strong opinon on how the implementation
of that interface looks. What you suggest below seems perfectly reasonable to me.
Quoted 6 lines
Quoted 4 lines
Right. thp_vma_allowable_orders() might require some care as discussed in other
context (cleanly separate dax and shmem handling/orders). But that would be follow-up cleanups.
Are you planning to do that, or do you want me to send a patch?
I'm planning on looking into some details, especially the interaction with large folios in the pagecache. I'll let you know once I have a better idea what actually should be done :)
Quoted 7 lines
Quoted 5 lines
Quoted 1 line
OK great - I'll scrub it from my todo list... really getting things done today :)
Resolved the khugepaged thiny already? :P
[khugepaged not active when only enabling the sub-size via the 2M folder IIRC]
Hmm... baby brain?
:)
I think I only mentioned it in a private mail at some point.
Quoted 3 lines
Sorry about that. I've been a bit useless lately. For some reason it wasn't on my list, but its there now. Will prioritise it, because I agree it's not good.
IIRC, if you do
echo never > /sys/kernel/mm/transparent_hugepage/enabled
echo always > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled
Quoted 6 lines
Quoted 3 lines
Quoted 1 line
khugepaged will not get activated.
khugepaged is controlled by the top level knob.
What do you mean by "top level knob"? I assume
/sys/kernel/mm/transparent_hugepage/enabled ?
Yes.
Quoted 14 lines
If so, that's not really a thing in its own right; its just the legacy PMD-size THP control, and we only take any notice of it if a per-size control is set to "inherit". So if we have:
# echo always > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled
Then by design, /sys/kernel/mm/transparent_hugepage/enabled should be ignored.
Quoted 4 lines
But the above setting
sounds confusing, can we disable the top level knob, but enable it on a per-order basis? TBH, it sounds weird and doesn't make too much sense to me.
Quoted 6 lines
Well that's the design and that's how its documented. It's done this way for back-compat. All controls are now per-size. But at boot, we default all per-size controls to "never" except for the PMD-sized control, which is defaulted to "inherit". That way, an unenlightened user-space can still control PMD-sized THP via the legacy (top-level) control. But enlightened apps can directly control per-size.

Quoted 7 lines

OK, good to know.
Quoted 3 lines
I'm not sure how your way would work, because you would have 2 controls competing to do the same thing?
I don't see how they compete if they are 2-level knobs.

I'm not sure I understand exactly how your 2-level proposal works. Could you explain in more detail?

The problem as I see it, is that the control can take multiple values; "never", "always" or "madvise". In a two-level scheme, what do we do when top level says "always" but per-size control says "madvise", or vice-versa? The scheme we adopted has clear and obvious (to me at least) semantics in this case.

The other problem is that the top-level control is still used to control file memory collapse (when CONFIG_READ_ONLY_THP_FOR_FS is configured). If you're advocating for a scheme where the top-level is set to the most permissive you want to allow, then the per-size controls are only able to further restrict, that would make it impossible to, for instance set all 2M THP (inc file-backed) to madvise, but set all 64K THP to always.

Quoted 12 lines

And I failed
to see how it achieved back-compat. For example, memcached reads /sys/kernel/mm/transparent_hugepage/enabled to determine whether it should manage memory in huge page (2M) granularity. If the setting is set to :
# echo never > /sys/kernel/mm/transparent_hugepage/enabled
# echo always > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled
memcached will manage memory in 4K granularity, but 2M THP is actually enabled unless memcached checks the per-order knobs.

Quoted 8 lines

If we use 2-level mode, memcached doesn't need check per-order setting at all in order to know whether THP is enabled or not. And it actually doesn't care about what orders are enabled, it assumes THP size is 2M (or PMD size). Even though 2M is not enabled but lower orders are enabled, memcached still can fully utilize the mTHP since the memory chunk managed by memcached is still 2M aligned in this setting. So unenlightened applications still can work well. Jemalloc should do the similar thing if I remember correctly.

I wonder why we didn't decide to just make /sys/kernel/mm/transparent_hugepage/enabled an alias for /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled ? That may have solved this problem more cleanly? But that would have made it difficult to introduce "auto" in future (the idea was to set all per-size to 'inherit' and then set top-level to 'auto').

Quoted 11 lines

Quoted 11 lines
Quoted 10 lines
Quoted 10 lines
--
Cheers,
David / dhildenb

Yang Shi

2024-07-09 19:02 UTC #24 lore raw

Quoted 157 lines · Ryan Roberts

Quoted 139 lines · Yang Shi
Quoted 108 lines · Ryan Roberts
Quoted 106 lines · Yang Shi
Quoted 105 lines · David Hildenbrand
Quoted 81 lines · Ryan Roberts
Quoted 80 lines · David Hildenbrand
Quoted 79 lines · Ryan Roberts
Quoted 77 lines · David Hildenbrand
Quoted 66 lines · Ryan Roberts
Quoted 65 lines · David Hildenbrand
Quoted 63 lines · Ryan Roberts
Quoted 55 lines · Baolin Wang
On 2024/7/1 15:55, Ryan Roberts wrote:
On 28/06/2024 11:49, Bang Li wrote:
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for
anonymous shmem"), we can configure different policies through
the multi-size THP sysfs interface for anonymous shmem. But
currently "THPeligible" indicates only whether the mapping is
eligible for allocating THP-pages as well as the THP is PMD
mappable or not for anonymous shmem, we need to support semantics
for mTHP with anonymous shmem similar to those for mTHP with
anonymous memory.
Signed-off-by: Bang Li <[email protected]>
---
fs/proc/task_mmu.c | 10 +++++++---
include/linux/huge_mm.h | 11 +++++++++++
mm/shmem.c | 9 +--------
3 files changed, 19 insertions(+), 11 deletions(-)
diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
index 93fb2c61b154..09b5db356886 100644
--- a/fs/proc/task_mmu.c
+++ b/fs/proc/task_mmu.c
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
{
struct vm_area_struct *vma = v;
struct mem_size_stats mss = {};
+ bool thp_eligible;
smap_gather_stats(vma, &mss, 0);
@@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void
*v)
__show_smap(m, &mss, false);
- seq_printf(m, "THPeligible: %8u\n",
- !!thp_vma_allowable_orders(vma, vma->vm_flags,
- TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
+ thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
+ TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL);
+ if (vma_is_anon_shmem(vma))
+ thp_eligible =
!!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+ vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I
would
have liked, but is there a reason why we need a separate function for
shmem?
Since shmem_allowable_huge_orders() only uses shmem specific logic to
determine
if huge orders are allowable, there is no need to complicate the
thp_vma_allowable_orders() function by adding more shmem related logic,
making
it more bloated. In my view, providing a dedicated helper
shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.
My point was really that a single interface (thp_vma_allowable_orders) should be
used to get this information. I have no strong opinon on how the implementation
of that interface looks. What you suggest below seems perfectly reasonable to me.
Quoted 6 lines
Quoted 4 lines
Right. thp_vma_allowable_orders() might require some care as discussed in other
context (cleanly separate dax and shmem handling/orders). But that would be follow-up cleanups.
Are you planning to do that, or do you want me to send a patch?
I'm planning on looking into some details, especially the interaction with large folios in the pagecache. I'll let you know once I have a better idea what actually should be done :)
Quoted 7 lines
Quoted 5 lines
Quoted 1 line
OK great - I'll scrub it from my todo list... really getting things done today :)
Resolved the khugepaged thiny already? :P
[khugepaged not active when only enabling the sub-size via the 2M folder IIRC]
Hmm... baby brain?
:)
I think I only mentioned it in a private mail at some point.
Quoted 3 lines
Sorry about that. I've been a bit useless lately. For some reason it wasn't on my list, but its there now. Will prioritise it, because I agree it's not good.
IIRC, if you do
echo never > /sys/kernel/mm/transparent_hugepage/enabled
echo always > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled
Quoted 6 lines
Quoted 3 lines
Quoted 1 line
khugepaged will not get activated.
khugepaged is controlled by the top level knob.
What do you mean by "top level knob"? I assume
/sys/kernel/mm/transparent_hugepage/enabled ?
Yes.
Quoted 14 lines
If so, that's not really a thing in its own right; its just the legacy PMD-size THP control, and we only take any notice of it if a per-size control is set to "inherit". So if we have:
# echo always > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled
Then by design, /sys/kernel/mm/transparent_hugepage/enabled should be ignored.
Quoted 4 lines
But the above setting
sounds confusing, can we disable the top level knob, but enable it on a per-order basis? TBH, it sounds weird and doesn't make too much sense to me.
Quoted 6 lines
Well that's the design and that's how its documented. It's done this way for back-compat. All controls are now per-size. But at boot, we default all per-size controls to "never" except for the PMD-sized control, which is defaulted to "inherit". That way, an unenlightened user-space can still control PMD-sized THP via the legacy (top-level) control. But enlightened apps can directly control per-size.
Quoted 7 lines
OK, good to know.
Quoted 3 lines
I'm not sure how your way would work, because you would have 2 controls competing to do the same thing?
I don't see how they compete if they are 2-level knobs.
I'm not sure I understand exactly how your 2-level proposal works. Could you explain in more detail?
The problem as I see it, is that the control can take multiple values; "never", "always" or "madvise". In a two-level scheme, what do we do when top level says "always" but per-size control says "madvise", or vice-versa? The scheme we adopted has clear and obvious (to me at least) semantics in this case.

Quoted 6 lines

The other problem is that the top-level control is still used to control file memory collapse (when CONFIG_READ_ONLY_THP_FOR_FS is configured). If you're advocating for a scheme where the top-level is set to the most permissive you want to allow, then the per-size controls are only able to further restrict, that would make it impossible to, for instance set all 2M THP (inc file-backed) to madvise, but set all 64K THP to always.

If we really have such usecases having different orders to have different modes (always or madvise), I think the top level knob can be actually simplified to on/off mode. For example, always/madvise maps to on, never maps to off since the allocation decision (shall allocate THP or not, and how hard to try) is actually done on a per order basis. The Memcached usecase actually doesn't care about always or madvise, as long as it is not never Memcached will manage memory in huge page granularity.

This might be useful, for example, having high order (i.e. 2M) set to madvise, low order (i.e. 64K) set to always. This may be able to achieve some balance between performance and memory usage.

Quoted 42 lines

Quoted 13 lines
And I failed
to see how it achieved back-compat. For example, memcached reads /sys/kernel/mm/transparent_hugepage/enabled to determine whether it should manage memory in huge page (2M) granularity. If the setting is set to :
# echo never > /sys/kernel/mm/transparent_hugepage/enabled
# echo always > /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled
memcached will manage memory in 4K granularity, but 2M THP is actually enabled unless memcached checks the per-order knobs.
Quoted 8 lines
If we use 2-level mode, memcached doesn't need check per-order setting at all in order to know whether THP is enabled or not. And it actually doesn't care about what orders are enabled, it assumes THP size is 2M (or PMD size). Even though 2M is not enabled but lower orders are enabled, memcached still can fully utilize the mTHP since the memory chunk managed by memcached is still 2M aligned in this setting. So unenlightened applications still can work well. Jemalloc should do the similar thing if I remember correctly.
I wonder why we didn't decide to just make /sys/kernel/mm/transparent_hugepage/enabled an alias for /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/enabled ? That may have solved this problem more cleanly? But that would have made it difficult to introduce "auto" in future (the idea was to set all per-size to 'inherit' and then set top-level to 'auto').
Quoted 11 lines
Quoted 11 lines
Quoted 10 lines
Quoted 10 lines
--
Cheers,
David / dhildenb

Bang Li

author

2024-07-01 09:43 UTC #25 lore raw

In reply to Baolin Wang

Hi, Baolin

Quoted 130 lines · Baolin Wang

Quoted 33 lines · Ryan Roberts
Quoted 32 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
fs/proc/task_mmu.c | 10 +++++++---
include/linux/huge_mm.h | 11 +++++++++++
mm/shmem.c | 9 +--------
  3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +2 -3
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
   {
       struct vm_area_struct *vma = v;
       struct mem_size_stats mss = {};
+    bool thp_eligible;
       smap_gather_stats(vma, &mss, 0);
@@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void *v)
       __show_smap(m, &mss, false);
-    seq_printf(m, "THPeligible:    %8u\n",
-           !!thp_vma_allowable_orders(vma, vma->vm_flags,
-               TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
+    thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
Quoted 8 lines
fs/proc/task_mmu.c static int show_smap(struct seq_file *m, void *v) · line 889
Quoted 5 lines
fs/proc/task_mmu.c +4 -0
+                        TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL);
+    if (vma_is_anon_shmem(vma))
+        thp_eligible = 
 !!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+                            vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I would have liked, but is there a reason why we need a separate function for shmem?
Since shmem_allowable_huge_orders() only uses shmem specific logic to determine if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.
IIUC, I agree with David's suggestion that the shmem_allowable_huge_orders() helper function could be used in thp_vma_allowable_orders() to support shmem mTHP. Something like:
mm/huge_memory.c +5 -4
index c7ce28f6b7f3..9677fe6cf478 100644
@@ -151,10 +151,13 @@ unsigned long __thp_vma_allowable_orders(struct 
 vm_area_struct *vma,
          * Must be done before hugepage flags check since shmem has its
          * own flags.
          */
-       if (!in_pf && shmem_file(vma->vm_file))
-               return shmem_is_huge(file_inode(vma->vm_file), 
 vma->vm_pgoff,
-                                    !enforce_sysfs, vma->vm_mm, vm_flags)
-                       ? orders : 0;
+       if (!in_pf && shmem_file(vma->vm_file)) {
+               bool global_huge = 
 shmem_is_huge(file_inode(vma->vm_file), vma->vm_pgoff,
+                                    !enforce_sysfs, vma->vm_mm, vm_flags);
+
+               return 
shmem_allowable_huge_orders(file_inode(vma->vm_file),
+                                       vma, vma->vm_pgoff, global_huge); +       }
        if (!vma_is_anonymous(vma)) {
                /*
Quoted 54 lines
Couldn't (shouldn't) thp_vma_allowable_orders() be taught to handle shmem too?
Quoted 52 lines
fs/proc/task_mmu.c +1 -0
+    seq_printf(m, "THPeligible:    %8u\n", thp_eligible);
       if (arch_pkeys_enabled())
           seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
include/linux/huge_mm.h +11 -0
index 212cca384d7e..f87136f38aa1 100644
@@ -267,6 +267,10 @@ unsigned long thp_vma_allowable_orders(struct 
 vm_area_struct *vma,
       return __thp_vma_allowable_orders(vma, vm_flags, tva_flags, orders);
   }
+unsigned long shmem_allowable_huge_orders(struct inode *inode,
+                struct vm_area_struct *vma, pgoff_t index,
+                bool global_huge);
+
   struct thpsize {
       struct kobject kobj;
       struct list_head node;
@@ -460,6 +464,13 @@ static inline unsigned long 
 thp_vma_allowable_orders(struct vm_area_struct *vma,
       return 0;
   }
+static inline unsigned long shmem_allowable_huge_orders(struct inode 
 *inode,
+                struct vm_area_struct *vma, pgoff_t index,
+                bool global_huge)
+{
+    return 0;
+}
+
   #define transparent_hugepage_flags 0UL
   #define thp_get_unmapped_area    NULL
mm/shmem.c +1 -1
index d495c0701a83..aa85df9c662a 100644
@@ -1622,7 +1622,7 @@ static gfp_t limit_gfp_mask(gfp_t huge_gfp, 
 gfp_t limit_gfp)
   }
   #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-static unsigned long shmem_allowable_huge_orders(struct inode *inode,
+unsigned long shmem_allowable_huge_orders(struct inode *inode,
                   struct vm_area_struct *vma, pgoff_t index,
                   bool global_huge)
   {
@@ -1707,13 +1707,6 @@ static unsigned long 
shmem_suitable_orders(struct inode *inode, struct vm_fault return orders;
}
#else
-static unsigned long shmem_allowable_huge_orders(struct inode *inode, - struct vm_area_struct *vma, pgoff_t index,

Quoted 8 lines

Quoted 8 lines
Quoted 8 lines
- bool global_huge)
mm/shmem.c +0 -4
-{
-    return 0;
-}
-
static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault *vmf, struct address_space *mapping, pgoff_t index, unsigned long orders)

Thanks for the reference code. Currently, we only implement the mTHP of anonymous shmem, so we only need to handle anonymous shmem specially. As shown in the following code:

mm/huge_memory.c +8 -4

@@ -151,10 +151,14 @@ unsigned long __thp_vma_allowable_orders(struct vm_area_struct *vma,
         * Must be done before hugepage flags check since shmem has its
         * own flags.
         */
-       if (!in_pf && shmem_file(vma->vm_file))
-               return shmem_is_huge(file_inode(vma->vm_file), vma->vm_pgoff,
-                                    !enforce_sysfs, vma->vm_mm, vm_flags)
-                       ? orders : 0;
+       if (!in_pf && shmem_file(vma->vm_file)) {
+               bool global_huge = shmem_is_huge(file_inode(vma->vm_file), vma->vm_pgoff,
+                                    !enforce_sysfs, vma->vm_mm, vm_flags);
+               if (!vma_is_anon_shmem(vma))
+                       return global_huge? orders : 0;
+               return shmem_allowable_huge_orders(file_inode(vma->vm_file),
+                                               vma, vma->vm_pgoff, global_huge);
+       }
 
        if (!vma_is_anonymous(vma)) {

Thanks,
Bang

Baolin Wang

reviewer

2024-07-01 11:12 UTC #26 lore raw

Quoted 155 lines · Bang Li

Hi, Baolin
Quoted 133 lines · Baolin Wang
Quoted 34 lines · Ryan Roberts
Quoted 33 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
fs/proc/task_mmu.c | 10 +++++++---
include/linux/huge_mm.h | 11 +++++++++++
mm/shmem.c | 9 +--------
  3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +3 -3
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
   {
       struct vm_area_struct *vma = v;
       struct mem_size_stats mss = {};
+    bool thp_eligible;
       smap_gather_stats(vma, &mss, 0);
@@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void *v)
       __show_smap(m, &mss, false);
-    seq_printf(m, "THPeligible:    %8u\n",
-           !!thp_vma_allowable_orders(vma, vma->vm_flags,
-               TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
+    thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
+                        TVA_SMAPS | TVA_ENFORCE_SYSFS, 
Quoted 8 lines
fs/proc/task_mmu.c static int show_smap(struct seq_file *m, void *v) · line 890
Quoted 5 lines
fs/proc/task_mmu.c +3 -0
 THP_ORDERS_ALL);
+    if (vma_is_anon_shmem(vma))
+        thp_eligible = 
 !!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+                            vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I would have liked, but is there a reason why we need a separate function for shmem?
Since shmem_allowable_huge_orders() only uses shmem specific logic to determine if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.
IIUC, I agree with David's suggestion that the shmem_allowable_huge_orders() helper function could be used in thp_vma_allowable_orders() to support shmem mTHP. Something like:
mm/huge_memory.c +2 -4
index c7ce28f6b7f3..9677fe6cf478 100644
@@ -151,10 +151,13 @@ unsigned long __thp_vma_allowable_orders(struct 
 vm_area_struct *vma,
           * Must be done before hugepage flags check since shmem has its
           * own flags.
           */
-       if (!in_pf && shmem_file(vma->vm_file))
-               return shmem_is_huge(file_inode(vma->vm_file), 
 vma->vm_pgoff,
-                                    !enforce_sysfs, vma->vm_mm, 
 vm_flags)
-                       ? orders : 0;
+       if (!in_pf && shmem_file(vma->vm_file)) {
+               bool global_huge = 
shmem_is_huge(file_inode(vma->vm_file), vma->vm_pgoff,
+                                    !enforce_sysfs, vma->vm_mm, vm_flags);
+
+               return shmem_allowable_huge_orders(file_inode(vma->vm_file),
+                                       vma, vma->vm_pgoff, global_huge); +       }
         if (!vma_is_anonymous(vma)) {
                 /*
Quoted 54 lines
Couldn't (shouldn't) thp_vma_allowable_orders() be taught to handle shmem too?
Quoted 52 lines
fs/proc/task_mmu.c +1 -0
+    seq_printf(m, "THPeligible:    %8u\n", thp_eligible);
       if (arch_pkeys_enabled())
           seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
include/linux/huge_mm.h +11 -0
index 212cca384d7e..f87136f38aa1 100644
@@ -267,6 +267,10 @@ unsigned long thp_vma_allowable_orders(struct 
 vm_area_struct *vma,
       return __thp_vma_allowable_orders(vma, vm_flags, tva_flags, orders);
   }
+unsigned long shmem_allowable_huge_orders(struct inode *inode,
+                struct vm_area_struct *vma, pgoff_t index,
+                bool global_huge);
+
   struct thpsize {
       struct kobject kobj;
       struct list_head node;
@@ -460,6 +464,13 @@ static inline unsigned long 
 thp_vma_allowable_orders(struct vm_area_struct *vma,
       return 0;
   }
+static inline unsigned long shmem_allowable_huge_orders(struct 
 inode *inode,
+                struct vm_area_struct *vma, pgoff_t index,
+                bool global_huge)
+{
+    return 0;
+}
+
   #define transparent_hugepage_flags 0UL
   #define thp_get_unmapped_area    NULL
mm/shmem.c +1 -1
index d495c0701a83..aa85df9c662a 100644
@@ -1622,7 +1622,7 @@ static gfp_t limit_gfp_mask(gfp_t huge_gfp, 
 gfp_t limit_gfp)
   }
   #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-static unsigned long shmem_allowable_huge_orders(struct inode *inode,
+unsigned long shmem_allowable_huge_orders(struct inode *inode,
                   struct vm_area_struct *vma, pgoff_t index,
                   bool global_huge)
   {
@@ -1707,13 +1707,6 @@ static unsigned long 
shmem_suitable_orders(struct inode *inode, struct vm_fault return orders;
}
#else
-static unsigned long shmem_allowable_huge_orders(struct inode *inode, - struct vm_area_struct *vma, pgoff_t index,
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
- bool global_huge)
mm/shmem.c +0 -4
-{
-    return 0;
-}
-
static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault *vmf, struct address_space *mapping, pgoff_t index, unsigned long orders)
Thanks for the reference code. Currently, we only implement the mTHP of anonymous shmem, so we only need to handle anonymous shmem specially. As shown in the following code:
mm/huge_memory.c +0 -0
@@ -151,10 +151,14 @@ unsigned long __thp_vma_allowable_orders(struct 
 vm_area_struct *vma,
          * Must be done before hugepage flags check since shmem has its
          * own flags.

mm/huge_memory.c unsigned long __thp_vma_allowable_orders(struct · line 159

Quoted 10 lines

mm/huge_memory.c +3 -4

          */
-       if (!in_pf && shmem_file(vma->vm_file))
-               return shmem_is_huge(file_inode(vma->vm_file), 
 vma->vm_pgoff,
-                                    !enforce_sysfs, vma->vm_mm, vm_flags)
-                       ? orders : 0;
+       if (!in_pf && shmem_file(vma->vm_file)) {
+               bool global_huge = 
 shmem_is_huge(file_inode(vma->vm_file), vma->vm_pgoff,
+                                    !enforce_sysfs, vma->vm_mm, vm_flags);

Nit: add a blank line after the declaration. Otherwise looks good to me.

Quoted 12 lines

mm/huge_memory.c +3 -0
+               if (!vma_is_anon_shmem(vma))
+                       return global_huge? orders : 0;
+               return 
shmem_allowable_huge_orders(file_inode(vma->vm_file),
+                                               vma, vma->vm_pgoff, global_huge);
+       }
        if (!vma_is_anonymous(vma)) {
Thanks,
Bang

Bang Li

author

2024-07-01 14:51 UTC #27 lore raw

Quoted 165 lines · Baolin Wang

Quoted 162 lines · Bang Li
Hi, Baolin
Quoted 134 lines · Baolin Wang
Quoted 34 lines · Ryan Roberts
Quoted 33 lines · Bang Li
After the commit 7fb1b252afb5 ("mm: shmem: add mTHP support for anonymous shmem"), we can configure different policies through the multi-size THP sysfs interface for anonymous shmem. But currently "THPeligible" indicates only whether the mapping is eligible for allocating THP-pages as well as the THP is PMD mappable or not for anonymous shmem, we need to support semantics for mTHP with anonymous shmem similar to those for mTHP with anonymous memory.
Signed-off-by Bang Li <[email protected]>
---
fs/proc/task_mmu.c | 10 +++++++---
include/linux/huge_mm.h | 11 +++++++++++
mm/shmem.c | 9 +--------
  3 files changed, 19 insertions(+), 11 deletions(-)
fs/proc/task_mmu.c +3 -3
index 93fb2c61b154..09b5db356886 100644
@@ -870,6 +870,7 @@ static int show_smap(struct seq_file *m, void *v)
   {
       struct vm_area_struct *vma = v;
       struct mem_size_stats mss = {};
+    bool thp_eligible;
       smap_gather_stats(vma, &mss, 0);
@@ -882,9 +883,12 @@ static int show_smap(struct seq_file *m, void *v)
       __show_smap(m, &mss, false);
-    seq_printf(m, "THPeligible:    %8u\n",
-           !!thp_vma_allowable_orders(vma, vma->vm_flags,
-               TVA_SMAPS | TVA_ENFORCE_SYSFS, THP_ORDERS_ALL));
+    thp_eligible = !!thp_vma_allowable_orders(vma, vma->vm_flags,
+                        TVA_SMAPS | TVA_ENFORCE_SYSFS, 
Quoted 8 lines
fs/proc/task_mmu.c static int show_smap(struct seq_file *m, void *v) · line 890
Quoted 5 lines
fs/proc/task_mmu.c +3 -0
 THP_ORDERS_ALL);
+    if (vma_is_anon_shmem(vma))
+        thp_eligible = 
 !!shmem_allowable_huge_orders(file_inode(vma->vm_file),
+                            vma, vma->vm_pgoff, thp_eligible);
Afraid I haven't been following the shmem mTHP support work as much as I would have liked, but is there a reason why we need a separate function for shmem?
Since shmem_allowable_huge_orders() only uses shmem specific logic to determine if huge orders are allowable, there is no need to complicate the thp_vma_allowable_orders() function by adding more shmem related logic, making it more bloated. In my view, providing a dedicated helper shmem_allowable_huge_orders(), specifically for shmem, simplifies the logic.
IIUC, I agree with David's suggestion that the shmem_allowable_huge_orders() helper function could be used in thp_vma_allowable_orders() to support shmem mTHP. Something like:
mm/huge_memory.c +2 -4
index c7ce28f6b7f3..9677fe6cf478 100644
@@ -151,10 +151,13 @@ unsigned long __thp_vma_allowable_orders(struct 
 vm_area_struct *vma,
           * Must be done before hugepage flags check since shmem has its
           * own flags.
           */
-       if (!in_pf && shmem_file(vma->vm_file))
-               return shmem_is_huge(file_inode(vma->vm_file), 
 vma->vm_pgoff,
-                                    !enforce_sysfs, vma->vm_mm, 
 vm_flags)
-                       ? orders : 0;
+       if (!in_pf && shmem_file(vma->vm_file)) {
+               bool global_huge = 
shmem_is_huge(file_inode(vma->vm_file), vma->vm_pgoff, +                                    !enforce_sysfs, vma->vm_mm, vm_flags);
+
+               return shmem_allowable_huge_orders(file_inode(vma->vm_file), +                                       vma, vma->vm_pgoff, global_huge);
+       }
         if (!vma_is_anonymous(vma)) {
                 /*
Quoted 54 lines
Couldn't (shouldn't) thp_vma_allowable_orders() be taught to handle shmem too?
Quoted 52 lines
fs/proc/task_mmu.c +1 -0
+    seq_printf(m, "THPeligible:    %8u\n", thp_eligible);
       if (arch_pkeys_enabled())
           seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
include/linux/huge_mm.h +11 -0
index 212cca384d7e..f87136f38aa1 100644
@@ -267,6 +267,10 @@ unsigned long thp_vma_allowable_orders(struct 
 vm_area_struct *vma,
       return __thp_vma_allowable_orders(vma, vm_flags, tva_flags, orders);
   }
+unsigned long shmem_allowable_huge_orders(struct inode *inode,
+                struct vm_area_struct *vma, pgoff_t index,
+                bool global_huge);
+
   struct thpsize {
       struct kobject kobj;
       struct list_head node;
@@ -460,6 +464,13 @@ static inline unsigned long 
 thp_vma_allowable_orders(struct vm_area_struct *vma,
       return 0;
   }
+static inline unsigned long shmem_allowable_huge_orders(struct 
 inode *inode,
+                struct vm_area_struct *vma, pgoff_t index,
+                bool global_huge)
+{
+    return 0;
+}
+
   #define transparent_hugepage_flags 0UL
   #define thp_get_unmapped_area    NULL
mm/shmem.c +1 -1
index d495c0701a83..aa85df9c662a 100644
@@ -1622,7 +1622,7 @@ static gfp_t limit_gfp_mask(gfp_t huge_gfp, 
 gfp_t limit_gfp)
   }
   #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-static unsigned long shmem_allowable_huge_orders(struct inode *inode,
+unsigned long shmem_allowable_huge_orders(struct inode *inode,
                   struct vm_area_struct *vma, pgoff_t index,
                   bool global_huge)
   {
@@ -1707,13 +1707,6 @@ static unsigned long 
shmem_suitable_orders(struct inode *inode, struct vm_fault return orders;
}
#else
-static unsigned long shmem_allowable_huge_orders(struct inode *inode, - struct vm_area_struct *vma, pgoff_t index,
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
- bool global_huge)
mm/shmem.c +0 -4
-{
-    return 0;
-}
-
static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault *vmf, struct address_space *mapping, pgoff_t index, unsigned long orders)
Thanks for the reference code. Currently, we only implement the mTHP of anonymous shmem, so we only need to handle anonymous shmem specially. As shown in the following code:
mm/huge_memory.c +0 -3
@@ -151,10 +151,14 @@ unsigned long __thp_vma_allowable_orders(struct 
 vm_area_struct *vma,
           * Must be done before hugepage flags check since shmem has its
           * own flags.
           */
-       if (!in_pf && shmem_file(vma->vm_file))
-               return shmem_is_huge(file_inode(vma->vm_file), 
 vma->vm_pgoff,
-                                    !enforce_sysfs, vma->vm_mm, 
 vm_flags)

Quoted 8 lines

Quoted 6 lines
mm/huge_memory.c +2 -1
-                       ? orders : 0;
+       if (!in_pf && shmem_file(vma->vm_file)) {
+               bool global_huge = 
shmem_is_huge(file_inode(vma->vm_file), vma->vm_pgoff, + !enforce_sysfs, vma->vm_mm, vm_flags);
Nit: add a blank line after the declaration. Otherwise looks good to me.

It doesn't matter to me whether I add spaces or not, thanks for your suggestion anyway.

Thanks,
Bang

Quoted 13 lines

Quoted 13 lines
+               if (!vma_is_anon_shmem(vma))
+                       return global_huge? orders : 0; +               return shmem_allowable_huge_orders(file_inode(vma->vm_file),
+                                               vma, vma->vm_pgoff, global_huge);
+       }
         if (!vma_is_anonymous(vma)) {
Thanks,
Bang