[PATCH 00/12] mempolicy: cleanups leading to NUMA mpol without vma

[PATCH 00/12] mempolicy: cleanups leading to NUMA mpol without vma

[Hugh Dickins]

Here is a series of patches based on v6.6-rc3: mostly cleanups in
mm/mempolicy.c, but finally removing the pseudo-vma from shmem folio
allocation, and removing the mmap_lock around folio migration for
mbind and migrate_pages syscalls.

Applies cleanly to any v6.6-rc so far, but not quite to mm-unstable or
linux-next: a couple of rejects in mm/mempolicy.c (clashing with the
vma_start_write() in do_mbind() and the foliation in mpol_misplaced())
and in mm/swap_state.c (clashing with zswap mods).  Adjustments will
also be needed if hugetlb page->index changes to PAGE_SIZE: interleave
has to co-operate with that.  After some review, a v2 will probably be
issued, based then on mm-unstable.

01/12 hugetlbfs: drop shared NUMA mempolicy pretence
02/12 kernfs: drop shared NUMA mempolicy hooks
03/12 mempolicy: fix migrate_pages(2) syscall return
04/12 mempolicy trivia: delete those ancient pr_debug()s
05/12 mempolicy trivia: slightly more consistent naming
06/12 mempolicy trivia: use pgoff_t in shared mempolicy tree
07/12 mempolicy: mpol_shared_policy_init() without pseudo-vma
08/12 mempolicy: remove confusing MPOL_MF_LAZY dead code
09/12 mm: add page_rmappable_folio() wrapper
10/12 mempolicy: alloc_pages_mpol() for NUMA policy without vma
11/12 mempolicy: mmap_lock is not needed while migrating folios
12/12 mempolicy: migration attempt to match interleave nodes

 fs/hugetlbfs/inode.c           |  41 +-
 fs/kernfs/file.c               |  49 --
 fs/proc/task_mmu.c             |   5 +-
 include/linux/gfp.h            |  10 +-
 include/linux/huge_mm.h        |  13 +
 include/linux/hugetlb.h        |  11 -
 include/linux/mempolicy.h      |  36 +-
 include/linux/mm.h             |   2 +-
 include/uapi/linux/mempolicy.h |   2 +-
 ipc/shm.c                      |  21 +-
 mm/hugetlb.c                   |  38 +-
 mm/mempolicy.c                 | 952 ++++++++++++++++-------------------
 mm/page_alloc.c                |   8 +-
 mm/shmem.c                     | 102 ++--
 mm/swap.h                      |   9 +-
 mm/swap_state.c                |  86 ++--
 16 files changed, 617 insertions(+), 768 deletions(-)

Hugh

[PATCH 01/12] hugetlbfs: drop shared NUMA mempolicy pretence

[Hugh Dickins]

hugetlbfs_fallocate() goes through the motions of pasting a shared NUMA
mempolicy onto its pseudo-vma, but how could there ever be a shared NUMA
mempolicy for this file?  hugetlb_vm_ops has never offered a set_policy
method, and hugetlbfs_parse_param() has never supported any mpol options
for a mount-wide default policy.

It's just an illusion: clean it away so as not to confuse others, giving
us more freedom to adjust shmem's set_policy/get_policy implementation.
But hugetlbfs_inode_info is still required, just to accommodate seals.

Yes, shared NUMA mempolicy support could be added to hugetlbfs, with a
set_policy method and/or mpol mount option (Andi's first posting did
include an admitted-unsatisfactory hugetlb_set_policy()); but it seems
that nobody has bothered to add that in the nineteen years since v2.6.7
made it possible, and there is at least one company that has invested
enough into hugetlbfs, that I guess they have learnt well enough how to
manage its NUMA, without needing shared mempolicy.

Signed-off-by: Hugh Dickins <hughd@google.com>
---
 fs/hugetlbfs/inode.c    | 41 +----------------------------------------
 include/linux/hugetlb.h |  2 --
 2 files changed, 1 insertion(+), 42 deletions(-)

diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
index 316c4cebd3f3..ffee27b10d42 100644
--- a/fs/hugetlbfs/inode.c
+++ b/fs/hugetlbfs/inode.c
@@ -83,29 +83,6 @@ static const struct fs_parameter_spec hugetlb_fs_parameters[] = {
 	{}
 };
 
-#ifdef CONFIG_NUMA
-static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma,
-					struct inode *inode, pgoff_t index)
-{
-	vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy,
-							index);
-}
-
-static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma)
-{
-	mpol_cond_put(vma->vm_policy);
-}
-#else
-static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma,
-					struct inode *inode, pgoff_t index)
-{
-}
-
-static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma)
-{
-}
-#endif
-
 /*
  * Mask used when checking the page offset value passed in via system
  * calls.  This value will be converted to a loff_t which is signed.
@@ -852,8 +829,7 @@ static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset,
 
 	/*
 	 * Initialize a pseudo vma as this is required by the huge page
-	 * allocation routines.  If NUMA is configured, use page index
-	 * as input to create an allocation policy.
+	 * allocation routines.
 	 */
 	vma_init(&pseudo_vma, mm);
 	vm_flags_init(&pseudo_vma, VM_HUGETLB | VM_MAYSHARE | VM_SHARED);
@@ -901,9 +877,7 @@ static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset,
 		 * folios in these areas, we need to consume the reserves
 		 * to keep reservation accounting consistent.
 		 */
-		hugetlb_set_vma_policy(&pseudo_vma, inode, index);
 		folio = alloc_hugetlb_folio(&pseudo_vma, addr, 0);
-		hugetlb_drop_vma_policy(&pseudo_vma);
 		if (IS_ERR(folio)) {
 			mutex_unlock(&hugetlb_fault_mutex_table[hash]);
 			error = PTR_ERR(folio);
@@ -1282,18 +1256,6 @@ static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 		hugetlbfs_inc_free_inodes(sbinfo);
 		return NULL;
 	}
-
-	/*
-	 * Any time after allocation, hugetlbfs_destroy_inode can be called
-	 * for the inode.  mpol_free_shared_policy is unconditionally called
-	 * as part of hugetlbfs_destroy_inode.  So, initialize policy here
-	 * in case of a quick call to destroy.
-	 *
-	 * Note that the policy is initialized even if we are creating a
-	 * private inode.  This simplifies hugetlbfs_destroy_inode.
-	 */
-	mpol_shared_policy_init(&p->policy, NULL);
-
 	return &p->vfs_inode;
 }
 
@@ -1305,7 +1267,6 @@ static void hugetlbfs_free_inode(struct inode *inode)
 static void hugetlbfs_destroy_inode(struct inode *inode)
 {
 	hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
-	mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 }
 
 static const struct address_space_operations hugetlbfs_aops = {
diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
index 5b2626063f4f..6522eb3cd007 100644
--- a/include/linux/hugetlb.h
+++ b/include/linux/hugetlb.h
@@ -30,7 +30,6 @@ void free_huge_folio(struct folio *folio);
 
 #ifdef CONFIG_HUGETLB_PAGE
 
-#include <linux/mempolicy.h>
 #include <linux/shm.h>
 #include <asm/tlbflush.h>
 
@@ -512,7 +511,6 @@ static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb)
 }
 
 struct hugetlbfs_inode_info {
-	struct shared_policy policy;
 	struct inode vfs_inode;
 	unsigned int seals;
 };
-- 
2.35.3

Re: [PATCH 01/12] hugetlbfs: drop shared NUMA mempolicy pretence

[Matthew Wilcox]

On Mon, Sep 25, 2023 at 01:21:10AM -0700, Hugh Dickins wrote:
> hugetlbfs_fallocate() goes through the motions of pasting a shared NUMA
> mempolicy onto its pseudo-vma, but how could there ever be a shared NUMA
> mempolicy for this file?  hugetlb_vm_ops has never offered a set_policy
> method, and hugetlbfs_parse_param() has never supported any mpol options
> for a mount-wide default policy.

Hah.  I was wondering, but never cared enough to investigate.  Thanks
for doing this.

Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>

Re: [PATCH 01/12] hugetlbfs: drop shared NUMA mempolicy pretence

[Andi Kleen]

On Mon, Sep 25, 2023 at 01:21:10AM -0700, Hugh Dickins wrote:
> hugetlbfs_fallocate() goes through the motions of pasting a shared NUMA
> mempolicy onto its pseudo-vma, but how could there ever be a shared NUMA
> mempolicy for this file?  hugetlb_vm_ops has never offered a set_policy
> method, and hugetlbfs_parse_param() has never supported any mpol options
> for a mount-wide default policy.
> 
> It's just an illusion: clean it away so as not to confuse others, giving
> us more freedom to adjust shmem's set_policy/get_policy implementation.
> But hugetlbfs_inode_info is still required, just to accommodate seals.
> 
> Yes, shared NUMA mempolicy support could be added to hugetlbfs, with a
> set_policy method and/or mpol mount option (Andi's first posting did
> include an admitted-unsatisfactory hugetlb_set_policy()); but it seems
> that nobody has bothered to add that in the nineteen years since v2.6.7
> made it possible, and there is at least one company that has invested
> enough into hugetlbfs, that I guess they have learnt well enough how to
> manage its NUMA, without needing shared mempolicy.

TBH i'm not sure people in general rely on shared mempolicy. The
original use case for it was to modify the numa policy of non anonymous
shared memory files without modifying the program (e.g. Oracle
database's shared memory segments)

But I don't think that particular usage model ever got any real
traction: at leas I haven't seen any real usage of it outside my tests.

I suspect people either are fine with just process policy or modify the
program, in which case it's not a big burden to modify every user,
so process policy or vma based mbind policy works fine.

Maybe it would be an interesting experiment to disable it everywhere
with some flag and see if anybody complains.

On the other hand it might be Hyrum'ed by now.

-Andi

Re: [PATCH 01/12] hugetlbfs: drop shared NUMA mempolicy pretence

[Hugh Dickins]

On Mon, 25 Sep 2023, Andi Kleen wrote:
> On Mon, Sep 25, 2023 at 01:21:10AM -0700, Hugh Dickins wrote:
> > hugetlbfs_fallocate() goes through the motions of pasting a shared NUMA
> > mempolicy onto its pseudo-vma, but how could there ever be a shared NUMA
> > mempolicy for this file?  hugetlb_vm_ops has never offered a set_policy
> > method, and hugetlbfs_parse_param() has never supported any mpol options
> > for a mount-wide default policy.
> > 
> > It's just an illusion: clean it away so as not to confuse others, giving
> > us more freedom to adjust shmem's set_policy/get_policy implementation.
> > But hugetlbfs_inode_info is still required, just to accommodate seals.
> > 
> > Yes, shared NUMA mempolicy support could be added to hugetlbfs, with a
> > set_policy method and/or mpol mount option (Andi's first posting did
> > include an admitted-unsatisfactory hugetlb_set_policy()); but it seems
> > that nobody has bothered to add that in the nineteen years since v2.6.7
> > made it possible, and there is at least one company that has invested
> > enough into hugetlbfs, that I guess they have learnt well enough how to
> > manage its NUMA, without needing shared mempolicy.
> 
> TBH i'm not sure people in general rely on shared mempolicy. The
> original use case for it was to modify the numa policy of non anonymous
> shared memory files without modifying the program (e.g. Oracle
> database's shared memory segments)

Ah, "without modifying the program": that makes a lot of sense, but
I had never thought of it that way - I just saw it as the right way to
manage the shared object (though an outlier, since we have so many other
msyscall()s which do not manage the underlying shared object in this way).

> 
> But I don't think that particular usage model ever got any real
> traction: at leas I haven't seen any real usage of it outside my tests.

If the hugetlbfs support had actually gone in, I imagine Oracle would
have managed it that way; but they seem to have survived well without.

> 
> I suspect people either are fine with just process policy or modify the
> program, in which case it's not a big burden to modify every user,
> so process policy or vma based mbind policy works fine.
> 
> Maybe it would be an interesting experiment to disable it everywhere
> with some flag and see if anybody complains.
> 
> On the other hand it might be Hyrum'ed by now.

This is interesting info, Andi, thank you for providing.

I'm torn.  shmem and mempolicy (and struct vm_operations_struct) would
certainly be simpler without shared mempolicy: but I frankly don't have
the time and courage to experiment with deprecating it now; and it is
fundamentally right that such policy should be kept with the object,
not with its mappings.  I've assumed for years that it has to stay.

Hugh

[PATCH 02/12] kernfs: drop shared NUMA mempolicy hooks

[Hugh Dickins]

It seems strange that kernfs should be an outlier with a set_policy and
get_policy in its kernfs_vm_ops.  Ah, it dates back to v2.6.30's commit
095160aee954 ("sysfs: fix some bin_vm_ops errors"), when I had crashed
on powerpc's pci_mmap_legacy_page_range() fallback to shmem_zero_setup().

Well, that was commendably thorough, to give sysfs-bin a set_policy and
get_policy, just to avoid the way it was coded resulting in EINVAL from
mmap when CONFIG_NUMA; but somehow feels a bit over-the-top to me now.

It's easier to say that nobody should expect to manage a shmem object's
shared NUMA mempolicy via some kernfs backdoor to that object: delete
that code (and there's no longer an EINVAL from mmap in the NUMA case).

This then leaves set_policy/get_policy as implemented only by shmem -
though importantly also by SysV SHM, which has to interface with shmem
which implements them, and with SHM_HUGETLB which does not.

Signed-off-by: Hugh Dickins <hughd@google.com>
---
 fs/kernfs/file.c | 49 ------------------------------------------------
 1 file changed, 49 deletions(-)

diff --git a/fs/kernfs/file.c b/fs/kernfs/file.c
index 180906c36f51..aaa76410e550 100644
--- a/fs/kernfs/file.c
+++ b/fs/kernfs/file.c
@@ -429,60 +429,11 @@ static int kernfs_vma_access(struct vm_area_struct *vma, unsigned long addr,
 	return ret;
 }
 
-#ifdef CONFIG_NUMA
-static int kernfs_vma_set_policy(struct vm_area_struct *vma,
-				 struct mempolicy *new)
-{
-	struct file *file = vma->vm_file;
-	struct kernfs_open_file *of = kernfs_of(file);
-	int ret;
-
-	if (!of->vm_ops)
-		return 0;
-
-	if (!kernfs_get_active(of->kn))
-		return -EINVAL;
-
-	ret = 0;
-	if (of->vm_ops->set_policy)
-		ret = of->vm_ops->set_policy(vma, new);
-
-	kernfs_put_active(of->kn);
-	return ret;
-}
-
-static struct mempolicy *kernfs_vma_get_policy(struct vm_area_struct *vma,
-					       unsigned long addr)
-{
-	struct file *file = vma->vm_file;
-	struct kernfs_open_file *of = kernfs_of(file);
-	struct mempolicy *pol;
-
-	if (!of->vm_ops)
-		return vma->vm_policy;
-
-	if (!kernfs_get_active(of->kn))
-		return vma->vm_policy;
-
-	pol = vma->vm_policy;
-	if (of->vm_ops->get_policy)
-		pol = of->vm_ops->get_policy(vma, addr);
-
-	kernfs_put_active(of->kn);
-	return pol;
-}
-
-#endif
-
 static const struct vm_operations_struct kernfs_vm_ops = {
 	.open		= kernfs_vma_open,
 	.fault		= kernfs_vma_fault,
 	.page_mkwrite	= kernfs_vma_page_mkwrite,
 	.access		= kernfs_vma_access,
-#ifdef CONFIG_NUMA
-	.set_policy	= kernfs_vma_set_policy,
-	.get_policy	= kernfs_vma_get_policy,
-#endif
 };
 
 static int kernfs_fop_mmap(struct file *file, struct vm_area_struct *vma)
-- 
2.35.3

Re: [PATCH 02/12] kernfs: drop shared NUMA mempolicy hooks

[Matthew Wilcox]

On Mon, Sep 25, 2023 at 01:22:27AM -0700, Hugh Dickins wrote:
> It seems strange that kernfs should be an outlier with a set_policy and
> get_policy in its kernfs_vm_ops.  Ah, it dates back to v2.6.30's commit
> 095160aee954 ("sysfs: fix some bin_vm_ops errors"), when I had crashed
> on powerpc's pci_mmap_legacy_page_range() fallback to shmem_zero_setup().
> 
> Well, that was commendably thorough, to give sysfs-bin a set_policy and
> get_policy, just to avoid the way it was coded resulting in EINVAL from
> mmap when CONFIG_NUMA; but somehow feels a bit over-the-top to me now.
> 
> It's easier to say that nobody should expect to manage a shmem object's
> shared NUMA mempolicy via some kernfs backdoor to that object: delete
> that code (and there's no longer an EINVAL from mmap in the NUMA case).
> 
> This then leaves set_policy/get_policy as implemented only by shmem -
> though importantly also by SysV SHM, which has to interface with shmem
> which implements them, and with SHM_HUGETLB which does not.
> 
> Signed-off-by: Hugh Dickins <hughd@google.com>

Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>

[PATCH 03/12] mempolicy: fix migrate_pages(2) syscall return nr_failed

[Hugh Dickins]

"man 2 migrate_pages" says "On success migrate_pages() returns the number
of pages that could not be moved".  Although 5.3 and 5.4 commits fixed
mbind(MPOL_MF_STRICT|MPOL_MF_MOVE*) to fail with EIO when not all pages
could be moved (because some could not be isolated for migration),
migrate_pages(2) was left still reporting only those pages failing at the
migration stage, forgetting those failing at the earlier isolation stage.

Fix that by accumulating a long nr_failed count in struct queue_pages,
returned by queue_pages_range() when it's not returning an error, for
adding on to the nr_failed count from migrate_pages() in mm/migrate.c.
A count of pages?  It's more a count of folios, but changing it to pages
would entail more work (also in mm/migrate.c): does not seem justified.

queue_pages_range() itself should only return -EIO in the "strictly
unmovable" case (STRICT without any MOVEs): in that case it's best to
break out as soon as nr_failed gets set; but otherwise it should continue
to isolate pages for MOVing even when nr_failed - as the mbind(2) manpage
promises.  This fixes mbind(MPOL_MF_STRICT|MPOL_MF_MOVE*) behavior left
over from 5.3 and 5.4, and the recent syzbot need for vma_start_write()
before mbind_range(): but both of those may be fixed by smaller patches.

There's a case when nr_failed should be incremented when it was missed:
queue_folios_pte_range() and queue_folios_hugetlb() count the transient
migration entries, like queue_folios_pmd() already did.  And there's a
case when nr_failed should not be incremented when it would have been:
in meeting later PTEs of the same large folio, which can only be isolated
once: fixed by recording the current large folio in struct queue_pages.

Clean up the affected functions, fixing or updating many comments.  Bool
migrate_folio_add(), without -EIO: true if adding, or if skipping shared
(but its arguable folio_estimated_sharers() heuristic left unchanged).
Use MPOL_MF_WRLOCK flag to queue_pages_range(), instead of bool lock_vma.
Use explicit STRICT|MOVE* flags where queue_pages_test_walk() checks for
skipping, instead of hiding them behind MPOL_MF_VALID.

Signed-off-by: Hugh Dickins <hughd@google.com>
---
 mm/mempolicy.c | 322 +++++++++++++++++++++++--------------------------
 1 file changed, 149 insertions(+), 173 deletions(-)

diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 42b5567e3773..937386409c28 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -111,7 +111,8 @@
 
 /* Internal flags */
 #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0)	/* Skip checks for continuous vmas */
-#define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1)		/* Invert check for nodemask */
+#define MPOL_MF_INVERT       (MPOL_MF_INTERNAL << 1)	/* Invert check for nodemask */
+#define MPOL_MF_WRLOCK       (MPOL_MF_INTERNAL << 2)	/* Write-lock walked vmas */
 
 static struct kmem_cache *policy_cache;
 static struct kmem_cache *sn_cache;
@@ -416,9 +417,19 @@ static const struct mempolicy_operations mpol_ops[MPOL_MAX] = {
 	},
 };
 
-static int migrate_folio_add(struct folio *folio, struct list_head *foliolist,
+static bool migrate_folio_add(struct folio *folio, struct list_head *foliolist,
 				unsigned long flags);
 
+static bool strictly_unmovable(unsigned long flags)
+{
+	/*
+	 * STRICT without MOVE flags lets do_mbind() fail immediately with -EIO
+	 * if any misplaced page is found.
+	 */
+	return (flags & (MPOL_MF_STRICT | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) ==
+			 MPOL_MF_STRICT;
+}
+
 struct queue_pages {
 	struct list_head *pagelist;
 	unsigned long flags;
@@ -426,6 +437,8 @@ struct queue_pages {
 	unsigned long start;
 	unsigned long end;
 	struct vm_area_struct *first;
+	struct folio *large;		/* note last large folio on pagelist */
+	long nr_failed;			/* could not be isolated at this time */
 };
 
 /*
@@ -443,50 +456,27 @@ static inline bool queue_folio_required(struct folio *folio,
 	return node_isset(nid, *qp->nmask) == !(flags & MPOL_MF_INVERT);
 }
 
-/*
- * queue_folios_pmd() has three possible return values:
- * 0 - folios are placed on the right node or queued successfully, or
- *     special page is met, i.e. huge zero page.
- * 1 - there is unmovable folio, and MPOL_MF_MOVE* & MPOL_MF_STRICT were
- *     specified.
- * -EIO - is migration entry or only MPOL_MF_STRICT was specified and an
- *        existing folio was already on a node that does not follow the
- *        policy.
- */
-static int queue_folios_pmd(pmd_t *pmd, spinlock_t *ptl, unsigned long addr,
+static void queue_folios_pmd(pmd_t *pmd, spinlock_t *ptl, unsigned long addr,
 				unsigned long end, struct mm_walk *walk)
-	__releases(ptl)
 {
-	int ret = 0;
 	struct folio *folio;
 	struct queue_pages *qp = walk->private;
-	unsigned long flags;
 
 	if (unlikely(is_pmd_migration_entry(*pmd))) {
-		ret = -EIO;
-		goto unlock;
+		qp->nr_failed++;
+		return;
 	}
 	folio = pfn_folio(pmd_pfn(*pmd));
 	if (is_huge_zero_page(&folio->page)) {
 		walk->action = ACTION_CONTINUE;
-		goto unlock;
+		return;
 	}
 	if (!queue_folio_required(folio, qp))
-		goto unlock;
-
-	flags = qp->flags;
-	/* go to folio migration */
-	if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
-		if (!vma_migratable(walk->vma) ||
-		    migrate_folio_add(folio, qp->pagelist, flags)) {
-			ret = 1;
-			goto unlock;
-		}
-	} else
-		ret = -EIO;
-unlock:
-	spin_unlock(ptl);
-	return ret;
+		return;
+	if (!(qp->flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) ||
+	    !vma_migratable(walk->vma) ||
+	    !migrate_folio_add(folio, qp->pagelist, qp->flags))
+		qp->nr_failed++;
 }
 
 /*
@@ -496,8 +486,6 @@ static int queue_folios_pmd(pmd_t *pmd, spinlock_t *ptl, unsigned long addr,
  * queue_folios_pte_range() has three possible return values:
  * 0 - folios are placed on the right node or queued successfully, or
  *     special page is met, i.e. zero page.
- * 1 - there is unmovable folio, and MPOL_MF_MOVE* & MPOL_MF_STRICT were
- *     specified.
  * -EIO - only MPOL_MF_STRICT was specified and an existing folio was already
  *        on a node that does not follow the policy.
  */
@@ -508,14 +496,16 @@ static int queue_folios_pte_range(pmd_t *pmd, unsigned long addr,
 	struct folio *folio;
 	struct queue_pages *qp = walk->private;
 	unsigned long flags = qp->flags;
-	bool has_unmovable = false;
 	pte_t *pte, *mapped_pte;
 	pte_t ptent;
 	spinlock_t *ptl;
 
 	ptl = pmd_trans_huge_lock(pmd, vma);
-	if (ptl)
-		return queue_folios_pmd(pmd, ptl, addr, end, walk);
+	if (ptl) {
+		queue_folios_pmd(pmd, ptl, addr, end, walk);
+		spin_unlock(ptl);
+		goto out;
+	}
 
 	mapped_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
 	if (!pte) {
@@ -524,8 +514,13 @@ static int queue_folios_pte_range(pmd_t *pmd, unsigned long addr,
 	}
 	for (; addr != end; pte++, addr += PAGE_SIZE) {
 		ptent = ptep_get(pte);
-		if (!pte_present(ptent))
+		if (pte_none(ptent))
 			continue;
+		if (!pte_present(ptent)) {
+			if (is_migration_entry(pte_to_swp_entry(ptent)))
+				qp->nr_failed++;
+			continue;
+		}
 		folio = vm_normal_folio(vma, addr, ptent);
 		if (!folio || folio_is_zone_device(folio))
 			continue;
@@ -537,97 +532,82 @@ static int queue_folios_pte_range(pmd_t *pmd, unsigned long addr,
 			continue;
 		if (!queue_folio_required(folio, qp))
 			continue;
-		if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
-			/* MPOL_MF_STRICT must be specified if we get here */
-			if (!vma_migratable(vma)) {
-				has_unmovable = true;
+		if (!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) ||
+		    !vma_migratable(vma)) {
+			qp->nr_failed++;
+			if (strictly_unmovable(flags))
 				break;
-			}
-
+		}
+		if (migrate_folio_add(folio, qp->pagelist, flags)) {
 			/*
-			 * Do not abort immediately since there may be
-			 * temporary off LRU pages in the range.  Still
-			 * need migrate other LRU pages.
+			 * A large folio can only be isolated from LRU once,
+			 * but may be mapped by many PTEs (and Copy-On-Write may
+			 * intersperse PTEs of other folios).  This is a common
+			 * case, so don't mistake it for failure (but of course
+			 * there can be other cases of multi-mapped pages which
+			 * this quick check does not help to filter out - and a
+			 * search of the pagelist might grow to be prohibitive).
 			 */
-			if (migrate_folio_add(folio, qp->pagelist, flags))
-				has_unmovable = true;
-		} else
-			break;
+			if (folio_test_large(folio))
+				qp->large = folio;
+		} else if (folio != qp->large) {
+			qp->nr_failed++;
+			if (strictly_unmovable(flags))
+				break;
+		}
 	}
 	pte_unmap_unlock(mapped_pte, ptl);
 	cond_resched();
-
-	if (has_unmovable)
-		return 1;
-
-	return addr != end ? -EIO : 0;
+out:
+	if (qp->nr_failed && strictly_unmovable(flags))
+		return -EIO;
+	return 0;
 }
 
 static int queue_folios_hugetlb(pte_t *pte, unsigned long hmask,
 			       unsigned long addr, unsigned long end,
 			       struct mm_walk *walk)
 {
-	int ret = 0;
 #ifdef CONFIG_HUGETLB_PAGE
 	struct queue_pages *qp = walk->private;
-	unsigned long flags = (qp->flags & MPOL_MF_VALID);
+	unsigned long flags = qp->flags;
 	struct folio *folio;
 	spinlock_t *ptl;
 	pte_t entry;
 
 	ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte);
 	entry = huge_ptep_get(pte);
-	if (!pte_present(entry))
+	if (!pte_present(entry)) {
+		if (unlikely(is_hugetlb_entry_migration(entry)))
+			qp->nr_failed++;
 		goto unlock;
+	}
 	folio = pfn_folio(pte_pfn(entry));
 	if (!queue_folio_required(folio, qp))
 		goto unlock;
-
-	if (flags == MPOL_MF_STRICT) {
-		/*
-		 * STRICT alone means only detecting misplaced folio and no
-		 * need to further check other vma.
-		 */
-		ret = -EIO;
+	if (!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) ||
+	    !vma_migratable(walk->vma)) {
+		qp->nr_failed++;
 		goto unlock;
 	}
-
-	if (!vma_migratable(walk->vma)) {
-		/*
-		 * Must be STRICT with MOVE*, otherwise .test_walk() have
-		 * stopped walking current vma.
-		 * Detecting misplaced folio but allow migrating folios which
-		 * have been queued.
-		 */
-		ret = 1;
-		goto unlock;
-	}
-
 	/*
-	 * With MPOL_MF_MOVE, we try to migrate only unshared folios. If it
-	 * is shared it is likely not worth migrating.
+	 * Unless MPOL_MF_MOVE_ALL, we try to avoid migrating a shared folio.
+	 * Choosing not to migrate a shared folio is not counted as a failure.
 	 *
 	 * To check if the folio is shared, ideally we want to make sure
 	 * every page is mapped to the same process. Doing that is very
-	 * expensive, so check the estimated mapcount of the folio instead.
+	 * expensive, so check the estimated sharers of the folio instead.
 	 */
-	if (flags & (MPOL_MF_MOVE_ALL) ||
-	    (flags & MPOL_MF_MOVE && folio_estimated_sharers(folio) == 1 &&
-	     !hugetlb_pmd_shared(pte))) {
-		if (!isolate_hugetlb(folio, qp->pagelist) &&
-			(flags & MPOL_MF_STRICT))
-			/*
-			 * Failed to isolate folio but allow migrating pages
-			 * which have been queued.
-			 */
-			ret = 1;
-	}
+	if ((flags & MPOL_MF_MOVE_ALL) ||
+	    (folio_estimated_sharers(folio) == 1 && !hugetlb_pmd_shared(pte)))
+		if (!isolate_hugetlb(folio, qp->pagelist))
+			qp->nr_failed++;
 unlock:
 	spin_unlock(ptl);
-#else
-	BUG();
+	if (qp->nr_failed && strictly_unmovable(flags))
+		return -EIO;
 #endif
-	return ret;
+	return 0;
 }
 
 #ifdef CONFIG_NUMA_BALANCING
@@ -708,8 +688,11 @@ static int queue_pages_test_walk(unsigned long start, unsigned long end,
 		return 1;
 	}
 
-	/* queue pages from current vma */
-	if (flags & MPOL_MF_VALID)
+	/*
+	 * Check page nodes, and queue pages to move, in the current vma.
+	 * But if no moving, and no strict checking, the scan can be skipped.
+	 */
+	if (flags & (MPOL_MF_STRICT | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
 		return 0;
 	return 1;
 }
@@ -731,22 +714,22 @@ static const struct mm_walk_ops queue_pages_lock_vma_walk_ops = {
 /*
  * Walk through page tables and collect pages to be migrated.
  *
- * If pages found in a given range are on a set of nodes (determined by
- * @nodes and @flags,) it's isolated and queued to the pagelist which is
- * passed via @private.
+ * If pages found in a given range are not on the required set of @nodes,
+ * and migration is allowed, they are isolated and queued to the pagelist
+ * which is passed via @private.
  *
- * queue_pages_range() has three possible return values:
- * 1 - there is unmovable page, but MPOL_MF_MOVE* & MPOL_MF_STRICT were
- *     specified.
- * 0 - queue pages successfully or no misplaced page.
- * errno - i.e. misplaced pages with MPOL_MF_STRICT specified (-EIO) or
- *         memory range specified by nodemask and maxnode points outside
- *         your accessible address space (-EFAULT)
+ * queue_pages_range() may return:
+ * 0 - all pages already on the right node, or successfully queued for moving
+ *     (or neither strict checking nor moving requested: only range checking).
+ * >0 - this number of misplaced folios could not be queued for moving
+ *      (a hugetlbfs page or a transparent huge page being counted as 1).
+ * -EIO - a misplaced page found, when MPOL_MF_STRICT specified without MOVEs.
+ * -EFAULT - a hole in the memory range, when MPOL_MF_DISCONTIG_OK unspecified.
  */
-static int
+static long
 queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end,
 		nodemask_t *nodes, unsigned long flags,
-		struct list_head *pagelist, bool lock_vma)
+		struct list_head *pagelist)
 {
 	int err;
 	struct queue_pages qp = {
@@ -757,7 +740,7 @@ queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end,
 		.end = end,
 		.first = NULL,
 	};
-	const struct mm_walk_ops *ops = lock_vma ?
+	const struct mm_walk_ops *ops = (flags & MPOL_MF_WRLOCK) ?
 			&queue_pages_lock_vma_walk_ops : &queue_pages_walk_ops;
 
 	err = walk_page_range(mm, start, end, ops, &qp);
@@ -766,7 +749,7 @@ queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end,
 		/* whole range in hole */
 		err = -EFAULT;
 
-	return err;
+	return err ? : qp.nr_failed;
 }
 
 /*
@@ -1029,16 +1012,16 @@ static long do_get_mempolicy(int *policy, nodemask_t *nmask,
 }
 
 #ifdef CONFIG_MIGRATION
-static int migrate_folio_add(struct folio *folio, struct list_head *foliolist,
+static bool migrate_folio_add(struct folio *folio, struct list_head *foliolist,
 				unsigned long flags)
 {
 	/*
-	 * We try to migrate only unshared folios. If it is shared it
-	 * is likely not worth migrating.
+	 * Unless MPOL_MF_MOVE_ALL, we try to avoid migrating a shared folio.
+	 * Choosing not to migrate a shared folio is not counted as a failure.
 	 *
 	 * To check if the folio is shared, ideally we want to make sure
 	 * every page is mapped to the same process. Doing that is very
-	 * expensive, so check the estimated mapcount of the folio instead.
+	 * expensive, so check the estimated sharers of the folio instead.
 	 */
 	if ((flags & MPOL_MF_MOVE_ALL) || folio_estimated_sharers(folio) == 1) {
 		if (folio_isolate_lru(folio)) {
@@ -1046,32 +1029,31 @@ static int migrate_folio_add(struct folio *folio, struct list_head *foliolist,
 			node_stat_mod_folio(folio,
 				NR_ISOLATED_ANON + folio_is_file_lru(folio),
 				folio_nr_pages(folio));
-		} else if (flags & MPOL_MF_STRICT) {
+		} else {
 			/*
 			 * Non-movable folio may reach here.  And, there may be
 			 * temporary off LRU folios or non-LRU movable folios.
 			 * Treat them as unmovable folios since they can't be
-			 * isolated, so they can't be moved at the moment.  It
-			 * should return -EIO for this case too.
+			 * isolated, so they can't be moved at the moment.
 			 */
-			return -EIO;
+			return false;
 		}
 	}
-
-	return 0;
+	return true;
 }
 
 /*
  * Migrate pages from one node to a target node.
  * Returns error or the number of pages not migrated.
  */
-static int migrate_to_node(struct mm_struct *mm, int source, int dest,
-			   int flags)
+static long migrate_to_node(struct mm_struct *mm, int source, int dest,
+			    int flags)
 {
 	nodemask_t nmask;
 	struct vm_area_struct *vma;
 	LIST_HEAD(pagelist);
-	int err = 0;
+	long nr_failed;
+	long err = 0;
 	struct migration_target_control mtc = {
 		.nid = dest,
 		.gfp_mask = GFP_HIGHUSER_MOVABLE | __GFP_THISNODE,
@@ -1080,23 +1062,27 @@ static int migrate_to_node(struct mm_struct *mm, int source, int dest,
 	nodes_clear(nmask);
 	node_set(source, nmask);
 
-	/*
-	 * This does not "check" the range but isolates all pages that
-	 * need migration.  Between passing in the full user address
-	 * space range and MPOL_MF_DISCONTIG_OK, this call can not fail.
-	 */
-	vma = find_vma(mm, 0);
 	VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)));
-	queue_pages_range(mm, vma->vm_start, mm->task_size, &nmask,
-			flags | MPOL_MF_DISCONTIG_OK, &pagelist, false);
+	vma = find_vma(mm, 0);
+
+	/*
+	 * This does not migrate the range, but isolates all pages that
+	 * need migration.  Between passing in the full user address
+	 * space range and MPOL_MF_DISCONTIG_OK, this call cannot fail,
+	 * but passes back the count of pages which could not be isolated.
+	 */
+	nr_failed = queue_pages_range(mm, vma->vm_start, mm->task_size, &nmask,
+				      flags | MPOL_MF_DISCONTIG_OK, &pagelist);
 
 	if (!list_empty(&pagelist)) {
 		err = migrate_pages(&pagelist, alloc_migration_target, NULL,
-				(unsigned long)&mtc, MIGRATE_SYNC, MR_SYSCALL, NULL);
+			(unsigned long)&mtc, MIGRATE_SYNC, MR_SYSCALL, NULL);
 		if (err)
 			putback_movable_pages(&pagelist);
 	}
 
+	if (err >= 0)
+		err += nr_failed;
 	return err;
 }
 
@@ -1109,8 +1095,8 @@ static int migrate_to_node(struct mm_struct *mm, int source, int dest,
 int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
 		     const nodemask_t *to, int flags)
 {
-	int busy = 0;
-	int err = 0;
+	long nr_failed = 0;
+	long err = 0;
 	nodemask_t tmp;
 
 	lru_cache_disable();
@@ -1192,7 +1178,7 @@ int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
 		node_clear(source, tmp);
 		err = migrate_to_node(mm, source, dest, flags);
 		if (err > 0)
-			busy += err;
+			nr_failed += err;
 		if (err < 0)
 			break;
 	}
@@ -1201,8 +1187,7 @@ int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
 	lru_cache_enable();
 	if (err < 0)
 		return err;
-	return busy;
-
+	return (nr_failed < INT_MAX) ? nr_failed : INT_MAX;
 }
 
 /*
@@ -1241,10 +1226,10 @@ static struct folio *new_folio(struct folio *src, unsigned long start)
 }
 #else
 
-static int migrate_folio_add(struct folio *folio, struct list_head *foliolist,
+static bool migrate_folio_add(struct folio *folio, struct list_head *foliolist,
 				unsigned long flags)
 {
-	return -EIO;
+	return false;
 }
 
 int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
@@ -1268,8 +1253,8 @@ static long do_mbind(unsigned long start, unsigned long len,
 	struct vma_iterator vmi;
 	struct mempolicy *new;
 	unsigned long end;
-	int err;
-	int ret;
+	long err;
+	long nr_failed;
 	LIST_HEAD(pagelist);
 
 	if (flags & ~(unsigned long)MPOL_MF_VALID)
@@ -1309,10 +1294,8 @@ static long do_mbind(unsigned long start, unsigned long len,
 		 start, start + len, mode, mode_flags,
 		 nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE);
 
-	if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
-
+	if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
 		lru_cache_disable();
-	}
 	{
 		NODEMASK_SCRATCH(scratch);
 		if (scratch) {
@@ -1328,44 +1311,37 @@ static long do_mbind(unsigned long start, unsigned long len,
 		goto mpol_out;
 
 	/*
-	 * Lock the VMAs before scanning for pages to migrate, to ensure we don't
-	 * miss a concurrently inserted page.
+	 * Lock the VMAs before scanning for pages to migrate,
+	 * to ensure we don't miss a concurrently inserted page.
 	 */
-	ret = queue_pages_range(mm, start, end, nmask,
-			  flags | MPOL_MF_INVERT, &pagelist, true);
+	nr_failed = queue_pages_range(mm, start, end, nmask,
+			flags | MPOL_MF_INVERT | MPOL_MF_WRLOCK, &pagelist);
 
-	if (ret < 0) {
-		err = ret;
-		goto up_out;
-	}
-
-	vma_iter_init(&vmi, mm, start);
-	prev = vma_prev(&vmi);
-	for_each_vma_range(vmi, vma, end) {
-		err = mbind_range(&vmi, vma, &prev, start, end, new);
-		if (err)
-			break;
+	if (nr_failed < 0) {
+		err = nr_failed;
+	} else {
+		vma_iter_init(&vmi, mm, start);
+		prev = vma_prev(&vmi);
+		for_each_vma_range(vmi, vma, end) {
+			err = mbind_range(&vmi, vma, &prev, start, end, new);
+			if (err)
+				break;
+		}
 	}
 
 	if (!err) {
-		int nr_failed = 0;
-
 		if (!list_empty(&pagelist)) {
 			WARN_ON_ONCE(flags & MPOL_MF_LAZY);
-			nr_failed = migrate_pages(&pagelist, new_folio, NULL,
+			nr_failed |= migrate_pages(&pagelist, new_folio, NULL,
 				start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND, NULL);
-			if (nr_failed)
-				putback_movable_pages(&pagelist);
 		}
-
-		if ((ret > 0) || (nr_failed && (flags & MPOL_MF_STRICT)))
+		if (nr_failed && (flags & MPOL_MF_STRICT))
 			err = -EIO;
-	} else {
-up_out:
-		if (!list_empty(&pagelist))
-			putback_movable_pages(&pagelist);
 	}
 
+	if (!list_empty(&pagelist))
+		putback_movable_pages(&pagelist);
+
 	mmap_write_unlock(mm);
 mpol_out:
 	mpol_put(new);
-- 
2.35.3

Re: [PATCH 03/12] mempolicy: fix migrate_pages(2) syscall return nr_failed

[Matthew Wilcox]

On Mon, Sep 25, 2023 at 01:24:02AM -0700, Hugh Dickins wrote:
> "man 2 migrate_pages" says "On success migrate_pages() returns the number
> of pages that could not be moved".  Although 5.3 and 5.4 commits fixed
> mbind(MPOL_MF_STRICT|MPOL_MF_MOVE*) to fail with EIO when not all pages
> could be moved (because some could not be isolated for migration),
> migrate_pages(2) was left still reporting only those pages failing at the
> migration stage, forgetting those failing at the earlier isolation stage.
> 
> Fix that by accumulating a long nr_failed count in struct queue_pages,
> returned by queue_pages_range() when it's not returning an error, for
> adding on to the nr_failed count from migrate_pages() in mm/migrate.c.
> A count of pages?  It's more a count of folios, but changing it to pages
> would entail more work (also in mm/migrate.c): does not seem justified.

I certainly see what you're saying.  If a folio is only partially mapped
(in an extreme case, the VMA is PAGE_SIZE and maps one page of a 512-page
folio), then setting nr_failed to folio_nr_pages() is misleading at best.

> +static void queue_folios_pmd(pmd_t *pmd, spinlock_t *ptl, unsigned long addr,
>  				unsigned long end, struct mm_walk *walk)
> -	__releases(ptl)
>  {
> -	int ret = 0;
>  	struct folio *folio;
>  	struct queue_pages *qp = walk->private;
> -	unsigned long flags;
>  
>  	if (unlikely(is_pmd_migration_entry(*pmd))) {
> -		ret = -EIO;
> -		goto unlock;
> +		qp->nr_failed++;
> +		return;
>  	}
>  	folio = pfn_folio(pmd_pfn(*pmd));
>  	if (is_huge_zero_page(&folio->page)) {
>  		walk->action = ACTION_CONTINUE;
> -		goto unlock;
> +		return;
>  	}
>  	if (!queue_folio_required(folio, qp))
> -		goto unlock;
> -
> -	flags = qp->flags;
> -	/* go to folio migration */
> -	if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
> -		if (!vma_migratable(walk->vma) ||
> -		    migrate_folio_add(folio, qp->pagelist, flags)) {
> -			ret = 1;
> -			goto unlock;
> -		}
> -	} else
> -		ret = -EIO;
> -unlock:
> -	spin_unlock(ptl);
> -	return ret;
> +		return;
> +	if (!(qp->flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) ||
> +	    !vma_migratable(walk->vma) ||
> +	    !migrate_folio_add(folio, qp->pagelist, qp->flags))
> +		qp->nr_failed++;

However, I think here, we would do well to increment by HPAGE_PMD_NR.
Or whatever equivalent is flavour of the week.

Bravo to the other changes.

Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>

Re: [PATCH 03/12] mempolicy: fix migrate_pages(2) syscall return nr_failed

[Hugh Dickins]

On Mon, 25 Sep 2023, Matthew Wilcox wrote:
> On Mon, Sep 25, 2023 at 01:24:02AM -0700, Hugh Dickins wrote:
> > "man 2 migrate_pages" says "On success migrate_pages() returns the number
> > of pages that could not be moved".  Although 5.3 and 5.4 commits fixed
> > mbind(MPOL_MF_STRICT|MPOL_MF_MOVE*) to fail with EIO when not all pages
> > could be moved (because some could not be isolated for migration),
> > migrate_pages(2) was left still reporting only those pages failing at the
> > migration stage, forgetting those failing at the earlier isolation stage.
> > 
> > Fix that by accumulating a long nr_failed count in struct queue_pages,
> > returned by queue_pages_range() when it's not returning an error, for
> > adding on to the nr_failed count from migrate_pages() in mm/migrate.c.
> > A count of pages?  It's more a count of folios, but changing it to pages
> > would entail more work (also in mm/migrate.c): does not seem justified.
> 
> I certainly see what you're saying.  If a folio is only partially mapped
> (in an extreme case, the VMA is PAGE_SIZE and maps one page of a 512-page
> folio), then setting nr_failed to folio_nr_pages() is misleading at best.

Actually, that wasn't what I was thinking when I said that: but thank you
for the comment, you've helped me to see that what I'm actually doing is
not what is claimed there.

What I was thinking, something I'm taking as an axiom, is that the units
of failure when isolating must match the units of failure when migrating,
whatever they are.

And migrate_pages(), the internal one, has this helpfully explicit comment:
 * Returns the number of {normal folio, large folio, hugetlb} that were not
 * migrated, or an error code. The number of large folio splits will be
 * considered as the number of non-migrated large folio, no matter how many
 * split folios of the large folio are migrated successfully.

(TBH I haven't spent long enough to actually understand what the second
sentence is saying: I do realize that splits complicate the issue, but the
function wouldn't be expected to return a "number of large folio splits"
anyway.  One day, I should work out what the code is actually doing, and
try to reword that sentence better.)

So above I was trying to say that migrate_pages(), the syscall, returns
that quantity: totalling the failed-isolation and failed-migration folios.

But you've alerted me to how in fact I'm doing an nr_failed++ for each
PTE of a failing-to-isolate folio, not as claimed.  It looks like I need
to record "qp->large" in the case of failure as well as success.  (And
then bother about when isolation fails on the first PTE, but succeeds
by the time of a later PTE? maybe, or maybe that just gets silly.)
I must fix that in v2.

> 
> > +static void queue_folios_pmd(pmd_t *pmd, spinlock_t *ptl, unsigned long addr,
> >  				unsigned long end, struct mm_walk *walk)
...
> > +	if (!(qp->flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) ||
> > +	    !vma_migratable(walk->vma) ||
> > +	    !migrate_folio_add(folio, qp->pagelist, qp->flags))
> > +		qp->nr_failed++;
> 
> However, I think here, we would do well to increment by HPAGE_PMD_NR.
> Or whatever equivalent is flavour of the week.

I *really* wanted to do that (and increment nr_failed PTE by PTE as I'm
doing, rather than as I claimed), and gave it some thought: but I don't
think it can be done - or not without abandoning the axiom (in which
case it's impossible to say what migrate_pages(2) is counting), or
adding a layer of complication which simply isn't justifiable.

Certainly we could change the definition of what migrate_pages(internal)
returns (though I haven't researched who depends on it: IIRC-long-ago
there's maybe only one other caller who cares, to update a stat); but
that still would not help.  Because whether migrate_pages(internal)
returns 1 or HPAGE_PMD_NR for an unmigratable and unsplittable THP, it
has no idea whether that THP got into the pagelist via a PMD or via one
or some number more of PTEs.  More info would have to be passed down
separately, folio by folio: an auxiliary xarray perhaps, but let's not.

If it turns out that I'm deluded, and it can be easily done, please
clarify one point: you made this comment on queue_folios_pmd(), but
what about queue_folios_hugetlb()?  Would you nowadays prefer hugetlb
to count 1 or folio_nr_pages()?  I think the latter.

> 
> Bravo to the other changes.

Thanks - I'm guessing your enthusiasm is mainly due to that "qp->large"
realization, which we ought to have thought of before.  I'm afraid it's
going to get more complicated, once COWs are feeding on Ryan's ALFalfA -
might need large[MAX_ORDER], or some better way.  But no great hurry,
nothing will crash if it's occasionally not-quite-right.

> 
> Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>

Many thanks for all these rapid and encouraging reviews.

Hugh

Re: [PATCH 03/12] mempolicy: fix migrate_pages(2) syscall return nr_failed

[Huang, Ying]

Hugh Dickins <hughd@google.com> writes:

> "man 2 migrate_pages" says "On success migrate_pages() returns the number
> of pages that could not be moved".  Although 5.3 and 5.4 commits fixed
> mbind(MPOL_MF_STRICT|MPOL_MF_MOVE*) to fail with EIO when not all pages
> could be moved (because some could not be isolated for migration),
> migrate_pages(2) was left still reporting only those pages failing at the
> migration stage, forgetting those failing at the earlier isolation stage.
>
> Fix that by accumulating a long nr_failed count in struct queue_pages,
> returned by queue_pages_range() when it's not returning an error, for
> adding on to the nr_failed count from migrate_pages() in mm/migrate.c.
> A count of pages?  It's more a count of folios, but changing it to pages
> would entail more work (also in mm/migrate.c): does not seem justified.
>
> queue_pages_range() itself should only return -EIO in the "strictly
> unmovable" case (STRICT without any MOVEs): in that case it's best to
> break out as soon as nr_failed gets set; but otherwise it should continue
> to isolate pages for MOVing even when nr_failed - as the mbind(2) manpage
> promises.  This fixes mbind(MPOL_MF_STRICT|MPOL_MF_MOVE*) behavior left
> over from 5.3 and 5.4, and the recent syzbot need for vma_start_write()
> before mbind_range(): but both of those may be fixed by smaller patches.
>
> There's a case when nr_failed should be incremented when it was missed:
> queue_folios_pte_range() and queue_folios_hugetlb() count the transient
> migration entries, like queue_folios_pmd() already did.  And there's a
> case when nr_failed should not be incremented when it would have been:
> in meeting later PTEs of the same large folio, which can only be isolated
> once: fixed by recording the current large folio in struct queue_pages.
>
> Clean up the affected functions, fixing or updating many comments.  Bool
> migrate_folio_add(), without -EIO: true if adding, or if skipping shared
> (but its arguable folio_estimated_sharers() heuristic left unchanged).
> Use MPOL_MF_WRLOCK flag to queue_pages_range(), instead of bool lock_vma.
> Use explicit STRICT|MOVE* flags where queue_pages_test_walk() checks for
> skipping, instead of hiding them behind MPOL_MF_VALID.
>
> Signed-off-by: Hugh Dickins <hughd@google.com>
> ---
>  mm/mempolicy.c | 322 +++++++++++++++++++++++--------------------------
>  1 file changed, 149 insertions(+), 173 deletions(-)
>
> diff --git a/mm/mempolicy.c b/mm/mempolicy.c
> index 42b5567e3773..937386409c28 100644
> --- a/mm/mempolicy.c
> +++ b/mm/mempolicy.c
> @@ -111,7 +111,8 @@
>  
>  /* Internal flags */
>  #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0)	/* Skip checks for continuous vmas */
> -#define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1)		/* Invert check for nodemask */
> +#define MPOL_MF_INVERT       (MPOL_MF_INTERNAL << 1)	/* Invert check for nodemask */
> +#define MPOL_MF_WRLOCK       (MPOL_MF_INTERNAL << 2)	/* Write-lock walked vmas */
>  
>  static struct kmem_cache *policy_cache;
>  static struct kmem_cache *sn_cache;
> @@ -416,9 +417,19 @@ static const struct mempolicy_operations mpol_ops[MPOL_MAX] = {
>  	},
>  };
>  
> -static int migrate_folio_add(struct folio *folio, struct list_head *foliolist,
> +static bool migrate_folio_add(struct folio *folio, struct list_head *foliolist,
>  				unsigned long flags);
>  
> +static bool strictly_unmovable(unsigned long flags)
> +{
> +	/*
> +	 * STRICT without MOVE flags lets do_mbind() fail immediately with -EIO
> +	 * if any misplaced page is found.
> +	 */
> +	return (flags & (MPOL_MF_STRICT | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) ==
> +			 MPOL_MF_STRICT;
> +}
> +
>  struct queue_pages {
>  	struct list_head *pagelist;
>  	unsigned long flags;
> @@ -426,6 +437,8 @@ struct queue_pages {
>  	unsigned long start;
>  	unsigned long end;
>  	struct vm_area_struct *first;
> +	struct folio *large;		/* note last large folio on pagelist */
> +	long nr_failed;			/* could not be isolated at this time */
>  };
>  
>  /*
> @@ -443,50 +456,27 @@ static inline bool queue_folio_required(struct folio *folio,
>  	return node_isset(nid, *qp->nmask) == !(flags & MPOL_MF_INVERT);
>  }
>  
> -/*
> - * queue_folios_pmd() has three possible return values:
> - * 0 - folios are placed on the right node or queued successfully, or
> - *     special page is met, i.e. huge zero page.
> - * 1 - there is unmovable folio, and MPOL_MF_MOVE* & MPOL_MF_STRICT were
> - *     specified.
> - * -EIO - is migration entry or only MPOL_MF_STRICT was specified and an
> - *        existing folio was already on a node that does not follow the
> - *        policy.
> - */
> -static int queue_folios_pmd(pmd_t *pmd, spinlock_t *ptl, unsigned long addr,
> +static void queue_folios_pmd(pmd_t *pmd, spinlock_t *ptl, unsigned long addr,

I don't find that "ptl" is used in the function now.  So, remove it?

>  				unsigned long end, struct mm_walk *walk)
> -	__releases(ptl)
>  {
> -	int ret = 0;
>  	struct folio *folio;
>  	struct queue_pages *qp = walk->private;
> -	unsigned long flags;
>  
>  	if (unlikely(is_pmd_migration_entry(*pmd))) {
> -		ret = -EIO;
> -		goto unlock;
> +		qp->nr_failed++;
> +		return;
>  	}
>  	folio = pfn_folio(pmd_pfn(*pmd));
>  	if (is_huge_zero_page(&folio->page)) {
>  		walk->action = ACTION_CONTINUE;
> -		goto unlock;
> +		return;
>  	}
>  	if (!queue_folio_required(folio, qp))
> -		goto unlock;
> -
> -	flags = qp->flags;
> -	/* go to folio migration */
> -	if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
> -		if (!vma_migratable(walk->vma) ||
> -		    migrate_folio_add(folio, qp->pagelist, flags)) {
> -			ret = 1;
> -			goto unlock;
> -		}
> -	} else
> -		ret = -EIO;
> -unlock:
> -	spin_unlock(ptl);
> -	return ret;
> +		return;
> +	if (!(qp->flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) ||
> +	    !vma_migratable(walk->vma) ||
> +	    !migrate_folio_add(folio, qp->pagelist, qp->flags))
> +		qp->nr_failed++;
>  }
>  
>  /*
> @@ -496,8 +486,6 @@ static int queue_folios_pmd(pmd_t *pmd, spinlock_t *ptl, unsigned long addr,
>   * queue_folios_pte_range() has three possible return values:
>   * 0 - folios are placed on the right node or queued successfully, or
>   *     special page is met, i.e. zero page.
> - * 1 - there is unmovable folio, and MPOL_MF_MOVE* & MPOL_MF_STRICT were
> - *     specified.
>   * -EIO - only MPOL_MF_STRICT was specified and an existing folio was already
>   *        on a node that does not follow the policy.
>   */
> @@ -508,14 +496,16 @@ static int queue_folios_pte_range(pmd_t *pmd, unsigned long addr,
>  	struct folio *folio;
>  	struct queue_pages *qp = walk->private;
>  	unsigned long flags = qp->flags;
> -	bool has_unmovable = false;
>  	pte_t *pte, *mapped_pte;
>  	pte_t ptent;
>  	spinlock_t *ptl;
>  
>  	ptl = pmd_trans_huge_lock(pmd, vma);
> -	if (ptl)
> -		return queue_folios_pmd(pmd, ptl, addr, end, walk);
> +	if (ptl) {
> +		queue_folios_pmd(pmd, ptl, addr, end, walk);
> +		spin_unlock(ptl);
> +		goto out;
> +	}
>  
>  	mapped_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
>  	if (!pte) {
> @@ -524,8 +514,13 @@ static int queue_folios_pte_range(pmd_t *pmd, unsigned long addr,
>  	}
>  	for (; addr != end; pte++, addr += PAGE_SIZE) {
>  		ptent = ptep_get(pte);
> -		if (!pte_present(ptent))
> +		if (pte_none(ptent))
>  			continue;
> +		if (!pte_present(ptent)) {
> +			if (is_migration_entry(pte_to_swp_entry(ptent)))
> +				qp->nr_failed++;
> +			continue;
> +		}
>  		folio = vm_normal_folio(vma, addr, ptent);
>  		if (!folio || folio_is_zone_device(folio))
>  			continue;
> @@ -537,97 +532,82 @@ static int queue_folios_pte_range(pmd_t *pmd, unsigned long addr,
>  			continue;
>  		if (!queue_folio_required(folio, qp))
>  			continue;
> -		if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
> -			/* MPOL_MF_STRICT must be specified if we get here */
> -			if (!vma_migratable(vma)) {
> -				has_unmovable = true;
> +		if (!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) ||
> +		    !vma_migratable(vma)) {
> +			qp->nr_failed++;
> +			if (strictly_unmovable(flags))
>  				break;
> -			}
> -
> +		}

IIUC, even if !(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) or
!vma_migratable(vma), the folio will be isolated in migrate_folio_add()
below.  Is this the expected behavior?

> +		if (migrate_folio_add(folio, qp->pagelist, flags)) {
>  			/*
> -			 * Do not abort immediately since there may be
> -			 * temporary off LRU pages in the range.  Still
> -			 * need migrate other LRU pages.
> +			 * A large folio can only be isolated from LRU once,
> +			 * but may be mapped by many PTEs (and Copy-On-Write may
> +			 * intersperse PTEs of other folios).  This is a common
> +			 * case, so don't mistake it for failure (but of course
> +			 * there can be other cases of multi-mapped pages which
> +			 * this quick check does not help to filter out - and a
> +			 * search of the pagelist might grow to be prohibitive).
>  			 */
> -			if (migrate_folio_add(folio, qp->pagelist, flags))
> -				has_unmovable = true;
> -		} else
> -			break;
> +			if (folio_test_large(folio))
> +				qp->large = folio;
> +		} else if (folio != qp->large) {
> +			qp->nr_failed++;
> +			if (strictly_unmovable(flags))
> +				break;
> +		}
>  	}
>  	pte_unmap_unlock(mapped_pte, ptl);
>  	cond_resched();
> -
> -	if (has_unmovable)
> -		return 1;
> -
> -	return addr != end ? -EIO : 0;
> +out:
> +	if (qp->nr_failed && strictly_unmovable(flags))
> +		return -EIO;
> +	return 0;
>  }
>  
>  static int queue_folios_hugetlb(pte_t *pte, unsigned long hmask,
>  			       unsigned long addr, unsigned long end,
>  			       struct mm_walk *walk)
>  {
> -	int ret = 0;
>  #ifdef CONFIG_HUGETLB_PAGE
>  	struct queue_pages *qp = walk->private;
> -	unsigned long flags = (qp->flags & MPOL_MF_VALID);
> +	unsigned long flags = qp->flags;
>  	struct folio *folio;
>  	spinlock_t *ptl;
>  	pte_t entry;
>  
>  	ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte);
>  	entry = huge_ptep_get(pte);
> -	if (!pte_present(entry))
> +	if (!pte_present(entry)) {
> +		if (unlikely(is_hugetlb_entry_migration(entry)))
> +			qp->nr_failed++;
>  		goto unlock;
> +	}
>  	folio = pfn_folio(pte_pfn(entry));
>  	if (!queue_folio_required(folio, qp))
>  		goto unlock;
> -
> -	if (flags == MPOL_MF_STRICT) {
> -		/*
> -		 * STRICT alone means only detecting misplaced folio and no
> -		 * need to further check other vma.
> -		 */
> -		ret = -EIO;
> +	if (!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) ||
> +	    !vma_migratable(walk->vma)) {
> +		qp->nr_failed++;
>  		goto unlock;
>  	}
> -
> -	if (!vma_migratable(walk->vma)) {
> -		/*
> -		 * Must be STRICT with MOVE*, otherwise .test_walk() have
> -		 * stopped walking current vma.
> -		 * Detecting misplaced folio but allow migrating folios which
> -		 * have been queued.
> -		 */
> -		ret = 1;
> -		goto unlock;
> -	}
> -
>  	/*
> -	 * With MPOL_MF_MOVE, we try to migrate only unshared folios. If it
> -	 * is shared it is likely not worth migrating.
> +	 * Unless MPOL_MF_MOVE_ALL, we try to avoid migrating a shared folio.
> +	 * Choosing not to migrate a shared folio is not counted as a failure.
>  	 *
>  	 * To check if the folio is shared, ideally we want to make sure
>  	 * every page is mapped to the same process. Doing that is very
> -	 * expensive, so check the estimated mapcount of the folio instead.
> +	 * expensive, so check the estimated sharers of the folio instead.
>  	 */
> -	if (flags & (MPOL_MF_MOVE_ALL) ||
> -	    (flags & MPOL_MF_MOVE && folio_estimated_sharers(folio) == 1 &&
> -	     !hugetlb_pmd_shared(pte))) {
> -		if (!isolate_hugetlb(folio, qp->pagelist) &&
> -			(flags & MPOL_MF_STRICT))
> -			/*
> -			 * Failed to isolate folio but allow migrating pages
> -			 * which have been queued.
> -			 */
> -			ret = 1;
> -	}
> +	if ((flags & MPOL_MF_MOVE_ALL) ||
> +	    (folio_estimated_sharers(folio) == 1 && !hugetlb_pmd_shared(pte)))
> +		if (!isolate_hugetlb(folio, qp->pagelist))
> +			qp->nr_failed++;
>  unlock:
>  	spin_unlock(ptl);
> -#else
> -	BUG();
> +	if (qp->nr_failed && strictly_unmovable(flags))
> +		return -EIO;
>  #endif
> -	return ret;
> +	return 0;
>  }
>  
>  #ifdef CONFIG_NUMA_BALANCING
> @@ -708,8 +688,11 @@ static int queue_pages_test_walk(unsigned long start, unsigned long end,
>  		return 1;
>  	}
>  
> -	/* queue pages from current vma */
> -	if (flags & MPOL_MF_VALID)
> +	/*
> +	 * Check page nodes, and queue pages to move, in the current vma.
> +	 * But if no moving, and no strict checking, the scan can be skipped.
> +	 */
> +	if (flags & (MPOL_MF_STRICT | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
>  		return 0;
>  	return 1;
>  }
> @@ -731,22 +714,22 @@ static const struct mm_walk_ops queue_pages_lock_vma_walk_ops = {
>  /*
>   * Walk through page tables and collect pages to be migrated.
>   *
> - * If pages found in a given range are on a set of nodes (determined by
> - * @nodes and @flags,) it's isolated and queued to the pagelist which is
> - * passed via @private.
> + * If pages found in a given range are not on the required set of @nodes,
> + * and migration is allowed, they are isolated and queued to the pagelist
> + * which is passed via @private.

s/@private/@pagelist/

>   *
> - * queue_pages_range() has three possible return values:
> - * 1 - there is unmovable page, but MPOL_MF_MOVE* & MPOL_MF_STRICT were
> - *     specified.
> - * 0 - queue pages successfully or no misplaced page.
> - * errno - i.e. misplaced pages with MPOL_MF_STRICT specified (-EIO) or
> - *         memory range specified by nodemask and maxnode points outside
> - *         your accessible address space (-EFAULT)
> + * queue_pages_range() may return:
> + * 0 - all pages already on the right node, or successfully queued for moving
> + *     (or neither strict checking nor moving requested: only range checking).
> + * >0 - this number of misplaced folios could not be queued for moving
> + *      (a hugetlbfs page or a transparent huge page being counted as 1).
> + * -EIO - a misplaced page found, when MPOL_MF_STRICT specified without MOVEs.
> + * -EFAULT - a hole in the memory range, when MPOL_MF_DISCONTIG_OK unspecified.
>   */
> -static int
> +static long
>  queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end,
>  		nodemask_t *nodes, unsigned long flags,
> -		struct list_head *pagelist, bool lock_vma)
> +		struct list_head *pagelist)
>  {
>  	int err;
>  	struct queue_pages qp = {
> @@ -757,7 +740,7 @@ queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end,
>  		.end = end,
>  		.first = NULL,
>  	};
> -	const struct mm_walk_ops *ops = lock_vma ?
> +	const struct mm_walk_ops *ops = (flags & MPOL_MF_WRLOCK) ?
>  			&queue_pages_lock_vma_walk_ops : &queue_pages_walk_ops;
>  
>  	err = walk_page_range(mm, start, end, ops, &qp);
> @@ -766,7 +749,7 @@ queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end,
>  		/* whole range in hole */
>  		err = -EFAULT;
>  
> -	return err;
> +	return err ? : qp.nr_failed;
>  }
>  
>  /*
> @@ -1029,16 +1012,16 @@ static long do_get_mempolicy(int *policy, nodemask_t *nmask,
>  }
>  
>  #ifdef CONFIG_MIGRATION
> -static int migrate_folio_add(struct folio *folio, struct list_head *foliolist,
> +static bool migrate_folio_add(struct folio *folio, struct list_head *foliolist,
>  				unsigned long flags)
>  {
>  	/*
> -	 * We try to migrate only unshared folios. If it is shared it
> -	 * is likely not worth migrating.
> +	 * Unless MPOL_MF_MOVE_ALL, we try to avoid migrating a shared folio.
> +	 * Choosing not to migrate a shared folio is not counted as a failure.
>  	 *
>  	 * To check if the folio is shared, ideally we want to make sure
>  	 * every page is mapped to the same process. Doing that is very
> -	 * expensive, so check the estimated mapcount of the folio instead.
> +	 * expensive, so check the estimated sharers of the folio instead.
>  	 */
>  	if ((flags & MPOL_MF_MOVE_ALL) || folio_estimated_sharers(folio) == 1) {
>  		if (folio_isolate_lru(folio)) {
> @@ -1046,32 +1029,31 @@ static int migrate_folio_add(struct folio *folio, struct list_head *foliolist,
>  			node_stat_mod_folio(folio,
>  				NR_ISOLATED_ANON + folio_is_file_lru(folio),
>  				folio_nr_pages(folio));
> -		} else if (flags & MPOL_MF_STRICT) {
> +		} else {
>  			/*
>  			 * Non-movable folio may reach here.  And, there may be
>  			 * temporary off LRU folios or non-LRU movable folios.
>  			 * Treat them as unmovable folios since they can't be
> -			 * isolated, so they can't be moved at the moment.  It
> -			 * should return -EIO for this case too.
> +			 * isolated, so they can't be moved at the moment.
>  			 */
> -			return -EIO;
> +			return false;
>  		}
>  	}
> -
> -	return 0;
> +	return true;
>  }
>  
>  /*
>   * Migrate pages from one node to a target node.
>   * Returns error or the number of pages not migrated.
>   */
> -static int migrate_to_node(struct mm_struct *mm, int source, int dest,
> -			   int flags)
> +static long migrate_to_node(struct mm_struct *mm, int source, int dest,
> +			    int flags)
>  {
>  	nodemask_t nmask;
>  	struct vm_area_struct *vma;
>  	LIST_HEAD(pagelist);
> -	int err = 0;
> +	long nr_failed;
> +	long err = 0;
>  	struct migration_target_control mtc = {
>  		.nid = dest,
>  		.gfp_mask = GFP_HIGHUSER_MOVABLE | __GFP_THISNODE,
> @@ -1080,23 +1062,27 @@ static int migrate_to_node(struct mm_struct *mm, int source, int dest,
>  	nodes_clear(nmask);
>  	node_set(source, nmask);
>  
> -	/*
> -	 * This does not "check" the range but isolates all pages that
> -	 * need migration.  Between passing in the full user address
> -	 * space range and MPOL_MF_DISCONTIG_OK, this call can not fail.
> -	 */
> -	vma = find_vma(mm, 0);
>  	VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)));
> -	queue_pages_range(mm, vma->vm_start, mm->task_size, &nmask,
> -			flags | MPOL_MF_DISCONTIG_OK, &pagelist, false);
> +	vma = find_vma(mm, 0);
> +
> +	/*
> +	 * This does not migrate the range, but isolates all pages that
> +	 * need migration.  Between passing in the full user address
> +	 * space range and MPOL_MF_DISCONTIG_OK, this call cannot fail,
> +	 * but passes back the count of pages which could not be isolated.
> +	 */
> +	nr_failed = queue_pages_range(mm, vma->vm_start, mm->task_size, &nmask,
> +				      flags | MPOL_MF_DISCONTIG_OK, &pagelist);
>  
>  	if (!list_empty(&pagelist)) {
>  		err = migrate_pages(&pagelist, alloc_migration_target, NULL,
> -				(unsigned long)&mtc, MIGRATE_SYNC, MR_SYSCALL, NULL);
> +			(unsigned long)&mtc, MIGRATE_SYNC, MR_SYSCALL, NULL);
>  		if (err)
>  			putback_movable_pages(&pagelist);
>  	}
>  
> +	if (err >= 0)
> +		err += nr_failed;
>  	return err;
>  }
>  
> @@ -1109,8 +1095,8 @@ static int migrate_to_node(struct mm_struct *mm, int source, int dest,
>  int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
>  		     const nodemask_t *to, int flags)
>  {
> -	int busy = 0;
> -	int err = 0;
> +	long nr_failed = 0;
> +	long err = 0;
>  	nodemask_t tmp;
>  
>  	lru_cache_disable();
> @@ -1192,7 +1178,7 @@ int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
>  		node_clear(source, tmp);
>  		err = migrate_to_node(mm, source, dest, flags);
>  		if (err > 0)
> -			busy += err;
> +			nr_failed += err;
>  		if (err < 0)
>  			break;
>  	}
> @@ -1201,8 +1187,7 @@ int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
>  	lru_cache_enable();
>  	if (err < 0)
>  		return err;
> -	return busy;
> -
> +	return (nr_failed < INT_MAX) ? nr_failed : INT_MAX;

        return min_t(long, nr_failed, INT_MAX);
?

>  }
>  
>  /*
> @@ -1241,10 +1226,10 @@ static struct folio *new_folio(struct folio *src, unsigned long start)
>  }
>  #else
>  
> -static int migrate_folio_add(struct folio *folio, struct list_head *foliolist,
> +static bool migrate_folio_add(struct folio *folio, struct list_head *foliolist,
>  				unsigned long flags)
>  {
> -	return -EIO;
> +	return false;
>  }
>  
>  int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
> @@ -1268,8 +1253,8 @@ static long do_mbind(unsigned long start, unsigned long len,
>  	struct vma_iterator vmi;
>  	struct mempolicy *new;
>  	unsigned long end;
> -	int err;
> -	int ret;
> +	long err;
> +	long nr_failed;
>  	LIST_HEAD(pagelist);
>  
>  	if (flags & ~(unsigned long)MPOL_MF_VALID)
> @@ -1309,10 +1294,8 @@ static long do_mbind(unsigned long start, unsigned long len,
>  		 start, start + len, mode, mode_flags,
>  		 nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE);
>  
> -	if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
> -
> +	if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
>  		lru_cache_disable();
> -	}
>  	{
>  		NODEMASK_SCRATCH(scratch);
>  		if (scratch) {
> @@ -1328,44 +1311,37 @@ static long do_mbind(unsigned long start, unsigned long len,
>  		goto mpol_out;
>  
>  	/*
> -	 * Lock the VMAs before scanning for pages to migrate, to ensure we don't
> -	 * miss a concurrently inserted page.
> +	 * Lock the VMAs before scanning for pages to migrate,
> +	 * to ensure we don't miss a concurrently inserted page.
>  	 */
> -	ret = queue_pages_range(mm, start, end, nmask,
> -			  flags | MPOL_MF_INVERT, &pagelist, true);
> +	nr_failed = queue_pages_range(mm, start, end, nmask,
> +			flags | MPOL_MF_INVERT | MPOL_MF_WRLOCK, &pagelist);
>  
> -	if (ret < 0) {
> -		err = ret;
> -		goto up_out;
> -	}
> -
> -	vma_iter_init(&vmi, mm, start);
> -	prev = vma_prev(&vmi);
> -	for_each_vma_range(vmi, vma, end) {
> -		err = mbind_range(&vmi, vma, &prev, start, end, new);
> -		if (err)
> -			break;
> +	if (nr_failed < 0) {
> +		err = nr_failed;
> +	} else {
> +		vma_iter_init(&vmi, mm, start);
> +		prev = vma_prev(&vmi);
> +		for_each_vma_range(vmi, vma, end) {
> +			err = mbind_range(&vmi, vma, &prev, start, end, new);
> +			if (err)
> +				break;
> +		}
>  	}
>  
>  	if (!err) {
> -		int nr_failed = 0;
> -
>  		if (!list_empty(&pagelist)) {
>  			WARN_ON_ONCE(flags & MPOL_MF_LAZY);
> -			nr_failed = migrate_pages(&pagelist, new_folio, NULL,
> +			nr_failed |= migrate_pages(&pagelist, new_folio, NULL,
>  				start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND, NULL);
> -			if (nr_failed)
> -				putback_movable_pages(&pagelist);
>  		}
> -
> -		if ((ret > 0) || (nr_failed && (flags & MPOL_MF_STRICT)))
> +		if (nr_failed && (flags & MPOL_MF_STRICT))
>  			err = -EIO;
> -	} else {
> -up_out:
> -		if (!list_empty(&pagelist))
> -			putback_movable_pages(&pagelist);
>  	}
>  
> +	if (!list_empty(&pagelist))
> +		putback_movable_pages(&pagelist);
> +
>  	mmap_write_unlock(mm);
>  mpol_out:
>  	mpol_put(new);

--
Best Regards,
Huang, Ying

Re: [PATCH 03/12] mempolicy: fix migrate_pages(2) syscall return nr_failed

[Hugh Dickins]

On Wed, 27 Sep 2023, Huang, Ying wrote:
> Hugh Dickins <hughd@google.com> writes:
> > -static int queue_folios_pmd(pmd_t *pmd, spinlock_t *ptl, unsigned long addr,
> > +static void queue_folios_pmd(pmd_t *pmd, spinlock_t *ptl, unsigned long addr,
> 
> I don't find that "ptl" is used in the function now.  So, remove it?

Yes indeed, thanks for spotting that, I shall remove it.
And I think addr, end are also unused: I'll remove them too.

...
> > +		if (!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) ||
> > +		    !vma_migratable(vma)) {
> > +			qp->nr_failed++;
> > +			if (strictly_unmovable(flags))
> >  				break;
> > -			}
> > -
> > +		}
> 
> IIUC, even if !(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) or
> !vma_migratable(vma), the folio will be isolated in migrate_folio_add()
> below.  Is this the expected behavior?

Yikes, I think you're right, thanks a lot: it was okay up until I realized
the "qp->large" issue, and had to separate out the migrate_folio_add()
failure case.  Late changes...

> 
> > +		if (migrate_folio_add(folio, qp->pagelist, flags)) {
...
> > @@ -731,22 +714,22 @@ static const struct mm_walk_ops queue_pages_lock_vma_walk_ops = {
> >  /*
> >   * Walk through page tables and collect pages to be migrated.
> >   *
> > - * If pages found in a given range are on a set of nodes (determined by
> > - * @nodes and @flags,) it's isolated and queued to the pagelist which is
> > - * passed via @private.
> > + * If pages found in a given range are not on the required set of @nodes,
> > + * and migration is allowed, they are isolated and queued to the pagelist
> > + * which is passed via @private.
> 
> s/@private/@pagelist/

Right: or even better, simply "queued to @pagelist."

...
> > @@ -1201,8 +1187,7 @@ int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
> >  	lru_cache_enable();
> >  	if (err < 0)
> >  		return err;
> > -	return busy;
> > -
> > +	return (nr_failed < INT_MAX) ? nr_failed : INT_MAX;
> 
>         return min_t(long, nr_failed, INT_MAX);
> ?

I may be the odd man out, but I read it more easily how it is:
so will stick with that, unless others object.

Very helpful comments: thank you,

Hugh

[PATCH 04/12] mempolicy trivia: delete those ancient pr_debug()s

[Hugh Dickins]

Delete those ancient pr_debug()s - PDprintk()s in Andi Kleen's original
submission of core NUMA API, and useful when debugging shared mempolicy
lifetime back then, but not used recently.

Signed-off-by: Hugh Dickins <hughd@google.com>
---
 mm/mempolicy.c | 21 ---------------------
 1 file changed, 21 deletions(-)

diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 937386409c28..b2573921b78f 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -264,9 +264,6 @@ static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags,
 {
 	struct mempolicy *policy;
 
-	pr_debug("setting mode %d flags %d nodes[0] %lx\n",
-		 mode, flags, nodes ? nodes_addr(*nodes)[0] : NUMA_NO_NODE);
-
 	if (mode == MPOL_DEFAULT) {
 		if (nodes && !nodes_empty(*nodes))
 			return ERR_PTR(-EINVAL);
@@ -765,11 +762,6 @@ static int vma_replace_policy(struct vm_area_struct *vma,
 
 	vma_assert_write_locked(vma);
 
-	pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n",
-		 vma->vm_start, vma->vm_end, vma->vm_pgoff,
-		 vma->vm_ops, vma->vm_file,
-		 vma->vm_ops ? vma->vm_ops->set_policy : NULL);
-
 	new = mpol_dup(pol);
 	if (IS_ERR(new))
 		return PTR_ERR(new);
@@ -1290,10 +1282,6 @@ static long do_mbind(unsigned long start, unsigned long len,
 	if (!new)
 		flags |= MPOL_MF_DISCONTIG_OK;
 
-	pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n",
-		 start, start + len, mode, mode_flags,
-		 nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE);
-
 	if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
 		lru_cache_disable();
 	{
@@ -2511,8 +2499,6 @@ static void sp_insert(struct shared_policy *sp, struct sp_node *new)
 	}
 	rb_link_node(&new->nd, parent, p);
 	rb_insert_color(&new->nd, &sp->root);
-	pr_debug("inserting %lx-%lx: %d\n", new->start, new->end,
-		 new->policy ? new->policy->mode : 0);
 }
 
 /* Find shared policy intersecting idx */
@@ -2649,7 +2635,6 @@ void mpol_put_task_policy(struct task_struct *task)
 
 static void sp_delete(struct shared_policy *sp, struct sp_node *n)
 {
-	pr_debug("deleting %lx-l%lx\n", n->start, n->end);
 	rb_erase(&n->nd, &sp->root);
 	sp_free(n);
 }
@@ -2806,12 +2791,6 @@ int mpol_set_shared_policy(struct shared_policy *info,
 	struct sp_node *new = NULL;
 	unsigned long sz = vma_pages(vma);
 
-	pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n",
-		 vma->vm_pgoff,
-		 sz, npol ? npol->mode : -1,
-		 npol ? npol->flags : -1,
-		 npol ? nodes_addr(npol->nodes)[0] : NUMA_NO_NODE);
-
 	if (npol) {
 		new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol);
 		if (!new)
-- 
2.35.3

Re: [PATCH 04/12] mempolicy trivia: delete those ancient pr_debug()s

[Matthew Wilcox]

On Mon, Sep 25, 2023 at 01:25:09AM -0700, Hugh Dickins wrote:
> Delete those ancient pr_debug()s - PDprintk()s in Andi Kleen's original
> submission of core NUMA API, and useful when debugging shared mempolicy
> lifetime back then, but not used recently.

Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>

[PATCH 05/12] mempolicy trivia: slightly more consistent naming

[Hugh Dickins]

Before getting down to work, do a little cleanup, mainly of inconsistent
variable naming.  I gave up trying to rationalize mpol versus pol versus
policy, and node versus nid, but let's avoid p and nd.  Remove a few
superfluous blank lines, but add one; and here prefer vma->vm_policy to
vma_policy(vma) - the latter being appropriate in other sources, which
have to allow for !CONFIG_NUMA.  That intriguing line about KERNEL_DS?
should have gone in v2.6.15, when numa_policy_init() stopped using
set_mempolicy(2)'s system call handler.

Signed-off-by: Hugh Dickins <hughd@google.com>
---
 include/linux/mempolicy.h | 11 +++---
 mm/mempolicy.c            | 73 ++++++++++++++++++---------------------
 2 files changed, 38 insertions(+), 46 deletions(-)

diff --git a/include/linux/mempolicy.h b/include/linux/mempolicy.h
index d232de7cdc56..8013d716dc46 100644
--- a/include/linux/mempolicy.h
+++ b/include/linux/mempolicy.h
@@ -126,10 +126,9 @@ struct shared_policy {
 
 int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst);
 void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol);
-int mpol_set_shared_policy(struct shared_policy *info,
-				struct vm_area_struct *vma,
-				struct mempolicy *new);
-void mpol_free_shared_policy(struct shared_policy *p);
+int mpol_set_shared_policy(struct shared_policy *sp,
+			   struct vm_area_struct *vma, struct mempolicy *mpol);
+void mpol_free_shared_policy(struct shared_policy *sp);
 struct mempolicy *mpol_shared_policy_lookup(struct shared_policy *sp,
 					    unsigned long idx);
 
@@ -193,7 +192,7 @@ static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b)
 	return true;
 }
 
-static inline void mpol_put(struct mempolicy *p)
+static inline void mpol_put(struct mempolicy *pol)
 {
 }
 
@@ -212,7 +211,7 @@ static inline void mpol_shared_policy_init(struct shared_policy *sp,
 {
 }
 
-static inline void mpol_free_shared_policy(struct shared_policy *p)
+static inline void mpol_free_shared_policy(struct shared_policy *sp)
 {
 }
 
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index b2573921b78f..121bb490481b 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -25,7 +25,7 @@
  *                to the last. It would be better if bind would truly restrict
  *                the allocation to memory nodes instead
  *
- * preferred       Try a specific node first before normal fallback.
+ * preferred      Try a specific node first before normal fallback.
  *                As a special case NUMA_NO_NODE here means do the allocation
  *                on the local CPU. This is normally identical to default,
  *                but useful to set in a VMA when you have a non default
@@ -52,7 +52,7 @@
  * on systems with highmem kernel lowmem allocation don't get policied.
  * Same with GFP_DMA allocations.
  *
- * For shmfs/tmpfs/hugetlbfs shared memory the policy is shared between
+ * For shmem/tmpfs shared memory the policy is shared between
  * all users and remembered even when nobody has memory mapped.
  */
 
@@ -291,6 +291,7 @@ static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags,
 			return ERR_PTR(-EINVAL);
 	} else if (nodes_empty(*nodes))
 		return ERR_PTR(-EINVAL);
+
 	policy = kmem_cache_alloc(policy_cache, GFP_KERNEL);
 	if (!policy)
 		return ERR_PTR(-ENOMEM);
@@ -303,11 +304,11 @@ static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags,
 }
 
 /* Slow path of a mpol destructor. */
-void __mpol_put(struct mempolicy *p)
+void __mpol_put(struct mempolicy *pol)
 {
-	if (!atomic_dec_and_test(&p->refcnt))
+	if (!atomic_dec_and_test(&pol->refcnt))
 		return;
-	kmem_cache_free(policy_cache, p);
+	kmem_cache_free(policy_cache, pol);
 }
 
 static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes)
@@ -364,7 +365,6 @@ static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask)
  *
  * Called with task's alloc_lock held.
  */
-
 void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new)
 {
 	mpol_rebind_policy(tsk->mempolicy, new);
@@ -375,7 +375,6 @@ void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new)
  *
  * Call holding a reference to mm.  Takes mm->mmap_lock during call.
  */
-
 void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
 {
 	struct vm_area_struct *vma;
@@ -754,7 +753,7 @@ queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end,
  * This must be called with the mmap_lock held for writing.
  */
 static int vma_replace_policy(struct vm_area_struct *vma,
-						struct mempolicy *pol)
+				struct mempolicy *pol)
 {
 	int err;
 	struct mempolicy *old;
@@ -800,7 +799,7 @@ static int mbind_range(struct vma_iterator *vmi, struct vm_area_struct *vma,
 		vmstart = vma->vm_start;
 	}
 
-	if (mpol_equal(vma_policy(vma), new_pol)) {
+	if (mpol_equal(vma->vm_policy, new_pol)) {
 		*prev = vma;
 		return 0;
 	}
@@ -872,18 +871,18 @@ static long do_set_mempolicy(unsigned short mode, unsigned short flags,
  *
  * Called with task's alloc_lock held
  */
-static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes)
+static void get_policy_nodemask(struct mempolicy *pol, nodemask_t *nodes)
 {
 	nodes_clear(*nodes);
-	if (p == &default_policy)
+	if (pol == &default_policy)
 		return;
 
-	switch (p->mode) {
+	switch (pol->mode) {
 	case MPOL_BIND:
 	case MPOL_INTERLEAVE:
 	case MPOL_PREFERRED:
 	case MPOL_PREFERRED_MANY:
-		*nodes = p->nodes;
+		*nodes = pol->nodes;
 		break;
 	case MPOL_LOCAL:
 		/* return empty node mask for local allocation */
@@ -1649,7 +1648,6 @@ static int kernel_migrate_pages(pid_t pid, unsigned long maxnode,
 out_put:
 	put_task_struct(task);
 	goto out;
-
 }
 
 SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
@@ -1659,7 +1657,6 @@ SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
 	return kernel_migrate_pages(pid, maxnode, old_nodes, new_nodes);
 }
 
-
 /* Retrieve NUMA policy */
 static int kernel_get_mempolicy(int __user *policy,
 				unsigned long __user *nmask,
@@ -1842,10 +1839,10 @@ nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
  * policy_node() is always coupled with policy_nodemask(), which
  * secures the nodemask limit for 'bind' and 'prefer-many' policy.
  */
-static int policy_node(gfp_t gfp, struct mempolicy *policy, int nd)
+static int policy_node(gfp_t gfp, struct mempolicy *policy, int nid)
 {
 	if (policy->mode == MPOL_PREFERRED) {
-		nd = first_node(policy->nodes);
+		nid = first_node(policy->nodes);
 	} else {
 		/*
 		 * __GFP_THISNODE shouldn't even be used with the bind policy
@@ -1860,19 +1857,18 @@ static int policy_node(gfp_t gfp, struct mempolicy *policy, int nd)
 	    policy->home_node != NUMA_NO_NODE)
 		return policy->home_node;
 
-	return nd;
+	return nid;
 }
 
 /* Do dynamic interleaving for a process */
-static unsigned interleave_nodes(struct mempolicy *policy)
+static unsigned int interleave_nodes(struct mempolicy *policy)
 {
-	unsigned next;
-	struct task_struct *me = current;
+	unsigned int nid;
 
-	next = next_node_in(me->il_prev, policy->nodes);
-	if (next < MAX_NUMNODES)
-		me->il_prev = next;
-	return next;
+	nid = next_node_in(current->il_prev, policy->nodes);
+	if (nid < MAX_NUMNODES)
+		current->il_prev = nid;
+	return nid;
 }
 
 /*
@@ -2362,7 +2358,7 @@ unsigned long alloc_pages_bulk_array_mempolicy(gfp_t gfp,
 
 int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst)
 {
-	struct mempolicy *pol = mpol_dup(vma_policy(src));
+	struct mempolicy *pol = mpol_dup(src->vm_policy);
 
 	if (IS_ERR(pol))
 		return PTR_ERR(pol);
@@ -2784,40 +2780,40 @@ void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
 	}
 }
 
-int mpol_set_shared_policy(struct shared_policy *info,
-			struct vm_area_struct *vma, struct mempolicy *npol)
+int mpol_set_shared_policy(struct shared_policy *sp,
+			struct vm_area_struct *vma, struct mempolicy *pol)
 {
 	int err;
 	struct sp_node *new = NULL;
 	unsigned long sz = vma_pages(vma);
 
-	if (npol) {
-		new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol);
+	if (pol) {
+		new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, pol);
 		if (!new)
 			return -ENOMEM;
 	}
-	err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new);
+	err = shared_policy_replace(sp, vma->vm_pgoff, vma->vm_pgoff + sz, new);
 	if (err && new)
 		sp_free(new);
 	return err;
 }
 
 /* Free a backing policy store on inode delete. */
-void mpol_free_shared_policy(struct shared_policy *p)
+void mpol_free_shared_policy(struct shared_policy *sp)
 {
 	struct sp_node *n;
 	struct rb_node *next;
 
-	if (!p->root.rb_node)
+	if (!sp->root.rb_node)
 		return;
-	write_lock(&p->lock);
-	next = rb_first(&p->root);
+	write_lock(&sp->lock);
+	next = rb_first(&sp->root);
 	while (next) {
 		n = rb_entry(next, struct sp_node, nd);
 		next = rb_next(&n->nd);
-		sp_delete(p, n);
+		sp_delete(sp, n);
 	}
-	write_unlock(&p->lock);
+	write_unlock(&sp->lock);
 }
 
 #ifdef CONFIG_NUMA_BALANCING
@@ -2867,7 +2863,6 @@ static inline void __init check_numabalancing_enable(void)
 }
 #endif /* CONFIG_NUMA_BALANCING */
 
-/* assumes fs == KERNEL_DS */
 void __init numa_policy_init(void)
 {
 	nodemask_t interleave_nodes;
@@ -2930,7 +2925,6 @@ void numa_default_policy(void)
 /*
  * Parse and format mempolicy from/to strings
  */
-
 static const char * const policy_modes[] =
 {
 	[MPOL_DEFAULT]    = "default",
@@ -2941,7 +2935,6 @@ static const char * const policy_modes[] =
 	[MPOL_PREFERRED_MANY]  = "prefer (many)",
 };
 
-
 #ifdef CONFIG_TMPFS
 /**
  * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option.
-- 
2.35.3

Re: [PATCH 05/12] mempolicy trivia: slightly more consistent naming

[Matthew Wilcox]

On Mon, Sep 25, 2023 at 01:26:55AM -0700, Hugh Dickins wrote:
> Before getting down to work, do a little cleanup, mainly of inconsistent
> variable naming.  I gave up trying to rationalize mpol versus pol versus
> policy, and node versus nid, but let's avoid p and nd.  Remove a few
> superfluous blank lines, but add one; and here prefer vma->vm_policy to
> vma_policy(vma) - the latter being appropriate in other sources, which
> have to allow for !CONFIG_NUMA.  That intriguing line about KERNEL_DS?
> should have gone in v2.6.15, when numa_policy_init() stopped using
> set_mempolicy(2)'s system call handler.

Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>

[PATCH 06/12] mempolicy trivia: use pgoff_t in shared mempolicy tree

[Hugh Dickins]

Prefer the more explicit "pgoff_t" to "unsigned long" when dealing with
a shared mempolicy tree.  Delete confusing comment about pseudo mm vmas.

Signed-off-by: Hugh Dickins <hughd@google.com>
---
 include/linux/mempolicy.h | 12 +++---------
 mm/mempolicy.c            |  8 ++++----
 2 files changed, 7 insertions(+), 13 deletions(-)

diff --git a/include/linux/mempolicy.h b/include/linux/mempolicy.h
index 8013d716dc46..12f7dc74a457 100644
--- a/include/linux/mempolicy.h
+++ b/include/linux/mempolicy.h
@@ -107,18 +107,12 @@ static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b)
 
 /*
  * Tree of shared policies for a shared memory region.
- * Maintain the policies in a pseudo mm that contains vmas. The vmas
- * carry the policy. As a special twist the pseudo mm is indexed in pages, not
- * bytes, so that we can work with shared memory segments bigger than
- * unsigned long.
  */
-
 struct sp_node {
 	struct rb_node nd;
-	unsigned long start, end;
+	pgoff_t start, end;
 	struct mempolicy *policy;
 };
-
 struct shared_policy {
 	struct rb_root root;
 	rwlock_t lock;
@@ -130,7 +124,7 @@ int mpol_set_shared_policy(struct shared_policy *sp,
 			   struct vm_area_struct *vma, struct mempolicy *mpol);
 void mpol_free_shared_policy(struct shared_policy *sp);
 struct mempolicy *mpol_shared_policy_lookup(struct shared_policy *sp,
-					    unsigned long idx);
+					    pgoff_t idx);
 
 struct mempolicy *get_task_policy(struct task_struct *p);
 struct mempolicy *__get_vma_policy(struct vm_area_struct *vma,
@@ -216,7 +210,7 @@ static inline void mpol_free_shared_policy(struct shared_policy *sp)
 }
 
 static inline struct mempolicy *
-mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
+mpol_shared_policy_lookup(struct shared_policy *sp, pgoff_t idx)
 {
 	return NULL;
 }
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 121bb490481b..065e886ec9b6 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -2444,7 +2444,7 @@ bool __mpol_equal(struct mempolicy *a, struct mempolicy *b)
  * reading or for writing
  */
 static struct sp_node *
-sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
+sp_lookup(struct shared_policy *sp, pgoff_t start, pgoff_t end)
 {
 	struct rb_node *n = sp->root.rb_node;
 
@@ -2499,7 +2499,7 @@ static void sp_insert(struct shared_policy *sp, struct sp_node *new)
 
 /* Find shared policy intersecting idx */
 struct mempolicy *
-mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
+mpol_shared_policy_lookup(struct shared_policy *sp, pgoff_t idx)
 {
 	struct mempolicy *pol = NULL;
 	struct sp_node *sn;
@@ -2665,8 +2665,8 @@ static struct sp_node *sp_alloc(unsigned long start, unsigned long end,
 }
 
 /* Replace a policy range. */
-static int shared_policy_replace(struct shared_policy *sp, unsigned long start,
-				 unsigned long end, struct sp_node *new)
+static int shared_policy_replace(struct shared_policy *sp, pgoff_t start,
+				 pgoff_t end, struct sp_node *new)
 {
 	struct sp_node *n;
 	struct sp_node *n_new = NULL;
-- 
2.35.3

Re: [PATCH 06/12] mempolicy trivia: use pgoff_t in shared mempolicy tree

[Matthew Wilcox]

On Mon, Sep 25, 2023 at 01:28:14AM -0700, Hugh Dickins wrote:
> Prefer the more explicit "pgoff_t" to "unsigned long" when dealing with
> a shared mempolicy tree.  Delete confusing comment about pseudo mm vmas.

Yes, with three quibbles

>  struct sp_node {
>  	struct rb_node nd;
> -	unsigned long start, end;
> +	pgoff_t start, end;
>  	struct mempolicy *policy;
>  };
> -
>  struct shared_policy {

Did you intend to delete the blank line between these two structs?
That's not our normal style.

> +++ b/mm/mempolicy.c
> @@ -2444,7 +2444,7 @@ bool __mpol_equal(struct mempolicy *a, struct mempolicy *b)
>   * reading or for writing
>   */
>  static struct sp_node *
> -sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
> +sp_lookup(struct shared_policy *sp, pgoff_t start, pgoff_t end)

While you're reformatting anyway, mind joining these two lines?

> @@ -2499,7 +2499,7 @@ static void sp_insert(struct shared_policy *sp, struct sp_node *new)
>  
>  /* Find shared policy intersecting idx */
>  struct mempolicy *
> -mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
> +mpol_shared_policy_lookup(struct shared_policy *sp, pgoff_t idx)

Ditto

Re: [PATCH 06/12] mempolicy trivia: use pgoff_t in shared mempolicy tree

[Matthew Wilcox]

On Mon, Sep 25, 2023 at 11:31:40PM +0100, Matthew Wilcox wrote:
> On Mon, Sep 25, 2023 at 01:28:14AM -0700, Hugh Dickins wrote:
> > Prefer the more explicit "pgoff_t" to "unsigned long" when dealing with
> > a shared mempolicy tree.  Delete confusing comment about pseudo mm vmas.
> 
> Yes, with three quibbles

Actually, a fourth has occurred to me

> >  struct sp_node {
> >  	struct rb_node nd;
> > -	unsigned long start, end;
> > +	pgoff_t start, end;
> >  	struct mempolicy *policy;
> >  };

This data structure is unused outside mempolicy.c today, and you don't
add any.  Perhaps we could move it from mempolicy.h to mempolicy.c?

Re: [PATCH 06/12] mempolicy trivia: use pgoff_t in shared mempolicy tree

[Hugh Dickins]

On Mon, 25 Sep 2023, Matthew Wilcox wrote:
> On Mon, Sep 25, 2023 at 01:28:14AM -0700, Hugh Dickins wrote:
> > Prefer the more explicit "pgoff_t" to "unsigned long" when dealing with
> > a shared mempolicy tree.  Delete confusing comment about pseudo mm vmas.
> 
> Yes, with three quibbles
> 
> >  struct sp_node {
> >  	struct rb_node nd;
> > -	unsigned long start, end;
> > +	pgoff_t start, end;
> >  	struct mempolicy *policy;
> >  };
> > -
> >  struct shared_policy {
> 
> Did you intend to delete the blank line between these two structs?
> That's not our normal style.

I think I did intend it actually, to join both of those structs to the
"Tree of shared policies" comment above them.  But now that I'm looking
again, I think what I'd most like to do (and going against all of your
suggestions e.g. move sp_node to mempolicy.c: good observation, but isn't
keeping them together more helpful to the reader?) is swap those structs
around - struct shared_policy first for the root of the tree, then
struct sp_node showing the nodes of the tree (and still without blank
line).  Wouldn't that be the most helpful way to present them?

I'll knuckle down and do exactly as you have suggested, if you say so:
but above is my own preference.

> 
> > +++ b/mm/mempolicy.c
> > @@ -2444,7 +2444,7 @@ bool __mpol_equal(struct mempolicy *a, struct mempolicy *b)
> >   * reading or for writing
> >   */
> >  static struct sp_node *
> > -sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
> > +sp_lookup(struct shared_policy *sp, pgoff_t start, pgoff_t end)
> 
> While you're reformatting anyway, mind joining these two lines?
> 
> > @@ -2499,7 +2499,7 @@ static void sp_insert(struct shared_policy *sp, struct sp_node *new)
> >  
> >  /* Find shared policy intersecting idx */
> >  struct mempolicy *
> > -mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
> > +mpol_shared_policy_lookup(struct shared_policy *sp, pgoff_t idx)
> 
> Ditto

Sure, will do (I recall Linus much prefers them as you suggest).
But we have different indentation habits: I think it's spaced exactitude
which irritates you the most, would the style below be okay?
Or maybe you'd prefer to go over-80 with these two.

struct mempolicy *mpol_shared_policy_lookup(struct shared_policy *sp,
						pgoff_t idx)

Hugh

[PATCH 07/12] mempolicy: mpol_shared_policy_init() without pseudo-vma

[Hugh Dickins]

mpol_shared_policy_init() does not need to use a pseudo-vma: it can use
sp_alloc() and sp_insert() directly, since the object's shared policy
tree is empty and inaccessible (needing no lock) at get_inode() time.

Signed-off-by: Hugh Dickins <hughd@google.com>
---
 mm/mempolicy.c | 11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 065e886ec9b6..a22b641cfd6b 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -2749,7 +2749,7 @@ void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
 	rwlock_init(&sp->lock);
 
 	if (mpol) {
-		struct vm_area_struct pvma;
+		struct sp_node *n;
 		struct mempolicy *new;
 		NODEMASK_SCRATCH(scratch);
 
@@ -2766,11 +2766,10 @@ void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
 		if (ret)
 			goto put_new;
 
-		/* Create pseudo-vma that contains just the policy */
-		vma_init(&pvma, NULL);
-		pvma.vm_end = TASK_SIZE;	/* policy covers entire file */
-		mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
-
+		/* alloc node covering entire file; adds ref to new */
+		n = sp_alloc(0, MAX_LFS_FILESIZE >> PAGE_SHIFT, new);
+		if (n)
+			sp_insert(sp, n);
 put_new:
 		mpol_put(new);			/* drop initial ref */
 free_scratch:
-- 
2.35.3

Re: [PATCH 07/12] mempolicy: mpol_shared_policy_init() without pseudo-vma

[Matthew Wilcox]

On Mon, Sep 25, 2023 at 01:29:28AM -0700, Hugh Dickins wrote:
> +		/* alloc node covering entire file; adds ref to new */

This comment is confusing.  sp_alloc initialises the refcount of 'n' to 1.
Which is the same memory referred to by the name 'new' in __mpol_dup(),
but in this function, the name "new" refers to the mempolicy called
"old" in __mpol_dup().

> +		n = sp_alloc(0, MAX_LFS_FILESIZE >> PAGE_SHIFT, new);
> +		if (n)
> +			sp_insert(sp, n);
>  put_new:
>  		mpol_put(new);			/* drop initial ref */
>  free_scratch:

This is all a bit inefficient, really.  We call mpol_new() to get a
new mpol, then we set it up, then we dup it, then we free it.  It'd
be nice if we could donate it instead of copying it.  Maybe you'll
do something like that later.

Re: [PATCH 07/12] mempolicy: mpol_shared_policy_init() without pseudo-vma

[Hugh Dickins]

On Mon, 25 Sep 2023, Matthew Wilcox wrote:
> On Mon, Sep 25, 2023 at 01:29:28AM -0700, Hugh Dickins wrote:
> > +		/* alloc node covering entire file; adds ref to new */
> 
> This comment is confusing.  sp_alloc initialises the refcount of 'n' to 1.
> Which is the same memory referred to by the name 'new' in __mpol_dup(),
> but in this function, the name "new" refers to the mempolicy called
> "old" in __mpol_dup().

No promises, but I'll see if I can make it look better in v2.

> 
> > +		n = sp_alloc(0, MAX_LFS_FILESIZE >> PAGE_SHIFT, new);
> > +		if (n)
> > +			sp_insert(sp, n);
> >  put_new:
> >  		mpol_put(new);			/* drop initial ref */
> >  free_scratch:
> 
> This is all a bit inefficient, really.  We call mpol_new() to get a
> new mpol, then we set it up, then we dup it, then we free it.  It'd
> be nice if we could donate it instead of copying it.  Maybe you'll
> do something like that later.

"later" is probably the operative word.  I do have an unincluded 2017
patch where I had that same realization, and wrote "I suspect that this
series of commits may be adding to an absurdity of over-mpol_dup()ing:
but that's for some other future cleanup, right now I'm just happy not
to be corrupting or leaking mpols."

Hugh

[PATCH 08/12] mempolicy: remove confusing MPOL_MF_LAZY dead code

[Hugh Dickins]

v3.8 commit b24f53a0bea3 ("mm: mempolicy: Add MPOL_MF_LAZY") introduced
MPOL_MF_LAZY, and included it in the MPOL_MF_VALID flags; but a720094ded8
("mm: mempolicy: Hide MPOL_NOOP and MPOL_MF_LAZY from userspace for now")
immediately removed it from MPOL_MF_VALID flags, pending further review.
"This will need to be revisited", but it has not been reinstated.

The present state is confusing: there is dead code in mm/mempolicy.c to
handle MPOL_MF_LAZY cases which can never occur.  Remove that: it can be
resurrected later if necessary.  But keep the definition of MPOL_MF_LAZY,
which must remain in the UAPI, even though it always fails with EINVAL.

https://lore.kernel.org/linux-mm/1553041659-46787-1-git-send-email-yang.shi@linux.alibaba.com/
links to a previous request to remove MPOL_MF_LAZY.

Signed-off-by: Hugh Dickins <hughd@google.com>
---
 include/uapi/linux/mempolicy.h |  2 +-
 mm/mempolicy.c                 | 18 ------------------
 2 files changed, 1 insertion(+), 19 deletions(-)

diff --git a/include/uapi/linux/mempolicy.h b/include/uapi/linux/mempolicy.h
index 046d0ccba4cd..a8963f7ef4c2 100644
--- a/include/uapi/linux/mempolicy.h
+++ b/include/uapi/linux/mempolicy.h
@@ -48,7 +48,7 @@ enum {
 #define MPOL_MF_MOVE	 (1<<1)	/* Move pages owned by this process to conform
 				   to policy */
 #define MPOL_MF_MOVE_ALL (1<<2)	/* Move every page to conform to policy */
-#define MPOL_MF_LAZY	 (1<<3)	/* Modifies '_MOVE:  lazy migrate on fault */
+#define MPOL_MF_LAZY	 (1<<3)	/* UNSUPPORTED FLAG: Lazy migrate on fault */
 #define MPOL_MF_INTERNAL (1<<4)	/* Internal flags start here */
 
 #define MPOL_MF_VALID	(MPOL_MF_STRICT   | 	\
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index a22b641cfd6b..7ab6102d7da4 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -632,12 +632,6 @@ unsigned long change_prot_numa(struct vm_area_struct *vma,
 
 	return nr_updated;
 }
-#else
-static unsigned long change_prot_numa(struct vm_area_struct *vma,
-			unsigned long addr, unsigned long end)
-{
-	return 0;
-}
 #endif /* CONFIG_NUMA_BALANCING */
 
 static int queue_pages_test_walk(unsigned long start, unsigned long end,
@@ -676,14 +670,6 @@ static int queue_pages_test_walk(unsigned long start, unsigned long end,
 	if (endvma > end)
 		endvma = end;
 
-	if (flags & MPOL_MF_LAZY) {
-		/* Similar to task_numa_work, skip inaccessible VMAs */
-		if (!is_vm_hugetlb_page(vma) && vma_is_accessible(vma) &&
-			!(vma->vm_flags & VM_MIXEDMAP))
-			change_prot_numa(vma, start, endvma);
-		return 1;
-	}
-
 	/*
 	 * Check page nodes, and queue pages to move, in the current vma.
 	 * But if no moving, and no strict checking, the scan can be skipped.
@@ -1271,9 +1257,6 @@ static long do_mbind(unsigned long start, unsigned long len,
 	if (IS_ERR(new))
 		return PTR_ERR(new);
 
-	if (flags & MPOL_MF_LAZY)
-		new->flags |= MPOL_F_MOF;
-
 	/*
 	 * If we are using the default policy then operation
 	 * on discontinuous address spaces is okay after all
@@ -1318,7 +1301,6 @@ static long do_mbind(unsigned long start, unsigned long len,
 
 	if (!err) {
 		if (!list_empty(&pagelist)) {
-			WARN_ON_ONCE(flags & MPOL_MF_LAZY);
 			nr_failed |= migrate_pages(&pagelist, new_folio, NULL,
 				start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND, NULL);
 		}
-- 
2.35.3

Re: [PATCH 08/12] mempolicy: remove confusing MPOL_MF_LAZY dead code

[Matthew Wilcox]

On Mon, Sep 25, 2023 at 01:30:51AM -0700, Hugh Dickins wrote:
> v3.8 commit b24f53a0bea3 ("mm: mempolicy: Add MPOL_MF_LAZY") introduced
> MPOL_MF_LAZY, and included it in the MPOL_MF_VALID flags; but a720094ded8
> ("mm: mempolicy: Hide MPOL_NOOP and MPOL_MF_LAZY from userspace for now")
> immediately removed it from MPOL_MF_VALID flags, pending further review.
> "This will need to be revisited", but it has not been reinstated.
> 
> The present state is confusing: there is dead code in mm/mempolicy.c to
> handle MPOL_MF_LAZY cases which can never occur.  Remove that: it can be
> resurrected later if necessary.  But keep the definition of MPOL_MF_LAZY,
> which must remain in the UAPI, even though it always fails with EINVAL.
> 
> https://lore.kernel.org/linux-mm/1553041659-46787-1-git-send-email-yang.shi@linux.alibaba.com/
> links to a previous request to remove MPOL_MF_LAZY.
> 
> Signed-off-by: Hugh Dickins <hughd@google.com>

Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>

[PATCH 09/12] mm: add page_rmappable_folio() wrapper

[Hugh Dickins]

folio_prep_large_rmappable() is being used repeatedly along with a
conversion from page to folio, a check non-NULL, a check order > 1:
wrap it all up into struct folio *page_rmappable_folio(struct page *).

Signed-off-by: Hugh Dickins <hughd@google.com>
---
 include/linux/huge_mm.h | 13 +++++++++++++
 mm/mempolicy.c          | 17 +++--------------
 mm/page_alloc.c         |  8 ++------
 3 files changed, 18 insertions(+), 20 deletions(-)

diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
index fa0350b0812a..58e7662a8a62 100644
--- a/include/linux/huge_mm.h
+++ b/include/linux/huge_mm.h
@@ -141,6 +141,15 @@ unsigned long thp_get_unmapped_area(struct file *filp, unsigned long addr,
 		unsigned long len, unsigned long pgoff, unsigned long flags);
 
 void folio_prep_large_rmappable(struct folio *folio);
+static inline struct folio *page_rmappable_folio(struct page *page)
+{
+	struct folio *folio = (struct folio *)page;
+
+	if (folio && folio_order(folio) > 1)
+		folio_prep_large_rmappable(folio);
+	return folio;
+}
+
 bool can_split_folio(struct folio *folio, int *pextra_pins);
 int split_huge_page_to_list(struct page *page, struct list_head *list);
 static inline int split_huge_page(struct page *page)
@@ -281,6 +290,10 @@ static inline bool hugepage_vma_check(struct vm_area_struct *vma,
 }
 
 static inline void folio_prep_large_rmappable(struct folio *folio) {}
+static inline struct folio *page_rmappable_folio(struct page *page)
+{
+	return (struct folio *)page;
+}
 
 #define transparent_hugepage_flags 0UL
 
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 7ab6102d7da4..4c3b3f535630 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -2137,10 +2137,7 @@ struct folio *vma_alloc_folio(gfp_t gfp, int order, struct vm_area_struct *vma,
 		mpol_cond_put(pol);
 		gfp |= __GFP_COMP;
 		page = alloc_page_interleave(gfp, order, nid);
-		folio = (struct folio *)page;
-		if (folio && order > 1)
-			folio_prep_large_rmappable(folio);
-		goto out;
+		return page_rmappable_folio(page);
 	}
 
 	if (pol->mode == MPOL_PREFERRED_MANY) {
@@ -2150,10 +2147,7 @@ struct folio *vma_alloc_folio(gfp_t gfp, int order, struct vm_area_struct *vma,
 		gfp |= __GFP_COMP;
 		page = alloc_pages_preferred_many(gfp, order, node, pol);
 		mpol_cond_put(pol);
-		folio = (struct folio *)page;
-		if (folio && order > 1)
-			folio_prep_large_rmappable(folio);
-		goto out;
+		return page_rmappable_folio(page);
 	}
 
 	if (unlikely(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && hugepage)) {
@@ -2247,12 +2241,7 @@ EXPORT_SYMBOL(alloc_pages);
 
 struct folio *folio_alloc(gfp_t gfp, unsigned order)
 {
-	struct page *page = alloc_pages(gfp | __GFP_COMP, order);
-	struct folio *folio = (struct folio *)page;
-
-	if (folio && order > 1)
-		folio_prep_large_rmappable(folio);
-	return folio;
+	return page_rmappable_folio(alloc_pages(gfp | __GFP_COMP, order));
 }
 EXPORT_SYMBOL(folio_alloc);
 
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 95546f376302..5b1707d9025a 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -4456,12 +4456,8 @@ struct folio *__folio_alloc(gfp_t gfp, unsigned int order, int preferred_nid,
 		nodemask_t *nodemask)
 {
 	struct page *page = __alloc_pages(gfp | __GFP_COMP, order,
-			preferred_nid, nodemask);
-	struct folio *folio = (struct folio *)page;
-
-	if (folio && order > 1)
-		folio_prep_large_rmappable(folio);
-	return folio;
+					  preferred_nid, nodemask);
+	return page_rmappable_folio(page);
 }
 EXPORT_SYMBOL(__folio_alloc);
 
-- 
2.35.3

Re: [PATCH 09/12] mm: add page_rmappable_folio() wrapper

[Matthew Wilcox]

On Mon, Sep 25, 2023 at 01:32:02AM -0700, Hugh Dickins wrote:
>  {
>  	struct page *page = __alloc_pages(gfp | __GFP_COMP, order,
> -			preferred_nid, nodemask);
> +					  preferred_nid, nodemask);

I really prefer not to do this "align arguments with opening bracket"
style.  As long as they're indented enough to make them visually distinct
from indentation-for-if-blocks, I find it annoying when functions get
renamed to something with a different length and somebody then wastes
time reindenting all the arguments to match.

> +	return page_rmappable_folio(page);

I don't particularly object to the main thrust of this patch.  I'm not
sure I like it in huge_mm.h though.  Maybe in mm/internal.h?  I
wouldn't want anyone outside mm/ calling it.

Re: [PATCH 09/12] mm: add page_rmappable_folio() wrapper

[Hugh Dickins]

On Mon, 25 Sep 2023, Matthew Wilcox wrote:
> On Mon, Sep 25, 2023 at 01:32:02AM -0700, Hugh Dickins wrote:
> >  {
> >  	struct page *page = __alloc_pages(gfp | __GFP_COMP, order,
> > -			preferred_nid, nodemask);
> > +					  preferred_nid, nodemask);
> 
> I really prefer not to do this "align arguments with opening bracket"
> style.  As long as they're indented enough to make them visually distinct
> from indentation-for-if-blocks, I find it annoying when functions get
> renamed to something with a different length and somebody then wastes
> time reindenting all the arguments to match.

Okay, I don't care much about inserting spaces to align with the bracket,
but didn't like those continuation args leftward of function name above.
I'll adjust in v2, and eventually we reach a compromise.

> 
> > +	return page_rmappable_folio(page);
> 
> I don't particularly object to the main thrust of this patch.  I'm not
> sure I like it in huge_mm.h though.  Maybe in mm/internal.h?  I
> wouldn't want anyone outside mm/ calling it.

I was expecting more resistance :)  Right, I put it in huge_mm.h to be
next to your folio_prep_large_rmappable() declaration, but it doesn't
have to be there.  Ooh, there's a folio_undo_large_rmappable() in
mm/internal.h, I'll move page_rmappable_folio() next to that.

Hugh

[PATCH 10/12] mempolicy: alloc_pages_mpol() for NUMA policy without vma

[Hugh Dickins]

Shrink shmem's stack usage by eliminating the pseudo-vma from its folio
allocation.  alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the
principal actor for passing mempolicy choice down to __alloc_pages(),
rather than vma_alloc_folio(gfp, order, vma, addr, hugepage).

vma_alloc_folio() and alloc_pages() remain, but as wrappers around
alloc_pages_mpol().  alloc_pages_bulk_*() untouched, except to provide
the additional args to policy_nodemask(), which subsumes policy_node().
Cleanup throughout, cutting out some unhelpful "helpers".

It would all be much simpler without MPOL_INTERLEAVE, but that adds a
dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd
("tmpfs: distribute interleave better across nodes"), which added ino
bias to the interleave, hidden from mm/mempolicy.c until this commit.

Hence "ilx" throughout, the "interleave index".  Originally I thought it
could be done just with nid, but that's wrong: the nodemask may come from
the shared policy layer below a shmem vma, or it may come from the task
layer above a shmem vma; and without the final nodemask then nodeid
cannot be decided.  And how ilx is applied depends also on page order.

The interleave index is almost always irrelevant unless MPOL_INTERLEAVE:
with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX
passed down from vma-less alloc_pages() is also used as hint not to use
THP-style hugepage allocation - to avoid the overhead of a hugepage arg
(though I don't understand why we never just added a GFP bit for THP -
if it actually needs a different allocation strategy from other pages of
the same order).  vma_alloc_folio() still carries its hugepage arg here,
but it is not used, and should be removed when agreed.

get_vma_policy() no longer allows a NULL vma: over time I believe we've
eradicated all the places which used to need it e.g. swapoff and madvise
used to pass NULL vma to read_swap_cache_async(), but now know the vma.

Signed-off-by: Hugh Dickins <hughd@google.com>
---
 fs/proc/task_mmu.c        |   5 +-
 include/linux/gfp.h       |  10 +-
 include/linux/mempolicy.h |  13 +-
 include/linux/mm.h        |   2 +-
 ipc/shm.c                 |  21 +--
 mm/mempolicy.c            | 383 ++++++++++++++++----------------------
 mm/shmem.c                | 102 +++++-----
 mm/swap.h                 |   9 +-
 mm/swap_state.c           |  86 +++++----
 9 files changed, 304 insertions(+), 327 deletions(-)

diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
index 3dd5be96691b..b0955a20e95f 100644
--- a/fs/proc/task_mmu.c
+++ b/fs/proc/task_mmu.c
@@ -1945,8 +1945,9 @@ static int show_numa_map(struct seq_file *m, void *v)
 	struct numa_maps *md = &numa_priv->md;
 	struct file *file = vma->vm_file;
 	struct mm_struct *mm = vma->vm_mm;
-	struct mempolicy *pol;
 	char buffer[64];
+	struct mempolicy *pol;
+	pgoff_t ilx;
 	int nid;
 
 	if (!mm)
@@ -1955,7 +1956,7 @@ static int show_numa_map(struct seq_file *m, void *v)
 	/* Ensure we start with an empty set of numa_maps statistics. */
 	memset(md, 0, sizeof(*md));
 
-	pol = __get_vma_policy(vma, vma->vm_start);
+	pol = __get_vma_policy(vma, vma->vm_start, &ilx);
 	if (pol) {
 		mpol_to_str(buffer, sizeof(buffer), pol);
 		mpol_cond_put(pol);
diff --git a/include/linux/gfp.h b/include/linux/gfp.h
index 665f06675c83..f74f8d05b053 100644
--- a/include/linux/gfp.h
+++ b/include/linux/gfp.h
@@ -8,6 +8,7 @@
 #include <linux/topology.h>
 
 struct vm_area_struct;
+struct mempolicy;
 
 /* Convert GFP flags to their corresponding migrate type */
 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
@@ -262,7 +263,9 @@ static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
 
 #ifdef CONFIG_NUMA
 struct page *alloc_pages(gfp_t gfp, unsigned int order);
-struct folio *folio_alloc(gfp_t gfp, unsigned order);
+struct page *alloc_pages_mpol(gfp_t gfp, unsigned int order,
+		struct mempolicy *mpol, pgoff_t ilx, int nid);
+struct folio *folio_alloc(gfp_t gfp, unsigned int order);
 struct folio *vma_alloc_folio(gfp_t gfp, int order, struct vm_area_struct *vma,
 		unsigned long addr, bool hugepage);
 #else
@@ -270,6 +273,11 @@ static inline struct page *alloc_pages(gfp_t gfp_mask, unsigned int order)
 {
 	return alloc_pages_node(numa_node_id(), gfp_mask, order);
 }
+static inline struct page *alloc_pages_mpol(gfp_t gfp, unsigned int order,
+		struct mempolicy *mpol, pgoff_t ilx, int nid)
+{
+	return alloc_pages(gfp, order);
+}
 static inline struct folio *folio_alloc(gfp_t gfp, unsigned int order)
 {
 	return __folio_alloc_node(gfp, order, numa_node_id());
diff --git a/include/linux/mempolicy.h b/include/linux/mempolicy.h
index 12f7dc74a457..ad93f23434bb 100644
--- a/include/linux/mempolicy.h
+++ b/include/linux/mempolicy.h
@@ -128,7 +128,9 @@ struct mempolicy *mpol_shared_policy_lookup(struct shared_policy *sp,
 
 struct mempolicy *get_task_policy(struct task_struct *p);
 struct mempolicy *__get_vma_policy(struct vm_area_struct *vma,
-		unsigned long addr);
+		unsigned long addr, pgoff_t *ilx);
+struct mempolicy *get_vma_policy(struct vm_area_struct *vma,
+		unsigned long addr, int order, pgoff_t *ilx);
 bool vma_policy_mof(struct vm_area_struct *vma);
 
 extern void numa_default_policy(void);
@@ -142,8 +144,6 @@ extern int huge_node(struct vm_area_struct *vma,
 extern bool init_nodemask_of_mempolicy(nodemask_t *mask);
 extern bool mempolicy_in_oom_domain(struct task_struct *tsk,
 				const nodemask_t *mask);
-extern nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy);
-
 extern unsigned int mempolicy_slab_node(void);
 
 extern enum zone_type policy_zone;
@@ -215,6 +215,13 @@ mpol_shared_policy_lookup(struct shared_policy *sp, pgoff_t idx)
 	return NULL;
 }
 
+static inline struct mempolicy *get_vma_policy(struct vm_area_struct *vma,
+				unsigned long addr, int order, pgoff_t *ilx)
+{
+	*ilx = 0;
+	return NULL;
+}
+
 #define vma_policy(vma) NULL
 
 static inline int
diff --git a/include/linux/mm.h b/include/linux/mm.h
index bf5d0b1b16f4..456f060b2475 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -619,7 +619,7 @@ struct vm_operations_struct {
 	 * policy.
 	 */
 	struct mempolicy *(*get_policy)(struct vm_area_struct *vma,
-					unsigned long addr);
+					unsigned long addr, pgoff_t *ilx);
 #endif
 	/*
 	 * Called by vm_normal_page() for special PTEs to find the
diff --git a/ipc/shm.c b/ipc/shm.c
index 576a543b7cff..222aaf035afb 100644
--- a/ipc/shm.c
+++ b/ipc/shm.c
@@ -562,30 +562,25 @@ static unsigned long shm_pagesize(struct vm_area_struct *vma)
 }
 
 #ifdef CONFIG_NUMA
-static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
+static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
 {
-	struct file *file = vma->vm_file;
-	struct shm_file_data *sfd = shm_file_data(file);
+	struct shm_file_data *sfd = shm_file_data(vma->vm_file);
 	int err = 0;
 
 	if (sfd->vm_ops->set_policy)
-		err = sfd->vm_ops->set_policy(vma, new);
+		err = sfd->vm_ops->set_policy(vma, mpol);
 	return err;
 }
 
 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
-					unsigned long addr)
+					unsigned long addr, pgoff_t *ilx)
 {
-	struct file *file = vma->vm_file;
-	struct shm_file_data *sfd = shm_file_data(file);
-	struct mempolicy *pol = NULL;
+	struct shm_file_data *sfd = shm_file_data(vma->vm_file);
+	struct mempolicy *mpol = vma->vm_policy;
 
 	if (sfd->vm_ops->get_policy)
-		pol = sfd->vm_ops->get_policy(vma, addr);
-	else if (vma->vm_policy)
-		pol = vma->vm_policy;
-
-	return pol;
+		mpol = sfd->vm_ops->get_policy(vma, addr, ilx);
+	return mpol;
 }
 #endif
 
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 4c3b3f535630..d74df1e1b14a 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -114,6 +114,8 @@
 #define MPOL_MF_INVERT       (MPOL_MF_INTERNAL << 1)	/* Invert check for nodemask */
 #define MPOL_MF_WRLOCK       (MPOL_MF_INTERNAL << 2)	/* Write-lock walked vmas */
 
+#define NO_INTERLEAVE_INDEX (-1UL)
+
 static struct kmem_cache *policy_cache;
 static struct kmem_cache *sn_cache;
 
@@ -915,6 +917,7 @@ static long do_get_mempolicy(int *policy, nodemask_t *nmask,
 	}
 
 	if (flags & MPOL_F_ADDR) {
+		pgoff_t ilx;		/* ignored here */
 		/*
 		 * Do NOT fall back to task policy if the
 		 * vma/shared policy at addr is NULL.  We
@@ -926,10 +929,7 @@ static long do_get_mempolicy(int *policy, nodemask_t *nmask,
 			mmap_read_unlock(mm);
 			return -EFAULT;
 		}
-		if (vma->vm_ops && vma->vm_ops->get_policy)
-			pol = vma->vm_ops->get_policy(vma, addr);
-		else
-			pol = vma->vm_policy;
+		pol = __get_vma_policy(vma, addr, &ilx);
 	} else if (addr)
 		return -EINVAL;
 
@@ -1187,6 +1187,15 @@ static struct folio *new_folio(struct folio *src, unsigned long start)
 			break;
 	}
 
+	/*
+	 * __get_vma_policy() now expects a genuine non-NULL vma. Return NULL
+	 * when the page can no longer be located in a vma: that is not ideal
+	 * (migrate_pages() will give up early, presuming ENOMEM), but good
+	 * enough to avoid a crash by syzkaller or concurrent holepunch.
+	 */
+	if (!vma)
+		return NULL;
+
 	if (folio_test_hugetlb(src)) {
 		return alloc_hugetlb_folio_vma(folio_hstate(src),
 				vma, address);
@@ -1195,9 +1204,6 @@ static struct folio *new_folio(struct folio *src, unsigned long start)
 	if (folio_test_large(src))
 		gfp = GFP_TRANSHUGE;
 
-	/*
-	 * if !vma, vma_alloc_folio() will use task or system default policy
-	 */
 	return vma_alloc_folio(gfp, folio_order(src), vma, address,
 			folio_test_large(src));
 }
@@ -1705,34 +1711,19 @@ bool vma_migratable(struct vm_area_struct *vma)
 }
 
 struct mempolicy *__get_vma_policy(struct vm_area_struct *vma,
-						unsigned long addr)
+				   unsigned long addr, pgoff_t *ilx)
 {
-	struct mempolicy *pol = NULL;
-
-	if (vma) {
-		if (vma->vm_ops && vma->vm_ops->get_policy) {
-			pol = vma->vm_ops->get_policy(vma, addr);
-		} else if (vma->vm_policy) {
-			pol = vma->vm_policy;
-
-			/*
-			 * shmem_alloc_page() passes MPOL_F_SHARED policy with
-			 * a pseudo vma whose vma->vm_ops=NULL. Take a reference
-			 * count on these policies which will be dropped by
-			 * mpol_cond_put() later
-			 */
-			if (mpol_needs_cond_ref(pol))
-				mpol_get(pol);
-		}
-	}
-
-	return pol;
+	*ilx = 0;
+	return (vma->vm_ops && vma->vm_ops->get_policy) ?
+		vma->vm_ops->get_policy(vma, addr, ilx) : vma->vm_policy;
 }
 
 /*
- * get_vma_policy(@vma, @addr)
+ * get_vma_policy(@vma, @addr, @order, @ilx)
  * @vma: virtual memory area whose policy is sought
  * @addr: address in @vma for shared policy lookup
+ * @order: 0, or appropriate huge_page_order for interleaving
+ * @ilx: interleave index (output), for use only when MPOL_INTERLEAVE
  *
  * Returns effective policy for a VMA at specified address.
  * Falls back to current->mempolicy or system default policy, as necessary.
@@ -1741,14 +1732,18 @@ struct mempolicy *__get_vma_policy(struct vm_area_struct *vma,
  * freeing by another task.  It is the caller's responsibility to free the
  * extra reference for shared policies.
  */
-static struct mempolicy *get_vma_policy(struct vm_area_struct *vma,
-						unsigned long addr)
+struct mempolicy *get_vma_policy(struct vm_area_struct *vma,
+				 unsigned long addr, int order, pgoff_t *ilx)
 {
-	struct mempolicy *pol = __get_vma_policy(vma, addr);
+	struct mempolicy *pol;
 
+	pol = __get_vma_policy(vma, addr, ilx);
 	if (!pol)
 		pol = get_task_policy(current);
-
+	if (pol->mode == MPOL_INTERLEAVE) {
+		*ilx += vma->vm_pgoff >> order;
+		*ilx += (addr - vma->vm_start) >> (PAGE_SHIFT + order);
+	}
 	return pol;
 }
 
@@ -1758,8 +1753,9 @@ bool vma_policy_mof(struct vm_area_struct *vma)
 
 	if (vma->vm_ops && vma->vm_ops->get_policy) {
 		bool ret = false;
+		pgoff_t ilx;		/* ignored here */
 
-		pol = vma->vm_ops->get_policy(vma, vma->vm_start);
+		pol = vma->vm_ops->get_policy(vma, vma->vm_start, &ilx);
 		if (pol && (pol->flags & MPOL_F_MOF))
 			ret = true;
 		mpol_cond_put(pol);
@@ -1794,54 +1790,6 @@ bool apply_policy_zone(struct mempolicy *policy, enum zone_type zone)
 	return zone >= dynamic_policy_zone;
 }
 
-/*
- * Return a nodemask representing a mempolicy for filtering nodes for
- * page allocation
- */
-nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
-{
-	int mode = policy->mode;
-
-	/* Lower zones don't get a nodemask applied for MPOL_BIND */
-	if (unlikely(mode == MPOL_BIND) &&
-		apply_policy_zone(policy, gfp_zone(gfp)) &&
-		cpuset_nodemask_valid_mems_allowed(&policy->nodes))
-		return &policy->nodes;
-
-	if (mode == MPOL_PREFERRED_MANY)
-		return &policy->nodes;
-
-	return NULL;
-}
-
-/*
- * Return the  preferred node id for 'prefer' mempolicy, and return
- * the given id for all other policies.
- *
- * policy_node() is always coupled with policy_nodemask(), which
- * secures the nodemask limit for 'bind' and 'prefer-many' policy.
- */
-static int policy_node(gfp_t gfp, struct mempolicy *policy, int nid)
-{
-	if (policy->mode == MPOL_PREFERRED) {
-		nid = first_node(policy->nodes);
-	} else {
-		/*
-		 * __GFP_THISNODE shouldn't even be used with the bind policy
-		 * because we might easily break the expectation to stay on the
-		 * requested node and not break the policy.
-		 */
-		WARN_ON_ONCE(policy->mode == MPOL_BIND && (gfp & __GFP_THISNODE));
-	}
-
-	if ((policy->mode == MPOL_BIND ||
-	     policy->mode == MPOL_PREFERRED_MANY) &&
-	    policy->home_node != NUMA_NO_NODE)
-		return policy->home_node;
-
-	return nid;
-}
-
 /* Do dynamic interleaving for a process */
 static unsigned int interleave_nodes(struct mempolicy *policy)
 {
@@ -1901,11 +1849,11 @@ unsigned int mempolicy_slab_node(void)
 }
 
 /*
- * Do static interleaving for a VMA with known offset @n.  Returns the n'th
- * node in pol->nodes (starting from n=0), wrapping around if n exceeds the
- * number of present nodes.
+ * Do static interleaving for interleave index @ilx.  Returns the ilx'th
+ * node in pol->nodes (starting from ilx=0), wrapping around if ilx
+ * exceeds the number of present nodes.
  */
-static unsigned offset_il_node(struct mempolicy *pol, unsigned long n)
+static unsigned int interleave_nid(struct mempolicy *pol, pgoff_t ilx)
 {
 	nodemask_t nodemask = pol->nodes;
 	unsigned int target, nnodes;
@@ -1923,33 +1871,54 @@ static unsigned offset_il_node(struct mempolicy *pol, unsigned long n)
 	nnodes = nodes_weight(nodemask);
 	if (!nnodes)
 		return numa_node_id();
-	target = (unsigned int)n % nnodes;
+	target = ilx % nnodes;
 	nid = first_node(nodemask);
 	for (i = 0; i < target; i++)
 		nid = next_node(nid, nodemask);
 	return nid;
 }
 
-/* Determine a node number for interleave */
-static inline unsigned interleave_nid(struct mempolicy *pol,
-		 struct vm_area_struct *vma, unsigned long addr, int shift)
+/*
+ * Return a nodemask representing a mempolicy for filtering nodes for
+ * page allocation, together with preferred node id (or the input node id).
+ */
+static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *pol,
+				   pgoff_t ilx, int *nid)
 {
-	if (vma) {
-		unsigned long off;
+	nodemask_t *nodemask = NULL;
 
+	switch (pol->mode) {
+	case MPOL_PREFERRED:
+		/* Override input node id */
+		*nid = first_node(pol->nodes);
+		break;
+	case MPOL_PREFERRED_MANY:
+		nodemask = &pol->nodes;
+		if (pol->home_node != NUMA_NO_NODE)
+			*nid = pol->home_node;
+		break;
+	case MPOL_BIND:
+		/* Restrict to nodemask (but not on lower zones) */
+		if (apply_policy_zone(pol, gfp_zone(gfp)) &&
+		    cpuset_nodemask_valid_mems_allowed(&pol->nodes))
+			nodemask = &pol->nodes;
+		if (pol->home_node != NUMA_NO_NODE)
+			*nid = pol->home_node;
 		/*
-		 * for small pages, there is no difference between
-		 * shift and PAGE_SHIFT, so the bit-shift is safe.
-		 * for huge pages, since vm_pgoff is in units of small
-		 * pages, we need to shift off the always 0 bits to get
-		 * a useful offset.
+		 * __GFP_THISNODE shouldn't even be used with the bind policy
+		 * because we might easily break the expectation to stay on the
+		 * requested node and not break the policy.
 		 */
-		BUG_ON(shift < PAGE_SHIFT);
-		off = vma->vm_pgoff >> (shift - PAGE_SHIFT);
-		off += (addr - vma->vm_start) >> shift;
-		return offset_il_node(pol, off);
-	} else
-		return interleave_nodes(pol);
+		WARN_ON_ONCE(gfp & __GFP_THISNODE);
+		break;
+	case MPOL_INTERLEAVE:
+		/* Override input node id */
+		*nid = (ilx == NO_INTERLEAVE_INDEX) ?
+			interleave_nodes(pol) : interleave_nid(pol, ilx);
+		break;
+	}
+
+	return nodemask;
 }
 
 #ifdef CONFIG_HUGETLBFS
@@ -1965,27 +1934,16 @@ static inline unsigned interleave_nid(struct mempolicy *pol,
  * to the struct mempolicy for conditional unref after allocation.
  * If the effective policy is 'bind' or 'prefer-many', returns a pointer
  * to the mempolicy's @nodemask for filtering the zonelist.
- *
- * Must be protected by read_mems_allowed_begin()
  */
 int huge_node(struct vm_area_struct *vma, unsigned long addr, gfp_t gfp_flags,
-				struct mempolicy **mpol, nodemask_t **nodemask)
+		struct mempolicy **mpol, nodemask_t **nodemask)
 {
+	pgoff_t ilx;
 	int nid;
-	int mode;
 
-	*mpol = get_vma_policy(vma, addr);
-	*nodemask = NULL;
-	mode = (*mpol)->mode;
-
-	if (unlikely(mode == MPOL_INTERLEAVE)) {
-		nid = interleave_nid(*mpol, vma, addr,
-					huge_page_shift(hstate_vma(vma)));
-	} else {
-		nid = policy_node(gfp_flags, *mpol, numa_node_id());
-		if (mode == MPOL_BIND || mode == MPOL_PREFERRED_MANY)
-			*nodemask = &(*mpol)->nodes;
-	}
+	nid = numa_node_id();
+	*mpol = get_vma_policy(vma, addr, hstate_vma(vma)->order, &ilx);
+	*nodemask = policy_nodemask(gfp_flags, *mpol, ilx, &nid);
 	return nid;
 }
 
@@ -2063,27 +2021,8 @@ bool mempolicy_in_oom_domain(struct task_struct *tsk,
 	return ret;
 }
 
-/* Allocate a page in interleaved policy.
-   Own path because it needs to do special accounting. */
-static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
-					unsigned nid)
-{
-	struct page *page;
-
-	page = __alloc_pages(gfp, order, nid, NULL);
-	/* skip NUMA_INTERLEAVE_HIT counter update if numa stats is disabled */
-	if (!static_branch_likely(&vm_numa_stat_key))
-		return page;
-	if (page && page_to_nid(page) == nid) {
-		preempt_disable();
-		__count_numa_event(page_zone(page), NUMA_INTERLEAVE_HIT);
-		preempt_enable();
-	}
-	return page;
-}
-
 static struct page *alloc_pages_preferred_many(gfp_t gfp, unsigned int order,
-						int nid, struct mempolicy *pol)
+						int nid, nodemask_t *nodemask)
 {
 	struct page *page;
 	gfp_t preferred_gfp;
@@ -2096,7 +2035,7 @@ static struct page *alloc_pages_preferred_many(gfp_t gfp, unsigned int order,
 	 */
 	preferred_gfp = gfp | __GFP_NOWARN;
 	preferred_gfp &= ~(__GFP_DIRECT_RECLAIM | __GFP_NOFAIL);
-	page = __alloc_pages(preferred_gfp, order, nid, &pol->nodes);
+	page = __alloc_pages(preferred_gfp, order, nid, nodemask);
 	if (!page)
 		page = __alloc_pages(gfp, order, nid, NULL);
 
@@ -2104,55 +2043,29 @@ static struct page *alloc_pages_preferred_many(gfp_t gfp, unsigned int order,
 }
 
 /**
- * vma_alloc_folio - Allocate a folio for a VMA.
+ * alloc_pages_mpol - Allocate pages according to NUMA mempolicy.
  * @gfp: GFP flags.
- * @order: Order of the folio.
- * @vma: Pointer to VMA or NULL if not available.
- * @addr: Virtual address of the allocation.  Must be inside @vma.
- * @hugepage: For hugepages try only the preferred node if possible.
+ * @order: Order of the page allocation.
+ * @pol: Pointer to the NUMA mempolicy.
+ * @ilx: Index for interleave mempolicy (also distinguishes alloc_pages()).
+ * @nid: Preferred node (usually numa_node_id() but @mpol may override it).
  *
- * Allocate a folio for a specific address in @vma, using the appropriate
- * NUMA policy.  When @vma is not NULL the caller must hold the mmap_lock
- * of the mm_struct of the VMA to prevent it from going away.  Should be
- * used for all allocations for folios that will be mapped into user space.
- *
- * Return: The folio on success or NULL if allocation fails.
+ * Return: The page on success or NULL if allocation fails.
  */
-struct folio *vma_alloc_folio(gfp_t gfp, int order, struct vm_area_struct *vma,
-		unsigned long addr, bool hugepage)
+struct page *alloc_pages_mpol(gfp_t gfp, unsigned int order,
+		struct mempolicy *pol, pgoff_t ilx, int nid)
 {
-	struct mempolicy *pol;
-	int node = numa_node_id();
-	struct folio *folio;
-	int preferred_nid;
-	nodemask_t *nmask;
+	nodemask_t *nodemask;
+	struct page *page;
 
-	pol = get_vma_policy(vma, addr);
+	nodemask = policy_nodemask(gfp, pol, ilx, &nid);
 
-	if (pol->mode == MPOL_INTERLEAVE) {
-		struct page *page;
-		unsigned nid;
-
-		nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order);
-		mpol_cond_put(pol);
-		gfp |= __GFP_COMP;
-		page = alloc_page_interleave(gfp, order, nid);
-		return page_rmappable_folio(page);
-	}
-
-	if (pol->mode == MPOL_PREFERRED_MANY) {
-		struct page *page;
-
-		node = policy_node(gfp, pol, node);
-		gfp |= __GFP_COMP;
-		page = alloc_pages_preferred_many(gfp, order, node, pol);
-		mpol_cond_put(pol);
-		return page_rmappable_folio(page);
-	}
-
-	if (unlikely(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && hugepage)) {
-		int hpage_node = node;
+	if (pol->mode == MPOL_PREFERRED_MANY)
+		return alloc_pages_preferred_many(gfp, order, nid, nodemask);
 
+	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
+	    /* filter "hugepage" allocation, unless from alloc_pages() */
+	    order == HPAGE_PMD_ORDER && ilx != NO_INTERLEAVE_INDEX) {
 		/*
 		 * For hugepage allocation and non-interleave policy which
 		 * allows the current node (or other explicitly preferred
@@ -2163,39 +2076,68 @@ struct folio *vma_alloc_folio(gfp_t gfp, int order, struct vm_area_struct *vma,
 		 * If the policy is interleave or does not allow the current
 		 * node in its nodemask, we allocate the standard way.
 		 */
-		if (pol->mode == MPOL_PREFERRED)
-			hpage_node = first_node(pol->nodes);
-
-		nmask = policy_nodemask(gfp, pol);
-		if (!nmask || node_isset(hpage_node, *nmask)) {
-			mpol_cond_put(pol);
+		if (pol->mode != MPOL_INTERLEAVE &&
+		    (!nodemask || node_isset(nid, *nodemask))) {
 			/*
 			 * First, try to allocate THP only on local node, but
 			 * don't reclaim unnecessarily, just compact.
 			 */
-			folio = __folio_alloc_node(gfp | __GFP_THISNODE |
-					__GFP_NORETRY, order, hpage_node);
-
+			page = __alloc_pages_node(nid,
+				gfp | __GFP_THISNODE | __GFP_NORETRY, order);
+			if (page || !(gfp & __GFP_DIRECT_RECLAIM))
+				return page;
 			/*
 			 * If hugepage allocations are configured to always
 			 * synchronous compact or the vma has been madvised
 			 * to prefer hugepage backing, retry allowing remote
 			 * memory with both reclaim and compact as well.
 			 */
-			if (!folio && (gfp & __GFP_DIRECT_RECLAIM))
-				folio = __folio_alloc(gfp, order, hpage_node,
-						      nmask);
-
-			goto out;
 		}
 	}
 
-	nmask = policy_nodemask(gfp, pol);
-	preferred_nid = policy_node(gfp, pol, node);
-	folio = __folio_alloc(gfp, order, preferred_nid, nmask);
+	page = __alloc_pages(gfp, order, nid, nodemask);
+
+	if (unlikely(pol->mode == MPOL_INTERLEAVE) && page) {
+		/* skip NUMA_INTERLEAVE_HIT update if numa stats is disabled */
+		if (static_branch_likely(&vm_numa_stat_key) &&
+		    page_to_nid(page) == nid) {
+			preempt_disable();
+			__count_numa_event(page_zone(page), NUMA_INTERLEAVE_HIT);
+			preempt_enable();
+		}
+	}
+
+	return page;
+}
+
+/**
+ * vma_alloc_folio - Allocate a folio for a VMA.
+ * @gfp: GFP flags.
+ * @order: Order of the folio.
+ * @vma: Pointer to VMA.
+ * @addr: Virtual address of the allocation.  Must be inside @vma.
+ * @hugepage: Unused (was: For hugepages try only preferred node if possible).
+ *
+ * Allocate a folio for a specific address in @vma, using the appropriate
+ * NUMA policy.  The caller must hold the mmap_lock of the mm_struct of the
+ * VMA to prevent it from going away.  Should be used for all allocations
+ * for folios that will be mapped into user space, excepting hugetlbfs, and
+ * excepting where direct use of alloc_pages_mpol() is more appropriate.
+ *
+ * Return: The folio on success or NULL if allocation fails.
+ */
+struct folio *vma_alloc_folio(gfp_t gfp, int order, struct vm_area_struct *vma,
+		unsigned long addr, bool hugepage)
+{
+	struct mempolicy *pol;
+	pgoff_t ilx;
+	struct page *page;
+
+	pol = get_vma_policy(vma, addr, order, &ilx);
+	page = alloc_pages_mpol(gfp | __GFP_COMP, order,
+				pol, ilx, numa_node_id());
 	mpol_cond_put(pol);
-out:
-	return folio;
+	return page_rmappable_folio(page);
 }
 EXPORT_SYMBOL(vma_alloc_folio);
 
@@ -2213,33 +2155,23 @@ EXPORT_SYMBOL(vma_alloc_folio);
  * flags are used.
  * Return: The page on success or NULL if allocation fails.
  */
-struct page *alloc_pages(gfp_t gfp, unsigned order)
+struct page *alloc_pages(gfp_t gfp, unsigned int order)
 {
 	struct mempolicy *pol = &default_policy;
-	struct page *page;
-
-	if (!in_interrupt() && !(gfp & __GFP_THISNODE))
-		pol = get_task_policy(current);
 
 	/*
 	 * No reference counting needed for current->mempolicy
 	 * nor system default_policy
 	 */
-	if (pol->mode == MPOL_INTERLEAVE)
-		page = alloc_page_interleave(gfp, order, interleave_nodes(pol));
-	else if (pol->mode == MPOL_PREFERRED_MANY)
-		page = alloc_pages_preferred_many(gfp, order,
-				  policy_node(gfp, pol, numa_node_id()), pol);
-	else
-		page = __alloc_pages(gfp, order,
-				policy_node(gfp, pol, numa_node_id()),
-				policy_nodemask(gfp, pol));
+	if (!in_interrupt() && !(gfp & __GFP_THISNODE))
+		pol = get_task_policy(current);
 
-	return page;
+	return alloc_pages_mpol(gfp, order,
+				pol, NO_INTERLEAVE_INDEX, numa_node_id());
 }
 EXPORT_SYMBOL(alloc_pages);
 
-struct folio *folio_alloc(gfp_t gfp, unsigned order)
+struct folio *folio_alloc(gfp_t gfp, unsigned int order)
 {
 	return page_rmappable_folio(alloc_pages(gfp | __GFP_COMP, order));
 }
@@ -2310,6 +2242,8 @@ unsigned long alloc_pages_bulk_array_mempolicy(gfp_t gfp,
 		unsigned long nr_pages, struct page **page_array)
 {
 	struct mempolicy *pol = &default_policy;
+	nodemask_t *nodemask;
+	int nid;
 
 	if (!in_interrupt() && !(gfp & __GFP_THISNODE))
 		pol = get_task_policy(current);
@@ -2322,9 +2256,10 @@ unsigned long alloc_pages_bulk_array_mempolicy(gfp_t gfp,
 		return alloc_pages_bulk_array_preferred_many(gfp,
 				numa_node_id(), pol, nr_pages, page_array);
 
-	return __alloc_pages_bulk(gfp, policy_node(gfp, pol, numa_node_id()),
-				  policy_nodemask(gfp, pol), nr_pages, NULL,
-				  page_array);
+	nid = numa_node_id();
+	nodemask = policy_nodemask(gfp, pol, NO_INTERLEAVE_INDEX, &nid);
+	return __alloc_pages_bulk(gfp, nid, nodemask,
+				  nr_pages, NULL, page_array);
 }
 
 int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst)
@@ -2510,23 +2445,21 @@ static void sp_free(struct sp_node *n)
 int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr)
 {
 	struct mempolicy *pol;
+	pgoff_t ilx;
 	struct zoneref *z;
 	int curnid = page_to_nid(page);
-	unsigned long pgoff;
 	int thiscpu = raw_smp_processor_id();
 	int thisnid = cpu_to_node(thiscpu);
 	int polnid = NUMA_NO_NODE;
 	int ret = NUMA_NO_NODE;
 
-	pol = get_vma_policy(vma, addr);
+	pol = get_vma_policy(vma, addr, compound_order(page), &ilx);
 	if (!(pol->flags & MPOL_F_MOF))
 		goto out;
 
 	switch (pol->mode) {
 	case MPOL_INTERLEAVE:
-		pgoff = vma->vm_pgoff;
-		pgoff += (addr - vma->vm_start) >> PAGE_SHIFT;
-		polnid = offset_il_node(pol, pgoff);
+		polnid = interleave_nid(pol, ilx);
 		break;
 
 	case MPOL_PREFERRED:
diff --git a/mm/shmem.c b/mm/shmem.c
index 69595d341882..aaf44aec2826 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -1565,38 +1565,20 @@ static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
 	return NULL;
 }
 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
-#ifndef CONFIG_NUMA
-#define vm_policy vm_private_data
-#endif
 
-static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
-		struct shmem_inode_info *info, pgoff_t index)
-{
-	/* Create a pseudo vma that just contains the policy */
-	vma_init(vma, NULL);
-	/* Bias interleave by inode number to distribute better across nodes */
-	vma->vm_pgoff = index + info->vfs_inode.i_ino;
-	vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
-}
+static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
+			pgoff_t index, unsigned int order, pgoff_t *ilx);
 
-static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
-{
-	/* Drop reference taken by mpol_shared_policy_lookup() */
-	mpol_cond_put(vma->vm_policy);
-}
-
-static struct folio *shmem_swapin(swp_entry_t swap, gfp_t gfp,
+static struct folio *shmem_swapin_cluster(swp_entry_t swap, gfp_t gfp,
 			struct shmem_inode_info *info, pgoff_t index)
 {
-	struct vm_area_struct pvma;
+	struct mempolicy *mpol;
+	pgoff_t ilx;
 	struct page *page;
-	struct vm_fault vmf = {
-		.vma = &pvma,
-	};
 
-	shmem_pseudo_vma_init(&pvma, info, index);
-	page = swap_cluster_readahead(swap, gfp, &vmf);
-	shmem_pseudo_vma_destroy(&pvma);
+	mpol = shmem_get_pgoff_policy(info, index, 0, &ilx);
+	page = swap_cluster_readahead(swap, gfp, mpol, ilx);
+	mpol_cond_put(mpol);
 
 	if (!page)
 		return NULL;
@@ -1630,35 +1612,37 @@ static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
 static struct folio *shmem_alloc_hugefolio(gfp_t gfp,
 		struct shmem_inode_info *info, pgoff_t index)
 {
-	struct vm_area_struct pvma;
 	struct address_space *mapping = info->vfs_inode.i_mapping;
-	pgoff_t hindex;
-	struct folio *folio;
+	struct mempolicy *mpol;
+	pgoff_t ilx;
+	struct page *page;
 
-	hindex = round_down(index, HPAGE_PMD_NR);
-	if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
+	index = round_down(index, HPAGE_PMD_NR);
+	if (xa_find(&mapping->i_pages, &index, index + HPAGE_PMD_NR - 1,
 								XA_PRESENT))
 		return NULL;
 
-	shmem_pseudo_vma_init(&pvma, info, hindex);
-	folio = vma_alloc_folio(gfp, HPAGE_PMD_ORDER, &pvma, 0, true);
-	shmem_pseudo_vma_destroy(&pvma);
-	if (!folio)
+	mpol = shmem_get_pgoff_policy(info, index, HPAGE_PMD_ORDER, &ilx);
+	page = alloc_pages_mpol(gfp, HPAGE_PMD_ORDER, mpol, ilx, numa_node_id());
+	mpol_cond_put(mpol);
+
+	if (!page)
 		count_vm_event(THP_FILE_FALLBACK);
-	return folio;
+	return page_rmappable_folio(page);
 }
 
 static struct folio *shmem_alloc_folio(gfp_t gfp,
-			struct shmem_inode_info *info, pgoff_t index)
+		struct shmem_inode_info *info, pgoff_t index)
 {
-	struct vm_area_struct pvma;
-	struct folio *folio;
+	struct mempolicy *mpol;
+	pgoff_t ilx;
+	struct page *page;
 
-	shmem_pseudo_vma_init(&pvma, info, index);
-	folio = vma_alloc_folio(gfp, 0, &pvma, 0, false);
-	shmem_pseudo_vma_destroy(&pvma);
+	mpol = shmem_get_pgoff_policy(info, index, 0, &ilx);
+	page = alloc_pages_mpol(gfp, 0, mpol, ilx, numa_node_id());
+	mpol_cond_put(mpol);
 
-	return folio;
+	return (struct folio *)page;
 }
 
 static struct folio *shmem_alloc_and_acct_folio(gfp_t gfp, struct inode *inode,
@@ -1848,7 +1832,7 @@ static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
 			count_memcg_event_mm(charge_mm, PGMAJFAULT);
 		}
 		/* Here we actually start the io */
-		folio = shmem_swapin(swap, gfp, info, index);
+		folio = shmem_swapin_cluster(swap, gfp, info, index);
 		if (!folio) {
 			error = -ENOMEM;
 			goto failed;
@@ -2330,15 +2314,41 @@ static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
 }
 
 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
-					  unsigned long addr)
+					  unsigned long addr, pgoff_t *ilx)
 {
 	struct inode *inode = file_inode(vma->vm_file);
 	pgoff_t index;
 
+	/*
+	 * Bias interleave by inode number to distribute better across nodes;
+	 * but this interface is independent of which page order is used, so
+	 * supplies only that bias, letting caller apply the offset (adjusted
+	 * by page order, as in shmem_get_pgoff_policy() and get_vma_policy()).
+	 */
+	*ilx = inode->i_ino;
 	index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 	return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
 }
-#endif
+
+static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
+			pgoff_t index, unsigned int order, pgoff_t *ilx)
+{
+	struct mempolicy *mpol;
+
+	/* Bias interleave by inode number to distribute better across nodes */
+	*ilx = info->vfs_inode.i_ino + (index >> order);
+
+	mpol = mpol_shared_policy_lookup(&info->policy, index);
+	return mpol ? mpol : get_task_policy(current);
+}
+#else
+static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
+			pgoff_t index, unsigned int order, pgoff_t *ilx)
+{
+	*ilx = 0;
+	return NULL;
+}
+#endif /* CONFIG_NUMA */
 
 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
 {
diff --git a/mm/swap.h b/mm/swap.h
index 8a3c7a0ace4f..73c332ee4d91 100644
--- a/mm/swap.h
+++ b/mm/swap.h
@@ -2,6 +2,8 @@
 #ifndef _MM_SWAP_H
 #define _MM_SWAP_H
 
+struct mempolicy;
+
 #ifdef CONFIG_SWAP
 #include <linux/blk_types.h> /* for bio_end_io_t */
 
@@ -48,11 +50,10 @@ struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
 				   unsigned long addr,
 				   struct swap_iocb **plug);
 struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
-				     struct vm_area_struct *vma,
-				     unsigned long addr,
+				     struct mempolicy *mpol, pgoff_t ilx,
 				     bool *new_page_allocated);
 struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t flag,
-				    struct vm_fault *vmf);
+				    struct mempolicy *mpol, pgoff_t ilx);
 struct page *swapin_readahead(swp_entry_t entry, gfp_t flag,
 			      struct vm_fault *vmf);
 
@@ -80,7 +81,7 @@ static inline void show_swap_cache_info(void)
 }
 
 static inline struct page *swap_cluster_readahead(swp_entry_t entry,
-				gfp_t gfp_mask, struct vm_fault *vmf)
+			gfp_t gfp_mask, struct mempolicy *mpol, pgoff_t ilx)
 {
 	return NULL;
 }
diff --git a/mm/swap_state.c b/mm/swap_state.c
index b3b14bd0dd64..116d4d8a930e 100644
--- a/mm/swap_state.c
+++ b/mm/swap_state.c
@@ -10,6 +10,7 @@
 #include <linux/mm.h>
 #include <linux/gfp.h>
 #include <linux/kernel_stat.h>
+#include <linux/mempolicy.h>
 #include <linux/swap.h>
 #include <linux/swapops.h>
 #include <linux/init.h>
@@ -410,8 +411,8 @@ struct folio *filemap_get_incore_folio(struct address_space *mapping,
 }
 
 struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
-			struct vm_area_struct *vma, unsigned long addr,
-			bool *new_page_allocated)
+				     struct mempolicy *mpol, pgoff_t ilx,
+				     bool *new_page_allocated)
 {
 	struct swap_info_struct *si;
 	struct folio *folio;
@@ -453,7 +454,8 @@ struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
 		 * before marking swap_map SWAP_HAS_CACHE, when -EEXIST will
 		 * cause any racers to loop around until we add it to cache.
 		 */
-		folio = vma_alloc_folio(gfp_mask, 0, vma, addr, false);
+		folio = (struct folio *)alloc_pages_mpol(gfp_mask, 0,
+						mpol, ilx, numa_node_id());
 		if (!folio)
                         goto fail_put_swap;
 
@@ -528,14 +530,19 @@ struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
 				   struct vm_area_struct *vma,
 				   unsigned long addr, struct swap_iocb **plug)
 {
-	bool page_was_allocated;
-	struct page *retpage = __read_swap_cache_async(entry, gfp_mask,
-			vma, addr, &page_was_allocated);
+	bool page_allocated;
+	struct mempolicy *mpol;
+	pgoff_t ilx;
+	struct page *page;
 
-	if (page_was_allocated)
-		swap_readpage(retpage, false, plug);
+	mpol = get_vma_policy(vma, addr, 0, &ilx);
+	page = __read_swap_cache_async(entry, gfp_mask, mpol, ilx,
+					&page_allocated);
+	mpol_cond_put(mpol);
 
-	return retpage;
+	if (page_allocated)
+		swap_readpage(page, false, plug);
+	return page;
 }
 
 static unsigned int __swapin_nr_pages(unsigned long prev_offset,
@@ -603,7 +610,8 @@ static unsigned long swapin_nr_pages(unsigned long offset)
  * swap_cluster_readahead - swap in pages in hope we need them soon
  * @entry: swap entry of this memory
  * @gfp_mask: memory allocation flags
- * @vmf: fault information
+ * @mpol: NUMA memory allocation policy to be applied
+ * @ilx: NUMA interleave index, for use only when MPOL_INTERLEAVE
  *
  * Returns the struct page for entry and addr, after queueing swapin.
  *
@@ -612,13 +620,12 @@ static unsigned long swapin_nr_pages(unsigned long offset)
  * because it doesn't cost us any seek time.  We also make sure to queue
  * the 'original' request together with the readahead ones...
  *
- * This has been extended to use the NUMA policies from the mm triggering
- * the readahead.
- *
- * Caller must hold read mmap_lock if vmf->vma is not NULL.
+ * Note: it is intentional that the same NUMA policy and interleave index
+ * are used for every page of the readahead: neighbouring pages on swap
+ * are fairly likely to have been swapped out from the same node.
  */
 struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
-				struct vm_fault *vmf)
+				    struct mempolicy *mpol, pgoff_t ilx)
 {
 	struct page *page;
 	unsigned long entry_offset = swp_offset(entry);
@@ -629,8 +636,6 @@ struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
 	struct blk_plug plug;
 	struct swap_iocb *splug = NULL;
 	bool page_allocated;
-	struct vm_area_struct *vma = vmf->vma;
-	unsigned long addr = vmf->address;
 
 	mask = swapin_nr_pages(offset) - 1;
 	if (!mask)
@@ -648,8 +653,8 @@ struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
 	for (offset = start_offset; offset <= end_offset ; offset++) {
 		/* Ok, do the async read-ahead now */
 		page = __read_swap_cache_async(
-			swp_entry(swp_type(entry), offset),
-			gfp_mask, vma, addr, &page_allocated);
+				swp_entry(swp_type(entry), offset),
+				gfp_mask, mpol, ilx, &page_allocated);
 		if (!page)
 			continue;
 		if (page_allocated) {
@@ -663,11 +668,14 @@ struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
 	}
 	blk_finish_plug(&plug);
 	swap_read_unplug(splug);
-
 	lru_add_drain();	/* Push any new pages onto the LRU now */
 skip:
 	/* The page was likely read above, so no need for plugging here */
-	return read_swap_cache_async(entry, gfp_mask, vma, addr, NULL);
+	page = __read_swap_cache_async(entry, gfp_mask, mpol, ilx,
+					&page_allocated);
+	if (page_allocated)
+		swap_readpage(page, false, NULL);
+	return page;
 }
 
 int init_swap_address_space(unsigned int type, unsigned long nr_pages)
@@ -765,8 +773,10 @@ static void swap_ra_info(struct vm_fault *vmf,
 
 /**
  * swap_vma_readahead - swap in pages in hope we need them soon
- * @fentry: swap entry of this memory
+ * @targ_entry: swap entry of the targeted memory
  * @gfp_mask: memory allocation flags
+ * @mpol: NUMA memory allocation policy to be applied
+ * @targ_ilx: NUMA interleave index, for use only when MPOL_INTERLEAVE
  * @vmf: fault information
  *
  * Returns the struct page for entry and addr, after queueing swapin.
@@ -777,16 +787,17 @@ static void swap_ra_info(struct vm_fault *vmf,
  * Caller must hold read mmap_lock if vmf->vma is not NULL.
  *
  */
-static struct page *swap_vma_readahead(swp_entry_t fentry, gfp_t gfp_mask,
+static struct page *swap_vma_readahead(swp_entry_t targ_entry, gfp_t gfp_mask,
+				       struct mempolicy *mpol, pgoff_t targ_ilx,
 				       struct vm_fault *vmf)
 {
 	struct blk_plug plug;
 	struct swap_iocb *splug = NULL;
-	struct vm_area_struct *vma = vmf->vma;
 	struct page *page;
 	pte_t *pte = NULL, pentry;
 	unsigned long addr;
 	swp_entry_t entry;
+	pgoff_t ilx;
 	unsigned int i;
 	bool page_allocated;
 	struct vma_swap_readahead ra_info = {
@@ -798,9 +809,10 @@ static struct page *swap_vma_readahead(swp_entry_t fentry, gfp_t gfp_mask,
 		goto skip;
 
 	addr = vmf->address - (ra_info.offset * PAGE_SIZE);
+	ilx = targ_ilx - ra_info.offset;
 
 	blk_start_plug(&plug);
-	for (i = 0; i < ra_info.nr_pte; i++, addr += PAGE_SIZE) {
+	for (i = 0; i < ra_info.nr_pte; i++, ilx++, addr += PAGE_SIZE) {
 		if (!pte++) {
 			pte = pte_offset_map(vmf->pmd, addr);
 			if (!pte)
@@ -814,8 +826,8 @@ static struct page *swap_vma_readahead(swp_entry_t fentry, gfp_t gfp_mask,
 			continue;
 		pte_unmap(pte);
 		pte = NULL;
-		page = __read_swap_cache_async(entry, gfp_mask, vma,
-					       addr, &page_allocated);
+		page = __read_swap_cache_async(entry, gfp_mask, mpol, ilx,
+						&page_allocated);
 		if (!page)
 			continue;
 		if (page_allocated) {
@@ -834,8 +846,11 @@ static struct page *swap_vma_readahead(swp_entry_t fentry, gfp_t gfp_mask,
 	lru_add_drain();
 skip:
 	/* The page was likely read above, so no need for plugging here */
-	return read_swap_cache_async(fentry, gfp_mask, vma, vmf->address,
-				     NULL);
+	page = __read_swap_cache_async(targ_entry, gfp_mask, mpol, targ_ilx,
+					&page_allocated);
+	if (page_allocated)
+		swap_readpage(page, false, NULL);
+	return page;
 }
 
 /**
@@ -853,9 +868,16 @@ static struct page *swap_vma_readahead(swp_entry_t fentry, gfp_t gfp_mask,
 struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
 				struct vm_fault *vmf)
 {
-	return swap_use_vma_readahead() ?
-			swap_vma_readahead(entry, gfp_mask, vmf) :
-			swap_cluster_readahead(entry, gfp_mask, vmf);
+	struct mempolicy *mpol;
+	pgoff_t ilx;
+	struct page *page;
+
+	mpol = get_vma_policy(vmf->vma, vmf->address, 0, &ilx);
+	page = swap_use_vma_readahead() ?
+		swap_vma_readahead(entry, gfp_mask, mpol, ilx, vmf) :
+		swap_cluster_readahead(entry, gfp_mask, mpol, ilx);
+	mpol_cond_put(mpol);
+	return page;
 }
 
 #ifdef CONFIG_SYSFS
-- 
2.35.3

[PATCH 11/12] mempolicy: mmap_lock is not needed while migrating folios

[Hugh Dickins]

mbind(2) holds down_write of current task's mmap_lock throughout
(exclusive because it needs to set the new mempolicy on the vmas);
migrate_pages(2) holds down_read of pid's mmap_lock throughout.

They both hold mmap_lock across the internal migrate_pages(), under which
all new page allocations (huge or small) are made.  I'm nervous about it;
and migrate_pages() certainly does not need mmap_lock itself.  It's done
this way for mbind(2), because its page allocator is vma_alloc_folio() or
alloc_hugetlb_folio_vma(), both of which depend on vma and address.

Now that we have alloc_pages_mpol(), depending on (refcounted) memory
policy and interleave index, mbind(2) can be modified to use that or
alloc_hugetlb_folio_nodemask(), and then not need mmap_lock across the
internal migrate_pages() at all: add alloc_migration_target_by_mpol()
to replace mbind's new_page().

(After that change, alloc_hugetlb_folio_vma() is used by nothing but a
userfaultfd function: move it out of hugetlb.h and into the #ifdef.)

migrate_pages(2) has chosen its target node before migrating, so can
continue to use the standard alloc_migration_target(); but let it take
and drop mmap_lock just around migrate_to_node()'s queue_pages_range():
neither the node-to-node calculations nor the page migrations need it.

It seems unlikely, but it is conceivable that some userspace depends on
the kernel's mmap_lock exclusion here, instead of doing its own locking:
more likely in a testsuite than in real life.  It is also possible, of
course, that some pages on the list will be munmapped by another thread
before they are migrated, or a newer memory policy applied to the range
by that time: but such races could happen before, as soon as mmap_lock
was dropped, so it does not appear to be a concern.

Signed-off-by: Hugh Dickins <hughd@google.com>
---
 include/linux/hugetlb.h |  9 -----
 mm/hugetlb.c            | 38 +++++++++---------
 mm/mempolicy.c          | 85 +++++++++++++++++++++--------------------
 3 files changed, 64 insertions(+), 68 deletions(-)

diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
index 6522eb3cd007..9c4265c73f76 100644
--- a/include/linux/hugetlb.h
+++ b/include/linux/hugetlb.h
@@ -714,8 +714,6 @@ struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
 				unsigned long addr, int avoid_reserve);
 struct folio *alloc_hugetlb_folio_nodemask(struct hstate *h, int preferred_nid,
 				nodemask_t *nmask, gfp_t gfp_mask);
-struct folio *alloc_hugetlb_folio_vma(struct hstate *h, struct vm_area_struct *vma,
-				unsigned long address);
 int hugetlb_add_to_page_cache(struct folio *folio, struct address_space *mapping,
 			pgoff_t idx);
 void restore_reserve_on_error(struct hstate *h, struct vm_area_struct *vma,
@@ -1024,13 +1022,6 @@ alloc_hugetlb_folio_nodemask(struct hstate *h, int preferred_nid,
 	return NULL;
 }
 
-static inline struct folio *alloc_hugetlb_folio_vma(struct hstate *h,
-					       struct vm_area_struct *vma,
-					       unsigned long address)
-{
-	return NULL;
-}
-
 static inline int __alloc_bootmem_huge_page(struct hstate *h)
 {
 	return 0;
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index ba6d39b71cb1..1af54dbbd7cc 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -2479,24 +2479,6 @@ struct folio *alloc_hugetlb_folio_nodemask(struct hstate *h, int preferred_nid,
 	return alloc_migrate_hugetlb_folio(h, gfp_mask, preferred_nid, nmask);
 }
 
-/* mempolicy aware migration callback */
-struct folio *alloc_hugetlb_folio_vma(struct hstate *h, struct vm_area_struct *vma,
-		unsigned long address)
-{
-	struct mempolicy *mpol;
-	nodemask_t *nodemask;
-	struct folio *folio;
-	gfp_t gfp_mask;
-	int node;
-
-	gfp_mask = htlb_alloc_mask(h);
-	node = huge_node(vma, address, gfp_mask, &mpol, &nodemask);
-	folio = alloc_hugetlb_folio_nodemask(h, node, nodemask, gfp_mask);
-	mpol_cond_put(mpol);
-
-	return folio;
-}
-
 /*
  * Increase the hugetlb pool such that it can accommodate a reservation
  * of size 'delta'.
@@ -6225,6 +6207,26 @@ vm_fault_t hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 }
 
 #ifdef CONFIG_USERFAULTFD
+/*
+ * Can probably be eliminated, but still used by hugetlb_mfill_atomic_pte().
+ */
+static struct folio *alloc_hugetlb_folio_vma(struct hstate *h,
+		struct vm_area_struct *vma, unsigned long address)
+{
+	struct mempolicy *mpol;
+	nodemask_t *nodemask;
+	struct folio *folio;
+	gfp_t gfp_mask;
+	int node;
+
+	gfp_mask = htlb_alloc_mask(h);
+	node = huge_node(vma, address, gfp_mask, &mpol, &nodemask);
+	folio = alloc_hugetlb_folio_nodemask(h, node, nodemask, gfp_mask);
+	mpol_cond_put(mpol);
+
+	return folio;
+}
+
 /*
  * Used by userfaultfd UFFDIO_* ioctls. Based on userfaultfd's mfill_atomic_pte
  * with modifications for hugetlb pages.
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index d74df1e1b14a..74b1894d29c1 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -417,6 +417,8 @@ static const struct mempolicy_operations mpol_ops[MPOL_MAX] = {
 
 static bool migrate_folio_add(struct folio *folio, struct list_head *foliolist,
 				unsigned long flags);
+static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *pol,
+				pgoff_t ilx, int *nid);
 
 static bool strictly_unmovable(unsigned long flags)
 {
@@ -1040,6 +1042,8 @@ static long migrate_to_node(struct mm_struct *mm, int source, int dest,
 	node_set(source, nmask);
 
 	VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)));
+
+	mmap_read_lock(mm);
 	vma = find_vma(mm, 0);
 
 	/*
@@ -1050,6 +1054,7 @@ static long migrate_to_node(struct mm_struct *mm, int source, int dest,
 	 */
 	nr_failed = queue_pages_range(mm, vma->vm_start, mm->task_size, &nmask,
 				      flags | MPOL_MF_DISCONTIG_OK, &pagelist);
+	mmap_read_unlock(mm);
 
 	if (!list_empty(&pagelist)) {
 		err = migrate_pages(&pagelist, alloc_migration_target, NULL,
@@ -1078,8 +1083,6 @@ int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
 
 	lru_cache_disable();
 
-	mmap_read_lock(mm);
-
 	/*
 	 * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
 	 * bit in 'to' is not also set in 'tmp'.  Clear the found 'source'
@@ -1159,7 +1162,6 @@ int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
 		if (err < 0)
 			break;
 	}
-	mmap_read_unlock(mm);
 
 	lru_cache_enable();
 	if (err < 0)
@@ -1168,44 +1170,38 @@ int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
 }
 
 /*
- * Allocate a new page for page migration based on vma policy.
- * Start by assuming the page is mapped by the same vma as contains @start.
- * Search forward from there, if not.  N.B., this assumes that the
- * list of pages handed to migrate_pages()--which is how we get here--
- * is in virtual address order.
+ * Allocate a new folio for page migration, according to NUMA mempolicy.
  */
-static struct folio *new_folio(struct folio *src, unsigned long start)
+static struct folio *alloc_migration_target_by_mpol(struct folio *src,
+						    unsigned long private)
 {
-	struct vm_area_struct *vma;
-	unsigned long address;
-	VMA_ITERATOR(vmi, current->mm, start);
-	gfp_t gfp = GFP_HIGHUSER_MOVABLE | __GFP_RETRY_MAYFAIL;
-
-	for_each_vma(vmi, vma) {
-		address = page_address_in_vma(&src->page, vma);
-		if (address != -EFAULT)
-			break;
-	}
-
-	/*
-	 * __get_vma_policy() now expects a genuine non-NULL vma. Return NULL
-	 * when the page can no longer be located in a vma: that is not ideal
-	 * (migrate_pages() will give up early, presuming ENOMEM), but good
-	 * enough to avoid a crash by syzkaller or concurrent holepunch.
-	 */
-	if (!vma)
-		return NULL;
+	struct mempolicy *pol = (struct mempolicy *)private;
+	pgoff_t ilx = 0;	/* improve on this later */
+	struct page *page;
+	unsigned int order;
+	int nid = numa_node_id();
+	gfp_t gfp;
 
 	if (folio_test_hugetlb(src)) {
-		return alloc_hugetlb_folio_vma(folio_hstate(src),
-				vma, address);
+		nodemask_t *nodemask;
+		struct hstate *h;
+
+		ilx += src->index;	/* HugeTLBfs indexes in hpage_size */
+		h = folio_hstate(src);
+		gfp = htlb_alloc_mask(h);
+		nodemask = policy_nodemask(gfp, pol, ilx, &nid);
+		return alloc_hugetlb_folio_nodemask(h, nid, nodemask, gfp);
 	}
 
 	if (folio_test_large(src))
 		gfp = GFP_TRANSHUGE;
+	else
+		gfp = GFP_HIGHUSER_MOVABLE | __GFP_RETRY_MAYFAIL | __GFP_COMP;
 
-	return vma_alloc_folio(gfp, folio_order(src), vma, address,
-			folio_test_large(src));
+	order = folio_order(src);
+	ilx += src->index >> order;
+	page = alloc_pages_mpol(gfp, order, pol, ilx, nid);
+	return page_rmappable_folio(page);
 }
 #else
 
@@ -1221,7 +1217,8 @@ int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
 	return -ENOSYS;
 }
 
-static struct folio *new_folio(struct folio *src, unsigned long start)
+static struct folio *alloc_migration_target_by_mpol(struct folio *src,
+						    unsigned long private)
 {
 	return NULL;
 }
@@ -1295,6 +1292,7 @@ static long do_mbind(unsigned long start, unsigned long len,
 
 	if (nr_failed < 0) {
 		err = nr_failed;
+		nr_failed = 0;
 	} else {
 		vma_iter_init(&vmi, mm, start);
 		prev = vma_prev(&vmi);
@@ -1305,19 +1303,24 @@ static long do_mbind(unsigned long start, unsigned long len,
 		}
 	}
 
-	if (!err) {
-		if (!list_empty(&pagelist)) {
-			nr_failed |= migrate_pages(&pagelist, new_folio, NULL,
-				start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND, NULL);
+	mmap_write_unlock(mm);
+
+	if (!err && !list_empty(&pagelist)) {
+		/* Convert MPOL_DEFAULT's NULL to task or default policy */
+		if (!new) {
+			new = get_task_policy(current);
+			mpol_get(new);
 		}
-		if (nr_failed && (flags & MPOL_MF_STRICT))
-			err = -EIO;
+		nr_failed |= migrate_pages(&pagelist,
+				alloc_migration_target_by_mpol, NULL,
+				(unsigned long)new, MIGRATE_SYNC,
+				MR_MEMPOLICY_MBIND, NULL);
 	}
 
+	if (nr_failed && (flags & MPOL_MF_STRICT))
+		err = -EIO;
 	if (!list_empty(&pagelist))
 		putback_movable_pages(&pagelist);
-
-	mmap_write_unlock(mm);
 mpol_out:
 	mpol_put(new);
 	if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
-- 
2.35.3

[PATCH 12/12] mempolicy: migration attempt to match interleave nodes

[Hugh Dickins]

Improve alloc_migration_target_by_mpol()'s treatment of MPOL_INTERLEAVE.

Make an effort in do_mbind(), to identify the correct interleave index
for the first page to be migrated, so that it and all subsequent pages
from the same vma will be targeted to precisely their intended nodes.
Pages from following vmas will still be interleaved from the requested
nodemask, but perhaps starting from a different base.

Whether this is worth doing at all, or worth improving further, is
arguable: queue_folio_required() is right not to care about the precise
placement on interleaved nodes; but this little effort seems appropriate.

Signed-off-by: Hugh Dickins <hughd@google.com>
---
 mm/mempolicy.c | 52 +++++++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 49 insertions(+), 3 deletions(-)

diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 74b1894d29c1..7bb9ff69879b 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -430,6 +430,11 @@ static bool strictly_unmovable(unsigned long flags)
 			 MPOL_MF_STRICT;
 }
 
+struct migration_mpol {		/* for alloc_migration_target_by_mpol() */
+	struct mempolicy *pol;
+	pgoff_t ilx;
+};
+
 struct queue_pages {
 	struct list_head *pagelist;
 	unsigned long flags;
@@ -1175,8 +1180,9 @@ int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
 static struct folio *alloc_migration_target_by_mpol(struct folio *src,
 						    unsigned long private)
 {
-	struct mempolicy *pol = (struct mempolicy *)private;
-	pgoff_t ilx = 0;	/* improve on this later */
+	struct migration_mpol *mmpol = (struct migration_mpol *)private;
+	struct mempolicy *pol = mmpol->pol;
+	pgoff_t ilx = mmpol->ilx;
 	struct page *page;
 	unsigned int order;
 	int nid = numa_node_id();
@@ -1231,6 +1237,7 @@ static long do_mbind(unsigned long start, unsigned long len,
 	struct mm_struct *mm = current->mm;
 	struct vm_area_struct *vma, *prev;
 	struct vma_iterator vmi;
+	struct migration_mpol mmpol;
 	struct mempolicy *new;
 	unsigned long end;
 	long err;
@@ -1311,9 +1318,48 @@ static long do_mbind(unsigned long start, unsigned long len,
 			new = get_task_policy(current);
 			mpol_get(new);
 		}
+		mmpol.pol = new;
+		mmpol.ilx = 0;
+
+		/*
+		 * In the interleaved case, attempt to allocate on exactly the
+		 * targeted nodes, for the first VMA to be migrated; for later
+		 * VMAs, the nodes will still be interleaved from the targeted
+		 * nodemask, but one by one may be selected differently.
+		 */
+		if (new->mode == MPOL_INTERLEAVE) {
+			struct page *page;
+			unsigned int order;
+			unsigned long addr = -EFAULT;
+
+			list_for_each_entry(page, &pagelist, lru) {
+				if (!PageKsm(page))
+					break;
+			}
+			if (!list_entry_is_head(page, &pagelist, lru)) {
+				vma_iter_init(&vmi, mm, start);
+				for_each_vma_range(vmi, vma, end) {
+					addr = page_address_in_vma(page, vma);
+					if (addr != -EFAULT)
+						break;
+				}
+			}
+			if (addr != -EFAULT) {
+				order = compound_order(page);
+				/* We already know the pol, but not the ilx */
+				mpol_cond_put(get_vma_policy(vma, addr, order,
+							     &mmpol.ilx));
+				/* HugeTLBfs indexes in hpage_size */
+				if (order && PageHuge(page))
+					order = 0;
+				/* Set base from which to increment by index */
+				mmpol.ilx -= page->index >> order;
+			}
+		}
+
 		nr_failed |= migrate_pages(&pagelist,
 				alloc_migration_target_by_mpol, NULL,
-				(unsigned long)new, MIGRATE_SYNC,
+				(unsigned long)&mmpol, MIGRATE_SYNC,
 				MR_MEMPOLICY_MBIND, NULL);
 	}
 
-- 
2.35.3