diff mbox series

[RFC,2/3] mm: hugetlb: Refactor vma_has_reserves() to should_use_hstate_resv()

Message ID 3d1946d01f63104de913c0979b5a596e2add1672.1728684491.git.ackerleytng@google.com (mailing list archive)
State New
Headers show
Series Reduce dependence on vmas deep in hugetlb allocation code | expand

Commit Message

Ackerley Tng Oct. 11, 2024, 11:22 p.m. UTC 
With the addition of the chg parameter, vma_has_reserves() no longer
just determines whether the vma has reserves.

The comment in the vma->vm_flags & VM_NORESERVE block indicates that
this function actually computes whether or not the reserved count
should be decremented.

This refactoring also takes into account the allocation's request
parameter avoid_reserve, which helps to further simplify the calling
function alloc_hugetlb_folio().

Signed-off-by: Ackerley Tng <ackerleytng@google.com>
---
 mm/hugetlb.c | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)

--
2.47.0.rc1.288.g06298d1525-goog

Comments

Peter Xu Nov. 5, 2024, 6:46 p.m. UTC | #1 
On Fri, Oct 11, 2024 at 11:22:37PM +0000, Ackerley Tng wrote:
> With the addition of the chg parameter, vma_has_reserves() no longer
> just determines whether the vma has reserves.
> 
> The comment in the vma->vm_flags & VM_NORESERVE block indicates that
> this function actually computes whether or not the reserved count
> should be decremented.
> 
> This refactoring also takes into account the allocation's request
> parameter avoid_reserve, which helps to further simplify the calling
> function alloc_hugetlb_folio().
> 
> Signed-off-by: Ackerley Tng <ackerleytng@google.com>

I wonder if this patch could be merged with the 3rd, IIUC this can
fundamentally be seen as a movement of what patch 3 was removed.

Furthermore, I do feel like should_use_hstate_resv() could be redundant on
its own on many things.

Let me try to justify.  Firstly, after 3 patches applied, now it looks like
this (I removed all comments to make things shorter..):

static bool should_use_hstate_resv(struct vm_area_struct *vma, long chg,
				   bool avoid_reserve)
{
	if (avoid_reserve)
		return false;

	if (vma->vm_flags & VM_NORESERVE) {
		if (vma->vm_flags & VM_MAYSHARE && chg == 0)
			return true;
		else
			return false;
	}

	if (vma->vm_flags & VM_MAYSHARE) {
		if (chg)
			return false;
		else
			return true;
	}

	if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) {
		if (chg)
			return false;
		else
			return true;
	}

	return false;
}

Then let's look at chg==0 processing all above: what does it say?  I
suppose it means "we don't need another global reservation".  It means if
chg==0 we always will use an existing reservation.  From math POV it also
is true, so it can already be moved out ahead, IIUC, like this:

static bool should_use_hstate_resv(struct vm_area_struct *vma, long chg,
				   bool avoid_reserve)
{
	if (avoid_reserve)
		return false;

        if (chg == 0)
                return true;

	if (vma->vm_flags & VM_NORESERVE)
                return false;

	if (vma->vm_flags & VM_MAYSHARE)
                return false;

	if (is_vma_resv_set(vma, HPAGE_RESV_OWNER))
                return false;

	return false;             <--------------------- [1]
}

Move on.  If I read it right, above [1] is exactly for avoid_reserve==1
case, where it basically says "it's !NORESERVE, private, and it's not the
vma resv owner, either fork() or CoW".  If my reading is correct, it means
after your patch 2, [1] should never be reachable at all.. I would hope
adding a panic() right above would be ok.

And irrelevant of whether my understanding is correct.. math-wise above is
also already the same as:

static bool should_use_hstate_resv(struct vm_area_struct *vma, long chg,
				   bool avoid_reserve)
{
	if (avoid_reserve)
		return false;

        if (chg == 0)
                return true;

	return false;
}

Then it makes a lot more sense now, because afaict, gbl_chg is exactly what
should_use_hstate_resv() is looking for.. only except avoid_reserve==true.

Would above make sense to you?

In short, it's about whether a patch on top of your series would further
simply this whole thing, like below:

===8<===
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 60e72214d5bf..4b1c5c4ed7be 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -1245,80 +1245,6 @@ void clear_vma_resv_huge_pages(struct vm_area_struct *vma)
 	hugetlb_dup_vma_private(vma);
 }
 
-/*
- * Returns true if this allocation should use (debit) hstate reservations, based on
- *
- * @vma: VMA config
- * @chg: Whether the page requirement can be satisfied using subpool reservations
- * @avoid_reserve: Whether allocation was requested to avoid using reservations
- */
-static bool should_use_hstate_resv(struct vm_area_struct *vma, long chg,
-				   bool avoid_reserve)
-{
-	if (avoid_reserve)
-		return false;
-
-	if (vma->vm_flags & VM_NORESERVE) {
-		/*
-		 * This address is already reserved by other process(chg == 0),
-		 * so, we should decrement reserved count. Without decrementing,
-		 * reserve count remains after releasing inode, because this
-		 * allocated page will go into page cache and is regarded as
-		 * coming from reserved pool in releasing step.  Currently, we
-		 * don't have any other solution to deal with this situation
-		 * properly, so add work-around here.
-		 */
-		if (vma->vm_flags & VM_MAYSHARE && chg == 0)
-			return true;
-		else
-			return false;
-	}
-
-	/* Shared mappings always use reserves */
-	if (vma->vm_flags & VM_MAYSHARE) {
-		/*
-		 * We know VM_NORESERVE is not set.  Therefore, there SHOULD
-		 * be a region map for all pages.  The only situation where
-		 * there is no region map is if a hole was punched via
-		 * fallocate.  In this case, there really are no reserves to
-		 * use.  This situation is indicated if chg != 0.
-		 */
-		if (chg)
-			return false;
-		else
-			return true;
-	}
-
-	/*
-	 * Only the process that called mmap() has reserves for private
-	 * mappings. A child process with MAP_PRIVATE mappings created by their
-	 * parent have no page reserves.
-	 */
-	if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) {
-		/*
-		 * Like the shared case above, a hole punch or truncate
-		 * could have been performed on the private mapping.
-		 * Examine the value of chg to determine if reserves
-		 * actually exist or were previously consumed.
-		 * Very Subtle - The value of chg comes from a previous
-		 * call to vma_needs_reserves().  The reserve map for
-		 * private mappings has different (opposite) semantics
-		 * than that of shared mappings.  vma_needs_reserves()
-		 * has already taken this difference in semantics into
-		 * account.  Therefore, the meaning of chg is the same
-		 * as in the shared case above.  Code could easily be
-		 * combined, but keeping it separate draws attention to
-		 * subtle differences.
-		 */
-		if (chg)
-			return false;
-		else
-			return true;
-	}
-
-	return false;
-}
-
 static void enqueue_hugetlb_folio(struct hstate *h, struct folio *folio)
 {
 	int nid = folio_nid(folio);
@@ -3255,7 +3181,7 @@ struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
 
 	}
 
-	use_hstate_resv = should_use_hstate_resv(vma, gbl_chg, avoid_reserve);
+	use_hstate_resv = avoid_reserve || !gbl_chg;
 
 	/*
 	 * charge_cgroup_reservation if this allocation is not consuming a
===8<===

Thanks,
Oscar Salvador Nov. 11, 2024, 9:19 a.m. UTC | #2 
On Tue, Nov 05, 2024 at 01:46:39PM -0500, Peter Xu wrote:
> I wonder if this patch could be merged with the 3rd, IIUC this can
> fundamentally be seen as a movement of what patch 3 was removed.

I think it makes sense to merge it, yes.

> Furthermore, I do feel like should_use_hstate_resv() could be redundant on
> its own on many things.

...
 
> Then let's look at chg==0 processing all above: what does it say?  I
> suppose it means "we don't need another global reservation".  It means if
> chg==0 we always will use an existing reservation.  From math POV it also
> is true, so it can already be moved out ahead, IIUC, like this:
> 
> static bool should_use_hstate_resv(struct vm_area_struct *vma, long chg,
> 				   bool avoid_reserve)
> {
> 	if (avoid_reserve)
> 		return false;
> 
>         if (chg == 0)
>                 return true;
> 
> 	if (vma->vm_flags & VM_NORESERVE)
>                 return false;
> 
> 	if (vma->vm_flags & VM_MAYSHARE)
>                 return false;
> 
> 	if (is_vma_resv_set(vma, HPAGE_RESV_OWNER))
>                 return false;
> 
> 	return false;             <--------------------- [1]
> }
> 
> Move on.  If I read it right, above [1] is exactly for avoid_reserve==1
> case, where it basically says "it's !NORESERVE, private, and it's not the
> vma resv owner, either fork() or CoW".  If my reading is correct, it means
> after your patch 2, [1] should never be reachable at all.. I would hope
> adding a panic() right above would be ok.
> 
> And irrelevant of whether my understanding is correct.. math-wise above is
> also already the same as:
> 
> static bool should_use_hstate_resv(struct vm_area_struct *vma, long chg,
> 				   bool avoid_reserve)
> {
> 	if (avoid_reserve)
> 		return false;
> 
>         if (chg == 0)
>                 return true;
> 
> 	return false;
> }

I have been looking into this because hugetlb reservations always make
me uneasy, but I think you are right.

CoW and fork both set avoid_reserve to 1,

 copy_hugetlb_range
  ...
  alloc_hugetlb_folio(dst_vma, addr, 1)

 hugetlb_wp
  outside_reserve = 1
  alloc_hugetlb_folio(..., outside_reserve)

So I think you are right and this can be simplified.

I would not add a panic though, maybe some kind of warning (VM_DEBUG?).
diff mbox series

Patch

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 73165c670739..47c421eba112 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -1246,9 +1246,19 @@  void clear_vma_resv_huge_pages(struct vm_area_struct *vma)
 	hugetlb_dup_vma_private(vma);
 }

-/* Returns true if the VMA has associated reserve pages */
-static bool vma_has_reserves(struct vm_area_struct *vma, long chg)
+/*
+ * Returns true if this allocation should use (debit) hstate reservations, based on
+ *
+ * @vma: VMA config
+ * @chg: Whether the page requirement can be satisfied using subpool reservations
+ * @avoid_reserve: Whether allocation was requested to avoid using reservations
+ */
+static bool should_use_hstate_resv(struct vm_area_struct *vma, long chg,
+				   bool avoid_reserve)
 {
+	if (avoid_reserve)
+		return false;
+
 	if (vma->vm_flags & VM_NORESERVE) {
 		/*
 		 * This address is already reserved by other process(chg == 0),
@@ -3025,7 +3035,7 @@  struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
 	if (ret)
 		goto out_uncharge_cgroup_reservation;

-	use_hstate_resv = !avoid_reserve && vma_has_reserves(vma, gbl_chg);
+	use_hstate_resv = should_use_hstate_resv(vma, gbl_chg, avoid_reserve);

 	spin_lock_irq(&hugetlb_lock);
 	folio = dequeue_hugetlb_folio_vma(h, vma, addr, use_hstate_resv);