Re: [RFC PATCH 12/57] mm/collapse: allocate a destination per candidate

On Sun, Aug 16, 2026 at 11:45:24PM +0100, Kiryl Shutsemau wrote:

From: "Kiryl Shutsemau (Meta)" <kas@kernel.org>

Fill in the allocation, which happens on both sides of the freeze.

A destination is a folio of the candidate's order, charged to the memcg,
with the memcg's deferred-split list entry taken up front while sleeping
is still allowed: the PMD-order install would otherwise need one under
the pmd lock.

collapse_alloc() does all of that for one candidate with the gfp it is
handed, and counts nothing when it fails: what a miss means is up to the
caller.

collapse_provision() is the caller inside the window.  The sources are
frozen by then and a faulter on any of them is waiting, so it asks
without __GFP_DIRECT_RECLAIM: reclaim entered there would be paid for by
that faulter.

[...]

+/*
+ * Allocate one candidate's destination with @gfp: a folio of its order, charged,
+ * with the memcg's deferred-split list heads in place so the install cannot need
+ * to allocate under the pmd lock.  Those heads cost only the first collapse in a
+ * memcg.
+ *
+ * A failure counts nothing and changes nothing: what a miss means is the caller's
+ * policy.
+ */
+static enum scan_result collapse_alloc(struct mm_struct *mm,
+				       struct collapse_control *cc,
+				       struct collapse_candidate *cand,
+				       gfp_t gfp)
+{
+	struct folio *folio;
+
+	folio = __folio_alloc(gfp, cand->order, collapse_find_target_node(cc),
+			      &cc->alloc_nmask);
+	if (!folio)
+		return SCAN_ALLOC_HUGE_PAGE_FAIL;
+
+	if (unlikely(mem_cgroup_charge(folio, mm, gfp)) ||
+	    folio_memcg_alloc_deferred(folio)) {

One small nit: folio_memcg_alloc_deferred() passes GFP_KERNEL to the
deferred_split_lru allocation.

At the full series tip, collapse_round() invokes it after
collapse_freeze() and before collapse_putback(), while faults on the
source migration entries wait:

static void collapse_round(struct mm_struct *mm, unsigned long pmd_addr,
			   struct collapse_control *cc)
{
...
	collapse_freeze(vma, cc, pmd);
	collapse_provision(mm, cc);
...
	collapse_putback(vma, cc);
...
}

collapse_provision() strips direct reclaim before calling
collapse_alloc():

static void collapse_provision(struct mm_struct *mm,
			       struct collapse_control *cc)
{
	const gfp_t gfp = cc->policy.gfp & ~__GFP_DIRECT_RECLAIM;
...
		result = collapse_alloc(mm, cc, cand, gfp);
...
}

collapse_alloc() uses that gfp for the charge, then calls
folio_memcg_alloc_deferred():

static enum scan_result collapse_alloc(struct mm_struct *mm,
				       struct collapse_control *cc,
				       struct collapse_candidate *cand,
				       gfp_t gfp)
{
...
	if (unlikely(mem_cgroup_charge(folio, mm, gfp)) ||
	    folio_memcg_alloc_deferred(folio)) {
...
	}
...
}

The helper substitutes GFP_KERNEL:

int folio_memcg_alloc_deferred(struct folio *folio)
{
...
	return folio_memcg_list_lru_alloc(folio, &deferred_split_lru, GFP_KERNEL);
}

When this metadata has not been allocated for a memcg yet,
folio_memcg_list_lru_alloc() can take its blocking slow branch:

int folio_memcg_list_lru_alloc(struct folio *folio, struct list_lru *lru,
			       gfp_t gfp)
{
...
	/* Fast path when list_lru heads already exist */
	rcu_read_lock();
	memcg = folio_memcg(folio);
	res = memcg_list_lru_allocated(memcg, lru);
	rcu_read_unlock();
	if (likely(res))
		return 0;

	/* Allocation may block, pin the memcg */
	memcg = get_mem_cgroup_from_folio(folio);
	res = __memcg_list_lru_alloc(memcg, lru, gfp);
...
}

GFP_KERNEL includes __GFP_DIRECT_RECLAIM:

...
#define __GFP_DIRECT_RECLAIM	((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
#define __GFP_KSWAPD_RECLAIM	((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
#define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))

...
#define GFP_KERNEL	(__GFP_RECLAIM | __GFP_IO | __GFP_FS)

This only affects calls that find the metadata missing; after it has been
installed, later calls take the fast branch. Still, it bypasses the gfp
with direct reclaim stripped for the freeze window. Could
folio_memcg_alloc_deferred() take that gfp here?

[...]

Cheers, Lance