[v2] mempolicy: alloc_pages_mpol() for NUMA policy without vma: fix

Commit Message

Hugh Dickins Oct. 24, 2023, 4:09 p.m. UTC

mm-unstable commit 48a7bd12d57f ("mempolicy: alloc_pages_mpol() for NUMA
policy without vma") ended read_swap_cache_async() supporting NULL vma -
okay; but missed the NULL mpol being passed to __read_swap_cache_async()
by zswap_writeback_entry() - oops!

Since its other callers all give good mpol, add get_task_policy(current)
there in mm/zswap.c, to produce the same good-enough behaviour as before
(and task policy, acted on in current task, does not require the refcount
to be dup'ed).

But if that policy is (quite reasonably) MPOL_INTERLEAVE, then ilx must
be NO_INTERLEAVE_INDEX rather than 0, to provide the same distribution
as before: move that definition from mempolicy.c to mempolicy.h.

Reported-by: Domenico Cerasuolo <[e-mail]>
Closes: https://lore.kernel.org/linux-mm/[e-mail]/T/#mf08c877d1884fc7867f9e328cdf02257ff3b3ae9
Suggested-by: Johannes Weiner <[e-mail]>
Fixes: 48a7bd12d57f ("mempolicy: alloc_pages_mpol() for NUMA policy without vma")
Signed-off-by: Hugh Dickins <[e-mail]>
---
v2: !CONFIG_NUMA builds with a get_task_policy() added in mempolicy.h

 include/linux/mempolicy.h | 7 +++++++
 mm/mempolicy.c            | 2 --
 mm/zswap.c                | 7 +++++--
 3 files changed, 12 insertions(+), 4 deletions(-)

Patch

diff --git a/include/linux/mempolicy.h b/include/linux/mempolicy.h
index 2801d5b0a4e9..931b118336f4 100644
--- a/include/linux/mempolicy.h
+++ b/include/linux/mempolicy.h
@@ -17,6 +17,8 @@ 
 
 struct mm_struct;
 
+#define NO_INTERLEAVE_INDEX (-1UL)	/* use task il_prev for interleaving */
+
 #ifdef CONFIG_NUMA
 
 /*
@@ -179,6 +181,11 @@  extern bool apply_policy_zone(struct mempolicy *policy, enum zone_type zone);
 
 struct mempolicy {};
 
+static inline struct mempolicy *get_task_policy(struct task_struct *p)
+{
+	return NULL;
+}
+
 static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b)
 {
 	return true;
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 898ee2e3c85b..989293180eb6 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -114,8 +114,6 @@ 
 #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;
 
diff --git a/mm/zswap.c b/mm/zswap.c
index 37d2b1cb2ecb..060857adca76 100644
--- a/mm/zswap.c
+++ b/mm/zswap.c
@@ -24,6 +24,7 @@ 
 #include <linux/swap.h>
 #include <linux/crypto.h>
 #include <linux/scatterlist.h>
+#include <linux/mempolicy.h>
 #include <linux/mempool.h>
 #include <linux/zpool.h>
 #include <crypto/acompress.h>
@@ -1057,6 +1058,7 @@  static int zswap_writeback_entry(struct zswap_entry *entry,
 {
 	swp_entry_t swpentry = entry->swpentry;
 	struct page *page;
+	struct mempolicy *mpol;
 	struct scatterlist input, output;
 	struct crypto_acomp_ctx *acomp_ctx;
 	struct zpool *pool = zswap_find_zpool(entry);
@@ -1075,8 +1077,9 @@  static int zswap_writeback_entry(struct zswap_entry *entry,
 	}
 
 	/* try to allocate swap cache page */
-	page = __read_swap_cache_async(swpentry, GFP_KERNEL, NULL, 0,
-				       &page_was_allocated);
+	mpol = get_task_policy(current);
+	page = __read_swap_cache_async(swpentry, GFP_KERNEL, mpol,
+				NO_INTERLEAVE_INDEX, &page_was_allocated);
 	if (!page) {
 		ret = -ENOMEM;
 		goto fail;