Linux: Report reclaimable memory to kernel as such (#16385)

Linux provides SLAB_RECLAIM_ACCOUNT and __GFP_RECLAIMABLE flags to
mark memory allocations that can be freed via shinker calls.  It
should allow kernel to tune and group such allocations for lower
memory fragmentation and better reclamation under pressure.

This patch marks as reclaimable most of ARC memory, directly
evictable via ZFS shrinker, plus also dnode/znode/sa memory,
indirectly evictable via kernel's superblock shrinker.

Signed-off-by:	Alexander Motin <mav@FreeBSD.org>
Sponsored by:	iXsystems, Inc.
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Allan Jude <allan@klarasystems.com>

include/os/freebsd/spl/sys/kmem.h

@@ -49,6 +49,7 @@ MALLOC_DECLARE(M_SOLARIS);
 #define	KM_NOSLEEP		M_NOWAIT
 #define	KM_NORMALPRI		0
 #define	KMC_NODEBUG		UMA_ZONE_NODUMP
+#define	KMC_RECLAIMABLE		0x0
 
 typedef struct vmem vmem_t;
 

include/os/linux/spl/sys/kmem_cache.h

@@ -45,6 +45,7 @@ typedef enum kmc_bit {
 	KMC_BIT_TOTAL		= 18,	/* Proc handler helper bit */
 	KMC_BIT_ALLOC		= 19,	/* Proc handler helper bit */
 	KMC_BIT_MAX		= 20,	/* Proc handler helper bit */
+	KMC_BIT_RECLAIMABLE	= 21,	/* Can be freed by shrinker */
 } kmc_bit_t;
 
 /* kmem move callback return values */
@@ -66,9 +67,7 @@ typedef enum kmem_cbrc {
 #define	KMC_TOTAL		(1 << KMC_BIT_TOTAL)
 #define	KMC_ALLOC		(1 << KMC_BIT_ALLOC)
 #define	KMC_MAX			(1 << KMC_BIT_MAX)
-
+#define	KMC_RECLAIMABLE		(1 << KMC_BIT_RECLAIMABLE)
-#define	KMC_REAP_CHUNK		INT_MAX
-#define	KMC_DEFAULT_SEEKS	1
 
 extern struct list_head spl_kmem_cache_list;
 extern struct rw_semaphore spl_kmem_cache_sem;

include/sys/zfs_context.h

@@ -413,6 +413,7 @@ void procfs_list_add(procfs_list_t *procfs_list, void *p);
 #define	KM_NORMALPRI		0	/* not needed with UMEM_DEFAULT */
 #define	KMC_NODEBUG		UMC_NODEBUG
 #define	KMC_KVMEM		0x0
+#define	KMC_RECLAIMABLE		0x0
 #define	kmem_alloc(_s, _f)	umem_alloc(_s, _f)
 #define	kmem_zalloc(_s, _f)	umem_zalloc(_s, _f)
 #define	kmem_free(_b, _s)	umem_free(_b, _s)

module/os/freebsd/zfs/abd_os.c

@@ -300,7 +300,7 @@ void
 abd_init(void)
 {
 	abd_chunk_cache = kmem_cache_create("abd_chunk", PAGE_SIZE, 0,
-	    NULL, NULL, NULL, NULL, 0, KMC_NODEBUG);
+	    NULL, NULL, NULL, NULL, 0, KMC_NODEBUG | KMC_RECLAIMABLE);
 
 	wmsum_init(&abd_sums.abdstat_struct_size, 0);
 	wmsum_init(&abd_sums.abdstat_scatter_cnt, 0);

module/os/freebsd/zfs/zfs_znode.c

@@ -236,7 +236,7 @@ zfs_znode_init(void)
 	ASSERT3P(znode_cache, ==, NULL);
 	znode_cache = kmem_cache_create("zfs_znode_cache",
 	    sizeof (znode_t), 0, zfs_znode_cache_constructor,
-	    zfs_znode_cache_destructor, NULL, NULL, NULL, 0);
+	    zfs_znode_cache_destructor, NULL, NULL, NULL, KMC_RECLAIMABLE);
 }
 
 static znode_t *

module/os/linux/spl/spl-kmem-cache.c

@@ -144,6 +144,8 @@ kv_alloc(spl_kmem_cache_t *skc, int size, int flags)
 	gfp_t lflags = kmem_flags_convert(flags);
 	void *ptr;
 
+	if (skc->skc_flags & KMC_RECLAIMABLE)
+		lflags |= __GFP_RECLAIMABLE;
 	ptr = spl_vmalloc(size, lflags | __GFP_HIGHMEM);
 
 	/* Resulting allocated memory will be page aligned */
@@ -424,6 +426,8 @@ spl_emergency_alloc(spl_kmem_cache_t *skc, int flags, void **obj)
 	if (!empty)
 		return (-EEXIST);
 
+	if (skc->skc_flags & KMC_RECLAIMABLE)
+		lflags |= __GFP_RECLAIMABLE;
 	ske = kmalloc(sizeof (*ske), lflags);
 	if (ske == NULL)
 		return (-ENOMEM);
@@ -663,6 +667,7 @@ spl_magazine_destroy(spl_kmem_cache_t *skc)
  *	KMC_KVMEM       Force kvmem backed SPL cache
  *	KMC_SLAB        Force Linux slab backed cache
  *	KMC_NODEBUG	Disable debugging (unsupported)
+ *	KMC_RECLAIMABLE	Memory can be freed under pressure
  */
 spl_kmem_cache_t *
 spl_kmem_cache_create(const char *name, size_t size, size_t align,
@@ -780,6 +785,9 @@ spl_kmem_cache_create(const char *name, size_t size, size_t align,
 		if (size > spl_kmem_cache_slab_limit)
 			goto out;
 
+		if (skc->skc_flags & KMC_RECLAIMABLE)
+			slabflags |= SLAB_RECLAIM_ACCOUNT;
+
 #if defined(SLAB_USERCOPY)
 		/*
 		 * Required for PAX-enabled kernels if the slab is to be

module/os/linux/spl/spl-zlib.c

@@ -202,7 +202,7 @@ spl_zlib_init(void)
 	zlib_workspace_cache = kmem_cache_create(
 	    "spl_zlib_workspace_cache",
 	    size, 0, NULL, NULL, NULL, NULL, NULL,
-	    KMC_KVMEM);
+	    KMC_KVMEM | KMC_RECLAIMABLE);
 	if (!zlib_workspace_cache)
 		return (-ENOMEM);
 

module/os/linux/zfs/abd_os.c

@@ -281,7 +281,7 @@ abd_alloc_chunks(abd_t *abd, size_t size)
 	struct sg_table table;
 	struct scatterlist *sg;
 	struct page *page, *tmp_page = NULL;
-	gfp_t gfp = __GFP_NOWARN | GFP_NOIO;
+	gfp_t gfp = __GFP_RECLAIMABLE | __GFP_NOWARN | GFP_NOIO;
 	gfp_t gfp_comp = (gfp | __GFP_NORETRY | __GFP_COMP) & ~__GFP_RECLAIM;
 	unsigned int max_order = MIN(zfs_abd_scatter_max_order,
 	    ABD_MAX_ORDER - 1);
@@ -403,7 +403,7 @@ abd_alloc_chunks(abd_t *abd, size_t size)
 	struct scatterlist *sg = NULL;
 	struct sg_table table;
 	struct page *page;
-	gfp_t gfp = __GFP_NOWARN | GFP_NOIO;
+	gfp_t gfp = __GFP_RECLAIMABLE | __GFP_NOWARN | GFP_NOIO;
 	int nr_pages = abd_chunkcnt_for_bytes(size);
 	int i = 0;
 
@@ -762,7 +762,7 @@ abd_init(void)
 	int i;
 
 	abd_cache = kmem_cache_create("abd_t", sizeof (abd_t),
-	    0, NULL, NULL, NULL, NULL, NULL, 0);
+	    0, NULL, NULL, NULL, NULL, NULL, KMC_RECLAIMABLE);
 
 	wmsum_init(&abd_sums.abdstat_struct_size, 0);
 	wmsum_init(&abd_sums.abdstat_linear_cnt, 0);

module/os/linux/zfs/zfs_znode.c

@@ -194,7 +194,8 @@ zfs_znode_init(void)
 	ASSERT(znode_cache == NULL);
 	znode_cache = kmem_cache_create("zfs_znode_cache",
 	    sizeof (znode_t), 0, zfs_znode_cache_constructor,
-	    zfs_znode_cache_destructor, NULL, NULL, NULL, KMC_SLAB);
+	    zfs_znode_cache_destructor, NULL, NULL, NULL,
+	    KMC_SLAB | KMC_RECLAIMABLE);
 
 	ASSERT(znode_hold_cache == NULL);
 	znode_hold_cache = kmem_cache_create("zfs_znode_hold_cache",

module/zfs/arc.c

@@ -1258,7 +1258,7 @@ retry:
 	}
 
 	hdr_full_cache = kmem_cache_create("arc_buf_hdr_t_full", HDR_FULL_SIZE,
-	    0, hdr_full_cons, hdr_full_dest, NULL, NULL, NULL, 0);
+	    0, hdr_full_cons, hdr_full_dest, NULL, NULL, NULL, KMC_RECLAIMABLE);
 	hdr_l2only_cache = kmem_cache_create("arc_buf_hdr_t_l2only",
 	    HDR_L2ONLY_SIZE, 0, hdr_l2only_cons, hdr_l2only_dest, NULL,
 	    NULL, NULL, 0);

module/zfs/dnode.c

@@ -306,7 +306,7 @@ dnode_init(void)
 {
 	ASSERT(dnode_cache == NULL);
 	dnode_cache = kmem_cache_create("dnode_t", sizeof (dnode_t),
-	    0, dnode_cons, dnode_dest, NULL, NULL, NULL, 0);
+	    0, dnode_cons, dnode_dest, NULL, NULL, NULL, KMC_RECLAIMABLE);
 	kmem_cache_set_move(dnode_cache, dnode_move);
 
 	wmsum_init(&dnode_sums.dnode_hold_dbuf_hold, 0);

module/zfs/lz4_zfs.c

@@ -867,7 +867,8 @@ void
 lz4_init(void)
 {
 	lz4_cache = kmem_cache_create("lz4_cache",
-	    sizeof (struct refTables), 0, NULL, NULL, NULL, NULL, NULL, 0);
+	    sizeof (struct refTables), 0, NULL, NULL, NULL, NULL, NULL,
+	    KMC_RECLAIMABLE);
 }
 
 void

module/zfs/sa.c

@@ -236,7 +236,7 @@ sa_cache_init(void)
 {
 	sa_cache = kmem_cache_create("sa_cache",
 	    sizeof (sa_handle_t), 0, sa_cache_constructor,
-	    sa_cache_destructor, NULL, NULL, NULL, 0);
+	    sa_cache_destructor, NULL, NULL, NULL, KMC_RECLAIMABLE);
 }
 
 void

module/zfs/zio.c

@@ -194,6 +194,10 @@ zio_init(void)
 		cflags = (zio_exclude_metadata || size > zio_buf_debug_limit) ?
 		    KMC_NODEBUG : 0;
 		data_cflags = KMC_NODEBUG;
+		if (abd_size_alloc_linear(size)) {
+			cflags |= KMC_RECLAIMABLE;
+			data_cflags |= KMC_RECLAIMABLE;
+		}
 		if (cflags == data_cflags) {
 			/*
 			 * Resulting kmem caches would be identical.