MFC r258137:

Introduce allocation cache to store LZ4 compression contexts without kicking
VM subsystem twice for every written record.

Tests on 24-core system show double reduction of CPU time spent on copying
single large well-compressed file.

This patch is not really needed on illumos (while not harm either) since
their memory allocator by default uses caching for all requests up to 128K.

3 files changed, 28 insertions, 2 deletions

diff --git a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/lz4.c b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/lz4.c
index 0f55cc8..a7691d6 100644
--- a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/lz4.c
+++ b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/lz4.c
@@ -44,6 +44,8 @@ static int LZ4_compressCtx(void *ctx, const char *source, char *dest,
 static int LZ4_compress64kCtx(void *ctx, const char *source, char *dest,
     int isize, int osize);
 
+static kmem_cache_t *lz4_ctx_cache;
+
 /*ARGSUSED*/
 size_t
 lz4_compress(void *s_start, void *d_start, size_t s_len, size_t d_len, int n)
@@ -840,7 +842,7 @@ static int
 real_LZ4_compress(const char *source, char *dest, int isize, int osize)
 {
 #if HEAPMODE
-	void *ctx = kmem_zalloc(sizeof (struct refTables), KM_NOSLEEP);
+	void *ctx = kmem_cache_alloc(lz4_ctx_cache, KM_NOSLEEP);
 	int result;
 
 	/*
@@ -850,12 +852,13 @@ real_LZ4_compress(const char *source, char *dest, int isize, int osize)
 	if (ctx == NULL)
 		return (0);
 
+	bzero(ctx, sizeof(struct refTables));
 	if (isize < LZ4_64KLIMIT)
 		result = LZ4_compress64kCtx(ctx, source, dest, isize, osize);
 	else
 		result = LZ4_compressCtx(ctx, source, dest, isize, osize);
 
-	kmem_free(ctx, sizeof (struct refTables));
+	kmem_cache_free(lz4_ctx_cache, ctx);
 	return (result);
 #else
 	if (isize < (int)LZ4_64KLIMIT)
@@ -1001,3 +1004,22 @@ LZ4_uncompress_unknownOutputSize(const char *source, char *dest, int isize,
 	_output_error:
 	return (int)(-(((char *)ip) - source));
 }
+
+extern void
+lz4_init(void)
+{
+
+#if HEAPMODE
+	lz4_ctx_cache = kmem_cache_create("lz4_ctx", sizeof(struct refTables),
+	    0, NULL, NULL, NULL, NULL, NULL, 0);
+#endif
+}
+
+extern void
+lz4_fini(void)
+{
+
+#if HEAPMODE
+	kmem_cache_destroy(lz4_ctx_cache);
+#endif
+}
diff --git a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c
index d92bcf0..17983bf 100644
--- a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c
+++ b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c
@@ -1739,6 +1739,7 @@ spa_init(int mode)
 	unique_init();
 	space_map_init();
 	zio_init();
+	lz4_init();
 	dmu_init();
 	zil_init();
 	vdev_cache_stat_init();
@@ -1764,6 +1765,7 @@ spa_fini(void)
 	vdev_cache_stat_fini();
 	zil_fini();
 	dmu_fini();
+	lz4_fini();
 	zio_fini();
 	space_map_fini();
 	unique_fini();
diff --git a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zio_compress.h b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zio_compress.h
index 85b8a99..c0be314 100644
--- a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zio_compress.h
+++ b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zio_compress.h
@@ -70,6 +70,8 @@ extern size_t zle_compress(void *src, void *dst, size_t s_len, size_t d_len,
     int level);
 extern int zle_decompress(void *src, void *dst, size_t s_len, size_t d_len,
     int level);
+extern void lz4_init(void);
+extern void lz4_fini(void);
 extern size_t lz4_compress(void *src, void *dst, size_t s_len, size_t d_len,
     int level);
 extern int lz4_decompress(void *src, void *dst, size_t s_len, size_t d_len,