L2ARC: Rework write throttling with DWPD rate limiting and parallel writes

include/sys/arc.h

@@ -354,8 +354,6 @@ boolean_t l2arc_range_check_overlap(uint64_t bottom, uint64_t top,
     uint64_t check);
 void l2arc_init(void);
 void l2arc_fini(void);
-void l2arc_start(void);
-void l2arc_stop(void);
 void l2arc_spa_rebuild_start(spa_t *spa);
 void l2arc_spa_rebuild_stop(spa_t *spa);
 

include/sys/arc_impl.h

@@ -41,6 +41,27 @@
 extern "C" {
 #endif
 
+/*
+ * We can feed L2ARC from two states of ARC buffers, mru and mfu,
+ * and each of the state has two types: data and metadata.
+ */
+#define	L2ARC_FEED_TYPES	4
+
+/*
+ * L2ARC state and statistics for persistent marker management.
+ */
+typedef struct l2arc_info {
+	arc_buf_hdr_t	**l2arc_markers[L2ARC_FEED_TYPES];
+	uint64_t	l2arc_total_writes;	/* total writes for reset */
+	uint64_t	l2arc_total_capacity;	/* total L2ARC capacity */
+	uint64_t	l2arc_smallest_capacity; /* smallest device capacity */
+	/*
+	 * Per-device thread coordination for sublist processing
+	 */
+	boolean_t	*l2arc_sublist_busy[L2ARC_FEED_TYPES];
+	kmutex_t	l2arc_sublist_lock;	/* protects busy flags */
+} l2arc_info_t;
+
 /*
  * Note that buffers can be in one of 6 states:
  *	ARC_anon	- anonymous (discussed below)
@@ -421,6 +442,19 @@ typedef struct l2arc_dev {
 	 */
 	zfs_refcount_t		l2ad_lb_count;
 	boolean_t		l2ad_trim_all; /* TRIM whole device */
+	/*
+	 * DWPD tracking with daily reset
+	 */
+	uint64_t		l2ad_dwpd_writes;	/* 24h bytes written */
+	uint64_t		l2ad_dwpd_start;	/* 24h period start */
+	uint64_t		l2ad_dwpd_accumulated;	/* Accumulated */
+	/*
+	 * Per-device feed thread for parallel L2ARC writes
+	 */
+	kthread_t		*l2ad_feed_thread;	/* feed thread handle */
+	boolean_t		l2ad_thread_exit;	/* signal thread exit */
+	kmutex_t		l2ad_feed_thr_lock;	/* thread sleep/wake */
+	kcondvar_t		l2ad_feed_cv;		/* thread wakeup cv */
 } l2arc_dev_t;
 
 /*

include/sys/spa_impl.h

@@ -52,6 +52,7 @@
 #include <sys/dsl_crypt.h>
 #include <sys/zfeature.h>
 #include <sys/zthr.h>
+#include <sys/arc_impl.h>
 #include <sys/dsl_deadlist.h>
 #include <zfeature_common.h>
 
@@ -283,6 +284,7 @@ struct spa {
 	spa_aux_vdev_t	spa_spares;		/* hot spares */
 	spa_aux_vdev_t	spa_l2cache;		/* L2ARC cache devices */
 	boolean_t	spa_aux_sync_uber;	/* need to sync aux uber */
+	l2arc_info_t	spa_l2arc_info;		/* L2ARC state and stats */
 	nvlist_t	*spa_label_features;	/* Features for reading MOS */
 	uint64_t	spa_config_object;	/* MOS object for pool config */
 	uint64_t	spa_config_generation;	/* config generation number */

man/man4/zfs.4

@@ -109,8 +109,7 @@ Seconds between L2ARC writing.
 .
 .It Sy l2arc_headroom Ns = Ns Sy 8 Pq u64
 How far through the ARC lists to search for L2ARC cacheable content,
-expressed as a multiplier of
-.Sy l2arc_write_max .
+expressed as a multiplier of the effective write size.
 ARC persistence across reboots can be achieved with persistent L2ARC
 by setting this parameter to
 .Sy 0 ,
@@ -125,6 +124,19 @@ A value of
 .Sy 100
 disables this feature.
 .
+.It Sy l2arc_dwpd_limit Ns = Ns Sy 100 Pq uint
+Drive Writes Per Day limit for L2ARC devices to protect SSD endurance,
+specified as a percentage where 100 equals 1.0 DWPD.
+A value of 100 means each L2ARC device can write its own capacity once per day.
+Lower values support fractional DWPD
+(50 = 0.5 DWPD, 30 = 0.3 DWPD for QLC SSDs).
+Higher values allow more writes (300 = 3.0 DWPD).
+The effective write rate is always bounded by
+.Sy l2arc_write_max .
+A value of 0 disables DWPD rate limiting entirely.
+DWPD limiting only applies after the initial fill pass completes and when
+total L2ARC capacity meets the persist threshold.
+.
 .It Sy l2arc_exclude_special Ns = Ns Sy 0 Ns | Ns 1 Pq int
 Controls whether buffers present on special vdevs are eligible for caching
 into L2ARC.
@@ -179,9 +191,8 @@ can render it slow or unusable.
 This parameter limits L2ARC writes and rebuilds to achieve the target.
 .
 .It Sy l2arc_trim_ahead Ns = Ns Sy 0 Ns % Pq u64
-Trims ahead of the current write size
-.Pq Sy l2arc_write_max
-on L2ARC devices by this percentage of write size if we have filled the device.
+Trims ahead of the current write size on L2ARC devices by this percentage
+of write size if we have filled the device.
 If set to
 .Sy 100
 we TRIM twice the space required to accommodate upcoming writes.
@@ -216,13 +227,12 @@ to enable caching/reading prefetches to/from L2ARC.
 .It Sy l2arc_norw Ns = Ns Sy 0 Ns | Ns 1 Pq int
 No reads during writes.
 .
-.It Sy l2arc_write_boost Ns = Ns Sy 33554432 Ns B Po 32 MiB Pc Pq u64
-Cold L2ARC devices will have
-.Sy l2arc_write_max
-increased by this amount while they remain cold.
-.
 .It Sy l2arc_write_max Ns = Ns Sy 33554432 Ns B Po 32 MiB Pc Pq u64
-Max write bytes per interval.
+Maximum write rate in bytes per second for each L2ARC device.
+Used directly during initial fill, when DWPD limiting is disabled,
+or for non-persistent L2ARC.
+When DWPD limiting is active, writes are capped by this rate.
+Total L2ARC throughput scales with the number of cache devices in a pool.
 .
 .It Sy l2arc_rebuild_enabled Ns = Ns Sy 1 Ns | Ns 0 Pq int
 Rebuild the L2ARC when importing a pool (persistent L2ARC).