lightnvm: manage open and closed blocks separately

LightNVM targets need to know the state of the flash block when doing
flash optimizations. An example is implementing a write buffer to
respect the flash page size. Currently, block state is not accounted
for; the media manager only differentiates among free, bad and in-use
blocks.

This patch adds the logic in the generic media manager to enable
targets manage blocks into open and close separately, and it implements
such management in rrpc. It also adds a set of flags to describe the
state of the block (open, closed, free, bad).

In order to avoid taking two locks (nvm_lun and rrpc_lun) consecutively,
we introduce lockless get_/put_block primitives so that the open and
close list locks and future common logic is handled within the nvm_lun
lock.

Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>

drivers/lightnvm/core.c

@@ -167,6 +167,20 @@ static struct nvm_dev *nvm_find_nvm_dev(const char *name)
 	return NULL;
 }
 
+struct nvm_block *nvm_get_blk_unlocked(struct nvm_dev *dev, struct nvm_lun *lun,
+							unsigned long flags)
+{
+	return dev->mt->get_blk_unlocked(dev, lun, flags);
+}
+EXPORT_SYMBOL(nvm_get_blk_unlocked);
+
+/* Assumes that all valid pages have already been moved on release to bm */
+void nvm_put_blk_unlocked(struct nvm_dev *dev, struct nvm_block *blk)
+{
+	return dev->mt->put_blk_unlocked(dev, blk);
+}
+EXPORT_SYMBOL(nvm_put_blk_unlocked);
+
 struct nvm_block *nvm_get_blk(struct nvm_dev *dev, struct nvm_lun *lun,
 							unsigned long flags)
 {

drivers/lightnvm/gennvm.c

@@ -60,7 +60,8 @@ static int gennvm_luns_init(struct nvm_dev *dev, struct gen_nvm *gn)
 		lun->vlun.lun_id = i % dev->luns_per_chnl;
 		lun->vlun.chnl_id = i / dev->luns_per_chnl;
 		lun->vlun.nr_free_blocks = dev->blks_per_lun;
-		lun->vlun.nr_inuse_blocks = 0;
+		lun->vlun.nr_open_blocks = 0;
+		lun->vlun.nr_closed_blocks = 0;
 		lun->vlun.nr_bad_blocks = 0;
 	}
 	return 0;
@@ -134,15 +135,15 @@ static int gennvm_block_map(u64 slba, u32 nlb, __le64 *entries, void *private)
 		pba = pba - (dev->sec_per_lun * lun_id);
 		blk = &lun->vlun.blocks[div_u64(pba, dev->sec_per_blk)];
 
-		if (!blk->type) {
+		if (!blk->state) {
 			/* at this point, we don't know anything about the
 			 * block. It's up to the FTL on top to re-etablish the
-			 * block state
+			 * block state. The block is assumed to be open.
 			 */
 			list_move_tail(&blk->list, &lun->used_list);
-			blk->type = 1;
+			blk->state = NVM_BLK_ST_OPEN;
 			lun->vlun.nr_free_blocks--;
-			lun->vlun.nr_inuse_blocks++;
+			lun->vlun.nr_open_blocks++;
 		}
 	}
 
@@ -256,14 +257,14 @@ static void gennvm_unregister(struct nvm_dev *dev)
 	module_put(THIS_MODULE);
 }
 
-static struct nvm_block *gennvm_get_blk(struct nvm_dev *dev,
+static struct nvm_block *gennvm_get_blk_unlocked(struct nvm_dev *dev,
 				struct nvm_lun *vlun, unsigned long flags)
 {
 	struct gen_lun *lun = container_of(vlun, struct gen_lun, vlun);
 	struct nvm_block *blk = NULL;
 	int is_gc = flags & NVM_IOTYPE_GC;
 
-	spin_lock(&vlun->lock);
+	assert_spin_locked(&vlun->lock);
 
 	if (list_empty(&lun->free_list)) {
 		pr_err_ratelimited("gennvm: lun %u have no free pages available",
@@ -276,44 +277,63 @@ static struct nvm_block *gennvm_get_blk(struct nvm_dev *dev,
 
 	blk = list_first_entry(&lun->free_list, struct nvm_block, list);
 	list_move_tail(&blk->list, &lun->used_list);
-	blk->type = 1;
+	blk->state = NVM_BLK_ST_OPEN;
 
 	lun->vlun.nr_free_blocks--;
-	lun->vlun.nr_inuse_blocks++;
+	lun->vlun.nr_open_blocks++;
 
 out:
+	return blk;
+}
+
+static struct nvm_block *gennvm_get_blk(struct nvm_dev *dev,
+				struct nvm_lun *vlun, unsigned long flags)
+{
+	struct nvm_block *blk;
+
+	spin_lock(&vlun->lock);
+	blk = gennvm_get_blk_unlocked(dev, vlun, flags);
 	spin_unlock(&vlun->lock);
 	return blk;
 }
 
+static void gennvm_put_blk_unlocked(struct nvm_dev *dev, struct nvm_block *blk)
+{
+	struct nvm_lun *vlun = blk->lun;
+	struct gen_lun *lun = container_of(vlun, struct gen_lun, vlun);
+
+	assert_spin_locked(&vlun->lock);
+
+	if (blk->state & NVM_BLK_ST_OPEN) {
+		list_move_tail(&blk->list, &lun->free_list);
+		lun->vlun.nr_open_blocks--;
+		lun->vlun.nr_free_blocks++;
+		blk->state = NVM_BLK_ST_FREE;
+	} else if (blk->state & NVM_BLK_ST_CLOSED) {
+		list_move_tail(&blk->list, &lun->free_list);
+		lun->vlun.nr_closed_blocks--;
+		lun->vlun.nr_free_blocks++;
+		blk->state = NVM_BLK_ST_FREE;
+	} else if (blk->state & NVM_BLK_ST_BAD) {
+		list_move_tail(&blk->list, &lun->bb_list);
+		lun->vlun.nr_bad_blocks++;
+		blk->state = NVM_BLK_ST_BAD;
+	} else {
+		WARN_ON_ONCE(1);
+		pr_err("gennvm: erroneous block type (%lu -> %u)\n",
+							blk->id, blk->state);
+		list_move_tail(&blk->list, &lun->bb_list);
+		lun->vlun.nr_bad_blocks++;
+		blk->state = NVM_BLK_ST_BAD;
+	}
+}
+
 static void gennvm_put_blk(struct nvm_dev *dev, struct nvm_block *blk)
 {
 	struct nvm_lun *vlun = blk->lun;
-	struct gen_lun *lun = container_of(vlun, struct gen_lun, vlun);
 
 	spin_lock(&vlun->lock);
-
+	gennvm_put_blk_unlocked(dev, blk);
-	switch (blk->type) {
-	case 1:
-		list_move_tail(&blk->list, &lun->free_list);
-		lun->vlun.nr_free_blocks++;
-		lun->vlun.nr_inuse_blocks--;
-		blk->type = 0;
-		break;
-	case 2:
-		list_move_tail(&blk->list, &lun->bb_list);
-		lun->vlun.nr_bad_blocks++;
-		lun->vlun.nr_inuse_blocks--;
-		break;
-	default:
-		WARN_ON_ONCE(1);
-		pr_err("gennvm: erroneous block type (%lu -> %u)\n",
-							blk->id, blk->type);
-		list_move_tail(&blk->list, &lun->bb_list);
-		lun->vlun.nr_bad_blocks++;
-		lun->vlun.nr_inuse_blocks--;
-	}
-
 	spin_unlock(&vlun->lock);
 }
 
@@ -339,7 +359,7 @@ static void gennvm_blk_set_type(struct nvm_dev *dev, struct ppa_addr *ppa,
 	blk = &lun->vlun.blocks[ppa->g.blk];
 
 	/* will be moved to bb list on put_blk from target */
-	blk->type = type;
+	blk->state = type;
 }
 
 /* mark block bad. It is expected the target recover from the error. */
@@ -358,9 +378,10 @@ static void gennvm_mark_blk_bad(struct nvm_dev *dev, struct nvm_rq *rqd)
 	/* look up blocks and mark them as bad */
 	if (rqd->nr_pages > 1)
 		for (i = 0; i < rqd->nr_pages; i++)
-			gennvm_blk_set_type(dev, &rqd->ppa_list[i], 2);
+			gennvm_blk_set_type(dev, &rqd->ppa_list[i],
+						NVM_BLK_ST_BAD);
 	else
-		gennvm_blk_set_type(dev, &rqd->ppa_addr, 2);
+		gennvm_blk_set_type(dev, &rqd->ppa_addr, NVM_BLK_ST_BAD);
 }
 
 static void gennvm_end_io(struct nvm_rq *rqd)
@@ -416,10 +437,11 @@ static void gennvm_lun_info_print(struct nvm_dev *dev)
 	gennvm_for_each_lun(gn, lun, i) {
 		spin_lock(&lun->vlun.lock);
 
-		pr_info("%s: lun%8u\t%u\t%u\t%u\n",
+		pr_info("%s: lun%8u\t%u\t%u\t%u\t%u\n",
 				dev->name, i,
 				lun->vlun.nr_free_blocks,
-				lun->vlun.nr_inuse_blocks,
+				lun->vlun.nr_open_blocks,
+				lun->vlun.nr_closed_blocks,
 				lun->vlun.nr_bad_blocks);
 
 		spin_unlock(&lun->vlun.lock);
@@ -427,20 +449,23 @@ static void gennvm_lun_info_print(struct nvm_dev *dev)
 }
 
 static struct nvmm_type gennvm = {
-	.name		= "gennvm",
+	.name			= "gennvm",
-	.version	= {0, 1, 0},
+	.version		= {0, 1, 0},
 
-	.register_mgr	= gennvm_register,
+	.register_mgr		= gennvm_register,
-	.unregister_mgr	= gennvm_unregister,
+	.unregister_mgr		= gennvm_unregister,
 
-	.get_blk	= gennvm_get_blk,
+	.get_blk_unlocked	= gennvm_get_blk_unlocked,
-	.put_blk	= gennvm_put_blk,
+	.put_blk_unlocked	= gennvm_put_blk_unlocked,
 
-	.submit_io	= gennvm_submit_io,
+	.get_blk		= gennvm_get_blk,
-	.erase_blk	= gennvm_erase_blk,
+	.put_blk		= gennvm_put_blk,
 
-	.get_lun	= gennvm_get_lun,
+	.submit_io		= gennvm_submit_io,
-	.lun_info_print = gennvm_lun_info_print,
+	.erase_blk		= gennvm_erase_blk,
+
+	.get_lun		= gennvm_get_lun,
+	.lun_info_print		= gennvm_lun_info_print,
 };
 
 static int __init gennvm_module_init(void)

drivers/lightnvm/rrpc.c

@@ -179,16 +179,23 @@ static void rrpc_set_lun_cur(struct rrpc_lun *rlun, struct rrpc_block *rblk)
 static struct rrpc_block *rrpc_get_blk(struct rrpc *rrpc, struct rrpc_lun *rlun,
 							unsigned long flags)
 {
+	struct nvm_lun *lun = rlun->parent;
 	struct nvm_block *blk;
 	struct rrpc_block *rblk;
 
-	blk = nvm_get_blk(rrpc->dev, rlun->parent, flags);
+	spin_lock(&lun->lock);
-	if (!blk)
+	blk = nvm_get_blk_unlocked(rrpc->dev, rlun->parent, flags);
+	if (!blk) {
+		pr_err("nvm: rrpc: cannot get new block from media manager\n");
+		spin_unlock(&lun->lock);
 		return NULL;
+	}
 
 	rblk = &rlun->blocks[blk->id];
-	blk->priv = rblk;
+	list_add_tail(&rblk->list, &rlun->open_list);
+	spin_unlock(&lun->lock);
 
+	blk->priv = rblk;
 	bitmap_zero(rblk->invalid_pages, rrpc->dev->pgs_per_blk);
 	rblk->next_page = 0;
 	rblk->nr_invalid_pages = 0;
@@ -199,7 +206,13 @@ static struct rrpc_block *rrpc_get_blk(struct rrpc *rrpc, struct rrpc_lun *rlun,
 
 static void rrpc_put_blk(struct rrpc *rrpc, struct rrpc_block *rblk)
 {
-	nvm_put_blk(rrpc->dev, rblk->parent);
+	struct rrpc_lun *rlun = rblk->rlun;
+	struct nvm_lun *lun = rlun->parent;
+
+	spin_lock(&lun->lock);
+	nvm_put_blk_unlocked(rrpc->dev, rblk->parent);
+	list_del(&rblk->list);
+	spin_unlock(&lun->lock);
 }
 
 static void rrpc_put_blks(struct rrpc *rrpc)
@@ -653,8 +666,20 @@ static void rrpc_end_io_write(struct rrpc *rrpc, struct rrpc_rq *rrqd,
 		lun = rblk->parent->lun;
 
 		cmnt_size = atomic_inc_return(&rblk->data_cmnt_size);
-		if (unlikely(cmnt_size == rrpc->dev->pgs_per_blk))
+		if (unlikely(cmnt_size == rrpc->dev->pgs_per_blk)) {
+			struct nvm_block *blk = rblk->parent;
+			struct rrpc_lun *rlun = rblk->rlun;
+
+			spin_lock(&lun->lock);
+			lun->nr_open_blocks--;
+			lun->nr_closed_blocks++;
+			blk->state &= ~NVM_BLK_ST_OPEN;
+			blk->state |= NVM_BLK_ST_CLOSED;
+			list_move_tail(&rblk->list, &rlun->closed_list);
+			spin_unlock(&lun->lock);
+
 			rrpc_run_gc(rrpc, rblk);
+		}
 	}
 }
 
@@ -1134,6 +1159,9 @@ static int rrpc_luns_init(struct rrpc *rrpc, int lun_begin, int lun_end)
 		rlun->rrpc = rrpc;
 		rlun->parent = lun;
 		INIT_LIST_HEAD(&rlun->prio_list);
+		INIT_LIST_HEAD(&rlun->open_list);
+		INIT_LIST_HEAD(&rlun->closed_list);
+
 		INIT_WORK(&rlun->ws_gc, rrpc_lun_gc);
 		spin_lock_init(&rlun->lock);
 

drivers/lightnvm/rrpc.h

@@ -56,6 +56,7 @@ struct rrpc_block {
 	struct nvm_block *parent;
 	struct rrpc_lun *rlun;
 	struct list_head prio;
+	struct list_head list;
 
 #define MAX_INVALID_PAGES_STORAGE 8
 	/* Bitmap for invalid page intries */
@@ -74,7 +75,16 @@ struct rrpc_lun {
 	struct nvm_lun *parent;
 	struct rrpc_block *cur, *gc_cur;
 	struct rrpc_block *blocks;	/* Reference to block allocation */
-	struct list_head prio_list;		/* Blocks that may be GC'ed */
+
+	struct list_head prio_list;	/* Blocks that may be GC'ed */
+	struct list_head open_list;	/* In-use open blocks. These are blocks
+					 * that can be both written to and read
+					 * from
+					 */
+	struct list_head closed_list;	/* In-use closed blocks. These are
+					 * blocks that can _only_ be read from
+					 */
+
 	struct work_struct ws_gc;
 
 	spinlock_t lock;

include/linux/lightnvm.h

@@ -229,12 +229,25 @@ struct nvm_lun {
 	int lun_id;
 	int chnl_id;
 
-	unsigned int nr_inuse_blocks;	/* Number of used blocks */
+	/* It is up to the target to mark blocks as closed. If the target does
+	 * not do it, all blocks are marked as open, and nr_open_blocks
+	 * represents the number of blocks in use
+	 */
+	unsigned int nr_open_blocks;	/* Number of used, writable blocks */
+	unsigned int nr_closed_blocks;	/* Number of used, read-only blocks */
 	unsigned int nr_free_blocks;	/* Number of unused blocks */
 	unsigned int nr_bad_blocks;	/* Number of bad blocks */
-	struct nvm_block *blocks;
 
 	spinlock_t lock;
+
+	struct nvm_block *blocks;
+};
+
+enum {
+	NVM_BLK_ST_FREE =	0x1,	/* Free block */
+	NVM_BLK_ST_OPEN =	0x2,	/* Open block - read-write */
+	NVM_BLK_ST_CLOSED =	0x4,	/* Closed block - read-only */
+	NVM_BLK_ST_BAD =	0x8,	/* Bad block */
 };
 
 struct nvm_block {
@@ -243,7 +256,7 @@ struct nvm_block {
 	unsigned long id;
 
 	void *priv;
-	int type;
+	int state;
 };
 
 struct nvm_dev {
@@ -404,6 +417,8 @@ struct nvmm_type {
 	nvmm_unregister_fn *unregister_mgr;
 
 	/* Block administration callbacks */
+	nvmm_get_blk_fn *get_blk_unlocked;
+	nvmm_put_blk_fn *put_blk_unlocked;
 	nvmm_get_blk_fn *get_blk;
 	nvmm_put_blk_fn *put_blk;
 	nvmm_open_blk_fn *open_blk;
@@ -424,6 +439,10 @@ struct nvmm_type {
 extern int nvm_register_mgr(struct nvmm_type *);
 extern void nvm_unregister_mgr(struct nvmm_type *);
 
+extern struct nvm_block *nvm_get_blk_unlocked(struct nvm_dev *,
+					struct nvm_lun *, unsigned long);
+extern void nvm_put_blk_unlocked(struct nvm_dev *, struct nvm_block *);
+
 extern struct nvm_block *nvm_get_blk(struct nvm_dev *, struct nvm_lun *,
 								unsigned long);
 extern void nvm_put_blk(struct nvm_dev *, struct nvm_block *);