lightnvm: physical block device (pblk) target

This patch introduces pblk, a host-side translation layer for
Open-Channel SSDs to expose them like block devices. The translation
layer allows data placement decisions, and I/O scheduling to be
managed by the host, enabling users to optimize the SSD for their
specific workloads.

An open-channel SSD has a set of LUNs (parallel units) and a
collection of blocks. Each block can be read in any order, but
writes must be sequential. Writes may also fail, and if a block
requires it, must also be reset before new writes can be
applied.

To manage the constraints, pblk maintains a logical to
physical address (L2P) table,  write cache, garbage
collection logic, recovery scheme, and logic to rate-limit
user I/Os versus garbage collection I/Os.

The L2P table is fully-associative and manages sectors at a
4KB granularity. Pblk stores the L2P table in two places, in
the out-of-band area of the media and on the last page of a
line. In the cause of a power failure, pblk will perform a
scan to recover the L2P table.

The user data is organized into lines. A line is data
striped across blocks and LUNs. The lines enable the host to
reduce the amount of metadata to maintain besides the user
data and makes it easier to implement RAID or erasure coding
in the future.

pblk implements multi-tenant support and can be instantiated
multiple times on the same drive. Each instance owns a
portion of the SSD - both regarding I/O bandwidth and
capacity - providing I/O isolation for each case.

Finally, pblk also exposes a sysfs interface that allows
user-space to peek into the internals of pblk. The interface
is available at /dev/block/*/pblk/ where * is the block
device name exposed.

This work also contains contributions from:
  Matias Bjørling <matias@cnexlabs.com>
  Simon A. F. Lund <slund@cnexlabs.com>
  Young Tack Jin <youngtack.jin@gmail.com>
  Huaicheng Li <huaicheng@cs.uchicago.edu>

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

14 files changed, 8023 insertions, 0 deletions

diff --git a/drivers/lightnvm/Kconfig b/drivers/lightnvm/Kconfig
index 052714106b7b..ead61a93cb4e 100644
--- a/drivers/lightnvm/Kconfig
+++ b/drivers/lightnvm/Kconfig
@@ -33,4 +33,13 @@ config NVM_RRPC
 	host. The target is implemented using a linear mapping table and
 	cost-based garbage collection. It is optimized for 4K IO sizes.
 
+config NVM_PBLK
+	tristate "Physical Block Device Open-Channel SSD target"
+	---help---
+	Allows an open-channel SSD to be exposed as a block device to the
+	host. The target assumes the device exposes raw flash and must be
+	explicitly managed by the host.
+
+	Please note the disk format is considered EXPERIMENTAL for now.
+
 endif # NVM
diff --git a/drivers/lightnvm/Makefile b/drivers/lightnvm/Makefile
index b2a39e2d2895..82d1a117fb27 100644
--- a/drivers/lightnvm/Makefile
+++ b/drivers/lightnvm/Makefile
@@ -4,3 +4,8 @@
 
 obj-$(CONFIG_NVM)		:= core.o
 obj-$(CONFIG_NVM_RRPC)		+= rrpc.o
+obj-$(CONFIG_NVM_PBLK)		+= pblk.o
+pblk-y				:= pblk-init.o pblk-core.o pblk-rb.o \
+				   pblk-write.o pblk-cache.o pblk-read.o \
+				   pblk-gc.o pblk-recovery.o pblk-map.o \
+				   pblk-rl.o pblk-sysfs.o
diff --git a/drivers/lightnvm/pblk-cache.c b/drivers/lightnvm/pblk-cache.c
new file mode 100644
index 000000000000..59bcea88db84
--- /dev/null
+++ b/drivers/lightnvm/pblk-cache.c
@@ -0,0 +1,114 @@
+/*
+ * Copyright (C) 2016 CNEX Labs
+ * Initial release: Javier Gonzalez <javier@cnexlabs.com>
+ *                  Matias Bjorling <matias@cnexlabs.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License version
+ * 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * pblk-cache.c - pblk's write cache
+ */
+
+#include "pblk.h"
+
+int pblk_write_to_cache(struct pblk *pblk, struct bio *bio, unsigned long flags)
+{
+	struct pblk_w_ctx w_ctx;
+	sector_t lba = pblk_get_lba(bio);
+	unsigned int bpos, pos;
+	int nr_entries = pblk_get_secs(bio);
+	int i, ret;
+
+	/* Update the write buffer head (mem) with the entries that we can
+	 * write. The write in itself cannot fail, so there is no need to
+	 * rollback from here on.
+	 */
+retry:
+	ret = pblk_rb_may_write_user(&pblk->rwb, bio, nr_entries, &bpos);
+	if (ret == NVM_IO_REQUEUE) {
+		io_schedule();
+		goto retry;
+	}
+
+	if (unlikely(!bio_has_data(bio)))
+		goto out;
+
+	w_ctx.flags = flags;
+	pblk_ppa_set_empty(&w_ctx.ppa);
+
+	for (i = 0; i < nr_entries; i++) {
+		void *data = bio_data(bio);
+
+		w_ctx.lba = lba + i;
+
+		pos = pblk_rb_wrap_pos(&pblk->rwb, bpos + i);
+		pblk_rb_write_entry_user(&pblk->rwb, data, w_ctx, pos);
+
+		bio_advance(bio, PBLK_EXPOSED_PAGE_SIZE);
+	}
+
+#ifdef CONFIG_NVM_DEBUG
+	atomic_long_add(nr_entries, &pblk->inflight_writes);
+	atomic_long_add(nr_entries, &pblk->req_writes);
+#endif
+
+out:
+	pblk_write_should_kick(pblk);
+	return ret;
+}
+
+/*
+ * On GC the incoming lbas are not necessarily sequential. Also, some of the
+ * lbas might not be valid entries, which are marked as empty by the GC thread
+ */
+int pblk_write_gc_to_cache(struct pblk *pblk, void *data, u64 *lba_list,
+			   unsigned int nr_entries, unsigned int nr_rec_entries,
+			   struct pblk_line *gc_line, unsigned long flags)
+{
+	struct pblk_w_ctx w_ctx;
+	unsigned int bpos, pos;
+	int i, valid_entries;
+
+	/* Update the write buffer head (mem) with the entries that we can
+	 * write. The write in itself cannot fail, so there is no need to
+	 * rollback from here on.
+	 */
+retry:
+	if (!pblk_rb_may_write_gc(&pblk->rwb, nr_rec_entries, &bpos)) {
+		io_schedule();
+		goto retry;
+	}
+
+	w_ctx.flags = flags;
+	pblk_ppa_set_empty(&w_ctx.ppa);
+
+	for (i = 0, valid_entries = 0; i < nr_entries; i++) {
+		if (lba_list[i] == ADDR_EMPTY)
+			continue;
+
+		w_ctx.lba = lba_list[i];
+
+		pos = pblk_rb_wrap_pos(&pblk->rwb, bpos + valid_entries);
+		pblk_rb_write_entry_gc(&pblk->rwb, data, w_ctx, gc_line, pos);
+
+		data += PBLK_EXPOSED_PAGE_SIZE;
+		valid_entries++;
+	}
+
+	WARN_ONCE(nr_rec_entries != valid_entries,
+					"pblk: inconsistent GC write\n");
+
+#ifdef CONFIG_NVM_DEBUG
+	atomic_long_add(valid_entries, &pblk->inflight_writes);
+	atomic_long_add(valid_entries, &pblk->recov_gc_writes);
+#endif
+
+	pblk_write_should_kick(pblk);
+	return NVM_IO_OK;
+}
diff --git a/drivers/lightnvm/pblk-core.c b/drivers/lightnvm/pblk-core.c
new file mode 100644
index 000000000000..a2bcd098babc
--- /dev/null
+++ b/drivers/lightnvm/pblk-core.c
@@ -0,0 +1,1655 @@
+/*
+ * Copyright (C) 2016 CNEX Labs
+ * Initial release: Javier Gonzalez <javier@cnexlabs.com>
+ *                  Matias Bjorling <matias@cnexlabs.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License version
+ * 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * pblk-core.c - pblk's core functionality
+ *
+ */
+
+#include "pblk.h"
+#include <linux/time.h>
+
+static void pblk_mark_bb(struct pblk *pblk, struct pblk_line *line,
+			 struct ppa_addr *ppa)
+{
+	struct nvm_tgt_dev *dev = pblk->dev;
+	struct nvm_geo *geo = &dev->geo;
+	int pos = pblk_dev_ppa_to_pos(geo, *ppa);
+
+	pr_debug("pblk: erase failed: line:%d, pos:%d\n", line->id, pos);
+	atomic_long_inc(&pblk->erase_failed);
+
+	if (test_and_set_bit(pos, line->blk_bitmap))
+		pr_err("pblk: attempted to erase bb: line:%d, pos:%d\n",
+							line->id, pos);
+
+	pblk_line_run_ws(pblk, NULL, ppa, pblk_line_mark_bb);
+}
+
+static void __pblk_end_io_erase(struct pblk *pblk, struct nvm_rq *rqd)
+{
+	struct pblk_line *line;
+
+	line = &pblk->lines[pblk_dev_ppa_to_line(rqd->ppa_addr)];
+	atomic_dec(&line->left_seblks);
+
+	if (rqd->error) {
+		struct ppa_addr *ppa;
+
+		ppa = kmalloc(sizeof(struct ppa_addr), GFP_ATOMIC);
+		if (!ppa)
+			return;
+
+		*ppa = rqd->ppa_addr;
+		pblk_mark_bb(pblk, line, ppa);
+	}
+}
+
+/* Erase completion assumes that only one block is erased at the time */
+static void pblk_end_io_erase(struct nvm_rq *rqd)
+{
+	struct pblk *pblk = rqd->private;
+
+	up(&pblk->erase_sem);
+	__pblk_end_io_erase(pblk, rqd);
+	mempool_free(rqd, pblk->r_rq_pool);
+}
+
+static void __pblk_map_invalidate(struct pblk *pblk, struct pblk_line *line,
+				  u64 paddr)
+{
+	struct pblk_line_mgmt *l_mg = &pblk->l_mg;
+	struct list_head *move_list = NULL;
+
+	/* Lines being reclaimed (GC'ed) cannot be invalidated. Before the L2P
+	 * table is modified with reclaimed sectors, a check is done to endure
+	 * that newer updates are not overwritten.
+	 */
+	spin_lock(&line->lock);
+	if (line->state == PBLK_LINESTATE_GC ||
+					line->state == PBLK_LINESTATE_FREE) {
+		spin_unlock(&line->lock);
+		return;
+	}
+
+	if (test_and_set_bit(paddr, line->invalid_bitmap)) {
+		WARN_ONCE(1, "pblk: double invalidate\n");
+		spin_unlock(&line->lock);
+		return;
+	}
+	line->vsc--;
+
+	if (line->state == PBLK_LINESTATE_CLOSED)
+		move_list = pblk_line_gc_list(pblk, line);
+	spin_unlock(&line->lock);
+
+	if (move_list) {
+		spin_lock(&l_mg->gc_lock);
+		spin_lock(&line->lock);
+		/* Prevent moving a line that has just been chosen for GC */
+		if (line->state == PBLK_LINESTATE_GC ||
+					line->state == PBLK_LINESTATE_FREE) {
+			spin_unlock(&line->lock);
+			spin_unlock(&l_mg->gc_lock);
+			return;
+		}
+		spin_unlock(&line->lock);
+
+		list_move_tail(&line->list, move_list);
+		spin_unlock(&l_mg->gc_lock);
+	}
+}
+
+void pblk_map_invalidate(struct pblk *pblk, struct ppa_addr ppa)
+{
+	struct pblk_line *line;
+	u64 paddr;
+	int line_id;
+
+#ifdef CONFIG_NVM_DEBUG
+	/* Callers must ensure that the ppa points to a device address */
+	BUG_ON(pblk_addr_in_cache(ppa));
+	BUG_ON(pblk_ppa_empty(ppa));
+#endif
+
+	line_id = pblk_tgt_ppa_to_line(ppa);
+	line = &pblk->lines[line_id];
+	paddr = pblk_dev_ppa_to_line_addr(pblk, ppa);
+
+	__pblk_map_invalidate(pblk, line, paddr);
+}
+
+void pblk_map_pad_invalidate(struct pblk *pblk, struct pblk_line *line,
+			     u64 paddr)
+{
+	__pblk_map_invalidate(pblk, line, paddr);
+
+	pblk_rb_sync_init(&pblk->rwb, NULL);
+	line->left_ssecs--;
+	if (!line->left_ssecs)
+		pblk_line_run_ws(pblk, line, NULL, pblk_line_close_ws);
+	pblk_rb_sync_end(&pblk->rwb, NULL);
+}
+
+static void pblk_invalidate_range(struct pblk *pblk, sector_t slba,
+				  unsigned int nr_secs)
+{
+	sector_t lba;
+
+	spin_lock(&pblk->trans_lock);
+	for (lba = slba; lba < slba + nr_secs; lba++) {
+		struct ppa_addr ppa;
+
+		ppa = pblk_trans_map_get(pblk, lba);
+
+		if (!pblk_addr_in_cache(ppa) && !pblk_ppa_empty(ppa))
+			pblk_map_invalidate(pblk, ppa);
+
+		pblk_ppa_set_empty(&ppa);
+		pblk_trans_map_set(pblk, lba, ppa);
+	}
+	spin_unlock(&pblk->trans_lock);
+}
+
+struct nvm_rq *pblk_alloc_rqd(struct pblk *pblk, int rw)
+{
+	mempool_t *pool;
+	struct nvm_rq *rqd;
+	int rq_size;
+
+	if (rw == WRITE) {
+		pool = pblk->w_rq_pool;
+		rq_size = pblk_w_rq_size;
+	} else {
+		pool = pblk->r_rq_pool;
+		rq_size = pblk_r_rq_size;
+	}
+
+	rqd = mempool_alloc(pool, GFP_KERNEL);
+	memset(rqd, 0, rq_size);
+
+	return rqd;
+}
+
+void pblk_free_rqd(struct pblk *pblk, struct nvm_rq *rqd, int rw)
+{
+	mempool_t *pool;
+
+	if (rw == WRITE)
+		pool = pblk->w_rq_pool;
+	else
+		pool = pblk->r_rq_pool;
+
+	mempool_free(rqd, pool);
+}
+
+void pblk_bio_free_pages(struct pblk *pblk, struct bio *bio, int off,
+			 int nr_pages)
+{
+	struct bio_vec bv;
+	int i;
+
+	WARN_ON(off + nr_pages != bio->bi_vcnt);
+
+	bio_advance(bio, off * PBLK_EXPOSED_PAGE_SIZE);
+	for (i = off; i < nr_pages + off; i++) {
+		bv = bio->bi_io_vec[i];
+		mempool_free(bv.bv_page, pblk->page_pool);
+	}
+}
+
+int pblk_bio_add_pages(struct pblk *pblk, struct bio *bio, gfp_t flags,
+		       int nr_pages)
+{
+	struct request_queue *q = pblk->dev->q;
+	struct page *page;
+	int i, ret;
+
+	for (i = 0; i < nr_pages; i++) {
+		page = mempool_alloc(pblk->page_pool, flags);
+		if (!page)
+			goto err;
+
+		ret = bio_add_pc_page(q, bio, page, PBLK_EXPOSED_PAGE_SIZE, 0);
+		if (ret != PBLK_EXPOSED_PAGE_SIZE) {
+			pr_err("pblk: could not add page to bio\n");
+			mempool_free(page, pblk->page_pool);
+			goto err;
+		}
+	}
+
+	return 0;
+err:
+	pblk_bio_free_pages(pblk, bio, 0, i - 1);
+	return -1;
+}
+
+static void pblk_write_kick(struct pblk *pblk)
+{
+	wake_up_process(pblk->writer_ts);
+	mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(1000));
+}
+
+void pblk_write_timer_fn(unsigned long data)
+{
+	struct pblk *pblk = (struct pblk *)data;
+
+	/* kick the write thread every tick to flush outstanding data */
+	pblk_write_kick(pblk);
+}
+
+void pblk_write_should_kick(struct pblk *pblk)
+{
+	unsigned int secs_avail = pblk_rb_read_count(&pblk->rwb);
+
+	if (secs_avail >= pblk->min_write_pgs)
+		pblk_write_kick(pblk);
+}
+
+void pblk_end_bio_sync(struct bio *bio)
+{
+	struct completion *waiting = bio->bi_private;
+
+	complete(waiting);
+}
+
+void pblk_end_io_sync(struct nvm_rq *rqd)
+{
+	struct completion *waiting = rqd->private;
+
+	complete(waiting);
+}
+
+void pblk_flush_writer(struct pblk *pblk)
+{
+	struct bio *bio;
+	int ret;
+	DECLARE_COMPLETION_ONSTACK(wait);
+
+	bio = bio_alloc(GFP_KERNEL, 1);
+	if (!bio)
+		return;
+
+	bio->bi_iter.bi_sector = 0; /* internal bio */
+	bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_OP_FLUSH);
+	bio->bi_private = &wait;
+	bio->bi_end_io = pblk_end_bio_sync;
+
+	ret = pblk_write_to_cache(pblk, bio, 0);
+	if (ret == NVM_IO_OK) {
+		if (!wait_for_completion_io_timeout(&wait,
+				msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
+			pr_err("pblk: flush cache timed out\n");
+		}
+	} else if (ret != NVM_IO_DONE) {
+		pr_err("pblk: tear down bio failed\n");
+	}
+
+	if (bio->bi_error)
+		pr_err("pblk: flush sync write failed (%u)\n", bio->bi_error);
+
+	bio_put(bio);
+}
+
+struct list_head *pblk_line_gc_list(struct pblk *pblk, struct pblk_line *line)
+{
+	struct pblk_line_meta *lm = &pblk->lm;
+	struct pblk_line_mgmt *l_mg = &pblk->l_mg;
+	struct list_head *move_list = NULL;
+
+	if (!line->vsc) {
+		if (line->gc_group != PBLK_LINEGC_FULL) {
+			line->gc_group = PBLK_LINEGC_FULL;
+			move_list = &l_mg->gc_full_list;
+		}
+	} else if (line->vsc < lm->mid_thrs) {
+		if (line->gc_group != PBLK_LINEGC_HIGH) {
+			line->gc_group = PBLK_LINEGC_HIGH;
+			move_list = &l_mg->gc_high_list;
+		}
+	} else if (line->vsc < lm->high_thrs) {
+		if (line->gc_group != PBLK_LINEGC_MID) {
+			line->gc_group = PBLK_LINEGC_MID;
+			move_list = &l_mg->gc_mid_list;
+		}
+	} else if (line->vsc < line->sec_in_line) {
+		if (line->gc_group != PBLK_LINEGC_LOW) {
+			line->gc_group = PBLK_LINEGC_LOW;
+			move_list = &l_mg->gc_low_list;
+		}
+	} else if (line->vsc == line->sec_in_line) {
+		if (line->gc_group != PBLK_LINEGC_EMPTY) {
+			line->gc_group = PBLK_LINEGC_EMPTY;
+			move_list = &l_mg->gc_empty_list;
+		}
+	} else {
+		line->state = PBLK_LINESTATE_CORRUPT;
+		line->gc_group = PBLK_LINEGC_NONE;
+		move_list =  &l_mg->corrupt_list;
+		pr_err("pblk: corrupted vsc for line %d, vsc:%d (%d/%d/%d)\n",
+						line->id, line->vsc,
+						line->sec_in_line,
+						lm->high_thrs, lm->mid_thrs);
+	}
+
+	return move_list;
+}
+
+void pblk_discard(struct pblk *pblk, struct bio *bio)
+{
+	sector_t slba = pblk_get_lba(bio);
+	sector_t nr_secs = pblk_get_secs(bio);
+
+	pblk_invalidate_range(pblk, slba, nr_secs);
+}
+
+struct ppa_addr pblk_get_lba_map(struct pblk *pblk, sector_t lba)
+{
+	struct ppa_addr ppa;
+
+	spin_lock(&pblk->trans_lock);
+	ppa = pblk_trans_map_get(pblk, lba);
+	spin_unlock(&pblk->trans_lock);
+
+	return ppa;
+}
+
+void pblk_log_write_err(struct pblk *pblk, struct nvm_rq *rqd)
+{
+	atomic_long_inc(&pblk->write_failed);
+#ifdef CONFIG_NVM_DEBUG
+	pblk_print_failed_rqd(pblk, rqd, rqd->error);
+#endif
+}
+
+void pblk_log_read_err(struct pblk *pblk, struct nvm_rq *rqd)
+{
+	/* Empty page read is not necessarily an error (e.g., L2P recovery) */
+	if (rqd->error == NVM_RSP_ERR_EMPTYPAGE) {
+		atomic_long_inc(&pblk->read_empty);
+		return;
+	}
+
+	switch (rqd->error) {
+	case NVM_RSP_WARN_HIGHECC:
+		atomic_long_inc(&pblk->read_high_ecc);
+		break;
+	case NVM_RSP_ERR_FAILECC:
+	case NVM_RSP_ERR_FAILCRC:
+		atomic_long_inc(&pblk->read_failed);
+		break;
+	default:
+		pr_err("pblk: unknown read error:%d\n", rqd->error);
+	}
+#ifdef CONFIG_NVM_DEBUG
+	pblk_print_failed_rqd(pblk, rqd, rqd->error);
+#endif
+}
+
+int pblk_submit_io(struct pblk *pblk, struct nvm_rq *rqd)
+{
+	struct nvm_tgt_dev *dev = pblk->dev;
+
+#ifdef CONFIG_NVM_DEBUG
+	struct ppa_addr *ppa_list;
+
+	ppa_list = (rqd->nr_ppas > 1) ? rqd->ppa_list : &rqd->ppa_addr;
+	if (pblk_boundary_ppa_checks(dev, ppa_list, rqd->nr_ppas)) {
+		WARN_ON(1);
+		return -EINVAL;
+	}
+
+	if (rqd->opcode == NVM_OP_PWRITE) {
+		struct pblk_line *line;
+		struct ppa_addr ppa;
+		int i;
+
+		for (i = 0; i < rqd->nr_ppas; i++) {
+			ppa = ppa_list[i];
+			line = &pblk->lines[pblk_dev_ppa_to_line(ppa)];
+
+			spin_lock(&line->lock);
+			if (line->state != PBLK_LINESTATE_OPEN) {
+				pr_err("pblk: bad ppa: line:%d,state:%d\n",
+							line->id, line->state);
+				WARN_ON(1);
+				spin_unlock(&line->lock);
+				return -EINVAL;
+			}
+			spin_unlock(&line->lock);
+		}
+	}
+#endif
+	return nvm_submit_io(dev, rqd);
+}
+
+struct bio *pblk_bio_map_addr(struct pblk *pblk, void *data,
+			      unsigned int nr_secs, unsigned int len,
+			      gfp_t gfp_mask)
+{
+	struct nvm_tgt_dev *dev = pblk->dev;
+	struct pblk_line_mgmt *l_mg = &pblk->l_mg;
+	void *kaddr = data;
+	struct page *page;
+	struct bio *bio;
+	int i, ret;
+
+	if (l_mg->emeta_alloc_type == PBLK_KMALLOC_META)
+		return bio_map_kern(dev->q, kaddr, len, gfp_mask);
+
+	bio = bio_kmalloc(gfp_mask, nr_secs);
+	if (!bio)
+		return ERR_PTR(-ENOMEM);
+
+	for (i = 0; i < nr_secs; i++) {
+		page = vmalloc_to_page(kaddr);
+		if (!page) {
+			pr_err("pblk: could not map vmalloc bio\n");
+			bio_put(bio);
+			bio = ERR_PTR(-ENOMEM);
+			goto out;
+		}
+
+		ret = bio_add_pc_page(dev->q, bio, page, PAGE_SIZE, 0);
+		if (ret != PAGE_SIZE) {
+			pr_err("pblk: could not add page to bio\n");
+			bio_put(bio);
+			bio = ERR_PTR(-ENOMEM);
+			goto out;
+		}
+
+		kaddr += PAGE_SIZE;
+	}
+out:
+	return bio;
+}
+
+int pblk_calc_secs(struct pblk *pblk, unsigned long secs_avail,
+		   unsigned long secs_to_flush)
+{
+	int max = pblk->max_write_pgs;
+	int min = pblk->min_write_pgs;
+	int secs_to_sync = 0;
+
+	if (secs_avail >= max)
+		secs_to_sync = max;
+	else if (secs_avail >= min)
+		secs_to_sync = min * (secs_avail / min);
+	else if (secs_to_flush)
+		secs_to_sync = min;
+
+	return secs_to_sync;
+}
+
+static u64 __pblk_alloc_page(struct pblk *pblk, struct pblk_line *line,
+			     int nr_secs)
+{
+	u64 addr;
+	int i;
+
+	/* logic error: ppa out-of-bounds. Prevent generating bad address */
+	if (line->cur_sec + nr_secs > pblk->lm.sec_per_line) {
+		WARN(1, "pblk: page allocation out of bounds\n");
+		nr_secs = pblk->lm.sec_per_line - line->cur_sec;
+	}
+
+	line->cur_sec = addr = find_next_zero_bit(line->map_bitmap,
+					pblk->lm.sec_per_line, line->cur_sec);
+	for (i = 0; i < nr_secs; i++, line->cur_sec++)
+		WARN_ON(test_and_set_bit(line->cur_sec, line->map_bitmap));
+
+	return addr;
+}
+
+u64 pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
+{
+	u64 addr;
+
+	/* Lock needed in case a write fails and a recovery needs to remap
+	 * failed write buffer entries
+	 */
+	spin_lock(&line->lock);
+	addr = __pblk_alloc_page(pblk, line, nr_secs);
+	line->left_msecs -= nr_secs;
+	WARN(line->left_msecs < 0, "pblk: page allocation out of bounds\n");
+	spin_unlock(&line->lock);
+
+	return addr;
+}
+
+/*
+ * Submit emeta to one LUN in the raid line at the time to avoid a deadlock when
+ * taking the per LUN semaphore.
+ */
+static int pblk_line_submit_emeta_io(struct pblk *pblk, struct pblk_line *line,
+				     u64 paddr, int dir)
+{
+	struct nvm_tgt_dev *dev = pblk->dev;
+	struct nvm_geo *geo = &dev->geo;
+	struct pblk_line_meta *lm = &pblk->lm;
+	struct bio *bio;
+	struct nvm_rq rqd;
+	struct ppa_addr *ppa_list;
+	dma_addr_t dma_ppa_list;
+	void *emeta = line->emeta;
+	int min = pblk->min_write_pgs;
+	int left_ppas = lm->emeta_sec;
+	int id = line->id;
+	int rq_ppas, rq_len;
+	int cmd_op, bio_op;
+	int flags;
+	int i, j;
+	int ret;
+	DECLARE_COMPLETION_ONSTACK(wait);
+
+	if (dir == WRITE) {
+		bio_op = REQ_OP_WRITE;
+		cmd_op = NVM_OP_PWRITE;
+		flags = pblk_set_progr_mode(pblk, WRITE);
+	} else if (dir == READ) {
+		bio_op = REQ_OP_READ;
+		cmd_op = NVM_OP_PREAD;
+		flags = pblk_set_read_mode(pblk);
+	} else
+		return -EINVAL;
+
+	ppa_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL, &dma_ppa_list);
+	if (!ppa_list)
+		return -ENOMEM;
+
+next_rq:
+	memset(&rqd, 0, sizeof(struct nvm_rq));
+
+	rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
+	rq_len = rq_ppas * geo->sec_size;
+
+	bio = pblk_bio_map_addr(pblk, emeta, rq_ppas, rq_len, GFP_KERNEL);
+	if (IS_ERR(bio)) {
+		ret = PTR_ERR(bio);
+		goto free_rqd_dma;
+	}
+
+	bio->bi_iter.bi_sector = 0; /* internal bio */
+	bio_set_op_attrs(bio, bio_op, 0);
+
+	rqd.bio = bio;
+	rqd.opcode = cmd_op;
+	rqd.flags = flags;
+	rqd.nr_ppas = rq_ppas;
+	rqd.ppa_list = ppa_list;
+	rqd.dma_ppa_list = dma_ppa_list;
+	rqd.end_io = pblk_end_io_sync;
+	rqd.private = &wait;
+
+	if (dir == WRITE) {
+		for (i = 0; i < rqd.nr_ppas; ) {
+			spin_lock(&line->lock);
+			paddr = __pblk_alloc_page(pblk, line, min);
+			spin_unlock(&line->lock);
+			for (j = 0; j < min; j++, i++, paddr++)
+				rqd.ppa_list[i] =
+					addr_to_gen_ppa(pblk, paddr, id);
+		}
+	} else {
+		for (i = 0; i < rqd.nr_ppas; ) {
+			struct ppa_addr ppa = addr_to_gen_ppa(pblk, paddr, id);
+			int pos = pblk_dev_ppa_to_pos(geo, ppa);
+
+			while (test_bit(pos, line->blk_bitmap)) {
+				paddr += min;
+				if (pblk_boundary_paddr_checks(pblk, paddr)) {
+					pr_err("pblk: corrupt emeta line:%d\n",
+								line->id);
+					bio_put(bio);
+					ret = -EINTR;
+					goto free_rqd_dma;
+				}
+
+				ppa = addr_to_gen_ppa(pblk, paddr, id);
+				pos = pblk_dev_ppa_to_pos(geo, ppa);
+			}
+
+			if (pblk_boundary_paddr_checks(pblk, paddr + min)) {
+				pr_err("pblk: corrupt emeta line:%d\n",
+								line->id);
+				bio_put(bio);
+				ret = -EINTR;
+				goto free_rqd_dma;
+			}
+
+			for (j = 0; j < min; j++, i++, paddr++)
+				rqd.ppa_list[i] =
+					addr_to_gen_ppa(pblk, paddr, line->id);
+		}
+	}
+
+	ret = pblk_submit_io(pblk, &rqd);
+	if (ret) {
+		pr_err("pblk: emeta I/O submission failed: %d\n", ret);
+		bio_put(bio);
+		goto free_rqd_dma;
+	}
+
+	if (!wait_for_completion_io_timeout(&wait,
+				msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
+		pr_err("pblk: emeta I/O timed out\n");
+	}
+	reinit_completion(&wait);
+
+	bio_put(bio);
+
+	if (rqd.error) {
+		if (dir == WRITE)
+			pblk_log_write_err(pblk, &rqd);
+		else
+			pblk_log_read_err(pblk, &rqd);
+	}
+
+	emeta += rq_len;
+	left_ppas -= rq_ppas;
+	if (left_ppas)
+		goto next_rq;
+free_rqd_dma:
+	nvm_dev_dma_free(dev->parent, ppa_list, dma_ppa_list);
+	return ret;
+}
+
+u64 pblk_line_smeta_start(struct pblk *pblk, struct pblk_line *line)
+{
+	struct nvm_tgt_dev *dev = pblk->dev;
+	struct nvm_geo *geo = &dev->geo;
+	struct pblk_line_meta *lm = &pblk->lm;
+	int bit;
+
+	/* This usually only happens on bad lines */
+	bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
+	if (bit >= lm->blk_per_line)
+		return -1;
+
+	return bit * geo->sec_per_pl;
+}
+
+static int pblk_line_submit_smeta_io(struct pblk *pblk, struct pblk_line *line,
+				     u64 paddr, int dir)
+{
+	struct nvm_tgt_dev *dev = pblk->dev;
+	struct pblk_line_meta *lm = &pblk->lm;
+	struct bio *bio;
+	struct nvm_rq rqd;
+	__le64 *lba_list = NULL;
+	int i, ret;
+	int cmd_op, bio_op;
+	int flags;
+	DECLARE_COMPLETION_ONSTACK(wait);
+
+	if (dir == WRITE) {
+		bio_op = REQ_OP_WRITE;
+		cmd_op = NVM_OP_PWRITE;
+		flags = pblk_set_progr_mode(pblk, WRITE);
+		lba_list = pblk_line_emeta_to_lbas(line->emeta);
+	} else if (dir == READ) {
+		bio_op = REQ_OP_READ;
+		cmd_op = NVM_OP_PREAD;
+		flags = pblk_set_read_mode(pblk);
+	} else
+		return -EINVAL;
+
+	memset(&rqd, 0, sizeof(struct nvm_rq));
+
+	rqd.ppa_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
+							&rqd.dma_ppa_list);
+	if (!rqd.ppa_list)
+		return -ENOMEM;
+
+	bio = bio_map_kern(dev->q, line->smeta, lm->smeta_len, GFP_KERNEL);
+	if (IS_ERR(bio)) {
+		ret = PTR_ERR(bio);
+		goto free_ppa_list;
+	}
+
+	bio->bi_iter.bi_sector = 0; /* internal bio */
+	bio_set_op_attrs(bio, bio_op, 0);
+
+	rqd.bio = bio;
+	rqd.opcode = cmd_op;
+	rqd.flags = flags;
+	rqd.nr_ppas = lm->smeta_sec;
+	rqd.end_io = pblk_end_io_sync;
+	rqd.private = &wait;
+
+	for (i = 0; i < lm->smeta_sec; i++, paddr++) {
+		rqd.ppa_list[i] = addr_to_gen_ppa(pblk, paddr, line->id);
+		if (dir == WRITE)
+			lba_list[paddr] = cpu_to_le64(ADDR_EMPTY);
+	}
+
+	/*
+	 * This I/O is sent by the write thread when a line is replace. Since
+	 * the write thread is the only one sending write and erase commands,
+	 * there is no need to take the LUN semaphore.
+	 */
+	ret = pblk_submit_io(pblk, &rqd);
+	if (ret) {
+		pr_err("pblk: smeta I/O submission failed: %d\n", ret);
+		bio_put(bio);
+		goto free_ppa_list;
+	}
+
+	if (!wait_for_completion_io_timeout(&wait,
+				msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
+		pr_err("pblk: smeta I/O timed out\n");
+	}
+
+	if (rqd.error) {
+		if (dir == WRITE)
+			pblk_log_write_err(pblk, &rqd);
+		else
+			pblk_log_read_err(pblk, &rqd);
+	}
+
+free_ppa_list:
+	nvm_dev_dma_free(dev->parent, rqd.ppa_list, rqd.dma_ppa_list);
+
+	return ret;
+}
+
+int pblk_line_read_smeta(struct pblk *pblk, struct pblk_line *line)
+{
+	u64 bpaddr = pblk_line_smeta_start(pblk, line);
+
+	return pblk_line_submit_smeta_io(pblk, line, bpaddr, READ);
+}
+
+int pblk_line_read_emeta(struct pblk *pblk, struct pblk_line *line)
+{
+	return pblk_line_submit_emeta_io(pblk, line, line->emeta_ssec, READ);
+}
+
+static void pblk_setup_e_rq(struct pblk *pblk, struct nvm_rq *rqd,
+			    struct ppa_addr ppa)
+{
+	rqd->opcode = NVM_OP_ERASE;
+	rqd->ppa_addr = ppa;
+	rqd->nr_ppas = 1;
+	rqd->flags = pblk_set_progr_mode(pblk, ERASE);
+	rqd->bio = NULL;
+}
+
+static int pblk_blk_erase_sync(struct pblk *pblk, struct ppa_addr ppa)
+{
+	struct nvm_rq rqd;
+	int ret;
+	DECLARE_COMPLETION_ONSTACK(wait);
+
+	memset(&rqd, 0, sizeof(struct nvm_rq));
+
+	pblk_setup_e_rq(pblk, &rqd, ppa);
+
+	rqd.end_io = pblk_end_io_sync;
+	rqd.private = &wait;
+
+	/* The write thread schedules erases so that it minimizes disturbances
+	 * with writes. Thus, there is no need to take the LUN semaphore.
+	 */
+	ret = pblk_submit_io(pblk, &rqd);
+	if (ret) {
+		struct nvm_tgt_dev *dev = pblk->dev;
+		struct nvm_geo *geo = &dev->geo;
+
+		pr_err("pblk: could not sync erase line:%d,blk:%d\n",
+					pblk_dev_ppa_to_line(ppa),
+					pblk_dev_ppa_to_pos(geo, ppa));
+
+		rqd.error = ret;
+		goto out;
+	}
+
+	if (!wait_for_completion_io_timeout(&wait,
+				msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
+		pr_err("pblk: sync erase timed out\n");
+	}
+
+out:
+	rqd.private = pblk;
+	__pblk_end_io_erase(pblk, &rqd);
+
+	return 0;
+}
+
+int pblk_line_erase(struct pblk *pblk, struct pblk_line *line)
+{
+	struct pblk_line_meta *lm = &pblk->lm;
+	struct ppa_addr ppa;
+	int bit = -1;
+
+	/* Erase one block at the time and only erase good blocks */
+	while ((bit = find_next_zero_bit(line->erase_bitmap, lm->blk_per_line,
+						bit + 1)) < lm->blk_per_line) {
+		ppa = pblk->luns[bit].bppa; /* set ch and lun */
+		ppa.g.blk = line->id;
+
+		/* If the erase fails, the block is bad and should be marked */
+		line->left_eblks--;
+		WARN_ON(test_and_set_bit(bit, line->erase_bitmap));
+
+		if (pblk_blk_erase_sync(pblk, ppa)) {
+			pr_err("pblk: failed to erase line %d\n", line->id);
+			return -ENOMEM;
+		}
+	}
+
+	return 0;
+}
+
+/* For