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: NJavier González <javier@cnexlabs.com>
Signed-off-by: NMatias Bjørling <matias@cnexlabs.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

Documentation/lightnvm/pblk.txt

+pblk: Physical Block Device Target
+==================================
+
+pblk implements a fully associative, host-based FTL that exposes a traditional
+block I/O interface. Its primary responsibilities are:
+
+  - Map logical addresses onto physical addresses (4KB granularity) in a
+    logical-to-physical (L2P) table.
+  - Maintain the integrity and consistency of the L2P table as well as its
+    recovery from normal tear down and power outage.
+  - Deal with controller- and media-specific constrains.
+  - Handle I/O errors.
+  - Implement garbage collection.
+  - Maintain consistency across the I/O stack during synchronization points.
+
+For more information please refer to:
+
+  http://lightnvm.io
+
+which maintains updated FAQs, manual pages, technical documentation, tools,
+contacts, etc.

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

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

drivers/lightnvm/pblk-cache.c

+/*
+ * 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;
+}

drivers/lightnvm/pblk-gc.c

+/*
+ * 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-gc.c - pblk's garbage collector
+ */
+
+#include "pblk.h"
+#include <linux/delay.h>
+
+static void pblk_gc_free_gc_rq(struct pblk_gc_rq *gc_rq)
+{
+	kfree(gc_rq->data);
+	kfree(gc_rq->lba_list);
+	kfree(gc_rq);
+}
+
+static int pblk_gc_write(struct pblk *pblk)
+{
+	struct pblk_gc *gc = &pblk->gc;
+	struct pblk_gc_rq *gc_rq, *tgc_rq;
+	LIST_HEAD(w_list);
+
+	spin_lock(&gc->w_lock);
+	if (list_empty(&gc->w_list)) {
+		spin_unlock(&gc->w_lock);
+		return 1;
+	}
+
+	list_for_each_entry_safe(gc_rq, tgc_rq, &gc->w_list, list) {
+		list_move_tail(&gc_rq->list, &w_list);
+		gc->w_entries--;
+	}
+	spin_unlock(&gc->w_lock);
+
+	list_for_each_entry_safe(gc_rq, tgc_rq, &w_list, list) {
+		pblk_write_gc_to_cache(pblk, gc_rq->data, gc_rq->lba_list,
+				gc_rq->nr_secs, gc_rq->secs_to_gc,
+				gc_rq->line, PBLK_IOTYPE_GC);
+
+		kref_put(&gc_rq->line->ref, pblk_line_put);
+
+		list_del(&gc_rq->list);
+		pblk_gc_free_gc_rq(gc_rq);
+	}
+
+	return 0;
+}
+
+static void pblk_gc_writer_kick(struct pblk_gc *gc)
+{
+	wake_up_process(gc->gc_writer_ts);
+}
+
+/*
+ * Responsible for managing all memory related to a gc request. Also in case of
+ * failure
+ */
+static int pblk_gc_move_valid_secs(struct pblk *pblk, struct pblk_line *line,
+				   u64 *lba_list, unsigned int nr_secs)
+{
+	struct nvm_tgt_dev *dev = pblk->dev;
+	struct nvm_geo *geo = &dev->geo;
+	struct pblk_gc *gc = &pblk->gc;
+	struct pblk_gc_rq *gc_rq;
+	void *data;
+	unsigned int secs_to_gc;
+	int ret = NVM_IO_OK;
+
+	data = kmalloc(nr_secs * geo->sec_size, GFP_KERNEL);
+	if (!data) {
+		ret = NVM_IO_ERR;
+		goto free_lba_list;
+	}
+
+	/* Read from GC victim block */
+	if (pblk_submit_read_gc(pblk, lba_list, data, nr_secs,
+							&secs_to_gc, line)) {
+		ret = NVM_IO_ERR;
+		goto free_data;
+	}
+
+	if (!secs_to_gc)
+		goto free_data;
+
+	gc_rq = kmalloc(sizeof(struct pblk_gc_rq), GFP_KERNEL);
+	if (!gc_rq) {
+		ret = NVM_IO_ERR;
+		goto free_data;
+	}
+
+	gc_rq->line = line;
+	gc_rq->data = data;
+	gc_rq->lba_list = lba_list;
+	gc_rq->nr_secs = nr_secs;
+	gc_rq->secs_to_gc = secs_to_gc;
+
+	kref_get(&line->ref);
+
+retry:
+	spin_lock(&gc->w_lock);
+	if (gc->w_entries > 256) {
+		spin_unlock(&gc->w_lock);
+		usleep_range(256, 1024);
+		goto retry;
+	}
+	gc->w_entries++;
+	list_add_tail(&gc_rq->list, &gc->w_list);
+	spin_unlock(&gc->w_lock);
+
+	pblk_gc_writer_kick(&pblk->gc);
+
+	return NVM_IO_OK;
+
+free_data:
+	kfree(data);
+free_lba_list:
+	kfree(lba_list);
+
+	return ret;
+}
+
+static void pblk_put_line_back(struct pblk *pblk, struct pblk_line *line)
+{
+	struct pblk_line_mgmt *l_mg = &pblk->l_mg;
+	struct list_head *move_list;
+
+	spin_lock(&line->lock);
+	WARN_ON(line->state != PBLK_LINESTATE_GC);
+	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);
+		list_add_tail(&line->list, move_list);
+		spin_unlock(&l_mg->gc_lock);
+	}
+}
+
+static void pblk_gc_line_ws(struct work_struct *work)
+{
+	struct pblk_line_ws *line_ws = container_of(work, struct pblk_line_ws,
+									ws);
+	struct pblk *pblk = line_ws->pblk;
+	struct pblk_line_mgmt *l_mg = &pblk->l_mg;
+	struct pblk_line *line = line_ws->line;
+	struct pblk_line_meta *lm = &pblk->lm;
+	__le64 *lba_list = line_ws->priv;
+	u64 *gc_list;
+	int sec_left;
+	int nr_ppas, bit;
+	int put_line = 1;
+
+	pr_debug("pblk: line '%d' being reclaimed for GC\n", line->id);
+
+	spin_lock(&line->lock);
+	sec_left = line->vsc;
+	if (!sec_left) {
+		/* Lines are erased before being used (l_mg->data_/log_next) */
+		spin_unlock(&line->lock);
+		goto out;
+	}
+	spin_unlock(&line->lock);
+
+	if (sec_left < 0) {
+		pr_err("pblk: corrupted GC line (%d)\n", line->id);
+		put_line = 0;
+		pblk_put_line_back(pblk, line);
+		goto out;
+	}
+
+	bit = -1;
+next_rq:
+	gc_list = kmalloc_array(pblk->max_write_pgs, sizeof(u64), GFP_KERNEL);
+	if (!gc_list) {
+		put_line = 0;
+		pblk_put_line_back(pblk, line);
+		goto out;
+	}
+
+	nr_ppas = 0;
+	do {
+		bit = find_next_zero_bit(line->invalid_bitmap, lm->sec_per_line,
+								bit + 1);
+		if (bit > line->emeta_ssec)
+			break;
+
+		gc_list[nr_ppas++] = le64_to_cpu(lba_list[bit]);
+	} while (nr_ppas < pblk->max_write_pgs);
+
+	if (unlikely(!nr_ppas)) {
+		kfree(gc_list);
+		goto out;
+	}
+
+	if (pblk_gc_move_valid_secs(pblk, line, gc_list, nr_ppas)) {
+		pr_err("pblk: could not GC all sectors: line:%d (%d/%d/%d)\n",
+						line->id, line->vsc,
+						nr_ppas, nr_ppas);
+		put_line = 0;
+		pblk_put_line_back(pblk, line);
+		goto out;
+	}
+
+	sec_left -= nr_ppas;
+	if (sec_left > 0)
+		goto next_rq;
+
+out:
+	pblk_mfree(line->emeta, l_mg->emeta_alloc_type);
+	mempool_free(line_ws, pblk->line_ws_pool);
+	atomic_dec(&pblk->gc.inflight_gc);
+	if (put_line)
+		kref_put(&line->ref, pblk_line_put);
+}
+
+static int pblk_gc_line(struct pblk *pblk, struct pblk_line *line)
+{
+	struct pblk_line_mgmt *l_mg = &pblk->l_mg;
+	struct pblk_line_meta *lm = &pblk->lm;
+	struct pblk_line_ws *line_ws;
+	__le64 *lba_list;
+	int ret;
+
+	line_ws = mempool_alloc(pblk->line_ws_pool, GFP_KERNEL);
+	line->emeta = pblk_malloc(lm->emeta_len, l_mg->emeta_alloc_type,
+								GFP_KERNEL);
+	if (!line->emeta) {
+		pr_err("pblk: cannot use GC emeta\n");
+		goto fail_free_ws;
+	}
+
+	ret = pblk_line_read_emeta(pblk, line);
+	if (ret) {
+		pr_err("pblk: line %d read emeta failed (%d)\n", line->id, ret);
+		goto fail_free_emeta;
+	}
+
+	/* If this read fails, it means that emeta is corrupted. For now, leave
+	 * the line untouched. TODO: Implement a recovery routine that scans and
+	 * moves all sectors on the line.
+	 */
+	lba_list = pblk_recov_get_lba_list(pblk, line->emeta);
+	if (!lba_list) {
+		pr_err("pblk: could not interpret emeta (line %d)\n", line->id);
+		goto fail_free_emeta;
+	}
+
+	line_ws->pblk = pblk;
+	line_ws->line = line;
+	line_ws->priv = lba_list;
+
+	INIT_WORK(&line_ws->ws, pblk_gc_line_ws);
+	queue_work(pblk->gc.gc_reader_wq, &line_ws->ws);
+
+	return 0;
+
+fail_free_emeta:
+	pblk_mfree(line->emeta, l_mg->emeta_alloc_type);
+fail_free_ws:
+	mempool_free(line_ws, pblk->line_ws_pool);
+	pblk_put_line_back(pblk, line);
+
+	return 1;
+}
+
+static void pblk_gc_lines(struct pblk *pblk, struct list_head *gc_list)
+{
+	struct pblk_line *line, *tline;
+
+	list_for_each_entry_safe(line, tline, gc_list, list) {
+		if (pblk_gc_line(pblk, line))
+			pr_err("pblk: failed to GC line %d\n", line->id);
+		list_del(&line->list);
+	}
+}
+
+/*
+ * Lines with no valid sectors will be returned to the free list immediately. If
+ * GC is activated - either because the free block count is under the determined
+ * threshold, or because it is being forced from user space - only lines with a
+ * high count of invalid sectors will be recycled.
+ */
+static void pblk_gc_run(struct pblk *pblk)
+{
+	struct pblk_line_mgmt *l_mg = &pblk->l_mg;
+	struct pblk_gc *gc = &pblk->gc;
+	struct pblk_line *line, *tline;
+	unsigned int nr_blocks_free, nr_blocks_need;
+	struct list_head *group_list;
+	int run_gc, gc_group = 0;
+	int prev_gc = 0;
+	int inflight_gc = atomic_read(&gc->inflight_gc);
+	LIST_HEAD(gc_list);
+
+	spin_lock(&l_mg->gc_lock);
+	list_for_each_entry_safe(line, tline, &l_mg->gc_full_list, list) {
+		spin_lock(&line->lock);
+		WARN_ON(line->state != PBLK_LINESTATE_CLOSED);
+		line->state = PBLK_LINESTATE_GC;
+		spin_unlock(&line->lock);
+
+		list_del(&line->list);
+		kref_put(&line->ref, pblk_line_put);
+	}
+	spin_unlock(&l_mg->gc_lock);
+
+	nr_blocks_need = pblk_rl_gc_thrs(&pblk->rl);
+	nr_blocks_free = pblk_rl_nr_free_blks(&pblk->rl);
+	run_gc = (nr_blocks_need > nr_blocks_free || gc->gc_forced);
+
+next_gc_group:
+	group_list = l_mg->gc_lists[gc_group++];
+	spin_lock(&l_mg->gc_lock);
+	while (run_gc && !list_empty(group_list)) {
+		/* No need to queue up more GC lines than we can handle */
+		if (!run_gc || inflight_gc > gc->gc_jobs_active) {
+			spin_unlock(&l_mg->gc_lock);
+			pblk_gc_lines(pblk, &gc_list);
+			return;
+		}
+
+		line = list_first_entry(group_list, struct pblk_line, list);
+		nr_blocks_free += line->blk_in_line;
+
+		spin_lock(&line->lock);
+		WARN_ON(line->state != PBLK_LINESTATE_CLOSED);
+		line->state = PBLK_LINESTATE_GC;
+		list_move_tail(&line->list, &gc_list);
+		atomic_inc(&gc->inflight_gc);
+		inflight_gc++;
+		spin_unlock(&line->lock);
+
+		prev_gc = 1;
+		run_gc = (nr_blocks_need > nr_blocks_free || gc->gc_forced);
+	}
+	spin_unlock(&l_mg->gc_lock);
+
+	pblk_gc_lines(pblk, &gc_list);
+
+	if (!prev_gc && pblk->rl.rb_state > gc_group &&
+						gc_group < PBLK_NR_GC_LISTS)
+		goto next_gc_group;
+}
+
+
+static void pblk_gc_kick(struct pblk *pblk)
+{
+	struct pblk_gc *gc = &pblk->gc;
+
+	wake_up_process(gc->gc_ts);
+	pblk_gc_writer_kick(gc);
+	mod_timer(&gc->gc_timer, jiffies + msecs_to_jiffies(GC_TIME_MSECS));
+}
+
+static void pblk_gc_timer(unsigned long data)
+{
+	struct pblk *pblk = (struct pblk *)data;
+
+	pblk_gc_kick(pblk);
+}
+
+static int pblk_gc_ts(void *data)
+{
+	struct pblk *pblk = data;
+
+	while (!kthread_should_stop()) {
+		pblk_gc_run(pblk);
+		set_current_state(TASK_INTERRUPTIBLE);
+		io_schedule();
+	}
+
+	return 0;
+}
+
+static int pblk_gc_writer_ts(void *data)
+{
+	struct pblk *pblk = data;
+
+	while (!kthread_should_stop()) {
+		if (!pblk_gc_write(pblk))
+			continue;
+		set_current_state(TASK_INTERRUPTIBLE);
+		io_schedule();
+	}
+
+	return 0;
+}
+
+static void pblk_gc_start(struct pblk *pblk)
+{
+	pblk->gc.gc_active = 1;
+
+	pr_debug("pblk: gc start\n");
+}
+
+int pblk_gc_status(struct pblk *pblk)
+{
+	struct pblk_gc *gc = &pblk->gc;
+	int ret;
+
+	spin_lock(&gc->lock);
+	ret = gc->gc_active;
+	spin_unlock(&gc->lock);
+
+	return ret;
+}
+
+static void __pblk_gc_should_start(struct pblk *pblk)
+{
+	struct pblk_gc *gc = &pblk->gc;
+
+	lockdep_assert_held(&gc->lock);
+
+	if (gc->gc_enabled && !gc->gc_active)
+		pblk_gc_start(pblk);
+}
+
+void pblk_gc_should_start(struct pblk *pblk)
+{
+	struct pblk_gc *gc = &pblk->gc;
+
+	spin_lock(&gc->lock);
+	__pblk_gc_should_start(pblk);
+	spin_unlock(&gc->lock);
+}
+
+/*
+ * If flush_wq == 1 then no lock should be held by the caller since
+ * flush_workqueue can sleep
+ */
+static void pblk_gc_stop(struct pblk *pblk, int flush_wq)
+{
+	spin_lock(&pblk->gc.lock);
+	pblk->gc.gc_active = 0;
+	spin_unlock(&pblk->gc.lock);
+
+	pr_debug("pblk: gc stop\n");
+}
+
+void pblk_gc_should_stop(struct pblk *pblk)
+{
+	struct pblk_gc *gc = &pblk->gc;
+
+	if (gc->gc_active && !gc->gc_forced)
+		pblk_gc_stop(pblk, 0);
+}
+
+void pblk_gc_sysfs_state_show(struct pblk *pblk, int *gc_enabled,
+			      int *gc_active)
+{
+	struct pblk_gc *gc = &pblk->gc;
+
+	spin_lock(&gc->lock);
+	*gc_enabled = gc->gc_enabled;
+	*gc_active = gc->gc_active;
+	spin_unlock(&gc->lock);
+}
+
+void pblk_gc_sysfs_force(struct pblk *pblk, int force)
+{
+	struct pblk_gc *gc = &pblk->gc;
+	int rsv = 0;
+
+	spin_lock(&gc->lock);
+	if (force) {
+		gc->gc_enabled = 1;
+		rsv = 64;
+	}
+	pblk_rl_set_gc_rsc(&pblk->rl, rsv);
+	gc->gc_forced = force;
+	__pblk_gc_should_start(pblk);
+	spin_unlock(&gc->lock);
+}
+
+int pblk_gc_init(struct pblk *pblk)
+{
+	struct pblk_gc *gc = &pblk->gc;
+	int ret;
+
+	gc->gc_ts = kthread_create(pblk_gc_ts, pblk, "pblk-gc-ts");
+	if (IS_ERR(gc->gc_ts)) {
+		pr_err("pblk: could not allocate GC main kthread\n");
+		return PTR_ERR(gc->gc_ts);
+	}
+
+	gc->gc_writer_ts = kthread_create(pblk_gc_writer_ts, pblk,
+							"pblk-gc-writer-ts");
+	if (IS_ERR(gc->gc_writer_ts)) {
+		pr_err("pblk: could not allocate GC writer kthread\n");
+		ret = PTR_ERR(gc->gc_writer_ts);
+		goto fail_free_main_kthread;
+	}
+
+	setup_timer(&gc->gc_timer, pblk_gc_timer, (unsigned long)pblk);
+	mod_timer(&gc->gc_timer, jiffies + msecs_to_jiffies(GC_TIME_MSECS));
+
+	gc->gc_active = 0;
+	gc->gc_forced = 0;
+	gc->gc_enabled = 1;
+	gc->gc_jobs_active = 8;
+	gc->w_entries = 0;
+	atomic_set(&gc->inflight_gc, 0);
+
+	gc->gc_reader_wq = alloc_workqueue("pblk-gc-reader-wq",
+			WQ_MEM_RECLAIM | WQ_UNBOUND, gc->gc_jobs_active);
+	if (!gc->gc_reader_wq) {
+		pr_err("pblk: could not allocate GC reader workqueue\n");
+		ret = -ENOMEM;
+		goto fail_free_writer_kthread;
+	}
+
+	spin_lock_init(&gc->lock);
+	spin_lock_init(&gc->w_lock);
+	INIT_LIST_HEAD(&gc->w_list);
+
+	return 0;
+
+fail_free_main_kthread:
+	kthread_stop(gc->gc_ts);
+fail_free_writer_kthread:
+	kthread_stop(gc->gc_writer_ts);
+
+	return ret;
+}
+
+void pblk_gc_exit(struct pblk *pblk)
+{
+	struct pblk_gc *gc = &pblk->gc;
+
+	flush_workqueue(gc->gc_reader_wq);
+
+	del_timer(&gc->gc_timer);
+	pblk_gc_stop(pblk, 1);
+
+	if (gc->gc_ts)
+		kthread_stop(gc->gc_ts);
+
+	if (pblk->gc.gc_reader_wq)
+		destroy_workqueue(pblk->gc.gc_reader_wq);
+
+	if (gc->gc_writer_ts)
+		kthread_stop(gc->gc_writer_ts);
+}

drivers/lightnvm/pblk-map.c

+/*
+ * 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-map.c - pblk's lba-ppa mapping strategy
+ *
+ */
+
+#include "pblk.h"
+
+static void pblk_map_page_data(struct pblk *pblk, unsigned int sentry,
+			       struct ppa_addr *ppa_list,
+			       unsigned long *lun_bitmap,
+			       struct pblk_sec_meta *meta_list,
+			       unsigned int valid_secs)
+{
+	struct pblk_line *line = pblk_line_get_data(pblk);
+	struct line_emeta *emeta = line->emeta;
+	struct pblk_w_ctx *w_ctx;
+	__le64 *lba_list = pblk_line_emeta_to_lbas(emeta);
+	u64 paddr;
+	int nr_secs = pblk->min_write_pgs;
+	int i;
+
+	paddr = pblk_alloc_page(pblk, line, nr_secs);
+
+	for (i = 0; i < nr_secs; i++, paddr++) {
+		/* ppa to be sent to the device */
+		ppa_list[i] = addr_to_gen_ppa(pblk, paddr, line->id);
+
+		/* Write context for target bio completion on write buffer. Note
+		 * that the write buffer is protected by the sync backpointer,
+		 * and a single writer thread have access to each specific entry
+		 * at a time. Thus, it is safe to modify the context for the
+		 * entry we are setting up for submission without taking any
+		 * lock or memory barrier.
+		 */
+		if (i < valid_secs) {
+			kref_get(&line->ref);
+			w_ctx = pblk_rb_w_ctx(&pblk->rwb, sentry + i);
+			w_ctx->ppa = ppa_list[i];
+			meta_list[i].lba = cpu_to_le64(w_ctx->lba);
+			lba_list[paddr] = cpu_to_le64(w_ctx->lba);
+			le64_add_cpu(&line->emeta->nr_valid_lbas, 1);
+		} else {
+			meta_list[i].lba = cpu_to_le64(ADDR_EMPTY);
+			lba_list[paddr] = cpu_to_le64(ADDR_EMPTY);
+			pblk_map_pad_invalidate(pblk, line, paddr);
+		}
+	}
+
+	if (pblk_line_is_full(line)) {
+		line = pblk_line_replace_data(pblk);
+		if (!line)
+			return;
+	}
+
+	pblk_down_rq(pblk, ppa_list, nr_secs, lun_bitmap);
+}
+
+void pblk_map_rq(struct pblk *pblk, struct nvm_rq *rqd, unsigned int sentry,
+		 unsigned long *lun_bitmap, unsigned int valid_secs,
+		 unsigned int off)
+{
+	struct pblk_sec_meta *meta_list = rqd->meta_list;
+	unsigned int map_secs;
+	int min = pblk->min_write_pgs;
+	int i;
+
+	for (i = off; i < rqd->nr_ppas; i += min) {
+		map_secs = (i + min > valid_secs) ? (valid_secs % min) : min;
+		pblk_map_page_data(pblk, sentry + i, &rqd->ppa_list[i],
+					lun_bitmap, &meta_list[i], map_secs);
+	}
+}
+
+/* only if erase_ppa is set, acquire erase semaphore */
+void pblk_map_erase_rq(struct pblk *pblk, struct nvm_rq *rqd,
+		       unsigned int sentry, unsigned long *lun_bitmap,
+		       unsigned int valid_secs, struct ppa_addr *erase_ppa)
+{
+	struct nvm_tgt_dev *dev = pblk->dev;
+	struct nvm_geo *geo = &dev->geo;
+	struct pblk_line *e_line = pblk_line_get_data_next(pblk);
+	struct pblk_sec_meta *meta_list = rqd->meta_list;
+	unsigned int map_secs;
+	int min = pblk->min_write_pgs;
+	int i, erase_lun;
+
+	for (i = 0; i < rqd->nr_ppas; i += min) {
+		map_secs = (i + min > valid_secs) ? (valid_secs % min) : min;
+		pblk_map_page_data(pblk, sentry + i, &rqd->ppa_list[i],
+					lun_bitmap, &meta_list[i], map_secs);
+
+		erase_lun = rqd->ppa_list[i].g.lun * geo->nr_chnls +
+							rqd->ppa_list[i].g.ch;
+
+		if (!test_bit(erase_lun, e_line->erase_bitmap)) {
+			if (down_trylock(&pblk->erase_sem))
+				continue;
+
+			set_bit(erase_lun, e_line->erase_bitmap);
+			e_line->left_eblks--;
+			*erase_ppa = rqd->ppa_list[i];
+			erase_ppa->g.blk = e_line->id;
+
+			/* Avoid evaluating e_line->left_eblks */
+			return pblk_map_rq(pblk, rqd, sentry, lun_bitmap,
+							valid_secs, i + min);
+		}
+	}
+
+	/* Erase blocks that are bad in this line but might not be in next */
+	if (unlikely(ppa_empty(*erase_ppa))) {
+		struct pblk_line_meta *lm = &pblk->lm;
+
+		i = find_first_zero_bit(e_line->erase_bitmap, lm->blk_per_line);
+		if (i == lm->blk_per_line)
+			return;
+
+		set_bit(i, e_line->erase_bitmap);
+		e_line->left_eblks--;
+		*erase_ppa = pblk->luns[i].bppa; /* set ch and lun */
+		erase_ppa->g.blk = e_line->id;
+	}
+}

drivers/lightnvm/pblk-read.c

+/*
+ * 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-read.c - pblk's read path
+ */
+
+#include "pblk.h"
+
+/*
+ * There is no guarantee that the value read from cache has not been updated and
+ * resides at another location in the cache. We guarantee though that if the
+ * value is read from the cache, it belongs to the mapped lba. In order to
+ * guarantee and order between writes and reads are ordered, a flush must be
+ * issued.
+ */
+static int pblk_read_from_cache(struct pblk *pblk, struct bio *bio,
+				sector_t lba, struct ppa_addr ppa,
+				int bio_iter)
+{
+#ifdef CONFIG_NVM_DEBUG
+	/* Callers must ensure that the ppa points to a cache address */
+	BUG_ON(pblk_ppa_empty(ppa));
+	BUG_ON(!pblk_addr_in_cache(ppa));
+#endif
+
+	return pblk_rb_copy_to_bio(&pblk->rwb, bio, lba,
+					pblk_addr_to_cacheline(ppa), bio_iter);
+}
+
+static void pblk_read_ppalist_rq(struct pblk *pblk, struct nvm_rq *rqd,
+				 unsigned long *read_bitmap)
+{
+	struct bio *bio = rqd->bio;
+	struct ppa_addr ppas[PBLK_MAX_REQ_ADDRS];
+	sector_t blba = pblk_get_lba(bio);
+	int nr_secs = rqd->nr_ppas;
+	int advanced_bio = 0;
+	int i, j = 0;
+
+	/* logic error: lba out-of-bounds. Ignore read request */
+	if (!(blba + nr_secs < pblk->rl.nr_secs)) {
+		WARN_ON("pblk: read lbas out of bounds\n");
+		return;
+	}
+
+	pblk_lookup_l2p_seq(pblk, ppas, blba, nr_secs);
+
+	for (i = 0; i < nr_secs; i++) {
+		struct ppa_addr p = ppas[i];
+		sector_t lba = blba + i;
+
+retry:
+		if (pblk_ppa_empty(p)) {
+			WARN_ON(test_and_set_bit(i, read_bitmap));
+			continue;
+		}
+
+		/* Try to read from write buffer. The address is later checked
+		 * on the write buffer to prevent retrieving overwritten data.
+		 */
+		if (pblk_addr_in_cache(p)) {
+			if (!pblk_read_from_cache(pblk, bio, lba, p, i)) {
+				pblk_lookup_l2p_seq(pblk, &p, lba, 1);
+				goto retry;
+			}
+			WARN_ON(test_and_set_bit(i, read_bitmap));
+			advanced_bio = 1;
+		} else {
+			/* Read from media non-cached sectors */
+			rqd->ppa_list[j++] = p;
+		}
+
+		if (advanced_bio)
+			bio_advance(bio, PBLK_EXPOSED_PAGE_SIZE);
+	}
+
+#ifdef CONFIG_NVM_DEBUG
+	atomic_long_add(nr_secs, &pblk->inflight_reads);
+#endif
+}
+
+static int pblk_submit_read_io(struct pblk *pblk, struct nvm_rq *rqd)
+{
+	int err;
+
+	rqd->flags = pblk_set_read_mode(pblk);
+
+	err = pblk_submit_io(pblk, rqd);
+	if (err)
+		return NVM_IO_ERR;
+
+	return NVM_IO_OK;
+}
+
+static void pblk_end_io_read(struct nvm_rq *rqd)
+{
+	struct pblk *pblk = rqd->private;
+	struct nvm_tgt_dev *dev = pblk->dev;
+	struct pblk_r_ctx *r_ctx = nvm_rq_to_pdu(rqd);
+	struct bio *bio = rqd->bio;
+
+	if (rqd->error)
+		pblk_log_read_err(pblk, rqd);
+#ifdef CONFIG_NVM_DEBUG
+	else
+		WARN_ONCE(bio->bi_error, "pblk: corrupted read error\n");
+#endif
+
+	if (rqd->nr_ppas > 1)
+		nvm_dev_dma_free(dev->parent, rqd->ppa_list, rqd->dma_ppa_list);
+
+	bio_put(bio);
+	if (r_ctx->orig_bio) {
+#ifdef CONFIG_NVM_DEBUG
+		WARN_ONCE(r_ctx->orig_bio->bi_error,
+						"pblk: corrupted read bio\n");
+#endif
+		bio_endio(r_ctx->orig_bio);
+		bio_put(r_ctx->orig_bio);
+	}
+
+#ifdef CONFIG_NVM_DEBUG
+	atomic_long_add(rqd->nr_ppas, &pblk->sync_reads);
+	atomic_long_sub(rqd->nr_ppas, &pblk->inflight_reads);
+#endif
+
+	pblk_free_rqd(pblk, rqd, READ);
+}
+
+static int pblk_fill_partial_read_bio(struct pblk *pblk, struct nvm_rq *rqd,
+				      unsigned int bio_init_idx,
+				      unsigned long *read_bitmap)
+{
+	struct bio *new_bio, *bio = rqd->bio;
+	struct bio_vec src_bv, dst_bv;
+	void *ppa_ptr = NULL;
+	void *src_p, *dst_p;
+	dma_addr_t dma_ppa_list = 0;
+	int nr_secs = rqd->nr_ppas;
+	int nr_holes = nr_secs - bitmap_weight(read_bitmap, nr_secs);
+	int i, ret, hole;
+	DECLARE_COMPLETION_ONSTACK(wait);
+
+	new_bio = bio_alloc(GFP_KERNEL, nr_holes);
+	if (!new_bio) {
+		pr_err("pblk: could not alloc read bio\n");
+		return NVM_IO_ERR;
+	}
+
+	if (pblk_bio_add_pages(pblk, new_bio, GFP_KERNEL, nr_holes))
+		goto err;
+
+	if (nr_holes != new_bio->bi_vcnt) {
+		pr_err("pblk: malformed bio\n");
+		goto err;
+	}
+
+	new_bio->bi_iter.bi_sector = 0; /* internal bio */
+	bio_set_op_attrs(new_bio, REQ_OP_READ, 0);
+	new_bio->bi_private = &wait;
+	new_bio->bi_end_io = pblk_end_bio_sync;
+
+	rqd->bio = new_bio;
+	rqd->nr_ppas = nr_holes;
+	rqd->end_io = NULL;
+
+	if (unlikely(nr_secs > 1 && nr_holes == 1)) {
+		ppa_ptr = rqd->ppa_list;
+		dma_ppa_list = rqd->dma_ppa_list;
+		rqd->ppa_addr = rqd->ppa_list[0];
+	}
+
+	ret = pblk_submit_read_io(pblk, rqd);
+	if (ret) {
+		bio_put(rqd->bio);
+		pr_err("pblk: read IO submission failed\n");
+		goto err;
+	}
+
+	if (!wait_for_completion_io_timeout(&wait,
+				msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
+		pr_err("pblk: partial read I/O timed out\n");
+	}
+
+	if (rqd->error) {
+		atomic_long_inc(&pblk->read_failed);
+#ifdef CONFIG_NVM_DEBUG
+		pblk_print_failed_rqd(pblk, rqd, rqd->error);
+#endif
+	}
+
+	if (unlikely(nr_secs > 1 && nr_holes == 1)) {
+		rqd->ppa_list = ppa_ptr;
+		rqd->dma_ppa_list = dma_ppa_list;
+	}
+
+	/* Fill the holes in the original bio */
+	i = 0;
+	hole = find_first_zero_bit(read_bitmap, nr_secs);
+	do {
+		src_bv = new_bio->bi_io_vec[i++];
+		dst_bv = bio->bi_io_vec[bio_init_idx + hole];
+
+		src_p = kmap_atomic(src_bv.bv_page);
+		dst_p = kmap_atomic(dst_bv.bv_page);
+
+		memcpy(dst_p + dst_bv.bv_offset,
+			src_p + src_bv.bv_offset,
+			PBLK_EXPOSED_PAGE_SIZE);
+
+		kunmap_atomic(src_p);
+		kunmap_atomic(dst_p);
+
+		mempool_free(src_bv.bv_page, pblk->page_pool);
+
+		hole = find_next_zero_bit(read_bitmap, nr_secs, hole + 1);
+	} while (hole < nr_secs);
+
+	bio_put(new_bio);
+
+	/* Complete the original bio and associated request */
+	rqd->bio = bio;
+	rqd->nr_ppas = nr_secs;
+	rqd->private = pblk;
+
+	bio_endio(bio);
+	pblk_end_io_read(rqd);
+	return NVM_IO_OK;
+
+err:
+	/* Free allocated pages in new bio */
+	pblk_bio_free_pages(pblk, bio, 0, new_bio->bi_vcnt);
+	rqd->private = pblk;
+	pblk_end_io_read(rqd);
+	return NVM_IO_ERR;
+}
+
+static void pblk_read_rq(struct pblk *pblk, struct nvm_rq *rqd,
+			 unsigned long *read_bitmap)
+{
+	struct bio *bio = rqd->bio;
+	struct ppa_addr ppa;
+	sector_t lba = pblk_get_lba(bio);
+
+	/* logic error: lba out-of-bounds. Ignore read request */
+	if (!(lba < pblk->rl.nr_secs)) {
+		WARN_ON("pblk: read lba out of bounds\n");
+		return;
+	}
+
+	pblk_lookup_l2p_seq(pblk, &ppa, lba, 1);
+
+#ifdef CONFIG_NVM_DEBUG
+	atomic_long_inc(&pblk->inflight_reads);
+#endif
+
+retry:
+	if (pblk_ppa_empty(ppa)) {
+		WARN_ON(test_and_set_bit(0, read_bitmap));
+		return;
+	}
+
+	/* Try to read from write buffer. The address is later checked on the
+	 * write buffer to prevent retrieving overwritten data.
+	 */
+	if (pblk_addr_in_cache(ppa)) {
+		if (!pblk_read_from_cache(pblk, bio, lba, ppa, 0)) {
+			pblk_lookup_l2p_seq(pblk, &ppa, lba, 1);
+			goto retry;
+		}
+		WARN_ON(test_and_set_bit(0, read_bitmap));
+	} else {
+		rqd->ppa_addr = ppa;
+	}
+}
+
+int pblk_submit_read(struct pblk *pblk, struct bio *bio)
+{
+	struct nvm_tgt_dev *dev = pblk->dev;
+	int nr_secs = pblk_get_secs(bio);
+	struct nvm_rq *rqd;
+	unsigned long read_bitmap; /* Max 64 ppas per request */
+	unsigned int bio_init_idx;
+	int ret = NVM_IO_ERR;
+
+	if (nr_secs > PBLK_MAX_REQ_ADDRS)
+		return NVM_IO_ERR;
+
+	bitmap_zero(&read_bitmap, nr_secs);
+
+	rqd = pblk_alloc_rqd(pblk, READ);
+	if (IS_ERR(rqd)) {
+		pr_err_ratelimited("pblk: not able to alloc rqd");
+		return NVM_IO_ERR;
+	}
+
+	rqd->opcode = NVM_OP_PREAD;
+	rqd->bio = bio;
+	rqd->nr_ppas = nr_secs;
+	rqd->private = pblk;
+	rqd->end_io = pblk_end_io_read;
+
+	/* Save the index for this bio's start. This is needed in case
+	 * we need to fill a partial read.
+	 */
+	bio_init_idx = pblk_get_bi_idx(bio);
+
+	if (nr_secs > 1) {
+		rqd->ppa_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
+						&rqd->dma_ppa_list);
+		if (!rqd->ppa_list) {
+			pr_err("pblk: not able to allocate ppa list\n");
+			goto fail_rqd_free;
+		}
+
+		pblk_read_ppalist_rq(pblk, rqd, &read_bitmap);
+	} else {
+		pblk_read_rq(pblk, rqd, &read_bitmap);
+	}
+
+	bio_get(bio);
+	if (bitmap_full(&read_bitmap, nr_secs)) {
+		bio_endio(bio);
+		pblk_end_io_read(rqd);
+		return NVM_IO_OK;
+	}
+
+	/* All sectors are to be read from the device */
+	if (bitmap_empty(&read_bitmap, rqd->nr_ppas)) {
+		struct bio *int_bio = NULL;
+		struct pblk_r_ctx *r_ctx = nvm_rq_to_pdu(rqd);
+
+		/* Clone read bio to deal with read errors internally */
+		int_bio = bio_clone_bioset(bio, GFP_KERNEL, fs_bio_set);
+		if (!int_bio) {
+			pr_err("pblk: could not clone read bio\n");
+			return NVM_IO_ERR;
+		}
+
+		rqd->bio = int_bio;
+		r_ctx->orig_bio = bio;
+
+		ret = pblk_submit_read_io(pblk, rqd);
+		if (ret) {
+			pr_err("pblk: read IO submission failed\n");
+			if (int_bio)
+				bio_put(int_bio);
+			return ret;
+		}
+
+		return NVM_IO_OK;
+	}
+
+	/* The read bio request could be partially filled by the write buffer,
+	 * but there are some holes that need to be read from the drive.
+	 */
+	ret = pblk_fill_partial_read_bio(pblk, rqd, bio_init_idx, &read_bitmap);
+	if (ret) {
+		pr_err("pblk: failed to perform partial read\n");
+		return ret;
+	}
+
+	return NVM_IO_OK;
+
+fail_rqd_free:
+	pblk_free_rqd(pblk, rqd, READ);
+	return ret;
+}
+
+static int read_ppalist_rq_gc(struct pblk *pblk, struct nvm_rq *rqd,
+			      struct pblk_line *line, u64 *lba_list,
+			      unsigned int nr_secs)
+{
+	struct ppa_addr ppas[PBLK_MAX_REQ_ADDRS];
+	int valid_secs = 0;
+	int i;
+
+	pblk_lookup_l2p_rand(pblk, ppas, lba_list, nr_secs);
+
+	for (i = 0; i < nr_secs; i++) {
+		if (pblk_addr_in_cache(ppas[i]) || ppas[i].g.blk != line->id ||
+						pblk_ppa_empty(ppas[i])) {
+			lba_list[i] = ADDR_EMPTY;
+			continue;
+		}
+
+		rqd->ppa_list[valid_secs++] = ppas[i];
+	}
+
+#ifdef CONFIG_NVM_DEBUG
+	atomic_long_add(valid_secs, &pblk->inflight_reads);
+#endif
+	return valid_secs;
+}
+
+static int read_rq_gc(struct pblk *pblk, struct nvm_rq *rqd,
+		      struct pblk_line *line, sector_t lba)
+{
+	struct ppa_addr ppa;
+	int valid_secs = 0;
+
+	/* logic error: lba out-of-bounds */
+	if (!(lba < pblk->rl.nr_secs)) {
+		WARN_ON("pblk: read lba out of bounds\n");
+		goto out;
+	}
+
+	if (lba == ADDR_EMPTY)
+		goto out;
+
+	spin_lock(&pblk->trans_lock);
+	ppa = pblk_trans_map_get(pblk, lba);
+	spin_unlock(&pblk->trans_lock);
+
+	/* Ignore updated values until the moment */
+	if (pblk_addr_in_cache(ppa) || ppa.g.blk != line->id ||
+							pblk_ppa_empty(ppa))
+		goto out;
+
+	rqd->ppa_addr = ppa;
+	valid_secs = 1;
+
+#ifdef CONFIG_NVM_DEBUG
+	atomic_long_inc(&pblk->inflight_reads);
+#endif
+
+out:
+	return valid_secs;
+}
+
+int pblk_submit_read_gc(struct pblk *pblk, u64 *lba_list, void *data,
+			unsigned int nr_secs, unsigned int *secs_to_gc,
+			struct pblk_line *line)
+{
+	struct nvm_tgt_dev *dev = pblk->dev;
+	struct nvm_geo *geo = &dev->geo;
+	struct request_queue *q = dev->q;
+	struct bio *bio;
+	struct nvm_rq rqd;
+	int ret, data_len;
+	DECLARE_COMPLETION_ONSTACK(wait);
+
+	memset(&rqd, 0, sizeof(struct nvm_rq));
+
+	if (nr_secs > 1) {
+		rqd.ppa_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
+							&rqd.dma_ppa_list);
+		if (!rqd.ppa_list)
+			return NVM_IO_ERR;
+
+		*secs_to_gc = read_ppalist_rq_gc(pblk, &rqd, line, lba_list,
+								nr_secs);
+		if (*secs_to_gc == 1) {
+			struct ppa_addr ppa;
+
+			ppa = rqd.ppa_list[0];
+			nvm_dev_dma_free(dev->parent, rqd.ppa_list,
+							rqd.dma_ppa_list);
+			rqd.ppa_addr = ppa;
+		}
+	} else {
+		*secs_to_gc = read_rq_gc(pblk, &rqd, line, lba_list[0]);
+	}
+
+	if (!(*secs_to_gc))
+		goto out;
+
+	data_len = (*secs_to_gc) * geo->sec_size;
+	bio = bio_map_kern(q, data, data_len, GFP_KERNEL);
+	if (IS_ERR(bio)) {
+		pr_err("pblk: could not allocate GC bio (%lu)\n", PTR_ERR(bio));
+		goto err_free_dma;
+	}
+
+	bio->bi_iter.bi_sector = 0; /* internal bio */
+	bio_set_op_attrs(bio, REQ_OP_READ, 0);
+
+	rqd.opcode = NVM_OP_PREAD;
+	rqd.end_io = pblk_end_io_sync;
+	rqd.private = &wait;
+	rqd.nr_ppas = *secs_to_gc;
+	rqd.bio = bio;
+
+	ret = pblk_submit_read_io(pblk, &rqd);
+	if (ret) {
+		bio_endio(bio);
+		pr_err("pblk: GC read request failed\n");
+		goto err_free_dma;
+	}
+
+	if (!wait_for_completion_io_timeout(&wait,
+				msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
+		pr_err("pblk: GC read I/O timed out\n");
+	}
+
+	if (rqd.error) {
+		atomic_long_inc(&pblk->read_failed_gc);
+#ifdef CONFIG_NVM_DEBUG
+		pblk_print_failed_rqd(pblk, &rqd, rqd.error);
+#endif
+	}
+
+#ifdef CONFIG_NVM_DEBUG
+	atomic_long_add(*secs_to_gc, &pblk->sync_reads);
+	atomic_long_add(*secs_to_gc, &pblk->recov_gc_reads);
+	atomic_long_sub(*secs_to_gc, &pblk->inflight_reads);
+#endif
+
+out:
+	if (rqd.nr_ppas > 1)
+		nvm_dev_dma_free(dev->parent, rqd.ppa_list, rqd.dma_ppa_list);
+	return NVM_IO_OK;
+
+err_free_dma:
+	if (rqd.nr_ppas > 1)
+		nvm_dev_dma_free(dev->parent, rqd.ppa_list, rqd.dma_ppa_list);
+	return NVM_IO_ERR;
+}