[PATCH 1/7] lightnvm: fix missing grown bad block type

[PATCH 1/7] lightnvm: fix missing grown bad block type

[Matias Bjørling]

The get/set bad block interface defines good block, factory bad block,
grown bad block, device reserved block, and host reserved block.
Unfortunately the grown bad block was missing, leaving the offsets wrong
for device and host side reserved blocks.

This patch adds the missing type and corrects the offsets.

Signed-off-by: Matias Bjørling <m@bjorling.me>
---
 include/linux/lightnvm.h | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/include/linux/lightnvm.h b/include/linux/lightnvm.h
index 96c8ad1..aa35907 100644
--- a/include/linux/lightnvm.h
+++ b/include/linux/lightnvm.h
@@ -58,8 +58,9 @@ enum {
 	/* Block Types */
 	NVM_BLK_T_FREE		= 0x0,
 	NVM_BLK_T_BAD		= 0x1,
-	NVM_BLK_T_DEV		= 0x2,
-	NVM_BLK_T_HOST		= 0x4,
+	NVM_BLK_T_GRWN_BAD	= 0x2,
+	NVM_BLK_T_DEV		= 0x4,
+	NVM_BLK_T_HOST		= 0x8,
 };
 
 struct nvm_id_group {
-- 
2.1.4

[PATCH 2/7] lightnvm: add mccap support

[Matias Bjørling]

Some flash media has extended capabilities, such as programming SLC
pages on MLC/TLC flash, erase/program suspend, scramble and encryption.
MCCAP is introduced to detect support for these capabilities in the
command set.

Signed-off-by: Matias Bjørling <m@bjorling.me>
---
 drivers/lightnvm/core.c  | 1 +
 include/linux/lightnvm.h | 7 +++++++
 2 files changed, 8 insertions(+)

diff --git a/drivers/lightnvm/core.c b/drivers/lightnvm/core.c
index e5e3963..5199c12 100644
--- a/drivers/lightnvm/core.c
+++ b/drivers/lightnvm/core.c
@@ -376,6 +376,7 @@ static int nvm_core_init(struct nvm_dev *dev)
 	dev->sec_size = grp->csecs;
 	dev->oob_size = grp->sos;
 	dev->sec_per_pg = grp->fpg_sz / grp->csecs;
+	dev->mccap = grp->mccap;
 	memcpy(&dev->ppaf, &id->ppaf, sizeof(struct nvm_addr_format));
 
 	dev->plane_mode = NVM_PLANE_SINGLE;
diff --git a/include/linux/lightnvm.h b/include/linux/lightnvm.h
index aa35907..b90d283 100644
--- a/include/linux/lightnvm.h
+++ b/include/linux/lightnvm.h
@@ -61,6 +61,12 @@ enum {
 	NVM_BLK_T_GRWN_BAD	= 0x2,
 	NVM_BLK_T_DEV		= 0x4,
 	NVM_BLK_T_HOST		= 0x8,
+
+	/* Memory capabilities */
+	NVM_ID_CAP_SLC		= 0x1,
+	NVM_ID_CAP_CMD_SUSPEND	= 0x2,
+	NVM_ID_CAP_SCRAMBLE	= 0x4,
+	NVM_ID_CAP_ENCRYPT	= 0x8,
 };
 
 struct nvm_id_group {
@@ -278,6 +284,7 @@ struct nvm_dev {
 	int blks_per_lun;
 	int sec_size;
 	int oob_size;
+	int mccap;
 	struct nvm_addr_format ppaf;
 
 	/* Calculated/Cached values. These do not reflect the actual usable
-- 
2.1.4

[PATCH 3/7] lightnvm: introduce mlc lower page table mappings

[Matias Bjørling]

NAND MLC memories have both lower and upper pages. When programming,
both of these must be written, before data can be read. However,
these lower and upper pages might not placed at even and odd flash
pages, but can be skipped. Therefore each flash memory has its lower
pages defined, which can then be used when programming and to know when
padding are necessary.

This patch implements the lower page definition in the specification,
and exposes it through a simple lookup table at dev->lptbl.

Signed-off-by: Matias Bjørling <m@bjorling.me>
---
 drivers/lightnvm/core.c      | 61 +++++++++++++++++++++++++++++++++++++++++++-
 drivers/nvme/host/lightnvm.c | 22 +++++++++++++++-
 include/linux/lightnvm.h     | 20 +++++++++++++++
 3 files changed, 101 insertions(+), 2 deletions(-)

diff --git a/drivers/lightnvm/core.c b/drivers/lightnvm/core.c
index 5199c12..1f302cc 100644
--- a/drivers/lightnvm/core.c
+++ b/drivers/lightnvm/core.c
@@ -362,6 +362,51 @@ int nvm_submit_ppa(struct nvm_dev *dev, struct ppa_addr *ppa, int nr_ppas,
 }
 EXPORT_SYMBOL(nvm_submit_ppa);
 
+static int nvm_init_slc_tbl(struct nvm_dev *dev, struct nvm_id_group *grp)
+{
+	int i;
+
+	dev->lps_per_blk = dev->pgs_per_blk;
+	dev->lptbl = kcalloc(dev->lps_per_blk, sizeof(int), GFP_KERNEL);
+	if (!dev->lptbl)
+		return -ENOMEM;
+
+	/* Just a linear array */
+	for (i = 0; i < dev->lps_per_blk; i++)
+		dev->lptbl[i] = i;
+
+	return 0;
+}
+
+static int nvm_init_mlc_tbl(struct nvm_dev *dev, struct nvm_id_group *grp)
+{
+	int i, p;
+	struct nvm_id_lp_mlc *mlc = &grp->lptbl.mlc;
+
+	if (!mlc->num_pairs)
+		return 0;
+
+	dev->lps_per_blk = mlc->num_pairs;
+	dev->lptbl = kcalloc(dev->lps_per_blk, sizeof(int), GFP_KERNEL);
+	if (!dev->lptbl)
+		return -ENOMEM;
+
+	/* The lower page table encoding consists of a list of bytes, where each
+	 * has a lower and an upper half. The first half byte maintains the
+	 * increment value and every value after is an offset added to the
+	 * previous incrementation value */
+	dev->lptbl[0] = mlc->pairs[0] & 0xF;
+	for (i = 1; i < dev->lps_per_blk; i++) {
+		p = mlc->pairs[i >> 1];
+		if (i & 0x1) /* upper */
+			dev->lptbl[i] = dev->lptbl[i - 1] + ((p & 0xF0) >> 4);
+		else /* lower */
+			dev->lptbl[i] = dev->lptbl[i - 1] + (p & 0xF);
+	}
+
+	return 0;
+}
+
 static int nvm_core_init(struct nvm_dev *dev)
 {
 	struct nvm_id *id = &dev->identity;
@@ -387,11 +432,23 @@ static int nvm_core_init(struct nvm_dev *dev)
 		return -EINVAL;
 	}
 
-	if (grp->fmtype != 0 && grp->fmtype != 1) {
+	switch (grp->fmtype) {
+	case NVM_ID_FMTYPE_SLC:
+		if (nvm_init_slc_tbl(dev, grp))
+			return -ENOMEM;
+		break;
+	case NVM_ID_FMTYPE_MLC:
+		if (nvm_init_mlc_tbl(dev, grp))
+			return -ENOMEM;
+		break;
+	default:
 		pr_err("nvm: flash type not supported\n");
 		return -EINVAL;
 	}
 
+	if (!dev->lps_per_blk)
+		pr_info("nvm: lower page programming table missing\n");
+
 	if (grp->mpos & 0x020202)
 		dev->plane_mode = NVM_PLANE_DOUBLE;
 	if (grp->mpos & 0x040404)
@@ -420,6 +477,8 @@ static void nvm_free(struct nvm_dev *dev)
 
 	if (dev->mt)
 		dev->mt->unregister_mgr(dev);
+
+	kfree(dev->lptbl);
 }
 
 static int nvm_init(struct nvm_dev *dev)
diff --git a/drivers/nvme/host/lightnvm.c b/drivers/nvme/host/lightnvm.c
index b112f02..215c025 100644
--- a/drivers/nvme/host/lightnvm.c
+++ b/drivers/nvme/host/lightnvm.c
@@ -146,6 +146,16 @@ struct nvme_nvm_command {
 	};
 };
 
+struct nvme_nvm_lp_mlc {
+	__u16			num_pairs;
+	__u8			pairs[886];
+};
+
+struct nvme_nvm_lp_tbl {
+	__u8			id[8];
+	struct nvme_nvm_lp_mlc	mlc;
+};
+
 struct nvme_nvm_id_group {
 	__u8			mtype;
 	__u8			fmtype;
@@ -169,7 +179,8 @@ struct nvme_nvm_id_group {
 	__le32			mpos;
 	__le32			mccap;
 	__le16			cpar;
-	__u8			reserved[906];
+	__u8			reserved[10];
+	struct nvme_nvm_lp_tbl lptbl;
 } __packed;
 
 struct nvme_nvm_addr_format {
@@ -266,6 +277,15 @@ static int init_grps(struct nvm_id *nvm_id, struct nvme_nvm_id *nvme_nvm_id)
 		dst->mccap = le32_to_cpu(src->mccap);
 
 		dst->cpar = le16_to_cpu(src->cpar);
+
+		if (dst->fmtype == NVM_ID_FMTYPE_MLC) {
+			memcpy(dst->lptbl.id, src->lptbl.id, 8);
+			dst->lptbl.mlc.num_pairs =
+					le16_to_cpu(src->lptbl.mlc.num_pairs);
+			/* 4 bits per pair */
+			memcpy(dst->lptbl.mlc.pairs, src->lptbl.mlc.pairs,
+						dst->lptbl.mlc.num_pairs >> 1);
+		}
 	}
 
 	return 0;
diff --git a/include/linux/lightnvm.h b/include/linux/lightnvm.h
index b90d283..678fd91 100644
--- a/include/linux/lightnvm.h
+++ b/include/linux/lightnvm.h
@@ -67,6 +67,20 @@ enum {
 	NVM_ID_CAP_CMD_SUSPEND	= 0x2,
 	NVM_ID_CAP_SCRAMBLE	= 0x4,
 	NVM_ID_CAP_ENCRYPT	= 0x8,
+
+	/* Memory types */
+	NVM_ID_FMTYPE_SLC	= 0,
+	NVM_ID_FMTYPE_MLC	= 1,
+};
+
+struct nvm_id_lp_mlc {
+	u16	num_pairs;
+	u8	pairs[886];
+};
+
+struct nvm_id_lp_tbl {
+	__u8	id[8];
+	struct nvm_id_lp_mlc mlc;
 };
 
 struct nvm_id_group {
@@ -89,6 +103,8 @@ struct nvm_id_group {
 	u32	mpos;
 	u32	mccap;
 	u16	cpar;
+
+	struct nvm_id_lp_tbl lptbl;
 };
 
 struct nvm_addr_format {
@@ -297,6 +313,10 @@ struct nvm_dev {
 	int sec_per_blk;
 	int sec_per_lun;
 
+	/* lower page table */
+	int lps_per_blk;
+	int *lptbl;
+
 	unsigned long total_pages;
 	unsigned long total_blocks;
 	int nr_luns;
-- 
2.1.4

[PATCH v2 4/7] lightnvm: core on-disk initialization

[Matias Bjørling]

An Open-Channel SSD shall be initialized before use. To initialize, we
define an on-disk format, that keeps a small set of metadata to bring up
the media manager on top of the device.

The initial step is introduced to allow a user to format the disks for a
given media manager. During format, a system block is stored on one to
three separate luns on the device. Each lun has the system block
duplicated. During initialization, the system block can be retrieved and
the appropriate media manager can initialized.

The on-disk format currently covers (struct nvm_system_block):

 - Magic value "NVMS".
 - Monotonic increasing sequence number.
 - The physical block erase count.
 - Version of the system block format.
 - Media manager type.
 - Media manager superblock physical address.

The interface provides three functions to manage the system block:

 int nvm_init_sysblock(struct nvm_dev *, struct nvm_sb_info *)
 int nvm_get_sysblock(struct nvm *dev, struct nvm_sb_info *)
 int nvm_update_sysblock(struct nvm *dev, struct nvm_sb_info *)

Each implement a part of the logic to manage the system block. The
initialization creates the first system blocks and mark them on the
device. Get retrieves the latest system block by scanning all pages in
the associated system blocks. The update sysblock writes new metadata
and allocates new block if necessary.

Signed-off-by: Matias Bjørling <m@bjorling.me>
---
 drivers/lightnvm/Makefile     |   2 +-
 drivers/lightnvm/core.c       |   1 +
 drivers/lightnvm/sysblk.c     | 562 ++++++++++++++++++++++++++++++++++++++++++
 include/linux/lightnvm.h      |  35 +++
 include/uapi/linux/lightnvm.h |   1 +
 5 files changed, 600 insertions(+), 1 deletion(-)
 create mode 100644 drivers/lightnvm/sysblk.c

diff --git a/drivers/lightnvm/Makefile b/drivers/lightnvm/Makefile
index 7e0f42a..a7a0a22 100644
--- a/drivers/lightnvm/Makefile
+++ b/drivers/lightnvm/Makefile
@@ -2,6 +2,6 @@
 # Makefile for Open-Channel SSDs.
 #
 
-obj-$(CONFIG_NVM)		:= core.o
+obj-$(CONFIG_NVM)		:= core.o sysblk.o
 obj-$(CONFIG_NVM_GENNVM) 	+= gennvm.o
 obj-$(CONFIG_NVM_RRPC)		+= rrpc.o
diff --git a/drivers/lightnvm/core.c b/drivers/lightnvm/core.c
index 1f302cc..73b8ae1 100644
--- a/drivers/lightnvm/core.c
+++ b/drivers/lightnvm/core.c
@@ -466,6 +466,7 @@ static int nvm_core_init(struct nvm_dev *dev)
 				dev->nr_chnls;
 	dev->total_pages = dev->total_blocks * dev->pgs_per_blk;
 	INIT_LIST_HEAD(&dev->online_targets);
+	mutex_init(&dev->mlock);
 
 	return 0;
 }
diff --git a/drivers/lightnvm/sysblk.c b/drivers/lightnvm/sysblk.c
new file mode 100644
index 0000000..b8489f4
--- /dev/null
+++ b/drivers/lightnvm/sysblk.c
@@ -0,0 +1,562 @@
+/*
+ * Copyright (C) 2015 Matias Bjorling. All rights reserved.
+ *
+ * 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.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; see the file COPYING.  If not, write to
+ * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
+ * USA.
+ *
+ */
+
+#include <linux/lightnvm.h>
+
+#define MAX_SYSBLKS 3	/* remember to update mapping scheme on change */
+#define MAX_BLKS_PR_SYSBLK 2 /* 2 blks with 256 pages and 3000 erases
+			      * enables ~1.5M updates per sysblk unit
+			      */
+
+struct sysblk_scan {
+	/* A row is a collection of flash blocks for a system block. */
+	int nr_rows;
+	int row;
+	int act_blk[MAX_SYSBLKS];
+
+	int nr_ppas;
+	struct ppa_addr ppas[MAX_SYSBLKS * MAX_BLKS_PR_SYSBLK];/* all sysblks */
+};
+
+static inline int scan_ppa_idx(int row, int blkid)
+{
+	return (row * MAX_BLKS_PR_SYSBLK) + blkid;
+}
+
+void nvm_sysblk_to_cpu(struct nvm_sb_info *info, struct nvm_system_block *sb)
+{
+	info->seqnr = be32_to_cpu(sb->seqnr);
+	info->erase_cnt = be32_to_cpu(sb->erase_cnt);
+	info->version = be16_to_cpu(sb->version);
+	strncpy(info->mmtype, sb->mmtype, NVM_MMTYPE_LEN);
+	info->fs_ppa.ppa = be64_to_cpu(sb->fs_ppa);
+}
+
+void nvm_cpu_to_sysblk(struct nvm_system_block *sb, struct nvm_sb_info *info)
+{
+	sb->magic = cpu_to_be32(NVM_SYSBLK_MAGIC);
+	sb->seqnr = cpu_to_be32(info->seqnr);
+	sb->erase_cnt = cpu_to_be32(info->erase_cnt);
+	sb->version = cpu_to_be16(info->version);
+	strncpy(sb->mmtype, info->mmtype, NVM_MMTYPE_LEN);
+	sb->fs_ppa = cpu_to_be64(info->fs_ppa.ppa);
+}
+
+static int nvm_setup_sysblks(struct nvm_dev *dev, struct ppa_addr *sysblk_ppas)
+{
+	int nr_rows = min_t(int, MAX_SYSBLKS, dev->nr_chnls);
+	int i;
+
+	for (i = 0; i < nr_rows; i++)
+		sysblk_ppas[i].ppa = 0;
+
+	/* if possible, place sysblk at first channel, middle channel and last
+	 * channel of the device. If not, create only one or two sys blocks
+	 */
+	switch (dev->nr_chnls) {
+	case 2:
+		sysblk_ppas[1].g.ch = 1;
+		/* fall-through */
+	case 1:
+		sysblk_ppas[0].g.ch = 0;
+		break;
+	default:
+		sysblk_ppas[0].g.ch = 0;
+		sysblk_ppas[1].g.ch = dev->nr_chnls / 2;
+		sysblk_ppas[2].g.ch = dev->nr_chnls - 1;
+		break;
+	}
+
+	return nr_rows;
+}
+
+void nvm_setup_sysblk_scan(struct nvm_dev *dev, struct sysblk_scan *s,
+						struct ppa_addr *sysblk_ppas)
+{
+	memset(s, 0, sizeof(struct sysblk_scan));
+	s->nr_rows = nvm_setup_sysblks(dev, sysblk_ppas);
+}
+
+static int sysblk_get_host_blks(struct ppa_addr ppa, int nr_blks, u8 *blks,
+								void *private)
+{
+	struct sysblk_scan *s = private;
+	int i, nr_sysblk = 0;
+
+	for (i = 0; i < nr_blks; i++) {
+		if (blks[i] != NVM_BLK_T_HOST)
+			continue;
+
+		if (s->nr_ppas == MAX_BLKS_PR_SYSBLK * MAX_SYSBLKS) {
+			pr_err("nvm: too many host blks\n");
+			return -EINVAL;
+		}
+
+		ppa.g.blk = i;
+
+		s->ppas[scan_ppa_idx(s->row, nr_sysblk)] = ppa;
+		s->nr_ppas++;
+		nr_sysblk++;
+	}
+
+	return 0;
+}
+
+static int nvm_get_all_sysblks(struct nvm_dev *dev, struct sysblk_scan *s,
+				struct ppa_addr *ppas, nvm_bb_update_fn *fn)
+{
+	struct ppa_addr dppa;
+	int i, ret;
+
+	s->nr_ppas = 0;
+
+	for (i = 0; i < s->nr_rows; i++) {
+		dppa = generic_to_dev_addr(dev, ppas[i]);
+		s->row = i;
+
+		ret = dev->ops->get_bb_tbl(dev, dppa, dev->blks_per_lun, fn, s);
+		if (ret) {
+			pr_err("nvm: failed bb tbl for ppa (%u %u)\n",
+							ppas[i].g.ch,
+							ppas[i].g.blk);
+			return ret;
+		}
+	}
+
+	return ret;
+}
+
+/*
+ * scans a block for latest sysblk.
+ * Returns:
+ *	0 - newer sysblk not found. PPA is updated to latest page.
+ *	1 - newer sysblk found and stored in *cur. PPA is updated to
+ *	    next valid page.
+ *	<0- error.
+ */
+static int nvm_scan_block(struct nvm_dev *dev, struct ppa_addr *ppa,
+						struct nvm_system_block *sblk)
+{
+	struct nvm_system_block *cur;
+	int pg, cursz, ret, found = 0;
+
+	/* the full buffer for a flash page is allocated. Only the first of it
+	 * contains the system block information
+	 */
+	cursz = dev->sec_size * dev->sec_per_pg * dev->nr_planes;
+	cur = kmalloc(cursz, GFP_KERNEL);
+	if (!cur)
+		return -ENOMEM;
+
+	/* perform linear scan through the block */
+	for (pg = 0; pg < dev->lps_per_blk; pg++) {
+		ppa->g.pg = ppa_to_slc(dev, pg);
+
+		ret = nvm_submit_ppa(dev, ppa, 1, NVM_OP_PREAD, NVM_IO_SLC_MODE,
+								cur, cursz);
+		if (ret) {
+			if (ret == NVM_RSP_ERR_EMPTYPAGE) {
+				pr_debug("nvm: sysblk scan empty ppa (%u %u %u %u)\n",
+							ppa->g.ch,
+							ppa->g.lun,
+							ppa->g.blk,
+							ppa->g.pg);
+				break;
+			}
+			pr_err("nvm: read failed (%x) for ppa (%u %u %u %u)",
+							ret,
+							ppa->g.ch,
+							ppa->g.lun,
+							ppa->g.blk,
+							ppa->g.pg);
+			break; /* if we can't read a page, continue to the
+				* next blk
+				*/
+		}
+
+		if (be32_to_cpu(cur->magic) != NVM_SYSBLK_MAGIC) {
+			pr_debug("nvm: scan break for ppa (%u %u %u %u)\n",
+							ppa->g.ch,
+							ppa->g.lun,
+							ppa->g.blk,
+							ppa->g.pg);
+			break; /* last valid page already found */
+		}
+
+		if (be32_to_cpu(cur->seqnr) < be32_to_cpu(sblk->seqnr))
+			continue;
+
+		memcpy(sblk, cur, sizeof(struct nvm_system_block));
+		found = 1;
+	}
+
+	kfree(cur);
+
+	return found;
+}
+
+static int nvm_set_bb_tbl(struct nvm_dev *dev, struct sysblk_scan *s, int type)
+{
+	struct nvm_rq rqd;
+	int ret;
+
+	if (s->nr_ppas > dev->ops->max_phys_sect) {
+		pr_err("nvm: unable to update all sysblocks atomically\n");
+		return -EINVAL;
+	}
+
+	memset(&rqd, 0, sizeof(struct nvm_rq));
+
+	nvm_set_rqd_ppalist(dev, &rqd, s->ppas, s->nr_ppas);
+	nvm_generic_to_addr_mode(dev, &rqd);
+
+	ret = dev->ops->set_bb_tbl(dev, &rqd, type);
+	nvm_free_rqd_ppalist(dev, &rqd);
+	if (ret) {
+		pr_err("nvm: sysblk failed bb mark\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int sysblk_get_free_blks(struct ppa_addr ppa, int nr_blks, u8 *blks,
+								void *private)
+{
+	struct sysblk_scan *s = private;
+	struct ppa_addr *sppa;
+	int i, blkid = 0;
+
+	for (i = 0; i < nr_blks; i++) {
+		if (blks[i] == NVM_BLK_T_HOST)
+			return -EEXIST;
+
+		if (blks[i] != NVM_BLK_T_FREE)
+			continue;
+
+		sppa = &s->ppas[scan_ppa_idx(s->row, blkid)];
+		sppa->g.ch = ppa.g.ch;
+		sppa->g.lun = ppa.g.lun;
+		sppa->g.blk = i;
+		s->nr_ppas++;
+		blkid++;
+
+		pr_debug("nvm: use (%u %u %u) as sysblk\n",
+					sppa->g.ch, sppa->g.lun, sppa->g.blk);
+		if (blkid > MAX_BLKS_PR_SYSBLK - 1)
+			return 0;
+	}
+
+	pr_err("nvm: sysblk failed get sysblk\n");
+	return -EINVAL;
+}
+
+static int nvm_write_and_verify(struct nvm_dev *dev, struct nvm_sb_info *info,
+							struct sysblk_scan *s)
+{
+	struct nvm_system_block nvmsb;
+	void *buf;
+	int i, sect, ret, bufsz;
+	struct ppa_addr *ppas;
+
+	nvm_cpu_to_sysblk(&nvmsb, info);
+
+	/* buffer for flash page */
+	bufsz = dev->sec_size * dev->sec_per_pg * dev->nr_planes;
+	buf = kzalloc(bufsz, GFP_KERNEL);
+	if (!buf)
+		return -ENOMEM;
+	memcpy(buf, &nvmsb, sizeof(struct nvm_system_block));
+
+	ppas = kcalloc(dev->sec_per_pg, sizeof(struct ppa_addr), GFP_KERNEL);
+	if (!ppas) {
+		ret = -ENOMEM;
+		goto err;
+	}
+
+	/* Write and verify */
+	for (i = 0; i < s->nr_rows; i++) {
+		ppas[0] = s->ppas[scan_ppa_idx(i, s->act_blk[i])];
+
+		pr_debug("nvm: writing sysblk to ppa (%u %u %u %u)\n",
+							ppas[0].g.ch,
+							ppas[0].g.lun,
+							ppas[0].g.blk,
+							ppas[0].g.pg);
+
+		/* Expand to all sectors within a flash page */
+		if (dev->sec_per_pg > 1) {
+			for (sect = 1; sect < dev->sec_per_pg; sect++) {
+				ppas[sect].ppa = ppas[0].ppa;
+				ppas[sect].g.sec = sect;
+			}
+		}
+
+		ret = nvm_submit_ppa(dev, ppas, dev->sec_per_pg, NVM_OP_PWRITE,
+						NVM_IO_SLC_MODE, buf, bufsz);
+		if (ret) {
+			pr_err("nvm: sysblk failed program (%u %u %u)\n",
+							ppas[0].g.ch,
+							ppas[0].g.lun,
+							ppas[0].g.blk);
+			break;
+		}
+
+		ret = nvm_submit_ppa(dev, ppas, dev->sec_per_pg, NVM_OP_PREAD,
+						NVM_IO_SLC_MODE, buf, bufsz);
+		if (ret) {
+			pr_err("nvm: sysblk failed read (%u %u %u)\n",
+							ppas[0].g.ch,
+							ppas[0].g.lun,
+							ppas[0].g.blk);
+			break;
+		}
+
+		if (memcmp(buf, &nvmsb, sizeof(struct nvm_system_block))) {
+			pr_err("nvm: sysblk failed verify (%u %u %u)\n",
+							ppas[0].g.ch,
+							ppas[0].g.lun,
+							ppas[0].g.blk);
+			ret = -EINVAL;
+			break;
+		}
+	}
+
+	kfree(ppas);
+err:
+	kfree(buf);
+
+	return ret;
+}
+
+static int nvm_prepare_new_sysblks(struct nvm_dev *dev, struct sysblk_scan *s)
+{
+	int i, ret;
+	unsigned long nxt_blk;
+	struct ppa_addr *ppa;
+
+	for (i = 0; i < s->nr_rows; i++) {
+		nxt_blk = (s->act_blk[i] + 1) % MAX_BLKS_PR_SYSBLK;
+		ppa = &s->ppas[scan_ppa_idx(i, nxt_blk)];
+		ppa->g.pg = ppa_to_slc(dev, 0);
+
+		ret = nvm_erase_ppa(dev, ppa, 1);
+		if (ret)
+			return ret;
+
+		s->act_blk[i] = nxt_blk;
+	}
+
+	return 0;
+}
+
+int nvm_get_sysblock(struct nvm_dev *dev, struct nvm_sb_info *info)
+{
+	struct ppa_addr sysblk_ppas[MAX_SYSBLKS];
+	struct sysblk_scan s;
+	struct nvm_system_block *cur;
+	int i, j, found = 0;
+	int ret = -ENOMEM;
+
+	/*
+	 * 1. setup sysblk locations
+	 * 2. get bad block list
+	 * 3. filter on host-specific (type 3)
+	 * 4. iterate through all and find the highest seq nr.
+	 * 5. return superblock information
+	 */
+
+	if (!dev->ops->get_bb_tbl)
+		return -EINVAL;
+
+	nvm_setup_sysblk_scan(dev, &s, sysblk_ppas);
+
+	mutex_lock(&dev->mlock);
+	ret = nvm_get_all_sysblks(dev, &s, sysblk_ppas, sysblk_get_host_blks);
+	if (ret)
+		goto err_sysblk;
+
+	/* no sysblocks initialized */
+	if (!s.nr_ppas)
+		goto err_sysblk;
+
+	cur = kzalloc(sizeof(struct nvm_system_block), GFP_KERNEL);
+	if (!cur)
+		goto err_sysblk;
+
+	/* find the latest block across all sysblocks */
+	for (i = 0; i < s.nr_rows; i++) {
+		for (j = 0; j < MAX_BLKS_PR_SYSBLK; j++) {
+			struct ppa_addr ppa = s.ppas[scan_ppa_idx(i, j)];
+
+			ret = nvm_scan_block(dev, &ppa, cur);
+			if (ret > 0)
+				found = 1;
+			else if (ret < 0)
+				break;
+		}
+	}
+
+	nvm_sysblk_to_cpu(info, cur);
+
+	kfree(cur);
+err_sysblk:
+	mutex_unlock(&dev->mlock);
+
+	if (found)
+		return 1;
+	return ret;
+}
+
+int nvm_update_sysblock(struct nvm_dev *dev, struct nvm_sb_info *new)
+{
+	/* 1. for each latest superblock
+	 * 2. if room
+	 *    a. write new flash page entry with the updated information
+	 * 3. if no room
+	 *    a. find next available block on lun (linear search)
+	 *       if none, continue to next lun
+	 *       if none at all, report error. also report that it wasn't
+	 *       possible to write to all superblocks.
+	 *    c. write data to block.
+	 */
+	struct ppa_addr sysblk_ppas[MAX_SYSBLKS];
+	struct sysblk_scan s;
+	struct nvm_system_block *cur;
+	int i, j, ppaidx, found = 0;
+	int ret = -ENOMEM;
+
+	if (!dev->ops->get_bb_tbl)
+		return -EINVAL;
+
+	nvm_setup_sysblk_scan(dev, &s, sysblk_ppas);
+
+	mutex_lock(&dev->mlock);
+	ret = nvm_get_all_sysblks(dev, &s, sysblk_ppas, sysblk_get_host_blks);
+	if (ret)
+		goto err_sysblk;
+
+	cur = kzalloc(sizeof(struct nvm_system_block), GFP_KERNEL);
+	if (!cur)
+		goto err_sysblk;
+
+	/* Get the latest sysblk for each sysblk row */
+	for (i = 0; i < s.nr_rows; i++) {
+		found = 0;
+		for (j = 0; j < MAX_BLKS_PR_SYSBLK; j++) {
+			ppaidx = scan_ppa_idx(i, j);
+			ret = nvm_scan_block(dev, &s.ppas[ppaidx], cur);
+			if (ret > 0) {
+				s.act_blk[i] = j;
+				found = 1;
+			} else if (ret < 0)
+				break;
+		}
+	}
+
+	if (!found) {
+		pr_err("nvm: no valid sysblks found to update\n");
+		ret = -EINVAL;
+		goto err_cur;
+	}
+
+	/*
+	 * All sysblocks found. Check that they have same page id in their flash
+	 * blocks
+	 */
+	for (i = 1; i < s.nr_rows; i++) {
+		struct ppa_addr l = s.ppas[scan_ppa_idx(0, s.act_blk[0])];
+		struct ppa_addr r = s.ppas[scan_ppa_idx(i, s.act_blk[i])];
+
+		if (l.g.pg != r.g.pg) {
+			pr_err("nvm: sysblks not on same page. Previous update failed.\n");
+			ret = -EINVAL;
+			goto err_cur;
+		}
+	}
+
+	/*
+	 * Check that there haven't been another update to the seqnr since we
+	 * began
+	 */
+	if ((new->seqnr - 1) != be32_to_cpu(cur->seqnr)) {
+		pr_err("nvm: seq is not sequential\n");
+		ret = -EINVAL;
+		goto err_cur;
+	}
+
+	/*
+	 * When all pages in a block has been written, a new block is selected
+	 * and writing is performed on the new block.
+	 */
+	if (s.ppas[scan_ppa_idx(0, s.act_blk[0])].g.pg ==
+						dev->lps_per_blk - 1) {
+		ret = nvm_prepare_new_sysblks(dev, &s);
+		if (ret)
+			goto err_cur;
+	}
+
+	ret = nvm_write_and_verify(dev, new, &s);
+err_cur:
+	kfree(cur);
+err_sysblk:
+	mutex_unlock(&dev->mlock);
+
+	return ret;
+}
+
+int nvm_init_sysblock(struct nvm_dev *dev, struct nvm_sb_info *info)
+{
+	struct ppa_addr sysblk_ppas[MAX_SYSBLKS];
+	struct sysblk_scan s;
+	int ret;
+
+	/*
+	 * 1. select master blocks and select first available blks
+	 * 2. get bad block list
+	 * 3. mark MAX_SYSBLKS block as host-based device allocated.
+	 * 4. write and verify data to block
+	 */
+
+	if (!dev->ops->get_bb_tbl || !dev->ops->set_bb_tbl)
+		return -EINVAL;
+
+	if (!(dev->mccap & NVM_ID_CAP_SLC) || !dev->lps_per_blk) {
+		pr_err("nvm: memory does not support SLC access\n");
+		return -EINVAL;
+	}
+
+	/* Index all sysblocks and mark them as host-driven */
+	nvm_setup_sysblk_scan(dev, &s, sysblk_ppas);
+
+	mutex_lock(&dev->mlock);
+	ret = nvm_get_all_sysblks(dev, &s, sysblk_ppas, sysblk_get_free_blks);
+	if (ret)
+		goto err_mark;
+
+	ret = nvm_set_bb_tbl(dev, &s, NVM_BLK_T_HOST);
+	if (ret)
+		goto err_mark;
+
+	/* Write to the first block of each row */
+	ret = nvm_write_and_verify(dev, info, &s);
+err_mark:
+	mutex_unlock(&dev->mlock);
+	return ret;
+}
diff --git a/include/linux/lightnvm.h b/include/linux/lightnvm.h
index 678fd91..7ad22d3 100644
--- a/include/linux/lightnvm.h
+++ b/include/linux/lightnvm.h
@@ -17,6 +17,7 @@ enum {
 #include <linux/types.h>
 #include <linux/file.h>
 #include <linux/dmapool.h>
+#include <uapi/linux/lightnvm.h>
 
 enum {
 	/* HW Responsibilities */
@@ -281,6 +282,15 @@ struct nvm_block {
 	int state;
 };
 
+/* system block cpu representation */
+struct nvm_sb_info {
+	unsigned long		seqnr;
+	unsigned long		erase_cnt;
+	unsigned int		version;
+	char			mmtype[NVM_MMTYPE_LEN];
+	struct ppa_addr		fs_ppa;
+};
+
 struct nvm_dev {
 	struct nvm_dev_ops *ops;
 
@@ -329,6 +339,8 @@ struct nvm_dev {
 	/* Backend device */
 	struct request_queue *q;
 	char name[DISK_NAME_LEN];
+
+	struct mutex mlock;
 };
 
 static inline struct ppa_addr generic_to_dev_addr(struct nvm_dev *dev,
@@ -394,6 +406,11 @@ static inline struct ppa_addr block_to_ppa(struct nvm_dev *dev,
 	return ppa;
 }
 
+static inline int ppa_to_slc(struct nvm_dev *dev, int slc_pg)
+{
+	return dev->lptbl[slc_pg];
+}
+
 typedef blk_qc_t (nvm_tgt_make_rq_fn)(struct request_queue *, struct bio *);
 typedef sector_t (nvm_tgt_capacity_fn)(void *);
 typedef void *(nvm_tgt_init_fn)(struct nvm_dev *, struct gendisk *, int, int);
@@ -489,6 +506,24 @@ extern int nvm_erase_blk(struct nvm_dev *, struct nvm_block *);
 extern void nvm_end_io(struct nvm_rq *, int);
 extern int nvm_submit_ppa(struct nvm_dev *, struct ppa_addr *, int, int, int,
 								void *, int);
+
+/* sysblk.c */
+#define NVM_SYSBLK_MAGIC 0x4E564D53 /* "NVMS" */
+
+/* system block on disk representation */
+struct nvm_system_block {
+	__be32			magic;		/* magic signature */
+	__be32			seqnr;		/* sequence number */
+	__be32			erase_cnt;	/* erase count */
+	__be16			version;	/* version number */
+	u8			mmtype[NVM_MMTYPE_LEN]; /* media manager name */
+	__be64			fs_ppa;		/* PPA for media manager
+						 * superblock */
+};
+
+extern int nvm_get_sysblock(struct nvm_dev *, struct nvm_sb_info *);
+extern int nvm_update_sysblock(struct nvm_dev *, struct nvm_sb_info *);
+extern int nvm_init_sysblock(struct nvm_dev *, struct nvm_sb_info *);
 #else /* CONFIG_NVM */
 struct nvm_dev_ops;
 
diff --git a/include/uapi/linux/lightnvm.h b/include/uapi/linux/lightnvm.h
index 928f989..0171b85 100644
--- a/include/uapi/linux/lightnvm.h
+++ b/include/uapi/linux/lightnvm.h
@@ -33,6 +33,7 @@
 
 #define NVM_TTYPE_NAME_MAX 48
 #define NVM_TTYPE_MAX 63
+#define NVM_MMTYPE_LEN 8
 
 #define NVM_CTRL_FILE "/dev/lightnvm/control"
 
-- 
2.1.4

[PATCH 5/7] lightnvm: introduce ioctl to initialize device

[Matias Bjørling]

Based on the previous patch, we now introduce an ioctl to initialize the
device using nvm_init_sysblock and create the necessary system blocks.
The user may specify the media manager that they wish to instantiate on
top. Default from user-space will be "gennvm".

Signed-off-by: Matias Bjørling <m@bjorling.me>
---
 drivers/lightnvm/core.c       | 50 +++++++++++++++++++++++++++++++++++++++++++
 include/uapi/linux/lightnvm.h | 11 ++++++++++
 2 files changed, 61 insertions(+)

diff --git a/drivers/lightnvm/core.c b/drivers/lightnvm/core.c
index 73b8ae1..ee08fac 100644
--- a/drivers/lightnvm/core.c
+++ b/drivers/lightnvm/core.c
@@ -1019,6 +1019,54 @@ static long nvm_ioctl_dev_remove(struct file *file, void __user *arg)
 	return __nvm_configure_remove(&remove);
 }
 
+static void nvm_setup_nvm_sb_info(struct nvm_sb_info *info)
+{
+	info->seqnr = 1;
+	info->erase_cnt = 0;
+	info->version = 1;
+}
+
+static long __nvm_ioctl_dev_init(struct nvm_ioctl_dev_init *init)
+{
+	struct nvm_dev *dev;
+	struct nvm_sb_info info;
+
+	down_write(&nvm_lock);
+	dev = nvm_find_nvm_dev(init->dev);
+	up_write(&nvm_lock);
+	if (!dev) {
+		pr_err("nvm: device not found\n");
+		return -EINVAL;
+	}
+
+	nvm_setup_nvm_sb_info(&info);
+
+	strncpy(info.mmtype, init->mmtype, NVM_MMTYPE_LEN);
+	info.fs_ppa.ppa = -1;
+
+	return nvm_init_sysblock(dev, &info);
+}
+
+static long nvm_ioctl_dev_init(struct file *file, void __user *arg)
+{
+	struct nvm_ioctl_dev_init init;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	if (copy_from_user(&init, arg, sizeof(struct nvm_ioctl_dev_init)))
+		return -EFAULT;
+
+	if (init.flags != 0) {
+		pr_err("nvm: no flags supported\n");
+		return -EINVAL;
+	}
+
+	init.dev[DISK_NAME_LEN - 1] = '\0';
+
+	return __nvm_ioctl_dev_init(&init);
+}
+
 static long nvm_ctl_ioctl(struct file *file, uint cmd, unsigned long arg)
 {
 	void __user *argp = (void __user *)arg;
@@ -1032,6 +1080,8 @@ static long nvm_ctl_ioctl(struct file *file, uint cmd, unsigned long arg)
 		return nvm_ioctl_dev_create(file, argp);
 	case NVM_DEV_REMOVE:
 		return nvm_ioctl_dev_remove(file, argp);
+	case NVM_DEV_INIT:
+		return nvm_ioctl_dev_init(file, argp);
 	}
 	return 0;
 }
diff --git a/include/uapi/linux/lightnvm.h b/include/uapi/linux/lightnvm.h
index 0171b85..56339e2 100644
--- a/include/uapi/linux/lightnvm.h
+++ b/include/uapi/linux/lightnvm.h
@@ -101,6 +101,12 @@ struct nvm_ioctl_remove {
 	__u32 flags;
 };
 
+struct nvm_ioctl_dev_init {
+	char dev[DISK_NAME_LEN];		/* open-channel SSD device */
+	char mmtype[NVM_MMTYPE_LEN];		/* register to media manager */
+
+	__u32 flags;
+};
 
 /* The ioctl type, 'L', 0x20 - 0x2F documented in ioctl-number.txt */
 enum {
@@ -111,6 +117,9 @@ enum {
 	/* device level cmds */
 	NVM_DEV_CREATE_CMD,
 	NVM_DEV_REMOVE_CMD,
+
+	/* Init a device to support LightNVM media managers */
+	NVM_DEV_INIT_CMD,
 };
 
 #define NVM_IOCTL 'L' /* 0x4c */
@@ -123,6 +132,8 @@ enum {
 						struct nvm_ioctl_create)
 #define NVM_DEV_REMOVE		_IOW(NVM_IOCTL, NVM_DEV_REMOVE_CMD, \
 						struct nvm_ioctl_remove)
+#define NVM_DEV_INIT		_IOW(NVM_IOCTL, NVM_DEV_INIT_CMD, \
+						struct nvm_ioctl_dev_init)
 
 #define NVM_VERSION_MAJOR	1
 #define NVM_VERSION_MINOR	0
-- 
2.1.4

[PATCH 6/7] lightnvm: use system block for mm initialization

[Matias Bjørling]

Use system block information to register the appropriate media manager.
This enables the LightNVM subsystem to instantiate a media manager
selected by the user, instead of relying on automatic detection by each
media manager loaded in the kernel.

A device must now be initialized before it can proceed to initialize its
media manager. Upon initialization, the configured media manager is
automatically initialized as well.

Signed-off-by: Matias Bjørling <m@bjorling.me>
---
 drivers/lightnvm/core.c  | 25 +++++++++++++++++++++++--
 include/linux/lightnvm.h |  3 +++
 2 files changed, 26 insertions(+), 2 deletions(-)

diff --git a/drivers/lightnvm/core.c b/drivers/lightnvm/core.c
index ee08fac..9e5712d 100644
--- a/drivers/lightnvm/core.c
+++ b/drivers/lightnvm/core.c
@@ -106,6 +106,9 @@ struct nvmm_type *nvm_init_mgr(struct nvm_dev *dev)
 	lockdep_assert_held(&nvm_lock);
 
 	list_for_each_entry(mt, &nvm_mgrs, list) {
+		if (strncmp(dev->sb.mmtype, mt->name, NVM_MMTYPE_LEN))
+			continue;
+
 		ret = mt->register_mgr(dev);
 		if (ret < 0) {
 			pr_err("nvm: media mgr failed to init (%d) on dev %s\n",
@@ -569,9 +572,16 @@ int nvm_register(struct request_queue *q, char *disk_name,
 		}
 	}
 
+	ret = nvm_get_sysblock(dev, &dev->sb);
+	if (!ret)
+		pr_err("nvm: device not initialized.\n");
+	else if (ret < 0)
+		pr_err("nvm: err (%d) on device initialization\n", ret);
+
 	/* register device with a supported media manager */
 	down_write(&nvm_lock);
-	dev->mt = nvm_init_mgr(dev);
+	if (ret > 0)
+		dev->mt = nvm_init_mgr(dev);
 	list_add(&dev->devices, &nvm_devices);
 	up_write(&nvm_lock);
 
@@ -1030,6 +1040,7 @@ static long __nvm_ioctl_dev_init(struct nvm_ioctl_dev_init *init)
 {
 	struct nvm_dev *dev;
 	struct nvm_sb_info info;
+	int ret;
 
 	down_write(&nvm_lock);
 	dev = nvm_find_nvm_dev(init->dev);
@@ -1044,7 +1055,17 @@ static long __nvm_ioctl_dev_init(struct nvm_ioctl_dev_init *init)
 	strncpy(info.mmtype, init->mmtype, NVM_MMTYPE_LEN);
 	info.fs_ppa.ppa = -1;
 
-	return nvm_init_sysblock(dev, &info);
+	ret = nvm_init_sysblock(dev, &info);
+	if (ret)
+		return ret;
+
+	memcpy(&dev->sb, &info, sizeof(struct nvm_sb_info));
+
+	down_write(&nvm_lock);
+	dev->mt = nvm_init_mgr(dev);
+	up_write(&nvm_lock);
+
+	return 0;
 }
 
 static long nvm_ioctl_dev_init(struct file *file, void __user *arg)
diff --git a/include/linux/lightnvm.h b/include/linux/lightnvm.h
index 7ad22d3..02f36bd 100644
--- a/include/linux/lightnvm.h
+++ b/include/linux/lightnvm.h
@@ -301,6 +301,9 @@ struct nvm_dev {
 	struct nvmm_type *mt;
 	void *mp;
 
+	/* System blocks */
+	struct nvm_sb_info sb;
+
 	/* Device information */
 	int nr_chnls;
 	int nr_planes;
-- 
2.1.4

[PATCH 7/7] lightnvm: introduce factory reset

[Matias Bjørling]

Now that a device can be managed using the system blocks, a method to
reset the device is necessary as well. This patch introduces logic to
reset the device easily to factory state and exposes it through an
ioctl.

The ioctl takes the following flags:

  NVM_FACTORY_ERASE_ONLY_USER
      By default all blocks, except host-reserved blocks are erased upon
      factory reset. Instead of this, only erase host-reserved blocks.
  NVM_FACTORY_RESET_HOST_BLKS
      Mark host-reserved blocks to be erased and set their type to free.
  NVM_FACTORY_RESET_GRWN_BBLKS
      Mark "grown bad blocks" to be erased and set their type to free.

Signed-off-by: Matias Bjørling <m@bjorling.me>
---
 drivers/lightnvm/core.c       |  34 ++++++++
 drivers/lightnvm/sysblk.c     | 179 ++++++++++++++++++++++++++++++++++++++++++
 include/linux/lightnvm.h      |   2 +
 include/uapi/linux/lightnvm.h |  19 +++++
 4 files changed, 234 insertions(+)

diff --git a/drivers/lightnvm/core.c b/drivers/lightnvm/core.c
index 9e5712d..33224cb 100644
--- a/drivers/lightnvm/core.c
+++ b/drivers/lightnvm/core.c
@@ -1088,6 +1088,38 @@ static long nvm_ioctl_dev_init(struct file *file, void __user *arg)
 	return __nvm_ioctl_dev_init(&init);
 }
 
+static long nvm_ioctl_dev_factory(struct file *file, void __user *arg)
+{
+	struct nvm_ioctl_dev_factory fact;
+	struct nvm_dev *dev;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	if (copy_from_user(&fact, arg, sizeof(struct nvm_ioctl_dev_factory)))
+		return -EFAULT;
+
+	fact.dev[DISK_NAME_LEN - 1] = '\0';
+
+	if (fact.flags & ~(NVM_FACTORY_NR_BITS - 1))
+		return -EINVAL;
+
+	down_write(&nvm_lock);
+	dev = nvm_find_nvm_dev(fact.dev);
+	up_write(&nvm_lock);
+	if (!dev) {
+		pr_err("nvm: device not found\n");
+		return -EINVAL;
+	}
+
+	if (dev->mt) {
+		dev->mt->unregister_mgr(dev);
+		dev->mt = NULL;
+	}
+
+	return nvm_dev_factory(dev, fact.flags);
+}
+
 static long nvm_ctl_ioctl(struct file *file, uint cmd, unsigned long arg)
 {
 	void __user *argp = (void __user *)arg;
@@ -1103,6 +1135,8 @@ static long nvm_ctl_ioctl(struct file *file, uint cmd, unsigned long arg)
 		return nvm_ioctl_dev_remove(file, argp);
 	case NVM_DEV_INIT:
 		return nvm_ioctl_dev_init(file, argp);
+	case NVM_DEV_FACTORY:
+		return nvm_ioctl_dev_factory(file, argp);
 	}
 	return 0;
 }
diff --git a/drivers/lightnvm/sysblk.c b/drivers/lightnvm/sysblk.c
index b8489f4..321de1f 100644
--- a/drivers/lightnvm/sysblk.c
+++ b/drivers/lightnvm/sysblk.c
@@ -560,3 +560,182 @@ err_mark:
 	mutex_unlock(&dev->mlock);
 	return ret;
 }
+
+struct factory_blks {
+	struct nvm_dev *dev;
+	int flags;
+	unsigned long *blks;
+};
+
+static int factory_nblks(int nblks)
+{
+	/* Round up to nearest BITS_PER_LONG */
+	return (nblks + (BITS_PER_LONG - 1)) & ~(BITS_PER_LONG - 1);
+}
+
+static unsigned int factory_blk_offset(struct nvm_dev *dev, int ch, int lun)
+{
+	int nblks = factory_nblks(dev->blks_per_lun);
+
+	return ((ch * dev->luns_per_chnl * nblks) + (lun * nblks)) /
+								BITS_PER_LONG;
+}
+
+static int nvm_factory_blks(struct ppa_addr ppa, int nr_blks, u8 *blks,
+								void *private)
+{
+	struct factory_blks *f = private;
+	struct nvm_dev *dev = f->dev;
+	int i, lunoff;
+
+	lunoff = factory_blk_offset(dev, ppa.g.ch, ppa.g.lun);
+
+	/* non-set bits correspond to the block must be erased */
+	for (i = 0; i < nr_blks; i++) {
+		switch (blks[i]) {
+		case NVM_BLK_T_FREE:
+			if (f->flags & NVM_FACTORY_ERASE_ONLY_USER)
+				set_bit(i, &f->blks[lunoff]);
+			break;
+		case NVM_BLK_T_HOST:
+			if (!(f->flags & NVM_FACTORY_RESET_HOST_BLKS))
+				set_bit(i, &f->blks[lunoff]);
+			break;
+		case NVM_BLK_T_GRWN_BAD:
+			if (!(f->flags & NVM_FACTORY_RESET_GRWN_BBLKS))
+				set_bit(i, &f->blks[lunoff]);
+			break;
+		default:
+			set_bit(i, &f->blks[lunoff]);
+			break;
+		}
+	}
+
+	return 0;
+}
+
+static int nvm_fact_get_blks(struct nvm_dev *dev, struct ppa_addr *erase_list,
+					int max_ppas, struct factory_blks *f)
+{
+	struct ppa_addr ppa;
+	int ch, lun, blkid, idx, done = 0, ppa_cnt = 0;
+	unsigned long *offset;
+
+	while (!done) {
+		done = 1;
+		for (ch = 0; ch < dev->nr_chnls; ch++) {
+			for (lun = 0; lun < dev->luns_per_chnl; lun++) {
+				idx = factory_blk_offset(dev, ch, lun);
+				offset = &f->blks[idx];
+
+				blkid = find_first_zero_bit(offset,
+							dev->blks_per_lun);
+				if (blkid >= dev->blks_per_lun)
+					continue;
+				set_bit(blkid, offset);
+
+				ppa.ppa = 0;
+				ppa.g.ch = ch;
+				ppa.g.lun = lun;
+				ppa.g.blk = blkid;
+				pr_debug("nvm: erase ppa (%u %u %u)\n",
+								ppa.g.ch,
+								ppa.g.lun,
+								ppa.g.blk);
+
+				erase_list[ppa_cnt] = ppa;
+				ppa_cnt++;
+				done = 0;
+
+				if (ppa_cnt == max_ppas)
+					return ppa_cnt;
+			}
+		}
+	}
+
+	return ppa_cnt;
+}
+
+static int nvm_fact_get_bb_tbl(struct nvm_dev *dev, struct ppa_addr ppa,
+					nvm_bb_update_fn *fn, void *priv)
+{
+	struct ppa_addr dev_ppa;
+	int ret;
+
+	dev_ppa = generic_to_dev_addr(dev, ppa);
+
+	ret = dev->ops->get_bb_tbl(dev, dev_ppa, dev->blks_per_lun, fn, priv);
+	if (ret)
+		pr_err("nvm: failed bb tbl for ch%u lun%u\n",
+							ppa.g.ch, ppa.g.blk);
+	return ret;
+}
+
+static int nvm_fact_select_blks(struct nvm_dev *dev, struct factory_blks *f)
+{
+	int ch, lun, ret;
+	struct ppa_addr ppa;
+
+	ppa.ppa = 0;
+	for (ch = 0; ch < dev->nr_chnls; ch++) {
+		for (lun = 0; lun < dev->luns_per_chnl; lun++) {
+			ppa.g.ch = ch;
+			ppa.g.lun = lun;
+
+			ret = nvm_fact_get_bb_tbl(dev, ppa, nvm_factory_blks,
+									f);
+			if (ret)
+				return ret;
+		}
+	}
+
+	return 0;
+}
+
+int nvm_dev_factory(struct nvm_dev *dev, int flags)
+{
+	struct factory_blks f;
+	struct ppa_addr *ppas;
+	int ppa_cnt, ret = -ENOMEM;
+	int max_ppas = dev->ops->max_phys_sect / dev->nr_planes;
+	struct ppa_addr sysblk_ppas[MAX_SYSBLKS];
+	struct sysblk_scan s;
+
+	f.blks = kzalloc(factory_nblks(dev->blks_per_lun) * dev->nr_luns,
+								GFP_KERNEL);
+	if (!f.blks)
+		return ret;
+
+	ppas = kcalloc(max_ppas, sizeof(struct ppa_addr), GFP_KERNEL);
+	if (!ppas)
+		goto err_blks;
+
+	f.dev = dev;
+	f.flags = flags;
+
+	/* create list of blks to be erased */
+	ret = nvm_fact_select_blks(dev, &f);
+	if (ret)
+		goto err_ppas;
+
+	/* continue to erase until list of blks until empty */
+	while ((ppa_cnt = nvm_fact_get_blks(dev, ppas, max_ppas, &f)) > 0)
+		nvm_erase_ppa(dev, ppas, ppa_cnt);
+
+	/* mark host reserved blocks free */
+	if (flags & NVM_FACTORY_RESET_HOST_BLKS) {
+		nvm_setup_sysblk_scan(dev, &s, sysblk_ppas);
+		mutex_lock(&dev->mlock);
+		ret = nvm_get_all_sysblks(dev, &s, sysblk_ppas,
+							sysblk_get_host_blks);
+		if (!ret)
+			ret = nvm_set_bb_tbl(dev, &s, NVM_BLK_T_FREE);
+		mutex_unlock(&dev->mlock);
+	}
+err_ppas:
+	kfree(ppas);
+err_blks:
+	kfree(f.blks);
+	return ret;
+}
+EXPORT_SYMBOL(nvm_dev_factory);
diff --git a/include/linux/lightnvm.h b/include/linux/lightnvm.h
index 02f36bd..fc0e7c9 100644
--- a/include/linux/lightnvm.h
+++ b/include/linux/lightnvm.h
@@ -527,6 +527,8 @@ struct nvm_system_block {
 extern int nvm_get_sysblock(struct nvm_dev *, struct nvm_sb_info *);
 extern int nvm_update_sysblock(struct nvm_dev *, struct nvm_sb_info *);
 extern int nvm_init_sysblock(struct nvm_dev *, struct nvm_sb_info *);
+
+extern int nvm_dev_factory(struct nvm_dev *, int flags);
 #else /* CONFIG_NVM */
 struct nvm_dev_ops;
 
diff --git a/include/uapi/linux/lightnvm.h b/include/uapi/linux/lightnvm.h
index 56339e2..774a431 100644
--- a/include/uapi/linux/lightnvm.h
+++ b/include/uapi/linux/lightnvm.h
@@ -108,6 +108,20 @@ struct nvm_ioctl_dev_init {
 	__u32 flags;
 };
 
+enum {
+	NVM_FACTORY_ERASE_ONLY_USER	= 1 << 0, /* erase only blocks used as
+						   * host blks or grown blks */
+	NVM_FACTORY_RESET_HOST_BLKS	= 1 << 1, /* remove host blk marks */
+	NVM_FACTORY_RESET_GRWN_BBLKS	= 1 << 2, /* remove grown blk marks */
+	NVM_FACTORY_NR_BITS		= 1 << 3, /* stops here */
+};
+
+struct nvm_ioctl_dev_factory {
+	char dev[DISK_NAME_LEN];
+
+	__u32 flags;
+};
+
 /* The ioctl type, 'L', 0x20 - 0x2F documented in ioctl-number.txt */
 enum {
 	/* top level cmds */
@@ -120,6 +134,9 @@ enum {
 
 	/* Init a device to support LightNVM media managers */
 	NVM_DEV_INIT_CMD,
+
+	/* Factory reset device */
+	NVM_DEV_FACTORY_CMD,
 };
 
 #define NVM_IOCTL 'L' /* 0x4c */
@@ -134,6 +151,8 @@ enum {
 						struct nvm_ioctl_remove)
 #define NVM_DEV_INIT		_IOW(NVM_IOCTL, NVM_DEV_INIT_CMD, \
 						struct nvm_ioctl_dev_init)
+#define NVM_DEV_FACTORY		_IOW(NVM_IOCTL, NVM_DEV_FACTORY_CMD, \
+						struct nvm_ioctl_dev_factory)
 
 #define NVM_VERSION_MAJOR	1
 #define NVM_VERSION_MINOR	0
-- 
2.1.4