Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1307030 > unrolled thread

[PATCH 00/25] LightNVM patches for 4.5

Back to article view | Back to linux.kernel

Contents

  [PATCH 00/25] LightNVM patches for 4.5 Matias Bjørling <m@bjorling.me> - 2016-01-12 08:00 +0100
    [PATCH 24/25] lightnvm: use system block for mm initialization Matias Bjørling <m@bjorling.me> - 2016-01-12 08:00 +0100
    [PATCH 03/25] lightnvm: add check after mempool allocation Matias Bjørling <m@bjorling.me> - 2016-01-12 08:00 +0100
    [PATCH 07/25] lightnvm: refactor end_io functions for sync Matias Bjørling <m@bjorling.me> - 2016-01-12 08:00 +0100
    [PATCH 20/25] lightnvm: add mccap support Matias Bjørling <m@bjorling.me> - 2016-01-12 08:00 +0100
    [PATCH 05/25] lightnvm: move ppa erase logic to core Matias Bjørling <m@bjorling.me> - 2016-01-12 08:00 +0100
    [PATCH 09/25] lightnvm: check bi_error in gc Matias Bjørling <m@bjorling.me> - 2016-01-12 08:00 +0100
    [PATCH 22/25] lightnvm: core on-disk initialization Matias Bjørling <m@bjorling.me> - 2016-01-12 08:00 +0100
    Re: [PATCH 00/25] LightNVM patches for 4.5 Jens Axboe <axboe@fb.com> - 2016-01-12 16:20 +0100
      Re: [PATCH 00/25] LightNVM patches for 4.5 Matias Bjorling <m@bjorling.me> - 2016-01-13 01:10 +0100

#1307030 — [PATCH 00/25] LightNVM patches for 4.5

Hi Jens,

Here is the collected patches for LightNVM ready for the 4.5 window.
They covers cover the following:

  - Several error handling fixes by Wenwei, and Javier.
  - Statistics bug reported and fixed by Chao.
  - Added two preparation patches for host-side write buffering, by
    Javier.
  - Fix by Wenwei to correctly place block upon reclaim fail.
  - Added sync support and introduced nvm_submit_ppa.
  - Fixed ordering bug when plane ppas were sent before sectors ppas.
  - Added missing bad block type.
  - Added mccap and lower page table support to support SLC.
  - Added on disk initialization to allow specific media manager to be
    loaded upon device scan.
  - Added factory reset to erase flash blocks on a device.
  - Added ioctls for initialization and factory reset.
  - Several refactoring patches to prepare for initialization and
    factory reset.

They are based on top of 4.4-rc8.

Please pick up when convenient.

Thank you,
Matias

Chao Yu (1):
  lightnvm: fix incorrect nr_free_blocks stat

Javier Gonzalez (1):
  lightnvm: add check after mempool allocation

Javier González (2):
  lightnvm: reference rrpc lun in rrpc block
  lightnvm: manage open and closed blocks separately

Matias Bjørling (15):
  lightnvm: move ppa erase logic to core
  lightnvm: refactor rqd ppa list into set/free
  lightnvm: refactor end_io functions for sync
  lightnvm: return the get_bb_tbl return value
  lightnvm: sectors first in ppa list
  lightnvm: support multiple ppas in nvm_erase_ppa
  lightnvm: move rq->error to nvm_rq->error
  lightnvm: introduce nvm_submit_ppa
  lightnvm: fix missing grown bad block type
  lightnvm: add mccap support
  lightnvm: introduce mlc lower page table mappings
  lightnvm: core on-disk initialization
  lightnvm: introduce ioctl to initialize device
  lightnvm: use system block for mm initialization
  lightnvm: introduce factory reset

Wenwei Tao (6):
  lightnvm: fix bio submission issue
  lightnvm: unlock rq and free ppa_list on submission fail
  lightnvm: check bi_error in gc
  lightnvm: put block back to gc list on its reclaim fail
  lightnvm: fix locking and mempool in rrpc_lun_gc
  lightnvm: move the pages per block check out of the loop

 drivers/block/null_blk.c      |   3 +-
 drivers/lightnvm/Makefile     |   2 +-
 drivers/lightnvm/core.c       | 340 ++++++++++++++++++-
 drivers/lightnvm/gennvm.c     | 202 +++++-------
 drivers/lightnvm/rrpc.c       | 104 ++++--
 drivers/lightnvm/rrpc.h       |  13 +-
 drivers/lightnvm/sysblk.c     | 741 ++++++++++++++++++++++++++++++++++++++++++
 drivers/nvme/host/lightnvm.c  |  32 +-
 include/linux/lightnvm.h      | 119 ++++++-
 include/uapi/linux/lightnvm.h |  31 ++
 10 files changed, 1415 insertions(+), 172 deletions(-)
 create mode 100644 drivers/lightnvm/sysblk.c

-- 
2.1.4

[toc] | [next] | [standalone]

#1307031 — [PATCH 24/25] lightnvm: use system block for mm initialization

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

[toc] | [prev] | [next] | [standalone]

#1307032 — [PATCH 03/25] lightnvm: add check after mempool allocation

From: Javier Gonzalez <javier@javigon.com>

The mempool allocation might fail. Make sure to return error when it
does, instead of causing a kernel panic.

Signed-off-by: Javier Gonzalez <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
---
 drivers/lightnvm/rrpc.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/lightnvm/rrpc.c b/drivers/lightnvm/rrpc.c
index a1e7488..f4bc986 100644
--- a/drivers/lightnvm/rrpc.c
+++ b/drivers/lightnvm/rrpc.c
@@ -287,6 +287,8 @@ static int rrpc_move_valid_pages(struct rrpc *rrpc, struct rrpc_block *rblk)
 	}
 
 	page = mempool_alloc(rrpc->page_pool, GFP_NOIO);
+	if (!page)
+		return -ENOMEM;
 
 	while ((slot = find_first_zero_bit(rblk->invalid_pages,
 					    nr_pgs_per_blk)) < nr_pgs_per_blk) {
-- 
2.1.4

[toc] | [prev] | [next] | [standalone]

#1307033 — [PATCH 07/25] lightnvm: refactor end_io functions for sync

To implement sync I/O support within the LightNVM core, the end_io
functions are refactored to take an end_io function pointer instead of
testing for initialized media manager, followed by calling its end_io
function.

Sync I/O can then be implemented using a callback that signal I/O
completion. This is similar to the logic found in blk_to_execute_io().
By implementing it this way, the underlying device I/Os submission logic
is abstracted away from core, targets, and media managers.

Signed-off-by: Matias Bjørling <m@bjorling.me>
---
 drivers/block/null_blk.c     |  3 +--
 drivers/lightnvm/core.c      | 16 ++++++++++++++++
 drivers/lightnvm/gennvm.c    | 34 ++++++++++++++--------------------
 drivers/lightnvm/rrpc.c      |  6 ++----
 drivers/nvme/host/lightnvm.c |  5 +----
 include/linux/lightnvm.h     | 12 ++++++++----
 6 files changed, 42 insertions(+), 34 deletions(-)

diff --git a/drivers/block/null_blk.c b/drivers/block/null_blk.c
index 09e3c0d..e6cad40 100644
--- a/drivers/block/null_blk.c
+++ b/drivers/block/null_blk.c
@@ -436,9 +436,8 @@ static void null_del_dev(struct nullb *nullb)
 static void null_lnvm_end_io(struct request *rq, int error)
 {
 	struct nvm_rq *rqd = rq->end_io_data;
-	struct nvm_dev *dev = rqd->dev;
 
-	dev->mt->end_io(rqd, error);
+	nvm_end_io(rqd, error);
 
 	blk_put_request(rq);
 }
diff --git a/drivers/lightnvm/core.c b/drivers/lightnvm/core.c
index 081b0f5..fa1a052 100644
--- a/drivers/lightnvm/core.c
+++ b/drivers/lightnvm/core.c
@@ -27,6 +27,7 @@
 #include <linux/module.h>
 #include <linux/miscdevice.h>
 #include <linux/lightnvm.h>
+#include <linux/sched/sysctl.h>
 #include <uapi/linux/lightnvm.h>
 
 static LIST_HEAD(nvm_targets);
@@ -288,6 +289,21 @@ int nvm_erase_ppa(struct nvm_dev *dev, struct ppa_addr ppa)
 }
 EXPORT_SYMBOL(nvm_erase_ppa);
 
+void nvm_end_io(struct nvm_rq *rqd, int error)
+{
+	rqd->end_io(rqd, error);
+}
+EXPORT_SYMBOL(nvm_end_io);
+
+static void nvm_end_io_sync(struct nvm_rq *rqd, int errors)
+{
+	struct completion *waiting = rqd->wait;
+
+	rqd->wait = NULL;
+
+	complete(waiting);
+}
+
 static int nvm_core_init(struct nvm_dev *dev)
 {
 	struct nvm_id *id = &dev->identity;
diff --git a/drivers/lightnvm/gennvm.c b/drivers/lightnvm/gennvm.c
index 373be72..12ddcaa 100644
--- a/drivers/lightnvm/gennvm.c
+++ b/drivers/lightnvm/gennvm.c
@@ -317,18 +317,6 @@ static void gennvm_put_blk(struct nvm_dev *dev, struct nvm_block *blk)
 	spin_unlock(&vlun->lock);
 }
 
-static int gennvm_submit_io(struct nvm_dev *dev, struct nvm_rq *rqd)
-{
-	if (!dev->ops->submit_io)
-		return -ENODEV;
-
-	/* Convert address space */
-	nvm_generic_to_addr_mode(dev, rqd);
-
-	rqd->dev = dev;
-	return dev->ops->submit_io(dev, rqd);
-}
-
 static void gennvm_blk_set_type(struct nvm_dev *dev, struct ppa_addr *ppa,
 								int type)
 {
@@ -375,25 +363,32 @@ static void gennvm_mark_blk_bad(struct nvm_dev *dev, struct nvm_rq *rqd)
 		gennvm_blk_set_type(dev, &rqd->ppa_addr, 2);
 }
 
-static int gennvm_end_io(struct nvm_rq *rqd, int error)
+static void gennvm_end_io(struct nvm_rq *rqd, int error)
 {
 	struct nvm_tgt_instance *ins = rqd->ins;
-	int ret = 0;
 
 	switch (error) {
 	case NVM_RSP_SUCCESS:
-		break;
 	case NVM_RSP_ERR_EMPTYPAGE:
 		break;
 	case NVM_RSP_ERR_FAILWRITE:
 		gennvm_mark_blk_bad(rqd->dev, rqd);
-	default:
-		ret++;
 	}
 
-	ret += ins->tt->end_io(rqd, error);
+	ins->tt->end_io(rqd, error);
+}
 
-	return ret;
+static int gennvm_submit_io(struct nvm_dev *dev, struct nvm_rq *rqd)
+{
+	if (!dev->ops->submit_io)
+		return -ENODEV;
+
+	/* Convert address space */
+	nvm_generic_to_addr_mode(dev, rqd);
+
+	rqd->dev = dev;
+	rqd->end_io = gennvm_end_io;
+	return dev->ops->submit_io(dev, rqd);
 }
 
 static int gennvm_erase_blk(struct nvm_dev *dev, struct nvm_block *blk,
@@ -442,7 +437,6 @@ static struct nvmm_type gennvm = {
 	.put_blk	= gennvm_put_blk,
 
 	.submit_io	= gennvm_submit_io,
-	.end_io		= gennvm_end_io,
 	.erase_blk	= gennvm_erase_blk,
 
 	.get_lun	= gennvm_get_lun,
diff --git a/drivers/lightnvm/rrpc.c b/drivers/lightnvm/rrpc.c
index 748cab49..661c6f3 100644
--- a/drivers/lightnvm/rrpc.c
+++ b/drivers/lightnvm/rrpc.c
@@ -642,7 +642,7 @@ static void rrpc_end_io_write(struct rrpc *rrpc, struct rrpc_rq *rrqd,
 	}
 }
 
-static int rrpc_end_io(struct nvm_rq *rqd, int error)
+static void rrpc_end_io(struct nvm_rq *rqd, int error)
 {
 	struct rrpc *rrpc = container_of(rqd->ins, struct rrpc, instance);
 	struct rrpc_rq *rrqd = nvm_rq_to_pdu(rqd);
@@ -655,7 +655,7 @@ static int rrpc_end_io(struct nvm_rq *rqd, int error)
 	bio_put(rqd->bio);
 
 	if (rrqd->flags & NVM_IOTYPE_GC)
-		return 0;
+		return;
 
 	rrpc_unlock_rq(rrpc, rqd);
 
@@ -665,8 +665,6 @@ static int rrpc_end_io(struct nvm_rq *rqd, int error)
 		nvm_dev_dma_free(rrpc->dev, rqd->metadata, rqd->dma_metadata);
 
 	mempool_free(rqd, rrpc->rq_pool);
-
-	return 0;
 }
 
 static int rrpc_read_ppalist_rq(struct rrpc *rrpc, struct bio *bio,
diff --git a/drivers/nvme/host/lightnvm.c b/drivers/nvme/host/lightnvm.c
index 15f2acb..1d1830e 100644
--- a/drivers/nvme/host/lightnvm.c
+++ b/drivers/nvme/host/lightnvm.c
@@ -453,11 +453,8 @@ static inline void nvme_nvm_rqtocmd(struct request *rq, struct nvm_rq *rqd,
 static void nvme_nvm_end_io(struct request *rq, int error)
 {
 	struct nvm_rq *rqd = rq->end_io_data;
-	struct nvm_dev *dev = rqd->dev;
 
-	if (dev->mt && dev->mt->end_io(rqd, error))
-		pr_err("nvme: err status: %x result: %lx\n",
-				rq->errors, (unsigned long)rq->special);
+	nvm_end_io(rqd, error);
 
 	kfree(rq->cmd);
 	blk_mq_free_request(rq);
diff --git a/include/linux/lightnvm.h b/include/linux/lightnvm.h
index 2fd6871..9c9fe9c 100644
--- a/include/linux/lightnvm.h
+++ b/include/linux/lightnvm.h
@@ -148,6 +148,9 @@ struct ppa_addr {
 	};
 };
 
+struct nvm_rq;
+typedef void (nvm_end_io_fn)(struct nvm_rq *, int);
+
 struct nvm_rq {
 	struct nvm_tgt_instance *ins;
 	struct nvm_dev *dev;
@@ -164,6 +167,9 @@ struct nvm_rq {
 	void *metadata;
 	dma_addr_t dma_metadata;
 
+	struct completion *wait;
+	nvm_end_io_fn *end_io;
+
 	uint8_t opcode;
 	uint16_t nr_pages;
 	uint16_t flags;
@@ -347,7 +353,6 @@ static inline struct ppa_addr block_to_ppa(struct nvm_dev *dev,
 
 typedef blk_qc_t (nvm_tgt_make_rq_fn)(struct request_queue *, struct bio *);
 typedef sector_t (nvm_tgt_capacity_fn)(void *);
-typedef int (nvm_tgt_end_io_fn)(struct nvm_rq *, int);
 typedef void *(nvm_tgt_init_fn)(struct nvm_dev *, struct gendisk *, int, int);
 typedef void (nvm_tgt_exit_fn)(void *);
 
@@ -358,7 +363,7 @@ struct nvm_tgt_type {
 	/* target entry points */
 	nvm_tgt_make_rq_fn *make_rq;
 	nvm_tgt_capacity_fn *capacity;
-	nvm_tgt_end_io_fn *end_io;
+	nvm_end_io_fn *end_io;
 
 	/* module-specific init/teardown */
 	nvm_tgt_init_fn *init;
@@ -383,7 +388,6 @@ typedef int (nvmm_open_blk_fn)(struct nvm_dev *, struct nvm_block *);
 typedef int (nvmm_close_blk_fn)(struct nvm_dev *, struct nvm_block *);
 typedef void (nvmm_flush_blk_fn)(struct nvm_dev *, struct nvm_block *);
 typedef int (nvmm_submit_io_fn)(struct nvm_dev *, struct nvm_rq *);
-typedef int (nvmm_end_io_fn)(struct nvm_rq *, int);
 typedef int (nvmm_erase_blk_fn)(struct nvm_dev *, struct nvm_block *,
 								unsigned long);
 typedef struct nvm_lun *(nvmm_get_lun_fn)(struct nvm_dev *, int);
@@ -404,7 +408,6 @@ struct nvmm_type {
 	nvmm_flush_blk_fn *flush_blk;
 
 	nvmm_submit_io_fn *submit_io;
-	nvmm_end_io_fn *end_io;
 	nvmm_erase_blk_fn *erase_blk;
 
 	/* Configuration management */
@@ -434,6 +437,7 @@ extern int nvm_set_rqd_ppalist(struct nvm_dev *, struct nvm_rq *,
 extern void nvm_free_rqd_ppalist(struct nvm_dev *, struct nvm_rq *);
 extern int nvm_erase_ppa(struct nvm_dev *, struct ppa_addr);
 extern int nvm_erase_blk(struct nvm_dev *, struct nvm_block *);
+extern void nvm_end_io(struct nvm_rq *, int);
 #else /* CONFIG_NVM */
 struct nvm_dev_ops;
 
-- 
2.1.4

[toc] | [prev] | [next] | [standalone]

#1307034 — [PATCH 20/25] lightnvm: add mccap support

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

[toc] | [prev] | [next] | [standalone]

#1307035 — [PATCH 05/25] lightnvm: move ppa erase logic to core

A device may function in single, dual or quad plane mode. The gennvm
media manager manages this with explicit helpers. They convert a single
ppa to 1, 2 or 4 separate ppas in a ppa list. To aid implementation of
recovery and system blocks, this functionality can be moved directly
into the core.

Signed-off-by: Matias Bjørling <m@bjorling.me>
---
 drivers/lightnvm/core.c   | 67 ++++++++++++++++++++++++++++++++++++++++++++++
 drivers/lightnvm/gennvm.c | 68 +++--------------------------------------------
 include/linux/lightnvm.h  |  3 +++
 3 files changed, 74 insertions(+), 64 deletions(-)

diff --git a/drivers/lightnvm/core.c b/drivers/lightnvm/core.c
index 8f41b24..6134339 100644
--- a/drivers/lightnvm/core.c
+++ b/drivers/lightnvm/core.c
@@ -192,6 +192,73 @@ int nvm_erase_blk(struct nvm_dev *dev, struct nvm_block *blk)
 }
 EXPORT_SYMBOL(nvm_erase_blk);
 
+void nvm_addr_to_generic_mode(struct nvm_dev *dev, struct nvm_rq *rqd)
+{
+	int i;
+
+	if (rqd->nr_pages > 1) {
+		for (i = 0; i < rqd->nr_pages; i++)
+			rqd->ppa_list[i] = dev_to_generic_addr(dev,
+							rqd->ppa_list[i]);
+	} else {
+		rqd->ppa_addr = dev_to_generic_addr(dev, rqd->ppa_addr);
+	}
+}
+EXPORT_SYMBOL(nvm_addr_to_generic_mode);
+
+void nvm_generic_to_addr_mode(struct nvm_dev *dev, struct nvm_rq *rqd)
+{
+	int i;
+
+	if (rqd->nr_pages > 1) {
+		for (i = 0; i < rqd->nr_pages; i++)
+			rqd->ppa_list[i] = generic_to_dev_addr(dev,
+							rqd->ppa_list[i]);
+	} else {
+		rqd->ppa_addr = generic_to_dev_addr(dev, rqd->ppa_addr);
+	}
+}
+EXPORT_SYMBOL(nvm_generic_to_addr_mode);
+
+int nvm_erase_ppa(struct nvm_dev *dev, struct ppa_addr ppa)
+{
+	int plane_cnt = 0, pl_idx, ret;
+	struct nvm_rq rqd;
+
+	if (!dev->ops->erase_block)
+		return 0;
+
+	if (dev->plane_mode == NVM_PLANE_SINGLE) {
+		rqd.nr_pages = 1;
+		rqd.ppa_addr = ppa;
+	} else {
+		plane_cnt = (1 << dev->plane_mode);
+		rqd.nr_pages = plane_cnt;
+
+		rqd.ppa_list = nvm_dev_dma_alloc(dev, GFP_KERNEL,
+							&rqd.dma_ppa_list);
+		if (!rqd.ppa_list) {
+			pr_err("nvm: failed to allocate dma memory\n");
+			return -ENOMEM;
+		}
+
+		for (pl_idx = 0; pl_idx < plane_cnt; pl_idx++) {
+			ppa.g.pl = pl_idx;
+			rqd.ppa_list[pl_idx] = ppa;
+		}
+	}
+
+	nvm_generic_to_addr_mode(dev, &rqd);
+
+	ret = dev->ops->erase_block(dev, &rqd);
+
+	if (plane_cnt)
+		nvm_dev_dma_free(dev, rqd.ppa_list, rqd.dma_ppa_list);
+
+	return ret;
+}
+EXPORT_SYMBOL(nvm_erase_ppa);
+
 static int nvm_core_init(struct nvm_dev *dev)
 {
 	struct nvm_id *id = &dev->identity;
diff --git a/drivers/lightnvm/gennvm.c b/drivers/lightnvm/gennvm.c
index 2a96ff6..373be72 100644
--- a/drivers/lightnvm/gennvm.c
+++ b/drivers/lightnvm/gennvm.c
@@ -317,39 +317,13 @@ static void gennvm_put_blk(struct nvm_dev *dev, struct nvm_block *blk)
 	spin_unlock(&vlun->lock);
 }
 
-static void gennvm_addr_to_generic_mode(struct nvm_dev *dev, struct nvm_rq *rqd)
-{
-	int i;
-
-	if (rqd->nr_pages > 1) {
-		for (i = 0; i < rqd->nr_pages; i++)
-			rqd->ppa_list[i] = dev_to_generic_addr(dev,
-							rqd->ppa_list[i]);
-	} else {
-		rqd->ppa_addr = dev_to_generic_addr(dev, rqd->ppa_addr);
-	}
-}
-
-static void gennvm_generic_to_addr_mode(struct nvm_dev *dev, struct nvm_rq *rqd)
-{
-	int i;
-
-	if (rqd->nr_pages > 1) {
-		for (i = 0; i < rqd->nr_pages; i++)
-			rqd->ppa_list[i] = generic_to_dev_addr(dev,
-							rqd->ppa_list[i]);
-	} else {
-		rqd->ppa_addr = generic_to_dev_addr(dev, rqd->ppa_addr);
-	}
-}
-
 static int gennvm_submit_io(struct nvm_dev *dev, struct nvm_rq *rqd)
 {
 	if (!dev->ops->submit_io)
 		return -ENODEV;
 
 	/* Convert address space */
-	gennvm_generic_to_addr_mode(dev, rqd);
+	nvm_generic_to_addr_mode(dev, rqd);
 
 	rqd->dev = dev;
 	return dev->ops->submit_io(dev, rqd);
@@ -391,7 +365,7 @@ static void gennvm_mark_blk_bad(struct nvm_dev *dev, struct nvm_rq *rqd)
 	if (dev->ops->set_bb_tbl(dev, rqd, 1))
 		return;
 
-	gennvm_addr_to_generic_mode(dev, rqd);
+	nvm_addr_to_generic_mode(dev, rqd);
 
 	/* look up blocks and mark them as bad */
 	if (rqd->nr_pages > 1)
@@ -425,43 +399,9 @@ static int gennvm_end_io(struct nvm_rq *rqd, int error)
 static int gennvm_erase_blk(struct nvm_dev *dev, struct nvm_block *blk,
 							unsigned long flags)
 {
-	int plane_cnt = 0, pl_idx, ret;
-	struct ppa_addr addr;
-	struct nvm_rq rqd;
+	struct ppa_addr addr = block_to_ppa(dev, blk);
 
-	if (!dev->ops->erase_block)
-		return 0;
-
-	addr = block_to_ppa(dev, blk);
-
-	if (dev->plane_mode == NVM_PLANE_SINGLE) {
-		rqd.nr_pages = 1;
-		rqd.ppa_addr = addr;
-	} else {
-		plane_cnt = (1 << dev->plane_mode);
-		rqd.nr_pages = plane_cnt;
-
-		rqd.ppa_list = nvm_dev_dma_alloc(dev, GFP_KERNEL,
-							&rqd.dma_ppa_list);
-		if (!rqd.ppa_list) {
-			pr_err("gennvm: failed to allocate dma memory\n");
-			return -ENOMEM;
-		}
-
-		for (pl_idx = 0; pl_idx < plane_cnt; pl_idx++) {
-			addr.g.pl = pl_idx;
-			rqd.ppa_list[pl_idx] = addr;
-		}
-	}
-
-	gennvm_generic_to_addr_mode(dev, &rqd);
-
-	ret = dev->ops->erase_block(dev, &rqd);
-
-	if (plane_cnt)
-		nvm_dev_dma_free(dev, rqd.ppa_list, rqd.dma_ppa_list);
-
-	return ret;
+	return nvm_erase_ppa(dev, addr);
 }
 
 static struct nvm_lun *gennvm_get_lun(struct nvm_dev *dev, int lunid)
diff --git a/include/linux/lightnvm.h b/include/linux/lightnvm.h
index 034117b..c228dbc 100644
--- a/include/linux/lightnvm.h
+++ b/include/linux/lightnvm.h
@@ -427,6 +427,9 @@ extern int nvm_register(struct request_queue *, char *,
 extern void nvm_unregister(char *);
 
 extern int nvm_submit_io(struct nvm_dev *, struct nvm_rq *);
+extern void nvm_generic_to_addr_mode(struct nvm_dev *, struct nvm_rq *);
+extern void nvm_addr_to_generic_mode(struct nvm_dev *, struct nvm_rq *);
+extern int nvm_erase_ppa(struct nvm_dev *, struct ppa_addr);
 extern int nvm_erase_blk(struct nvm_dev *, struct nvm_block *);
 #else /* CONFIG_NVM */
 struct nvm_dev_ops;
-- 
2.1.4

[toc] | [prev] | [next] | [standalone]

#1307036 — [PATCH 09/25] lightnvm: check bi_error in gc

From: Wenwei Tao <ww.tao0320@gmail.com>

We should check last io completion status before
starting another one.

Signed-off-by: Wenwei Tao <ww.tao0320@gmail.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
---
 drivers/lightnvm/rrpc.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/lightnvm/rrpc.c b/drivers/lightnvm/rrpc.c
index 661c6f3..5df4a69 100644
--- a/drivers/lightnvm/rrpc.c
+++ b/drivers/lightnvm/rrpc.c
@@ -330,6 +330,10 @@ try:
 			goto finished;
 		}
 		wait_for_completion_io(&wait);
+		if (bio->bi_error) {
+			rrpc_inflight_laddr_release(rrpc, rqd);
+			goto finished;
+		}
 
 		bio_reset(bio);
 		reinit_completion(&wait);
@@ -352,6 +356,8 @@ try:
 		wait_for_completion_io(&wait);
 
 		rrpc_inflight_laddr_release(rrpc, rqd);
+		if (bio->bi_error)
+			goto finished;
 
 		bio_reset(bio);
 	}
-- 
2.1.4

[toc] | [prev] | [next] | [standalone]

#1307037 — [PATCH 22/25] lightnvm: core on-disk initialization

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

[toc] | [prev] | [next] | [standalone]

#1307528

On 01/11/2016 11:49 PM, Matias Bjørling wrote:
> Hi Jens,
>
> Here is the collected patches for LightNVM ready for the 4.5 window.

This can't possibly be for the 4.5 window, since that opened two days 
ago. The code isn't in linux-next either.

I'll be nice and make an exception since this is the first window post 
inclusion, but in the future, the big update of patches for a given 
kernel revision has to be posted and reviewed at least 1-2 weeks before 
the window opens.

-- 
Jens Axboe

[toc] | [prev] | [next] | [standalone]

#1307930

On 01/12/2016 04:19 PM, Jens Axboe wrote:
> On 01/11/2016 11:49 PM, Matias Bjørling wrote:
>> Hi Jens,
>>
>> Here is the collected patches for LightNVM ready for the 4.5 window.
> 
> This can't possibly be for the 4.5 window, since that opened two days
> ago. The code isn't in linux-next either.
> 
> I'll be nice and make an exception since this is the first window post
> inclusion, but in the future, the big update of patches for a given
> kernel revision has to be posted and reviewed at least 1-2 weeks before
> the window opens.
> 

Thanks Jens. Won't happen again.

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web