s390: add scm block driver

Block device driver for Storage Class Memory (SCM). This driver
provides a block device interface for each available SCM increment.

Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

5 files changed, 597 insertions, 0 deletions

diff --git a/drivers/s390/block/Kconfig b/drivers/s390/block/Kconfig
index 8e477bb1f3f6..18178b64e89a 100644
--- a/drivers/s390/block/Kconfig
+++ b/drivers/s390/block/Kconfig
@@ -70,3 +70,14 @@ config DASD_EER
 	  This driver provides a character device interface to the
 	  DASD extended error reporting. This is only needed if you want to
 	  use applications written for the EER facility.
+
+config SCM_BLOCK
+	def_tristate m
+	prompt "Support for Storage Class Memory"
+	depends on S390 && BLOCK && EADM_SCH && SCM_BUS
+	help
+	  Block device driver for Storage Class Memory (SCM). This driver
+	  provides a block device interface for each available SCM increment.
+
+	  To compile this driver as a module, choose M here: the
+	  module will be called scm_block.
diff --git a/drivers/s390/block/Makefile b/drivers/s390/block/Makefile
index 0a89e080b389..b64e2b32c753 100644
--- a/drivers/s390/block/Makefile
+++ b/drivers/s390/block/Makefile
@@ -17,3 +17,6 @@ obj-$(CONFIG_DASD_ECKD) += dasd_eckd_mod.o
 obj-$(CONFIG_DASD_FBA)  += dasd_fba_mod.o
 obj-$(CONFIG_BLK_DEV_XPRAM) += xpram.o
 obj-$(CONFIG_DCSSBLK) += dcssblk.o
+
+scm_block-objs := scm_drv.o scm_blk.o
+obj-$(CONFIG_SCM_BLOCK) += scm_block.o
diff --git a/drivers/s390/block/scm_blk.c b/drivers/s390/block/scm_blk.c
new file mode 100644
index 000000000000..634ad58cbef6
--- /dev/null
+++ b/drivers/s390/block/scm_blk.c
@@ -0,0 +1,414 @@
+/*
+ * Block driver for s390 storage class memory.
+ *
+ * Copyright IBM Corp. 2012
+ * Author(s): Sebastian Ott <sebott@linux.vnet.ibm.com>
+ */
+
+#define KMSG_COMPONENT "scm_block"
+#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
+
+#include <linux/interrupt.h>
+#include <linux/spinlock.h>
+#include <linux/module.h>
+#include <linux/blkdev.h>
+#include <linux/genhd.h>
+#include <linux/slab.h>
+#include <linux/list.h>
+#include <asm/eadm.h>
+#include "scm_blk.h"
+
+debug_info_t *scm_debug;
+static int scm_major;
+static DEFINE_SPINLOCK(list_lock);
+static LIST_HEAD(inactive_requests);
+static unsigned int nr_requests = 64;
+static atomic_t nr_devices = ATOMIC_INIT(0);
+module_param(nr_requests, uint, S_IRUGO);
+MODULE_PARM_DESC(nr_requests, "Number of parallel requests.");
+
+MODULE_DESCRIPTION("Block driver for s390 storage class memory.");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS("scm:scmdev*");
+
+static void __scm_free_rq(struct scm_request *scmrq)
+{
+	struct aob_rq_header *aobrq = to_aobrq(scmrq);
+
+	free_page((unsigned long) scmrq->aob);
+	free_page((unsigned long) scmrq->aidaw);
+	kfree(aobrq);
+}
+
+static void scm_free_rqs(void)
+{
+	struct list_head *iter, *safe;
+	struct scm_request *scmrq;
+
+	spin_lock_irq(&list_lock);
+	list_for_each_safe(iter, safe, &inactive_requests) {
+		scmrq = list_entry(iter, struct scm_request, list);
+		list_del(&scmrq->list);
+		__scm_free_rq(scmrq);
+	}
+	spin_unlock_irq(&list_lock);
+}
+
+static int __scm_alloc_rq(void)
+{
+	struct aob_rq_header *aobrq;
+	struct scm_request *scmrq;
+
+	aobrq = kzalloc(sizeof(*aobrq) + sizeof(*scmrq), GFP_KERNEL);
+	if (!aobrq)
+		return -ENOMEM;
+
+	scmrq = (void *) aobrq->data;
+	scmrq->aidaw = (void *) get_zeroed_page(GFP_DMA);
+	scmrq->aob = (void *) get_zeroed_page(GFP_DMA);
+	if (!scmrq->aob || !scmrq->aidaw) {
+		__scm_free_rq(scmrq);
+		return -ENOMEM;
+	}
+	INIT_LIST_HEAD(&scmrq->list);
+	spin_lock_irq(&list_lock);
+	list_add(&scmrq->list, &inactive_requests);
+	spin_unlock_irq(&list_lock);
+
+	return 0;
+}
+
+static int scm_alloc_rqs(unsigned int nrqs)
+{
+	int ret = 0;
+
+	while (nrqs-- && !ret)
+		ret = __scm_alloc_rq();
+
+	return ret;
+}
+
+static struct scm_request *scm_request_fetch(void)
+{
+	struct scm_request *scmrq = NULL;
+
+	spin_lock(&list_lock);
+	if (list_empty(&inactive_requests))
+		goto out;
+	scmrq = list_first_entry(&inactive_requests, struct scm_request, list);
+	list_del(&scmrq->list);
+out:
+	spin_unlock(&list_lock);
+	return scmrq;
+}
+
+static void scm_request_done(struct scm_request *scmrq)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&list_lock, flags);
+	list_add(&scmrq->list, &inactive_requests);
+	spin_unlock_irqrestore(&list_lock, flags);
+}
+
+static int scm_open(struct block_device *blkdev, fmode_t mode)
+{
+	return scm_get_ref();
+}
+
+static int scm_release(struct gendisk *gendisk, fmode_t mode)
+{
+	scm_put_ref();
+	return 0;
+}
+
+static const struct block_device_operations scm_blk_devops = {
+	.owner = THIS_MODULE,
+	.open = scm_open,
+	.release = scm_release,
+};
+
+static void scm_request_prepare(struct scm_request *scmrq)
+{
+	struct scm_blk_dev *bdev = scmrq->bdev;
+	struct scm_device *scmdev = bdev->gendisk->private_data;
+	struct aidaw *aidaw = scmrq->aidaw;
+	struct msb *msb = &scmrq->aob->msb[0];
+	struct req_iterator iter;
+	struct bio_vec *bv;
+
+	msb->bs = MSB_BS_4K;
+	scmrq->aob->request.msb_count = 1;
+	msb->scm_addr = scmdev->address +
+		((u64) blk_rq_pos(scmrq->request) << 9);
+	msb->oc = (rq_data_dir(scmrq->request) == READ) ?
+		MSB_OC_READ : MSB_OC_WRITE;
+	msb->flags |= MSB_FLAG_IDA;
+	msb->data_addr = (u64) aidaw;
+
+	rq_for_each_segment(bv, scmrq->request, iter) {
+		WARN_ON(bv->bv_offset);
+		msb->blk_count += bv->bv_len >> 12;
+		aidaw->data_addr = (u64) page_address(bv->bv_page);
+		aidaw++;
+	}
+}
+
+static inline void scm_request_init(struct scm_blk_dev *bdev,
+				    struct scm_request *scmrq,
+				    struct request *req)
+{
+	struct aob_rq_header *aobrq = to_aobrq(scmrq);
+	struct aob *aob = scmrq->aob;
+
+	memset(aob, 0, sizeof(*aob));
+	memset(scmrq->aidaw, 0, PAGE_SIZE);
+	aobrq->scmdev = bdev->scmdev;
+	aob->request.cmd_code = ARQB_CMD_MOVE;
+	aob->request.data = (u64) aobrq;
+	scmrq->request = req;
+	scmrq->bdev = bdev;
+	scmrq->retries = 4;
+	scmrq->error = 0;
+}
+
+static void scm_ensure_queue_restart(struct scm_blk_dev *bdev)
+{
+	if (atomic_read(&bdev->queued_reqs)) {
+		/* Queue restart is triggered by the next interrupt. */
+		return;
+	}
+	blk_delay_queue(bdev->rq, SCM_QUEUE_DELAY);
+}
+
+static void scm_request_requeue(struct scm_request *scmrq)
+{
+	struct scm_blk_dev *bdev = scmrq->bdev;
+
+	blk_requeue_request(bdev->rq, scmrq->request);
+	scm_request_done(scmrq);
+	scm_ensure_queue_restart(bdev);
+}
+
+static void scm_request_finish(struct scm_request *scmrq)
+{
+	blk_end_request_all(scmrq->request, scmrq->error);
+	scm_request_done(scmrq);
+}
+
+static void scm_blk_request(struct request_queue *rq)
+{
+	struct scm_device *scmdev = rq->queuedata;
+	struct scm_blk_dev *bdev = dev_get_drvdata(&scmdev->dev);
+	struct scm_request *scmrq;
+	struct request *req;
+	int ret;
+
+	while ((req = blk_peek_request(rq))) {
+		if (req->cmd_type != REQ_TYPE_FS)
+			continue;
+
+		scmrq = scm_request_fetch();
+		if (!scmrq) {
+			SCM_LOG(5, "no request");
+			scm_ensure_queue_restart(bdev);
+			return;
+		}
+		scm_request_init(bdev, scmrq, req);
+		scm_request_prepare(scmrq);
+		blk_start_request(req);
+
+		ret = scm_start_aob(scmrq->aob);
+		if (ret) {
+			SCM_LOG(5, "no subchannel");
+			scm_request_requeue(scmrq);
+			return;
+		}
+		atomic_inc(&bdev->queued_reqs);
+	}
+}
+
+static void __scmrq_log_error(struct scm_request *scmrq)
+{
+	struct aob *aob = scmrq->aob;
+
+	if (scmrq->error == -ETIMEDOUT)
+		SCM_LOG(1, "Request timeout");
+	else {
+		SCM_LOG(1, "Request error");
+		SCM_LOG_HEX(1, &aob->response, sizeof(aob->response));
+	}
+	if (scmrq->retries)
+		SCM_LOG(1, "Retry request");
+	else
+		pr_err("An I/O operation to SCM failed with rc=%d\n",
+		       scmrq->error);
+}
+
+void scm_blk_irq(struct scm_device *scmdev, void *data, int error)
+{
+	struct scm_request *scmrq = data;
+	struct scm_blk_dev *bdev = scmrq->bdev;
+
+	scmrq->error = error;
+	if (error)
+		__scmrq_log_error(scmrq);
+
+	spin_lock(&bdev->lock);
+	list_add_tail(&scmrq->list, &bdev->finished_requests);
+	spin_unlock(&bdev->lock);
+	tasklet_hi_schedule(&bdev->tasklet);
+}
+
+static void scm_blk_tasklet(struct scm_blk_dev *bdev)
+{
+	struct scm_request *scmrq;
+	unsigned long flags;
+
+	spin_lock_irqsave(&bdev->lock, flags);
+	while (!list_empty(&bdev->finished_requests)) {
+		scmrq = list_first_entry(&bdev->finished_requests,
+					 struct scm_request, list);
+		list_del(&scmrq->list);
+		spin_unlock_irqrestore(&bdev->lock, flags);
+
+		if (scmrq->error && scmrq->retries-- > 0) {
+			if (scm_start_aob(scmrq->aob)) {
+				spin_lock_irqsave(&bdev->rq_lock, flags);
+				scm_request_requeue(scmrq);
+				spin_unlock_irqrestore(&bdev->rq_lock, flags);
+			}
+			/* Request restarted or requeued, handle next. */
+			spin_lock_irqsave(&bdev->lock, flags);
+			continue;
+		}
+		scm_request_finish(scmrq);
+		atomic_dec(&bdev->queued_reqs);
+		spin_lock_irqsave(&bdev->lock, flags);
+	}
+	spin_unlock_irqrestore(&bdev->lock, flags);
+	/* Look out for more requests. */
+	blk_run_queue(bdev->rq);
+}
+
+int scm_blk_dev_setup(struct scm_blk_dev *bdev, struct scm_device *scmdev)
+{
+	struct request_queue *rq;
+	int len, ret = -ENOMEM;
+	unsigned int devindex, nr_max_blk;
+
+	devindex = atomic_inc_return(&nr_devices) - 1;
+	/* scma..scmz + scmaa..scmzz */
+	if (devindex > 701) {
+		ret = -ENODEV;
+		goto out;
+	}
+
+	bdev->scmdev = scmdev;
+	spin_lock_init(&bdev->rq_lock);
+	spin_lock_init(&bdev->lock);
+	INIT_LIST_HEAD(&bdev->finished_requests);
+	atomic_set(&bdev->queued_reqs, 0);
+	tasklet_init(&bdev->tasklet,
+		     (void (*)(unsigned long)) scm_blk_tasklet,
+		     (unsigned long) bdev);
+
+	rq = blk_init_queue(scm_blk_request, &bdev->rq_lock);
+	if (!rq)
+		goto out;
+
+	bdev->rq = rq;
+	nr_max_blk = min(scmdev->nr_max_block,
+			 (unsigned int) (PAGE_SIZE / sizeof(struct aidaw)));
+
+	blk_queue_logical_block_size(rq, 1 << 12);
+	blk_queue_max_hw_sectors(rq, nr_max_blk << 3); /* 8 * 512 = blk_size */
+	blk_queue_max_segments(rq, nr_max_blk);
+	queue_flag_set_unlocked(QUEUE_FLAG_NONROT, rq);
+
+	bdev->gendisk = alloc_disk(SCM_NR_PARTS);
+	if (!bdev->gendisk)
+		goto out_queue;
+
+	rq->queuedata = scmdev;
+	bdev->gendisk->driverfs_dev = &scmdev->dev;
+	bdev->gendisk->private_data = scmdev;
+	bdev->gendisk->fops = &scm_blk_devops;
+	bdev->gendisk->queue = rq;
+	bdev->gendisk->major = scm_major;
+	bdev->gendisk->first_minor = devindex * SCM_NR_PARTS;
+
+	len = snprintf(bdev->gendisk->disk_name, DISK_NAME_LEN, "scm");
+	if (devindex > 25) {
+		len += snprintf(bdev->gendisk->disk_name + len,
+				DISK_NAME_LEN - len, "%c",
+				'a' + (devindex / 26) - 1);
+		devindex = devindex % 26;
+	}
+	snprintf(bdev->gendisk->disk_name + len, DISK_NAME_LEN - len, "%c",
+		 'a' + devindex);
+
+	/* 512 byte sectors */
+	set_capacity(bdev->gendisk, scmdev->size >> 9);
+	add_disk(bdev->gendisk);
+	return 0;
+
+out_queue:
+	blk_cleanup_queue(rq);
+out:
+	atomic_dec(&nr_devices);
+	return ret;
+}
+
+void scm_blk_dev_cleanup(struct scm_blk_dev *bdev)
+{
+	tasklet_kill(&bdev->tasklet);
+	del_gendisk(bdev->gendisk);
+	blk_cleanup_queue(bdev->gendisk->queue);
+	put_disk(bdev->gendisk);
+}
+
+static int __init scm_blk_init(void)
+{
+	int ret;
+
+	ret = register_blkdev(0, "scm");
+	if (ret < 0)
+		goto out;
+
+	scm_major = ret;
+	if (scm_alloc_rqs(nr_requests))
+		goto out_unreg;
+
+	scm_debug = debug_register("scm_log", 16, 1, 16);
+	if (!scm_debug)
+		goto out_free;
+
+	debug_register_view(scm_debug, &debug_hex_ascii_view);
+	debug_set_level(scm_debug, 2);
+
+	ret = scm_drv_init();
+	if (ret)
+		goto out_dbf;
+
+	return ret;
+
+out_dbf:
+	debug_unregister(scm_debug);
+out_free:
+	scm_free_rqs();
+out_unreg:
+	unregister_blkdev(scm_major, "scm");
+out:
+	return ret;
+}
+module_init(scm_blk_init);
+
+static void __exit scm_blk_cleanup(void)
+{
+	scm_drv_cleanup();
+	debug_unregister(scm_debug);
+	scm_free_rqs();
+	unregister_blkdev(scm_major, "scm");
+}
+module_exit(scm_blk_cleanup);
diff --git a/drivers/s390/block/scm_blk.h b/drivers/s390/block/scm_blk.h
new file mode 100644
index 000000000000..5aba5612588f
--- /dev/null
+++ b/drivers/s390/block/scm_blk.h
@@ -0,0 +1,79 @@
+#ifndef SCM_BLK_H
+#define SCM_BLK_H
+
+#include <linux/interrupt.h>
+#include <linux/spinlock.h>
+#include <linux/blkdev.h>
+#include <linux/genhd.h>
+#include <linux/list.h>
+
+#include <asm/debug.h>
+#include <asm/eadm.h>
+
+#define SCM_NR_PARTS 8
+#define SCM_QUEUE_DELAY 5
+
+struct scm_blk_dev {
+	struct tasklet_struct tasklet;
+	struct request_queue *rq;
+	struct gendisk *gendisk;
+	struct scm_device *scmdev;
+	spinlock_t rq_lock;	/* guard the request queue */
+	spinlock_t lock;	/* guard the rest of the blockdev */
+	atomic_t queued_reqs;
+	struct list_head finished_requests;
+};
+
+struct scm_request {
+	struct scm_blk_dev *bdev;
+	struct request *request;
+	struct aidaw *aidaw;
+	struct aob *aob;
+	struct list_head list;
+	u8 retries;
+	int error;
+};
+
+#define to_aobrq(rq) container_of((void *) rq, struct aob_rq_header, data)
+
+int scm_blk_dev_setup(struct scm_blk_dev *, struct scm_device *);
+void scm_blk_dev_cleanup(struct scm_blk_dev *);
+void scm_blk_irq(struct scm_device *, void *, int);
+
+int scm_drv_init(void);
+void scm_drv_cleanup(void);
+
+
+extern debug_info_t *scm_debug;
+
+#define SCM_LOG(imp, txt) do {					\
+		debug_text_event(scm_debug, imp, txt);		\
+	} while (0)
+
+static inline void SCM_LOG_HEX(int level, void *data, int length)
+{
+	if (level > scm_debug->level)
+		return;
+	while (length > 0) {
+		debug_event(scm_debug, level, data, length);
+		length -= scm_debug->buf_size;
+		data += scm_debug->buf_size;
+	}
+}
+
+static inline void SCM_LOG_STATE(int level, struct scm_device *scmdev)
+{
+	struct {
+		u64 address;
+		u8 oper_state;
+		u8 rank;
+	} __packed data = {
+		.address = scmdev->address,
+		.oper_state = scmdev->attrs.oper_state,
+		.rank = scmdev->attrs.rank,
+	};
+
+	SCM_LOG_HEX(level, &data, sizeof(data));
+}
+
+#endif /* SCM_BLK_H */
diff --git a/drivers/s390/block/scm_drv.c b/drivers/s390/block/scm_drv.c
new file mode 100644
index 000000000000..fce711a63060
--- /dev/null
+++ b/drivers/s390/block/scm_drv.c
@@ -0,0 +1,90 @@
+/*
+ * Device driver for s390 storage class memory.
+ *
+ * Copyright IBM Corp. 2012
+ * Author(s): Sebastian Ott <sebott@linux.vnet.ibm.com>
+ */
+
+#define KMSG_COMPONENT "scm_block"
+#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
+
+#include <linux/module.h>
+#include <linux/spinlock.h>
+#include <linux/slab.h>
+#include <asm/eadm.h>
+#include "scm_blk.h"
+
+static void notify(struct scm_device *scmdev)
+{
+	pr_info("%lu: The capabilities of the SCM increment changed\n",
+		(unsigned long) scmdev->address);
+	SCM_LOG(2, "State changed");
+	SCM_LOG_STATE(2, scmdev);
+}
+
+static int scm_probe(struct scm_device *scmdev)
+{
+	struct scm_blk_dev *bdev;
+	int ret;
+
+	SCM_LOG(2, "probe");
+	SCM_LOG_STATE(2, scmdev);
+
+	if (scmdev->attrs.oper_state != OP_STATE_GOOD)
+		return -EINVAL;
+
+	bdev = kzalloc(sizeof(*bdev), GFP_KERNEL);
+	if (!bdev)
+		return -ENOMEM;
+
+	spin_lock_irq(&scmdev->lock);
+	dev_set_drvdata(&scmdev->dev, bdev);
+	spin_unlock_irq(&scmdev->lock);
+
+	ret = scm_blk_dev_setup(bdev, scmdev);
+	if (ret) {
+		spin_lock_irq(&scmdev->lock);
+		dev_set_drvdata(&scmdev->dev, NULL);
+		spin_unlock_irq(&scmdev->lock);
+		kfree(bdev);
+		goto out;
+	}
+
+out:
+	return ret;
+}
+
+static int scm_remove(struct scm_device *scmdev)
+{
+	struct scm_blk_dev *bdev;
+
+	spin_lock_irq(&scmdev->lock);
+	bdev = dev_get_drvdata(&scmdev->dev);
+	dev_set_drvdata(&scmdev->dev, NULL);
+	spin_unlock_irq(&scmdev->lock);
+	scm_blk_dev_cleanup(bdev);
+	kfree(bdev);
+
+	return 0;
+}
+
+static struct scm_driver scm_drv = {
+	.drv = {
+		.name = "scm_block",
+		.owner = THIS_MODULE,
+	},
+	.notify = notify,
+	.probe = scm_probe,
+	.remove = scm_remove,
+	.handler = scm_blk_irq,
+};
+
+int __init scm_drv_init(void)
+{
+	return scm_driver_register(&scm_drv);
+}
+
+void scm_drv_cleanup(void)
+{
+	scm_driver_unregister(&scm_drv);
+}