These changes are the raw update to linux-4.4.6-rt14. Kernel sources
[kvmfornfv.git] / kernel / drivers / nvdimm / region_devs.c
diff --git a/kernel/drivers/nvdimm/region_devs.c b/kernel/drivers/nvdimm/region_devs.c
new file mode 100644 (file)
index 0000000..9521696
--- /dev/null
@@ -0,0 +1,756 @@
+/*
+ * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of version 2 of the GNU General Public License 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.
+ */
+#include <linux/scatterlist.h>
+#include <linux/highmem.h>
+#include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/sort.h>
+#include <linux/io.h>
+#include <linux/nd.h>
+#include "nd-core.h"
+#include "nd.h"
+
+static DEFINE_IDA(region_ida);
+
+static void nd_region_release(struct device *dev)
+{
+       struct nd_region *nd_region = to_nd_region(dev);
+       u16 i;
+
+       for (i = 0; i < nd_region->ndr_mappings; i++) {
+               struct nd_mapping *nd_mapping = &nd_region->mapping[i];
+               struct nvdimm *nvdimm = nd_mapping->nvdimm;
+
+               put_device(&nvdimm->dev);
+       }
+       free_percpu(nd_region->lane);
+       ida_simple_remove(&region_ida, nd_region->id);
+       if (is_nd_blk(dev))
+               kfree(to_nd_blk_region(dev));
+       else
+               kfree(nd_region);
+}
+
+static struct device_type nd_blk_device_type = {
+       .name = "nd_blk",
+       .release = nd_region_release,
+};
+
+static struct device_type nd_pmem_device_type = {
+       .name = "nd_pmem",
+       .release = nd_region_release,
+};
+
+static struct device_type nd_volatile_device_type = {
+       .name = "nd_volatile",
+       .release = nd_region_release,
+};
+
+bool is_nd_pmem(struct device *dev)
+{
+       return dev ? dev->type == &nd_pmem_device_type : false;
+}
+
+bool is_nd_blk(struct device *dev)
+{
+       return dev ? dev->type == &nd_blk_device_type : false;
+}
+
+struct nd_region *to_nd_region(struct device *dev)
+{
+       struct nd_region *nd_region = container_of(dev, struct nd_region, dev);
+
+       WARN_ON(dev->type->release != nd_region_release);
+       return nd_region;
+}
+EXPORT_SYMBOL_GPL(to_nd_region);
+
+struct nd_blk_region *to_nd_blk_region(struct device *dev)
+{
+       struct nd_region *nd_region = to_nd_region(dev);
+
+       WARN_ON(!is_nd_blk(dev));
+       return container_of(nd_region, struct nd_blk_region, nd_region);
+}
+EXPORT_SYMBOL_GPL(to_nd_blk_region);
+
+void *nd_region_provider_data(struct nd_region *nd_region)
+{
+       return nd_region->provider_data;
+}
+EXPORT_SYMBOL_GPL(nd_region_provider_data);
+
+void *nd_blk_region_provider_data(struct nd_blk_region *ndbr)
+{
+       return ndbr->blk_provider_data;
+}
+EXPORT_SYMBOL_GPL(nd_blk_region_provider_data);
+
+void nd_blk_region_set_provider_data(struct nd_blk_region *ndbr, void *data)
+{
+       ndbr->blk_provider_data = data;
+}
+EXPORT_SYMBOL_GPL(nd_blk_region_set_provider_data);
+
+/**
+ * nd_region_to_nstype() - region to an integer namespace type
+ * @nd_region: region-device to interrogate
+ *
+ * This is the 'nstype' attribute of a region as well, an input to the
+ * MODALIAS for namespace devices, and bit number for a nvdimm_bus to match
+ * namespace devices with namespace drivers.
+ */
+int nd_region_to_nstype(struct nd_region *nd_region)
+{
+       if (is_nd_pmem(&nd_region->dev)) {
+               u16 i, alias;
+
+               for (i = 0, alias = 0; i < nd_region->ndr_mappings; i++) {
+                       struct nd_mapping *nd_mapping = &nd_region->mapping[i];
+                       struct nvdimm *nvdimm = nd_mapping->nvdimm;
+
+                       if (nvdimm->flags & NDD_ALIASING)
+                               alias++;
+               }
+               if (alias)
+                       return ND_DEVICE_NAMESPACE_PMEM;
+               else
+                       return ND_DEVICE_NAMESPACE_IO;
+       } else if (is_nd_blk(&nd_region->dev)) {
+               return ND_DEVICE_NAMESPACE_BLK;
+       }
+
+       return 0;
+}
+EXPORT_SYMBOL(nd_region_to_nstype);
+
+static ssize_t size_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       struct nd_region *nd_region = to_nd_region(dev);
+       unsigned long long size = 0;
+
+       if (is_nd_pmem(dev)) {
+               size = nd_region->ndr_size;
+       } else if (nd_region->ndr_mappings == 1) {
+               struct nd_mapping *nd_mapping = &nd_region->mapping[0];
+
+               size = nd_mapping->size;
+       }
+
+       return sprintf(buf, "%llu\n", size);
+}
+static DEVICE_ATTR_RO(size);
+
+static ssize_t mappings_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       struct nd_region *nd_region = to_nd_region(dev);
+
+       return sprintf(buf, "%d\n", nd_region->ndr_mappings);
+}
+static DEVICE_ATTR_RO(mappings);
+
+static ssize_t nstype_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       struct nd_region *nd_region = to_nd_region(dev);
+
+       return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
+}
+static DEVICE_ATTR_RO(nstype);
+
+static ssize_t set_cookie_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       struct nd_region *nd_region = to_nd_region(dev);
+       struct nd_interleave_set *nd_set = nd_region->nd_set;
+
+       if (is_nd_pmem(dev) && nd_set)
+               /* pass, should be precluded by region_visible */;
+       else
+               return -ENXIO;
+
+       return sprintf(buf, "%#llx\n", nd_set->cookie);
+}
+static DEVICE_ATTR_RO(set_cookie);
+
+resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
+{
+       resource_size_t blk_max_overlap = 0, available, overlap;
+       int i;
+
+       WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
+
+ retry:
+       available = 0;
+       overlap = blk_max_overlap;
+       for (i = 0; i < nd_region->ndr_mappings; i++) {
+               struct nd_mapping *nd_mapping = &nd_region->mapping[i];
+               struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
+
+               /* if a dimm is disabled the available capacity is zero */
+               if (!ndd)
+                       return 0;
+
+               if (is_nd_pmem(&nd_region->dev)) {
+                       available += nd_pmem_available_dpa(nd_region,
+                                       nd_mapping, &overlap);
+                       if (overlap > blk_max_overlap) {
+                               blk_max_overlap = overlap;
+                               goto retry;
+                       }
+               } else if (is_nd_blk(&nd_region->dev)) {
+                       available += nd_blk_available_dpa(nd_mapping);
+               }
+       }
+
+       return available;
+}
+
+static ssize_t available_size_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       struct nd_region *nd_region = to_nd_region(dev);
+       unsigned long long available = 0;
+
+       /*
+        * Flush in-flight updates and grab a snapshot of the available
+        * size.  Of course, this value is potentially invalidated the
+        * memory nvdimm_bus_lock() is dropped, but that's userspace's
+        * problem to not race itself.
+        */
+       nvdimm_bus_lock(dev);
+       wait_nvdimm_bus_probe_idle(dev);
+       available = nd_region_available_dpa(nd_region);
+       nvdimm_bus_unlock(dev);
+
+       return sprintf(buf, "%llu\n", available);
+}
+static DEVICE_ATTR_RO(available_size);
+
+static ssize_t init_namespaces_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       struct nd_region_namespaces *num_ns = dev_get_drvdata(dev);
+       ssize_t rc;
+
+       nvdimm_bus_lock(dev);
+       if (num_ns)
+               rc = sprintf(buf, "%d/%d\n", num_ns->active, num_ns->count);
+       else
+               rc = -ENXIO;
+       nvdimm_bus_unlock(dev);
+
+       return rc;
+}
+static DEVICE_ATTR_RO(init_namespaces);
+
+static ssize_t namespace_seed_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       struct nd_region *nd_region = to_nd_region(dev);
+       ssize_t rc;
+
+       nvdimm_bus_lock(dev);
+       if (nd_region->ns_seed)
+               rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
+       else
+               rc = sprintf(buf, "\n");
+       nvdimm_bus_unlock(dev);
+       return rc;
+}
+static DEVICE_ATTR_RO(namespace_seed);
+
+static ssize_t btt_seed_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       struct nd_region *nd_region = to_nd_region(dev);
+       ssize_t rc;
+
+       nvdimm_bus_lock(dev);
+       if (nd_region->btt_seed)
+               rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
+       else
+               rc = sprintf(buf, "\n");
+       nvdimm_bus_unlock(dev);
+
+       return rc;
+}
+static DEVICE_ATTR_RO(btt_seed);
+
+static ssize_t pfn_seed_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       struct nd_region *nd_region = to_nd_region(dev);
+       ssize_t rc;
+
+       nvdimm_bus_lock(dev);
+       if (nd_region->pfn_seed)
+               rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
+       else
+               rc = sprintf(buf, "\n");
+       nvdimm_bus_unlock(dev);
+
+       return rc;
+}
+static DEVICE_ATTR_RO(pfn_seed);
+
+static ssize_t read_only_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       struct nd_region *nd_region = to_nd_region(dev);
+
+       return sprintf(buf, "%d\n", nd_region->ro);
+}
+
+static ssize_t read_only_store(struct device *dev,
+               struct device_attribute *attr, const char *buf, size_t len)
+{
+       bool ro;
+       int rc = strtobool(buf, &ro);
+       struct nd_region *nd_region = to_nd_region(dev);
+
+       if (rc)
+               return rc;
+
+       nd_region->ro = ro;
+       return len;
+}
+static DEVICE_ATTR_RW(read_only);
+
+static struct attribute *nd_region_attributes[] = {
+       &dev_attr_size.attr,
+       &dev_attr_nstype.attr,
+       &dev_attr_mappings.attr,
+       &dev_attr_btt_seed.attr,
+       &dev_attr_pfn_seed.attr,
+       &dev_attr_read_only.attr,
+       &dev_attr_set_cookie.attr,
+       &dev_attr_available_size.attr,
+       &dev_attr_namespace_seed.attr,
+       &dev_attr_init_namespaces.attr,
+       NULL,
+};
+
+static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
+{
+       struct device *dev = container_of(kobj, typeof(*dev), kobj);
+       struct nd_region *nd_region = to_nd_region(dev);
+       struct nd_interleave_set *nd_set = nd_region->nd_set;
+       int type = nd_region_to_nstype(nd_region);
+
+       if (a != &dev_attr_set_cookie.attr
+                       && a != &dev_attr_available_size.attr)
+               return a->mode;
+
+       if ((type == ND_DEVICE_NAMESPACE_PMEM
+                               || type == ND_DEVICE_NAMESPACE_BLK)
+                       && a == &dev_attr_available_size.attr)
+               return a->mode;
+       else if (is_nd_pmem(dev) && nd_set)
+               return a->mode;
+
+       return 0;
+}
+
+struct attribute_group nd_region_attribute_group = {
+       .attrs = nd_region_attributes,
+       .is_visible = region_visible,
+};
+EXPORT_SYMBOL_GPL(nd_region_attribute_group);
+
+u64 nd_region_interleave_set_cookie(struct nd_region *nd_region)
+{
+       struct nd_interleave_set *nd_set = nd_region->nd_set;
+
+       if (nd_set)
+               return nd_set->cookie;
+       return 0;
+}
+
+/*
+ * Upon successful probe/remove, take/release a reference on the
+ * associated interleave set (if present), and plant new btt + namespace
+ * seeds.  Also, on the removal of a BLK region, notify the provider to
+ * disable the region.
+ */
+static void nd_region_notify_driver_action(struct nvdimm_bus *nvdimm_bus,
+               struct device *dev, bool probe)
+{
+       struct nd_region *nd_region;
+
+       if (!probe && (is_nd_pmem(dev) || is_nd_blk(dev))) {
+               int i;
+
+               nd_region = to_nd_region(dev);
+               for (i = 0; i < nd_region->ndr_mappings; i++) {
+                       struct nd_mapping *nd_mapping = &nd_region->mapping[i];
+                       struct nvdimm_drvdata *ndd = nd_mapping->ndd;
+                       struct nvdimm *nvdimm = nd_mapping->nvdimm;
+
+                       kfree(nd_mapping->labels);
+                       nd_mapping->labels = NULL;
+                       put_ndd(ndd);
+                       nd_mapping->ndd = NULL;
+                       if (ndd)
+                               atomic_dec(&nvdimm->busy);
+               }
+
+               if (is_nd_pmem(dev))
+                       return;
+
+               to_nd_blk_region(dev)->disable(nvdimm_bus, dev);
+       }
+       if (dev->parent && is_nd_blk(dev->parent) && probe) {
+               nd_region = to_nd_region(dev->parent);
+               nvdimm_bus_lock(dev);
+               if (nd_region->ns_seed == dev)
+                       nd_region_create_blk_seed(nd_region);
+               nvdimm_bus_unlock(dev);
+       }
+       if (is_nd_btt(dev) && probe) {
+               struct nd_btt *nd_btt = to_nd_btt(dev);
+
+               nd_region = to_nd_region(dev->parent);
+               nvdimm_bus_lock(dev);
+               if (nd_region->btt_seed == dev)
+                       nd_region_create_btt_seed(nd_region);
+               if (nd_region->ns_seed == &nd_btt->ndns->dev &&
+                               is_nd_blk(dev->parent))
+                       nd_region_create_blk_seed(nd_region);
+               nvdimm_bus_unlock(dev);
+       }
+}
+
+void nd_region_probe_success(struct nvdimm_bus *nvdimm_bus, struct device *dev)
+{
+       nd_region_notify_driver_action(nvdimm_bus, dev, true);
+}
+
+void nd_region_disable(struct nvdimm_bus *nvdimm_bus, struct device *dev)
+{
+       nd_region_notify_driver_action(nvdimm_bus, dev, false);
+}
+
+static ssize_t mappingN(struct device *dev, char *buf, int n)
+{
+       struct nd_region *nd_region = to_nd_region(dev);
+       struct nd_mapping *nd_mapping;
+       struct nvdimm *nvdimm;
+
+       if (n >= nd_region->ndr_mappings)
+               return -ENXIO;
+       nd_mapping = &nd_region->mapping[n];
+       nvdimm = nd_mapping->nvdimm;
+
+       return sprintf(buf, "%s,%llu,%llu\n", dev_name(&nvdimm->dev),
+                       nd_mapping->start, nd_mapping->size);
+}
+
+#define REGION_MAPPING(idx) \
+static ssize_t mapping##idx##_show(struct device *dev,         \
+               struct device_attribute *attr, char *buf)       \
+{                                                              \
+       return mappingN(dev, buf, idx);                         \
+}                                                              \
+static DEVICE_ATTR_RO(mapping##idx)
+
+/*
+ * 32 should be enough for a while, even in the presence of socket
+ * interleave a 32-way interleave set is a degenerate case.
+ */
+REGION_MAPPING(0);
+REGION_MAPPING(1);
+REGION_MAPPING(2);
+REGION_MAPPING(3);
+REGION_MAPPING(4);
+REGION_MAPPING(5);
+REGION_MAPPING(6);
+REGION_MAPPING(7);
+REGION_MAPPING(8);
+REGION_MAPPING(9);
+REGION_MAPPING(10);
+REGION_MAPPING(11);
+REGION_MAPPING(12);
+REGION_MAPPING(13);
+REGION_MAPPING(14);
+REGION_MAPPING(15);
+REGION_MAPPING(16);
+REGION_MAPPING(17);
+REGION_MAPPING(18);
+REGION_MAPPING(19);
+REGION_MAPPING(20);
+REGION_MAPPING(21);
+REGION_MAPPING(22);
+REGION_MAPPING(23);
+REGION_MAPPING(24);
+REGION_MAPPING(25);
+REGION_MAPPING(26);
+REGION_MAPPING(27);
+REGION_MAPPING(28);
+REGION_MAPPING(29);
+REGION_MAPPING(30);
+REGION_MAPPING(31);
+
+static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
+{
+       struct device *dev = container_of(kobj, struct device, kobj);
+       struct nd_region *nd_region = to_nd_region(dev);
+
+       if (n < nd_region->ndr_mappings)
+               return a->mode;
+       return 0;
+}
+
+static struct attribute *mapping_attributes[] = {
+       &dev_attr_mapping0.attr,
+       &dev_attr_mapping1.attr,
+       &dev_attr_mapping2.attr,
+       &dev_attr_mapping3.attr,
+       &dev_attr_mapping4.attr,
+       &dev_attr_mapping5.attr,
+       &dev_attr_mapping6.attr,
+       &dev_attr_mapping7.attr,
+       &dev_attr_mapping8.attr,
+       &dev_attr_mapping9.attr,
+       &dev_attr_mapping10.attr,
+       &dev_attr_mapping11.attr,
+       &dev_attr_mapping12.attr,
+       &dev_attr_mapping13.attr,
+       &dev_attr_mapping14.attr,
+       &dev_attr_mapping15.attr,
+       &dev_attr_mapping16.attr,
+       &dev_attr_mapping17.attr,
+       &dev_attr_mapping18.attr,
+       &dev_attr_mapping19.attr,
+       &dev_attr_mapping20.attr,
+       &dev_attr_mapping21.attr,
+       &dev_attr_mapping22.attr,
+       &dev_attr_mapping23.attr,
+       &dev_attr_mapping24.attr,
+       &dev_attr_mapping25.attr,
+       &dev_attr_mapping26.attr,
+       &dev_attr_mapping27.attr,
+       &dev_attr_mapping28.attr,
+       &dev_attr_mapping29.attr,
+       &dev_attr_mapping30.attr,
+       &dev_attr_mapping31.attr,
+       NULL,
+};
+
+struct attribute_group nd_mapping_attribute_group = {
+       .is_visible = mapping_visible,
+       .attrs = mapping_attributes,
+};
+EXPORT_SYMBOL_GPL(nd_mapping_attribute_group);
+
+int nd_blk_region_init(struct nd_region *nd_region)
+{
+       struct device *dev = &nd_region->dev;
+       struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
+
+       if (!is_nd_blk(dev))
+               return 0;
+
+       if (nd_region->ndr_mappings < 1) {
+               dev_err(dev, "invalid BLK region\n");
+               return -ENXIO;
+       }
+
+       return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
+}
+
+/**
+ * nd_region_acquire_lane - allocate and lock a lane
+ * @nd_region: region id and number of lanes possible
+ *
+ * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
+ * We optimize for the common case where there are 256 lanes, one
+ * per-cpu.  For larger systems we need to lock to share lanes.  For now
+ * this implementation assumes the cost of maintaining an allocator for
+ * free lanes is on the order of the lock hold time, so it implements a
+ * static lane = cpu % num_lanes mapping.
+ *
+ * In the case of a BTT instance on top of a BLK namespace a lane may be
+ * acquired recursively.  We lock on the first instance.
+ *
+ * In the case of a BTT instance on top of PMEM, we only acquire a lane
+ * for the BTT metadata updates.
+ */
+unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
+{
+       unsigned int cpu, lane;
+
+       cpu = get_cpu();
+       if (nd_region->num_lanes < nr_cpu_ids) {
+               struct nd_percpu_lane *ndl_lock, *ndl_count;
+
+               lane = cpu % nd_region->num_lanes;
+               ndl_count = per_cpu_ptr(nd_region->lane, cpu);
+               ndl_lock = per_cpu_ptr(nd_region->lane, lane);
+               if (ndl_count->count++ == 0)
+                       spin_lock(&ndl_lock->lock);
+       } else
+               lane = cpu;
+
+       return lane;
+}
+EXPORT_SYMBOL(nd_region_acquire_lane);
+
+void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
+{
+       if (nd_region->num_lanes < nr_cpu_ids) {
+               unsigned int cpu = get_cpu();
+               struct nd_percpu_lane *ndl_lock, *ndl_count;
+
+               ndl_count = per_cpu_ptr(nd_region->lane, cpu);
+               ndl_lock = per_cpu_ptr(nd_region->lane, lane);
+               if (--ndl_count->count == 0)
+                       spin_unlock(&ndl_lock->lock);
+               put_cpu();
+       }
+       put_cpu();
+}
+EXPORT_SYMBOL(nd_region_release_lane);
+
+static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
+               struct nd_region_desc *ndr_desc, struct device_type *dev_type,
+               const char *caller)
+{
+       struct nd_region *nd_region;
+       struct device *dev;
+       void *region_buf;
+       unsigned int i;
+       int ro = 0;
+
+       for (i = 0; i < ndr_desc->num_mappings; i++) {
+               struct nd_mapping *nd_mapping = &ndr_desc->nd_mapping[i];
+               struct nvdimm *nvdimm = nd_mapping->nvdimm;
+
+               if ((nd_mapping->start | nd_mapping->size) % SZ_4K) {
+                       dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not 4K aligned\n",
+                                       caller, dev_name(&nvdimm->dev), i);
+
+                       return NULL;
+               }
+
+               if (nvdimm->flags & NDD_UNARMED)
+                       ro = 1;
+       }
+
+       if (dev_type == &nd_blk_device_type) {
+               struct nd_blk_region_desc *ndbr_desc;
+               struct nd_blk_region *ndbr;
+
+               ndbr_desc = to_blk_region_desc(ndr_desc);
+               ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
+                               * ndr_desc->num_mappings,
+                               GFP_KERNEL);
+               if (ndbr) {
+                       nd_region = &ndbr->nd_region;
+                       ndbr->enable = ndbr_desc->enable;
+                       ndbr->disable = ndbr_desc->disable;
+                       ndbr->do_io = ndbr_desc->do_io;
+               }
+               region_buf = ndbr;
+       } else {
+               nd_region = kzalloc(sizeof(struct nd_region)
+                               + sizeof(struct nd_mapping)
+                               * ndr_desc->num_mappings,
+                               GFP_KERNEL);
+               region_buf = nd_region;
+       }
+
+       if (!region_buf)
+               return NULL;
+       nd_region->id = ida_simple_get(&region_ida, 0, 0, GFP_KERNEL);
+       if (nd_region->id < 0)
+               goto err_id;
+
+       nd_region->lane = alloc_percpu(struct nd_percpu_lane);
+       if (!nd_region->lane)
+               goto err_percpu;
+
+        for (i = 0; i < nr_cpu_ids; i++) {
+               struct nd_percpu_lane *ndl;
+
+               ndl = per_cpu_ptr(nd_region->lane, i);
+               spin_lock_init(&ndl->lock);
+               ndl->count = 0;
+       }
+
+       memcpy(nd_region->mapping, ndr_desc->nd_mapping,
+                       sizeof(struct nd_mapping) * ndr_desc->num_mappings);
+       for (i = 0; i < ndr_desc->num_mappings; i++) {
+               struct nd_mapping *nd_mapping = &ndr_desc->nd_mapping[i];
+               struct nvdimm *nvdimm = nd_mapping->nvdimm;
+
+               get_device(&nvdimm->dev);
+       }
+       nd_region->ndr_mappings = ndr_desc->num_mappings;
+       nd_region->provider_data = ndr_desc->provider_data;
+       nd_region->nd_set = ndr_desc->nd_set;
+       nd_region->num_lanes = ndr_desc->num_lanes;
+       nd_region->flags = ndr_desc->flags;
+       nd_region->ro = ro;
+       nd_region->numa_node = ndr_desc->numa_node;
+       ida_init(&nd_region->ns_ida);
+       ida_init(&nd_region->btt_ida);
+       ida_init(&nd_region->pfn_ida);
+       dev = &nd_region->dev;
+       dev_set_name(dev, "region%d", nd_region->id);
+       dev->parent = &nvdimm_bus->dev;
+       dev->type = dev_type;
+       dev->groups = ndr_desc->attr_groups;
+       nd_region->ndr_size = resource_size(ndr_desc->res);
+       nd_region->ndr_start = ndr_desc->res->start;
+       nd_device_register(dev);
+
+       return nd_region;
+
+ err_percpu:
+       ida_simple_remove(&region_ida, nd_region->id);
+ err_id:
+       kfree(region_buf);
+       return NULL;
+}
+
+struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
+               struct nd_region_desc *ndr_desc)
+{
+       ndr_desc->num_lanes = ND_MAX_LANES;
+       return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
+                       __func__);
+}
+EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
+
+struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
+               struct nd_region_desc *ndr_desc)
+{
+       if (ndr_desc->num_mappings > 1)
+               return NULL;
+       ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
+       return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
+                       __func__);
+}
+EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
+
+struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
+               struct nd_region_desc *ndr_desc)
+{
+       ndr_desc->num_lanes = ND_MAX_LANES;
+       return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
+                       __func__);
+}
+EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);