Merge tag 'drm-amdkfd-fixes-2015-07-09' of git://people.freedesktop.org/~gabbayo...
[firefly-linux-kernel-4.4.55.git] / drivers / nvdimm / namespace_devs.c
index 4f653d1e61adce1fdc0016d6dd1856d19a91c3cc..fef0dd80d4adb18c10a62a89ae0a53a78c9ea091 100644 (file)
@@ -14,6 +14,7 @@
 #include <linux/device.h>
 #include <linux/slab.h>
 #include <linux/nd.h>
+#include "nd-core.h"
 #include "nd.h"
 
 static void namespace_io_release(struct device *dev)
@@ -23,11 +24,82 @@ static void namespace_io_release(struct device *dev)
        kfree(nsio);
 }
 
+static void namespace_pmem_release(struct device *dev)
+{
+       struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
+
+       kfree(nspm->alt_name);
+       kfree(nspm->uuid);
+       kfree(nspm);
+}
+
+static void namespace_blk_release(struct device *dev)
+{
+       struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
+       struct nd_region *nd_region = to_nd_region(dev->parent);
+
+       if (nsblk->id >= 0)
+               ida_simple_remove(&nd_region->ns_ida, nsblk->id);
+       kfree(nsblk->alt_name);
+       kfree(nsblk->uuid);
+       kfree(nsblk->res);
+       kfree(nsblk);
+}
+
 static struct device_type namespace_io_device_type = {
        .name = "nd_namespace_io",
        .release = namespace_io_release,
 };
 
+static struct device_type namespace_pmem_device_type = {
+       .name = "nd_namespace_pmem",
+       .release = namespace_pmem_release,
+};
+
+static struct device_type namespace_blk_device_type = {
+       .name = "nd_namespace_blk",
+       .release = namespace_blk_release,
+};
+
+static bool is_namespace_pmem(struct device *dev)
+{
+       return dev ? dev->type == &namespace_pmem_device_type : false;
+}
+
+static bool is_namespace_blk(struct device *dev)
+{
+       return dev ? dev->type == &namespace_blk_device_type : false;
+}
+
+static bool is_namespace_io(struct device *dev)
+{
+       return dev ? dev->type == &namespace_io_device_type : false;
+}
+
+const char *nvdimm_namespace_disk_name(struct nd_namespace_common *ndns,
+               char *name)
+{
+       struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
+       const char *suffix = "";
+
+       if (ndns->claim && is_nd_btt(ndns->claim))
+               suffix = "s";
+
+       if (is_namespace_pmem(&ndns->dev) || is_namespace_io(&ndns->dev))
+               sprintf(name, "pmem%d%s", nd_region->id, suffix);
+       else if (is_namespace_blk(&ndns->dev)) {
+               struct nd_namespace_blk *nsblk;
+
+               nsblk = to_nd_namespace_blk(&ndns->dev);
+               sprintf(name, "ndblk%d.%d%s", nd_region->id, nsblk->id, suffix);
+       } else {
+               return NULL;
+       }
+
+       return name;
+}
+EXPORT_SYMBOL(nvdimm_namespace_disk_name);
+
 static ssize_t nstype_show(struct device *dev,
                struct device_attribute *attr, char *buf)
 {
@@ -37,21 +109,1193 @@ static ssize_t nstype_show(struct device *dev,
 }
 static DEVICE_ATTR_RO(nstype);
 
+static ssize_t __alt_name_store(struct device *dev, const char *buf,
+               const size_t len)
+{
+       char *input, *pos, *alt_name, **ns_altname;
+       ssize_t rc;
+
+       if (is_namespace_pmem(dev)) {
+               struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
+
+               ns_altname = &nspm->alt_name;
+       } else if (is_namespace_blk(dev)) {
+               struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
+
+               ns_altname = &nsblk->alt_name;
+       } else
+               return -ENXIO;
+
+       if (dev->driver || to_ndns(dev)->claim)
+               return -EBUSY;
+
+       input = kmemdup(buf, len + 1, GFP_KERNEL);
+       if (!input)
+               return -ENOMEM;
+
+       input[len] = '\0';
+       pos = strim(input);
+       if (strlen(pos) + 1 > NSLABEL_NAME_LEN) {
+               rc = -EINVAL;
+               goto out;
+       }
+
+       alt_name = kzalloc(NSLABEL_NAME_LEN, GFP_KERNEL);
+       if (!alt_name) {
+               rc = -ENOMEM;
+               goto out;
+       }
+       kfree(*ns_altname);
+       *ns_altname = alt_name;
+       sprintf(*ns_altname, "%s", pos);
+       rc = len;
+
+out:
+       kfree(input);
+       return rc;
+}
+
+static resource_size_t nd_namespace_blk_size(struct nd_namespace_blk *nsblk)
+{
+       struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
+       struct nd_mapping *nd_mapping = &nd_region->mapping[0];
+       struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
+       struct nd_label_id label_id;
+       resource_size_t size = 0;
+       struct resource *res;
+
+       if (!nsblk->uuid)
+               return 0;
+       nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
+       for_each_dpa_resource(ndd, res)
+               if (strcmp(res->name, label_id.id) == 0)
+                       size += resource_size(res);
+       return size;
+}
+
+static bool __nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
+{
+       struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
+       struct nd_mapping *nd_mapping = &nd_region->mapping[0];
+       struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
+       struct nd_label_id label_id;
+       struct resource *res;
+       int count, i;
+
+       if (!nsblk->uuid || !nsblk->lbasize || !ndd)
+               return false;
+
+       count = 0;
+       nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
+       for_each_dpa_resource(ndd, res) {
+               if (strcmp(res->name, label_id.id) != 0)
+                       continue;
+               /*
+                * Resources with unacknoweldged adjustments indicate a
+                * failure to update labels
+                */
+               if (res->flags & DPA_RESOURCE_ADJUSTED)
+                       return false;
+               count++;
+       }
+
+       /* These values match after a successful label update */
+       if (count != nsblk->num_resources)
+               return false;
+
+       for (i = 0; i < nsblk->num_resources; i++) {
+               struct resource *found = NULL;
+
+               for_each_dpa_resource(ndd, res)
+                       if (res == nsblk->res[i]) {
+                               found = res;
+                               break;
+                       }
+               /* stale resource */
+               if (!found)
+                       return false;
+       }
+
+       return true;
+}
+
+resource_size_t nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
+{
+       resource_size_t size;
+
+       nvdimm_bus_lock(&nsblk->common.dev);
+       size = __nd_namespace_blk_validate(nsblk);
+       nvdimm_bus_unlock(&nsblk->common.dev);
+
+       return size;
+}
+EXPORT_SYMBOL(nd_namespace_blk_validate);
+
+
+static int nd_namespace_label_update(struct nd_region *nd_region,
+               struct device *dev)
+{
+       dev_WARN_ONCE(dev, dev->driver || to_ndns(dev)->claim,
+                       "namespace must be idle during label update\n");
+       if (dev->driver || to_ndns(dev)->claim)
+               return 0;
+
+       /*
+        * Only allow label writes that will result in a valid namespace
+        * or deletion of an existing namespace.
+        */
+       if (is_namespace_pmem(dev)) {
+               struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
+               resource_size_t size = resource_size(&nspm->nsio.res);
+
+               if (size == 0 && nspm->uuid)
+                       /* delete allocation */;
+               else if (!nspm->uuid)
+                       return 0;
+
+               return nd_pmem_namespace_label_update(nd_region, nspm, size);
+       } else if (is_namespace_blk(dev)) {
+               struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
+               resource_size_t size = nd_namespace_blk_size(nsblk);
+
+               if (size == 0 && nsblk->uuid)
+                       /* delete allocation */;
+               else if (!nsblk->uuid || !nsblk->lbasize)
+                       return 0;
+
+               return nd_blk_namespace_label_update(nd_region, nsblk, size);
+       } else
+               return -ENXIO;
+}
+
+static ssize_t alt_name_store(struct device *dev,
+               struct device_attribute *attr, const char *buf, size_t len)
+{
+       struct nd_region *nd_region = to_nd_region(dev->parent);
+       ssize_t rc;
+
+       device_lock(dev);
+       nvdimm_bus_lock(dev);
+       wait_nvdimm_bus_probe_idle(dev);
+       rc = __alt_name_store(dev, buf, len);
+       if (rc >= 0)
+               rc = nd_namespace_label_update(nd_region, dev);
+       dev_dbg(dev, "%s: %s(%zd)\n", __func__, rc < 0 ? "fail " : "", rc);
+       nvdimm_bus_unlock(dev);
+       device_unlock(dev);
+
+       return rc < 0 ? rc : len;
+}
+
+static ssize_t alt_name_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       char *ns_altname;
+
+       if (is_namespace_pmem(dev)) {
+               struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
+
+               ns_altname = nspm->alt_name;
+       } else if (is_namespace_blk(dev)) {
+               struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
+
+               ns_altname = nsblk->alt_name;
+       } else
+               return -ENXIO;
+
+       return sprintf(buf, "%s\n", ns_altname ? ns_altname : "");
+}
+static DEVICE_ATTR_RW(alt_name);
+
+static int scan_free(struct nd_region *nd_region,
+               struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
+               resource_size_t n)
+{
+       bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
+       struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
+       int rc = 0;
+
+       while (n) {
+               struct resource *res, *last;
+               resource_size_t new_start;
+
+               last = NULL;
+               for_each_dpa_resource(ndd, res)
+                       if (strcmp(res->name, label_id->id) == 0)
+                               last = res;
+               res = last;
+               if (!res)
+                       return 0;
+
+               if (n >= resource_size(res)) {
+                       n -= resource_size(res);
+                       nd_dbg_dpa(nd_region, ndd, res, "delete %d\n", rc);
+                       nvdimm_free_dpa(ndd, res);
+                       /* retry with last resource deleted */
+                       continue;
+               }
+
+               /*
+                * Keep BLK allocations relegated to high DPA as much as
+                * possible
+                */
+               if (is_blk)
+                       new_start = res->start + n;
+               else
+                       new_start = res->start;
+
+               rc = adjust_resource(res, new_start, resource_size(res) - n);
+               if (rc == 0)
+                       res->flags |= DPA_RESOURCE_ADJUSTED;
+               nd_dbg_dpa(nd_region, ndd, res, "shrink %d\n", rc);
+               break;
+       }
+
+       return rc;
+}
+
+/**
+ * shrink_dpa_allocation - for each dimm in region free n bytes for label_id
+ * @nd_region: the set of dimms to reclaim @n bytes from
+ * @label_id: unique identifier for the namespace consuming this dpa range
+ * @n: number of bytes per-dimm to release
+ *
+ * Assumes resources are ordered.  Starting from the end try to
+ * adjust_resource() the allocation to @n, but if @n is larger than the
+ * allocation delete it and find the 'new' last allocation in the label
+ * set.
+ */
+static int shrink_dpa_allocation(struct nd_region *nd_region,
+               struct nd_label_id *label_id, resource_size_t n)
+{
+       int i;
+
+       for (i = 0; i < nd_region->ndr_mappings; i++) {
+               struct nd_mapping *nd_mapping = &nd_region->mapping[i];
+               int rc;
+
+               rc = scan_free(nd_region, nd_mapping, label_id, n);
+               if (rc)
+                       return rc;
+       }
+
+       return 0;
+}
+
+static resource_size_t init_dpa_allocation(struct nd_label_id *label_id,
+               struct nd_region *nd_region, struct nd_mapping *nd_mapping,
+               resource_size_t n)
+{
+       bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
+       struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
+       resource_size_t first_dpa;
+       struct resource *res;
+       int rc = 0;
+
+       /* allocate blk from highest dpa first */
+       if (is_blk)
+               first_dpa = nd_mapping->start + nd_mapping->size - n;
+       else
+               first_dpa = nd_mapping->start;
+
+       /* first resource allocation for this label-id or dimm */
+       res = nvdimm_allocate_dpa(ndd, label_id, first_dpa, n);
+       if (!res)
+               rc = -EBUSY;
+
+       nd_dbg_dpa(nd_region, ndd, res, "init %d\n", rc);
+       return rc ? n : 0;
+}
+
+static bool space_valid(bool is_pmem, bool is_reserve,
+               struct nd_label_id *label_id, struct resource *res)
+{
+       /*
+        * For BLK-space any space is valid, for PMEM-space, it must be
+        * contiguous with an existing allocation unless we are
+        * reserving pmem.
+        */
+       if (is_reserve || !is_pmem)
+               return true;
+       if (!res || strcmp(res->name, label_id->id) == 0)
+               return true;
+       return false;
+}
+
+enum alloc_loc {
+       ALLOC_ERR = 0, ALLOC_BEFORE, ALLOC_MID, ALLOC_AFTER,
+};
+
+static resource_size_t scan_allocate(struct nd_region *nd_region,
+               struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
+               resource_size_t n)
+{
+       resource_size_t mapping_end = nd_mapping->start + nd_mapping->size - 1;
+       bool is_reserve = strcmp(label_id->id, "pmem-reserve") == 0;
+       bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
+       struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
+       const resource_size_t to_allocate = n;
+       struct resource *res;
+       int first;
+
+ retry:
+       first = 0;
+       for_each_dpa_resource(ndd, res) {
+               resource_size_t allocate, available = 0, free_start, free_end;
+               struct resource *next = res->sibling, *new_res = NULL;
+               enum alloc_loc loc = ALLOC_ERR;
+               const char *action;
+               int rc = 0;
+
+               /* ignore resources outside this nd_mapping */
+               if (res->start > mapping_end)
+                       continue;
+               if (res->end < nd_mapping->start)
+                       continue;
+
+               /* space at the beginning of the mapping */
+               if (!first++ && res->start > nd_mapping->start) {
+                       free_start = nd_mapping->start;
+                       available = res->start - free_start;
+                       if (space_valid(is_pmem, is_reserve, label_id, NULL))
+                               loc = ALLOC_BEFORE;
+               }
+
+               /* space between allocations */
+               if (!loc && next) {
+                       free_start = res->start + resource_size(res);
+                       free_end = min(mapping_end, next->start - 1);
+                       if (space_valid(is_pmem, is_reserve, label_id, res)
+                                       && free_start < free_end) {
+                               available = free_end + 1 - free_start;
+                               loc = ALLOC_MID;
+                       }
+               }
+
+               /* space at the end of the mapping */
+               if (!loc && !next) {
+                       free_start = res->start + resource_size(res);
+                       free_end = mapping_end;
+                       if (space_valid(is_pmem, is_reserve, label_id, res)
+                                       && free_start < free_end) {
+                               available = free_end + 1 - free_start;
+                               loc = ALLOC_AFTER;
+                       }
+               }
+
+               if (!loc || !available)
+                       continue;
+               allocate = min(available, n);
+               switch (loc) {
+               case ALLOC_BEFORE:
+                       if (strcmp(res->name, label_id->id) == 0) {
+                               /* adjust current resource up */
+                               if (is_pmem && !is_reserve)
+                                       return n;
+                               rc = adjust_resource(res, res->start - allocate,
+                                               resource_size(res) + allocate);
+                               action = "cur grow up";
+                       } else
+                               action = "allocate";
+                       break;
+               case ALLOC_MID:
+                       if (strcmp(next->name, label_id->id) == 0) {
+                               /* adjust next resource up */
+                               if (is_pmem && !is_reserve)
+                                       return n;
+                               rc = adjust_resource(next, next->start
+                                               - allocate, resource_size(next)
+                                               + allocate);
+                               new_res = next;
+                               action = "next grow up";
+                       } else if (strcmp(res->name, label_id->id) == 0) {
+                               action = "grow down";
+                       } else
+                               action = "allocate";
+                       break;
+               case ALLOC_AFTER:
+                       if (strcmp(res->name, label_id->id) == 0)
+                               action = "grow down";
+                       else
+                               action = "allocate";
+                       break;
+               default:
+                       return n;
+               }
+
+               if (strcmp(action, "allocate") == 0) {
+                       /* BLK allocate bottom up */
+                       if (!is_pmem)
+                               free_start += available - allocate;
+                       else if (!is_reserve && free_start != nd_mapping->start)
+                               return n;
+
+                       new_res = nvdimm_allocate_dpa(ndd, label_id,
+                                       free_start, allocate);
+                       if (!new_res)
+                               rc = -EBUSY;
+               } else if (strcmp(action, "grow down") == 0) {
+                       /* adjust current resource down */
+                       rc = adjust_resource(res, res->start, resource_size(res)
+                                       + allocate);
+                       if (rc == 0)
+                               res->flags |= DPA_RESOURCE_ADJUSTED;
+               }
+
+               if (!new_res)
+                       new_res = res;
+
+               nd_dbg_dpa(nd_region, ndd, new_res, "%s(%d) %d\n",
+                               action, loc, rc);
+
+               if (rc)
+                       return n;
+
+               n -= allocate;
+               if (n) {
+                       /*
+                        * Retry scan with newly inserted resources.
+                        * For example, if we did an ALLOC_BEFORE
+                        * insertion there may also have been space
+                        * available for an ALLOC_AFTER insertion, so we
+                        * need to check this same resource again
+                        */
+                       goto retry;
+               } else
+                       return 0;
+       }
+
+       /*
+        * If we allocated nothing in the BLK case it may be because we are in
+        * an initial "pmem-reserve pass".  Only do an initial BLK allocation
+        * when none of the DPA space is reserved.
+        */
+       if ((is_pmem || !ndd->dpa.child) && n == to_allocate)
+               return init_dpa_allocation(label_id, nd_region, nd_mapping, n);
+       return n;
+}
+
+static int merge_dpa(struct nd_region *nd_region,
+               struct nd_mapping *nd_mapping, struct nd_label_id *label_id)
+{
+       struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
+       struct resource *res;
+
+       if (strncmp("pmem", label_id->id, 4) == 0)
+               return 0;
+ retry:
+       for_each_dpa_resource(ndd, res) {
+               int rc;
+               struct resource *next = res->sibling;
+               resource_size_t end = res->start + resource_size(res);
+
+               if (!next || strcmp(res->name, label_id->id) != 0
+                               || strcmp(next->name, label_id->id) != 0
+                               || end != next->start)
+                       continue;
+               end += resource_size(next);
+               nvdimm_free_dpa(ndd, next);
+               rc = adjust_resource(res, res->start, end - res->start);
+               nd_dbg_dpa(nd_region, ndd, res, "merge %d\n", rc);
+               if (rc)
+                       return rc;
+               res->flags |= DPA_RESOURCE_ADJUSTED;
+               goto retry;
+       }
+
+       return 0;
+}
+
+static int __reserve_free_pmem(struct device *dev, void *data)
+{
+       struct nvdimm *nvdimm = data;
+       struct nd_region *nd_region;
+       struct nd_label_id label_id;
+       int i;
+
+       if (!is_nd_pmem(dev))
+               return 0;
+
+       nd_region = to_nd_region(dev);
+       if (nd_region->ndr_mappings == 0)
+               return 0;
+
+       memset(&label_id, 0, sizeof(label_id));
+       strcat(label_id.id, "pmem-reserve");
+       for (i = 0; i < nd_region->ndr_mappings; i++) {
+               struct nd_mapping *nd_mapping = &nd_region->mapping[i];
+               resource_size_t n, rem = 0;
+
+               if (nd_mapping->nvdimm != nvdimm)
+                       continue;
+
+               n = nd_pmem_available_dpa(nd_region, nd_mapping, &rem);
+               if (n == 0)
+                       return 0;
+               rem = scan_allocate(nd_region, nd_mapping, &label_id, n);
+               dev_WARN_ONCE(&nd_region->dev, rem,
+                               "pmem reserve underrun: %#llx of %#llx bytes\n",
+                               (unsigned long long) n - rem,
+                               (unsigned long long) n);
+               return rem ? -ENXIO : 0;
+       }
+
+       return 0;
+}
+
+static void release_free_pmem(struct nvdimm_bus *nvdimm_bus,
+               struct nd_mapping *nd_mapping)
+{
+       struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
+       struct resource *res, *_res;
+
+       for_each_dpa_resource_safe(ndd, res, _res)
+               if (strcmp(res->name, "pmem-reserve") == 0)
+                       nvdimm_free_dpa(ndd, res);
+}
+
+static int reserve_free_pmem(struct nvdimm_bus *nvdimm_bus,
+               struct nd_mapping *nd_mapping)
+{
+       struct nvdimm *nvdimm = nd_mapping->nvdimm;
+       int rc;
+
+       rc = device_for_each_child(&nvdimm_bus->dev, nvdimm,
+                       __reserve_free_pmem);
+       if (rc)
+               release_free_pmem(nvdimm_bus, nd_mapping);
+       return rc;
+}
+
+/**
+ * grow_dpa_allocation - for each dimm allocate n bytes for @label_id
+ * @nd_region: the set of dimms to allocate @n more bytes from
+ * @label_id: unique identifier for the namespace consuming this dpa range
+ * @n: number of bytes per-dimm to add to the existing allocation
+ *
+ * Assumes resources are ordered.  For BLK regions, first consume
+ * BLK-only available DPA free space, then consume PMEM-aliased DPA
+ * space starting at the highest DPA.  For PMEM regions start
+ * allocations from the start of an interleave set and end at the first
+ * BLK allocation or the end of the interleave set, whichever comes
+ * first.
+ */
+static int grow_dpa_allocation(struct nd_region *nd_region,
+               struct nd_label_id *label_id, resource_size_t n)
+{
+       struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
+       bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
+       int i;
+
+       for (i = 0; i < nd_region->ndr_mappings; i++) {
+               struct nd_mapping *nd_mapping = &nd_region->mapping[i];
+               resource_size_t rem = n;
+               int rc, j;
+
+               /*
+                * In the BLK case try once with all unallocated PMEM
+                * reserved, and once without
+                */
+               for (j = is_pmem; j < 2; j++) {
+                       bool blk_only = j == 0;
+
+                       if (blk_only) {
+                               rc = reserve_free_pmem(nvdimm_bus, nd_mapping);
+                               if (rc)
+                                       return rc;
+                       }
+                       rem = scan_allocate(nd_region, nd_mapping,
+                                       label_id, rem);
+                       if (blk_only)
+                               release_free_pmem(nvdimm_bus, nd_mapping);
+
+                       /* try again and allow encroachments into PMEM */
+                       if (rem == 0)
+                               break;
+               }
+
+               dev_WARN_ONCE(&nd_region->dev, rem,
+                               "allocation underrun: %#llx of %#llx bytes\n",
+                               (unsigned long long) n - rem,
+                               (unsigned long long) n);
+               if (rem)
+                       return -ENXIO;
+
+               rc = merge_dpa(nd_region, nd_mapping, label_id);
+               if (rc)
+                       return rc;
+       }
+
+       return 0;
+}
+
+static void nd_namespace_pmem_set_size(struct nd_region *nd_region,
+               struct nd_namespace_pmem *nspm, resource_size_t size)
+{
+       struct resource *res = &nspm->nsio.res;
+
+       res->start = nd_region->ndr_start;
+       res->end = nd_region->ndr_start + size - 1;
+}
+
+static ssize_t __size_store(struct device *dev, unsigned long long val)
+{
+       resource_size_t allocated = 0, available = 0;
+       struct nd_region *nd_region = to_nd_region(dev->parent);
+       struct nd_mapping *nd_mapping;
+       struct nvdimm_drvdata *ndd;
+       struct nd_label_id label_id;
+       u32 flags = 0, remainder;
+       u8 *uuid = NULL;
+       int rc, i;
+
+       if (dev->driver || to_ndns(dev)->claim)
+               return -EBUSY;
+
+       if (is_namespace_pmem(dev)) {
+               struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
+
+               uuid = nspm->uuid;
+       } else if (is_namespace_blk(dev)) {
+               struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
+
+               uuid = nsblk->uuid;
+               flags = NSLABEL_FLAG_LOCAL;
+       }
+
+       /*
+        * We need a uuid for the allocation-label and dimm(s) on which
+        * to store the label.
+        */
+       if (!uuid || nd_region->ndr_mappings == 0)
+               return -ENXIO;
+
+       div_u64_rem(val, SZ_4K * nd_region->ndr_mappings, &remainder);
+       if (remainder) {
+               dev_dbg(dev, "%llu is not %dK aligned\n", val,
+                               (SZ_4K * nd_region->ndr_mappings) / SZ_1K);
+               return -EINVAL;
+       }
+
+       nd_label_gen_id(&label_id, uuid, flags);
+       for (i = 0; i < nd_region->ndr_mappings; i++) {
+               nd_mapping = &nd_region->mapping[i];
+               ndd = to_ndd(nd_mapping);
+
+               /*
+                * All dimms in an interleave set, or the base dimm for a blk
+                * region, need to be enabled for the size to be changed.
+                */
+               if (!ndd)
+                       return -ENXIO;
+
+               allocated += nvdimm_allocated_dpa(ndd, &label_id);
+       }
+       available = nd_region_available_dpa(nd_region);
+
+       if (val > available + allocated)
+               return -ENOSPC;
+
+       if (val == allocated)
+               return 0;
+
+       val = div_u64(val, nd_region->ndr_mappings);
+       allocated = div_u64(allocated, nd_region->ndr_mappings);
+       if (val < allocated)
+               rc = shrink_dpa_allocation(nd_region, &label_id,
+                               allocated - val);
+       else
+               rc = grow_dpa_allocation(nd_region, &label_id, val - allocated);
+
+       if (rc)
+               return rc;
+
+       if (is_namespace_pmem(dev)) {
+               struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
+
+               nd_namespace_pmem_set_size(nd_region, nspm,
+                               val * nd_region->ndr_mappings);
+       } else if (is_namespace_blk(dev)) {
+               struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
+
+               /*
+                * Try to delete the namespace if we deleted all of its
+                * allocation, this is not the seed device for the
+                * region, and it is not actively claimed by a btt
+                * instance.
+                */
+               if (val == 0 && nd_region->ns_seed != dev
+                               && !nsblk->common.claim)
+                       nd_device_unregister(dev, ND_ASYNC);
+       }
+
+       return rc;
+}
+
+static ssize_t size_store(struct device *dev,
+               struct device_attribute *attr, const char *buf, size_t len)
+{
+       struct nd_region *nd_region = to_nd_region(dev->parent);
+       unsigned long long val;
+       u8 **uuid = NULL;
+       int rc;
+
+       rc = kstrtoull(buf, 0, &val);
+       if (rc)
+               return rc;
+
+       device_lock(dev);
+       nvdimm_bus_lock(dev);
+       wait_nvdimm_bus_probe_idle(dev);
+       rc = __size_store(dev, val);
+       if (rc >= 0)
+               rc = nd_namespace_label_update(nd_region, dev);
+
+       if (is_namespace_pmem(dev)) {
+               struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
+
+               uuid = &nspm->uuid;
+       } else if (is_namespace_blk(dev)) {
+               struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
+
+               uuid = &nsblk->uuid;
+       }
+
+       if (rc == 0 && val == 0 && uuid) {
+               /* setting size zero == 'delete namespace' */
+               kfree(*uuid);
+               *uuid = NULL;
+       }
+
+       dev_dbg(dev, "%s: %llx %s (%d)\n", __func__, val, rc < 0
+                       ? "fail" : "success", rc);
+
+       nvdimm_bus_unlock(dev);
+       device_unlock(dev);
+
+       return rc < 0 ? rc : len;
+}
+
+resource_size_t __nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
+{
+       struct device *dev = &ndns->dev;
+
+       if (is_namespace_pmem(dev)) {
+               struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
+
+               return resource_size(&nspm->nsio.res);
+       } else if (is_namespace_blk(dev)) {
+               return nd_namespace_blk_size(to_nd_namespace_blk(dev));
+       } else if (is_namespace_io(dev)) {
+               struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
+
+               return resource_size(&nsio->res);
+       } else
+               WARN_ONCE(1, "unknown namespace type\n");
+       return 0;
+}
+
+resource_size_t nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
+{
+       resource_size_t size;
+
+       nvdimm_bus_lock(&ndns->dev);
+       size = __nvdimm_namespace_capacity(ndns);
+       nvdimm_bus_unlock(&ndns->dev);
+
+       return size;
+}
+EXPORT_SYMBOL(nvdimm_namespace_capacity);
+
+static ssize_t size_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       return sprintf(buf, "%llu\n", (unsigned long long)
+                       nvdimm_namespace_capacity(to_ndns(dev)));
+}
+static DEVICE_ATTR(size, S_IRUGO, size_show, size_store);
+
+static ssize_t uuid_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       u8 *uuid;
+
+       if (is_namespace_pmem(dev)) {
+               struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
+
+               uuid = nspm->uuid;
+       } else if (is_namespace_blk(dev)) {
+               struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
+
+               uuid = nsblk->uuid;
+       } else
+               return -ENXIO;
+
+       if (uuid)
+               return sprintf(buf, "%pUb\n", uuid);
+       return sprintf(buf, "\n");
+}
+
+/**
+ * namespace_update_uuid - check for a unique uuid and whether we're "renaming"
+ * @nd_region: parent region so we can updates all dimms in the set
+ * @dev: namespace type for generating label_id
+ * @new_uuid: incoming uuid
+ * @old_uuid: reference to the uuid storage location in the namespace object
+ */
+static int namespace_update_uuid(struct nd_region *nd_region,
+               struct device *dev, u8 *new_uuid, u8 **old_uuid)
+{
+       u32 flags = is_namespace_blk(dev) ? NSLABEL_FLAG_LOCAL : 0;
+       struct nd_label_id old_label_id;
+       struct nd_label_id new_label_id;
+       int i;
+
+       if (!nd_is_uuid_unique(dev, new_uuid))
+               return -EINVAL;
+
+       if (*old_uuid == NULL)
+               goto out;
+
+       /*
+        * If we've already written a label with this uuid, then it's
+        * too late to rename because we can't reliably update the uuid
+        * without losing the old namespace.  Userspace must delete this
+        * namespace to abandon the old uuid.
+        */
+       for (i = 0; i < nd_region->ndr_mappings; i++) {
+               struct nd_mapping *nd_mapping = &nd_region->mapping[i];
+
+               /*
+                * This check by itself is sufficient because old_uuid
+                * would be NULL above if this uuid did not exist in the
+                * currently written set.
+                *
+                * FIXME: can we delete uuid with zero dpa allocated?
+                */
+               if (nd_mapping->labels)
+                       return -EBUSY;
+       }
+
+       nd_label_gen_id(&old_label_id, *old_uuid, flags);
+       nd_label_gen_id(&new_label_id, new_uuid, flags);
+       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);
+               struct resource *res;
+
+               for_each_dpa_resource(ndd, res)
+                       if (strcmp(res->name, old_label_id.id) == 0)
+                               sprintf((void *) res->name, "%s",
+                                               new_label_id.id);
+       }
+       kfree(*old_uuid);
+ out:
+       *old_uuid = new_uuid;
+       return 0;
+}
+
+static ssize_t uuid_store(struct device *dev,
+               struct device_attribute *attr, const char *buf, size_t len)
+{
+       struct nd_region *nd_region = to_nd_region(dev->parent);
+       u8 *uuid = NULL;
+       ssize_t rc = 0;
+       u8 **ns_uuid;
+
+       if (is_namespace_pmem(dev)) {
+               struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
+
+               ns_uuid = &nspm->uuid;
+       } else if (is_namespace_blk(dev)) {
+               struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
+
+               ns_uuid = &nsblk->uuid;
+       } else
+               return -ENXIO;
+
+       device_lock(dev);
+       nvdimm_bus_lock(dev);
+       wait_nvdimm_bus_probe_idle(dev);
+       if (to_ndns(dev)->claim)
+               rc = -EBUSY;
+       if (rc >= 0)
+               rc = nd_uuid_store(dev, &uuid, buf, len);
+       if (rc >= 0)
+               rc = namespace_update_uuid(nd_region, dev, uuid, ns_uuid);
+       if (rc >= 0)
+               rc = nd_namespace_label_update(nd_region, dev);
+       else
+               kfree(uuid);
+       dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
+                       rc, buf, buf[len - 1] == '\n' ? "" : "\n");
+       nvdimm_bus_unlock(dev);
+       device_unlock(dev);
+
+       return rc < 0 ? rc : len;
+}
+static DEVICE_ATTR_RW(uuid);
+
+static ssize_t resource_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       struct resource *res;
+
+       if (is_namespace_pmem(dev)) {
+               struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
+
+               res = &nspm->nsio.res;
+       } else if (is_namespace_io(dev)) {
+               struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
+
+               res = &nsio->res;
+       } else
+               return -ENXIO;
+
+       /* no address to convey if the namespace has no allocation */
+       if (resource_size(res) == 0)
+               return -ENXIO;
+       return sprintf(buf, "%#llx\n", (unsigned long long) res->start);
+}
+static DEVICE_ATTR_RO(resource);
+
+static const unsigned long ns_lbasize_supported[] = { 512, 520, 528,
+       4096, 4104, 4160, 4224, 0 };
+
+static ssize_t sector_size_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
+
+       if (!is_namespace_blk(dev))
+               return -ENXIO;
+
+       return nd_sector_size_show(nsblk->lbasize, ns_lbasize_supported, buf);
+}
+
+static ssize_t sector_size_store(struct device *dev,
+               struct device_attribute *attr, const char *buf, size_t len)
+{
+       struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
+       struct nd_region *nd_region = to_nd_region(dev->parent);
+       ssize_t rc = 0;
+
+       if (!is_namespace_blk(dev))
+               return -ENXIO;
+
+       device_lock(dev);
+       nvdimm_bus_lock(dev);
+       if (to_ndns(dev)->claim)
+               rc = -EBUSY;
+       if (rc >= 0)
+               rc = nd_sector_size_store(dev, buf, &nsblk->lbasize,
+                               ns_lbasize_supported);
+       if (rc >= 0)
+               rc = nd_namespace_label_update(nd_region, dev);
+       dev_dbg(dev, "%s: result: %zd %s: %s%s", __func__,
+                       rc, rc < 0 ? "tried" : "wrote", buf,
+                       buf[len - 1] == '\n' ? "" : "\n");
+       nvdimm_bus_unlock(dev);
+       device_unlock(dev);
+
+       return rc ? rc : len;
+}
+static DEVICE_ATTR_RW(sector_size);
+
+static ssize_t dpa_extents_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       struct nd_region *nd_region = to_nd_region(dev->parent);
+       struct nd_label_id label_id;
+       int count = 0, i;
+       u8 *uuid = NULL;
+       u32 flags = 0;
+
+       nvdimm_bus_lock(dev);
+       if (is_namespace_pmem(dev)) {
+               struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
+
+               uuid = nspm->uuid;
+               flags = 0;
+       } else if (is_namespace_blk(dev)) {
+               struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
+
+               uuid = nsblk->uuid;
+               flags = NSLABEL_FLAG_LOCAL;
+       }
+
+       if (!uuid)
+               goto out;
+
+       nd_label_gen_id(&label_id, uuid, flags);
+       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);
+               struct resource *res;
+
+               for_each_dpa_resource(ndd, res)
+                       if (strcmp(res->name, label_id.id) == 0)
+                               count++;
+       }
+ out:
+       nvdimm_bus_unlock(dev);
+
+       return sprintf(buf, "%d\n", count);
+}
+static DEVICE_ATTR_RO(dpa_extents);
+
+static ssize_t holder_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       struct nd_namespace_common *ndns = to_ndns(dev);
+       ssize_t rc;
+
+       device_lock(dev);
+       rc = sprintf(buf, "%s\n", ndns->claim ? dev_name(ndns->claim) : "");
+       device_unlock(dev);
+
+       return rc;
+}
+static DEVICE_ATTR_RO(holder);
+
+static ssize_t force_raw_store(struct device *dev,
+               struct device_attribute *attr, const char *buf, size_t len)
+{
+       bool force_raw;
+       int rc = strtobool(buf, &force_raw);
+
+       if (rc)
+               return rc;
+
+       to_ndns(dev)->force_raw = force_raw;
+       return len;
+}
+
+static ssize_t force_raw_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       return sprintf(buf, "%d\n", to_ndns(dev)->force_raw);
+}
+static DEVICE_ATTR_RW(force_raw);
+
 static struct attribute *nd_namespace_attributes[] = {
        &dev_attr_nstype.attr,
+       &dev_attr_size.attr,
+       &dev_attr_uuid.attr,
+       &dev_attr_holder.attr,
+       &dev_attr_resource.attr,
+       &dev_attr_alt_name.attr,
+       &dev_attr_force_raw.attr,
+       &dev_attr_sector_size.attr,
+       &dev_attr_dpa_extents.attr,
        NULL,
 };
 
+static umode_t namespace_visible(struct kobject *kobj,
+               struct attribute *a, int n)
+{
+       struct device *dev = container_of(kobj, struct device, kobj);
+
+       if (a == &dev_attr_resource.attr) {
+               if (is_namespace_blk(dev))
+                       return 0;
+               return a->mode;
+       }
+
+       if (is_namespace_pmem(dev) || is_namespace_blk(dev)) {
+               if (a == &dev_attr_size.attr)
+                       return S_IWUSR | S_IRUGO;
+
+               if (is_namespace_pmem(dev) && a == &dev_attr_sector_size.attr)
+                       return 0;
+
+               return a->mode;
+       }
+
+       if (a == &dev_attr_nstype.attr || a == &dev_attr_size.attr
+                       || a == &dev_attr_holder.attr
+                       || a == &dev_attr_force_raw.attr)
+               return a->mode;
+
+       return 0;
+}
+
 static struct attribute_group nd_namespace_attribute_group = {
        .attrs = nd_namespace_attributes,
+       .is_visible = namespace_visible,
 };
 
 static const struct attribute_group *nd_namespace_attribute_groups[] = {
        &nd_device_attribute_group,
        &nd_namespace_attribute_group,
+       &nd_numa_attribute_group,
        NULL,
 };
 
+struct nd_namespace_common *nvdimm_namespace_common_probe(struct device *dev)
+{
+       struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
+       struct nd_namespace_common *ndns;
+       resource_size_t size;
+
+       if (nd_btt) {
+               ndns = nd_btt->ndns;
+               if (!ndns)
+                       return ERR_PTR(-ENODEV);
+
+               /*
+                * Flush any in-progess probes / removals in the driver
+                * for the raw personality of this namespace.
+                */
+               device_lock(&ndns->dev);
+               device_unlock(&ndns->dev);
+               if (ndns->dev.driver) {
+                       dev_dbg(&ndns->dev, "is active, can't bind %s\n",
+                                       dev_name(&nd_btt->dev));
+                       return ERR_PTR(-EBUSY);
+               }
+               if (dev_WARN_ONCE(&ndns->dev, ndns->claim != &nd_btt->dev,
+                                       "host (%s) vs claim (%s) mismatch\n",
+                                       dev_name(&nd_btt->dev),
+                                       dev_name(ndns->claim)))
+                       return ERR_PTR(-ENXIO);
+       } else {
+               ndns = to_ndns(dev);
+               if (ndns->claim) {
+                       dev_dbg(dev, "claimed by %s, failing probe\n",
+                               dev_name(ndns->claim));
+
+                       return ERR_PTR(-ENXIO);
+               }
+       }
+
+       size = nvdimm_namespace_capacity(ndns);
+       if (size < ND_MIN_NAMESPACE_SIZE) {
+               dev_dbg(&ndns->dev, "%pa, too small must be at least %#x\n",
+                               &size, ND_MIN_NAMESPACE_SIZE);
+               return ERR_PTR(-ENODEV);
+       }
+
+       if (is_namespace_pmem(&ndns->dev)) {
+               struct nd_namespace_pmem *nspm;
+
+               nspm = to_nd_namespace_pmem(&ndns->dev);
+               if (!nspm->uuid) {
+                       dev_dbg(&ndns->dev, "%s: uuid not set\n", __func__);
+                       return ERR_PTR(-ENODEV);
+               }
+       } else if (is_namespace_blk(&ndns->dev)) {
+               struct nd_namespace_blk *nsblk;
+
+               nsblk = to_nd_namespace_blk(&ndns->dev);
+               if (!nd_namespace_blk_validate(nsblk))
+                       return ERR_PTR(-ENODEV);
+       }
+
+       return ndns;
+}
+EXPORT_SYMBOL(nvdimm_namespace_common_probe);
+
 static struct device **create_namespace_io(struct nd_region *nd_region)
 {
        struct nd_namespace_io *nsio;
@@ -68,7 +1312,7 @@ static struct device **create_namespace_io(struct nd_region *nd_region)
                return NULL;
        }
 
-       dev = &nsio->dev;
+       dev = &nsio->common.dev;
        dev->type = &namespace_io_device_type;
        dev->parent = &nd_region->dev;
        res = &nsio->res;
@@ -81,31 +1325,546 @@ static struct device **create_namespace_io(struct nd_region *nd_region)
        return devs;
 }
 
+static bool has_uuid_at_pos(struct nd_region *nd_region, u8 *uuid,
+               u64 cookie, u16 pos)
+{
+       struct nd_namespace_label *found = NULL;
+       int i;
+
+       for (i = 0; i < nd_region->ndr_mappings; i++) {
+               struct nd_mapping *nd_mapping = &nd_region->mapping[i];
+               struct nd_namespace_label *nd_label;
+               bool found_uuid = false;
+               int l;
+
+               for_each_label(l, nd_label, nd_mapping->labels) {
+                       u64 isetcookie = __le64_to_cpu(nd_label->isetcookie);
+                       u16 position = __le16_to_cpu(nd_label->position);
+                       u16 nlabel = __le16_to_cpu(nd_label->nlabel);
+
+                       if (isetcookie != cookie)
+                               continue;
+
+                       if (memcmp(nd_label->uuid, uuid, NSLABEL_UUID_LEN) != 0)
+                               continue;
+
+                       if (found_uuid) {
+                               dev_dbg(to_ndd(nd_mapping)->dev,
+                                               "%s duplicate entry for uuid\n",
+                                               __func__);
+                               return false;
+                       }
+                       found_uuid = true;
+                       if (nlabel != nd_region->ndr_mappings)
+                               continue;
+                       if (position != pos)
+                               continue;
+                       found = nd_label;
+                       break;
+               }
+               if (found)
+                       break;
+       }
+       return found != NULL;
+}
+
+static int select_pmem_id(struct nd_region *nd_region, u8 *pmem_id)
+{
+       struct nd_namespace_label *select = NULL;
+       int i;
+
+       if (!pmem_id)
+               return -ENODEV;
+
+       for (i = 0; i < nd_region->ndr_mappings; i++) {
+               struct nd_mapping *nd_mapping = &nd_region->mapping[i];
+               struct nd_namespace_label *nd_label;
+               u64 hw_start, hw_end, pmem_start, pmem_end;
+               int l;
+
+               for_each_label(l, nd_label, nd_mapping->labels)
+                       if (memcmp(nd_label->uuid, pmem_id, NSLABEL_UUID_LEN) == 0)
+                               break;
+
+               if (!nd_label) {
+                       WARN_ON(1);
+                       return -EINVAL;
+               }
+
+               select = nd_label;
+               /*
+                * Check that this label is compliant with the dpa
+                * range published in NFIT
+                */
+               hw_start = nd_mapping->start;
+               hw_end = hw_start + nd_mapping->size;
+               pmem_start = __le64_to_cpu(select->dpa);
+               pmem_end = pmem_start + __le64_to_cpu(select->rawsize);
+               if (pmem_start == hw_start && pmem_end <= hw_end)
+                       /* pass */;
+               else
+                       return -EINVAL;
+
+               nd_mapping->labels[0] = select;
+               nd_mapping->labels[1] = NULL;
+       }
+       return 0;
+}
+
+/**
+ * find_pmem_label_set - validate interleave set labelling, retrieve label0
+ * @nd_region: region with mappings to validate
+ */
+static int find_pmem_label_set(struct nd_region *nd_region,
+               struct nd_namespace_pmem *nspm)
+{
+       u64 cookie = nd_region_interleave_set_cookie(nd_region);
+       struct nd_namespace_label *nd_label;
+       u8 select_id[NSLABEL_UUID_LEN];
+       resource_size_t size = 0;
+       u8 *pmem_id = NULL;
+       int rc = -ENODEV, l;
+       u16 i;
+
+       if (cookie == 0)
+               return -ENXIO;
+
+       /*
+        * Find a complete set of labels by uuid.  By definition we can start
+        * with any mapping as the reference label
+        */
+       for_each_label(l, nd_label, nd_region->mapping[0].labels) {
+               u64 isetcookie = __le64_to_cpu(nd_label->isetcookie);
+
+               if (isetcookie != cookie)
+                       continue;
+
+               for (i = 0; nd_region->ndr_mappings; i++)
+                       if (!has_uuid_at_pos(nd_region, nd_label->uuid,
+                                               cookie, i))
+                               break;
+               if (i < nd_region->ndr_mappings) {
+                       /*
+                        * Give up if we don't find an instance of a
+                        * uuid at each position (from 0 to
+                        * nd_region->ndr_mappings - 1), or if we find a
+                        * dimm with two instances of the same uuid.
+                        */
+                       rc = -EINVAL;
+                       goto err;
+               } else if (pmem_id) {
+                       /*
+                        * If there is more than one valid uuid set, we
+                        * need userspace to clean this up.
+                        */
+                       rc = -EBUSY;
+                       goto err;
+               }
+               memcpy(select_id, nd_label->uuid, NSLABEL_UUID_LEN);
+               pmem_id = select_id;
+       }
+
+       /*
+        * Fix up each mapping's 'labels' to have the validated pmem label for
+        * that position at labels[0], and NULL at labels[1].  In the process,
+        * check that the namespace aligns with interleave-set.  We know
+        * that it does not overlap with any blk namespaces by virtue of
+        * the dimm being enabled (i.e. nd_label_reserve_dpa()
+        * succeeded).
+        */
+       rc = select_pmem_id(nd_region, pmem_id);
+       if (rc)
+               goto err;
+
+       /* Calculate total size and populate namespace properties from label0 */
+       for (i = 0; i < nd_region->ndr_mappings; i++) {
+               struct nd_mapping *nd_mapping = &nd_region->mapping[i];
+               struct nd_namespace_label *label0 = nd_mapping->labels[0];
+
+               size += __le64_to_cpu(label0->rawsize);
+               if (__le16_to_cpu(label0->position) != 0)
+                       continue;
+               WARN_ON(nspm->alt_name || nspm->uuid);
+               nspm->alt_name = kmemdup((void __force *) label0->name,
+                               NSLABEL_NAME_LEN, GFP_KERNEL);
+               nspm->uuid = kmemdup((void __force *) label0->uuid,
+                               NSLABEL_UUID_LEN, GFP_KERNEL);
+       }
+
+       if (!nspm->alt_name || !nspm->uuid) {
+               rc = -ENOMEM;
+               goto err;
+       }
+
+       nd_namespace_pmem_set_size(nd_region, nspm, size);
+
+       return 0;
+ err:
+       switch (rc) {
+       case -EINVAL:
+               dev_dbg(&nd_region->dev, "%s: invalid label(s)\n", __func__);
+               break;
+       case -ENODEV:
+               dev_dbg(&nd_region->dev, "%s: label not found\n", __func__);
+               break;
+       default:
+               dev_dbg(&nd_region->dev, "%s: unexpected err: %d\n",
+                               __func__, rc);
+               break;
+       }
+       return rc;
+}
+
+static struct device **create_namespace_pmem(struct nd_region *nd_region)
+{
+       struct nd_namespace_pmem *nspm;
+       struct device *dev, **devs;
+       struct resource *res;
+       int rc;
+
+       nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
+       if (!nspm)
+               return NULL;
+
+       dev = &nspm->nsio.common.dev;
+       dev->type = &namespace_pmem_device_type;
+       dev->parent = &nd_region->dev;
+       res = &nspm->nsio.res;
+       res->name = dev_name(&nd_region->dev);
+       res->flags = IORESOURCE_MEM;
+       rc = find_pmem_label_set(nd_region, nspm);
+       if (rc == -ENODEV) {
+               int i;
+
+               /* Pass, try to permit namespace creation... */
+               for (i = 0; i < nd_region->ndr_mappings; i++) {
+                       struct nd_mapping *nd_mapping = &nd_region->mapping[i];
+
+                       kfree(nd_mapping->labels);
+                       nd_mapping->labels = NULL;
+               }
+
+               /* Publish a zero-sized namespace for userspace to configure. */
+               nd_namespace_pmem_set_size(nd_region, nspm, 0);
+
+               rc = 0;
+       } else if (rc)
+               goto err;
+
+       devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL);
+       if (!devs)
+               goto err;
+
+       devs[0] = dev;
+       return devs;
+
+ err:
+       namespace_pmem_release(&nspm->nsio.common.dev);
+       return NULL;
+}
+
+struct resource *nsblk_add_resource(struct nd_region *nd_region,
+               struct nvdimm_drvdata *ndd, struct nd_namespace_blk *nsblk,
+               resource_size_t start)
+{
+       struct nd_label_id label_id;
+       struct resource *res;
+
+       nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
+       res = krealloc(nsblk->res,
+                       sizeof(void *) * (nsblk->num_resources + 1),
+                       GFP_KERNEL);
+       if (!res)
+               return NULL;
+       nsblk->res = (struct resource **) res;
+       for_each_dpa_resource(ndd, res)
+               if (strcmp(res->name, label_id.id) == 0
+                               && res->start == start) {
+                       nsblk->res[nsblk->num_resources++] = res;
+                       return res;
+               }
+       return NULL;
+}
+
+static struct device *nd_namespace_blk_create(struct nd_region *nd_region)
+{
+       struct nd_namespace_blk *nsblk;
+       struct device *dev;
+
+       if (!is_nd_blk(&nd_region->dev))
+               return NULL;
+
+       nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
+       if (!nsblk)
+               return NULL;
+
+       dev = &nsblk->common.dev;
+       dev->type = &namespace_blk_device_type;
+       nsblk->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
+       if (nsblk->id < 0) {
+               kfree(nsblk);
+               return NULL;
+       }
+       dev_set_name(dev, "namespace%d.%d", nd_region->id, nsblk->id);
+       dev->parent = &nd_region->dev;
+       dev->groups = nd_namespace_attribute_groups;
+
+       return &nsblk->common.dev;
+}
+
+void nd_region_create_blk_seed(struct nd_region *nd_region)
+{
+       WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
+       nd_region->ns_seed = nd_namespace_blk_create(nd_region);
+       /*
+        * Seed creation failures are not fatal, provisioning is simply
+        * disabled until memory becomes available
+        */
+       if (!nd_region->ns_seed)
+               dev_err(&nd_region->dev, "failed to create blk namespace\n");
+       else
+               nd_device_register(nd_region->ns_seed);
+}
+
+void nd_region_create_btt_seed(struct nd_region *nd_region)
+{
+       WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
+       nd_region->btt_seed = nd_btt_create(nd_region);
+       /*
+        * Seed creation failures are not fatal, provisioning is simply
+        * disabled until memory becomes available
+        */
+       if (!nd_region->btt_seed)
+               dev_err(&nd_region->dev, "failed to create btt namespace\n");
+}
+
+static struct device **create_namespace_blk(struct nd_region *nd_region)
+{
+       struct nd_mapping *nd_mapping = &nd_region->mapping[0];
+       struct nd_namespace_label *nd_label;
+       struct device *dev, **devs = NULL;
+       struct nd_namespace_blk *nsblk;
+       struct nvdimm_drvdata *ndd;
+       int i, l, count = 0;
+       struct resource *res;
+
+       if (nd_region->ndr_mappings == 0)
+               return NULL;
+
+       ndd = to_ndd(nd_mapping);
+       for_each_label(l, nd_label, nd_mapping->labels) {
+               u32 flags = __le32_to_cpu(nd_label->flags);
+               char *name[NSLABEL_NAME_LEN];
+               struct device **__devs;
+
+               if (flags & NSLABEL_FLAG_LOCAL)
+                       /* pass */;
+               else
+                       continue;
+
+               for (i = 0; i < count; i++) {
+                       nsblk = to_nd_namespace_blk(devs[i]);
+                       if (memcmp(nsblk->uuid, nd_label->uuid,
+                                               NSLABEL_UUID_LEN) == 0) {
+                               res = nsblk_add_resource(nd_region, ndd, nsblk,
+                                               __le64_to_cpu(nd_label->dpa));
+                               if (!res)
+                                       goto err;
+                               nd_dbg_dpa(nd_region, ndd, res, "%s assign\n",
+                                       dev_name(&nsblk->common.dev));
+                               break;
+                       }
+               }
+               if (i < count)
+                       continue;
+               __devs = kcalloc(count + 2, sizeof(dev), GFP_KERNEL);
+               if (!__devs)
+                       goto err;
+               memcpy(__devs, devs, sizeof(dev) * count);
+               kfree(devs);
+               devs = __devs;
+
+               nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
+               if (!nsblk)
+                       goto err;
+               dev = &nsblk->common.dev;
+               dev->type = &namespace_blk_device_type;
+               dev->parent = &nd_region->dev;
+               dev_set_name(dev, "namespace%d.%d", nd_region->id, count);
+               devs[count++] = dev;
+               nsblk->id = -1;
+               nsblk->lbasize = __le64_to_cpu(nd_label->lbasize);
+               nsblk->uuid = kmemdup(nd_label->uuid, NSLABEL_UUID_LEN,
+                               GFP_KERNEL);
+               if (!nsblk->uuid)
+                       goto err;
+               memcpy(name, nd_label->name, NSLABEL_NAME_LEN);
+               if (name[0])
+                       nsblk->alt_name = kmemdup(name, NSLABEL_NAME_LEN,
+                                       GFP_KERNEL);
+               res = nsblk_add_resource(nd_region, ndd, nsblk,
+                               __le64_to_cpu(nd_label->dpa));
+               if (!res)
+                       goto err;
+               nd_dbg_dpa(nd_region, ndd, res, "%s assign\n",
+                               dev_name(&nsblk->common.dev));
+       }
+
+       dev_dbg(&nd_region->dev, "%s: discovered %d blk namespace%s\n",
+                       __func__, count, count == 1 ? "" : "s");
+
+       if (count == 0) {
+               /* Publish a zero-sized namespace for userspace to configure. */
+               for (i = 0; i < nd_region->ndr_mappings; i++) {
+                       struct nd_mapping *nd_mapping = &nd_region->mapping[i];
+
+                       kfree(nd_mapping->labels);
+                       nd_mapping->labels = NULL;
+               }
+
+               devs = kcalloc(2, sizeof(dev), GFP_KERNEL);
+               if (!devs)
+                       goto err;
+               nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
+               if (!nsblk)
+                       goto err;
+               dev = &nsblk->common.dev;
+               dev->type = &namespace_blk_device_type;
+               dev->parent = &nd_region->dev;
+               devs[count++] = dev;
+       }
+
+       return devs;
+
+err:
+       for (i = 0; i < count; i++) {
+               nsblk = to_nd_namespace_blk(devs[i]);
+               namespace_blk_release(&nsblk->common.dev);
+       }
+       kfree(devs);
+       return NULL;
+}
+
+static int init_active_labels(struct nd_region *nd_region)
+{
+       int i;
+
+       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);
+               struct nvdimm *nvdimm = nd_mapping->nvdimm;
+               int count, j;
+
+               /*
+                * If the dimm is disabled then prevent the region from
+                * being activated if it aliases DPA.
+                */
+               if (!ndd) {
+                       if ((nvdimm->flags & NDD_ALIASING) == 0)
+                               return 0;
+                       dev_dbg(&nd_region->dev, "%s: is disabled, failing probe\n",
+                                       dev_name(&nd_mapping->nvdimm->dev));
+                       return -ENXIO;
+               }
+               nd_mapping->ndd = ndd;
+               atomic_inc(&nvdimm->busy);
+               get_ndd(ndd);
+
+               count = nd_label_active_count(ndd);
+               dev_dbg(ndd->dev, "%s: %d\n", __func__, count);
+               if (!count)
+                       continue;
+               nd_mapping->labels = kcalloc(count + 1, sizeof(void *),
+                               GFP_KERNEL);
+               if (!nd_mapping->labels)
+                       return -ENOMEM;
+               for (j = 0; j < count; j++) {
+                       struct nd_namespace_label *label;
+
+                       label = nd_label_active(ndd, j);
+                       nd_mapping->labels[j] = label;
+               }
+       }
+
+       return 0;
+}
+
 int nd_region_register_namespaces(struct nd_region *nd_region, int *err)
 {
        struct device **devs = NULL;
-       int i;
+       int i, rc = 0, type;
 
        *err = 0;
-       switch (nd_region_to_nstype(nd_region)) {
+       nvdimm_bus_lock(&nd_region->dev);
+       rc = init_active_labels(nd_region);
+       if (rc) {
+               nvdimm_bus_unlock(&nd_region->dev);
+               return rc;
+       }
+
+       type = nd_region_to_nstype(nd_region);
+       switch (type) {
        case ND_DEVICE_NAMESPACE_IO:
                devs = create_namespace_io(nd_region);
                break;
+       case ND_DEVICE_NAMESPACE_PMEM:
+               devs = create_namespace_pmem(nd_region);
+               break;
+       case ND_DEVICE_NAMESPACE_BLK:
+               devs = create_namespace_blk(nd_region);
+               break;
        default:
                break;
        }
+       nvdimm_bus_unlock(&nd_region->dev);
 
        if (!devs)
                return -ENODEV;
 
        for (i = 0; devs[i]; i++) {
                struct device *dev = devs[i];
+               int id;
 
-               dev_set_name(dev, "namespace%d.%d", nd_region->id, i);
+               if (type == ND_DEVICE_NAMESPACE_BLK) {
+                       struct nd_namespace_blk *nsblk;
+
+                       nsblk = to_nd_namespace_blk(dev);
+                       id = ida_simple_get(&nd_region->ns_ida, 0, 0,
+                                       GFP_KERNEL);
+                       nsblk->id = id;
+               } else
+                       id = i;
+
+               if (id < 0)
+                       break;
+               dev_set_name(dev, "namespace%d.%d", nd_region->id, id);
                dev->groups = nd_namespace_attribute_groups;
                nd_device_register(dev);
        }
+       if (i)
+               nd_region->ns_seed = devs[0];
+
+       if (devs[i]) {
+               int j;
+
+               for (j = i; devs[j]; j++) {
+                       struct device *dev = devs[j];
+
+                       device_initialize(dev);
+                       put_device(dev);
+               }
+               *err = j - i;
+               /*
+                * All of the namespaces we tried to register failed, so
+                * fail region activation.
+                */
+               if (*err == 0)
+                       rc = -ENODEV;
+       }
        kfree(devs);
 
+       if (rc == -ENODEV)
+               return rc;
+
        return i;
 }