libnvdimm: Add a poison list and export badblocks
authorVishal Verma <vishal.l.verma@intel.com>
Fri, 25 Dec 2015 02:21:43 +0000 (19:21 -0700)
committerDan Williams <dan.j.williams@intel.com>
Sat, 9 Jan 2016 16:39:03 +0000 (08:39 -0800)
During region creation, perform Address Range Scrubs (ARS) for the SPA
(System Physical Address) ranges to retrieve known poison locations from
firmware. Add a new data structure 'nd_poison' which is used as a list
in nvdimm_bus to store these poison locations.

When creating a pmem namespace, if there is any known poison associated
with its physical address space, convert the poison ranges to bad sectors
that are exposed using the badblocks interface.

Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
drivers/acpi/nfit.c
drivers/nvdimm/core.c
drivers/nvdimm/nd-core.h
drivers/nvdimm/nd.h
drivers/nvdimm/pmem.c
include/linux/libnvdimm.h

index e7ed39bab97d5d0a97d03919bd615eb5fba4105e..e1dbc8da09b7ca477ef3c02b9a0486a66b2fcc28 100644 (file)
@@ -15,6 +15,7 @@
 #include <linux/module.h>
 #include <linux/mutex.h>
 #include <linux/ndctl.h>
+#include <linux/delay.h>
 #include <linux/list.h>
 #include <linux/acpi.h>
 #include <linux/sort.h>
@@ -1473,6 +1474,201 @@ static void acpi_nfit_blk_region_disable(struct nvdimm_bus *nvdimm_bus,
        /* devm will free nfit_blk */
 }
 
+static int ars_get_cap(struct nvdimm_bus_descriptor *nd_desc,
+               struct nd_cmd_ars_cap *cmd, u64 addr, u64 length)
+{
+       cmd->address = addr;
+       cmd->length = length;
+
+       return nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
+                       sizeof(*cmd));
+}
+
+static int ars_do_start(struct nvdimm_bus_descriptor *nd_desc,
+               struct nd_cmd_ars_start *cmd, u64 addr, u64 length)
+{
+       int rc;
+
+       cmd->address = addr;
+       cmd->length = length;
+       cmd->type = ND_ARS_PERSISTENT;
+
+       while (1) {
+               rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, cmd,
+                               sizeof(*cmd));
+               if (rc)
+                       return rc;
+               switch (cmd->status) {
+               case 0:
+                       return 0;
+               case 1:
+                       /* ARS unsupported, but we should never get here */
+                       return 0;
+               case 2:
+                       return -EINVAL;
+               case 3:
+                       /* ARS is in progress */
+                       msleep(1000);
+                       break;
+               default:
+                       return -ENXIO;
+               }
+       }
+}
+
+static int ars_get_status(struct nvdimm_bus_descriptor *nd_desc,
+               struct nd_cmd_ars_status *cmd)
+{
+       int rc;
+
+       while (1) {
+               rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, cmd,
+                       sizeof(*cmd));
+               if (rc || cmd->status & 0xffff)
+                       return -ENXIO;
+
+               /* Check extended status (Upper two bytes) */
+               switch (cmd->status >> 16) {
+               case 0:
+                       return 0;
+               case 1:
+                       /* ARS is in progress */
+                       msleep(1000);
+                       break;
+               case 2:
+                       /* No ARS performed for the current boot */
+                       return 0;
+               default:
+                       return -ENXIO;
+               }
+       }
+}
+
+static int ars_status_process_records(struct nvdimm_bus *nvdimm_bus,
+               struct nd_cmd_ars_status *ars_status, u64 start)
+{
+       int rc;
+       u32 i;
+
+       /*
+        * The address field returned by ars_status should be either
+        * less than or equal to the address we last started ARS for.
+        * The (start, length) returned by ars_status should also have
+        * non-zero overlap with the range we started ARS for.
+        * If this is not the case, bail.
+        */
+       if (ars_status->address > start ||
+                       (ars_status->address + ars_status->length < start))
+               return -ENXIO;
+
+       for (i = 0; i < ars_status->num_records; i++) {
+               rc = nvdimm_bus_add_poison(nvdimm_bus,
+                               ars_status->records[i].err_address,
+                               ars_status->records[i].length);
+               if (rc)
+                       return rc;
+       }
+
+       return 0;
+}
+
+static int acpi_nfit_find_poison(struct acpi_nfit_desc *acpi_desc,
+               struct nd_region_desc *ndr_desc)
+{
+       struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
+       struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
+       struct nd_cmd_ars_status *ars_status = NULL;
+       struct nd_cmd_ars_start *ars_start = NULL;
+       struct nd_cmd_ars_cap *ars_cap = NULL;
+       u64 start, len, cur, remaining;
+       int rc;
+
+       ars_cap = kzalloc(sizeof(*ars_cap), GFP_KERNEL);
+       if (!ars_cap)
+               return -ENOMEM;
+
+       start = ndr_desc->res->start;
+       len = ndr_desc->res->end - ndr_desc->res->start + 1;
+
+       rc = ars_get_cap(nd_desc, ars_cap, start, len);
+       if (rc)
+               goto out;
+
+       /*
+        * If ARS is unsupported, or if the 'Persistent Memory Scrub' flag in
+        * extended status is not set, skip this but continue initialization
+        */
+       if ((ars_cap->status & 0xffff) ||
+               !(ars_cap->status >> 16 & ND_ARS_PERSISTENT)) {
+               dev_warn(acpi_desc->dev,
+                       "ARS unsupported (status: 0x%x), won't create an error list\n",
+                       ars_cap->status);
+               goto out;
+       }
+
+       /*
+        * Check if a full-range ARS has been run. If so, use those results
+        * without having to start a new ARS.
+        */
+       ars_status = kzalloc(ars_cap->max_ars_out + sizeof(*ars_status),
+                       GFP_KERNEL);
+       if (!ars_status) {
+               rc = -ENOMEM;
+               goto out;
+       }
+
+       rc = ars_get_status(nd_desc, ars_status);
+       if (rc)
+               goto out;
+
+       if (ars_status->address <= start &&
+               (ars_status->address + ars_status->length >= start + len)) {
+               rc = ars_status_process_records(nvdimm_bus, ars_status, start);
+               goto out;
+       }
+
+       /*
+        * ARS_STATUS can overflow if the number of poison entries found is
+        * greater than the maximum buffer size (ars_cap->max_ars_out)
+        * To detect overflow, check if the length field of ars_status
+        * is less than the length we supplied. If so, process the
+        * error entries we got, adjust the start point, and start again
+        */
+       ars_start = kzalloc(sizeof(*ars_start), GFP_KERNEL);
+       if (!ars_start)
+               return -ENOMEM;
+
+       cur = start;
+       remaining = len;
+       do {
+               u64 done, end;
+
+               rc = ars_do_start(nd_desc, ars_start, cur, remaining);
+               if (rc)
+                       goto out;
+
+               rc = ars_get_status(nd_desc, ars_status);
+               if (rc)
+                       goto out;
+
+               rc = ars_status_process_records(nvdimm_bus, ars_status, cur);
+               if (rc)
+                       goto out;
+
+               end = min(cur + remaining,
+                       ars_status->address + ars_status->length);
+               done = end - cur;
+               cur += done;
+               remaining -= done;
+       } while (remaining);
+
+ out:
+       kfree(ars_cap);
+       kfree(ars_start);
+       kfree(ars_status);
+       return rc;
+}
+
 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
                struct nd_mapping *nd_mapping, struct nd_region_desc *ndr_desc,
                struct acpi_nfit_memory_map *memdev,
@@ -1585,6 +1781,13 @@ static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
 
        nvdimm_bus = acpi_desc->nvdimm_bus;
        if (nfit_spa_type(spa) == NFIT_SPA_PM) {
+               rc = acpi_nfit_find_poison(acpi_desc, ndr_desc);
+               if (rc) {
+                       dev_err(acpi_desc->dev,
+                               "error while performing ARS to find poison: %d\n",
+                               rc);
+                       return rc;
+               }
                if (!nvdimm_pmem_region_create(nvdimm_bus, ndr_desc))
                        return -ENOMEM;
        } else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE) {
index 82c49bb870555fc6636ebc30106c396807f12d7b..21003b7f0b384923ebb5c35c38047329ed7c660c 100644 (file)
@@ -325,6 +325,7 @@ struct nvdimm_bus *__nvdimm_bus_register(struct device *parent,
        if (!nvdimm_bus)
                return NULL;
        INIT_LIST_HEAD(&nvdimm_bus->list);
+       INIT_LIST_HEAD(&nvdimm_bus->poison_list);
        init_waitqueue_head(&nvdimm_bus->probe_wait);
        nvdimm_bus->id = ida_simple_get(&nd_ida, 0, 0, GFP_KERNEL);
        mutex_init(&nvdimm_bus->reconfig_mutex);
@@ -359,6 +360,191 @@ struct nvdimm_bus *__nvdimm_bus_register(struct device *parent,
 }
 EXPORT_SYMBOL_GPL(__nvdimm_bus_register);
 
+/**
+ * __add_badblock_range() - Convert a physical address range to bad sectors
+ * @disk:      the disk associated with the namespace
+ * @ns_offset: namespace offset where the error range begins (in bytes)
+ * @len:       number of bytes of poison to be added
+ *
+ * This assumes that the range provided with (ns_offset, len) is within
+ * the bounds of physical addresses for this namespace, i.e. lies in the
+ * interval [ns_start, ns_start + ns_size)
+ */
+static int __add_badblock_range(struct gendisk *disk, u64 ns_offset, u64 len)
+{
+       unsigned int sector_size = queue_logical_block_size(disk->queue);
+       sector_t start_sector;
+       u64 num_sectors;
+       u32 rem;
+       int rc;
+
+       start_sector = div_u64(ns_offset, sector_size);
+       num_sectors = div_u64_rem(len, sector_size, &rem);
+       if (rem)
+               num_sectors++;
+
+       if (!disk->bb) {
+               rc = disk_alloc_badblocks(disk);
+               if (rc)
+                       return rc;
+       }
+
+       if (unlikely(num_sectors > (u64)INT_MAX)) {
+               u64 remaining = num_sectors;
+               sector_t s = start_sector;
+
+               while (remaining) {
+                       int done = min_t(u64, remaining, INT_MAX);
+
+                       rc = disk_set_badblocks(disk, s, done);
+                       if (rc)
+                               return rc;
+                       remaining -= done;
+                       s += done;
+               }
+               return 0;
+       } else
+               return disk_set_badblocks(disk, start_sector, num_sectors);
+}
+
+/**
+ * nvdimm_namespace_add_poison() - Convert a list of poison ranges to badblocks
+ * @disk:      the gendisk associated with the namespace where badblocks
+ *             will be stored
+ * @offset:    offset at the start of the namespace before 'sector 0'
+ * @ndns:      the namespace containing poison ranges
+ *
+ * The poison list generated during NFIT initialization may contain multiple,
+ * possibly overlapping ranges in the SPA (System Physical Address) space.
+ * Compare each of these ranges to the namespace currently being initialized,
+ * and add badblocks to the gendisk for all matching sub-ranges
+ *
+ * Return:
+ * 0 - Success
+ */
+int nvdimm_namespace_add_poison(struct gendisk *disk, resource_size_t offset,
+               struct nd_namespace_common *ndns)
+{
+       struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
+       struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
+       struct nvdimm_bus *nvdimm_bus;
+       struct list_head *poison_list;
+       u64 ns_start, ns_end, ns_size;
+       struct nd_poison *pl;
+       int rc;
+
+       ns_size = nvdimm_namespace_capacity(ndns) - offset;
+       ns_start = nsio->res.start + offset;
+       ns_end = nsio->res.end;
+
+       nvdimm_bus = to_nvdimm_bus(nd_region->dev.parent);
+       poison_list = &nvdimm_bus->poison_list;
+       if (list_empty(poison_list))
+               return 0;
+
+       list_for_each_entry(pl, poison_list, list) {
+               u64 pl_end = pl->start + pl->length - 1;
+
+               /* Discard intervals with no intersection */
+               if (pl_end < ns_start)
+                       continue;
+               if (pl->start > ns_end)
+                       continue;
+               /* Deal with any overlap after start of the namespace */
+               if (pl->start >= ns_start) {
+                       u64 start = pl->start;
+                       u64 len;
+
+                       if (pl_end <= ns_end)
+                               len = pl->length;
+                       else
+                               len = ns_start + ns_size - pl->start;
+
+                       rc = __add_badblock_range(disk, start - ns_start, len);
+                       if (rc)
+                               return rc;
+                       dev_info(&nvdimm_bus->dev,
+                               "Found a poison range (0x%llx, 0x%llx)\n",
+                               start, len);
+                       continue;
+               }
+               /* Deal with overlap for poison starting before the namespace */
+               if (pl->start < ns_start) {
+                       u64 len;
+
+                       if (pl_end < ns_end)
+                               len = pl->start + pl->length - ns_start;
+                       else
+                               len = ns_size;
+
+                       rc = __add_badblock_range(disk, 0, len);
+                       if (rc)
+                               return rc;
+                       dev_info(&nvdimm_bus->dev,
+                               "Found a poison range (0x%llx, 0x%llx)\n",
+                               pl->start, len);
+               }
+       }
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(nvdimm_namespace_add_poison);
+
+static int __add_poison(struct nvdimm_bus *nvdimm_bus, u64 addr, u64 length)
+{
+       struct nd_poison *pl;
+
+       pl = kzalloc(sizeof(*pl), GFP_KERNEL);
+       if (!pl)
+               return -ENOMEM;
+
+       pl->start = addr;
+       pl->length = length;
+       list_add_tail(&pl->list, &nvdimm_bus->poison_list);
+
+       return 0;
+}
+
+int nvdimm_bus_add_poison(struct nvdimm_bus *nvdimm_bus, u64 addr, u64 length)
+{
+       struct nd_poison *pl;
+
+       if (list_empty(&nvdimm_bus->poison_list))
+               return __add_poison(nvdimm_bus, addr, length);
+
+       /*
+        * There is a chance this is a duplicate, check for those first.
+        * This will be the common case as ARS_STATUS returns all known
+        * errors in the SPA space, and we can't query it per region
+        */
+       list_for_each_entry(pl, &nvdimm_bus->poison_list, list)
+               if (pl->start == addr) {
+                       /* If length has changed, update this list entry */
+                       if (pl->length != length)
+                               pl->length = length;
+                       return 0;
+               }
+
+       /*
+        * If not a duplicate or a simple length update, add the entry as is,
+        * as any overlapping ranges will get resolved when the list is consumed
+        * and converted to badblocks
+        */
+       return __add_poison(nvdimm_bus, addr, length);
+}
+EXPORT_SYMBOL_GPL(nvdimm_bus_add_poison);
+
+static void free_poison_list(struct list_head *poison_list)
+{
+       struct nd_poison *pl, *next;
+
+       list_for_each_entry_safe(pl, next, poison_list, list) {
+               list_del(&pl->list);
+               kfree(pl);
+       }
+       list_del_init(poison_list);
+}
+
 static int child_unregister(struct device *dev, void *data)
 {
        /*
@@ -385,6 +571,7 @@ void nvdimm_bus_unregister(struct nvdimm_bus *nvdimm_bus)
 
        nd_synchronize();
        device_for_each_child(&nvdimm_bus->dev, NULL, child_unregister);
+       free_poison_list(&nvdimm_bus->poison_list);
        nvdimm_bus_destroy_ndctl(nvdimm_bus);
 
        device_unregister(&nvdimm_bus->dev);
index 159aed53204249ce6e9da13fcaa0b58406a1f2f6..d3b7ea78df96bdba7e3ce18707cc56a941b4da57 100644 (file)
@@ -30,6 +30,7 @@ struct nvdimm_bus {
        struct list_head list;
        struct device dev;
        int id, probe_active;
+       struct list_head poison_list;
        struct mutex reconfig_mutex;
 };
 
@@ -89,4 +90,6 @@ bool __nd_attach_ndns(struct device *dev, struct nd_namespace_common *attach,
 ssize_t nd_namespace_store(struct device *dev,
                struct nd_namespace_common **_ndns, const char *buf,
                size_t len);
+int nvdimm_namespace_add_poison(struct gendisk *disk, resource_size_t offset,
+               struct nd_namespace_common *ndns);
 #endif /* __ND_CORE_H__ */
index 417e521d299cb4645f5739f7608cc2d0d84a701c..ba91fcd5818db1be3541e60e671fa00fbf92fe2e 100644 (file)
@@ -38,6 +38,12 @@ enum {
 #endif
 };
 
+struct nd_poison {
+       u64 start;
+       u64 length;
+       struct list_head list;
+};
+
 struct nvdimm_drvdata {
        struct device *dev;
        int nsindex_size;
index 8ee79893d2f5f6f5165558baa22f236738b99c45..5b95043443a39ebbeb77392169292172b1382052 100644 (file)
@@ -27,6 +27,7 @@
 #include <linux/slab.h>
 #include <linux/pmem.h>
 #include <linux/nd.h>
+#include "nd-core.h"
 #include "pfn.h"
 #include "nd.h"
 
@@ -168,6 +169,7 @@ static int pmem_attach_disk(struct device *dev,
 {
        int nid = dev_to_node(dev);
        struct gendisk *disk;
+       int ret;
 
        pmem->pmem_queue = blk_alloc_queue_node(GFP_KERNEL, nid);
        if (!pmem->pmem_queue)
@@ -196,6 +198,10 @@ static int pmem_attach_disk(struct device *dev,
        set_capacity(disk, (pmem->size - pmem->data_offset) / 512);
        pmem->pmem_disk = disk;
 
+       ret = nvdimm_namespace_add_poison(disk, pmem->data_offset, ndns);
+       if (ret)
+               return ret;
+
        add_disk(disk);
        revalidate_disk(disk);
 
index 3f021dc5da8c34e59768dca61dda323035368c0f..bed40dff0e86999b5c39c82ebadc29f25e0060ec 100644 (file)
@@ -116,6 +116,7 @@ static inline struct nd_blk_region_desc *to_blk_region_desc(
 
 }
 
+int nvdimm_bus_add_poison(struct nvdimm_bus *nvdimm_bus, u64 addr, u64 length);
 struct nvdimm_bus *__nvdimm_bus_register(struct device *parent,
                struct nvdimm_bus_descriptor *nfit_desc, struct module *module);
 #define nvdimm_bus_register(parent, desc) \