Expose known poison in SPA ranges to the block layer

index 51cf825..298805a 100644
@@ -1088,6 +1088,8 @@ static void nfit_test1_setup(struct nfit_test *t)
 	struct acpi_nfit_memory_map *memdev;
 	struct acpi_nfit_control_region *dcr;
 	struct acpi_nfit_system_address *spa;
+	struct nvdimm_bus_descriptor *nd_desc;
+	struct acpi_nfit_desc *acpi_desc;
 
 	offset = 0;
 	/* spa0 (flat range with no bdw aliasing) */

@@ -1135,6 +1137,13 @@ static void nfit_test1_setup(struct nfit_test *t)
 	dcr->command_size = 0;
 	dcr->status_offset = 0;
 	dcr->status_size = 0;
+
+	acpi_desc = &t->acpi_desc;
+	set_bit(ND_CMD_ARS_CAP, &acpi_desc->bus_dsm_force_en);
+	set_bit(ND_CMD_ARS_START, &acpi_desc->bus_dsm_force_en);
+	set_bit(ND_CMD_ARS_STATUS, &acpi_desc->bus_dsm_force_en);
+	nd_desc = &acpi_desc->nd_desc;
+	nd_desc->ndctl = nfit_test_ctl;
 }
 
 static int nfit_test_blk_do_io(struct nd_blk_region *ndbr, resource_size_t dpa,

index e7ed39b..bd8726d 100644
@@ -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 +1474,195 @@ 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 & 0xffff0000) {
+		case 0:
+			return 0;
+		case 1:
+			/* ARS is in progress */
+			msleep(1000);
+			break;
+		case 2:
+			/* No ARS performed for the current boot */
+		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, skip this but continue initialization */
+	if (ars_cap->status & 0xffff) {
+		dev_warn(acpi_desc->dev,
+			"ARS unsupported, won't construct an error list\n");
+		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 +1775,9 @@ 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)
+			return rc;
 		if (!nvdimm_pmem_region_create(nvdimm_bus, ndr_desc))
 			return -ENOMEM;
 	} else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE) {

81 lines not on this page · raw message

index 159aed5..29acdaa 100644
@@ -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;
 };
 

index 417e521..ba91fcd 100644
@@ -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 3f021dc..43b82a0 100644
@@ -116,6 +116,8 @@ static inline struct nd_blk_region_desc *to_blk_region_desc(
 
 }
 
+struct list_head *nvdimm_bus_get_poison_list(struct nvdimm_bus *nvdimm_bus);
+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) \

index 8ee7989..462570f 100644
@@ -19,14 +19,18 @@
 #include <linux/blkdev.h>
 #include <linux/hdreg.h>
 #include <linux/init.h>
+#include <linux/list.h>
 #include <linux/platform_device.h>
 #include <linux/module.h>
 #include <linux/memory_hotplug.h>
 #include <linux/moduleparam.h>
+#include <linux/libnvdimm.h>
 #include <linux/vmalloc.h>
+#include <linux/blkdev.h>
 #include <linux/slab.h>
 #include <linux/pmem.h>
 #include <linux/nd.h>
+#include "nd-core.h"
 #include "pfn.h"
 #include "nd.h"
 

@@ -163,11 +167,122 @@ static void pmem_detach_disk(struct pmem_device *pmem)
 	blk_cleanup_queue(pmem->pmem_queue);
 }
 
+/**
+ * pmem_add_badblock_range() - Convert a physical address range to bad sectors
+ * @disk:	the disk associated with the pmem namespace
+ * @start:	start of the physical address range
+ * @length:	number of bytes of poison to be added
+ *
+ * This assumes that the range provided with (start, length) is already within
+ * the bounds of physical addresses for this namespace, i.e. lies in the
+ * interval [pmem->phys_addr, pmem->phys_addr + pmem->size)
+ */
+static int pmem_add_badblock_range(struct gendisk *disk, u64 start, u64 length)
+{
+	unsigned int sector_size = queue_logical_block_size(disk->queue);
+	struct pmem_device *pmem = disk->private_data;
+	sector_t start_sector;
+	u64 num_sectors;
+	u32 rem;
+
+	start_sector = div_u64(start - pmem->phys_addr, sector_size);
+	num_sectors = div_u64_rem(length, sector_size, &rem);
+	if (rem)
+		num_sectors++;
+
+	if (unlikely(num_sectors > (u64)INT_MAX)) {
+		u64 remaining = num_sectors;
+		sector_t s = start_sector;
+		int rc;
+
+		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);
+}
+
+/*
+ * 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 pmem namespace currently being
+ * initialized, and add badblocks for the sub-ranges that match
+ */
+static int pmem_add_poison(struct gendisk *disk, struct nvdimm_bus *nvdimm_bus)
+{
+	struct pmem_device *pmem = disk->private_data;
+	struct list_head *poison_list;
+	struct nd_poison *pl;
+	int rc;
+
+	poison_list = nvdimm_bus_get_poison_list(nvdimm_bus);
+	if (list_empty(poison_list))
+		return 0;
+
+	list_for_each_entry(pl, poison_list, list) {
+		u64 pl_end = pl->start + pl->length - 1;
+		u64 pmem_end = pmem->phys_addr + pmem->size - 1;
+
+		/* Discard intervals with no intersection */
+		if (pl_end < pmem->phys_addr)
+			continue;
+		if (pl->start > pmem_end)
+			continue;
+		/* Deal with any overlap after start of the pmem range */
+		if (pl->start >= pmem->phys_addr) {
+			u64 start = pl->start;
+			u64 len;
+
+			if (pl_end <= pmem_end)
+				len = pl->length;
+			else
+				len = pmem->phys_addr + pmem->size - pl->start;
+
+			rc = pmem_add_badblock_range(disk, 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 pmem range */
+		if (pl->start < pmem->phys_addr) {
+			u64 start = pmem->phys_addr;
+			u64 len;
+
+			if (pl_end < pmem_end)
+				len = pl->start + pl->length - pmem->phys_addr;
+			else
+				len = pmem->size;
+
+			rc = pmem_add_badblock_range(disk, start, len);
+			if (rc)
+				return rc;
+			dev_info(&nvdimm_bus->dev,
+				"Found a poison range (0x%llx, 0x%llx)\n",
+				start, len);
+		}
+	}
+
+	return 0;
+}
+
 static int pmem_attach_disk(struct device *dev,
 		struct nd_namespace_common *ndns, struct pmem_device *pmem)
 {
+	struct nd_region *nd_region = to_nd_region(dev->parent);
+	struct nvdimm_bus *nvdimm_bus;
 	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 +311,15 @@ static int pmem_attach_disk(struct device *dev,
 	set_capacity(disk, (pmem->size - pmem->data_offset) / 512);
 	pmem->pmem_disk = disk;
 
+	ret = disk_alloc_badblocks(disk);
+	if (ret)
+		return ret;
+
+	nvdimm_bus = to_nvdimm_bus(nd_region->dev.parent);
+	ret = pmem_add_poison(disk, nvdimm_bus);
+	if (ret)
+		return ret;
+
 	add_disk(disk);
 	revalidate_disk(disk);