nfit, libnvdimm: allow an ARS scrub to be triggered on demand
authorVishal Verma <vishal.l.verma@intel.com>
Sun, 24 Jul 2016 04:51:42 +0000 (21:51 -0700)
committerDan Williams <dan.j.williams@intel.com>
Sun, 24 Jul 2016 04:51:42 +0000 (21:51 -0700)
Normally, an ARS (Address Range Scrub) only happens at
boot/initialization time. There can however arise situations where a
bus-wide rescan is needed - notably, in the case of discovering a latent
media error, we should do a full rescan to figure out what other sectors
are bad, and thus potentially avoid triggering an mce on them in the
future. Also provide a sysfs trigger to start a bus-wide scrub.

Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
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/acpi/nfit.h
drivers/nvdimm/core.c
include/linux/libnvdimm.h

index be7c2fd..19d0dfd 100644 (file)
@@ -15,6 +15,7 @@
 #include <linux/module.h>
 #include <linux/mutex.h>
 #include <linux/ndctl.h>
+#include <linux/sysfs.h>
 #include <linux/delay.h>
 #include <linux/list.h>
 #include <linux/acpi.h>
@@ -874,14 +875,87 @@ static ssize_t revision_show(struct device *dev,
 }
 static DEVICE_ATTR_RO(revision);
 
+/*
+ * This shows the number of full Address Range Scrubs that have been
+ * completed since driver load time. Userspace can wait on this using
+ * select/poll etc. A '+' at the end indicates an ARS is in progress
+ */
+static ssize_t scrub_show(struct device *dev,
+               struct device_attribute *attr, char *buf)
+{
+       struct nvdimm_bus_descriptor *nd_desc;
+       ssize_t rc = -ENXIO;
+
+       device_lock(dev);
+       nd_desc = dev_get_drvdata(dev);
+       if (nd_desc) {
+               struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
+
+               rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
+                               (work_busy(&acpi_desc->work)) ? "+\n" : "\n");
+       }
+       device_unlock(dev);
+       return rc;
+}
+
+static int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc);
+
+static ssize_t scrub_store(struct device *dev,
+               struct device_attribute *attr, const char *buf, size_t size)
+{
+       struct nvdimm_bus_descriptor *nd_desc;
+       ssize_t rc;
+       long val;
+
+       rc = kstrtol(buf, 0, &val);
+       if (rc)
+               return rc;
+       if (val != 1)
+               return -EINVAL;
+
+       device_lock(dev);
+       nd_desc = dev_get_drvdata(dev);
+       if (nd_desc) {
+               struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
+
+               rc = acpi_nfit_ars_rescan(acpi_desc);
+       }
+       device_unlock(dev);
+       if (rc)
+               return rc;
+       return size;
+}
+static DEVICE_ATTR_RW(scrub);
+
+static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
+{
+       struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
+       const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
+               | 1 << ND_CMD_ARS_STATUS;
+
+       return (nd_desc->cmd_mask & mask) == mask;
+}
+
+static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
+{
+       struct device *dev = container_of(kobj, struct device, kobj);
+       struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
+
+       if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
+               return 0;
+       return a->mode;
+}
+
 static struct attribute *acpi_nfit_attributes[] = {
        &dev_attr_revision.attr,
+       &dev_attr_scrub.attr,
        NULL,
 };
 
 static struct attribute_group acpi_nfit_attribute_group = {
        .name = "nfit",
        .attrs = acpi_nfit_attributes,
+       .is_visible = nfit_visible,
 };
 
 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
@@ -2054,7 +2128,7 @@ static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
        unsigned int tmo = scrub_timeout;
        int rc;
 
-       if (nfit_spa->ars_done || !nfit_spa->nd_region)
+       if (!nfit_spa->ars_required || !nfit_spa->nd_region)
                return;
 
        rc = ars_start(acpi_desc, nfit_spa);
@@ -2143,7 +2217,9 @@ static void acpi_nfit_scrub(struct work_struct *work)
         * firmware initiated scrubs to complete and then we go search for the
         * affected spa regions to mark them scanned.  In the second phase we
         * initiate a directed scrub for every range that was not scrubbed in
-        * phase 1.
+        * phase 1. If we're called for a 'rescan', we harmlessly pass through
+        * the first phase, but really only care about running phase 2, where
+        * regions can be notified of new poison.
         */
 
        /* process platform firmware initiated scrubs */
@@ -2246,14 +2322,17 @@ static void acpi_nfit_scrub(struct work_struct *work)
                 * Flag all the ranges that still need scrubbing, but
                 * register them now to make data available.
                 */
-               if (nfit_spa->nd_region)
-                       nfit_spa->ars_done = 1;
-               else
+               if (!nfit_spa->nd_region) {
+                       nfit_spa->ars_required = 1;
                        acpi_nfit_register_region(acpi_desc, nfit_spa);
+               }
        }
 
        list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
                acpi_nfit_async_scrub(acpi_desc, nfit_spa);
+       acpi_desc->scrub_count++;
+       if (acpi_desc->scrub_count_state)
+               sysfs_notify_dirent(acpi_desc->scrub_count_state);
        mutex_unlock(&acpi_desc->init_mutex);
 }
 
@@ -2291,12 +2370,48 @@ static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
        return 0;
 }
 
+static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
+{
+       struct device *dev = acpi_desc->dev;
+       struct kernfs_node *nfit;
+       struct device *bus_dev;
+
+       if (!ars_supported(acpi_desc->nvdimm_bus))
+               return 0;
+
+       bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
+       nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
+       if (!nfit) {
+               dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
+               return -ENODEV;
+       }
+       acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
+       sysfs_put(nfit);
+       if (!acpi_desc->scrub_count_state) {
+               dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
+               return -ENODEV;
+       }
+
+       return 0;
+}
+
 static void acpi_nfit_destruct(void *data)
 {
        struct acpi_nfit_desc *acpi_desc = data;
+       struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
 
        acpi_desc->cancel = 1;
+       /*
+        * Bounce the nvdimm bus lock to make sure any in-flight
+        * acpi_nfit_ars_rescan() submissions have had a chance to
+        * either submit or see ->cancel set.
+        */
+       device_lock(bus_dev);
+       device_unlock(bus_dev);
+
        flush_workqueue(nfit_wq);
+       if (acpi_desc->scrub_count_state)
+               sysfs_put(acpi_desc->scrub_count_state);
        nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
        acpi_desc->nvdimm_bus = NULL;
 }
@@ -2309,14 +2424,21 @@ int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
        int rc;
 
        if (!acpi_desc->nvdimm_bus) {
+               acpi_nfit_init_dsms(acpi_desc);
+
                acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
                                &acpi_desc->nd_desc);
                if (!acpi_desc->nvdimm_bus)
                        return -ENOMEM;
+
                rc = devm_add_action_or_reset(dev, acpi_nfit_destruct,
                                acpi_desc);
                if (rc)
                        return rc;
+
+               rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
+               if (rc)
+                       return rc;
        }
 
        mutex_lock(&acpi_desc->init_mutex);
@@ -2360,8 +2482,6 @@ int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
        if (rc)
                goto out_unlock;
 
-       acpi_nfit_init_dsms(acpi_desc);
-
        rc = acpi_nfit_register_dimms(acpi_desc);
        if (rc)
                goto out_unlock;
@@ -2429,6 +2549,33 @@ static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
        return 0;
 }
 
+static int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc)
+{
+       struct device *dev = acpi_desc->dev;
+       struct nfit_spa *nfit_spa;
+
+       if (work_busy(&acpi_desc->work))
+               return -EBUSY;
+
+       if (acpi_desc->cancel)
+               return 0;
+
+       mutex_lock(&acpi_desc->init_mutex);
+       list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
+               struct acpi_nfit_system_address *spa = nfit_spa->spa;
+
+               if (nfit_spa_type(spa) != NFIT_SPA_PM)
+                       continue;
+
+               nfit_spa->ars_required = 1;
+       }
+       queue_work(nfit_wq, &acpi_desc->work);
+       dev_dbg(dev, "%s: ars_scan triggered\n", __func__);
+       mutex_unlock(&acpi_desc->init_mutex);
+
+       return 0;
+}
+
 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
 {
        struct nvdimm_bus_descriptor *nd_desc;
index b63a583..6ecf337 100644 (file)
@@ -80,7 +80,7 @@ enum {
 struct nfit_spa {
        struct list_head list;
        struct nd_region *nd_region;
-       unsigned int ars_done:1;
+       unsigned int ars_required:1;
        u32 clear_err_unit;
        u32 max_ars;
        struct acpi_nfit_system_address spa[0];
@@ -148,6 +148,8 @@ struct acpi_nfit_desc {
        struct nd_cmd_ars_status *ars_status;
        size_t ars_status_size;
        struct work_struct work;
+       struct kernfs_node *scrub_count_state;
+       unsigned int scrub_count;
        unsigned int cancel:1;
        unsigned long dimm_cmd_force_en;
        unsigned long bus_cmd_force_en;
index 2c98f95..715583f 100644 (file)
@@ -201,6 +201,13 @@ struct nvdimm_bus_descriptor *to_nd_desc(struct nvdimm_bus *nvdimm_bus)
 }
 EXPORT_SYMBOL_GPL(to_nd_desc);
 
+struct device *to_nvdimm_bus_dev(struct nvdimm_bus *nvdimm_bus)
+{
+       /* struct nvdimm_bus definition is private to libnvdimm */
+       return &nvdimm_bus->dev;
+}
+EXPORT_SYMBOL_GPL(to_nvdimm_bus_dev);
+
 static bool is_uuid_sep(char sep)
 {
        if (sep == '\n' || sep == '-' || sep == ':' || sep == '\0')
index 2ab869d..b519e13 100644 (file)
@@ -137,6 +137,7 @@ struct nvdimm *to_nvdimm(struct device *dev);
 struct nd_region *to_nd_region(struct device *dev);
 struct nd_blk_region *to_nd_blk_region(struct device *dev);
 struct nvdimm_bus_descriptor *to_nd_desc(struct nvdimm_bus *nvdimm_bus);
+struct device *to_nvdimm_bus_dev(struct nvdimm_bus *nvdimm_bus);
 const char *nvdimm_name(struct nvdimm *nvdimm);
 unsigned long nvdimm_cmd_mask(struct nvdimm *nvdimm);
 void *nvdimm_provider_data(struct nvdimm *nvdimm);