--- /dev/null
+/*
+ * PowerNV OPAL Dump Interface
+ *
+ * Copyright 2013,2014 IBM Corp.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/kobject.h>
+#include <linux/mm.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/pagemap.h>
+#include <linux/delay.h>
+
+#include <asm/opal.h>
+
+#define DUMP_TYPE_FSP  0x01
+
+struct dump_obj {
+       struct kobject  kobj;
+       struct bin_attribute dump_attr;
+       uint32_t        id;  /* becomes object name */
+       uint32_t        type;
+       uint32_t        size;
+       char            *buffer;
+};
+#define to_dump_obj(x) container_of(x, struct dump_obj, kobj)
+
+struct dump_attribute {
+       struct attribute attr;
+       ssize_t (*show)(struct dump_obj *dump, struct dump_attribute *attr,
+                       char *buf);
+       ssize_t (*store)(struct dump_obj *dump, struct dump_attribute *attr,
+                        const char *buf, size_t count);
+};
+#define to_dump_attr(x) container_of(x, struct dump_attribute, attr)
+
+static ssize_t dump_id_show(struct dump_obj *dump_obj,
+                           struct dump_attribute *attr,
+                           char *buf)
+{
+       return sprintf(buf, "0x%x\n", dump_obj->id);
+}
+
+static const char* dump_type_to_string(uint32_t type)
+{
+       switch (type) {
+       case 0x01: return "SP Dump";
+       case 0x02: return "System/Platform Dump";
+       case 0x03: return "SMA Dump";
+       default: return "unknown";
+       }
+}
+
+static ssize_t dump_type_show(struct dump_obj *dump_obj,
+                             struct dump_attribute *attr,
+                             char *buf)
+{
+       
+       return sprintf(buf, "0x%x %s\n", dump_obj->type,
+                      dump_type_to_string(dump_obj->type));
+}
+
+static ssize_t dump_ack_show(struct dump_obj *dump_obj,
+                            struct dump_attribute *attr,
+                            char *buf)
+{
+       return sprintf(buf, "ack - acknowledge dump\n");
+}
+
+/*
+ * Send acknowledgement to OPAL
+ */
+static int64_t dump_send_ack(uint32_t dump_id)
+{
+       int rc;
+
+       rc = opal_dump_ack(dump_id);
+       if (rc)
+               pr_warn("%s: Failed to send ack to Dump ID 0x%x (%d)\n",
+                       __func__, dump_id, rc);
+       return rc;
+}
+
+static void delay_release_kobj(void *kobj)
+{
+       kobject_put((struct kobject *)kobj);
+}
+
+static ssize_t dump_ack_store(struct dump_obj *dump_obj,
+                             struct dump_attribute *attr,
+                             const char *buf,
+                             size_t count)
+{
+       dump_send_ack(dump_obj->id);
+       sysfs_schedule_callback(&dump_obj->kobj, delay_release_kobj,
+                               &dump_obj->kobj, THIS_MODULE);
+       return count;
+}
+
+/* Attributes of a dump
+ * The binary attribute of the dump itself is dynamic
+ * due to the dynamic size of the dump
+ */
+static struct dump_attribute id_attribute =
+       __ATTR(id, 0666, dump_id_show, NULL);
+static struct dump_attribute type_attribute =
+       __ATTR(type, 0666, dump_type_show, NULL);
+static struct dump_attribute ack_attribute =
+       __ATTR(acknowledge, 0660, dump_ack_show, dump_ack_store);
+
+static ssize_t init_dump_show(struct dump_obj *dump_obj,
+                             struct dump_attribute *attr,
+                             char *buf)
+{
+       return sprintf(buf, "1 - initiate dump\n");
+}
+
+static int64_t dump_fips_init(uint8_t type)
+{
+       int rc;
+
+       rc = opal_dump_init(type);
+       if (rc)
+               pr_warn("%s: Failed to initiate FipS dump (%d)\n",
+                       __func__, rc);
+       return rc;
+}
+
+static ssize_t init_dump_store(struct dump_obj *dump_obj,
+                              struct dump_attribute *attr,
+                              const char *buf,
+                              size_t count)
+{
+       dump_fips_init(DUMP_TYPE_FSP);
+       pr_info("%s: Initiated FSP dump\n", __func__);
+       return count;
+}
+
+static struct dump_attribute initiate_attribute =
+       __ATTR(initiate_dump, 0600, init_dump_show, init_dump_store);
+
+static struct attribute *initiate_attrs[] = {
+       &initiate_attribute.attr,
+       NULL,
+};
+
+static struct attribute_group initiate_attr_group = {
+       .attrs = initiate_attrs,
+};
+
+static struct kset *dump_kset;
+
+static ssize_t dump_attr_show(struct kobject *kobj,
+                             struct attribute *attr,
+                             char *buf)
+{
+       struct dump_attribute *attribute;
+       struct dump_obj *dump;
+
+       attribute = to_dump_attr(attr);
+       dump = to_dump_obj(kobj);
+
+       if (!attribute->show)
+               return -EIO;
+
+       return attribute->show(dump, attribute, buf);
+}
+
+static ssize_t dump_attr_store(struct kobject *kobj,
+                              struct attribute *attr,
+                              const char *buf, size_t len)
+{
+       struct dump_attribute *attribute;
+       struct dump_obj *dump;
+
+       attribute = to_dump_attr(attr);
+       dump = to_dump_obj(kobj);
+
+       if (!attribute->store)
+               return -EIO;
+
+       return attribute->store(dump, attribute, buf, len);
+}
+
+static const struct sysfs_ops dump_sysfs_ops = {
+       .show = dump_attr_show,
+       .store = dump_attr_store,
+};
+
+static void dump_release(struct kobject *kobj)
+{
+       struct dump_obj *dump;
+
+       dump = to_dump_obj(kobj);
+       vfree(dump->buffer);
+       kfree(dump);
+}
+
+static struct attribute *dump_default_attrs[] = {
+       &id_attribute.attr,
+       &type_attribute.attr,
+       &ack_attribute.attr,
+       NULL,
+};
+
+static struct kobj_type dump_ktype = {
+       .sysfs_ops = &dump_sysfs_ops,
+       .release = &dump_release,
+       .default_attrs = dump_default_attrs,
+};
+
+static void free_dump_sg_list(struct opal_sg_list *list)
+{
+       struct opal_sg_list *sg1;
+       while (list) {
+               sg1 = list->next;
+               kfree(list);
+               list = sg1;
+       }
+       list = NULL;
+}
+
+static struct opal_sg_list *dump_data_to_sglist(struct dump_obj *dump)
+{
+       struct opal_sg_list *sg1, *list = NULL;
+       void *addr;
+       int64_t size;
+
+       addr = dump->buffer;
+       size = dump->size;
+
+       sg1 = kzalloc(PAGE_SIZE, GFP_KERNEL);
+       if (!sg1)
+               goto nomem;
+
+       list = sg1;
+       sg1->num_entries = 0;
+       while (size > 0) {
+               /* Translate virtual address to physical address */
+               sg1->entry[sg1->num_entries].data =
+                       (void *)(vmalloc_to_pfn(addr) << PAGE_SHIFT);
+
+               if (size > PAGE_SIZE)
+                       sg1->entry[sg1->num_entries].length = PAGE_SIZE;
+               else
+                       sg1->entry[sg1->num_entries].length = size;
+
+               sg1->num_entries++;
+               if (sg1->num_entries >= SG_ENTRIES_PER_NODE) {
+                       sg1->next = kzalloc(PAGE_SIZE, GFP_KERNEL);
+                       if (!sg1->next)
+                               goto nomem;
+
+                       sg1 = sg1->next;
+                       sg1->num_entries = 0;
+               }
+               addr += PAGE_SIZE;
+               size -= PAGE_SIZE;
+       }
+       return list;
+
+nomem:
+       pr_err("%s : Failed to allocate memory\n", __func__);
+       free_dump_sg_list(list);
+       return NULL;
+}
+
+static void sglist_to_phy_addr(struct opal_sg_list *list)
+{
+       struct opal_sg_list *sg, *next;
+
+       for (sg = list; sg; sg = next) {
+               next = sg->next;
+               /* Don't translate NULL pointer for last entry */
+               if (sg->next)
+                       sg->next = (struct opal_sg_list *)__pa(sg->next);
+               else
+                       sg->next = NULL;
+
+               /* Convert num_entries to length */
+               sg->num_entries =
+                       sg->num_entries * sizeof(struct opal_sg_entry) + 16;
+       }
+}
+
+static int64_t dump_read_info(uint32_t *id, uint32_t *size, uint32_t *type)
+{
+       int rc;
+       *type = 0xffffffff;
+
+       rc = opal_dump_info2(id, size, type);
+
+       if (rc == OPAL_PARAMETER)
+               rc = opal_dump_info(id, size);
+
+       if (rc)
+               pr_warn("%s: Failed to get dump info (%d)\n",
+                       __func__, rc);
+       return rc;
+}
+
+static int64_t dump_read_data(struct dump_obj *dump)
+{
+       struct opal_sg_list *list;
+       uint64_t addr;
+       int64_t rc;
+
+       /* Allocate memory */
+       dump->buffer = vzalloc(PAGE_ALIGN(dump->size));
+       if (!dump->buffer) {
+               pr_err("%s : Failed to allocate memory\n", __func__);
+               rc = -ENOMEM;
+               goto out;
+       }
+
+       /* Generate SG list */
+       list = dump_data_to_sglist(dump);
+       if (!list) {
+               rc = -ENOMEM;
+               goto out;
+       }
+
+       /* Translate sg list addr to real address */
+       sglist_to_phy_addr(list);
+
+       /* First entry address */
+       addr = __pa(list);
+
+       /* Fetch data */
+       rc = OPAL_BUSY_EVENT;
+       while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
+               rc = opal_dump_read(dump->id, addr);
+               if (rc == OPAL_BUSY_EVENT) {
+                       opal_poll_events(NULL);
+                       msleep(20);
+               }
+       }
+
+       if (rc != OPAL_SUCCESS && rc != OPAL_PARTIAL)
+               pr_warn("%s: Extract dump failed for ID 0x%x\n",
+                       __func__, dump->id);
+
+       /* Free SG list */
+       free_dump_sg_list(list);
+
+out:
+       return rc;
+}
+
+static ssize_t dump_attr_read(struct file *filep, struct kobject *kobj,
+                             struct bin_attribute *bin_attr,
+                             char *buffer, loff_t pos, size_t count)
+{
+       ssize_t rc;
+
+       struct dump_obj *dump = to_dump_obj(kobj);
+
+       if (!dump->buffer) {
+               rc = dump_read_data(dump);
+
+               if (rc != OPAL_SUCCESS && rc != OPAL_PARTIAL) {
+                       vfree(dump->buffer);
+                       dump->buffer = NULL;
+
+                       return -EIO;
+               }
+               if (rc == OPAL_PARTIAL) {
+                       /* On a partial read, we just return EIO
+                        * and rely on userspace to ask us to try
+                        * again.
+                        */
+                       pr_info("%s: Platform dump partially read.ID = 0x%x\n",
+                               __func__, dump->id);
+                       return -EIO;
+               }
+       }
+
+       memcpy(buffer, dump->buffer + pos, count);
+
+       /* You may think we could free the dump buffer now and retrieve
+        * it again later if needed, but due to current firmware limitation,
+        * that's not the case. So, once read into userspace once,
+        * we keep the dump around until it's acknowledged by userspace.
+        */
+
+       return count;
+}
+
+static struct dump_obj *create_dump_obj(uint32_t id, size_t size,
+                                       uint32_t type)
+{
+       struct dump_obj *dump;
+       int rc;
+
+       dump = kzalloc(sizeof(*dump), GFP_KERNEL);
+       if (!dump)
+               return NULL;
+
+       dump->kobj.kset = dump_kset;
+
+       kobject_init(&dump->kobj, &dump_ktype);
+
+       sysfs_bin_attr_init(&dump->dump_attr);
+
+       dump->dump_attr.attr.name = "dump";
+       dump->dump_attr.attr.mode = 0400;
+       dump->dump_attr.size = size;
+       dump->dump_attr.read = dump_attr_read;
+
+       dump->id = id;
+       dump->size = size;
+       dump->type = type;
+
+       rc = kobject_add(&dump->kobj, NULL, "0x%x-0x%x", type, id);
+       if (rc) {
+               kobject_put(&dump->kobj);
+               return NULL;
+       }
+
+       rc = sysfs_create_bin_file(&dump->kobj, &dump->dump_attr);
+       if (rc) {
+               kobject_put(&dump->kobj);
+               return NULL;
+       }
+
+       pr_info("%s: New platform dump. ID = 0x%x Size %u\n",
+               __func__, dump->id, dump->size);
+
+       kobject_uevent(&dump->kobj, KOBJ_ADD);
+
+       return dump;
+}
+
+static int process_dump(void)
+{
+       int rc;
+       uint32_t dump_id, dump_size, dump_type;
+       struct dump_obj *dump;
+       char name[22];
+
+       rc = dump_read_info(&dump_id, &dump_size, &dump_type);
+       if (rc != OPAL_SUCCESS)
+               return rc;
+
+       sprintf(name, "0x%x-0x%x", dump_type, dump_id);
+
+       /* we may get notified twice, let's handle
+        * that gracefully and not create two conflicting
+        * entries.
+        */
+       if (kset_find_obj(dump_kset, name))
+               return 0;
+
+       dump = create_dump_obj(dump_id, dump_size, dump_type);
+       if (!dump)
+               return -1;
+
+       return 0;
+}
+
+static void dump_work_fn(struct work_struct *work)
+{
+       process_dump();
+}
+
+static DECLARE_WORK(dump_work, dump_work_fn);
+
+static void schedule_process_dump(void)
+{
+       schedule_work(&dump_work);
+}
+
+/*
+ * New dump available notification
+ *
+ * Once we get notification, we add sysfs entries for it.
+ * We only fetch the dump on demand, and create sysfs asynchronously.
+ */
+static int dump_event(struct notifier_block *nb,
+                     unsigned long events, void *change)
+{
+       if (events & OPAL_EVENT_DUMP_AVAIL)
+               schedule_process_dump();
+
+       return 0;
+}
+
+static struct notifier_block dump_nb = {
+       .notifier_call  = dump_event,
+       .next           = NULL,
+       .priority       = 0
+};
+
+void __init opal_platform_dump_init(void)
+{
+       int rc;
+
+       dump_kset = kset_create_and_add("dump", NULL, opal_kobj);
+       if (!dump_kset) {
+               pr_warn("%s: Failed to create dump kset\n", __func__);
+               return;
+       }
+
+       rc = sysfs_create_group(&dump_kset->kobj, &initiate_attr_group);
+       if (rc) {
+               pr_warn("%s: Failed to create initiate dump attr group\n",
+                       __func__);
+               kobject_put(&dump_kset->kobj);
+               return;
+       }
+
+       rc = opal_notifier_register(&dump_nb);
+       if (rc) {
+               pr_warn("%s: Can't register OPAL event notifier (%d)\n",
+                       __func__, rc);
+               return;
+       }
+
+       opal_dump_resend_notification();
+}