]> nv-tegra.nvidia Code Review - linux-2.6.git/blobdiff - drivers/base/memory.c
PM / Sleep: Add user space interface for manipulating wakeup sources, v3
[linux-2.6.git] / drivers / base / memory.c
index cb99daeae936777b45f128783c65fd5622258513..7dda4f790f00d58145cc3b58ce4607c42ea829b1 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * drivers/base/memory.c - basic Memory class support
+ * Memory subsystem support
  *
  * Written by Matt Tolentino <matthew.e.tolentino@intel.com>
  *            Dave Hansen <haveblue@us.ibm.com>
@@ -10,7 +10,6 @@
  * SPARSEMEM should be contained here, or in mm/memory_hotplug.c.
  */
 
-#include <linux/sysdev.h>
 #include <linux/module.h>
 #include <linux/init.h>
 #include <linux/topology.h>
 #include <linux/kobject.h>
 #include <linux/memory_hotplug.h>
 #include <linux/mm.h>
-#include <asm/atomic.h>
+#include <linux/mutex.h>
+#include <linux/stat.h>
+#include <linux/slab.h>
+
+#include <linux/atomic.h>
 #include <asm/uaccess.h>
 
+static DEFINE_MUTEX(mem_sysfs_mutex);
+
 #define MEMORY_CLASS_NAME      "memory"
 
-static struct sysdev_class memory_sysdev_class = {
-       set_kset_name(MEMORY_CLASS_NAME),
-};
+static int sections_per_block;
 
-static const char *memory_uevent_name(struct kset *kset, struct kobject *kobj)
+static inline int base_memory_block_id(int section_nr)
 {
-       return MEMORY_CLASS_NAME;
+       return section_nr / sections_per_block;
 }
 
-static int memory_uevent(struct kset *kset, struct kobj_uevent_env *env)
-{
-       int retval = 0;
-
-       return retval;
-}
-
-static struct kset_uevent_ops memory_uevent_ops = {
-       .name           = memory_uevent_name,
-       .uevent         = memory_uevent,
+static struct bus_type memory_subsys = {
+       .name = MEMORY_CLASS_NAME,
+       .dev_name = MEMORY_CLASS_NAME,
 };
 
 static BLOCKING_NOTIFIER_HEAD(memory_chain);
@@ -52,44 +48,71 @@ int register_memory_notifier(struct notifier_block *nb)
 {
         return blocking_notifier_chain_register(&memory_chain, nb);
 }
+EXPORT_SYMBOL(register_memory_notifier);
 
 void unregister_memory_notifier(struct notifier_block *nb)
 {
         blocking_notifier_chain_unregister(&memory_chain, nb);
 }
+EXPORT_SYMBOL(unregister_memory_notifier);
+
+static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain);
+
+int register_memory_isolate_notifier(struct notifier_block *nb)
+{
+       return atomic_notifier_chain_register(&memory_isolate_chain, nb);
+}
+EXPORT_SYMBOL(register_memory_isolate_notifier);
+
+void unregister_memory_isolate_notifier(struct notifier_block *nb)
+{
+       atomic_notifier_chain_unregister(&memory_isolate_chain, nb);
+}
+EXPORT_SYMBOL(unregister_memory_isolate_notifier);
 
 /*
  * register_memory - Setup a sysfs device for a memory block
  */
-int register_memory(struct memory_block *memory, struct mem_section *section,
-               struct node *root)
+static
+int register_memory(struct memory_block *memory)
 {
        int error;
 
-       memory->sysdev.cls = &memory_sysdev_class;
-       memory->sysdev.id = __section_nr(section);
-
-       error = sysdev_register(&memory->sysdev);
-
-       if (root && !error)
-               error = sysfs_create_link(&root->sysdev.kobj,
-                                         &memory->sysdev.kobj,
-                                         kobject_name(&memory->sysdev.kobj));
+       memory->dev.bus = &memory_subsys;
+       memory->dev.id = memory->start_section_nr / sections_per_block;
 
+       error = device_register(&memory->dev);
        return error;
 }
 
 static void
-unregister_memory(struct memory_block *memory, struct mem_section *section,
-               struct node *root)
+unregister_memory(struct memory_block *memory)
 {
-       BUG_ON(memory->sysdev.cls != &memory_sysdev_class);
-       BUG_ON(memory->sysdev.id != __section_nr(section));
+       BUG_ON(memory->dev.bus != &memory_subsys);
 
-       sysdev_unregister(&memory->sysdev);
-       if (root)
-               sysfs_remove_link(&root->sysdev.kobj,
-                                 kobject_name(&memory->sysdev.kobj));
+       /* drop the ref. we got in remove_memory_block() */
+       kobject_put(&memory->dev.kobj);
+       device_unregister(&memory->dev);
+}
+
+unsigned long __weak memory_block_size_bytes(void)
+{
+       return MIN_MEMORY_BLOCK_SIZE;
+}
+
+static unsigned long get_memory_block_size(void)
+{
+       unsigned long block_sz;
+
+       block_sz = memory_block_size_bytes();
+
+       /* Validate blk_sz is a power of 2 and not less than section size */
+       if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) {
+               WARN_ON(1);
+               block_sz = MIN_MEMORY_BLOCK_SIZE;
+       }
+
+       return block_sz;
 }
 
 /*
@@ -97,20 +120,55 @@ unregister_memory(struct memory_block *memory, struct mem_section *section,
  * uses.
  */
 
-static ssize_t show_mem_phys_index(struct sys_device *dev, char *buf)
+static ssize_t show_mem_start_phys_index(struct device *dev,
+                       struct device_attribute *attr, char *buf)
+{
+       struct memory_block *mem =
+               container_of(dev, struct memory_block, dev);
+       unsigned long phys_index;
+
+       phys_index = mem->start_section_nr / sections_per_block;
+       return sprintf(buf, "%08lx\n", phys_index);
+}
+
+static ssize_t show_mem_end_phys_index(struct device *dev,
+                       struct device_attribute *attr, char *buf)
 {
        struct memory_block *mem =
-               container_of(dev, struct memory_block, sysdev);
-       return sprintf(buf, "%08lx\n", mem->phys_index);
+               container_of(dev, struct memory_block, dev);
+       unsigned long phys_index;
+
+       phys_index = mem->end_section_nr / sections_per_block;
+       return sprintf(buf, "%08lx\n", phys_index);
+}
+
+/*
+ * Show whether the section of memory is likely to be hot-removable
+ */
+static ssize_t show_mem_removable(struct device *dev,
+                       struct device_attribute *attr, char *buf)
+{
+       unsigned long i, pfn;
+       int ret = 1;
+       struct memory_block *mem =
+               container_of(dev, struct memory_block, dev);
+
+       for (i = 0; i < sections_per_block; i++) {
+               pfn = section_nr_to_pfn(mem->start_section_nr + i);
+               ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION);
+       }
+
+       return sprintf(buf, "%d\n", ret);
 }
 
 /*
  * online, offline, going offline, etc.
  */
-static ssize_t show_mem_state(struct sys_device *dev, char *buf)
+static ssize_t show_mem_state(struct device *dev,
+                       struct device_attribute *attr, char *buf)
 {
        struct memory_block *mem =
-               container_of(dev, struct memory_block, sysdev);
+               container_of(dev, struct memory_block, dev);
        ssize_t len = 0;
 
        /*
@@ -137,73 +195,85 @@ static ssize_t show_mem_state(struct sys_device *dev, char *buf)
        return len;
 }
 
-static inline int memory_notify(unsigned long val, void *v)
+int memory_notify(unsigned long val, void *v)
 {
        return blocking_notifier_call_chain(&memory_chain, val, v);
 }
 
+int memory_isolate_notify(unsigned long val, void *v)
+{
+       return atomic_notifier_call_chain(&memory_isolate_chain, val, v);
+}
+
 /*
- * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
- * OK to have direct references to sparsemem variables in here.
+ * The probe routines leave the pages reserved, just as the bootmem code does.
+ * Make sure they're still that way.
  */
-static int
-memory_block_action(struct memory_block *mem, unsigned long action)
+static bool pages_correctly_reserved(unsigned long start_pfn,
+                                       unsigned long nr_pages)
 {
-       int i;
-       unsigned long psection;
-       unsigned long start_pfn, start_paddr;
-       struct page *first_page;
-       int ret;
-       int old_state = mem->state;
-
-       psection = mem->phys_index;
-       first_page = pfn_to_page(psection << PFN_SECTION_SHIFT);
+       int i, j;
+       struct page *page;
+       unsigned long pfn = start_pfn;
 
        /*
-        * The probe routines leave the pages reserved, just
-        * as the bootmem code does.  Make sure they're still
-        * that way.
+        * memmap between sections is not contiguous except with
+        * SPARSEMEM_VMEMMAP. We lookup the page once per section
+        * and assume memmap is contiguous within each section
         */
-       if (action == MEM_ONLINE) {
-               for (i = 0; i < PAGES_PER_SECTION; i++) {
-                       if (PageReserved(first_page+i))
+       for (i = 0; i < sections_per_block; i++, pfn += PAGES_PER_SECTION) {
+               if (WARN_ON_ONCE(!pfn_valid(pfn)))
+                       return false;
+               page = pfn_to_page(pfn);
+
+               for (j = 0; j < PAGES_PER_SECTION; j++) {
+                       if (PageReserved(page + j))
                                continue;
 
                        printk(KERN_WARNING "section number %ld page number %d "
-                               "not reserved, was it already online? \n",
-                               psection, i);
-                       return -EBUSY;
+                               "not reserved, was it already online?\n",
+                               pfn_to_section_nr(pfn), j);
+
+                       return false;
                }
        }
 
+       return true;
+}
+
+/*
+ * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
+ * OK to have direct references to sparsemem variables in here.
+ */
+static int
+memory_block_action(unsigned long phys_index, unsigned long action)
+{
+       unsigned long start_pfn, start_paddr;
+       unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
+       struct page *first_page;
+       int ret;
+
+       first_page = pfn_to_page(phys_index << PFN_SECTION_SHIFT);
+
        switch (action) {
                case MEM_ONLINE:
                        start_pfn = page_to_pfn(first_page);
-                       ret = online_pages(start_pfn, PAGES_PER_SECTION);
+
+                       if (!pages_correctly_reserved(start_pfn, nr_pages))
+                               return -EBUSY;
+
+                       ret = online_pages(start_pfn, nr_pages);
                        break;
                case MEM_OFFLINE:
-                       mem->state = MEM_GOING_OFFLINE;
-                       memory_notify(MEM_GOING_OFFLINE, NULL);
                        start_paddr = page_to_pfn(first_page) << PAGE_SHIFT;
                        ret = remove_memory(start_paddr,
-                                           PAGES_PER_SECTION << PAGE_SHIFT);
-                       if (ret) {
-                               mem->state = old_state;
-                               break;
-                       }
-                       memory_notify(MEM_MAPPING_INVALID, NULL);
+                                           nr_pages << PAGE_SHIFT);
                        break;
                default:
-                       printk(KERN_WARNING "%s(%p, %ld) unknown action: %ld\n",
-                                       __FUNCTION__, mem, action, action);
-                       WARN_ON(1);
+                       WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
+                            "%ld\n", __func__, phys_index, action, action);
                        ret = -EINVAL;
        }
-       /*
-        * For now, only notify on successful memory operations
-        */
-       if (!ret)
-               memory_notify(action, NULL);
 
        return ret;
 }
@@ -212,40 +282,54 @@ static int memory_block_change_state(struct memory_block *mem,
                unsigned long to_state, unsigned long from_state_req)
 {
        int ret = 0;
-       down(&mem->state_sem);
+
+       mutex_lock(&mem->state_mutex);
 
        if (mem->state != from_state_req) {
                ret = -EINVAL;
                goto out;
        }
 
-       ret = memory_block_action(mem, to_state);
-       if (!ret)
-               mem->state = to_state;
+       if (to_state == MEM_OFFLINE)
+               mem->state = MEM_GOING_OFFLINE;
+
+       ret = memory_block_action(mem->start_section_nr, to_state);
 
+       if (ret) {
+               mem->state = from_state_req;
+               goto out;
+       }
+
+       mem->state = to_state;
+       switch (mem->state) {
+       case MEM_OFFLINE:
+               kobject_uevent(&mem->dev.kobj, KOBJ_OFFLINE);
+               break;
+       case MEM_ONLINE:
+               kobject_uevent(&mem->dev.kobj, KOBJ_ONLINE);
+               break;
+       default:
+               break;
+       }
 out:
-       up(&mem->state_sem);
+       mutex_unlock(&mem->state_mutex);
        return ret;
 }
 
 static ssize_t
-store_mem_state(struct sys_device *dev, const char *buf, size_t count)
+store_mem_state(struct device *dev,
+               struct device_attribute *attr, const char *buf, size_t count)
 {
        struct memory_block *mem;
-       unsigned int phys_section_nr;
        int ret = -EINVAL;
 
-       mem = container_of(dev, struct memory_block, sysdev);
-       phys_section_nr = mem->phys_index;
-
-       if (!valid_section_nr(phys_section_nr))
-               goto out;
+       mem = container_of(dev, struct memory_block, dev);
 
        if (!strncmp(buf, "online", min((int)count, 6)))
                ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
        else if(!strncmp(buf, "offline", min((int)count, 7)))
                ret = memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
-out:
+
        if (ret)
                return ret;
        return count;
@@ -260,37 +344,41 @@ out:
  * s.t. if I offline all of these sections I can then
  * remove the physical device?
  */
-static ssize_t show_phys_device(struct sys_device *dev, char *buf)
+static ssize_t show_phys_device(struct device *dev,
+                               struct device_attribute *attr, char *buf)
 {
        struct memory_block *mem =
-               container_of(dev, struct memory_block, sysdev);
+               container_of(dev, struct memory_block, dev);
        return sprintf(buf, "%d\n", mem->phys_device);
 }
 
-static SYSDEV_ATTR(phys_index, 0444, show_mem_phys_index, NULL);
-static SYSDEV_ATTR(state, 0644, show_mem_state, store_mem_state);
-static SYSDEV_ATTR(phys_device, 0444, show_phys_device, NULL);
+static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL);
+static DEVICE_ATTR(end_phys_index, 0444, show_mem_end_phys_index, NULL);
+static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state);
+static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL);
+static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL);
 
 #define mem_create_simple_file(mem, attr_name) \
-       sysdev_create_file(&mem->sysdev, &attr_##attr_name)
+       device_create_file(&mem->dev, &dev_attr_##attr_name)
 #define mem_remove_simple_file(mem, attr_name) \
-       sysdev_remove_file(&mem->sysdev, &attr_##attr_name)
+       device_remove_file(&mem->dev, &dev_attr_##attr_name)
 
 /*
  * Block size attribute stuff
  */
 static ssize_t
-print_block_size(struct class *class, char *buf)
+print_block_size(struct device *dev, struct device_attribute *attr,
+                char *buf)
 {
-       return sprintf(buf, "%lx\n", (unsigned long)PAGES_PER_SECTION * PAGE_SIZE);
+       return sprintf(buf, "%lx\n", get_memory_block_size());
 }
 
-static CLASS_ATTR(block_size_bytes, 0444, print_block_size, NULL);
+static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL);
 
 static int block_size_init(void)
 {
-       return sysfs_create_file(&memory_sysdev_class.kset.kobj,
-                               &class_attr_block_size_bytes.attr);
+       return device_create_file(memory_subsys.dev_root,
+                                 &dev_attr_block_size_bytes);
 }
 
 /*
@@ -301,28 +389,38 @@ static int block_size_init(void)
  */
 #ifdef CONFIG_ARCH_MEMORY_PROBE
 static ssize_t
-memory_probe_store(struct class *class, const char *buf, size_t count)
+memory_probe_store(struct device *dev, struct device_attribute *attr,
+                  const char *buf, size_t count)
 {
        u64 phys_addr;
        int nid;
-       int ret;
+       int i, ret;
+       unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
 
        phys_addr = simple_strtoull(buf, NULL, 0);
 
-       nid = memory_add_physaddr_to_nid(phys_addr);
-       ret = add_memory(nid, phys_addr, PAGES_PER_SECTION << PAGE_SHIFT);
+       if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
+               return -EINVAL;
 
-       if (ret)
-               count = ret;
+       for (i = 0; i < sections_per_block; i++) {
+               nid = memory_add_physaddr_to_nid(phys_addr);
+               ret = add_memory(nid, phys_addr,
+                                PAGES_PER_SECTION << PAGE_SHIFT);
+               if (ret)
+                       goto out;
 
-       return count;
+               phys_addr += MIN_MEMORY_BLOCK_SIZE;
+       }
+
+       ret = count;
+out:
+       return ret;
 }
-static CLASS_ATTR(probe, 0700, NULL, memory_probe_store);
+static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store);
 
 static int memory_probe_init(void)
 {
-       return sysfs_create_file(&memory_sysdev_class.kset.kobj,
-                               &class_attr_probe.attr);
+       return device_create_file(memory_subsys.dev_root, &dev_attr_probe);
 }
 #else
 static inline int memory_probe_init(void)
@@ -331,66 +429,188 @@ static inline int memory_probe_init(void)
 }
 #endif
 
+#ifdef CONFIG_MEMORY_FAILURE
+/*
+ * Support for offlining pages of memory
+ */
+
+/* Soft offline a page */
+static ssize_t
+store_soft_offline_page(struct device *dev,
+                       struct device_attribute *attr,
+                       const char *buf, size_t count)
+{
+       int ret;
+       u64 pfn;
+       if (!capable(CAP_SYS_ADMIN))
+               return -EPERM;
+       if (strict_strtoull(buf, 0, &pfn) < 0)
+               return -EINVAL;
+       pfn >>= PAGE_SHIFT;
+       if (!pfn_valid(pfn))
+               return -ENXIO;
+       ret = soft_offline_page(pfn_to_page(pfn), 0);
+       return ret == 0 ? count : ret;
+}
+
+/* Forcibly offline a page, including killing processes. */
+static ssize_t
+store_hard_offline_page(struct device *dev,
+                       struct device_attribute *attr,
+                       const char *buf, size_t count)
+{
+       int ret;
+       u64 pfn;
+       if (!capable(CAP_SYS_ADMIN))
+               return -EPERM;
+       if (strict_strtoull(buf, 0, &pfn) < 0)
+               return -EINVAL;
+       pfn >>= PAGE_SHIFT;
+       ret = memory_failure(pfn, 0, 0);
+       return ret ? ret : count;
+}
+
+static DEVICE_ATTR(soft_offline_page, 0644, NULL, store_soft_offline_page);
+static DEVICE_ATTR(hard_offline_page, 0644, NULL, store_hard_offline_page);
+
+static __init int memory_fail_init(void)
+{
+       int err;
+
+       err = device_create_file(memory_subsys.dev_root,
+                               &dev_attr_soft_offline_page);
+       if (!err)
+               err = device_create_file(memory_subsys.dev_root,
+                               &dev_attr_hard_offline_page);
+       return err;
+}
+#else
+static inline int memory_fail_init(void)
+{
+       return 0;
+}
+#endif
+
 /*
  * Note that phys_device is optional.  It is here to allow for
  * differentiation between which *physical* devices each
  * section belongs to...
  */
+int __weak arch_get_memory_phys_device(unsigned long start_pfn)
+{
+       return 0;
+}
+
+/*
+ * A reference for the returned object is held and the reference for the
+ * hinted object is released.
+ */
+struct memory_block *find_memory_block_hinted(struct mem_section *section,
+                                             struct memory_block *hint)
+{
+       int block_id = base_memory_block_id(__section_nr(section));
+       struct device *hintdev = hint ? &hint->dev : NULL;
+       struct device *dev;
+
+       dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev);
+       if (hint)
+               put_device(&hint->dev);
+       if (!dev)
+               return NULL;
+       return container_of(dev, struct memory_block, dev);
+}
+
+/*
+ * For now, we have a linear search to go find the appropriate
+ * memory_block corresponding to a particular phys_index. If
+ * this gets to be a real problem, we can always use a radix
+ * tree or something here.
+ *
+ * This could be made generic for all device subsystems.
+ */
+struct memory_block *find_memory_block(struct mem_section *section)
+{
+       return find_memory_block_hinted(section, NULL);
+}
 
-static int add_memory_block(unsigned long node_id, struct mem_section *section,
-                    unsigned long state, int phys_device)
+static int init_memory_block(struct memory_block **memory,
+                            struct mem_section *section, unsigned long state)
 {
-       struct memory_block *mem = kzalloc(sizeof(*mem), GFP_KERNEL);
+       struct memory_block *mem;
+       unsigned long start_pfn;
+       int scn_nr;
        int ret = 0;
 
+       mem = kzalloc(sizeof(*mem), GFP_KERNEL);
        if (!mem)
                return -ENOMEM;
 
-       mem->phys_index = __section_nr(section);
+       scn_nr = __section_nr(section);
+       mem->start_section_nr =
+                       base_memory_block_id(scn_nr) * sections_per_block;
+       mem->end_section_nr = mem->start_section_nr + sections_per_block - 1;
        mem->state = state;
-       init_MUTEX(&mem->state_sem);
-       mem->phys_device = phys_device;
+       mem->section_count++;
+       mutex_init(&mem->state_mutex);
+       start_pfn = section_nr_to_pfn(mem->start_section_nr);
+       mem->phys_device = arch_get_memory_phys_device(start_pfn);
 
-       ret = register_memory(mem, section, NULL);
+       ret = register_memory(mem);
        if (!ret)
                ret = mem_create_simple_file(mem, phys_index);
+       if (!ret)
+               ret = mem_create_simple_file(mem, end_phys_index);
        if (!ret)
                ret = mem_create_simple_file(mem, state);
        if (!ret)
                ret = mem_create_simple_file(mem, phys_device);
+       if (!ret)
+               ret = mem_create_simple_file(mem, removable);
 
+       *memory = mem;
        return ret;
 }
 
-/*
- * For now, we have a linear search to go find the appropriate
- * memory_block corresponding to a particular phys_index. If
- * this gets to be a real problem, we can always use a radix
- * tree or something here.
- *
- * This could be made generic for all sysdev classes.
- */
-static struct memory_block *find_memory_block(struct mem_section *section)
+static int add_memory_section(int nid, struct mem_section *section,
+                       struct memory_block **mem_p,
+                       unsigned long state, enum mem_add_context context)
 {
-       struct kobject *kobj;
-       struct sys_device *sysdev;
-       struct memory_block *mem;
-       char name[sizeof(MEMORY_CLASS_NAME) + 9 + 1];
+       struct memory_block *mem = NULL;
+       int scn_nr = __section_nr(section);
+       int ret = 0;
 
-       /*
-        * This only works because we know that section == sysdev->id
-        * slightly redundant with sysdev_register()
-        */
-       sprintf(&name[0], "%s%d", MEMORY_CLASS_NAME, __section_nr(section));
+       mutex_lock(&mem_sysfs_mutex);
 
-       kobj = kset_find_obj(&memory_sysdev_class.kset, name);
-       if (!kobj)
-               return NULL;
+       if (context == BOOT) {
+               /* same memory block ? */
+               if (mem_p && *mem_p)
+                       if (scn_nr >= (*mem_p)->start_section_nr &&
+                           scn_nr <= (*mem_p)->end_section_nr) {
+                               mem = *mem_p;
+                               kobject_get(&mem->dev.kobj);
+                       }
+       } else
+               mem = find_memory_block(section);
+
+       if (mem) {
+               mem->section_count++;
+               kobject_put(&mem->dev.kobj);
+       } else {
+               ret = init_memory_block(&mem, section, state);
+               /* store memory_block pointer for next loop */
+               if (!ret && context == BOOT)
+                       if (mem_p)
+                               *mem_p = mem;
+       }
 
-       sysdev = container_of(kobj, struct sys_device, kobj);
-       mem = container_of(sysdev, struct memory_block, sysdev);
+       if (!ret) {
+               if (context == HOTPLUG &&
+                   mem->section_count == sections_per_block)
+                       ret = register_mem_sect_under_node(mem, nid);
+       }
 
-       return mem;
+       mutex_unlock(&mem_sysfs_mutex);
+       return ret;
 }
 
 int remove_memory_block(unsigned long node_id, struct mem_section *section,
@@ -398,12 +618,23 @@ int remove_memory_block(unsigned long node_id, struct mem_section *section,
 {
        struct memory_block *mem;
 
+       mutex_lock(&mem_sysfs_mutex);
        mem = find_memory_block(section);
-       mem_remove_simple_file(mem, phys_index);
-       mem_remove_simple_file(mem, state);
-       mem_remove_simple_file(mem, phys_device);
-       unregister_memory(mem, section, NULL);
-
+       unregister_mem_sect_under_nodes(mem, __section_nr(section));
+
+       mem->section_count--;
+       if (mem->section_count == 0) {
+               mem_remove_simple_file(mem, phys_index);
+               mem_remove_simple_file(mem, end_phys_index);
+               mem_remove_simple_file(mem, state);
+               mem_remove_simple_file(mem, phys_device);
+               mem_remove_simple_file(mem, removable);
+               unregister_memory(mem);
+               kfree(mem);
+       } else
+               kobject_put(&mem->dev.kobj);
+
+       mutex_unlock(&mem_sysfs_mutex);
        return 0;
 }
 
@@ -411,14 +642,14 @@ int remove_memory_block(unsigned long node_id, struct mem_section *section,
  * need an interface for the VM to add new memory regions,
  * but without onlining it.
  */
-int register_new_memory(struct mem_section *section)
+int register_new_memory(int nid, struct mem_section *section)
 {
-       return add_memory_block(0, section, MEM_OFFLINE, 0);
+       return add_memory_section(nid, section, NULL, MEM_OFFLINE, HOTPLUG);
 }
 
 int unregister_memory_section(struct mem_section *section)
 {
-       if (!valid_section(section))
+       if (!present_section(section))
                return -EINVAL;
 
        return remove_memory_block(0, section, 0);
@@ -432,25 +663,36 @@ int __init memory_dev_init(void)
        unsigned int i;
        int ret;
        int err;
+       unsigned long block_sz;
+       struct memory_block *mem = NULL;
 
-       memory_sysdev_class.kset.uevent_ops = &memory_uevent_ops;
-       ret = sysdev_class_register(&memory_sysdev_class);
+       ret = subsys_system_register(&memory_subsys, NULL);
        if (ret)
                goto out;
 
+       block_sz = get_memory_block_size();
+       sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
+
        /*
         * Create entries for memory sections that were found
         * during boot and have been initialized
         */
        for (i = 0; i < NR_MEM_SECTIONS; i++) {
-               if (!valid_section_nr(i))
+               if (!present_section_nr(i))
                        continue;
-               err = add_memory_block(0, __nr_to_section(i), MEM_ONLINE, 0);
+               /* don't need to reuse memory_block if only one per block */
+               err = add_memory_section(0, __nr_to_section(i),
+                                (sections_per_block == 1) ? NULL : &mem,
+                                        MEM_ONLINE,
+                                        BOOT);
                if (!ret)
                        ret = err;
        }
 
        err = memory_probe_init();
+       if (!ret)
+               ret = err;
+       err = memory_fail_init();
        if (!ret)
                ret = err;
        err = block_size_init();
@@ -458,6 +700,6 @@ int __init memory_dev_init(void)
                ret = err;
 out:
        if (ret)
-               printk(KERN_ERR "%s() failed: %d\n", __FUNCTION__, ret);
+               printk(KERN_ERR "%s() failed: %d\n", __func__, ret);
        return ret;
 }