Merge branch 'master' into for-linus
[safe/jmp/linux-2.6] / drivers / base / node.c
index 39f3d1b..91d4087 100644 (file)
@@ -6,6 +6,7 @@
 #include <linux/module.h>
 #include <linux/init.h>
 #include <linux/mm.h>
+#include <linux/memory.h>
 #include <linux/node.h>
 #include <linux/hugetlb.h>
 #include <linux/cpumask.h>
@@ -13,6 +14,7 @@
 #include <linux/nodemask.h>
 #include <linux/cpu.h>
 #include <linux/device.h>
+#include <linux/swap.h>
 
 static struct sysdev_class node_class = {
        .name = "node",
@@ -22,25 +24,27 @@ static struct sysdev_class node_class = {
 static ssize_t node_read_cpumap(struct sys_device *dev, int type, char *buf)
 {
        struct node *node_dev = to_node(dev);
-       node_to_cpumask_ptr(mask, node_dev->sysdev.id);
+       const struct cpumask *mask = cpumask_of_node(node_dev->sysdev.id);
        int len;
 
        /* 2008/04/07: buf currently PAGE_SIZE, need 9 chars per 32 bits. */
        BUILD_BUG_ON((NR_CPUS/32 * 9) > (PAGE_SIZE-1));
 
        len = type?
-               cpulist_scnprintf(buf, PAGE_SIZE-2, *mask):
-               cpumask_scnprintf(buf, PAGE_SIZE-2, *mask);
+               cpulist_scnprintf(buf, PAGE_SIZE-2, mask) :
+               cpumask_scnprintf(buf, PAGE_SIZE-2, mask);
        buf[len++] = '\n';
        buf[len] = '\0';
        return len;
 }
 
-static inline ssize_t node_read_cpumask(struct sys_device *dev, char *buf)
+static inline ssize_t node_read_cpumask(struct sys_device *dev,
+                               struct sysdev_attribute *attr, char *buf)
 {
        return node_read_cpumap(dev, 0, buf);
 }
-static inline ssize_t node_read_cpulist(struct sys_device *dev, char *buf)
+static inline ssize_t node_read_cpulist(struct sys_device *dev,
+                               struct sysdev_attribute *attr, char *buf)
 {
        return node_read_cpumap(dev, 1, buf);
 }
@@ -49,7 +53,8 @@ static SYSDEV_ATTR(cpumap,  S_IRUGO, node_read_cpumask, NULL);
 static SYSDEV_ATTR(cpulist, S_IRUGO, node_read_cpulist, NULL);
 
 #define K(x) ((x) << (PAGE_SHIFT - 10))
-static ssize_t node_read_meminfo(struct sys_device * dev, char * buf)
+static ssize_t node_read_meminfo(struct sys_device * dev,
+                       struct sysdev_attribute *attr, char * buf)
 {
        int n;
        int nid = dev->id;
@@ -58,34 +63,48 @@ static ssize_t node_read_meminfo(struct sys_device * dev, char * buf)
        si_meminfo_node(&i, nid);
 
        n = sprintf(buf, "\n"
-                      "Node %d MemTotal:     %8lu kB\n"
-                      "Node %d MemFree:      %8lu kB\n"
-                      "Node %d MemUsed:      %8lu kB\n"
-                      "Node %d Active:       %8lu kB\n"
-                      "Node %d Inactive:     %8lu kB\n"
+                      "Node %d MemTotal:       %8lu kB\n"
+                      "Node %d MemFree:        %8lu kB\n"
+                      "Node %d MemUsed:        %8lu kB\n"
+                      "Node %d Active:         %8lu kB\n"
+                      "Node %d Inactive:       %8lu kB\n"
+                      "Node %d Active(anon):   %8lu kB\n"
+                      "Node %d Inactive(anon): %8lu kB\n"
+                      "Node %d Active(file):   %8lu kB\n"
+                      "Node %d Inactive(file): %8lu kB\n"
+                      "Node %d Unevictable:    %8lu kB\n"
+                      "Node %d Mlocked:        %8lu kB\n"
 #ifdef CONFIG_HIGHMEM
-                      "Node %d HighTotal:    %8lu kB\n"
-                      "Node %d HighFree:     %8lu kB\n"
-                      "Node %d LowTotal:     %8lu kB\n"
-                      "Node %d LowFree:      %8lu kB\n"
+                      "Node %d HighTotal:      %8lu kB\n"
+                      "Node %d HighFree:       %8lu kB\n"
+                      "Node %d LowTotal:       %8lu kB\n"
+                      "Node %d LowFree:        %8lu kB\n"
 #endif
-                      "Node %d Dirty:        %8lu kB\n"
-                      "Node %d Writeback:    %8lu kB\n"
-                      "Node %d FilePages:    %8lu kB\n"
-                      "Node %d Mapped:       %8lu kB\n"
-                      "Node %d AnonPages:    %8lu kB\n"
-                      "Node %d PageTables:   %8lu kB\n"
-                      "Node %d NFS_Unstable: %8lu kB\n"
-                      "Node %d Bounce:       %8lu kB\n"
-                      "Node %d WritebackTmp: %8lu kB\n"
-                      "Node %d Slab:         %8lu kB\n"
-                      "Node %d SReclaimable: %8lu kB\n"
-                      "Node %d SUnreclaim:   %8lu kB\n",
+                      "Node %d Dirty:          %8lu kB\n"
+                      "Node %d Writeback:      %8lu kB\n"
+                      "Node %d FilePages:      %8lu kB\n"
+                      "Node %d Mapped:         %8lu kB\n"
+                      "Node %d AnonPages:      %8lu kB\n"
+                      "Node %d PageTables:     %8lu kB\n"
+                      "Node %d NFS_Unstable:   %8lu kB\n"
+                      "Node %d Bounce:         %8lu kB\n"
+                      "Node %d WritebackTmp:   %8lu kB\n"
+                      "Node %d Slab:           %8lu kB\n"
+                      "Node %d SReclaimable:   %8lu kB\n"
+                      "Node %d SUnreclaim:     %8lu kB\n",
                       nid, K(i.totalram),
                       nid, K(i.freeram),
                       nid, K(i.totalram - i.freeram),
-                      nid, node_page_state(nid, NR_ACTIVE),
-                      nid, node_page_state(nid, NR_INACTIVE),
+                      nid, K(node_page_state(nid, NR_ACTIVE_ANON) +
+                               node_page_state(nid, NR_ACTIVE_FILE)),
+                      nid, K(node_page_state(nid, NR_INACTIVE_ANON) +
+                               node_page_state(nid, NR_INACTIVE_FILE)),
+                      nid, K(node_page_state(nid, NR_ACTIVE_ANON)),
+                      nid, K(node_page_state(nid, NR_INACTIVE_ANON)),
+                      nid, K(node_page_state(nid, NR_ACTIVE_FILE)),
+                      nid, K(node_page_state(nid, NR_INACTIVE_FILE)),
+                      nid, K(node_page_state(nid, NR_UNEVICTABLE)),
+                      nid, K(node_page_state(nid, NR_MLOCK)),
 #ifdef CONFIG_HIGHMEM
                       nid, K(i.totalhigh),
                       nid, K(i.freehigh),
@@ -112,7 +131,8 @@ static ssize_t node_read_meminfo(struct sys_device * dev, char * buf)
 #undef K
 static SYSDEV_ATTR(meminfo, S_IRUGO, node_read_meminfo, NULL);
 
-static ssize_t node_read_numastat(struct sys_device * dev, char * buf)
+static ssize_t node_read_numastat(struct sys_device * dev,
+                               struct sysdev_attribute *attr, char * buf)
 {
        return sprintf(buf,
                       "numa_hit %lu\n"
@@ -130,7 +150,8 @@ static ssize_t node_read_numastat(struct sys_device * dev, char * buf)
 }
 static SYSDEV_ATTR(numastat, S_IRUGO, node_read_numastat, NULL);
 
-static ssize_t node_read_distance(struct sys_device * dev, char * buf)
+static ssize_t node_read_distance(struct sys_device * dev,
+                       struct sysdev_attribute *attr, char * buf)
 {
        int nid = dev->id;
        int len = 0;
@@ -168,6 +189,8 @@ int register_node(struct node *node, int num, struct node *parent)
                sysdev_create_file(&node->sysdev, &attr_meminfo);
                sysdev_create_file(&node->sysdev, &attr_numastat);
                sysdev_create_file(&node->sysdev, &attr_distance);
+
+               scan_unevictable_register_node(node);
        }
        return error;
 }
@@ -187,6 +210,8 @@ void unregister_node(struct node *node)
        sysdev_remove_file(&node->sysdev, &attr_numastat);
        sysdev_remove_file(&node->sysdev, &attr_distance);
 
+       scan_unevictable_unregister_node(node);
+
        sysdev_unregister(&node->sysdev);
 }
 
@@ -220,6 +245,105 @@ int unregister_cpu_under_node(unsigned int cpu, unsigned int nid)
        return 0;
 }
 
+#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
+#define page_initialized(page)  (page->lru.next)
+
+static int get_nid_for_pfn(unsigned long pfn)
+{
+       struct page *page;
+
+       if (!pfn_valid_within(pfn))
+               return -1;
+       page = pfn_to_page(pfn);
+       if (!page_initialized(page))
+               return -1;
+       return pfn_to_nid(pfn);
+}
+
+/* register memory section under specified node if it spans that node */
+int register_mem_sect_under_node(struct memory_block *mem_blk, int nid)
+{
+       unsigned long pfn, sect_start_pfn, sect_end_pfn;
+
+       if (!mem_blk)
+               return -EFAULT;
+       if (!node_online(nid))
+               return 0;
+       sect_start_pfn = section_nr_to_pfn(mem_blk->phys_index);
+       sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
+       for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
+               int page_nid;
+
+               page_nid = get_nid_for_pfn(pfn);
+               if (page_nid < 0)
+                       continue;
+               if (page_nid != nid)
+                       continue;
+               return sysfs_create_link_nowarn(&node_devices[nid].sysdev.kobj,
+                                       &mem_blk->sysdev.kobj,
+                                       kobject_name(&mem_blk->sysdev.kobj));
+       }
+       /* mem section does not span the specified node */
+       return 0;
+}
+
+/* unregister memory section under all nodes that it spans */
+int unregister_mem_sect_under_nodes(struct memory_block *mem_blk)
+{
+       nodemask_t unlinked_nodes;
+       unsigned long pfn, sect_start_pfn, sect_end_pfn;
+
+       if (!mem_blk)
+               return -EFAULT;
+       nodes_clear(unlinked_nodes);
+       sect_start_pfn = section_nr_to_pfn(mem_blk->phys_index);
+       sect_end_pfn = sect_start_pfn + PAGES_PER_SECTION - 1;
+       for (pfn = sect_start_pfn; pfn <= sect_end_pfn; pfn++) {
+               int nid;
+
+               nid = get_nid_for_pfn(pfn);
+               if (nid < 0)
+                       continue;
+               if (!node_online(nid))
+                       continue;
+               if (node_test_and_set(nid, unlinked_nodes))
+                       continue;
+               sysfs_remove_link(&node_devices[nid].sysdev.kobj,
+                        kobject_name(&mem_blk->sysdev.kobj));
+       }
+       return 0;
+}
+
+static int link_mem_sections(int nid)
+{
+       unsigned long start_pfn = NODE_DATA(nid)->node_start_pfn;
+       unsigned long end_pfn = start_pfn + NODE_DATA(nid)->node_spanned_pages;
+       unsigned long pfn;
+       int err = 0;
+
+       for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
+               unsigned long section_nr = pfn_to_section_nr(pfn);
+               struct mem_section *mem_sect;
+               struct memory_block *mem_blk;
+               int ret;
+
+               if (!present_section_nr(section_nr))
+                       continue;
+               mem_sect = __nr_to_section(section_nr);
+               mem_blk = find_memory_block(mem_sect);
+               ret = register_mem_sect_under_node(mem_blk, nid);
+               if (!err)
+                       err = ret;
+
+               /* discard ref obtained in find_memory_block() */
+               kobject_put(&mem_blk->sysdev.kobj);
+       }
+       return err;
+}
+#else
+static int link_mem_sections(int nid) { return 0; }
+#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
+
 int register_one_node(int nid)
 {
        int error = 0;
@@ -239,6 +363,9 @@ int register_one_node(int nid)
                        if (cpu_to_node(cpu) == nid)
                                register_cpu_under_node(cpu, nid);
                }
+
+               /* link memory sections under this node */
+               error = link_mem_sections(nid);
        }
 
        return error;