mm: cleanup descriptions of filler arg
[linux-2.6.git] / mm / sparse.c
index 5d9dbbb..858e1df 100644 (file)
@@ -2,6 +2,7 @@
  * sparse memory mappings.
  */
 #include <linux/mm.h>
+#include <linux/slab.h>
 #include <linux/mmzone.h>
 #include <linux/bootmem.h>
 #include <linux/highmem.h>
@@ -12,7 +13,6 @@
 #include <asm/dma.h>
 #include <asm/pgalloc.h>
 #include <asm/pgtable.h>
-#include "internal.h"
 
 /*
  * Permanent SPARSEMEM data:
@@ -40,7 +40,7 @@ static u8 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned;
 static u16 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned;
 #endif
 
-int page_to_nid(struct page *page)
+int page_to_nid(const struct page *page)
 {
        return section_to_node_table[page_to_section(page)];
 }
@@ -63,9 +63,12 @@ static struct mem_section noinline __init_refok *sparse_index_alloc(int nid)
        unsigned long array_size = SECTIONS_PER_ROOT *
                                   sizeof(struct mem_section);
 
-       if (slab_is_available())
-               section = kmalloc_node(array_size, GFP_KERNEL, nid);
-       else
+       if (slab_is_available()) {
+               if (node_state(nid, N_HIGH_MEMORY))
+                       section = kmalloc_node(array_size, GFP_KERNEL, nid);
+               else
+                       section = kmalloc(array_size, GFP_KERNEL);
+       } else
                section = alloc_bootmem_node(NODE_DATA(nid), array_size);
 
        if (section)
@@ -165,9 +168,7 @@ void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn,
                WARN_ON_ONCE(1);
                *start_pfn = max_sparsemem_pfn;
                *end_pfn = max_sparsemem_pfn;
-       }
-
-       if (*end_pfn > max_sparsemem_pfn) {
+       } else if (*end_pfn > max_sparsemem_pfn) {
                mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
                        "End of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
                        *start_pfn, *end_pfn, max_sparsemem_pfn);
@@ -271,7 +272,8 @@ static unsigned long *__kmalloc_section_usemap(void)
 
 #ifdef CONFIG_MEMORY_HOTREMOVE
 static unsigned long * __init
-sparse_early_usemap_alloc_pgdat_section(struct pglist_data *pgdat)
+sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
+                                        unsigned long count)
 {
        unsigned long section_nr;
 
@@ -286,7 +288,7 @@ sparse_early_usemap_alloc_pgdat_section(struct pglist_data *pgdat)
         * this problem.
         */
        section_nr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT);
-       return alloc_bootmem_section(usemap_size(), section_nr);
+       return alloc_bootmem_section(usemap_size() * count, section_nr);
 }
 
 static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
@@ -329,7 +331,8 @@ static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
 }
 #else
 static unsigned long * __init
-sparse_early_usemap_alloc_pgdat_section(struct pglist_data *pgdat)
+sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
+                                        unsigned long count)
 {
        return NULL;
 }
@@ -339,44 +342,117 @@ static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
 }
 #endif /* CONFIG_MEMORY_HOTREMOVE */
 
-static unsigned long *__init sparse_early_usemap_alloc(unsigned long pnum)
+static void __init sparse_early_usemaps_alloc_node(unsigned long**usemap_map,
+                                unsigned long pnum_begin,
+                                unsigned long pnum_end,
+                                unsigned long usemap_count, int nodeid)
 {
-       unsigned long *usemap;
-       struct mem_section *ms = __nr_to_section(pnum);
-       int nid = sparse_early_nid(ms);
-
-       usemap = sparse_early_usemap_alloc_pgdat_section(NODE_DATA(nid));
-       if (usemap)
-               return usemap;
+       void *usemap;
+       unsigned long pnum;
+       int size = usemap_size();
 
-       usemap = alloc_bootmem_node(NODE_DATA(nid), usemap_size());
+       usemap = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nodeid),
+                                                                usemap_count);
        if (usemap) {
-               check_usemap_section_nr(nid, usemap);
-               return usemap;
+               for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
+                       if (!present_section_nr(pnum))
+                               continue;
+                       usemap_map[pnum] = usemap;
+                       usemap += size;
+               }
+               return;
        }
 
-       /* Stupid: suppress gcc warning for SPARSEMEM && !NUMA */
-       nid = 0;
+       usemap = alloc_bootmem_node(NODE_DATA(nodeid), size * usemap_count);
+       if (usemap) {
+               for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
+                       if (!present_section_nr(pnum))
+                               continue;
+                       usemap_map[pnum] = usemap;
+                       usemap += size;
+                       check_usemap_section_nr(nodeid, usemap_map[pnum]);
+               }
+               return;
+       }
 
        printk(KERN_WARNING "%s: allocation failed\n", __func__);
-       return NULL;
 }
 
 #ifndef CONFIG_SPARSEMEM_VMEMMAP
 struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid)
 {
        struct page *map;
+       unsigned long size;
 
        map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION);
        if (map)
                return map;
 
-       map = alloc_bootmem_pages_node(NODE_DATA(nid),
-                      PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION));
+       size = PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION);
+       map = __alloc_bootmem_node_high(NODE_DATA(nid), size,
+                                        PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
        return map;
 }
+void __init sparse_mem_maps_populate_node(struct page **map_map,
+                                         unsigned long pnum_begin,
+                                         unsigned long pnum_end,
+                                         unsigned long map_count, int nodeid)
+{
+       void *map;
+       unsigned long pnum;
+       unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
+
+       map = alloc_remap(nodeid, size * map_count);
+       if (map) {
+               for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
+                       if (!present_section_nr(pnum))
+                               continue;
+                       map_map[pnum] = map;
+                       map += size;
+               }
+               return;
+       }
+
+       size = PAGE_ALIGN(size);
+       map = __alloc_bootmem_node_high(NODE_DATA(nodeid), size * map_count,
+                                        PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
+       if (map) {
+               for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
+                       if (!present_section_nr(pnum))
+                               continue;
+                       map_map[pnum] = map;
+                       map += size;
+               }
+               return;
+       }
+
+       /* fallback */
+       for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
+               struct mem_section *ms;
+
+               if (!present_section_nr(pnum))
+                       continue;
+               map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
+               if (map_map[pnum])
+                       continue;
+               ms = __nr_to_section(pnum);
+               printk(KERN_ERR "%s: sparsemem memory map backing failed "
+                       "some memory will not be available.\n", __func__);
+               ms->section_mem_map = 0;
+       }
+}
 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
 
+#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
+static void __init sparse_early_mem_maps_alloc_node(struct page **map_map,
+                                unsigned long pnum_begin,
+                                unsigned long pnum_end,
+                                unsigned long map_count, int nodeid)
+{
+       sparse_mem_maps_populate_node(map_map, pnum_begin, pnum_end,
+                                        map_count, nodeid);
+}
+#else
 static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum)
 {
        struct page *map;
@@ -392,10 +468,12 @@ static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum)
        ms->section_mem_map = 0;
        return NULL;
 }
+#endif
 
 void __attribute__((weak)) __meminit vmemmap_populate_print_last(void)
 {
 }
+
 /*
  * Allocate the accumulated non-linear sections, allocate a mem_map
  * for each and record the physical to section mapping.
@@ -407,6 +485,14 @@ void __init sparse_init(void)
        unsigned long *usemap;
        unsigned long **usemap_map;
        int size;
+       int nodeid_begin = 0;
+       unsigned long pnum_begin = 0;
+       unsigned long usemap_count;
+#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
+       unsigned long map_count;
+       int size2;
+       struct page **map_map;
+#endif
 
        /*
         * map is using big page (aka 2M in x86 64 bit)
@@ -414,7 +500,7 @@ void __init sparse_init(void)
         * so alloc 2M (with 2M align) and 24 bytes in turn will
         * make next 2M slip to one more 2M later.
         * then in big system, the memory will have a lot of holes...
-        * here try to allocate 2M pages continously.
+        * here try to allocate 2M pages continuously.
         *
         * powerpc need to call sparse_init_one_section right after each
         * sparse_early_mem_map_alloc, so allocate usemap_map at first.
@@ -425,10 +511,81 @@ void __init sparse_init(void)
                panic("can not allocate usemap_map\n");
 
        for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
+               struct mem_section *ms;
+
+               if (!present_section_nr(pnum))
+                       continue;
+               ms = __nr_to_section(pnum);
+               nodeid_begin = sparse_early_nid(ms);
+               pnum_begin = pnum;
+               break;
+       }
+       usemap_count = 1;
+       for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
+               struct mem_section *ms;
+               int nodeid;
+
                if (!present_section_nr(pnum))
                        continue;
-               usemap_map[pnum] = sparse_early_usemap_alloc(pnum);
+               ms = __nr_to_section(pnum);
+               nodeid = sparse_early_nid(ms);
+               if (nodeid == nodeid_begin) {
+                       usemap_count++;
+                       continue;
+               }
+               /* ok, we need to take cake of from pnum_begin to pnum - 1*/
+               sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, pnum,
+                                                usemap_count, nodeid_begin);
+               /* new start, update count etc*/
+               nodeid_begin = nodeid;
+               pnum_begin = pnum;
+               usemap_count = 1;
        }
+       /* ok, last chunk */
+       sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, NR_MEM_SECTIONS,
+                                        usemap_count, nodeid_begin);
+
+#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
+       size2 = sizeof(struct page *) * NR_MEM_SECTIONS;
+       map_map = alloc_bootmem(size2);
+       if (!map_map)
+               panic("can not allocate map_map\n");
+
+       for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
+               struct mem_section *ms;
+
+               if (!present_section_nr(pnum))
+                       continue;
+               ms = __nr_to_section(pnum);
+               nodeid_begin = sparse_early_nid(ms);
+               pnum_begin = pnum;
+               break;
+       }
+       map_count = 1;
+       for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
+               struct mem_section *ms;
+               int nodeid;
+
+               if (!present_section_nr(pnum))
+                       continue;
+               ms = __nr_to_section(pnum);
+               nodeid = sparse_early_nid(ms);
+               if (nodeid == nodeid_begin) {
+                       map_count++;
+                       continue;
+               }
+               /* ok, we need to take cake of from pnum_begin to pnum - 1*/
+               sparse_early_mem_maps_alloc_node(map_map, pnum_begin, pnum,
+                                                map_count, nodeid_begin);
+               /* new start, update count etc*/
+               nodeid_begin = nodeid;
+               pnum_begin = pnum;
+               map_count = 1;
+       }
+       /* ok, last chunk */
+       sparse_early_mem_maps_alloc_node(map_map, pnum_begin, NR_MEM_SECTIONS,
+                                        map_count, nodeid_begin);
+#endif
 
        for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
                if (!present_section_nr(pnum))
@@ -438,7 +595,11 @@ void __init sparse_init(void)
                if (!usemap)
                        continue;
 
+#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
+               map = map_map[pnum];
+#else
                map = sparse_early_mem_map_alloc(pnum);
+#endif
                if (!map)
                        continue;
 
@@ -448,6 +609,9 @@ void __init sparse_init(void)
 
        vmemmap_populate_print_last();
 
+#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
+       free_bootmem(__pa(map_map), size2);
+#endif
        free_bootmem(__pa(usemap_map), size);
 }
 
@@ -507,10 +671,10 @@ static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages)
 static void free_map_bootmem(struct page *page, unsigned long nr_pages)
 {
        unsigned long maps_section_nr, removing_section_nr, i;
-       int magic;
+       unsigned long magic;
 
        for (i = 0; i < nr_pages; i++, page++) {
-               magic = atomic_read(&page->_mapcount);
+               magic = (unsigned long) page->lru.next;
 
                BUG_ON(magic == NODE_INFO);
 
@@ -571,7 +735,7 @@ static void free_section_usemap(struct page *memmap, unsigned long *usemap)
  * set.  If this is <=0, then that means that the passed-in
  * map was not consumed and must be freed.
  */
-int sparse_add_one_section(struct zone *zone, unsigned long start_pfn,
+int __meminit sparse_add_one_section(struct zone *zone, unsigned long start_pfn,
                           int nr_pages)
 {
        unsigned long section_nr = pfn_to_section_nr(start_pfn);