VServer 1.9.2 (patch-2.6.8.1-vs1.9.2.diff)
[linux-2.6.git] / arch / parisc / mm / init.c
index e7a8b1b..871d7df 100644 (file)
@@ -5,6 +5,7 @@
  *  Copyright 1999 SuSE GmbH
  *    changed by Philipp Rumpf
  *  Copyright 1999 Philipp Rumpf (prumpf@tux.org)
+ *  Copyright 2004 Randolph Chung (tausq@debian.org)
  *
  */
 
@@ -23,6 +24,7 @@
 #include <asm/pgalloc.h>
 #include <asm/tlb.h>
 #include <asm/pdc_chassis.h>
+#include <asm/mmzone.h>
 
 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
 
@@ -32,10 +34,9 @@ extern char _end;    /* end of BSS, defined by linker */
 extern char __init_begin, __init_end;
 
 #ifdef CONFIG_DISCONTIGMEM
-struct node_map_data node_data[MAX_PHYSMEM_RANGES];
-bootmem_data_t bmem_data[MAX_PHYSMEM_RANGES];
-unsigned char *chunkmap;
-unsigned int maxchunkmap;
+struct node_map_data node_data[MAX_NUMNODES];
+bootmem_data_t bmem_data[MAX_NUMNODES];
+unsigned char pfnnid_map[PFNNID_MAP_MAX];
 #endif
 
 static struct resource data_resource = {
@@ -119,21 +120,6 @@ static void __init setup_bootmem(void)
 
        disable_sr_hashing(); /* Turn off space register hashing */
 
-#ifdef CONFIG_DISCONTIGMEM
-       /*
-        * The below is still true as of 2.4.2. If this is ever fixed,
-        * we can remove this warning!
-        */
-
-       printk(KERN_WARNING "\n\n");
-       printk(KERN_WARNING "CONFIG_DISCONTIGMEM is enabled, which is probably a mistake. This\n");
-       printk(KERN_WARNING "option can lead to heavy swapping, even when there are gigabytes\n");
-       printk(KERN_WARNING "of free memory.\n\n");
-#endif
-
-#ifdef __LP64__
-
-#ifndef CONFIG_DISCONTIGMEM
        /*
         * Sort the ranges. Since the number of ranges is typically
         * small, and performance is not an issue here, just do
@@ -160,11 +146,10 @@ static void __init setup_bootmem(void)
                }
        }
 
+#ifndef CONFIG_DISCONTIGMEM
        /*
         * Throw out ranges that are too far apart (controlled by
-        * MAX_GAP). If CONFIG_DISCONTIGMEM wasn't implemented so
-        * poorly, we would recommend enabling that option, but,
-        * until it is fixed, this is the best way to go.
+        * MAX_GAP).
         */
 
        for (i = 1; i < npmem_ranges; i++) {
@@ -172,6 +157,11 @@ static void __init setup_bootmem(void)
                        (pmem_ranges[i-1].start_pfn +
                         pmem_ranges[i-1].pages) > MAX_GAP) {
                        npmem_ranges = i;
+                       printk("Large gap in memory detected (%ld pages). "
+                              "Consider turning on CONFIG_DISCONTIGMEM\n",
+                              pmem_ranges[i].start_pfn -
+                              (pmem_ranges[i-1].start_pfn +
+                               pmem_ranges[i-1].pages));
                        break;
                }
        }
@@ -194,8 +184,6 @@ static void __init setup_bootmem(void)
                }
        }
 
-#endif /* __LP64__ */
-
        sysram_resource_count = npmem_ranges;
        for (i = 0; i < sysram_resource_count; i++) {
                struct resource *res = &sysram_resources[i];
@@ -218,6 +206,7 @@ static void __init setup_bootmem(void)
        mem_limit_func();       /* check for "mem=" argument */
 
        mem_max = 0;
+       num_physpages = 0;
        for (i = 0; i < npmem_ranges; i++) {
                unsigned long rsize;
 
@@ -232,15 +221,16 @@ static void __init setup_bootmem(void)
                                npmem_ranges = i + 1;
                                mem_max = mem_limit;
                        }
+               num_physpages += pmem_ranges[i].pages;
                        break;
                }
+           num_physpages += pmem_ranges[i].pages;
                mem_max += rsize;
        }
 
        printk(KERN_INFO "Total Memory: %ld Mb\n",mem_max >> 20);
 
 #ifndef CONFIG_DISCONTIGMEM
-
        /* Merge the ranges, keeping track of the holes */
 
        {
@@ -272,9 +262,18 @@ static void __init setup_bootmem(void)
        bootmap_start_pfn = PAGE_ALIGN(__pa((unsigned long) &_end)) >> PAGE_SHIFT;
 
 #ifdef CONFIG_DISCONTIGMEM
+       for (i = 0; i < MAX_PHYSMEM_RANGES; i++) {
+               memset(NODE_DATA(i), 0, sizeof(pg_data_t));
+               NODE_DATA(i)->bdata = &bmem_data[i];
+       }
+       memset(pfnnid_map, 0xff, sizeof(pfnnid_map));
+
+       numnodes = npmem_ranges;
+
        for (i = 0; i < npmem_ranges; i++)
-               node_data[i].pg_data.bdata = &bmem_data[i];
+               node_set_online(i);
 #endif
+
        /*
         * Initialize and free the full range of memory in each range.
         * Note that the only writing these routines do are to the bootmap,
@@ -443,16 +442,20 @@ unsigned long pcxl_dma_start;
 
 void __init mem_init(void)
 {
-       int i;
-
        high_memory = __va((max_pfn << PAGE_SHIFT));
-       max_mapnr = (virt_to_page(high_memory - 1) - mem_map) + 1;
 
-       num_physpages = 0;
-       mem_map = zone_table[0]->zone_mem_map;
-       for (i = 0; i < npmem_ranges; i++)
-               num_physpages += free_all_bootmem_node(NODE_DATA(i));
-       totalram_pages = num_physpages;
+#ifndef CONFIG_DISCONTIGMEM
+       max_mapnr = page_to_pfn(virt_to_page(high_memory - 1)) + 1;
+       mem_map = zone_table[ZONE_DMA]->zone_mem_map;
+       totalram_pages += free_all_bootmem();
+#else
+       {
+               int i;
+
+               for (i = 0; i < npmem_ranges; i++)
+                       totalram_pages += free_all_bootmem_node(NODE_DATA(i));
+       }
+#endif
 
        printk(KERN_INFO "Memory: %luk available\n", num_physpages << (PAGE_SHIFT-10));
 
@@ -484,7 +487,9 @@ void show_mem(void)
 
        printk(KERN_INFO "Mem-info:\n");
        show_free_areas();
-       printk(KERN_INFO "Free swap:     %6dkB\n",nr_swap_pages<<(PAGE_SHIFT-10));
+       printk(KERN_INFO "Free swap:     %6ldkB\n",
+                               nr_swap_pages<<(PAGE_SHIFT-10));
+#ifndef CONFIG_DISCONTIGMEM
        i = max_mapnr;
        while (i-- > 0) {
                total++;
@@ -492,15 +497,55 @@ void show_mem(void)
                        reserved++;
                else if (PageSwapCache(mem_map+i))
                        cached++;
-               else if (!atomic_read(&mem_map[i].count))
+               else if (!page_count(&mem_map[i]))
                        free++;
                else
-                       shared += atomic_read(&mem_map[i].count) - 1;
+                       shared += page_count(&mem_map[i]) - 1;
+       }
+#else
+       for (i = 0; i < npmem_ranges; i++) {
+               int j;
+
+               for (j = node_start_pfn(i); j < node_end_pfn(i); j++) {
+                       struct page *p;
+
+                       p = node_mem_map(i) + j - node_start_pfn(i);
+
+                       total++;
+                       if (PageReserved(p))
+                               reserved++;
+                       else if (PageSwapCache(p))
+                               cached++;
+                       else if (!page_count(p))
+                               free++;
+                       else
+                               shared += page_count(p) - 1;
+               }
        }
+#endif
        printk(KERN_INFO "%d pages of RAM\n", total);
        printk(KERN_INFO "%d reserved pages\n", reserved);
        printk(KERN_INFO "%d pages shared\n", shared);
        printk(KERN_INFO "%d pages swap cached\n", cached);
+
+
+#ifdef CONFIG_DISCONTIGMEM
+       {
+               struct zonelist *zl;
+               int i, j, k;
+
+               for (i = 0; i < npmem_ranges; i++) {
+                       for (j = 0; j < MAX_NR_ZONES; j++) {
+                               zl = NODE_DATA(i)->node_zonelists + j;
+
+                               printk("Zone list for zone %d on node %d: ", j, i);
+                               for (k = 0; zl->zones[k] != NULL; k++) 
+                                       printk("[%d/%s] ", zl->zones[k]->zone_pgdat->node_id, zl->zones[k]->name);
+                               printk("\n");
+                       }
+               }
+       }
+#endif
 }
 
 
@@ -543,7 +588,7 @@ static void __init map_pages(unsigned long start_vaddr, unsigned long start_padd
 #if PTRS_PER_PMD == 1
                pmd = (pmd_t *)__pa(pg_dir);
 #else
-               pmd = (pmd_t *) (PAGE_MASK & pgd_val(*pg_dir));
+               pmd = (pmd_t *)pgd_address(*pg_dir);
 
                /*
                 * pmd is physical at this point
@@ -554,7 +599,7 @@ static void __init map_pages(unsigned long start_vaddr, unsigned long start_padd
                        pmd = (pmd_t *) __pa(pmd);
                }
 
-               pgd_val(*pg_dir) = _PAGE_TABLE | (unsigned long) pmd;
+               pgd_populate(NULL, pg_dir, __va(pmd));
 #endif
                pg_dir++;
 
@@ -567,15 +612,14 @@ static void __init map_pages(unsigned long start_vaddr, unsigned long start_padd
                         * pg_table is physical at this point
                         */
 
-                       pg_table = (pte_t *) (PAGE_MASK & pmd_val(*pmd));
+                       pg_table = (pte_t *)pmd_address(*pmd);
                        if (!pg_table) {
                                pg_table = (pte_t *)
                                        alloc_bootmem_low_pages_node(NODE_DATA(0),PAGE_SIZE);
                                pg_table = (pte_t *) __pa(pg_table);
                        }
 
-                       pmd_val(*pmd) = _PAGE_TABLE |
-                                          (unsigned long) pg_table;
+                       pmd_populate_kernel(NULL, pmd, __va(pg_table));
 
                        /* now change pg_table to kernel virtual addresses */
 
@@ -583,8 +627,6 @@ static void __init map_pages(unsigned long start_vaddr, unsigned long start_padd
                        for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++,pg_table++) {
                                pte_t pte;
 
-#if !defined(CONFIG_STI_CONSOLE)
-#warning STI console should explicitly allocate executable pages but does not
                                /*
                                 * Map the fault vector writable so we can
                                 * write the HPMC checksum.
@@ -594,7 +636,6 @@ static void __init map_pages(unsigned long start_vaddr, unsigned long start_padd
                                                        && address != gw_addr)
                                    pte = __mk_pte(address, PAGE_KERNEL_RO);
                                else
-#endif
                                    pte = __mk_pte(address, pgprot);
 
                                if (address >= end_paddr)
@@ -757,61 +798,26 @@ void __init paging_init(void)
        flush_tlb_all_local();
 
        for (i = 0; i < npmem_ranges; i++) {
-               unsigned long zones_size[MAX_NR_ZONES] = { 0, 0, 0, };
+               unsigned long zones_size[MAX_NR_ZONES] = { 0, 0, 0 };
 
+               /* We have an IOMMU, so all memory can go into a single
+                  ZONE_DMA zone. */
                zones_size[ZONE_DMA] = pmem_ranges[i].pages;
+
                free_area_init_node(i,NODE_DATA(i),NULL,zones_size,
-                               (pmem_ranges[i].start_pfn << PAGE_SHIFT),0);
-       }
+                               pmem_ranges[i].start_pfn, 0);
 
 #ifdef CONFIG_DISCONTIGMEM
-       /*
-        * Initialize support for virt_to_page() macro.
-        *
-        * Note that MAX_ADDRESS is the largest virtual address that
-        * we can map. However, since we map all physical memory into
-        * the kernel address space, it also has an effect on the maximum
-        * physical address we can map (MAX_ADDRESS - PAGE_OFFSET).
-        */
-
-       maxchunkmap = MAX_ADDRESS >> CHUNKSHIFT;
-       chunkmap = (unsigned char *)alloc_bootmem(maxchunkmap);
-
-       for (i = 0; i < maxchunkmap; i++)
-           chunkmap[i] = BADCHUNK;
-
-       for (i = 0; i < npmem_ranges; i++) {
-
-               ADJ_NODE_MEM_MAP(i) = NODE_MEM_MAP(i) - pmem_ranges[i].start_pfn;
                {
-                       unsigned long chunk_paddr;
-                       unsigned long end_paddr;
-                       int chunknum;
-
-                       chunk_paddr = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
-                       end_paddr = chunk_paddr + (pmem_ranges[i].pages << PAGE_SHIFT);
-                       chunk_paddr &= CHUNKMASK;
-
-                       chunknum = (int)CHUNKNUM(chunk_paddr);
-                       while (chunk_paddr < end_paddr) {
-                               if (chunknum >= maxchunkmap)
-                                       goto badchunkmap1;
-                               if (chunkmap[chunknum] != BADCHUNK)
-                                       goto badchunkmap2;
-                               chunkmap[chunknum] = (unsigned char)i;
-                               chunk_paddr += CHUNKSZ;
-                               chunknum++;
-                       }
+                   int j;
+                   for (j = (node_start_pfn(i) >> PFNNID_SHIFT);
+                        j <= (node_end_pfn(i) >> PFNNID_SHIFT);
+                        j++) {
+                       pfnnid_map[j] = i;
+                   }
                }
-       }
-
-       return;
-
-badchunkmap1:
-       panic("paging_init: Physical address exceeds maximum address space!\n");
-badchunkmap2:
-       panic("paging_init: Collision in chunk map array. CHUNKSZ needs to be smaller\n");
 #endif
+       }
 }
 
 #ifdef CONFIG_PA20