arch/parisc/mm/init.c

   1 /*
   2  *  linux/arch/parisc/mm/init.c
   3  *
   4  *  Copyright (C) 1995  Linus Torvalds
   5  *  Copyright 1999 SuSE GmbH
   6  *    changed by Philipp Rumpf
   7  *  Copyright 1999 Philipp Rumpf (prumpf@tux.org)
   8  *  Copyright 2004 Randolph Chung (tausq@debian.org)
   9  *  Copyright 2006-2007 Helge Deller (deller@gmx.de)
  10  *
  11  */
  12
  13
  14 #include <linux/module.h>
  15 #include <linux/mm.h>
  16 #include <linux/bootmem.h>
  17 #include <linux/gfp.h>
  18 #include <linux/delay.h>
  19 #include <linux/init.h>
  20 #include <linux/pci.h>          /* for hppa_dma_ops and pcxl_dma_ops */
  21 #include <linux/initrd.h>
  22 #include <linux/swap.h>
  23 #include <linux/unistd.h>
  24 #include <linux/nodemask.h>     /* for node_online_map */
  25 #include <linux/pagemap.h>      /* for release_pages and page_cache_release */
  26
  27 #include <asm/pgalloc.h>
  28 #include <asm/pgtable.h>
  29 #include <asm/tlb.h>
  30 #include <asm/pdc_chassis.h>
  31 #include <asm/mmzone.h>
  32 #include <asm/sections.h>
  33
  34 extern int  data_start;
  35
  36 #ifdef CONFIG_DISCONTIGMEM
  37 struct node_map_data node_data[MAX_NUMNODES] __read_mostly;
  38 unsigned char pfnnid_map[PFNNID_MAP_MAX] __read_mostly;
  39 #endif
  40
  41 static struct resource data_resource = {
  42         .name   = "Kernel data",
  43         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM,
  44 };
  45
  46 static struct resource code_resource = {
  47         .name   = "Kernel code",
  48         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM,
  49 };
  50
  51 static struct resource pdcdata_resource = {
  52         .name   = "PDC data (Page Zero)",
  53         .start  = 0,
  54         .end    = 0x9ff,
  55         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM,
  56 };
  57
  58 static struct resource sysram_resources[MAX_PHYSMEM_RANGES] __read_mostly;
  59
  60 /* The following array is initialized from the firmware specific
  61  * information retrieved in kernel/inventory.c.
  62  */
  63
  64 physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES] __read_mostly;
  65 int npmem_ranges __read_mostly;
  66
  67 #ifdef CONFIG_64BIT
  68 #define MAX_MEM         (~0UL)
  69 #else /* !CONFIG_64BIT */
  70 #define MAX_MEM         (3584U*1024U*1024U)
  71 #endif /* !CONFIG_64BIT */
  72
  73 static unsigned long mem_limit __read_mostly = MAX_MEM;
  74
  75 static void __init mem_limit_func(void)
  76 {
  77         char *cp, *end;
  78         unsigned long limit;
  79
  80         /* We need this before __setup() functions are called */
  81
  82         limit = MAX_MEM;
  83         for (cp = boot_command_line; *cp; ) {
  84                 if (memcmp(cp, "mem=", 4) == 0) {
  85                         cp += 4;
  86                         limit = memparse(cp, &end);
  87                         if (end != cp)
  88                                 break;
  89                         cp = end;
  90                 } else {
  91                         while (*cp != ' ' && *cp)
  92                                 ++cp;
  93                         while (*cp == ' ')
  94                                 ++cp;
  95                 }
  96         }
  97
  98         if (limit < mem_limit)
  99                 mem_limit = limit;
 100 }
 101
 102 #define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
 103
 104 static void __init setup_bootmem(void)
 105 {
 106         unsigned long bootmap_size;
 107         unsigned long mem_max;
 108         unsigned long bootmap_pages;
 109         unsigned long bootmap_start_pfn;
 110         unsigned long bootmap_pfn;
 111 #ifndef CONFIG_DISCONTIGMEM
 112         physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1];
 113         int npmem_holes;
 114 #endif
 115         int i, sysram_resource_count;
 116
 117         disable_sr_hashing(); /* Turn off space register hashing */
 118
 119         /*
 120          * Sort the ranges. Since the number of ranges is typically
 121          * small, and performance is not an issue here, just do
 122          * a simple insertion sort.
 123          */
 124
 125         for (i = 1; i < npmem_ranges; i++) {
 126                 int j;
 127
 128                 for (j = i; j > 0; j--) {
 129                         unsigned long tmp;
 130
 131                         if (pmem_ranges[j-1].start_pfn <
 132                             pmem_ranges[j].start_pfn) {
 133
 134                                 break;
 135                         }
 136                         tmp = pmem_ranges[j-1].start_pfn;
 137                         pmem_ranges[j-1].start_pfn = pmem_ranges[j].start_pfn;
 138                         pmem_ranges[j].start_pfn = tmp;
 139                         tmp = pmem_ranges[j-1].pages;
 140                         pmem_ranges[j-1].pages = pmem_ranges[j].pages;
 141                         pmem_ranges[j].pages = tmp;
 142                 }
 143         }
 144
 145 #ifndef CONFIG_DISCONTIGMEM
 146         /*
 147          * Throw out ranges that are too far apart (controlled by
 148          * MAX_GAP).
 149          */
 150
 151         for (i = 1; i < npmem_ranges; i++) {
 152                 if (pmem_ranges[i].start_pfn -
 153                         (pmem_ranges[i-1].start_pfn +
 154                          pmem_ranges[i-1].pages) > MAX_GAP) {
 155                         npmem_ranges = i;
 156                         printk("Large gap in memory detected (%ld pages). "
 157                                "Consider turning on CONFIG_DISCONTIGMEM\n",
 158                                pmem_ranges[i].start_pfn -
 159                                (pmem_ranges[i-1].start_pfn +
 160                                 pmem_ranges[i-1].pages));
 161                         break;
 162                 }
 163         }
 164 #endif
 165
 166         if (npmem_ranges > 1) {
 167
 168                 /* Print the memory ranges */
 169
 170                 printk(KERN_INFO "Memory Ranges:\n");
 171
 172                 for (i = 0; i < npmem_ranges; i++) {
 173                         unsigned long start;
 174                         unsigned long size;
 175
 176                         size = (pmem_ranges[i].pages << PAGE_SHIFT);
 177                         start = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
 178                         printk(KERN_INFO "%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n",
 179                                 i,start, start + (size - 1), size >> 20);
 180                 }
 181         }
 182
 183         sysram_resource_count = npmem_ranges;
 184         for (i = 0; i < sysram_resource_count; i++) {
 185                 struct resource *res = &sysram_resources[i];
 186                 res->name = "System RAM";
 187                 res->start = pmem_ranges[i].start_pfn << PAGE_SHIFT;
 188                 res->end = res->start + (pmem_ranges[i].pages << PAGE_SHIFT)-1;
 189                 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
 190                 request_resource(&iomem_resource, res);
 191         }
 192
 193         /*
 194          * For 32 bit kernels we limit the amount of memory we can
 195          * support, in order to preserve enough kernel address space
 196          * for other purposes. For 64 bit kernels we don't normally
 197          * limit the memory, but this mechanism can be used to
 198          * artificially limit the amount of memory (and it is written
 199          * to work with multiple memory ranges).
 200          */
 201
 202         mem_limit_func();       /* check for "mem=" argument */
 203
 204         mem_max = 0;
 205         num_physpages = 0;
 206         for (i = 0; i < npmem_ranges; i++) {
 207                 unsigned long rsize;
 208
 209                 rsize = pmem_ranges[i].pages << PAGE_SHIFT;
 210                 if ((mem_max + rsize) > mem_limit) {
 211                         printk(KERN_WARNING "Memory truncated to %ld MB\n", mem_limit >> 20);
 212                         if (mem_max == mem_limit)
 213                                 npmem_ranges = i;
 214                         else {
 215                                 pmem_ranges[i].pages =   (mem_limit >> PAGE_SHIFT)
 216                                                        - (mem_max >> PAGE_SHIFT);
 217                                 npmem_ranges = i + 1;
 218                                 mem_max = mem_limit;
 219                         }
 220                 num_physpages += pmem_ranges[i].pages;
 221                         break;
 222                 }
 223             num_physpages += pmem_ranges[i].pages;
 224                 mem_max += rsize;
 225         }
 226
 227         printk(KERN_INFO "Total Memory: %ld MB\n",mem_max >> 20);
 228
 229 #ifndef CONFIG_DISCONTIGMEM
 230         /* Merge the ranges, keeping track of the holes */
 231
 232         {
 233                 unsigned long end_pfn;
 234                 unsigned long hole_pages;
 235
 236                 npmem_holes = 0;
 237                 end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages;
 238                 for (i = 1; i < npmem_ranges; i++) {
 239
 240                         hole_pages = pmem_ranges[i].start_pfn - end_pfn;
 241                         if (hole_pages) {
 242                                 pmem_holes[npmem_holes].start_pfn = end_pfn;
 243                                 pmem_holes[npmem_holes++].pages = hole_pages;
 244                                 end_pfn += hole_pages;
 245                         }
 246                         end_pfn += pmem_ranges[i].pages;
 247                 }
 248
 249                 pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn;
 250                 npmem_ranges = 1;
 251         }
 252 #endif
 253
 254         bootmap_pages = 0;
 255         for (i = 0; i < npmem_ranges; i++)
 256                 bootmap_pages += bootmem_bootmap_pages(pmem_ranges[i].pages);
 257
 258         bootmap_start_pfn = PAGE_ALIGN(__pa((unsigned long) &_end)) >> PAGE_SHIFT;
 259
 260 #ifdef CONFIG_DISCONTIGMEM
 261         for (i = 0; i < MAX_PHYSMEM_RANGES; i++) {
 262                 memset(NODE_DATA(i), 0, sizeof(pg_data_t));
 263                 NODE_DATA(i)->bdata = &bootmem_node_data[i];
 264         }
 265         memset(pfnnid_map, 0xff, sizeof(pfnnid_map));
 266
 267         for (i = 0; i < npmem_ranges; i++) {
 268                 node_set_state(i, N_NORMAL_MEMORY);
 269                 node_set_online(i);
 270         }
 271 #endif
 272
 273         /*
 274          * Initialize and free the full range of memory in each range.
 275          * Note that the only writing these routines do are to the bootmap,
 276          * and we've made sure to locate the bootmap properly so that they
 277          * won't be writing over anything important.
 278          */
 279
 280         bootmap_pfn = bootmap_start_pfn;
 281         max_pfn = 0;
 282         for (i = 0; i < npmem_ranges; i++) {
 283                 unsigned long start_pfn;
 284                 unsigned long npages;
 285
 286                 start_pfn = pmem_ranges[i].start_pfn;
 287                 npages = pmem_ranges[i].pages;
 288
 289                 bootmap_size = init_bootmem_node(NODE_DATA(i),
 290                                                 bootmap_pfn,
 291                                                 start_pfn,
 292                                                 (start_pfn + npages) );
 293                 free_bootmem_node(NODE_DATA(i),
 294                                   (start_pfn << PAGE_SHIFT),
 295                                   (npages << PAGE_SHIFT) );
 296                 bootmap_pfn += (bootmap_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 297                 if ((start_pfn + npages) > max_pfn)
 298                         max_pfn = start_pfn + npages;
 299         }
 300
 301         /* IOMMU is always used to access "high mem" on those boxes
 302          * that can support enough mem that a PCI device couldn't
 303          * directly DMA to any physical addresses.
 304          * ISA DMA support will need to revisit this.
 305          */
 306         max_low_pfn = max_pfn;
 307
 308         /* bootmap sizing messed up? */
 309         BUG_ON((bootmap_pfn - bootmap_start_pfn) != bootmap_pages);
 310
 311         /* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
 312
 313 #define PDC_CONSOLE_IO_IODC_SIZE 32768
 314
 315         reserve_bootmem_node(NODE_DATA(0), 0UL,
 316                         (unsigned long)(PAGE0->mem_free +
 317                                 PDC_CONSOLE_IO_IODC_SIZE), BOOTMEM_DEFAULT);
 318         reserve_bootmem_node(NODE_DATA(0), __pa((unsigned long)_text),
 319                         (unsigned long)(_end - _text), BOOTMEM_DEFAULT);
 320         reserve_bootmem_node(NODE_DATA(0), (bootmap_start_pfn << PAGE_SHIFT),
 321                         ((bootmap_pfn - bootmap_start_pfn) << PAGE_SHIFT),
 322                         BOOTMEM_DEFAULT);
 323
 324 #ifndef CONFIG_DISCONTIGMEM
 325
 326         /* reserve the holes */
 327
 328         for (i = 0; i < npmem_holes; i++) {
 329                 reserve_bootmem_node(NODE_DATA(0),
 330                                 (pmem_holes[i].start_pfn << PAGE_SHIFT),
 331                                 (pmem_holes[i].pages << PAGE_SHIFT),
 332                                 BOOTMEM_DEFAULT);
 333         }
 334 #endif
 335
 336 #ifdef CONFIG_BLK_DEV_INITRD
 337         if (initrd_start) {
 338                 printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
 339                 if (__pa(initrd_start) < mem_max) {
 340                         unsigned long initrd_reserve;
 341
 342                         if (__pa(initrd_end) > mem_max) {
 343                                 initrd_reserve = mem_max - __pa(initrd_start);
 344                         } else {
 345                                 initrd_reserve = initrd_end - initrd_start;
 346                         }
 347                         initrd_below_start_ok = 1;
 348                         printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
 349
 350                         reserve_bootmem_node(NODE_DATA(0), __pa(initrd_start),
 351                                         initrd_reserve, BOOTMEM_DEFAULT);
 352                 }
 353         }
 354 #endif
 355
 356         data_resource.start =  virt_to_phys(&data_start);
 357         data_resource.end = virt_to_phys(_end) - 1;
 358         code_resource.start = virt_to_phys(_text);
 359         code_resource.end = virt_to_phys(&data_start)-1;
 360
 361         /* We don't know which region the kernel will be in, so try
 362          * all of them.
 363          */
 364         for (i = 0; i < sysram_resource_count; i++) {
 365                 struct resource *res = &sysram_resources[i];
 366                 request_resource(res, &code_resource);
 367                 request_resource(res, &data_resource);
 368         }
 369         request_resource(&sysram_resources[0], &pdcdata_resource);
 370 }
 371
 372 static void __init map_pages(unsigned long start_vaddr,
 373                              unsigned long start_paddr, unsigned long size,
 374                              pgprot_t pgprot, int force)
 375 {
 376         pgd_t *pg_dir;
 377         pmd_t *pmd;
 378         pte_t *pg_table;
 379         unsigned long end_paddr;
 380         unsigned long start_pmd;
 381         unsigned long start_pte;
 382         unsigned long tmp1;
 383         unsigned long tmp2;
 384         unsigned long address;
 385         unsigned long vaddr;
 386         unsigned long ro_start;
 387         unsigned long ro_end;
 388         unsigned long fv_addr;
 389         unsigned long gw_addr;
 390         extern const unsigned long fault_vector_20;
 391         extern void * const linux_gateway_page;
 392
 393         ro_start = __pa((unsigned long)_text);
 394         ro_end   = __pa((unsigned long)&data_start);
 395         fv_addr  = __pa((unsigned long)&fault_vector_20) & PAGE_MASK;
 396         gw_addr  = __pa((unsigned long)&linux_gateway_page) & PAGE_MASK;
 397
 398         end_paddr = start_paddr + size;
 399
 400         pg_dir = pgd_offset_k(start_vaddr);
 401
 402 #if PTRS_PER_PMD == 1
 403         start_pmd = 0;
 404 #else
 405         start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
 406 #endif
 407         start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
 408
 409         address = start_paddr;
 410         vaddr = start_vaddr;
 411         while (address < end_paddr) {
 412 #if PTRS_PER_PMD == 1
 413                 pmd = (pmd_t *)__pa(pg_dir);
 414 #else
 415                 pmd = (pmd_t *)pgd_address(*pg_dir);
 416
 417                 /*
 418                  * pmd is physical at this point
 419                  */
 420
 421                 if (!pmd) {
 422                         pmd = (pmd_t *) alloc_bootmem_low_pages_node(NODE_DATA(0), PAGE_SIZE << PMD_ORDER);
 423                         pmd = (pmd_t *) __pa(pmd);
 424                 }
 425
 426                 pgd_populate(NULL, pg_dir, __va(pmd));
 427 #endif
 428                 pg_dir++;
 429
 430                 /* now change pmd to kernel virtual addresses */
 431
 432                 pmd = (pmd_t *)__va(pmd) + start_pmd;
 433                 for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++, pmd++) {
 434
 435                         /*
 436                          * pg_table is physical at this point
 437                          */
 438
 439                         pg_table = (pte_t *)pmd_address(*pmd);
 440                         if (!pg_table) {
 441                                 pg_table = (pte_t *)
 442                                         alloc_bootmem_low_pages_node(NODE_DATA(0), PAGE_SIZE);
 443                                 pg_table = (pte_t *) __pa(pg_table);
 444                         }
 445
 446                         pmd_populate_kernel(NULL, pmd, __va(pg_table));
 447
 448                         /* now change pg_table to kernel virtual addresses */
 449
 450                         pg_table = (pte_t *) __va(pg_table) + start_pte;
 451                         for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++, pg_table++) {
 452                                 pte_t pte;
 453
 454                                 /*
 455                                  * Map the fault vector writable so we can
 456                                  * write the HPMC checksum.
 457                                  */
 458                                 if (force)
 459                                         pte =  __mk_pte(address, pgprot);
 460                                 else if (core_kernel_text(vaddr) &&
 461                                          address != fv_addr)
 462                                         pte = __mk_pte(address, PAGE_KERNEL_EXEC);
 463                                 else
 464 #if defined(CONFIG_PARISC_PAGE_SIZE_4KB)
 465                                 if (address >= ro_start && address < ro_end
 466                                                         && address != fv_addr
 467                                                         && address != gw_addr)
 468                                         pte = __mk_pte(address, PAGE_KERNEL_RO);
 469                                 else
 470 #endif
 471                                         pte = __mk_pte(address, pgprot);
 472
 473                                 if (address >= end_paddr) {
 474                                         if (force)
 475                                                 break;
 476                                         else
 477                                                 pte_val(pte) = 0;
 478                                 }
 479
 480                                 set_pte(pg_table, pte);
 481
 482                                 address += PAGE_SIZE;
 483                                 vaddr += PAGE_SIZE;
 484                         }
 485                         start_pte = 0;
 486
 487                         if (address >= end_paddr)
 488                             break;
 489                 }
 490                 start_pmd = 0;
 491         }
 492 }
 493
 494 void free_initmem(void)
 495 {
 496         unsigned long addr;
 497         unsigned long init_begin = (unsigned long)__init_begin;
 498         unsigned long init_end = (unsigned long)__init_end;
 499
 500         /* The init text pages are marked R-X.  We have to
 501          * flush the icache and mark them RW-
 502          *
 503          * This is tricky, because map_pages is in the init section.
 504          * Do a dummy remap of the data section first (the data
 505          * section is already PAGE_KERNEL) to pull in the TLB entries
 506          * for map_kernel */
 507         map_pages(init_begin, __pa(init_begin), init_end - init_begin,
 508                   PAGE_KERNEL_RWX, 1);
 509         /* now remap at PAGE_KERNEL since the TLB is pre-primed to execute
 510          * map_pages */
 511         map_pages(init_begin, __pa(init_begin), init_end - init_begin,
 512                   PAGE_KERNEL, 1);
 513
 514         /* force the kernel to see the new TLB entries */
 515         __flush_tlb_range(0, init_begin, init_end);
 516         /* Attempt to catch anyone trying to execute code here
 517          * by filling the page with BRK insns.
 518          */
 519         memset((void *)init_begin, 0x00, init_end - init_begin);
 520         /* finally dump all the instructions which were cached, since the
 521          * pages are no-longer executable */
 522         flush_icache_range(init_begin, init_end);
 523
 524         for (addr = init_begin; addr < init_end; addr += PAGE_SIZE) {
 525                 ClearPageReserved(virt_to_page(addr));
 526                 init_page_count(virt_to_page(addr));
 527                 free_page(addr);
 528                 num_physpages++;
 529                 totalram_pages++;
 530         }
 531
 532         /* set up a new led state on systems shipped LED State panel */
 533         pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
 534
 535         printk(KERN_INFO "Freeing unused kernel memory: %luk freed\n",
 536                 (init_end - init_begin) >> 10);
 537 }
 538
 539
 540 #ifdef CONFIG_DEBUG_RODATA
 541 void mark_rodata_ro(void)
 542 {
 543         /* rodata memory was already mapped with KERNEL_RO access rights by
 544            pagetable_init() and map_pages(). No need to do additional stuff here */
 545         printk (KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 546                 (unsigned long)(__end_rodata - __start_rodata) >> 10);
 547 }
 548 #endif
 549
 550
 551 /*
 552  * Just an arbitrary offset to serve as a "hole" between mapping areas
 553  * (between top of physical memory and a potential pcxl dma mapping
 554  * area, and below the vmalloc mapping area).
 555  *
 556  * The current 32K value just means that there will be a 32K "hole"
 557  * between mapping areas. That means that  any out-of-bounds memory
 558  * accesses will hopefully be caught. The vmalloc() routines leaves
 559  * a hole of 4kB between each vmalloced area for the same reason.
 560  */
 561
 562  /* Leave room for gateway page expansion */
 563 #if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
 564 #error KERNEL_MAP_START is in gateway reserved region
 565 #endif
 566 #define MAP_START (KERNEL_MAP_START)
 567
 568 #define VM_MAP_OFFSET  (32*1024)
 569 #define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
 570                                      & ~(VM_MAP_OFFSET-1)))
 571
 572 void *parisc_vmalloc_start __read_mostly;
 573 EXPORT_SYMBOL(parisc_vmalloc_start);
 574
 575 #ifdef CONFIG_PA11
 576 unsigned long pcxl_dma_start __read_mostly;
 577 #endif
 578
 579 void __init mem_init(void)
 580 {
 581         int codesize, reservedpages, datasize, initsize;
 582
 583         /* Do sanity checks on page table constants */
 584         BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t));
 585         BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t));
 586         BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t));
 587         BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD
 588                         > BITS_PER_LONG);
 589
 590         high_memory = __va((max_pfn << PAGE_SHIFT));
 591
 592 #ifndef CONFIG_DISCONTIGMEM
 593         max_mapnr = page_to_pfn(virt_to_page(high_memory - 1)) + 1;
 594         totalram_pages += free_all_bootmem();
 595 #else
 596         {
 597                 int i;
 598
 599                 for (i = 0; i < npmem_ranges; i++)
 600                         totalram_pages += free_all_bootmem_node(NODE_DATA(i));
 601         }
 602 #endif
 603
 604         codesize = (unsigned long)_etext - (unsigned long)_text;
 605         datasize = (unsigned long)_edata - (unsigned long)_etext;
 606         initsize = (unsigned long)__init_end - (unsigned long)__init_begin;
 607
 608         reservedpages = 0;
 609 {
 610         unsigned long pfn;
 611 #ifdef CONFIG_DISCONTIGMEM
 612         int i;
 613
 614         for (i = 0; i < npmem_ranges; i++) {
 615                 for (pfn = node_start_pfn(i); pfn < node_end_pfn(i); pfn++) {
 616                         if (PageReserved(pfn_to_page(pfn)))
 617                                 reservedpages++;
 618                 }
 619         }
 620 #else /* !CONFIG_DISCONTIGMEM */
 621         for (pfn = 0; pfn < max_pfn; pfn++) {
 622                 /*
 623                  * Only count reserved RAM pages
 624                  */
 625                 if (PageReserved(pfn_to_page(pfn)))
 626                         reservedpages++;
 627         }
 628 #endif
 629 }
 630
 631 #ifdef CONFIG_PA11
 632         if (hppa_dma_ops == &pcxl_dma_ops) {
 633                 pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
 634                 parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start
 635                                                 + PCXL_DMA_MAP_SIZE);
 636         } else {
 637                 pcxl_dma_start = 0;
 638                 parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
 639         }
 640 #else
 641         parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
 642 #endif
 643
 644         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init)\n",
 645                 nr_free_pages() << (PAGE_SHIFT-10),
 646                 num_physpages << (PAGE_SHIFT-10),
 647                 codesize >> 10,
 648                 reservedpages << (PAGE_SHIFT-10),
 649                 datasize >> 10,
 650                 initsize >> 10
 651         );
 652
 653 #ifdef CONFIG_DEBUG_KERNEL /* double-sanity-check paranoia */
 654         printk("virtual kernel memory layout:\n"
 655                "    vmalloc : 0x%p - 0x%p   (%4ld MB)\n"
 656                "    memory  : 0x%p - 0x%p   (%4ld MB)\n"
 657                "      .init : 0x%p - 0x%p   (%4ld kB)\n"
 658                "      .data : 0x%p - 0x%p   (%4ld kB)\n"
 659                "      .text : 0x%p - 0x%p   (%4ld kB)\n",
 660
 661                (void*)VMALLOC_START, (void*)VMALLOC_END,
 662                (VMALLOC_END - VMALLOC_START) >> 20,
 663
 664                __va(0), high_memory,
 665                ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 666
 667                __init_begin, __init_end,
 668                ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10,
 669
 670                _etext, _edata,
 671                ((unsigned long)_edata - (unsigned long)_etext) >> 10,
 672
 673                _text, _etext,
 674                ((unsigned long)_etext - (unsigned long)_text) >> 10);
 675 #endif
 676 }
 677
 678 unsigned long *empty_zero_page __read_mostly;
 679 EXPORT_SYMBOL(empty_zero_page);
 680
 681 void show_mem(unsigned int filter)
 682 {
 683         int i,free = 0,total = 0,reserved = 0;
 684         int shared = 0, cached = 0;
 685
 686         printk(KERN_INFO "Mem-info:\n");
 687         show_free_areas(filter);
 688 #ifndef CONFIG_DISCONTIGMEM
 689         i = max_mapnr;
 690         while (i-- > 0) {
 691                 total++;
 692                 if (PageReserved(mem_map+i))
 693                         reserved++;
 694                 else if (PageSwapCache(mem_map+i))
 695                         cached++;
 696                 else if (!page_count(&mem_map[i]))
 697                         free++;
 698                 else
 699                         shared += page_count(&mem_map[i]) - 1;
 700         }
 701 #else
 702         for (i = 0; i < npmem_ranges; i++) {
 703                 int j;
 704
 705                 for (j = node_start_pfn(i); j < node_end_pfn(i); j++) {
 706                         struct page *p;
 707                         unsigned long flags;
 708
 709                         pgdat_resize_lock(NODE_DATA(i), &flags);
 710                         p = nid_page_nr(i, j) - node_start_pfn(i);
 711
 712                         total++;
 713                         if (PageReserved(p))
 714                                 reserved++;
 715                         else if (PageSwapCache(p))
 716                                 cached++;
 717                         else if (!page_count(p))
 718                                 free++;
 719                         else
 720                                 shared += page_count(p) - 1;
 721                         pgdat_resize_unlock(NODE_DATA(i), &flags);
 722                 }
 723         }
 724 #endif
 725         printk(KERN_INFO "%d pages of RAM\n", total);
 726         printk(KERN_INFO "%d reserved pages\n", reserved);
 727         printk(KERN_INFO "%d pages shared\n", shared);
 728         printk(KERN_INFO "%d pages swap cached\n", cached);
 729
 730
 731 #ifdef CONFIG_DISCONTIGMEM
 732         {
 733                 struct zonelist *zl;
 734                 int i, j;
 735
 736                 for (i = 0; i < npmem_ranges; i++) {
 737                         zl = node_zonelist(i, 0);
 738                         for (j = 0; j < MAX_NR_ZONES; j++) {
 739                                 struct zoneref *z;
 740                                 struct zone *zone;
 741
 742                                 printk("Zone list for zone %d on node %d: ", j, i);
 743                                 for_each_zone_zonelist(zone, z, zl, j)
 744                                         printk("[%d/%s] ", zone_to_nid(zone),
 745                                                                 zone->name);
 746                                 printk("\n");
 747                         }
 748                 }
 749         }
 750 #endif
 751 }
 752
 753 /*
 754  * pagetable_init() sets up the page tables
 755  *
 756  * Note that gateway_init() places the Linux gateway page at page 0.
 757  * Since gateway pages cannot be dereferenced this has the desirable
 758  * side effect of trapping those pesky NULL-reference errors in the
 759  * kernel.
 760  */
 761 static void __init pagetable_init(void)
 762 {
 763         int range;
 764
 765         /* Map each physical memory range to its kernel vaddr */
 766
 767         for (range = 0; range < npmem_ranges; range++) {
 768                 unsigned long start_paddr;
 769                 unsigned long end_paddr;
 770                 unsigned long size;
 771
 772                 start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
 773                 end_paddr = start_paddr + (pmem_ranges[range].pages << PAGE_SHIFT);
 774                 size = pmem_ranges[range].pages << PAGE_SHIFT;
 775
 776                 map_pages((unsigned long)__va(start_paddr), start_paddr,
 777                           size, PAGE_KERNEL, 0);
 778         }
 779
 780 #ifdef CONFIG_BLK_DEV_INITRD
 781         if (initrd_end && initrd_end > mem_limit) {
 782                 printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
 783                 map_pages(initrd_start, __pa(initrd_start),
 784                           initrd_end - initrd_start, PAGE_KERNEL, 0);
 785         }
 786 #endif
 787
 788         empty_zero_page = alloc_bootmem_pages(PAGE_SIZE);
 789         memset(empty_zero_page, 0, PAGE_SIZE);
 790 }
 791
 792 static void __init gateway_init(void)
 793 {
 794         unsigned long linux_gateway_page_addr;
 795         /* FIXME: This is 'const' in order to trick the compiler
 796            into not treating it as DP-relative data. */
 797         extern void * const linux_gateway_page;
 798
 799         linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
 800
 801         /*
 802          * Setup Linux Gateway page.
 803          *
 804          * The Linux gateway page will reside in kernel space (on virtual
 805          * page 0), so it doesn't need to be aliased into user space.
 806          */
 807
 808         map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
 809                   PAGE_SIZE, PAGE_GATEWAY, 1);
 810 }
 811
 812 #ifdef CONFIG_HPUX
 813 void
 814 map_hpux_gateway_page(struct task_struct *tsk, struct mm_struct *mm)
 815 {
 816         pgd_t *pg_dir;
 817         pmd_t *pmd;
 818         pte_t *pg_table;
 819         unsigned long start_pmd;
 820         unsigned long start_pte;
 821         unsigned long address;
 822         unsigned long hpux_gw_page_addr;
 823         /* FIXME: This is 'const' in order to trick the compiler
 824            into not treating it as DP-relative data. */
 825         extern void * const hpux_gateway_page;
 826
 827         hpux_gw_page_addr = HPUX_GATEWAY_ADDR & PAGE_MASK;
 828
 829         /*
 830          * Setup HP-UX Gateway page.
 831          *
 832          * The HP-UX gateway page resides in the user address space,
 833          * so it needs to be aliased into each process.
 834          */
 835
 836         pg_dir = pgd_offset(mm,hpux_gw_page_addr);
 837
 838 #if PTRS_PER_PMD == 1
 839         start_pmd = 0;
 840 #else
 841         start_pmd = ((hpux_gw_page_addr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
 842 #endif
 843         start_pte = ((hpux_gw_page_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
 844
 845         address = __pa(&hpux_gateway_page);
 846 #if PTRS_PER_PMD == 1
 847         pmd = (pmd_t *)__pa(pg_dir);
 848 #else
 849         pmd = (pmd_t *) pgd_address(*pg_dir);
 850
 851         /*
 852          * pmd is physical at this point
 853          */
 854
 855         if (!pmd) {
 856                 pmd = (pmd_t *) get_zeroed_page(GFP_KERNEL);
 857                 pmd = (pmd_t *) __pa(pmd);
 858         }
 859
 860         __pgd_val_set(*pg_dir, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pmd);
 861 #endif
 862         /* now change pmd to kernel virtual addresses */
 863
 864         pmd = (pmd_t *)__va(pmd) + start_pmd;
 865
 866         /*
 867          * pg_table is physical at this point
 868          */
 869
 870         pg_table = (pte_t *) pmd_address(*pmd);
 871         if (!pg_table)
 872                 pg_table = (pte_t *) __pa(get_zeroed_page(GFP_KERNEL));
 873
 874         __pmd_val_set(*pmd, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pg_table);
 875
 876         /* now change pg_table to kernel virtual addresses */
 877
 878         pg_table = (pte_t *) __va(pg_table) + start_pte;
 879         set_pte(pg_table, __mk_pte(address, PAGE_GATEWAY));
 880 }
 881 EXPORT_SYMBOL(map_hpux_gateway_page);
 882 #endif
 883
 884 void __init paging_init(void)
 885 {
 886         int i;
 887
 888         setup_bootmem();
 889         pagetable_init();
 890         gateway_init();
 891         flush_cache_all_local(); /* start with known state */
 892         flush_tlb_all_local(NULL);
 893
 894         for (i = 0; i < npmem_ranges; i++) {
 895                 unsigned long zones_size[MAX_NR_ZONES] = { 0, };
 896
 897                 zones_size[ZONE_NORMAL] = pmem_ranges[i].pages;
 898
 899 #ifdef CONFIG_DISCONTIGMEM
 900                 /* Need to initialize the pfnnid_map before we can initialize
 901                    the zone */
 902                 {
 903                     int j;
 904                     for (j = (pmem_ranges[i].start_pfn >> PFNNID_SHIFT);
 905                          j <= ((pmem_ranges[i].start_pfn + pmem_ranges[i].pages) >> PFNNID_SHIFT);
 906                          j++) {
 907                         pfnnid_map[j] = i;
 908                     }
 909                 }
 910 #endif
 911
 912                 free_area_init_node(i, zones_size,
 913                                 pmem_ranges[i].start_pfn, NULL);
 914         }
 915 }
 916
 917 #ifdef CONFIG_PA20
 918
 919 /*
 920  * Currently, all PA20 chips have 18 bit protection IDs, which is the
 921  * limiting factor (space ids are 32 bits).
 922  */
 923
 924 #define NR_SPACE_IDS 262144
 925
 926 #else
 927
 928 /*
 929  * Currently we have a one-to-one relationship between space IDs and
 930  * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
 931  * support 15 bit protection IDs, so that is the limiting factor.
 932  * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
 933  * probably not worth the effort for a special case here.
 934  */
 935
 936 #define NR_SPACE_IDS 32768
 937
 938 #endif  /* !CONFIG_PA20 */
 939
 940 #define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
 941 #define SID_ARRAY_SIZE  (NR_SPACE_IDS / (8 * sizeof(long)))
 942
 943 static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
 944 static unsigned long dirty_space_id[SID_ARRAY_SIZE];
 945 static unsigned long space_id_index;
 946 static unsigned long free_space_ids = NR_SPACE_IDS - 1;
 947 static unsigned long dirty_space_ids = 0;
 948
 949 static DEFINE_SPINLOCK(sid_lock);
 950
 951 unsigned long alloc_sid(void)
 952 {
 953         unsigned long index;
 954
 955         spin_lock(&sid_lock);
 956
 957         if (free_space_ids == 0) {
 958                 if (dirty_space_ids != 0) {
 959                         spin_unlock(&sid_lock);
 960                         flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
 961                         spin_lock(&sid_lock);
 962                 }
 963                 BUG_ON(free_space_ids == 0);
 964         }
 965
 966         free_space_ids--;
 967
 968         index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
 969         space_id[index >> SHIFT_PER_LONG] |= (1L << (index & (BITS_PER_LONG - 1)));
 970         space_id_index = index;
 971
 972         spin_unlock(&sid_lock);
 973
 974         return index << SPACEID_SHIFT;
 975 }
 976
 977 void free_sid(unsigned long spaceid)
 978 {
 979         unsigned long index = spaceid >> SPACEID_SHIFT;
 980         unsigned long *dirty_space_offset;
 981
 982         dirty_space_offset = dirty_space_id + (index >> SHIFT_PER_LONG);
 983         index &= (BITS_PER_LONG - 1);
 984
 985         spin_lock(&sid_lock);
 986
 987         BUG_ON(*dirty_space_offset & (1L << index)); /* attempt to free space id twice */
 988
 989         *dirty_space_offset |= (1L << index);
 990         dirty_space_ids++;
 991
 992         spin_unlock(&sid_lock);
 993 }
 994
 995
 996 #ifdef CONFIG_SMP
 997 static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
 998 {
 999         int i;
1000
1001         /* NOTE: sid_lock must be held upon entry */
1002
1003         *ndirtyptr = dirty_space_ids;
1004         if (dirty_space_ids != 0) {
1005             for (i = 0; i < SID_ARRAY_SIZE; i++) {
1006                 dirty_array[i] = dirty_space_id[i];
1007                 dirty_space_id[i] = 0;
1008             }
1009             dirty_space_ids = 0;
1010         }
1011
1012         return;
1013 }
1014
1015 static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
1016 {
1017         int i;
1018
1019         /* NOTE: sid_lock must be held upon entry */
1020
1021         if (ndirty != 0) {
1022                 for (i = 0; i < SID_ARRAY_SIZE; i++) {
1023                         space_id[i] ^= dirty_array[i];
1024                 }
1025
1026                 free_space_ids += ndirty;
1027                 space_id_index = 0;
1028         }
1029 }
1030
1031 #else /* CONFIG_SMP */
1032
1033 static void recycle_sids(void)
1034 {
1035         int i;
1036
1037         /* NOTE: sid_lock must be held upon entry */
1038
1039         if (dirty_space_ids != 0) {
1040                 for (i = 0; i < SID_ARRAY_SIZE; i++) {
1041                         space_id[i] ^= dirty_space_id[i];
1042                         dirty_space_id[i] = 0;
1043                 }
1044
1045                 free_space_ids += dirty_space_ids;
1046                 dirty_space_ids = 0;
1047                 space_id_index = 0;
1048         }
1049 }
1050 #endif
1051
1052 /*
1053  * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
1054  * purged, we can safely reuse the space ids that were released but
1055  * not flushed from the tlb.
1056  */
1057
1058 #ifdef CONFIG_SMP
1059
1060 static unsigned long recycle_ndirty;
1061 static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
1062 static unsigned int recycle_inuse;
1063
1064 void flush_tlb_all(void)
1065 {
1066         int do_recycle;
1067
1068         do_recycle = 0;
1069         spin_lock(&sid_lock);
1070         if (dirty_space_ids > RECYCLE_THRESHOLD) {
1071             BUG_ON(recycle_inuse);  /* FIXME: Use a semaphore/wait queue here */
1072             get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
1073             recycle_inuse++;
1074             do_recycle++;
1075         }
1076         spin_unlock(&sid_lock);
1077         on_each_cpu(flush_tlb_all_local, NULL, 1);
1078         if (do_recycle) {
1079             spin_lock(&sid_lock);
1080             recycle_sids(recycle_ndirty,recycle_dirty_array);
1081             recycle_inuse = 0;
1082             spin_unlock(&sid_lock);
1083         }
1084 }
1085 #else
1086 void flush_tlb_all(void)
1087 {
1088         spin_lock(&sid_lock);
1089         flush_tlb_all_local(NULL);
1090         recycle_sids();
1091         spin_unlock(&sid_lock);
1092 }
1093 #endif
1094
1095 #ifdef CONFIG_BLK_DEV_INITRD
1096 void free_initrd_mem(unsigned long start, unsigned long end)
1097 {
1098         if (start >= end)
1099                 return;
1100         printk(KERN_INFO "Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
1101         for (; start < end; start += PAGE_SIZE) {
1102                 ClearPageReserved(virt_to_page(start));
1103                 init_page_count(virt_to_page(start));
1104                 free_page(start);
1105                 num_physpages++;
1106                 totalram_pages++;
1107         }
1108 }
1109 #endif