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  *
  10  */
  11
  12 #include <linux/config.h>
  13
  14 #include <linux/module.h>
  15 #include <linux/mm.h>
  16 #include <linux/bootmem.h>
  17 #include <linux/delay.h>
  18 #include <linux/init.h>
  19 #include <linux/pci.h>          /* for hppa_dma_ops and pcxl_dma_ops */
  20 #include <linux/initrd.h>
  21 #include <linux/swap.h>
  22 #include <linux/unistd.h>
  23 #include <linux/nodemask.h>     /* for node_online_map */
  24
  25 #include <asm/pgalloc.h>
  26 #include <asm/tlb.h>
  27 #include <asm/pdc_chassis.h>
  28 #include <asm/mmzone.h>
  29
  30 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  31
  32 extern char _text;      /* start of kernel code, defined by linker */
  33 extern int  data_start;
  34 extern char _end;       /* end of BSS, defined by linker */
  35 extern char __init_begin, __init_end;
  36
  37 #ifdef CONFIG_DISCONTIGMEM
  38 struct node_map_data node_data[MAX_NUMNODES];
  39 bootmem_data_t bmem_data[MAX_NUMNODES];
  40 unsigned char pfnnid_map[PFNNID_MAP_MAX];
  41 #endif
  42
  43 static struct resource data_resource = {
  44         .name   = "Kernel data",
  45         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM,
  46 };
  47
  48 static struct resource code_resource = {
  49         .name   = "Kernel code",
  50         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM,
  51 };
  52
  53 static struct resource pdcdata_resource = {
  54         .name   = "PDC data (Page Zero)",
  55         .start  = 0,
  56         .end    = 0x9ff,
  57         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM,
  58 };
  59
  60 static struct resource sysram_resources[MAX_PHYSMEM_RANGES];
  61
  62 static unsigned long max_pfn;
  63
  64 /* The following array is initialized from the firmware specific
  65  * information retrieved in kernel/inventory.c.
  66  */
  67
  68 physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES];
  69 int npmem_ranges;
  70
  71 #ifdef __LP64__
  72 #define MAX_MEM         (~0UL)
  73 #else /* !__LP64__ */
  74 #define MAX_MEM         (3584U*1024U*1024U)
  75 #endif /* !__LP64__ */
  76
  77 static unsigned long mem_limit = MAX_MEM;
  78
  79 static void __init mem_limit_func(void)
  80 {
  81         char *cp, *end;
  82         unsigned long limit;
  83         extern char saved_command_line[];
  84
  85         /* We need this before __setup() functions are called */
  86
  87         limit = MAX_MEM;
  88         for (cp = saved_command_line; *cp; ) {
  89                 if (memcmp(cp, "mem=", 4) == 0) {
  90                         cp += 4;
  91                         limit = memparse(cp, &end);
  92                         if (end != cp)
  93                                 break;
  94                         cp = end;
  95                 } else {
  96                         while (*cp != ' ' && *cp)
  97                                 ++cp;
  98                         while (*cp == ' ')
  99                                 ++cp;
 100                 }
 101         }
 102
 103         if (limit < mem_limit)
 104                 mem_limit = limit;
 105 }
 106
 107 #define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
 108
 109 static void __init setup_bootmem(void)
 110 {
 111         unsigned long bootmap_size;
 112         unsigned long mem_max;
 113         unsigned long bootmap_pages;
 114         unsigned long bootmap_start_pfn;
 115         unsigned long bootmap_pfn;
 116 #ifndef CONFIG_DISCONTIGMEM
 117         physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1];
 118         int npmem_holes;
 119 #endif
 120         int i, sysram_resource_count;
 121
 122         disable_sr_hashing(); /* Turn off space register hashing */
 123
 124         /*
 125          * Sort the ranges. Since the number of ranges is typically
 126          * small, and performance is not an issue here, just do
 127          * a simple insertion sort.
 128          */
 129
 130         for (i = 1; i < npmem_ranges; i++) {
 131                 int j;
 132
 133                 for (j = i; j > 0; j--) {
 134                         unsigned long tmp;
 135
 136                         if (pmem_ranges[j-1].start_pfn <
 137                             pmem_ranges[j].start_pfn) {
 138
 139                                 break;
 140                         }
 141                         tmp = pmem_ranges[j-1].start_pfn;
 142                         pmem_ranges[j-1].start_pfn = pmem_ranges[j].start_pfn;
 143                         pmem_ranges[j].start_pfn = tmp;
 144                         tmp = pmem_ranges[j-1].pages;
 145                         pmem_ranges[j-1].pages = pmem_ranges[j].pages;
 146                         pmem_ranges[j].pages = tmp;
 147                 }
 148         }
 149
 150 #ifndef CONFIG_DISCONTIGMEM
 151         /*
 152          * Throw out ranges that are too far apart (controlled by
 153          * MAX_GAP).
 154          */
 155
 156         for (i = 1; i < npmem_ranges; i++) {
 157                 if (pmem_ranges[i].start_pfn -
 158                         (pmem_ranges[i-1].start_pfn +
 159                          pmem_ranges[i-1].pages) > MAX_GAP) {
 160                         npmem_ranges = i;
 161                         printk("Large gap in memory detected (%ld pages). "
 162                                "Consider turning on CONFIG_DISCONTIGMEM\n",
 163                                pmem_ranges[i].start_pfn -
 164                                (pmem_ranges[i-1].start_pfn +
 165                                 pmem_ranges[i-1].pages));
 166                         break;
 167                 }
 168         }
 169 #endif
 170
 171         if (npmem_ranges > 1) {
 172
 173                 /* Print the memory ranges */
 174
 175                 printk(KERN_INFO "Memory Ranges:\n");
 176
 177                 for (i = 0; i < npmem_ranges; i++) {
 178                         unsigned long start;
 179                         unsigned long size;
 180
 181                         size = (pmem_ranges[i].pages << PAGE_SHIFT);
 182                         start = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
 183                         printk(KERN_INFO "%2d) Start 0x%016lx End 0x%016lx Size %6ld Mb\n",
 184                                 i,start, start + (size - 1), size >> 20);
 185                 }
 186         }
 187
 188         sysram_resource_count = npmem_ranges;
 189         for (i = 0; i < sysram_resource_count; i++) {
 190                 struct resource *res = &sysram_resources[i];
 191                 res->name = "System RAM";
 192                 res->start = pmem_ranges[i].start_pfn << PAGE_SHIFT;
 193                 res->end = res->start + (pmem_ranges[i].pages << PAGE_SHIFT)-1;
 194                 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
 195                 request_resource(&iomem_resource, res);
 196         }
 197
 198         /*
 199          * For 32 bit kernels we limit the amount of memory we can
 200          * support, in order to preserve enough kernel address space
 201          * for other purposes. For 64 bit kernels we don't normally
 202          * limit the memory, but this mechanism can be used to
 203          * artificially limit the amount of memory (and it is written
 204          * to work with multiple memory ranges).
 205          */
 206
 207         mem_limit_func();       /* check for "mem=" argument */
 208
 209         mem_max = 0;
 210         num_physpages = 0;
 211         for (i = 0; i < npmem_ranges; i++) {
 212                 unsigned long rsize;
 213
 214                 rsize = pmem_ranges[i].pages << PAGE_SHIFT;
 215                 if ((mem_max + rsize) > mem_limit) {
 216                         printk(KERN_WARNING "Memory truncated to %ld Mb\n", mem_limit >> 20);
 217                         if (mem_max == mem_limit)
 218                                 npmem_ranges = i;
 219                         else {
 220                                 pmem_ranges[i].pages =   (mem_limit >> PAGE_SHIFT)
 221                                                        - (mem_max >> PAGE_SHIFT);
 222                                 npmem_ranges = i + 1;
 223                                 mem_max = mem_limit;
 224                         }
 225                 num_physpages += pmem_ranges[i].pages;
 226                         break;
 227                 }
 228             num_physpages += pmem_ranges[i].pages;
 229                 mem_max += rsize;
 230         }
 231
 232         printk(KERN_INFO "Total Memory: %ld Mb\n",mem_max >> 20);
 233
 234 #ifndef CONFIG_DISCONTIGMEM
 235         /* Merge the ranges, keeping track of the holes */
 236
 237         {
 238                 unsigned long end_pfn;
 239                 unsigned long hole_pages;
 240
 241                 npmem_holes = 0;
 242                 end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages;
 243                 for (i = 1; i < npmem_ranges; i++) {
 244
 245                         hole_pages = pmem_ranges[i].start_pfn - end_pfn;
 246                         if (hole_pages) {
 247                                 pmem_holes[npmem_holes].start_pfn = end_pfn;
 248                                 pmem_holes[npmem_holes++].pages = hole_pages;
 249                                 end_pfn += hole_pages;
 250                         }
 251                         end_pfn += pmem_ranges[i].pages;
 252                 }
 253
 254                 pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn;
 255                 npmem_ranges = 1;
 256         }
 257 #endif
 258
 259         bootmap_pages = 0;
 260         for (i = 0; i < npmem_ranges; i++)
 261                 bootmap_pages += bootmem_bootmap_pages(pmem_ranges[i].pages);
 262
 263         bootmap_start_pfn = PAGE_ALIGN(__pa((unsigned long) &_end)) >> PAGE_SHIFT;
 264
 265 #ifdef CONFIG_DISCONTIGMEM
 266         for (i = 0; i < MAX_PHYSMEM_RANGES; i++) {
 267                 memset(NODE_DATA(i), 0, sizeof(pg_data_t));
 268                 NODE_DATA(i)->bdata = &bmem_data[i];
 269         }
 270         memset(pfnnid_map, 0xff, sizeof(pfnnid_map));
 271
 272         numnodes = npmem_ranges;
 273
 274         for (i = 0; i < npmem_ranges; i++)
 275                 node_set_online(i);
 276 #endif
 277
 278         /*
 279          * Initialize and free the full range of memory in each range.
 280          * Note that the only writing these routines do are to the bootmap,
 281          * and we've made sure to locate the bootmap properly so that they
 282          * won't be writing over anything important.
 283          */
 284
 285         bootmap_pfn = bootmap_start_pfn;
 286         max_pfn = 0;
 287         for (i = 0; i < npmem_ranges; i++) {
 288                 unsigned long start_pfn;
 289                 unsigned long npages;
 290
 291                 start_pfn = pmem_ranges[i].start_pfn;
 292                 npages = pmem_ranges[i].pages;
 293
 294                 bootmap_size = init_bootmem_node(NODE_DATA(i),
 295                                                 bootmap_pfn,
 296                                                 start_pfn,
 297                                                 (start_pfn + npages) );
 298                 free_bootmem_node(NODE_DATA(i),
 299                                   (start_pfn << PAGE_SHIFT),
 300                                   (npages << PAGE_SHIFT) );
 301                 bootmap_pfn += (bootmap_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 302                 if ((start_pfn + npages) > max_pfn)
 303                         max_pfn = start_pfn + npages;
 304         }
 305
 306         if ((bootmap_pfn - bootmap_start_pfn) != bootmap_pages) {
 307                 printk(KERN_WARNING "WARNING! bootmap sizing is messed up!\n");
 308                 BUG();
 309         }
 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 + PDC_CONSOLE_IO_IODC_SIZE));
 317         reserve_bootmem_node(NODE_DATA(0),__pa((unsigned long)&_text),
 318                         (unsigned long)(&_end - &_text));
 319         reserve_bootmem_node(NODE_DATA(0), (bootmap_start_pfn << PAGE_SHIFT),
 320                         ((bootmap_pfn - bootmap_start_pfn) << PAGE_SHIFT));
 321
 322 #ifndef CONFIG_DISCONTIGMEM
 323
 324         /* reserve the holes */
 325
 326         for (i = 0; i < npmem_holes; i++) {
 327                 reserve_bootmem_node(NODE_DATA(0),
 328                                 (pmem_holes[i].start_pfn << PAGE_SHIFT),
 329                                 (pmem_holes[i].pages << PAGE_SHIFT));
 330         }
 331 #endif
 332
 333 #ifdef CONFIG_BLK_DEV_INITRD
 334         if (initrd_start) {
 335                 printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
 336                 if (__pa(initrd_start) < mem_max) {
 337                         unsigned long initrd_reserve;
 338
 339                         if (__pa(initrd_end) > mem_max) {
 340                                 initrd_reserve = mem_max - __pa(initrd_start);
 341                         } else {
 342                                 initrd_reserve = initrd_end - initrd_start;
 343                         }
 344                         initrd_below_start_ok = 1;
 345                         printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
 346
 347                         reserve_bootmem_node(NODE_DATA(0),__pa(initrd_start), initrd_reserve);
 348                 }
 349         }
 350 #endif
 351
 352         data_resource.start =  virt_to_phys(&data_start);
 353         data_resource.end = virt_to_phys(&_end)-1;
 354         code_resource.start = virt_to_phys(&_text);
 355         code_resource.end = virt_to_phys(&data_start)-1;
 356
 357         /* We don't know which region the kernel will be in, so try
 358          * all of them.
 359          */
 360         for (i = 0; i < sysram_resource_count; i++) {
 361                 struct resource *res = &sysram_resources[i];
 362                 request_resource(res, &code_resource);
 363                 request_resource(res, &data_resource);
 364         }
 365         request_resource(&sysram_resources[0], &pdcdata_resource);
 366 }
 367
 368 void free_initmem(void)
 369 {
 370         /* FIXME: */
 371 #if 0
 372         printk(KERN_INFO "NOT FREEING INITMEM (%dk)\n",
 373                         (&__init_end - &__init_begin) >> 10);
 374         return;
 375 #else
 376         unsigned long addr;
 377
 378         printk(KERN_INFO "Freeing unused kernel memory: ");
 379
 380 #if 1
 381         /* Attempt to catch anyone trying to execute code here
 382          * by filling the page with BRK insns.
 383          *
 384          * If we disable interrupts for all CPUs, then IPI stops working.
 385          * Kinda breaks the global cache flushing.
 386          */
 387         local_irq_disable();
 388
 389         memset(&__init_begin, 0x00,
 390                 (unsigned long)&__init_end - (unsigned long)&__init_begin);
 391
 392         flush_data_cache();
 393         asm volatile("sync" : : );
 394         flush_icache_range((unsigned long)&__init_begin, (unsigned long)&__init_end);
 395         asm volatile("sync" : : );
 396
 397         local_irq_enable();
 398 #endif
 399
 400         addr = (unsigned long)(&__init_begin);
 401         for (; addr < (unsigned long)(&__init_end); addr += PAGE_SIZE) {
 402                 ClearPageReserved(virt_to_page(addr));
 403                 set_page_count(virt_to_page(addr), 1);
 404                 free_page(addr);
 405                 num_physpages++;
 406                 totalram_pages++;
 407         }
 408
 409         /* set up a new led state on systems shipped LED State panel */
 410         pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
 411
 412         printk("%luk freed\n", (unsigned long)(&__init_end - &__init_begin) >> 10);
 413 #endif
 414 }
 415
 416 /*
 417  * Just an arbitrary offset to serve as a "hole" between mapping areas
 418  * (between top of physical memory and a potential pcxl dma mapping
 419  * area, and below the vmalloc mapping area).
 420  *
 421  * The current 32K value just means that there will be a 32K "hole"
 422  * between mapping areas. That means that  any out-of-bounds memory
 423  * accesses will hopefully be caught. The vmalloc() routines leaves
 424  * a hole of 4kB between each vmalloced area for the same reason.
 425  */
 426
 427  /* Leave room for gateway page expansion */
 428 #if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
 429 #error KERNEL_MAP_START is in gateway reserved region
 430 #endif
 431 #define MAP_START (KERNEL_MAP_START)
 432
 433 #define VM_MAP_OFFSET  (32*1024)
 434 #define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
 435                                      & ~(VM_MAP_OFFSET-1)))
 436
 437 void *vmalloc_start;
 438 EXPORT_SYMBOL(vmalloc_start);
 439
 440 #ifdef CONFIG_PA11
 441 unsigned long pcxl_dma_start;
 442 #endif
 443
 444 void __init mem_init(void)
 445 {
 446         high_memory = __va((max_pfn << PAGE_SHIFT));
 447
 448 #ifndef CONFIG_DISCONTIGMEM
 449         max_mapnr = page_to_pfn(virt_to_page(high_memory - 1)) + 1;
 450         mem_map = zone_table[ZONE_DMA]->zone_mem_map;
 451         totalram_pages += free_all_bootmem();
 452 #else
 453         {
 454                 int i;
 455
 456                 for (i = 0; i < npmem_ranges; i++)
 457                         totalram_pages += free_all_bootmem_node(NODE_DATA(i));
 458         }
 459 #endif
 460
 461         printk(KERN_INFO "Memory: %luk available\n", num_physpages << (PAGE_SHIFT-10));
 462
 463 #ifdef CONFIG_PA11
 464         if (hppa_dma_ops == &pcxl_dma_ops) {
 465                 pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
 466                 vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start + PCXL_DMA_MAP_SIZE);
 467         } else {
 468                 pcxl_dma_start = 0;
 469                 vmalloc_start = SET_MAP_OFFSET(MAP_START);
 470         }
 471 #else
 472         vmalloc_start = SET_MAP_OFFSET(MAP_START);
 473 #endif
 474
 475 }
 476
 477 int do_check_pgt_cache(int low, int high)
 478 {
 479         return 0;
 480 }
 481
 482 unsigned long *empty_zero_page;
 483
 484 void show_mem(void)
 485 {
 486         int i,free = 0,total = 0,reserved = 0;
 487         int shared = 0, cached = 0;
 488
 489         printk(KERN_INFO "Mem-info:\n");
 490         show_free_areas();
 491         printk(KERN_INFO "Free swap:     %6ldkB\n",
 492                                 nr_swap_pages<<(PAGE_SHIFT-10));
 493 #ifndef CONFIG_DISCONTIGMEM
 494         i = max_mapnr;
 495         while (i-- > 0) {
 496                 total++;
 497                 if (PageReserved(mem_map+i))
 498                         reserved++;
 499                 else if (PageSwapCache(mem_map+i))
 500                         cached++;
 501                 else if (!page_count(&mem_map[i]))
 502                         free++;
 503                 else
 504                         shared += page_count(&mem_map[i]) - 1;
 505         }
 506 #else
 507         for (i = 0; i < npmem_ranges; i++) {
 508                 int j;
 509
 510                 for (j = node_start_pfn(i); j < node_end_pfn(i); j++) {
 511                         struct page *p;
 512
 513                         p = node_mem_map(i) + j - node_start_pfn(i);
 514
 515                         total++;
 516                         if (PageReserved(p))
 517                                 reserved++;
 518                         else if (PageSwapCache(p))
 519                                 cached++;
 520                         else if (!page_count(p))
 521                                 free++;
 522                         else
 523                                 shared += page_count(p) - 1;
 524                 }
 525         }
 526 #endif
 527         printk(KERN_INFO "%d pages of RAM\n", total);
 528         printk(KERN_INFO "%d reserved pages\n", reserved);
 529         printk(KERN_INFO "%d pages shared\n", shared);
 530         printk(KERN_INFO "%d pages swap cached\n", cached);
 531
 532
 533 #ifdef CONFIG_DISCONTIGMEM
 534         {
 535                 struct zonelist *zl;
 536                 int i, j, k;
 537
 538                 for (i = 0; i < npmem_ranges; i++) {
 539                         for (j = 0; j < MAX_NR_ZONES; j++) {
 540                                 zl = NODE_DATA(i)->node_zonelists + j;
 541
 542                                 printk("Zone list for zone %d on node %d: ", j, i);
 543                                 for (k = 0; zl->zones[k] != NULL; k++)
 544                                         printk("[%d/%s] ", zl->zones[k]->zone_pgdat->node_id, zl->zones[k]->name);
 545                                 printk("\n");
 546                         }
 547                 }
 548         }
 549 #endif
 550 }
 551
 552
 553 static void __init map_pages(unsigned long start_vaddr, unsigned long start_paddr, unsigned long size, pgprot_t pgprot)
 554 {
 555         pgd_t *pg_dir;
 556         pmd_t *pmd;
 557         pte_t *pg_table;
 558         unsigned long end_paddr;
 559         unsigned long start_pmd;
 560         unsigned long start_pte;
 561         unsigned long tmp1;
 562         unsigned long tmp2;
 563         unsigned long address;
 564         unsigned long ro_start;
 565         unsigned long ro_end;
 566         unsigned long fv_addr;
 567         unsigned long gw_addr;
 568         extern const unsigned long fault_vector_20;
 569         extern void * const linux_gateway_page;
 570
 571         ro_start = __pa((unsigned long)&_text);
 572         ro_end   = __pa((unsigned long)&data_start);
 573         fv_addr  = __pa((unsigned long)&fault_vector_20) & PAGE_MASK;
 574         gw_addr  = __pa((unsigned long)&linux_gateway_page) & PAGE_MASK;
 575
 576         end_paddr = start_paddr + size;
 577
 578         pg_dir = pgd_offset_k(start_vaddr);
 579
 580 #if PTRS_PER_PMD == 1
 581         start_pmd = 0;
 582 #else
 583         start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
 584 #endif
 585         start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
 586
 587         address = start_paddr;
 588         while (address < end_paddr) {
 589 #if PTRS_PER_PMD == 1
 590                 pmd = (pmd_t *)__pa(pg_dir);
 591 #else
 592                 pmd = (pmd_t *)pgd_address(*pg_dir);
 593
 594                 /*
 595                  * pmd is physical at this point
 596                  */
 597
 598                 if (!pmd) {
 599                         pmd = (pmd_t *) alloc_bootmem_low_pages_node(NODE_DATA(0),PAGE_SIZE << PMD_ORDER);
 600                         pmd = (pmd_t *) __pa(pmd);
 601                 }
 602
 603                 pgd_populate(NULL, pg_dir, __va(pmd));
 604 #endif
 605                 pg_dir++;
 606
 607                 /* now change pmd to kernel virtual addresses */
 608
 609                 pmd = (pmd_t *)__va(pmd) + start_pmd;
 610                 for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++,pmd++) {
 611
 612                         /*
 613                          * pg_table is physical at this point
 614                          */
 615
 616                         pg_table = (pte_t *)pmd_address(*pmd);
 617                         if (!pg_table) {
 618                                 pg_table = (pte_t *)
 619                                         alloc_bootmem_low_pages_node(NODE_DATA(0),PAGE_SIZE);
 620                                 pg_table = (pte_t *) __pa(pg_table);
 621                         }
 622
 623                         pmd_populate_kernel(NULL, pmd, __va(pg_table));
 624
 625                         /* now change pg_table to kernel virtual addresses */
 626
 627                         pg_table = (pte_t *) __va(pg_table) + start_pte;
 628                         for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++,pg_table++) {
 629                                 pte_t pte;
 630
 631                                 /*
 632                                  * Map the fault vector writable so we can
 633                                  * write the HPMC checksum.
 634                                  */
 635                                 if (address >= ro_start && address < ro_end
 636                                                         && address != fv_addr
 637                                                         && address != gw_addr)
 638                                     pte = __mk_pte(address, PAGE_KERNEL_RO);
 639                                 else
 640                                     pte = __mk_pte(address, pgprot);
 641
 642                                 if (address >= end_paddr)
 643                                         pte_val(pte) = 0;
 644
 645                                 set_pte(pg_table, pte);
 646
 647                                 address += PAGE_SIZE;
 648                         }
 649                         start_pte = 0;
 650
 651                         if (address >= end_paddr)
 652                             break;
 653                 }
 654                 start_pmd = 0;
 655         }
 656 }
 657
 658 /*
 659  * pagetable_init() sets up the page tables
 660  *
 661  * Note that gateway_init() places the Linux gateway page at page 0.
 662  * Since gateway pages cannot be dereferenced this has the desirable
 663  * side effect of trapping those pesky NULL-reference errors in the
 664  * kernel.
 665  */
 666 static void __init pagetable_init(void)
 667 {
 668         int range;
 669
 670         /* Map each physical memory range to its kernel vaddr */
 671
 672         for (range = 0; range < npmem_ranges; range++) {
 673                 unsigned long start_paddr;
 674                 unsigned long end_paddr;
 675                 unsigned long size;
 676
 677                 start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
 678                 end_paddr = start_paddr + (pmem_ranges[range].pages << PAGE_SHIFT);
 679                 size = pmem_ranges[range].pages << PAGE_SHIFT;
 680
 681                 map_pages((unsigned long)__va(start_paddr), start_paddr,
 682                         size, PAGE_KERNEL);
 683         }
 684
 685 #ifdef CONFIG_BLK_DEV_INITRD
 686         if (initrd_end && initrd_end > mem_limit) {
 687                 printk("initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
 688                 map_pages(initrd_start, __pa(initrd_start),
 689                         initrd_end - initrd_start, PAGE_KERNEL);
 690         }
 691 #endif
 692
 693         empty_zero_page = alloc_bootmem_pages(PAGE_SIZE);
 694         memset(empty_zero_page, 0, PAGE_SIZE);
 695 }
 696
 697 static void __init gateway_init(void)
 698 {
 699         unsigned long linux_gateway_page_addr;
 700         /* FIXME: This is 'const' in order to trick the compiler
 701            into not treating it as DP-relative data. */
 702         extern void * const linux_gateway_page;
 703
 704         linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
 705
 706         /*
 707          * Setup Linux Gateway page.
 708          *
 709          * The Linux gateway page will reside in kernel space (on virtual
 710          * page 0), so it doesn't need to be aliased into user space.
 711          */
 712
 713         map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
 714                 PAGE_SIZE, PAGE_GATEWAY);
 715 }
 716
 717 #ifdef CONFIG_HPUX
 718 void
 719 map_hpux_gateway_page(struct task_struct *tsk, struct mm_struct *mm)
 720 {
 721         pgd_t *pg_dir;
 722         pmd_t *pmd;
 723         pte_t *pg_table;
 724         unsigned long start_pmd;
 725         unsigned long start_pte;
 726         unsigned long address;
 727         unsigned long hpux_gw_page_addr;
 728         /* FIXME: This is 'const' in order to trick the compiler
 729            into not treating it as DP-relative data. */
 730         extern void * const hpux_gateway_page;
 731
 732         hpux_gw_page_addr = HPUX_GATEWAY_ADDR & PAGE_MASK;
 733
 734         /*
 735          * Setup HP-UX Gateway page.
 736          *
 737          * The HP-UX gateway page resides in the user address space,
 738          * so it needs to be aliased into each process.
 739          */
 740
 741         pg_dir = pgd_offset(mm,hpux_gw_page_addr);
 742
 743 #if PTRS_PER_PMD == 1
 744         start_pmd = 0;
 745 #else
 746         start_pmd = ((hpux_gw_page_addr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
 747 #endif
 748         start_pte = ((hpux_gw_page_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
 749
 750         address = __pa(&hpux_gateway_page);
 751 #if PTRS_PER_PMD == 1
 752         pmd = (pmd_t *)__pa(pg_dir);
 753 #else
 754         pmd = (pmd_t *) pgd_address(*pg_dir);
 755
 756         /*
 757          * pmd is physical at this point
 758          */
 759
 760         if (!pmd) {
 761                 pmd = (pmd_t *) get_zeroed_page(GFP_KERNEL);
 762                 pmd = (pmd_t *) __pa(pmd);
 763         }
 764
 765         __pgd_val_set(*pg_dir, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pmd);
 766 #endif
 767         /* now change pmd to kernel virtual addresses */
 768
 769         pmd = (pmd_t *)__va(pmd) + start_pmd;
 770
 771         /*
 772          * pg_table is physical at this point
 773          */
 774
 775         pg_table = (pte_t *) pmd_address(*pmd);
 776         if (!pg_table)
 777                 pg_table = (pte_t *) __pa(get_zeroed_page(GFP_KERNEL));
 778
 779         __pmd_val_set(*pmd, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pg_table);
 780
 781         /* now change pg_table to kernel virtual addresses */
 782
 783         pg_table = (pte_t *) __va(pg_table) + start_pte;
 784         set_pte(pg_table, __mk_pte(address, PAGE_GATEWAY));
 785 }
 786 EXPORT_SYMBOL(map_hpux_gateway_page);
 787 #endif
 788
 789 extern void flush_tlb_all_local(void);
 790
 791 void __init paging_init(void)
 792 {
 793         int i;
 794
 795         setup_bootmem();
 796         pagetable_init();
 797         gateway_init();
 798         flush_cache_all_local(); /* start with known state */
 799         flush_tlb_all_local();
 800
 801         for (i = 0; i < npmem_ranges; i++) {
 802                 unsigned long zones_size[MAX_NR_ZONES] = { 0, 0, 0 };
 803
 804                 /* We have an IOMMU, so all memory can go into a single
 805                    ZONE_DMA zone. */
 806                 zones_size[ZONE_DMA] = pmem_ranges[i].pages;
 807
 808 #ifdef CONFIG_DISCONTIGMEM
 809                 /* Need to initialize the pfnnid_map before we can initialize
 810                    the zone */
 811                 {
 812                     int j;
 813                     for (j = (pmem_ranges[i].start_pfn >> PFNNID_SHIFT);
 814                          j <= ((pmem_ranges[i].start_pfn + pmem_ranges[i].pages) >> PFNNID_SHIFT);
 815                          j++) {
 816                         pfnnid_map[j] = i;
 817                     }
 818                 }
 819 #endif
 820
 821                 free_area_init_node(i, NODE_DATA(i), zones_size,
 822                                 pmem_ranges[i].start_pfn, NULL);
 823         }
 824 }
 825
 826 #ifdef CONFIG_PA20
 827
 828 /*
 829  * Currently, all PA20 chips have 18 bit protection id's, which is the
 830  * limiting factor (space ids are 32 bits).
 831  */
 832
 833 #define NR_SPACE_IDS 262144
 834
 835 #else
 836
 837 /*
 838  * Currently we have a one-to-one relationship between space id's and
 839  * protection id's. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
 840  * support 15 bit protection id's, so that is the limiting factor.
 841  * PCXT' has 18 bit protection id's, but only 16 bit spaceids, so it's
 842  * probably not worth the effort for a special case here.
 843  */
 844
 845 #define NR_SPACE_IDS 32768
 846
 847 #endif  /* !CONFIG_PA20 */
 848
 849 #define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
 850 #define SID_ARRAY_SIZE  (NR_SPACE_IDS / (8 * sizeof(long)))
 851
 852 static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
 853 static unsigned long dirty_space_id[SID_ARRAY_SIZE];
 854 static unsigned long space_id_index;
 855 static unsigned long free_space_ids = NR_SPACE_IDS - 1;
 856 static unsigned long dirty_space_ids = 0;
 857
 858 static spinlock_t sid_lock = SPIN_LOCK_UNLOCKED;
 859
 860 unsigned long alloc_sid(void)
 861 {
 862         unsigned long index;
 863
 864         spin_lock(&sid_lock);
 865
 866         if (free_space_ids == 0) {
 867                 if (dirty_space_ids != 0) {
 868                         spin_unlock(&sid_lock);
 869                         flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
 870                         spin_lock(&sid_lock);
 871                 }
 872                 if (free_space_ids == 0)
 873                         BUG();
 874         }
 875
 876         free_space_ids--;
 877
 878         index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
 879         space_id[index >> SHIFT_PER_LONG] |= (1L << (index & (BITS_PER_LONG - 1)));
 880         space_id_index = index;
 881
 882         spin_unlock(&sid_lock);
 883
 884         return index << SPACEID_SHIFT;
 885 }
 886
 887 void free_sid(unsigned long spaceid)
 888 {
 889         unsigned long index = spaceid >> SPACEID_SHIFT;
 890         unsigned long *dirty_space_offset;
 891
 892         dirty_space_offset = dirty_space_id + (index >> SHIFT_PER_LONG);
 893         index &= (BITS_PER_LONG - 1);
 894
 895         spin_lock(&sid_lock);
 896
 897         if (*dirty_space_offset & (1L << index))
 898             BUG(); /* attempt to free space id twice */
 899
 900         *dirty_space_offset |= (1L << index);
 901         dirty_space_ids++;
 902
 903         spin_unlock(&sid_lock);
 904 }
 905
 906
 907 #ifdef CONFIG_SMP
 908 static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
 909 {
 910         int i;
 911
 912         /* NOTE: sid_lock must be held upon entry */
 913
 914         *ndirtyptr = dirty_space_ids;
 915         if (dirty_space_ids != 0) {
 916             for (i = 0; i < SID_ARRAY_SIZE; i++) {
 917                 dirty_array[i] = dirty_space_id[i];
 918                 dirty_space_id[i] = 0;
 919             }
 920             dirty_space_ids = 0;
 921         }
 922
 923         return;
 924 }
 925
 926 static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
 927 {
 928         int i;
 929
 930         /* NOTE: sid_lock must be held upon entry */
 931
 932         if (ndirty != 0) {
 933                 for (i = 0; i < SID_ARRAY_SIZE; i++) {
 934                         space_id[i] ^= dirty_array[i];
 935                 }
 936
 937                 free_space_ids += ndirty;
 938                 space_id_index = 0;
 939         }
 940 }
 941
 942 #else /* CONFIG_SMP */
 943
 944 static void recycle_sids(void)
 945 {
 946         int i;
 947
 948         /* NOTE: sid_lock must be held upon entry */
 949
 950         if (dirty_space_ids != 0) {
 951                 for (i = 0; i < SID_ARRAY_SIZE; i++) {
 952                         space_id[i] ^= dirty_space_id[i];
 953                         dirty_space_id[i] = 0;
 954                 }
 955
 956                 free_space_ids += dirty_space_ids;
 957                 dirty_space_ids = 0;
 958                 space_id_index = 0;
 959         }
 960 }
 961 #endif
 962
 963 /*
 964  * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
 965  * purged, we can safely reuse the space ids that were released but
 966  * not flushed from the tlb.
 967  */
 968
 969 #ifdef CONFIG_SMP
 970
 971 static unsigned long recycle_ndirty;
 972 static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
 973 static unsigned int recycle_inuse = 0;
 974
 975 void flush_tlb_all(void)
 976 {
 977         int do_recycle;
 978
 979         do_recycle = 0;
 980         spin_lock(&sid_lock);
 981         if (dirty_space_ids > RECYCLE_THRESHOLD) {
 982             if (recycle_inuse) {
 983                 BUG();  /* FIXME: Use a semaphore/wait queue here */
 984             }
 985             get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
 986             recycle_inuse++;
 987             do_recycle++;
 988         }
 989         spin_unlock(&sid_lock);
 990         on_each_cpu((void (*)(void *))flush_tlb_all_local, NULL, 1, 1);
 991         if (do_recycle) {
 992             spin_lock(&sid_lock);
 993             recycle_sids(recycle_ndirty,recycle_dirty_array);
 994             recycle_inuse = 0;
 995             spin_unlock(&sid_lock);
 996         }
 997 }
 998 #else
 999 void flush_tlb_all(void)
1000 {
1001         spin_lock(&sid_lock);
1002         flush_tlb_all_local();
1003         recycle_sids();
1004         spin_unlock(&sid_lock);
1005 }
1006 #endif
1007
1008 #ifdef CONFIG_BLK_DEV_INITRD
1009 void free_initrd_mem(unsigned long start, unsigned long end)
1010 {
1011 #if 0
1012         if (start < end)
1013                 printk(KERN_INFO "Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
1014         for (; start < end; start += PAGE_SIZE) {
1015                 ClearPageReserved(virt_to_page(start));
1016                 set_page_count(virt_to_page(start), 1);
1017                 free_page(start);
1018                 num_physpages++;
1019                 totalram_pages++;
1020         }
1021 #endif
1022 }
1023 #endif