arch/parisc/mm/init.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/arch/parisc/mm/init.c
   4  *
   5  *  Copyright (C) 1995  Linus Torvalds
   6  *  Copyright 1999 SuSE GmbH
   7  *    changed by Philipp Rumpf
   8  *  Copyright 1999 Philipp Rumpf (prumpf@tux.org)
   9  *  Copyright 2004 Randolph Chung (tausq@debian.org)
  10  *  Copyright 2006-2007 Helge Deller (deller@gmx.de)
  11  *
  12  */
  13
  14
  15 #include <linux/module.h>
  16 #include <linux/mm.h>
  17 #include <linux/bootmem.h>
  18 #include <linux/memblock.h>
  19 #include <linux/gfp.h>
  20 #include <linux/delay.h>
  21 #include <linux/init.h>
  22 #include <linux/pci.h>          /* for hppa_dma_ops and pcxl_dma_ops */
  23 #include <linux/initrd.h>
  24 #include <linux/swap.h>
  25 #include <linux/unistd.h>
  26 #include <linux/nodemask.h>     /* for node_online_map */
  27 #include <linux/pagemap.h>      /* for release_pages */
  28 #include <linux/compat.h>
  29
  30 #include <asm/pgalloc.h>
  31 #include <asm/pgtable.h>
  32 #include <asm/tlb.h>
  33 #include <asm/pdc_chassis.h>
  34 #include <asm/mmzone.h>
  35 #include <asm/sections.h>
  36 #include <asm/msgbuf.h>
  37
  38 extern int  data_start;
  39 extern void parisc_kernel_start(void);  /* Kernel entry point in head.S */
  40
  41 #if CONFIG_PGTABLE_LEVELS == 3
  42 /* NOTE: This layout exactly conforms to the hybrid L2/L3 page table layout
  43  * with the first pmd adjacent to the pgd and below it. gcc doesn't actually
  44  * guarantee that global objects will be laid out in memory in the same order
  45  * as the order of declaration, so put these in different sections and use
  46  * the linker script to order them. */
  47 pmd_t pmd0[PTRS_PER_PMD] __attribute__ ((__section__ (".data..vm0.pmd"), aligned(PAGE_SIZE)));
  48 #endif
  49
  50 pgd_t swapper_pg_dir[PTRS_PER_PGD] __attribute__ ((__section__ (".data..vm0.pgd"), aligned(PAGE_SIZE)));
  51 pte_t pg0[PT_INITIAL * PTRS_PER_PTE] __attribute__ ((__section__ (".data..vm0.pte"), aligned(PAGE_SIZE)));
  52
  53 #ifdef CONFIG_DISCONTIGMEM
  54 struct node_map_data node_data[MAX_NUMNODES] __read_mostly;
  55 signed char pfnnid_map[PFNNID_MAP_MAX] __read_mostly;
  56 #endif
  57
  58 static struct resource data_resource = {
  59         .name   = "Kernel data",
  60         .flags  = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
  61 };
  62
  63 static struct resource code_resource = {
  64         .name   = "Kernel code",
  65         .flags  = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
  66 };
  67
  68 static struct resource pdcdata_resource = {
  69         .name   = "PDC data (Page Zero)",
  70         .start  = 0,
  71         .end    = 0x9ff,
  72         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM,
  73 };
  74
  75 static struct resource sysram_resources[MAX_PHYSMEM_RANGES] __read_mostly;
  76
  77 /* The following array is initialized from the firmware specific
  78  * information retrieved in kernel/inventory.c.
  79  */
  80
  81 physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES] __read_mostly;
  82 int npmem_ranges __read_mostly;
  83
  84 /*
  85  * get_memblock() allocates pages via memblock.
  86  * We can't use memblock_find_in_range(0, KERNEL_INITIAL_SIZE) here since it
  87  * doesn't allocate from bottom to top which is needed because we only created
  88  * the initial mapping up to KERNEL_INITIAL_SIZE in the assembly bootup code.
  89  */
  90 static void * __init get_memblock(unsigned long size)
  91 {
  92         static phys_addr_t search_addr __initdata;
  93         phys_addr_t phys;
  94
  95         if (!search_addr)
  96                 search_addr = PAGE_ALIGN(__pa((unsigned long) &_end));
  97         search_addr = ALIGN(search_addr, size);
  98         while (!memblock_is_region_memory(search_addr, size) ||
  99                 memblock_is_region_reserved(search_addr, size)) {
 100                 search_addr += size;
 101         }
 102         phys = search_addr;
 103
 104         if (phys)
 105                 memblock_reserve(phys, size);
 106         else
 107                 panic("get_memblock() failed.\n");
 108
 109         memset(__va(phys), 0, size);
 110
 111         return __va(phys);
 112 }
 113
 114 #ifdef CONFIG_64BIT
 115 #define MAX_MEM         (~0UL)
 116 #else /* !CONFIG_64BIT */
 117 #define MAX_MEM         (3584U*1024U*1024U)
 118 #endif /* !CONFIG_64BIT */
 119
 120 static unsigned long mem_limit __read_mostly = MAX_MEM;
 121
 122 static void __init mem_limit_func(void)
 123 {
 124         char *cp, *end;
 125         unsigned long limit;
 126
 127         /* We need this before __setup() functions are called */
 128
 129         limit = MAX_MEM;
 130         for (cp = boot_command_line; *cp; ) {
 131                 if (memcmp(cp, "mem=", 4) == 0) {
 132                         cp += 4;
 133                         limit = memparse(cp, &end);
 134                         if (end != cp)
 135                                 break;
 136                         cp = end;
 137                 } else {
 138                         while (*cp != ' ' && *cp)
 139                                 ++cp;
 140                         while (*cp == ' ')
 141                                 ++cp;
 142                 }
 143         }
 144
 145         if (limit < mem_limit)
 146                 mem_limit = limit;
 147 }
 148
 149 #define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
 150
 151 static void __init setup_bootmem(void)
 152 {
 153         unsigned long mem_max;
 154 #ifndef CONFIG_DISCONTIGMEM
 155         physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1];
 156         int npmem_holes;
 157 #endif
 158         int i, sysram_resource_count;
 159
 160         disable_sr_hashing(); /* Turn off space register hashing */
 161
 162         /*
 163          * Sort the ranges. Since the number of ranges is typically
 164          * small, and performance is not an issue here, just do
 165          * a simple insertion sort.
 166          */
 167
 168         for (i = 1; i < npmem_ranges; i++) {
 169                 int j;
 170
 171                 for (j = i; j > 0; j--) {
 172                         unsigned long tmp;
 173
 174                         if (pmem_ranges[j-1].start_pfn <
 175                             pmem_ranges[j].start_pfn) {
 176
 177                                 break;
 178                         }
 179                         tmp = pmem_ranges[j-1].start_pfn;
 180                         pmem_ranges[j-1].start_pfn = pmem_ranges[j].start_pfn;
 181                         pmem_ranges[j].start_pfn = tmp;
 182                         tmp = pmem_ranges[j-1].pages;
 183                         pmem_ranges[j-1].pages = pmem_ranges[j].pages;
 184                         pmem_ranges[j].pages = tmp;
 185                 }
 186         }
 187
 188 #ifndef CONFIG_DISCONTIGMEM
 189         /*
 190          * Throw out ranges that are too far apart (controlled by
 191          * MAX_GAP).
 192          */
 193
 194         for (i = 1; i < npmem_ranges; i++) {
 195                 if (pmem_ranges[i].start_pfn -
 196                         (pmem_ranges[i-1].start_pfn +
 197                          pmem_ranges[i-1].pages) > MAX_GAP) {
 198                         npmem_ranges = i;
 199                         printk("Large gap in memory detected (%ld pages). "
 200                                "Consider turning on CONFIG_DISCONTIGMEM\n",
 201                                pmem_ranges[i].start_pfn -
 202                                (pmem_ranges[i-1].start_pfn +
 203                                 pmem_ranges[i-1].pages));
 204                         break;
 205                 }
 206         }
 207 #endif
 208
 209         /* Print the memory ranges */
 210         pr_info("Memory Ranges:\n");
 211
 212         for (i = 0; i < npmem_ranges; i++) {
 213                 struct resource *res = &sysram_resources[i];
 214                 unsigned long start;
 215                 unsigned long size;
 216
 217                 size = (pmem_ranges[i].pages << PAGE_SHIFT);
 218                 start = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
 219                 pr_info("%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n",
 220                         i, start, start + (size - 1), size >> 20);
 221
 222                 /* request memory resource */
 223                 res->name = "System RAM";
 224                 res->start = start;
 225                 res->end = start + size - 1;
 226                 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 227                 request_resource(&iomem_resource, res);
 228         }
 229
 230         sysram_resource_count = npmem_ranges;
 231
 232         /*
 233          * For 32 bit kernels we limit the amount of memory we can
 234          * support, in order to preserve enough kernel address space
 235          * for other purposes. For 64 bit kernels we don't normally
 236          * limit the memory, but this mechanism can be used to
 237          * artificially limit the amount of memory (and it is written
 238          * to work with multiple memory ranges).
 239          */
 240
 241         mem_limit_func();       /* check for "mem=" argument */
 242
 243         mem_max = 0;
 244         for (i = 0; i < npmem_ranges; i++) {
 245                 unsigned long rsize;
 246
 247                 rsize = pmem_ranges[i].pages << PAGE_SHIFT;
 248                 if ((mem_max + rsize) > mem_limit) {
 249                         printk(KERN_WARNING "Memory truncated to %ld MB\n", mem_limit >> 20);
 250                         if (mem_max == mem_limit)
 251                                 npmem_ranges = i;
 252                         else {
 253                                 pmem_ranges[i].pages =   (mem_limit >> PAGE_SHIFT)
 254                                                        - (mem_max >> PAGE_SHIFT);
 255                                 npmem_ranges = i + 1;
 256                                 mem_max = mem_limit;
 257                         }
 258                         break;
 259                 }
 260                 mem_max += rsize;
 261         }
 262
 263         printk(KERN_INFO "Total Memory: %ld MB\n",mem_max >> 20);
 264
 265 #ifndef CONFIG_DISCONTIGMEM
 266         /* Merge the ranges, keeping track of the holes */
 267
 268         {
 269                 unsigned long end_pfn;
 270                 unsigned long hole_pages;
 271
 272                 npmem_holes = 0;
 273                 end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages;
 274                 for (i = 1; i < npmem_ranges; i++) {
 275
 276                         hole_pages = pmem_ranges[i].start_pfn - end_pfn;
 277                         if (hole_pages) {
 278                                 pmem_holes[npmem_holes].start_pfn = end_pfn;
 279                                 pmem_holes[npmem_holes++].pages = hole_pages;
 280                                 end_pfn += hole_pages;
 281                         }
 282                         end_pfn += pmem_ranges[i].pages;
 283                 }
 284
 285                 pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn;
 286                 npmem_ranges = 1;
 287         }
 288 #endif
 289
 290 #ifdef CONFIG_DISCONTIGMEM
 291         for (i = 0; i < MAX_PHYSMEM_RANGES; i++) {
 292                 memset(NODE_DATA(i), 0, sizeof(pg_data_t));
 293         }
 294         memset(pfnnid_map, 0xff, sizeof(pfnnid_map));
 295
 296         for (i = 0; i < npmem_ranges; i++) {
 297                 node_set_state(i, N_NORMAL_MEMORY);
 298                 node_set_online(i);
 299         }
 300 #endif
 301
 302         /*
 303          * Initialize and free the full range of memory in each range.
 304          */
 305
 306         max_pfn = 0;
 307         for (i = 0; i < npmem_ranges; i++) {
 308                 unsigned long start_pfn;
 309                 unsigned long npages;
 310                 unsigned long start;
 311                 unsigned long size;
 312
 313                 start_pfn = pmem_ranges[i].start_pfn;
 314                 npages = pmem_ranges[i].pages;
 315
 316                 start = start_pfn << PAGE_SHIFT;
 317                 size = npages << PAGE_SHIFT;
 318
 319                 /* add system RAM memblock */
 320                 memblock_add(start, size);
 321
 322                 if ((start_pfn + npages) > max_pfn)
 323                         max_pfn = start_pfn + npages;
 324         }
 325
 326         /* IOMMU is always used to access "high mem" on those boxes
 327          * that can support enough mem that a PCI device couldn't
 328          * directly DMA to any physical addresses.
 329          * ISA DMA support will need to revisit this.
 330          */
 331         max_low_pfn = max_pfn;
 332
 333         /* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
 334
 335 #define PDC_CONSOLE_IO_IODC_SIZE 32768
 336
 337         memblock_reserve(0UL, (unsigned long)(PAGE0->mem_free +
 338                                 PDC_CONSOLE_IO_IODC_SIZE));
 339         memblock_reserve(__pa(KERNEL_BINARY_TEXT_START),
 340                         (unsigned long)(_end - KERNEL_BINARY_TEXT_START));
 341
 342 #ifndef CONFIG_DISCONTIGMEM
 343
 344         /* reserve the holes */
 345
 346         for (i = 0; i < npmem_holes; i++) {
 347                 memblock_reserve((pmem_holes[i].start_pfn << PAGE_SHIFT),
 348                                 (pmem_holes[i].pages << PAGE_SHIFT));
 349         }
 350 #endif
 351
 352 #ifdef CONFIG_BLK_DEV_INITRD
 353         if (initrd_start) {
 354                 printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
 355                 if (__pa(initrd_start) < mem_max) {
 356                         unsigned long initrd_reserve;
 357
 358                         if (__pa(initrd_end) > mem_max) {
 359                                 initrd_reserve = mem_max - __pa(initrd_start);
 360                         } else {
 361                                 initrd_reserve = initrd_end - initrd_start;
 362                         }
 363                         initrd_below_start_ok = 1;
 364                         printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
 365
 366                         memblock_reserve(__pa(initrd_start), initrd_reserve);
 367                 }
 368         }
 369 #endif
 370
 371         data_resource.start =  virt_to_phys(&data_start);
 372         data_resource.end = virt_to_phys(_end) - 1;
 373         code_resource.start = virt_to_phys(_text);
 374         code_resource.end = virt_to_phys(&data_start)-1;
 375
 376         /* We don't know which region the kernel will be in, so try
 377          * all of them.
 378          */
 379         for (i = 0; i < sysram_resource_count; i++) {
 380                 struct resource *res = &sysram_resources[i];
 381                 request_resource(res, &code_resource);
 382                 request_resource(res, &data_resource);
 383         }
 384         request_resource(&sysram_resources[0], &pdcdata_resource);
 385
 386         /* Initialize Page Deallocation Table (PDT) and check for bad memory. */
 387         pdc_pdt_init();
 388 }
 389
 390 static int __init parisc_text_address(unsigned long vaddr)
 391 {
 392         static unsigned long head_ptr __initdata;
 393
 394         if (!head_ptr)
 395                 head_ptr = PAGE_MASK & (unsigned long)
 396                         dereference_function_descriptor(&parisc_kernel_start);
 397
 398         return core_kernel_text(vaddr) || vaddr == head_ptr;
 399 }
 400
 401 static void __init map_pages(unsigned long start_vaddr,
 402                              unsigned long start_paddr, unsigned long size,
 403                              pgprot_t pgprot, int force)
 404 {
 405         pgd_t *pg_dir;
 406         pmd_t *pmd;
 407         pte_t *pg_table;
 408         unsigned long end_paddr;
 409         unsigned long start_pmd;
 410         unsigned long start_pte;
 411         unsigned long tmp1;
 412         unsigned long tmp2;
 413         unsigned long address;
 414         unsigned long vaddr;
 415         unsigned long ro_start;
 416         unsigned long ro_end;
 417         unsigned long kernel_end;
 418
 419         ro_start = __pa((unsigned long)_text);
 420         ro_end   = __pa((unsigned long)&data_start);
 421         kernel_end  = __pa((unsigned long)&_end);
 422
 423         end_paddr = start_paddr + size;
 424
 425         pg_dir = pgd_offset_k(start_vaddr);
 426
 427 #if PTRS_PER_PMD == 1
 428         start_pmd = 0;
 429 #else
 430         start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
 431 #endif
 432         start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
 433
 434         address = start_paddr;
 435         vaddr = start_vaddr;
 436         while (address < end_paddr) {
 437 #if PTRS_PER_PMD == 1
 438                 pmd = (pmd_t *)__pa(pg_dir);
 439 #else
 440                 pmd = (pmd_t *)pgd_address(*pg_dir);
 441
 442                 /*
 443                  * pmd is physical at this point
 444                  */
 445
 446                 if (!pmd) {
 447                         pmd = (pmd_t *) get_memblock(PAGE_SIZE << PMD_ORDER);
 448                         pmd = (pmd_t *) __pa(pmd);
 449                 }
 450
 451                 pgd_populate(NULL, pg_dir, __va(pmd));
 452 #endif
 453                 pg_dir++;
 454
 455                 /* now change pmd to kernel virtual addresses */
 456
 457                 pmd = (pmd_t *)__va(pmd) + start_pmd;
 458                 for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++, pmd++) {
 459
 460                         /*
 461                          * pg_table is physical at this point
 462                          */
 463
 464                         pg_table = (pte_t *)pmd_address(*pmd);
 465                         if (!pg_table) {
 466                                 pg_table = (pte_t *) get_memblock(PAGE_SIZE);
 467                                 pg_table = (pte_t *) __pa(pg_table);
 468                         }
 469
 470                         pmd_populate_kernel(NULL, pmd, __va(pg_table));
 471
 472                         /* now change pg_table to kernel virtual addresses */
 473
 474                         pg_table = (pte_t *) __va(pg_table) + start_pte;
 475                         for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++, pg_table++) {
 476                                 pte_t pte;
 477
 478                                 if (force)
 479                                         pte =  __mk_pte(address, pgprot);
 480                                 else if (parisc_text_address(vaddr)) {
 481                                         pte = __mk_pte(address, PAGE_KERNEL_EXEC);
 482                                         if (address >= ro_start && address < kernel_end)
 483                                                 pte = pte_mkhuge(pte);
 484                                 }
 485                                 else
 486 #if defined(CONFIG_PARISC_PAGE_SIZE_4KB)
 487                                 if (address >= ro_start && address < ro_end) {
 488                                         pte = __mk_pte(address, PAGE_KERNEL_EXEC);
 489                                         pte = pte_mkhuge(pte);
 490                                 } else
 491 #endif
 492                                 {
 493                                         pte = __mk_pte(address, pgprot);
 494                                         if (address >= ro_start && address < kernel_end)
 495                                                 pte = pte_mkhuge(pte);
 496                                 }
 497
 498                                 if (address >= end_paddr) {
 499                                         if (force)
 500                                                 break;
 501                                         else
 502                                                 pte_val(pte) = 0;
 503                                 }
 504
 505                                 set_pte(pg_table, pte);
 506
 507                                 address += PAGE_SIZE;
 508                                 vaddr += PAGE_SIZE;
 509                         }
 510                         start_pte = 0;
 511
 512                         if (address >= end_paddr)
 513                             break;
 514                 }
 515                 start_pmd = 0;
 516         }
 517 }
 518
 519 void free_initmem(void)
 520 {
 521         unsigned long init_begin = (unsigned long)__init_begin;
 522         unsigned long init_end = (unsigned long)__init_end;
 523
 524         /* The init text pages are marked R-X.  We have to
 525          * flush the icache and mark them RW-
 526          *
 527          * This is tricky, because map_pages is in the init section.
 528          * Do a dummy remap of the data section first (the data
 529          * section is already PAGE_KERNEL) to pull in the TLB entries
 530          * for map_kernel */
 531         map_pages(init_begin, __pa(init_begin), init_end - init_begin,
 532                   PAGE_KERNEL_RWX, 1);
 533         /* now remap at PAGE_KERNEL since the TLB is pre-primed to execute
 534          * map_pages */
 535         map_pages(init_begin, __pa(init_begin), init_end - init_begin,
 536                   PAGE_KERNEL, 1);
 537
 538         /* force the kernel to see the new TLB entries */
 539         __flush_tlb_range(0, init_begin, init_end);
 540
 541         /* finally dump all the instructions which were cached, since the
 542          * pages are no-longer executable */
 543         flush_icache_range(init_begin, init_end);
 544
 545         free_initmem_default(POISON_FREE_INITMEM);
 546
 547         /* set up a new led state on systems shipped LED State panel */
 548         pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
 549 }
 550
 551
 552 #ifdef CONFIG_STRICT_KERNEL_RWX
 553 void mark_rodata_ro(void)
 554 {
 555         /* rodata memory was already mapped with KERNEL_RO access rights by
 556            pagetable_init() and map_pages(). No need to do additional stuff here */
 557         printk (KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 558                 (unsigned long)(__end_rodata - __start_rodata) >> 10);
 559 }
 560 #endif
 561
 562
 563 /*
 564  * Just an arbitrary offset to serve as a "hole" between mapping areas
 565  * (between top of physical memory and a potential pcxl dma mapping
 566  * area, and below the vmalloc mapping area).
 567  *
 568  * The current 32K value just means that there will be a 32K "hole"
 569  * between mapping areas. That means that  any out-of-bounds memory
 570  * accesses will hopefully be caught. The vmalloc() routines leaves
 571  * a hole of 4kB between each vmalloced area for the same reason.
 572  */
 573
 574  /* Leave room for gateway page expansion */
 575 #if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
 576 #error KERNEL_MAP_START is in gateway reserved region
 577 #endif
 578 #define MAP_START (KERNEL_MAP_START)
 579
 580 #define VM_MAP_OFFSET  (32*1024)
 581 #define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
 582                                      & ~(VM_MAP_OFFSET-1)))
 583
 584 void *parisc_vmalloc_start __read_mostly;
 585 EXPORT_SYMBOL(parisc_vmalloc_start);
 586
 587 #ifdef CONFIG_PA11
 588 unsigned long pcxl_dma_start __read_mostly;
 589 #endif
 590
 591 void __init mem_init(void)
 592 {
 593         /* Do sanity checks on IPC (compat) structures */
 594         BUILD_BUG_ON(sizeof(struct ipc64_perm) != 48);
 595 #ifndef CONFIG_64BIT
 596         BUILD_BUG_ON(sizeof(struct semid64_ds) != 80);
 597         BUILD_BUG_ON(sizeof(struct msqid64_ds) != 104);
 598         BUILD_BUG_ON(sizeof(struct shmid64_ds) != 104);
 599 #endif
 600 #ifdef CONFIG_COMPAT
 601         BUILD_BUG_ON(sizeof(struct compat_ipc64_perm) != sizeof(struct ipc64_perm));
 602         BUILD_BUG_ON(sizeof(struct compat_semid64_ds) != 80);
 603         BUILD_BUG_ON(sizeof(struct compat_msqid64_ds) != 104);
 604         BUILD_BUG_ON(sizeof(struct compat_shmid64_ds) != 104);
 605 #endif
 606
 607         /* Do sanity checks on page table constants */
 608         BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t));
 609         BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t));
 610         BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t));
 611         BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD
 612                         > BITS_PER_LONG);
 613
 614         high_memory = __va((max_pfn << PAGE_SHIFT));
 615         set_max_mapnr(page_to_pfn(virt_to_page(high_memory - 1)) + 1);
 616         free_all_bootmem();
 617
 618 #ifdef CONFIG_PA11
 619         if (hppa_dma_ops == &pcxl_dma_ops) {
 620                 pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
 621                 parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start
 622                                                 + PCXL_DMA_MAP_SIZE);
 623         } else {
 624                 pcxl_dma_start = 0;
 625                 parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
 626         }
 627 #else
 628         parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
 629 #endif
 630
 631         mem_init_print_info(NULL);
 632
 633 #if 0
 634         /*
 635          * Do not expose the virtual kernel memory layout to userspace.
 636          * But keep code for debugging purposes.
 637          */
 638         printk("virtual kernel memory layout:\n"
 639                "    vmalloc : 0x%px - 0x%px   (%4ld MB)\n"
 640                "    memory  : 0x%px - 0x%px   (%4ld MB)\n"
 641                "      .init : 0x%px - 0x%px   (%4ld kB)\n"
 642                "      .data : 0x%px - 0x%px   (%4ld kB)\n"
 643                "      .text : 0x%px - 0x%px   (%4ld kB)\n",
 644
 645                (void*)VMALLOC_START, (void*)VMALLOC_END,
 646                (VMALLOC_END - VMALLOC_START) >> 20,
 647
 648                __va(0), high_memory,
 649                ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 650
 651                __init_begin, __init_end,
 652                ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10,
 653
 654                _etext, _edata,
 655                ((unsigned long)_edata - (unsigned long)_etext) >> 10,
 656
 657                _text, _etext,
 658                ((unsigned long)_etext - (unsigned long)_text) >> 10);
 659 #endif
 660 }
 661
 662 unsigned long *empty_zero_page __read_mostly;
 663 EXPORT_SYMBOL(empty_zero_page);
 664
 665 /*
 666  * pagetable_init() sets up the page tables
 667  *
 668  * Note that gateway_init() places the Linux gateway page at page 0.
 669  * Since gateway pages cannot be dereferenced this has the desirable
 670  * side effect of trapping those pesky NULL-reference errors in the
 671  * kernel.
 672  */
 673 static void __init pagetable_init(void)
 674 {
 675         int range;
 676
 677         /* Map each physical memory range to its kernel vaddr */
 678
 679         for (range = 0; range < npmem_ranges; range++) {
 680                 unsigned long start_paddr;
 681                 unsigned long end_paddr;
 682                 unsigned long size;
 683
 684                 start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
 685                 size = pmem_ranges[range].pages << PAGE_SHIFT;
 686                 end_paddr = start_paddr + size;
 687
 688                 map_pages((unsigned long)__va(start_paddr), start_paddr,
 689                           size, PAGE_KERNEL, 0);
 690         }
 691
 692 #ifdef CONFIG_BLK_DEV_INITRD
 693         if (initrd_end && initrd_end > mem_limit) {
 694                 printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
 695                 map_pages(initrd_start, __pa(initrd_start),
 696                           initrd_end - initrd_start, PAGE_KERNEL, 0);
 697         }
 698 #endif
 699
 700         empty_zero_page = get_memblock(PAGE_SIZE);
 701 }
 702
 703 static void __init gateway_init(void)
 704 {
 705         unsigned long linux_gateway_page_addr;
 706         /* FIXME: This is 'const' in order to trick the compiler
 707            into not treating it as DP-relative data. */
 708         extern void * const linux_gateway_page;
 709
 710         linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
 711
 712         /*
 713          * Setup Linux Gateway page.
 714          *
 715          * The Linux gateway page will reside in kernel space (on virtual
 716          * page 0), so it doesn't need to be aliased into user space.
 717          */
 718
 719         map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
 720                   PAGE_SIZE, PAGE_GATEWAY, 1);
 721 }
 722
 723 void __init paging_init(void)
 724 {
 725         int i;
 726
 727         setup_bootmem();
 728         pagetable_init();
 729         gateway_init();
 730         flush_cache_all_local(); /* start with known state */
 731         flush_tlb_all_local(NULL);
 732
 733         for (i = 0; i < npmem_ranges; i++) {
 734                 unsigned long zones_size[MAX_NR_ZONES] = { 0, };
 735
 736                 zones_size[ZONE_NORMAL] = pmem_ranges[i].pages;
 737
 738 #ifdef CONFIG_DISCONTIGMEM
 739                 /* Need to initialize the pfnnid_map before we can initialize
 740                    the zone */
 741                 {
 742                     int j;
 743                     for (j = (pmem_ranges[i].start_pfn >> PFNNID_SHIFT);
 744                          j <= ((pmem_ranges[i].start_pfn + pmem_ranges[i].pages) >> PFNNID_SHIFT);
 745                          j++) {
 746                         pfnnid_map[j] = i;
 747                     }
 748                 }
 749 #endif
 750
 751                 free_area_init_node(i, zones_size,
 752                                 pmem_ranges[i].start_pfn, NULL);
 753         }
 754 }
 755
 756 #ifdef CONFIG_PA20
 757
 758 /*
 759  * Currently, all PA20 chips have 18 bit protection IDs, which is the
 760  * limiting factor (space ids are 32 bits).
 761  */
 762
 763 #define NR_SPACE_IDS 262144
 764
 765 #else
 766
 767 /*
 768  * Currently we have a one-to-one relationship between space IDs and
 769  * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
 770  * support 15 bit protection IDs, so that is the limiting factor.
 771  * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
 772  * probably not worth the effort for a special case here.
 773  */
 774
 775 #define NR_SPACE_IDS 32768
 776
 777 #endif  /* !CONFIG_PA20 */
 778
 779 #define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
 780 #define SID_ARRAY_SIZE  (NR_SPACE_IDS / (8 * sizeof(long)))
 781
 782 static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
 783 static unsigned long dirty_space_id[SID_ARRAY_SIZE];
 784 static unsigned long space_id_index;
 785 static unsigned long free_space_ids = NR_SPACE_IDS - 1;
 786 static unsigned long dirty_space_ids = 0;
 787
 788 static DEFINE_SPINLOCK(sid_lock);
 789
 790 unsigned long alloc_sid(void)
 791 {
 792         unsigned long index;
 793
 794         spin_lock(&sid_lock);
 795
 796         if (free_space_ids == 0) {
 797                 if (dirty_space_ids != 0) {
 798                         spin_unlock(&sid_lock);
 799                         flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
 800                         spin_lock(&sid_lock);
 801                 }
 802                 BUG_ON(free_space_ids == 0);
 803         }
 804
 805         free_space_ids--;
 806
 807         index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
 808         space_id[index >> SHIFT_PER_LONG] |= (1L << (index & (BITS_PER_LONG - 1)));
 809         space_id_index = index;
 810
 811         spin_unlock(&sid_lock);
 812
 813         return index << SPACEID_SHIFT;
 814 }
 815
 816 void free_sid(unsigned long spaceid)
 817 {
 818         unsigned long index = spaceid >> SPACEID_SHIFT;
 819         unsigned long *dirty_space_offset;
 820
 821         dirty_space_offset = dirty_space_id + (index >> SHIFT_PER_LONG);
 822         index &= (BITS_PER_LONG - 1);
 823
 824         spin_lock(&sid_lock);
 825
 826         BUG_ON(*dirty_space_offset & (1L << index)); /* attempt to free space id twice */
 827
 828         *dirty_space_offset |= (1L << index);
 829         dirty_space_ids++;
 830
 831         spin_unlock(&sid_lock);
 832 }
 833
 834
 835 #ifdef CONFIG_SMP
 836 static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
 837 {
 838         int i;
 839
 840         /* NOTE: sid_lock must be held upon entry */
 841
 842         *ndirtyptr = dirty_space_ids;
 843         if (dirty_space_ids != 0) {
 844             for (i = 0; i < SID_ARRAY_SIZE; i++) {
 845                 dirty_array[i] = dirty_space_id[i];
 846                 dirty_space_id[i] = 0;
 847             }
 848             dirty_space_ids = 0;
 849         }
 850
 851         return;
 852 }
 853
 854 static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
 855 {
 856         int i;
 857
 858         /* NOTE: sid_lock must be held upon entry */
 859
 860         if (ndirty != 0) {
 861                 for (i = 0; i < SID_ARRAY_SIZE; i++) {
 862                         space_id[i] ^= dirty_array[i];
 863                 }
 864
 865                 free_space_ids += ndirty;
 866                 space_id_index = 0;
 867         }
 868 }
 869
 870 #else /* CONFIG_SMP */
 871
 872 static void recycle_sids(void)
 873 {
 874         int i;
 875
 876         /* NOTE: sid_lock must be held upon entry */
 877
 878         if (dirty_space_ids != 0) {
 879                 for (i = 0; i < SID_ARRAY_SIZE; i++) {
 880                         space_id[i] ^= dirty_space_id[i];
 881                         dirty_space_id[i] = 0;
 882                 }
 883
 884                 free_space_ids += dirty_space_ids;
 885                 dirty_space_ids = 0;
 886                 space_id_index = 0;
 887         }
 888 }
 889 #endif
 890
 891 /*
 892  * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
 893  * purged, we can safely reuse the space ids that were released but
 894  * not flushed from the tlb.
 895  */
 896
 897 #ifdef CONFIG_SMP
 898
 899 static unsigned long recycle_ndirty;
 900 static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
 901 static unsigned int recycle_inuse;
 902
 903 void flush_tlb_all(void)
 904 {
 905         int do_recycle;
 906
 907         __inc_irq_stat(irq_tlb_count);
 908         do_recycle = 0;
 909         spin_lock(&sid_lock);
 910         if (dirty_space_ids > RECYCLE_THRESHOLD) {
 911             BUG_ON(recycle_inuse);  /* FIXME: Use a semaphore/wait queue here */
 912             get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
 913             recycle_inuse++;
 914             do_recycle++;
 915         }
 916         spin_unlock(&sid_lock);
 917         on_each_cpu(flush_tlb_all_local, NULL, 1);
 918         if (do_recycle) {
 919             spin_lock(&sid_lock);
 920             recycle_sids(recycle_ndirty,recycle_dirty_array);
 921             recycle_inuse = 0;
 922             spin_unlock(&sid_lock);
 923         }
 924 }
 925 #else
 926 void flush_tlb_all(void)
 927 {
 928         __inc_irq_stat(irq_tlb_count);
 929         spin_lock(&sid_lock);
 930         flush_tlb_all_local(NULL);
 931         recycle_sids();
 932         spin_unlock(&sid_lock);
 933 }
 934 #endif
 935
 936 #ifdef CONFIG_BLK_DEV_INITRD
 937 void free_initrd_mem(unsigned long start, unsigned long end)
 938 {
 939         free_reserved_area((void *)start, (void *)end, -1, "initrd");
 940 }
 941 #endif