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                                         break;
 500
 501                                 set_pte(pg_table, pte);
 502
 503                                 address += PAGE_SIZE;
 504                                 vaddr += PAGE_SIZE;
 505                         }
 506                         start_pte = 0;
 507
 508                         if (address >= end_paddr)
 509                             break;
 510                 }
 511                 start_pmd = 0;
 512         }
 513 }
 514
 515 void free_initmem(void)
 516 {
 517         unsigned long init_begin = (unsigned long)__init_begin;
 518         unsigned long init_end = (unsigned long)__init_end;
 519
 520         /* The init text pages are marked R-X.  We have to
 521          * flush the icache and mark them RW-
 522          *
 523          * This is tricky, because map_pages is in the init section.
 524          * Do a dummy remap of the data section first (the data
 525          * section is already PAGE_KERNEL) to pull in the TLB entries
 526          * for map_kernel */
 527         map_pages(init_begin, __pa(init_begin), init_end - init_begin,
 528                   PAGE_KERNEL_RWX, 1);
 529         /* now remap at PAGE_KERNEL since the TLB is pre-primed to execute
 530          * map_pages */
 531         map_pages(init_begin, __pa(init_begin), init_end - init_begin,
 532                   PAGE_KERNEL, 1);
 533
 534         /* force the kernel to see the new TLB entries */
 535         __flush_tlb_range(0, init_begin, init_end);
 536
 537         /* finally dump all the instructions which were cached, since the
 538          * pages are no-longer executable */
 539         flush_icache_range(init_begin, init_end);
 540
 541         free_initmem_default(POISON_FREE_INITMEM);
 542
 543         /* set up a new led state on systems shipped LED State panel */
 544         pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
 545 }
 546
 547
 548 #ifdef CONFIG_STRICT_KERNEL_RWX
 549 void mark_rodata_ro(void)
 550 {
 551         /* rodata memory was already mapped with KERNEL_RO access rights by
 552            pagetable_init() and map_pages(). No need to do additional stuff here */
 553         printk (KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 554                 (unsigned long)(__end_rodata - __start_rodata) >> 10);
 555 }
 556 #endif
 557
 558
 559 /*
 560  * Just an arbitrary offset to serve as a "hole" between mapping areas
 561  * (between top of physical memory and a potential pcxl dma mapping
 562  * area, and below the vmalloc mapping area).
 563  *
 564  * The current 32K value just means that there will be a 32K "hole"
 565  * between mapping areas. That means that  any out-of-bounds memory
 566  * accesses will hopefully be caught. The vmalloc() routines leaves
 567  * a hole of 4kB between each vmalloced area for the same reason.
 568  */
 569
 570  /* Leave room for gateway page expansion */
 571 #if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
 572 #error KERNEL_MAP_START is in gateway reserved region
 573 #endif
 574 #define MAP_START (KERNEL_MAP_START)
 575
 576 #define VM_MAP_OFFSET  (32*1024)
 577 #define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
 578                                      & ~(VM_MAP_OFFSET-1)))
 579
 580 void *parisc_vmalloc_start __read_mostly;
 581 EXPORT_SYMBOL(parisc_vmalloc_start);
 582
 583 #ifdef CONFIG_PA11
 584 unsigned long pcxl_dma_start __read_mostly;
 585 #endif
 586
 587 void __init mem_init(void)
 588 {
 589         /* Do sanity checks on IPC (compat) structures */
 590         BUILD_BUG_ON(sizeof(struct ipc64_perm) != 48);
 591 #ifndef CONFIG_64BIT
 592         BUILD_BUG_ON(sizeof(struct semid64_ds) != 80);
 593         BUILD_BUG_ON(sizeof(struct msqid64_ds) != 104);
 594         BUILD_BUG_ON(sizeof(struct shmid64_ds) != 104);
 595 #endif
 596 #ifdef CONFIG_COMPAT
 597         BUILD_BUG_ON(sizeof(struct compat_ipc64_perm) != sizeof(struct ipc64_perm));
 598         BUILD_BUG_ON(sizeof(struct compat_semid64_ds) != 80);
 599         BUILD_BUG_ON(sizeof(struct compat_msqid64_ds) != 104);
 600         BUILD_BUG_ON(sizeof(struct compat_shmid64_ds) != 104);
 601 #endif
 602
 603         /* Do sanity checks on page table constants */
 604         BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t));
 605         BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t));
 606         BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t));
 607         BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD
 608                         > BITS_PER_LONG);
 609
 610         high_memory = __va((max_pfn << PAGE_SHIFT));
 611         set_max_mapnr(max_low_pfn);
 612         free_all_bootmem();
 613
 614 #ifdef CONFIG_PA11
 615         if (hppa_dma_ops == &pcxl_dma_ops) {
 616                 pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
 617                 parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start
 618                                                 + PCXL_DMA_MAP_SIZE);
 619         } else {
 620                 pcxl_dma_start = 0;
 621                 parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
 622         }
 623 #else
 624         parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
 625 #endif
 626
 627         mem_init_print_info(NULL);
 628 #ifdef CONFIG_DEBUG_KERNEL /* double-sanity-check paranoia */
 629         printk("virtual kernel memory layout:\n"
 630                "    vmalloc : 0x%p - 0x%p   (%4ld MB)\n"
 631                "    memory  : 0x%p - 0x%p   (%4ld MB)\n"
 632                "      .init : 0x%p - 0x%p   (%4ld kB)\n"
 633                "      .data : 0x%p - 0x%p   (%4ld kB)\n"
 634                "      .text : 0x%p - 0x%p   (%4ld kB)\n",
 635
 636                (void*)VMALLOC_START, (void*)VMALLOC_END,
 637                (VMALLOC_END - VMALLOC_START) >> 20,
 638
 639                __va(0), high_memory,
 640                ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 641
 642                __init_begin, __init_end,
 643                ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10,
 644
 645                _etext, _edata,
 646                ((unsigned long)_edata - (unsigned long)_etext) >> 10,
 647
 648                _text, _etext,
 649                ((unsigned long)_etext - (unsigned long)_text) >> 10);
 650 #endif
 651 }
 652
 653 unsigned long *empty_zero_page __read_mostly;
 654 EXPORT_SYMBOL(empty_zero_page);
 655
 656 /*
 657  * pagetable_init() sets up the page tables
 658  *
 659  * Note that gateway_init() places the Linux gateway page at page 0.
 660  * Since gateway pages cannot be dereferenced this has the desirable
 661  * side effect of trapping those pesky NULL-reference errors in the
 662  * kernel.
 663  */
 664 static void __init pagetable_init(void)
 665 {
 666         int range;
 667
 668         /* Map each physical memory range to its kernel vaddr */
 669
 670         for (range = 0; range < npmem_ranges; range++) {
 671                 unsigned long start_paddr;
 672                 unsigned long end_paddr;
 673                 unsigned long size;
 674
 675                 start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
 676                 size = pmem_ranges[range].pages << PAGE_SHIFT;
 677                 end_paddr = start_paddr + size;
 678
 679                 map_pages((unsigned long)__va(start_paddr), start_paddr,
 680                           size, PAGE_KERNEL, 0);
 681         }
 682
 683 #ifdef CONFIG_BLK_DEV_INITRD
 684         if (initrd_end && initrd_end > mem_limit) {
 685                 printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
 686                 map_pages(initrd_start, __pa(initrd_start),
 687                           initrd_end - initrd_start, PAGE_KERNEL, 0);
 688         }
 689 #endif
 690
 691         empty_zero_page = get_memblock(PAGE_SIZE);
 692 }
 693
 694 static void __init gateway_init(void)
 695 {
 696         unsigned long linux_gateway_page_addr;
 697         /* FIXME: This is 'const' in order to trick the compiler
 698            into not treating it as DP-relative data. */
 699         extern void * const linux_gateway_page;
 700
 701         linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
 702
 703         /*
 704          * Setup Linux Gateway page.
 705          *
 706          * The Linux gateway page will reside in kernel space (on virtual
 707          * page 0), so it doesn't need to be aliased into user space.
 708          */
 709
 710         map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
 711                   PAGE_SIZE, PAGE_GATEWAY, 1);
 712 }
 713
 714 void __init paging_init(void)
 715 {
 716         int i;
 717
 718         setup_bootmem();
 719         pagetable_init();
 720         gateway_init();
 721         flush_cache_all_local(); /* start with known state */
 722         flush_tlb_all_local(NULL);
 723
 724         for (i = 0; i < npmem_ranges; i++) {
 725                 unsigned long zones_size[MAX_NR_ZONES] = { 0, };
 726
 727                 zones_size[ZONE_NORMAL] = pmem_ranges[i].pages;
 728
 729 #ifdef CONFIG_DISCONTIGMEM
 730                 /* Need to initialize the pfnnid_map before we can initialize
 731                    the zone */
 732                 {
 733                     int j;
 734                     for (j = (pmem_ranges[i].start_pfn >> PFNNID_SHIFT);
 735                          j <= ((pmem_ranges[i].start_pfn + pmem_ranges[i].pages) >> PFNNID_SHIFT);
 736                          j++) {
 737                         pfnnid_map[j] = i;
 738                     }
 739                 }
 740 #endif
 741
 742                 free_area_init_node(i, zones_size,
 743                                 pmem_ranges[i].start_pfn, NULL);
 744         }
 745 }
 746
 747 #ifdef CONFIG_PA20
 748
 749 /*
 750  * Currently, all PA20 chips have 18 bit protection IDs, which is the
 751  * limiting factor (space ids are 32 bits).
 752  */
 753
 754 #define NR_SPACE_IDS 262144
 755
 756 #else
 757
 758 /*
 759  * Currently we have a one-to-one relationship between space IDs and
 760  * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
 761  * support 15 bit protection IDs, so that is the limiting factor.
 762  * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
 763  * probably not worth the effort for a special case here.
 764  */
 765
 766 #define NR_SPACE_IDS 32768
 767
 768 #endif  /* !CONFIG_PA20 */
 769
 770 #define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
 771 #define SID_ARRAY_SIZE  (NR_SPACE_IDS / (8 * sizeof(long)))
 772
 773 static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
 774 static unsigned long dirty_space_id[SID_ARRAY_SIZE];
 775 static unsigned long space_id_index;
 776 static unsigned long free_space_ids = NR_SPACE_IDS - 1;
 777 static unsigned long dirty_space_ids = 0;
 778
 779 static DEFINE_SPINLOCK(sid_lock);
 780
 781 unsigned long alloc_sid(void)
 782 {
 783         unsigned long index;
 784
 785         spin_lock(&sid_lock);
 786
 787         if (free_space_ids == 0) {
 788                 if (dirty_space_ids != 0) {
 789                         spin_unlock(&sid_lock);
 790                         flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
 791                         spin_lock(&sid_lock);
 792                 }
 793                 BUG_ON(free_space_ids == 0);
 794         }
 795
 796         free_space_ids--;
 797
 798         index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
 799         space_id[index >> SHIFT_PER_LONG] |= (1L << (index & (BITS_PER_LONG - 1)));
 800         space_id_index = index;
 801
 802         spin_unlock(&sid_lock);
 803
 804         return index << SPACEID_SHIFT;
 805 }
 806
 807 void free_sid(unsigned long spaceid)
 808 {
 809         unsigned long index = spaceid >> SPACEID_SHIFT;
 810         unsigned long *dirty_space_offset;
 811
 812         dirty_space_offset = dirty_space_id + (index >> SHIFT_PER_LONG);
 813         index &= (BITS_PER_LONG - 1);
 814
 815         spin_lock(&sid_lock);
 816
 817         BUG_ON(*dirty_space_offset & (1L << index)); /* attempt to free space id twice */
 818
 819         *dirty_space_offset |= (1L << index);
 820         dirty_space_ids++;
 821
 822         spin_unlock(&sid_lock);
 823 }
 824
 825
 826 #ifdef CONFIG_SMP
 827 static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
 828 {
 829         int i;
 830
 831         /* NOTE: sid_lock must be held upon entry */
 832
 833         *ndirtyptr = dirty_space_ids;
 834         if (dirty_space_ids != 0) {
 835             for (i = 0; i < SID_ARRAY_SIZE; i++) {
 836                 dirty_array[i] = dirty_space_id[i];
 837                 dirty_space_id[i] = 0;
 838             }
 839             dirty_space_ids = 0;
 840         }
 841
 842         return;
 843 }
 844
 845 static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
 846 {
 847         int i;
 848
 849         /* NOTE: sid_lock must be held upon entry */
 850
 851         if (ndirty != 0) {
 852                 for (i = 0; i < SID_ARRAY_SIZE; i++) {
 853                         space_id[i] ^= dirty_array[i];
 854                 }
 855
 856                 free_space_ids += ndirty;
 857                 space_id_index = 0;
 858         }
 859 }
 860
 861 #else /* CONFIG_SMP */
 862
 863 static void recycle_sids(void)
 864 {
 865         int i;
 866
 867         /* NOTE: sid_lock must be held upon entry */
 868
 869         if (dirty_space_ids != 0) {
 870                 for (i = 0; i < SID_ARRAY_SIZE; i++) {
 871                         space_id[i] ^= dirty_space_id[i];
 872                         dirty_space_id[i] = 0;
 873                 }
 874
 875                 free_space_ids += dirty_space_ids;
 876                 dirty_space_ids = 0;
 877                 space_id_index = 0;
 878         }
 879 }
 880 #endif
 881
 882 /*
 883  * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
 884  * purged, we can safely reuse the space ids that were released but
 885  * not flushed from the tlb.
 886  */
 887
 888 #ifdef CONFIG_SMP
 889
 890 static unsigned long recycle_ndirty;
 891 static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
 892 static unsigned int recycle_inuse;
 893
 894 void flush_tlb_all(void)
 895 {
 896         int do_recycle;
 897
 898         do_recycle = 0;
 899         spin_lock(&sid_lock);
 900         __inc_irq_stat(irq_tlb_count);
 901         if (dirty_space_ids > RECYCLE_THRESHOLD) {
 902             BUG_ON(recycle_inuse);  /* FIXME: Use a semaphore/wait queue here */
 903             get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
 904             recycle_inuse++;
 905             do_recycle++;
 906         }
 907         spin_unlock(&sid_lock);
 908         on_each_cpu(flush_tlb_all_local, NULL, 1);
 909         if (do_recycle) {
 910             spin_lock(&sid_lock);
 911             recycle_sids(recycle_ndirty,recycle_dirty_array);
 912             recycle_inuse = 0;
 913             spin_unlock(&sid_lock);
 914         }
 915 }
 916 #else
 917 void flush_tlb_all(void)
 918 {
 919         spin_lock(&sid_lock);
 920         __inc_irq_stat(irq_tlb_count);
 921         flush_tlb_all_local(NULL);
 922         recycle_sids();
 923         spin_unlock(&sid_lock);
 924 }
 925 #endif
 926
 927 #ifdef CONFIG_BLK_DEV_INITRD
 928 void free_initrd_mem(unsigned long start, unsigned long end)
 929 {
 930         free_reserved_area((void *)start, (void *)end, -1, "initrd");
 931 }
 932 #endif