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/memblock.h>
  18 #include <linux/gfp.h>
  19 #include <linux/delay.h>
  20 #include <linux/init.h>
  21 #include <linux/initrd.h>
  22 #include <linux/swap.h>
  23 #include <linux/unistd.h>
  24 #include <linux/nodemask.h>     /* for node_online_map */
  25 #include <linux/pagemap.h>      /* for release_pages */
  26 #include <linux/compat.h>
  27
  28 #include <asm/pgalloc.h>
  29 #include <asm/tlb.h>
  30 #include <asm/pdc_chassis.h>
  31 #include <asm/mmzone.h>
  32 #include <asm/sections.h>
  33 #include <asm/msgbuf.h>
  34 #include <asm/sparsemem.h>
  35
  36 extern int  data_start;
  37 extern void parisc_kernel_start(void);  /* Kernel entry point in head.S */
  38
  39 #if CONFIG_PGTABLE_LEVELS == 3
  40 pmd_t pmd0[PTRS_PER_PMD] __section(".data..vm0.pmd") __attribute__ ((aligned(PAGE_SIZE)));
  41 #endif
  42
  43 pgd_t swapper_pg_dir[PTRS_PER_PGD] __section(".data..vm0.pgd") __attribute__ ((aligned(PAGE_SIZE)));
  44 pte_t pg0[PT_INITIAL * PTRS_PER_PTE] __section(".data..vm0.pte") __attribute__ ((aligned(PAGE_SIZE)));
  45
  46 static struct resource data_resource = {
  47         .name   = "Kernel data",
  48         .flags  = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
  49 };
  50
  51 static struct resource code_resource = {
  52         .name   = "Kernel code",
  53         .flags  = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
  54 };
  55
  56 static struct resource pdcdata_resource = {
  57         .name   = "PDC data (Page Zero)",
  58         .start  = 0,
  59         .end    = 0x9ff,
  60         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM,
  61 };
  62
  63 static struct resource sysram_resources[MAX_PHYSMEM_RANGES] __ro_after_init;
  64
  65 /* The following array is initialized from the firmware specific
  66  * information retrieved in kernel/inventory.c.
  67  */
  68
  69 physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES] __initdata;
  70 int npmem_ranges __initdata;
  71
  72 #ifdef CONFIG_64BIT
  73 #define MAX_MEM         (1UL << MAX_PHYSMEM_BITS)
  74 #else /* !CONFIG_64BIT */
  75 #define MAX_MEM         (3584U*1024U*1024U)
  76 #endif /* !CONFIG_64BIT */
  77
  78 static unsigned long mem_limit __read_mostly = MAX_MEM;
  79
  80 static void __init mem_limit_func(void)
  81 {
  82         char *cp, *end;
  83         unsigned long limit;
  84
  85         /* We need this before __setup() functions are called */
  86
  87         limit = MAX_MEM;
  88         for (cp = boot_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 mem_max;
 112 #ifndef CONFIG_SPARSEMEM
 113         physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1];
 114         int npmem_holes;
 115 #endif
 116         int i, sysram_resource_count;
 117
 118         disable_sr_hashing(); /* Turn off space register hashing */
 119
 120         /*
 121          * Sort the ranges. Since the number of ranges is typically
 122          * small, and performance is not an issue here, just do
 123          * a simple insertion sort.
 124          */
 125
 126         for (i = 1; i < npmem_ranges; i++) {
 127                 int j;
 128
 129                 for (j = i; j > 0; j--) {
 130                         if (pmem_ranges[j-1].start_pfn <
 131                             pmem_ranges[j].start_pfn) {
 132
 133                                 break;
 134                         }
 135                         swap(pmem_ranges[j-1], pmem_ranges[j]);
 136                 }
 137         }
 138
 139 #ifndef CONFIG_SPARSEMEM
 140         /*
 141          * Throw out ranges that are too far apart (controlled by
 142          * MAX_GAP).
 143          */
 144
 145         for (i = 1; i < npmem_ranges; i++) {
 146                 if (pmem_ranges[i].start_pfn -
 147                         (pmem_ranges[i-1].start_pfn +
 148                          pmem_ranges[i-1].pages) > MAX_GAP) {
 149                         npmem_ranges = i;
 150                         printk("Large gap in memory detected (%ld pages). "
 151                                "Consider turning on CONFIG_SPARSEMEM\n",
 152                                pmem_ranges[i].start_pfn -
 153                                (pmem_ranges[i-1].start_pfn +
 154                                 pmem_ranges[i-1].pages));
 155                         break;
 156                 }
 157         }
 158 #endif
 159
 160         /* Print the memory ranges */
 161         pr_info("Memory Ranges:\n");
 162
 163         for (i = 0; i < npmem_ranges; i++) {
 164                 struct resource *res = &sysram_resources[i];
 165                 unsigned long start;
 166                 unsigned long size;
 167
 168                 size = (pmem_ranges[i].pages << PAGE_SHIFT);
 169                 start = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
 170                 pr_info("%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n",
 171                         i, start, start + (size - 1), size >> 20);
 172
 173                 /* request memory resource */
 174                 res->name = "System RAM";
 175                 res->start = start;
 176                 res->end = start + size - 1;
 177                 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 178                 request_resource(&iomem_resource, res);
 179         }
 180
 181         sysram_resource_count = npmem_ranges;
 182
 183         /*
 184          * For 32 bit kernels we limit the amount of memory we can
 185          * support, in order to preserve enough kernel address space
 186          * for other purposes. For 64 bit kernels we don't normally
 187          * limit the memory, but this mechanism can be used to
 188          * artificially limit the amount of memory (and it is written
 189          * to work with multiple memory ranges).
 190          */
 191
 192         mem_limit_func();       /* check for "mem=" argument */
 193
 194         mem_max = 0;
 195         for (i = 0; i < npmem_ranges; i++) {
 196                 unsigned long rsize;
 197
 198                 rsize = pmem_ranges[i].pages << PAGE_SHIFT;
 199                 if ((mem_max + rsize) > mem_limit) {
 200                         printk(KERN_WARNING "Memory truncated to %ld MB\n", mem_limit >> 20);
 201                         if (mem_max == mem_limit)
 202                                 npmem_ranges = i;
 203                         else {
 204                                 pmem_ranges[i].pages =   (mem_limit >> PAGE_SHIFT)
 205                                                        - (mem_max >> PAGE_SHIFT);
 206                                 npmem_ranges = i + 1;
 207                                 mem_max = mem_limit;
 208                         }
 209                         break;
 210                 }
 211                 mem_max += rsize;
 212         }
 213
 214         printk(KERN_INFO "Total Memory: %ld MB\n",mem_max >> 20);
 215
 216 #ifndef CONFIG_SPARSEMEM
 217         /* Merge the ranges, keeping track of the holes */
 218         {
 219                 unsigned long end_pfn;
 220                 unsigned long hole_pages;
 221
 222                 npmem_holes = 0;
 223                 end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages;
 224                 for (i = 1; i < npmem_ranges; i++) {
 225
 226                         hole_pages = pmem_ranges[i].start_pfn - end_pfn;
 227                         if (hole_pages) {
 228                                 pmem_holes[npmem_holes].start_pfn = end_pfn;
 229                                 pmem_holes[npmem_holes++].pages = hole_pages;
 230                                 end_pfn += hole_pages;
 231                         }
 232                         end_pfn += pmem_ranges[i].pages;
 233                 }
 234
 235                 pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn;
 236                 npmem_ranges = 1;
 237         }
 238 #endif
 239
 240         /*
 241          * Initialize and free the full range of memory in each range.
 242          */
 243
 244         max_pfn = 0;
 245         for (i = 0; i < npmem_ranges; i++) {
 246                 unsigned long start_pfn;
 247                 unsigned long npages;
 248                 unsigned long start;
 249                 unsigned long size;
 250
 251                 start_pfn = pmem_ranges[i].start_pfn;
 252                 npages = pmem_ranges[i].pages;
 253
 254                 start = start_pfn << PAGE_SHIFT;
 255                 size = npages << PAGE_SHIFT;
 256
 257                 /* add system RAM memblock */
 258                 memblock_add(start, size);
 259
 260                 if ((start_pfn + npages) > max_pfn)
 261                         max_pfn = start_pfn + npages;
 262         }
 263
 264         /*
 265          * We can't use memblock top-down allocations because we only
 266          * created the initial mapping up to KERNEL_INITIAL_SIZE in
 267          * the assembly bootup code.
 268          */
 269         memblock_set_bottom_up(true);
 270
 271         /* IOMMU is always used to access "high mem" on those boxes
 272          * that can support enough mem that a PCI device couldn't
 273          * directly DMA to any physical addresses.
 274          * ISA DMA support will need to revisit this.
 275          */
 276         max_low_pfn = max_pfn;
 277
 278         /* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
 279
 280 #define PDC_CONSOLE_IO_IODC_SIZE 32768
 281
 282         memblock_reserve(0UL, (unsigned long)(PAGE0->mem_free +
 283                                 PDC_CONSOLE_IO_IODC_SIZE));
 284         memblock_reserve(__pa(KERNEL_BINARY_TEXT_START),
 285                         (unsigned long)(_end - KERNEL_BINARY_TEXT_START));
 286
 287 #ifndef CONFIG_SPARSEMEM
 288
 289         /* reserve the holes */
 290
 291         for (i = 0; i < npmem_holes; i++) {
 292                 memblock_reserve((pmem_holes[i].start_pfn << PAGE_SHIFT),
 293                                 (pmem_holes[i].pages << PAGE_SHIFT));
 294         }
 295 #endif
 296
 297 #ifdef CONFIG_BLK_DEV_INITRD
 298         if (initrd_start) {
 299                 printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
 300                 if (__pa(initrd_start) < mem_max) {
 301                         unsigned long initrd_reserve;
 302
 303                         if (__pa(initrd_end) > mem_max) {
 304                                 initrd_reserve = mem_max - __pa(initrd_start);
 305                         } else {
 306                                 initrd_reserve = initrd_end - initrd_start;
 307                         }
 308                         initrd_below_start_ok = 1;
 309                         printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
 310
 311                         memblock_reserve(__pa(initrd_start), initrd_reserve);
 312                 }
 313         }
 314 #endif
 315
 316         data_resource.start =  virt_to_phys(&data_start);
 317         data_resource.end = virt_to_phys(_end) - 1;
 318         code_resource.start = virt_to_phys(_text);
 319         code_resource.end = virt_to_phys(&data_start)-1;
 320
 321         /* We don't know which region the kernel will be in, so try
 322          * all of them.
 323          */
 324         for (i = 0; i < sysram_resource_count; i++) {
 325                 struct resource *res = &sysram_resources[i];
 326                 request_resource(res, &code_resource);
 327                 request_resource(res, &data_resource);
 328         }
 329         request_resource(&sysram_resources[0], &pdcdata_resource);
 330
 331         /* Initialize Page Deallocation Table (PDT) and check for bad memory. */
 332         pdc_pdt_init();
 333
 334         memblock_allow_resize();
 335         memblock_dump_all();
 336 }
 337
 338 static bool kernel_set_to_readonly;
 339
 340 static void __ref map_pages(unsigned long start_vaddr,
 341                             unsigned long start_paddr, unsigned long size,
 342                             pgprot_t pgprot, int force)
 343 {
 344         pmd_t *pmd;
 345         pte_t *pg_table;
 346         unsigned long end_paddr;
 347         unsigned long start_pmd;
 348         unsigned long start_pte;
 349         unsigned long tmp1;
 350         unsigned long tmp2;
 351         unsigned long address;
 352         unsigned long vaddr;
 353         unsigned long ro_start;
 354         unsigned long ro_end;
 355         unsigned long kernel_start, kernel_end;
 356
 357         ro_start = __pa((unsigned long)_text);
 358         ro_end   = __pa((unsigned long)&data_start);
 359         kernel_start = __pa((unsigned long)&__init_begin);
 360         kernel_end  = __pa((unsigned long)&_end);
 361
 362         end_paddr = start_paddr + size;
 363
 364         /* for 2-level configuration PTRS_PER_PMD is 0 so start_pmd will be 0 */
 365         start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
 366         start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
 367
 368         address = start_paddr;
 369         vaddr = start_vaddr;
 370         while (address < end_paddr) {
 371                 pgd_t *pgd = pgd_offset_k(vaddr);
 372                 p4d_t *p4d = p4d_offset(pgd, vaddr);
 373                 pud_t *pud = pud_offset(p4d, vaddr);
 374
 375 #if CONFIG_PGTABLE_LEVELS == 3
 376                 if (pud_none(*pud)) {
 377                         pmd = memblock_alloc(PAGE_SIZE << PMD_TABLE_ORDER,
 378                                              PAGE_SIZE << PMD_TABLE_ORDER);
 379                         if (!pmd)
 380                                 panic("pmd allocation failed.\n");
 381                         pud_populate(NULL, pud, pmd);
 382                 }
 383 #endif
 384
 385                 pmd = pmd_offset(pud, vaddr);
 386                 for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++, pmd++) {
 387                         if (pmd_none(*pmd)) {
 388                                 pg_table = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
 389                                 if (!pg_table)
 390                                         panic("page table allocation failed\n");
 391                                 pmd_populate_kernel(NULL, pmd, pg_table);
 392                         }
 393
 394                         pg_table = pte_offset_kernel(pmd, vaddr);
 395                         for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++, pg_table++) {
 396                                 pte_t pte;
 397                                 pgprot_t prot;
 398                                 bool huge = false;
 399
 400                                 if (force) {
 401                                         prot = pgprot;
 402                                 } else if (address < kernel_start || address >= kernel_end) {
 403                                         /* outside kernel memory */
 404                                         prot = PAGE_KERNEL;
 405                                 } else if (!kernel_set_to_readonly) {
 406                                         /* still initializing, allow writing to RO memory */
 407                                         prot = PAGE_KERNEL_RWX;
 408                                         huge = true;
 409                                 } else if (address >= ro_start) {
 410                                         /* Code (ro) and Data areas */
 411                                         prot = (address < ro_end) ?
 412                                                 PAGE_KERNEL_EXEC : PAGE_KERNEL;
 413                                         huge = true;
 414                                 } else {
 415                                         prot = PAGE_KERNEL;
 416                                 }
 417
 418                                 pte = __mk_pte(address, prot);
 419                                 if (huge)
 420                                         pte = pte_mkhuge(pte);
 421
 422                                 if (address >= end_paddr)
 423                                         break;
 424
 425                                 set_pte(pg_table, pte);
 426
 427                                 address += PAGE_SIZE;
 428                                 vaddr += PAGE_SIZE;
 429                         }
 430                         start_pte = 0;
 431
 432                         if (address >= end_paddr)
 433                             break;
 434                 }
 435                 start_pmd = 0;
 436         }
 437 }
 438
 439 void __init set_kernel_text_rw(int enable_read_write)
 440 {
 441         unsigned long start = (unsigned long) __init_begin;
 442         unsigned long end   = (unsigned long) &data_start;
 443
 444         map_pages(start, __pa(start), end-start,
 445                 PAGE_KERNEL_RWX, enable_read_write ? 1:0);
 446
 447         /* force the kernel to see the new page table entries */
 448         flush_cache_all();
 449         flush_tlb_all();
 450 }
 451
 452 void free_initmem(void)
 453 {
 454         unsigned long init_begin = (unsigned long)__init_begin;
 455         unsigned long init_end = (unsigned long)__init_end;
 456         unsigned long kernel_end  = (unsigned long)&_end;
 457
 458         /* Remap kernel text and data, but do not touch init section yet. */
 459         kernel_set_to_readonly = true;
 460         map_pages(init_end, __pa(init_end), kernel_end - init_end,
 461                   PAGE_KERNEL, 0);
 462
 463         /* The init text pages are marked R-X.  We have to
 464          * flush the icache and mark them RW-
 465          *
 466          * Do a dummy remap of the data section first (the data
 467          * section is already PAGE_KERNEL) to pull in the TLB entries
 468          * for map_kernel */
 469         map_pages(init_begin, __pa(init_begin), init_end - init_begin,
 470                   PAGE_KERNEL_RWX, 1);
 471         /* now remap at PAGE_KERNEL since the TLB is pre-primed to execute
 472          * map_pages */
 473         map_pages(init_begin, __pa(init_begin), init_end - init_begin,
 474                   PAGE_KERNEL, 1);
 475
 476         /* force the kernel to see the new TLB entries */
 477         __flush_tlb_range(0, init_begin, kernel_end);
 478
 479         /* finally dump all the instructions which were cached, since the
 480          * pages are no-longer executable */
 481         flush_icache_range(init_begin, init_end);
 482
 483         free_initmem_default(POISON_FREE_INITMEM);
 484
 485         /* set up a new led state on systems shipped LED State panel */
 486         pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
 487 }
 488
 489
 490 #ifdef CONFIG_STRICT_KERNEL_RWX
 491 void mark_rodata_ro(void)
 492 {
 493         /* rodata memory was already mapped with KERNEL_RO access rights by
 494            pagetable_init() and map_pages(). No need to do additional stuff here */
 495         unsigned long roai_size = __end_ro_after_init - __start_ro_after_init;
 496
 497         pr_info("Write protected read-only-after-init data: %luk\n", roai_size >> 10);
 498 }
 499 #endif
 500
 501
 502 /*
 503  * Just an arbitrary offset to serve as a "hole" between mapping areas
 504  * (between top of physical memory and a potential pcxl dma mapping
 505  * area, and below the vmalloc mapping area).
 506  *
 507  * The current 32K value just means that there will be a 32K "hole"
 508  * between mapping areas. That means that  any out-of-bounds memory
 509  * accesses will hopefully be caught. The vmalloc() routines leaves
 510  * a hole of 4kB between each vmalloced area for the same reason.
 511  */
 512
 513  /* Leave room for gateway page expansion */
 514 #if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
 515 #error KERNEL_MAP_START is in gateway reserved region
 516 #endif
 517 #define MAP_START (KERNEL_MAP_START)
 518
 519 #define VM_MAP_OFFSET  (32*1024)
 520 #define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
 521                                      & ~(VM_MAP_OFFSET-1)))
 522
 523 void *parisc_vmalloc_start __ro_after_init;
 524 EXPORT_SYMBOL(parisc_vmalloc_start);
 525
 526 void __init mem_init(void)
 527 {
 528         /* Do sanity checks on IPC (compat) structures */
 529         BUILD_BUG_ON(sizeof(struct ipc64_perm) != 48);
 530 #ifndef CONFIG_64BIT
 531         BUILD_BUG_ON(sizeof(struct semid64_ds) != 80);
 532         BUILD_BUG_ON(sizeof(struct msqid64_ds) != 104);
 533         BUILD_BUG_ON(sizeof(struct shmid64_ds) != 104);
 534 #endif
 535 #ifdef CONFIG_COMPAT
 536         BUILD_BUG_ON(sizeof(struct compat_ipc64_perm) != sizeof(struct ipc64_perm));
 537         BUILD_BUG_ON(sizeof(struct compat_semid64_ds) != 80);
 538         BUILD_BUG_ON(sizeof(struct compat_msqid64_ds) != 104);
 539         BUILD_BUG_ON(sizeof(struct compat_shmid64_ds) != 104);
 540 #endif
 541
 542         /* Do sanity checks on page table constants */
 543         BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t));
 544         BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t));
 545         BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t));
 546         BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD
 547                         > BITS_PER_LONG);
 548 #if CONFIG_PGTABLE_LEVELS == 3
 549         BUILD_BUG_ON(PT_INITIAL > PTRS_PER_PMD);
 550 #else
 551         BUILD_BUG_ON(PT_INITIAL > PTRS_PER_PGD);
 552 #endif
 553
 554 #ifdef CONFIG_64BIT
 555         /* avoid ldil_%L() asm statements to sign-extend into upper 32-bits */
 556         BUILD_BUG_ON(__PAGE_OFFSET >= 0x80000000);
 557         BUILD_BUG_ON(TMPALIAS_MAP_START >= 0x80000000);
 558 #endif
 559
 560         high_memory = __va((max_pfn << PAGE_SHIFT));
 561         set_max_mapnr(max_low_pfn);
 562         memblock_free_all();
 563
 564 #ifdef CONFIG_PA11
 565         if (boot_cpu_data.cpu_type == pcxl2 || boot_cpu_data.cpu_type == pcxl) {
 566                 pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
 567                 parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start
 568                                                 + PCXL_DMA_MAP_SIZE);
 569         } else
 570 #endif
 571                 parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
 572
 573 #if 0
 574         /*
 575          * Do not expose the virtual kernel memory layout to userspace.
 576          * But keep code for debugging purposes.
 577          */
 578         printk("virtual kernel memory layout:\n"
 579                "     vmalloc : 0x%px - 0x%px   (%4ld MB)\n"
 580                "     fixmap  : 0x%px - 0x%px   (%4ld kB)\n"
 581                "     memory  : 0x%px - 0x%px   (%4ld MB)\n"
 582                "       .init : 0x%px - 0x%px   (%4ld kB)\n"
 583                "       .data : 0x%px - 0x%px   (%4ld kB)\n"
 584                "       .text : 0x%px - 0x%px   (%4ld kB)\n",
 585
 586                (void*)VMALLOC_START, (void*)VMALLOC_END,
 587                (VMALLOC_END - VMALLOC_START) >> 20,
 588
 589                (void *)FIXMAP_START, (void *)(FIXMAP_START + FIXMAP_SIZE),
 590                (unsigned long)(FIXMAP_SIZE / 1024),
 591
 592                __va(0), high_memory,
 593                ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 594
 595                __init_begin, __init_end,
 596                ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10,
 597
 598                _etext, _edata,
 599                ((unsigned long)_edata - (unsigned long)_etext) >> 10,
 600
 601                _text, _etext,
 602                ((unsigned long)_etext - (unsigned long)_text) >> 10);
 603 #endif
 604 }
 605
 606 unsigned long *empty_zero_page __ro_after_init;
 607 EXPORT_SYMBOL(empty_zero_page);
 608
 609 /*
 610  * pagetable_init() sets up the page tables
 611  *
 612  * Note that gateway_init() places the Linux gateway page at page 0.
 613  * Since gateway pages cannot be dereferenced this has the desirable
 614  * side effect of trapping those pesky NULL-reference errors in the
 615  * kernel.
 616  */
 617 static void __init pagetable_init(void)
 618 {
 619         int range;
 620
 621         /* Map each physical memory range to its kernel vaddr */
 622
 623         for (range = 0; range < npmem_ranges; range++) {
 624                 unsigned long start_paddr;
 625                 unsigned long size;
 626
 627                 start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
 628                 size = pmem_ranges[range].pages << PAGE_SHIFT;
 629
 630                 map_pages((unsigned long)__va(start_paddr), start_paddr,
 631                           size, PAGE_KERNEL, 0);
 632         }
 633
 634 #ifdef CONFIG_BLK_DEV_INITRD
 635         if (initrd_end && initrd_end > mem_limit) {
 636                 printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
 637                 map_pages(initrd_start, __pa(initrd_start),
 638                           initrd_end - initrd_start, PAGE_KERNEL, 0);
 639         }
 640 #endif
 641
 642         empty_zero_page = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
 643         if (!empty_zero_page)
 644                 panic("zero page allocation failed.\n");
 645
 646 }
 647
 648 static void __init gateway_init(void)
 649 {
 650         unsigned long linux_gateway_page_addr;
 651         /* FIXME: This is 'const' in order to trick the compiler
 652            into not treating it as DP-relative data. */
 653         extern void * const linux_gateway_page;
 654
 655         linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
 656
 657         /*
 658          * Setup Linux Gateway page.
 659          *
 660          * The Linux gateway page will reside in kernel space (on virtual
 661          * page 0), so it doesn't need to be aliased into user space.
 662          */
 663
 664         map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
 665                   PAGE_SIZE, PAGE_GATEWAY, 1);
 666 }
 667
 668 static void __init fixmap_init(void)
 669 {
 670         unsigned long addr = FIXMAP_START;
 671         unsigned long end = FIXMAP_START + FIXMAP_SIZE;
 672         pgd_t *pgd = pgd_offset_k(addr);
 673         p4d_t *p4d = p4d_offset(pgd, addr);
 674         pud_t *pud = pud_offset(p4d, addr);
 675         pmd_t *pmd;
 676
 677         BUILD_BUG_ON(FIXMAP_SIZE > PMD_SIZE);
 678
 679 #if CONFIG_PGTABLE_LEVELS == 3
 680         if (pud_none(*pud)) {
 681                 pmd = memblock_alloc(PAGE_SIZE << PMD_TABLE_ORDER,
 682                                      PAGE_SIZE << PMD_TABLE_ORDER);
 683                 if (!pmd)
 684                         panic("fixmap: pmd allocation failed.\n");
 685                 pud_populate(NULL, pud, pmd);
 686         }
 687 #endif
 688
 689         pmd = pmd_offset(pud, addr);
 690         do {
 691                 pte_t *pte = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
 692                 if (!pte)
 693                         panic("fixmap: pte allocation failed.\n");
 694
 695                 pmd_populate_kernel(&init_mm, pmd, pte);
 696
 697                 addr += PAGE_SIZE;
 698         } while (addr < end);
 699 }
 700
 701 static void __init parisc_bootmem_free(void)
 702 {
 703         unsigned long max_zone_pfn[MAX_NR_ZONES] = { 0, };
 704
 705         max_zone_pfn[0] = memblock_end_of_DRAM();
 706
 707         free_area_init(max_zone_pfn);
 708 }
 709
 710 void __init paging_init(void)
 711 {
 712         setup_bootmem();
 713         pagetable_init();
 714         gateway_init();
 715         fixmap_init();
 716         flush_cache_all_local(); /* start with known state */
 717         flush_tlb_all_local(NULL);
 718
 719         sparse_init();
 720         parisc_bootmem_free();
 721 }
 722
 723 #ifdef CONFIG_PA20
 724
 725 /*
 726  * Currently, all PA20 chips have 18 bit protection IDs, which is the
 727  * limiting factor (space ids are 32 bits).
 728  */
 729
 730 #define NR_SPACE_IDS 262144
 731
 732 #else
 733
 734 /*
 735  * Currently we have a one-to-one relationship between space IDs and
 736  * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
 737  * support 15 bit protection IDs, so that is the limiting factor.
 738  * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
 739  * probably not worth the effort for a special case here.
 740  */
 741
 742 #define NR_SPACE_IDS 32768
 743
 744 #endif  /* !CONFIG_PA20 */
 745
 746 #define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
 747 #define SID_ARRAY_SIZE  (NR_SPACE_IDS / (8 * sizeof(long)))
 748
 749 static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
 750 static unsigned long dirty_space_id[SID_ARRAY_SIZE];
 751 static unsigned long space_id_index;
 752 static unsigned long free_space_ids = NR_SPACE_IDS - 1;
 753 static unsigned long dirty_space_ids;
 754
 755 static DEFINE_SPINLOCK(sid_lock);
 756
 757 unsigned long alloc_sid(void)
 758 {
 759         unsigned long index;
 760
 761         spin_lock(&sid_lock);
 762
 763         if (free_space_ids == 0) {
 764                 if (dirty_space_ids != 0) {
 765                         spin_unlock(&sid_lock);
 766                         flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
 767                         spin_lock(&sid_lock);
 768                 }
 769                 BUG_ON(free_space_ids == 0);
 770         }
 771
 772         free_space_ids--;
 773
 774         index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
 775         space_id[BIT_WORD(index)] |= BIT_MASK(index);
 776         space_id_index = index;
 777
 778         spin_unlock(&sid_lock);
 779
 780         return index << SPACEID_SHIFT;
 781 }
 782
 783 void free_sid(unsigned long spaceid)
 784 {
 785         unsigned long index = spaceid >> SPACEID_SHIFT;
 786         unsigned long *dirty_space_offset, mask;
 787
 788         dirty_space_offset = &dirty_space_id[BIT_WORD(index)];
 789         mask = BIT_MASK(index);
 790
 791         spin_lock(&sid_lock);
 792
 793         BUG_ON(*dirty_space_offset & mask); /* attempt to free space id twice */
 794
 795         *dirty_space_offset |= mask;
 796         dirty_space_ids++;
 797
 798         spin_unlock(&sid_lock);
 799 }
 800
 801
 802 #ifdef CONFIG_SMP
 803 static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
 804 {
 805         int i;
 806
 807         /* NOTE: sid_lock must be held upon entry */
 808
 809         *ndirtyptr = dirty_space_ids;
 810         if (dirty_space_ids != 0) {
 811             for (i = 0; i < SID_ARRAY_SIZE; i++) {
 812                 dirty_array[i] = dirty_space_id[i];
 813                 dirty_space_id[i] = 0;
 814             }
 815             dirty_space_ids = 0;
 816         }
 817
 818         return;
 819 }
 820
 821 static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
 822 {
 823         int i;
 824
 825         /* NOTE: sid_lock must be held upon entry */
 826
 827         if (ndirty != 0) {
 828                 for (i = 0; i < SID_ARRAY_SIZE; i++) {
 829                         space_id[i] ^= dirty_array[i];
 830                 }
 831
 832                 free_space_ids += ndirty;
 833                 space_id_index = 0;
 834         }
 835 }
 836
 837 #else /* CONFIG_SMP */
 838
 839 static void recycle_sids(void)
 840 {
 841         int i;
 842
 843         /* NOTE: sid_lock must be held upon entry */
 844
 845         if (dirty_space_ids != 0) {
 846                 for (i = 0; i < SID_ARRAY_SIZE; i++) {
 847                         space_id[i] ^= dirty_space_id[i];
 848                         dirty_space_id[i] = 0;
 849                 }
 850
 851                 free_space_ids += dirty_space_ids;
 852                 dirty_space_ids = 0;
 853                 space_id_index = 0;
 854         }
 855 }
 856 #endif
 857
 858 /*
 859  * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
 860  * purged, we can safely reuse the space ids that were released but
 861  * not flushed from the tlb.
 862  */
 863
 864 #ifdef CONFIG_SMP
 865
 866 static unsigned long recycle_ndirty;
 867 static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
 868 static unsigned int recycle_inuse;
 869
 870 void flush_tlb_all(void)
 871 {
 872         int do_recycle;
 873
 874         do_recycle = 0;
 875         spin_lock(&sid_lock);
 876         __inc_irq_stat(irq_tlb_count);
 877         if (dirty_space_ids > RECYCLE_THRESHOLD) {
 878             BUG_ON(recycle_inuse);  /* FIXME: Use a semaphore/wait queue here */
 879             get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
 880             recycle_inuse++;
 881             do_recycle++;
 882         }
 883         spin_unlock(&sid_lock);
 884         on_each_cpu(flush_tlb_all_local, NULL, 1);
 885         if (do_recycle) {
 886             spin_lock(&sid_lock);
 887             recycle_sids(recycle_ndirty,recycle_dirty_array);
 888             recycle_inuse = 0;
 889             spin_unlock(&sid_lock);
 890         }
 891 }
 892 #else
 893 void flush_tlb_all(void)
 894 {
 895         spin_lock(&sid_lock);
 896         __inc_irq_stat(irq_tlb_count);
 897         flush_tlb_all_local(NULL);
 898         recycle_sids();
 899         spin_unlock(&sid_lock);
 900 }
 901 #endif
 902
 903 static const pgprot_t protection_map[16] = {
 904         [VM_NONE]                                       = PAGE_NONE,
 905         [VM_READ]                                       = PAGE_READONLY,
 906         [VM_WRITE]                                      = PAGE_NONE,
 907         [VM_WRITE | VM_READ]                            = PAGE_READONLY,
 908         [VM_EXEC]                                       = PAGE_EXECREAD,
 909         [VM_EXEC | VM_READ]                             = PAGE_EXECREAD,
 910         [VM_EXEC | VM_WRITE]                            = PAGE_EXECREAD,
 911         [VM_EXEC | VM_WRITE | VM_READ]                  = PAGE_EXECREAD,
 912         [VM_SHARED]                                     = PAGE_NONE,
 913         [VM_SHARED | VM_READ]                           = PAGE_READONLY,
 914         [VM_SHARED | VM_WRITE]                          = PAGE_WRITEONLY,
 915         [VM_SHARED | VM_WRITE | VM_READ]                = PAGE_SHARED,
 916         [VM_SHARED | VM_EXEC]                           = PAGE_EXECREAD,
 917         [VM_SHARED | VM_EXEC | VM_READ]                 = PAGE_EXECREAD,
 918         [VM_SHARED | VM_EXEC | VM_WRITE]                = PAGE_RWX,
 919         [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ]      = PAGE_RWX
 920 };
 921 DECLARE_VM_GET_PAGE_PROT