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/pgtable.h>
  30 #include <asm/tlb.h>
  31 #include <asm/pdc_chassis.h>
  32 #include <asm/mmzone.h>
  33 #include <asm/sections.h>
  34 #include <asm/msgbuf.h>
  35 #include <asm/sparsemem.h>
  36
  37 extern int  data_start;
  38 extern void parisc_kernel_start(void);  /* Kernel entry point in head.S */
  39
  40 #if CONFIG_PGTABLE_LEVELS == 3
  41 /* NOTE: This layout exactly conforms to the hybrid L2/L3 page table layout
  42  * with the first pmd adjacent to the pgd and below it. gcc doesn't actually
  43  * guarantee that global objects will be laid out in memory in the same order
  44  * as the order of declaration, so put these in different sections and use
  45  * the linker script to order them. */
  46 pmd_t pmd0[PTRS_PER_PMD] __attribute__ ((__section__ (".data..vm0.pmd"), aligned(PAGE_SIZE)));
  47 #endif
  48
  49 pgd_t swapper_pg_dir[PTRS_PER_PGD] __attribute__ ((__section__ (".data..vm0.pgd"), aligned(PAGE_SIZE)));
  50 pte_t pg0[PT_INITIAL * PTRS_PER_PTE] __attribute__ ((__section__ (".data..vm0.pte"), aligned(PAGE_SIZE)));
  51
  52 static struct resource data_resource = {
  53         .name   = "Kernel data",
  54         .flags  = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
  55 };
  56
  57 static struct resource code_resource = {
  58         .name   = "Kernel code",
  59         .flags  = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
  60 };
  61
  62 static struct resource pdcdata_resource = {
  63         .name   = "PDC data (Page Zero)",
  64         .start  = 0,
  65         .end    = 0x9ff,
  66         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM,
  67 };
  68
  69 static struct resource sysram_resources[MAX_PHYSMEM_RANGES] __ro_after_init;
  70
  71 /* The following array is initialized from the firmware specific
  72  * information retrieved in kernel/inventory.c.
  73  */
  74
  75 physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES] __initdata;
  76 int npmem_ranges __initdata;
  77
  78 #ifdef CONFIG_64BIT
  79 #define MAX_MEM         (1UL << MAX_PHYSMEM_BITS)
  80 #else /* !CONFIG_64BIT */
  81 #define MAX_MEM         (3584U*1024U*1024U)
  82 #endif /* !CONFIG_64BIT */
  83
  84 static unsigned long mem_limit __read_mostly = MAX_MEM;
  85
  86 static void __init mem_limit_func(void)
  87 {
  88         char *cp, *end;
  89         unsigned long limit;
  90
  91         /* We need this before __setup() functions are called */
  92
  93         limit = MAX_MEM;
  94         for (cp = boot_command_line; *cp; ) {
  95                 if (memcmp(cp, "mem=", 4) == 0) {
  96                         cp += 4;
  97                         limit = memparse(cp, &end);
  98                         if (end != cp)
  99                                 break;
 100                         cp = end;
 101                 } else {
 102                         while (*cp != ' ' && *cp)
 103                                 ++cp;
 104                         while (*cp == ' ')
 105                                 ++cp;
 106                 }
 107         }
 108
 109         if (limit < mem_limit)
 110                 mem_limit = limit;
 111 }
 112
 113 #define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
 114
 115 static void __init setup_bootmem(void)
 116 {
 117         unsigned long mem_max;
 118 #ifndef CONFIG_SPARSEMEM
 119         physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1];
 120         int npmem_holes;
 121 #endif
 122         int i, sysram_resource_count;
 123
 124         disable_sr_hashing(); /* Turn off space register hashing */
 125
 126         /*
 127          * Sort the ranges. Since the number of ranges is typically
 128          * small, and performance is not an issue here, just do
 129          * a simple insertion sort.
 130          */
 131
 132         for (i = 1; i < npmem_ranges; i++) {
 133                 int j;
 134
 135                 for (j = i; j > 0; j--) {
 136                         physmem_range_t tmp;
 137
 138                         if (pmem_ranges[j-1].start_pfn <
 139                             pmem_ranges[j].start_pfn) {
 140
 141                                 break;
 142                         }
 143                         tmp = pmem_ranges[j-1];
 144                         pmem_ranges[j-1] = pmem_ranges[j];
 145                         pmem_ranges[j] = tmp;
 146                 }
 147         }
 148
 149 #ifndef CONFIG_SPARSEMEM
 150         /*
 151          * Throw out ranges that are too far apart (controlled by
 152          * MAX_GAP).
 153          */
 154
 155         for (i = 1; i < npmem_ranges; i++) {
 156                 if (pmem_ranges[i].start_pfn -
 157                         (pmem_ranges[i-1].start_pfn +
 158                          pmem_ranges[i-1].pages) > MAX_GAP) {
 159                         npmem_ranges = i;
 160                         printk("Large gap in memory detected (%ld pages). "
 161                                "Consider turning on CONFIG_SPARSEMEM\n",
 162                                pmem_ranges[i].start_pfn -
 163                                (pmem_ranges[i-1].start_pfn +
 164                                 pmem_ranges[i-1].pages));
 165                         break;
 166                 }
 167         }
 168 #endif
 169
 170         /* Print the memory ranges */
 171         pr_info("Memory Ranges:\n");
 172
 173         for (i = 0; i < npmem_ranges; i++) {
 174                 struct resource *res = &sysram_resources[i];
 175                 unsigned long start;
 176                 unsigned long size;
 177
 178                 size = (pmem_ranges[i].pages << PAGE_SHIFT);
 179                 start = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
 180                 pr_info("%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n",
 181                         i, start, start + (size - 1), size >> 20);
 182
 183                 /* request memory resource */
 184                 res->name = "System RAM";
 185                 res->start = start;
 186                 res->end = start + size - 1;
 187                 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 188                 request_resource(&iomem_resource, res);
 189         }
 190
 191         sysram_resource_count = npmem_ranges;
 192
 193         /*
 194          * For 32 bit kernels we limit the amount of memory we can
 195          * support, in order to preserve enough kernel address space
 196          * for other purposes. For 64 bit kernels we don't normally
 197          * limit the memory, but this mechanism can be used to
 198          * artificially limit the amount of memory (and it is written
 199          * to work with multiple memory ranges).
 200          */
 201
 202         mem_limit_func();       /* check for "mem=" argument */
 203
 204         mem_max = 0;
 205         for (i = 0; i < npmem_ranges; i++) {
 206                 unsigned long rsize;
 207
 208                 rsize = pmem_ranges[i].pages << PAGE_SHIFT;
 209                 if ((mem_max + rsize) > mem_limit) {
 210                         printk(KERN_WARNING "Memory truncated to %ld MB\n", mem_limit >> 20);
 211                         if (mem_max == mem_limit)
 212                                 npmem_ranges = i;
 213                         else {
 214                                 pmem_ranges[i].pages =   (mem_limit >> PAGE_SHIFT)
 215                                                        - (mem_max >> PAGE_SHIFT);
 216                                 npmem_ranges = i + 1;
 217                                 mem_max = mem_limit;
 218                         }
 219                         break;
 220                 }
 221                 mem_max += rsize;
 222         }
 223
 224         printk(KERN_INFO "Total Memory: %ld MB\n",mem_max >> 20);
 225
 226 #ifndef CONFIG_SPARSEMEM
 227         /* Merge the ranges, keeping track of the holes */
 228         {
 229                 unsigned long end_pfn;
 230                 unsigned long hole_pages;
 231
 232                 npmem_holes = 0;
 233                 end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages;
 234                 for (i = 1; i < npmem_ranges; i++) {
 235
 236                         hole_pages = pmem_ranges[i].start_pfn - end_pfn;
 237                         if (hole_pages) {
 238                                 pmem_holes[npmem_holes].start_pfn = end_pfn;
 239                                 pmem_holes[npmem_holes++].pages = hole_pages;
 240                                 end_pfn += hole_pages;
 241                         }
 242                         end_pfn += pmem_ranges[i].pages;
 243                 }
 244
 245                 pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn;
 246                 npmem_ranges = 1;
 247         }
 248 #endif
 249
 250         /*
 251          * Initialize and free the full range of memory in each range.
 252          */
 253
 254         max_pfn = 0;
 255         for (i = 0; i < npmem_ranges; i++) {
 256                 unsigned long start_pfn;
 257                 unsigned long npages;
 258                 unsigned long start;
 259                 unsigned long size;
 260
 261                 start_pfn = pmem_ranges[i].start_pfn;
 262                 npages = pmem_ranges[i].pages;
 263
 264                 start = start_pfn << PAGE_SHIFT;
 265                 size = npages << PAGE_SHIFT;
 266
 267                 /* add system RAM memblock */
 268                 memblock_add(start, size);
 269
 270                 if ((start_pfn + npages) > max_pfn)
 271                         max_pfn = start_pfn + npages;
 272         }
 273
 274         /*
 275          * We can't use memblock top-down allocations because we only
 276          * created the initial mapping up to KERNEL_INITIAL_SIZE in
 277          * the assembly bootup code.
 278          */
 279         memblock_set_bottom_up(true);
 280
 281         /* IOMMU is always used to access "high mem" on those boxes
 282          * that can support enough mem that a PCI device couldn't
 283          * directly DMA to any physical addresses.
 284          * ISA DMA support will need to revisit this.
 285          */
 286         max_low_pfn = max_pfn;
 287
 288         /* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
 289
 290 #define PDC_CONSOLE_IO_IODC_SIZE 32768
 291
 292         memblock_reserve(0UL, (unsigned long)(PAGE0->mem_free +
 293                                 PDC_CONSOLE_IO_IODC_SIZE));
 294         memblock_reserve(__pa(KERNEL_BINARY_TEXT_START),
 295                         (unsigned long)(_end - KERNEL_BINARY_TEXT_START));
 296
 297 #ifndef CONFIG_SPARSEMEM
 298
 299         /* reserve the holes */
 300
 301         for (i = 0; i < npmem_holes; i++) {
 302                 memblock_reserve((pmem_holes[i].start_pfn << PAGE_SHIFT),
 303                                 (pmem_holes[i].pages << PAGE_SHIFT));
 304         }
 305 #endif
 306
 307 #ifdef CONFIG_BLK_DEV_INITRD
 308         if (initrd_start) {
 309                 printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
 310                 if (__pa(initrd_start) < mem_max) {
 311                         unsigned long initrd_reserve;
 312
 313                         if (__pa(initrd_end) > mem_max) {
 314                                 initrd_reserve = mem_max - __pa(initrd_start);
 315                         } else {
 316                                 initrd_reserve = initrd_end - initrd_start;
 317                         }
 318                         initrd_below_start_ok = 1;
 319                         printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
 320
 321                         memblock_reserve(__pa(initrd_start), initrd_reserve);
 322                 }
 323         }
 324 #endif
 325
 326         data_resource.start =  virt_to_phys(&data_start);
 327         data_resource.end = virt_to_phys(_end) - 1;
 328         code_resource.start = virt_to_phys(_text);
 329         code_resource.end = virt_to_phys(&data_start)-1;
 330
 331         /* We don't know which region the kernel will be in, so try
 332          * all of them.
 333          */
 334         for (i = 0; i < sysram_resource_count; i++) {
 335                 struct resource *res = &sysram_resources[i];
 336                 request_resource(res, &code_resource);
 337                 request_resource(res, &data_resource);
 338         }
 339         request_resource(&sysram_resources[0], &pdcdata_resource);
 340
 341         /* Initialize Page Deallocation Table (PDT) and check for bad memory. */
 342         pdc_pdt_init();
 343
 344         memblock_allow_resize();
 345         memblock_dump_all();
 346 }
 347
 348 static bool kernel_set_to_readonly;
 349
 350 static void __init map_pages(unsigned long start_vaddr,
 351                              unsigned long start_paddr, unsigned long size,
 352                              pgprot_t pgprot, int force)
 353 {
 354         pgd_t *pg_dir;
 355         pmd_t *pmd;
 356         pte_t *pg_table;
 357         unsigned long end_paddr;
 358         unsigned long start_pmd;
 359         unsigned long start_pte;
 360         unsigned long tmp1;
 361         unsigned long tmp2;
 362         unsigned long address;
 363         unsigned long vaddr;
 364         unsigned long ro_start;
 365         unsigned long ro_end;
 366         unsigned long kernel_start, kernel_end;
 367
 368         ro_start = __pa((unsigned long)_text);
 369         ro_end   = __pa((unsigned long)&data_start);
 370         kernel_start = __pa((unsigned long)&__init_begin);
 371         kernel_end  = __pa((unsigned long)&_end);
 372
 373         end_paddr = start_paddr + size;
 374
 375         pg_dir = pgd_offset_k(start_vaddr);
 376
 377 #if PTRS_PER_PMD == 1
 378         start_pmd = 0;
 379 #else
 380         start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
 381 #endif
 382         start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
 383
 384         address = start_paddr;
 385         vaddr = start_vaddr;
 386         while (address < end_paddr) {
 387 #if PTRS_PER_PMD == 1
 388                 pmd = (pmd_t *)__pa(pg_dir);
 389 #else
 390                 pmd = (pmd_t *)pgd_address(*pg_dir);
 391
 392                 /*
 393                  * pmd is physical at this point
 394                  */
 395
 396                 if (!pmd) {
 397                         pmd = memblock_alloc(PAGE_SIZE << PMD_ORDER,
 398                                              PAGE_SIZE << PMD_ORDER);
 399                         if (!pmd)
 400                                 panic("pmd allocation failed.\n");
 401                         pmd = (pmd_t *) __pa(pmd);
 402                 }
 403
 404                 pud_populate(NULL, (pud_t *)pg_dir, __va(pmd));
 405 #endif
 406                 pg_dir++;
 407
 408                 /* now change pmd to kernel virtual addresses */
 409
 410                 pmd = (pmd_t *)__va(pmd) + start_pmd;
 411                 for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++, pmd++) {
 412
 413                         /*
 414                          * pg_table is physical at this point
 415                          */
 416
 417                         pg_table = (pte_t *)pmd_address(*pmd);
 418                         if (!pg_table) {
 419                                 pg_table = memblock_alloc(PAGE_SIZE,
 420                                                           PAGE_SIZE);
 421                                 if (!pg_table)
 422                                         panic("page table allocation failed\n");
 423                                 pg_table = (pte_t *) __pa(pg_table);
 424                         }
 425
 426                         pmd_populate_kernel(NULL, pmd, __va(pg_table));
 427
 428                         /* now change pg_table to kernel virtual addresses */
 429
 430                         pg_table = (pte_t *) __va(pg_table) + start_pte;
 431                         for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++, pg_table++) {
 432                                 pte_t pte;
 433                                 pgprot_t prot;
 434                                 bool huge = false;
 435
 436                                 if (force) {
 437                                         prot = pgprot;
 438                                 } else if (address < kernel_start || address >= kernel_end) {
 439                                         /* outside kernel memory */
 440                                         prot = PAGE_KERNEL;
 441                                 } else if (!kernel_set_to_readonly) {
 442                                         /* still initializing, allow writing to RO memory */
 443                                         prot = PAGE_KERNEL_RWX;
 444                                         huge = true;
 445                                 } else if (address >= ro_start) {
 446                                         /* Code (ro) and Data areas */
 447                                         prot = (address < ro_end) ?
 448                                                 PAGE_KERNEL_EXEC : PAGE_KERNEL;
 449                                         huge = true;
 450                                 } else {
 451                                         prot = PAGE_KERNEL;
 452                                 }
 453
 454                                 pte = __mk_pte(address, prot);
 455                                 if (huge)
 456                                         pte = pte_mkhuge(pte);
 457
 458                                 if (address >= end_paddr)
 459                                         break;
 460
 461                                 set_pte(pg_table, pte);
 462
 463                                 address += PAGE_SIZE;
 464                                 vaddr += PAGE_SIZE;
 465                         }
 466                         start_pte = 0;
 467
 468                         if (address >= end_paddr)
 469                             break;
 470                 }
 471                 start_pmd = 0;
 472         }
 473 }
 474
 475 void __init set_kernel_text_rw(int enable_read_write)
 476 {
 477         unsigned long start = (unsigned long) __init_begin;
 478         unsigned long end   = (unsigned long) &data_start;
 479
 480         map_pages(start, __pa(start), end-start,
 481                 PAGE_KERNEL_RWX, enable_read_write ? 1:0);
 482
 483         /* force the kernel to see the new page table entries */
 484         flush_cache_all();
 485         flush_tlb_all();
 486 }
 487
 488 void __ref free_initmem(void)
 489 {
 490         unsigned long init_begin = (unsigned long)__init_begin;
 491         unsigned long init_end = (unsigned long)__init_end;
 492         unsigned long kernel_end  = (unsigned long)&_end;
 493
 494         /* Remap kernel text and data, but do not touch init section yet. */
 495         kernel_set_to_readonly = true;
 496         map_pages(init_end, __pa(init_end), kernel_end - init_end,
 497                   PAGE_KERNEL, 0);
 498
 499         /* The init text pages are marked R-X.  We have to
 500          * flush the icache and mark them RW-
 501          *
 502          * This is tricky, because map_pages is in the init section.
 503          * Do a dummy remap of the data section first (the data
 504          * section is already PAGE_KERNEL) to pull in the TLB entries
 505          * for map_kernel */
 506         map_pages(init_begin, __pa(init_begin), init_end - init_begin,
 507                   PAGE_KERNEL_RWX, 1);
 508         /* now remap at PAGE_KERNEL since the TLB is pre-primed to execute
 509          * map_pages */
 510         map_pages(init_begin, __pa(init_begin), init_end - init_begin,
 511                   PAGE_KERNEL, 1);
 512
 513         /* force the kernel to see the new TLB entries */
 514         __flush_tlb_range(0, init_begin, kernel_end);
 515
 516         /* finally dump all the instructions which were cached, since the
 517          * pages are no-longer executable */
 518         flush_icache_range(init_begin, init_end);
 519
 520         free_initmem_default(POISON_FREE_INITMEM);
 521
 522         /* set up a new led state on systems shipped LED State panel */
 523         pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
 524 }
 525
 526
 527 #ifdef CONFIG_STRICT_KERNEL_RWX
 528 void mark_rodata_ro(void)
 529 {
 530         /* rodata memory was already mapped with KERNEL_RO access rights by
 531            pagetable_init() and map_pages(). No need to do additional stuff here */
 532         unsigned long roai_size = __end_ro_after_init - __start_ro_after_init;
 533
 534         pr_info("Write protected read-only-after-init data: %luk\n", roai_size >> 10);
 535 }
 536 #endif
 537
 538
 539 /*
 540  * Just an arbitrary offset to serve as a "hole" between mapping areas
 541  * (between top of physical memory and a potential pcxl dma mapping
 542  * area, and below the vmalloc mapping area).
 543  *
 544  * The current 32K value just means that there will be a 32K "hole"
 545  * between mapping areas. That means that  any out-of-bounds memory
 546  * accesses will hopefully be caught. The vmalloc() routines leaves
 547  * a hole of 4kB between each vmalloced area for the same reason.
 548  */
 549
 550  /* Leave room for gateway page expansion */
 551 #if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
 552 #error KERNEL_MAP_START is in gateway reserved region
 553 #endif
 554 #define MAP_START (KERNEL_MAP_START)
 555
 556 #define VM_MAP_OFFSET  (32*1024)
 557 #define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
 558                                      & ~(VM_MAP_OFFSET-1)))
 559
 560 void *parisc_vmalloc_start __ro_after_init;
 561 EXPORT_SYMBOL(parisc_vmalloc_start);
 562
 563 #ifdef CONFIG_PA11
 564 unsigned long pcxl_dma_start __ro_after_init;
 565 #endif
 566
 567 void __init mem_init(void)
 568 {
 569         /* Do sanity checks on IPC (compat) structures */
 570         BUILD_BUG_ON(sizeof(struct ipc64_perm) != 48);
 571 #ifndef CONFIG_64BIT
 572         BUILD_BUG_ON(sizeof(struct semid64_ds) != 80);
 573         BUILD_BUG_ON(sizeof(struct msqid64_ds) != 104);
 574         BUILD_BUG_ON(sizeof(struct shmid64_ds) != 104);
 575 #endif
 576 #ifdef CONFIG_COMPAT
 577         BUILD_BUG_ON(sizeof(struct compat_ipc64_perm) != sizeof(struct ipc64_perm));
 578         BUILD_BUG_ON(sizeof(struct compat_semid64_ds) != 80);
 579         BUILD_BUG_ON(sizeof(struct compat_msqid64_ds) != 104);
 580         BUILD_BUG_ON(sizeof(struct compat_shmid64_ds) != 104);
 581 #endif
 582
 583         /* Do sanity checks on page table constants */
 584         BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t));
 585         BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t));
 586         BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t));
 587         BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD
 588                         > BITS_PER_LONG);
 589
 590         high_memory = __va((max_pfn << PAGE_SHIFT));
 591         set_max_mapnr(page_to_pfn(virt_to_page(high_memory - 1)) + 1);
 592         memblock_free_all();
 593
 594 #ifdef CONFIG_PA11
 595         if (boot_cpu_data.cpu_type == pcxl2 || boot_cpu_data.cpu_type == pcxl) {
 596                 pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
 597                 parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start
 598                                                 + PCXL_DMA_MAP_SIZE);
 599         } else
 600 #endif
 601                 parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
 602
 603         mem_init_print_info(NULL);
 604
 605 #if 0
 606         /*
 607          * Do not expose the virtual kernel memory layout to userspace.
 608          * But keep code for debugging purposes.
 609          */
 610         printk("virtual kernel memory layout:\n"
 611                "     vmalloc : 0x%px - 0x%px   (%4ld MB)\n"
 612                "     fixmap  : 0x%px - 0x%px   (%4ld kB)\n"
 613                "     memory  : 0x%px - 0x%px   (%4ld MB)\n"
 614                "       .init : 0x%px - 0x%px   (%4ld kB)\n"
 615                "       .data : 0x%px - 0x%px   (%4ld kB)\n"
 616                "       .text : 0x%px - 0x%px   (%4ld kB)\n",
 617
 618                (void*)VMALLOC_START, (void*)VMALLOC_END,
 619                (VMALLOC_END - VMALLOC_START) >> 20,
 620
 621                (void *)FIXMAP_START, (void *)(FIXMAP_START + FIXMAP_SIZE),
 622                (unsigned long)(FIXMAP_SIZE / 1024),
 623
 624                __va(0), high_memory,
 625                ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 626
 627                __init_begin, __init_end,
 628                ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10,
 629
 630                _etext, _edata,
 631                ((unsigned long)_edata - (unsigned long)_etext) >> 10,
 632
 633                _text, _etext,
 634                ((unsigned long)_etext - (unsigned long)_text) >> 10);
 635 #endif
 636 }
 637
 638 unsigned long *empty_zero_page __ro_after_init;
 639 EXPORT_SYMBOL(empty_zero_page);
 640
 641 /*
 642  * pagetable_init() sets up the page tables
 643  *
 644  * Note that gateway_init() places the Linux gateway page at page 0.
 645  * Since gateway pages cannot be dereferenced this has the desirable
 646  * side effect of trapping those pesky NULL-reference errors in the
 647  * kernel.
 648  */
 649 static void __init pagetable_init(void)
 650 {
 651         int range;
 652
 653         /* Map each physical memory range to its kernel vaddr */
 654
 655         for (range = 0; range < npmem_ranges; range++) {
 656                 unsigned long start_paddr;
 657                 unsigned long end_paddr;
 658                 unsigned long size;
 659
 660                 start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
 661                 size = pmem_ranges[range].pages << PAGE_SHIFT;
 662                 end_paddr = start_paddr + size;
 663
 664                 map_pages((unsigned long)__va(start_paddr), start_paddr,
 665                           size, PAGE_KERNEL, 0);
 666         }
 667
 668 #ifdef CONFIG_BLK_DEV_INITRD
 669         if (initrd_end && initrd_end > mem_limit) {
 670                 printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
 671                 map_pages(initrd_start, __pa(initrd_start),
 672                           initrd_end - initrd_start, PAGE_KERNEL, 0);
 673         }
 674 #endif
 675
 676         empty_zero_page = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
 677         if (!empty_zero_page)
 678                 panic("zero page allocation failed.\n");
 679
 680 }
 681
 682 static void __init gateway_init(void)
 683 {
 684         unsigned long linux_gateway_page_addr;
 685         /* FIXME: This is 'const' in order to trick the compiler
 686            into not treating it as DP-relative data. */
 687         extern void * const linux_gateway_page;
 688
 689         linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
 690
 691         /*
 692          * Setup Linux Gateway page.
 693          *
 694          * The Linux gateway page will reside in kernel space (on virtual
 695          * page 0), so it doesn't need to be aliased into user space.
 696          */
 697
 698         map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
 699                   PAGE_SIZE, PAGE_GATEWAY, 1);
 700 }
 701
 702 static void __init parisc_bootmem_free(void)
 703 {
 704         unsigned long zones_size[MAX_NR_ZONES] = { 0, };
 705         unsigned long holes_size[MAX_NR_ZONES] = { 0, };
 706         unsigned long mem_start_pfn = ~0UL, mem_end_pfn = 0, mem_size_pfn = 0;
 707         int i;
 708
 709         for (i = 0; i < npmem_ranges; i++) {
 710                 unsigned long start = pmem_ranges[i].start_pfn;
 711                 unsigned long size = pmem_ranges[i].pages;
 712                 unsigned long end = start + size;
 713
 714                 if (mem_start_pfn > start)
 715                         mem_start_pfn = start;
 716                 if (mem_end_pfn < end)
 717                         mem_end_pfn = end;
 718                 mem_size_pfn += size;
 719         }
 720
 721         zones_size[0] = mem_end_pfn - mem_start_pfn;
 722         holes_size[0] = zones_size[0] - mem_size_pfn;
 723
 724         free_area_init_node(0, zones_size, mem_start_pfn, holes_size);
 725 }
 726
 727 void __init paging_init(void)
 728 {
 729         setup_bootmem();
 730         pagetable_init();
 731         gateway_init();
 732         flush_cache_all_local(); /* start with known state */
 733         flush_tlb_all_local(NULL);
 734
 735         /*
 736          * Mark all memblocks as present for sparsemem using
 737          * memory_present() and then initialize sparsemem.
 738          */
 739         memblocks_present();
 740         sparse_init();
 741         parisc_bootmem_free();
 742 }
 743
 744 #ifdef CONFIG_PA20
 745
 746 /*
 747  * Currently, all PA20 chips have 18 bit protection IDs, which is the
 748  * limiting factor (space ids are 32 bits).
 749  */
 750
 751 #define NR_SPACE_IDS 262144
 752
 753 #else
 754
 755 /*
 756  * Currently we have a one-to-one relationship between space IDs and
 757  * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
 758  * support 15 bit protection IDs, so that is the limiting factor.
 759  * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
 760  * probably not worth the effort for a special case here.
 761  */
 762
 763 #define NR_SPACE_IDS 32768
 764
 765 #endif  /* !CONFIG_PA20 */
 766
 767 #define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
 768 #define SID_ARRAY_SIZE  (NR_SPACE_IDS / (8 * sizeof(long)))
 769
 770 static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
 771 static unsigned long dirty_space_id[SID_ARRAY_SIZE];
 772 static unsigned long space_id_index;
 773 static unsigned long free_space_ids = NR_SPACE_IDS - 1;
 774 static unsigned long dirty_space_ids = 0;
 775
 776 static DEFINE_SPINLOCK(sid_lock);
 777
 778 unsigned long alloc_sid(void)
 779 {
 780         unsigned long index;
 781
 782         spin_lock(&sid_lock);
 783
 784         if (free_space_ids == 0) {
 785                 if (dirty_space_ids != 0) {
 786                         spin_unlock(&sid_lock);
 787                         flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
 788                         spin_lock(&sid_lock);
 789                 }
 790                 BUG_ON(free_space_ids == 0);
 791         }
 792
 793         free_space_ids--;
 794
 795         index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
 796         space_id[index >> SHIFT_PER_LONG] |= (1L << (index & (BITS_PER_LONG - 1)));
 797         space_id_index = index;
 798
 799         spin_unlock(&sid_lock);
 800
 801         return index << SPACEID_SHIFT;
 802 }
 803
 804 void free_sid(unsigned long spaceid)
 805 {
 806         unsigned long index = spaceid >> SPACEID_SHIFT;
 807         unsigned long *dirty_space_offset;
 808
 809         dirty_space_offset = dirty_space_id + (index >> SHIFT_PER_LONG);
 810         index &= (BITS_PER_LONG - 1);
 811
 812         spin_lock(&sid_lock);
 813
 814         BUG_ON(*dirty_space_offset & (1L << index)); /* attempt to free space id twice */
 815
 816         *dirty_space_offset |= (1L << index);
 817         dirty_space_ids++;
 818
 819         spin_unlock(&sid_lock);
 820 }
 821
 822
 823 #ifdef CONFIG_SMP
 824 static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
 825 {
 826         int i;
 827
 828         /* NOTE: sid_lock must be held upon entry */
 829
 830         *ndirtyptr = dirty_space_ids;
 831         if (dirty_space_ids != 0) {
 832             for (i = 0; i < SID_ARRAY_SIZE; i++) {
 833                 dirty_array[i] = dirty_space_id[i];
 834                 dirty_space_id[i] = 0;
 835             }
 836             dirty_space_ids = 0;
 837         }
 838
 839         return;
 840 }
 841
 842 static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
 843 {
 844         int i;
 845
 846         /* NOTE: sid_lock must be held upon entry */
 847
 848         if (ndirty != 0) {
 849                 for (i = 0; i < SID_ARRAY_SIZE; i++) {
 850                         space_id[i] ^= dirty_array[i];
 851                 }
 852
 853                 free_space_ids += ndirty;
 854                 space_id_index = 0;
 855         }
 856 }
 857
 858 #else /* CONFIG_SMP */
 859
 860 static void recycle_sids(void)
 861 {
 862         int i;
 863
 864         /* NOTE: sid_lock must be held upon entry */
 865
 866         if (dirty_space_ids != 0) {
 867                 for (i = 0; i < SID_ARRAY_SIZE; i++) {
 868                         space_id[i] ^= dirty_space_id[i];
 869                         dirty_space_id[i] = 0;
 870                 }
 871
 872                 free_space_ids += dirty_space_ids;
 873                 dirty_space_ids = 0;
 874                 space_id_index = 0;
 875         }
 876 }
 877 #endif
 878
 879 /*
 880  * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
 881  * purged, we can safely reuse the space ids that were released but
 882  * not flushed from the tlb.
 883  */
 884
 885 #ifdef CONFIG_SMP
 886
 887 static unsigned long recycle_ndirty;
 888 static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
 889 static unsigned int recycle_inuse;
 890
 891 void flush_tlb_all(void)
 892 {
 893         int do_recycle;
 894
 895         __inc_irq_stat(irq_tlb_count);
 896         do_recycle = 0;
 897         spin_lock(&sid_lock);
 898         if (dirty_space_ids > RECYCLE_THRESHOLD) {
 899             BUG_ON(recycle_inuse);  /* FIXME: Use a semaphore/wait queue here */
 900             get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
 901             recycle_inuse++;
 902             do_recycle++;
 903         }
 904         spin_unlock(&sid_lock);
 905         on_each_cpu(flush_tlb_all_local, NULL, 1);
 906         if (do_recycle) {
 907             spin_lock(&sid_lock);
 908             recycle_sids(recycle_ndirty,recycle_dirty_array);
 909             recycle_inuse = 0;
 910             spin_unlock(&sid_lock);
 911         }
 912 }
 913 #else
 914 void flush_tlb_all(void)
 915 {
 916         __inc_irq_stat(irq_tlb_count);
 917         spin_lock(&sid_lock);
 918         flush_tlb_all_local(NULL);
 919         recycle_sids();
 920         spin_unlock(&sid_lock);
 921 }
 922 #endif