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] __read_mostly; 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 int __init parisc_text_address(unsigned long vaddr) 349 { 350 static unsigned long head_ptr __initdata; 351 352 if (!head_ptr) 353 head_ptr = PAGE_MASK & (unsigned long) 354 dereference_function_descriptor(&parisc_kernel_start); 355 356 return core_kernel_text(vaddr) || vaddr == head_ptr; 357 } 358 359 static void __init map_pages(unsigned long start_vaddr, 360 unsigned long start_paddr, unsigned long size, 361 pgprot_t pgprot, int force) 362 { 363 pgd_t *pg_dir; 364 pmd_t *pmd; 365 pte_t *pg_table; 366 unsigned long end_paddr; 367 unsigned long start_pmd; 368 unsigned long start_pte; 369 unsigned long tmp1; 370 unsigned long tmp2; 371 unsigned long address; 372 unsigned long vaddr; 373 unsigned long ro_start; 374 unsigned long ro_end; 375 unsigned long kernel_end; 376 377 ro_start = __pa((unsigned long)_text); 378 ro_end = __pa((unsigned long)&data_start); 379 kernel_end = __pa((unsigned long)&_end); 380 381 end_paddr = start_paddr + size; 382 383 pg_dir = pgd_offset_k(start_vaddr); 384 385 #if PTRS_PER_PMD == 1 386 start_pmd = 0; 387 #else 388 start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1)); 389 #endif 390 start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)); 391 392 address = start_paddr; 393 vaddr = start_vaddr; 394 while (address < end_paddr) { 395 #if PTRS_PER_PMD == 1 396 pmd = (pmd_t *)__pa(pg_dir); 397 #else 398 pmd = (pmd_t *)pgd_address(*pg_dir); 399 400 /* 401 * pmd is physical at this point 402 */ 403 404 if (!pmd) { 405 pmd = memblock_alloc(PAGE_SIZE << PMD_ORDER, 406 PAGE_SIZE << PMD_ORDER); 407 if (!pmd) 408 panic("pmd allocation failed.\n"); 409 pmd = (pmd_t *) __pa(pmd); 410 } 411 412 pgd_populate(NULL, pg_dir, __va(pmd)); 413 #endif 414 pg_dir++; 415 416 /* now change pmd to kernel virtual addresses */ 417 418 pmd = (pmd_t *)__va(pmd) + start_pmd; 419 for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++, pmd++) { 420 421 /* 422 * pg_table is physical at this point 423 */ 424 425 pg_table = (pte_t *)pmd_address(*pmd); 426 if (!pg_table) { 427 pg_table = memblock_alloc(PAGE_SIZE, 428 PAGE_SIZE); 429 if (!pg_table) 430 panic("page table allocation failed\n"); 431 pg_table = (pte_t *) __pa(pg_table); 432 } 433 434 pmd_populate_kernel(NULL, pmd, __va(pg_table)); 435 436 /* now change pg_table to kernel virtual addresses */ 437 438 pg_table = (pte_t *) __va(pg_table) + start_pte; 439 for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++, pg_table++) { 440 pte_t pte; 441 442 if (force) 443 pte = __mk_pte(address, pgprot); 444 else if (parisc_text_address(vaddr)) { 445 pte = __mk_pte(address, PAGE_KERNEL_EXEC); 446 if (address >= ro_start && address < kernel_end) 447 pte = pte_mkhuge(pte); 448 } 449 else 450 #if defined(CONFIG_PARISC_PAGE_SIZE_4KB) 451 if (address >= ro_start && address < ro_end) { 452 pte = __mk_pte(address, PAGE_KERNEL_EXEC); 453 pte = pte_mkhuge(pte); 454 } else 455 #endif 456 { 457 pte = __mk_pte(address, pgprot); 458 if (address >= ro_start && address < kernel_end) 459 pte = pte_mkhuge(pte); 460 } 461 462 if (address >= end_paddr) 463 break; 464 465 set_pte(pg_table, pte); 466 467 address += PAGE_SIZE; 468 vaddr += PAGE_SIZE; 469 } 470 start_pte = 0; 471 472 if (address >= end_paddr) 473 break; 474 } 475 start_pmd = 0; 476 } 477 } 478 479 void __init set_kernel_text_rw(int enable_read_write) 480 { 481 unsigned long start = (unsigned long) __init_begin; 482 unsigned long end = (unsigned long) &data_start; 483 484 map_pages(start, __pa(start), end-start, 485 PAGE_KERNEL_RWX, enable_read_write ? 1:0); 486 487 /* force the kernel to see the new page table entries */ 488 flush_cache_all(); 489 flush_tlb_all(); 490 } 491 492 void __ref free_initmem(void) 493 { 494 unsigned long init_begin = (unsigned long)__init_begin; 495 unsigned long init_end = (unsigned long)__init_end; 496 497 /* The init text pages are marked R-X. We have to 498 * flush the icache and mark them RW- 499 * 500 * This is tricky, because map_pages is in the init section. 501 * Do a dummy remap of the data section first (the data 502 * section is already PAGE_KERNEL) to pull in the TLB entries 503 * for map_kernel */ 504 map_pages(init_begin, __pa(init_begin), init_end - init_begin, 505 PAGE_KERNEL_RWX, 1); 506 /* now remap at PAGE_KERNEL since the TLB is pre-primed to execute 507 * map_pages */ 508 map_pages(init_begin, __pa(init_begin), init_end - init_begin, 509 PAGE_KERNEL, 1); 510 511 /* force the kernel to see the new TLB entries */ 512 __flush_tlb_range(0, init_begin, init_end); 513 514 /* finally dump all the instructions which were cached, since the 515 * pages are no-longer executable */ 516 flush_icache_range(init_begin, init_end); 517 518 free_initmem_default(POISON_FREE_INITMEM); 519 520 /* set up a new led state on systems shipped LED State panel */ 521 pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE); 522 } 523 524 525 #ifdef CONFIG_STRICT_KERNEL_RWX 526 void mark_rodata_ro(void) 527 { 528 /* rodata memory was already mapped with KERNEL_RO access rights by 529 pagetable_init() and map_pages(). No need to do additional stuff here */ 530 printk (KERN_INFO "Write protecting the kernel read-only data: %luk\n", 531 (unsigned long)(__end_rodata - __start_rodata) >> 10); 532 } 533 #endif 534 535 536 /* 537 * Just an arbitrary offset to serve as a "hole" between mapping areas 538 * (between top of physical memory and a potential pcxl dma mapping 539 * area, and below the vmalloc mapping area). 540 * 541 * The current 32K value just means that there will be a 32K "hole" 542 * between mapping areas. That means that any out-of-bounds memory 543 * accesses will hopefully be caught. The vmalloc() routines leaves 544 * a hole of 4kB between each vmalloced area for the same reason. 545 */ 546 547 /* Leave room for gateway page expansion */ 548 #if KERNEL_MAP_START < GATEWAY_PAGE_SIZE 549 #error KERNEL_MAP_START is in gateway reserved region 550 #endif 551 #define MAP_START (KERNEL_MAP_START) 552 553 #define VM_MAP_OFFSET (32*1024) 554 #define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \ 555 & ~(VM_MAP_OFFSET-1))) 556 557 void *parisc_vmalloc_start __read_mostly; 558 EXPORT_SYMBOL(parisc_vmalloc_start); 559 560 #ifdef CONFIG_PA11 561 unsigned long pcxl_dma_start __read_mostly; 562 #endif 563 564 void __init mem_init(void) 565 { 566 /* Do sanity checks on IPC (compat) structures */ 567 BUILD_BUG_ON(sizeof(struct ipc64_perm) != 48); 568 #ifndef CONFIG_64BIT 569 BUILD_BUG_ON(sizeof(struct semid64_ds) != 80); 570 BUILD_BUG_ON(sizeof(struct msqid64_ds) != 104); 571 BUILD_BUG_ON(sizeof(struct shmid64_ds) != 104); 572 #endif 573 #ifdef CONFIG_COMPAT 574 BUILD_BUG_ON(sizeof(struct compat_ipc64_perm) != sizeof(struct ipc64_perm)); 575 BUILD_BUG_ON(sizeof(struct compat_semid64_ds) != 80); 576 BUILD_BUG_ON(sizeof(struct compat_msqid64_ds) != 104); 577 BUILD_BUG_ON(sizeof(struct compat_shmid64_ds) != 104); 578 #endif 579 580 /* Do sanity checks on page table constants */ 581 BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t)); 582 BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t)); 583 BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t)); 584 BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD 585 > BITS_PER_LONG); 586 587 high_memory = __va((max_pfn << PAGE_SHIFT)); 588 set_max_mapnr(page_to_pfn(virt_to_page(high_memory - 1)) + 1); 589 memblock_free_all(); 590 591 #ifdef CONFIG_PA11 592 if (boot_cpu_data.cpu_type == pcxl2 || boot_cpu_data.cpu_type == pcxl) { 593 pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START); 594 parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start 595 + PCXL_DMA_MAP_SIZE); 596 } else 597 #endif 598 parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START); 599 600 mem_init_print_info(NULL); 601 602 #if 0 603 /* 604 * Do not expose the virtual kernel memory layout to userspace. 605 * But keep code for debugging purposes. 606 */ 607 printk("virtual kernel memory layout:\n" 608 " vmalloc : 0x%px - 0x%px (%4ld MB)\n" 609 " fixmap : 0x%px - 0x%px (%4ld kB)\n" 610 " memory : 0x%px - 0x%px (%4ld MB)\n" 611 " .init : 0x%px - 0x%px (%4ld kB)\n" 612 " .data : 0x%px - 0x%px (%4ld kB)\n" 613 " .text : 0x%px - 0x%px (%4ld kB)\n", 614 615 (void*)VMALLOC_START, (void*)VMALLOC_END, 616 (VMALLOC_END - VMALLOC_START) >> 20, 617 618 (void *)FIXMAP_START, (void *)(FIXMAP_START + FIXMAP_SIZE), 619 (unsigned long)(FIXMAP_SIZE / 1024), 620 621 __va(0), high_memory, 622 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20, 623 624 __init_begin, __init_end, 625 ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10, 626 627 _etext, _edata, 628 ((unsigned long)_edata - (unsigned long)_etext) >> 10, 629 630 _text, _etext, 631 ((unsigned long)_etext - (unsigned long)_text) >> 10); 632 #endif 633 } 634 635 unsigned long *empty_zero_page __read_mostly; 636 EXPORT_SYMBOL(empty_zero_page); 637 638 /* 639 * pagetable_init() sets up the page tables 640 * 641 * Note that gateway_init() places the Linux gateway page at page 0. 642 * Since gateway pages cannot be dereferenced this has the desirable 643 * side effect of trapping those pesky NULL-reference errors in the 644 * kernel. 645 */ 646 static void __init pagetable_init(void) 647 { 648 int range; 649 650 /* Map each physical memory range to its kernel vaddr */ 651 652 for (range = 0; range < npmem_ranges; range++) { 653 unsigned long start_paddr; 654 unsigned long end_paddr; 655 unsigned long size; 656 657 start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT; 658 size = pmem_ranges[range].pages << PAGE_SHIFT; 659 end_paddr = start_paddr + size; 660 661 map_pages((unsigned long)__va(start_paddr), start_paddr, 662 size, PAGE_KERNEL, 0); 663 } 664 665 #ifdef CONFIG_BLK_DEV_INITRD 666 if (initrd_end && initrd_end > mem_limit) { 667 printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end); 668 map_pages(initrd_start, __pa(initrd_start), 669 initrd_end - initrd_start, PAGE_KERNEL, 0); 670 } 671 #endif 672 673 empty_zero_page = memblock_alloc(PAGE_SIZE, PAGE_SIZE); 674 if (!empty_zero_page) 675 panic("zero page allocation failed.\n"); 676 677 } 678 679 static void __init gateway_init(void) 680 { 681 unsigned long linux_gateway_page_addr; 682 /* FIXME: This is 'const' in order to trick the compiler 683 into not treating it as DP-relative data. */ 684 extern void * const linux_gateway_page; 685 686 linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK; 687 688 /* 689 * Setup Linux Gateway page. 690 * 691 * The Linux gateway page will reside in kernel space (on virtual 692 * page 0), so it doesn't need to be aliased into user space. 693 */ 694 695 map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page), 696 PAGE_SIZE, PAGE_GATEWAY, 1); 697 } 698 699 static void __init parisc_bootmem_free(void) 700 { 701 unsigned long zones_size[MAX_NR_ZONES] = { 0, }; 702 unsigned long holes_size[MAX_NR_ZONES] = { 0, }; 703 unsigned long mem_start_pfn = ~0UL, mem_end_pfn = 0, mem_size_pfn = 0; 704 int i; 705 706 for (i = 0; i < npmem_ranges; i++) { 707 unsigned long start = pmem_ranges[i].start_pfn; 708 unsigned long size = pmem_ranges[i].pages; 709 unsigned long end = start + size; 710 711 if (mem_start_pfn > start) 712 mem_start_pfn = start; 713 if (mem_end_pfn < end) 714 mem_end_pfn = end; 715 mem_size_pfn += size; 716 } 717 718 zones_size[0] = mem_end_pfn - mem_start_pfn; 719 holes_size[0] = zones_size[0] - mem_size_pfn; 720 721 free_area_init_node(0, zones_size, mem_start_pfn, holes_size); 722 } 723 724 void __init paging_init(void) 725 { 726 setup_bootmem(); 727 pagetable_init(); 728 gateway_init(); 729 flush_cache_all_local(); /* start with known state */ 730 flush_tlb_all_local(NULL); 731 732 /* 733 * Mark all memblocks as present for sparsemem using 734 * memory_present() and then initialize sparsemem. 735 */ 736 memblocks_present(); 737 sparse_init(); 738 parisc_bootmem_free(); 739 } 740 741 #ifdef CONFIG_PA20 742 743 /* 744 * Currently, all PA20 chips have 18 bit protection IDs, which is the 745 * limiting factor (space ids are 32 bits). 746 */ 747 748 #define NR_SPACE_IDS 262144 749 750 #else 751 752 /* 753 * Currently we have a one-to-one relationship between space IDs and 754 * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only 755 * support 15 bit protection IDs, so that is the limiting factor. 756 * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's 757 * probably not worth the effort for a special case here. 758 */ 759 760 #define NR_SPACE_IDS 32768 761 762 #endif /* !CONFIG_PA20 */ 763 764 #define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2) 765 #define SID_ARRAY_SIZE (NR_SPACE_IDS / (8 * sizeof(long))) 766 767 static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */ 768 static unsigned long dirty_space_id[SID_ARRAY_SIZE]; 769 static unsigned long space_id_index; 770 static unsigned long free_space_ids = NR_SPACE_IDS - 1; 771 static unsigned long dirty_space_ids = 0; 772 773 static DEFINE_SPINLOCK(sid_lock); 774 775 unsigned long alloc_sid(void) 776 { 777 unsigned long index; 778 779 spin_lock(&sid_lock); 780 781 if (free_space_ids == 0) { 782 if (dirty_space_ids != 0) { 783 spin_unlock(&sid_lock); 784 flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */ 785 spin_lock(&sid_lock); 786 } 787 BUG_ON(free_space_ids == 0); 788 } 789 790 free_space_ids--; 791 792 index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index); 793 space_id[index >> SHIFT_PER_LONG] |= (1L << (index & (BITS_PER_LONG - 1))); 794 space_id_index = index; 795 796 spin_unlock(&sid_lock); 797 798 return index << SPACEID_SHIFT; 799 } 800 801 void free_sid(unsigned long spaceid) 802 { 803 unsigned long index = spaceid >> SPACEID_SHIFT; 804 unsigned long *dirty_space_offset; 805 806 dirty_space_offset = dirty_space_id + (index >> SHIFT_PER_LONG); 807 index &= (BITS_PER_LONG - 1); 808 809 spin_lock(&sid_lock); 810 811 BUG_ON(*dirty_space_offset & (1L << index)); /* attempt to free space id twice */ 812 813 *dirty_space_offset |= (1L << index); 814 dirty_space_ids++; 815 816 spin_unlock(&sid_lock); 817 } 818 819 820 #ifdef CONFIG_SMP 821 static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array) 822 { 823 int i; 824 825 /* NOTE: sid_lock must be held upon entry */ 826 827 *ndirtyptr = dirty_space_ids; 828 if (dirty_space_ids != 0) { 829 for (i = 0; i < SID_ARRAY_SIZE; i++) { 830 dirty_array[i] = dirty_space_id[i]; 831 dirty_space_id[i] = 0; 832 } 833 dirty_space_ids = 0; 834 } 835 836 return; 837 } 838 839 static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array) 840 { 841 int i; 842 843 /* NOTE: sid_lock must be held upon entry */ 844 845 if (ndirty != 0) { 846 for (i = 0; i < SID_ARRAY_SIZE; i++) { 847 space_id[i] ^= dirty_array[i]; 848 } 849 850 free_space_ids += ndirty; 851 space_id_index = 0; 852 } 853 } 854 855 #else /* CONFIG_SMP */ 856 857 static void recycle_sids(void) 858 { 859 int i; 860 861 /* NOTE: sid_lock must be held upon entry */ 862 863 if (dirty_space_ids != 0) { 864 for (i = 0; i < SID_ARRAY_SIZE; i++) { 865 space_id[i] ^= dirty_space_id[i]; 866 dirty_space_id[i] = 0; 867 } 868 869 free_space_ids += dirty_space_ids; 870 dirty_space_ids = 0; 871 space_id_index = 0; 872 } 873 } 874 #endif 875 876 /* 877 * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is 878 * purged, we can safely reuse the space ids that were released but 879 * not flushed from the tlb. 880 */ 881 882 #ifdef CONFIG_SMP 883 884 static unsigned long recycle_ndirty; 885 static unsigned long recycle_dirty_array[SID_ARRAY_SIZE]; 886 static unsigned int recycle_inuse; 887 888 void flush_tlb_all(void) 889 { 890 int do_recycle; 891 892 __inc_irq_stat(irq_tlb_count); 893 do_recycle = 0; 894 spin_lock(&sid_lock); 895 if (dirty_space_ids > RECYCLE_THRESHOLD) { 896 BUG_ON(recycle_inuse); /* FIXME: Use a semaphore/wait queue here */ 897 get_dirty_sids(&recycle_ndirty,recycle_dirty_array); 898 recycle_inuse++; 899 do_recycle++; 900 } 901 spin_unlock(&sid_lock); 902 on_each_cpu(flush_tlb_all_local, NULL, 1); 903 if (do_recycle) { 904 spin_lock(&sid_lock); 905 recycle_sids(recycle_ndirty,recycle_dirty_array); 906 recycle_inuse = 0; 907 spin_unlock(&sid_lock); 908 } 909 } 910 #else 911 void flush_tlb_all(void) 912 { 913 __inc_irq_stat(irq_tlb_count); 914 spin_lock(&sid_lock); 915 flush_tlb_all_local(NULL); 916 recycle_sids(); 917 spin_unlock(&sid_lock); 918 } 919 #endif 920