arch/mips/mm/init.c

   1 /*
   2  * This file is subject to the terms and conditions of the GNU General Public
   3  * License.  See the file "COPYING" in the main directory of this archive
   4  * for more details.
   5  *
   6  * Copyright (C) 1994 - 2000 Ralf Baechle
   7  * Copyright (C) 1999, 2000 Silicon Graphics, Inc.
   8  * Kevin D. Kissell, [email protected] and Carsten Langgaard, [email protected]
   9  * Copyright (C) 2000 MIPS Technologies, Inc.  All rights reserved.
  10  */
  11 #include <linux/bug.h>
  12 #include <linux/init.h>
  13 #include <linux/export.h>
  14 #include <linux/signal.h>
  15 #include <linux/sched.h>
  16 #include <linux/smp.h>
  17 #include <linux/kernel.h>
  18 #include <linux/errno.h>
  19 #include <linux/string.h>
  20 #include <linux/types.h>
  21 #include <linux/pagemap.h>
  22 #include <linux/ptrace.h>
  23 #include <linux/mman.h>
  24 #include <linux/mm.h>
  25 #include <linux/memblock.h>
  26 #include <linux/highmem.h>
  27 #include <linux/swap.h>
  28 #include <linux/proc_fs.h>
  29 #include <linux/pfn.h>
  30 #include <linux/hardirq.h>
  31 #include <linux/gfp.h>
  32 #include <linux/kcore.h>
  33 #include <linux/initrd.h>
  34 #include <linux/execmem.h>
  35
  36 #include <asm/bootinfo.h>
  37 #include <asm/cachectl.h>
  38 #include <asm/cpu.h>
  39 #include <asm/dma.h>
  40 #include <asm/maar.h>
  41 #include <asm/mmu_context.h>
  42 #include <asm/mmzone.h>
  43 #include <asm/sections.h>
  44 #include <asm/pgalloc.h>
  45 #include <asm/tlb.h>
  46 #include <asm/fixmap.h>
  47
  48 /*
  49  * We have up to 8 empty zeroed pages so we can map one of the right colour
  50  * when needed.  This is necessary only on R4000 / R4400 SC and MC versions
  51  * where we have to avoid VCED / VECI exceptions for good performance at
  52  * any price.  Since page is never written to after the initialization we
  53  * don't have to care about aliases on other CPUs.
  54  */
  55 unsigned long empty_zero_page, zero_page_mask;
  56 EXPORT_SYMBOL_GPL(empty_zero_page);
  57 EXPORT_SYMBOL(zero_page_mask);
  58
  59 /*
  60  * Not static inline because used by IP27 special magic initialization code
  61  */
  62 void setup_zero_pages(void)
  63 {
  64         unsigned int order, i;
  65         struct page *page;
  66
  67         if (cpu_has_vce)
  68                 order = 3;
  69         else
  70                 order = 0;
  71
  72         empty_zero_page = __get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
  73         if (!empty_zero_page)
  74                 panic("Oh boy, that early out of memory?");
  75
  76         page = virt_to_page((void *)empty_zero_page);
  77         split_page(page, order);
  78         for (i = 0; i < (1 << order); i++, page++)
  79                 mark_page_reserved(page);
  80
  81         zero_page_mask = ((PAGE_SIZE << order) - 1) & PAGE_MASK;
  82 }
  83
  84 static void *__kmap_pgprot(struct page *page, unsigned long addr, pgprot_t prot)
  85 {
  86         enum fixed_addresses idx;
  87         unsigned int old_mmid;
  88         unsigned long vaddr, flags, entrylo;
  89         unsigned long old_ctx;
  90         pte_t pte;
  91         int tlbidx;
  92
  93         BUG_ON(folio_test_dcache_dirty(page_folio(page)));
  94
  95         preempt_disable();
  96         pagefault_disable();
  97         idx = (addr >> PAGE_SHIFT) & (FIX_N_COLOURS - 1);
  98         idx += in_interrupt() ? FIX_N_COLOURS : 0;
  99         vaddr = __fix_to_virt(FIX_CMAP_END - idx);
 100         pte = mk_pte(page, prot);
 101 #if defined(CONFIG_XPA)
 102         entrylo = pte_to_entrylo(pte.pte_high);
 103 #elif defined(CONFIG_PHYS_ADDR_T_64BIT) && defined(CONFIG_CPU_MIPS32)
 104         entrylo = pte.pte_high;
 105 #else
 106         entrylo = pte_to_entrylo(pte_val(pte));
 107 #endif
 108
 109         local_irq_save(flags);
 110         old_ctx = read_c0_entryhi();
 111         write_c0_entryhi(vaddr & (PAGE_MASK << 1));
 112         write_c0_entrylo0(entrylo);
 113         write_c0_entrylo1(entrylo);
 114         if (cpu_has_mmid) {
 115                 old_mmid = read_c0_memorymapid();
 116                 write_c0_memorymapid(MMID_KERNEL_WIRED);
 117         }
 118 #ifdef CONFIG_XPA
 119         if (cpu_has_xpa) {
 120                 entrylo = (pte.pte_low & _PFNX_MASK);
 121                 writex_c0_entrylo0(entrylo);
 122                 writex_c0_entrylo1(entrylo);
 123         }
 124 #endif
 125         tlbidx = num_wired_entries();
 126         write_c0_wired(tlbidx + 1);
 127         write_c0_index(tlbidx);
 128         mtc0_tlbw_hazard();
 129         tlb_write_indexed();
 130         tlbw_use_hazard();
 131         write_c0_entryhi(old_ctx);
 132         if (cpu_has_mmid)
 133                 write_c0_memorymapid(old_mmid);
 134         local_irq_restore(flags);
 135
 136         return (void*) vaddr;
 137 }
 138
 139 void *kmap_coherent(struct page *page, unsigned long addr)
 140 {
 141         return __kmap_pgprot(page, addr, PAGE_KERNEL);
 142 }
 143
 144 void *kmap_noncoherent(struct page *page, unsigned long addr)
 145 {
 146         return __kmap_pgprot(page, addr, PAGE_KERNEL_NC);
 147 }
 148
 149 void kunmap_coherent(void)
 150 {
 151         unsigned int wired;
 152         unsigned long flags, old_ctx;
 153
 154         local_irq_save(flags);
 155         old_ctx = read_c0_entryhi();
 156         wired = num_wired_entries() - 1;
 157         write_c0_wired(wired);
 158         write_c0_index(wired);
 159         write_c0_entryhi(UNIQUE_ENTRYHI(wired));
 160         write_c0_entrylo0(0);
 161         write_c0_entrylo1(0);
 162         mtc0_tlbw_hazard();
 163         tlb_write_indexed();
 164         tlbw_use_hazard();
 165         write_c0_entryhi(old_ctx);
 166         local_irq_restore(flags);
 167         pagefault_enable();
 168         preempt_enable();
 169 }
 170
 171 void copy_user_highpage(struct page *to, struct page *from,
 172         unsigned long vaddr, struct vm_area_struct *vma)
 173 {
 174         struct folio *src = page_folio(from);
 175         void *vfrom, *vto;
 176
 177         vto = kmap_atomic(to);
 178         if (cpu_has_dc_aliases &&
 179             folio_mapped(src) && !folio_test_dcache_dirty(src)) {
 180                 vfrom = kmap_coherent(from, vaddr);
 181                 copy_page(vto, vfrom);
 182                 kunmap_coherent();
 183         } else {
 184                 vfrom = kmap_atomic(from);
 185                 copy_page(vto, vfrom);
 186                 kunmap_atomic(vfrom);
 187         }
 188         if ((!cpu_has_ic_fills_f_dc) ||
 189             pages_do_alias((unsigned long)vto, vaddr & PAGE_MASK))
 190                 flush_data_cache_page((unsigned long)vto);
 191         kunmap_atomic(vto);
 192         /* Make sure this page is cleared on other CPU's too before using it */
 193         smp_wmb();
 194 }
 195
 196 void copy_to_user_page(struct vm_area_struct *vma,
 197         struct page *page, unsigned long vaddr, void *dst, const void *src,
 198         unsigned long len)
 199 {
 200         struct folio *folio = page_folio(page);
 201
 202         if (cpu_has_dc_aliases &&
 203             folio_mapped(folio) && !folio_test_dcache_dirty(folio)) {
 204                 void *vto = kmap_coherent(page, vaddr) + (vaddr & ~PAGE_MASK);
 205                 memcpy(vto, src, len);
 206                 kunmap_coherent();
 207         } else {
 208                 memcpy(dst, src, len);
 209                 if (cpu_has_dc_aliases)
 210                         folio_set_dcache_dirty(folio);
 211         }
 212         if (vma->vm_flags & VM_EXEC)
 213                 flush_cache_page(vma, vaddr, page_to_pfn(page));
 214 }
 215
 216 void copy_from_user_page(struct vm_area_struct *vma,
 217         struct page *page, unsigned long vaddr, void *dst, const void *src,
 218         unsigned long len)
 219 {
 220         struct folio *folio = page_folio(page);
 221
 222         if (cpu_has_dc_aliases &&
 223             folio_mapped(folio) && !folio_test_dcache_dirty(folio)) {
 224                 void *vfrom = kmap_coherent(page, vaddr) + (vaddr & ~PAGE_MASK);
 225                 memcpy(dst, vfrom, len);
 226                 kunmap_coherent();
 227         } else {
 228                 memcpy(dst, src, len);
 229                 if (cpu_has_dc_aliases)
 230                         folio_set_dcache_dirty(folio);
 231         }
 232 }
 233 EXPORT_SYMBOL_GPL(copy_from_user_page);
 234
 235 void __init fixrange_init(unsigned long start, unsigned long end,
 236         pgd_t *pgd_base)
 237 {
 238 #ifdef CONFIG_HIGHMEM
 239         pgd_t *pgd;
 240         pud_t *pud;
 241         pmd_t *pmd;
 242         pte_t *pte;
 243         int i, j, k;
 244         unsigned long vaddr;
 245
 246         vaddr = start;
 247         i = pgd_index(vaddr);
 248         j = pud_index(vaddr);
 249         k = pmd_index(vaddr);
 250         pgd = pgd_base + i;
 251
 252         for ( ; (i < PTRS_PER_PGD) && (vaddr < end); pgd++, i++) {
 253                 pud = (pud_t *)pgd;
 254                 for ( ; (j < PTRS_PER_PUD) && (vaddr < end); pud++, j++) {
 255                         pmd = (pmd_t *)pud;
 256                         for (; (k < PTRS_PER_PMD) && (vaddr < end); pmd++, k++) {
 257                                 if (pmd_none(*pmd)) {
 258                                         pte = (pte_t *) memblock_alloc_low(PAGE_SIZE,
 259                                                                            PAGE_SIZE);
 260                                         if (!pte)
 261                                                 panic("%s: Failed to allocate %lu bytes align=%lx\n",
 262                                                       __func__, PAGE_SIZE,
 263                                                       PAGE_SIZE);
 264
 265                                         set_pmd(pmd, __pmd((unsigned long)pte));
 266                                         BUG_ON(pte != pte_offset_kernel(pmd, 0));
 267                                 }
 268                                 vaddr += PMD_SIZE;
 269                         }
 270                         k = 0;
 271                 }
 272                 j = 0;
 273         }
 274 #endif
 275 }
 276
 277 struct maar_walk_info {
 278         struct maar_config cfg[16];
 279         unsigned int num_cfg;
 280 };
 281
 282 static int maar_res_walk(unsigned long start_pfn, unsigned long nr_pages,
 283                          void *data)
 284 {
 285         struct maar_walk_info *wi = data;
 286         struct maar_config *cfg = &wi->cfg[wi->num_cfg];
 287         unsigned int maar_align;
 288
 289         /* MAAR registers hold physical addresses right shifted by 4 bits */
 290         maar_align = BIT(MIPS_MAAR_ADDR_SHIFT + 4);
 291
 292         /* Fill in the MAAR config entry */
 293         cfg->lower = ALIGN(PFN_PHYS(start_pfn), maar_align);
 294         cfg->upper = ALIGN_DOWN(PFN_PHYS(start_pfn + nr_pages), maar_align) - 1;
 295         cfg->attrs = MIPS_MAAR_S;
 296
 297         /* Ensure we don't overflow the cfg array */
 298         if (!WARN_ON(wi->num_cfg >= ARRAY_SIZE(wi->cfg)))
 299                 wi->num_cfg++;
 300
 301         return 0;
 302 }
 303
 304
 305 unsigned __weak platform_maar_init(unsigned num_pairs)
 306 {
 307         unsigned int num_configured;
 308         struct maar_walk_info wi;
 309
 310         wi.num_cfg = 0;
 311         walk_system_ram_range(0, max_pfn, &wi, maar_res_walk);
 312
 313         num_configured = maar_config(wi.cfg, wi.num_cfg, num_pairs);
 314         if (num_configured < wi.num_cfg)
 315                 pr_warn("Not enough MAAR pairs (%u) for all memory regions (%u)\n",
 316                         num_pairs, wi.num_cfg);
 317
 318         return num_configured;
 319 }
 320
 321 void maar_init(void)
 322 {
 323         unsigned num_maars, used, i;
 324         phys_addr_t lower, upper, attr;
 325         static struct {
 326                 struct maar_config cfgs[3];
 327                 unsigned used;
 328         } recorded = { { { 0 } }, 0 };
 329
 330         if (!cpu_has_maar)
 331                 return;
 332
 333         /* Detect the number of MAARs */
 334         write_c0_maari(~0);
 335         back_to_back_c0_hazard();
 336         num_maars = read_c0_maari() + 1;
 337
 338         /* MAARs should be in pairs */
 339         WARN_ON(num_maars % 2);
 340
 341         /* Set MAARs using values we recorded already */
 342         if (recorded.used) {
 343                 used = maar_config(recorded.cfgs, recorded.used, num_maars / 2);
 344                 BUG_ON(used != recorded.used);
 345         } else {
 346                 /* Configure the required MAARs */
 347                 used = platform_maar_init(num_maars / 2);
 348         }
 349
 350         /* Disable any further MAARs */
 351         for (i = (used * 2); i < num_maars; i++) {
 352                 write_c0_maari(i);
 353                 back_to_back_c0_hazard();
 354                 write_c0_maar(0);
 355                 back_to_back_c0_hazard();
 356         }
 357
 358         if (recorded.used)
 359                 return;
 360
 361         pr_info("MAAR configuration:\n");
 362         for (i = 0; i < num_maars; i += 2) {
 363                 write_c0_maari(i);
 364                 back_to_back_c0_hazard();
 365                 upper = read_c0_maar();
 366 #ifdef CONFIG_XPA
 367                 upper |= (phys_addr_t)readx_c0_maar() << MIPS_MAARX_ADDR_SHIFT;
 368 #endif
 369
 370                 write_c0_maari(i + 1);
 371                 back_to_back_c0_hazard();
 372                 lower = read_c0_maar();
 373 #ifdef CONFIG_XPA
 374                 lower |= (phys_addr_t)readx_c0_maar() << MIPS_MAARX_ADDR_SHIFT;
 375 #endif
 376
 377                 attr = lower & upper;
 378                 lower = (lower & MIPS_MAAR_ADDR) << 4;
 379                 upper = ((upper & MIPS_MAAR_ADDR) << 4) | 0xffff;
 380
 381                 pr_info("  [%d]: ", i / 2);
 382                 if ((attr & MIPS_MAAR_V) != MIPS_MAAR_V) {
 383                         pr_cont("disabled\n");
 384                         continue;
 385                 }
 386
 387                 pr_cont("%pa-%pa", &lower, &upper);
 388
 389                 if (attr & MIPS_MAAR_S)
 390                         pr_cont(" speculate");
 391
 392                 pr_cont("\n");
 393
 394                 /* Record the setup for use on secondary CPUs */
 395                 if (used <= ARRAY_SIZE(recorded.cfgs)) {
 396                         recorded.cfgs[recorded.used].lower = lower;
 397                         recorded.cfgs[recorded.used].upper = upper;
 398                         recorded.cfgs[recorded.used].attrs = attr;
 399                         recorded.used++;
 400                 }
 401         }
 402 }
 403
 404 #ifndef CONFIG_NUMA
 405 void __init paging_init(void)
 406 {
 407         unsigned long max_zone_pfns[MAX_NR_ZONES];
 408
 409         pagetable_init();
 410
 411 #ifdef CONFIG_ZONE_DMA
 412         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
 413 #endif
 414 #ifdef CONFIG_ZONE_DMA32
 415         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
 416 #endif
 417         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 418 #ifdef CONFIG_HIGHMEM
 419         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 420
 421         if (cpu_has_dc_aliases && max_low_pfn != highend_pfn) {
 422                 printk(KERN_WARNING "This processor doesn't support highmem."
 423                        " %ldk highmem ignored\n",
 424                        (highend_pfn - max_low_pfn) << (PAGE_SHIFT - 10));
 425                 max_zone_pfns[ZONE_HIGHMEM] = max_low_pfn;
 426
 427                 max_mapnr = max_low_pfn;
 428         } else if (highend_pfn) {
 429                 max_mapnr = highend_pfn;
 430         } else {
 431                 max_mapnr = max_low_pfn;
 432         }
 433 #else
 434         max_mapnr = max_low_pfn;
 435 #endif
 436         high_memory = (void *) __va(max_low_pfn << PAGE_SHIFT);
 437
 438         free_area_init(max_zone_pfns);
 439 }
 440
 441 #ifdef CONFIG_64BIT
 442 static struct kcore_list kcore_kseg0;
 443 #endif
 444
 445 static inline void __init mem_init_free_highmem(void)
 446 {
 447 #ifdef CONFIG_HIGHMEM
 448         unsigned long tmp;
 449
 450         if (cpu_has_dc_aliases)
 451                 return;
 452
 453         for (tmp = highstart_pfn; tmp < highend_pfn; tmp++) {
 454                 struct page *page = pfn_to_page(tmp);
 455
 456                 if (!memblock_is_memory(PFN_PHYS(tmp)))
 457                         SetPageReserved(page);
 458                 else
 459                         free_highmem_page(page);
 460         }
 461 #endif
 462 }
 463
 464 void __init mem_init(void)
 465 {
 466         /*
 467          * When PFN_PTE_SHIFT is greater than PAGE_SHIFT we won't have enough PTE
 468          * bits to hold a full 32b physical address on MIPS32 systems.
 469          */
 470         BUILD_BUG_ON(IS_ENABLED(CONFIG_32BIT) && (PFN_PTE_SHIFT > PAGE_SHIFT));
 471
 472         maar_init();
 473         memblock_free_all();
 474         setup_zero_pages();     /* Setup zeroed pages.  */
 475         mem_init_free_highmem();
 476
 477 #ifdef CONFIG_64BIT
 478         if ((unsigned long) &_text > (unsigned long) CKSEG0)
 479                 /* The -4 is a hack so that user tools don't have to handle
 480                    the overflow.  */
 481                 kclist_add(&kcore_kseg0, (void *) CKSEG0,
 482                                 0x80000000 - 4, KCORE_TEXT);
 483 #endif
 484 }
 485 #endif /* !CONFIG_NUMA */
 486
 487 void free_init_pages(const char *what, unsigned long begin, unsigned long end)
 488 {
 489         unsigned long pfn;
 490
 491         for (pfn = PFN_UP(begin); pfn < PFN_DOWN(end); pfn++) {
 492                 struct page *page = pfn_to_page(pfn);
 493                 void *addr = phys_to_virt(PFN_PHYS(pfn));
 494
 495                 memset(addr, POISON_FREE_INITMEM, PAGE_SIZE);
 496                 free_reserved_page(page);
 497         }
 498         printk(KERN_INFO "Freeing %s: %ldk freed\n", what, (end - begin) >> 10);
 499 }
 500
 501 void (*free_init_pages_eva)(void *begin, void *end) = NULL;
 502
 503 void __weak __init prom_free_prom_memory(void)
 504 {
 505         /* nothing to do */
 506 }
 507
 508 void __ref free_initmem(void)
 509 {
 510         prom_free_prom_memory();
 511         /*
 512          * Let the platform define a specific function to free the
 513          * init section since EVA may have used any possible mapping
 514          * between virtual and physical addresses.
 515          */
 516         if (free_init_pages_eva)
 517                 free_init_pages_eva((void *)&__init_begin, (void *)&__init_end);
 518         else
 519                 free_initmem_default(POISON_FREE_INITMEM);
 520 }
 521
 522 #ifdef CONFIG_HAVE_SETUP_PER_CPU_AREA
 523 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
 524 EXPORT_SYMBOL(__per_cpu_offset);
 525
 526 static int __init pcpu_cpu_distance(unsigned int from, unsigned int to)
 527 {
 528         return node_distance(cpu_to_node(from), cpu_to_node(to));
 529 }
 530
 531 static int __init pcpu_cpu_to_node(int cpu)
 532 {
 533         return cpu_to_node(cpu);
 534 }
 535
 536 void __init setup_per_cpu_areas(void)
 537 {
 538         unsigned long delta;
 539         unsigned int cpu;
 540         int rc;
 541
 542         /*
 543          * Always reserve area for module percpu variables.  That's
 544          * what the legacy allocator did.
 545          */
 546         rc = pcpu_embed_first_chunk(PERCPU_MODULE_RESERVE,
 547                                     PERCPU_DYNAMIC_RESERVE, PAGE_SIZE,
 548                                     pcpu_cpu_distance,
 549                                     pcpu_cpu_to_node);
 550         if (rc < 0)
 551                 panic("Failed to initialize percpu areas.");
 552
 553         delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
 554         for_each_possible_cpu(cpu)
 555                 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
 556 }
 557 #endif
 558
 559 #ifndef CONFIG_MIPS_PGD_C0_CONTEXT
 560 unsigned long pgd_current[NR_CPUS];
 561 #endif
 562
 563 /*
 564  * Align swapper_pg_dir in to 64K, allows its address to be loaded
 565  * with a single LUI instruction in the TLB handlers.  If we used
 566  * __aligned(64K), its size would get rounded up to the alignment
 567  * size, and waste space.  So we place it in its own section and align
 568  * it in the linker script.
 569  */
 570 pgd_t swapper_pg_dir[PTRS_PER_PGD] __section(".bss..swapper_pg_dir");
 571 #ifndef __PAGETABLE_PUD_FOLDED
 572 pud_t invalid_pud_table[PTRS_PER_PUD] __page_aligned_bss;
 573 #endif
 574 #ifndef __PAGETABLE_PMD_FOLDED
 575 pmd_t invalid_pmd_table[PTRS_PER_PMD] __page_aligned_bss;
 576 EXPORT_SYMBOL_GPL(invalid_pmd_table);
 577 #endif
 578 pte_t invalid_pte_table[PTRS_PER_PTE] __page_aligned_bss;
 579 EXPORT_SYMBOL(invalid_pte_table);
 580
 581 #ifdef CONFIG_EXECMEM
 582 #ifdef MODULES_VADDR
 583 static struct execmem_info execmem_info __ro_after_init;
 584
 585 struct execmem_info __init *execmem_arch_setup(void)
 586 {
 587         execmem_info = (struct execmem_info){
 588                 .ranges = {
 589                         [EXECMEM_DEFAULT] = {
 590                                 .start  = MODULES_VADDR,
 591                                 .end    = MODULES_END,
 592                                 .pgprot = PAGE_KERNEL,
 593                                 .alignment = 1,
 594                         },
 595                 },
 596         };
 597
 598         return &execmem_info;
 599 }
 600 #endif
 601 #endif /* CONFIG_EXECMEM */