1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Initialize MMU support. 4 * 5 * Copyright (C) 1998-2003 Hewlett-Packard Co 6 * David Mosberger-Tang <davidm@hpl.hp.com> 7 */ 8 #include <linux/kernel.h> 9 #include <linux/init.h> 10 11 #include <linux/dma-map-ops.h> 12 #include <linux/dmar.h> 13 #include <linux/efi.h> 14 #include <linux/elf.h> 15 #include <linux/memblock.h> 16 #include <linux/mm.h> 17 #include <linux/sched/signal.h> 18 #include <linux/mmzone.h> 19 #include <linux/module.h> 20 #include <linux/personality.h> 21 #include <linux/reboot.h> 22 #include <linux/slab.h> 23 #include <linux/swap.h> 24 #include <linux/proc_fs.h> 25 #include <linux/bitops.h> 26 #include <linux/kexec.h> 27 #include <linux/swiotlb.h> 28 29 #include <asm/dma.h> 30 #include <asm/efi.h> 31 #include <asm/io.h> 32 #include <asm/numa.h> 33 #include <asm/patch.h> 34 #include <asm/pgalloc.h> 35 #include <asm/sal.h> 36 #include <asm/sections.h> 37 #include <asm/tlb.h> 38 #include <linux/uaccess.h> 39 #include <asm/unistd.h> 40 #include <asm/mca.h> 41 42 extern void ia64_tlb_init (void); 43 44 unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL; 45 46 struct page *zero_page_memmap_ptr; /* map entry for zero page */ 47 EXPORT_SYMBOL(zero_page_memmap_ptr); 48 49 void 50 __ia64_sync_icache_dcache (pte_t pte) 51 { 52 unsigned long addr; 53 struct page *page; 54 55 page = pte_page(pte); 56 addr = (unsigned long) page_address(page); 57 58 if (test_bit(PG_arch_1, &page->flags)) 59 return; /* i-cache is already coherent with d-cache */ 60 61 flush_icache_range(addr, addr + page_size(page)); 62 set_bit(PG_arch_1, &page->flags); /* mark page as clean */ 63 } 64 65 /* 66 * Since DMA is i-cache coherent, any (complete) pages that were written via 67 * DMA can be marked as "clean" so that lazy_mmu_prot_update() doesn't have to 68 * flush them when they get mapped into an executable vm-area. 69 */ 70 void arch_dma_mark_clean(phys_addr_t paddr, size_t size) 71 { 72 unsigned long pfn = PHYS_PFN(paddr); 73 74 do { 75 set_bit(PG_arch_1, &pfn_to_page(pfn)->flags); 76 } while (++pfn <= PHYS_PFN(paddr + size - 1)); 77 } 78 79 inline void 80 ia64_set_rbs_bot (void) 81 { 82 unsigned long stack_size = rlimit_max(RLIMIT_STACK) & -16; 83 84 if (stack_size > MAX_USER_STACK_SIZE) 85 stack_size = MAX_USER_STACK_SIZE; 86 current->thread.rbs_bot = PAGE_ALIGN(current->mm->start_stack - stack_size); 87 } 88 89 /* 90 * This performs some platform-dependent address space initialization. 91 * On IA-64, we want to setup the VM area for the register backing 92 * store (which grows upwards) and install the gateway page which is 93 * used for signal trampolines, etc. 94 */ 95 void 96 ia64_init_addr_space (void) 97 { 98 struct vm_area_struct *vma; 99 100 ia64_set_rbs_bot(); 101 102 /* 103 * If we're out of memory and kmem_cache_alloc() returns NULL, we simply ignore 104 * the problem. When the process attempts to write to the register backing store 105 * for the first time, it will get a SEGFAULT in this case. 106 */ 107 vma = vm_area_alloc(current->mm); 108 if (vma) { 109 vma_set_anonymous(vma); 110 vma->vm_start = current->thread.rbs_bot & PAGE_MASK; 111 vma->vm_end = vma->vm_start + PAGE_SIZE; 112 vm_flags_init(vma, VM_DATA_DEFAULT_FLAGS|VM_GROWSUP|VM_ACCOUNT); 113 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 114 mmap_write_lock(current->mm); 115 if (insert_vm_struct(current->mm, vma)) { 116 mmap_write_unlock(current->mm); 117 vm_area_free(vma); 118 return; 119 } 120 mmap_write_unlock(current->mm); 121 } 122 123 /* map NaT-page at address zero to speed up speculative dereferencing of NULL: */ 124 if (!(current->personality & MMAP_PAGE_ZERO)) { 125 vma = vm_area_alloc(current->mm); 126 if (vma) { 127 vma_set_anonymous(vma); 128 vma->vm_end = PAGE_SIZE; 129 vma->vm_page_prot = __pgprot(pgprot_val(PAGE_READONLY) | _PAGE_MA_NAT); 130 vm_flags_init(vma, VM_READ | VM_MAYREAD | VM_IO | 131 VM_DONTEXPAND | VM_DONTDUMP); 132 mmap_write_lock(current->mm); 133 if (insert_vm_struct(current->mm, vma)) { 134 mmap_write_unlock(current->mm); 135 vm_area_free(vma); 136 return; 137 } 138 mmap_write_unlock(current->mm); 139 } 140 } 141 } 142 143 void 144 free_initmem (void) 145 { 146 free_reserved_area(ia64_imva(__init_begin), ia64_imva(__init_end), 147 -1, "unused kernel"); 148 } 149 150 void __init 151 free_initrd_mem (unsigned long start, unsigned long end) 152 { 153 /* 154 * EFI uses 4KB pages while the kernel can use 4KB or bigger. 155 * Thus EFI and the kernel may have different page sizes. It is 156 * therefore possible to have the initrd share the same page as 157 * the end of the kernel (given current setup). 158 * 159 * To avoid freeing/using the wrong page (kernel sized) we: 160 * - align up the beginning of initrd 161 * - align down the end of initrd 162 * 163 * | | 164 * |=============| a000 165 * | | 166 * | | 167 * | | 9000 168 * |/////////////| 169 * |/////////////| 170 * |=============| 8000 171 * |///INITRD////| 172 * |/////////////| 173 * |/////////////| 7000 174 * | | 175 * |KKKKKKKKKKKKK| 176 * |=============| 6000 177 * |KKKKKKKKKKKKK| 178 * |KKKKKKKKKKKKK| 179 * K=kernel using 8KB pages 180 * 181 * In this example, we must free page 8000 ONLY. So we must align up 182 * initrd_start and keep initrd_end as is. 183 */ 184 start = PAGE_ALIGN(start); 185 end = end & PAGE_MASK; 186 187 if (start < end) 188 printk(KERN_INFO "Freeing initrd memory: %ldkB freed\n", (end - start) >> 10); 189 190 for (; start < end; start += PAGE_SIZE) { 191 if (!virt_addr_valid(start)) 192 continue; 193 free_reserved_page(virt_to_page(start)); 194 } 195 } 196 197 /* 198 * This installs a clean page in the kernel's page table. 199 */ 200 static struct page * __init 201 put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot) 202 { 203 pgd_t *pgd; 204 p4d_t *p4d; 205 pud_t *pud; 206 pmd_t *pmd; 207 pte_t *pte; 208 209 pgd = pgd_offset_k(address); /* note: this is NOT pgd_offset()! */ 210 211 { 212 p4d = p4d_alloc(&init_mm, pgd, address); 213 if (!p4d) 214 goto out; 215 pud = pud_alloc(&init_mm, p4d, address); 216 if (!pud) 217 goto out; 218 pmd = pmd_alloc(&init_mm, pud, address); 219 if (!pmd) 220 goto out; 221 pte = pte_alloc_kernel(pmd, address); 222 if (!pte) 223 goto out; 224 if (!pte_none(*pte)) 225 goto out; 226 set_pte(pte, mk_pte(page, pgprot)); 227 } 228 out: 229 /* no need for flush_tlb */ 230 return page; 231 } 232 233 static void __init 234 setup_gate (void) 235 { 236 struct page *page; 237 238 /* 239 * Map the gate page twice: once read-only to export the ELF 240 * headers etc. and once execute-only page to enable 241 * privilege-promotion via "epc": 242 */ 243 page = virt_to_page(ia64_imva(__start_gate_section)); 244 put_kernel_page(page, GATE_ADDR, PAGE_READONLY); 245 #ifdef HAVE_BUGGY_SEGREL 246 page = virt_to_page(ia64_imva(__start_gate_section + PAGE_SIZE)); 247 put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE); 248 #else 249 put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE); 250 /* Fill in the holes (if any) with read-only zero pages: */ 251 { 252 unsigned long addr; 253 254 for (addr = GATE_ADDR + PAGE_SIZE; 255 addr < GATE_ADDR + PERCPU_PAGE_SIZE; 256 addr += PAGE_SIZE) 257 { 258 put_kernel_page(ZERO_PAGE(0), addr, 259 PAGE_READONLY); 260 put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE, 261 PAGE_READONLY); 262 } 263 } 264 #endif 265 ia64_patch_gate(); 266 } 267 268 static struct vm_area_struct gate_vma; 269 270 static int __init gate_vma_init(void) 271 { 272 vma_init(&gate_vma, NULL); 273 gate_vma.vm_start = FIXADDR_USER_START; 274 gate_vma.vm_end = FIXADDR_USER_END; 275 vm_flags_init(&gate_vma, VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC); 276 gate_vma.vm_page_prot = __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_RX); 277 278 return 0; 279 } 280 __initcall(gate_vma_init); 281 282 struct vm_area_struct *get_gate_vma(struct mm_struct *mm) 283 { 284 return &gate_vma; 285 } 286 287 int in_gate_area_no_mm(unsigned long addr) 288 { 289 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) 290 return 1; 291 return 0; 292 } 293 294 int in_gate_area(struct mm_struct *mm, unsigned long addr) 295 { 296 return in_gate_area_no_mm(addr); 297 } 298 299 void ia64_mmu_init(void *my_cpu_data) 300 { 301 unsigned long pta, impl_va_bits; 302 extern void tlb_init(void); 303 304 #ifdef CONFIG_DISABLE_VHPT 305 # define VHPT_ENABLE_BIT 0 306 #else 307 # define VHPT_ENABLE_BIT 1 308 #endif 309 310 /* 311 * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped 312 * address space. The IA-64 architecture guarantees that at least 50 bits of 313 * virtual address space are implemented but if we pick a large enough page size 314 * (e.g., 64KB), the mapped address space is big enough that it will overlap with 315 * VMLPT. I assume that once we run on machines big enough to warrant 64KB pages, 316 * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a 317 * problem in practice. Alternatively, we could truncate the top of the mapped 318 * address space to not permit mappings that would overlap with the VMLPT. 319 * --davidm 00/12/06 320 */ 321 # define pte_bits 3 322 # define mapped_space_bits (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT) 323 /* 324 * The virtual page table has to cover the entire implemented address space within 325 * a region even though not all of this space may be mappable. The reason for 326 * this is that the Access bit and Dirty bit fault handlers perform 327 * non-speculative accesses to the virtual page table, so the address range of the 328 * virtual page table itself needs to be covered by virtual page table. 329 */ 330 # define vmlpt_bits (impl_va_bits - PAGE_SHIFT + pte_bits) 331 # define POW2(n) (1ULL << (n)) 332 333 impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61))); 334 335 if (impl_va_bits < 51 || impl_va_bits > 61) 336 panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1); 337 /* 338 * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need, 339 * which must fit into "vmlpt_bits - pte_bits" slots. Second half of 340 * the test makes sure that our mapped space doesn't overlap the 341 * unimplemented hole in the middle of the region. 342 */ 343 if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) || 344 (mapped_space_bits > impl_va_bits - 1)) 345 panic("Cannot build a big enough virtual-linear page table" 346 " to cover mapped address space.\n" 347 " Try using a smaller page size.\n"); 348 349 350 /* place the VMLPT at the end of each page-table mapped region: */ 351 pta = POW2(61) - POW2(vmlpt_bits); 352 353 /* 354 * Set the (virtually mapped linear) page table address. Bit 355 * 8 selects between the short and long format, bits 2-7 the 356 * size of the table, and bit 0 whether the VHPT walker is 357 * enabled. 358 */ 359 ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT); 360 361 ia64_tlb_init(); 362 363 #ifdef CONFIG_HUGETLB_PAGE 364 ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2); 365 ia64_srlz_d(); 366 #endif 367 } 368 369 int __init register_active_ranges(u64 start, u64 len, int nid) 370 { 371 u64 end = start + len; 372 373 #ifdef CONFIG_KEXEC 374 if (start > crashk_res.start && start < crashk_res.end) 375 start = crashk_res.end; 376 if (end > crashk_res.start && end < crashk_res.end) 377 end = crashk_res.start; 378 #endif 379 380 if (start < end) 381 memblock_add_node(__pa(start), end - start, nid, MEMBLOCK_NONE); 382 return 0; 383 } 384 385 int 386 find_max_min_low_pfn (u64 start, u64 end, void *arg) 387 { 388 unsigned long pfn_start, pfn_end; 389 #ifdef CONFIG_FLATMEM 390 pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT; 391 pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT; 392 #else 393 pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT; 394 pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT; 395 #endif 396 min_low_pfn = min(min_low_pfn, pfn_start); 397 max_low_pfn = max(max_low_pfn, pfn_end); 398 return 0; 399 } 400 401 /* 402 * Boot command-line option "nolwsys" can be used to disable the use of any light-weight 403 * system call handler. When this option is in effect, all fsyscalls will end up bubbling 404 * down into the kernel and calling the normal (heavy-weight) syscall handler. This is 405 * useful for performance testing, but conceivably could also come in handy for debugging 406 * purposes. 407 */ 408 409 static int nolwsys __initdata; 410 411 static int __init 412 nolwsys_setup (char *s) 413 { 414 nolwsys = 1; 415 return 1; 416 } 417 418 __setup("nolwsys", nolwsys_setup); 419 420 void __init 421 mem_init (void) 422 { 423 int i; 424 425 BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE); 426 BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE); 427 BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE); 428 429 /* 430 * This needs to be called _after_ the command line has been parsed but 431 * _before_ any drivers that may need the PCI DMA interface are 432 * initialized or bootmem has been freed. 433 */ 434 do { 435 #ifdef CONFIG_INTEL_IOMMU 436 detect_intel_iommu(); 437 if (iommu_detected) 438 break; 439 #endif 440 swiotlb_init(true, SWIOTLB_VERBOSE); 441 } while (0); 442 443 #ifdef CONFIG_FLATMEM 444 BUG_ON(!mem_map); 445 #endif 446 447 set_max_mapnr(max_low_pfn); 448 high_memory = __va(max_low_pfn * PAGE_SIZE); 449 memblock_free_all(); 450 451 /* 452 * For fsyscall entrypoints with no light-weight handler, use the ordinary 453 * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry 454 * code can tell them apart. 455 */ 456 for (i = 0; i < NR_syscalls; ++i) { 457 extern unsigned long fsyscall_table[NR_syscalls]; 458 extern unsigned long sys_call_table[NR_syscalls]; 459 460 if (!fsyscall_table[i] || nolwsys) 461 fsyscall_table[i] = sys_call_table[i] | 1; 462 } 463 setup_gate(); 464 } 465 466 #ifdef CONFIG_MEMORY_HOTPLUG 467 int arch_add_memory(int nid, u64 start, u64 size, 468 struct mhp_params *params) 469 { 470 unsigned long start_pfn = start >> PAGE_SHIFT; 471 unsigned long nr_pages = size >> PAGE_SHIFT; 472 int ret; 473 474 if (WARN_ON_ONCE(params->pgprot.pgprot != PAGE_KERNEL.pgprot)) 475 return -EINVAL; 476 477 ret = __add_pages(nid, start_pfn, nr_pages, params); 478 if (ret) 479 printk("%s: Problem encountered in __add_pages() as ret=%d\n", 480 __func__, ret); 481 482 return ret; 483 } 484 485 void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap) 486 { 487 unsigned long start_pfn = start >> PAGE_SHIFT; 488 unsigned long nr_pages = size >> PAGE_SHIFT; 489 490 __remove_pages(start_pfn, nr_pages, altmap); 491 } 492 #endif 493 494 static const pgprot_t protection_map[16] = { 495 [VM_NONE] = PAGE_NONE, 496 [VM_READ] = PAGE_READONLY, 497 [VM_WRITE] = PAGE_READONLY, 498 [VM_WRITE | VM_READ] = PAGE_READONLY, 499 [VM_EXEC] = __pgprot(__ACCESS_BITS | _PAGE_PL_3 | 500 _PAGE_AR_X_RX), 501 [VM_EXEC | VM_READ] = __pgprot(__ACCESS_BITS | _PAGE_PL_3 | 502 _PAGE_AR_RX), 503 [VM_EXEC | VM_WRITE] = PAGE_COPY_EXEC, 504 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_COPY_EXEC, 505 [VM_SHARED] = PAGE_NONE, 506 [VM_SHARED | VM_READ] = PAGE_READONLY, 507 [VM_SHARED | VM_WRITE] = PAGE_SHARED, 508 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED, 509 [VM_SHARED | VM_EXEC] = __pgprot(__ACCESS_BITS | _PAGE_PL_3 | 510 _PAGE_AR_X_RX), 511 [VM_SHARED | VM_EXEC | VM_READ] = __pgprot(__ACCESS_BITS | _PAGE_PL_3 | 512 _PAGE_AR_RX), 513 [VM_SHARED | VM_EXEC | VM_WRITE] = __pgprot(__ACCESS_BITS | _PAGE_PL_3 | 514 _PAGE_AR_RWX), 515 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = __pgprot(__ACCESS_BITS | _PAGE_PL_3 | 516 _PAGE_AR_RWX) 517 }; 518 DECLARE_VM_GET_PAGE_PROT 519