1 /* 2 * PowerPC64 port by Mike Corrigan and Dave Engebretsen 3 * {mikejc|engebret}@us.ibm.com 4 * 5 * Copyright (c) 2000 Mike Corrigan <mikejc@us.ibm.com> 6 * 7 * SMP scalability work: 8 * Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM 9 * 10 * Module name: htab.c 11 * 12 * Description: 13 * PowerPC Hashed Page Table functions 14 * 15 * This program is free software; you can redistribute it and/or 16 * modify it under the terms of the GNU General Public License 17 * as published by the Free Software Foundation; either version 18 * 2 of the License, or (at your option) any later version. 19 */ 20 21 #undef DEBUG 22 #undef DEBUG_LOW 23 24 #define pr_fmt(fmt) "hash-mmu: " fmt 25 #include <linux/spinlock.h> 26 #include <linux/errno.h> 27 #include <linux/sched/mm.h> 28 #include <linux/proc_fs.h> 29 #include <linux/stat.h> 30 #include <linux/sysctl.h> 31 #include <linux/export.h> 32 #include <linux/ctype.h> 33 #include <linux/cache.h> 34 #include <linux/init.h> 35 #include <linux/signal.h> 36 #include <linux/memblock.h> 37 #include <linux/context_tracking.h> 38 #include <linux/libfdt.h> 39 #include <linux/pkeys.h> 40 #include <linux/hugetlb.h> 41 42 #include <asm/debugfs.h> 43 #include <asm/processor.h> 44 #include <asm/pgtable.h> 45 #include <asm/mmu.h> 46 #include <asm/mmu_context.h> 47 #include <asm/page.h> 48 #include <asm/types.h> 49 #include <linux/uaccess.h> 50 #include <asm/machdep.h> 51 #include <asm/prom.h> 52 #include <asm/io.h> 53 #include <asm/eeh.h> 54 #include <asm/tlb.h> 55 #include <asm/cacheflush.h> 56 #include <asm/cputable.h> 57 #include <asm/sections.h> 58 #include <asm/copro.h> 59 #include <asm/udbg.h> 60 #include <asm/code-patching.h> 61 #include <asm/fadump.h> 62 #include <asm/firmware.h> 63 #include <asm/tm.h> 64 #include <asm/trace.h> 65 #include <asm/ps3.h> 66 #include <asm/pte-walk.h> 67 #include <asm/asm-prototypes.h> 68 69 #include <mm/mmu_decl.h> 70 71 #ifdef DEBUG 72 #define DBG(fmt...) udbg_printf(fmt) 73 #else 74 #define DBG(fmt...) 75 #endif 76 77 #ifdef DEBUG_LOW 78 #define DBG_LOW(fmt...) udbg_printf(fmt) 79 #else 80 #define DBG_LOW(fmt...) 81 #endif 82 83 #define KB (1024) 84 #define MB (1024*KB) 85 #define GB (1024L*MB) 86 87 /* 88 * Note: pte --> Linux PTE 89 * HPTE --> PowerPC Hashed Page Table Entry 90 * 91 * Execution context: 92 * htab_initialize is called with the MMU off (of course), but 93 * the kernel has been copied down to zero so it can directly 94 * reference global data. At this point it is very difficult 95 * to print debug info. 96 * 97 */ 98 99 static unsigned long _SDR1; 100 struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT]; 101 EXPORT_SYMBOL_GPL(mmu_psize_defs); 102 103 u8 hpte_page_sizes[1 << LP_BITS]; 104 EXPORT_SYMBOL_GPL(hpte_page_sizes); 105 106 struct hash_pte *htab_address; 107 unsigned long htab_size_bytes; 108 unsigned long htab_hash_mask; 109 EXPORT_SYMBOL_GPL(htab_hash_mask); 110 int mmu_linear_psize = MMU_PAGE_4K; 111 EXPORT_SYMBOL_GPL(mmu_linear_psize); 112 int mmu_virtual_psize = MMU_PAGE_4K; 113 int mmu_vmalloc_psize = MMU_PAGE_4K; 114 #ifdef CONFIG_SPARSEMEM_VMEMMAP 115 int mmu_vmemmap_psize = MMU_PAGE_4K; 116 #endif 117 int mmu_io_psize = MMU_PAGE_4K; 118 int mmu_kernel_ssize = MMU_SEGSIZE_256M; 119 EXPORT_SYMBOL_GPL(mmu_kernel_ssize); 120 int mmu_highuser_ssize = MMU_SEGSIZE_256M; 121 u16 mmu_slb_size = 64; 122 EXPORT_SYMBOL_GPL(mmu_slb_size); 123 #ifdef CONFIG_PPC_64K_PAGES 124 int mmu_ci_restrictions; 125 #endif 126 #ifdef CONFIG_DEBUG_PAGEALLOC 127 static u8 *linear_map_hash_slots; 128 static unsigned long linear_map_hash_count; 129 static DEFINE_SPINLOCK(linear_map_hash_lock); 130 #endif /* CONFIG_DEBUG_PAGEALLOC */ 131 struct mmu_hash_ops mmu_hash_ops; 132 EXPORT_SYMBOL(mmu_hash_ops); 133 134 /* 135 * These are definitions of page sizes arrays to be used when none 136 * is provided by the firmware. 137 */ 138 139 /* 140 * Fallback (4k pages only) 141 */ 142 static struct mmu_psize_def mmu_psize_defaults[] = { 143 [MMU_PAGE_4K] = { 144 .shift = 12, 145 .sllp = 0, 146 .penc = {[MMU_PAGE_4K] = 0, [1 ... MMU_PAGE_COUNT - 1] = -1}, 147 .avpnm = 0, 148 .tlbiel = 0, 149 }, 150 }; 151 152 /* 153 * POWER4, GPUL, POWER5 154 * 155 * Support for 16Mb large pages 156 */ 157 static struct mmu_psize_def mmu_psize_defaults_gp[] = { 158 [MMU_PAGE_4K] = { 159 .shift = 12, 160 .sllp = 0, 161 .penc = {[MMU_PAGE_4K] = 0, [1 ... MMU_PAGE_COUNT - 1] = -1}, 162 .avpnm = 0, 163 .tlbiel = 1, 164 }, 165 [MMU_PAGE_16M] = { 166 .shift = 24, 167 .sllp = SLB_VSID_L, 168 .penc = {[0 ... MMU_PAGE_16M - 1] = -1, [MMU_PAGE_16M] = 0, 169 [MMU_PAGE_16M + 1 ... MMU_PAGE_COUNT - 1] = -1 }, 170 .avpnm = 0x1UL, 171 .tlbiel = 0, 172 }, 173 }; 174 175 /* 176 * 'R' and 'C' update notes: 177 * - Under pHyp or KVM, the updatepp path will not set C, thus it *will* 178 * create writeable HPTEs without C set, because the hcall H_PROTECT 179 * that we use in that case will not update C 180 * - The above is however not a problem, because we also don't do that 181 * fancy "no flush" variant of eviction and we use H_REMOVE which will 182 * do the right thing and thus we don't have the race I described earlier 183 * 184 * - Under bare metal, we do have the race, so we need R and C set 185 * - We make sure R is always set and never lost 186 * - C is _PAGE_DIRTY, and *should* always be set for a writeable mapping 187 */ 188 unsigned long htab_convert_pte_flags(unsigned long pteflags) 189 { 190 unsigned long rflags = 0; 191 192 /* _PAGE_EXEC -> NOEXEC */ 193 if ((pteflags & _PAGE_EXEC) == 0) 194 rflags |= HPTE_R_N; 195 /* 196 * PPP bits: 197 * Linux uses slb key 0 for kernel and 1 for user. 198 * kernel RW areas are mapped with PPP=0b000 199 * User area is mapped with PPP=0b010 for read/write 200 * or PPP=0b011 for read-only (including writeable but clean pages). 201 */ 202 if (pteflags & _PAGE_PRIVILEGED) { 203 /* 204 * Kernel read only mapped with ppp bits 0b110 205 */ 206 if (!(pteflags & _PAGE_WRITE)) { 207 if (mmu_has_feature(MMU_FTR_KERNEL_RO)) 208 rflags |= (HPTE_R_PP0 | 0x2); 209 else 210 rflags |= 0x3; 211 } 212 } else { 213 if (pteflags & _PAGE_RWX) 214 rflags |= 0x2; 215 if (!((pteflags & _PAGE_WRITE) && (pteflags & _PAGE_DIRTY))) 216 rflags |= 0x1; 217 } 218 /* 219 * We can't allow hardware to update hpte bits. Hence always 220 * set 'R' bit and set 'C' if it is a write fault 221 */ 222 rflags |= HPTE_R_R; 223 224 if (pteflags & _PAGE_DIRTY) 225 rflags |= HPTE_R_C; 226 /* 227 * Add in WIG bits 228 */ 229 230 if ((pteflags & _PAGE_CACHE_CTL) == _PAGE_TOLERANT) 231 rflags |= HPTE_R_I; 232 else if ((pteflags & _PAGE_CACHE_CTL) == _PAGE_NON_IDEMPOTENT) 233 rflags |= (HPTE_R_I | HPTE_R_G); 234 else if ((pteflags & _PAGE_CACHE_CTL) == _PAGE_SAO) 235 rflags |= (HPTE_R_W | HPTE_R_I | HPTE_R_M); 236 else 237 /* 238 * Add memory coherence if cache inhibited is not set 239 */ 240 rflags |= HPTE_R_M; 241 242 rflags |= pte_to_hpte_pkey_bits(pteflags); 243 return rflags; 244 } 245 246 int htab_bolt_mapping(unsigned long vstart, unsigned long vend, 247 unsigned long pstart, unsigned long prot, 248 int psize, int ssize) 249 { 250 unsigned long vaddr, paddr; 251 unsigned int step, shift; 252 int ret = 0; 253 254 shift = mmu_psize_defs[psize].shift; 255 step = 1 << shift; 256 257 prot = htab_convert_pte_flags(prot); 258 259 DBG("htab_bolt_mapping(%lx..%lx -> %lx (%lx,%d,%d)\n", 260 vstart, vend, pstart, prot, psize, ssize); 261 262 for (vaddr = vstart, paddr = pstart; vaddr < vend; 263 vaddr += step, paddr += step) { 264 unsigned long hash, hpteg; 265 unsigned long vsid = get_kernel_vsid(vaddr, ssize); 266 unsigned long vpn = hpt_vpn(vaddr, vsid, ssize); 267 unsigned long tprot = prot; 268 269 /* 270 * If we hit a bad address return error. 271 */ 272 if (!vsid) 273 return -1; 274 /* Make kernel text executable */ 275 if (overlaps_kernel_text(vaddr, vaddr + step)) 276 tprot &= ~HPTE_R_N; 277 278 /* Make kvm guest trampolines executable */ 279 if (overlaps_kvm_tmp(vaddr, vaddr + step)) 280 tprot &= ~HPTE_R_N; 281 282 /* 283 * If relocatable, check if it overlaps interrupt vectors that 284 * are copied down to real 0. For relocatable kernel 285 * (e.g. kdump case) we copy interrupt vectors down to real 286 * address 0. Mark that region as executable. This is 287 * because on p8 system with relocation on exception feature 288 * enabled, exceptions are raised with MMU (IR=DR=1) ON. Hence 289 * in order to execute the interrupt handlers in virtual 290 * mode the vector region need to be marked as executable. 291 */ 292 if ((PHYSICAL_START > MEMORY_START) && 293 overlaps_interrupt_vector_text(vaddr, vaddr + step)) 294 tprot &= ~HPTE_R_N; 295 296 hash = hpt_hash(vpn, shift, ssize); 297 hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP); 298 299 BUG_ON(!mmu_hash_ops.hpte_insert); 300 ret = mmu_hash_ops.hpte_insert(hpteg, vpn, paddr, tprot, 301 HPTE_V_BOLTED, psize, psize, 302 ssize); 303 304 if (ret < 0) 305 break; 306 307 #ifdef CONFIG_DEBUG_PAGEALLOC 308 if (debug_pagealloc_enabled() && 309 (paddr >> PAGE_SHIFT) < linear_map_hash_count) 310 linear_map_hash_slots[paddr >> PAGE_SHIFT] = ret | 0x80; 311 #endif /* CONFIG_DEBUG_PAGEALLOC */ 312 } 313 return ret < 0 ? ret : 0; 314 } 315 316 int htab_remove_mapping(unsigned long vstart, unsigned long vend, 317 int psize, int ssize) 318 { 319 unsigned long vaddr; 320 unsigned int step, shift; 321 int rc; 322 int ret = 0; 323 324 shift = mmu_psize_defs[psize].shift; 325 step = 1 << shift; 326 327 if (!mmu_hash_ops.hpte_removebolted) 328 return -ENODEV; 329 330 for (vaddr = vstart; vaddr < vend; vaddr += step) { 331 rc = mmu_hash_ops.hpte_removebolted(vaddr, psize, ssize); 332 if (rc == -ENOENT) { 333 ret = -ENOENT; 334 continue; 335 } 336 if (rc < 0) 337 return rc; 338 } 339 340 return ret; 341 } 342 343 static bool disable_1tb_segments = false; 344 345 static int __init parse_disable_1tb_segments(char *p) 346 { 347 disable_1tb_segments = true; 348 return 0; 349 } 350 early_param("disable_1tb_segments", parse_disable_1tb_segments); 351 352 static int __init htab_dt_scan_seg_sizes(unsigned long node, 353 const char *uname, int depth, 354 void *data) 355 { 356 const char *type = of_get_flat_dt_prop(node, "device_type", NULL); 357 const __be32 *prop; 358 int size = 0; 359 360 /* We are scanning "cpu" nodes only */ 361 if (type == NULL || strcmp(type, "cpu") != 0) 362 return 0; 363 364 prop = of_get_flat_dt_prop(node, "ibm,processor-segment-sizes", &size); 365 if (prop == NULL) 366 return 0; 367 for (; size >= 4; size -= 4, ++prop) { 368 if (be32_to_cpu(prop[0]) == 40) { 369 DBG("1T segment support detected\n"); 370 371 if (disable_1tb_segments) { 372 DBG("1T segments disabled by command line\n"); 373 break; 374 } 375 376 cur_cpu_spec->mmu_features |= MMU_FTR_1T_SEGMENT; 377 return 1; 378 } 379 } 380 cur_cpu_spec->mmu_features &= ~MMU_FTR_NO_SLBIE_B; 381 return 0; 382 } 383 384 static int __init get_idx_from_shift(unsigned int shift) 385 { 386 int idx = -1; 387 388 switch (shift) { 389 case 0xc: 390 idx = MMU_PAGE_4K; 391 break; 392 case 0x10: 393 idx = MMU_PAGE_64K; 394 break; 395 case 0x14: 396 idx = MMU_PAGE_1M; 397 break; 398 case 0x18: 399 idx = MMU_PAGE_16M; 400 break; 401 case 0x22: 402 idx = MMU_PAGE_16G; 403 break; 404 } 405 return idx; 406 } 407 408 static int __init htab_dt_scan_page_sizes(unsigned long node, 409 const char *uname, int depth, 410 void *data) 411 { 412 const char *type = of_get_flat_dt_prop(node, "device_type", NULL); 413 const __be32 *prop; 414 int size = 0; 415 416 /* We are scanning "cpu" nodes only */ 417 if (type == NULL || strcmp(type, "cpu") != 0) 418 return 0; 419 420 prop = of_get_flat_dt_prop(node, "ibm,segment-page-sizes", &size); 421 if (!prop) 422 return 0; 423 424 pr_info("Page sizes from device-tree:\n"); 425 size /= 4; 426 cur_cpu_spec->mmu_features &= ~(MMU_FTR_16M_PAGE); 427 while(size > 0) { 428 unsigned int base_shift = be32_to_cpu(prop[0]); 429 unsigned int slbenc = be32_to_cpu(prop[1]); 430 unsigned int lpnum = be32_to_cpu(prop[2]); 431 struct mmu_psize_def *def; 432 int idx, base_idx; 433 434 size -= 3; prop += 3; 435 base_idx = get_idx_from_shift(base_shift); 436 if (base_idx < 0) { 437 /* skip the pte encoding also */ 438 prop += lpnum * 2; size -= lpnum * 2; 439 continue; 440 } 441 def = &mmu_psize_defs[base_idx]; 442 if (base_idx == MMU_PAGE_16M) 443 cur_cpu_spec->mmu_features |= MMU_FTR_16M_PAGE; 444 445 def->shift = base_shift; 446 if (base_shift <= 23) 447 def->avpnm = 0; 448 else 449 def->avpnm = (1 << (base_shift - 23)) - 1; 450 def->sllp = slbenc; 451 /* 452 * We don't know for sure what's up with tlbiel, so 453 * for now we only set it for 4K and 64K pages 454 */ 455 if (base_idx == MMU_PAGE_4K || base_idx == MMU_PAGE_64K) 456 def->tlbiel = 1; 457 else 458 def->tlbiel = 0; 459 460 while (size > 0 && lpnum) { 461 unsigned int shift = be32_to_cpu(prop[0]); 462 int penc = be32_to_cpu(prop[1]); 463 464 prop += 2; size -= 2; 465 lpnum--; 466 467 idx = get_idx_from_shift(shift); 468 if (idx < 0) 469 continue; 470 471 if (penc == -1) 472 pr_err("Invalid penc for base_shift=%d " 473 "shift=%d\n", base_shift, shift); 474 475 def->penc[idx] = penc; 476 pr_info("base_shift=%d: shift=%d, sllp=0x%04lx," 477 " avpnm=0x%08lx, tlbiel=%d, penc=%d\n", 478 base_shift, shift, def->sllp, 479 def->avpnm, def->tlbiel, def->penc[idx]); 480 } 481 } 482 483 return 1; 484 } 485 486 #ifdef CONFIG_HUGETLB_PAGE 487 /* 488 * Scan for 16G memory blocks that have been set aside for huge pages 489 * and reserve those blocks for 16G huge pages. 490 */ 491 static int __init htab_dt_scan_hugepage_blocks(unsigned long node, 492 const char *uname, int depth, 493 void *data) { 494 const char *type = of_get_flat_dt_prop(node, "device_type", NULL); 495 const __be64 *addr_prop; 496 const __be32 *page_count_prop; 497 unsigned int expected_pages; 498 long unsigned int phys_addr; 499 long unsigned int block_size; 500 501 /* We are scanning "memory" nodes only */ 502 if (type == NULL || strcmp(type, "memory") != 0) 503 return 0; 504 505 /* 506 * This property is the log base 2 of the number of virtual pages that 507 * will represent this memory block. 508 */ 509 page_count_prop = of_get_flat_dt_prop(node, "ibm,expected#pages", NULL); 510 if (page_count_prop == NULL) 511 return 0; 512 expected_pages = (1 << be32_to_cpu(page_count_prop[0])); 513 addr_prop = of_get_flat_dt_prop(node, "reg", NULL); 514 if (addr_prop == NULL) 515 return 0; 516 phys_addr = be64_to_cpu(addr_prop[0]); 517 block_size = be64_to_cpu(addr_prop[1]); 518 if (block_size != (16 * GB)) 519 return 0; 520 printk(KERN_INFO "Huge page(16GB) memory: " 521 "addr = 0x%lX size = 0x%lX pages = %d\n", 522 phys_addr, block_size, expected_pages); 523 if (phys_addr + block_size * expected_pages <= memblock_end_of_DRAM()) { 524 memblock_reserve(phys_addr, block_size * expected_pages); 525 pseries_add_gpage(phys_addr, block_size, expected_pages); 526 } 527 return 0; 528 } 529 #endif /* CONFIG_HUGETLB_PAGE */ 530 531 static void mmu_psize_set_default_penc(void) 532 { 533 int bpsize, apsize; 534 for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++) 535 for (apsize = 0; apsize < MMU_PAGE_COUNT; apsize++) 536 mmu_psize_defs[bpsize].penc[apsize] = -1; 537 } 538 539 #ifdef CONFIG_PPC_64K_PAGES 540 541 static bool might_have_hea(void) 542 { 543 /* 544 * The HEA ethernet adapter requires awareness of the 545 * GX bus. Without that awareness we can easily assume 546 * we will never see an HEA ethernet device. 547 */ 548 #ifdef CONFIG_IBMEBUS 549 return !cpu_has_feature(CPU_FTR_ARCH_207S) && 550 firmware_has_feature(FW_FEATURE_SPLPAR); 551 #else 552 return false; 553 #endif 554 } 555 556 #endif /* #ifdef CONFIG_PPC_64K_PAGES */ 557 558 static void __init htab_scan_page_sizes(void) 559 { 560 int rc; 561 562 /* se the invalid penc to -1 */ 563 mmu_psize_set_default_penc(); 564 565 /* Default to 4K pages only */ 566 memcpy(mmu_psize_defs, mmu_psize_defaults, 567 sizeof(mmu_psize_defaults)); 568 569 /* 570 * Try to find the available page sizes in the device-tree 571 */ 572 rc = of_scan_flat_dt(htab_dt_scan_page_sizes, NULL); 573 if (rc == 0 && early_mmu_has_feature(MMU_FTR_16M_PAGE)) { 574 /* 575 * Nothing in the device-tree, but the CPU supports 16M pages, 576 * so let's fallback on a known size list for 16M capable CPUs. 577 */ 578 memcpy(mmu_psize_defs, mmu_psize_defaults_gp, 579 sizeof(mmu_psize_defaults_gp)); 580 } 581 582 #ifdef CONFIG_HUGETLB_PAGE 583 if (!hugetlb_disabled) { 584 /* Reserve 16G huge page memory sections for huge pages */ 585 of_scan_flat_dt(htab_dt_scan_hugepage_blocks, NULL); 586 } 587 #endif /* CONFIG_HUGETLB_PAGE */ 588 } 589 590 /* 591 * Fill in the hpte_page_sizes[] array. 592 * We go through the mmu_psize_defs[] array looking for all the 593 * supported base/actual page size combinations. Each combination 594 * has a unique pagesize encoding (penc) value in the low bits of 595 * the LP field of the HPTE. For actual page sizes less than 1MB, 596 * some of the upper LP bits are used for RPN bits, meaning that 597 * we need to fill in several entries in hpte_page_sizes[]. 598 * 599 * In diagrammatic form, with r = RPN bits and z = page size bits: 600 * PTE LP actual page size 601 * rrrr rrrz >=8KB 602 * rrrr rrzz >=16KB 603 * rrrr rzzz >=32KB 604 * rrrr zzzz >=64KB 605 * ... 606 * 607 * The zzzz bits are implementation-specific but are chosen so that 608 * no encoding for a larger page size uses the same value in its 609 * low-order N bits as the encoding for the 2^(12+N) byte page size 610 * (if it exists). 611 */ 612 static void init_hpte_page_sizes(void) 613 { 614 long int ap, bp; 615 long int shift, penc; 616 617 for (bp = 0; bp < MMU_PAGE_COUNT; ++bp) { 618 if (!mmu_psize_defs[bp].shift) 619 continue; /* not a supported page size */ 620 for (ap = bp; ap < MMU_PAGE_COUNT; ++ap) { 621 penc = mmu_psize_defs[bp].penc[ap]; 622 if (penc == -1 || !mmu_psize_defs[ap].shift) 623 continue; 624 shift = mmu_psize_defs[ap].shift - LP_SHIFT; 625 if (shift <= 0) 626 continue; /* should never happen */ 627 /* 628 * For page sizes less than 1MB, this loop 629 * replicates the entry for all possible values 630 * of the rrrr bits. 631 */ 632 while (penc < (1 << LP_BITS)) { 633 hpte_page_sizes[penc] = (ap << 4) | bp; 634 penc += 1 << shift; 635 } 636 } 637 } 638 } 639 640 static void __init htab_init_page_sizes(void) 641 { 642 init_hpte_page_sizes(); 643 644 if (!debug_pagealloc_enabled()) { 645 /* 646 * Pick a size for the linear mapping. Currently, we only 647 * support 16M, 1M and 4K which is the default 648 */ 649 if (mmu_psize_defs[MMU_PAGE_16M].shift) 650 mmu_linear_psize = MMU_PAGE_16M; 651 else if (mmu_psize_defs[MMU_PAGE_1M].shift) 652 mmu_linear_psize = MMU_PAGE_1M; 653 } 654 655 #ifdef CONFIG_PPC_64K_PAGES 656 /* 657 * Pick a size for the ordinary pages. Default is 4K, we support 658 * 64K for user mappings and vmalloc if supported by the processor. 659 * We only use 64k for ioremap if the processor 660 * (and firmware) support cache-inhibited large pages. 661 * If not, we use 4k and set mmu_ci_restrictions so that 662 * hash_page knows to switch processes that use cache-inhibited 663 * mappings to 4k pages. 664 */ 665 if (mmu_psize_defs[MMU_PAGE_64K].shift) { 666 mmu_virtual_psize = MMU_PAGE_64K; 667 mmu_vmalloc_psize = MMU_PAGE_64K; 668 if (mmu_linear_psize == MMU_PAGE_4K) 669 mmu_linear_psize = MMU_PAGE_64K; 670 if (mmu_has_feature(MMU_FTR_CI_LARGE_PAGE)) { 671 /* 672 * When running on pSeries using 64k pages for ioremap 673 * would stop us accessing the HEA ethernet. So if we 674 * have the chance of ever seeing one, stay at 4k. 675 */ 676 if (!might_have_hea()) 677 mmu_io_psize = MMU_PAGE_64K; 678 } else 679 mmu_ci_restrictions = 1; 680 } 681 #endif /* CONFIG_PPC_64K_PAGES */ 682 683 #ifdef CONFIG_SPARSEMEM_VMEMMAP 684 /* 685 * We try to use 16M pages for vmemmap if that is supported 686 * and we have at least 1G of RAM at boot 687 */ 688 if (mmu_psize_defs[MMU_PAGE_16M].shift && 689 memblock_phys_mem_size() >= 0x40000000) 690 mmu_vmemmap_psize = MMU_PAGE_16M; 691 else if (mmu_psize_defs[MMU_PAGE_64K].shift) 692 mmu_vmemmap_psize = MMU_PAGE_64K; 693 else 694 mmu_vmemmap_psize = MMU_PAGE_4K; 695 #endif /* CONFIG_SPARSEMEM_VMEMMAP */ 696 697 printk(KERN_DEBUG "Page orders: linear mapping = %d, " 698 "virtual = %d, io = %d" 699 #ifdef CONFIG_SPARSEMEM_VMEMMAP 700 ", vmemmap = %d" 701 #endif 702 "\n", 703 mmu_psize_defs[mmu_linear_psize].shift, 704 mmu_psize_defs[mmu_virtual_psize].shift, 705 mmu_psize_defs[mmu_io_psize].shift 706 #ifdef CONFIG_SPARSEMEM_VMEMMAP 707 ,mmu_psize_defs[mmu_vmemmap_psize].shift 708 #endif 709 ); 710 } 711 712 static int __init htab_dt_scan_pftsize(unsigned long node, 713 const char *uname, int depth, 714 void *data) 715 { 716 const char *type = of_get_flat_dt_prop(node, "device_type", NULL); 717 const __be32 *prop; 718 719 /* We are scanning "cpu" nodes only */ 720 if (type == NULL || strcmp(type, "cpu") != 0) 721 return 0; 722 723 prop = of_get_flat_dt_prop(node, "ibm,pft-size", NULL); 724 if (prop != NULL) { 725 /* pft_size[0] is the NUMA CEC cookie */ 726 ppc64_pft_size = be32_to_cpu(prop[1]); 727 return 1; 728 } 729 return 0; 730 } 731 732 unsigned htab_shift_for_mem_size(unsigned long mem_size) 733 { 734 unsigned memshift = __ilog2(mem_size); 735 unsigned pshift = mmu_psize_defs[mmu_virtual_psize].shift; 736 unsigned pteg_shift; 737 738 /* round mem_size up to next power of 2 */ 739 if ((1UL << memshift) < mem_size) 740 memshift += 1; 741 742 /* aim for 2 pages / pteg */ 743 pteg_shift = memshift - (pshift + 1); 744 745 /* 746 * 2^11 PTEGS of 128 bytes each, ie. 2^18 bytes is the minimum htab 747 * size permitted by the architecture. 748 */ 749 return max(pteg_shift + 7, 18U); 750 } 751 752 static unsigned long __init htab_get_table_size(void) 753 { 754 /* 755 * If hash size isn't already provided by the platform, we try to 756 * retrieve it from the device-tree. If it's not there neither, we 757 * calculate it now based on the total RAM size 758 */ 759 if (ppc64_pft_size == 0) 760 of_scan_flat_dt(htab_dt_scan_pftsize, NULL); 761 if (ppc64_pft_size) 762 return 1UL << ppc64_pft_size; 763 764 return 1UL << htab_shift_for_mem_size(memblock_phys_mem_size()); 765 } 766 767 #ifdef CONFIG_MEMORY_HOTPLUG 768 int resize_hpt_for_hotplug(unsigned long new_mem_size) 769 { 770 unsigned target_hpt_shift; 771 772 if (!mmu_hash_ops.resize_hpt) 773 return 0; 774 775 target_hpt_shift = htab_shift_for_mem_size(new_mem_size); 776 777 /* 778 * To avoid lots of HPT resizes if memory size is fluctuating 779 * across a boundary, we deliberately have some hysterisis 780 * here: we immediately increase the HPT size if the target 781 * shift exceeds the current shift, but we won't attempt to 782 * reduce unless the target shift is at least 2 below the 783 * current shift 784 */ 785 if (target_hpt_shift > ppc64_pft_size || 786 target_hpt_shift < ppc64_pft_size - 1) 787 return mmu_hash_ops.resize_hpt(target_hpt_shift); 788 789 return 0; 790 } 791 792 int hash__create_section_mapping(unsigned long start, unsigned long end, int nid) 793 { 794 int rc; 795 796 if (end >= H_VMALLOC_START) { 797 pr_warn("Outside the supported range\n"); 798 return -1; 799 } 800 801 rc = htab_bolt_mapping(start, end, __pa(start), 802 pgprot_val(PAGE_KERNEL), mmu_linear_psize, 803 mmu_kernel_ssize); 804 805 if (rc < 0) { 806 int rc2 = htab_remove_mapping(start, end, mmu_linear_psize, 807 mmu_kernel_ssize); 808 BUG_ON(rc2 && (rc2 != -ENOENT)); 809 } 810 return rc; 811 } 812 813 int hash__remove_section_mapping(unsigned long start, unsigned long end) 814 { 815 int rc = htab_remove_mapping(start, end, mmu_linear_psize, 816 mmu_kernel_ssize); 817 WARN_ON(rc < 0); 818 return rc; 819 } 820 #endif /* CONFIG_MEMORY_HOTPLUG */ 821 822 static void __init hash_init_partition_table(phys_addr_t hash_table, 823 unsigned long htab_size) 824 { 825 mmu_partition_table_init(); 826 827 /* 828 * PS field (VRMA page size) is not used for LPID 0, hence set to 0. 829 * For now, UPRT is 0 and we have no segment table. 830 */ 831 htab_size = __ilog2(htab_size) - 18; 832 mmu_partition_table_set_entry(0, hash_table | htab_size, 0); 833 pr_info("Partition table %p\n", partition_tb); 834 } 835 836 static void __init htab_initialize(void) 837 { 838 unsigned long table; 839 unsigned long pteg_count; 840 unsigned long prot; 841 unsigned long base = 0, size = 0; 842 struct memblock_region *reg; 843 844 DBG(" -> htab_initialize()\n"); 845 846 if (mmu_has_feature(MMU_FTR_1T_SEGMENT)) { 847 mmu_kernel_ssize = MMU_SEGSIZE_1T; 848 mmu_highuser_ssize = MMU_SEGSIZE_1T; 849 printk(KERN_INFO "Using 1TB segments\n"); 850 } 851 852 /* 853 * Calculate the required size of the htab. We want the number of 854 * PTEGs to equal one half the number of real pages. 855 */ 856 htab_size_bytes = htab_get_table_size(); 857 pteg_count = htab_size_bytes >> 7; 858 859 htab_hash_mask = pteg_count - 1; 860 861 if (firmware_has_feature(FW_FEATURE_LPAR) || 862 firmware_has_feature(FW_FEATURE_PS3_LV1)) { 863 /* Using a hypervisor which owns the htab */ 864 htab_address = NULL; 865 _SDR1 = 0; 866 /* 867 * On POWER9, we need to do a H_REGISTER_PROC_TBL hcall 868 * to inform the hypervisor that we wish to use the HPT. 869 */ 870 if (cpu_has_feature(CPU_FTR_ARCH_300)) 871 register_process_table(0, 0, 0); 872 #ifdef CONFIG_FA_DUMP 873 /* 874 * If firmware assisted dump is active firmware preserves 875 * the contents of htab along with entire partition memory. 876 * Clear the htab if firmware assisted dump is active so 877 * that we dont end up using old mappings. 878 */ 879 if (is_fadump_active() && mmu_hash_ops.hpte_clear_all) 880 mmu_hash_ops.hpte_clear_all(); 881 #endif 882 } else { 883 unsigned long limit = MEMBLOCK_ALLOC_ANYWHERE; 884 885 #ifdef CONFIG_PPC_CELL 886 /* 887 * Cell may require the hash table down low when using the 888 * Axon IOMMU in order to fit the dynamic region over it, see 889 * comments in cell/iommu.c 890 */ 891 if (fdt_subnode_offset(initial_boot_params, 0, "axon") > 0) { 892 limit = 0x80000000; 893 pr_info("Hash table forced below 2G for Axon IOMMU\n"); 894 } 895 #endif /* CONFIG_PPC_CELL */ 896 897 table = memblock_phys_alloc_range(htab_size_bytes, 898 htab_size_bytes, 899 0, limit); 900 if (!table) 901 panic("ERROR: Failed to allocate %pa bytes below %pa\n", 902 &htab_size_bytes, &limit); 903 904 DBG("Hash table allocated at %lx, size: %lx\n", table, 905 htab_size_bytes); 906 907 htab_address = __va(table); 908 909 /* htab absolute addr + encoded htabsize */ 910 _SDR1 = table + __ilog2(htab_size_bytes) - 18; 911 912 /* Initialize the HPT with no entries */ 913 memset((void *)table, 0, htab_size_bytes); 914 915 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 916 /* Set SDR1 */ 917 mtspr(SPRN_SDR1, _SDR1); 918 else 919 hash_init_partition_table(table, htab_size_bytes); 920 } 921 922 prot = pgprot_val(PAGE_KERNEL); 923 924 #ifdef CONFIG_DEBUG_PAGEALLOC 925 if (debug_pagealloc_enabled()) { 926 linear_map_hash_count = memblock_end_of_DRAM() >> PAGE_SHIFT; 927 linear_map_hash_slots = memblock_alloc_try_nid( 928 linear_map_hash_count, 1, MEMBLOCK_LOW_LIMIT, 929 ppc64_rma_size, NUMA_NO_NODE); 930 if (!linear_map_hash_slots) 931 panic("%s: Failed to allocate %lu bytes max_addr=%pa\n", 932 __func__, linear_map_hash_count, &ppc64_rma_size); 933 } 934 #endif /* CONFIG_DEBUG_PAGEALLOC */ 935 936 /* create bolted the linear mapping in the hash table */ 937 for_each_memblock(memory, reg) { 938 base = (unsigned long)__va(reg->base); 939 size = reg->size; 940 941 DBG("creating mapping for region: %lx..%lx (prot: %lx)\n", 942 base, size, prot); 943 944 if ((base + size) >= H_VMALLOC_START) { 945 pr_warn("Outside the supported range\n"); 946 continue; 947 } 948 949 BUG_ON(htab_bolt_mapping(base, base + size, __pa(base), 950 prot, mmu_linear_psize, mmu_kernel_ssize)); 951 } 952 memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE); 953 954 /* 955 * If we have a memory_limit and we've allocated TCEs then we need to 956 * explicitly map the TCE area at the top of RAM. We also cope with the 957 * case that the TCEs start below memory_limit. 958 * tce_alloc_start/end are 16MB aligned so the mapping should work 959 * for either 4K or 16MB pages. 960 */ 961 if (tce_alloc_start) { 962 tce_alloc_start = (unsigned long)__va(tce_alloc_start); 963 tce_alloc_end = (unsigned long)__va(tce_alloc_end); 964 965 if (base + size >= tce_alloc_start) 966 tce_alloc_start = base + size + 1; 967 968 BUG_ON(htab_bolt_mapping(tce_alloc_start, tce_alloc_end, 969 __pa(tce_alloc_start), prot, 970 mmu_linear_psize, mmu_kernel_ssize)); 971 } 972 973 974 DBG(" <- htab_initialize()\n"); 975 } 976 #undef KB 977 #undef MB 978 979 void __init hash__early_init_devtree(void) 980 { 981 /* Initialize segment sizes */ 982 of_scan_flat_dt(htab_dt_scan_seg_sizes, NULL); 983 984 /* Initialize page sizes */ 985 htab_scan_page_sizes(); 986 } 987 988 struct hash_mm_context init_hash_mm_context; 989 void __init hash__early_init_mmu(void) 990 { 991 #ifndef CONFIG_PPC_64K_PAGES 992 /* 993 * We have code in __hash_page_4K() and elsewhere, which assumes it can 994 * do the following: 995 * new_pte |= (slot << H_PAGE_F_GIX_SHIFT) & (H_PAGE_F_SECOND | H_PAGE_F_GIX); 996 * 997 * Where the slot number is between 0-15, and values of 8-15 indicate 998 * the secondary bucket. For that code to work H_PAGE_F_SECOND and 999 * H_PAGE_F_GIX must occupy four contiguous bits in the PTE, and 1000 * H_PAGE_F_SECOND must be placed above H_PAGE_F_GIX. Assert that here 1001 * with a BUILD_BUG_ON(). 1002 */ 1003 BUILD_BUG_ON(H_PAGE_F_SECOND != (1ul << (H_PAGE_F_GIX_SHIFT + 3))); 1004 #endif /* CONFIG_PPC_64K_PAGES */ 1005 1006 htab_init_page_sizes(); 1007 1008 /* 1009 * initialize page table size 1010 */ 1011 __pte_frag_nr = H_PTE_FRAG_NR; 1012 __pte_frag_size_shift = H_PTE_FRAG_SIZE_SHIFT; 1013 __pmd_frag_nr = H_PMD_FRAG_NR; 1014 __pmd_frag_size_shift = H_PMD_FRAG_SIZE_SHIFT; 1015 1016 __pte_index_size = H_PTE_INDEX_SIZE; 1017 __pmd_index_size = H_PMD_INDEX_SIZE; 1018 __pud_index_size = H_PUD_INDEX_SIZE; 1019 __pgd_index_size = H_PGD_INDEX_SIZE; 1020 __pud_cache_index = H_PUD_CACHE_INDEX; 1021 __pte_table_size = H_PTE_TABLE_SIZE; 1022 __pmd_table_size = H_PMD_TABLE_SIZE; 1023 __pud_table_size = H_PUD_TABLE_SIZE; 1024 __pgd_table_size = H_PGD_TABLE_SIZE; 1025 /* 1026 * 4k use hugepd format, so for hash set then to 1027 * zero 1028 */ 1029 __pmd_val_bits = HASH_PMD_VAL_BITS; 1030 __pud_val_bits = HASH_PUD_VAL_BITS; 1031 __pgd_val_bits = HASH_PGD_VAL_BITS; 1032 1033 __kernel_virt_start = H_KERN_VIRT_START; 1034 __vmalloc_start = H_VMALLOC_START; 1035 __vmalloc_end = H_VMALLOC_END; 1036 __kernel_io_start = H_KERN_IO_START; 1037 __kernel_io_end = H_KERN_IO_END; 1038 vmemmap = (struct page *)H_VMEMMAP_START; 1039 ioremap_bot = IOREMAP_BASE; 1040 1041 #ifdef CONFIG_PCI 1042 pci_io_base = ISA_IO_BASE; 1043 #endif 1044 1045 /* Select appropriate backend */ 1046 if (firmware_has_feature(FW_FEATURE_PS3_LV1)) 1047 ps3_early_mm_init(); 1048 else if (firmware_has_feature(FW_FEATURE_LPAR)) 1049 hpte_init_pseries(); 1050 else if (IS_ENABLED(CONFIG_PPC_NATIVE)) 1051 hpte_init_native(); 1052 1053 if (!mmu_hash_ops.hpte_insert) 1054 panic("hash__early_init_mmu: No MMU hash ops defined!\n"); 1055 1056 /* 1057 * Initialize the MMU Hash table and create the linear mapping 1058 * of memory. Has to be done before SLB initialization as this is 1059 * currently where the page size encoding is obtained. 1060 */ 1061 htab_initialize(); 1062 1063 init_mm.context.hash_context = &init_hash_mm_context; 1064 mm_ctx_set_slb_addr_limit(&init_mm.context, SLB_ADDR_LIMIT_DEFAULT); 1065 1066 pr_info("Initializing hash mmu with SLB\n"); 1067 /* Initialize SLB management */ 1068 slb_initialize(); 1069 1070 if (cpu_has_feature(CPU_FTR_ARCH_206) 1071 && cpu_has_feature(CPU_FTR_HVMODE)) 1072 tlbiel_all(); 1073 } 1074 1075 #ifdef CONFIG_SMP 1076 void hash__early_init_mmu_secondary(void) 1077 { 1078 /* Initialize hash table for that CPU */ 1079 if (!firmware_has_feature(FW_FEATURE_LPAR)) { 1080 1081 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 1082 mtspr(SPRN_SDR1, _SDR1); 1083 else 1084 mtspr(SPRN_PTCR, 1085 __pa(partition_tb) | (PATB_SIZE_SHIFT - 12)); 1086 } 1087 /* Initialize SLB */ 1088 slb_initialize(); 1089 1090 if (cpu_has_feature(CPU_FTR_ARCH_206) 1091 && cpu_has_feature(CPU_FTR_HVMODE)) 1092 tlbiel_all(); 1093 } 1094 #endif /* CONFIG_SMP */ 1095 1096 /* 1097 * Called by asm hashtable.S for doing lazy icache flush 1098 */ 1099 unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap) 1100 { 1101 struct page *page; 1102 1103 if (!pfn_valid(pte_pfn(pte))) 1104 return pp; 1105 1106 page = pte_page(pte); 1107 1108 /* page is dirty */ 1109 if (!test_bit(PG_arch_1, &page->flags) && !PageReserved(page)) { 1110 if (trap == 0x400) { 1111 flush_dcache_icache_page(page); 1112 set_bit(PG_arch_1, &page->flags); 1113 } else 1114 pp |= HPTE_R_N; 1115 } 1116 return pp; 1117 } 1118 1119 #ifdef CONFIG_PPC_MM_SLICES 1120 static unsigned int get_paca_psize(unsigned long addr) 1121 { 1122 unsigned char *psizes; 1123 unsigned long index, mask_index; 1124 1125 if (addr < SLICE_LOW_TOP) { 1126 psizes = get_paca()->mm_ctx_low_slices_psize; 1127 index = GET_LOW_SLICE_INDEX(addr); 1128 } else { 1129 psizes = get_paca()->mm_ctx_high_slices_psize; 1130 index = GET_HIGH_SLICE_INDEX(addr); 1131 } 1132 mask_index = index & 0x1; 1133 return (psizes[index >> 1] >> (mask_index * 4)) & 0xF; 1134 } 1135 1136 #else 1137 unsigned int get_paca_psize(unsigned long addr) 1138 { 1139 return get_paca()->mm_ctx_user_psize; 1140 } 1141 #endif 1142 1143 /* 1144 * Demote a segment to using 4k pages. 1145 * For now this makes the whole process use 4k pages. 1146 */ 1147 #ifdef CONFIG_PPC_64K_PAGES 1148 void demote_segment_4k(struct mm_struct *mm, unsigned long addr) 1149 { 1150 if (get_slice_psize(mm, addr) == MMU_PAGE_4K) 1151 return; 1152 slice_set_range_psize(mm, addr, 1, MMU_PAGE_4K); 1153 copro_flush_all_slbs(mm); 1154 if ((get_paca_psize(addr) != MMU_PAGE_4K) && (current->mm == mm)) { 1155 1156 copy_mm_to_paca(mm); 1157 slb_flush_and_restore_bolted(); 1158 } 1159 } 1160 #endif /* CONFIG_PPC_64K_PAGES */ 1161 1162 #ifdef CONFIG_PPC_SUBPAGE_PROT 1163 /* 1164 * This looks up a 2-bit protection code for a 4k subpage of a 64k page. 1165 * Userspace sets the subpage permissions using the subpage_prot system call. 1166 * 1167 * Result is 0: full permissions, _PAGE_RW: read-only, 1168 * _PAGE_RWX: no access. 1169 */ 1170 static int subpage_protection(struct mm_struct *mm, unsigned long ea) 1171 { 1172 struct subpage_prot_table *spt = mm_ctx_subpage_prot(&mm->context); 1173 u32 spp = 0; 1174 u32 **sbpm, *sbpp; 1175 1176 if (!spt) 1177 return 0; 1178 1179 if (ea >= spt->maxaddr) 1180 return 0; 1181 if (ea < 0x100000000UL) { 1182 /* addresses below 4GB use spt->low_prot */ 1183 sbpm = spt->low_prot; 1184 } else { 1185 sbpm = spt->protptrs[ea >> SBP_L3_SHIFT]; 1186 if (!sbpm) 1187 return 0; 1188 } 1189 sbpp = sbpm[(ea >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1)]; 1190 if (!sbpp) 1191 return 0; 1192 spp = sbpp[(ea >> PAGE_SHIFT) & (SBP_L1_COUNT - 1)]; 1193 1194 /* extract 2-bit bitfield for this 4k subpage */ 1195 spp >>= 30 - 2 * ((ea >> 12) & 0xf); 1196 1197 /* 1198 * 0 -> full premission 1199 * 1 -> Read only 1200 * 2 -> no access. 1201 * We return the flag that need to be cleared. 1202 */ 1203 spp = ((spp & 2) ? _PAGE_RWX : 0) | ((spp & 1) ? _PAGE_WRITE : 0); 1204 return spp; 1205 } 1206 1207 #else /* CONFIG_PPC_SUBPAGE_PROT */ 1208 static inline int subpage_protection(struct mm_struct *mm, unsigned long ea) 1209 { 1210 return 0; 1211 } 1212 #endif 1213 1214 void hash_failure_debug(unsigned long ea, unsigned long access, 1215 unsigned long vsid, unsigned long trap, 1216 int ssize, int psize, int lpsize, unsigned long pte) 1217 { 1218 if (!printk_ratelimit()) 1219 return; 1220 pr_info("mm: Hashing failure ! EA=0x%lx access=0x%lx current=%s\n", 1221 ea, access, current->comm); 1222 pr_info(" trap=0x%lx vsid=0x%lx ssize=%d base psize=%d psize %d pte=0x%lx\n", 1223 trap, vsid, ssize, psize, lpsize, pte); 1224 } 1225 1226 static void check_paca_psize(unsigned long ea, struct mm_struct *mm, 1227 int psize, bool user_region) 1228 { 1229 if (user_region) { 1230 if (psize != get_paca_psize(ea)) { 1231 copy_mm_to_paca(mm); 1232 slb_flush_and_restore_bolted(); 1233 } 1234 } else if (get_paca()->vmalloc_sllp != 1235 mmu_psize_defs[mmu_vmalloc_psize].sllp) { 1236 get_paca()->vmalloc_sllp = 1237 mmu_psize_defs[mmu_vmalloc_psize].sllp; 1238 slb_vmalloc_update(); 1239 } 1240 } 1241 1242 /* 1243 * Result code is: 1244 * 0 - handled 1245 * 1 - normal page fault 1246 * -1 - critical hash insertion error 1247 * -2 - access not permitted by subpage protection mechanism 1248 */ 1249 int hash_page_mm(struct mm_struct *mm, unsigned long ea, 1250 unsigned long access, unsigned long trap, 1251 unsigned long flags) 1252 { 1253 bool is_thp; 1254 enum ctx_state prev_state = exception_enter(); 1255 pgd_t *pgdir; 1256 unsigned long vsid; 1257 pte_t *ptep; 1258 unsigned hugeshift; 1259 int rc, user_region = 0; 1260 int psize, ssize; 1261 1262 DBG_LOW("hash_page(ea=%016lx, access=%lx, trap=%lx\n", 1263 ea, access, trap); 1264 trace_hash_fault(ea, access, trap); 1265 1266 /* Get region & vsid */ 1267 switch (get_region_id(ea)) { 1268 case USER_REGION_ID: 1269 user_region = 1; 1270 if (! mm) { 1271 DBG_LOW(" user region with no mm !\n"); 1272 rc = 1; 1273 goto bail; 1274 } 1275 psize = get_slice_psize(mm, ea); 1276 ssize = user_segment_size(ea); 1277 vsid = get_user_vsid(&mm->context, ea, ssize); 1278 break; 1279 case VMALLOC_REGION_ID: 1280 vsid = get_kernel_vsid(ea, mmu_kernel_ssize); 1281 psize = mmu_vmalloc_psize; 1282 ssize = mmu_kernel_ssize; 1283 break; 1284 1285 case IO_REGION_ID: 1286 vsid = get_kernel_vsid(ea, mmu_kernel_ssize); 1287 psize = mmu_io_psize; 1288 ssize = mmu_kernel_ssize; 1289 break; 1290 default: 1291 /* 1292 * Not a valid range 1293 * Send the problem up to do_page_fault() 1294 */ 1295 rc = 1; 1296 goto bail; 1297 } 1298 DBG_LOW(" mm=%p, mm->pgdir=%p, vsid=%016lx\n", mm, mm->pgd, vsid); 1299 1300 /* Bad address. */ 1301 if (!vsid) { 1302 DBG_LOW("Bad address!\n"); 1303 rc = 1; 1304 goto bail; 1305 } 1306 /* Get pgdir */ 1307 pgdir = mm->pgd; 1308 if (pgdir == NULL) { 1309 rc = 1; 1310 goto bail; 1311 } 1312 1313 /* Check CPU locality */ 1314 if (user_region && mm_is_thread_local(mm)) 1315 flags |= HPTE_LOCAL_UPDATE; 1316 1317 #ifndef CONFIG_PPC_64K_PAGES 1318 /* 1319 * If we use 4K pages and our psize is not 4K, then we might 1320 * be hitting a special driver mapping, and need to align the 1321 * address before we fetch the PTE. 1322 * 1323 * It could also be a hugepage mapping, in which case this is 1324 * not necessary, but it's not harmful, either. 1325 */ 1326 if (psize != MMU_PAGE_4K) 1327 ea &= ~((1ul << mmu_psize_defs[psize].shift) - 1); 1328 #endif /* CONFIG_PPC_64K_PAGES */ 1329 1330 /* Get PTE and page size from page tables */ 1331 ptep = find_linux_pte(pgdir, ea, &is_thp, &hugeshift); 1332 if (ptep == NULL || !pte_present(*ptep)) { 1333 DBG_LOW(" no PTE !\n"); 1334 rc = 1; 1335 goto bail; 1336 } 1337 1338 /* Add _PAGE_PRESENT to the required access perm */ 1339 access |= _PAGE_PRESENT; 1340 1341 /* 1342 * Pre-check access permissions (will be re-checked atomically 1343 * in __hash_page_XX but this pre-check is a fast path 1344 */ 1345 if (!check_pte_access(access, pte_val(*ptep))) { 1346 DBG_LOW(" no access !\n"); 1347 rc = 1; 1348 goto bail; 1349 } 1350 1351 if (hugeshift) { 1352 if (is_thp) 1353 rc = __hash_page_thp(ea, access, vsid, (pmd_t *)ptep, 1354 trap, flags, ssize, psize); 1355 #ifdef CONFIG_HUGETLB_PAGE 1356 else 1357 rc = __hash_page_huge(ea, access, vsid, ptep, trap, 1358 flags, ssize, hugeshift, psize); 1359 #else 1360 else { 1361 /* 1362 * if we have hugeshift, and is not transhuge with 1363 * hugetlb disabled, something is really wrong. 1364 */ 1365 rc = 1; 1366 WARN_ON(1); 1367 } 1368 #endif 1369 if (current->mm == mm) 1370 check_paca_psize(ea, mm, psize, user_region); 1371 1372 goto bail; 1373 } 1374 1375 #ifndef CONFIG_PPC_64K_PAGES 1376 DBG_LOW(" i-pte: %016lx\n", pte_val(*ptep)); 1377 #else 1378 DBG_LOW(" i-pte: %016lx %016lx\n", pte_val(*ptep), 1379 pte_val(*(ptep + PTRS_PER_PTE))); 1380 #endif 1381 /* Do actual hashing */ 1382 #ifdef CONFIG_PPC_64K_PAGES 1383 /* If H_PAGE_4K_PFN is set, make sure this is a 4k segment */ 1384 if ((pte_val(*ptep) & H_PAGE_4K_PFN) && psize == MMU_PAGE_64K) { 1385 demote_segment_4k(mm, ea); 1386 psize = MMU_PAGE_4K; 1387 } 1388 1389 /* 1390 * If this PTE is non-cacheable and we have restrictions on 1391 * using non cacheable large pages, then we switch to 4k 1392 */ 1393 if (mmu_ci_restrictions && psize == MMU_PAGE_64K && pte_ci(*ptep)) { 1394 if (user_region) { 1395 demote_segment_4k(mm, ea); 1396 psize = MMU_PAGE_4K; 1397 } else if (ea < VMALLOC_END) { 1398 /* 1399 * some driver did a non-cacheable mapping 1400 * in vmalloc space, so switch vmalloc 1401 * to 4k pages 1402 */ 1403 printk(KERN_ALERT "Reducing vmalloc segment " 1404 "to 4kB pages because of " 1405 "non-cacheable mapping\n"); 1406 psize = mmu_vmalloc_psize = MMU_PAGE_4K; 1407 copro_flush_all_slbs(mm); 1408 } 1409 } 1410 1411 #endif /* CONFIG_PPC_64K_PAGES */ 1412 1413 if (current->mm == mm) 1414 check_paca_psize(ea, mm, psize, user_region); 1415 1416 #ifdef CONFIG_PPC_64K_PAGES 1417 if (psize == MMU_PAGE_64K) 1418 rc = __hash_page_64K(ea, access, vsid, ptep, trap, 1419 flags, ssize); 1420 else 1421 #endif /* CONFIG_PPC_64K_PAGES */ 1422 { 1423 int spp = subpage_protection(mm, ea); 1424 if (access & spp) 1425 rc = -2; 1426 else 1427 rc = __hash_page_4K(ea, access, vsid, ptep, trap, 1428 flags, ssize, spp); 1429 } 1430 1431 /* 1432 * Dump some info in case of hash insertion failure, they should 1433 * never happen so it is really useful to know if/when they do 1434 */ 1435 if (rc == -1) 1436 hash_failure_debug(ea, access, vsid, trap, ssize, psize, 1437 psize, pte_val(*ptep)); 1438 #ifndef CONFIG_PPC_64K_PAGES 1439 DBG_LOW(" o-pte: %016lx\n", pte_val(*ptep)); 1440 #else 1441 DBG_LOW(" o-pte: %016lx %016lx\n", pte_val(*ptep), 1442 pte_val(*(ptep + PTRS_PER_PTE))); 1443 #endif 1444 DBG_LOW(" -> rc=%d\n", rc); 1445 1446 bail: 1447 exception_exit(prev_state); 1448 return rc; 1449 } 1450 EXPORT_SYMBOL_GPL(hash_page_mm); 1451 1452 int hash_page(unsigned long ea, unsigned long access, unsigned long trap, 1453 unsigned long dsisr) 1454 { 1455 unsigned long flags = 0; 1456 struct mm_struct *mm = current->mm; 1457 1458 if ((get_region_id(ea) == VMALLOC_REGION_ID) || 1459 (get_region_id(ea) == IO_REGION_ID)) 1460 mm = &init_mm; 1461 1462 if (dsisr & DSISR_NOHPTE) 1463 flags |= HPTE_NOHPTE_UPDATE; 1464 1465 return hash_page_mm(mm, ea, access, trap, flags); 1466 } 1467 EXPORT_SYMBOL_GPL(hash_page); 1468 1469 int __hash_page(unsigned long ea, unsigned long msr, unsigned long trap, 1470 unsigned long dsisr) 1471 { 1472 unsigned long access = _PAGE_PRESENT | _PAGE_READ; 1473 unsigned long flags = 0; 1474 struct mm_struct *mm = current->mm; 1475 unsigned int region_id = get_region_id(ea); 1476 1477 if ((region_id == VMALLOC_REGION_ID) || (region_id == IO_REGION_ID)) 1478 mm = &init_mm; 1479 1480 if (dsisr & DSISR_NOHPTE) 1481 flags |= HPTE_NOHPTE_UPDATE; 1482 1483 if (dsisr & DSISR_ISSTORE) 1484 access |= _PAGE_WRITE; 1485 /* 1486 * We set _PAGE_PRIVILEGED only when 1487 * kernel mode access kernel space. 1488 * 1489 * _PAGE_PRIVILEGED is NOT set 1490 * 1) when kernel mode access user space 1491 * 2) user space access kernel space. 1492 */ 1493 access |= _PAGE_PRIVILEGED; 1494 if ((msr & MSR_PR) || (region_id == USER_REGION_ID)) 1495 access &= ~_PAGE_PRIVILEGED; 1496 1497 if (trap == 0x400) 1498 access |= _PAGE_EXEC; 1499 1500 return hash_page_mm(mm, ea, access, trap, flags); 1501 } 1502 1503 #ifdef CONFIG_PPC_MM_SLICES 1504 static bool should_hash_preload(struct mm_struct *mm, unsigned long ea) 1505 { 1506 int psize = get_slice_psize(mm, ea); 1507 1508 /* We only prefault standard pages for now */ 1509 if (unlikely(psize != mm_ctx_user_psize(&mm->context))) 1510 return false; 1511 1512 /* 1513 * Don't prefault if subpage protection is enabled for the EA. 1514 */ 1515 if (unlikely((psize == MMU_PAGE_4K) && subpage_protection(mm, ea))) 1516 return false; 1517 1518 return true; 1519 } 1520 #else 1521 static bool should_hash_preload(struct mm_struct *mm, unsigned long ea) 1522 { 1523 return true; 1524 } 1525 #endif 1526 1527 void hash_preload(struct mm_struct *mm, unsigned long ea, 1528 bool is_exec, unsigned long trap) 1529 { 1530 int hugepage_shift; 1531 unsigned long vsid; 1532 pgd_t *pgdir; 1533 pte_t *ptep; 1534 unsigned long flags; 1535 int rc, ssize, update_flags = 0; 1536 unsigned long access = _PAGE_PRESENT | _PAGE_READ | (is_exec ? _PAGE_EXEC : 0); 1537 1538 BUG_ON(get_region_id(ea) != USER_REGION_ID); 1539 1540 if (!should_hash_preload(mm, ea)) 1541 return; 1542 1543 DBG_LOW("hash_preload(mm=%p, mm->pgdir=%p, ea=%016lx, access=%lx," 1544 " trap=%lx\n", mm, mm->pgd, ea, access, trap); 1545 1546 /* Get Linux PTE if available */ 1547 pgdir = mm->pgd; 1548 if (pgdir == NULL) 1549 return; 1550 1551 /* Get VSID */ 1552 ssize = user_segment_size(ea); 1553 vsid = get_user_vsid(&mm->context, ea, ssize); 1554 if (!vsid) 1555 return; 1556 /* 1557 * Hash doesn't like irqs. Walking linux page table with irq disabled 1558 * saves us from holding multiple locks. 1559 */ 1560 local_irq_save(flags); 1561 1562 /* 1563 * THP pages use update_mmu_cache_pmd. We don't do 1564 * hash preload there. Hence can ignore THP here 1565 */ 1566 ptep = find_current_mm_pte(pgdir, ea, NULL, &hugepage_shift); 1567 if (!ptep) 1568 goto out_exit; 1569 1570 WARN_ON(hugepage_shift); 1571 #ifdef CONFIG_PPC_64K_PAGES 1572 /* If either H_PAGE_4K_PFN or cache inhibited is set (and we are on 1573 * a 64K kernel), then we don't preload, hash_page() will take 1574 * care of it once we actually try to access the page. 1575 * That way we don't have to duplicate all of the logic for segment 1576 * page size demotion here 1577 */ 1578 if ((pte_val(*ptep) & H_PAGE_4K_PFN) || pte_ci(*ptep)) 1579 goto out_exit; 1580 #endif /* CONFIG_PPC_64K_PAGES */ 1581 1582 /* Is that local to this CPU ? */ 1583 if (mm_is_thread_local(mm)) 1584 update_flags |= HPTE_LOCAL_UPDATE; 1585 1586 /* Hash it in */ 1587 #ifdef CONFIG_PPC_64K_PAGES 1588 if (mm_ctx_user_psize(&mm->context) == MMU_PAGE_64K) 1589 rc = __hash_page_64K(ea, access, vsid, ptep, trap, 1590 update_flags, ssize); 1591 else 1592 #endif /* CONFIG_PPC_64K_PAGES */ 1593 rc = __hash_page_4K(ea, access, vsid, ptep, trap, update_flags, 1594 ssize, subpage_protection(mm, ea)); 1595 1596 /* Dump some info in case of hash insertion failure, they should 1597 * never happen so it is really useful to know if/when they do 1598 */ 1599 if (rc == -1) 1600 hash_failure_debug(ea, access, vsid, trap, ssize, 1601 mm_ctx_user_psize(&mm->context), 1602 mm_ctx_user_psize(&mm->context), 1603 pte_val(*ptep)); 1604 out_exit: 1605 local_irq_restore(flags); 1606 } 1607 1608 #ifdef CONFIG_PPC_MEM_KEYS 1609 /* 1610 * Return the protection key associated with the given address and the 1611 * mm_struct. 1612 */ 1613 u16 get_mm_addr_key(struct mm_struct *mm, unsigned long address) 1614 { 1615 pte_t *ptep; 1616 u16 pkey = 0; 1617 unsigned long flags; 1618 1619 if (!mm || !mm->pgd) 1620 return 0; 1621 1622 local_irq_save(flags); 1623 ptep = find_linux_pte(mm->pgd, address, NULL, NULL); 1624 if (ptep) 1625 pkey = pte_to_pkey_bits(pte_val(READ_ONCE(*ptep))); 1626 local_irq_restore(flags); 1627 1628 return pkey; 1629 } 1630 #endif /* CONFIG_PPC_MEM_KEYS */ 1631 1632 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 1633 static inline void tm_flush_hash_page(int local) 1634 { 1635 /* 1636 * Transactions are not aborted by tlbiel, only tlbie. Without, syncing a 1637 * page back to a block device w/PIO could pick up transactional data 1638 * (bad!) so we force an abort here. Before the sync the page will be 1639 * made read-only, which will flush_hash_page. BIG ISSUE here: if the 1640 * kernel uses a page from userspace without unmapping it first, it may 1641 * see the speculated version. 1642 */ 1643 if (local && cpu_has_feature(CPU_FTR_TM) && current->thread.regs && 1644 MSR_TM_ACTIVE(current->thread.regs->msr)) { 1645 tm_enable(); 1646 tm_abort(TM_CAUSE_TLBI); 1647 } 1648 } 1649 #else 1650 static inline void tm_flush_hash_page(int local) 1651 { 1652 } 1653 #endif 1654 1655 /* 1656 * Return the global hash slot, corresponding to the given PTE, which contains 1657 * the HPTE. 1658 */ 1659 unsigned long pte_get_hash_gslot(unsigned long vpn, unsigned long shift, 1660 int ssize, real_pte_t rpte, unsigned int subpg_index) 1661 { 1662 unsigned long hash, gslot, hidx; 1663 1664 hash = hpt_hash(vpn, shift, ssize); 1665 hidx = __rpte_to_hidx(rpte, subpg_index); 1666 if (hidx & _PTEIDX_SECONDARY) 1667 hash = ~hash; 1668 gslot = (hash & htab_hash_mask) * HPTES_PER_GROUP; 1669 gslot += hidx & _PTEIDX_GROUP_IX; 1670 return gslot; 1671 } 1672 1673 /* 1674 * WARNING: This is called from hash_low_64.S, if you change this prototype, 1675 * do not forget to update the assembly call site ! 1676 */ 1677 void flush_hash_page(unsigned long vpn, real_pte_t pte, int psize, int ssize, 1678 unsigned long flags) 1679 { 1680 unsigned long index, shift, gslot; 1681 int local = flags & HPTE_LOCAL_UPDATE; 1682 1683 DBG_LOW("flush_hash_page(vpn=%016lx)\n", vpn); 1684 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) { 1685 gslot = pte_get_hash_gslot(vpn, shift, ssize, pte, index); 1686 DBG_LOW(" sub %ld: gslot=%lx\n", index, gslot); 1687 /* 1688 * We use same base page size and actual psize, because we don't 1689 * use these functions for hugepage 1690 */ 1691 mmu_hash_ops.hpte_invalidate(gslot, vpn, psize, psize, 1692 ssize, local); 1693 } pte_iterate_hashed_end(); 1694 1695 tm_flush_hash_page(local); 1696 } 1697 1698 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1699 void flush_hash_hugepage(unsigned long vsid, unsigned long addr, 1700 pmd_t *pmdp, unsigned int psize, int ssize, 1701 unsigned long flags) 1702 { 1703 int i, max_hpte_count, valid; 1704 unsigned long s_addr; 1705 unsigned char *hpte_slot_array; 1706 unsigned long hidx, shift, vpn, hash, slot; 1707 int local = flags & HPTE_LOCAL_UPDATE; 1708 1709 s_addr = addr & HPAGE_PMD_MASK; 1710 hpte_slot_array = get_hpte_slot_array(pmdp); 1711 /* 1712 * IF we try to do a HUGE PTE update after a withdraw is done. 1713 * we will find the below NULL. This happens when we do 1714 * split_huge_page_pmd 1715 */ 1716 if (!hpte_slot_array) 1717 return; 1718 1719 if (mmu_hash_ops.hugepage_invalidate) { 1720 mmu_hash_ops.hugepage_invalidate(vsid, s_addr, hpte_slot_array, 1721 psize, ssize, local); 1722 goto tm_abort; 1723 } 1724 /* 1725 * No bluk hpte removal support, invalidate each entry 1726 */ 1727 shift = mmu_psize_defs[psize].shift; 1728 max_hpte_count = HPAGE_PMD_SIZE >> shift; 1729 for (i = 0; i < max_hpte_count; i++) { 1730 /* 1731 * 8 bits per each hpte entries 1732 * 000| [ secondary group (one bit) | hidx (3 bits) | valid bit] 1733 */ 1734 valid = hpte_valid(hpte_slot_array, i); 1735 if (!valid) 1736 continue; 1737 hidx = hpte_hash_index(hpte_slot_array, i); 1738 1739 /* get the vpn */ 1740 addr = s_addr + (i * (1ul << shift)); 1741 vpn = hpt_vpn(addr, vsid, ssize); 1742 hash = hpt_hash(vpn, shift, ssize); 1743 if (hidx & _PTEIDX_SECONDARY) 1744 hash = ~hash; 1745 1746 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP; 1747 slot += hidx & _PTEIDX_GROUP_IX; 1748 mmu_hash_ops.hpte_invalidate(slot, vpn, psize, 1749 MMU_PAGE_16M, ssize, local); 1750 } 1751 tm_abort: 1752 tm_flush_hash_page(local); 1753 } 1754 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1755 1756 void flush_hash_range(unsigned long number, int local) 1757 { 1758 if (mmu_hash_ops.flush_hash_range) 1759 mmu_hash_ops.flush_hash_range(number, local); 1760 else { 1761 int i; 1762 struct ppc64_tlb_batch *batch = 1763 this_cpu_ptr(&ppc64_tlb_batch); 1764 1765 for (i = 0; i < number; i++) 1766 flush_hash_page(batch->vpn[i], batch->pte[i], 1767 batch->psize, batch->ssize, local); 1768 } 1769 } 1770 1771 /* 1772 * low_hash_fault is called when we the low level hash code failed 1773 * to instert a PTE due to an hypervisor error 1774 */ 1775 void low_hash_fault(struct pt_regs *regs, unsigned long address, int rc) 1776 { 1777 enum ctx_state prev_state = exception_enter(); 1778 1779 if (user_mode(regs)) { 1780 #ifdef CONFIG_PPC_SUBPAGE_PROT 1781 if (rc == -2) 1782 _exception(SIGSEGV, regs, SEGV_ACCERR, address); 1783 else 1784 #endif 1785 _exception(SIGBUS, regs, BUS_ADRERR, address); 1786 } else 1787 bad_page_fault(regs, address, SIGBUS); 1788 1789 exception_exit(prev_state); 1790 } 1791 1792 long hpte_insert_repeating(unsigned long hash, unsigned long vpn, 1793 unsigned long pa, unsigned long rflags, 1794 unsigned long vflags, int psize, int ssize) 1795 { 1796 unsigned long hpte_group; 1797 long slot; 1798 1799 repeat: 1800 hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP; 1801 1802 /* Insert into the hash table, primary slot */ 1803 slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa, rflags, vflags, 1804 psize, psize, ssize); 1805 1806 /* Primary is full, try the secondary */ 1807 if (unlikely(slot == -1)) { 1808 hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP; 1809 slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa, rflags, 1810 vflags | HPTE_V_SECONDARY, 1811 psize, psize, ssize); 1812 if (slot == -1) { 1813 if (mftb() & 0x1) 1814 hpte_group = (hash & htab_hash_mask) * 1815 HPTES_PER_GROUP; 1816 1817 mmu_hash_ops.hpte_remove(hpte_group); 1818 goto repeat; 1819 } 1820 } 1821 1822 return slot; 1823 } 1824 1825 #ifdef CONFIG_DEBUG_PAGEALLOC 1826 static void kernel_map_linear_page(unsigned long vaddr, unsigned long lmi) 1827 { 1828 unsigned long hash; 1829 unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize); 1830 unsigned long vpn = hpt_vpn(vaddr, vsid, mmu_kernel_ssize); 1831 unsigned long mode = htab_convert_pte_flags(pgprot_val(PAGE_KERNEL)); 1832 long ret; 1833 1834 hash = hpt_hash(vpn, PAGE_SHIFT, mmu_kernel_ssize); 1835 1836 /* Don't create HPTE entries for bad address */ 1837 if (!vsid) 1838 return; 1839 1840 ret = hpte_insert_repeating(hash, vpn, __pa(vaddr), mode, 1841 HPTE_V_BOLTED, 1842 mmu_linear_psize, mmu_kernel_ssize); 1843 1844 BUG_ON (ret < 0); 1845 spin_lock(&linear_map_hash_lock); 1846 BUG_ON(linear_map_hash_slots[lmi] & 0x80); 1847 linear_map_hash_slots[lmi] = ret | 0x80; 1848 spin_unlock(&linear_map_hash_lock); 1849 } 1850 1851 static void kernel_unmap_linear_page(unsigned long vaddr, unsigned long lmi) 1852 { 1853 unsigned long hash, hidx, slot; 1854 unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize); 1855 unsigned long vpn = hpt_vpn(vaddr, vsid, mmu_kernel_ssize); 1856 1857 hash = hpt_hash(vpn, PAGE_SHIFT, mmu_kernel_ssize); 1858 spin_lock(&linear_map_hash_lock); 1859 BUG_ON(!(linear_map_hash_slots[lmi] & 0x80)); 1860 hidx = linear_map_hash_slots[lmi] & 0x7f; 1861 linear_map_hash_slots[lmi] = 0; 1862 spin_unlock(&linear_map_hash_lock); 1863 if (hidx & _PTEIDX_SECONDARY) 1864 hash = ~hash; 1865 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP; 1866 slot += hidx & _PTEIDX_GROUP_IX; 1867 mmu_hash_ops.hpte_invalidate(slot, vpn, mmu_linear_psize, 1868 mmu_linear_psize, 1869 mmu_kernel_ssize, 0); 1870 } 1871 1872 void __kernel_map_pages(struct page *page, int numpages, int enable) 1873 { 1874 unsigned long flags, vaddr, lmi; 1875 int i; 1876 1877 local_irq_save(flags); 1878 for (i = 0; i < numpages; i++, page++) { 1879 vaddr = (unsigned long)page_address(page); 1880 lmi = __pa(vaddr) >> PAGE_SHIFT; 1881 if (lmi >= linear_map_hash_count) 1882 continue; 1883 if (enable) 1884 kernel_map_linear_page(vaddr, lmi); 1885 else 1886 kernel_unmap_linear_page(vaddr, lmi); 1887 } 1888 local_irq_restore(flags); 1889 } 1890 #endif /* CONFIG_DEBUG_PAGEALLOC */ 1891 1892 void hash__setup_initial_memory_limit(phys_addr_t first_memblock_base, 1893 phys_addr_t first_memblock_size) 1894 { 1895 /* 1896 * We don't currently support the first MEMBLOCK not mapping 0 1897 * physical on those processors 1898 */ 1899 BUG_ON(first_memblock_base != 0); 1900 1901 /* 1902 * On virtualized systems the first entry is our RMA region aka VRMA, 1903 * non-virtualized 64-bit hash MMU systems don't have a limitation 1904 * on real mode access. 1905 * 1906 * For guests on platforms before POWER9, we clamp the it limit to 1G 1907 * to avoid some funky things such as RTAS bugs etc... 1908 */ 1909 if (!early_cpu_has_feature(CPU_FTR_HVMODE)) { 1910 ppc64_rma_size = first_memblock_size; 1911 if (!early_cpu_has_feature(CPU_FTR_ARCH_300)) 1912 ppc64_rma_size = min_t(u64, ppc64_rma_size, 0x40000000); 1913 1914 /* Finally limit subsequent allocations */ 1915 memblock_set_current_limit(ppc64_rma_size); 1916 } else { 1917 ppc64_rma_size = ULONG_MAX; 1918 } 1919 } 1920 1921 #ifdef CONFIG_DEBUG_FS 1922 1923 static int hpt_order_get(void *data, u64 *val) 1924 { 1925 *val = ppc64_pft_size; 1926 return 0; 1927 } 1928 1929 static int hpt_order_set(void *data, u64 val) 1930 { 1931 if (!mmu_hash_ops.resize_hpt) 1932 return -ENODEV; 1933 1934 return mmu_hash_ops.resize_hpt(val); 1935 } 1936 1937 DEFINE_DEBUGFS_ATTRIBUTE(fops_hpt_order, hpt_order_get, hpt_order_set, "%llu\n"); 1938 1939 static int __init hash64_debugfs(void) 1940 { 1941 if (!debugfs_create_file_unsafe("hpt_order", 0600, powerpc_debugfs_root, 1942 NULL, &fops_hpt_order)) { 1943 pr_err("lpar: unable to create hpt_order debugsfs file\n"); 1944 } 1945 1946 return 0; 1947 } 1948 machine_device_initcall(pseries, hash64_debugfs); 1949 #endif /* CONFIG_DEBUG_FS */ 1950 1951 void __init print_system_hash_info(void) 1952 { 1953 pr_info("ppc64_pft_size = 0x%llx\n", ppc64_pft_size); 1954 1955 if (htab_hash_mask) 1956 pr_info("htab_hash_mask = 0x%lx\n", htab_hash_mask); 1957 pr_info("kernel vmalloc start = 0x%lx\n", KERN_VIRT_START); 1958 pr_info("kernel IO start = 0x%lx\n", KERN_IO_START); 1959 pr_info("kernel vmemmap start = 0x%lx\n", (unsigned long)vmemmap); 1960 } 1961