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#include <linux/config.h> 22#include <linux/spinlock.h> 23#include <linux/errno.h> 24#include <linux/sched.h> 25#include <linux/proc_fs.h> 26#include <linux/stat.h> 27#include <linux/sysctl.h> 28#include <linux/ctype.h> 29#include <linux/cache.h> 30 31#include <asm/ppcdebug.h> 32#include <asm/processor.h> 33#include <asm/pgtable.h> 34#include <asm/mmu.h> 35#include <asm/mmu_context.h> 36#include <asm/page.h> 37#include <asm/types.h> 38#include <asm/uaccess.h> 39#include <asm/naca.h> 40#include <asm/pmc.h> 41#include <asm/machdep.h> 42#include <asm/lmb.h> 43#include <asm/abs_addr.h> 44#include <asm/io.h> 45#include <asm/eeh.h> 46#include <asm/hvcall.h> 47#include <asm/iSeries/LparData.h> 48#include <asm/iSeries/HvCallHpt.h> 49 50/* 51 * Note: pte --> Linux PTE 52 * HPTE --> PowerPC Hashed Page Table Entry 53 * 54 * Execution context: 55 * htab_initialize is called with the MMU off (of course), but 56 * the kernel has been copied down to zero so it can directly 57 * reference global data. At this point it is very difficult 58 * to print debug info. 59 * 60 */ 61 62HTAB htab_data = {NULL, 0, 0, 0, 0}; 63 64extern unsigned long _SDR1; 65extern unsigned long klimit; 66 67void make_pte(HPTE *htab, unsigned long va, unsigned long pa, 68 int mode, unsigned long hash_mask, int large); 69long plpar_pte_enter(unsigned long flags, 70 unsigned long ptex, 71 unsigned long new_pteh, unsigned long new_ptel, 72 unsigned long *old_pteh_ret, unsigned long *old_ptel_ret); 73static long hpte_remove(unsigned long hpte_group); 74static long rpa_lpar_hpte_remove(unsigned long hpte_group); 75static long iSeries_hpte_remove(unsigned long hpte_group); 76 77static spinlock_t pSeries_tlbie_lock = SPIN_LOCK_UNLOCKED; 78static spinlock_t pSeries_lpar_tlbie_lock = SPIN_LOCK_UNLOCKED; 79spinlock_t hash_table_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED; 80 81#define KB (1024) 82#define MB (1024*KB) 83 84static inline void 85loop_forever(void) 86{ 87 volatile unsigned long x = 1; 88 for(;x;x|=1) 89 ; 90} 91 92static inline void 93create_pte_mapping(unsigned long start, unsigned long end, 94 unsigned long mode, unsigned long mask, int large) 95{ 96 unsigned long addr; 97 HPTE *htab = (HPTE *)__v2a(htab_data.htab); 98 unsigned int step; 99 100 if (large) 101 step = 16*MB; 102 else 103 step = 4*KB; 104 105 for (addr = start; addr < end; addr += step) { 106 unsigned long vsid = get_kernel_vsid(addr); 107 unsigned long va = (vsid << 28) | (addr & 0xfffffff); 108 make_pte(htab, va, (unsigned long)__v2a(addr), 109 mode, mask, large); 110 } 111} 112 113void 114htab_initialize(void) 115{ 116 unsigned long table, htab_size_bytes; 117 unsigned long pteg_count; 118 unsigned long mode_rw, mask; 119 120 /* 121 * Calculate the required size of the htab. We want the number of 122 * PTEGs to equal one half the number of real pages. 123 */ 124 htab_size_bytes = 1UL << naca->pftSize; 125 pteg_count = htab_size_bytes >> 7; 126 127 /* For debug, make the HTAB 1/8 as big as it normally would be. */ 128 ifppcdebug(PPCDBG_HTABSIZE) { 129 pteg_count >>= 3; 130 htab_size_bytes = pteg_count << 7; 131 } 132 133 htab_data.htab_num_ptegs = pteg_count; 134 htab_data.htab_hash_mask = pteg_count - 1; 135 136 if(naca->platform == PLATFORM_PSERIES) { 137 /* Find storage for the HPT. Must be contiguous in 138 * the absolute address space. 139 */ 140 table = lmb_alloc(htab_size_bytes, htab_size_bytes); 141 if ( !table ) { 142 ppc64_terminate_msg(0x20, "hpt space"); 143 loop_forever(); 144 } 145 htab_data.htab = (HPTE *)__a2v(table); 146 147 /* htab absolute addr + encoded htabsize */ 148 _SDR1 = table + __ilog2(pteg_count) - 11; 149 150 /* Initialize the HPT with no entries */ 151 memset((void *)table, 0, htab_size_bytes); 152 } else { 153 /* Using a hypervisor which owns the htab */ 154 htab_data.htab = NULL; 155 _SDR1 = 0; 156 } 157 158 mode_rw = _PAGE_ACCESSED | _PAGE_COHERENT | PP_RWXX; 159 mask = pteg_count-1; 160 161 if ((naca->platform & PLATFORM_PSERIES) && 162 cpu_has_largepage() && (naca->physicalMemorySize > 256*MB)) { 163 create_pte_mapping((unsigned long)KERNELBASE, 164 KERNELBASE + 256*MB, mode_rw, mask, 0); 165 create_pte_mapping((unsigned long)KERNELBASE + 256*MB, 166 KERNELBASE + (naca->physicalMemorySize), 167 mode_rw, mask, 1); 168 } else { 169 create_pte_mapping((unsigned long)KERNELBASE, 170 KERNELBASE+(naca->physicalMemorySize), 171 mode_rw, mask, 0); 172 } 173} 174#undef KB 175#undef MB 176 177/* 178 * Create a pte. Used during initialization only. 179 * We assume the PTE will fit in the primary PTEG. 180 */ 181void make_pte(HPTE *htab, unsigned long va, unsigned long pa, 182 int mode, unsigned long hash_mask, int large) 183{ 184 HPTE *hptep, local_hpte, rhpte; 185 unsigned long hash, vpn, flags, lpar_rc; 186 unsigned long i, dummy1, dummy2; 187 long slot; 188 189 if (large) 190 vpn = va >> LARGE_PAGE_SHIFT; 191 else 192 vpn = va >> PAGE_SHIFT; 193 194 hash = hpt_hash(vpn, large); 195 196 local_hpte.dw1.dword1 = pa | mode; 197 local_hpte.dw0.dword0 = 0; 198 local_hpte.dw0.dw0.avpn = va >> 23; 199 local_hpte.dw0.dw0.bolted = 1; /* bolted */ 200 if (large) { 201 local_hpte.dw0.dw0.l = 1; /* large page */ 202 local_hpte.dw0.dw0.avpn &= ~0x1UL; 203 } 204 local_hpte.dw0.dw0.v = 1; 205 206 if (naca->platform == PLATFORM_PSERIES) { 207 hptep = htab + ((hash & hash_mask)*HPTES_PER_GROUP); 208 209 for (i = 0; i < 8; ++i, ++hptep) { 210 if (hptep->dw0.dw0.v == 0) { /* !valid */ 211 *hptep = local_hpte; 212 return; 213 } 214 } 215 } else if (naca->platform == PLATFORM_PSERIES_LPAR) { 216 slot = ((hash & hash_mask)*HPTES_PER_GROUP); 217 218 /* Set CEC cookie to 0 */ 219 /* Zero page = 0 */ 220 /* I-cache Invalidate = 0 */ 221 /* I-cache synchronize = 0 */ 222 /* Exact = 0 - modify any entry in group */ 223 flags = 0; 224 225 lpar_rc = plpar_pte_enter(flags, slot, local_hpte.dw0.dword0, 226 local_hpte.dw1.dword1, 227 &dummy1, &dummy2); 228 if (lpar_rc != H_Success) { 229 ppc64_terminate_msg(0x21, "hpte enter"); 230 loop_forever(); 231 } 232 return; 233 } else if (naca->platform == PLATFORM_ISERIES_LPAR) { 234 slot = HvCallHpt_findValid(&rhpte, vpn); 235 if (slot < 0) { 236 /* Must find space in primary group */ 237 panic("hash_page: hpte already exists\n"); 238 } 239 HvCallHpt_addValidate(slot, 0, (HPTE *)&local_hpte ); 240 return; 241 } 242 243 /* We should _never_ get here and too early to call xmon. */ 244 ppc64_terminate_msg(0x22, "hpte platform"); 245 loop_forever(); 246} 247 248/* 249 * find_linux_pte returns the address of a linux pte for a given 250 * effective address and directory. If not found, it returns zero. 251 */ 252pte_t *find_linux_pte(pgd_t *pgdir, unsigned long ea) 253{ 254 pgd_t *pg; 255 pmd_t *pm; 256 pte_t *pt = NULL; 257 pte_t pte; 258 259 pg = pgdir + pgd_index(ea); 260 if (!pgd_none(*pg)) { 261 pm = pmd_offset(pg, ea); 262 if (!pmd_none(*pm)) { 263 pt = pte_offset(pm, ea); 264 pte = *pt; 265 if (!pte_present(pte)) 266 pt = NULL; 267 } 268 } 269 270 return pt; 271} 272 273static inline unsigned long computeHptePP(unsigned long pte) 274{ 275 return (pte & _PAGE_USER) | 276 (((pte & _PAGE_USER) >> 1) & 277 ((~((pte >> 2) & /* _PAGE_RW */ 278 (pte >> 7))) & /* _PAGE_DIRTY */ 279 1)); 280} 281 282/* 283 * Handle a fault by adding an HPTE. If the address can't be determined 284 * to be valid via Linux page tables, return 1. If handled return 0 285 */ 286int __hash_page(unsigned long ea, unsigned long access, 287 unsigned long vsid, pte_t *ptep) 288{ 289 unsigned long va, vpn; 290 unsigned long newpp, prpn; 291 unsigned long hpteflags; 292 long slot; 293 pte_t old_pte, new_pte; 294 295 /* Search the Linux page table for a match with va */ 296 va = (vsid << 28) | (ea & 0x0fffffff); 297 vpn = va >> PAGE_SHIFT; 298 299 /* Acquire the hash table lock to guarantee that the linux 300 * pte we fetch will not change 301 */ 302 spin_lock( &hash_table_lock ); 303 304 /* 305 * Check the user's access rights to the page. If access should be 306 * prevented then send the problem up to do_page_fault. 307 */ 308 access |= _PAGE_PRESENT; 309 if (unlikely(access & ~(pte_val(*ptep)))) { 310 spin_unlock( &hash_table_lock ); 311 return 1; 312 } 313 314 /* 315 * We have found a pte (which was present). 316 * The spinlocks prevent this status from changing 317 * The hash_table_lock prevents the _PAGE_HASHPTE status 318 * from changing (RPN, DIRTY and ACCESSED too) 319 * The page_table_lock prevents the pte from being 320 * invalidated or modified 321 */ 322 323 /* 324 * At this point, we have a pte (old_pte) which can be used to build 325 * or update an HPTE. There are 2 cases: 326 * 327 * 1. There is a valid (present) pte with no associated HPTE (this is 328 * the most common case) 329 * 2. There is a valid (present) pte with an associated HPTE. The 330 * current values of the pp bits in the HPTE prevent access 331 * because we are doing software DIRTY bit management and the 332 * page is currently not DIRTY. 333 */ 334 335 old_pte = *ptep; 336 new_pte = old_pte; 337 338 /* If the attempted access was a store */ 339 if (access & _PAGE_RW) 340 pte_val(new_pte) |= _PAGE_ACCESSED | _PAGE_DIRTY; 341 else 342 pte_val(new_pte) |= _PAGE_ACCESSED; 343 344 newpp = computeHptePP(pte_val(new_pte)); 345 346 /* Check if pte already has an hpte (case 2) */ 347 if (unlikely(pte_val(old_pte) & _PAGE_HASHPTE)) { 348 /* There MIGHT be an HPTE for this pte */ 349 unsigned long hash, slot, secondary; 350 351 hash = hpt_hash(vpn, 0); 352 secondary = (pte_val(old_pte) & _PAGE_SECONDARY) >> 15; 353 if (secondary) 354 hash = ~hash; 355 slot = (hash & htab_data.htab_hash_mask) * HPTES_PER_GROUP; 356 slot += (pte_val(old_pte) & _PAGE_GROUP_IX) >> 12; 357 358 if (ppc_md.hpte_updatepp(slot, secondary, 359 newpp, va, 0) == -1) { 360 pte_val(old_pte) &= ~_PAGE_HPTEFLAGS; 361 } else { 362 if (!pte_same(old_pte, new_pte)) { 363 *ptep = new_pte; 364 } 365 } 366 } 367 368 if (likely(!(pte_val(old_pte) & _PAGE_HASHPTE))) { 369 /* Update the linux pte with the HPTE slot */ 370 pte_val(new_pte) &= ~_PAGE_HPTEFLAGS; 371 pte_val(new_pte) |= _PAGE_HASHPTE; 372 prpn = pte_val(old_pte) >> PTE_SHIFT; 373 374 /* copy appropriate flags from linux pte */ 375 hpteflags = (pte_val(new_pte) & 0x1f8) | newpp; 376 377 slot = ppc_md.hpte_insert(vpn, prpn, hpteflags, 0, 0); 378 379 pte_val(new_pte) |= ((slot<<12) & 380 (_PAGE_GROUP_IX | _PAGE_SECONDARY)); 381 382 *ptep = new_pte; 383 } 384 385 spin_unlock(&hash_table_lock); 386 387 return 0; 388} 389 390/* 391 * Handle a fault by adding an HPTE. If the address can't be determined 392 * to be valid via Linux page tables, return 1. If handled return 0 393 */ 394int hash_page(unsigned long ea, unsigned long access) 395{ 396 void *pgdir; 397 unsigned long vsid; 398 struct mm_struct *mm; 399 pte_t *ptep; 400 int ret; 401 402 /* Check for invalid addresses. */ 403 if (!IS_VALID_EA(ea)) return 1; 404 405 switch (REGION_ID(ea)) { 406 case USER_REGION_ID: 407 mm = current->mm; 408 if (mm == NULL) return 1; 409 vsid = get_vsid(mm->context, ea); 410 break; 411 case IO_REGION_ID: 412 mm = &ioremap_mm; 413 vsid = get_kernel_vsid(ea); 414 break; 415 case VMALLOC_REGION_ID: 416 mm = &init_mm; 417 vsid = get_kernel_vsid(ea); 418 break; 419 case IO_UNMAPPED_REGION_ID: 420 udbg_printf("EEH Error ea = 0x%lx\n", ea); 421 PPCDBG_ENTER_DEBUGGER(); 422 panic("EEH Error ea = 0x%lx\n", ea); 423 break; 424 case KERNEL_REGION_ID: 425 /* 426 * As htab_initialize is now, we shouldn't ever get here since 427 * we're bolting the entire 0xC0... region. 428 */ 429 udbg_printf("Little faulted on kernel address 0x%lx\n", ea); 430 PPCDBG_ENTER_DEBUGGER(); 431 panic("Little faulted on kernel address 0x%lx\n", ea); 432 break; 433 default: 434 /* Not a valid range, send the problem up to do_page_fault */ 435 return 1; 436 break; 437 } 438 439 pgdir = mm->pgd; 440 if (pgdir == NULL) return 1; 441 442 /* 443 * Lock the Linux page table to prevent mmap and kswapd 444 * from modifying entries while we search and update 445 */ 446 spin_lock(&mm->page_table_lock); 447 448 ptep = find_linux_pte(pgdir, ea); 449 /* 450 * If no pte found or not present, send the problem up to 451 * do_page_fault 452 */ 453 if (ptep && pte_present(*ptep)) { 454 ret = __hash_page(ea, access, vsid, ptep); 455 } else { 456 /* If no pte, send the problem up to do_page_fault */ 457 ret = 1; 458 } 459 460 spin_unlock(&mm->page_table_lock); 461 462 return ret; 463} 464 465void flush_hash_page(unsigned long context, unsigned long ea, pte_t *ptep) 466{ 467 unsigned long vsid, vpn, va, hash, secondary, slot, flags; 468 unsigned long large = 0, local = 0; 469 pte_t pte; 470 471 if ((ea >= USER_START) && (ea <= USER_END)) 472 vsid = get_vsid(context, ea); 473 else 474 vsid = get_kernel_vsid(ea); 475 476 va = (vsid << 28) | (ea & 0x0fffffff); 477 if (large) 478 vpn = va >> LARGE_PAGE_SHIFT; 479 else 480 vpn = va >> PAGE_SHIFT; 481 hash = hpt_hash(vpn, large); 482 483 spin_lock_irqsave( &hash_table_lock, flags); 484 485 pte = __pte(pte_update(ptep, _PAGE_HPTEFLAGS, 0)); 486 secondary = (pte_val(pte) & _PAGE_SECONDARY) >> 15; 487 if (secondary) hash = ~hash; 488 slot = (hash & htab_data.htab_hash_mask) * HPTES_PER_GROUP; 489 slot += (pte_val(pte) & _PAGE_GROUP_IX) >> 12; 490 491 if (pte_val(pte) & _PAGE_HASHPTE) { 492 ppc_md.hpte_invalidate(slot, secondary, va, large, local); 493 } 494 495 spin_unlock_irqrestore( &hash_table_lock, flags ); 496} 497 498long plpar_pte_enter(unsigned long flags, 499 unsigned long ptex, 500 unsigned long new_pteh, unsigned long new_ptel, 501 unsigned long *old_pteh_ret, unsigned long *old_ptel_ret) 502{ 503 unsigned long dummy, ret; 504 ret = plpar_hcall(H_ENTER, flags, ptex, new_pteh, new_ptel, 505 old_pteh_ret, old_ptel_ret, &dummy); 506 return(ret); 507} 508 509long plpar_pte_remove(unsigned long flags, 510 unsigned long ptex, 511 unsigned long avpn, 512 unsigned long *old_pteh_ret, unsigned long *old_ptel_ret) 513{ 514 unsigned long dummy; 515 return plpar_hcall(H_REMOVE, flags, ptex, avpn, 0, 516 old_pteh_ret, old_ptel_ret, &dummy); 517} 518 519long plpar_pte_read(unsigned long flags, 520 unsigned long ptex, 521 unsigned long *old_pteh_ret, unsigned long *old_ptel_ret) 522{ 523 unsigned long dummy; 524 return plpar_hcall(H_READ, flags, ptex, 0, 0, 525 old_pteh_ret, old_ptel_ret, &dummy); 526} 527 528long plpar_pte_protect(unsigned long flags, 529 unsigned long ptex, 530 unsigned long avpn) 531{ 532 return plpar_hcall_norets(H_PROTECT, flags, ptex, avpn); 533} 534 535static __inline__ void set_pp_bit(unsigned long pp, HPTE *addr) 536{ 537 unsigned long old; 538 unsigned long *p = &addr->dw1.dword1; 539 540 __asm__ __volatile__( 541 "1: ldarx %0,0,%3\n\ 542 rldimi %0,%2,0,62\n\ 543 stdcx. %0,0,%3\n\ 544 bne 1b" 545 : "=&r" (old), "=m" (*p) 546 : "r" (pp), "r" (p), "m" (*p) 547 : "cc"); 548} 549 550/* 551 * Functions used to retrieve word 0 of a given page table entry. 552 * 553 * Input : slot : PTE index within the page table of the entry to retrieve 554 * Output: Contents of word 0 of the specified entry 555 */ 556static unsigned long rpa_lpar_hpte_getword0(unsigned long slot) 557{ 558 unsigned long dword0; 559 unsigned long lpar_rc; 560 unsigned long dummy_word1; 561 unsigned long flags; 562 563 /* Read 1 pte at a time */ 564 /* Do not need RPN to logical page translation */ 565 /* No cross CEC PFT access */ 566 flags = 0; 567 568 lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1); 569 570 if (lpar_rc != H_Success) 571 panic("Error on pte read in get_hpte0 rc = %lx\n", lpar_rc); 572 573 return dword0; 574} 575 576unsigned long iSeries_hpte_getword0(unsigned long slot) 577{ 578 unsigned long dword0; 579 580 HPTE hpte; 581 HvCallHpt_get(&hpte, slot); 582 dword0 = hpte.dw0.dword0; 583 584 return dword0; 585} 586 587/* 588 * Functions used to find the PTE for a particular virtual address. 589 * Only used during boot when bolting pages. 590 * 591 * Input : vpn : virtual page number 592 * Output: PTE index within the page table of the entry 593 * -1 on failure 594 */ 595static long hpte_find(unsigned long vpn) 596{ 597 HPTE *hptep; 598 unsigned long hash; 599 unsigned long i, j; 600 long slot; 601 Hpte_dword0 dw0; 602 603 hash = hpt_hash(vpn, 0); 604 605 for (j = 0; j < 2; j++) { 606 slot = (hash & htab_data.htab_hash_mask) * HPTES_PER_GROUP; 607 for (i = 0; i < HPTES_PER_GROUP; i++) { 608 hptep = htab_data.htab + slot; 609 dw0 = hptep->dw0.dw0; 610 611 if ((dw0.avpn == (vpn >> 11)) && dw0.v && 612 (dw0.h == j)) { 613 /* HPTE matches */ 614 if (j) 615 slot = -slot; 616 return slot; 617 } 618 ++slot; 619 } 620 hash = ~hash; 621 } 622 623 return -1; 624} 625 626static long rpa_lpar_hpte_find(unsigned long vpn) 627{ 628 unsigned long hash; 629 unsigned long i, j; 630 long slot; 631 union { 632 unsigned long dword0; 633 Hpte_dword0 dw0; 634 } hpte_dw0; 635 Hpte_dword0 dw0; 636 637 hash = hpt_hash(vpn, 0); 638 639 for (j = 0; j < 2; j++) { 640 slot = (hash & htab_data.htab_hash_mask) * HPTES_PER_GROUP; 641 for (i = 0; i < HPTES_PER_GROUP; i++) { 642 hpte_dw0.dword0 = rpa_lpar_hpte_getword0(slot); 643 dw0 = hpte_dw0.dw0; 644 645 if ((dw0.avpn == (vpn >> 11)) && dw0.v && 646 (dw0.h == j)) { 647 /* HPTE matches */ 648 if (j) 649 slot = -slot; 650 return slot; 651 } 652 ++slot; 653 } 654 hash = ~hash; 655 } 656 657 return -1; 658} 659 660static long iSeries_hpte_find(unsigned long vpn) 661{ 662 HPTE hpte; 663 long slot; 664 665 /* 666 * The HvCallHpt_findValid interface is as follows: 667 * 0xffffffffffffffff : No entry found. 668 * 0x00000000xxxxxxxx : Entry found in primary group, slot x 669 * 0x80000000xxxxxxxx : Entry found in secondary group, slot x 670 */ 671 slot = HvCallHpt_findValid(&hpte, vpn); 672 if (hpte.dw0.dw0.v) { 673 if (slot < 0) { 674 slot &= 0x7fffffffffffffff; 675 slot = -slot; 676 } 677 } else { 678 slot = -1; 679 } 680 681 return slot; 682} 683 684/* 685 * Functions used to invalidate a page table entry from the page table 686 * and tlb. 687 * 688 * Input : slot : PTE index within the page table of the entry to invalidated 689 * va : Virtual address of the entry being invalidated 690 * large : 1 = large page (16M) 691 * local : 1 = Use tlbiel to only invalidate the local tlb 692 */ 693static void hpte_invalidate(unsigned long slot, 694 unsigned long secondary, 695 unsigned long va, 696 int large, int local) 697{ 698 HPTE *hptep = htab_data.htab + slot; 699 Hpte_dword0 dw0; 700 unsigned long vpn, avpn; 701 unsigned long flags; 702 703 if (large) 704 vpn = va >> LARGE_PAGE_SHIFT; 705 else 706 vpn = va >> PAGE_SHIFT; 707 708 avpn = vpn >> 11; 709 710 dw0 = hptep->dw0.dw0; 711 712 /* 713 * Do not remove bolted entries. Alternatively, we could check 714 * the AVPN, hash group, and valid bits. By doing it this way, 715 * it is common with the pSeries LPAR optimal path. 716 */ 717 if (dw0.bolted) return; 718 719 /* Invalidate the hpte. */ 720 hptep->dw0.dword0 = 0; 721 722 /* Invalidate the tlb */ 723 spin_lock_irqsave(&pSeries_tlbie_lock, flags); 724 _tlbie(va, large); 725 spin_unlock_irqrestore(&pSeries_tlbie_lock, flags); 726} 727 728static void rpa_lpar_hpte_invalidate(unsigned long slot, 729 unsigned long secondary, 730 unsigned long va, 731 int large, int local) 732{ 733 unsigned long lpar_rc; 734 unsigned long dummy1, dummy2; 735 736 /* 737 * Don't remove a bolted entry. This case can occur when we bolt 738 * pages dynamically after initial boot. 739 */ 740 lpar_rc = plpar_pte_remove(H_ANDCOND, slot, (0x1UL << 4), 741 &dummy1, &dummy2); 742 743 if (lpar_rc != H_Success) 744 panic("Bad return code from invalidate rc = %lx\n", lpar_rc); 745} 746 747static void iSeries_hpte_invalidate(unsigned long slot, 748 unsigned long secondary, 749 unsigned long va, 750 int large, int local) 751{ 752 HPTE lhpte; 753 unsigned long vpn, avpn; 754 755 if (large) 756 vpn = va >> LARGE_PAGE_SHIFT; 757 else 758 vpn = va >> PAGE_SHIFT; 759 760 avpn = vpn >> 11; 761 762 lhpte.dw0.dword0 = iSeries_hpte_getword0(slot); 763 764 if ((lhpte.dw0.dw0.avpn == avpn) && 765 (lhpte.dw0.dw0.v) && 766 (lhpte.dw0.dw0.h == secondary)) { 767 HvCallHpt_invalidateSetSwBitsGet(slot, 0, 0); 768 } 769} 770 771/* 772 * Functions used to update page protection bits. 773 * 774 * Input : slot : PTE index within the page table of the entry to update 775 * newpp : new page protection bits 776 * va : Virtual address of the entry being updated 777 * large : 1 = large page (16M) 778 * Output: 0 on success, -1 on failure 779 */ 780static long hpte_updatepp(unsigned long slot, 781 unsigned long secondary, 782 unsigned long newpp, 783 unsigned long va, int large) 784{ 785 HPTE *hptep = htab_data.htab + slot; 786 Hpte_dword0 dw0; 787 Hpte_dword1 dw1; 788 unsigned long vpn, avpn; 789 unsigned long flags; 790 791 if (large) 792 vpn = va >> LARGE_PAGE_SHIFT; 793 else 794 vpn = va >> PAGE_SHIFT; 795 796 avpn = vpn >> 11; 797 798 dw0 = hptep->dw0.dw0; 799 if ((dw0.avpn == avpn) && 800 (dw0.v) && (dw0.h == secondary)) { 801 /* Turn off valid bit in HPTE */ 802 dw0.v = 0; 803 hptep->dw0.dw0 = dw0; 804 805 /* Ensure it is out of the tlb too */ 806 spin_lock_irqsave(&pSeries_tlbie_lock, flags); 807 _tlbie(va, large); 808 spin_unlock_irqrestore(&pSeries_tlbie_lock, flags); 809 810 /* Insert the new pp bits into the HPTE */ 811 dw1 = hptep->dw1.dw1; 812 dw1.pp = newpp; 813 hptep->dw1.dw1 = dw1; 814 815 /* Ensure it is visible before validating */ 816 __asm__ __volatile__ ("eieio" : : : "memory"); 817 818 /* Turn the valid bit back on in HPTE */ 819 dw0.v = 1; 820 hptep->dw0.dw0 = dw0; 821 822 __asm__ __volatile__ ("ptesync" : : : "memory"); 823 824 return 0; 825 } 826 827 return -1; 828} 829 830static long rpa_lpar_hpte_updatepp(unsigned long slot, 831 unsigned long secondary, 832 unsigned long newpp, 833 unsigned long va, int large) 834{ 835 unsigned long lpar_rc; 836 unsigned long flags = (newpp & 7); 837 unsigned long avpn = va >> 23; 838 HPTE hpte; 839 840 lpar_rc = plpar_pte_read(0, slot, &hpte.dw0.dword0, &hpte.dw1.dword1); 841 842 if ((hpte.dw0.dw0.avpn == avpn) && 843 (hpte.dw0.dw0.v) && 844 (hpte.dw0.dw0.h == secondary)) { 845 lpar_rc = plpar_pte_protect(flags, slot, 0); 846 if (lpar_rc != H_Success) 847 panic("bad return code from pte protect rc = %lx\n", 848 lpar_rc); 849 return 0; 850 } 851 852 return -1; 853} 854 855static long iSeries_hpte_updatepp(unsigned long slot, 856 unsigned long secondary, 857 unsigned long newpp, 858 unsigned long va, int large) 859{ 860 unsigned long vpn, avpn; 861 HPTE hpte; 862 863 if (large) 864 vpn = va >> LARGE_PAGE_SHIFT; 865 else 866 vpn = va >> PAGE_SHIFT; 867 868 avpn = vpn >> 11; 869 870 HvCallHpt_get(&hpte, slot); 871 if ((hpte.dw0.dw0.avpn == avpn) && 872 (hpte.dw0.dw0.v) && 873 (hpte.dw0.dw0.h == secondary)) { 874 HvCallHpt_setPp(slot, newpp); 875 return 0; 876 } 877 return -1; 878} 879 880/* 881 * Functions used to update the page protection bits. Intended to be used 882 * to create guard pages for kernel data structures on pages which are bolted 883 * in the HPT. Assumes pages being operated on will not be stolen. 884 * Does not work on large pages. No need to lock here because we are the 885 * only user. 886 * 887 * Input : newpp : page protection flags 888 * ea : effective kernel address to bolt. 889 */ 890static void hpte_updateboltedpp(unsigned long newpp, unsigned long ea) 891{ 892 unsigned long vsid, va, vpn, flags; 893 long slot; 894 HPTE *hptep; 895 896 vsid = get_kernel_vsid(ea); 897 va = (vsid << 28) | (ea & 0x0fffffff); 898 vpn = va >> PAGE_SHIFT; 899 900 slot = hpte_find(vpn); 901 if (slot == -1) 902 panic("could not find page to bolt\n"); 903 hptep = htab_data.htab + slot; 904 905 set_pp_bit(newpp, hptep); 906 907 /* Ensure it is out of the tlb too */ 908 spin_lock_irqsave(&pSeries_tlbie_lock, flags); 909 _tlbie(va, 0); 910 spin_unlock_irqrestore(&pSeries_tlbie_lock, flags); 911} 912 913static void rpa_lpar_hpte_updateboltedpp(unsigned long newpp, unsigned long ea) 914{ 915 unsigned long lpar_rc; 916 unsigned long vsid, va, vpn, flags; 917 long slot; 918 919 vsid = get_kernel_vsid(ea); 920 va = (vsid << 28) | (ea & 0x0fffffff); 921 vpn = va >> PAGE_SHIFT; 922 923 slot = rpa_lpar_hpte_find(vpn); 924 if (slot == -1) 925 panic("updateboltedpp: Could not find page to bolt\n"); 926 927 flags = newpp & 3; 928 lpar_rc = plpar_pte_protect(flags, slot, 0); 929 930 if (lpar_rc != H_Success) 931 panic("Bad return code from pte bolted protect rc = %lx\n", 932 lpar_rc); 933} 934 935void iSeries_hpte_updateboltedpp(unsigned long newpp, unsigned long ea) 936{ 937 unsigned long vsid,va,vpn; 938 long slot; 939 940 vsid = get_kernel_vsid( ea ); 941 va = ( vsid << 28 ) | ( ea & 0x0fffffff ); 942 vpn = va >> PAGE_SHIFT; 943 944 slot = iSeries_hpte_find(vpn); 945 if (slot == -1) 946 panic("updateboltedpp: Could not find page to bolt\n"); 947 948 HvCallHpt_setPp(slot, newpp); 949} 950 951/* 952 * Functions used to insert new hardware page table entries. 953 * Will castout non-bolted entries as necessary using a random 954 * algorithm. 955 * 956 * Input : vpn : virtual page number 957 * prpn : real page number in absolute space 958 * hpteflags: page protection flags 959 * bolted : 1 = bolt the page 960 * large : 1 = large page (16M) 961 * Output: hsss, where h = hash group, sss = slot within that group 962 */ 963static long hpte_insert(unsigned long vpn, unsigned long prpn, 964 unsigned long hpteflags, int bolted, int large) 965{ 966 HPTE *hptep; 967 Hpte_dword0 dw0; 968 HPTE lhpte; 969 int i, secondary; 970 unsigned long hash = hpt_hash(vpn, 0); 971 unsigned long avpn = vpn >> 11; 972 unsigned long arpn = physRpn_to_absRpn(prpn); 973 unsigned long hpte_group; 974 975repeat: 976 secondary = 0; 977 hpte_group = ((hash & htab_data.htab_hash_mask) * 978 HPTES_PER_GROUP) & ~0x7UL; 979 hptep = htab_data.htab + hpte_group; 980 981 for (i = 0; i < HPTES_PER_GROUP; i++) { 982 dw0 = hptep->dw0.dw0; 983 if (!dw0.v) { 984 /* retry with lock held */ 985 dw0 = hptep->dw0.dw0; 986 if (!dw0.v) 987 break; 988 } 989 hptep++; 990 } 991 992 if (i == HPTES_PER_GROUP) { 993 secondary = 1; 994 hpte_group = ((~hash & htab_data.htab_hash_mask) * 995 HPTES_PER_GROUP) & ~0x7UL; 996 hptep = htab_data.htab + hpte_group; 997 998 for (i = 0; i < HPTES_PER_GROUP; i++) { 999 dw0 = hptep->dw0.dw0; 1000 if (!dw0.v) { 1001 /* retry with lock held */ 1002 dw0 = hptep->dw0.dw0; 1003 if (!dw0.v) 1004 break; 1005 } 1006 hptep++; 1007 } 1008 if (i == HPTES_PER_GROUP) { 1009 if (mftb() & 0x1) 1010 hpte_group=((hash & htab_data.htab_hash_mask)* 1011 HPTES_PER_GROUP) & ~0x7UL; 1012 1013 hpte_remove(hpte_group); 1014 goto repeat; 1015 } 1016 } 1017 1018 lhpte.dw1.dword1 = 0; 1019 lhpte.dw1.dw1.rpn = arpn; 1020 lhpte.dw1.flags.flags = hpteflags; 1021 1022 lhpte.dw0.dword0 = 0; 1023 lhpte.dw0.dw0.avpn = avpn; 1024 lhpte.dw0.dw0.h = secondary; 1025 lhpte.dw0.dw0.bolted = bolted; 1026 lhpte.dw0.dw0.v = 1; 1027 1028 if (large) lhpte.dw0.dw0.l = 1; 1029 1030 hptep->dw1.dword1 = lhpte.dw1.dword1; 1031 1032 /* Guarantee the second dword is visible before the valid bit */ 1033 __asm__ __volatile__ ("eieio" : : : "memory"); 1034 1035 /* 1036 * Now set the first dword including the valid bit 1037 * NOTE: this also unlocks the hpte 1038 */ 1039 hptep->dw0.dword0 = lhpte.dw0.dword0; 1040 1041 __asm__ __volatile__ ("ptesync" : : : "memory"); 1042 1043 return ((secondary << 3) | i); 1044} 1045 1046static long rpa_lpar_hpte_insert(unsigned long vpn, unsigned long prpn, 1047 unsigned long hpteflags, 1048 int bolted, int large) 1049{ 1050 unsigned long lpar_rc; 1051 unsigned long flags; 1052 unsigned long slot; 1053 HPTE lhpte; 1054 int secondary; 1055 unsigned long hash = hpt_hash(vpn, 0); 1056 unsigned long avpn = vpn >> 11; 1057 unsigned long arpn = physRpn_to_absRpn(prpn); 1058 unsigned long hpte_group; 1059 1060 /* Fill in the local HPTE with absolute rpn, avpn and flags */ 1061 lhpte.dw1.dword1 = 0; 1062 lhpte.dw1.dw1.rpn = arpn; 1063 lhpte.dw1.flags.flags = hpteflags; 1064 1065 lhpte.dw0.dword0 = 0; 1066 lhpte.dw0.dw0.avpn = avpn; 1067 lhpte.dw0.dw0.bolted = bolted; 1068 lhpte.dw0.dw0.v = 1; 1069 1070 if (large) lhpte.dw0.dw0.l = 1; 1071 1072 /* Now fill in the actual HPTE */ 1073 /* Set CEC cookie to 0 */ 1074 /* Large page = 0 */ 1075 /* Zero page = 0 */ 1076 /* I-cache Invalidate = 0 */ 1077 /* I-cache synchronize = 0 */ 1078 /* Exact = 0 */ 1079 flags = 0; 1080 1081 /* -- Because at one point we hit a case where non cachable 1082 * pages where marked coherent & this is rejected by the HV. 1083 * Perhaps it is no longer an issue ... DRENG. 1084 */ 1085 if (hpteflags & (_PAGE_GUARDED|_PAGE_NO_CACHE)) 1086 lhpte.dw1.flags.flags &= ~_PAGE_COHERENT; 1087 1088repeat: 1089 secondary = 0; 1090 lhpte.dw0.dw0.h = secondary; 1091 hpte_group = ((hash & htab_data.htab_hash_mask) * 1092 HPTES_PER_GROUP) & ~0x7UL; 1093 1094 __asm__ __volatile__ ( 1095 H_ENTER_r3 1096 "mr 4, %2\n" 1097 "mr 5, %3\n" 1098 "mr 6, %4\n" 1099 "mr 7, %5\n" 1100 HSC 1101 "mr %0, 3\n" 1102 "mr %1, 4\n" 1103 : "=r" (lpar_rc), "=r" (slot) 1104 : "r" (flags), "r" (hpte_group), "r" (lhpte.dw0.dword0), 1105 "r" (lhpte.dw1.dword1) 1106 : "r0", "r3", "r4", "r5", "r6", "r7", 1107 "r8", "r9", "r10", "r11", "r12", "cc"); 1108 1109 if (lpar_rc == H_PTEG_Full) { 1110 secondary = 1; 1111 lhpte.dw0.dw0.h = secondary; 1112 hpte_group = ((~hash & htab_data.htab_hash_mask) * 1113 HPTES_PER_GROUP) & ~0x7UL; 1114 1115 __asm__ __volatile__ ( 1116 H_ENTER_r3 1117 "mr 4, %2\n" 1118 "mr 5, %3\n" 1119 "mr 6, %4\n" 1120 "mr 7, %5\n" 1121 HSC 1122 "mr %0, 3\n" 1123 "mr %1, 4\n" 1124 : "=r" (lpar_rc), "=r" (slot) 1125 : "r" (flags), "r" (hpte_group), "r" (lhpte.dw0.dword0), 1126 "r" (lhpte.dw1.dword1) 1127 : "r0", "r3", "r4", "r5", "r6", "r7", 1128 "r8", "r9", "r10", "r11", "r12", "cc"); 1129 if (lpar_rc == H_PTEG_Full) { 1130 if (mftb() & 0x1) 1131 hpte_group=((hash & htab_data.htab_hash_mask)* 1132 HPTES_PER_GROUP) & ~0x7UL; 1133 1134 rpa_lpar_hpte_remove(hpte_group); 1135 goto repeat; 1136 } 1137 } 1138 1139 if (lpar_rc != H_Success) 1140 panic("Bad return code from pte enter rc = %lx\n", lpar_rc); 1141 1142 return ((secondary << 3) | (slot & 0x7)); 1143} 1144 1145static long iSeries_hpte_insert(unsigned long vpn, unsigned long prpn, 1146 unsigned long hpteflags, 1147 int bolted, int large) 1148{ 1149 HPTE lhpte; 1150 unsigned long hash, hpte_group; 1151 unsigned long avpn = vpn >> 11; 1152 unsigned long arpn = physRpn_to_absRpn( prpn ); 1153 int secondary = 0; 1154 long slot; 1155 1156 hash = hpt_hash(vpn, 0); 1157 1158repeat: 1159 slot = HvCallHpt_findValid(&lhpte, vpn); 1160 if (lhpte.dw0.dw0.v) { 1161 panic("select_hpte_slot found entry already valid\n"); 1162 } 1163 1164 if (slot == -1) { /* No available entry found in either group */ 1165 if (mftb() & 0x1) { 1166 hpte_group=((hash & htab_data.htab_hash_mask)* 1167 HPTES_PER_GROUP) & ~0x7UL; 1168 } else { 1169 hpte_group=((~hash & htab_data.htab_hash_mask)* 1170 HPTES_PER_GROUP) & ~0x7UL; 1171 } 1172 1173 hash = hpt_hash(vpn, 0); 1174 iSeries_hpte_remove(hpte_group); 1175 goto repeat; 1176 } else if (slot < 0) { 1177 slot &= 0x7fffffffffffffff; 1178 secondary = 1; 1179 } 1180 1181 /* Create the HPTE */ 1182 lhpte.dw1.dword1 = 0; 1183 lhpte.dw1.dw1.rpn = arpn; 1184 lhpte.dw1.flags.flags = hpteflags; 1185 1186 lhpte.dw0.dword0 = 0; 1187 lhpte.dw0.dw0.avpn = avpn; 1188 lhpte.dw0.dw0.h = secondary; 1189 lhpte.dw0.dw0.bolted = bolted; 1190 lhpte.dw0.dw0.v = 1; 1191 1192 /* Now fill in the actual HPTE */ 1193 HvCallHpt_addValidate(slot, secondary, (HPTE *)&lhpte); 1194 return ((secondary << 3) | (slot & 0x7)); 1195} 1196 1197/* 1198 * Functions used to remove hardware page table entries. 1199 * 1200 * Input : hpte_group: PTE index of the first entry in a group 1201 * Output: offset within the group of the entry removed or 1202 * -1 on failure 1203 */ 1204static long hpte_remove(unsigned long hpte_group) 1205{ 1206 HPTE *hptep; 1207 Hpte_dword0 dw0; 1208 int i; 1209 int slot_offset; 1210 unsigned long vsid, group, pi, pi_high; 1211 unsigned long slot; 1212 unsigned long flags; 1213 int large; 1214 unsigned long va; 1215 1216 /* pick a random slot to start at */ 1217 slot_offset = mftb() & 0x7; 1218 1219 for (i = 0; i < HPTES_PER_GROUP; i++) { 1220 hptep = htab_data.htab + hpte_group + slot_offset; 1221 dw0 = hptep->dw0.dw0; 1222 1223 if (dw0.v && !dw0.bolted) { 1224 /* retry with lock held */ 1225 dw0 = hptep->dw0.dw0; 1226 if (dw0.v && !dw0.bolted) 1227 break; 1228 } 1229 1230 slot_offset++; 1231 slot_offset &= 0x7; 1232 } 1233 1234 if (i == HPTES_PER_GROUP) 1235 return -1; 1236 1237 large = dw0.l; 1238 1239 /* Invalidate the hpte. NOTE: this also unlocks it */ 1240 hptep->dw0.dword0 = 0; 1241 1242 /* Invalidate the tlb */ 1243 vsid = dw0.avpn >> 5; 1244 slot = hptep - htab_data.htab; 1245 group = slot >> 3; 1246 if (dw0.h) 1247 group = ~group; 1248 pi = (vsid ^ group) & 0x7ff; 1249 pi_high = (dw0.avpn & 0x1f) << 11; 1250 pi |= pi_high; 1251 1252 if (large) 1253 va = pi << LARGE_PAGE_SHIFT; 1254 else 1255 va = pi << PAGE_SHIFT; 1256 1257 spin_lock_irqsave(&pSeries_tlbie_lock, flags); 1258 _tlbie(va, large); 1259 spin_unlock_irqrestore(&pSeries_tlbie_lock, flags); 1260 1261 return i; 1262} 1263 1264static long rpa_lpar_hpte_remove(unsigned long hpte_group) 1265{ 1266 unsigned long slot_offset; 1267 unsigned long lpar_rc; 1268 int i; 1269 unsigned long dummy1, dummy2; 1270 1271 /* pick a random slot to start at */ 1272 slot_offset = mftb() & 0x7; 1273 1274 for (i = 0; i < HPTES_PER_GROUP; i++) { 1275 1276 /* Don't remove a bolted entry */ 1277 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset, 1278 (0x1UL << 4), &dummy1, &dummy2); 1279 1280 if (lpar_rc == H_Success) 1281 return i; 1282 1283 if (lpar_rc != H_Not_Found) 1284 panic("Bad return code from pte remove rc = %lx\n", 1285 lpar_rc); 1286 1287 slot_offset++; 1288 slot_offset &= 0x7; 1289 } 1290 1291 return -1; 1292} 1293 1294static long iSeries_hpte_remove(unsigned long hpte_group) 1295{ 1296 unsigned long slot_offset; 1297 int i; 1298 HPTE lhpte; 1299 1300 /* Pick a random slot to start at */ 1301 slot_offset = mftb() & 0x7; 1302 1303 for (i = 0; i < HPTES_PER_GROUP; i++) { 1304 lhpte.dw0.dword0 = 1305 iSeries_hpte_getword0(hpte_group + slot_offset); 1306 1307 if (!lhpte.dw0.dw0.bolted) { 1308 HvCallHpt_invalidateSetSwBitsGet(hpte_group + 1309 slot_offset, 0, 0); 1310 return i; 1311 } 1312 1313 slot_offset++; 1314 slot_offset &= 0x7; 1315 } 1316 1317 return -1; 1318} 1319 1320void hpte_init_pSeries(void) 1321{ 1322 ppc_md.hpte_invalidate = hpte_invalidate; 1323 ppc_md.hpte_updatepp = hpte_updatepp; 1324 ppc_md.hpte_updateboltedpp = hpte_updateboltedpp; 1325 ppc_md.hpte_insert = hpte_insert; 1326 ppc_md.hpte_remove = hpte_remove; 1327} 1328 1329void pSeries_lpar_mm_init(void) 1330{ 1331 ppc_md.hpte_invalidate = rpa_lpar_hpte_invalidate; 1332 ppc_md.hpte_updatepp = rpa_lpar_hpte_updatepp; 1333 ppc_md.hpte_updateboltedpp = rpa_lpar_hpte_updateboltedpp; 1334 ppc_md.hpte_insert = rpa_lpar_hpte_insert; 1335 ppc_md.hpte_remove = rpa_lpar_hpte_remove; 1336} 1337 1338void hpte_init_iSeries(void) 1339{ 1340 ppc_md.hpte_invalidate = iSeries_hpte_invalidate; 1341 ppc_md.hpte_updatepp = iSeries_hpte_updatepp; 1342 ppc_md.hpte_updateboltedpp = iSeries_hpte_updateboltedpp; 1343 ppc_md.hpte_insert = iSeries_hpte_insert; 1344 ppc_md.hpte_remove = iSeries_hpte_remove; 1345} 1346 1347