arch/powerpc/include/asm/book3s/64/mmu-hash.h

   1 #ifndef _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_
   2 #define _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_
   3 /*
   4  * PowerPC64 memory management structures
   5  *
   6  * Dave Engebretsen & Mike Corrigan <{engebret|mikejc}@us.ibm.com>
   7  *   PPC64 rework.
   8  *
   9  * This program is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU General Public License
  11  * as published by the Free Software Foundation; either version
  12  * 2 of the License, or (at your option) any later version.
  13  */
  14
  15 #include <asm/asm-compat.h>
  16 #include <asm/page.h>
  17 #include <asm/bug.h>
  18
  19 /*
  20  * This is necessary to get the definition of PGTABLE_RANGE which we
  21  * need for various slices related matters. Note that this isn't the
  22  * complete pgtable.h but only a portion of it.
  23  */
  24 #include <asm/book3s/64/pgtable.h>
  25 #include <asm/bug.h>
  26 #include <asm/processor.h>
  27 #include <asm/cpu_has_feature.h>
  28
  29 /*
  30  * SLB
  31  */
  32
  33 #define SLB_NUM_BOLTED          3
  34 #define SLB_CACHE_ENTRIES       8
  35 #define SLB_MIN_SIZE            32
  36
  37 /* Bits in the SLB ESID word */
  38 #define SLB_ESID_V              ASM_CONST(0x0000000008000000) /* valid */
  39
  40 /* Bits in the SLB VSID word */
  41 #define SLB_VSID_SHIFT          12
  42 #define SLB_VSID_SHIFT_256M     SLB_VSID_SHIFT
  43 #define SLB_VSID_SHIFT_1T       24
  44 #define SLB_VSID_SSIZE_SHIFT    62
  45 #define SLB_VSID_B              ASM_CONST(0xc000000000000000)
  46 #define SLB_VSID_B_256M         ASM_CONST(0x0000000000000000)
  47 #define SLB_VSID_B_1T           ASM_CONST(0x4000000000000000)
  48 #define SLB_VSID_KS             ASM_CONST(0x0000000000000800)
  49 #define SLB_VSID_KP             ASM_CONST(0x0000000000000400)
  50 #define SLB_VSID_N              ASM_CONST(0x0000000000000200) /* no-execute */
  51 #define SLB_VSID_L              ASM_CONST(0x0000000000000100)
  52 #define SLB_VSID_C              ASM_CONST(0x0000000000000080) /* class */
  53 #define SLB_VSID_LP             ASM_CONST(0x0000000000000030)
  54 #define SLB_VSID_LP_00          ASM_CONST(0x0000000000000000)
  55 #define SLB_VSID_LP_01          ASM_CONST(0x0000000000000010)
  56 #define SLB_VSID_LP_10          ASM_CONST(0x0000000000000020)
  57 #define SLB_VSID_LP_11          ASM_CONST(0x0000000000000030)
  58 #define SLB_VSID_LLP            (SLB_VSID_L|SLB_VSID_LP)
  59
  60 #define SLB_VSID_KERNEL         (SLB_VSID_KP)
  61 #define SLB_VSID_USER           (SLB_VSID_KP|SLB_VSID_KS|SLB_VSID_C)
  62
  63 #define SLBIE_C                 (0x08000000)
  64 #define SLBIE_SSIZE_SHIFT       25
  65
  66 /*
  67  * Hash table
  68  */
  69
  70 #define HPTES_PER_GROUP 8
  71
  72 #define HPTE_V_SSIZE_SHIFT      62
  73 #define HPTE_V_AVPN_SHIFT       7
  74 #define HPTE_V_COMMON_BITS      ASM_CONST(0x000fffffffffffff)
  75 #define HPTE_V_AVPN             ASM_CONST(0x3fffffffffffff80)
  76 #define HPTE_V_AVPN_3_0         ASM_CONST(0x000fffffffffff80)
  77 #define HPTE_V_AVPN_VAL(x)      (((x) & HPTE_V_AVPN) >> HPTE_V_AVPN_SHIFT)
  78 #define HPTE_V_COMPARE(x,y)     (!(((x) ^ (y)) & 0xffffffffffffff80UL))
  79 #define HPTE_V_BOLTED           ASM_CONST(0x0000000000000010)
  80 #define HPTE_V_LOCK             ASM_CONST(0x0000000000000008)
  81 #define HPTE_V_LARGE            ASM_CONST(0x0000000000000004)
  82 #define HPTE_V_SECONDARY        ASM_CONST(0x0000000000000002)
  83 #define HPTE_V_VALID            ASM_CONST(0x0000000000000001)
  84
  85 /*
  86  * ISA 3.0 has a different HPTE format.
  87  */
  88 #define HPTE_R_3_0_SSIZE_SHIFT  58
  89 #define HPTE_R_3_0_SSIZE_MASK   (3ull << HPTE_R_3_0_SSIZE_SHIFT)
  90 #define HPTE_R_PP0              ASM_CONST(0x8000000000000000)
  91 #define HPTE_R_TS               ASM_CONST(0x4000000000000000)
  92 #define HPTE_R_KEY_HI           ASM_CONST(0x3000000000000000)
  93 #define HPTE_R_KEY_BIT0         ASM_CONST(0x2000000000000000)
  94 #define HPTE_R_KEY_BIT1         ASM_CONST(0x1000000000000000)
  95 #define HPTE_R_RPN_SHIFT        12
  96 #define HPTE_R_RPN              ASM_CONST(0x0ffffffffffff000)
  97 #define HPTE_R_RPN_3_0          ASM_CONST(0x01fffffffffff000)
  98 #define HPTE_R_PP               ASM_CONST(0x0000000000000003)
  99 #define HPTE_R_PPP              ASM_CONST(0x8000000000000003)
 100 #define HPTE_R_N                ASM_CONST(0x0000000000000004)
 101 #define HPTE_R_G                ASM_CONST(0x0000000000000008)
 102 #define HPTE_R_M                ASM_CONST(0x0000000000000010)
 103 #define HPTE_R_I                ASM_CONST(0x0000000000000020)
 104 #define HPTE_R_W                ASM_CONST(0x0000000000000040)
 105 #define HPTE_R_WIMG             ASM_CONST(0x0000000000000078)
 106 #define HPTE_R_C                ASM_CONST(0x0000000000000080)
 107 #define HPTE_R_R                ASM_CONST(0x0000000000000100)
 108 #define HPTE_R_KEY_LO           ASM_CONST(0x0000000000000e00)
 109 #define HPTE_R_KEY_BIT2         ASM_CONST(0x0000000000000800)
 110 #define HPTE_R_KEY_BIT3         ASM_CONST(0x0000000000000400)
 111 #define HPTE_R_KEY_BIT4         ASM_CONST(0x0000000000000200)
 112 #define HPTE_R_KEY              (HPTE_R_KEY_LO | HPTE_R_KEY_HI)
 113
 114 #define HPTE_V_1TB_SEG          ASM_CONST(0x4000000000000000)
 115 #define HPTE_V_VRMA_MASK        ASM_CONST(0x4001ffffff000000)
 116
 117 /* Values for PP (assumes Ks=0, Kp=1) */
 118 #define PP_RWXX 0       /* Supervisor read/write, User none */
 119 #define PP_RWRX 1       /* Supervisor read/write, User read */
 120 #define PP_RWRW 2       /* Supervisor read/write, User read/write */
 121 #define PP_RXRX 3       /* Supervisor read,       User read */
 122 #define PP_RXXX (HPTE_R_PP0 | 2)        /* Supervisor read, user none */
 123
 124 /* Fields for tlbiel instruction in architecture 2.06 */
 125 #define TLBIEL_INVAL_SEL_MASK   0xc00   /* invalidation selector */
 126 #define  TLBIEL_INVAL_PAGE      0x000   /* invalidate a single page */
 127 #define  TLBIEL_INVAL_SET_LPID  0x800   /* invalidate a set for current LPID */
 128 #define  TLBIEL_INVAL_SET       0xc00   /* invalidate a set for all LPIDs */
 129 #define TLBIEL_INVAL_SET_MASK   0xfff000        /* set number to inval. */
 130 #define TLBIEL_INVAL_SET_SHIFT  12
 131
 132 #define POWER7_TLB_SETS         128     /* # sets in POWER7 TLB */
 133 #define POWER8_TLB_SETS         512     /* # sets in POWER8 TLB */
 134 #define POWER9_TLB_SETS_HASH    256     /* # sets in POWER9 TLB Hash mode */
 135 #define POWER9_TLB_SETS_RADIX   128     /* # sets in POWER9 TLB Radix mode */
 136
 137 #ifndef __ASSEMBLY__
 138
 139 struct mmu_hash_ops {
 140         void            (*hpte_invalidate)(unsigned long slot,
 141                                            unsigned long vpn,
 142                                            int bpsize, int apsize,
 143                                            int ssize, int local);
 144         long            (*hpte_updatepp)(unsigned long slot,
 145                                          unsigned long newpp,
 146                                          unsigned long vpn,
 147                                          int bpsize, int apsize,
 148                                          int ssize, unsigned long flags);
 149         void            (*hpte_updateboltedpp)(unsigned long newpp,
 150                                                unsigned long ea,
 151                                                int psize, int ssize);
 152         long            (*hpte_insert)(unsigned long hpte_group,
 153                                        unsigned long vpn,
 154                                        unsigned long prpn,
 155                                        unsigned long rflags,
 156                                        unsigned long vflags,
 157                                        int psize, int apsize,
 158                                        int ssize);
 159         long            (*hpte_remove)(unsigned long hpte_group);
 160         int             (*hpte_removebolted)(unsigned long ea,
 161                                              int psize, int ssize);
 162         void            (*flush_hash_range)(unsigned long number, int local);
 163         void            (*hugepage_invalidate)(unsigned long vsid,
 164                                                unsigned long addr,
 165                                                unsigned char *hpte_slot_array,
 166                                                int psize, int ssize, int local);
 167         int             (*resize_hpt)(unsigned long shift);
 168         /*
 169          * Special for kexec.
 170          * To be called in real mode with interrupts disabled. No locks are
 171          * taken as such, concurrent access on pre POWER5 hardware could result
 172          * in a deadlock.
 173          * The linear mapping is destroyed as well.
 174          */
 175         void            (*hpte_clear_all)(void);
 176 };
 177 extern struct mmu_hash_ops mmu_hash_ops;
 178
 179 struct hash_pte {
 180         __be64 v;
 181         __be64 r;
 182 };
 183
 184 extern struct hash_pte *htab_address;
 185 extern unsigned long htab_size_bytes;
 186 extern unsigned long htab_hash_mask;
 187
 188
 189 static inline int shift_to_mmu_psize(unsigned int shift)
 190 {
 191         int psize;
 192
 193         for (psize = 0; psize < MMU_PAGE_COUNT; ++psize)
 194                 if (mmu_psize_defs[psize].shift == shift)
 195                         return psize;
 196         return -1;
 197 }
 198
 199 static inline unsigned int mmu_psize_to_shift(unsigned int mmu_psize)
 200 {
 201         if (mmu_psize_defs[mmu_psize].shift)
 202                 return mmu_psize_defs[mmu_psize].shift;
 203         BUG();
 204 }
 205
 206 static inline unsigned long get_sllp_encoding(int psize)
 207 {
 208         unsigned long sllp;
 209
 210         sllp = ((mmu_psize_defs[psize].sllp & SLB_VSID_L) >> 6) |
 211                 ((mmu_psize_defs[psize].sllp & SLB_VSID_LP) >> 4);
 212         return sllp;
 213 }
 214
 215 #endif /* __ASSEMBLY__ */
 216
 217 /*
 218  * Segment sizes.
 219  * These are the values used by hardware in the B field of
 220  * SLB entries and the first dword of MMU hashtable entries.
 221  * The B field is 2 bits; the values 2 and 3 are unused and reserved.
 222  */
 223 #define MMU_SEGSIZE_256M        0
 224 #define MMU_SEGSIZE_1T          1
 225
 226 /*
 227  * encode page number shift.
 228  * in order to fit the 78 bit va in a 64 bit variable we shift the va by
 229  * 12 bits. This enable us to address upto 76 bit va.
 230  * For hpt hash from a va we can ignore the page size bits of va and for
 231  * hpte encoding we ignore up to 23 bits of va. So ignoring lower 12 bits ensure
 232  * we work in all cases including 4k page size.
 233  */
 234 #define VPN_SHIFT       12
 235
 236 /*
 237  * HPTE Large Page (LP) details
 238  */
 239 #define LP_SHIFT        12
 240 #define LP_BITS         8
 241 #define LP_MASK(i)      ((0xFF >> (i)) << LP_SHIFT)
 242
 243 #ifndef __ASSEMBLY__
 244
 245 static inline int slb_vsid_shift(int ssize)
 246 {
 247         if (ssize == MMU_SEGSIZE_256M)
 248                 return SLB_VSID_SHIFT;
 249         return SLB_VSID_SHIFT_1T;
 250 }
 251
 252 static inline int segment_shift(int ssize)
 253 {
 254         if (ssize == MMU_SEGSIZE_256M)
 255                 return SID_SHIFT;
 256         return SID_SHIFT_1T;
 257 }
 258
 259 /*
 260  * This array is indexed by the LP field of the HPTE second dword.
 261  * Since this field may contain some RPN bits, some entries are
 262  * replicated so that we get the same value irrespective of RPN.
 263  * The top 4 bits are the page size index (MMU_PAGE_*) for the
 264  * actual page size, the bottom 4 bits are the base page size.
 265  */
 266 extern u8 hpte_page_sizes[1 << LP_BITS];
 267
 268 static inline unsigned long __hpte_page_size(unsigned long h, unsigned long l,
 269                                              bool is_base_size)
 270 {
 271         unsigned int i, lp;
 272
 273         if (!(h & HPTE_V_LARGE))
 274                 return 1ul << 12;
 275
 276         /* Look at the 8 bit LP value */
 277         lp = (l >> LP_SHIFT) & ((1 << LP_BITS) - 1);
 278         i = hpte_page_sizes[lp];
 279         if (!i)
 280                 return 0;
 281         if (!is_base_size)
 282                 i >>= 4;
 283         return 1ul << mmu_psize_defs[i & 0xf].shift;
 284 }
 285
 286 static inline unsigned long hpte_page_size(unsigned long h, unsigned long l)
 287 {
 288         return __hpte_page_size(h, l, 0);
 289 }
 290
 291 static inline unsigned long hpte_base_page_size(unsigned long h, unsigned long l)
 292 {
 293         return __hpte_page_size(h, l, 1);
 294 }
 295
 296 /*
 297  * The current system page and segment sizes
 298  */
 299 extern int mmu_kernel_ssize;
 300 extern int mmu_highuser_ssize;
 301 extern u16 mmu_slb_size;
 302 extern unsigned long tce_alloc_start, tce_alloc_end;
 303
 304 /*
 305  * If the processor supports 64k normal pages but not 64k cache
 306  * inhibited pages, we have to be prepared to switch processes
 307  * to use 4k pages when they create cache-inhibited mappings.
 308  * If this is the case, mmu_ci_restrictions will be set to 1.
 309  */
 310 extern int mmu_ci_restrictions;
 311
 312 /*
 313  * This computes the AVPN and B fields of the first dword of a HPTE,
 314  * for use when we want to match an existing PTE.  The bottom 7 bits
 315  * of the returned value are zero.
 316  */
 317 static inline unsigned long hpte_encode_avpn(unsigned long vpn, int psize,
 318                                              int ssize)
 319 {
 320         unsigned long v;
 321         /*
 322          * The AVA field omits the low-order 23 bits of the 78 bits VA.
 323          * These bits are not needed in the PTE, because the
 324          * low-order b of these bits are part of the byte offset
 325          * into the virtual page and, if b < 23, the high-order
 326          * 23-b of these bits are always used in selecting the
 327          * PTEGs to be searched
 328          */
 329         v = (vpn >> (23 - VPN_SHIFT)) & ~(mmu_psize_defs[psize].avpnm);
 330         v <<= HPTE_V_AVPN_SHIFT;
 331         v |= ((unsigned long) ssize) << HPTE_V_SSIZE_SHIFT;
 332         return v;
 333 }
 334
 335 /*
 336  * ISA v3.0 defines a new HPTE format, which differs from the old
 337  * format in having smaller AVPN and ARPN fields, and the B field
 338  * in the second dword instead of the first.
 339  */
 340 static inline unsigned long hpte_old_to_new_v(unsigned long v)
 341 {
 342         /* trim AVPN, drop B */
 343         return v & HPTE_V_COMMON_BITS;
 344 }
 345
 346 static inline unsigned long hpte_old_to_new_r(unsigned long v, unsigned long r)
 347 {
 348         /* move B field from 1st to 2nd dword, trim ARPN */
 349         return (r & ~HPTE_R_3_0_SSIZE_MASK) |
 350                 (((v) >> HPTE_V_SSIZE_SHIFT) << HPTE_R_3_0_SSIZE_SHIFT);
 351 }
 352
 353 static inline unsigned long hpte_new_to_old_v(unsigned long v, unsigned long r)
 354 {
 355         /* insert B field */
 356         return (v & HPTE_V_COMMON_BITS) |
 357                 ((r & HPTE_R_3_0_SSIZE_MASK) <<
 358                  (HPTE_V_SSIZE_SHIFT - HPTE_R_3_0_SSIZE_SHIFT));
 359 }
 360
 361 static inline unsigned long hpte_new_to_old_r(unsigned long r)
 362 {
 363         /* clear out B field */
 364         return r & ~HPTE_R_3_0_SSIZE_MASK;
 365 }
 366
 367 /*
 368  * This function sets the AVPN and L fields of the HPTE  appropriately
 369  * using the base page size and actual page size.
 370  */
 371 static inline unsigned long hpte_encode_v(unsigned long vpn, int base_psize,
 372                                           int actual_psize, int ssize)
 373 {
 374         unsigned long v;
 375         v = hpte_encode_avpn(vpn, base_psize, ssize);
 376         if (actual_psize != MMU_PAGE_4K)
 377                 v |= HPTE_V_LARGE;
 378         return v;
 379 }
 380
 381 /*
 382  * This function sets the ARPN, and LP fields of the HPTE appropriately
 383  * for the page size. We assume the pa is already "clean" that is properly
 384  * aligned for the requested page size
 385  */
 386 static inline unsigned long hpte_encode_r(unsigned long pa, int base_psize,
 387                                           int actual_psize)
 388 {
 389         /* A 4K page needs no special encoding */
 390         if (actual_psize == MMU_PAGE_4K)
 391                 return pa & HPTE_R_RPN;
 392         else {
 393                 unsigned int penc = mmu_psize_defs[base_psize].penc[actual_psize];
 394                 unsigned int shift = mmu_psize_defs[actual_psize].shift;
 395                 return (pa & ~((1ul << shift) - 1)) | (penc << LP_SHIFT);
 396         }
 397 }
 398
 399 /*
 400  * Build a VPN_SHIFT bit shifted va given VSID, EA and segment size.
 401  */
 402 static inline unsigned long hpt_vpn(unsigned long ea,
 403                                     unsigned long vsid, int ssize)
 404 {
 405         unsigned long mask;
 406         int s_shift = segment_shift(ssize);
 407
 408         mask = (1ul << (s_shift - VPN_SHIFT)) - 1;
 409         return (vsid << (s_shift - VPN_SHIFT)) | ((ea >> VPN_SHIFT) & mask);
 410 }
 411
 412 /*
 413  * This hashes a virtual address
 414  */
 415 static inline unsigned long hpt_hash(unsigned long vpn,
 416                                      unsigned int shift, int ssize)
 417 {
 418         unsigned long mask;
 419         unsigned long hash, vsid;
 420
 421         /* VPN_SHIFT can be atmost 12 */
 422         if (ssize == MMU_SEGSIZE_256M) {
 423                 mask = (1ul << (SID_SHIFT - VPN_SHIFT)) - 1;
 424                 hash = (vpn >> (SID_SHIFT - VPN_SHIFT)) ^
 425                         ((vpn & mask) >> (shift - VPN_SHIFT));
 426         } else {
 427                 mask = (1ul << (SID_SHIFT_1T - VPN_SHIFT)) - 1;
 428                 vsid = vpn >> (SID_SHIFT_1T - VPN_SHIFT);
 429                 hash = vsid ^ (vsid << 25) ^
 430                         ((vpn & mask) >> (shift - VPN_SHIFT)) ;
 431         }
 432         return hash & 0x7fffffffffUL;
 433 }
 434
 435 #define HPTE_LOCAL_UPDATE       0x1
 436 #define HPTE_NOHPTE_UPDATE      0x2
 437
 438 extern int __hash_page_4K(unsigned long ea, unsigned long access,
 439                           unsigned long vsid, pte_t *ptep, unsigned long trap,
 440                           unsigned long flags, int ssize, int subpage_prot);
 441 extern int __hash_page_64K(unsigned long ea, unsigned long access,
 442                            unsigned long vsid, pte_t *ptep, unsigned long trap,
 443                            unsigned long flags, int ssize);
 444 struct mm_struct;
 445 unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap);
 446 extern int hash_page_mm(struct mm_struct *mm, unsigned long ea,
 447                         unsigned long access, unsigned long trap,
 448                         unsigned long flags);
 449 extern int hash_page(unsigned long ea, unsigned long access, unsigned long trap,
 450                      unsigned long dsisr);
 451 int __hash_page_huge(unsigned long ea, unsigned long access, unsigned long vsid,
 452                      pte_t *ptep, unsigned long trap, unsigned long flags,
 453                      int ssize, unsigned int shift, unsigned int mmu_psize);
 454 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 455 extern int __hash_page_thp(unsigned long ea, unsigned long access,
 456                            unsigned long vsid, pmd_t *pmdp, unsigned long trap,
 457                            unsigned long flags, int ssize, unsigned int psize);
 458 #else
 459 static inline int __hash_page_thp(unsigned long ea, unsigned long access,
 460                                   unsigned long vsid, pmd_t *pmdp,
 461                                   unsigned long trap, unsigned long flags,
 462                                   int ssize, unsigned int psize)
 463 {
 464         BUG();
 465         return -1;
 466 }
 467 #endif
 468 extern void hash_failure_debug(unsigned long ea, unsigned long access,
 469                                unsigned long vsid, unsigned long trap,
 470                                int ssize, int psize, int lpsize,
 471                                unsigned long pte);
 472 extern int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
 473                              unsigned long pstart, unsigned long prot,
 474                              int psize, int ssize);
 475 int htab_remove_mapping(unsigned long vstart, unsigned long vend,
 476                         int psize, int ssize);
 477 extern void pseries_add_gpage(u64 addr, u64 page_size, unsigned long number_of_pages);
 478 extern void demote_segment_4k(struct mm_struct *mm, unsigned long addr);
 479
 480 #ifdef CONFIG_PPC_PSERIES
 481 void hpte_init_pseries(void);
 482 #else
 483 static inline void hpte_init_pseries(void) { }
 484 #endif
 485
 486 extern void hpte_init_native(void);
 487
 488 extern void slb_initialize(void);
 489 extern void slb_flush_and_rebolt(void);
 490
 491 extern void slb_vmalloc_update(void);
 492 extern void slb_set_size(u16 size);
 493 #endif /* __ASSEMBLY__ */
 494
 495 /*
 496  * VSID allocation (256MB segment)
 497  *
 498  * We first generate a 37-bit "proto-VSID". Proto-VSIDs are generated
 499  * from mmu context id and effective segment id of the address.
 500  *
 501  * For user processes max context id is limited to MAX_USER_CONTEXT.
 502
 503  * For kernel space, we use context ids 1-4 to map addresses as below:
 504  * NOTE: each context only support 64TB now.
 505  * 0x00001 -  [ 0xc000000000000000 - 0xc0003fffffffffff ]
 506  * 0x00002 -  [ 0xd000000000000000 - 0xd0003fffffffffff ]
 507  * 0x00003 -  [ 0xe000000000000000 - 0xe0003fffffffffff ]
 508  * 0x00004 -  [ 0xf000000000000000 - 0xf0003fffffffffff ]
 509  *
 510  * The proto-VSIDs are then scrambled into real VSIDs with the
 511  * multiplicative hash:
 512  *
 513  *      VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS
 514  *
 515  * VSID_MULTIPLIER is prime, so in particular it is
 516  * co-prime to VSID_MODULUS, making this a 1:1 scrambling function.
 517  * Because the modulus is 2^n-1 we can compute it efficiently without
 518  * a divide or extra multiply (see below). The scramble function gives
 519  * robust scattering in the hash table (at least based on some initial
 520  * results).
 521  *
 522  * We use VSID 0 to indicate an invalid VSID. The means we can't use context id
 523  * 0, because a context id of 0 and an EA of 0 gives a proto-VSID of 0, which
 524  * will produce a VSID of 0.
 525  *
 526  * We also need to avoid the last segment of the last context, because that
 527  * would give a protovsid of 0x1fffffffff. That will result in a VSID 0
 528  * because of the modulo operation in vsid scramble.
 529  */
 530
 531 /*
 532  * Max Va bits we support as of now is 68 bits. We want 19 bit
 533  * context ID.
 534  * Restrictions:
 535  * GPU has restrictions of not able to access beyond 128TB
 536  * (47 bit effective address). We also cannot do more than 20bit PID.
 537  * For p4 and p5 which can only do 65 bit VA, we restrict our CONTEXT_BITS
 538  * to 16 bits (ie, we can only have 2^16 pids at the same time).
 539  */
 540 #define VA_BITS                 68
 541 #define CONTEXT_BITS            19
 542 #define ESID_BITS               (VA_BITS - (SID_SHIFT + CONTEXT_BITS))
 543 #define ESID_BITS_1T            (VA_BITS - (SID_SHIFT_1T + CONTEXT_BITS))
 544
 545 #define ESID_BITS_MASK          ((1 << ESID_BITS) - 1)
 546 #define ESID_BITS_1T_MASK       ((1 << ESID_BITS_1T) - 1)
 547
 548 /*
 549  * 256MB segment
 550  * The proto-VSID space has 2^(CONTEX_BITS + ESID_BITS) - 1 segments
 551  * available for user + kernel mapping. VSID 0 is reserved as invalid, contexts
 552  * 1-4 are used for kernel mapping. Each segment contains 2^28 bytes. Each
 553  * context maps 2^49 bytes (512TB).
 554  *
 555  * We also need to avoid the last segment of the last context, because that
 556  * would give a protovsid of 0x1fffffffff. That will result in a VSID 0
 557  * because of the modulo operation in vsid scramble.
 558  */
 559 #define MAX_USER_CONTEXT        ((ASM_CONST(1) << CONTEXT_BITS) - 2)
 560 #define MIN_USER_CONTEXT        (5)
 561
 562 /* Would be nice to use KERNEL_REGION_ID here */
 563 #define KERNEL_REGION_CONTEXT_OFFSET    (0xc - 1)
 564
 565 /*
 566  * For platforms that support on 65bit VA we limit the context bits
 567  */
 568 #define MAX_USER_CONTEXT_65BIT_VA ((ASM_CONST(1) << (65 - (SID_SHIFT + ESID_BITS))) - 2)
 569
 570 /*
 571  * This should be computed such that protovosid * vsid_mulitplier
 572  * doesn't overflow 64 bits. The vsid_mutliplier should also be
 573  * co-prime to vsid_modulus. We also need to make sure that number
 574  * of bits in multiplied result (dividend) is less than twice the number of
 575  * protovsid bits for our modulus optmization to work.
 576  *
 577  * The below table shows the current values used.
 578  * |-------+------------+----------------------+------------+-------------------|
 579  * |       | Prime Bits | proto VSID_BITS_65VA | Total Bits | 2* prot VSID_BITS |
 580  * |-------+------------+----------------------+------------+-------------------|
 581  * | 1T    |         24 |                   25 |         49 |                50 |
 582  * |-------+------------+----------------------+------------+-------------------|
 583  * | 256MB |         24 |                   37 |         61 |                74 |
 584  * |-------+------------+----------------------+------------+-------------------|
 585  *
 586  * |-------+------------+----------------------+------------+--------------------|
 587  * |       | Prime Bits | proto VSID_BITS_68VA | Total Bits | 2* proto VSID_BITS |
 588  * |-------+------------+----------------------+------------+--------------------|
 589  * | 1T    |         24 |                   28 |         52 |                 56 |
 590  * |-------+------------+----------------------+------------+--------------------|
 591  * | 256MB |         24 |                   40 |         64 |                 80 |
 592  * |-------+------------+----------------------+------------+--------------------|
 593  *
 594  */
 595 #define VSID_MULTIPLIER_256M    ASM_CONST(12538073)     /* 24-bit prime */
 596 #define VSID_BITS_256M          (VA_BITS - SID_SHIFT)
 597 #define VSID_BITS_65_256M       (65 - SID_SHIFT)
 598 /*
 599  * Modular multiplicative inverse of VSID_MULTIPLIER under modulo VSID_MODULUS
 600  */
 601 #define VSID_MULINV_256M        ASM_CONST(665548017062)
 602
 603 #define VSID_MULTIPLIER_1T      ASM_CONST(12538073)     /* 24-bit prime */
 604 #define VSID_BITS_1T            (VA_BITS - SID_SHIFT_1T)
 605 #define VSID_BITS_65_1T         (65 - SID_SHIFT_1T)
 606 #define VSID_MULINV_1T          ASM_CONST(209034062)
 607
 608 /* 1TB VSID reserved for VRMA */
 609 #define VRMA_VSID       0x1ffffffUL
 610 #define USER_VSID_RANGE (1UL << (ESID_BITS + SID_SHIFT))
 611
 612 /* 4 bits per slice and we have one slice per 1TB */
 613 #define SLICE_ARRAY_SIZE        (H_PGTABLE_RANGE >> 41)
 614 #define TASK_SLICE_ARRAY_SZ(x)  ((x)->context.slb_addr_limit >> 41)
 615
 616 #ifndef __ASSEMBLY__
 617
 618 #ifdef CONFIG_PPC_SUBPAGE_PROT
 619 /*
 620  * For the sub-page protection option, we extend the PGD with one of
 621  * these.  Basically we have a 3-level tree, with the top level being
 622  * the protptrs array.  To optimize speed and memory consumption when
 623  * only addresses < 4GB are being protected, pointers to the first
 624  * four pages of sub-page protection words are stored in the low_prot
 625  * array.
 626  * Each page of sub-page protection words protects 1GB (4 bytes
 627  * protects 64k).  For the 3-level tree, each page of pointers then
 628  * protects 8TB.
 629  */
 630 struct subpage_prot_table {
 631         unsigned long maxaddr;  /* only addresses < this are protected */
 632         unsigned int **protptrs[(TASK_SIZE_USER64 >> 43)];
 633         unsigned int *low_prot[4];
 634 };
 635
 636 #define SBP_L1_BITS             (PAGE_SHIFT - 2)
 637 #define SBP_L2_BITS             (PAGE_SHIFT - 3)
 638 #define SBP_L1_COUNT            (1 << SBP_L1_BITS)
 639 #define SBP_L2_COUNT            (1 << SBP_L2_BITS)
 640 #define SBP_L2_SHIFT            (PAGE_SHIFT + SBP_L1_BITS)
 641 #define SBP_L3_SHIFT            (SBP_L2_SHIFT + SBP_L2_BITS)
 642
 643 extern void subpage_prot_free(struct mm_struct *mm);
 644 extern void subpage_prot_init_new_context(struct mm_struct *mm);
 645 #else
 646 static inline void subpage_prot_free(struct mm_struct *mm) {}
 647 static inline void subpage_prot_init_new_context(struct mm_struct *mm) { }
 648 #endif /* CONFIG_PPC_SUBPAGE_PROT */
 649
 650 #if 0
 651 /*
 652  * The code below is equivalent to this function for arguments
 653  * < 2^VSID_BITS, which is all this should ever be called
 654  * with.  However gcc is not clever enough to compute the
 655  * modulus (2^n-1) without a second multiply.
 656  */
 657 #define vsid_scramble(protovsid, size) \
 658         ((((protovsid) * VSID_MULTIPLIER_##size) % VSID_MODULUS_##size))
 659
 660 /* simplified form avoiding mod operation */
 661 #define vsid_scramble(protovsid, size) \
 662         ({                                                               \
 663                 unsigned long x;                                         \
 664                 x = (protovsid) * VSID_MULTIPLIER_##size;                \
 665                 x = (x >> VSID_BITS_##size) + (x & VSID_MODULUS_##size); \
 666                 (x + ((x+1) >> VSID_BITS_##size)) & VSID_MODULUS_##size; \
 667         })
 668
 669 #else /* 1 */
 670 static inline unsigned long vsid_scramble(unsigned long protovsid,
 671                                   unsigned long vsid_multiplier, int vsid_bits)
 672 {
 673         unsigned long vsid;
 674         unsigned long vsid_modulus = ((1UL << vsid_bits) - 1);
 675         /*
 676          * We have same multipler for both 256 and 1T segements now
 677          */
 678         vsid = protovsid * vsid_multiplier;
 679         vsid = (vsid >> vsid_bits) + (vsid & vsid_modulus);
 680         return (vsid + ((vsid + 1) >> vsid_bits)) & vsid_modulus;
 681 }
 682
 683 #endif /* 1 */
 684
 685 /* Returns the segment size indicator for a user address */
 686 static inline int user_segment_size(unsigned long addr)
 687 {
 688         /* Use 1T segments if possible for addresses >= 1T */
 689         if (addr >= (1UL << SID_SHIFT_1T))
 690                 return mmu_highuser_ssize;
 691         return MMU_SEGSIZE_256M;
 692 }
 693
 694 static inline unsigned long get_vsid(unsigned long context, unsigned long ea,
 695                                      int ssize)
 696 {
 697         unsigned long va_bits = VA_BITS;
 698         unsigned long vsid_bits;
 699         unsigned long protovsid;
 700
 701         /*
 702          * Bad address. We return VSID 0 for that
 703          */
 704         if ((ea & ~REGION_MASK) >= H_PGTABLE_RANGE)
 705                 return 0;
 706
 707         if (!mmu_has_feature(MMU_FTR_68_BIT_VA))
 708                 va_bits = 65;
 709
 710         if (ssize == MMU_SEGSIZE_256M) {
 711                 vsid_bits = va_bits - SID_SHIFT;
 712                 protovsid = (context << ESID_BITS) |
 713                         ((ea >> SID_SHIFT) & ESID_BITS_MASK);
 714                 return vsid_scramble(protovsid, VSID_MULTIPLIER_256M, vsid_bits);
 715         }
 716         /* 1T segment */
 717         vsid_bits = va_bits - SID_SHIFT_1T;
 718         protovsid = (context << ESID_BITS_1T) |
 719                 ((ea >> SID_SHIFT_1T) & ESID_BITS_1T_MASK);
 720         return vsid_scramble(protovsid, VSID_MULTIPLIER_1T, vsid_bits);
 721 }
 722
 723 /*
 724  * This is only valid for addresses >= PAGE_OFFSET
 725  */
 726 static inline unsigned long get_kernel_vsid(unsigned long ea, int ssize)
 727 {
 728         unsigned long context;
 729
 730         if (!is_kernel_addr(ea))
 731                 return 0;
 732
 733         /*
 734          * For kernel space, we use context ids 1-4 to map the address space as
 735          * below:
 736          *
 737          * 0x00001 -  [ 0xc000000000000000 - 0xc0003fffffffffff ]
 738          * 0x00002 -  [ 0xd000000000000000 - 0xd0003fffffffffff ]
 739          * 0x00003 -  [ 0xe000000000000000 - 0xe0003fffffffffff ]
 740          * 0x00004 -  [ 0xf000000000000000 - 0xf0003fffffffffff ]
 741          *
 742          * So we can compute the context from the region (top nibble) by
 743          * subtracting 11, or 0xc - 1.
 744          */
 745         context = (ea >> 60) - KERNEL_REGION_CONTEXT_OFFSET;
 746
 747         return get_vsid(context, ea, ssize);
 748 }
 749
 750 unsigned htab_shift_for_mem_size(unsigned long mem_size);
 751
 752 #endif /* __ASSEMBLY__ */
 753
 754 #endif /* _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_ */