include/asm-powerpc/mmu.h

   1 #ifndef _ASM_POWERPC_MMU_H_
   2 #define _ASM_POWERPC_MMU_H_
   3 #ifdef __KERNEL__
   4
   5 #ifndef CONFIG_PPC64
   6 #include <asm-ppc/mmu.h>
   7 #else
   8
   9 /*
  10  * PowerPC memory management structures
  11  *
  12  * Dave Engebretsen & Mike Corrigan <{engebret|mikejc}@us.ibm.com>
  13  *   PPC64 rework.
  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 <asm/asm-compat.h>
  22 #include <asm/page.h>
  23
  24 /*
  25  * Segment table
  26  */
  27
  28 #define STE_ESID_V      0x80
  29 #define STE_ESID_KS     0x20
  30 #define STE_ESID_KP     0x10
  31 #define STE_ESID_N      0x08
  32
  33 #define STE_VSID_SHIFT  12
  34
  35 /* Location of cpu0's segment table */
  36 #define STAB0_PAGE      0x6
  37 #define STAB0_OFFSET    (STAB0_PAGE << 12)
  38 #define STAB0_PHYS_ADDR (STAB0_OFFSET + PHYSICAL_START)
  39
  40 #ifndef __ASSEMBLY__
  41 extern char initial_stab[];
  42 #endif /* ! __ASSEMBLY */
  43
  44 /*
  45  * SLB
  46  */
  47
  48 #define SLB_NUM_BOLTED          3
  49 #define SLB_CACHE_ENTRIES       8
  50
  51 /* Bits in the SLB ESID word */
  52 #define SLB_ESID_V              ASM_CONST(0x0000000008000000) /* valid */
  53
  54 /* Bits in the SLB VSID word */
  55 #define SLB_VSID_SHIFT          12
  56 #define SLB_VSID_B              ASM_CONST(0xc000000000000000)
  57 #define SLB_VSID_B_256M         ASM_CONST(0x0000000000000000)
  58 #define SLB_VSID_B_1T           ASM_CONST(0x4000000000000000)
  59 #define SLB_VSID_KS             ASM_CONST(0x0000000000000800)
  60 #define SLB_VSID_KP             ASM_CONST(0x0000000000000400)
  61 #define SLB_VSID_N              ASM_CONST(0x0000000000000200) /* no-execute */
  62 #define SLB_VSID_L              ASM_CONST(0x0000000000000100)
  63 #define SLB_VSID_C              ASM_CONST(0x0000000000000080) /* class */
  64 #define SLB_VSID_LP             ASM_CONST(0x0000000000000030)
  65 #define SLB_VSID_LP_00          ASM_CONST(0x0000000000000000)
  66 #define SLB_VSID_LP_01          ASM_CONST(0x0000000000000010)
  67 #define SLB_VSID_LP_10          ASM_CONST(0x0000000000000020)
  68 #define SLB_VSID_LP_11          ASM_CONST(0x0000000000000030)
  69 #define SLB_VSID_LLP            (SLB_VSID_L|SLB_VSID_LP)
  70
  71 #define SLB_VSID_KERNEL         (SLB_VSID_KP)
  72 #define SLB_VSID_USER           (SLB_VSID_KP|SLB_VSID_KS|SLB_VSID_C)
  73
  74 #define SLBIE_C                 (0x08000000)
  75
  76 /*
  77  * Hash table
  78  */
  79
  80 #define HPTES_PER_GROUP 8
  81
  82 #define HPTE_V_AVPN_SHIFT       7
  83 #define HPTE_V_AVPN             ASM_CONST(0xffffffffffffff80)
  84 #define HPTE_V_AVPN_VAL(x)      (((x) & HPTE_V_AVPN) >> HPTE_V_AVPN_SHIFT)
  85 #define HPTE_V_COMPARE(x,y)     (!(((x) ^ (y)) & HPTE_V_AVPN))
  86 #define HPTE_V_BOLTED           ASM_CONST(0x0000000000000010)
  87 #define HPTE_V_LOCK             ASM_CONST(0x0000000000000008)
  88 #define HPTE_V_LARGE            ASM_CONST(0x0000000000000004)
  89 #define HPTE_V_SECONDARY        ASM_CONST(0x0000000000000002)
  90 #define HPTE_V_VALID            ASM_CONST(0x0000000000000001)
  91
  92 #define HPTE_R_PP0              ASM_CONST(0x8000000000000000)
  93 #define HPTE_R_TS               ASM_CONST(0x4000000000000000)
  94 #define HPTE_R_RPN_SHIFT        12
  95 #define HPTE_R_RPN              ASM_CONST(0x3ffffffffffff000)
  96 #define HPTE_R_FLAGS            ASM_CONST(0x00000000000003ff)
  97 #define HPTE_R_PP               ASM_CONST(0x0000000000000003)
  98 #define HPTE_R_N                ASM_CONST(0x0000000000000004)
  99 #define HPTE_R_C                ASM_CONST(0x0000000000000080)
 100 #define HPTE_R_R                ASM_CONST(0x0000000000000100)
 101
 102 /* Values for PP (assumes Ks=0, Kp=1) */
 103 /* pp0 will always be 0 for linux     */
 104 #define PP_RWXX 0       /* Supervisor read/write, User none */
 105 #define PP_RWRX 1       /* Supervisor read/write, User read */
 106 #define PP_RWRW 2       /* Supervisor read/write, User read/write */
 107 #define PP_RXRX 3       /* Supervisor read,       User read */
 108
 109 #ifndef __ASSEMBLY__
 110
 111 typedef struct {
 112         unsigned long v;
 113         unsigned long r;
 114 } hpte_t;
 115
 116 extern hpte_t *htab_address;
 117 extern unsigned long htab_size_bytes;
 118 extern unsigned long htab_hash_mask;
 119
 120 /*
 121  * Page size definition
 122  *
 123  *    shift : is the "PAGE_SHIFT" value for that page size
 124  *    sllp  : is a bit mask with the value of SLB L || LP to be or'ed
 125  *            directly to a slbmte "vsid" value
 126  *    penc  : is the HPTE encoding mask for the "LP" field:
 127  *
 128  */
 129 struct mmu_psize_def
 130 {
 131         unsigned int    shift;  /* number of bits */
 132         unsigned int    penc;   /* HPTE encoding */
 133         unsigned int    tlbiel; /* tlbiel supported for that page size */
 134         unsigned long   avpnm;  /* bits to mask out in AVPN in the HPTE */
 135         unsigned long   sllp;   /* SLB L||LP (exact mask to use in slbmte) */
 136 };
 137
 138 #endif /* __ASSEMBLY__ */
 139
 140 /*
 141  * The kernel use the constants below to index in the page sizes array.
 142  * The use of fixed constants for this purpose is better for performances
 143  * of the low level hash refill handlers.
 144  *
 145  * A non supported page size has a "shift" field set to 0
 146  *
 147  * Any new page size being implemented can get a new entry in here. Whether
 148  * the kernel will use it or not is a different matter though. The actual page
 149  * size used by hugetlbfs is not defined here and may be made variable
 150  */
 151
 152 #define MMU_PAGE_4K             0       /* 4K */
 153 #define MMU_PAGE_64K            1       /* 64K */
 154 #define MMU_PAGE_64K_AP         2       /* 64K Admixed (in a 4K segment) */
 155 #define MMU_PAGE_1M             3       /* 1M */
 156 #define MMU_PAGE_16M            4       /* 16M */
 157 #define MMU_PAGE_16G            5       /* 16G */
 158 #define MMU_PAGE_COUNT          6
 159
 160 #ifndef __ASSEMBLY__
 161
 162 /*
 163  * The current system page sizes
 164  */
 165 extern struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT];
 166 extern int mmu_linear_psize;
 167 extern int mmu_virtual_psize;
 168 extern int mmu_vmalloc_psize;
 169 extern int mmu_io_psize;
 170
 171 /*
 172  * If the processor supports 64k normal pages but not 64k cache
 173  * inhibited pages, we have to be prepared to switch processes
 174  * to use 4k pages when they create cache-inhibited mappings.
 175  * If this is the case, mmu_ci_restrictions will be set to 1.
 176  */
 177 extern int mmu_ci_restrictions;
 178
 179 #ifdef CONFIG_HUGETLB_PAGE
 180 /*
 181  * The page size index of the huge pages for use by hugetlbfs
 182  */
 183 extern int mmu_huge_psize;
 184
 185 #endif /* CONFIG_HUGETLB_PAGE */
 186
 187 /*
 188  * This function sets the AVPN and L fields of the HPTE  appropriately
 189  * for the page size
 190  */
 191 static inline unsigned long hpte_encode_v(unsigned long va, int psize)
 192 {
 193         unsigned long v =
 194         v = (va >> 23) & ~(mmu_psize_defs[psize].avpnm);
 195         v <<= HPTE_V_AVPN_SHIFT;
 196         if (psize != MMU_PAGE_4K)
 197                 v |= HPTE_V_LARGE;
 198         return v;
 199 }
 200
 201 /*
 202  * This function sets the ARPN, and LP fields of the HPTE appropriately
 203  * for the page size. We assume the pa is already "clean" that is properly
 204  * aligned for the requested page size
 205  */
 206 static inline unsigned long hpte_encode_r(unsigned long pa, int psize)
 207 {
 208         unsigned long r;
 209
 210         /* A 4K page needs no special encoding */
 211         if (psize == MMU_PAGE_4K)
 212                 return pa & HPTE_R_RPN;
 213         else {
 214                 unsigned int penc = mmu_psize_defs[psize].penc;
 215                 unsigned int shift = mmu_psize_defs[psize].shift;
 216                 return (pa & ~((1ul << shift) - 1)) | (penc << 12);
 217         }
 218         return r;
 219 }
 220
 221 /*
 222  * This hashes a virtual address for a 256Mb segment only for now
 223  */
 224
 225 static inline unsigned long hpt_hash(unsigned long va, unsigned int shift)
 226 {
 227         return ((va >> 28) & 0x7fffffffffUL) ^ ((va & 0x0fffffffUL) >> shift);
 228 }
 229
 230 extern int __hash_page_4K(unsigned long ea, unsigned long access,
 231                           unsigned long vsid, pte_t *ptep, unsigned long trap,
 232                           unsigned int local);
 233 extern int __hash_page_64K(unsigned long ea, unsigned long access,
 234                            unsigned long vsid, pte_t *ptep, unsigned long trap,
 235                            unsigned int local);
 236 struct mm_struct;
 237 extern int hash_huge_page(struct mm_struct *mm, unsigned long access,
 238                           unsigned long ea, unsigned long vsid, int local,
 239                           unsigned long trap);
 240
 241 extern int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
 242                              unsigned long pstart, unsigned long mode,
 243                              int psize);
 244
 245 extern void htab_initialize(void);
 246 extern void htab_initialize_secondary(void);
 247 extern void hpte_init_native(void);
 248 extern void hpte_init_lpar(void);
 249 extern void hpte_init_iSeries(void);
 250 extern void hpte_init_beat(void);
 251
 252 extern void stabs_alloc(void);
 253 extern void slb_initialize(void);
 254 extern void slb_flush_and_rebolt(void);
 255 extern void stab_initialize(unsigned long stab);
 256
 257 #endif /* __ASSEMBLY__ */
 258
 259 /*
 260  * VSID allocation
 261  *
 262  * We first generate a 36-bit "proto-VSID".  For kernel addresses this
 263  * is equal to the ESID, for user addresses it is:
 264  *      (context << 15) | (esid & 0x7fff)
 265  *
 266  * The two forms are distinguishable because the top bit is 0 for user
 267  * addresses, whereas the top two bits are 1 for kernel addresses.
 268  * Proto-VSIDs with the top two bits equal to 0b10 are reserved for
 269  * now.
 270  *
 271  * The proto-VSIDs are then scrambled into real VSIDs with the
 272  * multiplicative hash:
 273  *
 274  *      VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS
 275  *      where   VSID_MULTIPLIER = 268435399 = 0xFFFFFC7
 276  *              VSID_MODULUS = 2^36-1 = 0xFFFFFFFFF
 277  *
 278  * This scramble is only well defined for proto-VSIDs below
 279  * 0xFFFFFFFFF, so both proto-VSID and actual VSID 0xFFFFFFFFF are
 280  * reserved.  VSID_MULTIPLIER is prime, so in particular it is
 281  * co-prime to VSID_MODULUS, making this a 1:1 scrambling function.
 282  * Because the modulus is 2^n-1 we can compute it efficiently without
 283  * a divide or extra multiply (see below).
 284  *
 285  * This scheme has several advantages over older methods:
 286  *
 287  *      - We have VSIDs allocated for every kernel address
 288  * (i.e. everything above 0xC000000000000000), except the very top
 289  * segment, which simplifies several things.
 290  *
 291  *      - We allow for 15 significant bits of ESID and 20 bits of
 292  * context for user addresses.  i.e. 8T (43 bits) of address space for
 293  * up to 1M contexts (although the page table structure and context
 294  * allocation will need changes to take advantage of this).
 295  *
 296  *      - The scramble function gives robust scattering in the hash
 297  * table (at least based on some initial results).  The previous
 298  * method was more susceptible to pathological cases giving excessive
 299  * hash collisions.
 300  */
 301 /*
 302  * WARNING - If you change these you must make sure the asm
 303  * implementations in slb_allocate (slb_low.S), do_stab_bolted
 304  * (head.S) and ASM_VSID_SCRAMBLE (below) are changed accordingly.
 305  *
 306  * You'll also need to change the precomputed VSID values in head.S
 307  * which are used by the iSeries firmware.
 308  */
 309
 310 #define VSID_MULTIPLIER ASM_CONST(200730139)    /* 28-bit prime */
 311 #define VSID_BITS       36
 312 #define VSID_MODULUS    ((1UL<<VSID_BITS)-1)
 313
 314 #define CONTEXT_BITS    19
 315 #define USER_ESID_BITS  16
 316
 317 #define USER_VSID_RANGE (1UL << (USER_ESID_BITS + SID_SHIFT))
 318
 319 /*
 320  * This macro generates asm code to compute the VSID scramble
 321  * function.  Used in slb_allocate() and do_stab_bolted.  The function
 322  * computed is: (protovsid*VSID_MULTIPLIER) % VSID_MODULUS
 323  *
 324  *      rt = register continaing the proto-VSID and into which the
 325  *              VSID will be stored
 326  *      rx = scratch register (clobbered)
 327  *
 328  *      - rt and rx must be different registers
 329  *      - The answer will end up in the low 36 bits of rt.  The higher
 330  *        bits may contain other garbage, so you may need to mask the
 331  *        result.
 332  */
 333 #define ASM_VSID_SCRAMBLE(rt, rx)       \
 334         lis     rx,VSID_MULTIPLIER@h;                                   \
 335         ori     rx,rx,VSID_MULTIPLIER@l;                                \
 336         mulld   rt,rt,rx;               /* rt = rt * MULTIPLIER */      \
 337                                                                         \
 338         srdi    rx,rt,VSID_BITS;                                        \
 339         clrldi  rt,rt,(64-VSID_BITS);                                   \
 340         add     rt,rt,rx;               /* add high and low bits */     \
 341         /* Now, r3 == VSID (mod 2^36-1), and lies between 0 and         \
 342          * 2^36-1+2^28-1.  That in particular means that if r3 >=       \
 343          * 2^36-1, then r3+1 has the 2^36 bit set.  So, if r3+1 has     \
 344          * the bit clear, r3 already has the answer we want, if it      \
 345          * doesn't, the answer is the low 36 bits of r3+1.  So in all   \
 346          * cases the answer is the low 36 bits of (r3 + ((r3+1) >> 36))*/\
 347         addi    rx,rt,1;                                                \
 348         srdi    rx,rx,VSID_BITS;        /* extract 2^36 bit */          \
 349         add     rt,rt,rx
 350
 351
 352 #ifndef __ASSEMBLY__
 353
 354 typedef unsigned long mm_context_id_t;
 355
 356 typedef struct {
 357         mm_context_id_t id;
 358         u16 user_psize;                 /* page size index */
 359         u16 sllp;                       /* SLB entry page size encoding */
 360 #ifdef CONFIG_HUGETLB_PAGE
 361         u16 low_htlb_areas, high_htlb_areas;
 362 #endif
 363         unsigned long vdso_base;
 364 } mm_context_t;
 365
 366
 367 static inline unsigned long vsid_scramble(unsigned long protovsid)
 368 {
 369 #if 0
 370         /* The code below is equivalent to this function for arguments
 371          * < 2^VSID_BITS, which is all this should ever be called
 372          * with.  However gcc is not clever enough to compute the
 373          * modulus (2^n-1) without a second multiply. */
 374         return ((protovsid * VSID_MULTIPLIER) % VSID_MODULUS);
 375 #else /* 1 */
 376         unsigned long x;
 377
 378         x = protovsid * VSID_MULTIPLIER;
 379         x = (x >> VSID_BITS) + (x & VSID_MODULUS);
 380         return (x + ((x+1) >> VSID_BITS)) & VSID_MODULUS;
 381 #endif /* 1 */
 382 }
 383
 384 /* This is only valid for addresses >= KERNELBASE */
 385 static inline unsigned long get_kernel_vsid(unsigned long ea)
 386 {
 387         return vsid_scramble(ea >> SID_SHIFT);
 388 }
 389
 390 /* This is only valid for user addresses (which are below 2^41) */
 391 static inline unsigned long get_vsid(unsigned long context, unsigned long ea)
 392 {
 393         return vsid_scramble((context << USER_ESID_BITS)
 394                              | (ea >> SID_SHIFT));
 395 }
 396
 397 #define VSID_SCRAMBLE(pvsid)    (((pvsid) * VSID_MULTIPLIER) % VSID_MODULUS)
 398 #define KERNEL_VSID(ea)         VSID_SCRAMBLE(GET_ESID(ea))
 399
 400 /* Physical address used by some IO functions */
 401 typedef unsigned long phys_addr_t;
 402
 403
 404 #endif /* __ASSEMBLY */
 405
 406 #endif /* CONFIG_PPC64 */
 407 #endif /* __KERNEL__ */
 408 #endif /* _ASM_POWERPC_MMU_H_ */