include/asm-ppc64/mmu.h

   1 /*
   2  * PowerPC memory management structures
   3  *
   4  * Dave Engebretsen & Mike Corrigan <{engebret|mikejc}@us.ibm.com>
   5  *   PPC64 rework.
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License
   9  * as published by the Free Software Foundation; either version
  10  * 2 of the License, or (at your option) any later version.
  11  */
  12
  13 #ifndef _PPC64_MMU_H_
  14 #define _PPC64_MMU_H_
  15
  16 #include <linux/config.h>
  17 #include <asm/page.h>
  18
  19 /*
  20  * Segment table
  21  */
  22
  23 #define STE_ESID_V      0x80
  24 #define STE_ESID_KS     0x20
  25 #define STE_ESID_KP     0x10
  26 #define STE_ESID_N      0x08
  27
  28 #define STE_VSID_SHIFT  12
  29
  30 /* Location of cpu0's segment table */
  31 #define STAB0_PAGE      0x6
  32 #define STAB0_PHYS_ADDR (STAB0_PAGE<<PAGE_SHIFT)
  33
  34 #ifndef __ASSEMBLY__
  35 extern char initial_stab[];
  36 #endif /* ! __ASSEMBLY */
  37
  38 /*
  39  * SLB
  40  */
  41
  42 #define SLB_NUM_BOLTED          3
  43 #define SLB_CACHE_ENTRIES       8
  44
  45 /* Bits in the SLB ESID word */
  46 #define SLB_ESID_V              ASM_CONST(0x0000000008000000) /* valid */
  47
  48 /* Bits in the SLB VSID word */
  49 #define SLB_VSID_SHIFT          12
  50 #define SLB_VSID_KS             ASM_CONST(0x0000000000000800)
  51 #define SLB_VSID_KP             ASM_CONST(0x0000000000000400)
  52 #define SLB_VSID_N              ASM_CONST(0x0000000000000200) /* no-execute */
  53 #define SLB_VSID_L              ASM_CONST(0x0000000000000100) /* largepage */
  54 #define SLB_VSID_C              ASM_CONST(0x0000000000000080) /* class */
  55 #define SLB_VSID_LS             ASM_CONST(0x0000000000000070) /* size of largepage */
  56
  57 #define SLB_VSID_KERNEL         (SLB_VSID_KP|SLB_VSID_C)
  58 #define SLB_VSID_USER           (SLB_VSID_KP|SLB_VSID_KS)
  59
  60 /*
  61  * Hash table
  62  */
  63
  64 #define HPTES_PER_GROUP 8
  65
  66 #define HPTE_V_AVPN_SHIFT       7
  67 #define HPTE_V_AVPN             ASM_CONST(0xffffffffffffff80)
  68 #define HPTE_V_AVPN_VAL(x)      (((x) & HPTE_V_AVPN) >> HPTE_V_AVPN_SHIFT)
  69 #define HPTE_V_BOLTED           ASM_CONST(0x0000000000000010)
  70 #define HPTE_V_LOCK             ASM_CONST(0x0000000000000008)
  71 #define HPTE_V_LARGE            ASM_CONST(0x0000000000000004)
  72 #define HPTE_V_SECONDARY        ASM_CONST(0x0000000000000002)
  73 #define HPTE_V_VALID            ASM_CONST(0x0000000000000001)
  74
  75 #define HPTE_R_PP0              ASM_CONST(0x8000000000000000)
  76 #define HPTE_R_TS               ASM_CONST(0x4000000000000000)
  77 #define HPTE_R_RPN_SHIFT        12
  78 #define HPTE_R_RPN              ASM_CONST(0x3ffffffffffff000)
  79 #define HPTE_R_FLAGS            ASM_CONST(0x00000000000003ff)
  80 #define HPTE_R_PP               ASM_CONST(0x0000000000000003)
  81
  82 /* Values for PP (assumes Ks=0, Kp=1) */
  83 /* pp0 will always be 0 for linux     */
  84 #define PP_RWXX 0       /* Supervisor read/write, User none */
  85 #define PP_RWRX 1       /* Supervisor read/write, User read */
  86 #define PP_RWRW 2       /* Supervisor read/write, User read/write */
  87 #define PP_RXRX 3       /* Supervisor read,       User read */
  88
  89 #ifndef __ASSEMBLY__
  90
  91 typedef struct {
  92         unsigned long v;
  93         unsigned long r;
  94 } hpte_t;
  95
  96 extern hpte_t *htab_address;
  97 extern unsigned long htab_hash_mask;
  98
  99 static inline unsigned long hpt_hash(unsigned long vpn, int large)
 100 {
 101         unsigned long vsid;
 102         unsigned long page;
 103
 104         if (large) {
 105                 vsid = vpn >> 4;
 106                 page = vpn & 0xf;
 107         } else {
 108                 vsid = vpn >> 16;
 109                 page = vpn & 0xffff;
 110         }
 111
 112         return (vsid & 0x7fffffffffUL) ^ page;
 113 }
 114
 115 static inline void __tlbie(unsigned long va, int large)
 116 {
 117         /* clear top 16 bits, non SLS segment */
 118         va &= ~(0xffffULL << 48);
 119
 120         if (large) {
 121                 va &= HPAGE_MASK;
 122                 asm volatile("tlbie %0,1" : : "r"(va) : "memory");
 123         } else {
 124                 va &= PAGE_MASK;
 125                 asm volatile("tlbie %0,0" : : "r"(va) : "memory");
 126         }
 127 }
 128
 129 static inline void tlbie(unsigned long va, int large)
 130 {
 131         asm volatile("ptesync": : :"memory");
 132         __tlbie(va, large);
 133         asm volatile("eieio; tlbsync; ptesync": : :"memory");
 134 }
 135
 136 static inline void __tlbiel(unsigned long va)
 137 {
 138         /* clear top 16 bits, non SLS segment */
 139         va &= ~(0xffffULL << 48);
 140         va &= PAGE_MASK;
 141
 142         /*
 143          * Thanks to Alan Modra we are now able to use machine specific
 144          * assembly instructions (like tlbiel) by using the gas -many flag.
 145          * However we have to support older toolchains so for the moment
 146          * we hardwire it.
 147          */
 148 #if 0
 149         asm volatile("tlbiel %0" : : "r"(va) : "memory");
 150 #else
 151         asm volatile(".long 0x7c000224 | (%0 << 11)" : : "r"(va) : "memory");
 152 #endif
 153 }
 154
 155 static inline void tlbiel(unsigned long va)
 156 {
 157         asm volatile("ptesync": : :"memory");
 158         __tlbiel(va);
 159         asm volatile("ptesync": : :"memory");
 160 }
 161
 162 static inline unsigned long slot2va(unsigned long hpte_v, unsigned long slot)
 163 {
 164         unsigned long avpn = HPTE_V_AVPN_VAL(hpte_v);
 165         unsigned long va;
 166
 167         va = avpn << 23;
 168
 169         if (! (hpte_v & HPTE_V_LARGE)) {
 170                 unsigned long vpi, pteg;
 171
 172                 pteg = slot / HPTES_PER_GROUP;
 173                 if (hpte_v & HPTE_V_SECONDARY)
 174                         pteg = ~pteg;
 175
 176                 vpi = ((va >> 28) ^ pteg) & htab_hash_mask;
 177
 178                 va |= vpi << PAGE_SHIFT;
 179         }
 180
 181         return va;
 182 }
 183
 184 /*
 185  * Handle a fault by adding an HPTE. If the address can't be determined
 186  * to be valid via Linux page tables, return 1. If handled return 0
 187  */
 188 extern int __hash_page(unsigned long ea, unsigned long access,
 189                        unsigned long vsid, pte_t *ptep, unsigned long trap,
 190                        int local);
 191
 192 extern void htab_finish_init(void);
 193
 194 extern void hpte_init_native(void);
 195 extern void hpte_init_lpar(void);
 196 extern void hpte_init_iSeries(void);
 197
 198 extern long pSeries_lpar_hpte_insert(unsigned long hpte_group,
 199                                      unsigned long va, unsigned long prpn,
 200                                      unsigned long vflags,
 201                                      unsigned long rflags);
 202 extern long native_hpte_insert(unsigned long hpte_group, unsigned long va,
 203                                unsigned long prpn,
 204                                unsigned long vflags, unsigned long rflags);
 205
 206 extern void stabs_alloc(void);
 207
 208 #endif /* __ASSEMBLY__ */
 209
 210 /*
 211  * VSID allocation
 212  *
 213  * We first generate a 36-bit "proto-VSID".  For kernel addresses this
 214  * is equal to the ESID, for user addresses it is:
 215  *      (context << 15) | (esid & 0x7fff)
 216  *
 217  * The two forms are distinguishable because the top bit is 0 for user
 218  * addresses, whereas the top two bits are 1 for kernel addresses.
 219  * Proto-VSIDs with the top two bits equal to 0b10 are reserved for
 220  * now.
 221  *
 222  * The proto-VSIDs are then scrambled into real VSIDs with the
 223  * multiplicative hash:
 224  *
 225  *      VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS
 226  *      where   VSID_MULTIPLIER = 268435399 = 0xFFFFFC7
 227  *              VSID_MODULUS = 2^36-1 = 0xFFFFFFFFF
 228  *
 229  * This scramble is only well defined for proto-VSIDs below
 230  * 0xFFFFFFFFF, so both proto-VSID and actual VSID 0xFFFFFFFFF are
 231  * reserved.  VSID_MULTIPLIER is prime, so in particular it is
 232  * co-prime to VSID_MODULUS, making this a 1:1 scrambling function.
 233  * Because the modulus is 2^n-1 we can compute it efficiently without
 234  * a divide or extra multiply (see below).
 235  *
 236  * This scheme has several advantages over older methods:
 237  *
 238  *      - We have VSIDs allocated for every kernel address
 239  * (i.e. everything above 0xC000000000000000), except the very top
 240  * segment, which simplifies several things.
 241  *
 242  *      - We allow for 15 significant bits of ESID and 20 bits of
 243  * context for user addresses.  i.e. 8T (43 bits) of address space for
 244  * up to 1M contexts (although the page table structure and context
 245  * allocation will need changes to take advantage of this).
 246  *
 247  *      - The scramble function gives robust scattering in the hash
 248  * table (at least based on some initial results).  The previous
 249  * method was more susceptible to pathological cases giving excessive
 250  * hash collisions.
 251  */
 252 /*
 253  * WARNING - If you change these you must make sure the asm
 254  * implementations in slb_allocate (slb_low.S), do_stab_bolted
 255  * (head.S) and ASM_VSID_SCRAMBLE (below) are changed accordingly.
 256  *
 257  * You'll also need to change the precomputed VSID values in head.S
 258  * which are used by the iSeries firmware.
 259  */
 260
 261 #define VSID_MULTIPLIER ASM_CONST(200730139)    /* 28-bit prime */
 262 #define VSID_BITS       36
 263 #define VSID_MODULUS    ((1UL<<VSID_BITS)-1)
 264
 265 #define CONTEXT_BITS    19
 266 #define USER_ESID_BITS  16
 267
 268 #define USER_VSID_RANGE (1UL << (USER_ESID_BITS + SID_SHIFT))
 269
 270 /*
 271  * This macro generates asm code to compute the VSID scramble
 272  * function.  Used in slb_allocate() and do_stab_bolted.  The function
 273  * computed is: (protovsid*VSID_MULTIPLIER) % VSID_MODULUS
 274  *
 275  *      rt = register continaing the proto-VSID and into which the
 276  *              VSID will be stored
 277  *      rx = scratch register (clobbered)
 278  *
 279  *      - rt and rx must be different registers
 280  *      - The answer will end up in the low 36 bits of rt.  The higher
 281  *        bits may contain other garbage, so you may need to mask the
 282  *        result.
 283  */
 284 #define ASM_VSID_SCRAMBLE(rt, rx)       \
 285         lis     rx,VSID_MULTIPLIER@h;                                   \
 286         ori     rx,rx,VSID_MULTIPLIER@l;                                \
 287         mulld   rt,rt,rx;               /* rt = rt * MULTIPLIER */      \
 288                                                                         \
 289         srdi    rx,rt,VSID_BITS;                                        \
 290         clrldi  rt,rt,(64-VSID_BITS);                                   \
 291         add     rt,rt,rx;               /* add high and low bits */     \
 292         /* Now, r3 == VSID (mod 2^36-1), and lies between 0 and         \
 293          * 2^36-1+2^28-1.  That in particular means that if r3 >=       \
 294          * 2^36-1, then r3+1 has the 2^36 bit set.  So, if r3+1 has     \
 295          * the bit clear, r3 already has the answer we want, if it      \
 296          * doesn't, the answer is the low 36 bits of r3+1.  So in all   \
 297          * cases the answer is the low 36 bits of (r3 + ((r3+1) >> 36))*/\
 298         addi    rx,rt,1;                                                \
 299         srdi    rx,rx,VSID_BITS;        /* extract 2^36 bit */          \
 300         add     rt,rt,rx
 301
 302
 303 #ifndef __ASSEMBLY__
 304
 305 typedef unsigned long mm_context_id_t;
 306
 307 typedef struct {
 308         mm_context_id_t id;
 309 #ifdef CONFIG_HUGETLB_PAGE
 310         u16 low_htlb_areas, high_htlb_areas;
 311 #endif
 312 } mm_context_t;
 313
 314
 315 static inline unsigned long vsid_scramble(unsigned long protovsid)
 316 {
 317 #if 0
 318         /* The code below is equivalent to this function for arguments
 319          * < 2^VSID_BITS, which is all this should ever be called
 320          * with.  However gcc is not clever enough to compute the
 321          * modulus (2^n-1) without a second multiply. */
 322         return ((protovsid * VSID_MULTIPLIER) % VSID_MODULUS);
 323 #else /* 1 */
 324         unsigned long x;
 325
 326         x = protovsid * VSID_MULTIPLIER;
 327         x = (x >> VSID_BITS) + (x & VSID_MODULUS);
 328         return (x + ((x+1) >> VSID_BITS)) & VSID_MODULUS;
 329 #endif /* 1 */
 330 }
 331
 332 /* This is only valid for addresses >= KERNELBASE */
 333 static inline unsigned long get_kernel_vsid(unsigned long ea)
 334 {
 335         return vsid_scramble(ea >> SID_SHIFT);
 336 }
 337
 338 /* This is only valid for user addresses (which are below 2^41) */
 339 static inline unsigned long get_vsid(unsigned long context, unsigned long ea)
 340 {
 341         return vsid_scramble((context << USER_ESID_BITS)
 342                              | (ea >> SID_SHIFT));
 343 }
 344
 345 #define VSID_SCRAMBLE(pvsid)    (((pvsid) * VSID_MULTIPLIER) % VSID_MODULUS)
 346 #define KERNEL_VSID(ea)         VSID_SCRAMBLE(GET_ESID(ea))
 347
 348 #endif /* __ASSEMBLY */
 349
 350 #endif /* _PPC64_MMU_H_ */