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      0x9
  32 #define STAB0_PHYS_ADDR (STAB0_PAGE<<PAGE_SHIFT)
  33 #define STAB0_VIRT_ADDR (KERNELBASE+STAB0_PHYS_ADDR)
  34
  35 /*
  36  * SLB
  37  */
  38
  39 #define SLB_NUM_BOLTED          3
  40 #define SLB_CACHE_ENTRIES       8
  41
  42 /* Bits in the SLB ESID word */
  43 #define SLB_ESID_V              ASM_CONST(0x0000000008000000) /* valid */
  44
  45 /* Bits in the SLB VSID word */
  46 #define SLB_VSID_SHIFT          12
  47 #define SLB_VSID_KS             ASM_CONST(0x0000000000000800)
  48 #define SLB_VSID_KP             ASM_CONST(0x0000000000000400)
  49 #define SLB_VSID_N              ASM_CONST(0x0000000000000200) /* no-execute */
  50 #define SLB_VSID_L              ASM_CONST(0x0000000000000100) /* largepage */
  51 #define SLB_VSID_C              ASM_CONST(0x0000000000000080) /* class */
  52 #define SLB_VSID_LS             ASM_CONST(0x0000000000000070) /* size of largepage */
  53
  54 #define SLB_VSID_KERNEL         (SLB_VSID_KP|SLB_VSID_C)
  55 #define SLB_VSID_USER           (SLB_VSID_KP|SLB_VSID_KS)
  56
  57 /*
  58  * Hash table
  59  */
  60
  61 #define HPTES_PER_GROUP 8
  62
  63 #define HPTE_V_AVPN_SHIFT       7
  64 #define HPTE_V_AVPN             ASM_CONST(0xffffffffffffff80)
  65 #define HPTE_V_AVPN_VAL(x)      (((x) & HPTE_V_AVPN) >> HPTE_V_AVPN_SHIFT)
  66 #define HPTE_V_BOLTED           ASM_CONST(0x0000000000000010)
  67 #define HPTE_V_LOCK             ASM_CONST(0x0000000000000008)
  68 #define HPTE_V_LARGE            ASM_CONST(0x0000000000000004)
  69 #define HPTE_V_SECONDARY        ASM_CONST(0x0000000000000002)
  70 #define HPTE_V_VALID            ASM_CONST(0x0000000000000001)
  71
  72 #define HPTE_R_PP0              ASM_CONST(0x8000000000000000)
  73 #define HPTE_R_TS               ASM_CONST(0x4000000000000000)
  74 #define HPTE_R_RPN_SHIFT        12
  75 #define HPTE_R_RPN              ASM_CONST(0x3ffffffffffff000)
  76 #define HPTE_R_FLAGS            ASM_CONST(0x00000000000003ff)
  77 #define HPTE_R_PP               ASM_CONST(0x0000000000000003)
  78
  79 /* Values for PP (assumes Ks=0, Kp=1) */
  80 /* pp0 will always be 0 for linux     */
  81 #define PP_RWXX 0       /* Supervisor read/write, User none */
  82 #define PP_RWRX 1       /* Supervisor read/write, User read */
  83 #define PP_RWRW 2       /* Supervisor read/write, User read/write */
  84 #define PP_RXRX 3       /* Supervisor read,       User read */
  85
  86 #ifndef __ASSEMBLY__
  87
  88 typedef struct {
  89         unsigned long v;
  90         unsigned long r;
  91 } hpte_t;
  92
  93 extern hpte_t *htab_address;
  94 extern unsigned long htab_hash_mask;
  95
  96 static inline unsigned long hpt_hash(unsigned long vpn, int large)
  97 {
  98         unsigned long vsid;
  99         unsigned long page;
 100
 101         if (large) {
 102                 vsid = vpn >> 4;
 103                 page = vpn & 0xf;
 104         } else {
 105                 vsid = vpn >> 16;
 106                 page = vpn & 0xffff;
 107         }
 108
 109         return (vsid & 0x7fffffffffUL) ^ page;
 110 }
 111
 112 static inline void __tlbie(unsigned long va, int large)
 113 {
 114         /* clear top 16 bits, non SLS segment */
 115         va &= ~(0xffffULL << 48);
 116
 117         if (large) {
 118                 va &= HPAGE_MASK;
 119                 asm volatile("tlbie %0,1" : : "r"(va) : "memory");
 120         } else {
 121                 va &= PAGE_MASK;
 122                 asm volatile("tlbie %0,0" : : "r"(va) : "memory");
 123         }
 124 }
 125
 126 static inline void tlbie(unsigned long va, int large)
 127 {
 128         asm volatile("ptesync": : :"memory");
 129         __tlbie(va, large);
 130         asm volatile("eieio; tlbsync; ptesync": : :"memory");
 131 }
 132
 133 static inline void __tlbiel(unsigned long va)
 134 {
 135         /* clear top 16 bits, non SLS segment */
 136         va &= ~(0xffffULL << 48);
 137         va &= PAGE_MASK;
 138
 139         /*
 140          * Thanks to Alan Modra we are now able to use machine specific
 141          * assembly instructions (like tlbiel) by using the gas -many flag.
 142          * However we have to support older toolchains so for the moment
 143          * we hardwire it.
 144          */
 145 #if 0
 146         asm volatile("tlbiel %0" : : "r"(va) : "memory");
 147 #else
 148         asm volatile(".long 0x7c000224 | (%0 << 11)" : : "r"(va) : "memory");
 149 #endif
 150 }
 151
 152 static inline void tlbiel(unsigned long va)
 153 {
 154         asm volatile("ptesync": : :"memory");
 155         __tlbiel(va);
 156         asm volatile("ptesync": : :"memory");
 157 }
 158
 159 static inline unsigned long slot2va(unsigned long hpte_v, unsigned long slot)
 160 {
 161         unsigned long avpn = HPTE_V_AVPN_VAL(hpte_v);
 162         unsigned long va;
 163
 164         va = avpn << 23;
 165
 166         if (! (hpte_v & HPTE_V_LARGE)) {
 167                 unsigned long vpi, pteg;
 168
 169                 pteg = slot / HPTES_PER_GROUP;
 170                 if (hpte_v & HPTE_V_SECONDARY)
 171                         pteg = ~pteg;
 172
 173                 vpi = ((va >> 28) ^ pteg) & htab_hash_mask;
 174
 175                 va |= vpi << PAGE_SHIFT;
 176         }
 177
 178         return va;
 179 }
 180
 181 /*
 182  * Handle a fault by adding an HPTE. If the address can't be determined
 183  * to be valid via Linux page tables, return 1. If handled return 0
 184  */
 185 extern int __hash_page(unsigned long ea, unsigned long access,
 186                        unsigned long vsid, pte_t *ptep, unsigned long trap,
 187                        int local);
 188
 189 extern void htab_finish_init(void);
 190
 191 extern void hpte_init_native(void);
 192 extern void hpte_init_lpar(void);
 193 extern void hpte_init_iSeries(void);
 194
 195 extern long pSeries_lpar_hpte_insert(unsigned long hpte_group,
 196                                      unsigned long va, unsigned long prpn,
 197                                      unsigned long vflags,
 198                                      unsigned long rflags);
 199 extern long native_hpte_insert(unsigned long hpte_group, unsigned long va,
 200                                unsigned long prpn,
 201                                unsigned long vflags, unsigned long rflags);
 202
 203 extern void stabs_alloc(void);
 204
 205 #endif /* __ASSEMBLY__ */
 206
 207 /*
 208  * VSID allocation
 209  *
 210  * We first generate a 36-bit "proto-VSID".  For kernel addresses this
 211  * is equal to the ESID, for user addresses it is:
 212  *      (context << 15) | (esid & 0x7fff)
 213  *
 214  * The two forms are distinguishable because the top bit is 0 for user
 215  * addresses, whereas the top two bits are 1 for kernel addresses.
 216  * Proto-VSIDs with the top two bits equal to 0b10 are reserved for
 217  * now.
 218  *
 219  * The proto-VSIDs are then scrambled into real VSIDs with the
 220  * multiplicative hash:
 221  *
 222  *      VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS
 223  *      where   VSID_MULTIPLIER = 268435399 = 0xFFFFFC7
 224  *              VSID_MODULUS = 2^36-1 = 0xFFFFFFFFF
 225  *
 226  * This scramble is only well defined for proto-VSIDs below
 227  * 0xFFFFFFFFF, so both proto-VSID and actual VSID 0xFFFFFFFFF are
 228  * reserved.  VSID_MULTIPLIER is prime, so in particular it is
 229  * co-prime to VSID_MODULUS, making this a 1:1 scrambling function.
 230  * Because the modulus is 2^n-1 we can compute it efficiently without
 231  * a divide or extra multiply (see below).
 232  *
 233  * This scheme has several advantages over older methods:
 234  *
 235  *      - We have VSIDs allocated for every kernel address
 236  * (i.e. everything above 0xC000000000000000), except the very top
 237  * segment, which simplifies several things.
 238  *
 239  *      - We allow for 15 significant bits of ESID and 20 bits of
 240  * context for user addresses.  i.e. 8T (43 bits) of address space for
 241  * up to 1M contexts (although the page table structure and context
 242  * allocation will need changes to take advantage of this).
 243  *
 244  *      - The scramble function gives robust scattering in the hash
 245  * table (at least based on some initial results).  The previous
 246  * method was more susceptible to pathological cases giving excessive
 247  * hash collisions.
 248  */
 249 /*
 250  * WARNING - If you change these you must make sure the asm
 251  * implementations in slb_allocate (slb_low.S), do_stab_bolted
 252  * (head.S) and ASM_VSID_SCRAMBLE (below) are changed accordingly.
 253  *
 254  * You'll also need to change the precomputed VSID values in head.S
 255  * which are used by the iSeries firmware.
 256  */
 257
 258 #define VSID_MULTIPLIER ASM_CONST(200730139)    /* 28-bit prime */
 259 #define VSID_BITS       36
 260 #define VSID_MODULUS    ((1UL<<VSID_BITS)-1)
 261
 262 #define CONTEXT_BITS    20
 263 #define USER_ESID_BITS  15
 264
 265 /*
 266  * This macro generates asm code to compute the VSID scramble
 267  * function.  Used in slb_allocate() and do_stab_bolted.  The function
 268  * computed is: (protovsid*VSID_MULTIPLIER) % VSID_MODULUS
 269  *
 270  *      rt = register continaing the proto-VSID and into which the
 271  *              VSID will be stored
 272  *      rx = scratch register (clobbered)
 273  *
 274  *      - rt and rx must be different registers
 275  *      - The answer will end up in the low 36 bits of rt.  The higher
 276  *        bits may contain other garbage, so you may need to mask the
 277  *        result.
 278  */
 279 #define ASM_VSID_SCRAMBLE(rt, rx)       \
 280         lis     rx,VSID_MULTIPLIER@h;                                   \
 281         ori     rx,rx,VSID_MULTIPLIER@l;                                \
 282         mulld   rt,rt,rx;               /* rt = rt * MULTIPLIER */      \
 283                                                                         \
 284         srdi    rx,rt,VSID_BITS;                                        \
 285         clrldi  rt,rt,(64-VSID_BITS);                                   \
 286         add     rt,rt,rx;               /* add high and low bits */     \
 287         /* Now, r3 == VSID (mod 2^36-1), and lies between 0 and         \
 288          * 2^36-1+2^28-1.  That in particular means that if r3 >=       \
 289          * 2^36-1, then r3+1 has the 2^36 bit set.  So, if r3+1 has     \
 290          * the bit clear, r3 already has the answer we want, if it      \
 291          * doesn't, the answer is the low 36 bits of r3+1.  So in all   \
 292          * cases the answer is the low 36 bits of (r3 + ((r3+1) >> 36))*/\
 293         addi    rx,rt,1;                                                \
 294         srdi    rx,rx,VSID_BITS;        /* extract 2^36 bit */          \
 295         add     rt,rt,rx
 296
 297
 298 #ifndef __ASSEMBLY__
 299
 300 typedef unsigned long mm_context_id_t;
 301
 302 typedef struct {
 303         mm_context_id_t id;
 304 #ifdef CONFIG_HUGETLB_PAGE
 305         pgd_t *huge_pgdir;
 306         u16 htlb_segs; /* bitmask */
 307 #endif
 308 } mm_context_t;
 309
 310
 311 static inline unsigned long vsid_scramble(unsigned long protovsid)
 312 {
 313 #if 0
 314         /* The code below is equivalent to this function for arguments
 315          * < 2^VSID_BITS, which is all this should ever be called
 316          * with.  However gcc is not clever enough to compute the
 317          * modulus (2^n-1) without a second multiply. */
 318         return ((protovsid * VSID_MULTIPLIER) % VSID_MODULUS);
 319 #else /* 1 */
 320         unsigned long x;
 321
 322         x = protovsid * VSID_MULTIPLIER;
 323         x = (x >> VSID_BITS) + (x & VSID_MODULUS);
 324         return (x + ((x+1) >> VSID_BITS)) & VSID_MODULUS;
 325 #endif /* 1 */
 326 }
 327
 328 /* This is only valid for addresses >= KERNELBASE */
 329 static inline unsigned long get_kernel_vsid(unsigned long ea)
 330 {
 331         return vsid_scramble(ea >> SID_SHIFT);
 332 }
 333
 334 /* This is only valid for user addresses (which are below 2^41) */
 335 static inline unsigned long get_vsid(unsigned long context, unsigned long ea)
 336 {
 337         return vsid_scramble((context << USER_ESID_BITS)
 338                              | (ea >> SID_SHIFT));
 339 }
 340
 341 #define VSID_SCRAMBLE(pvsid)    (((pvsid) * VSID_MULTIPLIER) % VSID_MODULUS)
 342 #define KERNEL_VSID(ea)         VSID_SCRAMBLE(GET_ESID(ea))
 343
 344 #endif /* __ASSEMBLY */
 345
 346 #endif /* _PPC64_MMU_H_ */