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)
  58 #define SLB_VSID_USER           (SLB_VSID_KP|SLB_VSID_KS|SLB_VSID_C)
  59
  60 #define SLBIE_C                 (0x08000000)
  61
  62 /*
  63  * Hash table
  64  */
  65
  66 #define HPTES_PER_GROUP 8
  67
  68 #define HPTE_V_AVPN_SHIFT       7
  69 #define HPTE_V_AVPN             ASM_CONST(0xffffffffffffff80)
  70 #define HPTE_V_AVPN_VAL(x)      (((x) & HPTE_V_AVPN) >> HPTE_V_AVPN_SHIFT)
  71 #define HPTE_V_BOLTED           ASM_CONST(0x0000000000000010)
  72 #define HPTE_V_LOCK             ASM_CONST(0x0000000000000008)
  73 #define HPTE_V_LARGE            ASM_CONST(0x0000000000000004)
  74 #define HPTE_V_SECONDARY        ASM_CONST(0x0000000000000002)
  75 #define HPTE_V_VALID            ASM_CONST(0x0000000000000001)
  76
  77 #define HPTE_R_PP0              ASM_CONST(0x8000000000000000)
  78 #define HPTE_R_TS               ASM_CONST(0x4000000000000000)
  79 #define HPTE_R_RPN_SHIFT        12
  80 #define HPTE_R_RPN              ASM_CONST(0x3ffffffffffff000)
  81 #define HPTE_R_FLAGS            ASM_CONST(0x00000000000003ff)
  82 #define HPTE_R_PP               ASM_CONST(0x0000000000000003)
  83
  84 /* Values for PP (assumes Ks=0, Kp=1) */
  85 /* pp0 will always be 0 for linux     */
  86 #define PP_RWXX 0       /* Supervisor read/write, User none */
  87 #define PP_RWRX 1       /* Supervisor read/write, User read */
  88 #define PP_RWRW 2       /* Supervisor read/write, User read/write */
  89 #define PP_RXRX 3       /* Supervisor read,       User read */
  90
  91 #ifndef __ASSEMBLY__
  92
  93 typedef struct {
  94         unsigned long v;
  95         unsigned long r;
  96 } hpte_t;
  97
  98 extern hpte_t *htab_address;
  99 extern unsigned long htab_hash_mask;
 100
 101 static inline unsigned long hpt_hash(unsigned long vpn, int large)
 102 {
 103         unsigned long vsid;
 104         unsigned long page;
 105
 106         if (large) {
 107                 vsid = vpn >> 4;
 108                 page = vpn & 0xf;
 109         } else {
 110                 vsid = vpn >> 16;
 111                 page = vpn & 0xffff;
 112         }
 113
 114         return (vsid & 0x7fffffffffUL) ^ page;
 115 }
 116
 117 static inline void __tlbie(unsigned long va, int large)
 118 {
 119         /* clear top 16 bits, non SLS segment */
 120         va &= ~(0xffffULL << 48);
 121
 122         if (large) {
 123                 va &= HPAGE_MASK;
 124                 asm volatile("tlbie %0,1" : : "r"(va) : "memory");
 125         } else {
 126                 va &= PAGE_MASK;
 127                 asm volatile("tlbie %0,0" : : "r"(va) : "memory");
 128         }
 129 }
 130
 131 static inline void tlbie(unsigned long va, int large)
 132 {
 133         asm volatile("ptesync": : :"memory");
 134         __tlbie(va, large);
 135         asm volatile("eieio; tlbsync; ptesync": : :"memory");
 136 }
 137
 138 static inline void __tlbiel(unsigned long va)
 139 {
 140         /* clear top 16 bits, non SLS segment */
 141         va &= ~(0xffffULL << 48);
 142         va &= PAGE_MASK;
 143
 144         /*
 145          * Thanks to Alan Modra we are now able to use machine specific
 146          * assembly instructions (like tlbiel) by using the gas -many flag.
 147          * However we have to support older toolchains so for the moment
 148          * we hardwire it.
 149          */
 150 #if 0
 151         asm volatile("tlbiel %0" : : "r"(va) : "memory");
 152 #else
 153         asm volatile(".long 0x7c000224 | (%0 << 11)" : : "r"(va) : "memory");
 154 #endif
 155 }
 156
 157 static inline void tlbiel(unsigned long va)
 158 {
 159         asm volatile("ptesync": : :"memory");
 160         __tlbiel(va);
 161         asm volatile("ptesync": : :"memory");
 162 }
 163
 164 static inline unsigned long slot2va(unsigned long hpte_v, unsigned long slot)
 165 {
 166         unsigned long avpn = HPTE_V_AVPN_VAL(hpte_v);
 167         unsigned long va;
 168
 169         va = avpn << 23;
 170
 171         if (! (hpte_v & HPTE_V_LARGE)) {
 172                 unsigned long vpi, pteg;
 173
 174                 pteg = slot / HPTES_PER_GROUP;
 175                 if (hpte_v & HPTE_V_SECONDARY)
 176                         pteg = ~pteg;
 177
 178                 vpi = ((va >> 28) ^ pteg) & htab_hash_mask;
 179
 180                 va |= vpi << PAGE_SHIFT;
 181         }
 182
 183         return va;
 184 }
 185
 186 /*
 187  * Handle a fault by adding an HPTE. If the address can't be determined
 188  * to be valid via Linux page tables, return 1. If handled return 0
 189  */
 190 extern int __hash_page(unsigned long ea, unsigned long access,
 191                        unsigned long vsid, pte_t *ptep, unsigned long trap,
 192                        int local);
 193
 194 extern void htab_finish_init(void);
 195
 196 extern void hpte_init_native(void);
 197 extern void hpte_init_lpar(void);
 198 extern void hpte_init_iSeries(void);
 199
 200 extern long pSeries_lpar_hpte_insert(unsigned long hpte_group,
 201                                      unsigned long va, unsigned long prpn,
 202                                      unsigned long vflags,
 203                                      unsigned long rflags);
 204 extern long native_hpte_insert(unsigned long hpte_group, unsigned long va,
 205                                unsigned long prpn,
 206                                unsigned long vflags, unsigned long rflags);
 207
 208 extern void stabs_alloc(void);
 209
 210 #endif /* __ASSEMBLY__ */
 211
 212 /*
 213  * VSID allocation
 214  *
 215  * We first generate a 36-bit "proto-VSID".  For kernel addresses this
 216  * is equal to the ESID, for user addresses it is:
 217  *      (context << 15) | (esid & 0x7fff)
 218  *
 219  * The two forms are distinguishable because the top bit is 0 for user
 220  * addresses, whereas the top two bits are 1 for kernel addresses.
 221  * Proto-VSIDs with the top two bits equal to 0b10 are reserved for
 222  * now.
 223  *
 224  * The proto-VSIDs are then scrambled into real VSIDs with the
 225  * multiplicative hash:
 226  *
 227  *      VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS
 228  *      where   VSID_MULTIPLIER = 268435399 = 0xFFFFFC7
 229  *              VSID_MODULUS = 2^36-1 = 0xFFFFFFFFF
 230  *
 231  * This scramble is only well defined for proto-VSIDs below
 232  * 0xFFFFFFFFF, so both proto-VSID and actual VSID 0xFFFFFFFFF are
 233  * reserved.  VSID_MULTIPLIER is prime, so in particular it is
 234  * co-prime to VSID_MODULUS, making this a 1:1 scrambling function.
 235  * Because the modulus is 2^n-1 we can compute it efficiently without
 236  * a divide or extra multiply (see below).
 237  *
 238  * This scheme has several advantages over older methods:
 239  *
 240  *      - We have VSIDs allocated for every kernel address
 241  * (i.e. everything above 0xC000000000000000), except the very top
 242  * segment, which simplifies several things.
 243  *
 244  *      - We allow for 15 significant bits of ESID and 20 bits of
 245  * context for user addresses.  i.e. 8T (43 bits) of address space for
 246  * up to 1M contexts (although the page table structure and context
 247  * allocation will need changes to take advantage of this).
 248  *
 249  *      - The scramble function gives robust scattering in the hash
 250  * table (at least based on some initial results).  The previous
 251  * method was more susceptible to pathological cases giving excessive
 252  * hash collisions.
 253  */
 254 /*
 255  * WARNING - If you change these you must make sure the asm
 256  * implementations in slb_allocate (slb_low.S), do_stab_bolted
 257  * (head.S) and ASM_VSID_SCRAMBLE (below) are changed accordingly.
 258  *
 259  * You'll also need to change the precomputed VSID values in head.S
 260  * which are used by the iSeries firmware.
 261  */
 262
 263 #define VSID_MULTIPLIER ASM_CONST(200730139)    /* 28-bit prime */
 264 #define VSID_BITS       36
 265 #define VSID_MODULUS    ((1UL<<VSID_BITS)-1)
 266
 267 #define CONTEXT_BITS    19
 268 #define USER_ESID_BITS  16
 269
 270 #define USER_VSID_RANGE (1UL << (USER_ESID_BITS + SID_SHIFT))
 271
 272 /*
 273  * This macro generates asm code to compute the VSID scramble
 274  * function.  Used in slb_allocate() and do_stab_bolted.  The function
 275  * computed is: (protovsid*VSID_MULTIPLIER) % VSID_MODULUS
 276  *
 277  *      rt = register continaing the proto-VSID and into which the
 278  *              VSID will be stored
 279  *      rx = scratch register (clobbered)
 280  *
 281  *      - rt and rx must be different registers
 282  *      - The answer will end up in the low 36 bits of rt.  The higher
 283  *        bits may contain other garbage, so you may need to mask the
 284  *        result.
 285  */
 286 #define ASM_VSID_SCRAMBLE(rt, rx)       \
 287         lis     rx,VSID_MULTIPLIER@h;                                   \
 288         ori     rx,rx,VSID_MULTIPLIER@l;                                \
 289         mulld   rt,rt,rx;               /* rt = rt * MULTIPLIER */      \
 290                                                                         \
 291         srdi    rx,rt,VSID_BITS;                                        \
 292         clrldi  rt,rt,(64-VSID_BITS);                                   \
 293         add     rt,rt,rx;               /* add high and low bits */     \
 294         /* Now, r3 == VSID (mod 2^36-1), and lies between 0 and         \
 295          * 2^36-1+2^28-1.  That in particular means that if r3 >=       \
 296          * 2^36-1, then r3+1 has the 2^36 bit set.  So, if r3+1 has     \
 297          * the bit clear, r3 already has the answer we want, if it      \
 298          * doesn't, the answer is the low 36 bits of r3+1.  So in all   \
 299          * cases the answer is the low 36 bits of (r3 + ((r3+1) >> 36))*/\
 300         addi    rx,rt,1;                                                \
 301         srdi    rx,rx,VSID_BITS;        /* extract 2^36 bit */          \
 302         add     rt,rt,rx
 303
 304
 305 #ifndef __ASSEMBLY__
 306
 307 typedef unsigned long mm_context_id_t;
 308
 309 typedef struct {
 310         mm_context_id_t id;
 311 #ifdef CONFIG_HUGETLB_PAGE
 312         u16 low_htlb_areas, high_htlb_areas;
 313 #endif
 314 } mm_context_t;
 315
 316
 317 static inline unsigned long vsid_scramble(unsigned long protovsid)
 318 {
 319 #if 0
 320         /* The code below is equivalent to this function for arguments
 321          * < 2^VSID_BITS, which is all this should ever be called
 322          * with.  However gcc is not clever enough to compute the
 323          * modulus (2^n-1) without a second multiply. */
 324         return ((protovsid * VSID_MULTIPLIER) % VSID_MODULUS);
 325 #else /* 1 */
 326         unsigned long x;
 327
 328         x = protovsid * VSID_MULTIPLIER;
 329         x = (x >> VSID_BITS) + (x & VSID_MODULUS);
 330         return (x + ((x+1) >> VSID_BITS)) & VSID_MODULUS;
 331 #endif /* 1 */
 332 }
 333
 334 /* This is only valid for addresses >= KERNELBASE */
 335 static inline unsigned long get_kernel_vsid(unsigned long ea)
 336 {
 337         return vsid_scramble(ea >> SID_SHIFT);
 338 }
 339
 340 /* This is only valid for user addresses (which are below 2^41) */
 341 static inline unsigned long get_vsid(unsigned long context, unsigned long ea)
 342 {
 343         return vsid_scramble((context << USER_ESID_BITS)
 344                              | (ea >> SID_SHIFT));
 345 }
 346
 347 #define VSID_SCRAMBLE(pvsid)    (((pvsid) * VSID_MULTIPLIER) % VSID_MODULUS)
 348 #define KERNEL_VSID(ea)         VSID_SCRAMBLE(GET_ESID(ea))
 349
 350 #endif /* __ASSEMBLY */
 351
 352 #endif /* _PPC64_MMU_H_ */