linux/arch/powerpc/mm/book3s32/mmu.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * This file contains the routines for handling the MMU on those
   4 * PowerPC implementations where the MMU substantially follows the
   5 * architecture specification.  This includes the 6xx, 7xx, 7xxx,
   6 * and 8260 implementations but excludes the 8xx and 4xx.
   7 *  -- paulus
   8 *
   9 *  Derived from arch/ppc/mm/init.c:
  10 *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
  11 *
  12 *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
  13 *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
  14 *    Copyright (C) 1996 Paul Mackerras
  15 *
  16 *  Derived from "arch/i386/mm/init.c"
  17 *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
  18 */
  19
  20#include <linux/kernel.h>
  21#include <linux/mm.h>
  22#include <linux/init.h>
  23#include <linux/highmem.h>
  24#include <linux/memblock.h>
  25
  26#include <asm/prom.h>
  27#include <asm/mmu.h>
  28#include <asm/machdep.h>
  29#include <asm/code-patching.h>
  30#include <asm/sections.h>
  31
  32#include <mm/mmu_decl.h>
  33
  34struct hash_pte *Hash;
  35static unsigned long Hash_size, Hash_mask;
  36unsigned long _SDR1;
  37static unsigned int hash_mb, hash_mb2;
  38
  39struct ppc_bat BATS[8][2];      /* 8 pairs of IBAT, DBAT */
  40
  41struct batrange {               /* stores address ranges mapped by BATs */
  42        unsigned long start;
  43        unsigned long limit;
  44        phys_addr_t phys;
  45} bat_addrs[8];
  46
  47/*
  48 * Return PA for this VA if it is mapped by a BAT, or 0
  49 */
  50phys_addr_t v_block_mapped(unsigned long va)
  51{
  52        int b;
  53        for (b = 0; b < ARRAY_SIZE(bat_addrs); ++b)
  54                if (va >= bat_addrs[b].start && va < bat_addrs[b].limit)
  55                        return bat_addrs[b].phys + (va - bat_addrs[b].start);
  56        return 0;
  57}
  58
  59/*
  60 * Return VA for a given PA or 0 if not mapped
  61 */
  62unsigned long p_block_mapped(phys_addr_t pa)
  63{
  64        int b;
  65        for (b = 0; b < ARRAY_SIZE(bat_addrs); ++b)
  66                if (pa >= bat_addrs[b].phys
  67                    && pa < (bat_addrs[b].limit-bat_addrs[b].start)
  68                              +bat_addrs[b].phys)
  69                        return bat_addrs[b].start+(pa-bat_addrs[b].phys);
  70        return 0;
  71}
  72
  73static int find_free_bat(void)
  74{
  75        int b;
  76
  77        if (cpu_has_feature(CPU_FTR_601)) {
  78                for (b = 0; b < 4; b++) {
  79                        struct ppc_bat *bat = BATS[b];
  80
  81                        if (!(bat[0].batl & 0x40))
  82                                return b;
  83                }
  84        } else {
  85                int n = mmu_has_feature(MMU_FTR_USE_HIGH_BATS) ? 8 : 4;
  86
  87                for (b = 0; b < n; b++) {
  88                        struct ppc_bat *bat = BATS[b];
  89
  90                        if (!(bat[1].batu & 3))
  91                                return b;
  92                }
  93        }
  94        return -1;
  95}
  96
  97/*
  98 * This function calculates the size of the larger block usable to map the
  99 * beginning of an area based on the start address and size of that area:
 100 * - max block size is 8M on 601 and 256 on other 6xx.
 101 * - base address must be aligned to the block size. So the maximum block size
 102 *   is identified by the lowest bit set to 1 in the base address (for instance
 103 *   if base is 0x16000000, max size is 0x02000000).
 104 * - block size has to be a power of two. This is calculated by finding the
 105 *   highest bit set to 1.
 106 */
 107static unsigned int block_size(unsigned long base, unsigned long top)
 108{
 109        unsigned int max_size = (cpu_has_feature(CPU_FTR_601) ? 8 : 256) << 20;
 110        unsigned int base_shift = (ffs(base) - 1) & 31;
 111        unsigned int block_shift = (fls(top - base) - 1) & 31;
 112
 113        return min3(max_size, 1U << base_shift, 1U << block_shift);
 114}
 115
 116/*
 117 * Set up one of the IBAT (block address translation) register pairs.
 118 * The parameters are not checked; in particular size must be a power
 119 * of 2 between 128k and 256M.
 120 * Only for 603+ ...
 121 */
 122static void setibat(int index, unsigned long virt, phys_addr_t phys,
 123                    unsigned int size, pgprot_t prot)
 124{
 125        unsigned int bl = (size >> 17) - 1;
 126        int wimgxpp;
 127        struct ppc_bat *bat = BATS[index];
 128        unsigned long flags = pgprot_val(prot);
 129
 130        if (!cpu_has_feature(CPU_FTR_NEED_COHERENT))
 131                flags &= ~_PAGE_COHERENT;
 132
 133        wimgxpp = (flags & _PAGE_COHERENT) | (_PAGE_EXEC ? BPP_RX : BPP_XX);
 134        bat[0].batu = virt | (bl << 2) | 2; /* Vs=1, Vp=0 */
 135        bat[0].batl = BAT_PHYS_ADDR(phys) | wimgxpp;
 136        if (flags & _PAGE_USER)
 137                bat[0].batu |= 1;       /* Vp = 1 */
 138}
 139
 140static void clearibat(int index)
 141{
 142        struct ppc_bat *bat = BATS[index];
 143
 144        bat[0].batu = 0;
 145        bat[0].batl = 0;
 146}
 147
 148static unsigned long __init __mmu_mapin_ram(unsigned long base, unsigned long top)
 149{
 150        int idx;
 151
 152        while ((idx = find_free_bat()) != -1 && base != top) {
 153                unsigned int size = block_size(base, top);
 154
 155                if (size < 128 << 10)
 156                        break;
 157                setbat(idx, PAGE_OFFSET + base, base, size, PAGE_KERNEL_X);
 158                base += size;
 159        }
 160
 161        return base;
 162}
 163
 164unsigned long __init mmu_mapin_ram(unsigned long base, unsigned long top)
 165{
 166        unsigned long done;
 167        unsigned long border = (unsigned long)__init_begin - PAGE_OFFSET;
 168
 169        if (__map_without_bats) {
 170                pr_debug("RAM mapped without BATs\n");
 171                return base;
 172        }
 173
 174        if (!strict_kernel_rwx_enabled() || base >= border || top <= border)
 175                return __mmu_mapin_ram(base, top);
 176
 177        done = __mmu_mapin_ram(base, border);
 178        if (done != border)
 179                return done;
 180
 181        return __mmu_mapin_ram(border, top);
 182}
 183
 184void mmu_mark_initmem_nx(void)
 185{
 186        int nb = mmu_has_feature(MMU_FTR_USE_HIGH_BATS) ? 8 : 4;
 187        int i;
 188        unsigned long base = (unsigned long)_stext - PAGE_OFFSET;
 189        unsigned long top = (unsigned long)_etext - PAGE_OFFSET;
 190        unsigned long size;
 191
 192        if (cpu_has_feature(CPU_FTR_601))
 193                return;
 194
 195        for (i = 0; i < nb - 1 && base < top && top - base > (128 << 10);) {
 196                size = block_size(base, top);
 197                setibat(i++, PAGE_OFFSET + base, base, size, PAGE_KERNEL_TEXT);
 198                base += size;
 199        }
 200        if (base < top) {
 201                size = block_size(base, top);
 202                size = max(size, 128UL << 10);
 203                if ((top - base) > size) {
 204                        if (strict_kernel_rwx_enabled())
 205                                pr_warn("Kernel _etext not properly aligned\n");
 206                        size <<= 1;
 207                }
 208                setibat(i++, PAGE_OFFSET + base, base, size, PAGE_KERNEL_TEXT);
 209                base += size;
 210        }
 211        for (; i < nb; i++)
 212                clearibat(i);
 213
 214        update_bats();
 215
 216        for (i = TASK_SIZE >> 28; i < 16; i++) {
 217                /* Do not set NX on VM space for modules */
 218                if (IS_ENABLED(CONFIG_MODULES) &&
 219                    (VMALLOC_START & 0xf0000000) == i << 28)
 220                        break;
 221                mtsrin(mfsrin(i << 28) | 0x10000000, i << 28);
 222        }
 223}
 224
 225void mmu_mark_rodata_ro(void)
 226{
 227        int nb = mmu_has_feature(MMU_FTR_USE_HIGH_BATS) ? 8 : 4;
 228        int i;
 229
 230        if (cpu_has_feature(CPU_FTR_601))
 231                return;
 232
 233        for (i = 0; i < nb; i++) {
 234                struct ppc_bat *bat = BATS[i];
 235
 236                if (bat_addrs[i].start < (unsigned long)__init_begin)
 237                        bat[1].batl = (bat[1].batl & ~BPP_RW) | BPP_RX;
 238        }
 239
 240        update_bats();
 241}
 242
 243/*
 244 * Set up one of the I/D BAT (block address translation) register pairs.
 245 * The parameters are not checked; in particular size must be a power
 246 * of 2 between 128k and 256M.
 247 * On 603+, only set IBAT when _PAGE_EXEC is set
 248 */
 249void __init setbat(int index, unsigned long virt, phys_addr_t phys,
 250                   unsigned int size, pgprot_t prot)
 251{
 252        unsigned int bl;
 253        int wimgxpp;
 254        struct ppc_bat *bat = BATS[index];
 255        unsigned long flags = pgprot_val(prot);
 256
 257        if ((flags & _PAGE_NO_CACHE) ||
 258            (cpu_has_feature(CPU_FTR_NEED_COHERENT) == 0))
 259                flags &= ~_PAGE_COHERENT;
 260
 261        bl = (size >> 17) - 1;
 262        if (PVR_VER(mfspr(SPRN_PVR)) != 1) {
 263                /* 603, 604, etc. */
 264                /* Do DBAT first */
 265                wimgxpp = flags & (_PAGE_WRITETHRU | _PAGE_NO_CACHE
 266                                   | _PAGE_COHERENT | _PAGE_GUARDED);
 267                wimgxpp |= (flags & _PAGE_RW)? BPP_RW: BPP_RX;
 268                bat[1].batu = virt | (bl << 2) | 2; /* Vs=1, Vp=0 */
 269                bat[1].batl = BAT_PHYS_ADDR(phys) | wimgxpp;
 270                if (flags & _PAGE_USER)
 271                        bat[1].batu |= 1;       /* Vp = 1 */
 272                if (flags & _PAGE_GUARDED) {
 273                        /* G bit must be zero in IBATs */
 274                        flags &= ~_PAGE_EXEC;
 275                }
 276                if (flags & _PAGE_EXEC)
 277                        bat[0] = bat[1];
 278                else
 279                        bat[0].batu = bat[0].batl = 0;
 280        } else {
 281                /* 601 cpu */
 282                if (bl > BL_8M)
 283                        bl = BL_8M;
 284                wimgxpp = flags & (_PAGE_WRITETHRU | _PAGE_NO_CACHE
 285                                   | _PAGE_COHERENT);
 286                wimgxpp |= (flags & _PAGE_RW)?
 287                        ((flags & _PAGE_USER)? PP_RWRW: PP_RWXX): PP_RXRX;
 288                bat->batu = virt | wimgxpp | 4; /* Ks=0, Ku=1 */
 289                bat->batl = phys | bl | 0x40;   /* V=1 */
 290        }
 291
 292        bat_addrs[index].start = virt;
 293        bat_addrs[index].limit = virt + ((bl + 1) << 17) - 1;
 294        bat_addrs[index].phys = phys;
 295}
 296
 297/*
 298 * Preload a translation in the hash table
 299 */
 300void hash_preload(struct mm_struct *mm, unsigned long ea,
 301                  bool is_exec, unsigned long trap)
 302{
 303        pmd_t *pmd;
 304
 305        if (!Hash)
 306                return;
 307        pmd = pmd_offset(pud_offset(pgd_offset(mm, ea), ea), ea);
 308        if (!pmd_none(*pmd))
 309                add_hash_page(mm->context.id, ea, pmd_val(*pmd));
 310}
 311
 312/*
 313 * Initialize the hash table and patch the instructions in hashtable.S.
 314 */
 315void __init MMU_init_hw(void)
 316{
 317        unsigned int n_hpteg, lg_n_hpteg;
 318
 319        if (!mmu_has_feature(MMU_FTR_HPTE_TABLE))
 320                return;
 321
 322        if ( ppc_md.progress ) ppc_md.progress("hash:enter", 0x105);
 323
 324#define LG_HPTEG_SIZE   6               /* 64 bytes per HPTEG */
 325#define SDR1_LOW_BITS   ((n_hpteg - 1) >> 10)
 326#define MIN_N_HPTEG     1024            /* min 64kB hash table */
 327
 328        /*
 329         * Allow 1 HPTE (1/8 HPTEG) for each page of memory.
 330         * This is less than the recommended amount, but then
 331         * Linux ain't AIX.
 332         */
 333        n_hpteg = total_memory / (PAGE_SIZE * 8);
 334        if (n_hpteg < MIN_N_HPTEG)
 335                n_hpteg = MIN_N_HPTEG;
 336        lg_n_hpteg = __ilog2(n_hpteg);
 337        if (n_hpteg & (n_hpteg - 1)) {
 338                ++lg_n_hpteg;           /* round up if not power of 2 */
 339                n_hpteg = 1 << lg_n_hpteg;
 340        }
 341        Hash_size = n_hpteg << LG_HPTEG_SIZE;
 342
 343        /*
 344         * Find some memory for the hash table.
 345         */
 346        if ( ppc_md.progress ) ppc_md.progress("hash:find piece", 0x322);
 347        Hash = memblock_alloc(Hash_size, Hash_size);
 348        if (!Hash)
 349                panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
 350                      __func__, Hash_size, Hash_size);
 351        _SDR1 = __pa(Hash) | SDR1_LOW_BITS;
 352
 353        pr_info("Total memory = %lldMB; using %ldkB for hash table\n",
 354                (unsigned long long)(total_memory >> 20), Hash_size >> 10);
 355
 356
 357        Hash_mask = n_hpteg - 1;
 358        hash_mb2 = hash_mb = 32 - LG_HPTEG_SIZE - lg_n_hpteg;
 359        if (lg_n_hpteg > 16)
 360                hash_mb2 = 16 - LG_HPTEG_SIZE;
 361}
 362
 363void __init MMU_init_hw_patch(void)
 364{
 365        unsigned int hmask = Hash_mask >> (16 - LG_HPTEG_SIZE);
 366
 367        if (ppc_md.progress)
 368                ppc_md.progress("hash:patch", 0x345);
 369        if (ppc_md.progress)
 370                ppc_md.progress("hash:done", 0x205);
 371
 372        /* WARNING: Make sure nothing can trigger a KASAN check past this point */
 373
 374        /*
 375         * Patch up the instructions in hashtable.S:create_hpte
 376         */
 377        modify_instruction_site(&patch__hash_page_A0, 0xffff,
 378                                ((unsigned int)Hash - PAGE_OFFSET) >> 16);
 379        modify_instruction_site(&patch__hash_page_A1, 0x7c0, hash_mb << 6);
 380        modify_instruction_site(&patch__hash_page_A2, 0x7c0, hash_mb2 << 6);
 381        modify_instruction_site(&patch__hash_page_B, 0xffff, hmask);
 382        modify_instruction_site(&patch__hash_page_C, 0xffff, hmask);
 383
 384        /*
 385         * Patch up the instructions in hashtable.S:flush_hash_page
 386         */
 387        modify_instruction_site(&patch__flush_hash_A0, 0xffff,
 388                                ((unsigned int)Hash - PAGE_OFFSET) >> 16);
 389        modify_instruction_site(&patch__flush_hash_A1, 0x7c0, hash_mb << 6);
 390        modify_instruction_site(&patch__flush_hash_A2, 0x7c0, hash_mb2 << 6);
 391        modify_instruction_site(&patch__flush_hash_B, 0xffff, hmask);
 392}
 393
 394void setup_initial_memory_limit(phys_addr_t first_memblock_base,
 395                                phys_addr_t first_memblock_size)
 396{
 397        /* We don't currently support the first MEMBLOCK not mapping 0
 398         * physical on those processors
 399         */
 400        BUG_ON(first_memblock_base != 0);
 401
 402        /* 601 can only access 16MB at the moment */
 403        if (PVR_VER(mfspr(SPRN_PVR)) == 1)
 404                memblock_set_current_limit(min_t(u64, first_memblock_size, 0x01000000));
 405        else /* Anything else has 256M mapped */
 406                memblock_set_current_limit(min_t(u64, first_memblock_size, 0x10000000));
 407}
 408
 409void __init print_system_hash_info(void)
 410{
 411        pr_info("Hash_size         = 0x%lx\n", Hash_size);
 412        if (Hash_mask)
 413                pr_info("Hash_mask         = 0x%lx\n", Hash_mask);
 414}
 415
 416#ifdef CONFIG_PPC_KUEP
 417void __init setup_kuep(bool disabled)
 418{
 419        pr_info("Activating Kernel Userspace Execution Prevention\n");
 420
 421        if (cpu_has_feature(CPU_FTR_601))
 422                pr_warn("KUEP is not working on powerpc 601 (No NX bit in Seg Regs)\n");
 423
 424        if (disabled)
 425                pr_warn("KUEP cannot be disabled yet on 6xx when compiled in\n");
 426}
 427#endif
 428
 429#ifdef CONFIG_PPC_KUAP
 430void __init setup_kuap(bool disabled)
 431{
 432        pr_info("Activating Kernel Userspace Access Protection\n");
 433
 434        if (disabled)
 435                pr_warn("KUAP cannot be disabled yet on 6xx when compiled in\n");
 436}
 437#endif
 438