LXR linux/arch/powerpc/platforms/cell/beat

   1/*
   2 * "Cell Reference Set" HTAB support.
   3 *
   4 * (C) Copyright 2006-2007 TOSHIBA CORPORATION
   5 *
   6 * This code is based on arch/powerpc/platforms/pseries/lpar.c:
   7 *  Copyright (C) 2001 Todd Inglett, IBM Corporation
   8 *
   9 * This program is free software; you can redistribute it and/or modify
  10 * it under the terms of the GNU General Public License as published by
  11 * the Free Software Foundation; either version 2 of the License, or
  12 * (at your option) any later version.
  13 *
  14 * This program is distributed in the hope that it will be useful,
  15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17 * GNU General Public License for more details.
  18 *
  19 * You should have received a copy of the GNU General Public License along
  20 * with this program; if not, write to the Free Software Foundation, Inc.,
  21 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  22 */
  23
  24#undef DEBUG_LOW
  25
  26#include <linux/kernel.h>
  27#include <linux/spinlock.h>
  28
  29#include <asm/mmu.h>
  30#include <asm/page.h>
  31#include <asm/pgtable.h>
  32#include <asm/machdep.h>
  33#include <asm/udbg.h>
  34
  35#include "beat_wrapper.h"
  36
  37#ifdef DEBUG_LOW
  38#define DBG_LOW(fmt...) do { udbg_printf(fmt); } while (0)
  39#else
  40#define DBG_LOW(fmt...) do { } while (0)
  41#endif
  42
  43static DEFINE_RAW_SPINLOCK(beat_htab_lock);
  44
  45static inline unsigned int beat_read_mask(unsigned hpte_group)
  46{
  47        unsigned long rmask = 0;
  48        u64 hpte_v[5];
  49
  50        beat_read_htab_entries(0, hpte_group + 0, hpte_v);
  51        if (!(hpte_v[0] & HPTE_V_BOLTED))
  52                rmask |= 0x8000;
  53        if (!(hpte_v[1] & HPTE_V_BOLTED))
  54                rmask |= 0x4000;
  55        if (!(hpte_v[2] & HPTE_V_BOLTED))
  56                rmask |= 0x2000;
  57        if (!(hpte_v[3] & HPTE_V_BOLTED))
  58                rmask |= 0x1000;
  59        beat_read_htab_entries(0, hpte_group + 4, hpte_v);
  60        if (!(hpte_v[0] & HPTE_V_BOLTED))
  61                rmask |= 0x0800;
  62        if (!(hpte_v[1] & HPTE_V_BOLTED))
  63                rmask |= 0x0400;
  64        if (!(hpte_v[2] & HPTE_V_BOLTED))
  65                rmask |= 0x0200;
  66        if (!(hpte_v[3] & HPTE_V_BOLTED))
  67                rmask |= 0x0100;
  68        hpte_group = ~hpte_group & (htab_hash_mask * HPTES_PER_GROUP);
  69        beat_read_htab_entries(0, hpte_group + 0, hpte_v);
  70        if (!(hpte_v[0] & HPTE_V_BOLTED))
  71                rmask |= 0x80;
  72        if (!(hpte_v[1] & HPTE_V_BOLTED))
  73                rmask |= 0x40;
  74        if (!(hpte_v[2] & HPTE_V_BOLTED))
  75                rmask |= 0x20;
  76        if (!(hpte_v[3] & HPTE_V_BOLTED))
  77                rmask |= 0x10;
  78        beat_read_htab_entries(0, hpte_group + 4, hpte_v);
  79        if (!(hpte_v[0] & HPTE_V_BOLTED))
  80                rmask |= 0x08;
  81        if (!(hpte_v[1] & HPTE_V_BOLTED))
  82                rmask |= 0x04;
  83        if (!(hpte_v[2] & HPTE_V_BOLTED))
  84                rmask |= 0x02;
  85        if (!(hpte_v[3] & HPTE_V_BOLTED))
  86                rmask |= 0x01;
  87        return rmask;
  88}
  89
  90static long beat_lpar_hpte_insert(unsigned long hpte_group,
  91                                  unsigned long va, unsigned long pa,
  92                                  unsigned long rflags, unsigned long vflags,
  93                                  int psize, int ssize)
  94{
  95        unsigned long lpar_rc;
  96        u64 hpte_v, hpte_r, slot;
  97
  98        /* same as iseries */
  99        if (vflags & HPTE_V_SECONDARY)
 100                return -1;
 101
 102        if (!(vflags & HPTE_V_BOLTED))
 103                DBG_LOW("hpte_insert(group=%lx, va=%016lx, pa=%016lx, "
 104                        "rflags=%lx, vflags=%lx, psize=%d)\n",
 105                hpte_group, va, pa, rflags, vflags, psize);
 106
 107        hpte_v = hpte_encode_v(va, psize, MMU_SEGSIZE_256M) |
 108                vflags | HPTE_V_VALID;
 109        hpte_r = hpte_encode_r(pa, psize) | rflags;
 110
 111        if (!(vflags & HPTE_V_BOLTED))
 112                DBG_LOW(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
 113
 114        if (rflags & _PAGE_NO_CACHE)
 115                hpte_r &= ~_PAGE_COHERENT;
 116
 117        raw_spin_lock(&beat_htab_lock);
 118        lpar_rc = beat_read_mask(hpte_group);
 119        if (lpar_rc == 0) {
 120                if (!(vflags & HPTE_V_BOLTED))
 121                        DBG_LOW(" full\n");
 122                raw_spin_unlock(&beat_htab_lock);
 123                return -1;
 124        }
 125
 126        lpar_rc = beat_insert_htab_entry(0, hpte_group, lpar_rc << 48,
 127                hpte_v, hpte_r, &slot);
 128        raw_spin_unlock(&beat_htab_lock);
 129
 130        /*
 131         * Since we try and ioremap PHBs we don't own, the pte insert
 132         * will fail. However we must catch the failure in hash_page
 133         * or we will loop forever, so return -2 in this case.
 134         */
 135        if (unlikely(lpar_rc != 0)) {
 136                if (!(vflags & HPTE_V_BOLTED))
 137                        DBG_LOW(" lpar err %lx\n", lpar_rc);
 138                return -2;
 139        }
 140        if (!(vflags & HPTE_V_BOLTED))
 141                DBG_LOW(" -> slot: %lx\n", slot);
 142
 143        /* We have to pass down the secondary bucket bit here as well */
 144        return (slot ^ hpte_group) & 15;
 145}
 146
 147static long beat_lpar_hpte_remove(unsigned long hpte_group)
 148{
 149        DBG_LOW("hpte_remove(group=%lx)\n", hpte_group);
 150        return -1;
 151}
 152
 153static unsigned long beat_lpar_hpte_getword0(unsigned long slot)
 154{
 155        unsigned long dword0;
 156        unsigned long lpar_rc;
 157        u64 dword[5];
 158
 159        lpar_rc = beat_read_htab_entries(0, slot & ~3UL, dword);
 160
 161        dword0 = dword[slot&3];
 162
 163        BUG_ON(lpar_rc != 0);
 164
 165        return dword0;
 166}
 167
 168static void beat_lpar_hptab_clear(void)
 169{
 170        unsigned long size_bytes = 1UL << ppc64_pft_size;
 171        unsigned long hpte_count = size_bytes >> 4;
 172        int i;
 173        u64 dummy0, dummy1;
 174
 175        /* TODO: Use bulk call */
 176        for (i = 0; i < hpte_count; i++)
 177                beat_write_htab_entry(0, i, 0, 0, -1UL, -1UL, &dummy0, &dummy1);
 178}
 179
 180/*
 181 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
 182 * the low 3 bits of flags happen to line up.  So no transform is needed.
 183 * We can probably optimize here and assume the high bits of newpp are
 184 * already zero.  For now I am paranoid.
 185 */
 186static long beat_lpar_hpte_updatepp(unsigned long slot,
 187                                    unsigned long newpp,
 188                                    unsigned long va,
 189                                    int psize, int ssize, int local)
 190{
 191        unsigned long lpar_rc;
 192        u64 dummy0, dummy1;
 193        unsigned long want_v;
 194
 195        want_v = hpte_encode_v(va, psize, MMU_SEGSIZE_256M);
 196
 197        DBG_LOW("    update: "
 198                "avpnv=%016lx, slot=%016lx, psize: %d, newpp %016lx ... ",
 199                want_v & HPTE_V_AVPN, slot, psize, newpp);
 200
 201        raw_spin_lock(&beat_htab_lock);
 202        dummy0 = beat_lpar_hpte_getword0(slot);
 203        if ((dummy0 & ~0x7FUL) != (want_v & ~0x7FUL)) {
 204                DBG_LOW("not found !\n");
 205                raw_spin_unlock(&beat_htab_lock);
 206                return -1;
 207        }
 208
 209        lpar_rc = beat_write_htab_entry(0, slot, 0, newpp, 0, 7, &dummy0,
 210                                        &dummy1);
 211        raw_spin_unlock(&beat_htab_lock);
 212        if (lpar_rc != 0 || dummy0 == 0) {
 213                DBG_LOW("not found !\n");
 214                return -1;
 215        }
 216
 217        DBG_LOW("ok %lx %lx\n", dummy0, dummy1);
 218
 219        BUG_ON(lpar_rc != 0);
 220
 221        return 0;
 222}
 223
 224static long beat_lpar_hpte_find(unsigned long va, int psize)
 225{
 226        unsigned long hash;
 227        unsigned long i, j;
 228        long slot;
 229        unsigned long want_v, hpte_v;
 230
 231        hash = hpt_hash(va, mmu_psize_defs[psize].shift, MMU_SEGSIZE_256M);
 232        want_v = hpte_encode_v(va, psize, MMU_SEGSIZE_256M);
 233
 234        for (j = 0; j < 2; j++) {
 235                slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 236                for (i = 0; i < HPTES_PER_GROUP; i++) {
 237                        hpte_v = beat_lpar_hpte_getword0(slot);
 238
 239                        if (HPTE_V_COMPARE(hpte_v, want_v)
 240                            && (hpte_v & HPTE_V_VALID)
 241                            && (!!(hpte_v & HPTE_V_SECONDARY) == j)) {
 242                                /* HPTE matches */
 243                                if (j)
 244                                        slot = -slot;
 245                                return slot;
 246                        }
 247                        ++slot;
 248                }
 249                hash = ~hash;
 250        }
 251
 252        return -1;
 253}
 254
 255static void beat_lpar_hpte_updateboltedpp(unsigned long newpp,
 256                                          unsigned long ea,
 257                                          int psize, int ssize)
 258{
 259        unsigned long lpar_rc, slot, vsid, va;
 260        u64 dummy0, dummy1;
 261
 262        vsid = get_kernel_vsid(ea, MMU_SEGSIZE_256M);
 263        va = (vsid << 28) | (ea & 0x0fffffff);
 264
 265        raw_spin_lock(&beat_htab_lock);
 266        slot = beat_lpar_hpte_find(va, psize);
 267        BUG_ON(slot == -1);
 268
 269        lpar_rc = beat_write_htab_entry(0, slot, 0, newpp, 0, 7,
 270                &dummy0, &dummy1);
 271        raw_spin_unlock(&beat_htab_lock);
 272
 273        BUG_ON(lpar_rc != 0);
 274}
 275
 276static void beat_lpar_hpte_invalidate(unsigned long slot, unsigned long va,
 277                                         int psize, int ssize, int local)
 278{
 279        unsigned long want_v;
 280        unsigned long lpar_rc;
 281        u64 dummy1, dummy2;
 282        unsigned long flags;
 283
 284        DBG_LOW("    inval : slot=%lx, va=%016lx, psize: %d, local: %d\n",
 285                slot, va, psize, local);
 286        want_v = hpte_encode_v(va, psize, MMU_SEGSIZE_256M);
 287
 288        raw_spin_lock_irqsave(&beat_htab_lock, flags);
 289        dummy1 = beat_lpar_hpte_getword0(slot);
 290
 291        if ((dummy1 & ~0x7FUL) != (want_v & ~0x7FUL)) {
 292                DBG_LOW("not found !\n");
 293                raw_spin_unlock_irqrestore(&beat_htab_lock, flags);
 294                return;
 295        }
 296
 297        lpar_rc = beat_write_htab_entry(0, slot, 0, 0, HPTE_V_VALID, 0,
 298                &dummy1, &dummy2);
 299        raw_spin_unlock_irqrestore(&beat_htab_lock, flags);
 300
 301        BUG_ON(lpar_rc != 0);
 302}
 303
 304void __init hpte_init_beat(void)
 305{
 306        ppc_md.hpte_invalidate  = beat_lpar_hpte_invalidate;
 307        ppc_md.hpte_updatepp    = beat_lpar_hpte_updatepp;
 308        ppc_md.hpte_updateboltedpp = beat_lpar_hpte_updateboltedpp;
 309        ppc_md.hpte_insert      = beat_lpar_hpte_insert;
 310        ppc_md.hpte_remove      = beat_lpar_hpte_remove;
 311        ppc_md.hpte_clear_all   = beat_lpar_hptab_clear;
 312}
 313
 314static long beat_lpar_hpte_insert_v3(unsigned long hpte_group,
 315                                  unsigned long va, unsigned long pa,
 316                                  unsigned long rflags, unsigned long vflags,
 317                                  int psize, int ssize)
 318{
 319        unsigned long lpar_rc;
 320        u64 hpte_v, hpte_r, slot;
 321
 322        /* same as iseries */
 323        if (vflags & HPTE_V_SECONDARY)
 324                return -1;
 325
 326        if (!(vflags & HPTE_V_BOLTED))
 327                DBG_LOW("hpte_insert(group=%lx, va=%016lx, pa=%016lx, "
 328                        "rflags=%lx, vflags=%lx, psize=%d)\n",
 329                hpte_group, va, pa, rflags, vflags, psize);
 330
 331        hpte_v = hpte_encode_v(va, psize, MMU_SEGSIZE_256M) |
 332                vflags | HPTE_V_VALID;
 333        hpte_r = hpte_encode_r(pa, psize) | rflags;
 334
 335        if (!(vflags & HPTE_V_BOLTED))
 336                DBG_LOW(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
 337
 338        if (rflags & _PAGE_NO_CACHE)
 339                hpte_r &= ~_PAGE_COHERENT;
 340
 341        /* insert into not-volted entry */
 342        lpar_rc = beat_insert_htab_entry3(0, hpte_group, hpte_v, hpte_r,
 343                HPTE_V_BOLTED, 0, &slot);
 344        /*
 345         * Since we try and ioremap PHBs we don't own, the pte insert
 346         * will fail. However we must catch the failure in hash_page
 347         * or we will loop forever, so return -2 in this case.
 348         */
 349        if (unlikely(lpar_rc != 0)) {
 350                if (!(vflags & HPTE_V_BOLTED))
 351                        DBG_LOW(" lpar err %lx\n", lpar_rc);
 352                return -2;
 353        }
 354        if (!(vflags & HPTE_V_BOLTED))
 355                DBG_LOW(" -> slot: %lx\n", slot);
 356
 357        /* We have to pass down the secondary bucket bit here as well */
 358        return (slot ^ hpte_group) & 15;
 359}
 360
 361/*
 362 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
 363 * the low 3 bits of flags happen to line up.  So no transform is needed.
 364 * We can probably optimize here and assume the high bits of newpp are
 365 * already zero.  For now I am paranoid.
 366 */
 367static long beat_lpar_hpte_updatepp_v3(unsigned long slot,
 368                                    unsigned long newpp,
 369                                    unsigned long va,
 370                                    int psize, int ssize, int local)
 371{
 372        unsigned long lpar_rc;
 373        unsigned long want_v;
 374        unsigned long pss;
 375
 376        want_v = hpte_encode_v(va, psize, MMU_SEGSIZE_256M);
 377        pss = (psize == MMU_PAGE_4K) ? -1UL : mmu_psize_defs[psize].penc;
 378
 379        DBG_LOW("    update: "
 380                "avpnv=%016lx, slot=%016lx, psize: %d, newpp %016lx ... ",
 381                want_v & HPTE_V_AVPN, slot, psize, newpp);
 382
 383        lpar_rc = beat_update_htab_permission3(0, slot, want_v, pss, 7, newpp);
 384
 385        if (lpar_rc == 0xfffffff7) {
 386                DBG_LOW("not found !\n");
 387                return -1;
 388        }
 389
 390        DBG_LOW("ok\n");
 391
 392        BUG_ON(lpar_rc != 0);
 393
 394        return 0;
 395}
 396
 397static void beat_lpar_hpte_invalidate_v3(unsigned long slot, unsigned long va,
 398                                         int psize, int ssize, int local)
 399{
 400        unsigned long want_v;
 401        unsigned long lpar_rc;
 402        unsigned long pss;
 403
 404        DBG_LOW("    inval : slot=%lx, va=%016lx, psize: %d, local: %d\n",
 405                slot, va, psize, local);
 406        want_v = hpte_encode_v(va, psize, MMU_SEGSIZE_256M);
 407        pss = (psize == MMU_PAGE_4K) ? -1UL : mmu_psize_defs[psize].penc;
 408
 409        lpar_rc = beat_invalidate_htab_entry3(0, slot, want_v, pss);
 410
 411        /* E_busy can be valid output: page may be already replaced */
 412        BUG_ON(lpar_rc != 0 && lpar_rc != 0xfffffff7);
 413}
 414
 415static int64_t _beat_lpar_hptab_clear_v3(void)
 416{
 417        return beat_clear_htab3(0);
 418}
 419
 420static void beat_lpar_hptab_clear_v3(void)
 421{
 422        _beat_lpar_hptab_clear_v3();
 423}
 424
 425void __init hpte_init_beat_v3(void)
 426{
 427        if (_beat_lpar_hptab_clear_v3() == 0) {
 428                ppc_md.hpte_invalidate  = beat_lpar_hpte_invalidate_v3;
 429                ppc_md.hpte_updatepp    = beat_lpar_hpte_updatepp_v3;
 430                ppc_md.hpte_updateboltedpp = beat_lpar_hpte_updateboltedpp;
 431                ppc_md.hpte_insert      = beat_lpar_hpte_insert_v3;
 432                ppc_md.hpte_remove      = beat_lpar_hpte_remove;
 433                ppc_md.hpte_clear_all   = beat_lpar_hptab_clear_v3;
 434        } else {
 435                ppc_md.hpte_invalidate  = beat_lpar_hpte_invalidate;
 436                ppc_md.hpte_updatepp    = beat_lpar_hpte_updatepp;
 437                ppc_md.hpte_updateboltedpp = beat_lpar_hpte_updateboltedpp;
 438                ppc_md.hpte_insert      = beat_lpar_hpte_insert;
 439                ppc_md.hpte_remove      = beat_lpar_hpte_remove;
 440                ppc_md.hpte_clear_all   = beat_lpar_hptab_clear;
 441        }
 442}
 443