arch/powerpc/platforms/cell/beat_htab.c

   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
  43 static DEFINE_SPINLOCK(beat_htab_lock);
  44
  45 static inline unsigned int beat_read_mask(unsigned hpte_group)
  46 {
  47         unsigned long hpte_v[5];
  48         unsigned long rmask = 0;
  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
  90 static 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         unsigned long slot;
  97         unsigned long hpte_v, hpte_r;
  98
  99         /* same as iseries */
 100         if (vflags & HPTE_V_SECONDARY)
 101                 return -1;
 102
 103         if (!(vflags & HPTE_V_BOLTED))
 104                 DBG_LOW("hpte_insert(group=%lx, va=%016lx, pa=%016lx, "
 105                         "rflags=%lx, vflags=%lx, psize=%d)\n",
 106                 hpte_group, va, pa, rflags, vflags, psize);
 107
 108         hpte_v = hpte_encode_v(va, psize, MMU_SEGSIZE_256M) |
 109                 vflags | HPTE_V_VALID;
 110         hpte_r = hpte_encode_r(pa, psize) | rflags;
 111
 112         if (!(vflags & HPTE_V_BOLTED))
 113                 DBG_LOW(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
 114
 115         if (rflags & _PAGE_NO_CACHE)
 116                 hpte_r &= ~_PAGE_COHERENT;
 117
 118         spin_lock(&beat_htab_lock);
 119         lpar_rc = beat_read_mask(hpte_group);
 120         if (lpar_rc == 0) {
 121                 if (!(vflags & HPTE_V_BOLTED))
 122                         DBG_LOW(" full\n");
 123                 spin_unlock(&beat_htab_lock);
 124                 return -1;
 125         }
 126
 127         lpar_rc = beat_insert_htab_entry(0, hpte_group, lpar_rc << 48,
 128                 hpte_v, hpte_r, &slot);
 129         spin_unlock(&beat_htab_lock);
 130
 131         /*
 132          * Since we try and ioremap PHBs we don't own, the pte insert
 133          * will fail. However we must catch the failure in hash_page
 134          * or we will loop forever, so return -2 in this case.
 135          */
 136         if (unlikely(lpar_rc != 0)) {
 137                 if (!(vflags & HPTE_V_BOLTED))
 138                         DBG_LOW(" lpar err %lx\n", lpar_rc);
 139                 return -2;
 140         }
 141         if (!(vflags & HPTE_V_BOLTED))
 142                 DBG_LOW(" -> slot: %lx\n", slot);
 143
 144         /* We have to pass down the secondary bucket bit here as well */
 145         return (slot ^ hpte_group) & 15;
 146 }
 147
 148 static long beat_lpar_hpte_remove(unsigned long hpte_group)
 149 {
 150         DBG_LOW("hpte_remove(group=%lx)\n", hpte_group);
 151         return -1;
 152 }
 153
 154 static unsigned long beat_lpar_hpte_getword0(unsigned long slot)
 155 {
 156         unsigned long dword0, dword[5];
 157         unsigned long lpar_rc;
 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
 168 static 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         unsigned long 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  */
 186 static 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         unsigned long dummy0, dummy1, want_v;
 193
 194         want_v = hpte_encode_v(va, psize, MMU_SEGSIZE_256M);
 195
 196         DBG_LOW("    update: "
 197                 "avpnv=%016lx, slot=%016lx, psize: %d, newpp %016lx ... ",
 198                 want_v & HPTE_V_AVPN, slot, psize, newpp);
 199
 200         spin_lock(&beat_htab_lock);
 201         dummy0 = beat_lpar_hpte_getword0(slot);
 202         if ((dummy0 & ~0x7FUL) != (want_v & ~0x7FUL)) {
 203                 DBG_LOW("not found !\n");
 204                 spin_unlock(&beat_htab_lock);
 205                 return -1;
 206         }
 207
 208         lpar_rc = beat_write_htab_entry(0, slot, 0, newpp, 0, 7, &dummy0,
 209                                         &dummy1);
 210         spin_unlock(&beat_htab_lock);
 211         if (lpar_rc != 0 || dummy0 == 0) {
 212                 DBG_LOW("not found !\n");
 213                 return -1;
 214         }
 215
 216         DBG_LOW("ok %lx %lx\n", dummy0, dummy1);
 217
 218         BUG_ON(lpar_rc != 0);
 219
 220         return 0;
 221 }
 222
 223 static long beat_lpar_hpte_find(unsigned long va, int psize)
 224 {
 225         unsigned long hash;
 226         unsigned long i, j;
 227         long slot;
 228         unsigned long want_v, hpte_v;
 229
 230         hash = hpt_hash(va, mmu_psize_defs[psize].shift, MMU_SEGSIZE_256M);
 231         want_v = hpte_encode_v(va, psize, MMU_SEGSIZE_256M);
 232
 233         for (j = 0; j < 2; j++) {
 234                 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 235                 for (i = 0; i < HPTES_PER_GROUP; i++) {
 236                         hpte_v = beat_lpar_hpte_getword0(slot);
 237
 238                         if (HPTE_V_COMPARE(hpte_v, want_v)
 239                             && (hpte_v & HPTE_V_VALID)
 240                             && (!!(hpte_v & HPTE_V_SECONDARY) == j)) {
 241                                 /* HPTE matches */
 242                                 if (j)
 243                                         slot = -slot;
 244                                 return slot;
 245                         }
 246                         ++slot;
 247                 }
 248                 hash = ~hash;
 249         }
 250
 251         return -1;
 252 }
 253
 254 static void beat_lpar_hpte_updateboltedpp(unsigned long newpp,
 255                                           unsigned long ea,
 256                                           int psize, int ssize)
 257 {
 258         unsigned long lpar_rc, slot, vsid, va, dummy0, dummy1;
 259
 260         vsid = get_kernel_vsid(ea, MMU_SEGSIZE_256M);
 261         va = (vsid << 28) | (ea & 0x0fffffff);
 262
 263         spin_lock(&beat_htab_lock);
 264         slot = beat_lpar_hpte_find(va, psize);
 265         BUG_ON(slot == -1);
 266
 267         lpar_rc = beat_write_htab_entry(0, slot, 0, newpp, 0, 7,
 268                 &dummy0, &dummy1);
 269         spin_unlock(&beat_htab_lock);
 270
 271         BUG_ON(lpar_rc != 0);
 272 }
 273
 274 static void beat_lpar_hpte_invalidate(unsigned long slot, unsigned long va,
 275                                          int psize, int ssize, int local)
 276 {
 277         unsigned long want_v;
 278         unsigned long lpar_rc;
 279         unsigned long dummy1, dummy2;
 280         unsigned long flags;
 281
 282         DBG_LOW("    inval : slot=%lx, va=%016lx, psize: %d, local: %d\n",
 283                 slot, va, psize, local);
 284         want_v = hpte_encode_v(va, psize, MMU_SEGSIZE_256M);
 285
 286         spin_lock_irqsave(&beat_htab_lock, flags);
 287         dummy1 = beat_lpar_hpte_getword0(slot);
 288
 289         if ((dummy1 & ~0x7FUL) != (want_v & ~0x7FUL)) {
 290                 DBG_LOW("not found !\n");
 291                 spin_unlock_irqrestore(&beat_htab_lock, flags);
 292                 return;
 293         }
 294
 295         lpar_rc = beat_write_htab_entry(0, slot, 0, 0, HPTE_V_VALID, 0,
 296                 &dummy1, &dummy2);
 297         spin_unlock_irqrestore(&beat_htab_lock, flags);
 298
 299         BUG_ON(lpar_rc != 0);
 300 }
 301
 302 void __init hpte_init_beat(void)
 303 {
 304         ppc_md.hpte_invalidate  = beat_lpar_hpte_invalidate;
 305         ppc_md.hpte_updatepp    = beat_lpar_hpte_updatepp;
 306         ppc_md.hpte_updateboltedpp = beat_lpar_hpte_updateboltedpp;
 307         ppc_md.hpte_insert      = beat_lpar_hpte_insert;
 308         ppc_md.hpte_remove      = beat_lpar_hpte_remove;
 309         ppc_md.hpte_clear_all   = beat_lpar_hptab_clear;
 310 }
 311
 312 static long beat_lpar_hpte_insert_v3(unsigned long hpte_group,
 313                                   unsigned long va, unsigned long pa,
 314                                   unsigned long rflags, unsigned long vflags,
 315                                   int psize, int ssize)
 316 {
 317         unsigned long lpar_rc;
 318         unsigned long slot;
 319         unsigned long hpte_v, hpte_r;
 320
 321         /* same as iseries */
 322         if (vflags & HPTE_V_SECONDARY)
 323                 return -1;
 324
 325         if (!(vflags & HPTE_V_BOLTED))
 326                 DBG_LOW("hpte_insert(group=%lx, va=%016lx, pa=%016lx, "
 327                         "rflags=%lx, vflags=%lx, psize=%d)\n",
 328                 hpte_group, va, pa, rflags, vflags, psize);
 329
 330         hpte_v = hpte_encode_v(va, psize, MMU_SEGSIZE_256M) |
 331                 vflags | HPTE_V_VALID;
 332         hpte_r = hpte_encode_r(pa, psize) | rflags;
 333
 334         if (!(vflags & HPTE_V_BOLTED))
 335                 DBG_LOW(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
 336
 337         if (rflags & _PAGE_NO_CACHE)
 338                 hpte_r &= ~_PAGE_COHERENT;
 339
 340         /* insert into not-volted entry */
 341         lpar_rc = beat_insert_htab_entry3(0, hpte_group, hpte_v, hpte_r,
 342                 HPTE_V_BOLTED, 0, &slot);
 343         /*
 344          * Since we try and ioremap PHBs we don't own, the pte insert
 345          * will fail. However we must catch the failure in hash_page
 346          * or we will loop forever, so return -2 in this case.
 347          */
 348         if (unlikely(lpar_rc != 0)) {
 349                 if (!(vflags & HPTE_V_BOLTED))
 350                         DBG_LOW(" lpar err %lx\n", lpar_rc);
 351                 return -2;
 352         }
 353         if (!(vflags & HPTE_V_BOLTED))
 354                 DBG_LOW(" -> slot: %lx\n", slot);
 355
 356         /* We have to pass down the secondary bucket bit here as well */
 357         return (slot ^ hpte_group) & 15;
 358 }
 359
 360 /*
 361  * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
 362  * the low 3 bits of flags happen to line up.  So no transform is needed.
 363  * We can probably optimize here and assume the high bits of newpp are
 364  * already zero.  For now I am paranoid.
 365  */
 366 static long beat_lpar_hpte_updatepp_v3(unsigned long slot,
 367                                     unsigned long newpp,
 368                                     unsigned long va,
 369                                     int psize, int ssize, int local)
 370 {
 371         unsigned long lpar_rc;
 372         unsigned long want_v;
 373         unsigned long pss;
 374
 375         want_v = hpte_encode_v(va, psize, MMU_SEGSIZE_256M);
 376         pss = (psize == MMU_PAGE_4K) ? -1UL : mmu_psize_defs[psize].penc;
 377
 378         DBG_LOW("    update: "
 379                 "avpnv=%016lx, slot=%016lx, psize: %d, newpp %016lx ... ",
 380                 want_v & HPTE_V_AVPN, slot, psize, newpp);
 381
 382         lpar_rc = beat_update_htab_permission3(0, slot, want_v, pss, 7, newpp);
 383
 384         if (lpar_rc == 0xfffffff7) {
 385                 DBG_LOW("not found !\n");
 386                 return -1;
 387         }
 388
 389         DBG_LOW("ok\n");
 390
 391         BUG_ON(lpar_rc != 0);
 392
 393         return 0;
 394 }
 395
 396 static void beat_lpar_hpte_invalidate_v3(unsigned long slot, unsigned long va,
 397                                          int psize, int ssize, int local)
 398 {
 399         unsigned long want_v;
 400         unsigned long lpar_rc;
 401         unsigned long pss;
 402
 403         DBG_LOW("    inval : slot=%lx, va=%016lx, psize: %d, local: %d\n",
 404                 slot, va, psize, local);
 405         want_v = hpte_encode_v(va, psize, MMU_SEGSIZE_256M);
 406         pss = (psize == MMU_PAGE_4K) ? -1UL : mmu_psize_defs[psize].penc;
 407
 408         lpar_rc = beat_invalidate_htab_entry3(0, slot, want_v, pss);
 409
 410         /* E_busy can be valid output: page may be already replaced */
 411         BUG_ON(lpar_rc != 0 && lpar_rc != 0xfffffff7);
 412 }
 413
 414 static int64_t _beat_lpar_hptab_clear_v3(void)
 415 {
 416         return beat_clear_htab3(0);
 417 }
 418
 419 static void beat_lpar_hptab_clear_v3(void)
 420 {
 421         _beat_lpar_hptab_clear_v3();
 422 }
 423
 424 void __init hpte_init_beat_v3(void)
 425 {
 426         if (_beat_lpar_hptab_clear_v3() == 0) {
 427                 ppc_md.hpte_invalidate  = beat_lpar_hpte_invalidate_v3;
 428                 ppc_md.hpte_updatepp    = beat_lpar_hpte_updatepp_v3;
 429                 ppc_md.hpte_updateboltedpp = beat_lpar_hpte_updateboltedpp;
 430                 ppc_md.hpte_insert      = beat_lpar_hpte_insert_v3;
 431                 ppc_md.hpte_remove      = beat_lpar_hpte_remove;
 432                 ppc_md.hpte_clear_all   = beat_lpar_hptab_clear_v3;
 433         } else {
 434                 ppc_md.hpte_invalidate  = beat_lpar_hpte_invalidate;
 435                 ppc_md.hpte_updatepp    = beat_lpar_hpte_updatepp;
 436                 ppc_md.hpte_updateboltedpp = beat_lpar_hpte_updateboltedpp;
 437                 ppc_md.hpte_insert      = beat_lpar_hpte_insert;
 438                 ppc_md.hpte_remove      = beat_lpar_hpte_remove;
 439                 ppc_md.hpte_clear_all   = beat_lpar_hptab_clear;
 440         }
 441 }