arch/powerpc/platforms/pseries/lpar.c

   1 /*
   2  * pSeries_lpar.c
   3  * Copyright (C) 2001 Todd Inglett, IBM Corporation
   4  *
   5  * pSeries LPAR support.
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License as published by
   9  * the Free Software Foundation; either version 2 of the License, or
  10  * (at your option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program; if not, write to the Free Software
  19  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  20  */
  21
  22 /* Enables debugging of low-level hash table routines - careful! */
  23 #undef DEBUG
  24
  25 #include <linux/kernel.h>
  26 #include <linux/dma-mapping.h>
  27 #include <linux/console.h>
  28 #include <linux/export.h>
  29 #include <linux/jump_label.h>
  30 #include <asm/processor.h>
  31 #include <asm/mmu.h>
  32 #include <asm/page.h>
  33 #include <asm/pgtable.h>
  34 #include <asm/machdep.h>
  35 #include <asm/mmu_context.h>
  36 #include <asm/iommu.h>
  37 #include <asm/tlbflush.h>
  38 #include <asm/tlb.h>
  39 #include <asm/prom.h>
  40 #include <asm/cputable.h>
  41 #include <asm/udbg.h>
  42 #include <asm/smp.h>
  43 #include <asm/trace.h>
  44 #include <asm/firmware.h>
  45 #include <asm/plpar_wrappers.h>
  46 #include <asm/kexec.h>
  47 #include <asm/fadump.h>
  48 #include <asm/asm-prototypes.h>
  49
  50 #include "pseries.h"
  51
  52 /* Flag bits for H_BULK_REMOVE */
  53 #define HBR_REQUEST     0x4000000000000000UL
  54 #define HBR_RESPONSE    0x8000000000000000UL
  55 #define HBR_END         0xc000000000000000UL
  56 #define HBR_AVPN        0x0200000000000000UL
  57 #define HBR_ANDCOND     0x0100000000000000UL
  58
  59
  60 /* in hvCall.S */
  61 EXPORT_SYMBOL(plpar_hcall);
  62 EXPORT_SYMBOL(plpar_hcall9);
  63 EXPORT_SYMBOL(plpar_hcall_norets);
  64
  65 void vpa_init(int cpu)
  66 {
  67         int hwcpu = get_hard_smp_processor_id(cpu);
  68         unsigned long addr;
  69         long ret;
  70         struct paca_struct *pp;
  71         struct dtl_entry *dtl;
  72
  73         /*
  74          * The spec says it "may be problematic" if CPU x registers the VPA of
  75          * CPU y. We should never do that, but wail if we ever do.
  76          */
  77         WARN_ON(cpu != smp_processor_id());
  78
  79         if (cpu_has_feature(CPU_FTR_ALTIVEC))
  80                 lppaca_of(cpu).vmxregs_in_use = 1;
  81
  82         if (cpu_has_feature(CPU_FTR_ARCH_207S))
  83                 lppaca_of(cpu).ebb_regs_in_use = 1;
  84
  85         addr = __pa(&lppaca_of(cpu));
  86         ret = register_vpa(hwcpu, addr);
  87
  88         if (ret) {
  89                 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
  90                        "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
  91                 return;
  92         }
  93
  94 #ifdef CONFIG_PPC_STD_MMU_64
  95         /*
  96          * PAPR says this feature is SLB-Buffer but firmware never
  97          * reports that.  All SPLPAR support SLB shadow buffer.
  98          */
  99         if (!radix_enabled() && firmware_has_feature(FW_FEATURE_SPLPAR)) {
 100                 addr = __pa(paca[cpu].slb_shadow_ptr);
 101                 ret = register_slb_shadow(hwcpu, addr);
 102                 if (ret)
 103                         pr_err("WARNING: SLB shadow buffer registration for "
 104                                "cpu %d (hw %d) of area %lx failed with %ld\n",
 105                                cpu, hwcpu, addr, ret);
 106         }
 107 #endif /* CONFIG_PPC_STD_MMU_64 */
 108
 109         /*
 110          * Register dispatch trace log, if one has been allocated.
 111          */
 112         pp = &paca[cpu];
 113         dtl = pp->dispatch_log;
 114         if (dtl) {
 115                 pp->dtl_ridx = 0;
 116                 pp->dtl_curr = dtl;
 117                 lppaca_of(cpu).dtl_idx = 0;
 118
 119                 /* hypervisor reads buffer length from this field */
 120                 dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
 121                 ret = register_dtl(hwcpu, __pa(dtl));
 122                 if (ret)
 123                         pr_err("WARNING: DTL registration of cpu %d (hw %d) "
 124                                "failed with %ld\n", smp_processor_id(),
 125                                hwcpu, ret);
 126                 lppaca_of(cpu).dtl_enable_mask = 2;
 127         }
 128 }
 129
 130 #ifdef CONFIG_PPC_STD_MMU_64
 131
 132 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
 133                                      unsigned long vpn, unsigned long pa,
 134                                      unsigned long rflags, unsigned long vflags,
 135                                      int psize, int apsize, int ssize)
 136 {
 137         unsigned long lpar_rc;
 138         unsigned long flags;
 139         unsigned long slot;
 140         unsigned long hpte_v, hpte_r;
 141
 142         if (!(vflags & HPTE_V_BOLTED))
 143                 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
 144                          "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
 145                          hpte_group, vpn,  pa, rflags, vflags, psize);
 146
 147         hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
 148         hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
 149
 150         if (!(vflags & HPTE_V_BOLTED))
 151                 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
 152
 153         /* Now fill in the actual HPTE */
 154         /* Set CEC cookie to 0         */
 155         /* Zero page = 0               */
 156         /* I-cache Invalidate = 0      */
 157         /* I-cache synchronize = 0     */
 158         /* Exact = 0                   */
 159         flags = 0;
 160
 161         if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
 162                 flags |= H_COALESCE_CAND;
 163
 164         lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
 165         if (unlikely(lpar_rc == H_PTEG_FULL)) {
 166                 if (!(vflags & HPTE_V_BOLTED))
 167                         pr_devel(" full\n");
 168                 return -1;
 169         }
 170
 171         /*
 172          * Since we try and ioremap PHBs we don't own, the pte insert
 173          * will fail. However we must catch the failure in hash_page
 174          * or we will loop forever, so return -2 in this case.
 175          */
 176         if (unlikely(lpar_rc != H_SUCCESS)) {
 177                 if (!(vflags & HPTE_V_BOLTED))
 178                         pr_devel(" lpar err %ld\n", lpar_rc);
 179                 return -2;
 180         }
 181         if (!(vflags & HPTE_V_BOLTED))
 182                 pr_devel(" -> slot: %lu\n", slot & 7);
 183
 184         /* Because of iSeries, we have to pass down the secondary
 185          * bucket bit here as well
 186          */
 187         return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
 188 }
 189
 190 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
 191
 192 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
 193 {
 194         unsigned long slot_offset;
 195         unsigned long lpar_rc;
 196         int i;
 197         unsigned long dummy1, dummy2;
 198
 199         /* pick a random slot to start at */
 200         slot_offset = mftb() & 0x7;
 201
 202         for (i = 0; i < HPTES_PER_GROUP; i++) {
 203
 204                 /* don't remove a bolted entry */
 205                 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
 206                                            (0x1UL << 4), &dummy1, &dummy2);
 207                 if (lpar_rc == H_SUCCESS)
 208                         return i;
 209
 210                 /*
 211                  * The test for adjunct partition is performed before the
 212                  * ANDCOND test.  H_RESOURCE may be returned, so we need to
 213                  * check for that as well.
 214                  */
 215                 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
 216
 217                 slot_offset++;
 218                 slot_offset &= 0x7;
 219         }
 220
 221         return -1;
 222 }
 223
 224 static void manual_hpte_clear_all(void)
 225 {
 226         unsigned long size_bytes = 1UL << ppc64_pft_size;
 227         unsigned long hpte_count = size_bytes >> 4;
 228         struct {
 229                 unsigned long pteh;
 230                 unsigned long ptel;
 231         } ptes[4];
 232         long lpar_rc;
 233         unsigned long i, j;
 234
 235         /* Read in batches of 4,
 236          * invalidate only valid entries not in the VRMA
 237          * hpte_count will be a multiple of 4
 238          */
 239         for (i = 0; i < hpte_count; i += 4) {
 240                 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
 241                 if (lpar_rc != H_SUCCESS)
 242                         continue;
 243                 for (j = 0; j < 4; j++){
 244                         if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
 245                                 HPTE_V_VRMA_MASK)
 246                                 continue;
 247                         if (ptes[j].pteh & HPTE_V_VALID)
 248                                 plpar_pte_remove_raw(0, i + j, 0,
 249                                         &(ptes[j].pteh), &(ptes[j].ptel));
 250                 }
 251         }
 252 }
 253
 254 static int hcall_hpte_clear_all(void)
 255 {
 256         int rc;
 257
 258         do {
 259                 rc = plpar_hcall_norets(H_CLEAR_HPT);
 260         } while (rc == H_CONTINUE);
 261
 262         return rc;
 263 }
 264
 265 static void pseries_hpte_clear_all(void)
 266 {
 267         int rc;
 268
 269         rc = hcall_hpte_clear_all();
 270         if (rc != H_SUCCESS)
 271                 manual_hpte_clear_all();
 272
 273 #ifdef __LITTLE_ENDIAN__
 274         /*
 275          * Reset exceptions to big endian.
 276          *
 277          * FIXME this is a hack for kexec, we need to reset the exception
 278          * endian before starting the new kernel and this is a convenient place
 279          * to do it.
 280          *
 281          * This is also called on boot when a fadump happens. In that case we
 282          * must not change the exception endian mode.
 283          */
 284         if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active())
 285                 pseries_big_endian_exceptions();
 286 #endif
 287 }
 288
 289 /*
 290  * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
 291  * the low 3 bits of flags happen to line up.  So no transform is needed.
 292  * We can probably optimize here and assume the high bits of newpp are
 293  * already zero.  For now I am paranoid.
 294  */
 295 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
 296                                        unsigned long newpp,
 297                                        unsigned long vpn,
 298                                        int psize, int apsize,
 299                                        int ssize, unsigned long inv_flags)
 300 {
 301         unsigned long lpar_rc;
 302         unsigned long flags = (newpp & 7) | H_AVPN;
 303         unsigned long want_v;
 304
 305         want_v = hpte_encode_avpn(vpn, psize, ssize);
 306
 307         pr_devel("    update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
 308                  want_v, slot, flags, psize);
 309
 310         lpar_rc = plpar_pte_protect(flags, slot, want_v);
 311
 312         if (lpar_rc == H_NOT_FOUND) {
 313                 pr_devel("not found !\n");
 314                 return -1;
 315         }
 316
 317         pr_devel("ok\n");
 318
 319         BUG_ON(lpar_rc != H_SUCCESS);
 320
 321         return 0;
 322 }
 323
 324 static long __pSeries_lpar_hpte_find(unsigned long want_v, unsigned long hpte_group)
 325 {
 326         long lpar_rc;
 327         unsigned long i, j;
 328         struct {
 329                 unsigned long pteh;
 330                 unsigned long ptel;
 331         } ptes[4];
 332
 333         for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
 334
 335                 lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
 336                 if (lpar_rc != H_SUCCESS)
 337                         continue;
 338
 339                 for (j = 0; j < 4; j++) {
 340                         if (HPTE_V_COMPARE(ptes[j].pteh, want_v) &&
 341                             (ptes[j].pteh & HPTE_V_VALID))
 342                                 return i + j;
 343                 }
 344         }
 345
 346         return -1;
 347 }
 348
 349 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
 350 {
 351         long slot;
 352         unsigned long hash;
 353         unsigned long want_v;
 354         unsigned long hpte_group;
 355
 356         hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
 357         want_v = hpte_encode_avpn(vpn, psize, ssize);
 358
 359         /* Bolted entries are always in the primary group */
 360         hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 361         slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
 362         if (slot < 0)
 363                 return -1;
 364         return hpte_group + slot;
 365 }
 366
 367 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
 368                                              unsigned long ea,
 369                                              int psize, int ssize)
 370 {
 371         unsigned long vpn;
 372         unsigned long lpar_rc, slot, vsid, flags;
 373
 374         vsid = get_kernel_vsid(ea, ssize);
 375         vpn = hpt_vpn(ea, vsid, ssize);
 376
 377         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
 378         BUG_ON(slot == -1);
 379
 380         flags = newpp & 7;
 381         lpar_rc = plpar_pte_protect(flags, slot, 0);
 382
 383         BUG_ON(lpar_rc != H_SUCCESS);
 384 }
 385
 386 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
 387                                          int psize, int apsize,
 388                                          int ssize, int local)
 389 {
 390         unsigned long want_v;
 391         unsigned long lpar_rc;
 392         unsigned long dummy1, dummy2;
 393
 394         pr_devel("    inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
 395                  slot, vpn, psize, local);
 396
 397         want_v = hpte_encode_avpn(vpn, psize, ssize);
 398         lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
 399         if (lpar_rc == H_NOT_FOUND)
 400                 return;
 401
 402         BUG_ON(lpar_rc != H_SUCCESS);
 403 }
 404
 405 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 406 /*
 407  * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
 408  * to make sure that we avoid bouncing the hypervisor tlbie lock.
 409  */
 410 #define PPC64_HUGE_HPTE_BATCH 12
 411
 412 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
 413                                              unsigned long *vpn, int count,
 414                                              int psize, int ssize)
 415 {
 416         unsigned long param[PLPAR_HCALL9_BUFSIZE];
 417         int i = 0, pix = 0, rc;
 418         unsigned long flags = 0;
 419         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
 420
 421         if (lock_tlbie)
 422                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
 423
 424         for (i = 0; i < count; i++) {
 425
 426                 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 427                         pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
 428                                                      ssize, 0);
 429                 } else {
 430                         param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
 431                         param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
 432                         pix += 2;
 433                         if (pix == 8) {
 434                                 rc = plpar_hcall9(H_BULK_REMOVE, param,
 435                                                   param[0], param[1], param[2],
 436                                                   param[3], param[4], param[5],
 437                                                   param[6], param[7]);
 438                                 BUG_ON(rc != H_SUCCESS);
 439                                 pix = 0;
 440                         }
 441                 }
 442         }
 443         if (pix) {
 444                 param[pix] = HBR_END;
 445                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
 446                                   param[2], param[3], param[4], param[5],
 447                                   param[6], param[7]);
 448                 BUG_ON(rc != H_SUCCESS);
 449         }
 450
 451         if (lock_tlbie)
 452                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
 453 }
 454
 455 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
 456                                              unsigned long addr,
 457                                              unsigned char *hpte_slot_array,
 458                                              int psize, int ssize, int local)
 459 {
 460         int i, index = 0;
 461         unsigned long s_addr = addr;
 462         unsigned int max_hpte_count, valid;
 463         unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
 464         unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
 465         unsigned long shift, hidx, vpn = 0, hash, slot;
 466
 467         shift = mmu_psize_defs[psize].shift;
 468         max_hpte_count = 1U << (PMD_SHIFT - shift);
 469
 470         for (i = 0; i < max_hpte_count; i++) {
 471                 valid = hpte_valid(hpte_slot_array, i);
 472                 if (!valid)
 473                         continue;
 474                 hidx =  hpte_hash_index(hpte_slot_array, i);
 475
 476                 /* get the vpn */
 477                 addr = s_addr + (i * (1ul << shift));
 478                 vpn = hpt_vpn(addr, vsid, ssize);
 479                 hash = hpt_hash(vpn, shift, ssize);
 480                 if (hidx & _PTEIDX_SECONDARY)
 481                         hash = ~hash;
 482
 483                 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 484                 slot += hidx & _PTEIDX_GROUP_IX;
 485
 486                 slot_array[index] = slot;
 487                 vpn_array[index] = vpn;
 488                 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
 489                         /*
 490                          * Now do a bluk invalidate
 491                          */
 492                         __pSeries_lpar_hugepage_invalidate(slot_array,
 493                                                            vpn_array,
 494                                                            PPC64_HUGE_HPTE_BATCH,
 495                                                            psize, ssize);
 496                         index = 0;
 497                 } else
 498                         index++;
 499         }
 500         if (index)
 501                 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
 502                                                    index, psize, ssize);
 503 }
 504 #else
 505 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
 506                                              unsigned long addr,
 507                                              unsigned char *hpte_slot_array,
 508                                              int psize, int ssize, int local)
 509 {
 510         WARN(1, "%s called without THP support\n", __func__);
 511 }
 512 #endif
 513
 514 static int pSeries_lpar_hpte_removebolted(unsigned long ea,
 515                                           int psize, int ssize)
 516 {
 517         unsigned long vpn;
 518         unsigned long slot, vsid;
 519
 520         vsid = get_kernel_vsid(ea, ssize);
 521         vpn = hpt_vpn(ea, vsid, ssize);
 522
 523         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
 524         if (slot == -1)
 525                 return -ENOENT;
 526
 527         /*
 528          * lpar doesn't use the passed actual page size
 529          */
 530         pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
 531         return 0;
 532 }
 533
 534 /*
 535  * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
 536  * lock.
 537  */
 538 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
 539 {
 540         unsigned long vpn;
 541         unsigned long i, pix, rc;
 542         unsigned long flags = 0;
 543         struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
 544         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
 545         unsigned long param[PLPAR_HCALL9_BUFSIZE];
 546         unsigned long hash, index, shift, hidx, slot;
 547         real_pte_t pte;
 548         int psize, ssize;
 549
 550         if (lock_tlbie)
 551                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
 552
 553         psize = batch->psize;
 554         ssize = batch->ssize;
 555         pix = 0;
 556         for (i = 0; i < number; i++) {
 557                 vpn = batch->vpn[i];
 558                 pte = batch->pte[i];
 559                 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
 560                         hash = hpt_hash(vpn, shift, ssize);
 561                         hidx = __rpte_to_hidx(pte, index);
 562                         if (hidx & _PTEIDX_SECONDARY)
 563                                 hash = ~hash;
 564                         slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 565                         slot += hidx & _PTEIDX_GROUP_IX;
 566                         if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 567                                 /*
 568                                  * lpar doesn't use the passed actual page size
 569                                  */
 570                                 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
 571                                                              0, ssize, local);
 572                         } else {
 573                                 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
 574                                 param[pix+1] = hpte_encode_avpn(vpn, psize,
 575                                                                 ssize);
 576                                 pix += 2;
 577                                 if (pix == 8) {
 578                                         rc = plpar_hcall9(H_BULK_REMOVE, param,
 579                                                 param[0], param[1], param[2],
 580                                                 param[3], param[4], param[5],
 581                                                 param[6], param[7]);
 582                                         BUG_ON(rc != H_SUCCESS);
 583                                         pix = 0;
 584                                 }
 585                         }
 586                 } pte_iterate_hashed_end();
 587         }
 588         if (pix) {
 589                 param[pix] = HBR_END;
 590                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
 591                                   param[2], param[3], param[4], param[5],
 592                                   param[6], param[7]);
 593                 BUG_ON(rc != H_SUCCESS);
 594         }
 595
 596         if (lock_tlbie)
 597                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
 598 }
 599
 600 static int __init disable_bulk_remove(char *str)
 601 {
 602         if (strcmp(str, "off") == 0 &&
 603             firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 604                         printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
 605                         powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
 606         }
 607         return 1;
 608 }
 609
 610 __setup("bulk_remove=", disable_bulk_remove);
 611
 612 void __init hpte_init_pseries(void)
 613 {
 614         mmu_hash_ops.hpte_invalidate     = pSeries_lpar_hpte_invalidate;
 615         mmu_hash_ops.hpte_updatepp       = pSeries_lpar_hpte_updatepp;
 616         mmu_hash_ops.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
 617         mmu_hash_ops.hpte_insert         = pSeries_lpar_hpte_insert;
 618         mmu_hash_ops.hpte_remove         = pSeries_lpar_hpte_remove;
 619         mmu_hash_ops.hpte_removebolted   = pSeries_lpar_hpte_removebolted;
 620         mmu_hash_ops.flush_hash_range    = pSeries_lpar_flush_hash_range;
 621         mmu_hash_ops.hpte_clear_all      = pseries_hpte_clear_all;
 622         mmu_hash_ops.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
 623 }
 624
 625 #ifdef CONFIG_PPC_SMLPAR
 626 #define CMO_FREE_HINT_DEFAULT 1
 627 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
 628
 629 static int __init cmo_free_hint(char *str)
 630 {
 631         char *parm;
 632         parm = strstrip(str);
 633
 634         if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
 635                 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
 636                 cmo_free_hint_flag = 0;
 637                 return 1;
 638         }
 639
 640         cmo_free_hint_flag = 1;
 641         printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
 642
 643         if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
 644                 return 1;
 645
 646         return 0;
 647 }
 648
 649 __setup("cmo_free_hint=", cmo_free_hint);
 650
 651 static void pSeries_set_page_state(struct page *page, int order,
 652                                    unsigned long state)
 653 {
 654         int i, j;
 655         unsigned long cmo_page_sz, addr;
 656
 657         cmo_page_sz = cmo_get_page_size();
 658         addr = __pa((unsigned long)page_address(page));
 659
 660         for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
 661                 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
 662                         plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
 663         }
 664 }
 665
 666 void arch_free_page(struct page *page, int order)
 667 {
 668         if (radix_enabled())
 669                 return;
 670         if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
 671                 return;
 672
 673         pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
 674 }
 675 EXPORT_SYMBOL(arch_free_page);
 676
 677 #endif /* CONFIG_PPC_SMLPAR */
 678 #endif /* CONFIG_PPC_STD_MMU_64 */
 679
 680 #ifdef CONFIG_TRACEPOINTS
 681 #ifdef HAVE_JUMP_LABEL
 682 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
 683
 684 void hcall_tracepoint_regfunc(void)
 685 {
 686         static_key_slow_inc(&hcall_tracepoint_key);
 687 }
 688
 689 void hcall_tracepoint_unregfunc(void)
 690 {
 691         static_key_slow_dec(&hcall_tracepoint_key);
 692 }
 693 #else
 694 /*
 695  * We optimise our hcall path by placing hcall_tracepoint_refcount
 696  * directly in the TOC so we can check if the hcall tracepoints are
 697  * enabled via a single load.
 698  */
 699
 700 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
 701 extern long hcall_tracepoint_refcount;
 702
 703 void hcall_tracepoint_regfunc(void)
 704 {
 705         hcall_tracepoint_refcount++;
 706 }
 707
 708 void hcall_tracepoint_unregfunc(void)
 709 {
 710         hcall_tracepoint_refcount--;
 711 }
 712 #endif
 713
 714 /*
 715  * Since the tracing code might execute hcalls we need to guard against
 716  * recursion. One example of this are spinlocks calling H_YIELD on
 717  * shared processor partitions.
 718  */
 719 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
 720
 721
 722 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
 723 {
 724         unsigned long flags;
 725         unsigned int *depth;
 726
 727         /*
 728          * We cannot call tracepoints inside RCU idle regions which
 729          * means we must not trace H_CEDE.
 730          */
 731         if (opcode == H_CEDE)
 732                 return;
 733
 734         local_irq_save(flags);
 735
 736         depth = this_cpu_ptr(&hcall_trace_depth);
 737
 738         if (*depth)
 739                 goto out;
 740
 741         (*depth)++;
 742         preempt_disable();
 743         trace_hcall_entry(opcode, args);
 744         (*depth)--;
 745
 746 out:
 747         local_irq_restore(flags);
 748 }
 749
 750 void __trace_hcall_exit(long opcode, unsigned long retval,
 751                         unsigned long *retbuf)
 752 {
 753         unsigned long flags;
 754         unsigned int *depth;
 755
 756         if (opcode == H_CEDE)
 757                 return;
 758
 759         local_irq_save(flags);
 760
 761         depth = this_cpu_ptr(&hcall_trace_depth);
 762
 763         if (*depth)
 764                 goto out;
 765
 766         (*depth)++;
 767         trace_hcall_exit(opcode, retval, retbuf);
 768         preempt_enable();
 769         (*depth)--;
 770
 771 out:
 772         local_irq_restore(flags);
 773 }
 774 #endif
 775
 776 /**
 777  * h_get_mpp
 778  * H_GET_MPP hcall returns info in 7 parms
 779  */
 780 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
 781 {
 782         int rc;
 783         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
 784
 785         rc = plpar_hcall9(H_GET_MPP, retbuf);
 786
 787         mpp_data->entitled_mem = retbuf[0];
 788         mpp_data->mapped_mem = retbuf[1];
 789
 790         mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
 791         mpp_data->pool_num = retbuf[2] & 0xffff;
 792
 793         mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
 794         mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
 795         mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
 796
 797         mpp_data->pool_size = retbuf[4];
 798         mpp_data->loan_request = retbuf[5];
 799         mpp_data->backing_mem = retbuf[6];
 800
 801         return rc;
 802 }
 803 EXPORT_SYMBOL(h_get_mpp);
 804
 805 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
 806 {
 807         int rc;
 808         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
 809
 810         rc = plpar_hcall9(H_GET_MPP_X, retbuf);
 811
 812         mpp_x_data->coalesced_bytes = retbuf[0];
 813         mpp_x_data->pool_coalesced_bytes = retbuf[1];
 814         mpp_x_data->pool_purr_cycles = retbuf[2];
 815         mpp_x_data->pool_spurr_cycles = retbuf[3];
 816
 817         return rc;
 818 }