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, ssize) | 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 pSeries_lpar_hptab_clear(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 #ifdef __LITTLE_ENDIAN__
 254         /*
 255          * Reset exceptions to big endian.
 256          *
 257          * FIXME this is a hack for kexec, we need to reset the exception
 258          * endian before starting the new kernel and this is a convenient place
 259          * to do it.
 260          *
 261          * This is also called on boot when a fadump happens. In that case we
 262          * must not change the exception endian mode.
 263          */
 264         if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active())
 265                 pseries_big_endian_exceptions();
 266 #endif
 267 }
 268
 269 /*
 270  * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
 271  * the low 3 bits of flags happen to line up.  So no transform is needed.
 272  * We can probably optimize here and assume the high bits of newpp are
 273  * already zero.  For now I am paranoid.
 274  */
 275 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
 276                                        unsigned long newpp,
 277                                        unsigned long vpn,
 278                                        int psize, int apsize,
 279                                        int ssize, unsigned long inv_flags)
 280 {
 281         unsigned long lpar_rc;
 282         unsigned long flags = (newpp & 7) | H_AVPN;
 283         unsigned long want_v;
 284
 285         want_v = hpte_encode_avpn(vpn, psize, ssize);
 286
 287         pr_devel("    update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
 288                  want_v, slot, flags, psize);
 289
 290         lpar_rc = plpar_pte_protect(flags, slot, want_v);
 291
 292         if (lpar_rc == H_NOT_FOUND) {
 293                 pr_devel("not found !\n");
 294                 return -1;
 295         }
 296
 297         pr_devel("ok\n");
 298
 299         BUG_ON(lpar_rc != H_SUCCESS);
 300
 301         return 0;
 302 }
 303
 304 static long __pSeries_lpar_hpte_find(unsigned long want_v, unsigned long hpte_group)
 305 {
 306         long lpar_rc;
 307         unsigned long i, j;
 308         struct {
 309                 unsigned long pteh;
 310                 unsigned long ptel;
 311         } ptes[4];
 312
 313         for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
 314
 315                 lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
 316                 if (lpar_rc != H_SUCCESS)
 317                         continue;
 318
 319                 for (j = 0; j < 4; j++) {
 320                         if (HPTE_V_COMPARE(ptes[j].pteh, want_v) &&
 321                             (ptes[j].pteh & HPTE_V_VALID))
 322                                 return i + j;
 323                 }
 324         }
 325
 326         return -1;
 327 }
 328
 329 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
 330 {
 331         long slot;
 332         unsigned long hash;
 333         unsigned long want_v;
 334         unsigned long hpte_group;
 335
 336         hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
 337         want_v = hpte_encode_avpn(vpn, psize, ssize);
 338
 339         /* Bolted entries are always in the primary group */
 340         hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 341         slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
 342         if (slot < 0)
 343                 return -1;
 344         return hpte_group + slot;
 345 }
 346
 347 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
 348                                              unsigned long ea,
 349                                              int psize, int ssize)
 350 {
 351         unsigned long vpn;
 352         unsigned long lpar_rc, slot, vsid, flags;
 353
 354         vsid = get_kernel_vsid(ea, ssize);
 355         vpn = hpt_vpn(ea, vsid, ssize);
 356
 357         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
 358         BUG_ON(slot == -1);
 359
 360         flags = newpp & 7;
 361         lpar_rc = plpar_pte_protect(flags, slot, 0);
 362
 363         BUG_ON(lpar_rc != H_SUCCESS);
 364 }
 365
 366 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
 367                                          int psize, int apsize,
 368                                          int ssize, int local)
 369 {
 370         unsigned long want_v;
 371         unsigned long lpar_rc;
 372         unsigned long dummy1, dummy2;
 373
 374         pr_devel("    inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
 375                  slot, vpn, psize, local);
 376
 377         want_v = hpte_encode_avpn(vpn, psize, ssize);
 378         lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
 379         if (lpar_rc == H_NOT_FOUND)
 380                 return;
 381
 382         BUG_ON(lpar_rc != H_SUCCESS);
 383 }
 384
 385 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 386 /*
 387  * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
 388  * to make sure that we avoid bouncing the hypervisor tlbie lock.
 389  */
 390 #define PPC64_HUGE_HPTE_BATCH 12
 391
 392 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
 393                                              unsigned long *vpn, int count,
 394                                              int psize, int ssize)
 395 {
 396         unsigned long param[PLPAR_HCALL9_BUFSIZE];
 397         int i = 0, pix = 0, rc;
 398         unsigned long flags = 0;
 399         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
 400
 401         if (lock_tlbie)
 402                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
 403
 404         for (i = 0; i < count; i++) {
 405
 406                 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 407                         pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
 408                                                      ssize, 0);
 409                 } else {
 410                         param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
 411                         param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
 412                         pix += 2;
 413                         if (pix == 8) {
 414                                 rc = plpar_hcall9(H_BULK_REMOVE, param,
 415                                                   param[0], param[1], param[2],
 416                                                   param[3], param[4], param[5],
 417                                                   param[6], param[7]);
 418                                 BUG_ON(rc != H_SUCCESS);
 419                                 pix = 0;
 420                         }
 421                 }
 422         }
 423         if (pix) {
 424                 param[pix] = HBR_END;
 425                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
 426                                   param[2], param[3], param[4], param[5],
 427                                   param[6], param[7]);
 428                 BUG_ON(rc != H_SUCCESS);
 429         }
 430
 431         if (lock_tlbie)
 432                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
 433 }
 434
 435 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
 436                                              unsigned long addr,
 437                                              unsigned char *hpte_slot_array,
 438                                              int psize, int ssize, int local)
 439 {
 440         int i, index = 0;
 441         unsigned long s_addr = addr;
 442         unsigned int max_hpte_count, valid;
 443         unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
 444         unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
 445         unsigned long shift, hidx, vpn = 0, hash, slot;
 446
 447         shift = mmu_psize_defs[psize].shift;
 448         max_hpte_count = 1U << (PMD_SHIFT - shift);
 449
 450         for (i = 0; i < max_hpte_count; i++) {
 451                 valid = hpte_valid(hpte_slot_array, i);
 452                 if (!valid)
 453                         continue;
 454                 hidx =  hpte_hash_index(hpte_slot_array, i);
 455
 456                 /* get the vpn */
 457                 addr = s_addr + (i * (1ul << shift));
 458                 vpn = hpt_vpn(addr, vsid, ssize);
 459                 hash = hpt_hash(vpn, shift, ssize);
 460                 if (hidx & _PTEIDX_SECONDARY)
 461                         hash = ~hash;
 462
 463                 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 464                 slot += hidx & _PTEIDX_GROUP_IX;
 465
 466                 slot_array[index] = slot;
 467                 vpn_array[index] = vpn;
 468                 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
 469                         /*
 470                          * Now do a bluk invalidate
 471                          */
 472                         __pSeries_lpar_hugepage_invalidate(slot_array,
 473                                                            vpn_array,
 474                                                            PPC64_HUGE_HPTE_BATCH,
 475                                                            psize, ssize);
 476                         index = 0;
 477                 } else
 478                         index++;
 479         }
 480         if (index)
 481                 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
 482                                                    index, psize, ssize);
 483 }
 484 #else
 485 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
 486                                              unsigned long addr,
 487                                              unsigned char *hpte_slot_array,
 488                                              int psize, int ssize, int local)
 489 {
 490         WARN(1, "%s called without THP support\n", __func__);
 491 }
 492 #endif
 493
 494 static int pSeries_lpar_hpte_removebolted(unsigned long ea,
 495                                           int psize, int ssize)
 496 {
 497         unsigned long vpn;
 498         unsigned long slot, vsid;
 499
 500         vsid = get_kernel_vsid(ea, ssize);
 501         vpn = hpt_vpn(ea, vsid, ssize);
 502
 503         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
 504         if (slot == -1)
 505                 return -ENOENT;
 506
 507         /*
 508          * lpar doesn't use the passed actual page size
 509          */
 510         pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
 511         return 0;
 512 }
 513
 514 /*
 515  * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
 516  * lock.
 517  */
 518 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
 519 {
 520         unsigned long vpn;
 521         unsigned long i, pix, rc;
 522         unsigned long flags = 0;
 523         struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
 524         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
 525         unsigned long param[PLPAR_HCALL9_BUFSIZE];
 526         unsigned long hash, index, shift, hidx, slot;
 527         real_pte_t pte;
 528         int psize, ssize;
 529
 530         if (lock_tlbie)
 531                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
 532
 533         psize = batch->psize;
 534         ssize = batch->ssize;
 535         pix = 0;
 536         for (i = 0; i < number; i++) {
 537                 vpn = batch->vpn[i];
 538                 pte = batch->pte[i];
 539                 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
 540                         hash = hpt_hash(vpn, shift, ssize);
 541                         hidx = __rpte_to_hidx(pte, index);
 542                         if (hidx & _PTEIDX_SECONDARY)
 543                                 hash = ~hash;
 544                         slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 545                         slot += hidx & _PTEIDX_GROUP_IX;
 546                         if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 547                                 /*
 548                                  * lpar doesn't use the passed actual page size
 549                                  */
 550                                 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
 551                                                              0, ssize, local);
 552                         } else {
 553                                 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
 554                                 param[pix+1] = hpte_encode_avpn(vpn, psize,
 555                                                                 ssize);
 556                                 pix += 2;
 557                                 if (pix == 8) {
 558                                         rc = plpar_hcall9(H_BULK_REMOVE, param,
 559                                                 param[0], param[1], param[2],
 560                                                 param[3], param[4], param[5],
 561                                                 param[6], param[7]);
 562                                         BUG_ON(rc != H_SUCCESS);
 563                                         pix = 0;
 564                                 }
 565                         }
 566                 } pte_iterate_hashed_end();
 567         }
 568         if (pix) {
 569                 param[pix] = HBR_END;
 570                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
 571                                   param[2], param[3], param[4], param[5],
 572                                   param[6], param[7]);
 573                 BUG_ON(rc != H_SUCCESS);
 574         }
 575
 576         if (lock_tlbie)
 577                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
 578 }
 579
 580 static int __init disable_bulk_remove(char *str)
 581 {
 582         if (strcmp(str, "off") == 0 &&
 583             firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 584                         printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
 585                         powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
 586         }
 587         return 1;
 588 }
 589
 590 __setup("bulk_remove=", disable_bulk_remove);
 591
 592 void __init hpte_init_pseries(void)
 593 {
 594         mmu_hash_ops.hpte_invalidate     = pSeries_lpar_hpte_invalidate;
 595         mmu_hash_ops.hpte_updatepp       = pSeries_lpar_hpte_updatepp;
 596         mmu_hash_ops.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
 597         mmu_hash_ops.hpte_insert         = pSeries_lpar_hpte_insert;
 598         mmu_hash_ops.hpte_remove         = pSeries_lpar_hpte_remove;
 599         mmu_hash_ops.hpte_removebolted   = pSeries_lpar_hpte_removebolted;
 600         mmu_hash_ops.flush_hash_range    = pSeries_lpar_flush_hash_range;
 601         mmu_hash_ops.hpte_clear_all      = pSeries_lpar_hptab_clear;
 602         mmu_hash_ops.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
 603 }
 604
 605 #ifdef CONFIG_PPC_SMLPAR
 606 #define CMO_FREE_HINT_DEFAULT 1
 607 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
 608
 609 static int __init cmo_free_hint(char *str)
 610 {
 611         char *parm;
 612         parm = strstrip(str);
 613
 614         if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
 615                 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
 616                 cmo_free_hint_flag = 0;
 617                 return 1;
 618         }
 619
 620         cmo_free_hint_flag = 1;
 621         printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
 622
 623         if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
 624                 return 1;
 625
 626         return 0;
 627 }
 628
 629 __setup("cmo_free_hint=", cmo_free_hint);
 630
 631 static void pSeries_set_page_state(struct page *page, int order,
 632                                    unsigned long state)
 633 {
 634         int i, j;
 635         unsigned long cmo_page_sz, addr;
 636
 637         cmo_page_sz = cmo_get_page_size();
 638         addr = __pa((unsigned long)page_address(page));
 639
 640         for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
 641                 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
 642                         plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
 643         }
 644 }
 645
 646 void arch_free_page(struct page *page, int order)
 647 {
 648         if (radix_enabled())
 649                 return;
 650         if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
 651                 return;
 652
 653         pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
 654 }
 655 EXPORT_SYMBOL(arch_free_page);
 656
 657 #endif /* CONFIG_PPC_SMLPAR */
 658 #endif /* CONFIG_PPC_STD_MMU_64 */
 659
 660 #ifdef CONFIG_TRACEPOINTS
 661 #ifdef HAVE_JUMP_LABEL
 662 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
 663
 664 int hcall_tracepoint_regfunc(void)
 665 {
 666         static_key_slow_inc(&hcall_tracepoint_key);
 667         return 0;
 668 }
 669
 670 void hcall_tracepoint_unregfunc(void)
 671 {
 672         static_key_slow_dec(&hcall_tracepoint_key);
 673 }
 674 #else
 675 /*
 676  * We optimise our hcall path by placing hcall_tracepoint_refcount
 677  * directly in the TOC so we can check if the hcall tracepoints are
 678  * enabled via a single load.
 679  */
 680
 681 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
 682 extern long hcall_tracepoint_refcount;
 683
 684 int hcall_tracepoint_regfunc(void)
 685 {
 686         hcall_tracepoint_refcount++;
 687         return 0;
 688 }
 689
 690 void hcall_tracepoint_unregfunc(void)
 691 {
 692         hcall_tracepoint_refcount--;
 693 }
 694 #endif
 695
 696 /*
 697  * Since the tracing code might execute hcalls we need to guard against
 698  * recursion. One example of this are spinlocks calling H_YIELD on
 699  * shared processor partitions.
 700  */
 701 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
 702
 703
 704 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
 705 {
 706         unsigned long flags;
 707         unsigned int *depth;
 708
 709         /*
 710          * We cannot call tracepoints inside RCU idle regions which
 711          * means we must not trace H_CEDE.
 712          */
 713         if (opcode == H_CEDE)
 714                 return;
 715
 716         local_irq_save(flags);
 717
 718         depth = this_cpu_ptr(&hcall_trace_depth);
 719
 720         if (*depth)
 721                 goto out;
 722
 723         (*depth)++;
 724         preempt_disable();
 725         trace_hcall_entry(opcode, args);
 726         (*depth)--;
 727
 728 out:
 729         local_irq_restore(flags);
 730 }
 731
 732 void __trace_hcall_exit(long opcode, unsigned long retval,
 733                         unsigned long *retbuf)
 734 {
 735         unsigned long flags;
 736         unsigned int *depth;
 737
 738         if (opcode == H_CEDE)
 739                 return;
 740
 741         local_irq_save(flags);
 742
 743         depth = this_cpu_ptr(&hcall_trace_depth);
 744
 745         if (*depth)
 746                 goto out;
 747
 748         (*depth)++;
 749         trace_hcall_exit(opcode, retval, retbuf);
 750         preempt_enable();
 751         (*depth)--;
 752
 753 out:
 754         local_irq_restore(flags);
 755 }
 756 #endif
 757
 758 /**
 759  * h_get_mpp
 760  * H_GET_MPP hcall returns info in 7 parms
 761  */
 762 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
 763 {
 764         int rc;
 765         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
 766
 767         rc = plpar_hcall9(H_GET_MPP, retbuf);
 768
 769         mpp_data->entitled_mem = retbuf[0];
 770         mpp_data->mapped_mem = retbuf[1];
 771
 772         mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
 773         mpp_data->pool_num = retbuf[2] & 0xffff;
 774
 775         mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
 776         mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
 777         mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
 778
 779         mpp_data->pool_size = retbuf[4];
 780         mpp_data->loan_request = retbuf[5];
 781         mpp_data->backing_mem = retbuf[6];
 782
 783         return rc;
 784 }
 785 EXPORT_SYMBOL(h_get_mpp);
 786
 787 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
 788 {
 789         int rc;
 790         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
 791
 792         rc = plpar_hcall9(H_GET_MPP_X, retbuf);
 793
 794         mpp_x_data->coalesced_bytes = retbuf[0];
 795         mpp_x_data->pool_coalesced_bytes = retbuf[1];
 796         mpp_x_data->pool_purr_cycles = retbuf[2];
 797         mpp_x_data->pool_spurr_cycles = retbuf[3];
 798
 799         return rc;
 800 }