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 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;
 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         flags = (newpp & 7) | H_AVPN;
 291         if (mmu_has_feature(MMU_FTR_KERNEL_RO))
 292                 /* Move pp0 into bit 8 (IBM 55) */
 293                 flags |= (newpp & HPTE_R_PP0) >> 55;
 294
 295         lpar_rc = plpar_pte_protect(flags, slot, want_v);
 296
 297         if (lpar_rc == H_NOT_FOUND) {
 298                 pr_devel("not found !\n");
 299                 return -1;
 300         }
 301
 302         pr_devel("ok\n");
 303
 304         BUG_ON(lpar_rc != H_SUCCESS);
 305
 306         return 0;
 307 }
 308
 309 static long __pSeries_lpar_hpte_find(unsigned long want_v, unsigned long hpte_group)
 310 {
 311         long lpar_rc;
 312         unsigned long i, j;
 313         struct {
 314                 unsigned long pteh;
 315                 unsigned long ptel;
 316         } ptes[4];
 317
 318         for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
 319
 320                 lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
 321                 if (lpar_rc != H_SUCCESS)
 322                         continue;
 323
 324                 for (j = 0; j < 4; j++) {
 325                         if (HPTE_V_COMPARE(ptes[j].pteh, want_v) &&
 326                             (ptes[j].pteh & HPTE_V_VALID))
 327                                 return i + j;
 328                 }
 329         }
 330
 331         return -1;
 332 }
 333
 334 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
 335 {
 336         long slot;
 337         unsigned long hash;
 338         unsigned long want_v;
 339         unsigned long hpte_group;
 340
 341         hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
 342         want_v = hpte_encode_avpn(vpn, psize, ssize);
 343
 344         /* Bolted entries are always in the primary group */
 345         hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 346         slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
 347         if (slot < 0)
 348                 return -1;
 349         return hpte_group + slot;
 350 }
 351
 352 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
 353                                              unsigned long ea,
 354                                              int psize, int ssize)
 355 {
 356         unsigned long vpn;
 357         unsigned long lpar_rc, slot, vsid, flags;
 358
 359         vsid = get_kernel_vsid(ea, ssize);
 360         vpn = hpt_vpn(ea, vsid, ssize);
 361
 362         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
 363         BUG_ON(slot == -1);
 364
 365         flags = newpp & 7;
 366         if (mmu_has_feature(MMU_FTR_KERNEL_RO))
 367                 /* Move pp0 into bit 8 (IBM 55) */
 368                 flags |= (newpp & HPTE_R_PP0) >> 55;
 369
 370         lpar_rc = plpar_pte_protect(flags, slot, 0);
 371
 372         BUG_ON(lpar_rc != H_SUCCESS);
 373 }
 374
 375 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
 376                                          int psize, int apsize,
 377                                          int ssize, int local)
 378 {
 379         unsigned long want_v;
 380         unsigned long lpar_rc;
 381         unsigned long dummy1, dummy2;
 382
 383         pr_devel("    inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
 384                  slot, vpn, psize, local);
 385
 386         want_v = hpte_encode_avpn(vpn, psize, ssize);
 387         lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
 388         if (lpar_rc == H_NOT_FOUND)
 389                 return;
 390
 391         BUG_ON(lpar_rc != H_SUCCESS);
 392 }
 393
 394 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 395 /*
 396  * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
 397  * to make sure that we avoid bouncing the hypervisor tlbie lock.
 398  */
 399 #define PPC64_HUGE_HPTE_BATCH 12
 400
 401 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
 402                                              unsigned long *vpn, int count,
 403                                              int psize, int ssize)
 404 {
 405         unsigned long param[PLPAR_HCALL9_BUFSIZE];
 406         int i = 0, pix = 0, rc;
 407         unsigned long flags = 0;
 408         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
 409
 410         if (lock_tlbie)
 411                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
 412
 413         for (i = 0; i < count; i++) {
 414
 415                 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 416                         pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
 417                                                      ssize, 0);
 418                 } else {
 419                         param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
 420                         param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
 421                         pix += 2;
 422                         if (pix == 8) {
 423                                 rc = plpar_hcall9(H_BULK_REMOVE, param,
 424                                                   param[0], param[1], param[2],
 425                                                   param[3], param[4], param[5],
 426                                                   param[6], param[7]);
 427                                 BUG_ON(rc != H_SUCCESS);
 428                                 pix = 0;
 429                         }
 430                 }
 431         }
 432         if (pix) {
 433                 param[pix] = HBR_END;
 434                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
 435                                   param[2], param[3], param[4], param[5],
 436                                   param[6], param[7]);
 437                 BUG_ON(rc != H_SUCCESS);
 438         }
 439
 440         if (lock_tlbie)
 441                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
 442 }
 443
 444 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
 445                                              unsigned long addr,
 446                                              unsigned char *hpte_slot_array,
 447                                              int psize, int ssize, int local)
 448 {
 449         int i, index = 0;
 450         unsigned long s_addr = addr;
 451         unsigned int max_hpte_count, valid;
 452         unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
 453         unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
 454         unsigned long shift, hidx, vpn = 0, hash, slot;
 455
 456         shift = mmu_psize_defs[psize].shift;
 457         max_hpte_count = 1U << (PMD_SHIFT - shift);
 458
 459         for (i = 0; i < max_hpte_count; i++) {
 460                 valid = hpte_valid(hpte_slot_array, i);
 461                 if (!valid)
 462                         continue;
 463                 hidx =  hpte_hash_index(hpte_slot_array, i);
 464
 465                 /* get the vpn */
 466                 addr = s_addr + (i * (1ul << shift));
 467                 vpn = hpt_vpn(addr, vsid, ssize);
 468                 hash = hpt_hash(vpn, shift, ssize);
 469                 if (hidx & _PTEIDX_SECONDARY)
 470                         hash = ~hash;
 471
 472                 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 473                 slot += hidx & _PTEIDX_GROUP_IX;
 474
 475                 slot_array[index] = slot;
 476                 vpn_array[index] = vpn;
 477                 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
 478                         /*
 479                          * Now do a bluk invalidate
 480                          */
 481                         __pSeries_lpar_hugepage_invalidate(slot_array,
 482                                                            vpn_array,
 483                                                            PPC64_HUGE_HPTE_BATCH,
 484                                                            psize, ssize);
 485                         index = 0;
 486                 } else
 487                         index++;
 488         }
 489         if (index)
 490                 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
 491                                                    index, psize, ssize);
 492 }
 493 #else
 494 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
 495                                              unsigned long addr,
 496                                              unsigned char *hpte_slot_array,
 497                                              int psize, int ssize, int local)
 498 {
 499         WARN(1, "%s called without THP support\n", __func__);
 500 }
 501 #endif
 502
 503 static int pSeries_lpar_hpte_removebolted(unsigned long ea,
 504                                           int psize, int ssize)
 505 {
 506         unsigned long vpn;
 507         unsigned long slot, vsid;
 508
 509         vsid = get_kernel_vsid(ea, ssize);
 510         vpn = hpt_vpn(ea, vsid, ssize);
 511
 512         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
 513         if (slot == -1)
 514                 return -ENOENT;
 515
 516         /*
 517          * lpar doesn't use the passed actual page size
 518          */
 519         pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
 520         return 0;
 521 }
 522
 523 /*
 524  * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
 525  * lock.
 526  */
 527 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
 528 {
 529         unsigned long vpn;
 530         unsigned long i, pix, rc;
 531         unsigned long flags = 0;
 532         struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
 533         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
 534         unsigned long param[PLPAR_HCALL9_BUFSIZE];
 535         unsigned long hash, index, shift, hidx, slot;
 536         real_pte_t pte;
 537         int psize, ssize;
 538
 539         if (lock_tlbie)
 540                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
 541
 542         psize = batch->psize;
 543         ssize = batch->ssize;
 544         pix = 0;
 545         for (i = 0; i < number; i++) {
 546                 vpn = batch->vpn[i];
 547                 pte = batch->pte[i];
 548                 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
 549                         hash = hpt_hash(vpn, shift, ssize);
 550                         hidx = __rpte_to_hidx(pte, index);
 551                         if (hidx & _PTEIDX_SECONDARY)
 552                                 hash = ~hash;
 553                         slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 554                         slot += hidx & _PTEIDX_GROUP_IX;
 555                         if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 556                                 /*
 557                                  * lpar doesn't use the passed actual page size
 558                                  */
 559                                 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
 560                                                              0, ssize, local);
 561                         } else {
 562                                 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
 563                                 param[pix+1] = hpte_encode_avpn(vpn, psize,
 564                                                                 ssize);
 565                                 pix += 2;
 566                                 if (pix == 8) {
 567                                         rc = plpar_hcall9(H_BULK_REMOVE, param,
 568                                                 param[0], param[1], param[2],
 569                                                 param[3], param[4], param[5],
 570                                                 param[6], param[7]);
 571                                         BUG_ON(rc != H_SUCCESS);
 572                                         pix = 0;
 573                                 }
 574                         }
 575                 } pte_iterate_hashed_end();
 576         }
 577         if (pix) {
 578                 param[pix] = HBR_END;
 579                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
 580                                   param[2], param[3], param[4], param[5],
 581                                   param[6], param[7]);
 582                 BUG_ON(rc != H_SUCCESS);
 583         }
 584
 585         if (lock_tlbie)
 586                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
 587 }
 588
 589 static int __init disable_bulk_remove(char *str)
 590 {
 591         if (strcmp(str, "off") == 0 &&
 592             firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 593                         printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
 594                         powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
 595         }
 596         return 1;
 597 }
 598
 599 __setup("bulk_remove=", disable_bulk_remove);
 600
 601 void __init hpte_init_pseries(void)
 602 {
 603         mmu_hash_ops.hpte_invalidate     = pSeries_lpar_hpte_invalidate;
 604         mmu_hash_ops.hpte_updatepp       = pSeries_lpar_hpte_updatepp;
 605         mmu_hash_ops.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
 606         mmu_hash_ops.hpte_insert         = pSeries_lpar_hpte_insert;
 607         mmu_hash_ops.hpte_remove         = pSeries_lpar_hpte_remove;
 608         mmu_hash_ops.hpte_removebolted   = pSeries_lpar_hpte_removebolted;
 609         mmu_hash_ops.flush_hash_range    = pSeries_lpar_flush_hash_range;
 610         mmu_hash_ops.hpte_clear_all      = pSeries_lpar_hptab_clear;
 611         mmu_hash_ops.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
 612 }
 613
 614 #ifdef CONFIG_PPC_SMLPAR
 615 #define CMO_FREE_HINT_DEFAULT 1
 616 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
 617
 618 static int __init cmo_free_hint(char *str)
 619 {
 620         char *parm;
 621         parm = strstrip(str);
 622
 623         if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
 624                 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
 625                 cmo_free_hint_flag = 0;
 626                 return 1;
 627         }
 628
 629         cmo_free_hint_flag = 1;
 630         printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
 631
 632         if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
 633                 return 1;
 634
 635         return 0;
 636 }
 637
 638 __setup("cmo_free_hint=", cmo_free_hint);
 639
 640 static void pSeries_set_page_state(struct page *page, int order,
 641                                    unsigned long state)
 642 {
 643         int i, j;
 644         unsigned long cmo_page_sz, addr;
 645
 646         cmo_page_sz = cmo_get_page_size();
 647         addr = __pa((unsigned long)page_address(page));
 648
 649         for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
 650                 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
 651                         plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
 652         }
 653 }
 654
 655 void arch_free_page(struct page *page, int order)
 656 {
 657         if (radix_enabled())
 658                 return;
 659         if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
 660                 return;
 661
 662         pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
 663 }
 664 EXPORT_SYMBOL(arch_free_page);
 665
 666 #endif /* CONFIG_PPC_SMLPAR */
 667 #endif /* CONFIG_PPC_STD_MMU_64 */
 668
 669 #ifdef CONFIG_TRACEPOINTS
 670 #ifdef HAVE_JUMP_LABEL
 671 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
 672
 673 void hcall_tracepoint_regfunc(void)
 674 {
 675         static_key_slow_inc(&hcall_tracepoint_key);
 676 }
 677
 678 void hcall_tracepoint_unregfunc(void)
 679 {
 680         static_key_slow_dec(&hcall_tracepoint_key);
 681 }
 682 #else
 683 /*
 684  * We optimise our hcall path by placing hcall_tracepoint_refcount
 685  * directly in the TOC so we can check if the hcall tracepoints are
 686  * enabled via a single load.
 687  */
 688
 689 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
 690 extern long hcall_tracepoint_refcount;
 691
 692 void hcall_tracepoint_regfunc(void)
 693 {
 694         hcall_tracepoint_refcount++;
 695 }
 696
 697 void hcall_tracepoint_unregfunc(void)
 698 {
 699         hcall_tracepoint_refcount--;
 700 }
 701 #endif
 702
 703 /*
 704  * Since the tracing code might execute hcalls we need to guard against
 705  * recursion. One example of this are spinlocks calling H_YIELD on
 706  * shared processor partitions.
 707  */
 708 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
 709
 710
 711 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
 712 {
 713         unsigned long flags;
 714         unsigned int *depth;
 715
 716         /*
 717          * We cannot call tracepoints inside RCU idle regions which
 718          * means we must not trace H_CEDE.
 719          */
 720         if (opcode == H_CEDE)
 721                 return;
 722
 723         local_irq_save(flags);
 724
 725         depth = this_cpu_ptr(&hcall_trace_depth);
 726
 727         if (*depth)
 728                 goto out;
 729
 730         (*depth)++;
 731         preempt_disable();
 732         trace_hcall_entry(opcode, args);
 733         (*depth)--;
 734
 735 out:
 736         local_irq_restore(flags);
 737 }
 738
 739 void __trace_hcall_exit(long opcode, unsigned long retval,
 740                         unsigned long *retbuf)
 741 {
 742         unsigned long flags;
 743         unsigned int *depth;
 744
 745         if (opcode == H_CEDE)
 746                 return;
 747
 748         local_irq_save(flags);
 749
 750         depth = this_cpu_ptr(&hcall_trace_depth);
 751
 752         if (*depth)
 753                 goto out;
 754
 755         (*depth)++;
 756         trace_hcall_exit(opcode, retval, retbuf);
 757         preempt_enable();
 758         (*depth)--;
 759
 760 out:
 761         local_irq_restore(flags);
 762 }
 763 #endif
 764
 765 /**
 766  * h_get_mpp
 767  * H_GET_MPP hcall returns info in 7 parms
 768  */
 769 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
 770 {
 771         int rc;
 772         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
 773
 774         rc = plpar_hcall9(H_GET_MPP, retbuf);
 775
 776         mpp_data->entitled_mem = retbuf[0];
 777         mpp_data->mapped_mem = retbuf[1];
 778
 779         mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
 780         mpp_data->pool_num = retbuf[2] & 0xffff;
 781
 782         mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
 783         mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
 784         mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
 785
 786         mpp_data->pool_size = retbuf[4];
 787         mpp_data->loan_request = retbuf[5];
 788         mpp_data->backing_mem = retbuf[6];
 789
 790         return rc;
 791 }
 792 EXPORT_SYMBOL(h_get_mpp);
 793
 794 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
 795 {
 796         int rc;
 797         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
 798
 799         rc = plpar_hcall9(H_GET_MPP_X, retbuf);
 800
 801         mpp_x_data->coalesced_bytes = retbuf[0];
 802         mpp_x_data->pool_coalesced_bytes = retbuf[1];
 803         mpp_x_data->pool_purr_cycles = retbuf[2];
 804         mpp_x_data->pool_spurr_cycles = retbuf[3];
 805
 806         return rc;
 807 }