kernel/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
  49 #include "pseries.h"
  50
  51 /* Flag bits for H_BULK_REMOVE */
  52 #define HBR_REQUEST     0x4000000000000000UL
  53 #define HBR_RESPONSE    0x8000000000000000UL
  54 #define HBR_END         0xc000000000000000UL
  55 #define HBR_AVPN        0x0200000000000000UL
  56 #define HBR_ANDCOND     0x0100000000000000UL
  57
  58
  59 /* in hvCall.S */
  60 EXPORT_SYMBOL(plpar_hcall);
  61 EXPORT_SYMBOL(plpar_hcall9);
  62 EXPORT_SYMBOL(plpar_hcall_norets);
  63
  64 void vpa_init(int cpu)
  65 {
  66         int hwcpu = get_hard_smp_processor_id(cpu);
  67         unsigned long addr;
  68         long ret;
  69         struct paca_struct *pp;
  70         struct dtl_entry *dtl;
  71
  72         /*
  73          * The spec says it "may be problematic" if CPU x registers the VPA of
  74          * CPU y. We should never do that, but wail if we ever do.
  75          */
  76         WARN_ON(cpu != smp_processor_id());
  77
  78         if (cpu_has_feature(CPU_FTR_ALTIVEC))
  79                 lppaca_of(cpu).vmxregs_in_use = 1;
  80
  81         if (cpu_has_feature(CPU_FTR_ARCH_207S))
  82                 lppaca_of(cpu).ebb_regs_in_use = 1;
  83
  84         addr = __pa(&lppaca_of(cpu));
  85         ret = register_vpa(hwcpu, addr);
  86
  87         if (ret) {
  88                 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
  89                        "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
  90                 return;
  91         }
  92         /*
  93          * PAPR says this feature is SLB-Buffer but firmware never
  94          * reports that.  All SPLPAR support SLB shadow buffer.
  95          */
  96         addr = __pa(paca[cpu].slb_shadow_ptr);
  97         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
  98                 ret = register_slb_shadow(hwcpu, addr);
  99                 if (ret)
 100                         pr_err("WARNING: SLB shadow buffer registration for "
 101                                "cpu %d (hw %d) of area %lx failed with %ld\n",
 102                                cpu, hwcpu, addr, ret);
 103         }
 104
 105         /*
 106          * Register dispatch trace log, if one has been allocated.
 107          */
 108         pp = &paca[cpu];
 109         dtl = pp->dispatch_log;
 110         if (dtl) {
 111                 pp->dtl_ridx = 0;
 112                 pp->dtl_curr = dtl;
 113                 lppaca_of(cpu).dtl_idx = 0;
 114
 115                 /* hypervisor reads buffer length from this field */
 116                 dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
 117                 ret = register_dtl(hwcpu, __pa(dtl));
 118                 if (ret)
 119                         pr_err("WARNING: DTL registration of cpu %d (hw %d) "
 120                                "failed with %ld\n", smp_processor_id(),
 121                                hwcpu, ret);
 122                 lppaca_of(cpu).dtl_enable_mask = 2;
 123         }
 124 }
 125
 126 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
 127                                      unsigned long vpn, unsigned long pa,
 128                                      unsigned long rflags, unsigned long vflags,
 129                                      int psize, int apsize, int ssize)
 130 {
 131         unsigned long lpar_rc;
 132         unsigned long flags;
 133         unsigned long slot;
 134         unsigned long hpte_v, hpte_r;
 135
 136         if (!(vflags & HPTE_V_BOLTED))
 137                 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
 138                          "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
 139                          hpte_group, vpn,  pa, rflags, vflags, psize);
 140
 141         hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
 142         hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
 143
 144         if (!(vflags & HPTE_V_BOLTED))
 145                 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
 146
 147         /* Now fill in the actual HPTE */
 148         /* Set CEC cookie to 0         */
 149         /* Zero page = 0               */
 150         /* I-cache Invalidate = 0      */
 151         /* I-cache synchronize = 0     */
 152         /* Exact = 0                   */
 153         flags = 0;
 154
 155         /* Make pHyp happy */
 156         if ((rflags & _PAGE_NO_CACHE) && !(rflags & _PAGE_WRITETHRU))
 157                 hpte_r &= ~HPTE_R_M;
 158
 159         if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
 160                 flags |= H_COALESCE_CAND;
 161
 162         lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
 163         if (unlikely(lpar_rc == H_PTEG_FULL)) {
 164                 if (!(vflags & HPTE_V_BOLTED))
 165                         pr_devel(" full\n");
 166                 return -1;
 167         }
 168
 169         /*
 170          * Since we try and ioremap PHBs we don't own, the pte insert
 171          * will fail. However we must catch the failure in hash_page
 172          * or we will loop forever, so return -2 in this case.
 173          */
 174         if (unlikely(lpar_rc != H_SUCCESS)) {
 175                 if (!(vflags & HPTE_V_BOLTED))
 176                         pr_devel(" lpar err %ld\n", lpar_rc);
 177                 return -2;
 178         }
 179         if (!(vflags & HPTE_V_BOLTED))
 180                 pr_devel(" -> slot: %lu\n", slot & 7);
 181
 182         /* Because of iSeries, we have to pass down the secondary
 183          * bucket bit here as well
 184          */
 185         return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
 186 }
 187
 188 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
 189
 190 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
 191 {
 192         unsigned long slot_offset;
 193         unsigned long lpar_rc;
 194         int i;
 195         unsigned long dummy1, dummy2;
 196
 197         /* pick a random slot to start at */
 198         slot_offset = mftb() & 0x7;
 199
 200         for (i = 0; i < HPTES_PER_GROUP; i++) {
 201
 202                 /* don't remove a bolted entry */
 203                 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
 204                                            (0x1UL << 4), &dummy1, &dummy2);
 205                 if (lpar_rc == H_SUCCESS)
 206                         return i;
 207
 208                 /*
 209                  * The test for adjunct partition is performed before the
 210                  * ANDCOND test.  H_RESOURCE may be returned, so we need to
 211                  * check for that as well.
 212                  */
 213                 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
 214
 215                 slot_offset++;
 216                 slot_offset &= 0x7;
 217         }
 218
 219         return -1;
 220 }
 221
 222 static void pSeries_lpar_hptab_clear(void)
 223 {
 224         unsigned long size_bytes = 1UL << ppc64_pft_size;
 225         unsigned long hpte_count = size_bytes >> 4;
 226         struct {
 227                 unsigned long pteh;
 228                 unsigned long ptel;
 229         } ptes[4];
 230         long lpar_rc;
 231         unsigned long i, j;
 232
 233         /* Read in batches of 4,
 234          * invalidate only valid entries not in the VRMA
 235          * hpte_count will be a multiple of 4
 236          */
 237         for (i = 0; i < hpte_count; i += 4) {
 238                 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
 239                 if (lpar_rc != H_SUCCESS)
 240                         continue;
 241                 for (j = 0; j < 4; j++){
 242                         if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
 243                                 HPTE_V_VRMA_MASK)
 244                                 continue;
 245                         if (ptes[j].pteh & HPTE_V_VALID)
 246                                 plpar_pte_remove_raw(0, i + j, 0,
 247                                         &(ptes[j].pteh), &(ptes[j].ptel));
 248                 }
 249         }
 250
 251 #ifdef __LITTLE_ENDIAN__
 252         /*
 253          * Reset exceptions to big endian.
 254          *
 255          * FIXME this is a hack for kexec, we need to reset the exception
 256          * endian before starting the new kernel and this is a convenient place
 257          * to do it.
 258          *
 259          * This is also called on boot when a fadump happens. In that case we
 260          * must not change the exception endian mode.
 261          */
 262         if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active()) {
 263                 long rc;
 264
 265                 rc = pseries_big_endian_exceptions();
 266                 /*
 267                  * At this point it is unlikely panic() will get anything
 268                  * out to the user, but at least this will stop us from
 269                  * continuing on further and creating an even more
 270                  * difficult to debug situation.
 271                  *
 272                  * There is a known problem when kdump'ing, if cpus are offline
 273                  * the above call will fail. Rather than panicking again, keep
 274                  * going and hope the kdump kernel is also little endian, which
 275                  * it usually is.
 276                  */
 277                 if (rc && !kdump_in_progress())
 278                         panic("Could not enable big endian exceptions");
 279         }
 280 #endif
 281 }
 282
 283 /*
 284  * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
 285  * the low 3 bits of flags happen to line up.  So no transform is needed.
 286  * We can probably optimize here and assume the high bits of newpp are
 287  * already zero.  For now I am paranoid.
 288  */
 289 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
 290                                        unsigned long newpp,
 291                                        unsigned long vpn,
 292                                        int psize, int apsize,
 293                                        int ssize, unsigned long inv_flags)
 294 {
 295         unsigned long lpar_rc;
 296         unsigned long flags = (newpp & 7) | H_AVPN;
 297         unsigned long want_v;
 298
 299         want_v = hpte_encode_avpn(vpn, psize, ssize);
 300
 301         pr_devel("    update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
 302                  want_v, slot, flags, psize);
 303
 304         lpar_rc = plpar_pte_protect(flags, slot, want_v);
 305
 306         if (lpar_rc == H_NOT_FOUND) {
 307                 pr_devel("not found !\n");
 308                 return -1;
 309         }
 310
 311         pr_devel("ok\n");
 312
 313         BUG_ON(lpar_rc != H_SUCCESS);
 314
 315         return 0;
 316 }
 317
 318 static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
 319 {
 320         unsigned long dword0;
 321         unsigned long lpar_rc;
 322         unsigned long dummy_word1;
 323         unsigned long flags;
 324
 325         /* Read 1 pte at a time                        */
 326         /* Do not need RPN to logical page translation */
 327         /* No cross CEC PFT access                     */
 328         flags = 0;
 329
 330         lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
 331
 332         BUG_ON(lpar_rc != H_SUCCESS);
 333
 334         return dword0;
 335 }
 336
 337 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
 338 {
 339         unsigned long hash;
 340         unsigned long i;
 341         long slot;
 342         unsigned long want_v, hpte_v;
 343
 344         hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
 345         want_v = hpte_encode_avpn(vpn, psize, ssize);
 346
 347         /* Bolted entries are always in the primary group */
 348         slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 349         for (i = 0; i < HPTES_PER_GROUP; i++) {
 350                 hpte_v = pSeries_lpar_hpte_getword0(slot);
 351
 352                 if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
 353                         /* HPTE matches */
 354                         return slot;
 355                 ++slot;
 356         }
 357
 358         return -1;
 359 }
 360
 361 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
 362                                              unsigned long ea,
 363                                              int psize, int ssize)
 364 {
 365         unsigned long vpn;
 366         unsigned long lpar_rc, slot, vsid, flags;
 367
 368         vsid = get_kernel_vsid(ea, ssize);
 369         vpn = hpt_vpn(ea, vsid, ssize);
 370
 371         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
 372         BUG_ON(slot == -1);
 373
 374         flags = newpp & 7;
 375         lpar_rc = plpar_pte_protect(flags, slot, 0);
 376
 377         BUG_ON(lpar_rc != H_SUCCESS);
 378 }
 379
 380 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
 381                                          int psize, int apsize,
 382                                          int ssize, int local)
 383 {
 384         unsigned long want_v;
 385         unsigned long lpar_rc;
 386         unsigned long dummy1, dummy2;
 387
 388         pr_devel("    inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
 389                  slot, vpn, psize, local);
 390
 391         want_v = hpte_encode_avpn(vpn, psize, ssize);
 392         lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
 393         if (lpar_rc == H_NOT_FOUND)
 394                 return;
 395
 396         BUG_ON(lpar_rc != H_SUCCESS);
 397 }
 398
 399 /*
 400  * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
 401  * to make sure that we avoid bouncing the hypervisor tlbie lock.
 402  */
 403 #define PPC64_HUGE_HPTE_BATCH 12
 404
 405 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
 406                                              unsigned long *vpn, int count,
 407                                              int psize, int ssize)
 408 {
 409         unsigned long param[8];
 410         int i = 0, pix = 0, rc;
 411         unsigned long flags = 0;
 412         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
 413
 414         if (lock_tlbie)
 415                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
 416
 417         for (i = 0; i < count; i++) {
 418
 419                 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 420                         pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
 421                                                      ssize, 0);
 422                 } else {
 423                         param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
 424                         param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
 425                         pix += 2;
 426                         if (pix == 8) {
 427                                 rc = plpar_hcall9(H_BULK_REMOVE, param,
 428                                                   param[0], param[1], param[2],
 429                                                   param[3], param[4], param[5],
 430                                                   param[6], param[7]);
 431                                 BUG_ON(rc != H_SUCCESS);
 432                                 pix = 0;
 433                         }
 434                 }
 435         }
 436         if (pix) {
 437                 param[pix] = HBR_END;
 438                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
 439                                   param[2], param[3], param[4], param[5],
 440                                   param[6], param[7]);
 441                 BUG_ON(rc != H_SUCCESS);
 442         }
 443
 444         if (lock_tlbie)
 445                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
 446 }
 447
 448 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
 449                                              unsigned long addr,
 450                                              unsigned char *hpte_slot_array,
 451                                              int psize, int ssize, int local)
 452 {
 453         int i, index = 0;
 454         unsigned long s_addr = addr;
 455         unsigned int max_hpte_count, valid;
 456         unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
 457         unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
 458         unsigned long shift, hidx, vpn = 0, hash, slot;
 459
 460         shift = mmu_psize_defs[psize].shift;
 461         max_hpte_count = 1U << (PMD_SHIFT - shift);
 462
 463         for (i = 0; i < max_hpte_count; i++) {
 464                 valid = hpte_valid(hpte_slot_array, i);
 465                 if (!valid)
 466                         continue;
 467                 hidx =  hpte_hash_index(hpte_slot_array, i);
 468
 469                 /* get the vpn */
 470                 addr = s_addr + (i * (1ul << shift));
 471                 vpn = hpt_vpn(addr, vsid, ssize);
 472                 hash = hpt_hash(vpn, shift, ssize);
 473                 if (hidx & _PTEIDX_SECONDARY)
 474                         hash = ~hash;
 475
 476                 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 477                 slot += hidx & _PTEIDX_GROUP_IX;
 478
 479                 slot_array[index] = slot;
 480                 vpn_array[index] = vpn;
 481                 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
 482                         /*
 483                          * Now do a bluk invalidate
 484                          */
 485                         __pSeries_lpar_hugepage_invalidate(slot_array,
 486                                                            vpn_array,
 487                                                            PPC64_HUGE_HPTE_BATCH,
 488                                                            psize, ssize);
 489                         index = 0;
 490                 } else
 491                         index++;
 492         }
 493         if (index)
 494                 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
 495                                                    index, psize, ssize);
 496 }
 497
 498 static void pSeries_lpar_hpte_removebolted(unsigned long ea,
 499                                            int psize, int ssize)
 500 {
 501         unsigned long vpn;
 502         unsigned long slot, vsid;
 503
 504         vsid = get_kernel_vsid(ea, ssize);
 505         vpn = hpt_vpn(ea, vsid, ssize);
 506
 507         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
 508         BUG_ON(slot == -1);
 509         /*
 510          * lpar doesn't use the passed actual page size
 511          */
 512         pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
 513 }
 514
 515 /*
 516  * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
 517  * lock.
 518  */
 519 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
 520 {
 521         unsigned long vpn;
 522         unsigned long i, pix, rc;
 523         unsigned long flags = 0;
 524         struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
 525         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
 526         unsigned long param[9];
 527         unsigned long hash, index, shift, hidx, slot;
 528         real_pte_t pte;
 529         int psize, ssize;
 530
 531         if (lock_tlbie)
 532                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
 533
 534         psize = batch->psize;
 535         ssize = batch->ssize;
 536         pix = 0;
 537         for (i = 0; i < number; i++) {
 538                 vpn = batch->vpn[i];
 539                 pte = batch->pte[i];
 540                 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
 541                         hash = hpt_hash(vpn, shift, ssize);
 542                         hidx = __rpte_to_hidx(pte, index);
 543                         if (hidx & _PTEIDX_SECONDARY)
 544                                 hash = ~hash;
 545                         slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 546                         slot += hidx & _PTEIDX_GROUP_IX;
 547                         if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 548                                 /*
 549                                  * lpar doesn't use the passed actual page size
 550                                  */
 551                                 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
 552                                                              0, ssize, local);
 553                         } else {
 554                                 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
 555                                 param[pix+1] = hpte_encode_avpn(vpn, psize,
 556                                                                 ssize);
 557                                 pix += 2;
 558                                 if (pix == 8) {
 559                                         rc = plpar_hcall9(H_BULK_REMOVE, param,
 560                                                 param[0], param[1], param[2],
 561                                                 param[3], param[4], param[5],
 562                                                 param[6], param[7]);
 563                                         BUG_ON(rc != H_SUCCESS);
 564                                         pix = 0;
 565                                 }
 566                         }
 567                 } pte_iterate_hashed_end();
 568         }
 569         if (pix) {
 570                 param[pix] = HBR_END;
 571                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
 572                                   param[2], param[3], param[4], param[5],
 573                                   param[6], param[7]);
 574                 BUG_ON(rc != H_SUCCESS);
 575         }
 576
 577         if (lock_tlbie)
 578                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
 579 }
 580
 581 static int __init disable_bulk_remove(char *str)
 582 {
 583         if (strcmp(str, "off") == 0 &&
 584             firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 585                         printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
 586                         powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
 587         }
 588         return 1;
 589 }
 590
 591 __setup("bulk_remove=", disable_bulk_remove);
 592
 593 void __init hpte_init_lpar(void)
 594 {
 595         ppc_md.hpte_invalidate  = pSeries_lpar_hpte_invalidate;
 596         ppc_md.hpte_updatepp    = pSeries_lpar_hpte_updatepp;
 597         ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
 598         ppc_md.hpte_insert      = pSeries_lpar_hpte_insert;
 599         ppc_md.hpte_remove      = pSeries_lpar_hpte_remove;
 600         ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
 601         ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range;
 602         ppc_md.hpte_clear_all   = pSeries_lpar_hptab_clear;
 603         ppc_md.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
 604 }
 605
 606 #ifdef CONFIG_PPC_SMLPAR
 607 #define CMO_FREE_HINT_DEFAULT 1
 608 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
 609
 610 static int __init cmo_free_hint(char *str)
 611 {
 612         char *parm;
 613         parm = strstrip(str);
 614
 615         if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
 616                 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
 617                 cmo_free_hint_flag = 0;
 618                 return 1;
 619         }
 620
 621         cmo_free_hint_flag = 1;
 622         printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
 623
 624         if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
 625                 return 1;
 626
 627         return 0;
 628 }
 629
 630 __setup("cmo_free_hint=", cmo_free_hint);
 631
 632 static void pSeries_set_page_state(struct page *page, int order,
 633                                    unsigned long state)
 634 {
 635         int i, j;
 636         unsigned long cmo_page_sz, addr;
 637
 638         cmo_page_sz = cmo_get_page_size();
 639         addr = __pa((unsigned long)page_address(page));
 640
 641         for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
 642                 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
 643                         plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
 644         }
 645 }
 646
 647 void arch_free_page(struct page *page, int order)
 648 {
 649         if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
 650                 return;
 651
 652         pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
 653 }
 654 EXPORT_SYMBOL(arch_free_page);
 655
 656 #endif
 657
 658 #ifdef CONFIG_TRACEPOINTS
 659 #ifdef HAVE_JUMP_LABEL
 660 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
 661
 662 void hcall_tracepoint_regfunc(void)
 663 {
 664         static_key_slow_inc(&hcall_tracepoint_key);
 665 }
 666
 667 void hcall_tracepoint_unregfunc(void)
 668 {
 669         static_key_slow_dec(&hcall_tracepoint_key);
 670 }
 671 #else
 672 /*
 673  * We optimise our hcall path by placing hcall_tracepoint_refcount
 674  * directly in the TOC so we can check if the hcall tracepoints are
 675  * enabled via a single load.
 676  */
 677
 678 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
 679 extern long hcall_tracepoint_refcount;
 680
 681 void hcall_tracepoint_regfunc(void)
 682 {
 683         hcall_tracepoint_refcount++;
 684 }
 685
 686 void hcall_tracepoint_unregfunc(void)
 687 {
 688         hcall_tracepoint_refcount--;
 689 }
 690 #endif
 691
 692 /*
 693  * Since the tracing code might execute hcalls we need to guard against
 694  * recursion. One example of this are spinlocks calling H_YIELD on
 695  * shared processor partitions.
 696  */
 697 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
 698
 699
 700 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
 701 {
 702         unsigned long flags;
 703         unsigned int *depth;
 704
 705         /*
 706          * We cannot call tracepoints inside RCU idle regions which
 707          * means we must not trace H_CEDE.
 708          */
 709         if (opcode == H_CEDE)
 710                 return;
 711
 712         local_irq_save(flags);
 713
 714         depth = this_cpu_ptr(&hcall_trace_depth);
 715
 716         if (*depth)
 717                 goto out;
 718
 719         (*depth)++;
 720         preempt_disable();
 721         trace_hcall_entry(opcode, args);
 722         (*depth)--;
 723
 724 out:
 725         local_irq_restore(flags);
 726 }
 727
 728 void __trace_hcall_exit(long opcode, unsigned long retval,
 729                         unsigned long *retbuf)
 730 {
 731         unsigned long flags;
 732         unsigned int *depth;
 733
 734         if (opcode == H_CEDE)
 735                 return;
 736
 737         local_irq_save(flags);
 738
 739         depth = this_cpu_ptr(&hcall_trace_depth);
 740
 741         if (*depth)
 742                 goto out;
 743
 744         (*depth)++;
 745         trace_hcall_exit(opcode, retval, retbuf);
 746         preempt_enable();
 747         (*depth)--;
 748
 749 out:
 750         local_irq_restore(flags);
 751 }
 752 #endif
 753
 754 /**
 755  * h_get_mpp
 756  * H_GET_MPP hcall returns info in 7 parms
 757  */
 758 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
 759 {
 760         int rc;
 761         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
 762
 763         rc = plpar_hcall9(H_GET_MPP, retbuf);
 764
 765         mpp_data->entitled_mem = retbuf[0];
 766         mpp_data->mapped_mem = retbuf[1];
 767
 768         mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
 769         mpp_data->pool_num = retbuf[2] & 0xffff;
 770
 771         mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
 772         mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
 773         mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
 774
 775         mpp_data->pool_size = retbuf[4];
 776         mpp_data->loan_request = retbuf[5];
 777         mpp_data->backing_mem = retbuf[6];
 778
 779         return rc;
 780 }
 781 EXPORT_SYMBOL(h_get_mpp);
 782
 783 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
 784 {
 785         int rc;
 786         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
 787
 788         rc = plpar_hcall9(H_GET_MPP_X, retbuf);
 789
 790         mpp_x_data->coalesced_bytes = retbuf[0];
 791         mpp_x_data->pool_coalesced_bytes = retbuf[1];
 792         mpp_x_data->pool_purr_cycles = retbuf[2];
 793         mpp_x_data->pool_spurr_cycles = retbuf[3];
 794
 795         return rc;
 796 }