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/static_key.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
  47 #include "pseries.h"
  48
  49 /* Flag bits for H_BULK_REMOVE */
  50 #define HBR_REQUEST     0x4000000000000000UL
  51 #define HBR_RESPONSE    0x8000000000000000UL
  52 #define HBR_END         0xc000000000000000UL
  53 #define HBR_AVPN        0x0200000000000000UL
  54 #define HBR_ANDCOND     0x0100000000000000UL
  55
  56
  57 /* in hvCall.S */
  58 EXPORT_SYMBOL(plpar_hcall);
  59 EXPORT_SYMBOL(plpar_hcall9);
  60 EXPORT_SYMBOL(plpar_hcall_norets);
  61
  62 extern void pSeries_find_serial_port(void);
  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         /* Reset exceptions to big endian */
 253         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
 254                 long rc;
 255
 256                 rc = pseries_big_endian_exceptions();
 257                 /*
 258                  * At this point it is unlikely panic() will get anything
 259                  * out to the user, but at least this will stop us from
 260                  * continuing on further and creating an even more
 261                  * difficult to debug situation.
 262                  */
 263                 if (rc)
 264                         panic("Could not enable big endian exceptions");
 265         }
 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, int local)
 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 unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
 305 {
 306         unsigned long dword0;
 307         unsigned long lpar_rc;
 308         unsigned long dummy_word1;
 309         unsigned long flags;
 310
 311         /* Read 1 pte at a time                        */
 312         /* Do not need RPN to logical page translation */
 313         /* No cross CEC PFT access                     */
 314         flags = 0;
 315
 316         lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
 317
 318         BUG_ON(lpar_rc != H_SUCCESS);
 319
 320         return dword0;
 321 }
 322
 323 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
 324 {
 325         unsigned long hash;
 326         unsigned long i;
 327         long slot;
 328         unsigned long want_v, hpte_v;
 329
 330         hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
 331         want_v = hpte_encode_avpn(vpn, psize, ssize);
 332
 333         /* Bolted entries are always in the primary group */
 334         slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 335         for (i = 0; i < HPTES_PER_GROUP; i++) {
 336                 hpte_v = pSeries_lpar_hpte_getword0(slot);
 337
 338                 if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
 339                         /* HPTE matches */
 340                         return slot;
 341                 ++slot;
 342         }
 343
 344         return -1;
 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 /*
 386  * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
 387  * to make sure that we avoid bouncing the hypervisor tlbie lock.
 388  */
 389 #define PPC64_HUGE_HPTE_BATCH 12
 390
 391 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
 392                                              unsigned long *vpn, int count,
 393                                              int psize, int ssize)
 394 {
 395         unsigned long param[8];
 396         int i = 0, pix = 0, rc;
 397         unsigned long flags = 0;
 398         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
 399
 400         if (lock_tlbie)
 401                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
 402
 403         for (i = 0; i < count; i++) {
 404
 405                 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 406                         pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
 407                                                      ssize, 0);
 408                 } else {
 409                         param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
 410                         param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
 411                         pix += 2;
 412                         if (pix == 8) {
 413                                 rc = plpar_hcall9(H_BULK_REMOVE, param,
 414                                                   param[0], param[1], param[2],
 415                                                   param[3], param[4], param[5],
 416                                                   param[6], param[7]);
 417                                 BUG_ON(rc != H_SUCCESS);
 418                                 pix = 0;
 419                         }
 420                 }
 421         }
 422         if (pix) {
 423                 param[pix] = HBR_END;
 424                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
 425                                   param[2], param[3], param[4], param[5],
 426                                   param[6], param[7]);
 427                 BUG_ON(rc != H_SUCCESS);
 428         }
 429
 430         if (lock_tlbie)
 431                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
 432 }
 433
 434 static void pSeries_lpar_hugepage_invalidate(struct mm_struct *mm,
 435                                        unsigned char *hpte_slot_array,
 436                                        unsigned long addr, int psize)
 437 {
 438         int ssize = 0, i, index = 0;
 439         unsigned long s_addr = addr;
 440         unsigned int max_hpte_count, valid;
 441         unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
 442         unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
 443         unsigned long shift, hidx, vpn = 0, vsid, hash, slot;
 444
 445         shift = mmu_psize_defs[psize].shift;
 446         max_hpte_count = 1U << (PMD_SHIFT - shift);
 447
 448         for (i = 0; i < max_hpte_count; i++) {
 449                 valid = hpte_valid(hpte_slot_array, i);
 450                 if (!valid)
 451                         continue;
 452                 hidx =  hpte_hash_index(hpte_slot_array, i);
 453
 454                 /* get the vpn */
 455                 addr = s_addr + (i * (1ul << shift));
 456                 if (!is_kernel_addr(addr)) {
 457                         ssize = user_segment_size(addr);
 458                         vsid = get_vsid(mm->context.id, addr, ssize);
 459                         WARN_ON(vsid == 0);
 460                 } else {
 461                         vsid = get_kernel_vsid(addr, mmu_kernel_ssize);
 462                         ssize = mmu_kernel_ssize;
 463                 }
 464
 465                 vpn = hpt_vpn(addr, vsid, ssize);
 466                 hash = hpt_hash(vpn, shift, ssize);
 467                 if (hidx & _PTEIDX_SECONDARY)
 468                         hash = ~hash;
 469
 470                 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 471                 slot += hidx & _PTEIDX_GROUP_IX;
 472
 473                 slot_array[index] = slot;
 474                 vpn_array[index] = vpn;
 475                 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
 476                         /*
 477                          * Now do a bluk invalidate
 478                          */
 479                         __pSeries_lpar_hugepage_invalidate(slot_array,
 480                                                            vpn_array,
 481                                                            PPC64_HUGE_HPTE_BATCH,
 482                                                            psize, ssize);
 483                         index = 0;
 484                 } else
 485                         index++;
 486         }
 487         if (index)
 488                 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
 489                                                    index, psize, ssize);
 490 }
 491
 492 static void pSeries_lpar_hpte_removebolted(unsigned long ea,
 493                                            int psize, int ssize)
 494 {
 495         unsigned long vpn;
 496         unsigned long slot, vsid;
 497
 498         vsid = get_kernel_vsid(ea, ssize);
 499         vpn = hpt_vpn(ea, vsid, ssize);
 500
 501         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
 502         BUG_ON(slot == -1);
 503         /*
 504          * lpar doesn't use the passed actual page size
 505          */
 506         pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
 507 }
 508
 509 /*
 510  * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
 511  * lock.
 512  */
 513 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
 514 {
 515         unsigned long vpn;
 516         unsigned long i, pix, rc;
 517         unsigned long flags = 0;
 518         struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
 519         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
 520         unsigned long param[9];
 521         unsigned long hash, index, shift, hidx, slot;
 522         real_pte_t pte;
 523         int psize, ssize;
 524
 525         if (lock_tlbie)
 526                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
 527
 528         psize = batch->psize;
 529         ssize = batch->ssize;
 530         pix = 0;
 531         for (i = 0; i < number; i++) {
 532                 vpn = batch->vpn[i];
 533                 pte = batch->pte[i];
 534                 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
 535                         hash = hpt_hash(vpn, shift, ssize);
 536                         hidx = __rpte_to_hidx(pte, index);
 537                         if (hidx & _PTEIDX_SECONDARY)
 538                                 hash = ~hash;
 539                         slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 540                         slot += hidx & _PTEIDX_GROUP_IX;
 541                         if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 542                                 /*
 543                                  * lpar doesn't use the passed actual page size
 544                                  */
 545                                 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
 546                                                              0, ssize, local);
 547                         } else {
 548                                 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
 549                                 param[pix+1] = hpte_encode_avpn(vpn, psize,
 550                                                                 ssize);
 551                                 pix += 2;
 552                                 if (pix == 8) {
 553                                         rc = plpar_hcall9(H_BULK_REMOVE, param,
 554                                                 param[0], param[1], param[2],
 555                                                 param[3], param[4], param[5],
 556                                                 param[6], param[7]);
 557                                         BUG_ON(rc != H_SUCCESS);
 558                                         pix = 0;
 559                                 }
 560                         }
 561                 } pte_iterate_hashed_end();
 562         }
 563         if (pix) {
 564                 param[pix] = HBR_END;
 565                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
 566                                   param[2], param[3], param[4], param[5],
 567                                   param[6], param[7]);
 568                 BUG_ON(rc != H_SUCCESS);
 569         }
 570
 571         if (lock_tlbie)
 572                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
 573 }
 574
 575 static int __init disable_bulk_remove(char *str)
 576 {
 577         if (strcmp(str, "off") == 0 &&
 578             firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 579                         printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
 580                         powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
 581         }
 582         return 1;
 583 }
 584
 585 __setup("bulk_remove=", disable_bulk_remove);
 586
 587 void __init hpte_init_lpar(void)
 588 {
 589         ppc_md.hpte_invalidate  = pSeries_lpar_hpte_invalidate;
 590         ppc_md.hpte_updatepp    = pSeries_lpar_hpte_updatepp;
 591         ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
 592         ppc_md.hpte_insert      = pSeries_lpar_hpte_insert;
 593         ppc_md.hpte_remove      = pSeries_lpar_hpte_remove;
 594         ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
 595         ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range;
 596         ppc_md.hpte_clear_all   = pSeries_lpar_hptab_clear;
 597         ppc_md.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
 598 }
 599
 600 #ifdef CONFIG_PPC_SMLPAR
 601 #define CMO_FREE_HINT_DEFAULT 1
 602 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
 603
 604 static int __init cmo_free_hint(char *str)
 605 {
 606         char *parm;
 607         parm = strstrip(str);
 608
 609         if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
 610                 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
 611                 cmo_free_hint_flag = 0;
 612                 return 1;
 613         }
 614
 615         cmo_free_hint_flag = 1;
 616         printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
 617
 618         if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
 619                 return 1;
 620
 621         return 0;
 622 }
 623
 624 __setup("cmo_free_hint=", cmo_free_hint);
 625
 626 static void pSeries_set_page_state(struct page *page, int order,
 627                                    unsigned long state)
 628 {
 629         int i, j;
 630         unsigned long cmo_page_sz, addr;
 631
 632         cmo_page_sz = cmo_get_page_size();
 633         addr = __pa((unsigned long)page_address(page));
 634
 635         for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
 636                 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
 637                         plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
 638         }
 639 }
 640
 641 void arch_free_page(struct page *page, int order)
 642 {
 643         if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
 644                 return;
 645
 646         pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
 647 }
 648 EXPORT_SYMBOL(arch_free_page);
 649
 650 #endif
 651
 652 #ifdef CONFIG_TRACEPOINTS
 653 #ifdef CONFIG_JUMP_LABEL
 654 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
 655
 656 void hcall_tracepoint_regfunc(void)
 657 {
 658         static_key_slow_inc(&hcall_tracepoint_key);
 659 }
 660
 661 void hcall_tracepoint_unregfunc(void)
 662 {
 663         static_key_slow_dec(&hcall_tracepoint_key);
 664 }
 665 #else
 666 /*
 667  * We optimise our hcall path by placing hcall_tracepoint_refcount
 668  * directly in the TOC so we can check if the hcall tracepoints are
 669  * enabled via a single load.
 670  */
 671
 672 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
 673 extern long hcall_tracepoint_refcount;
 674
 675 void hcall_tracepoint_regfunc(void)
 676 {
 677         hcall_tracepoint_refcount++;
 678 }
 679
 680 void hcall_tracepoint_unregfunc(void)
 681 {
 682         hcall_tracepoint_refcount--;
 683 }
 684 #endif
 685
 686 /*
 687  * Since the tracing code might execute hcalls we need to guard against
 688  * recursion. One example of this are spinlocks calling H_YIELD on
 689  * shared processor partitions.
 690  */
 691 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
 692
 693
 694 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
 695 {
 696         unsigned long flags;
 697         unsigned int *depth;
 698
 699         /*
 700          * We cannot call tracepoints inside RCU idle regions which
 701          * means we must not trace H_CEDE.
 702          */
 703         if (opcode == H_CEDE)
 704                 return;
 705
 706         local_irq_save(flags);
 707
 708         depth = &__get_cpu_var(hcall_trace_depth);
 709
 710         if (*depth)
 711                 goto out;
 712
 713         (*depth)++;
 714         preempt_disable();
 715         trace_hcall_entry(opcode, args);
 716         (*depth)--;
 717
 718 out:
 719         local_irq_restore(flags);
 720 }
 721
 722 void __trace_hcall_exit(long opcode, unsigned long retval,
 723                         unsigned long *retbuf)
 724 {
 725         unsigned long flags;
 726         unsigned int *depth;
 727
 728         if (opcode == H_CEDE)
 729                 return;
 730
 731         local_irq_save(flags);
 732
 733         depth = &__get_cpu_var(hcall_trace_depth);
 734
 735         if (*depth)
 736                 goto out;
 737
 738         (*depth)++;
 739         trace_hcall_exit(opcode, retval, retbuf);
 740         preempt_enable();
 741         (*depth)--;
 742
 743 out:
 744         local_irq_restore(flags);
 745 }
 746 #endif
 747
 748 /**
 749  * h_get_mpp
 750  * H_GET_MPP hcall returns info in 7 parms
 751  */
 752 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
 753 {
 754         int rc;
 755         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
 756
 757         rc = plpar_hcall9(H_GET_MPP, retbuf);
 758
 759         mpp_data->entitled_mem = retbuf[0];
 760         mpp_data->mapped_mem = retbuf[1];
 761
 762         mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
 763         mpp_data->pool_num = retbuf[2] & 0xffff;
 764
 765         mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
 766         mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
 767         mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
 768
 769         mpp_data->pool_size = retbuf[4];
 770         mpp_data->loan_request = retbuf[5];
 771         mpp_data->backing_mem = retbuf[6];
 772
 773         return rc;
 774 }
 775 EXPORT_SYMBOL(h_get_mpp);
 776
 777 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
 778 {
 779         int rc;
 780         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
 781
 782         rc = plpar_hcall9(H_GET_MPP_X, retbuf);
 783
 784         mpp_x_data->coalesced_bytes = retbuf[0];
 785         mpp_x_data->pool_coalesced_bytes = retbuf[1];
 786         mpp_x_data->pool_purr_cycles = retbuf[2];
 787         mpp_x_data->pool_spurr_cycles = retbuf[3];
 788
 789         return rc;
 790 }