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 #define pr_fmt(fmt) "lpar: " fmt
  25
  26 #include <linux/kernel.h>
  27 #include <linux/dma-mapping.h>
  28 #include <linux/console.h>
  29 #include <linux/export.h>
  30 #include <linux/jump_label.h>
  31 #include <linux/delay.h>
  32 #include <linux/stop_machine.h>
  33 #include <asm/processor.h>
  34 #include <asm/mmu.h>
  35 #include <asm/page.h>
  36 #include <asm/pgtable.h>
  37 #include <asm/machdep.h>
  38 #include <asm/mmu_context.h>
  39 #include <asm/iommu.h>
  40 #include <asm/tlb.h>
  41 #include <asm/prom.h>
  42 #include <asm/cputable.h>
  43 #include <asm/udbg.h>
  44 #include <asm/smp.h>
  45 #include <asm/trace.h>
  46 #include <asm/firmware.h>
  47 #include <asm/plpar_wrappers.h>
  48 #include <asm/kexec.h>
  49 #include <asm/fadump.h>
  50 #include <asm/asm-prototypes.h>
  51 #include <asm/debugfs.h>
  52
  53 #include "pseries.h"
  54
  55 /* Flag bits for H_BULK_REMOVE */
  56 #define HBR_REQUEST     0x4000000000000000UL
  57 #define HBR_RESPONSE    0x8000000000000000UL
  58 #define HBR_END         0xc000000000000000UL
  59 #define HBR_AVPN        0x0200000000000000UL
  60 #define HBR_ANDCOND     0x0100000000000000UL
  61
  62
  63 /* in hvCall.S */
  64 EXPORT_SYMBOL(plpar_hcall);
  65 EXPORT_SYMBOL(plpar_hcall9);
  66 EXPORT_SYMBOL(plpar_hcall_norets);
  67
  68 void vpa_init(int cpu)
  69 {
  70         int hwcpu = get_hard_smp_processor_id(cpu);
  71         unsigned long addr;
  72         long ret;
  73         struct paca_struct *pp;
  74         struct dtl_entry *dtl;
  75
  76         /*
  77          * The spec says it "may be problematic" if CPU x registers the VPA of
  78          * CPU y. We should never do that, but wail if we ever do.
  79          */
  80         WARN_ON(cpu != smp_processor_id());
  81
  82         if (cpu_has_feature(CPU_FTR_ALTIVEC))
  83                 lppaca_of(cpu).vmxregs_in_use = 1;
  84
  85         if (cpu_has_feature(CPU_FTR_ARCH_207S))
  86                 lppaca_of(cpu).ebb_regs_in_use = 1;
  87
  88         addr = __pa(&lppaca_of(cpu));
  89         ret = register_vpa(hwcpu, addr);
  90
  91         if (ret) {
  92                 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
  93                        "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
  94                 return;
  95         }
  96
  97 #ifdef CONFIG_PPC_BOOK3S_64
  98         /*
  99          * PAPR says this feature is SLB-Buffer but firmware never
 100          * reports that.  All SPLPAR support SLB shadow buffer.
 101          */
 102         if (!radix_enabled() && firmware_has_feature(FW_FEATURE_SPLPAR)) {
 103                 addr = __pa(paca_ptrs[cpu]->slb_shadow_ptr);
 104                 ret = register_slb_shadow(hwcpu, addr);
 105                 if (ret)
 106                         pr_err("WARNING: SLB shadow buffer registration for "
 107                                "cpu %d (hw %d) of area %lx failed with %ld\n",
 108                                cpu, hwcpu, addr, ret);
 109         }
 110 #endif /* CONFIG_PPC_BOOK3S_64 */
 111
 112         /*
 113          * Register dispatch trace log, if one has been allocated.
 114          */
 115         pp = paca_ptrs[cpu];
 116         dtl = pp->dispatch_log;
 117         if (dtl) {
 118                 pp->dtl_ridx = 0;
 119                 pp->dtl_curr = dtl;
 120                 lppaca_of(cpu).dtl_idx = 0;
 121
 122                 /* hypervisor reads buffer length from this field */
 123                 dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
 124                 ret = register_dtl(hwcpu, __pa(dtl));
 125                 if (ret)
 126                         pr_err("WARNING: DTL registration of cpu %d (hw %d) "
 127                                "failed with %ld\n", smp_processor_id(),
 128                                hwcpu, ret);
 129                 lppaca_of(cpu).dtl_enable_mask = 2;
 130         }
 131 }
 132
 133 #ifdef CONFIG_PPC_BOOK3S_64
 134
 135 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
 136                                      unsigned long vpn, unsigned long pa,
 137                                      unsigned long rflags, unsigned long vflags,
 138                                      int psize, int apsize, int ssize)
 139 {
 140         unsigned long lpar_rc;
 141         unsigned long flags;
 142         unsigned long slot;
 143         unsigned long hpte_v, hpte_r;
 144
 145         if (!(vflags & HPTE_V_BOLTED))
 146                 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
 147                          "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
 148                          hpte_group, vpn,  pa, rflags, vflags, psize);
 149
 150         hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
 151         hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
 152
 153         if (!(vflags & HPTE_V_BOLTED))
 154                 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
 155
 156         /* Now fill in the actual HPTE */
 157         /* Set CEC cookie to 0         */
 158         /* Zero page = 0               */
 159         /* I-cache Invalidate = 0      */
 160         /* I-cache synchronize = 0     */
 161         /* Exact = 0                   */
 162         flags = 0;
 163
 164         if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
 165                 flags |= H_COALESCE_CAND;
 166
 167         lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
 168         if (unlikely(lpar_rc == H_PTEG_FULL)) {
 169                 pr_devel("Hash table group is full\n");
 170                 return -1;
 171         }
 172
 173         /*
 174          * Since we try and ioremap PHBs we don't own, the pte insert
 175          * will fail. However we must catch the failure in hash_page
 176          * or we will loop forever, so return -2 in this case.
 177          */
 178         if (unlikely(lpar_rc != H_SUCCESS)) {
 179                 pr_err("Failed hash pte insert with error %ld\n", lpar_rc);
 180                 return -2;
 181         }
 182         if (!(vflags & HPTE_V_BOLTED))
 183                 pr_devel(" -> slot: %lu\n", slot & 7);
 184
 185         /* Because of iSeries, we have to pass down the secondary
 186          * bucket bit here as well
 187          */
 188         return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
 189 }
 190
 191 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
 192
 193 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
 194 {
 195         unsigned long slot_offset;
 196         unsigned long lpar_rc;
 197         int i;
 198         unsigned long dummy1, dummy2;
 199
 200         /* pick a random slot to start at */
 201         slot_offset = mftb() & 0x7;
 202
 203         for (i = 0; i < HPTES_PER_GROUP; i++) {
 204
 205                 /* don't remove a bolted entry */
 206                 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
 207                                            (0x1UL << 4), &dummy1, &dummy2);
 208                 if (lpar_rc == H_SUCCESS)
 209                         return i;
 210
 211                 /*
 212                  * The test for adjunct partition is performed before the
 213                  * ANDCOND test.  H_RESOURCE may be returned, so we need to
 214                  * check for that as well.
 215                  */
 216                 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
 217
 218                 slot_offset++;
 219                 slot_offset &= 0x7;
 220         }
 221
 222         return -1;
 223 }
 224
 225 static void manual_hpte_clear_all(void)
 226 {
 227         unsigned long size_bytes = 1UL << ppc64_pft_size;
 228         unsigned long hpte_count = size_bytes >> 4;
 229         struct {
 230                 unsigned long pteh;
 231                 unsigned long ptel;
 232         } ptes[4];
 233         long lpar_rc;
 234         unsigned long i, j;
 235
 236         /* Read in batches of 4,
 237          * invalidate only valid entries not in the VRMA
 238          * hpte_count will be a multiple of 4
 239          */
 240         for (i = 0; i < hpte_count; i += 4) {
 241                 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
 242                 if (lpar_rc != H_SUCCESS) {
 243                         pr_info("Failed to read hash page table at %ld err %ld\n",
 244                                 i, lpar_rc);
 245                         continue;
 246                 }
 247                 for (j = 0; j < 4; j++){
 248                         if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
 249                                 HPTE_V_VRMA_MASK)
 250                                 continue;
 251                         if (ptes[j].pteh & HPTE_V_VALID)
 252                                 plpar_pte_remove_raw(0, i + j, 0,
 253                                         &(ptes[j].pteh), &(ptes[j].ptel));
 254                 }
 255         }
 256 }
 257
 258 static int hcall_hpte_clear_all(void)
 259 {
 260         int rc;
 261
 262         do {
 263                 rc = plpar_hcall_norets(H_CLEAR_HPT);
 264         } while (rc == H_CONTINUE);
 265
 266         return rc;
 267 }
 268
 269 static void pseries_hpte_clear_all(void)
 270 {
 271         int rc;
 272
 273         rc = hcall_hpte_clear_all();
 274         if (rc != H_SUCCESS)
 275                 manual_hpte_clear_all();
 276
 277 #ifdef __LITTLE_ENDIAN__
 278         /*
 279          * Reset exceptions to big endian.
 280          *
 281          * FIXME this is a hack for kexec, we need to reset the exception
 282          * endian before starting the new kernel and this is a convenient place
 283          * to do it.
 284          *
 285          * This is also called on boot when a fadump happens. In that case we
 286          * must not change the exception endian mode.
 287          */
 288         if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active())
 289                 pseries_big_endian_exceptions();
 290 #endif
 291 }
 292
 293 /*
 294  * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
 295  * the low 3 bits of flags happen to line up.  So no transform is needed.
 296  * We can probably optimize here and assume the high bits of newpp are
 297  * already zero.  For now I am paranoid.
 298  */
 299 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
 300                                        unsigned long newpp,
 301                                        unsigned long vpn,
 302                                        int psize, int apsize,
 303                                        int ssize, unsigned long inv_flags)
 304 {
 305         unsigned long lpar_rc;
 306         unsigned long flags;
 307         unsigned long want_v;
 308
 309         want_v = hpte_encode_avpn(vpn, psize, ssize);
 310
 311         flags = (newpp & 7) | H_AVPN;
 312         if (mmu_has_feature(MMU_FTR_KERNEL_RO))
 313                 /* Move pp0 into bit 8 (IBM 55) */
 314                 flags |= (newpp & HPTE_R_PP0) >> 55;
 315
 316         pr_devel("    update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
 317                  want_v, slot, flags, psize);
 318
 319         lpar_rc = plpar_pte_protect(flags, slot, want_v);
 320
 321         if (lpar_rc == H_NOT_FOUND) {
 322                 pr_devel("not found !\n");
 323                 return -1;
 324         }
 325
 326         pr_devel("ok\n");
 327
 328         BUG_ON(lpar_rc != H_SUCCESS);
 329
 330         return 0;
 331 }
 332
 333 static long __pSeries_lpar_hpte_find(unsigned long want_v, unsigned long hpte_group)
 334 {
 335         long lpar_rc;
 336         unsigned long i, j;
 337         struct {
 338                 unsigned long pteh;
 339                 unsigned long ptel;
 340         } ptes[4];
 341
 342         for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
 343
 344                 lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
 345                 if (lpar_rc != H_SUCCESS) {
 346                         pr_info("Failed to read hash page table at %ld err %ld\n",
 347                                 hpte_group, lpar_rc);
 348                         continue;
 349                 }
 350
 351                 for (j = 0; j < 4; j++) {
 352                         if (HPTE_V_COMPARE(ptes[j].pteh, want_v) &&
 353                             (ptes[j].pteh & HPTE_V_VALID))
 354                                 return i + j;
 355                 }
 356         }
 357
 358         return -1;
 359 }
 360
 361 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
 362 {
 363         long slot;
 364         unsigned long hash;
 365         unsigned long want_v;
 366         unsigned long hpte_group;
 367
 368         hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
 369         want_v = hpte_encode_avpn(vpn, psize, ssize);
 370
 371         /* Bolted entries are always in the primary group */
 372         hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 373         slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
 374         if (slot < 0)
 375                 return -1;
 376         return hpte_group + slot;
 377 }
 378
 379 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
 380                                              unsigned long ea,
 381                                              int psize, int ssize)
 382 {
 383         unsigned long vpn;
 384         unsigned long lpar_rc, slot, vsid, flags;
 385
 386         vsid = get_kernel_vsid(ea, ssize);
 387         vpn = hpt_vpn(ea, vsid, ssize);
 388
 389         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
 390         BUG_ON(slot == -1);
 391
 392         flags = newpp & 7;
 393         if (mmu_has_feature(MMU_FTR_KERNEL_RO))
 394                 /* Move pp0 into bit 8 (IBM 55) */
 395                 flags |= (newpp & HPTE_R_PP0) >> 55;
 396
 397         lpar_rc = plpar_pte_protect(flags, slot, 0);
 398
 399         BUG_ON(lpar_rc != H_SUCCESS);
 400 }
 401
 402 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
 403                                          int psize, int apsize,
 404                                          int ssize, int local)
 405 {
 406         unsigned long want_v;
 407         unsigned long lpar_rc;
 408         unsigned long dummy1, dummy2;
 409
 410         pr_devel("    inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
 411                  slot, vpn, psize, local);
 412
 413         want_v = hpte_encode_avpn(vpn, psize, ssize);
 414         lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
 415         if (lpar_rc == H_NOT_FOUND)
 416                 return;
 417
 418         BUG_ON(lpar_rc != H_SUCCESS);
 419 }
 420
 421 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 422 /*
 423  * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
 424  * to make sure that we avoid bouncing the hypervisor tlbie lock.
 425  */
 426 #define PPC64_HUGE_HPTE_BATCH 12
 427
 428 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
 429                                              unsigned long *vpn, int count,
 430                                              int psize, int ssize)
 431 {
 432         unsigned long param[PLPAR_HCALL9_BUFSIZE];
 433         int i = 0, pix = 0, rc;
 434         unsigned long flags = 0;
 435         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
 436
 437         if (lock_tlbie)
 438                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
 439
 440         for (i = 0; i < count; i++) {
 441
 442                 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 443                         pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
 444                                                      ssize, 0);
 445                 } else {
 446                         param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
 447                         param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
 448                         pix += 2;
 449                         if (pix == 8) {
 450                                 rc = plpar_hcall9(H_BULK_REMOVE, param,
 451                                                   param[0], param[1], param[2],
 452                                                   param[3], param[4], param[5],
 453                                                   param[6], param[7]);
 454                                 BUG_ON(rc != H_SUCCESS);
 455                                 pix = 0;
 456                         }
 457                 }
 458         }
 459         if (pix) {
 460                 param[pix] = HBR_END;
 461                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
 462                                   param[2], param[3], param[4], param[5],
 463                                   param[6], param[7]);
 464                 BUG_ON(rc != H_SUCCESS);
 465         }
 466
 467         if (lock_tlbie)
 468                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
 469 }
 470
 471 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
 472                                              unsigned long addr,
 473                                              unsigned char *hpte_slot_array,
 474                                              int psize, int ssize, int local)
 475 {
 476         int i, index = 0;
 477         unsigned long s_addr = addr;
 478         unsigned int max_hpte_count, valid;
 479         unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
 480         unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
 481         unsigned long shift, hidx, vpn = 0, hash, slot;
 482
 483         shift = mmu_psize_defs[psize].shift;
 484         max_hpte_count = 1U << (PMD_SHIFT - shift);
 485
 486         for (i = 0; i < max_hpte_count; i++) {
 487                 valid = hpte_valid(hpte_slot_array, i);
 488                 if (!valid)
 489                         continue;
 490                 hidx =  hpte_hash_index(hpte_slot_array, i);
 491
 492                 /* get the vpn */
 493                 addr = s_addr + (i * (1ul << shift));
 494                 vpn = hpt_vpn(addr, vsid, ssize);
 495                 hash = hpt_hash(vpn, shift, ssize);
 496                 if (hidx & _PTEIDX_SECONDARY)
 497                         hash = ~hash;
 498
 499                 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 500                 slot += hidx & _PTEIDX_GROUP_IX;
 501
 502                 slot_array[index] = slot;
 503                 vpn_array[index] = vpn;
 504                 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
 505                         /*
 506                          * Now do a bluk invalidate
 507                          */
 508                         __pSeries_lpar_hugepage_invalidate(slot_array,
 509                                                            vpn_array,
 510                                                            PPC64_HUGE_HPTE_BATCH,
 511                                                            psize, ssize);
 512                         index = 0;
 513                 } else
 514                         index++;
 515         }
 516         if (index)
 517                 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
 518                                                    index, psize, ssize);
 519 }
 520 #else
 521 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
 522                                              unsigned long addr,
 523                                              unsigned char *hpte_slot_array,
 524                                              int psize, int ssize, int local)
 525 {
 526         WARN(1, "%s called without THP support\n", __func__);
 527 }
 528 #endif
 529
 530 static int pSeries_lpar_hpte_removebolted(unsigned long ea,
 531                                           int psize, int ssize)
 532 {
 533         unsigned long vpn;
 534         unsigned long slot, vsid;
 535
 536         vsid = get_kernel_vsid(ea, ssize);
 537         vpn = hpt_vpn(ea, vsid, ssize);
 538
 539         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
 540         if (slot == -1)
 541                 return -ENOENT;
 542
 543         /*
 544          * lpar doesn't use the passed actual page size
 545          */
 546         pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
 547         return 0;
 548 }
 549
 550 /*
 551  * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
 552  * lock.
 553  */
 554 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
 555 {
 556         unsigned long vpn;
 557         unsigned long i, pix, rc;
 558         unsigned long flags = 0;
 559         struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
 560         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
 561         unsigned long param[PLPAR_HCALL9_BUFSIZE];
 562         unsigned long hash, index, shift, hidx, slot;
 563         real_pte_t pte;
 564         int psize, ssize;
 565
 566         if (lock_tlbie)
 567                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
 568
 569         psize = batch->psize;
 570         ssize = batch->ssize;
 571         pix = 0;
 572         for (i = 0; i < number; i++) {
 573                 vpn = batch->vpn[i];
 574                 pte = batch->pte[i];
 575                 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
 576                         hash = hpt_hash(vpn, shift, ssize);
 577                         hidx = __rpte_to_hidx(pte, index);
 578                         if (hidx & _PTEIDX_SECONDARY)
 579                                 hash = ~hash;
 580                         slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 581                         slot += hidx & _PTEIDX_GROUP_IX;
 582                         if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 583                                 /*
 584                                  * lpar doesn't use the passed actual page size
 585                                  */
 586                                 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
 587                                                              0, ssize, local);
 588                         } else {
 589                                 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
 590                                 param[pix+1] = hpte_encode_avpn(vpn, psize,
 591                                                                 ssize);
 592                                 pix += 2;
 593                                 if (pix == 8) {
 594                                         rc = plpar_hcall9(H_BULK_REMOVE, param,
 595                                                 param[0], param[1], param[2],
 596                                                 param[3], param[4], param[5],
 597                                                 param[6], param[7]);
 598                                         BUG_ON(rc != H_SUCCESS);
 599                                         pix = 0;
 600                                 }
 601                         }
 602                 } pte_iterate_hashed_end();
 603         }
 604         if (pix) {
 605                 param[pix] = HBR_END;
 606                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
 607                                   param[2], param[3], param[4], param[5],
 608                                   param[6], param[7]);
 609                 BUG_ON(rc != H_SUCCESS);
 610         }
 611
 612         if (lock_tlbie)
 613                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
 614 }
 615
 616 static int __init disable_bulk_remove(char *str)
 617 {
 618         if (strcmp(str, "off") == 0 &&
 619             firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 620                 pr_info("Disabling BULK_REMOVE firmware feature");
 621                 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
 622         }
 623         return 1;
 624 }
 625
 626 __setup("bulk_remove=", disable_bulk_remove);
 627
 628 #define HPT_RESIZE_TIMEOUT      10000 /* ms */
 629
 630 struct hpt_resize_state {
 631         unsigned long shift;
 632         int commit_rc;
 633 };
 634
 635 static int pseries_lpar_resize_hpt_commit(void *data)
 636 {
 637         struct hpt_resize_state *state = data;
 638
 639         state->commit_rc = plpar_resize_hpt_commit(0, state->shift);
 640         if (state->commit_rc != H_SUCCESS)
 641                 return -EIO;
 642
 643         /* Hypervisor has transitioned the HTAB, update our globals */
 644         ppc64_pft_size = state->shift;
 645         htab_size_bytes = 1UL << ppc64_pft_size;
 646         htab_hash_mask = (htab_size_bytes >> 7) - 1;
 647
 648         return 0;
 649 }
 650
 651 /*
 652  * Must be called in process context. The caller must hold the
 653  * cpus_lock.
 654  */
 655 static int pseries_lpar_resize_hpt(unsigned long shift)
 656 {
 657         struct hpt_resize_state state = {
 658                 .shift = shift,
 659                 .commit_rc = H_FUNCTION,
 660         };
 661         unsigned int delay, total_delay = 0;
 662         int rc;
 663         ktime_t t0, t1, t2;
 664
 665         might_sleep();
 666
 667         if (!firmware_has_feature(FW_FEATURE_HPT_RESIZE))
 668                 return -ENODEV;
 669
 670         pr_info("Attempting to resize HPT to shift %lu\n", shift);
 671
 672         t0 = ktime_get();
 673
 674         rc = plpar_resize_hpt_prepare(0, shift);
 675         while (H_IS_LONG_BUSY(rc)) {
 676                 delay = get_longbusy_msecs(rc);
 677                 total_delay += delay;
 678                 if (total_delay > HPT_RESIZE_TIMEOUT) {
 679                         /* prepare with shift==0 cancels an in-progress resize */
 680                         rc = plpar_resize_hpt_prepare(0, 0);
 681                         if (rc != H_SUCCESS)
 682                                 pr_warn("Unexpected error %d cancelling timed out HPT resize\n",
 683                                        rc);
 684                         return -ETIMEDOUT;
 685                 }
 686                 msleep(delay);
 687                 rc = plpar_resize_hpt_prepare(0, shift);
 688         };
 689
 690         switch (rc) {
 691         case H_SUCCESS:
 692                 /* Continue on */
 693                 break;
 694
 695         case H_PARAMETER:
 696                 return -EINVAL;
 697         case H_RESOURCE:
 698                 return -EPERM;
 699         default:
 700                 pr_warn("Unexpected error %d from H_RESIZE_HPT_PREPARE\n", rc);
 701                 return -EIO;
 702         }
 703
 704         t1 = ktime_get();
 705
 706         rc = stop_machine_cpuslocked(pseries_lpar_resize_hpt_commit,
 707                                      &state, NULL);
 708
 709         t2 = ktime_get();
 710
 711         if (rc != 0) {
 712                 switch (state.commit_rc) {
 713                 case H_PTEG_FULL:
 714                         pr_warn("Hash collision while resizing HPT\n");
 715                         return -ENOSPC;
 716
 717                 default:
 718                         pr_warn("Unexpected error %d from H_RESIZE_HPT_COMMIT\n",
 719                                 state.commit_rc);
 720                         return -EIO;
 721                 };
 722         }
 723
 724         pr_info("HPT resize to shift %lu complete (%lld ms / %lld ms)\n",
 725                 shift, (long long) ktime_ms_delta(t1, t0),
 726                 (long long) ktime_ms_delta(t2, t1));
 727
 728         return 0;
 729 }
 730
 731 static int pseries_lpar_register_process_table(unsigned long base,
 732                         unsigned long page_size, unsigned long table_size)
 733 {
 734         long rc;
 735         unsigned long flags = 0;
 736
 737         if (table_size)
 738                 flags |= PROC_TABLE_NEW;
 739         if (radix_enabled())
 740                 flags |= PROC_TABLE_RADIX | PROC_TABLE_GTSE;
 741         else
 742                 flags |= PROC_TABLE_HPT_SLB;
 743         for (;;) {
 744                 rc = plpar_hcall_norets(H_REGISTER_PROC_TBL, flags, base,
 745                                         page_size, table_size);
 746                 if (!H_IS_LONG_BUSY(rc))
 747                         break;
 748                 mdelay(get_longbusy_msecs(rc));
 749         }
 750         if (rc != H_SUCCESS) {
 751                 pr_err("Failed to register process table (rc=%ld)\n", rc);
 752                 BUG();
 753         }
 754         return rc;
 755 }
 756
 757 void __init hpte_init_pseries(void)
 758 {
 759         mmu_hash_ops.hpte_invalidate     = pSeries_lpar_hpte_invalidate;
 760         mmu_hash_ops.hpte_updatepp       = pSeries_lpar_hpte_updatepp;
 761         mmu_hash_ops.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
 762         mmu_hash_ops.hpte_insert         = pSeries_lpar_hpte_insert;
 763         mmu_hash_ops.hpte_remove         = pSeries_lpar_hpte_remove;
 764         mmu_hash_ops.hpte_removebolted   = pSeries_lpar_hpte_removebolted;
 765         mmu_hash_ops.flush_hash_range    = pSeries_lpar_flush_hash_range;
 766         mmu_hash_ops.hpte_clear_all      = pseries_hpte_clear_all;
 767         mmu_hash_ops.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
 768         register_process_table           = pseries_lpar_register_process_table;
 769
 770         if (firmware_has_feature(FW_FEATURE_HPT_RESIZE))
 771                 mmu_hash_ops.resize_hpt = pseries_lpar_resize_hpt;
 772 }
 773
 774 void radix_init_pseries(void)
 775 {
 776         pr_info("Using radix MMU under hypervisor\n");
 777         register_process_table = pseries_lpar_register_process_table;
 778 }
 779
 780 #ifdef CONFIG_PPC_SMLPAR
 781 #define CMO_FREE_HINT_DEFAULT 1
 782 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
 783
 784 static int __init cmo_free_hint(char *str)
 785 {
 786         char *parm;
 787         parm = strstrip(str);
 788
 789         if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
 790                 pr_info("%s: CMO free page hinting is not active.\n", __func__);
 791                 cmo_free_hint_flag = 0;
 792                 return 1;
 793         }
 794
 795         cmo_free_hint_flag = 1;
 796         pr_info("%s: CMO free page hinting is active.\n", __func__);
 797
 798         if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
 799                 return 1;
 800
 801         return 0;
 802 }
 803
 804 __setup("cmo_free_hint=", cmo_free_hint);
 805
 806 static void pSeries_set_page_state(struct page *page, int order,
 807                                    unsigned long state)
 808 {
 809         int i, j;
 810         unsigned long cmo_page_sz, addr;
 811
 812         cmo_page_sz = cmo_get_page_size();
 813         addr = __pa((unsigned long)page_address(page));
 814
 815         for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
 816                 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
 817                         plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
 818         }
 819 }
 820
 821 void arch_free_page(struct page *page, int order)
 822 {
 823         if (radix_enabled())
 824                 return;
 825         if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
 826                 return;
 827
 828         pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
 829 }
 830 EXPORT_SYMBOL(arch_free_page);
 831
 832 #endif /* CONFIG_PPC_SMLPAR */
 833 #endif /* CONFIG_PPC_BOOK3S_64 */
 834
 835 #ifdef CONFIG_TRACEPOINTS
 836 #ifdef CONFIG_JUMP_LABEL
 837 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
 838
 839 int hcall_tracepoint_regfunc(void)
 840 {
 841         static_key_slow_inc(&hcall_tracepoint_key);
 842         return 0;
 843 }
 844
 845 void hcall_tracepoint_unregfunc(void)
 846 {
 847         static_key_slow_dec(&hcall_tracepoint_key);
 848 }
 849 #else
 850 /*
 851  * We optimise our hcall path by placing hcall_tracepoint_refcount
 852  * directly in the TOC so we can check if the hcall tracepoints are
 853  * enabled via a single load.
 854  */
 855
 856 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
 857 extern long hcall_tracepoint_refcount;
 858
 859 int hcall_tracepoint_regfunc(void)
 860 {
 861         hcall_tracepoint_refcount++;
 862         return 0;
 863 }
 864
 865 void hcall_tracepoint_unregfunc(void)
 866 {
 867         hcall_tracepoint_refcount--;
 868 }
 869 #endif
 870
 871 /*
 872  * Since the tracing code might execute hcalls we need to guard against
 873  * recursion. One example of this are spinlocks calling H_YIELD on
 874  * shared processor partitions.
 875  */
 876 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
 877
 878
 879 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
 880 {
 881         unsigned long flags;
 882         unsigned int *depth;
 883
 884         /*
 885          * We cannot call tracepoints inside RCU idle regions which
 886          * means we must not trace H_CEDE.
 887          */
 888         if (opcode == H_CEDE)
 889                 return;
 890
 891         local_irq_save(flags);
 892
 893         depth = this_cpu_ptr(&hcall_trace_depth);
 894
 895         if (*depth)
 896                 goto out;
 897
 898         (*depth)++;
 899         preempt_disable();
 900         trace_hcall_entry(opcode, args);
 901         (*depth)--;
 902
 903 out:
 904         local_irq_restore(flags);
 905 }
 906
 907 void __trace_hcall_exit(long opcode, long retval, unsigned long *retbuf)
 908 {
 909         unsigned long flags;
 910         unsigned int *depth;
 911
 912         if (opcode == H_CEDE)
 913                 return;
 914
 915         local_irq_save(flags);
 916
 917         depth = this_cpu_ptr(&hcall_trace_depth);
 918
 919         if (*depth)
 920                 goto out;
 921
 922         (*depth)++;
 923         trace_hcall_exit(opcode, retval, retbuf);
 924         preempt_enable();
 925         (*depth)--;
 926
 927 out:
 928         local_irq_restore(flags);
 929 }
 930 #endif
 931
 932 /**
 933  * h_get_mpp
 934  * H_GET_MPP hcall returns info in 7 parms
 935  */
 936 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
 937 {
 938         int rc;
 939         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
 940
 941         rc = plpar_hcall9(H_GET_MPP, retbuf);
 942
 943         mpp_data->entitled_mem = retbuf[0];
 944         mpp_data->mapped_mem = retbuf[1];
 945
 946         mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
 947         mpp_data->pool_num = retbuf[2] & 0xffff;
 948
 949         mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
 950         mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
 951         mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
 952
 953         mpp_data->pool_size = retbuf[4];
 954         mpp_data->loan_request = retbuf[5];
 955         mpp_data->backing_mem = retbuf[6];
 956
 957         return rc;
 958 }
 959 EXPORT_SYMBOL(h_get_mpp);
 960
 961 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
 962 {
 963         int rc;
 964         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
 965
 966         rc = plpar_hcall9(H_GET_MPP_X, retbuf);
 967
 968         mpp_x_data->coalesced_bytes = retbuf[0];
 969         mpp_x_data->pool_coalesced_bytes = retbuf[1];
 970         mpp_x_data->pool_purr_cycles = retbuf[2];
 971         mpp_x_data->pool_spurr_cycles = retbuf[3];
 972
 973         return rc;
 974 }
 975
 976 static unsigned long vsid_unscramble(unsigned long vsid, int ssize)
 977 {
 978         unsigned long protovsid;
 979         unsigned long va_bits = VA_BITS;
 980         unsigned long modinv, vsid_modulus;
 981         unsigned long max_mod_inv, tmp_modinv;
 982
 983         if (!mmu_has_feature(MMU_FTR_68_BIT_VA))
 984                 va_bits = 65;
 985
 986         if (ssize == MMU_SEGSIZE_256M) {
 987                 modinv = VSID_MULINV_256M;
 988                 vsid_modulus = ((1UL << (va_bits - SID_SHIFT)) - 1);
 989         } else {
 990                 modinv = VSID_MULINV_1T;
 991                 vsid_modulus = ((1UL << (va_bits - SID_SHIFT_1T)) - 1);
 992         }
 993
 994         /*
 995          * vsid outside our range.
 996          */
 997         if (vsid >= vsid_modulus)
 998                 return 0;
 999
1000         /*
1001          * If modinv is the modular multiplicate inverse of (x % vsid_modulus)
1002          * and vsid = (protovsid * x) % vsid_modulus, then we say:
1003          *   protovsid = (vsid * modinv) % vsid_modulus
1004          */
1005
1006         /* Check if (vsid * modinv) overflow (63 bits) */
1007         max_mod_inv = 0x7fffffffffffffffull / vsid;
1008         if (modinv < max_mod_inv)
1009                 return (vsid * modinv) % vsid_modulus;
1010
1011         tmp_modinv = modinv/max_mod_inv;
1012         modinv %= max_mod_inv;
1013
1014         protovsid = (((vsid * max_mod_inv) % vsid_modulus) * tmp_modinv) % vsid_modulus;
1015         protovsid = (protovsid + vsid * modinv) % vsid_modulus;
1016
1017         return protovsid;
1018 }
1019
1020 static int __init reserve_vrma_context_id(void)
1021 {
1022         unsigned long protovsid;
1023
1024         /*
1025          * Reserve context ids which map to reserved virtual addresses. For now
1026          * we only reserve the context id which maps to the VRMA VSID. We ignore
1027          * the addresses in "ibm,adjunct-virtual-addresses" because we don't
1028          * enable adjunct support via the "ibm,client-architecture-support"
1029          * interface.
1030          */
1031         protovsid = vsid_unscramble(VRMA_VSID, MMU_SEGSIZE_1T);
1032         hash__reserve_context_id(protovsid >> ESID_BITS_1T);
1033         return 0;
1034 }
1035 machine_device_initcall(pseries, reserve_vrma_context_id);
1036
1037 #ifdef CONFIG_DEBUG_FS
1038 /* debugfs file interface for vpa data */
1039 static ssize_t vpa_file_read(struct file *filp, char __user *buf, size_t len,
1040                               loff_t *pos)
1041 {
1042         int cpu = (long)filp->private_data;
1043         struct lppaca *lppaca = &lppaca_of(cpu);
1044
1045         return simple_read_from_buffer(buf, len, pos, lppaca,
1046                                 sizeof(struct lppaca));
1047 }
1048
1049 static const struct file_operations vpa_fops = {
1050         .open           = simple_open,
1051         .read           = vpa_file_read,
1052         .llseek         = default_llseek,
1053 };
1054
1055 static int __init vpa_debugfs_init(void)
1056 {
1057         char name[16];
1058         long i;
1059         struct dentry *vpa_dir;
1060
1061         if (!firmware_has_feature(FW_FEATURE_SPLPAR))
1062                 return 0;
1063
1064         vpa_dir = debugfs_create_dir("vpa", powerpc_debugfs_root);
1065         if (!vpa_dir) {
1066                 pr_warn("%s: can't create vpa root dir\n", __func__);
1067                 return -ENOMEM;
1068         }
1069
1070         /* set up the per-cpu vpa file*/
1071         for_each_possible_cpu(i) {
1072                 struct dentry *d;
1073
1074                 sprintf(name, "cpu-%ld", i);
1075
1076                 d = debugfs_create_file(name, 0400, vpa_dir, (void *)i,
1077                                         &vpa_fops);
1078                 if (!d) {
1079                         pr_warn("%s: can't create per-cpu vpa file\n",
1080                                         __func__);
1081                         return -ENOMEM;
1082                 }
1083         }
1084
1085         return 0;
1086 }
1087 machine_arch_initcall(pseries, vpa_debugfs_init);
1088 #endif /* CONFIG_DEBUG_FS */