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 <asm/processor.h>
  30 #include <asm/mmu.h>
  31 #include <asm/page.h>
  32 #include <asm/pgtable.h>
  33 #include <asm/machdep.h>
  34 #include <asm/mmu_context.h>
  35 #include <asm/iommu.h>
  36 #include <asm/tlbflush.h>
  37 #include <asm/tlb.h>
  38 #include <asm/prom.h>
  39 #include <asm/cputable.h>
  40 #include <asm/udbg.h>
  41 #include <asm/smp.h>
  42 #include <asm/trace.h>
  43 #include <asm/firmware.h>
  44 #include <asm/plpar_wrappers.h>
  45
  46 #include "pseries.h"
  47
  48 /* Flag bits for H_BULK_REMOVE */
  49 #define HBR_REQUEST     0x4000000000000000UL
  50 #define HBR_RESPONSE    0x8000000000000000UL
  51 #define HBR_END         0xc000000000000000UL
  52 #define HBR_AVPN        0x0200000000000000UL
  53 #define HBR_ANDCOND     0x0100000000000000UL
  54
  55
  56 /* in hvCall.S */
  57 EXPORT_SYMBOL(plpar_hcall);
  58 EXPORT_SYMBOL(plpar_hcall9);
  59 EXPORT_SYMBOL(plpar_hcall_norets);
  60
  61 extern void pSeries_find_serial_port(void);
  62
  63 void vpa_init(int cpu)
  64 {
  65         int hwcpu = get_hard_smp_processor_id(cpu);
  66         unsigned long addr;
  67         long ret;
  68         struct paca_struct *pp;
  69         struct dtl_entry *dtl;
  70
  71         /*
  72          * The spec says it "may be problematic" if CPU x registers the VPA of
  73          * CPU y. We should never do that, but wail if we ever do.
  74          */
  75         WARN_ON(cpu != smp_processor_id());
  76
  77         if (cpu_has_feature(CPU_FTR_ALTIVEC))
  78                 lppaca_of(cpu).vmxregs_in_use = 1;
  79
  80         if (cpu_has_feature(CPU_FTR_ARCH_207S))
  81                 lppaca_of(cpu).ebb_regs_in_use = 1;
  82
  83         addr = __pa(&lppaca_of(cpu));
  84         ret = register_vpa(hwcpu, addr);
  85
  86         if (ret) {
  87                 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
  88                        "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
  89                 return;
  90         }
  91         /*
  92          * PAPR says this feature is SLB-Buffer but firmware never
  93          * reports that.  All SPLPAR support SLB shadow buffer.
  94          */
  95         addr = __pa(&slb_shadow[cpu]);
  96         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
  97                 ret = register_slb_shadow(hwcpu, addr);
  98                 if (ret)
  99                         pr_err("WARNING: SLB shadow buffer registration for "
 100                                "cpu %d (hw %d) of area %lx failed with %ld\n",
 101                                cpu, hwcpu, addr, ret);
 102         }
 103
 104         /*
 105          * Register dispatch trace log, if one has been allocated.
 106          */
 107         pp = &paca[cpu];
 108         dtl = pp->dispatch_log;
 109         if (dtl) {
 110                 pp->dtl_ridx = 0;
 111                 pp->dtl_curr = dtl;
 112                 lppaca_of(cpu).dtl_idx = 0;
 113
 114                 /* hypervisor reads buffer length from this field */
 115                 dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
 116                 ret = register_dtl(hwcpu, __pa(dtl));
 117                 if (ret)
 118                         pr_err("WARNING: DTL registration of cpu %d (hw %d) "
 119                                "failed with %ld\n", smp_processor_id(),
 120                                hwcpu, ret);
 121                 lppaca_of(cpu).dtl_enable_mask = 2;
 122         }
 123 }
 124
 125 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
 126                                      unsigned long vpn, unsigned long pa,
 127                                      unsigned long rflags, unsigned long vflags,
 128                                      int psize, int apsize, int ssize)
 129 {
 130         unsigned long lpar_rc;
 131         unsigned long flags;
 132         unsigned long slot;
 133         unsigned long hpte_v, hpte_r;
 134
 135         if (!(vflags & HPTE_V_BOLTED))
 136                 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
 137                          "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
 138                          hpte_group, vpn,  pa, rflags, vflags, psize);
 139
 140         hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
 141         hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
 142
 143         if (!(vflags & HPTE_V_BOLTED))
 144                 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
 145
 146         /* Now fill in the actual HPTE */
 147         /* Set CEC cookie to 0         */
 148         /* Zero page = 0               */
 149         /* I-cache Invalidate = 0      */
 150         /* I-cache synchronize = 0     */
 151         /* Exact = 0                   */
 152         flags = 0;
 153
 154         /* Make pHyp happy */
 155         if ((rflags & _PAGE_NO_CACHE) && !(rflags & _PAGE_WRITETHRU))
 156                 hpte_r &= ~_PAGE_COHERENT;
 157         if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
 158                 flags |= H_COALESCE_CAND;
 159
 160         lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
 161         if (unlikely(lpar_rc == H_PTEG_FULL)) {
 162                 if (!(vflags & HPTE_V_BOLTED))
 163                         pr_devel(" full\n");
 164                 return -1;
 165         }
 166
 167         /*
 168          * Since we try and ioremap PHBs we don't own, the pte insert
 169          * will fail. However we must catch the failure in hash_page
 170          * or we will loop forever, so return -2 in this case.
 171          */
 172         if (unlikely(lpar_rc != H_SUCCESS)) {
 173                 if (!(vflags & HPTE_V_BOLTED))
 174                         pr_devel(" lpar err %ld\n", lpar_rc);
 175                 return -2;
 176         }
 177         if (!(vflags & HPTE_V_BOLTED))
 178                 pr_devel(" -> slot: %lu\n", slot & 7);
 179
 180         /* Because of iSeries, we have to pass down the secondary
 181          * bucket bit here as well
 182          */
 183         return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
 184 }
 185
 186 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
 187
 188 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
 189 {
 190         unsigned long slot_offset;
 191         unsigned long lpar_rc;
 192         int i;
 193         unsigned long dummy1, dummy2;
 194
 195         /* pick a random slot to start at */
 196         slot_offset = mftb() & 0x7;
 197
 198         for (i = 0; i < HPTES_PER_GROUP; i++) {
 199
 200                 /* don't remove a bolted entry */
 201                 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
 202                                            (0x1UL << 4), &dummy1, &dummy2);
 203                 if (lpar_rc == H_SUCCESS)
 204                         return i;
 205
 206                 /*
 207                  * The test for adjunct partition is performed before the
 208                  * ANDCOND test.  H_RESOURCE may be returned, so we need to
 209                  * check for that as well.
 210                  */
 211                 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
 212
 213                 slot_offset++;
 214                 slot_offset &= 0x7;
 215         }
 216
 217         return -1;
 218 }
 219
 220 static void pSeries_lpar_hptab_clear(void)
 221 {
 222         unsigned long size_bytes = 1UL << ppc64_pft_size;
 223         unsigned long hpte_count = size_bytes >> 4;
 224         struct {
 225                 unsigned long pteh;
 226                 unsigned long ptel;
 227         } ptes[4];
 228         long lpar_rc;
 229         unsigned long i, j;
 230
 231         /* Read in batches of 4,
 232          * invalidate only valid entries not in the VRMA
 233          * hpte_count will be a multiple of 4
 234          */
 235         for (i = 0; i < hpte_count; i += 4) {
 236                 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
 237                 if (lpar_rc != H_SUCCESS)
 238                         continue;
 239                 for (j = 0; j < 4; j++){
 240                         if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
 241                                 HPTE_V_VRMA_MASK)
 242                                 continue;
 243                         if (ptes[j].pteh & HPTE_V_VALID)
 244                                 plpar_pte_remove_raw(0, i + j, 0,
 245                                         &(ptes[j].pteh), &(ptes[j].ptel));
 246                 }
 247         }
 248
 249 #ifdef __LITTLE_ENDIAN__
 250         /* Reset exceptions to big endian */
 251         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
 252                 long rc;
 253
 254                 rc = pseries_big_endian_exceptions();
 255                 /*
 256                  * At this point it is unlikely panic() will get anything
 257                  * out to the user, but at least this will stop us from
 258                  * continuing on further and creating an even more
 259                  * difficult to debug situation.
 260                  */
 261                 if (rc)
 262                         panic("Could not enable big endian exceptions");
 263         }
 264 #endif
 265 }
 266
 267 /*
 268  * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
 269  * the low 3 bits of flags happen to line up.  So no transform is needed.
 270  * We can probably optimize here and assume the high bits of newpp are
 271  * already zero.  For now I am paranoid.
 272  */
 273 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
 274                                        unsigned long newpp,
 275                                        unsigned long vpn,
 276                                        int psize, int apsize,
 277                                        int ssize, int local)
 278 {
 279         unsigned long lpar_rc;
 280         unsigned long flags = (newpp & 7) | H_AVPN;
 281         unsigned long want_v;
 282
 283         want_v = hpte_encode_avpn(vpn, psize, ssize);
 284
 285         pr_devel("    update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
 286                  want_v, slot, flags, psize);
 287
 288         lpar_rc = plpar_pte_protect(flags, slot, want_v);
 289
 290         if (lpar_rc == H_NOT_FOUND) {
 291                 pr_devel("not found !\n");
 292                 return -1;
 293         }
 294
 295         pr_devel("ok\n");
 296
 297         BUG_ON(lpar_rc != H_SUCCESS);
 298
 299         return 0;
 300 }
 301
 302 static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
 303 {
 304         unsigned long dword0;
 305         unsigned long lpar_rc;
 306         unsigned long dummy_word1;
 307         unsigned long flags;
 308
 309         /* Read 1 pte at a time                        */
 310         /* Do not need RPN to logical page translation */
 311         /* No cross CEC PFT access                     */
 312         flags = 0;
 313
 314         lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
 315
 316         BUG_ON(lpar_rc != H_SUCCESS);
 317
 318         return dword0;
 319 }
 320
 321 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
 322 {
 323         unsigned long hash;
 324         unsigned long i;
 325         long slot;
 326         unsigned long want_v, hpte_v;
 327
 328         hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
 329         want_v = hpte_encode_avpn(vpn, psize, ssize);
 330
 331         /* Bolted entries are always in the primary group */
 332         slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 333         for (i = 0; i < HPTES_PER_GROUP; i++) {
 334                 hpte_v = pSeries_lpar_hpte_getword0(slot);
 335
 336                 if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
 337                         /* HPTE matches */
 338                         return slot;
 339                 ++slot;
 340         }
 341
 342         return -1;
 343 }
 344
 345 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
 346                                              unsigned long ea,
 347                                              int psize, int ssize)
 348 {
 349         unsigned long vpn;
 350         unsigned long lpar_rc, slot, vsid, flags;
 351
 352         vsid = get_kernel_vsid(ea, ssize);
 353         vpn = hpt_vpn(ea, vsid, ssize);
 354
 355         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
 356         BUG_ON(slot == -1);
 357
 358         flags = newpp & 7;
 359         lpar_rc = plpar_pte_protect(flags, slot, 0);
 360
 361         BUG_ON(lpar_rc != H_SUCCESS);
 362 }
 363
 364 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
 365                                          int psize, int apsize,
 366                                          int ssize, int local)
 367 {
 368         unsigned long want_v;
 369         unsigned long lpar_rc;
 370         unsigned long dummy1, dummy2;
 371
 372         pr_devel("    inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
 373                  slot, vpn, psize, local);
 374
 375         want_v = hpte_encode_avpn(vpn, psize, ssize);
 376         lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
 377         if (lpar_rc == H_NOT_FOUND)
 378                 return;
 379
 380         BUG_ON(lpar_rc != H_SUCCESS);
 381 }
 382
 383 /*
 384  * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
 385  * to make sure that we avoid bouncing the hypervisor tlbie lock.
 386  */
 387 #define PPC64_HUGE_HPTE_BATCH 12
 388
 389 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
 390                                              unsigned long *vpn, int count,
 391                                              int psize, int ssize)
 392 {
 393         unsigned long param[8];
 394         int i = 0, pix = 0, rc;
 395         unsigned long flags = 0;
 396         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
 397
 398         if (lock_tlbie)
 399                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
 400
 401         for (i = 0; i < count; i++) {
 402
 403                 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 404                         pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
 405                                                      ssize, 0);
 406                 } else {
 407                         param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
 408                         param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
 409                         pix += 2;
 410                         if (pix == 8) {
 411                                 rc = plpar_hcall9(H_BULK_REMOVE, param,
 412                                                   param[0], param[1], param[2],
 413                                                   param[3], param[4], param[5],
 414                                                   param[6], param[7]);
 415                                 BUG_ON(rc != H_SUCCESS);
 416                                 pix = 0;
 417                         }
 418                 }
 419         }
 420         if (pix) {
 421                 param[pix] = HBR_END;
 422                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
 423                                   param[2], param[3], param[4], param[5],
 424                                   param[6], param[7]);
 425                 BUG_ON(rc != H_SUCCESS);
 426         }
 427
 428         if (lock_tlbie)
 429                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
 430 }
 431
 432 static void pSeries_lpar_hugepage_invalidate(struct mm_struct *mm,
 433                                        unsigned char *hpte_slot_array,
 434                                        unsigned long addr, int psize)
 435 {
 436         int ssize = 0, i, index = 0;
 437         unsigned long s_addr = addr;
 438         unsigned int max_hpte_count, valid;
 439         unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
 440         unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
 441         unsigned long shift, hidx, vpn = 0, vsid, hash, slot;
 442
 443         shift = mmu_psize_defs[psize].shift;
 444         max_hpte_count = 1U << (PMD_SHIFT - shift);
 445
 446         for (i = 0; i < max_hpte_count; i++) {
 447                 valid = hpte_valid(hpte_slot_array, i);
 448                 if (!valid)
 449                         continue;
 450                 hidx =  hpte_hash_index(hpte_slot_array, i);
 451
 452                 /* get the vpn */
 453                 addr = s_addr + (i * (1ul << shift));
 454                 if (!is_kernel_addr(addr)) {
 455                         ssize = user_segment_size(addr);
 456                         vsid = get_vsid(mm->context.id, addr, ssize);
 457                         WARN_ON(vsid == 0);
 458                 } else {
 459                         vsid = get_kernel_vsid(addr, mmu_kernel_ssize);
 460                         ssize = mmu_kernel_ssize;
 461                 }
 462
 463                 vpn = hpt_vpn(addr, vsid, ssize);
 464                 hash = hpt_hash(vpn, shift, ssize);
 465                 if (hidx & _PTEIDX_SECONDARY)
 466                         hash = ~hash;
 467
 468                 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 469                 slot += hidx & _PTEIDX_GROUP_IX;
 470
 471                 slot_array[index] = slot;
 472                 vpn_array[index] = vpn;
 473                 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
 474                         /*
 475                          * Now do a bluk invalidate
 476                          */
 477                         __pSeries_lpar_hugepage_invalidate(slot_array,
 478                                                            vpn_array,
 479                                                            PPC64_HUGE_HPTE_BATCH,
 480                                                            psize, ssize);
 481                         index = 0;
 482                 } else
 483                         index++;
 484         }
 485         if (index)
 486                 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
 487                                                    index, psize, ssize);
 488 }
 489
 490 static void pSeries_lpar_hpte_removebolted(unsigned long ea,
 491                                            int psize, int ssize)
 492 {
 493         unsigned long vpn;
 494         unsigned long slot, vsid;
 495
 496         vsid = get_kernel_vsid(ea, ssize);
 497         vpn = hpt_vpn(ea, vsid, ssize);
 498
 499         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
 500         BUG_ON(slot == -1);
 501         /*
 502          * lpar doesn't use the passed actual page size
 503          */
 504         pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
 505 }
 506
 507 /*
 508  * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
 509  * lock.
 510  */
 511 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
 512 {
 513         unsigned long vpn;
 514         unsigned long i, pix, rc;
 515         unsigned long flags = 0;
 516         struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
 517         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
 518         unsigned long param[9];
 519         unsigned long hash, index, shift, hidx, slot;
 520         real_pte_t pte;
 521         int psize, ssize;
 522
 523         if (lock_tlbie)
 524                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
 525
 526         psize = batch->psize;
 527         ssize = batch->ssize;
 528         pix = 0;
 529         for (i = 0; i < number; i++) {
 530                 vpn = batch->vpn[i];
 531                 pte = batch->pte[i];
 532                 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
 533                         hash = hpt_hash(vpn, shift, ssize);
 534                         hidx = __rpte_to_hidx(pte, index);
 535                         if (hidx & _PTEIDX_SECONDARY)
 536                                 hash = ~hash;
 537                         slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 538                         slot += hidx & _PTEIDX_GROUP_IX;
 539                         if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 540                                 /*
 541                                  * lpar doesn't use the passed actual page size
 542                                  */
 543                                 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
 544                                                              0, ssize, local);
 545                         } else {
 546                                 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
 547                                 param[pix+1] = hpte_encode_avpn(vpn, psize,
 548                                                                 ssize);
 549                                 pix += 2;
 550                                 if (pix == 8) {
 551                                         rc = plpar_hcall9(H_BULK_REMOVE, param,
 552                                                 param[0], param[1], param[2],
 553                                                 param[3], param[4], param[5],
 554                                                 param[6], param[7]);
 555                                         BUG_ON(rc != H_SUCCESS);
 556                                         pix = 0;
 557                                 }
 558                         }
 559                 } pte_iterate_hashed_end();
 560         }
 561         if (pix) {
 562                 param[pix] = HBR_END;
 563                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
 564                                   param[2], param[3], param[4], param[5],
 565                                   param[6], param[7]);
 566                 BUG_ON(rc != H_SUCCESS);
 567         }
 568
 569         if (lock_tlbie)
 570                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
 571 }
 572
 573 static int __init disable_bulk_remove(char *str)
 574 {
 575         if (strcmp(str, "off") == 0 &&
 576             firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 577                         printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
 578                         powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
 579         }
 580         return 1;
 581 }
 582
 583 __setup("bulk_remove=", disable_bulk_remove);
 584
 585 void __init hpte_init_lpar(void)
 586 {
 587         ppc_md.hpte_invalidate  = pSeries_lpar_hpte_invalidate;
 588         ppc_md.hpte_updatepp    = pSeries_lpar_hpte_updatepp;
 589         ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
 590         ppc_md.hpte_insert      = pSeries_lpar_hpte_insert;
 591         ppc_md.hpte_remove      = pSeries_lpar_hpte_remove;
 592         ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
 593         ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range;
 594         ppc_md.hpte_clear_all   = pSeries_lpar_hptab_clear;
 595         ppc_md.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
 596 }
 597
 598 #ifdef CONFIG_PPC_SMLPAR
 599 #define CMO_FREE_HINT_DEFAULT 1
 600 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
 601
 602 static int __init cmo_free_hint(char *str)
 603 {
 604         char *parm;
 605         parm = strstrip(str);
 606
 607         if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
 608                 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
 609                 cmo_free_hint_flag = 0;
 610                 return 1;
 611         }
 612
 613         cmo_free_hint_flag = 1;
 614         printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
 615
 616         if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
 617                 return 1;
 618
 619         return 0;
 620 }
 621
 622 __setup("cmo_free_hint=", cmo_free_hint);
 623
 624 static void pSeries_set_page_state(struct page *page, int order,
 625                                    unsigned long state)
 626 {
 627         int i, j;
 628         unsigned long cmo_page_sz, addr;
 629
 630         cmo_page_sz = cmo_get_page_size();
 631         addr = __pa((unsigned long)page_address(page));
 632
 633         for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
 634                 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
 635                         plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
 636         }
 637 }
 638
 639 void arch_free_page(struct page *page, int order)
 640 {
 641         if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
 642                 return;
 643
 644         pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
 645 }
 646 EXPORT_SYMBOL(arch_free_page);
 647
 648 #endif
 649
 650 #ifdef CONFIG_TRACEPOINTS
 651 /*
 652  * We optimise our hcall path by placing hcall_tracepoint_refcount
 653  * directly in the TOC so we can check if the hcall tracepoints are
 654  * enabled via a single load.
 655  */
 656
 657 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
 658 extern long hcall_tracepoint_refcount;
 659
 660 /*
 661  * Since the tracing code might execute hcalls we need to guard against
 662  * recursion. One example of this are spinlocks calling H_YIELD on
 663  * shared processor partitions.
 664  */
 665 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
 666
 667 void hcall_tracepoint_regfunc(void)
 668 {
 669         hcall_tracepoint_refcount++;
 670 }
 671
 672 void hcall_tracepoint_unregfunc(void)
 673 {
 674         hcall_tracepoint_refcount--;
 675 }
 676
 677 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
 678 {
 679         unsigned long flags;
 680         unsigned int *depth;
 681
 682         /*
 683          * We cannot call tracepoints inside RCU idle regions which
 684          * means we must not trace H_CEDE.
 685          */
 686         if (opcode == H_CEDE)
 687                 return;
 688
 689         local_irq_save(flags);
 690
 691         depth = &__get_cpu_var(hcall_trace_depth);
 692
 693         if (*depth)
 694                 goto out;
 695
 696         (*depth)++;
 697         preempt_disable();
 698         trace_hcall_entry(opcode, args);
 699         (*depth)--;
 700
 701 out:
 702         local_irq_restore(flags);
 703 }
 704
 705 void __trace_hcall_exit(long opcode, unsigned long retval,
 706                         unsigned long *retbuf)
 707 {
 708         unsigned long flags;
 709         unsigned int *depth;
 710
 711         if (opcode == H_CEDE)
 712                 return;
 713
 714         local_irq_save(flags);
 715
 716         depth = &__get_cpu_var(hcall_trace_depth);
 717
 718         if (*depth)
 719                 goto out;
 720
 721         (*depth)++;
 722         trace_hcall_exit(opcode, retval, retbuf);
 723         preempt_enable();
 724         (*depth)--;
 725
 726 out:
 727         local_irq_restore(flags);
 728 }
 729 #endif
 730
 731 /**
 732  * h_get_mpp
 733  * H_GET_MPP hcall returns info in 7 parms
 734  */
 735 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
 736 {
 737         int rc;
 738         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
 739
 740         rc = plpar_hcall9(H_GET_MPP, retbuf);
 741
 742         mpp_data->entitled_mem = retbuf[0];
 743         mpp_data->mapped_mem = retbuf[1];
 744
 745         mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
 746         mpp_data->pool_num = retbuf[2] & 0xffff;
 747
 748         mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
 749         mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
 750         mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
 751
 752         mpp_data->pool_size = retbuf[4];
 753         mpp_data->loan_request = retbuf[5];
 754         mpp_data->backing_mem = retbuf[6];
 755
 756         return rc;
 757 }
 758 EXPORT_SYMBOL(h_get_mpp);
 759
 760 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
 761 {
 762         int rc;
 763         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
 764
 765         rc = plpar_hcall9(H_GET_MPP_X, retbuf);
 766
 767         mpp_x_data->coalesced_bytes = retbuf[0];
 768         mpp_x_data->pool_coalesced_bytes = retbuf[1];
 769         mpp_x_data->pool_purr_cycles = retbuf[2];
 770         mpp_x_data->pool_spurr_cycles = retbuf[3];
 771
 772         return rc;
 773 }