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