target/ppc/mmu-radix64.c

   1 /*
   2  *  PowerPC Radix MMU mulation helpers for QEMU.
   3  *
   4  *  Copyright (c) 2016 Suraj Jitindar Singh, IBM Corporation
   5  *
   6  * This library is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2.1 of the License, or (at your option) any later version.
  10  *
  11  * This library is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  18  */
  19
  20 #include "qemu/osdep.h"
  21 #include "cpu.h"
  22 #include "exec/exec-all.h"
  23 #include "qemu/error-report.h"
  24 #include "sysemu/kvm.h"
  25 #include "kvm_ppc.h"
  26 #include "exec/log.h"
  27 #include "internal.h"
  28 #include "mmu-radix64.h"
  29 #include "mmu-book3s-v3.h"
  30
  31 static bool ppc_radix64_get_fully_qualified_addr(const CPUPPCState *env,
  32                                                  vaddr eaddr,
  33                                                  uint64_t *lpid, uint64_t *pid)
  34 {
  35     /* When EA(2:11) are nonzero, raise a segment interrupt */
  36     if (eaddr & ~R_EADDR_VALID_MASK) {
  37         return false;
  38     }
  39
  40     if (FIELD_EX64(env->msr, MSR, HV)) { /* MSR[HV] -> Hypervisor/bare metal */
  41         switch (eaddr & R_EADDR_QUADRANT) {
  42         case R_EADDR_QUADRANT0:
  43             *lpid = 0;
  44             *pid = env->spr[SPR_BOOKS_PID];
  45             break;
  46         case R_EADDR_QUADRANT1:
  47             *lpid = env->spr[SPR_LPIDR];
  48             *pid = env->spr[SPR_BOOKS_PID];
  49             break;
  50         case R_EADDR_QUADRANT2:
  51             *lpid = env->spr[SPR_LPIDR];
  52             *pid = 0;
  53             break;
  54         case R_EADDR_QUADRANT3:
  55             *lpid = 0;
  56             *pid = 0;
  57             break;
  58         default:
  59             g_assert_not_reached();
  60         }
  61     } else {  /* !MSR[HV] -> Guest */
  62         switch (eaddr & R_EADDR_QUADRANT) {
  63         case R_EADDR_QUADRANT0: /* Guest application */
  64             *lpid = env->spr[SPR_LPIDR];
  65             *pid = env->spr[SPR_BOOKS_PID];
  66             break;
  67         case R_EADDR_QUADRANT1: /* Illegal */
  68         case R_EADDR_QUADRANT2:
  69             return false;
  70         case R_EADDR_QUADRANT3: /* Guest OS */
  71             *lpid = env->spr[SPR_LPIDR];
  72             *pid = 0; /* pid set to 0 -> addresses guest operating system */
  73             break;
  74         default:
  75             g_assert_not_reached();
  76         }
  77     }
  78
  79     return true;
  80 }
  81
  82 static void ppc_radix64_raise_segi(PowerPCCPU *cpu, MMUAccessType access_type,
  83                                    vaddr eaddr)
  84 {
  85     CPUState *cs = CPU(cpu);
  86     CPUPPCState *env = &cpu->env;
  87
  88     switch (access_type) {
  89     case MMU_INST_FETCH:
  90         /* Instruction Segment Interrupt */
  91         cs->exception_index = POWERPC_EXCP_ISEG;
  92         break;
  93     case MMU_DATA_STORE:
  94     case MMU_DATA_LOAD:
  95         /* Data Segment Interrupt */
  96         cs->exception_index = POWERPC_EXCP_DSEG;
  97         env->spr[SPR_DAR] = eaddr;
  98         break;
  99     default:
 100         g_assert_not_reached();
 101     }
 102     env->error_code = 0;
 103 }
 104
 105 static inline const char *access_str(MMUAccessType access_type)
 106 {
 107     return access_type == MMU_DATA_LOAD ? "reading" :
 108         (access_type == MMU_DATA_STORE ? "writing" : "execute");
 109 }
 110
 111 static void ppc_radix64_raise_si(PowerPCCPU *cpu, MMUAccessType access_type,
 112                                  vaddr eaddr, uint32_t cause)
 113 {
 114     CPUState *cs = CPU(cpu);
 115     CPUPPCState *env = &cpu->env;
 116
 117     qemu_log_mask(CPU_LOG_MMU, "%s for %s @0x%"VADDR_PRIx" cause %08x\n",
 118                   __func__, access_str(access_type),
 119                   eaddr, cause);
 120
 121     switch (access_type) {
 122     case MMU_INST_FETCH:
 123         /* Instruction Storage Interrupt */
 124         cs->exception_index = POWERPC_EXCP_ISI;
 125         env->error_code = cause;
 126         break;
 127     case MMU_DATA_STORE:
 128         cause |= DSISR_ISSTORE;
 129         /* fall through */
 130     case MMU_DATA_LOAD:
 131         /* Data Storage Interrupt */
 132         cs->exception_index = POWERPC_EXCP_DSI;
 133         env->spr[SPR_DSISR] = cause;
 134         env->spr[SPR_DAR] = eaddr;
 135         env->error_code = 0;
 136         break;
 137     default:
 138         g_assert_not_reached();
 139     }
 140 }
 141
 142 static void ppc_radix64_raise_hsi(PowerPCCPU *cpu, MMUAccessType access_type,
 143                                   vaddr eaddr, hwaddr g_raddr, uint32_t cause)
 144 {
 145     CPUState *cs = CPU(cpu);
 146     CPUPPCState *env = &cpu->env;
 147
 148     env->error_code = 0;
 149     if (cause & DSISR_PRTABLE_FAULT) {
 150         /* HDSI PRTABLE_FAULT gets the originating access type in error_code */
 151         env->error_code = access_type;
 152         access_type = MMU_DATA_LOAD;
 153     }
 154
 155     qemu_log_mask(CPU_LOG_MMU, "%s for %s @0x%"VADDR_PRIx" 0x%"
 156                   HWADDR_PRIx" cause %08x\n",
 157                   __func__, access_str(access_type),
 158                   eaddr, g_raddr, cause);
 159
 160     switch (access_type) {
 161     case MMU_INST_FETCH:
 162         /* H Instruction Storage Interrupt */
 163         cs->exception_index = POWERPC_EXCP_HISI;
 164         env->spr[SPR_ASDR] = g_raddr;
 165         env->error_code = cause;
 166         break;
 167     case MMU_DATA_STORE:
 168         cause |= DSISR_ISSTORE;
 169         /* fall through */
 170     case MMU_DATA_LOAD:
 171         /* H Data Storage Interrupt */
 172         cs->exception_index = POWERPC_EXCP_HDSI;
 173         env->spr[SPR_HDSISR] = cause;
 174         env->spr[SPR_HDAR] = eaddr;
 175         env->spr[SPR_ASDR] = g_raddr;
 176         break;
 177     default:
 178         g_assert_not_reached();
 179     }
 180 }
 181
 182 static bool ppc_radix64_check_prot(PowerPCCPU *cpu, MMUAccessType access_type,
 183                                    uint64_t pte, int *fault_cause, int *prot,
 184                                    int mmu_idx, bool partition_scoped)
 185 {
 186     CPUPPCState *env = &cpu->env;
 187     int need_prot;
 188
 189     /* Check Page Attributes (pte58:59) */
 190     if ((pte & R_PTE_ATT) == R_PTE_ATT_NI_IO && access_type == MMU_INST_FETCH) {
 191         /*
 192          * Radix PTE entries with the non-idempotent I/O attribute are treated
 193          * as guarded storage
 194          */
 195         *fault_cause |= SRR1_NOEXEC_GUARD;
 196         return true;
 197     }
 198
 199     /* Determine permissions allowed by Encoded Access Authority */
 200     if (!partition_scoped && (pte & R_PTE_EAA_PRIV) &&
 201         FIELD_EX64(env->msr, MSR, PR)) {
 202         *prot = 0;
 203     } else if (mmuidx_pr(mmu_idx) || (pte & R_PTE_EAA_PRIV) ||
 204                partition_scoped) {
 205         *prot = ppc_radix64_get_prot_eaa(pte);
 206     } else { /* !MSR_PR && !(pte & R_PTE_EAA_PRIV) && !partition_scoped */
 207         *prot = ppc_radix64_get_prot_eaa(pte);
 208         *prot &= ppc_radix64_get_prot_amr(cpu); /* Least combined permissions */
 209     }
 210
 211     /* Check if requested access type is allowed */
 212     need_prot = prot_for_access_type(access_type);
 213     if (need_prot & ~*prot) { /* Page Protected for that Access */
 214         *fault_cause |= access_type == MMU_INST_FETCH ? SRR1_NOEXEC_GUARD :
 215                                                         DSISR_PROTFAULT;
 216         return true;
 217     }
 218
 219     return false;
 220 }
 221
 222 static int ppc_radix64_check_rc(MMUAccessType access_type, uint64_t pte)
 223 {
 224     switch (access_type) {
 225     case MMU_DATA_STORE:
 226         if (!(pte & R_PTE_C)) {
 227             break;
 228         }
 229         /* fall through */
 230     case MMU_INST_FETCH:
 231     case MMU_DATA_LOAD:
 232         if (!(pte & R_PTE_R)) {
 233             break;
 234         }
 235
 236         /* R/C bits are already set appropriately for this access */
 237         return 0;
 238     }
 239
 240     return 1;
 241 }
 242
 243 static bool ppc_radix64_is_valid_level(int level, int psize, uint64_t nls)
 244 {
 245     bool ret;
 246
 247     /*
 248      * Check if this is a valid level, according to POWER9 and POWER10
 249      * Processor User's Manuals, sections 4.10.4.1 and 5.10.6.1, respectively:
 250      * Supported Radix Tree Configurations and Resulting Page Sizes.
 251      *
 252      * Note: these checks are specific to POWER9 and POWER10 CPUs. Any future
 253      * CPUs that supports a different Radix MMU configuration will need their
 254      * own implementation.
 255      */
 256     switch (level) {
 257     case 0:     /* Root Page Dir */
 258         ret = psize == 52 && nls == 13;
 259         break;
 260     case 1:
 261     case 2:
 262         ret = nls == 9;
 263         break;
 264     case 3:
 265         ret = nls == 9 || nls == 5;
 266         break;
 267     default:
 268         ret = false;
 269     }
 270
 271     if (unlikely(!ret)) {
 272         qemu_log_mask(LOG_GUEST_ERROR, "invalid radix configuration: "
 273                       "level %d size %d nls %"PRIu64"\n",
 274                       level, psize, nls);
 275     }
 276     return ret;
 277 }
 278
 279 static int ppc_radix64_next_level(AddressSpace *as, vaddr eaddr,
 280                                   uint64_t *pte_addr, uint64_t *nls,
 281                                   int *psize, uint64_t *pte, int *fault_cause)
 282 {
 283     uint64_t index, mask, nlb, pde;
 284
 285     /* Read page <directory/table> entry from guest address space */
 286     pde = ldq_phys(as, *pte_addr);
 287     if (!(pde & R_PTE_VALID)) {         /* Invalid Entry */
 288         *fault_cause |= DSISR_NOPTE;
 289         return 1;
 290     }
 291
 292     *pte = pde;
 293     *psize -= *nls;
 294     if (!(pde & R_PTE_LEAF)) { /* Prepare for next iteration */
 295         *nls = pde & R_PDE_NLS;
 296         index = eaddr >> (*psize - *nls);       /* Shift */
 297         index &= ((1UL << *nls) - 1);           /* Mask */
 298         nlb = pde & R_PDE_NLB;
 299         mask = MAKE_64BIT_MASK(0, *nls + 3);
 300
 301         if (nlb & mask) {
 302             qemu_log_mask(LOG_GUEST_ERROR,
 303                 "%s: misaligned page dir/table base: 0x"TARGET_FMT_lx
 304                 " page dir size: 0x"TARGET_FMT_lx"\n",
 305                 __func__, nlb, mask + 1);
 306             nlb &= ~mask;
 307         }
 308         *pte_addr = nlb + index * sizeof(pde);
 309     }
 310     return 0;
 311 }
 312
 313 static int ppc_radix64_walk_tree(AddressSpace *as, vaddr eaddr,
 314                                  uint64_t base_addr, uint64_t nls,
 315                                  hwaddr *raddr, int *psize, uint64_t *pte,
 316                                  int *fault_cause, hwaddr *pte_addr)
 317 {
 318     uint64_t index, pde, rpn, mask;
 319     int level = 0;
 320
 321     index = eaddr >> (*psize - nls);    /* Shift */
 322     index &= ((1UL << nls) - 1);        /* Mask */
 323     mask = MAKE_64BIT_MASK(0, nls + 3);
 324
 325     if (base_addr & mask) {
 326         qemu_log_mask(LOG_GUEST_ERROR,
 327             "%s: misaligned page dir base: 0x"TARGET_FMT_lx
 328             " page dir size: 0x"TARGET_FMT_lx"\n",
 329             __func__, base_addr, mask + 1);
 330         base_addr &= ~mask;
 331     }
 332     *pte_addr = base_addr + index * sizeof(pde);
 333
 334     do {
 335         int ret;
 336
 337         if (!ppc_radix64_is_valid_level(level++, *psize, nls)) {
 338             *fault_cause |= DSISR_R_BADCONFIG;
 339             return 1;
 340         }
 341
 342         ret = ppc_radix64_next_level(as, eaddr, pte_addr, &nls, psize, &pde,
 343                                      fault_cause);
 344         if (ret) {
 345             return ret;
 346         }
 347     } while (!(pde & R_PTE_LEAF));
 348
 349     *pte = pde;
 350     rpn = pde & R_PTE_RPN;
 351     mask = (1UL << *psize) - 1;
 352
 353     /* Or high bits of rpn and low bits to ea to form whole real addr */
 354     *raddr = (rpn & ~mask) | (eaddr & mask);
 355     return 0;
 356 }
 357
 358 static bool validate_pate(PowerPCCPU *cpu, uint64_t lpid, ppc_v3_pate_t *pate)
 359 {
 360     CPUPPCState *env = &cpu->env;
 361
 362     if (!(pate->dw0 & PATE0_HR)) {
 363         return false;
 364     }
 365     if (lpid == 0 && !FIELD_EX64(env->msr, MSR, HV)) {
 366         return false;
 367     }
 368     if ((pate->dw0 & PATE1_R_PRTS) < 5) {
 369         return false;
 370     }
 371     /* More checks ... */
 372     return true;
 373 }
 374
 375 static int ppc_radix64_partition_scoped_xlate(PowerPCCPU *cpu,
 376                                               MMUAccessType orig_access_type,
 377                                               vaddr eaddr, hwaddr g_raddr,
 378                                               ppc_v3_pate_t pate,
 379                                               hwaddr *h_raddr, int *h_prot,
 380                                               int *h_page_size, bool pde_addr,
 381                                               int mmu_idx, uint64_t lpid,
 382                                               bool guest_visible)
 383 {
 384     MMUAccessType access_type = orig_access_type;
 385     int fault_cause = 0;
 386     hwaddr pte_addr;
 387     uint64_t pte;
 388
 389     if (pde_addr) {
 390         /*
 391          * Translation of process-scoped tables/directories is performed as
 392          * a read-access.
 393          */
 394         access_type = MMU_DATA_LOAD;
 395     }
 396
 397     qemu_log_mask(CPU_LOG_MMU, "%s for %s @0x%"VADDR_PRIx
 398                   " mmu_idx %u 0x%"HWADDR_PRIx"\n",
 399                   __func__, access_str(access_type),
 400                   eaddr, mmu_idx, g_raddr);
 401
 402     *h_page_size = PRTBE_R_GET_RTS(pate.dw0);
 403     /* No valid pte or access denied due to protection */
 404     if (ppc_radix64_walk_tree(CPU(cpu)->as, g_raddr, pate.dw0 & PRTBE_R_RPDB,
 405                               pate.dw0 & PRTBE_R_RPDS, h_raddr, h_page_size,
 406                               &pte, &fault_cause, &pte_addr) ||
 407         ppc_radix64_check_prot(cpu, access_type, pte,
 408                                &fault_cause, h_prot, mmu_idx, true)) {
 409         if (pde_addr) { /* address being translated was that of a guest pde */
 410             fault_cause |= DSISR_PRTABLE_FAULT;
 411         }
 412         if (guest_visible) {
 413             ppc_radix64_raise_hsi(cpu, orig_access_type,
 414                                   eaddr, g_raddr, fault_cause);
 415         }
 416         return 1;
 417     }
 418
 419     if (guest_visible) {
 420         if (ppc_radix64_check_rc(access_type, pte)) {
 421             /*
 422              * Per ISA 3.1 Book III, 7.5.3 and 7.5.5, failure to set R/C during
 423              * partition-scoped translation when effLPID = 0 results in normal
 424              * (non-Hypervisor) Data and Instruction Storage Interrupts
 425              * respectively.
 426              *
 427              * ISA 3.0 is ambiguous about this, but tests on POWER9 hardware
 428              * seem to exhibit the same behavior.
 429              */
 430             if (lpid > 0) {
 431                 ppc_radix64_raise_hsi(cpu, access_type, eaddr, g_raddr,
 432                                       DSISR_ATOMIC_RC);
 433             } else {
 434                 ppc_radix64_raise_si(cpu, access_type, eaddr, DSISR_ATOMIC_RC);
 435             }
 436             return 1;
 437         }
 438     }
 439
 440     return 0;
 441 }
 442
 443 /*
 444  * The spapr vhc has a flat partition scope provided by qemu memory when
 445  * not nested.
 446  *
 447  * When running a nested guest, the addressing is 2-level radix on top of the
 448  * vhc memory, so it works practically identically to the bare metal 2-level
 449  * radix. So that code is selected directly. A cleaner and more flexible nested
 450  * hypervisor implementation would allow the vhc to provide a ->nested_xlate()
 451  * function but that is not required for the moment.
 452  */
 453 static bool vhyp_flat_addressing(PowerPCCPU *cpu)
 454 {
 455     if (cpu->vhyp) {
 456         return !vhyp_cpu_in_nested(cpu);
 457     }
 458     return false;
 459 }
 460
 461 static int ppc_radix64_process_scoped_xlate(PowerPCCPU *cpu,
 462                                             MMUAccessType access_type,
 463                                             vaddr eaddr, uint64_t pid,
 464                                             ppc_v3_pate_t pate, hwaddr *g_raddr,
 465                                             int *g_prot, int *g_page_size,
 466                                             int mmu_idx, uint64_t lpid,
 467                                             bool guest_visible)
 468 {
 469     CPUState *cs = CPU(cpu);
 470     CPUPPCState *env = &cpu->env;
 471     uint64_t offset, size, prtb, prtbe_addr, prtbe0, base_addr, nls, index, pte;
 472     int fault_cause = 0, h_page_size, h_prot;
 473     hwaddr h_raddr, pte_addr;
 474     int ret;
 475
 476     qemu_log_mask(CPU_LOG_MMU, "%s for %s @0x%"VADDR_PRIx
 477                   " mmu_idx %u pid %"PRIu64"\n",
 478                   __func__, access_str(access_type),
 479                   eaddr, mmu_idx, pid);
 480
 481     prtb = (pate.dw1 & PATE1_R_PRTB);
 482     size = 1ULL << ((pate.dw1 & PATE1_R_PRTS) + 12);
 483     if (prtb & (size - 1)) {
 484         /* Process Table not properly aligned */
 485         if (guest_visible) {
 486             ppc_radix64_raise_si(cpu, access_type, eaddr, DSISR_R_BADCONFIG);
 487         }
 488         return 1;
 489     }
 490
 491     /* Index Process Table by PID to Find Corresponding Process Table Entry */
 492     offset = pid * sizeof(struct prtb_entry);
 493     if (offset >= size) {
 494         /* offset exceeds size of the process table */
 495         if (guest_visible) {
 496             ppc_radix64_raise_si(cpu, access_type, eaddr, DSISR_NOPTE);
 497         }
 498         return 1;
 499     }
 500     prtbe_addr = prtb + offset;
 501
 502     if (vhyp_flat_addressing(cpu)) {
 503         prtbe0 = ldq_phys(cs->as, prtbe_addr);
 504     } else {
 505         /*
 506          * Process table addresses are subject to partition-scoped
 507          * translation
 508          *
 509          * On a Radix host, the partition-scoped page table for LPID=0
 510          * is only used to translate the effective addresses of the
 511          * process table entries.
 512          */
 513         /* mmu_idx is 5 because we're translating from hypervisor scope */
 514         ret = ppc_radix64_partition_scoped_xlate(cpu, access_type, eaddr,
 515                                                  prtbe_addr, pate, &h_raddr,
 516                                                  &h_prot, &h_page_size, true,
 517                                                  5, lpid, guest_visible);
 518         if (ret) {
 519             return ret;
 520         }
 521         prtbe0 = ldq_phys(cs->as, h_raddr);
 522     }
 523
 524     /* Walk Radix Tree from Process Table Entry to Convert EA to RA */
 525     *g_page_size = PRTBE_R_GET_RTS(prtbe0);
 526     base_addr = prtbe0 & PRTBE_R_RPDB;
 527     nls = prtbe0 & PRTBE_R_RPDS;
 528     if (FIELD_EX64(env->msr, MSR, HV) || vhyp_flat_addressing(cpu)) {
 529         /*
 530          * Can treat process table addresses as real addresses
 531          */
 532         ret = ppc_radix64_walk_tree(cs->as, eaddr & R_EADDR_MASK, base_addr,
 533                                     nls, g_raddr, g_page_size, &pte,
 534                                     &fault_cause, &pte_addr);
 535         if (ret) {
 536             /* No valid PTE */
 537             if (guest_visible) {
 538                 ppc_radix64_raise_si(cpu, access_type, eaddr, fault_cause);
 539             }
 540             return ret;
 541         }
 542     } else {
 543         uint64_t rpn, mask;
 544         int level = 0;
 545
 546         index = (eaddr & R_EADDR_MASK) >> (*g_page_size - nls); /* Shift */
 547         index &= ((1UL << nls) - 1);                            /* Mask */
 548         pte_addr = base_addr + (index * sizeof(pte));
 549
 550         /*
 551          * Each process table address is subject to a partition-scoped
 552          * translation
 553          */
 554         do {
 555             /* mmu_idx is 5 because we're translating from hypervisor scope */
 556             ret = ppc_radix64_partition_scoped_xlate(cpu, access_type, eaddr,
 557                                                      pte_addr, pate, &h_raddr,
 558                                                      &h_prot, &h_page_size,
 559                                                      true, 5, lpid,
 560                                                      guest_visible);
 561             if (ret) {
 562                 return ret;
 563             }
 564
 565             if (!ppc_radix64_is_valid_level(level++, *g_page_size, nls)) {
 566                 fault_cause |= DSISR_R_BADCONFIG;
 567                 ret = 1;
 568             } else {
 569                 ret = ppc_radix64_next_level(cs->as, eaddr & R_EADDR_MASK,
 570                                              &h_raddr, &nls, g_page_size,
 571                                              &pte, &fault_cause);
 572             }
 573
 574             if (ret) {
 575                 /* No valid pte */
 576                 if (guest_visible) {
 577                     ppc_radix64_raise_si(cpu, access_type, eaddr, fault_cause);
 578                 }
 579                 return ret;
 580             }
 581             pte_addr = h_raddr;
 582         } while (!(pte & R_PTE_LEAF));
 583
 584         rpn = pte & R_PTE_RPN;
 585         mask = (1UL << *g_page_size) - 1;
 586
 587         /* Or high bits of rpn and low bits to ea to form whole real addr */
 588         *g_raddr = (rpn & ~mask) | (eaddr & mask);
 589     }
 590
 591     if (ppc_radix64_check_prot(cpu, access_type, pte, &fault_cause,
 592                                g_prot, mmu_idx, false)) {
 593         /* Access denied due to protection */
 594         if (guest_visible) {
 595             ppc_radix64_raise_si(cpu, access_type, eaddr, fault_cause);
 596         }
 597         return 1;
 598     }
 599
 600     if (guest_visible) {
 601         /* R/C bits not appropriately set for access */
 602         if (ppc_radix64_check_rc(access_type, pte)) {
 603             ppc_radix64_raise_si(cpu, access_type, eaddr, DSISR_ATOMIC_RC);
 604             return 1;
 605         }
 606     }
 607
 608     return 0;
 609 }
 610
 611 /*
 612  * Radix tree translation is a 2 steps translation process:
 613  *
 614  * 1. Process-scoped translation:   Guest Eff Addr  -> Guest Real Addr
 615  * 2. Partition-scoped translation: Guest Real Addr -> Host Real Addr
 616  *
 617  *                                  MSR[HV]
 618  *              +-------------+----------------+---------------+
 619  *              |             |     HV = 0     |     HV = 1    |
 620  *              +-------------+----------------+---------------+
 621  *              | Relocation  |    Partition   |      No       |
 622  *              | = Off       |     Scoped     |  Translation  |
 623  *  Relocation  +-------------+----------------+---------------+
 624  *              | Relocation  |   Partition &  |    Process    |
 625  *              | = On        | Process Scoped |    Scoped     |
 626  *              +-------------+----------------+---------------+
 627  */
 628 static bool ppc_radix64_xlate_impl(PowerPCCPU *cpu, vaddr eaddr,
 629                                    MMUAccessType access_type, hwaddr *raddr,
 630                                    int *psizep, int *protp, int mmu_idx,
 631                                    bool guest_visible)
 632 {
 633     CPUPPCState *env = &cpu->env;
 634     uint64_t lpid, pid;
 635     ppc_v3_pate_t pate;
 636     int psize, prot;
 637     hwaddr g_raddr;
 638     bool relocation;
 639
 640     assert(!(mmuidx_hv(mmu_idx) && cpu->vhyp));
 641
 642     relocation = !mmuidx_real(mmu_idx);
 643
 644     /* HV or virtual hypervisor Real Mode Access */
 645     if (!relocation && (mmuidx_hv(mmu_idx) || vhyp_flat_addressing(cpu))) {
 646         /* In real mode top 4 effective addr bits (mostly) ignored */
 647         *raddr = eaddr & 0x0FFFFFFFFFFFFFFFULL;
 648
 649         /* In HV mode, add HRMOR if top EA bit is clear */
 650         if (mmuidx_hv(mmu_idx) || !env->has_hv_mode) {
 651             if (!(eaddr >> 63)) {
 652                 *raddr |= env->spr[SPR_HRMOR];
 653            }
 654         }
 655         *protp = PAGE_READ | PAGE_WRITE | PAGE_EXEC;
 656         *psizep = TARGET_PAGE_BITS;
 657         return true;
 658     }
 659
 660     /*
 661      * Check UPRT (we avoid the check in real mode to deal with
 662      * transitional states during kexec.
 663      */
 664     if (guest_visible && !ppc64_use_proc_tbl(cpu)) {
 665         qemu_log_mask(LOG_GUEST_ERROR,
 666                       "LPCR:UPRT not set in radix mode ! LPCR="
 667                       TARGET_FMT_lx "\n", env->spr[SPR_LPCR]);
 668     }
 669
 670     /* Virtual Mode Access - get the fully qualified address */
 671     if (!ppc_radix64_get_fully_qualified_addr(&cpu->env, eaddr, &lpid, &pid)) {
 672         if (guest_visible) {
 673             ppc_radix64_raise_segi(cpu, access_type, eaddr);
 674         }
 675         return false;
 676     }
 677
 678     /* Get Partition Table */
 679     if (cpu->vhyp) {
 680         PPCVirtualHypervisorClass *vhc;
 681         vhc = PPC_VIRTUAL_HYPERVISOR_GET_CLASS(cpu->vhyp);
 682         if (!vhc->get_pate(cpu->vhyp, cpu, lpid, &pate)) {
 683             if (guest_visible) {
 684                 ppc_radix64_raise_hsi(cpu, access_type, eaddr, eaddr,
 685                                       DSISR_R_BADCONFIG);
 686             }
 687             return false;
 688         }
 689     } else {
 690         if (!ppc64_v3_get_pate(cpu, lpid, &pate)) {
 691             if (guest_visible) {
 692                 ppc_radix64_raise_hsi(cpu, access_type, eaddr, eaddr,
 693                                       DSISR_R_BADCONFIG);
 694             }
 695             return false;
 696         }
 697         if (!validate_pate(cpu, lpid, &pate)) {
 698             if (guest_visible) {
 699                 ppc_radix64_raise_hsi(cpu, access_type, eaddr, eaddr,
 700                                       DSISR_R_BADCONFIG);
 701             }
 702             return false;
 703         }
 704     }
 705
 706     *psizep = INT_MAX;
 707     *protp = PAGE_READ | PAGE_WRITE | PAGE_EXEC;
 708
 709     /*
 710      * Perform process-scoped translation if relocation enabled.
 711      *
 712      * - Translates an effective address to a host real address in
 713      *   quadrants 0 and 3 when HV=1.
 714      *
 715      * - Translates an effective address to a guest real address.
 716      */
 717     if (relocation) {
 718         int ret = ppc_radix64_process_scoped_xlate(cpu, access_type, eaddr, pid,
 719                                                    pate, &g_raddr, &prot,
 720                                                    &psize, mmu_idx, lpid,
 721                                                    guest_visible);
 722         if (ret) {
 723             return false;
 724         }
 725         *psizep = MIN(*psizep, psize);
 726         *protp &= prot;
 727     } else {
 728         g_raddr = eaddr & R_EADDR_MASK;
 729     }
 730
 731     if (vhyp_flat_addressing(cpu)) {
 732         *raddr = g_raddr;
 733     } else {
 734         /*
 735          * Perform partition-scoped translation if !HV or HV access to
 736          * quadrants 1 or 2. Translates a guest real address to a host
 737          * real address.
 738          */
 739         if (lpid || !mmuidx_hv(mmu_idx)) {
 740             int ret;
 741
 742             ret = ppc_radix64_partition_scoped_xlate(cpu, access_type, eaddr,
 743                                                      g_raddr, pate, raddr,
 744                                                      &prot, &psize, false,
 745                                                      mmu_idx, lpid,
 746                                                      guest_visible);
 747             if (ret) {
 748                 return false;
 749             }
 750             *psizep = MIN(*psizep, psize);
 751             *protp &= prot;
 752         } else {
 753             *raddr = g_raddr;
 754         }
 755     }
 756
 757     return true;
 758 }
 759
 760 bool ppc_radix64_xlate(PowerPCCPU *cpu, vaddr eaddr, MMUAccessType access_type,
 761                        hwaddr *raddrp, int *psizep, int *protp, int mmu_idx,
 762                        bool guest_visible)
 763 {
 764     bool ret = ppc_radix64_xlate_impl(cpu, eaddr, access_type, raddrp,
 765                                       psizep, protp, mmu_idx, guest_visible);
 766
 767     qemu_log_mask(CPU_LOG_MMU, "%s for %s @0x%"VADDR_PRIx
 768                   " mmu_idx %u (prot %c%c%c) -> 0x%"HWADDR_PRIx"\n",
 769                   __func__, access_str(access_type),
 770                   eaddr, mmu_idx,
 771                   *protp & PAGE_READ ? 'r' : '-',
 772                   *protp & PAGE_WRITE ? 'w' : '-',
 773                   *protp & PAGE_EXEC ? 'x' : '-',
 774                   *raddrp);
 775
 776     return ret;
 777 }