1 #include "qemu/osdep.h" 2 #include "qapi/error.h" 3 #include "sysemu/hw_accel.h" 4 #include "sysemu/sysemu.h" 5 #include "qemu/log.h" 6 #include "qemu/error-report.h" 7 #include "cpu.h" 8 #include "exec/exec-all.h" 9 #include "helper_regs.h" 10 #include "hw/ppc/spapr.h" 11 #include "mmu-hash64.h" 12 #include "cpu-models.h" 13 #include "trace.h" 14 #include "kvm_ppc.h" 15 #include "hw/ppc/spapr_ovec.h" 16 #include "mmu-book3s-v3.h" 17 18 struct SPRSyncState { 19 int spr; 20 target_ulong value; 21 target_ulong mask; 22 }; 23 24 static void do_spr_sync(CPUState *cs, run_on_cpu_data arg) 25 { 26 struct SPRSyncState *s = arg.host_ptr; 27 PowerPCCPU *cpu = POWERPC_CPU(cs); 28 CPUPPCState *env = &cpu->env; 29 30 cpu_synchronize_state(cs); 31 env->spr[s->spr] &= ~s->mask; 32 env->spr[s->spr] |= s->value; 33 } 34 35 static void set_spr(CPUState *cs, int spr, target_ulong value, 36 target_ulong mask) 37 { 38 struct SPRSyncState s = { 39 .spr = spr, 40 .value = value, 41 .mask = mask 42 }; 43 run_on_cpu(cs, do_spr_sync, RUN_ON_CPU_HOST_PTR(&s)); 44 } 45 46 static bool has_spr(PowerPCCPU *cpu, int spr) 47 { 48 /* We can test whether the SPR is defined by checking for a valid name */ 49 return cpu->env.spr_cb[spr].name != NULL; 50 } 51 52 static inline bool valid_ptex(PowerPCCPU *cpu, target_ulong ptex) 53 { 54 /* 55 * hash value/pteg group index is normalized by HPT mask 56 */ 57 if (((ptex & ~7ULL) / HPTES_PER_GROUP) & ~ppc_hash64_hpt_mask(cpu)) { 58 return false; 59 } 60 return true; 61 } 62 63 static bool is_ram_address(sPAPRMachineState *spapr, hwaddr addr) 64 { 65 MachineState *machine = MACHINE(spapr); 66 MemoryHotplugState *hpms = &spapr->hotplug_memory; 67 68 if (addr < machine->ram_size) { 69 return true; 70 } 71 if ((addr >= hpms->base) 72 && ((addr - hpms->base) < memory_region_size(&hpms->mr))) { 73 return true; 74 } 75 76 return false; 77 } 78 79 static target_ulong h_enter(PowerPCCPU *cpu, sPAPRMachineState *spapr, 80 target_ulong opcode, target_ulong *args) 81 { 82 target_ulong flags = args[0]; 83 target_ulong ptex = args[1]; 84 target_ulong pteh = args[2]; 85 target_ulong ptel = args[3]; 86 unsigned apshift; 87 target_ulong raddr; 88 target_ulong slot; 89 const ppc_hash_pte64_t *hptes; 90 91 apshift = ppc_hash64_hpte_page_shift_noslb(cpu, pteh, ptel); 92 if (!apshift) { 93 /* Bad page size encoding */ 94 return H_PARAMETER; 95 } 96 97 raddr = (ptel & HPTE64_R_RPN) & ~((1ULL << apshift) - 1); 98 99 if (is_ram_address(spapr, raddr)) { 100 /* Regular RAM - should have WIMG=0010 */ 101 if ((ptel & HPTE64_R_WIMG) != HPTE64_R_M) { 102 return H_PARAMETER; 103 } 104 } else { 105 target_ulong wimg_flags; 106 /* Looks like an IO address */ 107 /* FIXME: What WIMG combinations could be sensible for IO? 108 * For now we allow WIMG=010x, but are there others? */ 109 /* FIXME: Should we check against registered IO addresses? */ 110 wimg_flags = (ptel & (HPTE64_R_W | HPTE64_R_I | HPTE64_R_M)); 111 112 if (wimg_flags != HPTE64_R_I && 113 wimg_flags != (HPTE64_R_I | HPTE64_R_M)) { 114 return H_PARAMETER; 115 } 116 } 117 118 pteh &= ~0x60ULL; 119 120 if (!valid_ptex(cpu, ptex)) { 121 return H_PARAMETER; 122 } 123 124 slot = ptex & 7ULL; 125 ptex = ptex & ~7ULL; 126 127 if (likely((flags & H_EXACT) == 0)) { 128 hptes = ppc_hash64_map_hptes(cpu, ptex, HPTES_PER_GROUP); 129 for (slot = 0; slot < 8; slot++) { 130 if (!(ppc_hash64_hpte0(cpu, hptes, slot) & HPTE64_V_VALID)) { 131 break; 132 } 133 } 134 ppc_hash64_unmap_hptes(cpu, hptes, ptex, HPTES_PER_GROUP); 135 if (slot == 8) { 136 return H_PTEG_FULL; 137 } 138 } else { 139 hptes = ppc_hash64_map_hptes(cpu, ptex + slot, 1); 140 if (ppc_hash64_hpte0(cpu, hptes, 0) & HPTE64_V_VALID) { 141 ppc_hash64_unmap_hptes(cpu, hptes, ptex + slot, 1); 142 return H_PTEG_FULL; 143 } 144 ppc_hash64_unmap_hptes(cpu, hptes, ptex, 1); 145 } 146 147 ppc_hash64_store_hpte(cpu, ptex + slot, pteh | HPTE64_V_HPTE_DIRTY, ptel); 148 149 args[0] = ptex + slot; 150 return H_SUCCESS; 151 } 152 153 typedef enum { 154 REMOVE_SUCCESS = 0, 155 REMOVE_NOT_FOUND = 1, 156 REMOVE_PARM = 2, 157 REMOVE_HW = 3, 158 } RemoveResult; 159 160 static RemoveResult remove_hpte(PowerPCCPU *cpu, target_ulong ptex, 161 target_ulong avpn, 162 target_ulong flags, 163 target_ulong *vp, target_ulong *rp) 164 { 165 const ppc_hash_pte64_t *hptes; 166 target_ulong v, r; 167 168 if (!valid_ptex(cpu, ptex)) { 169 return REMOVE_PARM; 170 } 171 172 hptes = ppc_hash64_map_hptes(cpu, ptex, 1); 173 v = ppc_hash64_hpte0(cpu, hptes, 0); 174 r = ppc_hash64_hpte1(cpu, hptes, 0); 175 ppc_hash64_unmap_hptes(cpu, hptes, ptex, 1); 176 177 if ((v & HPTE64_V_VALID) == 0 || 178 ((flags & H_AVPN) && (v & ~0x7fULL) != avpn) || 179 ((flags & H_ANDCOND) && (v & avpn) != 0)) { 180 return REMOVE_NOT_FOUND; 181 } 182 *vp = v; 183 *rp = r; 184 ppc_hash64_store_hpte(cpu, ptex, HPTE64_V_HPTE_DIRTY, 0); 185 ppc_hash64_tlb_flush_hpte(cpu, ptex, v, r); 186 return REMOVE_SUCCESS; 187 } 188 189 static target_ulong h_remove(PowerPCCPU *cpu, sPAPRMachineState *spapr, 190 target_ulong opcode, target_ulong *args) 191 { 192 CPUPPCState *env = &cpu->env; 193 target_ulong flags = args[0]; 194 target_ulong ptex = args[1]; 195 target_ulong avpn = args[2]; 196 RemoveResult ret; 197 198 ret = remove_hpte(cpu, ptex, avpn, flags, 199 &args[0], &args[1]); 200 201 switch (ret) { 202 case REMOVE_SUCCESS: 203 check_tlb_flush(env, true); 204 return H_SUCCESS; 205 206 case REMOVE_NOT_FOUND: 207 return H_NOT_FOUND; 208 209 case REMOVE_PARM: 210 return H_PARAMETER; 211 212 case REMOVE_HW: 213 return H_HARDWARE; 214 } 215 216 g_assert_not_reached(); 217 } 218 219 #define H_BULK_REMOVE_TYPE 0xc000000000000000ULL 220 #define H_BULK_REMOVE_REQUEST 0x4000000000000000ULL 221 #define H_BULK_REMOVE_RESPONSE 0x8000000000000000ULL 222 #define H_BULK_REMOVE_END 0xc000000000000000ULL 223 #define H_BULK_REMOVE_CODE 0x3000000000000000ULL 224 #define H_BULK_REMOVE_SUCCESS 0x0000000000000000ULL 225 #define H_BULK_REMOVE_NOT_FOUND 0x1000000000000000ULL 226 #define H_BULK_REMOVE_PARM 0x2000000000000000ULL 227 #define H_BULK_REMOVE_HW 0x3000000000000000ULL 228 #define H_BULK_REMOVE_RC 0x0c00000000000000ULL 229 #define H_BULK_REMOVE_FLAGS 0x0300000000000000ULL 230 #define H_BULK_REMOVE_ABSOLUTE 0x0000000000000000ULL 231 #define H_BULK_REMOVE_ANDCOND 0x0100000000000000ULL 232 #define H_BULK_REMOVE_AVPN 0x0200000000000000ULL 233 #define H_BULK_REMOVE_PTEX 0x00ffffffffffffffULL 234 235 #define H_BULK_REMOVE_MAX_BATCH 4 236 237 static target_ulong h_bulk_remove(PowerPCCPU *cpu, sPAPRMachineState *spapr, 238 target_ulong opcode, target_ulong *args) 239 { 240 CPUPPCState *env = &cpu->env; 241 int i; 242 target_ulong rc = H_SUCCESS; 243 244 for (i = 0; i < H_BULK_REMOVE_MAX_BATCH; i++) { 245 target_ulong *tsh = &args[i*2]; 246 target_ulong tsl = args[i*2 + 1]; 247 target_ulong v, r, ret; 248 249 if ((*tsh & H_BULK_REMOVE_TYPE) == H_BULK_REMOVE_END) { 250 break; 251 } else if ((*tsh & H_BULK_REMOVE_TYPE) != H_BULK_REMOVE_REQUEST) { 252 return H_PARAMETER; 253 } 254 255 *tsh &= H_BULK_REMOVE_PTEX | H_BULK_REMOVE_FLAGS; 256 *tsh |= H_BULK_REMOVE_RESPONSE; 257 258 if ((*tsh & H_BULK_REMOVE_ANDCOND) && (*tsh & H_BULK_REMOVE_AVPN)) { 259 *tsh |= H_BULK_REMOVE_PARM; 260 return H_PARAMETER; 261 } 262 263 ret = remove_hpte(cpu, *tsh & H_BULK_REMOVE_PTEX, tsl, 264 (*tsh & H_BULK_REMOVE_FLAGS) >> 26, 265 &v, &r); 266 267 *tsh |= ret << 60; 268 269 switch (ret) { 270 case REMOVE_SUCCESS: 271 *tsh |= (r & (HPTE64_R_C | HPTE64_R_R)) << 43; 272 break; 273 274 case REMOVE_PARM: 275 rc = H_PARAMETER; 276 goto exit; 277 278 case REMOVE_HW: 279 rc = H_HARDWARE; 280 goto exit; 281 } 282 } 283 exit: 284 check_tlb_flush(env, true); 285 286 return rc; 287 } 288 289 static target_ulong h_protect(PowerPCCPU *cpu, sPAPRMachineState *spapr, 290 target_ulong opcode, target_ulong *args) 291 { 292 CPUPPCState *env = &cpu->env; 293 target_ulong flags = args[0]; 294 target_ulong ptex = args[1]; 295 target_ulong avpn = args[2]; 296 const ppc_hash_pte64_t *hptes; 297 target_ulong v, r; 298 299 if (!valid_ptex(cpu, ptex)) { 300 return H_PARAMETER; 301 } 302 303 hptes = ppc_hash64_map_hptes(cpu, ptex, 1); 304 v = ppc_hash64_hpte0(cpu, hptes, 0); 305 r = ppc_hash64_hpte1(cpu, hptes, 0); 306 ppc_hash64_unmap_hptes(cpu, hptes, ptex, 1); 307 308 if ((v & HPTE64_V_VALID) == 0 || 309 ((flags & H_AVPN) && (v & ~0x7fULL) != avpn)) { 310 return H_NOT_FOUND; 311 } 312 313 r &= ~(HPTE64_R_PP0 | HPTE64_R_PP | HPTE64_R_N | 314 HPTE64_R_KEY_HI | HPTE64_R_KEY_LO); 315 r |= (flags << 55) & HPTE64_R_PP0; 316 r |= (flags << 48) & HPTE64_R_KEY_HI; 317 r |= flags & (HPTE64_R_PP | HPTE64_R_N | HPTE64_R_KEY_LO); 318 ppc_hash64_store_hpte(cpu, ptex, 319 (v & ~HPTE64_V_VALID) | HPTE64_V_HPTE_DIRTY, 0); 320 ppc_hash64_tlb_flush_hpte(cpu, ptex, v, r); 321 /* Flush the tlb */ 322 check_tlb_flush(env, true); 323 /* Don't need a memory barrier, due to qemu's global lock */ 324 ppc_hash64_store_hpte(cpu, ptex, v | HPTE64_V_HPTE_DIRTY, r); 325 return H_SUCCESS; 326 } 327 328 static target_ulong h_read(PowerPCCPU *cpu, sPAPRMachineState *spapr, 329 target_ulong opcode, target_ulong *args) 330 { 331 target_ulong flags = args[0]; 332 target_ulong ptex = args[1]; 333 uint8_t *hpte; 334 int i, ridx, n_entries = 1; 335 336 if (!valid_ptex(cpu, ptex)) { 337 return H_PARAMETER; 338 } 339 340 if (flags & H_READ_4) { 341 /* Clear the two low order bits */ 342 ptex &= ~(3ULL); 343 n_entries = 4; 344 } 345 346 hpte = spapr->htab + (ptex * HASH_PTE_SIZE_64); 347 348 for (i = 0, ridx = 0; i < n_entries; i++) { 349 args[ridx++] = ldq_p(hpte); 350 args[ridx++] = ldq_p(hpte + (HASH_PTE_SIZE_64/2)); 351 hpte += HASH_PTE_SIZE_64; 352 } 353 354 return H_SUCCESS; 355 } 356 357 struct sPAPRPendingHPT { 358 /* These fields are read-only after initialization */ 359 int shift; 360 QemuThread thread; 361 362 /* These fields are protected by the BQL */ 363 bool complete; 364 365 /* These fields are private to the preparation thread if 366 * !complete, otherwise protected by the BQL */ 367 int ret; 368 void *hpt; 369 }; 370 371 static void free_pending_hpt(sPAPRPendingHPT *pending) 372 { 373 if (pending->hpt) { 374 qemu_vfree(pending->hpt); 375 } 376 377 g_free(pending); 378 } 379 380 static void *hpt_prepare_thread(void *opaque) 381 { 382 sPAPRPendingHPT *pending = opaque; 383 size_t size = 1ULL << pending->shift; 384 385 pending->hpt = qemu_memalign(size, size); 386 if (pending->hpt) { 387 memset(pending->hpt, 0, size); 388 pending->ret = H_SUCCESS; 389 } else { 390 pending->ret = H_NO_MEM; 391 } 392 393 qemu_mutex_lock_iothread(); 394 395 if (SPAPR_MACHINE(qdev_get_machine())->pending_hpt == pending) { 396 /* Ready to go */ 397 pending->complete = true; 398 } else { 399 /* We've been cancelled, clean ourselves up */ 400 free_pending_hpt(pending); 401 } 402 403 qemu_mutex_unlock_iothread(); 404 return NULL; 405 } 406 407 /* Must be called with BQL held */ 408 static void cancel_hpt_prepare(sPAPRMachineState *spapr) 409 { 410 sPAPRPendingHPT *pending = spapr->pending_hpt; 411 412 /* Let the thread know it's cancelled */ 413 spapr->pending_hpt = NULL; 414 415 if (!pending) { 416 /* Nothing to do */ 417 return; 418 } 419 420 if (!pending->complete) { 421 /* thread will clean itself up */ 422 return; 423 } 424 425 free_pending_hpt(pending); 426 } 427 428 /* Convert a return code from the KVM ioctl()s implementing resize HPT 429 * into a PAPR hypercall return code */ 430 static target_ulong resize_hpt_convert_rc(int ret) 431 { 432 if (ret >= 100000) { 433 return H_LONG_BUSY_ORDER_100_SEC; 434 } else if (ret >= 10000) { 435 return H_LONG_BUSY_ORDER_10_SEC; 436 } else if (ret >= 1000) { 437 return H_LONG_BUSY_ORDER_1_SEC; 438 } else if (ret >= 100) { 439 return H_LONG_BUSY_ORDER_100_MSEC; 440 } else if (ret >= 10) { 441 return H_LONG_BUSY_ORDER_10_MSEC; 442 } else if (ret > 0) { 443 return H_LONG_BUSY_ORDER_1_MSEC; 444 } 445 446 switch (ret) { 447 case 0: 448 return H_SUCCESS; 449 case -EPERM: 450 return H_AUTHORITY; 451 case -EINVAL: 452 return H_PARAMETER; 453 case -ENXIO: 454 return H_CLOSED; 455 case -ENOSPC: 456 return H_PTEG_FULL; 457 case -EBUSY: 458 return H_BUSY; 459 case -ENOMEM: 460 return H_NO_MEM; 461 default: 462 return H_HARDWARE; 463 } 464 } 465 466 static target_ulong h_resize_hpt_prepare(PowerPCCPU *cpu, 467 sPAPRMachineState *spapr, 468 target_ulong opcode, 469 target_ulong *args) 470 { 471 target_ulong flags = args[0]; 472 int shift = args[1]; 473 sPAPRPendingHPT *pending = spapr->pending_hpt; 474 uint64_t current_ram_size; 475 int rc; 476 477 if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) { 478 return H_AUTHORITY; 479 } 480 481 if (!spapr->htab_shift) { 482 /* Radix guest, no HPT */ 483 return H_NOT_AVAILABLE; 484 } 485 486 trace_spapr_h_resize_hpt_prepare(flags, shift); 487 488 if (flags != 0) { 489 return H_PARAMETER; 490 } 491 492 if (shift && ((shift < 18) || (shift > 46))) { 493 return H_PARAMETER; 494 } 495 496 current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size(); 497 498 /* We only allow the guest to allocate an HPT one order above what 499 * we'd normally give them (to stop a small guest claiming a huge 500 * chunk of resources in the HPT */ 501 if (shift > (spapr_hpt_shift_for_ramsize(current_ram_size) + 1)) { 502 return H_RESOURCE; 503 } 504 505 rc = kvmppc_resize_hpt_prepare(cpu, flags, shift); 506 if (rc != -ENOSYS) { 507 return resize_hpt_convert_rc(rc); 508 } 509 510 if (pending) { 511 /* something already in progress */ 512 if (pending->shift == shift) { 513 /* and it's suitable */ 514 if (pending->complete) { 515 return pending->ret; 516 } else { 517 return H_LONG_BUSY_ORDER_100_MSEC; 518 } 519 } 520 521 /* not suitable, cancel and replace */ 522 cancel_hpt_prepare(spapr); 523 } 524 525 if (!shift) { 526 /* nothing to do */ 527 return H_SUCCESS; 528 } 529 530 /* start new prepare */ 531 532 pending = g_new0(sPAPRPendingHPT, 1); 533 pending->shift = shift; 534 pending->ret = H_HARDWARE; 535 536 qemu_thread_create(&pending->thread, "sPAPR HPT prepare", 537 hpt_prepare_thread, pending, QEMU_THREAD_DETACHED); 538 539 spapr->pending_hpt = pending; 540 541 /* In theory we could estimate the time more accurately based on 542 * the new size, but there's not much point */ 543 return H_LONG_BUSY_ORDER_100_MSEC; 544 } 545 546 static uint64_t new_hpte_load0(void *htab, uint64_t pteg, int slot) 547 { 548 uint8_t *addr = htab; 549 550 addr += pteg * HASH_PTEG_SIZE_64; 551 addr += slot * HASH_PTE_SIZE_64; 552 return ldq_p(addr); 553 } 554 555 static void new_hpte_store(void *htab, uint64_t pteg, int slot, 556 uint64_t pte0, uint64_t pte1) 557 { 558 uint8_t *addr = htab; 559 560 addr += pteg * HASH_PTEG_SIZE_64; 561 addr += slot * HASH_PTE_SIZE_64; 562 563 stq_p(addr, pte0); 564 stq_p(addr + HASH_PTE_SIZE_64 / 2, pte1); 565 } 566 567 static int rehash_hpte(PowerPCCPU *cpu, 568 const ppc_hash_pte64_t *hptes, 569 void *old_hpt, uint64_t oldsize, 570 void *new_hpt, uint64_t newsize, 571 uint64_t pteg, int slot) 572 { 573 uint64_t old_hash_mask = (oldsize >> 7) - 1; 574 uint64_t new_hash_mask = (newsize >> 7) - 1; 575 target_ulong pte0 = ppc_hash64_hpte0(cpu, hptes, slot); 576 target_ulong pte1; 577 uint64_t avpn; 578 unsigned base_pg_shift; 579 uint64_t hash, new_pteg, replace_pte0; 580 581 if (!(pte0 & HPTE64_V_VALID) || !(pte0 & HPTE64_V_BOLTED)) { 582 return H_SUCCESS; 583 } 584 585 pte1 = ppc_hash64_hpte1(cpu, hptes, slot); 586 587 base_pg_shift = ppc_hash64_hpte_page_shift_noslb(cpu, pte0, pte1); 588 assert(base_pg_shift); /* H_ENTER shouldn't allow a bad encoding */ 589 avpn = HPTE64_V_AVPN_VAL(pte0) & ~(((1ULL << base_pg_shift) - 1) >> 23); 590 591 if (pte0 & HPTE64_V_SECONDARY) { 592 pteg = ~pteg; 593 } 594 595 if ((pte0 & HPTE64_V_SSIZE) == HPTE64_V_SSIZE_256M) { 596 uint64_t offset, vsid; 597 598 /* We only have 28 - 23 bits of offset in avpn */ 599 offset = (avpn & 0x1f) << 23; 600 vsid = avpn >> 5; 601 /* We can find more bits from the pteg value */ 602 if (base_pg_shift < 23) { 603 offset |= ((vsid ^ pteg) & old_hash_mask) << base_pg_shift; 604 } 605 606 hash = vsid ^ (offset >> base_pg_shift); 607 } else if ((pte0 & HPTE64_V_SSIZE) == HPTE64_V_SSIZE_1T) { 608 uint64_t offset, vsid; 609 610 /* We only have 40 - 23 bits of seg_off in avpn */ 611 offset = (avpn & 0x1ffff) << 23; 612 vsid = avpn >> 17; 613 if (base_pg_shift < 23) { 614 offset |= ((vsid ^ (vsid << 25) ^ pteg) & old_hash_mask) 615 << base_pg_shift; 616 } 617 618 hash = vsid ^ (vsid << 25) ^ (offset >> base_pg_shift); 619 } else { 620 error_report("rehash_pte: Bad segment size in HPTE"); 621 return H_HARDWARE; 622 } 623 624 new_pteg = hash & new_hash_mask; 625 if (pte0 & HPTE64_V_SECONDARY) { 626 assert(~pteg == (hash & old_hash_mask)); 627 new_pteg = ~new_pteg; 628 } else { 629 assert(pteg == (hash & old_hash_mask)); 630 } 631 assert((oldsize != newsize) || (pteg == new_pteg)); 632 replace_pte0 = new_hpte_load0(new_hpt, new_pteg, slot); 633 /* 634 * Strictly speaking, we don't need all these tests, since we only 635 * ever rehash bolted HPTEs. We might in future handle non-bolted 636 * HPTEs, though so make the logic correct for those cases as 637 * well. 638 */ 639 if (replace_pte0 & HPTE64_V_VALID) { 640 assert(newsize < oldsize); 641 if (replace_pte0 & HPTE64_V_BOLTED) { 642 if (pte0 & HPTE64_V_BOLTED) { 643 /* Bolted collision, nothing we can do */ 644 return H_PTEG_FULL; 645 } else { 646 /* Discard this hpte */ 647 return H_SUCCESS; 648 } 649 } 650 } 651 652 new_hpte_store(new_hpt, new_pteg, slot, pte0, pte1); 653 return H_SUCCESS; 654 } 655 656 static int rehash_hpt(PowerPCCPU *cpu, 657 void *old_hpt, uint64_t oldsize, 658 void *new_hpt, uint64_t newsize) 659 { 660 uint64_t n_ptegs = oldsize >> 7; 661 uint64_t pteg; 662 int slot; 663 int rc; 664 665 for (pteg = 0; pteg < n_ptegs; pteg++) { 666 hwaddr ptex = pteg * HPTES_PER_GROUP; 667 const ppc_hash_pte64_t *hptes 668 = ppc_hash64_map_hptes(cpu, ptex, HPTES_PER_GROUP); 669 670 if (!hptes) { 671 return H_HARDWARE; 672 } 673 674 for (slot = 0; slot < HPTES_PER_GROUP; slot++) { 675 rc = rehash_hpte(cpu, hptes, old_hpt, oldsize, new_hpt, newsize, 676 pteg, slot); 677 if (rc != H_SUCCESS) { 678 ppc_hash64_unmap_hptes(cpu, hptes, ptex, HPTES_PER_GROUP); 679 return rc; 680 } 681 } 682 ppc_hash64_unmap_hptes(cpu, hptes, ptex, HPTES_PER_GROUP); 683 } 684 685 return H_SUCCESS; 686 } 687 688 static void do_push_sregs_to_kvm_pr(CPUState *cs, run_on_cpu_data data) 689 { 690 int ret; 691 692 cpu_synchronize_state(cs); 693 694 ret = kvmppc_put_books_sregs(POWERPC_CPU(cs)); 695 if (ret < 0) { 696 error_report("failed to push sregs to KVM: %s", strerror(-ret)); 697 exit(1); 698 } 699 } 700 701 static void push_sregs_to_kvm_pr(sPAPRMachineState *spapr) 702 { 703 CPUState *cs; 704 705 /* 706 * This is a hack for the benefit of KVM PR - it abuses the SDR1 707 * slot in kvm_sregs to communicate the userspace address of the 708 * HPT 709 */ 710 if (!kvm_enabled() || !spapr->htab) { 711 return; 712 } 713 714 CPU_FOREACH(cs) { 715 run_on_cpu(cs, do_push_sregs_to_kvm_pr, RUN_ON_CPU_NULL); 716 } 717 } 718 719 static target_ulong h_resize_hpt_commit(PowerPCCPU *cpu, 720 sPAPRMachineState *spapr, 721 target_ulong opcode, 722 target_ulong *args) 723 { 724 target_ulong flags = args[0]; 725 target_ulong shift = args[1]; 726 sPAPRPendingHPT *pending = spapr->pending_hpt; 727 int rc; 728 size_t newsize; 729 730 if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) { 731 return H_AUTHORITY; 732 } 733 734 if (!spapr->htab_shift) { 735 /* Radix guest, no HPT */ 736 return H_NOT_AVAILABLE; 737 } 738 739 trace_spapr_h_resize_hpt_commit(flags, shift); 740 741 rc = kvmppc_resize_hpt_commit(cpu, flags, shift); 742 if (rc != -ENOSYS) { 743 rc = resize_hpt_convert_rc(rc); 744 if (rc == H_SUCCESS) { 745 /* Need to set the new htab_shift in the machine state */ 746 spapr->htab_shift = shift; 747 } 748 return rc; 749 } 750 751 if (flags != 0) { 752 return H_PARAMETER; 753 } 754 755 if (!pending || (pending->shift != shift)) { 756 /* no matching prepare */ 757 return H_CLOSED; 758 } 759 760 if (!pending->complete) { 761 /* prepare has not completed */ 762 return H_BUSY; 763 } 764 765 /* Shouldn't have got past PREPARE without an HPT */ 766 g_assert(spapr->htab_shift); 767 768 newsize = 1ULL << pending->shift; 769 rc = rehash_hpt(cpu, spapr->htab, HTAB_SIZE(spapr), 770 pending->hpt, newsize); 771 if (rc == H_SUCCESS) { 772 qemu_vfree(spapr->htab); 773 spapr->htab = pending->hpt; 774 spapr->htab_shift = pending->shift; 775 776 push_sregs_to_kvm_pr(spapr); 777 778 pending->hpt = NULL; /* so it's not free()d */ 779 } 780 781 /* Clean up */ 782 spapr->pending_hpt = NULL; 783 free_pending_hpt(pending); 784 785 return rc; 786 } 787 788 static target_ulong h_set_sprg0(PowerPCCPU *cpu, sPAPRMachineState *spapr, 789 target_ulong opcode, target_ulong *args) 790 { 791 cpu_synchronize_state(CPU(cpu)); 792 cpu->env.spr[SPR_SPRG0] = args[0]; 793 794 return H_SUCCESS; 795 } 796 797 static target_ulong h_set_dabr(PowerPCCPU *cpu, sPAPRMachineState *spapr, 798 target_ulong opcode, target_ulong *args) 799 { 800 if (!has_spr(cpu, SPR_DABR)) { 801 return H_HARDWARE; /* DABR register not available */ 802 } 803 cpu_synchronize_state(CPU(cpu)); 804 805 if (has_spr(cpu, SPR_DABRX)) { 806 cpu->env.spr[SPR_DABRX] = 0x3; /* Use Problem and Privileged state */ 807 } else if (!(args[0] & 0x4)) { /* Breakpoint Translation set? */ 808 return H_RESERVED_DABR; 809 } 810 811 cpu->env.spr[SPR_DABR] = args[0]; 812 return H_SUCCESS; 813 } 814 815 static target_ulong h_set_xdabr(PowerPCCPU *cpu, sPAPRMachineState *spapr, 816 target_ulong opcode, target_ulong *args) 817 { 818 target_ulong dabrx = args[1]; 819 820 if (!has_spr(cpu, SPR_DABR) || !has_spr(cpu, SPR_DABRX)) { 821 return H_HARDWARE; 822 } 823 824 if ((dabrx & ~0xfULL) != 0 || (dabrx & H_DABRX_HYPERVISOR) != 0 825 || (dabrx & (H_DABRX_KERNEL | H_DABRX_USER)) == 0) { 826 return H_PARAMETER; 827 } 828 829 cpu_synchronize_state(CPU(cpu)); 830 cpu->env.spr[SPR_DABRX] = dabrx; 831 cpu->env.spr[SPR_DABR] = args[0]; 832 833 return H_SUCCESS; 834 } 835 836 static target_ulong h_page_init(PowerPCCPU *cpu, sPAPRMachineState *spapr, 837 target_ulong opcode, target_ulong *args) 838 { 839 target_ulong flags = args[0]; 840 hwaddr dst = args[1]; 841 hwaddr src = args[2]; 842 hwaddr len = TARGET_PAGE_SIZE; 843 uint8_t *pdst, *psrc; 844 target_long ret = H_SUCCESS; 845 846 if (flags & ~(H_ICACHE_SYNCHRONIZE | H_ICACHE_INVALIDATE 847 | H_COPY_PAGE | H_ZERO_PAGE)) { 848 qemu_log_mask(LOG_UNIMP, "h_page_init: Bad flags (" TARGET_FMT_lx "\n", 849 flags); 850 return H_PARAMETER; 851 } 852 853 /* Map-in destination */ 854 if (!is_ram_address(spapr, dst) || (dst & ~TARGET_PAGE_MASK) != 0) { 855 return H_PARAMETER; 856 } 857 pdst = cpu_physical_memory_map(dst, &len, 1); 858 if (!pdst || len != TARGET_PAGE_SIZE) { 859 return H_PARAMETER; 860 } 861 862 if (flags & H_COPY_PAGE) { 863 /* Map-in source, copy to destination, and unmap source again */ 864 if (!is_ram_address(spapr, src) || (src & ~TARGET_PAGE_MASK) != 0) { 865 ret = H_PARAMETER; 866 goto unmap_out; 867 } 868 psrc = cpu_physical_memory_map(src, &len, 0); 869 if (!psrc || len != TARGET_PAGE_SIZE) { 870 ret = H_PARAMETER; 871 goto unmap_out; 872 } 873 memcpy(pdst, psrc, len); 874 cpu_physical_memory_unmap(psrc, len, 0, len); 875 } else if (flags & H_ZERO_PAGE) { 876 memset(pdst, 0, len); /* Just clear the destination page */ 877 } 878 879 if (kvm_enabled() && (flags & H_ICACHE_SYNCHRONIZE) != 0) { 880 kvmppc_dcbst_range(cpu, pdst, len); 881 } 882 if (flags & (H_ICACHE_SYNCHRONIZE | H_ICACHE_INVALIDATE)) { 883 if (kvm_enabled()) { 884 kvmppc_icbi_range(cpu, pdst, len); 885 } else { 886 tb_flush(CPU(cpu)); 887 } 888 } 889 890 unmap_out: 891 cpu_physical_memory_unmap(pdst, TARGET_PAGE_SIZE, 1, len); 892 return ret; 893 } 894 895 #define FLAGS_REGISTER_VPA 0x0000200000000000ULL 896 #define FLAGS_REGISTER_DTL 0x0000400000000000ULL 897 #define FLAGS_REGISTER_SLBSHADOW 0x0000600000000000ULL 898 #define FLAGS_DEREGISTER_VPA 0x0000a00000000000ULL 899 #define FLAGS_DEREGISTER_DTL 0x0000c00000000000ULL 900 #define FLAGS_DEREGISTER_SLBSHADOW 0x0000e00000000000ULL 901 902 #define VPA_MIN_SIZE 640 903 #define VPA_SIZE_OFFSET 0x4 904 #define VPA_SHARED_PROC_OFFSET 0x9 905 #define VPA_SHARED_PROC_VAL 0x2 906 907 static target_ulong register_vpa(CPUPPCState *env, target_ulong vpa) 908 { 909 CPUState *cs = CPU(ppc_env_get_cpu(env)); 910 uint16_t size; 911 uint8_t tmp; 912 913 if (vpa == 0) { 914 hcall_dprintf("Can't cope with registering a VPA at logical 0\n"); 915 return H_HARDWARE; 916 } 917 918 if (vpa % env->dcache_line_size) { 919 return H_PARAMETER; 920 } 921 /* FIXME: bounds check the address */ 922 923 size = lduw_be_phys(cs->as, vpa + 0x4); 924 925 if (size < VPA_MIN_SIZE) { 926 return H_PARAMETER; 927 } 928 929 /* VPA is not allowed to cross a page boundary */ 930 if ((vpa / 4096) != ((vpa + size - 1) / 4096)) { 931 return H_PARAMETER; 932 } 933 934 env->vpa_addr = vpa; 935 936 tmp = ldub_phys(cs->as, env->vpa_addr + VPA_SHARED_PROC_OFFSET); 937 tmp |= VPA_SHARED_PROC_VAL; 938 stb_phys(cs->as, env->vpa_addr + VPA_SHARED_PROC_OFFSET, tmp); 939 940 return H_SUCCESS; 941 } 942 943 static target_ulong deregister_vpa(CPUPPCState *env, target_ulong vpa) 944 { 945 if (env->slb_shadow_addr) { 946 return H_RESOURCE; 947 } 948 949 if (env->dtl_addr) { 950 return H_RESOURCE; 951 } 952 953 env->vpa_addr = 0; 954 return H_SUCCESS; 955 } 956 957 static target_ulong register_slb_shadow(CPUPPCState *env, target_ulong addr) 958 { 959 CPUState *cs = CPU(ppc_env_get_cpu(env)); 960 uint32_t size; 961 962 if (addr == 0) { 963 hcall_dprintf("Can't cope with SLB shadow at logical 0\n"); 964 return H_HARDWARE; 965 } 966 967 size = ldl_be_phys(cs->as, addr + 0x4); 968 if (size < 0x8) { 969 return H_PARAMETER; 970 } 971 972 if ((addr / 4096) != ((addr + size - 1) / 4096)) { 973 return H_PARAMETER; 974 } 975 976 if (!env->vpa_addr) { 977 return H_RESOURCE; 978 } 979 980 env->slb_shadow_addr = addr; 981 env->slb_shadow_size = size; 982 983 return H_SUCCESS; 984 } 985 986 static target_ulong deregister_slb_shadow(CPUPPCState *env, target_ulong addr) 987 { 988 env->slb_shadow_addr = 0; 989 env->slb_shadow_size = 0; 990 return H_SUCCESS; 991 } 992 993 static target_ulong register_dtl(CPUPPCState *env, target_ulong addr) 994 { 995 CPUState *cs = CPU(ppc_env_get_cpu(env)); 996 uint32_t size; 997 998 if (addr == 0) { 999 hcall_dprintf("Can't cope with DTL at logical 0\n"); 1000 return H_HARDWARE; 1001 } 1002 1003 size = ldl_be_phys(cs->as, addr + 0x4); 1004 1005 if (size < 48) { 1006 return H_PARAMETER; 1007 } 1008 1009 if (!env->vpa_addr) { 1010 return H_RESOURCE; 1011 } 1012 1013 env->dtl_addr = addr; 1014 env->dtl_size = size; 1015 1016 return H_SUCCESS; 1017 } 1018 1019 static target_ulong deregister_dtl(CPUPPCState *env, target_ulong addr) 1020 { 1021 env->dtl_addr = 0; 1022 env->dtl_size = 0; 1023 1024 return H_SUCCESS; 1025 } 1026 1027 static target_ulong h_register_vpa(PowerPCCPU *cpu, sPAPRMachineState *spapr, 1028 target_ulong opcode, target_ulong *args) 1029 { 1030 target_ulong flags = args[0]; 1031 target_ulong procno = args[1]; 1032 target_ulong vpa = args[2]; 1033 target_ulong ret = H_PARAMETER; 1034 CPUPPCState *tenv; 1035 PowerPCCPU *tcpu; 1036 1037 tcpu = spapr_find_cpu(procno); 1038 if (!tcpu) { 1039 return H_PARAMETER; 1040 } 1041 tenv = &tcpu->env; 1042 1043 switch (flags) { 1044 case FLAGS_REGISTER_VPA: 1045 ret = register_vpa(tenv, vpa); 1046 break; 1047 1048 case FLAGS_DEREGISTER_VPA: 1049 ret = deregister_vpa(tenv, vpa); 1050 break; 1051 1052 case FLAGS_REGISTER_SLBSHADOW: 1053 ret = register_slb_shadow(tenv, vpa); 1054 break; 1055 1056 case FLAGS_DEREGISTER_SLBSHADOW: 1057 ret = deregister_slb_shadow(tenv, vpa); 1058 break; 1059 1060 case FLAGS_REGISTER_DTL: 1061 ret = register_dtl(tenv, vpa); 1062 break; 1063 1064 case FLAGS_DEREGISTER_DTL: 1065 ret = deregister_dtl(tenv, vpa); 1066 break; 1067 } 1068 1069 return ret; 1070 } 1071 1072 static target_ulong h_cede(PowerPCCPU *cpu, sPAPRMachineState *spapr, 1073 target_ulong opcode, target_ulong *args) 1074 { 1075 CPUPPCState *env = &cpu->env; 1076 CPUState *cs = CPU(cpu); 1077 1078 env->msr |= (1ULL << MSR_EE); 1079 hreg_compute_hflags(env); 1080 if (!cpu_has_work(cs)) { 1081 cs->halted = 1; 1082 cs->exception_index = EXCP_HLT; 1083 cs->exit_request = 1; 1084 } 1085 return H_SUCCESS; 1086 } 1087 1088 static target_ulong h_rtas(PowerPCCPU *cpu, sPAPRMachineState *spapr, 1089 target_ulong opcode, target_ulong *args) 1090 { 1091 target_ulong rtas_r3 = args[0]; 1092 uint32_t token = rtas_ld(rtas_r3, 0); 1093 uint32_t nargs = rtas_ld(rtas_r3, 1); 1094 uint32_t nret = rtas_ld(rtas_r3, 2); 1095 1096 return spapr_rtas_call(cpu, spapr, token, nargs, rtas_r3 + 12, 1097 nret, rtas_r3 + 12 + 4*nargs); 1098 } 1099 1100 static target_ulong h_logical_load(PowerPCCPU *cpu, sPAPRMachineState *spapr, 1101 target_ulong opcode, target_ulong *args) 1102 { 1103 CPUState *cs = CPU(cpu); 1104 target_ulong size = args[0]; 1105 target_ulong addr = args[1]; 1106 1107 switch (size) { 1108 case 1: 1109 args[0] = ldub_phys(cs->as, addr); 1110 return H_SUCCESS; 1111 case 2: 1112 args[0] = lduw_phys(cs->as, addr); 1113 return H_SUCCESS; 1114 case 4: 1115 args[0] = ldl_phys(cs->as, addr); 1116 return H_SUCCESS; 1117 case 8: 1118 args[0] = ldq_phys(cs->as, addr); 1119 return H_SUCCESS; 1120 } 1121 return H_PARAMETER; 1122 } 1123 1124 static target_ulong h_logical_store(PowerPCCPU *cpu, sPAPRMachineState *spapr, 1125 target_ulong opcode, target_ulong *args) 1126 { 1127 CPUState *cs = CPU(cpu); 1128 1129 target_ulong size = args[0]; 1130 target_ulong addr = args[1]; 1131 target_ulong val = args[2]; 1132 1133 switch (size) { 1134 case 1: 1135 stb_phys(cs->as, addr, val); 1136 return H_SUCCESS; 1137 case 2: 1138 stw_phys(cs->as, addr, val); 1139 return H_SUCCESS; 1140 case 4: 1141 stl_phys(cs->as, addr, val); 1142 return H_SUCCESS; 1143 case 8: 1144 stq_phys(cs->as, addr, val); 1145 return H_SUCCESS; 1146 } 1147 return H_PARAMETER; 1148 } 1149 1150 static target_ulong h_logical_memop(PowerPCCPU *cpu, sPAPRMachineState *spapr, 1151 target_ulong opcode, target_ulong *args) 1152 { 1153 CPUState *cs = CPU(cpu); 1154 1155 target_ulong dst = args[0]; /* Destination address */ 1156 target_ulong src = args[1]; /* Source address */ 1157 target_ulong esize = args[2]; /* Element size (0=1,1=2,2=4,3=8) */ 1158 target_ulong count = args[3]; /* Element count */ 1159 target_ulong op = args[4]; /* 0 = copy, 1 = invert */ 1160 uint64_t tmp; 1161 unsigned int mask = (1 << esize) - 1; 1162 int step = 1 << esize; 1163 1164 if (count > 0x80000000) { 1165 return H_PARAMETER; 1166 } 1167 1168 if ((dst & mask) || (src & mask) || (op > 1)) { 1169 return H_PARAMETER; 1170 } 1171 1172 if (dst >= src && dst < (src + (count << esize))) { 1173 dst = dst + ((count - 1) << esize); 1174 src = src + ((count - 1) << esize); 1175 step = -step; 1176 } 1177 1178 while (count--) { 1179 switch (esize) { 1180 case 0: 1181 tmp = ldub_phys(cs->as, src); 1182 break; 1183 case 1: 1184 tmp = lduw_phys(cs->as, src); 1185 break; 1186 case 2: 1187 tmp = ldl_phys(cs->as, src); 1188 break; 1189 case 3: 1190 tmp = ldq_phys(cs->as, src); 1191 break; 1192 default: 1193 return H_PARAMETER; 1194 } 1195 if (op == 1) { 1196 tmp = ~tmp; 1197 } 1198 switch (esize) { 1199 case 0: 1200 stb_phys(cs->as, dst, tmp); 1201 break; 1202 case 1: 1203 stw_phys(cs->as, dst, tmp); 1204 break; 1205 case 2: 1206 stl_phys(cs->as, dst, tmp); 1207 break; 1208 case 3: 1209 stq_phys(cs->as, dst, tmp); 1210 break; 1211 } 1212 dst = dst + step; 1213 src = src + step; 1214 } 1215 1216 return H_SUCCESS; 1217 } 1218 1219 static target_ulong h_logical_icbi(PowerPCCPU *cpu, sPAPRMachineState *spapr, 1220 target_ulong opcode, target_ulong *args) 1221 { 1222 /* Nothing to do on emulation, KVM will trap this in the kernel */ 1223 return H_SUCCESS; 1224 } 1225 1226 static target_ulong h_logical_dcbf(PowerPCCPU *cpu, sPAPRMachineState *spapr, 1227 target_ulong opcode, target_ulong *args) 1228 { 1229 /* Nothing to do on emulation, KVM will trap this in the kernel */ 1230 return H_SUCCESS; 1231 } 1232 1233 static target_ulong h_set_mode_resource_le(PowerPCCPU *cpu, 1234 target_ulong mflags, 1235 target_ulong value1, 1236 target_ulong value2) 1237 { 1238 CPUState *cs; 1239 1240 if (value1) { 1241 return H_P3; 1242 } 1243 if (value2) { 1244 return H_P4; 1245 } 1246 1247 switch (mflags) { 1248 case H_SET_MODE_ENDIAN_BIG: 1249 CPU_FOREACH(cs) { 1250 set_spr(cs, SPR_LPCR, 0, LPCR_ILE); 1251 } 1252 spapr_pci_switch_vga(true); 1253 return H_SUCCESS; 1254 1255 case H_SET_MODE_ENDIAN_LITTLE: 1256 CPU_FOREACH(cs) { 1257 set_spr(cs, SPR_LPCR, LPCR_ILE, LPCR_ILE); 1258 } 1259 spapr_pci_switch_vga(false); 1260 return H_SUCCESS; 1261 } 1262 1263 return H_UNSUPPORTED_FLAG; 1264 } 1265 1266 static target_ulong h_set_mode_resource_addr_trans_mode(PowerPCCPU *cpu, 1267 target_ulong mflags, 1268 target_ulong value1, 1269 target_ulong value2) 1270 { 1271 CPUState *cs; 1272 PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cpu); 1273 1274 if (!(pcc->insns_flags2 & PPC2_ISA207S)) { 1275 return H_P2; 1276 } 1277 if (value1) { 1278 return H_P3; 1279 } 1280 if (value2) { 1281 return H_P4; 1282 } 1283 1284 if (mflags == AIL_RESERVED) { 1285 return H_UNSUPPORTED_FLAG; 1286 } 1287 1288 CPU_FOREACH(cs) { 1289 set_spr(cs, SPR_LPCR, mflags << LPCR_AIL_SHIFT, LPCR_AIL); 1290 } 1291 1292 return H_SUCCESS; 1293 } 1294 1295 static target_ulong h_set_mode(PowerPCCPU *cpu, sPAPRMachineState *spapr, 1296 target_ulong opcode, target_ulong *args) 1297 { 1298 target_ulong resource = args[1]; 1299 target_ulong ret = H_P2; 1300 1301 switch (resource) { 1302 case H_SET_MODE_RESOURCE_LE: 1303 ret = h_set_mode_resource_le(cpu, args[0], args[2], args[3]); 1304 break; 1305 case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE: 1306 ret = h_set_mode_resource_addr_trans_mode(cpu, args[0], 1307 args[2], args[3]); 1308 break; 1309 } 1310 1311 return ret; 1312 } 1313 1314 static target_ulong h_clean_slb(PowerPCCPU *cpu, sPAPRMachineState *spapr, 1315 target_ulong opcode, target_ulong *args) 1316 { 1317 qemu_log_mask(LOG_UNIMP, "Unimplemented SPAPR hcall 0x"TARGET_FMT_lx"%s\n", 1318 opcode, " (H_CLEAN_SLB)"); 1319 return H_FUNCTION; 1320 } 1321 1322 static target_ulong h_invalidate_pid(PowerPCCPU *cpu, sPAPRMachineState *spapr, 1323 target_ulong opcode, target_ulong *args) 1324 { 1325 qemu_log_mask(LOG_UNIMP, "Unimplemented SPAPR hcall 0x"TARGET_FMT_lx"%s\n", 1326 opcode, " (H_INVALIDATE_PID)"); 1327 return H_FUNCTION; 1328 } 1329 1330 static void spapr_check_setup_free_hpt(sPAPRMachineState *spapr, 1331 uint64_t patbe_old, uint64_t patbe_new) 1332 { 1333 /* 1334 * We have 4 Options: 1335 * HASH->HASH || RADIX->RADIX || NOTHING->RADIX : Do Nothing 1336 * HASH->RADIX : Free HPT 1337 * RADIX->HASH : Allocate HPT 1338 * NOTHING->HASH : Allocate HPT 1339 * Note: NOTHING implies the case where we said the guest could choose 1340 * later and so assumed radix and now it's called H_REG_PROC_TBL 1341 */ 1342 1343 if ((patbe_old & PATBE1_GR) == (patbe_new & PATBE1_GR)) { 1344 /* We assume RADIX, so this catches all the "Do Nothing" cases */ 1345 } else if (!(patbe_old & PATBE1_GR)) { 1346 /* HASH->RADIX : Free HPT */ 1347 spapr_free_hpt(spapr); 1348 } else if (!(patbe_new & PATBE1_GR)) { 1349 /* RADIX->HASH || NOTHING->HASH : Allocate HPT */ 1350 spapr_setup_hpt_and_vrma(spapr); 1351 } 1352 return; 1353 } 1354 1355 #define FLAGS_MASK 0x01FULL 1356 #define FLAG_MODIFY 0x10 1357 #define FLAG_REGISTER 0x08 1358 #define FLAG_RADIX 0x04 1359 #define FLAG_HASH_PROC_TBL 0x02 1360 #define FLAG_GTSE 0x01 1361 1362 static target_ulong h_register_process_table(PowerPCCPU *cpu, 1363 sPAPRMachineState *spapr, 1364 target_ulong opcode, 1365 target_ulong *args) 1366 { 1367 CPUState *cs; 1368 target_ulong flags = args[0]; 1369 target_ulong proc_tbl = args[1]; 1370 target_ulong page_size = args[2]; 1371 target_ulong table_size = args[3]; 1372 uint64_t cproc; 1373 1374 if (flags & ~FLAGS_MASK) { /* Check no reserved bits are set */ 1375 return H_PARAMETER; 1376 } 1377 if (flags & FLAG_MODIFY) { 1378 if (flags & FLAG_REGISTER) { 1379 if (flags & FLAG_RADIX) { /* Register new RADIX process table */ 1380 if (proc_tbl & 0xfff || proc_tbl >> 60) { 1381 return H_P2; 1382 } else if (page_size) { 1383 return H_P3; 1384 } else if (table_size > 24) { 1385 return H_P4; 1386 } 1387 cproc = PATBE1_GR | proc_tbl | table_size; 1388 } else { /* Register new HPT process table */ 1389 if (flags & FLAG_HASH_PROC_TBL) { /* Hash with Segment Tables */ 1390 /* TODO - Not Supported */ 1391 /* Technically caused by flag bits => H_PARAMETER */ 1392 return H_PARAMETER; 1393 } else { /* Hash with SLB */ 1394 if (proc_tbl >> 38) { 1395 return H_P2; 1396 } else if (page_size & ~0x7) { 1397 return H_P3; 1398 } else if (table_size > 24) { 1399 return H_P4; 1400 } 1401 } 1402 cproc = (proc_tbl << 25) | page_size << 5 | table_size; 1403 } 1404 1405 } else { /* Deregister current process table */ 1406 /* Set to benign value: (current GR) | 0. This allows 1407 * deregistration in KVM to succeed even if the radix bit in flags 1408 * doesn't match the radix bit in the old PATB. */ 1409 cproc = spapr->patb_entry & PATBE1_GR; 1410 } 1411 } else { /* Maintain current registration */ 1412 if (!(flags & FLAG_RADIX) != !(spapr->patb_entry & PATBE1_GR)) { 1413 /* Technically caused by flag bits => H_PARAMETER */ 1414 return H_PARAMETER; /* Existing Process Table Mismatch */ 1415 } 1416 cproc = spapr->patb_entry; 1417 } 1418 1419 /* Check if we need to setup OR free the hpt */ 1420 spapr_check_setup_free_hpt(spapr, spapr->patb_entry, cproc); 1421 1422 spapr->patb_entry = cproc; /* Save new process table */ 1423 1424 /* Update the UPRT and GTSE bits in the LPCR for all cpus */ 1425 CPU_FOREACH(cs) { 1426 set_spr(cs, SPR_LPCR, 1427 ((flags & (FLAG_RADIX | FLAG_HASH_PROC_TBL)) ? LPCR_UPRT : 0) | 1428 ((flags & FLAG_GTSE) ? LPCR_GTSE : 0), 1429 LPCR_UPRT | LPCR_GTSE); 1430 } 1431 1432 if (kvm_enabled()) { 1433 return kvmppc_configure_v3_mmu(cpu, flags & FLAG_RADIX, 1434 flags & FLAG_GTSE, cproc); 1435 } 1436 return H_SUCCESS; 1437 } 1438 1439 #define H_SIGNAL_SYS_RESET_ALL -1 1440 #define H_SIGNAL_SYS_RESET_ALLBUTSELF -2 1441 1442 static target_ulong h_signal_sys_reset(PowerPCCPU *cpu, 1443 sPAPRMachineState *spapr, 1444 target_ulong opcode, target_ulong *args) 1445 { 1446 target_long target = args[0]; 1447 CPUState *cs; 1448 1449 if (target < 0) { 1450 /* Broadcast */ 1451 if (target < H_SIGNAL_SYS_RESET_ALLBUTSELF) { 1452 return H_PARAMETER; 1453 } 1454 1455 CPU_FOREACH(cs) { 1456 PowerPCCPU *c = POWERPC_CPU(cs); 1457 1458 if (target == H_SIGNAL_SYS_RESET_ALLBUTSELF) { 1459 if (c == cpu) { 1460 continue; 1461 } 1462 } 1463 run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL); 1464 } 1465 return H_SUCCESS; 1466 1467 } else { 1468 /* Unicast */ 1469 cs = CPU(spapr_find_cpu(target)); 1470 if (cs) { 1471 run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL); 1472 return H_SUCCESS; 1473 } 1474 return H_PARAMETER; 1475 } 1476 } 1477 1478 static uint32_t cas_check_pvr(sPAPRMachineState *spapr, PowerPCCPU *cpu, 1479 target_ulong *addr, bool *raw_mode_supported, 1480 Error **errp) 1481 { 1482 bool explicit_match = false; /* Matched the CPU's real PVR */ 1483 uint32_t max_compat = spapr->max_compat_pvr; 1484 uint32_t best_compat = 0; 1485 int i; 1486 1487 /* 1488 * We scan the supplied table of PVRs looking for two things 1489 * 1. Is our real CPU PVR in the list? 1490 * 2. What's the "best" listed logical PVR 1491 */ 1492 for (i = 0; i < 512; ++i) { 1493 uint32_t pvr, pvr_mask; 1494 1495 pvr_mask = ldl_be_phys(&address_space_memory, *addr); 1496 pvr = ldl_be_phys(&address_space_memory, *addr + 4); 1497 *addr += 8; 1498 1499 if (~pvr_mask & pvr) { 1500 break; /* Terminator record */ 1501 } 1502 1503 if ((cpu->env.spr[SPR_PVR] & pvr_mask) == (pvr & pvr_mask)) { 1504 explicit_match = true; 1505 } else { 1506 if (ppc_check_compat(cpu, pvr, best_compat, max_compat)) { 1507 best_compat = pvr; 1508 } 1509 } 1510 } 1511 1512 if ((best_compat == 0) && (!explicit_match || max_compat)) { 1513 /* We couldn't find a suitable compatibility mode, and either 1514 * the guest doesn't support "raw" mode for this CPU, or raw 1515 * mode is disabled because a maximum compat mode is set */ 1516 error_setg(errp, "Couldn't negotiate a suitable PVR during CAS"); 1517 return 0; 1518 } 1519 1520 *raw_mode_supported = explicit_match; 1521 1522 /* Parsing finished */ 1523 trace_spapr_cas_pvr(cpu->compat_pvr, explicit_match, best_compat); 1524 1525 return best_compat; 1526 } 1527 1528 static target_ulong h_client_architecture_support(PowerPCCPU *cpu, 1529 sPAPRMachineState *spapr, 1530 target_ulong opcode, 1531 target_ulong *args) 1532 { 1533 /* Working address in data buffer */ 1534 target_ulong addr = ppc64_phys_to_real(args[0]); 1535 target_ulong ov_table; 1536 uint32_t cas_pvr; 1537 sPAPROptionVector *ov1_guest, *ov5_guest, *ov5_cas_old, *ov5_updates; 1538 bool guest_radix; 1539 Error *local_err = NULL; 1540 bool raw_mode_supported = false; 1541 1542 cas_pvr = cas_check_pvr(spapr, cpu, &addr, &raw_mode_supported, &local_err); 1543 if (local_err) { 1544 error_report_err(local_err); 1545 return H_HARDWARE; 1546 } 1547 1548 /* Update CPUs */ 1549 if (cpu->compat_pvr != cas_pvr) { 1550 ppc_set_compat_all(cas_pvr, &local_err); 1551 if (local_err) { 1552 /* We fail to set compat mode (likely because running with KVM PR), 1553 * but maybe we can fallback to raw mode if the guest supports it. 1554 */ 1555 if (!raw_mode_supported) { 1556 error_report_err(local_err); 1557 return H_HARDWARE; 1558 } 1559 local_err = NULL; 1560 } 1561 } 1562 1563 /* For the future use: here @ov_table points to the first option vector */ 1564 ov_table = addr; 1565 1566 ov1_guest = spapr_ovec_parse_vector(ov_table, 1); 1567 ov5_guest = spapr_ovec_parse_vector(ov_table, 5); 1568 if (spapr_ovec_test(ov5_guest, OV5_MMU_BOTH)) { 1569 error_report("guest requested hash and radix MMU, which is invalid."); 1570 exit(EXIT_FAILURE); 1571 } 1572 /* The radix/hash bit in byte 24 requires special handling: */ 1573 guest_radix = spapr_ovec_test(ov5_guest, OV5_MMU_RADIX_300); 1574 spapr_ovec_clear(ov5_guest, OV5_MMU_RADIX_300); 1575 1576 /* 1577 * HPT resizing is a bit of a special case, because when enabled 1578 * we assume an HPT guest will support it until it says it 1579 * doesn't, instead of assuming it won't support it until it says 1580 * it does. Strictly speaking that approach could break for 1581 * guests which don't make a CAS call, but those are so old we 1582 * don't care about them. Without that assumption we'd have to 1583 * make at least a temporary allocation of an HPT sized for max 1584 * memory, which could be impossibly difficult under KVM HV if 1585 * maxram is large. 1586 */ 1587 if (!guest_radix && !spapr_ovec_test(ov5_guest, OV5_HPT_RESIZE)) { 1588 int maxshift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size); 1589 1590 if (spapr->resize_hpt == SPAPR_RESIZE_HPT_REQUIRED) { 1591 error_report( 1592 "h_client_architecture_support: Guest doesn't support HPT resizing, but resize-hpt=required"); 1593 exit(1); 1594 } 1595 1596 if (spapr->htab_shift < maxshift) { 1597 /* Guest doesn't know about HPT resizing, so we 1598 * pre-emptively resize for the maximum permitted RAM. At 1599 * the point this is called, nothing should have been 1600 * entered into the existing HPT */ 1601 spapr_reallocate_hpt(spapr, maxshift, &error_fatal); 1602 push_sregs_to_kvm_pr(spapr); 1603 } 1604 } 1605 1606 /* NOTE: there are actually a number of ov5 bits where input from the 1607 * guest is always zero, and the platform/QEMU enables them independently 1608 * of guest input. To model these properly we'd want some sort of mask, 1609 * but since they only currently apply to memory migration as defined 1610 * by LoPAPR 1.1, 14.5.4.8, which QEMU doesn't implement, we don't need 1611 * to worry about this for now. 1612 */ 1613 ov5_cas_old = spapr_ovec_clone(spapr->ov5_cas); 1614 1615 /* also clear the radix/hash bit from the current ov5_cas bits to 1616 * be in sync with the newly ov5 bits. Else the radix bit will be 1617 * seen as being removed and this will generate a reset loop 1618 */ 1619 spapr_ovec_clear(ov5_cas_old, OV5_MMU_RADIX_300); 1620 1621 /* full range of negotiated ov5 capabilities */ 1622 spapr_ovec_intersect(spapr->ov5_cas, spapr->ov5, ov5_guest); 1623 spapr_ovec_cleanup(ov5_guest); 1624 /* capabilities that have been added since CAS-generated guest reset. 1625 * if capabilities have since been removed, generate another reset 1626 */ 1627 ov5_updates = spapr_ovec_new(); 1628 spapr->cas_reboot = spapr_ovec_diff(ov5_updates, 1629 ov5_cas_old, spapr->ov5_cas); 1630 /* Now that processing is finished, set the radix/hash bit for the 1631 * guest if it requested a valid mode; otherwise terminate the boot. */ 1632 if (guest_radix) { 1633 if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) { 1634 error_report("Guest requested unavailable MMU mode (radix)."); 1635 exit(EXIT_FAILURE); 1636 } 1637 spapr_ovec_set(spapr->ov5_cas, OV5_MMU_RADIX_300); 1638 } else { 1639 if (kvm_enabled() && kvmppc_has_cap_mmu_radix() 1640 && !kvmppc_has_cap_mmu_hash_v3()) { 1641 error_report("Guest requested unavailable MMU mode (hash)."); 1642 exit(EXIT_FAILURE); 1643 } 1644 } 1645 spapr->cas_legacy_guest_workaround = !spapr_ovec_test(ov1_guest, 1646 OV1_PPC_3_00); 1647 if (!spapr->cas_reboot) { 1648 /* If spapr_machine_reset() did not set up a HPT but one is necessary 1649 * (because the guest isn't going to use radix) then set it up here. */ 1650 if ((spapr->patb_entry & PATBE1_GR) && !guest_radix) { 1651 /* legacy hash or new hash: */ 1652 spapr_setup_hpt_and_vrma(spapr); 1653 } 1654 spapr->cas_reboot = 1655 (spapr_h_cas_compose_response(spapr, args[1], args[2], 1656 ov5_updates) != 0); 1657 } 1658 spapr_ovec_cleanup(ov5_updates); 1659 1660 if (spapr->cas_reboot) { 1661 qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET); 1662 } 1663 1664 return H_SUCCESS; 1665 } 1666 1667 static target_ulong h_get_cpu_characteristics(PowerPCCPU *cpu, 1668 sPAPRMachineState *spapr, 1669 target_ulong opcode, 1670 target_ulong *args) 1671 { 1672 uint64_t characteristics = H_CPU_CHAR_HON_BRANCH_HINTS & 1673 ~H_CPU_CHAR_THR_RECONF_TRIG; 1674 uint64_t behaviour = H_CPU_BEHAV_FAVOUR_SECURITY; 1675 uint8_t safe_cache = spapr_get_cap(spapr, SPAPR_CAP_CFPC); 1676 uint8_t safe_bounds_check = spapr_get_cap(spapr, SPAPR_CAP_SBBC); 1677 uint8_t safe_indirect_branch = spapr_get_cap(spapr, SPAPR_CAP_IBS); 1678 1679 switch (safe_cache) { 1680 case SPAPR_CAP_WORKAROUND: 1681 characteristics |= H_CPU_CHAR_L1D_FLUSH_ORI30; 1682 characteristics |= H_CPU_CHAR_L1D_FLUSH_TRIG2; 1683 characteristics |= H_CPU_CHAR_L1D_THREAD_PRIV; 1684 behaviour |= H_CPU_BEHAV_L1D_FLUSH_PR; 1685 break; 1686 case SPAPR_CAP_FIXED: 1687 break; 1688 default: /* broken */ 1689 assert(safe_cache == SPAPR_CAP_BROKEN); 1690 behaviour |= H_CPU_BEHAV_L1D_FLUSH_PR; 1691 break; 1692 } 1693 1694 switch (safe_bounds_check) { 1695 case SPAPR_CAP_WORKAROUND: 1696 characteristics |= H_CPU_CHAR_SPEC_BAR_ORI31; 1697 behaviour |= H_CPU_BEHAV_BNDS_CHK_SPEC_BAR; 1698 break; 1699 case SPAPR_CAP_FIXED: 1700 break; 1701 default: /* broken */ 1702 assert(safe_bounds_check == SPAPR_CAP_BROKEN); 1703 behaviour |= H_CPU_BEHAV_BNDS_CHK_SPEC_BAR; 1704 break; 1705 } 1706 1707 switch (safe_indirect_branch) { 1708 case SPAPR_CAP_FIXED: 1709 characteristics |= H_CPU_CHAR_BCCTRL_SERIALISED; 1710 break; 1711 default: /* broken */ 1712 assert(safe_indirect_branch == SPAPR_CAP_BROKEN); 1713 break; 1714 } 1715 1716 args[0] = characteristics; 1717 args[1] = behaviour; 1718 1719 return H_SUCCESS; 1720 } 1721 1722 static spapr_hcall_fn papr_hypercall_table[(MAX_HCALL_OPCODE / 4) + 1]; 1723 static spapr_hcall_fn kvmppc_hypercall_table[KVMPPC_HCALL_MAX - KVMPPC_HCALL_BASE + 1]; 1724 1725 void spapr_register_hypercall(target_ulong opcode, spapr_hcall_fn fn) 1726 { 1727 spapr_hcall_fn *slot; 1728 1729 if (opcode <= MAX_HCALL_OPCODE) { 1730 assert((opcode & 0x3) == 0); 1731 1732 slot = &papr_hypercall_table[opcode / 4]; 1733 } else { 1734 assert((opcode >= KVMPPC_HCALL_BASE) && (opcode <= KVMPPC_HCALL_MAX)); 1735 1736 slot = &kvmppc_hypercall_table[opcode - KVMPPC_HCALL_BASE]; 1737 } 1738 1739 assert(!(*slot)); 1740 *slot = fn; 1741 } 1742 1743 target_ulong spapr_hypercall(PowerPCCPU *cpu, target_ulong opcode, 1744 target_ulong *args) 1745 { 1746 sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine()); 1747 1748 if ((opcode <= MAX_HCALL_OPCODE) 1749 && ((opcode & 0x3) == 0)) { 1750 spapr_hcall_fn fn = papr_hypercall_table[opcode / 4]; 1751 1752 if (fn) { 1753 return fn(cpu, spapr, opcode, args); 1754 } 1755 } else if ((opcode >= KVMPPC_HCALL_BASE) && 1756 (opcode <= KVMPPC_HCALL_MAX)) { 1757 spapr_hcall_fn fn = kvmppc_hypercall_table[opcode - KVMPPC_HCALL_BASE]; 1758 1759 if (fn) { 1760 return fn(cpu, spapr, opcode, args); 1761 } 1762 } 1763 1764 qemu_log_mask(LOG_UNIMP, "Unimplemented SPAPR hcall 0x" TARGET_FMT_lx "\n", 1765 opcode); 1766 return H_FUNCTION; 1767 } 1768 1769 static void hypercall_register_types(void) 1770 { 1771 /* hcall-pft */ 1772 spapr_register_hypercall(H_ENTER, h_enter); 1773 spapr_register_hypercall(H_REMOVE, h_remove); 1774 spapr_register_hypercall(H_PROTECT, h_protect); 1775 spapr_register_hypercall(H_READ, h_read); 1776 1777 /* hcall-bulk */ 1778 spapr_register_hypercall(H_BULK_REMOVE, h_bulk_remove); 1779 1780 /* hcall-hpt-resize */ 1781 spapr_register_hypercall(H_RESIZE_HPT_PREPARE, h_resize_hpt_prepare); 1782 spapr_register_hypercall(H_RESIZE_HPT_COMMIT, h_resize_hpt_commit); 1783 1784 /* hcall-splpar */ 1785 spapr_register_hypercall(H_REGISTER_VPA, h_register_vpa); 1786 spapr_register_hypercall(H_CEDE, h_cede); 1787 spapr_register_hypercall(H_SIGNAL_SYS_RESET, h_signal_sys_reset); 1788 1789 /* processor register resource access h-calls */ 1790 spapr_register_hypercall(H_SET_SPRG0, h_set_sprg0); 1791 spapr_register_hypercall(H_SET_DABR, h_set_dabr); 1792 spapr_register_hypercall(H_SET_XDABR, h_set_xdabr); 1793 spapr_register_hypercall(H_PAGE_INIT, h_page_init); 1794 spapr_register_hypercall(H_SET_MODE, h_set_mode); 1795 1796 /* In Memory Table MMU h-calls */ 1797 spapr_register_hypercall(H_CLEAN_SLB, h_clean_slb); 1798 spapr_register_hypercall(H_INVALIDATE_PID, h_invalidate_pid); 1799 spapr_register_hypercall(H_REGISTER_PROC_TBL, h_register_process_table); 1800 1801 /* hcall-get-cpu-characteristics */ 1802 spapr_register_hypercall(H_GET_CPU_CHARACTERISTICS, 1803 h_get_cpu_characteristics); 1804 1805 /* "debugger" hcalls (also used by SLOF). Note: We do -not- differenciate 1806 * here between the "CI" and the "CACHE" variants, they will use whatever 1807 * mapping attributes qemu is using. When using KVM, the kernel will 1808 * enforce the attributes more strongly 1809 */ 1810 spapr_register_hypercall(H_LOGICAL_CI_LOAD, h_logical_load); 1811 spapr_register_hypercall(H_LOGICAL_CI_STORE, h_logical_store); 1812 spapr_register_hypercall(H_LOGICAL_CACHE_LOAD, h_logical_load); 1813 spapr_register_hypercall(H_LOGICAL_CACHE_STORE, h_logical_store); 1814 spapr_register_hypercall(H_LOGICAL_ICBI, h_logical_icbi); 1815 spapr_register_hypercall(H_LOGICAL_DCBF, h_logical_dcbf); 1816 spapr_register_hypercall(KVMPPC_H_LOGICAL_MEMOP, h_logical_memop); 1817 1818 /* qemu/KVM-PPC specific hcalls */ 1819 spapr_register_hypercall(KVMPPC_H_RTAS, h_rtas); 1820 1821 /* ibm,client-architecture-support support */ 1822 spapr_register_hypercall(KVMPPC_H_CAS, h_client_architecture_support); 1823 } 1824 1825 type_init(hypercall_register_types) 1826