hw/ppc/spapr_caps.c

   1 /*
   2  * QEMU PowerPC pSeries Logical Partition capabilities handling
   3  *
   4  * Copyright (c) 2017 David Gibson, Red Hat Inc.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24
  25 #include "qemu/osdep.h"
  26 #include "qemu/error-report.h"
  27 #include "qapi/error.h"
  28 #include "qapi/visitor.h"
  29 #include "sysemu/hw_accel.h"
  30 #include "exec/ram_addr.h"
  31 #include "target/ppc/cpu.h"
  32 #include "target/ppc/mmu-hash64.h"
  33 #include "cpu-models.h"
  34 #include "kvm_ppc.h"
  35 #include "migration/vmstate.h"
  36 #include "sysemu/qtest.h"
  37 #include "sysemu/tcg.h"
  38
  39 #include "hw/ppc/spapr.h"
  40
  41 typedef struct SpaprCapPossible {
  42     int num;            /* size of vals array below */
  43     const char *help;   /* help text for vals */
  44     /*
  45      * Note:
  46      * - because of the way compatibility is determined vals MUST be ordered
  47      *   such that later options are a superset of all preceding options.
  48      * - the order of vals must be preserved, that is their index is important,
  49      *   however vals may be added to the end of the list so long as the above
  50      *   point is observed
  51      */
  52     const char *vals[];
  53 } SpaprCapPossible;
  54
  55 typedef struct SpaprCapabilityInfo {
  56     const char *name;
  57     const char *description;
  58     int index;
  59
  60     /* Getter and Setter Function Pointers */
  61     ObjectPropertyAccessor *get;
  62     ObjectPropertyAccessor *set;
  63     const char *type;
  64     /* Possible values if this is a custom string type */
  65     SpaprCapPossible *possible;
  66     /* Make sure the virtual hardware can support this capability */
  67     void (*apply)(SpaprMachineState *spapr, uint8_t val, Error **errp);
  68     void (*cpu_apply)(SpaprMachineState *spapr, PowerPCCPU *cpu,
  69                       uint8_t val, Error **errp);
  70     bool (*migrate_needed)(void *opaque);
  71 } SpaprCapabilityInfo;
  72
  73 static void spapr_cap_get_bool(Object *obj, Visitor *v, const char *name,
  74                                void *opaque, Error **errp)
  75 {
  76     SpaprCapabilityInfo *cap = opaque;
  77     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
  78     bool value = spapr_get_cap(spapr, cap->index) == SPAPR_CAP_ON;
  79
  80     visit_type_bool(v, name, &value, errp);
  81 }
  82
  83 static void spapr_cap_set_bool(Object *obj, Visitor *v, const char *name,
  84                                void *opaque, Error **errp)
  85 {
  86     SpaprCapabilityInfo *cap = opaque;
  87     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
  88     bool value;
  89     Error *local_err = NULL;
  90
  91     visit_type_bool(v, name, &value, &local_err);
  92     if (local_err) {
  93         error_propagate(errp, local_err);
  94         return;
  95     }
  96
  97     spapr->cmd_line_caps[cap->index] = true;
  98     spapr->eff.caps[cap->index] = value ? SPAPR_CAP_ON : SPAPR_CAP_OFF;
  99 }
 100
 101
 102 static void  spapr_cap_get_string(Object *obj, Visitor *v, const char *name,
 103                                   void *opaque, Error **errp)
 104 {
 105     SpaprCapabilityInfo *cap = opaque;
 106     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
 107     char *val = NULL;
 108     uint8_t value = spapr_get_cap(spapr, cap->index);
 109
 110     if (value >= cap->possible->num) {
 111         error_setg(errp, "Invalid value (%d) for cap-%s", value, cap->name);
 112         return;
 113     }
 114
 115     val = g_strdup(cap->possible->vals[value]);
 116
 117     visit_type_str(v, name, &val, errp);
 118     g_free(val);
 119 }
 120
 121 static void spapr_cap_set_string(Object *obj, Visitor *v, const char *name,
 122                                  void *opaque, Error **errp)
 123 {
 124     SpaprCapabilityInfo *cap = opaque;
 125     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
 126     Error *local_err = NULL;
 127     uint8_t i;
 128     char *val;
 129
 130     visit_type_str(v, name, &val, &local_err);
 131     if (local_err) {
 132         error_propagate(errp, local_err);
 133         return;
 134     }
 135
 136     if (!strcmp(val, "?")) {
 137         error_setg(errp, "%s", cap->possible->help);
 138         goto out;
 139     }
 140     for (i = 0; i < cap->possible->num; i++) {
 141         if (!strcasecmp(val, cap->possible->vals[i])) {
 142             spapr->cmd_line_caps[cap->index] = true;
 143             spapr->eff.caps[cap->index] = i;
 144             goto out;
 145         }
 146     }
 147
 148     error_setg(errp, "Invalid capability mode \"%s\" for cap-%s", val,
 149                cap->name);
 150 out:
 151     g_free(val);
 152 }
 153
 154 static void spapr_cap_get_pagesize(Object *obj, Visitor *v, const char *name,
 155                                    void *opaque, Error **errp)
 156 {
 157     SpaprCapabilityInfo *cap = opaque;
 158     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
 159     uint8_t val = spapr_get_cap(spapr, cap->index);
 160     uint64_t pagesize = (1ULL << val);
 161
 162     visit_type_size(v, name, &pagesize, errp);
 163 }
 164
 165 static void spapr_cap_set_pagesize(Object *obj, Visitor *v, const char *name,
 166                                    void *opaque, Error **errp)
 167 {
 168     SpaprCapabilityInfo *cap = opaque;
 169     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
 170     uint64_t pagesize;
 171     uint8_t val;
 172     Error *local_err = NULL;
 173
 174     visit_type_size(v, name, &pagesize, &local_err);
 175     if (local_err) {
 176         error_propagate(errp, local_err);
 177         return;
 178     }
 179
 180     if (!is_power_of_2(pagesize)) {
 181         error_setg(errp, "cap-%s must be a power of 2", cap->name);
 182         return;
 183     }
 184
 185     val = ctz64(pagesize);
 186     spapr->cmd_line_caps[cap->index] = true;
 187     spapr->eff.caps[cap->index] = val;
 188 }
 189
 190 static void cap_htm_apply(SpaprMachineState *spapr, uint8_t val, Error **errp)
 191 {
 192     if (!val) {
 193         /* TODO: We don't support disabling htm yet */
 194         return;
 195     }
 196     if (tcg_enabled()) {
 197         error_setg(errp,
 198                    "No Transactional Memory support in TCG,"
 199                    " try appending -machine cap-htm=off");
 200     } else if (kvm_enabled() && !kvmppc_has_cap_htm()) {
 201         error_setg(errp,
 202 "KVM implementation does not support Transactional Memory,"
 203                    " try appending -machine cap-htm=off"
 204             );
 205     }
 206 }
 207
 208 static void cap_vsx_apply(SpaprMachineState *spapr, uint8_t val, Error **errp)
 209 {
 210     PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
 211     CPUPPCState *env = &cpu->env;
 212
 213     if (!val) {
 214         /* TODO: We don't support disabling vsx yet */
 215         return;
 216     }
 217     /* Allowable CPUs in spapr_cpu_core.c should already have gotten
 218      * rid of anything that doesn't do VMX */
 219     g_assert(env->insns_flags & PPC_ALTIVEC);
 220     if (!(env->insns_flags2 & PPC2_VSX)) {
 221         error_setg(errp, "VSX support not available,"
 222                    " try appending -machine cap-vsx=off");
 223     }
 224 }
 225
 226 static void cap_dfp_apply(SpaprMachineState *spapr, uint8_t val, Error **errp)
 227 {
 228     PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
 229     CPUPPCState *env = &cpu->env;
 230
 231     if (!val) {
 232         /* TODO: We don't support disabling dfp yet */
 233         return;
 234     }
 235     if (!(env->insns_flags2 & PPC2_DFP)) {
 236         error_setg(errp, "DFP support not available,"
 237                    " try appending -machine cap-dfp=off");
 238     }
 239 }
 240
 241 SpaprCapPossible cap_cfpc_possible = {
 242     .num = 3,
 243     .vals = {"broken", "workaround", "fixed"},
 244     .help = "broken - no protection, workaround - workaround available,"
 245             " fixed - fixed in hardware",
 246 };
 247
 248 static void cap_safe_cache_apply(SpaprMachineState *spapr, uint8_t val,
 249                                  Error **errp)
 250 {
 251     Error *local_err = NULL;
 252     uint8_t kvm_val =  kvmppc_get_cap_safe_cache();
 253
 254     if (tcg_enabled() && val) {
 255         /* TCG only supports broken, allow other values and print a warning */
 256         error_setg(&local_err,
 257                    "TCG doesn't support requested feature, cap-cfpc=%s",
 258                    cap_cfpc_possible.vals[val]);
 259     } else if (kvm_enabled() && (val > kvm_val)) {
 260         error_setg(errp,
 261                    "Requested safe cache capability level not supported by kvm,"
 262                    " try appending -machine cap-cfpc=%s",
 263                    cap_cfpc_possible.vals[kvm_val]);
 264     }
 265
 266     if (local_err != NULL)
 267         warn_report_err(local_err);
 268 }
 269
 270 SpaprCapPossible cap_sbbc_possible = {
 271     .num = 3,
 272     .vals = {"broken", "workaround", "fixed"},
 273     .help = "broken - no protection, workaround - workaround available,"
 274             " fixed - fixed in hardware",
 275 };
 276
 277 static void cap_safe_bounds_check_apply(SpaprMachineState *spapr, uint8_t val,
 278                                         Error **errp)
 279 {
 280     Error *local_err = NULL;
 281     uint8_t kvm_val =  kvmppc_get_cap_safe_bounds_check();
 282
 283     if (tcg_enabled() && val) {
 284         /* TCG only supports broken, allow other values and print a warning */
 285         error_setg(&local_err,
 286                    "TCG doesn't support requested feature, cap-sbbc=%s",
 287                    cap_sbbc_possible.vals[val]);
 288     } else if (kvm_enabled() && (val > kvm_val)) {
 289         error_setg(errp,
 290 "Requested safe bounds check capability level not supported by kvm,"
 291                    " try appending -machine cap-sbbc=%s",
 292                    cap_sbbc_possible.vals[kvm_val]);
 293     }
 294
 295     if (local_err != NULL)
 296         warn_report_err(local_err);
 297 }
 298
 299 SpaprCapPossible cap_ibs_possible = {
 300     .num = 5,
 301     /* Note workaround only maintained for compatibility */
 302     .vals = {"broken", "workaround", "fixed-ibs", "fixed-ccd", "fixed-na"},
 303     .help = "broken - no protection, workaround - count cache flush"
 304             ", fixed-ibs - indirect branch serialisation,"
 305             " fixed-ccd - cache count disabled,"
 306             " fixed-na - fixed in hardware (no longer applicable)",
 307 };
 308
 309 static void cap_safe_indirect_branch_apply(SpaprMachineState *spapr,
 310                                            uint8_t val, Error **errp)
 311 {
 312     Error *local_err = NULL;
 313     uint8_t kvm_val = kvmppc_get_cap_safe_indirect_branch();
 314
 315     if (tcg_enabled() && val) {
 316         /* TCG only supports broken, allow other values and print a warning */
 317         error_setg(&local_err,
 318                    "TCG doesn't support requested feature, cap-ibs=%s",
 319                    cap_ibs_possible.vals[val]);
 320     } else if (kvm_enabled() && (val > kvm_val)) {
 321         error_setg(errp,
 322 "Requested safe indirect branch capability level not supported by kvm,"
 323                    " try appending -machine cap-ibs=%s",
 324                    cap_ibs_possible.vals[kvm_val]);
 325     }
 326
 327     if (local_err != NULL) {
 328         warn_report_err(local_err);
 329     }
 330 }
 331
 332 #define VALUE_DESC_TRISTATE     " (broken, workaround, fixed)"
 333
 334 void spapr_check_pagesize(SpaprMachineState *spapr, hwaddr pagesize,
 335                           Error **errp)
 336 {
 337     hwaddr maxpagesize = (1ULL << spapr->eff.caps[SPAPR_CAP_HPT_MAXPAGESIZE]);
 338
 339     if (!kvmppc_hpt_needs_host_contiguous_pages()) {
 340         return;
 341     }
 342
 343     if (maxpagesize > pagesize) {
 344         error_setg(errp,
 345                    "Can't support %"HWADDR_PRIu" kiB guest pages with %"
 346                    HWADDR_PRIu" kiB host pages with this KVM implementation",
 347                    maxpagesize >> 10, pagesize >> 10);
 348     }
 349 }
 350
 351 static void cap_hpt_maxpagesize_apply(SpaprMachineState *spapr,
 352                                       uint8_t val, Error **errp)
 353 {
 354     if (val < 12) {
 355         error_setg(errp, "Require at least 4kiB hpt-max-page-size");
 356         return;
 357     } else if (val < 16) {
 358         warn_report("Many guests require at least 64kiB hpt-max-page-size");
 359     }
 360
 361     spapr_check_pagesize(spapr, qemu_minrampagesize(), errp);
 362 }
 363
 364 static bool cap_hpt_maxpagesize_migrate_needed(void *opaque)
 365 {
 366     return !SPAPR_MACHINE_GET_CLASS(opaque)->pre_4_1_migration;
 367 }
 368
 369 static bool spapr_pagesize_cb(void *opaque, uint32_t seg_pshift,
 370                               uint32_t pshift)
 371 {
 372     unsigned maxshift = *((unsigned *)opaque);
 373
 374     assert(pshift >= seg_pshift);
 375
 376     /* Don't allow the guest to use pages bigger than the configured
 377      * maximum size */
 378     if (pshift > maxshift) {
 379         return false;
 380     }
 381
 382     /* For whatever reason, KVM doesn't allow multiple pagesizes
 383      * within a segment, *except* for the case of 16M pages in a 4k or
 384      * 64k segment.  Always exclude other cases, so that TCG and KVM
 385      * guests see a consistent environment */
 386     if ((pshift != seg_pshift) && (pshift != 24)) {
 387         return false;
 388     }
 389
 390     return true;
 391 }
 392
 393 static void cap_hpt_maxpagesize_cpu_apply(SpaprMachineState *spapr,
 394                                           PowerPCCPU *cpu,
 395                                           uint8_t val, Error **errp)
 396 {
 397     unsigned maxshift = val;
 398
 399     ppc_hash64_filter_pagesizes(cpu, spapr_pagesize_cb, &maxshift);
 400 }
 401
 402 static void cap_nested_kvm_hv_apply(SpaprMachineState *spapr,
 403                                     uint8_t val, Error **errp)
 404 {
 405     if (!val) {
 406         /* capability disabled by default */
 407         return;
 408     }
 409
 410     if (tcg_enabled()) {
 411         error_setg(errp,
 412                    "No Nested KVM-HV support in tcg,"
 413                    " try appending -machine cap-nested-hv=off");
 414     } else if (kvm_enabled()) {
 415         if (!kvmppc_has_cap_nested_kvm_hv()) {
 416             error_setg(errp,
 417 "KVM implementation does not support Nested KVM-HV,"
 418                        " try appending -machine cap-nested-hv=off");
 419         } else if (kvmppc_set_cap_nested_kvm_hv(val) < 0) {
 420                 error_setg(errp,
 421 "Error enabling cap-nested-hv with KVM, try cap-nested-hv=off");
 422         }
 423     }
 424 }
 425
 426 static void cap_large_decr_apply(SpaprMachineState *spapr,
 427                                  uint8_t val, Error **errp)
 428 {
 429     PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
 430     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cpu);
 431
 432     if (!val) {
 433         return; /* Disabled by default */
 434     }
 435
 436     if (tcg_enabled()) {
 437         if (!ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0,
 438                               spapr->max_compat_pvr)) {
 439             error_setg(errp,
 440                 "Large decrementer only supported on POWER9, try -cpu POWER9");
 441             return;
 442         }
 443     } else if (kvm_enabled()) {
 444         int kvm_nr_bits = kvmppc_get_cap_large_decr();
 445
 446         if (!kvm_nr_bits) {
 447             error_setg(errp,
 448                        "No large decrementer support,"
 449                         " try appending -machine cap-large-decr=off");
 450         } else if (pcc->lrg_decr_bits != kvm_nr_bits) {
 451             error_setg(errp,
 452 "KVM large decrementer size (%d) differs to model (%d),"
 453                 " try appending -machine cap-large-decr=off",
 454                 kvm_nr_bits, pcc->lrg_decr_bits);
 455         }
 456     }
 457 }
 458
 459 static void cap_large_decr_cpu_apply(SpaprMachineState *spapr,
 460                                      PowerPCCPU *cpu,
 461                                      uint8_t val, Error **errp)
 462 {
 463     CPUPPCState *env = &cpu->env;
 464     target_ulong lpcr = env->spr[SPR_LPCR];
 465
 466     if (kvm_enabled()) {
 467         if (kvmppc_enable_cap_large_decr(cpu, val)) {
 468             error_setg(errp,
 469                        "No large decrementer support,"
 470                        " try appending -machine cap-large-decr=off");
 471         }
 472     }
 473
 474     if (val) {
 475         lpcr |= LPCR_LD;
 476     } else {
 477         lpcr &= ~LPCR_LD;
 478     }
 479     ppc_store_lpcr(cpu, lpcr);
 480 }
 481
 482 static void cap_ccf_assist_apply(SpaprMachineState *spapr, uint8_t val,
 483                                  Error **errp)
 484 {
 485     uint8_t kvm_val = kvmppc_get_cap_count_cache_flush_assist();
 486
 487     if (tcg_enabled() && val) {
 488         /* TCG doesn't implement anything here, but allow with a warning */
 489         warn_report("TCG doesn't support requested feature, cap-ccf-assist=on");
 490     } else if (kvm_enabled() && (val > kvm_val)) {
 491         uint8_t kvm_ibs = kvmppc_get_cap_safe_indirect_branch();
 492
 493         if (kvm_ibs == SPAPR_CAP_FIXED_CCD) {
 494             /*
 495              * If we don't have CCF assist on the host, the assist
 496              * instruction is a harmless no-op.  It won't correctly
 497              * implement the cache count flush *but* if we have
 498              * count-cache-disabled in the host, that flush is
 499              * unnnecessary.  So, specifically allow this case.  This
 500              * allows us to have better performance on POWER9 DD2.3,
 501              * while still working on POWER9 DD2.2 and POWER8 host
 502              * cpus.
 503              */
 504             return;
 505         }
 506         error_setg(errp,
 507 "Requested count cache flush assist capability level not supported by kvm,"
 508                    " try appending -machine cap-ccf-assist=off");
 509     }
 510 }
 511
 512 static void cap_fwnmi_apply(SpaprMachineState *spapr, uint8_t val,
 513                                 Error **errp)
 514 {
 515     if (!val) {
 516         return; /* Disabled by default */
 517     }
 518
 519     if (kvm_enabled()) {
 520         if (!kvmppc_get_fwnmi()) {
 521             error_setg(errp,
 522 "Firmware Assisted Non-Maskable Interrupts(FWNMI) not supported by KVM.");
 523             error_append_hint(errp, "Try appending -machine cap-fwnmi=off\n");
 524         }
 525     }
 526 }
 527
 528 SpaprCapabilityInfo capability_table[SPAPR_CAP_NUM] = {
 529     [SPAPR_CAP_HTM] = {
 530         .name = "htm",
 531         .description = "Allow Hardware Transactional Memory (HTM)",
 532         .index = SPAPR_CAP_HTM,
 533         .get = spapr_cap_get_bool,
 534         .set = spapr_cap_set_bool,
 535         .type = "bool",
 536         .apply = cap_htm_apply,
 537     },
 538     [SPAPR_CAP_VSX] = {
 539         .name = "vsx",
 540         .description = "Allow Vector Scalar Extensions (VSX)",
 541         .index = SPAPR_CAP_VSX,
 542         .get = spapr_cap_get_bool,
 543         .set = spapr_cap_set_bool,
 544         .type = "bool",
 545         .apply = cap_vsx_apply,
 546     },
 547     [SPAPR_CAP_DFP] = {
 548         .name = "dfp",
 549         .description = "Allow Decimal Floating Point (DFP)",
 550         .index = SPAPR_CAP_DFP,
 551         .get = spapr_cap_get_bool,
 552         .set = spapr_cap_set_bool,
 553         .type = "bool",
 554         .apply = cap_dfp_apply,
 555     },
 556     [SPAPR_CAP_CFPC] = {
 557         .name = "cfpc",
 558         .description = "Cache Flush on Privilege Change" VALUE_DESC_TRISTATE,
 559         .index = SPAPR_CAP_CFPC,
 560         .get = spapr_cap_get_string,
 561         .set = spapr_cap_set_string,
 562         .type = "string",
 563         .possible = &cap_cfpc_possible,
 564         .apply = cap_safe_cache_apply,
 565     },
 566     [SPAPR_CAP_SBBC] = {
 567         .name = "sbbc",
 568         .description = "Speculation Barrier Bounds Checking" VALUE_DESC_TRISTATE,
 569         .index = SPAPR_CAP_SBBC,
 570         .get = spapr_cap_get_string,
 571         .set = spapr_cap_set_string,
 572         .type = "string",
 573         .possible = &cap_sbbc_possible,
 574         .apply = cap_safe_bounds_check_apply,
 575     },
 576     [SPAPR_CAP_IBS] = {
 577         .name = "ibs",
 578         .description =
 579             "Indirect Branch Speculation (broken, workaround, fixed-ibs,"
 580             "fixed-ccd, fixed-na)",
 581         .index = SPAPR_CAP_IBS,
 582         .get = spapr_cap_get_string,
 583         .set = spapr_cap_set_string,
 584         .type = "string",
 585         .possible = &cap_ibs_possible,
 586         .apply = cap_safe_indirect_branch_apply,
 587     },
 588     [SPAPR_CAP_HPT_MAXPAGESIZE] = {
 589         .name = "hpt-max-page-size",
 590         .description = "Maximum page size for Hash Page Table guests",
 591         .index = SPAPR_CAP_HPT_MAXPAGESIZE,
 592         .get = spapr_cap_get_pagesize,
 593         .set = spapr_cap_set_pagesize,
 594         .type = "int",
 595         .apply = cap_hpt_maxpagesize_apply,
 596         .cpu_apply = cap_hpt_maxpagesize_cpu_apply,
 597         .migrate_needed = cap_hpt_maxpagesize_migrate_needed,
 598     },
 599     [SPAPR_CAP_NESTED_KVM_HV] = {
 600         .name = "nested-hv",
 601         .description = "Allow Nested KVM-HV",
 602         .index = SPAPR_CAP_NESTED_KVM_HV,
 603         .get = spapr_cap_get_bool,
 604         .set = spapr_cap_set_bool,
 605         .type = "bool",
 606         .apply = cap_nested_kvm_hv_apply,
 607     },
 608     [SPAPR_CAP_LARGE_DECREMENTER] = {
 609         .name = "large-decr",
 610         .description = "Allow Large Decrementer",
 611         .index = SPAPR_CAP_LARGE_DECREMENTER,
 612         .get = spapr_cap_get_bool,
 613         .set = spapr_cap_set_bool,
 614         .type = "bool",
 615         .apply = cap_large_decr_apply,
 616         .cpu_apply = cap_large_decr_cpu_apply,
 617     },
 618     [SPAPR_CAP_CCF_ASSIST] = {
 619         .name = "ccf-assist",
 620         .description = "Count Cache Flush Assist via HW Instruction",
 621         .index = SPAPR_CAP_CCF_ASSIST,
 622         .get = spapr_cap_get_bool,
 623         .set = spapr_cap_set_bool,
 624         .type = "bool",
 625         .apply = cap_ccf_assist_apply,
 626     },
 627     [SPAPR_CAP_FWNMI] = {
 628         .name = "fwnmi",
 629         .description = "Implements PAPR FWNMI option",
 630         .index = SPAPR_CAP_FWNMI,
 631         .get = spapr_cap_get_bool,
 632         .set = spapr_cap_set_bool,
 633         .type = "bool",
 634         .apply = cap_fwnmi_apply,
 635     },
 636 };
 637
 638 static SpaprCapabilities default_caps_with_cpu(SpaprMachineState *spapr,
 639                                                const char *cputype)
 640 {
 641     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
 642     SpaprCapabilities caps;
 643
 644     caps = smc->default_caps;
 645
 646     if (!ppc_type_check_compat(cputype, CPU_POWERPC_LOGICAL_3_00,
 647                                0, spapr->max_compat_pvr)) {
 648         caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
 649     }
 650
 651     if (!ppc_type_check_compat(cputype, CPU_POWERPC_LOGICAL_2_07,
 652                                0, spapr->max_compat_pvr)) {
 653         caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
 654         caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
 655     }
 656
 657     if (!ppc_type_check_compat(cputype, CPU_POWERPC_LOGICAL_2_06_PLUS,
 658                                0, spapr->max_compat_pvr)) {
 659         caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
 660     }
 661
 662     if (!ppc_type_check_compat(cputype, CPU_POWERPC_LOGICAL_2_06,
 663                                0, spapr->max_compat_pvr)) {
 664         caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_OFF;
 665         caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_OFF;
 666         caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
 667     }
 668
 669     /* This is for pseries-2.12 and older */
 670     if (smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] == 0) {
 671         uint8_t mps;
 672
 673         if (kvmppc_hpt_needs_host_contiguous_pages()) {
 674             mps = ctz64(qemu_minrampagesize());
 675         } else {
 676             mps = 34; /* allow everything up to 16GiB, i.e. everything */
 677         }
 678
 679         caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = mps;
 680     }
 681
 682     return caps;
 683 }
 684
 685 int spapr_caps_pre_load(void *opaque)
 686 {
 687     SpaprMachineState *spapr = opaque;
 688
 689     /* Set to default so we can tell if this came in with the migration */
 690     spapr->mig = spapr->def;
 691     return 0;
 692 }
 693
 694 int spapr_caps_pre_save(void *opaque)
 695 {
 696     SpaprMachineState *spapr = opaque;
 697
 698     spapr->mig = spapr->eff;
 699     return 0;
 700 }
 701
 702 /* This has to be called from the top-level spapr post_load, not the
 703  * caps specific one.  Otherwise it wouldn't be called when the source
 704  * caps are all defaults, which could still conflict with overridden
 705  * caps on the destination */
 706 int spapr_caps_post_migration(SpaprMachineState *spapr)
 707 {
 708     int i;
 709     bool ok = true;
 710     SpaprCapabilities dstcaps = spapr->eff;
 711     SpaprCapabilities srccaps;
 712
 713     srccaps = default_caps_with_cpu(spapr, MACHINE(spapr)->cpu_type);
 714     for (i = 0; i < SPAPR_CAP_NUM; i++) {
 715         /* If not default value then assume came in with the migration */
 716         if (spapr->mig.caps[i] != spapr->def.caps[i]) {
 717             srccaps.caps[i] = spapr->mig.caps[i];
 718         }
 719     }
 720
 721     for (i = 0; i < SPAPR_CAP_NUM; i++) {
 722         SpaprCapabilityInfo *info = &capability_table[i];
 723
 724         if (srccaps.caps[i] > dstcaps.caps[i]) {
 725             error_report("cap-%s higher level (%d) in incoming stream than on destination (%d)",
 726                          info->name, srccaps.caps[i], dstcaps.caps[i]);
 727             ok = false;
 728         }
 729
 730         if (srccaps.caps[i] < dstcaps.caps[i]) {
 731             warn_report("cap-%s lower level (%d) in incoming stream than on destination (%d)",
 732                          info->name, srccaps.caps[i], dstcaps.caps[i]);
 733         }
 734     }
 735
 736     return ok ? 0 : -EINVAL;
 737 }
 738
 739 /* Used to generate the migration field and needed function for a spapr cap */
 740 #define SPAPR_CAP_MIG_STATE(sname, cap)                 \
 741 static bool spapr_cap_##sname##_needed(void *opaque)    \
 742 {                                                       \
 743     SpaprMachineState *spapr = opaque;                  \
 744     bool (*needed)(void *opaque) =                      \
 745         capability_table[cap].migrate_needed;           \
 746                                                         \
 747     return needed ? needed(opaque) : true &&            \
 748            spapr->cmd_line_caps[cap] &&                 \
 749            (spapr->eff.caps[cap] !=                     \
 750             spapr->def.caps[cap]);                      \
 751 }                                                       \
 752                                                         \
 753 const VMStateDescription vmstate_spapr_cap_##sname = {  \
 754     .name = "spapr/cap/" #sname,                        \
 755     .version_id = 1,                                    \
 756     .minimum_version_id = 1,                            \
 757     .needed = spapr_cap_##sname##_needed,               \
 758     .fields = (VMStateField[]) {                        \
 759         VMSTATE_UINT8(mig.caps[cap],                    \
 760                       SpaprMachineState),               \
 761         VMSTATE_END_OF_LIST()                           \
 762     },                                                  \
 763 }
 764
 765 SPAPR_CAP_MIG_STATE(htm, SPAPR_CAP_HTM);
 766 SPAPR_CAP_MIG_STATE(vsx, SPAPR_CAP_VSX);
 767 SPAPR_CAP_MIG_STATE(dfp, SPAPR_CAP_DFP);
 768 SPAPR_CAP_MIG_STATE(cfpc, SPAPR_CAP_CFPC);
 769 SPAPR_CAP_MIG_STATE(sbbc, SPAPR_CAP_SBBC);
 770 SPAPR_CAP_MIG_STATE(ibs, SPAPR_CAP_IBS);
 771 SPAPR_CAP_MIG_STATE(hpt_maxpagesize, SPAPR_CAP_HPT_MAXPAGESIZE);
 772 SPAPR_CAP_MIG_STATE(nested_kvm_hv, SPAPR_CAP_NESTED_KVM_HV);
 773 SPAPR_CAP_MIG_STATE(large_decr, SPAPR_CAP_LARGE_DECREMENTER);
 774 SPAPR_CAP_MIG_STATE(ccf_assist, SPAPR_CAP_CCF_ASSIST);
 775 SPAPR_CAP_MIG_STATE(fwnmi, SPAPR_CAP_FWNMI);
 776
 777 void spapr_caps_init(SpaprMachineState *spapr)
 778 {
 779     SpaprCapabilities default_caps;
 780     int i;
 781
 782     /* Compute the actual set of caps we should run with */
 783     default_caps = default_caps_with_cpu(spapr, MACHINE(spapr)->cpu_type);
 784
 785     for (i = 0; i < SPAPR_CAP_NUM; i++) {
 786         /* Store the defaults */
 787         spapr->def.caps[i] = default_caps.caps[i];
 788         /* If not set on the command line then apply the default value */
 789         if (!spapr->cmd_line_caps[i]) {
 790             spapr->eff.caps[i] = default_caps.caps[i];
 791         }
 792     }
 793 }
 794
 795 void spapr_caps_apply(SpaprMachineState *spapr)
 796 {
 797     int i;
 798
 799     for (i = 0; i < SPAPR_CAP_NUM; i++) {
 800         SpaprCapabilityInfo *info = &capability_table[i];
 801
 802         /*
 803          * If the apply function can't set the desired level and thinks it's
 804          * fatal, it should cause that.
 805          */
 806         info->apply(spapr, spapr->eff.caps[i], &error_fatal);
 807     }
 808 }
 809
 810 void spapr_caps_cpu_apply(SpaprMachineState *spapr, PowerPCCPU *cpu)
 811 {
 812     int i;
 813
 814     for (i = 0; i < SPAPR_CAP_NUM; i++) {
 815         SpaprCapabilityInfo *info = &capability_table[i];
 816
 817         /*
 818          * If the apply function can't set the desired level and thinks it's
 819          * fatal, it should cause that.
 820          */
 821         if (info->cpu_apply) {
 822             info->cpu_apply(spapr, cpu, spapr->eff.caps[i], &error_fatal);
 823         }
 824     }
 825 }
 826
 827 void spapr_caps_add_properties(SpaprMachineClass *smc, Error **errp)
 828 {
 829     Error *local_err = NULL;
 830     ObjectClass *klass = OBJECT_CLASS(smc);
 831     int i;
 832
 833     for (i = 0; i < ARRAY_SIZE(capability_table); i++) {
 834         SpaprCapabilityInfo *cap = &capability_table[i];
 835         char *name = g_strdup_printf("cap-%s", cap->name);
 836         char *desc;
 837
 838         object_class_property_add(klass, name, cap->type,
 839                                   cap->get, cap->set,
 840                                   NULL, cap, &local_err);
 841         if (local_err) {
 842             error_propagate(errp, local_err);
 843             g_free(name);
 844             return;
 845         }
 846
 847         desc = g_strdup_printf("%s", cap->description);
 848         object_class_property_set_description(klass, name, desc, &local_err);
 849         g_free(name);
 850         g_free(desc);
 851         if (local_err) {
 852             error_propagate(errp, local_err);
 853             return;
 854         }
 855     }
 856 }