arch/arm64/kvm/sys_regs.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) 2012,2013 - ARM Ltd
   4  * Author: Marc Zyngier <marc.zyngier@arm.com>
   5  *
   6  * Derived from arch/arm/kvm/coproc.c:
   7  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
   8  * Authors: Rusty Russell <rusty@rustcorp.com.au>
   9  *          Christoffer Dall <c.dall@virtualopensystems.com>
  10  */
  11
  12 #include <linux/bitfield.h>
  13 #include <linux/bsearch.h>
  14 #include <linux/kvm_host.h>
  15 #include <linux/mm.h>
  16 #include <linux/printk.h>
  17 #include <linux/uaccess.h>
  18
  19 #include <asm/cacheflush.h>
  20 #include <asm/cputype.h>
  21 #include <asm/debug-monitors.h>
  22 #include <asm/esr.h>
  23 #include <asm/kvm_arm.h>
  24 #include <asm/kvm_emulate.h>
  25 #include <asm/kvm_hyp.h>
  26 #include <asm/kvm_mmu.h>
  27 #include <asm/perf_event.h>
  28 #include <asm/sysreg.h>
  29
  30 #include <trace/events/kvm.h>
  31
  32 #include "sys_regs.h"
  33
  34 #include "trace.h"
  35
  36 /*
  37  * All of this file is extremely similar to the ARM coproc.c, but the
  38  * types are different. My gut feeling is that it should be pretty
  39  * easy to merge, but that would be an ABI breakage -- again. VFP
  40  * would also need to be abstracted.
  41  *
  42  * For AArch32, we only take care of what is being trapped. Anything
  43  * that has to do with init and userspace access has to go via the
  44  * 64bit interface.
  45  */
  46
  47 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id);
  48 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id);
  49 static u64 sys_reg_to_index(const struct sys_reg_desc *reg);
  50
  51 static bool read_from_write_only(struct kvm_vcpu *vcpu,
  52                                  struct sys_reg_params *params,
  53                                  const struct sys_reg_desc *r)
  54 {
  55         WARN_ONCE(1, "Unexpected sys_reg read to write-only register\n");
  56         print_sys_reg_instr(params);
  57         kvm_inject_undefined(vcpu);
  58         return false;
  59 }
  60
  61 static bool write_to_read_only(struct kvm_vcpu *vcpu,
  62                                struct sys_reg_params *params,
  63                                const struct sys_reg_desc *r)
  64 {
  65         WARN_ONCE(1, "Unexpected sys_reg write to read-only register\n");
  66         print_sys_reg_instr(params);
  67         kvm_inject_undefined(vcpu);
  68         return false;
  69 }
  70
  71 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg)
  72 {
  73         u64 val = 0x8badf00d8badf00d;
  74
  75         if (vcpu->arch.sysregs_loaded_on_cpu &&
  76             __vcpu_read_sys_reg_from_cpu(reg, &val))
  77                 return val;
  78
  79         return __vcpu_sys_reg(vcpu, reg);
  80 }
  81
  82 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg)
  83 {
  84         if (vcpu->arch.sysregs_loaded_on_cpu &&
  85             __vcpu_write_sys_reg_to_cpu(val, reg))
  86                 return;
  87
  88          __vcpu_sys_reg(vcpu, reg) = val;
  89 }
  90
  91 /* 3 bits per cache level, as per CLIDR, but non-existent caches always 0 */
  92 static u32 cache_levels;
  93
  94 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
  95 #define CSSELR_MAX 14
  96
  97 /* Which cache CCSIDR represents depends on CSSELR value. */
  98 static u32 get_ccsidr(u32 csselr)
  99 {
 100         u32 ccsidr;
 101
 102         /* Make sure noone else changes CSSELR during this! */
 103         local_irq_disable();
 104         write_sysreg(csselr, csselr_el1);
 105         isb();
 106         ccsidr = read_sysreg(ccsidr_el1);
 107         local_irq_enable();
 108
 109         return ccsidr;
 110 }
 111
 112 /*
 113  * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
 114  */
 115 static bool access_dcsw(struct kvm_vcpu *vcpu,
 116                         struct sys_reg_params *p,
 117                         const struct sys_reg_desc *r)
 118 {
 119         if (!p->is_write)
 120                 return read_from_write_only(vcpu, p, r);
 121
 122         /*
 123          * Only track S/W ops if we don't have FWB. It still indicates
 124          * that the guest is a bit broken (S/W operations should only
 125          * be done by firmware, knowing that there is only a single
 126          * CPU left in the system, and certainly not from non-secure
 127          * software).
 128          */
 129         if (!cpus_have_const_cap(ARM64_HAS_STAGE2_FWB))
 130                 kvm_set_way_flush(vcpu);
 131
 132         return true;
 133 }
 134
 135 static void get_access_mask(const struct sys_reg_desc *r, u64 *mask, u64 *shift)
 136 {
 137         switch (r->aarch32_map) {
 138         case AA32_LO:
 139                 *mask = GENMASK_ULL(31, 0);
 140                 *shift = 0;
 141                 break;
 142         case AA32_HI:
 143                 *mask = GENMASK_ULL(63, 32);
 144                 *shift = 32;
 145                 break;
 146         default:
 147                 *mask = GENMASK_ULL(63, 0);
 148                 *shift = 0;
 149                 break;
 150         }
 151 }
 152
 153 /*
 154  * Generic accessor for VM registers. Only called as long as HCR_TVM
 155  * is set. If the guest enables the MMU, we stop trapping the VM
 156  * sys_regs and leave it in complete control of the caches.
 157  */
 158 static bool access_vm_reg(struct kvm_vcpu *vcpu,
 159                           struct sys_reg_params *p,
 160                           const struct sys_reg_desc *r)
 161 {
 162         bool was_enabled = vcpu_has_cache_enabled(vcpu);
 163         u64 val, mask, shift;
 164
 165         BUG_ON(!p->is_write);
 166
 167         get_access_mask(r, &mask, &shift);
 168
 169         if (~mask) {
 170                 val = vcpu_read_sys_reg(vcpu, r->reg);
 171                 val &= ~mask;
 172         } else {
 173                 val = 0;
 174         }
 175
 176         val |= (p->regval & (mask >> shift)) << shift;
 177         vcpu_write_sys_reg(vcpu, val, r->reg);
 178
 179         kvm_toggle_cache(vcpu, was_enabled);
 180         return true;
 181 }
 182
 183 static bool access_actlr(struct kvm_vcpu *vcpu,
 184                          struct sys_reg_params *p,
 185                          const struct sys_reg_desc *r)
 186 {
 187         u64 mask, shift;
 188
 189         if (p->is_write)
 190                 return ignore_write(vcpu, p);
 191
 192         get_access_mask(r, &mask, &shift);
 193         p->regval = (vcpu_read_sys_reg(vcpu, r->reg) & mask) >> shift;
 194
 195         return true;
 196 }
 197
 198 /*
 199  * Trap handler for the GICv3 SGI generation system register.
 200  * Forward the request to the VGIC emulation.
 201  * The cp15_64 code makes sure this automatically works
 202  * for both AArch64 and AArch32 accesses.
 203  */
 204 static bool access_gic_sgi(struct kvm_vcpu *vcpu,
 205                            struct sys_reg_params *p,
 206                            const struct sys_reg_desc *r)
 207 {
 208         bool g1;
 209
 210         if (!p->is_write)
 211                 return read_from_write_only(vcpu, p, r);
 212
 213         /*
 214          * In a system where GICD_CTLR.DS=1, a ICC_SGI0R_EL1 access generates
 215          * Group0 SGIs only, while ICC_SGI1R_EL1 can generate either group,
 216          * depending on the SGI configuration. ICC_ASGI1R_EL1 is effectively
 217          * equivalent to ICC_SGI0R_EL1, as there is no "alternative" secure
 218          * group.
 219          */
 220         if (p->Op0 == 0) {              /* AArch32 */
 221                 switch (p->Op1) {
 222                 default:                /* Keep GCC quiet */
 223                 case 0:                 /* ICC_SGI1R */
 224                         g1 = true;
 225                         break;
 226                 case 1:                 /* ICC_ASGI1R */
 227                 case 2:                 /* ICC_SGI0R */
 228                         g1 = false;
 229                         break;
 230                 }
 231         } else {                        /* AArch64 */
 232                 switch (p->Op2) {
 233                 default:                /* Keep GCC quiet */
 234                 case 5:                 /* ICC_SGI1R_EL1 */
 235                         g1 = true;
 236                         break;
 237                 case 6:                 /* ICC_ASGI1R_EL1 */
 238                 case 7:                 /* ICC_SGI0R_EL1 */
 239                         g1 = false;
 240                         break;
 241                 }
 242         }
 243
 244         vgic_v3_dispatch_sgi(vcpu, p->regval, g1);
 245
 246         return true;
 247 }
 248
 249 static bool access_gic_sre(struct kvm_vcpu *vcpu,
 250                            struct sys_reg_params *p,
 251                            const struct sys_reg_desc *r)
 252 {
 253         if (p->is_write)
 254                 return ignore_write(vcpu, p);
 255
 256         p->regval = vcpu->arch.vgic_cpu.vgic_v3.vgic_sre;
 257         return true;
 258 }
 259
 260 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
 261                         struct sys_reg_params *p,
 262                         const struct sys_reg_desc *r)
 263 {
 264         if (p->is_write)
 265                 return ignore_write(vcpu, p);
 266         else
 267                 return read_zero(vcpu, p);
 268 }
 269
 270 /*
 271  * ARMv8.1 mandates at least a trivial LORegion implementation, where all the
 272  * RW registers are RES0 (which we can implement as RAZ/WI). On an ARMv8.0
 273  * system, these registers should UNDEF. LORID_EL1 being a RO register, we
 274  * treat it separately.
 275  */
 276 static bool trap_loregion(struct kvm_vcpu *vcpu,
 277                           struct sys_reg_params *p,
 278                           const struct sys_reg_desc *r)
 279 {
 280         u64 val = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
 281         u32 sr = reg_to_encoding(r);
 282
 283         if (!(val & (0xfUL << ID_AA64MMFR1_LOR_SHIFT))) {
 284                 kvm_inject_undefined(vcpu);
 285                 return false;
 286         }
 287
 288         if (p->is_write && sr == SYS_LORID_EL1)
 289                 return write_to_read_only(vcpu, p, r);
 290
 291         return trap_raz_wi(vcpu, p, r);
 292 }
 293
 294 static bool trap_oslar_el1(struct kvm_vcpu *vcpu,
 295                            struct sys_reg_params *p,
 296                            const struct sys_reg_desc *r)
 297 {
 298         u64 oslsr;
 299
 300         if (!p->is_write)
 301                 return read_from_write_only(vcpu, p, r);
 302
 303         /* Forward the OSLK bit to OSLSR */
 304         oslsr = __vcpu_sys_reg(vcpu, OSLSR_EL1) & ~SYS_OSLSR_OSLK;
 305         if (p->regval & SYS_OSLAR_OSLK)
 306                 oslsr |= SYS_OSLSR_OSLK;
 307
 308         __vcpu_sys_reg(vcpu, OSLSR_EL1) = oslsr;
 309         return true;
 310 }
 311
 312 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
 313                            struct sys_reg_params *p,
 314                            const struct sys_reg_desc *r)
 315 {
 316         if (p->is_write)
 317                 return write_to_read_only(vcpu, p, r);
 318
 319         p->regval = __vcpu_sys_reg(vcpu, r->reg);
 320         return true;
 321 }
 322
 323 static int set_oslsr_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
 324                          const struct kvm_one_reg *reg, void __user *uaddr)
 325 {
 326         u64 id = sys_reg_to_index(rd);
 327         u64 val;
 328         int err;
 329
 330         err = reg_from_user(&val, uaddr, id);
 331         if (err)
 332                 return err;
 333
 334         /*
 335          * The only modifiable bit is the OSLK bit. Refuse the write if
 336          * userspace attempts to change any other bit in the register.
 337          */
 338         if ((val ^ rd->val) & ~SYS_OSLSR_OSLK)
 339                 return -EINVAL;
 340
 341         __vcpu_sys_reg(vcpu, rd->reg) = val;
 342         return 0;
 343 }
 344
 345 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
 346                                    struct sys_reg_params *p,
 347                                    const struct sys_reg_desc *r)
 348 {
 349         if (p->is_write) {
 350                 return ignore_write(vcpu, p);
 351         } else {
 352                 p->regval = read_sysreg(dbgauthstatus_el1);
 353                 return true;
 354         }
 355 }
 356
 357 /*
 358  * We want to avoid world-switching all the DBG registers all the
 359  * time:
 360  *
 361  * - If we've touched any debug register, it is likely that we're
 362  *   going to touch more of them. It then makes sense to disable the
 363  *   traps and start doing the save/restore dance
 364  * - If debug is active (DBG_MDSCR_KDE or DBG_MDSCR_MDE set), it is
 365  *   then mandatory to save/restore the registers, as the guest
 366  *   depends on them.
 367  *
 368  * For this, we use a DIRTY bit, indicating the guest has modified the
 369  * debug registers, used as follow:
 370  *
 371  * On guest entry:
 372  * - If the dirty bit is set (because we're coming back from trapping),
 373  *   disable the traps, save host registers, restore guest registers.
 374  * - If debug is actively in use (DBG_MDSCR_KDE or DBG_MDSCR_MDE set),
 375  *   set the dirty bit, disable the traps, save host registers,
 376  *   restore guest registers.
 377  * - Otherwise, enable the traps
 378  *
 379  * On guest exit:
 380  * - If the dirty bit is set, save guest registers, restore host
 381  *   registers and clear the dirty bit. This ensure that the host can
 382  *   now use the debug registers.
 383  */
 384 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
 385                             struct sys_reg_params *p,
 386                             const struct sys_reg_desc *r)
 387 {
 388         if (p->is_write) {
 389                 vcpu_write_sys_reg(vcpu, p->regval, r->reg);
 390                 vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
 391         } else {
 392                 p->regval = vcpu_read_sys_reg(vcpu, r->reg);
 393         }
 394
 395         trace_trap_reg(__func__, r->reg, p->is_write, p->regval);
 396
 397         return true;
 398 }
 399
 400 /*
 401  * reg_to_dbg/dbg_to_reg
 402  *
 403  * A 32 bit write to a debug register leave top bits alone
 404  * A 32 bit read from a debug register only returns the bottom bits
 405  *
 406  * All writes will set the KVM_ARM64_DEBUG_DIRTY flag to ensure the
 407  * hyp.S code switches between host and guest values in future.
 408  */
 409 static void reg_to_dbg(struct kvm_vcpu *vcpu,
 410                        struct sys_reg_params *p,
 411                        const struct sys_reg_desc *rd,
 412                        u64 *dbg_reg)
 413 {
 414         u64 mask, shift, val;
 415
 416         get_access_mask(rd, &mask, &shift);
 417
 418         val = *dbg_reg;
 419         val &= ~mask;
 420         val |= (p->regval & (mask >> shift)) << shift;
 421         *dbg_reg = val;
 422
 423         vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
 424 }
 425
 426 static void dbg_to_reg(struct kvm_vcpu *vcpu,
 427                        struct sys_reg_params *p,
 428                        const struct sys_reg_desc *rd,
 429                        u64 *dbg_reg)
 430 {
 431         u64 mask, shift;
 432
 433         get_access_mask(rd, &mask, &shift);
 434         p->regval = (*dbg_reg & mask) >> shift;
 435 }
 436
 437 static bool trap_bvr(struct kvm_vcpu *vcpu,
 438                      struct sys_reg_params *p,
 439                      const struct sys_reg_desc *rd)
 440 {
 441         u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm];
 442
 443         if (p->is_write)
 444                 reg_to_dbg(vcpu, p, rd, dbg_reg);
 445         else
 446                 dbg_to_reg(vcpu, p, rd, dbg_reg);
 447
 448         trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
 449
 450         return true;
 451 }
 452
 453 static int set_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
 454                 const struct kvm_one_reg *reg, void __user *uaddr)
 455 {
 456         __u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm];
 457
 458         if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
 459                 return -EFAULT;
 460         return 0;
 461 }
 462
 463 static int get_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
 464         const struct kvm_one_reg *reg, void __user *uaddr)
 465 {
 466         __u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm];
 467
 468         if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
 469                 return -EFAULT;
 470         return 0;
 471 }
 472
 473 static void reset_bvr(struct kvm_vcpu *vcpu,
 474                       const struct sys_reg_desc *rd)
 475 {
 476         vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm] = rd->val;
 477 }
 478
 479 static bool trap_bcr(struct kvm_vcpu *vcpu,
 480                      struct sys_reg_params *p,
 481                      const struct sys_reg_desc *rd)
 482 {
 483         u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm];
 484
 485         if (p->is_write)
 486                 reg_to_dbg(vcpu, p, rd, dbg_reg);
 487         else
 488                 dbg_to_reg(vcpu, p, rd, dbg_reg);
 489
 490         trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
 491
 492         return true;
 493 }
 494
 495 static int set_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
 496                 const struct kvm_one_reg *reg, void __user *uaddr)
 497 {
 498         __u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm];
 499
 500         if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
 501                 return -EFAULT;
 502
 503         return 0;
 504 }
 505
 506 static int get_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
 507         const struct kvm_one_reg *reg, void __user *uaddr)
 508 {
 509         __u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm];
 510
 511         if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
 512                 return -EFAULT;
 513         return 0;
 514 }
 515
 516 static void reset_bcr(struct kvm_vcpu *vcpu,
 517                       const struct sys_reg_desc *rd)
 518 {
 519         vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm] = rd->val;
 520 }
 521
 522 static bool trap_wvr(struct kvm_vcpu *vcpu,
 523                      struct sys_reg_params *p,
 524                      const struct sys_reg_desc *rd)
 525 {
 526         u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm];
 527
 528         if (p->is_write)
 529                 reg_to_dbg(vcpu, p, rd, dbg_reg);
 530         else
 531                 dbg_to_reg(vcpu, p, rd, dbg_reg);
 532
 533         trace_trap_reg(__func__, rd->CRm, p->is_write,
 534                 vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm]);
 535
 536         return true;
 537 }
 538
 539 static int set_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
 540                 const struct kvm_one_reg *reg, void __user *uaddr)
 541 {
 542         __u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm];
 543
 544         if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
 545                 return -EFAULT;
 546         return 0;
 547 }
 548
 549 static int get_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
 550         const struct kvm_one_reg *reg, void __user *uaddr)
 551 {
 552         __u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm];
 553
 554         if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
 555                 return -EFAULT;
 556         return 0;
 557 }
 558
 559 static void reset_wvr(struct kvm_vcpu *vcpu,
 560                       const struct sys_reg_desc *rd)
 561 {
 562         vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm] = rd->val;
 563 }
 564
 565 static bool trap_wcr(struct kvm_vcpu *vcpu,
 566                      struct sys_reg_params *p,
 567                      const struct sys_reg_desc *rd)
 568 {
 569         u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm];
 570
 571         if (p->is_write)
 572                 reg_to_dbg(vcpu, p, rd, dbg_reg);
 573         else
 574                 dbg_to_reg(vcpu, p, rd, dbg_reg);
 575
 576         trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
 577
 578         return true;
 579 }
 580
 581 static int set_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
 582                 const struct kvm_one_reg *reg, void __user *uaddr)
 583 {
 584         __u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm];
 585
 586         if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
 587                 return -EFAULT;
 588         return 0;
 589 }
 590
 591 static int get_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
 592         const struct kvm_one_reg *reg, void __user *uaddr)
 593 {
 594         __u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm];
 595
 596         if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
 597                 return -EFAULT;
 598         return 0;
 599 }
 600
 601 static void reset_wcr(struct kvm_vcpu *vcpu,
 602                       const struct sys_reg_desc *rd)
 603 {
 604         vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm] = rd->val;
 605 }
 606
 607 static void reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
 608 {
 609         u64 amair = read_sysreg(amair_el1);
 610         vcpu_write_sys_reg(vcpu, amair, AMAIR_EL1);
 611 }
 612
 613 static void reset_actlr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
 614 {
 615         u64 actlr = read_sysreg(actlr_el1);
 616         vcpu_write_sys_reg(vcpu, actlr, ACTLR_EL1);
 617 }
 618
 619 static void reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
 620 {
 621         u64 mpidr;
 622
 623         /*
 624          * Map the vcpu_id into the first three affinity level fields of
 625          * the MPIDR. We limit the number of VCPUs in level 0 due to a
 626          * limitation to 16 CPUs in that level in the ICC_SGIxR registers
 627          * of the GICv3 to be able to address each CPU directly when
 628          * sending IPIs.
 629          */
 630         mpidr = (vcpu->vcpu_id & 0x0f) << MPIDR_LEVEL_SHIFT(0);
 631         mpidr |= ((vcpu->vcpu_id >> 4) & 0xff) << MPIDR_LEVEL_SHIFT(1);
 632         mpidr |= ((vcpu->vcpu_id >> 12) & 0xff) << MPIDR_LEVEL_SHIFT(2);
 633         vcpu_write_sys_reg(vcpu, (1ULL << 31) | mpidr, MPIDR_EL1);
 634 }
 635
 636 static unsigned int pmu_visibility(const struct kvm_vcpu *vcpu,
 637                                    const struct sys_reg_desc *r)
 638 {
 639         if (kvm_vcpu_has_pmu(vcpu))
 640                 return 0;
 641
 642         return REG_HIDDEN;
 643 }
 644
 645 static void reset_pmu_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
 646 {
 647         u64 n, mask = BIT(ARMV8_PMU_CYCLE_IDX);
 648
 649         /* No PMU available, any PMU reg may UNDEF... */
 650         if (!kvm_arm_support_pmu_v3())
 651                 return;
 652
 653         n = read_sysreg(pmcr_el0) >> ARMV8_PMU_PMCR_N_SHIFT;
 654         n &= ARMV8_PMU_PMCR_N_MASK;
 655         if (n)
 656                 mask |= GENMASK(n - 1, 0);
 657
 658         reset_unknown(vcpu, r);
 659         __vcpu_sys_reg(vcpu, r->reg) &= mask;
 660 }
 661
 662 static void reset_pmevcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
 663 {
 664         reset_unknown(vcpu, r);
 665         __vcpu_sys_reg(vcpu, r->reg) &= GENMASK(31, 0);
 666 }
 667
 668 static void reset_pmevtyper(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
 669 {
 670         reset_unknown(vcpu, r);
 671         __vcpu_sys_reg(vcpu, r->reg) &= ARMV8_PMU_EVTYPE_MASK;
 672 }
 673
 674 static void reset_pmselr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
 675 {
 676         reset_unknown(vcpu, r);
 677         __vcpu_sys_reg(vcpu, r->reg) &= ARMV8_PMU_COUNTER_MASK;
 678 }
 679
 680 static void reset_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
 681 {
 682         u64 pmcr, val;
 683
 684         /* No PMU available, PMCR_EL0 may UNDEF... */
 685         if (!kvm_arm_support_pmu_v3())
 686                 return;
 687
 688         pmcr = read_sysreg(pmcr_el0);
 689         /*
 690          * Writable bits of PMCR_EL0 (ARMV8_PMU_PMCR_MASK) are reset to UNKNOWN
 691          * except PMCR.E resetting to zero.
 692          */
 693         val = ((pmcr & ~ARMV8_PMU_PMCR_MASK)
 694                | (ARMV8_PMU_PMCR_MASK & 0xdecafbad)) & (~ARMV8_PMU_PMCR_E);
 695         if (!system_supports_32bit_el0())
 696                 val |= ARMV8_PMU_PMCR_LC;
 697         __vcpu_sys_reg(vcpu, r->reg) = val;
 698 }
 699
 700 static bool check_pmu_access_disabled(struct kvm_vcpu *vcpu, u64 flags)
 701 {
 702         u64 reg = __vcpu_sys_reg(vcpu, PMUSERENR_EL0);
 703         bool enabled = (reg & flags) || vcpu_mode_priv(vcpu);
 704
 705         if (!enabled)
 706                 kvm_inject_undefined(vcpu);
 707
 708         return !enabled;
 709 }
 710
 711 static bool pmu_access_el0_disabled(struct kvm_vcpu *vcpu)
 712 {
 713         return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_EN);
 714 }
 715
 716 static bool pmu_write_swinc_el0_disabled(struct kvm_vcpu *vcpu)
 717 {
 718         return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_SW | ARMV8_PMU_USERENR_EN);
 719 }
 720
 721 static bool pmu_access_cycle_counter_el0_disabled(struct kvm_vcpu *vcpu)
 722 {
 723         return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_CR | ARMV8_PMU_USERENR_EN);
 724 }
 725
 726 static bool pmu_access_event_counter_el0_disabled(struct kvm_vcpu *vcpu)
 727 {
 728         return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_ER | ARMV8_PMU_USERENR_EN);
 729 }
 730
 731 static bool access_pmcr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
 732                         const struct sys_reg_desc *r)
 733 {
 734         u64 val;
 735
 736         if (pmu_access_el0_disabled(vcpu))
 737                 return false;
 738
 739         if (p->is_write) {
 740                 /* Only update writeable bits of PMCR */
 741                 val = __vcpu_sys_reg(vcpu, PMCR_EL0);
 742                 val &= ~ARMV8_PMU_PMCR_MASK;
 743                 val |= p->regval & ARMV8_PMU_PMCR_MASK;
 744                 if (!system_supports_32bit_el0())
 745                         val |= ARMV8_PMU_PMCR_LC;
 746                 __vcpu_sys_reg(vcpu, PMCR_EL0) = val;
 747                 kvm_pmu_handle_pmcr(vcpu, val);
 748                 kvm_vcpu_pmu_restore_guest(vcpu);
 749         } else {
 750                 /* PMCR.P & PMCR.C are RAZ */
 751                 val = __vcpu_sys_reg(vcpu, PMCR_EL0)
 752                       & ~(ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C);
 753                 p->regval = val;
 754         }
 755
 756         return true;
 757 }
 758
 759 static bool access_pmselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
 760                           const struct sys_reg_desc *r)
 761 {
 762         if (pmu_access_event_counter_el0_disabled(vcpu))
 763                 return false;
 764
 765         if (p->is_write)
 766                 __vcpu_sys_reg(vcpu, PMSELR_EL0) = p->regval;
 767         else
 768                 /* return PMSELR.SEL field */
 769                 p->regval = __vcpu_sys_reg(vcpu, PMSELR_EL0)
 770                             & ARMV8_PMU_COUNTER_MASK;
 771
 772         return true;
 773 }
 774
 775 static bool access_pmceid(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
 776                           const struct sys_reg_desc *r)
 777 {
 778         u64 pmceid, mask, shift;
 779
 780         BUG_ON(p->is_write);
 781
 782         if (pmu_access_el0_disabled(vcpu))
 783                 return false;
 784
 785         get_access_mask(r, &mask, &shift);
 786
 787         pmceid = kvm_pmu_get_pmceid(vcpu, (p->Op2 & 1));
 788         pmceid &= mask;
 789         pmceid >>= shift;
 790
 791         p->regval = pmceid;
 792
 793         return true;
 794 }
 795
 796 static bool pmu_counter_idx_valid(struct kvm_vcpu *vcpu, u64 idx)
 797 {
 798         u64 pmcr, val;
 799
 800         pmcr = __vcpu_sys_reg(vcpu, PMCR_EL0);
 801         val = (pmcr >> ARMV8_PMU_PMCR_N_SHIFT) & ARMV8_PMU_PMCR_N_MASK;
 802         if (idx >= val && idx != ARMV8_PMU_CYCLE_IDX) {
 803                 kvm_inject_undefined(vcpu);
 804                 return false;
 805         }
 806
 807         return true;
 808 }
 809
 810 static bool access_pmu_evcntr(struct kvm_vcpu *vcpu,
 811                               struct sys_reg_params *p,
 812                               const struct sys_reg_desc *r)
 813 {
 814         u64 idx = ~0UL;
 815
 816         if (r->CRn == 9 && r->CRm == 13) {
 817                 if (r->Op2 == 2) {
 818                         /* PMXEVCNTR_EL0 */
 819                         if (pmu_access_event_counter_el0_disabled(vcpu))
 820                                 return false;
 821
 822                         idx = __vcpu_sys_reg(vcpu, PMSELR_EL0)
 823                               & ARMV8_PMU_COUNTER_MASK;
 824                 } else if (r->Op2 == 0) {
 825                         /* PMCCNTR_EL0 */
 826                         if (pmu_access_cycle_counter_el0_disabled(vcpu))
 827                                 return false;
 828
 829                         idx = ARMV8_PMU_CYCLE_IDX;
 830                 }
 831         } else if (r->CRn == 0 && r->CRm == 9) {
 832                 /* PMCCNTR */
 833                 if (pmu_access_event_counter_el0_disabled(vcpu))
 834                         return false;
 835
 836                 idx = ARMV8_PMU_CYCLE_IDX;
 837         } else if (r->CRn == 14 && (r->CRm & 12) == 8) {
 838                 /* PMEVCNTRn_EL0 */
 839                 if (pmu_access_event_counter_el0_disabled(vcpu))
 840                         return false;
 841
 842                 idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
 843         }
 844
 845         /* Catch any decoding mistake */
 846         WARN_ON(idx == ~0UL);
 847
 848         if (!pmu_counter_idx_valid(vcpu, idx))
 849                 return false;
 850
 851         if (p->is_write) {
 852                 if (pmu_access_el0_disabled(vcpu))
 853                         return false;
 854
 855                 kvm_pmu_set_counter_value(vcpu, idx, p->regval);
 856         } else {
 857                 p->regval = kvm_pmu_get_counter_value(vcpu, idx);
 858         }
 859
 860         return true;
 861 }
 862
 863 static bool access_pmu_evtyper(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
 864                                const struct sys_reg_desc *r)
 865 {
 866         u64 idx, reg;
 867
 868         if (pmu_access_el0_disabled(vcpu))
 869                 return false;
 870
 871         if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 1) {
 872                 /* PMXEVTYPER_EL0 */
 873                 idx = __vcpu_sys_reg(vcpu, PMSELR_EL0) & ARMV8_PMU_COUNTER_MASK;
 874                 reg = PMEVTYPER0_EL0 + idx;
 875         } else if (r->CRn == 14 && (r->CRm & 12) == 12) {
 876                 idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
 877                 if (idx == ARMV8_PMU_CYCLE_IDX)
 878                         reg = PMCCFILTR_EL0;
 879                 else
 880                         /* PMEVTYPERn_EL0 */
 881                         reg = PMEVTYPER0_EL0 + idx;
 882         } else {
 883                 BUG();
 884         }
 885
 886         if (!pmu_counter_idx_valid(vcpu, idx))
 887                 return false;
 888
 889         if (p->is_write) {
 890                 kvm_pmu_set_counter_event_type(vcpu, p->regval, idx);
 891                 __vcpu_sys_reg(vcpu, reg) = p->regval & ARMV8_PMU_EVTYPE_MASK;
 892                 kvm_vcpu_pmu_restore_guest(vcpu);
 893         } else {
 894                 p->regval = __vcpu_sys_reg(vcpu, reg) & ARMV8_PMU_EVTYPE_MASK;
 895         }
 896
 897         return true;
 898 }
 899
 900 static bool access_pmcnten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
 901                            const struct sys_reg_desc *r)
 902 {
 903         u64 val, mask;
 904
 905         if (pmu_access_el0_disabled(vcpu))
 906                 return false;
 907
 908         mask = kvm_pmu_valid_counter_mask(vcpu);
 909         if (p->is_write) {
 910                 val = p->regval & mask;
 911                 if (r->Op2 & 0x1) {
 912                         /* accessing PMCNTENSET_EL0 */
 913                         __vcpu_sys_reg(vcpu, PMCNTENSET_EL0) |= val;
 914                         kvm_pmu_enable_counter_mask(vcpu, val);
 915                         kvm_vcpu_pmu_restore_guest(vcpu);
 916                 } else {
 917                         /* accessing PMCNTENCLR_EL0 */
 918                         __vcpu_sys_reg(vcpu, PMCNTENSET_EL0) &= ~val;
 919                         kvm_pmu_disable_counter_mask(vcpu, val);
 920                 }
 921         } else {
 922                 p->regval = __vcpu_sys_reg(vcpu, PMCNTENSET_EL0);
 923         }
 924
 925         return true;
 926 }
 927
 928 static bool access_pminten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
 929                            const struct sys_reg_desc *r)
 930 {
 931         u64 mask = kvm_pmu_valid_counter_mask(vcpu);
 932
 933         if (check_pmu_access_disabled(vcpu, 0))
 934                 return false;
 935
 936         if (p->is_write) {
 937                 u64 val = p->regval & mask;
 938
 939                 if (r->Op2 & 0x1)
 940                         /* accessing PMINTENSET_EL1 */
 941                         __vcpu_sys_reg(vcpu, PMINTENSET_EL1) |= val;
 942                 else
 943                         /* accessing PMINTENCLR_EL1 */
 944                         __vcpu_sys_reg(vcpu, PMINTENSET_EL1) &= ~val;
 945         } else {
 946                 p->regval = __vcpu_sys_reg(vcpu, PMINTENSET_EL1);
 947         }
 948
 949         return true;
 950 }
 951
 952 static bool access_pmovs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
 953                          const struct sys_reg_desc *r)
 954 {
 955         u64 mask = kvm_pmu_valid_counter_mask(vcpu);
 956
 957         if (pmu_access_el0_disabled(vcpu))
 958                 return false;
 959
 960         if (p->is_write) {
 961                 if (r->CRm & 0x2)
 962                         /* accessing PMOVSSET_EL0 */
 963                         __vcpu_sys_reg(vcpu, PMOVSSET_EL0) |= (p->regval & mask);
 964                 else
 965                         /* accessing PMOVSCLR_EL0 */
 966                         __vcpu_sys_reg(vcpu, PMOVSSET_EL0) &= ~(p->regval & mask);
 967         } else {
 968                 p->regval = __vcpu_sys_reg(vcpu, PMOVSSET_EL0);
 969         }
 970
 971         return true;
 972 }
 973
 974 static bool access_pmswinc(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
 975                            const struct sys_reg_desc *r)
 976 {
 977         u64 mask;
 978
 979         if (!p->is_write)
 980                 return read_from_write_only(vcpu, p, r);
 981
 982         if (pmu_write_swinc_el0_disabled(vcpu))
 983                 return false;
 984
 985         mask = kvm_pmu_valid_counter_mask(vcpu);
 986         kvm_pmu_software_increment(vcpu, p->regval & mask);
 987         return true;
 988 }
 989
 990 static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
 991                              const struct sys_reg_desc *r)
 992 {
 993         if (p->is_write) {
 994                 if (!vcpu_mode_priv(vcpu)) {
 995                         kvm_inject_undefined(vcpu);
 996                         return false;
 997                 }
 998
 999                 __vcpu_sys_reg(vcpu, PMUSERENR_EL0) =
1000                                p->regval & ARMV8_PMU_USERENR_MASK;
1001         } else {
1002                 p->regval = __vcpu_sys_reg(vcpu, PMUSERENR_EL0)
1003                             & ARMV8_PMU_USERENR_MASK;
1004         }
1005
1006         return true;
1007 }
1008
1009 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
1010 #define DBG_BCR_BVR_WCR_WVR_EL1(n)                                      \
1011         { SYS_DESC(SYS_DBGBVRn_EL1(n)),                                 \
1012           trap_bvr, reset_bvr, 0, 0, get_bvr, set_bvr },                \
1013         { SYS_DESC(SYS_DBGBCRn_EL1(n)),                                 \
1014           trap_bcr, reset_bcr, 0, 0, get_bcr, set_bcr },                \
1015         { SYS_DESC(SYS_DBGWVRn_EL1(n)),                                 \
1016           trap_wvr, reset_wvr, 0, 0,  get_wvr, set_wvr },               \
1017         { SYS_DESC(SYS_DBGWCRn_EL1(n)),                                 \
1018           trap_wcr, reset_wcr, 0, 0,  get_wcr, set_wcr }
1019
1020 #define PMU_SYS_REG(r)                                          \
1021         SYS_DESC(r), .reset = reset_pmu_reg, .visibility = pmu_visibility
1022
1023 /* Macro to expand the PMEVCNTRn_EL0 register */
1024 #define PMU_PMEVCNTR_EL0(n)                                             \
1025         { PMU_SYS_REG(SYS_PMEVCNTRn_EL0(n)),                            \
1026           .reset = reset_pmevcntr,                                      \
1027           .access = access_pmu_evcntr, .reg = (PMEVCNTR0_EL0 + n), }
1028
1029 /* Macro to expand the PMEVTYPERn_EL0 register */
1030 #define PMU_PMEVTYPER_EL0(n)                                            \
1031         { PMU_SYS_REG(SYS_PMEVTYPERn_EL0(n)),                           \
1032           .reset = reset_pmevtyper,                                     \
1033           .access = access_pmu_evtyper, .reg = (PMEVTYPER0_EL0 + n), }
1034
1035 static bool undef_access(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1036                          const struct sys_reg_desc *r)
1037 {
1038         kvm_inject_undefined(vcpu);
1039
1040         return false;
1041 }
1042
1043 /* Macro to expand the AMU counter and type registers*/
1044 #define AMU_AMEVCNTR0_EL0(n) { SYS_DESC(SYS_AMEVCNTR0_EL0(n)), undef_access }
1045 #define AMU_AMEVTYPER0_EL0(n) { SYS_DESC(SYS_AMEVTYPER0_EL0(n)), undef_access }
1046 #define AMU_AMEVCNTR1_EL0(n) { SYS_DESC(SYS_AMEVCNTR1_EL0(n)), undef_access }
1047 #define AMU_AMEVTYPER1_EL0(n) { SYS_DESC(SYS_AMEVTYPER1_EL0(n)), undef_access }
1048
1049 static unsigned int ptrauth_visibility(const struct kvm_vcpu *vcpu,
1050                         const struct sys_reg_desc *rd)
1051 {
1052         return vcpu_has_ptrauth(vcpu) ? 0 : REG_HIDDEN;
1053 }
1054
1055 /*
1056  * If we land here on a PtrAuth access, that is because we didn't
1057  * fixup the access on exit by allowing the PtrAuth sysregs. The only
1058  * way this happens is when the guest does not have PtrAuth support
1059  * enabled.
1060  */
1061 #define __PTRAUTH_KEY(k)                                                \
1062         { SYS_DESC(SYS_## k), undef_access, reset_unknown, k,           \
1063         .visibility = ptrauth_visibility}
1064
1065 #define PTRAUTH_KEY(k)                                                  \
1066         __PTRAUTH_KEY(k ## KEYLO_EL1),                                  \
1067         __PTRAUTH_KEY(k ## KEYHI_EL1)
1068
1069 static bool access_arch_timer(struct kvm_vcpu *vcpu,
1070                               struct sys_reg_params *p,
1071                               const struct sys_reg_desc *r)
1072 {
1073         enum kvm_arch_timers tmr;
1074         enum kvm_arch_timer_regs treg;
1075         u64 reg = reg_to_encoding(r);
1076
1077         switch (reg) {
1078         case SYS_CNTP_TVAL_EL0:
1079         case SYS_AARCH32_CNTP_TVAL:
1080                 tmr = TIMER_PTIMER;
1081                 treg = TIMER_REG_TVAL;
1082                 break;
1083         case SYS_CNTP_CTL_EL0:
1084         case SYS_AARCH32_CNTP_CTL:
1085                 tmr = TIMER_PTIMER;
1086                 treg = TIMER_REG_CTL;
1087                 break;
1088         case SYS_CNTP_CVAL_EL0:
1089         case SYS_AARCH32_CNTP_CVAL:
1090                 tmr = TIMER_PTIMER;
1091                 treg = TIMER_REG_CVAL;
1092                 break;
1093         default:
1094                 BUG();
1095         }
1096
1097         if (p->is_write)
1098                 kvm_arm_timer_write_sysreg(vcpu, tmr, treg, p->regval);
1099         else
1100                 p->regval = kvm_arm_timer_read_sysreg(vcpu, tmr, treg);
1101
1102         return true;
1103 }
1104
1105 /* Read a sanitised cpufeature ID register by sys_reg_desc */
1106 static u64 read_id_reg(const struct kvm_vcpu *vcpu,
1107                 struct sys_reg_desc const *r, bool raz)
1108 {
1109         u32 id = reg_to_encoding(r);
1110         u64 val;
1111
1112         if (raz)
1113                 return 0;
1114
1115         val = read_sanitised_ftr_reg(id);
1116
1117         switch (id) {
1118         case SYS_ID_AA64PFR0_EL1:
1119                 if (!vcpu_has_sve(vcpu))
1120                         val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_SVE);
1121                 val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_AMU);
1122                 val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_CSV2);
1123                 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_CSV2), (u64)vcpu->kvm->arch.pfr0_csv2);
1124                 val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_CSV3);
1125                 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_CSV3), (u64)vcpu->kvm->arch.pfr0_csv3);
1126                 if (kvm_vgic_global_state.type == VGIC_V3) {
1127                         val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_GIC);
1128                         val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_GIC), 1);
1129                 }
1130                 break;
1131         case SYS_ID_AA64PFR1_EL1:
1132                 if (!kvm_has_mte(vcpu->kvm))
1133                         val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_MTE);
1134
1135                 val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_SME);
1136                 break;
1137         case SYS_ID_AA64ISAR1_EL1:
1138                 if (!vcpu_has_ptrauth(vcpu))
1139                         val &= ~(ARM64_FEATURE_MASK(ID_AA64ISAR1_APA) |
1140                                  ARM64_FEATURE_MASK(ID_AA64ISAR1_API) |
1141                                  ARM64_FEATURE_MASK(ID_AA64ISAR1_GPA) |
1142                                  ARM64_FEATURE_MASK(ID_AA64ISAR1_GPI));
1143                 break;
1144         case SYS_ID_AA64ISAR2_EL1:
1145                 if (!vcpu_has_ptrauth(vcpu))
1146                         val &= ~(ARM64_FEATURE_MASK(ID_AA64ISAR2_APA3) |
1147                                  ARM64_FEATURE_MASK(ID_AA64ISAR2_GPA3));
1148                 if (!cpus_have_final_cap(ARM64_HAS_WFXT))
1149                         val &= ~ARM64_FEATURE_MASK(ID_AA64ISAR2_WFXT);
1150                 break;
1151         case SYS_ID_AA64DFR0_EL1:
1152                 /* Limit debug to ARMv8.0 */
1153                 val &= ~ARM64_FEATURE_MASK(ID_AA64DFR0_DEBUGVER);
1154                 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64DFR0_DEBUGVER), 6);
1155                 /* Limit guests to PMUv3 for ARMv8.4 */
1156                 val = cpuid_feature_cap_perfmon_field(val,
1157                                                       ID_AA64DFR0_PMUVER_SHIFT,
1158                                                       kvm_vcpu_has_pmu(vcpu) ? ID_AA64DFR0_PMUVER_8_4 : 0);
1159                 /* Hide SPE from guests */
1160                 val &= ~ARM64_FEATURE_MASK(ID_AA64DFR0_PMSVER);
1161                 break;
1162         case SYS_ID_DFR0_EL1:
1163                 /* Limit guests to PMUv3 for ARMv8.4 */
1164                 val = cpuid_feature_cap_perfmon_field(val,
1165                                                       ID_DFR0_PERFMON_SHIFT,
1166                                                       kvm_vcpu_has_pmu(vcpu) ? ID_DFR0_PERFMON_8_4 : 0);
1167                 break;
1168         }
1169
1170         return val;
1171 }
1172
1173 static unsigned int id_visibility(const struct kvm_vcpu *vcpu,
1174                                   const struct sys_reg_desc *r)
1175 {
1176         u32 id = reg_to_encoding(r);
1177
1178         switch (id) {
1179         case SYS_ID_AA64ZFR0_EL1:
1180                 if (!vcpu_has_sve(vcpu))
1181                         return REG_RAZ;
1182                 break;
1183         }
1184
1185         return 0;
1186 }
1187
1188 /* cpufeature ID register access trap handlers */
1189
1190 static bool __access_id_reg(struct kvm_vcpu *vcpu,
1191                             struct sys_reg_params *p,
1192                             const struct sys_reg_desc *r,
1193                             bool raz)
1194 {
1195         if (p->is_write)
1196                 return write_to_read_only(vcpu, p, r);
1197
1198         p->regval = read_id_reg(vcpu, r, raz);
1199         return true;
1200 }
1201
1202 static bool access_id_reg(struct kvm_vcpu *vcpu,
1203                           struct sys_reg_params *p,
1204                           const struct sys_reg_desc *r)
1205 {
1206         bool raz = sysreg_visible_as_raz(vcpu, r);
1207
1208         return __access_id_reg(vcpu, p, r, raz);
1209 }
1210
1211 static bool access_raz_id_reg(struct kvm_vcpu *vcpu,
1212                               struct sys_reg_params *p,
1213                               const struct sys_reg_desc *r)
1214 {
1215         return __access_id_reg(vcpu, p, r, true);
1216 }
1217
1218 /* Visibility overrides for SVE-specific control registers */
1219 static unsigned int sve_visibility(const struct kvm_vcpu *vcpu,
1220                                    const struct sys_reg_desc *rd)
1221 {
1222         if (vcpu_has_sve(vcpu))
1223                 return 0;
1224
1225         return REG_HIDDEN;
1226 }
1227
1228 static int set_id_aa64pfr0_el1(struct kvm_vcpu *vcpu,
1229                                const struct sys_reg_desc *rd,
1230                                const struct kvm_one_reg *reg, void __user *uaddr)
1231 {
1232         const u64 id = sys_reg_to_index(rd);
1233         u8 csv2, csv3;
1234         int err;
1235         u64 val;
1236
1237         err = reg_from_user(&val, uaddr, id);
1238         if (err)
1239                 return err;
1240
1241         /*
1242          * Allow AA64PFR0_EL1.CSV2 to be set from userspace as long as
1243          * it doesn't promise more than what is actually provided (the
1244          * guest could otherwise be covered in ectoplasmic residue).
1245          */
1246         csv2 = cpuid_feature_extract_unsigned_field(val, ID_AA64PFR0_CSV2_SHIFT);
1247         if (csv2 > 1 ||
1248             (csv2 && arm64_get_spectre_v2_state() != SPECTRE_UNAFFECTED))
1249                 return -EINVAL;
1250
1251         /* Same thing for CSV3 */
1252         csv3 = cpuid_feature_extract_unsigned_field(val, ID_AA64PFR0_CSV3_SHIFT);
1253         if (csv3 > 1 ||
1254             (csv3 && arm64_get_meltdown_state() != SPECTRE_UNAFFECTED))
1255                 return -EINVAL;
1256
1257         /* We can only differ with CSV[23], and anything else is an error */
1258         val ^= read_id_reg(vcpu, rd, false);
1259         val &= ~((0xFUL << ID_AA64PFR0_CSV2_SHIFT) |
1260                  (0xFUL << ID_AA64PFR0_CSV3_SHIFT));
1261         if (val)
1262                 return -EINVAL;
1263
1264         vcpu->kvm->arch.pfr0_csv2 = csv2;
1265         vcpu->kvm->arch.pfr0_csv3 = csv3 ;
1266
1267         return 0;
1268 }
1269
1270 /*
1271  * cpufeature ID register user accessors
1272  *
1273  * For now, these registers are immutable for userspace, so no values
1274  * are stored, and for set_id_reg() we don't allow the effective value
1275  * to be changed.
1276  */
1277 static int __get_id_reg(const struct kvm_vcpu *vcpu,
1278                         const struct sys_reg_desc *rd, void __user *uaddr,
1279                         bool raz)
1280 {
1281         const u64 id = sys_reg_to_index(rd);
1282         const u64 val = read_id_reg(vcpu, rd, raz);
1283
1284         return reg_to_user(uaddr, &val, id);
1285 }
1286
1287 static int __set_id_reg(const struct kvm_vcpu *vcpu,
1288                         const struct sys_reg_desc *rd, void __user *uaddr,
1289                         bool raz)
1290 {
1291         const u64 id = sys_reg_to_index(rd);
1292         int err;
1293         u64 val;
1294
1295         err = reg_from_user(&val, uaddr, id);
1296         if (err)
1297                 return err;
1298
1299         /* This is what we mean by invariant: you can't change it. */
1300         if (val != read_id_reg(vcpu, rd, raz))
1301                 return -EINVAL;
1302
1303         return 0;
1304 }
1305
1306 static int get_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1307                       const struct kvm_one_reg *reg, void __user *uaddr)
1308 {
1309         bool raz = sysreg_visible_as_raz(vcpu, rd);
1310
1311         return __get_id_reg(vcpu, rd, uaddr, raz);
1312 }
1313
1314 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1315                       const struct kvm_one_reg *reg, void __user *uaddr)
1316 {
1317         bool raz = sysreg_visible_as_raz(vcpu, rd);
1318
1319         return __set_id_reg(vcpu, rd, uaddr, raz);
1320 }
1321
1322 static int set_raz_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1323                           const struct kvm_one_reg *reg, void __user *uaddr)
1324 {
1325         return __set_id_reg(vcpu, rd, uaddr, true);
1326 }
1327
1328 static int get_raz_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1329                        const struct kvm_one_reg *reg, void __user *uaddr)
1330 {
1331         const u64 id = sys_reg_to_index(rd);
1332         const u64 val = 0;
1333
1334         return reg_to_user(uaddr, &val, id);
1335 }
1336
1337 static int set_wi_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1338                       const struct kvm_one_reg *reg, void __user *uaddr)
1339 {
1340         int err;
1341         u64 val;
1342
1343         /* Perform the access even if we are going to ignore the value */
1344         err = reg_from_user(&val, uaddr, sys_reg_to_index(rd));
1345         if (err)
1346                 return err;
1347
1348         return 0;
1349 }
1350
1351 static bool access_ctr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1352                        const struct sys_reg_desc *r)
1353 {
1354         if (p->is_write)
1355                 return write_to_read_only(vcpu, p, r);
1356
1357         p->regval = read_sanitised_ftr_reg(SYS_CTR_EL0);
1358         return true;
1359 }
1360
1361 static bool access_clidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1362                          const struct sys_reg_desc *r)
1363 {
1364         if (p->is_write)
1365                 return write_to_read_only(vcpu, p, r);
1366
1367         p->regval = read_sysreg(clidr_el1);
1368         return true;
1369 }
1370
1371 static bool access_csselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1372                           const struct sys_reg_desc *r)
1373 {
1374         int reg = r->reg;
1375
1376         if (p->is_write)
1377                 vcpu_write_sys_reg(vcpu, p->regval, reg);
1378         else
1379                 p->regval = vcpu_read_sys_reg(vcpu, reg);
1380         return true;
1381 }
1382
1383 static bool access_ccsidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1384                           const struct sys_reg_desc *r)
1385 {
1386         u32 csselr;
1387
1388         if (p->is_write)
1389                 return write_to_read_only(vcpu, p, r);
1390
1391         csselr = vcpu_read_sys_reg(vcpu, CSSELR_EL1);
1392         p->regval = get_ccsidr(csselr);
1393
1394         /*
1395          * Guests should not be doing cache operations by set/way at all, and
1396          * for this reason, we trap them and attempt to infer the intent, so
1397          * that we can flush the entire guest's address space at the appropriate
1398          * time.
1399          * To prevent this trapping from causing performance problems, let's
1400          * expose the geometry of all data and unified caches (which are
1401          * guaranteed to be PIPT and thus non-aliasing) as 1 set and 1 way.
1402          * [If guests should attempt to infer aliasing properties from the
1403          * geometry (which is not permitted by the architecture), they would
1404          * only do so for virtually indexed caches.]
1405          */
1406         if (!(csselr & 1)) // data or unified cache
1407                 p->regval &= ~GENMASK(27, 3);
1408         return true;
1409 }
1410
1411 static unsigned int mte_visibility(const struct kvm_vcpu *vcpu,
1412                                    const struct sys_reg_desc *rd)
1413 {
1414         if (kvm_has_mte(vcpu->kvm))
1415                 return 0;
1416
1417         return REG_HIDDEN;
1418 }
1419
1420 #define MTE_REG(name) {                         \
1421         SYS_DESC(SYS_##name),                   \
1422         .access = undef_access,                 \
1423         .reset = reset_unknown,                 \
1424         .reg = name,                            \
1425         .visibility = mte_visibility,           \
1426 }
1427
1428 /* sys_reg_desc initialiser for known cpufeature ID registers */
1429 #define ID_SANITISED(name) {                    \
1430         SYS_DESC(SYS_##name),                   \
1431         .access = access_id_reg,                \
1432         .get_user = get_id_reg,                 \
1433         .set_user = set_id_reg,                 \
1434         .visibility = id_visibility,            \
1435 }
1436
1437 /*
1438  * sys_reg_desc initialiser for architecturally unallocated cpufeature ID
1439  * register with encoding Op0=3, Op1=0, CRn=0, CRm=crm, Op2=op2
1440  * (1 <= crm < 8, 0 <= Op2 < 8).
1441  */
1442 #define ID_UNALLOCATED(crm, op2) {                      \
1443         Op0(3), Op1(0), CRn(0), CRm(crm), Op2(op2),     \
1444         .access = access_raz_id_reg,                    \
1445         .get_user = get_raz_reg,                        \
1446         .set_user = set_raz_id_reg,                     \
1447 }
1448
1449 /*
1450  * sys_reg_desc initialiser for known ID registers that we hide from guests.
1451  * For now, these are exposed just like unallocated ID regs: they appear
1452  * RAZ for the guest.
1453  */
1454 #define ID_HIDDEN(name) {                       \
1455         SYS_DESC(SYS_##name),                   \
1456         .access = access_raz_id_reg,            \
1457         .get_user = get_raz_reg,                \
1458         .set_user = set_raz_id_reg,             \
1459 }
1460
1461 /*
1462  * Architected system registers.
1463  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
1464  *
1465  * Debug handling: We do trap most, if not all debug related system
1466  * registers. The implementation is good enough to ensure that a guest
1467  * can use these with minimal performance degradation. The drawback is
1468  * that we don't implement any of the external debug architecture.
1469  * This should be revisited if we ever encounter a more demanding
1470  * guest...
1471  */
1472 static const struct sys_reg_desc sys_reg_descs[] = {
1473         { SYS_DESC(SYS_DC_ISW), access_dcsw },
1474         { SYS_DESC(SYS_DC_CSW), access_dcsw },
1475         { SYS_DESC(SYS_DC_CISW), access_dcsw },
1476
1477         DBG_BCR_BVR_WCR_WVR_EL1(0),
1478         DBG_BCR_BVR_WCR_WVR_EL1(1),
1479         { SYS_DESC(SYS_MDCCINT_EL1), trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
1480         { SYS_DESC(SYS_MDSCR_EL1), trap_debug_regs, reset_val, MDSCR_EL1, 0 },
1481         DBG_BCR_BVR_WCR_WVR_EL1(2),
1482         DBG_BCR_BVR_WCR_WVR_EL1(3),
1483         DBG_BCR_BVR_WCR_WVR_EL1(4),
1484         DBG_BCR_BVR_WCR_WVR_EL1(5),
1485         DBG_BCR_BVR_WCR_WVR_EL1(6),
1486         DBG_BCR_BVR_WCR_WVR_EL1(7),
1487         DBG_BCR_BVR_WCR_WVR_EL1(8),
1488         DBG_BCR_BVR_WCR_WVR_EL1(9),
1489         DBG_BCR_BVR_WCR_WVR_EL1(10),
1490         DBG_BCR_BVR_WCR_WVR_EL1(11),
1491         DBG_BCR_BVR_WCR_WVR_EL1(12),
1492         DBG_BCR_BVR_WCR_WVR_EL1(13),
1493         DBG_BCR_BVR_WCR_WVR_EL1(14),
1494         DBG_BCR_BVR_WCR_WVR_EL1(15),
1495
1496         { SYS_DESC(SYS_MDRAR_EL1), trap_raz_wi },
1497         { SYS_DESC(SYS_OSLAR_EL1), trap_oslar_el1 },
1498         { SYS_DESC(SYS_OSLSR_EL1), trap_oslsr_el1, reset_val, OSLSR_EL1,
1499                 SYS_OSLSR_OSLM_IMPLEMENTED, .set_user = set_oslsr_el1, },
1500         { SYS_DESC(SYS_OSDLR_EL1), trap_raz_wi },
1501         { SYS_DESC(SYS_DBGPRCR_EL1), trap_raz_wi },
1502         { SYS_DESC(SYS_DBGCLAIMSET_EL1), trap_raz_wi },
1503         { SYS_DESC(SYS_DBGCLAIMCLR_EL1), trap_raz_wi },
1504         { SYS_DESC(SYS_DBGAUTHSTATUS_EL1), trap_dbgauthstatus_el1 },
1505
1506         { SYS_DESC(SYS_MDCCSR_EL0), trap_raz_wi },
1507         { SYS_DESC(SYS_DBGDTR_EL0), trap_raz_wi },
1508         // DBGDTR[TR]X_EL0 share the same encoding
1509         { SYS_DESC(SYS_DBGDTRTX_EL0), trap_raz_wi },
1510
1511         { SYS_DESC(SYS_DBGVCR32_EL2), NULL, reset_val, DBGVCR32_EL2, 0 },
1512
1513         { SYS_DESC(SYS_MPIDR_EL1), NULL, reset_mpidr, MPIDR_EL1 },
1514
1515         /*
1516          * ID regs: all ID_SANITISED() entries here must have corresponding
1517          * entries in arm64_ftr_regs[].
1518          */
1519
1520         /* AArch64 mappings of the AArch32 ID registers */
1521         /* CRm=1 */
1522         ID_SANITISED(ID_PFR0_EL1),
1523         ID_SANITISED(ID_PFR1_EL1),
1524         ID_SANITISED(ID_DFR0_EL1),
1525         ID_HIDDEN(ID_AFR0_EL1),
1526         ID_SANITISED(ID_MMFR0_EL1),
1527         ID_SANITISED(ID_MMFR1_EL1),
1528         ID_SANITISED(ID_MMFR2_EL1),
1529         ID_SANITISED(ID_MMFR3_EL1),
1530
1531         /* CRm=2 */
1532         ID_SANITISED(ID_ISAR0_EL1),
1533         ID_SANITISED(ID_ISAR1_EL1),
1534         ID_SANITISED(ID_ISAR2_EL1),
1535         ID_SANITISED(ID_ISAR3_EL1),
1536         ID_SANITISED(ID_ISAR4_EL1),
1537         ID_SANITISED(ID_ISAR5_EL1),
1538         ID_SANITISED(ID_MMFR4_EL1),
1539         ID_SANITISED(ID_ISAR6_EL1),
1540
1541         /* CRm=3 */
1542         ID_SANITISED(MVFR0_EL1),
1543         ID_SANITISED(MVFR1_EL1),
1544         ID_SANITISED(MVFR2_EL1),
1545         ID_UNALLOCATED(3,3),
1546         ID_SANITISED(ID_PFR2_EL1),
1547         ID_HIDDEN(ID_DFR1_EL1),
1548         ID_SANITISED(ID_MMFR5_EL1),
1549         ID_UNALLOCATED(3,7),
1550
1551         /* AArch64 ID registers */
1552         /* CRm=4 */
1553         { SYS_DESC(SYS_ID_AA64PFR0_EL1), .access = access_id_reg,
1554           .get_user = get_id_reg, .set_user = set_id_aa64pfr0_el1, },
1555         ID_SANITISED(ID_AA64PFR1_EL1),
1556         ID_UNALLOCATED(4,2),
1557         ID_UNALLOCATED(4,3),
1558         ID_SANITISED(ID_AA64ZFR0_EL1),
1559         ID_HIDDEN(ID_AA64SMFR0_EL1),
1560         ID_UNALLOCATED(4,6),
1561         ID_UNALLOCATED(4,7),
1562
1563         /* CRm=5 */
1564         ID_SANITISED(ID_AA64DFR0_EL1),
1565         ID_SANITISED(ID_AA64DFR1_EL1),
1566         ID_UNALLOCATED(5,2),
1567         ID_UNALLOCATED(5,3),
1568         ID_HIDDEN(ID_AA64AFR0_EL1),
1569         ID_HIDDEN(ID_AA64AFR1_EL1),
1570         ID_UNALLOCATED(5,6),
1571         ID_UNALLOCATED(5,7),
1572
1573         /* CRm=6 */
1574         ID_SANITISED(ID_AA64ISAR0_EL1),
1575         ID_SANITISED(ID_AA64ISAR1_EL1),
1576         ID_SANITISED(ID_AA64ISAR2_EL1),
1577         ID_UNALLOCATED(6,3),
1578         ID_UNALLOCATED(6,4),
1579         ID_UNALLOCATED(6,5),
1580         ID_UNALLOCATED(6,6),
1581         ID_UNALLOCATED(6,7),
1582
1583         /* CRm=7 */
1584         ID_SANITISED(ID_AA64MMFR0_EL1),
1585         ID_SANITISED(ID_AA64MMFR1_EL1),
1586         ID_SANITISED(ID_AA64MMFR2_EL1),
1587         ID_UNALLOCATED(7,3),
1588         ID_UNALLOCATED(7,4),
1589         ID_UNALLOCATED(7,5),
1590         ID_UNALLOCATED(7,6),
1591         ID_UNALLOCATED(7,7),
1592
1593         { SYS_DESC(SYS_SCTLR_EL1), access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
1594         { SYS_DESC(SYS_ACTLR_EL1), access_actlr, reset_actlr, ACTLR_EL1 },
1595         { SYS_DESC(SYS_CPACR_EL1), NULL, reset_val, CPACR_EL1, 0 },
1596
1597         MTE_REG(RGSR_EL1),
1598         MTE_REG(GCR_EL1),
1599
1600         { SYS_DESC(SYS_ZCR_EL1), NULL, reset_val, ZCR_EL1, 0, .visibility = sve_visibility },
1601         { SYS_DESC(SYS_TRFCR_EL1), undef_access },
1602         { SYS_DESC(SYS_SMPRI_EL1), undef_access },
1603         { SYS_DESC(SYS_SMCR_EL1), undef_access },
1604         { SYS_DESC(SYS_TTBR0_EL1), access_vm_reg, reset_unknown, TTBR0_EL1 },
1605         { SYS_DESC(SYS_TTBR1_EL1), access_vm_reg, reset_unknown, TTBR1_EL1 },
1606         { SYS_DESC(SYS_TCR_EL1), access_vm_reg, reset_val, TCR_EL1, 0 },
1607
1608         PTRAUTH_KEY(APIA),
1609         PTRAUTH_KEY(APIB),
1610         PTRAUTH_KEY(APDA),
1611         PTRAUTH_KEY(APDB),
1612         PTRAUTH_KEY(APGA),
1613
1614         { SYS_DESC(SYS_AFSR0_EL1), access_vm_reg, reset_unknown, AFSR0_EL1 },
1615         { SYS_DESC(SYS_AFSR1_EL1), access_vm_reg, reset_unknown, AFSR1_EL1 },
1616         { SYS_DESC(SYS_ESR_EL1), access_vm_reg, reset_unknown, ESR_EL1 },
1617
1618         { SYS_DESC(SYS_ERRIDR_EL1), trap_raz_wi },
1619         { SYS_DESC(SYS_ERRSELR_EL1), trap_raz_wi },
1620         { SYS_DESC(SYS_ERXFR_EL1), trap_raz_wi },
1621         { SYS_DESC(SYS_ERXCTLR_EL1), trap_raz_wi },
1622         { SYS_DESC(SYS_ERXSTATUS_EL1), trap_raz_wi },
1623         { SYS_DESC(SYS_ERXADDR_EL1), trap_raz_wi },
1624         { SYS_DESC(SYS_ERXMISC0_EL1), trap_raz_wi },
1625         { SYS_DESC(SYS_ERXMISC1_EL1), trap_raz_wi },
1626
1627         MTE_REG(TFSR_EL1),
1628         MTE_REG(TFSRE0_EL1),
1629
1630         { SYS_DESC(SYS_FAR_EL1), access_vm_reg, reset_unknown, FAR_EL1 },
1631         { SYS_DESC(SYS_PAR_EL1), NULL, reset_unknown, PAR_EL1 },
1632
1633         { SYS_DESC(SYS_PMSCR_EL1), undef_access },
1634         { SYS_DESC(SYS_PMSNEVFR_EL1), undef_access },
1635         { SYS_DESC(SYS_PMSICR_EL1), undef_access },
1636         { SYS_DESC(SYS_PMSIRR_EL1), undef_access },
1637         { SYS_DESC(SYS_PMSFCR_EL1), undef_access },
1638         { SYS_DESC(SYS_PMSEVFR_EL1), undef_access },
1639         { SYS_DESC(SYS_PMSLATFR_EL1), undef_access },
1640         { SYS_DESC(SYS_PMSIDR_EL1), undef_access },
1641         { SYS_DESC(SYS_PMBLIMITR_EL1), undef_access },
1642         { SYS_DESC(SYS_PMBPTR_EL1), undef_access },
1643         { SYS_DESC(SYS_PMBSR_EL1), undef_access },
1644         /* PMBIDR_EL1 is not trapped */
1645
1646         { PMU_SYS_REG(SYS_PMINTENSET_EL1),
1647           .access = access_pminten, .reg = PMINTENSET_EL1 },
1648         { PMU_SYS_REG(SYS_PMINTENCLR_EL1),
1649           .access = access_pminten, .reg = PMINTENSET_EL1 },
1650         { SYS_DESC(SYS_PMMIR_EL1), trap_raz_wi },
1651
1652         { SYS_DESC(SYS_MAIR_EL1), access_vm_reg, reset_unknown, MAIR_EL1 },
1653         { SYS_DESC(SYS_AMAIR_EL1), access_vm_reg, reset_amair_el1, AMAIR_EL1 },
1654
1655         { SYS_DESC(SYS_LORSA_EL1), trap_loregion },
1656         { SYS_DESC(SYS_LOREA_EL1), trap_loregion },
1657         { SYS_DESC(SYS_LORN_EL1), trap_loregion },
1658         { SYS_DESC(SYS_LORC_EL1), trap_loregion },
1659         { SYS_DESC(SYS_LORID_EL1), trap_loregion },
1660
1661         { SYS_DESC(SYS_VBAR_EL1), NULL, reset_val, VBAR_EL1, 0 },
1662         { SYS_DESC(SYS_DISR_EL1), NULL, reset_val, DISR_EL1, 0 },
1663
1664         { SYS_DESC(SYS_ICC_IAR0_EL1), write_to_read_only },
1665         { SYS_DESC(SYS_ICC_EOIR0_EL1), read_from_write_only },
1666         { SYS_DESC(SYS_ICC_HPPIR0_EL1), write_to_read_only },
1667         { SYS_DESC(SYS_ICC_DIR_EL1), read_from_write_only },
1668         { SYS_DESC(SYS_ICC_RPR_EL1), write_to_read_only },
1669         { SYS_DESC(SYS_ICC_SGI1R_EL1), access_gic_sgi },
1670         { SYS_DESC(SYS_ICC_ASGI1R_EL1), access_gic_sgi },
1671         { SYS_DESC(SYS_ICC_SGI0R_EL1), access_gic_sgi },
1672         { SYS_DESC(SYS_ICC_IAR1_EL1), write_to_read_only },
1673         { SYS_DESC(SYS_ICC_EOIR1_EL1), read_from_write_only },
1674         { SYS_DESC(SYS_ICC_HPPIR1_EL1), write_to_read_only },
1675         { SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
1676
1677         { SYS_DESC(SYS_CONTEXTIDR_EL1), access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
1678         { SYS_DESC(SYS_TPIDR_EL1), NULL, reset_unknown, TPIDR_EL1 },
1679
1680         { SYS_DESC(SYS_SCXTNUM_EL1), undef_access },
1681
1682         { SYS_DESC(SYS_CNTKCTL_EL1), NULL, reset_val, CNTKCTL_EL1, 0},
1683
1684         { SYS_DESC(SYS_CCSIDR_EL1), access_ccsidr },
1685         { SYS_DESC(SYS_CLIDR_EL1), access_clidr },
1686         { SYS_DESC(SYS_SMIDR_EL1), undef_access },
1687         { SYS_DESC(SYS_CSSELR_EL1), access_csselr, reset_unknown, CSSELR_EL1 },
1688         { SYS_DESC(SYS_CTR_EL0), access_ctr },
1689         { SYS_DESC(SYS_SVCR), undef_access },
1690
1691         { PMU_SYS_REG(SYS_PMCR_EL0), .access = access_pmcr,
1692           .reset = reset_pmcr, .reg = PMCR_EL0 },
1693         { PMU_SYS_REG(SYS_PMCNTENSET_EL0),
1694           .access = access_pmcnten, .reg = PMCNTENSET_EL0 },
1695         { PMU_SYS_REG(SYS_PMCNTENCLR_EL0),
1696           .access = access_pmcnten, .reg = PMCNTENSET_EL0 },
1697         { PMU_SYS_REG(SYS_PMOVSCLR_EL0),
1698           .access = access_pmovs, .reg = PMOVSSET_EL0 },
1699         /*
1700          * PM_SWINC_EL0 is exposed to userspace as RAZ/WI, as it was
1701          * previously (and pointlessly) advertised in the past...
1702          */
1703         { PMU_SYS_REG(SYS_PMSWINC_EL0),
1704           .get_user = get_raz_reg, .set_user = set_wi_reg,
1705           .access = access_pmswinc, .reset = NULL },
1706         { PMU_SYS_REG(SYS_PMSELR_EL0),
1707           .access = access_pmselr, .reset = reset_pmselr, .reg = PMSELR_EL0 },
1708         { PMU_SYS_REG(SYS_PMCEID0_EL0),
1709           .access = access_pmceid, .reset = NULL },
1710         { PMU_SYS_REG(SYS_PMCEID1_EL0),
1711           .access = access_pmceid, .reset = NULL },
1712         { PMU_SYS_REG(SYS_PMCCNTR_EL0),
1713           .access = access_pmu_evcntr, .reset = reset_unknown, .reg = PMCCNTR_EL0 },
1714         { PMU_SYS_REG(SYS_PMXEVTYPER_EL0),
1715           .access = access_pmu_evtyper, .reset = NULL },
1716         { PMU_SYS_REG(SYS_PMXEVCNTR_EL0),
1717           .access = access_pmu_evcntr, .reset = NULL },
1718         /*
1719          * PMUSERENR_EL0 resets as unknown in 64bit mode while it resets as zero
1720          * in 32bit mode. Here we choose to reset it as zero for consistency.
1721          */
1722         { PMU_SYS_REG(SYS_PMUSERENR_EL0), .access = access_pmuserenr,
1723           .reset = reset_val, .reg = PMUSERENR_EL0, .val = 0 },
1724         { PMU_SYS_REG(SYS_PMOVSSET_EL0),
1725           .access = access_pmovs, .reg = PMOVSSET_EL0 },
1726
1727         { SYS_DESC(SYS_TPIDR_EL0), NULL, reset_unknown, TPIDR_EL0 },
1728         { SYS_DESC(SYS_TPIDRRO_EL0), NULL, reset_unknown, TPIDRRO_EL0 },
1729         { SYS_DESC(SYS_TPIDR2_EL0), undef_access },
1730
1731         { SYS_DESC(SYS_SCXTNUM_EL0), undef_access },
1732
1733         { SYS_DESC(SYS_AMCR_EL0), undef_access },
1734         { SYS_DESC(SYS_AMCFGR_EL0), undef_access },
1735         { SYS_DESC(SYS_AMCGCR_EL0), undef_access },
1736         { SYS_DESC(SYS_AMUSERENR_EL0), undef_access },
1737         { SYS_DESC(SYS_AMCNTENCLR0_EL0), undef_access },
1738         { SYS_DESC(SYS_AMCNTENSET0_EL0), undef_access },
1739         { SYS_DESC(SYS_AMCNTENCLR1_EL0), undef_access },
1740         { SYS_DESC(SYS_AMCNTENSET1_EL0), undef_access },
1741         AMU_AMEVCNTR0_EL0(0),
1742         AMU_AMEVCNTR0_EL0(1),
1743         AMU_AMEVCNTR0_EL0(2),
1744         AMU_AMEVCNTR0_EL0(3),
1745         AMU_AMEVCNTR0_EL0(4),
1746         AMU_AMEVCNTR0_EL0(5),
1747         AMU_AMEVCNTR0_EL0(6),
1748         AMU_AMEVCNTR0_EL0(7),
1749         AMU_AMEVCNTR0_EL0(8),
1750         AMU_AMEVCNTR0_EL0(9),
1751         AMU_AMEVCNTR0_EL0(10),
1752         AMU_AMEVCNTR0_EL0(11),
1753         AMU_AMEVCNTR0_EL0(12),
1754         AMU_AMEVCNTR0_EL0(13),
1755         AMU_AMEVCNTR0_EL0(14),
1756         AMU_AMEVCNTR0_EL0(15),
1757         AMU_AMEVTYPER0_EL0(0),
1758         AMU_AMEVTYPER0_EL0(1),
1759         AMU_AMEVTYPER0_EL0(2),
1760         AMU_AMEVTYPER0_EL0(3),
1761         AMU_AMEVTYPER0_EL0(4),
1762         AMU_AMEVTYPER0_EL0(5),
1763         AMU_AMEVTYPER0_EL0(6),
1764         AMU_AMEVTYPER0_EL0(7),
1765         AMU_AMEVTYPER0_EL0(8),
1766         AMU_AMEVTYPER0_EL0(9),
1767         AMU_AMEVTYPER0_EL0(10),
1768         AMU_AMEVTYPER0_EL0(11),
1769         AMU_AMEVTYPER0_EL0(12),
1770         AMU_AMEVTYPER0_EL0(13),
1771         AMU_AMEVTYPER0_EL0(14),
1772         AMU_AMEVTYPER0_EL0(15),
1773         AMU_AMEVCNTR1_EL0(0),
1774         AMU_AMEVCNTR1_EL0(1),
1775         AMU_AMEVCNTR1_EL0(2),
1776         AMU_AMEVCNTR1_EL0(3),
1777         AMU_AMEVCNTR1_EL0(4),
1778         AMU_AMEVCNTR1_EL0(5),
1779         AMU_AMEVCNTR1_EL0(6),
1780         AMU_AMEVCNTR1_EL0(7),
1781         AMU_AMEVCNTR1_EL0(8),
1782         AMU_AMEVCNTR1_EL0(9),
1783         AMU_AMEVCNTR1_EL0(10),
1784         AMU_AMEVCNTR1_EL0(11),
1785         AMU_AMEVCNTR1_EL0(12),
1786         AMU_AMEVCNTR1_EL0(13),
1787         AMU_AMEVCNTR1_EL0(14),
1788         AMU_AMEVCNTR1_EL0(15),
1789         AMU_AMEVTYPER1_EL0(0),
1790         AMU_AMEVTYPER1_EL0(1),
1791         AMU_AMEVTYPER1_EL0(2),
1792         AMU_AMEVTYPER1_EL0(3),
1793         AMU_AMEVTYPER1_EL0(4),
1794         AMU_AMEVTYPER1_EL0(5),
1795         AMU_AMEVTYPER1_EL0(6),
1796         AMU_AMEVTYPER1_EL0(7),
1797         AMU_AMEVTYPER1_EL0(8),
1798         AMU_AMEVTYPER1_EL0(9),
1799         AMU_AMEVTYPER1_EL0(10),
1800         AMU_AMEVTYPER1_EL0(11),
1801         AMU_AMEVTYPER1_EL0(12),
1802         AMU_AMEVTYPER1_EL0(13),
1803         AMU_AMEVTYPER1_EL0(14),
1804         AMU_AMEVTYPER1_EL0(15),
1805
1806         { SYS_DESC(SYS_CNTP_TVAL_EL0), access_arch_timer },
1807         { SYS_DESC(SYS_CNTP_CTL_EL0), access_arch_timer },
1808         { SYS_DESC(SYS_CNTP_CVAL_EL0), access_arch_timer },
1809
1810         /* PMEVCNTRn_EL0 */
1811         PMU_PMEVCNTR_EL0(0),
1812         PMU_PMEVCNTR_EL0(1),
1813         PMU_PMEVCNTR_EL0(2),
1814         PMU_PMEVCNTR_EL0(3),
1815         PMU_PMEVCNTR_EL0(4),
1816         PMU_PMEVCNTR_EL0(5),
1817         PMU_PMEVCNTR_EL0(6),
1818         PMU_PMEVCNTR_EL0(7),
1819         PMU_PMEVCNTR_EL0(8),
1820         PMU_PMEVCNTR_EL0(9),
1821         PMU_PMEVCNTR_EL0(10),
1822         PMU_PMEVCNTR_EL0(11),
1823         PMU_PMEVCNTR_EL0(12),
1824         PMU_PMEVCNTR_EL0(13),
1825         PMU_PMEVCNTR_EL0(14),
1826         PMU_PMEVCNTR_EL0(15),
1827         PMU_PMEVCNTR_EL0(16),
1828         PMU_PMEVCNTR_EL0(17),
1829         PMU_PMEVCNTR_EL0(18),
1830         PMU_PMEVCNTR_EL0(19),
1831         PMU_PMEVCNTR_EL0(20),
1832         PMU_PMEVCNTR_EL0(21),
1833         PMU_PMEVCNTR_EL0(22),
1834         PMU_PMEVCNTR_EL0(23),
1835         PMU_PMEVCNTR_EL0(24),
1836         PMU_PMEVCNTR_EL0(25),
1837         PMU_PMEVCNTR_EL0(26),
1838         PMU_PMEVCNTR_EL0(27),
1839         PMU_PMEVCNTR_EL0(28),
1840         PMU_PMEVCNTR_EL0(29),
1841         PMU_PMEVCNTR_EL0(30),
1842         /* PMEVTYPERn_EL0 */
1843         PMU_PMEVTYPER_EL0(0),
1844         PMU_PMEVTYPER_EL0(1),
1845         PMU_PMEVTYPER_EL0(2),
1846         PMU_PMEVTYPER_EL0(3),
1847         PMU_PMEVTYPER_EL0(4),
1848         PMU_PMEVTYPER_EL0(5),
1849         PMU_PMEVTYPER_EL0(6),
1850         PMU_PMEVTYPER_EL0(7),
1851         PMU_PMEVTYPER_EL0(8),
1852         PMU_PMEVTYPER_EL0(9),
1853         PMU_PMEVTYPER_EL0(10),
1854         PMU_PMEVTYPER_EL0(11),
1855         PMU_PMEVTYPER_EL0(12),
1856         PMU_PMEVTYPER_EL0(13),
1857         PMU_PMEVTYPER_EL0(14),
1858         PMU_PMEVTYPER_EL0(15),
1859         PMU_PMEVTYPER_EL0(16),
1860         PMU_PMEVTYPER_EL0(17),
1861         PMU_PMEVTYPER_EL0(18),
1862         PMU_PMEVTYPER_EL0(19),
1863         PMU_PMEVTYPER_EL0(20),
1864         PMU_PMEVTYPER_EL0(21),
1865         PMU_PMEVTYPER_EL0(22),
1866         PMU_PMEVTYPER_EL0(23),
1867         PMU_PMEVTYPER_EL0(24),
1868         PMU_PMEVTYPER_EL0(25),
1869         PMU_PMEVTYPER_EL0(26),
1870         PMU_PMEVTYPER_EL0(27),
1871         PMU_PMEVTYPER_EL0(28),
1872         PMU_PMEVTYPER_EL0(29),
1873         PMU_PMEVTYPER_EL0(30),
1874         /*
1875          * PMCCFILTR_EL0 resets as unknown in 64bit mode while it resets as zero
1876          * in 32bit mode. Here we choose to reset it as zero for consistency.
1877          */
1878         { PMU_SYS_REG(SYS_PMCCFILTR_EL0), .access = access_pmu_evtyper,
1879           .reset = reset_val, .reg = PMCCFILTR_EL0, .val = 0 },
1880
1881         { SYS_DESC(SYS_DACR32_EL2), NULL, reset_unknown, DACR32_EL2 },
1882         { SYS_DESC(SYS_IFSR32_EL2), NULL, reset_unknown, IFSR32_EL2 },
1883         { SYS_DESC(SYS_FPEXC32_EL2), NULL, reset_val, FPEXC32_EL2, 0x700 },
1884 };
1885
1886 static bool trap_dbgdidr(struct kvm_vcpu *vcpu,
1887                         struct sys_reg_params *p,
1888                         const struct sys_reg_desc *r)
1889 {
1890         if (p->is_write) {
1891                 return ignore_write(vcpu, p);
1892         } else {
1893                 u64 dfr = read_sanitised_ftr_reg(SYS_ID_AA64DFR0_EL1);
1894                 u64 pfr = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1895                 u32 el3 = !!cpuid_feature_extract_unsigned_field(pfr, ID_AA64PFR0_EL3_SHIFT);
1896
1897                 p->regval = ((((dfr >> ID_AA64DFR0_WRPS_SHIFT) & 0xf) << 28) |
1898                              (((dfr >> ID_AA64DFR0_BRPS_SHIFT) & 0xf) << 24) |
1899                              (((dfr >> ID_AA64DFR0_CTX_CMPS_SHIFT) & 0xf) << 20)
1900                              | (6 << 16) | (1 << 15) | (el3 << 14) | (el3 << 12));
1901                 return true;
1902         }
1903 }
1904
1905 /*
1906  * AArch32 debug register mappings
1907  *
1908  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
1909  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
1910  *
1911  * None of the other registers share their location, so treat them as
1912  * if they were 64bit.
1913  */
1914 #define DBG_BCR_BVR_WCR_WVR(n)                                                \
1915         /* DBGBVRn */                                                         \
1916         { AA32(LO), Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, \
1917         /* DBGBCRn */                                                         \
1918         { Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },           \
1919         /* DBGWVRn */                                                         \
1920         { Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },           \
1921         /* DBGWCRn */                                                         \
1922         { Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
1923
1924 #define DBGBXVR(n)                                                            \
1925         { AA32(HI), Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_bvr, NULL, n }
1926
1927 /*
1928  * Trapped cp14 registers. We generally ignore most of the external
1929  * debug, on the principle that they don't really make sense to a
1930  * guest. Revisit this one day, would this principle change.
1931  */
1932 static const struct sys_reg_desc cp14_regs[] = {
1933         /* DBGDIDR */
1934         { Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgdidr },
1935         /* DBGDTRRXext */
1936         { Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
1937
1938         DBG_BCR_BVR_WCR_WVR(0),
1939         /* DBGDSCRint */
1940         { Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
1941         DBG_BCR_BVR_WCR_WVR(1),
1942         /* DBGDCCINT */
1943         { Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug_regs, NULL, MDCCINT_EL1 },
1944         /* DBGDSCRext */
1945         { Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug_regs, NULL, MDSCR_EL1 },
1946         DBG_BCR_BVR_WCR_WVR(2),
1947         /* DBGDTR[RT]Xint */
1948         { Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
1949         /* DBGDTR[RT]Xext */
1950         { Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
1951         DBG_BCR_BVR_WCR_WVR(3),
1952         DBG_BCR_BVR_WCR_WVR(4),
1953         DBG_BCR_BVR_WCR_WVR(5),
1954         /* DBGWFAR */
1955         { Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
1956         /* DBGOSECCR */
1957         { Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
1958         DBG_BCR_BVR_WCR_WVR(6),
1959         /* DBGVCR */
1960         { Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug_regs, NULL, DBGVCR32_EL2 },
1961         DBG_BCR_BVR_WCR_WVR(7),
1962         DBG_BCR_BVR_WCR_WVR(8),
1963         DBG_BCR_BVR_WCR_WVR(9),
1964         DBG_BCR_BVR_WCR_WVR(10),
1965         DBG_BCR_BVR_WCR_WVR(11),
1966         DBG_BCR_BVR_WCR_WVR(12),
1967         DBG_BCR_BVR_WCR_WVR(13),
1968         DBG_BCR_BVR_WCR_WVR(14),
1969         DBG_BCR_BVR_WCR_WVR(15),
1970
1971         /* DBGDRAR (32bit) */
1972         { Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
1973
1974         DBGBXVR(0),
1975         /* DBGOSLAR */
1976         { Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_oslar_el1 },
1977         DBGBXVR(1),
1978         /* DBGOSLSR */
1979         { Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1, NULL, OSLSR_EL1 },
1980         DBGBXVR(2),
1981         DBGBXVR(3),
1982         /* DBGOSDLR */
1983         { Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
1984         DBGBXVR(4),
1985         /* DBGPRCR */
1986         { Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
1987         DBGBXVR(5),
1988         DBGBXVR(6),
1989         DBGBXVR(7),
1990         DBGBXVR(8),
1991         DBGBXVR(9),
1992         DBGBXVR(10),
1993         DBGBXVR(11),
1994         DBGBXVR(12),
1995         DBGBXVR(13),
1996         DBGBXVR(14),
1997         DBGBXVR(15),
1998
1999         /* DBGDSAR (32bit) */
2000         { Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
2001
2002         /* DBGDEVID2 */
2003         { Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
2004         /* DBGDEVID1 */
2005         { Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
2006         /* DBGDEVID */
2007         { Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
2008         /* DBGCLAIMSET */
2009         { Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
2010         /* DBGCLAIMCLR */
2011         { Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
2012         /* DBGAUTHSTATUS */
2013         { Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
2014 };
2015
2016 /* Trapped cp14 64bit registers */
2017 static const struct sys_reg_desc cp14_64_regs[] = {
2018         /* DBGDRAR (64bit) */
2019         { Op1( 0), CRm( 1), .access = trap_raz_wi },
2020
2021         /* DBGDSAR (64bit) */
2022         { Op1( 0), CRm( 2), .access = trap_raz_wi },
2023 };
2024
2025 #define CP15_PMU_SYS_REG(_map, _Op1, _CRn, _CRm, _Op2)                  \
2026         AA32(_map),                                                     \
2027         Op1(_Op1), CRn(_CRn), CRm(_CRm), Op2(_Op2),                     \
2028         .visibility = pmu_visibility
2029
2030 /* Macro to expand the PMEVCNTRn register */
2031 #define PMU_PMEVCNTR(n)                                                 \
2032         { CP15_PMU_SYS_REG(DIRECT, 0, 0b1110,                           \
2033           (0b1000 | (((n) >> 3) & 0x3)), ((n) & 0x7)),                  \
2034           .access = access_pmu_evcntr }
2035
2036 /* Macro to expand the PMEVTYPERn register */
2037 #define PMU_PMEVTYPER(n)                                                \
2038         { CP15_PMU_SYS_REG(DIRECT, 0, 0b1110,                           \
2039           (0b1100 | (((n) >> 3) & 0x3)), ((n) & 0x7)),                  \
2040           .access = access_pmu_evtyper }
2041 /*
2042  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
2043  * depending on the way they are accessed (as a 32bit or a 64bit
2044  * register).
2045  */
2046 static const struct sys_reg_desc cp15_regs[] = {
2047         { Op1( 0), CRn( 0), CRm( 0), Op2( 1), access_ctr },
2048         { Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, SCTLR_EL1 },
2049         /* ACTLR */
2050         { AA32(LO), Op1( 0), CRn( 1), CRm( 0), Op2( 1), access_actlr, NULL, ACTLR_EL1 },
2051         /* ACTLR2 */
2052         { AA32(HI), Op1( 0), CRn( 1), CRm( 0), Op2( 3), access_actlr, NULL, ACTLR_EL1 },
2053         { Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
2054         { Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, TTBR1_EL1 },
2055         /* TTBCR */
2056         { AA32(LO), Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, TCR_EL1 },
2057         /* TTBCR2 */
2058         { AA32(HI), Op1( 0), CRn( 2), CRm( 0), Op2( 3), access_vm_reg, NULL, TCR_EL1 },
2059         { Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, DACR32_EL2 },
2060         /* DFSR */
2061         { Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, ESR_EL1 },
2062         { Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, IFSR32_EL2 },
2063         /* ADFSR */
2064         { Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, AFSR0_EL1 },
2065         /* AIFSR */
2066         { Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, AFSR1_EL1 },
2067         /* DFAR */
2068         { AA32(LO), Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, FAR_EL1 },
2069         /* IFAR */
2070         { AA32(HI), Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, FAR_EL1 },
2071
2072         /*
2073          * DC{C,I,CI}SW operations:
2074          */
2075         { Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
2076         { Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
2077         { Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
2078
2079         /* PMU */
2080         { CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 0), .access = access_pmcr },
2081         { CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 1), .access = access_pmcnten },
2082         { CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 2), .access = access_pmcnten },
2083         { CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 3), .access = access_pmovs },
2084         { CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 4), .access = access_pmswinc },
2085         { CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 5), .access = access_pmselr },
2086         { CP15_PMU_SYS_REG(LO,     0, 9, 12, 6), .access = access_pmceid },
2087         { CP15_PMU_SYS_REG(LO,     0, 9, 12, 7), .access = access_pmceid },
2088         { CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 0), .access = access_pmu_evcntr },
2089         { CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 1), .access = access_pmu_evtyper },
2090         { CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 2), .access = access_pmu_evcntr },
2091         { CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 0), .access = access_pmuserenr },
2092         { CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 1), .access = access_pminten },
2093         { CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 2), .access = access_pminten },
2094         { CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 3), .access = access_pmovs },
2095         { CP15_PMU_SYS_REG(HI,     0, 9, 14, 4), .access = access_pmceid },
2096         { CP15_PMU_SYS_REG(HI,     0, 9, 14, 5), .access = access_pmceid },
2097         /* PMMIR */
2098         { CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 6), .access = trap_raz_wi },
2099
2100         /* PRRR/MAIR0 */
2101         { AA32(LO), Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, MAIR_EL1 },
2102         /* NMRR/MAIR1 */
2103         { AA32(HI), Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, MAIR_EL1 },
2104         /* AMAIR0 */
2105         { AA32(LO), Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, AMAIR_EL1 },
2106         /* AMAIR1 */
2107         { AA32(HI), Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, AMAIR_EL1 },
2108
2109         /* ICC_SRE */
2110         { Op1( 0), CRn(12), CRm(12), Op2( 5), access_gic_sre },
2111
2112         { Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, CONTEXTIDR_EL1 },
2113
2114         /* Arch Tmers */
2115         { SYS_DESC(SYS_AARCH32_CNTP_TVAL), access_arch_timer },
2116         { SYS_DESC(SYS_AARCH32_CNTP_CTL), access_arch_timer },
2117
2118         /* PMEVCNTRn */
2119         PMU_PMEVCNTR(0),
2120         PMU_PMEVCNTR(1),
2121         PMU_PMEVCNTR(2),
2122         PMU_PMEVCNTR(3),
2123         PMU_PMEVCNTR(4),
2124         PMU_PMEVCNTR(5),
2125         PMU_PMEVCNTR(6),
2126         PMU_PMEVCNTR(7),
2127         PMU_PMEVCNTR(8),
2128         PMU_PMEVCNTR(9),
2129         PMU_PMEVCNTR(10),
2130         PMU_PMEVCNTR(11),
2131         PMU_PMEVCNTR(12),
2132         PMU_PMEVCNTR(13),
2133         PMU_PMEVCNTR(14),
2134         PMU_PMEVCNTR(15),
2135         PMU_PMEVCNTR(16),
2136         PMU_PMEVCNTR(17),
2137         PMU_PMEVCNTR(18),
2138         PMU_PMEVCNTR(19),
2139         PMU_PMEVCNTR(20),
2140         PMU_PMEVCNTR(21),
2141         PMU_PMEVCNTR(22),
2142         PMU_PMEVCNTR(23),
2143         PMU_PMEVCNTR(24),
2144         PMU_PMEVCNTR(25),
2145         PMU_PMEVCNTR(26),
2146         PMU_PMEVCNTR(27),
2147         PMU_PMEVCNTR(28),
2148         PMU_PMEVCNTR(29),
2149         PMU_PMEVCNTR(30),
2150         /* PMEVTYPERn */
2151         PMU_PMEVTYPER(0),
2152         PMU_PMEVTYPER(1),
2153         PMU_PMEVTYPER(2),
2154         PMU_PMEVTYPER(3),
2155         PMU_PMEVTYPER(4),
2156         PMU_PMEVTYPER(5),
2157         PMU_PMEVTYPER(6),
2158         PMU_PMEVTYPER(7),
2159         PMU_PMEVTYPER(8),
2160         PMU_PMEVTYPER(9),
2161         PMU_PMEVTYPER(10),
2162         PMU_PMEVTYPER(11),
2163         PMU_PMEVTYPER(12),
2164         PMU_PMEVTYPER(13),
2165         PMU_PMEVTYPER(14),
2166         PMU_PMEVTYPER(15),
2167         PMU_PMEVTYPER(16),
2168         PMU_PMEVTYPER(17),
2169         PMU_PMEVTYPER(18),
2170         PMU_PMEVTYPER(19),
2171         PMU_PMEVTYPER(20),
2172         PMU_PMEVTYPER(21),
2173         PMU_PMEVTYPER(22),
2174         PMU_PMEVTYPER(23),
2175         PMU_PMEVTYPER(24),
2176         PMU_PMEVTYPER(25),
2177         PMU_PMEVTYPER(26),
2178         PMU_PMEVTYPER(27),
2179         PMU_PMEVTYPER(28),
2180         PMU_PMEVTYPER(29),
2181         PMU_PMEVTYPER(30),
2182         /* PMCCFILTR */
2183         { CP15_PMU_SYS_REG(DIRECT, 0, 14, 15, 7), .access = access_pmu_evtyper },
2184
2185         { Op1(1), CRn( 0), CRm( 0), Op2(0), access_ccsidr },
2186         { Op1(1), CRn( 0), CRm( 0), Op2(1), access_clidr },
2187         { Op1(2), CRn( 0), CRm( 0), Op2(0), access_csselr, NULL, CSSELR_EL1 },
2188 };
2189
2190 static const struct sys_reg_desc cp15_64_regs[] = {
2191         { Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
2192         { CP15_PMU_SYS_REG(DIRECT, 0, 0, 9, 0), .access = access_pmu_evcntr },
2193         { Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI1R */
2194         { Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR1_EL1 },
2195         { Op1( 1), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_ASGI1R */
2196         { Op1( 2), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI0R */
2197         { SYS_DESC(SYS_AARCH32_CNTP_CVAL),    access_arch_timer },
2198 };
2199
2200 static bool check_sysreg_table(const struct sys_reg_desc *table, unsigned int n,
2201                                bool is_32)
2202 {
2203         unsigned int i;
2204
2205         for (i = 0; i < n; i++) {
2206                 if (!is_32 && table[i].reg && !table[i].reset) {
2207                         kvm_err("sys_reg table %pS entry %d lacks reset\n", &table[i], i);
2208                         return false;
2209                 }
2210
2211                 if (i && cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
2212                         kvm_err("sys_reg table %pS entry %d out of order\n", &table[i - 1], i - 1);
2213                         return false;
2214                 }
2215         }
2216
2217         return true;
2218 }
2219
2220 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu)
2221 {
2222         kvm_inject_undefined(vcpu);
2223         return 1;
2224 }
2225
2226 static void perform_access(struct kvm_vcpu *vcpu,
2227                            struct sys_reg_params *params,
2228                            const struct sys_reg_desc *r)
2229 {
2230         trace_kvm_sys_access(*vcpu_pc(vcpu), params, r);
2231
2232         /* Check for regs disabled by runtime config */
2233         if (sysreg_hidden(vcpu, r)) {
2234                 kvm_inject_undefined(vcpu);
2235                 return;
2236         }
2237
2238         /*
2239          * Not having an accessor means that we have configured a trap
2240          * that we don't know how to handle. This certainly qualifies
2241          * as a gross bug that should be fixed right away.
2242          */
2243         BUG_ON(!r->access);
2244
2245         /* Skip instruction if instructed so */
2246         if (likely(r->access(vcpu, params, r)))
2247                 kvm_incr_pc(vcpu);
2248 }
2249
2250 /*
2251  * emulate_cp --  tries to match a sys_reg access in a handling table, and
2252  *                call the corresponding trap handler.
2253  *
2254  * @params: pointer to the descriptor of the access
2255  * @table: array of trap descriptors
2256  * @num: size of the trap descriptor array
2257  *
2258  * Return true if the access has been handled, false if not.
2259  */
2260 static bool emulate_cp(struct kvm_vcpu *vcpu,
2261                        struct sys_reg_params *params,
2262                        const struct sys_reg_desc *table,
2263                        size_t num)
2264 {
2265         const struct sys_reg_desc *r;
2266
2267         if (!table)
2268                 return false;   /* Not handled */
2269
2270         r = find_reg(params, table, num);
2271
2272         if (r) {
2273                 perform_access(vcpu, params, r);
2274                 return true;
2275         }
2276
2277         /* Not handled */
2278         return false;
2279 }
2280
2281 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
2282                                 struct sys_reg_params *params)
2283 {
2284         u8 esr_ec = kvm_vcpu_trap_get_class(vcpu);
2285         int cp = -1;
2286
2287         switch (esr_ec) {
2288         case ESR_ELx_EC_CP15_32:
2289         case ESR_ELx_EC_CP15_64:
2290                 cp = 15;
2291                 break;
2292         case ESR_ELx_EC_CP14_MR:
2293         case ESR_ELx_EC_CP14_64:
2294                 cp = 14;
2295                 break;
2296         default:
2297                 WARN_ON(1);
2298         }
2299
2300         print_sys_reg_msg(params,
2301                           "Unsupported guest CP%d access at: %08lx [%08lx]\n",
2302                           cp, *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2303         kvm_inject_undefined(vcpu);
2304 }
2305
2306 /**
2307  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
2308  * @vcpu: The VCPU pointer
2309  * @run:  The kvm_run struct
2310  */
2311 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
2312                             const struct sys_reg_desc *global,
2313                             size_t nr_global)
2314 {
2315         struct sys_reg_params params;
2316         u64 esr = kvm_vcpu_get_esr(vcpu);
2317         int Rt = kvm_vcpu_sys_get_rt(vcpu);
2318         int Rt2 = (esr >> 10) & 0x1f;
2319
2320         params.CRm = (esr >> 1) & 0xf;
2321         params.is_write = ((esr & 1) == 0);
2322
2323         params.Op0 = 0;
2324         params.Op1 = (esr >> 16) & 0xf;
2325         params.Op2 = 0;
2326         params.CRn = 0;
2327
2328         /*
2329          * Make a 64-bit value out of Rt and Rt2. As we use the same trap
2330          * backends between AArch32 and AArch64, we get away with it.
2331          */
2332         if (params.is_write) {
2333                 params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
2334                 params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
2335         }
2336
2337         /*
2338          * If the table contains a handler, handle the
2339          * potential register operation in the case of a read and return
2340          * with success.
2341          */
2342         if (emulate_cp(vcpu, &params, global, nr_global)) {
2343                 /* Split up the value between registers for the read side */
2344                 if (!params.is_write) {
2345                         vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
2346                         vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
2347                 }
2348
2349                 return 1;
2350         }
2351
2352         unhandled_cp_access(vcpu, &params);
2353         return 1;
2354 }
2355
2356 static bool emulate_sys_reg(struct kvm_vcpu *vcpu, struct sys_reg_params *params);
2357
2358 /*
2359  * The CP10 ID registers are architecturally mapped to AArch64 feature
2360  * registers. Abuse that fact so we can rely on the AArch64 handler for accesses
2361  * from AArch32.
2362  */
2363 static bool kvm_esr_cp10_id_to_sys64(u64 esr, struct sys_reg_params *params)
2364 {
2365         u8 reg_id = (esr >> 10) & 0xf;
2366         bool valid;
2367
2368         params->is_write = ((esr & 1) == 0);
2369         params->Op0 = 3;
2370         params->Op1 = 0;
2371         params->CRn = 0;
2372         params->CRm = 3;
2373
2374         /* CP10 ID registers are read-only */
2375         valid = !params->is_write;
2376
2377         switch (reg_id) {
2378         /* MVFR0 */
2379         case 0b0111:
2380                 params->Op2 = 0;
2381                 break;
2382         /* MVFR1 */
2383         case 0b0110:
2384                 params->Op2 = 1;
2385                 break;
2386         /* MVFR2 */
2387         case 0b0101:
2388                 params->Op2 = 2;
2389                 break;
2390         default:
2391                 valid = false;
2392         }
2393
2394         if (valid)
2395                 return true;
2396
2397         kvm_pr_unimpl("Unhandled cp10 register %s: %u\n",
2398                       params->is_write ? "write" : "read", reg_id);
2399         return false;
2400 }
2401
2402 /**
2403  * kvm_handle_cp10_id() - Handles a VMRS trap on guest access to a 'Media and
2404  *                        VFP Register' from AArch32.
2405  * @vcpu: The vCPU pointer
2406  *
2407  * MVFR{0-2} are architecturally mapped to the AArch64 MVFR{0-2}_EL1 registers.
2408  * Work out the correct AArch64 system register encoding and reroute to the
2409  * AArch64 system register emulation.
2410  */
2411 int kvm_handle_cp10_id(struct kvm_vcpu *vcpu)
2412 {
2413         int Rt = kvm_vcpu_sys_get_rt(vcpu);
2414         u64 esr = kvm_vcpu_get_esr(vcpu);
2415         struct sys_reg_params params;
2416
2417         /* UNDEF on any unhandled register access */
2418         if (!kvm_esr_cp10_id_to_sys64(esr, &params)) {
2419                 kvm_inject_undefined(vcpu);
2420                 return 1;
2421         }
2422
2423         if (emulate_sys_reg(vcpu, &params))
2424                 vcpu_set_reg(vcpu, Rt, params.regval);
2425
2426         return 1;
2427 }
2428
2429 /**
2430  * kvm_emulate_cp15_id_reg() - Handles an MRC trap on a guest CP15 access where
2431  *                             CRn=0, which corresponds to the AArch32 feature
2432  *                             registers.
2433  * @vcpu: the vCPU pointer
2434  * @params: the system register access parameters.
2435  *
2436  * Our cp15 system register tables do not enumerate the AArch32 feature
2437  * registers. Conveniently, our AArch64 table does, and the AArch32 system
2438  * register encoding can be trivially remapped into the AArch64 for the feature
2439  * registers: Append op0=3, leaving op1, CRn, CRm, and op2 the same.
2440  *
2441  * According to DDI0487G.b G7.3.1, paragraph "Behavior of VMSAv8-32 32-bit
2442  * System registers with (coproc=0b1111, CRn==c0)", read accesses from this
2443  * range are either UNKNOWN or RES0. Rerouting remains architectural as we
2444  * treat undefined registers in this range as RAZ.
2445  */
2446 static int kvm_emulate_cp15_id_reg(struct kvm_vcpu *vcpu,
2447                                    struct sys_reg_params *params)
2448 {
2449         int Rt = kvm_vcpu_sys_get_rt(vcpu);
2450
2451         /* Treat impossible writes to RO registers as UNDEFINED */
2452         if (params->is_write) {
2453                 unhandled_cp_access(vcpu, params);
2454                 return 1;
2455         }
2456
2457         params->Op0 = 3;
2458
2459         /*
2460          * All registers where CRm > 3 are known to be UNKNOWN/RAZ from AArch32.
2461          * Avoid conflicting with future expansion of AArch64 feature registers
2462          * and simply treat them as RAZ here.
2463          */
2464         if (params->CRm > 3)
2465                 params->regval = 0;
2466         else if (!emulate_sys_reg(vcpu, params))
2467                 return 1;
2468
2469         vcpu_set_reg(vcpu, Rt, params->regval);
2470         return 1;
2471 }
2472
2473 /**
2474  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
2475  * @vcpu: The VCPU pointer
2476  * @run:  The kvm_run struct
2477  */
2478 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
2479                             struct sys_reg_params *params,
2480                             const struct sys_reg_desc *global,
2481                             size_t nr_global)
2482 {
2483         int Rt  = kvm_vcpu_sys_get_rt(vcpu);
2484
2485         params->regval = vcpu_get_reg(vcpu, Rt);
2486
2487         if (emulate_cp(vcpu, params, global, nr_global)) {
2488                 if (!params->is_write)
2489                         vcpu_set_reg(vcpu, Rt, params->regval);
2490                 return 1;
2491         }
2492
2493         unhandled_cp_access(vcpu, params);
2494         return 1;
2495 }
2496
2497 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu)
2498 {
2499         return kvm_handle_cp_64(vcpu, cp15_64_regs, ARRAY_SIZE(cp15_64_regs));
2500 }
2501
2502 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu)
2503 {
2504         struct sys_reg_params params;
2505
2506         params = esr_cp1x_32_to_params(kvm_vcpu_get_esr(vcpu));
2507
2508         /*
2509          * Certain AArch32 ID registers are handled by rerouting to the AArch64
2510          * system register table. Registers in the ID range where CRm=0 are
2511          * excluded from this scheme as they do not trivially map into AArch64
2512          * system register encodings.
2513          */
2514         if (params.Op1 == 0 && params.CRn == 0 && params.CRm)
2515                 return kvm_emulate_cp15_id_reg(vcpu, &params);
2516
2517         return kvm_handle_cp_32(vcpu, &params, cp15_regs, ARRAY_SIZE(cp15_regs));
2518 }
2519
2520 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu)
2521 {
2522         return kvm_handle_cp_64(vcpu, cp14_64_regs, ARRAY_SIZE(cp14_64_regs));
2523 }
2524
2525 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu)
2526 {
2527         struct sys_reg_params params;
2528
2529         params = esr_cp1x_32_to_params(kvm_vcpu_get_esr(vcpu));
2530
2531         return kvm_handle_cp_32(vcpu, &params, cp14_regs, ARRAY_SIZE(cp14_regs));
2532 }
2533
2534 static bool is_imp_def_sys_reg(struct sys_reg_params *params)
2535 {
2536         // See ARM DDI 0487E.a, section D12.3.2
2537         return params->Op0 == 3 && (params->CRn & 0b1011) == 0b1011;
2538 }
2539
2540 /**
2541  * emulate_sys_reg - Emulate a guest access to an AArch64 system register
2542  * @vcpu: The VCPU pointer
2543  * @params: Decoded system register parameters
2544  *
2545  * Return: true if the system register access was successful, false otherwise.
2546  */
2547 static bool emulate_sys_reg(struct kvm_vcpu *vcpu,
2548                            struct sys_reg_params *params)
2549 {
2550         const struct sys_reg_desc *r;
2551
2552         r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2553
2554         if (likely(r)) {
2555                 perform_access(vcpu, params, r);
2556                 return true;
2557         }
2558
2559         if (is_imp_def_sys_reg(params)) {
2560                 kvm_inject_undefined(vcpu);
2561         } else {
2562                 print_sys_reg_msg(params,
2563                                   "Unsupported guest sys_reg access at: %lx [%08lx]\n",
2564                                   *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2565                 kvm_inject_undefined(vcpu);
2566         }
2567         return false;
2568 }
2569
2570 /**
2571  * kvm_reset_sys_regs - sets system registers to reset value
2572  * @vcpu: The VCPU pointer
2573  *
2574  * This function finds the right table above and sets the registers on the
2575  * virtual CPU struct to their architecturally defined reset values.
2576  */
2577 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
2578 {
2579         unsigned long i;
2580
2581         for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++)
2582                 if (sys_reg_descs[i].reset)
2583                         sys_reg_descs[i].reset(vcpu, &sys_reg_descs[i]);
2584 }
2585
2586 /**
2587  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
2588  * @vcpu: The VCPU pointer
2589  */
2590 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu)
2591 {
2592         struct sys_reg_params params;
2593         unsigned long esr = kvm_vcpu_get_esr(vcpu);
2594         int Rt = kvm_vcpu_sys_get_rt(vcpu);
2595
2596         trace_kvm_handle_sys_reg(esr);
2597
2598         params = esr_sys64_to_params(esr);
2599         params.regval = vcpu_get_reg(vcpu, Rt);
2600
2601         if (!emulate_sys_reg(vcpu, &params))
2602                 return 1;
2603
2604         if (!params.is_write)
2605                 vcpu_set_reg(vcpu, Rt, params.regval);
2606         return 1;
2607 }
2608
2609 /******************************************************************************
2610  * Userspace API
2611  *****************************************************************************/
2612
2613 static bool index_to_params(u64 id, struct sys_reg_params *params)
2614 {
2615         switch (id & KVM_REG_SIZE_MASK) {
2616         case KVM_REG_SIZE_U64:
2617                 /* Any unused index bits means it's not valid. */
2618                 if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
2619                               | KVM_REG_ARM_COPROC_MASK
2620                               | KVM_REG_ARM64_SYSREG_OP0_MASK
2621                               | KVM_REG_ARM64_SYSREG_OP1_MASK
2622                               | KVM_REG_ARM64_SYSREG_CRN_MASK
2623                               | KVM_REG_ARM64_SYSREG_CRM_MASK
2624                               | KVM_REG_ARM64_SYSREG_OP2_MASK))
2625                         return false;
2626                 params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
2627                                >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
2628                 params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
2629                                >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
2630                 params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
2631                                >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
2632                 params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
2633                                >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
2634                 params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
2635                                >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
2636                 return true;
2637         default:
2638                 return false;
2639         }
2640 }
2641
2642 const struct sys_reg_desc *find_reg_by_id(u64 id,
2643                                           struct sys_reg_params *params,
2644                                           const struct sys_reg_desc table[],
2645                                           unsigned int num)
2646 {
2647         if (!index_to_params(id, params))
2648                 return NULL;
2649
2650         return find_reg(params, table, num);
2651 }
2652
2653 /* Decode an index value, and find the sys_reg_desc entry. */
2654 static const struct sys_reg_desc *index_to_sys_reg_desc(struct kvm_vcpu *vcpu,
2655                                                     u64 id)
2656 {
2657         const struct sys_reg_desc *r;
2658         struct sys_reg_params params;
2659
2660         /* We only do sys_reg for now. */
2661         if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
2662                 return NULL;
2663
2664         if (!index_to_params(id, &params))
2665                 return NULL;
2666
2667         r = find_reg(&params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2668
2669         /* Not saved in the sys_reg array and not otherwise accessible? */
2670         if (r && !(r->reg || r->get_user))
2671                 r = NULL;
2672
2673         return r;
2674 }
2675
2676 /*
2677  * These are the invariant sys_reg registers: we let the guest see the
2678  * host versions of these, so they're part of the guest state.
2679  *
2680  * A future CPU may provide a mechanism to present different values to
2681  * the guest, or a future kvm may trap them.
2682  */
2683
2684 #define FUNCTION_INVARIANT(reg)                                         \
2685         static void get_##reg(struct kvm_vcpu *v,                       \
2686                               const struct sys_reg_desc *r)             \
2687         {                                                               \
2688                 ((struct sys_reg_desc *)r)->val = read_sysreg(reg);     \
2689         }
2690
2691 FUNCTION_INVARIANT(midr_el1)
2692 FUNCTION_INVARIANT(revidr_el1)
2693 FUNCTION_INVARIANT(clidr_el1)
2694 FUNCTION_INVARIANT(aidr_el1)
2695
2696 static void get_ctr_el0(struct kvm_vcpu *v, const struct sys_reg_desc *r)
2697 {
2698         ((struct sys_reg_desc *)r)->val = read_sanitised_ftr_reg(SYS_CTR_EL0);
2699 }
2700
2701 /* ->val is filled in by kvm_sys_reg_table_init() */
2702 static struct sys_reg_desc invariant_sys_regs[] = {
2703         { SYS_DESC(SYS_MIDR_EL1), NULL, get_midr_el1 },
2704         { SYS_DESC(SYS_REVIDR_EL1), NULL, get_revidr_el1 },
2705         { SYS_DESC(SYS_CLIDR_EL1), NULL, get_clidr_el1 },
2706         { SYS_DESC(SYS_AIDR_EL1), NULL, get_aidr_el1 },
2707         { SYS_DESC(SYS_CTR_EL0), NULL, get_ctr_el0 },
2708 };
2709
2710 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id)
2711 {
2712         if (copy_from_user(val, uaddr, KVM_REG_SIZE(id)) != 0)
2713                 return -EFAULT;
2714         return 0;
2715 }
2716
2717 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id)
2718 {
2719         if (copy_to_user(uaddr, val, KVM_REG_SIZE(id)) != 0)
2720                 return -EFAULT;
2721         return 0;
2722 }
2723
2724 static int get_invariant_sys_reg(u64 id, void __user *uaddr)
2725 {
2726         struct sys_reg_params params;
2727         const struct sys_reg_desc *r;
2728
2729         r = find_reg_by_id(id, &params, invariant_sys_regs,
2730                            ARRAY_SIZE(invariant_sys_regs));
2731         if (!r)
2732                 return -ENOENT;
2733
2734         return reg_to_user(uaddr, &r->val, id);
2735 }
2736
2737 static int set_invariant_sys_reg(u64 id, void __user *uaddr)
2738 {
2739         struct sys_reg_params params;
2740         const struct sys_reg_desc *r;
2741         int err;
2742         u64 val = 0; /* Make sure high bits are 0 for 32-bit regs */
2743
2744         r = find_reg_by_id(id, &params, invariant_sys_regs,
2745                            ARRAY_SIZE(invariant_sys_regs));
2746         if (!r)
2747                 return -ENOENT;
2748
2749         err = reg_from_user(&val, uaddr, id);
2750         if (err)
2751                 return err;
2752
2753         /* This is what we mean by invariant: you can't change it. */
2754         if (r->val != val)
2755                 return -EINVAL;
2756
2757         return 0;
2758 }
2759
2760 static bool is_valid_cache(u32 val)
2761 {
2762         u32 level, ctype;
2763
2764         if (val >= CSSELR_MAX)
2765                 return false;
2766
2767         /* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
2768         level = (val >> 1);
2769         ctype = (cache_levels >> (level * 3)) & 7;
2770
2771         switch (ctype) {
2772         case 0: /* No cache */
2773                 return false;
2774         case 1: /* Instruction cache only */
2775                 return (val & 1);
2776         case 2: /* Data cache only */
2777         case 4: /* Unified cache */
2778                 return !(val & 1);
2779         case 3: /* Separate instruction and data caches */
2780                 return true;
2781         default: /* Reserved: we can't know instruction or data. */
2782                 return false;
2783         }
2784 }
2785
2786 static int demux_c15_get(u64 id, void __user *uaddr)
2787 {
2788         u32 val;
2789         u32 __user *uval = uaddr;
2790
2791         /* Fail if we have unknown bits set. */
2792         if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2793                    | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2794                 return -ENOENT;
2795
2796         switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2797         case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2798                 if (KVM_REG_SIZE(id) != 4)
2799                         return -ENOENT;
2800                 val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2801                         >> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2802                 if (!is_valid_cache(val))
2803                         return -ENOENT;
2804
2805                 return put_user(get_ccsidr(val), uval);
2806         default:
2807                 return -ENOENT;
2808         }
2809 }
2810
2811 static int demux_c15_set(u64 id, void __user *uaddr)
2812 {
2813         u32 val, newval;
2814         u32 __user *uval = uaddr;
2815
2816         /* Fail if we have unknown bits set. */
2817         if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2818                    | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2819                 return -ENOENT;
2820
2821         switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2822         case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2823                 if (KVM_REG_SIZE(id) != 4)
2824                         return -ENOENT;
2825                 val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2826                         >> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2827                 if (!is_valid_cache(val))
2828                         return -ENOENT;
2829
2830                 if (get_user(newval, uval))
2831                         return -EFAULT;
2832
2833                 /* This is also invariant: you can't change it. */
2834                 if (newval != get_ccsidr(val))
2835                         return -EINVAL;
2836                 return 0;
2837         default:
2838                 return -ENOENT;
2839         }
2840 }
2841
2842 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2843 {
2844         const struct sys_reg_desc *r;
2845         void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2846
2847         if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2848                 return demux_c15_get(reg->id, uaddr);
2849
2850         if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2851                 return -ENOENT;
2852
2853         r = index_to_sys_reg_desc(vcpu, reg->id);
2854         if (!r)
2855                 return get_invariant_sys_reg(reg->id, uaddr);
2856
2857         /* Check for regs disabled by runtime config */
2858         if (sysreg_hidden(vcpu, r))
2859                 return -ENOENT;
2860
2861         if (r->get_user)
2862                 return (r->get_user)(vcpu, r, reg, uaddr);
2863
2864         return reg_to_user(uaddr, &__vcpu_sys_reg(vcpu, r->reg), reg->id);
2865 }
2866
2867 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2868 {
2869         const struct sys_reg_desc *r;
2870         void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2871
2872         if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2873                 return demux_c15_set(reg->id, uaddr);
2874
2875         if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2876                 return -ENOENT;
2877
2878         r = index_to_sys_reg_desc(vcpu, reg->id);
2879         if (!r)
2880                 return set_invariant_sys_reg(reg->id, uaddr);
2881
2882         /* Check for regs disabled by runtime config */
2883         if (sysreg_hidden(vcpu, r))
2884                 return -ENOENT;
2885
2886         if (r->set_user)
2887                 return (r->set_user)(vcpu, r, reg, uaddr);
2888
2889         return reg_from_user(&__vcpu_sys_reg(vcpu, r->reg), uaddr, reg->id);
2890 }
2891
2892 static unsigned int num_demux_regs(void)
2893 {
2894         unsigned int i, count = 0;
2895
2896         for (i = 0; i < CSSELR_MAX; i++)
2897                 if (is_valid_cache(i))
2898                         count++;
2899
2900         return count;
2901 }
2902
2903 static int write_demux_regids(u64 __user *uindices)
2904 {
2905         u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
2906         unsigned int i;
2907
2908         val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
2909         for (i = 0; i < CSSELR_MAX; i++) {
2910                 if (!is_valid_cache(i))
2911                         continue;
2912                 if (put_user(val | i, uindices))
2913                         return -EFAULT;
2914                 uindices++;
2915         }
2916         return 0;
2917 }
2918
2919 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
2920 {
2921         return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
2922                 KVM_REG_ARM64_SYSREG |
2923                 (reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
2924                 (reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
2925                 (reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
2926                 (reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
2927                 (reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
2928 }
2929
2930 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
2931 {
2932         if (!*uind)
2933                 return true;
2934
2935         if (put_user(sys_reg_to_index(reg), *uind))
2936                 return false;
2937
2938         (*uind)++;
2939         return true;
2940 }
2941
2942 static int walk_one_sys_reg(const struct kvm_vcpu *vcpu,
2943                             const struct sys_reg_desc *rd,
2944                             u64 __user **uind,
2945                             unsigned int *total)
2946 {
2947         /*
2948          * Ignore registers we trap but don't save,
2949          * and for which no custom user accessor is provided.
2950          */
2951         if (!(rd->reg || rd->get_user))
2952                 return 0;
2953
2954         if (sysreg_hidden(vcpu, rd))
2955                 return 0;
2956
2957         if (!copy_reg_to_user(rd, uind))
2958                 return -EFAULT;
2959
2960         (*total)++;
2961         return 0;
2962 }
2963
2964 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
2965 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
2966 {
2967         const struct sys_reg_desc *i2, *end2;
2968         unsigned int total = 0;
2969         int err;
2970
2971         i2 = sys_reg_descs;
2972         end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
2973
2974         while (i2 != end2) {
2975                 err = walk_one_sys_reg(vcpu, i2++, &uind, &total);
2976                 if (err)
2977                         return err;
2978         }
2979         return total;
2980 }
2981
2982 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
2983 {
2984         return ARRAY_SIZE(invariant_sys_regs)
2985                 + num_demux_regs()
2986                 + walk_sys_regs(vcpu, (u64 __user *)NULL);
2987 }
2988
2989 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
2990 {
2991         unsigned int i;
2992         int err;
2993
2994         /* Then give them all the invariant registers' indices. */
2995         for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
2996                 if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
2997                         return -EFAULT;
2998                 uindices++;
2999         }
3000
3001         err = walk_sys_regs(vcpu, uindices);
3002         if (err < 0)
3003                 return err;
3004         uindices += err;
3005
3006         return write_demux_regids(uindices);
3007 }
3008
3009 int kvm_sys_reg_table_init(void)
3010 {
3011         bool valid = true;
3012         unsigned int i;
3013         struct sys_reg_desc clidr;
3014
3015         /* Make sure tables are unique and in order. */
3016         valid &= check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs), false);
3017         valid &= check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs), true);
3018         valid &= check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs), true);
3019         valid &= check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs), true);
3020         valid &= check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs), true);
3021         valid &= check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs), false);
3022
3023         if (!valid)
3024                 return -EINVAL;
3025
3026         /* We abuse the reset function to overwrite the table itself. */
3027         for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
3028                 invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
3029
3030         /*
3031          * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
3032          *
3033          *   If software reads the Cache Type fields from Ctype1
3034          *   upwards, once it has seen a value of 0b000, no caches
3035          *   exist at further-out levels of the hierarchy. So, for
3036          *   example, if Ctype3 is the first Cache Type field with a
3037          *   value of 0b000, the values of Ctype4 to Ctype7 must be
3038          *   ignored.
3039          */
3040         get_clidr_el1(NULL, &clidr); /* Ugly... */
3041         cache_levels = clidr.val;
3042         for (i = 0; i < 7; i++)
3043                 if (((cache_levels >> (i*3)) & 7) == 0)
3044                         break;
3045         /* Clear all higher bits. */
3046         cache_levels &= (1 << (i*3))-1;
3047
3048         return 0;
3049 }