arch/arm64/kvm/sys_regs.c

   1 /*
   2  * Copyright (C) 2012,2013 - ARM Ltd
   3  * Author: Marc Zyngier <marc.zyngier@arm.com>
   4  *
   5  * Derived from arch/arm/kvm/coproc.c:
   6  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
   7  * Authors: Rusty Russell <rusty@rustcorp.com.au>
   8  *          Christoffer Dall <c.dall@virtualopensystems.com>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License, version 2, as
  12  * published by the Free Software Foundation.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License
  20  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  21  */
  22
  23 #include <linux/mm.h>
  24 #include <linux/kvm_host.h>
  25 #include <linux/uaccess.h>
  26 #include <asm/kvm_arm.h>
  27 #include <asm/kvm_host.h>
  28 #include <asm/kvm_emulate.h>
  29 #include <asm/kvm_coproc.h>
  30 #include <asm/kvm_mmu.h>
  31 #include <asm/cacheflush.h>
  32 #include <asm/cputype.h>
  33 #include <asm/debug-monitors.h>
  34 #include <trace/events/kvm.h>
  35
  36 #include "sys_regs.h"
  37
  38 /*
  39  * All of this file is extremly similar to the ARM coproc.c, but the
  40  * types are different. My gut feeling is that it should be pretty
  41  * easy to merge, but that would be an ABI breakage -- again. VFP
  42  * would also need to be abstracted.
  43  *
  44  * For AArch32, we only take care of what is being trapped. Anything
  45  * that has to do with init and userspace access has to go via the
  46  * 64bit interface.
  47  */
  48
  49 /* 3 bits per cache level, as per CLIDR, but non-existent caches always 0 */
  50 static u32 cache_levels;
  51
  52 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
  53 #define CSSELR_MAX 12
  54
  55 /* Which cache CCSIDR represents depends on CSSELR value. */
  56 static u32 get_ccsidr(u32 csselr)
  57 {
  58         u32 ccsidr;
  59
  60         /* Make sure noone else changes CSSELR during this! */
  61         local_irq_disable();
  62         /* Put value into CSSELR */
  63         asm volatile("msr csselr_el1, %x0" : : "r" (csselr));
  64         isb();
  65         /* Read result out of CCSIDR */
  66         asm volatile("mrs %0, ccsidr_el1" : "=r" (ccsidr));
  67         local_irq_enable();
  68
  69         return ccsidr;
  70 }
  71
  72 static void do_dc_cisw(u32 val)
  73 {
  74         asm volatile("dc cisw, %x0" : : "r" (val));
  75         dsb(ish);
  76 }
  77
  78 static void do_dc_csw(u32 val)
  79 {
  80         asm volatile("dc csw, %x0" : : "r" (val));
  81         dsb(ish);
  82 }
  83
  84 /* See note at ARM ARM B1.14.4 */
  85 static bool access_dcsw(struct kvm_vcpu *vcpu,
  86                         const struct sys_reg_params *p,
  87                         const struct sys_reg_desc *r)
  88 {
  89         unsigned long val;
  90         int cpu;
  91
  92         if (!p->is_write)
  93                 return read_from_write_only(vcpu, p);
  94
  95         cpu = get_cpu();
  96
  97         cpumask_setall(&vcpu->arch.require_dcache_flush);
  98         cpumask_clear_cpu(cpu, &vcpu->arch.require_dcache_flush);
  99
 100         /* If we were already preempted, take the long way around */
 101         if (cpu != vcpu->arch.last_pcpu) {
 102                 flush_cache_all();
 103                 goto done;
 104         }
 105
 106         val = *vcpu_reg(vcpu, p->Rt);
 107
 108         switch (p->CRm) {
 109         case 6:                 /* Upgrade DCISW to DCCISW, as per HCR.SWIO */
 110         case 14:                /* DCCISW */
 111                 do_dc_cisw(val);
 112                 break;
 113
 114         case 10:                /* DCCSW */
 115                 do_dc_csw(val);
 116                 break;
 117         }
 118
 119 done:
 120         put_cpu();
 121
 122         return true;
 123 }
 124
 125 /*
 126  * Generic accessor for VM registers. Only called as long as HCR_TVM
 127  * is set.
 128  */
 129 static bool access_vm_reg(struct kvm_vcpu *vcpu,
 130                           const struct sys_reg_params *p,
 131                           const struct sys_reg_desc *r)
 132 {
 133         unsigned long val;
 134
 135         BUG_ON(!p->is_write);
 136
 137         val = *vcpu_reg(vcpu, p->Rt);
 138         if (!p->is_aarch32 || !p->is_32bit)
 139                 vcpu_sys_reg(vcpu, r->reg) = val;
 140         else
 141                 vcpu_cp15_64_low(vcpu, r->reg) = val & 0xffffffffUL;
 142
 143         return true;
 144 }
 145
 146 /*
 147  * SCTLR_EL1 accessor. Only called as long as HCR_TVM is set.  If the
 148  * guest enables the MMU, we stop trapping the VM sys_regs and leave
 149  * it in complete control of the caches.
 150  */
 151 static bool access_sctlr(struct kvm_vcpu *vcpu,
 152                          const struct sys_reg_params *p,
 153                          const struct sys_reg_desc *r)
 154 {
 155         access_vm_reg(vcpu, p, r);
 156
 157         if (vcpu_has_cache_enabled(vcpu)) {     /* MMU+Caches enabled? */
 158                 vcpu->arch.hcr_el2 &= ~HCR_TVM;
 159                 stage2_flush_vm(vcpu->kvm);
 160         }
 161
 162         return true;
 163 }
 164
 165 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
 166                         const struct sys_reg_params *p,
 167                         const struct sys_reg_desc *r)
 168 {
 169         if (p->is_write)
 170                 return ignore_write(vcpu, p);
 171         else
 172                 return read_zero(vcpu, p);
 173 }
 174
 175 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
 176                            const struct sys_reg_params *p,
 177                            const struct sys_reg_desc *r)
 178 {
 179         if (p->is_write) {
 180                 return ignore_write(vcpu, p);
 181         } else {
 182                 *vcpu_reg(vcpu, p->Rt) = (1 << 3);
 183                 return true;
 184         }
 185 }
 186
 187 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
 188                                    const struct sys_reg_params *p,
 189                                    const struct sys_reg_desc *r)
 190 {
 191         if (p->is_write) {
 192                 return ignore_write(vcpu, p);
 193         } else {
 194                 u32 val;
 195                 asm volatile("mrs %0, dbgauthstatus_el1" : "=r" (val));
 196                 *vcpu_reg(vcpu, p->Rt) = val;
 197                 return true;
 198         }
 199 }
 200
 201 /*
 202  * We want to avoid world-switching all the DBG registers all the
 203  * time:
 204  *
 205  * - If we've touched any debug register, it is likely that we're
 206  *   going to touch more of them. It then makes sense to disable the
 207  *   traps and start doing the save/restore dance
 208  * - If debug is active (DBG_MDSCR_KDE or DBG_MDSCR_MDE set), it is
 209  *   then mandatory to save/restore the registers, as the guest
 210  *   depends on them.
 211  *
 212  * For this, we use a DIRTY bit, indicating the guest has modified the
 213  * debug registers, used as follow:
 214  *
 215  * On guest entry:
 216  * - If the dirty bit is set (because we're coming back from trapping),
 217  *   disable the traps, save host registers, restore guest registers.
 218  * - If debug is actively in use (DBG_MDSCR_KDE or DBG_MDSCR_MDE set),
 219  *   set the dirty bit, disable the traps, save host registers,
 220  *   restore guest registers.
 221  * - Otherwise, enable the traps
 222  *
 223  * On guest exit:
 224  * - If the dirty bit is set, save guest registers, restore host
 225  *   registers and clear the dirty bit. This ensure that the host can
 226  *   now use the debug registers.
 227  */
 228 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
 229                             const struct sys_reg_params *p,
 230                             const struct sys_reg_desc *r)
 231 {
 232         if (p->is_write) {
 233                 vcpu_sys_reg(vcpu, r->reg) = *vcpu_reg(vcpu, p->Rt);
 234                 vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
 235         } else {
 236                 *vcpu_reg(vcpu, p->Rt) = vcpu_sys_reg(vcpu, r->reg);
 237         }
 238
 239         return true;
 240 }
 241
 242 static void reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
 243 {
 244         u64 amair;
 245
 246         asm volatile("mrs %0, amair_el1\n" : "=r" (amair));
 247         vcpu_sys_reg(vcpu, AMAIR_EL1) = amair;
 248 }
 249
 250 static void reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
 251 {
 252         /*
 253          * Simply map the vcpu_id into the Aff0 field of the MPIDR.
 254          */
 255         vcpu_sys_reg(vcpu, MPIDR_EL1) = (1UL << 31) | (vcpu->vcpu_id & 0xff);
 256 }
 257
 258 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
 259 #define DBG_BCR_BVR_WCR_WVR_EL1(n)                                      \
 260         /* DBGBVRn_EL1 */                                               \
 261         { Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b100),     \
 262           trap_debug_regs, reset_val, (DBGBVR0_EL1 + (n)), 0 },         \
 263         /* DBGBCRn_EL1 */                                               \
 264         { Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b101),     \
 265           trap_debug_regs, reset_val, (DBGBCR0_EL1 + (n)), 0 },         \
 266         /* DBGWVRn_EL1 */                                               \
 267         { Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b110),     \
 268           trap_debug_regs, reset_val, (DBGWVR0_EL1 + (n)), 0 },         \
 269         /* DBGWCRn_EL1 */                                               \
 270         { Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b111),     \
 271           trap_debug_regs, reset_val, (DBGWCR0_EL1 + (n)), 0 }
 272
 273 /*
 274  * Architected system registers.
 275  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
 276  *
 277  * We could trap ID_DFR0 and tell the guest we don't support performance
 278  * monitoring.  Unfortunately the patch to make the kernel check ID_DFR0 was
 279  * NAKed, so it will read the PMCR anyway.
 280  *
 281  * Therefore we tell the guest we have 0 counters.  Unfortunately, we
 282  * must always support PMCCNTR (the cycle counter): we just RAZ/WI for
 283  * all PM registers, which doesn't crash the guest kernel at least.
 284  *
 285  * Debug handling: We do trap most, if not all debug related system
 286  * registers. The implementation is good enough to ensure that a guest
 287  * can use these with minimal performance degradation. The drawback is
 288  * that we don't implement any of the external debug, none of the
 289  * OSlock protocol. This should be revisited if we ever encounter a
 290  * more demanding guest...
 291  */
 292 static const struct sys_reg_desc sys_reg_descs[] = {
 293         /* DC ISW */
 294         { Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b0110), Op2(0b010),
 295           access_dcsw },
 296         /* DC CSW */
 297         { Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b1010), Op2(0b010),
 298           access_dcsw },
 299         /* DC CISW */
 300         { Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b1110), Op2(0b010),
 301           access_dcsw },
 302
 303         DBG_BCR_BVR_WCR_WVR_EL1(0),
 304         DBG_BCR_BVR_WCR_WVR_EL1(1),
 305         /* MDCCINT_EL1 */
 306         { Op0(0b10), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b000),
 307           trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
 308         /* MDSCR_EL1 */
 309         { Op0(0b10), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b010),
 310           trap_debug_regs, reset_val, MDSCR_EL1, 0 },
 311         DBG_BCR_BVR_WCR_WVR_EL1(2),
 312         DBG_BCR_BVR_WCR_WVR_EL1(3),
 313         DBG_BCR_BVR_WCR_WVR_EL1(4),
 314         DBG_BCR_BVR_WCR_WVR_EL1(5),
 315         DBG_BCR_BVR_WCR_WVR_EL1(6),
 316         DBG_BCR_BVR_WCR_WVR_EL1(7),
 317         DBG_BCR_BVR_WCR_WVR_EL1(8),
 318         DBG_BCR_BVR_WCR_WVR_EL1(9),
 319         DBG_BCR_BVR_WCR_WVR_EL1(10),
 320         DBG_BCR_BVR_WCR_WVR_EL1(11),
 321         DBG_BCR_BVR_WCR_WVR_EL1(12),
 322         DBG_BCR_BVR_WCR_WVR_EL1(13),
 323         DBG_BCR_BVR_WCR_WVR_EL1(14),
 324         DBG_BCR_BVR_WCR_WVR_EL1(15),
 325
 326         /* MDRAR_EL1 */
 327         { Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b000),
 328           trap_raz_wi },
 329         /* OSLAR_EL1 */
 330         { Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b100),
 331           trap_raz_wi },
 332         /* OSLSR_EL1 */
 333         { Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0001), Op2(0b100),
 334           trap_oslsr_el1 },
 335         /* OSDLR_EL1 */
 336         { Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0011), Op2(0b100),
 337           trap_raz_wi },
 338         /* DBGPRCR_EL1 */
 339         { Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0100), Op2(0b100),
 340           trap_raz_wi },
 341         /* DBGCLAIMSET_EL1 */
 342         { Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1000), Op2(0b110),
 343           trap_raz_wi },
 344         /* DBGCLAIMCLR_EL1 */
 345         { Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1001), Op2(0b110),
 346           trap_raz_wi },
 347         /* DBGAUTHSTATUS_EL1 */
 348         { Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1110), Op2(0b110),
 349           trap_dbgauthstatus_el1 },
 350
 351         /* TEECR32_EL1 */
 352         { Op0(0b10), Op1(0b010), CRn(0b0000), CRm(0b0000), Op2(0b000),
 353           NULL, reset_val, TEECR32_EL1, 0 },
 354         /* TEEHBR32_EL1 */
 355         { Op0(0b10), Op1(0b010), CRn(0b0001), CRm(0b0000), Op2(0b000),
 356           NULL, reset_val, TEEHBR32_EL1, 0 },
 357
 358         /* MDCCSR_EL1 */
 359         { Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0001), Op2(0b000),
 360           trap_raz_wi },
 361         /* DBGDTR_EL0 */
 362         { Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0100), Op2(0b000),
 363           trap_raz_wi },
 364         /* DBGDTR[TR]X_EL0 */
 365         { Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0101), Op2(0b000),
 366           trap_raz_wi },
 367
 368         /* DBGVCR32_EL2 */
 369         { Op0(0b10), Op1(0b100), CRn(0b0000), CRm(0b0111), Op2(0b000),
 370           NULL, reset_val, DBGVCR32_EL2, 0 },
 371
 372         /* MPIDR_EL1 */
 373         { Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b101),
 374           NULL, reset_mpidr, MPIDR_EL1 },
 375         /* SCTLR_EL1 */
 376         { Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b000),
 377           access_sctlr, reset_val, SCTLR_EL1, 0x00C50078 },
 378         /* CPACR_EL1 */
 379         { Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b010),
 380           NULL, reset_val, CPACR_EL1, 0 },
 381         /* TTBR0_EL1 */
 382         { Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b000),
 383           access_vm_reg, reset_unknown, TTBR0_EL1 },
 384         /* TTBR1_EL1 */
 385         { Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b001),
 386           access_vm_reg, reset_unknown, TTBR1_EL1 },
 387         /* TCR_EL1 */
 388         { Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b010),
 389           access_vm_reg, reset_val, TCR_EL1, 0 },
 390
 391         /* AFSR0_EL1 */
 392         { Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0001), Op2(0b000),
 393           access_vm_reg, reset_unknown, AFSR0_EL1 },
 394         /* AFSR1_EL1 */
 395         { Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0001), Op2(0b001),
 396           access_vm_reg, reset_unknown, AFSR1_EL1 },
 397         /* ESR_EL1 */
 398         { Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0010), Op2(0b000),
 399           access_vm_reg, reset_unknown, ESR_EL1 },
 400         /* FAR_EL1 */
 401         { Op0(0b11), Op1(0b000), CRn(0b0110), CRm(0b0000), Op2(0b000),
 402           access_vm_reg, reset_unknown, FAR_EL1 },
 403         /* PAR_EL1 */
 404         { Op0(0b11), Op1(0b000), CRn(0b0111), CRm(0b0100), Op2(0b000),
 405           NULL, reset_unknown, PAR_EL1 },
 406
 407         /* PMINTENSET_EL1 */
 408         { Op0(0b11), Op1(0b000), CRn(0b1001), CRm(0b1110), Op2(0b001),
 409           trap_raz_wi },
 410         /* PMINTENCLR_EL1 */
 411         { Op0(0b11), Op1(0b000), CRn(0b1001), CRm(0b1110), Op2(0b010),
 412           trap_raz_wi },
 413
 414         /* MAIR_EL1 */
 415         { Op0(0b11), Op1(0b000), CRn(0b1010), CRm(0b0010), Op2(0b000),
 416           access_vm_reg, reset_unknown, MAIR_EL1 },
 417         /* AMAIR_EL1 */
 418         { Op0(0b11), Op1(0b000), CRn(0b1010), CRm(0b0011), Op2(0b000),
 419           access_vm_reg, reset_amair_el1, AMAIR_EL1 },
 420
 421         /* VBAR_EL1 */
 422         { Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b0000), Op2(0b000),
 423           NULL, reset_val, VBAR_EL1, 0 },
 424         /* CONTEXTIDR_EL1 */
 425         { Op0(0b11), Op1(0b000), CRn(0b1101), CRm(0b0000), Op2(0b001),
 426           access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
 427         /* TPIDR_EL1 */
 428         { Op0(0b11), Op1(0b000), CRn(0b1101), CRm(0b0000), Op2(0b100),
 429           NULL, reset_unknown, TPIDR_EL1 },
 430
 431         /* CNTKCTL_EL1 */
 432         { Op0(0b11), Op1(0b000), CRn(0b1110), CRm(0b0001), Op2(0b000),
 433           NULL, reset_val, CNTKCTL_EL1, 0},
 434
 435         /* CSSELR_EL1 */
 436         { Op0(0b11), Op1(0b010), CRn(0b0000), CRm(0b0000), Op2(0b000),
 437           NULL, reset_unknown, CSSELR_EL1 },
 438
 439         /* PMCR_EL0 */
 440         { Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b000),
 441           trap_raz_wi },
 442         /* PMCNTENSET_EL0 */
 443         { Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b001),
 444           trap_raz_wi },
 445         /* PMCNTENCLR_EL0 */
 446         { Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b010),
 447           trap_raz_wi },
 448         /* PMOVSCLR_EL0 */
 449         { Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b011),
 450           trap_raz_wi },
 451         /* PMSWINC_EL0 */
 452         { Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b100),
 453           trap_raz_wi },
 454         /* PMSELR_EL0 */
 455         { Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b101),
 456           trap_raz_wi },
 457         /* PMCEID0_EL0 */
 458         { Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b110),
 459           trap_raz_wi },
 460         /* PMCEID1_EL0 */
 461         { Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b111),
 462           trap_raz_wi },
 463         /* PMCCNTR_EL0 */
 464         { Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b000),
 465           trap_raz_wi },
 466         /* PMXEVTYPER_EL0 */
 467         { Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b001),
 468           trap_raz_wi },
 469         /* PMXEVCNTR_EL0 */
 470         { Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b010),
 471           trap_raz_wi },
 472         /* PMUSERENR_EL0 */
 473         { Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1110), Op2(0b000),
 474           trap_raz_wi },
 475         /* PMOVSSET_EL0 */
 476         { Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1110), Op2(0b011),
 477           trap_raz_wi },
 478
 479         /* TPIDR_EL0 */
 480         { Op0(0b11), Op1(0b011), CRn(0b1101), CRm(0b0000), Op2(0b010),
 481           NULL, reset_unknown, TPIDR_EL0 },
 482         /* TPIDRRO_EL0 */
 483         { Op0(0b11), Op1(0b011), CRn(0b1101), CRm(0b0000), Op2(0b011),
 484           NULL, reset_unknown, TPIDRRO_EL0 },
 485
 486         /* DACR32_EL2 */
 487         { Op0(0b11), Op1(0b100), CRn(0b0011), CRm(0b0000), Op2(0b000),
 488           NULL, reset_unknown, DACR32_EL2 },
 489         /* IFSR32_EL2 */
 490         { Op0(0b11), Op1(0b100), CRn(0b0101), CRm(0b0000), Op2(0b001),
 491           NULL, reset_unknown, IFSR32_EL2 },
 492         /* FPEXC32_EL2 */
 493         { Op0(0b11), Op1(0b100), CRn(0b0101), CRm(0b0011), Op2(0b000),
 494           NULL, reset_val, FPEXC32_EL2, 0x70 },
 495 };
 496
 497 static bool trap_dbgidr(struct kvm_vcpu *vcpu,
 498                         const struct sys_reg_params *p,
 499                         const struct sys_reg_desc *r)
 500 {
 501         if (p->is_write) {
 502                 return ignore_write(vcpu, p);
 503         } else {
 504                 u64 dfr = read_cpuid(ID_AA64DFR0_EL1);
 505                 u64 pfr = read_cpuid(ID_AA64PFR0_EL1);
 506                 u32 el3 = !!((pfr >> 12) & 0xf);
 507
 508                 *vcpu_reg(vcpu, p->Rt) = ((((dfr >> 20) & 0xf) << 28) |
 509                                           (((dfr >> 12) & 0xf) << 24) |
 510                                           (((dfr >> 28) & 0xf) << 20) |
 511                                           (6 << 16) | (el3 << 14) | (el3 << 12));
 512                 return true;
 513         }
 514 }
 515
 516 static bool trap_debug32(struct kvm_vcpu *vcpu,
 517                          const struct sys_reg_params *p,
 518                          const struct sys_reg_desc *r)
 519 {
 520         if (p->is_write) {
 521                 vcpu_cp14(vcpu, r->reg) = *vcpu_reg(vcpu, p->Rt);
 522                 vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
 523         } else {
 524                 *vcpu_reg(vcpu, p->Rt) = vcpu_cp14(vcpu, r->reg);
 525         }
 526
 527         return true;
 528 }
 529
 530 #define DBG_BCR_BVR_WCR_WVR(n)                                  \
 531         /* DBGBVRn */                                           \
 532         { Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_debug32,    \
 533           NULL, (cp14_DBGBVR0 + (n) * 2) },                     \
 534         /* DBGBCRn */                                           \
 535         { Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_debug32,    \
 536           NULL, (cp14_DBGBCR0 + (n) * 2) },                     \
 537         /* DBGWVRn */                                           \
 538         { Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_debug32,    \
 539           NULL, (cp14_DBGWVR0 + (n) * 2) },                     \
 540         /* DBGWCRn */                                           \
 541         { Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_debug32,    \
 542           NULL, (cp14_DBGWCR0 + (n) * 2) }
 543
 544 #define DBGBXVR(n)                                              \
 545         { Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_debug32,    \
 546           NULL, cp14_DBGBXVR0 + n * 2 }
 547
 548 /*
 549  * Trapped cp14 registers. We generally ignore most of the external
 550  * debug, on the principle that they don't really make sense to a
 551  * guest. Revisit this one day, whould this principle change.
 552  */
 553 static const struct sys_reg_desc cp14_regs[] = {
 554         /* DBGIDR */
 555         { Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgidr },
 556         /* DBGDTRRXext */
 557         { Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
 558
 559         DBG_BCR_BVR_WCR_WVR(0),
 560         /* DBGDSCRint */
 561         { Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
 562         DBG_BCR_BVR_WCR_WVR(1),
 563         /* DBGDCCINT */
 564         { Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug32 },
 565         /* DBGDSCRext */
 566         { Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug32 },
 567         DBG_BCR_BVR_WCR_WVR(2),
 568         /* DBGDTR[RT]Xint */
 569         { Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
 570         /* DBGDTR[RT]Xext */
 571         { Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
 572         DBG_BCR_BVR_WCR_WVR(3),
 573         DBG_BCR_BVR_WCR_WVR(4),
 574         DBG_BCR_BVR_WCR_WVR(5),
 575         /* DBGWFAR */
 576         { Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
 577         /* DBGOSECCR */
 578         { Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
 579         DBG_BCR_BVR_WCR_WVR(6),
 580         /* DBGVCR */
 581         { Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug32 },
 582         DBG_BCR_BVR_WCR_WVR(7),
 583         DBG_BCR_BVR_WCR_WVR(8),
 584         DBG_BCR_BVR_WCR_WVR(9),
 585         DBG_BCR_BVR_WCR_WVR(10),
 586         DBG_BCR_BVR_WCR_WVR(11),
 587         DBG_BCR_BVR_WCR_WVR(12),
 588         DBG_BCR_BVR_WCR_WVR(13),
 589         DBG_BCR_BVR_WCR_WVR(14),
 590         DBG_BCR_BVR_WCR_WVR(15),
 591
 592         /* DBGDRAR (32bit) */
 593         { Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
 594
 595         DBGBXVR(0),
 596         /* DBGOSLAR */
 597         { Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_raz_wi },
 598         DBGBXVR(1),
 599         /* DBGOSLSR */
 600         { Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1 },
 601         DBGBXVR(2),
 602         DBGBXVR(3),
 603         /* DBGOSDLR */
 604         { Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
 605         DBGBXVR(4),
 606         /* DBGPRCR */
 607         { Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
 608         DBGBXVR(5),
 609         DBGBXVR(6),
 610         DBGBXVR(7),
 611         DBGBXVR(8),
 612         DBGBXVR(9),
 613         DBGBXVR(10),
 614         DBGBXVR(11),
 615         DBGBXVR(12),
 616         DBGBXVR(13),
 617         DBGBXVR(14),
 618         DBGBXVR(15),
 619
 620         /* DBGDSAR (32bit) */
 621         { Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
 622
 623         /* DBGDEVID2 */
 624         { Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
 625         /* DBGDEVID1 */
 626         { Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
 627         /* DBGDEVID */
 628         { Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
 629         /* DBGCLAIMSET */
 630         { Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
 631         /* DBGCLAIMCLR */
 632         { Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
 633         /* DBGAUTHSTATUS */
 634         { Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
 635 };
 636
 637 /* Trapped cp14 64bit registers */
 638 static const struct sys_reg_desc cp14_64_regs[] = {
 639         /* DBGDRAR (64bit) */
 640         { Op1( 0), CRm( 1), .access = trap_raz_wi },
 641
 642         /* DBGDSAR (64bit) */
 643         { Op1( 0), CRm( 2), .access = trap_raz_wi },
 644 };
 645
 646 /*
 647  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
 648  * depending on the way they are accessed (as a 32bit or a 64bit
 649  * register).
 650  */
 651 static const struct sys_reg_desc cp15_regs[] = {
 652         { Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_sctlr, NULL, c1_SCTLR },
 653         { Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
 654         { Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, c2_TTBR1 },
 655         { Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, c2_TTBCR },
 656         { Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, c3_DACR },
 657         { Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, c5_DFSR },
 658         { Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, c5_IFSR },
 659         { Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, c5_ADFSR },
 660         { Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, c5_AIFSR },
 661         { Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, c6_DFAR },
 662         { Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, c6_IFAR },
 663
 664         /*
 665          * DC{C,I,CI}SW operations:
 666          */
 667         { Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
 668         { Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
 669         { Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
 670
 671         /* PMU */
 672         { Op1( 0), CRn( 9), CRm(12), Op2( 0), trap_raz_wi },
 673         { Op1( 0), CRn( 9), CRm(12), Op2( 1), trap_raz_wi },
 674         { Op1( 0), CRn( 9), CRm(12), Op2( 2), trap_raz_wi },
 675         { Op1( 0), CRn( 9), CRm(12), Op2( 3), trap_raz_wi },
 676         { Op1( 0), CRn( 9), CRm(12), Op2( 5), trap_raz_wi },
 677         { Op1( 0), CRn( 9), CRm(12), Op2( 6), trap_raz_wi },
 678         { Op1( 0), CRn( 9), CRm(12), Op2( 7), trap_raz_wi },
 679         { Op1( 0), CRn( 9), CRm(13), Op2( 0), trap_raz_wi },
 680         { Op1( 0), CRn( 9), CRm(13), Op2( 1), trap_raz_wi },
 681         { Op1( 0), CRn( 9), CRm(13), Op2( 2), trap_raz_wi },
 682         { Op1( 0), CRn( 9), CRm(14), Op2( 0), trap_raz_wi },
 683         { Op1( 0), CRn( 9), CRm(14), Op2( 1), trap_raz_wi },
 684         { Op1( 0), CRn( 9), CRm(14), Op2( 2), trap_raz_wi },
 685
 686         { Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, c10_PRRR },
 687         { Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, c10_NMRR },
 688         { Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, c10_AMAIR0 },
 689         { Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, c10_AMAIR1 },
 690         { Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, c13_CID },
 691 };
 692
 693 static const struct sys_reg_desc cp15_64_regs[] = {
 694         { Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
 695         { Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR1 },
 696 };
 697
 698 /* Target specific emulation tables */
 699 static struct kvm_sys_reg_target_table *target_tables[KVM_ARM_NUM_TARGETS];
 700
 701 void kvm_register_target_sys_reg_table(unsigned int target,
 702                                        struct kvm_sys_reg_target_table *table)
 703 {
 704         target_tables[target] = table;
 705 }
 706
 707 /* Get specific register table for this target. */
 708 static const struct sys_reg_desc *get_target_table(unsigned target,
 709                                                    bool mode_is_64,
 710                                                    size_t *num)
 711 {
 712         struct kvm_sys_reg_target_table *table;
 713
 714         table = target_tables[target];
 715         if (mode_is_64) {
 716                 *num = table->table64.num;
 717                 return table->table64.table;
 718         } else {
 719                 *num = table->table32.num;
 720                 return table->table32.table;
 721         }
 722 }
 723
 724 static const struct sys_reg_desc *find_reg(const struct sys_reg_params *params,
 725                                          const struct sys_reg_desc table[],
 726                                          unsigned int num)
 727 {
 728         unsigned int i;
 729
 730         for (i = 0; i < num; i++) {
 731                 const struct sys_reg_desc *r = &table[i];
 732
 733                 if (params->Op0 != r->Op0)
 734                         continue;
 735                 if (params->Op1 != r->Op1)
 736                         continue;
 737                 if (params->CRn != r->CRn)
 738                         continue;
 739                 if (params->CRm != r->CRm)
 740                         continue;
 741                 if (params->Op2 != r->Op2)
 742                         continue;
 743
 744                 return r;
 745         }
 746         return NULL;
 747 }
 748
 749 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu, struct kvm_run *run)
 750 {
 751         kvm_inject_undefined(vcpu);
 752         return 1;
 753 }
 754
 755 /*
 756  * emulate_cp --  tries to match a sys_reg access in a handling table, and
 757  *                call the corresponding trap handler.
 758  *
 759  * @params: pointer to the descriptor of the access
 760  * @table: array of trap descriptors
 761  * @num: size of the trap descriptor array
 762  *
 763  * Return 0 if the access has been handled, and -1 if not.
 764  */
 765 static int emulate_cp(struct kvm_vcpu *vcpu,
 766                       const struct sys_reg_params *params,
 767                       const struct sys_reg_desc *table,
 768                       size_t num)
 769 {
 770         const struct sys_reg_desc *r;
 771
 772         if (!table)
 773                 return -1;      /* Not handled */
 774
 775         r = find_reg(params, table, num);
 776
 777         if (r) {
 778                 /*
 779                  * Not having an accessor means that we have
 780                  * configured a trap that we don't know how to
 781                  * handle. This certainly qualifies as a gross bug
 782                  * that should be fixed right away.
 783                  */
 784                 BUG_ON(!r->access);
 785
 786                 if (likely(r->access(vcpu, params, r))) {
 787                         /* Skip instruction, since it was emulated */
 788                         kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
 789                 }
 790
 791                 /* Handled */
 792                 return 0;
 793         }
 794
 795         /* Not handled */
 796         return -1;
 797 }
 798
 799 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
 800                                 struct sys_reg_params *params)
 801 {
 802         u8 hsr_ec = kvm_vcpu_trap_get_class(vcpu);
 803         int cp;
 804
 805         switch(hsr_ec) {
 806         case ESR_EL2_EC_CP15_32:
 807         case ESR_EL2_EC_CP15_64:
 808                 cp = 15;
 809                 break;
 810         case ESR_EL2_EC_CP14_MR:
 811         case ESR_EL2_EC_CP14_64:
 812                 cp = 14;
 813                 break;
 814         default:
 815                 WARN_ON((cp = -1));
 816         }
 817
 818         kvm_err("Unsupported guest CP%d access at: %08lx\n",
 819                 cp, *vcpu_pc(vcpu));
 820         print_sys_reg_instr(params);
 821         kvm_inject_undefined(vcpu);
 822 }
 823
 824 /**
 825  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP15 access
 826  * @vcpu: The VCPU pointer
 827  * @run:  The kvm_run struct
 828  */
 829 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
 830                             const struct sys_reg_desc *global,
 831                             size_t nr_global,
 832                             const struct sys_reg_desc *target_specific,
 833                             size_t nr_specific)
 834 {
 835         struct sys_reg_params params;
 836         u32 hsr = kvm_vcpu_get_hsr(vcpu);
 837         int Rt2 = (hsr >> 10) & 0xf;
 838
 839         params.is_aarch32 = true;
 840         params.is_32bit = false;
 841         params.CRm = (hsr >> 1) & 0xf;
 842         params.Rt = (hsr >> 5) & 0xf;
 843         params.is_write = ((hsr & 1) == 0);
 844
 845         params.Op0 = 0;
 846         params.Op1 = (hsr >> 16) & 0xf;
 847         params.Op2 = 0;
 848         params.CRn = 0;
 849
 850         /*
 851          * Massive hack here. Store Rt2 in the top 32bits so we only
 852          * have one register to deal with. As we use the same trap
 853          * backends between AArch32 and AArch64, we get away with it.
 854          */
 855         if (params.is_write) {
 856                 u64 val = *vcpu_reg(vcpu, params.Rt);
 857                 val &= 0xffffffff;
 858                 val |= *vcpu_reg(vcpu, Rt2) << 32;
 859                 *vcpu_reg(vcpu, params.Rt) = val;
 860         }
 861
 862         if (!emulate_cp(vcpu, &params, target_specific, nr_specific))
 863                 goto out;
 864         if (!emulate_cp(vcpu, &params, global, nr_global))
 865                 goto out;
 866
 867         unhandled_cp_access(vcpu, &params);
 868
 869 out:
 870         /* Do the opposite hack for the read side */
 871         if (!params.is_write) {
 872                 u64 val = *vcpu_reg(vcpu, params.Rt);
 873                 val >>= 32;
 874                 *vcpu_reg(vcpu, Rt2) = val;
 875         }
 876
 877         return 1;
 878 }
 879
 880 /**
 881  * kvm_handle_cp15_32 -- handles a mrc/mcr trap on a guest CP15 access
 882  * @vcpu: The VCPU pointer
 883  * @run:  The kvm_run struct
 884  */
 885 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
 886                             const struct sys_reg_desc *global,
 887                             size_t nr_global,
 888                             const struct sys_reg_desc *target_specific,
 889                             size_t nr_specific)
 890 {
 891         struct sys_reg_params params;
 892         u32 hsr = kvm_vcpu_get_hsr(vcpu);
 893
 894         params.is_aarch32 = true;
 895         params.is_32bit = true;
 896         params.CRm = (hsr >> 1) & 0xf;
 897         params.Rt  = (hsr >> 5) & 0xf;
 898         params.is_write = ((hsr & 1) == 0);
 899         params.CRn = (hsr >> 10) & 0xf;
 900         params.Op0 = 0;
 901         params.Op1 = (hsr >> 14) & 0x7;
 902         params.Op2 = (hsr >> 17) & 0x7;
 903
 904         if (!emulate_cp(vcpu, &params, target_specific, nr_specific))
 905                 return 1;
 906         if (!emulate_cp(vcpu, &params, global, nr_global))
 907                 return 1;
 908
 909         unhandled_cp_access(vcpu, &params);
 910         return 1;
 911 }
 912
 913 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
 914 {
 915         const struct sys_reg_desc *target_specific;
 916         size_t num;
 917
 918         target_specific = get_target_table(vcpu->arch.target, false, &num);
 919         return kvm_handle_cp_64(vcpu,
 920                                 cp15_64_regs, ARRAY_SIZE(cp15_64_regs),
 921                                 target_specific, num);
 922 }
 923
 924 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
 925 {
 926         const struct sys_reg_desc *target_specific;
 927         size_t num;
 928
 929         target_specific = get_target_table(vcpu->arch.target, false, &num);
 930         return kvm_handle_cp_32(vcpu,
 931                                 cp15_regs, ARRAY_SIZE(cp15_regs),
 932                                 target_specific, num);
 933 }
 934
 935 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
 936 {
 937         return kvm_handle_cp_64(vcpu,
 938                                 cp14_64_regs, ARRAY_SIZE(cp14_64_regs),
 939                                 NULL, 0);
 940 }
 941
 942 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
 943 {
 944         return kvm_handle_cp_32(vcpu,
 945                                 cp14_regs, ARRAY_SIZE(cp14_regs),
 946                                 NULL, 0);
 947 }
 948
 949 static int emulate_sys_reg(struct kvm_vcpu *vcpu,
 950                            const struct sys_reg_params *params)
 951 {
 952         size_t num;
 953         const struct sys_reg_desc *table, *r;
 954
 955         table = get_target_table(vcpu->arch.target, true, &num);
 956
 957         /* Search target-specific then generic table. */
 958         r = find_reg(params, table, num);
 959         if (!r)
 960                 r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
 961
 962         if (likely(r)) {
 963                 /*
 964                  * Not having an accessor means that we have
 965                  * configured a trap that we don't know how to
 966                  * handle. This certainly qualifies as a gross bug
 967                  * that should be fixed right away.
 968                  */
 969                 BUG_ON(!r->access);
 970
 971                 if (likely(r->access(vcpu, params, r))) {
 972                         /* Skip instruction, since it was emulated */
 973                         kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
 974                         return 1;
 975                 }
 976                 /* If access function fails, it should complain. */
 977         } else {
 978                 kvm_err("Unsupported guest sys_reg access at: %lx\n",
 979                         *vcpu_pc(vcpu));
 980                 print_sys_reg_instr(params);
 981         }
 982         kvm_inject_undefined(vcpu);
 983         return 1;
 984 }
 985
 986 static void reset_sys_reg_descs(struct kvm_vcpu *vcpu,
 987                               const struct sys_reg_desc *table, size_t num)
 988 {
 989         unsigned long i;
 990
 991         for (i = 0; i < num; i++)
 992                 if (table[i].reset)
 993                         table[i].reset(vcpu, &table[i]);
 994 }
 995
 996 /**
 997  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
 998  * @vcpu: The VCPU pointer
 999  * @run:  The kvm_run struct
1000  */
1001 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu, struct kvm_run *run)
1002 {
1003         struct sys_reg_params params;
1004         unsigned long esr = kvm_vcpu_get_hsr(vcpu);
1005
1006         params.is_aarch32 = false;
1007         params.is_32bit = false;
1008         params.Op0 = (esr >> 20) & 3;
1009         params.Op1 = (esr >> 14) & 0x7;
1010         params.CRn = (esr >> 10) & 0xf;
1011         params.CRm = (esr >> 1) & 0xf;
1012         params.Op2 = (esr >> 17) & 0x7;
1013         params.Rt = (esr >> 5) & 0x1f;
1014         params.is_write = !(esr & 1);
1015
1016         return emulate_sys_reg(vcpu, &params);
1017 }
1018
1019 /******************************************************************************
1020  * Userspace API
1021  *****************************************************************************/
1022
1023 static bool index_to_params(u64 id, struct sys_reg_params *params)
1024 {
1025         switch (id & KVM_REG_SIZE_MASK) {
1026         case KVM_REG_SIZE_U64:
1027                 /* Any unused index bits means it's not valid. */
1028                 if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
1029                               | KVM_REG_ARM_COPROC_MASK
1030                               | KVM_REG_ARM64_SYSREG_OP0_MASK
1031                               | KVM_REG_ARM64_SYSREG_OP1_MASK
1032                               | KVM_REG_ARM64_SYSREG_CRN_MASK
1033                               | KVM_REG_ARM64_SYSREG_CRM_MASK
1034                               | KVM_REG_ARM64_SYSREG_OP2_MASK))
1035                         return false;
1036                 params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
1037                                >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
1038                 params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
1039                                >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
1040                 params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
1041                                >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
1042                 params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
1043                                >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
1044                 params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
1045                                >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
1046                 return true;
1047         default:
1048                 return false;
1049         }
1050 }
1051
1052 /* Decode an index value, and find the sys_reg_desc entry. */
1053 static const struct sys_reg_desc *index_to_sys_reg_desc(struct kvm_vcpu *vcpu,
1054                                                     u64 id)
1055 {
1056         size_t num;
1057         const struct sys_reg_desc *table, *r;
1058         struct sys_reg_params params;
1059
1060         /* We only do sys_reg for now. */
1061         if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
1062                 return NULL;
1063
1064         if (!index_to_params(id, &params))
1065                 return NULL;
1066
1067         table = get_target_table(vcpu->arch.target, true, &num);
1068         r = find_reg(&params, table, num);
1069         if (!r)
1070                 r = find_reg(&params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
1071
1072         /* Not saved in the sys_reg array? */
1073         if (r && !r->reg)
1074                 r = NULL;
1075
1076         return r;
1077 }
1078
1079 /*
1080  * These are the invariant sys_reg registers: we let the guest see the
1081  * host versions of these, so they're part of the guest state.
1082  *
1083  * A future CPU may provide a mechanism to present different values to
1084  * the guest, or a future kvm may trap them.
1085  */
1086
1087 #define FUNCTION_INVARIANT(reg)                                         \
1088         static void get_##reg(struct kvm_vcpu *v,                       \
1089                               const struct sys_reg_desc *r)             \
1090         {                                                               \
1091                 u64 val;                                                \
1092                                                                         \
1093                 asm volatile("mrs %0, " __stringify(reg) "\n"           \
1094                              : "=r" (val));                             \
1095                 ((struct sys_reg_desc *)r)->val = val;                  \
1096         }
1097
1098 FUNCTION_INVARIANT(midr_el1)
1099 FUNCTION_INVARIANT(ctr_el0)
1100 FUNCTION_INVARIANT(revidr_el1)
1101 FUNCTION_INVARIANT(id_pfr0_el1)
1102 FUNCTION_INVARIANT(id_pfr1_el1)
1103 FUNCTION_INVARIANT(id_dfr0_el1)
1104 FUNCTION_INVARIANT(id_afr0_el1)
1105 FUNCTION_INVARIANT(id_mmfr0_el1)
1106 FUNCTION_INVARIANT(id_mmfr1_el1)
1107 FUNCTION_INVARIANT(id_mmfr2_el1)
1108 FUNCTION_INVARIANT(id_mmfr3_el1)
1109 FUNCTION_INVARIANT(id_isar0_el1)
1110 FUNCTION_INVARIANT(id_isar1_el1)
1111 FUNCTION_INVARIANT(id_isar2_el1)
1112 FUNCTION_INVARIANT(id_isar3_el1)
1113 FUNCTION_INVARIANT(id_isar4_el1)
1114 FUNCTION_INVARIANT(id_isar5_el1)
1115 FUNCTION_INVARIANT(clidr_el1)
1116 FUNCTION_INVARIANT(aidr_el1)
1117
1118 /* ->val is filled in by kvm_sys_reg_table_init() */
1119 static struct sys_reg_desc invariant_sys_regs[] = {
1120         { Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b000),
1121           NULL, get_midr_el1 },
1122         { Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b110),
1123           NULL, get_revidr_el1 },
1124         { Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b000),
1125           NULL, get_id_pfr0_el1 },
1126         { Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b001),
1127           NULL, get_id_pfr1_el1 },
1128         { Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b010),
1129           NULL, get_id_dfr0_el1 },
1130         { Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b011),
1131           NULL, get_id_afr0_el1 },
1132         { Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b100),
1133           NULL, get_id_mmfr0_el1 },
1134         { Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b101),
1135           NULL, get_id_mmfr1_el1 },
1136         { Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b110),
1137           NULL, get_id_mmfr2_el1 },
1138         { Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b111),
1139           NULL, get_id_mmfr3_el1 },
1140         { Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b000),
1141           NULL, get_id_isar0_el1 },
1142         { Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b001),
1143           NULL, get_id_isar1_el1 },
1144         { Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b010),
1145           NULL, get_id_isar2_el1 },
1146         { Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b011),
1147           NULL, get_id_isar3_el1 },
1148         { Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b100),
1149           NULL, get_id_isar4_el1 },
1150         { Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b101),
1151           NULL, get_id_isar5_el1 },
1152         { Op0(0b11), Op1(0b001), CRn(0b0000), CRm(0b0000), Op2(0b001),
1153           NULL, get_clidr_el1 },
1154         { Op0(0b11), Op1(0b001), CRn(0b0000), CRm(0b0000), Op2(0b111),
1155           NULL, get_aidr_el1 },
1156         { Op0(0b11), Op1(0b011), CRn(0b0000), CRm(0b0000), Op2(0b001),
1157           NULL, get_ctr_el0 },
1158 };
1159
1160 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id)
1161 {
1162         if (copy_from_user(val, uaddr, KVM_REG_SIZE(id)) != 0)
1163                 return -EFAULT;
1164         return 0;
1165 }
1166
1167 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id)
1168 {
1169         if (copy_to_user(uaddr, val, KVM_REG_SIZE(id)) != 0)
1170                 return -EFAULT;
1171         return 0;
1172 }
1173
1174 static int get_invariant_sys_reg(u64 id, void __user *uaddr)
1175 {
1176         struct sys_reg_params params;
1177         const struct sys_reg_desc *r;
1178
1179         if (!index_to_params(id, &params))
1180                 return -ENOENT;
1181
1182         r = find_reg(&params, invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs));
1183         if (!r)
1184                 return -ENOENT;
1185
1186         return reg_to_user(uaddr, &r->val, id);
1187 }
1188
1189 static int set_invariant_sys_reg(u64 id, void __user *uaddr)
1190 {
1191         struct sys_reg_params params;
1192         const struct sys_reg_desc *r;
1193         int err;
1194         u64 val = 0; /* Make sure high bits are 0 for 32-bit regs */
1195
1196         if (!index_to_params(id, &params))
1197                 return -ENOENT;
1198         r = find_reg(&params, invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs));
1199         if (!r)
1200                 return -ENOENT;
1201
1202         err = reg_from_user(&val, uaddr, id);
1203         if (err)
1204                 return err;
1205
1206         /* This is what we mean by invariant: you can't change it. */
1207         if (r->val != val)
1208                 return -EINVAL;
1209
1210         return 0;
1211 }
1212
1213 static bool is_valid_cache(u32 val)
1214 {
1215         u32 level, ctype;
1216
1217         if (val >= CSSELR_MAX)
1218                 return -ENOENT;
1219
1220         /* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
1221         level = (val >> 1);
1222         ctype = (cache_levels >> (level * 3)) & 7;
1223
1224         switch (ctype) {
1225         case 0: /* No cache */
1226                 return false;
1227         case 1: /* Instruction cache only */
1228                 return (val & 1);
1229         case 2: /* Data cache only */
1230         case 4: /* Unified cache */
1231                 return !(val & 1);
1232         case 3: /* Separate instruction and data caches */
1233                 return true;
1234         default: /* Reserved: we can't know instruction or data. */
1235                 return false;
1236         }
1237 }
1238
1239 static int demux_c15_get(u64 id, void __user *uaddr)
1240 {
1241         u32 val;
1242         u32 __user *uval = uaddr;
1243
1244         /* Fail if we have unknown bits set. */
1245         if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
1246                    | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
1247                 return -ENOENT;
1248
1249         switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
1250         case KVM_REG_ARM_DEMUX_ID_CCSIDR:
1251                 if (KVM_REG_SIZE(id) != 4)
1252                         return -ENOENT;
1253                 val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
1254                         >> KVM_REG_ARM_DEMUX_VAL_SHIFT;
1255                 if (!is_valid_cache(val))
1256                         return -ENOENT;
1257
1258                 return put_user(get_ccsidr(val), uval);
1259         default:
1260                 return -ENOENT;
1261         }
1262 }
1263
1264 static int demux_c15_set(u64 id, void __user *uaddr)
1265 {
1266         u32 val, newval;
1267         u32 __user *uval = uaddr;
1268
1269         /* Fail if we have unknown bits set. */
1270         if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
1271                    | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
1272                 return -ENOENT;
1273
1274         switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
1275         case KVM_REG_ARM_DEMUX_ID_CCSIDR:
1276                 if (KVM_REG_SIZE(id) != 4)
1277                         return -ENOENT;
1278                 val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
1279                         >> KVM_REG_ARM_DEMUX_VAL_SHIFT;
1280                 if (!is_valid_cache(val))
1281                         return -ENOENT;
1282
1283                 if (get_user(newval, uval))
1284                         return -EFAULT;
1285
1286                 /* This is also invariant: you can't change it. */
1287                 if (newval != get_ccsidr(val))
1288                         return -EINVAL;
1289                 return 0;
1290         default:
1291                 return -ENOENT;
1292         }
1293 }
1294
1295 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
1296 {
1297         const struct sys_reg_desc *r;
1298         void __user *uaddr = (void __user *)(unsigned long)reg->addr;
1299
1300         if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
1301                 return demux_c15_get(reg->id, uaddr);
1302
1303         if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
1304                 return -ENOENT;
1305
1306         r = index_to_sys_reg_desc(vcpu, reg->id);
1307         if (!r)
1308                 return get_invariant_sys_reg(reg->id, uaddr);
1309
1310         return reg_to_user(uaddr, &vcpu_sys_reg(vcpu, r->reg), reg->id);
1311 }
1312
1313 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
1314 {
1315         const struct sys_reg_desc *r;
1316         void __user *uaddr = (void __user *)(unsigned long)reg->addr;
1317
1318         if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
1319                 return demux_c15_set(reg->id, uaddr);
1320
1321         if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
1322                 return -ENOENT;
1323
1324         r = index_to_sys_reg_desc(vcpu, reg->id);
1325         if (!r)
1326                 return set_invariant_sys_reg(reg->id, uaddr);
1327
1328         return reg_from_user(&vcpu_sys_reg(vcpu, r->reg), uaddr, reg->id);
1329 }
1330
1331 static unsigned int num_demux_regs(void)
1332 {
1333         unsigned int i, count = 0;
1334
1335         for (i = 0; i < CSSELR_MAX; i++)
1336                 if (is_valid_cache(i))
1337                         count++;
1338
1339         return count;
1340 }
1341
1342 static int write_demux_regids(u64 __user *uindices)
1343 {
1344         u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
1345         unsigned int i;
1346
1347         val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
1348         for (i = 0; i < CSSELR_MAX; i++) {
1349                 if (!is_valid_cache(i))
1350                         continue;
1351                 if (put_user(val | i, uindices))
1352                         return -EFAULT;
1353                 uindices++;
1354         }
1355         return 0;
1356 }
1357
1358 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
1359 {
1360         return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
1361                 KVM_REG_ARM64_SYSREG |
1362                 (reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
1363                 (reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
1364                 (reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
1365                 (reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
1366                 (reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
1367 }
1368
1369 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
1370 {
1371         if (!*uind)
1372                 return true;
1373
1374         if (put_user(sys_reg_to_index(reg), *uind))
1375                 return false;
1376
1377         (*uind)++;
1378         return true;
1379 }
1380
1381 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
1382 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
1383 {
1384         const struct sys_reg_desc *i1, *i2, *end1, *end2;
1385         unsigned int total = 0;
1386         size_t num;
1387
1388         /* We check for duplicates here, to allow arch-specific overrides. */
1389         i1 = get_target_table(vcpu->arch.target, true, &num);
1390         end1 = i1 + num;
1391         i2 = sys_reg_descs;
1392         end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
1393
1394         BUG_ON(i1 == end1 || i2 == end2);
1395
1396         /* Walk carefully, as both tables may refer to the same register. */
1397         while (i1 || i2) {
1398                 int cmp = cmp_sys_reg(i1, i2);
1399                 /* target-specific overrides generic entry. */
1400                 if (cmp <= 0) {
1401                         /* Ignore registers we trap but don't save. */
1402                         if (i1->reg) {
1403                                 if (!copy_reg_to_user(i1, &uind))
1404                                         return -EFAULT;
1405                                 total++;
1406                         }
1407                 } else {
1408                         /* Ignore registers we trap but don't save. */
1409                         if (i2->reg) {
1410                                 if (!copy_reg_to_user(i2, &uind))
1411                                         return -EFAULT;
1412                                 total++;
1413                         }
1414                 }
1415
1416                 if (cmp <= 0 && ++i1 == end1)
1417                         i1 = NULL;
1418                 if (cmp >= 0 && ++i2 == end2)
1419                         i2 = NULL;
1420         }
1421         return total;
1422 }
1423
1424 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
1425 {
1426         return ARRAY_SIZE(invariant_sys_regs)
1427                 + num_demux_regs()
1428                 + walk_sys_regs(vcpu, (u64 __user *)NULL);
1429 }
1430
1431 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
1432 {
1433         unsigned int i;
1434         int err;
1435
1436         /* Then give them all the invariant registers' indices. */
1437         for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
1438                 if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
1439                         return -EFAULT;
1440                 uindices++;
1441         }
1442
1443         err = walk_sys_regs(vcpu, uindices);
1444         if (err < 0)
1445                 return err;
1446         uindices += err;
1447
1448         return write_demux_regids(uindices);
1449 }
1450
1451 static int check_sysreg_table(const struct sys_reg_desc *table, unsigned int n)
1452 {
1453         unsigned int i;
1454
1455         for (i = 1; i < n; i++) {
1456                 if (cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
1457                         kvm_err("sys_reg table %p out of order (%d)\n", table, i - 1);
1458                         return 1;
1459                 }
1460         }
1461
1462         return 0;
1463 }
1464
1465 void kvm_sys_reg_table_init(void)
1466 {
1467         unsigned int i;
1468         struct sys_reg_desc clidr;
1469
1470         /* Make sure tables are unique and in order. */
1471         BUG_ON(check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs)));
1472         BUG_ON(check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs)));
1473         BUG_ON(check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs)));
1474         BUG_ON(check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs)));
1475         BUG_ON(check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs)));
1476         BUG_ON(check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs)));
1477
1478         /* We abuse the reset function to overwrite the table itself. */
1479         for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
1480                 invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
1481
1482         /*
1483          * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
1484          *
1485          *   If software reads the Cache Type fields from Ctype1
1486          *   upwards, once it has seen a value of 0b000, no caches
1487          *   exist at further-out levels of the hierarchy. So, for
1488          *   example, if Ctype3 is the first Cache Type field with a
1489          *   value of 0b000, the values of Ctype4 to Ctype7 must be
1490          *   ignored.
1491          */
1492         get_clidr_el1(NULL, &clidr); /* Ugly... */
1493         cache_levels = clidr.val;
1494         for (i = 0; i < 7; i++)
1495                 if (((cache_levels >> (i*3)) & 7) == 0)
1496                         break;
1497         /* Clear all higher bits. */
1498         cache_levels &= (1 << (i*3))-1;
1499 }
1500
1501 /**
1502  * kvm_reset_sys_regs - sets system registers to reset value
1503  * @vcpu: The VCPU pointer
1504  *
1505  * This function finds the right table above and sets the registers on the
1506  * virtual CPU struct to their architecturally defined reset values.
1507  */
1508 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
1509 {
1510         size_t num;
1511         const struct sys_reg_desc *table;
1512
1513         /* Catch someone adding a register without putting in reset entry. */
1514         memset(&vcpu->arch.ctxt.sys_regs, 0x42, sizeof(vcpu->arch.ctxt.sys_regs));
1515
1516         /* Generic chip reset first (so target could override). */
1517         reset_sys_reg_descs(vcpu, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
1518
1519         table = get_target_table(vcpu->arch.target, true, &num);
1520         reset_sys_reg_descs(vcpu, table, num);
1521
1522         for (num = 1; num < NR_SYS_REGS; num++)
1523                 if (vcpu_sys_reg(vcpu, num) == 0x4242424242424242)
1524                         panic("Didn't reset vcpu_sys_reg(%zi)", num);
1525 }