1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2015 - ARM Ltd 4 * Author: Marc Zyngier <marc.zyngier@arm.com> 5 */ 6 7 #include <hyp/switch.h> 8 #include <hyp/sysreg-sr.h> 9 10 #include <linux/arm-smccc.h> 11 #include <linux/kvm_host.h> 12 #include <linux/types.h> 13 #include <linux/jump_label.h> 14 #include <uapi/linux/psci.h> 15 16 #include <kvm/arm_psci.h> 17 18 #include <asm/barrier.h> 19 #include <asm/cpufeature.h> 20 #include <asm/kprobes.h> 21 #include <asm/kvm_asm.h> 22 #include <asm/kvm_emulate.h> 23 #include <asm/kvm_hyp.h> 24 #include <asm/kvm_mmu.h> 25 #include <asm/fpsimd.h> 26 #include <asm/debug-monitors.h> 27 #include <asm/processor.h> 28 29 #include <nvhe/fixed_config.h> 30 #include <nvhe/mem_protect.h> 31 32 /* Non-VHE specific context */ 33 DEFINE_PER_CPU(struct kvm_host_data, kvm_host_data); 34 DEFINE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt); 35 DEFINE_PER_CPU(unsigned long, kvm_hyp_vector); 36 37 extern void kvm_nvhe_prepare_backtrace(unsigned long fp, unsigned long pc); 38 39 static void __activate_traps(struct kvm_vcpu *vcpu) 40 { 41 u64 val; 42 43 ___activate_traps(vcpu); 44 __activate_traps_common(vcpu); 45 46 val = vcpu->arch.cptr_el2; 47 val |= CPTR_EL2_TTA | CPTR_EL2_TAM; 48 if (!guest_owns_fp_regs(vcpu)) { 49 val |= CPTR_EL2_TFP | CPTR_EL2_TZ; 50 __activate_traps_fpsimd32(vcpu); 51 } 52 if (cpus_have_final_cap(ARM64_SME)) 53 val |= CPTR_EL2_TSM; 54 55 write_sysreg(val, cptr_el2); 56 write_sysreg(__this_cpu_read(kvm_hyp_vector), vbar_el2); 57 58 if (cpus_have_final_cap(ARM64_SME)) { 59 val = read_sysreg_s(SYS_HFGRTR_EL2); 60 val &= ~(HFGxTR_EL2_nTPIDR2_EL0_MASK | 61 HFGxTR_EL2_nSMPRI_EL1_MASK); 62 write_sysreg_s(val, SYS_HFGRTR_EL2); 63 64 val = read_sysreg_s(SYS_HFGWTR_EL2); 65 val &= ~(HFGxTR_EL2_nTPIDR2_EL0_MASK | 66 HFGxTR_EL2_nSMPRI_EL1_MASK); 67 write_sysreg_s(val, SYS_HFGWTR_EL2); 68 } 69 70 if (cpus_have_final_cap(ARM64_WORKAROUND_SPECULATIVE_AT)) { 71 struct kvm_cpu_context *ctxt = &vcpu->arch.ctxt; 72 73 isb(); 74 /* 75 * At this stage, and thanks to the above isb(), S2 is 76 * configured and enabled. We can now restore the guest's S1 77 * configuration: SCTLR, and only then TCR. 78 */ 79 write_sysreg_el1(ctxt_sys_reg(ctxt, SCTLR_EL1), SYS_SCTLR); 80 isb(); 81 write_sysreg_el1(ctxt_sys_reg(ctxt, TCR_EL1), SYS_TCR); 82 } 83 } 84 85 static void __deactivate_traps(struct kvm_vcpu *vcpu) 86 { 87 extern char __kvm_hyp_host_vector[]; 88 u64 cptr; 89 90 ___deactivate_traps(vcpu); 91 92 if (cpus_have_final_cap(ARM64_WORKAROUND_SPECULATIVE_AT)) { 93 u64 val; 94 95 /* 96 * Set the TCR and SCTLR registers in the exact opposite 97 * sequence as __activate_traps (first prevent walks, 98 * then force the MMU on). A generous sprinkling of isb() 99 * ensure that things happen in this exact order. 100 */ 101 val = read_sysreg_el1(SYS_TCR); 102 write_sysreg_el1(val | TCR_EPD1_MASK | TCR_EPD0_MASK, SYS_TCR); 103 isb(); 104 val = read_sysreg_el1(SYS_SCTLR); 105 write_sysreg_el1(val | SCTLR_ELx_M, SYS_SCTLR); 106 isb(); 107 } 108 109 __deactivate_traps_common(vcpu); 110 111 write_sysreg(this_cpu_ptr(&kvm_init_params)->hcr_el2, hcr_el2); 112 113 if (cpus_have_final_cap(ARM64_SME)) { 114 u64 val; 115 116 val = read_sysreg_s(SYS_HFGRTR_EL2); 117 val |= HFGxTR_EL2_nTPIDR2_EL0_MASK | 118 HFGxTR_EL2_nSMPRI_EL1_MASK; 119 write_sysreg_s(val, SYS_HFGRTR_EL2); 120 121 val = read_sysreg_s(SYS_HFGWTR_EL2); 122 val |= HFGxTR_EL2_nTPIDR2_EL0_MASK | 123 HFGxTR_EL2_nSMPRI_EL1_MASK; 124 write_sysreg_s(val, SYS_HFGWTR_EL2); 125 } 126 127 cptr = CPTR_EL2_DEFAULT; 128 if (vcpu_has_sve(vcpu) && (vcpu->arch.fp_state == FP_STATE_GUEST_OWNED)) 129 cptr |= CPTR_EL2_TZ; 130 if (cpus_have_final_cap(ARM64_SME)) 131 cptr &= ~CPTR_EL2_TSM; 132 133 write_sysreg(cptr, cptr_el2); 134 write_sysreg(__kvm_hyp_host_vector, vbar_el2); 135 } 136 137 /* Save VGICv3 state on non-VHE systems */ 138 static void __hyp_vgic_save_state(struct kvm_vcpu *vcpu) 139 { 140 if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif)) { 141 __vgic_v3_save_state(&vcpu->arch.vgic_cpu.vgic_v3); 142 __vgic_v3_deactivate_traps(&vcpu->arch.vgic_cpu.vgic_v3); 143 } 144 } 145 146 /* Restore VGICv3 state on non-VHE systems */ 147 static void __hyp_vgic_restore_state(struct kvm_vcpu *vcpu) 148 { 149 if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif)) { 150 __vgic_v3_activate_traps(&vcpu->arch.vgic_cpu.vgic_v3); 151 __vgic_v3_restore_state(&vcpu->arch.vgic_cpu.vgic_v3); 152 } 153 } 154 155 /* 156 * Disable host events, enable guest events 157 */ 158 #ifdef CONFIG_HW_PERF_EVENTS 159 static bool __pmu_switch_to_guest(struct kvm_vcpu *vcpu) 160 { 161 struct kvm_pmu_events *pmu = &vcpu->arch.pmu.events; 162 163 if (pmu->events_host) 164 write_sysreg(pmu->events_host, pmcntenclr_el0); 165 166 if (pmu->events_guest) 167 write_sysreg(pmu->events_guest, pmcntenset_el0); 168 169 return (pmu->events_host || pmu->events_guest); 170 } 171 172 /* 173 * Disable guest events, enable host events 174 */ 175 static void __pmu_switch_to_host(struct kvm_vcpu *vcpu) 176 { 177 struct kvm_pmu_events *pmu = &vcpu->arch.pmu.events; 178 179 if (pmu->events_guest) 180 write_sysreg(pmu->events_guest, pmcntenclr_el0); 181 182 if (pmu->events_host) 183 write_sysreg(pmu->events_host, pmcntenset_el0); 184 } 185 #else 186 #define __pmu_switch_to_guest(v) ({ false; }) 187 #define __pmu_switch_to_host(v) do {} while (0) 188 #endif 189 190 /* 191 * Handler for protected VM MSR, MRS or System instruction execution in AArch64. 192 * 193 * Returns true if the hypervisor has handled the exit, and control should go 194 * back to the guest, or false if it hasn't. 195 */ 196 static bool kvm_handle_pvm_sys64(struct kvm_vcpu *vcpu, u64 *exit_code) 197 { 198 /* 199 * Make sure we handle the exit for workarounds and ptrauth 200 * before the pKVM handling, as the latter could decide to 201 * UNDEF. 202 */ 203 return (kvm_hyp_handle_sysreg(vcpu, exit_code) || 204 kvm_handle_pvm_sysreg(vcpu, exit_code)); 205 } 206 207 static const exit_handler_fn hyp_exit_handlers[] = { 208 [0 ... ESR_ELx_EC_MAX] = NULL, 209 [ESR_ELx_EC_CP15_32] = kvm_hyp_handle_cp15_32, 210 [ESR_ELx_EC_SYS64] = kvm_hyp_handle_sysreg, 211 [ESR_ELx_EC_SVE] = kvm_hyp_handle_fpsimd, 212 [ESR_ELx_EC_FP_ASIMD] = kvm_hyp_handle_fpsimd, 213 [ESR_ELx_EC_IABT_LOW] = kvm_hyp_handle_iabt_low, 214 [ESR_ELx_EC_DABT_LOW] = kvm_hyp_handle_dabt_low, 215 [ESR_ELx_EC_PAC] = kvm_hyp_handle_ptrauth, 216 }; 217 218 static const exit_handler_fn pvm_exit_handlers[] = { 219 [0 ... ESR_ELx_EC_MAX] = NULL, 220 [ESR_ELx_EC_SYS64] = kvm_handle_pvm_sys64, 221 [ESR_ELx_EC_SVE] = kvm_handle_pvm_restricted, 222 [ESR_ELx_EC_FP_ASIMD] = kvm_hyp_handle_fpsimd, 223 [ESR_ELx_EC_IABT_LOW] = kvm_hyp_handle_iabt_low, 224 [ESR_ELx_EC_DABT_LOW] = kvm_hyp_handle_dabt_low, 225 [ESR_ELx_EC_PAC] = kvm_hyp_handle_ptrauth, 226 }; 227 228 static const exit_handler_fn *kvm_get_exit_handler_array(struct kvm_vcpu *vcpu) 229 { 230 if (unlikely(kvm_vm_is_protected(kern_hyp_va(vcpu->kvm)))) 231 return pvm_exit_handlers; 232 233 return hyp_exit_handlers; 234 } 235 236 /* 237 * Some guests (e.g., protected VMs) are not be allowed to run in AArch32. 238 * The ARMv8 architecture does not give the hypervisor a mechanism to prevent a 239 * guest from dropping to AArch32 EL0 if implemented by the CPU. If the 240 * hypervisor spots a guest in such a state ensure it is handled, and don't 241 * trust the host to spot or fix it. The check below is based on the one in 242 * kvm_arch_vcpu_ioctl_run(). 243 * 244 * Returns false if the guest ran in AArch32 when it shouldn't have, and 245 * thus should exit to the host, or true if a the guest run loop can continue. 246 */ 247 static void early_exit_filter(struct kvm_vcpu *vcpu, u64 *exit_code) 248 { 249 struct kvm *kvm = kern_hyp_va(vcpu->kvm); 250 251 if (kvm_vm_is_protected(kvm) && vcpu_mode_is_32bit(vcpu)) { 252 /* 253 * As we have caught the guest red-handed, decide that it isn't 254 * fit for purpose anymore by making the vcpu invalid. The VMM 255 * can try and fix it by re-initializing the vcpu with 256 * KVM_ARM_VCPU_INIT, however, this is likely not possible for 257 * protected VMs. 258 */ 259 vcpu->arch.target = -1; 260 *exit_code &= BIT(ARM_EXIT_WITH_SERROR_BIT); 261 *exit_code |= ARM_EXCEPTION_IL; 262 } 263 } 264 265 /* Switch to the guest for legacy non-VHE systems */ 266 int __kvm_vcpu_run(struct kvm_vcpu *vcpu) 267 { 268 struct kvm_cpu_context *host_ctxt; 269 struct kvm_cpu_context *guest_ctxt; 270 struct kvm_s2_mmu *mmu; 271 bool pmu_switch_needed; 272 u64 exit_code; 273 274 /* 275 * Having IRQs masked via PMR when entering the guest means the GIC 276 * will not signal the CPU of interrupts of lower priority, and the 277 * only way to get out will be via guest exceptions. 278 * Naturally, we want to avoid this. 279 */ 280 if (system_uses_irq_prio_masking()) { 281 gic_write_pmr(GIC_PRIO_IRQON | GIC_PRIO_PSR_I_SET); 282 pmr_sync(); 283 } 284 285 host_ctxt = &this_cpu_ptr(&kvm_host_data)->host_ctxt; 286 host_ctxt->__hyp_running_vcpu = vcpu; 287 guest_ctxt = &vcpu->arch.ctxt; 288 289 pmu_switch_needed = __pmu_switch_to_guest(vcpu); 290 291 __sysreg_save_state_nvhe(host_ctxt); 292 /* 293 * We must flush and disable the SPE buffer for nVHE, as 294 * the translation regime(EL1&0) is going to be loaded with 295 * that of the guest. And we must do this before we change the 296 * translation regime to EL2 (via MDCR_EL2_E2PB == 0) and 297 * before we load guest Stage1. 298 */ 299 __debug_save_host_buffers_nvhe(vcpu); 300 301 __kvm_adjust_pc(vcpu); 302 303 /* 304 * We must restore the 32-bit state before the sysregs, thanks 305 * to erratum #852523 (Cortex-A57) or #853709 (Cortex-A72). 306 * 307 * Also, and in order to be able to deal with erratum #1319537 (A57) 308 * and #1319367 (A72), we must ensure that all VM-related sysreg are 309 * restored before we enable S2 translation. 310 */ 311 __sysreg32_restore_state(vcpu); 312 __sysreg_restore_state_nvhe(guest_ctxt); 313 314 mmu = kern_hyp_va(vcpu->arch.hw_mmu); 315 __load_stage2(mmu, kern_hyp_va(mmu->arch)); 316 __activate_traps(vcpu); 317 318 __hyp_vgic_restore_state(vcpu); 319 __timer_enable_traps(vcpu); 320 321 __debug_switch_to_guest(vcpu); 322 323 do { 324 /* Jump in the fire! */ 325 exit_code = __guest_enter(vcpu); 326 327 /* And we're baaack! */ 328 } while (fixup_guest_exit(vcpu, &exit_code)); 329 330 __sysreg_save_state_nvhe(guest_ctxt); 331 __sysreg32_save_state(vcpu); 332 __timer_disable_traps(vcpu); 333 __hyp_vgic_save_state(vcpu); 334 335 __deactivate_traps(vcpu); 336 __load_host_stage2(); 337 338 __sysreg_restore_state_nvhe(host_ctxt); 339 340 if (vcpu->arch.fp_state == FP_STATE_GUEST_OWNED) 341 __fpsimd_save_fpexc32(vcpu); 342 343 __debug_switch_to_host(vcpu); 344 /* 345 * This must come after restoring the host sysregs, since a non-VHE 346 * system may enable SPE here and make use of the TTBRs. 347 */ 348 __debug_restore_host_buffers_nvhe(vcpu); 349 350 if (pmu_switch_needed) 351 __pmu_switch_to_host(vcpu); 352 353 /* Returning to host will clear PSR.I, remask PMR if needed */ 354 if (system_uses_irq_prio_masking()) 355 gic_write_pmr(GIC_PRIO_IRQOFF); 356 357 host_ctxt->__hyp_running_vcpu = NULL; 358 359 return exit_code; 360 } 361 362 asmlinkage void __noreturn hyp_panic(void) 363 { 364 u64 spsr = read_sysreg_el2(SYS_SPSR); 365 u64 elr = read_sysreg_el2(SYS_ELR); 366 u64 par = read_sysreg_par(); 367 struct kvm_cpu_context *host_ctxt; 368 struct kvm_vcpu *vcpu; 369 370 host_ctxt = &this_cpu_ptr(&kvm_host_data)->host_ctxt; 371 vcpu = host_ctxt->__hyp_running_vcpu; 372 373 if (vcpu) { 374 __timer_disable_traps(vcpu); 375 __deactivate_traps(vcpu); 376 __load_host_stage2(); 377 __sysreg_restore_state_nvhe(host_ctxt); 378 } 379 380 /* Prepare to dump kvm nvhe hyp stacktrace */ 381 kvm_nvhe_prepare_backtrace((unsigned long)__builtin_frame_address(0), 382 _THIS_IP_); 383 384 __hyp_do_panic(host_ctxt, spsr, elr, par); 385 unreachable(); 386 } 387 388 asmlinkage void __noreturn hyp_panic_bad_stack(void) 389 { 390 hyp_panic(); 391 } 392 393 asmlinkage void kvm_unexpected_el2_exception(void) 394 { 395 __kvm_unexpected_el2_exception(); 396 } 397