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/handle_exit.c: 7 * Copyright (C) 2012 - Virtual Open Systems and Columbia University 8 * Author: Christoffer Dall <c.dall@virtualopensystems.com> 9 */ 10 11 #include <linux/kvm.h> 12 #include <linux/kvm_host.h> 13 14 #include <asm/esr.h> 15 #include <asm/exception.h> 16 #include <asm/kvm_asm.h> 17 #include <asm/kvm_emulate.h> 18 #include <asm/kvm_mmu.h> 19 #include <asm/debug-monitors.h> 20 #include <asm/stacktrace/nvhe.h> 21 #include <asm/traps.h> 22 23 #include <kvm/arm_hypercalls.h> 24 25 #define CREATE_TRACE_POINTS 26 #include "trace_handle_exit.h" 27 28 typedef int (*exit_handle_fn)(struct kvm_vcpu *); 29 30 static void kvm_handle_guest_serror(struct kvm_vcpu *vcpu, u64 esr) 31 { 32 if (!arm64_is_ras_serror(esr) || arm64_is_fatal_ras_serror(NULL, esr)) 33 kvm_inject_vabt(vcpu); 34 } 35 36 static int handle_hvc(struct kvm_vcpu *vcpu) 37 { 38 int ret; 39 40 trace_kvm_hvc_arm64(*vcpu_pc(vcpu), vcpu_get_reg(vcpu, 0), 41 kvm_vcpu_hvc_get_imm(vcpu)); 42 vcpu->stat.hvc_exit_stat++; 43 44 ret = kvm_hvc_call_handler(vcpu); 45 if (ret < 0) { 46 vcpu_set_reg(vcpu, 0, ~0UL); 47 return 1; 48 } 49 50 return ret; 51 } 52 53 static int handle_smc(struct kvm_vcpu *vcpu) 54 { 55 /* 56 * "If an SMC instruction executed at Non-secure EL1 is 57 * trapped to EL2 because HCR_EL2.TSC is 1, the exception is a 58 * Trap exception, not a Secure Monitor Call exception [...]" 59 * 60 * We need to advance the PC after the trap, as it would 61 * otherwise return to the same address... 62 */ 63 vcpu_set_reg(vcpu, 0, ~0UL); 64 kvm_incr_pc(vcpu); 65 return 1; 66 } 67 68 /* 69 * Guest access to FP/ASIMD registers are routed to this handler only 70 * when the system doesn't support FP/ASIMD. 71 */ 72 static int handle_no_fpsimd(struct kvm_vcpu *vcpu) 73 { 74 kvm_inject_undefined(vcpu); 75 return 1; 76 } 77 78 /** 79 * kvm_handle_wfx - handle a wait-for-interrupts or wait-for-event 80 * instruction executed by a guest 81 * 82 * @vcpu: the vcpu pointer 83 * 84 * WFE[T]: Yield the CPU and come back to this vcpu when the scheduler 85 * decides to. 86 * WFI: Simply call kvm_vcpu_halt(), which will halt execution of 87 * world-switches and schedule other host processes until there is an 88 * incoming IRQ or FIQ to the VM. 89 * WFIT: Same as WFI, with a timed wakeup implemented as a background timer 90 * 91 * WF{I,E}T can immediately return if the deadline has already expired. 92 */ 93 static int kvm_handle_wfx(struct kvm_vcpu *vcpu) 94 { 95 u64 esr = kvm_vcpu_get_esr(vcpu); 96 97 if (esr & ESR_ELx_WFx_ISS_WFE) { 98 trace_kvm_wfx_arm64(*vcpu_pc(vcpu), true); 99 vcpu->stat.wfe_exit_stat++; 100 } else { 101 trace_kvm_wfx_arm64(*vcpu_pc(vcpu), false); 102 vcpu->stat.wfi_exit_stat++; 103 } 104 105 if (esr & ESR_ELx_WFx_ISS_WFxT) { 106 if (esr & ESR_ELx_WFx_ISS_RV) { 107 u64 val, now; 108 109 now = kvm_arm_timer_get_reg(vcpu, KVM_REG_ARM_TIMER_CNT); 110 val = vcpu_get_reg(vcpu, kvm_vcpu_sys_get_rt(vcpu)); 111 112 if (now >= val) 113 goto out; 114 } else { 115 /* Treat WFxT as WFx if RN is invalid */ 116 esr &= ~ESR_ELx_WFx_ISS_WFxT; 117 } 118 } 119 120 if (esr & ESR_ELx_WFx_ISS_WFE) { 121 kvm_vcpu_on_spin(vcpu, vcpu_mode_priv(vcpu)); 122 } else { 123 if (esr & ESR_ELx_WFx_ISS_WFxT) 124 vcpu_set_flag(vcpu, IN_WFIT); 125 126 kvm_vcpu_wfi(vcpu); 127 } 128 out: 129 kvm_incr_pc(vcpu); 130 131 return 1; 132 } 133 134 /** 135 * kvm_handle_guest_debug - handle a debug exception instruction 136 * 137 * @vcpu: the vcpu pointer 138 * 139 * We route all debug exceptions through the same handler. If both the 140 * guest and host are using the same debug facilities it will be up to 141 * userspace to re-inject the correct exception for guest delivery. 142 * 143 * @return: 0 (while setting vcpu->run->exit_reason) 144 */ 145 static int kvm_handle_guest_debug(struct kvm_vcpu *vcpu) 146 { 147 struct kvm_run *run = vcpu->run; 148 u64 esr = kvm_vcpu_get_esr(vcpu); 149 150 run->exit_reason = KVM_EXIT_DEBUG; 151 run->debug.arch.hsr = lower_32_bits(esr); 152 run->debug.arch.hsr_high = upper_32_bits(esr); 153 run->flags = KVM_DEBUG_ARCH_HSR_HIGH_VALID; 154 155 switch (ESR_ELx_EC(esr)) { 156 case ESR_ELx_EC_WATCHPT_LOW: 157 run->debug.arch.far = vcpu->arch.fault.far_el2; 158 break; 159 case ESR_ELx_EC_SOFTSTP_LOW: 160 vcpu_clear_flag(vcpu, DBG_SS_ACTIVE_PENDING); 161 break; 162 } 163 164 return 0; 165 } 166 167 static int kvm_handle_unknown_ec(struct kvm_vcpu *vcpu) 168 { 169 u64 esr = kvm_vcpu_get_esr(vcpu); 170 171 kvm_pr_unimpl("Unknown exception class: esr: %#016llx -- %s\n", 172 esr, esr_get_class_string(esr)); 173 174 kvm_inject_undefined(vcpu); 175 return 1; 176 } 177 178 /* 179 * Guest access to SVE registers should be routed to this handler only 180 * when the system doesn't support SVE. 181 */ 182 static int handle_sve(struct kvm_vcpu *vcpu) 183 { 184 kvm_inject_undefined(vcpu); 185 return 1; 186 } 187 188 /* 189 * Guest usage of a ptrauth instruction (which the guest EL1 did not turn into 190 * a NOP). If we get here, it is that we didn't fixup ptrauth on exit, and all 191 * that we can do is give the guest an UNDEF. 192 */ 193 static int kvm_handle_ptrauth(struct kvm_vcpu *vcpu) 194 { 195 kvm_inject_undefined(vcpu); 196 return 1; 197 } 198 199 static exit_handle_fn arm_exit_handlers[] = { 200 [0 ... ESR_ELx_EC_MAX] = kvm_handle_unknown_ec, 201 [ESR_ELx_EC_WFx] = kvm_handle_wfx, 202 [ESR_ELx_EC_CP15_32] = kvm_handle_cp15_32, 203 [ESR_ELx_EC_CP15_64] = kvm_handle_cp15_64, 204 [ESR_ELx_EC_CP14_MR] = kvm_handle_cp14_32, 205 [ESR_ELx_EC_CP14_LS] = kvm_handle_cp14_load_store, 206 [ESR_ELx_EC_CP10_ID] = kvm_handle_cp10_id, 207 [ESR_ELx_EC_CP14_64] = kvm_handle_cp14_64, 208 [ESR_ELx_EC_HVC32] = handle_hvc, 209 [ESR_ELx_EC_SMC32] = handle_smc, 210 [ESR_ELx_EC_HVC64] = handle_hvc, 211 [ESR_ELx_EC_SMC64] = handle_smc, 212 [ESR_ELx_EC_SYS64] = kvm_handle_sys_reg, 213 [ESR_ELx_EC_SVE] = handle_sve, 214 [ESR_ELx_EC_IABT_LOW] = kvm_handle_guest_abort, 215 [ESR_ELx_EC_DABT_LOW] = kvm_handle_guest_abort, 216 [ESR_ELx_EC_SOFTSTP_LOW]= kvm_handle_guest_debug, 217 [ESR_ELx_EC_WATCHPT_LOW]= kvm_handle_guest_debug, 218 [ESR_ELx_EC_BREAKPT_LOW]= kvm_handle_guest_debug, 219 [ESR_ELx_EC_BKPT32] = kvm_handle_guest_debug, 220 [ESR_ELx_EC_BRK64] = kvm_handle_guest_debug, 221 [ESR_ELx_EC_FP_ASIMD] = handle_no_fpsimd, 222 [ESR_ELx_EC_PAC] = kvm_handle_ptrauth, 223 }; 224 225 static exit_handle_fn kvm_get_exit_handler(struct kvm_vcpu *vcpu) 226 { 227 u64 esr = kvm_vcpu_get_esr(vcpu); 228 u8 esr_ec = ESR_ELx_EC(esr); 229 230 return arm_exit_handlers[esr_ec]; 231 } 232 233 /* 234 * We may be single-stepping an emulated instruction. If the emulation 235 * has been completed in the kernel, we can return to userspace with a 236 * KVM_EXIT_DEBUG, otherwise userspace needs to complete its 237 * emulation first. 238 */ 239 static int handle_trap_exceptions(struct kvm_vcpu *vcpu) 240 { 241 int handled; 242 243 /* 244 * See ARM ARM B1.14.1: "Hyp traps on instructions 245 * that fail their condition code check" 246 */ 247 if (!kvm_condition_valid(vcpu)) { 248 kvm_incr_pc(vcpu); 249 handled = 1; 250 } else { 251 exit_handle_fn exit_handler; 252 253 exit_handler = kvm_get_exit_handler(vcpu); 254 handled = exit_handler(vcpu); 255 } 256 257 return handled; 258 } 259 260 /* 261 * Return > 0 to return to guest, < 0 on error, 0 (and set exit_reason) on 262 * proper exit to userspace. 263 */ 264 int handle_exit(struct kvm_vcpu *vcpu, int exception_index) 265 { 266 struct kvm_run *run = vcpu->run; 267 268 if (ARM_SERROR_PENDING(exception_index)) { 269 /* 270 * The SError is handled by handle_exit_early(). If the guest 271 * survives it will re-execute the original instruction. 272 */ 273 return 1; 274 } 275 276 exception_index = ARM_EXCEPTION_CODE(exception_index); 277 278 switch (exception_index) { 279 case ARM_EXCEPTION_IRQ: 280 return 1; 281 case ARM_EXCEPTION_EL1_SERROR: 282 return 1; 283 case ARM_EXCEPTION_TRAP: 284 return handle_trap_exceptions(vcpu); 285 case ARM_EXCEPTION_HYP_GONE: 286 /* 287 * EL2 has been reset to the hyp-stub. This happens when a guest 288 * is pre-emptied by kvm_reboot()'s shutdown call. 289 */ 290 run->exit_reason = KVM_EXIT_FAIL_ENTRY; 291 return 0; 292 case ARM_EXCEPTION_IL: 293 /* 294 * We attempted an illegal exception return. Guest state must 295 * have been corrupted somehow. Give up. 296 */ 297 run->exit_reason = KVM_EXIT_FAIL_ENTRY; 298 return -EINVAL; 299 default: 300 kvm_pr_unimpl("Unsupported exception type: %d", 301 exception_index); 302 run->exit_reason = KVM_EXIT_INTERNAL_ERROR; 303 return 0; 304 } 305 } 306 307 /* For exit types that need handling before we can be preempted */ 308 void handle_exit_early(struct kvm_vcpu *vcpu, int exception_index) 309 { 310 if (ARM_SERROR_PENDING(exception_index)) { 311 if (this_cpu_has_cap(ARM64_HAS_RAS_EXTN)) { 312 u64 disr = kvm_vcpu_get_disr(vcpu); 313 314 kvm_handle_guest_serror(vcpu, disr_to_esr(disr)); 315 } else { 316 kvm_inject_vabt(vcpu); 317 } 318 319 return; 320 } 321 322 exception_index = ARM_EXCEPTION_CODE(exception_index); 323 324 if (exception_index == ARM_EXCEPTION_EL1_SERROR) 325 kvm_handle_guest_serror(vcpu, kvm_vcpu_get_esr(vcpu)); 326 } 327 328 void __noreturn __cold nvhe_hyp_panic_handler(u64 esr, u64 spsr, 329 u64 elr_virt, u64 elr_phys, 330 u64 par, uintptr_t vcpu, 331 u64 far, u64 hpfar) { 332 u64 elr_in_kimg = __phys_to_kimg(elr_phys); 333 u64 hyp_offset = elr_in_kimg - kaslr_offset() - elr_virt; 334 u64 mode = spsr & PSR_MODE_MASK; 335 u64 panic_addr = elr_virt + hyp_offset; 336 337 if (mode != PSR_MODE_EL2t && mode != PSR_MODE_EL2h) { 338 kvm_err("Invalid host exception to nVHE hyp!\n"); 339 } else if (ESR_ELx_EC(esr) == ESR_ELx_EC_BRK64 && 340 (esr & ESR_ELx_BRK64_ISS_COMMENT_MASK) == BUG_BRK_IMM) { 341 const char *file = NULL; 342 unsigned int line = 0; 343 344 /* All hyp bugs, including warnings, are treated as fatal. */ 345 if (!is_protected_kvm_enabled() || 346 IS_ENABLED(CONFIG_NVHE_EL2_DEBUG)) { 347 struct bug_entry *bug = find_bug(elr_in_kimg); 348 349 if (bug) 350 bug_get_file_line(bug, &file, &line); 351 } 352 353 if (file) 354 kvm_err("nVHE hyp BUG at: %s:%u!\n", file, line); 355 else 356 kvm_err("nVHE hyp BUG at: [<%016llx>] %pB!\n", panic_addr, 357 (void *)(panic_addr + kaslr_offset())); 358 } else { 359 kvm_err("nVHE hyp panic at: [<%016llx>] %pB!\n", panic_addr, 360 (void *)(panic_addr + kaslr_offset())); 361 } 362 363 /* Dump the nVHE hypervisor backtrace */ 364 kvm_nvhe_dump_backtrace(hyp_offset); 365 366 /* 367 * Hyp has panicked and we're going to handle that by panicking the 368 * kernel. The kernel offset will be revealed in the panic so we're 369 * also safe to reveal the hyp offset as a debugging aid for translating 370 * hyp VAs to vmlinux addresses. 371 */ 372 kvm_err("Hyp Offset: 0x%llx\n", hyp_offset); 373 374 panic("HYP panic:\nPS:%08llx PC:%016llx ESR:%016llx\nFAR:%016llx HPFAR:%016llx PAR:%016llx\nVCPU:%016lx\n", 375 spsr, elr_virt, esr, far, hpfar, par, vcpu); 376 } 377