1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2015 - ARM Ltd
4 * Author: Marc Zyngier <marc.zyngier@arm.com>
7 #ifndef __ARM64_KVM_HYP_SWITCH_H__
8 #define __ARM64_KVM_HYP_SWITCH_H__
10 #include <hyp/adjust_pc.h>
11 #include <hyp/fault.h>
13 #include <linux/arm-smccc.h>
14 #include <linux/kvm_host.h>
15 #include <linux/types.h>
16 #include <linux/jump_label.h>
17 #include <uapi/linux/psci.h>
19 #include <kvm/arm_psci.h>
21 #include <asm/barrier.h>
22 #include <asm/cpufeature.h>
23 #include <asm/extable.h>
24 #include <asm/kprobes.h>
25 #include <asm/kvm_asm.h>
26 #include <asm/kvm_emulate.h>
27 #include <asm/kvm_hyp.h>
28 #include <asm/kvm_mmu.h>
29 #include <asm/kvm_nested.h>
30 #include <asm/fpsimd.h>
31 #include <asm/debug-monitors.h>
32 #include <asm/processor.h>
34 struct kvm_exception_table_entry {
38 extern struct kvm_exception_table_entry __start___kvm_ex_table;
39 extern struct kvm_exception_table_entry __stop___kvm_ex_table;
41 /* Check whether the FP regs are owned by the guest */
42 static inline bool guest_owns_fp_regs(struct kvm_vcpu *vcpu)
44 return vcpu->arch.fp_state == FP_STATE_GUEST_OWNED;
47 /* Save the 32-bit only FPSIMD system register state */
48 static inline void __fpsimd_save_fpexc32(struct kvm_vcpu *vcpu)
50 if (!vcpu_el1_is_32bit(vcpu))
53 __vcpu_sys_reg(vcpu, FPEXC32_EL2) = read_sysreg(fpexc32_el2);
56 static inline void __activate_traps_fpsimd32(struct kvm_vcpu *vcpu)
59 * We are about to set CPTR_EL2.TFP to trap all floating point
60 * register accesses to EL2, however, the ARM ARM clearly states that
61 * traps are only taken to EL2 if the operation would not otherwise
62 * trap to EL1. Therefore, always make sure that for 32-bit guests,
63 * we set FPEXC.EN to prevent traps to EL1, when setting the TFP bit.
64 * If FP/ASIMD is not implemented, FPEXC is UNDEFINED and any access to
65 * it will cause an exception.
67 if (vcpu_el1_is_32bit(vcpu) && system_supports_fpsimd()) {
68 write_sysreg(1 << 30, fpexc32_el2);
73 static inline void __activate_traps_common(struct kvm_vcpu *vcpu)
75 /* Trap on AArch32 cp15 c15 (impdef sysregs) accesses (EL1 or EL0) */
76 write_sysreg(1 << 15, hstr_el2);
79 * Make sure we trap PMU access from EL0 to EL2. Also sanitize
80 * PMSELR_EL0 to make sure it never contains the cycle
81 * counter, which could make a PMXEVCNTR_EL0 access UNDEF at
82 * EL1 instead of being trapped to EL2.
84 if (kvm_arm_support_pmu_v3()) {
85 write_sysreg(0, pmselr_el0);
86 write_sysreg(ARMV8_PMU_USERENR_MASK, pmuserenr_el0);
89 vcpu->arch.mdcr_el2_host = read_sysreg(mdcr_el2);
90 write_sysreg(vcpu->arch.mdcr_el2, mdcr_el2);
92 if (cpus_have_final_cap(ARM64_SME)) {
93 sysreg_clear_set_s(SYS_HFGRTR_EL2,
94 HFGxTR_EL2_nSMPRI_EL1_MASK |
95 HFGxTR_EL2_nTPIDR2_EL0_MASK,
97 sysreg_clear_set_s(SYS_HFGWTR_EL2,
98 HFGxTR_EL2_nSMPRI_EL1_MASK |
99 HFGxTR_EL2_nTPIDR2_EL0_MASK,
104 static inline void __deactivate_traps_common(struct kvm_vcpu *vcpu)
106 write_sysreg(vcpu->arch.mdcr_el2_host, mdcr_el2);
108 write_sysreg(0, hstr_el2);
109 if (kvm_arm_support_pmu_v3())
110 write_sysreg(0, pmuserenr_el0);
112 if (cpus_have_final_cap(ARM64_SME)) {
113 sysreg_clear_set_s(SYS_HFGRTR_EL2, 0,
114 HFGxTR_EL2_nSMPRI_EL1_MASK |
115 HFGxTR_EL2_nTPIDR2_EL0_MASK);
116 sysreg_clear_set_s(SYS_HFGWTR_EL2, 0,
117 HFGxTR_EL2_nSMPRI_EL1_MASK |
118 HFGxTR_EL2_nTPIDR2_EL0_MASK);
122 static inline void ___activate_traps(struct kvm_vcpu *vcpu)
124 u64 hcr = vcpu->arch.hcr_el2;
126 if (cpus_have_final_cap(ARM64_WORKAROUND_CAVIUM_TX2_219_TVM))
129 write_sysreg(hcr, hcr_el2);
131 if (cpus_have_final_cap(ARM64_HAS_RAS_EXTN) && (hcr & HCR_VSE))
132 write_sysreg_s(vcpu->arch.vsesr_el2, SYS_VSESR_EL2);
135 static inline void ___deactivate_traps(struct kvm_vcpu *vcpu)
138 * If we pended a virtual abort, preserve it until it gets
139 * cleared. See D1.14.3 (Virtual Interrupts) for details, but
140 * the crucial bit is "On taking a vSError interrupt,
141 * HCR_EL2.VSE is cleared to 0."
143 if (vcpu->arch.hcr_el2 & HCR_VSE) {
144 vcpu->arch.hcr_el2 &= ~HCR_VSE;
145 vcpu->arch.hcr_el2 |= read_sysreg(hcr_el2) & HCR_VSE;
149 static inline bool __populate_fault_info(struct kvm_vcpu *vcpu)
151 return __get_fault_info(vcpu->arch.fault.esr_el2, &vcpu->arch.fault);
154 static inline void __hyp_sve_restore_guest(struct kvm_vcpu *vcpu)
156 sve_cond_update_zcr_vq(vcpu_sve_max_vq(vcpu) - 1, SYS_ZCR_EL2);
157 __sve_restore_state(vcpu_sve_pffr(vcpu),
158 &vcpu->arch.ctxt.fp_regs.fpsr);
159 write_sysreg_el1(__vcpu_sys_reg(vcpu, ZCR_EL1), SYS_ZCR);
163 * We trap the first access to the FP/SIMD to save the host context and
164 * restore the guest context lazily.
165 * If FP/SIMD is not implemented, handle the trap and inject an undefined
166 * instruction exception to the guest. Similarly for trapped SVE accesses.
168 static bool kvm_hyp_handle_fpsimd(struct kvm_vcpu *vcpu, u64 *exit_code)
174 if (!system_supports_fpsimd())
177 sve_guest = vcpu_has_sve(vcpu);
178 esr_ec = kvm_vcpu_trap_get_class(vcpu);
180 /* Only handle traps the vCPU can support here: */
182 case ESR_ELx_EC_FP_ASIMD:
192 /* Valid trap. Switch the context: */
194 /* First disable enough traps to allow us to update the registers */
195 if (has_vhe() || has_hvhe()) {
196 reg = CPACR_EL1_FPEN_EL0EN | CPACR_EL1_FPEN_EL1EN;
198 reg |= CPACR_EL1_ZEN_EL0EN | CPACR_EL1_ZEN_EL1EN;
200 sysreg_clear_set(cpacr_el1, 0, reg);
206 sysreg_clear_set(cptr_el2, reg, 0);
210 /* Write out the host state if it's in the registers */
211 if (vcpu->arch.fp_state == FP_STATE_HOST_OWNED)
212 __fpsimd_save_state(vcpu->arch.host_fpsimd_state);
214 /* Restore the guest state */
216 __hyp_sve_restore_guest(vcpu);
218 __fpsimd_restore_state(&vcpu->arch.ctxt.fp_regs);
220 /* Skip restoring fpexc32 for AArch64 guests */
221 if (!(read_sysreg(hcr_el2) & HCR_RW))
222 write_sysreg(__vcpu_sys_reg(vcpu, FPEXC32_EL2), fpexc32_el2);
224 vcpu->arch.fp_state = FP_STATE_GUEST_OWNED;
229 static inline bool handle_tx2_tvm(struct kvm_vcpu *vcpu)
231 u32 sysreg = esr_sys64_to_sysreg(kvm_vcpu_get_esr(vcpu));
232 int rt = kvm_vcpu_sys_get_rt(vcpu);
233 u64 val = vcpu_get_reg(vcpu, rt);
236 * The normal sysreg handling code expects to see the traps,
237 * let's not do anything here.
239 if (vcpu->arch.hcr_el2 & HCR_TVM)
244 write_sysreg_el1(val, SYS_SCTLR);
247 write_sysreg_el1(val, SYS_TTBR0);
250 write_sysreg_el1(val, SYS_TTBR1);
253 write_sysreg_el1(val, SYS_TCR);
256 write_sysreg_el1(val, SYS_ESR);
259 write_sysreg_el1(val, SYS_FAR);
262 write_sysreg_el1(val, SYS_AFSR0);
265 write_sysreg_el1(val, SYS_AFSR1);
268 write_sysreg_el1(val, SYS_MAIR);
271 write_sysreg_el1(val, SYS_AMAIR);
273 case SYS_CONTEXTIDR_EL1:
274 write_sysreg_el1(val, SYS_CONTEXTIDR);
280 __kvm_skip_instr(vcpu);
284 static inline bool esr_is_ptrauth_trap(u64 esr)
286 switch (esr_sys64_to_sysreg(esr)) {
287 case SYS_APIAKEYLO_EL1:
288 case SYS_APIAKEYHI_EL1:
289 case SYS_APIBKEYLO_EL1:
290 case SYS_APIBKEYHI_EL1:
291 case SYS_APDAKEYLO_EL1:
292 case SYS_APDAKEYHI_EL1:
293 case SYS_APDBKEYLO_EL1:
294 case SYS_APDBKEYHI_EL1:
295 case SYS_APGAKEYLO_EL1:
296 case SYS_APGAKEYHI_EL1:
303 #define __ptrauth_save_key(ctxt, key) \
306 __val = read_sysreg_s(SYS_ ## key ## KEYLO_EL1); \
307 ctxt_sys_reg(ctxt, key ## KEYLO_EL1) = __val; \
308 __val = read_sysreg_s(SYS_ ## key ## KEYHI_EL1); \
309 ctxt_sys_reg(ctxt, key ## KEYHI_EL1) = __val; \
312 DECLARE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt);
314 static bool kvm_hyp_handle_ptrauth(struct kvm_vcpu *vcpu, u64 *exit_code)
316 struct kvm_cpu_context *ctxt;
319 if (!vcpu_has_ptrauth(vcpu))
322 ctxt = this_cpu_ptr(&kvm_hyp_ctxt);
323 __ptrauth_save_key(ctxt, APIA);
324 __ptrauth_save_key(ctxt, APIB);
325 __ptrauth_save_key(ctxt, APDA);
326 __ptrauth_save_key(ctxt, APDB);
327 __ptrauth_save_key(ctxt, APGA);
329 vcpu_ptrauth_enable(vcpu);
331 val = read_sysreg(hcr_el2);
332 val |= (HCR_API | HCR_APK);
333 write_sysreg(val, hcr_el2);
338 static bool kvm_hyp_handle_cntpct(struct kvm_vcpu *vcpu)
340 struct arch_timer_context *ctxt;
345 * We only get here for 64bit guests, 32bit guests will hit
346 * the long and winding road all the way to the standard
347 * handling. Yes, it sucks to be irrelevant.
349 sysreg = esr_sys64_to_sysreg(kvm_vcpu_get_esr(vcpu));
353 case SYS_CNTPCTSS_EL0:
354 if (vcpu_has_nv(vcpu)) {
355 if (is_hyp_ctxt(vcpu)) {
356 ctxt = vcpu_hptimer(vcpu);
360 /* Check for guest hypervisor trapping */
361 val = __vcpu_sys_reg(vcpu, CNTHCTL_EL2);
362 if (!vcpu_el2_e2h_is_set(vcpu))
363 val = (val & CNTHCTL_EL1PCTEN) << 10;
365 if (!(val & (CNTHCTL_EL1PCTEN << 10)))
369 ctxt = vcpu_ptimer(vcpu);
375 val = arch_timer_read_cntpct_el0();
377 if (ctxt->offset.vm_offset)
378 val -= *kern_hyp_va(ctxt->offset.vm_offset);
379 if (ctxt->offset.vcpu_offset)
380 val -= *kern_hyp_va(ctxt->offset.vcpu_offset);
382 vcpu_set_reg(vcpu, kvm_vcpu_sys_get_rt(vcpu), val);
383 __kvm_skip_instr(vcpu);
387 static bool kvm_hyp_handle_sysreg(struct kvm_vcpu *vcpu, u64 *exit_code)
389 if (cpus_have_final_cap(ARM64_WORKAROUND_CAVIUM_TX2_219_TVM) &&
390 handle_tx2_tvm(vcpu))
393 if (static_branch_unlikely(&vgic_v3_cpuif_trap) &&
394 __vgic_v3_perform_cpuif_access(vcpu) == 1)
397 if (esr_is_ptrauth_trap(kvm_vcpu_get_esr(vcpu)))
398 return kvm_hyp_handle_ptrauth(vcpu, exit_code);
400 if (kvm_hyp_handle_cntpct(vcpu))
406 static bool kvm_hyp_handle_cp15_32(struct kvm_vcpu *vcpu, u64 *exit_code)
408 if (static_branch_unlikely(&vgic_v3_cpuif_trap) &&
409 __vgic_v3_perform_cpuif_access(vcpu) == 1)
415 static bool kvm_hyp_handle_iabt_low(struct kvm_vcpu *vcpu, u64 *exit_code)
417 if (!__populate_fault_info(vcpu))
423 static bool kvm_hyp_handle_dabt_low(struct kvm_vcpu *vcpu, u64 *exit_code)
425 if (!__populate_fault_info(vcpu))
428 if (static_branch_unlikely(&vgic_v2_cpuif_trap)) {
431 valid = kvm_vcpu_trap_get_fault_type(vcpu) == ESR_ELx_FSC_FAULT &&
432 kvm_vcpu_dabt_isvalid(vcpu) &&
433 !kvm_vcpu_abt_issea(vcpu) &&
434 !kvm_vcpu_abt_iss1tw(vcpu);
437 int ret = __vgic_v2_perform_cpuif_access(vcpu);
442 /* Promote an illegal access to an SError.*/
444 *exit_code = ARM_EXCEPTION_EL1_SERROR;
451 typedef bool (*exit_handler_fn)(struct kvm_vcpu *, u64 *);
453 static const exit_handler_fn *kvm_get_exit_handler_array(struct kvm_vcpu *vcpu);
455 static void early_exit_filter(struct kvm_vcpu *vcpu, u64 *exit_code);
458 * Allow the hypervisor to handle the exit with an exit handler if it has one.
460 * Returns true if the hypervisor handled the exit, and control should go back
461 * to the guest, or false if it hasn't.
463 static inline bool kvm_hyp_handle_exit(struct kvm_vcpu *vcpu, u64 *exit_code)
465 const exit_handler_fn *handlers = kvm_get_exit_handler_array(vcpu);
468 fn = handlers[kvm_vcpu_trap_get_class(vcpu)];
471 return fn(vcpu, exit_code);
476 static inline void synchronize_vcpu_pstate(struct kvm_vcpu *vcpu, u64 *exit_code)
479 * Check for the conditions of Cortex-A510's #2077057. When these occur
480 * SPSR_EL2 can't be trusted, but isn't needed either as it is
481 * unchanged from the value in vcpu_gp_regs(vcpu)->pstate.
482 * Are we single-stepping the guest, and took a PAC exception from the
483 * active-not-pending state?
485 if (cpus_have_final_cap(ARM64_WORKAROUND_2077057) &&
486 vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
487 *vcpu_cpsr(vcpu) & DBG_SPSR_SS &&
488 ESR_ELx_EC(read_sysreg_el2(SYS_ESR)) == ESR_ELx_EC_PAC)
489 write_sysreg_el2(*vcpu_cpsr(vcpu), SYS_SPSR);
491 vcpu->arch.ctxt.regs.pstate = read_sysreg_el2(SYS_SPSR);
495 * Return true when we were able to fixup the guest exit and should return to
496 * the guest, false when we should restore the host state and return to the
499 static inline bool fixup_guest_exit(struct kvm_vcpu *vcpu, u64 *exit_code)
502 * Save PSTATE early so that we can evaluate the vcpu mode
505 synchronize_vcpu_pstate(vcpu, exit_code);
508 * Check whether we want to repaint the state one way or
511 early_exit_filter(vcpu, exit_code);
513 if (ARM_EXCEPTION_CODE(*exit_code) != ARM_EXCEPTION_IRQ)
514 vcpu->arch.fault.esr_el2 = read_sysreg_el2(SYS_ESR);
516 if (ARM_SERROR_PENDING(*exit_code) &&
517 ARM_EXCEPTION_CODE(*exit_code) != ARM_EXCEPTION_IRQ) {
518 u8 esr_ec = kvm_vcpu_trap_get_class(vcpu);
521 * HVC already have an adjusted PC, which we need to
522 * correct in order to return to after having injected
525 * SMC, on the other hand, is *trapped*, meaning its
526 * preferred return address is the SMC itself.
528 if (esr_ec == ESR_ELx_EC_HVC32 || esr_ec == ESR_ELx_EC_HVC64)
529 write_sysreg_el2(read_sysreg_el2(SYS_ELR) - 4, SYS_ELR);
533 * We're using the raw exception code in order to only process
534 * the trap if no SError is pending. We will come back to the
535 * same PC once the SError has been injected, and replay the
536 * trapping instruction.
538 if (*exit_code != ARM_EXCEPTION_TRAP)
541 /* Check if there's an exit handler and allow it to handle the exit. */
542 if (kvm_hyp_handle_exit(vcpu, exit_code))
545 /* Return to the host kernel and handle the exit */
549 /* Re-enter the guest */
550 asm(ALTERNATIVE("nop", "dmb sy", ARM64_WORKAROUND_1508412));
554 static inline void __kvm_unexpected_el2_exception(void)
556 extern char __guest_exit_panic[];
557 unsigned long addr, fixup;
558 struct kvm_exception_table_entry *entry, *end;
559 unsigned long elr_el2 = read_sysreg(elr_el2);
561 entry = &__start___kvm_ex_table;
562 end = &__stop___kvm_ex_table;
564 while (entry < end) {
565 addr = (unsigned long)&entry->insn + entry->insn;
566 fixup = (unsigned long)&entry->fixup + entry->fixup;
568 if (addr != elr_el2) {
573 write_sysreg(fixup, elr_el2);
577 /* Trigger a panic after restoring the hyp context. */
578 write_sysreg(__guest_exit_panic, elr_el2);
581 #endif /* __ARM64_KVM_HYP_SWITCH_H__ */