2 * Copyright (C) 2015 - ARM Ltd
3 * Author: Marc Zyngier <marc.zyngier@arm.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License version 2 as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program. If not, see <http://www.gnu.org/licenses/>.
18 #include <linux/arm-smccc.h>
19 #include <linux/types.h>
20 #include <linux/jump_label.h>
21 #include <uapi/linux/psci.h>
23 #include <kvm/arm_psci.h>
25 #include <asm/cpufeature.h>
26 #include <asm/extable.h>
27 #include <asm/kprobes.h>
28 #include <asm/kvm_asm.h>
29 #include <asm/kvm_emulate.h>
30 #include <asm/kvm_host.h>
31 #include <asm/kvm_hyp.h>
32 #include <asm/kvm_mmu.h>
33 #include <asm/fpsimd.h>
34 #include <asm/debug-monitors.h>
35 #include <asm/processor.h>
36 #include <asm/thread_info.h>
38 extern struct exception_table_entry __start___kvm_ex_table;
39 extern struct exception_table_entry __stop___kvm_ex_table;
41 /* Check whether the FP regs were dirtied while in the host-side run loop: */
42 static bool __hyp_text update_fp_enabled(struct kvm_vcpu *vcpu)
45 * When the system doesn't support FP/SIMD, we cannot rely on
46 * the _TIF_FOREIGN_FPSTATE flag. However, we always inject an
47 * abort on the very first access to FP and thus we should never
48 * see KVM_ARM64_FP_ENABLED. For added safety, make sure we always
51 if (!system_supports_fpsimd() ||
52 vcpu->arch.host_thread_info->flags & _TIF_FOREIGN_FPSTATE)
53 vcpu->arch.flags &= ~(KVM_ARM64_FP_ENABLED |
56 return !!(vcpu->arch.flags & KVM_ARM64_FP_ENABLED);
59 /* Save the 32-bit only FPSIMD system register state */
60 static void __hyp_text __fpsimd_save_fpexc32(struct kvm_vcpu *vcpu)
62 if (!vcpu_el1_is_32bit(vcpu))
65 vcpu->arch.ctxt.sys_regs[FPEXC32_EL2] = read_sysreg(fpexc32_el2);
68 static void __hyp_text __activate_traps_fpsimd32(struct kvm_vcpu *vcpu)
71 * We are about to set CPTR_EL2.TFP to trap all floating point
72 * register accesses to EL2, however, the ARM ARM clearly states that
73 * traps are only taken to EL2 if the operation would not otherwise
74 * trap to EL1. Therefore, always make sure that for 32-bit guests,
75 * we set FPEXC.EN to prevent traps to EL1, when setting the TFP bit.
76 * If FP/ASIMD is not implemented, FPEXC is UNDEFINED and any access to
77 * it will cause an exception.
79 if (vcpu_el1_is_32bit(vcpu) && system_supports_fpsimd()) {
80 write_sysreg(1 << 30, fpexc32_el2);
85 static void __hyp_text __activate_traps_common(struct kvm_vcpu *vcpu)
87 /* Trap on AArch32 cp15 c15 (impdef sysregs) accesses (EL1 or EL0) */
88 write_sysreg(1 << 15, hstr_el2);
91 * Make sure we trap PMU access from EL0 to EL2. Also sanitize
92 * PMSELR_EL0 to make sure it never contains the cycle
93 * counter, which could make a PMXEVCNTR_EL0 access UNDEF at
94 * EL1 instead of being trapped to EL2.
96 write_sysreg(0, pmselr_el0);
97 write_sysreg(ARMV8_PMU_USERENR_MASK, pmuserenr_el0);
98 write_sysreg(vcpu->arch.mdcr_el2, mdcr_el2);
101 static void __hyp_text __deactivate_traps_common(void)
103 write_sysreg(0, hstr_el2);
104 write_sysreg(0, pmuserenr_el0);
107 static void activate_traps_vhe(struct kvm_vcpu *vcpu)
111 val = read_sysreg(cpacr_el1);
112 val |= CPACR_EL1_TTA;
113 val &= ~CPACR_EL1_ZEN;
114 if (!update_fp_enabled(vcpu)) {
115 val &= ~CPACR_EL1_FPEN;
116 __activate_traps_fpsimd32(vcpu);
119 write_sysreg(val, cpacr_el1);
121 write_sysreg(kvm_get_hyp_vector(), vbar_el1);
123 NOKPROBE_SYMBOL(activate_traps_vhe);
125 static void __hyp_text __activate_traps_nvhe(struct kvm_vcpu *vcpu)
129 __activate_traps_common(vcpu);
131 val = CPTR_EL2_DEFAULT;
132 val |= CPTR_EL2_TTA | CPTR_EL2_TZ;
133 if (!update_fp_enabled(vcpu)) {
135 __activate_traps_fpsimd32(vcpu);
138 write_sysreg(val, cptr_el2);
141 static void __hyp_text __activate_traps(struct kvm_vcpu *vcpu)
143 u64 hcr = vcpu->arch.hcr_el2;
145 write_sysreg(hcr, hcr_el2);
147 if (cpus_have_const_cap(ARM64_HAS_RAS_EXTN) && (hcr & HCR_VSE))
148 write_sysreg_s(vcpu->arch.vsesr_el2, SYS_VSESR_EL2);
151 activate_traps_vhe(vcpu);
153 __activate_traps_nvhe(vcpu);
156 static void deactivate_traps_vhe(void)
158 extern char vectors[]; /* kernel exception vectors */
159 write_sysreg(HCR_HOST_VHE_FLAGS, hcr_el2);
160 write_sysreg(CPACR_EL1_DEFAULT, cpacr_el1);
161 write_sysreg(vectors, vbar_el1);
163 NOKPROBE_SYMBOL(deactivate_traps_vhe);
165 static void __hyp_text __deactivate_traps_nvhe(void)
167 u64 mdcr_el2 = read_sysreg(mdcr_el2);
169 __deactivate_traps_common();
171 mdcr_el2 &= MDCR_EL2_HPMN_MASK;
172 mdcr_el2 |= MDCR_EL2_E2PB_MASK << MDCR_EL2_E2PB_SHIFT;
174 write_sysreg(mdcr_el2, mdcr_el2);
175 write_sysreg(HCR_HOST_NVHE_FLAGS, hcr_el2);
176 write_sysreg(CPTR_EL2_DEFAULT, cptr_el2);
179 static void __hyp_text __deactivate_traps(struct kvm_vcpu *vcpu)
182 * If we pended a virtual abort, preserve it until it gets
183 * cleared. See D1.14.3 (Virtual Interrupts) for details, but
184 * the crucial bit is "On taking a vSError interrupt,
185 * HCR_EL2.VSE is cleared to 0."
187 if (vcpu->arch.hcr_el2 & HCR_VSE)
188 vcpu->arch.hcr_el2 = read_sysreg(hcr_el2);
191 deactivate_traps_vhe();
193 __deactivate_traps_nvhe();
196 void activate_traps_vhe_load(struct kvm_vcpu *vcpu)
198 __activate_traps_common(vcpu);
201 void deactivate_traps_vhe_put(void)
203 u64 mdcr_el2 = read_sysreg(mdcr_el2);
205 mdcr_el2 &= MDCR_EL2_HPMN_MASK |
206 MDCR_EL2_E2PB_MASK << MDCR_EL2_E2PB_SHIFT |
209 write_sysreg(mdcr_el2, mdcr_el2);
211 __deactivate_traps_common();
214 static void __hyp_text __activate_vm(struct kvm *kvm)
216 write_sysreg(kvm->arch.vttbr, vttbr_el2);
219 static void __hyp_text __deactivate_vm(struct kvm_vcpu *vcpu)
221 write_sysreg(0, vttbr_el2);
224 /* Save VGICv3 state on non-VHE systems */
225 static void __hyp_text __hyp_vgic_save_state(struct kvm_vcpu *vcpu)
227 if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif)) {
228 __vgic_v3_save_state(vcpu);
229 __vgic_v3_deactivate_traps(vcpu);
233 /* Restore VGICv3 state on non_VEH systems */
234 static void __hyp_text __hyp_vgic_restore_state(struct kvm_vcpu *vcpu)
236 if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif)) {
237 __vgic_v3_activate_traps(vcpu);
238 __vgic_v3_restore_state(vcpu);
242 static bool __hyp_text __true_value(void)
247 static bool __hyp_text __false_value(void)
252 static hyp_alternate_select(__check_arm_834220,
253 __false_value, __true_value,
254 ARM64_WORKAROUND_834220);
256 static bool __hyp_text __translate_far_to_hpfar(u64 far, u64 *hpfar)
261 * Resolve the IPA the hard way using the guest VA.
263 * Stage-1 translation already validated the memory access
264 * rights. As such, we can use the EL1 translation regime, and
265 * don't have to distinguish between EL0 and EL1 access.
267 * We do need to save/restore PAR_EL1 though, as we haven't
268 * saved the guest context yet, and we may return early...
270 par = read_sysreg(par_el1);
271 if (!__kvm_at("s1e1r", far))
272 tmp = read_sysreg(par_el1);
274 tmp = 1; /* back to the guest */
275 write_sysreg(par, par_el1);
277 if (unlikely(tmp & 1))
278 return false; /* Translation failed, back to guest */
280 /* Convert PAR to HPFAR format */
281 *hpfar = ((tmp >> 12) & ((1UL << 36) - 1)) << 4;
285 static bool __hyp_text __populate_fault_info(struct kvm_vcpu *vcpu)
291 esr = vcpu->arch.fault.esr_el2;
292 ec = ESR_ELx_EC(esr);
294 if (ec != ESR_ELx_EC_DABT_LOW && ec != ESR_ELx_EC_IABT_LOW)
297 far = read_sysreg_el2(far);
300 * The HPFAR can be invalid if the stage 2 fault did not
301 * happen during a stage 1 page table walk (the ESR_EL2.S1PTW
302 * bit is clear) and one of the two following cases are true:
303 * 1. The fault was due to a permission fault
304 * 2. The processor carries errata 834220
306 * Therefore, for all non S1PTW faults where we either have a
307 * permission fault or the errata workaround is enabled, we
308 * resolve the IPA using the AT instruction.
310 if (!(esr & ESR_ELx_S1PTW) &&
311 (__check_arm_834220()() || (esr & ESR_ELx_FSC_TYPE) == FSC_PERM)) {
312 if (!__translate_far_to_hpfar(far, &hpfar))
315 hpfar = read_sysreg(hpfar_el2);
318 vcpu->arch.fault.far_el2 = far;
319 vcpu->arch.fault.hpfar_el2 = hpfar;
323 /* Skip an instruction which has been emulated. Returns true if
324 * execution can continue or false if we need to exit hyp mode because
325 * single-step was in effect.
327 static bool __hyp_text __skip_instr(struct kvm_vcpu *vcpu)
329 *vcpu_pc(vcpu) = read_sysreg_el2(elr);
331 if (vcpu_mode_is_32bit(vcpu)) {
332 vcpu->arch.ctxt.gp_regs.regs.pstate = read_sysreg_el2(spsr);
333 kvm_skip_instr32(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
334 write_sysreg_el2(vcpu->arch.ctxt.gp_regs.regs.pstate, spsr);
339 write_sysreg_el2(*vcpu_pc(vcpu), elr);
341 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
342 vcpu->arch.fault.esr_el2 =
343 (ESR_ELx_EC_SOFTSTP_LOW << ESR_ELx_EC_SHIFT) | 0x22;
350 static bool __hyp_text __hyp_switch_fpsimd(struct kvm_vcpu *vcpu)
352 struct user_fpsimd_state *host_fpsimd = vcpu->arch.host_fpsimd_state;
355 write_sysreg(read_sysreg(cpacr_el1) | CPACR_EL1_FPEN,
358 write_sysreg(read_sysreg(cptr_el2) & ~(u64)CPTR_EL2_TFP,
363 if (vcpu->arch.flags & KVM_ARM64_FP_HOST) {
365 * In the SVE case, VHE is assumed: it is enforced by
366 * Kconfig and kvm_arch_init().
368 if (system_supports_sve() &&
369 (vcpu->arch.flags & KVM_ARM64_HOST_SVE_IN_USE)) {
370 struct thread_struct *thread = container_of(
372 struct thread_struct, uw.fpsimd_state);
374 sve_save_state(sve_pffr(thread), &host_fpsimd->fpsr);
376 __fpsimd_save_state(host_fpsimd);
379 vcpu->arch.flags &= ~KVM_ARM64_FP_HOST;
382 __fpsimd_restore_state(&vcpu->arch.ctxt.gp_regs.fp_regs);
384 /* Skip restoring fpexc32 for AArch64 guests */
385 if (!(read_sysreg(hcr_el2) & HCR_RW))
386 write_sysreg(vcpu->arch.ctxt.sys_regs[FPEXC32_EL2],
389 vcpu->arch.flags |= KVM_ARM64_FP_ENABLED;
395 * Return true when we were able to fixup the guest exit and should return to
396 * the guest, false when we should restore the host state and return to the
399 static bool __hyp_text fixup_guest_exit(struct kvm_vcpu *vcpu, u64 *exit_code)
401 if (ARM_EXCEPTION_CODE(*exit_code) != ARM_EXCEPTION_IRQ)
402 vcpu->arch.fault.esr_el2 = read_sysreg_el2(esr);
405 * We're using the raw exception code in order to only process
406 * the trap if no SError is pending. We will come back to the
407 * same PC once the SError has been injected, and replay the
408 * trapping instruction.
410 if (*exit_code != ARM_EXCEPTION_TRAP)
414 * We trap the first access to the FP/SIMD to save the host context
415 * and restore the guest context lazily.
416 * If FP/SIMD is not implemented, handle the trap and inject an
417 * undefined instruction exception to the guest.
419 if (system_supports_fpsimd() &&
420 kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_FP_ASIMD)
421 return __hyp_switch_fpsimd(vcpu);
423 if (!__populate_fault_info(vcpu))
426 if (static_branch_unlikely(&vgic_v2_cpuif_trap)) {
429 valid = kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_DABT_LOW &&
430 kvm_vcpu_trap_get_fault_type(vcpu) == FSC_FAULT &&
431 kvm_vcpu_dabt_isvalid(vcpu) &&
432 !kvm_vcpu_dabt_isextabt(vcpu) &&
433 !kvm_vcpu_abt_iss1tw(vcpu);
436 int ret = __vgic_v2_perform_cpuif_access(vcpu);
438 if (ret == 1 && __skip_instr(vcpu))
442 /* Promote an illegal access to an
443 * SError. If we would be returning
444 * due to single-step clear the SS
445 * bit so handle_exit knows what to
446 * do after dealing with the error.
448 if (!__skip_instr(vcpu))
449 *vcpu_cpsr(vcpu) &= ~DBG_SPSR_SS;
450 *exit_code = ARM_EXCEPTION_EL1_SERROR;
457 if (static_branch_unlikely(&vgic_v3_cpuif_trap) &&
458 (kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_SYS64 ||
459 kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_CP15_32)) {
460 int ret = __vgic_v3_perform_cpuif_access(vcpu);
462 if (ret == 1 && __skip_instr(vcpu))
467 /* Return to the host kernel and handle the exit */
471 static inline bool __hyp_text __needs_ssbd_off(struct kvm_vcpu *vcpu)
473 if (!cpus_have_const_cap(ARM64_SSBD))
476 return !(vcpu->arch.workaround_flags & VCPU_WORKAROUND_2_FLAG);
479 static void __hyp_text __set_guest_arch_workaround_state(struct kvm_vcpu *vcpu)
481 #ifdef CONFIG_ARM64_SSBD
483 * The host runs with the workaround always present. If the
484 * guest wants it disabled, so be it...
486 if (__needs_ssbd_off(vcpu) &&
487 __hyp_this_cpu_read(arm64_ssbd_callback_required))
488 arm_smccc_1_1_smc(ARM_SMCCC_ARCH_WORKAROUND_2, 0, NULL);
492 static void __hyp_text __set_host_arch_workaround_state(struct kvm_vcpu *vcpu)
494 #ifdef CONFIG_ARM64_SSBD
496 * If the guest has disabled the workaround, bring it back on.
498 if (__needs_ssbd_off(vcpu) &&
499 __hyp_this_cpu_read(arm64_ssbd_callback_required))
500 arm_smccc_1_1_smc(ARM_SMCCC_ARCH_WORKAROUND_2, 1, NULL);
504 /* Switch to the guest for VHE systems running in EL2 */
505 int kvm_vcpu_run_vhe(struct kvm_vcpu *vcpu)
507 struct kvm_cpu_context *host_ctxt;
508 struct kvm_cpu_context *guest_ctxt;
511 host_ctxt = vcpu->arch.host_cpu_context;
512 host_ctxt->__hyp_running_vcpu = vcpu;
513 guest_ctxt = &vcpu->arch.ctxt;
515 sysreg_save_host_state_vhe(host_ctxt);
517 __activate_traps(vcpu);
518 __activate_vm(vcpu->kvm);
520 sysreg_restore_guest_state_vhe(guest_ctxt);
521 __debug_switch_to_guest(vcpu);
523 __set_guest_arch_workaround_state(vcpu);
526 /* Jump in the fire! */
527 exit_code = __guest_enter(vcpu, host_ctxt);
529 /* And we're baaack! */
530 } while (fixup_guest_exit(vcpu, &exit_code));
532 __set_host_arch_workaround_state(vcpu);
534 sysreg_save_guest_state_vhe(guest_ctxt);
536 __deactivate_traps(vcpu);
538 sysreg_restore_host_state_vhe(host_ctxt);
540 if (vcpu->arch.flags & KVM_ARM64_FP_ENABLED)
541 __fpsimd_save_fpexc32(vcpu);
543 __debug_switch_to_host(vcpu);
547 NOKPROBE_SYMBOL(kvm_vcpu_run_vhe);
549 /* Switch to the guest for legacy non-VHE systems */
550 int __hyp_text __kvm_vcpu_run_nvhe(struct kvm_vcpu *vcpu)
552 struct kvm_cpu_context *host_ctxt;
553 struct kvm_cpu_context *guest_ctxt;
556 vcpu = kern_hyp_va(vcpu);
558 host_ctxt = kern_hyp_va(vcpu->arch.host_cpu_context);
559 host_ctxt->__hyp_running_vcpu = vcpu;
560 guest_ctxt = &vcpu->arch.ctxt;
562 __sysreg_save_state_nvhe(host_ctxt);
563 __debug_save_host_buffers_nvhe(vcpu);
565 __activate_traps(vcpu);
566 __activate_vm(kern_hyp_va(vcpu->kvm));
568 __hyp_vgic_restore_state(vcpu);
569 __timer_enable_traps(vcpu);
572 * We must restore the 32-bit state before the sysregs, thanks
573 * to erratum #852523 (Cortex-A57) or #853709 (Cortex-A72).
575 __sysreg32_restore_state(vcpu);
576 __sysreg_restore_state_nvhe(guest_ctxt);
577 __debug_switch_to_guest(vcpu);
579 __set_guest_arch_workaround_state(vcpu);
582 /* Jump in the fire! */
583 exit_code = __guest_enter(vcpu, host_ctxt);
585 /* And we're baaack! */
586 } while (fixup_guest_exit(vcpu, &exit_code));
588 __set_host_arch_workaround_state(vcpu);
590 __sysreg_save_state_nvhe(guest_ctxt);
591 __sysreg32_save_state(vcpu);
592 __timer_disable_traps(vcpu);
593 __hyp_vgic_save_state(vcpu);
595 __deactivate_traps(vcpu);
596 __deactivate_vm(vcpu);
598 __sysreg_restore_state_nvhe(host_ctxt);
600 if (vcpu->arch.flags & KVM_ARM64_FP_ENABLED)
601 __fpsimd_save_fpexc32(vcpu);
603 __debug_switch_to_host(vcpu);
605 * This must come after restoring the host sysregs, since a non-VHE
606 * system may enable SPE here and make use of the TTBRs.
608 __debug_restore_host_buffers_nvhe(vcpu);
613 static const char __hyp_panic_string[] = "HYP panic:\nPS:%08llx PC:%016llx ESR:%08llx\nFAR:%016llx HPFAR:%016llx PAR:%016llx\nVCPU:%p\n";
615 static void __hyp_text __hyp_call_panic_nvhe(u64 spsr, u64 elr, u64 par,
616 struct kvm_cpu_context *__host_ctxt)
618 struct kvm_vcpu *vcpu;
619 unsigned long str_va;
621 vcpu = __host_ctxt->__hyp_running_vcpu;
623 if (read_sysreg(vttbr_el2)) {
624 __timer_disable_traps(vcpu);
625 __deactivate_traps(vcpu);
626 __deactivate_vm(vcpu);
627 __sysreg_restore_state_nvhe(__host_ctxt);
631 * Force the panic string to be loaded from the literal pool,
632 * making sure it is a kernel address and not a PC-relative
635 asm volatile("ldr %0, =%1" : "=r" (str_va) : "S" (__hyp_panic_string));
637 __hyp_do_panic(str_va,
639 read_sysreg(esr_el2), read_sysreg_el2(far),
640 read_sysreg(hpfar_el2), par, vcpu);
643 static void __hyp_call_panic_vhe(u64 spsr, u64 elr, u64 par,
644 struct kvm_cpu_context *host_ctxt)
646 struct kvm_vcpu *vcpu;
647 vcpu = host_ctxt->__hyp_running_vcpu;
649 __deactivate_traps(vcpu);
650 sysreg_restore_host_state_vhe(host_ctxt);
652 panic(__hyp_panic_string,
654 read_sysreg_el2(esr), read_sysreg_el2(far),
655 read_sysreg(hpfar_el2), par, vcpu);
657 NOKPROBE_SYMBOL(__hyp_call_panic_vhe);
659 void __hyp_text __noreturn hyp_panic(struct kvm_cpu_context *host_ctxt)
661 u64 spsr = read_sysreg_el2(spsr);
662 u64 elr = read_sysreg_el2(elr);
663 u64 par = read_sysreg(par_el1);
666 __hyp_call_panic_nvhe(spsr, elr, par, host_ctxt);
668 __hyp_call_panic_vhe(spsr, elr, par, host_ctxt);
673 asmlinkage void __hyp_text kvm_unexpected_el2_exception(void)
675 unsigned long addr, fixup;
676 struct kvm_cpu_context *host_ctxt;
677 struct exception_table_entry *entry, *end;
678 unsigned long elr_el2 = read_sysreg(elr_el2);
680 entry = hyp_symbol_addr(__start___kvm_ex_table);
681 end = hyp_symbol_addr(__stop___kvm_ex_table);
682 host_ctxt = __hyp_this_cpu_ptr(kvm_host_cpu_state);
684 while (entry < end) {
685 addr = (unsigned long)&entry->insn + entry->insn;
686 fixup = (unsigned long)&entry->fixup + entry->fixup;
688 if (addr != elr_el2) {
693 write_sysreg(fixup, elr_el2);
697 hyp_panic(host_ctxt);