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/kvm_asm.h>
27 #include <asm/kvm_emulate.h>
28 #include <asm/kvm_host.h>
29 #include <asm/kvm_hyp.h>
30 #include <asm/kvm_mmu.h>
31 #include <asm/fpsimd.h>
32 #include <asm/debug-monitors.h>
33 #include <asm/processor.h>
34 #include <asm/thread_info.h>
36 /* Check whether the FP regs were dirtied while in the host-side run loop: */
37 static bool __hyp_text update_fp_enabled(struct kvm_vcpu *vcpu)
39 if (vcpu->arch.host_thread_info->flags & _TIF_FOREIGN_FPSTATE)
40 vcpu->arch.flags &= ~(KVM_ARM64_FP_ENABLED |
43 return !!(vcpu->arch.flags & KVM_ARM64_FP_ENABLED);
46 /* Save the 32-bit only FPSIMD system register state */
47 static void __hyp_text __fpsimd_save_fpexc32(struct kvm_vcpu *vcpu)
49 if (!vcpu_el1_is_32bit(vcpu))
52 vcpu->arch.ctxt.sys_regs[FPEXC32_EL2] = read_sysreg(fpexc32_el2);
55 static void __hyp_text __activate_traps_fpsimd32(struct kvm_vcpu *vcpu)
58 * We are about to set CPTR_EL2.TFP to trap all floating point
59 * register accesses to EL2, however, the ARM ARM clearly states that
60 * traps are only taken to EL2 if the operation would not otherwise
61 * trap to EL1. Therefore, always make sure that for 32-bit guests,
62 * we set FPEXC.EN to prevent traps to EL1, when setting the TFP bit.
63 * If FP/ASIMD is not implemented, FPEXC is UNDEFINED and any access to
64 * it will cause an exception.
66 if (vcpu_el1_is_32bit(vcpu) && system_supports_fpsimd()) {
67 write_sysreg(1 << 30, fpexc32_el2);
72 static void __hyp_text __activate_traps_common(struct kvm_vcpu *vcpu)
74 /* Trap on AArch32 cp15 c15 (impdef sysregs) accesses (EL1 or EL0) */
75 write_sysreg(1 << 15, hstr_el2);
78 * Make sure we trap PMU access from EL0 to EL2. Also sanitize
79 * PMSELR_EL0 to make sure it never contains the cycle
80 * counter, which could make a PMXEVCNTR_EL0 access UNDEF at
81 * EL1 instead of being trapped to EL2.
83 write_sysreg(0, pmselr_el0);
84 write_sysreg(ARMV8_PMU_USERENR_MASK, pmuserenr_el0);
85 write_sysreg(vcpu->arch.mdcr_el2, mdcr_el2);
88 static void __hyp_text __deactivate_traps_common(void)
90 write_sysreg(0, hstr_el2);
91 write_sysreg(0, pmuserenr_el0);
94 static void activate_traps_vhe(struct kvm_vcpu *vcpu)
98 val = read_sysreg(cpacr_el1);
100 val &= ~CPACR_EL1_ZEN;
101 if (!update_fp_enabled(vcpu))
102 val &= ~CPACR_EL1_FPEN;
104 write_sysreg(val, cpacr_el1);
106 write_sysreg(kvm_get_hyp_vector(), vbar_el1);
109 static void __hyp_text __activate_traps_nvhe(struct kvm_vcpu *vcpu)
113 __activate_traps_common(vcpu);
115 val = CPTR_EL2_DEFAULT;
116 val |= CPTR_EL2_TTA | CPTR_EL2_TZ;
117 if (!update_fp_enabled(vcpu))
120 write_sysreg(val, cptr_el2);
123 static void __hyp_text __activate_traps(struct kvm_vcpu *vcpu)
125 u64 hcr = vcpu->arch.hcr_el2;
127 write_sysreg(hcr, hcr_el2);
129 if (cpus_have_const_cap(ARM64_HAS_RAS_EXTN) && (hcr & HCR_VSE))
130 write_sysreg_s(vcpu->arch.vsesr_el2, SYS_VSESR_EL2);
132 __activate_traps_fpsimd32(vcpu);
134 activate_traps_vhe(vcpu);
136 __activate_traps_nvhe(vcpu);
139 static void deactivate_traps_vhe(void)
141 extern char vectors[]; /* kernel exception vectors */
142 write_sysreg(HCR_HOST_VHE_FLAGS, hcr_el2);
143 write_sysreg(CPACR_EL1_DEFAULT, cpacr_el1);
144 write_sysreg(vectors, vbar_el1);
147 static void __hyp_text __deactivate_traps_nvhe(void)
149 u64 mdcr_el2 = read_sysreg(mdcr_el2);
151 __deactivate_traps_common();
153 mdcr_el2 &= MDCR_EL2_HPMN_MASK;
154 mdcr_el2 |= MDCR_EL2_E2PB_MASK << MDCR_EL2_E2PB_SHIFT;
156 write_sysreg(mdcr_el2, mdcr_el2);
157 write_sysreg(HCR_RW, hcr_el2);
158 write_sysreg(CPTR_EL2_DEFAULT, cptr_el2);
161 static void __hyp_text __deactivate_traps(struct kvm_vcpu *vcpu)
164 * If we pended a virtual abort, preserve it until it gets
165 * cleared. See D1.14.3 (Virtual Interrupts) for details, but
166 * the crucial bit is "On taking a vSError interrupt,
167 * HCR_EL2.VSE is cleared to 0."
169 if (vcpu->arch.hcr_el2 & HCR_VSE)
170 vcpu->arch.hcr_el2 = read_sysreg(hcr_el2);
173 deactivate_traps_vhe();
175 __deactivate_traps_nvhe();
178 void activate_traps_vhe_load(struct kvm_vcpu *vcpu)
180 __activate_traps_common(vcpu);
183 void deactivate_traps_vhe_put(void)
185 u64 mdcr_el2 = read_sysreg(mdcr_el2);
187 mdcr_el2 &= MDCR_EL2_HPMN_MASK |
188 MDCR_EL2_E2PB_MASK << MDCR_EL2_E2PB_SHIFT |
191 write_sysreg(mdcr_el2, mdcr_el2);
193 __deactivate_traps_common();
196 static void __hyp_text __activate_vm(struct kvm *kvm)
198 write_sysreg(kvm->arch.vttbr, vttbr_el2);
201 static void __hyp_text __deactivate_vm(struct kvm_vcpu *vcpu)
203 write_sysreg(0, vttbr_el2);
206 /* Save VGICv3 state on non-VHE systems */
207 static void __hyp_text __hyp_vgic_save_state(struct kvm_vcpu *vcpu)
209 if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif)) {
210 __vgic_v3_save_state(vcpu);
211 __vgic_v3_deactivate_traps(vcpu);
215 /* Restore VGICv3 state on non_VEH systems */
216 static void __hyp_text __hyp_vgic_restore_state(struct kvm_vcpu *vcpu)
218 if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif)) {
219 __vgic_v3_activate_traps(vcpu);
220 __vgic_v3_restore_state(vcpu);
224 static bool __hyp_text __true_value(void)
229 static bool __hyp_text __false_value(void)
234 static hyp_alternate_select(__check_arm_834220,
235 __false_value, __true_value,
236 ARM64_WORKAROUND_834220);
238 static bool __hyp_text __translate_far_to_hpfar(u64 far, u64 *hpfar)
243 * Resolve the IPA the hard way using the guest VA.
245 * Stage-1 translation already validated the memory access
246 * rights. As such, we can use the EL1 translation regime, and
247 * don't have to distinguish between EL0 and EL1 access.
249 * We do need to save/restore PAR_EL1 though, as we haven't
250 * saved the guest context yet, and we may return early...
252 par = read_sysreg(par_el1);
253 asm volatile("at s1e1r, %0" : : "r" (far));
256 tmp = read_sysreg(par_el1);
257 write_sysreg(par, par_el1);
259 if (unlikely(tmp & 1))
260 return false; /* Translation failed, back to guest */
262 /* Convert PAR to HPFAR format */
263 *hpfar = ((tmp >> 12) & ((1UL << 36) - 1)) << 4;
267 static bool __hyp_text __populate_fault_info(struct kvm_vcpu *vcpu)
273 esr = vcpu->arch.fault.esr_el2;
274 ec = ESR_ELx_EC(esr);
276 if (ec != ESR_ELx_EC_DABT_LOW && ec != ESR_ELx_EC_IABT_LOW)
279 far = read_sysreg_el2(far);
282 * The HPFAR can be invalid if the stage 2 fault did not
283 * happen during a stage 1 page table walk (the ESR_EL2.S1PTW
284 * bit is clear) and one of the two following cases are true:
285 * 1. The fault was due to a permission fault
286 * 2. The processor carries errata 834220
288 * Therefore, for all non S1PTW faults where we either have a
289 * permission fault or the errata workaround is enabled, we
290 * resolve the IPA using the AT instruction.
292 if (!(esr & ESR_ELx_S1PTW) &&
293 (__check_arm_834220()() || (esr & ESR_ELx_FSC_TYPE) == FSC_PERM)) {
294 if (!__translate_far_to_hpfar(far, &hpfar))
297 hpfar = read_sysreg(hpfar_el2);
300 vcpu->arch.fault.far_el2 = far;
301 vcpu->arch.fault.hpfar_el2 = hpfar;
305 /* Skip an instruction which has been emulated. Returns true if
306 * execution can continue or false if we need to exit hyp mode because
307 * single-step was in effect.
309 static bool __hyp_text __skip_instr(struct kvm_vcpu *vcpu)
311 *vcpu_pc(vcpu) = read_sysreg_el2(elr);
313 if (vcpu_mode_is_32bit(vcpu)) {
314 vcpu->arch.ctxt.gp_regs.regs.pstate = read_sysreg_el2(spsr);
315 kvm_skip_instr32(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
316 write_sysreg_el2(vcpu->arch.ctxt.gp_regs.regs.pstate, spsr);
321 write_sysreg_el2(*vcpu_pc(vcpu), elr);
323 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
324 vcpu->arch.fault.esr_el2 =
325 (ESR_ELx_EC_SOFTSTP_LOW << ESR_ELx_EC_SHIFT) | 0x22;
332 static bool __hyp_text __hyp_switch_fpsimd(struct kvm_vcpu *vcpu)
334 struct user_fpsimd_state *host_fpsimd = vcpu->arch.host_fpsimd_state;
337 write_sysreg(read_sysreg(cpacr_el1) | CPACR_EL1_FPEN,
340 write_sysreg(read_sysreg(cptr_el2) & ~(u64)CPTR_EL2_TFP,
345 if (vcpu->arch.flags & KVM_ARM64_FP_HOST) {
347 * In the SVE case, VHE is assumed: it is enforced by
348 * Kconfig and kvm_arch_init().
350 if (system_supports_sve() &&
351 (vcpu->arch.flags & KVM_ARM64_HOST_SVE_IN_USE)) {
352 struct thread_struct *thread = container_of(
354 struct thread_struct, uw.fpsimd_state);
356 sve_save_state(sve_pffr(thread), &host_fpsimd->fpsr);
358 __fpsimd_save_state(host_fpsimd);
361 vcpu->arch.flags &= ~KVM_ARM64_FP_HOST;
364 __fpsimd_restore_state(&vcpu->arch.ctxt.gp_regs.fp_regs);
366 /* Skip restoring fpexc32 for AArch64 guests */
367 if (!(read_sysreg(hcr_el2) & HCR_RW))
368 write_sysreg(vcpu->arch.ctxt.sys_regs[FPEXC32_EL2],
371 vcpu->arch.flags |= KVM_ARM64_FP_ENABLED;
377 * Return true when we were able to fixup the guest exit and should return to
378 * the guest, false when we should restore the host state and return to the
381 static bool __hyp_text fixup_guest_exit(struct kvm_vcpu *vcpu, u64 *exit_code)
383 if (ARM_EXCEPTION_CODE(*exit_code) != ARM_EXCEPTION_IRQ)
384 vcpu->arch.fault.esr_el2 = read_sysreg_el2(esr);
387 * We're using the raw exception code in order to only process
388 * the trap if no SError is pending. We will come back to the
389 * same PC once the SError has been injected, and replay the
390 * trapping instruction.
392 if (*exit_code != ARM_EXCEPTION_TRAP)
396 * We trap the first access to the FP/SIMD to save the host context
397 * and restore the guest context lazily.
398 * If FP/SIMD is not implemented, handle the trap and inject an
399 * undefined instruction exception to the guest.
401 if (system_supports_fpsimd() &&
402 kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_FP_ASIMD)
403 return __hyp_switch_fpsimd(vcpu);
405 if (!__populate_fault_info(vcpu))
408 if (static_branch_unlikely(&vgic_v2_cpuif_trap)) {
411 valid = kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_DABT_LOW &&
412 kvm_vcpu_trap_get_fault_type(vcpu) == FSC_FAULT &&
413 kvm_vcpu_dabt_isvalid(vcpu) &&
414 !kvm_vcpu_dabt_isextabt(vcpu) &&
415 !kvm_vcpu_dabt_iss1tw(vcpu);
418 int ret = __vgic_v2_perform_cpuif_access(vcpu);
420 if (ret == 1 && __skip_instr(vcpu))
424 /* Promote an illegal access to an
425 * SError. If we would be returning
426 * due to single-step clear the SS
427 * bit so handle_exit knows what to
428 * do after dealing with the error.
430 if (!__skip_instr(vcpu))
431 *vcpu_cpsr(vcpu) &= ~DBG_SPSR_SS;
432 *exit_code = ARM_EXCEPTION_EL1_SERROR;
439 if (static_branch_unlikely(&vgic_v3_cpuif_trap) &&
440 (kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_SYS64 ||
441 kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_CP15_32)) {
442 int ret = __vgic_v3_perform_cpuif_access(vcpu);
444 if (ret == 1 && __skip_instr(vcpu))
449 /* Return to the host kernel and handle the exit */
453 static inline bool __hyp_text __needs_ssbd_off(struct kvm_vcpu *vcpu)
455 if (!cpus_have_const_cap(ARM64_SSBD))
458 return !(vcpu->arch.workaround_flags & VCPU_WORKAROUND_2_FLAG);
461 static void __hyp_text __set_guest_arch_workaround_state(struct kvm_vcpu *vcpu)
463 #ifdef CONFIG_ARM64_SSBD
465 * The host runs with the workaround always present. If the
466 * guest wants it disabled, so be it...
468 if (__needs_ssbd_off(vcpu) &&
469 __hyp_this_cpu_read(arm64_ssbd_callback_required))
470 arm_smccc_1_1_smc(ARM_SMCCC_ARCH_WORKAROUND_2, 0, NULL);
474 static void __hyp_text __set_host_arch_workaround_state(struct kvm_vcpu *vcpu)
476 #ifdef CONFIG_ARM64_SSBD
478 * If the guest has disabled the workaround, bring it back on.
480 if (__needs_ssbd_off(vcpu) &&
481 __hyp_this_cpu_read(arm64_ssbd_callback_required))
482 arm_smccc_1_1_smc(ARM_SMCCC_ARCH_WORKAROUND_2, 1, NULL);
486 /* Switch to the guest for VHE systems running in EL2 */
487 int kvm_vcpu_run_vhe(struct kvm_vcpu *vcpu)
489 struct kvm_cpu_context *host_ctxt;
490 struct kvm_cpu_context *guest_ctxt;
493 host_ctxt = vcpu->arch.host_cpu_context;
494 host_ctxt->__hyp_running_vcpu = vcpu;
495 guest_ctxt = &vcpu->arch.ctxt;
497 sysreg_save_host_state_vhe(host_ctxt);
499 __activate_traps(vcpu);
500 __activate_vm(vcpu->kvm);
502 sysreg_restore_guest_state_vhe(guest_ctxt);
503 __debug_switch_to_guest(vcpu);
505 __set_guest_arch_workaround_state(vcpu);
508 /* Jump in the fire! */
509 exit_code = __guest_enter(vcpu, host_ctxt);
511 /* And we're baaack! */
512 } while (fixup_guest_exit(vcpu, &exit_code));
514 __set_host_arch_workaround_state(vcpu);
516 sysreg_save_guest_state_vhe(guest_ctxt);
518 __deactivate_traps(vcpu);
520 sysreg_restore_host_state_vhe(host_ctxt);
522 if (vcpu->arch.flags & KVM_ARM64_FP_ENABLED)
523 __fpsimd_save_fpexc32(vcpu);
525 __debug_switch_to_host(vcpu);
530 /* Switch to the guest for legacy non-VHE systems */
531 int __hyp_text __kvm_vcpu_run_nvhe(struct kvm_vcpu *vcpu)
533 struct kvm_cpu_context *host_ctxt;
534 struct kvm_cpu_context *guest_ctxt;
537 vcpu = kern_hyp_va(vcpu);
539 host_ctxt = kern_hyp_va(vcpu->arch.host_cpu_context);
540 host_ctxt->__hyp_running_vcpu = vcpu;
541 guest_ctxt = &vcpu->arch.ctxt;
543 __sysreg_save_state_nvhe(host_ctxt);
545 __activate_traps(vcpu);
546 __activate_vm(kern_hyp_va(vcpu->kvm));
548 __hyp_vgic_restore_state(vcpu);
549 __timer_enable_traps(vcpu);
552 * We must restore the 32-bit state before the sysregs, thanks
553 * to erratum #852523 (Cortex-A57) or #853709 (Cortex-A72).
555 __sysreg32_restore_state(vcpu);
556 __sysreg_restore_state_nvhe(guest_ctxt);
557 __debug_switch_to_guest(vcpu);
559 __set_guest_arch_workaround_state(vcpu);
562 /* Jump in the fire! */
563 exit_code = __guest_enter(vcpu, host_ctxt);
565 /* And we're baaack! */
566 } while (fixup_guest_exit(vcpu, &exit_code));
568 __set_host_arch_workaround_state(vcpu);
570 __sysreg_save_state_nvhe(guest_ctxt);
571 __sysreg32_save_state(vcpu);
572 __timer_disable_traps(vcpu);
573 __hyp_vgic_save_state(vcpu);
575 __deactivate_traps(vcpu);
576 __deactivate_vm(vcpu);
578 __sysreg_restore_state_nvhe(host_ctxt);
580 if (vcpu->arch.flags & KVM_ARM64_FP_ENABLED)
581 __fpsimd_save_fpexc32(vcpu);
584 * This must come after restoring the host sysregs, since a non-VHE
585 * system may enable SPE here and make use of the TTBRs.
587 __debug_switch_to_host(vcpu);
592 static const char __hyp_panic_string[] = "HYP panic:\nPS:%08llx PC:%016llx ESR:%08llx\nFAR:%016llx HPFAR:%016llx PAR:%016llx\nVCPU:%p\n";
594 static void __hyp_text __hyp_call_panic_nvhe(u64 spsr, u64 elr, u64 par,
595 struct kvm_cpu_context *__host_ctxt)
597 struct kvm_vcpu *vcpu;
598 unsigned long str_va;
600 vcpu = __host_ctxt->__hyp_running_vcpu;
602 if (read_sysreg(vttbr_el2)) {
603 __timer_disable_traps(vcpu);
604 __deactivate_traps(vcpu);
605 __deactivate_vm(vcpu);
606 __sysreg_restore_state_nvhe(__host_ctxt);
610 * Force the panic string to be loaded from the literal pool,
611 * making sure it is a kernel address and not a PC-relative
614 asm volatile("ldr %0, =__hyp_panic_string" : "=r" (str_va));
616 __hyp_do_panic(str_va,
618 read_sysreg(esr_el2), read_sysreg_el2(far),
619 read_sysreg(hpfar_el2), par, vcpu);
622 static void __hyp_call_panic_vhe(u64 spsr, u64 elr, u64 par,
623 struct kvm_cpu_context *host_ctxt)
625 struct kvm_vcpu *vcpu;
626 vcpu = host_ctxt->__hyp_running_vcpu;
628 __deactivate_traps(vcpu);
629 sysreg_restore_host_state_vhe(host_ctxt);
631 panic(__hyp_panic_string,
633 read_sysreg_el2(esr), read_sysreg_el2(far),
634 read_sysreg(hpfar_el2), par, vcpu);
637 void __hyp_text __noreturn hyp_panic(struct kvm_cpu_context *host_ctxt)
639 u64 spsr = read_sysreg_el2(spsr);
640 u64 elr = read_sysreg_el2(elr);
641 u64 par = read_sysreg(par_el1);
644 __hyp_call_panic_nvhe(spsr, elr, par, host_ctxt);
646 __hyp_call_panic_vhe(spsr, elr, par, host_ctxt);