xref: /openbmc/linux/arch/arm64/kvm/hyp/nvhe/switch.c (revision e15a5365)
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 #include <asm/thread_info.h>
29 
30 /* Non-VHE specific context */
31 DEFINE_PER_CPU(struct kvm_host_data, kvm_host_data);
32 DEFINE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt);
33 DEFINE_PER_CPU(unsigned long, kvm_hyp_vector);
34 
35 static void __activate_traps(struct kvm_vcpu *vcpu)
36 {
37 	u64 val;
38 
39 	___activate_traps(vcpu);
40 	__activate_traps_common(vcpu);
41 
42 	val = CPTR_EL2_DEFAULT;
43 	val |= CPTR_EL2_TTA | CPTR_EL2_TZ | CPTR_EL2_TAM;
44 	if (!update_fp_enabled(vcpu)) {
45 		val |= CPTR_EL2_TFP;
46 		__activate_traps_fpsimd32(vcpu);
47 	}
48 
49 	write_sysreg(val, cptr_el2);
50 	write_sysreg(__this_cpu_read(kvm_hyp_vector), vbar_el2);
51 
52 	if (cpus_have_final_cap(ARM64_WORKAROUND_SPECULATIVE_AT)) {
53 		struct kvm_cpu_context *ctxt = &vcpu->arch.ctxt;
54 
55 		isb();
56 		/*
57 		 * At this stage, and thanks to the above isb(), S2 is
58 		 * configured and enabled. We can now restore the guest's S1
59 		 * configuration: SCTLR, and only then TCR.
60 		 */
61 		write_sysreg_el1(ctxt_sys_reg(ctxt, SCTLR_EL1),	SYS_SCTLR);
62 		isb();
63 		write_sysreg_el1(ctxt_sys_reg(ctxt, TCR_EL1),	SYS_TCR);
64 	}
65 }
66 
67 static void __deactivate_traps(struct kvm_vcpu *vcpu)
68 {
69 	extern char __kvm_hyp_host_vector[];
70 	u64 mdcr_el2;
71 
72 	___deactivate_traps(vcpu);
73 
74 	mdcr_el2 = read_sysreg(mdcr_el2);
75 
76 	if (cpus_have_final_cap(ARM64_WORKAROUND_SPECULATIVE_AT)) {
77 		u64 val;
78 
79 		/*
80 		 * Set the TCR and SCTLR registers in the exact opposite
81 		 * sequence as __activate_traps (first prevent walks,
82 		 * then force the MMU on). A generous sprinkling of isb()
83 		 * ensure that things happen in this exact order.
84 		 */
85 		val = read_sysreg_el1(SYS_TCR);
86 		write_sysreg_el1(val | TCR_EPD1_MASK | TCR_EPD0_MASK, SYS_TCR);
87 		isb();
88 		val = read_sysreg_el1(SYS_SCTLR);
89 		write_sysreg_el1(val | SCTLR_ELx_M, SYS_SCTLR);
90 		isb();
91 	}
92 
93 	__deactivate_traps_common();
94 
95 	mdcr_el2 &= MDCR_EL2_HPMN_MASK;
96 	mdcr_el2 |= MDCR_EL2_E2PB_MASK << MDCR_EL2_E2PB_SHIFT;
97 
98 	write_sysreg(mdcr_el2, mdcr_el2);
99 	write_sysreg(HCR_HOST_NVHE_FLAGS, hcr_el2);
100 	write_sysreg(CPTR_EL2_DEFAULT, cptr_el2);
101 	write_sysreg(__kvm_hyp_host_vector, vbar_el2);
102 }
103 
104 static void __load_host_stage2(void)
105 {
106 	write_sysreg(0, vttbr_el2);
107 }
108 
109 /* Save VGICv3 state on non-VHE systems */
110 static void __hyp_vgic_save_state(struct kvm_vcpu *vcpu)
111 {
112 	if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif)) {
113 		__vgic_v3_save_state(&vcpu->arch.vgic_cpu.vgic_v3);
114 		__vgic_v3_deactivate_traps(&vcpu->arch.vgic_cpu.vgic_v3);
115 	}
116 }
117 
118 /* Restore VGICv3 state on non_VEH systems */
119 static void __hyp_vgic_restore_state(struct kvm_vcpu *vcpu)
120 {
121 	if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif)) {
122 		__vgic_v3_activate_traps(&vcpu->arch.vgic_cpu.vgic_v3);
123 		__vgic_v3_restore_state(&vcpu->arch.vgic_cpu.vgic_v3);
124 	}
125 }
126 
127 /**
128  * Disable host events, enable guest events
129  */
130 static bool __pmu_switch_to_guest(struct kvm_cpu_context *host_ctxt)
131 {
132 	struct kvm_host_data *host;
133 	struct kvm_pmu_events *pmu;
134 
135 	host = container_of(host_ctxt, struct kvm_host_data, host_ctxt);
136 	pmu = &host->pmu_events;
137 
138 	if (pmu->events_host)
139 		write_sysreg(pmu->events_host, pmcntenclr_el0);
140 
141 	if (pmu->events_guest)
142 		write_sysreg(pmu->events_guest, pmcntenset_el0);
143 
144 	return (pmu->events_host || pmu->events_guest);
145 }
146 
147 /**
148  * Disable guest events, enable host events
149  */
150 static void __pmu_switch_to_host(struct kvm_cpu_context *host_ctxt)
151 {
152 	struct kvm_host_data *host;
153 	struct kvm_pmu_events *pmu;
154 
155 	host = container_of(host_ctxt, struct kvm_host_data, host_ctxt);
156 	pmu = &host->pmu_events;
157 
158 	if (pmu->events_guest)
159 		write_sysreg(pmu->events_guest, pmcntenclr_el0);
160 
161 	if (pmu->events_host)
162 		write_sysreg(pmu->events_host, pmcntenset_el0);
163 }
164 
165 /* Switch to the guest for legacy non-VHE systems */
166 int __kvm_vcpu_run(struct kvm_vcpu *vcpu)
167 {
168 	struct kvm_cpu_context *host_ctxt;
169 	struct kvm_cpu_context *guest_ctxt;
170 	bool pmu_switch_needed;
171 	u64 exit_code;
172 
173 	/*
174 	 * Having IRQs masked via PMR when entering the guest means the GIC
175 	 * will not signal the CPU of interrupts of lower priority, and the
176 	 * only way to get out will be via guest exceptions.
177 	 * Naturally, we want to avoid this.
178 	 */
179 	if (system_uses_irq_prio_masking()) {
180 		gic_write_pmr(GIC_PRIO_IRQON | GIC_PRIO_PSR_I_SET);
181 		pmr_sync();
182 	}
183 
184 	host_ctxt = &this_cpu_ptr(&kvm_host_data)->host_ctxt;
185 	host_ctxt->__hyp_running_vcpu = vcpu;
186 	guest_ctxt = &vcpu->arch.ctxt;
187 
188 	pmu_switch_needed = __pmu_switch_to_guest(host_ctxt);
189 
190 	__sysreg_save_state_nvhe(host_ctxt);
191 
192 	/*
193 	 * We must restore the 32-bit state before the sysregs, thanks
194 	 * to erratum #852523 (Cortex-A57) or #853709 (Cortex-A72).
195 	 *
196 	 * Also, and in order to be able to deal with erratum #1319537 (A57)
197 	 * and #1319367 (A72), we must ensure that all VM-related sysreg are
198 	 * restored before we enable S2 translation.
199 	 */
200 	__sysreg32_restore_state(vcpu);
201 	__sysreg_restore_state_nvhe(guest_ctxt);
202 
203 	__load_guest_stage2(kern_hyp_va(vcpu->arch.hw_mmu));
204 	__activate_traps(vcpu);
205 
206 	__hyp_vgic_restore_state(vcpu);
207 	__timer_enable_traps(vcpu);
208 
209 	__debug_switch_to_guest(vcpu);
210 
211 	do {
212 		/* Jump in the fire! */
213 		exit_code = __guest_enter(vcpu);
214 
215 		/* And we're baaack! */
216 	} while (fixup_guest_exit(vcpu, &exit_code));
217 
218 	__sysreg_save_state_nvhe(guest_ctxt);
219 	__sysreg32_save_state(vcpu);
220 	__timer_disable_traps(vcpu);
221 	__hyp_vgic_save_state(vcpu);
222 
223 	__deactivate_traps(vcpu);
224 	__load_host_stage2();
225 
226 	__sysreg_restore_state_nvhe(host_ctxt);
227 
228 	if (vcpu->arch.flags & KVM_ARM64_FP_ENABLED)
229 		__fpsimd_save_fpexc32(vcpu);
230 
231 	/*
232 	 * This must come after restoring the host sysregs, since a non-VHE
233 	 * system may enable SPE here and make use of the TTBRs.
234 	 */
235 	__debug_switch_to_host(vcpu);
236 
237 	if (pmu_switch_needed)
238 		__pmu_switch_to_host(host_ctxt);
239 
240 	/* Returning to host will clear PSR.I, remask PMR if needed */
241 	if (system_uses_irq_prio_masking())
242 		gic_write_pmr(GIC_PRIO_IRQOFF);
243 
244 	host_ctxt->__hyp_running_vcpu = NULL;
245 
246 	return exit_code;
247 }
248 
249 void __noreturn hyp_panic(void)
250 {
251 	u64 spsr = read_sysreg_el2(SYS_SPSR);
252 	u64 elr = read_sysreg_el2(SYS_ELR);
253 	u64 par = read_sysreg_par();
254 	bool restore_host = true;
255 	struct kvm_cpu_context *host_ctxt;
256 	struct kvm_vcpu *vcpu;
257 
258 	host_ctxt = &this_cpu_ptr(&kvm_host_data)->host_ctxt;
259 	vcpu = host_ctxt->__hyp_running_vcpu;
260 
261 	if (vcpu) {
262 		__timer_disable_traps(vcpu);
263 		__deactivate_traps(vcpu);
264 		__load_host_stage2();
265 		__sysreg_restore_state_nvhe(host_ctxt);
266 	}
267 
268 	__hyp_do_panic(restore_host, spsr, elr, par);
269 	unreachable();
270 }
271 
272 asmlinkage void kvm_unexpected_el2_exception(void)
273 {
274 	return __kvm_unexpected_el2_exception();
275 }
276