xref: /openbmc/linux/arch/arm64/kvm/hyp/nvhe/switch.c (revision 94d964e5)
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 
29 #include <nvhe/fixed_config.h>
30 #include <nvhe/mem_protect.h>
31 
32 /* Non-VHE specific context */
33 DEFINE_PER_CPU(struct kvm_host_data, kvm_host_data);
34 DEFINE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt);
35 DEFINE_PER_CPU(unsigned long, kvm_hyp_vector);
36 
37 static void __activate_traps(struct kvm_vcpu *vcpu)
38 {
39 	u64 val;
40 
41 	___activate_traps(vcpu);
42 	__activate_traps_common(vcpu);
43 
44 	val = vcpu->arch.cptr_el2;
45 	val |= CPTR_EL2_TTA | CPTR_EL2_TAM;
46 	if (!update_fp_enabled(vcpu)) {
47 		val |= CPTR_EL2_TFP | CPTR_EL2_TZ;
48 		__activate_traps_fpsimd32(vcpu);
49 	}
50 
51 	write_sysreg(val, cptr_el2);
52 	write_sysreg(__this_cpu_read(kvm_hyp_vector), vbar_el2);
53 
54 	if (cpus_have_final_cap(ARM64_WORKAROUND_SPECULATIVE_AT)) {
55 		struct kvm_cpu_context *ctxt = &vcpu->arch.ctxt;
56 
57 		isb();
58 		/*
59 		 * At this stage, and thanks to the above isb(), S2 is
60 		 * configured and enabled. We can now restore the guest's S1
61 		 * configuration: SCTLR, and only then TCR.
62 		 */
63 		write_sysreg_el1(ctxt_sys_reg(ctxt, SCTLR_EL1),	SYS_SCTLR);
64 		isb();
65 		write_sysreg_el1(ctxt_sys_reg(ctxt, TCR_EL1),	SYS_TCR);
66 	}
67 }
68 
69 static void __deactivate_traps(struct kvm_vcpu *vcpu)
70 {
71 	extern char __kvm_hyp_host_vector[];
72 	u64 cptr;
73 
74 	___deactivate_traps(vcpu);
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(vcpu);
94 
95 	write_sysreg(this_cpu_ptr(&kvm_init_params)->hcr_el2, hcr_el2);
96 
97 	cptr = CPTR_EL2_DEFAULT;
98 	if (vcpu_has_sve(vcpu) && (vcpu->arch.flags & KVM_ARM64_FP_ENABLED))
99 		cptr |= CPTR_EL2_TZ;
100 
101 	write_sysreg(cptr, cptr_el2);
102 	write_sysreg(__kvm_hyp_host_vector, vbar_el2);
103 }
104 
105 /* Save VGICv3 state on non-VHE systems */
106 static void __hyp_vgic_save_state(struct kvm_vcpu *vcpu)
107 {
108 	if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif)) {
109 		__vgic_v3_save_state(&vcpu->arch.vgic_cpu.vgic_v3);
110 		__vgic_v3_deactivate_traps(&vcpu->arch.vgic_cpu.vgic_v3);
111 	}
112 }
113 
114 /* Restore VGICv3 state on non_VEH systems */
115 static void __hyp_vgic_restore_state(struct kvm_vcpu *vcpu)
116 {
117 	if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif)) {
118 		__vgic_v3_activate_traps(&vcpu->arch.vgic_cpu.vgic_v3);
119 		__vgic_v3_restore_state(&vcpu->arch.vgic_cpu.vgic_v3);
120 	}
121 }
122 
123 /**
124  * Disable host events, enable guest events
125  */
126 static bool __pmu_switch_to_guest(struct kvm_cpu_context *host_ctxt)
127 {
128 	struct kvm_host_data *host;
129 	struct kvm_pmu_events *pmu;
130 
131 	host = container_of(host_ctxt, struct kvm_host_data, host_ctxt);
132 	pmu = &host->pmu_events;
133 
134 	if (pmu->events_host)
135 		write_sysreg(pmu->events_host, pmcntenclr_el0);
136 
137 	if (pmu->events_guest)
138 		write_sysreg(pmu->events_guest, pmcntenset_el0);
139 
140 	return (pmu->events_host || pmu->events_guest);
141 }
142 
143 /**
144  * Disable guest events, enable host events
145  */
146 static void __pmu_switch_to_host(struct kvm_cpu_context *host_ctxt)
147 {
148 	struct kvm_host_data *host;
149 	struct kvm_pmu_events *pmu;
150 
151 	host = container_of(host_ctxt, struct kvm_host_data, host_ctxt);
152 	pmu = &host->pmu_events;
153 
154 	if (pmu->events_guest)
155 		write_sysreg(pmu->events_guest, pmcntenclr_el0);
156 
157 	if (pmu->events_host)
158 		write_sysreg(pmu->events_host, pmcntenset_el0);
159 }
160 
161 /**
162  * Handler for protected VM MSR, MRS or System instruction execution in AArch64.
163  *
164  * Returns true if the hypervisor has handled the exit, and control should go
165  * back to the guest, or false if it hasn't.
166  */
167 static bool kvm_handle_pvm_sys64(struct kvm_vcpu *vcpu, u64 *exit_code)
168 {
169 	/*
170 	 * Make sure we handle the exit for workarounds and ptrauth
171 	 * before the pKVM handling, as the latter could decide to
172 	 * UNDEF.
173 	 */
174 	return (kvm_hyp_handle_sysreg(vcpu, exit_code) ||
175 		kvm_handle_pvm_sysreg(vcpu, exit_code));
176 }
177 
178 /**
179  * Handler for protected floating-point and Advanced SIMD accesses.
180  *
181  * Returns true if the hypervisor has handled the exit, and control should go
182  * back to the guest, or false if it hasn't.
183  */
184 static bool kvm_handle_pvm_fpsimd(struct kvm_vcpu *vcpu, u64 *exit_code)
185 {
186 	/* Linux guests assume support for floating-point and Advanced SIMD. */
187 	BUILD_BUG_ON(!FIELD_GET(ARM64_FEATURE_MASK(ID_AA64PFR0_FP),
188 				PVM_ID_AA64PFR0_ALLOW));
189 	BUILD_BUG_ON(!FIELD_GET(ARM64_FEATURE_MASK(ID_AA64PFR0_ASIMD),
190 				PVM_ID_AA64PFR0_ALLOW));
191 
192 	return kvm_hyp_handle_fpsimd(vcpu, exit_code);
193 }
194 
195 static const exit_handler_fn hyp_exit_handlers[] = {
196 	[0 ... ESR_ELx_EC_MAX]		= NULL,
197 	[ESR_ELx_EC_CP15_32]		= kvm_hyp_handle_cp15_32,
198 	[ESR_ELx_EC_SYS64]		= kvm_hyp_handle_sysreg,
199 	[ESR_ELx_EC_SVE]		= kvm_hyp_handle_fpsimd,
200 	[ESR_ELx_EC_FP_ASIMD]		= kvm_hyp_handle_fpsimd,
201 	[ESR_ELx_EC_IABT_LOW]		= kvm_hyp_handle_iabt_low,
202 	[ESR_ELx_EC_DABT_LOW]		= kvm_hyp_handle_dabt_low,
203 	[ESR_ELx_EC_PAC]		= kvm_hyp_handle_ptrauth,
204 };
205 
206 static const exit_handler_fn pvm_exit_handlers[] = {
207 	[0 ... ESR_ELx_EC_MAX]		= NULL,
208 	[ESR_ELx_EC_SYS64]		= kvm_handle_pvm_sys64,
209 	[ESR_ELx_EC_SVE]		= kvm_handle_pvm_restricted,
210 	[ESR_ELx_EC_FP_ASIMD]		= kvm_handle_pvm_fpsimd,
211 	[ESR_ELx_EC_IABT_LOW]		= kvm_hyp_handle_iabt_low,
212 	[ESR_ELx_EC_DABT_LOW]		= kvm_hyp_handle_dabt_low,
213 	[ESR_ELx_EC_PAC]		= kvm_hyp_handle_ptrauth,
214 };
215 
216 static const exit_handler_fn *kvm_get_exit_handler_array(struct kvm_vcpu *vcpu)
217 {
218 	if (unlikely(kvm_vm_is_protected(kern_hyp_va(vcpu->kvm))))
219 		return pvm_exit_handlers;
220 
221 	return hyp_exit_handlers;
222 }
223 
224 /*
225  * Some guests (e.g., protected VMs) are not be allowed to run in AArch32.
226  * The ARMv8 architecture does not give the hypervisor a mechanism to prevent a
227  * guest from dropping to AArch32 EL0 if implemented by the CPU. If the
228  * hypervisor spots a guest in such a state ensure it is handled, and don't
229  * trust the host to spot or fix it.  The check below is based on the one in
230  * kvm_arch_vcpu_ioctl_run().
231  *
232  * Returns false if the guest ran in AArch32 when it shouldn't have, and
233  * thus should exit to the host, or true if a the guest run loop can continue.
234  */
235 static void early_exit_filter(struct kvm_vcpu *vcpu, u64 *exit_code)
236 {
237 	struct kvm *kvm = kern_hyp_va(vcpu->kvm);
238 
239 	if (kvm_vm_is_protected(kvm) && vcpu_mode_is_32bit(vcpu)) {
240 		/*
241 		 * As we have caught the guest red-handed, decide that it isn't
242 		 * fit for purpose anymore by making the vcpu invalid. The VMM
243 		 * can try and fix it by re-initializing the vcpu with
244 		 * KVM_ARM_VCPU_INIT, however, this is likely not possible for
245 		 * protected VMs.
246 		 */
247 		vcpu->arch.target = -1;
248 		*exit_code &= BIT(ARM_EXIT_WITH_SERROR_BIT);
249 		*exit_code |= ARM_EXCEPTION_IL;
250 	}
251 }
252 
253 /* Switch to the guest for legacy non-VHE systems */
254 int __kvm_vcpu_run(struct kvm_vcpu *vcpu)
255 {
256 	struct kvm_cpu_context *host_ctxt;
257 	struct kvm_cpu_context *guest_ctxt;
258 	struct kvm_s2_mmu *mmu;
259 	bool pmu_switch_needed;
260 	u64 exit_code;
261 
262 	/*
263 	 * Having IRQs masked via PMR when entering the guest means the GIC
264 	 * will not signal the CPU of interrupts of lower priority, and the
265 	 * only way to get out will be via guest exceptions.
266 	 * Naturally, we want to avoid this.
267 	 */
268 	if (system_uses_irq_prio_masking()) {
269 		gic_write_pmr(GIC_PRIO_IRQON | GIC_PRIO_PSR_I_SET);
270 		pmr_sync();
271 	}
272 
273 	host_ctxt = &this_cpu_ptr(&kvm_host_data)->host_ctxt;
274 	host_ctxt->__hyp_running_vcpu = vcpu;
275 	guest_ctxt = &vcpu->arch.ctxt;
276 
277 	pmu_switch_needed = __pmu_switch_to_guest(host_ctxt);
278 
279 	__sysreg_save_state_nvhe(host_ctxt);
280 	/*
281 	 * We must flush and disable the SPE buffer for nVHE, as
282 	 * the translation regime(EL1&0) is going to be loaded with
283 	 * that of the guest. And we must do this before we change the
284 	 * translation regime to EL2 (via MDCR_EL2_E2PB == 0) and
285 	 * before we load guest Stage1.
286 	 */
287 	__debug_save_host_buffers_nvhe(vcpu);
288 
289 	__kvm_adjust_pc(vcpu);
290 
291 	/*
292 	 * We must restore the 32-bit state before the sysregs, thanks
293 	 * to erratum #852523 (Cortex-A57) or #853709 (Cortex-A72).
294 	 *
295 	 * Also, and in order to be able to deal with erratum #1319537 (A57)
296 	 * and #1319367 (A72), we must ensure that all VM-related sysreg are
297 	 * restored before we enable S2 translation.
298 	 */
299 	__sysreg32_restore_state(vcpu);
300 	__sysreg_restore_state_nvhe(guest_ctxt);
301 
302 	mmu = kern_hyp_va(vcpu->arch.hw_mmu);
303 	__load_stage2(mmu, kern_hyp_va(mmu->arch));
304 	__activate_traps(vcpu);
305 
306 	__hyp_vgic_restore_state(vcpu);
307 	__timer_enable_traps(vcpu);
308 
309 	__debug_switch_to_guest(vcpu);
310 
311 	do {
312 		/* Jump in the fire! */
313 		exit_code = __guest_enter(vcpu);
314 
315 		/* And we're baaack! */
316 	} while (fixup_guest_exit(vcpu, &exit_code));
317 
318 	__sysreg_save_state_nvhe(guest_ctxt);
319 	__sysreg32_save_state(vcpu);
320 	__timer_disable_traps(vcpu);
321 	__hyp_vgic_save_state(vcpu);
322 
323 	__deactivate_traps(vcpu);
324 	__load_host_stage2();
325 
326 	__sysreg_restore_state_nvhe(host_ctxt);
327 
328 	if (vcpu->arch.flags & KVM_ARM64_FP_ENABLED)
329 		__fpsimd_save_fpexc32(vcpu);
330 
331 	__debug_switch_to_host(vcpu);
332 	/*
333 	 * This must come after restoring the host sysregs, since a non-VHE
334 	 * system may enable SPE here and make use of the TTBRs.
335 	 */
336 	__debug_restore_host_buffers_nvhe(vcpu);
337 
338 	if (pmu_switch_needed)
339 		__pmu_switch_to_host(host_ctxt);
340 
341 	/* Returning to host will clear PSR.I, remask PMR if needed */
342 	if (system_uses_irq_prio_masking())
343 		gic_write_pmr(GIC_PRIO_IRQOFF);
344 
345 	host_ctxt->__hyp_running_vcpu = NULL;
346 
347 	return exit_code;
348 }
349 
350 void __noreturn hyp_panic(void)
351 {
352 	u64 spsr = read_sysreg_el2(SYS_SPSR);
353 	u64 elr = read_sysreg_el2(SYS_ELR);
354 	u64 par = read_sysreg_par();
355 	struct kvm_cpu_context *host_ctxt;
356 	struct kvm_vcpu *vcpu;
357 
358 	host_ctxt = &this_cpu_ptr(&kvm_host_data)->host_ctxt;
359 	vcpu = host_ctxt->__hyp_running_vcpu;
360 
361 	if (vcpu) {
362 		__timer_disable_traps(vcpu);
363 		__deactivate_traps(vcpu);
364 		__load_host_stage2();
365 		__sysreg_restore_state_nvhe(host_ctxt);
366 	}
367 
368 	__hyp_do_panic(host_ctxt, spsr, elr, par);
369 	unreachable();
370 }
371 
372 asmlinkage void kvm_unexpected_el2_exception(void)
373 {
374 	return __kvm_unexpected_el2_exception();
375 }
376