xref: /openbmc/linux/arch/arm64/kvm/hyp/nvhe/switch.c (revision fe7498ef)
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 #include <nvhe/fixed_config.h>
31 #include <nvhe/mem_protect.h>
32 
33 /* Non-VHE specific context */
34 DEFINE_PER_CPU(struct kvm_host_data, kvm_host_data);
35 DEFINE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt);
36 DEFINE_PER_CPU(unsigned long, kvm_hyp_vector);
37 
38 static void __activate_traps(struct kvm_vcpu *vcpu)
39 {
40 	u64 val;
41 
42 	___activate_traps(vcpu);
43 	__activate_traps_common(vcpu);
44 
45 	val = vcpu->arch.cptr_el2;
46 	val |= CPTR_EL2_TTA | CPTR_EL2_TAM;
47 	if (!update_fp_enabled(vcpu)) {
48 		val |= CPTR_EL2_TFP | CPTR_EL2_TZ;
49 		__activate_traps_fpsimd32(vcpu);
50 	}
51 
52 	write_sysreg(val, cptr_el2);
53 	write_sysreg(__this_cpu_read(kvm_hyp_vector), vbar_el2);
54 
55 	if (cpus_have_final_cap(ARM64_WORKAROUND_SPECULATIVE_AT)) {
56 		struct kvm_cpu_context *ctxt = &vcpu->arch.ctxt;
57 
58 		isb();
59 		/*
60 		 * At this stage, and thanks to the above isb(), S2 is
61 		 * configured and enabled. We can now restore the guest's S1
62 		 * configuration: SCTLR, and only then TCR.
63 		 */
64 		write_sysreg_el1(ctxt_sys_reg(ctxt, SCTLR_EL1),	SYS_SCTLR);
65 		isb();
66 		write_sysreg_el1(ctxt_sys_reg(ctxt, TCR_EL1),	SYS_TCR);
67 	}
68 }
69 
70 static void __deactivate_traps(struct kvm_vcpu *vcpu)
71 {
72 	extern char __kvm_hyp_host_vector[];
73 	u64 cptr;
74 
75 	___deactivate_traps(vcpu);
76 
77 	if (cpus_have_final_cap(ARM64_WORKAROUND_SPECULATIVE_AT)) {
78 		u64 val;
79 
80 		/*
81 		 * Set the TCR and SCTLR registers in the exact opposite
82 		 * sequence as __activate_traps (first prevent walks,
83 		 * then force the MMU on). A generous sprinkling of isb()
84 		 * ensure that things happen in this exact order.
85 		 */
86 		val = read_sysreg_el1(SYS_TCR);
87 		write_sysreg_el1(val | TCR_EPD1_MASK | TCR_EPD0_MASK, SYS_TCR);
88 		isb();
89 		val = read_sysreg_el1(SYS_SCTLR);
90 		write_sysreg_el1(val | SCTLR_ELx_M, SYS_SCTLR);
91 		isb();
92 	}
93 
94 	__deactivate_traps_common(vcpu);
95 
96 	write_sysreg(this_cpu_ptr(&kvm_init_params)->hcr_el2, hcr_el2);
97 
98 	cptr = CPTR_EL2_DEFAULT;
99 	if (vcpu_has_sve(vcpu) && (vcpu->arch.flags & KVM_ARM64_FP_ENABLED))
100 		cptr |= CPTR_EL2_TZ;
101 
102 	write_sysreg(cptr, cptr_el2);
103 	write_sysreg(__kvm_hyp_host_vector, vbar_el2);
104 }
105 
106 /* Save VGICv3 state on non-VHE systems */
107 static void __hyp_vgic_save_state(struct kvm_vcpu *vcpu)
108 {
109 	if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif)) {
110 		__vgic_v3_save_state(&vcpu->arch.vgic_cpu.vgic_v3);
111 		__vgic_v3_deactivate_traps(&vcpu->arch.vgic_cpu.vgic_v3);
112 	}
113 }
114 
115 /* Restore VGICv3 state on non_VEH systems */
116 static void __hyp_vgic_restore_state(struct kvm_vcpu *vcpu)
117 {
118 	if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif)) {
119 		__vgic_v3_activate_traps(&vcpu->arch.vgic_cpu.vgic_v3);
120 		__vgic_v3_restore_state(&vcpu->arch.vgic_cpu.vgic_v3);
121 	}
122 }
123 
124 /**
125  * Disable host events, enable guest events
126  */
127 static bool __pmu_switch_to_guest(struct kvm_cpu_context *host_ctxt)
128 {
129 	struct kvm_host_data *host;
130 	struct kvm_pmu_events *pmu;
131 
132 	host = container_of(host_ctxt, struct kvm_host_data, host_ctxt);
133 	pmu = &host->pmu_events;
134 
135 	if (pmu->events_host)
136 		write_sysreg(pmu->events_host, pmcntenclr_el0);
137 
138 	if (pmu->events_guest)
139 		write_sysreg(pmu->events_guest, pmcntenset_el0);
140 
141 	return (pmu->events_host || pmu->events_guest);
142 }
143 
144 /**
145  * Disable guest events, enable host events
146  */
147 static void __pmu_switch_to_host(struct kvm_cpu_context *host_ctxt)
148 {
149 	struct kvm_host_data *host;
150 	struct kvm_pmu_events *pmu;
151 
152 	host = container_of(host_ctxt, struct kvm_host_data, host_ctxt);
153 	pmu = &host->pmu_events;
154 
155 	if (pmu->events_guest)
156 		write_sysreg(pmu->events_guest, pmcntenclr_el0);
157 
158 	if (pmu->events_host)
159 		write_sysreg(pmu->events_host, pmcntenset_el0);
160 }
161 
162 /**
163  * Handler for protected VM MSR, MRS or System instruction execution in AArch64.
164  *
165  * Returns true if the hypervisor has handled the exit, and control should go
166  * back to the guest, or false if it hasn't.
167  */
168 static bool kvm_handle_pvm_sys64(struct kvm_vcpu *vcpu, u64 *exit_code)
169 {
170 	/*
171 	 * Make sure we handle the exit for workarounds and ptrauth
172 	 * before the pKVM handling, as the latter could decide to
173 	 * UNDEF.
174 	 */
175 	return (kvm_hyp_handle_sysreg(vcpu, exit_code) ||
176 		kvm_handle_pvm_sysreg(vcpu, exit_code));
177 }
178 
179 /**
180  * Handler for protected floating-point and Advanced SIMD accesses.
181  *
182  * Returns true if the hypervisor has handled the exit, and control should go
183  * back to the guest, or false if it hasn't.
184  */
185 static bool kvm_handle_pvm_fpsimd(struct kvm_vcpu *vcpu, u64 *exit_code)
186 {
187 	/* Linux guests assume support for floating-point and Advanced SIMD. */
188 	BUILD_BUG_ON(!FIELD_GET(ARM64_FEATURE_MASK(ID_AA64PFR0_FP),
189 				PVM_ID_AA64PFR0_ALLOW));
190 	BUILD_BUG_ON(!FIELD_GET(ARM64_FEATURE_MASK(ID_AA64PFR0_ASIMD),
191 				PVM_ID_AA64PFR0_ALLOW));
192 
193 	return kvm_hyp_handle_fpsimd(vcpu, exit_code);
194 }
195 
196 static const exit_handler_fn hyp_exit_handlers[] = {
197 	[0 ... ESR_ELx_EC_MAX]		= NULL,
198 	[ESR_ELx_EC_CP15_32]		= kvm_hyp_handle_cp15_32,
199 	[ESR_ELx_EC_SYS64]		= kvm_hyp_handle_sysreg,
200 	[ESR_ELx_EC_SVE]		= kvm_hyp_handle_fpsimd,
201 	[ESR_ELx_EC_FP_ASIMD]		= kvm_hyp_handle_fpsimd,
202 	[ESR_ELx_EC_IABT_LOW]		= kvm_hyp_handle_iabt_low,
203 	[ESR_ELx_EC_DABT_LOW]		= kvm_hyp_handle_dabt_low,
204 	[ESR_ELx_EC_PAC]		= kvm_hyp_handle_ptrauth,
205 };
206 
207 static const exit_handler_fn pvm_exit_handlers[] = {
208 	[0 ... ESR_ELx_EC_MAX]		= NULL,
209 	[ESR_ELx_EC_SYS64]		= kvm_handle_pvm_sys64,
210 	[ESR_ELx_EC_SVE]		= kvm_handle_pvm_restricted,
211 	[ESR_ELx_EC_FP_ASIMD]		= kvm_handle_pvm_fpsimd,
212 	[ESR_ELx_EC_IABT_LOW]		= kvm_hyp_handle_iabt_low,
213 	[ESR_ELx_EC_DABT_LOW]		= kvm_hyp_handle_dabt_low,
214 	[ESR_ELx_EC_PAC]		= kvm_hyp_handle_ptrauth,
215 };
216 
217 static const exit_handler_fn *kvm_get_exit_handler_array(struct kvm_vcpu *vcpu)
218 {
219 	if (unlikely(kvm_vm_is_protected(kern_hyp_va(vcpu->kvm))))
220 		return pvm_exit_handlers;
221 
222 	return hyp_exit_handlers;
223 }
224 
225 /*
226  * Some guests (e.g., protected VMs) are not be allowed to run in AArch32.
227  * The ARMv8 architecture does not give the hypervisor a mechanism to prevent a
228  * guest from dropping to AArch32 EL0 if implemented by the CPU. If the
229  * hypervisor spots a guest in such a state ensure it is handled, and don't
230  * trust the host to spot or fix it.  The check below is based on the one in
231  * kvm_arch_vcpu_ioctl_run().
232  *
233  * Returns false if the guest ran in AArch32 when it shouldn't have, and
234  * thus should exit to the host, or true if a the guest run loop can continue.
235  */
236 static bool handle_aarch32_guest(struct kvm_vcpu *vcpu, u64 *exit_code)
237 {
238 	struct kvm *kvm = kern_hyp_va(vcpu->kvm);
239 
240 	if (kvm_vm_is_protected(kvm) && vcpu_mode_is_32bit(vcpu)) {
241 		/*
242 		 * As we have caught the guest red-handed, decide that it isn't
243 		 * fit for purpose anymore by making the vcpu invalid. The VMM
244 		 * can try and fix it by re-initializing the vcpu with
245 		 * KVM_ARM_VCPU_INIT, however, this is likely not possible for
246 		 * protected VMs.
247 		 */
248 		vcpu->arch.target = -1;
249 		*exit_code &= BIT(ARM_EXIT_WITH_SERROR_BIT);
250 		*exit_code |= ARM_EXCEPTION_IL;
251 		return false;
252 	}
253 
254 	return true;
255 }
256 
257 /* Switch to the guest for legacy non-VHE systems */
258 int __kvm_vcpu_run(struct kvm_vcpu *vcpu)
259 {
260 	struct kvm_cpu_context *host_ctxt;
261 	struct kvm_cpu_context *guest_ctxt;
262 	struct kvm_s2_mmu *mmu;
263 	bool pmu_switch_needed;
264 	u64 exit_code;
265 
266 	/*
267 	 * Having IRQs masked via PMR when entering the guest means the GIC
268 	 * will not signal the CPU of interrupts of lower priority, and the
269 	 * only way to get out will be via guest exceptions.
270 	 * Naturally, we want to avoid this.
271 	 */
272 	if (system_uses_irq_prio_masking()) {
273 		gic_write_pmr(GIC_PRIO_IRQON | GIC_PRIO_PSR_I_SET);
274 		pmr_sync();
275 	}
276 
277 	host_ctxt = &this_cpu_ptr(&kvm_host_data)->host_ctxt;
278 	host_ctxt->__hyp_running_vcpu = vcpu;
279 	guest_ctxt = &vcpu->arch.ctxt;
280 
281 	pmu_switch_needed = __pmu_switch_to_guest(host_ctxt);
282 
283 	__sysreg_save_state_nvhe(host_ctxt);
284 	/*
285 	 * We must flush and disable the SPE buffer for nVHE, as
286 	 * the translation regime(EL1&0) is going to be loaded with
287 	 * that of the guest. And we must do this before we change the
288 	 * translation regime to EL2 (via MDCR_EL2_E2PB == 0) and
289 	 * before we load guest Stage1.
290 	 */
291 	__debug_save_host_buffers_nvhe(vcpu);
292 
293 	__kvm_adjust_pc(vcpu);
294 
295 	/*
296 	 * We must restore the 32-bit state before the sysregs, thanks
297 	 * to erratum #852523 (Cortex-A57) or #853709 (Cortex-A72).
298 	 *
299 	 * Also, and in order to be able to deal with erratum #1319537 (A57)
300 	 * and #1319367 (A72), we must ensure that all VM-related sysreg are
301 	 * restored before we enable S2 translation.
302 	 */
303 	__sysreg32_restore_state(vcpu);
304 	__sysreg_restore_state_nvhe(guest_ctxt);
305 
306 	mmu = kern_hyp_va(vcpu->arch.hw_mmu);
307 	__load_stage2(mmu, kern_hyp_va(mmu->arch));
308 	__activate_traps(vcpu);
309 
310 	__hyp_vgic_restore_state(vcpu);
311 	__timer_enable_traps(vcpu);
312 
313 	__debug_switch_to_guest(vcpu);
314 
315 	do {
316 		/* Jump in the fire! */
317 		exit_code = __guest_enter(vcpu);
318 
319 		if (unlikely(!handle_aarch32_guest(vcpu, &exit_code)))
320 			break;
321 
322 		/* And we're baaack! */
323 	} while (fixup_guest_exit(vcpu, &exit_code));
324 
325 	__sysreg_save_state_nvhe(guest_ctxt);
326 	__sysreg32_save_state(vcpu);
327 	__timer_disable_traps(vcpu);
328 	__hyp_vgic_save_state(vcpu);
329 
330 	__deactivate_traps(vcpu);
331 	__load_host_stage2();
332 
333 	__sysreg_restore_state_nvhe(host_ctxt);
334 
335 	if (vcpu->arch.flags & KVM_ARM64_FP_ENABLED)
336 		__fpsimd_save_fpexc32(vcpu);
337 
338 	__debug_switch_to_host(vcpu);
339 	/*
340 	 * This must come after restoring the host sysregs, since a non-VHE
341 	 * system may enable SPE here and make use of the TTBRs.
342 	 */
343 	__debug_restore_host_buffers_nvhe(vcpu);
344 
345 	if (pmu_switch_needed)
346 		__pmu_switch_to_host(host_ctxt);
347 
348 	/* Returning to host will clear PSR.I, remask PMR if needed */
349 	if (system_uses_irq_prio_masking())
350 		gic_write_pmr(GIC_PRIO_IRQOFF);
351 
352 	host_ctxt->__hyp_running_vcpu = NULL;
353 
354 	return exit_code;
355 }
356 
357 void __noreturn hyp_panic(void)
358 {
359 	u64 spsr = read_sysreg_el2(SYS_SPSR);
360 	u64 elr = read_sysreg_el2(SYS_ELR);
361 	u64 par = read_sysreg_par();
362 	struct kvm_cpu_context *host_ctxt;
363 	struct kvm_vcpu *vcpu;
364 
365 	host_ctxt = &this_cpu_ptr(&kvm_host_data)->host_ctxt;
366 	vcpu = host_ctxt->__hyp_running_vcpu;
367 
368 	if (vcpu) {
369 		__timer_disable_traps(vcpu);
370 		__deactivate_traps(vcpu);
371 		__load_host_stage2();
372 		__sysreg_restore_state_nvhe(host_ctxt);
373 	}
374 
375 	__hyp_do_panic(host_ctxt, spsr, elr, par);
376 	unreachable();
377 }
378 
379 asmlinkage void kvm_unexpected_el2_exception(void)
380 {
381 	return __kvm_unexpected_el2_exception();
382 }
383