xref: /linux/arch/arm64/kvm/hyp/vhe/switch.c (revision ca93bf607a44c1f009283dac4af7df0d9ae5e357)
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 
9 #include <linux/arm-smccc.h>
10 #include <linux/kvm_host.h>
11 #include <linux/types.h>
12 #include <linux/jump_label.h>
13 #include <linux/percpu.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 #include <asm/vectors.h>
30 
31 /* VHE specific context */
32 DEFINE_PER_CPU(struct kvm_host_data, kvm_host_data);
33 DEFINE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt);
34 DEFINE_PER_CPU(unsigned long, kvm_hyp_vector);
35 
36 static void __activate_traps(struct kvm_vcpu *vcpu)
37 {
38 	u64 val;
39 
40 	___activate_traps(vcpu);
41 
42 	if (has_cntpoff()) {
43 		struct timer_map map;
44 
45 		get_timer_map(vcpu, &map);
46 
47 		/*
48 		 * We're entrering the guest. Reload the correct
49 		 * values from memory now that TGE is clear.
50 		 */
51 		if (map.direct_ptimer == vcpu_ptimer(vcpu))
52 			val = __vcpu_sys_reg(vcpu, CNTP_CVAL_EL0);
53 		if (map.direct_ptimer == vcpu_hptimer(vcpu))
54 			val = __vcpu_sys_reg(vcpu, CNTHP_CVAL_EL2);
55 
56 		if (map.direct_ptimer) {
57 			write_sysreg_el0(val, SYS_CNTP_CVAL);
58 			isb();
59 		}
60 	}
61 
62 	val = read_sysreg(cpacr_el1);
63 	val |= CPACR_ELx_TTA;
64 	val &= ~(CPACR_EL1_ZEN_EL0EN | CPACR_EL1_ZEN_EL1EN |
65 		 CPACR_EL1_SMEN_EL0EN | CPACR_EL1_SMEN_EL1EN);
66 
67 	/*
68 	 * With VHE (HCR.E2H == 1), accesses to CPACR_EL1 are routed to
69 	 * CPTR_EL2. In general, CPACR_EL1 has the same layout as CPTR_EL2,
70 	 * except for some missing controls, such as TAM.
71 	 * In this case, CPTR_EL2.TAM has the same position with or without
72 	 * VHE (HCR.E2H == 1) which allows us to use here the CPTR_EL2.TAM
73 	 * shift value for trapping the AMU accesses.
74 	 */
75 
76 	val |= CPTR_EL2_TAM;
77 
78 	if (guest_owns_fp_regs(vcpu)) {
79 		if (vcpu_has_sve(vcpu))
80 			val |= CPACR_EL1_ZEN_EL0EN | CPACR_EL1_ZEN_EL1EN;
81 	} else {
82 		val &= ~(CPACR_EL1_FPEN_EL0EN | CPACR_EL1_FPEN_EL1EN);
83 		__activate_traps_fpsimd32(vcpu);
84 	}
85 
86 	write_sysreg(val, cpacr_el1);
87 
88 	write_sysreg(__this_cpu_read(kvm_hyp_vector), vbar_el1);
89 }
90 NOKPROBE_SYMBOL(__activate_traps);
91 
92 static void __deactivate_traps(struct kvm_vcpu *vcpu)
93 {
94 	const char *host_vectors = vectors;
95 
96 	___deactivate_traps(vcpu);
97 
98 	write_sysreg(HCR_HOST_VHE_FLAGS, hcr_el2);
99 
100 	if (has_cntpoff()) {
101 		struct timer_map map;
102 		u64 val, offset;
103 
104 		get_timer_map(vcpu, &map);
105 
106 		/*
107 		 * We're exiting the guest. Save the latest CVAL value
108 		 * to memory and apply the offset now that TGE is set.
109 		 */
110 		val = read_sysreg_el0(SYS_CNTP_CVAL);
111 		if (map.direct_ptimer == vcpu_ptimer(vcpu))
112 			__vcpu_sys_reg(vcpu, CNTP_CVAL_EL0) = val;
113 		if (map.direct_ptimer == vcpu_hptimer(vcpu))
114 			__vcpu_sys_reg(vcpu, CNTHP_CVAL_EL2) = val;
115 
116 		offset = read_sysreg_s(SYS_CNTPOFF_EL2);
117 
118 		if (map.direct_ptimer && offset) {
119 			write_sysreg_el0(val + offset, SYS_CNTP_CVAL);
120 			isb();
121 		}
122 	}
123 
124 	/*
125 	 * ARM errata 1165522 and 1530923 require the actual execution of the
126 	 * above before we can switch to the EL2/EL0 translation regime used by
127 	 * the host.
128 	 */
129 	asm(ALTERNATIVE("nop", "isb", ARM64_WORKAROUND_SPECULATIVE_AT));
130 
131 	kvm_reset_cptr_el2(vcpu);
132 
133 	if (!arm64_kernel_unmapped_at_el0())
134 		host_vectors = __this_cpu_read(this_cpu_vector);
135 	write_sysreg(host_vectors, vbar_el1);
136 }
137 NOKPROBE_SYMBOL(__deactivate_traps);
138 
139 /*
140  * Disable IRQs in __vcpu_{load,put}_{activate,deactivate}_traps() to
141  * prevent a race condition between context switching of PMUSERENR_EL0
142  * in __{activate,deactivate}_traps_common() and IPIs that attempts to
143  * update PMUSERENR_EL0. See also kvm_set_pmuserenr().
144  */
145 static void __vcpu_load_activate_traps(struct kvm_vcpu *vcpu)
146 {
147 	unsigned long flags;
148 
149 	local_irq_save(flags);
150 	__activate_traps_common(vcpu);
151 	local_irq_restore(flags);
152 }
153 
154 static void __vcpu_put_deactivate_traps(struct kvm_vcpu *vcpu)
155 {
156 	unsigned long flags;
157 
158 	local_irq_save(flags);
159 	__deactivate_traps_common(vcpu);
160 	local_irq_restore(flags);
161 }
162 
163 void kvm_vcpu_load_vhe(struct kvm_vcpu *vcpu)
164 {
165 	__vcpu_load_switch_sysregs(vcpu);
166 	__vcpu_load_activate_traps(vcpu);
167 	__load_stage2(vcpu->arch.hw_mmu, vcpu->arch.hw_mmu->arch);
168 }
169 
170 void kvm_vcpu_put_vhe(struct kvm_vcpu *vcpu)
171 {
172 	__vcpu_put_deactivate_traps(vcpu);
173 	__vcpu_put_switch_sysregs(vcpu);
174 }
175 
176 static const exit_handler_fn hyp_exit_handlers[] = {
177 	[0 ... ESR_ELx_EC_MAX]		= NULL,
178 	[ESR_ELx_EC_CP15_32]		= kvm_hyp_handle_cp15_32,
179 	[ESR_ELx_EC_SYS64]		= kvm_hyp_handle_sysreg,
180 	[ESR_ELx_EC_SVE]		= kvm_hyp_handle_fpsimd,
181 	[ESR_ELx_EC_FP_ASIMD]		= kvm_hyp_handle_fpsimd,
182 	[ESR_ELx_EC_IABT_LOW]		= kvm_hyp_handle_iabt_low,
183 	[ESR_ELx_EC_DABT_LOW]		= kvm_hyp_handle_dabt_low,
184 	[ESR_ELx_EC_WATCHPT_LOW]	= kvm_hyp_handle_watchpt_low,
185 	[ESR_ELx_EC_PAC]		= kvm_hyp_handle_ptrauth,
186 	[ESR_ELx_EC_MOPS]		= kvm_hyp_handle_mops,
187 };
188 
189 static const exit_handler_fn *kvm_get_exit_handler_array(struct kvm_vcpu *vcpu)
190 {
191 	return hyp_exit_handlers;
192 }
193 
194 static void early_exit_filter(struct kvm_vcpu *vcpu, u64 *exit_code)
195 {
196 	/*
197 	 * If we were in HYP context on entry, adjust the PSTATE view
198 	 * so that the usual helpers work correctly.
199 	 */
200 	if (unlikely(vcpu_get_flag(vcpu, VCPU_HYP_CONTEXT))) {
201 		u64 mode = *vcpu_cpsr(vcpu) & (PSR_MODE_MASK | PSR_MODE32_BIT);
202 
203 		switch (mode) {
204 		case PSR_MODE_EL1t:
205 			mode = PSR_MODE_EL2t;
206 			break;
207 		case PSR_MODE_EL1h:
208 			mode = PSR_MODE_EL2h;
209 			break;
210 		}
211 
212 		*vcpu_cpsr(vcpu) &= ~(PSR_MODE_MASK | PSR_MODE32_BIT);
213 		*vcpu_cpsr(vcpu) |= mode;
214 	}
215 }
216 
217 /* Switch to the guest for VHE systems running in EL2 */
218 static int __kvm_vcpu_run_vhe(struct kvm_vcpu *vcpu)
219 {
220 	struct kvm_cpu_context *host_ctxt;
221 	struct kvm_cpu_context *guest_ctxt;
222 	u64 exit_code;
223 
224 	host_ctxt = &this_cpu_ptr(&kvm_host_data)->host_ctxt;
225 	host_ctxt->__hyp_running_vcpu = vcpu;
226 	guest_ctxt = &vcpu->arch.ctxt;
227 
228 	sysreg_save_host_state_vhe(host_ctxt);
229 
230 	/*
231 	 * Note that ARM erratum 1165522 requires us to configure both stage 1
232 	 * and stage 2 translation for the guest context before we clear
233 	 * HCR_EL2.TGE. The stage 1 and stage 2 guest context has already been
234 	 * loaded on the CPU in kvm_vcpu_load_vhe().
235 	 */
236 	__activate_traps(vcpu);
237 
238 	__kvm_adjust_pc(vcpu);
239 
240 	sysreg_restore_guest_state_vhe(guest_ctxt);
241 	__debug_switch_to_guest(vcpu);
242 
243 	if (is_hyp_ctxt(vcpu))
244 		vcpu_set_flag(vcpu, VCPU_HYP_CONTEXT);
245 	else
246 		vcpu_clear_flag(vcpu, VCPU_HYP_CONTEXT);
247 
248 	do {
249 		/* Jump in the fire! */
250 		exit_code = __guest_enter(vcpu);
251 
252 		/* And we're baaack! */
253 	} while (fixup_guest_exit(vcpu, &exit_code));
254 
255 	sysreg_save_guest_state_vhe(guest_ctxt);
256 
257 	__deactivate_traps(vcpu);
258 
259 	sysreg_restore_host_state_vhe(host_ctxt);
260 
261 	if (vcpu->arch.fp_state == FP_STATE_GUEST_OWNED)
262 		__fpsimd_save_fpexc32(vcpu);
263 
264 	__debug_switch_to_host(vcpu);
265 
266 	return exit_code;
267 }
268 NOKPROBE_SYMBOL(__kvm_vcpu_run_vhe);
269 
270 int __kvm_vcpu_run(struct kvm_vcpu *vcpu)
271 {
272 	int ret;
273 
274 	local_daif_mask();
275 
276 	/*
277 	 * Having IRQs masked via PMR when entering the guest means the GIC
278 	 * will not signal the CPU of interrupts of lower priority, and the
279 	 * only way to get out will be via guest exceptions.
280 	 * Naturally, we want to avoid this.
281 	 *
282 	 * local_daif_mask() already sets GIC_PRIO_PSR_I_SET, we just need a
283 	 * dsb to ensure the redistributor is forwards EL2 IRQs to the CPU.
284 	 */
285 	pmr_sync();
286 
287 	ret = __kvm_vcpu_run_vhe(vcpu);
288 
289 	/*
290 	 * local_daif_restore() takes care to properly restore PSTATE.DAIF
291 	 * and the GIC PMR if the host is using IRQ priorities.
292 	 */
293 	local_daif_restore(DAIF_PROCCTX_NOIRQ);
294 
295 	/*
296 	 * When we exit from the guest we change a number of CPU configuration
297 	 * parameters, such as traps.  We rely on the isb() in kvm_call_hyp*()
298 	 * to make sure these changes take effect before running the host or
299 	 * additional guests.
300 	 */
301 	return ret;
302 }
303 
304 static void __hyp_call_panic(u64 spsr, u64 elr, u64 par)
305 {
306 	struct kvm_cpu_context *host_ctxt;
307 	struct kvm_vcpu *vcpu;
308 
309 	host_ctxt = &this_cpu_ptr(&kvm_host_data)->host_ctxt;
310 	vcpu = host_ctxt->__hyp_running_vcpu;
311 
312 	__deactivate_traps(vcpu);
313 	sysreg_restore_host_state_vhe(host_ctxt);
314 
315 	panic("HYP panic:\nPS:%08llx PC:%016llx ESR:%08llx\nFAR:%016llx HPFAR:%016llx PAR:%016llx\nVCPU:%p\n",
316 	      spsr, elr,
317 	      read_sysreg_el2(SYS_ESR), read_sysreg_el2(SYS_FAR),
318 	      read_sysreg(hpfar_el2), par, vcpu);
319 }
320 NOKPROBE_SYMBOL(__hyp_call_panic);
321 
322 void __noreturn hyp_panic(void)
323 {
324 	u64 spsr = read_sysreg_el2(SYS_SPSR);
325 	u64 elr = read_sysreg_el2(SYS_ELR);
326 	u64 par = read_sysreg_par();
327 
328 	__hyp_call_panic(spsr, elr, par);
329 	unreachable();
330 }
331 
332 asmlinkage void kvm_unexpected_el2_exception(void)
333 {
334 	__kvm_unexpected_el2_exception();
335 }
336