xref: /openbmc/linux/arch/arm64/kvm/hyp/vhe/switch.c (revision 83146efc)
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 {activate,deactivate}_traps_vhe_{load,put}() 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 void activate_traps_vhe_load(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 void deactivate_traps_vhe_put(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 static const exit_handler_fn hyp_exit_handlers[] = {
164 	[0 ... ESR_ELx_EC_MAX]		= NULL,
165 	[ESR_ELx_EC_CP15_32]		= kvm_hyp_handle_cp15_32,
166 	[ESR_ELx_EC_SYS64]		= kvm_hyp_handle_sysreg,
167 	[ESR_ELx_EC_SVE]		= kvm_hyp_handle_fpsimd,
168 	[ESR_ELx_EC_FP_ASIMD]		= kvm_hyp_handle_fpsimd,
169 	[ESR_ELx_EC_IABT_LOW]		= kvm_hyp_handle_iabt_low,
170 	[ESR_ELx_EC_DABT_LOW]		= kvm_hyp_handle_dabt_low,
171 	[ESR_ELx_EC_WATCHPT_LOW]	= kvm_hyp_handle_watchpt_low,
172 	[ESR_ELx_EC_PAC]		= kvm_hyp_handle_ptrauth,
173 };
174 
175 static const exit_handler_fn *kvm_get_exit_handler_array(struct kvm_vcpu *vcpu)
176 {
177 	return hyp_exit_handlers;
178 }
179 
180 static void early_exit_filter(struct kvm_vcpu *vcpu, u64 *exit_code)
181 {
182 	/*
183 	 * If we were in HYP context on entry, adjust the PSTATE view
184 	 * so that the usual helpers work correctly.
185 	 */
186 	if (unlikely(vcpu_get_flag(vcpu, VCPU_HYP_CONTEXT))) {
187 		u64 mode = *vcpu_cpsr(vcpu) & (PSR_MODE_MASK | PSR_MODE32_BIT);
188 
189 		switch (mode) {
190 		case PSR_MODE_EL1t:
191 			mode = PSR_MODE_EL2t;
192 			break;
193 		case PSR_MODE_EL1h:
194 			mode = PSR_MODE_EL2h;
195 			break;
196 		}
197 
198 		*vcpu_cpsr(vcpu) &= ~(PSR_MODE_MASK | PSR_MODE32_BIT);
199 		*vcpu_cpsr(vcpu) |= mode;
200 	}
201 }
202 
203 /* Switch to the guest for VHE systems running in EL2 */
204 static int __kvm_vcpu_run_vhe(struct kvm_vcpu *vcpu)
205 {
206 	struct kvm_cpu_context *host_ctxt;
207 	struct kvm_cpu_context *guest_ctxt;
208 	u64 exit_code;
209 
210 	host_ctxt = &this_cpu_ptr(&kvm_host_data)->host_ctxt;
211 	host_ctxt->__hyp_running_vcpu = vcpu;
212 	guest_ctxt = &vcpu->arch.ctxt;
213 
214 	sysreg_save_host_state_vhe(host_ctxt);
215 
216 	/*
217 	 * ARM erratum 1165522 requires us to configure both stage 1 and
218 	 * stage 2 translation for the guest context before we clear
219 	 * HCR_EL2.TGE.
220 	 *
221 	 * We have already configured the guest's stage 1 translation in
222 	 * kvm_vcpu_load_sysregs_vhe above.  We must now call
223 	 * __load_stage2 before __activate_traps, because
224 	 * __load_stage2 configures stage 2 translation, and
225 	 * __activate_traps clear HCR_EL2.TGE (among other things).
226 	 */
227 	__load_stage2(vcpu->arch.hw_mmu, vcpu->arch.hw_mmu->arch);
228 	__activate_traps(vcpu);
229 
230 	__kvm_adjust_pc(vcpu);
231 
232 	sysreg_restore_guest_state_vhe(guest_ctxt);
233 	__debug_switch_to_guest(vcpu);
234 
235 	if (is_hyp_ctxt(vcpu))
236 		vcpu_set_flag(vcpu, VCPU_HYP_CONTEXT);
237 	else
238 		vcpu_clear_flag(vcpu, VCPU_HYP_CONTEXT);
239 
240 	do {
241 		/* Jump in the fire! */
242 		exit_code = __guest_enter(vcpu);
243 
244 		/* And we're baaack! */
245 	} while (fixup_guest_exit(vcpu, &exit_code));
246 
247 	sysreg_save_guest_state_vhe(guest_ctxt);
248 
249 	__deactivate_traps(vcpu);
250 
251 	sysreg_restore_host_state_vhe(host_ctxt);
252 
253 	if (vcpu->arch.fp_state == FP_STATE_GUEST_OWNED)
254 		__fpsimd_save_fpexc32(vcpu);
255 
256 	__debug_switch_to_host(vcpu);
257 
258 	return exit_code;
259 }
260 NOKPROBE_SYMBOL(__kvm_vcpu_run_vhe);
261 
262 int __kvm_vcpu_run(struct kvm_vcpu *vcpu)
263 {
264 	int ret;
265 
266 	local_daif_mask();
267 
268 	/*
269 	 * Having IRQs masked via PMR when entering the guest means the GIC
270 	 * will not signal the CPU of interrupts of lower priority, and the
271 	 * only way to get out will be via guest exceptions.
272 	 * Naturally, we want to avoid this.
273 	 *
274 	 * local_daif_mask() already sets GIC_PRIO_PSR_I_SET, we just need a
275 	 * dsb to ensure the redistributor is forwards EL2 IRQs to the CPU.
276 	 */
277 	pmr_sync();
278 
279 	ret = __kvm_vcpu_run_vhe(vcpu);
280 
281 	/*
282 	 * local_daif_restore() takes care to properly restore PSTATE.DAIF
283 	 * and the GIC PMR if the host is using IRQ priorities.
284 	 */
285 	local_daif_restore(DAIF_PROCCTX_NOIRQ);
286 
287 	/*
288 	 * When we exit from the guest we change a number of CPU configuration
289 	 * parameters, such as traps.  We rely on the isb() in kvm_call_hyp*()
290 	 * to make sure these changes take effect before running the host or
291 	 * additional guests.
292 	 */
293 	return ret;
294 }
295 
296 static void __hyp_call_panic(u64 spsr, u64 elr, u64 par)
297 {
298 	struct kvm_cpu_context *host_ctxt;
299 	struct kvm_vcpu *vcpu;
300 
301 	host_ctxt = &this_cpu_ptr(&kvm_host_data)->host_ctxt;
302 	vcpu = host_ctxt->__hyp_running_vcpu;
303 
304 	__deactivate_traps(vcpu);
305 	sysreg_restore_host_state_vhe(host_ctxt);
306 
307 	panic("HYP panic:\nPS:%08llx PC:%016llx ESR:%08llx\nFAR:%016llx HPFAR:%016llx PAR:%016llx\nVCPU:%p\n",
308 	      spsr, elr,
309 	      read_sysreg_el2(SYS_ESR), read_sysreg_el2(SYS_FAR),
310 	      read_sysreg(hpfar_el2), par, vcpu);
311 }
312 NOKPROBE_SYMBOL(__hyp_call_panic);
313 
314 void __noreturn hyp_panic(void)
315 {
316 	u64 spsr = read_sysreg_el2(SYS_SPSR);
317 	u64 elr = read_sysreg_el2(SYS_ELR);
318 	u64 par = read_sysreg_par();
319 
320 	__hyp_call_panic(spsr, elr, par);
321 	unreachable();
322 }
323 
324 asmlinkage void kvm_unexpected_el2_exception(void)
325 {
326 	__kvm_unexpected_el2_exception();
327 }
328