1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2015 - ARM Ltd
4  * Author: Marc Zyngier <[email protected]>
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 /*
37  * HCR_EL2 bits that the NV guest can freely change (no RES0/RES1
38  * semantics, irrespective of the configuration), but that cannot be
39  * applied to the actual HW as things would otherwise break badly.
40  *
41  * - TGE: we want the guest to use EL1, which is incompatible with
42  *   this bit being set
43  *
44  * - API/APK: they are already accounted for by vcpu_load(), and can
45  *   only take effect across a load/put cycle (such as ERET)
46  */
47 #define NV_HCR_GUEST_EXCLUDE	(HCR_TGE | HCR_API | HCR_APK)
48 
__compute_hcr(struct kvm_vcpu * vcpu)49 static u64 __compute_hcr(struct kvm_vcpu *vcpu)
50 {
51 	u64 hcr = vcpu->arch.hcr_el2;
52 
53 	if (!vcpu_has_nv(vcpu))
54 		return hcr;
55 
56 	if (is_hyp_ctxt(vcpu)) {
57 		hcr |= HCR_NV | HCR_NV2 | HCR_AT | HCR_TTLB;
58 
59 		if (!vcpu_el2_e2h_is_set(vcpu))
60 			hcr |= HCR_NV1;
61 
62 		write_sysreg_s(vcpu->arch.ctxt.vncr_array, SYS_VNCR_EL2);
63 	}
64 
65 	return hcr | (__vcpu_sys_reg(vcpu, HCR_EL2) & ~NV_HCR_GUEST_EXCLUDE);
66 }
67 
__activate_cptr_traps(struct kvm_vcpu * vcpu)68 static void __activate_cptr_traps(struct kvm_vcpu *vcpu)
69 {
70 	u64 cptr;
71 
72 	/*
73 	 * With VHE (HCR.E2H == 1), accesses to CPACR_EL1 are routed to
74 	 * CPTR_EL2. In general, CPACR_EL1 has the same layout as CPTR_EL2,
75 	 * except for some missing controls, such as TAM.
76 	 * In this case, CPTR_EL2.TAM has the same position with or without
77 	 * VHE (HCR.E2H == 1) which allows us to use here the CPTR_EL2.TAM
78 	 * shift value for trapping the AMU accesses.
79 	 */
80 	u64 val = CPACR_EL1_TTA | CPTR_EL2_TAM;
81 
82 	if (guest_owns_fp_regs()) {
83 		val |= CPACR_EL1_FPEN;
84 		if (vcpu_has_sve(vcpu))
85 			val |= CPACR_EL1_ZEN;
86 	} else {
87 		__activate_traps_fpsimd32(vcpu);
88 	}
89 
90 	if (!vcpu_has_nv(vcpu))
91 		goto write;
92 
93 	/*
94 	 * The architecture is a bit crap (what a surprise): an EL2 guest
95 	 * writing to CPTR_EL2 via CPACR_EL1 can't set any of TCPAC or TTA,
96 	 * as they are RES0 in the guest's view. To work around it, trap the
97 	 * sucker using the very same bit it can't set...
98 	 */
99 	if (vcpu_el2_e2h_is_set(vcpu) && is_hyp_ctxt(vcpu))
100 		val |= CPTR_EL2_TCPAC;
101 
102 	/*
103 	 * Layer the guest hypervisor's trap configuration on top of our own if
104 	 * we're in a nested context.
105 	 */
106 	if (is_hyp_ctxt(vcpu))
107 		goto write;
108 
109 	cptr = vcpu_sanitised_cptr_el2(vcpu);
110 
111 	/*
112 	 * Pay attention, there's some interesting detail here.
113 	 *
114 	 * The CPTR_EL2.xEN fields are 2 bits wide, although there are only two
115 	 * meaningful trap states when HCR_EL2.TGE = 0 (running a nested guest):
116 	 *
117 	 *  - CPTR_EL2.xEN = x0, traps are enabled
118 	 *  - CPTR_EL2.xEN = x1, traps are disabled
119 	 *
120 	 * In other words, bit[0] determines if guest accesses trap or not. In
121 	 * the interest of simplicity, clear the entire field if the guest
122 	 * hypervisor has traps enabled to dispel any illusion of something more
123 	 * complicated taking place.
124 	 */
125 	if (!(SYS_FIELD_GET(CPACR_EL1, FPEN, cptr) & BIT(0)))
126 		val &= ~CPACR_EL1_FPEN;
127 	if (!(SYS_FIELD_GET(CPACR_EL1, ZEN, cptr) & BIT(0)))
128 		val &= ~CPACR_EL1_ZEN;
129 
130 	if (kvm_has_feat(vcpu->kvm, ID_AA64MMFR3_EL1, S2POE, IMP))
131 		val |= cptr & CPACR_EL1_E0POE;
132 
133 	val |= cptr & CPTR_EL2_TCPAC;
134 
135 write:
136 	write_sysreg(val, cpacr_el1);
137 }
138 
__deactivate_cptr_traps(struct kvm_vcpu * vcpu)139 static void __deactivate_cptr_traps(struct kvm_vcpu *vcpu)
140 {
141 	u64 val = CPACR_EL1_FPEN | CPACR_EL1_ZEN_EL1EN;
142 
143 	if (cpus_have_final_cap(ARM64_SME))
144 		val |= CPACR_EL1_SMEN_EL1EN;
145 
146 	write_sysreg(val, cpacr_el1);
147 }
148 
__activate_traps(struct kvm_vcpu * vcpu)149 static void __activate_traps(struct kvm_vcpu *vcpu)
150 {
151 	u64 val;
152 
153 	___activate_traps(vcpu, __compute_hcr(vcpu));
154 
155 	if (has_cntpoff()) {
156 		struct timer_map map;
157 
158 		get_timer_map(vcpu, &map);
159 
160 		/*
161 		 * We're entrering the guest. Reload the correct
162 		 * values from memory now that TGE is clear.
163 		 */
164 		if (map.direct_ptimer == vcpu_ptimer(vcpu))
165 			val = __vcpu_sys_reg(vcpu, CNTP_CVAL_EL0);
166 		if (map.direct_ptimer == vcpu_hptimer(vcpu))
167 			val = __vcpu_sys_reg(vcpu, CNTHP_CVAL_EL2);
168 
169 		if (map.direct_ptimer) {
170 			write_sysreg_el0(val, SYS_CNTP_CVAL);
171 			isb();
172 		}
173 	}
174 
175 	__activate_cptr_traps(vcpu);
176 
177 	write_sysreg(__this_cpu_read(kvm_hyp_vector), vbar_el1);
178 }
179 NOKPROBE_SYMBOL(__activate_traps);
180 
__deactivate_traps(struct kvm_vcpu * vcpu)181 static void __deactivate_traps(struct kvm_vcpu *vcpu)
182 {
183 	const char *host_vectors = vectors;
184 
185 	___deactivate_traps(vcpu);
186 
187 	write_sysreg(HCR_HOST_VHE_FLAGS, hcr_el2);
188 
189 	if (has_cntpoff()) {
190 		struct timer_map map;
191 		u64 val, offset;
192 
193 		get_timer_map(vcpu, &map);
194 
195 		/*
196 		 * We're exiting the guest. Save the latest CVAL value
197 		 * to memory and apply the offset now that TGE is set.
198 		 */
199 		val = read_sysreg_el0(SYS_CNTP_CVAL);
200 		if (map.direct_ptimer == vcpu_ptimer(vcpu))
201 			__vcpu_sys_reg(vcpu, CNTP_CVAL_EL0) = val;
202 		if (map.direct_ptimer == vcpu_hptimer(vcpu))
203 			__vcpu_sys_reg(vcpu, CNTHP_CVAL_EL2) = val;
204 
205 		offset = read_sysreg_s(SYS_CNTPOFF_EL2);
206 
207 		if (map.direct_ptimer && offset) {
208 			write_sysreg_el0(val + offset, SYS_CNTP_CVAL);
209 			isb();
210 		}
211 	}
212 
213 	/*
214 	 * ARM errata 1165522 and 1530923 require the actual execution of the
215 	 * above before we can switch to the EL2/EL0 translation regime used by
216 	 * the host.
217 	 */
218 	asm(ALTERNATIVE("nop", "isb", ARM64_WORKAROUND_SPECULATIVE_AT));
219 
220 	__deactivate_cptr_traps(vcpu);
221 
222 	if (!arm64_kernel_unmapped_at_el0())
223 		host_vectors = __this_cpu_read(this_cpu_vector);
224 	write_sysreg(host_vectors, vbar_el1);
225 }
226 NOKPROBE_SYMBOL(__deactivate_traps);
227 
228 /*
229  * Disable IRQs in __vcpu_{load,put}_{activate,deactivate}_traps() to
230  * prevent a race condition between context switching of PMUSERENR_EL0
231  * in __{activate,deactivate}_traps_common() and IPIs that attempts to
232  * update PMUSERENR_EL0. See also kvm_set_pmuserenr().
233  */
__vcpu_load_activate_traps(struct kvm_vcpu * vcpu)234 static void __vcpu_load_activate_traps(struct kvm_vcpu *vcpu)
235 {
236 	unsigned long flags;
237 
238 	local_irq_save(flags);
239 	__activate_traps_common(vcpu);
240 	local_irq_restore(flags);
241 }
242 
__vcpu_put_deactivate_traps(struct kvm_vcpu * vcpu)243 static void __vcpu_put_deactivate_traps(struct kvm_vcpu *vcpu)
244 {
245 	unsigned long flags;
246 
247 	local_irq_save(flags);
248 	__deactivate_traps_common(vcpu);
249 	local_irq_restore(flags);
250 }
251 
kvm_vcpu_load_vhe(struct kvm_vcpu * vcpu)252 void kvm_vcpu_load_vhe(struct kvm_vcpu *vcpu)
253 {
254 	host_data_ptr(host_ctxt)->__hyp_running_vcpu = vcpu;
255 
256 	__vcpu_load_switch_sysregs(vcpu);
257 	__vcpu_load_activate_traps(vcpu);
258 	__load_stage2(vcpu->arch.hw_mmu, vcpu->arch.hw_mmu->arch);
259 }
260 
kvm_vcpu_put_vhe(struct kvm_vcpu * vcpu)261 void kvm_vcpu_put_vhe(struct kvm_vcpu *vcpu)
262 {
263 	__vcpu_put_deactivate_traps(vcpu);
264 	__vcpu_put_switch_sysregs(vcpu);
265 
266 	host_data_ptr(host_ctxt)->__hyp_running_vcpu = NULL;
267 }
268 
compute_emulated_cntx_ctl_el0(struct kvm_vcpu * vcpu,enum vcpu_sysreg reg)269 static u64 compute_emulated_cntx_ctl_el0(struct kvm_vcpu *vcpu,
270 					 enum vcpu_sysreg reg)
271 {
272 	unsigned long ctl;
273 	u64 cval, cnt;
274 	bool stat;
275 
276 	switch (reg) {
277 	case CNTP_CTL_EL0:
278 		cval = __vcpu_sys_reg(vcpu, CNTP_CVAL_EL0);
279 		ctl  = __vcpu_sys_reg(vcpu, CNTP_CTL_EL0);
280 		cnt  = compute_counter_value(vcpu_ptimer(vcpu));
281 		break;
282 	case CNTV_CTL_EL0:
283 		cval = __vcpu_sys_reg(vcpu, CNTV_CVAL_EL0);
284 		ctl  = __vcpu_sys_reg(vcpu, CNTV_CTL_EL0);
285 		cnt  = compute_counter_value(vcpu_vtimer(vcpu));
286 		break;
287 	default:
288 		BUG();
289 	}
290 
291 	stat = cval <= cnt;
292 	__assign_bit(__ffs(ARCH_TIMER_CTRL_IT_STAT), &ctl, stat);
293 
294 	return ctl;
295 }
296 
kvm_hyp_handle_timer(struct kvm_vcpu * vcpu,u64 * exit_code)297 static bool kvm_hyp_handle_timer(struct kvm_vcpu *vcpu, u64 *exit_code)
298 {
299 	u64 esr, val;
300 
301 	/*
302 	 * Having FEAT_ECV allows for a better quality of timer emulation.
303 	 * However, this comes at a huge cost in terms of traps. Try and
304 	 * satisfy the reads from guest's hypervisor context without
305 	 * returning to the kernel if we can.
306 	 */
307 	if (!is_hyp_ctxt(vcpu))
308 		return false;
309 
310 	esr = kvm_vcpu_get_esr(vcpu);
311 	if ((esr & ESR_ELx_SYS64_ISS_DIR_MASK) != ESR_ELx_SYS64_ISS_DIR_READ)
312 		return false;
313 
314 	switch (esr_sys64_to_sysreg(esr)) {
315 	case SYS_CNTP_CTL_EL02:
316 		val = compute_emulated_cntx_ctl_el0(vcpu, CNTP_CTL_EL0);
317 		break;
318 	case SYS_CNTP_CTL_EL0:
319 		if (vcpu_el2_e2h_is_set(vcpu))
320 			val = read_sysreg_el0(SYS_CNTP_CTL);
321 		else
322 			val = compute_emulated_cntx_ctl_el0(vcpu, CNTP_CTL_EL0);
323 		break;
324 	case SYS_CNTP_CVAL_EL02:
325 		val = __vcpu_sys_reg(vcpu, CNTP_CVAL_EL0);
326 		break;
327 	case SYS_CNTP_CVAL_EL0:
328 		if (vcpu_el2_e2h_is_set(vcpu)) {
329 			val = read_sysreg_el0(SYS_CNTP_CVAL);
330 
331 			if (!has_cntpoff())
332 				val -= timer_get_offset(vcpu_hptimer(vcpu));
333 		} else {
334 			val = __vcpu_sys_reg(vcpu, CNTP_CVAL_EL0);
335 		}
336 		break;
337 	case SYS_CNTPCT_EL0:
338 	case SYS_CNTPCTSS_EL0:
339 		val = compute_counter_value(vcpu_hptimer(vcpu));
340 		break;
341 	case SYS_CNTV_CTL_EL02:
342 		val = compute_emulated_cntx_ctl_el0(vcpu, CNTV_CTL_EL0);
343 		break;
344 	case SYS_CNTV_CTL_EL0:
345 		if (vcpu_el2_e2h_is_set(vcpu))
346 			val = read_sysreg_el0(SYS_CNTV_CTL);
347 		else
348 			val = compute_emulated_cntx_ctl_el0(vcpu, CNTV_CTL_EL0);
349 		break;
350 	case SYS_CNTV_CVAL_EL02:
351 		val = __vcpu_sys_reg(vcpu, CNTV_CVAL_EL0);
352 		break;
353 	case SYS_CNTV_CVAL_EL0:
354 		if (vcpu_el2_e2h_is_set(vcpu))
355 			val = read_sysreg_el0(SYS_CNTV_CVAL);
356 		else
357 			val = __vcpu_sys_reg(vcpu, CNTV_CVAL_EL0);
358 		break;
359 	case SYS_CNTVCT_EL0:
360 	case SYS_CNTVCTSS_EL0:
361 		val = compute_counter_value(vcpu_hvtimer(vcpu));
362 		break;
363 	default:
364 		return false;
365 	}
366 
367 	vcpu_set_reg(vcpu, kvm_vcpu_sys_get_rt(vcpu), val);
368 	__kvm_skip_instr(vcpu);
369 
370 	return true;
371 }
372 
kvm_hyp_handle_eret(struct kvm_vcpu * vcpu,u64 * exit_code)373 static bool kvm_hyp_handle_eret(struct kvm_vcpu *vcpu, u64 *exit_code)
374 {
375 	u64 esr = kvm_vcpu_get_esr(vcpu);
376 	u64 spsr, elr, mode;
377 
378 	/*
379 	 * Going through the whole put/load motions is a waste of time
380 	 * if this is a VHE guest hypervisor returning to its own
381 	 * userspace, or the hypervisor performing a local exception
382 	 * return. No need to save/restore registers, no need to
383 	 * switch S2 MMU. Just do the canonical ERET.
384 	 *
385 	 * Unless the trap has to be forwarded further down the line,
386 	 * of course...
387 	 */
388 	if ((__vcpu_sys_reg(vcpu, HCR_EL2) & HCR_NV) ||
389 	    (__vcpu_sys_reg(vcpu, HFGITR_EL2) & HFGITR_EL2_ERET))
390 		return false;
391 
392 	spsr = read_sysreg_el1(SYS_SPSR);
393 	mode = spsr & (PSR_MODE_MASK | PSR_MODE32_BIT);
394 
395 	switch (mode) {
396 	case PSR_MODE_EL0t:
397 		if (!(vcpu_el2_e2h_is_set(vcpu) && vcpu_el2_tge_is_set(vcpu)))
398 			return false;
399 		break;
400 	case PSR_MODE_EL2t:
401 		mode = PSR_MODE_EL1t;
402 		break;
403 	case PSR_MODE_EL2h:
404 		mode = PSR_MODE_EL1h;
405 		break;
406 	default:
407 		return false;
408 	}
409 
410 	/* If ERETAx fails, take the slow path */
411 	if (esr_iss_is_eretax(esr)) {
412 		if (!(vcpu_has_ptrauth(vcpu) && kvm_auth_eretax(vcpu, &elr)))
413 			return false;
414 	} else {
415 		elr = read_sysreg_el1(SYS_ELR);
416 	}
417 
418 	spsr = (spsr & ~(PSR_MODE_MASK | PSR_MODE32_BIT)) | mode;
419 
420 	write_sysreg_el2(spsr, SYS_SPSR);
421 	write_sysreg_el2(elr, SYS_ELR);
422 
423 	return true;
424 }
425 
kvm_hyp_handle_tlbi_el2(struct kvm_vcpu * vcpu,u64 * exit_code)426 static bool kvm_hyp_handle_tlbi_el2(struct kvm_vcpu *vcpu, u64 *exit_code)
427 {
428 	int ret = -EINVAL;
429 	u32 instr;
430 	u64 val;
431 
432 	/*
433 	 * Ideally, we would never trap on EL2 S1 TLB invalidations using
434 	 * the EL1 instructions when the guest's HCR_EL2.{E2H,TGE}=={1,1}.
435 	 * But "thanks" to FEAT_NV2, we don't trap writes to HCR_EL2,
436 	 * meaning that we can't track changes to the virtual TGE bit. So we
437 	 * have to leave HCR_EL2.TTLB set on the host. Oopsie...
438 	 *
439 	 * Try and handle these invalidation as quickly as possible, without
440 	 * fully exiting. Note that we don't need to consider any forwarding
441 	 * here, as having E2H+TGE set is the very definition of being
442 	 * InHost.
443 	 *
444 	 * For the lesser hypervisors out there that have failed to get on
445 	 * with the VHE program, we can also handle the nVHE style of EL2
446 	 * invalidation.
447 	 */
448 	if (!(is_hyp_ctxt(vcpu)))
449 		return false;
450 
451 	instr = esr_sys64_to_sysreg(kvm_vcpu_get_esr(vcpu));
452 	val = vcpu_get_reg(vcpu, kvm_vcpu_sys_get_rt(vcpu));
453 
454 	if ((kvm_supported_tlbi_s1e1_op(vcpu, instr) &&
455 	     vcpu_el2_e2h_is_set(vcpu) && vcpu_el2_tge_is_set(vcpu)) ||
456 	    kvm_supported_tlbi_s1e2_op (vcpu, instr))
457 		ret = __kvm_tlbi_s1e2(NULL, val, instr);
458 
459 	if (ret)
460 		return false;
461 
462 	__kvm_skip_instr(vcpu);
463 
464 	return true;
465 }
466 
kvm_hyp_handle_cpacr_el1(struct kvm_vcpu * vcpu,u64 * exit_code)467 static bool kvm_hyp_handle_cpacr_el1(struct kvm_vcpu *vcpu, u64 *exit_code)
468 {
469 	u64 esr = kvm_vcpu_get_esr(vcpu);
470 	int rt;
471 
472 	if (!is_hyp_ctxt(vcpu) || esr_sys64_to_sysreg(esr) != SYS_CPACR_EL1)
473 		return false;
474 
475 	rt = kvm_vcpu_sys_get_rt(vcpu);
476 
477 	if ((esr & ESR_ELx_SYS64_ISS_DIR_MASK) == ESR_ELx_SYS64_ISS_DIR_READ) {
478 		vcpu_set_reg(vcpu, rt, __vcpu_sys_reg(vcpu, CPTR_EL2));
479 	} else {
480 		vcpu_write_sys_reg(vcpu, vcpu_get_reg(vcpu, rt), CPTR_EL2);
481 		__activate_cptr_traps(vcpu);
482 	}
483 
484 	__kvm_skip_instr(vcpu);
485 
486 	return true;
487 }
488 
kvm_hyp_handle_zcr_el2(struct kvm_vcpu * vcpu,u64 * exit_code)489 static bool kvm_hyp_handle_zcr_el2(struct kvm_vcpu *vcpu, u64 *exit_code)
490 {
491 	u32 sysreg = esr_sys64_to_sysreg(kvm_vcpu_get_esr(vcpu));
492 
493 	if (!vcpu_has_nv(vcpu))
494 		return false;
495 
496 	if (sysreg != SYS_ZCR_EL2)
497 		return false;
498 
499 	if (guest_owns_fp_regs())
500 		return false;
501 
502 	/*
503 	 * ZCR_EL2 traps are handled in the slow path, with the expectation
504 	 * that the guest's FP context has already been loaded onto the CPU.
505 	 *
506 	 * Load the guest's FP context and unconditionally forward to the
507 	 * slow path for handling (i.e. return false).
508 	 */
509 	kvm_hyp_handle_fpsimd(vcpu, exit_code);
510 	return false;
511 }
512 
kvm_hyp_handle_sysreg_vhe(struct kvm_vcpu * vcpu,u64 * exit_code)513 static bool kvm_hyp_handle_sysreg_vhe(struct kvm_vcpu *vcpu, u64 *exit_code)
514 {
515 	if (kvm_hyp_handle_tlbi_el2(vcpu, exit_code))
516 		return true;
517 
518 	if (kvm_hyp_handle_timer(vcpu, exit_code))
519 		return true;
520 
521 	if (kvm_hyp_handle_cpacr_el1(vcpu, exit_code))
522 		return true;
523 
524 	if (kvm_hyp_handle_zcr_el2(vcpu, exit_code))
525 		return true;
526 
527 	return kvm_hyp_handle_sysreg(vcpu, exit_code);
528 }
529 
530 static const exit_handler_fn hyp_exit_handlers[] = {
531 	[0 ... ESR_ELx_EC_MAX]		= NULL,
532 	[ESR_ELx_EC_CP15_32]		= kvm_hyp_handle_cp15_32,
533 	[ESR_ELx_EC_SYS64]		= kvm_hyp_handle_sysreg_vhe,
534 	[ESR_ELx_EC_SVE]		= kvm_hyp_handle_fpsimd,
535 	[ESR_ELx_EC_FP_ASIMD]		= kvm_hyp_handle_fpsimd,
536 	[ESR_ELx_EC_IABT_LOW]		= kvm_hyp_handle_iabt_low,
537 	[ESR_ELx_EC_DABT_LOW]		= kvm_hyp_handle_dabt_low,
538 	[ESR_ELx_EC_WATCHPT_LOW]	= kvm_hyp_handle_watchpt_low,
539 	[ESR_ELx_EC_ERET]		= kvm_hyp_handle_eret,
540 	[ESR_ELx_EC_MOPS]		= kvm_hyp_handle_mops,
541 };
542 
fixup_guest_exit(struct kvm_vcpu * vcpu,u64 * exit_code)543 static inline bool fixup_guest_exit(struct kvm_vcpu *vcpu, u64 *exit_code)
544 {
545 	synchronize_vcpu_pstate(vcpu, exit_code);
546 
547 	/*
548 	 * If we were in HYP context on entry, adjust the PSTATE view
549 	 * so that the usual helpers work correctly.
550 	 */
551 	if (vcpu_has_nv(vcpu) && (read_sysreg(hcr_el2) & HCR_NV)) {
552 		u64 mode = *vcpu_cpsr(vcpu) & (PSR_MODE_MASK | PSR_MODE32_BIT);
553 
554 		switch (mode) {
555 		case PSR_MODE_EL1t:
556 			mode = PSR_MODE_EL2t;
557 			break;
558 		case PSR_MODE_EL1h:
559 			mode = PSR_MODE_EL2h;
560 			break;
561 		}
562 
563 		*vcpu_cpsr(vcpu) &= ~(PSR_MODE_MASK | PSR_MODE32_BIT);
564 		*vcpu_cpsr(vcpu) |= mode;
565 	}
566 
567 	return __fixup_guest_exit(vcpu, exit_code, hyp_exit_handlers);
568 }
569 
570 /* Switch to the guest for VHE systems running in EL2 */
__kvm_vcpu_run_vhe(struct kvm_vcpu * vcpu)571 static int __kvm_vcpu_run_vhe(struct kvm_vcpu *vcpu)
572 {
573 	struct kvm_cpu_context *host_ctxt;
574 	struct kvm_cpu_context *guest_ctxt;
575 	u64 exit_code;
576 
577 	host_ctxt = host_data_ptr(host_ctxt);
578 	guest_ctxt = &vcpu->arch.ctxt;
579 
580 	sysreg_save_host_state_vhe(host_ctxt);
581 
582 	fpsimd_lazy_switch_to_guest(vcpu);
583 
584 	/*
585 	 * Note that ARM erratum 1165522 requires us to configure both stage 1
586 	 * and stage 2 translation for the guest context before we clear
587 	 * HCR_EL2.TGE. The stage 1 and stage 2 guest context has already been
588 	 * loaded on the CPU in kvm_vcpu_load_vhe().
589 	 */
590 	__activate_traps(vcpu);
591 
592 	__kvm_adjust_pc(vcpu);
593 
594 	sysreg_restore_guest_state_vhe(guest_ctxt);
595 	__debug_switch_to_guest(vcpu);
596 
597 	do {
598 		/* Jump in the fire! */
599 		exit_code = __guest_enter(vcpu);
600 
601 		/* And we're baaack! */
602 	} while (fixup_guest_exit(vcpu, &exit_code));
603 
604 	sysreg_save_guest_state_vhe(guest_ctxt);
605 
606 	__deactivate_traps(vcpu);
607 
608 	fpsimd_lazy_switch_to_host(vcpu);
609 
610 	sysreg_restore_host_state_vhe(host_ctxt);
611 
612 	if (guest_owns_fp_regs())
613 		__fpsimd_save_fpexc32(vcpu);
614 
615 	__debug_switch_to_host(vcpu);
616 
617 	return exit_code;
618 }
619 NOKPROBE_SYMBOL(__kvm_vcpu_run_vhe);
620 
__kvm_vcpu_run(struct kvm_vcpu * vcpu)621 int __kvm_vcpu_run(struct kvm_vcpu *vcpu)
622 {
623 	int ret;
624 
625 	local_daif_mask();
626 
627 	/*
628 	 * Having IRQs masked via PMR when entering the guest means the GIC
629 	 * will not signal the CPU of interrupts of lower priority, and the
630 	 * only way to get out will be via guest exceptions.
631 	 * Naturally, we want to avoid this.
632 	 *
633 	 * local_daif_mask() already sets GIC_PRIO_PSR_I_SET, we just need a
634 	 * dsb to ensure the redistributor is forwards EL2 IRQs to the CPU.
635 	 */
636 	pmr_sync();
637 
638 	ret = __kvm_vcpu_run_vhe(vcpu);
639 
640 	/*
641 	 * local_daif_restore() takes care to properly restore PSTATE.DAIF
642 	 * and the GIC PMR if the host is using IRQ priorities.
643 	 */
644 	local_daif_restore(DAIF_PROCCTX_NOIRQ);
645 
646 	/*
647 	 * When we exit from the guest we change a number of CPU configuration
648 	 * parameters, such as traps.  We rely on the isb() in kvm_call_hyp*()
649 	 * to make sure these changes take effect before running the host or
650 	 * additional guests.
651 	 */
652 	return ret;
653 }
654 
__hyp_call_panic(u64 spsr,u64 elr,u64 par)655 static void __noreturn __hyp_call_panic(u64 spsr, u64 elr, u64 par)
656 {
657 	struct kvm_cpu_context *host_ctxt;
658 	struct kvm_vcpu *vcpu;
659 
660 	host_ctxt = host_data_ptr(host_ctxt);
661 	vcpu = host_ctxt->__hyp_running_vcpu;
662 
663 	__deactivate_traps(vcpu);
664 	sysreg_restore_host_state_vhe(host_ctxt);
665 
666 	panic("HYP panic:\nPS:%08llx PC:%016llx ESR:%08llx\nFAR:%016llx HPFAR:%016llx PAR:%016llx\nVCPU:%p\n",
667 	      spsr, elr,
668 	      read_sysreg_el2(SYS_ESR), read_sysreg_el2(SYS_FAR),
669 	      read_sysreg(hpfar_el2), par, vcpu);
670 }
671 NOKPROBE_SYMBOL(__hyp_call_panic);
672 
hyp_panic(void)673 void __noreturn hyp_panic(void)
674 {
675 	u64 spsr = read_sysreg_el2(SYS_SPSR);
676 	u64 elr = read_sysreg_el2(SYS_ELR);
677 	u64 par = read_sysreg_par();
678 
679 	__hyp_call_panic(spsr, elr, par);
680 }
681 
kvm_unexpected_el2_exception(void)682 asmlinkage void kvm_unexpected_el2_exception(void)
683 {
684 	__kvm_unexpected_el2_exception();
685 }
686