xref: /openbmc/linux/arch/arm64/kvm/hyp/include/hyp/switch.h (revision 9134211f)
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 #ifndef __ARM64_KVM_HYP_SWITCH_H__
8 #define __ARM64_KVM_HYP_SWITCH_H__
9 
10 #include <hyp/adjust_pc.h>
11 #include <hyp/fault.h>
12 
13 #include <linux/arm-smccc.h>
14 #include <linux/kvm_host.h>
15 #include <linux/types.h>
16 #include <linux/jump_label.h>
17 #include <uapi/linux/psci.h>
18 
19 #include <kvm/arm_psci.h>
20 
21 #include <asm/barrier.h>
22 #include <asm/cpufeature.h>
23 #include <asm/extable.h>
24 #include <asm/kprobes.h>
25 #include <asm/kvm_asm.h>
26 #include <asm/kvm_emulate.h>
27 #include <asm/kvm_hyp.h>
28 #include <asm/kvm_mmu.h>
29 #include <asm/kvm_nested.h>
30 #include <asm/fpsimd.h>
31 #include <asm/debug-monitors.h>
32 #include <asm/processor.h>
33 
34 struct kvm_exception_table_entry {
35 	int insn, fixup;
36 };
37 
38 extern struct kvm_exception_table_entry __start___kvm_ex_table;
39 extern struct kvm_exception_table_entry __stop___kvm_ex_table;
40 
41 /* Check whether the FP regs are owned by the guest */
42 static inline bool guest_owns_fp_regs(struct kvm_vcpu *vcpu)
43 {
44 	return vcpu->arch.fp_state == FP_STATE_GUEST_OWNED;
45 }
46 
47 /* Save the 32-bit only FPSIMD system register state */
48 static inline void __fpsimd_save_fpexc32(struct kvm_vcpu *vcpu)
49 {
50 	if (!vcpu_el1_is_32bit(vcpu))
51 		return;
52 
53 	__vcpu_sys_reg(vcpu, FPEXC32_EL2) = read_sysreg(fpexc32_el2);
54 }
55 
56 static inline void __activate_traps_fpsimd32(struct kvm_vcpu *vcpu)
57 {
58 	/*
59 	 * We are about to set CPTR_EL2.TFP to trap all floating point
60 	 * register accesses to EL2, however, the ARM ARM clearly states that
61 	 * traps are only taken to EL2 if the operation would not otherwise
62 	 * trap to EL1.  Therefore, always make sure that for 32-bit guests,
63 	 * we set FPEXC.EN to prevent traps to EL1, when setting the TFP bit.
64 	 * If FP/ASIMD is not implemented, FPEXC is UNDEFINED and any access to
65 	 * it will cause an exception.
66 	 */
67 	if (vcpu_el1_is_32bit(vcpu) && system_supports_fpsimd()) {
68 		write_sysreg(1 << 30, fpexc32_el2);
69 		isb();
70 	}
71 }
72 
73 static inline bool __hfgxtr_traps_required(void)
74 {
75 	if (cpus_have_final_cap(ARM64_SME))
76 		return true;
77 
78 	if (cpus_have_final_cap(ARM64_WORKAROUND_AMPERE_AC03_CPU_38))
79 		return true;
80 
81 	return false;
82 }
83 
84 static inline void __activate_traps_hfgxtr(void)
85 {
86 	u64 r_clr = 0, w_clr = 0, r_set = 0, w_set = 0, tmp;
87 
88 	if (cpus_have_final_cap(ARM64_SME)) {
89 		tmp = HFGxTR_EL2_nSMPRI_EL1_MASK | HFGxTR_EL2_nTPIDR2_EL0_MASK;
90 
91 		r_clr |= tmp;
92 		w_clr |= tmp;
93 	}
94 
95 	/*
96 	 * Trap guest writes to TCR_EL1 to prevent it from enabling HA or HD.
97 	 */
98 	if (cpus_have_final_cap(ARM64_WORKAROUND_AMPERE_AC03_CPU_38))
99 		w_set |= HFGxTR_EL2_TCR_EL1_MASK;
100 
101 	sysreg_clear_set_s(SYS_HFGRTR_EL2, r_clr, r_set);
102 	sysreg_clear_set_s(SYS_HFGWTR_EL2, w_clr, w_set);
103 }
104 
105 static inline void __deactivate_traps_hfgxtr(void)
106 {
107 	u64 r_clr = 0, w_clr = 0, r_set = 0, w_set = 0, tmp;
108 
109 	if (cpus_have_final_cap(ARM64_SME)) {
110 		tmp = HFGxTR_EL2_nSMPRI_EL1_MASK | HFGxTR_EL2_nTPIDR2_EL0_MASK;
111 
112 		r_set |= tmp;
113 		w_set |= tmp;
114 	}
115 
116 	if (cpus_have_final_cap(ARM64_WORKAROUND_AMPERE_AC03_CPU_38))
117 		w_clr |= HFGxTR_EL2_TCR_EL1_MASK;
118 
119 	sysreg_clear_set_s(SYS_HFGRTR_EL2, r_clr, r_set);
120 	sysreg_clear_set_s(SYS_HFGWTR_EL2, w_clr, w_set);
121 }
122 
123 static inline void __activate_traps_common(struct kvm_vcpu *vcpu)
124 {
125 	/* Trap on AArch32 cp15 c15 (impdef sysregs) accesses (EL1 or EL0) */
126 	write_sysreg(1 << 15, hstr_el2);
127 
128 	/*
129 	 * Make sure we trap PMU access from EL0 to EL2. Also sanitize
130 	 * PMSELR_EL0 to make sure it never contains the cycle
131 	 * counter, which could make a PMXEVCNTR_EL0 access UNDEF at
132 	 * EL1 instead of being trapped to EL2.
133 	 */
134 	if (kvm_arm_support_pmu_v3()) {
135 		struct kvm_cpu_context *hctxt;
136 
137 		write_sysreg(0, pmselr_el0);
138 
139 		hctxt = &this_cpu_ptr(&kvm_host_data)->host_ctxt;
140 		ctxt_sys_reg(hctxt, PMUSERENR_EL0) = read_sysreg(pmuserenr_el0);
141 		write_sysreg(ARMV8_PMU_USERENR_MASK, pmuserenr_el0);
142 		vcpu_set_flag(vcpu, PMUSERENR_ON_CPU);
143 	}
144 
145 	vcpu->arch.mdcr_el2_host = read_sysreg(mdcr_el2);
146 	write_sysreg(vcpu->arch.mdcr_el2, mdcr_el2);
147 
148 	if (__hfgxtr_traps_required())
149 		__activate_traps_hfgxtr();
150 }
151 
152 static inline void __deactivate_traps_common(struct kvm_vcpu *vcpu)
153 {
154 	write_sysreg(vcpu->arch.mdcr_el2_host, mdcr_el2);
155 
156 	write_sysreg(0, hstr_el2);
157 	if (kvm_arm_support_pmu_v3()) {
158 		struct kvm_cpu_context *hctxt;
159 
160 		hctxt = &this_cpu_ptr(&kvm_host_data)->host_ctxt;
161 		write_sysreg(ctxt_sys_reg(hctxt, PMUSERENR_EL0), pmuserenr_el0);
162 		vcpu_clear_flag(vcpu, PMUSERENR_ON_CPU);
163 	}
164 
165 	if (__hfgxtr_traps_required())
166 		__deactivate_traps_hfgxtr();
167 }
168 
169 static inline void ___activate_traps(struct kvm_vcpu *vcpu)
170 {
171 	u64 hcr = vcpu->arch.hcr_el2;
172 
173 	if (cpus_have_final_cap(ARM64_WORKAROUND_CAVIUM_TX2_219_TVM))
174 		hcr |= HCR_TVM;
175 
176 	write_sysreg(hcr, hcr_el2);
177 
178 	if (cpus_have_final_cap(ARM64_HAS_RAS_EXTN) && (hcr & HCR_VSE))
179 		write_sysreg_s(vcpu->arch.vsesr_el2, SYS_VSESR_EL2);
180 
181 	if (cpus_have_final_cap(ARM64_HAS_HCX))
182 		write_sysreg_s(HCRX_GUEST_FLAGS, SYS_HCRX_EL2);
183 }
184 
185 static inline void ___deactivate_traps(struct kvm_vcpu *vcpu)
186 {
187 	/*
188 	 * If we pended a virtual abort, preserve it until it gets
189 	 * cleared. See D1.14.3 (Virtual Interrupts) for details, but
190 	 * the crucial bit is "On taking a vSError interrupt,
191 	 * HCR_EL2.VSE is cleared to 0."
192 	 */
193 	if (vcpu->arch.hcr_el2 & HCR_VSE) {
194 		vcpu->arch.hcr_el2 &= ~HCR_VSE;
195 		vcpu->arch.hcr_el2 |= read_sysreg(hcr_el2) & HCR_VSE;
196 	}
197 
198 	if (cpus_have_final_cap(ARM64_HAS_HCX))
199 		write_sysreg_s(HCRX_HOST_FLAGS, SYS_HCRX_EL2);
200 }
201 
202 static inline bool __populate_fault_info(struct kvm_vcpu *vcpu)
203 {
204 	return __get_fault_info(vcpu->arch.fault.esr_el2, &vcpu->arch.fault);
205 }
206 
207 static inline void __hyp_sve_restore_guest(struct kvm_vcpu *vcpu)
208 {
209 	sve_cond_update_zcr_vq(vcpu_sve_max_vq(vcpu) - 1, SYS_ZCR_EL2);
210 	__sve_restore_state(vcpu_sve_pffr(vcpu),
211 			    &vcpu->arch.ctxt.fp_regs.fpsr);
212 	write_sysreg_el1(__vcpu_sys_reg(vcpu, ZCR_EL1), SYS_ZCR);
213 }
214 
215 /*
216  * We trap the first access to the FP/SIMD to save the host context and
217  * restore the guest context lazily.
218  * If FP/SIMD is not implemented, handle the trap and inject an undefined
219  * instruction exception to the guest. Similarly for trapped SVE accesses.
220  */
221 static bool kvm_hyp_handle_fpsimd(struct kvm_vcpu *vcpu, u64 *exit_code)
222 {
223 	bool sve_guest;
224 	u8 esr_ec;
225 	u64 reg;
226 
227 	if (!system_supports_fpsimd())
228 		return false;
229 
230 	sve_guest = vcpu_has_sve(vcpu);
231 	esr_ec = kvm_vcpu_trap_get_class(vcpu);
232 
233 	/* Only handle traps the vCPU can support here: */
234 	switch (esr_ec) {
235 	case ESR_ELx_EC_FP_ASIMD:
236 		break;
237 	case ESR_ELx_EC_SVE:
238 		if (!sve_guest)
239 			return false;
240 		break;
241 	default:
242 		return false;
243 	}
244 
245 	/* Valid trap.  Switch the context: */
246 
247 	/* First disable enough traps to allow us to update the registers */
248 	if (has_vhe() || has_hvhe()) {
249 		reg = CPACR_EL1_FPEN_EL0EN | CPACR_EL1_FPEN_EL1EN;
250 		if (sve_guest)
251 			reg |= CPACR_EL1_ZEN_EL0EN | CPACR_EL1_ZEN_EL1EN;
252 
253 		sysreg_clear_set(cpacr_el1, 0, reg);
254 	} else {
255 		reg = CPTR_EL2_TFP;
256 		if (sve_guest)
257 			reg |= CPTR_EL2_TZ;
258 
259 		sysreg_clear_set(cptr_el2, reg, 0);
260 	}
261 	isb();
262 
263 	/* Write out the host state if it's in the registers */
264 	if (vcpu->arch.fp_state == FP_STATE_HOST_OWNED)
265 		__fpsimd_save_state(vcpu->arch.host_fpsimd_state);
266 
267 	/* Restore the guest state */
268 	if (sve_guest)
269 		__hyp_sve_restore_guest(vcpu);
270 	else
271 		__fpsimd_restore_state(&vcpu->arch.ctxt.fp_regs);
272 
273 	/* Skip restoring fpexc32 for AArch64 guests */
274 	if (!(read_sysreg(hcr_el2) & HCR_RW))
275 		write_sysreg(__vcpu_sys_reg(vcpu, FPEXC32_EL2), fpexc32_el2);
276 
277 	vcpu->arch.fp_state = FP_STATE_GUEST_OWNED;
278 
279 	return true;
280 }
281 
282 static inline bool handle_tx2_tvm(struct kvm_vcpu *vcpu)
283 {
284 	u32 sysreg = esr_sys64_to_sysreg(kvm_vcpu_get_esr(vcpu));
285 	int rt = kvm_vcpu_sys_get_rt(vcpu);
286 	u64 val = vcpu_get_reg(vcpu, rt);
287 
288 	/*
289 	 * The normal sysreg handling code expects to see the traps,
290 	 * let's not do anything here.
291 	 */
292 	if (vcpu->arch.hcr_el2 & HCR_TVM)
293 		return false;
294 
295 	switch (sysreg) {
296 	case SYS_SCTLR_EL1:
297 		write_sysreg_el1(val, SYS_SCTLR);
298 		break;
299 	case SYS_TTBR0_EL1:
300 		write_sysreg_el1(val, SYS_TTBR0);
301 		break;
302 	case SYS_TTBR1_EL1:
303 		write_sysreg_el1(val, SYS_TTBR1);
304 		break;
305 	case SYS_TCR_EL1:
306 		write_sysreg_el1(val, SYS_TCR);
307 		break;
308 	case SYS_ESR_EL1:
309 		write_sysreg_el1(val, SYS_ESR);
310 		break;
311 	case SYS_FAR_EL1:
312 		write_sysreg_el1(val, SYS_FAR);
313 		break;
314 	case SYS_AFSR0_EL1:
315 		write_sysreg_el1(val, SYS_AFSR0);
316 		break;
317 	case SYS_AFSR1_EL1:
318 		write_sysreg_el1(val, SYS_AFSR1);
319 		break;
320 	case SYS_MAIR_EL1:
321 		write_sysreg_el1(val, SYS_MAIR);
322 		break;
323 	case SYS_AMAIR_EL1:
324 		write_sysreg_el1(val, SYS_AMAIR);
325 		break;
326 	case SYS_CONTEXTIDR_EL1:
327 		write_sysreg_el1(val, SYS_CONTEXTIDR);
328 		break;
329 	default:
330 		return false;
331 	}
332 
333 	__kvm_skip_instr(vcpu);
334 	return true;
335 }
336 
337 static inline bool esr_is_ptrauth_trap(u64 esr)
338 {
339 	switch (esr_sys64_to_sysreg(esr)) {
340 	case SYS_APIAKEYLO_EL1:
341 	case SYS_APIAKEYHI_EL1:
342 	case SYS_APIBKEYLO_EL1:
343 	case SYS_APIBKEYHI_EL1:
344 	case SYS_APDAKEYLO_EL1:
345 	case SYS_APDAKEYHI_EL1:
346 	case SYS_APDBKEYLO_EL1:
347 	case SYS_APDBKEYHI_EL1:
348 	case SYS_APGAKEYLO_EL1:
349 	case SYS_APGAKEYHI_EL1:
350 		return true;
351 	}
352 
353 	return false;
354 }
355 
356 #define __ptrauth_save_key(ctxt, key)					\
357 	do {								\
358 	u64 __val;                                                      \
359 	__val = read_sysreg_s(SYS_ ## key ## KEYLO_EL1);                \
360 	ctxt_sys_reg(ctxt, key ## KEYLO_EL1) = __val;                   \
361 	__val = read_sysreg_s(SYS_ ## key ## KEYHI_EL1);                \
362 	ctxt_sys_reg(ctxt, key ## KEYHI_EL1) = __val;                   \
363 } while(0)
364 
365 DECLARE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt);
366 
367 static bool kvm_hyp_handle_ptrauth(struct kvm_vcpu *vcpu, u64 *exit_code)
368 {
369 	struct kvm_cpu_context *ctxt;
370 	u64 val;
371 
372 	if (!vcpu_has_ptrauth(vcpu))
373 		return false;
374 
375 	ctxt = this_cpu_ptr(&kvm_hyp_ctxt);
376 	__ptrauth_save_key(ctxt, APIA);
377 	__ptrauth_save_key(ctxt, APIB);
378 	__ptrauth_save_key(ctxt, APDA);
379 	__ptrauth_save_key(ctxt, APDB);
380 	__ptrauth_save_key(ctxt, APGA);
381 
382 	vcpu_ptrauth_enable(vcpu);
383 
384 	val = read_sysreg(hcr_el2);
385 	val |= (HCR_API | HCR_APK);
386 	write_sysreg(val, hcr_el2);
387 
388 	return true;
389 }
390 
391 static bool kvm_hyp_handle_cntpct(struct kvm_vcpu *vcpu)
392 {
393 	struct arch_timer_context *ctxt;
394 	u32 sysreg;
395 	u64 val;
396 
397 	/*
398 	 * We only get here for 64bit guests, 32bit guests will hit
399 	 * the long and winding road all the way to the standard
400 	 * handling. Yes, it sucks to be irrelevant.
401 	 */
402 	sysreg = esr_sys64_to_sysreg(kvm_vcpu_get_esr(vcpu));
403 
404 	switch (sysreg) {
405 	case SYS_CNTPCT_EL0:
406 	case SYS_CNTPCTSS_EL0:
407 		if (vcpu_has_nv(vcpu)) {
408 			if (is_hyp_ctxt(vcpu)) {
409 				ctxt = vcpu_hptimer(vcpu);
410 				break;
411 			}
412 
413 			/* Check for guest hypervisor trapping */
414 			val = __vcpu_sys_reg(vcpu, CNTHCTL_EL2);
415 			if (!vcpu_el2_e2h_is_set(vcpu))
416 				val = (val & CNTHCTL_EL1PCTEN) << 10;
417 
418 			if (!(val & (CNTHCTL_EL1PCTEN << 10)))
419 				return false;
420 		}
421 
422 		ctxt = vcpu_ptimer(vcpu);
423 		break;
424 	default:
425 		return false;
426 	}
427 
428 	val = arch_timer_read_cntpct_el0();
429 
430 	if (ctxt->offset.vm_offset)
431 		val -= *kern_hyp_va(ctxt->offset.vm_offset);
432 	if (ctxt->offset.vcpu_offset)
433 		val -= *kern_hyp_va(ctxt->offset.vcpu_offset);
434 
435 	vcpu_set_reg(vcpu, kvm_vcpu_sys_get_rt(vcpu), val);
436 	__kvm_skip_instr(vcpu);
437 	return true;
438 }
439 
440 static bool handle_ampere1_tcr(struct kvm_vcpu *vcpu)
441 {
442 	u32 sysreg = esr_sys64_to_sysreg(kvm_vcpu_get_esr(vcpu));
443 	int rt = kvm_vcpu_sys_get_rt(vcpu);
444 	u64 val = vcpu_get_reg(vcpu, rt);
445 
446 	if (sysreg != SYS_TCR_EL1)
447 		return false;
448 
449 	/*
450 	 * Affected parts do not advertise support for hardware Access Flag /
451 	 * Dirty state management in ID_AA64MMFR1_EL1.HAFDBS, but the underlying
452 	 * control bits are still functional. The architecture requires these be
453 	 * RES0 on systems that do not implement FEAT_HAFDBS.
454 	 *
455 	 * Uphold the requirements of the architecture by masking guest writes
456 	 * to TCR_EL1.{HA,HD} here.
457 	 */
458 	val &= ~(TCR_HD | TCR_HA);
459 	write_sysreg_el1(val, SYS_TCR);
460 	return true;
461 }
462 
463 static bool kvm_hyp_handle_sysreg(struct kvm_vcpu *vcpu, u64 *exit_code)
464 {
465 	if (cpus_have_final_cap(ARM64_WORKAROUND_CAVIUM_TX2_219_TVM) &&
466 	    handle_tx2_tvm(vcpu))
467 		return true;
468 
469 	if (cpus_have_final_cap(ARM64_WORKAROUND_AMPERE_AC03_CPU_38) &&
470 	    handle_ampere1_tcr(vcpu))
471 		return true;
472 
473 	if (static_branch_unlikely(&vgic_v3_cpuif_trap) &&
474 	    __vgic_v3_perform_cpuif_access(vcpu) == 1)
475 		return true;
476 
477 	if (esr_is_ptrauth_trap(kvm_vcpu_get_esr(vcpu)))
478 		return kvm_hyp_handle_ptrauth(vcpu, exit_code);
479 
480 	if (kvm_hyp_handle_cntpct(vcpu))
481 		return true;
482 
483 	return false;
484 }
485 
486 static bool kvm_hyp_handle_cp15_32(struct kvm_vcpu *vcpu, u64 *exit_code)
487 {
488 	if (static_branch_unlikely(&vgic_v3_cpuif_trap) &&
489 	    __vgic_v3_perform_cpuif_access(vcpu) == 1)
490 		return true;
491 
492 	return false;
493 }
494 
495 static bool kvm_hyp_handle_memory_fault(struct kvm_vcpu *vcpu, u64 *exit_code)
496 {
497 	if (!__populate_fault_info(vcpu))
498 		return true;
499 
500 	return false;
501 }
502 static bool kvm_hyp_handle_iabt_low(struct kvm_vcpu *vcpu, u64 *exit_code)
503 	__alias(kvm_hyp_handle_memory_fault);
504 static bool kvm_hyp_handle_watchpt_low(struct kvm_vcpu *vcpu, u64 *exit_code)
505 	__alias(kvm_hyp_handle_memory_fault);
506 
507 static bool kvm_hyp_handle_dabt_low(struct kvm_vcpu *vcpu, u64 *exit_code)
508 {
509 	if (kvm_hyp_handle_memory_fault(vcpu, exit_code))
510 		return true;
511 
512 	if (static_branch_unlikely(&vgic_v2_cpuif_trap)) {
513 		bool valid;
514 
515 		valid = kvm_vcpu_trap_get_fault_type(vcpu) == ESR_ELx_FSC_FAULT &&
516 			kvm_vcpu_dabt_isvalid(vcpu) &&
517 			!kvm_vcpu_abt_issea(vcpu) &&
518 			!kvm_vcpu_abt_iss1tw(vcpu);
519 
520 		if (valid) {
521 			int ret = __vgic_v2_perform_cpuif_access(vcpu);
522 
523 			if (ret == 1)
524 				return true;
525 
526 			/* Promote an illegal access to an SError.*/
527 			if (ret == -1)
528 				*exit_code = ARM_EXCEPTION_EL1_SERROR;
529 		}
530 	}
531 
532 	return false;
533 }
534 
535 typedef bool (*exit_handler_fn)(struct kvm_vcpu *, u64 *);
536 
537 static const exit_handler_fn *kvm_get_exit_handler_array(struct kvm_vcpu *vcpu);
538 
539 static void early_exit_filter(struct kvm_vcpu *vcpu, u64 *exit_code);
540 
541 /*
542  * Allow the hypervisor to handle the exit with an exit handler if it has one.
543  *
544  * Returns true if the hypervisor handled the exit, and control should go back
545  * to the guest, or false if it hasn't.
546  */
547 static inline bool kvm_hyp_handle_exit(struct kvm_vcpu *vcpu, u64 *exit_code)
548 {
549 	const exit_handler_fn *handlers = kvm_get_exit_handler_array(vcpu);
550 	exit_handler_fn fn;
551 
552 	fn = handlers[kvm_vcpu_trap_get_class(vcpu)];
553 
554 	if (fn)
555 		return fn(vcpu, exit_code);
556 
557 	return false;
558 }
559 
560 static inline void synchronize_vcpu_pstate(struct kvm_vcpu *vcpu, u64 *exit_code)
561 {
562 	/*
563 	 * Check for the conditions of Cortex-A510's #2077057. When these occur
564 	 * SPSR_EL2 can't be trusted, but isn't needed either as it is
565 	 * unchanged from the value in vcpu_gp_regs(vcpu)->pstate.
566 	 * Are we single-stepping the guest, and took a PAC exception from the
567 	 * active-not-pending state?
568 	 */
569 	if (cpus_have_final_cap(ARM64_WORKAROUND_2077057)		&&
570 	    vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP			&&
571 	    *vcpu_cpsr(vcpu) & DBG_SPSR_SS				&&
572 	    ESR_ELx_EC(read_sysreg_el2(SYS_ESR)) == ESR_ELx_EC_PAC)
573 		write_sysreg_el2(*vcpu_cpsr(vcpu), SYS_SPSR);
574 
575 	vcpu->arch.ctxt.regs.pstate = read_sysreg_el2(SYS_SPSR);
576 }
577 
578 /*
579  * Return true when we were able to fixup the guest exit and should return to
580  * the guest, false when we should restore the host state and return to the
581  * main run loop.
582  */
583 static inline bool fixup_guest_exit(struct kvm_vcpu *vcpu, u64 *exit_code)
584 {
585 	/*
586 	 * Save PSTATE early so that we can evaluate the vcpu mode
587 	 * early on.
588 	 */
589 	synchronize_vcpu_pstate(vcpu, exit_code);
590 
591 	/*
592 	 * Check whether we want to repaint the state one way or
593 	 * another.
594 	 */
595 	early_exit_filter(vcpu, exit_code);
596 
597 	if (ARM_EXCEPTION_CODE(*exit_code) != ARM_EXCEPTION_IRQ)
598 		vcpu->arch.fault.esr_el2 = read_sysreg_el2(SYS_ESR);
599 
600 	if (ARM_SERROR_PENDING(*exit_code) &&
601 	    ARM_EXCEPTION_CODE(*exit_code) != ARM_EXCEPTION_IRQ) {
602 		u8 esr_ec = kvm_vcpu_trap_get_class(vcpu);
603 
604 		/*
605 		 * HVC already have an adjusted PC, which we need to
606 		 * correct in order to return to after having injected
607 		 * the SError.
608 		 *
609 		 * SMC, on the other hand, is *trapped*, meaning its
610 		 * preferred return address is the SMC itself.
611 		 */
612 		if (esr_ec == ESR_ELx_EC_HVC32 || esr_ec == ESR_ELx_EC_HVC64)
613 			write_sysreg_el2(read_sysreg_el2(SYS_ELR) - 4, SYS_ELR);
614 	}
615 
616 	/*
617 	 * We're using the raw exception code in order to only process
618 	 * the trap if no SError is pending. We will come back to the
619 	 * same PC once the SError has been injected, and replay the
620 	 * trapping instruction.
621 	 */
622 	if (*exit_code != ARM_EXCEPTION_TRAP)
623 		goto exit;
624 
625 	/* Check if there's an exit handler and allow it to handle the exit. */
626 	if (kvm_hyp_handle_exit(vcpu, exit_code))
627 		goto guest;
628 exit:
629 	/* Return to the host kernel and handle the exit */
630 	return false;
631 
632 guest:
633 	/* Re-enter the guest */
634 	asm(ALTERNATIVE("nop", "dmb sy", ARM64_WORKAROUND_1508412));
635 	return true;
636 }
637 
638 static inline void __kvm_unexpected_el2_exception(void)
639 {
640 	extern char __guest_exit_panic[];
641 	unsigned long addr, fixup;
642 	struct kvm_exception_table_entry *entry, *end;
643 	unsigned long elr_el2 = read_sysreg(elr_el2);
644 
645 	entry = &__start___kvm_ex_table;
646 	end = &__stop___kvm_ex_table;
647 
648 	while (entry < end) {
649 		addr = (unsigned long)&entry->insn + entry->insn;
650 		fixup = (unsigned long)&entry->fixup + entry->fixup;
651 
652 		if (addr != elr_el2) {
653 			entry++;
654 			continue;
655 		}
656 
657 		write_sysreg(fixup, elr_el2);
658 		return;
659 	}
660 
661 	/* Trigger a panic after restoring the hyp context. */
662 	write_sysreg(__guest_exit_panic, elr_el2);
663 }
664 
665 #endif /* __ARM64_KVM_HYP_SWITCH_H__ */
666