xref: /openbmc/linux/arch/arm64/kvm/hyp/include/hyp/switch.h (revision 76a4f7cc) 
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 
12 #include <linux/arm-smccc.h>
13 #include <linux/kvm_host.h>
14 #include <linux/types.h>
15 #include <linux/jump_label.h>
16 #include <uapi/linux/psci.h>
17 
18 #include <kvm/arm_psci.h>
19 
20 #include <asm/barrier.h>
21 #include <asm/cpufeature.h>
22 #include <asm/extable.h>
23 #include <asm/kprobes.h>
24 #include <asm/kvm_asm.h>
25 #include <asm/kvm_emulate.h>
26 #include <asm/kvm_hyp.h>
27 #include <asm/kvm_mmu.h>
28 #include <asm/fpsimd.h>
29 #include <asm/debug-monitors.h>
30 #include <asm/processor.h>
31 #include <asm/thread_info.h>
32 
33 extern struct exception_table_entry __start___kvm_ex_table;
34 extern struct exception_table_entry __stop___kvm_ex_table;
35 
36 /* Check whether the FP regs were dirtied while in the host-side run loop: */
37 static inline bool update_fp_enabled(struct kvm_vcpu *vcpu)
38 {
39 	/*
40 	 * When the system doesn't support FP/SIMD, we cannot rely on
41 	 * the _TIF_FOREIGN_FPSTATE flag. However, we always inject an
42 	 * abort on the very first access to FP and thus we should never
43 	 * see KVM_ARM64_FP_ENABLED. For added safety, make sure we always
44 	 * trap the accesses.
45 	 */
46 	if (!system_supports_fpsimd() ||
47 	    vcpu->arch.host_thread_info->flags & _TIF_FOREIGN_FPSTATE)
48 		vcpu->arch.flags &= ~(KVM_ARM64_FP_ENABLED |
49 				      KVM_ARM64_FP_HOST);
50 
51 	return !!(vcpu->arch.flags & KVM_ARM64_FP_ENABLED);
52 }
53 
54 /* Save the 32-bit only FPSIMD system register state */
55 static inline void __fpsimd_save_fpexc32(struct kvm_vcpu *vcpu)
56 {
57 	if (!vcpu_el1_is_32bit(vcpu))
58 		return;
59 
60 	__vcpu_sys_reg(vcpu, FPEXC32_EL2) = read_sysreg(fpexc32_el2);
61 }
62 
63 static inline void __activate_traps_fpsimd32(struct kvm_vcpu *vcpu)
64 {
65 	/*
66 	 * We are about to set CPTR_EL2.TFP to trap all floating point
67 	 * register accesses to EL2, however, the ARM ARM clearly states that
68 	 * traps are only taken to EL2 if the operation would not otherwise
69 	 * trap to EL1.  Therefore, always make sure that for 32-bit guests,
70 	 * we set FPEXC.EN to prevent traps to EL1, when setting the TFP bit.
71 	 * If FP/ASIMD is not implemented, FPEXC is UNDEFINED and any access to
72 	 * it will cause an exception.
73 	 */
74 	if (vcpu_el1_is_32bit(vcpu) && system_supports_fpsimd()) {
75 		write_sysreg(1 << 30, fpexc32_el2);
76 		isb();
77 	}
78 }
79 
80 static inline void __activate_traps_common(struct kvm_vcpu *vcpu)
81 {
82 	/* Trap on AArch32 cp15 c15 (impdef sysregs) accesses (EL1 or EL0) */
83 	write_sysreg(1 << 15, hstr_el2);
84 
85 	/*
86 	 * Make sure we trap PMU access from EL0 to EL2. Also sanitize
87 	 * PMSELR_EL0 to make sure it never contains the cycle
88 	 * counter, which could make a PMXEVCNTR_EL0 access UNDEF at
89 	 * EL1 instead of being trapped to EL2.
90 	 */
91 	if (kvm_arm_support_pmu_v3()) {
92 		write_sysreg(0, pmselr_el0);
93 		write_sysreg(ARMV8_PMU_USERENR_MASK, pmuserenr_el0);
94 	}
95 
96 	vcpu->arch.mdcr_el2_host = read_sysreg(mdcr_el2);
97 	write_sysreg(vcpu->arch.mdcr_el2, mdcr_el2);
98 }
99 
100 static inline void __deactivate_traps_common(struct kvm_vcpu *vcpu)
101 {
102 	write_sysreg(vcpu->arch.mdcr_el2_host, mdcr_el2);
103 
104 	write_sysreg(0, hstr_el2);
105 	if (kvm_arm_support_pmu_v3())
106 		write_sysreg(0, pmuserenr_el0);
107 }
108 
109 static inline void ___activate_traps(struct kvm_vcpu *vcpu)
110 {
111 	u64 hcr = vcpu->arch.hcr_el2;
112 
113 	if (cpus_have_final_cap(ARM64_WORKAROUND_CAVIUM_TX2_219_TVM))
114 		hcr |= HCR_TVM;
115 
116 	write_sysreg(hcr, hcr_el2);
117 
118 	if (cpus_have_final_cap(ARM64_HAS_RAS_EXTN) && (hcr & HCR_VSE))
119 		write_sysreg_s(vcpu->arch.vsesr_el2, SYS_VSESR_EL2);
120 }
121 
122 static inline void ___deactivate_traps(struct kvm_vcpu *vcpu)
123 {
124 	/*
125 	 * If we pended a virtual abort, preserve it until it gets
126 	 * cleared. See D1.14.3 (Virtual Interrupts) for details, but
127 	 * the crucial bit is "On taking a vSError interrupt,
128 	 * HCR_EL2.VSE is cleared to 0."
129 	 */
130 	if (vcpu->arch.hcr_el2 & HCR_VSE) {
131 		vcpu->arch.hcr_el2 &= ~HCR_VSE;
132 		vcpu->arch.hcr_el2 |= read_sysreg(hcr_el2) & HCR_VSE;
133 	}
134 }
135 
136 static inline bool __translate_far_to_hpfar(u64 far, u64 *hpfar)
137 {
138 	u64 par, tmp;
139 
140 	/*
141 	 * Resolve the IPA the hard way using the guest VA.
142 	 *
143 	 * Stage-1 translation already validated the memory access
144 	 * rights. As such, we can use the EL1 translation regime, and
145 	 * don't have to distinguish between EL0 and EL1 access.
146 	 *
147 	 * We do need to save/restore PAR_EL1 though, as we haven't
148 	 * saved the guest context yet, and we may return early...
149 	 */
150 	par = read_sysreg_par();
151 	if (!__kvm_at("s1e1r", far))
152 		tmp = read_sysreg_par();
153 	else
154 		tmp = SYS_PAR_EL1_F; /* back to the guest */
155 	write_sysreg(par, par_el1);
156 
157 	if (unlikely(tmp & SYS_PAR_EL1_F))
158 		return false; /* Translation failed, back to guest */
159 
160 	/* Convert PAR to HPFAR format */
161 	*hpfar = PAR_TO_HPFAR(tmp);
162 	return true;
163 }
164 
165 static inline bool __get_fault_info(u64 esr, struct kvm_vcpu_fault_info *fault)
166 {
167 	u64 hpfar, far;
168 
169 	far = read_sysreg_el2(SYS_FAR);
170 
171 	/*
172 	 * The HPFAR can be invalid if the stage 2 fault did not
173 	 * happen during a stage 1 page table walk (the ESR_EL2.S1PTW
174 	 * bit is clear) and one of the two following cases are true:
175 	 *   1. The fault was due to a permission fault
176 	 *   2. The processor carries errata 834220
177 	 *
178 	 * Therefore, for all non S1PTW faults where we either have a
179 	 * permission fault or the errata workaround is enabled, we
180 	 * resolve the IPA using the AT instruction.
181 	 */
182 	if (!(esr & ESR_ELx_S1PTW) &&
183 	    (cpus_have_final_cap(ARM64_WORKAROUND_834220) ||
184 	     (esr & ESR_ELx_FSC_TYPE) == FSC_PERM)) {
185 		if (!__translate_far_to_hpfar(far, &hpfar))
186 			return false;
187 	} else {
188 		hpfar = read_sysreg(hpfar_el2);
189 	}
190 
191 	fault->far_el2 = far;
192 	fault->hpfar_el2 = hpfar;
193 	return true;
194 }
195 
196 static inline bool __populate_fault_info(struct kvm_vcpu *vcpu)
197 {
198 	u8 ec;
199 	u64 esr;
200 
201 	esr = vcpu->arch.fault.esr_el2;
202 	ec = ESR_ELx_EC(esr);
203 
204 	if (ec != ESR_ELx_EC_DABT_LOW && ec != ESR_ELx_EC_IABT_LOW)
205 		return true;
206 
207 	return __get_fault_info(esr, &vcpu->arch.fault);
208 }
209 
210 static inline void __hyp_sve_save_host(struct kvm_vcpu *vcpu)
211 {
212 	struct thread_struct *thread;
213 
214 	thread = container_of(vcpu->arch.host_fpsimd_state, struct thread_struct,
215 			      uw.fpsimd_state);
216 
217 	__sve_save_state(sve_pffr(thread), &vcpu->arch.host_fpsimd_state->fpsr);
218 }
219 
220 static inline void __hyp_sve_restore_guest(struct kvm_vcpu *vcpu)
221 {
222 	sve_cond_update_zcr_vq(vcpu_sve_max_vq(vcpu) - 1, SYS_ZCR_EL2);
223 	__sve_restore_state(vcpu_sve_pffr(vcpu),
224 			    &vcpu->arch.ctxt.fp_regs.fpsr);
225 	write_sysreg_el1(__vcpu_sys_reg(vcpu, ZCR_EL1), SYS_ZCR);
226 }
227 
228 /* Check for an FPSIMD/SVE trap and handle as appropriate */
229 static inline bool __hyp_handle_fpsimd(struct kvm_vcpu *vcpu)
230 {
231 	bool sve_guest, sve_host;
232 	u8 esr_ec;
233 	u64 reg;
234 
235 	if (!system_supports_fpsimd())
236 		return false;
237 
238 	if (system_supports_sve()) {
239 		sve_guest = vcpu_has_sve(vcpu);
240 		sve_host = vcpu->arch.flags & KVM_ARM64_HOST_SVE_IN_USE;
241 	} else {
242 		sve_guest = false;
243 		sve_host = false;
244 	}
245 
246 	esr_ec = kvm_vcpu_trap_get_class(vcpu);
247 	if (esr_ec != ESR_ELx_EC_FP_ASIMD &&
248 	    esr_ec != ESR_ELx_EC_SVE)
249 		return false;
250 
251 	/* Don't handle SVE traps for non-SVE vcpus here: */
252 	if (!sve_guest && esr_ec != ESR_ELx_EC_FP_ASIMD)
253 		return false;
254 
255 	/* Valid trap.  Switch the context: */
256 	if (has_vhe()) {
257 		reg = CPACR_EL1_FPEN;
258 		if (sve_guest)
259 			reg |= CPACR_EL1_ZEN;
260 
261 		sysreg_clear_set(cpacr_el1, 0, reg);
262 	} else {
263 		reg = CPTR_EL2_TFP;
264 		if (sve_guest)
265 			reg |= CPTR_EL2_TZ;
266 
267 		sysreg_clear_set(cptr_el2, reg, 0);
268 	}
269 	isb();
270 
271 	if (vcpu->arch.flags & KVM_ARM64_FP_HOST) {
272 		if (sve_host)
273 			__hyp_sve_save_host(vcpu);
274 		else
275 			__fpsimd_save_state(vcpu->arch.host_fpsimd_state);
276 
277 		vcpu->arch.flags &= ~KVM_ARM64_FP_HOST;
278 	}
279 
280 	if (sve_guest)
281 		__hyp_sve_restore_guest(vcpu);
282 	else
283 		__fpsimd_restore_state(&vcpu->arch.ctxt.fp_regs);
284 
285 	/* Skip restoring fpexc32 for AArch64 guests */
286 	if (!(read_sysreg(hcr_el2) & HCR_RW))
287 		write_sysreg(__vcpu_sys_reg(vcpu, FPEXC32_EL2), fpexc32_el2);
288 
289 	vcpu->arch.flags |= KVM_ARM64_FP_ENABLED;
290 
291 	return true;
292 }
293 
294 static inline bool handle_tx2_tvm(struct kvm_vcpu *vcpu)
295 {
296 	u32 sysreg = esr_sys64_to_sysreg(kvm_vcpu_get_esr(vcpu));
297 	int rt = kvm_vcpu_sys_get_rt(vcpu);
298 	u64 val = vcpu_get_reg(vcpu, rt);
299 
300 	/*
301 	 * The normal sysreg handling code expects to see the traps,
302 	 * let's not do anything here.
303 	 */
304 	if (vcpu->arch.hcr_el2 & HCR_TVM)
305 		return false;
306 
307 	switch (sysreg) {
308 	case SYS_SCTLR_EL1:
309 		write_sysreg_el1(val, SYS_SCTLR);
310 		break;
311 	case SYS_TTBR0_EL1:
312 		write_sysreg_el1(val, SYS_TTBR0);
313 		break;
314 	case SYS_TTBR1_EL1:
315 		write_sysreg_el1(val, SYS_TTBR1);
316 		break;
317 	case SYS_TCR_EL1:
318 		write_sysreg_el1(val, SYS_TCR);
319 		break;
320 	case SYS_ESR_EL1:
321 		write_sysreg_el1(val, SYS_ESR);
322 		break;
323 	case SYS_FAR_EL1:
324 		write_sysreg_el1(val, SYS_FAR);
325 		break;
326 	case SYS_AFSR0_EL1:
327 		write_sysreg_el1(val, SYS_AFSR0);
328 		break;
329 	case SYS_AFSR1_EL1:
330 		write_sysreg_el1(val, SYS_AFSR1);
331 		break;
332 	case SYS_MAIR_EL1:
333 		write_sysreg_el1(val, SYS_MAIR);
334 		break;
335 	case SYS_AMAIR_EL1:
336 		write_sysreg_el1(val, SYS_AMAIR);
337 		break;
338 	case SYS_CONTEXTIDR_EL1:
339 		write_sysreg_el1(val, SYS_CONTEXTIDR);
340 		break;
341 	default:
342 		return false;
343 	}
344 
345 	__kvm_skip_instr(vcpu);
346 	return true;
347 }
348 
349 static inline bool esr_is_ptrauth_trap(u32 esr)
350 {
351 	u32 ec = ESR_ELx_EC(esr);
352 
353 	if (ec == ESR_ELx_EC_PAC)
354 		return true;
355 
356 	if (ec != ESR_ELx_EC_SYS64)
357 		return false;
358 
359 	switch (esr_sys64_to_sysreg(esr)) {
360 	case SYS_APIAKEYLO_EL1:
361 	case SYS_APIAKEYHI_EL1:
362 	case SYS_APIBKEYLO_EL1:
363 	case SYS_APIBKEYHI_EL1:
364 	case SYS_APDAKEYLO_EL1:
365 	case SYS_APDAKEYHI_EL1:
366 	case SYS_APDBKEYLO_EL1:
367 	case SYS_APDBKEYHI_EL1:
368 	case SYS_APGAKEYLO_EL1:
369 	case SYS_APGAKEYHI_EL1:
370 		return true;
371 	}
372 
373 	return false;
374 }
375 
376 #define __ptrauth_save_key(ctxt, key)					\
377 	do {								\
378 	u64 __val;                                                      \
379 	__val = read_sysreg_s(SYS_ ## key ## KEYLO_EL1);                \
380 	ctxt_sys_reg(ctxt, key ## KEYLO_EL1) = __val;                   \
381 	__val = read_sysreg_s(SYS_ ## key ## KEYHI_EL1);                \
382 	ctxt_sys_reg(ctxt, key ## KEYHI_EL1) = __val;                   \
383 } while(0)
384 
385 DECLARE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt);
386 
387 static inline bool __hyp_handle_ptrauth(struct kvm_vcpu *vcpu)
388 {
389 	struct kvm_cpu_context *ctxt;
390 	u64 val;
391 
392 	if (!vcpu_has_ptrauth(vcpu) ||
393 	    !esr_is_ptrauth_trap(kvm_vcpu_get_esr(vcpu)))
394 		return false;
395 
396 	ctxt = this_cpu_ptr(&kvm_hyp_ctxt);
397 	__ptrauth_save_key(ctxt, APIA);
398 	__ptrauth_save_key(ctxt, APIB);
399 	__ptrauth_save_key(ctxt, APDA);
400 	__ptrauth_save_key(ctxt, APDB);
401 	__ptrauth_save_key(ctxt, APGA);
402 
403 	vcpu_ptrauth_enable(vcpu);
404 
405 	val = read_sysreg(hcr_el2);
406 	val |= (HCR_API | HCR_APK);
407 	write_sysreg(val, hcr_el2);
408 
409 	return true;
410 }
411 
412 /*
413  * Return true when we were able to fixup the guest exit and should return to
414  * the guest, false when we should restore the host state and return to the
415  * main run loop.
416  */
417 static inline bool fixup_guest_exit(struct kvm_vcpu *vcpu, u64 *exit_code)
418 {
419 	if (ARM_EXCEPTION_CODE(*exit_code) != ARM_EXCEPTION_IRQ)
420 		vcpu->arch.fault.esr_el2 = read_sysreg_el2(SYS_ESR);
421 
422 	if (ARM_SERROR_PENDING(*exit_code)) {
423 		u8 esr_ec = kvm_vcpu_trap_get_class(vcpu);
424 
425 		/*
426 		 * HVC already have an adjusted PC, which we need to
427 		 * correct in order to return to after having injected
428 		 * the SError.
429 		 *
430 		 * SMC, on the other hand, is *trapped*, meaning its
431 		 * preferred return address is the SMC itself.
432 		 */
433 		if (esr_ec == ESR_ELx_EC_HVC32 || esr_ec == ESR_ELx_EC_HVC64)
434 			write_sysreg_el2(read_sysreg_el2(SYS_ELR) - 4, SYS_ELR);
435 	}
436 
437 	/*
438 	 * We're using the raw exception code in order to only process
439 	 * the trap if no SError is pending. We will come back to the
440 	 * same PC once the SError has been injected, and replay the
441 	 * trapping instruction.
442 	 */
443 	if (*exit_code != ARM_EXCEPTION_TRAP)
444 		goto exit;
445 
446 	if (cpus_have_final_cap(ARM64_WORKAROUND_CAVIUM_TX2_219_TVM) &&
447 	    kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_SYS64 &&
448 	    handle_tx2_tvm(vcpu))
449 		goto guest;
450 
451 	/*
452 	 * We trap the first access to the FP/SIMD to save the host context
453 	 * and restore the guest context lazily.
454 	 * If FP/SIMD is not implemented, handle the trap and inject an
455 	 * undefined instruction exception to the guest.
456 	 * Similarly for trapped SVE accesses.
457 	 */
458 	if (__hyp_handle_fpsimd(vcpu))
459 		goto guest;
460 
461 	if (__hyp_handle_ptrauth(vcpu))
462 		goto guest;
463 
464 	if (!__populate_fault_info(vcpu))
465 		goto guest;
466 
467 	if (static_branch_unlikely(&vgic_v2_cpuif_trap)) {
468 		bool valid;
469 
470 		valid = kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_DABT_LOW &&
471 			kvm_vcpu_trap_get_fault_type(vcpu) == FSC_FAULT &&
472 			kvm_vcpu_dabt_isvalid(vcpu) &&
473 			!kvm_vcpu_abt_issea(vcpu) &&
474 			!kvm_vcpu_abt_iss1tw(vcpu);
475 
476 		if (valid) {
477 			int ret = __vgic_v2_perform_cpuif_access(vcpu);
478 
479 			if (ret == 1)
480 				goto guest;
481 
482 			/* Promote an illegal access to an SError.*/
483 			if (ret == -1)
484 				*exit_code = ARM_EXCEPTION_EL1_SERROR;
485 
486 			goto exit;
487 		}
488 	}
489 
490 	if (static_branch_unlikely(&vgic_v3_cpuif_trap) &&
491 	    (kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_SYS64 ||
492 	     kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_CP15_32)) {
493 		int ret = __vgic_v3_perform_cpuif_access(vcpu);
494 
495 		if (ret == 1)
496 			goto guest;
497 	}
498 
499 exit:
500 	/* Return to the host kernel and handle the exit */
501 	return false;
502 
503 guest:
504 	/* Re-enter the guest */
505 	asm(ALTERNATIVE("nop", "dmb sy", ARM64_WORKAROUND_1508412));
506 	return true;
507 }
508 
509 static inline void __kvm_unexpected_el2_exception(void)
510 {
511 	extern char __guest_exit_panic[];
512 	unsigned long addr, fixup;
513 	struct exception_table_entry *entry, *end;
514 	unsigned long elr_el2 = read_sysreg(elr_el2);
515 
516 	entry = &__start___kvm_ex_table;
517 	end = &__stop___kvm_ex_table;
518 
519 	while (entry < end) {
520 		addr = (unsigned long)&entry->insn + entry->insn;
521 		fixup = (unsigned long)&entry->fixup + entry->fixup;
522 
523 		if (addr != elr_el2) {
524 			entry++;
525 			continue;
526 		}
527 
528 		write_sysreg(fixup, elr_el2);
529 		return;
530 	}
531 
532 	/* Trigger a panic after restoring the hyp context. */
533 	write_sysreg(__guest_exit_panic, elr_el2);
534 }
535 
536 #endif /* __ARM64_KVM_HYP_SWITCH_H__ */
537