xref: /openbmc/linux/arch/arm64/kvm/hyp/include/hyp/switch.h (revision 65b96377) 
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/fpsimd.h>
30 #include <asm/debug-monitors.h>
31 #include <asm/processor.h>
32 
33 struct kvm_exception_table_entry {
34 	int insn, fixup;
35 };
36 
37 extern struct kvm_exception_table_entry __start___kvm_ex_table;
38 extern struct kvm_exception_table_entry __stop___kvm_ex_table;
39 
40 /* Check whether the FP regs were dirtied while in the host-side run loop: */
41 static inline bool update_fp_enabled(struct kvm_vcpu *vcpu)
42 {
43 	/*
44 	 * When the system doesn't support FP/SIMD, we cannot rely on
45 	 * the _TIF_FOREIGN_FPSTATE flag. However, we always inject an
46 	 * abort on the very first access to FP and thus we should never
47 	 * see KVM_ARM64_FP_ENABLED. For added safety, make sure we always
48 	 * trap the accesses.
49 	 */
50 	if (!system_supports_fpsimd() ||
51 	    vcpu->arch.flags & KVM_ARM64_FP_FOREIGN_FPSTATE)
52 		vcpu->arch.flags &= ~(KVM_ARM64_FP_ENABLED |
53 				      KVM_ARM64_FP_HOST);
54 
55 	return !!(vcpu->arch.flags & KVM_ARM64_FP_ENABLED);
56 }
57 
58 /* Save the 32-bit only FPSIMD system register state */
59 static inline void __fpsimd_save_fpexc32(struct kvm_vcpu *vcpu)
60 {
61 	if (!vcpu_el1_is_32bit(vcpu))
62 		return;
63 
64 	__vcpu_sys_reg(vcpu, FPEXC32_EL2) = read_sysreg(fpexc32_el2);
65 }
66 
67 static inline void __activate_traps_fpsimd32(struct kvm_vcpu *vcpu)
68 {
69 	/*
70 	 * We are about to set CPTR_EL2.TFP to trap all floating point
71 	 * register accesses to EL2, however, the ARM ARM clearly states that
72 	 * traps are only taken to EL2 if the operation would not otherwise
73 	 * trap to EL1.  Therefore, always make sure that for 32-bit guests,
74 	 * we set FPEXC.EN to prevent traps to EL1, when setting the TFP bit.
75 	 * If FP/ASIMD is not implemented, FPEXC is UNDEFINED and any access to
76 	 * it will cause an exception.
77 	 */
78 	if (vcpu_el1_is_32bit(vcpu) && system_supports_fpsimd()) {
79 		write_sysreg(1 << 30, fpexc32_el2);
80 		isb();
81 	}
82 }
83 
84 static inline void __activate_traps_common(struct kvm_vcpu *vcpu)
85 {
86 	/* Trap on AArch32 cp15 c15 (impdef sysregs) accesses (EL1 or EL0) */
87 	write_sysreg(1 << 15, hstr_el2);
88 
89 	/*
90 	 * Make sure we trap PMU access from EL0 to EL2. Also sanitize
91 	 * PMSELR_EL0 to make sure it never contains the cycle
92 	 * counter, which could make a PMXEVCNTR_EL0 access UNDEF at
93 	 * EL1 instead of being trapped to EL2.
94 	 */
95 	if (kvm_arm_support_pmu_v3()) {
96 		write_sysreg(0, pmselr_el0);
97 		write_sysreg(ARMV8_PMU_USERENR_MASK, pmuserenr_el0);
98 	}
99 
100 	vcpu->arch.mdcr_el2_host = read_sysreg(mdcr_el2);
101 	write_sysreg(vcpu->arch.mdcr_el2, mdcr_el2);
102 }
103 
104 static inline void __deactivate_traps_common(struct kvm_vcpu *vcpu)
105 {
106 	write_sysreg(vcpu->arch.mdcr_el2_host, mdcr_el2);
107 
108 	write_sysreg(0, hstr_el2);
109 	if (kvm_arm_support_pmu_v3())
110 		write_sysreg(0, pmuserenr_el0);
111 }
112 
113 static inline void ___activate_traps(struct kvm_vcpu *vcpu)
114 {
115 	u64 hcr = vcpu->arch.hcr_el2;
116 
117 	if (cpus_have_final_cap(ARM64_WORKAROUND_CAVIUM_TX2_219_TVM))
118 		hcr |= HCR_TVM;
119 
120 	write_sysreg(hcr, hcr_el2);
121 
122 	if (cpus_have_final_cap(ARM64_HAS_RAS_EXTN) && (hcr & HCR_VSE))
123 		write_sysreg_s(vcpu->arch.vsesr_el2, SYS_VSESR_EL2);
124 }
125 
126 static inline void ___deactivate_traps(struct kvm_vcpu *vcpu)
127 {
128 	/*
129 	 * If we pended a virtual abort, preserve it until it gets
130 	 * cleared. See D1.14.3 (Virtual Interrupts) for details, but
131 	 * the crucial bit is "On taking a vSError interrupt,
132 	 * HCR_EL2.VSE is cleared to 0."
133 	 */
134 	if (vcpu->arch.hcr_el2 & HCR_VSE) {
135 		vcpu->arch.hcr_el2 &= ~HCR_VSE;
136 		vcpu->arch.hcr_el2 |= read_sysreg(hcr_el2) & HCR_VSE;
137 	}
138 }
139 
140 static inline bool __populate_fault_info(struct kvm_vcpu *vcpu)
141 {
142 	return __get_fault_info(vcpu->arch.fault.esr_el2, &vcpu->arch.fault);
143 }
144 
145 static inline void __hyp_sve_restore_guest(struct kvm_vcpu *vcpu)
146 {
147 	sve_cond_update_zcr_vq(vcpu_sve_max_vq(vcpu) - 1, SYS_ZCR_EL2);
148 	__sve_restore_state(vcpu_sve_pffr(vcpu),
149 			    &vcpu->arch.ctxt.fp_regs.fpsr);
150 	write_sysreg_el1(__vcpu_sys_reg(vcpu, ZCR_EL1), SYS_ZCR);
151 }
152 
153 /*
154  * We trap the first access to the FP/SIMD to save the host context and
155  * restore the guest context lazily.
156  * If FP/SIMD is not implemented, handle the trap and inject an undefined
157  * instruction exception to the guest. Similarly for trapped SVE accesses.
158  */
159 static bool kvm_hyp_handle_fpsimd(struct kvm_vcpu *vcpu, u64 *exit_code)
160 {
161 	bool sve_guest;
162 	u8 esr_ec;
163 	u64 reg;
164 
165 	if (!system_supports_fpsimd())
166 		return false;
167 
168 	sve_guest = vcpu_has_sve(vcpu);
169 	esr_ec = kvm_vcpu_trap_get_class(vcpu);
170 
171 	/* Don't handle SVE traps for non-SVE vcpus here: */
172 	if (!sve_guest && esr_ec != ESR_ELx_EC_FP_ASIMD)
173 		return false;
174 
175 	/* Valid trap.  Switch the context: */
176 
177 	/* First disable enough traps to allow us to update the registers */
178 	if (has_vhe()) {
179 		reg = CPACR_EL1_FPEN_EL0EN | CPACR_EL1_FPEN_EL1EN;
180 		if (sve_guest)
181 			reg |= CPACR_EL1_ZEN_EL0EN | CPACR_EL1_ZEN_EL1EN;
182 
183 		sysreg_clear_set(cpacr_el1, 0, reg);
184 	} else {
185 		reg = CPTR_EL2_TFP;
186 		if (sve_guest)
187 			reg |= CPTR_EL2_TZ;
188 
189 		sysreg_clear_set(cptr_el2, reg, 0);
190 	}
191 	isb();
192 
193 	/* Write out the host state if it's in the registers */
194 	if (vcpu->arch.flags & KVM_ARM64_FP_HOST) {
195 		__fpsimd_save_state(vcpu->arch.host_fpsimd_state);
196 		vcpu->arch.flags &= ~KVM_ARM64_FP_HOST;
197 	}
198 
199 	/* Restore the guest state */
200 	if (sve_guest)
201 		__hyp_sve_restore_guest(vcpu);
202 	else
203 		__fpsimd_restore_state(&vcpu->arch.ctxt.fp_regs);
204 
205 	/* Skip restoring fpexc32 for AArch64 guests */
206 	if (!(read_sysreg(hcr_el2) & HCR_RW))
207 		write_sysreg(__vcpu_sys_reg(vcpu, FPEXC32_EL2), fpexc32_el2);
208 
209 	vcpu->arch.flags |= KVM_ARM64_FP_ENABLED;
210 
211 	return true;
212 }
213 
214 static inline bool handle_tx2_tvm(struct kvm_vcpu *vcpu)
215 {
216 	u32 sysreg = esr_sys64_to_sysreg(kvm_vcpu_get_esr(vcpu));
217 	int rt = kvm_vcpu_sys_get_rt(vcpu);
218 	u64 val = vcpu_get_reg(vcpu, rt);
219 
220 	/*
221 	 * The normal sysreg handling code expects to see the traps,
222 	 * let's not do anything here.
223 	 */
224 	if (vcpu->arch.hcr_el2 & HCR_TVM)
225 		return false;
226 
227 	switch (sysreg) {
228 	case SYS_SCTLR_EL1:
229 		write_sysreg_el1(val, SYS_SCTLR);
230 		break;
231 	case SYS_TTBR0_EL1:
232 		write_sysreg_el1(val, SYS_TTBR0);
233 		break;
234 	case SYS_TTBR1_EL1:
235 		write_sysreg_el1(val, SYS_TTBR1);
236 		break;
237 	case SYS_TCR_EL1:
238 		write_sysreg_el1(val, SYS_TCR);
239 		break;
240 	case SYS_ESR_EL1:
241 		write_sysreg_el1(val, SYS_ESR);
242 		break;
243 	case SYS_FAR_EL1:
244 		write_sysreg_el1(val, SYS_FAR);
245 		break;
246 	case SYS_AFSR0_EL1:
247 		write_sysreg_el1(val, SYS_AFSR0);
248 		break;
249 	case SYS_AFSR1_EL1:
250 		write_sysreg_el1(val, SYS_AFSR1);
251 		break;
252 	case SYS_MAIR_EL1:
253 		write_sysreg_el1(val, SYS_MAIR);
254 		break;
255 	case SYS_AMAIR_EL1:
256 		write_sysreg_el1(val, SYS_AMAIR);
257 		break;
258 	case SYS_CONTEXTIDR_EL1:
259 		write_sysreg_el1(val, SYS_CONTEXTIDR);
260 		break;
261 	default:
262 		return false;
263 	}
264 
265 	__kvm_skip_instr(vcpu);
266 	return true;
267 }
268 
269 static inline bool esr_is_ptrauth_trap(u32 esr)
270 {
271 	switch (esr_sys64_to_sysreg(esr)) {
272 	case SYS_APIAKEYLO_EL1:
273 	case SYS_APIAKEYHI_EL1:
274 	case SYS_APIBKEYLO_EL1:
275 	case SYS_APIBKEYHI_EL1:
276 	case SYS_APDAKEYLO_EL1:
277 	case SYS_APDAKEYHI_EL1:
278 	case SYS_APDBKEYLO_EL1:
279 	case SYS_APDBKEYHI_EL1:
280 	case SYS_APGAKEYLO_EL1:
281 	case SYS_APGAKEYHI_EL1:
282 		return true;
283 	}
284 
285 	return false;
286 }
287 
288 #define __ptrauth_save_key(ctxt, key)					\
289 	do {								\
290 	u64 __val;                                                      \
291 	__val = read_sysreg_s(SYS_ ## key ## KEYLO_EL1);                \
292 	ctxt_sys_reg(ctxt, key ## KEYLO_EL1) = __val;                   \
293 	__val = read_sysreg_s(SYS_ ## key ## KEYHI_EL1);                \
294 	ctxt_sys_reg(ctxt, key ## KEYHI_EL1) = __val;                   \
295 } while(0)
296 
297 DECLARE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt);
298 
299 static bool kvm_hyp_handle_ptrauth(struct kvm_vcpu *vcpu, u64 *exit_code)
300 {
301 	struct kvm_cpu_context *ctxt;
302 	u64 val;
303 
304 	if (!vcpu_has_ptrauth(vcpu))
305 		return false;
306 
307 	ctxt = this_cpu_ptr(&kvm_hyp_ctxt);
308 	__ptrauth_save_key(ctxt, APIA);
309 	__ptrauth_save_key(ctxt, APIB);
310 	__ptrauth_save_key(ctxt, APDA);
311 	__ptrauth_save_key(ctxt, APDB);
312 	__ptrauth_save_key(ctxt, APGA);
313 
314 	vcpu_ptrauth_enable(vcpu);
315 
316 	val = read_sysreg(hcr_el2);
317 	val |= (HCR_API | HCR_APK);
318 	write_sysreg(val, hcr_el2);
319 
320 	return true;
321 }
322 
323 static bool kvm_hyp_handle_sysreg(struct kvm_vcpu *vcpu, u64 *exit_code)
324 {
325 	if (cpus_have_final_cap(ARM64_WORKAROUND_CAVIUM_TX2_219_TVM) &&
326 	    handle_tx2_tvm(vcpu))
327 		return true;
328 
329 	if (static_branch_unlikely(&vgic_v3_cpuif_trap) &&
330 	    __vgic_v3_perform_cpuif_access(vcpu) == 1)
331 		return true;
332 
333 	if (esr_is_ptrauth_trap(kvm_vcpu_get_esr(vcpu)))
334 		return kvm_hyp_handle_ptrauth(vcpu, exit_code);
335 
336 	return false;
337 }
338 
339 static bool kvm_hyp_handle_cp15_32(struct kvm_vcpu *vcpu, u64 *exit_code)
340 {
341 	if (static_branch_unlikely(&vgic_v3_cpuif_trap) &&
342 	    __vgic_v3_perform_cpuif_access(vcpu) == 1)
343 		return true;
344 
345 	return false;
346 }
347 
348 static bool kvm_hyp_handle_iabt_low(struct kvm_vcpu *vcpu, u64 *exit_code)
349 {
350 	if (!__populate_fault_info(vcpu))
351 		return true;
352 
353 	return false;
354 }
355 
356 static bool kvm_hyp_handle_dabt_low(struct kvm_vcpu *vcpu, u64 *exit_code)
357 {
358 	if (!__populate_fault_info(vcpu))
359 		return true;
360 
361 	if (static_branch_unlikely(&vgic_v2_cpuif_trap)) {
362 		bool valid;
363 
364 		valid = kvm_vcpu_trap_get_fault_type(vcpu) == FSC_FAULT &&
365 			kvm_vcpu_dabt_isvalid(vcpu) &&
366 			!kvm_vcpu_abt_issea(vcpu) &&
367 			!kvm_vcpu_abt_iss1tw(vcpu);
368 
369 		if (valid) {
370 			int ret = __vgic_v2_perform_cpuif_access(vcpu);
371 
372 			if (ret == 1)
373 				return true;
374 
375 			/* Promote an illegal access to an SError.*/
376 			if (ret == -1)
377 				*exit_code = ARM_EXCEPTION_EL1_SERROR;
378 		}
379 	}
380 
381 	return false;
382 }
383 
384 typedef bool (*exit_handler_fn)(struct kvm_vcpu *, u64 *);
385 
386 static const exit_handler_fn *kvm_get_exit_handler_array(struct kvm_vcpu *vcpu);
387 
388 static void early_exit_filter(struct kvm_vcpu *vcpu, u64 *exit_code);
389 
390 /*
391  * Allow the hypervisor to handle the exit with an exit handler if it has one.
392  *
393  * Returns true if the hypervisor handled the exit, and control should go back
394  * to the guest, or false if it hasn't.
395  */
396 static inline bool kvm_hyp_handle_exit(struct kvm_vcpu *vcpu, u64 *exit_code)
397 {
398 	const exit_handler_fn *handlers = kvm_get_exit_handler_array(vcpu);
399 	exit_handler_fn fn;
400 
401 	fn = handlers[kvm_vcpu_trap_get_class(vcpu)];
402 
403 	if (fn)
404 		return fn(vcpu, exit_code);
405 
406 	return false;
407 }
408 
409 static inline void synchronize_vcpu_pstate(struct kvm_vcpu *vcpu, u64 *exit_code)
410 {
411 	/*
412 	 * Check for the conditions of Cortex-A510's #2077057. When these occur
413 	 * SPSR_EL2 can't be trusted, but isn't needed either as it is
414 	 * unchanged from the value in vcpu_gp_regs(vcpu)->pstate.
415 	 * Are we single-stepping the guest, and took a PAC exception from the
416 	 * active-not-pending state?
417 	 */
418 	if (cpus_have_final_cap(ARM64_WORKAROUND_2077057)		&&
419 	    vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP			&&
420 	    *vcpu_cpsr(vcpu) & DBG_SPSR_SS				&&
421 	    ESR_ELx_EC(read_sysreg_el2(SYS_ESR)) == ESR_ELx_EC_PAC)
422 		write_sysreg_el2(*vcpu_cpsr(vcpu), SYS_SPSR);
423 
424 	vcpu->arch.ctxt.regs.pstate = read_sysreg_el2(SYS_SPSR);
425 }
426 
427 /*
428  * Return true when we were able to fixup the guest exit and should return to
429  * the guest, false when we should restore the host state and return to the
430  * main run loop.
431  */
432 static inline bool fixup_guest_exit(struct kvm_vcpu *vcpu, u64 *exit_code)
433 {
434 	/*
435 	 * Save PSTATE early so that we can evaluate the vcpu mode
436 	 * early on.
437 	 */
438 	synchronize_vcpu_pstate(vcpu, exit_code);
439 
440 	/*
441 	 * Check whether we want to repaint the state one way or
442 	 * another.
443 	 */
444 	early_exit_filter(vcpu, exit_code);
445 
446 	if (ARM_EXCEPTION_CODE(*exit_code) != ARM_EXCEPTION_IRQ)
447 		vcpu->arch.fault.esr_el2 = read_sysreg_el2(SYS_ESR);
448 
449 	if (ARM_SERROR_PENDING(*exit_code) &&
450 	    ARM_EXCEPTION_CODE(*exit_code) != ARM_EXCEPTION_IRQ) {
451 		u8 esr_ec = kvm_vcpu_trap_get_class(vcpu);
452 
453 		/*
454 		 * HVC already have an adjusted PC, which we need to
455 		 * correct in order to return to after having injected
456 		 * the SError.
457 		 *
458 		 * SMC, on the other hand, is *trapped*, meaning its
459 		 * preferred return address is the SMC itself.
460 		 */
461 		if (esr_ec == ESR_ELx_EC_HVC32 || esr_ec == ESR_ELx_EC_HVC64)
462 			write_sysreg_el2(read_sysreg_el2(SYS_ELR) - 4, SYS_ELR);
463 	}
464 
465 	/*
466 	 * We're using the raw exception code in order to only process
467 	 * the trap if no SError is pending. We will come back to the
468 	 * same PC once the SError has been injected, and replay the
469 	 * trapping instruction.
470 	 */
471 	if (*exit_code != ARM_EXCEPTION_TRAP)
472 		goto exit;
473 
474 	/* Check if there's an exit handler and allow it to handle the exit. */
475 	if (kvm_hyp_handle_exit(vcpu, exit_code))
476 		goto guest;
477 exit:
478 	/* Return to the host kernel and handle the exit */
479 	return false;
480 
481 guest:
482 	/* Re-enter the guest */
483 	asm(ALTERNATIVE("nop", "dmb sy", ARM64_WORKAROUND_1508412));
484 	return true;
485 }
486 
487 static inline void __kvm_unexpected_el2_exception(void)
488 {
489 	extern char __guest_exit_panic[];
490 	unsigned long addr, fixup;
491 	struct kvm_exception_table_entry *entry, *end;
492 	unsigned long elr_el2 = read_sysreg(elr_el2);
493 
494 	entry = &__start___kvm_ex_table;
495 	end = &__stop___kvm_ex_table;
496 
497 	while (entry < end) {
498 		addr = (unsigned long)&entry->insn + entry->insn;
499 		fixup = (unsigned long)&entry->fixup + entry->fixup;
500 
501 		if (addr != elr_el2) {
502 			entry++;
503 			continue;
504 		}
505 
506 		write_sysreg(fixup, elr_el2);
507 		return;
508 	}
509 
510 	/* Trigger a panic after restoring the hyp context. */
511 	write_sysreg(__guest_exit_panic, elr_el2);
512 }
513 
514 #endif /* __ARM64_KVM_HYP_SWITCH_H__ */
515