xref: /linux/arch/arm64/kvm/handle_exit.c (revision d5859510d35d8e7d63fed5169f1775317f40fb03)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2012,2013 - ARM Ltd
4  * Author: Marc Zyngier <marc.zyngier@arm.com>
5  *
6  * Derived from arch/arm/kvm/handle_exit.c:
7  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
8  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
9  */
10 
11 #include <linux/kvm.h>
12 #include <linux/kvm_host.h>
13 
14 #include <asm/esr.h>
15 #include <asm/exception.h>
16 #include <asm/kvm_asm.h>
17 #include <asm/kvm_emulate.h>
18 #include <asm/kvm_mmu.h>
19 #include <asm/kvm_nested.h>
20 #include <asm/debug-monitors.h>
21 #include <asm/stacktrace/nvhe.h>
22 #include <asm/traps.h>
23 
24 #include <kvm/arm_hypercalls.h>
25 
26 #define CREATE_TRACE_POINTS
27 #include "trace_handle_exit.h"
28 
29 typedef int (*exit_handle_fn)(struct kvm_vcpu *);
30 
31 static void kvm_handle_guest_serror(struct kvm_vcpu *vcpu, u64 esr)
32 {
33 	if (!arm64_is_ras_serror(esr) || arm64_is_fatal_ras_serror(NULL, esr))
34 		kvm_inject_vabt(vcpu);
35 }
36 
37 static int handle_hvc(struct kvm_vcpu *vcpu)
38 {
39 	trace_kvm_hvc_arm64(*vcpu_pc(vcpu), vcpu_get_reg(vcpu, 0),
40 			    kvm_vcpu_hvc_get_imm(vcpu));
41 	vcpu->stat.hvc_exit_stat++;
42 
43 	/* Forward hvc instructions to the virtual EL2 if the guest has EL2. */
44 	if (vcpu_has_nv(vcpu)) {
45 		if (vcpu_read_sys_reg(vcpu, HCR_EL2) & HCR_HCD)
46 			kvm_inject_undefined(vcpu);
47 		else
48 			kvm_inject_nested_sync(vcpu, kvm_vcpu_get_esr(vcpu));
49 
50 		return 1;
51 	}
52 
53 	return kvm_smccc_call_handler(vcpu);
54 }
55 
56 static int handle_smc(struct kvm_vcpu *vcpu)
57 {
58 	/*
59 	 * Forward this trapped smc instruction to the virtual EL2 if
60 	 * the guest has asked for it.
61 	 */
62 	if (forward_smc_trap(vcpu))
63 		return 1;
64 
65 	/*
66 	 * "If an SMC instruction executed at Non-secure EL1 is
67 	 * trapped to EL2 because HCR_EL2.TSC is 1, the exception is a
68 	 * Trap exception, not a Secure Monitor Call exception [...]"
69 	 *
70 	 * We need to advance the PC after the trap, as it would
71 	 * otherwise return to the same address. Furthermore, pre-incrementing
72 	 * the PC before potentially exiting to userspace maintains the same
73 	 * abstraction for both SMCs and HVCs.
74 	 */
75 	kvm_incr_pc(vcpu);
76 
77 	/*
78 	 * SMCs with a nonzero immediate are reserved according to DEN0028E 2.9
79 	 * "SMC and HVC immediate value".
80 	 */
81 	if (kvm_vcpu_hvc_get_imm(vcpu)) {
82 		vcpu_set_reg(vcpu, 0, ~0UL);
83 		return 1;
84 	}
85 
86 	/*
87 	 * If imm is zero then it is likely an SMCCC call.
88 	 *
89 	 * Note that on ARMv8.3, even if EL3 is not implemented, SMC executed
90 	 * at Non-secure EL1 is trapped to EL2 if HCR_EL2.TSC==1, rather than
91 	 * being treated as UNDEFINED.
92 	 */
93 	return kvm_smccc_call_handler(vcpu);
94 }
95 
96 /*
97  * Guest access to FP/ASIMD registers are routed to this handler only
98  * when the system doesn't support FP/ASIMD.
99  */
100 static int handle_no_fpsimd(struct kvm_vcpu *vcpu)
101 {
102 	kvm_inject_undefined(vcpu);
103 	return 1;
104 }
105 
106 /**
107  * kvm_handle_wfx - handle a wait-for-interrupts or wait-for-event
108  *		    instruction executed by a guest
109  *
110  * @vcpu:	the vcpu pointer
111  *
112  * WFE[T]: Yield the CPU and come back to this vcpu when the scheduler
113  * decides to.
114  * WFI: Simply call kvm_vcpu_halt(), which will halt execution of
115  * world-switches and schedule other host processes until there is an
116  * incoming IRQ or FIQ to the VM.
117  * WFIT: Same as WFI, with a timed wakeup implemented as a background timer
118  *
119  * WF{I,E}T can immediately return if the deadline has already expired.
120  */
121 static int kvm_handle_wfx(struct kvm_vcpu *vcpu)
122 {
123 	u64 esr = kvm_vcpu_get_esr(vcpu);
124 
125 	if (esr & ESR_ELx_WFx_ISS_WFE) {
126 		trace_kvm_wfx_arm64(*vcpu_pc(vcpu), true);
127 		vcpu->stat.wfe_exit_stat++;
128 	} else {
129 		trace_kvm_wfx_arm64(*vcpu_pc(vcpu), false);
130 		vcpu->stat.wfi_exit_stat++;
131 	}
132 
133 	if (esr & ESR_ELx_WFx_ISS_WFxT) {
134 		if (esr & ESR_ELx_WFx_ISS_RV) {
135 			u64 val, now;
136 
137 			now = kvm_arm_timer_get_reg(vcpu, KVM_REG_ARM_TIMER_CNT);
138 			val = vcpu_get_reg(vcpu, kvm_vcpu_sys_get_rt(vcpu));
139 
140 			if (now >= val)
141 				goto out;
142 		} else {
143 			/* Treat WFxT as WFx if RN is invalid */
144 			esr &= ~ESR_ELx_WFx_ISS_WFxT;
145 		}
146 	}
147 
148 	if (esr & ESR_ELx_WFx_ISS_WFE) {
149 		kvm_vcpu_on_spin(vcpu, vcpu_mode_priv(vcpu));
150 	} else {
151 		if (esr & ESR_ELx_WFx_ISS_WFxT)
152 			vcpu_set_flag(vcpu, IN_WFIT);
153 
154 		kvm_vcpu_wfi(vcpu);
155 	}
156 out:
157 	kvm_incr_pc(vcpu);
158 
159 	return 1;
160 }
161 
162 /**
163  * kvm_handle_guest_debug - handle a debug exception instruction
164  *
165  * @vcpu:	the vcpu pointer
166  *
167  * We route all debug exceptions through the same handler. If both the
168  * guest and host are using the same debug facilities it will be up to
169  * userspace to re-inject the correct exception for guest delivery.
170  *
171  * @return: 0 (while setting vcpu->run->exit_reason)
172  */
173 static int kvm_handle_guest_debug(struct kvm_vcpu *vcpu)
174 {
175 	struct kvm_run *run = vcpu->run;
176 	u64 esr = kvm_vcpu_get_esr(vcpu);
177 
178 	run->exit_reason = KVM_EXIT_DEBUG;
179 	run->debug.arch.hsr = lower_32_bits(esr);
180 	run->debug.arch.hsr_high = upper_32_bits(esr);
181 	run->flags = KVM_DEBUG_ARCH_HSR_HIGH_VALID;
182 
183 	switch (ESR_ELx_EC(esr)) {
184 	case ESR_ELx_EC_WATCHPT_LOW:
185 		run->debug.arch.far = vcpu->arch.fault.far_el2;
186 		break;
187 	case ESR_ELx_EC_SOFTSTP_LOW:
188 		vcpu_clear_flag(vcpu, DBG_SS_ACTIVE_PENDING);
189 		break;
190 	}
191 
192 	return 0;
193 }
194 
195 static int kvm_handle_unknown_ec(struct kvm_vcpu *vcpu)
196 {
197 	u64 esr = kvm_vcpu_get_esr(vcpu);
198 
199 	kvm_pr_unimpl("Unknown exception class: esr: %#016llx -- %s\n",
200 		      esr, esr_get_class_string(esr));
201 
202 	kvm_inject_undefined(vcpu);
203 	return 1;
204 }
205 
206 /*
207  * Guest access to SVE registers should be routed to this handler only
208  * when the system doesn't support SVE.
209  */
210 static int handle_sve(struct kvm_vcpu *vcpu)
211 {
212 	kvm_inject_undefined(vcpu);
213 	return 1;
214 }
215 
216 /*
217  * Two possibilities to handle a trapping ptrauth instruction:
218  *
219  * - Guest usage of a ptrauth instruction (which the guest EL1 did not
220  *   turn into a NOP). If we get here, it is because we didn't enable
221  *   ptrauth for the guest. This results in an UNDEF, as it isn't
222  *   supposed to use ptrauth without being told it could.
223  *
224  * - Running an L2 NV guest while L1 has left HCR_EL2.API==0, and for
225  *   which we reinject the exception into L1.
226  *
227  * Anything else is an emulation bug (hence the WARN_ON + UNDEF).
228  */
229 static int kvm_handle_ptrauth(struct kvm_vcpu *vcpu)
230 {
231 	if (!vcpu_has_ptrauth(vcpu)) {
232 		kvm_inject_undefined(vcpu);
233 		return 1;
234 	}
235 
236 	if (vcpu_has_nv(vcpu) && !is_hyp_ctxt(vcpu)) {
237 		kvm_inject_nested_sync(vcpu, kvm_vcpu_get_esr(vcpu));
238 		return 1;
239 	}
240 
241 	/* Really shouldn't be here! */
242 	WARN_ON_ONCE(1);
243 	kvm_inject_undefined(vcpu);
244 	return 1;
245 }
246 
247 static int kvm_handle_eret(struct kvm_vcpu *vcpu)
248 {
249 	if (esr_iss_is_eretax(kvm_vcpu_get_esr(vcpu)) &&
250 	    !vcpu_has_ptrauth(vcpu))
251 		return kvm_handle_ptrauth(vcpu);
252 
253 	/*
254 	 * If we got here, two possibilities:
255 	 *
256 	 * - the guest is in EL2, and we need to fully emulate ERET
257 	 *
258 	 * - the guest is in EL1, and we need to reinject the
259          *   exception into the L1 hypervisor.
260 	 *
261 	 * If KVM ever traps ERET for its own use, we'll have to
262 	 * revisit this.
263 	 */
264 	if (is_hyp_ctxt(vcpu))
265 		kvm_emulate_nested_eret(vcpu);
266 	else
267 		kvm_inject_nested_sync(vcpu, kvm_vcpu_get_esr(vcpu));
268 
269 	return 1;
270 }
271 
272 static int handle_svc(struct kvm_vcpu *vcpu)
273 {
274 	/*
275 	 * So far, SVC traps only for NV via HFGITR_EL2. A SVC from a
276 	 * 32bit guest would be caught by vpcu_mode_is_bad_32bit(), so
277 	 * we should only have to deal with a 64 bit exception.
278 	 */
279 	kvm_inject_nested_sync(vcpu, kvm_vcpu_get_esr(vcpu));
280 	return 1;
281 }
282 
283 static exit_handle_fn arm_exit_handlers[] = {
284 	[0 ... ESR_ELx_EC_MAX]	= kvm_handle_unknown_ec,
285 	[ESR_ELx_EC_WFx]	= kvm_handle_wfx,
286 	[ESR_ELx_EC_CP15_32]	= kvm_handle_cp15_32,
287 	[ESR_ELx_EC_CP15_64]	= kvm_handle_cp15_64,
288 	[ESR_ELx_EC_CP14_MR]	= kvm_handle_cp14_32,
289 	[ESR_ELx_EC_CP14_LS]	= kvm_handle_cp14_load_store,
290 	[ESR_ELx_EC_CP10_ID]	= kvm_handle_cp10_id,
291 	[ESR_ELx_EC_CP14_64]	= kvm_handle_cp14_64,
292 	[ESR_ELx_EC_HVC32]	= handle_hvc,
293 	[ESR_ELx_EC_SMC32]	= handle_smc,
294 	[ESR_ELx_EC_HVC64]	= handle_hvc,
295 	[ESR_ELx_EC_SMC64]	= handle_smc,
296 	[ESR_ELx_EC_SVC64]	= handle_svc,
297 	[ESR_ELx_EC_SYS64]	= kvm_handle_sys_reg,
298 	[ESR_ELx_EC_SVE]	= handle_sve,
299 	[ESR_ELx_EC_ERET]	= kvm_handle_eret,
300 	[ESR_ELx_EC_IABT_LOW]	= kvm_handle_guest_abort,
301 	[ESR_ELx_EC_DABT_LOW]	= kvm_handle_guest_abort,
302 	[ESR_ELx_EC_SOFTSTP_LOW]= kvm_handle_guest_debug,
303 	[ESR_ELx_EC_WATCHPT_LOW]= kvm_handle_guest_debug,
304 	[ESR_ELx_EC_BREAKPT_LOW]= kvm_handle_guest_debug,
305 	[ESR_ELx_EC_BKPT32]	= kvm_handle_guest_debug,
306 	[ESR_ELx_EC_BRK64]	= kvm_handle_guest_debug,
307 	[ESR_ELx_EC_FP_ASIMD]	= handle_no_fpsimd,
308 	[ESR_ELx_EC_PAC]	= kvm_handle_ptrauth,
309 };
310 
311 static exit_handle_fn kvm_get_exit_handler(struct kvm_vcpu *vcpu)
312 {
313 	u64 esr = kvm_vcpu_get_esr(vcpu);
314 	u8 esr_ec = ESR_ELx_EC(esr);
315 
316 	return arm_exit_handlers[esr_ec];
317 }
318 
319 /*
320  * We may be single-stepping an emulated instruction. If the emulation
321  * has been completed in the kernel, we can return to userspace with a
322  * KVM_EXIT_DEBUG, otherwise userspace needs to complete its
323  * emulation first.
324  */
325 static int handle_trap_exceptions(struct kvm_vcpu *vcpu)
326 {
327 	int handled;
328 
329 	/*
330 	 * See ARM ARM B1.14.1: "Hyp traps on instructions
331 	 * that fail their condition code check"
332 	 */
333 	if (!kvm_condition_valid(vcpu)) {
334 		kvm_incr_pc(vcpu);
335 		handled = 1;
336 	} else {
337 		exit_handle_fn exit_handler;
338 
339 		exit_handler = kvm_get_exit_handler(vcpu);
340 		handled = exit_handler(vcpu);
341 	}
342 
343 	return handled;
344 }
345 
346 /*
347  * Return > 0 to return to guest, < 0 on error, 0 (and set exit_reason) on
348  * proper exit to userspace.
349  */
350 int handle_exit(struct kvm_vcpu *vcpu, int exception_index)
351 {
352 	struct kvm_run *run = vcpu->run;
353 
354 	if (ARM_SERROR_PENDING(exception_index)) {
355 		/*
356 		 * The SError is handled by handle_exit_early(). If the guest
357 		 * survives it will re-execute the original instruction.
358 		 */
359 		return 1;
360 	}
361 
362 	exception_index = ARM_EXCEPTION_CODE(exception_index);
363 
364 	switch (exception_index) {
365 	case ARM_EXCEPTION_IRQ:
366 		return 1;
367 	case ARM_EXCEPTION_EL1_SERROR:
368 		return 1;
369 	case ARM_EXCEPTION_TRAP:
370 		return handle_trap_exceptions(vcpu);
371 	case ARM_EXCEPTION_HYP_GONE:
372 		/*
373 		 * EL2 has been reset to the hyp-stub. This happens when a guest
374 		 * is pre-emptied by kvm_reboot()'s shutdown call.
375 		 */
376 		run->exit_reason = KVM_EXIT_FAIL_ENTRY;
377 		return 0;
378 	case ARM_EXCEPTION_IL:
379 		/*
380 		 * We attempted an illegal exception return.  Guest state must
381 		 * have been corrupted somehow.  Give up.
382 		 */
383 		run->exit_reason = KVM_EXIT_FAIL_ENTRY;
384 		return -EINVAL;
385 	default:
386 		kvm_pr_unimpl("Unsupported exception type: %d",
387 			      exception_index);
388 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
389 		return 0;
390 	}
391 }
392 
393 /* For exit types that need handling before we can be preempted */
394 void handle_exit_early(struct kvm_vcpu *vcpu, int exception_index)
395 {
396 	if (ARM_SERROR_PENDING(exception_index)) {
397 		if (this_cpu_has_cap(ARM64_HAS_RAS_EXTN)) {
398 			u64 disr = kvm_vcpu_get_disr(vcpu);
399 
400 			kvm_handle_guest_serror(vcpu, disr_to_esr(disr));
401 		} else {
402 			kvm_inject_vabt(vcpu);
403 		}
404 
405 		return;
406 	}
407 
408 	exception_index = ARM_EXCEPTION_CODE(exception_index);
409 
410 	if (exception_index == ARM_EXCEPTION_EL1_SERROR)
411 		kvm_handle_guest_serror(vcpu, kvm_vcpu_get_esr(vcpu));
412 }
413 
414 void __noreturn __cold nvhe_hyp_panic_handler(u64 esr, u64 spsr,
415 					      u64 elr_virt, u64 elr_phys,
416 					      u64 par, uintptr_t vcpu,
417 					      u64 far, u64 hpfar) {
418 	u64 elr_in_kimg = __phys_to_kimg(elr_phys);
419 	u64 hyp_offset = elr_in_kimg - kaslr_offset() - elr_virt;
420 	u64 mode = spsr & PSR_MODE_MASK;
421 	u64 panic_addr = elr_virt + hyp_offset;
422 
423 	if (mode != PSR_MODE_EL2t && mode != PSR_MODE_EL2h) {
424 		kvm_err("Invalid host exception to nVHE hyp!\n");
425 	} else if (ESR_ELx_EC(esr) == ESR_ELx_EC_BRK64 &&
426 		   (esr & ESR_ELx_BRK64_ISS_COMMENT_MASK) == BUG_BRK_IMM) {
427 		const char *file = NULL;
428 		unsigned int line = 0;
429 
430 		/* All hyp bugs, including warnings, are treated as fatal. */
431 		if (!is_protected_kvm_enabled() ||
432 		    IS_ENABLED(CONFIG_NVHE_EL2_DEBUG)) {
433 			struct bug_entry *bug = find_bug(elr_in_kimg);
434 
435 			if (bug)
436 				bug_get_file_line(bug, &file, &line);
437 		}
438 
439 		if (file)
440 			kvm_err("nVHE hyp BUG at: %s:%u!\n", file, line);
441 		else
442 			kvm_err("nVHE hyp BUG at: [<%016llx>] %pB!\n", panic_addr,
443 					(void *)(panic_addr + kaslr_offset()));
444 	} else {
445 		kvm_err("nVHE hyp panic at: [<%016llx>] %pB!\n", panic_addr,
446 				(void *)(panic_addr + kaslr_offset()));
447 	}
448 
449 	/* Dump the nVHE hypervisor backtrace */
450 	kvm_nvhe_dump_backtrace(hyp_offset);
451 
452 	/*
453 	 * Hyp has panicked and we're going to handle that by panicking the
454 	 * kernel. The kernel offset will be revealed in the panic so we're
455 	 * also safe to reveal the hyp offset as a debugging aid for translating
456 	 * hyp VAs to vmlinux addresses.
457 	 */
458 	kvm_err("Hyp Offset: 0x%llx\n", hyp_offset);
459 
460 	panic("HYP panic:\nPS:%08llx PC:%016llx ESR:%016llx\nFAR:%016llx HPFAR:%016llx PAR:%016llx\nVCPU:%016lx\n",
461 	      spsr, elr_virt, esr, far, hpfar, par, vcpu);
462 }
463