xref: /openbmc/linux/arch/arm64/kvm/handle_exit.c (revision 2dd6532e)
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/debug-monitors.h>
20 #include <asm/traps.h>
21 
22 #include <kvm/arm_hypercalls.h>
23 
24 #define CREATE_TRACE_POINTS
25 #include "trace_handle_exit.h"
26 
27 typedef int (*exit_handle_fn)(struct kvm_vcpu *);
28 
29 static void kvm_handle_guest_serror(struct kvm_vcpu *vcpu, u64 esr)
30 {
31 	if (!arm64_is_ras_serror(esr) || arm64_is_fatal_ras_serror(NULL, esr))
32 		kvm_inject_vabt(vcpu);
33 }
34 
35 static int handle_hvc(struct kvm_vcpu *vcpu)
36 {
37 	int ret;
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 	ret = kvm_hvc_call_handler(vcpu);
44 	if (ret < 0) {
45 		vcpu_set_reg(vcpu, 0, ~0UL);
46 		return 1;
47 	}
48 
49 	return ret;
50 }
51 
52 static int handle_smc(struct kvm_vcpu *vcpu)
53 {
54 	/*
55 	 * "If an SMC instruction executed at Non-secure EL1 is
56 	 * trapped to EL2 because HCR_EL2.TSC is 1, the exception is a
57 	 * Trap exception, not a Secure Monitor Call exception [...]"
58 	 *
59 	 * We need to advance the PC after the trap, as it would
60 	 * otherwise return to the same address...
61 	 */
62 	vcpu_set_reg(vcpu, 0, ~0UL);
63 	kvm_incr_pc(vcpu);
64 	return 1;
65 }
66 
67 /*
68  * Guest access to FP/ASIMD registers are routed to this handler only
69  * when the system doesn't support FP/ASIMD.
70  */
71 static int handle_no_fpsimd(struct kvm_vcpu *vcpu)
72 {
73 	kvm_inject_undefined(vcpu);
74 	return 1;
75 }
76 
77 /**
78  * kvm_handle_wfx - handle a wait-for-interrupts or wait-for-event
79  *		    instruction executed by a guest
80  *
81  * @vcpu:	the vcpu pointer
82  *
83  * WFE[T]: Yield the CPU and come back to this vcpu when the scheduler
84  * decides to.
85  * WFI: Simply call kvm_vcpu_halt(), which will halt execution of
86  * world-switches and schedule other host processes until there is an
87  * incoming IRQ or FIQ to the VM.
88  * WFIT: Same as WFI, with a timed wakeup implemented as a background timer
89  *
90  * WF{I,E}T can immediately return if the deadline has already expired.
91  */
92 static int kvm_handle_wfx(struct kvm_vcpu *vcpu)
93 {
94 	u64 esr = kvm_vcpu_get_esr(vcpu);
95 
96 	if (esr & ESR_ELx_WFx_ISS_WFE) {
97 		trace_kvm_wfx_arm64(*vcpu_pc(vcpu), true);
98 		vcpu->stat.wfe_exit_stat++;
99 	} else {
100 		trace_kvm_wfx_arm64(*vcpu_pc(vcpu), false);
101 		vcpu->stat.wfi_exit_stat++;
102 	}
103 
104 	if (esr & ESR_ELx_WFx_ISS_WFxT) {
105 		if (esr & ESR_ELx_WFx_ISS_RV) {
106 			u64 val, now;
107 
108 			now = kvm_arm_timer_get_reg(vcpu, KVM_REG_ARM_TIMER_CNT);
109 			val = vcpu_get_reg(vcpu, kvm_vcpu_sys_get_rt(vcpu));
110 
111 			if (now >= val)
112 				goto out;
113 		} else {
114 			/* Treat WFxT as WFx if RN is invalid */
115 			esr &= ~ESR_ELx_WFx_ISS_WFxT;
116 		}
117 	}
118 
119 	if (esr & ESR_ELx_WFx_ISS_WFE) {
120 		kvm_vcpu_on_spin(vcpu, vcpu_mode_priv(vcpu));
121 	} else {
122 		if (esr & ESR_ELx_WFx_ISS_WFxT)
123 			vcpu->arch.flags |= KVM_ARM64_WFIT;
124 
125 		kvm_vcpu_wfi(vcpu);
126 	}
127 out:
128 	kvm_incr_pc(vcpu);
129 
130 	return 1;
131 }
132 
133 /**
134  * kvm_handle_guest_debug - handle a debug exception instruction
135  *
136  * @vcpu:	the vcpu pointer
137  *
138  * We route all debug exceptions through the same handler. If both the
139  * guest and host are using the same debug facilities it will be up to
140  * userspace to re-inject the correct exception for guest delivery.
141  *
142  * @return: 0 (while setting vcpu->run->exit_reason)
143  */
144 static int kvm_handle_guest_debug(struct kvm_vcpu *vcpu)
145 {
146 	struct kvm_run *run = vcpu->run;
147 	u64 esr = kvm_vcpu_get_esr(vcpu);
148 
149 	run->exit_reason = KVM_EXIT_DEBUG;
150 	run->debug.arch.hsr = lower_32_bits(esr);
151 	run->debug.arch.hsr_high = upper_32_bits(esr);
152 	run->flags = KVM_DEBUG_ARCH_HSR_HIGH_VALID;
153 
154 	if (ESR_ELx_EC(esr) == ESR_ELx_EC_WATCHPT_LOW)
155 		run->debug.arch.far = vcpu->arch.fault.far_el2;
156 
157 	return 0;
158 }
159 
160 static int kvm_handle_unknown_ec(struct kvm_vcpu *vcpu)
161 {
162 	u64 esr = kvm_vcpu_get_esr(vcpu);
163 
164 	kvm_pr_unimpl("Unknown exception class: esr: %#016llx -- %s\n",
165 		      esr, esr_get_class_string(esr));
166 
167 	kvm_inject_undefined(vcpu);
168 	return 1;
169 }
170 
171 /*
172  * Guest access to SVE registers should be routed to this handler only
173  * when the system doesn't support SVE.
174  */
175 static int handle_sve(struct kvm_vcpu *vcpu)
176 {
177 	kvm_inject_undefined(vcpu);
178 	return 1;
179 }
180 
181 /*
182  * Guest usage of a ptrauth instruction (which the guest EL1 did not turn into
183  * a NOP). If we get here, it is that we didn't fixup ptrauth on exit, and all
184  * that we can do is give the guest an UNDEF.
185  */
186 static int kvm_handle_ptrauth(struct kvm_vcpu *vcpu)
187 {
188 	kvm_inject_undefined(vcpu);
189 	return 1;
190 }
191 
192 static exit_handle_fn arm_exit_handlers[] = {
193 	[0 ... ESR_ELx_EC_MAX]	= kvm_handle_unknown_ec,
194 	[ESR_ELx_EC_WFx]	= kvm_handle_wfx,
195 	[ESR_ELx_EC_CP15_32]	= kvm_handle_cp15_32,
196 	[ESR_ELx_EC_CP15_64]	= kvm_handle_cp15_64,
197 	[ESR_ELx_EC_CP14_MR]	= kvm_handle_cp14_32,
198 	[ESR_ELx_EC_CP14_LS]	= kvm_handle_cp14_load_store,
199 	[ESR_ELx_EC_CP10_ID]	= kvm_handle_cp10_id,
200 	[ESR_ELx_EC_CP14_64]	= kvm_handle_cp14_64,
201 	[ESR_ELx_EC_HVC32]	= handle_hvc,
202 	[ESR_ELx_EC_SMC32]	= handle_smc,
203 	[ESR_ELx_EC_HVC64]	= handle_hvc,
204 	[ESR_ELx_EC_SMC64]	= handle_smc,
205 	[ESR_ELx_EC_SYS64]	= kvm_handle_sys_reg,
206 	[ESR_ELx_EC_SVE]	= handle_sve,
207 	[ESR_ELx_EC_IABT_LOW]	= kvm_handle_guest_abort,
208 	[ESR_ELx_EC_DABT_LOW]	= kvm_handle_guest_abort,
209 	[ESR_ELx_EC_SOFTSTP_LOW]= kvm_handle_guest_debug,
210 	[ESR_ELx_EC_WATCHPT_LOW]= kvm_handle_guest_debug,
211 	[ESR_ELx_EC_BREAKPT_LOW]= kvm_handle_guest_debug,
212 	[ESR_ELx_EC_BKPT32]	= kvm_handle_guest_debug,
213 	[ESR_ELx_EC_BRK64]	= kvm_handle_guest_debug,
214 	[ESR_ELx_EC_FP_ASIMD]	= handle_no_fpsimd,
215 	[ESR_ELx_EC_PAC]	= kvm_handle_ptrauth,
216 };
217 
218 static exit_handle_fn kvm_get_exit_handler(struct kvm_vcpu *vcpu)
219 {
220 	u64 esr = kvm_vcpu_get_esr(vcpu);
221 	u8 esr_ec = ESR_ELx_EC(esr);
222 
223 	return arm_exit_handlers[esr_ec];
224 }
225 
226 /*
227  * We may be single-stepping an emulated instruction. If the emulation
228  * has been completed in the kernel, we can return to userspace with a
229  * KVM_EXIT_DEBUG, otherwise userspace needs to complete its
230  * emulation first.
231  */
232 static int handle_trap_exceptions(struct kvm_vcpu *vcpu)
233 {
234 	int handled;
235 
236 	/*
237 	 * See ARM ARM B1.14.1: "Hyp traps on instructions
238 	 * that fail their condition code check"
239 	 */
240 	if (!kvm_condition_valid(vcpu)) {
241 		kvm_incr_pc(vcpu);
242 		handled = 1;
243 	} else {
244 		exit_handle_fn exit_handler;
245 
246 		exit_handler = kvm_get_exit_handler(vcpu);
247 		handled = exit_handler(vcpu);
248 	}
249 
250 	return handled;
251 }
252 
253 /*
254  * Return > 0 to return to guest, < 0 on error, 0 (and set exit_reason) on
255  * proper exit to userspace.
256  */
257 int handle_exit(struct kvm_vcpu *vcpu, int exception_index)
258 {
259 	struct kvm_run *run = vcpu->run;
260 
261 	if (ARM_SERROR_PENDING(exception_index)) {
262 		/*
263 		 * The SError is handled by handle_exit_early(). If the guest
264 		 * survives it will re-execute the original instruction.
265 		 */
266 		return 1;
267 	}
268 
269 	exception_index = ARM_EXCEPTION_CODE(exception_index);
270 
271 	switch (exception_index) {
272 	case ARM_EXCEPTION_IRQ:
273 		return 1;
274 	case ARM_EXCEPTION_EL1_SERROR:
275 		return 1;
276 	case ARM_EXCEPTION_TRAP:
277 		return handle_trap_exceptions(vcpu);
278 	case ARM_EXCEPTION_HYP_GONE:
279 		/*
280 		 * EL2 has been reset to the hyp-stub. This happens when a guest
281 		 * is pre-emptied by kvm_reboot()'s shutdown call.
282 		 */
283 		run->exit_reason = KVM_EXIT_FAIL_ENTRY;
284 		return 0;
285 	case ARM_EXCEPTION_IL:
286 		/*
287 		 * We attempted an illegal exception return.  Guest state must
288 		 * have been corrupted somehow.  Give up.
289 		 */
290 		run->exit_reason = KVM_EXIT_FAIL_ENTRY;
291 		return -EINVAL;
292 	default:
293 		kvm_pr_unimpl("Unsupported exception type: %d",
294 			      exception_index);
295 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
296 		return 0;
297 	}
298 }
299 
300 /* For exit types that need handling before we can be preempted */
301 void handle_exit_early(struct kvm_vcpu *vcpu, int exception_index)
302 {
303 	if (ARM_SERROR_PENDING(exception_index)) {
304 		if (this_cpu_has_cap(ARM64_HAS_RAS_EXTN)) {
305 			u64 disr = kvm_vcpu_get_disr(vcpu);
306 
307 			kvm_handle_guest_serror(vcpu, disr_to_esr(disr));
308 		} else {
309 			kvm_inject_vabt(vcpu);
310 		}
311 
312 		return;
313 	}
314 
315 	exception_index = ARM_EXCEPTION_CODE(exception_index);
316 
317 	if (exception_index == ARM_EXCEPTION_EL1_SERROR)
318 		kvm_handle_guest_serror(vcpu, kvm_vcpu_get_esr(vcpu));
319 }
320 
321 void __noreturn __cold nvhe_hyp_panic_handler(u64 esr, u64 spsr,
322 					      u64 elr_virt, u64 elr_phys,
323 					      u64 par, uintptr_t vcpu,
324 					      u64 far, u64 hpfar) {
325 	u64 elr_in_kimg = __phys_to_kimg(elr_phys);
326 	u64 hyp_offset = elr_in_kimg - kaslr_offset() - elr_virt;
327 	u64 mode = spsr & PSR_MODE_MASK;
328 	u64 panic_addr = elr_virt + hyp_offset;
329 
330 	if (mode != PSR_MODE_EL2t && mode != PSR_MODE_EL2h) {
331 		kvm_err("Invalid host exception to nVHE hyp!\n");
332 	} else if (ESR_ELx_EC(esr) == ESR_ELx_EC_BRK64 &&
333 		   (esr & ESR_ELx_BRK64_ISS_COMMENT_MASK) == BUG_BRK_IMM) {
334 		const char *file = NULL;
335 		unsigned int line = 0;
336 
337 		/* All hyp bugs, including warnings, are treated as fatal. */
338 		if (!is_protected_kvm_enabled() ||
339 		    IS_ENABLED(CONFIG_NVHE_EL2_DEBUG)) {
340 			struct bug_entry *bug = find_bug(elr_in_kimg);
341 
342 			if (bug)
343 				bug_get_file_line(bug, &file, &line);
344 		}
345 
346 		if (file)
347 			kvm_err("nVHE hyp BUG at: %s:%u!\n", file, line);
348 		else
349 			kvm_err("nVHE hyp BUG at: [<%016llx>] %pB!\n", panic_addr,
350 					(void *)panic_addr);
351 	} else {
352 		kvm_err("nVHE hyp panic at: [<%016llx>] %pB!\n", panic_addr,
353 				(void *)panic_addr);
354 	}
355 
356 	/*
357 	 * Hyp has panicked and we're going to handle that by panicking the
358 	 * kernel. The kernel offset will be revealed in the panic so we're
359 	 * also safe to reveal the hyp offset as a debugging aid for translating
360 	 * hyp VAs to vmlinux addresses.
361 	 */
362 	kvm_err("Hyp Offset: 0x%llx\n", hyp_offset);
363 
364 	panic("HYP panic:\nPS:%08llx PC:%016llx ESR:%016llx\nFAR:%016llx HPFAR:%016llx PAR:%016llx\nVCPU:%016lx\n",
365 	      spsr, elr_virt, esr, far, hpfar, par, vcpu);
366 }
367