xref: /openbmc/linux/arch/arm64/kvm/handle_exit.c (revision 8dda2eac)
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, u32 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: Yield the CPU and come back to this vcpu when the scheduler
84  * decides to.
85  * WFI: Simply call kvm_vcpu_block(), 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  */
89 static int kvm_handle_wfx(struct kvm_vcpu *vcpu)
90 {
91 	if (kvm_vcpu_get_esr(vcpu) & ESR_ELx_WFx_ISS_WFE) {
92 		trace_kvm_wfx_arm64(*vcpu_pc(vcpu), true);
93 		vcpu->stat.wfe_exit_stat++;
94 		kvm_vcpu_on_spin(vcpu, vcpu_mode_priv(vcpu));
95 	} else {
96 		trace_kvm_wfx_arm64(*vcpu_pc(vcpu), false);
97 		vcpu->stat.wfi_exit_stat++;
98 		kvm_vcpu_block(vcpu);
99 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
100 	}
101 
102 	kvm_incr_pc(vcpu);
103 
104 	return 1;
105 }
106 
107 /**
108  * kvm_handle_guest_debug - handle a debug exception instruction
109  *
110  * @vcpu:	the vcpu pointer
111  *
112  * We route all debug exceptions through the same handler. If both the
113  * guest and host are using the same debug facilities it will be up to
114  * userspace to re-inject the correct exception for guest delivery.
115  *
116  * @return: 0 (while setting vcpu->run->exit_reason), -1 for error
117  */
118 static int kvm_handle_guest_debug(struct kvm_vcpu *vcpu)
119 {
120 	struct kvm_run *run = vcpu->run;
121 	u32 esr = kvm_vcpu_get_esr(vcpu);
122 	int ret = 0;
123 
124 	run->exit_reason = KVM_EXIT_DEBUG;
125 	run->debug.arch.hsr = esr;
126 
127 	switch (ESR_ELx_EC(esr)) {
128 	case ESR_ELx_EC_WATCHPT_LOW:
129 		run->debug.arch.far = vcpu->arch.fault.far_el2;
130 		fallthrough;
131 	case ESR_ELx_EC_SOFTSTP_LOW:
132 	case ESR_ELx_EC_BREAKPT_LOW:
133 	case ESR_ELx_EC_BKPT32:
134 	case ESR_ELx_EC_BRK64:
135 		break;
136 	default:
137 		kvm_err("%s: un-handled case esr: %#08x\n",
138 			__func__, (unsigned int) esr);
139 		ret = -1;
140 		break;
141 	}
142 
143 	return ret;
144 }
145 
146 static int kvm_handle_unknown_ec(struct kvm_vcpu *vcpu)
147 {
148 	u32 esr = kvm_vcpu_get_esr(vcpu);
149 
150 	kvm_pr_unimpl("Unknown exception class: esr: %#08x -- %s\n",
151 		      esr, esr_get_class_string(esr));
152 
153 	kvm_inject_undefined(vcpu);
154 	return 1;
155 }
156 
157 static int handle_sve(struct kvm_vcpu *vcpu)
158 {
159 	/* Until SVE is supported for guests: */
160 	kvm_inject_undefined(vcpu);
161 	return 1;
162 }
163 
164 /*
165  * Guest usage of a ptrauth instruction (which the guest EL1 did not turn into
166  * a NOP). If we get here, it is that we didn't fixup ptrauth on exit, and all
167  * that we can do is give the guest an UNDEF.
168  */
169 static int kvm_handle_ptrauth(struct kvm_vcpu *vcpu)
170 {
171 	kvm_inject_undefined(vcpu);
172 	return 1;
173 }
174 
175 static exit_handle_fn arm_exit_handlers[] = {
176 	[0 ... ESR_ELx_EC_MAX]	= kvm_handle_unknown_ec,
177 	[ESR_ELx_EC_WFx]	= kvm_handle_wfx,
178 	[ESR_ELx_EC_CP15_32]	= kvm_handle_cp15_32,
179 	[ESR_ELx_EC_CP15_64]	= kvm_handle_cp15_64,
180 	[ESR_ELx_EC_CP14_MR]	= kvm_handle_cp14_32,
181 	[ESR_ELx_EC_CP14_LS]	= kvm_handle_cp14_load_store,
182 	[ESR_ELx_EC_CP14_64]	= kvm_handle_cp14_64,
183 	[ESR_ELx_EC_HVC32]	= handle_hvc,
184 	[ESR_ELx_EC_SMC32]	= handle_smc,
185 	[ESR_ELx_EC_HVC64]	= handle_hvc,
186 	[ESR_ELx_EC_SMC64]	= handle_smc,
187 	[ESR_ELx_EC_SYS64]	= kvm_handle_sys_reg,
188 	[ESR_ELx_EC_SVE]	= handle_sve,
189 	[ESR_ELx_EC_IABT_LOW]	= kvm_handle_guest_abort,
190 	[ESR_ELx_EC_DABT_LOW]	= kvm_handle_guest_abort,
191 	[ESR_ELx_EC_SOFTSTP_LOW]= kvm_handle_guest_debug,
192 	[ESR_ELx_EC_WATCHPT_LOW]= kvm_handle_guest_debug,
193 	[ESR_ELx_EC_BREAKPT_LOW]= kvm_handle_guest_debug,
194 	[ESR_ELx_EC_BKPT32]	= kvm_handle_guest_debug,
195 	[ESR_ELx_EC_BRK64]	= kvm_handle_guest_debug,
196 	[ESR_ELx_EC_FP_ASIMD]	= handle_no_fpsimd,
197 	[ESR_ELx_EC_PAC]	= kvm_handle_ptrauth,
198 };
199 
200 static exit_handle_fn kvm_get_exit_handler(struct kvm_vcpu *vcpu)
201 {
202 	u32 esr = kvm_vcpu_get_esr(vcpu);
203 	u8 esr_ec = ESR_ELx_EC(esr);
204 
205 	return arm_exit_handlers[esr_ec];
206 }
207 
208 /*
209  * We may be single-stepping an emulated instruction. If the emulation
210  * has been completed in the kernel, we can return to userspace with a
211  * KVM_EXIT_DEBUG, otherwise userspace needs to complete its
212  * emulation first.
213  */
214 static int handle_trap_exceptions(struct kvm_vcpu *vcpu)
215 {
216 	int handled;
217 
218 	/*
219 	 * See ARM ARM B1.14.1: "Hyp traps on instructions
220 	 * that fail their condition code check"
221 	 */
222 	if (!kvm_condition_valid(vcpu)) {
223 		kvm_incr_pc(vcpu);
224 		handled = 1;
225 	} else {
226 		exit_handle_fn exit_handler;
227 
228 		exit_handler = kvm_get_exit_handler(vcpu);
229 		handled = exit_handler(vcpu);
230 	}
231 
232 	return handled;
233 }
234 
235 /*
236  * Return > 0 to return to guest, < 0 on error, 0 (and set exit_reason) on
237  * proper exit to userspace.
238  */
239 int handle_exit(struct kvm_vcpu *vcpu, int exception_index)
240 {
241 	struct kvm_run *run = vcpu->run;
242 
243 	exception_index = ARM_EXCEPTION_CODE(exception_index);
244 
245 	switch (exception_index) {
246 	case ARM_EXCEPTION_IRQ:
247 		return 1;
248 	case ARM_EXCEPTION_EL1_SERROR:
249 		return 1;
250 	case ARM_EXCEPTION_TRAP:
251 		return handle_trap_exceptions(vcpu);
252 	case ARM_EXCEPTION_HYP_GONE:
253 		/*
254 		 * EL2 has been reset to the hyp-stub. This happens when a guest
255 		 * is pre-empted by kvm_reboot()'s shutdown call.
256 		 */
257 		run->exit_reason = KVM_EXIT_FAIL_ENTRY;
258 		return 0;
259 	case ARM_EXCEPTION_IL:
260 		/*
261 		 * We attempted an illegal exception return.  Guest state must
262 		 * have been corrupted somehow.  Give up.
263 		 */
264 		run->exit_reason = KVM_EXIT_FAIL_ENTRY;
265 		return -EINVAL;
266 	default:
267 		kvm_pr_unimpl("Unsupported exception type: %d",
268 			      exception_index);
269 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
270 		return 0;
271 	}
272 }
273 
274 /* For exit types that need handling before we can be preempted */
275 void handle_exit_early(struct kvm_vcpu *vcpu, int exception_index)
276 {
277 	if (ARM_SERROR_PENDING(exception_index)) {
278 		if (this_cpu_has_cap(ARM64_HAS_RAS_EXTN)) {
279 			u64 disr = kvm_vcpu_get_disr(vcpu);
280 
281 			kvm_handle_guest_serror(vcpu, disr_to_esr(disr));
282 		} else {
283 			kvm_inject_vabt(vcpu);
284 		}
285 
286 		return;
287 	}
288 
289 	exception_index = ARM_EXCEPTION_CODE(exception_index);
290 
291 	if (exception_index == ARM_EXCEPTION_EL1_SERROR)
292 		kvm_handle_guest_serror(vcpu, kvm_vcpu_get_esr(vcpu));
293 }
294 
295 void __noreturn __cold nvhe_hyp_panic_handler(u64 esr, u64 spsr, u64 elr,
296 					      u64 par, uintptr_t vcpu,
297 					      u64 far, u64 hpfar) {
298 	u64 elr_in_kimg = __phys_to_kimg(__hyp_pa(elr));
299 	u64 hyp_offset = elr_in_kimg - kaslr_offset() - elr;
300 	u64 mode = spsr & PSR_MODE_MASK;
301 
302 	/*
303 	 * The nVHE hyp symbols are not included by kallsyms to avoid issues
304 	 * with aliasing. That means that the symbols cannot be printed with the
305 	 * "%pS" format specifier, so fall back to the vmlinux address if
306 	 * there's no better option.
307 	 */
308 	if (mode != PSR_MODE_EL2t && mode != PSR_MODE_EL2h) {
309 		kvm_err("Invalid host exception to nVHE hyp!\n");
310 	} else if (ESR_ELx_EC(esr) == ESR_ELx_EC_BRK64 &&
311 		   (esr & ESR_ELx_BRK64_ISS_COMMENT_MASK) == BUG_BRK_IMM) {
312 		struct bug_entry *bug = find_bug(elr_in_kimg);
313 		const char *file = NULL;
314 		unsigned int line = 0;
315 
316 		/* All hyp bugs, including warnings, are treated as fatal. */
317 		if (bug)
318 			bug_get_file_line(bug, &file, &line);
319 
320 		if (file)
321 			kvm_err("nVHE hyp BUG at: %s:%u!\n", file, line);
322 		else
323 			kvm_err("nVHE hyp BUG at: %016llx!\n", elr + hyp_offset);
324 	} else {
325 		kvm_err("nVHE hyp panic at: %016llx!\n", elr + hyp_offset);
326 	}
327 
328 	/*
329 	 * Hyp has panicked and we're going to handle that by panicking the
330 	 * kernel. The kernel offset will be revealed in the panic so we're
331 	 * also safe to reveal the hyp offset as a debugging aid for translating
332 	 * hyp VAs to vmlinux addresses.
333 	 */
334 	kvm_err("Hyp Offset: 0x%llx\n", hyp_offset);
335 
336 	panic("HYP panic:\nPS:%08llx PC:%016llx ESR:%08llx\nFAR:%016llx HPFAR:%016llx PAR:%016llx\nVCPU:%016lx\n",
337 	      spsr, elr, esr, far, hpfar, par, vcpu);
338 }
339