xref: /openbmc/linux/arch/arm64/kvm/handle_exit.c (revision dfc66bef)
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)
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 
123 	run->exit_reason = KVM_EXIT_DEBUG;
124 	run->debug.arch.hsr = esr;
125 
126 	if (ESR_ELx_EC(esr) == ESR_ELx_EC_WATCHPT_LOW)
127 		run->debug.arch.far = vcpu->arch.fault.far_el2;
128 
129 	return 0;
130 }
131 
132 static int kvm_handle_unknown_ec(struct kvm_vcpu *vcpu)
133 {
134 	u32 esr = kvm_vcpu_get_esr(vcpu);
135 
136 	kvm_pr_unimpl("Unknown exception class: esr: %#08x -- %s\n",
137 		      esr, esr_get_class_string(esr));
138 
139 	kvm_inject_undefined(vcpu);
140 	return 1;
141 }
142 
143 /*
144  * Guest access to SVE registers should be routed to this handler only
145  * when the system doesn't support SVE.
146  */
147 static int handle_sve(struct kvm_vcpu *vcpu)
148 {
149 	kvm_inject_undefined(vcpu);
150 	return 1;
151 }
152 
153 /*
154  * Guest usage of a ptrauth instruction (which the guest EL1 did not turn into
155  * a NOP). If we get here, it is that we didn't fixup ptrauth on exit, and all
156  * that we can do is give the guest an UNDEF.
157  */
158 static int kvm_handle_ptrauth(struct kvm_vcpu *vcpu)
159 {
160 	kvm_inject_undefined(vcpu);
161 	return 1;
162 }
163 
164 static exit_handle_fn arm_exit_handlers[] = {
165 	[0 ... ESR_ELx_EC_MAX]	= kvm_handle_unknown_ec,
166 	[ESR_ELx_EC_WFx]	= kvm_handle_wfx,
167 	[ESR_ELx_EC_CP15_32]	= kvm_handle_cp15_32,
168 	[ESR_ELx_EC_CP15_64]	= kvm_handle_cp15_64,
169 	[ESR_ELx_EC_CP14_MR]	= kvm_handle_cp14_32,
170 	[ESR_ELx_EC_CP14_LS]	= kvm_handle_cp14_load_store,
171 	[ESR_ELx_EC_CP14_64]	= kvm_handle_cp14_64,
172 	[ESR_ELx_EC_HVC32]	= handle_hvc,
173 	[ESR_ELx_EC_SMC32]	= handle_smc,
174 	[ESR_ELx_EC_HVC64]	= handle_hvc,
175 	[ESR_ELx_EC_SMC64]	= handle_smc,
176 	[ESR_ELx_EC_SYS64]	= kvm_handle_sys_reg,
177 	[ESR_ELx_EC_SVE]	= handle_sve,
178 	[ESR_ELx_EC_IABT_LOW]	= kvm_handle_guest_abort,
179 	[ESR_ELx_EC_DABT_LOW]	= kvm_handle_guest_abort,
180 	[ESR_ELx_EC_SOFTSTP_LOW]= kvm_handle_guest_debug,
181 	[ESR_ELx_EC_WATCHPT_LOW]= kvm_handle_guest_debug,
182 	[ESR_ELx_EC_BREAKPT_LOW]= kvm_handle_guest_debug,
183 	[ESR_ELx_EC_BKPT32]	= kvm_handle_guest_debug,
184 	[ESR_ELx_EC_BRK64]	= kvm_handle_guest_debug,
185 	[ESR_ELx_EC_FP_ASIMD]	= handle_no_fpsimd,
186 	[ESR_ELx_EC_PAC]	= kvm_handle_ptrauth,
187 };
188 
189 static exit_handle_fn kvm_get_exit_handler(struct kvm_vcpu *vcpu)
190 {
191 	u32 esr = kvm_vcpu_get_esr(vcpu);
192 	u8 esr_ec = ESR_ELx_EC(esr);
193 
194 	return arm_exit_handlers[esr_ec];
195 }
196 
197 /*
198  * We may be single-stepping an emulated instruction. If the emulation
199  * has been completed in the kernel, we can return to userspace with a
200  * KVM_EXIT_DEBUG, otherwise userspace needs to complete its
201  * emulation first.
202  */
203 static int handle_trap_exceptions(struct kvm_vcpu *vcpu)
204 {
205 	int handled;
206 
207 	/*
208 	 * See ARM ARM B1.14.1: "Hyp traps on instructions
209 	 * that fail their condition code check"
210 	 */
211 	if (!kvm_condition_valid(vcpu)) {
212 		kvm_incr_pc(vcpu);
213 		handled = 1;
214 	} else {
215 		exit_handle_fn exit_handler;
216 
217 		exit_handler = kvm_get_exit_handler(vcpu);
218 		handled = exit_handler(vcpu);
219 	}
220 
221 	return handled;
222 }
223 
224 /*
225  * Return > 0 to return to guest, < 0 on error, 0 (and set exit_reason) on
226  * proper exit to userspace.
227  */
228 int handle_exit(struct kvm_vcpu *vcpu, int exception_index)
229 {
230 	struct kvm_run *run = vcpu->run;
231 
232 	exception_index = ARM_EXCEPTION_CODE(exception_index);
233 
234 	switch (exception_index) {
235 	case ARM_EXCEPTION_IRQ:
236 		return 1;
237 	case ARM_EXCEPTION_EL1_SERROR:
238 		return 1;
239 	case ARM_EXCEPTION_TRAP:
240 		return handle_trap_exceptions(vcpu);
241 	case ARM_EXCEPTION_HYP_GONE:
242 		/*
243 		 * EL2 has been reset to the hyp-stub. This happens when a guest
244 		 * is pre-empted by kvm_reboot()'s shutdown call.
245 		 */
246 		run->exit_reason = KVM_EXIT_FAIL_ENTRY;
247 		return 0;
248 	case ARM_EXCEPTION_IL:
249 		/*
250 		 * We attempted an illegal exception return.  Guest state must
251 		 * have been corrupted somehow.  Give up.
252 		 */
253 		run->exit_reason = KVM_EXIT_FAIL_ENTRY;
254 		return -EINVAL;
255 	default:
256 		kvm_pr_unimpl("Unsupported exception type: %d",
257 			      exception_index);
258 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
259 		return 0;
260 	}
261 }
262 
263 /* For exit types that need handling before we can be preempted */
264 void handle_exit_early(struct kvm_vcpu *vcpu, int exception_index)
265 {
266 	if (ARM_SERROR_PENDING(exception_index)) {
267 		if (this_cpu_has_cap(ARM64_HAS_RAS_EXTN)) {
268 			u64 disr = kvm_vcpu_get_disr(vcpu);
269 
270 			kvm_handle_guest_serror(vcpu, disr_to_esr(disr));
271 		} else {
272 			kvm_inject_vabt(vcpu);
273 		}
274 
275 		return;
276 	}
277 
278 	exception_index = ARM_EXCEPTION_CODE(exception_index);
279 
280 	if (exception_index == ARM_EXCEPTION_EL1_SERROR)
281 		kvm_handle_guest_serror(vcpu, kvm_vcpu_get_esr(vcpu));
282 }
283 
284 void __noreturn __cold nvhe_hyp_panic_handler(u64 esr, u64 spsr,
285 					      u64 elr_virt, u64 elr_phys,
286 					      u64 par, uintptr_t vcpu,
287 					      u64 far, u64 hpfar) {
288 	u64 elr_in_kimg = __phys_to_kimg(elr_phys);
289 	u64 hyp_offset = elr_in_kimg - kaslr_offset() - elr_virt;
290 	u64 mode = spsr & PSR_MODE_MASK;
291 
292 	/*
293 	 * The nVHE hyp symbols are not included by kallsyms to avoid issues
294 	 * with aliasing. That means that the symbols cannot be printed with the
295 	 * "%pS" format specifier, so fall back to the vmlinux address if
296 	 * there's no better option.
297 	 */
298 	if (mode != PSR_MODE_EL2t && mode != PSR_MODE_EL2h) {
299 		kvm_err("Invalid host exception to nVHE hyp!\n");
300 	} else if (ESR_ELx_EC(esr) == ESR_ELx_EC_BRK64 &&
301 		   (esr & ESR_ELx_BRK64_ISS_COMMENT_MASK) == BUG_BRK_IMM) {
302 		const char *file = NULL;
303 		unsigned int line = 0;
304 
305 		/* All hyp bugs, including warnings, are treated as fatal. */
306 		if (!is_protected_kvm_enabled() ||
307 		    IS_ENABLED(CONFIG_NVHE_EL2_DEBUG)) {
308 			struct bug_entry *bug = find_bug(elr_in_kimg);
309 
310 			if (bug)
311 				bug_get_file_line(bug, &file, &line);
312 		}
313 
314 		if (file)
315 			kvm_err("nVHE hyp BUG at: %s:%u!\n", file, line);
316 		else
317 			kvm_err("nVHE hyp BUG at: %016llx!\n", elr_virt + hyp_offset);
318 	} else {
319 		kvm_err("nVHE hyp panic at: %016llx!\n", elr_virt + hyp_offset);
320 	}
321 
322 	/*
323 	 * Hyp has panicked and we're going to handle that by panicking the
324 	 * kernel. The kernel offset will be revealed in the panic so we're
325 	 * also safe to reveal the hyp offset as a debugging aid for translating
326 	 * hyp VAs to vmlinux addresses.
327 	 */
328 	kvm_err("Hyp Offset: 0x%llx\n", hyp_offset);
329 
330 	panic("HYP panic:\nPS:%08llx PC:%016llx ESR:%08llx\nFAR:%016llx HPFAR:%016llx PAR:%016llx\nVCPU:%016lx\n",
331 	      spsr, elr_virt, esr, far, hpfar, par, vcpu);
332 }
333