xref: /openbmc/linux/arch/arm64/kvm/psci.c (revision 63705da3)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2012 - ARM Ltd
4  * Author: Marc Zyngier <marc.zyngier@arm.com>
5  */
6 
7 #include <linux/arm-smccc.h>
8 #include <linux/preempt.h>
9 #include <linux/kvm_host.h>
10 #include <linux/uaccess.h>
11 #include <linux/wait.h>
12 
13 #include <asm/cputype.h>
14 #include <asm/kvm_emulate.h>
15 
16 #include <kvm/arm_psci.h>
17 #include <kvm/arm_hypercalls.h>
18 
19 /*
20  * This is an implementation of the Power State Coordination Interface
21  * as described in ARM document number ARM DEN 0022A.
22  */
23 
24 #define AFFINITY_MASK(level)	~((0x1UL << ((level) * MPIDR_LEVEL_BITS)) - 1)
25 
26 static unsigned long psci_affinity_mask(unsigned long affinity_level)
27 {
28 	if (affinity_level <= 3)
29 		return MPIDR_HWID_BITMASK & AFFINITY_MASK(affinity_level);
30 
31 	return 0;
32 }
33 
34 static unsigned long kvm_psci_vcpu_suspend(struct kvm_vcpu *vcpu)
35 {
36 	/*
37 	 * NOTE: For simplicity, we make VCPU suspend emulation to be
38 	 * same-as WFI (Wait-for-interrupt) emulation.
39 	 *
40 	 * This means for KVM the wakeup events are interrupts and
41 	 * this is consistent with intended use of StateID as described
42 	 * in section 5.4.1 of PSCI v0.2 specification (ARM DEN 0022A).
43 	 *
44 	 * Further, we also treat power-down request to be same as
45 	 * stand-by request as-per section 5.4.2 clause 3 of PSCI v0.2
46 	 * specification (ARM DEN 0022A). This means all suspend states
47 	 * for KVM will preserve the register state.
48 	 */
49 	kvm_vcpu_block(vcpu);
50 	kvm_clear_request(KVM_REQ_UNHALT, vcpu);
51 
52 	return PSCI_RET_SUCCESS;
53 }
54 
55 static void kvm_psci_vcpu_off(struct kvm_vcpu *vcpu)
56 {
57 	vcpu->arch.power_off = true;
58 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
59 	kvm_vcpu_kick(vcpu);
60 }
61 
62 static inline bool kvm_psci_valid_affinity(struct kvm_vcpu *vcpu,
63 					   unsigned long affinity)
64 {
65 	return !(affinity & ~MPIDR_HWID_BITMASK);
66 }
67 
68 static unsigned long kvm_psci_vcpu_on(struct kvm_vcpu *source_vcpu)
69 {
70 	struct vcpu_reset_state *reset_state;
71 	struct kvm *kvm = source_vcpu->kvm;
72 	struct kvm_vcpu *vcpu = NULL;
73 	unsigned long cpu_id;
74 
75 	cpu_id = smccc_get_arg1(source_vcpu);
76 	if (!kvm_psci_valid_affinity(source_vcpu, cpu_id))
77 		return PSCI_RET_INVALID_PARAMS;
78 
79 	vcpu = kvm_mpidr_to_vcpu(kvm, cpu_id);
80 
81 	/*
82 	 * Make sure the caller requested a valid CPU and that the CPU is
83 	 * turned off.
84 	 */
85 	if (!vcpu)
86 		return PSCI_RET_INVALID_PARAMS;
87 	if (!vcpu->arch.power_off) {
88 		if (kvm_psci_version(source_vcpu, kvm) != KVM_ARM_PSCI_0_1)
89 			return PSCI_RET_ALREADY_ON;
90 		else
91 			return PSCI_RET_INVALID_PARAMS;
92 	}
93 
94 	reset_state = &vcpu->arch.reset_state;
95 
96 	reset_state->pc = smccc_get_arg2(source_vcpu);
97 
98 	/* Propagate caller endianness */
99 	reset_state->be = kvm_vcpu_is_be(source_vcpu);
100 
101 	/*
102 	 * NOTE: We always update r0 (or x0) because for PSCI v0.1
103 	 * the general purpose registers are undefined upon CPU_ON.
104 	 */
105 	reset_state->r0 = smccc_get_arg3(source_vcpu);
106 
107 	WRITE_ONCE(reset_state->reset, true);
108 	kvm_make_request(KVM_REQ_VCPU_RESET, vcpu);
109 
110 	/*
111 	 * Make sure the reset request is observed if the change to
112 	 * power_state is observed.
113 	 */
114 	smp_wmb();
115 
116 	vcpu->arch.power_off = false;
117 	kvm_vcpu_wake_up(vcpu);
118 
119 	return PSCI_RET_SUCCESS;
120 }
121 
122 static unsigned long kvm_psci_vcpu_affinity_info(struct kvm_vcpu *vcpu)
123 {
124 	int i, matching_cpus = 0;
125 	unsigned long mpidr;
126 	unsigned long target_affinity;
127 	unsigned long target_affinity_mask;
128 	unsigned long lowest_affinity_level;
129 	struct kvm *kvm = vcpu->kvm;
130 	struct kvm_vcpu *tmp;
131 
132 	target_affinity = smccc_get_arg1(vcpu);
133 	lowest_affinity_level = smccc_get_arg2(vcpu);
134 
135 	if (!kvm_psci_valid_affinity(vcpu, target_affinity))
136 		return PSCI_RET_INVALID_PARAMS;
137 
138 	/* Determine target affinity mask */
139 	target_affinity_mask = psci_affinity_mask(lowest_affinity_level);
140 	if (!target_affinity_mask)
141 		return PSCI_RET_INVALID_PARAMS;
142 
143 	/* Ignore other bits of target affinity */
144 	target_affinity &= target_affinity_mask;
145 
146 	/*
147 	 * If one or more VCPU matching target affinity are running
148 	 * then ON else OFF
149 	 */
150 	kvm_for_each_vcpu(i, tmp, kvm) {
151 		mpidr = kvm_vcpu_get_mpidr_aff(tmp);
152 		if ((mpidr & target_affinity_mask) == target_affinity) {
153 			matching_cpus++;
154 			if (!tmp->arch.power_off)
155 				return PSCI_0_2_AFFINITY_LEVEL_ON;
156 		}
157 	}
158 
159 	if (!matching_cpus)
160 		return PSCI_RET_INVALID_PARAMS;
161 
162 	return PSCI_0_2_AFFINITY_LEVEL_OFF;
163 }
164 
165 static void kvm_prepare_system_event(struct kvm_vcpu *vcpu, u32 type)
166 {
167 	int i;
168 	struct kvm_vcpu *tmp;
169 
170 	/*
171 	 * The KVM ABI specifies that a system event exit may call KVM_RUN
172 	 * again and may perform shutdown/reboot at a later time that when the
173 	 * actual request is made.  Since we are implementing PSCI and a
174 	 * caller of PSCI reboot and shutdown expects that the system shuts
175 	 * down or reboots immediately, let's make sure that VCPUs are not run
176 	 * after this call is handled and before the VCPUs have been
177 	 * re-initialized.
178 	 */
179 	kvm_for_each_vcpu(i, tmp, vcpu->kvm)
180 		tmp->arch.power_off = true;
181 	kvm_make_all_cpus_request(vcpu->kvm, KVM_REQ_SLEEP);
182 
183 	memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
184 	vcpu->run->system_event.type = type;
185 	vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
186 }
187 
188 static void kvm_psci_system_off(struct kvm_vcpu *vcpu)
189 {
190 	kvm_prepare_system_event(vcpu, KVM_SYSTEM_EVENT_SHUTDOWN);
191 }
192 
193 static void kvm_psci_system_reset(struct kvm_vcpu *vcpu)
194 {
195 	kvm_prepare_system_event(vcpu, KVM_SYSTEM_EVENT_RESET);
196 }
197 
198 static void kvm_psci_narrow_to_32bit(struct kvm_vcpu *vcpu)
199 {
200 	int i;
201 
202 	/*
203 	 * Zero the input registers' upper 32 bits. They will be fully
204 	 * zeroed on exit, so we're fine changing them in place.
205 	 */
206 	for (i = 1; i < 4; i++)
207 		vcpu_set_reg(vcpu, i, lower_32_bits(vcpu_get_reg(vcpu, i)));
208 }
209 
210 static unsigned long kvm_psci_check_allowed_function(struct kvm_vcpu *vcpu, u32 fn)
211 {
212 	switch(fn) {
213 	case PSCI_0_2_FN64_CPU_SUSPEND:
214 	case PSCI_0_2_FN64_CPU_ON:
215 	case PSCI_0_2_FN64_AFFINITY_INFO:
216 		/* Disallow these functions for 32bit guests */
217 		if (vcpu_mode_is_32bit(vcpu))
218 			return PSCI_RET_NOT_SUPPORTED;
219 		break;
220 	}
221 
222 	return 0;
223 }
224 
225 static int kvm_psci_0_2_call(struct kvm_vcpu *vcpu)
226 {
227 	struct kvm *kvm = vcpu->kvm;
228 	u32 psci_fn = smccc_get_function(vcpu);
229 	unsigned long val;
230 	int ret = 1;
231 
232 	val = kvm_psci_check_allowed_function(vcpu, psci_fn);
233 	if (val)
234 		goto out;
235 
236 	switch (psci_fn) {
237 	case PSCI_0_2_FN_PSCI_VERSION:
238 		/*
239 		 * Bits[31:16] = Major Version = 0
240 		 * Bits[15:0] = Minor Version = 2
241 		 */
242 		val = KVM_ARM_PSCI_0_2;
243 		break;
244 	case PSCI_0_2_FN_CPU_SUSPEND:
245 	case PSCI_0_2_FN64_CPU_SUSPEND:
246 		val = kvm_psci_vcpu_suspend(vcpu);
247 		break;
248 	case PSCI_0_2_FN_CPU_OFF:
249 		kvm_psci_vcpu_off(vcpu);
250 		val = PSCI_RET_SUCCESS;
251 		break;
252 	case PSCI_0_2_FN_CPU_ON:
253 		kvm_psci_narrow_to_32bit(vcpu);
254 		fallthrough;
255 	case PSCI_0_2_FN64_CPU_ON:
256 		mutex_lock(&kvm->lock);
257 		val = kvm_psci_vcpu_on(vcpu);
258 		mutex_unlock(&kvm->lock);
259 		break;
260 	case PSCI_0_2_FN_AFFINITY_INFO:
261 		kvm_psci_narrow_to_32bit(vcpu);
262 		fallthrough;
263 	case PSCI_0_2_FN64_AFFINITY_INFO:
264 		val = kvm_psci_vcpu_affinity_info(vcpu);
265 		break;
266 	case PSCI_0_2_FN_MIGRATE_INFO_TYPE:
267 		/*
268 		 * Trusted OS is MP hence does not require migration
269 	         * or
270 		 * Trusted OS is not present
271 		 */
272 		val = PSCI_0_2_TOS_MP;
273 		break;
274 	case PSCI_0_2_FN_SYSTEM_OFF:
275 		kvm_psci_system_off(vcpu);
276 		/*
277 		 * We shouldn't be going back to guest VCPU after
278 		 * receiving SYSTEM_OFF request.
279 		 *
280 		 * If user space accidentally/deliberately resumes
281 		 * guest VCPU after SYSTEM_OFF request then guest
282 		 * VCPU should see internal failure from PSCI return
283 		 * value. To achieve this, we preload r0 (or x0) with
284 		 * PSCI return value INTERNAL_FAILURE.
285 		 */
286 		val = PSCI_RET_INTERNAL_FAILURE;
287 		ret = 0;
288 		break;
289 	case PSCI_0_2_FN_SYSTEM_RESET:
290 		kvm_psci_system_reset(vcpu);
291 		/*
292 		 * Same reason as SYSTEM_OFF for preloading r0 (or x0)
293 		 * with PSCI return value INTERNAL_FAILURE.
294 		 */
295 		val = PSCI_RET_INTERNAL_FAILURE;
296 		ret = 0;
297 		break;
298 	default:
299 		val = PSCI_RET_NOT_SUPPORTED;
300 		break;
301 	}
302 
303 out:
304 	smccc_set_retval(vcpu, val, 0, 0, 0);
305 	return ret;
306 }
307 
308 static int kvm_psci_1_0_call(struct kvm_vcpu *vcpu)
309 {
310 	u32 psci_fn = smccc_get_function(vcpu);
311 	u32 feature;
312 	unsigned long val;
313 	int ret = 1;
314 
315 	switch(psci_fn) {
316 	case PSCI_0_2_FN_PSCI_VERSION:
317 		val = KVM_ARM_PSCI_1_0;
318 		break;
319 	case PSCI_1_0_FN_PSCI_FEATURES:
320 		feature = smccc_get_arg1(vcpu);
321 		val = kvm_psci_check_allowed_function(vcpu, feature);
322 		if (val)
323 			break;
324 
325 		switch(feature) {
326 		case PSCI_0_2_FN_PSCI_VERSION:
327 		case PSCI_0_2_FN_CPU_SUSPEND:
328 		case PSCI_0_2_FN64_CPU_SUSPEND:
329 		case PSCI_0_2_FN_CPU_OFF:
330 		case PSCI_0_2_FN_CPU_ON:
331 		case PSCI_0_2_FN64_CPU_ON:
332 		case PSCI_0_2_FN_AFFINITY_INFO:
333 		case PSCI_0_2_FN64_AFFINITY_INFO:
334 		case PSCI_0_2_FN_MIGRATE_INFO_TYPE:
335 		case PSCI_0_2_FN_SYSTEM_OFF:
336 		case PSCI_0_2_FN_SYSTEM_RESET:
337 		case PSCI_1_0_FN_PSCI_FEATURES:
338 		case ARM_SMCCC_VERSION_FUNC_ID:
339 			val = 0;
340 			break;
341 		default:
342 			val = PSCI_RET_NOT_SUPPORTED;
343 			break;
344 		}
345 		break;
346 	default:
347 		return kvm_psci_0_2_call(vcpu);
348 	}
349 
350 	smccc_set_retval(vcpu, val, 0, 0, 0);
351 	return ret;
352 }
353 
354 static int kvm_psci_0_1_call(struct kvm_vcpu *vcpu)
355 {
356 	struct kvm *kvm = vcpu->kvm;
357 	u32 psci_fn = smccc_get_function(vcpu);
358 	unsigned long val;
359 
360 	switch (psci_fn) {
361 	case KVM_PSCI_FN_CPU_OFF:
362 		kvm_psci_vcpu_off(vcpu);
363 		val = PSCI_RET_SUCCESS;
364 		break;
365 	case KVM_PSCI_FN_CPU_ON:
366 		mutex_lock(&kvm->lock);
367 		val = kvm_psci_vcpu_on(vcpu);
368 		mutex_unlock(&kvm->lock);
369 		break;
370 	default:
371 		val = PSCI_RET_NOT_SUPPORTED;
372 		break;
373 	}
374 
375 	smccc_set_retval(vcpu, val, 0, 0, 0);
376 	return 1;
377 }
378 
379 /**
380  * kvm_psci_call - handle PSCI call if r0 value is in range
381  * @vcpu: Pointer to the VCPU struct
382  *
383  * Handle PSCI calls from guests through traps from HVC instructions.
384  * The calling convention is similar to SMC calls to the secure world
385  * where the function number is placed in r0.
386  *
387  * This function returns: > 0 (success), 0 (success but exit to user
388  * space), and < 0 (errors)
389  *
390  * Errors:
391  * -EINVAL: Unrecognized PSCI function
392  */
393 int kvm_psci_call(struct kvm_vcpu *vcpu)
394 {
395 	switch (kvm_psci_version(vcpu, vcpu->kvm)) {
396 	case KVM_ARM_PSCI_1_0:
397 		return kvm_psci_1_0_call(vcpu);
398 	case KVM_ARM_PSCI_0_2:
399 		return kvm_psci_0_2_call(vcpu);
400 	case KVM_ARM_PSCI_0_1:
401 		return kvm_psci_0_1_call(vcpu);
402 	default:
403 		return -EINVAL;
404 	};
405 }
406 
407 int kvm_arm_get_fw_num_regs(struct kvm_vcpu *vcpu)
408 {
409 	return 3;		/* PSCI version and two workaround registers */
410 }
411 
412 int kvm_arm_copy_fw_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
413 {
414 	if (put_user(KVM_REG_ARM_PSCI_VERSION, uindices++))
415 		return -EFAULT;
416 
417 	if (put_user(KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1, uindices++))
418 		return -EFAULT;
419 
420 	if (put_user(KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2, uindices++))
421 		return -EFAULT;
422 
423 	return 0;
424 }
425 
426 #define KVM_REG_FEATURE_LEVEL_WIDTH	4
427 #define KVM_REG_FEATURE_LEVEL_MASK	(BIT(KVM_REG_FEATURE_LEVEL_WIDTH) - 1)
428 
429 /*
430  * Convert the workaround level into an easy-to-compare number, where higher
431  * values mean better protection.
432  */
433 static int get_kernel_wa_level(u64 regid)
434 {
435 	switch (regid) {
436 	case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1:
437 		switch (arm64_get_spectre_v2_state()) {
438 		case SPECTRE_VULNERABLE:
439 			return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1_NOT_AVAIL;
440 		case SPECTRE_MITIGATED:
441 			return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1_AVAIL;
442 		case SPECTRE_UNAFFECTED:
443 			return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1_NOT_REQUIRED;
444 		}
445 		return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1_NOT_AVAIL;
446 	case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2:
447 		switch (arm64_get_spectre_v4_state()) {
448 		case SPECTRE_MITIGATED:
449 			/*
450 			 * As for the hypercall discovery, we pretend we
451 			 * don't have any FW mitigation if SSBS is there at
452 			 * all times.
453 			 */
454 			if (cpus_have_final_cap(ARM64_SSBS))
455 				return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_AVAIL;
456 			fallthrough;
457 		case SPECTRE_UNAFFECTED:
458 			return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_REQUIRED;
459 		case SPECTRE_VULNERABLE:
460 			return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_AVAIL;
461 		}
462 	}
463 
464 	return -EINVAL;
465 }
466 
467 int kvm_arm_get_fw_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
468 {
469 	void __user *uaddr = (void __user *)(long)reg->addr;
470 	u64 val;
471 
472 	switch (reg->id) {
473 	case KVM_REG_ARM_PSCI_VERSION:
474 		val = kvm_psci_version(vcpu, vcpu->kvm);
475 		break;
476 	case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1:
477 	case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2:
478 		val = get_kernel_wa_level(reg->id) & KVM_REG_FEATURE_LEVEL_MASK;
479 		break;
480 	default:
481 		return -ENOENT;
482 	}
483 
484 	if (copy_to_user(uaddr, &val, KVM_REG_SIZE(reg->id)))
485 		return -EFAULT;
486 
487 	return 0;
488 }
489 
490 int kvm_arm_set_fw_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
491 {
492 	void __user *uaddr = (void __user *)(long)reg->addr;
493 	u64 val;
494 	int wa_level;
495 
496 	if (copy_from_user(&val, uaddr, KVM_REG_SIZE(reg->id)))
497 		return -EFAULT;
498 
499 	switch (reg->id) {
500 	case KVM_REG_ARM_PSCI_VERSION:
501 	{
502 		bool wants_02;
503 
504 		wants_02 = test_bit(KVM_ARM_VCPU_PSCI_0_2, vcpu->arch.features);
505 
506 		switch (val) {
507 		case KVM_ARM_PSCI_0_1:
508 			if (wants_02)
509 				return -EINVAL;
510 			vcpu->kvm->arch.psci_version = val;
511 			return 0;
512 		case KVM_ARM_PSCI_0_2:
513 		case KVM_ARM_PSCI_1_0:
514 			if (!wants_02)
515 				return -EINVAL;
516 			vcpu->kvm->arch.psci_version = val;
517 			return 0;
518 		}
519 		break;
520 	}
521 
522 	case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1:
523 		if (val & ~KVM_REG_FEATURE_LEVEL_MASK)
524 			return -EINVAL;
525 
526 		if (get_kernel_wa_level(reg->id) < val)
527 			return -EINVAL;
528 
529 		return 0;
530 
531 	case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2:
532 		if (val & ~(KVM_REG_FEATURE_LEVEL_MASK |
533 			    KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_ENABLED))
534 			return -EINVAL;
535 
536 		/* The enabled bit must not be set unless the level is AVAIL. */
537 		if ((val & KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_ENABLED) &&
538 		    (val & KVM_REG_FEATURE_LEVEL_MASK) != KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_AVAIL)
539 			return -EINVAL;
540 
541 		/*
542 		 * Map all the possible incoming states to the only two we
543 		 * really want to deal with.
544 		 */
545 		switch (val & KVM_REG_FEATURE_LEVEL_MASK) {
546 		case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_AVAIL:
547 		case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_UNKNOWN:
548 			wa_level = KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_AVAIL;
549 			break;
550 		case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_AVAIL:
551 		case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_REQUIRED:
552 			wa_level = KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_REQUIRED;
553 			break;
554 		default:
555 			return -EINVAL;
556 		}
557 
558 		/*
559 		 * We can deal with NOT_AVAIL on NOT_REQUIRED, but not the
560 		 * other way around.
561 		 */
562 		if (get_kernel_wa_level(reg->id) < wa_level)
563 			return -EINVAL;
564 
565 		return 0;
566 	default:
567 		return -ENOENT;
568 	}
569 
570 	return -EINVAL;
571 }
572