xref: /linux/arch/arm64/kvm/hypercalls.c (revision 0e685c3e7158d35626d6d76b9f859eae806d87fa)
1 // SPDX-License-Identifier: GPL-2.0
2 // Copyright (C) 2019 Arm Ltd.
3 
4 #include <linux/arm-smccc.h>
5 #include <linux/kvm_host.h>
6 
7 #include <asm/kvm_emulate.h>
8 
9 #include <kvm/arm_hypercalls.h>
10 #include <kvm/arm_psci.h>
11 
12 #define KVM_ARM_SMCCC_STD_FEATURES				\
13 	GENMASK(KVM_REG_ARM_STD_BMAP_BIT_COUNT - 1, 0)
14 #define KVM_ARM_SMCCC_STD_HYP_FEATURES				\
15 	GENMASK(KVM_REG_ARM_STD_HYP_BMAP_BIT_COUNT - 1, 0)
16 #define KVM_ARM_SMCCC_VENDOR_HYP_FEATURES			\
17 	GENMASK(KVM_REG_ARM_VENDOR_HYP_BMAP_BIT_COUNT - 1, 0)
18 
19 static void kvm_ptp_get_time(struct kvm_vcpu *vcpu, u64 *val)
20 {
21 	struct system_time_snapshot systime_snapshot;
22 	u64 cycles = ~0UL;
23 	u32 feature;
24 
25 	/*
26 	 * system time and counter value must captured at the same
27 	 * time to keep consistency and precision.
28 	 */
29 	ktime_get_snapshot(&systime_snapshot);
30 
31 	/*
32 	 * This is only valid if the current clocksource is the
33 	 * architected counter, as this is the only one the guest
34 	 * can see.
35 	 */
36 	if (systime_snapshot.cs_id != CSID_ARM_ARCH_COUNTER)
37 		return;
38 
39 	/*
40 	 * The guest selects one of the two reference counters
41 	 * (virtual or physical) with the first argument of the SMCCC
42 	 * call. In case the identifier is not supported, error out.
43 	 */
44 	feature = smccc_get_arg1(vcpu);
45 	switch (feature) {
46 	case KVM_PTP_VIRT_COUNTER:
47 		cycles = systime_snapshot.cycles - vcpu_read_sys_reg(vcpu, CNTVOFF_EL2);
48 		break;
49 	case KVM_PTP_PHYS_COUNTER:
50 		cycles = systime_snapshot.cycles;
51 		break;
52 	default:
53 		return;
54 	}
55 
56 	/*
57 	 * This relies on the top bit of val[0] never being set for
58 	 * valid values of system time, because that is *really* far
59 	 * in the future (about 292 years from 1970, and at that stage
60 	 * nobody will give a damn about it).
61 	 */
62 	val[0] = upper_32_bits(systime_snapshot.real);
63 	val[1] = lower_32_bits(systime_snapshot.real);
64 	val[2] = upper_32_bits(cycles);
65 	val[3] = lower_32_bits(cycles);
66 }
67 
68 static bool kvm_hvc_call_default_allowed(u32 func_id)
69 {
70 	switch (func_id) {
71 	/*
72 	 * List of function-ids that are not gated with the bitmapped
73 	 * feature firmware registers, and are to be allowed for
74 	 * servicing the call by default.
75 	 */
76 	case ARM_SMCCC_VERSION_FUNC_ID:
77 	case ARM_SMCCC_ARCH_FEATURES_FUNC_ID:
78 		return true;
79 	default:
80 		/* PSCI 0.2 and up is in the 0:0x1f range */
81 		if (ARM_SMCCC_OWNER_NUM(func_id) == ARM_SMCCC_OWNER_STANDARD &&
82 		    ARM_SMCCC_FUNC_NUM(func_id) <= 0x1f)
83 			return true;
84 
85 		/*
86 		 * KVM's PSCI 0.1 doesn't comply with SMCCC, and has
87 		 * its own function-id base and range
88 		 */
89 		if (func_id >= KVM_PSCI_FN(0) && func_id <= KVM_PSCI_FN(3))
90 			return true;
91 
92 		return false;
93 	}
94 }
95 
96 static bool kvm_hvc_call_allowed(struct kvm_vcpu *vcpu, u32 func_id)
97 {
98 	struct kvm_smccc_features *smccc_feat = &vcpu->kvm->arch.smccc_feat;
99 
100 	switch (func_id) {
101 	case ARM_SMCCC_TRNG_VERSION:
102 	case ARM_SMCCC_TRNG_FEATURES:
103 	case ARM_SMCCC_TRNG_GET_UUID:
104 	case ARM_SMCCC_TRNG_RND32:
105 	case ARM_SMCCC_TRNG_RND64:
106 		return test_bit(KVM_REG_ARM_STD_BIT_TRNG_V1_0,
107 				&smccc_feat->std_bmap);
108 	case ARM_SMCCC_HV_PV_TIME_FEATURES:
109 	case ARM_SMCCC_HV_PV_TIME_ST:
110 		return test_bit(KVM_REG_ARM_STD_HYP_BIT_PV_TIME,
111 				&smccc_feat->std_hyp_bmap);
112 	case ARM_SMCCC_VENDOR_HYP_KVM_FEATURES_FUNC_ID:
113 	case ARM_SMCCC_VENDOR_HYP_CALL_UID_FUNC_ID:
114 		return test_bit(KVM_REG_ARM_VENDOR_HYP_BIT_FUNC_FEAT,
115 				&smccc_feat->vendor_hyp_bmap);
116 	case ARM_SMCCC_VENDOR_HYP_KVM_PTP_FUNC_ID:
117 		return test_bit(KVM_REG_ARM_VENDOR_HYP_BIT_PTP,
118 				&smccc_feat->vendor_hyp_bmap);
119 	default:
120 		return kvm_hvc_call_default_allowed(func_id);
121 	}
122 }
123 
124 int kvm_hvc_call_handler(struct kvm_vcpu *vcpu)
125 {
126 	struct kvm_smccc_features *smccc_feat = &vcpu->kvm->arch.smccc_feat;
127 	u32 func_id = smccc_get_function(vcpu);
128 	u64 val[4] = {SMCCC_RET_NOT_SUPPORTED};
129 	u32 feature;
130 	gpa_t gpa;
131 
132 	if (!kvm_hvc_call_allowed(vcpu, func_id))
133 		goto out;
134 
135 	switch (func_id) {
136 	case ARM_SMCCC_VERSION_FUNC_ID:
137 		val[0] = ARM_SMCCC_VERSION_1_1;
138 		break;
139 	case ARM_SMCCC_ARCH_FEATURES_FUNC_ID:
140 		feature = smccc_get_arg1(vcpu);
141 		switch (feature) {
142 		case ARM_SMCCC_ARCH_WORKAROUND_1:
143 			switch (arm64_get_spectre_v2_state()) {
144 			case SPECTRE_VULNERABLE:
145 				break;
146 			case SPECTRE_MITIGATED:
147 				val[0] = SMCCC_RET_SUCCESS;
148 				break;
149 			case SPECTRE_UNAFFECTED:
150 				val[0] = SMCCC_ARCH_WORKAROUND_RET_UNAFFECTED;
151 				break;
152 			}
153 			break;
154 		case ARM_SMCCC_ARCH_WORKAROUND_2:
155 			switch (arm64_get_spectre_v4_state()) {
156 			case SPECTRE_VULNERABLE:
157 				break;
158 			case SPECTRE_MITIGATED:
159 				/*
160 				 * SSBS everywhere: Indicate no firmware
161 				 * support, as the SSBS support will be
162 				 * indicated to the guest and the default is
163 				 * safe.
164 				 *
165 				 * Otherwise, expose a permanent mitigation
166 				 * to the guest, and hide SSBS so that the
167 				 * guest stays protected.
168 				 */
169 				if (cpus_have_final_cap(ARM64_SSBS))
170 					break;
171 				fallthrough;
172 			case SPECTRE_UNAFFECTED:
173 				val[0] = SMCCC_RET_NOT_REQUIRED;
174 				break;
175 			}
176 			break;
177 		case ARM_SMCCC_ARCH_WORKAROUND_3:
178 			switch (arm64_get_spectre_bhb_state()) {
179 			case SPECTRE_VULNERABLE:
180 				break;
181 			case SPECTRE_MITIGATED:
182 				val[0] = SMCCC_RET_SUCCESS;
183 				break;
184 			case SPECTRE_UNAFFECTED:
185 				val[0] = SMCCC_ARCH_WORKAROUND_RET_UNAFFECTED;
186 				break;
187 			}
188 			break;
189 		case ARM_SMCCC_HV_PV_TIME_FEATURES:
190 			if (test_bit(KVM_REG_ARM_STD_HYP_BIT_PV_TIME,
191 				     &smccc_feat->std_hyp_bmap))
192 				val[0] = SMCCC_RET_SUCCESS;
193 			break;
194 		}
195 		break;
196 	case ARM_SMCCC_HV_PV_TIME_FEATURES:
197 		val[0] = kvm_hypercall_pv_features(vcpu);
198 		break;
199 	case ARM_SMCCC_HV_PV_TIME_ST:
200 		gpa = kvm_init_stolen_time(vcpu);
201 		if (gpa != INVALID_GPA)
202 			val[0] = gpa;
203 		break;
204 	case ARM_SMCCC_VENDOR_HYP_CALL_UID_FUNC_ID:
205 		val[0] = ARM_SMCCC_VENDOR_HYP_UID_KVM_REG_0;
206 		val[1] = ARM_SMCCC_VENDOR_HYP_UID_KVM_REG_1;
207 		val[2] = ARM_SMCCC_VENDOR_HYP_UID_KVM_REG_2;
208 		val[3] = ARM_SMCCC_VENDOR_HYP_UID_KVM_REG_3;
209 		break;
210 	case ARM_SMCCC_VENDOR_HYP_KVM_FEATURES_FUNC_ID:
211 		val[0] = smccc_feat->vendor_hyp_bmap;
212 		break;
213 	case ARM_SMCCC_VENDOR_HYP_KVM_PTP_FUNC_ID:
214 		kvm_ptp_get_time(vcpu, val);
215 		break;
216 	case ARM_SMCCC_TRNG_VERSION:
217 	case ARM_SMCCC_TRNG_FEATURES:
218 	case ARM_SMCCC_TRNG_GET_UUID:
219 	case ARM_SMCCC_TRNG_RND32:
220 	case ARM_SMCCC_TRNG_RND64:
221 		return kvm_trng_call(vcpu);
222 	default:
223 		return kvm_psci_call(vcpu);
224 	}
225 
226 out:
227 	smccc_set_retval(vcpu, val[0], val[1], val[2], val[3]);
228 	return 1;
229 }
230 
231 static const u64 kvm_arm_fw_reg_ids[] = {
232 	KVM_REG_ARM_PSCI_VERSION,
233 	KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1,
234 	KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2,
235 	KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_3,
236 	KVM_REG_ARM_STD_BMAP,
237 	KVM_REG_ARM_STD_HYP_BMAP,
238 	KVM_REG_ARM_VENDOR_HYP_BMAP,
239 };
240 
241 void kvm_arm_init_hypercalls(struct kvm *kvm)
242 {
243 	struct kvm_smccc_features *smccc_feat = &kvm->arch.smccc_feat;
244 
245 	smccc_feat->std_bmap = KVM_ARM_SMCCC_STD_FEATURES;
246 	smccc_feat->std_hyp_bmap = KVM_ARM_SMCCC_STD_HYP_FEATURES;
247 	smccc_feat->vendor_hyp_bmap = KVM_ARM_SMCCC_VENDOR_HYP_FEATURES;
248 }
249 
250 int kvm_arm_get_fw_num_regs(struct kvm_vcpu *vcpu)
251 {
252 	return ARRAY_SIZE(kvm_arm_fw_reg_ids);
253 }
254 
255 int kvm_arm_copy_fw_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
256 {
257 	int i;
258 
259 	for (i = 0; i < ARRAY_SIZE(kvm_arm_fw_reg_ids); i++) {
260 		if (put_user(kvm_arm_fw_reg_ids[i], uindices++))
261 			return -EFAULT;
262 	}
263 
264 	return 0;
265 }
266 
267 #define KVM_REG_FEATURE_LEVEL_MASK	GENMASK(3, 0)
268 
269 /*
270  * Convert the workaround level into an easy-to-compare number, where higher
271  * values mean better protection.
272  */
273 static int get_kernel_wa_level(u64 regid)
274 {
275 	switch (regid) {
276 	case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1:
277 		switch (arm64_get_spectre_v2_state()) {
278 		case SPECTRE_VULNERABLE:
279 			return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1_NOT_AVAIL;
280 		case SPECTRE_MITIGATED:
281 			return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1_AVAIL;
282 		case SPECTRE_UNAFFECTED:
283 			return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1_NOT_REQUIRED;
284 		}
285 		return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1_NOT_AVAIL;
286 	case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2:
287 		switch (arm64_get_spectre_v4_state()) {
288 		case SPECTRE_MITIGATED:
289 			/*
290 			 * As for the hypercall discovery, we pretend we
291 			 * don't have any FW mitigation if SSBS is there at
292 			 * all times.
293 			 */
294 			if (cpus_have_final_cap(ARM64_SSBS))
295 				return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_AVAIL;
296 			fallthrough;
297 		case SPECTRE_UNAFFECTED:
298 			return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_REQUIRED;
299 		case SPECTRE_VULNERABLE:
300 			return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_AVAIL;
301 		}
302 		break;
303 	case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_3:
304 		switch (arm64_get_spectre_bhb_state()) {
305 		case SPECTRE_VULNERABLE:
306 			return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_3_NOT_AVAIL;
307 		case SPECTRE_MITIGATED:
308 			return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_3_AVAIL;
309 		case SPECTRE_UNAFFECTED:
310 			return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_3_NOT_REQUIRED;
311 		}
312 		return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_3_NOT_AVAIL;
313 	}
314 
315 	return -EINVAL;
316 }
317 
318 int kvm_arm_get_fw_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
319 {
320 	struct kvm_smccc_features *smccc_feat = &vcpu->kvm->arch.smccc_feat;
321 	void __user *uaddr = (void __user *)(long)reg->addr;
322 	u64 val;
323 
324 	switch (reg->id) {
325 	case KVM_REG_ARM_PSCI_VERSION:
326 		val = kvm_psci_version(vcpu);
327 		break;
328 	case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1:
329 	case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2:
330 	case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_3:
331 		val = get_kernel_wa_level(reg->id) & KVM_REG_FEATURE_LEVEL_MASK;
332 		break;
333 	case KVM_REG_ARM_STD_BMAP:
334 		val = READ_ONCE(smccc_feat->std_bmap);
335 		break;
336 	case KVM_REG_ARM_STD_HYP_BMAP:
337 		val = READ_ONCE(smccc_feat->std_hyp_bmap);
338 		break;
339 	case KVM_REG_ARM_VENDOR_HYP_BMAP:
340 		val = READ_ONCE(smccc_feat->vendor_hyp_bmap);
341 		break;
342 	default:
343 		return -ENOENT;
344 	}
345 
346 	if (copy_to_user(uaddr, &val, KVM_REG_SIZE(reg->id)))
347 		return -EFAULT;
348 
349 	return 0;
350 }
351 
352 static int kvm_arm_set_fw_reg_bmap(struct kvm_vcpu *vcpu, u64 reg_id, u64 val)
353 {
354 	int ret = 0;
355 	struct kvm *kvm = vcpu->kvm;
356 	struct kvm_smccc_features *smccc_feat = &kvm->arch.smccc_feat;
357 	unsigned long *fw_reg_bmap, fw_reg_features;
358 
359 	switch (reg_id) {
360 	case KVM_REG_ARM_STD_BMAP:
361 		fw_reg_bmap = &smccc_feat->std_bmap;
362 		fw_reg_features = KVM_ARM_SMCCC_STD_FEATURES;
363 		break;
364 	case KVM_REG_ARM_STD_HYP_BMAP:
365 		fw_reg_bmap = &smccc_feat->std_hyp_bmap;
366 		fw_reg_features = KVM_ARM_SMCCC_STD_HYP_FEATURES;
367 		break;
368 	case KVM_REG_ARM_VENDOR_HYP_BMAP:
369 		fw_reg_bmap = &smccc_feat->vendor_hyp_bmap;
370 		fw_reg_features = KVM_ARM_SMCCC_VENDOR_HYP_FEATURES;
371 		break;
372 	default:
373 		return -ENOENT;
374 	}
375 
376 	/* Check for unsupported bit */
377 	if (val & ~fw_reg_features)
378 		return -EINVAL;
379 
380 	mutex_lock(&kvm->lock);
381 
382 	if (test_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags) &&
383 	    val != *fw_reg_bmap) {
384 		ret = -EBUSY;
385 		goto out;
386 	}
387 
388 	WRITE_ONCE(*fw_reg_bmap, val);
389 out:
390 	mutex_unlock(&kvm->lock);
391 	return ret;
392 }
393 
394 int kvm_arm_set_fw_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
395 {
396 	void __user *uaddr = (void __user *)(long)reg->addr;
397 	u64 val;
398 	int wa_level;
399 
400 	if (copy_from_user(&val, uaddr, KVM_REG_SIZE(reg->id)))
401 		return -EFAULT;
402 
403 	switch (reg->id) {
404 	case KVM_REG_ARM_PSCI_VERSION:
405 	{
406 		bool wants_02;
407 
408 		wants_02 = test_bit(KVM_ARM_VCPU_PSCI_0_2, vcpu->arch.features);
409 
410 		switch (val) {
411 		case KVM_ARM_PSCI_0_1:
412 			if (wants_02)
413 				return -EINVAL;
414 			vcpu->kvm->arch.psci_version = val;
415 			return 0;
416 		case KVM_ARM_PSCI_0_2:
417 		case KVM_ARM_PSCI_1_0:
418 		case KVM_ARM_PSCI_1_1:
419 			if (!wants_02)
420 				return -EINVAL;
421 			vcpu->kvm->arch.psci_version = val;
422 			return 0;
423 		}
424 		break;
425 	}
426 
427 	case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1:
428 	case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_3:
429 		if (val & ~KVM_REG_FEATURE_LEVEL_MASK)
430 			return -EINVAL;
431 
432 		if (get_kernel_wa_level(reg->id) < val)
433 			return -EINVAL;
434 
435 		return 0;
436 
437 	case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2:
438 		if (val & ~(KVM_REG_FEATURE_LEVEL_MASK |
439 			    KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_ENABLED))
440 			return -EINVAL;
441 
442 		/* The enabled bit must not be set unless the level is AVAIL. */
443 		if ((val & KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_ENABLED) &&
444 		    (val & KVM_REG_FEATURE_LEVEL_MASK) != KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_AVAIL)
445 			return -EINVAL;
446 
447 		/*
448 		 * Map all the possible incoming states to the only two we
449 		 * really want to deal with.
450 		 */
451 		switch (val & KVM_REG_FEATURE_LEVEL_MASK) {
452 		case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_AVAIL:
453 		case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_UNKNOWN:
454 			wa_level = KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_AVAIL;
455 			break;
456 		case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_AVAIL:
457 		case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_REQUIRED:
458 			wa_level = KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_REQUIRED;
459 			break;
460 		default:
461 			return -EINVAL;
462 		}
463 
464 		/*
465 		 * We can deal with NOT_AVAIL on NOT_REQUIRED, but not the
466 		 * other way around.
467 		 */
468 		if (get_kernel_wa_level(reg->id) < wa_level)
469 			return -EINVAL;
470 
471 		return 0;
472 	case KVM_REG_ARM_STD_BMAP:
473 	case KVM_REG_ARM_STD_HYP_BMAP:
474 	case KVM_REG_ARM_VENDOR_HYP_BMAP:
475 		return kvm_arm_set_fw_reg_bmap(vcpu, reg->id, val);
476 	default:
477 		return -ENOENT;
478 	}
479 
480 	return -EINVAL;
481 }
482