xref: /linux/arch/arm64/kvm/reset.c (revision dfbe56bf48663ae383aba0eb756ba3615a3e9feb)
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/reset.c
7  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
8  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
9  */
10 
11 #include <linux/errno.h>
12 #include <linux/kernel.h>
13 #include <linux/kvm_host.h>
14 #include <linux/kvm.h>
15 #include <linux/hw_breakpoint.h>
16 #include <linux/slab.h>
17 #include <linux/string.h>
18 #include <linux/types.h>
19 
20 #include <kvm/arm_arch_timer.h>
21 
22 #include <asm/cpufeature.h>
23 #include <asm/cputype.h>
24 #include <asm/fpsimd.h>
25 #include <asm/ptrace.h>
26 #include <asm/kvm_arm.h>
27 #include <asm/kvm_asm.h>
28 #include <asm/kvm_emulate.h>
29 #include <asm/kvm_mmu.h>
30 #include <asm/virt.h>
31 
32 /* Maximum phys_shift supported for any VM on this host */
33 static u32 kvm_ipa_limit;
34 
35 /*
36  * ARMv8 Reset Values
37  */
38 #define VCPU_RESET_PSTATE_EL1	(PSR_MODE_EL1h | PSR_A_BIT | PSR_I_BIT | \
39 				 PSR_F_BIT | PSR_D_BIT)
40 
41 #define VCPU_RESET_PSTATE_SVC	(PSR_AA32_MODE_SVC | PSR_AA32_A_BIT | \
42 				 PSR_AA32_I_BIT | PSR_AA32_F_BIT)
43 
44 unsigned int kvm_sve_max_vl;
45 
46 int kvm_arm_init_sve(void)
47 {
48 	if (system_supports_sve()) {
49 		kvm_sve_max_vl = sve_max_virtualisable_vl;
50 
51 		/*
52 		 * The get_sve_reg()/set_sve_reg() ioctl interface will need
53 		 * to be extended with multiple register slice support in
54 		 * order to support vector lengths greater than
55 		 * SVE_VL_ARCH_MAX:
56 		 */
57 		if (WARN_ON(kvm_sve_max_vl > SVE_VL_ARCH_MAX))
58 			kvm_sve_max_vl = SVE_VL_ARCH_MAX;
59 
60 		/*
61 		 * Don't even try to make use of vector lengths that
62 		 * aren't available on all CPUs, for now:
63 		 */
64 		if (kvm_sve_max_vl < sve_max_vl)
65 			pr_warn("KVM: SVE vector length for guests limited to %u bytes\n",
66 				kvm_sve_max_vl);
67 	}
68 
69 	return 0;
70 }
71 
72 static int kvm_vcpu_enable_sve(struct kvm_vcpu *vcpu)
73 {
74 	if (!system_supports_sve())
75 		return -EINVAL;
76 
77 	/* Verify that KVM startup enforced this when SVE was detected: */
78 	if (WARN_ON(!has_vhe()))
79 		return -EINVAL;
80 
81 	vcpu->arch.sve_max_vl = kvm_sve_max_vl;
82 
83 	/*
84 	 * Userspace can still customize the vector lengths by writing
85 	 * KVM_REG_ARM64_SVE_VLS.  Allocation is deferred until
86 	 * kvm_arm_vcpu_finalize(), which freezes the configuration.
87 	 */
88 	vcpu->arch.flags |= KVM_ARM64_GUEST_HAS_SVE;
89 
90 	return 0;
91 }
92 
93 /*
94  * Finalize vcpu's maximum SVE vector length, allocating
95  * vcpu->arch.sve_state as necessary.
96  */
97 static int kvm_vcpu_finalize_sve(struct kvm_vcpu *vcpu)
98 {
99 	void *buf;
100 	unsigned int vl;
101 
102 	vl = vcpu->arch.sve_max_vl;
103 
104 	/*
105 	 * Responsibility for these properties is shared between
106 	 * kvm_arm_init_arch_resources(), kvm_vcpu_enable_sve() and
107 	 * set_sve_vls().  Double-check here just to be sure:
108 	 */
109 	if (WARN_ON(!sve_vl_valid(vl) || vl > sve_max_virtualisable_vl ||
110 		    vl > SVE_VL_ARCH_MAX))
111 		return -EIO;
112 
113 	buf = kzalloc(SVE_SIG_REGS_SIZE(sve_vq_from_vl(vl)), GFP_KERNEL);
114 	if (!buf)
115 		return -ENOMEM;
116 
117 	vcpu->arch.sve_state = buf;
118 	vcpu->arch.flags |= KVM_ARM64_VCPU_SVE_FINALIZED;
119 	return 0;
120 }
121 
122 int kvm_arm_vcpu_finalize(struct kvm_vcpu *vcpu, int feature)
123 {
124 	switch (feature) {
125 	case KVM_ARM_VCPU_SVE:
126 		if (!vcpu_has_sve(vcpu))
127 			return -EINVAL;
128 
129 		if (kvm_arm_vcpu_sve_finalized(vcpu))
130 			return -EPERM;
131 
132 		return kvm_vcpu_finalize_sve(vcpu);
133 	}
134 
135 	return -EINVAL;
136 }
137 
138 bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu)
139 {
140 	if (vcpu_has_sve(vcpu) && !kvm_arm_vcpu_sve_finalized(vcpu))
141 		return false;
142 
143 	return true;
144 }
145 
146 void kvm_arm_vcpu_destroy(struct kvm_vcpu *vcpu)
147 {
148 	kfree(vcpu->arch.sve_state);
149 }
150 
151 static void kvm_vcpu_reset_sve(struct kvm_vcpu *vcpu)
152 {
153 	if (vcpu_has_sve(vcpu))
154 		memset(vcpu->arch.sve_state, 0, vcpu_sve_state_size(vcpu));
155 }
156 
157 static int kvm_vcpu_enable_ptrauth(struct kvm_vcpu *vcpu)
158 {
159 	/*
160 	 * For now make sure that both address/generic pointer authentication
161 	 * features are requested by the userspace together and the system
162 	 * supports these capabilities.
163 	 */
164 	if (!test_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, vcpu->arch.features) ||
165 	    !test_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, vcpu->arch.features) ||
166 	    !system_has_full_ptr_auth())
167 		return -EINVAL;
168 
169 	vcpu->arch.flags |= KVM_ARM64_GUEST_HAS_PTRAUTH;
170 	return 0;
171 }
172 
173 /**
174  * kvm_reset_vcpu - sets core registers and sys_regs to reset value
175  * @vcpu: The VCPU pointer
176  *
177  * This function finds the right table above and sets the registers on
178  * the virtual CPU struct to their architecturally defined reset
179  * values, except for registers whose reset is deferred until
180  * kvm_arm_vcpu_finalize().
181  *
182  * Note: This function can be called from two paths: The KVM_ARM_VCPU_INIT
183  * ioctl or as part of handling a request issued by another VCPU in the PSCI
184  * handling code.  In the first case, the VCPU will not be loaded, and in the
185  * second case the VCPU will be loaded.  Because this function operates purely
186  * on the memory-backed values of system registers, we want to do a full put if
187  * we were loaded (handling a request) and load the values back at the end of
188  * the function.  Otherwise we leave the state alone.  In both cases, we
189  * disable preemption around the vcpu reset as we would otherwise race with
190  * preempt notifiers which also call put/load.
191  */
192 int kvm_reset_vcpu(struct kvm_vcpu *vcpu)
193 {
194 	int ret;
195 	bool loaded;
196 	u32 pstate;
197 
198 	/* Reset PMU outside of the non-preemptible section */
199 	kvm_pmu_vcpu_reset(vcpu);
200 
201 	preempt_disable();
202 	loaded = (vcpu->cpu != -1);
203 	if (loaded)
204 		kvm_arch_vcpu_put(vcpu);
205 
206 	if (!kvm_arm_vcpu_sve_finalized(vcpu)) {
207 		if (test_bit(KVM_ARM_VCPU_SVE, vcpu->arch.features)) {
208 			ret = kvm_vcpu_enable_sve(vcpu);
209 			if (ret)
210 				goto out;
211 		}
212 	} else {
213 		kvm_vcpu_reset_sve(vcpu);
214 	}
215 
216 	if (test_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, vcpu->arch.features) ||
217 	    test_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, vcpu->arch.features)) {
218 		if (kvm_vcpu_enable_ptrauth(vcpu)) {
219 			ret = -EINVAL;
220 			goto out;
221 		}
222 	}
223 
224 	switch (vcpu->arch.target) {
225 	default:
226 		if (test_bit(KVM_ARM_VCPU_EL1_32BIT, vcpu->arch.features)) {
227 			if (!cpus_have_const_cap(ARM64_HAS_32BIT_EL1)) {
228 				ret = -EINVAL;
229 				goto out;
230 			}
231 			pstate = VCPU_RESET_PSTATE_SVC;
232 		} else {
233 			pstate = VCPU_RESET_PSTATE_EL1;
234 		}
235 
236 		if (kvm_vcpu_has_pmu(vcpu) && !kvm_arm_support_pmu_v3()) {
237 			ret = -EINVAL;
238 			goto out;
239 		}
240 		break;
241 	}
242 
243 	/* Reset core registers */
244 	memset(vcpu_gp_regs(vcpu), 0, sizeof(*vcpu_gp_regs(vcpu)));
245 	vcpu_gp_regs(vcpu)->pstate = pstate;
246 
247 	/* Reset system registers */
248 	kvm_reset_sys_regs(vcpu);
249 
250 	/*
251 	 * Additional reset state handling that PSCI may have imposed on us.
252 	 * Must be done after all the sys_reg reset.
253 	 */
254 	if (vcpu->arch.reset_state.reset) {
255 		unsigned long target_pc = vcpu->arch.reset_state.pc;
256 
257 		/* Gracefully handle Thumb2 entry point */
258 		if (vcpu_mode_is_32bit(vcpu) && (target_pc & 1)) {
259 			target_pc &= ~1UL;
260 			vcpu_set_thumb(vcpu);
261 		}
262 
263 		/* Propagate caller endianness */
264 		if (vcpu->arch.reset_state.be)
265 			kvm_vcpu_set_be(vcpu);
266 
267 		*vcpu_pc(vcpu) = target_pc;
268 		vcpu_set_reg(vcpu, 0, vcpu->arch.reset_state.r0);
269 
270 		vcpu->arch.reset_state.reset = false;
271 	}
272 
273 	/* Reset timer */
274 	ret = kvm_timer_vcpu_reset(vcpu);
275 out:
276 	if (loaded)
277 		kvm_arch_vcpu_load(vcpu, smp_processor_id());
278 	preempt_enable();
279 	return ret;
280 }
281 
282 u32 get_kvm_ipa_limit(void)
283 {
284 	return kvm_ipa_limit;
285 }
286 
287 int kvm_set_ipa_limit(void)
288 {
289 	unsigned int parange, tgran_2;
290 	u64 mmfr0;
291 
292 	mmfr0 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
293 	parange = cpuid_feature_extract_unsigned_field(mmfr0,
294 				ID_AA64MMFR0_PARANGE_SHIFT);
295 
296 	/*
297 	 * Check with ARMv8.5-GTG that our PAGE_SIZE is supported at
298 	 * Stage-2. If not, things will stop very quickly.
299 	 */
300 	switch (PAGE_SIZE) {
301 	default:
302 	case SZ_4K:
303 		tgran_2 = ID_AA64MMFR0_TGRAN4_2_SHIFT;
304 		break;
305 	case SZ_16K:
306 		tgran_2 = ID_AA64MMFR0_TGRAN16_2_SHIFT;
307 		break;
308 	case SZ_64K:
309 		tgran_2 = ID_AA64MMFR0_TGRAN64_2_SHIFT;
310 		break;
311 	}
312 
313 	switch (cpuid_feature_extract_unsigned_field(mmfr0, tgran_2)) {
314 	case ID_AA64MMFR0_TGRAN_2_SUPPORTED_NONE:
315 		kvm_err("PAGE_SIZE not supported at Stage-2, giving up\n");
316 		return -EINVAL;
317 	case ID_AA64MMFR0_TGRAN_2_SUPPORTED_DEFAULT:
318 		kvm_debug("PAGE_SIZE supported at Stage-2 (default)\n");
319 		break;
320 	case ID_AA64MMFR0_TGRAN_2_SUPPORTED_MIN ... ID_AA64MMFR0_TGRAN_2_SUPPORTED_MAX:
321 		kvm_debug("PAGE_SIZE supported at Stage-2 (advertised)\n");
322 		break;
323 	default:
324 		kvm_err("Unsupported value for TGRAN_2, giving up\n");
325 		return -EINVAL;
326 	}
327 
328 	kvm_ipa_limit = id_aa64mmfr0_parange_to_phys_shift(parange);
329 	kvm_info("IPA Size Limit: %d bits%s\n", kvm_ipa_limit,
330 		 ((kvm_ipa_limit < KVM_PHYS_SHIFT) ?
331 		  " (Reduced IPA size, limited VM/VMM compatibility)" : ""));
332 
333 	return 0;
334 }
335 
336 /*
337  * Configure the VTCR_EL2 for this VM. The VTCR value is common
338  * across all the physical CPUs on the system. We use system wide
339  * sanitised values to fill in different fields, except for Hardware
340  * Management of Access Flags. HA Flag is set unconditionally on
341  * all CPUs, as it is safe to run with or without the feature and
342  * the bit is RES0 on CPUs that don't support it.
343  */
344 int kvm_arm_setup_stage2(struct kvm *kvm, unsigned long type)
345 {
346 	u64 vtcr = VTCR_EL2_FLAGS, mmfr0;
347 	u32 parange, phys_shift;
348 	u8 lvls;
349 
350 	if (type & ~KVM_VM_TYPE_ARM_IPA_SIZE_MASK)
351 		return -EINVAL;
352 
353 	phys_shift = KVM_VM_TYPE_ARM_IPA_SIZE(type);
354 	if (phys_shift) {
355 		if (phys_shift > kvm_ipa_limit ||
356 		    phys_shift < 32)
357 			return -EINVAL;
358 	} else {
359 		phys_shift = KVM_PHYS_SHIFT;
360 		if (phys_shift > kvm_ipa_limit) {
361 			pr_warn_once("%s using unsupported default IPA limit, upgrade your VMM\n",
362 				     current->comm);
363 			return -EINVAL;
364 		}
365 	}
366 
367 	mmfr0 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
368 	parange = cpuid_feature_extract_unsigned_field(mmfr0,
369 				ID_AA64MMFR0_PARANGE_SHIFT);
370 	if (parange > ID_AA64MMFR0_PARANGE_MAX)
371 		parange = ID_AA64MMFR0_PARANGE_MAX;
372 	vtcr |= parange << VTCR_EL2_PS_SHIFT;
373 
374 	vtcr |= VTCR_EL2_T0SZ(phys_shift);
375 	/*
376 	 * Use a minimum 2 level page table to prevent splitting
377 	 * host PMD huge pages at stage2.
378 	 */
379 	lvls = stage2_pgtable_levels(phys_shift);
380 	if (lvls < 2)
381 		lvls = 2;
382 	vtcr |= VTCR_EL2_LVLS_TO_SL0(lvls);
383 
384 	/*
385 	 * Enable the Hardware Access Flag management, unconditionally
386 	 * on all CPUs. The features is RES0 on CPUs without the support
387 	 * and must be ignored by the CPUs.
388 	 */
389 	vtcr |= VTCR_EL2_HA;
390 
391 	/* Set the vmid bits */
392 	vtcr |= (kvm_get_vmid_bits() == 16) ?
393 		VTCR_EL2_VS_16BIT :
394 		VTCR_EL2_VS_8BIT;
395 	kvm->arch.vtcr = vtcr;
396 	return 0;
397 }
398