xref: /linux/arch/arm64/kvm/reset.c (revision 0e685c3e7158d35626d6d76b9f859eae806d87fa)
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/kvm_nested.h>
31 #include <asm/virt.h>
32 
33 /* Maximum phys_shift supported for any VM on this host */
34 static u32 __ro_after_init kvm_ipa_limit;
35 
36 /*
37  * ARMv8 Reset Values
38  */
39 #define VCPU_RESET_PSTATE_EL1	(PSR_MODE_EL1h | PSR_A_BIT | PSR_I_BIT | \
40 				 PSR_F_BIT | PSR_D_BIT)
41 
42 #define VCPU_RESET_PSTATE_EL2	(PSR_MODE_EL2h | PSR_A_BIT | PSR_I_BIT | \
43 				 PSR_F_BIT | PSR_D_BIT)
44 
45 #define VCPU_RESET_PSTATE_SVC	(PSR_AA32_MODE_SVC | PSR_AA32_A_BIT | \
46 				 PSR_AA32_I_BIT | PSR_AA32_F_BIT)
47 
48 unsigned int __ro_after_init kvm_sve_max_vl;
49 
50 int __init kvm_arm_init_sve(void)
51 {
52 	if (system_supports_sve()) {
53 		kvm_sve_max_vl = sve_max_virtualisable_vl();
54 
55 		/*
56 		 * The get_sve_reg()/set_sve_reg() ioctl interface will need
57 		 * to be extended with multiple register slice support in
58 		 * order to support vector lengths greater than
59 		 * VL_ARCH_MAX:
60 		 */
61 		if (WARN_ON(kvm_sve_max_vl > VL_ARCH_MAX))
62 			kvm_sve_max_vl = VL_ARCH_MAX;
63 
64 		/*
65 		 * Don't even try to make use of vector lengths that
66 		 * aren't available on all CPUs, for now:
67 		 */
68 		if (kvm_sve_max_vl < sve_max_vl())
69 			pr_warn("KVM: SVE vector length for guests limited to %u bytes\n",
70 				kvm_sve_max_vl);
71 	}
72 
73 	return 0;
74 }
75 
76 static int kvm_vcpu_enable_sve(struct kvm_vcpu *vcpu)
77 {
78 	if (!system_supports_sve())
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_set_flag(vcpu, 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 	size_t reg_sz;
102 	int ret;
103 
104 	vl = vcpu->arch.sve_max_vl;
105 
106 	/*
107 	 * Responsibility for these properties is shared between
108 	 * kvm_arm_init_sve(), kvm_vcpu_enable_sve() and
109 	 * set_sve_vls().  Double-check here just to be sure:
110 	 */
111 	if (WARN_ON(!sve_vl_valid(vl) || vl > sve_max_virtualisable_vl() ||
112 		    vl > VL_ARCH_MAX))
113 		return -EIO;
114 
115 	reg_sz = vcpu_sve_state_size(vcpu);
116 	buf = kzalloc(reg_sz, GFP_KERNEL_ACCOUNT);
117 	if (!buf)
118 		return -ENOMEM;
119 
120 	ret = kvm_share_hyp(buf, buf + reg_sz);
121 	if (ret) {
122 		kfree(buf);
123 		return ret;
124 	}
125 
126 	vcpu->arch.sve_state = buf;
127 	vcpu_set_flag(vcpu, VCPU_SVE_FINALIZED);
128 	return 0;
129 }
130 
131 int kvm_arm_vcpu_finalize(struct kvm_vcpu *vcpu, int feature)
132 {
133 	switch (feature) {
134 	case KVM_ARM_VCPU_SVE:
135 		if (!vcpu_has_sve(vcpu))
136 			return -EINVAL;
137 
138 		if (kvm_arm_vcpu_sve_finalized(vcpu))
139 			return -EPERM;
140 
141 		return kvm_vcpu_finalize_sve(vcpu);
142 	}
143 
144 	return -EINVAL;
145 }
146 
147 bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu)
148 {
149 	if (vcpu_has_sve(vcpu) && !kvm_arm_vcpu_sve_finalized(vcpu))
150 		return false;
151 
152 	return true;
153 }
154 
155 void kvm_arm_vcpu_destroy(struct kvm_vcpu *vcpu)
156 {
157 	void *sve_state = vcpu->arch.sve_state;
158 
159 	kvm_vcpu_unshare_task_fp(vcpu);
160 	kvm_unshare_hyp(vcpu, vcpu + 1);
161 	if (sve_state)
162 		kvm_unshare_hyp(sve_state, sve_state + vcpu_sve_state_size(vcpu));
163 	kfree(sve_state);
164 	kfree(vcpu->arch.ccsidr);
165 }
166 
167 static void kvm_vcpu_reset_sve(struct kvm_vcpu *vcpu)
168 {
169 	if (vcpu_has_sve(vcpu))
170 		memset(vcpu->arch.sve_state, 0, vcpu_sve_state_size(vcpu));
171 }
172 
173 static int kvm_vcpu_enable_ptrauth(struct kvm_vcpu *vcpu)
174 {
175 	/*
176 	 * For now make sure that both address/generic pointer authentication
177 	 * features are requested by the userspace together and the system
178 	 * supports these capabilities.
179 	 */
180 	if (!test_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, vcpu->arch.features) ||
181 	    !test_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, vcpu->arch.features) ||
182 	    !system_has_full_ptr_auth())
183 		return -EINVAL;
184 
185 	vcpu_set_flag(vcpu, GUEST_HAS_PTRAUTH);
186 	return 0;
187 }
188 
189 /**
190  * kvm_set_vm_width() - set the register width for the guest
191  * @vcpu: Pointer to the vcpu being configured
192  *
193  * Set both KVM_ARCH_FLAG_EL1_32BIT and KVM_ARCH_FLAG_REG_WIDTH_CONFIGURED
194  * in the VM flags based on the vcpu's requested register width, the HW
195  * capabilities and other options (such as MTE).
196  * When REG_WIDTH_CONFIGURED is already set, the vcpu settings must be
197  * consistent with the value of the FLAG_EL1_32BIT bit in the flags.
198  *
199  * Return: 0 on success, negative error code on failure.
200  */
201 static int kvm_set_vm_width(struct kvm_vcpu *vcpu)
202 {
203 	struct kvm *kvm = vcpu->kvm;
204 	bool is32bit;
205 
206 	is32bit = vcpu_has_feature(vcpu, KVM_ARM_VCPU_EL1_32BIT);
207 
208 	lockdep_assert_held(&kvm->lock);
209 
210 	if (test_bit(KVM_ARCH_FLAG_REG_WIDTH_CONFIGURED, &kvm->arch.flags)) {
211 		/*
212 		 * The guest's register width is already configured.
213 		 * Make sure that the vcpu is consistent with it.
214 		 */
215 		if (is32bit == test_bit(KVM_ARCH_FLAG_EL1_32BIT, &kvm->arch.flags))
216 			return 0;
217 
218 		return -EINVAL;
219 	}
220 
221 	if (!cpus_have_const_cap(ARM64_HAS_32BIT_EL1) && is32bit)
222 		return -EINVAL;
223 
224 	/* MTE is incompatible with AArch32 */
225 	if (kvm_has_mte(kvm) && is32bit)
226 		return -EINVAL;
227 
228 	/* NV is incompatible with AArch32 */
229 	if (vcpu_has_nv(vcpu) && is32bit)
230 		return -EINVAL;
231 
232 	if (is32bit)
233 		set_bit(KVM_ARCH_FLAG_EL1_32BIT, &kvm->arch.flags);
234 
235 	set_bit(KVM_ARCH_FLAG_REG_WIDTH_CONFIGURED, &kvm->arch.flags);
236 
237 	return 0;
238 }
239 
240 /**
241  * kvm_reset_vcpu - sets core registers and sys_regs to reset value
242  * @vcpu: The VCPU pointer
243  *
244  * This function sets the registers on the virtual CPU struct to their
245  * architecturally defined reset values, except for registers whose reset is
246  * deferred until kvm_arm_vcpu_finalize().
247  *
248  * Note: This function can be called from two paths: The KVM_ARM_VCPU_INIT
249  * ioctl or as part of handling a request issued by another VCPU in the PSCI
250  * handling code.  In the first case, the VCPU will not be loaded, and in the
251  * second case the VCPU will be loaded.  Because this function operates purely
252  * on the memory-backed values of system registers, we want to do a full put if
253  * we were loaded (handling a request) and load the values back at the end of
254  * the function.  Otherwise we leave the state alone.  In both cases, we
255  * disable preemption around the vcpu reset as we would otherwise race with
256  * preempt notifiers which also call put/load.
257  */
258 int kvm_reset_vcpu(struct kvm_vcpu *vcpu)
259 {
260 	struct vcpu_reset_state reset_state;
261 	int ret;
262 	bool loaded;
263 	u32 pstate;
264 
265 	mutex_lock(&vcpu->kvm->lock);
266 	ret = kvm_set_vm_width(vcpu);
267 	if (!ret) {
268 		reset_state = vcpu->arch.reset_state;
269 		WRITE_ONCE(vcpu->arch.reset_state.reset, false);
270 	}
271 	mutex_unlock(&vcpu->kvm->lock);
272 
273 	if (ret)
274 		return ret;
275 
276 	/* Reset PMU outside of the non-preemptible section */
277 	kvm_pmu_vcpu_reset(vcpu);
278 
279 	preempt_disable();
280 	loaded = (vcpu->cpu != -1);
281 	if (loaded)
282 		kvm_arch_vcpu_put(vcpu);
283 
284 	/* Disallow NV+SVE for the time being */
285 	if (vcpu_has_nv(vcpu) && vcpu_has_feature(vcpu, KVM_ARM_VCPU_SVE)) {
286 		ret = -EINVAL;
287 		goto out;
288 	}
289 
290 	if (!kvm_arm_vcpu_sve_finalized(vcpu)) {
291 		if (test_bit(KVM_ARM_VCPU_SVE, vcpu->arch.features)) {
292 			ret = kvm_vcpu_enable_sve(vcpu);
293 			if (ret)
294 				goto out;
295 		}
296 	} else {
297 		kvm_vcpu_reset_sve(vcpu);
298 	}
299 
300 	if (test_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, vcpu->arch.features) ||
301 	    test_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, vcpu->arch.features)) {
302 		if (kvm_vcpu_enable_ptrauth(vcpu)) {
303 			ret = -EINVAL;
304 			goto out;
305 		}
306 	}
307 
308 	switch (vcpu->arch.target) {
309 	default:
310 		if (vcpu_el1_is_32bit(vcpu)) {
311 			pstate = VCPU_RESET_PSTATE_SVC;
312 		} else if (vcpu_has_nv(vcpu)) {
313 			pstate = VCPU_RESET_PSTATE_EL2;
314 		} else {
315 			pstate = VCPU_RESET_PSTATE_EL1;
316 		}
317 
318 		if (kvm_vcpu_has_pmu(vcpu) && !kvm_arm_support_pmu_v3()) {
319 			ret = -EINVAL;
320 			goto out;
321 		}
322 		break;
323 	}
324 
325 	/* Reset core registers */
326 	memset(vcpu_gp_regs(vcpu), 0, sizeof(*vcpu_gp_regs(vcpu)));
327 	memset(&vcpu->arch.ctxt.fp_regs, 0, sizeof(vcpu->arch.ctxt.fp_regs));
328 	vcpu->arch.ctxt.spsr_abt = 0;
329 	vcpu->arch.ctxt.spsr_und = 0;
330 	vcpu->arch.ctxt.spsr_irq = 0;
331 	vcpu->arch.ctxt.spsr_fiq = 0;
332 	vcpu_gp_regs(vcpu)->pstate = pstate;
333 
334 	/* Reset system registers */
335 	kvm_reset_sys_regs(vcpu);
336 
337 	/*
338 	 * Additional reset state handling that PSCI may have imposed on us.
339 	 * Must be done after all the sys_reg reset.
340 	 */
341 	if (reset_state.reset) {
342 		unsigned long target_pc = reset_state.pc;
343 
344 		/* Gracefully handle Thumb2 entry point */
345 		if (vcpu_mode_is_32bit(vcpu) && (target_pc & 1)) {
346 			target_pc &= ~1UL;
347 			vcpu_set_thumb(vcpu);
348 		}
349 
350 		/* Propagate caller endianness */
351 		if (reset_state.be)
352 			kvm_vcpu_set_be(vcpu);
353 
354 		*vcpu_pc(vcpu) = target_pc;
355 		vcpu_set_reg(vcpu, 0, reset_state.r0);
356 	}
357 
358 	/* Reset timer */
359 	ret = kvm_timer_vcpu_reset(vcpu);
360 out:
361 	if (loaded)
362 		kvm_arch_vcpu_load(vcpu, smp_processor_id());
363 	preempt_enable();
364 	return ret;
365 }
366 
367 u32 get_kvm_ipa_limit(void)
368 {
369 	return kvm_ipa_limit;
370 }
371 
372 int __init kvm_set_ipa_limit(void)
373 {
374 	unsigned int parange;
375 	u64 mmfr0;
376 
377 	mmfr0 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
378 	parange = cpuid_feature_extract_unsigned_field(mmfr0,
379 				ID_AA64MMFR0_EL1_PARANGE_SHIFT);
380 	/*
381 	 * IPA size beyond 48 bits could not be supported
382 	 * on either 4K or 16K page size. Hence let's cap
383 	 * it to 48 bits, in case it's reported as larger
384 	 * on the system.
385 	 */
386 	if (PAGE_SIZE != SZ_64K)
387 		parange = min(parange, (unsigned int)ID_AA64MMFR0_EL1_PARANGE_48);
388 
389 	/*
390 	 * Check with ARMv8.5-GTG that our PAGE_SIZE is supported at
391 	 * Stage-2. If not, things will stop very quickly.
392 	 */
393 	switch (cpuid_feature_extract_unsigned_field(mmfr0, ID_AA64MMFR0_EL1_TGRAN_2_SHIFT)) {
394 	case ID_AA64MMFR0_EL1_TGRAN_2_SUPPORTED_NONE:
395 		kvm_err("PAGE_SIZE not supported at Stage-2, giving up\n");
396 		return -EINVAL;
397 	case ID_AA64MMFR0_EL1_TGRAN_2_SUPPORTED_DEFAULT:
398 		kvm_debug("PAGE_SIZE supported at Stage-2 (default)\n");
399 		break;
400 	case ID_AA64MMFR0_EL1_TGRAN_2_SUPPORTED_MIN ... ID_AA64MMFR0_EL1_TGRAN_2_SUPPORTED_MAX:
401 		kvm_debug("PAGE_SIZE supported at Stage-2 (advertised)\n");
402 		break;
403 	default:
404 		kvm_err("Unsupported value for TGRAN_2, giving up\n");
405 		return -EINVAL;
406 	}
407 
408 	kvm_ipa_limit = id_aa64mmfr0_parange_to_phys_shift(parange);
409 	kvm_info("IPA Size Limit: %d bits%s\n", kvm_ipa_limit,
410 		 ((kvm_ipa_limit < KVM_PHYS_SHIFT) ?
411 		  " (Reduced IPA size, limited VM/VMM compatibility)" : ""));
412 
413 	return 0;
414 }
415