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_coproc.h>
29 #include <asm/kvm_emulate.h>
30 #include <asm/kvm_mmu.h>
31 #include <asm/virt.h>
32
33 /* Maximum phys_shift supported for any VM on this host */
34 static u32 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_SVC (PSR_AA32_MODE_SVC | PSR_AA32_A_BIT | \
43 PSR_AA32_I_BIT | PSR_AA32_F_BIT)
44
system_has_full_ptr_auth(void)45 static bool system_has_full_ptr_auth(void)
46 {
47 return system_supports_address_auth() && system_supports_generic_auth();
48 }
49
50 /**
51 * kvm_arch_vm_ioctl_check_extension
52 *
53 * We currently assume that the number of HW registers is uniform
54 * across all CPUs (see cpuinfo_sanity_check).
55 */
kvm_arch_vm_ioctl_check_extension(struct kvm * kvm,long ext)56 int kvm_arch_vm_ioctl_check_extension(struct kvm *kvm, long ext)
57 {
58 int r;
59
60 switch (ext) {
61 case KVM_CAP_ARM_EL1_32BIT:
62 r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
63 break;
64 case KVM_CAP_GUEST_DEBUG_HW_BPS:
65 r = get_num_brps();
66 break;
67 case KVM_CAP_GUEST_DEBUG_HW_WPS:
68 r = get_num_wrps();
69 break;
70 case KVM_CAP_ARM_PMU_V3:
71 r = kvm_arm_support_pmu_v3();
72 break;
73 case KVM_CAP_ARM_INJECT_SERROR_ESR:
74 r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
75 break;
76 case KVM_CAP_SET_GUEST_DEBUG:
77 case KVM_CAP_VCPU_ATTRIBUTES:
78 r = 1;
79 break;
80 case KVM_CAP_ARM_VM_IPA_SIZE:
81 r = kvm_ipa_limit;
82 break;
83 case KVM_CAP_ARM_SVE:
84 r = system_supports_sve();
85 break;
86 case KVM_CAP_ARM_PTRAUTH_ADDRESS:
87 case KVM_CAP_ARM_PTRAUTH_GENERIC:
88 r = system_has_full_ptr_auth();
89 break;
90 default:
91 r = 0;
92 }
93
94 return r;
95 }
96
97 unsigned int kvm_sve_max_vl;
98
kvm_arm_init_sve(void)99 int kvm_arm_init_sve(void)
100 {
101 if (system_supports_sve()) {
102 kvm_sve_max_vl = sve_max_virtualisable_vl;
103
104 /*
105 * The get_sve_reg()/set_sve_reg() ioctl interface will need
106 * to be extended with multiple register slice support in
107 * order to support vector lengths greater than
108 * SVE_VL_ARCH_MAX:
109 */
110 if (WARN_ON(kvm_sve_max_vl > SVE_VL_ARCH_MAX))
111 kvm_sve_max_vl = SVE_VL_ARCH_MAX;
112
113 /*
114 * Don't even try to make use of vector lengths that
115 * aren't available on all CPUs, for now:
116 */
117 if (kvm_sve_max_vl < sve_max_vl)
118 pr_warn("KVM: SVE vector length for guests limited to %u bytes\n",
119 kvm_sve_max_vl);
120 }
121
122 return 0;
123 }
124
kvm_vcpu_enable_sve(struct kvm_vcpu * vcpu)125 static int kvm_vcpu_enable_sve(struct kvm_vcpu *vcpu)
126 {
127 if (!system_supports_sve())
128 return -EINVAL;
129
130 /* Verify that KVM startup enforced this when SVE was detected: */
131 if (WARN_ON(!has_vhe()))
132 return -EINVAL;
133
134 vcpu->arch.sve_max_vl = kvm_sve_max_vl;
135
136 /*
137 * Userspace can still customize the vector lengths by writing
138 * KVM_REG_ARM64_SVE_VLS. Allocation is deferred until
139 * kvm_arm_vcpu_finalize(), which freezes the configuration.
140 */
141 vcpu->arch.flags |= KVM_ARM64_GUEST_HAS_SVE;
142
143 return 0;
144 }
145
146 /*
147 * Finalize vcpu's maximum SVE vector length, allocating
148 * vcpu->arch.sve_state as necessary.
149 */
kvm_vcpu_finalize_sve(struct kvm_vcpu * vcpu)150 static int kvm_vcpu_finalize_sve(struct kvm_vcpu *vcpu)
151 {
152 void *buf;
153 unsigned int vl;
154
155 vl = vcpu->arch.sve_max_vl;
156
157 /*
158 * Responsibility for these properties is shared between
159 * kvm_arm_init_arch_resources(), kvm_vcpu_enable_sve() and
160 * set_sve_vls(). Double-check here just to be sure:
161 */
162 if (WARN_ON(!sve_vl_valid(vl) || vl > sve_max_virtualisable_vl ||
163 vl > SVE_VL_ARCH_MAX))
164 return -EIO;
165
166 buf = kzalloc(SVE_SIG_REGS_SIZE(sve_vq_from_vl(vl)), GFP_KERNEL);
167 if (!buf)
168 return -ENOMEM;
169
170 vcpu->arch.sve_state = buf;
171 vcpu->arch.flags |= KVM_ARM64_VCPU_SVE_FINALIZED;
172 return 0;
173 }
174
kvm_arm_vcpu_finalize(struct kvm_vcpu * vcpu,int feature)175 int kvm_arm_vcpu_finalize(struct kvm_vcpu *vcpu, int feature)
176 {
177 switch (feature) {
178 case KVM_ARM_VCPU_SVE:
179 if (!vcpu_has_sve(vcpu))
180 return -EINVAL;
181
182 if (kvm_arm_vcpu_sve_finalized(vcpu))
183 return -EPERM;
184
185 return kvm_vcpu_finalize_sve(vcpu);
186 }
187
188 return -EINVAL;
189 }
190
kvm_arm_vcpu_is_finalized(struct kvm_vcpu * vcpu)191 bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu)
192 {
193 if (vcpu_has_sve(vcpu) && !kvm_arm_vcpu_sve_finalized(vcpu))
194 return false;
195
196 return true;
197 }
198
kvm_arm_vcpu_destroy(struct kvm_vcpu * vcpu)199 void kvm_arm_vcpu_destroy(struct kvm_vcpu *vcpu)
200 {
201 kfree(vcpu->arch.sve_state);
202 }
203
kvm_vcpu_reset_sve(struct kvm_vcpu * vcpu)204 static void kvm_vcpu_reset_sve(struct kvm_vcpu *vcpu)
205 {
206 if (vcpu_has_sve(vcpu))
207 memset(vcpu->arch.sve_state, 0, vcpu_sve_state_size(vcpu));
208 }
209
kvm_vcpu_enable_ptrauth(struct kvm_vcpu * vcpu)210 static int kvm_vcpu_enable_ptrauth(struct kvm_vcpu *vcpu)
211 {
212 /*
213 * For now make sure that both address/generic pointer authentication
214 * features are requested by the userspace together and the system
215 * supports these capabilities.
216 */
217 if (!test_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, vcpu->arch.features) ||
218 !test_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, vcpu->arch.features) ||
219 !system_has_full_ptr_auth())
220 return -EINVAL;
221
222 vcpu->arch.flags |= KVM_ARM64_GUEST_HAS_PTRAUTH;
223 return 0;
224 }
225
vcpu_allowed_register_width(struct kvm_vcpu * vcpu)226 static bool vcpu_allowed_register_width(struct kvm_vcpu *vcpu)
227 {
228 struct kvm_vcpu *tmp;
229 bool is32bit;
230 int i;
231
232 is32bit = vcpu_has_feature(vcpu, KVM_ARM_VCPU_EL1_32BIT);
233 if (!cpus_have_const_cap(ARM64_HAS_32BIT_EL1) && is32bit)
234 return false;
235
236 /* Check that the vcpus are either all 32bit or all 64bit */
237 kvm_for_each_vcpu(i, tmp, vcpu->kvm) {
238 if (vcpu_has_feature(tmp, KVM_ARM_VCPU_EL1_32BIT) != is32bit)
239 return false;
240 }
241
242 return true;
243 }
244
245 /**
246 * kvm_reset_vcpu - sets core registers and sys_regs to reset value
247 * @vcpu: The VCPU pointer
248 *
249 * This function finds the right table above and sets the registers on
250 * the virtual CPU struct to their architecturally defined reset
251 * values, except for registers whose reset is deferred until
252 * kvm_arm_vcpu_finalize().
253 *
254 * Note: This function can be called from two paths: The KVM_ARM_VCPU_INIT
255 * ioctl or as part of handling a request issued by another VCPU in the PSCI
256 * handling code. In the first case, the VCPU will not be loaded, and in the
257 * second case the VCPU will be loaded. Because this function operates purely
258 * on the memory-backed values of system registers, we want to do a full put if
259 * we were loaded (handling a request) and load the values back at the end of
260 * the function. Otherwise we leave the state alone. In both cases, we
261 * disable preemption around the vcpu reset as we would otherwise race with
262 * preempt notifiers which also call put/load.
263 */
kvm_reset_vcpu(struct kvm_vcpu * vcpu)264 int kvm_reset_vcpu(struct kvm_vcpu *vcpu)
265 {
266 struct vcpu_reset_state reset_state;
267 int ret;
268 bool loaded;
269 u32 pstate;
270
271 mutex_lock(&vcpu->kvm->lock);
272 reset_state = vcpu->arch.reset_state;
273 WRITE_ONCE(vcpu->arch.reset_state.reset, false);
274 mutex_unlock(&vcpu->kvm->lock);
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 if (!kvm_arm_vcpu_sve_finalized(vcpu)) {
285 if (test_bit(KVM_ARM_VCPU_SVE, vcpu->arch.features)) {
286 ret = kvm_vcpu_enable_sve(vcpu);
287 if (ret)
288 goto out;
289 }
290 } else {
291 kvm_vcpu_reset_sve(vcpu);
292 }
293
294 if (test_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, vcpu->arch.features) ||
295 test_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, vcpu->arch.features)) {
296 if (kvm_vcpu_enable_ptrauth(vcpu)) {
297 ret = -EINVAL;
298 goto out;
299 }
300 }
301
302 if (!vcpu_allowed_register_width(vcpu)) {
303 ret = -EINVAL;
304 goto out;
305 }
306
307 switch (vcpu->arch.target) {
308 default:
309 if (test_bit(KVM_ARM_VCPU_EL1_32BIT, vcpu->arch.features)) {
310 pstate = VCPU_RESET_PSTATE_SVC;
311 } else {
312 pstate = VCPU_RESET_PSTATE_EL1;
313 }
314
315 break;
316 }
317
318 /* Reset core registers */
319 memset(vcpu_gp_regs(vcpu), 0, sizeof(*vcpu_gp_regs(vcpu)));
320 memset(&vcpu->arch.ctxt.fp_regs, 0, sizeof(vcpu->arch.ctxt.fp_regs));
321 vcpu->arch.ctxt.spsr_abt = 0;
322 vcpu->arch.ctxt.spsr_und = 0;
323 vcpu->arch.ctxt.spsr_irq = 0;
324 vcpu->arch.ctxt.spsr_fiq = 0;
325 vcpu_gp_regs(vcpu)->pstate = pstate;
326
327 /* Reset system registers */
328 kvm_reset_sys_regs(vcpu);
329
330 /*
331 * Additional reset state handling that PSCI may have imposed on us.
332 * Must be done after all the sys_reg reset.
333 */
334 if (reset_state.reset) {
335 unsigned long target_pc = reset_state.pc;
336
337 /* Gracefully handle Thumb2 entry point */
338 if (vcpu_mode_is_32bit(vcpu) && (target_pc & 1)) {
339 target_pc &= ~1UL;
340 vcpu_set_thumb(vcpu);
341 }
342
343 /* Propagate caller endianness */
344 if (reset_state.be)
345 kvm_vcpu_set_be(vcpu);
346
347 *vcpu_pc(vcpu) = target_pc;
348 vcpu_set_reg(vcpu, 0, reset_state.r0);
349 }
350
351 /* Reset timer */
352 ret = kvm_timer_vcpu_reset(vcpu);
353 out:
354 if (loaded)
355 kvm_arch_vcpu_load(vcpu, smp_processor_id());
356 preempt_enable();
357 return ret;
358 }
359
get_kvm_ipa_limit(void)360 u32 get_kvm_ipa_limit(void)
361 {
362 return kvm_ipa_limit;
363 }
364
kvm_set_ipa_limit(void)365 int kvm_set_ipa_limit(void)
366 {
367 unsigned int parange, tgran_2;
368 u64 mmfr0;
369
370 mmfr0 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
371 parange = cpuid_feature_extract_unsigned_field(mmfr0,
372 ID_AA64MMFR0_PARANGE_SHIFT);
373 /*
374 * IPA size beyond 48 bits could not be supported
375 * on either 4K or 16K page size. Hence let's cap
376 * it to 48 bits, in case it's reported as larger
377 * on the system.
378 */
379 if (PAGE_SIZE != SZ_64K)
380 parange = min(parange, (unsigned int)ID_AA64MMFR0_PARANGE_48);
381
382 /*
383 * Check with ARMv8.5-GTG that our PAGE_SIZE is supported at
384 * Stage-2. If not, things will stop very quickly.
385 */
386 switch (PAGE_SIZE) {
387 default:
388 case SZ_4K:
389 tgran_2 = ID_AA64MMFR0_TGRAN4_2_SHIFT;
390 break;
391 case SZ_16K:
392 tgran_2 = ID_AA64MMFR0_TGRAN16_2_SHIFT;
393 break;
394 case SZ_64K:
395 tgran_2 = ID_AA64MMFR0_TGRAN64_2_SHIFT;
396 break;
397 }
398
399 switch (cpuid_feature_extract_unsigned_field(mmfr0, tgran_2)) {
400 case ID_AA64MMFR0_TGRAN_2_SUPPORTED_NONE:
401 kvm_err("PAGE_SIZE not supported at Stage-2, giving up\n");
402 return -EINVAL;
403 case ID_AA64MMFR0_TGRAN_2_SUPPORTED_DEFAULT:
404 kvm_debug("PAGE_SIZE supported at Stage-2 (default)\n");
405 break;
406 case ID_AA64MMFR0_TGRAN_2_SUPPORTED_MIN ... ID_AA64MMFR0_TGRAN_2_SUPPORTED_MAX:
407 kvm_debug("PAGE_SIZE supported at Stage-2 (advertised)\n");
408 break;
409 default:
410 kvm_err("Unsupported value for TGRAN_2, giving up\n");
411 return -EINVAL;
412 }
413
414 kvm_ipa_limit = id_aa64mmfr0_parange_to_phys_shift(parange);
415 kvm_info("IPA Size Limit: %d bits%s\n", kvm_ipa_limit,
416 ((kvm_ipa_limit < KVM_PHYS_SHIFT) ?
417 " (Reduced IPA size, limited VM/VMM compatibility)" : ""));
418
419 return 0;
420 }
421
422 /*
423 * Configure the VTCR_EL2 for this VM. The VTCR value is common
424 * across all the physical CPUs on the system. We use system wide
425 * sanitised values to fill in different fields, except for Hardware
426 * Management of Access Flags. HA Flag is set unconditionally on
427 * all CPUs, as it is safe to run with or without the feature and
428 * the bit is RES0 on CPUs that don't support it.
429 */
kvm_arm_setup_stage2(struct kvm * kvm,unsigned long type)430 int kvm_arm_setup_stage2(struct kvm *kvm, unsigned long type)
431 {
432 u64 vtcr = VTCR_EL2_FLAGS, mmfr0;
433 u32 parange, phys_shift;
434 u8 lvls;
435
436 if (type & ~KVM_VM_TYPE_ARM_IPA_SIZE_MASK)
437 return -EINVAL;
438
439 phys_shift = KVM_VM_TYPE_ARM_IPA_SIZE(type);
440 if (phys_shift) {
441 if (phys_shift > kvm_ipa_limit ||
442 phys_shift < 32)
443 return -EINVAL;
444 } else {
445 phys_shift = KVM_PHYS_SHIFT;
446 if (phys_shift > kvm_ipa_limit) {
447 pr_warn_once("%s using unsupported default IPA limit, upgrade your VMM\n",
448 current->comm);
449 return -EINVAL;
450 }
451 }
452
453 mmfr0 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
454 parange = cpuid_feature_extract_unsigned_field(mmfr0,
455 ID_AA64MMFR0_PARANGE_SHIFT);
456 if (parange > ID_AA64MMFR0_PARANGE_MAX)
457 parange = ID_AA64MMFR0_PARANGE_MAX;
458 vtcr |= parange << VTCR_EL2_PS_SHIFT;
459
460 vtcr |= VTCR_EL2_T0SZ(phys_shift);
461 /*
462 * Use a minimum 2 level page table to prevent splitting
463 * host PMD huge pages at stage2.
464 */
465 lvls = stage2_pgtable_levels(phys_shift);
466 if (lvls < 2)
467 lvls = 2;
468 vtcr |= VTCR_EL2_LVLS_TO_SL0(lvls);
469
470 /*
471 * Enable the Hardware Access Flag management, unconditionally
472 * on all CPUs. The features is RES0 on CPUs without the support
473 * and must be ignored by the CPUs.
474 */
475 vtcr |= VTCR_EL2_HA;
476
477 /* Set the vmid bits */
478 vtcr |= (kvm_get_vmid_bits() == 16) ?
479 VTCR_EL2_VS_16BIT :
480 VTCR_EL2_VS_8BIT;
481 kvm->arch.vtcr = vtcr;
482 return 0;
483 }
484