1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) 2012,2013 - ARM Ltd
4 * Author: Marc Zyngier <[email protected]>
5 *
6 * Derived from arch/arm/kvm/reset.c
7 * Copyright (C) 2012 - Virtual Open Systems and Columbia University
8 * Author: Christoffer Dall <[email protected]>
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 unsigned int __ro_after_init kvm_host_sve_max_vl;
36
37 /*
38 * ARMv8 Reset Values
39 */
40 #define VCPU_RESET_PSTATE_EL1 (PSR_MODE_EL1h | PSR_A_BIT | PSR_I_BIT | \
41 PSR_F_BIT | PSR_D_BIT)
42
43 #define VCPU_RESET_PSTATE_EL2 (PSR_MODE_EL2h | PSR_A_BIT | PSR_I_BIT | \
44 PSR_F_BIT | PSR_D_BIT)
45
46 #define VCPU_RESET_PSTATE_SVC (PSR_AA32_MODE_SVC | PSR_AA32_A_BIT | \
47 PSR_AA32_I_BIT | PSR_AA32_F_BIT)
48
49 unsigned int __ro_after_init kvm_sve_max_vl;
50
kvm_arm_init_sve(void)51 int __init kvm_arm_init_sve(void)
52 {
53 if (system_supports_sve()) {
54 kvm_sve_max_vl = sve_max_virtualisable_vl();
55 kvm_host_sve_max_vl = sve_max_vl();
56 kvm_nvhe_sym(kvm_host_sve_max_vl) = kvm_host_sve_max_vl;
57
58 /*
59 * The get_sve_reg()/set_sve_reg() ioctl interface will need
60 * to be extended with multiple register slice support in
61 * order to support vector lengths greater than
62 * VL_ARCH_MAX:
63 */
64 if (WARN_ON(kvm_sve_max_vl > VL_ARCH_MAX))
65 kvm_sve_max_vl = VL_ARCH_MAX;
66
67 /*
68 * Don't even try to make use of vector lengths that
69 * aren't available on all CPUs, for now:
70 */
71 if (kvm_sve_max_vl < sve_max_vl())
72 pr_warn("KVM: SVE vector length for guests limited to %u bytes\n",
73 kvm_sve_max_vl);
74 }
75
76 return 0;
77 }
78
kvm_vcpu_enable_sve(struct kvm_vcpu * vcpu)79 static void kvm_vcpu_enable_sve(struct kvm_vcpu *vcpu)
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 set_bit(KVM_ARCH_FLAG_GUEST_HAS_SVE, &vcpu->kvm->arch.flags);
89 }
90
91 /*
92 * Finalize vcpu's maximum SVE vector length, allocating
93 * vcpu->arch.sve_state as necessary.
94 */
kvm_vcpu_finalize_sve(struct kvm_vcpu * vcpu)95 static int kvm_vcpu_finalize_sve(struct kvm_vcpu *vcpu)
96 {
97 void *buf;
98 unsigned int vl;
99 size_t reg_sz;
100 int ret;
101
102 vl = vcpu->arch.sve_max_vl;
103
104 /*
105 * Responsibility for these properties is shared between
106 * kvm_arm_init_sve(), 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 > VL_ARCH_MAX))
111 return -EIO;
112
113 reg_sz = vcpu_sve_state_size(vcpu);
114 buf = kzalloc(reg_sz, GFP_KERNEL_ACCOUNT);
115 if (!buf)
116 return -ENOMEM;
117
118 ret = kvm_share_hyp(buf, buf + reg_sz);
119 if (ret) {
120 kfree(buf);
121 return ret;
122 }
123
124 vcpu->arch.sve_state = buf;
125 vcpu_set_flag(vcpu, VCPU_SVE_FINALIZED);
126 return 0;
127 }
128
kvm_arm_vcpu_finalize(struct kvm_vcpu * vcpu,int feature)129 int kvm_arm_vcpu_finalize(struct kvm_vcpu *vcpu, int feature)
130 {
131 switch (feature) {
132 case KVM_ARM_VCPU_SVE:
133 if (!vcpu_has_sve(vcpu))
134 return -EINVAL;
135
136 if (kvm_arm_vcpu_sve_finalized(vcpu))
137 return -EPERM;
138
139 return kvm_vcpu_finalize_sve(vcpu);
140 }
141
142 return -EINVAL;
143 }
144
kvm_arm_vcpu_is_finalized(struct kvm_vcpu * vcpu)145 bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu)
146 {
147 if (vcpu_has_sve(vcpu) && !kvm_arm_vcpu_sve_finalized(vcpu))
148 return false;
149
150 return true;
151 }
152
kvm_arm_vcpu_destroy(struct kvm_vcpu * vcpu)153 void kvm_arm_vcpu_destroy(struct kvm_vcpu *vcpu)
154 {
155 void *sve_state = vcpu->arch.sve_state;
156
157 kvm_unshare_hyp(vcpu, vcpu + 1);
158 if (sve_state)
159 kvm_unshare_hyp(sve_state, sve_state + vcpu_sve_state_size(vcpu));
160 kfree(sve_state);
161 kfree(vcpu->arch.ccsidr);
162 }
163
kvm_vcpu_reset_sve(struct kvm_vcpu * vcpu)164 static void kvm_vcpu_reset_sve(struct kvm_vcpu *vcpu)
165 {
166 if (vcpu_has_sve(vcpu))
167 memset(vcpu->arch.sve_state, 0, vcpu_sve_state_size(vcpu));
168 }
169
170 /**
171 * kvm_reset_vcpu - sets core registers and sys_regs to reset value
172 * @vcpu: The VCPU pointer
173 *
174 * This function sets the registers on the virtual CPU struct to their
175 * architecturally defined reset values, except for registers whose reset is
176 * deferred until kvm_arm_vcpu_finalize().
177 *
178 * Note: This function can be called from two paths: The KVM_ARM_VCPU_INIT
179 * ioctl or as part of handling a request issued by another VCPU in the PSCI
180 * handling code. In the first case, the VCPU will not be loaded, and in the
181 * second case the VCPU will be loaded. Because this function operates purely
182 * on the memory-backed values of system registers, we want to do a full put if
183 * we were loaded (handling a request) and load the values back at the end of
184 * the function. Otherwise we leave the state alone. In both cases, we
185 * disable preemption around the vcpu reset as we would otherwise race with
186 * preempt notifiers which also call put/load.
187 */
kvm_reset_vcpu(struct kvm_vcpu * vcpu)188 void kvm_reset_vcpu(struct kvm_vcpu *vcpu)
189 {
190 struct vcpu_reset_state reset_state;
191 bool loaded;
192 u32 pstate;
193
194 spin_lock(&vcpu->arch.mp_state_lock);
195 reset_state = vcpu->arch.reset_state;
196 vcpu->arch.reset_state.reset = false;
197 spin_unlock(&vcpu->arch.mp_state_lock);
198
199 /* Reset PMU outside of the non-preemptible section */
200 kvm_pmu_vcpu_reset(vcpu);
201
202 preempt_disable();
203 loaded = (vcpu->cpu != -1);
204 if (loaded)
205 kvm_arch_vcpu_put(vcpu);
206
207 if (!kvm_arm_vcpu_sve_finalized(vcpu)) {
208 if (vcpu_has_feature(vcpu, KVM_ARM_VCPU_SVE))
209 kvm_vcpu_enable_sve(vcpu);
210 } else {
211 kvm_vcpu_reset_sve(vcpu);
212 }
213
214 if (vcpu_el1_is_32bit(vcpu))
215 pstate = VCPU_RESET_PSTATE_SVC;
216 else if (vcpu_has_nv(vcpu))
217 pstate = VCPU_RESET_PSTATE_EL2;
218 else
219 pstate = VCPU_RESET_PSTATE_EL1;
220
221 /* Reset core registers */
222 memset(vcpu_gp_regs(vcpu), 0, sizeof(*vcpu_gp_regs(vcpu)));
223 memset(&vcpu->arch.ctxt.fp_regs, 0, sizeof(vcpu->arch.ctxt.fp_regs));
224 vcpu->arch.ctxt.spsr_abt = 0;
225 vcpu->arch.ctxt.spsr_und = 0;
226 vcpu->arch.ctxt.spsr_irq = 0;
227 vcpu->arch.ctxt.spsr_fiq = 0;
228 vcpu_gp_regs(vcpu)->pstate = pstate;
229
230 /* Reset system registers */
231 kvm_reset_sys_regs(vcpu);
232
233 /*
234 * Additional reset state handling that PSCI may have imposed on us.
235 * Must be done after all the sys_reg reset.
236 */
237 if (reset_state.reset) {
238 unsigned long target_pc = reset_state.pc;
239
240 /* Gracefully handle Thumb2 entry point */
241 if (vcpu_mode_is_32bit(vcpu) && (target_pc & 1)) {
242 target_pc &= ~1UL;
243 vcpu_set_thumb(vcpu);
244 }
245
246 /* Propagate caller endianness */
247 if (reset_state.be)
248 kvm_vcpu_set_be(vcpu);
249
250 *vcpu_pc(vcpu) = target_pc;
251 vcpu_set_reg(vcpu, 0, reset_state.r0);
252 }
253
254 /* Reset timer */
255 kvm_timer_vcpu_reset(vcpu);
256
257 if (loaded)
258 kvm_arch_vcpu_load(vcpu, smp_processor_id());
259 preempt_enable();
260 }
261
kvm_get_pa_bits(struct kvm * kvm)262 u32 kvm_get_pa_bits(struct kvm *kvm)
263 {
264 /* Fixed limit until we can configure ID_AA64MMFR0.PARange */
265 return kvm_ipa_limit;
266 }
267
get_kvm_ipa_limit(void)268 u32 get_kvm_ipa_limit(void)
269 {
270 return kvm_ipa_limit;
271 }
272
kvm_set_ipa_limit(void)273 int __init kvm_set_ipa_limit(void)
274 {
275 unsigned int parange;
276 u64 mmfr0;
277
278 mmfr0 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
279 parange = cpuid_feature_extract_unsigned_field(mmfr0,
280 ID_AA64MMFR0_EL1_PARANGE_SHIFT);
281 /*
282 * IPA size beyond 48 bits for 4K and 16K page size is only supported
283 * when LPA2 is available. So if we have LPA2, enable it, else cap to 48
284 * bits, in case it's reported as larger on the system.
285 */
286 if (!kvm_lpa2_is_enabled() && PAGE_SIZE != SZ_64K)
287 parange = min(parange, (unsigned int)ID_AA64MMFR0_EL1_PARANGE_48);
288
289 /*
290 * Check with ARMv8.5-GTG that our PAGE_SIZE is supported at
291 * Stage-2. If not, things will stop very quickly.
292 */
293 switch (cpuid_feature_extract_unsigned_field(mmfr0, ID_AA64MMFR0_EL1_TGRAN_2_SHIFT)) {
294 case ID_AA64MMFR0_EL1_TGRAN_2_SUPPORTED_NONE:
295 kvm_err("PAGE_SIZE not supported at Stage-2, giving up\n");
296 return -EINVAL;
297 case ID_AA64MMFR0_EL1_TGRAN_2_SUPPORTED_DEFAULT:
298 kvm_debug("PAGE_SIZE supported at Stage-2 (default)\n");
299 break;
300 case ID_AA64MMFR0_EL1_TGRAN_2_SUPPORTED_MIN ... ID_AA64MMFR0_EL1_TGRAN_2_SUPPORTED_MAX:
301 kvm_debug("PAGE_SIZE supported at Stage-2 (advertised)\n");
302 break;
303 default:
304 kvm_err("Unsupported value for TGRAN_2, giving up\n");
305 return -EINVAL;
306 }
307
308 kvm_ipa_limit = id_aa64mmfr0_parange_to_phys_shift(parange);
309 kvm_info("IPA Size Limit: %d bits%s\n", kvm_ipa_limit,
310 ((kvm_ipa_limit < KVM_PHYS_SHIFT) ?
311 " (Reduced IPA size, limited VM/VMM compatibility)" : ""));
312
313 return 0;
314 }
315