]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - arch/arm64/kvm/hyp/switch.c
arm64: Kill PSCI_GET_VERSION as a variant-2 workaround
[mirror_ubuntu-artful-kernel.git] / arch / arm64 / kvm / hyp / switch.c
1 /*
2 * Copyright (C) 2015 - ARM Ltd
3 * Author: Marc Zyngier <marc.zyngier@arm.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License version 2 as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program. If not, see <http://www.gnu.org/licenses/>.
16 */
17
18 #include <linux/types.h>
19 #include <linux/jump_label.h>
20 #include <uapi/linux/psci.h>
21
22 #include <kvm/arm_psci.h>
23
24 #include <asm/kvm_asm.h>
25 #include <asm/kvm_emulate.h>
26 #include <asm/kvm_hyp.h>
27 #include <asm/fpsimd.h>
28
29 static bool __hyp_text __fpsimd_enabled_nvhe(void)
30 {
31 return !(read_sysreg(cptr_el2) & CPTR_EL2_TFP);
32 }
33
34 static bool __hyp_text __fpsimd_enabled_vhe(void)
35 {
36 return !!(read_sysreg(cpacr_el1) & CPACR_EL1_FPEN);
37 }
38
39 static hyp_alternate_select(__fpsimd_is_enabled,
40 __fpsimd_enabled_nvhe, __fpsimd_enabled_vhe,
41 ARM64_HAS_VIRT_HOST_EXTN);
42
43 bool __hyp_text __fpsimd_enabled(void)
44 {
45 return __fpsimd_is_enabled()();
46 }
47
48 static void __hyp_text __activate_traps_vhe(void)
49 {
50 u64 val;
51
52 val = read_sysreg(cpacr_el1);
53 val |= CPACR_EL1_TTA;
54 val &= ~CPACR_EL1_FPEN;
55 write_sysreg(val, cpacr_el1);
56
57 write_sysreg(kvm_get_hyp_vector(), vbar_el1);
58 }
59
60 static void __hyp_text __activate_traps_nvhe(void)
61 {
62 u64 val;
63
64 val = CPTR_EL2_DEFAULT;
65 val |= CPTR_EL2_TTA | CPTR_EL2_TFP;
66 write_sysreg(val, cptr_el2);
67 }
68
69 static hyp_alternate_select(__activate_traps_arch,
70 __activate_traps_nvhe, __activate_traps_vhe,
71 ARM64_HAS_VIRT_HOST_EXTN);
72
73 static void __hyp_text __activate_traps(struct kvm_vcpu *vcpu)
74 {
75 u64 val;
76
77 /*
78 * We are about to set CPTR_EL2.TFP to trap all floating point
79 * register accesses to EL2, however, the ARM ARM clearly states that
80 * traps are only taken to EL2 if the operation would not otherwise
81 * trap to EL1. Therefore, always make sure that for 32-bit guests,
82 * we set FPEXC.EN to prevent traps to EL1, when setting the TFP bit.
83 * If FP/ASIMD is not implemented, FPEXC is UNDEFINED and any access to
84 * it will cause an exception.
85 */
86 val = vcpu->arch.hcr_el2;
87 if (!(val & HCR_RW) && system_supports_fpsimd()) {
88 write_sysreg(1 << 30, fpexc32_el2);
89 isb();
90 }
91 write_sysreg(val, hcr_el2);
92 /* Trap on AArch32 cp15 c15 accesses (EL1 or EL0) */
93 write_sysreg(1 << 15, hstr_el2);
94 /*
95 * Make sure we trap PMU access from EL0 to EL2. Also sanitize
96 * PMSELR_EL0 to make sure it never contains the cycle
97 * counter, which could make a PMXEVCNTR_EL0 access UNDEF at
98 * EL1 instead of being trapped to EL2.
99 */
100 write_sysreg(0, pmselr_el0);
101 write_sysreg(ARMV8_PMU_USERENR_MASK, pmuserenr_el0);
102 write_sysreg(vcpu->arch.mdcr_el2, mdcr_el2);
103 __activate_traps_arch()();
104 }
105
106 static void __hyp_text __deactivate_traps_vhe(void)
107 {
108 extern char vectors[]; /* kernel exception vectors */
109 u64 mdcr_el2 = read_sysreg(mdcr_el2);
110
111 mdcr_el2 &= MDCR_EL2_HPMN_MASK |
112 MDCR_EL2_E2PB_MASK << MDCR_EL2_E2PB_SHIFT |
113 MDCR_EL2_TPMS;
114
115 write_sysreg(mdcr_el2, mdcr_el2);
116 write_sysreg(HCR_HOST_VHE_FLAGS, hcr_el2);
117 write_sysreg(CPACR_EL1_FPEN, cpacr_el1);
118 write_sysreg(vectors, vbar_el1);
119 }
120
121 static void __hyp_text __deactivate_traps_nvhe(void)
122 {
123 u64 mdcr_el2 = read_sysreg(mdcr_el2);
124
125 mdcr_el2 &= MDCR_EL2_HPMN_MASK;
126 mdcr_el2 |= MDCR_EL2_E2PB_MASK << MDCR_EL2_E2PB_SHIFT;
127
128 write_sysreg(mdcr_el2, mdcr_el2);
129 write_sysreg(HCR_RW, hcr_el2);
130 write_sysreg(CPTR_EL2_DEFAULT, cptr_el2);
131 }
132
133 static hyp_alternate_select(__deactivate_traps_arch,
134 __deactivate_traps_nvhe, __deactivate_traps_vhe,
135 ARM64_HAS_VIRT_HOST_EXTN);
136
137 static void __hyp_text __deactivate_traps(struct kvm_vcpu *vcpu)
138 {
139 /*
140 * If we pended a virtual abort, preserve it until it gets
141 * cleared. See D1.14.3 (Virtual Interrupts) for details, but
142 * the crucial bit is "On taking a vSError interrupt,
143 * HCR_EL2.VSE is cleared to 0."
144 */
145 if (vcpu->arch.hcr_el2 & HCR_VSE)
146 vcpu->arch.hcr_el2 = read_sysreg(hcr_el2);
147
148 __deactivate_traps_arch()();
149 write_sysreg(0, hstr_el2);
150 write_sysreg(0, pmuserenr_el0);
151 }
152
153 static void __hyp_text __activate_vm(struct kvm_vcpu *vcpu)
154 {
155 struct kvm *kvm = kern_hyp_va(vcpu->kvm);
156 write_sysreg(kvm->arch.vttbr, vttbr_el2);
157 }
158
159 static void __hyp_text __deactivate_vm(struct kvm_vcpu *vcpu)
160 {
161 write_sysreg(0, vttbr_el2);
162 }
163
164 static void __hyp_text __vgic_save_state(struct kvm_vcpu *vcpu)
165 {
166 if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif))
167 __vgic_v3_save_state(vcpu);
168 else
169 __vgic_v2_save_state(vcpu);
170
171 write_sysreg(read_sysreg(hcr_el2) & ~HCR_INT_OVERRIDE, hcr_el2);
172 }
173
174 static void __hyp_text __vgic_restore_state(struct kvm_vcpu *vcpu)
175 {
176 u64 val;
177
178 val = read_sysreg(hcr_el2);
179 val |= HCR_INT_OVERRIDE;
180 val |= vcpu->arch.irq_lines;
181 write_sysreg(val, hcr_el2);
182
183 if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif))
184 __vgic_v3_restore_state(vcpu);
185 else
186 __vgic_v2_restore_state(vcpu);
187 }
188
189 static bool __hyp_text __true_value(void)
190 {
191 return true;
192 }
193
194 static bool __hyp_text __false_value(void)
195 {
196 return false;
197 }
198
199 static hyp_alternate_select(__check_arm_834220,
200 __false_value, __true_value,
201 ARM64_WORKAROUND_834220);
202
203 static bool __hyp_text __translate_far_to_hpfar(u64 far, u64 *hpfar)
204 {
205 u64 par, tmp;
206
207 /*
208 * Resolve the IPA the hard way using the guest VA.
209 *
210 * Stage-1 translation already validated the memory access
211 * rights. As such, we can use the EL1 translation regime, and
212 * don't have to distinguish between EL0 and EL1 access.
213 *
214 * We do need to save/restore PAR_EL1 though, as we haven't
215 * saved the guest context yet, and we may return early...
216 */
217 par = read_sysreg(par_el1);
218 asm volatile("at s1e1r, %0" : : "r" (far));
219 isb();
220
221 tmp = read_sysreg(par_el1);
222 write_sysreg(par, par_el1);
223
224 if (unlikely(tmp & 1))
225 return false; /* Translation failed, back to guest */
226
227 /* Convert PAR to HPFAR format */
228 *hpfar = ((tmp >> 12) & ((1UL << 36) - 1)) << 4;
229 return true;
230 }
231
232 static bool __hyp_text __populate_fault_info(struct kvm_vcpu *vcpu)
233 {
234 u64 esr = read_sysreg_el2(esr);
235 u8 ec = ESR_ELx_EC(esr);
236 u64 hpfar, far;
237
238 vcpu->arch.fault.esr_el2 = esr;
239
240 if (ec != ESR_ELx_EC_DABT_LOW && ec != ESR_ELx_EC_IABT_LOW)
241 return true;
242
243 far = read_sysreg_el2(far);
244
245 /*
246 * The HPFAR can be invalid if the stage 2 fault did not
247 * happen during a stage 1 page table walk (the ESR_EL2.S1PTW
248 * bit is clear) and one of the two following cases are true:
249 * 1. The fault was due to a permission fault
250 * 2. The processor carries errata 834220
251 *
252 * Therefore, for all non S1PTW faults where we either have a
253 * permission fault or the errata workaround is enabled, we
254 * resolve the IPA using the AT instruction.
255 */
256 if (!(esr & ESR_ELx_S1PTW) &&
257 (__check_arm_834220()() || (esr & ESR_ELx_FSC_TYPE) == FSC_PERM)) {
258 if (!__translate_far_to_hpfar(far, &hpfar))
259 return false;
260 } else {
261 hpfar = read_sysreg(hpfar_el2);
262 }
263
264 vcpu->arch.fault.far_el2 = far;
265 vcpu->arch.fault.hpfar_el2 = hpfar;
266 return true;
267 }
268
269 static void __hyp_text __skip_instr(struct kvm_vcpu *vcpu)
270 {
271 *vcpu_pc(vcpu) = read_sysreg_el2(elr);
272
273 if (vcpu_mode_is_32bit(vcpu)) {
274 vcpu->arch.ctxt.gp_regs.regs.pstate = read_sysreg_el2(spsr);
275 kvm_skip_instr32(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
276 write_sysreg_el2(vcpu->arch.ctxt.gp_regs.regs.pstate, spsr);
277 } else {
278 *vcpu_pc(vcpu) += 4;
279 }
280
281 write_sysreg_el2(*vcpu_pc(vcpu), elr);
282 }
283
284 int __hyp_text __kvm_vcpu_run(struct kvm_vcpu *vcpu)
285 {
286 struct kvm_cpu_context *host_ctxt;
287 struct kvm_cpu_context *guest_ctxt;
288 bool fp_enabled;
289 u64 exit_code;
290
291 vcpu = kern_hyp_va(vcpu);
292 write_sysreg(vcpu, tpidr_el2);
293
294 host_ctxt = kern_hyp_va(vcpu->arch.host_cpu_context);
295 guest_ctxt = &vcpu->arch.ctxt;
296
297 __sysreg_save_host_state(host_ctxt);
298 __debug_cond_save_host_state(vcpu);
299
300 __activate_traps(vcpu);
301 __activate_vm(vcpu);
302
303 __vgic_restore_state(vcpu);
304 __timer_restore_state(vcpu);
305
306 /*
307 * We must restore the 32-bit state before the sysregs, thanks
308 * to erratum #852523 (Cortex-A57) or #853709 (Cortex-A72).
309 */
310 __sysreg32_restore_state(vcpu);
311 __sysreg_restore_guest_state(guest_ctxt);
312 __debug_restore_state(vcpu, kern_hyp_va(vcpu->arch.debug_ptr), guest_ctxt);
313
314 /* Jump in the fire! */
315 again:
316 exit_code = __guest_enter(vcpu, host_ctxt);
317 /* And we're baaack! */
318
319 /*
320 * We're using the raw exception code in order to only process
321 * the trap if no SError is pending. We will come back to the
322 * same PC once the SError has been injected, and replay the
323 * trapping instruction.
324 */
325 if (exit_code == ARM_EXCEPTION_TRAP && !__populate_fault_info(vcpu))
326 goto again;
327
328 if (static_branch_unlikely(&vgic_v2_cpuif_trap) &&
329 exit_code == ARM_EXCEPTION_TRAP) {
330 bool valid;
331
332 valid = kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_DABT_LOW &&
333 kvm_vcpu_trap_get_fault_type(vcpu) == FSC_FAULT &&
334 kvm_vcpu_dabt_isvalid(vcpu) &&
335 !kvm_vcpu_dabt_isextabt(vcpu) &&
336 !kvm_vcpu_dabt_iss1tw(vcpu);
337
338 if (valid) {
339 int ret = __vgic_v2_perform_cpuif_access(vcpu);
340
341 if (ret == 1) {
342 __skip_instr(vcpu);
343 goto again;
344 }
345
346 if (ret == -1) {
347 /* Promote an illegal access to an SError */
348 __skip_instr(vcpu);
349 exit_code = ARM_EXCEPTION_EL1_SERROR;
350 }
351
352 /* 0 falls through to be handler out of EL2 */
353 }
354 }
355
356 if (static_branch_unlikely(&vgic_v3_cpuif_trap) &&
357 exit_code == ARM_EXCEPTION_TRAP &&
358 (kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_SYS64 ||
359 kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_CP15_32)) {
360 int ret = __vgic_v3_perform_cpuif_access(vcpu);
361
362 if (ret == 1) {
363 __skip_instr(vcpu);
364 goto again;
365 }
366
367 /* 0 falls through to be handled out of EL2 */
368 }
369
370 if (cpus_have_const_cap(ARM64_HARDEN_BP_POST_GUEST_EXIT)) {
371 u32 midr = read_cpuid_id();
372
373 /* Apply BTAC predictors mitigation to all Falkor chips */
374 if ((midr & MIDR_CPU_MODEL_MASK) == MIDR_QCOM_FALKOR_V1)
375 __qcom_hyp_sanitize_btac_predictors();
376 }
377
378 fp_enabled = __fpsimd_enabled();
379
380 __sysreg_save_guest_state(guest_ctxt);
381 __sysreg32_save_state(vcpu);
382 __timer_save_state(vcpu);
383 __vgic_save_state(vcpu);
384
385 __deactivate_traps(vcpu);
386 __deactivate_vm(vcpu);
387
388 __sysreg_restore_host_state(host_ctxt);
389
390 if (fp_enabled) {
391 __fpsimd_save_state(&guest_ctxt->gp_regs.fp_regs);
392 __fpsimd_restore_state(&host_ctxt->gp_regs.fp_regs);
393 }
394
395 __debug_save_state(vcpu, kern_hyp_va(vcpu->arch.debug_ptr), guest_ctxt);
396 /*
397 * This must come after restoring the host sysregs, since a non-VHE
398 * system may enable SPE here and make use of the TTBRs.
399 */
400 __debug_cond_restore_host_state(vcpu);
401
402 return exit_code;
403 }
404
405 static const char __hyp_panic_string[] = "HYP panic:\nPS:%08llx PC:%016llx ESR:%08llx\nFAR:%016llx HPFAR:%016llx PAR:%016llx\nVCPU:%p\n";
406
407 static void __hyp_text __hyp_call_panic_nvhe(u64 spsr, u64 elr, u64 par)
408 {
409 unsigned long str_va;
410
411 /*
412 * Force the panic string to be loaded from the literal pool,
413 * making sure it is a kernel address and not a PC-relative
414 * reference.
415 */
416 asm volatile("ldr %0, =__hyp_panic_string" : "=r" (str_va));
417
418 __hyp_do_panic(str_va,
419 spsr, elr,
420 read_sysreg(esr_el2), read_sysreg_el2(far),
421 read_sysreg(hpfar_el2), par,
422 (void *)read_sysreg(tpidr_el2));
423 }
424
425 static void __hyp_text __hyp_call_panic_vhe(u64 spsr, u64 elr, u64 par)
426 {
427 panic(__hyp_panic_string,
428 spsr, elr,
429 read_sysreg_el2(esr), read_sysreg_el2(far),
430 read_sysreg(hpfar_el2), par,
431 (void *)read_sysreg(tpidr_el2));
432 }
433
434 static hyp_alternate_select(__hyp_call_panic,
435 __hyp_call_panic_nvhe, __hyp_call_panic_vhe,
436 ARM64_HAS_VIRT_HOST_EXTN);
437
438 void __hyp_text __noreturn __hyp_panic(void)
439 {
440 u64 spsr = read_sysreg_el2(spsr);
441 u64 elr = read_sysreg_el2(elr);
442 u64 par = read_sysreg(par_el1);
443
444 if (read_sysreg(vttbr_el2)) {
445 struct kvm_vcpu *vcpu;
446 struct kvm_cpu_context *host_ctxt;
447
448 vcpu = (struct kvm_vcpu *)read_sysreg(tpidr_el2);
449 host_ctxt = kern_hyp_va(vcpu->arch.host_cpu_context);
450 __timer_save_state(vcpu);
451 __deactivate_traps(vcpu);
452 __deactivate_vm(vcpu);
453 __sysreg_restore_host_state(host_ctxt);
454 }
455
456 /* Call panic for real */
457 __hyp_call_panic()(spsr, elr, par);
458
459 unreachable();
460 }