]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - arch/arm64/kvm/hyp/switch.c
67f0ee24a406f901783b5e07b37939028474ac4e
[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 <asm/kvm_asm.h>
23 #include <asm/kvm_emulate.h>
24 #include <asm/kvm_hyp.h>
25 #include <asm/fpsimd.h>
26
27 static bool __hyp_text __fpsimd_enabled_nvhe(void)
28 {
29 return !(read_sysreg(cptr_el2) & CPTR_EL2_TFP);
30 }
31
32 static bool __hyp_text __fpsimd_enabled_vhe(void)
33 {
34 return !!(read_sysreg(cpacr_el1) & CPACR_EL1_FPEN);
35 }
36
37 static hyp_alternate_select(__fpsimd_is_enabled,
38 __fpsimd_enabled_nvhe, __fpsimd_enabled_vhe,
39 ARM64_HAS_VIRT_HOST_EXTN);
40
41 bool __hyp_text __fpsimd_enabled(void)
42 {
43 return __fpsimd_is_enabled()();
44 }
45
46 static void __hyp_text __activate_traps_vhe(void)
47 {
48 u64 val;
49
50 val = read_sysreg(cpacr_el1);
51 val |= CPACR_EL1_TTA;
52 val &= ~CPACR_EL1_FPEN;
53 write_sysreg(val, cpacr_el1);
54
55 write_sysreg(kvm_get_hyp_vector(), vbar_el1);
56 }
57
58 static void __hyp_text __activate_traps_nvhe(void)
59 {
60 u64 val;
61
62 val = CPTR_EL2_DEFAULT;
63 val |= CPTR_EL2_TTA | CPTR_EL2_TFP;
64 write_sysreg(val, cptr_el2);
65 }
66
67 static hyp_alternate_select(__activate_traps_arch,
68 __activate_traps_nvhe, __activate_traps_vhe,
69 ARM64_HAS_VIRT_HOST_EXTN);
70
71 static void __hyp_text __activate_traps(struct kvm_vcpu *vcpu)
72 {
73 u64 val;
74
75 /*
76 * We are about to set CPTR_EL2.TFP to trap all floating point
77 * register accesses to EL2, however, the ARM ARM clearly states that
78 * traps are only taken to EL2 if the operation would not otherwise
79 * trap to EL1. Therefore, always make sure that for 32-bit guests,
80 * we set FPEXC.EN to prevent traps to EL1, when setting the TFP bit.
81 * If FP/ASIMD is not implemented, FPEXC is UNDEFINED and any access to
82 * it will cause an exception.
83 */
84 val = vcpu->arch.hcr_el2;
85 if (!(val & HCR_RW) && system_supports_fpsimd()) {
86 write_sysreg(1 << 30, fpexc32_el2);
87 isb();
88 }
89 write_sysreg(val, hcr_el2);
90 /* Trap on AArch32 cp15 c15 accesses (EL1 or EL0) */
91 write_sysreg(1 << 15, hstr_el2);
92 /*
93 * Make sure we trap PMU access from EL0 to EL2. Also sanitize
94 * PMSELR_EL0 to make sure it never contains the cycle
95 * counter, which could make a PMXEVCNTR_EL0 access UNDEF at
96 * EL1 instead of being trapped to EL2.
97 */
98 write_sysreg(0, pmselr_el0);
99 write_sysreg(ARMV8_PMU_USERENR_MASK, pmuserenr_el0);
100 write_sysreg(vcpu->arch.mdcr_el2, mdcr_el2);
101 __activate_traps_arch()();
102 }
103
104 static void __hyp_text __deactivate_traps_vhe(void)
105 {
106 extern char vectors[]; /* kernel exception vectors */
107 u64 mdcr_el2 = read_sysreg(mdcr_el2);
108
109 mdcr_el2 &= MDCR_EL2_HPMN_MASK |
110 MDCR_EL2_E2PB_MASK << MDCR_EL2_E2PB_SHIFT |
111 MDCR_EL2_TPMS;
112
113 write_sysreg(mdcr_el2, mdcr_el2);
114 write_sysreg(HCR_HOST_VHE_FLAGS, hcr_el2);
115 write_sysreg(CPACR_EL1_FPEN, cpacr_el1);
116 write_sysreg(vectors, vbar_el1);
117 }
118
119 static void __hyp_text __deactivate_traps_nvhe(void)
120 {
121 u64 mdcr_el2 = read_sysreg(mdcr_el2);
122
123 mdcr_el2 &= MDCR_EL2_HPMN_MASK;
124 mdcr_el2 |= MDCR_EL2_E2PB_MASK << MDCR_EL2_E2PB_SHIFT;
125
126 write_sysreg(mdcr_el2, mdcr_el2);
127 write_sysreg(HCR_RW, hcr_el2);
128 write_sysreg(CPTR_EL2_DEFAULT, cptr_el2);
129 }
130
131 static hyp_alternate_select(__deactivate_traps_arch,
132 __deactivate_traps_nvhe, __deactivate_traps_vhe,
133 ARM64_HAS_VIRT_HOST_EXTN);
134
135 static void __hyp_text __deactivate_traps(struct kvm_vcpu *vcpu)
136 {
137 /*
138 * If we pended a virtual abort, preserve it until it gets
139 * cleared. See D1.14.3 (Virtual Interrupts) for details, but
140 * the crucial bit is "On taking a vSError interrupt,
141 * HCR_EL2.VSE is cleared to 0."
142 */
143 if (vcpu->arch.hcr_el2 & HCR_VSE)
144 vcpu->arch.hcr_el2 = read_sysreg(hcr_el2);
145
146 __deactivate_traps_arch()();
147 write_sysreg(0, hstr_el2);
148 write_sysreg(0, pmuserenr_el0);
149 }
150
151 static void __hyp_text __activate_vm(struct kvm_vcpu *vcpu)
152 {
153 struct kvm *kvm = kern_hyp_va(vcpu->kvm);
154 write_sysreg(kvm->arch.vttbr, vttbr_el2);
155 }
156
157 static void __hyp_text __deactivate_vm(struct kvm_vcpu *vcpu)
158 {
159 write_sysreg(0, vttbr_el2);
160 }
161
162 static void __hyp_text __vgic_save_state(struct kvm_vcpu *vcpu)
163 {
164 if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif))
165 __vgic_v3_save_state(vcpu);
166 else
167 __vgic_v2_save_state(vcpu);
168
169 write_sysreg(read_sysreg(hcr_el2) & ~HCR_INT_OVERRIDE, hcr_el2);
170 }
171
172 static void __hyp_text __vgic_restore_state(struct kvm_vcpu *vcpu)
173 {
174 u64 val;
175
176 val = read_sysreg(hcr_el2);
177 val |= HCR_INT_OVERRIDE;
178 val |= vcpu->arch.irq_lines;
179 write_sysreg(val, hcr_el2);
180
181 if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif))
182 __vgic_v3_restore_state(vcpu);
183 else
184 __vgic_v2_restore_state(vcpu);
185 }
186
187 static bool __hyp_text __true_value(void)
188 {
189 return true;
190 }
191
192 static bool __hyp_text __false_value(void)
193 {
194 return false;
195 }
196
197 static hyp_alternate_select(__check_arm_834220,
198 __false_value, __true_value,
199 ARM64_WORKAROUND_834220);
200
201 static bool __hyp_text __translate_far_to_hpfar(u64 far, u64 *hpfar)
202 {
203 u64 par, tmp;
204
205 /*
206 * Resolve the IPA the hard way using the guest VA.
207 *
208 * Stage-1 translation already validated the memory access
209 * rights. As such, we can use the EL1 translation regime, and
210 * don't have to distinguish between EL0 and EL1 access.
211 *
212 * We do need to save/restore PAR_EL1 though, as we haven't
213 * saved the guest context yet, and we may return early...
214 */
215 par = read_sysreg(par_el1);
216 asm volatile("at s1e1r, %0" : : "r" (far));
217 isb();
218
219 tmp = read_sysreg(par_el1);
220 write_sysreg(par, par_el1);
221
222 if (unlikely(tmp & 1))
223 return false; /* Translation failed, back to guest */
224
225 /* Convert PAR to HPFAR format */
226 *hpfar = ((tmp >> 12) & ((1UL << 36) - 1)) << 4;
227 return true;
228 }
229
230 static bool __hyp_text __populate_fault_info(struct kvm_vcpu *vcpu)
231 {
232 u64 esr = read_sysreg_el2(esr);
233 u8 ec = ESR_ELx_EC(esr);
234 u64 hpfar, far;
235
236 vcpu->arch.fault.esr_el2 = esr;
237
238 if (ec != ESR_ELx_EC_DABT_LOW && ec != ESR_ELx_EC_IABT_LOW)
239 return true;
240
241 far = read_sysreg_el2(far);
242
243 /*
244 * The HPFAR can be invalid if the stage 2 fault did not
245 * happen during a stage 1 page table walk (the ESR_EL2.S1PTW
246 * bit is clear) and one of the two following cases are true:
247 * 1. The fault was due to a permission fault
248 * 2. The processor carries errata 834220
249 *
250 * Therefore, for all non S1PTW faults where we either have a
251 * permission fault or the errata workaround is enabled, we
252 * resolve the IPA using the AT instruction.
253 */
254 if (!(esr & ESR_ELx_S1PTW) &&
255 (__check_arm_834220()() || (esr & ESR_ELx_FSC_TYPE) == FSC_PERM)) {
256 if (!__translate_far_to_hpfar(far, &hpfar))
257 return false;
258 } else {
259 hpfar = read_sysreg(hpfar_el2);
260 }
261
262 vcpu->arch.fault.far_el2 = far;
263 vcpu->arch.fault.hpfar_el2 = hpfar;
264 return true;
265 }
266
267 static void __hyp_text __skip_instr(struct kvm_vcpu *vcpu)
268 {
269 *vcpu_pc(vcpu) = read_sysreg_el2(elr);
270
271 if (vcpu_mode_is_32bit(vcpu)) {
272 vcpu->arch.ctxt.gp_regs.regs.pstate = read_sysreg_el2(spsr);
273 kvm_skip_instr32(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
274 write_sysreg_el2(vcpu->arch.ctxt.gp_regs.regs.pstate, spsr);
275 } else {
276 *vcpu_pc(vcpu) += 4;
277 }
278
279 write_sysreg_el2(*vcpu_pc(vcpu), elr);
280 }
281
282 int __hyp_text __kvm_vcpu_run(struct kvm_vcpu *vcpu)
283 {
284 struct kvm_cpu_context *host_ctxt;
285 struct kvm_cpu_context *guest_ctxt;
286 bool fp_enabled;
287 u64 exit_code;
288
289 vcpu = kern_hyp_va(vcpu);
290 write_sysreg(vcpu, tpidr_el2);
291
292 host_ctxt = kern_hyp_va(vcpu->arch.host_cpu_context);
293 guest_ctxt = &vcpu->arch.ctxt;
294
295 __sysreg_save_host_state(host_ctxt);
296 __debug_cond_save_host_state(vcpu);
297
298 __activate_traps(vcpu);
299 __activate_vm(vcpu);
300
301 __vgic_restore_state(vcpu);
302 __timer_restore_state(vcpu);
303
304 /*
305 * We must restore the 32-bit state before the sysregs, thanks
306 * to erratum #852523 (Cortex-A57) or #853709 (Cortex-A72).
307 */
308 __sysreg32_restore_state(vcpu);
309 __sysreg_restore_guest_state(guest_ctxt);
310 __debug_restore_state(vcpu, kern_hyp_va(vcpu->arch.debug_ptr), guest_ctxt);
311
312 /* Jump in the fire! */
313 again:
314 exit_code = __guest_enter(vcpu, host_ctxt);
315 /* And we're baaack! */
316
317 /*
318 * We're using the raw exception code in order to only process
319 * the trap if no SError is pending. We will come back to the
320 * same PC once the SError has been injected, and replay the
321 * trapping instruction.
322 */
323 if (exit_code == ARM_EXCEPTION_TRAP && !__populate_fault_info(vcpu))
324 goto again;
325
326 if (exit_code == ARM_EXCEPTION_TRAP &&
327 (kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_HVC64 ||
328 kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_HVC32) &&
329 vcpu_get_reg(vcpu, 0) == PSCI_0_2_FN_PSCI_VERSION) {
330 u64 val = PSCI_RET_NOT_SUPPORTED;
331 if (test_bit(KVM_ARM_VCPU_PSCI_0_2, vcpu->arch.features))
332 val = 2;
333
334 vcpu_set_reg(vcpu, 0, val);
335 goto again;
336 }
337
338 if (static_branch_unlikely(&vgic_v2_cpuif_trap) &&
339 exit_code == ARM_EXCEPTION_TRAP) {
340 bool valid;
341
342 valid = kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_DABT_LOW &&
343 kvm_vcpu_trap_get_fault_type(vcpu) == FSC_FAULT &&
344 kvm_vcpu_dabt_isvalid(vcpu) &&
345 !kvm_vcpu_dabt_isextabt(vcpu) &&
346 !kvm_vcpu_dabt_iss1tw(vcpu);
347
348 if (valid) {
349 int ret = __vgic_v2_perform_cpuif_access(vcpu);
350
351 if (ret == 1) {
352 __skip_instr(vcpu);
353 goto again;
354 }
355
356 if (ret == -1) {
357 /* Promote an illegal access to an SError */
358 __skip_instr(vcpu);
359 exit_code = ARM_EXCEPTION_EL1_SERROR;
360 }
361
362 /* 0 falls through to be handler out of EL2 */
363 }
364 }
365
366 if (static_branch_unlikely(&vgic_v3_cpuif_trap) &&
367 exit_code == ARM_EXCEPTION_TRAP &&
368 (kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_SYS64 ||
369 kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_CP15_32)) {
370 int ret = __vgic_v3_perform_cpuif_access(vcpu);
371
372 if (ret == 1) {
373 __skip_instr(vcpu);
374 goto again;
375 }
376
377 /* 0 falls through to be handled out of EL2 */
378 }
379
380 if (cpus_have_const_cap(ARM64_HARDEN_BP_POST_GUEST_EXIT)) {
381 u32 midr = read_cpuid_id();
382
383 /* Apply BTAC predictors mitigation to all Falkor chips */
384 if ((midr & MIDR_CPU_MODEL_MASK) == MIDR_QCOM_FALKOR_V1)
385 __qcom_hyp_sanitize_btac_predictors();
386 }
387
388 fp_enabled = __fpsimd_enabled();
389
390 __sysreg_save_guest_state(guest_ctxt);
391 __sysreg32_save_state(vcpu);
392 __timer_save_state(vcpu);
393 __vgic_save_state(vcpu);
394
395 __deactivate_traps(vcpu);
396 __deactivate_vm(vcpu);
397
398 __sysreg_restore_host_state(host_ctxt);
399
400 if (fp_enabled) {
401 __fpsimd_save_state(&guest_ctxt->gp_regs.fp_regs);
402 __fpsimd_restore_state(&host_ctxt->gp_regs.fp_regs);
403 }
404
405 __debug_save_state(vcpu, kern_hyp_va(vcpu->arch.debug_ptr), guest_ctxt);
406 /*
407 * This must come after restoring the host sysregs, since a non-VHE
408 * system may enable SPE here and make use of the TTBRs.
409 */
410 __debug_cond_restore_host_state(vcpu);
411
412 return exit_code;
413 }
414
415 static const char __hyp_panic_string[] = "HYP panic:\nPS:%08llx PC:%016llx ESR:%08llx\nFAR:%016llx HPFAR:%016llx PAR:%016llx\nVCPU:%p\n";
416
417 static void __hyp_text __hyp_call_panic_nvhe(u64 spsr, u64 elr, u64 par)
418 {
419 unsigned long str_va;
420
421 /*
422 * Force the panic string to be loaded from the literal pool,
423 * making sure it is a kernel address and not a PC-relative
424 * reference.
425 */
426 asm volatile("ldr %0, =__hyp_panic_string" : "=r" (str_va));
427
428 __hyp_do_panic(str_va,
429 spsr, elr,
430 read_sysreg(esr_el2), read_sysreg_el2(far),
431 read_sysreg(hpfar_el2), par,
432 (void *)read_sysreg(tpidr_el2));
433 }
434
435 static void __hyp_text __hyp_call_panic_vhe(u64 spsr, u64 elr, u64 par)
436 {
437 panic(__hyp_panic_string,
438 spsr, elr,
439 read_sysreg_el2(esr), read_sysreg_el2(far),
440 read_sysreg(hpfar_el2), par,
441 (void *)read_sysreg(tpidr_el2));
442 }
443
444 static hyp_alternate_select(__hyp_call_panic,
445 __hyp_call_panic_nvhe, __hyp_call_panic_vhe,
446 ARM64_HAS_VIRT_HOST_EXTN);
447
448 void __hyp_text __noreturn __hyp_panic(void)
449 {
450 u64 spsr = read_sysreg_el2(spsr);
451 u64 elr = read_sysreg_el2(elr);
452 u64 par = read_sysreg(par_el1);
453
454 if (read_sysreg(vttbr_el2)) {
455 struct kvm_vcpu *vcpu;
456 struct kvm_cpu_context *host_ctxt;
457
458 vcpu = (struct kvm_vcpu *)read_sysreg(tpidr_el2);
459 host_ctxt = kern_hyp_va(vcpu->arch.host_cpu_context);
460 __timer_save_state(vcpu);
461 __deactivate_traps(vcpu);
462 __deactivate_vm(vcpu);
463 __sysreg_restore_host_state(host_ctxt);
464 }
465
466 /* Call panic for real */
467 __hyp_call_panic()(spsr, elr, par);
468
469 unreachable();
470 }