Merge tag 'erofs-for-5.7-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/xiang...
[linux-block.git] / arch / powerpc / kvm / powerpc.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *
4  * Copyright IBM Corp. 2007
5  *
6  * Authors: Hollis Blanchard <hollisb@us.ibm.com>
7  *          Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
8  */
9
10 #include <linux/errno.h>
11 #include <linux/err.h>
12 #include <linux/kvm_host.h>
13 #include <linux/vmalloc.h>
14 #include <linux/hrtimer.h>
15 #include <linux/sched/signal.h>
16 #include <linux/fs.h>
17 #include <linux/slab.h>
18 #include <linux/file.h>
19 #include <linux/module.h>
20 #include <linux/irqbypass.h>
21 #include <linux/kvm_irqfd.h>
22 #include <asm/cputable.h>
23 #include <linux/uaccess.h>
24 #include <asm/kvm_ppc.h>
25 #include <asm/cputhreads.h>
26 #include <asm/irqflags.h>
27 #include <asm/iommu.h>
28 #include <asm/switch_to.h>
29 #include <asm/xive.h>
30 #ifdef CONFIG_PPC_PSERIES
31 #include <asm/hvcall.h>
32 #include <asm/plpar_wrappers.h>
33 #endif
34 #include <asm/ultravisor.h>
35 #include <asm/kvm_host.h>
36
37 #include "timing.h"
38 #include "irq.h"
39 #include "../mm/mmu_decl.h"
40
41 #define CREATE_TRACE_POINTS
42 #include "trace.h"
43
44 struct kvmppc_ops *kvmppc_hv_ops;
45 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
46 struct kvmppc_ops *kvmppc_pr_ops;
47 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
48
49
50 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
51 {
52         return !!(v->arch.pending_exceptions) || kvm_request_pending(v);
53 }
54
55 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
56 {
57         return kvm_arch_vcpu_runnable(vcpu);
58 }
59
60 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
61 {
62         return false;
63 }
64
65 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
66 {
67         return 1;
68 }
69
70 /*
71  * Common checks before entering the guest world.  Call with interrupts
72  * disabled.
73  *
74  * returns:
75  *
76  * == 1 if we're ready to go into guest state
77  * <= 0 if we need to go back to the host with return value
78  */
79 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
80 {
81         int r;
82
83         WARN_ON(irqs_disabled());
84         hard_irq_disable();
85
86         while (true) {
87                 if (need_resched()) {
88                         local_irq_enable();
89                         cond_resched();
90                         hard_irq_disable();
91                         continue;
92                 }
93
94                 if (signal_pending(current)) {
95                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
96                         vcpu->run->exit_reason = KVM_EXIT_INTR;
97                         r = -EINTR;
98                         break;
99                 }
100
101                 vcpu->mode = IN_GUEST_MODE;
102
103                 /*
104                  * Reading vcpu->requests must happen after setting vcpu->mode,
105                  * so we don't miss a request because the requester sees
106                  * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
107                  * before next entering the guest (and thus doesn't IPI).
108                  * This also orders the write to mode from any reads
109                  * to the page tables done while the VCPU is running.
110                  * Please see the comment in kvm_flush_remote_tlbs.
111                  */
112                 smp_mb();
113
114                 if (kvm_request_pending(vcpu)) {
115                         /* Make sure we process requests preemptable */
116                         local_irq_enable();
117                         trace_kvm_check_requests(vcpu);
118                         r = kvmppc_core_check_requests(vcpu);
119                         hard_irq_disable();
120                         if (r > 0)
121                                 continue;
122                         break;
123                 }
124
125                 if (kvmppc_core_prepare_to_enter(vcpu)) {
126                         /* interrupts got enabled in between, so we
127                            are back at square 1 */
128                         continue;
129                 }
130
131                 guest_enter_irqoff();
132                 return 1;
133         }
134
135         /* return to host */
136         local_irq_enable();
137         return r;
138 }
139 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
140
141 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
142 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
143 {
144         struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
145         int i;
146
147         shared->sprg0 = swab64(shared->sprg0);
148         shared->sprg1 = swab64(shared->sprg1);
149         shared->sprg2 = swab64(shared->sprg2);
150         shared->sprg3 = swab64(shared->sprg3);
151         shared->srr0 = swab64(shared->srr0);
152         shared->srr1 = swab64(shared->srr1);
153         shared->dar = swab64(shared->dar);
154         shared->msr = swab64(shared->msr);
155         shared->dsisr = swab32(shared->dsisr);
156         shared->int_pending = swab32(shared->int_pending);
157         for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
158                 shared->sr[i] = swab32(shared->sr[i]);
159 }
160 #endif
161
162 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
163 {
164         int nr = kvmppc_get_gpr(vcpu, 11);
165         int r;
166         unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
167         unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
168         unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
169         unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
170         unsigned long r2 = 0;
171
172         if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
173                 /* 32 bit mode */
174                 param1 &= 0xffffffff;
175                 param2 &= 0xffffffff;
176                 param3 &= 0xffffffff;
177                 param4 &= 0xffffffff;
178         }
179
180         switch (nr) {
181         case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
182         {
183 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
184                 /* Book3S can be little endian, find it out here */
185                 int shared_big_endian = true;
186                 if (vcpu->arch.intr_msr & MSR_LE)
187                         shared_big_endian = false;
188                 if (shared_big_endian != vcpu->arch.shared_big_endian)
189                         kvmppc_swab_shared(vcpu);
190                 vcpu->arch.shared_big_endian = shared_big_endian;
191 #endif
192
193                 if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
194                         /*
195                          * Older versions of the Linux magic page code had
196                          * a bug where they would map their trampoline code
197                          * NX. If that's the case, remove !PR NX capability.
198                          */
199                         vcpu->arch.disable_kernel_nx = true;
200                         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
201                 }
202
203                 vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
204                 vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
205
206 #ifdef CONFIG_PPC_64K_PAGES
207                 /*
208                  * Make sure our 4k magic page is in the same window of a 64k
209                  * page within the guest and within the host's page.
210                  */
211                 if ((vcpu->arch.magic_page_pa & 0xf000) !=
212                     ((ulong)vcpu->arch.shared & 0xf000)) {
213                         void *old_shared = vcpu->arch.shared;
214                         ulong shared = (ulong)vcpu->arch.shared;
215                         void *new_shared;
216
217                         shared &= PAGE_MASK;
218                         shared |= vcpu->arch.magic_page_pa & 0xf000;
219                         new_shared = (void*)shared;
220                         memcpy(new_shared, old_shared, 0x1000);
221                         vcpu->arch.shared = new_shared;
222                 }
223 #endif
224
225                 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
226
227                 r = EV_SUCCESS;
228                 break;
229         }
230         case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
231                 r = EV_SUCCESS;
232 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
233                 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
234 #endif
235
236                 /* Second return value is in r4 */
237                 break;
238         case EV_HCALL_TOKEN(EV_IDLE):
239                 r = EV_SUCCESS;
240                 kvm_vcpu_block(vcpu);
241                 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
242                 break;
243         default:
244                 r = EV_UNIMPLEMENTED;
245                 break;
246         }
247
248         kvmppc_set_gpr(vcpu, 4, r2);
249
250         return r;
251 }
252 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
253
254 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
255 {
256         int r = false;
257
258         /* We have to know what CPU to virtualize */
259         if (!vcpu->arch.pvr)
260                 goto out;
261
262         /* PAPR only works with book3s_64 */
263         if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
264                 goto out;
265
266         /* HV KVM can only do PAPR mode for now */
267         if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
268                 goto out;
269
270 #ifdef CONFIG_KVM_BOOKE_HV
271         if (!cpu_has_feature(CPU_FTR_EMB_HV))
272                 goto out;
273 #endif
274
275         r = true;
276
277 out:
278         vcpu->arch.sane = r;
279         return r ? 0 : -EINVAL;
280 }
281 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
282
283 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
284 {
285         enum emulation_result er;
286         int r;
287
288         er = kvmppc_emulate_loadstore(vcpu);
289         switch (er) {
290         case EMULATE_DONE:
291                 /* Future optimization: only reload non-volatiles if they were
292                  * actually modified. */
293                 r = RESUME_GUEST_NV;
294                 break;
295         case EMULATE_AGAIN:
296                 r = RESUME_GUEST;
297                 break;
298         case EMULATE_DO_MMIO:
299                 run->exit_reason = KVM_EXIT_MMIO;
300                 /* We must reload nonvolatiles because "update" load/store
301                  * instructions modify register state. */
302                 /* Future optimization: only reload non-volatiles if they were
303                  * actually modified. */
304                 r = RESUME_HOST_NV;
305                 break;
306         case EMULATE_FAIL:
307         {
308                 u32 last_inst;
309
310                 kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
311                 /* XXX Deliver Program interrupt to guest. */
312                 pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
313                 r = RESUME_HOST;
314                 break;
315         }
316         default:
317                 WARN_ON(1);
318                 r = RESUME_GUEST;
319         }
320
321         return r;
322 }
323 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
324
325 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
326               bool data)
327 {
328         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
329         struct kvmppc_pte pte;
330         int r = -EINVAL;
331
332         vcpu->stat.st++;
333
334         if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->store_to_eaddr)
335                 r = vcpu->kvm->arch.kvm_ops->store_to_eaddr(vcpu, eaddr, ptr,
336                                                             size);
337
338         if ((!r) || (r == -EAGAIN))
339                 return r;
340
341         r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
342                          XLATE_WRITE, &pte);
343         if (r < 0)
344                 return r;
345
346         *eaddr = pte.raddr;
347
348         if (!pte.may_write)
349                 return -EPERM;
350
351         /* Magic page override */
352         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
353             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
354             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
355                 void *magic = vcpu->arch.shared;
356                 magic += pte.eaddr & 0xfff;
357                 memcpy(magic, ptr, size);
358                 return EMULATE_DONE;
359         }
360
361         if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
362                 return EMULATE_DO_MMIO;
363
364         return EMULATE_DONE;
365 }
366 EXPORT_SYMBOL_GPL(kvmppc_st);
367
368 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
369                       bool data)
370 {
371         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
372         struct kvmppc_pte pte;
373         int rc = -EINVAL;
374
375         vcpu->stat.ld++;
376
377         if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->load_from_eaddr)
378                 rc = vcpu->kvm->arch.kvm_ops->load_from_eaddr(vcpu, eaddr, ptr,
379                                                               size);
380
381         if ((!rc) || (rc == -EAGAIN))
382                 return rc;
383
384         rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
385                           XLATE_READ, &pte);
386         if (rc)
387                 return rc;
388
389         *eaddr = pte.raddr;
390
391         if (!pte.may_read)
392                 return -EPERM;
393
394         if (!data && !pte.may_execute)
395                 return -ENOEXEC;
396
397         /* Magic page override */
398         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
399             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
400             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
401                 void *magic = vcpu->arch.shared;
402                 magic += pte.eaddr & 0xfff;
403                 memcpy(ptr, magic, size);
404                 return EMULATE_DONE;
405         }
406
407         if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
408                 return EMULATE_DO_MMIO;
409
410         return EMULATE_DONE;
411 }
412 EXPORT_SYMBOL_GPL(kvmppc_ld);
413
414 int kvm_arch_hardware_enable(void)
415 {
416         return 0;
417 }
418
419 int kvm_arch_hardware_setup(void)
420 {
421         return 0;
422 }
423
424 int kvm_arch_check_processor_compat(void)
425 {
426         return kvmppc_core_check_processor_compat();
427 }
428
429 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
430 {
431         struct kvmppc_ops *kvm_ops = NULL;
432         /*
433          * if we have both HV and PR enabled, default is HV
434          */
435         if (type == 0) {
436                 if (kvmppc_hv_ops)
437                         kvm_ops = kvmppc_hv_ops;
438                 else
439                         kvm_ops = kvmppc_pr_ops;
440                 if (!kvm_ops)
441                         goto err_out;
442         } else  if (type == KVM_VM_PPC_HV) {
443                 if (!kvmppc_hv_ops)
444                         goto err_out;
445                 kvm_ops = kvmppc_hv_ops;
446         } else if (type == KVM_VM_PPC_PR) {
447                 if (!kvmppc_pr_ops)
448                         goto err_out;
449                 kvm_ops = kvmppc_pr_ops;
450         } else
451                 goto err_out;
452
453         if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
454                 return -ENOENT;
455
456         kvm->arch.kvm_ops = kvm_ops;
457         return kvmppc_core_init_vm(kvm);
458 err_out:
459         return -EINVAL;
460 }
461
462 void kvm_arch_destroy_vm(struct kvm *kvm)
463 {
464         unsigned int i;
465         struct kvm_vcpu *vcpu;
466
467 #ifdef CONFIG_KVM_XICS
468         /*
469          * We call kick_all_cpus_sync() to ensure that all
470          * CPUs have executed any pending IPIs before we
471          * continue and free VCPUs structures below.
472          */
473         if (is_kvmppc_hv_enabled(kvm))
474                 kick_all_cpus_sync();
475 #endif
476
477         kvm_for_each_vcpu(i, vcpu, kvm)
478                 kvm_vcpu_destroy(vcpu);
479
480         mutex_lock(&kvm->lock);
481         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
482                 kvm->vcpus[i] = NULL;
483
484         atomic_set(&kvm->online_vcpus, 0);
485
486         kvmppc_core_destroy_vm(kvm);
487
488         mutex_unlock(&kvm->lock);
489
490         /* drop the module reference */
491         module_put(kvm->arch.kvm_ops->owner);
492 }
493
494 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
495 {
496         int r;
497         /* Assume we're using HV mode when the HV module is loaded */
498         int hv_enabled = kvmppc_hv_ops ? 1 : 0;
499
500         if (kvm) {
501                 /*
502                  * Hooray - we know which VM type we're running on. Depend on
503                  * that rather than the guess above.
504                  */
505                 hv_enabled = is_kvmppc_hv_enabled(kvm);
506         }
507
508         switch (ext) {
509 #ifdef CONFIG_BOOKE
510         case KVM_CAP_PPC_BOOKE_SREGS:
511         case KVM_CAP_PPC_BOOKE_WATCHDOG:
512         case KVM_CAP_PPC_EPR:
513 #else
514         case KVM_CAP_PPC_SEGSTATE:
515         case KVM_CAP_PPC_HIOR:
516         case KVM_CAP_PPC_PAPR:
517 #endif
518         case KVM_CAP_PPC_UNSET_IRQ:
519         case KVM_CAP_PPC_IRQ_LEVEL:
520         case KVM_CAP_ENABLE_CAP:
521         case KVM_CAP_ONE_REG:
522         case KVM_CAP_IOEVENTFD:
523         case KVM_CAP_DEVICE_CTRL:
524         case KVM_CAP_IMMEDIATE_EXIT:
525                 r = 1;
526                 break;
527         case KVM_CAP_PPC_GUEST_DEBUG_SSTEP:
528                 /* fall through */
529         case KVM_CAP_PPC_PAIRED_SINGLES:
530         case KVM_CAP_PPC_OSI:
531         case KVM_CAP_PPC_GET_PVINFO:
532 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
533         case KVM_CAP_SW_TLB:
534 #endif
535                 /* We support this only for PR */
536                 r = !hv_enabled;
537                 break;
538 #ifdef CONFIG_KVM_MPIC
539         case KVM_CAP_IRQ_MPIC:
540                 r = 1;
541                 break;
542 #endif
543
544 #ifdef CONFIG_PPC_BOOK3S_64
545         case KVM_CAP_SPAPR_TCE:
546         case KVM_CAP_SPAPR_TCE_64:
547                 r = 1;
548                 break;
549         case KVM_CAP_SPAPR_TCE_VFIO:
550                 r = !!cpu_has_feature(CPU_FTR_HVMODE);
551                 break;
552         case KVM_CAP_PPC_RTAS:
553         case KVM_CAP_PPC_FIXUP_HCALL:
554         case KVM_CAP_PPC_ENABLE_HCALL:
555 #ifdef CONFIG_KVM_XICS
556         case KVM_CAP_IRQ_XICS:
557 #endif
558         case KVM_CAP_PPC_GET_CPU_CHAR:
559                 r = 1;
560                 break;
561 #ifdef CONFIG_KVM_XIVE
562         case KVM_CAP_PPC_IRQ_XIVE:
563                 /*
564                  * We need XIVE to be enabled on the platform (implies
565                  * a POWER9 processor) and the PowerNV platform, as
566                  * nested is not yet supported.
567                  */
568                 r = xive_enabled() && !!cpu_has_feature(CPU_FTR_HVMODE) &&
569                         kvmppc_xive_native_supported();
570                 break;
571 #endif
572
573         case KVM_CAP_PPC_ALLOC_HTAB:
574                 r = hv_enabled;
575                 break;
576 #endif /* CONFIG_PPC_BOOK3S_64 */
577 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
578         case KVM_CAP_PPC_SMT:
579                 r = 0;
580                 if (kvm) {
581                         if (kvm->arch.emul_smt_mode > 1)
582                                 r = kvm->arch.emul_smt_mode;
583                         else
584                                 r = kvm->arch.smt_mode;
585                 } else if (hv_enabled) {
586                         if (cpu_has_feature(CPU_FTR_ARCH_300))
587                                 r = 1;
588                         else
589                                 r = threads_per_subcore;
590                 }
591                 break;
592         case KVM_CAP_PPC_SMT_POSSIBLE:
593                 r = 1;
594                 if (hv_enabled) {
595                         if (!cpu_has_feature(CPU_FTR_ARCH_300))
596                                 r = ((threads_per_subcore << 1) - 1);
597                         else
598                                 /* P9 can emulate dbells, so allow any mode */
599                                 r = 8 | 4 | 2 | 1;
600                 }
601                 break;
602         case KVM_CAP_PPC_RMA:
603                 r = 0;
604                 break;
605         case KVM_CAP_PPC_HWRNG:
606                 r = kvmppc_hwrng_present();
607                 break;
608         case KVM_CAP_PPC_MMU_RADIX:
609                 r = !!(hv_enabled && radix_enabled());
610                 break;
611         case KVM_CAP_PPC_MMU_HASH_V3:
612                 r = !!(hv_enabled && cpu_has_feature(CPU_FTR_ARCH_300) &&
613                        cpu_has_feature(CPU_FTR_HVMODE));
614                 break;
615         case KVM_CAP_PPC_NESTED_HV:
616                 r = !!(hv_enabled && kvmppc_hv_ops->enable_nested &&
617                        !kvmppc_hv_ops->enable_nested(NULL));
618                 break;
619 #endif
620         case KVM_CAP_SYNC_MMU:
621 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
622                 r = hv_enabled;
623 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
624                 r = 1;
625 #else
626                 r = 0;
627 #endif
628                 break;
629 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
630         case KVM_CAP_PPC_HTAB_FD:
631                 r = hv_enabled;
632                 break;
633 #endif
634         case KVM_CAP_NR_VCPUS:
635                 /*
636                  * Recommending a number of CPUs is somewhat arbitrary; we
637                  * return the number of present CPUs for -HV (since a host
638                  * will have secondary threads "offline"), and for other KVM
639                  * implementations just count online CPUs.
640                  */
641                 if (hv_enabled)
642                         r = num_present_cpus();
643                 else
644                         r = num_online_cpus();
645                 break;
646         case KVM_CAP_MAX_VCPUS:
647                 r = KVM_MAX_VCPUS;
648                 break;
649         case KVM_CAP_MAX_VCPU_ID:
650                 r = KVM_MAX_VCPU_ID;
651                 break;
652 #ifdef CONFIG_PPC_BOOK3S_64
653         case KVM_CAP_PPC_GET_SMMU_INFO:
654                 r = 1;
655                 break;
656         case KVM_CAP_SPAPR_MULTITCE:
657                 r = 1;
658                 break;
659         case KVM_CAP_SPAPR_RESIZE_HPT:
660                 r = !!hv_enabled;
661                 break;
662 #endif
663 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
664         case KVM_CAP_PPC_FWNMI:
665                 r = hv_enabled;
666                 break;
667 #endif
668 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
669         case KVM_CAP_PPC_HTM:
670                 r = !!(cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM) ||
671                      (hv_enabled && cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST));
672                 break;
673 #endif
674         default:
675                 r = 0;
676                 break;
677         }
678         return r;
679
680 }
681
682 long kvm_arch_dev_ioctl(struct file *filp,
683                         unsigned int ioctl, unsigned long arg)
684 {
685         return -EINVAL;
686 }
687
688 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
689                            struct kvm_memory_slot *dont)
690 {
691         kvmppc_core_free_memslot(kvm, free, dont);
692 }
693
694 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
695                             unsigned long npages)
696 {
697         return kvmppc_core_create_memslot(kvm, slot, npages);
698 }
699
700 int kvm_arch_prepare_memory_region(struct kvm *kvm,
701                                    struct kvm_memory_slot *memslot,
702                                    const struct kvm_userspace_memory_region *mem,
703                                    enum kvm_mr_change change)
704 {
705         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
706 }
707
708 void kvm_arch_commit_memory_region(struct kvm *kvm,
709                                    const struct kvm_userspace_memory_region *mem,
710                                    const struct kvm_memory_slot *old,
711                                    const struct kvm_memory_slot *new,
712                                    enum kvm_mr_change change)
713 {
714         kvmppc_core_commit_memory_region(kvm, mem, old, new, change);
715 }
716
717 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
718                                    struct kvm_memory_slot *slot)
719 {
720         kvmppc_core_flush_memslot(kvm, slot);
721 }
722
723 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
724 {
725         return 0;
726 }
727
728 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
729 {
730         struct kvm_vcpu *vcpu;
731
732         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
733         kvmppc_decrementer_func(vcpu);
734
735         return HRTIMER_NORESTART;
736 }
737
738 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
739 {
740         int err;
741
742         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
743         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
744         vcpu->arch.dec_expires = get_tb();
745
746 #ifdef CONFIG_KVM_EXIT_TIMING
747         mutex_init(&vcpu->arch.exit_timing_lock);
748 #endif
749         err = kvmppc_subarch_vcpu_init(vcpu);
750         if (err)
751                 return err;
752
753         err = kvmppc_core_vcpu_create(vcpu);
754         if (err)
755                 goto out_vcpu_uninit;
756
757         vcpu->arch.wqp = &vcpu->wq;
758         kvmppc_create_vcpu_debugfs(vcpu, vcpu->vcpu_id);
759         return 0;
760
761 out_vcpu_uninit:
762         kvmppc_subarch_vcpu_uninit(vcpu);
763         return err;
764 }
765
766 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
767 {
768 }
769
770 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
771 {
772         /* Make sure we're not using the vcpu anymore */
773         hrtimer_cancel(&vcpu->arch.dec_timer);
774
775         kvmppc_remove_vcpu_debugfs(vcpu);
776
777         switch (vcpu->arch.irq_type) {
778         case KVMPPC_IRQ_MPIC:
779                 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
780                 break;
781         case KVMPPC_IRQ_XICS:
782                 if (xics_on_xive())
783                         kvmppc_xive_cleanup_vcpu(vcpu);
784                 else
785                         kvmppc_xics_free_icp(vcpu);
786                 break;
787         case KVMPPC_IRQ_XIVE:
788                 kvmppc_xive_native_cleanup_vcpu(vcpu);
789                 break;
790         }
791
792         kvmppc_core_vcpu_free(vcpu);
793
794         kvmppc_subarch_vcpu_uninit(vcpu);
795 }
796
797 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
798 {
799         return kvmppc_core_pending_dec(vcpu);
800 }
801
802 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
803 {
804 #ifdef CONFIG_BOOKE
805         /*
806          * vrsave (formerly usprg0) isn't used by Linux, but may
807          * be used by the guest.
808          *
809          * On non-booke this is associated with Altivec and
810          * is handled by code in book3s.c.
811          */
812         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
813 #endif
814         kvmppc_core_vcpu_load(vcpu, cpu);
815 }
816
817 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
818 {
819         kvmppc_core_vcpu_put(vcpu);
820 #ifdef CONFIG_BOOKE
821         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
822 #endif
823 }
824
825 /*
826  * irq_bypass_add_producer and irq_bypass_del_producer are only
827  * useful if the architecture supports PCI passthrough.
828  * irq_bypass_stop and irq_bypass_start are not needed and so
829  * kvm_ops are not defined for them.
830  */
831 bool kvm_arch_has_irq_bypass(void)
832 {
833         return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
834                 (kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
835 }
836
837 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
838                                      struct irq_bypass_producer *prod)
839 {
840         struct kvm_kernel_irqfd *irqfd =
841                 container_of(cons, struct kvm_kernel_irqfd, consumer);
842         struct kvm *kvm = irqfd->kvm;
843
844         if (kvm->arch.kvm_ops->irq_bypass_add_producer)
845                 return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
846
847         return 0;
848 }
849
850 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
851                                       struct irq_bypass_producer *prod)
852 {
853         struct kvm_kernel_irqfd *irqfd =
854                 container_of(cons, struct kvm_kernel_irqfd, consumer);
855         struct kvm *kvm = irqfd->kvm;
856
857         if (kvm->arch.kvm_ops->irq_bypass_del_producer)
858                 kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
859 }
860
861 #ifdef CONFIG_VSX
862 static inline int kvmppc_get_vsr_dword_offset(int index)
863 {
864         int offset;
865
866         if ((index != 0) && (index != 1))
867                 return -1;
868
869 #ifdef __BIG_ENDIAN
870         offset =  index;
871 #else
872         offset = 1 - index;
873 #endif
874
875         return offset;
876 }
877
878 static inline int kvmppc_get_vsr_word_offset(int index)
879 {
880         int offset;
881
882         if ((index > 3) || (index < 0))
883                 return -1;
884
885 #ifdef __BIG_ENDIAN
886         offset = index;
887 #else
888         offset = 3 - index;
889 #endif
890         return offset;
891 }
892
893 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
894         u64 gpr)
895 {
896         union kvmppc_one_reg val;
897         int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
898         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
899
900         if (offset == -1)
901                 return;
902
903         if (index >= 32) {
904                 val.vval = VCPU_VSX_VR(vcpu, index - 32);
905                 val.vsxval[offset] = gpr;
906                 VCPU_VSX_VR(vcpu, index - 32) = val.vval;
907         } else {
908                 VCPU_VSX_FPR(vcpu, index, offset) = gpr;
909         }
910 }
911
912 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
913         u64 gpr)
914 {
915         union kvmppc_one_reg val;
916         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
917
918         if (index >= 32) {
919                 val.vval = VCPU_VSX_VR(vcpu, index - 32);
920                 val.vsxval[0] = gpr;
921                 val.vsxval[1] = gpr;
922                 VCPU_VSX_VR(vcpu, index - 32) = val.vval;
923         } else {
924                 VCPU_VSX_FPR(vcpu, index, 0) = gpr;
925                 VCPU_VSX_FPR(vcpu, index, 1) = gpr;
926         }
927 }
928
929 static inline void kvmppc_set_vsr_word_dump(struct kvm_vcpu *vcpu,
930         u32 gpr)
931 {
932         union kvmppc_one_reg val;
933         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
934
935         if (index >= 32) {
936                 val.vsx32val[0] = gpr;
937                 val.vsx32val[1] = gpr;
938                 val.vsx32val[2] = gpr;
939                 val.vsx32val[3] = gpr;
940                 VCPU_VSX_VR(vcpu, index - 32) = val.vval;
941         } else {
942                 val.vsx32val[0] = gpr;
943                 val.vsx32val[1] = gpr;
944                 VCPU_VSX_FPR(vcpu, index, 0) = val.vsxval[0];
945                 VCPU_VSX_FPR(vcpu, index, 1) = val.vsxval[0];
946         }
947 }
948
949 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
950         u32 gpr32)
951 {
952         union kvmppc_one_reg val;
953         int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
954         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
955         int dword_offset, word_offset;
956
957         if (offset == -1)
958                 return;
959
960         if (index >= 32) {
961                 val.vval = VCPU_VSX_VR(vcpu, index - 32);
962                 val.vsx32val[offset] = gpr32;
963                 VCPU_VSX_VR(vcpu, index - 32) = val.vval;
964         } else {
965                 dword_offset = offset / 2;
966                 word_offset = offset % 2;
967                 val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
968                 val.vsx32val[word_offset] = gpr32;
969                 VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
970         }
971 }
972 #endif /* CONFIG_VSX */
973
974 #ifdef CONFIG_ALTIVEC
975 static inline int kvmppc_get_vmx_offset_generic(struct kvm_vcpu *vcpu,
976                 int index, int element_size)
977 {
978         int offset;
979         int elts = sizeof(vector128)/element_size;
980
981         if ((index < 0) || (index >= elts))
982                 return -1;
983
984         if (kvmppc_need_byteswap(vcpu))
985                 offset = elts - index - 1;
986         else
987                 offset = index;
988
989         return offset;
990 }
991
992 static inline int kvmppc_get_vmx_dword_offset(struct kvm_vcpu *vcpu,
993                 int index)
994 {
995         return kvmppc_get_vmx_offset_generic(vcpu, index, 8);
996 }
997
998 static inline int kvmppc_get_vmx_word_offset(struct kvm_vcpu *vcpu,
999                 int index)
1000 {
1001         return kvmppc_get_vmx_offset_generic(vcpu, index, 4);
1002 }
1003
1004 static inline int kvmppc_get_vmx_hword_offset(struct kvm_vcpu *vcpu,
1005                 int index)
1006 {
1007         return kvmppc_get_vmx_offset_generic(vcpu, index, 2);
1008 }
1009
1010 static inline int kvmppc_get_vmx_byte_offset(struct kvm_vcpu *vcpu,
1011                 int index)
1012 {
1013         return kvmppc_get_vmx_offset_generic(vcpu, index, 1);
1014 }
1015
1016
1017 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu,
1018         u64 gpr)
1019 {
1020         union kvmppc_one_reg val;
1021         int offset = kvmppc_get_vmx_dword_offset(vcpu,
1022                         vcpu->arch.mmio_vmx_offset);
1023         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1024
1025         if (offset == -1)
1026                 return;
1027
1028         val.vval = VCPU_VSX_VR(vcpu, index);
1029         val.vsxval[offset] = gpr;
1030         VCPU_VSX_VR(vcpu, index) = val.vval;
1031 }
1032
1033 static inline void kvmppc_set_vmx_word(struct kvm_vcpu *vcpu,
1034         u32 gpr32)
1035 {
1036         union kvmppc_one_reg val;
1037         int offset = kvmppc_get_vmx_word_offset(vcpu,
1038                         vcpu->arch.mmio_vmx_offset);
1039         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1040
1041         if (offset == -1)
1042                 return;
1043
1044         val.vval = VCPU_VSX_VR(vcpu, index);
1045         val.vsx32val[offset] = gpr32;
1046         VCPU_VSX_VR(vcpu, index) = val.vval;
1047 }
1048
1049 static inline void kvmppc_set_vmx_hword(struct kvm_vcpu *vcpu,
1050         u16 gpr16)
1051 {
1052         union kvmppc_one_reg val;
1053         int offset = kvmppc_get_vmx_hword_offset(vcpu,
1054                         vcpu->arch.mmio_vmx_offset);
1055         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1056
1057         if (offset == -1)
1058                 return;
1059
1060         val.vval = VCPU_VSX_VR(vcpu, index);
1061         val.vsx16val[offset] = gpr16;
1062         VCPU_VSX_VR(vcpu, index) = val.vval;
1063 }
1064
1065 static inline void kvmppc_set_vmx_byte(struct kvm_vcpu *vcpu,
1066         u8 gpr8)
1067 {
1068         union kvmppc_one_reg val;
1069         int offset = kvmppc_get_vmx_byte_offset(vcpu,
1070                         vcpu->arch.mmio_vmx_offset);
1071         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1072
1073         if (offset == -1)
1074                 return;
1075
1076         val.vval = VCPU_VSX_VR(vcpu, index);
1077         val.vsx8val[offset] = gpr8;
1078         VCPU_VSX_VR(vcpu, index) = val.vval;
1079 }
1080 #endif /* CONFIG_ALTIVEC */
1081
1082 #ifdef CONFIG_PPC_FPU
1083 static inline u64 sp_to_dp(u32 fprs)
1084 {
1085         u64 fprd;
1086
1087         preempt_disable();
1088         enable_kernel_fp();
1089         asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs)
1090              : "fr0");
1091         preempt_enable();
1092         return fprd;
1093 }
1094
1095 static inline u32 dp_to_sp(u64 fprd)
1096 {
1097         u32 fprs;
1098
1099         preempt_disable();
1100         enable_kernel_fp();
1101         asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd)
1102              : "fr0");
1103         preempt_enable();
1104         return fprs;
1105 }
1106
1107 #else
1108 #define sp_to_dp(x)     (x)
1109 #define dp_to_sp(x)     (x)
1110 #endif /* CONFIG_PPC_FPU */
1111
1112 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
1113                                       struct kvm_run *run)
1114 {
1115         u64 uninitialized_var(gpr);
1116
1117         if (run->mmio.len > sizeof(gpr)) {
1118                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
1119                 return;
1120         }
1121
1122         if (!vcpu->arch.mmio_host_swabbed) {
1123                 switch (run->mmio.len) {
1124                 case 8: gpr = *(u64 *)run->mmio.data; break;
1125                 case 4: gpr = *(u32 *)run->mmio.data; break;
1126                 case 2: gpr = *(u16 *)run->mmio.data; break;
1127                 case 1: gpr = *(u8 *)run->mmio.data; break;
1128                 }
1129         } else {
1130                 switch (run->mmio.len) {
1131                 case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
1132                 case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
1133                 case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
1134                 case 1: gpr = *(u8 *)run->mmio.data; break;
1135                 }
1136         }
1137
1138         /* conversion between single and double precision */
1139         if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
1140                 gpr = sp_to_dp(gpr);
1141
1142         if (vcpu->arch.mmio_sign_extend) {
1143                 switch (run->mmio.len) {
1144 #ifdef CONFIG_PPC64
1145                 case 4:
1146                         gpr = (s64)(s32)gpr;
1147                         break;
1148 #endif
1149                 case 2:
1150                         gpr = (s64)(s16)gpr;
1151                         break;
1152                 case 1:
1153                         gpr = (s64)(s8)gpr;
1154                         break;
1155                 }
1156         }
1157
1158         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1159         case KVM_MMIO_REG_GPR:
1160                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1161                 break;
1162         case KVM_MMIO_REG_FPR:
1163                 if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1164                         vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_FP);
1165
1166                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1167                 break;
1168 #ifdef CONFIG_PPC_BOOK3S
1169         case KVM_MMIO_REG_QPR:
1170                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1171                 break;
1172         case KVM_MMIO_REG_FQPR:
1173                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1174                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1175                 break;
1176 #endif
1177 #ifdef CONFIG_VSX
1178         case KVM_MMIO_REG_VSX:
1179                 if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1180                         vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VSX);
1181
1182                 if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_DWORD)
1183                         kvmppc_set_vsr_dword(vcpu, gpr);
1184                 else if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_WORD)
1185                         kvmppc_set_vsr_word(vcpu, gpr);
1186                 else if (vcpu->arch.mmio_copy_type ==
1187                                 KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1188                         kvmppc_set_vsr_dword_dump(vcpu, gpr);
1189                 else if (vcpu->arch.mmio_copy_type ==
1190                                 KVMPPC_VSX_COPY_WORD_LOAD_DUMP)
1191                         kvmppc_set_vsr_word_dump(vcpu, gpr);
1192                 break;
1193 #endif
1194 #ifdef CONFIG_ALTIVEC
1195         case KVM_MMIO_REG_VMX:
1196                 if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1197                         vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VEC);
1198
1199                 if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_DWORD)
1200                         kvmppc_set_vmx_dword(vcpu, gpr);
1201                 else if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_WORD)
1202                         kvmppc_set_vmx_word(vcpu, gpr);
1203                 else if (vcpu->arch.mmio_copy_type ==
1204                                 KVMPPC_VMX_COPY_HWORD)
1205                         kvmppc_set_vmx_hword(vcpu, gpr);
1206                 else if (vcpu->arch.mmio_copy_type ==
1207                                 KVMPPC_VMX_COPY_BYTE)
1208                         kvmppc_set_vmx_byte(vcpu, gpr);
1209                 break;
1210 #endif
1211 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1212         case KVM_MMIO_REG_NESTED_GPR:
1213                 if (kvmppc_need_byteswap(vcpu))
1214                         gpr = swab64(gpr);
1215                 kvm_vcpu_write_guest(vcpu, vcpu->arch.nested_io_gpr, &gpr,
1216                                      sizeof(gpr));
1217                 break;
1218 #endif
1219         default:
1220                 BUG();
1221         }
1222 }
1223
1224 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1225                                 unsigned int rt, unsigned int bytes,
1226                                 int is_default_endian, int sign_extend)
1227 {
1228         int idx, ret;
1229         bool host_swabbed;
1230
1231         /* Pity C doesn't have a logical XOR operator */
1232         if (kvmppc_need_byteswap(vcpu)) {
1233                 host_swabbed = is_default_endian;
1234         } else {
1235                 host_swabbed = !is_default_endian;
1236         }
1237
1238         if (bytes > sizeof(run->mmio.data)) {
1239                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1240                        run->mmio.len);
1241         }
1242
1243         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1244         run->mmio.len = bytes;
1245         run->mmio.is_write = 0;
1246
1247         vcpu->arch.io_gpr = rt;
1248         vcpu->arch.mmio_host_swabbed = host_swabbed;
1249         vcpu->mmio_needed = 1;
1250         vcpu->mmio_is_write = 0;
1251         vcpu->arch.mmio_sign_extend = sign_extend;
1252
1253         idx = srcu_read_lock(&vcpu->kvm->srcu);
1254
1255         ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1256                               bytes, &run->mmio.data);
1257
1258         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1259
1260         if (!ret) {
1261                 kvmppc_complete_mmio_load(vcpu, run);
1262                 vcpu->mmio_needed = 0;
1263                 return EMULATE_DONE;
1264         }
1265
1266         return EMULATE_DO_MMIO;
1267 }
1268
1269 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1270                        unsigned int rt, unsigned int bytes,
1271                        int is_default_endian)
1272 {
1273         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
1274 }
1275 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1276
1277 /* Same as above, but sign extends */
1278 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
1279                         unsigned int rt, unsigned int bytes,
1280                         int is_default_endian)
1281 {
1282         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
1283 }
1284
1285 #ifdef CONFIG_VSX
1286 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1287                         unsigned int rt, unsigned int bytes,
1288                         int is_default_endian, int mmio_sign_extend)
1289 {
1290         enum emulation_result emulated = EMULATE_DONE;
1291
1292         /* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1293         if (vcpu->arch.mmio_vsx_copy_nums > 4)
1294                 return EMULATE_FAIL;
1295
1296         while (vcpu->arch.mmio_vsx_copy_nums) {
1297                 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1298                         is_default_endian, mmio_sign_extend);
1299
1300                 if (emulated != EMULATE_DONE)
1301                         break;
1302
1303                 vcpu->arch.paddr_accessed += run->mmio.len;
1304
1305                 vcpu->arch.mmio_vsx_copy_nums--;
1306                 vcpu->arch.mmio_vsx_offset++;
1307         }
1308         return emulated;
1309 }
1310 #endif /* CONFIG_VSX */
1311
1312 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1313                         u64 val, unsigned int bytes, int is_default_endian)
1314 {
1315         void *data = run->mmio.data;
1316         int idx, ret;
1317         bool host_swabbed;
1318
1319         /* Pity C doesn't have a logical XOR operator */
1320         if (kvmppc_need_byteswap(vcpu)) {
1321                 host_swabbed = is_default_endian;
1322         } else {
1323                 host_swabbed = !is_default_endian;
1324         }
1325
1326         if (bytes > sizeof(run->mmio.data)) {
1327                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1328                        run->mmio.len);
1329         }
1330
1331         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1332         run->mmio.len = bytes;
1333         run->mmio.is_write = 1;
1334         vcpu->mmio_needed = 1;
1335         vcpu->mmio_is_write = 1;
1336
1337         if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1338                 val = dp_to_sp(val);
1339
1340         /* Store the value at the lowest bytes in 'data'. */
1341         if (!host_swabbed) {
1342                 switch (bytes) {
1343                 case 8: *(u64 *)data = val; break;
1344                 case 4: *(u32 *)data = val; break;
1345                 case 2: *(u16 *)data = val; break;
1346                 case 1: *(u8  *)data = val; break;
1347                 }
1348         } else {
1349                 switch (bytes) {
1350                 case 8: *(u64 *)data = swab64(val); break;
1351                 case 4: *(u32 *)data = swab32(val); break;
1352                 case 2: *(u16 *)data = swab16(val); break;
1353                 case 1: *(u8  *)data = val; break;
1354                 }
1355         }
1356
1357         idx = srcu_read_lock(&vcpu->kvm->srcu);
1358
1359         ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1360                                bytes, &run->mmio.data);
1361
1362         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1363
1364         if (!ret) {
1365                 vcpu->mmio_needed = 0;
1366                 return EMULATE_DONE;
1367         }
1368
1369         return EMULATE_DO_MMIO;
1370 }
1371 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1372
1373 #ifdef CONFIG_VSX
1374 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1375 {
1376         u32 dword_offset, word_offset;
1377         union kvmppc_one_reg reg;
1378         int vsx_offset = 0;
1379         int copy_type = vcpu->arch.mmio_copy_type;
1380         int result = 0;
1381
1382         switch (copy_type) {
1383         case KVMPPC_VSX_COPY_DWORD:
1384                 vsx_offset =
1385                         kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1386
1387                 if (vsx_offset == -1) {
1388                         result = -1;
1389                         break;
1390                 }
1391
1392                 if (rs < 32) {
1393                         *val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1394                 } else {
1395                         reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1396                         *val = reg.vsxval[vsx_offset];
1397                 }
1398                 break;
1399
1400         case KVMPPC_VSX_COPY_WORD:
1401                 vsx_offset =
1402                         kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1403
1404                 if (vsx_offset == -1) {
1405                         result = -1;
1406                         break;
1407                 }
1408
1409                 if (rs < 32) {
1410                         dword_offset = vsx_offset / 2;
1411                         word_offset = vsx_offset % 2;
1412                         reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1413                         *val = reg.vsx32val[word_offset];
1414                 } else {
1415                         reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1416                         *val = reg.vsx32val[vsx_offset];
1417                 }
1418                 break;
1419
1420         default:
1421                 result = -1;
1422                 break;
1423         }
1424
1425         return result;
1426 }
1427
1428 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1429                         int rs, unsigned int bytes, int is_default_endian)
1430 {
1431         u64 val;
1432         enum emulation_result emulated = EMULATE_DONE;
1433
1434         vcpu->arch.io_gpr = rs;
1435
1436         /* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1437         if (vcpu->arch.mmio_vsx_copy_nums > 4)
1438                 return EMULATE_FAIL;
1439
1440         while (vcpu->arch.mmio_vsx_copy_nums) {
1441                 if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1442                         return EMULATE_FAIL;
1443
1444                 emulated = kvmppc_handle_store(run, vcpu,
1445                          val, bytes, is_default_endian);
1446
1447                 if (emulated != EMULATE_DONE)
1448                         break;
1449
1450                 vcpu->arch.paddr_accessed += run->mmio.len;
1451
1452                 vcpu->arch.mmio_vsx_copy_nums--;
1453                 vcpu->arch.mmio_vsx_offset++;
1454         }
1455
1456         return emulated;
1457 }
1458
1459 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu,
1460                         struct kvm_run *run)
1461 {
1462         enum emulation_result emulated = EMULATE_FAIL;
1463         int r;
1464
1465         vcpu->arch.paddr_accessed += run->mmio.len;
1466
1467         if (!vcpu->mmio_is_write) {
1468                 emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr,
1469                          run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1470         } else {
1471                 emulated = kvmppc_handle_vsx_store(run, vcpu,
1472                          vcpu->arch.io_gpr, run->mmio.len, 1);
1473         }
1474
1475         switch (emulated) {
1476         case EMULATE_DO_MMIO:
1477                 run->exit_reason = KVM_EXIT_MMIO;
1478                 r = RESUME_HOST;
1479                 break;
1480         case EMULATE_FAIL:
1481                 pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1482                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1483                 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1484                 r = RESUME_HOST;
1485                 break;
1486         default:
1487                 r = RESUME_GUEST;
1488                 break;
1489         }
1490         return r;
1491 }
1492 #endif /* CONFIG_VSX */
1493
1494 #ifdef CONFIG_ALTIVEC
1495 int kvmppc_handle_vmx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1496                 unsigned int rt, unsigned int bytes, int is_default_endian)
1497 {
1498         enum emulation_result emulated = EMULATE_DONE;
1499
1500         if (vcpu->arch.mmio_vsx_copy_nums > 2)
1501                 return EMULATE_FAIL;
1502
1503         while (vcpu->arch.mmio_vmx_copy_nums) {
1504                 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1505                                 is_default_endian, 0);
1506
1507                 if (emulated != EMULATE_DONE)
1508                         break;
1509
1510                 vcpu->arch.paddr_accessed += run->mmio.len;
1511                 vcpu->arch.mmio_vmx_copy_nums--;
1512                 vcpu->arch.mmio_vmx_offset++;
1513         }
1514
1515         return emulated;
1516 }
1517
1518 int kvmppc_get_vmx_dword(struct kvm_vcpu *vcpu, int index, u64 *val)
1519 {
1520         union kvmppc_one_reg reg;
1521         int vmx_offset = 0;
1522         int result = 0;
1523
1524         vmx_offset =
1525                 kvmppc_get_vmx_dword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1526
1527         if (vmx_offset == -1)
1528                 return -1;
1529
1530         reg.vval = VCPU_VSX_VR(vcpu, index);
1531         *val = reg.vsxval[vmx_offset];
1532
1533         return result;
1534 }
1535
1536 int kvmppc_get_vmx_word(struct kvm_vcpu *vcpu, int index, u64 *val)
1537 {
1538         union kvmppc_one_reg reg;
1539         int vmx_offset = 0;
1540         int result = 0;
1541
1542         vmx_offset =
1543                 kvmppc_get_vmx_word_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1544
1545         if (vmx_offset == -1)
1546                 return -1;
1547
1548         reg.vval = VCPU_VSX_VR(vcpu, index);
1549         *val = reg.vsx32val[vmx_offset];
1550
1551         return result;
1552 }
1553
1554 int kvmppc_get_vmx_hword(struct kvm_vcpu *vcpu, int index, u64 *val)
1555 {
1556         union kvmppc_one_reg reg;
1557         int vmx_offset = 0;
1558         int result = 0;
1559
1560         vmx_offset =
1561                 kvmppc_get_vmx_hword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1562
1563         if (vmx_offset == -1)
1564                 return -1;
1565
1566         reg.vval = VCPU_VSX_VR(vcpu, index);
1567         *val = reg.vsx16val[vmx_offset];
1568
1569         return result;
1570 }
1571
1572 int kvmppc_get_vmx_byte(struct kvm_vcpu *vcpu, int index, u64 *val)
1573 {
1574         union kvmppc_one_reg reg;
1575         int vmx_offset = 0;
1576         int result = 0;
1577
1578         vmx_offset =
1579                 kvmppc_get_vmx_byte_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1580
1581         if (vmx_offset == -1)
1582                 return -1;
1583
1584         reg.vval = VCPU_VSX_VR(vcpu, index);
1585         *val = reg.vsx8val[vmx_offset];
1586
1587         return result;
1588 }
1589
1590 int kvmppc_handle_vmx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1591                 unsigned int rs, unsigned int bytes, int is_default_endian)
1592 {
1593         u64 val = 0;
1594         unsigned int index = rs & KVM_MMIO_REG_MASK;
1595         enum emulation_result emulated = EMULATE_DONE;
1596
1597         if (vcpu->arch.mmio_vsx_copy_nums > 2)
1598                 return EMULATE_FAIL;
1599
1600         vcpu->arch.io_gpr = rs;
1601
1602         while (vcpu->arch.mmio_vmx_copy_nums) {
1603                 switch (vcpu->arch.mmio_copy_type) {
1604                 case KVMPPC_VMX_COPY_DWORD:
1605                         if (kvmppc_get_vmx_dword(vcpu, index, &val) == -1)
1606                                 return EMULATE_FAIL;
1607
1608                         break;
1609                 case KVMPPC_VMX_COPY_WORD:
1610                         if (kvmppc_get_vmx_word(vcpu, index, &val) == -1)
1611                                 return EMULATE_FAIL;
1612                         break;
1613                 case KVMPPC_VMX_COPY_HWORD:
1614                         if (kvmppc_get_vmx_hword(vcpu, index, &val) == -1)
1615                                 return EMULATE_FAIL;
1616                         break;
1617                 case KVMPPC_VMX_COPY_BYTE:
1618                         if (kvmppc_get_vmx_byte(vcpu, index, &val) == -1)
1619                                 return EMULATE_FAIL;
1620                         break;
1621                 default:
1622                         return EMULATE_FAIL;
1623                 }
1624
1625                 emulated = kvmppc_handle_store(run, vcpu, val, bytes,
1626                                 is_default_endian);
1627                 if (emulated != EMULATE_DONE)
1628                         break;
1629
1630                 vcpu->arch.paddr_accessed += run->mmio.len;
1631                 vcpu->arch.mmio_vmx_copy_nums--;
1632                 vcpu->arch.mmio_vmx_offset++;
1633         }
1634
1635         return emulated;
1636 }
1637
1638 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu,
1639                 struct kvm_run *run)
1640 {
1641         enum emulation_result emulated = EMULATE_FAIL;
1642         int r;
1643
1644         vcpu->arch.paddr_accessed += run->mmio.len;
1645
1646         if (!vcpu->mmio_is_write) {
1647                 emulated = kvmppc_handle_vmx_load(run, vcpu,
1648                                 vcpu->arch.io_gpr, run->mmio.len, 1);
1649         } else {
1650                 emulated = kvmppc_handle_vmx_store(run, vcpu,
1651                                 vcpu->arch.io_gpr, run->mmio.len, 1);
1652         }
1653
1654         switch (emulated) {
1655         case EMULATE_DO_MMIO:
1656                 run->exit_reason = KVM_EXIT_MMIO;
1657                 r = RESUME_HOST;
1658                 break;
1659         case EMULATE_FAIL:
1660                 pr_info("KVM: MMIO emulation failed (VMX repeat)\n");
1661                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1662                 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1663                 r = RESUME_HOST;
1664                 break;
1665         default:
1666                 r = RESUME_GUEST;
1667                 break;
1668         }
1669         return r;
1670 }
1671 #endif /* CONFIG_ALTIVEC */
1672
1673 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1674 {
1675         int r = 0;
1676         union kvmppc_one_reg val;
1677         int size;
1678
1679         size = one_reg_size(reg->id);
1680         if (size > sizeof(val))
1681                 return -EINVAL;
1682
1683         r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1684         if (r == -EINVAL) {
1685                 r = 0;
1686                 switch (reg->id) {
1687 #ifdef CONFIG_ALTIVEC
1688                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1689                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1690                                 r = -ENXIO;
1691                                 break;
1692                         }
1693                         val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1694                         break;
1695                 case KVM_REG_PPC_VSCR:
1696                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1697                                 r = -ENXIO;
1698                                 break;
1699                         }
1700                         val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1701                         break;
1702                 case KVM_REG_PPC_VRSAVE:
1703                         val = get_reg_val(reg->id, vcpu->arch.vrsave);
1704                         break;
1705 #endif /* CONFIG_ALTIVEC */
1706                 default:
1707                         r = -EINVAL;
1708                         break;
1709                 }
1710         }
1711
1712         if (r)
1713                 return r;
1714
1715         if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1716                 r = -EFAULT;
1717
1718         return r;
1719 }
1720
1721 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1722 {
1723         int r;
1724         union kvmppc_one_reg val;
1725         int size;
1726
1727         size = one_reg_size(reg->id);
1728         if (size > sizeof(val))
1729                 return -EINVAL;
1730
1731         if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1732                 return -EFAULT;
1733
1734         r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1735         if (r == -EINVAL) {
1736                 r = 0;
1737                 switch (reg->id) {
1738 #ifdef CONFIG_ALTIVEC
1739                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1740                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1741                                 r = -ENXIO;
1742                                 break;
1743                         }
1744                         vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1745                         break;
1746                 case KVM_REG_PPC_VSCR:
1747                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1748                                 r = -ENXIO;
1749                                 break;
1750                         }
1751                         vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1752                         break;
1753                 case KVM_REG_PPC_VRSAVE:
1754                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1755                                 r = -ENXIO;
1756                                 break;
1757                         }
1758                         vcpu->arch.vrsave = set_reg_val(reg->id, val);
1759                         break;
1760 #endif /* CONFIG_ALTIVEC */
1761                 default:
1762                         r = -EINVAL;
1763                         break;
1764                 }
1765         }
1766
1767         return r;
1768 }
1769
1770 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1771 {
1772         int r;
1773
1774         vcpu_load(vcpu);
1775
1776         if (vcpu->mmio_needed) {
1777                 vcpu->mmio_needed = 0;
1778                 if (!vcpu->mmio_is_write)
1779                         kvmppc_complete_mmio_load(vcpu, run);
1780 #ifdef CONFIG_VSX
1781                 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1782                         vcpu->arch.mmio_vsx_copy_nums--;
1783                         vcpu->arch.mmio_vsx_offset++;
1784                 }
1785
1786                 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1787                         r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run);
1788                         if (r == RESUME_HOST) {
1789                                 vcpu->mmio_needed = 1;
1790                                 goto out;
1791                         }
1792                 }
1793 #endif
1794 #ifdef CONFIG_ALTIVEC
1795                 if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1796                         vcpu->arch.mmio_vmx_copy_nums--;
1797                         vcpu->arch.mmio_vmx_offset++;
1798                 }
1799
1800                 if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1801                         r = kvmppc_emulate_mmio_vmx_loadstore(vcpu, run);
1802                         if (r == RESUME_HOST) {
1803                                 vcpu->mmio_needed = 1;
1804                                 goto out;
1805                         }
1806                 }
1807 #endif
1808         } else if (vcpu->arch.osi_needed) {
1809                 u64 *gprs = run->osi.gprs;
1810                 int i;
1811
1812                 for (i = 0; i < 32; i++)
1813                         kvmppc_set_gpr(vcpu, i, gprs[i]);
1814                 vcpu->arch.osi_needed = 0;
1815         } else if (vcpu->arch.hcall_needed) {
1816                 int i;
1817
1818                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1819                 for (i = 0; i < 9; ++i)
1820                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1821                 vcpu->arch.hcall_needed = 0;
1822 #ifdef CONFIG_BOOKE
1823         } else if (vcpu->arch.epr_needed) {
1824                 kvmppc_set_epr(vcpu, run->epr.epr);
1825                 vcpu->arch.epr_needed = 0;
1826 #endif
1827         }
1828
1829         kvm_sigset_activate(vcpu);
1830
1831         if (run->immediate_exit)
1832                 r = -EINTR;
1833         else
1834                 r = kvmppc_vcpu_run(run, vcpu);
1835
1836         kvm_sigset_deactivate(vcpu);
1837
1838 #ifdef CONFIG_ALTIVEC
1839 out:
1840 #endif
1841         vcpu_put(vcpu);
1842         return r;
1843 }
1844
1845 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1846 {
1847         if (irq->irq == KVM_INTERRUPT_UNSET) {
1848                 kvmppc_core_dequeue_external(vcpu);
1849                 return 0;
1850         }
1851
1852         kvmppc_core_queue_external(vcpu, irq);
1853
1854         kvm_vcpu_kick(vcpu);
1855
1856         return 0;
1857 }
1858
1859 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1860                                      struct kvm_enable_cap *cap)
1861 {
1862         int r;
1863
1864         if (cap->flags)
1865                 return -EINVAL;
1866
1867         switch (cap->cap) {
1868         case KVM_CAP_PPC_OSI:
1869                 r = 0;
1870                 vcpu->arch.osi_enabled = true;
1871                 break;
1872         case KVM_CAP_PPC_PAPR:
1873                 r = 0;
1874                 vcpu->arch.papr_enabled = true;
1875                 break;
1876         case KVM_CAP_PPC_EPR:
1877                 r = 0;
1878                 if (cap->args[0])
1879                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1880                 else
1881                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1882                 break;
1883 #ifdef CONFIG_BOOKE
1884         case KVM_CAP_PPC_BOOKE_WATCHDOG:
1885                 r = 0;
1886                 vcpu->arch.watchdog_enabled = true;
1887                 break;
1888 #endif
1889 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1890         case KVM_CAP_SW_TLB: {
1891                 struct kvm_config_tlb cfg;
1892                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1893
1894                 r = -EFAULT;
1895                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1896                         break;
1897
1898                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1899                 break;
1900         }
1901 #endif
1902 #ifdef CONFIG_KVM_MPIC
1903         case KVM_CAP_IRQ_MPIC: {
1904                 struct fd f;
1905                 struct kvm_device *dev;
1906
1907                 r = -EBADF;
1908                 f = fdget(cap->args[0]);
1909                 if (!f.file)
1910                         break;
1911
1912                 r = -EPERM;
1913                 dev = kvm_device_from_filp(f.file);
1914                 if (dev)
1915                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1916
1917                 fdput(f);
1918                 break;
1919         }
1920 #endif
1921 #ifdef CONFIG_KVM_XICS
1922         case KVM_CAP_IRQ_XICS: {
1923                 struct fd f;
1924                 struct kvm_device *dev;
1925
1926                 r = -EBADF;
1927                 f = fdget(cap->args[0]);
1928                 if (!f.file)
1929                         break;
1930
1931                 r = -EPERM;
1932                 dev = kvm_device_from_filp(f.file);
1933                 if (dev) {
1934                         if (xics_on_xive())
1935                                 r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1936                         else
1937                                 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1938                 }
1939
1940                 fdput(f);
1941                 break;
1942         }
1943 #endif /* CONFIG_KVM_XICS */
1944 #ifdef CONFIG_KVM_XIVE
1945         case KVM_CAP_PPC_IRQ_XIVE: {
1946                 struct fd f;
1947                 struct kvm_device *dev;
1948
1949                 r = -EBADF;
1950                 f = fdget(cap->args[0]);
1951                 if (!f.file)
1952                         break;
1953
1954                 r = -ENXIO;
1955                 if (!xive_enabled())
1956                         break;
1957
1958                 r = -EPERM;
1959                 dev = kvm_device_from_filp(f.file);
1960                 if (dev)
1961                         r = kvmppc_xive_native_connect_vcpu(dev, vcpu,
1962                                                             cap->args[1]);
1963
1964                 fdput(f);
1965                 break;
1966         }
1967 #endif /* CONFIG_KVM_XIVE */
1968 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1969         case KVM_CAP_PPC_FWNMI:
1970                 r = -EINVAL;
1971                 if (!is_kvmppc_hv_enabled(vcpu->kvm))
1972                         break;
1973                 r = 0;
1974                 vcpu->kvm->arch.fwnmi_enabled = true;
1975                 break;
1976 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1977         default:
1978                 r = -EINVAL;
1979                 break;
1980         }
1981
1982         if (!r)
1983                 r = kvmppc_sanity_check(vcpu);
1984
1985         return r;
1986 }
1987
1988 bool kvm_arch_intc_initialized(struct kvm *kvm)
1989 {
1990 #ifdef CONFIG_KVM_MPIC
1991         if (kvm->arch.mpic)
1992                 return true;
1993 #endif
1994 #ifdef CONFIG_KVM_XICS
1995         if (kvm->arch.xics || kvm->arch.xive)
1996                 return true;
1997 #endif
1998         return false;
1999 }
2000
2001 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2002                                     struct kvm_mp_state *mp_state)
2003 {
2004         return -EINVAL;
2005 }
2006
2007 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2008                                     struct kvm_mp_state *mp_state)
2009 {
2010         return -EINVAL;
2011 }
2012
2013 long kvm_arch_vcpu_async_ioctl(struct file *filp,
2014                                unsigned int ioctl, unsigned long arg)
2015 {
2016         struct kvm_vcpu *vcpu = filp->private_data;
2017         void __user *argp = (void __user *)arg;
2018
2019         if (ioctl == KVM_INTERRUPT) {
2020                 struct kvm_interrupt irq;
2021                 if (copy_from_user(&irq, argp, sizeof(irq)))
2022                         return -EFAULT;
2023                 return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2024         }
2025         return -ENOIOCTLCMD;
2026 }
2027
2028 long kvm_arch_vcpu_ioctl(struct file *filp,
2029                          unsigned int ioctl, unsigned long arg)
2030 {
2031         struct kvm_vcpu *vcpu = filp->private_data;
2032         void __user *argp = (void __user *)arg;
2033         long r;
2034
2035         switch (ioctl) {
2036         case KVM_ENABLE_CAP:
2037         {
2038                 struct kvm_enable_cap cap;
2039                 r = -EFAULT;
2040                 vcpu_load(vcpu);
2041                 if (copy_from_user(&cap, argp, sizeof(cap)))
2042                         goto out;
2043                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
2044                 vcpu_put(vcpu);
2045                 break;
2046         }
2047
2048         case KVM_SET_ONE_REG:
2049         case KVM_GET_ONE_REG:
2050         {
2051                 struct kvm_one_reg reg;
2052                 r = -EFAULT;
2053                 if (copy_from_user(&reg, argp, sizeof(reg)))
2054                         goto out;
2055                 if (ioctl == KVM_SET_ONE_REG)
2056                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
2057                 else
2058                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
2059                 break;
2060         }
2061
2062 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
2063         case KVM_DIRTY_TLB: {
2064                 struct kvm_dirty_tlb dirty;
2065                 r = -EFAULT;
2066                 vcpu_load(vcpu);
2067                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
2068                         goto out;
2069                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
2070                 vcpu_put(vcpu);
2071                 break;
2072         }
2073 #endif
2074         default:
2075                 r = -EINVAL;
2076         }
2077
2078 out:
2079         return r;
2080 }
2081
2082 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2083 {
2084         return VM_FAULT_SIGBUS;
2085 }
2086
2087 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
2088 {
2089         u32 inst_nop = 0x60000000;
2090 #ifdef CONFIG_KVM_BOOKE_HV
2091         u32 inst_sc1 = 0x44000022;
2092         pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
2093         pvinfo->hcall[1] = cpu_to_be32(inst_nop);
2094         pvinfo->hcall[2] = cpu_to_be32(inst_nop);
2095         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2096 #else
2097         u32 inst_lis = 0x3c000000;
2098         u32 inst_ori = 0x60000000;
2099         u32 inst_sc = 0x44000002;
2100         u32 inst_imm_mask = 0xffff;
2101
2102         /*
2103          * The hypercall to get into KVM from within guest context is as
2104          * follows:
2105          *
2106          *    lis r0, r0, KVM_SC_MAGIC_R0@h
2107          *    ori r0, KVM_SC_MAGIC_R0@l
2108          *    sc
2109          *    nop
2110          */
2111         pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
2112         pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
2113         pvinfo->hcall[2] = cpu_to_be32(inst_sc);
2114         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2115 #endif
2116
2117         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
2118
2119         return 0;
2120 }
2121
2122 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
2123                           bool line_status)
2124 {
2125         if (!irqchip_in_kernel(kvm))
2126                 return -ENXIO;
2127
2128         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2129                                         irq_event->irq, irq_event->level,
2130                                         line_status);
2131         return 0;
2132 }
2133
2134
2135 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
2136                             struct kvm_enable_cap *cap)
2137 {
2138         int r;
2139
2140         if (cap->flags)
2141                 return -EINVAL;
2142
2143         switch (cap->cap) {
2144 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
2145         case KVM_CAP_PPC_ENABLE_HCALL: {
2146                 unsigned long hcall = cap->args[0];
2147
2148                 r = -EINVAL;
2149                 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
2150                     cap->args[1] > 1)
2151                         break;
2152                 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
2153                         break;
2154                 if (cap->args[1])
2155                         set_bit(hcall / 4, kvm->arch.enabled_hcalls);
2156                 else
2157                         clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
2158                 r = 0;
2159                 break;
2160         }
2161         case KVM_CAP_PPC_SMT: {
2162                 unsigned long mode = cap->args[0];
2163                 unsigned long flags = cap->args[1];
2164
2165                 r = -EINVAL;
2166                 if (kvm->arch.kvm_ops->set_smt_mode)
2167                         r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
2168                 break;
2169         }
2170
2171         case KVM_CAP_PPC_NESTED_HV:
2172                 r = -EINVAL;
2173                 if (!is_kvmppc_hv_enabled(kvm) ||
2174                     !kvm->arch.kvm_ops->enable_nested)
2175                         break;
2176                 r = kvm->arch.kvm_ops->enable_nested(kvm);
2177                 break;
2178 #endif
2179         default:
2180                 r = -EINVAL;
2181                 break;
2182         }
2183
2184         return r;
2185 }
2186
2187 #ifdef CONFIG_PPC_BOOK3S_64
2188 /*
2189  * These functions check whether the underlying hardware is safe
2190  * against attacks based on observing the effects of speculatively
2191  * executed instructions, and whether it supplies instructions for
2192  * use in workarounds.  The information comes from firmware, either
2193  * via the device tree on powernv platforms or from an hcall on
2194  * pseries platforms.
2195  */
2196 #ifdef CONFIG_PPC_PSERIES
2197 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2198 {
2199         struct h_cpu_char_result c;
2200         unsigned long rc;
2201
2202         if (!machine_is(pseries))
2203                 return -ENOTTY;
2204
2205         rc = plpar_get_cpu_characteristics(&c);
2206         if (rc == H_SUCCESS) {
2207                 cp->character = c.character;
2208                 cp->behaviour = c.behaviour;
2209                 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2210                         KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2211                         KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2212                         KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2213                         KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2214                         KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
2215                         KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
2216                         KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2217                         KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2218                 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2219                         KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2220                         KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2221                         KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2222         }
2223         return 0;
2224 }
2225 #else
2226 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2227 {
2228         return -ENOTTY;
2229 }
2230 #endif
2231
2232 static inline bool have_fw_feat(struct device_node *fw_features,
2233                                 const char *state, const char *name)
2234 {
2235         struct device_node *np;
2236         bool r = false;
2237
2238         np = of_get_child_by_name(fw_features, name);
2239         if (np) {
2240                 r = of_property_read_bool(np, state);
2241                 of_node_put(np);
2242         }
2243         return r;
2244 }
2245
2246 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2247 {
2248         struct device_node *np, *fw_features;
2249         int r;
2250
2251         memset(cp, 0, sizeof(*cp));
2252         r = pseries_get_cpu_char(cp);
2253         if (r != -ENOTTY)
2254                 return r;
2255
2256         np = of_find_node_by_name(NULL, "ibm,opal");
2257         if (np) {
2258                 fw_features = of_get_child_by_name(np, "fw-features");
2259                 of_node_put(np);
2260                 if (!fw_features)
2261                         return 0;
2262                 if (have_fw_feat(fw_features, "enabled",
2263                                  "inst-spec-barrier-ori31,31,0"))
2264                         cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
2265                 if (have_fw_feat(fw_features, "enabled",
2266                                  "fw-bcctrl-serialized"))
2267                         cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
2268                 if (have_fw_feat(fw_features, "enabled",
2269                                  "inst-l1d-flush-ori30,30,0"))
2270                         cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
2271                 if (have_fw_feat(fw_features, "enabled",
2272                                  "inst-l1d-flush-trig2"))
2273                         cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
2274                 if (have_fw_feat(fw_features, "enabled",
2275                                  "fw-l1d-thread-split"))
2276                         cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
2277                 if (have_fw_feat(fw_features, "enabled",
2278                                  "fw-count-cache-disabled"))
2279                         cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2280                 if (have_fw_feat(fw_features, "enabled",
2281                                  "fw-count-cache-flush-bcctr2,0,0"))
2282                         cp->character |= KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2283                 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2284                         KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2285                         KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2286                         KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2287                         KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2288                         KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2289                         KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2290
2291                 if (have_fw_feat(fw_features, "enabled",
2292                                  "speculation-policy-favor-security"))
2293                         cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
2294                 if (!have_fw_feat(fw_features, "disabled",
2295                                   "needs-l1d-flush-msr-pr-0-to-1"))
2296                         cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
2297                 if (!have_fw_feat(fw_features, "disabled",
2298                                   "needs-spec-barrier-for-bound-checks"))
2299                         cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2300                 if (have_fw_feat(fw_features, "enabled",
2301                                  "needs-count-cache-flush-on-context-switch"))
2302                         cp->behaviour |= KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2303                 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2304                         KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2305                         KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2306                         KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2307
2308                 of_node_put(fw_features);
2309         }
2310
2311         return 0;
2312 }
2313 #endif
2314
2315 long kvm_arch_vm_ioctl(struct file *filp,
2316                        unsigned int ioctl, unsigned long arg)
2317 {
2318         struct kvm *kvm __maybe_unused = filp->private_data;
2319         void __user *argp = (void __user *)arg;
2320         long r;
2321
2322         switch (ioctl) {
2323         case KVM_PPC_GET_PVINFO: {
2324                 struct kvm_ppc_pvinfo pvinfo;
2325                 memset(&pvinfo, 0, sizeof(pvinfo));
2326                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
2327                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
2328                         r = -EFAULT;
2329                         goto out;
2330                 }
2331
2332                 break;
2333         }
2334 #ifdef CONFIG_SPAPR_TCE_IOMMU
2335         case KVM_CREATE_SPAPR_TCE_64: {
2336                 struct kvm_create_spapr_tce_64 create_tce_64;
2337
2338                 r = -EFAULT;
2339                 if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
2340                         goto out;
2341                 if (create_tce_64.flags) {
2342                         r = -EINVAL;
2343                         goto out;
2344                 }
2345                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2346                 goto out;
2347         }
2348         case KVM_CREATE_SPAPR_TCE: {
2349                 struct kvm_create_spapr_tce create_tce;
2350                 struct kvm_create_spapr_tce_64 create_tce_64;
2351
2352                 r = -EFAULT;
2353                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
2354                         goto out;
2355
2356                 create_tce_64.liobn = create_tce.liobn;
2357                 create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
2358                 create_tce_64.offset = 0;
2359                 create_tce_64.size = create_tce.window_size >>
2360                                 IOMMU_PAGE_SHIFT_4K;
2361                 create_tce_64.flags = 0;
2362                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2363                 goto out;
2364         }
2365 #endif
2366 #ifdef CONFIG_PPC_BOOK3S_64
2367         case KVM_PPC_GET_SMMU_INFO: {
2368                 struct kvm_ppc_smmu_info info;
2369                 struct kvm *kvm = filp->private_data;
2370
2371                 memset(&info, 0, sizeof(info));
2372                 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
2373                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2374                         r = -EFAULT;
2375                 break;
2376         }
2377         case KVM_PPC_RTAS_DEFINE_TOKEN: {
2378                 struct kvm *kvm = filp->private_data;
2379
2380                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
2381                 break;
2382         }
2383         case KVM_PPC_CONFIGURE_V3_MMU: {
2384                 struct kvm *kvm = filp->private_data;
2385                 struct kvm_ppc_mmuv3_cfg cfg;
2386
2387                 r = -EINVAL;
2388                 if (!kvm->arch.kvm_ops->configure_mmu)
2389                         goto out;
2390                 r = -EFAULT;
2391                 if (copy_from_user(&cfg, argp, sizeof(cfg)))
2392                         goto out;
2393                 r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
2394                 break;
2395         }
2396         case KVM_PPC_GET_RMMU_INFO: {
2397                 struct kvm *kvm = filp->private_data;
2398                 struct kvm_ppc_rmmu_info info;
2399
2400                 r = -EINVAL;
2401                 if (!kvm->arch.kvm_ops->get_rmmu_info)
2402                         goto out;
2403                 r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
2404                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2405                         r = -EFAULT;
2406                 break;
2407         }
2408         case KVM_PPC_GET_CPU_CHAR: {
2409                 struct kvm_ppc_cpu_char cpuchar;
2410
2411                 r = kvmppc_get_cpu_char(&cpuchar);
2412                 if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2413                         r = -EFAULT;
2414                 break;
2415         }
2416         case KVM_PPC_SVM_OFF: {
2417                 struct kvm *kvm = filp->private_data;
2418
2419                 r = 0;
2420                 if (!kvm->arch.kvm_ops->svm_off)
2421                         goto out;
2422
2423                 r = kvm->arch.kvm_ops->svm_off(kvm);
2424                 break;
2425         }
2426         default: {
2427                 struct kvm *kvm = filp->private_data;
2428                 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
2429         }
2430 #else /* CONFIG_PPC_BOOK3S_64 */
2431         default:
2432                 r = -ENOTTY;
2433 #endif
2434         }
2435 out:
2436         return r;
2437 }
2438
2439 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
2440 static unsigned long nr_lpids;
2441
2442 long kvmppc_alloc_lpid(void)
2443 {
2444         long lpid;
2445
2446         do {
2447                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
2448                 if (lpid >= nr_lpids) {
2449                         pr_err("%s: No LPIDs free\n", __func__);
2450                         return -ENOMEM;
2451                 }
2452         } while (test_and_set_bit(lpid, lpid_inuse));
2453
2454         return lpid;
2455 }
2456 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2457
2458 void kvmppc_claim_lpid(long lpid)
2459 {
2460         set_bit(lpid, lpid_inuse);
2461 }
2462 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
2463
2464 void kvmppc_free_lpid(long lpid)
2465 {
2466         clear_bit(lpid, lpid_inuse);
2467 }
2468 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2469
2470 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2471 {
2472         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
2473         memset(lpid_inuse, 0, sizeof(lpid_inuse));
2474 }
2475 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2476
2477 int kvm_arch_init(void *opaque)
2478 {
2479         return 0;
2480 }
2481
2482 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);