Merge tag 'v5.2-rc2-smb3-fixes' of git://git.samba.org/sfrench/cifs-2.6
[linux-2.6-microblaze.git] / arch / powerpc / kvm / powerpc.c
1 /*
2  * This program is free software; you can redistribute it and/or modify
3  * it under the terms of the GNU General Public License, version 2, as
4  * published by the Free Software Foundation.
5  *
6  * This program is distributed in the hope that it will be useful,
7  * but WITHOUT ANY WARRANTY; without even the implied warranty of
8  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9  * GNU General Public License for more details.
10  *
11  * You should have received a copy of the GNU General Public License
12  * along with this program; if not, write to the Free Software
13  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
14  *
15  * Copyright IBM Corp. 2007
16  *
17  * Authors: Hollis Blanchard <hollisb@us.ibm.com>
18  *          Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
19  */
20
21 #include <linux/errno.h>
22 #include <linux/err.h>
23 #include <linux/kvm_host.h>
24 #include <linux/vmalloc.h>
25 #include <linux/hrtimer.h>
26 #include <linux/sched/signal.h>
27 #include <linux/fs.h>
28 #include <linux/slab.h>
29 #include <linux/file.h>
30 #include <linux/module.h>
31 #include <linux/irqbypass.h>
32 #include <linux/kvm_irqfd.h>
33 #include <asm/cputable.h>
34 #include <linux/uaccess.h>
35 #include <asm/kvm_ppc.h>
36 #include <asm/cputhreads.h>
37 #include <asm/irqflags.h>
38 #include <asm/iommu.h>
39 #include <asm/switch_to.h>
40 #include <asm/xive.h>
41 #ifdef CONFIG_PPC_PSERIES
42 #include <asm/hvcall.h>
43 #include <asm/plpar_wrappers.h>
44 #endif
45
46 #include "timing.h"
47 #include "irq.h"
48 #include "../mm/mmu_decl.h"
49
50 #define CREATE_TRACE_POINTS
51 #include "trace.h"
52
53 struct kvmppc_ops *kvmppc_hv_ops;
54 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
55 struct kvmppc_ops *kvmppc_pr_ops;
56 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
57
58
59 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
60 {
61         return !!(v->arch.pending_exceptions) || kvm_request_pending(v);
62 }
63
64 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
65 {
66         return false;
67 }
68
69 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
70 {
71         return 1;
72 }
73
74 /*
75  * Common checks before entering the guest world.  Call with interrupts
76  * disabled.
77  *
78  * returns:
79  *
80  * == 1 if we're ready to go into guest state
81  * <= 0 if we need to go back to the host with return value
82  */
83 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
84 {
85         int r;
86
87         WARN_ON(irqs_disabled());
88         hard_irq_disable();
89
90         while (true) {
91                 if (need_resched()) {
92                         local_irq_enable();
93                         cond_resched();
94                         hard_irq_disable();
95                         continue;
96                 }
97
98                 if (signal_pending(current)) {
99                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
100                         vcpu->run->exit_reason = KVM_EXIT_INTR;
101                         r = -EINTR;
102                         break;
103                 }
104
105                 vcpu->mode = IN_GUEST_MODE;
106
107                 /*
108                  * Reading vcpu->requests must happen after setting vcpu->mode,
109                  * so we don't miss a request because the requester sees
110                  * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
111                  * before next entering the guest (and thus doesn't IPI).
112                  * This also orders the write to mode from any reads
113                  * to the page tables done while the VCPU is running.
114                  * Please see the comment in kvm_flush_remote_tlbs.
115                  */
116                 smp_mb();
117
118                 if (kvm_request_pending(vcpu)) {
119                         /* Make sure we process requests preemptable */
120                         local_irq_enable();
121                         trace_kvm_check_requests(vcpu);
122                         r = kvmppc_core_check_requests(vcpu);
123                         hard_irq_disable();
124                         if (r > 0)
125                                 continue;
126                         break;
127                 }
128
129                 if (kvmppc_core_prepare_to_enter(vcpu)) {
130                         /* interrupts got enabled in between, so we
131                            are back at square 1 */
132                         continue;
133                 }
134
135                 guest_enter_irqoff();
136                 return 1;
137         }
138
139         /* return to host */
140         local_irq_enable();
141         return r;
142 }
143 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
144
145 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
146 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
147 {
148         struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
149         int i;
150
151         shared->sprg0 = swab64(shared->sprg0);
152         shared->sprg1 = swab64(shared->sprg1);
153         shared->sprg2 = swab64(shared->sprg2);
154         shared->sprg3 = swab64(shared->sprg3);
155         shared->srr0 = swab64(shared->srr0);
156         shared->srr1 = swab64(shared->srr1);
157         shared->dar = swab64(shared->dar);
158         shared->msr = swab64(shared->msr);
159         shared->dsisr = swab32(shared->dsisr);
160         shared->int_pending = swab32(shared->int_pending);
161         for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
162                 shared->sr[i] = swab32(shared->sr[i]);
163 }
164 #endif
165
166 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
167 {
168         int nr = kvmppc_get_gpr(vcpu, 11);
169         int r;
170         unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
171         unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
172         unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
173         unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
174         unsigned long r2 = 0;
175
176         if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
177                 /* 32 bit mode */
178                 param1 &= 0xffffffff;
179                 param2 &= 0xffffffff;
180                 param3 &= 0xffffffff;
181                 param4 &= 0xffffffff;
182         }
183
184         switch (nr) {
185         case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
186         {
187 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
188                 /* Book3S can be little endian, find it out here */
189                 int shared_big_endian = true;
190                 if (vcpu->arch.intr_msr & MSR_LE)
191                         shared_big_endian = false;
192                 if (shared_big_endian != vcpu->arch.shared_big_endian)
193                         kvmppc_swab_shared(vcpu);
194                 vcpu->arch.shared_big_endian = shared_big_endian;
195 #endif
196
197                 if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
198                         /*
199                          * Older versions of the Linux magic page code had
200                          * a bug where they would map their trampoline code
201                          * NX. If that's the case, remove !PR NX capability.
202                          */
203                         vcpu->arch.disable_kernel_nx = true;
204                         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
205                 }
206
207                 vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
208                 vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
209
210 #ifdef CONFIG_PPC_64K_PAGES
211                 /*
212                  * Make sure our 4k magic page is in the same window of a 64k
213                  * page within the guest and within the host's page.
214                  */
215                 if ((vcpu->arch.magic_page_pa & 0xf000) !=
216                     ((ulong)vcpu->arch.shared & 0xf000)) {
217                         void *old_shared = vcpu->arch.shared;
218                         ulong shared = (ulong)vcpu->arch.shared;
219                         void *new_shared;
220
221                         shared &= PAGE_MASK;
222                         shared |= vcpu->arch.magic_page_pa & 0xf000;
223                         new_shared = (void*)shared;
224                         memcpy(new_shared, old_shared, 0x1000);
225                         vcpu->arch.shared = new_shared;
226                 }
227 #endif
228
229                 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
230
231                 r = EV_SUCCESS;
232                 break;
233         }
234         case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
235                 r = EV_SUCCESS;
236 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
237                 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
238 #endif
239
240                 /* Second return value is in r4 */
241                 break;
242         case EV_HCALL_TOKEN(EV_IDLE):
243                 r = EV_SUCCESS;
244                 kvm_vcpu_block(vcpu);
245                 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
246                 break;
247         default:
248                 r = EV_UNIMPLEMENTED;
249                 break;
250         }
251
252         kvmppc_set_gpr(vcpu, 4, r2);
253
254         return r;
255 }
256 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
257
258 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
259 {
260         int r = false;
261
262         /* We have to know what CPU to virtualize */
263         if (!vcpu->arch.pvr)
264                 goto out;
265
266         /* PAPR only works with book3s_64 */
267         if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
268                 goto out;
269
270         /* HV KVM can only do PAPR mode for now */
271         if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
272                 goto out;
273
274 #ifdef CONFIG_KVM_BOOKE_HV
275         if (!cpu_has_feature(CPU_FTR_EMB_HV))
276                 goto out;
277 #endif
278
279         r = true;
280
281 out:
282         vcpu->arch.sane = r;
283         return r ? 0 : -EINVAL;
284 }
285 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
286
287 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
288 {
289         enum emulation_result er;
290         int r;
291
292         er = kvmppc_emulate_loadstore(vcpu);
293         switch (er) {
294         case EMULATE_DONE:
295                 /* Future optimization: only reload non-volatiles if they were
296                  * actually modified. */
297                 r = RESUME_GUEST_NV;
298                 break;
299         case EMULATE_AGAIN:
300                 r = RESUME_GUEST;
301                 break;
302         case EMULATE_DO_MMIO:
303                 run->exit_reason = KVM_EXIT_MMIO;
304                 /* We must reload nonvolatiles because "update" load/store
305                  * instructions modify register state. */
306                 /* Future optimization: only reload non-volatiles if they were
307                  * actually modified. */
308                 r = RESUME_HOST_NV;
309                 break;
310         case EMULATE_FAIL:
311         {
312                 u32 last_inst;
313
314                 kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
315                 /* XXX Deliver Program interrupt to guest. */
316                 pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
317                 r = RESUME_HOST;
318                 break;
319         }
320         default:
321                 WARN_ON(1);
322                 r = RESUME_GUEST;
323         }
324
325         return r;
326 }
327 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
328
329 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
330               bool data)
331 {
332         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
333         struct kvmppc_pte pte;
334         int r = -EINVAL;
335
336         vcpu->stat.st++;
337
338         if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->store_to_eaddr)
339                 r = vcpu->kvm->arch.kvm_ops->store_to_eaddr(vcpu, eaddr, ptr,
340                                                             size);
341
342         if ((!r) || (r == -EAGAIN))
343                 return r;
344
345         r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
346                          XLATE_WRITE, &pte);
347         if (r < 0)
348                 return r;
349
350         *eaddr = pte.raddr;
351
352         if (!pte.may_write)
353                 return -EPERM;
354
355         /* Magic page override */
356         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
357             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
358             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
359                 void *magic = vcpu->arch.shared;
360                 magic += pte.eaddr & 0xfff;
361                 memcpy(magic, ptr, size);
362                 return EMULATE_DONE;
363         }
364
365         if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
366                 return EMULATE_DO_MMIO;
367
368         return EMULATE_DONE;
369 }
370 EXPORT_SYMBOL_GPL(kvmppc_st);
371
372 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
373                       bool data)
374 {
375         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
376         struct kvmppc_pte pte;
377         int rc = -EINVAL;
378
379         vcpu->stat.ld++;
380
381         if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->load_from_eaddr)
382                 rc = vcpu->kvm->arch.kvm_ops->load_from_eaddr(vcpu, eaddr, ptr,
383                                                               size);
384
385         if ((!rc) || (rc == -EAGAIN))
386                 return rc;
387
388         rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
389                           XLATE_READ, &pte);
390         if (rc)
391                 return rc;
392
393         *eaddr = pte.raddr;
394
395         if (!pte.may_read)
396                 return -EPERM;
397
398         if (!data && !pte.may_execute)
399                 return -ENOEXEC;
400
401         /* Magic page override */
402         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
403             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
404             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
405                 void *magic = vcpu->arch.shared;
406                 magic += pte.eaddr & 0xfff;
407                 memcpy(ptr, magic, size);
408                 return EMULATE_DONE;
409         }
410
411         if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
412                 return EMULATE_DO_MMIO;
413
414         return EMULATE_DONE;
415 }
416 EXPORT_SYMBOL_GPL(kvmppc_ld);
417
418 int kvm_arch_hardware_enable(void)
419 {
420         return 0;
421 }
422
423 int kvm_arch_hardware_setup(void)
424 {
425         return 0;
426 }
427
428 void kvm_arch_check_processor_compat(void *rtn)
429 {
430         *(int *)rtn = kvmppc_core_check_processor_compat();
431 }
432
433 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
434 {
435         struct kvmppc_ops *kvm_ops = NULL;
436         /*
437          * if we have both HV and PR enabled, default is HV
438          */
439         if (type == 0) {
440                 if (kvmppc_hv_ops)
441                         kvm_ops = kvmppc_hv_ops;
442                 else
443                         kvm_ops = kvmppc_pr_ops;
444                 if (!kvm_ops)
445                         goto err_out;
446         } else  if (type == KVM_VM_PPC_HV) {
447                 if (!kvmppc_hv_ops)
448                         goto err_out;
449                 kvm_ops = kvmppc_hv_ops;
450         } else if (type == KVM_VM_PPC_PR) {
451                 if (!kvmppc_pr_ops)
452                         goto err_out;
453                 kvm_ops = kvmppc_pr_ops;
454         } else
455                 goto err_out;
456
457         if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
458                 return -ENOENT;
459
460         kvm->arch.kvm_ops = kvm_ops;
461         return kvmppc_core_init_vm(kvm);
462 err_out:
463         return -EINVAL;
464 }
465
466 bool kvm_arch_has_vcpu_debugfs(void)
467 {
468         return false;
469 }
470
471 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
472 {
473         return 0;
474 }
475
476 void kvm_arch_destroy_vm(struct kvm *kvm)
477 {
478         unsigned int i;
479         struct kvm_vcpu *vcpu;
480
481 #ifdef CONFIG_KVM_XICS
482         /*
483          * We call kick_all_cpus_sync() to ensure that all
484          * CPUs have executed any pending IPIs before we
485          * continue and free VCPUs structures below.
486          */
487         if (is_kvmppc_hv_enabled(kvm))
488                 kick_all_cpus_sync();
489 #endif
490
491         kvm_for_each_vcpu(i, vcpu, kvm)
492                 kvm_arch_vcpu_free(vcpu);
493
494         mutex_lock(&kvm->lock);
495         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
496                 kvm->vcpus[i] = NULL;
497
498         atomic_set(&kvm->online_vcpus, 0);
499
500         kvmppc_core_destroy_vm(kvm);
501
502         mutex_unlock(&kvm->lock);
503
504         /* drop the module reference */
505         module_put(kvm->arch.kvm_ops->owner);
506 }
507
508 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
509 {
510         int r;
511         /* Assume we're using HV mode when the HV module is loaded */
512         int hv_enabled = kvmppc_hv_ops ? 1 : 0;
513
514         if (kvm) {
515                 /*
516                  * Hooray - we know which VM type we're running on. Depend on
517                  * that rather than the guess above.
518                  */
519                 hv_enabled = is_kvmppc_hv_enabled(kvm);
520         }
521
522         switch (ext) {
523 #ifdef CONFIG_BOOKE
524         case KVM_CAP_PPC_BOOKE_SREGS:
525         case KVM_CAP_PPC_BOOKE_WATCHDOG:
526         case KVM_CAP_PPC_EPR:
527 #else
528         case KVM_CAP_PPC_SEGSTATE:
529         case KVM_CAP_PPC_HIOR:
530         case KVM_CAP_PPC_PAPR:
531 #endif
532         case KVM_CAP_PPC_UNSET_IRQ:
533         case KVM_CAP_PPC_IRQ_LEVEL:
534         case KVM_CAP_ENABLE_CAP:
535         case KVM_CAP_ONE_REG:
536         case KVM_CAP_IOEVENTFD:
537         case KVM_CAP_DEVICE_CTRL:
538         case KVM_CAP_IMMEDIATE_EXIT:
539                 r = 1;
540                 break;
541         case KVM_CAP_PPC_PAIRED_SINGLES:
542         case KVM_CAP_PPC_OSI:
543         case KVM_CAP_PPC_GET_PVINFO:
544 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
545         case KVM_CAP_SW_TLB:
546 #endif
547                 /* We support this only for PR */
548                 r = !hv_enabled;
549                 break;
550 #ifdef CONFIG_KVM_MPIC
551         case KVM_CAP_IRQ_MPIC:
552                 r = 1;
553                 break;
554 #endif
555
556 #ifdef CONFIG_PPC_BOOK3S_64
557         case KVM_CAP_SPAPR_TCE:
558         case KVM_CAP_SPAPR_TCE_64:
559                 r = 1;
560                 break;
561         case KVM_CAP_SPAPR_TCE_VFIO:
562                 r = !!cpu_has_feature(CPU_FTR_HVMODE);
563                 break;
564         case KVM_CAP_PPC_RTAS:
565         case KVM_CAP_PPC_FIXUP_HCALL:
566         case KVM_CAP_PPC_ENABLE_HCALL:
567 #ifdef CONFIG_KVM_XICS
568         case KVM_CAP_IRQ_XICS:
569 #endif
570         case KVM_CAP_PPC_GET_CPU_CHAR:
571                 r = 1;
572                 break;
573 #ifdef CONFIG_KVM_XIVE
574         case KVM_CAP_PPC_IRQ_XIVE:
575                 /*
576                  * We need XIVE to be enabled on the platform (implies
577                  * a POWER9 processor) and the PowerNV platform, as
578                  * nested is not yet supported.
579                  */
580                 r = xive_enabled() && !!cpu_has_feature(CPU_FTR_HVMODE);
581                 break;
582 #endif
583
584         case KVM_CAP_PPC_ALLOC_HTAB:
585                 r = hv_enabled;
586                 break;
587 #endif /* CONFIG_PPC_BOOK3S_64 */
588 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
589         case KVM_CAP_PPC_SMT:
590                 r = 0;
591                 if (kvm) {
592                         if (kvm->arch.emul_smt_mode > 1)
593                                 r = kvm->arch.emul_smt_mode;
594                         else
595                                 r = kvm->arch.smt_mode;
596                 } else if (hv_enabled) {
597                         if (cpu_has_feature(CPU_FTR_ARCH_300))
598                                 r = 1;
599                         else
600                                 r = threads_per_subcore;
601                 }
602                 break;
603         case KVM_CAP_PPC_SMT_POSSIBLE:
604                 r = 1;
605                 if (hv_enabled) {
606                         if (!cpu_has_feature(CPU_FTR_ARCH_300))
607                                 r = ((threads_per_subcore << 1) - 1);
608                         else
609                                 /* P9 can emulate dbells, so allow any mode */
610                                 r = 8 | 4 | 2 | 1;
611                 }
612                 break;
613         case KVM_CAP_PPC_RMA:
614                 r = 0;
615                 break;
616         case KVM_CAP_PPC_HWRNG:
617                 r = kvmppc_hwrng_present();
618                 break;
619         case KVM_CAP_PPC_MMU_RADIX:
620                 r = !!(hv_enabled && radix_enabled());
621                 break;
622         case KVM_CAP_PPC_MMU_HASH_V3:
623                 r = !!(hv_enabled && cpu_has_feature(CPU_FTR_ARCH_300) &&
624                        cpu_has_feature(CPU_FTR_HVMODE));
625                 break;
626         case KVM_CAP_PPC_NESTED_HV:
627                 r = !!(hv_enabled && kvmppc_hv_ops->enable_nested &&
628                        !kvmppc_hv_ops->enable_nested(NULL));
629                 break;
630 #endif
631         case KVM_CAP_SYNC_MMU:
632 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
633                 r = hv_enabled;
634 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
635                 r = 1;
636 #else
637                 r = 0;
638 #endif
639                 break;
640 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
641         case KVM_CAP_PPC_HTAB_FD:
642                 r = hv_enabled;
643                 break;
644 #endif
645         case KVM_CAP_NR_VCPUS:
646                 /*
647                  * Recommending a number of CPUs is somewhat arbitrary; we
648                  * return the number of present CPUs for -HV (since a host
649                  * will have secondary threads "offline"), and for other KVM
650                  * implementations just count online CPUs.
651                  */
652                 if (hv_enabled)
653                         r = num_present_cpus();
654                 else
655                         r = num_online_cpus();
656                 break;
657         case KVM_CAP_MAX_VCPUS:
658                 r = KVM_MAX_VCPUS;
659                 break;
660 #ifdef CONFIG_PPC_BOOK3S_64
661         case KVM_CAP_PPC_GET_SMMU_INFO:
662                 r = 1;
663                 break;
664         case KVM_CAP_SPAPR_MULTITCE:
665                 r = 1;
666                 break;
667         case KVM_CAP_SPAPR_RESIZE_HPT:
668                 r = !!hv_enabled;
669                 break;
670 #endif
671 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
672         case KVM_CAP_PPC_FWNMI:
673                 r = hv_enabled;
674                 break;
675 #endif
676 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
677         case KVM_CAP_PPC_HTM:
678                 r = !!(cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM) ||
679                      (hv_enabled && cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST));
680                 break;
681 #endif
682         default:
683                 r = 0;
684                 break;
685         }
686         return r;
687
688 }
689
690 long kvm_arch_dev_ioctl(struct file *filp,
691                         unsigned int ioctl, unsigned long arg)
692 {
693         return -EINVAL;
694 }
695
696 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
697                            struct kvm_memory_slot *dont)
698 {
699         kvmppc_core_free_memslot(kvm, free, dont);
700 }
701
702 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
703                             unsigned long npages)
704 {
705         return kvmppc_core_create_memslot(kvm, slot, npages);
706 }
707
708 int kvm_arch_prepare_memory_region(struct kvm *kvm,
709                                    struct kvm_memory_slot *memslot,
710                                    const struct kvm_userspace_memory_region *mem,
711                                    enum kvm_mr_change change)
712 {
713         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
714 }
715
716 void kvm_arch_commit_memory_region(struct kvm *kvm,
717                                    const struct kvm_userspace_memory_region *mem,
718                                    const struct kvm_memory_slot *old,
719                                    const struct kvm_memory_slot *new,
720                                    enum kvm_mr_change change)
721 {
722         kvmppc_core_commit_memory_region(kvm, mem, old, new, change);
723 }
724
725 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
726                                    struct kvm_memory_slot *slot)
727 {
728         kvmppc_core_flush_memslot(kvm, slot);
729 }
730
731 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
732 {
733         struct kvm_vcpu *vcpu;
734         vcpu = kvmppc_core_vcpu_create(kvm, id);
735         if (!IS_ERR(vcpu)) {
736                 vcpu->arch.wqp = &vcpu->wq;
737                 kvmppc_create_vcpu_debugfs(vcpu, id);
738         }
739         return vcpu;
740 }
741
742 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
743 {
744 }
745
746 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
747 {
748         /* Make sure we're not using the vcpu anymore */
749         hrtimer_cancel(&vcpu->arch.dec_timer);
750
751         kvmppc_remove_vcpu_debugfs(vcpu);
752
753         switch (vcpu->arch.irq_type) {
754         case KVMPPC_IRQ_MPIC:
755                 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
756                 break;
757         case KVMPPC_IRQ_XICS:
758                 if (xics_on_xive())
759                         kvmppc_xive_cleanup_vcpu(vcpu);
760                 else
761                         kvmppc_xics_free_icp(vcpu);
762                 break;
763         case KVMPPC_IRQ_XIVE:
764                 kvmppc_xive_native_cleanup_vcpu(vcpu);
765                 break;
766         }
767
768         kvmppc_core_vcpu_free(vcpu);
769 }
770
771 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
772 {
773         kvm_arch_vcpu_free(vcpu);
774 }
775
776 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
777 {
778         return kvmppc_core_pending_dec(vcpu);
779 }
780
781 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
782 {
783         struct kvm_vcpu *vcpu;
784
785         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
786         kvmppc_decrementer_func(vcpu);
787
788         return HRTIMER_NORESTART;
789 }
790
791 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
792 {
793         int ret;
794
795         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
796         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
797         vcpu->arch.dec_expires = get_tb();
798
799 #ifdef CONFIG_KVM_EXIT_TIMING
800         mutex_init(&vcpu->arch.exit_timing_lock);
801 #endif
802         ret = kvmppc_subarch_vcpu_init(vcpu);
803         return ret;
804 }
805
806 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
807 {
808         kvmppc_mmu_destroy(vcpu);
809         kvmppc_subarch_vcpu_uninit(vcpu);
810 }
811
812 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
813 {
814 #ifdef CONFIG_BOOKE
815         /*
816          * vrsave (formerly usprg0) isn't used by Linux, but may
817          * be used by the guest.
818          *
819          * On non-booke this is associated with Altivec and
820          * is handled by code in book3s.c.
821          */
822         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
823 #endif
824         kvmppc_core_vcpu_load(vcpu, cpu);
825 }
826
827 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
828 {
829         kvmppc_core_vcpu_put(vcpu);
830 #ifdef CONFIG_BOOKE
831         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
832 #endif
833 }
834
835 /*
836  * irq_bypass_add_producer and irq_bypass_del_producer are only
837  * useful if the architecture supports PCI passthrough.
838  * irq_bypass_stop and irq_bypass_start are not needed and so
839  * kvm_ops are not defined for them.
840  */
841 bool kvm_arch_has_irq_bypass(void)
842 {
843         return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
844                 (kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
845 }
846
847 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
848                                      struct irq_bypass_producer *prod)
849 {
850         struct kvm_kernel_irqfd *irqfd =
851                 container_of(cons, struct kvm_kernel_irqfd, consumer);
852         struct kvm *kvm = irqfd->kvm;
853
854         if (kvm->arch.kvm_ops->irq_bypass_add_producer)
855                 return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
856
857         return 0;
858 }
859
860 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
861                                       struct irq_bypass_producer *prod)
862 {
863         struct kvm_kernel_irqfd *irqfd =
864                 container_of(cons, struct kvm_kernel_irqfd, consumer);
865         struct kvm *kvm = irqfd->kvm;
866
867         if (kvm->arch.kvm_ops->irq_bypass_del_producer)
868                 kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
869 }
870
871 #ifdef CONFIG_VSX
872 static inline int kvmppc_get_vsr_dword_offset(int index)
873 {
874         int offset;
875
876         if ((index != 0) && (index != 1))
877                 return -1;
878
879 #ifdef __BIG_ENDIAN
880         offset =  index;
881 #else
882         offset = 1 - index;
883 #endif
884
885         return offset;
886 }
887
888 static inline int kvmppc_get_vsr_word_offset(int index)
889 {
890         int offset;
891
892         if ((index > 3) || (index < 0))
893                 return -1;
894
895 #ifdef __BIG_ENDIAN
896         offset = index;
897 #else
898         offset = 3 - index;
899 #endif
900         return offset;
901 }
902
903 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
904         u64 gpr)
905 {
906         union kvmppc_one_reg val;
907         int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
908         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
909
910         if (offset == -1)
911                 return;
912
913         if (index >= 32) {
914                 val.vval = VCPU_VSX_VR(vcpu, index - 32);
915                 val.vsxval[offset] = gpr;
916                 VCPU_VSX_VR(vcpu, index - 32) = val.vval;
917         } else {
918                 VCPU_VSX_FPR(vcpu, index, offset) = gpr;
919         }
920 }
921
922 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
923         u64 gpr)
924 {
925         union kvmppc_one_reg val;
926         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
927
928         if (index >= 32) {
929                 val.vval = VCPU_VSX_VR(vcpu, index - 32);
930                 val.vsxval[0] = gpr;
931                 val.vsxval[1] = gpr;
932                 VCPU_VSX_VR(vcpu, index - 32) = val.vval;
933         } else {
934                 VCPU_VSX_FPR(vcpu, index, 0) = gpr;
935                 VCPU_VSX_FPR(vcpu, index, 1) = gpr;
936         }
937 }
938
939 static inline void kvmppc_set_vsr_word_dump(struct kvm_vcpu *vcpu,
940         u32 gpr)
941 {
942         union kvmppc_one_reg val;
943         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
944
945         if (index >= 32) {
946                 val.vsx32val[0] = gpr;
947                 val.vsx32val[1] = gpr;
948                 val.vsx32val[2] = gpr;
949                 val.vsx32val[3] = gpr;
950                 VCPU_VSX_VR(vcpu, index - 32) = val.vval;
951         } else {
952                 val.vsx32val[0] = gpr;
953                 val.vsx32val[1] = gpr;
954                 VCPU_VSX_FPR(vcpu, index, 0) = val.vsxval[0];
955                 VCPU_VSX_FPR(vcpu, index, 1) = val.vsxval[0];
956         }
957 }
958
959 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
960         u32 gpr32)
961 {
962         union kvmppc_one_reg val;
963         int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
964         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
965         int dword_offset, word_offset;
966
967         if (offset == -1)
968                 return;
969
970         if (index >= 32) {
971                 val.vval = VCPU_VSX_VR(vcpu, index - 32);
972                 val.vsx32val[offset] = gpr32;
973                 VCPU_VSX_VR(vcpu, index - 32) = val.vval;
974         } else {
975                 dword_offset = offset / 2;
976                 word_offset = offset % 2;
977                 val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
978                 val.vsx32val[word_offset] = gpr32;
979                 VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
980         }
981 }
982 #endif /* CONFIG_VSX */
983
984 #ifdef CONFIG_ALTIVEC
985 static inline int kvmppc_get_vmx_offset_generic(struct kvm_vcpu *vcpu,
986                 int index, int element_size)
987 {
988         int offset;
989         int elts = sizeof(vector128)/element_size;
990
991         if ((index < 0) || (index >= elts))
992                 return -1;
993
994         if (kvmppc_need_byteswap(vcpu))
995                 offset = elts - index - 1;
996         else
997                 offset = index;
998
999         return offset;
1000 }
1001
1002 static inline int kvmppc_get_vmx_dword_offset(struct kvm_vcpu *vcpu,
1003                 int index)
1004 {
1005         return kvmppc_get_vmx_offset_generic(vcpu, index, 8);
1006 }
1007
1008 static inline int kvmppc_get_vmx_word_offset(struct kvm_vcpu *vcpu,
1009                 int index)
1010 {
1011         return kvmppc_get_vmx_offset_generic(vcpu, index, 4);
1012 }
1013
1014 static inline int kvmppc_get_vmx_hword_offset(struct kvm_vcpu *vcpu,
1015                 int index)
1016 {
1017         return kvmppc_get_vmx_offset_generic(vcpu, index, 2);
1018 }
1019
1020 static inline int kvmppc_get_vmx_byte_offset(struct kvm_vcpu *vcpu,
1021                 int index)
1022 {
1023         return kvmppc_get_vmx_offset_generic(vcpu, index, 1);
1024 }
1025
1026
1027 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu,
1028         u64 gpr)
1029 {
1030         union kvmppc_one_reg val;
1031         int offset = kvmppc_get_vmx_dword_offset(vcpu,
1032                         vcpu->arch.mmio_vmx_offset);
1033         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1034
1035         if (offset == -1)
1036                 return;
1037
1038         val.vval = VCPU_VSX_VR(vcpu, index);
1039         val.vsxval[offset] = gpr;
1040         VCPU_VSX_VR(vcpu, index) = val.vval;
1041 }
1042
1043 static inline void kvmppc_set_vmx_word(struct kvm_vcpu *vcpu,
1044         u32 gpr32)
1045 {
1046         union kvmppc_one_reg val;
1047         int offset = kvmppc_get_vmx_word_offset(vcpu,
1048                         vcpu->arch.mmio_vmx_offset);
1049         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1050
1051         if (offset == -1)
1052                 return;
1053
1054         val.vval = VCPU_VSX_VR(vcpu, index);
1055         val.vsx32val[offset] = gpr32;
1056         VCPU_VSX_VR(vcpu, index) = val.vval;
1057 }
1058
1059 static inline void kvmppc_set_vmx_hword(struct kvm_vcpu *vcpu,
1060         u16 gpr16)
1061 {
1062         union kvmppc_one_reg val;
1063         int offset = kvmppc_get_vmx_hword_offset(vcpu,
1064                         vcpu->arch.mmio_vmx_offset);
1065         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1066
1067         if (offset == -1)
1068                 return;
1069
1070         val.vval = VCPU_VSX_VR(vcpu, index);
1071         val.vsx16val[offset] = gpr16;
1072         VCPU_VSX_VR(vcpu, index) = val.vval;
1073 }
1074
1075 static inline void kvmppc_set_vmx_byte(struct kvm_vcpu *vcpu,
1076         u8 gpr8)
1077 {
1078         union kvmppc_one_reg val;
1079         int offset = kvmppc_get_vmx_byte_offset(vcpu,
1080                         vcpu->arch.mmio_vmx_offset);
1081         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1082
1083         if (offset == -1)
1084                 return;
1085
1086         val.vval = VCPU_VSX_VR(vcpu, index);
1087         val.vsx8val[offset] = gpr8;
1088         VCPU_VSX_VR(vcpu, index) = val.vval;
1089 }
1090 #endif /* CONFIG_ALTIVEC */
1091
1092 #ifdef CONFIG_PPC_FPU
1093 static inline u64 sp_to_dp(u32 fprs)
1094 {
1095         u64 fprd;
1096
1097         preempt_disable();
1098         enable_kernel_fp();
1099         asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs)
1100              : "fr0");
1101         preempt_enable();
1102         return fprd;
1103 }
1104
1105 static inline u32 dp_to_sp(u64 fprd)
1106 {
1107         u32 fprs;
1108
1109         preempt_disable();
1110         enable_kernel_fp();
1111         asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd)
1112              : "fr0");
1113         preempt_enable();
1114         return fprs;
1115 }
1116
1117 #else
1118 #define sp_to_dp(x)     (x)
1119 #define dp_to_sp(x)     (x)
1120 #endif /* CONFIG_PPC_FPU */
1121
1122 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
1123                                       struct kvm_run *run)
1124 {
1125         u64 uninitialized_var(gpr);
1126
1127         if (run->mmio.len > sizeof(gpr)) {
1128                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
1129                 return;
1130         }
1131
1132         if (!vcpu->arch.mmio_host_swabbed) {
1133                 switch (run->mmio.len) {
1134                 case 8: gpr = *(u64 *)run->mmio.data; break;
1135                 case 4: gpr = *(u32 *)run->mmio.data; break;
1136                 case 2: gpr = *(u16 *)run->mmio.data; break;
1137                 case 1: gpr = *(u8 *)run->mmio.data; break;
1138                 }
1139         } else {
1140                 switch (run->mmio.len) {
1141                 case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
1142                 case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
1143                 case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
1144                 case 1: gpr = *(u8 *)run->mmio.data; break;
1145                 }
1146         }
1147
1148         /* conversion between single and double precision */
1149         if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
1150                 gpr = sp_to_dp(gpr);
1151
1152         if (vcpu->arch.mmio_sign_extend) {
1153                 switch (run->mmio.len) {
1154 #ifdef CONFIG_PPC64
1155                 case 4:
1156                         gpr = (s64)(s32)gpr;
1157                         break;
1158 #endif
1159                 case 2:
1160                         gpr = (s64)(s16)gpr;
1161                         break;
1162                 case 1:
1163                         gpr = (s64)(s8)gpr;
1164                         break;
1165                 }
1166         }
1167
1168         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1169         case KVM_MMIO_REG_GPR:
1170                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1171                 break;
1172         case KVM_MMIO_REG_FPR:
1173                 if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1174                         vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_FP);
1175
1176                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1177                 break;
1178 #ifdef CONFIG_PPC_BOOK3S
1179         case KVM_MMIO_REG_QPR:
1180                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1181                 break;
1182         case KVM_MMIO_REG_FQPR:
1183                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1184                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1185                 break;
1186 #endif
1187 #ifdef CONFIG_VSX
1188         case KVM_MMIO_REG_VSX:
1189                 if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1190                         vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VSX);
1191
1192                 if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_DWORD)
1193                         kvmppc_set_vsr_dword(vcpu, gpr);
1194                 else if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_WORD)
1195                         kvmppc_set_vsr_word(vcpu, gpr);
1196                 else if (vcpu->arch.mmio_copy_type ==
1197                                 KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1198                         kvmppc_set_vsr_dword_dump(vcpu, gpr);
1199                 else if (vcpu->arch.mmio_copy_type ==
1200                                 KVMPPC_VSX_COPY_WORD_LOAD_DUMP)
1201                         kvmppc_set_vsr_word_dump(vcpu, gpr);
1202                 break;
1203 #endif
1204 #ifdef CONFIG_ALTIVEC
1205         case KVM_MMIO_REG_VMX:
1206                 if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1207                         vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VEC);
1208
1209                 if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_DWORD)
1210                         kvmppc_set_vmx_dword(vcpu, gpr);
1211                 else if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_WORD)
1212                         kvmppc_set_vmx_word(vcpu, gpr);
1213                 else if (vcpu->arch.mmio_copy_type ==
1214                                 KVMPPC_VMX_COPY_HWORD)
1215                         kvmppc_set_vmx_hword(vcpu, gpr);
1216                 else if (vcpu->arch.mmio_copy_type ==
1217                                 KVMPPC_VMX_COPY_BYTE)
1218                         kvmppc_set_vmx_byte(vcpu, gpr);
1219                 break;
1220 #endif
1221 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1222         case KVM_MMIO_REG_NESTED_GPR:
1223                 if (kvmppc_need_byteswap(vcpu))
1224                         gpr = swab64(gpr);
1225                 kvm_vcpu_write_guest(vcpu, vcpu->arch.nested_io_gpr, &gpr,
1226                                      sizeof(gpr));
1227                 break;
1228 #endif
1229         default:
1230                 BUG();
1231         }
1232 }
1233
1234 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1235                                 unsigned int rt, unsigned int bytes,
1236                                 int is_default_endian, int sign_extend)
1237 {
1238         int idx, ret;
1239         bool host_swabbed;
1240
1241         /* Pity C doesn't have a logical XOR operator */
1242         if (kvmppc_need_byteswap(vcpu)) {
1243                 host_swabbed = is_default_endian;
1244         } else {
1245                 host_swabbed = !is_default_endian;
1246         }
1247
1248         if (bytes > sizeof(run->mmio.data)) {
1249                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1250                        run->mmio.len);
1251         }
1252
1253         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1254         run->mmio.len = bytes;
1255         run->mmio.is_write = 0;
1256
1257         vcpu->arch.io_gpr = rt;
1258         vcpu->arch.mmio_host_swabbed = host_swabbed;
1259         vcpu->mmio_needed = 1;
1260         vcpu->mmio_is_write = 0;
1261         vcpu->arch.mmio_sign_extend = sign_extend;
1262
1263         idx = srcu_read_lock(&vcpu->kvm->srcu);
1264
1265         ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1266                               bytes, &run->mmio.data);
1267
1268         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1269
1270         if (!ret) {
1271                 kvmppc_complete_mmio_load(vcpu, run);
1272                 vcpu->mmio_needed = 0;
1273                 return EMULATE_DONE;
1274         }
1275
1276         return EMULATE_DO_MMIO;
1277 }
1278
1279 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1280                        unsigned int rt, unsigned int bytes,
1281                        int is_default_endian)
1282 {
1283         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
1284 }
1285 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1286
1287 /* Same as above, but sign extends */
1288 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
1289                         unsigned int rt, unsigned int bytes,
1290                         int is_default_endian)
1291 {
1292         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
1293 }
1294
1295 #ifdef CONFIG_VSX
1296 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1297                         unsigned int rt, unsigned int bytes,
1298                         int is_default_endian, int mmio_sign_extend)
1299 {
1300         enum emulation_result emulated = EMULATE_DONE;
1301
1302         /* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1303         if (vcpu->arch.mmio_vsx_copy_nums > 4)
1304                 return EMULATE_FAIL;
1305
1306         while (vcpu->arch.mmio_vsx_copy_nums) {
1307                 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1308                         is_default_endian, mmio_sign_extend);
1309
1310                 if (emulated != EMULATE_DONE)
1311                         break;
1312
1313                 vcpu->arch.paddr_accessed += run->mmio.len;
1314
1315                 vcpu->arch.mmio_vsx_copy_nums--;
1316                 vcpu->arch.mmio_vsx_offset++;
1317         }
1318         return emulated;
1319 }
1320 #endif /* CONFIG_VSX */
1321
1322 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1323                         u64 val, unsigned int bytes, int is_default_endian)
1324 {
1325         void *data = run->mmio.data;
1326         int idx, ret;
1327         bool host_swabbed;
1328
1329         /* Pity C doesn't have a logical XOR operator */
1330         if (kvmppc_need_byteswap(vcpu)) {
1331                 host_swabbed = is_default_endian;
1332         } else {
1333                 host_swabbed = !is_default_endian;
1334         }
1335
1336         if (bytes > sizeof(run->mmio.data)) {
1337                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1338                        run->mmio.len);
1339         }
1340
1341         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1342         run->mmio.len = bytes;
1343         run->mmio.is_write = 1;
1344         vcpu->mmio_needed = 1;
1345         vcpu->mmio_is_write = 1;
1346
1347         if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1348                 val = dp_to_sp(val);
1349
1350         /* Store the value at the lowest bytes in 'data'. */
1351         if (!host_swabbed) {
1352                 switch (bytes) {
1353                 case 8: *(u64 *)data = val; break;
1354                 case 4: *(u32 *)data = val; break;
1355                 case 2: *(u16 *)data = val; break;
1356                 case 1: *(u8  *)data = val; break;
1357                 }
1358         } else {
1359                 switch (bytes) {
1360                 case 8: *(u64 *)data = swab64(val); break;
1361                 case 4: *(u32 *)data = swab32(val); break;
1362                 case 2: *(u16 *)data = swab16(val); break;
1363                 case 1: *(u8  *)data = val; break;
1364                 }
1365         }
1366
1367         idx = srcu_read_lock(&vcpu->kvm->srcu);
1368
1369         ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1370                                bytes, &run->mmio.data);
1371
1372         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1373
1374         if (!ret) {
1375                 vcpu->mmio_needed = 0;
1376                 return EMULATE_DONE;
1377         }
1378
1379         return EMULATE_DO_MMIO;
1380 }
1381 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1382
1383 #ifdef CONFIG_VSX
1384 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1385 {
1386         u32 dword_offset, word_offset;
1387         union kvmppc_one_reg reg;
1388         int vsx_offset = 0;
1389         int copy_type = vcpu->arch.mmio_copy_type;
1390         int result = 0;
1391
1392         switch (copy_type) {
1393         case KVMPPC_VSX_COPY_DWORD:
1394                 vsx_offset =
1395                         kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1396
1397                 if (vsx_offset == -1) {
1398                         result = -1;
1399                         break;
1400                 }
1401
1402                 if (rs < 32) {
1403                         *val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1404                 } else {
1405                         reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1406                         *val = reg.vsxval[vsx_offset];
1407                 }
1408                 break;
1409
1410         case KVMPPC_VSX_COPY_WORD:
1411                 vsx_offset =
1412                         kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1413
1414                 if (vsx_offset == -1) {
1415                         result = -1;
1416                         break;
1417                 }
1418
1419                 if (rs < 32) {
1420                         dword_offset = vsx_offset / 2;
1421                         word_offset = vsx_offset % 2;
1422                         reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1423                         *val = reg.vsx32val[word_offset];
1424                 } else {
1425                         reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1426                         *val = reg.vsx32val[vsx_offset];
1427                 }
1428                 break;
1429
1430         default:
1431                 result = -1;
1432                 break;
1433         }
1434
1435         return result;
1436 }
1437
1438 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1439                         int rs, unsigned int bytes, int is_default_endian)
1440 {
1441         u64 val;
1442         enum emulation_result emulated = EMULATE_DONE;
1443
1444         vcpu->arch.io_gpr = rs;
1445
1446         /* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1447         if (vcpu->arch.mmio_vsx_copy_nums > 4)
1448                 return EMULATE_FAIL;
1449
1450         while (vcpu->arch.mmio_vsx_copy_nums) {
1451                 if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1452                         return EMULATE_FAIL;
1453
1454                 emulated = kvmppc_handle_store(run, vcpu,
1455                          val, bytes, is_default_endian);
1456
1457                 if (emulated != EMULATE_DONE)
1458                         break;
1459
1460                 vcpu->arch.paddr_accessed += run->mmio.len;
1461
1462                 vcpu->arch.mmio_vsx_copy_nums--;
1463                 vcpu->arch.mmio_vsx_offset++;
1464         }
1465
1466         return emulated;
1467 }
1468
1469 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu,
1470                         struct kvm_run *run)
1471 {
1472         enum emulation_result emulated = EMULATE_FAIL;
1473         int r;
1474
1475         vcpu->arch.paddr_accessed += run->mmio.len;
1476
1477         if (!vcpu->mmio_is_write) {
1478                 emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr,
1479                          run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1480         } else {
1481                 emulated = kvmppc_handle_vsx_store(run, vcpu,
1482                          vcpu->arch.io_gpr, run->mmio.len, 1);
1483         }
1484
1485         switch (emulated) {
1486         case EMULATE_DO_MMIO:
1487                 run->exit_reason = KVM_EXIT_MMIO;
1488                 r = RESUME_HOST;
1489                 break;
1490         case EMULATE_FAIL:
1491                 pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1492                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1493                 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1494                 r = RESUME_HOST;
1495                 break;
1496         default:
1497                 r = RESUME_GUEST;
1498                 break;
1499         }
1500         return r;
1501 }
1502 #endif /* CONFIG_VSX */
1503
1504 #ifdef CONFIG_ALTIVEC
1505 int kvmppc_handle_vmx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1506                 unsigned int rt, unsigned int bytes, int is_default_endian)
1507 {
1508         enum emulation_result emulated = EMULATE_DONE;
1509
1510         if (vcpu->arch.mmio_vsx_copy_nums > 2)
1511                 return EMULATE_FAIL;
1512
1513         while (vcpu->arch.mmio_vmx_copy_nums) {
1514                 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1515                                 is_default_endian, 0);
1516
1517                 if (emulated != EMULATE_DONE)
1518                         break;
1519
1520                 vcpu->arch.paddr_accessed += run->mmio.len;
1521                 vcpu->arch.mmio_vmx_copy_nums--;
1522                 vcpu->arch.mmio_vmx_offset++;
1523         }
1524
1525         return emulated;
1526 }
1527
1528 int kvmppc_get_vmx_dword(struct kvm_vcpu *vcpu, int index, u64 *val)
1529 {
1530         union kvmppc_one_reg reg;
1531         int vmx_offset = 0;
1532         int result = 0;
1533
1534         vmx_offset =
1535                 kvmppc_get_vmx_dword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1536
1537         if (vmx_offset == -1)
1538                 return -1;
1539
1540         reg.vval = VCPU_VSX_VR(vcpu, index);
1541         *val = reg.vsxval[vmx_offset];
1542
1543         return result;
1544 }
1545
1546 int kvmppc_get_vmx_word(struct kvm_vcpu *vcpu, int index, u64 *val)
1547 {
1548         union kvmppc_one_reg reg;
1549         int vmx_offset = 0;
1550         int result = 0;
1551
1552         vmx_offset =
1553                 kvmppc_get_vmx_word_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1554
1555         if (vmx_offset == -1)
1556                 return -1;
1557
1558         reg.vval = VCPU_VSX_VR(vcpu, index);
1559         *val = reg.vsx32val[vmx_offset];
1560
1561         return result;
1562 }
1563
1564 int kvmppc_get_vmx_hword(struct kvm_vcpu *vcpu, int index, u64 *val)
1565 {
1566         union kvmppc_one_reg reg;
1567         int vmx_offset = 0;
1568         int result = 0;
1569
1570         vmx_offset =
1571                 kvmppc_get_vmx_hword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1572
1573         if (vmx_offset == -1)
1574                 return -1;
1575
1576         reg.vval = VCPU_VSX_VR(vcpu, index);
1577         *val = reg.vsx16val[vmx_offset];
1578
1579         return result;
1580 }
1581
1582 int kvmppc_get_vmx_byte(struct kvm_vcpu *vcpu, int index, u64 *val)
1583 {
1584         union kvmppc_one_reg reg;
1585         int vmx_offset = 0;
1586         int result = 0;
1587
1588         vmx_offset =
1589                 kvmppc_get_vmx_byte_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1590
1591         if (vmx_offset == -1)
1592                 return -1;
1593
1594         reg.vval = VCPU_VSX_VR(vcpu, index);
1595         *val = reg.vsx8val[vmx_offset];
1596
1597         return result;
1598 }
1599
1600 int kvmppc_handle_vmx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1601                 unsigned int rs, unsigned int bytes, int is_default_endian)
1602 {
1603         u64 val = 0;
1604         unsigned int index = rs & KVM_MMIO_REG_MASK;
1605         enum emulation_result emulated = EMULATE_DONE;
1606
1607         if (vcpu->arch.mmio_vsx_copy_nums > 2)
1608                 return EMULATE_FAIL;
1609
1610         vcpu->arch.io_gpr = rs;
1611
1612         while (vcpu->arch.mmio_vmx_copy_nums) {
1613                 switch (vcpu->arch.mmio_copy_type) {
1614                 case KVMPPC_VMX_COPY_DWORD:
1615                         if (kvmppc_get_vmx_dword(vcpu, index, &val) == -1)
1616                                 return EMULATE_FAIL;
1617
1618                         break;
1619                 case KVMPPC_VMX_COPY_WORD:
1620                         if (kvmppc_get_vmx_word(vcpu, index, &val) == -1)
1621                                 return EMULATE_FAIL;
1622                         break;
1623                 case KVMPPC_VMX_COPY_HWORD:
1624                         if (kvmppc_get_vmx_hword(vcpu, index, &val) == -1)
1625                                 return EMULATE_FAIL;
1626                         break;
1627                 case KVMPPC_VMX_COPY_BYTE:
1628                         if (kvmppc_get_vmx_byte(vcpu, index, &val) == -1)
1629                                 return EMULATE_FAIL;
1630                         break;
1631                 default:
1632                         return EMULATE_FAIL;
1633                 }
1634
1635                 emulated = kvmppc_handle_store(run, vcpu, val, bytes,
1636                                 is_default_endian);
1637                 if (emulated != EMULATE_DONE)
1638                         break;
1639
1640                 vcpu->arch.paddr_accessed += run->mmio.len;
1641                 vcpu->arch.mmio_vmx_copy_nums--;
1642                 vcpu->arch.mmio_vmx_offset++;
1643         }
1644
1645         return emulated;
1646 }
1647
1648 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu,
1649                 struct kvm_run *run)
1650 {
1651         enum emulation_result emulated = EMULATE_FAIL;
1652         int r;
1653
1654         vcpu->arch.paddr_accessed += run->mmio.len;
1655
1656         if (!vcpu->mmio_is_write) {
1657                 emulated = kvmppc_handle_vmx_load(run, vcpu,
1658                                 vcpu->arch.io_gpr, run->mmio.len, 1);
1659         } else {
1660                 emulated = kvmppc_handle_vmx_store(run, vcpu,
1661                                 vcpu->arch.io_gpr, run->mmio.len, 1);
1662         }
1663
1664         switch (emulated) {
1665         case EMULATE_DO_MMIO:
1666                 run->exit_reason = KVM_EXIT_MMIO;
1667                 r = RESUME_HOST;
1668                 break;
1669         case EMULATE_FAIL:
1670                 pr_info("KVM: MMIO emulation failed (VMX repeat)\n");
1671                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1672                 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1673                 r = RESUME_HOST;
1674                 break;
1675         default:
1676                 r = RESUME_GUEST;
1677                 break;
1678         }
1679         return r;
1680 }
1681 #endif /* CONFIG_ALTIVEC */
1682
1683 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1684 {
1685         int r = 0;
1686         union kvmppc_one_reg val;
1687         int size;
1688
1689         size = one_reg_size(reg->id);
1690         if (size > sizeof(val))
1691                 return -EINVAL;
1692
1693         r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1694         if (r == -EINVAL) {
1695                 r = 0;
1696                 switch (reg->id) {
1697 #ifdef CONFIG_ALTIVEC
1698                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1699                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1700                                 r = -ENXIO;
1701                                 break;
1702                         }
1703                         val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1704                         break;
1705                 case KVM_REG_PPC_VSCR:
1706                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1707                                 r = -ENXIO;
1708                                 break;
1709                         }
1710                         val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1711                         break;
1712                 case KVM_REG_PPC_VRSAVE:
1713                         val = get_reg_val(reg->id, vcpu->arch.vrsave);
1714                         break;
1715 #endif /* CONFIG_ALTIVEC */
1716                 default:
1717                         r = -EINVAL;
1718                         break;
1719                 }
1720         }
1721
1722         if (r)
1723                 return r;
1724
1725         if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1726                 r = -EFAULT;
1727
1728         return r;
1729 }
1730
1731 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1732 {
1733         int r;
1734         union kvmppc_one_reg val;
1735         int size;
1736
1737         size = one_reg_size(reg->id);
1738         if (size > sizeof(val))
1739                 return -EINVAL;
1740
1741         if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1742                 return -EFAULT;
1743
1744         r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1745         if (r == -EINVAL) {
1746                 r = 0;
1747                 switch (reg->id) {
1748 #ifdef CONFIG_ALTIVEC
1749                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1750                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1751                                 r = -ENXIO;
1752                                 break;
1753                         }
1754                         vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1755                         break;
1756                 case KVM_REG_PPC_VSCR:
1757                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1758                                 r = -ENXIO;
1759                                 break;
1760                         }
1761                         vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1762                         break;
1763                 case KVM_REG_PPC_VRSAVE:
1764                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1765                                 r = -ENXIO;
1766                                 break;
1767                         }
1768                         vcpu->arch.vrsave = set_reg_val(reg->id, val);
1769                         break;
1770 #endif /* CONFIG_ALTIVEC */
1771                 default:
1772                         r = -EINVAL;
1773                         break;
1774                 }
1775         }
1776
1777         return r;
1778 }
1779
1780 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1781 {
1782         int r;
1783
1784         vcpu_load(vcpu);
1785
1786         if (vcpu->mmio_needed) {
1787                 vcpu->mmio_needed = 0;
1788                 if (!vcpu->mmio_is_write)
1789                         kvmppc_complete_mmio_load(vcpu, run);
1790 #ifdef CONFIG_VSX
1791                 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1792                         vcpu->arch.mmio_vsx_copy_nums--;
1793                         vcpu->arch.mmio_vsx_offset++;
1794                 }
1795
1796                 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1797                         r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run);
1798                         if (r == RESUME_HOST) {
1799                                 vcpu->mmio_needed = 1;
1800                                 goto out;
1801                         }
1802                 }
1803 #endif
1804 #ifdef CONFIG_ALTIVEC
1805                 if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1806                         vcpu->arch.mmio_vmx_copy_nums--;
1807                         vcpu->arch.mmio_vmx_offset++;
1808                 }
1809
1810                 if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1811                         r = kvmppc_emulate_mmio_vmx_loadstore(vcpu, run);
1812                         if (r == RESUME_HOST) {
1813                                 vcpu->mmio_needed = 1;
1814                                 goto out;
1815                         }
1816                 }
1817 #endif
1818         } else if (vcpu->arch.osi_needed) {
1819                 u64 *gprs = run->osi.gprs;
1820                 int i;
1821
1822                 for (i = 0; i < 32; i++)
1823                         kvmppc_set_gpr(vcpu, i, gprs[i]);
1824                 vcpu->arch.osi_needed = 0;
1825         } else if (vcpu->arch.hcall_needed) {
1826                 int i;
1827
1828                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1829                 for (i = 0; i < 9; ++i)
1830                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1831                 vcpu->arch.hcall_needed = 0;
1832 #ifdef CONFIG_BOOKE
1833         } else if (vcpu->arch.epr_needed) {
1834                 kvmppc_set_epr(vcpu, run->epr.epr);
1835                 vcpu->arch.epr_needed = 0;
1836 #endif
1837         }
1838
1839         kvm_sigset_activate(vcpu);
1840
1841         if (run->immediate_exit)
1842                 r = -EINTR;
1843         else
1844                 r = kvmppc_vcpu_run(run, vcpu);
1845
1846         kvm_sigset_deactivate(vcpu);
1847
1848 #ifdef CONFIG_ALTIVEC
1849 out:
1850 #endif
1851         vcpu_put(vcpu);
1852         return r;
1853 }
1854
1855 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1856 {
1857         if (irq->irq == KVM_INTERRUPT_UNSET) {
1858                 kvmppc_core_dequeue_external(vcpu);
1859                 return 0;
1860         }
1861
1862         kvmppc_core_queue_external(vcpu, irq);
1863
1864         kvm_vcpu_kick(vcpu);
1865
1866         return 0;
1867 }
1868
1869 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1870                                      struct kvm_enable_cap *cap)
1871 {
1872         int r;
1873
1874         if (cap->flags)
1875                 return -EINVAL;
1876
1877         switch (cap->cap) {
1878         case KVM_CAP_PPC_OSI:
1879                 r = 0;
1880                 vcpu->arch.osi_enabled = true;
1881                 break;
1882         case KVM_CAP_PPC_PAPR:
1883                 r = 0;
1884                 vcpu->arch.papr_enabled = true;
1885                 break;
1886         case KVM_CAP_PPC_EPR:
1887                 r = 0;
1888                 if (cap->args[0])
1889                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1890                 else
1891                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1892                 break;
1893 #ifdef CONFIG_BOOKE
1894         case KVM_CAP_PPC_BOOKE_WATCHDOG:
1895                 r = 0;
1896                 vcpu->arch.watchdog_enabled = true;
1897                 break;
1898 #endif
1899 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1900         case KVM_CAP_SW_TLB: {
1901                 struct kvm_config_tlb cfg;
1902                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1903
1904                 r = -EFAULT;
1905                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1906                         break;
1907
1908                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1909                 break;
1910         }
1911 #endif
1912 #ifdef CONFIG_KVM_MPIC
1913         case KVM_CAP_IRQ_MPIC: {
1914                 struct fd f;
1915                 struct kvm_device *dev;
1916
1917                 r = -EBADF;
1918                 f = fdget(cap->args[0]);
1919                 if (!f.file)
1920                         break;
1921
1922                 r = -EPERM;
1923                 dev = kvm_device_from_filp(f.file);
1924                 if (dev)
1925                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1926
1927                 fdput(f);
1928                 break;
1929         }
1930 #endif
1931 #ifdef CONFIG_KVM_XICS
1932         case KVM_CAP_IRQ_XICS: {
1933                 struct fd f;
1934                 struct kvm_device *dev;
1935
1936                 r = -EBADF;
1937                 f = fdget(cap->args[0]);
1938                 if (!f.file)
1939                         break;
1940
1941                 r = -EPERM;
1942                 dev = kvm_device_from_filp(f.file);
1943                 if (dev) {
1944                         if (xics_on_xive())
1945                                 r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1946                         else
1947                                 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1948                 }
1949
1950                 fdput(f);
1951                 break;
1952         }
1953 #endif /* CONFIG_KVM_XICS */
1954 #ifdef CONFIG_KVM_XIVE
1955         case KVM_CAP_PPC_IRQ_XIVE: {
1956                 struct fd f;
1957                 struct kvm_device *dev;
1958
1959                 r = -EBADF;
1960                 f = fdget(cap->args[0]);
1961                 if (!f.file)
1962                         break;
1963
1964                 r = -ENXIO;
1965                 if (!xive_enabled())
1966                         break;
1967
1968                 r = -EPERM;
1969                 dev = kvm_device_from_filp(f.file);
1970                 if (dev)
1971                         r = kvmppc_xive_native_connect_vcpu(dev, vcpu,
1972                                                             cap->args[1]);
1973
1974                 fdput(f);
1975                 break;
1976         }
1977 #endif /* CONFIG_KVM_XIVE */
1978 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1979         case KVM_CAP_PPC_FWNMI:
1980                 r = -EINVAL;
1981                 if (!is_kvmppc_hv_enabled(vcpu->kvm))
1982                         break;
1983                 r = 0;
1984                 vcpu->kvm->arch.fwnmi_enabled = true;
1985                 break;
1986 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1987         default:
1988                 r = -EINVAL;
1989                 break;
1990         }
1991
1992         if (!r)
1993                 r = kvmppc_sanity_check(vcpu);
1994
1995         return r;
1996 }
1997
1998 bool kvm_arch_intc_initialized(struct kvm *kvm)
1999 {
2000 #ifdef CONFIG_KVM_MPIC
2001         if (kvm->arch.mpic)
2002                 return true;
2003 #endif
2004 #ifdef CONFIG_KVM_XICS
2005         if (kvm->arch.xics || kvm->arch.xive)
2006                 return true;
2007 #endif
2008         return false;
2009 }
2010
2011 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2012                                     struct kvm_mp_state *mp_state)
2013 {
2014         return -EINVAL;
2015 }
2016
2017 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2018                                     struct kvm_mp_state *mp_state)
2019 {
2020         return -EINVAL;
2021 }
2022
2023 long kvm_arch_vcpu_async_ioctl(struct file *filp,
2024                                unsigned int ioctl, unsigned long arg)
2025 {
2026         struct kvm_vcpu *vcpu = filp->private_data;
2027         void __user *argp = (void __user *)arg;
2028
2029         if (ioctl == KVM_INTERRUPT) {
2030                 struct kvm_interrupt irq;
2031                 if (copy_from_user(&irq, argp, sizeof(irq)))
2032                         return -EFAULT;
2033                 return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2034         }
2035         return -ENOIOCTLCMD;
2036 }
2037
2038 long kvm_arch_vcpu_ioctl(struct file *filp,
2039                          unsigned int ioctl, unsigned long arg)
2040 {
2041         struct kvm_vcpu *vcpu = filp->private_data;
2042         void __user *argp = (void __user *)arg;
2043         long r;
2044
2045         switch (ioctl) {
2046         case KVM_ENABLE_CAP:
2047         {
2048                 struct kvm_enable_cap cap;
2049                 r = -EFAULT;
2050                 vcpu_load(vcpu);
2051                 if (copy_from_user(&cap, argp, sizeof(cap)))
2052                         goto out;
2053                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
2054                 vcpu_put(vcpu);
2055                 break;
2056         }
2057
2058         case KVM_SET_ONE_REG:
2059         case KVM_GET_ONE_REG:
2060         {
2061                 struct kvm_one_reg reg;
2062                 r = -EFAULT;
2063                 if (copy_from_user(&reg, argp, sizeof(reg)))
2064                         goto out;
2065                 if (ioctl == KVM_SET_ONE_REG)
2066                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
2067                 else
2068                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
2069                 break;
2070         }
2071
2072 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
2073         case KVM_DIRTY_TLB: {
2074                 struct kvm_dirty_tlb dirty;
2075                 r = -EFAULT;
2076                 vcpu_load(vcpu);
2077                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
2078                         goto out;
2079                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
2080                 vcpu_put(vcpu);
2081                 break;
2082         }
2083 #endif
2084         default:
2085                 r = -EINVAL;
2086         }
2087
2088 out:
2089         return r;
2090 }
2091
2092 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2093 {
2094         return VM_FAULT_SIGBUS;
2095 }
2096
2097 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
2098 {
2099         u32 inst_nop = 0x60000000;
2100 #ifdef CONFIG_KVM_BOOKE_HV
2101         u32 inst_sc1 = 0x44000022;
2102         pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
2103         pvinfo->hcall[1] = cpu_to_be32(inst_nop);
2104         pvinfo->hcall[2] = cpu_to_be32(inst_nop);
2105         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2106 #else
2107         u32 inst_lis = 0x3c000000;
2108         u32 inst_ori = 0x60000000;
2109         u32 inst_sc = 0x44000002;
2110         u32 inst_imm_mask = 0xffff;
2111
2112         /*
2113          * The hypercall to get into KVM from within guest context is as
2114          * follows:
2115          *
2116          *    lis r0, r0, KVM_SC_MAGIC_R0@h
2117          *    ori r0, KVM_SC_MAGIC_R0@l
2118          *    sc
2119          *    nop
2120          */
2121         pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
2122         pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
2123         pvinfo->hcall[2] = cpu_to_be32(inst_sc);
2124         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2125 #endif
2126
2127         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
2128
2129         return 0;
2130 }
2131
2132 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
2133                           bool line_status)
2134 {
2135         if (!irqchip_in_kernel(kvm))
2136                 return -ENXIO;
2137
2138         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2139                                         irq_event->irq, irq_event->level,
2140                                         line_status);
2141         return 0;
2142 }
2143
2144
2145 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
2146                             struct kvm_enable_cap *cap)
2147 {
2148         int r;
2149
2150         if (cap->flags)
2151                 return -EINVAL;
2152
2153         switch (cap->cap) {
2154 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
2155         case KVM_CAP_PPC_ENABLE_HCALL: {
2156                 unsigned long hcall = cap->args[0];
2157
2158                 r = -EINVAL;
2159                 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
2160                     cap->args[1] > 1)
2161                         break;
2162                 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
2163                         break;
2164                 if (cap->args[1])
2165                         set_bit(hcall / 4, kvm->arch.enabled_hcalls);
2166                 else
2167                         clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
2168                 r = 0;
2169                 break;
2170         }
2171         case KVM_CAP_PPC_SMT: {
2172                 unsigned long mode = cap->args[0];
2173                 unsigned long flags = cap->args[1];
2174
2175                 r = -EINVAL;
2176                 if (kvm->arch.kvm_ops->set_smt_mode)
2177                         r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
2178                 break;
2179         }
2180
2181         case KVM_CAP_PPC_NESTED_HV:
2182                 r = -EINVAL;
2183                 if (!is_kvmppc_hv_enabled(kvm) ||
2184                     !kvm->arch.kvm_ops->enable_nested)
2185                         break;
2186                 r = kvm->arch.kvm_ops->enable_nested(kvm);
2187                 break;
2188 #endif
2189         default:
2190                 r = -EINVAL;
2191                 break;
2192         }
2193
2194         return r;
2195 }
2196
2197 #ifdef CONFIG_PPC_BOOK3S_64
2198 /*
2199  * These functions check whether the underlying hardware is safe
2200  * against attacks based on observing the effects of speculatively
2201  * executed instructions, and whether it supplies instructions for
2202  * use in workarounds.  The information comes from firmware, either
2203  * via the device tree on powernv platforms or from an hcall on
2204  * pseries platforms.
2205  */
2206 #ifdef CONFIG_PPC_PSERIES
2207 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2208 {
2209         struct h_cpu_char_result c;
2210         unsigned long rc;
2211
2212         if (!machine_is(pseries))
2213                 return -ENOTTY;
2214
2215         rc = plpar_get_cpu_characteristics(&c);
2216         if (rc == H_SUCCESS) {
2217                 cp->character = c.character;
2218                 cp->behaviour = c.behaviour;
2219                 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2220                         KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2221                         KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2222                         KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2223                         KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2224                         KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
2225                         KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
2226                         KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2227                         KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2228                 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2229                         KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2230                         KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2231                         KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2232         }
2233         return 0;
2234 }
2235 #else
2236 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2237 {
2238         return -ENOTTY;
2239 }
2240 #endif
2241
2242 static inline bool have_fw_feat(struct device_node *fw_features,
2243                                 const char *state, const char *name)
2244 {
2245         struct device_node *np;
2246         bool r = false;
2247
2248         np = of_get_child_by_name(fw_features, name);
2249         if (np) {
2250                 r = of_property_read_bool(np, state);
2251                 of_node_put(np);
2252         }
2253         return r;
2254 }
2255
2256 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2257 {
2258         struct device_node *np, *fw_features;
2259         int r;
2260
2261         memset(cp, 0, sizeof(*cp));
2262         r = pseries_get_cpu_char(cp);
2263         if (r != -ENOTTY)
2264                 return r;
2265
2266         np = of_find_node_by_name(NULL, "ibm,opal");
2267         if (np) {
2268                 fw_features = of_get_child_by_name(np, "fw-features");
2269                 of_node_put(np);
2270                 if (!fw_features)
2271                         return 0;
2272                 if (have_fw_feat(fw_features, "enabled",
2273                                  "inst-spec-barrier-ori31,31,0"))
2274                         cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
2275                 if (have_fw_feat(fw_features, "enabled",
2276                                  "fw-bcctrl-serialized"))
2277                         cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
2278                 if (have_fw_feat(fw_features, "enabled",
2279                                  "inst-l1d-flush-ori30,30,0"))
2280                         cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
2281                 if (have_fw_feat(fw_features, "enabled",
2282                                  "inst-l1d-flush-trig2"))
2283                         cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
2284                 if (have_fw_feat(fw_features, "enabled",
2285                                  "fw-l1d-thread-split"))
2286                         cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
2287                 if (have_fw_feat(fw_features, "enabled",
2288                                  "fw-count-cache-disabled"))
2289                         cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2290                 if (have_fw_feat(fw_features, "enabled",
2291                                  "fw-count-cache-flush-bcctr2,0,0"))
2292                         cp->character |= KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2293                 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2294                         KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2295                         KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2296                         KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2297                         KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2298                         KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2299                         KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2300
2301                 if (have_fw_feat(fw_features, "enabled",
2302                                  "speculation-policy-favor-security"))
2303                         cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
2304                 if (!have_fw_feat(fw_features, "disabled",
2305                                   "needs-l1d-flush-msr-pr-0-to-1"))
2306                         cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
2307                 if (!have_fw_feat(fw_features, "disabled",
2308                                   "needs-spec-barrier-for-bound-checks"))
2309                         cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2310                 if (have_fw_feat(fw_features, "enabled",
2311                                  "needs-count-cache-flush-on-context-switch"))
2312                         cp->behaviour |= KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2313                 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2314                         KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2315                         KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2316                         KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2317
2318                 of_node_put(fw_features);
2319         }
2320
2321         return 0;
2322 }
2323 #endif
2324
2325 long kvm_arch_vm_ioctl(struct file *filp,
2326                        unsigned int ioctl, unsigned long arg)
2327 {
2328         struct kvm *kvm __maybe_unused = filp->private_data;
2329         void __user *argp = (void __user *)arg;
2330         long r;
2331
2332         switch (ioctl) {
2333         case KVM_PPC_GET_PVINFO: {
2334                 struct kvm_ppc_pvinfo pvinfo;
2335                 memset(&pvinfo, 0, sizeof(pvinfo));
2336                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
2337                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
2338                         r = -EFAULT;
2339                         goto out;
2340                 }
2341
2342                 break;
2343         }
2344 #ifdef CONFIG_SPAPR_TCE_IOMMU
2345         case KVM_CREATE_SPAPR_TCE_64: {
2346                 struct kvm_create_spapr_tce_64 create_tce_64;
2347
2348                 r = -EFAULT;
2349                 if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
2350                         goto out;
2351                 if (create_tce_64.flags) {
2352                         r = -EINVAL;
2353                         goto out;
2354                 }
2355                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2356                 goto out;
2357         }
2358         case KVM_CREATE_SPAPR_TCE: {
2359                 struct kvm_create_spapr_tce create_tce;
2360                 struct kvm_create_spapr_tce_64 create_tce_64;
2361
2362                 r = -EFAULT;
2363                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
2364                         goto out;
2365
2366                 create_tce_64.liobn = create_tce.liobn;
2367                 create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
2368                 create_tce_64.offset = 0;
2369                 create_tce_64.size = create_tce.window_size >>
2370                                 IOMMU_PAGE_SHIFT_4K;
2371                 create_tce_64.flags = 0;
2372                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2373                 goto out;
2374         }
2375 #endif
2376 #ifdef CONFIG_PPC_BOOK3S_64
2377         case KVM_PPC_GET_SMMU_INFO: {
2378                 struct kvm_ppc_smmu_info info;
2379                 struct kvm *kvm = filp->private_data;
2380
2381                 memset(&info, 0, sizeof(info));
2382                 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
2383                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2384                         r = -EFAULT;
2385                 break;
2386         }
2387         case KVM_PPC_RTAS_DEFINE_TOKEN: {
2388                 struct kvm *kvm = filp->private_data;
2389
2390                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
2391                 break;
2392         }
2393         case KVM_PPC_CONFIGURE_V3_MMU: {
2394                 struct kvm *kvm = filp->private_data;
2395                 struct kvm_ppc_mmuv3_cfg cfg;
2396
2397                 r = -EINVAL;
2398                 if (!kvm->arch.kvm_ops->configure_mmu)
2399                         goto out;
2400                 r = -EFAULT;
2401                 if (copy_from_user(&cfg, argp, sizeof(cfg)))
2402                         goto out;
2403                 r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
2404                 break;
2405         }
2406         case KVM_PPC_GET_RMMU_INFO: {
2407                 struct kvm *kvm = filp->private_data;
2408                 struct kvm_ppc_rmmu_info info;
2409
2410                 r = -EINVAL;
2411                 if (!kvm->arch.kvm_ops->get_rmmu_info)
2412                         goto out;
2413                 r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
2414                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2415                         r = -EFAULT;
2416                 break;
2417         }
2418         case KVM_PPC_GET_CPU_CHAR: {
2419                 struct kvm_ppc_cpu_char cpuchar;
2420
2421                 r = kvmppc_get_cpu_char(&cpuchar);
2422                 if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2423                         r = -EFAULT;
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);