Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm...
[sfrench/cifs-2.6.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/tlbflush.h>
37 #include <asm/cputhreads.h>
38 #include <asm/irqflags.h>
39 #include <asm/iommu.h>
40 #include <asm/switch_to.h>
41 #include <asm/xive.h>
42
43 #include "timing.h"
44 #include "irq.h"
45 #include "../mm/mmu_decl.h"
46
47 #define CREATE_TRACE_POINTS
48 #include "trace.h"
49
50 struct kvmppc_ops *kvmppc_hv_ops;
51 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
52 struct kvmppc_ops *kvmppc_pr_ops;
53 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
54
55
56 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
57 {
58         return !!(v->arch.pending_exceptions) ||
59                v->requests;
60 }
61
62 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
63 {
64         return 1;
65 }
66
67 /*
68  * Common checks before entering the guest world.  Call with interrupts
69  * disabled.
70  *
71  * returns:
72  *
73  * == 1 if we're ready to go into guest state
74  * <= 0 if we need to go back to the host with return value
75  */
76 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
77 {
78         int r;
79
80         WARN_ON(irqs_disabled());
81         hard_irq_disable();
82
83         while (true) {
84                 if (need_resched()) {
85                         local_irq_enable();
86                         cond_resched();
87                         hard_irq_disable();
88                         continue;
89                 }
90
91                 if (signal_pending(current)) {
92                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
93                         vcpu->run->exit_reason = KVM_EXIT_INTR;
94                         r = -EINTR;
95                         break;
96                 }
97
98                 vcpu->mode = IN_GUEST_MODE;
99
100                 /*
101                  * Reading vcpu->requests must happen after setting vcpu->mode,
102                  * so we don't miss a request because the requester sees
103                  * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
104                  * before next entering the guest (and thus doesn't IPI).
105                  * This also orders the write to mode from any reads
106                  * to the page tables done while the VCPU is running.
107                  * Please see the comment in kvm_flush_remote_tlbs.
108                  */
109                 smp_mb();
110
111                 if (vcpu->requests) {
112                         /* Make sure we process requests preemptable */
113                         local_irq_enable();
114                         trace_kvm_check_requests(vcpu);
115                         r = kvmppc_core_check_requests(vcpu);
116                         hard_irq_disable();
117                         if (r > 0)
118                                 continue;
119                         break;
120                 }
121
122                 if (kvmppc_core_prepare_to_enter(vcpu)) {
123                         /* interrupts got enabled in between, so we
124                            are back at square 1 */
125                         continue;
126                 }
127
128                 guest_enter_irqoff();
129                 return 1;
130         }
131
132         /* return to host */
133         local_irq_enable();
134         return r;
135 }
136 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
137
138 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
139 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
140 {
141         struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
142         int i;
143
144         shared->sprg0 = swab64(shared->sprg0);
145         shared->sprg1 = swab64(shared->sprg1);
146         shared->sprg2 = swab64(shared->sprg2);
147         shared->sprg3 = swab64(shared->sprg3);
148         shared->srr0 = swab64(shared->srr0);
149         shared->srr1 = swab64(shared->srr1);
150         shared->dar = swab64(shared->dar);
151         shared->msr = swab64(shared->msr);
152         shared->dsisr = swab32(shared->dsisr);
153         shared->int_pending = swab32(shared->int_pending);
154         for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
155                 shared->sr[i] = swab32(shared->sr[i]);
156 }
157 #endif
158
159 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
160 {
161         int nr = kvmppc_get_gpr(vcpu, 11);
162         int r;
163         unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
164         unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
165         unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
166         unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
167         unsigned long r2 = 0;
168
169         if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
170                 /* 32 bit mode */
171                 param1 &= 0xffffffff;
172                 param2 &= 0xffffffff;
173                 param3 &= 0xffffffff;
174                 param4 &= 0xffffffff;
175         }
176
177         switch (nr) {
178         case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
179         {
180 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
181                 /* Book3S can be little endian, find it out here */
182                 int shared_big_endian = true;
183                 if (vcpu->arch.intr_msr & MSR_LE)
184                         shared_big_endian = false;
185                 if (shared_big_endian != vcpu->arch.shared_big_endian)
186                         kvmppc_swab_shared(vcpu);
187                 vcpu->arch.shared_big_endian = shared_big_endian;
188 #endif
189
190                 if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
191                         /*
192                          * Older versions of the Linux magic page code had
193                          * a bug where they would map their trampoline code
194                          * NX. If that's the case, remove !PR NX capability.
195                          */
196                         vcpu->arch.disable_kernel_nx = true;
197                         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
198                 }
199
200                 vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
201                 vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
202
203 #ifdef CONFIG_PPC_64K_PAGES
204                 /*
205                  * Make sure our 4k magic page is in the same window of a 64k
206                  * page within the guest and within the host's page.
207                  */
208                 if ((vcpu->arch.magic_page_pa & 0xf000) !=
209                     ((ulong)vcpu->arch.shared & 0xf000)) {
210                         void *old_shared = vcpu->arch.shared;
211                         ulong shared = (ulong)vcpu->arch.shared;
212                         void *new_shared;
213
214                         shared &= PAGE_MASK;
215                         shared |= vcpu->arch.magic_page_pa & 0xf000;
216                         new_shared = (void*)shared;
217                         memcpy(new_shared, old_shared, 0x1000);
218                         vcpu->arch.shared = new_shared;
219                 }
220 #endif
221
222                 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
223
224                 r = EV_SUCCESS;
225                 break;
226         }
227         case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
228                 r = EV_SUCCESS;
229 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
230                 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
231 #endif
232
233                 /* Second return value is in r4 */
234                 break;
235         case EV_HCALL_TOKEN(EV_IDLE):
236                 r = EV_SUCCESS;
237                 kvm_vcpu_block(vcpu);
238                 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
239                 break;
240         default:
241                 r = EV_UNIMPLEMENTED;
242                 break;
243         }
244
245         kvmppc_set_gpr(vcpu, 4, r2);
246
247         return r;
248 }
249 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
250
251 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
252 {
253         int r = false;
254
255         /* We have to know what CPU to virtualize */
256         if (!vcpu->arch.pvr)
257                 goto out;
258
259         /* PAPR only works with book3s_64 */
260         if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
261                 goto out;
262
263         /* HV KVM can only do PAPR mode for now */
264         if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
265                 goto out;
266
267 #ifdef CONFIG_KVM_BOOKE_HV
268         if (!cpu_has_feature(CPU_FTR_EMB_HV))
269                 goto out;
270 #endif
271
272         r = true;
273
274 out:
275         vcpu->arch.sane = r;
276         return r ? 0 : -EINVAL;
277 }
278 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
279
280 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
281 {
282         enum emulation_result er;
283         int r;
284
285         er = kvmppc_emulate_loadstore(vcpu);
286         switch (er) {
287         case EMULATE_DONE:
288                 /* Future optimization: only reload non-volatiles if they were
289                  * actually modified. */
290                 r = RESUME_GUEST_NV;
291                 break;
292         case EMULATE_AGAIN:
293                 r = RESUME_GUEST;
294                 break;
295         case EMULATE_DO_MMIO:
296                 run->exit_reason = KVM_EXIT_MMIO;
297                 /* We must reload nonvolatiles because "update" load/store
298                  * instructions modify register state. */
299                 /* Future optimization: only reload non-volatiles if they were
300                  * actually modified. */
301                 r = RESUME_HOST_NV;
302                 break;
303         case EMULATE_FAIL:
304         {
305                 u32 last_inst;
306
307                 kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
308                 /* XXX Deliver Program interrupt to guest. */
309                 pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
310                 r = RESUME_HOST;
311                 break;
312         }
313         default:
314                 WARN_ON(1);
315                 r = RESUME_GUEST;
316         }
317
318         return r;
319 }
320 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
321
322 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
323               bool data)
324 {
325         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
326         struct kvmppc_pte pte;
327         int r;
328
329         vcpu->stat.st++;
330
331         r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
332                          XLATE_WRITE, &pte);
333         if (r < 0)
334                 return r;
335
336         *eaddr = pte.raddr;
337
338         if (!pte.may_write)
339                 return -EPERM;
340
341         /* Magic page override */
342         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
343             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
344             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
345                 void *magic = vcpu->arch.shared;
346                 magic += pte.eaddr & 0xfff;
347                 memcpy(magic, ptr, size);
348                 return EMULATE_DONE;
349         }
350
351         if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
352                 return EMULATE_DO_MMIO;
353
354         return EMULATE_DONE;
355 }
356 EXPORT_SYMBOL_GPL(kvmppc_st);
357
358 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
359                       bool data)
360 {
361         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
362         struct kvmppc_pte pte;
363         int rc;
364
365         vcpu->stat.ld++;
366
367         rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
368                           XLATE_READ, &pte);
369         if (rc)
370                 return rc;
371
372         *eaddr = pte.raddr;
373
374         if (!pte.may_read)
375                 return -EPERM;
376
377         if (!data && !pte.may_execute)
378                 return -ENOEXEC;
379
380         /* Magic page override */
381         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
382             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
383             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
384                 void *magic = vcpu->arch.shared;
385                 magic += pte.eaddr & 0xfff;
386                 memcpy(ptr, magic, size);
387                 return EMULATE_DONE;
388         }
389
390         if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
391                 return EMULATE_DO_MMIO;
392
393         return EMULATE_DONE;
394 }
395 EXPORT_SYMBOL_GPL(kvmppc_ld);
396
397 int kvm_arch_hardware_enable(void)
398 {
399         return 0;
400 }
401
402 int kvm_arch_hardware_setup(void)
403 {
404         return 0;
405 }
406
407 void kvm_arch_check_processor_compat(void *rtn)
408 {
409         *(int *)rtn = kvmppc_core_check_processor_compat();
410 }
411
412 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
413 {
414         struct kvmppc_ops *kvm_ops = NULL;
415         /*
416          * if we have both HV and PR enabled, default is HV
417          */
418         if (type == 0) {
419                 if (kvmppc_hv_ops)
420                         kvm_ops = kvmppc_hv_ops;
421                 else
422                         kvm_ops = kvmppc_pr_ops;
423                 if (!kvm_ops)
424                         goto err_out;
425         } else  if (type == KVM_VM_PPC_HV) {
426                 if (!kvmppc_hv_ops)
427                         goto err_out;
428                 kvm_ops = kvmppc_hv_ops;
429         } else if (type == KVM_VM_PPC_PR) {
430                 if (!kvmppc_pr_ops)
431                         goto err_out;
432                 kvm_ops = kvmppc_pr_ops;
433         } else
434                 goto err_out;
435
436         if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
437                 return -ENOENT;
438
439         kvm->arch.kvm_ops = kvm_ops;
440         return kvmppc_core_init_vm(kvm);
441 err_out:
442         return -EINVAL;
443 }
444
445 bool kvm_arch_has_vcpu_debugfs(void)
446 {
447         return false;
448 }
449
450 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
451 {
452         return 0;
453 }
454
455 void kvm_arch_destroy_vm(struct kvm *kvm)
456 {
457         unsigned int i;
458         struct kvm_vcpu *vcpu;
459
460 #ifdef CONFIG_KVM_XICS
461         /*
462          * We call kick_all_cpus_sync() to ensure that all
463          * CPUs have executed any pending IPIs before we
464          * continue and free VCPUs structures below.
465          */
466         if (is_kvmppc_hv_enabled(kvm))
467                 kick_all_cpus_sync();
468 #endif
469
470         kvm_for_each_vcpu(i, vcpu, kvm)
471                 kvm_arch_vcpu_free(vcpu);
472
473         mutex_lock(&kvm->lock);
474         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
475                 kvm->vcpus[i] = NULL;
476
477         atomic_set(&kvm->online_vcpus, 0);
478
479         kvmppc_core_destroy_vm(kvm);
480
481         mutex_unlock(&kvm->lock);
482
483         /* drop the module reference */
484         module_put(kvm->arch.kvm_ops->owner);
485 }
486
487 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
488 {
489         int r;
490         /* Assume we're using HV mode when the HV module is loaded */
491         int hv_enabled = kvmppc_hv_ops ? 1 : 0;
492
493         if (kvm) {
494                 /*
495                  * Hooray - we know which VM type we're running on. Depend on
496                  * that rather than the guess above.
497                  */
498                 hv_enabled = is_kvmppc_hv_enabled(kvm);
499         }
500
501         switch (ext) {
502 #ifdef CONFIG_BOOKE
503         case KVM_CAP_PPC_BOOKE_SREGS:
504         case KVM_CAP_PPC_BOOKE_WATCHDOG:
505         case KVM_CAP_PPC_EPR:
506 #else
507         case KVM_CAP_PPC_SEGSTATE:
508         case KVM_CAP_PPC_HIOR:
509         case KVM_CAP_PPC_PAPR:
510 #endif
511         case KVM_CAP_PPC_UNSET_IRQ:
512         case KVM_CAP_PPC_IRQ_LEVEL:
513         case KVM_CAP_ENABLE_CAP:
514         case KVM_CAP_ENABLE_CAP_VM:
515         case KVM_CAP_ONE_REG:
516         case KVM_CAP_IOEVENTFD:
517         case KVM_CAP_DEVICE_CTRL:
518         case KVM_CAP_IMMEDIATE_EXIT:
519                 r = 1;
520                 break;
521         case KVM_CAP_PPC_PAIRED_SINGLES:
522         case KVM_CAP_PPC_OSI:
523         case KVM_CAP_PPC_GET_PVINFO:
524 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
525         case KVM_CAP_SW_TLB:
526 #endif
527                 /* We support this only for PR */
528                 r = !hv_enabled;
529                 break;
530 #ifdef CONFIG_KVM_MPIC
531         case KVM_CAP_IRQ_MPIC:
532                 r = 1;
533                 break;
534 #endif
535
536 #ifdef CONFIG_PPC_BOOK3S_64
537         case KVM_CAP_SPAPR_TCE:
538         case KVM_CAP_SPAPR_TCE_64:
539                 /* fallthrough */
540         case KVM_CAP_SPAPR_TCE_VFIO:
541         case KVM_CAP_PPC_RTAS:
542         case KVM_CAP_PPC_FIXUP_HCALL:
543         case KVM_CAP_PPC_ENABLE_HCALL:
544 #ifdef CONFIG_KVM_XICS
545         case KVM_CAP_IRQ_XICS:
546 #endif
547                 r = 1;
548                 break;
549
550         case KVM_CAP_PPC_ALLOC_HTAB:
551                 r = hv_enabled;
552                 break;
553 #endif /* CONFIG_PPC_BOOK3S_64 */
554 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
555         case KVM_CAP_PPC_SMT:
556                 r = 0;
557                 if (hv_enabled) {
558                         if (cpu_has_feature(CPU_FTR_ARCH_300))
559                                 r = 1;
560                         else
561                                 r = threads_per_subcore;
562                 }
563                 break;
564         case KVM_CAP_PPC_RMA:
565                 r = 0;
566                 break;
567         case KVM_CAP_PPC_HWRNG:
568                 r = kvmppc_hwrng_present();
569                 break;
570         case KVM_CAP_PPC_MMU_RADIX:
571                 r = !!(hv_enabled && radix_enabled());
572                 break;
573         case KVM_CAP_PPC_MMU_HASH_V3:
574                 r = !!(hv_enabled && !radix_enabled() &&
575                        cpu_has_feature(CPU_FTR_ARCH_300));
576                 break;
577 #endif
578         case KVM_CAP_SYNC_MMU:
579 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
580                 r = hv_enabled;
581 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
582                 r = 1;
583 #else
584                 r = 0;
585 #endif
586                 break;
587 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
588         case KVM_CAP_PPC_HTAB_FD:
589                 r = hv_enabled;
590                 break;
591 #endif
592         case KVM_CAP_NR_VCPUS:
593                 /*
594                  * Recommending a number of CPUs is somewhat arbitrary; we
595                  * return the number of present CPUs for -HV (since a host
596                  * will have secondary threads "offline"), and for other KVM
597                  * implementations just count online CPUs.
598                  */
599                 if (hv_enabled)
600                         r = num_present_cpus();
601                 else
602                         r = num_online_cpus();
603                 break;
604         case KVM_CAP_NR_MEMSLOTS:
605                 r = KVM_USER_MEM_SLOTS;
606                 break;
607         case KVM_CAP_MAX_VCPUS:
608                 r = KVM_MAX_VCPUS;
609                 break;
610 #ifdef CONFIG_PPC_BOOK3S_64
611         case KVM_CAP_PPC_GET_SMMU_INFO:
612                 r = 1;
613                 break;
614         case KVM_CAP_SPAPR_MULTITCE:
615                 r = 1;
616                 break;
617         case KVM_CAP_SPAPR_RESIZE_HPT:
618                 /* Disable this on POWER9 until code handles new HPTE format */
619                 r = !!hv_enabled && !cpu_has_feature(CPU_FTR_ARCH_300);
620                 break;
621 #endif
622         case KVM_CAP_PPC_HTM:
623                 r = cpu_has_feature(CPU_FTR_TM_COMP) &&
624                     is_kvmppc_hv_enabled(kvm);
625                 break;
626         default:
627                 r = 0;
628                 break;
629         }
630         return r;
631
632 }
633
634 long kvm_arch_dev_ioctl(struct file *filp,
635                         unsigned int ioctl, unsigned long arg)
636 {
637         return -EINVAL;
638 }
639
640 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
641                            struct kvm_memory_slot *dont)
642 {
643         kvmppc_core_free_memslot(kvm, free, dont);
644 }
645
646 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
647                             unsigned long npages)
648 {
649         return kvmppc_core_create_memslot(kvm, slot, npages);
650 }
651
652 int kvm_arch_prepare_memory_region(struct kvm *kvm,
653                                    struct kvm_memory_slot *memslot,
654                                    const struct kvm_userspace_memory_region *mem,
655                                    enum kvm_mr_change change)
656 {
657         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
658 }
659
660 void kvm_arch_commit_memory_region(struct kvm *kvm,
661                                    const struct kvm_userspace_memory_region *mem,
662                                    const struct kvm_memory_slot *old,
663                                    const struct kvm_memory_slot *new,
664                                    enum kvm_mr_change change)
665 {
666         kvmppc_core_commit_memory_region(kvm, mem, old, new);
667 }
668
669 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
670                                    struct kvm_memory_slot *slot)
671 {
672         kvmppc_core_flush_memslot(kvm, slot);
673 }
674
675 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
676 {
677         struct kvm_vcpu *vcpu;
678         vcpu = kvmppc_core_vcpu_create(kvm, id);
679         if (!IS_ERR(vcpu)) {
680                 vcpu->arch.wqp = &vcpu->wq;
681                 kvmppc_create_vcpu_debugfs(vcpu, id);
682         }
683         return vcpu;
684 }
685
686 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
687 {
688 }
689
690 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
691 {
692         /* Make sure we're not using the vcpu anymore */
693         hrtimer_cancel(&vcpu->arch.dec_timer);
694
695         kvmppc_remove_vcpu_debugfs(vcpu);
696
697         switch (vcpu->arch.irq_type) {
698         case KVMPPC_IRQ_MPIC:
699                 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
700                 break;
701         case KVMPPC_IRQ_XICS:
702                 if (xive_enabled())
703                         kvmppc_xive_cleanup_vcpu(vcpu);
704                 else
705                         kvmppc_xics_free_icp(vcpu);
706                 break;
707         }
708
709         kvmppc_core_vcpu_free(vcpu);
710 }
711
712 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
713 {
714         kvm_arch_vcpu_free(vcpu);
715 }
716
717 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
718 {
719         return kvmppc_core_pending_dec(vcpu);
720 }
721
722 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
723 {
724         struct kvm_vcpu *vcpu;
725
726         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
727         kvmppc_decrementer_func(vcpu);
728
729         return HRTIMER_NORESTART;
730 }
731
732 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
733 {
734         int ret;
735
736         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
737         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
738         vcpu->arch.dec_expires = ~(u64)0;
739
740 #ifdef CONFIG_KVM_EXIT_TIMING
741         mutex_init(&vcpu->arch.exit_timing_lock);
742 #endif
743         ret = kvmppc_subarch_vcpu_init(vcpu);
744         return ret;
745 }
746
747 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
748 {
749         kvmppc_mmu_destroy(vcpu);
750         kvmppc_subarch_vcpu_uninit(vcpu);
751 }
752
753 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
754 {
755 #ifdef CONFIG_BOOKE
756         /*
757          * vrsave (formerly usprg0) isn't used by Linux, but may
758          * be used by the guest.
759          *
760          * On non-booke this is associated with Altivec and
761          * is handled by code in book3s.c.
762          */
763         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
764 #endif
765         kvmppc_core_vcpu_load(vcpu, cpu);
766 }
767
768 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
769 {
770         kvmppc_core_vcpu_put(vcpu);
771 #ifdef CONFIG_BOOKE
772         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
773 #endif
774 }
775
776 /*
777  * irq_bypass_add_producer and irq_bypass_del_producer are only
778  * useful if the architecture supports PCI passthrough.
779  * irq_bypass_stop and irq_bypass_start are not needed and so
780  * kvm_ops are not defined for them.
781  */
782 bool kvm_arch_has_irq_bypass(void)
783 {
784         return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
785                 (kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
786 }
787
788 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
789                                      struct irq_bypass_producer *prod)
790 {
791         struct kvm_kernel_irqfd *irqfd =
792                 container_of(cons, struct kvm_kernel_irqfd, consumer);
793         struct kvm *kvm = irqfd->kvm;
794
795         if (kvm->arch.kvm_ops->irq_bypass_add_producer)
796                 return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
797
798         return 0;
799 }
800
801 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
802                                       struct irq_bypass_producer *prod)
803 {
804         struct kvm_kernel_irqfd *irqfd =
805                 container_of(cons, struct kvm_kernel_irqfd, consumer);
806         struct kvm *kvm = irqfd->kvm;
807
808         if (kvm->arch.kvm_ops->irq_bypass_del_producer)
809                 kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
810 }
811
812 #ifdef CONFIG_VSX
813 static inline int kvmppc_get_vsr_dword_offset(int index)
814 {
815         int offset;
816
817         if ((index != 0) && (index != 1))
818                 return -1;
819
820 #ifdef __BIG_ENDIAN
821         offset =  index;
822 #else
823         offset = 1 - index;
824 #endif
825
826         return offset;
827 }
828
829 static inline int kvmppc_get_vsr_word_offset(int index)
830 {
831         int offset;
832
833         if ((index > 3) || (index < 0))
834                 return -1;
835
836 #ifdef __BIG_ENDIAN
837         offset = index;
838 #else
839         offset = 3 - index;
840 #endif
841         return offset;
842 }
843
844 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
845         u64 gpr)
846 {
847         union kvmppc_one_reg val;
848         int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
849         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
850
851         if (offset == -1)
852                 return;
853
854         if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
855                 val.vval = VCPU_VSX_VR(vcpu, index);
856                 val.vsxval[offset] = gpr;
857                 VCPU_VSX_VR(vcpu, index) = val.vval;
858         } else {
859                 VCPU_VSX_FPR(vcpu, index, offset) = gpr;
860         }
861 }
862
863 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
864         u64 gpr)
865 {
866         union kvmppc_one_reg val;
867         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
868
869         if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
870                 val.vval = VCPU_VSX_VR(vcpu, index);
871                 val.vsxval[0] = gpr;
872                 val.vsxval[1] = gpr;
873                 VCPU_VSX_VR(vcpu, index) = val.vval;
874         } else {
875                 VCPU_VSX_FPR(vcpu, index, 0) = gpr;
876                 VCPU_VSX_FPR(vcpu, index, 1) = gpr;
877         }
878 }
879
880 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
881         u32 gpr32)
882 {
883         union kvmppc_one_reg val;
884         int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
885         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
886         int dword_offset, word_offset;
887
888         if (offset == -1)
889                 return;
890
891         if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
892                 val.vval = VCPU_VSX_VR(vcpu, index);
893                 val.vsx32val[offset] = gpr32;
894                 VCPU_VSX_VR(vcpu, index) = val.vval;
895         } else {
896                 dword_offset = offset / 2;
897                 word_offset = offset % 2;
898                 val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
899                 val.vsx32val[word_offset] = gpr32;
900                 VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
901         }
902 }
903 #endif /* CONFIG_VSX */
904
905 #ifdef CONFIG_PPC_FPU
906 static inline u64 sp_to_dp(u32 fprs)
907 {
908         u64 fprd;
909
910         preempt_disable();
911         enable_kernel_fp();
912         asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs)
913              : "fr0");
914         preempt_enable();
915         return fprd;
916 }
917
918 static inline u32 dp_to_sp(u64 fprd)
919 {
920         u32 fprs;
921
922         preempt_disable();
923         enable_kernel_fp();
924         asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd)
925              : "fr0");
926         preempt_enable();
927         return fprs;
928 }
929
930 #else
931 #define sp_to_dp(x)     (x)
932 #define dp_to_sp(x)     (x)
933 #endif /* CONFIG_PPC_FPU */
934
935 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
936                                       struct kvm_run *run)
937 {
938         u64 uninitialized_var(gpr);
939
940         if (run->mmio.len > sizeof(gpr)) {
941                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
942                 return;
943         }
944
945         if (!vcpu->arch.mmio_host_swabbed) {
946                 switch (run->mmio.len) {
947                 case 8: gpr = *(u64 *)run->mmio.data; break;
948                 case 4: gpr = *(u32 *)run->mmio.data; break;
949                 case 2: gpr = *(u16 *)run->mmio.data; break;
950                 case 1: gpr = *(u8 *)run->mmio.data; break;
951                 }
952         } else {
953                 switch (run->mmio.len) {
954                 case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
955                 case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
956                 case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
957                 case 1: gpr = *(u8 *)run->mmio.data; break;
958                 }
959         }
960
961         /* conversion between single and double precision */
962         if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
963                 gpr = sp_to_dp(gpr);
964
965         if (vcpu->arch.mmio_sign_extend) {
966                 switch (run->mmio.len) {
967 #ifdef CONFIG_PPC64
968                 case 4:
969                         gpr = (s64)(s32)gpr;
970                         break;
971 #endif
972                 case 2:
973                         gpr = (s64)(s16)gpr;
974                         break;
975                 case 1:
976                         gpr = (s64)(s8)gpr;
977                         break;
978                 }
979         }
980
981         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
982         case KVM_MMIO_REG_GPR:
983                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
984                 break;
985         case KVM_MMIO_REG_FPR:
986                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
987                 break;
988 #ifdef CONFIG_PPC_BOOK3S
989         case KVM_MMIO_REG_QPR:
990                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
991                 break;
992         case KVM_MMIO_REG_FQPR:
993                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
994                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
995                 break;
996 #endif
997 #ifdef CONFIG_VSX
998         case KVM_MMIO_REG_VSX:
999                 if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_DWORD)
1000                         kvmppc_set_vsr_dword(vcpu, gpr);
1001                 else if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_WORD)
1002                         kvmppc_set_vsr_word(vcpu, gpr);
1003                 else if (vcpu->arch.mmio_vsx_copy_type ==
1004                                 KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1005                         kvmppc_set_vsr_dword_dump(vcpu, gpr);
1006                 break;
1007 #endif
1008         default:
1009                 BUG();
1010         }
1011 }
1012
1013 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1014                                 unsigned int rt, unsigned int bytes,
1015                                 int is_default_endian, int sign_extend)
1016 {
1017         int idx, ret;
1018         bool host_swabbed;
1019
1020         /* Pity C doesn't have a logical XOR operator */
1021         if (kvmppc_need_byteswap(vcpu)) {
1022                 host_swabbed = is_default_endian;
1023         } else {
1024                 host_swabbed = !is_default_endian;
1025         }
1026
1027         if (bytes > sizeof(run->mmio.data)) {
1028                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1029                        run->mmio.len);
1030         }
1031
1032         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1033         run->mmio.len = bytes;
1034         run->mmio.is_write = 0;
1035
1036         vcpu->arch.io_gpr = rt;
1037         vcpu->arch.mmio_host_swabbed = host_swabbed;
1038         vcpu->mmio_needed = 1;
1039         vcpu->mmio_is_write = 0;
1040         vcpu->arch.mmio_sign_extend = sign_extend;
1041
1042         idx = srcu_read_lock(&vcpu->kvm->srcu);
1043
1044         ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1045                               bytes, &run->mmio.data);
1046
1047         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1048
1049         if (!ret) {
1050                 kvmppc_complete_mmio_load(vcpu, run);
1051                 vcpu->mmio_needed = 0;
1052                 return EMULATE_DONE;
1053         }
1054
1055         return EMULATE_DO_MMIO;
1056 }
1057
1058 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1059                        unsigned int rt, unsigned int bytes,
1060                        int is_default_endian)
1061 {
1062         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
1063 }
1064 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1065
1066 /* Same as above, but sign extends */
1067 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
1068                         unsigned int rt, unsigned int bytes,
1069                         int is_default_endian)
1070 {
1071         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
1072 }
1073
1074 #ifdef CONFIG_VSX
1075 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1076                         unsigned int rt, unsigned int bytes,
1077                         int is_default_endian, int mmio_sign_extend)
1078 {
1079         enum emulation_result emulated = EMULATE_DONE;
1080
1081         /* Currently, mmio_vsx_copy_nums only allowed to be less than 4 */
1082         if ( (vcpu->arch.mmio_vsx_copy_nums > 4) ||
1083                 (vcpu->arch.mmio_vsx_copy_nums < 0) ) {
1084                 return EMULATE_FAIL;
1085         }
1086
1087         while (vcpu->arch.mmio_vsx_copy_nums) {
1088                 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1089                         is_default_endian, mmio_sign_extend);
1090
1091                 if (emulated != EMULATE_DONE)
1092                         break;
1093
1094                 vcpu->arch.paddr_accessed += run->mmio.len;
1095
1096                 vcpu->arch.mmio_vsx_copy_nums--;
1097                 vcpu->arch.mmio_vsx_offset++;
1098         }
1099         return emulated;
1100 }
1101 #endif /* CONFIG_VSX */
1102
1103 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1104                         u64 val, unsigned int bytes, int is_default_endian)
1105 {
1106         void *data = run->mmio.data;
1107         int idx, ret;
1108         bool host_swabbed;
1109
1110         /* Pity C doesn't have a logical XOR operator */
1111         if (kvmppc_need_byteswap(vcpu)) {
1112                 host_swabbed = is_default_endian;
1113         } else {
1114                 host_swabbed = !is_default_endian;
1115         }
1116
1117         if (bytes > sizeof(run->mmio.data)) {
1118                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1119                        run->mmio.len);
1120         }
1121
1122         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1123         run->mmio.len = bytes;
1124         run->mmio.is_write = 1;
1125         vcpu->mmio_needed = 1;
1126         vcpu->mmio_is_write = 1;
1127
1128         if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1129                 val = dp_to_sp(val);
1130
1131         /* Store the value at the lowest bytes in 'data'. */
1132         if (!host_swabbed) {
1133                 switch (bytes) {
1134                 case 8: *(u64 *)data = val; break;
1135                 case 4: *(u32 *)data = val; break;
1136                 case 2: *(u16 *)data = val; break;
1137                 case 1: *(u8  *)data = val; break;
1138                 }
1139         } else {
1140                 switch (bytes) {
1141                 case 8: *(u64 *)data = swab64(val); break;
1142                 case 4: *(u32 *)data = swab32(val); break;
1143                 case 2: *(u16 *)data = swab16(val); break;
1144                 case 1: *(u8  *)data = val; break;
1145                 }
1146         }
1147
1148         idx = srcu_read_lock(&vcpu->kvm->srcu);
1149
1150         ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1151                                bytes, &run->mmio.data);
1152
1153         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1154
1155         if (!ret) {
1156                 vcpu->mmio_needed = 0;
1157                 return EMULATE_DONE;
1158         }
1159
1160         return EMULATE_DO_MMIO;
1161 }
1162 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1163
1164 #ifdef CONFIG_VSX
1165 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1166 {
1167         u32 dword_offset, word_offset;
1168         union kvmppc_one_reg reg;
1169         int vsx_offset = 0;
1170         int copy_type = vcpu->arch.mmio_vsx_copy_type;
1171         int result = 0;
1172
1173         switch (copy_type) {
1174         case KVMPPC_VSX_COPY_DWORD:
1175                 vsx_offset =
1176                         kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1177
1178                 if (vsx_offset == -1) {
1179                         result = -1;
1180                         break;
1181                 }
1182
1183                 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1184                         *val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1185                 } else {
1186                         reg.vval = VCPU_VSX_VR(vcpu, rs);
1187                         *val = reg.vsxval[vsx_offset];
1188                 }
1189                 break;
1190
1191         case KVMPPC_VSX_COPY_WORD:
1192                 vsx_offset =
1193                         kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1194
1195                 if (vsx_offset == -1) {
1196                         result = -1;
1197                         break;
1198                 }
1199
1200                 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1201                         dword_offset = vsx_offset / 2;
1202                         word_offset = vsx_offset % 2;
1203                         reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1204                         *val = reg.vsx32val[word_offset];
1205                 } else {
1206                         reg.vval = VCPU_VSX_VR(vcpu, rs);
1207                         *val = reg.vsx32val[vsx_offset];
1208                 }
1209                 break;
1210
1211         default:
1212                 result = -1;
1213                 break;
1214         }
1215
1216         return result;
1217 }
1218
1219 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1220                         int rs, unsigned int bytes, int is_default_endian)
1221 {
1222         u64 val;
1223         enum emulation_result emulated = EMULATE_DONE;
1224
1225         vcpu->arch.io_gpr = rs;
1226
1227         /* Currently, mmio_vsx_copy_nums only allowed to be less than 4 */
1228         if ( (vcpu->arch.mmio_vsx_copy_nums > 4) ||
1229                 (vcpu->arch.mmio_vsx_copy_nums < 0) ) {
1230                 return EMULATE_FAIL;
1231         }
1232
1233         while (vcpu->arch.mmio_vsx_copy_nums) {
1234                 if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1235                         return EMULATE_FAIL;
1236
1237                 emulated = kvmppc_handle_store(run, vcpu,
1238                          val, bytes, is_default_endian);
1239
1240                 if (emulated != EMULATE_DONE)
1241                         break;
1242
1243                 vcpu->arch.paddr_accessed += run->mmio.len;
1244
1245                 vcpu->arch.mmio_vsx_copy_nums--;
1246                 vcpu->arch.mmio_vsx_offset++;
1247         }
1248
1249         return emulated;
1250 }
1251
1252 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu,
1253                         struct kvm_run *run)
1254 {
1255         enum emulation_result emulated = EMULATE_FAIL;
1256         int r;
1257
1258         vcpu->arch.paddr_accessed += run->mmio.len;
1259
1260         if (!vcpu->mmio_is_write) {
1261                 emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr,
1262                          run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1263         } else {
1264                 emulated = kvmppc_handle_vsx_store(run, vcpu,
1265                          vcpu->arch.io_gpr, run->mmio.len, 1);
1266         }
1267
1268         switch (emulated) {
1269         case EMULATE_DO_MMIO:
1270                 run->exit_reason = KVM_EXIT_MMIO;
1271                 r = RESUME_HOST;
1272                 break;
1273         case EMULATE_FAIL:
1274                 pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1275                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1276                 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1277                 r = RESUME_HOST;
1278                 break;
1279         default:
1280                 r = RESUME_GUEST;
1281                 break;
1282         }
1283         return r;
1284 }
1285 #endif /* CONFIG_VSX */
1286
1287 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1288 {
1289         int r = 0;
1290         union kvmppc_one_reg val;
1291         int size;
1292
1293         size = one_reg_size(reg->id);
1294         if (size > sizeof(val))
1295                 return -EINVAL;
1296
1297         r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1298         if (r == -EINVAL) {
1299                 r = 0;
1300                 switch (reg->id) {
1301 #ifdef CONFIG_ALTIVEC
1302                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1303                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1304                                 r = -ENXIO;
1305                                 break;
1306                         }
1307                         val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1308                         break;
1309                 case KVM_REG_PPC_VSCR:
1310                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1311                                 r = -ENXIO;
1312                                 break;
1313                         }
1314                         val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1315                         break;
1316                 case KVM_REG_PPC_VRSAVE:
1317                         val = get_reg_val(reg->id, vcpu->arch.vrsave);
1318                         break;
1319 #endif /* CONFIG_ALTIVEC */
1320                 default:
1321                         r = -EINVAL;
1322                         break;
1323                 }
1324         }
1325
1326         if (r)
1327                 return r;
1328
1329         if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1330                 r = -EFAULT;
1331
1332         return r;
1333 }
1334
1335 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1336 {
1337         int r;
1338         union kvmppc_one_reg val;
1339         int size;
1340
1341         size = one_reg_size(reg->id);
1342         if (size > sizeof(val))
1343                 return -EINVAL;
1344
1345         if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1346                 return -EFAULT;
1347
1348         r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1349         if (r == -EINVAL) {
1350                 r = 0;
1351                 switch (reg->id) {
1352 #ifdef CONFIG_ALTIVEC
1353                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1354                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1355                                 r = -ENXIO;
1356                                 break;
1357                         }
1358                         vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1359                         break;
1360                 case KVM_REG_PPC_VSCR:
1361                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1362                                 r = -ENXIO;
1363                                 break;
1364                         }
1365                         vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1366                         break;
1367                 case KVM_REG_PPC_VRSAVE:
1368                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1369                                 r = -ENXIO;
1370                                 break;
1371                         }
1372                         vcpu->arch.vrsave = set_reg_val(reg->id, val);
1373                         break;
1374 #endif /* CONFIG_ALTIVEC */
1375                 default:
1376                         r = -EINVAL;
1377                         break;
1378                 }
1379         }
1380
1381         return r;
1382 }
1383
1384 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1385 {
1386         int r;
1387         sigset_t sigsaved;
1388
1389         if (vcpu->mmio_needed) {
1390                 vcpu->mmio_needed = 0;
1391                 if (!vcpu->mmio_is_write)
1392                         kvmppc_complete_mmio_load(vcpu, run);
1393 #ifdef CONFIG_VSX
1394                 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1395                         vcpu->arch.mmio_vsx_copy_nums--;
1396                         vcpu->arch.mmio_vsx_offset++;
1397                 }
1398
1399                 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1400                         r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run);
1401                         if (r == RESUME_HOST) {
1402                                 vcpu->mmio_needed = 1;
1403                                 return r;
1404                         }
1405                 }
1406 #endif
1407         } else if (vcpu->arch.osi_needed) {
1408                 u64 *gprs = run->osi.gprs;
1409                 int i;
1410
1411                 for (i = 0; i < 32; i++)
1412                         kvmppc_set_gpr(vcpu, i, gprs[i]);
1413                 vcpu->arch.osi_needed = 0;
1414         } else if (vcpu->arch.hcall_needed) {
1415                 int i;
1416
1417                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1418                 for (i = 0; i < 9; ++i)
1419                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1420                 vcpu->arch.hcall_needed = 0;
1421 #ifdef CONFIG_BOOKE
1422         } else if (vcpu->arch.epr_needed) {
1423                 kvmppc_set_epr(vcpu, run->epr.epr);
1424                 vcpu->arch.epr_needed = 0;
1425 #endif
1426         }
1427
1428         if (vcpu->sigset_active)
1429                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1430
1431         if (run->immediate_exit)
1432                 r = -EINTR;
1433         else
1434                 r = kvmppc_vcpu_run(run, vcpu);
1435
1436         if (vcpu->sigset_active)
1437                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1438
1439         return r;
1440 }
1441
1442 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1443 {
1444         if (irq->irq == KVM_INTERRUPT_UNSET) {
1445                 kvmppc_core_dequeue_external(vcpu);
1446                 return 0;
1447         }
1448
1449         kvmppc_core_queue_external(vcpu, irq);
1450
1451         kvm_vcpu_kick(vcpu);
1452
1453         return 0;
1454 }
1455
1456 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1457                                      struct kvm_enable_cap *cap)
1458 {
1459         int r;
1460
1461         if (cap->flags)
1462                 return -EINVAL;
1463
1464         switch (cap->cap) {
1465         case KVM_CAP_PPC_OSI:
1466                 r = 0;
1467                 vcpu->arch.osi_enabled = true;
1468                 break;
1469         case KVM_CAP_PPC_PAPR:
1470                 r = 0;
1471                 vcpu->arch.papr_enabled = true;
1472                 break;
1473         case KVM_CAP_PPC_EPR:
1474                 r = 0;
1475                 if (cap->args[0])
1476                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1477                 else
1478                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1479                 break;
1480 #ifdef CONFIG_BOOKE
1481         case KVM_CAP_PPC_BOOKE_WATCHDOG:
1482                 r = 0;
1483                 vcpu->arch.watchdog_enabled = true;
1484                 break;
1485 #endif
1486 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1487         case KVM_CAP_SW_TLB: {
1488                 struct kvm_config_tlb cfg;
1489                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1490
1491                 r = -EFAULT;
1492                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1493                         break;
1494
1495                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1496                 break;
1497         }
1498 #endif
1499 #ifdef CONFIG_KVM_MPIC
1500         case KVM_CAP_IRQ_MPIC: {
1501                 struct fd f;
1502                 struct kvm_device *dev;
1503
1504                 r = -EBADF;
1505                 f = fdget(cap->args[0]);
1506                 if (!f.file)
1507                         break;
1508
1509                 r = -EPERM;
1510                 dev = kvm_device_from_filp(f.file);
1511                 if (dev)
1512                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1513
1514                 fdput(f);
1515                 break;
1516         }
1517 #endif
1518 #ifdef CONFIG_KVM_XICS
1519         case KVM_CAP_IRQ_XICS: {
1520                 struct fd f;
1521                 struct kvm_device *dev;
1522
1523                 r = -EBADF;
1524                 f = fdget(cap->args[0]);
1525                 if (!f.file)
1526                         break;
1527
1528                 r = -EPERM;
1529                 dev = kvm_device_from_filp(f.file);
1530                 if (dev) {
1531                         if (xive_enabled())
1532                                 r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1533                         else
1534                                 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1535                 }
1536
1537                 fdput(f);
1538                 break;
1539         }
1540 #endif /* CONFIG_KVM_XICS */
1541         default:
1542                 r = -EINVAL;
1543                 break;
1544         }
1545
1546         if (!r)
1547                 r = kvmppc_sanity_check(vcpu);
1548
1549         return r;
1550 }
1551
1552 bool kvm_arch_intc_initialized(struct kvm *kvm)
1553 {
1554 #ifdef CONFIG_KVM_MPIC
1555         if (kvm->arch.mpic)
1556                 return true;
1557 #endif
1558 #ifdef CONFIG_KVM_XICS
1559         if (kvm->arch.xics || kvm->arch.xive)
1560                 return true;
1561 #endif
1562         return false;
1563 }
1564
1565 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1566                                     struct kvm_mp_state *mp_state)
1567 {
1568         return -EINVAL;
1569 }
1570
1571 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1572                                     struct kvm_mp_state *mp_state)
1573 {
1574         return -EINVAL;
1575 }
1576
1577 long kvm_arch_vcpu_ioctl(struct file *filp,
1578                          unsigned int ioctl, unsigned long arg)
1579 {
1580         struct kvm_vcpu *vcpu = filp->private_data;
1581         void __user *argp = (void __user *)arg;
1582         long r;
1583
1584         switch (ioctl) {
1585         case KVM_INTERRUPT: {
1586                 struct kvm_interrupt irq;
1587                 r = -EFAULT;
1588                 if (copy_from_user(&irq, argp, sizeof(irq)))
1589                         goto out;
1590                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1591                 goto out;
1592         }
1593
1594         case KVM_ENABLE_CAP:
1595         {
1596                 struct kvm_enable_cap cap;
1597                 r = -EFAULT;
1598                 if (copy_from_user(&cap, argp, sizeof(cap)))
1599                         goto out;
1600                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1601                 break;
1602         }
1603
1604         case KVM_SET_ONE_REG:
1605         case KVM_GET_ONE_REG:
1606         {
1607                 struct kvm_one_reg reg;
1608                 r = -EFAULT;
1609                 if (copy_from_user(&reg, argp, sizeof(reg)))
1610                         goto out;
1611                 if (ioctl == KVM_SET_ONE_REG)
1612                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
1613                 else
1614                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
1615                 break;
1616         }
1617
1618 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1619         case KVM_DIRTY_TLB: {
1620                 struct kvm_dirty_tlb dirty;
1621                 r = -EFAULT;
1622                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
1623                         goto out;
1624                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1625                 break;
1626         }
1627 #endif
1628         default:
1629                 r = -EINVAL;
1630         }
1631
1632 out:
1633         return r;
1634 }
1635
1636 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1637 {
1638         return VM_FAULT_SIGBUS;
1639 }
1640
1641 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1642 {
1643         u32 inst_nop = 0x60000000;
1644 #ifdef CONFIG_KVM_BOOKE_HV
1645         u32 inst_sc1 = 0x44000022;
1646         pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1647         pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1648         pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1649         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1650 #else
1651         u32 inst_lis = 0x3c000000;
1652         u32 inst_ori = 0x60000000;
1653         u32 inst_sc = 0x44000002;
1654         u32 inst_imm_mask = 0xffff;
1655
1656         /*
1657          * The hypercall to get into KVM from within guest context is as
1658          * follows:
1659          *
1660          *    lis r0, r0, KVM_SC_MAGIC_R0@h
1661          *    ori r0, KVM_SC_MAGIC_R0@l
1662          *    sc
1663          *    nop
1664          */
1665         pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1666         pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1667         pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1668         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1669 #endif
1670
1671         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1672
1673         return 0;
1674 }
1675
1676 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1677                           bool line_status)
1678 {
1679         if (!irqchip_in_kernel(kvm))
1680                 return -ENXIO;
1681
1682         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1683                                         irq_event->irq, irq_event->level,
1684                                         line_status);
1685         return 0;
1686 }
1687
1688
1689 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1690                                    struct kvm_enable_cap *cap)
1691 {
1692         int r;
1693
1694         if (cap->flags)
1695                 return -EINVAL;
1696
1697         switch (cap->cap) {
1698 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1699         case KVM_CAP_PPC_ENABLE_HCALL: {
1700                 unsigned long hcall = cap->args[0];
1701
1702                 r = -EINVAL;
1703                 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1704                     cap->args[1] > 1)
1705                         break;
1706                 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1707                         break;
1708                 if (cap->args[1])
1709                         set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1710                 else
1711                         clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1712                 r = 0;
1713                 break;
1714         }
1715 #endif
1716         default:
1717                 r = -EINVAL;
1718                 break;
1719         }
1720
1721         return r;
1722 }
1723
1724 long kvm_arch_vm_ioctl(struct file *filp,
1725                        unsigned int ioctl, unsigned long arg)
1726 {
1727         struct kvm *kvm __maybe_unused = filp->private_data;
1728         void __user *argp = (void __user *)arg;
1729         long r;
1730
1731         switch (ioctl) {
1732         case KVM_PPC_GET_PVINFO: {
1733                 struct kvm_ppc_pvinfo pvinfo;
1734                 memset(&pvinfo, 0, sizeof(pvinfo));
1735                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1736                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1737                         r = -EFAULT;
1738                         goto out;
1739                 }
1740
1741                 break;
1742         }
1743         case KVM_ENABLE_CAP:
1744         {
1745                 struct kvm_enable_cap cap;
1746                 r = -EFAULT;
1747                 if (copy_from_user(&cap, argp, sizeof(cap)))
1748                         goto out;
1749                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1750                 break;
1751         }
1752 #ifdef CONFIG_SPAPR_TCE_IOMMU
1753         case KVM_CREATE_SPAPR_TCE_64: {
1754                 struct kvm_create_spapr_tce_64 create_tce_64;
1755
1756                 r = -EFAULT;
1757                 if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
1758                         goto out;
1759                 if (create_tce_64.flags) {
1760                         r = -EINVAL;
1761                         goto out;
1762                 }
1763                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
1764                 goto out;
1765         }
1766         case KVM_CREATE_SPAPR_TCE: {
1767                 struct kvm_create_spapr_tce create_tce;
1768                 struct kvm_create_spapr_tce_64 create_tce_64;
1769
1770                 r = -EFAULT;
1771                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1772                         goto out;
1773
1774                 create_tce_64.liobn = create_tce.liobn;
1775                 create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
1776                 create_tce_64.offset = 0;
1777                 create_tce_64.size = create_tce.window_size >>
1778                                 IOMMU_PAGE_SHIFT_4K;
1779                 create_tce_64.flags = 0;
1780                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
1781                 goto out;
1782         }
1783 #endif
1784 #ifdef CONFIG_PPC_BOOK3S_64
1785         case KVM_PPC_GET_SMMU_INFO: {
1786                 struct kvm_ppc_smmu_info info;
1787                 struct kvm *kvm = filp->private_data;
1788
1789                 memset(&info, 0, sizeof(info));
1790                 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1791                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1792                         r = -EFAULT;
1793                 break;
1794         }
1795         case KVM_PPC_RTAS_DEFINE_TOKEN: {
1796                 struct kvm *kvm = filp->private_data;
1797
1798                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1799                 break;
1800         }
1801         case KVM_PPC_CONFIGURE_V3_MMU: {
1802                 struct kvm *kvm = filp->private_data;
1803                 struct kvm_ppc_mmuv3_cfg cfg;
1804
1805                 r = -EINVAL;
1806                 if (!kvm->arch.kvm_ops->configure_mmu)
1807                         goto out;
1808                 r = -EFAULT;
1809                 if (copy_from_user(&cfg, argp, sizeof(cfg)))
1810                         goto out;
1811                 r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
1812                 break;
1813         }
1814         case KVM_PPC_GET_RMMU_INFO: {
1815                 struct kvm *kvm = filp->private_data;
1816                 struct kvm_ppc_rmmu_info info;
1817
1818                 r = -EINVAL;
1819                 if (!kvm->arch.kvm_ops->get_rmmu_info)
1820                         goto out;
1821                 r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
1822                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1823                         r = -EFAULT;
1824                 break;
1825         }
1826         default: {
1827                 struct kvm *kvm = filp->private_data;
1828                 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1829         }
1830 #else /* CONFIG_PPC_BOOK3S_64 */
1831         default:
1832                 r = -ENOTTY;
1833 #endif
1834         }
1835 out:
1836         return r;
1837 }
1838
1839 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1840 static unsigned long nr_lpids;
1841
1842 long kvmppc_alloc_lpid(void)
1843 {
1844         long lpid;
1845
1846         do {
1847                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1848                 if (lpid >= nr_lpids) {
1849                         pr_err("%s: No LPIDs free\n", __func__);
1850                         return -ENOMEM;
1851                 }
1852         } while (test_and_set_bit(lpid, lpid_inuse));
1853
1854         return lpid;
1855 }
1856 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1857
1858 void kvmppc_claim_lpid(long lpid)
1859 {
1860         set_bit(lpid, lpid_inuse);
1861 }
1862 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1863
1864 void kvmppc_free_lpid(long lpid)
1865 {
1866         clear_bit(lpid, lpid_inuse);
1867 }
1868 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1869
1870 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1871 {
1872         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1873         memset(lpid_inuse, 0, sizeof(lpid_inuse));
1874 }
1875 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1876
1877 int kvm_arch_init(void *opaque)
1878 {
1879         return 0;
1880 }
1881
1882 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);