KVM: PPC: Book3S: Provide different CAPs based on HV or PR mode
[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/fs.h>
27 #include <linux/slab.h>
28 #include <linux/file.h>
29 #include <linux/module.h>
30 #include <asm/cputable.h>
31 #include <asm/uaccess.h>
32 #include <asm/kvm_ppc.h>
33 #include <asm/tlbflush.h>
34 #include <asm/cputhreads.h>
35 #include <asm/irqflags.h>
36 #include "timing.h"
37 #include "irq.h"
38 #include "../mm/mmu_decl.h"
39
40 #define CREATE_TRACE_POINTS
41 #include "trace.h"
42
43 struct kvmppc_ops *kvmppc_hv_ops;
44 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
45 struct kvmppc_ops *kvmppc_pr_ops;
46 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
47
48
49 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
50 {
51         return !!(v->arch.pending_exceptions) ||
52                v->requests;
53 }
54
55 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
56 {
57         return 1;
58 }
59
60 /*
61  * Common checks before entering the guest world.  Call with interrupts
62  * disabled.
63  *
64  * returns:
65  *
66  * == 1 if we're ready to go into guest state
67  * <= 0 if we need to go back to the host with return value
68  */
69 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
70 {
71         int r;
72
73         WARN_ON(irqs_disabled());
74         hard_irq_disable();
75
76         while (true) {
77                 if (need_resched()) {
78                         local_irq_enable();
79                         cond_resched();
80                         hard_irq_disable();
81                         continue;
82                 }
83
84                 if (signal_pending(current)) {
85                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
86                         vcpu->run->exit_reason = KVM_EXIT_INTR;
87                         r = -EINTR;
88                         break;
89                 }
90
91                 vcpu->mode = IN_GUEST_MODE;
92
93                 /*
94                  * Reading vcpu->requests must happen after setting vcpu->mode,
95                  * so we don't miss a request because the requester sees
96                  * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
97                  * before next entering the guest (and thus doesn't IPI).
98                  */
99                 smp_mb();
100
101                 if (vcpu->requests) {
102                         /* Make sure we process requests preemptable */
103                         local_irq_enable();
104                         trace_kvm_check_requests(vcpu);
105                         r = kvmppc_core_check_requests(vcpu);
106                         hard_irq_disable();
107                         if (r > 0)
108                                 continue;
109                         break;
110                 }
111
112                 if (kvmppc_core_prepare_to_enter(vcpu)) {
113                         /* interrupts got enabled in between, so we
114                            are back at square 1 */
115                         continue;
116                 }
117
118                 kvm_guest_enter();
119                 return 1;
120         }
121
122         /* return to host */
123         local_irq_enable();
124         return r;
125 }
126 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
127
128 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
129 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
130 {
131         struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
132         int i;
133
134         shared->sprg0 = swab64(shared->sprg0);
135         shared->sprg1 = swab64(shared->sprg1);
136         shared->sprg2 = swab64(shared->sprg2);
137         shared->sprg3 = swab64(shared->sprg3);
138         shared->srr0 = swab64(shared->srr0);
139         shared->srr1 = swab64(shared->srr1);
140         shared->dar = swab64(shared->dar);
141         shared->msr = swab64(shared->msr);
142         shared->dsisr = swab32(shared->dsisr);
143         shared->int_pending = swab32(shared->int_pending);
144         for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
145                 shared->sr[i] = swab32(shared->sr[i]);
146 }
147 #endif
148
149 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
150 {
151         int nr = kvmppc_get_gpr(vcpu, 11);
152         int r;
153         unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
154         unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
155         unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
156         unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
157         unsigned long r2 = 0;
158
159         if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
160                 /* 32 bit mode */
161                 param1 &= 0xffffffff;
162                 param2 &= 0xffffffff;
163                 param3 &= 0xffffffff;
164                 param4 &= 0xffffffff;
165         }
166
167         switch (nr) {
168         case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
169         {
170 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
171                 /* Book3S can be little endian, find it out here */
172                 int shared_big_endian = true;
173                 if (vcpu->arch.intr_msr & MSR_LE)
174                         shared_big_endian = false;
175                 if (shared_big_endian != vcpu->arch.shared_big_endian)
176                         kvmppc_swab_shared(vcpu);
177                 vcpu->arch.shared_big_endian = shared_big_endian;
178 #endif
179
180                 if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
181                         /*
182                          * Older versions of the Linux magic page code had
183                          * a bug where they would map their trampoline code
184                          * NX. If that's the case, remove !PR NX capability.
185                          */
186                         vcpu->arch.disable_kernel_nx = true;
187                         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
188                 }
189
190                 vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
191                 vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
192
193 #ifdef CONFIG_PPC_64K_PAGES
194                 /*
195                  * Make sure our 4k magic page is in the same window of a 64k
196                  * page within the guest and within the host's page.
197                  */
198                 if ((vcpu->arch.magic_page_pa & 0xf000) !=
199                     ((ulong)vcpu->arch.shared & 0xf000)) {
200                         void *old_shared = vcpu->arch.shared;
201                         ulong shared = (ulong)vcpu->arch.shared;
202                         void *new_shared;
203
204                         shared &= PAGE_MASK;
205                         shared |= vcpu->arch.magic_page_pa & 0xf000;
206                         new_shared = (void*)shared;
207                         memcpy(new_shared, old_shared, 0x1000);
208                         vcpu->arch.shared = new_shared;
209                 }
210 #endif
211
212                 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
213
214                 r = EV_SUCCESS;
215                 break;
216         }
217         case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
218                 r = EV_SUCCESS;
219 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
220                 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
221 #endif
222
223                 /* Second return value is in r4 */
224                 break;
225         case EV_HCALL_TOKEN(EV_IDLE):
226                 r = EV_SUCCESS;
227                 kvm_vcpu_block(vcpu);
228                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
229                 break;
230         default:
231                 r = EV_UNIMPLEMENTED;
232                 break;
233         }
234
235         kvmppc_set_gpr(vcpu, 4, r2);
236
237         return r;
238 }
239 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
240
241 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
242 {
243         int r = false;
244
245         /* We have to know what CPU to virtualize */
246         if (!vcpu->arch.pvr)
247                 goto out;
248
249         /* PAPR only works with book3s_64 */
250         if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
251                 goto out;
252
253         /* HV KVM can only do PAPR mode for now */
254         if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
255                 goto out;
256
257 #ifdef CONFIG_KVM_BOOKE_HV
258         if (!cpu_has_feature(CPU_FTR_EMB_HV))
259                 goto out;
260 #endif
261
262         r = true;
263
264 out:
265         vcpu->arch.sane = r;
266         return r ? 0 : -EINVAL;
267 }
268 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
269
270 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
271 {
272         enum emulation_result er;
273         int r;
274
275         er = kvmppc_emulate_instruction(run, vcpu);
276         switch (er) {
277         case EMULATE_DONE:
278                 /* Future optimization: only reload non-volatiles if they were
279                  * actually modified. */
280                 r = RESUME_GUEST_NV;
281                 break;
282         case EMULATE_AGAIN:
283                 r = RESUME_GUEST;
284                 break;
285         case EMULATE_DO_MMIO:
286                 run->exit_reason = KVM_EXIT_MMIO;
287                 /* We must reload nonvolatiles because "update" load/store
288                  * instructions modify register state. */
289                 /* Future optimization: only reload non-volatiles if they were
290                  * actually modified. */
291                 r = RESUME_HOST_NV;
292                 break;
293         case EMULATE_FAIL:
294         {
295                 u32 last_inst;
296
297                 kvmppc_get_last_inst(vcpu, false, &last_inst);
298                 /* XXX Deliver Program interrupt to guest. */
299                 pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
300                 r = RESUME_HOST;
301                 break;
302         }
303         default:
304                 WARN_ON(1);
305                 r = RESUME_GUEST;
306         }
307
308         return r;
309 }
310 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
311
312 int kvm_arch_hardware_enable(void *garbage)
313 {
314         return 0;
315 }
316
317 void kvm_arch_hardware_disable(void *garbage)
318 {
319 }
320
321 int kvm_arch_hardware_setup(void)
322 {
323         return 0;
324 }
325
326 void kvm_arch_hardware_unsetup(void)
327 {
328 }
329
330 void kvm_arch_check_processor_compat(void *rtn)
331 {
332         *(int *)rtn = kvmppc_core_check_processor_compat();
333 }
334
335 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
336 {
337         struct kvmppc_ops *kvm_ops = NULL;
338         /*
339          * if we have both HV and PR enabled, default is HV
340          */
341         if (type == 0) {
342                 if (kvmppc_hv_ops)
343                         kvm_ops = kvmppc_hv_ops;
344                 else
345                         kvm_ops = kvmppc_pr_ops;
346                 if (!kvm_ops)
347                         goto err_out;
348         } else  if (type == KVM_VM_PPC_HV) {
349                 if (!kvmppc_hv_ops)
350                         goto err_out;
351                 kvm_ops = kvmppc_hv_ops;
352         } else if (type == KVM_VM_PPC_PR) {
353                 if (!kvmppc_pr_ops)
354                         goto err_out;
355                 kvm_ops = kvmppc_pr_ops;
356         } else
357                 goto err_out;
358
359         if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
360                 return -ENOENT;
361
362         kvm->arch.kvm_ops = kvm_ops;
363         return kvmppc_core_init_vm(kvm);
364 err_out:
365         return -EINVAL;
366 }
367
368 void kvm_arch_destroy_vm(struct kvm *kvm)
369 {
370         unsigned int i;
371         struct kvm_vcpu *vcpu;
372
373         kvm_for_each_vcpu(i, vcpu, kvm)
374                 kvm_arch_vcpu_free(vcpu);
375
376         mutex_lock(&kvm->lock);
377         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
378                 kvm->vcpus[i] = NULL;
379
380         atomic_set(&kvm->online_vcpus, 0);
381
382         kvmppc_core_destroy_vm(kvm);
383
384         mutex_unlock(&kvm->lock);
385
386         /* drop the module reference */
387         module_put(kvm->arch.kvm_ops->owner);
388 }
389
390 void kvm_arch_sync_events(struct kvm *kvm)
391 {
392 }
393
394 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
395 {
396         int r;
397         /* Assume we're using HV mode when the HV module is loaded */
398         int hv_enabled = kvmppc_hv_ops ? 1 : 0;
399
400         if (kvm) {
401                 /*
402                  * Hooray - we know which VM type we're running on. Depend on
403                  * that rather than the guess above.
404                  */
405                 hv_enabled = is_kvmppc_hv_enabled(kvm);
406         }
407
408         switch (ext) {
409 #ifdef CONFIG_BOOKE
410         case KVM_CAP_PPC_BOOKE_SREGS:
411         case KVM_CAP_PPC_BOOKE_WATCHDOG:
412         case KVM_CAP_PPC_EPR:
413 #else
414         case KVM_CAP_PPC_SEGSTATE:
415         case KVM_CAP_PPC_HIOR:
416         case KVM_CAP_PPC_PAPR:
417 #endif
418         case KVM_CAP_PPC_UNSET_IRQ:
419         case KVM_CAP_PPC_IRQ_LEVEL:
420         case KVM_CAP_ENABLE_CAP:
421         case KVM_CAP_ENABLE_CAP_VM:
422         case KVM_CAP_ONE_REG:
423         case KVM_CAP_IOEVENTFD:
424         case KVM_CAP_DEVICE_CTRL:
425                 r = 1;
426                 break;
427         case KVM_CAP_PPC_PAIRED_SINGLES:
428         case KVM_CAP_PPC_OSI:
429         case KVM_CAP_PPC_GET_PVINFO:
430 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
431         case KVM_CAP_SW_TLB:
432 #endif
433                 /* We support this only for PR */
434                 r = !hv_enabled;
435                 break;
436 #ifdef CONFIG_KVM_MMIO
437         case KVM_CAP_COALESCED_MMIO:
438                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
439                 break;
440 #endif
441 #ifdef CONFIG_KVM_MPIC
442         case KVM_CAP_IRQ_MPIC:
443                 r = 1;
444                 break;
445 #endif
446
447 #ifdef CONFIG_PPC_BOOK3S_64
448         case KVM_CAP_SPAPR_TCE:
449         case KVM_CAP_PPC_ALLOC_HTAB:
450         case KVM_CAP_PPC_RTAS:
451         case KVM_CAP_PPC_FIXUP_HCALL:
452         case KVM_CAP_PPC_ENABLE_HCALL:
453 #ifdef CONFIG_KVM_XICS
454         case KVM_CAP_IRQ_XICS:
455 #endif
456                 r = 1;
457                 break;
458 #endif /* CONFIG_PPC_BOOK3S_64 */
459 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
460         case KVM_CAP_PPC_SMT:
461                 if (hv_enabled)
462                         r = threads_per_subcore;
463                 else
464                         r = 0;
465                 break;
466         case KVM_CAP_PPC_RMA:
467                 r = hv_enabled;
468                 /* PPC970 requires an RMA */
469                 if (r && cpu_has_feature(CPU_FTR_ARCH_201))
470                         r = 2;
471                 break;
472 #endif
473         case KVM_CAP_SYNC_MMU:
474 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
475                 if (hv_enabled)
476                         r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
477                 else
478                         r = 0;
479 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
480                 r = 1;
481 #else
482                 r = 0;
483 #endif
484                 break;
485 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
486         case KVM_CAP_PPC_HTAB_FD:
487                 r = hv_enabled;
488                 break;
489 #endif
490         case KVM_CAP_NR_VCPUS:
491                 /*
492                  * Recommending a number of CPUs is somewhat arbitrary; we
493                  * return the number of present CPUs for -HV (since a host
494                  * will have secondary threads "offline"), and for other KVM
495                  * implementations just count online CPUs.
496                  */
497                 if (hv_enabled)
498                         r = num_present_cpus();
499                 else
500                         r = num_online_cpus();
501                 break;
502         case KVM_CAP_MAX_VCPUS:
503                 r = KVM_MAX_VCPUS;
504                 break;
505 #ifdef CONFIG_PPC_BOOK3S_64
506         case KVM_CAP_PPC_GET_SMMU_INFO:
507                 r = 1;
508                 break;
509 #endif
510         default:
511                 r = 0;
512                 break;
513         }
514         return r;
515
516 }
517
518 long kvm_arch_dev_ioctl(struct file *filp,
519                         unsigned int ioctl, unsigned long arg)
520 {
521         return -EINVAL;
522 }
523
524 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
525                            struct kvm_memory_slot *dont)
526 {
527         kvmppc_core_free_memslot(kvm, free, dont);
528 }
529
530 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
531                             unsigned long npages)
532 {
533         return kvmppc_core_create_memslot(kvm, slot, npages);
534 }
535
536 void kvm_arch_memslots_updated(struct kvm *kvm)
537 {
538 }
539
540 int kvm_arch_prepare_memory_region(struct kvm *kvm,
541                                    struct kvm_memory_slot *memslot,
542                                    struct kvm_userspace_memory_region *mem,
543                                    enum kvm_mr_change change)
544 {
545         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
546 }
547
548 void kvm_arch_commit_memory_region(struct kvm *kvm,
549                                    struct kvm_userspace_memory_region *mem,
550                                    const struct kvm_memory_slot *old,
551                                    enum kvm_mr_change change)
552 {
553         kvmppc_core_commit_memory_region(kvm, mem, old);
554 }
555
556 void kvm_arch_flush_shadow_all(struct kvm *kvm)
557 {
558 }
559
560 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
561                                    struct kvm_memory_slot *slot)
562 {
563         kvmppc_core_flush_memslot(kvm, slot);
564 }
565
566 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
567 {
568         struct kvm_vcpu *vcpu;
569         vcpu = kvmppc_core_vcpu_create(kvm, id);
570         if (!IS_ERR(vcpu)) {
571                 vcpu->arch.wqp = &vcpu->wq;
572                 kvmppc_create_vcpu_debugfs(vcpu, id);
573         }
574         return vcpu;
575 }
576
577 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
578 {
579         return 0;
580 }
581
582 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
583 {
584         /* Make sure we're not using the vcpu anymore */
585         hrtimer_cancel(&vcpu->arch.dec_timer);
586         tasklet_kill(&vcpu->arch.tasklet);
587
588         kvmppc_remove_vcpu_debugfs(vcpu);
589
590         switch (vcpu->arch.irq_type) {
591         case KVMPPC_IRQ_MPIC:
592                 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
593                 break;
594         case KVMPPC_IRQ_XICS:
595                 kvmppc_xics_free_icp(vcpu);
596                 break;
597         }
598
599         kvmppc_core_vcpu_free(vcpu);
600 }
601
602 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
603 {
604         kvm_arch_vcpu_free(vcpu);
605 }
606
607 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
608 {
609         return kvmppc_core_pending_dec(vcpu);
610 }
611
612 /*
613  * low level hrtimer wake routine. Because this runs in hardirq context
614  * we schedule a tasklet to do the real work.
615  */
616 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
617 {
618         struct kvm_vcpu *vcpu;
619
620         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
621         tasklet_schedule(&vcpu->arch.tasklet);
622
623         return HRTIMER_NORESTART;
624 }
625
626 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
627 {
628         int ret;
629
630         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
631         tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
632         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
633         vcpu->arch.dec_expires = ~(u64)0;
634
635 #ifdef CONFIG_KVM_EXIT_TIMING
636         mutex_init(&vcpu->arch.exit_timing_lock);
637 #endif
638         ret = kvmppc_subarch_vcpu_init(vcpu);
639         return ret;
640 }
641
642 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
643 {
644         kvmppc_mmu_destroy(vcpu);
645         kvmppc_subarch_vcpu_uninit(vcpu);
646 }
647
648 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
649 {
650 #ifdef CONFIG_BOOKE
651         /*
652          * vrsave (formerly usprg0) isn't used by Linux, but may
653          * be used by the guest.
654          *
655          * On non-booke this is associated with Altivec and
656          * is handled by code in book3s.c.
657          */
658         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
659 #endif
660         kvmppc_core_vcpu_load(vcpu, cpu);
661 }
662
663 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
664 {
665         kvmppc_core_vcpu_put(vcpu);
666 #ifdef CONFIG_BOOKE
667         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
668 #endif
669 }
670
671 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
672                                      struct kvm_run *run)
673 {
674         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
675 }
676
677 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
678                                       struct kvm_run *run)
679 {
680         u64 uninitialized_var(gpr);
681
682         if (run->mmio.len > sizeof(gpr)) {
683                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
684                 return;
685         }
686
687         if (vcpu->arch.mmio_is_bigendian) {
688                 switch (run->mmio.len) {
689                 case 8: gpr = *(u64 *)run->mmio.data; break;
690                 case 4: gpr = *(u32 *)run->mmio.data; break;
691                 case 2: gpr = *(u16 *)run->mmio.data; break;
692                 case 1: gpr = *(u8 *)run->mmio.data; break;
693                 }
694         } else {
695                 /* Convert BE data from userland back to LE. */
696                 switch (run->mmio.len) {
697                 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
698                 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
699                 case 1: gpr = *(u8 *)run->mmio.data; break;
700                 }
701         }
702
703         if (vcpu->arch.mmio_sign_extend) {
704                 switch (run->mmio.len) {
705 #ifdef CONFIG_PPC64
706                 case 4:
707                         gpr = (s64)(s32)gpr;
708                         break;
709 #endif
710                 case 2:
711                         gpr = (s64)(s16)gpr;
712                         break;
713                 case 1:
714                         gpr = (s64)(s8)gpr;
715                         break;
716                 }
717         }
718
719         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
720
721         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
722         case KVM_MMIO_REG_GPR:
723                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
724                 break;
725         case KVM_MMIO_REG_FPR:
726                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
727                 break;
728 #ifdef CONFIG_PPC_BOOK3S
729         case KVM_MMIO_REG_QPR:
730                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
731                 break;
732         case KVM_MMIO_REG_FQPR:
733                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
734                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
735                 break;
736 #endif
737         default:
738                 BUG();
739         }
740 }
741
742 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
743                        unsigned int rt, unsigned int bytes,
744                        int is_default_endian)
745 {
746         int idx, ret;
747         int is_bigendian;
748
749         if (kvmppc_need_byteswap(vcpu)) {
750                 /* Default endianness is "little endian". */
751                 is_bigendian = !is_default_endian;
752         } else {
753                 /* Default endianness is "big endian". */
754                 is_bigendian = is_default_endian;
755         }
756
757         if (bytes > sizeof(run->mmio.data)) {
758                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
759                        run->mmio.len);
760         }
761
762         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
763         run->mmio.len = bytes;
764         run->mmio.is_write = 0;
765
766         vcpu->arch.io_gpr = rt;
767         vcpu->arch.mmio_is_bigendian = is_bigendian;
768         vcpu->mmio_needed = 1;
769         vcpu->mmio_is_write = 0;
770         vcpu->arch.mmio_sign_extend = 0;
771
772         idx = srcu_read_lock(&vcpu->kvm->srcu);
773
774         ret = kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
775                               bytes, &run->mmio.data);
776
777         srcu_read_unlock(&vcpu->kvm->srcu, idx);
778
779         if (!ret) {
780                 kvmppc_complete_mmio_load(vcpu, run);
781                 vcpu->mmio_needed = 0;
782                 return EMULATE_DONE;
783         }
784
785         return EMULATE_DO_MMIO;
786 }
787 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
788
789 /* Same as above, but sign extends */
790 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
791                         unsigned int rt, unsigned int bytes,
792                         int is_default_endian)
793 {
794         int r;
795
796         vcpu->arch.mmio_sign_extend = 1;
797         r = kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian);
798
799         return r;
800 }
801
802 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
803                         u64 val, unsigned int bytes, int is_default_endian)
804 {
805         void *data = run->mmio.data;
806         int idx, ret;
807         int is_bigendian;
808
809         if (kvmppc_need_byteswap(vcpu)) {
810                 /* Default endianness is "little endian". */
811                 is_bigendian = !is_default_endian;
812         } else {
813                 /* Default endianness is "big endian". */
814                 is_bigendian = is_default_endian;
815         }
816
817         if (bytes > sizeof(run->mmio.data)) {
818                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
819                        run->mmio.len);
820         }
821
822         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
823         run->mmio.len = bytes;
824         run->mmio.is_write = 1;
825         vcpu->mmio_needed = 1;
826         vcpu->mmio_is_write = 1;
827
828         /* Store the value at the lowest bytes in 'data'. */
829         if (is_bigendian) {
830                 switch (bytes) {
831                 case 8: *(u64 *)data = val; break;
832                 case 4: *(u32 *)data = val; break;
833                 case 2: *(u16 *)data = val; break;
834                 case 1: *(u8  *)data = val; break;
835                 }
836         } else {
837                 /* Store LE value into 'data'. */
838                 switch (bytes) {
839                 case 4: st_le32(data, val); break;
840                 case 2: st_le16(data, val); break;
841                 case 1: *(u8 *)data = val; break;
842                 }
843         }
844
845         idx = srcu_read_lock(&vcpu->kvm->srcu);
846
847         ret = kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
848                                bytes, &run->mmio.data);
849
850         srcu_read_unlock(&vcpu->kvm->srcu, idx);
851
852         if (!ret) {
853                 vcpu->mmio_needed = 0;
854                 return EMULATE_DONE;
855         }
856
857         return EMULATE_DO_MMIO;
858 }
859 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
860
861 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
862 {
863         int r;
864         sigset_t sigsaved;
865
866         if (vcpu->sigset_active)
867                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
868
869         if (vcpu->mmio_needed) {
870                 if (!vcpu->mmio_is_write)
871                         kvmppc_complete_mmio_load(vcpu, run);
872                 vcpu->mmio_needed = 0;
873         } else if (vcpu->arch.dcr_needed) {
874                 if (!vcpu->arch.dcr_is_write)
875                         kvmppc_complete_dcr_load(vcpu, run);
876                 vcpu->arch.dcr_needed = 0;
877         } else if (vcpu->arch.osi_needed) {
878                 u64 *gprs = run->osi.gprs;
879                 int i;
880
881                 for (i = 0; i < 32; i++)
882                         kvmppc_set_gpr(vcpu, i, gprs[i]);
883                 vcpu->arch.osi_needed = 0;
884         } else if (vcpu->arch.hcall_needed) {
885                 int i;
886
887                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
888                 for (i = 0; i < 9; ++i)
889                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
890                 vcpu->arch.hcall_needed = 0;
891 #ifdef CONFIG_BOOKE
892         } else if (vcpu->arch.epr_needed) {
893                 kvmppc_set_epr(vcpu, run->epr.epr);
894                 vcpu->arch.epr_needed = 0;
895 #endif
896         }
897
898         r = kvmppc_vcpu_run(run, vcpu);
899
900         if (vcpu->sigset_active)
901                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
902
903         return r;
904 }
905
906 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
907 {
908         if (irq->irq == KVM_INTERRUPT_UNSET) {
909                 kvmppc_core_dequeue_external(vcpu);
910                 return 0;
911         }
912
913         kvmppc_core_queue_external(vcpu, irq);
914
915         kvm_vcpu_kick(vcpu);
916
917         return 0;
918 }
919
920 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
921                                      struct kvm_enable_cap *cap)
922 {
923         int r;
924
925         if (cap->flags)
926                 return -EINVAL;
927
928         switch (cap->cap) {
929         case KVM_CAP_PPC_OSI:
930                 r = 0;
931                 vcpu->arch.osi_enabled = true;
932                 break;
933         case KVM_CAP_PPC_PAPR:
934                 r = 0;
935                 vcpu->arch.papr_enabled = true;
936                 break;
937         case KVM_CAP_PPC_EPR:
938                 r = 0;
939                 if (cap->args[0])
940                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
941                 else
942                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
943                 break;
944 #ifdef CONFIG_BOOKE
945         case KVM_CAP_PPC_BOOKE_WATCHDOG:
946                 r = 0;
947                 vcpu->arch.watchdog_enabled = true;
948                 break;
949 #endif
950 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
951         case KVM_CAP_SW_TLB: {
952                 struct kvm_config_tlb cfg;
953                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
954
955                 r = -EFAULT;
956                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
957                         break;
958
959                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
960                 break;
961         }
962 #endif
963 #ifdef CONFIG_KVM_MPIC
964         case KVM_CAP_IRQ_MPIC: {
965                 struct fd f;
966                 struct kvm_device *dev;
967
968                 r = -EBADF;
969                 f = fdget(cap->args[0]);
970                 if (!f.file)
971                         break;
972
973                 r = -EPERM;
974                 dev = kvm_device_from_filp(f.file);
975                 if (dev)
976                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
977
978                 fdput(f);
979                 break;
980         }
981 #endif
982 #ifdef CONFIG_KVM_XICS
983         case KVM_CAP_IRQ_XICS: {
984                 struct fd f;
985                 struct kvm_device *dev;
986
987                 r = -EBADF;
988                 f = fdget(cap->args[0]);
989                 if (!f.file)
990                         break;
991
992                 r = -EPERM;
993                 dev = kvm_device_from_filp(f.file);
994                 if (dev)
995                         r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
996
997                 fdput(f);
998                 break;
999         }
1000 #endif /* CONFIG_KVM_XICS */
1001         default:
1002                 r = -EINVAL;
1003                 break;
1004         }
1005
1006         if (!r)
1007                 r = kvmppc_sanity_check(vcpu);
1008
1009         return r;
1010 }
1011
1012 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1013                                     struct kvm_mp_state *mp_state)
1014 {
1015         return -EINVAL;
1016 }
1017
1018 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1019                                     struct kvm_mp_state *mp_state)
1020 {
1021         return -EINVAL;
1022 }
1023
1024 long kvm_arch_vcpu_ioctl(struct file *filp,
1025                          unsigned int ioctl, unsigned long arg)
1026 {
1027         struct kvm_vcpu *vcpu = filp->private_data;
1028         void __user *argp = (void __user *)arg;
1029         long r;
1030
1031         switch (ioctl) {
1032         case KVM_INTERRUPT: {
1033                 struct kvm_interrupt irq;
1034                 r = -EFAULT;
1035                 if (copy_from_user(&irq, argp, sizeof(irq)))
1036                         goto out;
1037                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1038                 goto out;
1039         }
1040
1041         case KVM_ENABLE_CAP:
1042         {
1043                 struct kvm_enable_cap cap;
1044                 r = -EFAULT;
1045                 if (copy_from_user(&cap, argp, sizeof(cap)))
1046                         goto out;
1047                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1048                 break;
1049         }
1050
1051         case KVM_SET_ONE_REG:
1052         case KVM_GET_ONE_REG:
1053         {
1054                 struct kvm_one_reg reg;
1055                 r = -EFAULT;
1056                 if (copy_from_user(&reg, argp, sizeof(reg)))
1057                         goto out;
1058                 if (ioctl == KVM_SET_ONE_REG)
1059                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
1060                 else
1061                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
1062                 break;
1063         }
1064
1065 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1066         case KVM_DIRTY_TLB: {
1067                 struct kvm_dirty_tlb dirty;
1068                 r = -EFAULT;
1069                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
1070                         goto out;
1071                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1072                 break;
1073         }
1074 #endif
1075         default:
1076                 r = -EINVAL;
1077         }
1078
1079 out:
1080         return r;
1081 }
1082
1083 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1084 {
1085         return VM_FAULT_SIGBUS;
1086 }
1087
1088 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1089 {
1090         u32 inst_nop = 0x60000000;
1091 #ifdef CONFIG_KVM_BOOKE_HV
1092         u32 inst_sc1 = 0x44000022;
1093         pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1094         pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1095         pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1096         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1097 #else
1098         u32 inst_lis = 0x3c000000;
1099         u32 inst_ori = 0x60000000;
1100         u32 inst_sc = 0x44000002;
1101         u32 inst_imm_mask = 0xffff;
1102
1103         /*
1104          * The hypercall to get into KVM from within guest context is as
1105          * follows:
1106          *
1107          *    lis r0, r0, KVM_SC_MAGIC_R0@h
1108          *    ori r0, KVM_SC_MAGIC_R0@l
1109          *    sc
1110          *    nop
1111          */
1112         pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1113         pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1114         pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1115         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1116 #endif
1117
1118         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1119
1120         return 0;
1121 }
1122
1123 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1124                           bool line_status)
1125 {
1126         if (!irqchip_in_kernel(kvm))
1127                 return -ENXIO;
1128
1129         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1130                                         irq_event->irq, irq_event->level,
1131                                         line_status);
1132         return 0;
1133 }
1134
1135
1136 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1137                                    struct kvm_enable_cap *cap)
1138 {
1139         int r;
1140
1141         if (cap->flags)
1142                 return -EINVAL;
1143
1144         switch (cap->cap) {
1145 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1146         case KVM_CAP_PPC_ENABLE_HCALL: {
1147                 unsigned long hcall = cap->args[0];
1148
1149                 r = -EINVAL;
1150                 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1151                     cap->args[1] > 1)
1152                         break;
1153                 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1154                         break;
1155                 if (cap->args[1])
1156                         set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1157                 else
1158                         clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1159                 r = 0;
1160                 break;
1161         }
1162 #endif
1163         default:
1164                 r = -EINVAL;
1165                 break;
1166         }
1167
1168         return r;
1169 }
1170
1171 long kvm_arch_vm_ioctl(struct file *filp,
1172                        unsigned int ioctl, unsigned long arg)
1173 {
1174         struct kvm *kvm __maybe_unused = filp->private_data;
1175         void __user *argp = (void __user *)arg;
1176         long r;
1177
1178         switch (ioctl) {
1179         case KVM_PPC_GET_PVINFO: {
1180                 struct kvm_ppc_pvinfo pvinfo;
1181                 memset(&pvinfo, 0, sizeof(pvinfo));
1182                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1183                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1184                         r = -EFAULT;
1185                         goto out;
1186                 }
1187
1188                 break;
1189         }
1190         case KVM_ENABLE_CAP:
1191         {
1192                 struct kvm_enable_cap cap;
1193                 r = -EFAULT;
1194                 if (copy_from_user(&cap, argp, sizeof(cap)))
1195                         goto out;
1196                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1197                 break;
1198         }
1199 #ifdef CONFIG_PPC_BOOK3S_64
1200         case KVM_CREATE_SPAPR_TCE: {
1201                 struct kvm_create_spapr_tce create_tce;
1202
1203                 r = -EFAULT;
1204                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1205                         goto out;
1206                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
1207                 goto out;
1208         }
1209         case KVM_PPC_GET_SMMU_INFO: {
1210                 struct kvm_ppc_smmu_info info;
1211                 struct kvm *kvm = filp->private_data;
1212
1213                 memset(&info, 0, sizeof(info));
1214                 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1215                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1216                         r = -EFAULT;
1217                 break;
1218         }
1219         case KVM_PPC_RTAS_DEFINE_TOKEN: {
1220                 struct kvm *kvm = filp->private_data;
1221
1222                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1223                 break;
1224         }
1225         default: {
1226                 struct kvm *kvm = filp->private_data;
1227                 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1228         }
1229 #else /* CONFIG_PPC_BOOK3S_64 */
1230         default:
1231                 r = -ENOTTY;
1232 #endif
1233         }
1234 out:
1235         return r;
1236 }
1237
1238 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1239 static unsigned long nr_lpids;
1240
1241 long kvmppc_alloc_lpid(void)
1242 {
1243         long lpid;
1244
1245         do {
1246                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1247                 if (lpid >= nr_lpids) {
1248                         pr_err("%s: No LPIDs free\n", __func__);
1249                         return -ENOMEM;
1250                 }
1251         } while (test_and_set_bit(lpid, lpid_inuse));
1252
1253         return lpid;
1254 }
1255 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1256
1257 void kvmppc_claim_lpid(long lpid)
1258 {
1259         set_bit(lpid, lpid_inuse);
1260 }
1261 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1262
1263 void kvmppc_free_lpid(long lpid)
1264 {
1265         clear_bit(lpid, lpid_inuse);
1266 }
1267 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1268
1269 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1270 {
1271         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1272         memset(lpid_inuse, 0, sizeof(lpid_inuse));
1273 }
1274 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1275
1276 int kvm_arch_init(void *opaque)
1277 {
1278         return 0;
1279 }
1280
1281 void kvm_arch_exit(void)
1282 {
1283
1284 }