KVM: Rename and add argument to check_extension
[linux-block.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         /* FIXME!!
398          * Should some of this be vm ioctl ? is it possible now ?
399          */
400         int hv_enabled = kvmppc_hv_ops ? 1 : 0;
401
402         switch (ext) {
403 #ifdef CONFIG_BOOKE
404         case KVM_CAP_PPC_BOOKE_SREGS:
405         case KVM_CAP_PPC_BOOKE_WATCHDOG:
406         case KVM_CAP_PPC_EPR:
407 #else
408         case KVM_CAP_PPC_SEGSTATE:
409         case KVM_CAP_PPC_HIOR:
410         case KVM_CAP_PPC_PAPR:
411 #endif
412         case KVM_CAP_PPC_UNSET_IRQ:
413         case KVM_CAP_PPC_IRQ_LEVEL:
414         case KVM_CAP_ENABLE_CAP:
415         case KVM_CAP_ENABLE_CAP_VM:
416         case KVM_CAP_ONE_REG:
417         case KVM_CAP_IOEVENTFD:
418         case KVM_CAP_DEVICE_CTRL:
419                 r = 1;
420                 break;
421         case KVM_CAP_PPC_PAIRED_SINGLES:
422         case KVM_CAP_PPC_OSI:
423         case KVM_CAP_PPC_GET_PVINFO:
424 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
425         case KVM_CAP_SW_TLB:
426 #endif
427                 /* We support this only for PR */
428                 r = !hv_enabled;
429                 break;
430 #ifdef CONFIG_KVM_MMIO
431         case KVM_CAP_COALESCED_MMIO:
432                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
433                 break;
434 #endif
435 #ifdef CONFIG_KVM_MPIC
436         case KVM_CAP_IRQ_MPIC:
437                 r = 1;
438                 break;
439 #endif
440
441 #ifdef CONFIG_PPC_BOOK3S_64
442         case KVM_CAP_SPAPR_TCE:
443         case KVM_CAP_PPC_ALLOC_HTAB:
444         case KVM_CAP_PPC_RTAS:
445         case KVM_CAP_PPC_FIXUP_HCALL:
446         case KVM_CAP_PPC_ENABLE_HCALL:
447 #ifdef CONFIG_KVM_XICS
448         case KVM_CAP_IRQ_XICS:
449 #endif
450                 r = 1;
451                 break;
452 #endif /* CONFIG_PPC_BOOK3S_64 */
453 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
454         case KVM_CAP_PPC_SMT:
455                 if (hv_enabled)
456                         r = threads_per_subcore;
457                 else
458                         r = 0;
459                 break;
460         case KVM_CAP_PPC_RMA:
461                 r = hv_enabled;
462                 /* PPC970 requires an RMA */
463                 if (r && cpu_has_feature(CPU_FTR_ARCH_201))
464                         r = 2;
465                 break;
466 #endif
467         case KVM_CAP_SYNC_MMU:
468 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
469                 if (hv_enabled)
470                         r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
471                 else
472                         r = 0;
473 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
474                 r = 1;
475 #else
476                 r = 0;
477 #endif
478                 break;
479 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
480         case KVM_CAP_PPC_HTAB_FD:
481                 r = hv_enabled;
482                 break;
483 #endif
484         case KVM_CAP_NR_VCPUS:
485                 /*
486                  * Recommending a number of CPUs is somewhat arbitrary; we
487                  * return the number of present CPUs for -HV (since a host
488                  * will have secondary threads "offline"), and for other KVM
489                  * implementations just count online CPUs.
490                  */
491                 if (hv_enabled)
492                         r = num_present_cpus();
493                 else
494                         r = num_online_cpus();
495                 break;
496         case KVM_CAP_MAX_VCPUS:
497                 r = KVM_MAX_VCPUS;
498                 break;
499 #ifdef CONFIG_PPC_BOOK3S_64
500         case KVM_CAP_PPC_GET_SMMU_INFO:
501                 r = 1;
502                 break;
503 #endif
504         default:
505                 r = 0;
506                 break;
507         }
508         return r;
509
510 }
511
512 long kvm_arch_dev_ioctl(struct file *filp,
513                         unsigned int ioctl, unsigned long arg)
514 {
515         return -EINVAL;
516 }
517
518 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
519                            struct kvm_memory_slot *dont)
520 {
521         kvmppc_core_free_memslot(kvm, free, dont);
522 }
523
524 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
525                             unsigned long npages)
526 {
527         return kvmppc_core_create_memslot(kvm, slot, npages);
528 }
529
530 void kvm_arch_memslots_updated(struct kvm *kvm)
531 {
532 }
533
534 int kvm_arch_prepare_memory_region(struct kvm *kvm,
535                                    struct kvm_memory_slot *memslot,
536                                    struct kvm_userspace_memory_region *mem,
537                                    enum kvm_mr_change change)
538 {
539         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
540 }
541
542 void kvm_arch_commit_memory_region(struct kvm *kvm,
543                                    struct kvm_userspace_memory_region *mem,
544                                    const struct kvm_memory_slot *old,
545                                    enum kvm_mr_change change)
546 {
547         kvmppc_core_commit_memory_region(kvm, mem, old);
548 }
549
550 void kvm_arch_flush_shadow_all(struct kvm *kvm)
551 {
552 }
553
554 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
555                                    struct kvm_memory_slot *slot)
556 {
557         kvmppc_core_flush_memslot(kvm, slot);
558 }
559
560 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
561 {
562         struct kvm_vcpu *vcpu;
563         vcpu = kvmppc_core_vcpu_create(kvm, id);
564         if (!IS_ERR(vcpu)) {
565                 vcpu->arch.wqp = &vcpu->wq;
566                 kvmppc_create_vcpu_debugfs(vcpu, id);
567         }
568         return vcpu;
569 }
570
571 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
572 {
573         return 0;
574 }
575
576 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
577 {
578         /* Make sure we're not using the vcpu anymore */
579         hrtimer_cancel(&vcpu->arch.dec_timer);
580         tasklet_kill(&vcpu->arch.tasklet);
581
582         kvmppc_remove_vcpu_debugfs(vcpu);
583
584         switch (vcpu->arch.irq_type) {
585         case KVMPPC_IRQ_MPIC:
586                 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
587                 break;
588         case KVMPPC_IRQ_XICS:
589                 kvmppc_xics_free_icp(vcpu);
590                 break;
591         }
592
593         kvmppc_core_vcpu_free(vcpu);
594 }
595
596 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
597 {
598         kvm_arch_vcpu_free(vcpu);
599 }
600
601 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
602 {
603         return kvmppc_core_pending_dec(vcpu);
604 }
605
606 /*
607  * low level hrtimer wake routine. Because this runs in hardirq context
608  * we schedule a tasklet to do the real work.
609  */
610 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
611 {
612         struct kvm_vcpu *vcpu;
613
614         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
615         tasklet_schedule(&vcpu->arch.tasklet);
616
617         return HRTIMER_NORESTART;
618 }
619
620 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
621 {
622         int ret;
623
624         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
625         tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
626         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
627         vcpu->arch.dec_expires = ~(u64)0;
628
629 #ifdef CONFIG_KVM_EXIT_TIMING
630         mutex_init(&vcpu->arch.exit_timing_lock);
631 #endif
632         ret = kvmppc_subarch_vcpu_init(vcpu);
633         return ret;
634 }
635
636 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
637 {
638         kvmppc_mmu_destroy(vcpu);
639         kvmppc_subarch_vcpu_uninit(vcpu);
640 }
641
642 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
643 {
644 #ifdef CONFIG_BOOKE
645         /*
646          * vrsave (formerly usprg0) isn't used by Linux, but may
647          * be used by the guest.
648          *
649          * On non-booke this is associated with Altivec and
650          * is handled by code in book3s.c.
651          */
652         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
653 #endif
654         kvmppc_core_vcpu_load(vcpu, cpu);
655 }
656
657 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
658 {
659         kvmppc_core_vcpu_put(vcpu);
660 #ifdef CONFIG_BOOKE
661         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
662 #endif
663 }
664
665 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
666                                      struct kvm_run *run)
667 {
668         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
669 }
670
671 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
672                                       struct kvm_run *run)
673 {
674         u64 uninitialized_var(gpr);
675
676         if (run->mmio.len > sizeof(gpr)) {
677                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
678                 return;
679         }
680
681         if (vcpu->arch.mmio_is_bigendian) {
682                 switch (run->mmio.len) {
683                 case 8: gpr = *(u64 *)run->mmio.data; break;
684                 case 4: gpr = *(u32 *)run->mmio.data; break;
685                 case 2: gpr = *(u16 *)run->mmio.data; break;
686                 case 1: gpr = *(u8 *)run->mmio.data; break;
687                 }
688         } else {
689                 /* Convert BE data from userland back to LE. */
690                 switch (run->mmio.len) {
691                 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
692                 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
693                 case 1: gpr = *(u8 *)run->mmio.data; break;
694                 }
695         }
696
697         if (vcpu->arch.mmio_sign_extend) {
698                 switch (run->mmio.len) {
699 #ifdef CONFIG_PPC64
700                 case 4:
701                         gpr = (s64)(s32)gpr;
702                         break;
703 #endif
704                 case 2:
705                         gpr = (s64)(s16)gpr;
706                         break;
707                 case 1:
708                         gpr = (s64)(s8)gpr;
709                         break;
710                 }
711         }
712
713         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
714
715         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
716         case KVM_MMIO_REG_GPR:
717                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
718                 break;
719         case KVM_MMIO_REG_FPR:
720                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
721                 break;
722 #ifdef CONFIG_PPC_BOOK3S
723         case KVM_MMIO_REG_QPR:
724                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
725                 break;
726         case KVM_MMIO_REG_FQPR:
727                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
728                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
729                 break;
730 #endif
731         default:
732                 BUG();
733         }
734 }
735
736 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
737                        unsigned int rt, unsigned int bytes,
738                        int is_default_endian)
739 {
740         int idx, ret;
741         int is_bigendian;
742
743         if (kvmppc_need_byteswap(vcpu)) {
744                 /* Default endianness is "little endian". */
745                 is_bigendian = !is_default_endian;
746         } else {
747                 /* Default endianness is "big endian". */
748                 is_bigendian = is_default_endian;
749         }
750
751         if (bytes > sizeof(run->mmio.data)) {
752                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
753                        run->mmio.len);
754         }
755
756         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
757         run->mmio.len = bytes;
758         run->mmio.is_write = 0;
759
760         vcpu->arch.io_gpr = rt;
761         vcpu->arch.mmio_is_bigendian = is_bigendian;
762         vcpu->mmio_needed = 1;
763         vcpu->mmio_is_write = 0;
764         vcpu->arch.mmio_sign_extend = 0;
765
766         idx = srcu_read_lock(&vcpu->kvm->srcu);
767
768         ret = kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
769                               bytes, &run->mmio.data);
770
771         srcu_read_unlock(&vcpu->kvm->srcu, idx);
772
773         if (!ret) {
774                 kvmppc_complete_mmio_load(vcpu, run);
775                 vcpu->mmio_needed = 0;
776                 return EMULATE_DONE;
777         }
778
779         return EMULATE_DO_MMIO;
780 }
781 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
782
783 /* Same as above, but sign extends */
784 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
785                         unsigned int rt, unsigned int bytes,
786                         int is_default_endian)
787 {
788         int r;
789
790         vcpu->arch.mmio_sign_extend = 1;
791         r = kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian);
792
793         return r;
794 }
795
796 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
797                         u64 val, unsigned int bytes, int is_default_endian)
798 {
799         void *data = run->mmio.data;
800         int idx, ret;
801         int is_bigendian;
802
803         if (kvmppc_need_byteswap(vcpu)) {
804                 /* Default endianness is "little endian". */
805                 is_bigendian = !is_default_endian;
806         } else {
807                 /* Default endianness is "big endian". */
808                 is_bigendian = is_default_endian;
809         }
810
811         if (bytes > sizeof(run->mmio.data)) {
812                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
813                        run->mmio.len);
814         }
815
816         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
817         run->mmio.len = bytes;
818         run->mmio.is_write = 1;
819         vcpu->mmio_needed = 1;
820         vcpu->mmio_is_write = 1;
821
822         /* Store the value at the lowest bytes in 'data'. */
823         if (is_bigendian) {
824                 switch (bytes) {
825                 case 8: *(u64 *)data = val; break;
826                 case 4: *(u32 *)data = val; break;
827                 case 2: *(u16 *)data = val; break;
828                 case 1: *(u8  *)data = val; break;
829                 }
830         } else {
831                 /* Store LE value into 'data'. */
832                 switch (bytes) {
833                 case 4: st_le32(data, val); break;
834                 case 2: st_le16(data, val); break;
835                 case 1: *(u8 *)data = val; break;
836                 }
837         }
838
839         idx = srcu_read_lock(&vcpu->kvm->srcu);
840
841         ret = kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
842                                bytes, &run->mmio.data);
843
844         srcu_read_unlock(&vcpu->kvm->srcu, idx);
845
846         if (!ret) {
847                 vcpu->mmio_needed = 0;
848                 return EMULATE_DONE;
849         }
850
851         return EMULATE_DO_MMIO;
852 }
853 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
854
855 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
856 {
857         int r;
858         sigset_t sigsaved;
859
860         if (vcpu->sigset_active)
861                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
862
863         if (vcpu->mmio_needed) {
864                 if (!vcpu->mmio_is_write)
865                         kvmppc_complete_mmio_load(vcpu, run);
866                 vcpu->mmio_needed = 0;
867         } else if (vcpu->arch.dcr_needed) {
868                 if (!vcpu->arch.dcr_is_write)
869                         kvmppc_complete_dcr_load(vcpu, run);
870                 vcpu->arch.dcr_needed = 0;
871         } else if (vcpu->arch.osi_needed) {
872                 u64 *gprs = run->osi.gprs;
873                 int i;
874
875                 for (i = 0; i < 32; i++)
876                         kvmppc_set_gpr(vcpu, i, gprs[i]);
877                 vcpu->arch.osi_needed = 0;
878         } else if (vcpu->arch.hcall_needed) {
879                 int i;
880
881                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
882                 for (i = 0; i < 9; ++i)
883                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
884                 vcpu->arch.hcall_needed = 0;
885 #ifdef CONFIG_BOOKE
886         } else if (vcpu->arch.epr_needed) {
887                 kvmppc_set_epr(vcpu, run->epr.epr);
888                 vcpu->arch.epr_needed = 0;
889 #endif
890         }
891
892         r = kvmppc_vcpu_run(run, vcpu);
893
894         if (vcpu->sigset_active)
895                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
896
897         return r;
898 }
899
900 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
901 {
902         if (irq->irq == KVM_INTERRUPT_UNSET) {
903                 kvmppc_core_dequeue_external(vcpu);
904                 return 0;
905         }
906
907         kvmppc_core_queue_external(vcpu, irq);
908
909         kvm_vcpu_kick(vcpu);
910
911         return 0;
912 }
913
914 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
915                                      struct kvm_enable_cap *cap)
916 {
917         int r;
918
919         if (cap->flags)
920                 return -EINVAL;
921
922         switch (cap->cap) {
923         case KVM_CAP_PPC_OSI:
924                 r = 0;
925                 vcpu->arch.osi_enabled = true;
926                 break;
927         case KVM_CAP_PPC_PAPR:
928                 r = 0;
929                 vcpu->arch.papr_enabled = true;
930                 break;
931         case KVM_CAP_PPC_EPR:
932                 r = 0;
933                 if (cap->args[0])
934                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
935                 else
936                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
937                 break;
938 #ifdef CONFIG_BOOKE
939         case KVM_CAP_PPC_BOOKE_WATCHDOG:
940                 r = 0;
941                 vcpu->arch.watchdog_enabled = true;
942                 break;
943 #endif
944 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
945         case KVM_CAP_SW_TLB: {
946                 struct kvm_config_tlb cfg;
947                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
948
949                 r = -EFAULT;
950                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
951                         break;
952
953                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
954                 break;
955         }
956 #endif
957 #ifdef CONFIG_KVM_MPIC
958         case KVM_CAP_IRQ_MPIC: {
959                 struct fd f;
960                 struct kvm_device *dev;
961
962                 r = -EBADF;
963                 f = fdget(cap->args[0]);
964                 if (!f.file)
965                         break;
966
967                 r = -EPERM;
968                 dev = kvm_device_from_filp(f.file);
969                 if (dev)
970                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
971
972                 fdput(f);
973                 break;
974         }
975 #endif
976 #ifdef CONFIG_KVM_XICS
977         case KVM_CAP_IRQ_XICS: {
978                 struct fd f;
979                 struct kvm_device *dev;
980
981                 r = -EBADF;
982                 f = fdget(cap->args[0]);
983                 if (!f.file)
984                         break;
985
986                 r = -EPERM;
987                 dev = kvm_device_from_filp(f.file);
988                 if (dev)
989                         r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
990
991                 fdput(f);
992                 break;
993         }
994 #endif /* CONFIG_KVM_XICS */
995         default:
996                 r = -EINVAL;
997                 break;
998         }
999
1000         if (!r)
1001                 r = kvmppc_sanity_check(vcpu);
1002
1003         return r;
1004 }
1005
1006 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1007                                     struct kvm_mp_state *mp_state)
1008 {
1009         return -EINVAL;
1010 }
1011
1012 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1013                                     struct kvm_mp_state *mp_state)
1014 {
1015         return -EINVAL;
1016 }
1017
1018 long kvm_arch_vcpu_ioctl(struct file *filp,
1019                          unsigned int ioctl, unsigned long arg)
1020 {
1021         struct kvm_vcpu *vcpu = filp->private_data;
1022         void __user *argp = (void __user *)arg;
1023         long r;
1024
1025         switch (ioctl) {
1026         case KVM_INTERRUPT: {
1027                 struct kvm_interrupt irq;
1028                 r = -EFAULT;
1029                 if (copy_from_user(&irq, argp, sizeof(irq)))
1030                         goto out;
1031                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1032                 goto out;
1033         }
1034
1035         case KVM_ENABLE_CAP:
1036         {
1037                 struct kvm_enable_cap cap;
1038                 r = -EFAULT;
1039                 if (copy_from_user(&cap, argp, sizeof(cap)))
1040                         goto out;
1041                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1042                 break;
1043         }
1044
1045         case KVM_SET_ONE_REG:
1046         case KVM_GET_ONE_REG:
1047         {
1048                 struct kvm_one_reg reg;
1049                 r = -EFAULT;
1050                 if (copy_from_user(&reg, argp, sizeof(reg)))
1051                         goto out;
1052                 if (ioctl == KVM_SET_ONE_REG)
1053                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
1054                 else
1055                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
1056                 break;
1057         }
1058
1059 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1060         case KVM_DIRTY_TLB: {
1061                 struct kvm_dirty_tlb dirty;
1062                 r = -EFAULT;
1063                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
1064                         goto out;
1065                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1066                 break;
1067         }
1068 #endif
1069         default:
1070                 r = -EINVAL;
1071         }
1072
1073 out:
1074         return r;
1075 }
1076
1077 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1078 {
1079         return VM_FAULT_SIGBUS;
1080 }
1081
1082 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1083 {
1084         u32 inst_nop = 0x60000000;
1085 #ifdef CONFIG_KVM_BOOKE_HV
1086         u32 inst_sc1 = 0x44000022;
1087         pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1088         pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1089         pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1090         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1091 #else
1092         u32 inst_lis = 0x3c000000;
1093         u32 inst_ori = 0x60000000;
1094         u32 inst_sc = 0x44000002;
1095         u32 inst_imm_mask = 0xffff;
1096
1097         /*
1098          * The hypercall to get into KVM from within guest context is as
1099          * follows:
1100          *
1101          *    lis r0, r0, KVM_SC_MAGIC_R0@h
1102          *    ori r0, KVM_SC_MAGIC_R0@l
1103          *    sc
1104          *    nop
1105          */
1106         pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1107         pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1108         pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1109         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1110 #endif
1111
1112         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1113
1114         return 0;
1115 }
1116
1117 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1118                           bool line_status)
1119 {
1120         if (!irqchip_in_kernel(kvm))
1121                 return -ENXIO;
1122
1123         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1124                                         irq_event->irq, irq_event->level,
1125                                         line_status);
1126         return 0;
1127 }
1128
1129
1130 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1131                                    struct kvm_enable_cap *cap)
1132 {
1133         int r;
1134
1135         if (cap->flags)
1136                 return -EINVAL;
1137
1138         switch (cap->cap) {
1139 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1140         case KVM_CAP_PPC_ENABLE_HCALL: {
1141                 unsigned long hcall = cap->args[0];
1142
1143                 r = -EINVAL;
1144                 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1145                     cap->args[1] > 1)
1146                         break;
1147                 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1148                         break;
1149                 if (cap->args[1])
1150                         set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1151                 else
1152                         clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1153                 r = 0;
1154                 break;
1155         }
1156 #endif
1157         default:
1158                 r = -EINVAL;
1159                 break;
1160         }
1161
1162         return r;
1163 }
1164
1165 long kvm_arch_vm_ioctl(struct file *filp,
1166                        unsigned int ioctl, unsigned long arg)
1167 {
1168         struct kvm *kvm __maybe_unused = filp->private_data;
1169         void __user *argp = (void __user *)arg;
1170         long r;
1171
1172         switch (ioctl) {
1173         case KVM_PPC_GET_PVINFO: {
1174                 struct kvm_ppc_pvinfo pvinfo;
1175                 memset(&pvinfo, 0, sizeof(pvinfo));
1176                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1177                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1178                         r = -EFAULT;
1179                         goto out;
1180                 }
1181
1182                 break;
1183         }
1184         case KVM_ENABLE_CAP:
1185         {
1186                 struct kvm_enable_cap cap;
1187                 r = -EFAULT;
1188                 if (copy_from_user(&cap, argp, sizeof(cap)))
1189                         goto out;
1190                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1191                 break;
1192         }
1193 #ifdef CONFIG_PPC_BOOK3S_64
1194         case KVM_CREATE_SPAPR_TCE: {
1195                 struct kvm_create_spapr_tce create_tce;
1196
1197                 r = -EFAULT;
1198                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1199                         goto out;
1200                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
1201                 goto out;
1202         }
1203         case KVM_PPC_GET_SMMU_INFO: {
1204                 struct kvm_ppc_smmu_info info;
1205                 struct kvm *kvm = filp->private_data;
1206
1207                 memset(&info, 0, sizeof(info));
1208                 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1209                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1210                         r = -EFAULT;
1211                 break;
1212         }
1213         case KVM_PPC_RTAS_DEFINE_TOKEN: {
1214                 struct kvm *kvm = filp->private_data;
1215
1216                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1217                 break;
1218         }
1219         default: {
1220                 struct kvm *kvm = filp->private_data;
1221                 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1222         }
1223 #else /* CONFIG_PPC_BOOK3S_64 */
1224         default:
1225                 r = -ENOTTY;
1226 #endif
1227         }
1228 out:
1229         return r;
1230 }
1231
1232 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1233 static unsigned long nr_lpids;
1234
1235 long kvmppc_alloc_lpid(void)
1236 {
1237         long lpid;
1238
1239         do {
1240                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1241                 if (lpid >= nr_lpids) {
1242                         pr_err("%s: No LPIDs free\n", __func__);
1243                         return -ENOMEM;
1244                 }
1245         } while (test_and_set_bit(lpid, lpid_inuse));
1246
1247         return lpid;
1248 }
1249 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1250
1251 void kvmppc_claim_lpid(long lpid)
1252 {
1253         set_bit(lpid, lpid_inuse);
1254 }
1255 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1256
1257 void kvmppc_free_lpid(long lpid)
1258 {
1259         clear_bit(lpid, lpid_inuse);
1260 }
1261 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1262
1263 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1264 {
1265         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1266         memset(lpid_inuse, 0, sizeof(lpid_inuse));
1267 }
1268 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1269
1270 int kvm_arch_init(void *opaque)
1271 {
1272         return 0;
1273 }
1274
1275 void kvm_arch_exit(void)
1276 {
1277
1278 }