]> Pileus Git - ~andy/linux/blob - arch/powerpc/kvm/powerpc.c
16b45954511c708c29e0d6b94c584ea41b8e96ca
[~andy/linux] / 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 <asm/cputable.h>
29 #include <asm/uaccess.h>
30 #include <asm/kvm_ppc.h>
31 #include <asm/tlbflush.h>
32 #include <asm/cputhreads.h>
33 #include <asm/irqflags.h>
34 #include "timing.h"
35 #include "../mm/mmu_decl.h"
36
37 #define CREATE_TRACE_POINTS
38 #include "trace.h"
39
40 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
41 {
42         return !!(v->arch.pending_exceptions) ||
43                v->requests;
44 }
45
46 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
47 {
48         return 1;
49 }
50
51 #ifndef CONFIG_KVM_BOOK3S_64_HV
52 /*
53  * Common checks before entering the guest world.  Call with interrupts
54  * disabled.
55  *
56  * returns:
57  *
58  * == 1 if we're ready to go into guest state
59  * <= 0 if we need to go back to the host with return value
60  */
61 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
62 {
63         int r = 1;
64
65         WARN_ON_ONCE(!irqs_disabled());
66         while (true) {
67                 if (need_resched()) {
68                         local_irq_enable();
69                         cond_resched();
70                         local_irq_disable();
71                         continue;
72                 }
73
74                 if (signal_pending(current)) {
75                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
76                         vcpu->run->exit_reason = KVM_EXIT_INTR;
77                         r = -EINTR;
78                         break;
79                 }
80
81                 vcpu->mode = IN_GUEST_MODE;
82
83                 /*
84                  * Reading vcpu->requests must happen after setting vcpu->mode,
85                  * so we don't miss a request because the requester sees
86                  * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
87                  * before next entering the guest (and thus doesn't IPI).
88                  */
89                 smp_mb();
90
91                 if (vcpu->requests) {
92                         /* Make sure we process requests preemptable */
93                         local_irq_enable();
94                         trace_kvm_check_requests(vcpu);
95                         r = kvmppc_core_check_requests(vcpu);
96                         local_irq_disable();
97                         if (r > 0)
98                                 continue;
99                         break;
100                 }
101
102                 if (kvmppc_core_prepare_to_enter(vcpu)) {
103                         /* interrupts got enabled in between, so we
104                            are back at square 1 */
105                         continue;
106                 }
107
108 #ifdef CONFIG_PPC64
109                 /* lazy EE magic */
110                 hard_irq_disable();
111                 if (lazy_irq_pending()) {
112                         /* Got an interrupt in between, try again */
113                         local_irq_enable();
114                         local_irq_disable();
115                         kvm_guest_exit();
116                         continue;
117                 }
118
119                 trace_hardirqs_on();
120 #endif
121
122                 kvm_guest_enter();
123                 break;
124         }
125
126         return r;
127 }
128 #endif /* CONFIG_KVM_BOOK3S_64_HV */
129
130 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
131 {
132         int nr = kvmppc_get_gpr(vcpu, 11);
133         int r;
134         unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
135         unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
136         unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
137         unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
138         unsigned long r2 = 0;
139
140         if (!(vcpu->arch.shared->msr & MSR_SF)) {
141                 /* 32 bit mode */
142                 param1 &= 0xffffffff;
143                 param2 &= 0xffffffff;
144                 param3 &= 0xffffffff;
145                 param4 &= 0xffffffff;
146         }
147
148         switch (nr) {
149         case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
150         {
151                 vcpu->arch.magic_page_pa = param1;
152                 vcpu->arch.magic_page_ea = param2;
153
154                 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
155
156                 r = EV_SUCCESS;
157                 break;
158         }
159         case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
160                 r = EV_SUCCESS;
161 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
162                 /* XXX Missing magic page on 44x */
163                 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
164 #endif
165
166                 /* Second return value is in r4 */
167                 break;
168         case EV_HCALL_TOKEN(EV_IDLE):
169                 r = EV_SUCCESS;
170                 kvm_vcpu_block(vcpu);
171                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
172                 break;
173         default:
174                 r = EV_UNIMPLEMENTED;
175                 break;
176         }
177
178         kvmppc_set_gpr(vcpu, 4, r2);
179
180         return r;
181 }
182
183 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
184 {
185         int r = false;
186
187         /* We have to know what CPU to virtualize */
188         if (!vcpu->arch.pvr)
189                 goto out;
190
191         /* PAPR only works with book3s_64 */
192         if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
193                 goto out;
194
195 #ifdef CONFIG_KVM_BOOK3S_64_HV
196         /* HV KVM can only do PAPR mode for now */
197         if (!vcpu->arch.papr_enabled)
198                 goto out;
199 #endif
200
201 #ifdef CONFIG_KVM_BOOKE_HV
202         if (!cpu_has_feature(CPU_FTR_EMB_HV))
203                 goto out;
204 #endif
205
206         r = true;
207
208 out:
209         vcpu->arch.sane = r;
210         return r ? 0 : -EINVAL;
211 }
212
213 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
214 {
215         enum emulation_result er;
216         int r;
217
218         er = kvmppc_emulate_instruction(run, vcpu);
219         switch (er) {
220         case EMULATE_DONE:
221                 /* Future optimization: only reload non-volatiles if they were
222                  * actually modified. */
223                 r = RESUME_GUEST_NV;
224                 break;
225         case EMULATE_DO_MMIO:
226                 run->exit_reason = KVM_EXIT_MMIO;
227                 /* We must reload nonvolatiles because "update" load/store
228                  * instructions modify register state. */
229                 /* Future optimization: only reload non-volatiles if they were
230                  * actually modified. */
231                 r = RESUME_HOST_NV;
232                 break;
233         case EMULATE_FAIL:
234                 /* XXX Deliver Program interrupt to guest. */
235                 printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__,
236                        kvmppc_get_last_inst(vcpu));
237                 r = RESUME_HOST;
238                 break;
239         default:
240                 WARN_ON(1);
241                 r = RESUME_GUEST;
242         }
243
244         return r;
245 }
246
247 int kvm_arch_hardware_enable(void *garbage)
248 {
249         return 0;
250 }
251
252 void kvm_arch_hardware_disable(void *garbage)
253 {
254 }
255
256 int kvm_arch_hardware_setup(void)
257 {
258         return 0;
259 }
260
261 void kvm_arch_hardware_unsetup(void)
262 {
263 }
264
265 void kvm_arch_check_processor_compat(void *rtn)
266 {
267         *(int *)rtn = kvmppc_core_check_processor_compat();
268 }
269
270 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
271 {
272         if (type)
273                 return -EINVAL;
274
275         return kvmppc_core_init_vm(kvm);
276 }
277
278 void kvm_arch_destroy_vm(struct kvm *kvm)
279 {
280         unsigned int i;
281         struct kvm_vcpu *vcpu;
282
283         kvm_for_each_vcpu(i, vcpu, kvm)
284                 kvm_arch_vcpu_free(vcpu);
285
286         mutex_lock(&kvm->lock);
287         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
288                 kvm->vcpus[i] = NULL;
289
290         atomic_set(&kvm->online_vcpus, 0);
291
292         kvmppc_core_destroy_vm(kvm);
293
294         mutex_unlock(&kvm->lock);
295 }
296
297 void kvm_arch_sync_events(struct kvm *kvm)
298 {
299 }
300
301 int kvm_dev_ioctl_check_extension(long ext)
302 {
303         int r;
304
305         switch (ext) {
306 #ifdef CONFIG_BOOKE
307         case KVM_CAP_PPC_BOOKE_SREGS:
308         case KVM_CAP_PPC_BOOKE_WATCHDOG:
309         case KVM_CAP_PPC_EPR:
310 #else
311         case KVM_CAP_PPC_SEGSTATE:
312         case KVM_CAP_PPC_HIOR:
313         case KVM_CAP_PPC_PAPR:
314 #endif
315         case KVM_CAP_PPC_UNSET_IRQ:
316         case KVM_CAP_PPC_IRQ_LEVEL:
317         case KVM_CAP_ENABLE_CAP:
318         case KVM_CAP_ONE_REG:
319         case KVM_CAP_IOEVENTFD:
320                 r = 1;
321                 break;
322 #ifndef CONFIG_KVM_BOOK3S_64_HV
323         case KVM_CAP_PPC_PAIRED_SINGLES:
324         case KVM_CAP_PPC_OSI:
325         case KVM_CAP_PPC_GET_PVINFO:
326 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
327         case KVM_CAP_SW_TLB:
328 #endif
329                 r = 1;
330                 break;
331         case KVM_CAP_COALESCED_MMIO:
332                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
333                 break;
334 #endif
335 #ifdef CONFIG_PPC_BOOK3S_64
336         case KVM_CAP_SPAPR_TCE:
337         case KVM_CAP_PPC_ALLOC_HTAB:
338                 r = 1;
339                 break;
340 #endif /* CONFIG_PPC_BOOK3S_64 */
341 #ifdef CONFIG_KVM_BOOK3S_64_HV
342         case KVM_CAP_PPC_SMT:
343                 r = threads_per_core;
344                 break;
345         case KVM_CAP_PPC_RMA:
346                 r = 1;
347                 /* PPC970 requires an RMA */
348                 if (cpu_has_feature(CPU_FTR_ARCH_201))
349                         r = 2;
350                 break;
351 #endif
352         case KVM_CAP_SYNC_MMU:
353 #ifdef CONFIG_KVM_BOOK3S_64_HV
354                 r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
355 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
356                 r = 1;
357 #else
358                 r = 0;
359                 break;
360 #endif
361 #ifdef CONFIG_KVM_BOOK3S_64_HV
362         case KVM_CAP_PPC_HTAB_FD:
363                 r = 1;
364                 break;
365 #endif
366                 break;
367         case KVM_CAP_NR_VCPUS:
368                 /*
369                  * Recommending a number of CPUs is somewhat arbitrary; we
370                  * return the number of present CPUs for -HV (since a host
371                  * will have secondary threads "offline"), and for other KVM
372                  * implementations just count online CPUs.
373                  */
374 #ifdef CONFIG_KVM_BOOK3S_64_HV
375                 r = num_present_cpus();
376 #else
377                 r = num_online_cpus();
378 #endif
379                 break;
380         case KVM_CAP_MAX_VCPUS:
381                 r = KVM_MAX_VCPUS;
382                 break;
383 #ifdef CONFIG_PPC_BOOK3S_64
384         case KVM_CAP_PPC_GET_SMMU_INFO:
385                 r = 1;
386                 break;
387 #endif
388         default:
389                 r = 0;
390                 break;
391         }
392         return r;
393
394 }
395
396 long kvm_arch_dev_ioctl(struct file *filp,
397                         unsigned int ioctl, unsigned long arg)
398 {
399         return -EINVAL;
400 }
401
402 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
403                            struct kvm_memory_slot *dont)
404 {
405         kvmppc_core_free_memslot(free, dont);
406 }
407
408 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
409 {
410         return kvmppc_core_create_memslot(slot, npages);
411 }
412
413 int kvm_arch_prepare_memory_region(struct kvm *kvm,
414                                    struct kvm_memory_slot *memslot,
415                                    struct kvm_userspace_memory_region *mem,
416                                    enum kvm_mr_change change)
417 {
418         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
419 }
420
421 void kvm_arch_commit_memory_region(struct kvm *kvm,
422                                    struct kvm_userspace_memory_region *mem,
423                                    const struct kvm_memory_slot *old,
424                                    enum kvm_mr_change change)
425 {
426         kvmppc_core_commit_memory_region(kvm, mem, old);
427 }
428
429 void kvm_arch_flush_shadow_all(struct kvm *kvm)
430 {
431 }
432
433 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
434                                    struct kvm_memory_slot *slot)
435 {
436         kvmppc_core_flush_memslot(kvm, slot);
437 }
438
439 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
440 {
441         struct kvm_vcpu *vcpu;
442         vcpu = kvmppc_core_vcpu_create(kvm, id);
443         if (!IS_ERR(vcpu)) {
444                 vcpu->arch.wqp = &vcpu->wq;
445                 kvmppc_create_vcpu_debugfs(vcpu, id);
446         }
447         return vcpu;
448 }
449
450 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
451 {
452         return 0;
453 }
454
455 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
456 {
457         /* Make sure we're not using the vcpu anymore */
458         hrtimer_cancel(&vcpu->arch.dec_timer);
459         tasklet_kill(&vcpu->arch.tasklet);
460
461         kvmppc_remove_vcpu_debugfs(vcpu);
462         kvmppc_core_vcpu_free(vcpu);
463 }
464
465 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
466 {
467         kvm_arch_vcpu_free(vcpu);
468 }
469
470 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
471 {
472         return kvmppc_core_pending_dec(vcpu);
473 }
474
475 /*
476  * low level hrtimer wake routine. Because this runs in hardirq context
477  * we schedule a tasklet to do the real work.
478  */
479 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
480 {
481         struct kvm_vcpu *vcpu;
482
483         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
484         tasklet_schedule(&vcpu->arch.tasklet);
485
486         return HRTIMER_NORESTART;
487 }
488
489 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
490 {
491         int ret;
492
493         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
494         tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
495         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
496         vcpu->arch.dec_expires = ~(u64)0;
497
498 #ifdef CONFIG_KVM_EXIT_TIMING
499         mutex_init(&vcpu->arch.exit_timing_lock);
500 #endif
501         ret = kvmppc_subarch_vcpu_init(vcpu);
502         return ret;
503 }
504
505 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
506 {
507         kvmppc_mmu_destroy(vcpu);
508         kvmppc_subarch_vcpu_uninit(vcpu);
509 }
510
511 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
512 {
513 #ifdef CONFIG_BOOKE
514         /*
515          * vrsave (formerly usprg0) isn't used by Linux, but may
516          * be used by the guest.
517          *
518          * On non-booke this is associated with Altivec and
519          * is handled by code in book3s.c.
520          */
521         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
522 #endif
523         kvmppc_core_vcpu_load(vcpu, cpu);
524 }
525
526 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
527 {
528         kvmppc_core_vcpu_put(vcpu);
529 #ifdef CONFIG_BOOKE
530         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
531 #endif
532 }
533
534 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
535                                         struct kvm_guest_debug *dbg)
536 {
537         return -EINVAL;
538 }
539
540 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
541                                      struct kvm_run *run)
542 {
543         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
544 }
545
546 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
547                                       struct kvm_run *run)
548 {
549         u64 uninitialized_var(gpr);
550
551         if (run->mmio.len > sizeof(gpr)) {
552                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
553                 return;
554         }
555
556         if (vcpu->arch.mmio_is_bigendian) {
557                 switch (run->mmio.len) {
558                 case 8: gpr = *(u64 *)run->mmio.data; break;
559                 case 4: gpr = *(u32 *)run->mmio.data; break;
560                 case 2: gpr = *(u16 *)run->mmio.data; break;
561                 case 1: gpr = *(u8 *)run->mmio.data; break;
562                 }
563         } else {
564                 /* Convert BE data from userland back to LE. */
565                 switch (run->mmio.len) {
566                 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
567                 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
568                 case 1: gpr = *(u8 *)run->mmio.data; break;
569                 }
570         }
571
572         if (vcpu->arch.mmio_sign_extend) {
573                 switch (run->mmio.len) {
574 #ifdef CONFIG_PPC64
575                 case 4:
576                         gpr = (s64)(s32)gpr;
577                         break;
578 #endif
579                 case 2:
580                         gpr = (s64)(s16)gpr;
581                         break;
582                 case 1:
583                         gpr = (s64)(s8)gpr;
584                         break;
585                 }
586         }
587
588         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
589
590         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
591         case KVM_MMIO_REG_GPR:
592                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
593                 break;
594         case KVM_MMIO_REG_FPR:
595                 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
596                 break;
597 #ifdef CONFIG_PPC_BOOK3S
598         case KVM_MMIO_REG_QPR:
599                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
600                 break;
601         case KVM_MMIO_REG_FQPR:
602                 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
603                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
604                 break;
605 #endif
606         default:
607                 BUG();
608         }
609 }
610
611 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
612                        unsigned int rt, unsigned int bytes, int is_bigendian)
613 {
614         if (bytes > sizeof(run->mmio.data)) {
615                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
616                        run->mmio.len);
617         }
618
619         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
620         run->mmio.len = bytes;
621         run->mmio.is_write = 0;
622
623         vcpu->arch.io_gpr = rt;
624         vcpu->arch.mmio_is_bigendian = is_bigendian;
625         vcpu->mmio_needed = 1;
626         vcpu->mmio_is_write = 0;
627         vcpu->arch.mmio_sign_extend = 0;
628
629         if (!kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
630                              bytes, &run->mmio.data)) {
631                 kvmppc_complete_mmio_load(vcpu, run);
632                 vcpu->mmio_needed = 0;
633                 return EMULATE_DONE;
634         }
635
636         return EMULATE_DO_MMIO;
637 }
638
639 /* Same as above, but sign extends */
640 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
641                         unsigned int rt, unsigned int bytes, int is_bigendian)
642 {
643         int r;
644
645         vcpu->arch.mmio_sign_extend = 1;
646         r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian);
647
648         return r;
649 }
650
651 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
652                         u64 val, unsigned int bytes, int is_bigendian)
653 {
654         void *data = run->mmio.data;
655
656         if (bytes > sizeof(run->mmio.data)) {
657                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
658                        run->mmio.len);
659         }
660
661         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
662         run->mmio.len = bytes;
663         run->mmio.is_write = 1;
664         vcpu->mmio_needed = 1;
665         vcpu->mmio_is_write = 1;
666
667         /* Store the value at the lowest bytes in 'data'. */
668         if (is_bigendian) {
669                 switch (bytes) {
670                 case 8: *(u64 *)data = val; break;
671                 case 4: *(u32 *)data = val; break;
672                 case 2: *(u16 *)data = val; break;
673                 case 1: *(u8  *)data = val; break;
674                 }
675         } else {
676                 /* Store LE value into 'data'. */
677                 switch (bytes) {
678                 case 4: st_le32(data, val); break;
679                 case 2: st_le16(data, val); break;
680                 case 1: *(u8 *)data = val; break;
681                 }
682         }
683
684         if (!kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
685                               bytes, &run->mmio.data)) {
686                 kvmppc_complete_mmio_load(vcpu, run);
687                 vcpu->mmio_needed = 0;
688                 return EMULATE_DONE;
689         }
690
691         return EMULATE_DO_MMIO;
692 }
693
694 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
695 {
696         int r;
697         sigset_t sigsaved;
698
699         if (vcpu->sigset_active)
700                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
701
702         if (vcpu->mmio_needed) {
703                 if (!vcpu->mmio_is_write)
704                         kvmppc_complete_mmio_load(vcpu, run);
705                 vcpu->mmio_needed = 0;
706         } else if (vcpu->arch.dcr_needed) {
707                 if (!vcpu->arch.dcr_is_write)
708                         kvmppc_complete_dcr_load(vcpu, run);
709                 vcpu->arch.dcr_needed = 0;
710         } else if (vcpu->arch.osi_needed) {
711                 u64 *gprs = run->osi.gprs;
712                 int i;
713
714                 for (i = 0; i < 32; i++)
715                         kvmppc_set_gpr(vcpu, i, gprs[i]);
716                 vcpu->arch.osi_needed = 0;
717         } else if (vcpu->arch.hcall_needed) {
718                 int i;
719
720                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
721                 for (i = 0; i < 9; ++i)
722                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
723                 vcpu->arch.hcall_needed = 0;
724 #ifdef CONFIG_BOOKE
725         } else if (vcpu->arch.epr_needed) {
726                 kvmppc_set_epr(vcpu, run->epr.epr);
727                 vcpu->arch.epr_needed = 0;
728 #endif
729         }
730
731         r = kvmppc_vcpu_run(run, vcpu);
732
733         if (vcpu->sigset_active)
734                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
735
736         return r;
737 }
738
739 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
740 {
741         if (irq->irq == KVM_INTERRUPT_UNSET) {
742                 kvmppc_core_dequeue_external(vcpu);
743                 return 0;
744         }
745
746         kvmppc_core_queue_external(vcpu, irq);
747
748         kvm_vcpu_kick(vcpu);
749
750         return 0;
751 }
752
753 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
754                                      struct kvm_enable_cap *cap)
755 {
756         int r;
757
758         if (cap->flags)
759                 return -EINVAL;
760
761         switch (cap->cap) {
762         case KVM_CAP_PPC_OSI:
763                 r = 0;
764                 vcpu->arch.osi_enabled = true;
765                 break;
766         case KVM_CAP_PPC_PAPR:
767                 r = 0;
768                 vcpu->arch.papr_enabled = true;
769                 break;
770         case KVM_CAP_PPC_EPR:
771                 r = 0;
772                 vcpu->arch.epr_enabled = cap->args[0];
773                 break;
774 #ifdef CONFIG_BOOKE
775         case KVM_CAP_PPC_BOOKE_WATCHDOG:
776                 r = 0;
777                 vcpu->arch.watchdog_enabled = true;
778                 break;
779 #endif
780 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
781         case KVM_CAP_SW_TLB: {
782                 struct kvm_config_tlb cfg;
783                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
784
785                 r = -EFAULT;
786                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
787                         break;
788
789                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
790                 break;
791         }
792 #endif
793         default:
794                 r = -EINVAL;
795                 break;
796         }
797
798         if (!r)
799                 r = kvmppc_sanity_check(vcpu);
800
801         return r;
802 }
803
804 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
805                                     struct kvm_mp_state *mp_state)
806 {
807         return -EINVAL;
808 }
809
810 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
811                                     struct kvm_mp_state *mp_state)
812 {
813         return -EINVAL;
814 }
815
816 long kvm_arch_vcpu_ioctl(struct file *filp,
817                          unsigned int ioctl, unsigned long arg)
818 {
819         struct kvm_vcpu *vcpu = filp->private_data;
820         void __user *argp = (void __user *)arg;
821         long r;
822
823         switch (ioctl) {
824         case KVM_INTERRUPT: {
825                 struct kvm_interrupt irq;
826                 r = -EFAULT;
827                 if (copy_from_user(&irq, argp, sizeof(irq)))
828                         goto out;
829                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
830                 goto out;
831         }
832
833         case KVM_ENABLE_CAP:
834         {
835                 struct kvm_enable_cap cap;
836                 r = -EFAULT;
837                 if (copy_from_user(&cap, argp, sizeof(cap)))
838                         goto out;
839                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
840                 break;
841         }
842
843         case KVM_SET_ONE_REG:
844         case KVM_GET_ONE_REG:
845         {
846                 struct kvm_one_reg reg;
847                 r = -EFAULT;
848                 if (copy_from_user(&reg, argp, sizeof(reg)))
849                         goto out;
850                 if (ioctl == KVM_SET_ONE_REG)
851                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
852                 else
853                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
854                 break;
855         }
856
857 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
858         case KVM_DIRTY_TLB: {
859                 struct kvm_dirty_tlb dirty;
860                 r = -EFAULT;
861                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
862                         goto out;
863                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
864                 break;
865         }
866 #endif
867         default:
868                 r = -EINVAL;
869         }
870
871 out:
872         return r;
873 }
874
875 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
876 {
877         return VM_FAULT_SIGBUS;
878 }
879
880 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
881 {
882         u32 inst_nop = 0x60000000;
883 #ifdef CONFIG_KVM_BOOKE_HV
884         u32 inst_sc1 = 0x44000022;
885         pvinfo->hcall[0] = inst_sc1;
886         pvinfo->hcall[1] = inst_nop;
887         pvinfo->hcall[2] = inst_nop;
888         pvinfo->hcall[3] = inst_nop;
889 #else
890         u32 inst_lis = 0x3c000000;
891         u32 inst_ori = 0x60000000;
892         u32 inst_sc = 0x44000002;
893         u32 inst_imm_mask = 0xffff;
894
895         /*
896          * The hypercall to get into KVM from within guest context is as
897          * follows:
898          *
899          *    lis r0, r0, KVM_SC_MAGIC_R0@h
900          *    ori r0, KVM_SC_MAGIC_R0@l
901          *    sc
902          *    nop
903          */
904         pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
905         pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
906         pvinfo->hcall[2] = inst_sc;
907         pvinfo->hcall[3] = inst_nop;
908 #endif
909
910         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
911
912         return 0;
913 }
914
915 long kvm_arch_vm_ioctl(struct file *filp,
916                        unsigned int ioctl, unsigned long arg)
917 {
918         void __user *argp = (void __user *)arg;
919         long r;
920
921         switch (ioctl) {
922         case KVM_PPC_GET_PVINFO: {
923                 struct kvm_ppc_pvinfo pvinfo;
924                 memset(&pvinfo, 0, sizeof(pvinfo));
925                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
926                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
927                         r = -EFAULT;
928                         goto out;
929                 }
930
931                 break;
932         }
933 #ifdef CONFIG_PPC_BOOK3S_64
934         case KVM_CREATE_SPAPR_TCE: {
935                 struct kvm_create_spapr_tce create_tce;
936                 struct kvm *kvm = filp->private_data;
937
938                 r = -EFAULT;
939                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
940                         goto out;
941                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
942                 goto out;
943         }
944 #endif /* CONFIG_PPC_BOOK3S_64 */
945
946 #ifdef CONFIG_KVM_BOOK3S_64_HV
947         case KVM_ALLOCATE_RMA: {
948                 struct kvm *kvm = filp->private_data;
949                 struct kvm_allocate_rma rma;
950
951                 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
952                 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
953                         r = -EFAULT;
954                 break;
955         }
956
957         case KVM_PPC_ALLOCATE_HTAB: {
958                 struct kvm *kvm = filp->private_data;
959                 u32 htab_order;
960
961                 r = -EFAULT;
962                 if (get_user(htab_order, (u32 __user *)argp))
963                         break;
964                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
965                 if (r)
966                         break;
967                 r = -EFAULT;
968                 if (put_user(htab_order, (u32 __user *)argp))
969                         break;
970                 r = 0;
971                 break;
972         }
973
974         case KVM_PPC_GET_HTAB_FD: {
975                 struct kvm *kvm = filp->private_data;
976                 struct kvm_get_htab_fd ghf;
977
978                 r = -EFAULT;
979                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
980                         break;
981                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
982                 break;
983         }
984 #endif /* CONFIG_KVM_BOOK3S_64_HV */
985
986 #ifdef CONFIG_PPC_BOOK3S_64
987         case KVM_PPC_GET_SMMU_INFO: {
988                 struct kvm *kvm = filp->private_data;
989                 struct kvm_ppc_smmu_info info;
990
991                 memset(&info, 0, sizeof(info));
992                 r = kvm_vm_ioctl_get_smmu_info(kvm, &info);
993                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
994                         r = -EFAULT;
995                 break;
996         }
997 #endif /* CONFIG_PPC_BOOK3S_64 */
998         default:
999                 r = -ENOTTY;
1000         }
1001
1002 out:
1003         return r;
1004 }
1005
1006 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1007 static unsigned long nr_lpids;
1008
1009 long kvmppc_alloc_lpid(void)
1010 {
1011         long lpid;
1012
1013         do {
1014                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1015                 if (lpid >= nr_lpids) {
1016                         pr_err("%s: No LPIDs free\n", __func__);
1017                         return -ENOMEM;
1018                 }
1019         } while (test_and_set_bit(lpid, lpid_inuse));
1020
1021         return lpid;
1022 }
1023
1024 void kvmppc_claim_lpid(long lpid)
1025 {
1026         set_bit(lpid, lpid_inuse);
1027 }
1028
1029 void kvmppc_free_lpid(long lpid)
1030 {
1031         clear_bit(lpid, lpid_inuse);
1032 }
1033
1034 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1035 {
1036         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1037         memset(lpid_inuse, 0, sizeof(lpid_inuse));
1038 }
1039
1040 int kvm_arch_init(void *opaque)
1041 {
1042         return 0;
1043 }
1044
1045 void kvm_arch_exit(void)
1046 {
1047 }