]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - arch/powerpc/kvm/powerpc.c
Merge tag 'nfs-for-3.17-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs
[mirror_ubuntu-zesty-kernel.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_loadstore(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 kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
313 bool data)
314 {
315 ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
316 struct kvmppc_pte pte;
317 int r;
318
319 vcpu->stat.st++;
320
321 r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
322 XLATE_WRITE, &pte);
323 if (r < 0)
324 return r;
325
326 *eaddr = pte.raddr;
327
328 if (!pte.may_write)
329 return -EPERM;
330
331 /* Magic page override */
332 if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
333 ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
334 !(kvmppc_get_msr(vcpu) & MSR_PR)) {
335 void *magic = vcpu->arch.shared;
336 magic += pte.eaddr & 0xfff;
337 memcpy(magic, ptr, size);
338 return EMULATE_DONE;
339 }
340
341 if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
342 return EMULATE_DO_MMIO;
343
344 return EMULATE_DONE;
345 }
346 EXPORT_SYMBOL_GPL(kvmppc_st);
347
348 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
349 bool data)
350 {
351 ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
352 struct kvmppc_pte pte;
353 int rc;
354
355 vcpu->stat.ld++;
356
357 rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
358 XLATE_READ, &pte);
359 if (rc)
360 return rc;
361
362 *eaddr = pte.raddr;
363
364 if (!pte.may_read)
365 return -EPERM;
366
367 if (!data && !pte.may_execute)
368 return -ENOEXEC;
369
370 /* Magic page override */
371 if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
372 ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
373 !(kvmppc_get_msr(vcpu) & MSR_PR)) {
374 void *magic = vcpu->arch.shared;
375 magic += pte.eaddr & 0xfff;
376 memcpy(ptr, magic, size);
377 return EMULATE_DONE;
378 }
379
380 if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
381 return EMULATE_DO_MMIO;
382
383 return EMULATE_DONE;
384 }
385 EXPORT_SYMBOL_GPL(kvmppc_ld);
386
387 int kvm_arch_hardware_enable(void *garbage)
388 {
389 return 0;
390 }
391
392 void kvm_arch_hardware_disable(void *garbage)
393 {
394 }
395
396 int kvm_arch_hardware_setup(void)
397 {
398 return 0;
399 }
400
401 void kvm_arch_hardware_unsetup(void)
402 {
403 }
404
405 void kvm_arch_check_processor_compat(void *rtn)
406 {
407 *(int *)rtn = kvmppc_core_check_processor_compat();
408 }
409
410 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
411 {
412 struct kvmppc_ops *kvm_ops = NULL;
413 /*
414 * if we have both HV and PR enabled, default is HV
415 */
416 if (type == 0) {
417 if (kvmppc_hv_ops)
418 kvm_ops = kvmppc_hv_ops;
419 else
420 kvm_ops = kvmppc_pr_ops;
421 if (!kvm_ops)
422 goto err_out;
423 } else if (type == KVM_VM_PPC_HV) {
424 if (!kvmppc_hv_ops)
425 goto err_out;
426 kvm_ops = kvmppc_hv_ops;
427 } else if (type == KVM_VM_PPC_PR) {
428 if (!kvmppc_pr_ops)
429 goto err_out;
430 kvm_ops = kvmppc_pr_ops;
431 } else
432 goto err_out;
433
434 if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
435 return -ENOENT;
436
437 kvm->arch.kvm_ops = kvm_ops;
438 return kvmppc_core_init_vm(kvm);
439 err_out:
440 return -EINVAL;
441 }
442
443 void kvm_arch_destroy_vm(struct kvm *kvm)
444 {
445 unsigned int i;
446 struct kvm_vcpu *vcpu;
447
448 kvm_for_each_vcpu(i, vcpu, kvm)
449 kvm_arch_vcpu_free(vcpu);
450
451 mutex_lock(&kvm->lock);
452 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
453 kvm->vcpus[i] = NULL;
454
455 atomic_set(&kvm->online_vcpus, 0);
456
457 kvmppc_core_destroy_vm(kvm);
458
459 mutex_unlock(&kvm->lock);
460
461 /* drop the module reference */
462 module_put(kvm->arch.kvm_ops->owner);
463 }
464
465 void kvm_arch_sync_events(struct kvm *kvm)
466 {
467 }
468
469 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
470 {
471 int r;
472 /* Assume we're using HV mode when the HV module is loaded */
473 int hv_enabled = kvmppc_hv_ops ? 1 : 0;
474
475 if (kvm) {
476 /*
477 * Hooray - we know which VM type we're running on. Depend on
478 * that rather than the guess above.
479 */
480 hv_enabled = is_kvmppc_hv_enabled(kvm);
481 }
482
483 switch (ext) {
484 #ifdef CONFIG_BOOKE
485 case KVM_CAP_PPC_BOOKE_SREGS:
486 case KVM_CAP_PPC_BOOKE_WATCHDOG:
487 case KVM_CAP_PPC_EPR:
488 #else
489 case KVM_CAP_PPC_SEGSTATE:
490 case KVM_CAP_PPC_HIOR:
491 case KVM_CAP_PPC_PAPR:
492 #endif
493 case KVM_CAP_PPC_UNSET_IRQ:
494 case KVM_CAP_PPC_IRQ_LEVEL:
495 case KVM_CAP_ENABLE_CAP:
496 case KVM_CAP_ENABLE_CAP_VM:
497 case KVM_CAP_ONE_REG:
498 case KVM_CAP_IOEVENTFD:
499 case KVM_CAP_DEVICE_CTRL:
500 r = 1;
501 break;
502 case KVM_CAP_PPC_PAIRED_SINGLES:
503 case KVM_CAP_PPC_OSI:
504 case KVM_CAP_PPC_GET_PVINFO:
505 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
506 case KVM_CAP_SW_TLB:
507 #endif
508 /* We support this only for PR */
509 r = !hv_enabled;
510 break;
511 #ifdef CONFIG_KVM_MMIO
512 case KVM_CAP_COALESCED_MMIO:
513 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
514 break;
515 #endif
516 #ifdef CONFIG_KVM_MPIC
517 case KVM_CAP_IRQ_MPIC:
518 r = 1;
519 break;
520 #endif
521
522 #ifdef CONFIG_PPC_BOOK3S_64
523 case KVM_CAP_SPAPR_TCE:
524 case KVM_CAP_PPC_ALLOC_HTAB:
525 case KVM_CAP_PPC_RTAS:
526 case KVM_CAP_PPC_FIXUP_HCALL:
527 case KVM_CAP_PPC_ENABLE_HCALL:
528 #ifdef CONFIG_KVM_XICS
529 case KVM_CAP_IRQ_XICS:
530 #endif
531 r = 1;
532 break;
533 #endif /* CONFIG_PPC_BOOK3S_64 */
534 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
535 case KVM_CAP_PPC_SMT:
536 if (hv_enabled)
537 r = threads_per_subcore;
538 else
539 r = 0;
540 break;
541 case KVM_CAP_PPC_RMA:
542 r = hv_enabled;
543 /* PPC970 requires an RMA */
544 if (r && cpu_has_feature(CPU_FTR_ARCH_201))
545 r = 2;
546 break;
547 #endif
548 case KVM_CAP_SYNC_MMU:
549 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
550 if (hv_enabled)
551 r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
552 else
553 r = 0;
554 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
555 r = 1;
556 #else
557 r = 0;
558 #endif
559 break;
560 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
561 case KVM_CAP_PPC_HTAB_FD:
562 r = hv_enabled;
563 break;
564 #endif
565 case KVM_CAP_NR_VCPUS:
566 /*
567 * Recommending a number of CPUs is somewhat arbitrary; we
568 * return the number of present CPUs for -HV (since a host
569 * will have secondary threads "offline"), and for other KVM
570 * implementations just count online CPUs.
571 */
572 if (hv_enabled)
573 r = num_present_cpus();
574 else
575 r = num_online_cpus();
576 break;
577 case KVM_CAP_MAX_VCPUS:
578 r = KVM_MAX_VCPUS;
579 break;
580 #ifdef CONFIG_PPC_BOOK3S_64
581 case KVM_CAP_PPC_GET_SMMU_INFO:
582 r = 1;
583 break;
584 #endif
585 default:
586 r = 0;
587 break;
588 }
589 return r;
590
591 }
592
593 long kvm_arch_dev_ioctl(struct file *filp,
594 unsigned int ioctl, unsigned long arg)
595 {
596 return -EINVAL;
597 }
598
599 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
600 struct kvm_memory_slot *dont)
601 {
602 kvmppc_core_free_memslot(kvm, free, dont);
603 }
604
605 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
606 unsigned long npages)
607 {
608 return kvmppc_core_create_memslot(kvm, slot, npages);
609 }
610
611 void kvm_arch_memslots_updated(struct kvm *kvm)
612 {
613 }
614
615 int kvm_arch_prepare_memory_region(struct kvm *kvm,
616 struct kvm_memory_slot *memslot,
617 struct kvm_userspace_memory_region *mem,
618 enum kvm_mr_change change)
619 {
620 return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
621 }
622
623 void kvm_arch_commit_memory_region(struct kvm *kvm,
624 struct kvm_userspace_memory_region *mem,
625 const struct kvm_memory_slot *old,
626 enum kvm_mr_change change)
627 {
628 kvmppc_core_commit_memory_region(kvm, mem, old);
629 }
630
631 void kvm_arch_flush_shadow_all(struct kvm *kvm)
632 {
633 }
634
635 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
636 struct kvm_memory_slot *slot)
637 {
638 kvmppc_core_flush_memslot(kvm, slot);
639 }
640
641 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
642 {
643 struct kvm_vcpu *vcpu;
644 vcpu = kvmppc_core_vcpu_create(kvm, id);
645 if (!IS_ERR(vcpu)) {
646 vcpu->arch.wqp = &vcpu->wq;
647 kvmppc_create_vcpu_debugfs(vcpu, id);
648 }
649 return vcpu;
650 }
651
652 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
653 {
654 return 0;
655 }
656
657 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
658 {
659 /* Make sure we're not using the vcpu anymore */
660 hrtimer_cancel(&vcpu->arch.dec_timer);
661 tasklet_kill(&vcpu->arch.tasklet);
662
663 kvmppc_remove_vcpu_debugfs(vcpu);
664
665 switch (vcpu->arch.irq_type) {
666 case KVMPPC_IRQ_MPIC:
667 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
668 break;
669 case KVMPPC_IRQ_XICS:
670 kvmppc_xics_free_icp(vcpu);
671 break;
672 }
673
674 kvmppc_core_vcpu_free(vcpu);
675 }
676
677 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
678 {
679 kvm_arch_vcpu_free(vcpu);
680 }
681
682 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
683 {
684 return kvmppc_core_pending_dec(vcpu);
685 }
686
687 /*
688 * low level hrtimer wake routine. Because this runs in hardirq context
689 * we schedule a tasklet to do the real work.
690 */
691 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
692 {
693 struct kvm_vcpu *vcpu;
694
695 vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
696 tasklet_schedule(&vcpu->arch.tasklet);
697
698 return HRTIMER_NORESTART;
699 }
700
701 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
702 {
703 int ret;
704
705 hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
706 tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
707 vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
708 vcpu->arch.dec_expires = ~(u64)0;
709
710 #ifdef CONFIG_KVM_EXIT_TIMING
711 mutex_init(&vcpu->arch.exit_timing_lock);
712 #endif
713 ret = kvmppc_subarch_vcpu_init(vcpu);
714 return ret;
715 }
716
717 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
718 {
719 kvmppc_mmu_destroy(vcpu);
720 kvmppc_subarch_vcpu_uninit(vcpu);
721 }
722
723 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
724 {
725 #ifdef CONFIG_BOOKE
726 /*
727 * vrsave (formerly usprg0) isn't used by Linux, but may
728 * be used by the guest.
729 *
730 * On non-booke this is associated with Altivec and
731 * is handled by code in book3s.c.
732 */
733 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
734 #endif
735 kvmppc_core_vcpu_load(vcpu, cpu);
736 }
737
738 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
739 {
740 kvmppc_core_vcpu_put(vcpu);
741 #ifdef CONFIG_BOOKE
742 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
743 #endif
744 }
745
746 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
747 struct kvm_run *run)
748 {
749 u64 uninitialized_var(gpr);
750
751 if (run->mmio.len > sizeof(gpr)) {
752 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
753 return;
754 }
755
756 if (vcpu->arch.mmio_is_bigendian) {
757 switch (run->mmio.len) {
758 case 8: gpr = *(u64 *)run->mmio.data; break;
759 case 4: gpr = *(u32 *)run->mmio.data; break;
760 case 2: gpr = *(u16 *)run->mmio.data; break;
761 case 1: gpr = *(u8 *)run->mmio.data; break;
762 }
763 } else {
764 /* Convert BE data from userland back to LE. */
765 switch (run->mmio.len) {
766 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
767 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
768 case 1: gpr = *(u8 *)run->mmio.data; break;
769 }
770 }
771
772 if (vcpu->arch.mmio_sign_extend) {
773 switch (run->mmio.len) {
774 #ifdef CONFIG_PPC64
775 case 4:
776 gpr = (s64)(s32)gpr;
777 break;
778 #endif
779 case 2:
780 gpr = (s64)(s16)gpr;
781 break;
782 case 1:
783 gpr = (s64)(s8)gpr;
784 break;
785 }
786 }
787
788 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
789
790 switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
791 case KVM_MMIO_REG_GPR:
792 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
793 break;
794 case KVM_MMIO_REG_FPR:
795 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
796 break;
797 #ifdef CONFIG_PPC_BOOK3S
798 case KVM_MMIO_REG_QPR:
799 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
800 break;
801 case KVM_MMIO_REG_FQPR:
802 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
803 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
804 break;
805 #endif
806 default:
807 BUG();
808 }
809 }
810
811 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
812 unsigned int rt, unsigned int bytes,
813 int is_default_endian)
814 {
815 int idx, ret;
816 int is_bigendian;
817
818 if (kvmppc_need_byteswap(vcpu)) {
819 /* Default endianness is "little endian". */
820 is_bigendian = !is_default_endian;
821 } else {
822 /* Default endianness is "big endian". */
823 is_bigendian = is_default_endian;
824 }
825
826 if (bytes > sizeof(run->mmio.data)) {
827 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
828 run->mmio.len);
829 }
830
831 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
832 run->mmio.len = bytes;
833 run->mmio.is_write = 0;
834
835 vcpu->arch.io_gpr = rt;
836 vcpu->arch.mmio_is_bigendian = is_bigendian;
837 vcpu->mmio_needed = 1;
838 vcpu->mmio_is_write = 0;
839 vcpu->arch.mmio_sign_extend = 0;
840
841 idx = srcu_read_lock(&vcpu->kvm->srcu);
842
843 ret = kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
844 bytes, &run->mmio.data);
845
846 srcu_read_unlock(&vcpu->kvm->srcu, idx);
847
848 if (!ret) {
849 kvmppc_complete_mmio_load(vcpu, run);
850 vcpu->mmio_needed = 0;
851 return EMULATE_DONE;
852 }
853
854 return EMULATE_DO_MMIO;
855 }
856 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
857
858 /* Same as above, but sign extends */
859 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
860 unsigned int rt, unsigned int bytes,
861 int is_default_endian)
862 {
863 int r;
864
865 vcpu->arch.mmio_sign_extend = 1;
866 r = kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian);
867
868 return r;
869 }
870
871 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
872 u64 val, unsigned int bytes, int is_default_endian)
873 {
874 void *data = run->mmio.data;
875 int idx, ret;
876 int is_bigendian;
877
878 if (kvmppc_need_byteswap(vcpu)) {
879 /* Default endianness is "little endian". */
880 is_bigendian = !is_default_endian;
881 } else {
882 /* Default endianness is "big endian". */
883 is_bigendian = is_default_endian;
884 }
885
886 if (bytes > sizeof(run->mmio.data)) {
887 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
888 run->mmio.len);
889 }
890
891 run->mmio.phys_addr = vcpu->arch.paddr_accessed;
892 run->mmio.len = bytes;
893 run->mmio.is_write = 1;
894 vcpu->mmio_needed = 1;
895 vcpu->mmio_is_write = 1;
896
897 /* Store the value at the lowest bytes in 'data'. */
898 if (is_bigendian) {
899 switch (bytes) {
900 case 8: *(u64 *)data = val; break;
901 case 4: *(u32 *)data = val; break;
902 case 2: *(u16 *)data = val; break;
903 case 1: *(u8 *)data = val; break;
904 }
905 } else {
906 /* Store LE value into 'data'. */
907 switch (bytes) {
908 case 4: st_le32(data, val); break;
909 case 2: st_le16(data, val); break;
910 case 1: *(u8 *)data = val; break;
911 }
912 }
913
914 idx = srcu_read_lock(&vcpu->kvm->srcu);
915
916 ret = kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
917 bytes, &run->mmio.data);
918
919 srcu_read_unlock(&vcpu->kvm->srcu, idx);
920
921 if (!ret) {
922 vcpu->mmio_needed = 0;
923 return EMULATE_DONE;
924 }
925
926 return EMULATE_DO_MMIO;
927 }
928 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
929
930 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
931 {
932 int r;
933 sigset_t sigsaved;
934
935 if (vcpu->sigset_active)
936 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
937
938 if (vcpu->mmio_needed) {
939 if (!vcpu->mmio_is_write)
940 kvmppc_complete_mmio_load(vcpu, run);
941 vcpu->mmio_needed = 0;
942 } else if (vcpu->arch.osi_needed) {
943 u64 *gprs = run->osi.gprs;
944 int i;
945
946 for (i = 0; i < 32; i++)
947 kvmppc_set_gpr(vcpu, i, gprs[i]);
948 vcpu->arch.osi_needed = 0;
949 } else if (vcpu->arch.hcall_needed) {
950 int i;
951
952 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
953 for (i = 0; i < 9; ++i)
954 kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
955 vcpu->arch.hcall_needed = 0;
956 #ifdef CONFIG_BOOKE
957 } else if (vcpu->arch.epr_needed) {
958 kvmppc_set_epr(vcpu, run->epr.epr);
959 vcpu->arch.epr_needed = 0;
960 #endif
961 }
962
963 r = kvmppc_vcpu_run(run, vcpu);
964
965 if (vcpu->sigset_active)
966 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
967
968 return r;
969 }
970
971 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
972 {
973 if (irq->irq == KVM_INTERRUPT_UNSET) {
974 kvmppc_core_dequeue_external(vcpu);
975 return 0;
976 }
977
978 kvmppc_core_queue_external(vcpu, irq);
979
980 kvm_vcpu_kick(vcpu);
981
982 return 0;
983 }
984
985 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
986 struct kvm_enable_cap *cap)
987 {
988 int r;
989
990 if (cap->flags)
991 return -EINVAL;
992
993 switch (cap->cap) {
994 case KVM_CAP_PPC_OSI:
995 r = 0;
996 vcpu->arch.osi_enabled = true;
997 break;
998 case KVM_CAP_PPC_PAPR:
999 r = 0;
1000 vcpu->arch.papr_enabled = true;
1001 break;
1002 case KVM_CAP_PPC_EPR:
1003 r = 0;
1004 if (cap->args[0])
1005 vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1006 else
1007 vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1008 break;
1009 #ifdef CONFIG_BOOKE
1010 case KVM_CAP_PPC_BOOKE_WATCHDOG:
1011 r = 0;
1012 vcpu->arch.watchdog_enabled = true;
1013 break;
1014 #endif
1015 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1016 case KVM_CAP_SW_TLB: {
1017 struct kvm_config_tlb cfg;
1018 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1019
1020 r = -EFAULT;
1021 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1022 break;
1023
1024 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1025 break;
1026 }
1027 #endif
1028 #ifdef CONFIG_KVM_MPIC
1029 case KVM_CAP_IRQ_MPIC: {
1030 struct fd f;
1031 struct kvm_device *dev;
1032
1033 r = -EBADF;
1034 f = fdget(cap->args[0]);
1035 if (!f.file)
1036 break;
1037
1038 r = -EPERM;
1039 dev = kvm_device_from_filp(f.file);
1040 if (dev)
1041 r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1042
1043 fdput(f);
1044 break;
1045 }
1046 #endif
1047 #ifdef CONFIG_KVM_XICS
1048 case KVM_CAP_IRQ_XICS: {
1049 struct fd f;
1050 struct kvm_device *dev;
1051
1052 r = -EBADF;
1053 f = fdget(cap->args[0]);
1054 if (!f.file)
1055 break;
1056
1057 r = -EPERM;
1058 dev = kvm_device_from_filp(f.file);
1059 if (dev)
1060 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1061
1062 fdput(f);
1063 break;
1064 }
1065 #endif /* CONFIG_KVM_XICS */
1066 default:
1067 r = -EINVAL;
1068 break;
1069 }
1070
1071 if (!r)
1072 r = kvmppc_sanity_check(vcpu);
1073
1074 return r;
1075 }
1076
1077 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1078 struct kvm_mp_state *mp_state)
1079 {
1080 return -EINVAL;
1081 }
1082
1083 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1084 struct kvm_mp_state *mp_state)
1085 {
1086 return -EINVAL;
1087 }
1088
1089 long kvm_arch_vcpu_ioctl(struct file *filp,
1090 unsigned int ioctl, unsigned long arg)
1091 {
1092 struct kvm_vcpu *vcpu = filp->private_data;
1093 void __user *argp = (void __user *)arg;
1094 long r;
1095
1096 switch (ioctl) {
1097 case KVM_INTERRUPT: {
1098 struct kvm_interrupt irq;
1099 r = -EFAULT;
1100 if (copy_from_user(&irq, argp, sizeof(irq)))
1101 goto out;
1102 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1103 goto out;
1104 }
1105
1106 case KVM_ENABLE_CAP:
1107 {
1108 struct kvm_enable_cap cap;
1109 r = -EFAULT;
1110 if (copy_from_user(&cap, argp, sizeof(cap)))
1111 goto out;
1112 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1113 break;
1114 }
1115
1116 case KVM_SET_ONE_REG:
1117 case KVM_GET_ONE_REG:
1118 {
1119 struct kvm_one_reg reg;
1120 r = -EFAULT;
1121 if (copy_from_user(&reg, argp, sizeof(reg)))
1122 goto out;
1123 if (ioctl == KVM_SET_ONE_REG)
1124 r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
1125 else
1126 r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
1127 break;
1128 }
1129
1130 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1131 case KVM_DIRTY_TLB: {
1132 struct kvm_dirty_tlb dirty;
1133 r = -EFAULT;
1134 if (copy_from_user(&dirty, argp, sizeof(dirty)))
1135 goto out;
1136 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1137 break;
1138 }
1139 #endif
1140 default:
1141 r = -EINVAL;
1142 }
1143
1144 out:
1145 return r;
1146 }
1147
1148 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1149 {
1150 return VM_FAULT_SIGBUS;
1151 }
1152
1153 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1154 {
1155 u32 inst_nop = 0x60000000;
1156 #ifdef CONFIG_KVM_BOOKE_HV
1157 u32 inst_sc1 = 0x44000022;
1158 pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1159 pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1160 pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1161 pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1162 #else
1163 u32 inst_lis = 0x3c000000;
1164 u32 inst_ori = 0x60000000;
1165 u32 inst_sc = 0x44000002;
1166 u32 inst_imm_mask = 0xffff;
1167
1168 /*
1169 * The hypercall to get into KVM from within guest context is as
1170 * follows:
1171 *
1172 * lis r0, r0, KVM_SC_MAGIC_R0@h
1173 * ori r0, KVM_SC_MAGIC_R0@l
1174 * sc
1175 * nop
1176 */
1177 pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1178 pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1179 pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1180 pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1181 #endif
1182
1183 pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1184
1185 return 0;
1186 }
1187
1188 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1189 bool line_status)
1190 {
1191 if (!irqchip_in_kernel(kvm))
1192 return -ENXIO;
1193
1194 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1195 irq_event->irq, irq_event->level,
1196 line_status);
1197 return 0;
1198 }
1199
1200
1201 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1202 struct kvm_enable_cap *cap)
1203 {
1204 int r;
1205
1206 if (cap->flags)
1207 return -EINVAL;
1208
1209 switch (cap->cap) {
1210 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1211 case KVM_CAP_PPC_ENABLE_HCALL: {
1212 unsigned long hcall = cap->args[0];
1213
1214 r = -EINVAL;
1215 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1216 cap->args[1] > 1)
1217 break;
1218 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1219 break;
1220 if (cap->args[1])
1221 set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1222 else
1223 clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1224 r = 0;
1225 break;
1226 }
1227 #endif
1228 default:
1229 r = -EINVAL;
1230 break;
1231 }
1232
1233 return r;
1234 }
1235
1236 long kvm_arch_vm_ioctl(struct file *filp,
1237 unsigned int ioctl, unsigned long arg)
1238 {
1239 struct kvm *kvm __maybe_unused = filp->private_data;
1240 void __user *argp = (void __user *)arg;
1241 long r;
1242
1243 switch (ioctl) {
1244 case KVM_PPC_GET_PVINFO: {
1245 struct kvm_ppc_pvinfo pvinfo;
1246 memset(&pvinfo, 0, sizeof(pvinfo));
1247 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1248 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1249 r = -EFAULT;
1250 goto out;
1251 }
1252
1253 break;
1254 }
1255 case KVM_ENABLE_CAP:
1256 {
1257 struct kvm_enable_cap cap;
1258 r = -EFAULT;
1259 if (copy_from_user(&cap, argp, sizeof(cap)))
1260 goto out;
1261 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1262 break;
1263 }
1264 #ifdef CONFIG_PPC_BOOK3S_64
1265 case KVM_CREATE_SPAPR_TCE: {
1266 struct kvm_create_spapr_tce create_tce;
1267
1268 r = -EFAULT;
1269 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1270 goto out;
1271 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
1272 goto out;
1273 }
1274 case KVM_PPC_GET_SMMU_INFO: {
1275 struct kvm_ppc_smmu_info info;
1276 struct kvm *kvm = filp->private_data;
1277
1278 memset(&info, 0, sizeof(info));
1279 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1280 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1281 r = -EFAULT;
1282 break;
1283 }
1284 case KVM_PPC_RTAS_DEFINE_TOKEN: {
1285 struct kvm *kvm = filp->private_data;
1286
1287 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1288 break;
1289 }
1290 default: {
1291 struct kvm *kvm = filp->private_data;
1292 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1293 }
1294 #else /* CONFIG_PPC_BOOK3S_64 */
1295 default:
1296 r = -ENOTTY;
1297 #endif
1298 }
1299 out:
1300 return r;
1301 }
1302
1303 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1304 static unsigned long nr_lpids;
1305
1306 long kvmppc_alloc_lpid(void)
1307 {
1308 long lpid;
1309
1310 do {
1311 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1312 if (lpid >= nr_lpids) {
1313 pr_err("%s: No LPIDs free\n", __func__);
1314 return -ENOMEM;
1315 }
1316 } while (test_and_set_bit(lpid, lpid_inuse));
1317
1318 return lpid;
1319 }
1320 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1321
1322 void kvmppc_claim_lpid(long lpid)
1323 {
1324 set_bit(lpid, lpid_inuse);
1325 }
1326 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1327
1328 void kvmppc_free_lpid(long lpid)
1329 {
1330 clear_bit(lpid, lpid_inuse);
1331 }
1332 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1333
1334 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1335 {
1336 nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1337 memset(lpid_inuse, 0, sizeof(lpid_inuse));
1338 }
1339 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1340
1341 int kvm_arch_init(void *opaque)
1342 {
1343 return 0;
1344 }
1345
1346 void kvm_arch_exit(void)
1347 {
1348
1349 }
1350
1351 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);