xref: /linux/arch/powerpc/kvm/powerpc.c (revision cc04a46f11ea046ed53e2c832ae29e4790f7e35f)
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, INST_GENERIC, &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)
388 {
389 	return 0;
390 }
391 
392 int kvm_arch_hardware_setup(void)
393 {
394 	return 0;
395 }
396 
397 void kvm_arch_check_processor_compat(void *rtn)
398 {
399 	*(int *)rtn = kvmppc_core_check_processor_compat();
400 }
401 
402 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
403 {
404 	struct kvmppc_ops *kvm_ops = NULL;
405 	/*
406 	 * if we have both HV and PR enabled, default is HV
407 	 */
408 	if (type == 0) {
409 		if (kvmppc_hv_ops)
410 			kvm_ops = kvmppc_hv_ops;
411 		else
412 			kvm_ops = kvmppc_pr_ops;
413 		if (!kvm_ops)
414 			goto err_out;
415 	} else	if (type == KVM_VM_PPC_HV) {
416 		if (!kvmppc_hv_ops)
417 			goto err_out;
418 		kvm_ops = kvmppc_hv_ops;
419 	} else if (type == KVM_VM_PPC_PR) {
420 		if (!kvmppc_pr_ops)
421 			goto err_out;
422 		kvm_ops = kvmppc_pr_ops;
423 	} else
424 		goto err_out;
425 
426 	if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
427 		return -ENOENT;
428 
429 	kvm->arch.kvm_ops = kvm_ops;
430 	return kvmppc_core_init_vm(kvm);
431 err_out:
432 	return -EINVAL;
433 }
434 
435 void kvm_arch_destroy_vm(struct kvm *kvm)
436 {
437 	unsigned int i;
438 	struct kvm_vcpu *vcpu;
439 
440 	kvm_for_each_vcpu(i, vcpu, kvm)
441 		kvm_arch_vcpu_free(vcpu);
442 
443 	mutex_lock(&kvm->lock);
444 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
445 		kvm->vcpus[i] = NULL;
446 
447 	atomic_set(&kvm->online_vcpus, 0);
448 
449 	kvmppc_core_destroy_vm(kvm);
450 
451 	mutex_unlock(&kvm->lock);
452 
453 	/* drop the module reference */
454 	module_put(kvm->arch.kvm_ops->owner);
455 }
456 
457 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
458 {
459 	int r;
460 	/* Assume we're using HV mode when the HV module is loaded */
461 	int hv_enabled = kvmppc_hv_ops ? 1 : 0;
462 
463 	if (kvm) {
464 		/*
465 		 * Hooray - we know which VM type we're running on. Depend on
466 		 * that rather than the guess above.
467 		 */
468 		hv_enabled = is_kvmppc_hv_enabled(kvm);
469 	}
470 
471 	switch (ext) {
472 #ifdef CONFIG_BOOKE
473 	case KVM_CAP_PPC_BOOKE_SREGS:
474 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
475 	case KVM_CAP_PPC_EPR:
476 #else
477 	case KVM_CAP_PPC_SEGSTATE:
478 	case KVM_CAP_PPC_HIOR:
479 	case KVM_CAP_PPC_PAPR:
480 #endif
481 	case KVM_CAP_PPC_UNSET_IRQ:
482 	case KVM_CAP_PPC_IRQ_LEVEL:
483 	case KVM_CAP_ENABLE_CAP:
484 	case KVM_CAP_ENABLE_CAP_VM:
485 	case KVM_CAP_ONE_REG:
486 	case KVM_CAP_IOEVENTFD:
487 	case KVM_CAP_DEVICE_CTRL:
488 		r = 1;
489 		break;
490 	case KVM_CAP_PPC_PAIRED_SINGLES:
491 	case KVM_CAP_PPC_OSI:
492 	case KVM_CAP_PPC_GET_PVINFO:
493 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
494 	case KVM_CAP_SW_TLB:
495 #endif
496 		/* We support this only for PR */
497 		r = !hv_enabled;
498 		break;
499 #ifdef CONFIG_KVM_MMIO
500 	case KVM_CAP_COALESCED_MMIO:
501 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
502 		break;
503 #endif
504 #ifdef CONFIG_KVM_MPIC
505 	case KVM_CAP_IRQ_MPIC:
506 		r = 1;
507 		break;
508 #endif
509 
510 #ifdef CONFIG_PPC_BOOK3S_64
511 	case KVM_CAP_SPAPR_TCE:
512 	case KVM_CAP_PPC_ALLOC_HTAB:
513 	case KVM_CAP_PPC_RTAS:
514 	case KVM_CAP_PPC_FIXUP_HCALL:
515 	case KVM_CAP_PPC_ENABLE_HCALL:
516 #ifdef CONFIG_KVM_XICS
517 	case KVM_CAP_IRQ_XICS:
518 #endif
519 		r = 1;
520 		break;
521 #endif /* CONFIG_PPC_BOOK3S_64 */
522 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
523 	case KVM_CAP_PPC_SMT:
524 		if (hv_enabled)
525 			r = threads_per_subcore;
526 		else
527 			r = 0;
528 		break;
529 	case KVM_CAP_PPC_RMA:
530 		r = 0;
531 		break;
532 	case KVM_CAP_PPC_HWRNG:
533 		r = kvmppc_hwrng_present();
534 		break;
535 #endif
536 	case KVM_CAP_SYNC_MMU:
537 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
538 		r = hv_enabled;
539 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
540 		r = 1;
541 #else
542 		r = 0;
543 #endif
544 		break;
545 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
546 	case KVM_CAP_PPC_HTAB_FD:
547 		r = hv_enabled;
548 		break;
549 #endif
550 	case KVM_CAP_NR_VCPUS:
551 		/*
552 		 * Recommending a number of CPUs is somewhat arbitrary; we
553 		 * return the number of present CPUs for -HV (since a host
554 		 * will have secondary threads "offline"), and for other KVM
555 		 * implementations just count online CPUs.
556 		 */
557 		if (hv_enabled)
558 			r = num_present_cpus();
559 		else
560 			r = num_online_cpus();
561 		break;
562 	case KVM_CAP_MAX_VCPUS:
563 		r = KVM_MAX_VCPUS;
564 		break;
565 #ifdef CONFIG_PPC_BOOK3S_64
566 	case KVM_CAP_PPC_GET_SMMU_INFO:
567 		r = 1;
568 		break;
569 #endif
570 	default:
571 		r = 0;
572 		break;
573 	}
574 	return r;
575 
576 }
577 
578 long kvm_arch_dev_ioctl(struct file *filp,
579                         unsigned int ioctl, unsigned long arg)
580 {
581 	return -EINVAL;
582 }
583 
584 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
585 			   struct kvm_memory_slot *dont)
586 {
587 	kvmppc_core_free_memslot(kvm, free, dont);
588 }
589 
590 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
591 			    unsigned long npages)
592 {
593 	return kvmppc_core_create_memslot(kvm, slot, npages);
594 }
595 
596 int kvm_arch_prepare_memory_region(struct kvm *kvm,
597 				   struct kvm_memory_slot *memslot,
598 				   const struct kvm_userspace_memory_region *mem,
599 				   enum kvm_mr_change change)
600 {
601 	return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
602 }
603 
604 void kvm_arch_commit_memory_region(struct kvm *kvm,
605 				   const struct kvm_userspace_memory_region *mem,
606 				   const struct kvm_memory_slot *old,
607 				   const struct kvm_memory_slot *new,
608 				   enum kvm_mr_change change)
609 {
610 	kvmppc_core_commit_memory_region(kvm, mem, old, new);
611 }
612 
613 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
614 				   struct kvm_memory_slot *slot)
615 {
616 	kvmppc_core_flush_memslot(kvm, slot);
617 }
618 
619 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
620 {
621 	struct kvm_vcpu *vcpu;
622 	vcpu = kvmppc_core_vcpu_create(kvm, id);
623 	if (!IS_ERR(vcpu)) {
624 		vcpu->arch.wqp = &vcpu->wq;
625 		kvmppc_create_vcpu_debugfs(vcpu, id);
626 	}
627 	return vcpu;
628 }
629 
630 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
631 {
632 }
633 
634 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
635 {
636 	/* Make sure we're not using the vcpu anymore */
637 	hrtimer_cancel(&vcpu->arch.dec_timer);
638 
639 	kvmppc_remove_vcpu_debugfs(vcpu);
640 
641 	switch (vcpu->arch.irq_type) {
642 	case KVMPPC_IRQ_MPIC:
643 		kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
644 		break;
645 	case KVMPPC_IRQ_XICS:
646 		kvmppc_xics_free_icp(vcpu);
647 		break;
648 	}
649 
650 	kvmppc_core_vcpu_free(vcpu);
651 }
652 
653 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
654 {
655 	kvm_arch_vcpu_free(vcpu);
656 }
657 
658 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
659 {
660 	return kvmppc_core_pending_dec(vcpu);
661 }
662 
663 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
664 {
665 	struct kvm_vcpu *vcpu;
666 
667 	vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
668 	kvmppc_decrementer_func(vcpu);
669 
670 	return HRTIMER_NORESTART;
671 }
672 
673 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
674 {
675 	int ret;
676 
677 	hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
678 	vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
679 	vcpu->arch.dec_expires = ~(u64)0;
680 
681 #ifdef CONFIG_KVM_EXIT_TIMING
682 	mutex_init(&vcpu->arch.exit_timing_lock);
683 #endif
684 	ret = kvmppc_subarch_vcpu_init(vcpu);
685 	return ret;
686 }
687 
688 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
689 {
690 	kvmppc_mmu_destroy(vcpu);
691 	kvmppc_subarch_vcpu_uninit(vcpu);
692 }
693 
694 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
695 {
696 #ifdef CONFIG_BOOKE
697 	/*
698 	 * vrsave (formerly usprg0) isn't used by Linux, but may
699 	 * be used by the guest.
700 	 *
701 	 * On non-booke this is associated with Altivec and
702 	 * is handled by code in book3s.c.
703 	 */
704 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
705 #endif
706 	kvmppc_core_vcpu_load(vcpu, cpu);
707 }
708 
709 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
710 {
711 	kvmppc_core_vcpu_put(vcpu);
712 #ifdef CONFIG_BOOKE
713 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
714 #endif
715 }
716 
717 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
718                                       struct kvm_run *run)
719 {
720 	u64 uninitialized_var(gpr);
721 
722 	if (run->mmio.len > sizeof(gpr)) {
723 		printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
724 		return;
725 	}
726 
727 	if (!vcpu->arch.mmio_host_swabbed) {
728 		switch (run->mmio.len) {
729 		case 8: gpr = *(u64 *)run->mmio.data; break;
730 		case 4: gpr = *(u32 *)run->mmio.data; break;
731 		case 2: gpr = *(u16 *)run->mmio.data; break;
732 		case 1: gpr = *(u8 *)run->mmio.data; break;
733 		}
734 	} else {
735 		switch (run->mmio.len) {
736 		case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
737 		case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
738 		case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
739 		case 1: gpr = *(u8 *)run->mmio.data; break;
740 		}
741 	}
742 
743 	if (vcpu->arch.mmio_sign_extend) {
744 		switch (run->mmio.len) {
745 #ifdef CONFIG_PPC64
746 		case 4:
747 			gpr = (s64)(s32)gpr;
748 			break;
749 #endif
750 		case 2:
751 			gpr = (s64)(s16)gpr;
752 			break;
753 		case 1:
754 			gpr = (s64)(s8)gpr;
755 			break;
756 		}
757 	}
758 
759 	kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
760 
761 	switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
762 	case KVM_MMIO_REG_GPR:
763 		kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
764 		break;
765 	case KVM_MMIO_REG_FPR:
766 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
767 		break;
768 #ifdef CONFIG_PPC_BOOK3S
769 	case KVM_MMIO_REG_QPR:
770 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
771 		break;
772 	case KVM_MMIO_REG_FQPR:
773 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
774 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
775 		break;
776 #endif
777 	default:
778 		BUG();
779 	}
780 }
781 
782 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
783 		       unsigned int rt, unsigned int bytes,
784 		       int is_default_endian)
785 {
786 	int idx, ret;
787 	bool host_swabbed;
788 
789 	/* Pity C doesn't have a logical XOR operator */
790 	if (kvmppc_need_byteswap(vcpu)) {
791 		host_swabbed = is_default_endian;
792 	} else {
793 		host_swabbed = !is_default_endian;
794 	}
795 
796 	if (bytes > sizeof(run->mmio.data)) {
797 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
798 		       run->mmio.len);
799 	}
800 
801 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
802 	run->mmio.len = bytes;
803 	run->mmio.is_write = 0;
804 
805 	vcpu->arch.io_gpr = rt;
806 	vcpu->arch.mmio_host_swabbed = host_swabbed;
807 	vcpu->mmio_needed = 1;
808 	vcpu->mmio_is_write = 0;
809 	vcpu->arch.mmio_sign_extend = 0;
810 
811 	idx = srcu_read_lock(&vcpu->kvm->srcu);
812 
813 	ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
814 			      bytes, &run->mmio.data);
815 
816 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
817 
818 	if (!ret) {
819 		kvmppc_complete_mmio_load(vcpu, run);
820 		vcpu->mmio_needed = 0;
821 		return EMULATE_DONE;
822 	}
823 
824 	return EMULATE_DO_MMIO;
825 }
826 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
827 
828 /* Same as above, but sign extends */
829 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
830 			unsigned int rt, unsigned int bytes,
831 			int is_default_endian)
832 {
833 	int r;
834 
835 	vcpu->arch.mmio_sign_extend = 1;
836 	r = kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian);
837 
838 	return r;
839 }
840 
841 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
842 			u64 val, unsigned int bytes, int is_default_endian)
843 {
844 	void *data = run->mmio.data;
845 	int idx, ret;
846 	bool host_swabbed;
847 
848 	/* Pity C doesn't have a logical XOR operator */
849 	if (kvmppc_need_byteswap(vcpu)) {
850 		host_swabbed = is_default_endian;
851 	} else {
852 		host_swabbed = !is_default_endian;
853 	}
854 
855 	if (bytes > sizeof(run->mmio.data)) {
856 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
857 		       run->mmio.len);
858 	}
859 
860 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
861 	run->mmio.len = bytes;
862 	run->mmio.is_write = 1;
863 	vcpu->mmio_needed = 1;
864 	vcpu->mmio_is_write = 1;
865 
866 	/* Store the value at the lowest bytes in 'data'. */
867 	if (!host_swabbed) {
868 		switch (bytes) {
869 		case 8: *(u64 *)data = val; break;
870 		case 4: *(u32 *)data = val; break;
871 		case 2: *(u16 *)data = val; break;
872 		case 1: *(u8  *)data = val; break;
873 		}
874 	} else {
875 		switch (bytes) {
876 		case 8: *(u64 *)data = swab64(val); break;
877 		case 4: *(u32 *)data = swab32(val); break;
878 		case 2: *(u16 *)data = swab16(val); break;
879 		case 1: *(u8  *)data = val; break;
880 		}
881 	}
882 
883 	idx = srcu_read_lock(&vcpu->kvm->srcu);
884 
885 	ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
886 			       bytes, &run->mmio.data);
887 
888 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
889 
890 	if (!ret) {
891 		vcpu->mmio_needed = 0;
892 		return EMULATE_DONE;
893 	}
894 
895 	return EMULATE_DO_MMIO;
896 }
897 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
898 
899 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
900 {
901 	int r = 0;
902 	union kvmppc_one_reg val;
903 	int size;
904 
905 	size = one_reg_size(reg->id);
906 	if (size > sizeof(val))
907 		return -EINVAL;
908 
909 	r = kvmppc_get_one_reg(vcpu, reg->id, &val);
910 	if (r == -EINVAL) {
911 		r = 0;
912 		switch (reg->id) {
913 #ifdef CONFIG_ALTIVEC
914 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
915 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
916 				r = -ENXIO;
917 				break;
918 			}
919 			vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
920 			break;
921 		case KVM_REG_PPC_VSCR:
922 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
923 				r = -ENXIO;
924 				break;
925 			}
926 			vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
927 			break;
928 		case KVM_REG_PPC_VRSAVE:
929 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
930 				r = -ENXIO;
931 				break;
932 			}
933 			vcpu->arch.vrsave = set_reg_val(reg->id, val);
934 			break;
935 #endif /* CONFIG_ALTIVEC */
936 		default:
937 			r = -EINVAL;
938 			break;
939 		}
940 	}
941 
942 	if (r)
943 		return r;
944 
945 	if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
946 		r = -EFAULT;
947 
948 	return r;
949 }
950 
951 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
952 {
953 	int r;
954 	union kvmppc_one_reg val;
955 	int size;
956 
957 	size = one_reg_size(reg->id);
958 	if (size > sizeof(val))
959 		return -EINVAL;
960 
961 	if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
962 		return -EFAULT;
963 
964 	r = kvmppc_set_one_reg(vcpu, reg->id, &val);
965 	if (r == -EINVAL) {
966 		r = 0;
967 		switch (reg->id) {
968 #ifdef CONFIG_ALTIVEC
969 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
970 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
971 				r = -ENXIO;
972 				break;
973 			}
974 			val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
975 			break;
976 		case KVM_REG_PPC_VSCR:
977 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
978 				r = -ENXIO;
979 				break;
980 			}
981 			val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
982 			break;
983 		case KVM_REG_PPC_VRSAVE:
984 			val = get_reg_val(reg->id, vcpu->arch.vrsave);
985 			break;
986 #endif /* CONFIG_ALTIVEC */
987 		default:
988 			r = -EINVAL;
989 			break;
990 		}
991 	}
992 
993 	return r;
994 }
995 
996 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
997 {
998 	int r;
999 	sigset_t sigsaved;
1000 
1001 	if (vcpu->sigset_active)
1002 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1003 
1004 	if (vcpu->mmio_needed) {
1005 		if (!vcpu->mmio_is_write)
1006 			kvmppc_complete_mmio_load(vcpu, run);
1007 		vcpu->mmio_needed = 0;
1008 	} else if (vcpu->arch.osi_needed) {
1009 		u64 *gprs = run->osi.gprs;
1010 		int i;
1011 
1012 		for (i = 0; i < 32; i++)
1013 			kvmppc_set_gpr(vcpu, i, gprs[i]);
1014 		vcpu->arch.osi_needed = 0;
1015 	} else if (vcpu->arch.hcall_needed) {
1016 		int i;
1017 
1018 		kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1019 		for (i = 0; i < 9; ++i)
1020 			kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1021 		vcpu->arch.hcall_needed = 0;
1022 #ifdef CONFIG_BOOKE
1023 	} else if (vcpu->arch.epr_needed) {
1024 		kvmppc_set_epr(vcpu, run->epr.epr);
1025 		vcpu->arch.epr_needed = 0;
1026 #endif
1027 	}
1028 
1029 	r = kvmppc_vcpu_run(run, vcpu);
1030 
1031 	if (vcpu->sigset_active)
1032 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1033 
1034 	return r;
1035 }
1036 
1037 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1038 {
1039 	if (irq->irq == KVM_INTERRUPT_UNSET) {
1040 		kvmppc_core_dequeue_external(vcpu);
1041 		return 0;
1042 	}
1043 
1044 	kvmppc_core_queue_external(vcpu, irq);
1045 
1046 	kvm_vcpu_kick(vcpu);
1047 
1048 	return 0;
1049 }
1050 
1051 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1052 				     struct kvm_enable_cap *cap)
1053 {
1054 	int r;
1055 
1056 	if (cap->flags)
1057 		return -EINVAL;
1058 
1059 	switch (cap->cap) {
1060 	case KVM_CAP_PPC_OSI:
1061 		r = 0;
1062 		vcpu->arch.osi_enabled = true;
1063 		break;
1064 	case KVM_CAP_PPC_PAPR:
1065 		r = 0;
1066 		vcpu->arch.papr_enabled = true;
1067 		break;
1068 	case KVM_CAP_PPC_EPR:
1069 		r = 0;
1070 		if (cap->args[0])
1071 			vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1072 		else
1073 			vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1074 		break;
1075 #ifdef CONFIG_BOOKE
1076 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
1077 		r = 0;
1078 		vcpu->arch.watchdog_enabled = true;
1079 		break;
1080 #endif
1081 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1082 	case KVM_CAP_SW_TLB: {
1083 		struct kvm_config_tlb cfg;
1084 		void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1085 
1086 		r = -EFAULT;
1087 		if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1088 			break;
1089 
1090 		r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1091 		break;
1092 	}
1093 #endif
1094 #ifdef CONFIG_KVM_MPIC
1095 	case KVM_CAP_IRQ_MPIC: {
1096 		struct fd f;
1097 		struct kvm_device *dev;
1098 
1099 		r = -EBADF;
1100 		f = fdget(cap->args[0]);
1101 		if (!f.file)
1102 			break;
1103 
1104 		r = -EPERM;
1105 		dev = kvm_device_from_filp(f.file);
1106 		if (dev)
1107 			r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1108 
1109 		fdput(f);
1110 		break;
1111 	}
1112 #endif
1113 #ifdef CONFIG_KVM_XICS
1114 	case KVM_CAP_IRQ_XICS: {
1115 		struct fd f;
1116 		struct kvm_device *dev;
1117 
1118 		r = -EBADF;
1119 		f = fdget(cap->args[0]);
1120 		if (!f.file)
1121 			break;
1122 
1123 		r = -EPERM;
1124 		dev = kvm_device_from_filp(f.file);
1125 		if (dev)
1126 			r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1127 
1128 		fdput(f);
1129 		break;
1130 	}
1131 #endif /* CONFIG_KVM_XICS */
1132 	default:
1133 		r = -EINVAL;
1134 		break;
1135 	}
1136 
1137 	if (!r)
1138 		r = kvmppc_sanity_check(vcpu);
1139 
1140 	return r;
1141 }
1142 
1143 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1144                                     struct kvm_mp_state *mp_state)
1145 {
1146 	return -EINVAL;
1147 }
1148 
1149 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1150                                     struct kvm_mp_state *mp_state)
1151 {
1152 	return -EINVAL;
1153 }
1154 
1155 long kvm_arch_vcpu_ioctl(struct file *filp,
1156                          unsigned int ioctl, unsigned long arg)
1157 {
1158 	struct kvm_vcpu *vcpu = filp->private_data;
1159 	void __user *argp = (void __user *)arg;
1160 	long r;
1161 
1162 	switch (ioctl) {
1163 	case KVM_INTERRUPT: {
1164 		struct kvm_interrupt irq;
1165 		r = -EFAULT;
1166 		if (copy_from_user(&irq, argp, sizeof(irq)))
1167 			goto out;
1168 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1169 		goto out;
1170 	}
1171 
1172 	case KVM_ENABLE_CAP:
1173 	{
1174 		struct kvm_enable_cap cap;
1175 		r = -EFAULT;
1176 		if (copy_from_user(&cap, argp, sizeof(cap)))
1177 			goto out;
1178 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1179 		break;
1180 	}
1181 
1182 	case KVM_SET_ONE_REG:
1183 	case KVM_GET_ONE_REG:
1184 	{
1185 		struct kvm_one_reg reg;
1186 		r = -EFAULT;
1187 		if (copy_from_user(&reg, argp, sizeof(reg)))
1188 			goto out;
1189 		if (ioctl == KVM_SET_ONE_REG)
1190 			r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
1191 		else
1192 			r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
1193 		break;
1194 	}
1195 
1196 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1197 	case KVM_DIRTY_TLB: {
1198 		struct kvm_dirty_tlb dirty;
1199 		r = -EFAULT;
1200 		if (copy_from_user(&dirty, argp, sizeof(dirty)))
1201 			goto out;
1202 		r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1203 		break;
1204 	}
1205 #endif
1206 	default:
1207 		r = -EINVAL;
1208 	}
1209 
1210 out:
1211 	return r;
1212 }
1213 
1214 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1215 {
1216 	return VM_FAULT_SIGBUS;
1217 }
1218 
1219 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1220 {
1221 	u32 inst_nop = 0x60000000;
1222 #ifdef CONFIG_KVM_BOOKE_HV
1223 	u32 inst_sc1 = 0x44000022;
1224 	pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1225 	pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1226 	pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1227 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1228 #else
1229 	u32 inst_lis = 0x3c000000;
1230 	u32 inst_ori = 0x60000000;
1231 	u32 inst_sc = 0x44000002;
1232 	u32 inst_imm_mask = 0xffff;
1233 
1234 	/*
1235 	 * The hypercall to get into KVM from within guest context is as
1236 	 * follows:
1237 	 *
1238 	 *    lis r0, r0, KVM_SC_MAGIC_R0@h
1239 	 *    ori r0, KVM_SC_MAGIC_R0@l
1240 	 *    sc
1241 	 *    nop
1242 	 */
1243 	pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1244 	pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1245 	pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1246 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1247 #endif
1248 
1249 	pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1250 
1251 	return 0;
1252 }
1253 
1254 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1255 			  bool line_status)
1256 {
1257 	if (!irqchip_in_kernel(kvm))
1258 		return -ENXIO;
1259 
1260 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1261 					irq_event->irq, irq_event->level,
1262 					line_status);
1263 	return 0;
1264 }
1265 
1266 
1267 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1268 				   struct kvm_enable_cap *cap)
1269 {
1270 	int r;
1271 
1272 	if (cap->flags)
1273 		return -EINVAL;
1274 
1275 	switch (cap->cap) {
1276 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1277 	case KVM_CAP_PPC_ENABLE_HCALL: {
1278 		unsigned long hcall = cap->args[0];
1279 
1280 		r = -EINVAL;
1281 		if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1282 		    cap->args[1] > 1)
1283 			break;
1284 		if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1285 			break;
1286 		if (cap->args[1])
1287 			set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1288 		else
1289 			clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1290 		r = 0;
1291 		break;
1292 	}
1293 #endif
1294 	default:
1295 		r = -EINVAL;
1296 		break;
1297 	}
1298 
1299 	return r;
1300 }
1301 
1302 long kvm_arch_vm_ioctl(struct file *filp,
1303                        unsigned int ioctl, unsigned long arg)
1304 {
1305 	struct kvm *kvm __maybe_unused = filp->private_data;
1306 	void __user *argp = (void __user *)arg;
1307 	long r;
1308 
1309 	switch (ioctl) {
1310 	case KVM_PPC_GET_PVINFO: {
1311 		struct kvm_ppc_pvinfo pvinfo;
1312 		memset(&pvinfo, 0, sizeof(pvinfo));
1313 		r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1314 		if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1315 			r = -EFAULT;
1316 			goto out;
1317 		}
1318 
1319 		break;
1320 	}
1321 	case KVM_ENABLE_CAP:
1322 	{
1323 		struct kvm_enable_cap cap;
1324 		r = -EFAULT;
1325 		if (copy_from_user(&cap, argp, sizeof(cap)))
1326 			goto out;
1327 		r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1328 		break;
1329 	}
1330 #ifdef CONFIG_PPC_BOOK3S_64
1331 	case KVM_CREATE_SPAPR_TCE: {
1332 		struct kvm_create_spapr_tce create_tce;
1333 
1334 		r = -EFAULT;
1335 		if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1336 			goto out;
1337 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
1338 		goto out;
1339 	}
1340 	case KVM_PPC_GET_SMMU_INFO: {
1341 		struct kvm_ppc_smmu_info info;
1342 		struct kvm *kvm = filp->private_data;
1343 
1344 		memset(&info, 0, sizeof(info));
1345 		r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1346 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1347 			r = -EFAULT;
1348 		break;
1349 	}
1350 	case KVM_PPC_RTAS_DEFINE_TOKEN: {
1351 		struct kvm *kvm = filp->private_data;
1352 
1353 		r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1354 		break;
1355 	}
1356 	default: {
1357 		struct kvm *kvm = filp->private_data;
1358 		r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1359 	}
1360 #else /* CONFIG_PPC_BOOK3S_64 */
1361 	default:
1362 		r = -ENOTTY;
1363 #endif
1364 	}
1365 out:
1366 	return r;
1367 }
1368 
1369 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1370 static unsigned long nr_lpids;
1371 
1372 long kvmppc_alloc_lpid(void)
1373 {
1374 	long lpid;
1375 
1376 	do {
1377 		lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1378 		if (lpid >= nr_lpids) {
1379 			pr_err("%s: No LPIDs free\n", __func__);
1380 			return -ENOMEM;
1381 		}
1382 	} while (test_and_set_bit(lpid, lpid_inuse));
1383 
1384 	return lpid;
1385 }
1386 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1387 
1388 void kvmppc_claim_lpid(long lpid)
1389 {
1390 	set_bit(lpid, lpid_inuse);
1391 }
1392 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1393 
1394 void kvmppc_free_lpid(long lpid)
1395 {
1396 	clear_bit(lpid, lpid_inuse);
1397 }
1398 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1399 
1400 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1401 {
1402 	nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1403 	memset(lpid_inuse, 0, sizeof(lpid_inuse));
1404 }
1405 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1406 
1407 int kvm_arch_init(void *opaque)
1408 {
1409 	return 0;
1410 }
1411 
1412 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);
1413