xref: /openbmc/linux/arch/powerpc/kvm/powerpc.c (revision 80483c3a)
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 <asm/iommu.h>
37 #include "timing.h"
38 #include "irq.h"
39 #include "../mm/mmu_decl.h"
40 
41 #define CREATE_TRACE_POINTS
42 #include "trace.h"
43 
44 struct kvmppc_ops *kvmppc_hv_ops;
45 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
46 struct kvmppc_ops *kvmppc_pr_ops;
47 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
48 
49 
50 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
51 {
52 	return !!(v->arch.pending_exceptions) ||
53 	       v->requests;
54 }
55 
56 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
57 {
58 	return 1;
59 }
60 
61 /*
62  * Common checks before entering the guest world.  Call with interrupts
63  * disabled.
64  *
65  * returns:
66  *
67  * == 1 if we're ready to go into guest state
68  * <= 0 if we need to go back to the host with return value
69  */
70 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
71 {
72 	int r;
73 
74 	WARN_ON(irqs_disabled());
75 	hard_irq_disable();
76 
77 	while (true) {
78 		if (need_resched()) {
79 			local_irq_enable();
80 			cond_resched();
81 			hard_irq_disable();
82 			continue;
83 		}
84 
85 		if (signal_pending(current)) {
86 			kvmppc_account_exit(vcpu, SIGNAL_EXITS);
87 			vcpu->run->exit_reason = KVM_EXIT_INTR;
88 			r = -EINTR;
89 			break;
90 		}
91 
92 		vcpu->mode = IN_GUEST_MODE;
93 
94 		/*
95 		 * Reading vcpu->requests must happen after setting vcpu->mode,
96 		 * so we don't miss a request because the requester sees
97 		 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
98 		 * before next entering the guest (and thus doesn't IPI).
99 		 * This also orders the write to mode from any reads
100 		 * to the page tables done while the VCPU is running.
101 		 * Please see the comment in kvm_flush_remote_tlbs.
102 		 */
103 		smp_mb();
104 
105 		if (vcpu->requests) {
106 			/* Make sure we process requests preemptable */
107 			local_irq_enable();
108 			trace_kvm_check_requests(vcpu);
109 			r = kvmppc_core_check_requests(vcpu);
110 			hard_irq_disable();
111 			if (r > 0)
112 				continue;
113 			break;
114 		}
115 
116 		if (kvmppc_core_prepare_to_enter(vcpu)) {
117 			/* interrupts got enabled in between, so we
118 			   are back at square 1 */
119 			continue;
120 		}
121 
122 		guest_enter_irqoff();
123 		return 1;
124 	}
125 
126 	/* return to host */
127 	local_irq_enable();
128 	return r;
129 }
130 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
131 
132 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
133 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
134 {
135 	struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
136 	int i;
137 
138 	shared->sprg0 = swab64(shared->sprg0);
139 	shared->sprg1 = swab64(shared->sprg1);
140 	shared->sprg2 = swab64(shared->sprg2);
141 	shared->sprg3 = swab64(shared->sprg3);
142 	shared->srr0 = swab64(shared->srr0);
143 	shared->srr1 = swab64(shared->srr1);
144 	shared->dar = swab64(shared->dar);
145 	shared->msr = swab64(shared->msr);
146 	shared->dsisr = swab32(shared->dsisr);
147 	shared->int_pending = swab32(shared->int_pending);
148 	for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
149 		shared->sr[i] = swab32(shared->sr[i]);
150 }
151 #endif
152 
153 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
154 {
155 	int nr = kvmppc_get_gpr(vcpu, 11);
156 	int r;
157 	unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
158 	unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
159 	unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
160 	unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
161 	unsigned long r2 = 0;
162 
163 	if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
164 		/* 32 bit mode */
165 		param1 &= 0xffffffff;
166 		param2 &= 0xffffffff;
167 		param3 &= 0xffffffff;
168 		param4 &= 0xffffffff;
169 	}
170 
171 	switch (nr) {
172 	case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
173 	{
174 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
175 		/* Book3S can be little endian, find it out here */
176 		int shared_big_endian = true;
177 		if (vcpu->arch.intr_msr & MSR_LE)
178 			shared_big_endian = false;
179 		if (shared_big_endian != vcpu->arch.shared_big_endian)
180 			kvmppc_swab_shared(vcpu);
181 		vcpu->arch.shared_big_endian = shared_big_endian;
182 #endif
183 
184 		if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
185 			/*
186 			 * Older versions of the Linux magic page code had
187 			 * a bug where they would map their trampoline code
188 			 * NX. If that's the case, remove !PR NX capability.
189 			 */
190 			vcpu->arch.disable_kernel_nx = true;
191 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
192 		}
193 
194 		vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
195 		vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
196 
197 #ifdef CONFIG_PPC_64K_PAGES
198 		/*
199 		 * Make sure our 4k magic page is in the same window of a 64k
200 		 * page within the guest and within the host's page.
201 		 */
202 		if ((vcpu->arch.magic_page_pa & 0xf000) !=
203 		    ((ulong)vcpu->arch.shared & 0xf000)) {
204 			void *old_shared = vcpu->arch.shared;
205 			ulong shared = (ulong)vcpu->arch.shared;
206 			void *new_shared;
207 
208 			shared &= PAGE_MASK;
209 			shared |= vcpu->arch.magic_page_pa & 0xf000;
210 			new_shared = (void*)shared;
211 			memcpy(new_shared, old_shared, 0x1000);
212 			vcpu->arch.shared = new_shared;
213 		}
214 #endif
215 
216 		r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
217 
218 		r = EV_SUCCESS;
219 		break;
220 	}
221 	case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
222 		r = EV_SUCCESS;
223 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
224 		r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
225 #endif
226 
227 		/* Second return value is in r4 */
228 		break;
229 	case EV_HCALL_TOKEN(EV_IDLE):
230 		r = EV_SUCCESS;
231 		kvm_vcpu_block(vcpu);
232 		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
233 		break;
234 	default:
235 		r = EV_UNIMPLEMENTED;
236 		break;
237 	}
238 
239 	kvmppc_set_gpr(vcpu, 4, r2);
240 
241 	return r;
242 }
243 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
244 
245 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
246 {
247 	int r = false;
248 
249 	/* We have to know what CPU to virtualize */
250 	if (!vcpu->arch.pvr)
251 		goto out;
252 
253 	/* PAPR only works with book3s_64 */
254 	if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
255 		goto out;
256 
257 	/* HV KVM can only do PAPR mode for now */
258 	if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
259 		goto out;
260 
261 #ifdef CONFIG_KVM_BOOKE_HV
262 	if (!cpu_has_feature(CPU_FTR_EMB_HV))
263 		goto out;
264 #endif
265 
266 	r = true;
267 
268 out:
269 	vcpu->arch.sane = r;
270 	return r ? 0 : -EINVAL;
271 }
272 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
273 
274 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
275 {
276 	enum emulation_result er;
277 	int r;
278 
279 	er = kvmppc_emulate_loadstore(vcpu);
280 	switch (er) {
281 	case EMULATE_DONE:
282 		/* Future optimization: only reload non-volatiles if they were
283 		 * actually modified. */
284 		r = RESUME_GUEST_NV;
285 		break;
286 	case EMULATE_AGAIN:
287 		r = RESUME_GUEST;
288 		break;
289 	case EMULATE_DO_MMIO:
290 		run->exit_reason = KVM_EXIT_MMIO;
291 		/* We must reload nonvolatiles because "update" load/store
292 		 * instructions modify register state. */
293 		/* Future optimization: only reload non-volatiles if they were
294 		 * actually modified. */
295 		r = RESUME_HOST_NV;
296 		break;
297 	case EMULATE_FAIL:
298 	{
299 		u32 last_inst;
300 
301 		kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
302 		/* XXX Deliver Program interrupt to guest. */
303 		pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
304 		r = RESUME_HOST;
305 		break;
306 	}
307 	default:
308 		WARN_ON(1);
309 		r = RESUME_GUEST;
310 	}
311 
312 	return r;
313 }
314 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
315 
316 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
317 	      bool data)
318 {
319 	ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
320 	struct kvmppc_pte pte;
321 	int r;
322 
323 	vcpu->stat.st++;
324 
325 	r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
326 			 XLATE_WRITE, &pte);
327 	if (r < 0)
328 		return r;
329 
330 	*eaddr = pte.raddr;
331 
332 	if (!pte.may_write)
333 		return -EPERM;
334 
335 	/* Magic page override */
336 	if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
337 	    ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
338 	    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
339 		void *magic = vcpu->arch.shared;
340 		magic += pte.eaddr & 0xfff;
341 		memcpy(magic, ptr, size);
342 		return EMULATE_DONE;
343 	}
344 
345 	if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
346 		return EMULATE_DO_MMIO;
347 
348 	return EMULATE_DONE;
349 }
350 EXPORT_SYMBOL_GPL(kvmppc_st);
351 
352 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
353 		      bool data)
354 {
355 	ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
356 	struct kvmppc_pte pte;
357 	int rc;
358 
359 	vcpu->stat.ld++;
360 
361 	rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
362 			  XLATE_READ, &pte);
363 	if (rc)
364 		return rc;
365 
366 	*eaddr = pte.raddr;
367 
368 	if (!pte.may_read)
369 		return -EPERM;
370 
371 	if (!data && !pte.may_execute)
372 		return -ENOEXEC;
373 
374 	/* Magic page override */
375 	if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
376 	    ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
377 	    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
378 		void *magic = vcpu->arch.shared;
379 		magic += pte.eaddr & 0xfff;
380 		memcpy(ptr, magic, size);
381 		return EMULATE_DONE;
382 	}
383 
384 	if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
385 		return EMULATE_DO_MMIO;
386 
387 	return EMULATE_DONE;
388 }
389 EXPORT_SYMBOL_GPL(kvmppc_ld);
390 
391 int kvm_arch_hardware_enable(void)
392 {
393 	return 0;
394 }
395 
396 int kvm_arch_hardware_setup(void)
397 {
398 	return 0;
399 }
400 
401 void kvm_arch_check_processor_compat(void *rtn)
402 {
403 	*(int *)rtn = kvmppc_core_check_processor_compat();
404 }
405 
406 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
407 {
408 	struct kvmppc_ops *kvm_ops = NULL;
409 	/*
410 	 * if we have both HV and PR enabled, default is HV
411 	 */
412 	if (type == 0) {
413 		if (kvmppc_hv_ops)
414 			kvm_ops = kvmppc_hv_ops;
415 		else
416 			kvm_ops = kvmppc_pr_ops;
417 		if (!kvm_ops)
418 			goto err_out;
419 	} else	if (type == KVM_VM_PPC_HV) {
420 		if (!kvmppc_hv_ops)
421 			goto err_out;
422 		kvm_ops = kvmppc_hv_ops;
423 	} else if (type == KVM_VM_PPC_PR) {
424 		if (!kvmppc_pr_ops)
425 			goto err_out;
426 		kvm_ops = kvmppc_pr_ops;
427 	} else
428 		goto err_out;
429 
430 	if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
431 		return -ENOENT;
432 
433 	kvm->arch.kvm_ops = kvm_ops;
434 	return kvmppc_core_init_vm(kvm);
435 err_out:
436 	return -EINVAL;
437 }
438 
439 void kvm_arch_destroy_vm(struct kvm *kvm)
440 {
441 	unsigned int i;
442 	struct kvm_vcpu *vcpu;
443 
444 #ifdef CONFIG_KVM_XICS
445 	/*
446 	 * We call kick_all_cpus_sync() to ensure that all
447 	 * CPUs have executed any pending IPIs before we
448 	 * continue and free VCPUs structures below.
449 	 */
450 	if (is_kvmppc_hv_enabled(kvm))
451 		kick_all_cpus_sync();
452 #endif
453 
454 	kvm_for_each_vcpu(i, vcpu, kvm)
455 		kvm_arch_vcpu_free(vcpu);
456 
457 	mutex_lock(&kvm->lock);
458 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
459 		kvm->vcpus[i] = NULL;
460 
461 	atomic_set(&kvm->online_vcpus, 0);
462 
463 	kvmppc_core_destroy_vm(kvm);
464 
465 	mutex_unlock(&kvm->lock);
466 
467 	/* drop the module reference */
468 	module_put(kvm->arch.kvm_ops->owner);
469 }
470 
471 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
472 {
473 	int r;
474 	/* Assume we're using HV mode when the HV module is loaded */
475 	int hv_enabled = kvmppc_hv_ops ? 1 : 0;
476 
477 	if (kvm) {
478 		/*
479 		 * Hooray - we know which VM type we're running on. Depend on
480 		 * that rather than the guess above.
481 		 */
482 		hv_enabled = is_kvmppc_hv_enabled(kvm);
483 	}
484 
485 	switch (ext) {
486 #ifdef CONFIG_BOOKE
487 	case KVM_CAP_PPC_BOOKE_SREGS:
488 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
489 	case KVM_CAP_PPC_EPR:
490 #else
491 	case KVM_CAP_PPC_SEGSTATE:
492 	case KVM_CAP_PPC_HIOR:
493 	case KVM_CAP_PPC_PAPR:
494 #endif
495 	case KVM_CAP_PPC_UNSET_IRQ:
496 	case KVM_CAP_PPC_IRQ_LEVEL:
497 	case KVM_CAP_ENABLE_CAP:
498 	case KVM_CAP_ENABLE_CAP_VM:
499 	case KVM_CAP_ONE_REG:
500 	case KVM_CAP_IOEVENTFD:
501 	case KVM_CAP_DEVICE_CTRL:
502 		r = 1;
503 		break;
504 	case KVM_CAP_PPC_PAIRED_SINGLES:
505 	case KVM_CAP_PPC_OSI:
506 	case KVM_CAP_PPC_GET_PVINFO:
507 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
508 	case KVM_CAP_SW_TLB:
509 #endif
510 		/* We support this only for PR */
511 		r = !hv_enabled;
512 		break;
513 #ifdef CONFIG_KVM_MMIO
514 	case KVM_CAP_COALESCED_MMIO:
515 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
516 		break;
517 #endif
518 #ifdef CONFIG_KVM_MPIC
519 	case KVM_CAP_IRQ_MPIC:
520 		r = 1;
521 		break;
522 #endif
523 
524 #ifdef CONFIG_PPC_BOOK3S_64
525 	case KVM_CAP_SPAPR_TCE:
526 	case KVM_CAP_SPAPR_TCE_64:
527 	case KVM_CAP_PPC_ALLOC_HTAB:
528 	case KVM_CAP_PPC_RTAS:
529 	case KVM_CAP_PPC_FIXUP_HCALL:
530 	case KVM_CAP_PPC_ENABLE_HCALL:
531 #ifdef CONFIG_KVM_XICS
532 	case KVM_CAP_IRQ_XICS:
533 #endif
534 		r = 1;
535 		break;
536 #endif /* CONFIG_PPC_BOOK3S_64 */
537 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
538 	case KVM_CAP_PPC_SMT:
539 		if (hv_enabled)
540 			r = threads_per_subcore;
541 		else
542 			r = 0;
543 		break;
544 	case KVM_CAP_PPC_RMA:
545 		r = 0;
546 		break;
547 	case KVM_CAP_PPC_HWRNG:
548 		r = kvmppc_hwrng_present();
549 		break;
550 #endif
551 	case KVM_CAP_SYNC_MMU:
552 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
553 		r = hv_enabled;
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_NR_MEMSLOTS:
578 		r = KVM_USER_MEM_SLOTS;
579 		break;
580 	case KVM_CAP_MAX_VCPUS:
581 		r = KVM_MAX_VCPUS;
582 		break;
583 #ifdef CONFIG_PPC_BOOK3S_64
584 	case KVM_CAP_PPC_GET_SMMU_INFO:
585 		r = 1;
586 		break;
587 	case KVM_CAP_SPAPR_MULTITCE:
588 		r = 1;
589 		break;
590 #endif
591 	case KVM_CAP_PPC_HTM:
592 		r = cpu_has_feature(CPU_FTR_TM_COMP) &&
593 		    is_kvmppc_hv_enabled(kvm);
594 		break;
595 	default:
596 		r = 0;
597 		break;
598 	}
599 	return r;
600 
601 }
602 
603 long kvm_arch_dev_ioctl(struct file *filp,
604                         unsigned int ioctl, unsigned long arg)
605 {
606 	return -EINVAL;
607 }
608 
609 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
610 			   struct kvm_memory_slot *dont)
611 {
612 	kvmppc_core_free_memslot(kvm, free, dont);
613 }
614 
615 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
616 			    unsigned long npages)
617 {
618 	return kvmppc_core_create_memslot(kvm, slot, npages);
619 }
620 
621 int kvm_arch_prepare_memory_region(struct kvm *kvm,
622 				   struct kvm_memory_slot *memslot,
623 				   const struct kvm_userspace_memory_region *mem,
624 				   enum kvm_mr_change change)
625 {
626 	return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
627 }
628 
629 void kvm_arch_commit_memory_region(struct kvm *kvm,
630 				   const struct kvm_userspace_memory_region *mem,
631 				   const struct kvm_memory_slot *old,
632 				   const struct kvm_memory_slot *new,
633 				   enum kvm_mr_change change)
634 {
635 	kvmppc_core_commit_memory_region(kvm, mem, old, new);
636 }
637 
638 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
639 				   struct kvm_memory_slot *slot)
640 {
641 	kvmppc_core_flush_memslot(kvm, slot);
642 }
643 
644 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
645 {
646 	struct kvm_vcpu *vcpu;
647 	vcpu = kvmppc_core_vcpu_create(kvm, id);
648 	if (!IS_ERR(vcpu)) {
649 		vcpu->arch.wqp = &vcpu->wq;
650 		kvmppc_create_vcpu_debugfs(vcpu, id);
651 	}
652 	return vcpu;
653 }
654 
655 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
656 {
657 }
658 
659 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
660 {
661 	/* Make sure we're not using the vcpu anymore */
662 	hrtimer_cancel(&vcpu->arch.dec_timer);
663 
664 	kvmppc_remove_vcpu_debugfs(vcpu);
665 
666 	switch (vcpu->arch.irq_type) {
667 	case KVMPPC_IRQ_MPIC:
668 		kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
669 		break;
670 	case KVMPPC_IRQ_XICS:
671 		kvmppc_xics_free_icp(vcpu);
672 		break;
673 	}
674 
675 	kvmppc_core_vcpu_free(vcpu);
676 }
677 
678 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
679 {
680 	kvm_arch_vcpu_free(vcpu);
681 }
682 
683 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
684 {
685 	return kvmppc_core_pending_dec(vcpu);
686 }
687 
688 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
689 {
690 	struct kvm_vcpu *vcpu;
691 
692 	vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
693 	kvmppc_decrementer_func(vcpu);
694 
695 	return HRTIMER_NORESTART;
696 }
697 
698 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
699 {
700 	int ret;
701 
702 	hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
703 	vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
704 	vcpu->arch.dec_expires = ~(u64)0;
705 
706 #ifdef CONFIG_KVM_EXIT_TIMING
707 	mutex_init(&vcpu->arch.exit_timing_lock);
708 #endif
709 	ret = kvmppc_subarch_vcpu_init(vcpu);
710 	return ret;
711 }
712 
713 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
714 {
715 	kvmppc_mmu_destroy(vcpu);
716 	kvmppc_subarch_vcpu_uninit(vcpu);
717 }
718 
719 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
720 {
721 #ifdef CONFIG_BOOKE
722 	/*
723 	 * vrsave (formerly usprg0) isn't used by Linux, but may
724 	 * be used by the guest.
725 	 *
726 	 * On non-booke this is associated with Altivec and
727 	 * is handled by code in book3s.c.
728 	 */
729 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
730 #endif
731 	kvmppc_core_vcpu_load(vcpu, cpu);
732 }
733 
734 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
735 {
736 	kvmppc_core_vcpu_put(vcpu);
737 #ifdef CONFIG_BOOKE
738 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
739 #endif
740 }
741 
742 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
743                                       struct kvm_run *run)
744 {
745 	u64 uninitialized_var(gpr);
746 
747 	if (run->mmio.len > sizeof(gpr)) {
748 		printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
749 		return;
750 	}
751 
752 	if (!vcpu->arch.mmio_host_swabbed) {
753 		switch (run->mmio.len) {
754 		case 8: gpr = *(u64 *)run->mmio.data; break;
755 		case 4: gpr = *(u32 *)run->mmio.data; break;
756 		case 2: gpr = *(u16 *)run->mmio.data; break;
757 		case 1: gpr = *(u8 *)run->mmio.data; break;
758 		}
759 	} else {
760 		switch (run->mmio.len) {
761 		case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
762 		case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
763 		case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
764 		case 1: gpr = *(u8 *)run->mmio.data; break;
765 		}
766 	}
767 
768 	if (vcpu->arch.mmio_sign_extend) {
769 		switch (run->mmio.len) {
770 #ifdef CONFIG_PPC64
771 		case 4:
772 			gpr = (s64)(s32)gpr;
773 			break;
774 #endif
775 		case 2:
776 			gpr = (s64)(s16)gpr;
777 			break;
778 		case 1:
779 			gpr = (s64)(s8)gpr;
780 			break;
781 		}
782 	}
783 
784 	kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
785 
786 	switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
787 	case KVM_MMIO_REG_GPR:
788 		kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
789 		break;
790 	case KVM_MMIO_REG_FPR:
791 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
792 		break;
793 #ifdef CONFIG_PPC_BOOK3S
794 	case KVM_MMIO_REG_QPR:
795 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
796 		break;
797 	case KVM_MMIO_REG_FQPR:
798 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
799 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
800 		break;
801 #endif
802 	default:
803 		BUG();
804 	}
805 }
806 
807 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
808 				unsigned int rt, unsigned int bytes,
809 				int is_default_endian, int sign_extend)
810 {
811 	int idx, ret;
812 	bool host_swabbed;
813 
814 	/* Pity C doesn't have a logical XOR operator */
815 	if (kvmppc_need_byteswap(vcpu)) {
816 		host_swabbed = is_default_endian;
817 	} else {
818 		host_swabbed = !is_default_endian;
819 	}
820 
821 	if (bytes > sizeof(run->mmio.data)) {
822 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
823 		       run->mmio.len);
824 	}
825 
826 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
827 	run->mmio.len = bytes;
828 	run->mmio.is_write = 0;
829 
830 	vcpu->arch.io_gpr = rt;
831 	vcpu->arch.mmio_host_swabbed = host_swabbed;
832 	vcpu->mmio_needed = 1;
833 	vcpu->mmio_is_write = 0;
834 	vcpu->arch.mmio_sign_extend = sign_extend;
835 
836 	idx = srcu_read_lock(&vcpu->kvm->srcu);
837 
838 	ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
839 			      bytes, &run->mmio.data);
840 
841 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
842 
843 	if (!ret) {
844 		kvmppc_complete_mmio_load(vcpu, run);
845 		vcpu->mmio_needed = 0;
846 		return EMULATE_DONE;
847 	}
848 
849 	return EMULATE_DO_MMIO;
850 }
851 
852 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
853 		       unsigned int rt, unsigned int bytes,
854 		       int is_default_endian)
855 {
856 	return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
857 }
858 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
859 
860 /* Same as above, but sign extends */
861 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
862 			unsigned int rt, unsigned int bytes,
863 			int is_default_endian)
864 {
865 	return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
866 }
867 
868 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
869 			u64 val, unsigned int bytes, int is_default_endian)
870 {
871 	void *data = run->mmio.data;
872 	int idx, ret;
873 	bool host_swabbed;
874 
875 	/* Pity C doesn't have a logical XOR operator */
876 	if (kvmppc_need_byteswap(vcpu)) {
877 		host_swabbed = is_default_endian;
878 	} else {
879 		host_swabbed = !is_default_endian;
880 	}
881 
882 	if (bytes > sizeof(run->mmio.data)) {
883 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
884 		       run->mmio.len);
885 	}
886 
887 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
888 	run->mmio.len = bytes;
889 	run->mmio.is_write = 1;
890 	vcpu->mmio_needed = 1;
891 	vcpu->mmio_is_write = 1;
892 
893 	/* Store the value at the lowest bytes in 'data'. */
894 	if (!host_swabbed) {
895 		switch (bytes) {
896 		case 8: *(u64 *)data = val; break;
897 		case 4: *(u32 *)data = val; break;
898 		case 2: *(u16 *)data = val; break;
899 		case 1: *(u8  *)data = val; break;
900 		}
901 	} else {
902 		switch (bytes) {
903 		case 8: *(u64 *)data = swab64(val); break;
904 		case 4: *(u32 *)data = swab32(val); break;
905 		case 2: *(u16 *)data = swab16(val); break;
906 		case 1: *(u8  *)data = val; break;
907 		}
908 	}
909 
910 	idx = srcu_read_lock(&vcpu->kvm->srcu);
911 
912 	ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
913 			       bytes, &run->mmio.data);
914 
915 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
916 
917 	if (!ret) {
918 		vcpu->mmio_needed = 0;
919 		return EMULATE_DONE;
920 	}
921 
922 	return EMULATE_DO_MMIO;
923 }
924 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
925 
926 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
927 {
928 	int r = 0;
929 	union kvmppc_one_reg val;
930 	int size;
931 
932 	size = one_reg_size(reg->id);
933 	if (size > sizeof(val))
934 		return -EINVAL;
935 
936 	r = kvmppc_get_one_reg(vcpu, reg->id, &val);
937 	if (r == -EINVAL) {
938 		r = 0;
939 		switch (reg->id) {
940 #ifdef CONFIG_ALTIVEC
941 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
942 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
943 				r = -ENXIO;
944 				break;
945 			}
946 			val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
947 			break;
948 		case KVM_REG_PPC_VSCR:
949 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
950 				r = -ENXIO;
951 				break;
952 			}
953 			val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
954 			break;
955 		case KVM_REG_PPC_VRSAVE:
956 			val = get_reg_val(reg->id, vcpu->arch.vrsave);
957 			break;
958 #endif /* CONFIG_ALTIVEC */
959 		default:
960 			r = -EINVAL;
961 			break;
962 		}
963 	}
964 
965 	if (r)
966 		return r;
967 
968 	if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
969 		r = -EFAULT;
970 
971 	return r;
972 }
973 
974 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
975 {
976 	int r;
977 	union kvmppc_one_reg val;
978 	int size;
979 
980 	size = one_reg_size(reg->id);
981 	if (size > sizeof(val))
982 		return -EINVAL;
983 
984 	if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
985 		return -EFAULT;
986 
987 	r = kvmppc_set_one_reg(vcpu, reg->id, &val);
988 	if (r == -EINVAL) {
989 		r = 0;
990 		switch (reg->id) {
991 #ifdef CONFIG_ALTIVEC
992 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
993 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
994 				r = -ENXIO;
995 				break;
996 			}
997 			vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
998 			break;
999 		case KVM_REG_PPC_VSCR:
1000 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1001 				r = -ENXIO;
1002 				break;
1003 			}
1004 			vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1005 			break;
1006 		case KVM_REG_PPC_VRSAVE:
1007 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1008 				r = -ENXIO;
1009 				break;
1010 			}
1011 			vcpu->arch.vrsave = set_reg_val(reg->id, val);
1012 			break;
1013 #endif /* CONFIG_ALTIVEC */
1014 		default:
1015 			r = -EINVAL;
1016 			break;
1017 		}
1018 	}
1019 
1020 	return r;
1021 }
1022 
1023 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1024 {
1025 	int r;
1026 	sigset_t sigsaved;
1027 
1028 	if (vcpu->sigset_active)
1029 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1030 
1031 	if (vcpu->mmio_needed) {
1032 		if (!vcpu->mmio_is_write)
1033 			kvmppc_complete_mmio_load(vcpu, run);
1034 		vcpu->mmio_needed = 0;
1035 	} else if (vcpu->arch.osi_needed) {
1036 		u64 *gprs = run->osi.gprs;
1037 		int i;
1038 
1039 		for (i = 0; i < 32; i++)
1040 			kvmppc_set_gpr(vcpu, i, gprs[i]);
1041 		vcpu->arch.osi_needed = 0;
1042 	} else if (vcpu->arch.hcall_needed) {
1043 		int i;
1044 
1045 		kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1046 		for (i = 0; i < 9; ++i)
1047 			kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1048 		vcpu->arch.hcall_needed = 0;
1049 #ifdef CONFIG_BOOKE
1050 	} else if (vcpu->arch.epr_needed) {
1051 		kvmppc_set_epr(vcpu, run->epr.epr);
1052 		vcpu->arch.epr_needed = 0;
1053 #endif
1054 	}
1055 
1056 	r = kvmppc_vcpu_run(run, vcpu);
1057 
1058 	if (vcpu->sigset_active)
1059 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1060 
1061 	return r;
1062 }
1063 
1064 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1065 {
1066 	if (irq->irq == KVM_INTERRUPT_UNSET) {
1067 		kvmppc_core_dequeue_external(vcpu);
1068 		return 0;
1069 	}
1070 
1071 	kvmppc_core_queue_external(vcpu, irq);
1072 
1073 	kvm_vcpu_kick(vcpu);
1074 
1075 	return 0;
1076 }
1077 
1078 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1079 				     struct kvm_enable_cap *cap)
1080 {
1081 	int r;
1082 
1083 	if (cap->flags)
1084 		return -EINVAL;
1085 
1086 	switch (cap->cap) {
1087 	case KVM_CAP_PPC_OSI:
1088 		r = 0;
1089 		vcpu->arch.osi_enabled = true;
1090 		break;
1091 	case KVM_CAP_PPC_PAPR:
1092 		r = 0;
1093 		vcpu->arch.papr_enabled = true;
1094 		break;
1095 	case KVM_CAP_PPC_EPR:
1096 		r = 0;
1097 		if (cap->args[0])
1098 			vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1099 		else
1100 			vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1101 		break;
1102 #ifdef CONFIG_BOOKE
1103 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
1104 		r = 0;
1105 		vcpu->arch.watchdog_enabled = true;
1106 		break;
1107 #endif
1108 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1109 	case KVM_CAP_SW_TLB: {
1110 		struct kvm_config_tlb cfg;
1111 		void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1112 
1113 		r = -EFAULT;
1114 		if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1115 			break;
1116 
1117 		r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1118 		break;
1119 	}
1120 #endif
1121 #ifdef CONFIG_KVM_MPIC
1122 	case KVM_CAP_IRQ_MPIC: {
1123 		struct fd f;
1124 		struct kvm_device *dev;
1125 
1126 		r = -EBADF;
1127 		f = fdget(cap->args[0]);
1128 		if (!f.file)
1129 			break;
1130 
1131 		r = -EPERM;
1132 		dev = kvm_device_from_filp(f.file);
1133 		if (dev)
1134 			r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1135 
1136 		fdput(f);
1137 		break;
1138 	}
1139 #endif
1140 #ifdef CONFIG_KVM_XICS
1141 	case KVM_CAP_IRQ_XICS: {
1142 		struct fd f;
1143 		struct kvm_device *dev;
1144 
1145 		r = -EBADF;
1146 		f = fdget(cap->args[0]);
1147 		if (!f.file)
1148 			break;
1149 
1150 		r = -EPERM;
1151 		dev = kvm_device_from_filp(f.file);
1152 		if (dev)
1153 			r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1154 
1155 		fdput(f);
1156 		break;
1157 	}
1158 #endif /* CONFIG_KVM_XICS */
1159 	default:
1160 		r = -EINVAL;
1161 		break;
1162 	}
1163 
1164 	if (!r)
1165 		r = kvmppc_sanity_check(vcpu);
1166 
1167 	return r;
1168 }
1169 
1170 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1171                                     struct kvm_mp_state *mp_state)
1172 {
1173 	return -EINVAL;
1174 }
1175 
1176 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1177                                     struct kvm_mp_state *mp_state)
1178 {
1179 	return -EINVAL;
1180 }
1181 
1182 long kvm_arch_vcpu_ioctl(struct file *filp,
1183                          unsigned int ioctl, unsigned long arg)
1184 {
1185 	struct kvm_vcpu *vcpu = filp->private_data;
1186 	void __user *argp = (void __user *)arg;
1187 	long r;
1188 
1189 	switch (ioctl) {
1190 	case KVM_INTERRUPT: {
1191 		struct kvm_interrupt irq;
1192 		r = -EFAULT;
1193 		if (copy_from_user(&irq, argp, sizeof(irq)))
1194 			goto out;
1195 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1196 		goto out;
1197 	}
1198 
1199 	case KVM_ENABLE_CAP:
1200 	{
1201 		struct kvm_enable_cap cap;
1202 		r = -EFAULT;
1203 		if (copy_from_user(&cap, argp, sizeof(cap)))
1204 			goto out;
1205 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1206 		break;
1207 	}
1208 
1209 	case KVM_SET_ONE_REG:
1210 	case KVM_GET_ONE_REG:
1211 	{
1212 		struct kvm_one_reg reg;
1213 		r = -EFAULT;
1214 		if (copy_from_user(&reg, argp, sizeof(reg)))
1215 			goto out;
1216 		if (ioctl == KVM_SET_ONE_REG)
1217 			r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
1218 		else
1219 			r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
1220 		break;
1221 	}
1222 
1223 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1224 	case KVM_DIRTY_TLB: {
1225 		struct kvm_dirty_tlb dirty;
1226 		r = -EFAULT;
1227 		if (copy_from_user(&dirty, argp, sizeof(dirty)))
1228 			goto out;
1229 		r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1230 		break;
1231 	}
1232 #endif
1233 	default:
1234 		r = -EINVAL;
1235 	}
1236 
1237 out:
1238 	return r;
1239 }
1240 
1241 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1242 {
1243 	return VM_FAULT_SIGBUS;
1244 }
1245 
1246 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1247 {
1248 	u32 inst_nop = 0x60000000;
1249 #ifdef CONFIG_KVM_BOOKE_HV
1250 	u32 inst_sc1 = 0x44000022;
1251 	pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1252 	pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1253 	pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1254 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1255 #else
1256 	u32 inst_lis = 0x3c000000;
1257 	u32 inst_ori = 0x60000000;
1258 	u32 inst_sc = 0x44000002;
1259 	u32 inst_imm_mask = 0xffff;
1260 
1261 	/*
1262 	 * The hypercall to get into KVM from within guest context is as
1263 	 * follows:
1264 	 *
1265 	 *    lis r0, r0, KVM_SC_MAGIC_R0@h
1266 	 *    ori r0, KVM_SC_MAGIC_R0@l
1267 	 *    sc
1268 	 *    nop
1269 	 */
1270 	pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1271 	pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1272 	pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1273 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1274 #endif
1275 
1276 	pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1277 
1278 	return 0;
1279 }
1280 
1281 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1282 			  bool line_status)
1283 {
1284 	if (!irqchip_in_kernel(kvm))
1285 		return -ENXIO;
1286 
1287 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1288 					irq_event->irq, irq_event->level,
1289 					line_status);
1290 	return 0;
1291 }
1292 
1293 
1294 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1295 				   struct kvm_enable_cap *cap)
1296 {
1297 	int r;
1298 
1299 	if (cap->flags)
1300 		return -EINVAL;
1301 
1302 	switch (cap->cap) {
1303 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1304 	case KVM_CAP_PPC_ENABLE_HCALL: {
1305 		unsigned long hcall = cap->args[0];
1306 
1307 		r = -EINVAL;
1308 		if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1309 		    cap->args[1] > 1)
1310 			break;
1311 		if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1312 			break;
1313 		if (cap->args[1])
1314 			set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1315 		else
1316 			clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1317 		r = 0;
1318 		break;
1319 	}
1320 #endif
1321 	default:
1322 		r = -EINVAL;
1323 		break;
1324 	}
1325 
1326 	return r;
1327 }
1328 
1329 long kvm_arch_vm_ioctl(struct file *filp,
1330                        unsigned int ioctl, unsigned long arg)
1331 {
1332 	struct kvm *kvm __maybe_unused = filp->private_data;
1333 	void __user *argp = (void __user *)arg;
1334 	long r;
1335 
1336 	switch (ioctl) {
1337 	case KVM_PPC_GET_PVINFO: {
1338 		struct kvm_ppc_pvinfo pvinfo;
1339 		memset(&pvinfo, 0, sizeof(pvinfo));
1340 		r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1341 		if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1342 			r = -EFAULT;
1343 			goto out;
1344 		}
1345 
1346 		break;
1347 	}
1348 	case KVM_ENABLE_CAP:
1349 	{
1350 		struct kvm_enable_cap cap;
1351 		r = -EFAULT;
1352 		if (copy_from_user(&cap, argp, sizeof(cap)))
1353 			goto out;
1354 		r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1355 		break;
1356 	}
1357 #ifdef CONFIG_PPC_BOOK3S_64
1358 	case KVM_CREATE_SPAPR_TCE_64: {
1359 		struct kvm_create_spapr_tce_64 create_tce_64;
1360 
1361 		r = -EFAULT;
1362 		if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
1363 			goto out;
1364 		if (create_tce_64.flags) {
1365 			r = -EINVAL;
1366 			goto out;
1367 		}
1368 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
1369 		goto out;
1370 	}
1371 	case KVM_CREATE_SPAPR_TCE: {
1372 		struct kvm_create_spapr_tce create_tce;
1373 		struct kvm_create_spapr_tce_64 create_tce_64;
1374 
1375 		r = -EFAULT;
1376 		if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1377 			goto out;
1378 
1379 		create_tce_64.liobn = create_tce.liobn;
1380 		create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
1381 		create_tce_64.offset = 0;
1382 		create_tce_64.size = create_tce.window_size >>
1383 				IOMMU_PAGE_SHIFT_4K;
1384 		create_tce_64.flags = 0;
1385 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
1386 		goto out;
1387 	}
1388 	case KVM_PPC_GET_SMMU_INFO: {
1389 		struct kvm_ppc_smmu_info info;
1390 		struct kvm *kvm = filp->private_data;
1391 
1392 		memset(&info, 0, sizeof(info));
1393 		r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1394 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1395 			r = -EFAULT;
1396 		break;
1397 	}
1398 	case KVM_PPC_RTAS_DEFINE_TOKEN: {
1399 		struct kvm *kvm = filp->private_data;
1400 
1401 		r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1402 		break;
1403 	}
1404 	default: {
1405 		struct kvm *kvm = filp->private_data;
1406 		r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1407 	}
1408 #else /* CONFIG_PPC_BOOK3S_64 */
1409 	default:
1410 		r = -ENOTTY;
1411 #endif
1412 	}
1413 out:
1414 	return r;
1415 }
1416 
1417 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1418 static unsigned long nr_lpids;
1419 
1420 long kvmppc_alloc_lpid(void)
1421 {
1422 	long lpid;
1423 
1424 	do {
1425 		lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1426 		if (lpid >= nr_lpids) {
1427 			pr_err("%s: No LPIDs free\n", __func__);
1428 			return -ENOMEM;
1429 		}
1430 	} while (test_and_set_bit(lpid, lpid_inuse));
1431 
1432 	return lpid;
1433 }
1434 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1435 
1436 void kvmppc_claim_lpid(long lpid)
1437 {
1438 	set_bit(lpid, lpid_inuse);
1439 }
1440 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1441 
1442 void kvmppc_free_lpid(long lpid)
1443 {
1444 	clear_bit(lpid, lpid_inuse);
1445 }
1446 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1447 
1448 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1449 {
1450 	nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1451 	memset(lpid_inuse, 0, sizeof(lpid_inuse));
1452 }
1453 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1454 
1455 int kvm_arch_init(void *opaque)
1456 {
1457 	return 0;
1458 }
1459 
1460 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);
1461