xref: /linux/arch/powerpc/kvm/powerpc.c (revision b7019ac550eb3916f34d79db583e9b7ea2524afa)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *
4  * Copyright IBM Corp. 2007
5  *
6  * Authors: Hollis Blanchard <hollisb@us.ibm.com>
7  *          Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
8  */
9 
10 #include <linux/errno.h>
11 #include <linux/err.h>
12 #include <linux/kvm_host.h>
13 #include <linux/vmalloc.h>
14 #include <linux/hrtimer.h>
15 #include <linux/sched/signal.h>
16 #include <linux/fs.h>
17 #include <linux/slab.h>
18 #include <linux/file.h>
19 #include <linux/module.h>
20 #include <linux/irqbypass.h>
21 #include <linux/kvm_irqfd.h>
22 #include <asm/cputable.h>
23 #include <linux/uaccess.h>
24 #include <asm/kvm_ppc.h>
25 #include <asm/cputhreads.h>
26 #include <asm/irqflags.h>
27 #include <asm/iommu.h>
28 #include <asm/switch_to.h>
29 #include <asm/xive.h>
30 #ifdef CONFIG_PPC_PSERIES
31 #include <asm/hvcall.h>
32 #include <asm/plpar_wrappers.h>
33 #endif
34 
35 #include "timing.h"
36 #include "irq.h"
37 #include "../mm/mmu_decl.h"
38 
39 #define CREATE_TRACE_POINTS
40 #include "trace.h"
41 
42 struct kvmppc_ops *kvmppc_hv_ops;
43 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
44 struct kvmppc_ops *kvmppc_pr_ops;
45 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
46 
47 
48 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
49 {
50 	return !!(v->arch.pending_exceptions) || kvm_request_pending(v);
51 }
52 
53 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
54 {
55 	return false;
56 }
57 
58 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
59 {
60 	return 1;
61 }
62 
63 /*
64  * Common checks before entering the guest world.  Call with interrupts
65  * disabled.
66  *
67  * returns:
68  *
69  * == 1 if we're ready to go into guest state
70  * <= 0 if we need to go back to the host with return value
71  */
72 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
73 {
74 	int r;
75 
76 	WARN_ON(irqs_disabled());
77 	hard_irq_disable();
78 
79 	while (true) {
80 		if (need_resched()) {
81 			local_irq_enable();
82 			cond_resched();
83 			hard_irq_disable();
84 			continue;
85 		}
86 
87 		if (signal_pending(current)) {
88 			kvmppc_account_exit(vcpu, SIGNAL_EXITS);
89 			vcpu->run->exit_reason = KVM_EXIT_INTR;
90 			r = -EINTR;
91 			break;
92 		}
93 
94 		vcpu->mode = IN_GUEST_MODE;
95 
96 		/*
97 		 * Reading vcpu->requests must happen after setting vcpu->mode,
98 		 * so we don't miss a request because the requester sees
99 		 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
100 		 * before next entering the guest (and thus doesn't IPI).
101 		 * This also orders the write to mode from any reads
102 		 * to the page tables done while the VCPU is running.
103 		 * Please see the comment in kvm_flush_remote_tlbs.
104 		 */
105 		smp_mb();
106 
107 		if (kvm_request_pending(vcpu)) {
108 			/* Make sure we process requests preemptable */
109 			local_irq_enable();
110 			trace_kvm_check_requests(vcpu);
111 			r = kvmppc_core_check_requests(vcpu);
112 			hard_irq_disable();
113 			if (r > 0)
114 				continue;
115 			break;
116 		}
117 
118 		if (kvmppc_core_prepare_to_enter(vcpu)) {
119 			/* interrupts got enabled in between, so we
120 			   are back at square 1 */
121 			continue;
122 		}
123 
124 		guest_enter_irqoff();
125 		return 1;
126 	}
127 
128 	/* return to host */
129 	local_irq_enable();
130 	return r;
131 }
132 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
133 
134 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
135 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
136 {
137 	struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
138 	int i;
139 
140 	shared->sprg0 = swab64(shared->sprg0);
141 	shared->sprg1 = swab64(shared->sprg1);
142 	shared->sprg2 = swab64(shared->sprg2);
143 	shared->sprg3 = swab64(shared->sprg3);
144 	shared->srr0 = swab64(shared->srr0);
145 	shared->srr1 = swab64(shared->srr1);
146 	shared->dar = swab64(shared->dar);
147 	shared->msr = swab64(shared->msr);
148 	shared->dsisr = swab32(shared->dsisr);
149 	shared->int_pending = swab32(shared->int_pending);
150 	for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
151 		shared->sr[i] = swab32(shared->sr[i]);
152 }
153 #endif
154 
155 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
156 {
157 	int nr = kvmppc_get_gpr(vcpu, 11);
158 	int r;
159 	unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
160 	unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
161 	unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
162 	unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
163 	unsigned long r2 = 0;
164 
165 	if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
166 		/* 32 bit mode */
167 		param1 &= 0xffffffff;
168 		param2 &= 0xffffffff;
169 		param3 &= 0xffffffff;
170 		param4 &= 0xffffffff;
171 	}
172 
173 	switch (nr) {
174 	case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
175 	{
176 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
177 		/* Book3S can be little endian, find it out here */
178 		int shared_big_endian = true;
179 		if (vcpu->arch.intr_msr & MSR_LE)
180 			shared_big_endian = false;
181 		if (shared_big_endian != vcpu->arch.shared_big_endian)
182 			kvmppc_swab_shared(vcpu);
183 		vcpu->arch.shared_big_endian = shared_big_endian;
184 #endif
185 
186 		if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
187 			/*
188 			 * Older versions of the Linux magic page code had
189 			 * a bug where they would map their trampoline code
190 			 * NX. If that's the case, remove !PR NX capability.
191 			 */
192 			vcpu->arch.disable_kernel_nx = true;
193 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
194 		}
195 
196 		vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
197 		vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
198 
199 #ifdef CONFIG_PPC_64K_PAGES
200 		/*
201 		 * Make sure our 4k magic page is in the same window of a 64k
202 		 * page within the guest and within the host's page.
203 		 */
204 		if ((vcpu->arch.magic_page_pa & 0xf000) !=
205 		    ((ulong)vcpu->arch.shared & 0xf000)) {
206 			void *old_shared = vcpu->arch.shared;
207 			ulong shared = (ulong)vcpu->arch.shared;
208 			void *new_shared;
209 
210 			shared &= PAGE_MASK;
211 			shared |= vcpu->arch.magic_page_pa & 0xf000;
212 			new_shared = (void*)shared;
213 			memcpy(new_shared, old_shared, 0x1000);
214 			vcpu->arch.shared = new_shared;
215 		}
216 #endif
217 
218 		r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
219 
220 		r = EV_SUCCESS;
221 		break;
222 	}
223 	case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
224 		r = EV_SUCCESS;
225 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
226 		r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
227 #endif
228 
229 		/* Second return value is in r4 */
230 		break;
231 	case EV_HCALL_TOKEN(EV_IDLE):
232 		r = EV_SUCCESS;
233 		kvm_vcpu_block(vcpu);
234 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
235 		break;
236 	default:
237 		r = EV_UNIMPLEMENTED;
238 		break;
239 	}
240 
241 	kvmppc_set_gpr(vcpu, 4, r2);
242 
243 	return r;
244 }
245 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
246 
247 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
248 {
249 	int r = false;
250 
251 	/* We have to know what CPU to virtualize */
252 	if (!vcpu->arch.pvr)
253 		goto out;
254 
255 	/* PAPR only works with book3s_64 */
256 	if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
257 		goto out;
258 
259 	/* HV KVM can only do PAPR mode for now */
260 	if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
261 		goto out;
262 
263 #ifdef CONFIG_KVM_BOOKE_HV
264 	if (!cpu_has_feature(CPU_FTR_EMB_HV))
265 		goto out;
266 #endif
267 
268 	r = true;
269 
270 out:
271 	vcpu->arch.sane = r;
272 	return r ? 0 : -EINVAL;
273 }
274 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
275 
276 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
277 {
278 	enum emulation_result er;
279 	int r;
280 
281 	er = kvmppc_emulate_loadstore(vcpu);
282 	switch (er) {
283 	case EMULATE_DONE:
284 		/* Future optimization: only reload non-volatiles if they were
285 		 * actually modified. */
286 		r = RESUME_GUEST_NV;
287 		break;
288 	case EMULATE_AGAIN:
289 		r = RESUME_GUEST;
290 		break;
291 	case EMULATE_DO_MMIO:
292 		run->exit_reason = KVM_EXIT_MMIO;
293 		/* We must reload nonvolatiles because "update" load/store
294 		 * instructions modify register state. */
295 		/* Future optimization: only reload non-volatiles if they were
296 		 * actually modified. */
297 		r = RESUME_HOST_NV;
298 		break;
299 	case EMULATE_FAIL:
300 	{
301 		u32 last_inst;
302 
303 		kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
304 		/* XXX Deliver Program interrupt to guest. */
305 		pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
306 		r = RESUME_HOST;
307 		break;
308 	}
309 	default:
310 		WARN_ON(1);
311 		r = RESUME_GUEST;
312 	}
313 
314 	return r;
315 }
316 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
317 
318 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
319 	      bool data)
320 {
321 	ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
322 	struct kvmppc_pte pte;
323 	int r = -EINVAL;
324 
325 	vcpu->stat.st++;
326 
327 	if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->store_to_eaddr)
328 		r = vcpu->kvm->arch.kvm_ops->store_to_eaddr(vcpu, eaddr, ptr,
329 							    size);
330 
331 	if ((!r) || (r == -EAGAIN))
332 		return r;
333 
334 	r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
335 			 XLATE_WRITE, &pte);
336 	if (r < 0)
337 		return r;
338 
339 	*eaddr = pte.raddr;
340 
341 	if (!pte.may_write)
342 		return -EPERM;
343 
344 	/* Magic page override */
345 	if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
346 	    ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
347 	    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
348 		void *magic = vcpu->arch.shared;
349 		magic += pte.eaddr & 0xfff;
350 		memcpy(magic, ptr, size);
351 		return EMULATE_DONE;
352 	}
353 
354 	if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
355 		return EMULATE_DO_MMIO;
356 
357 	return EMULATE_DONE;
358 }
359 EXPORT_SYMBOL_GPL(kvmppc_st);
360 
361 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
362 		      bool data)
363 {
364 	ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
365 	struct kvmppc_pte pte;
366 	int rc = -EINVAL;
367 
368 	vcpu->stat.ld++;
369 
370 	if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->load_from_eaddr)
371 		rc = vcpu->kvm->arch.kvm_ops->load_from_eaddr(vcpu, eaddr, ptr,
372 							      size);
373 
374 	if ((!rc) || (rc == -EAGAIN))
375 		return rc;
376 
377 	rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
378 			  XLATE_READ, &pte);
379 	if (rc)
380 		return rc;
381 
382 	*eaddr = pte.raddr;
383 
384 	if (!pte.may_read)
385 		return -EPERM;
386 
387 	if (!data && !pte.may_execute)
388 		return -ENOEXEC;
389 
390 	/* Magic page override */
391 	if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
392 	    ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
393 	    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
394 		void *magic = vcpu->arch.shared;
395 		magic += pte.eaddr & 0xfff;
396 		memcpy(ptr, magic, size);
397 		return EMULATE_DONE;
398 	}
399 
400 	if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
401 		return EMULATE_DO_MMIO;
402 
403 	return EMULATE_DONE;
404 }
405 EXPORT_SYMBOL_GPL(kvmppc_ld);
406 
407 int kvm_arch_hardware_enable(void)
408 {
409 	return 0;
410 }
411 
412 int kvm_arch_hardware_setup(void)
413 {
414 	return 0;
415 }
416 
417 void kvm_arch_check_processor_compat(void *rtn)
418 {
419 	*(int *)rtn = kvmppc_core_check_processor_compat();
420 }
421 
422 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
423 {
424 	struct kvmppc_ops *kvm_ops = NULL;
425 	/*
426 	 * if we have both HV and PR enabled, default is HV
427 	 */
428 	if (type == 0) {
429 		if (kvmppc_hv_ops)
430 			kvm_ops = kvmppc_hv_ops;
431 		else
432 			kvm_ops = kvmppc_pr_ops;
433 		if (!kvm_ops)
434 			goto err_out;
435 	} else	if (type == KVM_VM_PPC_HV) {
436 		if (!kvmppc_hv_ops)
437 			goto err_out;
438 		kvm_ops = kvmppc_hv_ops;
439 	} else if (type == KVM_VM_PPC_PR) {
440 		if (!kvmppc_pr_ops)
441 			goto err_out;
442 		kvm_ops = kvmppc_pr_ops;
443 	} else
444 		goto err_out;
445 
446 	if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
447 		return -ENOENT;
448 
449 	kvm->arch.kvm_ops = kvm_ops;
450 	return kvmppc_core_init_vm(kvm);
451 err_out:
452 	return -EINVAL;
453 }
454 
455 bool kvm_arch_has_vcpu_debugfs(void)
456 {
457 	return false;
458 }
459 
460 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
461 {
462 	return 0;
463 }
464 
465 void kvm_arch_destroy_vm(struct kvm *kvm)
466 {
467 	unsigned int i;
468 	struct kvm_vcpu *vcpu;
469 
470 #ifdef CONFIG_KVM_XICS
471 	/*
472 	 * We call kick_all_cpus_sync() to ensure that all
473 	 * CPUs have executed any pending IPIs before we
474 	 * continue and free VCPUs structures below.
475 	 */
476 	if (is_kvmppc_hv_enabled(kvm))
477 		kick_all_cpus_sync();
478 #endif
479 
480 	kvm_for_each_vcpu(i, vcpu, kvm)
481 		kvm_arch_vcpu_free(vcpu);
482 
483 	mutex_lock(&kvm->lock);
484 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
485 		kvm->vcpus[i] = NULL;
486 
487 	atomic_set(&kvm->online_vcpus, 0);
488 
489 	kvmppc_core_destroy_vm(kvm);
490 
491 	mutex_unlock(&kvm->lock);
492 
493 	/* drop the module reference */
494 	module_put(kvm->arch.kvm_ops->owner);
495 }
496 
497 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
498 {
499 	int r;
500 	/* Assume we're using HV mode when the HV module is loaded */
501 	int hv_enabled = kvmppc_hv_ops ? 1 : 0;
502 
503 	if (kvm) {
504 		/*
505 		 * Hooray - we know which VM type we're running on. Depend on
506 		 * that rather than the guess above.
507 		 */
508 		hv_enabled = is_kvmppc_hv_enabled(kvm);
509 	}
510 
511 	switch (ext) {
512 #ifdef CONFIG_BOOKE
513 	case KVM_CAP_PPC_BOOKE_SREGS:
514 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
515 	case KVM_CAP_PPC_EPR:
516 #else
517 	case KVM_CAP_PPC_SEGSTATE:
518 	case KVM_CAP_PPC_HIOR:
519 	case KVM_CAP_PPC_PAPR:
520 #endif
521 	case KVM_CAP_PPC_UNSET_IRQ:
522 	case KVM_CAP_PPC_IRQ_LEVEL:
523 	case KVM_CAP_ENABLE_CAP:
524 	case KVM_CAP_ONE_REG:
525 	case KVM_CAP_IOEVENTFD:
526 	case KVM_CAP_DEVICE_CTRL:
527 	case KVM_CAP_IMMEDIATE_EXIT:
528 		r = 1;
529 		break;
530 	case KVM_CAP_PPC_PAIRED_SINGLES:
531 	case KVM_CAP_PPC_OSI:
532 	case KVM_CAP_PPC_GET_PVINFO:
533 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
534 	case KVM_CAP_SW_TLB:
535 #endif
536 		/* We support this only for PR */
537 		r = !hv_enabled;
538 		break;
539 #ifdef CONFIG_KVM_MPIC
540 	case KVM_CAP_IRQ_MPIC:
541 		r = 1;
542 		break;
543 #endif
544 
545 #ifdef CONFIG_PPC_BOOK3S_64
546 	case KVM_CAP_SPAPR_TCE:
547 	case KVM_CAP_SPAPR_TCE_64:
548 		r = 1;
549 		break;
550 	case KVM_CAP_SPAPR_TCE_VFIO:
551 		r = !!cpu_has_feature(CPU_FTR_HVMODE);
552 		break;
553 	case KVM_CAP_PPC_RTAS:
554 	case KVM_CAP_PPC_FIXUP_HCALL:
555 	case KVM_CAP_PPC_ENABLE_HCALL:
556 #ifdef CONFIG_KVM_XICS
557 	case KVM_CAP_IRQ_XICS:
558 #endif
559 	case KVM_CAP_PPC_GET_CPU_CHAR:
560 		r = 1;
561 		break;
562 #ifdef CONFIG_KVM_XIVE
563 	case KVM_CAP_PPC_IRQ_XIVE:
564 		/*
565 		 * We need XIVE to be enabled on the platform (implies
566 		 * a POWER9 processor) and the PowerNV platform, as
567 		 * nested is not yet supported.
568 		 */
569 		r = xive_enabled() && !!cpu_has_feature(CPU_FTR_HVMODE);
570 		break;
571 #endif
572 
573 	case KVM_CAP_PPC_ALLOC_HTAB:
574 		r = hv_enabled;
575 		break;
576 #endif /* CONFIG_PPC_BOOK3S_64 */
577 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
578 	case KVM_CAP_PPC_SMT:
579 		r = 0;
580 		if (kvm) {
581 			if (kvm->arch.emul_smt_mode > 1)
582 				r = kvm->arch.emul_smt_mode;
583 			else
584 				r = kvm->arch.smt_mode;
585 		} else if (hv_enabled) {
586 			if (cpu_has_feature(CPU_FTR_ARCH_300))
587 				r = 1;
588 			else
589 				r = threads_per_subcore;
590 		}
591 		break;
592 	case KVM_CAP_PPC_SMT_POSSIBLE:
593 		r = 1;
594 		if (hv_enabled) {
595 			if (!cpu_has_feature(CPU_FTR_ARCH_300))
596 				r = ((threads_per_subcore << 1) - 1);
597 			else
598 				/* P9 can emulate dbells, so allow any mode */
599 				r = 8 | 4 | 2 | 1;
600 		}
601 		break;
602 	case KVM_CAP_PPC_RMA:
603 		r = 0;
604 		break;
605 	case KVM_CAP_PPC_HWRNG:
606 		r = kvmppc_hwrng_present();
607 		break;
608 	case KVM_CAP_PPC_MMU_RADIX:
609 		r = !!(hv_enabled && radix_enabled());
610 		break;
611 	case KVM_CAP_PPC_MMU_HASH_V3:
612 		r = !!(hv_enabled && cpu_has_feature(CPU_FTR_ARCH_300) &&
613 		       cpu_has_feature(CPU_FTR_HVMODE));
614 		break;
615 	case KVM_CAP_PPC_NESTED_HV:
616 		r = !!(hv_enabled && kvmppc_hv_ops->enable_nested &&
617 		       !kvmppc_hv_ops->enable_nested(NULL));
618 		break;
619 #endif
620 	case KVM_CAP_SYNC_MMU:
621 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
622 		r = hv_enabled;
623 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
624 		r = 1;
625 #else
626 		r = 0;
627 #endif
628 		break;
629 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
630 	case KVM_CAP_PPC_HTAB_FD:
631 		r = hv_enabled;
632 		break;
633 #endif
634 	case KVM_CAP_NR_VCPUS:
635 		/*
636 		 * Recommending a number of CPUs is somewhat arbitrary; we
637 		 * return the number of present CPUs for -HV (since a host
638 		 * will have secondary threads "offline"), and for other KVM
639 		 * implementations just count online CPUs.
640 		 */
641 		if (hv_enabled)
642 			r = num_present_cpus();
643 		else
644 			r = num_online_cpus();
645 		break;
646 	case KVM_CAP_MAX_VCPUS:
647 		r = KVM_MAX_VCPUS;
648 		break;
649 	case KVM_CAP_MAX_VCPU_ID:
650 		r = KVM_MAX_VCPU_ID;
651 		break;
652 #ifdef CONFIG_PPC_BOOK3S_64
653 	case KVM_CAP_PPC_GET_SMMU_INFO:
654 		r = 1;
655 		break;
656 	case KVM_CAP_SPAPR_MULTITCE:
657 		r = 1;
658 		break;
659 	case KVM_CAP_SPAPR_RESIZE_HPT:
660 		r = !!hv_enabled;
661 		break;
662 #endif
663 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
664 	case KVM_CAP_PPC_FWNMI:
665 		r = hv_enabled;
666 		break;
667 #endif
668 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
669 	case KVM_CAP_PPC_HTM:
670 		r = !!(cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM) ||
671 		     (hv_enabled && cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST));
672 		break;
673 #endif
674 	default:
675 		r = 0;
676 		break;
677 	}
678 	return r;
679 
680 }
681 
682 long kvm_arch_dev_ioctl(struct file *filp,
683                         unsigned int ioctl, unsigned long arg)
684 {
685 	return -EINVAL;
686 }
687 
688 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
689 			   struct kvm_memory_slot *dont)
690 {
691 	kvmppc_core_free_memslot(kvm, free, dont);
692 }
693 
694 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
695 			    unsigned long npages)
696 {
697 	return kvmppc_core_create_memslot(kvm, slot, npages);
698 }
699 
700 int kvm_arch_prepare_memory_region(struct kvm *kvm,
701 				   struct kvm_memory_slot *memslot,
702 				   const struct kvm_userspace_memory_region *mem,
703 				   enum kvm_mr_change change)
704 {
705 	return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
706 }
707 
708 void kvm_arch_commit_memory_region(struct kvm *kvm,
709 				   const struct kvm_userspace_memory_region *mem,
710 				   const struct kvm_memory_slot *old,
711 				   const struct kvm_memory_slot *new,
712 				   enum kvm_mr_change change)
713 {
714 	kvmppc_core_commit_memory_region(kvm, mem, old, new, change);
715 }
716 
717 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
718 				   struct kvm_memory_slot *slot)
719 {
720 	kvmppc_core_flush_memslot(kvm, slot);
721 }
722 
723 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
724 {
725 	struct kvm_vcpu *vcpu;
726 	vcpu = kvmppc_core_vcpu_create(kvm, id);
727 	if (!IS_ERR(vcpu)) {
728 		vcpu->arch.wqp = &vcpu->wq;
729 		kvmppc_create_vcpu_debugfs(vcpu, id);
730 	}
731 	return vcpu;
732 }
733 
734 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
735 {
736 }
737 
738 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
739 {
740 	/* Make sure we're not using the vcpu anymore */
741 	hrtimer_cancel(&vcpu->arch.dec_timer);
742 
743 	kvmppc_remove_vcpu_debugfs(vcpu);
744 
745 	switch (vcpu->arch.irq_type) {
746 	case KVMPPC_IRQ_MPIC:
747 		kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
748 		break;
749 	case KVMPPC_IRQ_XICS:
750 		if (xics_on_xive())
751 			kvmppc_xive_cleanup_vcpu(vcpu);
752 		else
753 			kvmppc_xics_free_icp(vcpu);
754 		break;
755 	case KVMPPC_IRQ_XIVE:
756 		kvmppc_xive_native_cleanup_vcpu(vcpu);
757 		break;
758 	}
759 
760 	kvmppc_core_vcpu_free(vcpu);
761 }
762 
763 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
764 {
765 	kvm_arch_vcpu_free(vcpu);
766 }
767 
768 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
769 {
770 	return kvmppc_core_pending_dec(vcpu);
771 }
772 
773 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
774 {
775 	struct kvm_vcpu *vcpu;
776 
777 	vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
778 	kvmppc_decrementer_func(vcpu);
779 
780 	return HRTIMER_NORESTART;
781 }
782 
783 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
784 {
785 	int ret;
786 
787 	hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
788 	vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
789 	vcpu->arch.dec_expires = get_tb();
790 
791 #ifdef CONFIG_KVM_EXIT_TIMING
792 	mutex_init(&vcpu->arch.exit_timing_lock);
793 #endif
794 	ret = kvmppc_subarch_vcpu_init(vcpu);
795 	return ret;
796 }
797 
798 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
799 {
800 	kvmppc_mmu_destroy(vcpu);
801 	kvmppc_subarch_vcpu_uninit(vcpu);
802 }
803 
804 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
805 {
806 #ifdef CONFIG_BOOKE
807 	/*
808 	 * vrsave (formerly usprg0) isn't used by Linux, but may
809 	 * be used by the guest.
810 	 *
811 	 * On non-booke this is associated with Altivec and
812 	 * is handled by code in book3s.c.
813 	 */
814 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
815 #endif
816 	kvmppc_core_vcpu_load(vcpu, cpu);
817 }
818 
819 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
820 {
821 	kvmppc_core_vcpu_put(vcpu);
822 #ifdef CONFIG_BOOKE
823 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
824 #endif
825 }
826 
827 /*
828  * irq_bypass_add_producer and irq_bypass_del_producer are only
829  * useful if the architecture supports PCI passthrough.
830  * irq_bypass_stop and irq_bypass_start are not needed and so
831  * kvm_ops are not defined for them.
832  */
833 bool kvm_arch_has_irq_bypass(void)
834 {
835 	return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
836 		(kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
837 }
838 
839 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
840 				     struct irq_bypass_producer *prod)
841 {
842 	struct kvm_kernel_irqfd *irqfd =
843 		container_of(cons, struct kvm_kernel_irqfd, consumer);
844 	struct kvm *kvm = irqfd->kvm;
845 
846 	if (kvm->arch.kvm_ops->irq_bypass_add_producer)
847 		return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
848 
849 	return 0;
850 }
851 
852 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
853 				      struct irq_bypass_producer *prod)
854 {
855 	struct kvm_kernel_irqfd *irqfd =
856 		container_of(cons, struct kvm_kernel_irqfd, consumer);
857 	struct kvm *kvm = irqfd->kvm;
858 
859 	if (kvm->arch.kvm_ops->irq_bypass_del_producer)
860 		kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
861 }
862 
863 #ifdef CONFIG_VSX
864 static inline int kvmppc_get_vsr_dword_offset(int index)
865 {
866 	int offset;
867 
868 	if ((index != 0) && (index != 1))
869 		return -1;
870 
871 #ifdef __BIG_ENDIAN
872 	offset =  index;
873 #else
874 	offset = 1 - index;
875 #endif
876 
877 	return offset;
878 }
879 
880 static inline int kvmppc_get_vsr_word_offset(int index)
881 {
882 	int offset;
883 
884 	if ((index > 3) || (index < 0))
885 		return -1;
886 
887 #ifdef __BIG_ENDIAN
888 	offset = index;
889 #else
890 	offset = 3 - index;
891 #endif
892 	return offset;
893 }
894 
895 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
896 	u64 gpr)
897 {
898 	union kvmppc_one_reg val;
899 	int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
900 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
901 
902 	if (offset == -1)
903 		return;
904 
905 	if (index >= 32) {
906 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
907 		val.vsxval[offset] = gpr;
908 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
909 	} else {
910 		VCPU_VSX_FPR(vcpu, index, offset) = gpr;
911 	}
912 }
913 
914 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
915 	u64 gpr)
916 {
917 	union kvmppc_one_reg val;
918 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
919 
920 	if (index >= 32) {
921 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
922 		val.vsxval[0] = gpr;
923 		val.vsxval[1] = gpr;
924 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
925 	} else {
926 		VCPU_VSX_FPR(vcpu, index, 0) = gpr;
927 		VCPU_VSX_FPR(vcpu, index, 1) = gpr;
928 	}
929 }
930 
931 static inline void kvmppc_set_vsr_word_dump(struct kvm_vcpu *vcpu,
932 	u32 gpr)
933 {
934 	union kvmppc_one_reg val;
935 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
936 
937 	if (index >= 32) {
938 		val.vsx32val[0] = gpr;
939 		val.vsx32val[1] = gpr;
940 		val.vsx32val[2] = gpr;
941 		val.vsx32val[3] = gpr;
942 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
943 	} else {
944 		val.vsx32val[0] = gpr;
945 		val.vsx32val[1] = gpr;
946 		VCPU_VSX_FPR(vcpu, index, 0) = val.vsxval[0];
947 		VCPU_VSX_FPR(vcpu, index, 1) = val.vsxval[0];
948 	}
949 }
950 
951 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
952 	u32 gpr32)
953 {
954 	union kvmppc_one_reg val;
955 	int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
956 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
957 	int dword_offset, word_offset;
958 
959 	if (offset == -1)
960 		return;
961 
962 	if (index >= 32) {
963 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
964 		val.vsx32val[offset] = gpr32;
965 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
966 	} else {
967 		dword_offset = offset / 2;
968 		word_offset = offset % 2;
969 		val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
970 		val.vsx32val[word_offset] = gpr32;
971 		VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
972 	}
973 }
974 #endif /* CONFIG_VSX */
975 
976 #ifdef CONFIG_ALTIVEC
977 static inline int kvmppc_get_vmx_offset_generic(struct kvm_vcpu *vcpu,
978 		int index, int element_size)
979 {
980 	int offset;
981 	int elts = sizeof(vector128)/element_size;
982 
983 	if ((index < 0) || (index >= elts))
984 		return -1;
985 
986 	if (kvmppc_need_byteswap(vcpu))
987 		offset = elts - index - 1;
988 	else
989 		offset = index;
990 
991 	return offset;
992 }
993 
994 static inline int kvmppc_get_vmx_dword_offset(struct kvm_vcpu *vcpu,
995 		int index)
996 {
997 	return kvmppc_get_vmx_offset_generic(vcpu, index, 8);
998 }
999 
1000 static inline int kvmppc_get_vmx_word_offset(struct kvm_vcpu *vcpu,
1001 		int index)
1002 {
1003 	return kvmppc_get_vmx_offset_generic(vcpu, index, 4);
1004 }
1005 
1006 static inline int kvmppc_get_vmx_hword_offset(struct kvm_vcpu *vcpu,
1007 		int index)
1008 {
1009 	return kvmppc_get_vmx_offset_generic(vcpu, index, 2);
1010 }
1011 
1012 static inline int kvmppc_get_vmx_byte_offset(struct kvm_vcpu *vcpu,
1013 		int index)
1014 {
1015 	return kvmppc_get_vmx_offset_generic(vcpu, index, 1);
1016 }
1017 
1018 
1019 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu,
1020 	u64 gpr)
1021 {
1022 	union kvmppc_one_reg val;
1023 	int offset = kvmppc_get_vmx_dword_offset(vcpu,
1024 			vcpu->arch.mmio_vmx_offset);
1025 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1026 
1027 	if (offset == -1)
1028 		return;
1029 
1030 	val.vval = VCPU_VSX_VR(vcpu, index);
1031 	val.vsxval[offset] = gpr;
1032 	VCPU_VSX_VR(vcpu, index) = val.vval;
1033 }
1034 
1035 static inline void kvmppc_set_vmx_word(struct kvm_vcpu *vcpu,
1036 	u32 gpr32)
1037 {
1038 	union kvmppc_one_reg val;
1039 	int offset = kvmppc_get_vmx_word_offset(vcpu,
1040 			vcpu->arch.mmio_vmx_offset);
1041 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1042 
1043 	if (offset == -1)
1044 		return;
1045 
1046 	val.vval = VCPU_VSX_VR(vcpu, index);
1047 	val.vsx32val[offset] = gpr32;
1048 	VCPU_VSX_VR(vcpu, index) = val.vval;
1049 }
1050 
1051 static inline void kvmppc_set_vmx_hword(struct kvm_vcpu *vcpu,
1052 	u16 gpr16)
1053 {
1054 	union kvmppc_one_reg val;
1055 	int offset = kvmppc_get_vmx_hword_offset(vcpu,
1056 			vcpu->arch.mmio_vmx_offset);
1057 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1058 
1059 	if (offset == -1)
1060 		return;
1061 
1062 	val.vval = VCPU_VSX_VR(vcpu, index);
1063 	val.vsx16val[offset] = gpr16;
1064 	VCPU_VSX_VR(vcpu, index) = val.vval;
1065 }
1066 
1067 static inline void kvmppc_set_vmx_byte(struct kvm_vcpu *vcpu,
1068 	u8 gpr8)
1069 {
1070 	union kvmppc_one_reg val;
1071 	int offset = kvmppc_get_vmx_byte_offset(vcpu,
1072 			vcpu->arch.mmio_vmx_offset);
1073 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1074 
1075 	if (offset == -1)
1076 		return;
1077 
1078 	val.vval = VCPU_VSX_VR(vcpu, index);
1079 	val.vsx8val[offset] = gpr8;
1080 	VCPU_VSX_VR(vcpu, index) = val.vval;
1081 }
1082 #endif /* CONFIG_ALTIVEC */
1083 
1084 #ifdef CONFIG_PPC_FPU
1085 static inline u64 sp_to_dp(u32 fprs)
1086 {
1087 	u64 fprd;
1088 
1089 	preempt_disable();
1090 	enable_kernel_fp();
1091 	asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs)
1092 	     : "fr0");
1093 	preempt_enable();
1094 	return fprd;
1095 }
1096 
1097 static inline u32 dp_to_sp(u64 fprd)
1098 {
1099 	u32 fprs;
1100 
1101 	preempt_disable();
1102 	enable_kernel_fp();
1103 	asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd)
1104 	     : "fr0");
1105 	preempt_enable();
1106 	return fprs;
1107 }
1108 
1109 #else
1110 #define sp_to_dp(x)	(x)
1111 #define dp_to_sp(x)	(x)
1112 #endif /* CONFIG_PPC_FPU */
1113 
1114 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
1115                                       struct kvm_run *run)
1116 {
1117 	u64 uninitialized_var(gpr);
1118 
1119 	if (run->mmio.len > sizeof(gpr)) {
1120 		printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
1121 		return;
1122 	}
1123 
1124 	if (!vcpu->arch.mmio_host_swabbed) {
1125 		switch (run->mmio.len) {
1126 		case 8: gpr = *(u64 *)run->mmio.data; break;
1127 		case 4: gpr = *(u32 *)run->mmio.data; break;
1128 		case 2: gpr = *(u16 *)run->mmio.data; break;
1129 		case 1: gpr = *(u8 *)run->mmio.data; break;
1130 		}
1131 	} else {
1132 		switch (run->mmio.len) {
1133 		case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
1134 		case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
1135 		case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
1136 		case 1: gpr = *(u8 *)run->mmio.data; break;
1137 		}
1138 	}
1139 
1140 	/* conversion between single and double precision */
1141 	if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
1142 		gpr = sp_to_dp(gpr);
1143 
1144 	if (vcpu->arch.mmio_sign_extend) {
1145 		switch (run->mmio.len) {
1146 #ifdef CONFIG_PPC64
1147 		case 4:
1148 			gpr = (s64)(s32)gpr;
1149 			break;
1150 #endif
1151 		case 2:
1152 			gpr = (s64)(s16)gpr;
1153 			break;
1154 		case 1:
1155 			gpr = (s64)(s8)gpr;
1156 			break;
1157 		}
1158 	}
1159 
1160 	switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1161 	case KVM_MMIO_REG_GPR:
1162 		kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1163 		break;
1164 	case KVM_MMIO_REG_FPR:
1165 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1166 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_FP);
1167 
1168 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1169 		break;
1170 #ifdef CONFIG_PPC_BOOK3S
1171 	case KVM_MMIO_REG_QPR:
1172 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1173 		break;
1174 	case KVM_MMIO_REG_FQPR:
1175 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1176 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1177 		break;
1178 #endif
1179 #ifdef CONFIG_VSX
1180 	case KVM_MMIO_REG_VSX:
1181 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1182 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VSX);
1183 
1184 		if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_DWORD)
1185 			kvmppc_set_vsr_dword(vcpu, gpr);
1186 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_WORD)
1187 			kvmppc_set_vsr_word(vcpu, gpr);
1188 		else if (vcpu->arch.mmio_copy_type ==
1189 				KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1190 			kvmppc_set_vsr_dword_dump(vcpu, gpr);
1191 		else if (vcpu->arch.mmio_copy_type ==
1192 				KVMPPC_VSX_COPY_WORD_LOAD_DUMP)
1193 			kvmppc_set_vsr_word_dump(vcpu, gpr);
1194 		break;
1195 #endif
1196 #ifdef CONFIG_ALTIVEC
1197 	case KVM_MMIO_REG_VMX:
1198 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1199 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VEC);
1200 
1201 		if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_DWORD)
1202 			kvmppc_set_vmx_dword(vcpu, gpr);
1203 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_WORD)
1204 			kvmppc_set_vmx_word(vcpu, gpr);
1205 		else if (vcpu->arch.mmio_copy_type ==
1206 				KVMPPC_VMX_COPY_HWORD)
1207 			kvmppc_set_vmx_hword(vcpu, gpr);
1208 		else if (vcpu->arch.mmio_copy_type ==
1209 				KVMPPC_VMX_COPY_BYTE)
1210 			kvmppc_set_vmx_byte(vcpu, gpr);
1211 		break;
1212 #endif
1213 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1214 	case KVM_MMIO_REG_NESTED_GPR:
1215 		if (kvmppc_need_byteswap(vcpu))
1216 			gpr = swab64(gpr);
1217 		kvm_vcpu_write_guest(vcpu, vcpu->arch.nested_io_gpr, &gpr,
1218 				     sizeof(gpr));
1219 		break;
1220 #endif
1221 	default:
1222 		BUG();
1223 	}
1224 }
1225 
1226 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1227 				unsigned int rt, unsigned int bytes,
1228 				int is_default_endian, int sign_extend)
1229 {
1230 	int idx, ret;
1231 	bool host_swabbed;
1232 
1233 	/* Pity C doesn't have a logical XOR operator */
1234 	if (kvmppc_need_byteswap(vcpu)) {
1235 		host_swabbed = is_default_endian;
1236 	} else {
1237 		host_swabbed = !is_default_endian;
1238 	}
1239 
1240 	if (bytes > sizeof(run->mmio.data)) {
1241 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1242 		       run->mmio.len);
1243 	}
1244 
1245 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1246 	run->mmio.len = bytes;
1247 	run->mmio.is_write = 0;
1248 
1249 	vcpu->arch.io_gpr = rt;
1250 	vcpu->arch.mmio_host_swabbed = host_swabbed;
1251 	vcpu->mmio_needed = 1;
1252 	vcpu->mmio_is_write = 0;
1253 	vcpu->arch.mmio_sign_extend = sign_extend;
1254 
1255 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1256 
1257 	ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1258 			      bytes, &run->mmio.data);
1259 
1260 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1261 
1262 	if (!ret) {
1263 		kvmppc_complete_mmio_load(vcpu, run);
1264 		vcpu->mmio_needed = 0;
1265 		return EMULATE_DONE;
1266 	}
1267 
1268 	return EMULATE_DO_MMIO;
1269 }
1270 
1271 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1272 		       unsigned int rt, unsigned int bytes,
1273 		       int is_default_endian)
1274 {
1275 	return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
1276 }
1277 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1278 
1279 /* Same as above, but sign extends */
1280 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
1281 			unsigned int rt, unsigned int bytes,
1282 			int is_default_endian)
1283 {
1284 	return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
1285 }
1286 
1287 #ifdef CONFIG_VSX
1288 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1289 			unsigned int rt, unsigned int bytes,
1290 			int is_default_endian, int mmio_sign_extend)
1291 {
1292 	enum emulation_result emulated = EMULATE_DONE;
1293 
1294 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1295 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1296 		return EMULATE_FAIL;
1297 
1298 	while (vcpu->arch.mmio_vsx_copy_nums) {
1299 		emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1300 			is_default_endian, mmio_sign_extend);
1301 
1302 		if (emulated != EMULATE_DONE)
1303 			break;
1304 
1305 		vcpu->arch.paddr_accessed += run->mmio.len;
1306 
1307 		vcpu->arch.mmio_vsx_copy_nums--;
1308 		vcpu->arch.mmio_vsx_offset++;
1309 	}
1310 	return emulated;
1311 }
1312 #endif /* CONFIG_VSX */
1313 
1314 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1315 			u64 val, unsigned int bytes, int is_default_endian)
1316 {
1317 	void *data = run->mmio.data;
1318 	int idx, ret;
1319 	bool host_swabbed;
1320 
1321 	/* Pity C doesn't have a logical XOR operator */
1322 	if (kvmppc_need_byteswap(vcpu)) {
1323 		host_swabbed = is_default_endian;
1324 	} else {
1325 		host_swabbed = !is_default_endian;
1326 	}
1327 
1328 	if (bytes > sizeof(run->mmio.data)) {
1329 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1330 		       run->mmio.len);
1331 	}
1332 
1333 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1334 	run->mmio.len = bytes;
1335 	run->mmio.is_write = 1;
1336 	vcpu->mmio_needed = 1;
1337 	vcpu->mmio_is_write = 1;
1338 
1339 	if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1340 		val = dp_to_sp(val);
1341 
1342 	/* Store the value at the lowest bytes in 'data'. */
1343 	if (!host_swabbed) {
1344 		switch (bytes) {
1345 		case 8: *(u64 *)data = val; break;
1346 		case 4: *(u32 *)data = val; break;
1347 		case 2: *(u16 *)data = val; break;
1348 		case 1: *(u8  *)data = val; break;
1349 		}
1350 	} else {
1351 		switch (bytes) {
1352 		case 8: *(u64 *)data = swab64(val); break;
1353 		case 4: *(u32 *)data = swab32(val); break;
1354 		case 2: *(u16 *)data = swab16(val); break;
1355 		case 1: *(u8  *)data = val; break;
1356 		}
1357 	}
1358 
1359 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1360 
1361 	ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1362 			       bytes, &run->mmio.data);
1363 
1364 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1365 
1366 	if (!ret) {
1367 		vcpu->mmio_needed = 0;
1368 		return EMULATE_DONE;
1369 	}
1370 
1371 	return EMULATE_DO_MMIO;
1372 }
1373 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1374 
1375 #ifdef CONFIG_VSX
1376 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1377 {
1378 	u32 dword_offset, word_offset;
1379 	union kvmppc_one_reg reg;
1380 	int vsx_offset = 0;
1381 	int copy_type = vcpu->arch.mmio_copy_type;
1382 	int result = 0;
1383 
1384 	switch (copy_type) {
1385 	case KVMPPC_VSX_COPY_DWORD:
1386 		vsx_offset =
1387 			kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1388 
1389 		if (vsx_offset == -1) {
1390 			result = -1;
1391 			break;
1392 		}
1393 
1394 		if (rs < 32) {
1395 			*val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1396 		} else {
1397 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1398 			*val = reg.vsxval[vsx_offset];
1399 		}
1400 		break;
1401 
1402 	case KVMPPC_VSX_COPY_WORD:
1403 		vsx_offset =
1404 			kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1405 
1406 		if (vsx_offset == -1) {
1407 			result = -1;
1408 			break;
1409 		}
1410 
1411 		if (rs < 32) {
1412 			dword_offset = vsx_offset / 2;
1413 			word_offset = vsx_offset % 2;
1414 			reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1415 			*val = reg.vsx32val[word_offset];
1416 		} else {
1417 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1418 			*val = reg.vsx32val[vsx_offset];
1419 		}
1420 		break;
1421 
1422 	default:
1423 		result = -1;
1424 		break;
1425 	}
1426 
1427 	return result;
1428 }
1429 
1430 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1431 			int rs, unsigned int bytes, int is_default_endian)
1432 {
1433 	u64 val;
1434 	enum emulation_result emulated = EMULATE_DONE;
1435 
1436 	vcpu->arch.io_gpr = rs;
1437 
1438 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1439 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1440 		return EMULATE_FAIL;
1441 
1442 	while (vcpu->arch.mmio_vsx_copy_nums) {
1443 		if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1444 			return EMULATE_FAIL;
1445 
1446 		emulated = kvmppc_handle_store(run, vcpu,
1447 			 val, bytes, is_default_endian);
1448 
1449 		if (emulated != EMULATE_DONE)
1450 			break;
1451 
1452 		vcpu->arch.paddr_accessed += run->mmio.len;
1453 
1454 		vcpu->arch.mmio_vsx_copy_nums--;
1455 		vcpu->arch.mmio_vsx_offset++;
1456 	}
1457 
1458 	return emulated;
1459 }
1460 
1461 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu,
1462 			struct kvm_run *run)
1463 {
1464 	enum emulation_result emulated = EMULATE_FAIL;
1465 	int r;
1466 
1467 	vcpu->arch.paddr_accessed += run->mmio.len;
1468 
1469 	if (!vcpu->mmio_is_write) {
1470 		emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr,
1471 			 run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1472 	} else {
1473 		emulated = kvmppc_handle_vsx_store(run, vcpu,
1474 			 vcpu->arch.io_gpr, run->mmio.len, 1);
1475 	}
1476 
1477 	switch (emulated) {
1478 	case EMULATE_DO_MMIO:
1479 		run->exit_reason = KVM_EXIT_MMIO;
1480 		r = RESUME_HOST;
1481 		break;
1482 	case EMULATE_FAIL:
1483 		pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1484 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1485 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1486 		r = RESUME_HOST;
1487 		break;
1488 	default:
1489 		r = RESUME_GUEST;
1490 		break;
1491 	}
1492 	return r;
1493 }
1494 #endif /* CONFIG_VSX */
1495 
1496 #ifdef CONFIG_ALTIVEC
1497 int kvmppc_handle_vmx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1498 		unsigned int rt, unsigned int bytes, int is_default_endian)
1499 {
1500 	enum emulation_result emulated = EMULATE_DONE;
1501 
1502 	if (vcpu->arch.mmio_vsx_copy_nums > 2)
1503 		return EMULATE_FAIL;
1504 
1505 	while (vcpu->arch.mmio_vmx_copy_nums) {
1506 		emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1507 				is_default_endian, 0);
1508 
1509 		if (emulated != EMULATE_DONE)
1510 			break;
1511 
1512 		vcpu->arch.paddr_accessed += run->mmio.len;
1513 		vcpu->arch.mmio_vmx_copy_nums--;
1514 		vcpu->arch.mmio_vmx_offset++;
1515 	}
1516 
1517 	return emulated;
1518 }
1519 
1520 int kvmppc_get_vmx_dword(struct kvm_vcpu *vcpu, int index, u64 *val)
1521 {
1522 	union kvmppc_one_reg reg;
1523 	int vmx_offset = 0;
1524 	int result = 0;
1525 
1526 	vmx_offset =
1527 		kvmppc_get_vmx_dword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1528 
1529 	if (vmx_offset == -1)
1530 		return -1;
1531 
1532 	reg.vval = VCPU_VSX_VR(vcpu, index);
1533 	*val = reg.vsxval[vmx_offset];
1534 
1535 	return result;
1536 }
1537 
1538 int kvmppc_get_vmx_word(struct kvm_vcpu *vcpu, int index, u64 *val)
1539 {
1540 	union kvmppc_one_reg reg;
1541 	int vmx_offset = 0;
1542 	int result = 0;
1543 
1544 	vmx_offset =
1545 		kvmppc_get_vmx_word_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1546 
1547 	if (vmx_offset == -1)
1548 		return -1;
1549 
1550 	reg.vval = VCPU_VSX_VR(vcpu, index);
1551 	*val = reg.vsx32val[vmx_offset];
1552 
1553 	return result;
1554 }
1555 
1556 int kvmppc_get_vmx_hword(struct kvm_vcpu *vcpu, int index, u64 *val)
1557 {
1558 	union kvmppc_one_reg reg;
1559 	int vmx_offset = 0;
1560 	int result = 0;
1561 
1562 	vmx_offset =
1563 		kvmppc_get_vmx_hword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1564 
1565 	if (vmx_offset == -1)
1566 		return -1;
1567 
1568 	reg.vval = VCPU_VSX_VR(vcpu, index);
1569 	*val = reg.vsx16val[vmx_offset];
1570 
1571 	return result;
1572 }
1573 
1574 int kvmppc_get_vmx_byte(struct kvm_vcpu *vcpu, int index, u64 *val)
1575 {
1576 	union kvmppc_one_reg reg;
1577 	int vmx_offset = 0;
1578 	int result = 0;
1579 
1580 	vmx_offset =
1581 		kvmppc_get_vmx_byte_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1582 
1583 	if (vmx_offset == -1)
1584 		return -1;
1585 
1586 	reg.vval = VCPU_VSX_VR(vcpu, index);
1587 	*val = reg.vsx8val[vmx_offset];
1588 
1589 	return result;
1590 }
1591 
1592 int kvmppc_handle_vmx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1593 		unsigned int rs, unsigned int bytes, int is_default_endian)
1594 {
1595 	u64 val = 0;
1596 	unsigned int index = rs & KVM_MMIO_REG_MASK;
1597 	enum emulation_result emulated = EMULATE_DONE;
1598 
1599 	if (vcpu->arch.mmio_vsx_copy_nums > 2)
1600 		return EMULATE_FAIL;
1601 
1602 	vcpu->arch.io_gpr = rs;
1603 
1604 	while (vcpu->arch.mmio_vmx_copy_nums) {
1605 		switch (vcpu->arch.mmio_copy_type) {
1606 		case KVMPPC_VMX_COPY_DWORD:
1607 			if (kvmppc_get_vmx_dword(vcpu, index, &val) == -1)
1608 				return EMULATE_FAIL;
1609 
1610 			break;
1611 		case KVMPPC_VMX_COPY_WORD:
1612 			if (kvmppc_get_vmx_word(vcpu, index, &val) == -1)
1613 				return EMULATE_FAIL;
1614 			break;
1615 		case KVMPPC_VMX_COPY_HWORD:
1616 			if (kvmppc_get_vmx_hword(vcpu, index, &val) == -1)
1617 				return EMULATE_FAIL;
1618 			break;
1619 		case KVMPPC_VMX_COPY_BYTE:
1620 			if (kvmppc_get_vmx_byte(vcpu, index, &val) == -1)
1621 				return EMULATE_FAIL;
1622 			break;
1623 		default:
1624 			return EMULATE_FAIL;
1625 		}
1626 
1627 		emulated = kvmppc_handle_store(run, vcpu, val, bytes,
1628 				is_default_endian);
1629 		if (emulated != EMULATE_DONE)
1630 			break;
1631 
1632 		vcpu->arch.paddr_accessed += run->mmio.len;
1633 		vcpu->arch.mmio_vmx_copy_nums--;
1634 		vcpu->arch.mmio_vmx_offset++;
1635 	}
1636 
1637 	return emulated;
1638 }
1639 
1640 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu,
1641 		struct kvm_run *run)
1642 {
1643 	enum emulation_result emulated = EMULATE_FAIL;
1644 	int r;
1645 
1646 	vcpu->arch.paddr_accessed += run->mmio.len;
1647 
1648 	if (!vcpu->mmio_is_write) {
1649 		emulated = kvmppc_handle_vmx_load(run, vcpu,
1650 				vcpu->arch.io_gpr, run->mmio.len, 1);
1651 	} else {
1652 		emulated = kvmppc_handle_vmx_store(run, vcpu,
1653 				vcpu->arch.io_gpr, run->mmio.len, 1);
1654 	}
1655 
1656 	switch (emulated) {
1657 	case EMULATE_DO_MMIO:
1658 		run->exit_reason = KVM_EXIT_MMIO;
1659 		r = RESUME_HOST;
1660 		break;
1661 	case EMULATE_FAIL:
1662 		pr_info("KVM: MMIO emulation failed (VMX repeat)\n");
1663 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1664 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1665 		r = RESUME_HOST;
1666 		break;
1667 	default:
1668 		r = RESUME_GUEST;
1669 		break;
1670 	}
1671 	return r;
1672 }
1673 #endif /* CONFIG_ALTIVEC */
1674 
1675 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1676 {
1677 	int r = 0;
1678 	union kvmppc_one_reg val;
1679 	int size;
1680 
1681 	size = one_reg_size(reg->id);
1682 	if (size > sizeof(val))
1683 		return -EINVAL;
1684 
1685 	r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1686 	if (r == -EINVAL) {
1687 		r = 0;
1688 		switch (reg->id) {
1689 #ifdef CONFIG_ALTIVEC
1690 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1691 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1692 				r = -ENXIO;
1693 				break;
1694 			}
1695 			val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1696 			break;
1697 		case KVM_REG_PPC_VSCR:
1698 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1699 				r = -ENXIO;
1700 				break;
1701 			}
1702 			val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1703 			break;
1704 		case KVM_REG_PPC_VRSAVE:
1705 			val = get_reg_val(reg->id, vcpu->arch.vrsave);
1706 			break;
1707 #endif /* CONFIG_ALTIVEC */
1708 		default:
1709 			r = -EINVAL;
1710 			break;
1711 		}
1712 	}
1713 
1714 	if (r)
1715 		return r;
1716 
1717 	if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1718 		r = -EFAULT;
1719 
1720 	return r;
1721 }
1722 
1723 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1724 {
1725 	int r;
1726 	union kvmppc_one_reg val;
1727 	int size;
1728 
1729 	size = one_reg_size(reg->id);
1730 	if (size > sizeof(val))
1731 		return -EINVAL;
1732 
1733 	if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1734 		return -EFAULT;
1735 
1736 	r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1737 	if (r == -EINVAL) {
1738 		r = 0;
1739 		switch (reg->id) {
1740 #ifdef CONFIG_ALTIVEC
1741 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1742 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1743 				r = -ENXIO;
1744 				break;
1745 			}
1746 			vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1747 			break;
1748 		case KVM_REG_PPC_VSCR:
1749 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1750 				r = -ENXIO;
1751 				break;
1752 			}
1753 			vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1754 			break;
1755 		case KVM_REG_PPC_VRSAVE:
1756 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1757 				r = -ENXIO;
1758 				break;
1759 			}
1760 			vcpu->arch.vrsave = set_reg_val(reg->id, val);
1761 			break;
1762 #endif /* CONFIG_ALTIVEC */
1763 		default:
1764 			r = -EINVAL;
1765 			break;
1766 		}
1767 	}
1768 
1769 	return r;
1770 }
1771 
1772 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1773 {
1774 	int r;
1775 
1776 	vcpu_load(vcpu);
1777 
1778 	if (vcpu->mmio_needed) {
1779 		vcpu->mmio_needed = 0;
1780 		if (!vcpu->mmio_is_write)
1781 			kvmppc_complete_mmio_load(vcpu, run);
1782 #ifdef CONFIG_VSX
1783 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1784 			vcpu->arch.mmio_vsx_copy_nums--;
1785 			vcpu->arch.mmio_vsx_offset++;
1786 		}
1787 
1788 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1789 			r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run);
1790 			if (r == RESUME_HOST) {
1791 				vcpu->mmio_needed = 1;
1792 				goto out;
1793 			}
1794 		}
1795 #endif
1796 #ifdef CONFIG_ALTIVEC
1797 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1798 			vcpu->arch.mmio_vmx_copy_nums--;
1799 			vcpu->arch.mmio_vmx_offset++;
1800 		}
1801 
1802 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1803 			r = kvmppc_emulate_mmio_vmx_loadstore(vcpu, run);
1804 			if (r == RESUME_HOST) {
1805 				vcpu->mmio_needed = 1;
1806 				goto out;
1807 			}
1808 		}
1809 #endif
1810 	} else if (vcpu->arch.osi_needed) {
1811 		u64 *gprs = run->osi.gprs;
1812 		int i;
1813 
1814 		for (i = 0; i < 32; i++)
1815 			kvmppc_set_gpr(vcpu, i, gprs[i]);
1816 		vcpu->arch.osi_needed = 0;
1817 	} else if (vcpu->arch.hcall_needed) {
1818 		int i;
1819 
1820 		kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1821 		for (i = 0; i < 9; ++i)
1822 			kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1823 		vcpu->arch.hcall_needed = 0;
1824 #ifdef CONFIG_BOOKE
1825 	} else if (vcpu->arch.epr_needed) {
1826 		kvmppc_set_epr(vcpu, run->epr.epr);
1827 		vcpu->arch.epr_needed = 0;
1828 #endif
1829 	}
1830 
1831 	kvm_sigset_activate(vcpu);
1832 
1833 	if (run->immediate_exit)
1834 		r = -EINTR;
1835 	else
1836 		r = kvmppc_vcpu_run(run, vcpu);
1837 
1838 	kvm_sigset_deactivate(vcpu);
1839 
1840 #ifdef CONFIG_ALTIVEC
1841 out:
1842 #endif
1843 	vcpu_put(vcpu);
1844 	return r;
1845 }
1846 
1847 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1848 {
1849 	if (irq->irq == KVM_INTERRUPT_UNSET) {
1850 		kvmppc_core_dequeue_external(vcpu);
1851 		return 0;
1852 	}
1853 
1854 	kvmppc_core_queue_external(vcpu, irq);
1855 
1856 	kvm_vcpu_kick(vcpu);
1857 
1858 	return 0;
1859 }
1860 
1861 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1862 				     struct kvm_enable_cap *cap)
1863 {
1864 	int r;
1865 
1866 	if (cap->flags)
1867 		return -EINVAL;
1868 
1869 	switch (cap->cap) {
1870 	case KVM_CAP_PPC_OSI:
1871 		r = 0;
1872 		vcpu->arch.osi_enabled = true;
1873 		break;
1874 	case KVM_CAP_PPC_PAPR:
1875 		r = 0;
1876 		vcpu->arch.papr_enabled = true;
1877 		break;
1878 	case KVM_CAP_PPC_EPR:
1879 		r = 0;
1880 		if (cap->args[0])
1881 			vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1882 		else
1883 			vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1884 		break;
1885 #ifdef CONFIG_BOOKE
1886 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
1887 		r = 0;
1888 		vcpu->arch.watchdog_enabled = true;
1889 		break;
1890 #endif
1891 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1892 	case KVM_CAP_SW_TLB: {
1893 		struct kvm_config_tlb cfg;
1894 		void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1895 
1896 		r = -EFAULT;
1897 		if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1898 			break;
1899 
1900 		r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1901 		break;
1902 	}
1903 #endif
1904 #ifdef CONFIG_KVM_MPIC
1905 	case KVM_CAP_IRQ_MPIC: {
1906 		struct fd f;
1907 		struct kvm_device *dev;
1908 
1909 		r = -EBADF;
1910 		f = fdget(cap->args[0]);
1911 		if (!f.file)
1912 			break;
1913 
1914 		r = -EPERM;
1915 		dev = kvm_device_from_filp(f.file);
1916 		if (dev)
1917 			r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1918 
1919 		fdput(f);
1920 		break;
1921 	}
1922 #endif
1923 #ifdef CONFIG_KVM_XICS
1924 	case KVM_CAP_IRQ_XICS: {
1925 		struct fd f;
1926 		struct kvm_device *dev;
1927 
1928 		r = -EBADF;
1929 		f = fdget(cap->args[0]);
1930 		if (!f.file)
1931 			break;
1932 
1933 		r = -EPERM;
1934 		dev = kvm_device_from_filp(f.file);
1935 		if (dev) {
1936 			if (xics_on_xive())
1937 				r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1938 			else
1939 				r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1940 		}
1941 
1942 		fdput(f);
1943 		break;
1944 	}
1945 #endif /* CONFIG_KVM_XICS */
1946 #ifdef CONFIG_KVM_XIVE
1947 	case KVM_CAP_PPC_IRQ_XIVE: {
1948 		struct fd f;
1949 		struct kvm_device *dev;
1950 
1951 		r = -EBADF;
1952 		f = fdget(cap->args[0]);
1953 		if (!f.file)
1954 			break;
1955 
1956 		r = -ENXIO;
1957 		if (!xive_enabled())
1958 			break;
1959 
1960 		r = -EPERM;
1961 		dev = kvm_device_from_filp(f.file);
1962 		if (dev)
1963 			r = kvmppc_xive_native_connect_vcpu(dev, vcpu,
1964 							    cap->args[1]);
1965 
1966 		fdput(f);
1967 		break;
1968 	}
1969 #endif /* CONFIG_KVM_XIVE */
1970 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1971 	case KVM_CAP_PPC_FWNMI:
1972 		r = -EINVAL;
1973 		if (!is_kvmppc_hv_enabled(vcpu->kvm))
1974 			break;
1975 		r = 0;
1976 		vcpu->kvm->arch.fwnmi_enabled = true;
1977 		break;
1978 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1979 	default:
1980 		r = -EINVAL;
1981 		break;
1982 	}
1983 
1984 	if (!r)
1985 		r = kvmppc_sanity_check(vcpu);
1986 
1987 	return r;
1988 }
1989 
1990 bool kvm_arch_intc_initialized(struct kvm *kvm)
1991 {
1992 #ifdef CONFIG_KVM_MPIC
1993 	if (kvm->arch.mpic)
1994 		return true;
1995 #endif
1996 #ifdef CONFIG_KVM_XICS
1997 	if (kvm->arch.xics || kvm->arch.xive)
1998 		return true;
1999 #endif
2000 	return false;
2001 }
2002 
2003 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2004                                     struct kvm_mp_state *mp_state)
2005 {
2006 	return -EINVAL;
2007 }
2008 
2009 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2010                                     struct kvm_mp_state *mp_state)
2011 {
2012 	return -EINVAL;
2013 }
2014 
2015 long kvm_arch_vcpu_async_ioctl(struct file *filp,
2016 			       unsigned int ioctl, unsigned long arg)
2017 {
2018 	struct kvm_vcpu *vcpu = filp->private_data;
2019 	void __user *argp = (void __user *)arg;
2020 
2021 	if (ioctl == KVM_INTERRUPT) {
2022 		struct kvm_interrupt irq;
2023 		if (copy_from_user(&irq, argp, sizeof(irq)))
2024 			return -EFAULT;
2025 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2026 	}
2027 	return -ENOIOCTLCMD;
2028 }
2029 
2030 long kvm_arch_vcpu_ioctl(struct file *filp,
2031                          unsigned int ioctl, unsigned long arg)
2032 {
2033 	struct kvm_vcpu *vcpu = filp->private_data;
2034 	void __user *argp = (void __user *)arg;
2035 	long r;
2036 
2037 	switch (ioctl) {
2038 	case KVM_ENABLE_CAP:
2039 	{
2040 		struct kvm_enable_cap cap;
2041 		r = -EFAULT;
2042 		vcpu_load(vcpu);
2043 		if (copy_from_user(&cap, argp, sizeof(cap)))
2044 			goto out;
2045 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
2046 		vcpu_put(vcpu);
2047 		break;
2048 	}
2049 
2050 	case KVM_SET_ONE_REG:
2051 	case KVM_GET_ONE_REG:
2052 	{
2053 		struct kvm_one_reg reg;
2054 		r = -EFAULT;
2055 		if (copy_from_user(&reg, argp, sizeof(reg)))
2056 			goto out;
2057 		if (ioctl == KVM_SET_ONE_REG)
2058 			r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
2059 		else
2060 			r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
2061 		break;
2062 	}
2063 
2064 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
2065 	case KVM_DIRTY_TLB: {
2066 		struct kvm_dirty_tlb dirty;
2067 		r = -EFAULT;
2068 		vcpu_load(vcpu);
2069 		if (copy_from_user(&dirty, argp, sizeof(dirty)))
2070 			goto out;
2071 		r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
2072 		vcpu_put(vcpu);
2073 		break;
2074 	}
2075 #endif
2076 	default:
2077 		r = -EINVAL;
2078 	}
2079 
2080 out:
2081 	return r;
2082 }
2083 
2084 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2085 {
2086 	return VM_FAULT_SIGBUS;
2087 }
2088 
2089 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
2090 {
2091 	u32 inst_nop = 0x60000000;
2092 #ifdef CONFIG_KVM_BOOKE_HV
2093 	u32 inst_sc1 = 0x44000022;
2094 	pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
2095 	pvinfo->hcall[1] = cpu_to_be32(inst_nop);
2096 	pvinfo->hcall[2] = cpu_to_be32(inst_nop);
2097 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2098 #else
2099 	u32 inst_lis = 0x3c000000;
2100 	u32 inst_ori = 0x60000000;
2101 	u32 inst_sc = 0x44000002;
2102 	u32 inst_imm_mask = 0xffff;
2103 
2104 	/*
2105 	 * The hypercall to get into KVM from within guest context is as
2106 	 * follows:
2107 	 *
2108 	 *    lis r0, r0, KVM_SC_MAGIC_R0@h
2109 	 *    ori r0, KVM_SC_MAGIC_R0@l
2110 	 *    sc
2111 	 *    nop
2112 	 */
2113 	pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
2114 	pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
2115 	pvinfo->hcall[2] = cpu_to_be32(inst_sc);
2116 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2117 #endif
2118 
2119 	pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
2120 
2121 	return 0;
2122 }
2123 
2124 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
2125 			  bool line_status)
2126 {
2127 	if (!irqchip_in_kernel(kvm))
2128 		return -ENXIO;
2129 
2130 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2131 					irq_event->irq, irq_event->level,
2132 					line_status);
2133 	return 0;
2134 }
2135 
2136 
2137 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
2138 			    struct kvm_enable_cap *cap)
2139 {
2140 	int r;
2141 
2142 	if (cap->flags)
2143 		return -EINVAL;
2144 
2145 	switch (cap->cap) {
2146 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
2147 	case KVM_CAP_PPC_ENABLE_HCALL: {
2148 		unsigned long hcall = cap->args[0];
2149 
2150 		r = -EINVAL;
2151 		if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
2152 		    cap->args[1] > 1)
2153 			break;
2154 		if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
2155 			break;
2156 		if (cap->args[1])
2157 			set_bit(hcall / 4, kvm->arch.enabled_hcalls);
2158 		else
2159 			clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
2160 		r = 0;
2161 		break;
2162 	}
2163 	case KVM_CAP_PPC_SMT: {
2164 		unsigned long mode = cap->args[0];
2165 		unsigned long flags = cap->args[1];
2166 
2167 		r = -EINVAL;
2168 		if (kvm->arch.kvm_ops->set_smt_mode)
2169 			r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
2170 		break;
2171 	}
2172 
2173 	case KVM_CAP_PPC_NESTED_HV:
2174 		r = -EINVAL;
2175 		if (!is_kvmppc_hv_enabled(kvm) ||
2176 		    !kvm->arch.kvm_ops->enable_nested)
2177 			break;
2178 		r = kvm->arch.kvm_ops->enable_nested(kvm);
2179 		break;
2180 #endif
2181 	default:
2182 		r = -EINVAL;
2183 		break;
2184 	}
2185 
2186 	return r;
2187 }
2188 
2189 #ifdef CONFIG_PPC_BOOK3S_64
2190 /*
2191  * These functions check whether the underlying hardware is safe
2192  * against attacks based on observing the effects of speculatively
2193  * executed instructions, and whether it supplies instructions for
2194  * use in workarounds.  The information comes from firmware, either
2195  * via the device tree on powernv platforms or from an hcall on
2196  * pseries platforms.
2197  */
2198 #ifdef CONFIG_PPC_PSERIES
2199 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2200 {
2201 	struct h_cpu_char_result c;
2202 	unsigned long rc;
2203 
2204 	if (!machine_is(pseries))
2205 		return -ENOTTY;
2206 
2207 	rc = plpar_get_cpu_characteristics(&c);
2208 	if (rc == H_SUCCESS) {
2209 		cp->character = c.character;
2210 		cp->behaviour = c.behaviour;
2211 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2212 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2213 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2214 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2215 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2216 			KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
2217 			KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
2218 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2219 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2220 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2221 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2222 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2223 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2224 	}
2225 	return 0;
2226 }
2227 #else
2228 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2229 {
2230 	return -ENOTTY;
2231 }
2232 #endif
2233 
2234 static inline bool have_fw_feat(struct device_node *fw_features,
2235 				const char *state, const char *name)
2236 {
2237 	struct device_node *np;
2238 	bool r = false;
2239 
2240 	np = of_get_child_by_name(fw_features, name);
2241 	if (np) {
2242 		r = of_property_read_bool(np, state);
2243 		of_node_put(np);
2244 	}
2245 	return r;
2246 }
2247 
2248 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2249 {
2250 	struct device_node *np, *fw_features;
2251 	int r;
2252 
2253 	memset(cp, 0, sizeof(*cp));
2254 	r = pseries_get_cpu_char(cp);
2255 	if (r != -ENOTTY)
2256 		return r;
2257 
2258 	np = of_find_node_by_name(NULL, "ibm,opal");
2259 	if (np) {
2260 		fw_features = of_get_child_by_name(np, "fw-features");
2261 		of_node_put(np);
2262 		if (!fw_features)
2263 			return 0;
2264 		if (have_fw_feat(fw_features, "enabled",
2265 				 "inst-spec-barrier-ori31,31,0"))
2266 			cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
2267 		if (have_fw_feat(fw_features, "enabled",
2268 				 "fw-bcctrl-serialized"))
2269 			cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
2270 		if (have_fw_feat(fw_features, "enabled",
2271 				 "inst-l1d-flush-ori30,30,0"))
2272 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
2273 		if (have_fw_feat(fw_features, "enabled",
2274 				 "inst-l1d-flush-trig2"))
2275 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
2276 		if (have_fw_feat(fw_features, "enabled",
2277 				 "fw-l1d-thread-split"))
2278 			cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
2279 		if (have_fw_feat(fw_features, "enabled",
2280 				 "fw-count-cache-disabled"))
2281 			cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2282 		if (have_fw_feat(fw_features, "enabled",
2283 				 "fw-count-cache-flush-bcctr2,0,0"))
2284 			cp->character |= KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2285 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2286 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2287 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2288 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2289 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2290 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2291 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2292 
2293 		if (have_fw_feat(fw_features, "enabled",
2294 				 "speculation-policy-favor-security"))
2295 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
2296 		if (!have_fw_feat(fw_features, "disabled",
2297 				  "needs-l1d-flush-msr-pr-0-to-1"))
2298 			cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
2299 		if (!have_fw_feat(fw_features, "disabled",
2300 				  "needs-spec-barrier-for-bound-checks"))
2301 			cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2302 		if (have_fw_feat(fw_features, "enabled",
2303 				 "needs-count-cache-flush-on-context-switch"))
2304 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2305 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2306 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2307 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2308 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2309 
2310 		of_node_put(fw_features);
2311 	}
2312 
2313 	return 0;
2314 }
2315 #endif
2316 
2317 long kvm_arch_vm_ioctl(struct file *filp,
2318                        unsigned int ioctl, unsigned long arg)
2319 {
2320 	struct kvm *kvm __maybe_unused = filp->private_data;
2321 	void __user *argp = (void __user *)arg;
2322 	long r;
2323 
2324 	switch (ioctl) {
2325 	case KVM_PPC_GET_PVINFO: {
2326 		struct kvm_ppc_pvinfo pvinfo;
2327 		memset(&pvinfo, 0, sizeof(pvinfo));
2328 		r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
2329 		if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
2330 			r = -EFAULT;
2331 			goto out;
2332 		}
2333 
2334 		break;
2335 	}
2336 #ifdef CONFIG_SPAPR_TCE_IOMMU
2337 	case KVM_CREATE_SPAPR_TCE_64: {
2338 		struct kvm_create_spapr_tce_64 create_tce_64;
2339 
2340 		r = -EFAULT;
2341 		if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
2342 			goto out;
2343 		if (create_tce_64.flags) {
2344 			r = -EINVAL;
2345 			goto out;
2346 		}
2347 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2348 		goto out;
2349 	}
2350 	case KVM_CREATE_SPAPR_TCE: {
2351 		struct kvm_create_spapr_tce create_tce;
2352 		struct kvm_create_spapr_tce_64 create_tce_64;
2353 
2354 		r = -EFAULT;
2355 		if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
2356 			goto out;
2357 
2358 		create_tce_64.liobn = create_tce.liobn;
2359 		create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
2360 		create_tce_64.offset = 0;
2361 		create_tce_64.size = create_tce.window_size >>
2362 				IOMMU_PAGE_SHIFT_4K;
2363 		create_tce_64.flags = 0;
2364 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2365 		goto out;
2366 	}
2367 #endif
2368 #ifdef CONFIG_PPC_BOOK3S_64
2369 	case KVM_PPC_GET_SMMU_INFO: {
2370 		struct kvm_ppc_smmu_info info;
2371 		struct kvm *kvm = filp->private_data;
2372 
2373 		memset(&info, 0, sizeof(info));
2374 		r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
2375 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2376 			r = -EFAULT;
2377 		break;
2378 	}
2379 	case KVM_PPC_RTAS_DEFINE_TOKEN: {
2380 		struct kvm *kvm = filp->private_data;
2381 
2382 		r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
2383 		break;
2384 	}
2385 	case KVM_PPC_CONFIGURE_V3_MMU: {
2386 		struct kvm *kvm = filp->private_data;
2387 		struct kvm_ppc_mmuv3_cfg cfg;
2388 
2389 		r = -EINVAL;
2390 		if (!kvm->arch.kvm_ops->configure_mmu)
2391 			goto out;
2392 		r = -EFAULT;
2393 		if (copy_from_user(&cfg, argp, sizeof(cfg)))
2394 			goto out;
2395 		r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
2396 		break;
2397 	}
2398 	case KVM_PPC_GET_RMMU_INFO: {
2399 		struct kvm *kvm = filp->private_data;
2400 		struct kvm_ppc_rmmu_info info;
2401 
2402 		r = -EINVAL;
2403 		if (!kvm->arch.kvm_ops->get_rmmu_info)
2404 			goto out;
2405 		r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
2406 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2407 			r = -EFAULT;
2408 		break;
2409 	}
2410 	case KVM_PPC_GET_CPU_CHAR: {
2411 		struct kvm_ppc_cpu_char cpuchar;
2412 
2413 		r = kvmppc_get_cpu_char(&cpuchar);
2414 		if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2415 			r = -EFAULT;
2416 		break;
2417 	}
2418 	default: {
2419 		struct kvm *kvm = filp->private_data;
2420 		r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
2421 	}
2422 #else /* CONFIG_PPC_BOOK3S_64 */
2423 	default:
2424 		r = -ENOTTY;
2425 #endif
2426 	}
2427 out:
2428 	return r;
2429 }
2430 
2431 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
2432 static unsigned long nr_lpids;
2433 
2434 long kvmppc_alloc_lpid(void)
2435 {
2436 	long lpid;
2437 
2438 	do {
2439 		lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
2440 		if (lpid >= nr_lpids) {
2441 			pr_err("%s: No LPIDs free\n", __func__);
2442 			return -ENOMEM;
2443 		}
2444 	} while (test_and_set_bit(lpid, lpid_inuse));
2445 
2446 	return lpid;
2447 }
2448 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2449 
2450 void kvmppc_claim_lpid(long lpid)
2451 {
2452 	set_bit(lpid, lpid_inuse);
2453 }
2454 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
2455 
2456 void kvmppc_free_lpid(long lpid)
2457 {
2458 	clear_bit(lpid, lpid_inuse);
2459 }
2460 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2461 
2462 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2463 {
2464 	nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
2465 	memset(lpid_inuse, 0, sizeof(lpid_inuse));
2466 }
2467 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2468 
2469 int kvm_arch_init(void *opaque)
2470 {
2471 	return 0;
2472 }
2473 
2474 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);
2475