xref: /openbmc/linux/arch/powerpc/kvm/powerpc.c (revision a16be368)
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 #include <asm/ultravisor.h>
35 
36 #include "timing.h"
37 #include "irq.h"
38 #include "../mm/mmu_decl.h"
39 
40 #define CREATE_TRACE_POINTS
41 #include "trace.h"
42 
43 struct kvmppc_ops *kvmppc_hv_ops;
44 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
45 struct kvmppc_ops *kvmppc_pr_ops;
46 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
47 
48 
49 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
50 {
51 	return !!(v->arch.pending_exceptions) || kvm_request_pending(v);
52 }
53 
54 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
55 {
56 	return kvm_arch_vcpu_runnable(vcpu);
57 }
58 
59 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
60 {
61 	return false;
62 }
63 
64 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
65 {
66 	return 1;
67 }
68 
69 /*
70  * Common checks before entering the guest world.  Call with interrupts
71  * disabled.
72  *
73  * returns:
74  *
75  * == 1 if we're ready to go into guest state
76  * <= 0 if we need to go back to the host with return value
77  */
78 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
79 {
80 	int r;
81 
82 	WARN_ON(irqs_disabled());
83 	hard_irq_disable();
84 
85 	while (true) {
86 		if (need_resched()) {
87 			local_irq_enable();
88 			cond_resched();
89 			hard_irq_disable();
90 			continue;
91 		}
92 
93 		if (signal_pending(current)) {
94 			kvmppc_account_exit(vcpu, SIGNAL_EXITS);
95 			vcpu->run->exit_reason = KVM_EXIT_INTR;
96 			r = -EINTR;
97 			break;
98 		}
99 
100 		vcpu->mode = IN_GUEST_MODE;
101 
102 		/*
103 		 * Reading vcpu->requests must happen after setting vcpu->mode,
104 		 * so we don't miss a request because the requester sees
105 		 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
106 		 * before next entering the guest (and thus doesn't IPI).
107 		 * This also orders the write to mode from any reads
108 		 * to the page tables done while the VCPU is running.
109 		 * Please see the comment in kvm_flush_remote_tlbs.
110 		 */
111 		smp_mb();
112 
113 		if (kvm_request_pending(vcpu)) {
114 			/* Make sure we process requests preemptable */
115 			local_irq_enable();
116 			trace_kvm_check_requests(vcpu);
117 			r = kvmppc_core_check_requests(vcpu);
118 			hard_irq_disable();
119 			if (r > 0)
120 				continue;
121 			break;
122 		}
123 
124 		if (kvmppc_core_prepare_to_enter(vcpu)) {
125 			/* interrupts got enabled in between, so we
126 			   are back at square 1 */
127 			continue;
128 		}
129 
130 		guest_enter_irqoff();
131 		return 1;
132 	}
133 
134 	/* return to host */
135 	local_irq_enable();
136 	return r;
137 }
138 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
139 
140 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
141 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
142 {
143 	struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
144 	int i;
145 
146 	shared->sprg0 = swab64(shared->sprg0);
147 	shared->sprg1 = swab64(shared->sprg1);
148 	shared->sprg2 = swab64(shared->sprg2);
149 	shared->sprg3 = swab64(shared->sprg3);
150 	shared->srr0 = swab64(shared->srr0);
151 	shared->srr1 = swab64(shared->srr1);
152 	shared->dar = swab64(shared->dar);
153 	shared->msr = swab64(shared->msr);
154 	shared->dsisr = swab32(shared->dsisr);
155 	shared->int_pending = swab32(shared->int_pending);
156 	for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
157 		shared->sr[i] = swab32(shared->sr[i]);
158 }
159 #endif
160 
161 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
162 {
163 	int nr = kvmppc_get_gpr(vcpu, 11);
164 	int r;
165 	unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
166 	unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
167 	unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
168 	unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
169 	unsigned long r2 = 0;
170 
171 	if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
172 		/* 32 bit mode */
173 		param1 &= 0xffffffff;
174 		param2 &= 0xffffffff;
175 		param3 &= 0xffffffff;
176 		param4 &= 0xffffffff;
177 	}
178 
179 	switch (nr) {
180 	case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
181 	{
182 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
183 		/* Book3S can be little endian, find it out here */
184 		int shared_big_endian = true;
185 		if (vcpu->arch.intr_msr & MSR_LE)
186 			shared_big_endian = false;
187 		if (shared_big_endian != vcpu->arch.shared_big_endian)
188 			kvmppc_swab_shared(vcpu);
189 		vcpu->arch.shared_big_endian = shared_big_endian;
190 #endif
191 
192 		if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
193 			/*
194 			 * Older versions of the Linux magic page code had
195 			 * a bug where they would map their trampoline code
196 			 * NX. If that's the case, remove !PR NX capability.
197 			 */
198 			vcpu->arch.disable_kernel_nx = true;
199 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
200 		}
201 
202 		vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
203 		vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
204 
205 #ifdef CONFIG_PPC_64K_PAGES
206 		/*
207 		 * Make sure our 4k magic page is in the same window of a 64k
208 		 * page within the guest and within the host's page.
209 		 */
210 		if ((vcpu->arch.magic_page_pa & 0xf000) !=
211 		    ((ulong)vcpu->arch.shared & 0xf000)) {
212 			void *old_shared = vcpu->arch.shared;
213 			ulong shared = (ulong)vcpu->arch.shared;
214 			void *new_shared;
215 
216 			shared &= PAGE_MASK;
217 			shared |= vcpu->arch.magic_page_pa & 0xf000;
218 			new_shared = (void*)shared;
219 			memcpy(new_shared, old_shared, 0x1000);
220 			vcpu->arch.shared = new_shared;
221 		}
222 #endif
223 
224 		r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
225 
226 		r = EV_SUCCESS;
227 		break;
228 	}
229 	case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
230 		r = EV_SUCCESS;
231 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
232 		r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
233 #endif
234 
235 		/* Second return value is in r4 */
236 		break;
237 	case EV_HCALL_TOKEN(EV_IDLE):
238 		r = EV_SUCCESS;
239 		kvm_vcpu_block(vcpu);
240 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
241 		break;
242 	default:
243 		r = EV_UNIMPLEMENTED;
244 		break;
245 	}
246 
247 	kvmppc_set_gpr(vcpu, 4, r2);
248 
249 	return r;
250 }
251 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
252 
253 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
254 {
255 	int r = false;
256 
257 	/* We have to know what CPU to virtualize */
258 	if (!vcpu->arch.pvr)
259 		goto out;
260 
261 	/* PAPR only works with book3s_64 */
262 	if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
263 		goto out;
264 
265 	/* HV KVM can only do PAPR mode for now */
266 	if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
267 		goto out;
268 
269 #ifdef CONFIG_KVM_BOOKE_HV
270 	if (!cpu_has_feature(CPU_FTR_EMB_HV))
271 		goto out;
272 #endif
273 
274 	r = true;
275 
276 out:
277 	vcpu->arch.sane = r;
278 	return r ? 0 : -EINVAL;
279 }
280 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
281 
282 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
283 {
284 	enum emulation_result er;
285 	int r;
286 
287 	er = kvmppc_emulate_loadstore(vcpu);
288 	switch (er) {
289 	case EMULATE_DONE:
290 		/* Future optimization: only reload non-volatiles if they were
291 		 * actually modified. */
292 		r = RESUME_GUEST_NV;
293 		break;
294 	case EMULATE_AGAIN:
295 		r = RESUME_GUEST;
296 		break;
297 	case EMULATE_DO_MMIO:
298 		run->exit_reason = KVM_EXIT_MMIO;
299 		/* We must reload nonvolatiles because "update" load/store
300 		 * instructions modify register state. */
301 		/* Future optimization: only reload non-volatiles if they were
302 		 * actually modified. */
303 		r = RESUME_HOST_NV;
304 		break;
305 	case EMULATE_FAIL:
306 	{
307 		u32 last_inst;
308 
309 		kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
310 		/* XXX Deliver Program interrupt to guest. */
311 		pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
312 		r = RESUME_HOST;
313 		break;
314 	}
315 	default:
316 		WARN_ON(1);
317 		r = RESUME_GUEST;
318 	}
319 
320 	return r;
321 }
322 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
323 
324 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
325 	      bool data)
326 {
327 	ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
328 	struct kvmppc_pte pte;
329 	int r = -EINVAL;
330 
331 	vcpu->stat.st++;
332 
333 	if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->store_to_eaddr)
334 		r = vcpu->kvm->arch.kvm_ops->store_to_eaddr(vcpu, eaddr, ptr,
335 							    size);
336 
337 	if ((!r) || (r == -EAGAIN))
338 		return r;
339 
340 	r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
341 			 XLATE_WRITE, &pte);
342 	if (r < 0)
343 		return r;
344 
345 	*eaddr = pte.raddr;
346 
347 	if (!pte.may_write)
348 		return -EPERM;
349 
350 	/* Magic page override */
351 	if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
352 	    ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
353 	    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
354 		void *magic = vcpu->arch.shared;
355 		magic += pte.eaddr & 0xfff;
356 		memcpy(magic, ptr, size);
357 		return EMULATE_DONE;
358 	}
359 
360 	if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
361 		return EMULATE_DO_MMIO;
362 
363 	return EMULATE_DONE;
364 }
365 EXPORT_SYMBOL_GPL(kvmppc_st);
366 
367 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
368 		      bool data)
369 {
370 	ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
371 	struct kvmppc_pte pte;
372 	int rc = -EINVAL;
373 
374 	vcpu->stat.ld++;
375 
376 	if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->load_from_eaddr)
377 		rc = vcpu->kvm->arch.kvm_ops->load_from_eaddr(vcpu, eaddr, ptr,
378 							      size);
379 
380 	if ((!rc) || (rc == -EAGAIN))
381 		return rc;
382 
383 	rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
384 			  XLATE_READ, &pte);
385 	if (rc)
386 		return rc;
387 
388 	*eaddr = pte.raddr;
389 
390 	if (!pte.may_read)
391 		return -EPERM;
392 
393 	if (!data && !pte.may_execute)
394 		return -ENOEXEC;
395 
396 	/* Magic page override */
397 	if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
398 	    ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
399 	    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
400 		void *magic = vcpu->arch.shared;
401 		magic += pte.eaddr & 0xfff;
402 		memcpy(ptr, magic, size);
403 		return EMULATE_DONE;
404 	}
405 
406 	if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
407 		return EMULATE_DO_MMIO;
408 
409 	return EMULATE_DONE;
410 }
411 EXPORT_SYMBOL_GPL(kvmppc_ld);
412 
413 int kvm_arch_hardware_enable(void)
414 {
415 	return 0;
416 }
417 
418 int kvm_arch_hardware_setup(void *opaque)
419 {
420 	return 0;
421 }
422 
423 int kvm_arch_check_processor_compat(void *opaque)
424 {
425 	return kvmppc_core_check_processor_compat();
426 }
427 
428 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
429 {
430 	struct kvmppc_ops *kvm_ops = NULL;
431 	/*
432 	 * if we have both HV and PR enabled, default is HV
433 	 */
434 	if (type == 0) {
435 		if (kvmppc_hv_ops)
436 			kvm_ops = kvmppc_hv_ops;
437 		else
438 			kvm_ops = kvmppc_pr_ops;
439 		if (!kvm_ops)
440 			goto err_out;
441 	} else	if (type == KVM_VM_PPC_HV) {
442 		if (!kvmppc_hv_ops)
443 			goto err_out;
444 		kvm_ops = kvmppc_hv_ops;
445 	} else if (type == KVM_VM_PPC_PR) {
446 		if (!kvmppc_pr_ops)
447 			goto err_out;
448 		kvm_ops = kvmppc_pr_ops;
449 	} else
450 		goto err_out;
451 
452 	if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
453 		return -ENOENT;
454 
455 	kvm->arch.kvm_ops = kvm_ops;
456 	return kvmppc_core_init_vm(kvm);
457 err_out:
458 	return -EINVAL;
459 }
460 
461 void kvm_arch_destroy_vm(struct kvm *kvm)
462 {
463 	unsigned int i;
464 	struct kvm_vcpu *vcpu;
465 
466 #ifdef CONFIG_KVM_XICS
467 	/*
468 	 * We call kick_all_cpus_sync() to ensure that all
469 	 * CPUs have executed any pending IPIs before we
470 	 * continue and free VCPUs structures below.
471 	 */
472 	if (is_kvmppc_hv_enabled(kvm))
473 		kick_all_cpus_sync();
474 #endif
475 
476 	kvm_for_each_vcpu(i, vcpu, kvm)
477 		kvm_vcpu_destroy(vcpu);
478 
479 	mutex_lock(&kvm->lock);
480 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
481 		kvm->vcpus[i] = NULL;
482 
483 	atomic_set(&kvm->online_vcpus, 0);
484 
485 	kvmppc_core_destroy_vm(kvm);
486 
487 	mutex_unlock(&kvm->lock);
488 
489 	/* drop the module reference */
490 	module_put(kvm->arch.kvm_ops->owner);
491 }
492 
493 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
494 {
495 	int r;
496 	/* Assume we're using HV mode when the HV module is loaded */
497 	int hv_enabled = kvmppc_hv_ops ? 1 : 0;
498 
499 	if (kvm) {
500 		/*
501 		 * Hooray - we know which VM type we're running on. Depend on
502 		 * that rather than the guess above.
503 		 */
504 		hv_enabled = is_kvmppc_hv_enabled(kvm);
505 	}
506 
507 	switch (ext) {
508 #ifdef CONFIG_BOOKE
509 	case KVM_CAP_PPC_BOOKE_SREGS:
510 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
511 	case KVM_CAP_PPC_EPR:
512 #else
513 	case KVM_CAP_PPC_SEGSTATE:
514 	case KVM_CAP_PPC_HIOR:
515 	case KVM_CAP_PPC_PAPR:
516 #endif
517 	case KVM_CAP_PPC_UNSET_IRQ:
518 	case KVM_CAP_PPC_IRQ_LEVEL:
519 	case KVM_CAP_ENABLE_CAP:
520 	case KVM_CAP_ONE_REG:
521 	case KVM_CAP_IOEVENTFD:
522 	case KVM_CAP_DEVICE_CTRL:
523 	case KVM_CAP_IMMEDIATE_EXIT:
524 	case KVM_CAP_SET_GUEST_DEBUG:
525 		r = 1;
526 		break;
527 	case KVM_CAP_PPC_GUEST_DEBUG_SSTEP:
528 	case KVM_CAP_PPC_PAIRED_SINGLES:
529 	case KVM_CAP_PPC_OSI:
530 	case KVM_CAP_PPC_GET_PVINFO:
531 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
532 	case KVM_CAP_SW_TLB:
533 #endif
534 		/* We support this only for PR */
535 		r = !hv_enabled;
536 		break;
537 #ifdef CONFIG_KVM_MPIC
538 	case KVM_CAP_IRQ_MPIC:
539 		r = 1;
540 		break;
541 #endif
542 
543 #ifdef CONFIG_PPC_BOOK3S_64
544 	case KVM_CAP_SPAPR_TCE:
545 	case KVM_CAP_SPAPR_TCE_64:
546 		r = 1;
547 		break;
548 	case KVM_CAP_SPAPR_TCE_VFIO:
549 		r = !!cpu_has_feature(CPU_FTR_HVMODE);
550 		break;
551 	case KVM_CAP_PPC_RTAS:
552 	case KVM_CAP_PPC_FIXUP_HCALL:
553 	case KVM_CAP_PPC_ENABLE_HCALL:
554 #ifdef CONFIG_KVM_XICS
555 	case KVM_CAP_IRQ_XICS:
556 #endif
557 	case KVM_CAP_PPC_GET_CPU_CHAR:
558 		r = 1;
559 		break;
560 #ifdef CONFIG_KVM_XIVE
561 	case KVM_CAP_PPC_IRQ_XIVE:
562 		/*
563 		 * We need XIVE to be enabled on the platform (implies
564 		 * a POWER9 processor) and the PowerNV platform, as
565 		 * nested is not yet supported.
566 		 */
567 		r = xive_enabled() && !!cpu_has_feature(CPU_FTR_HVMODE) &&
568 			kvmppc_xive_native_supported();
569 		break;
570 #endif
571 
572 	case KVM_CAP_PPC_ALLOC_HTAB:
573 		r = hv_enabled;
574 		break;
575 #endif /* CONFIG_PPC_BOOK3S_64 */
576 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
577 	case KVM_CAP_PPC_SMT:
578 		r = 0;
579 		if (kvm) {
580 			if (kvm->arch.emul_smt_mode > 1)
581 				r = kvm->arch.emul_smt_mode;
582 			else
583 				r = kvm->arch.smt_mode;
584 		} else if (hv_enabled) {
585 			if (cpu_has_feature(CPU_FTR_ARCH_300))
586 				r = 1;
587 			else
588 				r = threads_per_subcore;
589 		}
590 		break;
591 	case KVM_CAP_PPC_SMT_POSSIBLE:
592 		r = 1;
593 		if (hv_enabled) {
594 			if (!cpu_has_feature(CPU_FTR_ARCH_300))
595 				r = ((threads_per_subcore << 1) - 1);
596 			else
597 				/* P9 can emulate dbells, so allow any mode */
598 				r = 8 | 4 | 2 | 1;
599 		}
600 		break;
601 	case KVM_CAP_PPC_RMA:
602 		r = 0;
603 		break;
604 	case KVM_CAP_PPC_HWRNG:
605 		r = kvmppc_hwrng_present();
606 		break;
607 	case KVM_CAP_PPC_MMU_RADIX:
608 		r = !!(hv_enabled && radix_enabled());
609 		break;
610 	case KVM_CAP_PPC_MMU_HASH_V3:
611 		r = !!(hv_enabled && cpu_has_feature(CPU_FTR_ARCH_300) &&
612 		       cpu_has_feature(CPU_FTR_HVMODE));
613 		break;
614 	case KVM_CAP_PPC_NESTED_HV:
615 		r = !!(hv_enabled && kvmppc_hv_ops->enable_nested &&
616 		       !kvmppc_hv_ops->enable_nested(NULL));
617 		break;
618 #endif
619 	case KVM_CAP_SYNC_MMU:
620 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
621 		r = hv_enabled;
622 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
623 		r = 1;
624 #else
625 		r = 0;
626 #endif
627 		break;
628 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
629 	case KVM_CAP_PPC_HTAB_FD:
630 		r = hv_enabled;
631 		break;
632 #endif
633 	case KVM_CAP_NR_VCPUS:
634 		/*
635 		 * Recommending a number of CPUs is somewhat arbitrary; we
636 		 * return the number of present CPUs for -HV (since a host
637 		 * will have secondary threads "offline"), and for other KVM
638 		 * implementations just count online CPUs.
639 		 */
640 		if (hv_enabled)
641 			r = num_present_cpus();
642 		else
643 			r = num_online_cpus();
644 		break;
645 	case KVM_CAP_MAX_VCPUS:
646 		r = KVM_MAX_VCPUS;
647 		break;
648 	case KVM_CAP_MAX_VCPU_ID:
649 		r = KVM_MAX_VCPU_ID;
650 		break;
651 #ifdef CONFIG_PPC_BOOK3S_64
652 	case KVM_CAP_PPC_GET_SMMU_INFO:
653 		r = 1;
654 		break;
655 	case KVM_CAP_SPAPR_MULTITCE:
656 		r = 1;
657 		break;
658 	case KVM_CAP_SPAPR_RESIZE_HPT:
659 		r = !!hv_enabled;
660 		break;
661 #endif
662 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
663 	case KVM_CAP_PPC_FWNMI:
664 		r = hv_enabled;
665 		break;
666 #endif
667 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
668 	case KVM_CAP_PPC_HTM:
669 		r = !!(cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM) ||
670 		     (hv_enabled && cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST));
671 		break;
672 #endif
673 #if defined(CONFIG_KVM_BOOK3S_HV_POSSIBLE)
674 	case KVM_CAP_PPC_SECURE_GUEST:
675 		r = hv_enabled && kvmppc_hv_ops->enable_svm &&
676 			!kvmppc_hv_ops->enable_svm(NULL);
677 		break;
678 #endif
679 	default:
680 		r = 0;
681 		break;
682 	}
683 	return r;
684 
685 }
686 
687 long kvm_arch_dev_ioctl(struct file *filp,
688                         unsigned int ioctl, unsigned long arg)
689 {
690 	return -EINVAL;
691 }
692 
693 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
694 {
695 	kvmppc_core_free_memslot(kvm, slot);
696 }
697 
698 int kvm_arch_prepare_memory_region(struct kvm *kvm,
699 				   struct kvm_memory_slot *memslot,
700 				   const struct kvm_userspace_memory_region *mem,
701 				   enum kvm_mr_change change)
702 {
703 	return kvmppc_core_prepare_memory_region(kvm, memslot, mem, change);
704 }
705 
706 void kvm_arch_commit_memory_region(struct kvm *kvm,
707 				   const struct kvm_userspace_memory_region *mem,
708 				   struct kvm_memory_slot *old,
709 				   const struct kvm_memory_slot *new,
710 				   enum kvm_mr_change change)
711 {
712 	kvmppc_core_commit_memory_region(kvm, mem, old, new, change);
713 }
714 
715 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
716 				   struct kvm_memory_slot *slot)
717 {
718 	kvmppc_core_flush_memslot(kvm, slot);
719 }
720 
721 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
722 {
723 	return 0;
724 }
725 
726 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
727 {
728 	struct kvm_vcpu *vcpu;
729 
730 	vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
731 	kvmppc_decrementer_func(vcpu);
732 
733 	return HRTIMER_NORESTART;
734 }
735 
736 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
737 {
738 	int err;
739 
740 	hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
741 	vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
742 	vcpu->arch.dec_expires = get_tb();
743 
744 #ifdef CONFIG_KVM_EXIT_TIMING
745 	mutex_init(&vcpu->arch.exit_timing_lock);
746 #endif
747 	err = kvmppc_subarch_vcpu_init(vcpu);
748 	if (err)
749 		return err;
750 
751 	err = kvmppc_core_vcpu_create(vcpu);
752 	if (err)
753 		goto out_vcpu_uninit;
754 
755 	vcpu->arch.waitp = &vcpu->wait;
756 	kvmppc_create_vcpu_debugfs(vcpu, vcpu->vcpu_id);
757 	return 0;
758 
759 out_vcpu_uninit:
760 	kvmppc_subarch_vcpu_uninit(vcpu);
761 	return err;
762 }
763 
764 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
765 {
766 }
767 
768 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
769 {
770 	/* Make sure we're not using the vcpu anymore */
771 	hrtimer_cancel(&vcpu->arch.dec_timer);
772 
773 	kvmppc_remove_vcpu_debugfs(vcpu);
774 
775 	switch (vcpu->arch.irq_type) {
776 	case KVMPPC_IRQ_MPIC:
777 		kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
778 		break;
779 	case KVMPPC_IRQ_XICS:
780 		if (xics_on_xive())
781 			kvmppc_xive_cleanup_vcpu(vcpu);
782 		else
783 			kvmppc_xics_free_icp(vcpu);
784 		break;
785 	case KVMPPC_IRQ_XIVE:
786 		kvmppc_xive_native_cleanup_vcpu(vcpu);
787 		break;
788 	}
789 
790 	kvmppc_core_vcpu_free(vcpu);
791 
792 	kvmppc_subarch_vcpu_uninit(vcpu);
793 }
794 
795 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
796 {
797 	return kvmppc_core_pending_dec(vcpu);
798 }
799 
800 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
801 {
802 #ifdef CONFIG_BOOKE
803 	/*
804 	 * vrsave (formerly usprg0) isn't used by Linux, but may
805 	 * be used by the guest.
806 	 *
807 	 * On non-booke this is associated with Altivec and
808 	 * is handled by code in book3s.c.
809 	 */
810 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
811 #endif
812 	kvmppc_core_vcpu_load(vcpu, cpu);
813 }
814 
815 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
816 {
817 	kvmppc_core_vcpu_put(vcpu);
818 #ifdef CONFIG_BOOKE
819 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
820 #endif
821 }
822 
823 /*
824  * irq_bypass_add_producer and irq_bypass_del_producer are only
825  * useful if the architecture supports PCI passthrough.
826  * irq_bypass_stop and irq_bypass_start are not needed and so
827  * kvm_ops are not defined for them.
828  */
829 bool kvm_arch_has_irq_bypass(void)
830 {
831 	return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
832 		(kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
833 }
834 
835 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
836 				     struct irq_bypass_producer *prod)
837 {
838 	struct kvm_kernel_irqfd *irqfd =
839 		container_of(cons, struct kvm_kernel_irqfd, consumer);
840 	struct kvm *kvm = irqfd->kvm;
841 
842 	if (kvm->arch.kvm_ops->irq_bypass_add_producer)
843 		return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
844 
845 	return 0;
846 }
847 
848 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
849 				      struct irq_bypass_producer *prod)
850 {
851 	struct kvm_kernel_irqfd *irqfd =
852 		container_of(cons, struct kvm_kernel_irqfd, consumer);
853 	struct kvm *kvm = irqfd->kvm;
854 
855 	if (kvm->arch.kvm_ops->irq_bypass_del_producer)
856 		kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
857 }
858 
859 #ifdef CONFIG_VSX
860 static inline int kvmppc_get_vsr_dword_offset(int index)
861 {
862 	int offset;
863 
864 	if ((index != 0) && (index != 1))
865 		return -1;
866 
867 #ifdef __BIG_ENDIAN
868 	offset =  index;
869 #else
870 	offset = 1 - index;
871 #endif
872 
873 	return offset;
874 }
875 
876 static inline int kvmppc_get_vsr_word_offset(int index)
877 {
878 	int offset;
879 
880 	if ((index > 3) || (index < 0))
881 		return -1;
882 
883 #ifdef __BIG_ENDIAN
884 	offset = index;
885 #else
886 	offset = 3 - index;
887 #endif
888 	return offset;
889 }
890 
891 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
892 	u64 gpr)
893 {
894 	union kvmppc_one_reg val;
895 	int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
896 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
897 
898 	if (offset == -1)
899 		return;
900 
901 	if (index >= 32) {
902 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
903 		val.vsxval[offset] = gpr;
904 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
905 	} else {
906 		VCPU_VSX_FPR(vcpu, index, offset) = gpr;
907 	}
908 }
909 
910 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
911 	u64 gpr)
912 {
913 	union kvmppc_one_reg val;
914 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
915 
916 	if (index >= 32) {
917 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
918 		val.vsxval[0] = gpr;
919 		val.vsxval[1] = gpr;
920 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
921 	} else {
922 		VCPU_VSX_FPR(vcpu, index, 0) = gpr;
923 		VCPU_VSX_FPR(vcpu, index, 1) = gpr;
924 	}
925 }
926 
927 static inline void kvmppc_set_vsr_word_dump(struct kvm_vcpu *vcpu,
928 	u32 gpr)
929 {
930 	union kvmppc_one_reg val;
931 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
932 
933 	if (index >= 32) {
934 		val.vsx32val[0] = gpr;
935 		val.vsx32val[1] = gpr;
936 		val.vsx32val[2] = gpr;
937 		val.vsx32val[3] = gpr;
938 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
939 	} else {
940 		val.vsx32val[0] = gpr;
941 		val.vsx32val[1] = gpr;
942 		VCPU_VSX_FPR(vcpu, index, 0) = val.vsxval[0];
943 		VCPU_VSX_FPR(vcpu, index, 1) = val.vsxval[0];
944 	}
945 }
946 
947 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
948 	u32 gpr32)
949 {
950 	union kvmppc_one_reg val;
951 	int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
952 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
953 	int dword_offset, word_offset;
954 
955 	if (offset == -1)
956 		return;
957 
958 	if (index >= 32) {
959 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
960 		val.vsx32val[offset] = gpr32;
961 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
962 	} else {
963 		dword_offset = offset / 2;
964 		word_offset = offset % 2;
965 		val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
966 		val.vsx32val[word_offset] = gpr32;
967 		VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
968 	}
969 }
970 #endif /* CONFIG_VSX */
971 
972 #ifdef CONFIG_ALTIVEC
973 static inline int kvmppc_get_vmx_offset_generic(struct kvm_vcpu *vcpu,
974 		int index, int element_size)
975 {
976 	int offset;
977 	int elts = sizeof(vector128)/element_size;
978 
979 	if ((index < 0) || (index >= elts))
980 		return -1;
981 
982 	if (kvmppc_need_byteswap(vcpu))
983 		offset = elts - index - 1;
984 	else
985 		offset = index;
986 
987 	return offset;
988 }
989 
990 static inline int kvmppc_get_vmx_dword_offset(struct kvm_vcpu *vcpu,
991 		int index)
992 {
993 	return kvmppc_get_vmx_offset_generic(vcpu, index, 8);
994 }
995 
996 static inline int kvmppc_get_vmx_word_offset(struct kvm_vcpu *vcpu,
997 		int index)
998 {
999 	return kvmppc_get_vmx_offset_generic(vcpu, index, 4);
1000 }
1001 
1002 static inline int kvmppc_get_vmx_hword_offset(struct kvm_vcpu *vcpu,
1003 		int index)
1004 {
1005 	return kvmppc_get_vmx_offset_generic(vcpu, index, 2);
1006 }
1007 
1008 static inline int kvmppc_get_vmx_byte_offset(struct kvm_vcpu *vcpu,
1009 		int index)
1010 {
1011 	return kvmppc_get_vmx_offset_generic(vcpu, index, 1);
1012 }
1013 
1014 
1015 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu,
1016 	u64 gpr)
1017 {
1018 	union kvmppc_one_reg val;
1019 	int offset = kvmppc_get_vmx_dword_offset(vcpu,
1020 			vcpu->arch.mmio_vmx_offset);
1021 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1022 
1023 	if (offset == -1)
1024 		return;
1025 
1026 	val.vval = VCPU_VSX_VR(vcpu, index);
1027 	val.vsxval[offset] = gpr;
1028 	VCPU_VSX_VR(vcpu, index) = val.vval;
1029 }
1030 
1031 static inline void kvmppc_set_vmx_word(struct kvm_vcpu *vcpu,
1032 	u32 gpr32)
1033 {
1034 	union kvmppc_one_reg val;
1035 	int offset = kvmppc_get_vmx_word_offset(vcpu,
1036 			vcpu->arch.mmio_vmx_offset);
1037 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1038 
1039 	if (offset == -1)
1040 		return;
1041 
1042 	val.vval = VCPU_VSX_VR(vcpu, index);
1043 	val.vsx32val[offset] = gpr32;
1044 	VCPU_VSX_VR(vcpu, index) = val.vval;
1045 }
1046 
1047 static inline void kvmppc_set_vmx_hword(struct kvm_vcpu *vcpu,
1048 	u16 gpr16)
1049 {
1050 	union kvmppc_one_reg val;
1051 	int offset = kvmppc_get_vmx_hword_offset(vcpu,
1052 			vcpu->arch.mmio_vmx_offset);
1053 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1054 
1055 	if (offset == -1)
1056 		return;
1057 
1058 	val.vval = VCPU_VSX_VR(vcpu, index);
1059 	val.vsx16val[offset] = gpr16;
1060 	VCPU_VSX_VR(vcpu, index) = val.vval;
1061 }
1062 
1063 static inline void kvmppc_set_vmx_byte(struct kvm_vcpu *vcpu,
1064 	u8 gpr8)
1065 {
1066 	union kvmppc_one_reg val;
1067 	int offset = kvmppc_get_vmx_byte_offset(vcpu,
1068 			vcpu->arch.mmio_vmx_offset);
1069 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1070 
1071 	if (offset == -1)
1072 		return;
1073 
1074 	val.vval = VCPU_VSX_VR(vcpu, index);
1075 	val.vsx8val[offset] = gpr8;
1076 	VCPU_VSX_VR(vcpu, index) = val.vval;
1077 }
1078 #endif /* CONFIG_ALTIVEC */
1079 
1080 #ifdef CONFIG_PPC_FPU
1081 static inline u64 sp_to_dp(u32 fprs)
1082 {
1083 	u64 fprd;
1084 
1085 	preempt_disable();
1086 	enable_kernel_fp();
1087 	asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs)
1088 	     : "fr0");
1089 	preempt_enable();
1090 	return fprd;
1091 }
1092 
1093 static inline u32 dp_to_sp(u64 fprd)
1094 {
1095 	u32 fprs;
1096 
1097 	preempt_disable();
1098 	enable_kernel_fp();
1099 	asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd)
1100 	     : "fr0");
1101 	preempt_enable();
1102 	return fprs;
1103 }
1104 
1105 #else
1106 #define sp_to_dp(x)	(x)
1107 #define dp_to_sp(x)	(x)
1108 #endif /* CONFIG_PPC_FPU */
1109 
1110 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
1111                                       struct kvm_run *run)
1112 {
1113 	u64 uninitialized_var(gpr);
1114 
1115 	if (run->mmio.len > sizeof(gpr)) {
1116 		printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
1117 		return;
1118 	}
1119 
1120 	if (!vcpu->arch.mmio_host_swabbed) {
1121 		switch (run->mmio.len) {
1122 		case 8: gpr = *(u64 *)run->mmio.data; break;
1123 		case 4: gpr = *(u32 *)run->mmio.data; break;
1124 		case 2: gpr = *(u16 *)run->mmio.data; break;
1125 		case 1: gpr = *(u8 *)run->mmio.data; break;
1126 		}
1127 	} else {
1128 		switch (run->mmio.len) {
1129 		case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
1130 		case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
1131 		case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
1132 		case 1: gpr = *(u8 *)run->mmio.data; break;
1133 		}
1134 	}
1135 
1136 	/* conversion between single and double precision */
1137 	if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
1138 		gpr = sp_to_dp(gpr);
1139 
1140 	if (vcpu->arch.mmio_sign_extend) {
1141 		switch (run->mmio.len) {
1142 #ifdef CONFIG_PPC64
1143 		case 4:
1144 			gpr = (s64)(s32)gpr;
1145 			break;
1146 #endif
1147 		case 2:
1148 			gpr = (s64)(s16)gpr;
1149 			break;
1150 		case 1:
1151 			gpr = (s64)(s8)gpr;
1152 			break;
1153 		}
1154 	}
1155 
1156 	switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1157 	case KVM_MMIO_REG_GPR:
1158 		kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1159 		break;
1160 	case KVM_MMIO_REG_FPR:
1161 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1162 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_FP);
1163 
1164 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1165 		break;
1166 #ifdef CONFIG_PPC_BOOK3S
1167 	case KVM_MMIO_REG_QPR:
1168 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1169 		break;
1170 	case KVM_MMIO_REG_FQPR:
1171 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1172 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1173 		break;
1174 #endif
1175 #ifdef CONFIG_VSX
1176 	case KVM_MMIO_REG_VSX:
1177 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1178 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VSX);
1179 
1180 		if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_DWORD)
1181 			kvmppc_set_vsr_dword(vcpu, gpr);
1182 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_WORD)
1183 			kvmppc_set_vsr_word(vcpu, gpr);
1184 		else if (vcpu->arch.mmio_copy_type ==
1185 				KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1186 			kvmppc_set_vsr_dword_dump(vcpu, gpr);
1187 		else if (vcpu->arch.mmio_copy_type ==
1188 				KVMPPC_VSX_COPY_WORD_LOAD_DUMP)
1189 			kvmppc_set_vsr_word_dump(vcpu, gpr);
1190 		break;
1191 #endif
1192 #ifdef CONFIG_ALTIVEC
1193 	case KVM_MMIO_REG_VMX:
1194 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1195 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VEC);
1196 
1197 		if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_DWORD)
1198 			kvmppc_set_vmx_dword(vcpu, gpr);
1199 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_WORD)
1200 			kvmppc_set_vmx_word(vcpu, gpr);
1201 		else if (vcpu->arch.mmio_copy_type ==
1202 				KVMPPC_VMX_COPY_HWORD)
1203 			kvmppc_set_vmx_hword(vcpu, gpr);
1204 		else if (vcpu->arch.mmio_copy_type ==
1205 				KVMPPC_VMX_COPY_BYTE)
1206 			kvmppc_set_vmx_byte(vcpu, gpr);
1207 		break;
1208 #endif
1209 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1210 	case KVM_MMIO_REG_NESTED_GPR:
1211 		if (kvmppc_need_byteswap(vcpu))
1212 			gpr = swab64(gpr);
1213 		kvm_vcpu_write_guest(vcpu, vcpu->arch.nested_io_gpr, &gpr,
1214 				     sizeof(gpr));
1215 		break;
1216 #endif
1217 	default:
1218 		BUG();
1219 	}
1220 }
1221 
1222 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1223 				unsigned int rt, unsigned int bytes,
1224 				int is_default_endian, int sign_extend)
1225 {
1226 	int idx, ret;
1227 	bool host_swabbed;
1228 
1229 	/* Pity C doesn't have a logical XOR operator */
1230 	if (kvmppc_need_byteswap(vcpu)) {
1231 		host_swabbed = is_default_endian;
1232 	} else {
1233 		host_swabbed = !is_default_endian;
1234 	}
1235 
1236 	if (bytes > sizeof(run->mmio.data)) {
1237 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1238 		       run->mmio.len);
1239 	}
1240 
1241 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1242 	run->mmio.len = bytes;
1243 	run->mmio.is_write = 0;
1244 
1245 	vcpu->arch.io_gpr = rt;
1246 	vcpu->arch.mmio_host_swabbed = host_swabbed;
1247 	vcpu->mmio_needed = 1;
1248 	vcpu->mmio_is_write = 0;
1249 	vcpu->arch.mmio_sign_extend = sign_extend;
1250 
1251 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1252 
1253 	ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1254 			      bytes, &run->mmio.data);
1255 
1256 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1257 
1258 	if (!ret) {
1259 		kvmppc_complete_mmio_load(vcpu, run);
1260 		vcpu->mmio_needed = 0;
1261 		return EMULATE_DONE;
1262 	}
1263 
1264 	return EMULATE_DO_MMIO;
1265 }
1266 
1267 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1268 		       unsigned int rt, unsigned int bytes,
1269 		       int is_default_endian)
1270 {
1271 	return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
1272 }
1273 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1274 
1275 /* Same as above, but sign extends */
1276 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
1277 			unsigned int rt, unsigned int bytes,
1278 			int is_default_endian)
1279 {
1280 	return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
1281 }
1282 
1283 #ifdef CONFIG_VSX
1284 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1285 			unsigned int rt, unsigned int bytes,
1286 			int is_default_endian, int mmio_sign_extend)
1287 {
1288 	enum emulation_result emulated = EMULATE_DONE;
1289 
1290 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1291 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1292 		return EMULATE_FAIL;
1293 
1294 	while (vcpu->arch.mmio_vsx_copy_nums) {
1295 		emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1296 			is_default_endian, mmio_sign_extend);
1297 
1298 		if (emulated != EMULATE_DONE)
1299 			break;
1300 
1301 		vcpu->arch.paddr_accessed += run->mmio.len;
1302 
1303 		vcpu->arch.mmio_vsx_copy_nums--;
1304 		vcpu->arch.mmio_vsx_offset++;
1305 	}
1306 	return emulated;
1307 }
1308 #endif /* CONFIG_VSX */
1309 
1310 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1311 			u64 val, unsigned int bytes, int is_default_endian)
1312 {
1313 	void *data = run->mmio.data;
1314 	int idx, ret;
1315 	bool host_swabbed;
1316 
1317 	/* Pity C doesn't have a logical XOR operator */
1318 	if (kvmppc_need_byteswap(vcpu)) {
1319 		host_swabbed = is_default_endian;
1320 	} else {
1321 		host_swabbed = !is_default_endian;
1322 	}
1323 
1324 	if (bytes > sizeof(run->mmio.data)) {
1325 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1326 		       run->mmio.len);
1327 	}
1328 
1329 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1330 	run->mmio.len = bytes;
1331 	run->mmio.is_write = 1;
1332 	vcpu->mmio_needed = 1;
1333 	vcpu->mmio_is_write = 1;
1334 
1335 	if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1336 		val = dp_to_sp(val);
1337 
1338 	/* Store the value at the lowest bytes in 'data'. */
1339 	if (!host_swabbed) {
1340 		switch (bytes) {
1341 		case 8: *(u64 *)data = val; break;
1342 		case 4: *(u32 *)data = val; break;
1343 		case 2: *(u16 *)data = val; break;
1344 		case 1: *(u8  *)data = val; break;
1345 		}
1346 	} else {
1347 		switch (bytes) {
1348 		case 8: *(u64 *)data = swab64(val); break;
1349 		case 4: *(u32 *)data = swab32(val); break;
1350 		case 2: *(u16 *)data = swab16(val); break;
1351 		case 1: *(u8  *)data = val; break;
1352 		}
1353 	}
1354 
1355 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1356 
1357 	ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1358 			       bytes, &run->mmio.data);
1359 
1360 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1361 
1362 	if (!ret) {
1363 		vcpu->mmio_needed = 0;
1364 		return EMULATE_DONE;
1365 	}
1366 
1367 	return EMULATE_DO_MMIO;
1368 }
1369 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1370 
1371 #ifdef CONFIG_VSX
1372 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1373 {
1374 	u32 dword_offset, word_offset;
1375 	union kvmppc_one_reg reg;
1376 	int vsx_offset = 0;
1377 	int copy_type = vcpu->arch.mmio_copy_type;
1378 	int result = 0;
1379 
1380 	switch (copy_type) {
1381 	case KVMPPC_VSX_COPY_DWORD:
1382 		vsx_offset =
1383 			kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1384 
1385 		if (vsx_offset == -1) {
1386 			result = -1;
1387 			break;
1388 		}
1389 
1390 		if (rs < 32) {
1391 			*val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1392 		} else {
1393 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1394 			*val = reg.vsxval[vsx_offset];
1395 		}
1396 		break;
1397 
1398 	case KVMPPC_VSX_COPY_WORD:
1399 		vsx_offset =
1400 			kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1401 
1402 		if (vsx_offset == -1) {
1403 			result = -1;
1404 			break;
1405 		}
1406 
1407 		if (rs < 32) {
1408 			dword_offset = vsx_offset / 2;
1409 			word_offset = vsx_offset % 2;
1410 			reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1411 			*val = reg.vsx32val[word_offset];
1412 		} else {
1413 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1414 			*val = reg.vsx32val[vsx_offset];
1415 		}
1416 		break;
1417 
1418 	default:
1419 		result = -1;
1420 		break;
1421 	}
1422 
1423 	return result;
1424 }
1425 
1426 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1427 			int rs, unsigned int bytes, int is_default_endian)
1428 {
1429 	u64 val;
1430 	enum emulation_result emulated = EMULATE_DONE;
1431 
1432 	vcpu->arch.io_gpr = rs;
1433 
1434 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1435 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1436 		return EMULATE_FAIL;
1437 
1438 	while (vcpu->arch.mmio_vsx_copy_nums) {
1439 		if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1440 			return EMULATE_FAIL;
1441 
1442 		emulated = kvmppc_handle_store(run, vcpu,
1443 			 val, bytes, is_default_endian);
1444 
1445 		if (emulated != EMULATE_DONE)
1446 			break;
1447 
1448 		vcpu->arch.paddr_accessed += run->mmio.len;
1449 
1450 		vcpu->arch.mmio_vsx_copy_nums--;
1451 		vcpu->arch.mmio_vsx_offset++;
1452 	}
1453 
1454 	return emulated;
1455 }
1456 
1457 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu,
1458 			struct kvm_run *run)
1459 {
1460 	enum emulation_result emulated = EMULATE_FAIL;
1461 	int r;
1462 
1463 	vcpu->arch.paddr_accessed += run->mmio.len;
1464 
1465 	if (!vcpu->mmio_is_write) {
1466 		emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr,
1467 			 run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1468 	} else {
1469 		emulated = kvmppc_handle_vsx_store(run, vcpu,
1470 			 vcpu->arch.io_gpr, run->mmio.len, 1);
1471 	}
1472 
1473 	switch (emulated) {
1474 	case EMULATE_DO_MMIO:
1475 		run->exit_reason = KVM_EXIT_MMIO;
1476 		r = RESUME_HOST;
1477 		break;
1478 	case EMULATE_FAIL:
1479 		pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1480 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1481 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1482 		r = RESUME_HOST;
1483 		break;
1484 	default:
1485 		r = RESUME_GUEST;
1486 		break;
1487 	}
1488 	return r;
1489 }
1490 #endif /* CONFIG_VSX */
1491 
1492 #ifdef CONFIG_ALTIVEC
1493 int kvmppc_handle_vmx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1494 		unsigned int rt, unsigned int bytes, int is_default_endian)
1495 {
1496 	enum emulation_result emulated = EMULATE_DONE;
1497 
1498 	if (vcpu->arch.mmio_vsx_copy_nums > 2)
1499 		return EMULATE_FAIL;
1500 
1501 	while (vcpu->arch.mmio_vmx_copy_nums) {
1502 		emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1503 				is_default_endian, 0);
1504 
1505 		if (emulated != EMULATE_DONE)
1506 			break;
1507 
1508 		vcpu->arch.paddr_accessed += run->mmio.len;
1509 		vcpu->arch.mmio_vmx_copy_nums--;
1510 		vcpu->arch.mmio_vmx_offset++;
1511 	}
1512 
1513 	return emulated;
1514 }
1515 
1516 int kvmppc_get_vmx_dword(struct kvm_vcpu *vcpu, int index, u64 *val)
1517 {
1518 	union kvmppc_one_reg reg;
1519 	int vmx_offset = 0;
1520 	int result = 0;
1521 
1522 	vmx_offset =
1523 		kvmppc_get_vmx_dword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1524 
1525 	if (vmx_offset == -1)
1526 		return -1;
1527 
1528 	reg.vval = VCPU_VSX_VR(vcpu, index);
1529 	*val = reg.vsxval[vmx_offset];
1530 
1531 	return result;
1532 }
1533 
1534 int kvmppc_get_vmx_word(struct kvm_vcpu *vcpu, int index, u64 *val)
1535 {
1536 	union kvmppc_one_reg reg;
1537 	int vmx_offset = 0;
1538 	int result = 0;
1539 
1540 	vmx_offset =
1541 		kvmppc_get_vmx_word_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1542 
1543 	if (vmx_offset == -1)
1544 		return -1;
1545 
1546 	reg.vval = VCPU_VSX_VR(vcpu, index);
1547 	*val = reg.vsx32val[vmx_offset];
1548 
1549 	return result;
1550 }
1551 
1552 int kvmppc_get_vmx_hword(struct kvm_vcpu *vcpu, int index, u64 *val)
1553 {
1554 	union kvmppc_one_reg reg;
1555 	int vmx_offset = 0;
1556 	int result = 0;
1557 
1558 	vmx_offset =
1559 		kvmppc_get_vmx_hword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1560 
1561 	if (vmx_offset == -1)
1562 		return -1;
1563 
1564 	reg.vval = VCPU_VSX_VR(vcpu, index);
1565 	*val = reg.vsx16val[vmx_offset];
1566 
1567 	return result;
1568 }
1569 
1570 int kvmppc_get_vmx_byte(struct kvm_vcpu *vcpu, int index, u64 *val)
1571 {
1572 	union kvmppc_one_reg reg;
1573 	int vmx_offset = 0;
1574 	int result = 0;
1575 
1576 	vmx_offset =
1577 		kvmppc_get_vmx_byte_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1578 
1579 	if (vmx_offset == -1)
1580 		return -1;
1581 
1582 	reg.vval = VCPU_VSX_VR(vcpu, index);
1583 	*val = reg.vsx8val[vmx_offset];
1584 
1585 	return result;
1586 }
1587 
1588 int kvmppc_handle_vmx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1589 		unsigned int rs, unsigned int bytes, int is_default_endian)
1590 {
1591 	u64 val = 0;
1592 	unsigned int index = rs & KVM_MMIO_REG_MASK;
1593 	enum emulation_result emulated = EMULATE_DONE;
1594 
1595 	if (vcpu->arch.mmio_vsx_copy_nums > 2)
1596 		return EMULATE_FAIL;
1597 
1598 	vcpu->arch.io_gpr = rs;
1599 
1600 	while (vcpu->arch.mmio_vmx_copy_nums) {
1601 		switch (vcpu->arch.mmio_copy_type) {
1602 		case KVMPPC_VMX_COPY_DWORD:
1603 			if (kvmppc_get_vmx_dword(vcpu, index, &val) == -1)
1604 				return EMULATE_FAIL;
1605 
1606 			break;
1607 		case KVMPPC_VMX_COPY_WORD:
1608 			if (kvmppc_get_vmx_word(vcpu, index, &val) == -1)
1609 				return EMULATE_FAIL;
1610 			break;
1611 		case KVMPPC_VMX_COPY_HWORD:
1612 			if (kvmppc_get_vmx_hword(vcpu, index, &val) == -1)
1613 				return EMULATE_FAIL;
1614 			break;
1615 		case KVMPPC_VMX_COPY_BYTE:
1616 			if (kvmppc_get_vmx_byte(vcpu, index, &val) == -1)
1617 				return EMULATE_FAIL;
1618 			break;
1619 		default:
1620 			return EMULATE_FAIL;
1621 		}
1622 
1623 		emulated = kvmppc_handle_store(run, vcpu, val, bytes,
1624 				is_default_endian);
1625 		if (emulated != EMULATE_DONE)
1626 			break;
1627 
1628 		vcpu->arch.paddr_accessed += run->mmio.len;
1629 		vcpu->arch.mmio_vmx_copy_nums--;
1630 		vcpu->arch.mmio_vmx_offset++;
1631 	}
1632 
1633 	return emulated;
1634 }
1635 
1636 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu,
1637 		struct kvm_run *run)
1638 {
1639 	enum emulation_result emulated = EMULATE_FAIL;
1640 	int r;
1641 
1642 	vcpu->arch.paddr_accessed += run->mmio.len;
1643 
1644 	if (!vcpu->mmio_is_write) {
1645 		emulated = kvmppc_handle_vmx_load(run, vcpu,
1646 				vcpu->arch.io_gpr, run->mmio.len, 1);
1647 	} else {
1648 		emulated = kvmppc_handle_vmx_store(run, vcpu,
1649 				vcpu->arch.io_gpr, run->mmio.len, 1);
1650 	}
1651 
1652 	switch (emulated) {
1653 	case EMULATE_DO_MMIO:
1654 		run->exit_reason = KVM_EXIT_MMIO;
1655 		r = RESUME_HOST;
1656 		break;
1657 	case EMULATE_FAIL:
1658 		pr_info("KVM: MMIO emulation failed (VMX repeat)\n");
1659 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1660 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1661 		r = RESUME_HOST;
1662 		break;
1663 	default:
1664 		r = RESUME_GUEST;
1665 		break;
1666 	}
1667 	return r;
1668 }
1669 #endif /* CONFIG_ALTIVEC */
1670 
1671 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1672 {
1673 	int r = 0;
1674 	union kvmppc_one_reg val;
1675 	int size;
1676 
1677 	size = one_reg_size(reg->id);
1678 	if (size > sizeof(val))
1679 		return -EINVAL;
1680 
1681 	r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1682 	if (r == -EINVAL) {
1683 		r = 0;
1684 		switch (reg->id) {
1685 #ifdef CONFIG_ALTIVEC
1686 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1687 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1688 				r = -ENXIO;
1689 				break;
1690 			}
1691 			val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1692 			break;
1693 		case KVM_REG_PPC_VSCR:
1694 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1695 				r = -ENXIO;
1696 				break;
1697 			}
1698 			val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1699 			break;
1700 		case KVM_REG_PPC_VRSAVE:
1701 			val = get_reg_val(reg->id, vcpu->arch.vrsave);
1702 			break;
1703 #endif /* CONFIG_ALTIVEC */
1704 		default:
1705 			r = -EINVAL;
1706 			break;
1707 		}
1708 	}
1709 
1710 	if (r)
1711 		return r;
1712 
1713 	if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1714 		r = -EFAULT;
1715 
1716 	return r;
1717 }
1718 
1719 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1720 {
1721 	int r;
1722 	union kvmppc_one_reg val;
1723 	int size;
1724 
1725 	size = one_reg_size(reg->id);
1726 	if (size > sizeof(val))
1727 		return -EINVAL;
1728 
1729 	if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1730 		return -EFAULT;
1731 
1732 	r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1733 	if (r == -EINVAL) {
1734 		r = 0;
1735 		switch (reg->id) {
1736 #ifdef CONFIG_ALTIVEC
1737 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1738 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1739 				r = -ENXIO;
1740 				break;
1741 			}
1742 			vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1743 			break;
1744 		case KVM_REG_PPC_VSCR:
1745 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1746 				r = -ENXIO;
1747 				break;
1748 			}
1749 			vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1750 			break;
1751 		case KVM_REG_PPC_VRSAVE:
1752 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1753 				r = -ENXIO;
1754 				break;
1755 			}
1756 			vcpu->arch.vrsave = set_reg_val(reg->id, val);
1757 			break;
1758 #endif /* CONFIG_ALTIVEC */
1759 		default:
1760 			r = -EINVAL;
1761 			break;
1762 		}
1763 	}
1764 
1765 	return r;
1766 }
1767 
1768 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
1769 {
1770 	struct kvm_run *run = vcpu->run;
1771 	int r;
1772 
1773 	vcpu_load(vcpu);
1774 
1775 	if (vcpu->mmio_needed) {
1776 		vcpu->mmio_needed = 0;
1777 		if (!vcpu->mmio_is_write)
1778 			kvmppc_complete_mmio_load(vcpu, run);
1779 #ifdef CONFIG_VSX
1780 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1781 			vcpu->arch.mmio_vsx_copy_nums--;
1782 			vcpu->arch.mmio_vsx_offset++;
1783 		}
1784 
1785 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1786 			r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run);
1787 			if (r == RESUME_HOST) {
1788 				vcpu->mmio_needed = 1;
1789 				goto out;
1790 			}
1791 		}
1792 #endif
1793 #ifdef CONFIG_ALTIVEC
1794 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1795 			vcpu->arch.mmio_vmx_copy_nums--;
1796 			vcpu->arch.mmio_vmx_offset++;
1797 		}
1798 
1799 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1800 			r = kvmppc_emulate_mmio_vmx_loadstore(vcpu, run);
1801 			if (r == RESUME_HOST) {
1802 				vcpu->mmio_needed = 1;
1803 				goto out;
1804 			}
1805 		}
1806 #endif
1807 	} else if (vcpu->arch.osi_needed) {
1808 		u64 *gprs = run->osi.gprs;
1809 		int i;
1810 
1811 		for (i = 0; i < 32; i++)
1812 			kvmppc_set_gpr(vcpu, i, gprs[i]);
1813 		vcpu->arch.osi_needed = 0;
1814 	} else if (vcpu->arch.hcall_needed) {
1815 		int i;
1816 
1817 		kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1818 		for (i = 0; i < 9; ++i)
1819 			kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1820 		vcpu->arch.hcall_needed = 0;
1821 #ifdef CONFIG_BOOKE
1822 	} else if (vcpu->arch.epr_needed) {
1823 		kvmppc_set_epr(vcpu, run->epr.epr);
1824 		vcpu->arch.epr_needed = 0;
1825 #endif
1826 	}
1827 
1828 	kvm_sigset_activate(vcpu);
1829 
1830 	if (run->immediate_exit)
1831 		r = -EINTR;
1832 	else
1833 		r = kvmppc_vcpu_run(run, vcpu);
1834 
1835 	kvm_sigset_deactivate(vcpu);
1836 
1837 #ifdef CONFIG_ALTIVEC
1838 out:
1839 #endif
1840 	vcpu_put(vcpu);
1841 	return r;
1842 }
1843 
1844 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1845 {
1846 	if (irq->irq == KVM_INTERRUPT_UNSET) {
1847 		kvmppc_core_dequeue_external(vcpu);
1848 		return 0;
1849 	}
1850 
1851 	kvmppc_core_queue_external(vcpu, irq);
1852 
1853 	kvm_vcpu_kick(vcpu);
1854 
1855 	return 0;
1856 }
1857 
1858 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1859 				     struct kvm_enable_cap *cap)
1860 {
1861 	int r;
1862 
1863 	if (cap->flags)
1864 		return -EINVAL;
1865 
1866 	switch (cap->cap) {
1867 	case KVM_CAP_PPC_OSI:
1868 		r = 0;
1869 		vcpu->arch.osi_enabled = true;
1870 		break;
1871 	case KVM_CAP_PPC_PAPR:
1872 		r = 0;
1873 		vcpu->arch.papr_enabled = true;
1874 		break;
1875 	case KVM_CAP_PPC_EPR:
1876 		r = 0;
1877 		if (cap->args[0])
1878 			vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1879 		else
1880 			vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1881 		break;
1882 #ifdef CONFIG_BOOKE
1883 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
1884 		r = 0;
1885 		vcpu->arch.watchdog_enabled = true;
1886 		break;
1887 #endif
1888 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1889 	case KVM_CAP_SW_TLB: {
1890 		struct kvm_config_tlb cfg;
1891 		void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1892 
1893 		r = -EFAULT;
1894 		if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1895 			break;
1896 
1897 		r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1898 		break;
1899 	}
1900 #endif
1901 #ifdef CONFIG_KVM_MPIC
1902 	case KVM_CAP_IRQ_MPIC: {
1903 		struct fd f;
1904 		struct kvm_device *dev;
1905 
1906 		r = -EBADF;
1907 		f = fdget(cap->args[0]);
1908 		if (!f.file)
1909 			break;
1910 
1911 		r = -EPERM;
1912 		dev = kvm_device_from_filp(f.file);
1913 		if (dev)
1914 			r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1915 
1916 		fdput(f);
1917 		break;
1918 	}
1919 #endif
1920 #ifdef CONFIG_KVM_XICS
1921 	case KVM_CAP_IRQ_XICS: {
1922 		struct fd f;
1923 		struct kvm_device *dev;
1924 
1925 		r = -EBADF;
1926 		f = fdget(cap->args[0]);
1927 		if (!f.file)
1928 			break;
1929 
1930 		r = -EPERM;
1931 		dev = kvm_device_from_filp(f.file);
1932 		if (dev) {
1933 			if (xics_on_xive())
1934 				r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1935 			else
1936 				r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1937 		}
1938 
1939 		fdput(f);
1940 		break;
1941 	}
1942 #endif /* CONFIG_KVM_XICS */
1943 #ifdef CONFIG_KVM_XIVE
1944 	case KVM_CAP_PPC_IRQ_XIVE: {
1945 		struct fd f;
1946 		struct kvm_device *dev;
1947 
1948 		r = -EBADF;
1949 		f = fdget(cap->args[0]);
1950 		if (!f.file)
1951 			break;
1952 
1953 		r = -ENXIO;
1954 		if (!xive_enabled())
1955 			break;
1956 
1957 		r = -EPERM;
1958 		dev = kvm_device_from_filp(f.file);
1959 		if (dev)
1960 			r = kvmppc_xive_native_connect_vcpu(dev, vcpu,
1961 							    cap->args[1]);
1962 
1963 		fdput(f);
1964 		break;
1965 	}
1966 #endif /* CONFIG_KVM_XIVE */
1967 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1968 	case KVM_CAP_PPC_FWNMI:
1969 		r = -EINVAL;
1970 		if (!is_kvmppc_hv_enabled(vcpu->kvm))
1971 			break;
1972 		r = 0;
1973 		vcpu->kvm->arch.fwnmi_enabled = true;
1974 		break;
1975 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1976 	default:
1977 		r = -EINVAL;
1978 		break;
1979 	}
1980 
1981 	if (!r)
1982 		r = kvmppc_sanity_check(vcpu);
1983 
1984 	return r;
1985 }
1986 
1987 bool kvm_arch_intc_initialized(struct kvm *kvm)
1988 {
1989 #ifdef CONFIG_KVM_MPIC
1990 	if (kvm->arch.mpic)
1991 		return true;
1992 #endif
1993 #ifdef CONFIG_KVM_XICS
1994 	if (kvm->arch.xics || kvm->arch.xive)
1995 		return true;
1996 #endif
1997 	return false;
1998 }
1999 
2000 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2001                                     struct kvm_mp_state *mp_state)
2002 {
2003 	return -EINVAL;
2004 }
2005 
2006 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2007                                     struct kvm_mp_state *mp_state)
2008 {
2009 	return -EINVAL;
2010 }
2011 
2012 long kvm_arch_vcpu_async_ioctl(struct file *filp,
2013 			       unsigned int ioctl, unsigned long arg)
2014 {
2015 	struct kvm_vcpu *vcpu = filp->private_data;
2016 	void __user *argp = (void __user *)arg;
2017 
2018 	if (ioctl == KVM_INTERRUPT) {
2019 		struct kvm_interrupt irq;
2020 		if (copy_from_user(&irq, argp, sizeof(irq)))
2021 			return -EFAULT;
2022 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2023 	}
2024 	return -ENOIOCTLCMD;
2025 }
2026 
2027 long kvm_arch_vcpu_ioctl(struct file *filp,
2028                          unsigned int ioctl, unsigned long arg)
2029 {
2030 	struct kvm_vcpu *vcpu = filp->private_data;
2031 	void __user *argp = (void __user *)arg;
2032 	long r;
2033 
2034 	switch (ioctl) {
2035 	case KVM_ENABLE_CAP:
2036 	{
2037 		struct kvm_enable_cap cap;
2038 		r = -EFAULT;
2039 		vcpu_load(vcpu);
2040 		if (copy_from_user(&cap, argp, sizeof(cap)))
2041 			goto out;
2042 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
2043 		vcpu_put(vcpu);
2044 		break;
2045 	}
2046 
2047 	case KVM_SET_ONE_REG:
2048 	case KVM_GET_ONE_REG:
2049 	{
2050 		struct kvm_one_reg reg;
2051 		r = -EFAULT;
2052 		if (copy_from_user(&reg, argp, sizeof(reg)))
2053 			goto out;
2054 		if (ioctl == KVM_SET_ONE_REG)
2055 			r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
2056 		else
2057 			r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
2058 		break;
2059 	}
2060 
2061 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
2062 	case KVM_DIRTY_TLB: {
2063 		struct kvm_dirty_tlb dirty;
2064 		r = -EFAULT;
2065 		vcpu_load(vcpu);
2066 		if (copy_from_user(&dirty, argp, sizeof(dirty)))
2067 			goto out;
2068 		r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
2069 		vcpu_put(vcpu);
2070 		break;
2071 	}
2072 #endif
2073 	default:
2074 		r = -EINVAL;
2075 	}
2076 
2077 out:
2078 	return r;
2079 }
2080 
2081 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2082 {
2083 	return VM_FAULT_SIGBUS;
2084 }
2085 
2086 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
2087 {
2088 	u32 inst_nop = 0x60000000;
2089 #ifdef CONFIG_KVM_BOOKE_HV
2090 	u32 inst_sc1 = 0x44000022;
2091 	pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
2092 	pvinfo->hcall[1] = cpu_to_be32(inst_nop);
2093 	pvinfo->hcall[2] = cpu_to_be32(inst_nop);
2094 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2095 #else
2096 	u32 inst_lis = 0x3c000000;
2097 	u32 inst_ori = 0x60000000;
2098 	u32 inst_sc = 0x44000002;
2099 	u32 inst_imm_mask = 0xffff;
2100 
2101 	/*
2102 	 * The hypercall to get into KVM from within guest context is as
2103 	 * follows:
2104 	 *
2105 	 *    lis r0, r0, KVM_SC_MAGIC_R0@h
2106 	 *    ori r0, KVM_SC_MAGIC_R0@l
2107 	 *    sc
2108 	 *    nop
2109 	 */
2110 	pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
2111 	pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
2112 	pvinfo->hcall[2] = cpu_to_be32(inst_sc);
2113 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2114 #endif
2115 
2116 	pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
2117 
2118 	return 0;
2119 }
2120 
2121 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
2122 			  bool line_status)
2123 {
2124 	if (!irqchip_in_kernel(kvm))
2125 		return -ENXIO;
2126 
2127 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2128 					irq_event->irq, irq_event->level,
2129 					line_status);
2130 	return 0;
2131 }
2132 
2133 
2134 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
2135 			    struct kvm_enable_cap *cap)
2136 {
2137 	int r;
2138 
2139 	if (cap->flags)
2140 		return -EINVAL;
2141 
2142 	switch (cap->cap) {
2143 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
2144 	case KVM_CAP_PPC_ENABLE_HCALL: {
2145 		unsigned long hcall = cap->args[0];
2146 
2147 		r = -EINVAL;
2148 		if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
2149 		    cap->args[1] > 1)
2150 			break;
2151 		if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
2152 			break;
2153 		if (cap->args[1])
2154 			set_bit(hcall / 4, kvm->arch.enabled_hcalls);
2155 		else
2156 			clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
2157 		r = 0;
2158 		break;
2159 	}
2160 	case KVM_CAP_PPC_SMT: {
2161 		unsigned long mode = cap->args[0];
2162 		unsigned long flags = cap->args[1];
2163 
2164 		r = -EINVAL;
2165 		if (kvm->arch.kvm_ops->set_smt_mode)
2166 			r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
2167 		break;
2168 	}
2169 
2170 	case KVM_CAP_PPC_NESTED_HV:
2171 		r = -EINVAL;
2172 		if (!is_kvmppc_hv_enabled(kvm) ||
2173 		    !kvm->arch.kvm_ops->enable_nested)
2174 			break;
2175 		r = kvm->arch.kvm_ops->enable_nested(kvm);
2176 		break;
2177 #endif
2178 #if defined(CONFIG_KVM_BOOK3S_HV_POSSIBLE)
2179 	case KVM_CAP_PPC_SECURE_GUEST:
2180 		r = -EINVAL;
2181 		if (!is_kvmppc_hv_enabled(kvm) || !kvm->arch.kvm_ops->enable_svm)
2182 			break;
2183 		r = kvm->arch.kvm_ops->enable_svm(kvm);
2184 		break;
2185 #endif
2186 	default:
2187 		r = -EINVAL;
2188 		break;
2189 	}
2190 
2191 	return r;
2192 }
2193 
2194 #ifdef CONFIG_PPC_BOOK3S_64
2195 /*
2196  * These functions check whether the underlying hardware is safe
2197  * against attacks based on observing the effects of speculatively
2198  * executed instructions, and whether it supplies instructions for
2199  * use in workarounds.  The information comes from firmware, either
2200  * via the device tree on powernv platforms or from an hcall on
2201  * pseries platforms.
2202  */
2203 #ifdef CONFIG_PPC_PSERIES
2204 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2205 {
2206 	struct h_cpu_char_result c;
2207 	unsigned long rc;
2208 
2209 	if (!machine_is(pseries))
2210 		return -ENOTTY;
2211 
2212 	rc = plpar_get_cpu_characteristics(&c);
2213 	if (rc == H_SUCCESS) {
2214 		cp->character = c.character;
2215 		cp->behaviour = c.behaviour;
2216 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2217 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2218 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2219 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2220 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2221 			KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
2222 			KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
2223 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2224 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2225 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2226 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2227 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2228 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2229 	}
2230 	return 0;
2231 }
2232 #else
2233 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2234 {
2235 	return -ENOTTY;
2236 }
2237 #endif
2238 
2239 static inline bool have_fw_feat(struct device_node *fw_features,
2240 				const char *state, const char *name)
2241 {
2242 	struct device_node *np;
2243 	bool r = false;
2244 
2245 	np = of_get_child_by_name(fw_features, name);
2246 	if (np) {
2247 		r = of_property_read_bool(np, state);
2248 		of_node_put(np);
2249 	}
2250 	return r;
2251 }
2252 
2253 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2254 {
2255 	struct device_node *np, *fw_features;
2256 	int r;
2257 
2258 	memset(cp, 0, sizeof(*cp));
2259 	r = pseries_get_cpu_char(cp);
2260 	if (r != -ENOTTY)
2261 		return r;
2262 
2263 	np = of_find_node_by_name(NULL, "ibm,opal");
2264 	if (np) {
2265 		fw_features = of_get_child_by_name(np, "fw-features");
2266 		of_node_put(np);
2267 		if (!fw_features)
2268 			return 0;
2269 		if (have_fw_feat(fw_features, "enabled",
2270 				 "inst-spec-barrier-ori31,31,0"))
2271 			cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
2272 		if (have_fw_feat(fw_features, "enabled",
2273 				 "fw-bcctrl-serialized"))
2274 			cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
2275 		if (have_fw_feat(fw_features, "enabled",
2276 				 "inst-l1d-flush-ori30,30,0"))
2277 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
2278 		if (have_fw_feat(fw_features, "enabled",
2279 				 "inst-l1d-flush-trig2"))
2280 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
2281 		if (have_fw_feat(fw_features, "enabled",
2282 				 "fw-l1d-thread-split"))
2283 			cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
2284 		if (have_fw_feat(fw_features, "enabled",
2285 				 "fw-count-cache-disabled"))
2286 			cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2287 		if (have_fw_feat(fw_features, "enabled",
2288 				 "fw-count-cache-flush-bcctr2,0,0"))
2289 			cp->character |= KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2290 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2291 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2292 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2293 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2294 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2295 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2296 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2297 
2298 		if (have_fw_feat(fw_features, "enabled",
2299 				 "speculation-policy-favor-security"))
2300 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
2301 		if (!have_fw_feat(fw_features, "disabled",
2302 				  "needs-l1d-flush-msr-pr-0-to-1"))
2303 			cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
2304 		if (!have_fw_feat(fw_features, "disabled",
2305 				  "needs-spec-barrier-for-bound-checks"))
2306 			cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2307 		if (have_fw_feat(fw_features, "enabled",
2308 				 "needs-count-cache-flush-on-context-switch"))
2309 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2310 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2311 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2312 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2313 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2314 
2315 		of_node_put(fw_features);
2316 	}
2317 
2318 	return 0;
2319 }
2320 #endif
2321 
2322 long kvm_arch_vm_ioctl(struct file *filp,
2323                        unsigned int ioctl, unsigned long arg)
2324 {
2325 	struct kvm *kvm __maybe_unused = filp->private_data;
2326 	void __user *argp = (void __user *)arg;
2327 	long r;
2328 
2329 	switch (ioctl) {
2330 	case KVM_PPC_GET_PVINFO: {
2331 		struct kvm_ppc_pvinfo pvinfo;
2332 		memset(&pvinfo, 0, sizeof(pvinfo));
2333 		r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
2334 		if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
2335 			r = -EFAULT;
2336 			goto out;
2337 		}
2338 
2339 		break;
2340 	}
2341 #ifdef CONFIG_SPAPR_TCE_IOMMU
2342 	case KVM_CREATE_SPAPR_TCE_64: {
2343 		struct kvm_create_spapr_tce_64 create_tce_64;
2344 
2345 		r = -EFAULT;
2346 		if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
2347 			goto out;
2348 		if (create_tce_64.flags) {
2349 			r = -EINVAL;
2350 			goto out;
2351 		}
2352 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2353 		goto out;
2354 	}
2355 	case KVM_CREATE_SPAPR_TCE: {
2356 		struct kvm_create_spapr_tce create_tce;
2357 		struct kvm_create_spapr_tce_64 create_tce_64;
2358 
2359 		r = -EFAULT;
2360 		if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
2361 			goto out;
2362 
2363 		create_tce_64.liobn = create_tce.liobn;
2364 		create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
2365 		create_tce_64.offset = 0;
2366 		create_tce_64.size = create_tce.window_size >>
2367 				IOMMU_PAGE_SHIFT_4K;
2368 		create_tce_64.flags = 0;
2369 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2370 		goto out;
2371 	}
2372 #endif
2373 #ifdef CONFIG_PPC_BOOK3S_64
2374 	case KVM_PPC_GET_SMMU_INFO: {
2375 		struct kvm_ppc_smmu_info info;
2376 		struct kvm *kvm = filp->private_data;
2377 
2378 		memset(&info, 0, sizeof(info));
2379 		r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
2380 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2381 			r = -EFAULT;
2382 		break;
2383 	}
2384 	case KVM_PPC_RTAS_DEFINE_TOKEN: {
2385 		struct kvm *kvm = filp->private_data;
2386 
2387 		r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
2388 		break;
2389 	}
2390 	case KVM_PPC_CONFIGURE_V3_MMU: {
2391 		struct kvm *kvm = filp->private_data;
2392 		struct kvm_ppc_mmuv3_cfg cfg;
2393 
2394 		r = -EINVAL;
2395 		if (!kvm->arch.kvm_ops->configure_mmu)
2396 			goto out;
2397 		r = -EFAULT;
2398 		if (copy_from_user(&cfg, argp, sizeof(cfg)))
2399 			goto out;
2400 		r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
2401 		break;
2402 	}
2403 	case KVM_PPC_GET_RMMU_INFO: {
2404 		struct kvm *kvm = filp->private_data;
2405 		struct kvm_ppc_rmmu_info info;
2406 
2407 		r = -EINVAL;
2408 		if (!kvm->arch.kvm_ops->get_rmmu_info)
2409 			goto out;
2410 		r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
2411 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2412 			r = -EFAULT;
2413 		break;
2414 	}
2415 	case KVM_PPC_GET_CPU_CHAR: {
2416 		struct kvm_ppc_cpu_char cpuchar;
2417 
2418 		r = kvmppc_get_cpu_char(&cpuchar);
2419 		if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2420 			r = -EFAULT;
2421 		break;
2422 	}
2423 	case KVM_PPC_SVM_OFF: {
2424 		struct kvm *kvm = filp->private_data;
2425 
2426 		r = 0;
2427 		if (!kvm->arch.kvm_ops->svm_off)
2428 			goto out;
2429 
2430 		r = kvm->arch.kvm_ops->svm_off(kvm);
2431 		break;
2432 	}
2433 	default: {
2434 		struct kvm *kvm = filp->private_data;
2435 		r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
2436 	}
2437 #else /* CONFIG_PPC_BOOK3S_64 */
2438 	default:
2439 		r = -ENOTTY;
2440 #endif
2441 	}
2442 out:
2443 	return r;
2444 }
2445 
2446 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
2447 static unsigned long nr_lpids;
2448 
2449 long kvmppc_alloc_lpid(void)
2450 {
2451 	long lpid;
2452 
2453 	do {
2454 		lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
2455 		if (lpid >= nr_lpids) {
2456 			pr_err("%s: No LPIDs free\n", __func__);
2457 			return -ENOMEM;
2458 		}
2459 	} while (test_and_set_bit(lpid, lpid_inuse));
2460 
2461 	return lpid;
2462 }
2463 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2464 
2465 void kvmppc_claim_lpid(long lpid)
2466 {
2467 	set_bit(lpid, lpid_inuse);
2468 }
2469 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
2470 
2471 void kvmppc_free_lpid(long lpid)
2472 {
2473 	clear_bit(lpid, lpid_inuse);
2474 }
2475 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2476 
2477 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2478 {
2479 	nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
2480 	memset(lpid_inuse, 0, sizeof(lpid_inuse));
2481 }
2482 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2483 
2484 int kvm_arch_init(void *opaque)
2485 {
2486 	return 0;
2487 }
2488 
2489 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);
2490