xref: /openbmc/linux/arch/powerpc/kvm/powerpc.c (revision 15e3ae36)
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 		r = 1;
525 		break;
526 	case KVM_CAP_PPC_GUEST_DEBUG_SSTEP:
527 	case KVM_CAP_PPC_PAIRED_SINGLES:
528 	case KVM_CAP_PPC_OSI:
529 	case KVM_CAP_PPC_GET_PVINFO:
530 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
531 	case KVM_CAP_SW_TLB:
532 #endif
533 		/* We support this only for PR */
534 		r = !hv_enabled;
535 		break;
536 #ifdef CONFIG_KVM_MPIC
537 	case KVM_CAP_IRQ_MPIC:
538 		r = 1;
539 		break;
540 #endif
541 
542 #ifdef CONFIG_PPC_BOOK3S_64
543 	case KVM_CAP_SPAPR_TCE:
544 	case KVM_CAP_SPAPR_TCE_64:
545 		r = 1;
546 		break;
547 	case KVM_CAP_SPAPR_TCE_VFIO:
548 		r = !!cpu_has_feature(CPU_FTR_HVMODE);
549 		break;
550 	case KVM_CAP_PPC_RTAS:
551 	case KVM_CAP_PPC_FIXUP_HCALL:
552 	case KVM_CAP_PPC_ENABLE_HCALL:
553 #ifdef CONFIG_KVM_XICS
554 	case KVM_CAP_IRQ_XICS:
555 #endif
556 	case KVM_CAP_PPC_GET_CPU_CHAR:
557 		r = 1;
558 		break;
559 #ifdef CONFIG_KVM_XIVE
560 	case KVM_CAP_PPC_IRQ_XIVE:
561 		/*
562 		 * We need XIVE to be enabled on the platform (implies
563 		 * a POWER9 processor) and the PowerNV platform, as
564 		 * nested is not yet supported.
565 		 */
566 		r = xive_enabled() && !!cpu_has_feature(CPU_FTR_HVMODE) &&
567 			kvmppc_xive_native_supported();
568 		break;
569 #endif
570 
571 	case KVM_CAP_PPC_ALLOC_HTAB:
572 		r = hv_enabled;
573 		break;
574 #endif /* CONFIG_PPC_BOOK3S_64 */
575 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
576 	case KVM_CAP_PPC_SMT:
577 		r = 0;
578 		if (kvm) {
579 			if (kvm->arch.emul_smt_mode > 1)
580 				r = kvm->arch.emul_smt_mode;
581 			else
582 				r = kvm->arch.smt_mode;
583 		} else if (hv_enabled) {
584 			if (cpu_has_feature(CPU_FTR_ARCH_300))
585 				r = 1;
586 			else
587 				r = threads_per_subcore;
588 		}
589 		break;
590 	case KVM_CAP_PPC_SMT_POSSIBLE:
591 		r = 1;
592 		if (hv_enabled) {
593 			if (!cpu_has_feature(CPU_FTR_ARCH_300))
594 				r = ((threads_per_subcore << 1) - 1);
595 			else
596 				/* P9 can emulate dbells, so allow any mode */
597 				r = 8 | 4 | 2 | 1;
598 		}
599 		break;
600 	case KVM_CAP_PPC_RMA:
601 		r = 0;
602 		break;
603 	case KVM_CAP_PPC_HWRNG:
604 		r = kvmppc_hwrng_present();
605 		break;
606 	case KVM_CAP_PPC_MMU_RADIX:
607 		r = !!(hv_enabled && radix_enabled());
608 		break;
609 	case KVM_CAP_PPC_MMU_HASH_V3:
610 		r = !!(hv_enabled && cpu_has_feature(CPU_FTR_ARCH_300) &&
611 		       cpu_has_feature(CPU_FTR_HVMODE));
612 		break;
613 	case KVM_CAP_PPC_NESTED_HV:
614 		r = !!(hv_enabled && kvmppc_hv_ops->enable_nested &&
615 		       !kvmppc_hv_ops->enable_nested(NULL));
616 		break;
617 #endif
618 	case KVM_CAP_SYNC_MMU:
619 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
620 		r = hv_enabled;
621 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
622 		r = 1;
623 #else
624 		r = 0;
625 #endif
626 		break;
627 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
628 	case KVM_CAP_PPC_HTAB_FD:
629 		r = hv_enabled;
630 		break;
631 #endif
632 	case KVM_CAP_NR_VCPUS:
633 		/*
634 		 * Recommending a number of CPUs is somewhat arbitrary; we
635 		 * return the number of present CPUs for -HV (since a host
636 		 * will have secondary threads "offline"), and for other KVM
637 		 * implementations just count online CPUs.
638 		 */
639 		if (hv_enabled)
640 			r = num_present_cpus();
641 		else
642 			r = num_online_cpus();
643 		break;
644 	case KVM_CAP_MAX_VCPUS:
645 		r = KVM_MAX_VCPUS;
646 		break;
647 	case KVM_CAP_MAX_VCPU_ID:
648 		r = KVM_MAX_VCPU_ID;
649 		break;
650 #ifdef CONFIG_PPC_BOOK3S_64
651 	case KVM_CAP_PPC_GET_SMMU_INFO:
652 		r = 1;
653 		break;
654 	case KVM_CAP_SPAPR_MULTITCE:
655 		r = 1;
656 		break;
657 	case KVM_CAP_SPAPR_RESIZE_HPT:
658 		r = !!hv_enabled;
659 		break;
660 #endif
661 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
662 	case KVM_CAP_PPC_FWNMI:
663 		r = hv_enabled;
664 		break;
665 #endif
666 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
667 	case KVM_CAP_PPC_HTM:
668 		r = !!(cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM) ||
669 		     (hv_enabled && cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST));
670 		break;
671 #endif
672 #if defined(CONFIG_KVM_BOOK3S_HV_POSSIBLE)
673 	case KVM_CAP_PPC_SECURE_GUEST:
674 		r = hv_enabled && kvmppc_hv_ops->enable_svm &&
675 			!kvmppc_hv_ops->enable_svm(NULL);
676 		break;
677 #endif
678 	default:
679 		r = 0;
680 		break;
681 	}
682 	return r;
683 
684 }
685 
686 long kvm_arch_dev_ioctl(struct file *filp,
687                         unsigned int ioctl, unsigned long arg)
688 {
689 	return -EINVAL;
690 }
691 
692 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
693 {
694 	kvmppc_core_free_memslot(kvm, slot);
695 }
696 
697 int kvm_arch_prepare_memory_region(struct kvm *kvm,
698 				   struct kvm_memory_slot *memslot,
699 				   const struct kvm_userspace_memory_region *mem,
700 				   enum kvm_mr_change change)
701 {
702 	return kvmppc_core_prepare_memory_region(kvm, memslot, mem, change);
703 }
704 
705 void kvm_arch_commit_memory_region(struct kvm *kvm,
706 				   const struct kvm_userspace_memory_region *mem,
707 				   struct kvm_memory_slot *old,
708 				   const struct kvm_memory_slot *new,
709 				   enum kvm_mr_change change)
710 {
711 	kvmppc_core_commit_memory_region(kvm, mem, old, new, change);
712 }
713 
714 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
715 				   struct kvm_memory_slot *slot)
716 {
717 	kvmppc_core_flush_memslot(kvm, slot);
718 }
719 
720 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
721 {
722 	return 0;
723 }
724 
725 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
726 {
727 	struct kvm_vcpu *vcpu;
728 
729 	vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
730 	kvmppc_decrementer_func(vcpu);
731 
732 	return HRTIMER_NORESTART;
733 }
734 
735 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
736 {
737 	int err;
738 
739 	hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
740 	vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
741 	vcpu->arch.dec_expires = get_tb();
742 
743 #ifdef CONFIG_KVM_EXIT_TIMING
744 	mutex_init(&vcpu->arch.exit_timing_lock);
745 #endif
746 	err = kvmppc_subarch_vcpu_init(vcpu);
747 	if (err)
748 		return err;
749 
750 	err = kvmppc_core_vcpu_create(vcpu);
751 	if (err)
752 		goto out_vcpu_uninit;
753 
754 	vcpu->arch.wqp = &vcpu->wq;
755 	kvmppc_create_vcpu_debugfs(vcpu, vcpu->vcpu_id);
756 	return 0;
757 
758 out_vcpu_uninit:
759 	kvmppc_subarch_vcpu_uninit(vcpu);
760 	return err;
761 }
762 
763 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
764 {
765 }
766 
767 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
768 {
769 	/* Make sure we're not using the vcpu anymore */
770 	hrtimer_cancel(&vcpu->arch.dec_timer);
771 
772 	kvmppc_remove_vcpu_debugfs(vcpu);
773 
774 	switch (vcpu->arch.irq_type) {
775 	case KVMPPC_IRQ_MPIC:
776 		kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
777 		break;
778 	case KVMPPC_IRQ_XICS:
779 		if (xics_on_xive())
780 			kvmppc_xive_cleanup_vcpu(vcpu);
781 		else
782 			kvmppc_xics_free_icp(vcpu);
783 		break;
784 	case KVMPPC_IRQ_XIVE:
785 		kvmppc_xive_native_cleanup_vcpu(vcpu);
786 		break;
787 	}
788 
789 	kvmppc_core_vcpu_free(vcpu);
790 
791 	kvmppc_subarch_vcpu_uninit(vcpu);
792 }
793 
794 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
795 {
796 	return kvmppc_core_pending_dec(vcpu);
797 }
798 
799 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
800 {
801 #ifdef CONFIG_BOOKE
802 	/*
803 	 * vrsave (formerly usprg0) isn't used by Linux, but may
804 	 * be used by the guest.
805 	 *
806 	 * On non-booke this is associated with Altivec and
807 	 * is handled by code in book3s.c.
808 	 */
809 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
810 #endif
811 	kvmppc_core_vcpu_load(vcpu, cpu);
812 }
813 
814 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
815 {
816 	kvmppc_core_vcpu_put(vcpu);
817 #ifdef CONFIG_BOOKE
818 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
819 #endif
820 }
821 
822 /*
823  * irq_bypass_add_producer and irq_bypass_del_producer are only
824  * useful if the architecture supports PCI passthrough.
825  * irq_bypass_stop and irq_bypass_start are not needed and so
826  * kvm_ops are not defined for them.
827  */
828 bool kvm_arch_has_irq_bypass(void)
829 {
830 	return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
831 		(kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
832 }
833 
834 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
835 				     struct irq_bypass_producer *prod)
836 {
837 	struct kvm_kernel_irqfd *irqfd =
838 		container_of(cons, struct kvm_kernel_irqfd, consumer);
839 	struct kvm *kvm = irqfd->kvm;
840 
841 	if (kvm->arch.kvm_ops->irq_bypass_add_producer)
842 		return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
843 
844 	return 0;
845 }
846 
847 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
848 				      struct irq_bypass_producer *prod)
849 {
850 	struct kvm_kernel_irqfd *irqfd =
851 		container_of(cons, struct kvm_kernel_irqfd, consumer);
852 	struct kvm *kvm = irqfd->kvm;
853 
854 	if (kvm->arch.kvm_ops->irq_bypass_del_producer)
855 		kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
856 }
857 
858 #ifdef CONFIG_VSX
859 static inline int kvmppc_get_vsr_dword_offset(int index)
860 {
861 	int offset;
862 
863 	if ((index != 0) && (index != 1))
864 		return -1;
865 
866 #ifdef __BIG_ENDIAN
867 	offset =  index;
868 #else
869 	offset = 1 - index;
870 #endif
871 
872 	return offset;
873 }
874 
875 static inline int kvmppc_get_vsr_word_offset(int index)
876 {
877 	int offset;
878 
879 	if ((index > 3) || (index < 0))
880 		return -1;
881 
882 #ifdef __BIG_ENDIAN
883 	offset = index;
884 #else
885 	offset = 3 - index;
886 #endif
887 	return offset;
888 }
889 
890 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
891 	u64 gpr)
892 {
893 	union kvmppc_one_reg val;
894 	int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
895 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
896 
897 	if (offset == -1)
898 		return;
899 
900 	if (index >= 32) {
901 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
902 		val.vsxval[offset] = gpr;
903 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
904 	} else {
905 		VCPU_VSX_FPR(vcpu, index, offset) = gpr;
906 	}
907 }
908 
909 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
910 	u64 gpr)
911 {
912 	union kvmppc_one_reg val;
913 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
914 
915 	if (index >= 32) {
916 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
917 		val.vsxval[0] = gpr;
918 		val.vsxval[1] = gpr;
919 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
920 	} else {
921 		VCPU_VSX_FPR(vcpu, index, 0) = gpr;
922 		VCPU_VSX_FPR(vcpu, index, 1) = gpr;
923 	}
924 }
925 
926 static inline void kvmppc_set_vsr_word_dump(struct kvm_vcpu *vcpu,
927 	u32 gpr)
928 {
929 	union kvmppc_one_reg val;
930 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
931 
932 	if (index >= 32) {
933 		val.vsx32val[0] = gpr;
934 		val.vsx32val[1] = gpr;
935 		val.vsx32val[2] = gpr;
936 		val.vsx32val[3] = gpr;
937 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
938 	} else {
939 		val.vsx32val[0] = gpr;
940 		val.vsx32val[1] = gpr;
941 		VCPU_VSX_FPR(vcpu, index, 0) = val.vsxval[0];
942 		VCPU_VSX_FPR(vcpu, index, 1) = val.vsxval[0];
943 	}
944 }
945 
946 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
947 	u32 gpr32)
948 {
949 	union kvmppc_one_reg val;
950 	int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
951 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
952 	int dword_offset, word_offset;
953 
954 	if (offset == -1)
955 		return;
956 
957 	if (index >= 32) {
958 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
959 		val.vsx32val[offset] = gpr32;
960 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
961 	} else {
962 		dword_offset = offset / 2;
963 		word_offset = offset % 2;
964 		val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
965 		val.vsx32val[word_offset] = gpr32;
966 		VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
967 	}
968 }
969 #endif /* CONFIG_VSX */
970 
971 #ifdef CONFIG_ALTIVEC
972 static inline int kvmppc_get_vmx_offset_generic(struct kvm_vcpu *vcpu,
973 		int index, int element_size)
974 {
975 	int offset;
976 	int elts = sizeof(vector128)/element_size;
977 
978 	if ((index < 0) || (index >= elts))
979 		return -1;
980 
981 	if (kvmppc_need_byteswap(vcpu))
982 		offset = elts - index - 1;
983 	else
984 		offset = index;
985 
986 	return offset;
987 }
988 
989 static inline int kvmppc_get_vmx_dword_offset(struct kvm_vcpu *vcpu,
990 		int index)
991 {
992 	return kvmppc_get_vmx_offset_generic(vcpu, index, 8);
993 }
994 
995 static inline int kvmppc_get_vmx_word_offset(struct kvm_vcpu *vcpu,
996 		int index)
997 {
998 	return kvmppc_get_vmx_offset_generic(vcpu, index, 4);
999 }
1000 
1001 static inline int kvmppc_get_vmx_hword_offset(struct kvm_vcpu *vcpu,
1002 		int index)
1003 {
1004 	return kvmppc_get_vmx_offset_generic(vcpu, index, 2);
1005 }
1006 
1007 static inline int kvmppc_get_vmx_byte_offset(struct kvm_vcpu *vcpu,
1008 		int index)
1009 {
1010 	return kvmppc_get_vmx_offset_generic(vcpu, index, 1);
1011 }
1012 
1013 
1014 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu,
1015 	u64 gpr)
1016 {
1017 	union kvmppc_one_reg val;
1018 	int offset = kvmppc_get_vmx_dword_offset(vcpu,
1019 			vcpu->arch.mmio_vmx_offset);
1020 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1021 
1022 	if (offset == -1)
1023 		return;
1024 
1025 	val.vval = VCPU_VSX_VR(vcpu, index);
1026 	val.vsxval[offset] = gpr;
1027 	VCPU_VSX_VR(vcpu, index) = val.vval;
1028 }
1029 
1030 static inline void kvmppc_set_vmx_word(struct kvm_vcpu *vcpu,
1031 	u32 gpr32)
1032 {
1033 	union kvmppc_one_reg val;
1034 	int offset = kvmppc_get_vmx_word_offset(vcpu,
1035 			vcpu->arch.mmio_vmx_offset);
1036 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1037 
1038 	if (offset == -1)
1039 		return;
1040 
1041 	val.vval = VCPU_VSX_VR(vcpu, index);
1042 	val.vsx32val[offset] = gpr32;
1043 	VCPU_VSX_VR(vcpu, index) = val.vval;
1044 }
1045 
1046 static inline void kvmppc_set_vmx_hword(struct kvm_vcpu *vcpu,
1047 	u16 gpr16)
1048 {
1049 	union kvmppc_one_reg val;
1050 	int offset = kvmppc_get_vmx_hword_offset(vcpu,
1051 			vcpu->arch.mmio_vmx_offset);
1052 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1053 
1054 	if (offset == -1)
1055 		return;
1056 
1057 	val.vval = VCPU_VSX_VR(vcpu, index);
1058 	val.vsx16val[offset] = gpr16;
1059 	VCPU_VSX_VR(vcpu, index) = val.vval;
1060 }
1061 
1062 static inline void kvmppc_set_vmx_byte(struct kvm_vcpu *vcpu,
1063 	u8 gpr8)
1064 {
1065 	union kvmppc_one_reg val;
1066 	int offset = kvmppc_get_vmx_byte_offset(vcpu,
1067 			vcpu->arch.mmio_vmx_offset);
1068 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1069 
1070 	if (offset == -1)
1071 		return;
1072 
1073 	val.vval = VCPU_VSX_VR(vcpu, index);
1074 	val.vsx8val[offset] = gpr8;
1075 	VCPU_VSX_VR(vcpu, index) = val.vval;
1076 }
1077 #endif /* CONFIG_ALTIVEC */
1078 
1079 #ifdef CONFIG_PPC_FPU
1080 static inline u64 sp_to_dp(u32 fprs)
1081 {
1082 	u64 fprd;
1083 
1084 	preempt_disable();
1085 	enable_kernel_fp();
1086 	asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs)
1087 	     : "fr0");
1088 	preempt_enable();
1089 	return fprd;
1090 }
1091 
1092 static inline u32 dp_to_sp(u64 fprd)
1093 {
1094 	u32 fprs;
1095 
1096 	preempt_disable();
1097 	enable_kernel_fp();
1098 	asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd)
1099 	     : "fr0");
1100 	preempt_enable();
1101 	return fprs;
1102 }
1103 
1104 #else
1105 #define sp_to_dp(x)	(x)
1106 #define dp_to_sp(x)	(x)
1107 #endif /* CONFIG_PPC_FPU */
1108 
1109 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
1110                                       struct kvm_run *run)
1111 {
1112 	u64 uninitialized_var(gpr);
1113 
1114 	if (run->mmio.len > sizeof(gpr)) {
1115 		printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
1116 		return;
1117 	}
1118 
1119 	if (!vcpu->arch.mmio_host_swabbed) {
1120 		switch (run->mmio.len) {
1121 		case 8: gpr = *(u64 *)run->mmio.data; break;
1122 		case 4: gpr = *(u32 *)run->mmio.data; break;
1123 		case 2: gpr = *(u16 *)run->mmio.data; break;
1124 		case 1: gpr = *(u8 *)run->mmio.data; break;
1125 		}
1126 	} else {
1127 		switch (run->mmio.len) {
1128 		case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
1129 		case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
1130 		case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
1131 		case 1: gpr = *(u8 *)run->mmio.data; break;
1132 		}
1133 	}
1134 
1135 	/* conversion between single and double precision */
1136 	if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
1137 		gpr = sp_to_dp(gpr);
1138 
1139 	if (vcpu->arch.mmio_sign_extend) {
1140 		switch (run->mmio.len) {
1141 #ifdef CONFIG_PPC64
1142 		case 4:
1143 			gpr = (s64)(s32)gpr;
1144 			break;
1145 #endif
1146 		case 2:
1147 			gpr = (s64)(s16)gpr;
1148 			break;
1149 		case 1:
1150 			gpr = (s64)(s8)gpr;
1151 			break;
1152 		}
1153 	}
1154 
1155 	switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1156 	case KVM_MMIO_REG_GPR:
1157 		kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1158 		break;
1159 	case KVM_MMIO_REG_FPR:
1160 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1161 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_FP);
1162 
1163 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1164 		break;
1165 #ifdef CONFIG_PPC_BOOK3S
1166 	case KVM_MMIO_REG_QPR:
1167 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1168 		break;
1169 	case KVM_MMIO_REG_FQPR:
1170 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1171 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1172 		break;
1173 #endif
1174 #ifdef CONFIG_VSX
1175 	case KVM_MMIO_REG_VSX:
1176 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1177 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VSX);
1178 
1179 		if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_DWORD)
1180 			kvmppc_set_vsr_dword(vcpu, gpr);
1181 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_WORD)
1182 			kvmppc_set_vsr_word(vcpu, gpr);
1183 		else if (vcpu->arch.mmio_copy_type ==
1184 				KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1185 			kvmppc_set_vsr_dword_dump(vcpu, gpr);
1186 		else if (vcpu->arch.mmio_copy_type ==
1187 				KVMPPC_VSX_COPY_WORD_LOAD_DUMP)
1188 			kvmppc_set_vsr_word_dump(vcpu, gpr);
1189 		break;
1190 #endif
1191 #ifdef CONFIG_ALTIVEC
1192 	case KVM_MMIO_REG_VMX:
1193 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1194 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VEC);
1195 
1196 		if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_DWORD)
1197 			kvmppc_set_vmx_dword(vcpu, gpr);
1198 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_WORD)
1199 			kvmppc_set_vmx_word(vcpu, gpr);
1200 		else if (vcpu->arch.mmio_copy_type ==
1201 				KVMPPC_VMX_COPY_HWORD)
1202 			kvmppc_set_vmx_hword(vcpu, gpr);
1203 		else if (vcpu->arch.mmio_copy_type ==
1204 				KVMPPC_VMX_COPY_BYTE)
1205 			kvmppc_set_vmx_byte(vcpu, gpr);
1206 		break;
1207 #endif
1208 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1209 	case KVM_MMIO_REG_NESTED_GPR:
1210 		if (kvmppc_need_byteswap(vcpu))
1211 			gpr = swab64(gpr);
1212 		kvm_vcpu_write_guest(vcpu, vcpu->arch.nested_io_gpr, &gpr,
1213 				     sizeof(gpr));
1214 		break;
1215 #endif
1216 	default:
1217 		BUG();
1218 	}
1219 }
1220 
1221 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1222 				unsigned int rt, unsigned int bytes,
1223 				int is_default_endian, int sign_extend)
1224 {
1225 	int idx, ret;
1226 	bool host_swabbed;
1227 
1228 	/* Pity C doesn't have a logical XOR operator */
1229 	if (kvmppc_need_byteswap(vcpu)) {
1230 		host_swabbed = is_default_endian;
1231 	} else {
1232 		host_swabbed = !is_default_endian;
1233 	}
1234 
1235 	if (bytes > sizeof(run->mmio.data)) {
1236 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1237 		       run->mmio.len);
1238 	}
1239 
1240 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1241 	run->mmio.len = bytes;
1242 	run->mmio.is_write = 0;
1243 
1244 	vcpu->arch.io_gpr = rt;
1245 	vcpu->arch.mmio_host_swabbed = host_swabbed;
1246 	vcpu->mmio_needed = 1;
1247 	vcpu->mmio_is_write = 0;
1248 	vcpu->arch.mmio_sign_extend = sign_extend;
1249 
1250 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1251 
1252 	ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1253 			      bytes, &run->mmio.data);
1254 
1255 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1256 
1257 	if (!ret) {
1258 		kvmppc_complete_mmio_load(vcpu, run);
1259 		vcpu->mmio_needed = 0;
1260 		return EMULATE_DONE;
1261 	}
1262 
1263 	return EMULATE_DO_MMIO;
1264 }
1265 
1266 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1267 		       unsigned int rt, unsigned int bytes,
1268 		       int is_default_endian)
1269 {
1270 	return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
1271 }
1272 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1273 
1274 /* Same as above, but sign extends */
1275 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
1276 			unsigned int rt, unsigned int bytes,
1277 			int is_default_endian)
1278 {
1279 	return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
1280 }
1281 
1282 #ifdef CONFIG_VSX
1283 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1284 			unsigned int rt, unsigned int bytes,
1285 			int is_default_endian, int mmio_sign_extend)
1286 {
1287 	enum emulation_result emulated = EMULATE_DONE;
1288 
1289 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1290 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1291 		return EMULATE_FAIL;
1292 
1293 	while (vcpu->arch.mmio_vsx_copy_nums) {
1294 		emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1295 			is_default_endian, mmio_sign_extend);
1296 
1297 		if (emulated != EMULATE_DONE)
1298 			break;
1299 
1300 		vcpu->arch.paddr_accessed += run->mmio.len;
1301 
1302 		vcpu->arch.mmio_vsx_copy_nums--;
1303 		vcpu->arch.mmio_vsx_offset++;
1304 	}
1305 	return emulated;
1306 }
1307 #endif /* CONFIG_VSX */
1308 
1309 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1310 			u64 val, unsigned int bytes, int is_default_endian)
1311 {
1312 	void *data = run->mmio.data;
1313 	int idx, ret;
1314 	bool host_swabbed;
1315 
1316 	/* Pity C doesn't have a logical XOR operator */
1317 	if (kvmppc_need_byteswap(vcpu)) {
1318 		host_swabbed = is_default_endian;
1319 	} else {
1320 		host_swabbed = !is_default_endian;
1321 	}
1322 
1323 	if (bytes > sizeof(run->mmio.data)) {
1324 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1325 		       run->mmio.len);
1326 	}
1327 
1328 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1329 	run->mmio.len = bytes;
1330 	run->mmio.is_write = 1;
1331 	vcpu->mmio_needed = 1;
1332 	vcpu->mmio_is_write = 1;
1333 
1334 	if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1335 		val = dp_to_sp(val);
1336 
1337 	/* Store the value at the lowest bytes in 'data'. */
1338 	if (!host_swabbed) {
1339 		switch (bytes) {
1340 		case 8: *(u64 *)data = val; break;
1341 		case 4: *(u32 *)data = val; break;
1342 		case 2: *(u16 *)data = val; break;
1343 		case 1: *(u8  *)data = val; break;
1344 		}
1345 	} else {
1346 		switch (bytes) {
1347 		case 8: *(u64 *)data = swab64(val); break;
1348 		case 4: *(u32 *)data = swab32(val); break;
1349 		case 2: *(u16 *)data = swab16(val); break;
1350 		case 1: *(u8  *)data = val; break;
1351 		}
1352 	}
1353 
1354 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1355 
1356 	ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1357 			       bytes, &run->mmio.data);
1358 
1359 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1360 
1361 	if (!ret) {
1362 		vcpu->mmio_needed = 0;
1363 		return EMULATE_DONE;
1364 	}
1365 
1366 	return EMULATE_DO_MMIO;
1367 }
1368 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1369 
1370 #ifdef CONFIG_VSX
1371 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1372 {
1373 	u32 dword_offset, word_offset;
1374 	union kvmppc_one_reg reg;
1375 	int vsx_offset = 0;
1376 	int copy_type = vcpu->arch.mmio_copy_type;
1377 	int result = 0;
1378 
1379 	switch (copy_type) {
1380 	case KVMPPC_VSX_COPY_DWORD:
1381 		vsx_offset =
1382 			kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1383 
1384 		if (vsx_offset == -1) {
1385 			result = -1;
1386 			break;
1387 		}
1388 
1389 		if (rs < 32) {
1390 			*val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1391 		} else {
1392 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1393 			*val = reg.vsxval[vsx_offset];
1394 		}
1395 		break;
1396 
1397 	case KVMPPC_VSX_COPY_WORD:
1398 		vsx_offset =
1399 			kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1400 
1401 		if (vsx_offset == -1) {
1402 			result = -1;
1403 			break;
1404 		}
1405 
1406 		if (rs < 32) {
1407 			dword_offset = vsx_offset / 2;
1408 			word_offset = vsx_offset % 2;
1409 			reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1410 			*val = reg.vsx32val[word_offset];
1411 		} else {
1412 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1413 			*val = reg.vsx32val[vsx_offset];
1414 		}
1415 		break;
1416 
1417 	default:
1418 		result = -1;
1419 		break;
1420 	}
1421 
1422 	return result;
1423 }
1424 
1425 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1426 			int rs, unsigned int bytes, int is_default_endian)
1427 {
1428 	u64 val;
1429 	enum emulation_result emulated = EMULATE_DONE;
1430 
1431 	vcpu->arch.io_gpr = rs;
1432 
1433 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1434 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1435 		return EMULATE_FAIL;
1436 
1437 	while (vcpu->arch.mmio_vsx_copy_nums) {
1438 		if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1439 			return EMULATE_FAIL;
1440 
1441 		emulated = kvmppc_handle_store(run, vcpu,
1442 			 val, bytes, is_default_endian);
1443 
1444 		if (emulated != EMULATE_DONE)
1445 			break;
1446 
1447 		vcpu->arch.paddr_accessed += run->mmio.len;
1448 
1449 		vcpu->arch.mmio_vsx_copy_nums--;
1450 		vcpu->arch.mmio_vsx_offset++;
1451 	}
1452 
1453 	return emulated;
1454 }
1455 
1456 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu,
1457 			struct kvm_run *run)
1458 {
1459 	enum emulation_result emulated = EMULATE_FAIL;
1460 	int r;
1461 
1462 	vcpu->arch.paddr_accessed += run->mmio.len;
1463 
1464 	if (!vcpu->mmio_is_write) {
1465 		emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr,
1466 			 run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1467 	} else {
1468 		emulated = kvmppc_handle_vsx_store(run, vcpu,
1469 			 vcpu->arch.io_gpr, run->mmio.len, 1);
1470 	}
1471 
1472 	switch (emulated) {
1473 	case EMULATE_DO_MMIO:
1474 		run->exit_reason = KVM_EXIT_MMIO;
1475 		r = RESUME_HOST;
1476 		break;
1477 	case EMULATE_FAIL:
1478 		pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1479 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1480 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1481 		r = RESUME_HOST;
1482 		break;
1483 	default:
1484 		r = RESUME_GUEST;
1485 		break;
1486 	}
1487 	return r;
1488 }
1489 #endif /* CONFIG_VSX */
1490 
1491 #ifdef CONFIG_ALTIVEC
1492 int kvmppc_handle_vmx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1493 		unsigned int rt, unsigned int bytes, int is_default_endian)
1494 {
1495 	enum emulation_result emulated = EMULATE_DONE;
1496 
1497 	if (vcpu->arch.mmio_vsx_copy_nums > 2)
1498 		return EMULATE_FAIL;
1499 
1500 	while (vcpu->arch.mmio_vmx_copy_nums) {
1501 		emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1502 				is_default_endian, 0);
1503 
1504 		if (emulated != EMULATE_DONE)
1505 			break;
1506 
1507 		vcpu->arch.paddr_accessed += run->mmio.len;
1508 		vcpu->arch.mmio_vmx_copy_nums--;
1509 		vcpu->arch.mmio_vmx_offset++;
1510 	}
1511 
1512 	return emulated;
1513 }
1514 
1515 int kvmppc_get_vmx_dword(struct kvm_vcpu *vcpu, int index, u64 *val)
1516 {
1517 	union kvmppc_one_reg reg;
1518 	int vmx_offset = 0;
1519 	int result = 0;
1520 
1521 	vmx_offset =
1522 		kvmppc_get_vmx_dword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1523 
1524 	if (vmx_offset == -1)
1525 		return -1;
1526 
1527 	reg.vval = VCPU_VSX_VR(vcpu, index);
1528 	*val = reg.vsxval[vmx_offset];
1529 
1530 	return result;
1531 }
1532 
1533 int kvmppc_get_vmx_word(struct kvm_vcpu *vcpu, int index, u64 *val)
1534 {
1535 	union kvmppc_one_reg reg;
1536 	int vmx_offset = 0;
1537 	int result = 0;
1538 
1539 	vmx_offset =
1540 		kvmppc_get_vmx_word_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1541 
1542 	if (vmx_offset == -1)
1543 		return -1;
1544 
1545 	reg.vval = VCPU_VSX_VR(vcpu, index);
1546 	*val = reg.vsx32val[vmx_offset];
1547 
1548 	return result;
1549 }
1550 
1551 int kvmppc_get_vmx_hword(struct kvm_vcpu *vcpu, int index, u64 *val)
1552 {
1553 	union kvmppc_one_reg reg;
1554 	int vmx_offset = 0;
1555 	int result = 0;
1556 
1557 	vmx_offset =
1558 		kvmppc_get_vmx_hword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1559 
1560 	if (vmx_offset == -1)
1561 		return -1;
1562 
1563 	reg.vval = VCPU_VSX_VR(vcpu, index);
1564 	*val = reg.vsx16val[vmx_offset];
1565 
1566 	return result;
1567 }
1568 
1569 int kvmppc_get_vmx_byte(struct kvm_vcpu *vcpu, int index, u64 *val)
1570 {
1571 	union kvmppc_one_reg reg;
1572 	int vmx_offset = 0;
1573 	int result = 0;
1574 
1575 	vmx_offset =
1576 		kvmppc_get_vmx_byte_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1577 
1578 	if (vmx_offset == -1)
1579 		return -1;
1580 
1581 	reg.vval = VCPU_VSX_VR(vcpu, index);
1582 	*val = reg.vsx8val[vmx_offset];
1583 
1584 	return result;
1585 }
1586 
1587 int kvmppc_handle_vmx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1588 		unsigned int rs, unsigned int bytes, int is_default_endian)
1589 {
1590 	u64 val = 0;
1591 	unsigned int index = rs & KVM_MMIO_REG_MASK;
1592 	enum emulation_result emulated = EMULATE_DONE;
1593 
1594 	if (vcpu->arch.mmio_vsx_copy_nums > 2)
1595 		return EMULATE_FAIL;
1596 
1597 	vcpu->arch.io_gpr = rs;
1598 
1599 	while (vcpu->arch.mmio_vmx_copy_nums) {
1600 		switch (vcpu->arch.mmio_copy_type) {
1601 		case KVMPPC_VMX_COPY_DWORD:
1602 			if (kvmppc_get_vmx_dword(vcpu, index, &val) == -1)
1603 				return EMULATE_FAIL;
1604 
1605 			break;
1606 		case KVMPPC_VMX_COPY_WORD:
1607 			if (kvmppc_get_vmx_word(vcpu, index, &val) == -1)
1608 				return EMULATE_FAIL;
1609 			break;
1610 		case KVMPPC_VMX_COPY_HWORD:
1611 			if (kvmppc_get_vmx_hword(vcpu, index, &val) == -1)
1612 				return EMULATE_FAIL;
1613 			break;
1614 		case KVMPPC_VMX_COPY_BYTE:
1615 			if (kvmppc_get_vmx_byte(vcpu, index, &val) == -1)
1616 				return EMULATE_FAIL;
1617 			break;
1618 		default:
1619 			return EMULATE_FAIL;
1620 		}
1621 
1622 		emulated = kvmppc_handle_store(run, vcpu, val, bytes,
1623 				is_default_endian);
1624 		if (emulated != EMULATE_DONE)
1625 			break;
1626 
1627 		vcpu->arch.paddr_accessed += run->mmio.len;
1628 		vcpu->arch.mmio_vmx_copy_nums--;
1629 		vcpu->arch.mmio_vmx_offset++;
1630 	}
1631 
1632 	return emulated;
1633 }
1634 
1635 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu,
1636 		struct kvm_run *run)
1637 {
1638 	enum emulation_result emulated = EMULATE_FAIL;
1639 	int r;
1640 
1641 	vcpu->arch.paddr_accessed += run->mmio.len;
1642 
1643 	if (!vcpu->mmio_is_write) {
1644 		emulated = kvmppc_handle_vmx_load(run, vcpu,
1645 				vcpu->arch.io_gpr, run->mmio.len, 1);
1646 	} else {
1647 		emulated = kvmppc_handle_vmx_store(run, vcpu,
1648 				vcpu->arch.io_gpr, run->mmio.len, 1);
1649 	}
1650 
1651 	switch (emulated) {
1652 	case EMULATE_DO_MMIO:
1653 		run->exit_reason = KVM_EXIT_MMIO;
1654 		r = RESUME_HOST;
1655 		break;
1656 	case EMULATE_FAIL:
1657 		pr_info("KVM: MMIO emulation failed (VMX repeat)\n");
1658 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1659 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1660 		r = RESUME_HOST;
1661 		break;
1662 	default:
1663 		r = RESUME_GUEST;
1664 		break;
1665 	}
1666 	return r;
1667 }
1668 #endif /* CONFIG_ALTIVEC */
1669 
1670 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1671 {
1672 	int r = 0;
1673 	union kvmppc_one_reg val;
1674 	int size;
1675 
1676 	size = one_reg_size(reg->id);
1677 	if (size > sizeof(val))
1678 		return -EINVAL;
1679 
1680 	r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1681 	if (r == -EINVAL) {
1682 		r = 0;
1683 		switch (reg->id) {
1684 #ifdef CONFIG_ALTIVEC
1685 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1686 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1687 				r = -ENXIO;
1688 				break;
1689 			}
1690 			val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1691 			break;
1692 		case KVM_REG_PPC_VSCR:
1693 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1694 				r = -ENXIO;
1695 				break;
1696 			}
1697 			val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1698 			break;
1699 		case KVM_REG_PPC_VRSAVE:
1700 			val = get_reg_val(reg->id, vcpu->arch.vrsave);
1701 			break;
1702 #endif /* CONFIG_ALTIVEC */
1703 		default:
1704 			r = -EINVAL;
1705 			break;
1706 		}
1707 	}
1708 
1709 	if (r)
1710 		return r;
1711 
1712 	if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1713 		r = -EFAULT;
1714 
1715 	return r;
1716 }
1717 
1718 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1719 {
1720 	int r;
1721 	union kvmppc_one_reg val;
1722 	int size;
1723 
1724 	size = one_reg_size(reg->id);
1725 	if (size > sizeof(val))
1726 		return -EINVAL;
1727 
1728 	if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1729 		return -EFAULT;
1730 
1731 	r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1732 	if (r == -EINVAL) {
1733 		r = 0;
1734 		switch (reg->id) {
1735 #ifdef CONFIG_ALTIVEC
1736 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1737 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1738 				r = -ENXIO;
1739 				break;
1740 			}
1741 			vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1742 			break;
1743 		case KVM_REG_PPC_VSCR:
1744 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1745 				r = -ENXIO;
1746 				break;
1747 			}
1748 			vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1749 			break;
1750 		case KVM_REG_PPC_VRSAVE:
1751 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1752 				r = -ENXIO;
1753 				break;
1754 			}
1755 			vcpu->arch.vrsave = set_reg_val(reg->id, val);
1756 			break;
1757 #endif /* CONFIG_ALTIVEC */
1758 		default:
1759 			r = -EINVAL;
1760 			break;
1761 		}
1762 	}
1763 
1764 	return r;
1765 }
1766 
1767 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1768 {
1769 	int r;
1770 
1771 	vcpu_load(vcpu);
1772 
1773 	if (vcpu->mmio_needed) {
1774 		vcpu->mmio_needed = 0;
1775 		if (!vcpu->mmio_is_write)
1776 			kvmppc_complete_mmio_load(vcpu, run);
1777 #ifdef CONFIG_VSX
1778 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1779 			vcpu->arch.mmio_vsx_copy_nums--;
1780 			vcpu->arch.mmio_vsx_offset++;
1781 		}
1782 
1783 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1784 			r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run);
1785 			if (r == RESUME_HOST) {
1786 				vcpu->mmio_needed = 1;
1787 				goto out;
1788 			}
1789 		}
1790 #endif
1791 #ifdef CONFIG_ALTIVEC
1792 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1793 			vcpu->arch.mmio_vmx_copy_nums--;
1794 			vcpu->arch.mmio_vmx_offset++;
1795 		}
1796 
1797 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1798 			r = kvmppc_emulate_mmio_vmx_loadstore(vcpu, run);
1799 			if (r == RESUME_HOST) {
1800 				vcpu->mmio_needed = 1;
1801 				goto out;
1802 			}
1803 		}
1804 #endif
1805 	} else if (vcpu->arch.osi_needed) {
1806 		u64 *gprs = run->osi.gprs;
1807 		int i;
1808 
1809 		for (i = 0; i < 32; i++)
1810 			kvmppc_set_gpr(vcpu, i, gprs[i]);
1811 		vcpu->arch.osi_needed = 0;
1812 	} else if (vcpu->arch.hcall_needed) {
1813 		int i;
1814 
1815 		kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1816 		for (i = 0; i < 9; ++i)
1817 			kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1818 		vcpu->arch.hcall_needed = 0;
1819 #ifdef CONFIG_BOOKE
1820 	} else if (vcpu->arch.epr_needed) {
1821 		kvmppc_set_epr(vcpu, run->epr.epr);
1822 		vcpu->arch.epr_needed = 0;
1823 #endif
1824 	}
1825 
1826 	kvm_sigset_activate(vcpu);
1827 
1828 	if (run->immediate_exit)
1829 		r = -EINTR;
1830 	else
1831 		r = kvmppc_vcpu_run(run, vcpu);
1832 
1833 	kvm_sigset_deactivate(vcpu);
1834 
1835 #ifdef CONFIG_ALTIVEC
1836 out:
1837 #endif
1838 	vcpu_put(vcpu);
1839 	return r;
1840 }
1841 
1842 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1843 {
1844 	if (irq->irq == KVM_INTERRUPT_UNSET) {
1845 		kvmppc_core_dequeue_external(vcpu);
1846 		return 0;
1847 	}
1848 
1849 	kvmppc_core_queue_external(vcpu, irq);
1850 
1851 	kvm_vcpu_kick(vcpu);
1852 
1853 	return 0;
1854 }
1855 
1856 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1857 				     struct kvm_enable_cap *cap)
1858 {
1859 	int r;
1860 
1861 	if (cap->flags)
1862 		return -EINVAL;
1863 
1864 	switch (cap->cap) {
1865 	case KVM_CAP_PPC_OSI:
1866 		r = 0;
1867 		vcpu->arch.osi_enabled = true;
1868 		break;
1869 	case KVM_CAP_PPC_PAPR:
1870 		r = 0;
1871 		vcpu->arch.papr_enabled = true;
1872 		break;
1873 	case KVM_CAP_PPC_EPR:
1874 		r = 0;
1875 		if (cap->args[0])
1876 			vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1877 		else
1878 			vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1879 		break;
1880 #ifdef CONFIG_BOOKE
1881 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
1882 		r = 0;
1883 		vcpu->arch.watchdog_enabled = true;
1884 		break;
1885 #endif
1886 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1887 	case KVM_CAP_SW_TLB: {
1888 		struct kvm_config_tlb cfg;
1889 		void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1890 
1891 		r = -EFAULT;
1892 		if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1893 			break;
1894 
1895 		r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1896 		break;
1897 	}
1898 #endif
1899 #ifdef CONFIG_KVM_MPIC
1900 	case KVM_CAP_IRQ_MPIC: {
1901 		struct fd f;
1902 		struct kvm_device *dev;
1903 
1904 		r = -EBADF;
1905 		f = fdget(cap->args[0]);
1906 		if (!f.file)
1907 			break;
1908 
1909 		r = -EPERM;
1910 		dev = kvm_device_from_filp(f.file);
1911 		if (dev)
1912 			r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1913 
1914 		fdput(f);
1915 		break;
1916 	}
1917 #endif
1918 #ifdef CONFIG_KVM_XICS
1919 	case KVM_CAP_IRQ_XICS: {
1920 		struct fd f;
1921 		struct kvm_device *dev;
1922 
1923 		r = -EBADF;
1924 		f = fdget(cap->args[0]);
1925 		if (!f.file)
1926 			break;
1927 
1928 		r = -EPERM;
1929 		dev = kvm_device_from_filp(f.file);
1930 		if (dev) {
1931 			if (xics_on_xive())
1932 				r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1933 			else
1934 				r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1935 		}
1936 
1937 		fdput(f);
1938 		break;
1939 	}
1940 #endif /* CONFIG_KVM_XICS */
1941 #ifdef CONFIG_KVM_XIVE
1942 	case KVM_CAP_PPC_IRQ_XIVE: {
1943 		struct fd f;
1944 		struct kvm_device *dev;
1945 
1946 		r = -EBADF;
1947 		f = fdget(cap->args[0]);
1948 		if (!f.file)
1949 			break;
1950 
1951 		r = -ENXIO;
1952 		if (!xive_enabled())
1953 			break;
1954 
1955 		r = -EPERM;
1956 		dev = kvm_device_from_filp(f.file);
1957 		if (dev)
1958 			r = kvmppc_xive_native_connect_vcpu(dev, vcpu,
1959 							    cap->args[1]);
1960 
1961 		fdput(f);
1962 		break;
1963 	}
1964 #endif /* CONFIG_KVM_XIVE */
1965 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1966 	case KVM_CAP_PPC_FWNMI:
1967 		r = -EINVAL;
1968 		if (!is_kvmppc_hv_enabled(vcpu->kvm))
1969 			break;
1970 		r = 0;
1971 		vcpu->kvm->arch.fwnmi_enabled = true;
1972 		break;
1973 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1974 	default:
1975 		r = -EINVAL;
1976 		break;
1977 	}
1978 
1979 	if (!r)
1980 		r = kvmppc_sanity_check(vcpu);
1981 
1982 	return r;
1983 }
1984 
1985 bool kvm_arch_intc_initialized(struct kvm *kvm)
1986 {
1987 #ifdef CONFIG_KVM_MPIC
1988 	if (kvm->arch.mpic)
1989 		return true;
1990 #endif
1991 #ifdef CONFIG_KVM_XICS
1992 	if (kvm->arch.xics || kvm->arch.xive)
1993 		return true;
1994 #endif
1995 	return false;
1996 }
1997 
1998 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1999                                     struct kvm_mp_state *mp_state)
2000 {
2001 	return -EINVAL;
2002 }
2003 
2004 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2005                                     struct kvm_mp_state *mp_state)
2006 {
2007 	return -EINVAL;
2008 }
2009 
2010 long kvm_arch_vcpu_async_ioctl(struct file *filp,
2011 			       unsigned int ioctl, unsigned long arg)
2012 {
2013 	struct kvm_vcpu *vcpu = filp->private_data;
2014 	void __user *argp = (void __user *)arg;
2015 
2016 	if (ioctl == KVM_INTERRUPT) {
2017 		struct kvm_interrupt irq;
2018 		if (copy_from_user(&irq, argp, sizeof(irq)))
2019 			return -EFAULT;
2020 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2021 	}
2022 	return -ENOIOCTLCMD;
2023 }
2024 
2025 long kvm_arch_vcpu_ioctl(struct file *filp,
2026                          unsigned int ioctl, unsigned long arg)
2027 {
2028 	struct kvm_vcpu *vcpu = filp->private_data;
2029 	void __user *argp = (void __user *)arg;
2030 	long r;
2031 
2032 	switch (ioctl) {
2033 	case KVM_ENABLE_CAP:
2034 	{
2035 		struct kvm_enable_cap cap;
2036 		r = -EFAULT;
2037 		vcpu_load(vcpu);
2038 		if (copy_from_user(&cap, argp, sizeof(cap)))
2039 			goto out;
2040 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
2041 		vcpu_put(vcpu);
2042 		break;
2043 	}
2044 
2045 	case KVM_SET_ONE_REG:
2046 	case KVM_GET_ONE_REG:
2047 	{
2048 		struct kvm_one_reg reg;
2049 		r = -EFAULT;
2050 		if (copy_from_user(&reg, argp, sizeof(reg)))
2051 			goto out;
2052 		if (ioctl == KVM_SET_ONE_REG)
2053 			r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
2054 		else
2055 			r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
2056 		break;
2057 	}
2058 
2059 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
2060 	case KVM_DIRTY_TLB: {
2061 		struct kvm_dirty_tlb dirty;
2062 		r = -EFAULT;
2063 		vcpu_load(vcpu);
2064 		if (copy_from_user(&dirty, argp, sizeof(dirty)))
2065 			goto out;
2066 		r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
2067 		vcpu_put(vcpu);
2068 		break;
2069 	}
2070 #endif
2071 	default:
2072 		r = -EINVAL;
2073 	}
2074 
2075 out:
2076 	return r;
2077 }
2078 
2079 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2080 {
2081 	return VM_FAULT_SIGBUS;
2082 }
2083 
2084 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
2085 {
2086 	u32 inst_nop = 0x60000000;
2087 #ifdef CONFIG_KVM_BOOKE_HV
2088 	u32 inst_sc1 = 0x44000022;
2089 	pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
2090 	pvinfo->hcall[1] = cpu_to_be32(inst_nop);
2091 	pvinfo->hcall[2] = cpu_to_be32(inst_nop);
2092 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2093 #else
2094 	u32 inst_lis = 0x3c000000;
2095 	u32 inst_ori = 0x60000000;
2096 	u32 inst_sc = 0x44000002;
2097 	u32 inst_imm_mask = 0xffff;
2098 
2099 	/*
2100 	 * The hypercall to get into KVM from within guest context is as
2101 	 * follows:
2102 	 *
2103 	 *    lis r0, r0, KVM_SC_MAGIC_R0@h
2104 	 *    ori r0, KVM_SC_MAGIC_R0@l
2105 	 *    sc
2106 	 *    nop
2107 	 */
2108 	pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
2109 	pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
2110 	pvinfo->hcall[2] = cpu_to_be32(inst_sc);
2111 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2112 #endif
2113 
2114 	pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
2115 
2116 	return 0;
2117 }
2118 
2119 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
2120 			  bool line_status)
2121 {
2122 	if (!irqchip_in_kernel(kvm))
2123 		return -ENXIO;
2124 
2125 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2126 					irq_event->irq, irq_event->level,
2127 					line_status);
2128 	return 0;
2129 }
2130 
2131 
2132 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
2133 			    struct kvm_enable_cap *cap)
2134 {
2135 	int r;
2136 
2137 	if (cap->flags)
2138 		return -EINVAL;
2139 
2140 	switch (cap->cap) {
2141 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
2142 	case KVM_CAP_PPC_ENABLE_HCALL: {
2143 		unsigned long hcall = cap->args[0];
2144 
2145 		r = -EINVAL;
2146 		if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
2147 		    cap->args[1] > 1)
2148 			break;
2149 		if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
2150 			break;
2151 		if (cap->args[1])
2152 			set_bit(hcall / 4, kvm->arch.enabled_hcalls);
2153 		else
2154 			clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
2155 		r = 0;
2156 		break;
2157 	}
2158 	case KVM_CAP_PPC_SMT: {
2159 		unsigned long mode = cap->args[0];
2160 		unsigned long flags = cap->args[1];
2161 
2162 		r = -EINVAL;
2163 		if (kvm->arch.kvm_ops->set_smt_mode)
2164 			r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
2165 		break;
2166 	}
2167 
2168 	case KVM_CAP_PPC_NESTED_HV:
2169 		r = -EINVAL;
2170 		if (!is_kvmppc_hv_enabled(kvm) ||
2171 		    !kvm->arch.kvm_ops->enable_nested)
2172 			break;
2173 		r = kvm->arch.kvm_ops->enable_nested(kvm);
2174 		break;
2175 #endif
2176 #if defined(CONFIG_KVM_BOOK3S_HV_POSSIBLE)
2177 	case KVM_CAP_PPC_SECURE_GUEST:
2178 		r = -EINVAL;
2179 		if (!is_kvmppc_hv_enabled(kvm) || !kvm->arch.kvm_ops->enable_svm)
2180 			break;
2181 		r = kvm->arch.kvm_ops->enable_svm(kvm);
2182 		break;
2183 #endif
2184 	default:
2185 		r = -EINVAL;
2186 		break;
2187 	}
2188 
2189 	return r;
2190 }
2191 
2192 #ifdef CONFIG_PPC_BOOK3S_64
2193 /*
2194  * These functions check whether the underlying hardware is safe
2195  * against attacks based on observing the effects of speculatively
2196  * executed instructions, and whether it supplies instructions for
2197  * use in workarounds.  The information comes from firmware, either
2198  * via the device tree on powernv platforms or from an hcall on
2199  * pseries platforms.
2200  */
2201 #ifdef CONFIG_PPC_PSERIES
2202 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2203 {
2204 	struct h_cpu_char_result c;
2205 	unsigned long rc;
2206 
2207 	if (!machine_is(pseries))
2208 		return -ENOTTY;
2209 
2210 	rc = plpar_get_cpu_characteristics(&c);
2211 	if (rc == H_SUCCESS) {
2212 		cp->character = c.character;
2213 		cp->behaviour = c.behaviour;
2214 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2215 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2216 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2217 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2218 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2219 			KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
2220 			KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
2221 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2222 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2223 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2224 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2225 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2226 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2227 	}
2228 	return 0;
2229 }
2230 #else
2231 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2232 {
2233 	return -ENOTTY;
2234 }
2235 #endif
2236 
2237 static inline bool have_fw_feat(struct device_node *fw_features,
2238 				const char *state, const char *name)
2239 {
2240 	struct device_node *np;
2241 	bool r = false;
2242 
2243 	np = of_get_child_by_name(fw_features, name);
2244 	if (np) {
2245 		r = of_property_read_bool(np, state);
2246 		of_node_put(np);
2247 	}
2248 	return r;
2249 }
2250 
2251 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2252 {
2253 	struct device_node *np, *fw_features;
2254 	int r;
2255 
2256 	memset(cp, 0, sizeof(*cp));
2257 	r = pseries_get_cpu_char(cp);
2258 	if (r != -ENOTTY)
2259 		return r;
2260 
2261 	np = of_find_node_by_name(NULL, "ibm,opal");
2262 	if (np) {
2263 		fw_features = of_get_child_by_name(np, "fw-features");
2264 		of_node_put(np);
2265 		if (!fw_features)
2266 			return 0;
2267 		if (have_fw_feat(fw_features, "enabled",
2268 				 "inst-spec-barrier-ori31,31,0"))
2269 			cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
2270 		if (have_fw_feat(fw_features, "enabled",
2271 				 "fw-bcctrl-serialized"))
2272 			cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
2273 		if (have_fw_feat(fw_features, "enabled",
2274 				 "inst-l1d-flush-ori30,30,0"))
2275 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
2276 		if (have_fw_feat(fw_features, "enabled",
2277 				 "inst-l1d-flush-trig2"))
2278 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
2279 		if (have_fw_feat(fw_features, "enabled",
2280 				 "fw-l1d-thread-split"))
2281 			cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
2282 		if (have_fw_feat(fw_features, "enabled",
2283 				 "fw-count-cache-disabled"))
2284 			cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2285 		if (have_fw_feat(fw_features, "enabled",
2286 				 "fw-count-cache-flush-bcctr2,0,0"))
2287 			cp->character |= KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2288 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2289 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2290 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2291 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2292 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2293 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2294 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2295 
2296 		if (have_fw_feat(fw_features, "enabled",
2297 				 "speculation-policy-favor-security"))
2298 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
2299 		if (!have_fw_feat(fw_features, "disabled",
2300 				  "needs-l1d-flush-msr-pr-0-to-1"))
2301 			cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
2302 		if (!have_fw_feat(fw_features, "disabled",
2303 				  "needs-spec-barrier-for-bound-checks"))
2304 			cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2305 		if (have_fw_feat(fw_features, "enabled",
2306 				 "needs-count-cache-flush-on-context-switch"))
2307 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2308 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2309 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2310 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2311 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2312 
2313 		of_node_put(fw_features);
2314 	}
2315 
2316 	return 0;
2317 }
2318 #endif
2319 
2320 long kvm_arch_vm_ioctl(struct file *filp,
2321                        unsigned int ioctl, unsigned long arg)
2322 {
2323 	struct kvm *kvm __maybe_unused = filp->private_data;
2324 	void __user *argp = (void __user *)arg;
2325 	long r;
2326 
2327 	switch (ioctl) {
2328 	case KVM_PPC_GET_PVINFO: {
2329 		struct kvm_ppc_pvinfo pvinfo;
2330 		memset(&pvinfo, 0, sizeof(pvinfo));
2331 		r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
2332 		if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
2333 			r = -EFAULT;
2334 			goto out;
2335 		}
2336 
2337 		break;
2338 	}
2339 #ifdef CONFIG_SPAPR_TCE_IOMMU
2340 	case KVM_CREATE_SPAPR_TCE_64: {
2341 		struct kvm_create_spapr_tce_64 create_tce_64;
2342 
2343 		r = -EFAULT;
2344 		if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
2345 			goto out;
2346 		if (create_tce_64.flags) {
2347 			r = -EINVAL;
2348 			goto out;
2349 		}
2350 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2351 		goto out;
2352 	}
2353 	case KVM_CREATE_SPAPR_TCE: {
2354 		struct kvm_create_spapr_tce create_tce;
2355 		struct kvm_create_spapr_tce_64 create_tce_64;
2356 
2357 		r = -EFAULT;
2358 		if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
2359 			goto out;
2360 
2361 		create_tce_64.liobn = create_tce.liobn;
2362 		create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
2363 		create_tce_64.offset = 0;
2364 		create_tce_64.size = create_tce.window_size >>
2365 				IOMMU_PAGE_SHIFT_4K;
2366 		create_tce_64.flags = 0;
2367 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2368 		goto out;
2369 	}
2370 #endif
2371 #ifdef CONFIG_PPC_BOOK3S_64
2372 	case KVM_PPC_GET_SMMU_INFO: {
2373 		struct kvm_ppc_smmu_info info;
2374 		struct kvm *kvm = filp->private_data;
2375 
2376 		memset(&info, 0, sizeof(info));
2377 		r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
2378 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2379 			r = -EFAULT;
2380 		break;
2381 	}
2382 	case KVM_PPC_RTAS_DEFINE_TOKEN: {
2383 		struct kvm *kvm = filp->private_data;
2384 
2385 		r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
2386 		break;
2387 	}
2388 	case KVM_PPC_CONFIGURE_V3_MMU: {
2389 		struct kvm *kvm = filp->private_data;
2390 		struct kvm_ppc_mmuv3_cfg cfg;
2391 
2392 		r = -EINVAL;
2393 		if (!kvm->arch.kvm_ops->configure_mmu)
2394 			goto out;
2395 		r = -EFAULT;
2396 		if (copy_from_user(&cfg, argp, sizeof(cfg)))
2397 			goto out;
2398 		r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
2399 		break;
2400 	}
2401 	case KVM_PPC_GET_RMMU_INFO: {
2402 		struct kvm *kvm = filp->private_data;
2403 		struct kvm_ppc_rmmu_info info;
2404 
2405 		r = -EINVAL;
2406 		if (!kvm->arch.kvm_ops->get_rmmu_info)
2407 			goto out;
2408 		r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
2409 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2410 			r = -EFAULT;
2411 		break;
2412 	}
2413 	case KVM_PPC_GET_CPU_CHAR: {
2414 		struct kvm_ppc_cpu_char cpuchar;
2415 
2416 		r = kvmppc_get_cpu_char(&cpuchar);
2417 		if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2418 			r = -EFAULT;
2419 		break;
2420 	}
2421 	case KVM_PPC_SVM_OFF: {
2422 		struct kvm *kvm = filp->private_data;
2423 
2424 		r = 0;
2425 		if (!kvm->arch.kvm_ops->svm_off)
2426 			goto out;
2427 
2428 		r = kvm->arch.kvm_ops->svm_off(kvm);
2429 		break;
2430 	}
2431 	default: {
2432 		struct kvm *kvm = filp->private_data;
2433 		r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
2434 	}
2435 #else /* CONFIG_PPC_BOOK3S_64 */
2436 	default:
2437 		r = -ENOTTY;
2438 #endif
2439 	}
2440 out:
2441 	return r;
2442 }
2443 
2444 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
2445 static unsigned long nr_lpids;
2446 
2447 long kvmppc_alloc_lpid(void)
2448 {
2449 	long lpid;
2450 
2451 	do {
2452 		lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
2453 		if (lpid >= nr_lpids) {
2454 			pr_err("%s: No LPIDs free\n", __func__);
2455 			return -ENOMEM;
2456 		}
2457 	} while (test_and_set_bit(lpid, lpid_inuse));
2458 
2459 	return lpid;
2460 }
2461 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2462 
2463 void kvmppc_claim_lpid(long lpid)
2464 {
2465 	set_bit(lpid, lpid_inuse);
2466 }
2467 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
2468 
2469 void kvmppc_free_lpid(long lpid)
2470 {
2471 	clear_bit(lpid, lpid_inuse);
2472 }
2473 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2474 
2475 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2476 {
2477 	nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
2478 	memset(lpid_inuse, 0, sizeof(lpid_inuse));
2479 }
2480 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2481 
2482 int kvm_arch_init(void *opaque)
2483 {
2484 	return 0;
2485 }
2486 
2487 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);
2488