xref: /openbmc/linux/arch/powerpc/kvm/powerpc.c (revision e7f127b2)
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_halt(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_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 		vcpu->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 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
407 	rc = kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size);
408 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
409 	if (rc)
410 		return EMULATE_DO_MMIO;
411 
412 	return EMULATE_DONE;
413 }
414 EXPORT_SYMBOL_GPL(kvmppc_ld);
415 
416 int kvm_arch_hardware_enable(void)
417 {
418 	return 0;
419 }
420 
421 int kvm_arch_hardware_setup(void *opaque)
422 {
423 	return 0;
424 }
425 
426 int kvm_arch_check_processor_compat(void *opaque)
427 {
428 	return kvmppc_core_check_processor_compat();
429 }
430 
431 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
432 {
433 	struct kvmppc_ops *kvm_ops = NULL;
434 	/*
435 	 * if we have both HV and PR enabled, default is HV
436 	 */
437 	if (type == 0) {
438 		if (kvmppc_hv_ops)
439 			kvm_ops = kvmppc_hv_ops;
440 		else
441 			kvm_ops = kvmppc_pr_ops;
442 		if (!kvm_ops)
443 			goto err_out;
444 	} else	if (type == KVM_VM_PPC_HV) {
445 		if (!kvmppc_hv_ops)
446 			goto err_out;
447 		kvm_ops = kvmppc_hv_ops;
448 	} else if (type == KVM_VM_PPC_PR) {
449 		if (!kvmppc_pr_ops)
450 			goto err_out;
451 		kvm_ops = kvmppc_pr_ops;
452 	} else
453 		goto err_out;
454 
455 	if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
456 		return -ENOENT;
457 
458 	kvm->arch.kvm_ops = kvm_ops;
459 	return kvmppc_core_init_vm(kvm);
460 err_out:
461 	return -EINVAL;
462 }
463 
464 void kvm_arch_destroy_vm(struct kvm *kvm)
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_destroy_vcpus(kvm);
477 
478 	mutex_lock(&kvm->lock);
479 
480 	kvmppc_core_destroy_vm(kvm);
481 
482 	mutex_unlock(&kvm->lock);
483 
484 	/* drop the module reference */
485 	module_put(kvm->arch.kvm_ops->owner);
486 }
487 
488 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
489 {
490 	int r;
491 	/* Assume we're using HV mode when the HV module is loaded */
492 	int hv_enabled = kvmppc_hv_ops ? 1 : 0;
493 
494 	if (kvm) {
495 		/*
496 		 * Hooray - we know which VM type we're running on. Depend on
497 		 * that rather than the guess above.
498 		 */
499 		hv_enabled = is_kvmppc_hv_enabled(kvm);
500 	}
501 
502 	switch (ext) {
503 #ifdef CONFIG_BOOKE
504 	case KVM_CAP_PPC_BOOKE_SREGS:
505 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
506 	case KVM_CAP_PPC_EPR:
507 #else
508 	case KVM_CAP_PPC_SEGSTATE:
509 	case KVM_CAP_PPC_HIOR:
510 	case KVM_CAP_PPC_PAPR:
511 #endif
512 	case KVM_CAP_PPC_UNSET_IRQ:
513 	case KVM_CAP_PPC_IRQ_LEVEL:
514 	case KVM_CAP_ENABLE_CAP:
515 	case KVM_CAP_ONE_REG:
516 	case KVM_CAP_IOEVENTFD:
517 	case KVM_CAP_DEVICE_CTRL:
518 	case KVM_CAP_IMMEDIATE_EXIT:
519 	case KVM_CAP_SET_GUEST_DEBUG:
520 		r = 1;
521 		break;
522 	case KVM_CAP_PPC_GUEST_DEBUG_SSTEP:
523 	case KVM_CAP_PPC_PAIRED_SINGLES:
524 	case KVM_CAP_PPC_OSI:
525 	case KVM_CAP_PPC_GET_PVINFO:
526 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
527 	case KVM_CAP_SW_TLB:
528 #endif
529 		/* We support this only for PR */
530 		r = !hv_enabled;
531 		break;
532 #ifdef CONFIG_KVM_MPIC
533 	case KVM_CAP_IRQ_MPIC:
534 		r = 1;
535 		break;
536 #endif
537 
538 #ifdef CONFIG_PPC_BOOK3S_64
539 	case KVM_CAP_SPAPR_TCE:
540 	case KVM_CAP_SPAPR_TCE_64:
541 		r = 1;
542 		break;
543 	case KVM_CAP_SPAPR_TCE_VFIO:
544 		r = !!cpu_has_feature(CPU_FTR_HVMODE);
545 		break;
546 	case KVM_CAP_PPC_RTAS:
547 	case KVM_CAP_PPC_FIXUP_HCALL:
548 	case KVM_CAP_PPC_ENABLE_HCALL:
549 #ifdef CONFIG_KVM_XICS
550 	case KVM_CAP_IRQ_XICS:
551 #endif
552 	case KVM_CAP_PPC_GET_CPU_CHAR:
553 		r = 1;
554 		break;
555 #ifdef CONFIG_KVM_XIVE
556 	case KVM_CAP_PPC_IRQ_XIVE:
557 		/*
558 		 * We need XIVE to be enabled on the platform (implies
559 		 * a POWER9 processor) and the PowerNV platform, as
560 		 * nested is not yet supported.
561 		 */
562 		r = xive_enabled() && !!cpu_has_feature(CPU_FTR_HVMODE) &&
563 			kvmppc_xive_native_supported();
564 		break;
565 #endif
566 
567 	case KVM_CAP_PPC_ALLOC_HTAB:
568 		r = hv_enabled;
569 		break;
570 #endif /* CONFIG_PPC_BOOK3S_64 */
571 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
572 	case KVM_CAP_PPC_SMT:
573 		r = 0;
574 		if (kvm) {
575 			if (kvm->arch.emul_smt_mode > 1)
576 				r = kvm->arch.emul_smt_mode;
577 			else
578 				r = kvm->arch.smt_mode;
579 		} else if (hv_enabled) {
580 			if (cpu_has_feature(CPU_FTR_ARCH_300))
581 				r = 1;
582 			else
583 				r = threads_per_subcore;
584 		}
585 		break;
586 	case KVM_CAP_PPC_SMT_POSSIBLE:
587 		r = 1;
588 		if (hv_enabled) {
589 			if (!cpu_has_feature(CPU_FTR_ARCH_300))
590 				r = ((threads_per_subcore << 1) - 1);
591 			else
592 				/* P9 can emulate dbells, so allow any mode */
593 				r = 8 | 4 | 2 | 1;
594 		}
595 		break;
596 	case KVM_CAP_PPC_RMA:
597 		r = 0;
598 		break;
599 	case KVM_CAP_PPC_HWRNG:
600 		r = kvmppc_hwrng_present();
601 		break;
602 	case KVM_CAP_PPC_MMU_RADIX:
603 		r = !!(hv_enabled && radix_enabled());
604 		break;
605 	case KVM_CAP_PPC_MMU_HASH_V3:
606 		r = !!(hv_enabled && kvmppc_hv_ops->hash_v3_possible &&
607 		       kvmppc_hv_ops->hash_v3_possible());
608 		break;
609 	case KVM_CAP_PPC_NESTED_HV:
610 		r = !!(hv_enabled && kvmppc_hv_ops->enable_nested &&
611 		       !kvmppc_hv_ops->enable_nested(NULL));
612 		break;
613 #endif
614 	case KVM_CAP_SYNC_MMU:
615 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
616 		r = hv_enabled;
617 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
618 		r = 1;
619 #else
620 		r = 0;
621 #endif
622 		break;
623 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
624 	case KVM_CAP_PPC_HTAB_FD:
625 		r = hv_enabled;
626 		break;
627 #endif
628 	case KVM_CAP_NR_VCPUS:
629 		/*
630 		 * Recommending a number of CPUs is somewhat arbitrary; we
631 		 * return the number of present CPUs for -HV (since a host
632 		 * will have secondary threads "offline"), and for other KVM
633 		 * implementations just count online CPUs.
634 		 */
635 		if (hv_enabled)
636 			r = min_t(unsigned int, num_present_cpus(), KVM_MAX_VCPUS);
637 		else
638 			r = min_t(unsigned int, num_online_cpus(), KVM_MAX_VCPUS);
639 		break;
640 	case KVM_CAP_MAX_VCPUS:
641 		r = KVM_MAX_VCPUS;
642 		break;
643 	case KVM_CAP_MAX_VCPU_ID:
644 		r = KVM_MAX_VCPU_IDS;
645 		break;
646 #ifdef CONFIG_PPC_BOOK3S_64
647 	case KVM_CAP_PPC_GET_SMMU_INFO:
648 		r = 1;
649 		break;
650 	case KVM_CAP_SPAPR_MULTITCE:
651 		r = 1;
652 		break;
653 	case KVM_CAP_SPAPR_RESIZE_HPT:
654 		r = !!hv_enabled;
655 		break;
656 #endif
657 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
658 	case KVM_CAP_PPC_FWNMI:
659 		r = hv_enabled;
660 		break;
661 #endif
662 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
663 	case KVM_CAP_PPC_HTM:
664 		r = !!(cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM) ||
665 		     (hv_enabled && cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST));
666 		break;
667 #endif
668 #if defined(CONFIG_KVM_BOOK3S_HV_POSSIBLE)
669 	case KVM_CAP_PPC_SECURE_GUEST:
670 		r = hv_enabled && kvmppc_hv_ops->enable_svm &&
671 			!kvmppc_hv_ops->enable_svm(NULL);
672 		break;
673 	case KVM_CAP_PPC_DAWR1:
674 		r = !!(hv_enabled && kvmppc_hv_ops->enable_dawr1 &&
675 		       !kvmppc_hv_ops->enable_dawr1(NULL));
676 		break;
677 	case KVM_CAP_PPC_RPT_INVALIDATE:
678 		r = 1;
679 		break;
680 #endif
681 	default:
682 		r = 0;
683 		break;
684 	}
685 	return r;
686 
687 }
688 
689 long kvm_arch_dev_ioctl(struct file *filp,
690                         unsigned int ioctl, unsigned long arg)
691 {
692 	return -EINVAL;
693 }
694 
695 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
696 {
697 	kvmppc_core_free_memslot(kvm, slot);
698 }
699 
700 int kvm_arch_prepare_memory_region(struct kvm *kvm,
701 				   const struct kvm_memory_slot *old,
702 				   struct kvm_memory_slot *new,
703 				   enum kvm_mr_change change)
704 {
705 	return kvmppc_core_prepare_memory_region(kvm, old, new, change);
706 }
707 
708 void kvm_arch_commit_memory_region(struct kvm *kvm,
709 				   struct kvm_memory_slot *old,
710 				   const struct kvm_memory_slot *new,
711 				   enum kvm_mr_change change)
712 {
713 	kvmppc_core_commit_memory_region(kvm, old, new, change);
714 }
715 
716 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
717 				   struct kvm_memory_slot *slot)
718 {
719 	kvmppc_core_flush_memslot(kvm, slot);
720 }
721 
722 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
723 {
724 	return 0;
725 }
726 
727 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
728 {
729 	struct kvm_vcpu *vcpu;
730 
731 	vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
732 	kvmppc_decrementer_func(vcpu);
733 
734 	return HRTIMER_NORESTART;
735 }
736 
737 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
738 {
739 	int err;
740 
741 	hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
742 	vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
743 	vcpu->arch.dec_expires = get_tb();
744 
745 #ifdef CONFIG_KVM_EXIT_TIMING
746 	mutex_init(&vcpu->arch.exit_timing_lock);
747 #endif
748 	err = kvmppc_subarch_vcpu_init(vcpu);
749 	if (err)
750 		return err;
751 
752 	err = kvmppc_core_vcpu_create(vcpu);
753 	if (err)
754 		goto out_vcpu_uninit;
755 
756 	rcuwait_init(&vcpu->arch.wait);
757 	vcpu->arch.waitp = &vcpu->arch.wait;
758 	kvmppc_create_vcpu_debugfs(vcpu, vcpu->vcpu_id);
759 	return 0;
760 
761 out_vcpu_uninit:
762 	kvmppc_subarch_vcpu_uninit(vcpu);
763 	return err;
764 }
765 
766 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
767 {
768 }
769 
770 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
771 {
772 	/* Make sure we're not using the vcpu anymore */
773 	hrtimer_cancel(&vcpu->arch.dec_timer);
774 
775 	kvmppc_remove_vcpu_debugfs(vcpu);
776 
777 	switch (vcpu->arch.irq_type) {
778 	case KVMPPC_IRQ_MPIC:
779 		kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
780 		break;
781 	case KVMPPC_IRQ_XICS:
782 		if (xics_on_xive())
783 			kvmppc_xive_cleanup_vcpu(vcpu);
784 		else
785 			kvmppc_xics_free_icp(vcpu);
786 		break;
787 	case KVMPPC_IRQ_XIVE:
788 		kvmppc_xive_native_cleanup_vcpu(vcpu);
789 		break;
790 	}
791 
792 	kvmppc_core_vcpu_free(vcpu);
793 
794 	kvmppc_subarch_vcpu_uninit(vcpu);
795 }
796 
797 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
798 {
799 	return kvmppc_core_pending_dec(vcpu);
800 }
801 
802 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
803 {
804 #ifdef CONFIG_BOOKE
805 	/*
806 	 * vrsave (formerly usprg0) isn't used by Linux, but may
807 	 * be used by the guest.
808 	 *
809 	 * On non-booke this is associated with Altivec and
810 	 * is handled by code in book3s.c.
811 	 */
812 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
813 #endif
814 	kvmppc_core_vcpu_load(vcpu, cpu);
815 }
816 
817 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
818 {
819 	kvmppc_core_vcpu_put(vcpu);
820 #ifdef CONFIG_BOOKE
821 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
822 #endif
823 }
824 
825 /*
826  * irq_bypass_add_producer and irq_bypass_del_producer are only
827  * useful if the architecture supports PCI passthrough.
828  * irq_bypass_stop and irq_bypass_start are not needed and so
829  * kvm_ops are not defined for them.
830  */
831 bool kvm_arch_has_irq_bypass(void)
832 {
833 	return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
834 		(kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
835 }
836 
837 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
838 				     struct irq_bypass_producer *prod)
839 {
840 	struct kvm_kernel_irqfd *irqfd =
841 		container_of(cons, struct kvm_kernel_irqfd, consumer);
842 	struct kvm *kvm = irqfd->kvm;
843 
844 	if (kvm->arch.kvm_ops->irq_bypass_add_producer)
845 		return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
846 
847 	return 0;
848 }
849 
850 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
851 				      struct irq_bypass_producer *prod)
852 {
853 	struct kvm_kernel_irqfd *irqfd =
854 		container_of(cons, struct kvm_kernel_irqfd, consumer);
855 	struct kvm *kvm = irqfd->kvm;
856 
857 	if (kvm->arch.kvm_ops->irq_bypass_del_producer)
858 		kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
859 }
860 
861 #ifdef CONFIG_VSX
862 static inline int kvmppc_get_vsr_dword_offset(int index)
863 {
864 	int offset;
865 
866 	if ((index != 0) && (index != 1))
867 		return -1;
868 
869 #ifdef __BIG_ENDIAN
870 	offset =  index;
871 #else
872 	offset = 1 - index;
873 #endif
874 
875 	return offset;
876 }
877 
878 static inline int kvmppc_get_vsr_word_offset(int index)
879 {
880 	int offset;
881 
882 	if ((index > 3) || (index < 0))
883 		return -1;
884 
885 #ifdef __BIG_ENDIAN
886 	offset = index;
887 #else
888 	offset = 3 - index;
889 #endif
890 	return offset;
891 }
892 
893 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
894 	u64 gpr)
895 {
896 	union kvmppc_one_reg val;
897 	int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
898 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
899 
900 	if (offset == -1)
901 		return;
902 
903 	if (index >= 32) {
904 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
905 		val.vsxval[offset] = gpr;
906 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
907 	} else {
908 		VCPU_VSX_FPR(vcpu, index, offset) = gpr;
909 	}
910 }
911 
912 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
913 	u64 gpr)
914 {
915 	union kvmppc_one_reg val;
916 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
917 
918 	if (index >= 32) {
919 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
920 		val.vsxval[0] = gpr;
921 		val.vsxval[1] = gpr;
922 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
923 	} else {
924 		VCPU_VSX_FPR(vcpu, index, 0) = gpr;
925 		VCPU_VSX_FPR(vcpu, index, 1) = gpr;
926 	}
927 }
928 
929 static inline void kvmppc_set_vsr_word_dump(struct kvm_vcpu *vcpu,
930 	u32 gpr)
931 {
932 	union kvmppc_one_reg val;
933 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
934 
935 	if (index >= 32) {
936 		val.vsx32val[0] = gpr;
937 		val.vsx32val[1] = gpr;
938 		val.vsx32val[2] = gpr;
939 		val.vsx32val[3] = gpr;
940 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
941 	} else {
942 		val.vsx32val[0] = gpr;
943 		val.vsx32val[1] = gpr;
944 		VCPU_VSX_FPR(vcpu, index, 0) = val.vsxval[0];
945 		VCPU_VSX_FPR(vcpu, index, 1) = val.vsxval[0];
946 	}
947 }
948 
949 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
950 	u32 gpr32)
951 {
952 	union kvmppc_one_reg val;
953 	int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
954 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
955 	int dword_offset, word_offset;
956 
957 	if (offset == -1)
958 		return;
959 
960 	if (index >= 32) {
961 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
962 		val.vsx32val[offset] = gpr32;
963 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
964 	} else {
965 		dword_offset = offset / 2;
966 		word_offset = offset % 2;
967 		val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
968 		val.vsx32val[word_offset] = gpr32;
969 		VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
970 	}
971 }
972 #endif /* CONFIG_VSX */
973 
974 #ifdef CONFIG_ALTIVEC
975 static inline int kvmppc_get_vmx_offset_generic(struct kvm_vcpu *vcpu,
976 		int index, int element_size)
977 {
978 	int offset;
979 	int elts = sizeof(vector128)/element_size;
980 
981 	if ((index < 0) || (index >= elts))
982 		return -1;
983 
984 	if (kvmppc_need_byteswap(vcpu))
985 		offset = elts - index - 1;
986 	else
987 		offset = index;
988 
989 	return offset;
990 }
991 
992 static inline int kvmppc_get_vmx_dword_offset(struct kvm_vcpu *vcpu,
993 		int index)
994 {
995 	return kvmppc_get_vmx_offset_generic(vcpu, index, 8);
996 }
997 
998 static inline int kvmppc_get_vmx_word_offset(struct kvm_vcpu *vcpu,
999 		int index)
1000 {
1001 	return kvmppc_get_vmx_offset_generic(vcpu, index, 4);
1002 }
1003 
1004 static inline int kvmppc_get_vmx_hword_offset(struct kvm_vcpu *vcpu,
1005 		int index)
1006 {
1007 	return kvmppc_get_vmx_offset_generic(vcpu, index, 2);
1008 }
1009 
1010 static inline int kvmppc_get_vmx_byte_offset(struct kvm_vcpu *vcpu,
1011 		int index)
1012 {
1013 	return kvmppc_get_vmx_offset_generic(vcpu, index, 1);
1014 }
1015 
1016 
1017 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu,
1018 	u64 gpr)
1019 {
1020 	union kvmppc_one_reg val;
1021 	int offset = kvmppc_get_vmx_dword_offset(vcpu,
1022 			vcpu->arch.mmio_vmx_offset);
1023 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1024 
1025 	if (offset == -1)
1026 		return;
1027 
1028 	val.vval = VCPU_VSX_VR(vcpu, index);
1029 	val.vsxval[offset] = gpr;
1030 	VCPU_VSX_VR(vcpu, index) = val.vval;
1031 }
1032 
1033 static inline void kvmppc_set_vmx_word(struct kvm_vcpu *vcpu,
1034 	u32 gpr32)
1035 {
1036 	union kvmppc_one_reg val;
1037 	int offset = kvmppc_get_vmx_word_offset(vcpu,
1038 			vcpu->arch.mmio_vmx_offset);
1039 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1040 
1041 	if (offset == -1)
1042 		return;
1043 
1044 	val.vval = VCPU_VSX_VR(vcpu, index);
1045 	val.vsx32val[offset] = gpr32;
1046 	VCPU_VSX_VR(vcpu, index) = val.vval;
1047 }
1048 
1049 static inline void kvmppc_set_vmx_hword(struct kvm_vcpu *vcpu,
1050 	u16 gpr16)
1051 {
1052 	union kvmppc_one_reg val;
1053 	int offset = kvmppc_get_vmx_hword_offset(vcpu,
1054 			vcpu->arch.mmio_vmx_offset);
1055 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1056 
1057 	if (offset == -1)
1058 		return;
1059 
1060 	val.vval = VCPU_VSX_VR(vcpu, index);
1061 	val.vsx16val[offset] = gpr16;
1062 	VCPU_VSX_VR(vcpu, index) = val.vval;
1063 }
1064 
1065 static inline void kvmppc_set_vmx_byte(struct kvm_vcpu *vcpu,
1066 	u8 gpr8)
1067 {
1068 	union kvmppc_one_reg val;
1069 	int offset = kvmppc_get_vmx_byte_offset(vcpu,
1070 			vcpu->arch.mmio_vmx_offset);
1071 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1072 
1073 	if (offset == -1)
1074 		return;
1075 
1076 	val.vval = VCPU_VSX_VR(vcpu, index);
1077 	val.vsx8val[offset] = gpr8;
1078 	VCPU_VSX_VR(vcpu, index) = val.vval;
1079 }
1080 #endif /* CONFIG_ALTIVEC */
1081 
1082 #ifdef CONFIG_PPC_FPU
1083 static inline u64 sp_to_dp(u32 fprs)
1084 {
1085 	u64 fprd;
1086 
1087 	preempt_disable();
1088 	enable_kernel_fp();
1089 	asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m<>" (fprd) : "m<>" (fprs)
1090 	     : "fr0");
1091 	preempt_enable();
1092 	return fprd;
1093 }
1094 
1095 static inline u32 dp_to_sp(u64 fprd)
1096 {
1097 	u32 fprs;
1098 
1099 	preempt_disable();
1100 	enable_kernel_fp();
1101 	asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m<>" (fprs) : "m<>" (fprd)
1102 	     : "fr0");
1103 	preempt_enable();
1104 	return fprs;
1105 }
1106 
1107 #else
1108 #define sp_to_dp(x)	(x)
1109 #define dp_to_sp(x)	(x)
1110 #endif /* CONFIG_PPC_FPU */
1111 
1112 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu)
1113 {
1114 	struct kvm_run *run = vcpu->run;
1115 	u64 gpr;
1116 
1117 	if (run->mmio.len > sizeof(gpr)) {
1118 		printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
1119 		return;
1120 	}
1121 
1122 	if (!vcpu->arch.mmio_host_swabbed) {
1123 		switch (run->mmio.len) {
1124 		case 8: gpr = *(u64 *)run->mmio.data; break;
1125 		case 4: gpr = *(u32 *)run->mmio.data; break;
1126 		case 2: gpr = *(u16 *)run->mmio.data; break;
1127 		case 1: gpr = *(u8 *)run->mmio.data; break;
1128 		}
1129 	} else {
1130 		switch (run->mmio.len) {
1131 		case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
1132 		case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
1133 		case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
1134 		case 1: gpr = *(u8 *)run->mmio.data; break;
1135 		}
1136 	}
1137 
1138 	/* conversion between single and double precision */
1139 	if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
1140 		gpr = sp_to_dp(gpr);
1141 
1142 	if (vcpu->arch.mmio_sign_extend) {
1143 		switch (run->mmio.len) {
1144 #ifdef CONFIG_PPC64
1145 		case 4:
1146 			gpr = (s64)(s32)gpr;
1147 			break;
1148 #endif
1149 		case 2:
1150 			gpr = (s64)(s16)gpr;
1151 			break;
1152 		case 1:
1153 			gpr = (s64)(s8)gpr;
1154 			break;
1155 		}
1156 	}
1157 
1158 	switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1159 	case KVM_MMIO_REG_GPR:
1160 		kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1161 		break;
1162 	case KVM_MMIO_REG_FPR:
1163 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1164 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_FP);
1165 
1166 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1167 		break;
1168 #ifdef CONFIG_PPC_BOOK3S
1169 	case KVM_MMIO_REG_QPR:
1170 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1171 		break;
1172 	case KVM_MMIO_REG_FQPR:
1173 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1174 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1175 		break;
1176 #endif
1177 #ifdef CONFIG_VSX
1178 	case KVM_MMIO_REG_VSX:
1179 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1180 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VSX);
1181 
1182 		if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_DWORD)
1183 			kvmppc_set_vsr_dword(vcpu, gpr);
1184 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_WORD)
1185 			kvmppc_set_vsr_word(vcpu, gpr);
1186 		else if (vcpu->arch.mmio_copy_type ==
1187 				KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1188 			kvmppc_set_vsr_dword_dump(vcpu, gpr);
1189 		else if (vcpu->arch.mmio_copy_type ==
1190 				KVMPPC_VSX_COPY_WORD_LOAD_DUMP)
1191 			kvmppc_set_vsr_word_dump(vcpu, gpr);
1192 		break;
1193 #endif
1194 #ifdef CONFIG_ALTIVEC
1195 	case KVM_MMIO_REG_VMX:
1196 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1197 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VEC);
1198 
1199 		if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_DWORD)
1200 			kvmppc_set_vmx_dword(vcpu, gpr);
1201 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_WORD)
1202 			kvmppc_set_vmx_word(vcpu, gpr);
1203 		else if (vcpu->arch.mmio_copy_type ==
1204 				KVMPPC_VMX_COPY_HWORD)
1205 			kvmppc_set_vmx_hword(vcpu, gpr);
1206 		else if (vcpu->arch.mmio_copy_type ==
1207 				KVMPPC_VMX_COPY_BYTE)
1208 			kvmppc_set_vmx_byte(vcpu, gpr);
1209 		break;
1210 #endif
1211 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1212 	case KVM_MMIO_REG_NESTED_GPR:
1213 		if (kvmppc_need_byteswap(vcpu))
1214 			gpr = swab64(gpr);
1215 		kvm_vcpu_write_guest(vcpu, vcpu->arch.nested_io_gpr, &gpr,
1216 				     sizeof(gpr));
1217 		break;
1218 #endif
1219 	default:
1220 		BUG();
1221 	}
1222 }
1223 
1224 static int __kvmppc_handle_load(struct kvm_vcpu *vcpu,
1225 				unsigned int rt, unsigned int bytes,
1226 				int is_default_endian, int sign_extend)
1227 {
1228 	struct kvm_run *run = vcpu->run;
1229 	int idx, ret;
1230 	bool host_swabbed;
1231 
1232 	/* Pity C doesn't have a logical XOR operator */
1233 	if (kvmppc_need_byteswap(vcpu)) {
1234 		host_swabbed = is_default_endian;
1235 	} else {
1236 		host_swabbed = !is_default_endian;
1237 	}
1238 
1239 	if (bytes > sizeof(run->mmio.data)) {
1240 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1241 		       run->mmio.len);
1242 	}
1243 
1244 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1245 	run->mmio.len = bytes;
1246 	run->mmio.is_write = 0;
1247 
1248 	vcpu->arch.io_gpr = rt;
1249 	vcpu->arch.mmio_host_swabbed = host_swabbed;
1250 	vcpu->mmio_needed = 1;
1251 	vcpu->mmio_is_write = 0;
1252 	vcpu->arch.mmio_sign_extend = sign_extend;
1253 
1254 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1255 
1256 	ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1257 			      bytes, &run->mmio.data);
1258 
1259 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1260 
1261 	if (!ret) {
1262 		kvmppc_complete_mmio_load(vcpu);
1263 		vcpu->mmio_needed = 0;
1264 		return EMULATE_DONE;
1265 	}
1266 
1267 	return EMULATE_DO_MMIO;
1268 }
1269 
1270 int kvmppc_handle_load(struct kvm_vcpu *vcpu,
1271 		       unsigned int rt, unsigned int bytes,
1272 		       int is_default_endian)
1273 {
1274 	return __kvmppc_handle_load(vcpu, rt, bytes, is_default_endian, 0);
1275 }
1276 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1277 
1278 /* Same as above, but sign extends */
1279 int kvmppc_handle_loads(struct kvm_vcpu *vcpu,
1280 			unsigned int rt, unsigned int bytes,
1281 			int is_default_endian)
1282 {
1283 	return __kvmppc_handle_load(vcpu, rt, bytes, is_default_endian, 1);
1284 }
1285 
1286 #ifdef CONFIG_VSX
1287 int kvmppc_handle_vsx_load(struct kvm_vcpu *vcpu,
1288 			unsigned int rt, unsigned int bytes,
1289 			int is_default_endian, int mmio_sign_extend)
1290 {
1291 	enum emulation_result emulated = EMULATE_DONE;
1292 
1293 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1294 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1295 		return EMULATE_FAIL;
1296 
1297 	while (vcpu->arch.mmio_vsx_copy_nums) {
1298 		emulated = __kvmppc_handle_load(vcpu, rt, bytes,
1299 			is_default_endian, mmio_sign_extend);
1300 
1301 		if (emulated != EMULATE_DONE)
1302 			break;
1303 
1304 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1305 
1306 		vcpu->arch.mmio_vsx_copy_nums--;
1307 		vcpu->arch.mmio_vsx_offset++;
1308 	}
1309 	return emulated;
1310 }
1311 #endif /* CONFIG_VSX */
1312 
1313 int kvmppc_handle_store(struct kvm_vcpu *vcpu,
1314 			u64 val, unsigned int bytes, int is_default_endian)
1315 {
1316 	struct kvm_run *run = vcpu->run;
1317 	void *data = run->mmio.data;
1318 	int idx, ret;
1319 	bool host_swabbed;
1320 
1321 	/* Pity C doesn't have a logical XOR operator */
1322 	if (kvmppc_need_byteswap(vcpu)) {
1323 		host_swabbed = is_default_endian;
1324 	} else {
1325 		host_swabbed = !is_default_endian;
1326 	}
1327 
1328 	if (bytes > sizeof(run->mmio.data)) {
1329 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1330 		       run->mmio.len);
1331 	}
1332 
1333 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1334 	run->mmio.len = bytes;
1335 	run->mmio.is_write = 1;
1336 	vcpu->mmio_needed = 1;
1337 	vcpu->mmio_is_write = 1;
1338 
1339 	if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1340 		val = dp_to_sp(val);
1341 
1342 	/* Store the value at the lowest bytes in 'data'. */
1343 	if (!host_swabbed) {
1344 		switch (bytes) {
1345 		case 8: *(u64 *)data = val; break;
1346 		case 4: *(u32 *)data = val; break;
1347 		case 2: *(u16 *)data = val; break;
1348 		case 1: *(u8  *)data = val; break;
1349 		}
1350 	} else {
1351 		switch (bytes) {
1352 		case 8: *(u64 *)data = swab64(val); break;
1353 		case 4: *(u32 *)data = swab32(val); break;
1354 		case 2: *(u16 *)data = swab16(val); break;
1355 		case 1: *(u8  *)data = val; break;
1356 		}
1357 	}
1358 
1359 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1360 
1361 	ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1362 			       bytes, &run->mmio.data);
1363 
1364 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1365 
1366 	if (!ret) {
1367 		vcpu->mmio_needed = 0;
1368 		return EMULATE_DONE;
1369 	}
1370 
1371 	return EMULATE_DO_MMIO;
1372 }
1373 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1374 
1375 #ifdef CONFIG_VSX
1376 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1377 {
1378 	u32 dword_offset, word_offset;
1379 	union kvmppc_one_reg reg;
1380 	int vsx_offset = 0;
1381 	int copy_type = vcpu->arch.mmio_copy_type;
1382 	int result = 0;
1383 
1384 	switch (copy_type) {
1385 	case KVMPPC_VSX_COPY_DWORD:
1386 		vsx_offset =
1387 			kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1388 
1389 		if (vsx_offset == -1) {
1390 			result = -1;
1391 			break;
1392 		}
1393 
1394 		if (rs < 32) {
1395 			*val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1396 		} else {
1397 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1398 			*val = reg.vsxval[vsx_offset];
1399 		}
1400 		break;
1401 
1402 	case KVMPPC_VSX_COPY_WORD:
1403 		vsx_offset =
1404 			kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1405 
1406 		if (vsx_offset == -1) {
1407 			result = -1;
1408 			break;
1409 		}
1410 
1411 		if (rs < 32) {
1412 			dword_offset = vsx_offset / 2;
1413 			word_offset = vsx_offset % 2;
1414 			reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1415 			*val = reg.vsx32val[word_offset];
1416 		} else {
1417 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1418 			*val = reg.vsx32val[vsx_offset];
1419 		}
1420 		break;
1421 
1422 	default:
1423 		result = -1;
1424 		break;
1425 	}
1426 
1427 	return result;
1428 }
1429 
1430 int kvmppc_handle_vsx_store(struct kvm_vcpu *vcpu,
1431 			int rs, unsigned int bytes, int is_default_endian)
1432 {
1433 	u64 val;
1434 	enum emulation_result emulated = EMULATE_DONE;
1435 
1436 	vcpu->arch.io_gpr = rs;
1437 
1438 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1439 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1440 		return EMULATE_FAIL;
1441 
1442 	while (vcpu->arch.mmio_vsx_copy_nums) {
1443 		if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1444 			return EMULATE_FAIL;
1445 
1446 		emulated = kvmppc_handle_store(vcpu,
1447 			 val, bytes, is_default_endian);
1448 
1449 		if (emulated != EMULATE_DONE)
1450 			break;
1451 
1452 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1453 
1454 		vcpu->arch.mmio_vsx_copy_nums--;
1455 		vcpu->arch.mmio_vsx_offset++;
1456 	}
1457 
1458 	return emulated;
1459 }
1460 
1461 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu)
1462 {
1463 	struct kvm_run *run = vcpu->run;
1464 	enum emulation_result emulated = EMULATE_FAIL;
1465 	int r;
1466 
1467 	vcpu->arch.paddr_accessed += run->mmio.len;
1468 
1469 	if (!vcpu->mmio_is_write) {
1470 		emulated = kvmppc_handle_vsx_load(vcpu, vcpu->arch.io_gpr,
1471 			 run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1472 	} else {
1473 		emulated = kvmppc_handle_vsx_store(vcpu,
1474 			 vcpu->arch.io_gpr, run->mmio.len, 1);
1475 	}
1476 
1477 	switch (emulated) {
1478 	case EMULATE_DO_MMIO:
1479 		run->exit_reason = KVM_EXIT_MMIO;
1480 		r = RESUME_HOST;
1481 		break;
1482 	case EMULATE_FAIL:
1483 		pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1484 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1485 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1486 		r = RESUME_HOST;
1487 		break;
1488 	default:
1489 		r = RESUME_GUEST;
1490 		break;
1491 	}
1492 	return r;
1493 }
1494 #endif /* CONFIG_VSX */
1495 
1496 #ifdef CONFIG_ALTIVEC
1497 int kvmppc_handle_vmx_load(struct kvm_vcpu *vcpu,
1498 		unsigned int rt, unsigned int bytes, int is_default_endian)
1499 {
1500 	enum emulation_result emulated = EMULATE_DONE;
1501 
1502 	if (vcpu->arch.mmio_vsx_copy_nums > 2)
1503 		return EMULATE_FAIL;
1504 
1505 	while (vcpu->arch.mmio_vmx_copy_nums) {
1506 		emulated = __kvmppc_handle_load(vcpu, rt, bytes,
1507 				is_default_endian, 0);
1508 
1509 		if (emulated != EMULATE_DONE)
1510 			break;
1511 
1512 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1513 		vcpu->arch.mmio_vmx_copy_nums--;
1514 		vcpu->arch.mmio_vmx_offset++;
1515 	}
1516 
1517 	return emulated;
1518 }
1519 
1520 static int kvmppc_get_vmx_dword(struct kvm_vcpu *vcpu, int index, u64 *val)
1521 {
1522 	union kvmppc_one_reg reg;
1523 	int vmx_offset = 0;
1524 	int result = 0;
1525 
1526 	vmx_offset =
1527 		kvmppc_get_vmx_dword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1528 
1529 	if (vmx_offset == -1)
1530 		return -1;
1531 
1532 	reg.vval = VCPU_VSX_VR(vcpu, index);
1533 	*val = reg.vsxval[vmx_offset];
1534 
1535 	return result;
1536 }
1537 
1538 static int kvmppc_get_vmx_word(struct kvm_vcpu *vcpu, int index, u64 *val)
1539 {
1540 	union kvmppc_one_reg reg;
1541 	int vmx_offset = 0;
1542 	int result = 0;
1543 
1544 	vmx_offset =
1545 		kvmppc_get_vmx_word_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1546 
1547 	if (vmx_offset == -1)
1548 		return -1;
1549 
1550 	reg.vval = VCPU_VSX_VR(vcpu, index);
1551 	*val = reg.vsx32val[vmx_offset];
1552 
1553 	return result;
1554 }
1555 
1556 static int kvmppc_get_vmx_hword(struct kvm_vcpu *vcpu, int index, u64 *val)
1557 {
1558 	union kvmppc_one_reg reg;
1559 	int vmx_offset = 0;
1560 	int result = 0;
1561 
1562 	vmx_offset =
1563 		kvmppc_get_vmx_hword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1564 
1565 	if (vmx_offset == -1)
1566 		return -1;
1567 
1568 	reg.vval = VCPU_VSX_VR(vcpu, index);
1569 	*val = reg.vsx16val[vmx_offset];
1570 
1571 	return result;
1572 }
1573 
1574 static int kvmppc_get_vmx_byte(struct kvm_vcpu *vcpu, int index, u64 *val)
1575 {
1576 	union kvmppc_one_reg reg;
1577 	int vmx_offset = 0;
1578 	int result = 0;
1579 
1580 	vmx_offset =
1581 		kvmppc_get_vmx_byte_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1582 
1583 	if (vmx_offset == -1)
1584 		return -1;
1585 
1586 	reg.vval = VCPU_VSX_VR(vcpu, index);
1587 	*val = reg.vsx8val[vmx_offset];
1588 
1589 	return result;
1590 }
1591 
1592 int kvmppc_handle_vmx_store(struct kvm_vcpu *vcpu,
1593 		unsigned int rs, unsigned int bytes, int is_default_endian)
1594 {
1595 	u64 val = 0;
1596 	unsigned int index = rs & KVM_MMIO_REG_MASK;
1597 	enum emulation_result emulated = EMULATE_DONE;
1598 
1599 	if (vcpu->arch.mmio_vsx_copy_nums > 2)
1600 		return EMULATE_FAIL;
1601 
1602 	vcpu->arch.io_gpr = rs;
1603 
1604 	while (vcpu->arch.mmio_vmx_copy_nums) {
1605 		switch (vcpu->arch.mmio_copy_type) {
1606 		case KVMPPC_VMX_COPY_DWORD:
1607 			if (kvmppc_get_vmx_dword(vcpu, index, &val) == -1)
1608 				return EMULATE_FAIL;
1609 
1610 			break;
1611 		case KVMPPC_VMX_COPY_WORD:
1612 			if (kvmppc_get_vmx_word(vcpu, index, &val) == -1)
1613 				return EMULATE_FAIL;
1614 			break;
1615 		case KVMPPC_VMX_COPY_HWORD:
1616 			if (kvmppc_get_vmx_hword(vcpu, index, &val) == -1)
1617 				return EMULATE_FAIL;
1618 			break;
1619 		case KVMPPC_VMX_COPY_BYTE:
1620 			if (kvmppc_get_vmx_byte(vcpu, index, &val) == -1)
1621 				return EMULATE_FAIL;
1622 			break;
1623 		default:
1624 			return EMULATE_FAIL;
1625 		}
1626 
1627 		emulated = kvmppc_handle_store(vcpu, val, bytes,
1628 				is_default_endian);
1629 		if (emulated != EMULATE_DONE)
1630 			break;
1631 
1632 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1633 		vcpu->arch.mmio_vmx_copy_nums--;
1634 		vcpu->arch.mmio_vmx_offset++;
1635 	}
1636 
1637 	return emulated;
1638 }
1639 
1640 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu)
1641 {
1642 	struct kvm_run *run = vcpu->run;
1643 	enum emulation_result emulated = EMULATE_FAIL;
1644 	int r;
1645 
1646 	vcpu->arch.paddr_accessed += run->mmio.len;
1647 
1648 	if (!vcpu->mmio_is_write) {
1649 		emulated = kvmppc_handle_vmx_load(vcpu,
1650 				vcpu->arch.io_gpr, run->mmio.len, 1);
1651 	} else {
1652 		emulated = kvmppc_handle_vmx_store(vcpu,
1653 				vcpu->arch.io_gpr, run->mmio.len, 1);
1654 	}
1655 
1656 	switch (emulated) {
1657 	case EMULATE_DO_MMIO:
1658 		run->exit_reason = KVM_EXIT_MMIO;
1659 		r = RESUME_HOST;
1660 		break;
1661 	case EMULATE_FAIL:
1662 		pr_info("KVM: MMIO emulation failed (VMX repeat)\n");
1663 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1664 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1665 		r = RESUME_HOST;
1666 		break;
1667 	default:
1668 		r = RESUME_GUEST;
1669 		break;
1670 	}
1671 	return r;
1672 }
1673 #endif /* CONFIG_ALTIVEC */
1674 
1675 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1676 {
1677 	int r = 0;
1678 	union kvmppc_one_reg val;
1679 	int size;
1680 
1681 	size = one_reg_size(reg->id);
1682 	if (size > sizeof(val))
1683 		return -EINVAL;
1684 
1685 	r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1686 	if (r == -EINVAL) {
1687 		r = 0;
1688 		switch (reg->id) {
1689 #ifdef CONFIG_ALTIVEC
1690 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1691 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1692 				r = -ENXIO;
1693 				break;
1694 			}
1695 			val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1696 			break;
1697 		case KVM_REG_PPC_VSCR:
1698 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1699 				r = -ENXIO;
1700 				break;
1701 			}
1702 			val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1703 			break;
1704 		case KVM_REG_PPC_VRSAVE:
1705 			val = get_reg_val(reg->id, vcpu->arch.vrsave);
1706 			break;
1707 #endif /* CONFIG_ALTIVEC */
1708 		default:
1709 			r = -EINVAL;
1710 			break;
1711 		}
1712 	}
1713 
1714 	if (r)
1715 		return r;
1716 
1717 	if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1718 		r = -EFAULT;
1719 
1720 	return r;
1721 }
1722 
1723 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1724 {
1725 	int r;
1726 	union kvmppc_one_reg val;
1727 	int size;
1728 
1729 	size = one_reg_size(reg->id);
1730 	if (size > sizeof(val))
1731 		return -EINVAL;
1732 
1733 	if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1734 		return -EFAULT;
1735 
1736 	r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1737 	if (r == -EINVAL) {
1738 		r = 0;
1739 		switch (reg->id) {
1740 #ifdef CONFIG_ALTIVEC
1741 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1742 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1743 				r = -ENXIO;
1744 				break;
1745 			}
1746 			vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1747 			break;
1748 		case KVM_REG_PPC_VSCR:
1749 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1750 				r = -ENXIO;
1751 				break;
1752 			}
1753 			vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1754 			break;
1755 		case KVM_REG_PPC_VRSAVE:
1756 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1757 				r = -ENXIO;
1758 				break;
1759 			}
1760 			vcpu->arch.vrsave = set_reg_val(reg->id, val);
1761 			break;
1762 #endif /* CONFIG_ALTIVEC */
1763 		default:
1764 			r = -EINVAL;
1765 			break;
1766 		}
1767 	}
1768 
1769 	return r;
1770 }
1771 
1772 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
1773 {
1774 	struct kvm_run *run = vcpu->run;
1775 	int r;
1776 
1777 	vcpu_load(vcpu);
1778 
1779 	if (vcpu->mmio_needed) {
1780 		vcpu->mmio_needed = 0;
1781 		if (!vcpu->mmio_is_write)
1782 			kvmppc_complete_mmio_load(vcpu);
1783 #ifdef CONFIG_VSX
1784 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1785 			vcpu->arch.mmio_vsx_copy_nums--;
1786 			vcpu->arch.mmio_vsx_offset++;
1787 		}
1788 
1789 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1790 			r = kvmppc_emulate_mmio_vsx_loadstore(vcpu);
1791 			if (r == RESUME_HOST) {
1792 				vcpu->mmio_needed = 1;
1793 				goto out;
1794 			}
1795 		}
1796 #endif
1797 #ifdef CONFIG_ALTIVEC
1798 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1799 			vcpu->arch.mmio_vmx_copy_nums--;
1800 			vcpu->arch.mmio_vmx_offset++;
1801 		}
1802 
1803 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1804 			r = kvmppc_emulate_mmio_vmx_loadstore(vcpu);
1805 			if (r == RESUME_HOST) {
1806 				vcpu->mmio_needed = 1;
1807 				goto out;
1808 			}
1809 		}
1810 #endif
1811 	} else if (vcpu->arch.osi_needed) {
1812 		u64 *gprs = run->osi.gprs;
1813 		int i;
1814 
1815 		for (i = 0; i < 32; i++)
1816 			kvmppc_set_gpr(vcpu, i, gprs[i]);
1817 		vcpu->arch.osi_needed = 0;
1818 	} else if (vcpu->arch.hcall_needed) {
1819 		int i;
1820 
1821 		kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1822 		for (i = 0; i < 9; ++i)
1823 			kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1824 		vcpu->arch.hcall_needed = 0;
1825 #ifdef CONFIG_BOOKE
1826 	} else if (vcpu->arch.epr_needed) {
1827 		kvmppc_set_epr(vcpu, run->epr.epr);
1828 		vcpu->arch.epr_needed = 0;
1829 #endif
1830 	}
1831 
1832 	kvm_sigset_activate(vcpu);
1833 
1834 	if (run->immediate_exit)
1835 		r = -EINTR;
1836 	else
1837 		r = kvmppc_vcpu_run(vcpu);
1838 
1839 	kvm_sigset_deactivate(vcpu);
1840 
1841 #ifdef CONFIG_ALTIVEC
1842 out:
1843 #endif
1844 	vcpu_put(vcpu);
1845 	return r;
1846 }
1847 
1848 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1849 {
1850 	if (irq->irq == KVM_INTERRUPT_UNSET) {
1851 		kvmppc_core_dequeue_external(vcpu);
1852 		return 0;
1853 	}
1854 
1855 	kvmppc_core_queue_external(vcpu, irq);
1856 
1857 	kvm_vcpu_kick(vcpu);
1858 
1859 	return 0;
1860 }
1861 
1862 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1863 				     struct kvm_enable_cap *cap)
1864 {
1865 	int r;
1866 
1867 	if (cap->flags)
1868 		return -EINVAL;
1869 
1870 	switch (cap->cap) {
1871 	case KVM_CAP_PPC_OSI:
1872 		r = 0;
1873 		vcpu->arch.osi_enabled = true;
1874 		break;
1875 	case KVM_CAP_PPC_PAPR:
1876 		r = 0;
1877 		vcpu->arch.papr_enabled = true;
1878 		break;
1879 	case KVM_CAP_PPC_EPR:
1880 		r = 0;
1881 		if (cap->args[0])
1882 			vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1883 		else
1884 			vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1885 		break;
1886 #ifdef CONFIG_BOOKE
1887 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
1888 		r = 0;
1889 		vcpu->arch.watchdog_enabled = true;
1890 		break;
1891 #endif
1892 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1893 	case KVM_CAP_SW_TLB: {
1894 		struct kvm_config_tlb cfg;
1895 		void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1896 
1897 		r = -EFAULT;
1898 		if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1899 			break;
1900 
1901 		r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1902 		break;
1903 	}
1904 #endif
1905 #ifdef CONFIG_KVM_MPIC
1906 	case KVM_CAP_IRQ_MPIC: {
1907 		struct fd f;
1908 		struct kvm_device *dev;
1909 
1910 		r = -EBADF;
1911 		f = fdget(cap->args[0]);
1912 		if (!f.file)
1913 			break;
1914 
1915 		r = -EPERM;
1916 		dev = kvm_device_from_filp(f.file);
1917 		if (dev)
1918 			r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1919 
1920 		fdput(f);
1921 		break;
1922 	}
1923 #endif
1924 #ifdef CONFIG_KVM_XICS
1925 	case KVM_CAP_IRQ_XICS: {
1926 		struct fd f;
1927 		struct kvm_device *dev;
1928 
1929 		r = -EBADF;
1930 		f = fdget(cap->args[0]);
1931 		if (!f.file)
1932 			break;
1933 
1934 		r = -EPERM;
1935 		dev = kvm_device_from_filp(f.file);
1936 		if (dev) {
1937 			if (xics_on_xive())
1938 				r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1939 			else
1940 				r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1941 		}
1942 
1943 		fdput(f);
1944 		break;
1945 	}
1946 #endif /* CONFIG_KVM_XICS */
1947 #ifdef CONFIG_KVM_XIVE
1948 	case KVM_CAP_PPC_IRQ_XIVE: {
1949 		struct fd f;
1950 		struct kvm_device *dev;
1951 
1952 		r = -EBADF;
1953 		f = fdget(cap->args[0]);
1954 		if (!f.file)
1955 			break;
1956 
1957 		r = -ENXIO;
1958 		if (!xive_enabled())
1959 			break;
1960 
1961 		r = -EPERM;
1962 		dev = kvm_device_from_filp(f.file);
1963 		if (dev)
1964 			r = kvmppc_xive_native_connect_vcpu(dev, vcpu,
1965 							    cap->args[1]);
1966 
1967 		fdput(f);
1968 		break;
1969 	}
1970 #endif /* CONFIG_KVM_XIVE */
1971 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1972 	case KVM_CAP_PPC_FWNMI:
1973 		r = -EINVAL;
1974 		if (!is_kvmppc_hv_enabled(vcpu->kvm))
1975 			break;
1976 		r = 0;
1977 		vcpu->kvm->arch.fwnmi_enabled = true;
1978 		break;
1979 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1980 	default:
1981 		r = -EINVAL;
1982 		break;
1983 	}
1984 
1985 	if (!r)
1986 		r = kvmppc_sanity_check(vcpu);
1987 
1988 	return r;
1989 }
1990 
1991 bool kvm_arch_intc_initialized(struct kvm *kvm)
1992 {
1993 #ifdef CONFIG_KVM_MPIC
1994 	if (kvm->arch.mpic)
1995 		return true;
1996 #endif
1997 #ifdef CONFIG_KVM_XICS
1998 	if (kvm->arch.xics || kvm->arch.xive)
1999 		return true;
2000 #endif
2001 	return false;
2002 }
2003 
2004 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2005                                     struct kvm_mp_state *mp_state)
2006 {
2007 	return -EINVAL;
2008 }
2009 
2010 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2011                                     struct kvm_mp_state *mp_state)
2012 {
2013 	return -EINVAL;
2014 }
2015 
2016 long kvm_arch_vcpu_async_ioctl(struct file *filp,
2017 			       unsigned int ioctl, unsigned long arg)
2018 {
2019 	struct kvm_vcpu *vcpu = filp->private_data;
2020 	void __user *argp = (void __user *)arg;
2021 
2022 	if (ioctl == KVM_INTERRUPT) {
2023 		struct kvm_interrupt irq;
2024 		if (copy_from_user(&irq, argp, sizeof(irq)))
2025 			return -EFAULT;
2026 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2027 	}
2028 	return -ENOIOCTLCMD;
2029 }
2030 
2031 long kvm_arch_vcpu_ioctl(struct file *filp,
2032                          unsigned int ioctl, unsigned long arg)
2033 {
2034 	struct kvm_vcpu *vcpu = filp->private_data;
2035 	void __user *argp = (void __user *)arg;
2036 	long r;
2037 
2038 	switch (ioctl) {
2039 	case KVM_ENABLE_CAP:
2040 	{
2041 		struct kvm_enable_cap cap;
2042 		r = -EFAULT;
2043 		if (copy_from_user(&cap, argp, sizeof(cap)))
2044 			goto out;
2045 		vcpu_load(vcpu);
2046 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
2047 		vcpu_put(vcpu);
2048 		break;
2049 	}
2050 
2051 	case KVM_SET_ONE_REG:
2052 	case KVM_GET_ONE_REG:
2053 	{
2054 		struct kvm_one_reg reg;
2055 		r = -EFAULT;
2056 		if (copy_from_user(&reg, argp, sizeof(reg)))
2057 			goto out;
2058 		if (ioctl == KVM_SET_ONE_REG)
2059 			r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
2060 		else
2061 			r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
2062 		break;
2063 	}
2064 
2065 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
2066 	case KVM_DIRTY_TLB: {
2067 		struct kvm_dirty_tlb dirty;
2068 		r = -EFAULT;
2069 		if (copy_from_user(&dirty, argp, sizeof(dirty)))
2070 			goto out;
2071 		vcpu_load(vcpu);
2072 		r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
2073 		vcpu_put(vcpu);
2074 		break;
2075 	}
2076 #endif
2077 	default:
2078 		r = -EINVAL;
2079 	}
2080 
2081 out:
2082 	return r;
2083 }
2084 
2085 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2086 {
2087 	return VM_FAULT_SIGBUS;
2088 }
2089 
2090 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
2091 {
2092 	u32 inst_nop = 0x60000000;
2093 #ifdef CONFIG_KVM_BOOKE_HV
2094 	u32 inst_sc1 = 0x44000022;
2095 	pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
2096 	pvinfo->hcall[1] = cpu_to_be32(inst_nop);
2097 	pvinfo->hcall[2] = cpu_to_be32(inst_nop);
2098 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2099 #else
2100 	u32 inst_lis = 0x3c000000;
2101 	u32 inst_ori = 0x60000000;
2102 	u32 inst_sc = 0x44000002;
2103 	u32 inst_imm_mask = 0xffff;
2104 
2105 	/*
2106 	 * The hypercall to get into KVM from within guest context is as
2107 	 * follows:
2108 	 *
2109 	 *    lis r0, r0, KVM_SC_MAGIC_R0@h
2110 	 *    ori r0, KVM_SC_MAGIC_R0@l
2111 	 *    sc
2112 	 *    nop
2113 	 */
2114 	pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
2115 	pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
2116 	pvinfo->hcall[2] = cpu_to_be32(inst_sc);
2117 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2118 #endif
2119 
2120 	pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
2121 
2122 	return 0;
2123 }
2124 
2125 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
2126 			  bool line_status)
2127 {
2128 	if (!irqchip_in_kernel(kvm))
2129 		return -ENXIO;
2130 
2131 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2132 					irq_event->irq, irq_event->level,
2133 					line_status);
2134 	return 0;
2135 }
2136 
2137 
2138 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
2139 			    struct kvm_enable_cap *cap)
2140 {
2141 	int r;
2142 
2143 	if (cap->flags)
2144 		return -EINVAL;
2145 
2146 	switch (cap->cap) {
2147 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
2148 	case KVM_CAP_PPC_ENABLE_HCALL: {
2149 		unsigned long hcall = cap->args[0];
2150 
2151 		r = -EINVAL;
2152 		if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
2153 		    cap->args[1] > 1)
2154 			break;
2155 		if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
2156 			break;
2157 		if (cap->args[1])
2158 			set_bit(hcall / 4, kvm->arch.enabled_hcalls);
2159 		else
2160 			clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
2161 		r = 0;
2162 		break;
2163 	}
2164 	case KVM_CAP_PPC_SMT: {
2165 		unsigned long mode = cap->args[0];
2166 		unsigned long flags = cap->args[1];
2167 
2168 		r = -EINVAL;
2169 		if (kvm->arch.kvm_ops->set_smt_mode)
2170 			r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
2171 		break;
2172 	}
2173 
2174 	case KVM_CAP_PPC_NESTED_HV:
2175 		r = -EINVAL;
2176 		if (!is_kvmppc_hv_enabled(kvm) ||
2177 		    !kvm->arch.kvm_ops->enable_nested)
2178 			break;
2179 		r = kvm->arch.kvm_ops->enable_nested(kvm);
2180 		break;
2181 #endif
2182 #if defined(CONFIG_KVM_BOOK3S_HV_POSSIBLE)
2183 	case KVM_CAP_PPC_SECURE_GUEST:
2184 		r = -EINVAL;
2185 		if (!is_kvmppc_hv_enabled(kvm) || !kvm->arch.kvm_ops->enable_svm)
2186 			break;
2187 		r = kvm->arch.kvm_ops->enable_svm(kvm);
2188 		break;
2189 	case KVM_CAP_PPC_DAWR1:
2190 		r = -EINVAL;
2191 		if (!is_kvmppc_hv_enabled(kvm) || !kvm->arch.kvm_ops->enable_dawr1)
2192 			break;
2193 		r = kvm->arch.kvm_ops->enable_dawr1(kvm);
2194 		break;
2195 #endif
2196 	default:
2197 		r = -EINVAL;
2198 		break;
2199 	}
2200 
2201 	return r;
2202 }
2203 
2204 #ifdef CONFIG_PPC_BOOK3S_64
2205 /*
2206  * These functions check whether the underlying hardware is safe
2207  * against attacks based on observing the effects of speculatively
2208  * executed instructions, and whether it supplies instructions for
2209  * use in workarounds.  The information comes from firmware, either
2210  * via the device tree on powernv platforms or from an hcall on
2211  * pseries platforms.
2212  */
2213 #ifdef CONFIG_PPC_PSERIES
2214 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2215 {
2216 	struct h_cpu_char_result c;
2217 	unsigned long rc;
2218 
2219 	if (!machine_is(pseries))
2220 		return -ENOTTY;
2221 
2222 	rc = plpar_get_cpu_characteristics(&c);
2223 	if (rc == H_SUCCESS) {
2224 		cp->character = c.character;
2225 		cp->behaviour = c.behaviour;
2226 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2227 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2228 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2229 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2230 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2231 			KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
2232 			KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
2233 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2234 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2235 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2236 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2237 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2238 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2239 	}
2240 	return 0;
2241 }
2242 #else
2243 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2244 {
2245 	return -ENOTTY;
2246 }
2247 #endif
2248 
2249 static inline bool have_fw_feat(struct device_node *fw_features,
2250 				const char *state, const char *name)
2251 {
2252 	struct device_node *np;
2253 	bool r = false;
2254 
2255 	np = of_get_child_by_name(fw_features, name);
2256 	if (np) {
2257 		r = of_property_read_bool(np, state);
2258 		of_node_put(np);
2259 	}
2260 	return r;
2261 }
2262 
2263 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2264 {
2265 	struct device_node *np, *fw_features;
2266 	int r;
2267 
2268 	memset(cp, 0, sizeof(*cp));
2269 	r = pseries_get_cpu_char(cp);
2270 	if (r != -ENOTTY)
2271 		return r;
2272 
2273 	np = of_find_node_by_name(NULL, "ibm,opal");
2274 	if (np) {
2275 		fw_features = of_get_child_by_name(np, "fw-features");
2276 		of_node_put(np);
2277 		if (!fw_features)
2278 			return 0;
2279 		if (have_fw_feat(fw_features, "enabled",
2280 				 "inst-spec-barrier-ori31,31,0"))
2281 			cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
2282 		if (have_fw_feat(fw_features, "enabled",
2283 				 "fw-bcctrl-serialized"))
2284 			cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
2285 		if (have_fw_feat(fw_features, "enabled",
2286 				 "inst-l1d-flush-ori30,30,0"))
2287 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
2288 		if (have_fw_feat(fw_features, "enabled",
2289 				 "inst-l1d-flush-trig2"))
2290 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
2291 		if (have_fw_feat(fw_features, "enabled",
2292 				 "fw-l1d-thread-split"))
2293 			cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
2294 		if (have_fw_feat(fw_features, "enabled",
2295 				 "fw-count-cache-disabled"))
2296 			cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2297 		if (have_fw_feat(fw_features, "enabled",
2298 				 "fw-count-cache-flush-bcctr2,0,0"))
2299 			cp->character |= KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2300 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2301 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2302 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2303 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2304 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2305 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2306 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2307 
2308 		if (have_fw_feat(fw_features, "enabled",
2309 				 "speculation-policy-favor-security"))
2310 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
2311 		if (!have_fw_feat(fw_features, "disabled",
2312 				  "needs-l1d-flush-msr-pr-0-to-1"))
2313 			cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
2314 		if (!have_fw_feat(fw_features, "disabled",
2315 				  "needs-spec-barrier-for-bound-checks"))
2316 			cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2317 		if (have_fw_feat(fw_features, "enabled",
2318 				 "needs-count-cache-flush-on-context-switch"))
2319 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2320 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2321 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2322 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2323 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2324 
2325 		of_node_put(fw_features);
2326 	}
2327 
2328 	return 0;
2329 }
2330 #endif
2331 
2332 long kvm_arch_vm_ioctl(struct file *filp,
2333                        unsigned int ioctl, unsigned long arg)
2334 {
2335 	struct kvm *kvm __maybe_unused = filp->private_data;
2336 	void __user *argp = (void __user *)arg;
2337 	long r;
2338 
2339 	switch (ioctl) {
2340 	case KVM_PPC_GET_PVINFO: {
2341 		struct kvm_ppc_pvinfo pvinfo;
2342 		memset(&pvinfo, 0, sizeof(pvinfo));
2343 		r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
2344 		if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
2345 			r = -EFAULT;
2346 			goto out;
2347 		}
2348 
2349 		break;
2350 	}
2351 #ifdef CONFIG_SPAPR_TCE_IOMMU
2352 	case KVM_CREATE_SPAPR_TCE_64: {
2353 		struct kvm_create_spapr_tce_64 create_tce_64;
2354 
2355 		r = -EFAULT;
2356 		if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
2357 			goto out;
2358 		if (create_tce_64.flags) {
2359 			r = -EINVAL;
2360 			goto out;
2361 		}
2362 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2363 		goto out;
2364 	}
2365 	case KVM_CREATE_SPAPR_TCE: {
2366 		struct kvm_create_spapr_tce create_tce;
2367 		struct kvm_create_spapr_tce_64 create_tce_64;
2368 
2369 		r = -EFAULT;
2370 		if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
2371 			goto out;
2372 
2373 		create_tce_64.liobn = create_tce.liobn;
2374 		create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
2375 		create_tce_64.offset = 0;
2376 		create_tce_64.size = create_tce.window_size >>
2377 				IOMMU_PAGE_SHIFT_4K;
2378 		create_tce_64.flags = 0;
2379 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2380 		goto out;
2381 	}
2382 #endif
2383 #ifdef CONFIG_PPC_BOOK3S_64
2384 	case KVM_PPC_GET_SMMU_INFO: {
2385 		struct kvm_ppc_smmu_info info;
2386 		struct kvm *kvm = filp->private_data;
2387 
2388 		memset(&info, 0, sizeof(info));
2389 		r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
2390 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2391 			r = -EFAULT;
2392 		break;
2393 	}
2394 	case KVM_PPC_RTAS_DEFINE_TOKEN: {
2395 		struct kvm *kvm = filp->private_data;
2396 
2397 		r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
2398 		break;
2399 	}
2400 	case KVM_PPC_CONFIGURE_V3_MMU: {
2401 		struct kvm *kvm = filp->private_data;
2402 		struct kvm_ppc_mmuv3_cfg cfg;
2403 
2404 		r = -EINVAL;
2405 		if (!kvm->arch.kvm_ops->configure_mmu)
2406 			goto out;
2407 		r = -EFAULT;
2408 		if (copy_from_user(&cfg, argp, sizeof(cfg)))
2409 			goto out;
2410 		r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
2411 		break;
2412 	}
2413 	case KVM_PPC_GET_RMMU_INFO: {
2414 		struct kvm *kvm = filp->private_data;
2415 		struct kvm_ppc_rmmu_info info;
2416 
2417 		r = -EINVAL;
2418 		if (!kvm->arch.kvm_ops->get_rmmu_info)
2419 			goto out;
2420 		r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
2421 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2422 			r = -EFAULT;
2423 		break;
2424 	}
2425 	case KVM_PPC_GET_CPU_CHAR: {
2426 		struct kvm_ppc_cpu_char cpuchar;
2427 
2428 		r = kvmppc_get_cpu_char(&cpuchar);
2429 		if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2430 			r = -EFAULT;
2431 		break;
2432 	}
2433 	case KVM_PPC_SVM_OFF: {
2434 		struct kvm *kvm = filp->private_data;
2435 
2436 		r = 0;
2437 		if (!kvm->arch.kvm_ops->svm_off)
2438 			goto out;
2439 
2440 		r = kvm->arch.kvm_ops->svm_off(kvm);
2441 		break;
2442 	}
2443 	default: {
2444 		struct kvm *kvm = filp->private_data;
2445 		r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
2446 	}
2447 #else /* CONFIG_PPC_BOOK3S_64 */
2448 	default:
2449 		r = -ENOTTY;
2450 #endif
2451 	}
2452 out:
2453 	return r;
2454 }
2455 
2456 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
2457 static unsigned long nr_lpids;
2458 
2459 long kvmppc_alloc_lpid(void)
2460 {
2461 	long lpid;
2462 
2463 	do {
2464 		lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
2465 		if (lpid >= nr_lpids) {
2466 			pr_err("%s: No LPIDs free\n", __func__);
2467 			return -ENOMEM;
2468 		}
2469 	} while (test_and_set_bit(lpid, lpid_inuse));
2470 
2471 	return lpid;
2472 }
2473 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2474 
2475 void kvmppc_claim_lpid(long lpid)
2476 {
2477 	set_bit(lpid, lpid_inuse);
2478 }
2479 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
2480 
2481 void kvmppc_free_lpid(long lpid)
2482 {
2483 	clear_bit(lpid, lpid_inuse);
2484 }
2485 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2486 
2487 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2488 {
2489 	nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
2490 	memset(lpid_inuse, 0, sizeof(lpid_inuse));
2491 }
2492 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2493 
2494 int kvm_arch_init(void *opaque)
2495 {
2496 	return 0;
2497 }
2498 
2499 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);
2500