xref: /openbmc/linux/arch/powerpc/kvm/powerpc.c (revision 16c8d76a)
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 		kvm_debug_ratelimited("Guest access to device memory using unsupported instruction (opcode: %#08x)\n",
311 				      last_inst);
312 
313 		/*
314 		 * Injecting a Data Storage here is a bit more
315 		 * accurate since the instruction that caused the
316 		 * access could still be a valid one.
317 		 */
318 		if (!IS_ENABLED(CONFIG_BOOKE)) {
319 			ulong dsisr = DSISR_BADACCESS;
320 
321 			if (vcpu->mmio_is_write)
322 				dsisr |= DSISR_ISSTORE;
323 
324 			kvmppc_core_queue_data_storage(vcpu, vcpu->arch.vaddr_accessed, dsisr);
325 		} else {
326 			/*
327 			 * BookE does not send a SIGBUS on a bad
328 			 * fault, so use a Program interrupt instead
329 			 * to avoid a fault loop.
330 			 */
331 			kvmppc_core_queue_program(vcpu, 0);
332 		}
333 
334 		r = RESUME_GUEST;
335 		break;
336 	}
337 	default:
338 		WARN_ON(1);
339 		r = RESUME_GUEST;
340 	}
341 
342 	return r;
343 }
344 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
345 
346 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
347 	      bool data)
348 {
349 	ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
350 	struct kvmppc_pte pte;
351 	int r = -EINVAL;
352 
353 	vcpu->stat.st++;
354 
355 	if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->store_to_eaddr)
356 		r = vcpu->kvm->arch.kvm_ops->store_to_eaddr(vcpu, eaddr, ptr,
357 							    size);
358 
359 	if ((!r) || (r == -EAGAIN))
360 		return r;
361 
362 	r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
363 			 XLATE_WRITE, &pte);
364 	if (r < 0)
365 		return r;
366 
367 	*eaddr = pte.raddr;
368 
369 	if (!pte.may_write)
370 		return -EPERM;
371 
372 	/* Magic page override */
373 	if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
374 	    ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
375 	    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
376 		void *magic = vcpu->arch.shared;
377 		magic += pte.eaddr & 0xfff;
378 		memcpy(magic, ptr, size);
379 		return EMULATE_DONE;
380 	}
381 
382 	if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
383 		return EMULATE_DO_MMIO;
384 
385 	return EMULATE_DONE;
386 }
387 EXPORT_SYMBOL_GPL(kvmppc_st);
388 
389 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
390 		      bool data)
391 {
392 	ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
393 	struct kvmppc_pte pte;
394 	int rc = -EINVAL;
395 
396 	vcpu->stat.ld++;
397 
398 	if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->load_from_eaddr)
399 		rc = vcpu->kvm->arch.kvm_ops->load_from_eaddr(vcpu, eaddr, ptr,
400 							      size);
401 
402 	if ((!rc) || (rc == -EAGAIN))
403 		return rc;
404 
405 	rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
406 			  XLATE_READ, &pte);
407 	if (rc)
408 		return rc;
409 
410 	*eaddr = pte.raddr;
411 
412 	if (!pte.may_read)
413 		return -EPERM;
414 
415 	if (!data && !pte.may_execute)
416 		return -ENOEXEC;
417 
418 	/* Magic page override */
419 	if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
420 	    ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
421 	    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
422 		void *magic = vcpu->arch.shared;
423 		magic += pte.eaddr & 0xfff;
424 		memcpy(ptr, magic, size);
425 		return EMULATE_DONE;
426 	}
427 
428 	kvm_vcpu_srcu_read_lock(vcpu);
429 	rc = kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size);
430 	kvm_vcpu_srcu_read_unlock(vcpu);
431 	if (rc)
432 		return EMULATE_DO_MMIO;
433 
434 	return EMULATE_DONE;
435 }
436 EXPORT_SYMBOL_GPL(kvmppc_ld);
437 
438 int kvm_arch_hardware_enable(void)
439 {
440 	return 0;
441 }
442 
443 int kvm_arch_hardware_setup(void *opaque)
444 {
445 	return 0;
446 }
447 
448 int kvm_arch_check_processor_compat(void *opaque)
449 {
450 	return kvmppc_core_check_processor_compat();
451 }
452 
453 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
454 {
455 	struct kvmppc_ops *kvm_ops = NULL;
456 	int r;
457 
458 	/*
459 	 * if we have both HV and PR enabled, default is HV
460 	 */
461 	if (type == 0) {
462 		if (kvmppc_hv_ops)
463 			kvm_ops = kvmppc_hv_ops;
464 		else
465 			kvm_ops = kvmppc_pr_ops;
466 		if (!kvm_ops)
467 			goto err_out;
468 	} else	if (type == KVM_VM_PPC_HV) {
469 		if (!kvmppc_hv_ops)
470 			goto err_out;
471 		kvm_ops = kvmppc_hv_ops;
472 	} else if (type == KVM_VM_PPC_PR) {
473 		if (!kvmppc_pr_ops)
474 			goto err_out;
475 		kvm_ops = kvmppc_pr_ops;
476 	} else
477 		goto err_out;
478 
479 	if (!try_module_get(kvm_ops->owner))
480 		return -ENOENT;
481 
482 	kvm->arch.kvm_ops = kvm_ops;
483 	r = kvmppc_core_init_vm(kvm);
484 	if (r)
485 		module_put(kvm_ops->owner);
486 	return r;
487 err_out:
488 	return -EINVAL;
489 }
490 
491 void kvm_arch_destroy_vm(struct kvm *kvm)
492 {
493 #ifdef CONFIG_KVM_XICS
494 	/*
495 	 * We call kick_all_cpus_sync() to ensure that all
496 	 * CPUs have executed any pending IPIs before we
497 	 * continue and free VCPUs structures below.
498 	 */
499 	if (is_kvmppc_hv_enabled(kvm))
500 		kick_all_cpus_sync();
501 #endif
502 
503 	kvm_destroy_vcpus(kvm);
504 
505 	mutex_lock(&kvm->lock);
506 
507 	kvmppc_core_destroy_vm(kvm);
508 
509 	mutex_unlock(&kvm->lock);
510 
511 	/* drop the module reference */
512 	module_put(kvm->arch.kvm_ops->owner);
513 }
514 
515 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
516 {
517 	int r;
518 	/* Assume we're using HV mode when the HV module is loaded */
519 	int hv_enabled = kvmppc_hv_ops ? 1 : 0;
520 
521 	if (kvm) {
522 		/*
523 		 * Hooray - we know which VM type we're running on. Depend on
524 		 * that rather than the guess above.
525 		 */
526 		hv_enabled = is_kvmppc_hv_enabled(kvm);
527 	}
528 
529 	switch (ext) {
530 #ifdef CONFIG_BOOKE
531 	case KVM_CAP_PPC_BOOKE_SREGS:
532 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
533 	case KVM_CAP_PPC_EPR:
534 #else
535 	case KVM_CAP_PPC_SEGSTATE:
536 	case KVM_CAP_PPC_HIOR:
537 	case KVM_CAP_PPC_PAPR:
538 #endif
539 	case KVM_CAP_PPC_UNSET_IRQ:
540 	case KVM_CAP_PPC_IRQ_LEVEL:
541 	case KVM_CAP_ENABLE_CAP:
542 	case KVM_CAP_ONE_REG:
543 	case KVM_CAP_IOEVENTFD:
544 	case KVM_CAP_DEVICE_CTRL:
545 	case KVM_CAP_IMMEDIATE_EXIT:
546 	case KVM_CAP_SET_GUEST_DEBUG:
547 		r = 1;
548 		break;
549 	case KVM_CAP_PPC_GUEST_DEBUG_SSTEP:
550 	case KVM_CAP_PPC_PAIRED_SINGLES:
551 	case KVM_CAP_PPC_OSI:
552 	case KVM_CAP_PPC_GET_PVINFO:
553 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
554 	case KVM_CAP_SW_TLB:
555 #endif
556 		/* We support this only for PR */
557 		r = !hv_enabled;
558 		break;
559 #ifdef CONFIG_KVM_MPIC
560 	case KVM_CAP_IRQ_MPIC:
561 		r = 1;
562 		break;
563 #endif
564 
565 #ifdef CONFIG_PPC_BOOK3S_64
566 	case KVM_CAP_SPAPR_TCE:
567 	case KVM_CAP_SPAPR_TCE_64:
568 		r = 1;
569 		break;
570 	case KVM_CAP_SPAPR_TCE_VFIO:
571 		r = !!cpu_has_feature(CPU_FTR_HVMODE);
572 		break;
573 	case KVM_CAP_PPC_RTAS:
574 	case KVM_CAP_PPC_FIXUP_HCALL:
575 	case KVM_CAP_PPC_ENABLE_HCALL:
576 #ifdef CONFIG_KVM_XICS
577 	case KVM_CAP_IRQ_XICS:
578 #endif
579 	case KVM_CAP_PPC_GET_CPU_CHAR:
580 		r = 1;
581 		break;
582 #ifdef CONFIG_KVM_XIVE
583 	case KVM_CAP_PPC_IRQ_XIVE:
584 		/*
585 		 * We need XIVE to be enabled on the platform (implies
586 		 * a POWER9 processor) and the PowerNV platform, as
587 		 * nested is not yet supported.
588 		 */
589 		r = xive_enabled() && !!cpu_has_feature(CPU_FTR_HVMODE) &&
590 			kvmppc_xive_native_supported();
591 		break;
592 #endif
593 
594 	case KVM_CAP_PPC_ALLOC_HTAB:
595 		r = hv_enabled;
596 		break;
597 #endif /* CONFIG_PPC_BOOK3S_64 */
598 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
599 	case KVM_CAP_PPC_SMT:
600 		r = 0;
601 		if (kvm) {
602 			if (kvm->arch.emul_smt_mode > 1)
603 				r = kvm->arch.emul_smt_mode;
604 			else
605 				r = kvm->arch.smt_mode;
606 		} else if (hv_enabled) {
607 			if (cpu_has_feature(CPU_FTR_ARCH_300))
608 				r = 1;
609 			else
610 				r = threads_per_subcore;
611 		}
612 		break;
613 	case KVM_CAP_PPC_SMT_POSSIBLE:
614 		r = 1;
615 		if (hv_enabled) {
616 			if (!cpu_has_feature(CPU_FTR_ARCH_300))
617 				r = ((threads_per_subcore << 1) - 1);
618 			else
619 				/* P9 can emulate dbells, so allow any mode */
620 				r = 8 | 4 | 2 | 1;
621 		}
622 		break;
623 	case KVM_CAP_PPC_RMA:
624 		r = 0;
625 		break;
626 	case KVM_CAP_PPC_HWRNG:
627 		r = kvmppc_hwrng_present();
628 		break;
629 	case KVM_CAP_PPC_MMU_RADIX:
630 		r = !!(hv_enabled && radix_enabled());
631 		break;
632 	case KVM_CAP_PPC_MMU_HASH_V3:
633 		r = !!(hv_enabled && kvmppc_hv_ops->hash_v3_possible &&
634 		       kvmppc_hv_ops->hash_v3_possible());
635 		break;
636 	case KVM_CAP_PPC_NESTED_HV:
637 		r = !!(hv_enabled && kvmppc_hv_ops->enable_nested &&
638 		       !kvmppc_hv_ops->enable_nested(NULL));
639 		break;
640 #endif
641 	case KVM_CAP_SYNC_MMU:
642 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
643 		r = hv_enabled;
644 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
645 		r = 1;
646 #else
647 		r = 0;
648 #endif
649 		break;
650 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
651 	case KVM_CAP_PPC_HTAB_FD:
652 		r = hv_enabled;
653 		break;
654 #endif
655 	case KVM_CAP_NR_VCPUS:
656 		/*
657 		 * Recommending a number of CPUs is somewhat arbitrary; we
658 		 * return the number of present CPUs for -HV (since a host
659 		 * will have secondary threads "offline"), and for other KVM
660 		 * implementations just count online CPUs.
661 		 */
662 		if (hv_enabled)
663 			r = min_t(unsigned int, num_present_cpus(), KVM_MAX_VCPUS);
664 		else
665 			r = min_t(unsigned int, num_online_cpus(), KVM_MAX_VCPUS);
666 		break;
667 	case KVM_CAP_MAX_VCPUS:
668 		r = KVM_MAX_VCPUS;
669 		break;
670 	case KVM_CAP_MAX_VCPU_ID:
671 		r = KVM_MAX_VCPU_IDS;
672 		break;
673 #ifdef CONFIG_PPC_BOOK3S_64
674 	case KVM_CAP_PPC_GET_SMMU_INFO:
675 		r = 1;
676 		break;
677 	case KVM_CAP_SPAPR_MULTITCE:
678 		r = 1;
679 		break;
680 	case KVM_CAP_SPAPR_RESIZE_HPT:
681 		r = !!hv_enabled;
682 		break;
683 #endif
684 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
685 	case KVM_CAP_PPC_FWNMI:
686 		r = hv_enabled;
687 		break;
688 #endif
689 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
690 	case KVM_CAP_PPC_HTM:
691 		r = !!(cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM) ||
692 		     (hv_enabled && cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST));
693 		break;
694 #endif
695 #if defined(CONFIG_KVM_BOOK3S_HV_POSSIBLE)
696 	case KVM_CAP_PPC_SECURE_GUEST:
697 		r = hv_enabled && kvmppc_hv_ops->enable_svm &&
698 			!kvmppc_hv_ops->enable_svm(NULL);
699 		break;
700 	case KVM_CAP_PPC_DAWR1:
701 		r = !!(hv_enabled && kvmppc_hv_ops->enable_dawr1 &&
702 		       !kvmppc_hv_ops->enable_dawr1(NULL));
703 		break;
704 	case KVM_CAP_PPC_RPT_INVALIDATE:
705 		r = 1;
706 		break;
707 #endif
708 	case KVM_CAP_PPC_AIL_MODE_3:
709 		r = 0;
710 		/*
711 		 * KVM PR, POWER7, and some POWER9s don't support AIL=3 mode.
712 		 * The POWER9s can support it if the guest runs in hash mode,
713 		 * but QEMU doesn't necessarily query the capability in time.
714 		 */
715 		if (hv_enabled) {
716 			if (kvmhv_on_pseries()) {
717 				if (pseries_reloc_on_exception())
718 					r = 1;
719 			} else if (cpu_has_feature(CPU_FTR_ARCH_207S) &&
720 				  !cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG)) {
721 				r = 1;
722 			}
723 		}
724 		break;
725 	default:
726 		r = 0;
727 		break;
728 	}
729 	return r;
730 
731 }
732 
733 long kvm_arch_dev_ioctl(struct file *filp,
734                         unsigned int ioctl, unsigned long arg)
735 {
736 	return -EINVAL;
737 }
738 
739 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
740 {
741 	kvmppc_core_free_memslot(kvm, slot);
742 }
743 
744 int kvm_arch_prepare_memory_region(struct kvm *kvm,
745 				   const struct kvm_memory_slot *old,
746 				   struct kvm_memory_slot *new,
747 				   enum kvm_mr_change change)
748 {
749 	return kvmppc_core_prepare_memory_region(kvm, old, new, change);
750 }
751 
752 void kvm_arch_commit_memory_region(struct kvm *kvm,
753 				   struct kvm_memory_slot *old,
754 				   const struct kvm_memory_slot *new,
755 				   enum kvm_mr_change change)
756 {
757 	kvmppc_core_commit_memory_region(kvm, old, new, change);
758 }
759 
760 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
761 				   struct kvm_memory_slot *slot)
762 {
763 	kvmppc_core_flush_memslot(kvm, slot);
764 }
765 
766 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
767 {
768 	return 0;
769 }
770 
771 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
772 {
773 	struct kvm_vcpu *vcpu;
774 
775 	vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
776 	kvmppc_decrementer_func(vcpu);
777 
778 	return HRTIMER_NORESTART;
779 }
780 
781 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
782 {
783 	int err;
784 
785 	hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
786 	vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
787 	vcpu->arch.dec_expires = get_tb();
788 
789 #ifdef CONFIG_KVM_EXIT_TIMING
790 	mutex_init(&vcpu->arch.exit_timing_lock);
791 #endif
792 	err = kvmppc_subarch_vcpu_init(vcpu);
793 	if (err)
794 		return err;
795 
796 	err = kvmppc_core_vcpu_create(vcpu);
797 	if (err)
798 		goto out_vcpu_uninit;
799 
800 	rcuwait_init(&vcpu->arch.wait);
801 	vcpu->arch.waitp = &vcpu->arch.wait;
802 	return 0;
803 
804 out_vcpu_uninit:
805 	kvmppc_subarch_vcpu_uninit(vcpu);
806 	return err;
807 }
808 
809 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
810 {
811 }
812 
813 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
814 {
815 	/* Make sure we're not using the vcpu anymore */
816 	hrtimer_cancel(&vcpu->arch.dec_timer);
817 
818 	switch (vcpu->arch.irq_type) {
819 	case KVMPPC_IRQ_MPIC:
820 		kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
821 		break;
822 	case KVMPPC_IRQ_XICS:
823 		if (xics_on_xive())
824 			kvmppc_xive_cleanup_vcpu(vcpu);
825 		else
826 			kvmppc_xics_free_icp(vcpu);
827 		break;
828 	case KVMPPC_IRQ_XIVE:
829 		kvmppc_xive_native_cleanup_vcpu(vcpu);
830 		break;
831 	}
832 
833 	kvmppc_core_vcpu_free(vcpu);
834 
835 	kvmppc_subarch_vcpu_uninit(vcpu);
836 }
837 
838 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
839 {
840 	return kvmppc_core_pending_dec(vcpu);
841 }
842 
843 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
844 {
845 #ifdef CONFIG_BOOKE
846 	/*
847 	 * vrsave (formerly usprg0) isn't used by Linux, but may
848 	 * be used by the guest.
849 	 *
850 	 * On non-booke this is associated with Altivec and
851 	 * is handled by code in book3s.c.
852 	 */
853 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
854 #endif
855 	kvmppc_core_vcpu_load(vcpu, cpu);
856 }
857 
858 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
859 {
860 	kvmppc_core_vcpu_put(vcpu);
861 #ifdef CONFIG_BOOKE
862 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
863 #endif
864 }
865 
866 /*
867  * irq_bypass_add_producer and irq_bypass_del_producer are only
868  * useful if the architecture supports PCI passthrough.
869  * irq_bypass_stop and irq_bypass_start are not needed and so
870  * kvm_ops are not defined for them.
871  */
872 bool kvm_arch_has_irq_bypass(void)
873 {
874 	return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
875 		(kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
876 }
877 
878 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
879 				     struct irq_bypass_producer *prod)
880 {
881 	struct kvm_kernel_irqfd *irqfd =
882 		container_of(cons, struct kvm_kernel_irqfd, consumer);
883 	struct kvm *kvm = irqfd->kvm;
884 
885 	if (kvm->arch.kvm_ops->irq_bypass_add_producer)
886 		return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
887 
888 	return 0;
889 }
890 
891 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
892 				      struct irq_bypass_producer *prod)
893 {
894 	struct kvm_kernel_irqfd *irqfd =
895 		container_of(cons, struct kvm_kernel_irqfd, consumer);
896 	struct kvm *kvm = irqfd->kvm;
897 
898 	if (kvm->arch.kvm_ops->irq_bypass_del_producer)
899 		kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
900 }
901 
902 #ifdef CONFIG_VSX
903 static inline int kvmppc_get_vsr_dword_offset(int index)
904 {
905 	int offset;
906 
907 	if ((index != 0) && (index != 1))
908 		return -1;
909 
910 #ifdef __BIG_ENDIAN
911 	offset =  index;
912 #else
913 	offset = 1 - index;
914 #endif
915 
916 	return offset;
917 }
918 
919 static inline int kvmppc_get_vsr_word_offset(int index)
920 {
921 	int offset;
922 
923 	if ((index > 3) || (index < 0))
924 		return -1;
925 
926 #ifdef __BIG_ENDIAN
927 	offset = index;
928 #else
929 	offset = 3 - index;
930 #endif
931 	return offset;
932 }
933 
934 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
935 	u64 gpr)
936 {
937 	union kvmppc_one_reg val;
938 	int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
939 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
940 
941 	if (offset == -1)
942 		return;
943 
944 	if (index >= 32) {
945 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
946 		val.vsxval[offset] = gpr;
947 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
948 	} else {
949 		VCPU_VSX_FPR(vcpu, index, offset) = gpr;
950 	}
951 }
952 
953 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
954 	u64 gpr)
955 {
956 	union kvmppc_one_reg val;
957 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
958 
959 	if (index >= 32) {
960 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
961 		val.vsxval[0] = gpr;
962 		val.vsxval[1] = gpr;
963 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
964 	} else {
965 		VCPU_VSX_FPR(vcpu, index, 0) = gpr;
966 		VCPU_VSX_FPR(vcpu, index, 1) = gpr;
967 	}
968 }
969 
970 static inline void kvmppc_set_vsr_word_dump(struct kvm_vcpu *vcpu,
971 	u32 gpr)
972 {
973 	union kvmppc_one_reg val;
974 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
975 
976 	if (index >= 32) {
977 		val.vsx32val[0] = gpr;
978 		val.vsx32val[1] = gpr;
979 		val.vsx32val[2] = gpr;
980 		val.vsx32val[3] = gpr;
981 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
982 	} else {
983 		val.vsx32val[0] = gpr;
984 		val.vsx32val[1] = gpr;
985 		VCPU_VSX_FPR(vcpu, index, 0) = val.vsxval[0];
986 		VCPU_VSX_FPR(vcpu, index, 1) = val.vsxval[0];
987 	}
988 }
989 
990 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
991 	u32 gpr32)
992 {
993 	union kvmppc_one_reg val;
994 	int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
995 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
996 	int dword_offset, word_offset;
997 
998 	if (offset == -1)
999 		return;
1000 
1001 	if (index >= 32) {
1002 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
1003 		val.vsx32val[offset] = gpr32;
1004 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
1005 	} else {
1006 		dword_offset = offset / 2;
1007 		word_offset = offset % 2;
1008 		val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
1009 		val.vsx32val[word_offset] = gpr32;
1010 		VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
1011 	}
1012 }
1013 #endif /* CONFIG_VSX */
1014 
1015 #ifdef CONFIG_ALTIVEC
1016 static inline int kvmppc_get_vmx_offset_generic(struct kvm_vcpu *vcpu,
1017 		int index, int element_size)
1018 {
1019 	int offset;
1020 	int elts = sizeof(vector128)/element_size;
1021 
1022 	if ((index < 0) || (index >= elts))
1023 		return -1;
1024 
1025 	if (kvmppc_need_byteswap(vcpu))
1026 		offset = elts - index - 1;
1027 	else
1028 		offset = index;
1029 
1030 	return offset;
1031 }
1032 
1033 static inline int kvmppc_get_vmx_dword_offset(struct kvm_vcpu *vcpu,
1034 		int index)
1035 {
1036 	return kvmppc_get_vmx_offset_generic(vcpu, index, 8);
1037 }
1038 
1039 static inline int kvmppc_get_vmx_word_offset(struct kvm_vcpu *vcpu,
1040 		int index)
1041 {
1042 	return kvmppc_get_vmx_offset_generic(vcpu, index, 4);
1043 }
1044 
1045 static inline int kvmppc_get_vmx_hword_offset(struct kvm_vcpu *vcpu,
1046 		int index)
1047 {
1048 	return kvmppc_get_vmx_offset_generic(vcpu, index, 2);
1049 }
1050 
1051 static inline int kvmppc_get_vmx_byte_offset(struct kvm_vcpu *vcpu,
1052 		int index)
1053 {
1054 	return kvmppc_get_vmx_offset_generic(vcpu, index, 1);
1055 }
1056 
1057 
1058 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu,
1059 	u64 gpr)
1060 {
1061 	union kvmppc_one_reg val;
1062 	int offset = kvmppc_get_vmx_dword_offset(vcpu,
1063 			vcpu->arch.mmio_vmx_offset);
1064 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1065 
1066 	if (offset == -1)
1067 		return;
1068 
1069 	val.vval = VCPU_VSX_VR(vcpu, index);
1070 	val.vsxval[offset] = gpr;
1071 	VCPU_VSX_VR(vcpu, index) = val.vval;
1072 }
1073 
1074 static inline void kvmppc_set_vmx_word(struct kvm_vcpu *vcpu,
1075 	u32 gpr32)
1076 {
1077 	union kvmppc_one_reg val;
1078 	int offset = kvmppc_get_vmx_word_offset(vcpu,
1079 			vcpu->arch.mmio_vmx_offset);
1080 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1081 
1082 	if (offset == -1)
1083 		return;
1084 
1085 	val.vval = VCPU_VSX_VR(vcpu, index);
1086 	val.vsx32val[offset] = gpr32;
1087 	VCPU_VSX_VR(vcpu, index) = val.vval;
1088 }
1089 
1090 static inline void kvmppc_set_vmx_hword(struct kvm_vcpu *vcpu,
1091 	u16 gpr16)
1092 {
1093 	union kvmppc_one_reg val;
1094 	int offset = kvmppc_get_vmx_hword_offset(vcpu,
1095 			vcpu->arch.mmio_vmx_offset);
1096 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1097 
1098 	if (offset == -1)
1099 		return;
1100 
1101 	val.vval = VCPU_VSX_VR(vcpu, index);
1102 	val.vsx16val[offset] = gpr16;
1103 	VCPU_VSX_VR(vcpu, index) = val.vval;
1104 }
1105 
1106 static inline void kvmppc_set_vmx_byte(struct kvm_vcpu *vcpu,
1107 	u8 gpr8)
1108 {
1109 	union kvmppc_one_reg val;
1110 	int offset = kvmppc_get_vmx_byte_offset(vcpu,
1111 			vcpu->arch.mmio_vmx_offset);
1112 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1113 
1114 	if (offset == -1)
1115 		return;
1116 
1117 	val.vval = VCPU_VSX_VR(vcpu, index);
1118 	val.vsx8val[offset] = gpr8;
1119 	VCPU_VSX_VR(vcpu, index) = val.vval;
1120 }
1121 #endif /* CONFIG_ALTIVEC */
1122 
1123 #ifdef CONFIG_PPC_FPU
1124 static inline u64 sp_to_dp(u32 fprs)
1125 {
1126 	u64 fprd;
1127 
1128 	preempt_disable();
1129 	enable_kernel_fp();
1130 	asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m<>" (fprd) : "m<>" (fprs)
1131 	     : "fr0");
1132 	preempt_enable();
1133 	return fprd;
1134 }
1135 
1136 static inline u32 dp_to_sp(u64 fprd)
1137 {
1138 	u32 fprs;
1139 
1140 	preempt_disable();
1141 	enable_kernel_fp();
1142 	asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m<>" (fprs) : "m<>" (fprd)
1143 	     : "fr0");
1144 	preempt_enable();
1145 	return fprs;
1146 }
1147 
1148 #else
1149 #define sp_to_dp(x)	(x)
1150 #define dp_to_sp(x)	(x)
1151 #endif /* CONFIG_PPC_FPU */
1152 
1153 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu)
1154 {
1155 	struct kvm_run *run = vcpu->run;
1156 	u64 gpr;
1157 
1158 	if (run->mmio.len > sizeof(gpr))
1159 		return;
1160 
1161 	if (!vcpu->arch.mmio_host_swabbed) {
1162 		switch (run->mmio.len) {
1163 		case 8: gpr = *(u64 *)run->mmio.data; break;
1164 		case 4: gpr = *(u32 *)run->mmio.data; break;
1165 		case 2: gpr = *(u16 *)run->mmio.data; break;
1166 		case 1: gpr = *(u8 *)run->mmio.data; break;
1167 		}
1168 	} else {
1169 		switch (run->mmio.len) {
1170 		case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
1171 		case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
1172 		case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
1173 		case 1: gpr = *(u8 *)run->mmio.data; break;
1174 		}
1175 	}
1176 
1177 	/* conversion between single and double precision */
1178 	if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
1179 		gpr = sp_to_dp(gpr);
1180 
1181 	if (vcpu->arch.mmio_sign_extend) {
1182 		switch (run->mmio.len) {
1183 #ifdef CONFIG_PPC64
1184 		case 4:
1185 			gpr = (s64)(s32)gpr;
1186 			break;
1187 #endif
1188 		case 2:
1189 			gpr = (s64)(s16)gpr;
1190 			break;
1191 		case 1:
1192 			gpr = (s64)(s8)gpr;
1193 			break;
1194 		}
1195 	}
1196 
1197 	switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1198 	case KVM_MMIO_REG_GPR:
1199 		kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1200 		break;
1201 	case KVM_MMIO_REG_FPR:
1202 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1203 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_FP);
1204 
1205 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1206 		break;
1207 #ifdef CONFIG_PPC_BOOK3S
1208 	case KVM_MMIO_REG_QPR:
1209 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1210 		break;
1211 	case KVM_MMIO_REG_FQPR:
1212 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1213 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1214 		break;
1215 #endif
1216 #ifdef CONFIG_VSX
1217 	case KVM_MMIO_REG_VSX:
1218 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1219 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VSX);
1220 
1221 		if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_DWORD)
1222 			kvmppc_set_vsr_dword(vcpu, gpr);
1223 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_WORD)
1224 			kvmppc_set_vsr_word(vcpu, gpr);
1225 		else if (vcpu->arch.mmio_copy_type ==
1226 				KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1227 			kvmppc_set_vsr_dword_dump(vcpu, gpr);
1228 		else if (vcpu->arch.mmio_copy_type ==
1229 				KVMPPC_VSX_COPY_WORD_LOAD_DUMP)
1230 			kvmppc_set_vsr_word_dump(vcpu, gpr);
1231 		break;
1232 #endif
1233 #ifdef CONFIG_ALTIVEC
1234 	case KVM_MMIO_REG_VMX:
1235 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1236 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VEC);
1237 
1238 		if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_DWORD)
1239 			kvmppc_set_vmx_dword(vcpu, gpr);
1240 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_WORD)
1241 			kvmppc_set_vmx_word(vcpu, gpr);
1242 		else if (vcpu->arch.mmio_copy_type ==
1243 				KVMPPC_VMX_COPY_HWORD)
1244 			kvmppc_set_vmx_hword(vcpu, gpr);
1245 		else if (vcpu->arch.mmio_copy_type ==
1246 				KVMPPC_VMX_COPY_BYTE)
1247 			kvmppc_set_vmx_byte(vcpu, gpr);
1248 		break;
1249 #endif
1250 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1251 	case KVM_MMIO_REG_NESTED_GPR:
1252 		if (kvmppc_need_byteswap(vcpu))
1253 			gpr = swab64(gpr);
1254 		kvm_vcpu_write_guest(vcpu, vcpu->arch.nested_io_gpr, &gpr,
1255 				     sizeof(gpr));
1256 		break;
1257 #endif
1258 	default:
1259 		BUG();
1260 	}
1261 }
1262 
1263 static int __kvmppc_handle_load(struct kvm_vcpu *vcpu,
1264 				unsigned int rt, unsigned int bytes,
1265 				int is_default_endian, int sign_extend)
1266 {
1267 	struct kvm_run *run = vcpu->run;
1268 	int idx, ret;
1269 	bool host_swabbed;
1270 
1271 	/* Pity C doesn't have a logical XOR operator */
1272 	if (kvmppc_need_byteswap(vcpu)) {
1273 		host_swabbed = is_default_endian;
1274 	} else {
1275 		host_swabbed = !is_default_endian;
1276 	}
1277 
1278 	if (bytes > sizeof(run->mmio.data))
1279 		return EMULATE_FAIL;
1280 
1281 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1282 	run->mmio.len = bytes;
1283 	run->mmio.is_write = 0;
1284 
1285 	vcpu->arch.io_gpr = rt;
1286 	vcpu->arch.mmio_host_swabbed = host_swabbed;
1287 	vcpu->mmio_needed = 1;
1288 	vcpu->mmio_is_write = 0;
1289 	vcpu->arch.mmio_sign_extend = sign_extend;
1290 
1291 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1292 
1293 	ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1294 			      bytes, &run->mmio.data);
1295 
1296 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1297 
1298 	if (!ret) {
1299 		kvmppc_complete_mmio_load(vcpu);
1300 		vcpu->mmio_needed = 0;
1301 		return EMULATE_DONE;
1302 	}
1303 
1304 	return EMULATE_DO_MMIO;
1305 }
1306 
1307 int kvmppc_handle_load(struct kvm_vcpu *vcpu,
1308 		       unsigned int rt, unsigned int bytes,
1309 		       int is_default_endian)
1310 {
1311 	return __kvmppc_handle_load(vcpu, rt, bytes, is_default_endian, 0);
1312 }
1313 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1314 
1315 /* Same as above, but sign extends */
1316 int kvmppc_handle_loads(struct kvm_vcpu *vcpu,
1317 			unsigned int rt, unsigned int bytes,
1318 			int is_default_endian)
1319 {
1320 	return __kvmppc_handle_load(vcpu, rt, bytes, is_default_endian, 1);
1321 }
1322 
1323 #ifdef CONFIG_VSX
1324 int kvmppc_handle_vsx_load(struct kvm_vcpu *vcpu,
1325 			unsigned int rt, unsigned int bytes,
1326 			int is_default_endian, int mmio_sign_extend)
1327 {
1328 	enum emulation_result emulated = EMULATE_DONE;
1329 
1330 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1331 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1332 		return EMULATE_FAIL;
1333 
1334 	while (vcpu->arch.mmio_vsx_copy_nums) {
1335 		emulated = __kvmppc_handle_load(vcpu, rt, bytes,
1336 			is_default_endian, mmio_sign_extend);
1337 
1338 		if (emulated != EMULATE_DONE)
1339 			break;
1340 
1341 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1342 
1343 		vcpu->arch.mmio_vsx_copy_nums--;
1344 		vcpu->arch.mmio_vsx_offset++;
1345 	}
1346 	return emulated;
1347 }
1348 #endif /* CONFIG_VSX */
1349 
1350 int kvmppc_handle_store(struct kvm_vcpu *vcpu,
1351 			u64 val, unsigned int bytes, int is_default_endian)
1352 {
1353 	struct kvm_run *run = vcpu->run;
1354 	void *data = run->mmio.data;
1355 	int idx, ret;
1356 	bool host_swabbed;
1357 
1358 	/* Pity C doesn't have a logical XOR operator */
1359 	if (kvmppc_need_byteswap(vcpu)) {
1360 		host_swabbed = is_default_endian;
1361 	} else {
1362 		host_swabbed = !is_default_endian;
1363 	}
1364 
1365 	if (bytes > sizeof(run->mmio.data))
1366 		return EMULATE_FAIL;
1367 
1368 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1369 	run->mmio.len = bytes;
1370 	run->mmio.is_write = 1;
1371 	vcpu->mmio_needed = 1;
1372 	vcpu->mmio_is_write = 1;
1373 
1374 	if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1375 		val = dp_to_sp(val);
1376 
1377 	/* Store the value at the lowest bytes in 'data'. */
1378 	if (!host_swabbed) {
1379 		switch (bytes) {
1380 		case 8: *(u64 *)data = val; break;
1381 		case 4: *(u32 *)data = val; break;
1382 		case 2: *(u16 *)data = val; break;
1383 		case 1: *(u8  *)data = val; break;
1384 		}
1385 	} else {
1386 		switch (bytes) {
1387 		case 8: *(u64 *)data = swab64(val); break;
1388 		case 4: *(u32 *)data = swab32(val); break;
1389 		case 2: *(u16 *)data = swab16(val); break;
1390 		case 1: *(u8  *)data = val; break;
1391 		}
1392 	}
1393 
1394 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1395 
1396 	ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1397 			       bytes, &run->mmio.data);
1398 
1399 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1400 
1401 	if (!ret) {
1402 		vcpu->mmio_needed = 0;
1403 		return EMULATE_DONE;
1404 	}
1405 
1406 	return EMULATE_DO_MMIO;
1407 }
1408 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1409 
1410 #ifdef CONFIG_VSX
1411 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1412 {
1413 	u32 dword_offset, word_offset;
1414 	union kvmppc_one_reg reg;
1415 	int vsx_offset = 0;
1416 	int copy_type = vcpu->arch.mmio_copy_type;
1417 	int result = 0;
1418 
1419 	switch (copy_type) {
1420 	case KVMPPC_VSX_COPY_DWORD:
1421 		vsx_offset =
1422 			kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1423 
1424 		if (vsx_offset == -1) {
1425 			result = -1;
1426 			break;
1427 		}
1428 
1429 		if (rs < 32) {
1430 			*val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1431 		} else {
1432 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1433 			*val = reg.vsxval[vsx_offset];
1434 		}
1435 		break;
1436 
1437 	case KVMPPC_VSX_COPY_WORD:
1438 		vsx_offset =
1439 			kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1440 
1441 		if (vsx_offset == -1) {
1442 			result = -1;
1443 			break;
1444 		}
1445 
1446 		if (rs < 32) {
1447 			dword_offset = vsx_offset / 2;
1448 			word_offset = vsx_offset % 2;
1449 			reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1450 			*val = reg.vsx32val[word_offset];
1451 		} else {
1452 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1453 			*val = reg.vsx32val[vsx_offset];
1454 		}
1455 		break;
1456 
1457 	default:
1458 		result = -1;
1459 		break;
1460 	}
1461 
1462 	return result;
1463 }
1464 
1465 int kvmppc_handle_vsx_store(struct kvm_vcpu *vcpu,
1466 			int rs, unsigned int bytes, int is_default_endian)
1467 {
1468 	u64 val;
1469 	enum emulation_result emulated = EMULATE_DONE;
1470 
1471 	vcpu->arch.io_gpr = rs;
1472 
1473 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1474 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1475 		return EMULATE_FAIL;
1476 
1477 	while (vcpu->arch.mmio_vsx_copy_nums) {
1478 		if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1479 			return EMULATE_FAIL;
1480 
1481 		emulated = kvmppc_handle_store(vcpu,
1482 			 val, bytes, is_default_endian);
1483 
1484 		if (emulated != EMULATE_DONE)
1485 			break;
1486 
1487 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1488 
1489 		vcpu->arch.mmio_vsx_copy_nums--;
1490 		vcpu->arch.mmio_vsx_offset++;
1491 	}
1492 
1493 	return emulated;
1494 }
1495 
1496 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu)
1497 {
1498 	struct kvm_run *run = vcpu->run;
1499 	enum emulation_result emulated = EMULATE_FAIL;
1500 	int r;
1501 
1502 	vcpu->arch.paddr_accessed += run->mmio.len;
1503 
1504 	if (!vcpu->mmio_is_write) {
1505 		emulated = kvmppc_handle_vsx_load(vcpu, vcpu->arch.io_gpr,
1506 			 run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1507 	} else {
1508 		emulated = kvmppc_handle_vsx_store(vcpu,
1509 			 vcpu->arch.io_gpr, run->mmio.len, 1);
1510 	}
1511 
1512 	switch (emulated) {
1513 	case EMULATE_DO_MMIO:
1514 		run->exit_reason = KVM_EXIT_MMIO;
1515 		r = RESUME_HOST;
1516 		break;
1517 	case EMULATE_FAIL:
1518 		pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1519 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1520 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1521 		r = RESUME_HOST;
1522 		break;
1523 	default:
1524 		r = RESUME_GUEST;
1525 		break;
1526 	}
1527 	return r;
1528 }
1529 #endif /* CONFIG_VSX */
1530 
1531 #ifdef CONFIG_ALTIVEC
1532 int kvmppc_handle_vmx_load(struct kvm_vcpu *vcpu,
1533 		unsigned int rt, unsigned int bytes, int is_default_endian)
1534 {
1535 	enum emulation_result emulated = EMULATE_DONE;
1536 
1537 	if (vcpu->arch.mmio_vmx_copy_nums > 2)
1538 		return EMULATE_FAIL;
1539 
1540 	while (vcpu->arch.mmio_vmx_copy_nums) {
1541 		emulated = __kvmppc_handle_load(vcpu, rt, bytes,
1542 				is_default_endian, 0);
1543 
1544 		if (emulated != EMULATE_DONE)
1545 			break;
1546 
1547 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1548 		vcpu->arch.mmio_vmx_copy_nums--;
1549 		vcpu->arch.mmio_vmx_offset++;
1550 	}
1551 
1552 	return emulated;
1553 }
1554 
1555 static int kvmppc_get_vmx_dword(struct kvm_vcpu *vcpu, int index, u64 *val)
1556 {
1557 	union kvmppc_one_reg reg;
1558 	int vmx_offset = 0;
1559 	int result = 0;
1560 
1561 	vmx_offset =
1562 		kvmppc_get_vmx_dword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1563 
1564 	if (vmx_offset == -1)
1565 		return -1;
1566 
1567 	reg.vval = VCPU_VSX_VR(vcpu, index);
1568 	*val = reg.vsxval[vmx_offset];
1569 
1570 	return result;
1571 }
1572 
1573 static int kvmppc_get_vmx_word(struct kvm_vcpu *vcpu, int index, u64 *val)
1574 {
1575 	union kvmppc_one_reg reg;
1576 	int vmx_offset = 0;
1577 	int result = 0;
1578 
1579 	vmx_offset =
1580 		kvmppc_get_vmx_word_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1581 
1582 	if (vmx_offset == -1)
1583 		return -1;
1584 
1585 	reg.vval = VCPU_VSX_VR(vcpu, index);
1586 	*val = reg.vsx32val[vmx_offset];
1587 
1588 	return result;
1589 }
1590 
1591 static int kvmppc_get_vmx_hword(struct kvm_vcpu *vcpu, int index, u64 *val)
1592 {
1593 	union kvmppc_one_reg reg;
1594 	int vmx_offset = 0;
1595 	int result = 0;
1596 
1597 	vmx_offset =
1598 		kvmppc_get_vmx_hword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1599 
1600 	if (vmx_offset == -1)
1601 		return -1;
1602 
1603 	reg.vval = VCPU_VSX_VR(vcpu, index);
1604 	*val = reg.vsx16val[vmx_offset];
1605 
1606 	return result;
1607 }
1608 
1609 static int kvmppc_get_vmx_byte(struct kvm_vcpu *vcpu, int index, u64 *val)
1610 {
1611 	union kvmppc_one_reg reg;
1612 	int vmx_offset = 0;
1613 	int result = 0;
1614 
1615 	vmx_offset =
1616 		kvmppc_get_vmx_byte_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1617 
1618 	if (vmx_offset == -1)
1619 		return -1;
1620 
1621 	reg.vval = VCPU_VSX_VR(vcpu, index);
1622 	*val = reg.vsx8val[vmx_offset];
1623 
1624 	return result;
1625 }
1626 
1627 int kvmppc_handle_vmx_store(struct kvm_vcpu *vcpu,
1628 		unsigned int rs, unsigned int bytes, int is_default_endian)
1629 {
1630 	u64 val = 0;
1631 	unsigned int index = rs & KVM_MMIO_REG_MASK;
1632 	enum emulation_result emulated = EMULATE_DONE;
1633 
1634 	if (vcpu->arch.mmio_vmx_copy_nums > 2)
1635 		return EMULATE_FAIL;
1636 
1637 	vcpu->arch.io_gpr = rs;
1638 
1639 	while (vcpu->arch.mmio_vmx_copy_nums) {
1640 		switch (vcpu->arch.mmio_copy_type) {
1641 		case KVMPPC_VMX_COPY_DWORD:
1642 			if (kvmppc_get_vmx_dword(vcpu, index, &val) == -1)
1643 				return EMULATE_FAIL;
1644 
1645 			break;
1646 		case KVMPPC_VMX_COPY_WORD:
1647 			if (kvmppc_get_vmx_word(vcpu, index, &val) == -1)
1648 				return EMULATE_FAIL;
1649 			break;
1650 		case KVMPPC_VMX_COPY_HWORD:
1651 			if (kvmppc_get_vmx_hword(vcpu, index, &val) == -1)
1652 				return EMULATE_FAIL;
1653 			break;
1654 		case KVMPPC_VMX_COPY_BYTE:
1655 			if (kvmppc_get_vmx_byte(vcpu, index, &val) == -1)
1656 				return EMULATE_FAIL;
1657 			break;
1658 		default:
1659 			return EMULATE_FAIL;
1660 		}
1661 
1662 		emulated = kvmppc_handle_store(vcpu, val, bytes,
1663 				is_default_endian);
1664 		if (emulated != EMULATE_DONE)
1665 			break;
1666 
1667 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1668 		vcpu->arch.mmio_vmx_copy_nums--;
1669 		vcpu->arch.mmio_vmx_offset++;
1670 	}
1671 
1672 	return emulated;
1673 }
1674 
1675 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu)
1676 {
1677 	struct kvm_run *run = vcpu->run;
1678 	enum emulation_result emulated = EMULATE_FAIL;
1679 	int r;
1680 
1681 	vcpu->arch.paddr_accessed += run->mmio.len;
1682 
1683 	if (!vcpu->mmio_is_write) {
1684 		emulated = kvmppc_handle_vmx_load(vcpu,
1685 				vcpu->arch.io_gpr, run->mmio.len, 1);
1686 	} else {
1687 		emulated = kvmppc_handle_vmx_store(vcpu,
1688 				vcpu->arch.io_gpr, run->mmio.len, 1);
1689 	}
1690 
1691 	switch (emulated) {
1692 	case EMULATE_DO_MMIO:
1693 		run->exit_reason = KVM_EXIT_MMIO;
1694 		r = RESUME_HOST;
1695 		break;
1696 	case EMULATE_FAIL:
1697 		pr_info("KVM: MMIO emulation failed (VMX repeat)\n");
1698 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1699 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1700 		r = RESUME_HOST;
1701 		break;
1702 	default:
1703 		r = RESUME_GUEST;
1704 		break;
1705 	}
1706 	return r;
1707 }
1708 #endif /* CONFIG_ALTIVEC */
1709 
1710 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1711 {
1712 	int r = 0;
1713 	union kvmppc_one_reg val;
1714 	int size;
1715 
1716 	size = one_reg_size(reg->id);
1717 	if (size > sizeof(val))
1718 		return -EINVAL;
1719 
1720 	r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1721 	if (r == -EINVAL) {
1722 		r = 0;
1723 		switch (reg->id) {
1724 #ifdef CONFIG_ALTIVEC
1725 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1726 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1727 				r = -ENXIO;
1728 				break;
1729 			}
1730 			val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1731 			break;
1732 		case KVM_REG_PPC_VSCR:
1733 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1734 				r = -ENXIO;
1735 				break;
1736 			}
1737 			val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1738 			break;
1739 		case KVM_REG_PPC_VRSAVE:
1740 			val = get_reg_val(reg->id, vcpu->arch.vrsave);
1741 			break;
1742 #endif /* CONFIG_ALTIVEC */
1743 		default:
1744 			r = -EINVAL;
1745 			break;
1746 		}
1747 	}
1748 
1749 	if (r)
1750 		return r;
1751 
1752 	if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1753 		r = -EFAULT;
1754 
1755 	return r;
1756 }
1757 
1758 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1759 {
1760 	int r;
1761 	union kvmppc_one_reg val;
1762 	int size;
1763 
1764 	size = one_reg_size(reg->id);
1765 	if (size > sizeof(val))
1766 		return -EINVAL;
1767 
1768 	if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1769 		return -EFAULT;
1770 
1771 	r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1772 	if (r == -EINVAL) {
1773 		r = 0;
1774 		switch (reg->id) {
1775 #ifdef CONFIG_ALTIVEC
1776 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1777 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1778 				r = -ENXIO;
1779 				break;
1780 			}
1781 			vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1782 			break;
1783 		case KVM_REG_PPC_VSCR:
1784 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1785 				r = -ENXIO;
1786 				break;
1787 			}
1788 			vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1789 			break;
1790 		case KVM_REG_PPC_VRSAVE:
1791 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1792 				r = -ENXIO;
1793 				break;
1794 			}
1795 			vcpu->arch.vrsave = set_reg_val(reg->id, val);
1796 			break;
1797 #endif /* CONFIG_ALTIVEC */
1798 		default:
1799 			r = -EINVAL;
1800 			break;
1801 		}
1802 	}
1803 
1804 	return r;
1805 }
1806 
1807 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
1808 {
1809 	struct kvm_run *run = vcpu->run;
1810 	int r;
1811 
1812 	vcpu_load(vcpu);
1813 
1814 	if (vcpu->mmio_needed) {
1815 		vcpu->mmio_needed = 0;
1816 		if (!vcpu->mmio_is_write)
1817 			kvmppc_complete_mmio_load(vcpu);
1818 #ifdef CONFIG_VSX
1819 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1820 			vcpu->arch.mmio_vsx_copy_nums--;
1821 			vcpu->arch.mmio_vsx_offset++;
1822 		}
1823 
1824 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1825 			r = kvmppc_emulate_mmio_vsx_loadstore(vcpu);
1826 			if (r == RESUME_HOST) {
1827 				vcpu->mmio_needed = 1;
1828 				goto out;
1829 			}
1830 		}
1831 #endif
1832 #ifdef CONFIG_ALTIVEC
1833 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1834 			vcpu->arch.mmio_vmx_copy_nums--;
1835 			vcpu->arch.mmio_vmx_offset++;
1836 		}
1837 
1838 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1839 			r = kvmppc_emulate_mmio_vmx_loadstore(vcpu);
1840 			if (r == RESUME_HOST) {
1841 				vcpu->mmio_needed = 1;
1842 				goto out;
1843 			}
1844 		}
1845 #endif
1846 	} else if (vcpu->arch.osi_needed) {
1847 		u64 *gprs = run->osi.gprs;
1848 		int i;
1849 
1850 		for (i = 0; i < 32; i++)
1851 			kvmppc_set_gpr(vcpu, i, gprs[i]);
1852 		vcpu->arch.osi_needed = 0;
1853 	} else if (vcpu->arch.hcall_needed) {
1854 		int i;
1855 
1856 		kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1857 		for (i = 0; i < 9; ++i)
1858 			kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1859 		vcpu->arch.hcall_needed = 0;
1860 #ifdef CONFIG_BOOKE
1861 	} else if (vcpu->arch.epr_needed) {
1862 		kvmppc_set_epr(vcpu, run->epr.epr);
1863 		vcpu->arch.epr_needed = 0;
1864 #endif
1865 	}
1866 
1867 	kvm_sigset_activate(vcpu);
1868 
1869 	if (run->immediate_exit)
1870 		r = -EINTR;
1871 	else
1872 		r = kvmppc_vcpu_run(vcpu);
1873 
1874 	kvm_sigset_deactivate(vcpu);
1875 
1876 #ifdef CONFIG_ALTIVEC
1877 out:
1878 #endif
1879 
1880 	/*
1881 	 * We're already returning to userspace, don't pass the
1882 	 * RESUME_HOST flags along.
1883 	 */
1884 	if (r > 0)
1885 		r = 0;
1886 
1887 	vcpu_put(vcpu);
1888 	return r;
1889 }
1890 
1891 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1892 {
1893 	if (irq->irq == KVM_INTERRUPT_UNSET) {
1894 		kvmppc_core_dequeue_external(vcpu);
1895 		return 0;
1896 	}
1897 
1898 	kvmppc_core_queue_external(vcpu, irq);
1899 
1900 	kvm_vcpu_kick(vcpu);
1901 
1902 	return 0;
1903 }
1904 
1905 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1906 				     struct kvm_enable_cap *cap)
1907 {
1908 	int r;
1909 
1910 	if (cap->flags)
1911 		return -EINVAL;
1912 
1913 	switch (cap->cap) {
1914 	case KVM_CAP_PPC_OSI:
1915 		r = 0;
1916 		vcpu->arch.osi_enabled = true;
1917 		break;
1918 	case KVM_CAP_PPC_PAPR:
1919 		r = 0;
1920 		vcpu->arch.papr_enabled = true;
1921 		break;
1922 	case KVM_CAP_PPC_EPR:
1923 		r = 0;
1924 		if (cap->args[0])
1925 			vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1926 		else
1927 			vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1928 		break;
1929 #ifdef CONFIG_BOOKE
1930 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
1931 		r = 0;
1932 		vcpu->arch.watchdog_enabled = true;
1933 		break;
1934 #endif
1935 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1936 	case KVM_CAP_SW_TLB: {
1937 		struct kvm_config_tlb cfg;
1938 		void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1939 
1940 		r = -EFAULT;
1941 		if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1942 			break;
1943 
1944 		r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1945 		break;
1946 	}
1947 #endif
1948 #ifdef CONFIG_KVM_MPIC
1949 	case KVM_CAP_IRQ_MPIC: {
1950 		struct fd f;
1951 		struct kvm_device *dev;
1952 
1953 		r = -EBADF;
1954 		f = fdget(cap->args[0]);
1955 		if (!f.file)
1956 			break;
1957 
1958 		r = -EPERM;
1959 		dev = kvm_device_from_filp(f.file);
1960 		if (dev)
1961 			r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1962 
1963 		fdput(f);
1964 		break;
1965 	}
1966 #endif
1967 #ifdef CONFIG_KVM_XICS
1968 	case KVM_CAP_IRQ_XICS: {
1969 		struct fd f;
1970 		struct kvm_device *dev;
1971 
1972 		r = -EBADF;
1973 		f = fdget(cap->args[0]);
1974 		if (!f.file)
1975 			break;
1976 
1977 		r = -EPERM;
1978 		dev = kvm_device_from_filp(f.file);
1979 		if (dev) {
1980 			if (xics_on_xive())
1981 				r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1982 			else
1983 				r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1984 		}
1985 
1986 		fdput(f);
1987 		break;
1988 	}
1989 #endif /* CONFIG_KVM_XICS */
1990 #ifdef CONFIG_KVM_XIVE
1991 	case KVM_CAP_PPC_IRQ_XIVE: {
1992 		struct fd f;
1993 		struct kvm_device *dev;
1994 
1995 		r = -EBADF;
1996 		f = fdget(cap->args[0]);
1997 		if (!f.file)
1998 			break;
1999 
2000 		r = -ENXIO;
2001 		if (!xive_enabled())
2002 			break;
2003 
2004 		r = -EPERM;
2005 		dev = kvm_device_from_filp(f.file);
2006 		if (dev)
2007 			r = kvmppc_xive_native_connect_vcpu(dev, vcpu,
2008 							    cap->args[1]);
2009 
2010 		fdput(f);
2011 		break;
2012 	}
2013 #endif /* CONFIG_KVM_XIVE */
2014 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
2015 	case KVM_CAP_PPC_FWNMI:
2016 		r = -EINVAL;
2017 		if (!is_kvmppc_hv_enabled(vcpu->kvm))
2018 			break;
2019 		r = 0;
2020 		vcpu->kvm->arch.fwnmi_enabled = true;
2021 		break;
2022 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
2023 	default:
2024 		r = -EINVAL;
2025 		break;
2026 	}
2027 
2028 	if (!r)
2029 		r = kvmppc_sanity_check(vcpu);
2030 
2031 	return r;
2032 }
2033 
2034 bool kvm_arch_intc_initialized(struct kvm *kvm)
2035 {
2036 #ifdef CONFIG_KVM_MPIC
2037 	if (kvm->arch.mpic)
2038 		return true;
2039 #endif
2040 #ifdef CONFIG_KVM_XICS
2041 	if (kvm->arch.xics || kvm->arch.xive)
2042 		return true;
2043 #endif
2044 	return false;
2045 }
2046 
2047 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2048                                     struct kvm_mp_state *mp_state)
2049 {
2050 	return -EINVAL;
2051 }
2052 
2053 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2054                                     struct kvm_mp_state *mp_state)
2055 {
2056 	return -EINVAL;
2057 }
2058 
2059 long kvm_arch_vcpu_async_ioctl(struct file *filp,
2060 			       unsigned int ioctl, unsigned long arg)
2061 {
2062 	struct kvm_vcpu *vcpu = filp->private_data;
2063 	void __user *argp = (void __user *)arg;
2064 
2065 	if (ioctl == KVM_INTERRUPT) {
2066 		struct kvm_interrupt irq;
2067 		if (copy_from_user(&irq, argp, sizeof(irq)))
2068 			return -EFAULT;
2069 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2070 	}
2071 	return -ENOIOCTLCMD;
2072 }
2073 
2074 long kvm_arch_vcpu_ioctl(struct file *filp,
2075                          unsigned int ioctl, unsigned long arg)
2076 {
2077 	struct kvm_vcpu *vcpu = filp->private_data;
2078 	void __user *argp = (void __user *)arg;
2079 	long r;
2080 
2081 	switch (ioctl) {
2082 	case KVM_ENABLE_CAP:
2083 	{
2084 		struct kvm_enable_cap cap;
2085 		r = -EFAULT;
2086 		if (copy_from_user(&cap, argp, sizeof(cap)))
2087 			goto out;
2088 		vcpu_load(vcpu);
2089 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
2090 		vcpu_put(vcpu);
2091 		break;
2092 	}
2093 
2094 	case KVM_SET_ONE_REG:
2095 	case KVM_GET_ONE_REG:
2096 	{
2097 		struct kvm_one_reg reg;
2098 		r = -EFAULT;
2099 		if (copy_from_user(&reg, argp, sizeof(reg)))
2100 			goto out;
2101 		if (ioctl == KVM_SET_ONE_REG)
2102 			r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
2103 		else
2104 			r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
2105 		break;
2106 	}
2107 
2108 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
2109 	case KVM_DIRTY_TLB: {
2110 		struct kvm_dirty_tlb dirty;
2111 		r = -EFAULT;
2112 		if (copy_from_user(&dirty, argp, sizeof(dirty)))
2113 			goto out;
2114 		vcpu_load(vcpu);
2115 		r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
2116 		vcpu_put(vcpu);
2117 		break;
2118 	}
2119 #endif
2120 	default:
2121 		r = -EINVAL;
2122 	}
2123 
2124 out:
2125 	return r;
2126 }
2127 
2128 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2129 {
2130 	return VM_FAULT_SIGBUS;
2131 }
2132 
2133 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
2134 {
2135 	u32 inst_nop = 0x60000000;
2136 #ifdef CONFIG_KVM_BOOKE_HV
2137 	u32 inst_sc1 = 0x44000022;
2138 	pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
2139 	pvinfo->hcall[1] = cpu_to_be32(inst_nop);
2140 	pvinfo->hcall[2] = cpu_to_be32(inst_nop);
2141 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2142 #else
2143 	u32 inst_lis = 0x3c000000;
2144 	u32 inst_ori = 0x60000000;
2145 	u32 inst_sc = 0x44000002;
2146 	u32 inst_imm_mask = 0xffff;
2147 
2148 	/*
2149 	 * The hypercall to get into KVM from within guest context is as
2150 	 * follows:
2151 	 *
2152 	 *    lis r0, r0, KVM_SC_MAGIC_R0@h
2153 	 *    ori r0, KVM_SC_MAGIC_R0@l
2154 	 *    sc
2155 	 *    nop
2156 	 */
2157 	pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
2158 	pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
2159 	pvinfo->hcall[2] = cpu_to_be32(inst_sc);
2160 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2161 #endif
2162 
2163 	pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
2164 
2165 	return 0;
2166 }
2167 
2168 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
2169 			  bool line_status)
2170 {
2171 	if (!irqchip_in_kernel(kvm))
2172 		return -ENXIO;
2173 
2174 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2175 					irq_event->irq, irq_event->level,
2176 					line_status);
2177 	return 0;
2178 }
2179 
2180 
2181 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
2182 			    struct kvm_enable_cap *cap)
2183 {
2184 	int r;
2185 
2186 	if (cap->flags)
2187 		return -EINVAL;
2188 
2189 	switch (cap->cap) {
2190 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
2191 	case KVM_CAP_PPC_ENABLE_HCALL: {
2192 		unsigned long hcall = cap->args[0];
2193 
2194 		r = -EINVAL;
2195 		if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
2196 		    cap->args[1] > 1)
2197 			break;
2198 		if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
2199 			break;
2200 		if (cap->args[1])
2201 			set_bit(hcall / 4, kvm->arch.enabled_hcalls);
2202 		else
2203 			clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
2204 		r = 0;
2205 		break;
2206 	}
2207 	case KVM_CAP_PPC_SMT: {
2208 		unsigned long mode = cap->args[0];
2209 		unsigned long flags = cap->args[1];
2210 
2211 		r = -EINVAL;
2212 		if (kvm->arch.kvm_ops->set_smt_mode)
2213 			r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
2214 		break;
2215 	}
2216 
2217 	case KVM_CAP_PPC_NESTED_HV:
2218 		r = -EINVAL;
2219 		if (!is_kvmppc_hv_enabled(kvm) ||
2220 		    !kvm->arch.kvm_ops->enable_nested)
2221 			break;
2222 		r = kvm->arch.kvm_ops->enable_nested(kvm);
2223 		break;
2224 #endif
2225 #if defined(CONFIG_KVM_BOOK3S_HV_POSSIBLE)
2226 	case KVM_CAP_PPC_SECURE_GUEST:
2227 		r = -EINVAL;
2228 		if (!is_kvmppc_hv_enabled(kvm) || !kvm->arch.kvm_ops->enable_svm)
2229 			break;
2230 		r = kvm->arch.kvm_ops->enable_svm(kvm);
2231 		break;
2232 	case KVM_CAP_PPC_DAWR1:
2233 		r = -EINVAL;
2234 		if (!is_kvmppc_hv_enabled(kvm) || !kvm->arch.kvm_ops->enable_dawr1)
2235 			break;
2236 		r = kvm->arch.kvm_ops->enable_dawr1(kvm);
2237 		break;
2238 #endif
2239 	default:
2240 		r = -EINVAL;
2241 		break;
2242 	}
2243 
2244 	return r;
2245 }
2246 
2247 #ifdef CONFIG_PPC_BOOK3S_64
2248 /*
2249  * These functions check whether the underlying hardware is safe
2250  * against attacks based on observing the effects of speculatively
2251  * executed instructions, and whether it supplies instructions for
2252  * use in workarounds.  The information comes from firmware, either
2253  * via the device tree on powernv platforms or from an hcall on
2254  * pseries platforms.
2255  */
2256 #ifdef CONFIG_PPC_PSERIES
2257 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2258 {
2259 	struct h_cpu_char_result c;
2260 	unsigned long rc;
2261 
2262 	if (!machine_is(pseries))
2263 		return -ENOTTY;
2264 
2265 	rc = plpar_get_cpu_characteristics(&c);
2266 	if (rc == H_SUCCESS) {
2267 		cp->character = c.character;
2268 		cp->behaviour = c.behaviour;
2269 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2270 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2271 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2272 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2273 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2274 			KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
2275 			KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
2276 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2277 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2278 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2279 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2280 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2281 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2282 	}
2283 	return 0;
2284 }
2285 #else
2286 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2287 {
2288 	return -ENOTTY;
2289 }
2290 #endif
2291 
2292 static inline bool have_fw_feat(struct device_node *fw_features,
2293 				const char *state, const char *name)
2294 {
2295 	struct device_node *np;
2296 	bool r = false;
2297 
2298 	np = of_get_child_by_name(fw_features, name);
2299 	if (np) {
2300 		r = of_property_read_bool(np, state);
2301 		of_node_put(np);
2302 	}
2303 	return r;
2304 }
2305 
2306 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2307 {
2308 	struct device_node *np, *fw_features;
2309 	int r;
2310 
2311 	memset(cp, 0, sizeof(*cp));
2312 	r = pseries_get_cpu_char(cp);
2313 	if (r != -ENOTTY)
2314 		return r;
2315 
2316 	np = of_find_node_by_name(NULL, "ibm,opal");
2317 	if (np) {
2318 		fw_features = of_get_child_by_name(np, "fw-features");
2319 		of_node_put(np);
2320 		if (!fw_features)
2321 			return 0;
2322 		if (have_fw_feat(fw_features, "enabled",
2323 				 "inst-spec-barrier-ori31,31,0"))
2324 			cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
2325 		if (have_fw_feat(fw_features, "enabled",
2326 				 "fw-bcctrl-serialized"))
2327 			cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
2328 		if (have_fw_feat(fw_features, "enabled",
2329 				 "inst-l1d-flush-ori30,30,0"))
2330 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
2331 		if (have_fw_feat(fw_features, "enabled",
2332 				 "inst-l1d-flush-trig2"))
2333 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
2334 		if (have_fw_feat(fw_features, "enabled",
2335 				 "fw-l1d-thread-split"))
2336 			cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
2337 		if (have_fw_feat(fw_features, "enabled",
2338 				 "fw-count-cache-disabled"))
2339 			cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2340 		if (have_fw_feat(fw_features, "enabled",
2341 				 "fw-count-cache-flush-bcctr2,0,0"))
2342 			cp->character |= KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2343 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2344 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2345 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2346 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2347 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2348 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2349 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2350 
2351 		if (have_fw_feat(fw_features, "enabled",
2352 				 "speculation-policy-favor-security"))
2353 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
2354 		if (!have_fw_feat(fw_features, "disabled",
2355 				  "needs-l1d-flush-msr-pr-0-to-1"))
2356 			cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
2357 		if (!have_fw_feat(fw_features, "disabled",
2358 				  "needs-spec-barrier-for-bound-checks"))
2359 			cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2360 		if (have_fw_feat(fw_features, "enabled",
2361 				 "needs-count-cache-flush-on-context-switch"))
2362 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2363 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2364 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2365 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2366 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2367 
2368 		of_node_put(fw_features);
2369 	}
2370 
2371 	return 0;
2372 }
2373 #endif
2374 
2375 long kvm_arch_vm_ioctl(struct file *filp,
2376                        unsigned int ioctl, unsigned long arg)
2377 {
2378 	struct kvm *kvm __maybe_unused = filp->private_data;
2379 	void __user *argp = (void __user *)arg;
2380 	long r;
2381 
2382 	switch (ioctl) {
2383 	case KVM_PPC_GET_PVINFO: {
2384 		struct kvm_ppc_pvinfo pvinfo;
2385 		memset(&pvinfo, 0, sizeof(pvinfo));
2386 		r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
2387 		if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
2388 			r = -EFAULT;
2389 			goto out;
2390 		}
2391 
2392 		break;
2393 	}
2394 #ifdef CONFIG_SPAPR_TCE_IOMMU
2395 	case KVM_CREATE_SPAPR_TCE_64: {
2396 		struct kvm_create_spapr_tce_64 create_tce_64;
2397 
2398 		r = -EFAULT;
2399 		if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
2400 			goto out;
2401 		if (create_tce_64.flags) {
2402 			r = -EINVAL;
2403 			goto out;
2404 		}
2405 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2406 		goto out;
2407 	}
2408 	case KVM_CREATE_SPAPR_TCE: {
2409 		struct kvm_create_spapr_tce create_tce;
2410 		struct kvm_create_spapr_tce_64 create_tce_64;
2411 
2412 		r = -EFAULT;
2413 		if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
2414 			goto out;
2415 
2416 		create_tce_64.liobn = create_tce.liobn;
2417 		create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
2418 		create_tce_64.offset = 0;
2419 		create_tce_64.size = create_tce.window_size >>
2420 				IOMMU_PAGE_SHIFT_4K;
2421 		create_tce_64.flags = 0;
2422 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2423 		goto out;
2424 	}
2425 #endif
2426 #ifdef CONFIG_PPC_BOOK3S_64
2427 	case KVM_PPC_GET_SMMU_INFO: {
2428 		struct kvm_ppc_smmu_info info;
2429 		struct kvm *kvm = filp->private_data;
2430 
2431 		memset(&info, 0, sizeof(info));
2432 		r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
2433 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2434 			r = -EFAULT;
2435 		break;
2436 	}
2437 	case KVM_PPC_RTAS_DEFINE_TOKEN: {
2438 		struct kvm *kvm = filp->private_data;
2439 
2440 		r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
2441 		break;
2442 	}
2443 	case KVM_PPC_CONFIGURE_V3_MMU: {
2444 		struct kvm *kvm = filp->private_data;
2445 		struct kvm_ppc_mmuv3_cfg cfg;
2446 
2447 		r = -EINVAL;
2448 		if (!kvm->arch.kvm_ops->configure_mmu)
2449 			goto out;
2450 		r = -EFAULT;
2451 		if (copy_from_user(&cfg, argp, sizeof(cfg)))
2452 			goto out;
2453 		r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
2454 		break;
2455 	}
2456 	case KVM_PPC_GET_RMMU_INFO: {
2457 		struct kvm *kvm = filp->private_data;
2458 		struct kvm_ppc_rmmu_info info;
2459 
2460 		r = -EINVAL;
2461 		if (!kvm->arch.kvm_ops->get_rmmu_info)
2462 			goto out;
2463 		r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
2464 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2465 			r = -EFAULT;
2466 		break;
2467 	}
2468 	case KVM_PPC_GET_CPU_CHAR: {
2469 		struct kvm_ppc_cpu_char cpuchar;
2470 
2471 		r = kvmppc_get_cpu_char(&cpuchar);
2472 		if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2473 			r = -EFAULT;
2474 		break;
2475 	}
2476 	case KVM_PPC_SVM_OFF: {
2477 		struct kvm *kvm = filp->private_data;
2478 
2479 		r = 0;
2480 		if (!kvm->arch.kvm_ops->svm_off)
2481 			goto out;
2482 
2483 		r = kvm->arch.kvm_ops->svm_off(kvm);
2484 		break;
2485 	}
2486 	default: {
2487 		struct kvm *kvm = filp->private_data;
2488 		r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
2489 	}
2490 #else /* CONFIG_PPC_BOOK3S_64 */
2491 	default:
2492 		r = -ENOTTY;
2493 #endif
2494 	}
2495 out:
2496 	return r;
2497 }
2498 
2499 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
2500 static unsigned long nr_lpids;
2501 
2502 long kvmppc_alloc_lpid(void)
2503 {
2504 	long lpid;
2505 
2506 	do {
2507 		lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
2508 		if (lpid >= nr_lpids) {
2509 			pr_err("%s: No LPIDs free\n", __func__);
2510 			return -ENOMEM;
2511 		}
2512 	} while (test_and_set_bit(lpid, lpid_inuse));
2513 
2514 	return lpid;
2515 }
2516 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2517 
2518 void kvmppc_claim_lpid(long lpid)
2519 {
2520 	set_bit(lpid, lpid_inuse);
2521 }
2522 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
2523 
2524 void kvmppc_free_lpid(long lpid)
2525 {
2526 	clear_bit(lpid, lpid_inuse);
2527 }
2528 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2529 
2530 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2531 {
2532 	nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
2533 	memset(lpid_inuse, 0, sizeof(lpid_inuse));
2534 }
2535 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2536 
2537 int kvm_arch_init(void *opaque)
2538 {
2539 	return 0;
2540 }
2541 
2542 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);
2543 
2544 void kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry)
2545 {
2546 	if (vcpu->kvm->arch.kvm_ops->create_vcpu_debugfs)
2547 		vcpu->kvm->arch.kvm_ops->create_vcpu_debugfs(vcpu, debugfs_dentry);
2548 }
2549 
2550 int kvm_arch_create_vm_debugfs(struct kvm *kvm)
2551 {
2552 	if (kvm->arch.kvm_ops->create_vm_debugfs)
2553 		kvm->arch.kvm_ops->create_vm_debugfs(kvm);
2554 	return 0;
2555 }
2556