xref: /openbmc/linux/arch/powerpc/kvm/powerpc.c (revision c059ee9d)
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 <linux/of.h>
23 #include <asm/cputable.h>
24 #include <linux/uaccess.h>
25 #include <asm/kvm_ppc.h>
26 #include <asm/cputhreads.h>
27 #include <asm/irqflags.h>
28 #include <asm/iommu.h>
29 #include <asm/switch_to.h>
30 #include <asm/xive.h>
31 #ifdef CONFIG_PPC_PSERIES
32 #include <asm/hvcall.h>
33 #include <asm/plpar_wrappers.h>
34 #endif
35 #include <asm/ultravisor.h>
36 
37 #include "timing.h"
38 #include "irq.h"
39 #include "../mm/mmu_decl.h"
40 
41 #define CREATE_TRACE_POINTS
42 #include "trace.h"
43 
44 struct kvmppc_ops *kvmppc_hv_ops;
45 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
46 struct kvmppc_ops *kvmppc_pr_ops;
47 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
48 
49 
50 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
51 {
52 	return !!(v->arch.pending_exceptions) || kvm_request_pending(v);
53 }
54 
55 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
56 {
57 	return kvm_arch_vcpu_runnable(vcpu);
58 }
59 
60 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
61 {
62 	return false;
63 }
64 
65 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
66 {
67 	return 1;
68 }
69 
70 /*
71  * Common checks before entering the guest world.  Call with interrupts
72  * disabled.
73  *
74  * returns:
75  *
76  * == 1 if we're ready to go into guest state
77  * <= 0 if we need to go back to the host with return value
78  */
79 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
80 {
81 	int r;
82 
83 	WARN_ON(irqs_disabled());
84 	hard_irq_disable();
85 
86 	while (true) {
87 		if (need_resched()) {
88 			local_irq_enable();
89 			cond_resched();
90 			hard_irq_disable();
91 			continue;
92 		}
93 
94 		if (signal_pending(current)) {
95 			kvmppc_account_exit(vcpu, SIGNAL_EXITS);
96 			vcpu->run->exit_reason = KVM_EXIT_INTR;
97 			r = -EINTR;
98 			break;
99 		}
100 
101 		vcpu->mode = IN_GUEST_MODE;
102 
103 		/*
104 		 * Reading vcpu->requests must happen after setting vcpu->mode,
105 		 * so we don't miss a request because the requester sees
106 		 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
107 		 * before next entering the guest (and thus doesn't IPI).
108 		 * This also orders the write to mode from any reads
109 		 * to the page tables done while the VCPU is running.
110 		 * Please see the comment in kvm_flush_remote_tlbs.
111 		 */
112 		smp_mb();
113 
114 		if (kvm_request_pending(vcpu)) {
115 			/* Make sure we process requests preemptable */
116 			local_irq_enable();
117 			trace_kvm_check_requests(vcpu);
118 			r = kvmppc_core_check_requests(vcpu);
119 			hard_irq_disable();
120 			if (r > 0)
121 				continue;
122 			break;
123 		}
124 
125 		if (kvmppc_core_prepare_to_enter(vcpu)) {
126 			/* interrupts got enabled in between, so we
127 			   are back at square 1 */
128 			continue;
129 		}
130 
131 		guest_enter_irqoff();
132 		return 1;
133 	}
134 
135 	/* return to host */
136 	local_irq_enable();
137 	return r;
138 }
139 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
140 
141 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
142 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
143 {
144 	struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
145 	int i;
146 
147 	shared->sprg0 = swab64(shared->sprg0);
148 	shared->sprg1 = swab64(shared->sprg1);
149 	shared->sprg2 = swab64(shared->sprg2);
150 	shared->sprg3 = swab64(shared->sprg3);
151 	shared->srr0 = swab64(shared->srr0);
152 	shared->srr1 = swab64(shared->srr1);
153 	shared->dar = swab64(shared->dar);
154 	shared->msr = swab64(shared->msr);
155 	shared->dsisr = swab32(shared->dsisr);
156 	shared->int_pending = swab32(shared->int_pending);
157 	for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
158 		shared->sr[i] = swab32(shared->sr[i]);
159 }
160 #endif
161 
162 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
163 {
164 	int nr = kvmppc_get_gpr(vcpu, 11);
165 	int r;
166 	unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
167 	unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
168 	unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
169 	unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
170 	unsigned long r2 = 0;
171 
172 	if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
173 		/* 32 bit mode */
174 		param1 &= 0xffffffff;
175 		param2 &= 0xffffffff;
176 		param3 &= 0xffffffff;
177 		param4 &= 0xffffffff;
178 	}
179 
180 	switch (nr) {
181 	case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
182 	{
183 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
184 		/* Book3S can be little endian, find it out here */
185 		int shared_big_endian = true;
186 		if (vcpu->arch.intr_msr & MSR_LE)
187 			shared_big_endian = false;
188 		if (shared_big_endian != vcpu->arch.shared_big_endian)
189 			kvmppc_swab_shared(vcpu);
190 		vcpu->arch.shared_big_endian = shared_big_endian;
191 #endif
192 
193 		if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
194 			/*
195 			 * Older versions of the Linux magic page code had
196 			 * a bug where they would map their trampoline code
197 			 * NX. If that's the case, remove !PR NX capability.
198 			 */
199 			vcpu->arch.disable_kernel_nx = true;
200 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
201 		}
202 
203 		vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
204 		vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
205 
206 #ifdef CONFIG_PPC_64K_PAGES
207 		/*
208 		 * Make sure our 4k magic page is in the same window of a 64k
209 		 * page within the guest and within the host's page.
210 		 */
211 		if ((vcpu->arch.magic_page_pa & 0xf000) !=
212 		    ((ulong)vcpu->arch.shared & 0xf000)) {
213 			void *old_shared = vcpu->arch.shared;
214 			ulong shared = (ulong)vcpu->arch.shared;
215 			void *new_shared;
216 
217 			shared &= PAGE_MASK;
218 			shared |= vcpu->arch.magic_page_pa & 0xf000;
219 			new_shared = (void*)shared;
220 			memcpy(new_shared, old_shared, 0x1000);
221 			vcpu->arch.shared = new_shared;
222 		}
223 #endif
224 
225 		r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
226 
227 		r = EV_SUCCESS;
228 		break;
229 	}
230 	case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
231 		r = EV_SUCCESS;
232 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
233 		r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
234 #endif
235 
236 		/* Second return value is in r4 */
237 		break;
238 	case EV_HCALL_TOKEN(EV_IDLE):
239 		r = EV_SUCCESS;
240 		kvm_vcpu_halt(vcpu);
241 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
242 		break;
243 	default:
244 		r = EV_UNIMPLEMENTED;
245 		break;
246 	}
247 
248 	kvmppc_set_gpr(vcpu, 4, r2);
249 
250 	return r;
251 }
252 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
253 
254 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
255 {
256 	int r = false;
257 
258 	/* We have to know what CPU to virtualize */
259 	if (!vcpu->arch.pvr)
260 		goto out;
261 
262 	/* PAPR only works with book3s_64 */
263 	if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
264 		goto out;
265 
266 	/* HV KVM can only do PAPR mode for now */
267 	if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
268 		goto out;
269 
270 #ifdef CONFIG_KVM_BOOKE_HV
271 	if (!cpu_has_feature(CPU_FTR_EMB_HV))
272 		goto out;
273 #endif
274 
275 	r = true;
276 
277 out:
278 	vcpu->arch.sane = r;
279 	return r ? 0 : -EINVAL;
280 }
281 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
282 
283 int kvmppc_emulate_mmio(struct kvm_vcpu *vcpu)
284 {
285 	enum emulation_result er;
286 	int r;
287 
288 	er = kvmppc_emulate_loadstore(vcpu);
289 	switch (er) {
290 	case EMULATE_DONE:
291 		/* Future optimization: only reload non-volatiles if they were
292 		 * actually modified. */
293 		r = RESUME_GUEST_NV;
294 		break;
295 	case EMULATE_AGAIN:
296 		r = RESUME_GUEST;
297 		break;
298 	case EMULATE_DO_MMIO:
299 		vcpu->run->exit_reason = KVM_EXIT_MMIO;
300 		/* We must reload nonvolatiles because "update" load/store
301 		 * instructions modify register state. */
302 		/* Future optimization: only reload non-volatiles if they were
303 		 * actually modified. */
304 		r = RESUME_HOST_NV;
305 		break;
306 	case EMULATE_FAIL:
307 	{
308 		u32 last_inst;
309 
310 		kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
311 		kvm_debug_ratelimited("Guest access to device memory using unsupported instruction (opcode: %#08x)\n",
312 				      last_inst);
313 
314 		/*
315 		 * Injecting a Data Storage here is a bit more
316 		 * accurate since the instruction that caused the
317 		 * access could still be a valid one.
318 		 */
319 		if (!IS_ENABLED(CONFIG_BOOKE)) {
320 			ulong dsisr = DSISR_BADACCESS;
321 
322 			if (vcpu->mmio_is_write)
323 				dsisr |= DSISR_ISSTORE;
324 
325 			kvmppc_core_queue_data_storage(vcpu, vcpu->arch.vaddr_accessed, dsisr);
326 		} else {
327 			/*
328 			 * BookE does not send a SIGBUS on a bad
329 			 * fault, so use a Program interrupt instead
330 			 * to avoid a fault loop.
331 			 */
332 			kvmppc_core_queue_program(vcpu, 0);
333 		}
334 
335 		r = RESUME_GUEST;
336 		break;
337 	}
338 	default:
339 		WARN_ON(1);
340 		r = RESUME_GUEST;
341 	}
342 
343 	return r;
344 }
345 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
346 
347 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
348 	      bool data)
349 {
350 	ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
351 	struct kvmppc_pte pte;
352 	int r = -EINVAL;
353 
354 	vcpu->stat.st++;
355 
356 	if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->store_to_eaddr)
357 		r = vcpu->kvm->arch.kvm_ops->store_to_eaddr(vcpu, eaddr, ptr,
358 							    size);
359 
360 	if ((!r) || (r == -EAGAIN))
361 		return r;
362 
363 	r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
364 			 XLATE_WRITE, &pte);
365 	if (r < 0)
366 		return r;
367 
368 	*eaddr = pte.raddr;
369 
370 	if (!pte.may_write)
371 		return -EPERM;
372 
373 	/* Magic page override */
374 	if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
375 	    ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
376 	    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
377 		void *magic = vcpu->arch.shared;
378 		magic += pte.eaddr & 0xfff;
379 		memcpy(magic, ptr, size);
380 		return EMULATE_DONE;
381 	}
382 
383 	if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
384 		return EMULATE_DO_MMIO;
385 
386 	return EMULATE_DONE;
387 }
388 EXPORT_SYMBOL_GPL(kvmppc_st);
389 
390 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
391 		      bool data)
392 {
393 	ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
394 	struct kvmppc_pte pte;
395 	int rc = -EINVAL;
396 
397 	vcpu->stat.ld++;
398 
399 	if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->load_from_eaddr)
400 		rc = vcpu->kvm->arch.kvm_ops->load_from_eaddr(vcpu, eaddr, ptr,
401 							      size);
402 
403 	if ((!rc) || (rc == -EAGAIN))
404 		return rc;
405 
406 	rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
407 			  XLATE_READ, &pte);
408 	if (rc)
409 		return rc;
410 
411 	*eaddr = pte.raddr;
412 
413 	if (!pte.may_read)
414 		return -EPERM;
415 
416 	if (!data && !pte.may_execute)
417 		return -ENOEXEC;
418 
419 	/* Magic page override */
420 	if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
421 	    ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
422 	    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
423 		void *magic = vcpu->arch.shared;
424 		magic += pte.eaddr & 0xfff;
425 		memcpy(ptr, magic, size);
426 		return EMULATE_DONE;
427 	}
428 
429 	kvm_vcpu_srcu_read_lock(vcpu);
430 	rc = kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size);
431 	kvm_vcpu_srcu_read_unlock(vcpu);
432 	if (rc)
433 		return EMULATE_DO_MMIO;
434 
435 	return EMULATE_DONE;
436 }
437 EXPORT_SYMBOL_GPL(kvmppc_ld);
438 
439 int kvm_arch_hardware_enable(void)
440 {
441 	return 0;
442 }
443 
444 int kvm_arch_hardware_setup(void *opaque)
445 {
446 	return 0;
447 }
448 
449 int kvm_arch_check_processor_compat(void *opaque)
450 {
451 	return kvmppc_core_check_processor_compat();
452 }
453 
454 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
455 {
456 	struct kvmppc_ops *kvm_ops = NULL;
457 	int r;
458 
459 	/*
460 	 * if we have both HV and PR enabled, default is HV
461 	 */
462 	if (type == 0) {
463 		if (kvmppc_hv_ops)
464 			kvm_ops = kvmppc_hv_ops;
465 		else
466 			kvm_ops = kvmppc_pr_ops;
467 		if (!kvm_ops)
468 			goto err_out;
469 	} else	if (type == KVM_VM_PPC_HV) {
470 		if (!kvmppc_hv_ops)
471 			goto err_out;
472 		kvm_ops = kvmppc_hv_ops;
473 	} else if (type == KVM_VM_PPC_PR) {
474 		if (!kvmppc_pr_ops)
475 			goto err_out;
476 		kvm_ops = kvmppc_pr_ops;
477 	} else
478 		goto err_out;
479 
480 	if (!try_module_get(kvm_ops->owner))
481 		return -ENOENT;
482 
483 	kvm->arch.kvm_ops = kvm_ops;
484 	r = kvmppc_core_init_vm(kvm);
485 	if (r)
486 		module_put(kvm_ops->owner);
487 	return r;
488 err_out:
489 	return -EINVAL;
490 }
491 
492 void kvm_arch_destroy_vm(struct kvm *kvm)
493 {
494 #ifdef CONFIG_KVM_XICS
495 	/*
496 	 * We call kick_all_cpus_sync() to ensure that all
497 	 * CPUs have executed any pending IPIs before we
498 	 * continue and free VCPUs structures below.
499 	 */
500 	if (is_kvmppc_hv_enabled(kvm))
501 		kick_all_cpus_sync();
502 #endif
503 
504 	kvm_destroy_vcpus(kvm);
505 
506 	mutex_lock(&kvm->lock);
507 
508 	kvmppc_core_destroy_vm(kvm);
509 
510 	mutex_unlock(&kvm->lock);
511 
512 	/* drop the module reference */
513 	module_put(kvm->arch.kvm_ops->owner);
514 }
515 
516 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
517 {
518 	int r;
519 	/* Assume we're using HV mode when the HV module is loaded */
520 	int hv_enabled = kvmppc_hv_ops ? 1 : 0;
521 
522 	if (kvm) {
523 		/*
524 		 * Hooray - we know which VM type we're running on. Depend on
525 		 * that rather than the guess above.
526 		 */
527 		hv_enabled = is_kvmppc_hv_enabled(kvm);
528 	}
529 
530 	switch (ext) {
531 #ifdef CONFIG_BOOKE
532 	case KVM_CAP_PPC_BOOKE_SREGS:
533 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
534 	case KVM_CAP_PPC_EPR:
535 #else
536 	case KVM_CAP_PPC_SEGSTATE:
537 	case KVM_CAP_PPC_HIOR:
538 	case KVM_CAP_PPC_PAPR:
539 #endif
540 	case KVM_CAP_PPC_UNSET_IRQ:
541 	case KVM_CAP_PPC_IRQ_LEVEL:
542 	case KVM_CAP_ENABLE_CAP:
543 	case KVM_CAP_ONE_REG:
544 	case KVM_CAP_IOEVENTFD:
545 	case KVM_CAP_DEVICE_CTRL:
546 	case KVM_CAP_IMMEDIATE_EXIT:
547 	case KVM_CAP_SET_GUEST_DEBUG:
548 		r = 1;
549 		break;
550 	case KVM_CAP_PPC_GUEST_DEBUG_SSTEP:
551 	case KVM_CAP_PPC_PAIRED_SINGLES:
552 	case KVM_CAP_PPC_OSI:
553 	case KVM_CAP_PPC_GET_PVINFO:
554 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
555 	case KVM_CAP_SW_TLB:
556 #endif
557 		/* We support this only for PR */
558 		r = !hv_enabled;
559 		break;
560 #ifdef CONFIG_KVM_MPIC
561 	case KVM_CAP_IRQ_MPIC:
562 		r = 1;
563 		break;
564 #endif
565 
566 #ifdef CONFIG_PPC_BOOK3S_64
567 	case KVM_CAP_SPAPR_TCE:
568 	case KVM_CAP_SPAPR_TCE_64:
569 		r = 1;
570 		break;
571 	case KVM_CAP_SPAPR_TCE_VFIO:
572 		r = !!cpu_has_feature(CPU_FTR_HVMODE);
573 		break;
574 	case KVM_CAP_PPC_RTAS:
575 	case KVM_CAP_PPC_FIXUP_HCALL:
576 	case KVM_CAP_PPC_ENABLE_HCALL:
577 #ifdef CONFIG_KVM_XICS
578 	case KVM_CAP_IRQ_XICS:
579 #endif
580 	case KVM_CAP_PPC_GET_CPU_CHAR:
581 		r = 1;
582 		break;
583 #ifdef CONFIG_KVM_XIVE
584 	case KVM_CAP_PPC_IRQ_XIVE:
585 		/*
586 		 * We need XIVE to be enabled on the platform (implies
587 		 * a POWER9 processor) and the PowerNV platform, as
588 		 * nested is not yet supported.
589 		 */
590 		r = xive_enabled() && !!cpu_has_feature(CPU_FTR_HVMODE) &&
591 			kvmppc_xive_native_supported();
592 		break;
593 #endif
594 
595 	case KVM_CAP_PPC_ALLOC_HTAB:
596 		r = hv_enabled;
597 		break;
598 #endif /* CONFIG_PPC_BOOK3S_64 */
599 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
600 	case KVM_CAP_PPC_SMT:
601 		r = 0;
602 		if (kvm) {
603 			if (kvm->arch.emul_smt_mode > 1)
604 				r = kvm->arch.emul_smt_mode;
605 			else
606 				r = kvm->arch.smt_mode;
607 		} else if (hv_enabled) {
608 			if (cpu_has_feature(CPU_FTR_ARCH_300))
609 				r = 1;
610 			else
611 				r = threads_per_subcore;
612 		}
613 		break;
614 	case KVM_CAP_PPC_SMT_POSSIBLE:
615 		r = 1;
616 		if (hv_enabled) {
617 			if (!cpu_has_feature(CPU_FTR_ARCH_300))
618 				r = ((threads_per_subcore << 1) - 1);
619 			else
620 				/* P9 can emulate dbells, so allow any mode */
621 				r = 8 | 4 | 2 | 1;
622 		}
623 		break;
624 	case KVM_CAP_PPC_RMA:
625 		r = 0;
626 		break;
627 	case KVM_CAP_PPC_HWRNG:
628 		r = kvmppc_hwrng_present();
629 		break;
630 	case KVM_CAP_PPC_MMU_RADIX:
631 		r = !!(hv_enabled && radix_enabled());
632 		break;
633 	case KVM_CAP_PPC_MMU_HASH_V3:
634 		r = !!(hv_enabled && kvmppc_hv_ops->hash_v3_possible &&
635 		       kvmppc_hv_ops->hash_v3_possible());
636 		break;
637 	case KVM_CAP_PPC_NESTED_HV:
638 		r = !!(hv_enabled && kvmppc_hv_ops->enable_nested &&
639 		       !kvmppc_hv_ops->enable_nested(NULL));
640 		break;
641 #endif
642 	case KVM_CAP_SYNC_MMU:
643 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
644 		r = hv_enabled;
645 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
646 		r = 1;
647 #else
648 		r = 0;
649 #endif
650 		break;
651 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
652 	case KVM_CAP_PPC_HTAB_FD:
653 		r = hv_enabled;
654 		break;
655 #endif
656 	case KVM_CAP_NR_VCPUS:
657 		/*
658 		 * Recommending a number of CPUs is somewhat arbitrary; we
659 		 * return the number of present CPUs for -HV (since a host
660 		 * will have secondary threads "offline"), and for other KVM
661 		 * implementations just count online CPUs.
662 		 */
663 		if (hv_enabled)
664 			r = min_t(unsigned int, num_present_cpus(), KVM_MAX_VCPUS);
665 		else
666 			r = min_t(unsigned int, num_online_cpus(), KVM_MAX_VCPUS);
667 		break;
668 	case KVM_CAP_MAX_VCPUS:
669 		r = KVM_MAX_VCPUS;
670 		break;
671 	case KVM_CAP_MAX_VCPU_ID:
672 		r = KVM_MAX_VCPU_IDS;
673 		break;
674 #ifdef CONFIG_PPC_BOOK3S_64
675 	case KVM_CAP_PPC_GET_SMMU_INFO:
676 		r = 1;
677 		break;
678 	case KVM_CAP_SPAPR_MULTITCE:
679 		r = 1;
680 		break;
681 	case KVM_CAP_SPAPR_RESIZE_HPT:
682 		r = !!hv_enabled;
683 		break;
684 #endif
685 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
686 	case KVM_CAP_PPC_FWNMI:
687 		r = hv_enabled;
688 		break;
689 #endif
690 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
691 	case KVM_CAP_PPC_HTM:
692 		r = !!(cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM) ||
693 		     (hv_enabled && cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST));
694 		break;
695 #endif
696 #if defined(CONFIG_KVM_BOOK3S_HV_POSSIBLE)
697 	case KVM_CAP_PPC_SECURE_GUEST:
698 		r = hv_enabled && kvmppc_hv_ops->enable_svm &&
699 			!kvmppc_hv_ops->enable_svm(NULL);
700 		break;
701 	case KVM_CAP_PPC_DAWR1:
702 		r = !!(hv_enabled && kvmppc_hv_ops->enable_dawr1 &&
703 		       !kvmppc_hv_ops->enable_dawr1(NULL));
704 		break;
705 	case KVM_CAP_PPC_RPT_INVALIDATE:
706 		r = 1;
707 		break;
708 #endif
709 	case KVM_CAP_PPC_AIL_MODE_3:
710 		r = 0;
711 		/*
712 		 * KVM PR, POWER7, and some POWER9s don't support AIL=3 mode.
713 		 * The POWER9s can support it if the guest runs in hash mode,
714 		 * but QEMU doesn't necessarily query the capability in time.
715 		 */
716 		if (hv_enabled) {
717 			if (kvmhv_on_pseries()) {
718 				if (pseries_reloc_on_exception())
719 					r = 1;
720 			} else if (cpu_has_feature(CPU_FTR_ARCH_207S) &&
721 				  !cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG)) {
722 				r = 1;
723 			}
724 		}
725 		break;
726 	default:
727 		r = 0;
728 		break;
729 	}
730 	return r;
731 
732 }
733 
734 long kvm_arch_dev_ioctl(struct file *filp,
735                         unsigned int ioctl, unsigned long arg)
736 {
737 	return -EINVAL;
738 }
739 
740 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
741 {
742 	kvmppc_core_free_memslot(kvm, slot);
743 }
744 
745 int kvm_arch_prepare_memory_region(struct kvm *kvm,
746 				   const struct kvm_memory_slot *old,
747 				   struct kvm_memory_slot *new,
748 				   enum kvm_mr_change change)
749 {
750 	return kvmppc_core_prepare_memory_region(kvm, old, new, change);
751 }
752 
753 void kvm_arch_commit_memory_region(struct kvm *kvm,
754 				   struct kvm_memory_slot *old,
755 				   const struct kvm_memory_slot *new,
756 				   enum kvm_mr_change change)
757 {
758 	kvmppc_core_commit_memory_region(kvm, old, new, change);
759 }
760 
761 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
762 				   struct kvm_memory_slot *slot)
763 {
764 	kvmppc_core_flush_memslot(kvm, slot);
765 }
766 
767 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
768 {
769 	return 0;
770 }
771 
772 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
773 {
774 	struct kvm_vcpu *vcpu;
775 
776 	vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
777 	kvmppc_decrementer_func(vcpu);
778 
779 	return HRTIMER_NORESTART;
780 }
781 
782 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
783 {
784 	int err;
785 
786 	hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
787 	vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
788 	vcpu->arch.dec_expires = get_tb();
789 
790 #ifdef CONFIG_KVM_EXIT_TIMING
791 	mutex_init(&vcpu->arch.exit_timing_lock);
792 #endif
793 	err = kvmppc_subarch_vcpu_init(vcpu);
794 	if (err)
795 		return err;
796 
797 	err = kvmppc_core_vcpu_create(vcpu);
798 	if (err)
799 		goto out_vcpu_uninit;
800 
801 	rcuwait_init(&vcpu->arch.wait);
802 	vcpu->arch.waitp = &vcpu->arch.wait;
803 	return 0;
804 
805 out_vcpu_uninit:
806 	kvmppc_subarch_vcpu_uninit(vcpu);
807 	return err;
808 }
809 
810 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
811 {
812 }
813 
814 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
815 {
816 	/* Make sure we're not using the vcpu anymore */
817 	hrtimer_cancel(&vcpu->arch.dec_timer);
818 
819 	switch (vcpu->arch.irq_type) {
820 	case KVMPPC_IRQ_MPIC:
821 		kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
822 		break;
823 	case KVMPPC_IRQ_XICS:
824 		if (xics_on_xive())
825 			kvmppc_xive_cleanup_vcpu(vcpu);
826 		else
827 			kvmppc_xics_free_icp(vcpu);
828 		break;
829 	case KVMPPC_IRQ_XIVE:
830 		kvmppc_xive_native_cleanup_vcpu(vcpu);
831 		break;
832 	}
833 
834 	kvmppc_core_vcpu_free(vcpu);
835 
836 	kvmppc_subarch_vcpu_uninit(vcpu);
837 }
838 
839 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
840 {
841 	return kvmppc_core_pending_dec(vcpu);
842 }
843 
844 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
845 {
846 #ifdef CONFIG_BOOKE
847 	/*
848 	 * vrsave (formerly usprg0) isn't used by Linux, but may
849 	 * be used by the guest.
850 	 *
851 	 * On non-booke this is associated with Altivec and
852 	 * is handled by code in book3s.c.
853 	 */
854 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
855 #endif
856 	kvmppc_core_vcpu_load(vcpu, cpu);
857 }
858 
859 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
860 {
861 	kvmppc_core_vcpu_put(vcpu);
862 #ifdef CONFIG_BOOKE
863 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
864 #endif
865 }
866 
867 /*
868  * irq_bypass_add_producer and irq_bypass_del_producer are only
869  * useful if the architecture supports PCI passthrough.
870  * irq_bypass_stop and irq_bypass_start are not needed and so
871  * kvm_ops are not defined for them.
872  */
873 bool kvm_arch_has_irq_bypass(void)
874 {
875 	return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
876 		(kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
877 }
878 
879 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
880 				     struct irq_bypass_producer *prod)
881 {
882 	struct kvm_kernel_irqfd *irqfd =
883 		container_of(cons, struct kvm_kernel_irqfd, consumer);
884 	struct kvm *kvm = irqfd->kvm;
885 
886 	if (kvm->arch.kvm_ops->irq_bypass_add_producer)
887 		return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
888 
889 	return 0;
890 }
891 
892 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
893 				      struct irq_bypass_producer *prod)
894 {
895 	struct kvm_kernel_irqfd *irqfd =
896 		container_of(cons, struct kvm_kernel_irqfd, consumer);
897 	struct kvm *kvm = irqfd->kvm;
898 
899 	if (kvm->arch.kvm_ops->irq_bypass_del_producer)
900 		kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
901 }
902 
903 #ifdef CONFIG_VSX
904 static inline int kvmppc_get_vsr_dword_offset(int index)
905 {
906 	int offset;
907 
908 	if ((index != 0) && (index != 1))
909 		return -1;
910 
911 #ifdef __BIG_ENDIAN
912 	offset =  index;
913 #else
914 	offset = 1 - index;
915 #endif
916 
917 	return offset;
918 }
919 
920 static inline int kvmppc_get_vsr_word_offset(int index)
921 {
922 	int offset;
923 
924 	if ((index > 3) || (index < 0))
925 		return -1;
926 
927 #ifdef __BIG_ENDIAN
928 	offset = index;
929 #else
930 	offset = 3 - index;
931 #endif
932 	return offset;
933 }
934 
935 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
936 	u64 gpr)
937 {
938 	union kvmppc_one_reg val;
939 	int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
940 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
941 
942 	if (offset == -1)
943 		return;
944 
945 	if (index >= 32) {
946 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
947 		val.vsxval[offset] = gpr;
948 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
949 	} else {
950 		VCPU_VSX_FPR(vcpu, index, offset) = gpr;
951 	}
952 }
953 
954 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
955 	u64 gpr)
956 {
957 	union kvmppc_one_reg val;
958 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
959 
960 	if (index >= 32) {
961 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
962 		val.vsxval[0] = gpr;
963 		val.vsxval[1] = gpr;
964 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
965 	} else {
966 		VCPU_VSX_FPR(vcpu, index, 0) = gpr;
967 		VCPU_VSX_FPR(vcpu, index, 1) = gpr;
968 	}
969 }
970 
971 static inline void kvmppc_set_vsr_word_dump(struct kvm_vcpu *vcpu,
972 	u32 gpr)
973 {
974 	union kvmppc_one_reg val;
975 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
976 
977 	if (index >= 32) {
978 		val.vsx32val[0] = gpr;
979 		val.vsx32val[1] = gpr;
980 		val.vsx32val[2] = gpr;
981 		val.vsx32val[3] = gpr;
982 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
983 	} else {
984 		val.vsx32val[0] = gpr;
985 		val.vsx32val[1] = gpr;
986 		VCPU_VSX_FPR(vcpu, index, 0) = val.vsxval[0];
987 		VCPU_VSX_FPR(vcpu, index, 1) = val.vsxval[0];
988 	}
989 }
990 
991 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
992 	u32 gpr32)
993 {
994 	union kvmppc_one_reg val;
995 	int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
996 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
997 	int dword_offset, word_offset;
998 
999 	if (offset == -1)
1000 		return;
1001 
1002 	if (index >= 32) {
1003 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
1004 		val.vsx32val[offset] = gpr32;
1005 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
1006 	} else {
1007 		dword_offset = offset / 2;
1008 		word_offset = offset % 2;
1009 		val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
1010 		val.vsx32val[word_offset] = gpr32;
1011 		VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
1012 	}
1013 }
1014 #endif /* CONFIG_VSX */
1015 
1016 #ifdef CONFIG_ALTIVEC
1017 static inline int kvmppc_get_vmx_offset_generic(struct kvm_vcpu *vcpu,
1018 		int index, int element_size)
1019 {
1020 	int offset;
1021 	int elts = sizeof(vector128)/element_size;
1022 
1023 	if ((index < 0) || (index >= elts))
1024 		return -1;
1025 
1026 	if (kvmppc_need_byteswap(vcpu))
1027 		offset = elts - index - 1;
1028 	else
1029 		offset = index;
1030 
1031 	return offset;
1032 }
1033 
1034 static inline int kvmppc_get_vmx_dword_offset(struct kvm_vcpu *vcpu,
1035 		int index)
1036 {
1037 	return kvmppc_get_vmx_offset_generic(vcpu, index, 8);
1038 }
1039 
1040 static inline int kvmppc_get_vmx_word_offset(struct kvm_vcpu *vcpu,
1041 		int index)
1042 {
1043 	return kvmppc_get_vmx_offset_generic(vcpu, index, 4);
1044 }
1045 
1046 static inline int kvmppc_get_vmx_hword_offset(struct kvm_vcpu *vcpu,
1047 		int index)
1048 {
1049 	return kvmppc_get_vmx_offset_generic(vcpu, index, 2);
1050 }
1051 
1052 static inline int kvmppc_get_vmx_byte_offset(struct kvm_vcpu *vcpu,
1053 		int index)
1054 {
1055 	return kvmppc_get_vmx_offset_generic(vcpu, index, 1);
1056 }
1057 
1058 
1059 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu,
1060 	u64 gpr)
1061 {
1062 	union kvmppc_one_reg val;
1063 	int offset = kvmppc_get_vmx_dword_offset(vcpu,
1064 			vcpu->arch.mmio_vmx_offset);
1065 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1066 
1067 	if (offset == -1)
1068 		return;
1069 
1070 	val.vval = VCPU_VSX_VR(vcpu, index);
1071 	val.vsxval[offset] = gpr;
1072 	VCPU_VSX_VR(vcpu, index) = val.vval;
1073 }
1074 
1075 static inline void kvmppc_set_vmx_word(struct kvm_vcpu *vcpu,
1076 	u32 gpr32)
1077 {
1078 	union kvmppc_one_reg val;
1079 	int offset = kvmppc_get_vmx_word_offset(vcpu,
1080 			vcpu->arch.mmio_vmx_offset);
1081 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1082 
1083 	if (offset == -1)
1084 		return;
1085 
1086 	val.vval = VCPU_VSX_VR(vcpu, index);
1087 	val.vsx32val[offset] = gpr32;
1088 	VCPU_VSX_VR(vcpu, index) = val.vval;
1089 }
1090 
1091 static inline void kvmppc_set_vmx_hword(struct kvm_vcpu *vcpu,
1092 	u16 gpr16)
1093 {
1094 	union kvmppc_one_reg val;
1095 	int offset = kvmppc_get_vmx_hword_offset(vcpu,
1096 			vcpu->arch.mmio_vmx_offset);
1097 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1098 
1099 	if (offset == -1)
1100 		return;
1101 
1102 	val.vval = VCPU_VSX_VR(vcpu, index);
1103 	val.vsx16val[offset] = gpr16;
1104 	VCPU_VSX_VR(vcpu, index) = val.vval;
1105 }
1106 
1107 static inline void kvmppc_set_vmx_byte(struct kvm_vcpu *vcpu,
1108 	u8 gpr8)
1109 {
1110 	union kvmppc_one_reg val;
1111 	int offset = kvmppc_get_vmx_byte_offset(vcpu,
1112 			vcpu->arch.mmio_vmx_offset);
1113 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1114 
1115 	if (offset == -1)
1116 		return;
1117 
1118 	val.vval = VCPU_VSX_VR(vcpu, index);
1119 	val.vsx8val[offset] = gpr8;
1120 	VCPU_VSX_VR(vcpu, index) = val.vval;
1121 }
1122 #endif /* CONFIG_ALTIVEC */
1123 
1124 #ifdef CONFIG_PPC_FPU
1125 static inline u64 sp_to_dp(u32 fprs)
1126 {
1127 	u64 fprd;
1128 
1129 	preempt_disable();
1130 	enable_kernel_fp();
1131 	asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m<>" (fprd) : "m<>" (fprs)
1132 	     : "fr0");
1133 	preempt_enable();
1134 	return fprd;
1135 }
1136 
1137 static inline u32 dp_to_sp(u64 fprd)
1138 {
1139 	u32 fprs;
1140 
1141 	preempt_disable();
1142 	enable_kernel_fp();
1143 	asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m<>" (fprs) : "m<>" (fprd)
1144 	     : "fr0");
1145 	preempt_enable();
1146 	return fprs;
1147 }
1148 
1149 #else
1150 #define sp_to_dp(x)	(x)
1151 #define dp_to_sp(x)	(x)
1152 #endif /* CONFIG_PPC_FPU */
1153 
1154 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu)
1155 {
1156 	struct kvm_run *run = vcpu->run;
1157 	u64 gpr;
1158 
1159 	if (run->mmio.len > sizeof(gpr))
1160 		return;
1161 
1162 	if (!vcpu->arch.mmio_host_swabbed) {
1163 		switch (run->mmio.len) {
1164 		case 8: gpr = *(u64 *)run->mmio.data; break;
1165 		case 4: gpr = *(u32 *)run->mmio.data; break;
1166 		case 2: gpr = *(u16 *)run->mmio.data; break;
1167 		case 1: gpr = *(u8 *)run->mmio.data; break;
1168 		}
1169 	} else {
1170 		switch (run->mmio.len) {
1171 		case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
1172 		case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
1173 		case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
1174 		case 1: gpr = *(u8 *)run->mmio.data; break;
1175 		}
1176 	}
1177 
1178 	/* conversion between single and double precision */
1179 	if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
1180 		gpr = sp_to_dp(gpr);
1181 
1182 	if (vcpu->arch.mmio_sign_extend) {
1183 		switch (run->mmio.len) {
1184 #ifdef CONFIG_PPC64
1185 		case 4:
1186 			gpr = (s64)(s32)gpr;
1187 			break;
1188 #endif
1189 		case 2:
1190 			gpr = (s64)(s16)gpr;
1191 			break;
1192 		case 1:
1193 			gpr = (s64)(s8)gpr;
1194 			break;
1195 		}
1196 	}
1197 
1198 	switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1199 	case KVM_MMIO_REG_GPR:
1200 		kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1201 		break;
1202 	case KVM_MMIO_REG_FPR:
1203 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1204 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_FP);
1205 
1206 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1207 		break;
1208 #ifdef CONFIG_PPC_BOOK3S
1209 	case KVM_MMIO_REG_QPR:
1210 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1211 		break;
1212 	case KVM_MMIO_REG_FQPR:
1213 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1214 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1215 		break;
1216 #endif
1217 #ifdef CONFIG_VSX
1218 	case KVM_MMIO_REG_VSX:
1219 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1220 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VSX);
1221 
1222 		if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_DWORD)
1223 			kvmppc_set_vsr_dword(vcpu, gpr);
1224 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_WORD)
1225 			kvmppc_set_vsr_word(vcpu, gpr);
1226 		else if (vcpu->arch.mmio_copy_type ==
1227 				KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1228 			kvmppc_set_vsr_dword_dump(vcpu, gpr);
1229 		else if (vcpu->arch.mmio_copy_type ==
1230 				KVMPPC_VSX_COPY_WORD_LOAD_DUMP)
1231 			kvmppc_set_vsr_word_dump(vcpu, gpr);
1232 		break;
1233 #endif
1234 #ifdef CONFIG_ALTIVEC
1235 	case KVM_MMIO_REG_VMX:
1236 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1237 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VEC);
1238 
1239 		if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_DWORD)
1240 			kvmppc_set_vmx_dword(vcpu, gpr);
1241 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_WORD)
1242 			kvmppc_set_vmx_word(vcpu, gpr);
1243 		else if (vcpu->arch.mmio_copy_type ==
1244 				KVMPPC_VMX_COPY_HWORD)
1245 			kvmppc_set_vmx_hword(vcpu, gpr);
1246 		else if (vcpu->arch.mmio_copy_type ==
1247 				KVMPPC_VMX_COPY_BYTE)
1248 			kvmppc_set_vmx_byte(vcpu, gpr);
1249 		break;
1250 #endif
1251 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1252 	case KVM_MMIO_REG_NESTED_GPR:
1253 		if (kvmppc_need_byteswap(vcpu))
1254 			gpr = swab64(gpr);
1255 		kvm_vcpu_write_guest(vcpu, vcpu->arch.nested_io_gpr, &gpr,
1256 				     sizeof(gpr));
1257 		break;
1258 #endif
1259 	default:
1260 		BUG();
1261 	}
1262 }
1263 
1264 static int __kvmppc_handle_load(struct kvm_vcpu *vcpu,
1265 				unsigned int rt, unsigned int bytes,
1266 				int is_default_endian, int sign_extend)
1267 {
1268 	struct kvm_run *run = vcpu->run;
1269 	int idx, ret;
1270 	bool host_swabbed;
1271 
1272 	/* Pity C doesn't have a logical XOR operator */
1273 	if (kvmppc_need_byteswap(vcpu)) {
1274 		host_swabbed = is_default_endian;
1275 	} else {
1276 		host_swabbed = !is_default_endian;
1277 	}
1278 
1279 	if (bytes > sizeof(run->mmio.data))
1280 		return EMULATE_FAIL;
1281 
1282 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1283 	run->mmio.len = bytes;
1284 	run->mmio.is_write = 0;
1285 
1286 	vcpu->arch.io_gpr = rt;
1287 	vcpu->arch.mmio_host_swabbed = host_swabbed;
1288 	vcpu->mmio_needed = 1;
1289 	vcpu->mmio_is_write = 0;
1290 	vcpu->arch.mmio_sign_extend = sign_extend;
1291 
1292 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1293 
1294 	ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1295 			      bytes, &run->mmio.data);
1296 
1297 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1298 
1299 	if (!ret) {
1300 		kvmppc_complete_mmio_load(vcpu);
1301 		vcpu->mmio_needed = 0;
1302 		return EMULATE_DONE;
1303 	}
1304 
1305 	return EMULATE_DO_MMIO;
1306 }
1307 
1308 int kvmppc_handle_load(struct kvm_vcpu *vcpu,
1309 		       unsigned int rt, unsigned int bytes,
1310 		       int is_default_endian)
1311 {
1312 	return __kvmppc_handle_load(vcpu, rt, bytes, is_default_endian, 0);
1313 }
1314 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1315 
1316 /* Same as above, but sign extends */
1317 int kvmppc_handle_loads(struct kvm_vcpu *vcpu,
1318 			unsigned int rt, unsigned int bytes,
1319 			int is_default_endian)
1320 {
1321 	return __kvmppc_handle_load(vcpu, rt, bytes, is_default_endian, 1);
1322 }
1323 
1324 #ifdef CONFIG_VSX
1325 int kvmppc_handle_vsx_load(struct kvm_vcpu *vcpu,
1326 			unsigned int rt, unsigned int bytes,
1327 			int is_default_endian, int mmio_sign_extend)
1328 {
1329 	enum emulation_result emulated = EMULATE_DONE;
1330 
1331 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1332 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1333 		return EMULATE_FAIL;
1334 
1335 	while (vcpu->arch.mmio_vsx_copy_nums) {
1336 		emulated = __kvmppc_handle_load(vcpu, rt, bytes,
1337 			is_default_endian, mmio_sign_extend);
1338 
1339 		if (emulated != EMULATE_DONE)
1340 			break;
1341 
1342 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1343 
1344 		vcpu->arch.mmio_vsx_copy_nums--;
1345 		vcpu->arch.mmio_vsx_offset++;
1346 	}
1347 	return emulated;
1348 }
1349 #endif /* CONFIG_VSX */
1350 
1351 int kvmppc_handle_store(struct kvm_vcpu *vcpu,
1352 			u64 val, unsigned int bytes, int is_default_endian)
1353 {
1354 	struct kvm_run *run = vcpu->run;
1355 	void *data = run->mmio.data;
1356 	int idx, ret;
1357 	bool host_swabbed;
1358 
1359 	/* Pity C doesn't have a logical XOR operator */
1360 	if (kvmppc_need_byteswap(vcpu)) {
1361 		host_swabbed = is_default_endian;
1362 	} else {
1363 		host_swabbed = !is_default_endian;
1364 	}
1365 
1366 	if (bytes > sizeof(run->mmio.data))
1367 		return EMULATE_FAIL;
1368 
1369 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1370 	run->mmio.len = bytes;
1371 	run->mmio.is_write = 1;
1372 	vcpu->mmio_needed = 1;
1373 	vcpu->mmio_is_write = 1;
1374 
1375 	if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1376 		val = dp_to_sp(val);
1377 
1378 	/* Store the value at the lowest bytes in 'data'. */
1379 	if (!host_swabbed) {
1380 		switch (bytes) {
1381 		case 8: *(u64 *)data = val; break;
1382 		case 4: *(u32 *)data = val; break;
1383 		case 2: *(u16 *)data = val; break;
1384 		case 1: *(u8  *)data = val; break;
1385 		}
1386 	} else {
1387 		switch (bytes) {
1388 		case 8: *(u64 *)data = swab64(val); break;
1389 		case 4: *(u32 *)data = swab32(val); break;
1390 		case 2: *(u16 *)data = swab16(val); break;
1391 		case 1: *(u8  *)data = val; break;
1392 		}
1393 	}
1394 
1395 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1396 
1397 	ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1398 			       bytes, &run->mmio.data);
1399 
1400 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1401 
1402 	if (!ret) {
1403 		vcpu->mmio_needed = 0;
1404 		return EMULATE_DONE;
1405 	}
1406 
1407 	return EMULATE_DO_MMIO;
1408 }
1409 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1410 
1411 #ifdef CONFIG_VSX
1412 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1413 {
1414 	u32 dword_offset, word_offset;
1415 	union kvmppc_one_reg reg;
1416 	int vsx_offset = 0;
1417 	int copy_type = vcpu->arch.mmio_copy_type;
1418 	int result = 0;
1419 
1420 	switch (copy_type) {
1421 	case KVMPPC_VSX_COPY_DWORD:
1422 		vsx_offset =
1423 			kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1424 
1425 		if (vsx_offset == -1) {
1426 			result = -1;
1427 			break;
1428 		}
1429 
1430 		if (rs < 32) {
1431 			*val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1432 		} else {
1433 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1434 			*val = reg.vsxval[vsx_offset];
1435 		}
1436 		break;
1437 
1438 	case KVMPPC_VSX_COPY_WORD:
1439 		vsx_offset =
1440 			kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1441 
1442 		if (vsx_offset == -1) {
1443 			result = -1;
1444 			break;
1445 		}
1446 
1447 		if (rs < 32) {
1448 			dword_offset = vsx_offset / 2;
1449 			word_offset = vsx_offset % 2;
1450 			reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1451 			*val = reg.vsx32val[word_offset];
1452 		} else {
1453 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1454 			*val = reg.vsx32val[vsx_offset];
1455 		}
1456 		break;
1457 
1458 	default:
1459 		result = -1;
1460 		break;
1461 	}
1462 
1463 	return result;
1464 }
1465 
1466 int kvmppc_handle_vsx_store(struct kvm_vcpu *vcpu,
1467 			int rs, unsigned int bytes, int is_default_endian)
1468 {
1469 	u64 val;
1470 	enum emulation_result emulated = EMULATE_DONE;
1471 
1472 	vcpu->arch.io_gpr = rs;
1473 
1474 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1475 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1476 		return EMULATE_FAIL;
1477 
1478 	while (vcpu->arch.mmio_vsx_copy_nums) {
1479 		if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1480 			return EMULATE_FAIL;
1481 
1482 		emulated = kvmppc_handle_store(vcpu,
1483 			 val, bytes, is_default_endian);
1484 
1485 		if (emulated != EMULATE_DONE)
1486 			break;
1487 
1488 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1489 
1490 		vcpu->arch.mmio_vsx_copy_nums--;
1491 		vcpu->arch.mmio_vsx_offset++;
1492 	}
1493 
1494 	return emulated;
1495 }
1496 
1497 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu)
1498 {
1499 	struct kvm_run *run = vcpu->run;
1500 	enum emulation_result emulated = EMULATE_FAIL;
1501 	int r;
1502 
1503 	vcpu->arch.paddr_accessed += run->mmio.len;
1504 
1505 	if (!vcpu->mmio_is_write) {
1506 		emulated = kvmppc_handle_vsx_load(vcpu, vcpu->arch.io_gpr,
1507 			 run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1508 	} else {
1509 		emulated = kvmppc_handle_vsx_store(vcpu,
1510 			 vcpu->arch.io_gpr, run->mmio.len, 1);
1511 	}
1512 
1513 	switch (emulated) {
1514 	case EMULATE_DO_MMIO:
1515 		run->exit_reason = KVM_EXIT_MMIO;
1516 		r = RESUME_HOST;
1517 		break;
1518 	case EMULATE_FAIL:
1519 		pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1520 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1521 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1522 		r = RESUME_HOST;
1523 		break;
1524 	default:
1525 		r = RESUME_GUEST;
1526 		break;
1527 	}
1528 	return r;
1529 }
1530 #endif /* CONFIG_VSX */
1531 
1532 #ifdef CONFIG_ALTIVEC
1533 int kvmppc_handle_vmx_load(struct kvm_vcpu *vcpu,
1534 		unsigned int rt, unsigned int bytes, int is_default_endian)
1535 {
1536 	enum emulation_result emulated = EMULATE_DONE;
1537 
1538 	if (vcpu->arch.mmio_vmx_copy_nums > 2)
1539 		return EMULATE_FAIL;
1540 
1541 	while (vcpu->arch.mmio_vmx_copy_nums) {
1542 		emulated = __kvmppc_handle_load(vcpu, rt, bytes,
1543 				is_default_endian, 0);
1544 
1545 		if (emulated != EMULATE_DONE)
1546 			break;
1547 
1548 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1549 		vcpu->arch.mmio_vmx_copy_nums--;
1550 		vcpu->arch.mmio_vmx_offset++;
1551 	}
1552 
1553 	return emulated;
1554 }
1555 
1556 static int kvmppc_get_vmx_dword(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_dword_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.vsxval[vmx_offset];
1570 
1571 	return result;
1572 }
1573 
1574 static int kvmppc_get_vmx_word(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_word_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.vsx32val[vmx_offset];
1588 
1589 	return result;
1590 }
1591 
1592 static int kvmppc_get_vmx_hword(struct kvm_vcpu *vcpu, int index, u64 *val)
1593 {
1594 	union kvmppc_one_reg reg;
1595 	int vmx_offset = 0;
1596 	int result = 0;
1597 
1598 	vmx_offset =
1599 		kvmppc_get_vmx_hword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1600 
1601 	if (vmx_offset == -1)
1602 		return -1;
1603 
1604 	reg.vval = VCPU_VSX_VR(vcpu, index);
1605 	*val = reg.vsx16val[vmx_offset];
1606 
1607 	return result;
1608 }
1609 
1610 static int kvmppc_get_vmx_byte(struct kvm_vcpu *vcpu, int index, u64 *val)
1611 {
1612 	union kvmppc_one_reg reg;
1613 	int vmx_offset = 0;
1614 	int result = 0;
1615 
1616 	vmx_offset =
1617 		kvmppc_get_vmx_byte_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1618 
1619 	if (vmx_offset == -1)
1620 		return -1;
1621 
1622 	reg.vval = VCPU_VSX_VR(vcpu, index);
1623 	*val = reg.vsx8val[vmx_offset];
1624 
1625 	return result;
1626 }
1627 
1628 int kvmppc_handle_vmx_store(struct kvm_vcpu *vcpu,
1629 		unsigned int rs, unsigned int bytes, int is_default_endian)
1630 {
1631 	u64 val = 0;
1632 	unsigned int index = rs & KVM_MMIO_REG_MASK;
1633 	enum emulation_result emulated = EMULATE_DONE;
1634 
1635 	if (vcpu->arch.mmio_vmx_copy_nums > 2)
1636 		return EMULATE_FAIL;
1637 
1638 	vcpu->arch.io_gpr = rs;
1639 
1640 	while (vcpu->arch.mmio_vmx_copy_nums) {
1641 		switch (vcpu->arch.mmio_copy_type) {
1642 		case KVMPPC_VMX_COPY_DWORD:
1643 			if (kvmppc_get_vmx_dword(vcpu, index, &val) == -1)
1644 				return EMULATE_FAIL;
1645 
1646 			break;
1647 		case KVMPPC_VMX_COPY_WORD:
1648 			if (kvmppc_get_vmx_word(vcpu, index, &val) == -1)
1649 				return EMULATE_FAIL;
1650 			break;
1651 		case KVMPPC_VMX_COPY_HWORD:
1652 			if (kvmppc_get_vmx_hword(vcpu, index, &val) == -1)
1653 				return EMULATE_FAIL;
1654 			break;
1655 		case KVMPPC_VMX_COPY_BYTE:
1656 			if (kvmppc_get_vmx_byte(vcpu, index, &val) == -1)
1657 				return EMULATE_FAIL;
1658 			break;
1659 		default:
1660 			return EMULATE_FAIL;
1661 		}
1662 
1663 		emulated = kvmppc_handle_store(vcpu, val, bytes,
1664 				is_default_endian);
1665 		if (emulated != EMULATE_DONE)
1666 			break;
1667 
1668 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1669 		vcpu->arch.mmio_vmx_copy_nums--;
1670 		vcpu->arch.mmio_vmx_offset++;
1671 	}
1672 
1673 	return emulated;
1674 }
1675 
1676 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu)
1677 {
1678 	struct kvm_run *run = vcpu->run;
1679 	enum emulation_result emulated = EMULATE_FAIL;
1680 	int r;
1681 
1682 	vcpu->arch.paddr_accessed += run->mmio.len;
1683 
1684 	if (!vcpu->mmio_is_write) {
1685 		emulated = kvmppc_handle_vmx_load(vcpu,
1686 				vcpu->arch.io_gpr, run->mmio.len, 1);
1687 	} else {
1688 		emulated = kvmppc_handle_vmx_store(vcpu,
1689 				vcpu->arch.io_gpr, run->mmio.len, 1);
1690 	}
1691 
1692 	switch (emulated) {
1693 	case EMULATE_DO_MMIO:
1694 		run->exit_reason = KVM_EXIT_MMIO;
1695 		r = RESUME_HOST;
1696 		break;
1697 	case EMULATE_FAIL:
1698 		pr_info("KVM: MMIO emulation failed (VMX repeat)\n");
1699 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1700 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1701 		r = RESUME_HOST;
1702 		break;
1703 	default:
1704 		r = RESUME_GUEST;
1705 		break;
1706 	}
1707 	return r;
1708 }
1709 #endif /* CONFIG_ALTIVEC */
1710 
1711 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1712 {
1713 	int r = 0;
1714 	union kvmppc_one_reg val;
1715 	int size;
1716 
1717 	size = one_reg_size(reg->id);
1718 	if (size > sizeof(val))
1719 		return -EINVAL;
1720 
1721 	r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1722 	if (r == -EINVAL) {
1723 		r = 0;
1724 		switch (reg->id) {
1725 #ifdef CONFIG_ALTIVEC
1726 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1727 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1728 				r = -ENXIO;
1729 				break;
1730 			}
1731 			val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1732 			break;
1733 		case KVM_REG_PPC_VSCR:
1734 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1735 				r = -ENXIO;
1736 				break;
1737 			}
1738 			val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1739 			break;
1740 		case KVM_REG_PPC_VRSAVE:
1741 			val = get_reg_val(reg->id, vcpu->arch.vrsave);
1742 			break;
1743 #endif /* CONFIG_ALTIVEC */
1744 		default:
1745 			r = -EINVAL;
1746 			break;
1747 		}
1748 	}
1749 
1750 	if (r)
1751 		return r;
1752 
1753 	if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1754 		r = -EFAULT;
1755 
1756 	return r;
1757 }
1758 
1759 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1760 {
1761 	int r;
1762 	union kvmppc_one_reg val;
1763 	int size;
1764 
1765 	size = one_reg_size(reg->id);
1766 	if (size > sizeof(val))
1767 		return -EINVAL;
1768 
1769 	if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1770 		return -EFAULT;
1771 
1772 	r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1773 	if (r == -EINVAL) {
1774 		r = 0;
1775 		switch (reg->id) {
1776 #ifdef CONFIG_ALTIVEC
1777 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1778 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1779 				r = -ENXIO;
1780 				break;
1781 			}
1782 			vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1783 			break;
1784 		case KVM_REG_PPC_VSCR:
1785 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1786 				r = -ENXIO;
1787 				break;
1788 			}
1789 			vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1790 			break;
1791 		case KVM_REG_PPC_VRSAVE:
1792 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1793 				r = -ENXIO;
1794 				break;
1795 			}
1796 			vcpu->arch.vrsave = set_reg_val(reg->id, val);
1797 			break;
1798 #endif /* CONFIG_ALTIVEC */
1799 		default:
1800 			r = -EINVAL;
1801 			break;
1802 		}
1803 	}
1804 
1805 	return r;
1806 }
1807 
1808 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
1809 {
1810 	struct kvm_run *run = vcpu->run;
1811 	int r;
1812 
1813 	vcpu_load(vcpu);
1814 
1815 	if (vcpu->mmio_needed) {
1816 		vcpu->mmio_needed = 0;
1817 		if (!vcpu->mmio_is_write)
1818 			kvmppc_complete_mmio_load(vcpu);
1819 #ifdef CONFIG_VSX
1820 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1821 			vcpu->arch.mmio_vsx_copy_nums--;
1822 			vcpu->arch.mmio_vsx_offset++;
1823 		}
1824 
1825 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1826 			r = kvmppc_emulate_mmio_vsx_loadstore(vcpu);
1827 			if (r == RESUME_HOST) {
1828 				vcpu->mmio_needed = 1;
1829 				goto out;
1830 			}
1831 		}
1832 #endif
1833 #ifdef CONFIG_ALTIVEC
1834 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1835 			vcpu->arch.mmio_vmx_copy_nums--;
1836 			vcpu->arch.mmio_vmx_offset++;
1837 		}
1838 
1839 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1840 			r = kvmppc_emulate_mmio_vmx_loadstore(vcpu);
1841 			if (r == RESUME_HOST) {
1842 				vcpu->mmio_needed = 1;
1843 				goto out;
1844 			}
1845 		}
1846 #endif
1847 	} else if (vcpu->arch.osi_needed) {
1848 		u64 *gprs = run->osi.gprs;
1849 		int i;
1850 
1851 		for (i = 0; i < 32; i++)
1852 			kvmppc_set_gpr(vcpu, i, gprs[i]);
1853 		vcpu->arch.osi_needed = 0;
1854 	} else if (vcpu->arch.hcall_needed) {
1855 		int i;
1856 
1857 		kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1858 		for (i = 0; i < 9; ++i)
1859 			kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1860 		vcpu->arch.hcall_needed = 0;
1861 #ifdef CONFIG_BOOKE
1862 	} else if (vcpu->arch.epr_needed) {
1863 		kvmppc_set_epr(vcpu, run->epr.epr);
1864 		vcpu->arch.epr_needed = 0;
1865 #endif
1866 	}
1867 
1868 	kvm_sigset_activate(vcpu);
1869 
1870 	if (run->immediate_exit)
1871 		r = -EINTR;
1872 	else
1873 		r = kvmppc_vcpu_run(vcpu);
1874 
1875 	kvm_sigset_deactivate(vcpu);
1876 
1877 #ifdef CONFIG_ALTIVEC
1878 out:
1879 #endif
1880 
1881 	/*
1882 	 * We're already returning to userspace, don't pass the
1883 	 * RESUME_HOST flags along.
1884 	 */
1885 	if (r > 0)
1886 		r = 0;
1887 
1888 	vcpu_put(vcpu);
1889 	return r;
1890 }
1891 
1892 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1893 {
1894 	if (irq->irq == KVM_INTERRUPT_UNSET) {
1895 		kvmppc_core_dequeue_external(vcpu);
1896 		return 0;
1897 	}
1898 
1899 	kvmppc_core_queue_external(vcpu, irq);
1900 
1901 	kvm_vcpu_kick(vcpu);
1902 
1903 	return 0;
1904 }
1905 
1906 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1907 				     struct kvm_enable_cap *cap)
1908 {
1909 	int r;
1910 
1911 	if (cap->flags)
1912 		return -EINVAL;
1913 
1914 	switch (cap->cap) {
1915 	case KVM_CAP_PPC_OSI:
1916 		r = 0;
1917 		vcpu->arch.osi_enabled = true;
1918 		break;
1919 	case KVM_CAP_PPC_PAPR:
1920 		r = 0;
1921 		vcpu->arch.papr_enabled = true;
1922 		break;
1923 	case KVM_CAP_PPC_EPR:
1924 		r = 0;
1925 		if (cap->args[0])
1926 			vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1927 		else
1928 			vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1929 		break;
1930 #ifdef CONFIG_BOOKE
1931 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
1932 		r = 0;
1933 		vcpu->arch.watchdog_enabled = true;
1934 		break;
1935 #endif
1936 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1937 	case KVM_CAP_SW_TLB: {
1938 		struct kvm_config_tlb cfg;
1939 		void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1940 
1941 		r = -EFAULT;
1942 		if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1943 			break;
1944 
1945 		r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1946 		break;
1947 	}
1948 #endif
1949 #ifdef CONFIG_KVM_MPIC
1950 	case KVM_CAP_IRQ_MPIC: {
1951 		struct fd f;
1952 		struct kvm_device *dev;
1953 
1954 		r = -EBADF;
1955 		f = fdget(cap->args[0]);
1956 		if (!f.file)
1957 			break;
1958 
1959 		r = -EPERM;
1960 		dev = kvm_device_from_filp(f.file);
1961 		if (dev)
1962 			r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1963 
1964 		fdput(f);
1965 		break;
1966 	}
1967 #endif
1968 #ifdef CONFIG_KVM_XICS
1969 	case KVM_CAP_IRQ_XICS: {
1970 		struct fd f;
1971 		struct kvm_device *dev;
1972 
1973 		r = -EBADF;
1974 		f = fdget(cap->args[0]);
1975 		if (!f.file)
1976 			break;
1977 
1978 		r = -EPERM;
1979 		dev = kvm_device_from_filp(f.file);
1980 		if (dev) {
1981 			if (xics_on_xive())
1982 				r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1983 			else
1984 				r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1985 		}
1986 
1987 		fdput(f);
1988 		break;
1989 	}
1990 #endif /* CONFIG_KVM_XICS */
1991 #ifdef CONFIG_KVM_XIVE
1992 	case KVM_CAP_PPC_IRQ_XIVE: {
1993 		struct fd f;
1994 		struct kvm_device *dev;
1995 
1996 		r = -EBADF;
1997 		f = fdget(cap->args[0]);
1998 		if (!f.file)
1999 			break;
2000 
2001 		r = -ENXIO;
2002 		if (!xive_enabled())
2003 			break;
2004 
2005 		r = -EPERM;
2006 		dev = kvm_device_from_filp(f.file);
2007 		if (dev)
2008 			r = kvmppc_xive_native_connect_vcpu(dev, vcpu,
2009 							    cap->args[1]);
2010 
2011 		fdput(f);
2012 		break;
2013 	}
2014 #endif /* CONFIG_KVM_XIVE */
2015 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
2016 	case KVM_CAP_PPC_FWNMI:
2017 		r = -EINVAL;
2018 		if (!is_kvmppc_hv_enabled(vcpu->kvm))
2019 			break;
2020 		r = 0;
2021 		vcpu->kvm->arch.fwnmi_enabled = true;
2022 		break;
2023 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
2024 	default:
2025 		r = -EINVAL;
2026 		break;
2027 	}
2028 
2029 	if (!r)
2030 		r = kvmppc_sanity_check(vcpu);
2031 
2032 	return r;
2033 }
2034 
2035 bool kvm_arch_intc_initialized(struct kvm *kvm)
2036 {
2037 #ifdef CONFIG_KVM_MPIC
2038 	if (kvm->arch.mpic)
2039 		return true;
2040 #endif
2041 #ifdef CONFIG_KVM_XICS
2042 	if (kvm->arch.xics || kvm->arch.xive)
2043 		return true;
2044 #endif
2045 	return false;
2046 }
2047 
2048 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2049                                     struct kvm_mp_state *mp_state)
2050 {
2051 	return -EINVAL;
2052 }
2053 
2054 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2055                                     struct kvm_mp_state *mp_state)
2056 {
2057 	return -EINVAL;
2058 }
2059 
2060 long kvm_arch_vcpu_async_ioctl(struct file *filp,
2061 			       unsigned int ioctl, unsigned long arg)
2062 {
2063 	struct kvm_vcpu *vcpu = filp->private_data;
2064 	void __user *argp = (void __user *)arg;
2065 
2066 	if (ioctl == KVM_INTERRUPT) {
2067 		struct kvm_interrupt irq;
2068 		if (copy_from_user(&irq, argp, sizeof(irq)))
2069 			return -EFAULT;
2070 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2071 	}
2072 	return -ENOIOCTLCMD;
2073 }
2074 
2075 long kvm_arch_vcpu_ioctl(struct file *filp,
2076                          unsigned int ioctl, unsigned long arg)
2077 {
2078 	struct kvm_vcpu *vcpu = filp->private_data;
2079 	void __user *argp = (void __user *)arg;
2080 	long r;
2081 
2082 	switch (ioctl) {
2083 	case KVM_ENABLE_CAP:
2084 	{
2085 		struct kvm_enable_cap cap;
2086 		r = -EFAULT;
2087 		if (copy_from_user(&cap, argp, sizeof(cap)))
2088 			goto out;
2089 		vcpu_load(vcpu);
2090 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
2091 		vcpu_put(vcpu);
2092 		break;
2093 	}
2094 
2095 	case KVM_SET_ONE_REG:
2096 	case KVM_GET_ONE_REG:
2097 	{
2098 		struct kvm_one_reg reg;
2099 		r = -EFAULT;
2100 		if (copy_from_user(&reg, argp, sizeof(reg)))
2101 			goto out;
2102 		if (ioctl == KVM_SET_ONE_REG)
2103 			r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
2104 		else
2105 			r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
2106 		break;
2107 	}
2108 
2109 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
2110 	case KVM_DIRTY_TLB: {
2111 		struct kvm_dirty_tlb dirty;
2112 		r = -EFAULT;
2113 		if (copy_from_user(&dirty, argp, sizeof(dirty)))
2114 			goto out;
2115 		vcpu_load(vcpu);
2116 		r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
2117 		vcpu_put(vcpu);
2118 		break;
2119 	}
2120 #endif
2121 	default:
2122 		r = -EINVAL;
2123 	}
2124 
2125 out:
2126 	return r;
2127 }
2128 
2129 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2130 {
2131 	return VM_FAULT_SIGBUS;
2132 }
2133 
2134 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
2135 {
2136 	u32 inst_nop = 0x60000000;
2137 #ifdef CONFIG_KVM_BOOKE_HV
2138 	u32 inst_sc1 = 0x44000022;
2139 	pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
2140 	pvinfo->hcall[1] = cpu_to_be32(inst_nop);
2141 	pvinfo->hcall[2] = cpu_to_be32(inst_nop);
2142 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2143 #else
2144 	u32 inst_lis = 0x3c000000;
2145 	u32 inst_ori = 0x60000000;
2146 	u32 inst_sc = 0x44000002;
2147 	u32 inst_imm_mask = 0xffff;
2148 
2149 	/*
2150 	 * The hypercall to get into KVM from within guest context is as
2151 	 * follows:
2152 	 *
2153 	 *    lis r0, r0, KVM_SC_MAGIC_R0@h
2154 	 *    ori r0, KVM_SC_MAGIC_R0@l
2155 	 *    sc
2156 	 *    nop
2157 	 */
2158 	pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
2159 	pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
2160 	pvinfo->hcall[2] = cpu_to_be32(inst_sc);
2161 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2162 #endif
2163 
2164 	pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
2165 
2166 	return 0;
2167 }
2168 
2169 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
2170 			  bool line_status)
2171 {
2172 	if (!irqchip_in_kernel(kvm))
2173 		return -ENXIO;
2174 
2175 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2176 					irq_event->irq, irq_event->level,
2177 					line_status);
2178 	return 0;
2179 }
2180 
2181 
2182 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
2183 			    struct kvm_enable_cap *cap)
2184 {
2185 	int r;
2186 
2187 	if (cap->flags)
2188 		return -EINVAL;
2189 
2190 	switch (cap->cap) {
2191 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
2192 	case KVM_CAP_PPC_ENABLE_HCALL: {
2193 		unsigned long hcall = cap->args[0];
2194 
2195 		r = -EINVAL;
2196 		if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
2197 		    cap->args[1] > 1)
2198 			break;
2199 		if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
2200 			break;
2201 		if (cap->args[1])
2202 			set_bit(hcall / 4, kvm->arch.enabled_hcalls);
2203 		else
2204 			clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
2205 		r = 0;
2206 		break;
2207 	}
2208 	case KVM_CAP_PPC_SMT: {
2209 		unsigned long mode = cap->args[0];
2210 		unsigned long flags = cap->args[1];
2211 
2212 		r = -EINVAL;
2213 		if (kvm->arch.kvm_ops->set_smt_mode)
2214 			r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
2215 		break;
2216 	}
2217 
2218 	case KVM_CAP_PPC_NESTED_HV:
2219 		r = -EINVAL;
2220 		if (!is_kvmppc_hv_enabled(kvm) ||
2221 		    !kvm->arch.kvm_ops->enable_nested)
2222 			break;
2223 		r = kvm->arch.kvm_ops->enable_nested(kvm);
2224 		break;
2225 #endif
2226 #if defined(CONFIG_KVM_BOOK3S_HV_POSSIBLE)
2227 	case KVM_CAP_PPC_SECURE_GUEST:
2228 		r = -EINVAL;
2229 		if (!is_kvmppc_hv_enabled(kvm) || !kvm->arch.kvm_ops->enable_svm)
2230 			break;
2231 		r = kvm->arch.kvm_ops->enable_svm(kvm);
2232 		break;
2233 	case KVM_CAP_PPC_DAWR1:
2234 		r = -EINVAL;
2235 		if (!is_kvmppc_hv_enabled(kvm) || !kvm->arch.kvm_ops->enable_dawr1)
2236 			break;
2237 		r = kvm->arch.kvm_ops->enable_dawr1(kvm);
2238 		break;
2239 #endif
2240 	default:
2241 		r = -EINVAL;
2242 		break;
2243 	}
2244 
2245 	return r;
2246 }
2247 
2248 #ifdef CONFIG_PPC_BOOK3S_64
2249 /*
2250  * These functions check whether the underlying hardware is safe
2251  * against attacks based on observing the effects of speculatively
2252  * executed instructions, and whether it supplies instructions for
2253  * use in workarounds.  The information comes from firmware, either
2254  * via the device tree on powernv platforms or from an hcall on
2255  * pseries platforms.
2256  */
2257 #ifdef CONFIG_PPC_PSERIES
2258 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2259 {
2260 	struct h_cpu_char_result c;
2261 	unsigned long rc;
2262 
2263 	if (!machine_is(pseries))
2264 		return -ENOTTY;
2265 
2266 	rc = plpar_get_cpu_characteristics(&c);
2267 	if (rc == H_SUCCESS) {
2268 		cp->character = c.character;
2269 		cp->behaviour = c.behaviour;
2270 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2271 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2272 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2273 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2274 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2275 			KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
2276 			KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
2277 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2278 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2279 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2280 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2281 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2282 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2283 	}
2284 	return 0;
2285 }
2286 #else
2287 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2288 {
2289 	return -ENOTTY;
2290 }
2291 #endif
2292 
2293 static inline bool have_fw_feat(struct device_node *fw_features,
2294 				const char *state, const char *name)
2295 {
2296 	struct device_node *np;
2297 	bool r = false;
2298 
2299 	np = of_get_child_by_name(fw_features, name);
2300 	if (np) {
2301 		r = of_property_read_bool(np, state);
2302 		of_node_put(np);
2303 	}
2304 	return r;
2305 }
2306 
2307 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2308 {
2309 	struct device_node *np, *fw_features;
2310 	int r;
2311 
2312 	memset(cp, 0, sizeof(*cp));
2313 	r = pseries_get_cpu_char(cp);
2314 	if (r != -ENOTTY)
2315 		return r;
2316 
2317 	np = of_find_node_by_name(NULL, "ibm,opal");
2318 	if (np) {
2319 		fw_features = of_get_child_by_name(np, "fw-features");
2320 		of_node_put(np);
2321 		if (!fw_features)
2322 			return 0;
2323 		if (have_fw_feat(fw_features, "enabled",
2324 				 "inst-spec-barrier-ori31,31,0"))
2325 			cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
2326 		if (have_fw_feat(fw_features, "enabled",
2327 				 "fw-bcctrl-serialized"))
2328 			cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
2329 		if (have_fw_feat(fw_features, "enabled",
2330 				 "inst-l1d-flush-ori30,30,0"))
2331 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
2332 		if (have_fw_feat(fw_features, "enabled",
2333 				 "inst-l1d-flush-trig2"))
2334 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
2335 		if (have_fw_feat(fw_features, "enabled",
2336 				 "fw-l1d-thread-split"))
2337 			cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
2338 		if (have_fw_feat(fw_features, "enabled",
2339 				 "fw-count-cache-disabled"))
2340 			cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2341 		if (have_fw_feat(fw_features, "enabled",
2342 				 "fw-count-cache-flush-bcctr2,0,0"))
2343 			cp->character |= KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2344 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2345 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2346 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2347 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2348 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2349 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2350 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2351 
2352 		if (have_fw_feat(fw_features, "enabled",
2353 				 "speculation-policy-favor-security"))
2354 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
2355 		if (!have_fw_feat(fw_features, "disabled",
2356 				  "needs-l1d-flush-msr-pr-0-to-1"))
2357 			cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
2358 		if (!have_fw_feat(fw_features, "disabled",
2359 				  "needs-spec-barrier-for-bound-checks"))
2360 			cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2361 		if (have_fw_feat(fw_features, "enabled",
2362 				 "needs-count-cache-flush-on-context-switch"))
2363 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2364 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2365 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2366 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2367 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2368 
2369 		of_node_put(fw_features);
2370 	}
2371 
2372 	return 0;
2373 }
2374 #endif
2375 
2376 long kvm_arch_vm_ioctl(struct file *filp,
2377                        unsigned int ioctl, unsigned long arg)
2378 {
2379 	struct kvm *kvm __maybe_unused = filp->private_data;
2380 	void __user *argp = (void __user *)arg;
2381 	long r;
2382 
2383 	switch (ioctl) {
2384 	case KVM_PPC_GET_PVINFO: {
2385 		struct kvm_ppc_pvinfo pvinfo;
2386 		memset(&pvinfo, 0, sizeof(pvinfo));
2387 		r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
2388 		if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
2389 			r = -EFAULT;
2390 			goto out;
2391 		}
2392 
2393 		break;
2394 	}
2395 #ifdef CONFIG_SPAPR_TCE_IOMMU
2396 	case KVM_CREATE_SPAPR_TCE_64: {
2397 		struct kvm_create_spapr_tce_64 create_tce_64;
2398 
2399 		r = -EFAULT;
2400 		if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
2401 			goto out;
2402 		if (create_tce_64.flags) {
2403 			r = -EINVAL;
2404 			goto out;
2405 		}
2406 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2407 		goto out;
2408 	}
2409 	case KVM_CREATE_SPAPR_TCE: {
2410 		struct kvm_create_spapr_tce create_tce;
2411 		struct kvm_create_spapr_tce_64 create_tce_64;
2412 
2413 		r = -EFAULT;
2414 		if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
2415 			goto out;
2416 
2417 		create_tce_64.liobn = create_tce.liobn;
2418 		create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
2419 		create_tce_64.offset = 0;
2420 		create_tce_64.size = create_tce.window_size >>
2421 				IOMMU_PAGE_SHIFT_4K;
2422 		create_tce_64.flags = 0;
2423 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2424 		goto out;
2425 	}
2426 #endif
2427 #ifdef CONFIG_PPC_BOOK3S_64
2428 	case KVM_PPC_GET_SMMU_INFO: {
2429 		struct kvm_ppc_smmu_info info;
2430 		struct kvm *kvm = filp->private_data;
2431 
2432 		memset(&info, 0, sizeof(info));
2433 		r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
2434 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2435 			r = -EFAULT;
2436 		break;
2437 	}
2438 	case KVM_PPC_RTAS_DEFINE_TOKEN: {
2439 		struct kvm *kvm = filp->private_data;
2440 
2441 		r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
2442 		break;
2443 	}
2444 	case KVM_PPC_CONFIGURE_V3_MMU: {
2445 		struct kvm *kvm = filp->private_data;
2446 		struct kvm_ppc_mmuv3_cfg cfg;
2447 
2448 		r = -EINVAL;
2449 		if (!kvm->arch.kvm_ops->configure_mmu)
2450 			goto out;
2451 		r = -EFAULT;
2452 		if (copy_from_user(&cfg, argp, sizeof(cfg)))
2453 			goto out;
2454 		r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
2455 		break;
2456 	}
2457 	case KVM_PPC_GET_RMMU_INFO: {
2458 		struct kvm *kvm = filp->private_data;
2459 		struct kvm_ppc_rmmu_info info;
2460 
2461 		r = -EINVAL;
2462 		if (!kvm->arch.kvm_ops->get_rmmu_info)
2463 			goto out;
2464 		r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
2465 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2466 			r = -EFAULT;
2467 		break;
2468 	}
2469 	case KVM_PPC_GET_CPU_CHAR: {
2470 		struct kvm_ppc_cpu_char cpuchar;
2471 
2472 		r = kvmppc_get_cpu_char(&cpuchar);
2473 		if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2474 			r = -EFAULT;
2475 		break;
2476 	}
2477 	case KVM_PPC_SVM_OFF: {
2478 		struct kvm *kvm = filp->private_data;
2479 
2480 		r = 0;
2481 		if (!kvm->arch.kvm_ops->svm_off)
2482 			goto out;
2483 
2484 		r = kvm->arch.kvm_ops->svm_off(kvm);
2485 		break;
2486 	}
2487 	default: {
2488 		struct kvm *kvm = filp->private_data;
2489 		r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
2490 	}
2491 #else /* CONFIG_PPC_BOOK3S_64 */
2492 	default:
2493 		r = -ENOTTY;
2494 #endif
2495 	}
2496 out:
2497 	return r;
2498 }
2499 
2500 static DEFINE_IDA(lpid_inuse);
2501 static unsigned long nr_lpids;
2502 
2503 long kvmppc_alloc_lpid(void)
2504 {
2505 	int lpid;
2506 
2507 	/* The host LPID must always be 0 (allocation starts at 1) */
2508 	lpid = ida_alloc_range(&lpid_inuse, 1, nr_lpids - 1, GFP_KERNEL);
2509 	if (lpid < 0) {
2510 		if (lpid == -ENOMEM)
2511 			pr_err("%s: Out of memory\n", __func__);
2512 		else
2513 			pr_err("%s: No LPIDs free\n", __func__);
2514 		return -ENOMEM;
2515 	}
2516 
2517 	return lpid;
2518 }
2519 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2520 
2521 void kvmppc_free_lpid(long lpid)
2522 {
2523 	ida_free(&lpid_inuse, lpid);
2524 }
2525 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2526 
2527 /* nr_lpids_param includes the host LPID */
2528 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2529 {
2530 	nr_lpids = nr_lpids_param;
2531 }
2532 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2533 
2534 int kvm_arch_init(void *opaque)
2535 {
2536 	return 0;
2537 }
2538 
2539 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);
2540 
2541 void kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry)
2542 {
2543 	if (vcpu->kvm->arch.kvm_ops->create_vcpu_debugfs)
2544 		vcpu->kvm->arch.kvm_ops->create_vcpu_debugfs(vcpu, debugfs_dentry);
2545 }
2546 
2547 int kvm_arch_create_vm_debugfs(struct kvm *kvm)
2548 {
2549 	if (kvm->arch.kvm_ops->create_vm_debugfs)
2550 		kvm->arch.kvm_ops->create_vm_debugfs(kvm);
2551 	return 0;
2552 }
2553