xref: /openbmc/linux/arch/powerpc/kvm/powerpc.c (revision dfe94d40)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *
4  * Copyright IBM Corp. 2007
5  *
6  * Authors: Hollis Blanchard <hollisb@us.ibm.com>
7  *          Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
8  */
9 
10 #include <linux/errno.h>
11 #include <linux/err.h>
12 #include <linux/kvm_host.h>
13 #include <linux/vmalloc.h>
14 #include <linux/hrtimer.h>
15 #include <linux/sched/signal.h>
16 #include <linux/fs.h>
17 #include <linux/slab.h>
18 #include <linux/file.h>
19 #include <linux/module.h>
20 #include <linux/irqbypass.h>
21 #include <linux/kvm_irqfd.h>
22 #include <asm/cputable.h>
23 #include <linux/uaccess.h>
24 #include <asm/kvm_ppc.h>
25 #include <asm/cputhreads.h>
26 #include <asm/irqflags.h>
27 #include <asm/iommu.h>
28 #include <asm/switch_to.h>
29 #include <asm/xive.h>
30 #ifdef CONFIG_PPC_PSERIES
31 #include <asm/hvcall.h>
32 #include <asm/plpar_wrappers.h>
33 #endif
34 #include <asm/ultravisor.h>
35 
36 #include "timing.h"
37 #include "irq.h"
38 #include "../mm/mmu_decl.h"
39 
40 #define CREATE_TRACE_POINTS
41 #include "trace.h"
42 
43 struct kvmppc_ops *kvmppc_hv_ops;
44 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
45 struct kvmppc_ops *kvmppc_pr_ops;
46 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
47 
48 
49 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
50 {
51 	return !!(v->arch.pending_exceptions) || kvm_request_pending(v);
52 }
53 
54 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
55 {
56 	return kvm_arch_vcpu_runnable(vcpu);
57 }
58 
59 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
60 {
61 	return false;
62 }
63 
64 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
65 {
66 	return 1;
67 }
68 
69 /*
70  * Common checks before entering the guest world.  Call with interrupts
71  * disabled.
72  *
73  * returns:
74  *
75  * == 1 if we're ready to go into guest state
76  * <= 0 if we need to go back to the host with return value
77  */
78 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
79 {
80 	int r;
81 
82 	WARN_ON(irqs_disabled());
83 	hard_irq_disable();
84 
85 	while (true) {
86 		if (need_resched()) {
87 			local_irq_enable();
88 			cond_resched();
89 			hard_irq_disable();
90 			continue;
91 		}
92 
93 		if (signal_pending(current)) {
94 			kvmppc_account_exit(vcpu, SIGNAL_EXITS);
95 			vcpu->run->exit_reason = KVM_EXIT_INTR;
96 			r = -EINTR;
97 			break;
98 		}
99 
100 		vcpu->mode = IN_GUEST_MODE;
101 
102 		/*
103 		 * Reading vcpu->requests must happen after setting vcpu->mode,
104 		 * so we don't miss a request because the requester sees
105 		 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
106 		 * before next entering the guest (and thus doesn't IPI).
107 		 * This also orders the write to mode from any reads
108 		 * to the page tables done while the VCPU is running.
109 		 * Please see the comment in kvm_flush_remote_tlbs.
110 		 */
111 		smp_mb();
112 
113 		if (kvm_request_pending(vcpu)) {
114 			/* Make sure we process requests preemptable */
115 			local_irq_enable();
116 			trace_kvm_check_requests(vcpu);
117 			r = kvmppc_core_check_requests(vcpu);
118 			hard_irq_disable();
119 			if (r > 0)
120 				continue;
121 			break;
122 		}
123 
124 		if (kvmppc_core_prepare_to_enter(vcpu)) {
125 			/* interrupts got enabled in between, so we
126 			   are back at square 1 */
127 			continue;
128 		}
129 
130 		guest_enter_irqoff();
131 		return 1;
132 	}
133 
134 	/* return to host */
135 	local_irq_enable();
136 	return r;
137 }
138 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
139 
140 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
141 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
142 {
143 	struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
144 	int i;
145 
146 	shared->sprg0 = swab64(shared->sprg0);
147 	shared->sprg1 = swab64(shared->sprg1);
148 	shared->sprg2 = swab64(shared->sprg2);
149 	shared->sprg3 = swab64(shared->sprg3);
150 	shared->srr0 = swab64(shared->srr0);
151 	shared->srr1 = swab64(shared->srr1);
152 	shared->dar = swab64(shared->dar);
153 	shared->msr = swab64(shared->msr);
154 	shared->dsisr = swab32(shared->dsisr);
155 	shared->int_pending = swab32(shared->int_pending);
156 	for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
157 		shared->sr[i] = swab32(shared->sr[i]);
158 }
159 #endif
160 
161 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
162 {
163 	int nr = kvmppc_get_gpr(vcpu, 11);
164 	int r;
165 	unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
166 	unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
167 	unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
168 	unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
169 	unsigned long r2 = 0;
170 
171 	if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
172 		/* 32 bit mode */
173 		param1 &= 0xffffffff;
174 		param2 &= 0xffffffff;
175 		param3 &= 0xffffffff;
176 		param4 &= 0xffffffff;
177 	}
178 
179 	switch (nr) {
180 	case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
181 	{
182 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
183 		/* Book3S can be little endian, find it out here */
184 		int shared_big_endian = true;
185 		if (vcpu->arch.intr_msr & MSR_LE)
186 			shared_big_endian = false;
187 		if (shared_big_endian != vcpu->arch.shared_big_endian)
188 			kvmppc_swab_shared(vcpu);
189 		vcpu->arch.shared_big_endian = shared_big_endian;
190 #endif
191 
192 		if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
193 			/*
194 			 * Older versions of the Linux magic page code had
195 			 * a bug where they would map their trampoline code
196 			 * NX. If that's the case, remove !PR NX capability.
197 			 */
198 			vcpu->arch.disable_kernel_nx = true;
199 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
200 		}
201 
202 		vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
203 		vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
204 
205 #ifdef CONFIG_PPC_64K_PAGES
206 		/*
207 		 * Make sure our 4k magic page is in the same window of a 64k
208 		 * page within the guest and within the host's page.
209 		 */
210 		if ((vcpu->arch.magic_page_pa & 0xf000) !=
211 		    ((ulong)vcpu->arch.shared & 0xf000)) {
212 			void *old_shared = vcpu->arch.shared;
213 			ulong shared = (ulong)vcpu->arch.shared;
214 			void *new_shared;
215 
216 			shared &= PAGE_MASK;
217 			shared |= vcpu->arch.magic_page_pa & 0xf000;
218 			new_shared = (void*)shared;
219 			memcpy(new_shared, old_shared, 0x1000);
220 			vcpu->arch.shared = new_shared;
221 		}
222 #endif
223 
224 		r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
225 
226 		r = EV_SUCCESS;
227 		break;
228 	}
229 	case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
230 		r = EV_SUCCESS;
231 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
232 		r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
233 #endif
234 
235 		/* Second return value is in r4 */
236 		break;
237 	case EV_HCALL_TOKEN(EV_IDLE):
238 		r = EV_SUCCESS;
239 		kvm_vcpu_block(vcpu);
240 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
241 		break;
242 	default:
243 		r = EV_UNIMPLEMENTED;
244 		break;
245 	}
246 
247 	kvmppc_set_gpr(vcpu, 4, r2);
248 
249 	return r;
250 }
251 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
252 
253 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
254 {
255 	int r = false;
256 
257 	/* We have to know what CPU to virtualize */
258 	if (!vcpu->arch.pvr)
259 		goto out;
260 
261 	/* PAPR only works with book3s_64 */
262 	if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
263 		goto out;
264 
265 	/* HV KVM can only do PAPR mode for now */
266 	if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
267 		goto out;
268 
269 #ifdef CONFIG_KVM_BOOKE_HV
270 	if (!cpu_has_feature(CPU_FTR_EMB_HV))
271 		goto out;
272 #endif
273 
274 	r = true;
275 
276 out:
277 	vcpu->arch.sane = r;
278 	return r ? 0 : -EINVAL;
279 }
280 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
281 
282 int kvmppc_emulate_mmio(struct kvm_vcpu *vcpu)
283 {
284 	enum emulation_result er;
285 	int r;
286 
287 	er = kvmppc_emulate_loadstore(vcpu);
288 	switch (er) {
289 	case EMULATE_DONE:
290 		/* Future optimization: only reload non-volatiles if they were
291 		 * actually modified. */
292 		r = RESUME_GUEST_NV;
293 		break;
294 	case EMULATE_AGAIN:
295 		r = RESUME_GUEST;
296 		break;
297 	case EMULATE_DO_MMIO:
298 		vcpu->run->exit_reason = KVM_EXIT_MMIO;
299 		/* We must reload nonvolatiles because "update" load/store
300 		 * instructions modify register state. */
301 		/* Future optimization: only reload non-volatiles if they were
302 		 * actually modified. */
303 		r = RESUME_HOST_NV;
304 		break;
305 	case EMULATE_FAIL:
306 	{
307 		u32 last_inst;
308 
309 		kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
310 		/* XXX Deliver Program interrupt to guest. */
311 		pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
312 		r = RESUME_HOST;
313 		break;
314 	}
315 	default:
316 		WARN_ON(1);
317 		r = RESUME_GUEST;
318 	}
319 
320 	return r;
321 }
322 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
323 
324 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
325 	      bool data)
326 {
327 	ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
328 	struct kvmppc_pte pte;
329 	int r = -EINVAL;
330 
331 	vcpu->stat.st++;
332 
333 	if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->store_to_eaddr)
334 		r = vcpu->kvm->arch.kvm_ops->store_to_eaddr(vcpu, eaddr, ptr,
335 							    size);
336 
337 	if ((!r) || (r == -EAGAIN))
338 		return r;
339 
340 	r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
341 			 XLATE_WRITE, &pte);
342 	if (r < 0)
343 		return r;
344 
345 	*eaddr = pte.raddr;
346 
347 	if (!pte.may_write)
348 		return -EPERM;
349 
350 	/* Magic page override */
351 	if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
352 	    ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
353 	    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
354 		void *magic = vcpu->arch.shared;
355 		magic += pte.eaddr & 0xfff;
356 		memcpy(magic, ptr, size);
357 		return EMULATE_DONE;
358 	}
359 
360 	if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
361 		return EMULATE_DO_MMIO;
362 
363 	return EMULATE_DONE;
364 }
365 EXPORT_SYMBOL_GPL(kvmppc_st);
366 
367 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
368 		      bool data)
369 {
370 	ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
371 	struct kvmppc_pte pte;
372 	int rc = -EINVAL;
373 
374 	vcpu->stat.ld++;
375 
376 	if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->load_from_eaddr)
377 		rc = vcpu->kvm->arch.kvm_ops->load_from_eaddr(vcpu, eaddr, ptr,
378 							      size);
379 
380 	if ((!rc) || (rc == -EAGAIN))
381 		return rc;
382 
383 	rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
384 			  XLATE_READ, &pte);
385 	if (rc)
386 		return rc;
387 
388 	*eaddr = pte.raddr;
389 
390 	if (!pte.may_read)
391 		return -EPERM;
392 
393 	if (!data && !pte.may_execute)
394 		return -ENOEXEC;
395 
396 	/* Magic page override */
397 	if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
398 	    ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
399 	    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
400 		void *magic = vcpu->arch.shared;
401 		magic += pte.eaddr & 0xfff;
402 		memcpy(ptr, magic, size);
403 		return EMULATE_DONE;
404 	}
405 
406 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
407 	rc = kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size);
408 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
409 	if (rc)
410 		return EMULATE_DO_MMIO;
411 
412 	return EMULATE_DONE;
413 }
414 EXPORT_SYMBOL_GPL(kvmppc_ld);
415 
416 int kvm_arch_hardware_enable(void)
417 {
418 	return 0;
419 }
420 
421 int kvm_arch_hardware_setup(void *opaque)
422 {
423 	return 0;
424 }
425 
426 int kvm_arch_check_processor_compat(void *opaque)
427 {
428 	return kvmppc_core_check_processor_compat();
429 }
430 
431 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
432 {
433 	struct kvmppc_ops *kvm_ops = NULL;
434 	/*
435 	 * if we have both HV and PR enabled, default is HV
436 	 */
437 	if (type == 0) {
438 		if (kvmppc_hv_ops)
439 			kvm_ops = kvmppc_hv_ops;
440 		else
441 			kvm_ops = kvmppc_pr_ops;
442 		if (!kvm_ops)
443 			goto err_out;
444 	} else	if (type == KVM_VM_PPC_HV) {
445 		if (!kvmppc_hv_ops)
446 			goto err_out;
447 		kvm_ops = kvmppc_hv_ops;
448 	} else if (type == KVM_VM_PPC_PR) {
449 		if (!kvmppc_pr_ops)
450 			goto err_out;
451 		kvm_ops = kvmppc_pr_ops;
452 	} else
453 		goto err_out;
454 
455 	if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
456 		return -ENOENT;
457 
458 	kvm->arch.kvm_ops = kvm_ops;
459 	return kvmppc_core_init_vm(kvm);
460 err_out:
461 	return -EINVAL;
462 }
463 
464 void kvm_arch_destroy_vm(struct kvm *kvm)
465 {
466 	unsigned int i;
467 	struct kvm_vcpu *vcpu;
468 
469 #ifdef CONFIG_KVM_XICS
470 	/*
471 	 * We call kick_all_cpus_sync() to ensure that all
472 	 * CPUs have executed any pending IPIs before we
473 	 * continue and free VCPUs structures below.
474 	 */
475 	if (is_kvmppc_hv_enabled(kvm))
476 		kick_all_cpus_sync();
477 #endif
478 
479 	kvm_for_each_vcpu(i, vcpu, kvm)
480 		kvm_vcpu_destroy(vcpu);
481 
482 	mutex_lock(&kvm->lock);
483 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
484 		kvm->vcpus[i] = NULL;
485 
486 	atomic_set(&kvm->online_vcpus, 0);
487 
488 	kvmppc_core_destroy_vm(kvm);
489 
490 	mutex_unlock(&kvm->lock);
491 
492 	/* drop the module reference */
493 	module_put(kvm->arch.kvm_ops->owner);
494 }
495 
496 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
497 {
498 	int r;
499 	/* Assume we're using HV mode when the HV module is loaded */
500 	int hv_enabled = kvmppc_hv_ops ? 1 : 0;
501 
502 	if (kvm) {
503 		/*
504 		 * Hooray - we know which VM type we're running on. Depend on
505 		 * that rather than the guess above.
506 		 */
507 		hv_enabled = is_kvmppc_hv_enabled(kvm);
508 	}
509 
510 	switch (ext) {
511 #ifdef CONFIG_BOOKE
512 	case KVM_CAP_PPC_BOOKE_SREGS:
513 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
514 	case KVM_CAP_PPC_EPR:
515 #else
516 	case KVM_CAP_PPC_SEGSTATE:
517 	case KVM_CAP_PPC_HIOR:
518 	case KVM_CAP_PPC_PAPR:
519 #endif
520 	case KVM_CAP_PPC_UNSET_IRQ:
521 	case KVM_CAP_PPC_IRQ_LEVEL:
522 	case KVM_CAP_ENABLE_CAP:
523 	case KVM_CAP_ONE_REG:
524 	case KVM_CAP_IOEVENTFD:
525 	case KVM_CAP_DEVICE_CTRL:
526 	case KVM_CAP_IMMEDIATE_EXIT:
527 	case KVM_CAP_SET_GUEST_DEBUG:
528 		r = 1;
529 		break;
530 	case KVM_CAP_PPC_GUEST_DEBUG_SSTEP:
531 	case KVM_CAP_PPC_PAIRED_SINGLES:
532 	case KVM_CAP_PPC_OSI:
533 	case KVM_CAP_PPC_GET_PVINFO:
534 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
535 	case KVM_CAP_SW_TLB:
536 #endif
537 		/* We support this only for PR */
538 		r = !hv_enabled;
539 		break;
540 #ifdef CONFIG_KVM_MPIC
541 	case KVM_CAP_IRQ_MPIC:
542 		r = 1;
543 		break;
544 #endif
545 
546 #ifdef CONFIG_PPC_BOOK3S_64
547 	case KVM_CAP_SPAPR_TCE:
548 	case KVM_CAP_SPAPR_TCE_64:
549 		r = 1;
550 		break;
551 	case KVM_CAP_SPAPR_TCE_VFIO:
552 		r = !!cpu_has_feature(CPU_FTR_HVMODE);
553 		break;
554 	case KVM_CAP_PPC_RTAS:
555 	case KVM_CAP_PPC_FIXUP_HCALL:
556 	case KVM_CAP_PPC_ENABLE_HCALL:
557 #ifdef CONFIG_KVM_XICS
558 	case KVM_CAP_IRQ_XICS:
559 #endif
560 	case KVM_CAP_PPC_GET_CPU_CHAR:
561 		r = 1;
562 		break;
563 #ifdef CONFIG_KVM_XIVE
564 	case KVM_CAP_PPC_IRQ_XIVE:
565 		/*
566 		 * We need XIVE to be enabled on the platform (implies
567 		 * a POWER9 processor) and the PowerNV platform, as
568 		 * nested is not yet supported.
569 		 */
570 		r = xive_enabled() && !!cpu_has_feature(CPU_FTR_HVMODE) &&
571 			kvmppc_xive_native_supported();
572 		break;
573 #endif
574 
575 	case KVM_CAP_PPC_ALLOC_HTAB:
576 		r = hv_enabled;
577 		break;
578 #endif /* CONFIG_PPC_BOOK3S_64 */
579 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
580 	case KVM_CAP_PPC_SMT:
581 		r = 0;
582 		if (kvm) {
583 			if (kvm->arch.emul_smt_mode > 1)
584 				r = kvm->arch.emul_smt_mode;
585 			else
586 				r = kvm->arch.smt_mode;
587 		} else if (hv_enabled) {
588 			if (cpu_has_feature(CPU_FTR_ARCH_300))
589 				r = 1;
590 			else
591 				r = threads_per_subcore;
592 		}
593 		break;
594 	case KVM_CAP_PPC_SMT_POSSIBLE:
595 		r = 1;
596 		if (hv_enabled) {
597 			if (!cpu_has_feature(CPU_FTR_ARCH_300))
598 				r = ((threads_per_subcore << 1) - 1);
599 			else
600 				/* P9 can emulate dbells, so allow any mode */
601 				r = 8 | 4 | 2 | 1;
602 		}
603 		break;
604 	case KVM_CAP_PPC_RMA:
605 		r = 0;
606 		break;
607 	case KVM_CAP_PPC_HWRNG:
608 		r = kvmppc_hwrng_present();
609 		break;
610 	case KVM_CAP_PPC_MMU_RADIX:
611 		r = !!(hv_enabled && radix_enabled());
612 		break;
613 	case KVM_CAP_PPC_MMU_HASH_V3:
614 		r = !!(hv_enabled && cpu_has_feature(CPU_FTR_ARCH_300) &&
615 		       cpu_has_feature(CPU_FTR_HVMODE));
616 		break;
617 	case KVM_CAP_PPC_NESTED_HV:
618 		r = !!(hv_enabled && kvmppc_hv_ops->enable_nested &&
619 		       !kvmppc_hv_ops->enable_nested(NULL));
620 		break;
621 #endif
622 	case KVM_CAP_SYNC_MMU:
623 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
624 		r = hv_enabled;
625 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
626 		r = 1;
627 #else
628 		r = 0;
629 #endif
630 		break;
631 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
632 	case KVM_CAP_PPC_HTAB_FD:
633 		r = hv_enabled;
634 		break;
635 #endif
636 	case KVM_CAP_NR_VCPUS:
637 		/*
638 		 * Recommending a number of CPUs is somewhat arbitrary; we
639 		 * return the number of present CPUs for -HV (since a host
640 		 * will have secondary threads "offline"), and for other KVM
641 		 * implementations just count online CPUs.
642 		 */
643 		if (hv_enabled)
644 			r = num_present_cpus();
645 		else
646 			r = num_online_cpus();
647 		break;
648 	case KVM_CAP_MAX_VCPUS:
649 		r = KVM_MAX_VCPUS;
650 		break;
651 	case KVM_CAP_MAX_VCPU_ID:
652 		r = KVM_MAX_VCPU_ID;
653 		break;
654 #ifdef CONFIG_PPC_BOOK3S_64
655 	case KVM_CAP_PPC_GET_SMMU_INFO:
656 		r = 1;
657 		break;
658 	case KVM_CAP_SPAPR_MULTITCE:
659 		r = 1;
660 		break;
661 	case KVM_CAP_SPAPR_RESIZE_HPT:
662 		r = !!hv_enabled;
663 		break;
664 #endif
665 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
666 	case KVM_CAP_PPC_FWNMI:
667 		r = hv_enabled;
668 		break;
669 #endif
670 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
671 	case KVM_CAP_PPC_HTM:
672 		r = !!(cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM) ||
673 		     (hv_enabled && cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST));
674 		break;
675 #endif
676 #if defined(CONFIG_KVM_BOOK3S_HV_POSSIBLE)
677 	case KVM_CAP_PPC_SECURE_GUEST:
678 		r = hv_enabled && kvmppc_hv_ops->enable_svm &&
679 			!kvmppc_hv_ops->enable_svm(NULL);
680 		break;
681 #endif
682 	default:
683 		r = 0;
684 		break;
685 	}
686 	return r;
687 
688 }
689 
690 long kvm_arch_dev_ioctl(struct file *filp,
691                         unsigned int ioctl, unsigned long arg)
692 {
693 	return -EINVAL;
694 }
695 
696 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
697 {
698 	kvmppc_core_free_memslot(kvm, slot);
699 }
700 
701 int kvm_arch_prepare_memory_region(struct kvm *kvm,
702 				   struct kvm_memory_slot *memslot,
703 				   const struct kvm_userspace_memory_region *mem,
704 				   enum kvm_mr_change change)
705 {
706 	return kvmppc_core_prepare_memory_region(kvm, memslot, mem, change);
707 }
708 
709 void kvm_arch_commit_memory_region(struct kvm *kvm,
710 				   const struct kvm_userspace_memory_region *mem,
711 				   struct kvm_memory_slot *old,
712 				   const struct kvm_memory_slot *new,
713 				   enum kvm_mr_change change)
714 {
715 	kvmppc_core_commit_memory_region(kvm, mem, old, new, change);
716 }
717 
718 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
719 				   struct kvm_memory_slot *slot)
720 {
721 	kvmppc_core_flush_memslot(kvm, slot);
722 }
723 
724 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
725 {
726 	return 0;
727 }
728 
729 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
730 {
731 	struct kvm_vcpu *vcpu;
732 
733 	vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
734 	kvmppc_decrementer_func(vcpu);
735 
736 	return HRTIMER_NORESTART;
737 }
738 
739 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
740 {
741 	int err;
742 
743 	hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
744 	vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
745 	vcpu->arch.dec_expires = get_tb();
746 
747 #ifdef CONFIG_KVM_EXIT_TIMING
748 	mutex_init(&vcpu->arch.exit_timing_lock);
749 #endif
750 	err = kvmppc_subarch_vcpu_init(vcpu);
751 	if (err)
752 		return err;
753 
754 	err = kvmppc_core_vcpu_create(vcpu);
755 	if (err)
756 		goto out_vcpu_uninit;
757 
758 	vcpu->arch.waitp = &vcpu->wait;
759 	kvmppc_create_vcpu_debugfs(vcpu, vcpu->vcpu_id);
760 	return 0;
761 
762 out_vcpu_uninit:
763 	kvmppc_subarch_vcpu_uninit(vcpu);
764 	return err;
765 }
766 
767 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
768 {
769 }
770 
771 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
772 {
773 	/* Make sure we're not using the vcpu anymore */
774 	hrtimer_cancel(&vcpu->arch.dec_timer);
775 
776 	kvmppc_remove_vcpu_debugfs(vcpu);
777 
778 	switch (vcpu->arch.irq_type) {
779 	case KVMPPC_IRQ_MPIC:
780 		kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
781 		break;
782 	case KVMPPC_IRQ_XICS:
783 		if (xics_on_xive())
784 			kvmppc_xive_cleanup_vcpu(vcpu);
785 		else
786 			kvmppc_xics_free_icp(vcpu);
787 		break;
788 	case KVMPPC_IRQ_XIVE:
789 		kvmppc_xive_native_cleanup_vcpu(vcpu);
790 		break;
791 	}
792 
793 	kvmppc_core_vcpu_free(vcpu);
794 
795 	kvmppc_subarch_vcpu_uninit(vcpu);
796 }
797 
798 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
799 {
800 	return kvmppc_core_pending_dec(vcpu);
801 }
802 
803 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
804 {
805 #ifdef CONFIG_BOOKE
806 	/*
807 	 * vrsave (formerly usprg0) isn't used by Linux, but may
808 	 * be used by the guest.
809 	 *
810 	 * On non-booke this is associated with Altivec and
811 	 * is handled by code in book3s.c.
812 	 */
813 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
814 #endif
815 	kvmppc_core_vcpu_load(vcpu, cpu);
816 }
817 
818 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
819 {
820 	kvmppc_core_vcpu_put(vcpu);
821 #ifdef CONFIG_BOOKE
822 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
823 #endif
824 }
825 
826 /*
827  * irq_bypass_add_producer and irq_bypass_del_producer are only
828  * useful if the architecture supports PCI passthrough.
829  * irq_bypass_stop and irq_bypass_start are not needed and so
830  * kvm_ops are not defined for them.
831  */
832 bool kvm_arch_has_irq_bypass(void)
833 {
834 	return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
835 		(kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
836 }
837 
838 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
839 				     struct irq_bypass_producer *prod)
840 {
841 	struct kvm_kernel_irqfd *irqfd =
842 		container_of(cons, struct kvm_kernel_irqfd, consumer);
843 	struct kvm *kvm = irqfd->kvm;
844 
845 	if (kvm->arch.kvm_ops->irq_bypass_add_producer)
846 		return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
847 
848 	return 0;
849 }
850 
851 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
852 				      struct irq_bypass_producer *prod)
853 {
854 	struct kvm_kernel_irqfd *irqfd =
855 		container_of(cons, struct kvm_kernel_irqfd, consumer);
856 	struct kvm *kvm = irqfd->kvm;
857 
858 	if (kvm->arch.kvm_ops->irq_bypass_del_producer)
859 		kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
860 }
861 
862 #ifdef CONFIG_VSX
863 static inline int kvmppc_get_vsr_dword_offset(int index)
864 {
865 	int offset;
866 
867 	if ((index != 0) && (index != 1))
868 		return -1;
869 
870 #ifdef __BIG_ENDIAN
871 	offset =  index;
872 #else
873 	offset = 1 - index;
874 #endif
875 
876 	return offset;
877 }
878 
879 static inline int kvmppc_get_vsr_word_offset(int index)
880 {
881 	int offset;
882 
883 	if ((index > 3) || (index < 0))
884 		return -1;
885 
886 #ifdef __BIG_ENDIAN
887 	offset = index;
888 #else
889 	offset = 3 - index;
890 #endif
891 	return offset;
892 }
893 
894 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
895 	u64 gpr)
896 {
897 	union kvmppc_one_reg val;
898 	int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
899 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
900 
901 	if (offset == -1)
902 		return;
903 
904 	if (index >= 32) {
905 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
906 		val.vsxval[offset] = gpr;
907 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
908 	} else {
909 		VCPU_VSX_FPR(vcpu, index, offset) = gpr;
910 	}
911 }
912 
913 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
914 	u64 gpr)
915 {
916 	union kvmppc_one_reg val;
917 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
918 
919 	if (index >= 32) {
920 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
921 		val.vsxval[0] = gpr;
922 		val.vsxval[1] = gpr;
923 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
924 	} else {
925 		VCPU_VSX_FPR(vcpu, index, 0) = gpr;
926 		VCPU_VSX_FPR(vcpu, index, 1) = gpr;
927 	}
928 }
929 
930 static inline void kvmppc_set_vsr_word_dump(struct kvm_vcpu *vcpu,
931 	u32 gpr)
932 {
933 	union kvmppc_one_reg val;
934 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
935 
936 	if (index >= 32) {
937 		val.vsx32val[0] = gpr;
938 		val.vsx32val[1] = gpr;
939 		val.vsx32val[2] = gpr;
940 		val.vsx32val[3] = gpr;
941 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
942 	} else {
943 		val.vsx32val[0] = gpr;
944 		val.vsx32val[1] = gpr;
945 		VCPU_VSX_FPR(vcpu, index, 0) = val.vsxval[0];
946 		VCPU_VSX_FPR(vcpu, index, 1) = val.vsxval[0];
947 	}
948 }
949 
950 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
951 	u32 gpr32)
952 {
953 	union kvmppc_one_reg val;
954 	int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
955 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
956 	int dword_offset, word_offset;
957 
958 	if (offset == -1)
959 		return;
960 
961 	if (index >= 32) {
962 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
963 		val.vsx32val[offset] = gpr32;
964 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
965 	} else {
966 		dword_offset = offset / 2;
967 		word_offset = offset % 2;
968 		val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
969 		val.vsx32val[word_offset] = gpr32;
970 		VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
971 	}
972 }
973 #endif /* CONFIG_VSX */
974 
975 #ifdef CONFIG_ALTIVEC
976 static inline int kvmppc_get_vmx_offset_generic(struct kvm_vcpu *vcpu,
977 		int index, int element_size)
978 {
979 	int offset;
980 	int elts = sizeof(vector128)/element_size;
981 
982 	if ((index < 0) || (index >= elts))
983 		return -1;
984 
985 	if (kvmppc_need_byteswap(vcpu))
986 		offset = elts - index - 1;
987 	else
988 		offset = index;
989 
990 	return offset;
991 }
992 
993 static inline int kvmppc_get_vmx_dword_offset(struct kvm_vcpu *vcpu,
994 		int index)
995 {
996 	return kvmppc_get_vmx_offset_generic(vcpu, index, 8);
997 }
998 
999 static inline int kvmppc_get_vmx_word_offset(struct kvm_vcpu *vcpu,
1000 		int index)
1001 {
1002 	return kvmppc_get_vmx_offset_generic(vcpu, index, 4);
1003 }
1004 
1005 static inline int kvmppc_get_vmx_hword_offset(struct kvm_vcpu *vcpu,
1006 		int index)
1007 {
1008 	return kvmppc_get_vmx_offset_generic(vcpu, index, 2);
1009 }
1010 
1011 static inline int kvmppc_get_vmx_byte_offset(struct kvm_vcpu *vcpu,
1012 		int index)
1013 {
1014 	return kvmppc_get_vmx_offset_generic(vcpu, index, 1);
1015 }
1016 
1017 
1018 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu,
1019 	u64 gpr)
1020 {
1021 	union kvmppc_one_reg val;
1022 	int offset = kvmppc_get_vmx_dword_offset(vcpu,
1023 			vcpu->arch.mmio_vmx_offset);
1024 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1025 
1026 	if (offset == -1)
1027 		return;
1028 
1029 	val.vval = VCPU_VSX_VR(vcpu, index);
1030 	val.vsxval[offset] = gpr;
1031 	VCPU_VSX_VR(vcpu, index) = val.vval;
1032 }
1033 
1034 static inline void kvmppc_set_vmx_word(struct kvm_vcpu *vcpu,
1035 	u32 gpr32)
1036 {
1037 	union kvmppc_one_reg val;
1038 	int offset = kvmppc_get_vmx_word_offset(vcpu,
1039 			vcpu->arch.mmio_vmx_offset);
1040 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1041 
1042 	if (offset == -1)
1043 		return;
1044 
1045 	val.vval = VCPU_VSX_VR(vcpu, index);
1046 	val.vsx32val[offset] = gpr32;
1047 	VCPU_VSX_VR(vcpu, index) = val.vval;
1048 }
1049 
1050 static inline void kvmppc_set_vmx_hword(struct kvm_vcpu *vcpu,
1051 	u16 gpr16)
1052 {
1053 	union kvmppc_one_reg val;
1054 	int offset = kvmppc_get_vmx_hword_offset(vcpu,
1055 			vcpu->arch.mmio_vmx_offset);
1056 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1057 
1058 	if (offset == -1)
1059 		return;
1060 
1061 	val.vval = VCPU_VSX_VR(vcpu, index);
1062 	val.vsx16val[offset] = gpr16;
1063 	VCPU_VSX_VR(vcpu, index) = val.vval;
1064 }
1065 
1066 static inline void kvmppc_set_vmx_byte(struct kvm_vcpu *vcpu,
1067 	u8 gpr8)
1068 {
1069 	union kvmppc_one_reg val;
1070 	int offset = kvmppc_get_vmx_byte_offset(vcpu,
1071 			vcpu->arch.mmio_vmx_offset);
1072 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1073 
1074 	if (offset == -1)
1075 		return;
1076 
1077 	val.vval = VCPU_VSX_VR(vcpu, index);
1078 	val.vsx8val[offset] = gpr8;
1079 	VCPU_VSX_VR(vcpu, index) = val.vval;
1080 }
1081 #endif /* CONFIG_ALTIVEC */
1082 
1083 #ifdef CONFIG_PPC_FPU
1084 static inline u64 sp_to_dp(u32 fprs)
1085 {
1086 	u64 fprd;
1087 
1088 	preempt_disable();
1089 	enable_kernel_fp();
1090 	asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m"UPD_CONSTR (fprd) : "m"UPD_CONSTR (fprs)
1091 	     : "fr0");
1092 	preempt_enable();
1093 	return fprd;
1094 }
1095 
1096 static inline u32 dp_to_sp(u64 fprd)
1097 {
1098 	u32 fprs;
1099 
1100 	preempt_disable();
1101 	enable_kernel_fp();
1102 	asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m"UPD_CONSTR (fprs) : "m"UPD_CONSTR (fprd)
1103 	     : "fr0");
1104 	preempt_enable();
1105 	return fprs;
1106 }
1107 
1108 #else
1109 #define sp_to_dp(x)	(x)
1110 #define dp_to_sp(x)	(x)
1111 #endif /* CONFIG_PPC_FPU */
1112 
1113 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu)
1114 {
1115 	struct kvm_run *run = vcpu->run;
1116 	u64 gpr;
1117 
1118 	if (run->mmio.len > sizeof(gpr)) {
1119 		printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
1120 		return;
1121 	}
1122 
1123 	if (!vcpu->arch.mmio_host_swabbed) {
1124 		switch (run->mmio.len) {
1125 		case 8: gpr = *(u64 *)run->mmio.data; break;
1126 		case 4: gpr = *(u32 *)run->mmio.data; break;
1127 		case 2: gpr = *(u16 *)run->mmio.data; break;
1128 		case 1: gpr = *(u8 *)run->mmio.data; break;
1129 		}
1130 	} else {
1131 		switch (run->mmio.len) {
1132 		case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
1133 		case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
1134 		case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
1135 		case 1: gpr = *(u8 *)run->mmio.data; break;
1136 		}
1137 	}
1138 
1139 	/* conversion between single and double precision */
1140 	if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
1141 		gpr = sp_to_dp(gpr);
1142 
1143 	if (vcpu->arch.mmio_sign_extend) {
1144 		switch (run->mmio.len) {
1145 #ifdef CONFIG_PPC64
1146 		case 4:
1147 			gpr = (s64)(s32)gpr;
1148 			break;
1149 #endif
1150 		case 2:
1151 			gpr = (s64)(s16)gpr;
1152 			break;
1153 		case 1:
1154 			gpr = (s64)(s8)gpr;
1155 			break;
1156 		}
1157 	}
1158 
1159 	switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1160 	case KVM_MMIO_REG_GPR:
1161 		kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1162 		break;
1163 	case KVM_MMIO_REG_FPR:
1164 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1165 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_FP);
1166 
1167 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1168 		break;
1169 #ifdef CONFIG_PPC_BOOK3S
1170 	case KVM_MMIO_REG_QPR:
1171 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1172 		break;
1173 	case KVM_MMIO_REG_FQPR:
1174 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1175 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1176 		break;
1177 #endif
1178 #ifdef CONFIG_VSX
1179 	case KVM_MMIO_REG_VSX:
1180 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1181 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VSX);
1182 
1183 		if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_DWORD)
1184 			kvmppc_set_vsr_dword(vcpu, gpr);
1185 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_WORD)
1186 			kvmppc_set_vsr_word(vcpu, gpr);
1187 		else if (vcpu->arch.mmio_copy_type ==
1188 				KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1189 			kvmppc_set_vsr_dword_dump(vcpu, gpr);
1190 		else if (vcpu->arch.mmio_copy_type ==
1191 				KVMPPC_VSX_COPY_WORD_LOAD_DUMP)
1192 			kvmppc_set_vsr_word_dump(vcpu, gpr);
1193 		break;
1194 #endif
1195 #ifdef CONFIG_ALTIVEC
1196 	case KVM_MMIO_REG_VMX:
1197 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1198 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VEC);
1199 
1200 		if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_DWORD)
1201 			kvmppc_set_vmx_dword(vcpu, gpr);
1202 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_WORD)
1203 			kvmppc_set_vmx_word(vcpu, gpr);
1204 		else if (vcpu->arch.mmio_copy_type ==
1205 				KVMPPC_VMX_COPY_HWORD)
1206 			kvmppc_set_vmx_hword(vcpu, gpr);
1207 		else if (vcpu->arch.mmio_copy_type ==
1208 				KVMPPC_VMX_COPY_BYTE)
1209 			kvmppc_set_vmx_byte(vcpu, gpr);
1210 		break;
1211 #endif
1212 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1213 	case KVM_MMIO_REG_NESTED_GPR:
1214 		if (kvmppc_need_byteswap(vcpu))
1215 			gpr = swab64(gpr);
1216 		kvm_vcpu_write_guest(vcpu, vcpu->arch.nested_io_gpr, &gpr,
1217 				     sizeof(gpr));
1218 		break;
1219 #endif
1220 	default:
1221 		BUG();
1222 	}
1223 }
1224 
1225 static int __kvmppc_handle_load(struct kvm_vcpu *vcpu,
1226 				unsigned int rt, unsigned int bytes,
1227 				int is_default_endian, int sign_extend)
1228 {
1229 	struct kvm_run *run = vcpu->run;
1230 	int idx, ret;
1231 	bool host_swabbed;
1232 
1233 	/* Pity C doesn't have a logical XOR operator */
1234 	if (kvmppc_need_byteswap(vcpu)) {
1235 		host_swabbed = is_default_endian;
1236 	} else {
1237 		host_swabbed = !is_default_endian;
1238 	}
1239 
1240 	if (bytes > sizeof(run->mmio.data)) {
1241 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1242 		       run->mmio.len);
1243 	}
1244 
1245 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1246 	run->mmio.len = bytes;
1247 	run->mmio.is_write = 0;
1248 
1249 	vcpu->arch.io_gpr = rt;
1250 	vcpu->arch.mmio_host_swabbed = host_swabbed;
1251 	vcpu->mmio_needed = 1;
1252 	vcpu->mmio_is_write = 0;
1253 	vcpu->arch.mmio_sign_extend = sign_extend;
1254 
1255 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1256 
1257 	ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1258 			      bytes, &run->mmio.data);
1259 
1260 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1261 
1262 	if (!ret) {
1263 		kvmppc_complete_mmio_load(vcpu);
1264 		vcpu->mmio_needed = 0;
1265 		return EMULATE_DONE;
1266 	}
1267 
1268 	return EMULATE_DO_MMIO;
1269 }
1270 
1271 int kvmppc_handle_load(struct kvm_vcpu *vcpu,
1272 		       unsigned int rt, unsigned int bytes,
1273 		       int is_default_endian)
1274 {
1275 	return __kvmppc_handle_load(vcpu, rt, bytes, is_default_endian, 0);
1276 }
1277 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1278 
1279 /* Same as above, but sign extends */
1280 int kvmppc_handle_loads(struct kvm_vcpu *vcpu,
1281 			unsigned int rt, unsigned int bytes,
1282 			int is_default_endian)
1283 {
1284 	return __kvmppc_handle_load(vcpu, rt, bytes, is_default_endian, 1);
1285 }
1286 
1287 #ifdef CONFIG_VSX
1288 int kvmppc_handle_vsx_load(struct kvm_vcpu *vcpu,
1289 			unsigned int rt, unsigned int bytes,
1290 			int is_default_endian, int mmio_sign_extend)
1291 {
1292 	enum emulation_result emulated = EMULATE_DONE;
1293 
1294 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1295 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1296 		return EMULATE_FAIL;
1297 
1298 	while (vcpu->arch.mmio_vsx_copy_nums) {
1299 		emulated = __kvmppc_handle_load(vcpu, rt, bytes,
1300 			is_default_endian, mmio_sign_extend);
1301 
1302 		if (emulated != EMULATE_DONE)
1303 			break;
1304 
1305 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1306 
1307 		vcpu->arch.mmio_vsx_copy_nums--;
1308 		vcpu->arch.mmio_vsx_offset++;
1309 	}
1310 	return emulated;
1311 }
1312 #endif /* CONFIG_VSX */
1313 
1314 int kvmppc_handle_store(struct kvm_vcpu *vcpu,
1315 			u64 val, unsigned int bytes, int is_default_endian)
1316 {
1317 	struct kvm_run *run = vcpu->run;
1318 	void *data = run->mmio.data;
1319 	int idx, ret;
1320 	bool host_swabbed;
1321 
1322 	/* Pity C doesn't have a logical XOR operator */
1323 	if (kvmppc_need_byteswap(vcpu)) {
1324 		host_swabbed = is_default_endian;
1325 	} else {
1326 		host_swabbed = !is_default_endian;
1327 	}
1328 
1329 	if (bytes > sizeof(run->mmio.data)) {
1330 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1331 		       run->mmio.len);
1332 	}
1333 
1334 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1335 	run->mmio.len = bytes;
1336 	run->mmio.is_write = 1;
1337 	vcpu->mmio_needed = 1;
1338 	vcpu->mmio_is_write = 1;
1339 
1340 	if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1341 		val = dp_to_sp(val);
1342 
1343 	/* Store the value at the lowest bytes in 'data'. */
1344 	if (!host_swabbed) {
1345 		switch (bytes) {
1346 		case 8: *(u64 *)data = val; break;
1347 		case 4: *(u32 *)data = val; break;
1348 		case 2: *(u16 *)data = val; break;
1349 		case 1: *(u8  *)data = val; break;
1350 		}
1351 	} else {
1352 		switch (bytes) {
1353 		case 8: *(u64 *)data = swab64(val); break;
1354 		case 4: *(u32 *)data = swab32(val); break;
1355 		case 2: *(u16 *)data = swab16(val); break;
1356 		case 1: *(u8  *)data = val; break;
1357 		}
1358 	}
1359 
1360 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1361 
1362 	ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1363 			       bytes, &run->mmio.data);
1364 
1365 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1366 
1367 	if (!ret) {
1368 		vcpu->mmio_needed = 0;
1369 		return EMULATE_DONE;
1370 	}
1371 
1372 	return EMULATE_DO_MMIO;
1373 }
1374 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1375 
1376 #ifdef CONFIG_VSX
1377 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1378 {
1379 	u32 dword_offset, word_offset;
1380 	union kvmppc_one_reg reg;
1381 	int vsx_offset = 0;
1382 	int copy_type = vcpu->arch.mmio_copy_type;
1383 	int result = 0;
1384 
1385 	switch (copy_type) {
1386 	case KVMPPC_VSX_COPY_DWORD:
1387 		vsx_offset =
1388 			kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1389 
1390 		if (vsx_offset == -1) {
1391 			result = -1;
1392 			break;
1393 		}
1394 
1395 		if (rs < 32) {
1396 			*val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1397 		} else {
1398 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1399 			*val = reg.vsxval[vsx_offset];
1400 		}
1401 		break;
1402 
1403 	case KVMPPC_VSX_COPY_WORD:
1404 		vsx_offset =
1405 			kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1406 
1407 		if (vsx_offset == -1) {
1408 			result = -1;
1409 			break;
1410 		}
1411 
1412 		if (rs < 32) {
1413 			dword_offset = vsx_offset / 2;
1414 			word_offset = vsx_offset % 2;
1415 			reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1416 			*val = reg.vsx32val[word_offset];
1417 		} else {
1418 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1419 			*val = reg.vsx32val[vsx_offset];
1420 		}
1421 		break;
1422 
1423 	default:
1424 		result = -1;
1425 		break;
1426 	}
1427 
1428 	return result;
1429 }
1430 
1431 int kvmppc_handle_vsx_store(struct kvm_vcpu *vcpu,
1432 			int rs, unsigned int bytes, int is_default_endian)
1433 {
1434 	u64 val;
1435 	enum emulation_result emulated = EMULATE_DONE;
1436 
1437 	vcpu->arch.io_gpr = rs;
1438 
1439 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1440 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1441 		return EMULATE_FAIL;
1442 
1443 	while (vcpu->arch.mmio_vsx_copy_nums) {
1444 		if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1445 			return EMULATE_FAIL;
1446 
1447 		emulated = kvmppc_handle_store(vcpu,
1448 			 val, bytes, is_default_endian);
1449 
1450 		if (emulated != EMULATE_DONE)
1451 			break;
1452 
1453 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1454 
1455 		vcpu->arch.mmio_vsx_copy_nums--;
1456 		vcpu->arch.mmio_vsx_offset++;
1457 	}
1458 
1459 	return emulated;
1460 }
1461 
1462 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu)
1463 {
1464 	struct kvm_run *run = vcpu->run;
1465 	enum emulation_result emulated = EMULATE_FAIL;
1466 	int r;
1467 
1468 	vcpu->arch.paddr_accessed += run->mmio.len;
1469 
1470 	if (!vcpu->mmio_is_write) {
1471 		emulated = kvmppc_handle_vsx_load(vcpu, vcpu->arch.io_gpr,
1472 			 run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1473 	} else {
1474 		emulated = kvmppc_handle_vsx_store(vcpu,
1475 			 vcpu->arch.io_gpr, run->mmio.len, 1);
1476 	}
1477 
1478 	switch (emulated) {
1479 	case EMULATE_DO_MMIO:
1480 		run->exit_reason = KVM_EXIT_MMIO;
1481 		r = RESUME_HOST;
1482 		break;
1483 	case EMULATE_FAIL:
1484 		pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1485 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1486 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1487 		r = RESUME_HOST;
1488 		break;
1489 	default:
1490 		r = RESUME_GUEST;
1491 		break;
1492 	}
1493 	return r;
1494 }
1495 #endif /* CONFIG_VSX */
1496 
1497 #ifdef CONFIG_ALTIVEC
1498 int kvmppc_handle_vmx_load(struct kvm_vcpu *vcpu,
1499 		unsigned int rt, unsigned int bytes, int is_default_endian)
1500 {
1501 	enum emulation_result emulated = EMULATE_DONE;
1502 
1503 	if (vcpu->arch.mmio_vsx_copy_nums > 2)
1504 		return EMULATE_FAIL;
1505 
1506 	while (vcpu->arch.mmio_vmx_copy_nums) {
1507 		emulated = __kvmppc_handle_load(vcpu, rt, bytes,
1508 				is_default_endian, 0);
1509 
1510 		if (emulated != EMULATE_DONE)
1511 			break;
1512 
1513 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1514 		vcpu->arch.mmio_vmx_copy_nums--;
1515 		vcpu->arch.mmio_vmx_offset++;
1516 	}
1517 
1518 	return emulated;
1519 }
1520 
1521 int kvmppc_get_vmx_dword(struct kvm_vcpu *vcpu, int index, u64 *val)
1522 {
1523 	union kvmppc_one_reg reg;
1524 	int vmx_offset = 0;
1525 	int result = 0;
1526 
1527 	vmx_offset =
1528 		kvmppc_get_vmx_dword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1529 
1530 	if (vmx_offset == -1)
1531 		return -1;
1532 
1533 	reg.vval = VCPU_VSX_VR(vcpu, index);
1534 	*val = reg.vsxval[vmx_offset];
1535 
1536 	return result;
1537 }
1538 
1539 int kvmppc_get_vmx_word(struct kvm_vcpu *vcpu, int index, u64 *val)
1540 {
1541 	union kvmppc_one_reg reg;
1542 	int vmx_offset = 0;
1543 	int result = 0;
1544 
1545 	vmx_offset =
1546 		kvmppc_get_vmx_word_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1547 
1548 	if (vmx_offset == -1)
1549 		return -1;
1550 
1551 	reg.vval = VCPU_VSX_VR(vcpu, index);
1552 	*val = reg.vsx32val[vmx_offset];
1553 
1554 	return result;
1555 }
1556 
1557 int kvmppc_get_vmx_hword(struct kvm_vcpu *vcpu, int index, u64 *val)
1558 {
1559 	union kvmppc_one_reg reg;
1560 	int vmx_offset = 0;
1561 	int result = 0;
1562 
1563 	vmx_offset =
1564 		kvmppc_get_vmx_hword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1565 
1566 	if (vmx_offset == -1)
1567 		return -1;
1568 
1569 	reg.vval = VCPU_VSX_VR(vcpu, index);
1570 	*val = reg.vsx16val[vmx_offset];
1571 
1572 	return result;
1573 }
1574 
1575 int kvmppc_get_vmx_byte(struct kvm_vcpu *vcpu, int index, u64 *val)
1576 {
1577 	union kvmppc_one_reg reg;
1578 	int vmx_offset = 0;
1579 	int result = 0;
1580 
1581 	vmx_offset =
1582 		kvmppc_get_vmx_byte_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1583 
1584 	if (vmx_offset == -1)
1585 		return -1;
1586 
1587 	reg.vval = VCPU_VSX_VR(vcpu, index);
1588 	*val = reg.vsx8val[vmx_offset];
1589 
1590 	return result;
1591 }
1592 
1593 int kvmppc_handle_vmx_store(struct kvm_vcpu *vcpu,
1594 		unsigned int rs, unsigned int bytes, int is_default_endian)
1595 {
1596 	u64 val = 0;
1597 	unsigned int index = rs & KVM_MMIO_REG_MASK;
1598 	enum emulation_result emulated = EMULATE_DONE;
1599 
1600 	if (vcpu->arch.mmio_vsx_copy_nums > 2)
1601 		return EMULATE_FAIL;
1602 
1603 	vcpu->arch.io_gpr = rs;
1604 
1605 	while (vcpu->arch.mmio_vmx_copy_nums) {
1606 		switch (vcpu->arch.mmio_copy_type) {
1607 		case KVMPPC_VMX_COPY_DWORD:
1608 			if (kvmppc_get_vmx_dword(vcpu, index, &val) == -1)
1609 				return EMULATE_FAIL;
1610 
1611 			break;
1612 		case KVMPPC_VMX_COPY_WORD:
1613 			if (kvmppc_get_vmx_word(vcpu, index, &val) == -1)
1614 				return EMULATE_FAIL;
1615 			break;
1616 		case KVMPPC_VMX_COPY_HWORD:
1617 			if (kvmppc_get_vmx_hword(vcpu, index, &val) == -1)
1618 				return EMULATE_FAIL;
1619 			break;
1620 		case KVMPPC_VMX_COPY_BYTE:
1621 			if (kvmppc_get_vmx_byte(vcpu, index, &val) == -1)
1622 				return EMULATE_FAIL;
1623 			break;
1624 		default:
1625 			return EMULATE_FAIL;
1626 		}
1627 
1628 		emulated = kvmppc_handle_store(vcpu, val, bytes,
1629 				is_default_endian);
1630 		if (emulated != EMULATE_DONE)
1631 			break;
1632 
1633 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1634 		vcpu->arch.mmio_vmx_copy_nums--;
1635 		vcpu->arch.mmio_vmx_offset++;
1636 	}
1637 
1638 	return emulated;
1639 }
1640 
1641 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu)
1642 {
1643 	struct kvm_run *run = vcpu->run;
1644 	enum emulation_result emulated = EMULATE_FAIL;
1645 	int r;
1646 
1647 	vcpu->arch.paddr_accessed += run->mmio.len;
1648 
1649 	if (!vcpu->mmio_is_write) {
1650 		emulated = kvmppc_handle_vmx_load(vcpu,
1651 				vcpu->arch.io_gpr, run->mmio.len, 1);
1652 	} else {
1653 		emulated = kvmppc_handle_vmx_store(vcpu,
1654 				vcpu->arch.io_gpr, run->mmio.len, 1);
1655 	}
1656 
1657 	switch (emulated) {
1658 	case EMULATE_DO_MMIO:
1659 		run->exit_reason = KVM_EXIT_MMIO;
1660 		r = RESUME_HOST;
1661 		break;
1662 	case EMULATE_FAIL:
1663 		pr_info("KVM: MMIO emulation failed (VMX repeat)\n");
1664 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1665 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1666 		r = RESUME_HOST;
1667 		break;
1668 	default:
1669 		r = RESUME_GUEST;
1670 		break;
1671 	}
1672 	return r;
1673 }
1674 #endif /* CONFIG_ALTIVEC */
1675 
1676 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1677 {
1678 	int r = 0;
1679 	union kvmppc_one_reg val;
1680 	int size;
1681 
1682 	size = one_reg_size(reg->id);
1683 	if (size > sizeof(val))
1684 		return -EINVAL;
1685 
1686 	r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1687 	if (r == -EINVAL) {
1688 		r = 0;
1689 		switch (reg->id) {
1690 #ifdef CONFIG_ALTIVEC
1691 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1692 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1693 				r = -ENXIO;
1694 				break;
1695 			}
1696 			val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1697 			break;
1698 		case KVM_REG_PPC_VSCR:
1699 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1700 				r = -ENXIO;
1701 				break;
1702 			}
1703 			val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1704 			break;
1705 		case KVM_REG_PPC_VRSAVE:
1706 			val = get_reg_val(reg->id, vcpu->arch.vrsave);
1707 			break;
1708 #endif /* CONFIG_ALTIVEC */
1709 		default:
1710 			r = -EINVAL;
1711 			break;
1712 		}
1713 	}
1714 
1715 	if (r)
1716 		return r;
1717 
1718 	if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1719 		r = -EFAULT;
1720 
1721 	return r;
1722 }
1723 
1724 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1725 {
1726 	int r;
1727 	union kvmppc_one_reg val;
1728 	int size;
1729 
1730 	size = one_reg_size(reg->id);
1731 	if (size > sizeof(val))
1732 		return -EINVAL;
1733 
1734 	if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1735 		return -EFAULT;
1736 
1737 	r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1738 	if (r == -EINVAL) {
1739 		r = 0;
1740 		switch (reg->id) {
1741 #ifdef CONFIG_ALTIVEC
1742 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1743 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1744 				r = -ENXIO;
1745 				break;
1746 			}
1747 			vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1748 			break;
1749 		case KVM_REG_PPC_VSCR:
1750 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1751 				r = -ENXIO;
1752 				break;
1753 			}
1754 			vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1755 			break;
1756 		case KVM_REG_PPC_VRSAVE:
1757 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1758 				r = -ENXIO;
1759 				break;
1760 			}
1761 			vcpu->arch.vrsave = set_reg_val(reg->id, val);
1762 			break;
1763 #endif /* CONFIG_ALTIVEC */
1764 		default:
1765 			r = -EINVAL;
1766 			break;
1767 		}
1768 	}
1769 
1770 	return r;
1771 }
1772 
1773 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
1774 {
1775 	struct kvm_run *run = vcpu->run;
1776 	int r;
1777 
1778 	vcpu_load(vcpu);
1779 
1780 	if (vcpu->mmio_needed) {
1781 		vcpu->mmio_needed = 0;
1782 		if (!vcpu->mmio_is_write)
1783 			kvmppc_complete_mmio_load(vcpu);
1784 #ifdef CONFIG_VSX
1785 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1786 			vcpu->arch.mmio_vsx_copy_nums--;
1787 			vcpu->arch.mmio_vsx_offset++;
1788 		}
1789 
1790 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1791 			r = kvmppc_emulate_mmio_vsx_loadstore(vcpu);
1792 			if (r == RESUME_HOST) {
1793 				vcpu->mmio_needed = 1;
1794 				goto out;
1795 			}
1796 		}
1797 #endif
1798 #ifdef CONFIG_ALTIVEC
1799 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1800 			vcpu->arch.mmio_vmx_copy_nums--;
1801 			vcpu->arch.mmio_vmx_offset++;
1802 		}
1803 
1804 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1805 			r = kvmppc_emulate_mmio_vmx_loadstore(vcpu);
1806 			if (r == RESUME_HOST) {
1807 				vcpu->mmio_needed = 1;
1808 				goto out;
1809 			}
1810 		}
1811 #endif
1812 	} else if (vcpu->arch.osi_needed) {
1813 		u64 *gprs = run->osi.gprs;
1814 		int i;
1815 
1816 		for (i = 0; i < 32; i++)
1817 			kvmppc_set_gpr(vcpu, i, gprs[i]);
1818 		vcpu->arch.osi_needed = 0;
1819 	} else if (vcpu->arch.hcall_needed) {
1820 		int i;
1821 
1822 		kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1823 		for (i = 0; i < 9; ++i)
1824 			kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1825 		vcpu->arch.hcall_needed = 0;
1826 #ifdef CONFIG_BOOKE
1827 	} else if (vcpu->arch.epr_needed) {
1828 		kvmppc_set_epr(vcpu, run->epr.epr);
1829 		vcpu->arch.epr_needed = 0;
1830 #endif
1831 	}
1832 
1833 	kvm_sigset_activate(vcpu);
1834 
1835 	if (run->immediate_exit)
1836 		r = -EINTR;
1837 	else
1838 		r = kvmppc_vcpu_run(vcpu);
1839 
1840 	kvm_sigset_deactivate(vcpu);
1841 
1842 #ifdef CONFIG_ALTIVEC
1843 out:
1844 #endif
1845 	vcpu_put(vcpu);
1846 	return r;
1847 }
1848 
1849 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1850 {
1851 	if (irq->irq == KVM_INTERRUPT_UNSET) {
1852 		kvmppc_core_dequeue_external(vcpu);
1853 		return 0;
1854 	}
1855 
1856 	kvmppc_core_queue_external(vcpu, irq);
1857 
1858 	kvm_vcpu_kick(vcpu);
1859 
1860 	return 0;
1861 }
1862 
1863 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1864 				     struct kvm_enable_cap *cap)
1865 {
1866 	int r;
1867 
1868 	if (cap->flags)
1869 		return -EINVAL;
1870 
1871 	switch (cap->cap) {
1872 	case KVM_CAP_PPC_OSI:
1873 		r = 0;
1874 		vcpu->arch.osi_enabled = true;
1875 		break;
1876 	case KVM_CAP_PPC_PAPR:
1877 		r = 0;
1878 		vcpu->arch.papr_enabled = true;
1879 		break;
1880 	case KVM_CAP_PPC_EPR:
1881 		r = 0;
1882 		if (cap->args[0])
1883 			vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1884 		else
1885 			vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1886 		break;
1887 #ifdef CONFIG_BOOKE
1888 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
1889 		r = 0;
1890 		vcpu->arch.watchdog_enabled = true;
1891 		break;
1892 #endif
1893 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1894 	case KVM_CAP_SW_TLB: {
1895 		struct kvm_config_tlb cfg;
1896 		void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1897 
1898 		r = -EFAULT;
1899 		if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1900 			break;
1901 
1902 		r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1903 		break;
1904 	}
1905 #endif
1906 #ifdef CONFIG_KVM_MPIC
1907 	case KVM_CAP_IRQ_MPIC: {
1908 		struct fd f;
1909 		struct kvm_device *dev;
1910 
1911 		r = -EBADF;
1912 		f = fdget(cap->args[0]);
1913 		if (!f.file)
1914 			break;
1915 
1916 		r = -EPERM;
1917 		dev = kvm_device_from_filp(f.file);
1918 		if (dev)
1919 			r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1920 
1921 		fdput(f);
1922 		break;
1923 	}
1924 #endif
1925 #ifdef CONFIG_KVM_XICS
1926 	case KVM_CAP_IRQ_XICS: {
1927 		struct fd f;
1928 		struct kvm_device *dev;
1929 
1930 		r = -EBADF;
1931 		f = fdget(cap->args[0]);
1932 		if (!f.file)
1933 			break;
1934 
1935 		r = -EPERM;
1936 		dev = kvm_device_from_filp(f.file);
1937 		if (dev) {
1938 			if (xics_on_xive())
1939 				r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1940 			else
1941 				r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1942 		}
1943 
1944 		fdput(f);
1945 		break;
1946 	}
1947 #endif /* CONFIG_KVM_XICS */
1948 #ifdef CONFIG_KVM_XIVE
1949 	case KVM_CAP_PPC_IRQ_XIVE: {
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 = -ENXIO;
1959 		if (!xive_enabled())
1960 			break;
1961 
1962 		r = -EPERM;
1963 		dev = kvm_device_from_filp(f.file);
1964 		if (dev)
1965 			r = kvmppc_xive_native_connect_vcpu(dev, vcpu,
1966 							    cap->args[1]);
1967 
1968 		fdput(f);
1969 		break;
1970 	}
1971 #endif /* CONFIG_KVM_XIVE */
1972 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1973 	case KVM_CAP_PPC_FWNMI:
1974 		r = -EINVAL;
1975 		if (!is_kvmppc_hv_enabled(vcpu->kvm))
1976 			break;
1977 		r = 0;
1978 		vcpu->kvm->arch.fwnmi_enabled = true;
1979 		break;
1980 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1981 	default:
1982 		r = -EINVAL;
1983 		break;
1984 	}
1985 
1986 	if (!r)
1987 		r = kvmppc_sanity_check(vcpu);
1988 
1989 	return r;
1990 }
1991 
1992 bool kvm_arch_intc_initialized(struct kvm *kvm)
1993 {
1994 #ifdef CONFIG_KVM_MPIC
1995 	if (kvm->arch.mpic)
1996 		return true;
1997 #endif
1998 #ifdef CONFIG_KVM_XICS
1999 	if (kvm->arch.xics || kvm->arch.xive)
2000 		return true;
2001 #endif
2002 	return false;
2003 }
2004 
2005 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2006                                     struct kvm_mp_state *mp_state)
2007 {
2008 	return -EINVAL;
2009 }
2010 
2011 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2012                                     struct kvm_mp_state *mp_state)
2013 {
2014 	return -EINVAL;
2015 }
2016 
2017 long kvm_arch_vcpu_async_ioctl(struct file *filp,
2018 			       unsigned int ioctl, unsigned long arg)
2019 {
2020 	struct kvm_vcpu *vcpu = filp->private_data;
2021 	void __user *argp = (void __user *)arg;
2022 
2023 	if (ioctl == KVM_INTERRUPT) {
2024 		struct kvm_interrupt irq;
2025 		if (copy_from_user(&irq, argp, sizeof(irq)))
2026 			return -EFAULT;
2027 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2028 	}
2029 	return -ENOIOCTLCMD;
2030 }
2031 
2032 long kvm_arch_vcpu_ioctl(struct file *filp,
2033                          unsigned int ioctl, unsigned long arg)
2034 {
2035 	struct kvm_vcpu *vcpu = filp->private_data;
2036 	void __user *argp = (void __user *)arg;
2037 	long r;
2038 
2039 	switch (ioctl) {
2040 	case KVM_ENABLE_CAP:
2041 	{
2042 		struct kvm_enable_cap cap;
2043 		r = -EFAULT;
2044 		vcpu_load(vcpu);
2045 		if (copy_from_user(&cap, argp, sizeof(cap)))
2046 			goto out;
2047 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
2048 		vcpu_put(vcpu);
2049 		break;
2050 	}
2051 
2052 	case KVM_SET_ONE_REG:
2053 	case KVM_GET_ONE_REG:
2054 	{
2055 		struct kvm_one_reg reg;
2056 		r = -EFAULT;
2057 		if (copy_from_user(&reg, argp, sizeof(reg)))
2058 			goto out;
2059 		if (ioctl == KVM_SET_ONE_REG)
2060 			r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
2061 		else
2062 			r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
2063 		break;
2064 	}
2065 
2066 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
2067 	case KVM_DIRTY_TLB: {
2068 		struct kvm_dirty_tlb dirty;
2069 		r = -EFAULT;
2070 		vcpu_load(vcpu);
2071 		if (copy_from_user(&dirty, argp, sizeof(dirty)))
2072 			goto out;
2073 		r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
2074 		vcpu_put(vcpu);
2075 		break;
2076 	}
2077 #endif
2078 	default:
2079 		r = -EINVAL;
2080 	}
2081 
2082 out:
2083 	return r;
2084 }
2085 
2086 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2087 {
2088 	return VM_FAULT_SIGBUS;
2089 }
2090 
2091 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
2092 {
2093 	u32 inst_nop = 0x60000000;
2094 #ifdef CONFIG_KVM_BOOKE_HV
2095 	u32 inst_sc1 = 0x44000022;
2096 	pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
2097 	pvinfo->hcall[1] = cpu_to_be32(inst_nop);
2098 	pvinfo->hcall[2] = cpu_to_be32(inst_nop);
2099 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2100 #else
2101 	u32 inst_lis = 0x3c000000;
2102 	u32 inst_ori = 0x60000000;
2103 	u32 inst_sc = 0x44000002;
2104 	u32 inst_imm_mask = 0xffff;
2105 
2106 	/*
2107 	 * The hypercall to get into KVM from within guest context is as
2108 	 * follows:
2109 	 *
2110 	 *    lis r0, r0, KVM_SC_MAGIC_R0@h
2111 	 *    ori r0, KVM_SC_MAGIC_R0@l
2112 	 *    sc
2113 	 *    nop
2114 	 */
2115 	pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
2116 	pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
2117 	pvinfo->hcall[2] = cpu_to_be32(inst_sc);
2118 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2119 #endif
2120 
2121 	pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
2122 
2123 	return 0;
2124 }
2125 
2126 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
2127 			  bool line_status)
2128 {
2129 	if (!irqchip_in_kernel(kvm))
2130 		return -ENXIO;
2131 
2132 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2133 					irq_event->irq, irq_event->level,
2134 					line_status);
2135 	return 0;
2136 }
2137 
2138 
2139 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
2140 			    struct kvm_enable_cap *cap)
2141 {
2142 	int r;
2143 
2144 	if (cap->flags)
2145 		return -EINVAL;
2146 
2147 	switch (cap->cap) {
2148 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
2149 	case KVM_CAP_PPC_ENABLE_HCALL: {
2150 		unsigned long hcall = cap->args[0];
2151 
2152 		r = -EINVAL;
2153 		if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
2154 		    cap->args[1] > 1)
2155 			break;
2156 		if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
2157 			break;
2158 		if (cap->args[1])
2159 			set_bit(hcall / 4, kvm->arch.enabled_hcalls);
2160 		else
2161 			clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
2162 		r = 0;
2163 		break;
2164 	}
2165 	case KVM_CAP_PPC_SMT: {
2166 		unsigned long mode = cap->args[0];
2167 		unsigned long flags = cap->args[1];
2168 
2169 		r = -EINVAL;
2170 		if (kvm->arch.kvm_ops->set_smt_mode)
2171 			r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
2172 		break;
2173 	}
2174 
2175 	case KVM_CAP_PPC_NESTED_HV:
2176 		r = -EINVAL;
2177 		if (!is_kvmppc_hv_enabled(kvm) ||
2178 		    !kvm->arch.kvm_ops->enable_nested)
2179 			break;
2180 		r = kvm->arch.kvm_ops->enable_nested(kvm);
2181 		break;
2182 #endif
2183 #if defined(CONFIG_KVM_BOOK3S_HV_POSSIBLE)
2184 	case KVM_CAP_PPC_SECURE_GUEST:
2185 		r = -EINVAL;
2186 		if (!is_kvmppc_hv_enabled(kvm) || !kvm->arch.kvm_ops->enable_svm)
2187 			break;
2188 		r = kvm->arch.kvm_ops->enable_svm(kvm);
2189 		break;
2190 #endif
2191 	default:
2192 		r = -EINVAL;
2193 		break;
2194 	}
2195 
2196 	return r;
2197 }
2198 
2199 #ifdef CONFIG_PPC_BOOK3S_64
2200 /*
2201  * These functions check whether the underlying hardware is safe
2202  * against attacks based on observing the effects of speculatively
2203  * executed instructions, and whether it supplies instructions for
2204  * use in workarounds.  The information comes from firmware, either
2205  * via the device tree on powernv platforms or from an hcall on
2206  * pseries platforms.
2207  */
2208 #ifdef CONFIG_PPC_PSERIES
2209 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2210 {
2211 	struct h_cpu_char_result c;
2212 	unsigned long rc;
2213 
2214 	if (!machine_is(pseries))
2215 		return -ENOTTY;
2216 
2217 	rc = plpar_get_cpu_characteristics(&c);
2218 	if (rc == H_SUCCESS) {
2219 		cp->character = c.character;
2220 		cp->behaviour = c.behaviour;
2221 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2222 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2223 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2224 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2225 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2226 			KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
2227 			KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
2228 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2229 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2230 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2231 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2232 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2233 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2234 	}
2235 	return 0;
2236 }
2237 #else
2238 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2239 {
2240 	return -ENOTTY;
2241 }
2242 #endif
2243 
2244 static inline bool have_fw_feat(struct device_node *fw_features,
2245 				const char *state, const char *name)
2246 {
2247 	struct device_node *np;
2248 	bool r = false;
2249 
2250 	np = of_get_child_by_name(fw_features, name);
2251 	if (np) {
2252 		r = of_property_read_bool(np, state);
2253 		of_node_put(np);
2254 	}
2255 	return r;
2256 }
2257 
2258 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2259 {
2260 	struct device_node *np, *fw_features;
2261 	int r;
2262 
2263 	memset(cp, 0, sizeof(*cp));
2264 	r = pseries_get_cpu_char(cp);
2265 	if (r != -ENOTTY)
2266 		return r;
2267 
2268 	np = of_find_node_by_name(NULL, "ibm,opal");
2269 	if (np) {
2270 		fw_features = of_get_child_by_name(np, "fw-features");
2271 		of_node_put(np);
2272 		if (!fw_features)
2273 			return 0;
2274 		if (have_fw_feat(fw_features, "enabled",
2275 				 "inst-spec-barrier-ori31,31,0"))
2276 			cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
2277 		if (have_fw_feat(fw_features, "enabled",
2278 				 "fw-bcctrl-serialized"))
2279 			cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
2280 		if (have_fw_feat(fw_features, "enabled",
2281 				 "inst-l1d-flush-ori30,30,0"))
2282 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
2283 		if (have_fw_feat(fw_features, "enabled",
2284 				 "inst-l1d-flush-trig2"))
2285 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
2286 		if (have_fw_feat(fw_features, "enabled",
2287 				 "fw-l1d-thread-split"))
2288 			cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
2289 		if (have_fw_feat(fw_features, "enabled",
2290 				 "fw-count-cache-disabled"))
2291 			cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2292 		if (have_fw_feat(fw_features, "enabled",
2293 				 "fw-count-cache-flush-bcctr2,0,0"))
2294 			cp->character |= KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2295 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2296 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2297 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2298 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2299 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2300 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2301 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2302 
2303 		if (have_fw_feat(fw_features, "enabled",
2304 				 "speculation-policy-favor-security"))
2305 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
2306 		if (!have_fw_feat(fw_features, "disabled",
2307 				  "needs-l1d-flush-msr-pr-0-to-1"))
2308 			cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
2309 		if (!have_fw_feat(fw_features, "disabled",
2310 				  "needs-spec-barrier-for-bound-checks"))
2311 			cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2312 		if (have_fw_feat(fw_features, "enabled",
2313 				 "needs-count-cache-flush-on-context-switch"))
2314 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2315 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2316 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2317 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2318 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2319 
2320 		of_node_put(fw_features);
2321 	}
2322 
2323 	return 0;
2324 }
2325 #endif
2326 
2327 long kvm_arch_vm_ioctl(struct file *filp,
2328                        unsigned int ioctl, unsigned long arg)
2329 {
2330 	struct kvm *kvm __maybe_unused = filp->private_data;
2331 	void __user *argp = (void __user *)arg;
2332 	long r;
2333 
2334 	switch (ioctl) {
2335 	case KVM_PPC_GET_PVINFO: {
2336 		struct kvm_ppc_pvinfo pvinfo;
2337 		memset(&pvinfo, 0, sizeof(pvinfo));
2338 		r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
2339 		if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
2340 			r = -EFAULT;
2341 			goto out;
2342 		}
2343 
2344 		break;
2345 	}
2346 #ifdef CONFIG_SPAPR_TCE_IOMMU
2347 	case KVM_CREATE_SPAPR_TCE_64: {
2348 		struct kvm_create_spapr_tce_64 create_tce_64;
2349 
2350 		r = -EFAULT;
2351 		if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
2352 			goto out;
2353 		if (create_tce_64.flags) {
2354 			r = -EINVAL;
2355 			goto out;
2356 		}
2357 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2358 		goto out;
2359 	}
2360 	case KVM_CREATE_SPAPR_TCE: {
2361 		struct kvm_create_spapr_tce create_tce;
2362 		struct kvm_create_spapr_tce_64 create_tce_64;
2363 
2364 		r = -EFAULT;
2365 		if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
2366 			goto out;
2367 
2368 		create_tce_64.liobn = create_tce.liobn;
2369 		create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
2370 		create_tce_64.offset = 0;
2371 		create_tce_64.size = create_tce.window_size >>
2372 				IOMMU_PAGE_SHIFT_4K;
2373 		create_tce_64.flags = 0;
2374 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2375 		goto out;
2376 	}
2377 #endif
2378 #ifdef CONFIG_PPC_BOOK3S_64
2379 	case KVM_PPC_GET_SMMU_INFO: {
2380 		struct kvm_ppc_smmu_info info;
2381 		struct kvm *kvm = filp->private_data;
2382 
2383 		memset(&info, 0, sizeof(info));
2384 		r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
2385 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2386 			r = -EFAULT;
2387 		break;
2388 	}
2389 	case KVM_PPC_RTAS_DEFINE_TOKEN: {
2390 		struct kvm *kvm = filp->private_data;
2391 
2392 		r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
2393 		break;
2394 	}
2395 	case KVM_PPC_CONFIGURE_V3_MMU: {
2396 		struct kvm *kvm = filp->private_data;
2397 		struct kvm_ppc_mmuv3_cfg cfg;
2398 
2399 		r = -EINVAL;
2400 		if (!kvm->arch.kvm_ops->configure_mmu)
2401 			goto out;
2402 		r = -EFAULT;
2403 		if (copy_from_user(&cfg, argp, sizeof(cfg)))
2404 			goto out;
2405 		r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
2406 		break;
2407 	}
2408 	case KVM_PPC_GET_RMMU_INFO: {
2409 		struct kvm *kvm = filp->private_data;
2410 		struct kvm_ppc_rmmu_info info;
2411 
2412 		r = -EINVAL;
2413 		if (!kvm->arch.kvm_ops->get_rmmu_info)
2414 			goto out;
2415 		r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
2416 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2417 			r = -EFAULT;
2418 		break;
2419 	}
2420 	case KVM_PPC_GET_CPU_CHAR: {
2421 		struct kvm_ppc_cpu_char cpuchar;
2422 
2423 		r = kvmppc_get_cpu_char(&cpuchar);
2424 		if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2425 			r = -EFAULT;
2426 		break;
2427 	}
2428 	case KVM_PPC_SVM_OFF: {
2429 		struct kvm *kvm = filp->private_data;
2430 
2431 		r = 0;
2432 		if (!kvm->arch.kvm_ops->svm_off)
2433 			goto out;
2434 
2435 		r = kvm->arch.kvm_ops->svm_off(kvm);
2436 		break;
2437 	}
2438 	default: {
2439 		struct kvm *kvm = filp->private_data;
2440 		r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
2441 	}
2442 #else /* CONFIG_PPC_BOOK3S_64 */
2443 	default:
2444 		r = -ENOTTY;
2445 #endif
2446 	}
2447 out:
2448 	return r;
2449 }
2450 
2451 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
2452 static unsigned long nr_lpids;
2453 
2454 long kvmppc_alloc_lpid(void)
2455 {
2456 	long lpid;
2457 
2458 	do {
2459 		lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
2460 		if (lpid >= nr_lpids) {
2461 			pr_err("%s: No LPIDs free\n", __func__);
2462 			return -ENOMEM;
2463 		}
2464 	} while (test_and_set_bit(lpid, lpid_inuse));
2465 
2466 	return lpid;
2467 }
2468 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2469 
2470 void kvmppc_claim_lpid(long lpid)
2471 {
2472 	set_bit(lpid, lpid_inuse);
2473 }
2474 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
2475 
2476 void kvmppc_free_lpid(long lpid)
2477 {
2478 	clear_bit(lpid, lpid_inuse);
2479 }
2480 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2481 
2482 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2483 {
2484 	nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
2485 	memset(lpid_inuse, 0, sizeof(lpid_inuse));
2486 }
2487 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2488 
2489 int kvm_arch_init(void *opaque)
2490 {
2491 	return 0;
2492 }
2493 
2494 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);
2495