xref: /openbmc/linux/arch/powerpc/kvm/powerpc.c (revision 0aaa8977)
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 && kvmppc_hv_ops->hash_v3_possible &&
615 		       kvmppc_hv_ops->hash_v3_possible());
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 = min_t(unsigned int, num_present_cpus(), KVM_MAX_VCPUS);
645 		else
646 			r = min_t(unsigned int, num_online_cpus(), KVM_MAX_VCPUS);
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_IDS;
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 	case KVM_CAP_PPC_DAWR1:
682 		r = !!(hv_enabled && kvmppc_hv_ops->enable_dawr1 &&
683 		       !kvmppc_hv_ops->enable_dawr1(NULL));
684 		break;
685 	case KVM_CAP_PPC_RPT_INVALIDATE:
686 		r = 1;
687 		break;
688 #endif
689 	default:
690 		r = 0;
691 		break;
692 	}
693 	return r;
694 
695 }
696 
697 long kvm_arch_dev_ioctl(struct file *filp,
698                         unsigned int ioctl, unsigned long arg)
699 {
700 	return -EINVAL;
701 }
702 
703 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
704 {
705 	kvmppc_core_free_memslot(kvm, slot);
706 }
707 
708 int kvm_arch_prepare_memory_region(struct kvm *kvm,
709 				   struct kvm_memory_slot *memslot,
710 				   const struct kvm_userspace_memory_region *mem,
711 				   enum kvm_mr_change change)
712 {
713 	return kvmppc_core_prepare_memory_region(kvm, memslot, mem, change);
714 }
715 
716 void kvm_arch_commit_memory_region(struct kvm *kvm,
717 				   const struct kvm_userspace_memory_region *mem,
718 				   struct kvm_memory_slot *old,
719 				   const struct kvm_memory_slot *new,
720 				   enum kvm_mr_change change)
721 {
722 	kvmppc_core_commit_memory_region(kvm, mem, old, new, change);
723 }
724 
725 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
726 				   struct kvm_memory_slot *slot)
727 {
728 	kvmppc_core_flush_memslot(kvm, slot);
729 }
730 
731 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
732 {
733 	return 0;
734 }
735 
736 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
737 {
738 	struct kvm_vcpu *vcpu;
739 
740 	vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
741 	kvmppc_decrementer_func(vcpu);
742 
743 	return HRTIMER_NORESTART;
744 }
745 
746 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
747 {
748 	int err;
749 
750 	hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
751 	vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
752 	vcpu->arch.dec_expires = get_tb();
753 
754 #ifdef CONFIG_KVM_EXIT_TIMING
755 	mutex_init(&vcpu->arch.exit_timing_lock);
756 #endif
757 	err = kvmppc_subarch_vcpu_init(vcpu);
758 	if (err)
759 		return err;
760 
761 	err = kvmppc_core_vcpu_create(vcpu);
762 	if (err)
763 		goto out_vcpu_uninit;
764 
765 	vcpu->arch.waitp = &vcpu->wait;
766 	kvmppc_create_vcpu_debugfs(vcpu, vcpu->vcpu_id);
767 	return 0;
768 
769 out_vcpu_uninit:
770 	kvmppc_subarch_vcpu_uninit(vcpu);
771 	return err;
772 }
773 
774 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
775 {
776 }
777 
778 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
779 {
780 	/* Make sure we're not using the vcpu anymore */
781 	hrtimer_cancel(&vcpu->arch.dec_timer);
782 
783 	kvmppc_remove_vcpu_debugfs(vcpu);
784 
785 	switch (vcpu->arch.irq_type) {
786 	case KVMPPC_IRQ_MPIC:
787 		kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
788 		break;
789 	case KVMPPC_IRQ_XICS:
790 		if (xics_on_xive())
791 			kvmppc_xive_cleanup_vcpu(vcpu);
792 		else
793 			kvmppc_xics_free_icp(vcpu);
794 		break;
795 	case KVMPPC_IRQ_XIVE:
796 		kvmppc_xive_native_cleanup_vcpu(vcpu);
797 		break;
798 	}
799 
800 	kvmppc_core_vcpu_free(vcpu);
801 
802 	kvmppc_subarch_vcpu_uninit(vcpu);
803 }
804 
805 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
806 {
807 	return kvmppc_core_pending_dec(vcpu);
808 }
809 
810 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
811 {
812 #ifdef CONFIG_BOOKE
813 	/*
814 	 * vrsave (formerly usprg0) isn't used by Linux, but may
815 	 * be used by the guest.
816 	 *
817 	 * On non-booke this is associated with Altivec and
818 	 * is handled by code in book3s.c.
819 	 */
820 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
821 #endif
822 	kvmppc_core_vcpu_load(vcpu, cpu);
823 }
824 
825 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
826 {
827 	kvmppc_core_vcpu_put(vcpu);
828 #ifdef CONFIG_BOOKE
829 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
830 #endif
831 }
832 
833 /*
834  * irq_bypass_add_producer and irq_bypass_del_producer are only
835  * useful if the architecture supports PCI passthrough.
836  * irq_bypass_stop and irq_bypass_start are not needed and so
837  * kvm_ops are not defined for them.
838  */
839 bool kvm_arch_has_irq_bypass(void)
840 {
841 	return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
842 		(kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
843 }
844 
845 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
846 				     struct irq_bypass_producer *prod)
847 {
848 	struct kvm_kernel_irqfd *irqfd =
849 		container_of(cons, struct kvm_kernel_irqfd, consumer);
850 	struct kvm *kvm = irqfd->kvm;
851 
852 	if (kvm->arch.kvm_ops->irq_bypass_add_producer)
853 		return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
854 
855 	return 0;
856 }
857 
858 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
859 				      struct irq_bypass_producer *prod)
860 {
861 	struct kvm_kernel_irqfd *irqfd =
862 		container_of(cons, struct kvm_kernel_irqfd, consumer);
863 	struct kvm *kvm = irqfd->kvm;
864 
865 	if (kvm->arch.kvm_ops->irq_bypass_del_producer)
866 		kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
867 }
868 
869 #ifdef CONFIG_VSX
870 static inline int kvmppc_get_vsr_dword_offset(int index)
871 {
872 	int offset;
873 
874 	if ((index != 0) && (index != 1))
875 		return -1;
876 
877 #ifdef __BIG_ENDIAN
878 	offset =  index;
879 #else
880 	offset = 1 - index;
881 #endif
882 
883 	return offset;
884 }
885 
886 static inline int kvmppc_get_vsr_word_offset(int index)
887 {
888 	int offset;
889 
890 	if ((index > 3) || (index < 0))
891 		return -1;
892 
893 #ifdef __BIG_ENDIAN
894 	offset = index;
895 #else
896 	offset = 3 - index;
897 #endif
898 	return offset;
899 }
900 
901 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
902 	u64 gpr)
903 {
904 	union kvmppc_one_reg val;
905 	int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
906 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
907 
908 	if (offset == -1)
909 		return;
910 
911 	if (index >= 32) {
912 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
913 		val.vsxval[offset] = gpr;
914 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
915 	} else {
916 		VCPU_VSX_FPR(vcpu, index, offset) = gpr;
917 	}
918 }
919 
920 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
921 	u64 gpr)
922 {
923 	union kvmppc_one_reg val;
924 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
925 
926 	if (index >= 32) {
927 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
928 		val.vsxval[0] = gpr;
929 		val.vsxval[1] = gpr;
930 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
931 	} else {
932 		VCPU_VSX_FPR(vcpu, index, 0) = gpr;
933 		VCPU_VSX_FPR(vcpu, index, 1) = gpr;
934 	}
935 }
936 
937 static inline void kvmppc_set_vsr_word_dump(struct kvm_vcpu *vcpu,
938 	u32 gpr)
939 {
940 	union kvmppc_one_reg val;
941 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
942 
943 	if (index >= 32) {
944 		val.vsx32val[0] = gpr;
945 		val.vsx32val[1] = gpr;
946 		val.vsx32val[2] = gpr;
947 		val.vsx32val[3] = gpr;
948 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
949 	} else {
950 		val.vsx32val[0] = gpr;
951 		val.vsx32val[1] = gpr;
952 		VCPU_VSX_FPR(vcpu, index, 0) = val.vsxval[0];
953 		VCPU_VSX_FPR(vcpu, index, 1) = val.vsxval[0];
954 	}
955 }
956 
957 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
958 	u32 gpr32)
959 {
960 	union kvmppc_one_reg val;
961 	int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
962 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
963 	int dword_offset, word_offset;
964 
965 	if (offset == -1)
966 		return;
967 
968 	if (index >= 32) {
969 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
970 		val.vsx32val[offset] = gpr32;
971 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
972 	} else {
973 		dword_offset = offset / 2;
974 		word_offset = offset % 2;
975 		val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
976 		val.vsx32val[word_offset] = gpr32;
977 		VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
978 	}
979 }
980 #endif /* CONFIG_VSX */
981 
982 #ifdef CONFIG_ALTIVEC
983 static inline int kvmppc_get_vmx_offset_generic(struct kvm_vcpu *vcpu,
984 		int index, int element_size)
985 {
986 	int offset;
987 	int elts = sizeof(vector128)/element_size;
988 
989 	if ((index < 0) || (index >= elts))
990 		return -1;
991 
992 	if (kvmppc_need_byteswap(vcpu))
993 		offset = elts - index - 1;
994 	else
995 		offset = index;
996 
997 	return offset;
998 }
999 
1000 static inline int kvmppc_get_vmx_dword_offset(struct kvm_vcpu *vcpu,
1001 		int index)
1002 {
1003 	return kvmppc_get_vmx_offset_generic(vcpu, index, 8);
1004 }
1005 
1006 static inline int kvmppc_get_vmx_word_offset(struct kvm_vcpu *vcpu,
1007 		int index)
1008 {
1009 	return kvmppc_get_vmx_offset_generic(vcpu, index, 4);
1010 }
1011 
1012 static inline int kvmppc_get_vmx_hword_offset(struct kvm_vcpu *vcpu,
1013 		int index)
1014 {
1015 	return kvmppc_get_vmx_offset_generic(vcpu, index, 2);
1016 }
1017 
1018 static inline int kvmppc_get_vmx_byte_offset(struct kvm_vcpu *vcpu,
1019 		int index)
1020 {
1021 	return kvmppc_get_vmx_offset_generic(vcpu, index, 1);
1022 }
1023 
1024 
1025 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu,
1026 	u64 gpr)
1027 {
1028 	union kvmppc_one_reg val;
1029 	int offset = kvmppc_get_vmx_dword_offset(vcpu,
1030 			vcpu->arch.mmio_vmx_offset);
1031 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1032 
1033 	if (offset == -1)
1034 		return;
1035 
1036 	val.vval = VCPU_VSX_VR(vcpu, index);
1037 	val.vsxval[offset] = gpr;
1038 	VCPU_VSX_VR(vcpu, index) = val.vval;
1039 }
1040 
1041 static inline void kvmppc_set_vmx_word(struct kvm_vcpu *vcpu,
1042 	u32 gpr32)
1043 {
1044 	union kvmppc_one_reg val;
1045 	int offset = kvmppc_get_vmx_word_offset(vcpu,
1046 			vcpu->arch.mmio_vmx_offset);
1047 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1048 
1049 	if (offset == -1)
1050 		return;
1051 
1052 	val.vval = VCPU_VSX_VR(vcpu, index);
1053 	val.vsx32val[offset] = gpr32;
1054 	VCPU_VSX_VR(vcpu, index) = val.vval;
1055 }
1056 
1057 static inline void kvmppc_set_vmx_hword(struct kvm_vcpu *vcpu,
1058 	u16 gpr16)
1059 {
1060 	union kvmppc_one_reg val;
1061 	int offset = kvmppc_get_vmx_hword_offset(vcpu,
1062 			vcpu->arch.mmio_vmx_offset);
1063 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1064 
1065 	if (offset == -1)
1066 		return;
1067 
1068 	val.vval = VCPU_VSX_VR(vcpu, index);
1069 	val.vsx16val[offset] = gpr16;
1070 	VCPU_VSX_VR(vcpu, index) = val.vval;
1071 }
1072 
1073 static inline void kvmppc_set_vmx_byte(struct kvm_vcpu *vcpu,
1074 	u8 gpr8)
1075 {
1076 	union kvmppc_one_reg val;
1077 	int offset = kvmppc_get_vmx_byte_offset(vcpu,
1078 			vcpu->arch.mmio_vmx_offset);
1079 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1080 
1081 	if (offset == -1)
1082 		return;
1083 
1084 	val.vval = VCPU_VSX_VR(vcpu, index);
1085 	val.vsx8val[offset] = gpr8;
1086 	VCPU_VSX_VR(vcpu, index) = val.vval;
1087 }
1088 #endif /* CONFIG_ALTIVEC */
1089 
1090 #ifdef CONFIG_PPC_FPU
1091 static inline u64 sp_to_dp(u32 fprs)
1092 {
1093 	u64 fprd;
1094 
1095 	preempt_disable();
1096 	enable_kernel_fp();
1097 	asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m<>" (fprd) : "m<>" (fprs)
1098 	     : "fr0");
1099 	preempt_enable();
1100 	return fprd;
1101 }
1102 
1103 static inline u32 dp_to_sp(u64 fprd)
1104 {
1105 	u32 fprs;
1106 
1107 	preempt_disable();
1108 	enable_kernel_fp();
1109 	asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m<>" (fprs) : "m<>" (fprd)
1110 	     : "fr0");
1111 	preempt_enable();
1112 	return fprs;
1113 }
1114 
1115 #else
1116 #define sp_to_dp(x)	(x)
1117 #define dp_to_sp(x)	(x)
1118 #endif /* CONFIG_PPC_FPU */
1119 
1120 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu)
1121 {
1122 	struct kvm_run *run = vcpu->run;
1123 	u64 gpr;
1124 
1125 	if (run->mmio.len > sizeof(gpr)) {
1126 		printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
1127 		return;
1128 	}
1129 
1130 	if (!vcpu->arch.mmio_host_swabbed) {
1131 		switch (run->mmio.len) {
1132 		case 8: gpr = *(u64 *)run->mmio.data; break;
1133 		case 4: gpr = *(u32 *)run->mmio.data; break;
1134 		case 2: gpr = *(u16 *)run->mmio.data; break;
1135 		case 1: gpr = *(u8 *)run->mmio.data; break;
1136 		}
1137 	} else {
1138 		switch (run->mmio.len) {
1139 		case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
1140 		case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
1141 		case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
1142 		case 1: gpr = *(u8 *)run->mmio.data; break;
1143 		}
1144 	}
1145 
1146 	/* conversion between single and double precision */
1147 	if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
1148 		gpr = sp_to_dp(gpr);
1149 
1150 	if (vcpu->arch.mmio_sign_extend) {
1151 		switch (run->mmio.len) {
1152 #ifdef CONFIG_PPC64
1153 		case 4:
1154 			gpr = (s64)(s32)gpr;
1155 			break;
1156 #endif
1157 		case 2:
1158 			gpr = (s64)(s16)gpr;
1159 			break;
1160 		case 1:
1161 			gpr = (s64)(s8)gpr;
1162 			break;
1163 		}
1164 	}
1165 
1166 	switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1167 	case KVM_MMIO_REG_GPR:
1168 		kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1169 		break;
1170 	case KVM_MMIO_REG_FPR:
1171 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1172 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_FP);
1173 
1174 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1175 		break;
1176 #ifdef CONFIG_PPC_BOOK3S
1177 	case KVM_MMIO_REG_QPR:
1178 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1179 		break;
1180 	case KVM_MMIO_REG_FQPR:
1181 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1182 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1183 		break;
1184 #endif
1185 #ifdef CONFIG_VSX
1186 	case KVM_MMIO_REG_VSX:
1187 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1188 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VSX);
1189 
1190 		if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_DWORD)
1191 			kvmppc_set_vsr_dword(vcpu, gpr);
1192 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_WORD)
1193 			kvmppc_set_vsr_word(vcpu, gpr);
1194 		else if (vcpu->arch.mmio_copy_type ==
1195 				KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1196 			kvmppc_set_vsr_dword_dump(vcpu, gpr);
1197 		else if (vcpu->arch.mmio_copy_type ==
1198 				KVMPPC_VSX_COPY_WORD_LOAD_DUMP)
1199 			kvmppc_set_vsr_word_dump(vcpu, gpr);
1200 		break;
1201 #endif
1202 #ifdef CONFIG_ALTIVEC
1203 	case KVM_MMIO_REG_VMX:
1204 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1205 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VEC);
1206 
1207 		if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_DWORD)
1208 			kvmppc_set_vmx_dword(vcpu, gpr);
1209 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_WORD)
1210 			kvmppc_set_vmx_word(vcpu, gpr);
1211 		else if (vcpu->arch.mmio_copy_type ==
1212 				KVMPPC_VMX_COPY_HWORD)
1213 			kvmppc_set_vmx_hword(vcpu, gpr);
1214 		else if (vcpu->arch.mmio_copy_type ==
1215 				KVMPPC_VMX_COPY_BYTE)
1216 			kvmppc_set_vmx_byte(vcpu, gpr);
1217 		break;
1218 #endif
1219 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1220 	case KVM_MMIO_REG_NESTED_GPR:
1221 		if (kvmppc_need_byteswap(vcpu))
1222 			gpr = swab64(gpr);
1223 		kvm_vcpu_write_guest(vcpu, vcpu->arch.nested_io_gpr, &gpr,
1224 				     sizeof(gpr));
1225 		break;
1226 #endif
1227 	default:
1228 		BUG();
1229 	}
1230 }
1231 
1232 static int __kvmppc_handle_load(struct kvm_vcpu *vcpu,
1233 				unsigned int rt, unsigned int bytes,
1234 				int is_default_endian, int sign_extend)
1235 {
1236 	struct kvm_run *run = vcpu->run;
1237 	int idx, ret;
1238 	bool host_swabbed;
1239 
1240 	/* Pity C doesn't have a logical XOR operator */
1241 	if (kvmppc_need_byteswap(vcpu)) {
1242 		host_swabbed = is_default_endian;
1243 	} else {
1244 		host_swabbed = !is_default_endian;
1245 	}
1246 
1247 	if (bytes > sizeof(run->mmio.data)) {
1248 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1249 		       run->mmio.len);
1250 	}
1251 
1252 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1253 	run->mmio.len = bytes;
1254 	run->mmio.is_write = 0;
1255 
1256 	vcpu->arch.io_gpr = rt;
1257 	vcpu->arch.mmio_host_swabbed = host_swabbed;
1258 	vcpu->mmio_needed = 1;
1259 	vcpu->mmio_is_write = 0;
1260 	vcpu->arch.mmio_sign_extend = sign_extend;
1261 
1262 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1263 
1264 	ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1265 			      bytes, &run->mmio.data);
1266 
1267 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1268 
1269 	if (!ret) {
1270 		kvmppc_complete_mmio_load(vcpu);
1271 		vcpu->mmio_needed = 0;
1272 		return EMULATE_DONE;
1273 	}
1274 
1275 	return EMULATE_DO_MMIO;
1276 }
1277 
1278 int kvmppc_handle_load(struct kvm_vcpu *vcpu,
1279 		       unsigned int rt, unsigned int bytes,
1280 		       int is_default_endian)
1281 {
1282 	return __kvmppc_handle_load(vcpu, rt, bytes, is_default_endian, 0);
1283 }
1284 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1285 
1286 /* Same as above, but sign extends */
1287 int kvmppc_handle_loads(struct kvm_vcpu *vcpu,
1288 			unsigned int rt, unsigned int bytes,
1289 			int is_default_endian)
1290 {
1291 	return __kvmppc_handle_load(vcpu, rt, bytes, is_default_endian, 1);
1292 }
1293 
1294 #ifdef CONFIG_VSX
1295 int kvmppc_handle_vsx_load(struct kvm_vcpu *vcpu,
1296 			unsigned int rt, unsigned int bytes,
1297 			int is_default_endian, int mmio_sign_extend)
1298 {
1299 	enum emulation_result emulated = EMULATE_DONE;
1300 
1301 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1302 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1303 		return EMULATE_FAIL;
1304 
1305 	while (vcpu->arch.mmio_vsx_copy_nums) {
1306 		emulated = __kvmppc_handle_load(vcpu, rt, bytes,
1307 			is_default_endian, mmio_sign_extend);
1308 
1309 		if (emulated != EMULATE_DONE)
1310 			break;
1311 
1312 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1313 
1314 		vcpu->arch.mmio_vsx_copy_nums--;
1315 		vcpu->arch.mmio_vsx_offset++;
1316 	}
1317 	return emulated;
1318 }
1319 #endif /* CONFIG_VSX */
1320 
1321 int kvmppc_handle_store(struct kvm_vcpu *vcpu,
1322 			u64 val, unsigned int bytes, int is_default_endian)
1323 {
1324 	struct kvm_run *run = vcpu->run;
1325 	void *data = run->mmio.data;
1326 	int idx, ret;
1327 	bool host_swabbed;
1328 
1329 	/* Pity C doesn't have a logical XOR operator */
1330 	if (kvmppc_need_byteswap(vcpu)) {
1331 		host_swabbed = is_default_endian;
1332 	} else {
1333 		host_swabbed = !is_default_endian;
1334 	}
1335 
1336 	if (bytes > sizeof(run->mmio.data)) {
1337 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1338 		       run->mmio.len);
1339 	}
1340 
1341 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1342 	run->mmio.len = bytes;
1343 	run->mmio.is_write = 1;
1344 	vcpu->mmio_needed = 1;
1345 	vcpu->mmio_is_write = 1;
1346 
1347 	if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1348 		val = dp_to_sp(val);
1349 
1350 	/* Store the value at the lowest bytes in 'data'. */
1351 	if (!host_swabbed) {
1352 		switch (bytes) {
1353 		case 8: *(u64 *)data = val; break;
1354 		case 4: *(u32 *)data = val; break;
1355 		case 2: *(u16 *)data = val; break;
1356 		case 1: *(u8  *)data = val; break;
1357 		}
1358 	} else {
1359 		switch (bytes) {
1360 		case 8: *(u64 *)data = swab64(val); break;
1361 		case 4: *(u32 *)data = swab32(val); break;
1362 		case 2: *(u16 *)data = swab16(val); break;
1363 		case 1: *(u8  *)data = val; break;
1364 		}
1365 	}
1366 
1367 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1368 
1369 	ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1370 			       bytes, &run->mmio.data);
1371 
1372 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1373 
1374 	if (!ret) {
1375 		vcpu->mmio_needed = 0;
1376 		return EMULATE_DONE;
1377 	}
1378 
1379 	return EMULATE_DO_MMIO;
1380 }
1381 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1382 
1383 #ifdef CONFIG_VSX
1384 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1385 {
1386 	u32 dword_offset, word_offset;
1387 	union kvmppc_one_reg reg;
1388 	int vsx_offset = 0;
1389 	int copy_type = vcpu->arch.mmio_copy_type;
1390 	int result = 0;
1391 
1392 	switch (copy_type) {
1393 	case KVMPPC_VSX_COPY_DWORD:
1394 		vsx_offset =
1395 			kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1396 
1397 		if (vsx_offset == -1) {
1398 			result = -1;
1399 			break;
1400 		}
1401 
1402 		if (rs < 32) {
1403 			*val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1404 		} else {
1405 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1406 			*val = reg.vsxval[vsx_offset];
1407 		}
1408 		break;
1409 
1410 	case KVMPPC_VSX_COPY_WORD:
1411 		vsx_offset =
1412 			kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1413 
1414 		if (vsx_offset == -1) {
1415 			result = -1;
1416 			break;
1417 		}
1418 
1419 		if (rs < 32) {
1420 			dword_offset = vsx_offset / 2;
1421 			word_offset = vsx_offset % 2;
1422 			reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1423 			*val = reg.vsx32val[word_offset];
1424 		} else {
1425 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1426 			*val = reg.vsx32val[vsx_offset];
1427 		}
1428 		break;
1429 
1430 	default:
1431 		result = -1;
1432 		break;
1433 	}
1434 
1435 	return result;
1436 }
1437 
1438 int kvmppc_handle_vsx_store(struct kvm_vcpu *vcpu,
1439 			int rs, unsigned int bytes, int is_default_endian)
1440 {
1441 	u64 val;
1442 	enum emulation_result emulated = EMULATE_DONE;
1443 
1444 	vcpu->arch.io_gpr = rs;
1445 
1446 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1447 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1448 		return EMULATE_FAIL;
1449 
1450 	while (vcpu->arch.mmio_vsx_copy_nums) {
1451 		if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1452 			return EMULATE_FAIL;
1453 
1454 		emulated = kvmppc_handle_store(vcpu,
1455 			 val, bytes, is_default_endian);
1456 
1457 		if (emulated != EMULATE_DONE)
1458 			break;
1459 
1460 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1461 
1462 		vcpu->arch.mmio_vsx_copy_nums--;
1463 		vcpu->arch.mmio_vsx_offset++;
1464 	}
1465 
1466 	return emulated;
1467 }
1468 
1469 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu)
1470 {
1471 	struct kvm_run *run = vcpu->run;
1472 	enum emulation_result emulated = EMULATE_FAIL;
1473 	int r;
1474 
1475 	vcpu->arch.paddr_accessed += run->mmio.len;
1476 
1477 	if (!vcpu->mmio_is_write) {
1478 		emulated = kvmppc_handle_vsx_load(vcpu, vcpu->arch.io_gpr,
1479 			 run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1480 	} else {
1481 		emulated = kvmppc_handle_vsx_store(vcpu,
1482 			 vcpu->arch.io_gpr, run->mmio.len, 1);
1483 	}
1484 
1485 	switch (emulated) {
1486 	case EMULATE_DO_MMIO:
1487 		run->exit_reason = KVM_EXIT_MMIO;
1488 		r = RESUME_HOST;
1489 		break;
1490 	case EMULATE_FAIL:
1491 		pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1492 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1493 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1494 		r = RESUME_HOST;
1495 		break;
1496 	default:
1497 		r = RESUME_GUEST;
1498 		break;
1499 	}
1500 	return r;
1501 }
1502 #endif /* CONFIG_VSX */
1503 
1504 #ifdef CONFIG_ALTIVEC
1505 int kvmppc_handle_vmx_load(struct kvm_vcpu *vcpu,
1506 		unsigned int rt, unsigned int bytes, int is_default_endian)
1507 {
1508 	enum emulation_result emulated = EMULATE_DONE;
1509 
1510 	if (vcpu->arch.mmio_vsx_copy_nums > 2)
1511 		return EMULATE_FAIL;
1512 
1513 	while (vcpu->arch.mmio_vmx_copy_nums) {
1514 		emulated = __kvmppc_handle_load(vcpu, rt, bytes,
1515 				is_default_endian, 0);
1516 
1517 		if (emulated != EMULATE_DONE)
1518 			break;
1519 
1520 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1521 		vcpu->arch.mmio_vmx_copy_nums--;
1522 		vcpu->arch.mmio_vmx_offset++;
1523 	}
1524 
1525 	return emulated;
1526 }
1527 
1528 static int kvmppc_get_vmx_dword(struct kvm_vcpu *vcpu, int index, u64 *val)
1529 {
1530 	union kvmppc_one_reg reg;
1531 	int vmx_offset = 0;
1532 	int result = 0;
1533 
1534 	vmx_offset =
1535 		kvmppc_get_vmx_dword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1536 
1537 	if (vmx_offset == -1)
1538 		return -1;
1539 
1540 	reg.vval = VCPU_VSX_VR(vcpu, index);
1541 	*val = reg.vsxval[vmx_offset];
1542 
1543 	return result;
1544 }
1545 
1546 static int kvmppc_get_vmx_word(struct kvm_vcpu *vcpu, int index, u64 *val)
1547 {
1548 	union kvmppc_one_reg reg;
1549 	int vmx_offset = 0;
1550 	int result = 0;
1551 
1552 	vmx_offset =
1553 		kvmppc_get_vmx_word_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1554 
1555 	if (vmx_offset == -1)
1556 		return -1;
1557 
1558 	reg.vval = VCPU_VSX_VR(vcpu, index);
1559 	*val = reg.vsx32val[vmx_offset];
1560 
1561 	return result;
1562 }
1563 
1564 static int kvmppc_get_vmx_hword(struct kvm_vcpu *vcpu, int index, u64 *val)
1565 {
1566 	union kvmppc_one_reg reg;
1567 	int vmx_offset = 0;
1568 	int result = 0;
1569 
1570 	vmx_offset =
1571 		kvmppc_get_vmx_hword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1572 
1573 	if (vmx_offset == -1)
1574 		return -1;
1575 
1576 	reg.vval = VCPU_VSX_VR(vcpu, index);
1577 	*val = reg.vsx16val[vmx_offset];
1578 
1579 	return result;
1580 }
1581 
1582 static int kvmppc_get_vmx_byte(struct kvm_vcpu *vcpu, int index, u64 *val)
1583 {
1584 	union kvmppc_one_reg reg;
1585 	int vmx_offset = 0;
1586 	int result = 0;
1587 
1588 	vmx_offset =
1589 		kvmppc_get_vmx_byte_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1590 
1591 	if (vmx_offset == -1)
1592 		return -1;
1593 
1594 	reg.vval = VCPU_VSX_VR(vcpu, index);
1595 	*val = reg.vsx8val[vmx_offset];
1596 
1597 	return result;
1598 }
1599 
1600 int kvmppc_handle_vmx_store(struct kvm_vcpu *vcpu,
1601 		unsigned int rs, unsigned int bytes, int is_default_endian)
1602 {
1603 	u64 val = 0;
1604 	unsigned int index = rs & KVM_MMIO_REG_MASK;
1605 	enum emulation_result emulated = EMULATE_DONE;
1606 
1607 	if (vcpu->arch.mmio_vsx_copy_nums > 2)
1608 		return EMULATE_FAIL;
1609 
1610 	vcpu->arch.io_gpr = rs;
1611 
1612 	while (vcpu->arch.mmio_vmx_copy_nums) {
1613 		switch (vcpu->arch.mmio_copy_type) {
1614 		case KVMPPC_VMX_COPY_DWORD:
1615 			if (kvmppc_get_vmx_dword(vcpu, index, &val) == -1)
1616 				return EMULATE_FAIL;
1617 
1618 			break;
1619 		case KVMPPC_VMX_COPY_WORD:
1620 			if (kvmppc_get_vmx_word(vcpu, index, &val) == -1)
1621 				return EMULATE_FAIL;
1622 			break;
1623 		case KVMPPC_VMX_COPY_HWORD:
1624 			if (kvmppc_get_vmx_hword(vcpu, index, &val) == -1)
1625 				return EMULATE_FAIL;
1626 			break;
1627 		case KVMPPC_VMX_COPY_BYTE:
1628 			if (kvmppc_get_vmx_byte(vcpu, index, &val) == -1)
1629 				return EMULATE_FAIL;
1630 			break;
1631 		default:
1632 			return EMULATE_FAIL;
1633 		}
1634 
1635 		emulated = kvmppc_handle_store(vcpu, val, bytes,
1636 				is_default_endian);
1637 		if (emulated != EMULATE_DONE)
1638 			break;
1639 
1640 		vcpu->arch.paddr_accessed += vcpu->run->mmio.len;
1641 		vcpu->arch.mmio_vmx_copy_nums--;
1642 		vcpu->arch.mmio_vmx_offset++;
1643 	}
1644 
1645 	return emulated;
1646 }
1647 
1648 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu)
1649 {
1650 	struct kvm_run *run = vcpu->run;
1651 	enum emulation_result emulated = EMULATE_FAIL;
1652 	int r;
1653 
1654 	vcpu->arch.paddr_accessed += run->mmio.len;
1655 
1656 	if (!vcpu->mmio_is_write) {
1657 		emulated = kvmppc_handle_vmx_load(vcpu,
1658 				vcpu->arch.io_gpr, run->mmio.len, 1);
1659 	} else {
1660 		emulated = kvmppc_handle_vmx_store(vcpu,
1661 				vcpu->arch.io_gpr, run->mmio.len, 1);
1662 	}
1663 
1664 	switch (emulated) {
1665 	case EMULATE_DO_MMIO:
1666 		run->exit_reason = KVM_EXIT_MMIO;
1667 		r = RESUME_HOST;
1668 		break;
1669 	case EMULATE_FAIL:
1670 		pr_info("KVM: MMIO emulation failed (VMX repeat)\n");
1671 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1672 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1673 		r = RESUME_HOST;
1674 		break;
1675 	default:
1676 		r = RESUME_GUEST;
1677 		break;
1678 	}
1679 	return r;
1680 }
1681 #endif /* CONFIG_ALTIVEC */
1682 
1683 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1684 {
1685 	int r = 0;
1686 	union kvmppc_one_reg val;
1687 	int size;
1688 
1689 	size = one_reg_size(reg->id);
1690 	if (size > sizeof(val))
1691 		return -EINVAL;
1692 
1693 	r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1694 	if (r == -EINVAL) {
1695 		r = 0;
1696 		switch (reg->id) {
1697 #ifdef CONFIG_ALTIVEC
1698 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1699 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1700 				r = -ENXIO;
1701 				break;
1702 			}
1703 			val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1704 			break;
1705 		case KVM_REG_PPC_VSCR:
1706 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1707 				r = -ENXIO;
1708 				break;
1709 			}
1710 			val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1711 			break;
1712 		case KVM_REG_PPC_VRSAVE:
1713 			val = get_reg_val(reg->id, vcpu->arch.vrsave);
1714 			break;
1715 #endif /* CONFIG_ALTIVEC */
1716 		default:
1717 			r = -EINVAL;
1718 			break;
1719 		}
1720 	}
1721 
1722 	if (r)
1723 		return r;
1724 
1725 	if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1726 		r = -EFAULT;
1727 
1728 	return r;
1729 }
1730 
1731 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1732 {
1733 	int r;
1734 	union kvmppc_one_reg val;
1735 	int size;
1736 
1737 	size = one_reg_size(reg->id);
1738 	if (size > sizeof(val))
1739 		return -EINVAL;
1740 
1741 	if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1742 		return -EFAULT;
1743 
1744 	r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1745 	if (r == -EINVAL) {
1746 		r = 0;
1747 		switch (reg->id) {
1748 #ifdef CONFIG_ALTIVEC
1749 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1750 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1751 				r = -ENXIO;
1752 				break;
1753 			}
1754 			vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1755 			break;
1756 		case KVM_REG_PPC_VSCR:
1757 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1758 				r = -ENXIO;
1759 				break;
1760 			}
1761 			vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1762 			break;
1763 		case KVM_REG_PPC_VRSAVE:
1764 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1765 				r = -ENXIO;
1766 				break;
1767 			}
1768 			vcpu->arch.vrsave = set_reg_val(reg->id, val);
1769 			break;
1770 #endif /* CONFIG_ALTIVEC */
1771 		default:
1772 			r = -EINVAL;
1773 			break;
1774 		}
1775 	}
1776 
1777 	return r;
1778 }
1779 
1780 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
1781 {
1782 	struct kvm_run *run = vcpu->run;
1783 	int r;
1784 
1785 	vcpu_load(vcpu);
1786 
1787 	if (vcpu->mmio_needed) {
1788 		vcpu->mmio_needed = 0;
1789 		if (!vcpu->mmio_is_write)
1790 			kvmppc_complete_mmio_load(vcpu);
1791 #ifdef CONFIG_VSX
1792 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1793 			vcpu->arch.mmio_vsx_copy_nums--;
1794 			vcpu->arch.mmio_vsx_offset++;
1795 		}
1796 
1797 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1798 			r = kvmppc_emulate_mmio_vsx_loadstore(vcpu);
1799 			if (r == RESUME_HOST) {
1800 				vcpu->mmio_needed = 1;
1801 				goto out;
1802 			}
1803 		}
1804 #endif
1805 #ifdef CONFIG_ALTIVEC
1806 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1807 			vcpu->arch.mmio_vmx_copy_nums--;
1808 			vcpu->arch.mmio_vmx_offset++;
1809 		}
1810 
1811 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1812 			r = kvmppc_emulate_mmio_vmx_loadstore(vcpu);
1813 			if (r == RESUME_HOST) {
1814 				vcpu->mmio_needed = 1;
1815 				goto out;
1816 			}
1817 		}
1818 #endif
1819 	} else if (vcpu->arch.osi_needed) {
1820 		u64 *gprs = run->osi.gprs;
1821 		int i;
1822 
1823 		for (i = 0; i < 32; i++)
1824 			kvmppc_set_gpr(vcpu, i, gprs[i]);
1825 		vcpu->arch.osi_needed = 0;
1826 	} else if (vcpu->arch.hcall_needed) {
1827 		int i;
1828 
1829 		kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1830 		for (i = 0; i < 9; ++i)
1831 			kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1832 		vcpu->arch.hcall_needed = 0;
1833 #ifdef CONFIG_BOOKE
1834 	} else if (vcpu->arch.epr_needed) {
1835 		kvmppc_set_epr(vcpu, run->epr.epr);
1836 		vcpu->arch.epr_needed = 0;
1837 #endif
1838 	}
1839 
1840 	kvm_sigset_activate(vcpu);
1841 
1842 	if (run->immediate_exit)
1843 		r = -EINTR;
1844 	else
1845 		r = kvmppc_vcpu_run(vcpu);
1846 
1847 	kvm_sigset_deactivate(vcpu);
1848 
1849 #ifdef CONFIG_ALTIVEC
1850 out:
1851 #endif
1852 	vcpu_put(vcpu);
1853 	return r;
1854 }
1855 
1856 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1857 {
1858 	if (irq->irq == KVM_INTERRUPT_UNSET) {
1859 		kvmppc_core_dequeue_external(vcpu);
1860 		return 0;
1861 	}
1862 
1863 	kvmppc_core_queue_external(vcpu, irq);
1864 
1865 	kvm_vcpu_kick(vcpu);
1866 
1867 	return 0;
1868 }
1869 
1870 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1871 				     struct kvm_enable_cap *cap)
1872 {
1873 	int r;
1874 
1875 	if (cap->flags)
1876 		return -EINVAL;
1877 
1878 	switch (cap->cap) {
1879 	case KVM_CAP_PPC_OSI:
1880 		r = 0;
1881 		vcpu->arch.osi_enabled = true;
1882 		break;
1883 	case KVM_CAP_PPC_PAPR:
1884 		r = 0;
1885 		vcpu->arch.papr_enabled = true;
1886 		break;
1887 	case KVM_CAP_PPC_EPR:
1888 		r = 0;
1889 		if (cap->args[0])
1890 			vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1891 		else
1892 			vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1893 		break;
1894 #ifdef CONFIG_BOOKE
1895 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
1896 		r = 0;
1897 		vcpu->arch.watchdog_enabled = true;
1898 		break;
1899 #endif
1900 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1901 	case KVM_CAP_SW_TLB: {
1902 		struct kvm_config_tlb cfg;
1903 		void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1904 
1905 		r = -EFAULT;
1906 		if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1907 			break;
1908 
1909 		r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1910 		break;
1911 	}
1912 #endif
1913 #ifdef CONFIG_KVM_MPIC
1914 	case KVM_CAP_IRQ_MPIC: {
1915 		struct fd f;
1916 		struct kvm_device *dev;
1917 
1918 		r = -EBADF;
1919 		f = fdget(cap->args[0]);
1920 		if (!f.file)
1921 			break;
1922 
1923 		r = -EPERM;
1924 		dev = kvm_device_from_filp(f.file);
1925 		if (dev)
1926 			r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1927 
1928 		fdput(f);
1929 		break;
1930 	}
1931 #endif
1932 #ifdef CONFIG_KVM_XICS
1933 	case KVM_CAP_IRQ_XICS: {
1934 		struct fd f;
1935 		struct kvm_device *dev;
1936 
1937 		r = -EBADF;
1938 		f = fdget(cap->args[0]);
1939 		if (!f.file)
1940 			break;
1941 
1942 		r = -EPERM;
1943 		dev = kvm_device_from_filp(f.file);
1944 		if (dev) {
1945 			if (xics_on_xive())
1946 				r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1947 			else
1948 				r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1949 		}
1950 
1951 		fdput(f);
1952 		break;
1953 	}
1954 #endif /* CONFIG_KVM_XICS */
1955 #ifdef CONFIG_KVM_XIVE
1956 	case KVM_CAP_PPC_IRQ_XIVE: {
1957 		struct fd f;
1958 		struct kvm_device *dev;
1959 
1960 		r = -EBADF;
1961 		f = fdget(cap->args[0]);
1962 		if (!f.file)
1963 			break;
1964 
1965 		r = -ENXIO;
1966 		if (!xive_enabled())
1967 			break;
1968 
1969 		r = -EPERM;
1970 		dev = kvm_device_from_filp(f.file);
1971 		if (dev)
1972 			r = kvmppc_xive_native_connect_vcpu(dev, vcpu,
1973 							    cap->args[1]);
1974 
1975 		fdput(f);
1976 		break;
1977 	}
1978 #endif /* CONFIG_KVM_XIVE */
1979 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1980 	case KVM_CAP_PPC_FWNMI:
1981 		r = -EINVAL;
1982 		if (!is_kvmppc_hv_enabled(vcpu->kvm))
1983 			break;
1984 		r = 0;
1985 		vcpu->kvm->arch.fwnmi_enabled = true;
1986 		break;
1987 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1988 	default:
1989 		r = -EINVAL;
1990 		break;
1991 	}
1992 
1993 	if (!r)
1994 		r = kvmppc_sanity_check(vcpu);
1995 
1996 	return r;
1997 }
1998 
1999 bool kvm_arch_intc_initialized(struct kvm *kvm)
2000 {
2001 #ifdef CONFIG_KVM_MPIC
2002 	if (kvm->arch.mpic)
2003 		return true;
2004 #endif
2005 #ifdef CONFIG_KVM_XICS
2006 	if (kvm->arch.xics || kvm->arch.xive)
2007 		return true;
2008 #endif
2009 	return false;
2010 }
2011 
2012 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2013                                     struct kvm_mp_state *mp_state)
2014 {
2015 	return -EINVAL;
2016 }
2017 
2018 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2019                                     struct kvm_mp_state *mp_state)
2020 {
2021 	return -EINVAL;
2022 }
2023 
2024 long kvm_arch_vcpu_async_ioctl(struct file *filp,
2025 			       unsigned int ioctl, unsigned long arg)
2026 {
2027 	struct kvm_vcpu *vcpu = filp->private_data;
2028 	void __user *argp = (void __user *)arg;
2029 
2030 	if (ioctl == KVM_INTERRUPT) {
2031 		struct kvm_interrupt irq;
2032 		if (copy_from_user(&irq, argp, sizeof(irq)))
2033 			return -EFAULT;
2034 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2035 	}
2036 	return -ENOIOCTLCMD;
2037 }
2038 
2039 long kvm_arch_vcpu_ioctl(struct file *filp,
2040                          unsigned int ioctl, unsigned long arg)
2041 {
2042 	struct kvm_vcpu *vcpu = filp->private_data;
2043 	void __user *argp = (void __user *)arg;
2044 	long r;
2045 
2046 	switch (ioctl) {
2047 	case KVM_ENABLE_CAP:
2048 	{
2049 		struct kvm_enable_cap cap;
2050 		r = -EFAULT;
2051 		if (copy_from_user(&cap, argp, sizeof(cap)))
2052 			goto out;
2053 		vcpu_load(vcpu);
2054 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
2055 		vcpu_put(vcpu);
2056 		break;
2057 	}
2058 
2059 	case KVM_SET_ONE_REG:
2060 	case KVM_GET_ONE_REG:
2061 	{
2062 		struct kvm_one_reg reg;
2063 		r = -EFAULT;
2064 		if (copy_from_user(&reg, argp, sizeof(reg)))
2065 			goto out;
2066 		if (ioctl == KVM_SET_ONE_REG)
2067 			r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
2068 		else
2069 			r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
2070 		break;
2071 	}
2072 
2073 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
2074 	case KVM_DIRTY_TLB: {
2075 		struct kvm_dirty_tlb dirty;
2076 		r = -EFAULT;
2077 		if (copy_from_user(&dirty, argp, sizeof(dirty)))
2078 			goto out;
2079 		vcpu_load(vcpu);
2080 		r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
2081 		vcpu_put(vcpu);
2082 		break;
2083 	}
2084 #endif
2085 	default:
2086 		r = -EINVAL;
2087 	}
2088 
2089 out:
2090 	return r;
2091 }
2092 
2093 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2094 {
2095 	return VM_FAULT_SIGBUS;
2096 }
2097 
2098 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
2099 {
2100 	u32 inst_nop = 0x60000000;
2101 #ifdef CONFIG_KVM_BOOKE_HV
2102 	u32 inst_sc1 = 0x44000022;
2103 	pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
2104 	pvinfo->hcall[1] = cpu_to_be32(inst_nop);
2105 	pvinfo->hcall[2] = cpu_to_be32(inst_nop);
2106 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2107 #else
2108 	u32 inst_lis = 0x3c000000;
2109 	u32 inst_ori = 0x60000000;
2110 	u32 inst_sc = 0x44000002;
2111 	u32 inst_imm_mask = 0xffff;
2112 
2113 	/*
2114 	 * The hypercall to get into KVM from within guest context is as
2115 	 * follows:
2116 	 *
2117 	 *    lis r0, r0, KVM_SC_MAGIC_R0@h
2118 	 *    ori r0, KVM_SC_MAGIC_R0@l
2119 	 *    sc
2120 	 *    nop
2121 	 */
2122 	pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
2123 	pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
2124 	pvinfo->hcall[2] = cpu_to_be32(inst_sc);
2125 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2126 #endif
2127 
2128 	pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
2129 
2130 	return 0;
2131 }
2132 
2133 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
2134 			  bool line_status)
2135 {
2136 	if (!irqchip_in_kernel(kvm))
2137 		return -ENXIO;
2138 
2139 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2140 					irq_event->irq, irq_event->level,
2141 					line_status);
2142 	return 0;
2143 }
2144 
2145 
2146 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
2147 			    struct kvm_enable_cap *cap)
2148 {
2149 	int r;
2150 
2151 	if (cap->flags)
2152 		return -EINVAL;
2153 
2154 	switch (cap->cap) {
2155 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
2156 	case KVM_CAP_PPC_ENABLE_HCALL: {
2157 		unsigned long hcall = cap->args[0];
2158 
2159 		r = -EINVAL;
2160 		if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
2161 		    cap->args[1] > 1)
2162 			break;
2163 		if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
2164 			break;
2165 		if (cap->args[1])
2166 			set_bit(hcall / 4, kvm->arch.enabled_hcalls);
2167 		else
2168 			clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
2169 		r = 0;
2170 		break;
2171 	}
2172 	case KVM_CAP_PPC_SMT: {
2173 		unsigned long mode = cap->args[0];
2174 		unsigned long flags = cap->args[1];
2175 
2176 		r = -EINVAL;
2177 		if (kvm->arch.kvm_ops->set_smt_mode)
2178 			r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
2179 		break;
2180 	}
2181 
2182 	case KVM_CAP_PPC_NESTED_HV:
2183 		r = -EINVAL;
2184 		if (!is_kvmppc_hv_enabled(kvm) ||
2185 		    !kvm->arch.kvm_ops->enable_nested)
2186 			break;
2187 		r = kvm->arch.kvm_ops->enable_nested(kvm);
2188 		break;
2189 #endif
2190 #if defined(CONFIG_KVM_BOOK3S_HV_POSSIBLE)
2191 	case KVM_CAP_PPC_SECURE_GUEST:
2192 		r = -EINVAL;
2193 		if (!is_kvmppc_hv_enabled(kvm) || !kvm->arch.kvm_ops->enable_svm)
2194 			break;
2195 		r = kvm->arch.kvm_ops->enable_svm(kvm);
2196 		break;
2197 	case KVM_CAP_PPC_DAWR1:
2198 		r = -EINVAL;
2199 		if (!is_kvmppc_hv_enabled(kvm) || !kvm->arch.kvm_ops->enable_dawr1)
2200 			break;
2201 		r = kvm->arch.kvm_ops->enable_dawr1(kvm);
2202 		break;
2203 #endif
2204 	default:
2205 		r = -EINVAL;
2206 		break;
2207 	}
2208 
2209 	return r;
2210 }
2211 
2212 #ifdef CONFIG_PPC_BOOK3S_64
2213 /*
2214  * These functions check whether the underlying hardware is safe
2215  * against attacks based on observing the effects of speculatively
2216  * executed instructions, and whether it supplies instructions for
2217  * use in workarounds.  The information comes from firmware, either
2218  * via the device tree on powernv platforms or from an hcall on
2219  * pseries platforms.
2220  */
2221 #ifdef CONFIG_PPC_PSERIES
2222 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2223 {
2224 	struct h_cpu_char_result c;
2225 	unsigned long rc;
2226 
2227 	if (!machine_is(pseries))
2228 		return -ENOTTY;
2229 
2230 	rc = plpar_get_cpu_characteristics(&c);
2231 	if (rc == H_SUCCESS) {
2232 		cp->character = c.character;
2233 		cp->behaviour = c.behaviour;
2234 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2235 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2236 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2237 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2238 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2239 			KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
2240 			KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
2241 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2242 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2243 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2244 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2245 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2246 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2247 	}
2248 	return 0;
2249 }
2250 #else
2251 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2252 {
2253 	return -ENOTTY;
2254 }
2255 #endif
2256 
2257 static inline bool have_fw_feat(struct device_node *fw_features,
2258 				const char *state, const char *name)
2259 {
2260 	struct device_node *np;
2261 	bool r = false;
2262 
2263 	np = of_get_child_by_name(fw_features, name);
2264 	if (np) {
2265 		r = of_property_read_bool(np, state);
2266 		of_node_put(np);
2267 	}
2268 	return r;
2269 }
2270 
2271 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2272 {
2273 	struct device_node *np, *fw_features;
2274 	int r;
2275 
2276 	memset(cp, 0, sizeof(*cp));
2277 	r = pseries_get_cpu_char(cp);
2278 	if (r != -ENOTTY)
2279 		return r;
2280 
2281 	np = of_find_node_by_name(NULL, "ibm,opal");
2282 	if (np) {
2283 		fw_features = of_get_child_by_name(np, "fw-features");
2284 		of_node_put(np);
2285 		if (!fw_features)
2286 			return 0;
2287 		if (have_fw_feat(fw_features, "enabled",
2288 				 "inst-spec-barrier-ori31,31,0"))
2289 			cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
2290 		if (have_fw_feat(fw_features, "enabled",
2291 				 "fw-bcctrl-serialized"))
2292 			cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
2293 		if (have_fw_feat(fw_features, "enabled",
2294 				 "inst-l1d-flush-ori30,30,0"))
2295 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
2296 		if (have_fw_feat(fw_features, "enabled",
2297 				 "inst-l1d-flush-trig2"))
2298 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
2299 		if (have_fw_feat(fw_features, "enabled",
2300 				 "fw-l1d-thread-split"))
2301 			cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
2302 		if (have_fw_feat(fw_features, "enabled",
2303 				 "fw-count-cache-disabled"))
2304 			cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2305 		if (have_fw_feat(fw_features, "enabled",
2306 				 "fw-count-cache-flush-bcctr2,0,0"))
2307 			cp->character |= KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2308 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2309 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2310 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2311 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2312 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2313 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2314 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2315 
2316 		if (have_fw_feat(fw_features, "enabled",
2317 				 "speculation-policy-favor-security"))
2318 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
2319 		if (!have_fw_feat(fw_features, "disabled",
2320 				  "needs-l1d-flush-msr-pr-0-to-1"))
2321 			cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
2322 		if (!have_fw_feat(fw_features, "disabled",
2323 				  "needs-spec-barrier-for-bound-checks"))
2324 			cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2325 		if (have_fw_feat(fw_features, "enabled",
2326 				 "needs-count-cache-flush-on-context-switch"))
2327 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2328 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2329 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2330 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2331 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2332 
2333 		of_node_put(fw_features);
2334 	}
2335 
2336 	return 0;
2337 }
2338 #endif
2339 
2340 long kvm_arch_vm_ioctl(struct file *filp,
2341                        unsigned int ioctl, unsigned long arg)
2342 {
2343 	struct kvm *kvm __maybe_unused = filp->private_data;
2344 	void __user *argp = (void __user *)arg;
2345 	long r;
2346 
2347 	switch (ioctl) {
2348 	case KVM_PPC_GET_PVINFO: {
2349 		struct kvm_ppc_pvinfo pvinfo;
2350 		memset(&pvinfo, 0, sizeof(pvinfo));
2351 		r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
2352 		if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
2353 			r = -EFAULT;
2354 			goto out;
2355 		}
2356 
2357 		break;
2358 	}
2359 #ifdef CONFIG_SPAPR_TCE_IOMMU
2360 	case KVM_CREATE_SPAPR_TCE_64: {
2361 		struct kvm_create_spapr_tce_64 create_tce_64;
2362 
2363 		r = -EFAULT;
2364 		if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
2365 			goto out;
2366 		if (create_tce_64.flags) {
2367 			r = -EINVAL;
2368 			goto out;
2369 		}
2370 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2371 		goto out;
2372 	}
2373 	case KVM_CREATE_SPAPR_TCE: {
2374 		struct kvm_create_spapr_tce create_tce;
2375 		struct kvm_create_spapr_tce_64 create_tce_64;
2376 
2377 		r = -EFAULT;
2378 		if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
2379 			goto out;
2380 
2381 		create_tce_64.liobn = create_tce.liobn;
2382 		create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
2383 		create_tce_64.offset = 0;
2384 		create_tce_64.size = create_tce.window_size >>
2385 				IOMMU_PAGE_SHIFT_4K;
2386 		create_tce_64.flags = 0;
2387 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2388 		goto out;
2389 	}
2390 #endif
2391 #ifdef CONFIG_PPC_BOOK3S_64
2392 	case KVM_PPC_GET_SMMU_INFO: {
2393 		struct kvm_ppc_smmu_info info;
2394 		struct kvm *kvm = filp->private_data;
2395 
2396 		memset(&info, 0, sizeof(info));
2397 		r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
2398 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2399 			r = -EFAULT;
2400 		break;
2401 	}
2402 	case KVM_PPC_RTAS_DEFINE_TOKEN: {
2403 		struct kvm *kvm = filp->private_data;
2404 
2405 		r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
2406 		break;
2407 	}
2408 	case KVM_PPC_CONFIGURE_V3_MMU: {
2409 		struct kvm *kvm = filp->private_data;
2410 		struct kvm_ppc_mmuv3_cfg cfg;
2411 
2412 		r = -EINVAL;
2413 		if (!kvm->arch.kvm_ops->configure_mmu)
2414 			goto out;
2415 		r = -EFAULT;
2416 		if (copy_from_user(&cfg, argp, sizeof(cfg)))
2417 			goto out;
2418 		r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
2419 		break;
2420 	}
2421 	case KVM_PPC_GET_RMMU_INFO: {
2422 		struct kvm *kvm = filp->private_data;
2423 		struct kvm_ppc_rmmu_info info;
2424 
2425 		r = -EINVAL;
2426 		if (!kvm->arch.kvm_ops->get_rmmu_info)
2427 			goto out;
2428 		r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
2429 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2430 			r = -EFAULT;
2431 		break;
2432 	}
2433 	case KVM_PPC_GET_CPU_CHAR: {
2434 		struct kvm_ppc_cpu_char cpuchar;
2435 
2436 		r = kvmppc_get_cpu_char(&cpuchar);
2437 		if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2438 			r = -EFAULT;
2439 		break;
2440 	}
2441 	case KVM_PPC_SVM_OFF: {
2442 		struct kvm *kvm = filp->private_data;
2443 
2444 		r = 0;
2445 		if (!kvm->arch.kvm_ops->svm_off)
2446 			goto out;
2447 
2448 		r = kvm->arch.kvm_ops->svm_off(kvm);
2449 		break;
2450 	}
2451 	default: {
2452 		struct kvm *kvm = filp->private_data;
2453 		r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
2454 	}
2455 #else /* CONFIG_PPC_BOOK3S_64 */
2456 	default:
2457 		r = -ENOTTY;
2458 #endif
2459 	}
2460 out:
2461 	return r;
2462 }
2463 
2464 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
2465 static unsigned long nr_lpids;
2466 
2467 long kvmppc_alloc_lpid(void)
2468 {
2469 	long lpid;
2470 
2471 	do {
2472 		lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
2473 		if (lpid >= nr_lpids) {
2474 			pr_err("%s: No LPIDs free\n", __func__);
2475 			return -ENOMEM;
2476 		}
2477 	} while (test_and_set_bit(lpid, lpid_inuse));
2478 
2479 	return lpid;
2480 }
2481 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2482 
2483 void kvmppc_claim_lpid(long lpid)
2484 {
2485 	set_bit(lpid, lpid_inuse);
2486 }
2487 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
2488 
2489 void kvmppc_free_lpid(long lpid)
2490 {
2491 	clear_bit(lpid, lpid_inuse);
2492 }
2493 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2494 
2495 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2496 {
2497 	nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
2498 	memset(lpid_inuse, 0, sizeof(lpid_inuse));
2499 }
2500 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2501 
2502 int kvm_arch_init(void *opaque)
2503 {
2504 	return 0;
2505 }
2506 
2507 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);
2508