xref: /openbmc/linux/arch/powerpc/kvm/powerpc.c (revision fb8d6c8d)
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 
35 #include "timing.h"
36 #include "irq.h"
37 #include "../mm/mmu_decl.h"
38 
39 #define CREATE_TRACE_POINTS
40 #include "trace.h"
41 
42 struct kvmppc_ops *kvmppc_hv_ops;
43 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
44 struct kvmppc_ops *kvmppc_pr_ops;
45 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
46 
47 
48 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
49 {
50 	return !!(v->arch.pending_exceptions) || kvm_request_pending(v);
51 }
52 
53 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
54 {
55 	return kvm_arch_vcpu_runnable(vcpu);
56 }
57 
58 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
59 {
60 	return false;
61 }
62 
63 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
64 {
65 	return 1;
66 }
67 
68 /*
69  * Common checks before entering the guest world.  Call with interrupts
70  * disabled.
71  *
72  * returns:
73  *
74  * == 1 if we're ready to go into guest state
75  * <= 0 if we need to go back to the host with return value
76  */
77 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
78 {
79 	int r;
80 
81 	WARN_ON(irqs_disabled());
82 	hard_irq_disable();
83 
84 	while (true) {
85 		if (need_resched()) {
86 			local_irq_enable();
87 			cond_resched();
88 			hard_irq_disable();
89 			continue;
90 		}
91 
92 		if (signal_pending(current)) {
93 			kvmppc_account_exit(vcpu, SIGNAL_EXITS);
94 			vcpu->run->exit_reason = KVM_EXIT_INTR;
95 			r = -EINTR;
96 			break;
97 		}
98 
99 		vcpu->mode = IN_GUEST_MODE;
100 
101 		/*
102 		 * Reading vcpu->requests must happen after setting vcpu->mode,
103 		 * so we don't miss a request because the requester sees
104 		 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
105 		 * before next entering the guest (and thus doesn't IPI).
106 		 * This also orders the write to mode from any reads
107 		 * to the page tables done while the VCPU is running.
108 		 * Please see the comment in kvm_flush_remote_tlbs.
109 		 */
110 		smp_mb();
111 
112 		if (kvm_request_pending(vcpu)) {
113 			/* Make sure we process requests preemptable */
114 			local_irq_enable();
115 			trace_kvm_check_requests(vcpu);
116 			r = kvmppc_core_check_requests(vcpu);
117 			hard_irq_disable();
118 			if (r > 0)
119 				continue;
120 			break;
121 		}
122 
123 		if (kvmppc_core_prepare_to_enter(vcpu)) {
124 			/* interrupts got enabled in between, so we
125 			   are back at square 1 */
126 			continue;
127 		}
128 
129 		guest_enter_irqoff();
130 		return 1;
131 	}
132 
133 	/* return to host */
134 	local_irq_enable();
135 	return r;
136 }
137 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
138 
139 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
140 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
141 {
142 	struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
143 	int i;
144 
145 	shared->sprg0 = swab64(shared->sprg0);
146 	shared->sprg1 = swab64(shared->sprg1);
147 	shared->sprg2 = swab64(shared->sprg2);
148 	shared->sprg3 = swab64(shared->sprg3);
149 	shared->srr0 = swab64(shared->srr0);
150 	shared->srr1 = swab64(shared->srr1);
151 	shared->dar = swab64(shared->dar);
152 	shared->msr = swab64(shared->msr);
153 	shared->dsisr = swab32(shared->dsisr);
154 	shared->int_pending = swab32(shared->int_pending);
155 	for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
156 		shared->sr[i] = swab32(shared->sr[i]);
157 }
158 #endif
159 
160 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
161 {
162 	int nr = kvmppc_get_gpr(vcpu, 11);
163 	int r;
164 	unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
165 	unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
166 	unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
167 	unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
168 	unsigned long r2 = 0;
169 
170 	if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
171 		/* 32 bit mode */
172 		param1 &= 0xffffffff;
173 		param2 &= 0xffffffff;
174 		param3 &= 0xffffffff;
175 		param4 &= 0xffffffff;
176 	}
177 
178 	switch (nr) {
179 	case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
180 	{
181 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
182 		/* Book3S can be little endian, find it out here */
183 		int shared_big_endian = true;
184 		if (vcpu->arch.intr_msr & MSR_LE)
185 			shared_big_endian = false;
186 		if (shared_big_endian != vcpu->arch.shared_big_endian)
187 			kvmppc_swab_shared(vcpu);
188 		vcpu->arch.shared_big_endian = shared_big_endian;
189 #endif
190 
191 		if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
192 			/*
193 			 * Older versions of the Linux magic page code had
194 			 * a bug where they would map their trampoline code
195 			 * NX. If that's the case, remove !PR NX capability.
196 			 */
197 			vcpu->arch.disable_kernel_nx = true;
198 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
199 		}
200 
201 		vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
202 		vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
203 
204 #ifdef CONFIG_PPC_64K_PAGES
205 		/*
206 		 * Make sure our 4k magic page is in the same window of a 64k
207 		 * page within the guest and within the host's page.
208 		 */
209 		if ((vcpu->arch.magic_page_pa & 0xf000) !=
210 		    ((ulong)vcpu->arch.shared & 0xf000)) {
211 			void *old_shared = vcpu->arch.shared;
212 			ulong shared = (ulong)vcpu->arch.shared;
213 			void *new_shared;
214 
215 			shared &= PAGE_MASK;
216 			shared |= vcpu->arch.magic_page_pa & 0xf000;
217 			new_shared = (void*)shared;
218 			memcpy(new_shared, old_shared, 0x1000);
219 			vcpu->arch.shared = new_shared;
220 		}
221 #endif
222 
223 		r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
224 
225 		r = EV_SUCCESS;
226 		break;
227 	}
228 	case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
229 		r = EV_SUCCESS;
230 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
231 		r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
232 #endif
233 
234 		/* Second return value is in r4 */
235 		break;
236 	case EV_HCALL_TOKEN(EV_IDLE):
237 		r = EV_SUCCESS;
238 		kvm_vcpu_block(vcpu);
239 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
240 		break;
241 	default:
242 		r = EV_UNIMPLEMENTED;
243 		break;
244 	}
245 
246 	kvmppc_set_gpr(vcpu, 4, r2);
247 
248 	return r;
249 }
250 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
251 
252 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
253 {
254 	int r = false;
255 
256 	/* We have to know what CPU to virtualize */
257 	if (!vcpu->arch.pvr)
258 		goto out;
259 
260 	/* PAPR only works with book3s_64 */
261 	if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
262 		goto out;
263 
264 	/* HV KVM can only do PAPR mode for now */
265 	if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
266 		goto out;
267 
268 #ifdef CONFIG_KVM_BOOKE_HV
269 	if (!cpu_has_feature(CPU_FTR_EMB_HV))
270 		goto out;
271 #endif
272 
273 	r = true;
274 
275 out:
276 	vcpu->arch.sane = r;
277 	return r ? 0 : -EINVAL;
278 }
279 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
280 
281 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
282 {
283 	enum emulation_result er;
284 	int r;
285 
286 	er = kvmppc_emulate_loadstore(vcpu);
287 	switch (er) {
288 	case EMULATE_DONE:
289 		/* Future optimization: only reload non-volatiles if they were
290 		 * actually modified. */
291 		r = RESUME_GUEST_NV;
292 		break;
293 	case EMULATE_AGAIN:
294 		r = RESUME_GUEST;
295 		break;
296 	case EMULATE_DO_MMIO:
297 		run->exit_reason = KVM_EXIT_MMIO;
298 		/* We must reload nonvolatiles because "update" load/store
299 		 * instructions modify register state. */
300 		/* Future optimization: only reload non-volatiles if they were
301 		 * actually modified. */
302 		r = RESUME_HOST_NV;
303 		break;
304 	case EMULATE_FAIL:
305 	{
306 		u32 last_inst;
307 
308 		kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
309 		/* XXX Deliver Program interrupt to guest. */
310 		pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
311 		r = RESUME_HOST;
312 		break;
313 	}
314 	default:
315 		WARN_ON(1);
316 		r = RESUME_GUEST;
317 	}
318 
319 	return r;
320 }
321 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
322 
323 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
324 	      bool data)
325 {
326 	ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
327 	struct kvmppc_pte pte;
328 	int r = -EINVAL;
329 
330 	vcpu->stat.st++;
331 
332 	if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->store_to_eaddr)
333 		r = vcpu->kvm->arch.kvm_ops->store_to_eaddr(vcpu, eaddr, ptr,
334 							    size);
335 
336 	if ((!r) || (r == -EAGAIN))
337 		return r;
338 
339 	r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
340 			 XLATE_WRITE, &pte);
341 	if (r < 0)
342 		return r;
343 
344 	*eaddr = pte.raddr;
345 
346 	if (!pte.may_write)
347 		return -EPERM;
348 
349 	/* Magic page override */
350 	if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
351 	    ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
352 	    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
353 		void *magic = vcpu->arch.shared;
354 		magic += pte.eaddr & 0xfff;
355 		memcpy(magic, ptr, size);
356 		return EMULATE_DONE;
357 	}
358 
359 	if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
360 		return EMULATE_DO_MMIO;
361 
362 	return EMULATE_DONE;
363 }
364 EXPORT_SYMBOL_GPL(kvmppc_st);
365 
366 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
367 		      bool data)
368 {
369 	ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
370 	struct kvmppc_pte pte;
371 	int rc = -EINVAL;
372 
373 	vcpu->stat.ld++;
374 
375 	if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->load_from_eaddr)
376 		rc = vcpu->kvm->arch.kvm_ops->load_from_eaddr(vcpu, eaddr, ptr,
377 							      size);
378 
379 	if ((!rc) || (rc == -EAGAIN))
380 		return rc;
381 
382 	rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
383 			  XLATE_READ, &pte);
384 	if (rc)
385 		return rc;
386 
387 	*eaddr = pte.raddr;
388 
389 	if (!pte.may_read)
390 		return -EPERM;
391 
392 	if (!data && !pte.may_execute)
393 		return -ENOEXEC;
394 
395 	/* Magic page override */
396 	if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
397 	    ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
398 	    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
399 		void *magic = vcpu->arch.shared;
400 		magic += pte.eaddr & 0xfff;
401 		memcpy(ptr, magic, size);
402 		return EMULATE_DONE;
403 	}
404 
405 	if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
406 		return EMULATE_DO_MMIO;
407 
408 	return EMULATE_DONE;
409 }
410 EXPORT_SYMBOL_GPL(kvmppc_ld);
411 
412 int kvm_arch_hardware_enable(void)
413 {
414 	return 0;
415 }
416 
417 int kvm_arch_hardware_setup(void)
418 {
419 	return 0;
420 }
421 
422 int kvm_arch_check_processor_compat(void)
423 {
424 	return kvmppc_core_check_processor_compat();
425 }
426 
427 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
428 {
429 	struct kvmppc_ops *kvm_ops = NULL;
430 	/*
431 	 * if we have both HV and PR enabled, default is HV
432 	 */
433 	if (type == 0) {
434 		if (kvmppc_hv_ops)
435 			kvm_ops = kvmppc_hv_ops;
436 		else
437 			kvm_ops = kvmppc_pr_ops;
438 		if (!kvm_ops)
439 			goto err_out;
440 	} else	if (type == KVM_VM_PPC_HV) {
441 		if (!kvmppc_hv_ops)
442 			goto err_out;
443 		kvm_ops = kvmppc_hv_ops;
444 	} else if (type == KVM_VM_PPC_PR) {
445 		if (!kvmppc_pr_ops)
446 			goto err_out;
447 		kvm_ops = kvmppc_pr_ops;
448 	} else
449 		goto err_out;
450 
451 	if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
452 		return -ENOENT;
453 
454 	kvm->arch.kvm_ops = kvm_ops;
455 	return kvmppc_core_init_vm(kvm);
456 err_out:
457 	return -EINVAL;
458 }
459 
460 void kvm_arch_destroy_vm(struct kvm *kvm)
461 {
462 	unsigned int i;
463 	struct kvm_vcpu *vcpu;
464 
465 #ifdef CONFIG_KVM_XICS
466 	/*
467 	 * We call kick_all_cpus_sync() to ensure that all
468 	 * CPUs have executed any pending IPIs before we
469 	 * continue and free VCPUs structures below.
470 	 */
471 	if (is_kvmppc_hv_enabled(kvm))
472 		kick_all_cpus_sync();
473 #endif
474 
475 	kvm_for_each_vcpu(i, vcpu, kvm)
476 		kvm_arch_vcpu_free(vcpu);
477 
478 	mutex_lock(&kvm->lock);
479 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
480 		kvm->vcpus[i] = NULL;
481 
482 	atomic_set(&kvm->online_vcpus, 0);
483 
484 	kvmppc_core_destroy_vm(kvm);
485 
486 	mutex_unlock(&kvm->lock);
487 
488 	/* drop the module reference */
489 	module_put(kvm->arch.kvm_ops->owner);
490 }
491 
492 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
493 {
494 	int r;
495 	/* Assume we're using HV mode when the HV module is loaded */
496 	int hv_enabled = kvmppc_hv_ops ? 1 : 0;
497 
498 	if (kvm) {
499 		/*
500 		 * Hooray - we know which VM type we're running on. Depend on
501 		 * that rather than the guess above.
502 		 */
503 		hv_enabled = is_kvmppc_hv_enabled(kvm);
504 	}
505 
506 	switch (ext) {
507 #ifdef CONFIG_BOOKE
508 	case KVM_CAP_PPC_BOOKE_SREGS:
509 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
510 	case KVM_CAP_PPC_EPR:
511 #else
512 	case KVM_CAP_PPC_SEGSTATE:
513 	case KVM_CAP_PPC_HIOR:
514 	case KVM_CAP_PPC_PAPR:
515 #endif
516 	case KVM_CAP_PPC_UNSET_IRQ:
517 	case KVM_CAP_PPC_IRQ_LEVEL:
518 	case KVM_CAP_ENABLE_CAP:
519 	case KVM_CAP_ONE_REG:
520 	case KVM_CAP_IOEVENTFD:
521 	case KVM_CAP_DEVICE_CTRL:
522 	case KVM_CAP_IMMEDIATE_EXIT:
523 		r = 1;
524 		break;
525 	case KVM_CAP_PPC_PAIRED_SINGLES:
526 	case KVM_CAP_PPC_OSI:
527 	case KVM_CAP_PPC_GET_PVINFO:
528 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
529 	case KVM_CAP_SW_TLB:
530 #endif
531 		/* We support this only for PR */
532 		r = !hv_enabled;
533 		break;
534 #ifdef CONFIG_KVM_MPIC
535 	case KVM_CAP_IRQ_MPIC:
536 		r = 1;
537 		break;
538 #endif
539 
540 #ifdef CONFIG_PPC_BOOK3S_64
541 	case KVM_CAP_SPAPR_TCE:
542 	case KVM_CAP_SPAPR_TCE_64:
543 		r = 1;
544 		break;
545 	case KVM_CAP_SPAPR_TCE_VFIO:
546 		r = !!cpu_has_feature(CPU_FTR_HVMODE);
547 		break;
548 	case KVM_CAP_PPC_RTAS:
549 	case KVM_CAP_PPC_FIXUP_HCALL:
550 	case KVM_CAP_PPC_ENABLE_HCALL:
551 #ifdef CONFIG_KVM_XICS
552 	case KVM_CAP_IRQ_XICS:
553 #endif
554 	case KVM_CAP_PPC_GET_CPU_CHAR:
555 		r = 1;
556 		break;
557 #ifdef CONFIG_KVM_XIVE
558 	case KVM_CAP_PPC_IRQ_XIVE:
559 		/*
560 		 * We need XIVE to be enabled on the platform (implies
561 		 * a POWER9 processor) and the PowerNV platform, as
562 		 * nested is not yet supported.
563 		 */
564 		r = xive_enabled() && !!cpu_has_feature(CPU_FTR_HVMODE) &&
565 			kvmppc_xive_native_supported();
566 		break;
567 #endif
568 
569 	case KVM_CAP_PPC_ALLOC_HTAB:
570 		r = hv_enabled;
571 		break;
572 #endif /* CONFIG_PPC_BOOK3S_64 */
573 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
574 	case KVM_CAP_PPC_SMT:
575 		r = 0;
576 		if (kvm) {
577 			if (kvm->arch.emul_smt_mode > 1)
578 				r = kvm->arch.emul_smt_mode;
579 			else
580 				r = kvm->arch.smt_mode;
581 		} else if (hv_enabled) {
582 			if (cpu_has_feature(CPU_FTR_ARCH_300))
583 				r = 1;
584 			else
585 				r = threads_per_subcore;
586 		}
587 		break;
588 	case KVM_CAP_PPC_SMT_POSSIBLE:
589 		r = 1;
590 		if (hv_enabled) {
591 			if (!cpu_has_feature(CPU_FTR_ARCH_300))
592 				r = ((threads_per_subcore << 1) - 1);
593 			else
594 				/* P9 can emulate dbells, so allow any mode */
595 				r = 8 | 4 | 2 | 1;
596 		}
597 		break;
598 	case KVM_CAP_PPC_RMA:
599 		r = 0;
600 		break;
601 	case KVM_CAP_PPC_HWRNG:
602 		r = kvmppc_hwrng_present();
603 		break;
604 	case KVM_CAP_PPC_MMU_RADIX:
605 		r = !!(hv_enabled && radix_enabled());
606 		break;
607 	case KVM_CAP_PPC_MMU_HASH_V3:
608 		r = !!(hv_enabled && cpu_has_feature(CPU_FTR_ARCH_300) &&
609 		       cpu_has_feature(CPU_FTR_HVMODE));
610 		break;
611 	case KVM_CAP_PPC_NESTED_HV:
612 		r = !!(hv_enabled && kvmppc_hv_ops->enable_nested &&
613 		       !kvmppc_hv_ops->enable_nested(NULL));
614 		break;
615 #endif
616 	case KVM_CAP_SYNC_MMU:
617 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
618 		r = hv_enabled;
619 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
620 		r = 1;
621 #else
622 		r = 0;
623 #endif
624 		break;
625 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
626 	case KVM_CAP_PPC_HTAB_FD:
627 		r = hv_enabled;
628 		break;
629 #endif
630 	case KVM_CAP_NR_VCPUS:
631 		/*
632 		 * Recommending a number of CPUs is somewhat arbitrary; we
633 		 * return the number of present CPUs for -HV (since a host
634 		 * will have secondary threads "offline"), and for other KVM
635 		 * implementations just count online CPUs.
636 		 */
637 		if (hv_enabled)
638 			r = num_present_cpus();
639 		else
640 			r = num_online_cpus();
641 		break;
642 	case KVM_CAP_MAX_VCPUS:
643 		r = KVM_MAX_VCPUS;
644 		break;
645 	case KVM_CAP_MAX_VCPU_ID:
646 		r = KVM_MAX_VCPU_ID;
647 		break;
648 #ifdef CONFIG_PPC_BOOK3S_64
649 	case KVM_CAP_PPC_GET_SMMU_INFO:
650 		r = 1;
651 		break;
652 	case KVM_CAP_SPAPR_MULTITCE:
653 		r = 1;
654 		break;
655 	case KVM_CAP_SPAPR_RESIZE_HPT:
656 		r = !!hv_enabled;
657 		break;
658 #endif
659 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
660 	case KVM_CAP_PPC_FWNMI:
661 		r = hv_enabled;
662 		break;
663 #endif
664 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
665 	case KVM_CAP_PPC_HTM:
666 		r = !!(cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM) ||
667 		     (hv_enabled && cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST));
668 		break;
669 #endif
670 	default:
671 		r = 0;
672 		break;
673 	}
674 	return r;
675 
676 }
677 
678 long kvm_arch_dev_ioctl(struct file *filp,
679                         unsigned int ioctl, unsigned long arg)
680 {
681 	return -EINVAL;
682 }
683 
684 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
685 			   struct kvm_memory_slot *dont)
686 {
687 	kvmppc_core_free_memslot(kvm, free, dont);
688 }
689 
690 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
691 			    unsigned long npages)
692 {
693 	return kvmppc_core_create_memslot(kvm, slot, npages);
694 }
695 
696 int kvm_arch_prepare_memory_region(struct kvm *kvm,
697 				   struct kvm_memory_slot *memslot,
698 				   const struct kvm_userspace_memory_region *mem,
699 				   enum kvm_mr_change change)
700 {
701 	return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
702 }
703 
704 void kvm_arch_commit_memory_region(struct kvm *kvm,
705 				   const struct kvm_userspace_memory_region *mem,
706 				   const struct kvm_memory_slot *old,
707 				   const struct kvm_memory_slot *new,
708 				   enum kvm_mr_change change)
709 {
710 	kvmppc_core_commit_memory_region(kvm, mem, old, new, change);
711 }
712 
713 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
714 				   struct kvm_memory_slot *slot)
715 {
716 	kvmppc_core_flush_memslot(kvm, slot);
717 }
718 
719 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
720 {
721 	struct kvm_vcpu *vcpu;
722 	vcpu = kvmppc_core_vcpu_create(kvm, id);
723 	if (!IS_ERR(vcpu)) {
724 		vcpu->arch.wqp = &vcpu->wq;
725 		kvmppc_create_vcpu_debugfs(vcpu, id);
726 	}
727 	return vcpu;
728 }
729 
730 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
731 {
732 }
733 
734 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
735 {
736 	/* Make sure we're not using the vcpu anymore */
737 	hrtimer_cancel(&vcpu->arch.dec_timer);
738 
739 	kvmppc_remove_vcpu_debugfs(vcpu);
740 
741 	switch (vcpu->arch.irq_type) {
742 	case KVMPPC_IRQ_MPIC:
743 		kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
744 		break;
745 	case KVMPPC_IRQ_XICS:
746 		if (xics_on_xive())
747 			kvmppc_xive_cleanup_vcpu(vcpu);
748 		else
749 			kvmppc_xics_free_icp(vcpu);
750 		break;
751 	case KVMPPC_IRQ_XIVE:
752 		kvmppc_xive_native_cleanup_vcpu(vcpu);
753 		break;
754 	}
755 
756 	kvmppc_core_vcpu_free(vcpu);
757 }
758 
759 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
760 {
761 	kvm_arch_vcpu_free(vcpu);
762 }
763 
764 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
765 {
766 	return kvmppc_core_pending_dec(vcpu);
767 }
768 
769 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
770 {
771 	struct kvm_vcpu *vcpu;
772 
773 	vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
774 	kvmppc_decrementer_func(vcpu);
775 
776 	return HRTIMER_NORESTART;
777 }
778 
779 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
780 {
781 	int ret;
782 
783 	hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
784 	vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
785 	vcpu->arch.dec_expires = get_tb();
786 
787 #ifdef CONFIG_KVM_EXIT_TIMING
788 	mutex_init(&vcpu->arch.exit_timing_lock);
789 #endif
790 	ret = kvmppc_subarch_vcpu_init(vcpu);
791 	return ret;
792 }
793 
794 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
795 {
796 	kvmppc_mmu_destroy(vcpu);
797 	kvmppc_subarch_vcpu_uninit(vcpu);
798 }
799 
800 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
801 {
802 #ifdef CONFIG_BOOKE
803 	/*
804 	 * vrsave (formerly usprg0) isn't used by Linux, but may
805 	 * be used by the guest.
806 	 *
807 	 * On non-booke this is associated with Altivec and
808 	 * is handled by code in book3s.c.
809 	 */
810 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
811 #endif
812 	kvmppc_core_vcpu_load(vcpu, cpu);
813 }
814 
815 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
816 {
817 	kvmppc_core_vcpu_put(vcpu);
818 #ifdef CONFIG_BOOKE
819 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
820 #endif
821 }
822 
823 /*
824  * irq_bypass_add_producer and irq_bypass_del_producer are only
825  * useful if the architecture supports PCI passthrough.
826  * irq_bypass_stop and irq_bypass_start are not needed and so
827  * kvm_ops are not defined for them.
828  */
829 bool kvm_arch_has_irq_bypass(void)
830 {
831 	return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
832 		(kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
833 }
834 
835 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
836 				     struct irq_bypass_producer *prod)
837 {
838 	struct kvm_kernel_irqfd *irqfd =
839 		container_of(cons, struct kvm_kernel_irqfd, consumer);
840 	struct kvm *kvm = irqfd->kvm;
841 
842 	if (kvm->arch.kvm_ops->irq_bypass_add_producer)
843 		return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
844 
845 	return 0;
846 }
847 
848 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
849 				      struct irq_bypass_producer *prod)
850 {
851 	struct kvm_kernel_irqfd *irqfd =
852 		container_of(cons, struct kvm_kernel_irqfd, consumer);
853 	struct kvm *kvm = irqfd->kvm;
854 
855 	if (kvm->arch.kvm_ops->irq_bypass_del_producer)
856 		kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
857 }
858 
859 #ifdef CONFIG_VSX
860 static inline int kvmppc_get_vsr_dword_offset(int index)
861 {
862 	int offset;
863 
864 	if ((index != 0) && (index != 1))
865 		return -1;
866 
867 #ifdef __BIG_ENDIAN
868 	offset =  index;
869 #else
870 	offset = 1 - index;
871 #endif
872 
873 	return offset;
874 }
875 
876 static inline int kvmppc_get_vsr_word_offset(int index)
877 {
878 	int offset;
879 
880 	if ((index > 3) || (index < 0))
881 		return -1;
882 
883 #ifdef __BIG_ENDIAN
884 	offset = index;
885 #else
886 	offset = 3 - index;
887 #endif
888 	return offset;
889 }
890 
891 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
892 	u64 gpr)
893 {
894 	union kvmppc_one_reg val;
895 	int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
896 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
897 
898 	if (offset == -1)
899 		return;
900 
901 	if (index >= 32) {
902 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
903 		val.vsxval[offset] = gpr;
904 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
905 	} else {
906 		VCPU_VSX_FPR(vcpu, index, offset) = gpr;
907 	}
908 }
909 
910 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
911 	u64 gpr)
912 {
913 	union kvmppc_one_reg val;
914 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
915 
916 	if (index >= 32) {
917 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
918 		val.vsxval[0] = gpr;
919 		val.vsxval[1] = gpr;
920 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
921 	} else {
922 		VCPU_VSX_FPR(vcpu, index, 0) = gpr;
923 		VCPU_VSX_FPR(vcpu, index, 1) = gpr;
924 	}
925 }
926 
927 static inline void kvmppc_set_vsr_word_dump(struct kvm_vcpu *vcpu,
928 	u32 gpr)
929 {
930 	union kvmppc_one_reg val;
931 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
932 
933 	if (index >= 32) {
934 		val.vsx32val[0] = gpr;
935 		val.vsx32val[1] = gpr;
936 		val.vsx32val[2] = gpr;
937 		val.vsx32val[3] = gpr;
938 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
939 	} else {
940 		val.vsx32val[0] = gpr;
941 		val.vsx32val[1] = gpr;
942 		VCPU_VSX_FPR(vcpu, index, 0) = val.vsxval[0];
943 		VCPU_VSX_FPR(vcpu, index, 1) = val.vsxval[0];
944 	}
945 }
946 
947 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
948 	u32 gpr32)
949 {
950 	union kvmppc_one_reg val;
951 	int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
952 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
953 	int dword_offset, word_offset;
954 
955 	if (offset == -1)
956 		return;
957 
958 	if (index >= 32) {
959 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
960 		val.vsx32val[offset] = gpr32;
961 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
962 	} else {
963 		dword_offset = offset / 2;
964 		word_offset = offset % 2;
965 		val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
966 		val.vsx32val[word_offset] = gpr32;
967 		VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
968 	}
969 }
970 #endif /* CONFIG_VSX */
971 
972 #ifdef CONFIG_ALTIVEC
973 static inline int kvmppc_get_vmx_offset_generic(struct kvm_vcpu *vcpu,
974 		int index, int element_size)
975 {
976 	int offset;
977 	int elts = sizeof(vector128)/element_size;
978 
979 	if ((index < 0) || (index >= elts))
980 		return -1;
981 
982 	if (kvmppc_need_byteswap(vcpu))
983 		offset = elts - index - 1;
984 	else
985 		offset = index;
986 
987 	return offset;
988 }
989 
990 static inline int kvmppc_get_vmx_dword_offset(struct kvm_vcpu *vcpu,
991 		int index)
992 {
993 	return kvmppc_get_vmx_offset_generic(vcpu, index, 8);
994 }
995 
996 static inline int kvmppc_get_vmx_word_offset(struct kvm_vcpu *vcpu,
997 		int index)
998 {
999 	return kvmppc_get_vmx_offset_generic(vcpu, index, 4);
1000 }
1001 
1002 static inline int kvmppc_get_vmx_hword_offset(struct kvm_vcpu *vcpu,
1003 		int index)
1004 {
1005 	return kvmppc_get_vmx_offset_generic(vcpu, index, 2);
1006 }
1007 
1008 static inline int kvmppc_get_vmx_byte_offset(struct kvm_vcpu *vcpu,
1009 		int index)
1010 {
1011 	return kvmppc_get_vmx_offset_generic(vcpu, index, 1);
1012 }
1013 
1014 
1015 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu,
1016 	u64 gpr)
1017 {
1018 	union kvmppc_one_reg val;
1019 	int offset = kvmppc_get_vmx_dword_offset(vcpu,
1020 			vcpu->arch.mmio_vmx_offset);
1021 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1022 
1023 	if (offset == -1)
1024 		return;
1025 
1026 	val.vval = VCPU_VSX_VR(vcpu, index);
1027 	val.vsxval[offset] = gpr;
1028 	VCPU_VSX_VR(vcpu, index) = val.vval;
1029 }
1030 
1031 static inline void kvmppc_set_vmx_word(struct kvm_vcpu *vcpu,
1032 	u32 gpr32)
1033 {
1034 	union kvmppc_one_reg val;
1035 	int offset = kvmppc_get_vmx_word_offset(vcpu,
1036 			vcpu->arch.mmio_vmx_offset);
1037 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1038 
1039 	if (offset == -1)
1040 		return;
1041 
1042 	val.vval = VCPU_VSX_VR(vcpu, index);
1043 	val.vsx32val[offset] = gpr32;
1044 	VCPU_VSX_VR(vcpu, index) = val.vval;
1045 }
1046 
1047 static inline void kvmppc_set_vmx_hword(struct kvm_vcpu *vcpu,
1048 	u16 gpr16)
1049 {
1050 	union kvmppc_one_reg val;
1051 	int offset = kvmppc_get_vmx_hword_offset(vcpu,
1052 			vcpu->arch.mmio_vmx_offset);
1053 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1054 
1055 	if (offset == -1)
1056 		return;
1057 
1058 	val.vval = VCPU_VSX_VR(vcpu, index);
1059 	val.vsx16val[offset] = gpr16;
1060 	VCPU_VSX_VR(vcpu, index) = val.vval;
1061 }
1062 
1063 static inline void kvmppc_set_vmx_byte(struct kvm_vcpu *vcpu,
1064 	u8 gpr8)
1065 {
1066 	union kvmppc_one_reg val;
1067 	int offset = kvmppc_get_vmx_byte_offset(vcpu,
1068 			vcpu->arch.mmio_vmx_offset);
1069 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1070 
1071 	if (offset == -1)
1072 		return;
1073 
1074 	val.vval = VCPU_VSX_VR(vcpu, index);
1075 	val.vsx8val[offset] = gpr8;
1076 	VCPU_VSX_VR(vcpu, index) = val.vval;
1077 }
1078 #endif /* CONFIG_ALTIVEC */
1079 
1080 #ifdef CONFIG_PPC_FPU
1081 static inline u64 sp_to_dp(u32 fprs)
1082 {
1083 	u64 fprd;
1084 
1085 	preempt_disable();
1086 	enable_kernel_fp();
1087 	asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs)
1088 	     : "fr0");
1089 	preempt_enable();
1090 	return fprd;
1091 }
1092 
1093 static inline u32 dp_to_sp(u64 fprd)
1094 {
1095 	u32 fprs;
1096 
1097 	preempt_disable();
1098 	enable_kernel_fp();
1099 	asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd)
1100 	     : "fr0");
1101 	preempt_enable();
1102 	return fprs;
1103 }
1104 
1105 #else
1106 #define sp_to_dp(x)	(x)
1107 #define dp_to_sp(x)	(x)
1108 #endif /* CONFIG_PPC_FPU */
1109 
1110 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
1111                                       struct kvm_run *run)
1112 {
1113 	u64 uninitialized_var(gpr);
1114 
1115 	if (run->mmio.len > sizeof(gpr)) {
1116 		printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
1117 		return;
1118 	}
1119 
1120 	if (!vcpu->arch.mmio_host_swabbed) {
1121 		switch (run->mmio.len) {
1122 		case 8: gpr = *(u64 *)run->mmio.data; break;
1123 		case 4: gpr = *(u32 *)run->mmio.data; break;
1124 		case 2: gpr = *(u16 *)run->mmio.data; break;
1125 		case 1: gpr = *(u8 *)run->mmio.data; break;
1126 		}
1127 	} else {
1128 		switch (run->mmio.len) {
1129 		case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
1130 		case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
1131 		case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
1132 		case 1: gpr = *(u8 *)run->mmio.data; break;
1133 		}
1134 	}
1135 
1136 	/* conversion between single and double precision */
1137 	if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
1138 		gpr = sp_to_dp(gpr);
1139 
1140 	if (vcpu->arch.mmio_sign_extend) {
1141 		switch (run->mmio.len) {
1142 #ifdef CONFIG_PPC64
1143 		case 4:
1144 			gpr = (s64)(s32)gpr;
1145 			break;
1146 #endif
1147 		case 2:
1148 			gpr = (s64)(s16)gpr;
1149 			break;
1150 		case 1:
1151 			gpr = (s64)(s8)gpr;
1152 			break;
1153 		}
1154 	}
1155 
1156 	switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1157 	case KVM_MMIO_REG_GPR:
1158 		kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1159 		break;
1160 	case KVM_MMIO_REG_FPR:
1161 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1162 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_FP);
1163 
1164 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1165 		break;
1166 #ifdef CONFIG_PPC_BOOK3S
1167 	case KVM_MMIO_REG_QPR:
1168 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1169 		break;
1170 	case KVM_MMIO_REG_FQPR:
1171 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1172 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1173 		break;
1174 #endif
1175 #ifdef CONFIG_VSX
1176 	case KVM_MMIO_REG_VSX:
1177 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1178 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VSX);
1179 
1180 		if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_DWORD)
1181 			kvmppc_set_vsr_dword(vcpu, gpr);
1182 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_WORD)
1183 			kvmppc_set_vsr_word(vcpu, gpr);
1184 		else if (vcpu->arch.mmio_copy_type ==
1185 				KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1186 			kvmppc_set_vsr_dword_dump(vcpu, gpr);
1187 		else if (vcpu->arch.mmio_copy_type ==
1188 				KVMPPC_VSX_COPY_WORD_LOAD_DUMP)
1189 			kvmppc_set_vsr_word_dump(vcpu, gpr);
1190 		break;
1191 #endif
1192 #ifdef CONFIG_ALTIVEC
1193 	case KVM_MMIO_REG_VMX:
1194 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1195 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VEC);
1196 
1197 		if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_DWORD)
1198 			kvmppc_set_vmx_dword(vcpu, gpr);
1199 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_WORD)
1200 			kvmppc_set_vmx_word(vcpu, gpr);
1201 		else if (vcpu->arch.mmio_copy_type ==
1202 				KVMPPC_VMX_COPY_HWORD)
1203 			kvmppc_set_vmx_hword(vcpu, gpr);
1204 		else if (vcpu->arch.mmio_copy_type ==
1205 				KVMPPC_VMX_COPY_BYTE)
1206 			kvmppc_set_vmx_byte(vcpu, gpr);
1207 		break;
1208 #endif
1209 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1210 	case KVM_MMIO_REG_NESTED_GPR:
1211 		if (kvmppc_need_byteswap(vcpu))
1212 			gpr = swab64(gpr);
1213 		kvm_vcpu_write_guest(vcpu, vcpu->arch.nested_io_gpr, &gpr,
1214 				     sizeof(gpr));
1215 		break;
1216 #endif
1217 	default:
1218 		BUG();
1219 	}
1220 }
1221 
1222 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1223 				unsigned int rt, unsigned int bytes,
1224 				int is_default_endian, int sign_extend)
1225 {
1226 	int idx, ret;
1227 	bool host_swabbed;
1228 
1229 	/* Pity C doesn't have a logical XOR operator */
1230 	if (kvmppc_need_byteswap(vcpu)) {
1231 		host_swabbed = is_default_endian;
1232 	} else {
1233 		host_swabbed = !is_default_endian;
1234 	}
1235 
1236 	if (bytes > sizeof(run->mmio.data)) {
1237 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1238 		       run->mmio.len);
1239 	}
1240 
1241 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1242 	run->mmio.len = bytes;
1243 	run->mmio.is_write = 0;
1244 
1245 	vcpu->arch.io_gpr = rt;
1246 	vcpu->arch.mmio_host_swabbed = host_swabbed;
1247 	vcpu->mmio_needed = 1;
1248 	vcpu->mmio_is_write = 0;
1249 	vcpu->arch.mmio_sign_extend = sign_extend;
1250 
1251 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1252 
1253 	ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1254 			      bytes, &run->mmio.data);
1255 
1256 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1257 
1258 	if (!ret) {
1259 		kvmppc_complete_mmio_load(vcpu, run);
1260 		vcpu->mmio_needed = 0;
1261 		return EMULATE_DONE;
1262 	}
1263 
1264 	return EMULATE_DO_MMIO;
1265 }
1266 
1267 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1268 		       unsigned int rt, unsigned int bytes,
1269 		       int is_default_endian)
1270 {
1271 	return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
1272 }
1273 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1274 
1275 /* Same as above, but sign extends */
1276 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
1277 			unsigned int rt, unsigned int bytes,
1278 			int is_default_endian)
1279 {
1280 	return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
1281 }
1282 
1283 #ifdef CONFIG_VSX
1284 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1285 			unsigned int rt, unsigned int bytes,
1286 			int is_default_endian, int mmio_sign_extend)
1287 {
1288 	enum emulation_result emulated = EMULATE_DONE;
1289 
1290 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1291 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1292 		return EMULATE_FAIL;
1293 
1294 	while (vcpu->arch.mmio_vsx_copy_nums) {
1295 		emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1296 			is_default_endian, mmio_sign_extend);
1297 
1298 		if (emulated != EMULATE_DONE)
1299 			break;
1300 
1301 		vcpu->arch.paddr_accessed += run->mmio.len;
1302 
1303 		vcpu->arch.mmio_vsx_copy_nums--;
1304 		vcpu->arch.mmio_vsx_offset++;
1305 	}
1306 	return emulated;
1307 }
1308 #endif /* CONFIG_VSX */
1309 
1310 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1311 			u64 val, unsigned int bytes, int is_default_endian)
1312 {
1313 	void *data = run->mmio.data;
1314 	int idx, ret;
1315 	bool host_swabbed;
1316 
1317 	/* Pity C doesn't have a logical XOR operator */
1318 	if (kvmppc_need_byteswap(vcpu)) {
1319 		host_swabbed = is_default_endian;
1320 	} else {
1321 		host_swabbed = !is_default_endian;
1322 	}
1323 
1324 	if (bytes > sizeof(run->mmio.data)) {
1325 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1326 		       run->mmio.len);
1327 	}
1328 
1329 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1330 	run->mmio.len = bytes;
1331 	run->mmio.is_write = 1;
1332 	vcpu->mmio_needed = 1;
1333 	vcpu->mmio_is_write = 1;
1334 
1335 	if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1336 		val = dp_to_sp(val);
1337 
1338 	/* Store the value at the lowest bytes in 'data'. */
1339 	if (!host_swabbed) {
1340 		switch (bytes) {
1341 		case 8: *(u64 *)data = val; break;
1342 		case 4: *(u32 *)data = val; break;
1343 		case 2: *(u16 *)data = val; break;
1344 		case 1: *(u8  *)data = val; break;
1345 		}
1346 	} else {
1347 		switch (bytes) {
1348 		case 8: *(u64 *)data = swab64(val); break;
1349 		case 4: *(u32 *)data = swab32(val); break;
1350 		case 2: *(u16 *)data = swab16(val); break;
1351 		case 1: *(u8  *)data = val; break;
1352 		}
1353 	}
1354 
1355 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1356 
1357 	ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1358 			       bytes, &run->mmio.data);
1359 
1360 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1361 
1362 	if (!ret) {
1363 		vcpu->mmio_needed = 0;
1364 		return EMULATE_DONE;
1365 	}
1366 
1367 	return EMULATE_DO_MMIO;
1368 }
1369 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1370 
1371 #ifdef CONFIG_VSX
1372 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1373 {
1374 	u32 dword_offset, word_offset;
1375 	union kvmppc_one_reg reg;
1376 	int vsx_offset = 0;
1377 	int copy_type = vcpu->arch.mmio_copy_type;
1378 	int result = 0;
1379 
1380 	switch (copy_type) {
1381 	case KVMPPC_VSX_COPY_DWORD:
1382 		vsx_offset =
1383 			kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1384 
1385 		if (vsx_offset == -1) {
1386 			result = -1;
1387 			break;
1388 		}
1389 
1390 		if (rs < 32) {
1391 			*val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1392 		} else {
1393 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1394 			*val = reg.vsxval[vsx_offset];
1395 		}
1396 		break;
1397 
1398 	case KVMPPC_VSX_COPY_WORD:
1399 		vsx_offset =
1400 			kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1401 
1402 		if (vsx_offset == -1) {
1403 			result = -1;
1404 			break;
1405 		}
1406 
1407 		if (rs < 32) {
1408 			dword_offset = vsx_offset / 2;
1409 			word_offset = vsx_offset % 2;
1410 			reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1411 			*val = reg.vsx32val[word_offset];
1412 		} else {
1413 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1414 			*val = reg.vsx32val[vsx_offset];
1415 		}
1416 		break;
1417 
1418 	default:
1419 		result = -1;
1420 		break;
1421 	}
1422 
1423 	return result;
1424 }
1425 
1426 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1427 			int rs, unsigned int bytes, int is_default_endian)
1428 {
1429 	u64 val;
1430 	enum emulation_result emulated = EMULATE_DONE;
1431 
1432 	vcpu->arch.io_gpr = rs;
1433 
1434 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1435 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1436 		return EMULATE_FAIL;
1437 
1438 	while (vcpu->arch.mmio_vsx_copy_nums) {
1439 		if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1440 			return EMULATE_FAIL;
1441 
1442 		emulated = kvmppc_handle_store(run, vcpu,
1443 			 val, bytes, is_default_endian);
1444 
1445 		if (emulated != EMULATE_DONE)
1446 			break;
1447 
1448 		vcpu->arch.paddr_accessed += run->mmio.len;
1449 
1450 		vcpu->arch.mmio_vsx_copy_nums--;
1451 		vcpu->arch.mmio_vsx_offset++;
1452 	}
1453 
1454 	return emulated;
1455 }
1456 
1457 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu,
1458 			struct kvm_run *run)
1459 {
1460 	enum emulation_result emulated = EMULATE_FAIL;
1461 	int r;
1462 
1463 	vcpu->arch.paddr_accessed += run->mmio.len;
1464 
1465 	if (!vcpu->mmio_is_write) {
1466 		emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr,
1467 			 run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1468 	} else {
1469 		emulated = kvmppc_handle_vsx_store(run, vcpu,
1470 			 vcpu->arch.io_gpr, run->mmio.len, 1);
1471 	}
1472 
1473 	switch (emulated) {
1474 	case EMULATE_DO_MMIO:
1475 		run->exit_reason = KVM_EXIT_MMIO;
1476 		r = RESUME_HOST;
1477 		break;
1478 	case EMULATE_FAIL:
1479 		pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1480 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1481 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1482 		r = RESUME_HOST;
1483 		break;
1484 	default:
1485 		r = RESUME_GUEST;
1486 		break;
1487 	}
1488 	return r;
1489 }
1490 #endif /* CONFIG_VSX */
1491 
1492 #ifdef CONFIG_ALTIVEC
1493 int kvmppc_handle_vmx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1494 		unsigned int rt, unsigned int bytes, int is_default_endian)
1495 {
1496 	enum emulation_result emulated = EMULATE_DONE;
1497 
1498 	if (vcpu->arch.mmio_vsx_copy_nums > 2)
1499 		return EMULATE_FAIL;
1500 
1501 	while (vcpu->arch.mmio_vmx_copy_nums) {
1502 		emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1503 				is_default_endian, 0);
1504 
1505 		if (emulated != EMULATE_DONE)
1506 			break;
1507 
1508 		vcpu->arch.paddr_accessed += run->mmio.len;
1509 		vcpu->arch.mmio_vmx_copy_nums--;
1510 		vcpu->arch.mmio_vmx_offset++;
1511 	}
1512 
1513 	return emulated;
1514 }
1515 
1516 int kvmppc_get_vmx_dword(struct kvm_vcpu *vcpu, int index, u64 *val)
1517 {
1518 	union kvmppc_one_reg reg;
1519 	int vmx_offset = 0;
1520 	int result = 0;
1521 
1522 	vmx_offset =
1523 		kvmppc_get_vmx_dword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1524 
1525 	if (vmx_offset == -1)
1526 		return -1;
1527 
1528 	reg.vval = VCPU_VSX_VR(vcpu, index);
1529 	*val = reg.vsxval[vmx_offset];
1530 
1531 	return result;
1532 }
1533 
1534 int kvmppc_get_vmx_word(struct kvm_vcpu *vcpu, int index, u64 *val)
1535 {
1536 	union kvmppc_one_reg reg;
1537 	int vmx_offset = 0;
1538 	int result = 0;
1539 
1540 	vmx_offset =
1541 		kvmppc_get_vmx_word_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1542 
1543 	if (vmx_offset == -1)
1544 		return -1;
1545 
1546 	reg.vval = VCPU_VSX_VR(vcpu, index);
1547 	*val = reg.vsx32val[vmx_offset];
1548 
1549 	return result;
1550 }
1551 
1552 int kvmppc_get_vmx_hword(struct kvm_vcpu *vcpu, int index, u64 *val)
1553 {
1554 	union kvmppc_one_reg reg;
1555 	int vmx_offset = 0;
1556 	int result = 0;
1557 
1558 	vmx_offset =
1559 		kvmppc_get_vmx_hword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1560 
1561 	if (vmx_offset == -1)
1562 		return -1;
1563 
1564 	reg.vval = VCPU_VSX_VR(vcpu, index);
1565 	*val = reg.vsx16val[vmx_offset];
1566 
1567 	return result;
1568 }
1569 
1570 int kvmppc_get_vmx_byte(struct kvm_vcpu *vcpu, int index, u64 *val)
1571 {
1572 	union kvmppc_one_reg reg;
1573 	int vmx_offset = 0;
1574 	int result = 0;
1575 
1576 	vmx_offset =
1577 		kvmppc_get_vmx_byte_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1578 
1579 	if (vmx_offset == -1)
1580 		return -1;
1581 
1582 	reg.vval = VCPU_VSX_VR(vcpu, index);
1583 	*val = reg.vsx8val[vmx_offset];
1584 
1585 	return result;
1586 }
1587 
1588 int kvmppc_handle_vmx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1589 		unsigned int rs, unsigned int bytes, int is_default_endian)
1590 {
1591 	u64 val = 0;
1592 	unsigned int index = rs & KVM_MMIO_REG_MASK;
1593 	enum emulation_result emulated = EMULATE_DONE;
1594 
1595 	if (vcpu->arch.mmio_vsx_copy_nums > 2)
1596 		return EMULATE_FAIL;
1597 
1598 	vcpu->arch.io_gpr = rs;
1599 
1600 	while (vcpu->arch.mmio_vmx_copy_nums) {
1601 		switch (vcpu->arch.mmio_copy_type) {
1602 		case KVMPPC_VMX_COPY_DWORD:
1603 			if (kvmppc_get_vmx_dword(vcpu, index, &val) == -1)
1604 				return EMULATE_FAIL;
1605 
1606 			break;
1607 		case KVMPPC_VMX_COPY_WORD:
1608 			if (kvmppc_get_vmx_word(vcpu, index, &val) == -1)
1609 				return EMULATE_FAIL;
1610 			break;
1611 		case KVMPPC_VMX_COPY_HWORD:
1612 			if (kvmppc_get_vmx_hword(vcpu, index, &val) == -1)
1613 				return EMULATE_FAIL;
1614 			break;
1615 		case KVMPPC_VMX_COPY_BYTE:
1616 			if (kvmppc_get_vmx_byte(vcpu, index, &val) == -1)
1617 				return EMULATE_FAIL;
1618 			break;
1619 		default:
1620 			return EMULATE_FAIL;
1621 		}
1622 
1623 		emulated = kvmppc_handle_store(run, vcpu, val, bytes,
1624 				is_default_endian);
1625 		if (emulated != EMULATE_DONE)
1626 			break;
1627 
1628 		vcpu->arch.paddr_accessed += run->mmio.len;
1629 		vcpu->arch.mmio_vmx_copy_nums--;
1630 		vcpu->arch.mmio_vmx_offset++;
1631 	}
1632 
1633 	return emulated;
1634 }
1635 
1636 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu,
1637 		struct kvm_run *run)
1638 {
1639 	enum emulation_result emulated = EMULATE_FAIL;
1640 	int r;
1641 
1642 	vcpu->arch.paddr_accessed += run->mmio.len;
1643 
1644 	if (!vcpu->mmio_is_write) {
1645 		emulated = kvmppc_handle_vmx_load(run, vcpu,
1646 				vcpu->arch.io_gpr, run->mmio.len, 1);
1647 	} else {
1648 		emulated = kvmppc_handle_vmx_store(run, vcpu,
1649 				vcpu->arch.io_gpr, run->mmio.len, 1);
1650 	}
1651 
1652 	switch (emulated) {
1653 	case EMULATE_DO_MMIO:
1654 		run->exit_reason = KVM_EXIT_MMIO;
1655 		r = RESUME_HOST;
1656 		break;
1657 	case EMULATE_FAIL:
1658 		pr_info("KVM: MMIO emulation failed (VMX repeat)\n");
1659 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1660 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1661 		r = RESUME_HOST;
1662 		break;
1663 	default:
1664 		r = RESUME_GUEST;
1665 		break;
1666 	}
1667 	return r;
1668 }
1669 #endif /* CONFIG_ALTIVEC */
1670 
1671 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1672 {
1673 	int r = 0;
1674 	union kvmppc_one_reg val;
1675 	int size;
1676 
1677 	size = one_reg_size(reg->id);
1678 	if (size > sizeof(val))
1679 		return -EINVAL;
1680 
1681 	r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1682 	if (r == -EINVAL) {
1683 		r = 0;
1684 		switch (reg->id) {
1685 #ifdef CONFIG_ALTIVEC
1686 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1687 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1688 				r = -ENXIO;
1689 				break;
1690 			}
1691 			val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1692 			break;
1693 		case KVM_REG_PPC_VSCR:
1694 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1695 				r = -ENXIO;
1696 				break;
1697 			}
1698 			val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1699 			break;
1700 		case KVM_REG_PPC_VRSAVE:
1701 			val = get_reg_val(reg->id, vcpu->arch.vrsave);
1702 			break;
1703 #endif /* CONFIG_ALTIVEC */
1704 		default:
1705 			r = -EINVAL;
1706 			break;
1707 		}
1708 	}
1709 
1710 	if (r)
1711 		return r;
1712 
1713 	if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1714 		r = -EFAULT;
1715 
1716 	return r;
1717 }
1718 
1719 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1720 {
1721 	int r;
1722 	union kvmppc_one_reg val;
1723 	int size;
1724 
1725 	size = one_reg_size(reg->id);
1726 	if (size > sizeof(val))
1727 		return -EINVAL;
1728 
1729 	if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1730 		return -EFAULT;
1731 
1732 	r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1733 	if (r == -EINVAL) {
1734 		r = 0;
1735 		switch (reg->id) {
1736 #ifdef CONFIG_ALTIVEC
1737 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1738 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1739 				r = -ENXIO;
1740 				break;
1741 			}
1742 			vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1743 			break;
1744 		case KVM_REG_PPC_VSCR:
1745 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1746 				r = -ENXIO;
1747 				break;
1748 			}
1749 			vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1750 			break;
1751 		case KVM_REG_PPC_VRSAVE:
1752 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1753 				r = -ENXIO;
1754 				break;
1755 			}
1756 			vcpu->arch.vrsave = set_reg_val(reg->id, val);
1757 			break;
1758 #endif /* CONFIG_ALTIVEC */
1759 		default:
1760 			r = -EINVAL;
1761 			break;
1762 		}
1763 	}
1764 
1765 	return r;
1766 }
1767 
1768 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1769 {
1770 	int r;
1771 
1772 	vcpu_load(vcpu);
1773 
1774 	if (vcpu->mmio_needed) {
1775 		vcpu->mmio_needed = 0;
1776 		if (!vcpu->mmio_is_write)
1777 			kvmppc_complete_mmio_load(vcpu, run);
1778 #ifdef CONFIG_VSX
1779 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1780 			vcpu->arch.mmio_vsx_copy_nums--;
1781 			vcpu->arch.mmio_vsx_offset++;
1782 		}
1783 
1784 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1785 			r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run);
1786 			if (r == RESUME_HOST) {
1787 				vcpu->mmio_needed = 1;
1788 				goto out;
1789 			}
1790 		}
1791 #endif
1792 #ifdef CONFIG_ALTIVEC
1793 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1794 			vcpu->arch.mmio_vmx_copy_nums--;
1795 			vcpu->arch.mmio_vmx_offset++;
1796 		}
1797 
1798 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1799 			r = kvmppc_emulate_mmio_vmx_loadstore(vcpu, run);
1800 			if (r == RESUME_HOST) {
1801 				vcpu->mmio_needed = 1;
1802 				goto out;
1803 			}
1804 		}
1805 #endif
1806 	} else if (vcpu->arch.osi_needed) {
1807 		u64 *gprs = run->osi.gprs;
1808 		int i;
1809 
1810 		for (i = 0; i < 32; i++)
1811 			kvmppc_set_gpr(vcpu, i, gprs[i]);
1812 		vcpu->arch.osi_needed = 0;
1813 	} else if (vcpu->arch.hcall_needed) {
1814 		int i;
1815 
1816 		kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1817 		for (i = 0; i < 9; ++i)
1818 			kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1819 		vcpu->arch.hcall_needed = 0;
1820 #ifdef CONFIG_BOOKE
1821 	} else if (vcpu->arch.epr_needed) {
1822 		kvmppc_set_epr(vcpu, run->epr.epr);
1823 		vcpu->arch.epr_needed = 0;
1824 #endif
1825 	}
1826 
1827 	kvm_sigset_activate(vcpu);
1828 
1829 	if (run->immediate_exit)
1830 		r = -EINTR;
1831 	else
1832 		r = kvmppc_vcpu_run(run, vcpu);
1833 
1834 	kvm_sigset_deactivate(vcpu);
1835 
1836 #ifdef CONFIG_ALTIVEC
1837 out:
1838 #endif
1839 	vcpu_put(vcpu);
1840 	return r;
1841 }
1842 
1843 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1844 {
1845 	if (irq->irq == KVM_INTERRUPT_UNSET) {
1846 		kvmppc_core_dequeue_external(vcpu);
1847 		return 0;
1848 	}
1849 
1850 	kvmppc_core_queue_external(vcpu, irq);
1851 
1852 	kvm_vcpu_kick(vcpu);
1853 
1854 	return 0;
1855 }
1856 
1857 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1858 				     struct kvm_enable_cap *cap)
1859 {
1860 	int r;
1861 
1862 	if (cap->flags)
1863 		return -EINVAL;
1864 
1865 	switch (cap->cap) {
1866 	case KVM_CAP_PPC_OSI:
1867 		r = 0;
1868 		vcpu->arch.osi_enabled = true;
1869 		break;
1870 	case KVM_CAP_PPC_PAPR:
1871 		r = 0;
1872 		vcpu->arch.papr_enabled = true;
1873 		break;
1874 	case KVM_CAP_PPC_EPR:
1875 		r = 0;
1876 		if (cap->args[0])
1877 			vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1878 		else
1879 			vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1880 		break;
1881 #ifdef CONFIG_BOOKE
1882 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
1883 		r = 0;
1884 		vcpu->arch.watchdog_enabled = true;
1885 		break;
1886 #endif
1887 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1888 	case KVM_CAP_SW_TLB: {
1889 		struct kvm_config_tlb cfg;
1890 		void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1891 
1892 		r = -EFAULT;
1893 		if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1894 			break;
1895 
1896 		r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1897 		break;
1898 	}
1899 #endif
1900 #ifdef CONFIG_KVM_MPIC
1901 	case KVM_CAP_IRQ_MPIC: {
1902 		struct fd f;
1903 		struct kvm_device *dev;
1904 
1905 		r = -EBADF;
1906 		f = fdget(cap->args[0]);
1907 		if (!f.file)
1908 			break;
1909 
1910 		r = -EPERM;
1911 		dev = kvm_device_from_filp(f.file);
1912 		if (dev)
1913 			r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1914 
1915 		fdput(f);
1916 		break;
1917 	}
1918 #endif
1919 #ifdef CONFIG_KVM_XICS
1920 	case KVM_CAP_IRQ_XICS: {
1921 		struct fd f;
1922 		struct kvm_device *dev;
1923 
1924 		r = -EBADF;
1925 		f = fdget(cap->args[0]);
1926 		if (!f.file)
1927 			break;
1928 
1929 		r = -EPERM;
1930 		dev = kvm_device_from_filp(f.file);
1931 		if (dev) {
1932 			if (xics_on_xive())
1933 				r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1934 			else
1935 				r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1936 		}
1937 
1938 		fdput(f);
1939 		break;
1940 	}
1941 #endif /* CONFIG_KVM_XICS */
1942 #ifdef CONFIG_KVM_XIVE
1943 	case KVM_CAP_PPC_IRQ_XIVE: {
1944 		struct fd f;
1945 		struct kvm_device *dev;
1946 
1947 		r = -EBADF;
1948 		f = fdget(cap->args[0]);
1949 		if (!f.file)
1950 			break;
1951 
1952 		r = -ENXIO;
1953 		if (!xive_enabled())
1954 			break;
1955 
1956 		r = -EPERM;
1957 		dev = kvm_device_from_filp(f.file);
1958 		if (dev)
1959 			r = kvmppc_xive_native_connect_vcpu(dev, vcpu,
1960 							    cap->args[1]);
1961 
1962 		fdput(f);
1963 		break;
1964 	}
1965 #endif /* CONFIG_KVM_XIVE */
1966 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1967 	case KVM_CAP_PPC_FWNMI:
1968 		r = -EINVAL;
1969 		if (!is_kvmppc_hv_enabled(vcpu->kvm))
1970 			break;
1971 		r = 0;
1972 		vcpu->kvm->arch.fwnmi_enabled = true;
1973 		break;
1974 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1975 	default:
1976 		r = -EINVAL;
1977 		break;
1978 	}
1979 
1980 	if (!r)
1981 		r = kvmppc_sanity_check(vcpu);
1982 
1983 	return r;
1984 }
1985 
1986 bool kvm_arch_intc_initialized(struct kvm *kvm)
1987 {
1988 #ifdef CONFIG_KVM_MPIC
1989 	if (kvm->arch.mpic)
1990 		return true;
1991 #endif
1992 #ifdef CONFIG_KVM_XICS
1993 	if (kvm->arch.xics || kvm->arch.xive)
1994 		return true;
1995 #endif
1996 	return false;
1997 }
1998 
1999 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2000                                     struct kvm_mp_state *mp_state)
2001 {
2002 	return -EINVAL;
2003 }
2004 
2005 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2006                                     struct kvm_mp_state *mp_state)
2007 {
2008 	return -EINVAL;
2009 }
2010 
2011 long kvm_arch_vcpu_async_ioctl(struct file *filp,
2012 			       unsigned int ioctl, unsigned long arg)
2013 {
2014 	struct kvm_vcpu *vcpu = filp->private_data;
2015 	void __user *argp = (void __user *)arg;
2016 
2017 	if (ioctl == KVM_INTERRUPT) {
2018 		struct kvm_interrupt irq;
2019 		if (copy_from_user(&irq, argp, sizeof(irq)))
2020 			return -EFAULT;
2021 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2022 	}
2023 	return -ENOIOCTLCMD;
2024 }
2025 
2026 long kvm_arch_vcpu_ioctl(struct file *filp,
2027                          unsigned int ioctl, unsigned long arg)
2028 {
2029 	struct kvm_vcpu *vcpu = filp->private_data;
2030 	void __user *argp = (void __user *)arg;
2031 	long r;
2032 
2033 	switch (ioctl) {
2034 	case KVM_ENABLE_CAP:
2035 	{
2036 		struct kvm_enable_cap cap;
2037 		r = -EFAULT;
2038 		vcpu_load(vcpu);
2039 		if (copy_from_user(&cap, argp, sizeof(cap)))
2040 			goto out;
2041 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
2042 		vcpu_put(vcpu);
2043 		break;
2044 	}
2045 
2046 	case KVM_SET_ONE_REG:
2047 	case KVM_GET_ONE_REG:
2048 	{
2049 		struct kvm_one_reg reg;
2050 		r = -EFAULT;
2051 		if (copy_from_user(&reg, argp, sizeof(reg)))
2052 			goto out;
2053 		if (ioctl == KVM_SET_ONE_REG)
2054 			r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
2055 		else
2056 			r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
2057 		break;
2058 	}
2059 
2060 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
2061 	case KVM_DIRTY_TLB: {
2062 		struct kvm_dirty_tlb dirty;
2063 		r = -EFAULT;
2064 		vcpu_load(vcpu);
2065 		if (copy_from_user(&dirty, argp, sizeof(dirty)))
2066 			goto out;
2067 		r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
2068 		vcpu_put(vcpu);
2069 		break;
2070 	}
2071 #endif
2072 	default:
2073 		r = -EINVAL;
2074 	}
2075 
2076 out:
2077 	return r;
2078 }
2079 
2080 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2081 {
2082 	return VM_FAULT_SIGBUS;
2083 }
2084 
2085 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
2086 {
2087 	u32 inst_nop = 0x60000000;
2088 #ifdef CONFIG_KVM_BOOKE_HV
2089 	u32 inst_sc1 = 0x44000022;
2090 	pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
2091 	pvinfo->hcall[1] = cpu_to_be32(inst_nop);
2092 	pvinfo->hcall[2] = cpu_to_be32(inst_nop);
2093 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2094 #else
2095 	u32 inst_lis = 0x3c000000;
2096 	u32 inst_ori = 0x60000000;
2097 	u32 inst_sc = 0x44000002;
2098 	u32 inst_imm_mask = 0xffff;
2099 
2100 	/*
2101 	 * The hypercall to get into KVM from within guest context is as
2102 	 * follows:
2103 	 *
2104 	 *    lis r0, r0, KVM_SC_MAGIC_R0@h
2105 	 *    ori r0, KVM_SC_MAGIC_R0@l
2106 	 *    sc
2107 	 *    nop
2108 	 */
2109 	pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
2110 	pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
2111 	pvinfo->hcall[2] = cpu_to_be32(inst_sc);
2112 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2113 #endif
2114 
2115 	pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
2116 
2117 	return 0;
2118 }
2119 
2120 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
2121 			  bool line_status)
2122 {
2123 	if (!irqchip_in_kernel(kvm))
2124 		return -ENXIO;
2125 
2126 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2127 					irq_event->irq, irq_event->level,
2128 					line_status);
2129 	return 0;
2130 }
2131 
2132 
2133 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
2134 			    struct kvm_enable_cap *cap)
2135 {
2136 	int r;
2137 
2138 	if (cap->flags)
2139 		return -EINVAL;
2140 
2141 	switch (cap->cap) {
2142 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
2143 	case KVM_CAP_PPC_ENABLE_HCALL: {
2144 		unsigned long hcall = cap->args[0];
2145 
2146 		r = -EINVAL;
2147 		if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
2148 		    cap->args[1] > 1)
2149 			break;
2150 		if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
2151 			break;
2152 		if (cap->args[1])
2153 			set_bit(hcall / 4, kvm->arch.enabled_hcalls);
2154 		else
2155 			clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
2156 		r = 0;
2157 		break;
2158 	}
2159 	case KVM_CAP_PPC_SMT: {
2160 		unsigned long mode = cap->args[0];
2161 		unsigned long flags = cap->args[1];
2162 
2163 		r = -EINVAL;
2164 		if (kvm->arch.kvm_ops->set_smt_mode)
2165 			r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
2166 		break;
2167 	}
2168 
2169 	case KVM_CAP_PPC_NESTED_HV:
2170 		r = -EINVAL;
2171 		if (!is_kvmppc_hv_enabled(kvm) ||
2172 		    !kvm->arch.kvm_ops->enable_nested)
2173 			break;
2174 		r = kvm->arch.kvm_ops->enable_nested(kvm);
2175 		break;
2176 #endif
2177 	default:
2178 		r = -EINVAL;
2179 		break;
2180 	}
2181 
2182 	return r;
2183 }
2184 
2185 #ifdef CONFIG_PPC_BOOK3S_64
2186 /*
2187  * These functions check whether the underlying hardware is safe
2188  * against attacks based on observing the effects of speculatively
2189  * executed instructions, and whether it supplies instructions for
2190  * use in workarounds.  The information comes from firmware, either
2191  * via the device tree on powernv platforms or from an hcall on
2192  * pseries platforms.
2193  */
2194 #ifdef CONFIG_PPC_PSERIES
2195 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2196 {
2197 	struct h_cpu_char_result c;
2198 	unsigned long rc;
2199 
2200 	if (!machine_is(pseries))
2201 		return -ENOTTY;
2202 
2203 	rc = plpar_get_cpu_characteristics(&c);
2204 	if (rc == H_SUCCESS) {
2205 		cp->character = c.character;
2206 		cp->behaviour = c.behaviour;
2207 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2208 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2209 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2210 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2211 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2212 			KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
2213 			KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
2214 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2215 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2216 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2217 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2218 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2219 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2220 	}
2221 	return 0;
2222 }
2223 #else
2224 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2225 {
2226 	return -ENOTTY;
2227 }
2228 #endif
2229 
2230 static inline bool have_fw_feat(struct device_node *fw_features,
2231 				const char *state, const char *name)
2232 {
2233 	struct device_node *np;
2234 	bool r = false;
2235 
2236 	np = of_get_child_by_name(fw_features, name);
2237 	if (np) {
2238 		r = of_property_read_bool(np, state);
2239 		of_node_put(np);
2240 	}
2241 	return r;
2242 }
2243 
2244 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2245 {
2246 	struct device_node *np, *fw_features;
2247 	int r;
2248 
2249 	memset(cp, 0, sizeof(*cp));
2250 	r = pseries_get_cpu_char(cp);
2251 	if (r != -ENOTTY)
2252 		return r;
2253 
2254 	np = of_find_node_by_name(NULL, "ibm,opal");
2255 	if (np) {
2256 		fw_features = of_get_child_by_name(np, "fw-features");
2257 		of_node_put(np);
2258 		if (!fw_features)
2259 			return 0;
2260 		if (have_fw_feat(fw_features, "enabled",
2261 				 "inst-spec-barrier-ori31,31,0"))
2262 			cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
2263 		if (have_fw_feat(fw_features, "enabled",
2264 				 "fw-bcctrl-serialized"))
2265 			cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
2266 		if (have_fw_feat(fw_features, "enabled",
2267 				 "inst-l1d-flush-ori30,30,0"))
2268 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
2269 		if (have_fw_feat(fw_features, "enabled",
2270 				 "inst-l1d-flush-trig2"))
2271 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
2272 		if (have_fw_feat(fw_features, "enabled",
2273 				 "fw-l1d-thread-split"))
2274 			cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
2275 		if (have_fw_feat(fw_features, "enabled",
2276 				 "fw-count-cache-disabled"))
2277 			cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2278 		if (have_fw_feat(fw_features, "enabled",
2279 				 "fw-count-cache-flush-bcctr2,0,0"))
2280 			cp->character |= KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2281 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2282 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2283 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2284 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2285 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2286 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS |
2287 			KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST;
2288 
2289 		if (have_fw_feat(fw_features, "enabled",
2290 				 "speculation-policy-favor-security"))
2291 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
2292 		if (!have_fw_feat(fw_features, "disabled",
2293 				  "needs-l1d-flush-msr-pr-0-to-1"))
2294 			cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
2295 		if (!have_fw_feat(fw_features, "disabled",
2296 				  "needs-spec-barrier-for-bound-checks"))
2297 			cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2298 		if (have_fw_feat(fw_features, "enabled",
2299 				 "needs-count-cache-flush-on-context-switch"))
2300 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2301 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2302 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2303 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR |
2304 			KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE;
2305 
2306 		of_node_put(fw_features);
2307 	}
2308 
2309 	return 0;
2310 }
2311 #endif
2312 
2313 long kvm_arch_vm_ioctl(struct file *filp,
2314                        unsigned int ioctl, unsigned long arg)
2315 {
2316 	struct kvm *kvm __maybe_unused = filp->private_data;
2317 	void __user *argp = (void __user *)arg;
2318 	long r;
2319 
2320 	switch (ioctl) {
2321 	case KVM_PPC_GET_PVINFO: {
2322 		struct kvm_ppc_pvinfo pvinfo;
2323 		memset(&pvinfo, 0, sizeof(pvinfo));
2324 		r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
2325 		if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
2326 			r = -EFAULT;
2327 			goto out;
2328 		}
2329 
2330 		break;
2331 	}
2332 #ifdef CONFIG_SPAPR_TCE_IOMMU
2333 	case KVM_CREATE_SPAPR_TCE_64: {
2334 		struct kvm_create_spapr_tce_64 create_tce_64;
2335 
2336 		r = -EFAULT;
2337 		if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
2338 			goto out;
2339 		if (create_tce_64.flags) {
2340 			r = -EINVAL;
2341 			goto out;
2342 		}
2343 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2344 		goto out;
2345 	}
2346 	case KVM_CREATE_SPAPR_TCE: {
2347 		struct kvm_create_spapr_tce create_tce;
2348 		struct kvm_create_spapr_tce_64 create_tce_64;
2349 
2350 		r = -EFAULT;
2351 		if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
2352 			goto out;
2353 
2354 		create_tce_64.liobn = create_tce.liobn;
2355 		create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
2356 		create_tce_64.offset = 0;
2357 		create_tce_64.size = create_tce.window_size >>
2358 				IOMMU_PAGE_SHIFT_4K;
2359 		create_tce_64.flags = 0;
2360 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2361 		goto out;
2362 	}
2363 #endif
2364 #ifdef CONFIG_PPC_BOOK3S_64
2365 	case KVM_PPC_GET_SMMU_INFO: {
2366 		struct kvm_ppc_smmu_info info;
2367 		struct kvm *kvm = filp->private_data;
2368 
2369 		memset(&info, 0, sizeof(info));
2370 		r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
2371 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2372 			r = -EFAULT;
2373 		break;
2374 	}
2375 	case KVM_PPC_RTAS_DEFINE_TOKEN: {
2376 		struct kvm *kvm = filp->private_data;
2377 
2378 		r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
2379 		break;
2380 	}
2381 	case KVM_PPC_CONFIGURE_V3_MMU: {
2382 		struct kvm *kvm = filp->private_data;
2383 		struct kvm_ppc_mmuv3_cfg cfg;
2384 
2385 		r = -EINVAL;
2386 		if (!kvm->arch.kvm_ops->configure_mmu)
2387 			goto out;
2388 		r = -EFAULT;
2389 		if (copy_from_user(&cfg, argp, sizeof(cfg)))
2390 			goto out;
2391 		r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
2392 		break;
2393 	}
2394 	case KVM_PPC_GET_RMMU_INFO: {
2395 		struct kvm *kvm = filp->private_data;
2396 		struct kvm_ppc_rmmu_info info;
2397 
2398 		r = -EINVAL;
2399 		if (!kvm->arch.kvm_ops->get_rmmu_info)
2400 			goto out;
2401 		r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
2402 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2403 			r = -EFAULT;
2404 		break;
2405 	}
2406 	case KVM_PPC_GET_CPU_CHAR: {
2407 		struct kvm_ppc_cpu_char cpuchar;
2408 
2409 		r = kvmppc_get_cpu_char(&cpuchar);
2410 		if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2411 			r = -EFAULT;
2412 		break;
2413 	}
2414 	default: {
2415 		struct kvm *kvm = filp->private_data;
2416 		r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
2417 	}
2418 #else /* CONFIG_PPC_BOOK3S_64 */
2419 	default:
2420 		r = -ENOTTY;
2421 #endif
2422 	}
2423 out:
2424 	return r;
2425 }
2426 
2427 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
2428 static unsigned long nr_lpids;
2429 
2430 long kvmppc_alloc_lpid(void)
2431 {
2432 	long lpid;
2433 
2434 	do {
2435 		lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
2436 		if (lpid >= nr_lpids) {
2437 			pr_err("%s: No LPIDs free\n", __func__);
2438 			return -ENOMEM;
2439 		}
2440 	} while (test_and_set_bit(lpid, lpid_inuse));
2441 
2442 	return lpid;
2443 }
2444 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2445 
2446 void kvmppc_claim_lpid(long lpid)
2447 {
2448 	set_bit(lpid, lpid_inuse);
2449 }
2450 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
2451 
2452 void kvmppc_free_lpid(long lpid)
2453 {
2454 	clear_bit(lpid, lpid_inuse);
2455 }
2456 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2457 
2458 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2459 {
2460 	nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
2461 	memset(lpid_inuse, 0, sizeof(lpid_inuse));
2462 }
2463 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2464 
2465 int kvm_arch_init(void *opaque)
2466 {
2467 	return 0;
2468 }
2469 
2470 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);
2471