xref: /openbmc/linux/arch/powerpc/kvm/powerpc.c (revision 83268fa6)
1 /*
2  * This program is free software; you can redistribute it and/or modify
3  * it under the terms of the GNU General Public License, version 2, as
4  * published by the Free Software Foundation.
5  *
6  * This program is distributed in the hope that it will be useful,
7  * but WITHOUT ANY WARRANTY; without even the implied warranty of
8  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9  * GNU General Public License for more details.
10  *
11  * You should have received a copy of the GNU General Public License
12  * along with this program; if not, write to the Free Software
13  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
14  *
15  * Copyright IBM Corp. 2007
16  *
17  * Authors: Hollis Blanchard <hollisb@us.ibm.com>
18  *          Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
19  */
20 
21 #include <linux/errno.h>
22 #include <linux/err.h>
23 #include <linux/kvm_host.h>
24 #include <linux/vmalloc.h>
25 #include <linux/hrtimer.h>
26 #include <linux/sched/signal.h>
27 #include <linux/fs.h>
28 #include <linux/slab.h>
29 #include <linux/file.h>
30 #include <linux/module.h>
31 #include <linux/irqbypass.h>
32 #include <linux/kvm_irqfd.h>
33 #include <asm/cputable.h>
34 #include <linux/uaccess.h>
35 #include <asm/kvm_ppc.h>
36 #include <asm/cputhreads.h>
37 #include <asm/irqflags.h>
38 #include <asm/iommu.h>
39 #include <asm/switch_to.h>
40 #include <asm/xive.h>
41 #ifdef CONFIG_PPC_PSERIES
42 #include <asm/hvcall.h>
43 #include <asm/plpar_wrappers.h>
44 #endif
45 
46 #include "timing.h"
47 #include "irq.h"
48 #include "../mm/mmu_decl.h"
49 
50 #define CREATE_TRACE_POINTS
51 #include "trace.h"
52 
53 struct kvmppc_ops *kvmppc_hv_ops;
54 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
55 struct kvmppc_ops *kvmppc_pr_ops;
56 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
57 
58 
59 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
60 {
61 	return !!(v->arch.pending_exceptions) || kvm_request_pending(v);
62 }
63 
64 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
65 {
66 	return false;
67 }
68 
69 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
70 {
71 	return 1;
72 }
73 
74 /*
75  * Common checks before entering the guest world.  Call with interrupts
76  * disabled.
77  *
78  * returns:
79  *
80  * == 1 if we're ready to go into guest state
81  * <= 0 if we need to go back to the host with return value
82  */
83 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
84 {
85 	int r;
86 
87 	WARN_ON(irqs_disabled());
88 	hard_irq_disable();
89 
90 	while (true) {
91 		if (need_resched()) {
92 			local_irq_enable();
93 			cond_resched();
94 			hard_irq_disable();
95 			continue;
96 		}
97 
98 		if (signal_pending(current)) {
99 			kvmppc_account_exit(vcpu, SIGNAL_EXITS);
100 			vcpu->run->exit_reason = KVM_EXIT_INTR;
101 			r = -EINTR;
102 			break;
103 		}
104 
105 		vcpu->mode = IN_GUEST_MODE;
106 
107 		/*
108 		 * Reading vcpu->requests must happen after setting vcpu->mode,
109 		 * so we don't miss a request because the requester sees
110 		 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
111 		 * before next entering the guest (and thus doesn't IPI).
112 		 * This also orders the write to mode from any reads
113 		 * to the page tables done while the VCPU is running.
114 		 * Please see the comment in kvm_flush_remote_tlbs.
115 		 */
116 		smp_mb();
117 
118 		if (kvm_request_pending(vcpu)) {
119 			/* Make sure we process requests preemptable */
120 			local_irq_enable();
121 			trace_kvm_check_requests(vcpu);
122 			r = kvmppc_core_check_requests(vcpu);
123 			hard_irq_disable();
124 			if (r > 0)
125 				continue;
126 			break;
127 		}
128 
129 		if (kvmppc_core_prepare_to_enter(vcpu)) {
130 			/* interrupts got enabled in between, so we
131 			   are back at square 1 */
132 			continue;
133 		}
134 
135 		guest_enter_irqoff();
136 		return 1;
137 	}
138 
139 	/* return to host */
140 	local_irq_enable();
141 	return r;
142 }
143 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
144 
145 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
146 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
147 {
148 	struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
149 	int i;
150 
151 	shared->sprg0 = swab64(shared->sprg0);
152 	shared->sprg1 = swab64(shared->sprg1);
153 	shared->sprg2 = swab64(shared->sprg2);
154 	shared->sprg3 = swab64(shared->sprg3);
155 	shared->srr0 = swab64(shared->srr0);
156 	shared->srr1 = swab64(shared->srr1);
157 	shared->dar = swab64(shared->dar);
158 	shared->msr = swab64(shared->msr);
159 	shared->dsisr = swab32(shared->dsisr);
160 	shared->int_pending = swab32(shared->int_pending);
161 	for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
162 		shared->sr[i] = swab32(shared->sr[i]);
163 }
164 #endif
165 
166 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
167 {
168 	int nr = kvmppc_get_gpr(vcpu, 11);
169 	int r;
170 	unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
171 	unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
172 	unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
173 	unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
174 	unsigned long r2 = 0;
175 
176 	if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
177 		/* 32 bit mode */
178 		param1 &= 0xffffffff;
179 		param2 &= 0xffffffff;
180 		param3 &= 0xffffffff;
181 		param4 &= 0xffffffff;
182 	}
183 
184 	switch (nr) {
185 	case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
186 	{
187 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
188 		/* Book3S can be little endian, find it out here */
189 		int shared_big_endian = true;
190 		if (vcpu->arch.intr_msr & MSR_LE)
191 			shared_big_endian = false;
192 		if (shared_big_endian != vcpu->arch.shared_big_endian)
193 			kvmppc_swab_shared(vcpu);
194 		vcpu->arch.shared_big_endian = shared_big_endian;
195 #endif
196 
197 		if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
198 			/*
199 			 * Older versions of the Linux magic page code had
200 			 * a bug where they would map their trampoline code
201 			 * NX. If that's the case, remove !PR NX capability.
202 			 */
203 			vcpu->arch.disable_kernel_nx = true;
204 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
205 		}
206 
207 		vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
208 		vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
209 
210 #ifdef CONFIG_PPC_64K_PAGES
211 		/*
212 		 * Make sure our 4k magic page is in the same window of a 64k
213 		 * page within the guest and within the host's page.
214 		 */
215 		if ((vcpu->arch.magic_page_pa & 0xf000) !=
216 		    ((ulong)vcpu->arch.shared & 0xf000)) {
217 			void *old_shared = vcpu->arch.shared;
218 			ulong shared = (ulong)vcpu->arch.shared;
219 			void *new_shared;
220 
221 			shared &= PAGE_MASK;
222 			shared |= vcpu->arch.magic_page_pa & 0xf000;
223 			new_shared = (void*)shared;
224 			memcpy(new_shared, old_shared, 0x1000);
225 			vcpu->arch.shared = new_shared;
226 		}
227 #endif
228 
229 		r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
230 
231 		r = EV_SUCCESS;
232 		break;
233 	}
234 	case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
235 		r = EV_SUCCESS;
236 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
237 		r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
238 #endif
239 
240 		/* Second return value is in r4 */
241 		break;
242 	case EV_HCALL_TOKEN(EV_IDLE):
243 		r = EV_SUCCESS;
244 		kvm_vcpu_block(vcpu);
245 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
246 		break;
247 	default:
248 		r = EV_UNIMPLEMENTED;
249 		break;
250 	}
251 
252 	kvmppc_set_gpr(vcpu, 4, r2);
253 
254 	return r;
255 }
256 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
257 
258 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
259 {
260 	int r = false;
261 
262 	/* We have to know what CPU to virtualize */
263 	if (!vcpu->arch.pvr)
264 		goto out;
265 
266 	/* PAPR only works with book3s_64 */
267 	if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
268 		goto out;
269 
270 	/* HV KVM can only do PAPR mode for now */
271 	if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
272 		goto out;
273 
274 #ifdef CONFIG_KVM_BOOKE_HV
275 	if (!cpu_has_feature(CPU_FTR_EMB_HV))
276 		goto out;
277 #endif
278 
279 	r = true;
280 
281 out:
282 	vcpu->arch.sane = r;
283 	return r ? 0 : -EINVAL;
284 }
285 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
286 
287 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
288 {
289 	enum emulation_result er;
290 	int r;
291 
292 	er = kvmppc_emulate_loadstore(vcpu);
293 	switch (er) {
294 	case EMULATE_DONE:
295 		/* Future optimization: only reload non-volatiles if they were
296 		 * actually modified. */
297 		r = RESUME_GUEST_NV;
298 		break;
299 	case EMULATE_AGAIN:
300 		r = RESUME_GUEST;
301 		break;
302 	case EMULATE_DO_MMIO:
303 		run->exit_reason = KVM_EXIT_MMIO;
304 		/* We must reload nonvolatiles because "update" load/store
305 		 * instructions modify register state. */
306 		/* Future optimization: only reload non-volatiles if they were
307 		 * actually modified. */
308 		r = RESUME_HOST_NV;
309 		break;
310 	case EMULATE_FAIL:
311 	{
312 		u32 last_inst;
313 
314 		kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
315 		/* XXX Deliver Program interrupt to guest. */
316 		pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
317 		r = RESUME_HOST;
318 		break;
319 	}
320 	default:
321 		WARN_ON(1);
322 		r = RESUME_GUEST;
323 	}
324 
325 	return r;
326 }
327 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
328 
329 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
330 	      bool data)
331 {
332 	ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
333 	struct kvmppc_pte pte;
334 	int r;
335 
336 	vcpu->stat.st++;
337 
338 	r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
339 			 XLATE_WRITE, &pte);
340 	if (r < 0)
341 		return r;
342 
343 	*eaddr = pte.raddr;
344 
345 	if (!pte.may_write)
346 		return -EPERM;
347 
348 	/* Magic page override */
349 	if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
350 	    ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
351 	    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
352 		void *magic = vcpu->arch.shared;
353 		magic += pte.eaddr & 0xfff;
354 		memcpy(magic, ptr, size);
355 		return EMULATE_DONE;
356 	}
357 
358 	if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
359 		return EMULATE_DO_MMIO;
360 
361 	return EMULATE_DONE;
362 }
363 EXPORT_SYMBOL_GPL(kvmppc_st);
364 
365 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
366 		      bool data)
367 {
368 	ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
369 	struct kvmppc_pte pte;
370 	int rc;
371 
372 	vcpu->stat.ld++;
373 
374 	rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
375 			  XLATE_READ, &pte);
376 	if (rc)
377 		return rc;
378 
379 	*eaddr = pte.raddr;
380 
381 	if (!pte.may_read)
382 		return -EPERM;
383 
384 	if (!data && !pte.may_execute)
385 		return -ENOEXEC;
386 
387 	/* Magic page override */
388 	if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
389 	    ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
390 	    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
391 		void *magic = vcpu->arch.shared;
392 		magic += pte.eaddr & 0xfff;
393 		memcpy(ptr, magic, size);
394 		return EMULATE_DONE;
395 	}
396 
397 	if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
398 		return EMULATE_DO_MMIO;
399 
400 	return EMULATE_DONE;
401 }
402 EXPORT_SYMBOL_GPL(kvmppc_ld);
403 
404 int kvm_arch_hardware_enable(void)
405 {
406 	return 0;
407 }
408 
409 int kvm_arch_hardware_setup(void)
410 {
411 	return 0;
412 }
413 
414 void kvm_arch_check_processor_compat(void *rtn)
415 {
416 	*(int *)rtn = kvmppc_core_check_processor_compat();
417 }
418 
419 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
420 {
421 	struct kvmppc_ops *kvm_ops = NULL;
422 	/*
423 	 * if we have both HV and PR enabled, default is HV
424 	 */
425 	if (type == 0) {
426 		if (kvmppc_hv_ops)
427 			kvm_ops = kvmppc_hv_ops;
428 		else
429 			kvm_ops = kvmppc_pr_ops;
430 		if (!kvm_ops)
431 			goto err_out;
432 	} else	if (type == KVM_VM_PPC_HV) {
433 		if (!kvmppc_hv_ops)
434 			goto err_out;
435 		kvm_ops = kvmppc_hv_ops;
436 	} else if (type == KVM_VM_PPC_PR) {
437 		if (!kvmppc_pr_ops)
438 			goto err_out;
439 		kvm_ops = kvmppc_pr_ops;
440 	} else
441 		goto err_out;
442 
443 	if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
444 		return -ENOENT;
445 
446 	kvm->arch.kvm_ops = kvm_ops;
447 	return kvmppc_core_init_vm(kvm);
448 err_out:
449 	return -EINVAL;
450 }
451 
452 bool kvm_arch_has_vcpu_debugfs(void)
453 {
454 	return false;
455 }
456 
457 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
458 {
459 	return 0;
460 }
461 
462 void kvm_arch_destroy_vm(struct kvm *kvm)
463 {
464 	unsigned int i;
465 	struct kvm_vcpu *vcpu;
466 
467 #ifdef CONFIG_KVM_XICS
468 	/*
469 	 * We call kick_all_cpus_sync() to ensure that all
470 	 * CPUs have executed any pending IPIs before we
471 	 * continue and free VCPUs structures below.
472 	 */
473 	if (is_kvmppc_hv_enabled(kvm))
474 		kick_all_cpus_sync();
475 #endif
476 
477 	kvm_for_each_vcpu(i, vcpu, kvm)
478 		kvm_arch_vcpu_free(vcpu);
479 
480 	mutex_lock(&kvm->lock);
481 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
482 		kvm->vcpus[i] = NULL;
483 
484 	atomic_set(&kvm->online_vcpus, 0);
485 
486 	kvmppc_core_destroy_vm(kvm);
487 
488 	mutex_unlock(&kvm->lock);
489 
490 	/* drop the module reference */
491 	module_put(kvm->arch.kvm_ops->owner);
492 }
493 
494 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
495 {
496 	int r;
497 	/* Assume we're using HV mode when the HV module is loaded */
498 	int hv_enabled = kvmppc_hv_ops ? 1 : 0;
499 
500 	if (kvm) {
501 		/*
502 		 * Hooray - we know which VM type we're running on. Depend on
503 		 * that rather than the guess above.
504 		 */
505 		hv_enabled = is_kvmppc_hv_enabled(kvm);
506 	}
507 
508 	switch (ext) {
509 #ifdef CONFIG_BOOKE
510 	case KVM_CAP_PPC_BOOKE_SREGS:
511 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
512 	case KVM_CAP_PPC_EPR:
513 #else
514 	case KVM_CAP_PPC_SEGSTATE:
515 	case KVM_CAP_PPC_HIOR:
516 	case KVM_CAP_PPC_PAPR:
517 #endif
518 	case KVM_CAP_PPC_UNSET_IRQ:
519 	case KVM_CAP_PPC_IRQ_LEVEL:
520 	case KVM_CAP_ENABLE_CAP:
521 	case KVM_CAP_ENABLE_CAP_VM:
522 	case KVM_CAP_ONE_REG:
523 	case KVM_CAP_IOEVENTFD:
524 	case KVM_CAP_DEVICE_CTRL:
525 	case KVM_CAP_IMMEDIATE_EXIT:
526 		r = 1;
527 		break;
528 	case KVM_CAP_PPC_PAIRED_SINGLES:
529 	case KVM_CAP_PPC_OSI:
530 	case KVM_CAP_PPC_GET_PVINFO:
531 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
532 	case KVM_CAP_SW_TLB:
533 #endif
534 		/* We support this only for PR */
535 		r = !hv_enabled;
536 		break;
537 #ifdef CONFIG_KVM_MPIC
538 	case KVM_CAP_IRQ_MPIC:
539 		r = 1;
540 		break;
541 #endif
542 
543 #ifdef CONFIG_PPC_BOOK3S_64
544 	case KVM_CAP_SPAPR_TCE:
545 	case KVM_CAP_SPAPR_TCE_64:
546 		/* fallthrough */
547 	case KVM_CAP_SPAPR_TCE_VFIO:
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 
558 	case KVM_CAP_PPC_ALLOC_HTAB:
559 		r = hv_enabled;
560 		break;
561 #endif /* CONFIG_PPC_BOOK3S_64 */
562 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
563 	case KVM_CAP_PPC_SMT:
564 		r = 0;
565 		if (kvm) {
566 			if (kvm->arch.emul_smt_mode > 1)
567 				r = kvm->arch.emul_smt_mode;
568 			else
569 				r = kvm->arch.smt_mode;
570 		} else if (hv_enabled) {
571 			if (cpu_has_feature(CPU_FTR_ARCH_300))
572 				r = 1;
573 			else
574 				r = threads_per_subcore;
575 		}
576 		break;
577 	case KVM_CAP_PPC_SMT_POSSIBLE:
578 		r = 1;
579 		if (hv_enabled) {
580 			if (!cpu_has_feature(CPU_FTR_ARCH_300))
581 				r = ((threads_per_subcore << 1) - 1);
582 			else
583 				/* P9 can emulate dbells, so allow any mode */
584 				r = 8 | 4 | 2 | 1;
585 		}
586 		break;
587 	case KVM_CAP_PPC_RMA:
588 		r = 0;
589 		break;
590 	case KVM_CAP_PPC_HWRNG:
591 		r = kvmppc_hwrng_present();
592 		break;
593 	case KVM_CAP_PPC_MMU_RADIX:
594 		r = !!(hv_enabled && radix_enabled());
595 		break;
596 	case KVM_CAP_PPC_MMU_HASH_V3:
597 		r = !!(hv_enabled && cpu_has_feature(CPU_FTR_ARCH_300) &&
598 		       cpu_has_feature(CPU_FTR_HVMODE));
599 		break;
600 	case KVM_CAP_PPC_NESTED_HV:
601 		r = !!(hv_enabled && kvmppc_hv_ops->enable_nested &&
602 		       !kvmppc_hv_ops->enable_nested(NULL));
603 		break;
604 #endif
605 	case KVM_CAP_SYNC_MMU:
606 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
607 		r = hv_enabled;
608 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
609 		r = 1;
610 #else
611 		r = 0;
612 #endif
613 		break;
614 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
615 	case KVM_CAP_PPC_HTAB_FD:
616 		r = hv_enabled;
617 		break;
618 #endif
619 	case KVM_CAP_NR_VCPUS:
620 		/*
621 		 * Recommending a number of CPUs is somewhat arbitrary; we
622 		 * return the number of present CPUs for -HV (since a host
623 		 * will have secondary threads "offline"), and for other KVM
624 		 * implementations just count online CPUs.
625 		 */
626 		if (hv_enabled)
627 			r = num_present_cpus();
628 		else
629 			r = num_online_cpus();
630 		break;
631 	case KVM_CAP_NR_MEMSLOTS:
632 		r = KVM_USER_MEM_SLOTS;
633 		break;
634 	case KVM_CAP_MAX_VCPUS:
635 		r = KVM_MAX_VCPUS;
636 		break;
637 #ifdef CONFIG_PPC_BOOK3S_64
638 	case KVM_CAP_PPC_GET_SMMU_INFO:
639 		r = 1;
640 		break;
641 	case KVM_CAP_SPAPR_MULTITCE:
642 		r = 1;
643 		break;
644 	case KVM_CAP_SPAPR_RESIZE_HPT:
645 		r = !!hv_enabled;
646 		break;
647 #endif
648 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
649 	case KVM_CAP_PPC_FWNMI:
650 		r = hv_enabled;
651 		break;
652 #endif
653 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
654 	case KVM_CAP_PPC_HTM:
655 		r = !!(cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM) ||
656 		     (hv_enabled && cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST));
657 		break;
658 #endif
659 	default:
660 		r = 0;
661 		break;
662 	}
663 	return r;
664 
665 }
666 
667 long kvm_arch_dev_ioctl(struct file *filp,
668                         unsigned int ioctl, unsigned long arg)
669 {
670 	return -EINVAL;
671 }
672 
673 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
674 			   struct kvm_memory_slot *dont)
675 {
676 	kvmppc_core_free_memslot(kvm, free, dont);
677 }
678 
679 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
680 			    unsigned long npages)
681 {
682 	return kvmppc_core_create_memslot(kvm, slot, npages);
683 }
684 
685 int kvm_arch_prepare_memory_region(struct kvm *kvm,
686 				   struct kvm_memory_slot *memslot,
687 				   const struct kvm_userspace_memory_region *mem,
688 				   enum kvm_mr_change change)
689 {
690 	return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
691 }
692 
693 void kvm_arch_commit_memory_region(struct kvm *kvm,
694 				   const struct kvm_userspace_memory_region *mem,
695 				   const struct kvm_memory_slot *old,
696 				   const struct kvm_memory_slot *new,
697 				   enum kvm_mr_change change)
698 {
699 	kvmppc_core_commit_memory_region(kvm, mem, old, new);
700 }
701 
702 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
703 				   struct kvm_memory_slot *slot)
704 {
705 	kvmppc_core_flush_memslot(kvm, slot);
706 }
707 
708 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
709 {
710 	struct kvm_vcpu *vcpu;
711 	vcpu = kvmppc_core_vcpu_create(kvm, id);
712 	if (!IS_ERR(vcpu)) {
713 		vcpu->arch.wqp = &vcpu->wq;
714 		kvmppc_create_vcpu_debugfs(vcpu, id);
715 	}
716 	return vcpu;
717 }
718 
719 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
720 {
721 }
722 
723 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
724 {
725 	/* Make sure we're not using the vcpu anymore */
726 	hrtimer_cancel(&vcpu->arch.dec_timer);
727 
728 	kvmppc_remove_vcpu_debugfs(vcpu);
729 
730 	switch (vcpu->arch.irq_type) {
731 	case KVMPPC_IRQ_MPIC:
732 		kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
733 		break;
734 	case KVMPPC_IRQ_XICS:
735 		if (xive_enabled())
736 			kvmppc_xive_cleanup_vcpu(vcpu);
737 		else
738 			kvmppc_xics_free_icp(vcpu);
739 		break;
740 	}
741 
742 	kvmppc_core_vcpu_free(vcpu);
743 }
744 
745 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
746 {
747 	kvm_arch_vcpu_free(vcpu);
748 }
749 
750 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
751 {
752 	return kvmppc_core_pending_dec(vcpu);
753 }
754 
755 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
756 {
757 	struct kvm_vcpu *vcpu;
758 
759 	vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
760 	kvmppc_decrementer_func(vcpu);
761 
762 	return HRTIMER_NORESTART;
763 }
764 
765 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
766 {
767 	int ret;
768 
769 	hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
770 	vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
771 	vcpu->arch.dec_expires = get_tb();
772 
773 #ifdef CONFIG_KVM_EXIT_TIMING
774 	mutex_init(&vcpu->arch.exit_timing_lock);
775 #endif
776 	ret = kvmppc_subarch_vcpu_init(vcpu);
777 	return ret;
778 }
779 
780 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
781 {
782 	kvmppc_mmu_destroy(vcpu);
783 	kvmppc_subarch_vcpu_uninit(vcpu);
784 }
785 
786 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
787 {
788 #ifdef CONFIG_BOOKE
789 	/*
790 	 * vrsave (formerly usprg0) isn't used by Linux, but may
791 	 * be used by the guest.
792 	 *
793 	 * On non-booke this is associated with Altivec and
794 	 * is handled by code in book3s.c.
795 	 */
796 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
797 #endif
798 	kvmppc_core_vcpu_load(vcpu, cpu);
799 }
800 
801 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
802 {
803 	kvmppc_core_vcpu_put(vcpu);
804 #ifdef CONFIG_BOOKE
805 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
806 #endif
807 }
808 
809 /*
810  * irq_bypass_add_producer and irq_bypass_del_producer are only
811  * useful if the architecture supports PCI passthrough.
812  * irq_bypass_stop and irq_bypass_start are not needed and so
813  * kvm_ops are not defined for them.
814  */
815 bool kvm_arch_has_irq_bypass(void)
816 {
817 	return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
818 		(kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
819 }
820 
821 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
822 				     struct irq_bypass_producer *prod)
823 {
824 	struct kvm_kernel_irqfd *irqfd =
825 		container_of(cons, struct kvm_kernel_irqfd, consumer);
826 	struct kvm *kvm = irqfd->kvm;
827 
828 	if (kvm->arch.kvm_ops->irq_bypass_add_producer)
829 		return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
830 
831 	return 0;
832 }
833 
834 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
835 				      struct irq_bypass_producer *prod)
836 {
837 	struct kvm_kernel_irqfd *irqfd =
838 		container_of(cons, struct kvm_kernel_irqfd, consumer);
839 	struct kvm *kvm = irqfd->kvm;
840 
841 	if (kvm->arch.kvm_ops->irq_bypass_del_producer)
842 		kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
843 }
844 
845 #ifdef CONFIG_VSX
846 static inline int kvmppc_get_vsr_dword_offset(int index)
847 {
848 	int offset;
849 
850 	if ((index != 0) && (index != 1))
851 		return -1;
852 
853 #ifdef __BIG_ENDIAN
854 	offset =  index;
855 #else
856 	offset = 1 - index;
857 #endif
858 
859 	return offset;
860 }
861 
862 static inline int kvmppc_get_vsr_word_offset(int index)
863 {
864 	int offset;
865 
866 	if ((index > 3) || (index < 0))
867 		return -1;
868 
869 #ifdef __BIG_ENDIAN
870 	offset = index;
871 #else
872 	offset = 3 - index;
873 #endif
874 	return offset;
875 }
876 
877 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
878 	u64 gpr)
879 {
880 	union kvmppc_one_reg val;
881 	int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
882 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
883 
884 	if (offset == -1)
885 		return;
886 
887 	if (index >= 32) {
888 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
889 		val.vsxval[offset] = gpr;
890 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
891 	} else {
892 		VCPU_VSX_FPR(vcpu, index, offset) = gpr;
893 	}
894 }
895 
896 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
897 	u64 gpr)
898 {
899 	union kvmppc_one_reg val;
900 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
901 
902 	if (index >= 32) {
903 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
904 		val.vsxval[0] = gpr;
905 		val.vsxval[1] = gpr;
906 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
907 	} else {
908 		VCPU_VSX_FPR(vcpu, index, 0) = gpr;
909 		VCPU_VSX_FPR(vcpu, index, 1) = gpr;
910 	}
911 }
912 
913 static inline void kvmppc_set_vsr_word_dump(struct kvm_vcpu *vcpu,
914 	u32 gpr)
915 {
916 	union kvmppc_one_reg val;
917 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
918 
919 	if (index >= 32) {
920 		val.vsx32val[0] = gpr;
921 		val.vsx32val[1] = gpr;
922 		val.vsx32val[2] = gpr;
923 		val.vsx32val[3] = gpr;
924 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
925 	} else {
926 		val.vsx32val[0] = gpr;
927 		val.vsx32val[1] = gpr;
928 		VCPU_VSX_FPR(vcpu, index, 0) = val.vsxval[0];
929 		VCPU_VSX_FPR(vcpu, index, 1) = val.vsxval[0];
930 	}
931 }
932 
933 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
934 	u32 gpr32)
935 {
936 	union kvmppc_one_reg val;
937 	int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
938 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
939 	int dword_offset, word_offset;
940 
941 	if (offset == -1)
942 		return;
943 
944 	if (index >= 32) {
945 		val.vval = VCPU_VSX_VR(vcpu, index - 32);
946 		val.vsx32val[offset] = gpr32;
947 		VCPU_VSX_VR(vcpu, index - 32) = val.vval;
948 	} else {
949 		dword_offset = offset / 2;
950 		word_offset = offset % 2;
951 		val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
952 		val.vsx32val[word_offset] = gpr32;
953 		VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
954 	}
955 }
956 #endif /* CONFIG_VSX */
957 
958 #ifdef CONFIG_ALTIVEC
959 static inline int kvmppc_get_vmx_offset_generic(struct kvm_vcpu *vcpu,
960 		int index, int element_size)
961 {
962 	int offset;
963 	int elts = sizeof(vector128)/element_size;
964 
965 	if ((index < 0) || (index >= elts))
966 		return -1;
967 
968 	if (kvmppc_need_byteswap(vcpu))
969 		offset = elts - index - 1;
970 	else
971 		offset = index;
972 
973 	return offset;
974 }
975 
976 static inline int kvmppc_get_vmx_dword_offset(struct kvm_vcpu *vcpu,
977 		int index)
978 {
979 	return kvmppc_get_vmx_offset_generic(vcpu, index, 8);
980 }
981 
982 static inline int kvmppc_get_vmx_word_offset(struct kvm_vcpu *vcpu,
983 		int index)
984 {
985 	return kvmppc_get_vmx_offset_generic(vcpu, index, 4);
986 }
987 
988 static inline int kvmppc_get_vmx_hword_offset(struct kvm_vcpu *vcpu,
989 		int index)
990 {
991 	return kvmppc_get_vmx_offset_generic(vcpu, index, 2);
992 }
993 
994 static inline int kvmppc_get_vmx_byte_offset(struct kvm_vcpu *vcpu,
995 		int index)
996 {
997 	return kvmppc_get_vmx_offset_generic(vcpu, index, 1);
998 }
999 
1000 
1001 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu,
1002 	u64 gpr)
1003 {
1004 	union kvmppc_one_reg val;
1005 	int offset = kvmppc_get_vmx_dword_offset(vcpu,
1006 			vcpu->arch.mmio_vmx_offset);
1007 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1008 
1009 	if (offset == -1)
1010 		return;
1011 
1012 	val.vval = VCPU_VSX_VR(vcpu, index);
1013 	val.vsxval[offset] = gpr;
1014 	VCPU_VSX_VR(vcpu, index) = val.vval;
1015 }
1016 
1017 static inline void kvmppc_set_vmx_word(struct kvm_vcpu *vcpu,
1018 	u32 gpr32)
1019 {
1020 	union kvmppc_one_reg val;
1021 	int offset = kvmppc_get_vmx_word_offset(vcpu,
1022 			vcpu->arch.mmio_vmx_offset);
1023 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1024 
1025 	if (offset == -1)
1026 		return;
1027 
1028 	val.vval = VCPU_VSX_VR(vcpu, index);
1029 	val.vsx32val[offset] = gpr32;
1030 	VCPU_VSX_VR(vcpu, index) = val.vval;
1031 }
1032 
1033 static inline void kvmppc_set_vmx_hword(struct kvm_vcpu *vcpu,
1034 	u16 gpr16)
1035 {
1036 	union kvmppc_one_reg val;
1037 	int offset = kvmppc_get_vmx_hword_offset(vcpu,
1038 			vcpu->arch.mmio_vmx_offset);
1039 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1040 
1041 	if (offset == -1)
1042 		return;
1043 
1044 	val.vval = VCPU_VSX_VR(vcpu, index);
1045 	val.vsx16val[offset] = gpr16;
1046 	VCPU_VSX_VR(vcpu, index) = val.vval;
1047 }
1048 
1049 static inline void kvmppc_set_vmx_byte(struct kvm_vcpu *vcpu,
1050 	u8 gpr8)
1051 {
1052 	union kvmppc_one_reg val;
1053 	int offset = kvmppc_get_vmx_byte_offset(vcpu,
1054 			vcpu->arch.mmio_vmx_offset);
1055 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1056 
1057 	if (offset == -1)
1058 		return;
1059 
1060 	val.vval = VCPU_VSX_VR(vcpu, index);
1061 	val.vsx8val[offset] = gpr8;
1062 	VCPU_VSX_VR(vcpu, index) = val.vval;
1063 }
1064 #endif /* CONFIG_ALTIVEC */
1065 
1066 #ifdef CONFIG_PPC_FPU
1067 static inline u64 sp_to_dp(u32 fprs)
1068 {
1069 	u64 fprd;
1070 
1071 	preempt_disable();
1072 	enable_kernel_fp();
1073 	asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs)
1074 	     : "fr0");
1075 	preempt_enable();
1076 	return fprd;
1077 }
1078 
1079 static inline u32 dp_to_sp(u64 fprd)
1080 {
1081 	u32 fprs;
1082 
1083 	preempt_disable();
1084 	enable_kernel_fp();
1085 	asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd)
1086 	     : "fr0");
1087 	preempt_enable();
1088 	return fprs;
1089 }
1090 
1091 #else
1092 #define sp_to_dp(x)	(x)
1093 #define dp_to_sp(x)	(x)
1094 #endif /* CONFIG_PPC_FPU */
1095 
1096 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
1097                                       struct kvm_run *run)
1098 {
1099 	u64 uninitialized_var(gpr);
1100 
1101 	if (run->mmio.len > sizeof(gpr)) {
1102 		printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
1103 		return;
1104 	}
1105 
1106 	if (!vcpu->arch.mmio_host_swabbed) {
1107 		switch (run->mmio.len) {
1108 		case 8: gpr = *(u64 *)run->mmio.data; break;
1109 		case 4: gpr = *(u32 *)run->mmio.data; break;
1110 		case 2: gpr = *(u16 *)run->mmio.data; break;
1111 		case 1: gpr = *(u8 *)run->mmio.data; break;
1112 		}
1113 	} else {
1114 		switch (run->mmio.len) {
1115 		case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
1116 		case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
1117 		case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
1118 		case 1: gpr = *(u8 *)run->mmio.data; break;
1119 		}
1120 	}
1121 
1122 	/* conversion between single and double precision */
1123 	if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
1124 		gpr = sp_to_dp(gpr);
1125 
1126 	if (vcpu->arch.mmio_sign_extend) {
1127 		switch (run->mmio.len) {
1128 #ifdef CONFIG_PPC64
1129 		case 4:
1130 			gpr = (s64)(s32)gpr;
1131 			break;
1132 #endif
1133 		case 2:
1134 			gpr = (s64)(s16)gpr;
1135 			break;
1136 		case 1:
1137 			gpr = (s64)(s8)gpr;
1138 			break;
1139 		}
1140 	}
1141 
1142 	switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1143 	case KVM_MMIO_REG_GPR:
1144 		kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1145 		break;
1146 	case KVM_MMIO_REG_FPR:
1147 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1148 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_FP);
1149 
1150 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1151 		break;
1152 #ifdef CONFIG_PPC_BOOK3S
1153 	case KVM_MMIO_REG_QPR:
1154 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1155 		break;
1156 	case KVM_MMIO_REG_FQPR:
1157 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1158 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1159 		break;
1160 #endif
1161 #ifdef CONFIG_VSX
1162 	case KVM_MMIO_REG_VSX:
1163 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1164 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VSX);
1165 
1166 		if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_DWORD)
1167 			kvmppc_set_vsr_dword(vcpu, gpr);
1168 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_WORD)
1169 			kvmppc_set_vsr_word(vcpu, gpr);
1170 		else if (vcpu->arch.mmio_copy_type ==
1171 				KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1172 			kvmppc_set_vsr_dword_dump(vcpu, gpr);
1173 		else if (vcpu->arch.mmio_copy_type ==
1174 				KVMPPC_VSX_COPY_WORD_LOAD_DUMP)
1175 			kvmppc_set_vsr_word_dump(vcpu, gpr);
1176 		break;
1177 #endif
1178 #ifdef CONFIG_ALTIVEC
1179 	case KVM_MMIO_REG_VMX:
1180 		if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1181 			vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VEC);
1182 
1183 		if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_DWORD)
1184 			kvmppc_set_vmx_dword(vcpu, gpr);
1185 		else if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_WORD)
1186 			kvmppc_set_vmx_word(vcpu, gpr);
1187 		else if (vcpu->arch.mmio_copy_type ==
1188 				KVMPPC_VMX_COPY_HWORD)
1189 			kvmppc_set_vmx_hword(vcpu, gpr);
1190 		else if (vcpu->arch.mmio_copy_type ==
1191 				KVMPPC_VMX_COPY_BYTE)
1192 			kvmppc_set_vmx_byte(vcpu, gpr);
1193 		break;
1194 #endif
1195 	default:
1196 		BUG();
1197 	}
1198 }
1199 
1200 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1201 				unsigned int rt, unsigned int bytes,
1202 				int is_default_endian, int sign_extend)
1203 {
1204 	int idx, ret;
1205 	bool host_swabbed;
1206 
1207 	/* Pity C doesn't have a logical XOR operator */
1208 	if (kvmppc_need_byteswap(vcpu)) {
1209 		host_swabbed = is_default_endian;
1210 	} else {
1211 		host_swabbed = !is_default_endian;
1212 	}
1213 
1214 	if (bytes > sizeof(run->mmio.data)) {
1215 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1216 		       run->mmio.len);
1217 	}
1218 
1219 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1220 	run->mmio.len = bytes;
1221 	run->mmio.is_write = 0;
1222 
1223 	vcpu->arch.io_gpr = rt;
1224 	vcpu->arch.mmio_host_swabbed = host_swabbed;
1225 	vcpu->mmio_needed = 1;
1226 	vcpu->mmio_is_write = 0;
1227 	vcpu->arch.mmio_sign_extend = sign_extend;
1228 
1229 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1230 
1231 	ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1232 			      bytes, &run->mmio.data);
1233 
1234 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1235 
1236 	if (!ret) {
1237 		kvmppc_complete_mmio_load(vcpu, run);
1238 		vcpu->mmio_needed = 0;
1239 		return EMULATE_DONE;
1240 	}
1241 
1242 	return EMULATE_DO_MMIO;
1243 }
1244 
1245 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1246 		       unsigned int rt, unsigned int bytes,
1247 		       int is_default_endian)
1248 {
1249 	return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
1250 }
1251 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1252 
1253 /* Same as above, but sign extends */
1254 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
1255 			unsigned int rt, unsigned int bytes,
1256 			int is_default_endian)
1257 {
1258 	return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
1259 }
1260 
1261 #ifdef CONFIG_VSX
1262 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1263 			unsigned int rt, unsigned int bytes,
1264 			int is_default_endian, int mmio_sign_extend)
1265 {
1266 	enum emulation_result emulated = EMULATE_DONE;
1267 
1268 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1269 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1270 		return EMULATE_FAIL;
1271 
1272 	while (vcpu->arch.mmio_vsx_copy_nums) {
1273 		emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1274 			is_default_endian, mmio_sign_extend);
1275 
1276 		if (emulated != EMULATE_DONE)
1277 			break;
1278 
1279 		vcpu->arch.paddr_accessed += run->mmio.len;
1280 
1281 		vcpu->arch.mmio_vsx_copy_nums--;
1282 		vcpu->arch.mmio_vsx_offset++;
1283 	}
1284 	return emulated;
1285 }
1286 #endif /* CONFIG_VSX */
1287 
1288 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1289 			u64 val, unsigned int bytes, int is_default_endian)
1290 {
1291 	void *data = run->mmio.data;
1292 	int idx, ret;
1293 	bool host_swabbed;
1294 
1295 	/* Pity C doesn't have a logical XOR operator */
1296 	if (kvmppc_need_byteswap(vcpu)) {
1297 		host_swabbed = is_default_endian;
1298 	} else {
1299 		host_swabbed = !is_default_endian;
1300 	}
1301 
1302 	if (bytes > sizeof(run->mmio.data)) {
1303 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1304 		       run->mmio.len);
1305 	}
1306 
1307 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1308 	run->mmio.len = bytes;
1309 	run->mmio.is_write = 1;
1310 	vcpu->mmio_needed = 1;
1311 	vcpu->mmio_is_write = 1;
1312 
1313 	if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1314 		val = dp_to_sp(val);
1315 
1316 	/* Store the value at the lowest bytes in 'data'. */
1317 	if (!host_swabbed) {
1318 		switch (bytes) {
1319 		case 8: *(u64 *)data = val; break;
1320 		case 4: *(u32 *)data = val; break;
1321 		case 2: *(u16 *)data = val; break;
1322 		case 1: *(u8  *)data = val; break;
1323 		}
1324 	} else {
1325 		switch (bytes) {
1326 		case 8: *(u64 *)data = swab64(val); break;
1327 		case 4: *(u32 *)data = swab32(val); break;
1328 		case 2: *(u16 *)data = swab16(val); break;
1329 		case 1: *(u8  *)data = val; break;
1330 		}
1331 	}
1332 
1333 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1334 
1335 	ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1336 			       bytes, &run->mmio.data);
1337 
1338 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1339 
1340 	if (!ret) {
1341 		vcpu->mmio_needed = 0;
1342 		return EMULATE_DONE;
1343 	}
1344 
1345 	return EMULATE_DO_MMIO;
1346 }
1347 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1348 
1349 #ifdef CONFIG_VSX
1350 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1351 {
1352 	u32 dword_offset, word_offset;
1353 	union kvmppc_one_reg reg;
1354 	int vsx_offset = 0;
1355 	int copy_type = vcpu->arch.mmio_copy_type;
1356 	int result = 0;
1357 
1358 	switch (copy_type) {
1359 	case KVMPPC_VSX_COPY_DWORD:
1360 		vsx_offset =
1361 			kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1362 
1363 		if (vsx_offset == -1) {
1364 			result = -1;
1365 			break;
1366 		}
1367 
1368 		if (rs < 32) {
1369 			*val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1370 		} else {
1371 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1372 			*val = reg.vsxval[vsx_offset];
1373 		}
1374 		break;
1375 
1376 	case KVMPPC_VSX_COPY_WORD:
1377 		vsx_offset =
1378 			kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1379 
1380 		if (vsx_offset == -1) {
1381 			result = -1;
1382 			break;
1383 		}
1384 
1385 		if (rs < 32) {
1386 			dword_offset = vsx_offset / 2;
1387 			word_offset = vsx_offset % 2;
1388 			reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1389 			*val = reg.vsx32val[word_offset];
1390 		} else {
1391 			reg.vval = VCPU_VSX_VR(vcpu, rs - 32);
1392 			*val = reg.vsx32val[vsx_offset];
1393 		}
1394 		break;
1395 
1396 	default:
1397 		result = -1;
1398 		break;
1399 	}
1400 
1401 	return result;
1402 }
1403 
1404 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1405 			int rs, unsigned int bytes, int is_default_endian)
1406 {
1407 	u64 val;
1408 	enum emulation_result emulated = EMULATE_DONE;
1409 
1410 	vcpu->arch.io_gpr = rs;
1411 
1412 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1413 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1414 		return EMULATE_FAIL;
1415 
1416 	while (vcpu->arch.mmio_vsx_copy_nums) {
1417 		if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1418 			return EMULATE_FAIL;
1419 
1420 		emulated = kvmppc_handle_store(run, vcpu,
1421 			 val, bytes, is_default_endian);
1422 
1423 		if (emulated != EMULATE_DONE)
1424 			break;
1425 
1426 		vcpu->arch.paddr_accessed += run->mmio.len;
1427 
1428 		vcpu->arch.mmio_vsx_copy_nums--;
1429 		vcpu->arch.mmio_vsx_offset++;
1430 	}
1431 
1432 	return emulated;
1433 }
1434 
1435 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu,
1436 			struct kvm_run *run)
1437 {
1438 	enum emulation_result emulated = EMULATE_FAIL;
1439 	int r;
1440 
1441 	vcpu->arch.paddr_accessed += run->mmio.len;
1442 
1443 	if (!vcpu->mmio_is_write) {
1444 		emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr,
1445 			 run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1446 	} else {
1447 		emulated = kvmppc_handle_vsx_store(run, vcpu,
1448 			 vcpu->arch.io_gpr, run->mmio.len, 1);
1449 	}
1450 
1451 	switch (emulated) {
1452 	case EMULATE_DO_MMIO:
1453 		run->exit_reason = KVM_EXIT_MMIO;
1454 		r = RESUME_HOST;
1455 		break;
1456 	case EMULATE_FAIL:
1457 		pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1458 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1459 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1460 		r = RESUME_HOST;
1461 		break;
1462 	default:
1463 		r = RESUME_GUEST;
1464 		break;
1465 	}
1466 	return r;
1467 }
1468 #endif /* CONFIG_VSX */
1469 
1470 #ifdef CONFIG_ALTIVEC
1471 int kvmppc_handle_vmx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1472 		unsigned int rt, unsigned int bytes, int is_default_endian)
1473 {
1474 	enum emulation_result emulated = EMULATE_DONE;
1475 
1476 	if (vcpu->arch.mmio_vsx_copy_nums > 2)
1477 		return EMULATE_FAIL;
1478 
1479 	while (vcpu->arch.mmio_vmx_copy_nums) {
1480 		emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1481 				is_default_endian, 0);
1482 
1483 		if (emulated != EMULATE_DONE)
1484 			break;
1485 
1486 		vcpu->arch.paddr_accessed += run->mmio.len;
1487 		vcpu->arch.mmio_vmx_copy_nums--;
1488 		vcpu->arch.mmio_vmx_offset++;
1489 	}
1490 
1491 	return emulated;
1492 }
1493 
1494 int kvmppc_get_vmx_dword(struct kvm_vcpu *vcpu, int index, u64 *val)
1495 {
1496 	union kvmppc_one_reg reg;
1497 	int vmx_offset = 0;
1498 	int result = 0;
1499 
1500 	vmx_offset =
1501 		kvmppc_get_vmx_dword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1502 
1503 	if (vmx_offset == -1)
1504 		return -1;
1505 
1506 	reg.vval = VCPU_VSX_VR(vcpu, index);
1507 	*val = reg.vsxval[vmx_offset];
1508 
1509 	return result;
1510 }
1511 
1512 int kvmppc_get_vmx_word(struct kvm_vcpu *vcpu, int index, u64 *val)
1513 {
1514 	union kvmppc_one_reg reg;
1515 	int vmx_offset = 0;
1516 	int result = 0;
1517 
1518 	vmx_offset =
1519 		kvmppc_get_vmx_word_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1520 
1521 	if (vmx_offset == -1)
1522 		return -1;
1523 
1524 	reg.vval = VCPU_VSX_VR(vcpu, index);
1525 	*val = reg.vsx32val[vmx_offset];
1526 
1527 	return result;
1528 }
1529 
1530 int kvmppc_get_vmx_hword(struct kvm_vcpu *vcpu, int index, u64 *val)
1531 {
1532 	union kvmppc_one_reg reg;
1533 	int vmx_offset = 0;
1534 	int result = 0;
1535 
1536 	vmx_offset =
1537 		kvmppc_get_vmx_hword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1538 
1539 	if (vmx_offset == -1)
1540 		return -1;
1541 
1542 	reg.vval = VCPU_VSX_VR(vcpu, index);
1543 	*val = reg.vsx16val[vmx_offset];
1544 
1545 	return result;
1546 }
1547 
1548 int kvmppc_get_vmx_byte(struct kvm_vcpu *vcpu, int index, u64 *val)
1549 {
1550 	union kvmppc_one_reg reg;
1551 	int vmx_offset = 0;
1552 	int result = 0;
1553 
1554 	vmx_offset =
1555 		kvmppc_get_vmx_byte_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1556 
1557 	if (vmx_offset == -1)
1558 		return -1;
1559 
1560 	reg.vval = VCPU_VSX_VR(vcpu, index);
1561 	*val = reg.vsx8val[vmx_offset];
1562 
1563 	return result;
1564 }
1565 
1566 int kvmppc_handle_vmx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1567 		unsigned int rs, unsigned int bytes, int is_default_endian)
1568 {
1569 	u64 val = 0;
1570 	unsigned int index = rs & KVM_MMIO_REG_MASK;
1571 	enum emulation_result emulated = EMULATE_DONE;
1572 
1573 	if (vcpu->arch.mmio_vsx_copy_nums > 2)
1574 		return EMULATE_FAIL;
1575 
1576 	vcpu->arch.io_gpr = rs;
1577 
1578 	while (vcpu->arch.mmio_vmx_copy_nums) {
1579 		switch (vcpu->arch.mmio_copy_type) {
1580 		case KVMPPC_VMX_COPY_DWORD:
1581 			if (kvmppc_get_vmx_dword(vcpu, index, &val) == -1)
1582 				return EMULATE_FAIL;
1583 
1584 			break;
1585 		case KVMPPC_VMX_COPY_WORD:
1586 			if (kvmppc_get_vmx_word(vcpu, index, &val) == -1)
1587 				return EMULATE_FAIL;
1588 			break;
1589 		case KVMPPC_VMX_COPY_HWORD:
1590 			if (kvmppc_get_vmx_hword(vcpu, index, &val) == -1)
1591 				return EMULATE_FAIL;
1592 			break;
1593 		case KVMPPC_VMX_COPY_BYTE:
1594 			if (kvmppc_get_vmx_byte(vcpu, index, &val) == -1)
1595 				return EMULATE_FAIL;
1596 			break;
1597 		default:
1598 			return EMULATE_FAIL;
1599 		}
1600 
1601 		emulated = kvmppc_handle_store(run, vcpu, val, bytes,
1602 				is_default_endian);
1603 		if (emulated != EMULATE_DONE)
1604 			break;
1605 
1606 		vcpu->arch.paddr_accessed += run->mmio.len;
1607 		vcpu->arch.mmio_vmx_copy_nums--;
1608 		vcpu->arch.mmio_vmx_offset++;
1609 	}
1610 
1611 	return emulated;
1612 }
1613 
1614 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu,
1615 		struct kvm_run *run)
1616 {
1617 	enum emulation_result emulated = EMULATE_FAIL;
1618 	int r;
1619 
1620 	vcpu->arch.paddr_accessed += run->mmio.len;
1621 
1622 	if (!vcpu->mmio_is_write) {
1623 		emulated = kvmppc_handle_vmx_load(run, vcpu,
1624 				vcpu->arch.io_gpr, run->mmio.len, 1);
1625 	} else {
1626 		emulated = kvmppc_handle_vmx_store(run, vcpu,
1627 				vcpu->arch.io_gpr, run->mmio.len, 1);
1628 	}
1629 
1630 	switch (emulated) {
1631 	case EMULATE_DO_MMIO:
1632 		run->exit_reason = KVM_EXIT_MMIO;
1633 		r = RESUME_HOST;
1634 		break;
1635 	case EMULATE_FAIL:
1636 		pr_info("KVM: MMIO emulation failed (VMX repeat)\n");
1637 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1638 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1639 		r = RESUME_HOST;
1640 		break;
1641 	default:
1642 		r = RESUME_GUEST;
1643 		break;
1644 	}
1645 	return r;
1646 }
1647 #endif /* CONFIG_ALTIVEC */
1648 
1649 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1650 {
1651 	int r = 0;
1652 	union kvmppc_one_reg val;
1653 	int size;
1654 
1655 	size = one_reg_size(reg->id);
1656 	if (size > sizeof(val))
1657 		return -EINVAL;
1658 
1659 	r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1660 	if (r == -EINVAL) {
1661 		r = 0;
1662 		switch (reg->id) {
1663 #ifdef CONFIG_ALTIVEC
1664 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1665 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1666 				r = -ENXIO;
1667 				break;
1668 			}
1669 			val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1670 			break;
1671 		case KVM_REG_PPC_VSCR:
1672 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1673 				r = -ENXIO;
1674 				break;
1675 			}
1676 			val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1677 			break;
1678 		case KVM_REG_PPC_VRSAVE:
1679 			val = get_reg_val(reg->id, vcpu->arch.vrsave);
1680 			break;
1681 #endif /* CONFIG_ALTIVEC */
1682 		default:
1683 			r = -EINVAL;
1684 			break;
1685 		}
1686 	}
1687 
1688 	if (r)
1689 		return r;
1690 
1691 	if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1692 		r = -EFAULT;
1693 
1694 	return r;
1695 }
1696 
1697 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1698 {
1699 	int r;
1700 	union kvmppc_one_reg val;
1701 	int size;
1702 
1703 	size = one_reg_size(reg->id);
1704 	if (size > sizeof(val))
1705 		return -EINVAL;
1706 
1707 	if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1708 		return -EFAULT;
1709 
1710 	r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1711 	if (r == -EINVAL) {
1712 		r = 0;
1713 		switch (reg->id) {
1714 #ifdef CONFIG_ALTIVEC
1715 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1716 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1717 				r = -ENXIO;
1718 				break;
1719 			}
1720 			vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1721 			break;
1722 		case KVM_REG_PPC_VSCR:
1723 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1724 				r = -ENXIO;
1725 				break;
1726 			}
1727 			vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1728 			break;
1729 		case KVM_REG_PPC_VRSAVE:
1730 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1731 				r = -ENXIO;
1732 				break;
1733 			}
1734 			vcpu->arch.vrsave = set_reg_val(reg->id, val);
1735 			break;
1736 #endif /* CONFIG_ALTIVEC */
1737 		default:
1738 			r = -EINVAL;
1739 			break;
1740 		}
1741 	}
1742 
1743 	return r;
1744 }
1745 
1746 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1747 {
1748 	int r;
1749 
1750 	vcpu_load(vcpu);
1751 
1752 	if (vcpu->mmio_needed) {
1753 		vcpu->mmio_needed = 0;
1754 		if (!vcpu->mmio_is_write)
1755 			kvmppc_complete_mmio_load(vcpu, run);
1756 #ifdef CONFIG_VSX
1757 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1758 			vcpu->arch.mmio_vsx_copy_nums--;
1759 			vcpu->arch.mmio_vsx_offset++;
1760 		}
1761 
1762 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1763 			r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run);
1764 			if (r == RESUME_HOST) {
1765 				vcpu->mmio_needed = 1;
1766 				goto out;
1767 			}
1768 		}
1769 #endif
1770 #ifdef CONFIG_ALTIVEC
1771 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1772 			vcpu->arch.mmio_vmx_copy_nums--;
1773 			vcpu->arch.mmio_vmx_offset++;
1774 		}
1775 
1776 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1777 			r = kvmppc_emulate_mmio_vmx_loadstore(vcpu, run);
1778 			if (r == RESUME_HOST) {
1779 				vcpu->mmio_needed = 1;
1780 				goto out;
1781 			}
1782 		}
1783 #endif
1784 	} else if (vcpu->arch.osi_needed) {
1785 		u64 *gprs = run->osi.gprs;
1786 		int i;
1787 
1788 		for (i = 0; i < 32; i++)
1789 			kvmppc_set_gpr(vcpu, i, gprs[i]);
1790 		vcpu->arch.osi_needed = 0;
1791 	} else if (vcpu->arch.hcall_needed) {
1792 		int i;
1793 
1794 		kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1795 		for (i = 0; i < 9; ++i)
1796 			kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1797 		vcpu->arch.hcall_needed = 0;
1798 #ifdef CONFIG_BOOKE
1799 	} else if (vcpu->arch.epr_needed) {
1800 		kvmppc_set_epr(vcpu, run->epr.epr);
1801 		vcpu->arch.epr_needed = 0;
1802 #endif
1803 	}
1804 
1805 	kvm_sigset_activate(vcpu);
1806 
1807 	if (run->immediate_exit)
1808 		r = -EINTR;
1809 	else
1810 		r = kvmppc_vcpu_run(run, vcpu);
1811 
1812 	kvm_sigset_deactivate(vcpu);
1813 
1814 #ifdef CONFIG_ALTIVEC
1815 out:
1816 #endif
1817 	vcpu_put(vcpu);
1818 	return r;
1819 }
1820 
1821 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1822 {
1823 	if (irq->irq == KVM_INTERRUPT_UNSET) {
1824 		kvmppc_core_dequeue_external(vcpu);
1825 		return 0;
1826 	}
1827 
1828 	kvmppc_core_queue_external(vcpu, irq);
1829 
1830 	kvm_vcpu_kick(vcpu);
1831 
1832 	return 0;
1833 }
1834 
1835 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1836 				     struct kvm_enable_cap *cap)
1837 {
1838 	int r;
1839 
1840 	if (cap->flags)
1841 		return -EINVAL;
1842 
1843 	switch (cap->cap) {
1844 	case KVM_CAP_PPC_OSI:
1845 		r = 0;
1846 		vcpu->arch.osi_enabled = true;
1847 		break;
1848 	case KVM_CAP_PPC_PAPR:
1849 		r = 0;
1850 		vcpu->arch.papr_enabled = true;
1851 		break;
1852 	case KVM_CAP_PPC_EPR:
1853 		r = 0;
1854 		if (cap->args[0])
1855 			vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1856 		else
1857 			vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1858 		break;
1859 #ifdef CONFIG_BOOKE
1860 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
1861 		r = 0;
1862 		vcpu->arch.watchdog_enabled = true;
1863 		break;
1864 #endif
1865 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1866 	case KVM_CAP_SW_TLB: {
1867 		struct kvm_config_tlb cfg;
1868 		void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1869 
1870 		r = -EFAULT;
1871 		if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1872 			break;
1873 
1874 		r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1875 		break;
1876 	}
1877 #endif
1878 #ifdef CONFIG_KVM_MPIC
1879 	case KVM_CAP_IRQ_MPIC: {
1880 		struct fd f;
1881 		struct kvm_device *dev;
1882 
1883 		r = -EBADF;
1884 		f = fdget(cap->args[0]);
1885 		if (!f.file)
1886 			break;
1887 
1888 		r = -EPERM;
1889 		dev = kvm_device_from_filp(f.file);
1890 		if (dev)
1891 			r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1892 
1893 		fdput(f);
1894 		break;
1895 	}
1896 #endif
1897 #ifdef CONFIG_KVM_XICS
1898 	case KVM_CAP_IRQ_XICS: {
1899 		struct fd f;
1900 		struct kvm_device *dev;
1901 
1902 		r = -EBADF;
1903 		f = fdget(cap->args[0]);
1904 		if (!f.file)
1905 			break;
1906 
1907 		r = -EPERM;
1908 		dev = kvm_device_from_filp(f.file);
1909 		if (dev) {
1910 			if (xive_enabled())
1911 				r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1912 			else
1913 				r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1914 		}
1915 
1916 		fdput(f);
1917 		break;
1918 	}
1919 #endif /* CONFIG_KVM_XICS */
1920 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1921 	case KVM_CAP_PPC_FWNMI:
1922 		r = -EINVAL;
1923 		if (!is_kvmppc_hv_enabled(vcpu->kvm))
1924 			break;
1925 		r = 0;
1926 		vcpu->kvm->arch.fwnmi_enabled = true;
1927 		break;
1928 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1929 	default:
1930 		r = -EINVAL;
1931 		break;
1932 	}
1933 
1934 	if (!r)
1935 		r = kvmppc_sanity_check(vcpu);
1936 
1937 	return r;
1938 }
1939 
1940 bool kvm_arch_intc_initialized(struct kvm *kvm)
1941 {
1942 #ifdef CONFIG_KVM_MPIC
1943 	if (kvm->arch.mpic)
1944 		return true;
1945 #endif
1946 #ifdef CONFIG_KVM_XICS
1947 	if (kvm->arch.xics || kvm->arch.xive)
1948 		return true;
1949 #endif
1950 	return false;
1951 }
1952 
1953 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1954                                     struct kvm_mp_state *mp_state)
1955 {
1956 	return -EINVAL;
1957 }
1958 
1959 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1960                                     struct kvm_mp_state *mp_state)
1961 {
1962 	return -EINVAL;
1963 }
1964 
1965 long kvm_arch_vcpu_async_ioctl(struct file *filp,
1966 			       unsigned int ioctl, unsigned long arg)
1967 {
1968 	struct kvm_vcpu *vcpu = filp->private_data;
1969 	void __user *argp = (void __user *)arg;
1970 
1971 	if (ioctl == KVM_INTERRUPT) {
1972 		struct kvm_interrupt irq;
1973 		if (copy_from_user(&irq, argp, sizeof(irq)))
1974 			return -EFAULT;
1975 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1976 	}
1977 	return -ENOIOCTLCMD;
1978 }
1979 
1980 long kvm_arch_vcpu_ioctl(struct file *filp,
1981                          unsigned int ioctl, unsigned long arg)
1982 {
1983 	struct kvm_vcpu *vcpu = filp->private_data;
1984 	void __user *argp = (void __user *)arg;
1985 	long r;
1986 
1987 	switch (ioctl) {
1988 	case KVM_ENABLE_CAP:
1989 	{
1990 		struct kvm_enable_cap cap;
1991 		r = -EFAULT;
1992 		vcpu_load(vcpu);
1993 		if (copy_from_user(&cap, argp, sizeof(cap)))
1994 			goto out;
1995 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1996 		vcpu_put(vcpu);
1997 		break;
1998 	}
1999 
2000 	case KVM_SET_ONE_REG:
2001 	case KVM_GET_ONE_REG:
2002 	{
2003 		struct kvm_one_reg reg;
2004 		r = -EFAULT;
2005 		if (copy_from_user(&reg, argp, sizeof(reg)))
2006 			goto out;
2007 		if (ioctl == KVM_SET_ONE_REG)
2008 			r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
2009 		else
2010 			r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
2011 		break;
2012 	}
2013 
2014 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
2015 	case KVM_DIRTY_TLB: {
2016 		struct kvm_dirty_tlb dirty;
2017 		r = -EFAULT;
2018 		vcpu_load(vcpu);
2019 		if (copy_from_user(&dirty, argp, sizeof(dirty)))
2020 			goto out;
2021 		r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
2022 		vcpu_put(vcpu);
2023 		break;
2024 	}
2025 #endif
2026 	default:
2027 		r = -EINVAL;
2028 	}
2029 
2030 out:
2031 	return r;
2032 }
2033 
2034 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2035 {
2036 	return VM_FAULT_SIGBUS;
2037 }
2038 
2039 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
2040 {
2041 	u32 inst_nop = 0x60000000;
2042 #ifdef CONFIG_KVM_BOOKE_HV
2043 	u32 inst_sc1 = 0x44000022;
2044 	pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
2045 	pvinfo->hcall[1] = cpu_to_be32(inst_nop);
2046 	pvinfo->hcall[2] = cpu_to_be32(inst_nop);
2047 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2048 #else
2049 	u32 inst_lis = 0x3c000000;
2050 	u32 inst_ori = 0x60000000;
2051 	u32 inst_sc = 0x44000002;
2052 	u32 inst_imm_mask = 0xffff;
2053 
2054 	/*
2055 	 * The hypercall to get into KVM from within guest context is as
2056 	 * follows:
2057 	 *
2058 	 *    lis r0, r0, KVM_SC_MAGIC_R0@h
2059 	 *    ori r0, KVM_SC_MAGIC_R0@l
2060 	 *    sc
2061 	 *    nop
2062 	 */
2063 	pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
2064 	pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
2065 	pvinfo->hcall[2] = cpu_to_be32(inst_sc);
2066 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2067 #endif
2068 
2069 	pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
2070 
2071 	return 0;
2072 }
2073 
2074 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
2075 			  bool line_status)
2076 {
2077 	if (!irqchip_in_kernel(kvm))
2078 		return -ENXIO;
2079 
2080 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2081 					irq_event->irq, irq_event->level,
2082 					line_status);
2083 	return 0;
2084 }
2085 
2086 
2087 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
2088 				   struct kvm_enable_cap *cap)
2089 {
2090 	int r;
2091 
2092 	if (cap->flags)
2093 		return -EINVAL;
2094 
2095 	switch (cap->cap) {
2096 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
2097 	case KVM_CAP_PPC_ENABLE_HCALL: {
2098 		unsigned long hcall = cap->args[0];
2099 
2100 		r = -EINVAL;
2101 		if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
2102 		    cap->args[1] > 1)
2103 			break;
2104 		if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
2105 			break;
2106 		if (cap->args[1])
2107 			set_bit(hcall / 4, kvm->arch.enabled_hcalls);
2108 		else
2109 			clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
2110 		r = 0;
2111 		break;
2112 	}
2113 	case KVM_CAP_PPC_SMT: {
2114 		unsigned long mode = cap->args[0];
2115 		unsigned long flags = cap->args[1];
2116 
2117 		r = -EINVAL;
2118 		if (kvm->arch.kvm_ops->set_smt_mode)
2119 			r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
2120 		break;
2121 	}
2122 
2123 	case KVM_CAP_PPC_NESTED_HV:
2124 		r = -EINVAL;
2125 		if (!is_kvmppc_hv_enabled(kvm) ||
2126 		    !kvm->arch.kvm_ops->enable_nested)
2127 			break;
2128 		r = kvm->arch.kvm_ops->enable_nested(kvm);
2129 		break;
2130 #endif
2131 	default:
2132 		r = -EINVAL;
2133 		break;
2134 	}
2135 
2136 	return r;
2137 }
2138 
2139 #ifdef CONFIG_PPC_BOOK3S_64
2140 /*
2141  * These functions check whether the underlying hardware is safe
2142  * against attacks based on observing the effects of speculatively
2143  * executed instructions, and whether it supplies instructions for
2144  * use in workarounds.  The information comes from firmware, either
2145  * via the device tree on powernv platforms or from an hcall on
2146  * pseries platforms.
2147  */
2148 #ifdef CONFIG_PPC_PSERIES
2149 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2150 {
2151 	struct h_cpu_char_result c;
2152 	unsigned long rc;
2153 
2154 	if (!machine_is(pseries))
2155 		return -ENOTTY;
2156 
2157 	rc = plpar_get_cpu_characteristics(&c);
2158 	if (rc == H_SUCCESS) {
2159 		cp->character = c.character;
2160 		cp->behaviour = c.behaviour;
2161 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2162 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2163 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2164 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2165 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2166 			KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
2167 			KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
2168 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2169 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2170 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2171 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2172 	}
2173 	return 0;
2174 }
2175 #else
2176 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2177 {
2178 	return -ENOTTY;
2179 }
2180 #endif
2181 
2182 static inline bool have_fw_feat(struct device_node *fw_features,
2183 				const char *state, const char *name)
2184 {
2185 	struct device_node *np;
2186 	bool r = false;
2187 
2188 	np = of_get_child_by_name(fw_features, name);
2189 	if (np) {
2190 		r = of_property_read_bool(np, state);
2191 		of_node_put(np);
2192 	}
2193 	return r;
2194 }
2195 
2196 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2197 {
2198 	struct device_node *np, *fw_features;
2199 	int r;
2200 
2201 	memset(cp, 0, sizeof(*cp));
2202 	r = pseries_get_cpu_char(cp);
2203 	if (r != -ENOTTY)
2204 		return r;
2205 
2206 	np = of_find_node_by_name(NULL, "ibm,opal");
2207 	if (np) {
2208 		fw_features = of_get_child_by_name(np, "fw-features");
2209 		of_node_put(np);
2210 		if (!fw_features)
2211 			return 0;
2212 		if (have_fw_feat(fw_features, "enabled",
2213 				 "inst-spec-barrier-ori31,31,0"))
2214 			cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
2215 		if (have_fw_feat(fw_features, "enabled",
2216 				 "fw-bcctrl-serialized"))
2217 			cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
2218 		if (have_fw_feat(fw_features, "enabled",
2219 				 "inst-l1d-flush-ori30,30,0"))
2220 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
2221 		if (have_fw_feat(fw_features, "enabled",
2222 				 "inst-l1d-flush-trig2"))
2223 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
2224 		if (have_fw_feat(fw_features, "enabled",
2225 				 "fw-l1d-thread-split"))
2226 			cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
2227 		if (have_fw_feat(fw_features, "enabled",
2228 				 "fw-count-cache-disabled"))
2229 			cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2230 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2231 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2232 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2233 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2234 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2235 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2236 
2237 		if (have_fw_feat(fw_features, "enabled",
2238 				 "speculation-policy-favor-security"))
2239 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
2240 		if (!have_fw_feat(fw_features, "disabled",
2241 				  "needs-l1d-flush-msr-pr-0-to-1"))
2242 			cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
2243 		if (!have_fw_feat(fw_features, "disabled",
2244 				  "needs-spec-barrier-for-bound-checks"))
2245 			cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2246 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2247 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2248 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2249 
2250 		of_node_put(fw_features);
2251 	}
2252 
2253 	return 0;
2254 }
2255 #endif
2256 
2257 long kvm_arch_vm_ioctl(struct file *filp,
2258                        unsigned int ioctl, unsigned long arg)
2259 {
2260 	struct kvm *kvm __maybe_unused = filp->private_data;
2261 	void __user *argp = (void __user *)arg;
2262 	long r;
2263 
2264 	switch (ioctl) {
2265 	case KVM_PPC_GET_PVINFO: {
2266 		struct kvm_ppc_pvinfo pvinfo;
2267 		memset(&pvinfo, 0, sizeof(pvinfo));
2268 		r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
2269 		if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
2270 			r = -EFAULT;
2271 			goto out;
2272 		}
2273 
2274 		break;
2275 	}
2276 	case KVM_ENABLE_CAP:
2277 	{
2278 		struct kvm_enable_cap cap;
2279 		r = -EFAULT;
2280 		if (copy_from_user(&cap, argp, sizeof(cap)))
2281 			goto out;
2282 		r = kvm_vm_ioctl_enable_cap(kvm, &cap);
2283 		break;
2284 	}
2285 #ifdef CONFIG_SPAPR_TCE_IOMMU
2286 	case KVM_CREATE_SPAPR_TCE_64: {
2287 		struct kvm_create_spapr_tce_64 create_tce_64;
2288 
2289 		r = -EFAULT;
2290 		if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
2291 			goto out;
2292 		if (create_tce_64.flags) {
2293 			r = -EINVAL;
2294 			goto out;
2295 		}
2296 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2297 		goto out;
2298 	}
2299 	case KVM_CREATE_SPAPR_TCE: {
2300 		struct kvm_create_spapr_tce create_tce;
2301 		struct kvm_create_spapr_tce_64 create_tce_64;
2302 
2303 		r = -EFAULT;
2304 		if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
2305 			goto out;
2306 
2307 		create_tce_64.liobn = create_tce.liobn;
2308 		create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
2309 		create_tce_64.offset = 0;
2310 		create_tce_64.size = create_tce.window_size >>
2311 				IOMMU_PAGE_SHIFT_4K;
2312 		create_tce_64.flags = 0;
2313 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2314 		goto out;
2315 	}
2316 #endif
2317 #ifdef CONFIG_PPC_BOOK3S_64
2318 	case KVM_PPC_GET_SMMU_INFO: {
2319 		struct kvm_ppc_smmu_info info;
2320 		struct kvm *kvm = filp->private_data;
2321 
2322 		memset(&info, 0, sizeof(info));
2323 		r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
2324 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2325 			r = -EFAULT;
2326 		break;
2327 	}
2328 	case KVM_PPC_RTAS_DEFINE_TOKEN: {
2329 		struct kvm *kvm = filp->private_data;
2330 
2331 		r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
2332 		break;
2333 	}
2334 	case KVM_PPC_CONFIGURE_V3_MMU: {
2335 		struct kvm *kvm = filp->private_data;
2336 		struct kvm_ppc_mmuv3_cfg cfg;
2337 
2338 		r = -EINVAL;
2339 		if (!kvm->arch.kvm_ops->configure_mmu)
2340 			goto out;
2341 		r = -EFAULT;
2342 		if (copy_from_user(&cfg, argp, sizeof(cfg)))
2343 			goto out;
2344 		r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
2345 		break;
2346 	}
2347 	case KVM_PPC_GET_RMMU_INFO: {
2348 		struct kvm *kvm = filp->private_data;
2349 		struct kvm_ppc_rmmu_info info;
2350 
2351 		r = -EINVAL;
2352 		if (!kvm->arch.kvm_ops->get_rmmu_info)
2353 			goto out;
2354 		r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
2355 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2356 			r = -EFAULT;
2357 		break;
2358 	}
2359 	case KVM_PPC_GET_CPU_CHAR: {
2360 		struct kvm_ppc_cpu_char cpuchar;
2361 
2362 		r = kvmppc_get_cpu_char(&cpuchar);
2363 		if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2364 			r = -EFAULT;
2365 		break;
2366 	}
2367 	default: {
2368 		struct kvm *kvm = filp->private_data;
2369 		r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
2370 	}
2371 #else /* CONFIG_PPC_BOOK3S_64 */
2372 	default:
2373 		r = -ENOTTY;
2374 #endif
2375 	}
2376 out:
2377 	return r;
2378 }
2379 
2380 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
2381 static unsigned long nr_lpids;
2382 
2383 long kvmppc_alloc_lpid(void)
2384 {
2385 	long lpid;
2386 
2387 	do {
2388 		lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
2389 		if (lpid >= nr_lpids) {
2390 			pr_err("%s: No LPIDs free\n", __func__);
2391 			return -ENOMEM;
2392 		}
2393 	} while (test_and_set_bit(lpid, lpid_inuse));
2394 
2395 	return lpid;
2396 }
2397 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2398 
2399 void kvmppc_claim_lpid(long lpid)
2400 {
2401 	set_bit(lpid, lpid_inuse);
2402 }
2403 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
2404 
2405 void kvmppc_free_lpid(long lpid)
2406 {
2407 	clear_bit(lpid, lpid_inuse);
2408 }
2409 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2410 
2411 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2412 {
2413 	nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
2414 	memset(lpid_inuse, 0, sizeof(lpid_inuse));
2415 }
2416 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2417 
2418 int kvm_arch_init(void *opaque)
2419 {
2420 	return 0;
2421 }
2422 
2423 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);
2424