xref: /openbmc/linux/arch/powerpc/kvm/powerpc.c (revision a9a08845)
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/tlbflush.h>
37 #include <asm/cputhreads.h>
38 #include <asm/irqflags.h>
39 #include <asm/iommu.h>
40 #include <asm/switch_to.h>
41 #include <asm/xive.h>
42 #ifdef CONFIG_PPC_PSERIES
43 #include <asm/hvcall.h>
44 #include <asm/plpar_wrappers.h>
45 #endif
46 
47 #include "timing.h"
48 #include "irq.h"
49 #include "../mm/mmu_decl.h"
50 
51 #define CREATE_TRACE_POINTS
52 #include "trace.h"
53 
54 struct kvmppc_ops *kvmppc_hv_ops;
55 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
56 struct kvmppc_ops *kvmppc_pr_ops;
57 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
58 
59 
60 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
61 {
62 	return !!(v->arch.pending_exceptions) || kvm_request_pending(v);
63 }
64 
65 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
66 {
67 	return false;
68 }
69 
70 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
71 {
72 	return 1;
73 }
74 
75 /*
76  * Common checks before entering the guest world.  Call with interrupts
77  * disabled.
78  *
79  * returns:
80  *
81  * == 1 if we're ready to go into guest state
82  * <= 0 if we need to go back to the host with return value
83  */
84 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
85 {
86 	int r;
87 
88 	WARN_ON(irqs_disabled());
89 	hard_irq_disable();
90 
91 	while (true) {
92 		if (need_resched()) {
93 			local_irq_enable();
94 			cond_resched();
95 			hard_irq_disable();
96 			continue;
97 		}
98 
99 		if (signal_pending(current)) {
100 			kvmppc_account_exit(vcpu, SIGNAL_EXITS);
101 			vcpu->run->exit_reason = KVM_EXIT_INTR;
102 			r = -EINTR;
103 			break;
104 		}
105 
106 		vcpu->mode = IN_GUEST_MODE;
107 
108 		/*
109 		 * Reading vcpu->requests must happen after setting vcpu->mode,
110 		 * so we don't miss a request because the requester sees
111 		 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
112 		 * before next entering the guest (and thus doesn't IPI).
113 		 * This also orders the write to mode from any reads
114 		 * to the page tables done while the VCPU is running.
115 		 * Please see the comment in kvm_flush_remote_tlbs.
116 		 */
117 		smp_mb();
118 
119 		if (kvm_request_pending(vcpu)) {
120 			/* Make sure we process requests preemptable */
121 			local_irq_enable();
122 			trace_kvm_check_requests(vcpu);
123 			r = kvmppc_core_check_requests(vcpu);
124 			hard_irq_disable();
125 			if (r > 0)
126 				continue;
127 			break;
128 		}
129 
130 		if (kvmppc_core_prepare_to_enter(vcpu)) {
131 			/* interrupts got enabled in between, so we
132 			   are back at square 1 */
133 			continue;
134 		}
135 
136 		guest_enter_irqoff();
137 		return 1;
138 	}
139 
140 	/* return to host */
141 	local_irq_enable();
142 	return r;
143 }
144 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
145 
146 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
147 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
148 {
149 	struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
150 	int i;
151 
152 	shared->sprg0 = swab64(shared->sprg0);
153 	shared->sprg1 = swab64(shared->sprg1);
154 	shared->sprg2 = swab64(shared->sprg2);
155 	shared->sprg3 = swab64(shared->sprg3);
156 	shared->srr0 = swab64(shared->srr0);
157 	shared->srr1 = swab64(shared->srr1);
158 	shared->dar = swab64(shared->dar);
159 	shared->msr = swab64(shared->msr);
160 	shared->dsisr = swab32(shared->dsisr);
161 	shared->int_pending = swab32(shared->int_pending);
162 	for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
163 		shared->sr[i] = swab32(shared->sr[i]);
164 }
165 #endif
166 
167 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
168 {
169 	int nr = kvmppc_get_gpr(vcpu, 11);
170 	int r;
171 	unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
172 	unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
173 	unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
174 	unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
175 	unsigned long r2 = 0;
176 
177 	if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
178 		/* 32 bit mode */
179 		param1 &= 0xffffffff;
180 		param2 &= 0xffffffff;
181 		param3 &= 0xffffffff;
182 		param4 &= 0xffffffff;
183 	}
184 
185 	switch (nr) {
186 	case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
187 	{
188 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
189 		/* Book3S can be little endian, find it out here */
190 		int shared_big_endian = true;
191 		if (vcpu->arch.intr_msr & MSR_LE)
192 			shared_big_endian = false;
193 		if (shared_big_endian != vcpu->arch.shared_big_endian)
194 			kvmppc_swab_shared(vcpu);
195 		vcpu->arch.shared_big_endian = shared_big_endian;
196 #endif
197 
198 		if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
199 			/*
200 			 * Older versions of the Linux magic page code had
201 			 * a bug where they would map their trampoline code
202 			 * NX. If that's the case, remove !PR NX capability.
203 			 */
204 			vcpu->arch.disable_kernel_nx = true;
205 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
206 		}
207 
208 		vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
209 		vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
210 
211 #ifdef CONFIG_PPC_64K_PAGES
212 		/*
213 		 * Make sure our 4k magic page is in the same window of a 64k
214 		 * page within the guest and within the host's page.
215 		 */
216 		if ((vcpu->arch.magic_page_pa & 0xf000) !=
217 		    ((ulong)vcpu->arch.shared & 0xf000)) {
218 			void *old_shared = vcpu->arch.shared;
219 			ulong shared = (ulong)vcpu->arch.shared;
220 			void *new_shared;
221 
222 			shared &= PAGE_MASK;
223 			shared |= vcpu->arch.magic_page_pa & 0xf000;
224 			new_shared = (void*)shared;
225 			memcpy(new_shared, old_shared, 0x1000);
226 			vcpu->arch.shared = new_shared;
227 		}
228 #endif
229 
230 		r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
231 
232 		r = EV_SUCCESS;
233 		break;
234 	}
235 	case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
236 		r = EV_SUCCESS;
237 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
238 		r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
239 #endif
240 
241 		/* Second return value is in r4 */
242 		break;
243 	case EV_HCALL_TOKEN(EV_IDLE):
244 		r = EV_SUCCESS;
245 		kvm_vcpu_block(vcpu);
246 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
247 		break;
248 	default:
249 		r = EV_UNIMPLEMENTED;
250 		break;
251 	}
252 
253 	kvmppc_set_gpr(vcpu, 4, r2);
254 
255 	return r;
256 }
257 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
258 
259 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
260 {
261 	int r = false;
262 
263 	/* We have to know what CPU to virtualize */
264 	if (!vcpu->arch.pvr)
265 		goto out;
266 
267 	/* PAPR only works with book3s_64 */
268 	if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
269 		goto out;
270 
271 	/* HV KVM can only do PAPR mode for now */
272 	if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
273 		goto out;
274 
275 #ifdef CONFIG_KVM_BOOKE_HV
276 	if (!cpu_has_feature(CPU_FTR_EMB_HV))
277 		goto out;
278 #endif
279 
280 	r = true;
281 
282 out:
283 	vcpu->arch.sane = r;
284 	return r ? 0 : -EINVAL;
285 }
286 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
287 
288 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
289 {
290 	enum emulation_result er;
291 	int r;
292 
293 	er = kvmppc_emulate_loadstore(vcpu);
294 	switch (er) {
295 	case EMULATE_DONE:
296 		/* Future optimization: only reload non-volatiles if they were
297 		 * actually modified. */
298 		r = RESUME_GUEST_NV;
299 		break;
300 	case EMULATE_AGAIN:
301 		r = RESUME_GUEST;
302 		break;
303 	case EMULATE_DO_MMIO:
304 		run->exit_reason = KVM_EXIT_MMIO;
305 		/* We must reload nonvolatiles because "update" load/store
306 		 * instructions modify register state. */
307 		/* Future optimization: only reload non-volatiles if they were
308 		 * actually modified. */
309 		r = RESUME_HOST_NV;
310 		break;
311 	case EMULATE_FAIL:
312 	{
313 		u32 last_inst;
314 
315 		kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
316 		/* XXX Deliver Program interrupt to guest. */
317 		pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
318 		r = RESUME_HOST;
319 		break;
320 	}
321 	default:
322 		WARN_ON(1);
323 		r = RESUME_GUEST;
324 	}
325 
326 	return r;
327 }
328 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
329 
330 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
331 	      bool data)
332 {
333 	ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
334 	struct kvmppc_pte pte;
335 	int r;
336 
337 	vcpu->stat.st++;
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;
372 
373 	vcpu->stat.ld++;
374 
375 	rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
376 			  XLATE_READ, &pte);
377 	if (rc)
378 		return rc;
379 
380 	*eaddr = pte.raddr;
381 
382 	if (!pte.may_read)
383 		return -EPERM;
384 
385 	if (!data && !pte.may_execute)
386 		return -ENOEXEC;
387 
388 	/* Magic page override */
389 	if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
390 	    ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
391 	    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
392 		void *magic = vcpu->arch.shared;
393 		magic += pte.eaddr & 0xfff;
394 		memcpy(ptr, magic, size);
395 		return EMULATE_DONE;
396 	}
397 
398 	if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
399 		return EMULATE_DO_MMIO;
400 
401 	return EMULATE_DONE;
402 }
403 EXPORT_SYMBOL_GPL(kvmppc_ld);
404 
405 int kvm_arch_hardware_enable(void)
406 {
407 	return 0;
408 }
409 
410 int kvm_arch_hardware_setup(void)
411 {
412 	return 0;
413 }
414 
415 void kvm_arch_check_processor_compat(void *rtn)
416 {
417 	*(int *)rtn = kvmppc_core_check_processor_compat();
418 }
419 
420 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
421 {
422 	struct kvmppc_ops *kvm_ops = NULL;
423 	/*
424 	 * if we have both HV and PR enabled, default is HV
425 	 */
426 	if (type == 0) {
427 		if (kvmppc_hv_ops)
428 			kvm_ops = kvmppc_hv_ops;
429 		else
430 			kvm_ops = kvmppc_pr_ops;
431 		if (!kvm_ops)
432 			goto err_out;
433 	} else	if (type == KVM_VM_PPC_HV) {
434 		if (!kvmppc_hv_ops)
435 			goto err_out;
436 		kvm_ops = kvmppc_hv_ops;
437 	} else if (type == KVM_VM_PPC_PR) {
438 		if (!kvmppc_pr_ops)
439 			goto err_out;
440 		kvm_ops = kvmppc_pr_ops;
441 	} else
442 		goto err_out;
443 
444 	if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
445 		return -ENOENT;
446 
447 	kvm->arch.kvm_ops = kvm_ops;
448 	return kvmppc_core_init_vm(kvm);
449 err_out:
450 	return -EINVAL;
451 }
452 
453 bool kvm_arch_has_vcpu_debugfs(void)
454 {
455 	return false;
456 }
457 
458 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
459 {
460 	return 0;
461 }
462 
463 void kvm_arch_destroy_vm(struct kvm *kvm)
464 {
465 	unsigned int i;
466 	struct kvm_vcpu *vcpu;
467 
468 #ifdef CONFIG_KVM_XICS
469 	/*
470 	 * We call kick_all_cpus_sync() to ensure that all
471 	 * CPUs have executed any pending IPIs before we
472 	 * continue and free VCPUs structures below.
473 	 */
474 	if (is_kvmppc_hv_enabled(kvm))
475 		kick_all_cpus_sync();
476 #endif
477 
478 	kvm_for_each_vcpu(i, vcpu, kvm)
479 		kvm_arch_vcpu_free(vcpu);
480 
481 	mutex_lock(&kvm->lock);
482 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
483 		kvm->vcpus[i] = NULL;
484 
485 	atomic_set(&kvm->online_vcpus, 0);
486 
487 	kvmppc_core_destroy_vm(kvm);
488 
489 	mutex_unlock(&kvm->lock);
490 
491 	/* drop the module reference */
492 	module_put(kvm->arch.kvm_ops->owner);
493 }
494 
495 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
496 {
497 	int r;
498 	/* Assume we're using HV mode when the HV module is loaded */
499 	int hv_enabled = kvmppc_hv_ops ? 1 : 0;
500 
501 	if (kvm) {
502 		/*
503 		 * Hooray - we know which VM type we're running on. Depend on
504 		 * that rather than the guess above.
505 		 */
506 		hv_enabled = is_kvmppc_hv_enabled(kvm);
507 	}
508 
509 	switch (ext) {
510 #ifdef CONFIG_BOOKE
511 	case KVM_CAP_PPC_BOOKE_SREGS:
512 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
513 	case KVM_CAP_PPC_EPR:
514 #else
515 	case KVM_CAP_PPC_SEGSTATE:
516 	case KVM_CAP_PPC_HIOR:
517 	case KVM_CAP_PPC_PAPR:
518 #endif
519 	case KVM_CAP_PPC_UNSET_IRQ:
520 	case KVM_CAP_PPC_IRQ_LEVEL:
521 	case KVM_CAP_ENABLE_CAP:
522 	case KVM_CAP_ENABLE_CAP_VM:
523 	case KVM_CAP_ONE_REG:
524 	case KVM_CAP_IOEVENTFD:
525 	case KVM_CAP_DEVICE_CTRL:
526 	case KVM_CAP_IMMEDIATE_EXIT:
527 		r = 1;
528 		break;
529 	case KVM_CAP_PPC_PAIRED_SINGLES:
530 	case KVM_CAP_PPC_OSI:
531 	case KVM_CAP_PPC_GET_PVINFO:
532 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
533 	case KVM_CAP_SW_TLB:
534 #endif
535 		/* We support this only for PR */
536 		r = !hv_enabled;
537 		break;
538 #ifdef CONFIG_KVM_MPIC
539 	case KVM_CAP_IRQ_MPIC:
540 		r = 1;
541 		break;
542 #endif
543 
544 #ifdef CONFIG_PPC_BOOK3S_64
545 	case KVM_CAP_SPAPR_TCE:
546 	case KVM_CAP_SPAPR_TCE_64:
547 		/* fallthrough */
548 	case KVM_CAP_SPAPR_TCE_VFIO:
549 	case KVM_CAP_PPC_RTAS:
550 	case KVM_CAP_PPC_FIXUP_HCALL:
551 	case KVM_CAP_PPC_ENABLE_HCALL:
552 #ifdef CONFIG_KVM_XICS
553 	case KVM_CAP_IRQ_XICS:
554 #endif
555 	case KVM_CAP_PPC_GET_CPU_CHAR:
556 		r = 1;
557 		break;
558 
559 	case KVM_CAP_PPC_ALLOC_HTAB:
560 		r = hv_enabled;
561 		break;
562 #endif /* CONFIG_PPC_BOOK3S_64 */
563 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
564 	case KVM_CAP_PPC_SMT:
565 		r = 0;
566 		if (kvm) {
567 			if (kvm->arch.emul_smt_mode > 1)
568 				r = kvm->arch.emul_smt_mode;
569 			else
570 				r = kvm->arch.smt_mode;
571 		} else if (hv_enabled) {
572 			if (cpu_has_feature(CPU_FTR_ARCH_300))
573 				r = 1;
574 			else
575 				r = threads_per_subcore;
576 		}
577 		break;
578 	case KVM_CAP_PPC_SMT_POSSIBLE:
579 		r = 1;
580 		if (hv_enabled) {
581 			if (!cpu_has_feature(CPU_FTR_ARCH_300))
582 				r = ((threads_per_subcore << 1) - 1);
583 			else
584 				/* P9 can emulate dbells, so allow any mode */
585 				r = 8 | 4 | 2 | 1;
586 		}
587 		break;
588 	case KVM_CAP_PPC_RMA:
589 		r = 0;
590 		break;
591 	case KVM_CAP_PPC_HWRNG:
592 		r = kvmppc_hwrng_present();
593 		break;
594 	case KVM_CAP_PPC_MMU_RADIX:
595 		r = !!(hv_enabled && radix_enabled());
596 		break;
597 	case KVM_CAP_PPC_MMU_HASH_V3:
598 		r = !!(hv_enabled && cpu_has_feature(CPU_FTR_ARCH_300));
599 		break;
600 #endif
601 	case KVM_CAP_SYNC_MMU:
602 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
603 		r = hv_enabled;
604 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
605 		r = 1;
606 #else
607 		r = 0;
608 #endif
609 		break;
610 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
611 	case KVM_CAP_PPC_HTAB_FD:
612 		r = hv_enabled;
613 		break;
614 #endif
615 	case KVM_CAP_NR_VCPUS:
616 		/*
617 		 * Recommending a number of CPUs is somewhat arbitrary; we
618 		 * return the number of present CPUs for -HV (since a host
619 		 * will have secondary threads "offline"), and for other KVM
620 		 * implementations just count online CPUs.
621 		 */
622 		if (hv_enabled)
623 			r = num_present_cpus();
624 		else
625 			r = num_online_cpus();
626 		break;
627 	case KVM_CAP_NR_MEMSLOTS:
628 		r = KVM_USER_MEM_SLOTS;
629 		break;
630 	case KVM_CAP_MAX_VCPUS:
631 		r = KVM_MAX_VCPUS;
632 		break;
633 #ifdef CONFIG_PPC_BOOK3S_64
634 	case KVM_CAP_PPC_GET_SMMU_INFO:
635 		r = 1;
636 		break;
637 	case KVM_CAP_SPAPR_MULTITCE:
638 		r = 1;
639 		break;
640 	case KVM_CAP_SPAPR_RESIZE_HPT:
641 		r = !!hv_enabled;
642 		break;
643 #endif
644 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
645 	case KVM_CAP_PPC_FWNMI:
646 		r = hv_enabled;
647 		break;
648 #endif
649 	case KVM_CAP_PPC_HTM:
650 		r = hv_enabled &&
651 		    (cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM_COMP);
652 		break;
653 	default:
654 		r = 0;
655 		break;
656 	}
657 	return r;
658 
659 }
660 
661 long kvm_arch_dev_ioctl(struct file *filp,
662                         unsigned int ioctl, unsigned long arg)
663 {
664 	return -EINVAL;
665 }
666 
667 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
668 			   struct kvm_memory_slot *dont)
669 {
670 	kvmppc_core_free_memslot(kvm, free, dont);
671 }
672 
673 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
674 			    unsigned long npages)
675 {
676 	return kvmppc_core_create_memslot(kvm, slot, npages);
677 }
678 
679 int kvm_arch_prepare_memory_region(struct kvm *kvm,
680 				   struct kvm_memory_slot *memslot,
681 				   const struct kvm_userspace_memory_region *mem,
682 				   enum kvm_mr_change change)
683 {
684 	return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
685 }
686 
687 void kvm_arch_commit_memory_region(struct kvm *kvm,
688 				   const struct kvm_userspace_memory_region *mem,
689 				   const struct kvm_memory_slot *old,
690 				   const struct kvm_memory_slot *new,
691 				   enum kvm_mr_change change)
692 {
693 	kvmppc_core_commit_memory_region(kvm, mem, old, new);
694 }
695 
696 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
697 				   struct kvm_memory_slot *slot)
698 {
699 	kvmppc_core_flush_memslot(kvm, slot);
700 }
701 
702 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
703 {
704 	struct kvm_vcpu *vcpu;
705 	vcpu = kvmppc_core_vcpu_create(kvm, id);
706 	if (!IS_ERR(vcpu)) {
707 		vcpu->arch.wqp = &vcpu->wq;
708 		kvmppc_create_vcpu_debugfs(vcpu, id);
709 	}
710 	return vcpu;
711 }
712 
713 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
714 {
715 }
716 
717 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
718 {
719 	/* Make sure we're not using the vcpu anymore */
720 	hrtimer_cancel(&vcpu->arch.dec_timer);
721 
722 	kvmppc_remove_vcpu_debugfs(vcpu);
723 
724 	switch (vcpu->arch.irq_type) {
725 	case KVMPPC_IRQ_MPIC:
726 		kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
727 		break;
728 	case KVMPPC_IRQ_XICS:
729 		if (xive_enabled())
730 			kvmppc_xive_cleanup_vcpu(vcpu);
731 		else
732 			kvmppc_xics_free_icp(vcpu);
733 		break;
734 	}
735 
736 	kvmppc_core_vcpu_free(vcpu);
737 }
738 
739 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
740 {
741 	kvm_arch_vcpu_free(vcpu);
742 }
743 
744 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
745 {
746 	return kvmppc_core_pending_dec(vcpu);
747 }
748 
749 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
750 {
751 	struct kvm_vcpu *vcpu;
752 
753 	vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
754 	kvmppc_decrementer_func(vcpu);
755 
756 	return HRTIMER_NORESTART;
757 }
758 
759 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
760 {
761 	int ret;
762 
763 	hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
764 	vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
765 	vcpu->arch.dec_expires = get_tb();
766 
767 #ifdef CONFIG_KVM_EXIT_TIMING
768 	mutex_init(&vcpu->arch.exit_timing_lock);
769 #endif
770 	ret = kvmppc_subarch_vcpu_init(vcpu);
771 	return ret;
772 }
773 
774 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
775 {
776 	kvmppc_mmu_destroy(vcpu);
777 	kvmppc_subarch_vcpu_uninit(vcpu);
778 }
779 
780 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
781 {
782 #ifdef CONFIG_BOOKE
783 	/*
784 	 * vrsave (formerly usprg0) isn't used by Linux, but may
785 	 * be used by the guest.
786 	 *
787 	 * On non-booke this is associated with Altivec and
788 	 * is handled by code in book3s.c.
789 	 */
790 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
791 #endif
792 	kvmppc_core_vcpu_load(vcpu, cpu);
793 }
794 
795 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
796 {
797 	kvmppc_core_vcpu_put(vcpu);
798 #ifdef CONFIG_BOOKE
799 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
800 #endif
801 }
802 
803 /*
804  * irq_bypass_add_producer and irq_bypass_del_producer are only
805  * useful if the architecture supports PCI passthrough.
806  * irq_bypass_stop and irq_bypass_start are not needed and so
807  * kvm_ops are not defined for them.
808  */
809 bool kvm_arch_has_irq_bypass(void)
810 {
811 	return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
812 		(kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
813 }
814 
815 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
816 				     struct irq_bypass_producer *prod)
817 {
818 	struct kvm_kernel_irqfd *irqfd =
819 		container_of(cons, struct kvm_kernel_irqfd, consumer);
820 	struct kvm *kvm = irqfd->kvm;
821 
822 	if (kvm->arch.kvm_ops->irq_bypass_add_producer)
823 		return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
824 
825 	return 0;
826 }
827 
828 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
829 				      struct irq_bypass_producer *prod)
830 {
831 	struct kvm_kernel_irqfd *irqfd =
832 		container_of(cons, struct kvm_kernel_irqfd, consumer);
833 	struct kvm *kvm = irqfd->kvm;
834 
835 	if (kvm->arch.kvm_ops->irq_bypass_del_producer)
836 		kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
837 }
838 
839 #ifdef CONFIG_VSX
840 static inline int kvmppc_get_vsr_dword_offset(int index)
841 {
842 	int offset;
843 
844 	if ((index != 0) && (index != 1))
845 		return -1;
846 
847 #ifdef __BIG_ENDIAN
848 	offset =  index;
849 #else
850 	offset = 1 - index;
851 #endif
852 
853 	return offset;
854 }
855 
856 static inline int kvmppc_get_vsr_word_offset(int index)
857 {
858 	int offset;
859 
860 	if ((index > 3) || (index < 0))
861 		return -1;
862 
863 #ifdef __BIG_ENDIAN
864 	offset = index;
865 #else
866 	offset = 3 - index;
867 #endif
868 	return offset;
869 }
870 
871 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
872 	u64 gpr)
873 {
874 	union kvmppc_one_reg val;
875 	int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
876 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
877 
878 	if (offset == -1)
879 		return;
880 
881 	if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
882 		val.vval = VCPU_VSX_VR(vcpu, index);
883 		val.vsxval[offset] = gpr;
884 		VCPU_VSX_VR(vcpu, index) = val.vval;
885 	} else {
886 		VCPU_VSX_FPR(vcpu, index, offset) = gpr;
887 	}
888 }
889 
890 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
891 	u64 gpr)
892 {
893 	union kvmppc_one_reg val;
894 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
895 
896 	if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
897 		val.vval = VCPU_VSX_VR(vcpu, index);
898 		val.vsxval[0] = gpr;
899 		val.vsxval[1] = gpr;
900 		VCPU_VSX_VR(vcpu, index) = val.vval;
901 	} else {
902 		VCPU_VSX_FPR(vcpu, index, 0) = gpr;
903 		VCPU_VSX_FPR(vcpu, index, 1) = gpr;
904 	}
905 }
906 
907 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
908 	u32 gpr32)
909 {
910 	union kvmppc_one_reg val;
911 	int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
912 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
913 	int dword_offset, word_offset;
914 
915 	if (offset == -1)
916 		return;
917 
918 	if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
919 		val.vval = VCPU_VSX_VR(vcpu, index);
920 		val.vsx32val[offset] = gpr32;
921 		VCPU_VSX_VR(vcpu, index) = val.vval;
922 	} else {
923 		dword_offset = offset / 2;
924 		word_offset = offset % 2;
925 		val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
926 		val.vsx32val[word_offset] = gpr32;
927 		VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
928 	}
929 }
930 #endif /* CONFIG_VSX */
931 
932 #ifdef CONFIG_ALTIVEC
933 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu,
934 		u64 gpr)
935 {
936 	int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
937 	u32 hi, lo;
938 	u32 di;
939 
940 #ifdef __BIG_ENDIAN
941 	hi = gpr >> 32;
942 	lo = gpr & 0xffffffff;
943 #else
944 	lo = gpr >> 32;
945 	hi = gpr & 0xffffffff;
946 #endif
947 
948 	di = 2 - vcpu->arch.mmio_vmx_copy_nums;		/* doubleword index */
949 	if (di > 1)
950 		return;
951 
952 	if (vcpu->arch.mmio_host_swabbed)
953 		di = 1 - di;
954 
955 	VCPU_VSX_VR(vcpu, index).u[di * 2] = hi;
956 	VCPU_VSX_VR(vcpu, index).u[di * 2 + 1] = lo;
957 }
958 #endif /* CONFIG_ALTIVEC */
959 
960 #ifdef CONFIG_PPC_FPU
961 static inline u64 sp_to_dp(u32 fprs)
962 {
963 	u64 fprd;
964 
965 	preempt_disable();
966 	enable_kernel_fp();
967 	asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs)
968 	     : "fr0");
969 	preempt_enable();
970 	return fprd;
971 }
972 
973 static inline u32 dp_to_sp(u64 fprd)
974 {
975 	u32 fprs;
976 
977 	preempt_disable();
978 	enable_kernel_fp();
979 	asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd)
980 	     : "fr0");
981 	preempt_enable();
982 	return fprs;
983 }
984 
985 #else
986 #define sp_to_dp(x)	(x)
987 #define dp_to_sp(x)	(x)
988 #endif /* CONFIG_PPC_FPU */
989 
990 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
991                                       struct kvm_run *run)
992 {
993 	u64 uninitialized_var(gpr);
994 
995 	if (run->mmio.len > sizeof(gpr)) {
996 		printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
997 		return;
998 	}
999 
1000 	if (!vcpu->arch.mmio_host_swabbed) {
1001 		switch (run->mmio.len) {
1002 		case 8: gpr = *(u64 *)run->mmio.data; break;
1003 		case 4: gpr = *(u32 *)run->mmio.data; break;
1004 		case 2: gpr = *(u16 *)run->mmio.data; break;
1005 		case 1: gpr = *(u8 *)run->mmio.data; break;
1006 		}
1007 	} else {
1008 		switch (run->mmio.len) {
1009 		case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
1010 		case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
1011 		case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
1012 		case 1: gpr = *(u8 *)run->mmio.data; break;
1013 		}
1014 	}
1015 
1016 	/* conversion between single and double precision */
1017 	if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
1018 		gpr = sp_to_dp(gpr);
1019 
1020 	if (vcpu->arch.mmio_sign_extend) {
1021 		switch (run->mmio.len) {
1022 #ifdef CONFIG_PPC64
1023 		case 4:
1024 			gpr = (s64)(s32)gpr;
1025 			break;
1026 #endif
1027 		case 2:
1028 			gpr = (s64)(s16)gpr;
1029 			break;
1030 		case 1:
1031 			gpr = (s64)(s8)gpr;
1032 			break;
1033 		}
1034 	}
1035 
1036 	switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1037 	case KVM_MMIO_REG_GPR:
1038 		kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1039 		break;
1040 	case KVM_MMIO_REG_FPR:
1041 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1042 		break;
1043 #ifdef CONFIG_PPC_BOOK3S
1044 	case KVM_MMIO_REG_QPR:
1045 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1046 		break;
1047 	case KVM_MMIO_REG_FQPR:
1048 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1049 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1050 		break;
1051 #endif
1052 #ifdef CONFIG_VSX
1053 	case KVM_MMIO_REG_VSX:
1054 		if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_DWORD)
1055 			kvmppc_set_vsr_dword(vcpu, gpr);
1056 		else if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_WORD)
1057 			kvmppc_set_vsr_word(vcpu, gpr);
1058 		else if (vcpu->arch.mmio_vsx_copy_type ==
1059 				KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1060 			kvmppc_set_vsr_dword_dump(vcpu, gpr);
1061 		break;
1062 #endif
1063 #ifdef CONFIG_ALTIVEC
1064 	case KVM_MMIO_REG_VMX:
1065 		kvmppc_set_vmx_dword(vcpu, gpr);
1066 		break;
1067 #endif
1068 	default:
1069 		BUG();
1070 	}
1071 }
1072 
1073 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1074 				unsigned int rt, unsigned int bytes,
1075 				int is_default_endian, int sign_extend)
1076 {
1077 	int idx, ret;
1078 	bool host_swabbed;
1079 
1080 	/* Pity C doesn't have a logical XOR operator */
1081 	if (kvmppc_need_byteswap(vcpu)) {
1082 		host_swabbed = is_default_endian;
1083 	} else {
1084 		host_swabbed = !is_default_endian;
1085 	}
1086 
1087 	if (bytes > sizeof(run->mmio.data)) {
1088 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1089 		       run->mmio.len);
1090 	}
1091 
1092 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1093 	run->mmio.len = bytes;
1094 	run->mmio.is_write = 0;
1095 
1096 	vcpu->arch.io_gpr = rt;
1097 	vcpu->arch.mmio_host_swabbed = host_swabbed;
1098 	vcpu->mmio_needed = 1;
1099 	vcpu->mmio_is_write = 0;
1100 	vcpu->arch.mmio_sign_extend = sign_extend;
1101 
1102 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1103 
1104 	ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1105 			      bytes, &run->mmio.data);
1106 
1107 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1108 
1109 	if (!ret) {
1110 		kvmppc_complete_mmio_load(vcpu, run);
1111 		vcpu->mmio_needed = 0;
1112 		return EMULATE_DONE;
1113 	}
1114 
1115 	return EMULATE_DO_MMIO;
1116 }
1117 
1118 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1119 		       unsigned int rt, unsigned int bytes,
1120 		       int is_default_endian)
1121 {
1122 	return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
1123 }
1124 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1125 
1126 /* Same as above, but sign extends */
1127 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
1128 			unsigned int rt, unsigned int bytes,
1129 			int is_default_endian)
1130 {
1131 	return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
1132 }
1133 
1134 #ifdef CONFIG_VSX
1135 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1136 			unsigned int rt, unsigned int bytes,
1137 			int is_default_endian, int mmio_sign_extend)
1138 {
1139 	enum emulation_result emulated = EMULATE_DONE;
1140 
1141 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1142 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1143 		return EMULATE_FAIL;
1144 
1145 	while (vcpu->arch.mmio_vsx_copy_nums) {
1146 		emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1147 			is_default_endian, mmio_sign_extend);
1148 
1149 		if (emulated != EMULATE_DONE)
1150 			break;
1151 
1152 		vcpu->arch.paddr_accessed += run->mmio.len;
1153 
1154 		vcpu->arch.mmio_vsx_copy_nums--;
1155 		vcpu->arch.mmio_vsx_offset++;
1156 	}
1157 	return emulated;
1158 }
1159 #endif /* CONFIG_VSX */
1160 
1161 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1162 			u64 val, unsigned int bytes, int is_default_endian)
1163 {
1164 	void *data = run->mmio.data;
1165 	int idx, ret;
1166 	bool host_swabbed;
1167 
1168 	/* Pity C doesn't have a logical XOR operator */
1169 	if (kvmppc_need_byteswap(vcpu)) {
1170 		host_swabbed = is_default_endian;
1171 	} else {
1172 		host_swabbed = !is_default_endian;
1173 	}
1174 
1175 	if (bytes > sizeof(run->mmio.data)) {
1176 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1177 		       run->mmio.len);
1178 	}
1179 
1180 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1181 	run->mmio.len = bytes;
1182 	run->mmio.is_write = 1;
1183 	vcpu->mmio_needed = 1;
1184 	vcpu->mmio_is_write = 1;
1185 
1186 	if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1187 		val = dp_to_sp(val);
1188 
1189 	/* Store the value at the lowest bytes in 'data'. */
1190 	if (!host_swabbed) {
1191 		switch (bytes) {
1192 		case 8: *(u64 *)data = val; break;
1193 		case 4: *(u32 *)data = val; break;
1194 		case 2: *(u16 *)data = val; break;
1195 		case 1: *(u8  *)data = val; break;
1196 		}
1197 	} else {
1198 		switch (bytes) {
1199 		case 8: *(u64 *)data = swab64(val); break;
1200 		case 4: *(u32 *)data = swab32(val); break;
1201 		case 2: *(u16 *)data = swab16(val); break;
1202 		case 1: *(u8  *)data = val; break;
1203 		}
1204 	}
1205 
1206 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1207 
1208 	ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1209 			       bytes, &run->mmio.data);
1210 
1211 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1212 
1213 	if (!ret) {
1214 		vcpu->mmio_needed = 0;
1215 		return EMULATE_DONE;
1216 	}
1217 
1218 	return EMULATE_DO_MMIO;
1219 }
1220 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1221 
1222 #ifdef CONFIG_VSX
1223 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1224 {
1225 	u32 dword_offset, word_offset;
1226 	union kvmppc_one_reg reg;
1227 	int vsx_offset = 0;
1228 	int copy_type = vcpu->arch.mmio_vsx_copy_type;
1229 	int result = 0;
1230 
1231 	switch (copy_type) {
1232 	case KVMPPC_VSX_COPY_DWORD:
1233 		vsx_offset =
1234 			kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1235 
1236 		if (vsx_offset == -1) {
1237 			result = -1;
1238 			break;
1239 		}
1240 
1241 		if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1242 			*val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1243 		} else {
1244 			reg.vval = VCPU_VSX_VR(vcpu, rs);
1245 			*val = reg.vsxval[vsx_offset];
1246 		}
1247 		break;
1248 
1249 	case KVMPPC_VSX_COPY_WORD:
1250 		vsx_offset =
1251 			kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1252 
1253 		if (vsx_offset == -1) {
1254 			result = -1;
1255 			break;
1256 		}
1257 
1258 		if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1259 			dword_offset = vsx_offset / 2;
1260 			word_offset = vsx_offset % 2;
1261 			reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1262 			*val = reg.vsx32val[word_offset];
1263 		} else {
1264 			reg.vval = VCPU_VSX_VR(vcpu, rs);
1265 			*val = reg.vsx32val[vsx_offset];
1266 		}
1267 		break;
1268 
1269 	default:
1270 		result = -1;
1271 		break;
1272 	}
1273 
1274 	return result;
1275 }
1276 
1277 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1278 			int rs, unsigned int bytes, int is_default_endian)
1279 {
1280 	u64 val;
1281 	enum emulation_result emulated = EMULATE_DONE;
1282 
1283 	vcpu->arch.io_gpr = rs;
1284 
1285 	/* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1286 	if (vcpu->arch.mmio_vsx_copy_nums > 4)
1287 		return EMULATE_FAIL;
1288 
1289 	while (vcpu->arch.mmio_vsx_copy_nums) {
1290 		if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1291 			return EMULATE_FAIL;
1292 
1293 		emulated = kvmppc_handle_store(run, vcpu,
1294 			 val, bytes, is_default_endian);
1295 
1296 		if (emulated != EMULATE_DONE)
1297 			break;
1298 
1299 		vcpu->arch.paddr_accessed += run->mmio.len;
1300 
1301 		vcpu->arch.mmio_vsx_copy_nums--;
1302 		vcpu->arch.mmio_vsx_offset++;
1303 	}
1304 
1305 	return emulated;
1306 }
1307 
1308 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu,
1309 			struct kvm_run *run)
1310 {
1311 	enum emulation_result emulated = EMULATE_FAIL;
1312 	int r;
1313 
1314 	vcpu->arch.paddr_accessed += run->mmio.len;
1315 
1316 	if (!vcpu->mmio_is_write) {
1317 		emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr,
1318 			 run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1319 	} else {
1320 		emulated = kvmppc_handle_vsx_store(run, vcpu,
1321 			 vcpu->arch.io_gpr, run->mmio.len, 1);
1322 	}
1323 
1324 	switch (emulated) {
1325 	case EMULATE_DO_MMIO:
1326 		run->exit_reason = KVM_EXIT_MMIO;
1327 		r = RESUME_HOST;
1328 		break;
1329 	case EMULATE_FAIL:
1330 		pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1331 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1332 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1333 		r = RESUME_HOST;
1334 		break;
1335 	default:
1336 		r = RESUME_GUEST;
1337 		break;
1338 	}
1339 	return r;
1340 }
1341 #endif /* CONFIG_VSX */
1342 
1343 #ifdef CONFIG_ALTIVEC
1344 /* handle quadword load access in two halves */
1345 int kvmppc_handle_load128_by2x64(struct kvm_run *run, struct kvm_vcpu *vcpu,
1346 		unsigned int rt, int is_default_endian)
1347 {
1348 	enum emulation_result emulated;
1349 
1350 	while (vcpu->arch.mmio_vmx_copy_nums) {
1351 		emulated = __kvmppc_handle_load(run, vcpu, rt, 8,
1352 				is_default_endian, 0);
1353 
1354 		if (emulated != EMULATE_DONE)
1355 			break;
1356 
1357 		vcpu->arch.paddr_accessed += run->mmio.len;
1358 		vcpu->arch.mmio_vmx_copy_nums--;
1359 	}
1360 
1361 	return emulated;
1362 }
1363 
1364 static inline int kvmppc_get_vmx_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1365 {
1366 	vector128 vrs = VCPU_VSX_VR(vcpu, rs);
1367 	u32 di;
1368 	u64 w0, w1;
1369 
1370 	di = 2 - vcpu->arch.mmio_vmx_copy_nums;		/* doubleword index */
1371 	if (di > 1)
1372 		return -1;
1373 
1374 	if (vcpu->arch.mmio_host_swabbed)
1375 		di = 1 - di;
1376 
1377 	w0 = vrs.u[di * 2];
1378 	w1 = vrs.u[di * 2 + 1];
1379 
1380 #ifdef __BIG_ENDIAN
1381 	*val = (w0 << 32) | w1;
1382 #else
1383 	*val = (w1 << 32) | w0;
1384 #endif
1385 	return 0;
1386 }
1387 
1388 /* handle quadword store in two halves */
1389 int kvmppc_handle_store128_by2x64(struct kvm_run *run, struct kvm_vcpu *vcpu,
1390 		unsigned int rs, int is_default_endian)
1391 {
1392 	u64 val = 0;
1393 	enum emulation_result emulated = EMULATE_DONE;
1394 
1395 	vcpu->arch.io_gpr = rs;
1396 
1397 	while (vcpu->arch.mmio_vmx_copy_nums) {
1398 		if (kvmppc_get_vmx_data(vcpu, rs, &val) == -1)
1399 			return EMULATE_FAIL;
1400 
1401 		emulated = kvmppc_handle_store(run, vcpu, val, 8,
1402 				is_default_endian);
1403 		if (emulated != EMULATE_DONE)
1404 			break;
1405 
1406 		vcpu->arch.paddr_accessed += run->mmio.len;
1407 		vcpu->arch.mmio_vmx_copy_nums--;
1408 	}
1409 
1410 	return emulated;
1411 }
1412 
1413 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu,
1414 		struct kvm_run *run)
1415 {
1416 	enum emulation_result emulated = EMULATE_FAIL;
1417 	int r;
1418 
1419 	vcpu->arch.paddr_accessed += run->mmio.len;
1420 
1421 	if (!vcpu->mmio_is_write) {
1422 		emulated = kvmppc_handle_load128_by2x64(run, vcpu,
1423 				vcpu->arch.io_gpr, 1);
1424 	} else {
1425 		emulated = kvmppc_handle_store128_by2x64(run, vcpu,
1426 				vcpu->arch.io_gpr, 1);
1427 	}
1428 
1429 	switch (emulated) {
1430 	case EMULATE_DO_MMIO:
1431 		run->exit_reason = KVM_EXIT_MMIO;
1432 		r = RESUME_HOST;
1433 		break;
1434 	case EMULATE_FAIL:
1435 		pr_info("KVM: MMIO emulation failed (VMX repeat)\n");
1436 		run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1437 		run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1438 		r = RESUME_HOST;
1439 		break;
1440 	default:
1441 		r = RESUME_GUEST;
1442 		break;
1443 	}
1444 	return r;
1445 }
1446 #endif /* CONFIG_ALTIVEC */
1447 
1448 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1449 {
1450 	int r = 0;
1451 	union kvmppc_one_reg val;
1452 	int size;
1453 
1454 	size = one_reg_size(reg->id);
1455 	if (size > sizeof(val))
1456 		return -EINVAL;
1457 
1458 	r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1459 	if (r == -EINVAL) {
1460 		r = 0;
1461 		switch (reg->id) {
1462 #ifdef CONFIG_ALTIVEC
1463 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1464 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1465 				r = -ENXIO;
1466 				break;
1467 			}
1468 			val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1469 			break;
1470 		case KVM_REG_PPC_VSCR:
1471 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1472 				r = -ENXIO;
1473 				break;
1474 			}
1475 			val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1476 			break;
1477 		case KVM_REG_PPC_VRSAVE:
1478 			val = get_reg_val(reg->id, vcpu->arch.vrsave);
1479 			break;
1480 #endif /* CONFIG_ALTIVEC */
1481 		default:
1482 			r = -EINVAL;
1483 			break;
1484 		}
1485 	}
1486 
1487 	if (r)
1488 		return r;
1489 
1490 	if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1491 		r = -EFAULT;
1492 
1493 	return r;
1494 }
1495 
1496 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1497 {
1498 	int r;
1499 	union kvmppc_one_reg val;
1500 	int size;
1501 
1502 	size = one_reg_size(reg->id);
1503 	if (size > sizeof(val))
1504 		return -EINVAL;
1505 
1506 	if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1507 		return -EFAULT;
1508 
1509 	r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1510 	if (r == -EINVAL) {
1511 		r = 0;
1512 		switch (reg->id) {
1513 #ifdef CONFIG_ALTIVEC
1514 		case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1515 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1516 				r = -ENXIO;
1517 				break;
1518 			}
1519 			vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1520 			break;
1521 		case KVM_REG_PPC_VSCR:
1522 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1523 				r = -ENXIO;
1524 				break;
1525 			}
1526 			vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1527 			break;
1528 		case KVM_REG_PPC_VRSAVE:
1529 			if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1530 				r = -ENXIO;
1531 				break;
1532 			}
1533 			vcpu->arch.vrsave = set_reg_val(reg->id, val);
1534 			break;
1535 #endif /* CONFIG_ALTIVEC */
1536 		default:
1537 			r = -EINVAL;
1538 			break;
1539 		}
1540 	}
1541 
1542 	return r;
1543 }
1544 
1545 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1546 {
1547 	int r;
1548 
1549 	vcpu_load(vcpu);
1550 
1551 	if (vcpu->mmio_needed) {
1552 		vcpu->mmio_needed = 0;
1553 		if (!vcpu->mmio_is_write)
1554 			kvmppc_complete_mmio_load(vcpu, run);
1555 #ifdef CONFIG_VSX
1556 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1557 			vcpu->arch.mmio_vsx_copy_nums--;
1558 			vcpu->arch.mmio_vsx_offset++;
1559 		}
1560 
1561 		if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1562 			r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run);
1563 			if (r == RESUME_HOST) {
1564 				vcpu->mmio_needed = 1;
1565 				goto out;
1566 			}
1567 		}
1568 #endif
1569 #ifdef CONFIG_ALTIVEC
1570 		if (vcpu->arch.mmio_vmx_copy_nums > 0)
1571 			vcpu->arch.mmio_vmx_copy_nums--;
1572 
1573 		if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1574 			r = kvmppc_emulate_mmio_vmx_loadstore(vcpu, run);
1575 			if (r == RESUME_HOST) {
1576 				vcpu->mmio_needed = 1;
1577 				goto out;
1578 			}
1579 		}
1580 #endif
1581 	} else if (vcpu->arch.osi_needed) {
1582 		u64 *gprs = run->osi.gprs;
1583 		int i;
1584 
1585 		for (i = 0; i < 32; i++)
1586 			kvmppc_set_gpr(vcpu, i, gprs[i]);
1587 		vcpu->arch.osi_needed = 0;
1588 	} else if (vcpu->arch.hcall_needed) {
1589 		int i;
1590 
1591 		kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1592 		for (i = 0; i < 9; ++i)
1593 			kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1594 		vcpu->arch.hcall_needed = 0;
1595 #ifdef CONFIG_BOOKE
1596 	} else if (vcpu->arch.epr_needed) {
1597 		kvmppc_set_epr(vcpu, run->epr.epr);
1598 		vcpu->arch.epr_needed = 0;
1599 #endif
1600 	}
1601 
1602 	kvm_sigset_activate(vcpu);
1603 
1604 	if (run->immediate_exit)
1605 		r = -EINTR;
1606 	else
1607 		r = kvmppc_vcpu_run(run, vcpu);
1608 
1609 	kvm_sigset_deactivate(vcpu);
1610 
1611 out:
1612 	vcpu_put(vcpu);
1613 	return r;
1614 }
1615 
1616 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1617 {
1618 	if (irq->irq == KVM_INTERRUPT_UNSET) {
1619 		kvmppc_core_dequeue_external(vcpu);
1620 		return 0;
1621 	}
1622 
1623 	kvmppc_core_queue_external(vcpu, irq);
1624 
1625 	kvm_vcpu_kick(vcpu);
1626 
1627 	return 0;
1628 }
1629 
1630 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1631 				     struct kvm_enable_cap *cap)
1632 {
1633 	int r;
1634 
1635 	if (cap->flags)
1636 		return -EINVAL;
1637 
1638 	switch (cap->cap) {
1639 	case KVM_CAP_PPC_OSI:
1640 		r = 0;
1641 		vcpu->arch.osi_enabled = true;
1642 		break;
1643 	case KVM_CAP_PPC_PAPR:
1644 		r = 0;
1645 		vcpu->arch.papr_enabled = true;
1646 		break;
1647 	case KVM_CAP_PPC_EPR:
1648 		r = 0;
1649 		if (cap->args[0])
1650 			vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1651 		else
1652 			vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1653 		break;
1654 #ifdef CONFIG_BOOKE
1655 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
1656 		r = 0;
1657 		vcpu->arch.watchdog_enabled = true;
1658 		break;
1659 #endif
1660 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1661 	case KVM_CAP_SW_TLB: {
1662 		struct kvm_config_tlb cfg;
1663 		void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1664 
1665 		r = -EFAULT;
1666 		if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1667 			break;
1668 
1669 		r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1670 		break;
1671 	}
1672 #endif
1673 #ifdef CONFIG_KVM_MPIC
1674 	case KVM_CAP_IRQ_MPIC: {
1675 		struct fd f;
1676 		struct kvm_device *dev;
1677 
1678 		r = -EBADF;
1679 		f = fdget(cap->args[0]);
1680 		if (!f.file)
1681 			break;
1682 
1683 		r = -EPERM;
1684 		dev = kvm_device_from_filp(f.file);
1685 		if (dev)
1686 			r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1687 
1688 		fdput(f);
1689 		break;
1690 	}
1691 #endif
1692 #ifdef CONFIG_KVM_XICS
1693 	case KVM_CAP_IRQ_XICS: {
1694 		struct fd f;
1695 		struct kvm_device *dev;
1696 
1697 		r = -EBADF;
1698 		f = fdget(cap->args[0]);
1699 		if (!f.file)
1700 			break;
1701 
1702 		r = -EPERM;
1703 		dev = kvm_device_from_filp(f.file);
1704 		if (dev) {
1705 			if (xive_enabled())
1706 				r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1707 			else
1708 				r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1709 		}
1710 
1711 		fdput(f);
1712 		break;
1713 	}
1714 #endif /* CONFIG_KVM_XICS */
1715 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1716 	case KVM_CAP_PPC_FWNMI:
1717 		r = -EINVAL;
1718 		if (!is_kvmppc_hv_enabled(vcpu->kvm))
1719 			break;
1720 		r = 0;
1721 		vcpu->kvm->arch.fwnmi_enabled = true;
1722 		break;
1723 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1724 	default:
1725 		r = -EINVAL;
1726 		break;
1727 	}
1728 
1729 	if (!r)
1730 		r = kvmppc_sanity_check(vcpu);
1731 
1732 	return r;
1733 }
1734 
1735 bool kvm_arch_intc_initialized(struct kvm *kvm)
1736 {
1737 #ifdef CONFIG_KVM_MPIC
1738 	if (kvm->arch.mpic)
1739 		return true;
1740 #endif
1741 #ifdef CONFIG_KVM_XICS
1742 	if (kvm->arch.xics || kvm->arch.xive)
1743 		return true;
1744 #endif
1745 	return false;
1746 }
1747 
1748 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1749                                     struct kvm_mp_state *mp_state)
1750 {
1751 	return -EINVAL;
1752 }
1753 
1754 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1755                                     struct kvm_mp_state *mp_state)
1756 {
1757 	return -EINVAL;
1758 }
1759 
1760 long kvm_arch_vcpu_async_ioctl(struct file *filp,
1761 			       unsigned int ioctl, unsigned long arg)
1762 {
1763 	struct kvm_vcpu *vcpu = filp->private_data;
1764 	void __user *argp = (void __user *)arg;
1765 
1766 	if (ioctl == KVM_INTERRUPT) {
1767 		struct kvm_interrupt irq;
1768 		if (copy_from_user(&irq, argp, sizeof(irq)))
1769 			return -EFAULT;
1770 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1771 	}
1772 	return -ENOIOCTLCMD;
1773 }
1774 
1775 long kvm_arch_vcpu_ioctl(struct file *filp,
1776                          unsigned int ioctl, unsigned long arg)
1777 {
1778 	struct kvm_vcpu *vcpu = filp->private_data;
1779 	void __user *argp = (void __user *)arg;
1780 	long r;
1781 
1782 	vcpu_load(vcpu);
1783 
1784 	switch (ioctl) {
1785 	case KVM_ENABLE_CAP:
1786 	{
1787 		struct kvm_enable_cap cap;
1788 		r = -EFAULT;
1789 		if (copy_from_user(&cap, argp, sizeof(cap)))
1790 			goto out;
1791 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1792 		break;
1793 	}
1794 
1795 	case KVM_SET_ONE_REG:
1796 	case KVM_GET_ONE_REG:
1797 	{
1798 		struct kvm_one_reg reg;
1799 		r = -EFAULT;
1800 		if (copy_from_user(&reg, argp, sizeof(reg)))
1801 			goto out;
1802 		if (ioctl == KVM_SET_ONE_REG)
1803 			r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
1804 		else
1805 			r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
1806 		break;
1807 	}
1808 
1809 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1810 	case KVM_DIRTY_TLB: {
1811 		struct kvm_dirty_tlb dirty;
1812 		r = -EFAULT;
1813 		if (copy_from_user(&dirty, argp, sizeof(dirty)))
1814 			goto out;
1815 		r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1816 		break;
1817 	}
1818 #endif
1819 	default:
1820 		r = -EINVAL;
1821 	}
1822 
1823 out:
1824 	vcpu_put(vcpu);
1825 	return r;
1826 }
1827 
1828 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1829 {
1830 	return VM_FAULT_SIGBUS;
1831 }
1832 
1833 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1834 {
1835 	u32 inst_nop = 0x60000000;
1836 #ifdef CONFIG_KVM_BOOKE_HV
1837 	u32 inst_sc1 = 0x44000022;
1838 	pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1839 	pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1840 	pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1841 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1842 #else
1843 	u32 inst_lis = 0x3c000000;
1844 	u32 inst_ori = 0x60000000;
1845 	u32 inst_sc = 0x44000002;
1846 	u32 inst_imm_mask = 0xffff;
1847 
1848 	/*
1849 	 * The hypercall to get into KVM from within guest context is as
1850 	 * follows:
1851 	 *
1852 	 *    lis r0, r0, KVM_SC_MAGIC_R0@h
1853 	 *    ori r0, KVM_SC_MAGIC_R0@l
1854 	 *    sc
1855 	 *    nop
1856 	 */
1857 	pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1858 	pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1859 	pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1860 	pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1861 #endif
1862 
1863 	pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1864 
1865 	return 0;
1866 }
1867 
1868 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1869 			  bool line_status)
1870 {
1871 	if (!irqchip_in_kernel(kvm))
1872 		return -ENXIO;
1873 
1874 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1875 					irq_event->irq, irq_event->level,
1876 					line_status);
1877 	return 0;
1878 }
1879 
1880 
1881 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1882 				   struct kvm_enable_cap *cap)
1883 {
1884 	int r;
1885 
1886 	if (cap->flags)
1887 		return -EINVAL;
1888 
1889 	switch (cap->cap) {
1890 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1891 	case KVM_CAP_PPC_ENABLE_HCALL: {
1892 		unsigned long hcall = cap->args[0];
1893 
1894 		r = -EINVAL;
1895 		if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1896 		    cap->args[1] > 1)
1897 			break;
1898 		if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1899 			break;
1900 		if (cap->args[1])
1901 			set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1902 		else
1903 			clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1904 		r = 0;
1905 		break;
1906 	}
1907 	case KVM_CAP_PPC_SMT: {
1908 		unsigned long mode = cap->args[0];
1909 		unsigned long flags = cap->args[1];
1910 
1911 		r = -EINVAL;
1912 		if (kvm->arch.kvm_ops->set_smt_mode)
1913 			r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
1914 		break;
1915 	}
1916 #endif
1917 	default:
1918 		r = -EINVAL;
1919 		break;
1920 	}
1921 
1922 	return r;
1923 }
1924 
1925 #ifdef CONFIG_PPC_BOOK3S_64
1926 /*
1927  * These functions check whether the underlying hardware is safe
1928  * against attacks based on observing the effects of speculatively
1929  * executed instructions, and whether it supplies instructions for
1930  * use in workarounds.  The information comes from firmware, either
1931  * via the device tree on powernv platforms or from an hcall on
1932  * pseries platforms.
1933  */
1934 #ifdef CONFIG_PPC_PSERIES
1935 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
1936 {
1937 	struct h_cpu_char_result c;
1938 	unsigned long rc;
1939 
1940 	if (!machine_is(pseries))
1941 		return -ENOTTY;
1942 
1943 	rc = plpar_get_cpu_characteristics(&c);
1944 	if (rc == H_SUCCESS) {
1945 		cp->character = c.character;
1946 		cp->behaviour = c.behaviour;
1947 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
1948 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
1949 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
1950 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
1951 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
1952 			KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
1953 			KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
1954 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
1955 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
1956 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
1957 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
1958 	}
1959 	return 0;
1960 }
1961 #else
1962 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
1963 {
1964 	return -ENOTTY;
1965 }
1966 #endif
1967 
1968 static inline bool have_fw_feat(struct device_node *fw_features,
1969 				const char *state, const char *name)
1970 {
1971 	struct device_node *np;
1972 	bool r = false;
1973 
1974 	np = of_get_child_by_name(fw_features, name);
1975 	if (np) {
1976 		r = of_property_read_bool(np, state);
1977 		of_node_put(np);
1978 	}
1979 	return r;
1980 }
1981 
1982 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
1983 {
1984 	struct device_node *np, *fw_features;
1985 	int r;
1986 
1987 	memset(cp, 0, sizeof(*cp));
1988 	r = pseries_get_cpu_char(cp);
1989 	if (r != -ENOTTY)
1990 		return r;
1991 
1992 	np = of_find_node_by_name(NULL, "ibm,opal");
1993 	if (np) {
1994 		fw_features = of_get_child_by_name(np, "fw-features");
1995 		of_node_put(np);
1996 		if (!fw_features)
1997 			return 0;
1998 		if (have_fw_feat(fw_features, "enabled",
1999 				 "inst-spec-barrier-ori31,31,0"))
2000 			cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
2001 		if (have_fw_feat(fw_features, "enabled",
2002 				 "fw-bcctrl-serialized"))
2003 			cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
2004 		if (have_fw_feat(fw_features, "enabled",
2005 				 "inst-l1d-flush-ori30,30,0"))
2006 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
2007 		if (have_fw_feat(fw_features, "enabled",
2008 				 "inst-l1d-flush-trig2"))
2009 			cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
2010 		if (have_fw_feat(fw_features, "enabled",
2011 				 "fw-l1d-thread-split"))
2012 			cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
2013 		if (have_fw_feat(fw_features, "enabled",
2014 				 "fw-count-cache-disabled"))
2015 			cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2016 		cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2017 			KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2018 			KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2019 			KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2020 			KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2021 			KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2022 
2023 		if (have_fw_feat(fw_features, "enabled",
2024 				 "speculation-policy-favor-security"))
2025 			cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
2026 		if (!have_fw_feat(fw_features, "disabled",
2027 				  "needs-l1d-flush-msr-pr-0-to-1"))
2028 			cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
2029 		if (!have_fw_feat(fw_features, "disabled",
2030 				  "needs-spec-barrier-for-bound-checks"))
2031 			cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2032 		cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2033 			KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2034 			KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2035 
2036 		of_node_put(fw_features);
2037 	}
2038 
2039 	return 0;
2040 }
2041 #endif
2042 
2043 long kvm_arch_vm_ioctl(struct file *filp,
2044                        unsigned int ioctl, unsigned long arg)
2045 {
2046 	struct kvm *kvm __maybe_unused = filp->private_data;
2047 	void __user *argp = (void __user *)arg;
2048 	long r;
2049 
2050 	switch (ioctl) {
2051 	case KVM_PPC_GET_PVINFO: {
2052 		struct kvm_ppc_pvinfo pvinfo;
2053 		memset(&pvinfo, 0, sizeof(pvinfo));
2054 		r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
2055 		if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
2056 			r = -EFAULT;
2057 			goto out;
2058 		}
2059 
2060 		break;
2061 	}
2062 	case KVM_ENABLE_CAP:
2063 	{
2064 		struct kvm_enable_cap cap;
2065 		r = -EFAULT;
2066 		if (copy_from_user(&cap, argp, sizeof(cap)))
2067 			goto out;
2068 		r = kvm_vm_ioctl_enable_cap(kvm, &cap);
2069 		break;
2070 	}
2071 #ifdef CONFIG_SPAPR_TCE_IOMMU
2072 	case KVM_CREATE_SPAPR_TCE_64: {
2073 		struct kvm_create_spapr_tce_64 create_tce_64;
2074 
2075 		r = -EFAULT;
2076 		if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
2077 			goto out;
2078 		if (create_tce_64.flags) {
2079 			r = -EINVAL;
2080 			goto out;
2081 		}
2082 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2083 		goto out;
2084 	}
2085 	case KVM_CREATE_SPAPR_TCE: {
2086 		struct kvm_create_spapr_tce create_tce;
2087 		struct kvm_create_spapr_tce_64 create_tce_64;
2088 
2089 		r = -EFAULT;
2090 		if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
2091 			goto out;
2092 
2093 		create_tce_64.liobn = create_tce.liobn;
2094 		create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
2095 		create_tce_64.offset = 0;
2096 		create_tce_64.size = create_tce.window_size >>
2097 				IOMMU_PAGE_SHIFT_4K;
2098 		create_tce_64.flags = 0;
2099 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2100 		goto out;
2101 	}
2102 #endif
2103 #ifdef CONFIG_PPC_BOOK3S_64
2104 	case KVM_PPC_GET_SMMU_INFO: {
2105 		struct kvm_ppc_smmu_info info;
2106 		struct kvm *kvm = filp->private_data;
2107 
2108 		memset(&info, 0, sizeof(info));
2109 		r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
2110 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2111 			r = -EFAULT;
2112 		break;
2113 	}
2114 	case KVM_PPC_RTAS_DEFINE_TOKEN: {
2115 		struct kvm *kvm = filp->private_data;
2116 
2117 		r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
2118 		break;
2119 	}
2120 	case KVM_PPC_CONFIGURE_V3_MMU: {
2121 		struct kvm *kvm = filp->private_data;
2122 		struct kvm_ppc_mmuv3_cfg cfg;
2123 
2124 		r = -EINVAL;
2125 		if (!kvm->arch.kvm_ops->configure_mmu)
2126 			goto out;
2127 		r = -EFAULT;
2128 		if (copy_from_user(&cfg, argp, sizeof(cfg)))
2129 			goto out;
2130 		r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
2131 		break;
2132 	}
2133 	case KVM_PPC_GET_RMMU_INFO: {
2134 		struct kvm *kvm = filp->private_data;
2135 		struct kvm_ppc_rmmu_info info;
2136 
2137 		r = -EINVAL;
2138 		if (!kvm->arch.kvm_ops->get_rmmu_info)
2139 			goto out;
2140 		r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
2141 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2142 			r = -EFAULT;
2143 		break;
2144 	}
2145 	case KVM_PPC_GET_CPU_CHAR: {
2146 		struct kvm_ppc_cpu_char cpuchar;
2147 
2148 		r = kvmppc_get_cpu_char(&cpuchar);
2149 		if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2150 			r = -EFAULT;
2151 		break;
2152 	}
2153 	default: {
2154 		struct kvm *kvm = filp->private_data;
2155 		r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
2156 	}
2157 #else /* CONFIG_PPC_BOOK3S_64 */
2158 	default:
2159 		r = -ENOTTY;
2160 #endif
2161 	}
2162 out:
2163 	return r;
2164 }
2165 
2166 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
2167 static unsigned long nr_lpids;
2168 
2169 long kvmppc_alloc_lpid(void)
2170 {
2171 	long lpid;
2172 
2173 	do {
2174 		lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
2175 		if (lpid >= nr_lpids) {
2176 			pr_err("%s: No LPIDs free\n", __func__);
2177 			return -ENOMEM;
2178 		}
2179 	} while (test_and_set_bit(lpid, lpid_inuse));
2180 
2181 	return lpid;
2182 }
2183 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2184 
2185 void kvmppc_claim_lpid(long lpid)
2186 {
2187 	set_bit(lpid, lpid_inuse);
2188 }
2189 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
2190 
2191 void kvmppc_free_lpid(long lpid)
2192 {
2193 	clear_bit(lpid, lpid_inuse);
2194 }
2195 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2196 
2197 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2198 {
2199 	nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
2200 	memset(lpid_inuse, 0, sizeof(lpid_inuse));
2201 }
2202 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2203 
2204 int kvm_arch_init(void *opaque)
2205 {
2206 	return 0;
2207 }
2208 
2209 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);
2210