xref: /openbmc/linux/arch/powerpc/kvm/powerpc.c (revision 77d84ff8)
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/fs.h>
27 #include <linux/slab.h>
28 #include <linux/file.h>
29 #include <linux/module.h>
30 #include <asm/cputable.h>
31 #include <asm/uaccess.h>
32 #include <asm/kvm_ppc.h>
33 #include <asm/tlbflush.h>
34 #include <asm/cputhreads.h>
35 #include <asm/irqflags.h>
36 #include "timing.h"
37 #include "irq.h"
38 #include "../mm/mmu_decl.h"
39 
40 #define CREATE_TRACE_POINTS
41 #include "trace.h"
42 
43 struct kvmppc_ops *kvmppc_hv_ops;
44 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
45 struct kvmppc_ops *kvmppc_pr_ops;
46 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
47 
48 
49 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
50 {
51 	return !!(v->arch.pending_exceptions) ||
52 	       v->requests;
53 }
54 
55 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
56 {
57 	return 1;
58 }
59 
60 /*
61  * Common checks before entering the guest world.  Call with interrupts
62  * disabled.
63  *
64  * returns:
65  *
66  * == 1 if we're ready to go into guest state
67  * <= 0 if we need to go back to the host with return value
68  */
69 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
70 {
71 	int r = 1;
72 
73 	WARN_ON_ONCE(!irqs_disabled());
74 	while (true) {
75 		if (need_resched()) {
76 			local_irq_enable();
77 			cond_resched();
78 			local_irq_disable();
79 			continue;
80 		}
81 
82 		if (signal_pending(current)) {
83 			kvmppc_account_exit(vcpu, SIGNAL_EXITS);
84 			vcpu->run->exit_reason = KVM_EXIT_INTR;
85 			r = -EINTR;
86 			break;
87 		}
88 
89 		vcpu->mode = IN_GUEST_MODE;
90 
91 		/*
92 		 * Reading vcpu->requests must happen after setting vcpu->mode,
93 		 * so we don't miss a request because the requester sees
94 		 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
95 		 * before next entering the guest (and thus doesn't IPI).
96 		 */
97 		smp_mb();
98 
99 		if (vcpu->requests) {
100 			/* Make sure we process requests preemptable */
101 			local_irq_enable();
102 			trace_kvm_check_requests(vcpu);
103 			r = kvmppc_core_check_requests(vcpu);
104 			local_irq_disable();
105 			if (r > 0)
106 				continue;
107 			break;
108 		}
109 
110 		if (kvmppc_core_prepare_to_enter(vcpu)) {
111 			/* interrupts got enabled in between, so we
112 			   are back at square 1 */
113 			continue;
114 		}
115 
116 #ifdef CONFIG_PPC64
117 		/* lazy EE magic */
118 		hard_irq_disable();
119 		if (lazy_irq_pending()) {
120 			/* Got an interrupt in between, try again */
121 			local_irq_enable();
122 			local_irq_disable();
123 			kvm_guest_exit();
124 			continue;
125 		}
126 #endif
127 
128 		kvm_guest_enter();
129 		break;
130 	}
131 
132 	return r;
133 }
134 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
135 
136 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
137 {
138 	int nr = kvmppc_get_gpr(vcpu, 11);
139 	int r;
140 	unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
141 	unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
142 	unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
143 	unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
144 	unsigned long r2 = 0;
145 
146 	if (!(vcpu->arch.shared->msr & MSR_SF)) {
147 		/* 32 bit mode */
148 		param1 &= 0xffffffff;
149 		param2 &= 0xffffffff;
150 		param3 &= 0xffffffff;
151 		param4 &= 0xffffffff;
152 	}
153 
154 	switch (nr) {
155 	case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
156 	{
157 		vcpu->arch.magic_page_pa = param1;
158 		vcpu->arch.magic_page_ea = param2;
159 
160 		r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
161 
162 		r = EV_SUCCESS;
163 		break;
164 	}
165 	case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
166 		r = EV_SUCCESS;
167 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
168 		/* XXX Missing magic page on 44x */
169 		r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
170 #endif
171 
172 		/* Second return value is in r4 */
173 		break;
174 	case EV_HCALL_TOKEN(EV_IDLE):
175 		r = EV_SUCCESS;
176 		kvm_vcpu_block(vcpu);
177 		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
178 		break;
179 	default:
180 		r = EV_UNIMPLEMENTED;
181 		break;
182 	}
183 
184 	kvmppc_set_gpr(vcpu, 4, r2);
185 
186 	return r;
187 }
188 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
189 
190 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
191 {
192 	int r = false;
193 
194 	/* We have to know what CPU to virtualize */
195 	if (!vcpu->arch.pvr)
196 		goto out;
197 
198 	/* PAPR only works with book3s_64 */
199 	if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
200 		goto out;
201 
202 	/* HV KVM can only do PAPR mode for now */
203 	if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
204 		goto out;
205 
206 #ifdef CONFIG_KVM_BOOKE_HV
207 	if (!cpu_has_feature(CPU_FTR_EMB_HV))
208 		goto out;
209 #endif
210 
211 	r = true;
212 
213 out:
214 	vcpu->arch.sane = r;
215 	return r ? 0 : -EINVAL;
216 }
217 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
218 
219 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
220 {
221 	enum emulation_result er;
222 	int r;
223 
224 	er = kvmppc_emulate_instruction(run, vcpu);
225 	switch (er) {
226 	case EMULATE_DONE:
227 		/* Future optimization: only reload non-volatiles if they were
228 		 * actually modified. */
229 		r = RESUME_GUEST_NV;
230 		break;
231 	case EMULATE_DO_MMIO:
232 		run->exit_reason = KVM_EXIT_MMIO;
233 		/* We must reload nonvolatiles because "update" load/store
234 		 * instructions modify register state. */
235 		/* Future optimization: only reload non-volatiles if they were
236 		 * actually modified. */
237 		r = RESUME_HOST_NV;
238 		break;
239 	case EMULATE_FAIL:
240 		/* XXX Deliver Program interrupt to guest. */
241 		printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__,
242 		       kvmppc_get_last_inst(vcpu));
243 		r = RESUME_HOST;
244 		break;
245 	default:
246 		WARN_ON(1);
247 		r = RESUME_GUEST;
248 	}
249 
250 	return r;
251 }
252 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
253 
254 int kvm_arch_hardware_enable(void *garbage)
255 {
256 	return 0;
257 }
258 
259 void kvm_arch_hardware_disable(void *garbage)
260 {
261 }
262 
263 int kvm_arch_hardware_setup(void)
264 {
265 	return 0;
266 }
267 
268 void kvm_arch_hardware_unsetup(void)
269 {
270 }
271 
272 void kvm_arch_check_processor_compat(void *rtn)
273 {
274 	*(int *)rtn = kvmppc_core_check_processor_compat();
275 }
276 
277 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
278 {
279 	struct kvmppc_ops *kvm_ops = NULL;
280 	/*
281 	 * if we have both HV and PR enabled, default is HV
282 	 */
283 	if (type == 0) {
284 		if (kvmppc_hv_ops)
285 			kvm_ops = kvmppc_hv_ops;
286 		else
287 			kvm_ops = kvmppc_pr_ops;
288 		if (!kvm_ops)
289 			goto err_out;
290 	} else	if (type == KVM_VM_PPC_HV) {
291 		if (!kvmppc_hv_ops)
292 			goto err_out;
293 		kvm_ops = kvmppc_hv_ops;
294 	} else if (type == KVM_VM_PPC_PR) {
295 		if (!kvmppc_pr_ops)
296 			goto err_out;
297 		kvm_ops = kvmppc_pr_ops;
298 	} else
299 		goto err_out;
300 
301 	if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
302 		return -ENOENT;
303 
304 	kvm->arch.kvm_ops = kvm_ops;
305 	return kvmppc_core_init_vm(kvm);
306 err_out:
307 	return -EINVAL;
308 }
309 
310 void kvm_arch_destroy_vm(struct kvm *kvm)
311 {
312 	unsigned int i;
313 	struct kvm_vcpu *vcpu;
314 
315 	kvm_for_each_vcpu(i, vcpu, kvm)
316 		kvm_arch_vcpu_free(vcpu);
317 
318 	mutex_lock(&kvm->lock);
319 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
320 		kvm->vcpus[i] = NULL;
321 
322 	atomic_set(&kvm->online_vcpus, 0);
323 
324 	kvmppc_core_destroy_vm(kvm);
325 
326 	mutex_unlock(&kvm->lock);
327 
328 	/* drop the module reference */
329 	module_put(kvm->arch.kvm_ops->owner);
330 }
331 
332 void kvm_arch_sync_events(struct kvm *kvm)
333 {
334 }
335 
336 int kvm_dev_ioctl_check_extension(long ext)
337 {
338 	int r;
339 	/* FIXME!!
340 	 * Should some of this be vm ioctl ? is it possible now ?
341 	 */
342 	int hv_enabled = kvmppc_hv_ops ? 1 : 0;
343 
344 	switch (ext) {
345 #ifdef CONFIG_BOOKE
346 	case KVM_CAP_PPC_BOOKE_SREGS:
347 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
348 	case KVM_CAP_PPC_EPR:
349 #else
350 	case KVM_CAP_PPC_SEGSTATE:
351 	case KVM_CAP_PPC_HIOR:
352 	case KVM_CAP_PPC_PAPR:
353 #endif
354 	case KVM_CAP_PPC_UNSET_IRQ:
355 	case KVM_CAP_PPC_IRQ_LEVEL:
356 	case KVM_CAP_ENABLE_CAP:
357 	case KVM_CAP_ONE_REG:
358 	case KVM_CAP_IOEVENTFD:
359 	case KVM_CAP_DEVICE_CTRL:
360 		r = 1;
361 		break;
362 	case KVM_CAP_PPC_PAIRED_SINGLES:
363 	case KVM_CAP_PPC_OSI:
364 	case KVM_CAP_PPC_GET_PVINFO:
365 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
366 	case KVM_CAP_SW_TLB:
367 #endif
368 		/* We support this only for PR */
369 		r = !hv_enabled;
370 		break;
371 #ifdef CONFIG_KVM_MMIO
372 	case KVM_CAP_COALESCED_MMIO:
373 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
374 		break;
375 #endif
376 #ifdef CONFIG_KVM_MPIC
377 	case KVM_CAP_IRQ_MPIC:
378 		r = 1;
379 		break;
380 #endif
381 
382 #ifdef CONFIG_PPC_BOOK3S_64
383 	case KVM_CAP_SPAPR_TCE:
384 	case KVM_CAP_PPC_ALLOC_HTAB:
385 	case KVM_CAP_PPC_RTAS:
386 #ifdef CONFIG_KVM_XICS
387 	case KVM_CAP_IRQ_XICS:
388 #endif
389 		r = 1;
390 		break;
391 #endif /* CONFIG_PPC_BOOK3S_64 */
392 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
393 	case KVM_CAP_PPC_SMT:
394 		if (hv_enabled)
395 			r = threads_per_core;
396 		else
397 			r = 0;
398 		break;
399 	case KVM_CAP_PPC_RMA:
400 		r = hv_enabled;
401 		/* PPC970 requires an RMA */
402 		if (r && cpu_has_feature(CPU_FTR_ARCH_201))
403 			r = 2;
404 		break;
405 #endif
406 	case KVM_CAP_SYNC_MMU:
407 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
408 		if (hv_enabled)
409 			r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
410 		else
411 			r = 0;
412 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
413 		r = 1;
414 #else
415 		r = 0;
416 #endif
417 		break;
418 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
419 	case KVM_CAP_PPC_HTAB_FD:
420 		r = hv_enabled;
421 		break;
422 #endif
423 	case KVM_CAP_NR_VCPUS:
424 		/*
425 		 * Recommending a number of CPUs is somewhat arbitrary; we
426 		 * return the number of present CPUs for -HV (since a host
427 		 * will have secondary threads "offline"), and for other KVM
428 		 * implementations just count online CPUs.
429 		 */
430 		if (hv_enabled)
431 			r = num_present_cpus();
432 		else
433 			r = num_online_cpus();
434 		break;
435 	case KVM_CAP_MAX_VCPUS:
436 		r = KVM_MAX_VCPUS;
437 		break;
438 #ifdef CONFIG_PPC_BOOK3S_64
439 	case KVM_CAP_PPC_GET_SMMU_INFO:
440 		r = 1;
441 		break;
442 #endif
443 	default:
444 		r = 0;
445 		break;
446 	}
447 	return r;
448 
449 }
450 
451 long kvm_arch_dev_ioctl(struct file *filp,
452                         unsigned int ioctl, unsigned long arg)
453 {
454 	return -EINVAL;
455 }
456 
457 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
458 			   struct kvm_memory_slot *dont)
459 {
460 	kvmppc_core_free_memslot(kvm, free, dont);
461 }
462 
463 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
464 			    unsigned long npages)
465 {
466 	return kvmppc_core_create_memslot(kvm, slot, npages);
467 }
468 
469 void kvm_arch_memslots_updated(struct kvm *kvm)
470 {
471 }
472 
473 int kvm_arch_prepare_memory_region(struct kvm *kvm,
474 				   struct kvm_memory_slot *memslot,
475 				   struct kvm_userspace_memory_region *mem,
476 				   enum kvm_mr_change change)
477 {
478 	return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
479 }
480 
481 void kvm_arch_commit_memory_region(struct kvm *kvm,
482 				   struct kvm_userspace_memory_region *mem,
483 				   const struct kvm_memory_slot *old,
484 				   enum kvm_mr_change change)
485 {
486 	kvmppc_core_commit_memory_region(kvm, mem, old);
487 }
488 
489 void kvm_arch_flush_shadow_all(struct kvm *kvm)
490 {
491 }
492 
493 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
494 				   struct kvm_memory_slot *slot)
495 {
496 	kvmppc_core_flush_memslot(kvm, slot);
497 }
498 
499 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
500 {
501 	struct kvm_vcpu *vcpu;
502 	vcpu = kvmppc_core_vcpu_create(kvm, id);
503 	if (!IS_ERR(vcpu)) {
504 		vcpu->arch.wqp = &vcpu->wq;
505 		kvmppc_create_vcpu_debugfs(vcpu, id);
506 	}
507 	return vcpu;
508 }
509 
510 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
511 {
512 	return 0;
513 }
514 
515 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
516 {
517 	/* Make sure we're not using the vcpu anymore */
518 	hrtimer_cancel(&vcpu->arch.dec_timer);
519 	tasklet_kill(&vcpu->arch.tasklet);
520 
521 	kvmppc_remove_vcpu_debugfs(vcpu);
522 
523 	switch (vcpu->arch.irq_type) {
524 	case KVMPPC_IRQ_MPIC:
525 		kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
526 		break;
527 	case KVMPPC_IRQ_XICS:
528 		kvmppc_xics_free_icp(vcpu);
529 		break;
530 	}
531 
532 	kvmppc_core_vcpu_free(vcpu);
533 }
534 
535 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
536 {
537 	kvm_arch_vcpu_free(vcpu);
538 }
539 
540 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
541 {
542 	return kvmppc_core_pending_dec(vcpu);
543 }
544 
545 /*
546  * low level hrtimer wake routine. Because this runs in hardirq context
547  * we schedule a tasklet to do the real work.
548  */
549 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
550 {
551 	struct kvm_vcpu *vcpu;
552 
553 	vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
554 	tasklet_schedule(&vcpu->arch.tasklet);
555 
556 	return HRTIMER_NORESTART;
557 }
558 
559 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
560 {
561 	int ret;
562 
563 	hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
564 	tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
565 	vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
566 	vcpu->arch.dec_expires = ~(u64)0;
567 
568 #ifdef CONFIG_KVM_EXIT_TIMING
569 	mutex_init(&vcpu->arch.exit_timing_lock);
570 #endif
571 	ret = kvmppc_subarch_vcpu_init(vcpu);
572 	return ret;
573 }
574 
575 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
576 {
577 	kvmppc_mmu_destroy(vcpu);
578 	kvmppc_subarch_vcpu_uninit(vcpu);
579 }
580 
581 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
582 {
583 #ifdef CONFIG_BOOKE
584 	/*
585 	 * vrsave (formerly usprg0) isn't used by Linux, but may
586 	 * be used by the guest.
587 	 *
588 	 * On non-booke this is associated with Altivec and
589 	 * is handled by code in book3s.c.
590 	 */
591 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
592 #endif
593 	kvmppc_core_vcpu_load(vcpu, cpu);
594 }
595 
596 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
597 {
598 	kvmppc_core_vcpu_put(vcpu);
599 #ifdef CONFIG_BOOKE
600 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
601 #endif
602 }
603 
604 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
605                                      struct kvm_run *run)
606 {
607 	kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
608 }
609 
610 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
611                                       struct kvm_run *run)
612 {
613 	u64 uninitialized_var(gpr);
614 
615 	if (run->mmio.len > sizeof(gpr)) {
616 		printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
617 		return;
618 	}
619 
620 	if (vcpu->arch.mmio_is_bigendian) {
621 		switch (run->mmio.len) {
622 		case 8: gpr = *(u64 *)run->mmio.data; break;
623 		case 4: gpr = *(u32 *)run->mmio.data; break;
624 		case 2: gpr = *(u16 *)run->mmio.data; break;
625 		case 1: gpr = *(u8 *)run->mmio.data; break;
626 		}
627 	} else {
628 		/* Convert BE data from userland back to LE. */
629 		switch (run->mmio.len) {
630 		case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
631 		case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
632 		case 1: gpr = *(u8 *)run->mmio.data; break;
633 		}
634 	}
635 
636 	if (vcpu->arch.mmio_sign_extend) {
637 		switch (run->mmio.len) {
638 #ifdef CONFIG_PPC64
639 		case 4:
640 			gpr = (s64)(s32)gpr;
641 			break;
642 #endif
643 		case 2:
644 			gpr = (s64)(s16)gpr;
645 			break;
646 		case 1:
647 			gpr = (s64)(s8)gpr;
648 			break;
649 		}
650 	}
651 
652 	kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
653 
654 	switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
655 	case KVM_MMIO_REG_GPR:
656 		kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
657 		break;
658 	case KVM_MMIO_REG_FPR:
659 		vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
660 		break;
661 #ifdef CONFIG_PPC_BOOK3S
662 	case KVM_MMIO_REG_QPR:
663 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
664 		break;
665 	case KVM_MMIO_REG_FQPR:
666 		vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
667 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
668 		break;
669 #endif
670 	default:
671 		BUG();
672 	}
673 }
674 
675 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
676                        unsigned int rt, unsigned int bytes, int is_bigendian)
677 {
678 	int idx, ret;
679 
680 	if (bytes > sizeof(run->mmio.data)) {
681 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
682 		       run->mmio.len);
683 	}
684 
685 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
686 	run->mmio.len = bytes;
687 	run->mmio.is_write = 0;
688 
689 	vcpu->arch.io_gpr = rt;
690 	vcpu->arch.mmio_is_bigendian = is_bigendian;
691 	vcpu->mmio_needed = 1;
692 	vcpu->mmio_is_write = 0;
693 	vcpu->arch.mmio_sign_extend = 0;
694 
695 	idx = srcu_read_lock(&vcpu->kvm->srcu);
696 
697 	ret = kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
698 			      bytes, &run->mmio.data);
699 
700 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
701 
702 	if (!ret) {
703 		kvmppc_complete_mmio_load(vcpu, run);
704 		vcpu->mmio_needed = 0;
705 		return EMULATE_DONE;
706 	}
707 
708 	return EMULATE_DO_MMIO;
709 }
710 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
711 
712 /* Same as above, but sign extends */
713 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
714                         unsigned int rt, unsigned int bytes, int is_bigendian)
715 {
716 	int r;
717 
718 	vcpu->arch.mmio_sign_extend = 1;
719 	r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian);
720 
721 	return r;
722 }
723 
724 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
725                         u64 val, unsigned int bytes, int is_bigendian)
726 {
727 	void *data = run->mmio.data;
728 	int idx, ret;
729 
730 	if (bytes > sizeof(run->mmio.data)) {
731 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
732 		       run->mmio.len);
733 	}
734 
735 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
736 	run->mmio.len = bytes;
737 	run->mmio.is_write = 1;
738 	vcpu->mmio_needed = 1;
739 	vcpu->mmio_is_write = 1;
740 
741 	/* Store the value at the lowest bytes in 'data'. */
742 	if (is_bigendian) {
743 		switch (bytes) {
744 		case 8: *(u64 *)data = val; break;
745 		case 4: *(u32 *)data = val; break;
746 		case 2: *(u16 *)data = val; break;
747 		case 1: *(u8  *)data = val; break;
748 		}
749 	} else {
750 		/* Store LE value into 'data'. */
751 		switch (bytes) {
752 		case 4: st_le32(data, val); break;
753 		case 2: st_le16(data, val); break;
754 		case 1: *(u8 *)data = val; break;
755 		}
756 	}
757 
758 	idx = srcu_read_lock(&vcpu->kvm->srcu);
759 
760 	ret = kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
761 			       bytes, &run->mmio.data);
762 
763 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
764 
765 	if (!ret) {
766 		vcpu->mmio_needed = 0;
767 		return EMULATE_DONE;
768 	}
769 
770 	return EMULATE_DO_MMIO;
771 }
772 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
773 
774 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
775 {
776 	int r;
777 	sigset_t sigsaved;
778 
779 	if (vcpu->sigset_active)
780 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
781 
782 	if (vcpu->mmio_needed) {
783 		if (!vcpu->mmio_is_write)
784 			kvmppc_complete_mmio_load(vcpu, run);
785 		vcpu->mmio_needed = 0;
786 	} else if (vcpu->arch.dcr_needed) {
787 		if (!vcpu->arch.dcr_is_write)
788 			kvmppc_complete_dcr_load(vcpu, run);
789 		vcpu->arch.dcr_needed = 0;
790 	} else if (vcpu->arch.osi_needed) {
791 		u64 *gprs = run->osi.gprs;
792 		int i;
793 
794 		for (i = 0; i < 32; i++)
795 			kvmppc_set_gpr(vcpu, i, gprs[i]);
796 		vcpu->arch.osi_needed = 0;
797 	} else if (vcpu->arch.hcall_needed) {
798 		int i;
799 
800 		kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
801 		for (i = 0; i < 9; ++i)
802 			kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
803 		vcpu->arch.hcall_needed = 0;
804 #ifdef CONFIG_BOOKE
805 	} else if (vcpu->arch.epr_needed) {
806 		kvmppc_set_epr(vcpu, run->epr.epr);
807 		vcpu->arch.epr_needed = 0;
808 #endif
809 	}
810 
811 	r = kvmppc_vcpu_run(run, vcpu);
812 
813 	if (vcpu->sigset_active)
814 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
815 
816 	return r;
817 }
818 
819 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
820 {
821 	if (irq->irq == KVM_INTERRUPT_UNSET) {
822 		kvmppc_core_dequeue_external(vcpu);
823 		return 0;
824 	}
825 
826 	kvmppc_core_queue_external(vcpu, irq);
827 
828 	kvm_vcpu_kick(vcpu);
829 
830 	return 0;
831 }
832 
833 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
834 				     struct kvm_enable_cap *cap)
835 {
836 	int r;
837 
838 	if (cap->flags)
839 		return -EINVAL;
840 
841 	switch (cap->cap) {
842 	case KVM_CAP_PPC_OSI:
843 		r = 0;
844 		vcpu->arch.osi_enabled = true;
845 		break;
846 	case KVM_CAP_PPC_PAPR:
847 		r = 0;
848 		vcpu->arch.papr_enabled = true;
849 		break;
850 	case KVM_CAP_PPC_EPR:
851 		r = 0;
852 		if (cap->args[0])
853 			vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
854 		else
855 			vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
856 		break;
857 #ifdef CONFIG_BOOKE
858 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
859 		r = 0;
860 		vcpu->arch.watchdog_enabled = true;
861 		break;
862 #endif
863 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
864 	case KVM_CAP_SW_TLB: {
865 		struct kvm_config_tlb cfg;
866 		void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
867 
868 		r = -EFAULT;
869 		if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
870 			break;
871 
872 		r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
873 		break;
874 	}
875 #endif
876 #ifdef CONFIG_KVM_MPIC
877 	case KVM_CAP_IRQ_MPIC: {
878 		struct fd f;
879 		struct kvm_device *dev;
880 
881 		r = -EBADF;
882 		f = fdget(cap->args[0]);
883 		if (!f.file)
884 			break;
885 
886 		r = -EPERM;
887 		dev = kvm_device_from_filp(f.file);
888 		if (dev)
889 			r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
890 
891 		fdput(f);
892 		break;
893 	}
894 #endif
895 #ifdef CONFIG_KVM_XICS
896 	case KVM_CAP_IRQ_XICS: {
897 		struct fd f;
898 		struct kvm_device *dev;
899 
900 		r = -EBADF;
901 		f = fdget(cap->args[0]);
902 		if (!f.file)
903 			break;
904 
905 		r = -EPERM;
906 		dev = kvm_device_from_filp(f.file);
907 		if (dev)
908 			r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
909 
910 		fdput(f);
911 		break;
912 	}
913 #endif /* CONFIG_KVM_XICS */
914 	default:
915 		r = -EINVAL;
916 		break;
917 	}
918 
919 	if (!r)
920 		r = kvmppc_sanity_check(vcpu);
921 
922 	return r;
923 }
924 
925 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
926                                     struct kvm_mp_state *mp_state)
927 {
928 	return -EINVAL;
929 }
930 
931 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
932                                     struct kvm_mp_state *mp_state)
933 {
934 	return -EINVAL;
935 }
936 
937 long kvm_arch_vcpu_ioctl(struct file *filp,
938                          unsigned int ioctl, unsigned long arg)
939 {
940 	struct kvm_vcpu *vcpu = filp->private_data;
941 	void __user *argp = (void __user *)arg;
942 	long r;
943 
944 	switch (ioctl) {
945 	case KVM_INTERRUPT: {
946 		struct kvm_interrupt irq;
947 		r = -EFAULT;
948 		if (copy_from_user(&irq, argp, sizeof(irq)))
949 			goto out;
950 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
951 		goto out;
952 	}
953 
954 	case KVM_ENABLE_CAP:
955 	{
956 		struct kvm_enable_cap cap;
957 		r = -EFAULT;
958 		if (copy_from_user(&cap, argp, sizeof(cap)))
959 			goto out;
960 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
961 		break;
962 	}
963 
964 	case KVM_SET_ONE_REG:
965 	case KVM_GET_ONE_REG:
966 	{
967 		struct kvm_one_reg reg;
968 		r = -EFAULT;
969 		if (copy_from_user(&reg, argp, sizeof(reg)))
970 			goto out;
971 		if (ioctl == KVM_SET_ONE_REG)
972 			r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
973 		else
974 			r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
975 		break;
976 	}
977 
978 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
979 	case KVM_DIRTY_TLB: {
980 		struct kvm_dirty_tlb dirty;
981 		r = -EFAULT;
982 		if (copy_from_user(&dirty, argp, sizeof(dirty)))
983 			goto out;
984 		r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
985 		break;
986 	}
987 #endif
988 	default:
989 		r = -EINVAL;
990 	}
991 
992 out:
993 	return r;
994 }
995 
996 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
997 {
998 	return VM_FAULT_SIGBUS;
999 }
1000 
1001 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1002 {
1003 	u32 inst_nop = 0x60000000;
1004 #ifdef CONFIG_KVM_BOOKE_HV
1005 	u32 inst_sc1 = 0x44000022;
1006 	pvinfo->hcall[0] = inst_sc1;
1007 	pvinfo->hcall[1] = inst_nop;
1008 	pvinfo->hcall[2] = inst_nop;
1009 	pvinfo->hcall[3] = inst_nop;
1010 #else
1011 	u32 inst_lis = 0x3c000000;
1012 	u32 inst_ori = 0x60000000;
1013 	u32 inst_sc = 0x44000002;
1014 	u32 inst_imm_mask = 0xffff;
1015 
1016 	/*
1017 	 * The hypercall to get into KVM from within guest context is as
1018 	 * follows:
1019 	 *
1020 	 *    lis r0, r0, KVM_SC_MAGIC_R0@h
1021 	 *    ori r0, KVM_SC_MAGIC_R0@l
1022 	 *    sc
1023 	 *    nop
1024 	 */
1025 	pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
1026 	pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
1027 	pvinfo->hcall[2] = inst_sc;
1028 	pvinfo->hcall[3] = inst_nop;
1029 #endif
1030 
1031 	pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1032 
1033 	return 0;
1034 }
1035 
1036 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1037 			  bool line_status)
1038 {
1039 	if (!irqchip_in_kernel(kvm))
1040 		return -ENXIO;
1041 
1042 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1043 					irq_event->irq, irq_event->level,
1044 					line_status);
1045 	return 0;
1046 }
1047 
1048 long kvm_arch_vm_ioctl(struct file *filp,
1049                        unsigned int ioctl, unsigned long arg)
1050 {
1051 	struct kvm *kvm __maybe_unused = filp->private_data;
1052 	void __user *argp = (void __user *)arg;
1053 	long r;
1054 
1055 	switch (ioctl) {
1056 	case KVM_PPC_GET_PVINFO: {
1057 		struct kvm_ppc_pvinfo pvinfo;
1058 		memset(&pvinfo, 0, sizeof(pvinfo));
1059 		r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1060 		if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1061 			r = -EFAULT;
1062 			goto out;
1063 		}
1064 
1065 		break;
1066 	}
1067 #ifdef CONFIG_PPC_BOOK3S_64
1068 	case KVM_CREATE_SPAPR_TCE: {
1069 		struct kvm_create_spapr_tce create_tce;
1070 
1071 		r = -EFAULT;
1072 		if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1073 			goto out;
1074 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
1075 		goto out;
1076 	}
1077 	case KVM_PPC_GET_SMMU_INFO: {
1078 		struct kvm_ppc_smmu_info info;
1079 		struct kvm *kvm = filp->private_data;
1080 
1081 		memset(&info, 0, sizeof(info));
1082 		r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1083 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1084 			r = -EFAULT;
1085 		break;
1086 	}
1087 	case KVM_PPC_RTAS_DEFINE_TOKEN: {
1088 		struct kvm *kvm = filp->private_data;
1089 
1090 		r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1091 		break;
1092 	}
1093 	default: {
1094 		struct kvm *kvm = filp->private_data;
1095 		r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1096 	}
1097 #else /* CONFIG_PPC_BOOK3S_64 */
1098 	default:
1099 		r = -ENOTTY;
1100 #endif
1101 	}
1102 out:
1103 	return r;
1104 }
1105 
1106 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1107 static unsigned long nr_lpids;
1108 
1109 long kvmppc_alloc_lpid(void)
1110 {
1111 	long lpid;
1112 
1113 	do {
1114 		lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1115 		if (lpid >= nr_lpids) {
1116 			pr_err("%s: No LPIDs free\n", __func__);
1117 			return -ENOMEM;
1118 		}
1119 	} while (test_and_set_bit(lpid, lpid_inuse));
1120 
1121 	return lpid;
1122 }
1123 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1124 
1125 void kvmppc_claim_lpid(long lpid)
1126 {
1127 	set_bit(lpid, lpid_inuse);
1128 }
1129 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1130 
1131 void kvmppc_free_lpid(long lpid)
1132 {
1133 	clear_bit(lpid, lpid_inuse);
1134 }
1135 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1136 
1137 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1138 {
1139 	nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1140 	memset(lpid_inuse, 0, sizeof(lpid_inuse));
1141 }
1142 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1143 
1144 int kvm_arch_init(void *opaque)
1145 {
1146 	return 0;
1147 }
1148 
1149 void kvm_arch_exit(void)
1150 {
1151 
1152 }
1153