xref: /openbmc/linux/arch/powerpc/kvm/powerpc.c (revision 37be287c)
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;
72 
73 	WARN_ON(irqs_disabled());
74 	hard_irq_disable();
75 
76 	while (true) {
77 		if (need_resched()) {
78 			local_irq_enable();
79 			cond_resched();
80 			hard_irq_disable();
81 			continue;
82 		}
83 
84 		if (signal_pending(current)) {
85 			kvmppc_account_exit(vcpu, SIGNAL_EXITS);
86 			vcpu->run->exit_reason = KVM_EXIT_INTR;
87 			r = -EINTR;
88 			break;
89 		}
90 
91 		vcpu->mode = IN_GUEST_MODE;
92 
93 		/*
94 		 * Reading vcpu->requests must happen after setting vcpu->mode,
95 		 * so we don't miss a request because the requester sees
96 		 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
97 		 * before next entering the guest (and thus doesn't IPI).
98 		 */
99 		smp_mb();
100 
101 		if (vcpu->requests) {
102 			/* Make sure we process requests preemptable */
103 			local_irq_enable();
104 			trace_kvm_check_requests(vcpu);
105 			r = kvmppc_core_check_requests(vcpu);
106 			hard_irq_disable();
107 			if (r > 0)
108 				continue;
109 			break;
110 		}
111 
112 		if (kvmppc_core_prepare_to_enter(vcpu)) {
113 			/* interrupts got enabled in between, so we
114 			   are back at square 1 */
115 			continue;
116 		}
117 
118 		kvm_guest_enter();
119 		return 1;
120 	}
121 
122 	/* return to host */
123 	local_irq_enable();
124 	return r;
125 }
126 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
127 
128 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
129 {
130 	int nr = kvmppc_get_gpr(vcpu, 11);
131 	int r;
132 	unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
133 	unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
134 	unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
135 	unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
136 	unsigned long r2 = 0;
137 
138 	if (!(vcpu->arch.shared->msr & MSR_SF)) {
139 		/* 32 bit mode */
140 		param1 &= 0xffffffff;
141 		param2 &= 0xffffffff;
142 		param3 &= 0xffffffff;
143 		param4 &= 0xffffffff;
144 	}
145 
146 	switch (nr) {
147 	case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
148 	{
149 		vcpu->arch.magic_page_pa = param1;
150 		vcpu->arch.magic_page_ea = param2;
151 
152 		r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
153 
154 		r = EV_SUCCESS;
155 		break;
156 	}
157 	case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
158 		r = EV_SUCCESS;
159 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
160 		/* XXX Missing magic page on 44x */
161 		r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
162 #endif
163 
164 		/* Second return value is in r4 */
165 		break;
166 	case EV_HCALL_TOKEN(EV_IDLE):
167 		r = EV_SUCCESS;
168 		kvm_vcpu_block(vcpu);
169 		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
170 		break;
171 	default:
172 		r = EV_UNIMPLEMENTED;
173 		break;
174 	}
175 
176 	kvmppc_set_gpr(vcpu, 4, r2);
177 
178 	return r;
179 }
180 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
181 
182 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
183 {
184 	int r = false;
185 
186 	/* We have to know what CPU to virtualize */
187 	if (!vcpu->arch.pvr)
188 		goto out;
189 
190 	/* PAPR only works with book3s_64 */
191 	if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
192 		goto out;
193 
194 	/* HV KVM can only do PAPR mode for now */
195 	if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
196 		goto out;
197 
198 #ifdef CONFIG_KVM_BOOKE_HV
199 	if (!cpu_has_feature(CPU_FTR_EMB_HV))
200 		goto out;
201 #endif
202 
203 	r = true;
204 
205 out:
206 	vcpu->arch.sane = r;
207 	return r ? 0 : -EINVAL;
208 }
209 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
210 
211 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
212 {
213 	enum emulation_result er;
214 	int r;
215 
216 	er = kvmppc_emulate_instruction(run, vcpu);
217 	switch (er) {
218 	case EMULATE_DONE:
219 		/* Future optimization: only reload non-volatiles if they were
220 		 * actually modified. */
221 		r = RESUME_GUEST_NV;
222 		break;
223 	case EMULATE_DO_MMIO:
224 		run->exit_reason = KVM_EXIT_MMIO;
225 		/* We must reload nonvolatiles because "update" load/store
226 		 * instructions modify register state. */
227 		/* Future optimization: only reload non-volatiles if they were
228 		 * actually modified. */
229 		r = RESUME_HOST_NV;
230 		break;
231 	case EMULATE_FAIL:
232 		/* XXX Deliver Program interrupt to guest. */
233 		printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__,
234 		       kvmppc_get_last_inst(vcpu));
235 		r = RESUME_HOST;
236 		break;
237 	default:
238 		WARN_ON(1);
239 		r = RESUME_GUEST;
240 	}
241 
242 	return r;
243 }
244 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
245 
246 int kvm_arch_hardware_enable(void *garbage)
247 {
248 	return 0;
249 }
250 
251 void kvm_arch_hardware_disable(void *garbage)
252 {
253 }
254 
255 int kvm_arch_hardware_setup(void)
256 {
257 	return 0;
258 }
259 
260 void kvm_arch_hardware_unsetup(void)
261 {
262 }
263 
264 void kvm_arch_check_processor_compat(void *rtn)
265 {
266 	*(int *)rtn = kvmppc_core_check_processor_compat();
267 }
268 
269 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
270 {
271 	struct kvmppc_ops *kvm_ops = NULL;
272 	/*
273 	 * if we have both HV and PR enabled, default is HV
274 	 */
275 	if (type == 0) {
276 		if (kvmppc_hv_ops)
277 			kvm_ops = kvmppc_hv_ops;
278 		else
279 			kvm_ops = kvmppc_pr_ops;
280 		if (!kvm_ops)
281 			goto err_out;
282 	} else	if (type == KVM_VM_PPC_HV) {
283 		if (!kvmppc_hv_ops)
284 			goto err_out;
285 		kvm_ops = kvmppc_hv_ops;
286 	} else if (type == KVM_VM_PPC_PR) {
287 		if (!kvmppc_pr_ops)
288 			goto err_out;
289 		kvm_ops = kvmppc_pr_ops;
290 	} else
291 		goto err_out;
292 
293 	if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
294 		return -ENOENT;
295 
296 	kvm->arch.kvm_ops = kvm_ops;
297 	return kvmppc_core_init_vm(kvm);
298 err_out:
299 	return -EINVAL;
300 }
301 
302 void kvm_arch_destroy_vm(struct kvm *kvm)
303 {
304 	unsigned int i;
305 	struct kvm_vcpu *vcpu;
306 
307 	kvm_for_each_vcpu(i, vcpu, kvm)
308 		kvm_arch_vcpu_free(vcpu);
309 
310 	mutex_lock(&kvm->lock);
311 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
312 		kvm->vcpus[i] = NULL;
313 
314 	atomic_set(&kvm->online_vcpus, 0);
315 
316 	kvmppc_core_destroy_vm(kvm);
317 
318 	mutex_unlock(&kvm->lock);
319 
320 	/* drop the module reference */
321 	module_put(kvm->arch.kvm_ops->owner);
322 }
323 
324 void kvm_arch_sync_events(struct kvm *kvm)
325 {
326 }
327 
328 int kvm_dev_ioctl_check_extension(long ext)
329 {
330 	int r;
331 	/* FIXME!!
332 	 * Should some of this be vm ioctl ? is it possible now ?
333 	 */
334 	int hv_enabled = kvmppc_hv_ops ? 1 : 0;
335 
336 	switch (ext) {
337 #ifdef CONFIG_BOOKE
338 	case KVM_CAP_PPC_BOOKE_SREGS:
339 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
340 	case KVM_CAP_PPC_EPR:
341 #else
342 	case KVM_CAP_PPC_SEGSTATE:
343 	case KVM_CAP_PPC_HIOR:
344 	case KVM_CAP_PPC_PAPR:
345 #endif
346 	case KVM_CAP_PPC_UNSET_IRQ:
347 	case KVM_CAP_PPC_IRQ_LEVEL:
348 	case KVM_CAP_ENABLE_CAP:
349 	case KVM_CAP_ONE_REG:
350 	case KVM_CAP_IOEVENTFD:
351 	case KVM_CAP_DEVICE_CTRL:
352 		r = 1;
353 		break;
354 	case KVM_CAP_PPC_PAIRED_SINGLES:
355 	case KVM_CAP_PPC_OSI:
356 	case KVM_CAP_PPC_GET_PVINFO:
357 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
358 	case KVM_CAP_SW_TLB:
359 #endif
360 		/* We support this only for PR */
361 		r = !hv_enabled;
362 		break;
363 #ifdef CONFIG_KVM_MMIO
364 	case KVM_CAP_COALESCED_MMIO:
365 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
366 		break;
367 #endif
368 #ifdef CONFIG_KVM_MPIC
369 	case KVM_CAP_IRQ_MPIC:
370 		r = 1;
371 		break;
372 #endif
373 
374 #ifdef CONFIG_PPC_BOOK3S_64
375 	case KVM_CAP_SPAPR_TCE:
376 	case KVM_CAP_PPC_ALLOC_HTAB:
377 	case KVM_CAP_PPC_RTAS:
378 #ifdef CONFIG_KVM_XICS
379 	case KVM_CAP_IRQ_XICS:
380 #endif
381 		r = 1;
382 		break;
383 #endif /* CONFIG_PPC_BOOK3S_64 */
384 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
385 	case KVM_CAP_PPC_SMT:
386 		if (hv_enabled)
387 			r = threads_per_core;
388 		else
389 			r = 0;
390 		break;
391 	case KVM_CAP_PPC_RMA:
392 		r = hv_enabled;
393 		/* PPC970 requires an RMA */
394 		if (r && cpu_has_feature(CPU_FTR_ARCH_201))
395 			r = 2;
396 		break;
397 #endif
398 	case KVM_CAP_SYNC_MMU:
399 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
400 		if (hv_enabled)
401 			r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
402 		else
403 			r = 0;
404 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
405 		r = 1;
406 #else
407 		r = 0;
408 #endif
409 		break;
410 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
411 	case KVM_CAP_PPC_HTAB_FD:
412 		r = hv_enabled;
413 		break;
414 #endif
415 	case KVM_CAP_NR_VCPUS:
416 		/*
417 		 * Recommending a number of CPUs is somewhat arbitrary; we
418 		 * return the number of present CPUs for -HV (since a host
419 		 * will have secondary threads "offline"), and for other KVM
420 		 * implementations just count online CPUs.
421 		 */
422 		if (hv_enabled)
423 			r = num_present_cpus();
424 		else
425 			r = num_online_cpus();
426 		break;
427 	case KVM_CAP_MAX_VCPUS:
428 		r = KVM_MAX_VCPUS;
429 		break;
430 #ifdef CONFIG_PPC_BOOK3S_64
431 	case KVM_CAP_PPC_GET_SMMU_INFO:
432 		r = 1;
433 		break;
434 #endif
435 	default:
436 		r = 0;
437 		break;
438 	}
439 	return r;
440 
441 }
442 
443 long kvm_arch_dev_ioctl(struct file *filp,
444                         unsigned int ioctl, unsigned long arg)
445 {
446 	return -EINVAL;
447 }
448 
449 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
450 			   struct kvm_memory_slot *dont)
451 {
452 	kvmppc_core_free_memslot(kvm, free, dont);
453 }
454 
455 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
456 			    unsigned long npages)
457 {
458 	return kvmppc_core_create_memslot(kvm, slot, npages);
459 }
460 
461 void kvm_arch_memslots_updated(struct kvm *kvm)
462 {
463 }
464 
465 int kvm_arch_prepare_memory_region(struct kvm *kvm,
466 				   struct kvm_memory_slot *memslot,
467 				   struct kvm_userspace_memory_region *mem,
468 				   enum kvm_mr_change change)
469 {
470 	return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
471 }
472 
473 void kvm_arch_commit_memory_region(struct kvm *kvm,
474 				   struct kvm_userspace_memory_region *mem,
475 				   const struct kvm_memory_slot *old,
476 				   enum kvm_mr_change change)
477 {
478 	kvmppc_core_commit_memory_region(kvm, mem, old);
479 }
480 
481 void kvm_arch_flush_shadow_all(struct kvm *kvm)
482 {
483 }
484 
485 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
486 				   struct kvm_memory_slot *slot)
487 {
488 	kvmppc_core_flush_memslot(kvm, slot);
489 }
490 
491 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
492 {
493 	struct kvm_vcpu *vcpu;
494 	vcpu = kvmppc_core_vcpu_create(kvm, id);
495 	if (!IS_ERR(vcpu)) {
496 		vcpu->arch.wqp = &vcpu->wq;
497 		kvmppc_create_vcpu_debugfs(vcpu, id);
498 	}
499 	return vcpu;
500 }
501 
502 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
503 {
504 	return 0;
505 }
506 
507 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
508 {
509 	/* Make sure we're not using the vcpu anymore */
510 	hrtimer_cancel(&vcpu->arch.dec_timer);
511 	tasklet_kill(&vcpu->arch.tasklet);
512 
513 	kvmppc_remove_vcpu_debugfs(vcpu);
514 
515 	switch (vcpu->arch.irq_type) {
516 	case KVMPPC_IRQ_MPIC:
517 		kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
518 		break;
519 	case KVMPPC_IRQ_XICS:
520 		kvmppc_xics_free_icp(vcpu);
521 		break;
522 	}
523 
524 	kvmppc_core_vcpu_free(vcpu);
525 }
526 
527 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
528 {
529 	kvm_arch_vcpu_free(vcpu);
530 }
531 
532 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
533 {
534 	return kvmppc_core_pending_dec(vcpu);
535 }
536 
537 /*
538  * low level hrtimer wake routine. Because this runs in hardirq context
539  * we schedule a tasklet to do the real work.
540  */
541 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
542 {
543 	struct kvm_vcpu *vcpu;
544 
545 	vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
546 	tasklet_schedule(&vcpu->arch.tasklet);
547 
548 	return HRTIMER_NORESTART;
549 }
550 
551 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
552 {
553 	int ret;
554 
555 	hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
556 	tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
557 	vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
558 	vcpu->arch.dec_expires = ~(u64)0;
559 
560 #ifdef CONFIG_KVM_EXIT_TIMING
561 	mutex_init(&vcpu->arch.exit_timing_lock);
562 #endif
563 	ret = kvmppc_subarch_vcpu_init(vcpu);
564 	return ret;
565 }
566 
567 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
568 {
569 	kvmppc_mmu_destroy(vcpu);
570 	kvmppc_subarch_vcpu_uninit(vcpu);
571 }
572 
573 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
574 {
575 #ifdef CONFIG_BOOKE
576 	/*
577 	 * vrsave (formerly usprg0) isn't used by Linux, but may
578 	 * be used by the guest.
579 	 *
580 	 * On non-booke this is associated with Altivec and
581 	 * is handled by code in book3s.c.
582 	 */
583 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
584 #endif
585 	kvmppc_core_vcpu_load(vcpu, cpu);
586 }
587 
588 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
589 {
590 	kvmppc_core_vcpu_put(vcpu);
591 #ifdef CONFIG_BOOKE
592 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
593 #endif
594 }
595 
596 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
597                                      struct kvm_run *run)
598 {
599 	kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
600 }
601 
602 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
603                                       struct kvm_run *run)
604 {
605 	u64 uninitialized_var(gpr);
606 
607 	if (run->mmio.len > sizeof(gpr)) {
608 		printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
609 		return;
610 	}
611 
612 	if (vcpu->arch.mmio_is_bigendian) {
613 		switch (run->mmio.len) {
614 		case 8: gpr = *(u64 *)run->mmio.data; break;
615 		case 4: gpr = *(u32 *)run->mmio.data; break;
616 		case 2: gpr = *(u16 *)run->mmio.data; break;
617 		case 1: gpr = *(u8 *)run->mmio.data; break;
618 		}
619 	} else {
620 		/* Convert BE data from userland back to LE. */
621 		switch (run->mmio.len) {
622 		case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
623 		case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
624 		case 1: gpr = *(u8 *)run->mmio.data; break;
625 		}
626 	}
627 
628 	if (vcpu->arch.mmio_sign_extend) {
629 		switch (run->mmio.len) {
630 #ifdef CONFIG_PPC64
631 		case 4:
632 			gpr = (s64)(s32)gpr;
633 			break;
634 #endif
635 		case 2:
636 			gpr = (s64)(s16)gpr;
637 			break;
638 		case 1:
639 			gpr = (s64)(s8)gpr;
640 			break;
641 		}
642 	}
643 
644 	kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
645 
646 	switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
647 	case KVM_MMIO_REG_GPR:
648 		kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
649 		break;
650 	case KVM_MMIO_REG_FPR:
651 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
652 		break;
653 #ifdef CONFIG_PPC_BOOK3S
654 	case KVM_MMIO_REG_QPR:
655 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
656 		break;
657 	case KVM_MMIO_REG_FQPR:
658 		VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
659 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
660 		break;
661 #endif
662 	default:
663 		BUG();
664 	}
665 }
666 
667 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
668 		       unsigned int rt, unsigned int bytes,
669 		       int is_default_endian)
670 {
671 	int idx, ret;
672 	int is_bigendian;
673 
674 	if (kvmppc_need_byteswap(vcpu)) {
675 		/* Default endianness is "little endian". */
676 		is_bigendian = !is_default_endian;
677 	} else {
678 		/* Default endianness is "big endian". */
679 		is_bigendian = is_default_endian;
680 	}
681 
682 	if (bytes > sizeof(run->mmio.data)) {
683 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
684 		       run->mmio.len);
685 	}
686 
687 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
688 	run->mmio.len = bytes;
689 	run->mmio.is_write = 0;
690 
691 	vcpu->arch.io_gpr = rt;
692 	vcpu->arch.mmio_is_bigendian = is_bigendian;
693 	vcpu->mmio_needed = 1;
694 	vcpu->mmio_is_write = 0;
695 	vcpu->arch.mmio_sign_extend = 0;
696 
697 	idx = srcu_read_lock(&vcpu->kvm->srcu);
698 
699 	ret = kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
700 			      bytes, &run->mmio.data);
701 
702 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
703 
704 	if (!ret) {
705 		kvmppc_complete_mmio_load(vcpu, run);
706 		vcpu->mmio_needed = 0;
707 		return EMULATE_DONE;
708 	}
709 
710 	return EMULATE_DO_MMIO;
711 }
712 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
713 
714 /* Same as above, but sign extends */
715 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
716 			unsigned int rt, unsigned int bytes,
717 			int is_default_endian)
718 {
719 	int r;
720 
721 	vcpu->arch.mmio_sign_extend = 1;
722 	r = kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian);
723 
724 	return r;
725 }
726 
727 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
728 			u64 val, unsigned int bytes, int is_default_endian)
729 {
730 	void *data = run->mmio.data;
731 	int idx, ret;
732 	int is_bigendian;
733 
734 	if (kvmppc_need_byteswap(vcpu)) {
735 		/* Default endianness is "little endian". */
736 		is_bigendian = !is_default_endian;
737 	} else {
738 		/* Default endianness is "big endian". */
739 		is_bigendian = is_default_endian;
740 	}
741 
742 	if (bytes > sizeof(run->mmio.data)) {
743 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
744 		       run->mmio.len);
745 	}
746 
747 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
748 	run->mmio.len = bytes;
749 	run->mmio.is_write = 1;
750 	vcpu->mmio_needed = 1;
751 	vcpu->mmio_is_write = 1;
752 
753 	/* Store the value at the lowest bytes in 'data'. */
754 	if (is_bigendian) {
755 		switch (bytes) {
756 		case 8: *(u64 *)data = val; break;
757 		case 4: *(u32 *)data = val; break;
758 		case 2: *(u16 *)data = val; break;
759 		case 1: *(u8  *)data = val; break;
760 		}
761 	} else {
762 		/* Store LE value into 'data'. */
763 		switch (bytes) {
764 		case 4: st_le32(data, val); break;
765 		case 2: st_le16(data, val); break;
766 		case 1: *(u8 *)data = val; break;
767 		}
768 	}
769 
770 	idx = srcu_read_lock(&vcpu->kvm->srcu);
771 
772 	ret = kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
773 			       bytes, &run->mmio.data);
774 
775 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
776 
777 	if (!ret) {
778 		vcpu->mmio_needed = 0;
779 		return EMULATE_DONE;
780 	}
781 
782 	return EMULATE_DO_MMIO;
783 }
784 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
785 
786 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
787 {
788 	int r;
789 	sigset_t sigsaved;
790 
791 	if (vcpu->sigset_active)
792 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
793 
794 	if (vcpu->mmio_needed) {
795 		if (!vcpu->mmio_is_write)
796 			kvmppc_complete_mmio_load(vcpu, run);
797 		vcpu->mmio_needed = 0;
798 	} else if (vcpu->arch.dcr_needed) {
799 		if (!vcpu->arch.dcr_is_write)
800 			kvmppc_complete_dcr_load(vcpu, run);
801 		vcpu->arch.dcr_needed = 0;
802 	} else if (vcpu->arch.osi_needed) {
803 		u64 *gprs = run->osi.gprs;
804 		int i;
805 
806 		for (i = 0; i < 32; i++)
807 			kvmppc_set_gpr(vcpu, i, gprs[i]);
808 		vcpu->arch.osi_needed = 0;
809 	} else if (vcpu->arch.hcall_needed) {
810 		int i;
811 
812 		kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
813 		for (i = 0; i < 9; ++i)
814 			kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
815 		vcpu->arch.hcall_needed = 0;
816 #ifdef CONFIG_BOOKE
817 	} else if (vcpu->arch.epr_needed) {
818 		kvmppc_set_epr(vcpu, run->epr.epr);
819 		vcpu->arch.epr_needed = 0;
820 #endif
821 	}
822 
823 	r = kvmppc_vcpu_run(run, vcpu);
824 
825 	if (vcpu->sigset_active)
826 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
827 
828 	return r;
829 }
830 
831 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
832 {
833 	if (irq->irq == KVM_INTERRUPT_UNSET) {
834 		kvmppc_core_dequeue_external(vcpu);
835 		return 0;
836 	}
837 
838 	kvmppc_core_queue_external(vcpu, irq);
839 
840 	kvm_vcpu_kick(vcpu);
841 
842 	return 0;
843 }
844 
845 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
846 				     struct kvm_enable_cap *cap)
847 {
848 	int r;
849 
850 	if (cap->flags)
851 		return -EINVAL;
852 
853 	switch (cap->cap) {
854 	case KVM_CAP_PPC_OSI:
855 		r = 0;
856 		vcpu->arch.osi_enabled = true;
857 		break;
858 	case KVM_CAP_PPC_PAPR:
859 		r = 0;
860 		vcpu->arch.papr_enabled = true;
861 		break;
862 	case KVM_CAP_PPC_EPR:
863 		r = 0;
864 		if (cap->args[0])
865 			vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
866 		else
867 			vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
868 		break;
869 #ifdef CONFIG_BOOKE
870 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
871 		r = 0;
872 		vcpu->arch.watchdog_enabled = true;
873 		break;
874 #endif
875 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
876 	case KVM_CAP_SW_TLB: {
877 		struct kvm_config_tlb cfg;
878 		void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
879 
880 		r = -EFAULT;
881 		if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
882 			break;
883 
884 		r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
885 		break;
886 	}
887 #endif
888 #ifdef CONFIG_KVM_MPIC
889 	case KVM_CAP_IRQ_MPIC: {
890 		struct fd f;
891 		struct kvm_device *dev;
892 
893 		r = -EBADF;
894 		f = fdget(cap->args[0]);
895 		if (!f.file)
896 			break;
897 
898 		r = -EPERM;
899 		dev = kvm_device_from_filp(f.file);
900 		if (dev)
901 			r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
902 
903 		fdput(f);
904 		break;
905 	}
906 #endif
907 #ifdef CONFIG_KVM_XICS
908 	case KVM_CAP_IRQ_XICS: {
909 		struct fd f;
910 		struct kvm_device *dev;
911 
912 		r = -EBADF;
913 		f = fdget(cap->args[0]);
914 		if (!f.file)
915 			break;
916 
917 		r = -EPERM;
918 		dev = kvm_device_from_filp(f.file);
919 		if (dev)
920 			r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
921 
922 		fdput(f);
923 		break;
924 	}
925 #endif /* CONFIG_KVM_XICS */
926 	default:
927 		r = -EINVAL;
928 		break;
929 	}
930 
931 	if (!r)
932 		r = kvmppc_sanity_check(vcpu);
933 
934 	return r;
935 }
936 
937 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
938                                     struct kvm_mp_state *mp_state)
939 {
940 	return -EINVAL;
941 }
942 
943 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
944                                     struct kvm_mp_state *mp_state)
945 {
946 	return -EINVAL;
947 }
948 
949 long kvm_arch_vcpu_ioctl(struct file *filp,
950                          unsigned int ioctl, unsigned long arg)
951 {
952 	struct kvm_vcpu *vcpu = filp->private_data;
953 	void __user *argp = (void __user *)arg;
954 	long r;
955 
956 	switch (ioctl) {
957 	case KVM_INTERRUPT: {
958 		struct kvm_interrupt irq;
959 		r = -EFAULT;
960 		if (copy_from_user(&irq, argp, sizeof(irq)))
961 			goto out;
962 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
963 		goto out;
964 	}
965 
966 	case KVM_ENABLE_CAP:
967 	{
968 		struct kvm_enable_cap cap;
969 		r = -EFAULT;
970 		if (copy_from_user(&cap, argp, sizeof(cap)))
971 			goto out;
972 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
973 		break;
974 	}
975 
976 	case KVM_SET_ONE_REG:
977 	case KVM_GET_ONE_REG:
978 	{
979 		struct kvm_one_reg reg;
980 		r = -EFAULT;
981 		if (copy_from_user(&reg, argp, sizeof(reg)))
982 			goto out;
983 		if (ioctl == KVM_SET_ONE_REG)
984 			r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
985 		else
986 			r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
987 		break;
988 	}
989 
990 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
991 	case KVM_DIRTY_TLB: {
992 		struct kvm_dirty_tlb dirty;
993 		r = -EFAULT;
994 		if (copy_from_user(&dirty, argp, sizeof(dirty)))
995 			goto out;
996 		r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
997 		break;
998 	}
999 #endif
1000 	default:
1001 		r = -EINVAL;
1002 	}
1003 
1004 out:
1005 	return r;
1006 }
1007 
1008 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1009 {
1010 	return VM_FAULT_SIGBUS;
1011 }
1012 
1013 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1014 {
1015 	u32 inst_nop = 0x60000000;
1016 #ifdef CONFIG_KVM_BOOKE_HV
1017 	u32 inst_sc1 = 0x44000022;
1018 	pvinfo->hcall[0] = inst_sc1;
1019 	pvinfo->hcall[1] = inst_nop;
1020 	pvinfo->hcall[2] = inst_nop;
1021 	pvinfo->hcall[3] = inst_nop;
1022 #else
1023 	u32 inst_lis = 0x3c000000;
1024 	u32 inst_ori = 0x60000000;
1025 	u32 inst_sc = 0x44000002;
1026 	u32 inst_imm_mask = 0xffff;
1027 
1028 	/*
1029 	 * The hypercall to get into KVM from within guest context is as
1030 	 * follows:
1031 	 *
1032 	 *    lis r0, r0, KVM_SC_MAGIC_R0@h
1033 	 *    ori r0, KVM_SC_MAGIC_R0@l
1034 	 *    sc
1035 	 *    nop
1036 	 */
1037 	pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
1038 	pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
1039 	pvinfo->hcall[2] = inst_sc;
1040 	pvinfo->hcall[3] = inst_nop;
1041 #endif
1042 
1043 	pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1044 
1045 	return 0;
1046 }
1047 
1048 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1049 			  bool line_status)
1050 {
1051 	if (!irqchip_in_kernel(kvm))
1052 		return -ENXIO;
1053 
1054 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1055 					irq_event->irq, irq_event->level,
1056 					line_status);
1057 	return 0;
1058 }
1059 
1060 long kvm_arch_vm_ioctl(struct file *filp,
1061                        unsigned int ioctl, unsigned long arg)
1062 {
1063 	struct kvm *kvm __maybe_unused = filp->private_data;
1064 	void __user *argp = (void __user *)arg;
1065 	long r;
1066 
1067 	switch (ioctl) {
1068 	case KVM_PPC_GET_PVINFO: {
1069 		struct kvm_ppc_pvinfo pvinfo;
1070 		memset(&pvinfo, 0, sizeof(pvinfo));
1071 		r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1072 		if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1073 			r = -EFAULT;
1074 			goto out;
1075 		}
1076 
1077 		break;
1078 	}
1079 #ifdef CONFIG_PPC_BOOK3S_64
1080 	case KVM_CREATE_SPAPR_TCE: {
1081 		struct kvm_create_spapr_tce create_tce;
1082 
1083 		r = -EFAULT;
1084 		if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1085 			goto out;
1086 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
1087 		goto out;
1088 	}
1089 	case KVM_PPC_GET_SMMU_INFO: {
1090 		struct kvm_ppc_smmu_info info;
1091 		struct kvm *kvm = filp->private_data;
1092 
1093 		memset(&info, 0, sizeof(info));
1094 		r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1095 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1096 			r = -EFAULT;
1097 		break;
1098 	}
1099 	case KVM_PPC_RTAS_DEFINE_TOKEN: {
1100 		struct kvm *kvm = filp->private_data;
1101 
1102 		r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1103 		break;
1104 	}
1105 	default: {
1106 		struct kvm *kvm = filp->private_data;
1107 		r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1108 	}
1109 #else /* CONFIG_PPC_BOOK3S_64 */
1110 	default:
1111 		r = -ENOTTY;
1112 #endif
1113 	}
1114 out:
1115 	return r;
1116 }
1117 
1118 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1119 static unsigned long nr_lpids;
1120 
1121 long kvmppc_alloc_lpid(void)
1122 {
1123 	long lpid;
1124 
1125 	do {
1126 		lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1127 		if (lpid >= nr_lpids) {
1128 			pr_err("%s: No LPIDs free\n", __func__);
1129 			return -ENOMEM;
1130 		}
1131 	} while (test_and_set_bit(lpid, lpid_inuse));
1132 
1133 	return lpid;
1134 }
1135 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1136 
1137 void kvmppc_claim_lpid(long lpid)
1138 {
1139 	set_bit(lpid, lpid_inuse);
1140 }
1141 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1142 
1143 void kvmppc_free_lpid(long lpid)
1144 {
1145 	clear_bit(lpid, lpid_inuse);
1146 }
1147 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1148 
1149 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1150 {
1151 	nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1152 	memset(lpid_inuse, 0, sizeof(lpid_inuse));
1153 }
1154 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1155 
1156 int kvm_arch_init(void *opaque)
1157 {
1158 	return 0;
1159 }
1160 
1161 void kvm_arch_exit(void)
1162 {
1163 
1164 }
1165