xref: /openbmc/linux/arch/powerpc/kvm/powerpc.c (revision 8fdff1dc)
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 <asm/cputable.h>
29 #include <asm/uaccess.h>
30 #include <asm/kvm_ppc.h>
31 #include <asm/tlbflush.h>
32 #include <asm/cputhreads.h>
33 #include <asm/irqflags.h>
34 #include "timing.h"
35 #include "../mm/mmu_decl.h"
36 
37 #define CREATE_TRACE_POINTS
38 #include "trace.h"
39 
40 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
41 {
42 	return !!(v->arch.pending_exceptions) ||
43 	       v->requests;
44 }
45 
46 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
47 {
48 	return 1;
49 }
50 
51 #ifndef CONFIG_KVM_BOOK3S_64_HV
52 /*
53  * Common checks before entering the guest world.  Call with interrupts
54  * disabled.
55  *
56  * returns:
57  *
58  * == 1 if we're ready to go into guest state
59  * <= 0 if we need to go back to the host with return value
60  */
61 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
62 {
63 	int r = 1;
64 
65 	WARN_ON_ONCE(!irqs_disabled());
66 	while (true) {
67 		if (need_resched()) {
68 			local_irq_enable();
69 			cond_resched();
70 			local_irq_disable();
71 			continue;
72 		}
73 
74 		if (signal_pending(current)) {
75 			kvmppc_account_exit(vcpu, SIGNAL_EXITS);
76 			vcpu->run->exit_reason = KVM_EXIT_INTR;
77 			r = -EINTR;
78 			break;
79 		}
80 
81 		vcpu->mode = IN_GUEST_MODE;
82 
83 		/*
84 		 * Reading vcpu->requests must happen after setting vcpu->mode,
85 		 * so we don't miss a request because the requester sees
86 		 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
87 		 * before next entering the guest (and thus doesn't IPI).
88 		 */
89 		smp_mb();
90 
91 		if (vcpu->requests) {
92 			/* Make sure we process requests preemptable */
93 			local_irq_enable();
94 			trace_kvm_check_requests(vcpu);
95 			r = kvmppc_core_check_requests(vcpu);
96 			local_irq_disable();
97 			if (r > 0)
98 				continue;
99 			break;
100 		}
101 
102 		if (kvmppc_core_prepare_to_enter(vcpu)) {
103 			/* interrupts got enabled in between, so we
104 			   are back at square 1 */
105 			continue;
106 		}
107 
108 #ifdef CONFIG_PPC64
109 		/* lazy EE magic */
110 		hard_irq_disable();
111 		if (lazy_irq_pending()) {
112 			/* Got an interrupt in between, try again */
113 			local_irq_enable();
114 			local_irq_disable();
115 			kvm_guest_exit();
116 			continue;
117 		}
118 
119 		trace_hardirqs_on();
120 #endif
121 
122 		kvm_guest_enter();
123 		break;
124 	}
125 
126 	return r;
127 }
128 #endif /* CONFIG_KVM_BOOK3S_64_HV */
129 
130 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
131 {
132 	int nr = kvmppc_get_gpr(vcpu, 11);
133 	int r;
134 	unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
135 	unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
136 	unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
137 	unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
138 	unsigned long r2 = 0;
139 
140 	if (!(vcpu->arch.shared->msr & MSR_SF)) {
141 		/* 32 bit mode */
142 		param1 &= 0xffffffff;
143 		param2 &= 0xffffffff;
144 		param3 &= 0xffffffff;
145 		param4 &= 0xffffffff;
146 	}
147 
148 	switch (nr) {
149 	case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
150 	{
151 		vcpu->arch.magic_page_pa = param1;
152 		vcpu->arch.magic_page_ea = param2;
153 
154 		r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
155 
156 		r = EV_SUCCESS;
157 		break;
158 	}
159 	case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
160 		r = EV_SUCCESS;
161 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
162 		/* XXX Missing magic page on 44x */
163 		r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
164 #endif
165 
166 		/* Second return value is in r4 */
167 		break;
168 	case EV_HCALL_TOKEN(EV_IDLE):
169 		r = EV_SUCCESS;
170 		kvm_vcpu_block(vcpu);
171 		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
172 		break;
173 	default:
174 		r = EV_UNIMPLEMENTED;
175 		break;
176 	}
177 
178 	kvmppc_set_gpr(vcpu, 4, r2);
179 
180 	return r;
181 }
182 
183 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
184 {
185 	int r = false;
186 
187 	/* We have to know what CPU to virtualize */
188 	if (!vcpu->arch.pvr)
189 		goto out;
190 
191 	/* PAPR only works with book3s_64 */
192 	if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
193 		goto out;
194 
195 #ifdef CONFIG_KVM_BOOK3S_64_HV
196 	/* HV KVM can only do PAPR mode for now */
197 	if (!vcpu->arch.papr_enabled)
198 		goto out;
199 #endif
200 
201 #ifdef CONFIG_KVM_BOOKE_HV
202 	if (!cpu_has_feature(CPU_FTR_EMB_HV))
203 		goto out;
204 #endif
205 
206 	r = true;
207 
208 out:
209 	vcpu->arch.sane = r;
210 	return r ? 0 : -EINVAL;
211 }
212 
213 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
214 {
215 	enum emulation_result er;
216 	int r;
217 
218 	er = kvmppc_emulate_instruction(run, vcpu);
219 	switch (er) {
220 	case EMULATE_DONE:
221 		/* Future optimization: only reload non-volatiles if they were
222 		 * actually modified. */
223 		r = RESUME_GUEST_NV;
224 		break;
225 	case EMULATE_DO_MMIO:
226 		run->exit_reason = KVM_EXIT_MMIO;
227 		/* We must reload nonvolatiles because "update" load/store
228 		 * instructions modify register state. */
229 		/* Future optimization: only reload non-volatiles if they were
230 		 * actually modified. */
231 		r = RESUME_HOST_NV;
232 		break;
233 	case EMULATE_FAIL:
234 		/* XXX Deliver Program interrupt to guest. */
235 		printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__,
236 		       kvmppc_get_last_inst(vcpu));
237 		r = RESUME_HOST;
238 		break;
239 	default:
240 		BUG();
241 	}
242 
243 	return r;
244 }
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 	if (type)
272 		return -EINVAL;
273 
274 	return kvmppc_core_init_vm(kvm);
275 }
276 
277 void kvm_arch_destroy_vm(struct kvm *kvm)
278 {
279 	unsigned int i;
280 	struct kvm_vcpu *vcpu;
281 
282 	kvm_for_each_vcpu(i, vcpu, kvm)
283 		kvm_arch_vcpu_free(vcpu);
284 
285 	mutex_lock(&kvm->lock);
286 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
287 		kvm->vcpus[i] = NULL;
288 
289 	atomic_set(&kvm->online_vcpus, 0);
290 
291 	kvmppc_core_destroy_vm(kvm);
292 
293 	mutex_unlock(&kvm->lock);
294 }
295 
296 void kvm_arch_sync_events(struct kvm *kvm)
297 {
298 }
299 
300 int kvm_dev_ioctl_check_extension(long ext)
301 {
302 	int r;
303 
304 	switch (ext) {
305 #ifdef CONFIG_BOOKE
306 	case KVM_CAP_PPC_BOOKE_SREGS:
307 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
308 #else
309 	case KVM_CAP_PPC_SEGSTATE:
310 	case KVM_CAP_PPC_HIOR:
311 	case KVM_CAP_PPC_PAPR:
312 #endif
313 	case KVM_CAP_PPC_UNSET_IRQ:
314 	case KVM_CAP_PPC_IRQ_LEVEL:
315 	case KVM_CAP_ENABLE_CAP:
316 	case KVM_CAP_ONE_REG:
317 	case KVM_CAP_IOEVENTFD:
318 		r = 1;
319 		break;
320 #ifndef CONFIG_KVM_BOOK3S_64_HV
321 	case KVM_CAP_PPC_PAIRED_SINGLES:
322 	case KVM_CAP_PPC_OSI:
323 	case KVM_CAP_PPC_GET_PVINFO:
324 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
325 	case KVM_CAP_SW_TLB:
326 #endif
327 		r = 1;
328 		break;
329 	case KVM_CAP_COALESCED_MMIO:
330 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
331 		break;
332 #endif
333 #ifdef CONFIG_PPC_BOOK3S_64
334 	case KVM_CAP_SPAPR_TCE:
335 	case KVM_CAP_PPC_ALLOC_HTAB:
336 		r = 1;
337 		break;
338 #endif /* CONFIG_PPC_BOOK3S_64 */
339 #ifdef CONFIG_KVM_BOOK3S_64_HV
340 	case KVM_CAP_PPC_SMT:
341 		r = threads_per_core;
342 		break;
343 	case KVM_CAP_PPC_RMA:
344 		r = 1;
345 		/* PPC970 requires an RMA */
346 		if (cpu_has_feature(CPU_FTR_ARCH_201))
347 			r = 2;
348 		break;
349 #endif
350 	case KVM_CAP_SYNC_MMU:
351 #ifdef CONFIG_KVM_BOOK3S_64_HV
352 		r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
353 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
354 		r = 1;
355 #else
356 		r = 0;
357 		break;
358 #endif
359 #ifdef CONFIG_KVM_BOOK3S_64_HV
360 	case KVM_CAP_PPC_HTAB_FD:
361 		r = 1;
362 		break;
363 #endif
364 		break;
365 	case KVM_CAP_NR_VCPUS:
366 		/*
367 		 * Recommending a number of CPUs is somewhat arbitrary; we
368 		 * return the number of present CPUs for -HV (since a host
369 		 * will have secondary threads "offline"), and for other KVM
370 		 * implementations just count online CPUs.
371 		 */
372 #ifdef CONFIG_KVM_BOOK3S_64_HV
373 		r = num_present_cpus();
374 #else
375 		r = num_online_cpus();
376 #endif
377 		break;
378 	case KVM_CAP_MAX_VCPUS:
379 		r = KVM_MAX_VCPUS;
380 		break;
381 #ifdef CONFIG_PPC_BOOK3S_64
382 	case KVM_CAP_PPC_GET_SMMU_INFO:
383 		r = 1;
384 		break;
385 #endif
386 	default:
387 		r = 0;
388 		break;
389 	}
390 	return r;
391 
392 }
393 
394 long kvm_arch_dev_ioctl(struct file *filp,
395                         unsigned int ioctl, unsigned long arg)
396 {
397 	return -EINVAL;
398 }
399 
400 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
401 			   struct kvm_memory_slot *dont)
402 {
403 	kvmppc_core_free_memslot(free, dont);
404 }
405 
406 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
407 {
408 	return kvmppc_core_create_memslot(slot, npages);
409 }
410 
411 int kvm_arch_prepare_memory_region(struct kvm *kvm,
412                                    struct kvm_memory_slot *memslot,
413                                    struct kvm_memory_slot old,
414                                    struct kvm_userspace_memory_region *mem,
415                                    int user_alloc)
416 {
417 	return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
418 }
419 
420 void kvm_arch_commit_memory_region(struct kvm *kvm,
421                struct kvm_userspace_memory_region *mem,
422                struct kvm_memory_slot old,
423                int user_alloc)
424 {
425 	kvmppc_core_commit_memory_region(kvm, mem, old);
426 }
427 
428 void kvm_arch_flush_shadow_all(struct kvm *kvm)
429 {
430 }
431 
432 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
433 				   struct kvm_memory_slot *slot)
434 {
435 	kvmppc_core_flush_memslot(kvm, slot);
436 }
437 
438 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
439 {
440 	struct kvm_vcpu *vcpu;
441 	vcpu = kvmppc_core_vcpu_create(kvm, id);
442 	if (!IS_ERR(vcpu)) {
443 		vcpu->arch.wqp = &vcpu->wq;
444 		kvmppc_create_vcpu_debugfs(vcpu, id);
445 	}
446 	return vcpu;
447 }
448 
449 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
450 {
451 	return 0;
452 }
453 
454 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
455 {
456 	/* Make sure we're not using the vcpu anymore */
457 	hrtimer_cancel(&vcpu->arch.dec_timer);
458 	tasklet_kill(&vcpu->arch.tasklet);
459 
460 	kvmppc_remove_vcpu_debugfs(vcpu);
461 	kvmppc_core_vcpu_free(vcpu);
462 }
463 
464 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
465 {
466 	kvm_arch_vcpu_free(vcpu);
467 }
468 
469 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
470 {
471 	return kvmppc_core_pending_dec(vcpu);
472 }
473 
474 /*
475  * low level hrtimer wake routine. Because this runs in hardirq context
476  * we schedule a tasklet to do the real work.
477  */
478 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
479 {
480 	struct kvm_vcpu *vcpu;
481 
482 	vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
483 	tasklet_schedule(&vcpu->arch.tasklet);
484 
485 	return HRTIMER_NORESTART;
486 }
487 
488 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
489 {
490 	int ret;
491 
492 	hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
493 	tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
494 	vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
495 	vcpu->arch.dec_expires = ~(u64)0;
496 
497 #ifdef CONFIG_KVM_EXIT_TIMING
498 	mutex_init(&vcpu->arch.exit_timing_lock);
499 #endif
500 	ret = kvmppc_subarch_vcpu_init(vcpu);
501 	return ret;
502 }
503 
504 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
505 {
506 	kvmppc_mmu_destroy(vcpu);
507 	kvmppc_subarch_vcpu_uninit(vcpu);
508 }
509 
510 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
511 {
512 #ifdef CONFIG_BOOKE
513 	/*
514 	 * vrsave (formerly usprg0) isn't used by Linux, but may
515 	 * be used by the guest.
516 	 *
517 	 * On non-booke this is associated with Altivec and
518 	 * is handled by code in book3s.c.
519 	 */
520 	mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
521 #endif
522 	kvmppc_core_vcpu_load(vcpu, cpu);
523 }
524 
525 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
526 {
527 	kvmppc_core_vcpu_put(vcpu);
528 #ifdef CONFIG_BOOKE
529 	vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
530 #endif
531 }
532 
533 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
534                                         struct kvm_guest_debug *dbg)
535 {
536 	return -EINVAL;
537 }
538 
539 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
540                                      struct kvm_run *run)
541 {
542 	kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
543 }
544 
545 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
546                                       struct kvm_run *run)
547 {
548 	u64 uninitialized_var(gpr);
549 
550 	if (run->mmio.len > sizeof(gpr)) {
551 		printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
552 		return;
553 	}
554 
555 	if (vcpu->arch.mmio_is_bigendian) {
556 		switch (run->mmio.len) {
557 		case 8: gpr = *(u64 *)run->mmio.data; break;
558 		case 4: gpr = *(u32 *)run->mmio.data; break;
559 		case 2: gpr = *(u16 *)run->mmio.data; break;
560 		case 1: gpr = *(u8 *)run->mmio.data; break;
561 		}
562 	} else {
563 		/* Convert BE data from userland back to LE. */
564 		switch (run->mmio.len) {
565 		case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
566 		case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
567 		case 1: gpr = *(u8 *)run->mmio.data; break;
568 		}
569 	}
570 
571 	if (vcpu->arch.mmio_sign_extend) {
572 		switch (run->mmio.len) {
573 #ifdef CONFIG_PPC64
574 		case 4:
575 			gpr = (s64)(s32)gpr;
576 			break;
577 #endif
578 		case 2:
579 			gpr = (s64)(s16)gpr;
580 			break;
581 		case 1:
582 			gpr = (s64)(s8)gpr;
583 			break;
584 		}
585 	}
586 
587 	kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
588 
589 	switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
590 	case KVM_MMIO_REG_GPR:
591 		kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
592 		break;
593 	case KVM_MMIO_REG_FPR:
594 		vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
595 		break;
596 #ifdef CONFIG_PPC_BOOK3S
597 	case KVM_MMIO_REG_QPR:
598 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
599 		break;
600 	case KVM_MMIO_REG_FQPR:
601 		vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
602 		vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
603 		break;
604 #endif
605 	default:
606 		BUG();
607 	}
608 }
609 
610 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
611                        unsigned int rt, unsigned int bytes, int is_bigendian)
612 {
613 	if (bytes > sizeof(run->mmio.data)) {
614 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
615 		       run->mmio.len);
616 	}
617 
618 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
619 	run->mmio.len = bytes;
620 	run->mmio.is_write = 0;
621 
622 	vcpu->arch.io_gpr = rt;
623 	vcpu->arch.mmio_is_bigendian = is_bigendian;
624 	vcpu->mmio_needed = 1;
625 	vcpu->mmio_is_write = 0;
626 	vcpu->arch.mmio_sign_extend = 0;
627 
628 	if (!kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
629 			     bytes, &run->mmio.data)) {
630 		kvmppc_complete_mmio_load(vcpu, run);
631 		vcpu->mmio_needed = 0;
632 		return EMULATE_DONE;
633 	}
634 
635 	return EMULATE_DO_MMIO;
636 }
637 
638 /* Same as above, but sign extends */
639 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
640                         unsigned int rt, unsigned int bytes, int is_bigendian)
641 {
642 	int r;
643 
644 	vcpu->arch.mmio_sign_extend = 1;
645 	r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian);
646 
647 	return r;
648 }
649 
650 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
651                         u64 val, unsigned int bytes, int is_bigendian)
652 {
653 	void *data = run->mmio.data;
654 
655 	if (bytes > sizeof(run->mmio.data)) {
656 		printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
657 		       run->mmio.len);
658 	}
659 
660 	run->mmio.phys_addr = vcpu->arch.paddr_accessed;
661 	run->mmio.len = bytes;
662 	run->mmio.is_write = 1;
663 	vcpu->mmio_needed = 1;
664 	vcpu->mmio_is_write = 1;
665 
666 	/* Store the value at the lowest bytes in 'data'. */
667 	if (is_bigendian) {
668 		switch (bytes) {
669 		case 8: *(u64 *)data = val; break;
670 		case 4: *(u32 *)data = val; break;
671 		case 2: *(u16 *)data = val; break;
672 		case 1: *(u8  *)data = val; break;
673 		}
674 	} else {
675 		/* Store LE value into 'data'. */
676 		switch (bytes) {
677 		case 4: st_le32(data, val); break;
678 		case 2: st_le16(data, val); break;
679 		case 1: *(u8 *)data = val; break;
680 		}
681 	}
682 
683 	if (!kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
684 			      bytes, &run->mmio.data)) {
685 		kvmppc_complete_mmio_load(vcpu, run);
686 		vcpu->mmio_needed = 0;
687 		return EMULATE_DONE;
688 	}
689 
690 	return EMULATE_DO_MMIO;
691 }
692 
693 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
694 {
695 	int r;
696 	sigset_t sigsaved;
697 
698 	if (vcpu->sigset_active)
699 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
700 
701 	if (vcpu->mmio_needed) {
702 		if (!vcpu->mmio_is_write)
703 			kvmppc_complete_mmio_load(vcpu, run);
704 		vcpu->mmio_needed = 0;
705 	} else if (vcpu->arch.dcr_needed) {
706 		if (!vcpu->arch.dcr_is_write)
707 			kvmppc_complete_dcr_load(vcpu, run);
708 		vcpu->arch.dcr_needed = 0;
709 	} else if (vcpu->arch.osi_needed) {
710 		u64 *gprs = run->osi.gprs;
711 		int i;
712 
713 		for (i = 0; i < 32; i++)
714 			kvmppc_set_gpr(vcpu, i, gprs[i]);
715 		vcpu->arch.osi_needed = 0;
716 	} else if (vcpu->arch.hcall_needed) {
717 		int i;
718 
719 		kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
720 		for (i = 0; i < 9; ++i)
721 			kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
722 		vcpu->arch.hcall_needed = 0;
723 	}
724 
725 	r = kvmppc_vcpu_run(run, vcpu);
726 
727 	if (vcpu->sigset_active)
728 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
729 
730 	return r;
731 }
732 
733 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
734 {
735 	if (irq->irq == KVM_INTERRUPT_UNSET) {
736 		kvmppc_core_dequeue_external(vcpu, irq);
737 		return 0;
738 	}
739 
740 	kvmppc_core_queue_external(vcpu, irq);
741 
742 	kvm_vcpu_kick(vcpu);
743 
744 	return 0;
745 }
746 
747 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
748 				     struct kvm_enable_cap *cap)
749 {
750 	int r;
751 
752 	if (cap->flags)
753 		return -EINVAL;
754 
755 	switch (cap->cap) {
756 	case KVM_CAP_PPC_OSI:
757 		r = 0;
758 		vcpu->arch.osi_enabled = true;
759 		break;
760 	case KVM_CAP_PPC_PAPR:
761 		r = 0;
762 		vcpu->arch.papr_enabled = true;
763 		break;
764 #ifdef CONFIG_BOOKE
765 	case KVM_CAP_PPC_BOOKE_WATCHDOG:
766 		r = 0;
767 		vcpu->arch.watchdog_enabled = true;
768 		break;
769 #endif
770 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
771 	case KVM_CAP_SW_TLB: {
772 		struct kvm_config_tlb cfg;
773 		void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
774 
775 		r = -EFAULT;
776 		if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
777 			break;
778 
779 		r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
780 		break;
781 	}
782 #endif
783 	default:
784 		r = -EINVAL;
785 		break;
786 	}
787 
788 	if (!r)
789 		r = kvmppc_sanity_check(vcpu);
790 
791 	return r;
792 }
793 
794 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
795                                     struct kvm_mp_state *mp_state)
796 {
797 	return -EINVAL;
798 }
799 
800 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
801                                     struct kvm_mp_state *mp_state)
802 {
803 	return -EINVAL;
804 }
805 
806 long kvm_arch_vcpu_ioctl(struct file *filp,
807                          unsigned int ioctl, unsigned long arg)
808 {
809 	struct kvm_vcpu *vcpu = filp->private_data;
810 	void __user *argp = (void __user *)arg;
811 	long r;
812 
813 	switch (ioctl) {
814 	case KVM_INTERRUPT: {
815 		struct kvm_interrupt irq;
816 		r = -EFAULT;
817 		if (copy_from_user(&irq, argp, sizeof(irq)))
818 			goto out;
819 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
820 		goto out;
821 	}
822 
823 	case KVM_ENABLE_CAP:
824 	{
825 		struct kvm_enable_cap cap;
826 		r = -EFAULT;
827 		if (copy_from_user(&cap, argp, sizeof(cap)))
828 			goto out;
829 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
830 		break;
831 	}
832 
833 	case KVM_SET_ONE_REG:
834 	case KVM_GET_ONE_REG:
835 	{
836 		struct kvm_one_reg reg;
837 		r = -EFAULT;
838 		if (copy_from_user(&reg, argp, sizeof(reg)))
839 			goto out;
840 		if (ioctl == KVM_SET_ONE_REG)
841 			r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
842 		else
843 			r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
844 		break;
845 	}
846 
847 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
848 	case KVM_DIRTY_TLB: {
849 		struct kvm_dirty_tlb dirty;
850 		r = -EFAULT;
851 		if (copy_from_user(&dirty, argp, sizeof(dirty)))
852 			goto out;
853 		r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
854 		break;
855 	}
856 #endif
857 	default:
858 		r = -EINVAL;
859 	}
860 
861 out:
862 	return r;
863 }
864 
865 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
866 {
867 	return VM_FAULT_SIGBUS;
868 }
869 
870 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
871 {
872 	u32 inst_nop = 0x60000000;
873 #ifdef CONFIG_KVM_BOOKE_HV
874 	u32 inst_sc1 = 0x44000022;
875 	pvinfo->hcall[0] = inst_sc1;
876 	pvinfo->hcall[1] = inst_nop;
877 	pvinfo->hcall[2] = inst_nop;
878 	pvinfo->hcall[3] = inst_nop;
879 #else
880 	u32 inst_lis = 0x3c000000;
881 	u32 inst_ori = 0x60000000;
882 	u32 inst_sc = 0x44000002;
883 	u32 inst_imm_mask = 0xffff;
884 
885 	/*
886 	 * The hypercall to get into KVM from within guest context is as
887 	 * follows:
888 	 *
889 	 *    lis r0, r0, KVM_SC_MAGIC_R0@h
890 	 *    ori r0, KVM_SC_MAGIC_R0@l
891 	 *    sc
892 	 *    nop
893 	 */
894 	pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
895 	pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
896 	pvinfo->hcall[2] = inst_sc;
897 	pvinfo->hcall[3] = inst_nop;
898 #endif
899 
900 	pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
901 
902 	return 0;
903 }
904 
905 long kvm_arch_vm_ioctl(struct file *filp,
906                        unsigned int ioctl, unsigned long arg)
907 {
908 	void __user *argp = (void __user *)arg;
909 	long r;
910 
911 	switch (ioctl) {
912 	case KVM_PPC_GET_PVINFO: {
913 		struct kvm_ppc_pvinfo pvinfo;
914 		memset(&pvinfo, 0, sizeof(pvinfo));
915 		r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
916 		if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
917 			r = -EFAULT;
918 			goto out;
919 		}
920 
921 		break;
922 	}
923 #ifdef CONFIG_PPC_BOOK3S_64
924 	case KVM_CREATE_SPAPR_TCE: {
925 		struct kvm_create_spapr_tce create_tce;
926 		struct kvm *kvm = filp->private_data;
927 
928 		r = -EFAULT;
929 		if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
930 			goto out;
931 		r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
932 		goto out;
933 	}
934 #endif /* CONFIG_PPC_BOOK3S_64 */
935 
936 #ifdef CONFIG_KVM_BOOK3S_64_HV
937 	case KVM_ALLOCATE_RMA: {
938 		struct kvm *kvm = filp->private_data;
939 		struct kvm_allocate_rma rma;
940 
941 		r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
942 		if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
943 			r = -EFAULT;
944 		break;
945 	}
946 
947 	case KVM_PPC_ALLOCATE_HTAB: {
948 		struct kvm *kvm = filp->private_data;
949 		u32 htab_order;
950 
951 		r = -EFAULT;
952 		if (get_user(htab_order, (u32 __user *)argp))
953 			break;
954 		r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
955 		if (r)
956 			break;
957 		r = -EFAULT;
958 		if (put_user(htab_order, (u32 __user *)argp))
959 			break;
960 		r = 0;
961 		break;
962 	}
963 
964 	case KVM_PPC_GET_HTAB_FD: {
965 		struct kvm *kvm = filp->private_data;
966 		struct kvm_get_htab_fd ghf;
967 
968 		r = -EFAULT;
969 		if (copy_from_user(&ghf, argp, sizeof(ghf)))
970 			break;
971 		r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
972 		break;
973 	}
974 #endif /* CONFIG_KVM_BOOK3S_64_HV */
975 
976 #ifdef CONFIG_PPC_BOOK3S_64
977 	case KVM_PPC_GET_SMMU_INFO: {
978 		struct kvm *kvm = filp->private_data;
979 		struct kvm_ppc_smmu_info info;
980 
981 		memset(&info, 0, sizeof(info));
982 		r = kvm_vm_ioctl_get_smmu_info(kvm, &info);
983 		if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
984 			r = -EFAULT;
985 		break;
986 	}
987 #endif /* CONFIG_PPC_BOOK3S_64 */
988 	default:
989 		r = -ENOTTY;
990 	}
991 
992 out:
993 	return r;
994 }
995 
996 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
997 static unsigned long nr_lpids;
998 
999 long kvmppc_alloc_lpid(void)
1000 {
1001 	long lpid;
1002 
1003 	do {
1004 		lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1005 		if (lpid >= nr_lpids) {
1006 			pr_err("%s: No LPIDs free\n", __func__);
1007 			return -ENOMEM;
1008 		}
1009 	} while (test_and_set_bit(lpid, lpid_inuse));
1010 
1011 	return lpid;
1012 }
1013 
1014 void kvmppc_claim_lpid(long lpid)
1015 {
1016 	set_bit(lpid, lpid_inuse);
1017 }
1018 
1019 void kvmppc_free_lpid(long lpid)
1020 {
1021 	clear_bit(lpid, lpid_inuse);
1022 }
1023 
1024 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1025 {
1026 	nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1027 	memset(lpid_inuse, 0, sizeof(lpid_inuse));
1028 }
1029 
1030 int kvm_arch_init(void *opaque)
1031 {
1032 	return 0;
1033 }
1034 
1035 void kvm_arch_exit(void)
1036 {
1037 }
1038