xref: /openbmc/linux/arch/powerpc/kvm/book3s.c (revision df2634f43f5106947f3735a0b61a6527a4b278cd)
1 /*
2  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
3  *
4  * Authors:
5  *    Alexander Graf <agraf@suse.de>
6  *    Kevin Wolf <mail@kevin-wolf.de>
7  *
8  * Description:
9  * This file is derived from arch/powerpc/kvm/44x.c,
10  * by Hollis Blanchard <hollisb@us.ibm.com>.
11  *
12  * This program is free software; you can redistribute it and/or modify
13  * it under the terms of the GNU General Public License, version 2, as
14  * published by the Free Software Foundation.
15  */
16 
17 #include <linux/kvm_host.h>
18 #include <linux/err.h>
19 #include <linux/slab.h>
20 #include "trace.h"
21 
22 #include <asm/reg.h>
23 #include <asm/cputable.h>
24 #include <asm/cacheflush.h>
25 #include <asm/tlbflush.h>
26 #include <asm/uaccess.h>
27 #include <asm/io.h>
28 #include <asm/kvm_ppc.h>
29 #include <asm/kvm_book3s.h>
30 #include <asm/mmu_context.h>
31 #include <linux/gfp.h>
32 #include <linux/sched.h>
33 #include <linux/vmalloc.h>
34 #include <linux/highmem.h>
35 
36 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
37 
38 /* #define EXIT_DEBUG */
39 /* #define DEBUG_EXT */
40 
41 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
42 			     ulong msr);
43 
44 /* Some compatibility defines */
45 #ifdef CONFIG_PPC_BOOK3S_32
46 #define MSR_USER32 MSR_USER
47 #define MSR_USER64 MSR_USER
48 #define HW_PAGE_SIZE PAGE_SIZE
49 #endif
50 
51 struct kvm_stats_debugfs_item debugfs_entries[] = {
52 	{ "exits",       VCPU_STAT(sum_exits) },
53 	{ "mmio",        VCPU_STAT(mmio_exits) },
54 	{ "sig",         VCPU_STAT(signal_exits) },
55 	{ "sysc",        VCPU_STAT(syscall_exits) },
56 	{ "inst_emu",    VCPU_STAT(emulated_inst_exits) },
57 	{ "dec",         VCPU_STAT(dec_exits) },
58 	{ "ext_intr",    VCPU_STAT(ext_intr_exits) },
59 	{ "queue_intr",  VCPU_STAT(queue_intr) },
60 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
61 	{ "pf_storage",  VCPU_STAT(pf_storage) },
62 	{ "sp_storage",  VCPU_STAT(sp_storage) },
63 	{ "pf_instruc",  VCPU_STAT(pf_instruc) },
64 	{ "sp_instruc",  VCPU_STAT(sp_instruc) },
65 	{ "ld",          VCPU_STAT(ld) },
66 	{ "ld_slow",     VCPU_STAT(ld_slow) },
67 	{ "st",          VCPU_STAT(st) },
68 	{ "st_slow",     VCPU_STAT(st_slow) },
69 	{ NULL }
70 };
71 
72 void kvmppc_core_load_host_debugstate(struct kvm_vcpu *vcpu)
73 {
74 }
75 
76 void kvmppc_core_load_guest_debugstate(struct kvm_vcpu *vcpu)
77 {
78 }
79 
80 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
81 {
82 #ifdef CONFIG_PPC_BOOK3S_64
83 	memcpy(to_svcpu(vcpu)->slb, to_book3s(vcpu)->slb_shadow, sizeof(to_svcpu(vcpu)->slb));
84 	memcpy(&get_paca()->shadow_vcpu, to_book3s(vcpu)->shadow_vcpu,
85 	       sizeof(get_paca()->shadow_vcpu));
86 	to_svcpu(vcpu)->slb_max = to_book3s(vcpu)->slb_shadow_max;
87 #endif
88 
89 #ifdef CONFIG_PPC_BOOK3S_32
90 	current->thread.kvm_shadow_vcpu = to_book3s(vcpu)->shadow_vcpu;
91 #endif
92 }
93 
94 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
95 {
96 #ifdef CONFIG_PPC_BOOK3S_64
97 	memcpy(to_book3s(vcpu)->slb_shadow, to_svcpu(vcpu)->slb, sizeof(to_svcpu(vcpu)->slb));
98 	memcpy(to_book3s(vcpu)->shadow_vcpu, &get_paca()->shadow_vcpu,
99 	       sizeof(get_paca()->shadow_vcpu));
100 	to_book3s(vcpu)->slb_shadow_max = to_svcpu(vcpu)->slb_max;
101 #endif
102 
103 	kvmppc_giveup_ext(vcpu, MSR_FP);
104 	kvmppc_giveup_ext(vcpu, MSR_VEC);
105 	kvmppc_giveup_ext(vcpu, MSR_VSX);
106 }
107 
108 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
109 {
110 	ulong smsr = vcpu->arch.shared->msr;
111 
112 	/* Guest MSR values */
113 	smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_DE;
114 	/* Process MSR values */
115 	smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
116 	/* External providers the guest reserved */
117 	smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
118 	/* 64-bit Process MSR values */
119 #ifdef CONFIG_PPC_BOOK3S_64
120 	smsr |= MSR_ISF | MSR_HV;
121 #endif
122 	vcpu->arch.shadow_msr = smsr;
123 }
124 
125 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
126 {
127 	ulong old_msr = vcpu->arch.shared->msr;
128 
129 #ifdef EXIT_DEBUG
130 	printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
131 #endif
132 
133 	msr &= to_book3s(vcpu)->msr_mask;
134 	vcpu->arch.shared->msr = msr;
135 	kvmppc_recalc_shadow_msr(vcpu);
136 
137 	if (msr & MSR_POW) {
138 		if (!vcpu->arch.pending_exceptions) {
139 			kvm_vcpu_block(vcpu);
140 			vcpu->stat.halt_wakeup++;
141 
142 			/* Unset POW bit after we woke up */
143 			msr &= ~MSR_POW;
144 			vcpu->arch.shared->msr = msr;
145 		}
146 	}
147 
148 	if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
149 		   (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
150 		kvmppc_mmu_flush_segments(vcpu);
151 		kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
152 
153 		/* Preload magic page segment when in kernel mode */
154 		if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
155 			struct kvm_vcpu_arch *a = &vcpu->arch;
156 
157 			if (msr & MSR_DR)
158 				kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
159 			else
160 				kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
161 		}
162 	}
163 
164 	/* Preload FPU if it's enabled */
165 	if (vcpu->arch.shared->msr & MSR_FP)
166 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
167 }
168 
169 void kvmppc_inject_interrupt(struct kvm_vcpu *vcpu, int vec, u64 flags)
170 {
171 	vcpu->arch.shared->srr0 = kvmppc_get_pc(vcpu);
172 	vcpu->arch.shared->srr1 = vcpu->arch.shared->msr | flags;
173 	kvmppc_set_pc(vcpu, to_book3s(vcpu)->hior + vec);
174 	vcpu->arch.mmu.reset_msr(vcpu);
175 }
176 
177 static int kvmppc_book3s_vec2irqprio(unsigned int vec)
178 {
179 	unsigned int prio;
180 
181 	switch (vec) {
182 	case 0x100: prio = BOOK3S_IRQPRIO_SYSTEM_RESET;		break;
183 	case 0x200: prio = BOOK3S_IRQPRIO_MACHINE_CHECK;	break;
184 	case 0x300: prio = BOOK3S_IRQPRIO_DATA_STORAGE;		break;
185 	case 0x380: prio = BOOK3S_IRQPRIO_DATA_SEGMENT;		break;
186 	case 0x400: prio = BOOK3S_IRQPRIO_INST_STORAGE;		break;
187 	case 0x480: prio = BOOK3S_IRQPRIO_INST_SEGMENT;		break;
188 	case 0x500: prio = BOOK3S_IRQPRIO_EXTERNAL;		break;
189 	case 0x501: prio = BOOK3S_IRQPRIO_EXTERNAL_LEVEL;	break;
190 	case 0x600: prio = BOOK3S_IRQPRIO_ALIGNMENT;		break;
191 	case 0x700: prio = BOOK3S_IRQPRIO_PROGRAM;		break;
192 	case 0x800: prio = BOOK3S_IRQPRIO_FP_UNAVAIL;		break;
193 	case 0x900: prio = BOOK3S_IRQPRIO_DECREMENTER;		break;
194 	case 0xc00: prio = BOOK3S_IRQPRIO_SYSCALL;		break;
195 	case 0xd00: prio = BOOK3S_IRQPRIO_DEBUG;		break;
196 	case 0xf20: prio = BOOK3S_IRQPRIO_ALTIVEC;		break;
197 	case 0xf40: prio = BOOK3S_IRQPRIO_VSX;			break;
198 	default:    prio = BOOK3S_IRQPRIO_MAX;			break;
199 	}
200 
201 	return prio;
202 }
203 
204 static void kvmppc_book3s_dequeue_irqprio(struct kvm_vcpu *vcpu,
205 					  unsigned int vec)
206 {
207 	clear_bit(kvmppc_book3s_vec2irqprio(vec),
208 		  &vcpu->arch.pending_exceptions);
209 
210 	if (!vcpu->arch.pending_exceptions)
211 		vcpu->arch.shared->int_pending = 0;
212 }
213 
214 void kvmppc_book3s_queue_irqprio(struct kvm_vcpu *vcpu, unsigned int vec)
215 {
216 	vcpu->stat.queue_intr++;
217 
218 	set_bit(kvmppc_book3s_vec2irqprio(vec),
219 		&vcpu->arch.pending_exceptions);
220 #ifdef EXIT_DEBUG
221 	printk(KERN_INFO "Queueing interrupt %x\n", vec);
222 #endif
223 }
224 
225 
226 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu, ulong flags)
227 {
228 	to_book3s(vcpu)->prog_flags = flags;
229 	kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_PROGRAM);
230 }
231 
232 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
233 {
234 	kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
235 }
236 
237 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
238 {
239 	return test_bit(BOOK3S_INTERRUPT_DECREMENTER >> 7, &vcpu->arch.pending_exceptions);
240 }
241 
242 void kvmppc_core_dequeue_dec(struct kvm_vcpu *vcpu)
243 {
244 	kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
245 }
246 
247 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
248                                 struct kvm_interrupt *irq)
249 {
250 	unsigned int vec = BOOK3S_INTERRUPT_EXTERNAL;
251 
252 	if (irq->irq == KVM_INTERRUPT_SET_LEVEL)
253 		vec = BOOK3S_INTERRUPT_EXTERNAL_LEVEL;
254 
255 	kvmppc_book3s_queue_irqprio(vcpu, vec);
256 }
257 
258 void kvmppc_core_dequeue_external(struct kvm_vcpu *vcpu,
259                                   struct kvm_interrupt *irq)
260 {
261 	kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL);
262 	kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL_LEVEL);
263 }
264 
265 int kvmppc_book3s_irqprio_deliver(struct kvm_vcpu *vcpu, unsigned int priority)
266 {
267 	int deliver = 1;
268 	int vec = 0;
269 	ulong flags = 0ULL;
270 	ulong crit_raw = vcpu->arch.shared->critical;
271 	ulong crit_r1 = kvmppc_get_gpr(vcpu, 1);
272 	bool crit;
273 
274 	/* Truncate crit indicators in 32 bit mode */
275 	if (!(vcpu->arch.shared->msr & MSR_SF)) {
276 		crit_raw &= 0xffffffff;
277 		crit_r1 &= 0xffffffff;
278 	}
279 
280 	/* Critical section when crit == r1 */
281 	crit = (crit_raw == crit_r1);
282 	/* ... and we're in supervisor mode */
283 	crit = crit && !(vcpu->arch.shared->msr & MSR_PR);
284 
285 	switch (priority) {
286 	case BOOK3S_IRQPRIO_DECREMENTER:
287 		deliver = (vcpu->arch.shared->msr & MSR_EE) && !crit;
288 		vec = BOOK3S_INTERRUPT_DECREMENTER;
289 		break;
290 	case BOOK3S_IRQPRIO_EXTERNAL:
291 	case BOOK3S_IRQPRIO_EXTERNAL_LEVEL:
292 		deliver = (vcpu->arch.shared->msr & MSR_EE) && !crit;
293 		vec = BOOK3S_INTERRUPT_EXTERNAL;
294 		break;
295 	case BOOK3S_IRQPRIO_SYSTEM_RESET:
296 		vec = BOOK3S_INTERRUPT_SYSTEM_RESET;
297 		break;
298 	case BOOK3S_IRQPRIO_MACHINE_CHECK:
299 		vec = BOOK3S_INTERRUPT_MACHINE_CHECK;
300 		break;
301 	case BOOK3S_IRQPRIO_DATA_STORAGE:
302 		vec = BOOK3S_INTERRUPT_DATA_STORAGE;
303 		break;
304 	case BOOK3S_IRQPRIO_INST_STORAGE:
305 		vec = BOOK3S_INTERRUPT_INST_STORAGE;
306 		break;
307 	case BOOK3S_IRQPRIO_DATA_SEGMENT:
308 		vec = BOOK3S_INTERRUPT_DATA_SEGMENT;
309 		break;
310 	case BOOK3S_IRQPRIO_INST_SEGMENT:
311 		vec = BOOK3S_INTERRUPT_INST_SEGMENT;
312 		break;
313 	case BOOK3S_IRQPRIO_ALIGNMENT:
314 		vec = BOOK3S_INTERRUPT_ALIGNMENT;
315 		break;
316 	case BOOK3S_IRQPRIO_PROGRAM:
317 		vec = BOOK3S_INTERRUPT_PROGRAM;
318 		flags = to_book3s(vcpu)->prog_flags;
319 		break;
320 	case BOOK3S_IRQPRIO_VSX:
321 		vec = BOOK3S_INTERRUPT_VSX;
322 		break;
323 	case BOOK3S_IRQPRIO_ALTIVEC:
324 		vec = BOOK3S_INTERRUPT_ALTIVEC;
325 		break;
326 	case BOOK3S_IRQPRIO_FP_UNAVAIL:
327 		vec = BOOK3S_INTERRUPT_FP_UNAVAIL;
328 		break;
329 	case BOOK3S_IRQPRIO_SYSCALL:
330 		vec = BOOK3S_INTERRUPT_SYSCALL;
331 		break;
332 	case BOOK3S_IRQPRIO_DEBUG:
333 		vec = BOOK3S_INTERRUPT_TRACE;
334 		break;
335 	case BOOK3S_IRQPRIO_PERFORMANCE_MONITOR:
336 		vec = BOOK3S_INTERRUPT_PERFMON;
337 		break;
338 	default:
339 		deliver = 0;
340 		printk(KERN_ERR "KVM: Unknown interrupt: 0x%x\n", priority);
341 		break;
342 	}
343 
344 #if 0
345 	printk(KERN_INFO "Deliver interrupt 0x%x? %x\n", vec, deliver);
346 #endif
347 
348 	if (deliver)
349 		kvmppc_inject_interrupt(vcpu, vec, flags);
350 
351 	return deliver;
352 }
353 
354 /*
355  * This function determines if an irqprio should be cleared once issued.
356  */
357 static bool clear_irqprio(struct kvm_vcpu *vcpu, unsigned int priority)
358 {
359 	switch (priority) {
360 		case BOOK3S_IRQPRIO_DECREMENTER:
361 			/* DEC interrupts get cleared by mtdec */
362 			return false;
363 		case BOOK3S_IRQPRIO_EXTERNAL_LEVEL:
364 			/* External interrupts get cleared by userspace */
365 			return false;
366 	}
367 
368 	return true;
369 }
370 
371 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
372 {
373 	unsigned long *pending = &vcpu->arch.pending_exceptions;
374 	unsigned long old_pending = vcpu->arch.pending_exceptions;
375 	unsigned int priority;
376 
377 #ifdef EXIT_DEBUG
378 	if (vcpu->arch.pending_exceptions)
379 		printk(KERN_EMERG "KVM: Check pending: %lx\n", vcpu->arch.pending_exceptions);
380 #endif
381 	priority = __ffs(*pending);
382 	while (priority < BOOK3S_IRQPRIO_MAX) {
383 		if (kvmppc_book3s_irqprio_deliver(vcpu, priority) &&
384 		    clear_irqprio(vcpu, priority)) {
385 			clear_bit(priority, &vcpu->arch.pending_exceptions);
386 			break;
387 		}
388 
389 		priority = find_next_bit(pending,
390 					 BITS_PER_BYTE * sizeof(*pending),
391 					 priority + 1);
392 	}
393 
394 	/* Tell the guest about our interrupt status */
395 	if (*pending)
396 		vcpu->arch.shared->int_pending = 1;
397 	else if (old_pending)
398 		vcpu->arch.shared->int_pending = 0;
399 }
400 
401 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
402 {
403 	u32 host_pvr;
404 
405 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
406 	vcpu->arch.pvr = pvr;
407 #ifdef CONFIG_PPC_BOOK3S_64
408 	if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
409 		kvmppc_mmu_book3s_64_init(vcpu);
410 		to_book3s(vcpu)->hior = 0xfff00000;
411 		to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
412 	} else
413 #endif
414 	{
415 		kvmppc_mmu_book3s_32_init(vcpu);
416 		to_book3s(vcpu)->hior = 0;
417 		to_book3s(vcpu)->msr_mask = 0xffffffffULL;
418 	}
419 
420 	/* If we are in hypervisor level on 970, we can tell the CPU to
421 	 * treat DCBZ as 32 bytes store */
422 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
423 	if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
424 	    !strcmp(cur_cpu_spec->platform, "ppc970"))
425 		vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
426 
427 	/* Cell performs badly if MSR_FEx are set. So let's hope nobody
428 	   really needs them in a VM on Cell and force disable them. */
429 	if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
430 		to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
431 
432 #ifdef CONFIG_PPC_BOOK3S_32
433 	/* 32 bit Book3S always has 32 byte dcbz */
434 	vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
435 #endif
436 
437 	/* On some CPUs we can execute paired single operations natively */
438 	asm ( "mfpvr %0" : "=r"(host_pvr));
439 	switch (host_pvr) {
440 	case 0x00080200:	/* lonestar 2.0 */
441 	case 0x00088202:	/* lonestar 2.2 */
442 	case 0x70000100:	/* gekko 1.0 */
443 	case 0x00080100:	/* gekko 2.0 */
444 	case 0x00083203:	/* gekko 2.3a */
445 	case 0x00083213:	/* gekko 2.3b */
446 	case 0x00083204:	/* gekko 2.4 */
447 	case 0x00083214:	/* gekko 2.4e (8SE) - retail HW2 */
448 	case 0x00087200:	/* broadway */
449 		vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
450 		/* Enable HID2.PSE - in case we need it later */
451 		mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
452 	}
453 }
454 
455 pfn_t kvmppc_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
456 {
457 	ulong mp_pa = vcpu->arch.magic_page_pa;
458 
459 	/* Magic page override */
460 	if (unlikely(mp_pa) &&
461 	    unlikely(((gfn << PAGE_SHIFT) & KVM_PAM) ==
462 		     ((mp_pa & PAGE_MASK) & KVM_PAM))) {
463 		ulong shared_page = ((ulong)vcpu->arch.shared) & PAGE_MASK;
464 		pfn_t pfn;
465 
466 		pfn = (pfn_t)virt_to_phys((void*)shared_page) >> PAGE_SHIFT;
467 		get_page(pfn_to_page(pfn));
468 		return pfn;
469 	}
470 
471 	return gfn_to_pfn(vcpu->kvm, gfn);
472 }
473 
474 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
475  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
476  * emulate 32 bytes dcbz length.
477  *
478  * The Book3s_64 inventors also realized this case and implemented a special bit
479  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
480  *
481  * My approach here is to patch the dcbz instruction on executing pages.
482  */
483 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
484 {
485 	struct page *hpage;
486 	u64 hpage_offset;
487 	u32 *page;
488 	int i;
489 
490 	hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
491 	if (is_error_page(hpage)) {
492 		kvm_release_page_clean(hpage);
493 		return;
494 	}
495 
496 	hpage_offset = pte->raddr & ~PAGE_MASK;
497 	hpage_offset &= ~0xFFFULL;
498 	hpage_offset /= 4;
499 
500 	get_page(hpage);
501 	page = kmap_atomic(hpage, KM_USER0);
502 
503 	/* patch dcbz into reserved instruction, so we trap */
504 	for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
505 		if ((page[i] & 0xff0007ff) == INS_DCBZ)
506 			page[i] &= 0xfffffff7;
507 
508 	kunmap_atomic(page, KM_USER0);
509 	put_page(hpage);
510 }
511 
512 static int kvmppc_xlate(struct kvm_vcpu *vcpu, ulong eaddr, bool data,
513 			 struct kvmppc_pte *pte)
514 {
515 	int relocated = (vcpu->arch.shared->msr & (data ? MSR_DR : MSR_IR));
516 	int r;
517 
518 	if (relocated) {
519 		r = vcpu->arch.mmu.xlate(vcpu, eaddr, pte, data);
520 	} else {
521 		pte->eaddr = eaddr;
522 		pte->raddr = eaddr & KVM_PAM;
523 		pte->vpage = VSID_REAL | eaddr >> 12;
524 		pte->may_read = true;
525 		pte->may_write = true;
526 		pte->may_execute = true;
527 		r = 0;
528 	}
529 
530 	return r;
531 }
532 
533 static hva_t kvmppc_bad_hva(void)
534 {
535 	return PAGE_OFFSET;
536 }
537 
538 static hva_t kvmppc_pte_to_hva(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte,
539 			       bool read)
540 {
541 	hva_t hpage;
542 
543 	if (read && !pte->may_read)
544 		goto err;
545 
546 	if (!read && !pte->may_write)
547 		goto err;
548 
549 	hpage = gfn_to_hva(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
550 	if (kvm_is_error_hva(hpage))
551 		goto err;
552 
553 	return hpage | (pte->raddr & ~PAGE_MASK);
554 err:
555 	return kvmppc_bad_hva();
556 }
557 
558 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
559 	      bool data)
560 {
561 	struct kvmppc_pte pte;
562 
563 	vcpu->stat.st++;
564 
565 	if (kvmppc_xlate(vcpu, *eaddr, data, &pte))
566 		return -ENOENT;
567 
568 	*eaddr = pte.raddr;
569 
570 	if (!pte.may_write)
571 		return -EPERM;
572 
573 	if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
574 		return EMULATE_DO_MMIO;
575 
576 	return EMULATE_DONE;
577 }
578 
579 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
580 		      bool data)
581 {
582 	struct kvmppc_pte pte;
583 	hva_t hva = *eaddr;
584 
585 	vcpu->stat.ld++;
586 
587 	if (kvmppc_xlate(vcpu, *eaddr, data, &pte))
588 		goto nopte;
589 
590 	*eaddr = pte.raddr;
591 
592 	hva = kvmppc_pte_to_hva(vcpu, &pte, true);
593 	if (kvm_is_error_hva(hva))
594 		goto mmio;
595 
596 	if (copy_from_user(ptr, (void __user *)hva, size)) {
597 		printk(KERN_INFO "kvmppc_ld at 0x%lx failed\n", hva);
598 		goto mmio;
599 	}
600 
601 	return EMULATE_DONE;
602 
603 nopte:
604 	return -ENOENT;
605 mmio:
606 	return EMULATE_DO_MMIO;
607 }
608 
609 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
610 {
611 	ulong mp_pa = vcpu->arch.magic_page_pa;
612 
613 	if (unlikely(mp_pa) &&
614 	    unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
615 		return 1;
616 	}
617 
618 	return kvm_is_visible_gfn(vcpu->kvm, gfn);
619 }
620 
621 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
622 			    ulong eaddr, int vec)
623 {
624 	bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
625 	int r = RESUME_GUEST;
626 	int relocated;
627 	int page_found = 0;
628 	struct kvmppc_pte pte;
629 	bool is_mmio = false;
630 	bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
631 	bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
632 	u64 vsid;
633 
634 	relocated = data ? dr : ir;
635 
636 	/* Resolve real address if translation turned on */
637 	if (relocated) {
638 		page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
639 	} else {
640 		pte.may_execute = true;
641 		pte.may_read = true;
642 		pte.may_write = true;
643 		pte.raddr = eaddr & KVM_PAM;
644 		pte.eaddr = eaddr;
645 		pte.vpage = eaddr >> 12;
646 	}
647 
648 	switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
649 	case 0:
650 		pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
651 		break;
652 	case MSR_DR:
653 	case MSR_IR:
654 		vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
655 
656 		if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
657 			pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
658 		else
659 			pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
660 		pte.vpage |= vsid;
661 
662 		if (vsid == -1)
663 			page_found = -EINVAL;
664 		break;
665 	}
666 
667 	if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
668 	   (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
669 		/*
670 		 * If we do the dcbz hack, we have to NX on every execution,
671 		 * so we can patch the executing code. This renders our guest
672 		 * NX-less.
673 		 */
674 		pte.may_execute = !data;
675 	}
676 
677 	if (page_found == -ENOENT) {
678 		/* Page not found in guest PTE entries */
679 		vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
680 		vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
681 		vcpu->arch.shared->msr |=
682 			(to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
683 		kvmppc_book3s_queue_irqprio(vcpu, vec);
684 	} else if (page_found == -EPERM) {
685 		/* Storage protection */
686 		vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
687 		vcpu->arch.shared->dsisr =
688 			to_svcpu(vcpu)->fault_dsisr & ~DSISR_NOHPTE;
689 		vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
690 		vcpu->arch.shared->msr |=
691 			(to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
692 		kvmppc_book3s_queue_irqprio(vcpu, vec);
693 	} else if (page_found == -EINVAL) {
694 		/* Page not found in guest SLB */
695 		vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
696 		kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
697 	} else if (!is_mmio &&
698 		   kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
699 		/* The guest's PTE is not mapped yet. Map on the host */
700 		kvmppc_mmu_map_page(vcpu, &pte);
701 		if (data)
702 			vcpu->stat.sp_storage++;
703 		else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
704 			(!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
705 			kvmppc_patch_dcbz(vcpu, &pte);
706 	} else {
707 		/* MMIO */
708 		vcpu->stat.mmio_exits++;
709 		vcpu->arch.paddr_accessed = pte.raddr;
710 		r = kvmppc_emulate_mmio(run, vcpu);
711 		if ( r == RESUME_HOST_NV )
712 			r = RESUME_HOST;
713 	}
714 
715 	return r;
716 }
717 
718 static inline int get_fpr_index(int i)
719 {
720 #ifdef CONFIG_VSX
721 	i *= 2;
722 #endif
723 	return i;
724 }
725 
726 /* Give up external provider (FPU, Altivec, VSX) */
727 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
728 {
729 	struct thread_struct *t = &current->thread;
730 	u64 *vcpu_fpr = vcpu->arch.fpr;
731 #ifdef CONFIG_VSX
732 	u64 *vcpu_vsx = vcpu->arch.vsr;
733 #endif
734 	u64 *thread_fpr = (u64*)t->fpr;
735 	int i;
736 
737 	if (!(vcpu->arch.guest_owned_ext & msr))
738 		return;
739 
740 #ifdef DEBUG_EXT
741 	printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
742 #endif
743 
744 	switch (msr) {
745 	case MSR_FP:
746 		giveup_fpu(current);
747 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
748 			vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
749 
750 		vcpu->arch.fpscr = t->fpscr.val;
751 		break;
752 	case MSR_VEC:
753 #ifdef CONFIG_ALTIVEC
754 		giveup_altivec(current);
755 		memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
756 		vcpu->arch.vscr = t->vscr;
757 #endif
758 		break;
759 	case MSR_VSX:
760 #ifdef CONFIG_VSX
761 		__giveup_vsx(current);
762 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
763 			vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
764 #endif
765 		break;
766 	default:
767 		BUG();
768 	}
769 
770 	vcpu->arch.guest_owned_ext &= ~msr;
771 	current->thread.regs->msr &= ~msr;
772 	kvmppc_recalc_shadow_msr(vcpu);
773 }
774 
775 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
776 {
777 	ulong srr0 = kvmppc_get_pc(vcpu);
778 	u32 last_inst = kvmppc_get_last_inst(vcpu);
779 	int ret;
780 
781 	ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
782 	if (ret == -ENOENT) {
783 		ulong msr = vcpu->arch.shared->msr;
784 
785 		msr = kvmppc_set_field(msr, 33, 33, 1);
786 		msr = kvmppc_set_field(msr, 34, 36, 0);
787 		vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
788 		kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
789 		return EMULATE_AGAIN;
790 	}
791 
792 	return EMULATE_DONE;
793 }
794 
795 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
796 {
797 
798 	/* Need to do paired single emulation? */
799 	if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
800 		return EMULATE_DONE;
801 
802 	/* Read out the instruction */
803 	if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
804 		/* Need to emulate */
805 		return EMULATE_FAIL;
806 
807 	return EMULATE_AGAIN;
808 }
809 
810 /* Handle external providers (FPU, Altivec, VSX) */
811 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
812 			     ulong msr)
813 {
814 	struct thread_struct *t = &current->thread;
815 	u64 *vcpu_fpr = vcpu->arch.fpr;
816 #ifdef CONFIG_VSX
817 	u64 *vcpu_vsx = vcpu->arch.vsr;
818 #endif
819 	u64 *thread_fpr = (u64*)t->fpr;
820 	int i;
821 
822 	/* When we have paired singles, we emulate in software */
823 	if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
824 		return RESUME_GUEST;
825 
826 	if (!(vcpu->arch.shared->msr & msr)) {
827 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
828 		return RESUME_GUEST;
829 	}
830 
831 	/* We already own the ext */
832 	if (vcpu->arch.guest_owned_ext & msr) {
833 		return RESUME_GUEST;
834 	}
835 
836 #ifdef DEBUG_EXT
837 	printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
838 #endif
839 
840 	current->thread.regs->msr |= msr;
841 
842 	switch (msr) {
843 	case MSR_FP:
844 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
845 			thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
846 
847 		t->fpscr.val = vcpu->arch.fpscr;
848 		t->fpexc_mode = 0;
849 		kvmppc_load_up_fpu();
850 		break;
851 	case MSR_VEC:
852 #ifdef CONFIG_ALTIVEC
853 		memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
854 		t->vscr = vcpu->arch.vscr;
855 		t->vrsave = -1;
856 		kvmppc_load_up_altivec();
857 #endif
858 		break;
859 	case MSR_VSX:
860 #ifdef CONFIG_VSX
861 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
862 			thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
863 		kvmppc_load_up_vsx();
864 #endif
865 		break;
866 	default:
867 		BUG();
868 	}
869 
870 	vcpu->arch.guest_owned_ext |= msr;
871 
872 	kvmppc_recalc_shadow_msr(vcpu);
873 
874 	return RESUME_GUEST;
875 }
876 
877 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
878                        unsigned int exit_nr)
879 {
880 	int r = RESUME_HOST;
881 
882 	vcpu->stat.sum_exits++;
883 
884 	run->exit_reason = KVM_EXIT_UNKNOWN;
885 	run->ready_for_interrupt_injection = 1;
886 
887 	trace_kvm_book3s_exit(exit_nr, vcpu);
888 	kvm_resched(vcpu);
889 	switch (exit_nr) {
890 	case BOOK3S_INTERRUPT_INST_STORAGE:
891 		vcpu->stat.pf_instruc++;
892 
893 #ifdef CONFIG_PPC_BOOK3S_32
894 		/* We set segments as unused segments when invalidating them. So
895 		 * treat the respective fault as segment fault. */
896 		if (to_svcpu(vcpu)->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT]
897 		    == SR_INVALID) {
898 			kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
899 			r = RESUME_GUEST;
900 			break;
901 		}
902 #endif
903 
904 		/* only care about PTEG not found errors, but leave NX alone */
905 		if (to_svcpu(vcpu)->shadow_srr1 & 0x40000000) {
906 			r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
907 			vcpu->stat.sp_instruc++;
908 		} else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
909 			  (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
910 			/*
911 			 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
912 			 *     so we can't use the NX bit inside the guest. Let's cross our fingers,
913 			 *     that no guest that needs the dcbz hack does NX.
914 			 */
915 			kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
916 			r = RESUME_GUEST;
917 		} else {
918 			vcpu->arch.shared->msr |=
919 				to_svcpu(vcpu)->shadow_srr1 & 0x58000000;
920 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
921 			r = RESUME_GUEST;
922 		}
923 		break;
924 	case BOOK3S_INTERRUPT_DATA_STORAGE:
925 	{
926 		ulong dar = kvmppc_get_fault_dar(vcpu);
927 		vcpu->stat.pf_storage++;
928 
929 #ifdef CONFIG_PPC_BOOK3S_32
930 		/* We set segments as unused segments when invalidating them. So
931 		 * treat the respective fault as segment fault. */
932 		if ((to_svcpu(vcpu)->sr[dar >> SID_SHIFT]) == SR_INVALID) {
933 			kvmppc_mmu_map_segment(vcpu, dar);
934 			r = RESUME_GUEST;
935 			break;
936 		}
937 #endif
938 
939 		/* The only case we need to handle is missing shadow PTEs */
940 		if (to_svcpu(vcpu)->fault_dsisr & DSISR_NOHPTE) {
941 			r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
942 		} else {
943 			vcpu->arch.shared->dar = dar;
944 			vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
945 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
946 			r = RESUME_GUEST;
947 		}
948 		break;
949 	}
950 	case BOOK3S_INTERRUPT_DATA_SEGMENT:
951 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
952 			vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
953 			kvmppc_book3s_queue_irqprio(vcpu,
954 				BOOK3S_INTERRUPT_DATA_SEGMENT);
955 		}
956 		r = RESUME_GUEST;
957 		break;
958 	case BOOK3S_INTERRUPT_INST_SEGMENT:
959 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
960 			kvmppc_book3s_queue_irqprio(vcpu,
961 				BOOK3S_INTERRUPT_INST_SEGMENT);
962 		}
963 		r = RESUME_GUEST;
964 		break;
965 	/* We're good on these - the host merely wanted to get our attention */
966 	case BOOK3S_INTERRUPT_DECREMENTER:
967 		vcpu->stat.dec_exits++;
968 		r = RESUME_GUEST;
969 		break;
970 	case BOOK3S_INTERRUPT_EXTERNAL:
971 		vcpu->stat.ext_intr_exits++;
972 		r = RESUME_GUEST;
973 		break;
974 	case BOOK3S_INTERRUPT_PERFMON:
975 		r = RESUME_GUEST;
976 		break;
977 	case BOOK3S_INTERRUPT_PROGRAM:
978 	{
979 		enum emulation_result er;
980 		ulong flags;
981 
982 program_interrupt:
983 		flags = to_svcpu(vcpu)->shadow_srr1 & 0x1f0000ull;
984 
985 		if (vcpu->arch.shared->msr & MSR_PR) {
986 #ifdef EXIT_DEBUG
987 			printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
988 #endif
989 			if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
990 			    (INS_DCBZ & 0xfffffff7)) {
991 				kvmppc_core_queue_program(vcpu, flags);
992 				r = RESUME_GUEST;
993 				break;
994 			}
995 		}
996 
997 		vcpu->stat.emulated_inst_exits++;
998 		er = kvmppc_emulate_instruction(run, vcpu);
999 		switch (er) {
1000 		case EMULATE_DONE:
1001 			r = RESUME_GUEST_NV;
1002 			break;
1003 		case EMULATE_AGAIN:
1004 			r = RESUME_GUEST;
1005 			break;
1006 		case EMULATE_FAIL:
1007 			printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
1008 			       __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
1009 			kvmppc_core_queue_program(vcpu, flags);
1010 			r = RESUME_GUEST;
1011 			break;
1012 		case EMULATE_DO_MMIO:
1013 			run->exit_reason = KVM_EXIT_MMIO;
1014 			r = RESUME_HOST_NV;
1015 			break;
1016 		default:
1017 			BUG();
1018 		}
1019 		break;
1020 	}
1021 	case BOOK3S_INTERRUPT_SYSCALL:
1022 		if (vcpu->arch.osi_enabled &&
1023 		    (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
1024 		    (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
1025 			/* MOL hypercalls */
1026 			u64 *gprs = run->osi.gprs;
1027 			int i;
1028 
1029 			run->exit_reason = KVM_EXIT_OSI;
1030 			for (i = 0; i < 32; i++)
1031 				gprs[i] = kvmppc_get_gpr(vcpu, i);
1032 			vcpu->arch.osi_needed = 1;
1033 			r = RESUME_HOST_NV;
1034 		} else if (!(vcpu->arch.shared->msr & MSR_PR) &&
1035 		    (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
1036 			/* KVM PV hypercalls */
1037 			kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
1038 			r = RESUME_GUEST;
1039 		} else {
1040 			/* Guest syscalls */
1041 			vcpu->stat.syscall_exits++;
1042 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1043 			r = RESUME_GUEST;
1044 		}
1045 		break;
1046 	case BOOK3S_INTERRUPT_FP_UNAVAIL:
1047 	case BOOK3S_INTERRUPT_ALTIVEC:
1048 	case BOOK3S_INTERRUPT_VSX:
1049 	{
1050 		int ext_msr = 0;
1051 
1052 		switch (exit_nr) {
1053 		case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
1054 		case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
1055 		case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
1056 		}
1057 
1058 		switch (kvmppc_check_ext(vcpu, exit_nr)) {
1059 		case EMULATE_DONE:
1060 			/* everything ok - let's enable the ext */
1061 			r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
1062 			break;
1063 		case EMULATE_FAIL:
1064 			/* we need to emulate this instruction */
1065 			goto program_interrupt;
1066 			break;
1067 		default:
1068 			/* nothing to worry about - go again */
1069 			break;
1070 		}
1071 		break;
1072 	}
1073 	case BOOK3S_INTERRUPT_ALIGNMENT:
1074 		if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
1075 			vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
1076 				kvmppc_get_last_inst(vcpu));
1077 			vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
1078 				kvmppc_get_last_inst(vcpu));
1079 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1080 		}
1081 		r = RESUME_GUEST;
1082 		break;
1083 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1084 	case BOOK3S_INTERRUPT_TRACE:
1085 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1086 		r = RESUME_GUEST;
1087 		break;
1088 	default:
1089 		/* Ugh - bork here! What did we get? */
1090 		printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1091 			exit_nr, kvmppc_get_pc(vcpu), to_svcpu(vcpu)->shadow_srr1);
1092 		r = RESUME_HOST;
1093 		BUG();
1094 		break;
1095 	}
1096 
1097 
1098 	if (!(r & RESUME_HOST)) {
1099 		/* To avoid clobbering exit_reason, only check for signals if
1100 		 * we aren't already exiting to userspace for some other
1101 		 * reason. */
1102 		if (signal_pending(current)) {
1103 #ifdef EXIT_DEBUG
1104 			printk(KERN_EMERG "KVM: Going back to host\n");
1105 #endif
1106 			vcpu->stat.signal_exits++;
1107 			run->exit_reason = KVM_EXIT_INTR;
1108 			r = -EINTR;
1109 		} else {
1110 			/* In case an interrupt came in that was triggered
1111 			 * from userspace (like DEC), we need to check what
1112 			 * to inject now! */
1113 			kvmppc_core_deliver_interrupts(vcpu);
1114 		}
1115 	}
1116 
1117 	trace_kvm_book3s_reenter(r, vcpu);
1118 
1119 	return r;
1120 }
1121 
1122 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1123 {
1124 	return 0;
1125 }
1126 
1127 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1128 {
1129 	int i;
1130 
1131 	regs->pc = kvmppc_get_pc(vcpu);
1132 	regs->cr = kvmppc_get_cr(vcpu);
1133 	regs->ctr = kvmppc_get_ctr(vcpu);
1134 	regs->lr = kvmppc_get_lr(vcpu);
1135 	regs->xer = kvmppc_get_xer(vcpu);
1136 	regs->msr = vcpu->arch.shared->msr;
1137 	regs->srr0 = vcpu->arch.shared->srr0;
1138 	regs->srr1 = vcpu->arch.shared->srr1;
1139 	regs->pid = vcpu->arch.pid;
1140 	regs->sprg0 = vcpu->arch.shared->sprg0;
1141 	regs->sprg1 = vcpu->arch.shared->sprg1;
1142 	regs->sprg2 = vcpu->arch.shared->sprg2;
1143 	regs->sprg3 = vcpu->arch.shared->sprg3;
1144 	regs->sprg5 = vcpu->arch.sprg4;
1145 	regs->sprg6 = vcpu->arch.sprg5;
1146 	regs->sprg7 = vcpu->arch.sprg6;
1147 
1148 	for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1149 		regs->gpr[i] = kvmppc_get_gpr(vcpu, i);
1150 
1151 	return 0;
1152 }
1153 
1154 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1155 {
1156 	int i;
1157 
1158 	kvmppc_set_pc(vcpu, regs->pc);
1159 	kvmppc_set_cr(vcpu, regs->cr);
1160 	kvmppc_set_ctr(vcpu, regs->ctr);
1161 	kvmppc_set_lr(vcpu, regs->lr);
1162 	kvmppc_set_xer(vcpu, regs->xer);
1163 	kvmppc_set_msr(vcpu, regs->msr);
1164 	vcpu->arch.shared->srr0 = regs->srr0;
1165 	vcpu->arch.shared->srr1 = regs->srr1;
1166 	vcpu->arch.shared->sprg0 = regs->sprg0;
1167 	vcpu->arch.shared->sprg1 = regs->sprg1;
1168 	vcpu->arch.shared->sprg2 = regs->sprg2;
1169 	vcpu->arch.shared->sprg3 = regs->sprg3;
1170 	vcpu->arch.sprg5 = regs->sprg4;
1171 	vcpu->arch.sprg6 = regs->sprg5;
1172 	vcpu->arch.sprg7 = regs->sprg6;
1173 
1174 	for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1175 		kvmppc_set_gpr(vcpu, i, regs->gpr[i]);
1176 
1177 	return 0;
1178 }
1179 
1180 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1181                                   struct kvm_sregs *sregs)
1182 {
1183 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1184 	int i;
1185 
1186 	sregs->pvr = vcpu->arch.pvr;
1187 
1188 	sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1189 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1190 		for (i = 0; i < 64; i++) {
1191 			sregs->u.s.ppc64.slb[i].slbe = vcpu3s->slb[i].orige | i;
1192 			sregs->u.s.ppc64.slb[i].slbv = vcpu3s->slb[i].origv;
1193 		}
1194 	} else {
1195 		for (i = 0; i < 16; i++)
1196 			sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
1197 
1198 		for (i = 0; i < 8; i++) {
1199 			sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1200 			sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1201 		}
1202 	}
1203 
1204 	return 0;
1205 }
1206 
1207 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1208                                   struct kvm_sregs *sregs)
1209 {
1210 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1211 	int i;
1212 
1213 	kvmppc_set_pvr(vcpu, sregs->pvr);
1214 
1215 	vcpu3s->sdr1 = sregs->u.s.sdr1;
1216 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1217 		for (i = 0; i < 64; i++) {
1218 			vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1219 						    sregs->u.s.ppc64.slb[i].slbe);
1220 		}
1221 	} else {
1222 		for (i = 0; i < 16; i++) {
1223 			vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1224 		}
1225 		for (i = 0; i < 8; i++) {
1226 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1227 				       (u32)sregs->u.s.ppc32.ibat[i]);
1228 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1229 				       (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1230 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1231 				       (u32)sregs->u.s.ppc32.dbat[i]);
1232 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1233 				       (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1234 		}
1235 	}
1236 
1237 	/* Flush the MMU after messing with the segments */
1238 	kvmppc_mmu_pte_flush(vcpu, 0, 0);
1239 
1240 	return 0;
1241 }
1242 
1243 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1244 {
1245 	return -ENOTSUPP;
1246 }
1247 
1248 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1249 {
1250 	return -ENOTSUPP;
1251 }
1252 
1253 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1254                                   struct kvm_translation *tr)
1255 {
1256 	return 0;
1257 }
1258 
1259 /*
1260  * Get (and clear) the dirty memory log for a memory slot.
1261  */
1262 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1263 				      struct kvm_dirty_log *log)
1264 {
1265 	struct kvm_memory_slot *memslot;
1266 	struct kvm_vcpu *vcpu;
1267 	ulong ga, ga_end;
1268 	int is_dirty = 0;
1269 	int r;
1270 	unsigned long n;
1271 
1272 	mutex_lock(&kvm->slots_lock);
1273 
1274 	r = kvm_get_dirty_log(kvm, log, &is_dirty);
1275 	if (r)
1276 		goto out;
1277 
1278 	/* If nothing is dirty, don't bother messing with page tables. */
1279 	if (is_dirty) {
1280 		memslot = &kvm->memslots->memslots[log->slot];
1281 
1282 		ga = memslot->base_gfn << PAGE_SHIFT;
1283 		ga_end = ga + (memslot->npages << PAGE_SHIFT);
1284 
1285 		kvm_for_each_vcpu(n, vcpu, kvm)
1286 			kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1287 
1288 		n = kvm_dirty_bitmap_bytes(memslot);
1289 		memset(memslot->dirty_bitmap, 0, n);
1290 	}
1291 
1292 	r = 0;
1293 out:
1294 	mutex_unlock(&kvm->slots_lock);
1295 	return r;
1296 }
1297 
1298 int kvmppc_core_check_processor_compat(void)
1299 {
1300 	return 0;
1301 }
1302 
1303 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
1304 {
1305 	struct kvmppc_vcpu_book3s *vcpu_book3s;
1306 	struct kvm_vcpu *vcpu;
1307 	int err = -ENOMEM;
1308 	unsigned long p;
1309 
1310 	vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1311 	if (!vcpu_book3s)
1312 		goto out;
1313 
1314 	vcpu_book3s->shadow_vcpu = (struct kvmppc_book3s_shadow_vcpu *)
1315 		kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
1316 	if (!vcpu_book3s->shadow_vcpu)
1317 		goto free_vcpu;
1318 
1319 	vcpu = &vcpu_book3s->vcpu;
1320 	err = kvm_vcpu_init(vcpu, kvm, id);
1321 	if (err)
1322 		goto free_shadow_vcpu;
1323 
1324 	p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1325 	/* the real shared page fills the last 4k of our page */
1326 	vcpu->arch.shared = (void*)(p + PAGE_SIZE - 4096);
1327 	if (!p)
1328 		goto uninit_vcpu;
1329 
1330 	vcpu->arch.host_retip = kvm_return_point;
1331 	vcpu->arch.host_msr = mfmsr();
1332 #ifdef CONFIG_PPC_BOOK3S_64
1333 	/* default to book3s_64 (970fx) */
1334 	vcpu->arch.pvr = 0x3C0301;
1335 #else
1336 	/* default to book3s_32 (750) */
1337 	vcpu->arch.pvr = 0x84202;
1338 #endif
1339 	kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
1340 	vcpu_book3s->slb_nr = 64;
1341 
1342 	/* remember where some real-mode handlers are */
1343 	vcpu->arch.trampoline_lowmem = kvmppc_trampoline_lowmem;
1344 	vcpu->arch.trampoline_enter = kvmppc_trampoline_enter;
1345 	vcpu->arch.highmem_handler = (ulong)kvmppc_handler_highmem;
1346 #ifdef CONFIG_PPC_BOOK3S_64
1347 	vcpu->arch.rmcall = *(ulong*)kvmppc_rmcall;
1348 #else
1349 	vcpu->arch.rmcall = (ulong)kvmppc_rmcall;
1350 #endif
1351 
1352 	vcpu->arch.shadow_msr = MSR_USER64;
1353 
1354 	err = kvmppc_mmu_init(vcpu);
1355 	if (err < 0)
1356 		goto uninit_vcpu;
1357 
1358 	return vcpu;
1359 
1360 uninit_vcpu:
1361 	kvm_vcpu_uninit(vcpu);
1362 free_shadow_vcpu:
1363 	kfree(vcpu_book3s->shadow_vcpu);
1364 free_vcpu:
1365 	vfree(vcpu_book3s);
1366 out:
1367 	return ERR_PTR(err);
1368 }
1369 
1370 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
1371 {
1372 	struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1373 
1374 	free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1375 	kvm_vcpu_uninit(vcpu);
1376 	kfree(vcpu_book3s->shadow_vcpu);
1377 	vfree(vcpu_book3s);
1378 }
1379 
1380 extern int __kvmppc_vcpu_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
1381 int __kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1382 {
1383 	int ret;
1384 	double fpr[32][TS_FPRWIDTH];
1385 	unsigned int fpscr;
1386 	int fpexc_mode;
1387 #ifdef CONFIG_ALTIVEC
1388 	vector128 vr[32];
1389 	vector128 vscr;
1390 	unsigned long uninitialized_var(vrsave);
1391 	int used_vr;
1392 #endif
1393 #ifdef CONFIG_VSX
1394 	int used_vsr;
1395 #endif
1396 	ulong ext_msr;
1397 
1398 	/* No need to go into the guest when all we do is going out */
1399 	if (signal_pending(current)) {
1400 		kvm_run->exit_reason = KVM_EXIT_INTR;
1401 		return -EINTR;
1402 	}
1403 
1404 	/* Save FPU state in stack */
1405 	if (current->thread.regs->msr & MSR_FP)
1406 		giveup_fpu(current);
1407 	memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
1408 	fpscr = current->thread.fpscr.val;
1409 	fpexc_mode = current->thread.fpexc_mode;
1410 
1411 #ifdef CONFIG_ALTIVEC
1412 	/* Save Altivec state in stack */
1413 	used_vr = current->thread.used_vr;
1414 	if (used_vr) {
1415 		if (current->thread.regs->msr & MSR_VEC)
1416 			giveup_altivec(current);
1417 		memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
1418 		vscr = current->thread.vscr;
1419 		vrsave = current->thread.vrsave;
1420 	}
1421 #endif
1422 
1423 #ifdef CONFIG_VSX
1424 	/* Save VSX state in stack */
1425 	used_vsr = current->thread.used_vsr;
1426 	if (used_vsr && (current->thread.regs->msr & MSR_VSX))
1427 			__giveup_vsx(current);
1428 #endif
1429 
1430 	/* Remember the MSR with disabled extensions */
1431 	ext_msr = current->thread.regs->msr;
1432 
1433 	/* XXX we get called with irq disabled - change that! */
1434 	local_irq_enable();
1435 
1436 	/* Preload FPU if it's enabled */
1437 	if (vcpu->arch.shared->msr & MSR_FP)
1438 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1439 
1440 	ret = __kvmppc_vcpu_entry(kvm_run, vcpu);
1441 
1442 	local_irq_disable();
1443 
1444 	current->thread.regs->msr = ext_msr;
1445 
1446 	/* Make sure we save the guest FPU/Altivec/VSX state */
1447 	kvmppc_giveup_ext(vcpu, MSR_FP);
1448 	kvmppc_giveup_ext(vcpu, MSR_VEC);
1449 	kvmppc_giveup_ext(vcpu, MSR_VSX);
1450 
1451 	/* Restore FPU state from stack */
1452 	memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
1453 	current->thread.fpscr.val = fpscr;
1454 	current->thread.fpexc_mode = fpexc_mode;
1455 
1456 #ifdef CONFIG_ALTIVEC
1457 	/* Restore Altivec state from stack */
1458 	if (used_vr && current->thread.used_vr) {
1459 		memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
1460 		current->thread.vscr = vscr;
1461 		current->thread.vrsave = vrsave;
1462 	}
1463 	current->thread.used_vr = used_vr;
1464 #endif
1465 
1466 #ifdef CONFIG_VSX
1467 	current->thread.used_vsr = used_vsr;
1468 #endif
1469 
1470 	return ret;
1471 }
1472 
1473 static int kvmppc_book3s_init(void)
1474 {
1475 	int r;
1476 
1477 	r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), 0,
1478 		     THIS_MODULE);
1479 
1480 	if (r)
1481 		return r;
1482 
1483 	r = kvmppc_mmu_hpte_sysinit();
1484 
1485 	return r;
1486 }
1487 
1488 static void kvmppc_book3s_exit(void)
1489 {
1490 	kvmppc_mmu_hpte_sysexit();
1491 	kvm_exit();
1492 }
1493 
1494 module_init(kvmppc_book3s_init);
1495 module_exit(kvmppc_book3s_exit);
1496