xref: /openbmc/linux/arch/powerpc/kvm/book3s_pr.c (revision 97da55fc)
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  *    Paul Mackerras <paulus@samba.org>
8  *
9  * Description:
10  * Functions relating to running KVM on Book 3S processors where
11  * we don't have access to hypervisor mode, and we run the guest
12  * in problem state (user mode).
13  *
14  * This file is derived from arch/powerpc/kvm/44x.c,
15  * by Hollis Blanchard <hollisb@us.ibm.com>.
16  *
17  * This program is free software; you can redistribute it and/or modify
18  * it under the terms of the GNU General Public License, version 2, as
19  * published by the Free Software Foundation.
20  */
21 
22 #include <linux/kvm_host.h>
23 #include <linux/export.h>
24 #include <linux/err.h>
25 #include <linux/slab.h>
26 
27 #include <asm/reg.h>
28 #include <asm/cputable.h>
29 #include <asm/cacheflush.h>
30 #include <asm/tlbflush.h>
31 #include <asm/uaccess.h>
32 #include <asm/io.h>
33 #include <asm/kvm_ppc.h>
34 #include <asm/kvm_book3s.h>
35 #include <asm/mmu_context.h>
36 #include <asm/switch_to.h>
37 #include <asm/firmware.h>
38 #include <asm/hvcall.h>
39 #include <linux/gfp.h>
40 #include <linux/sched.h>
41 #include <linux/vmalloc.h>
42 #include <linux/highmem.h>
43 
44 #include "trace.h"
45 
46 /* #define EXIT_DEBUG */
47 /* #define DEBUG_EXT */
48 
49 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
50 			     ulong msr);
51 
52 /* Some compatibility defines */
53 #ifdef CONFIG_PPC_BOOK3S_32
54 #define MSR_USER32 MSR_USER
55 #define MSR_USER64 MSR_USER
56 #define HW_PAGE_SIZE PAGE_SIZE
57 #endif
58 
59 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
60 {
61 #ifdef CONFIG_PPC_BOOK3S_64
62 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
63 	memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb));
64 	memcpy(&get_paca()->shadow_vcpu, to_book3s(vcpu)->shadow_vcpu,
65 	       sizeof(get_paca()->shadow_vcpu));
66 	svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max;
67 	svcpu_put(svcpu);
68 #endif
69 	vcpu->cpu = smp_processor_id();
70 #ifdef CONFIG_PPC_BOOK3S_32
71 	current->thread.kvm_shadow_vcpu = to_book3s(vcpu)->shadow_vcpu;
72 #endif
73 }
74 
75 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
76 {
77 #ifdef CONFIG_PPC_BOOK3S_64
78 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
79 	memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
80 	memcpy(to_book3s(vcpu)->shadow_vcpu, &get_paca()->shadow_vcpu,
81 	       sizeof(get_paca()->shadow_vcpu));
82 	to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max;
83 	svcpu_put(svcpu);
84 #endif
85 
86 	kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
87 	vcpu->cpu = -1;
88 }
89 
90 int kvmppc_core_check_requests(struct kvm_vcpu *vcpu)
91 {
92 	int r = 1; /* Indicate we want to get back into the guest */
93 
94 	/* We misuse TLB_FLUSH to indicate that we want to clear
95 	   all shadow cache entries */
96 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
97 		kvmppc_mmu_pte_flush(vcpu, 0, 0);
98 
99 	return r;
100 }
101 
102 /************* MMU Notifiers *************/
103 
104 int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
105 {
106 	trace_kvm_unmap_hva(hva);
107 
108 	/*
109 	 * Flush all shadow tlb entries everywhere. This is slow, but
110 	 * we are 100% sure that we catch the to be unmapped page
111 	 */
112 	kvm_flush_remote_tlbs(kvm);
113 
114 	return 0;
115 }
116 
117 int kvm_unmap_hva_range(struct kvm *kvm, unsigned long start, unsigned long end)
118 {
119 	/* kvm_unmap_hva flushes everything anyways */
120 	kvm_unmap_hva(kvm, start);
121 
122 	return 0;
123 }
124 
125 int kvm_age_hva(struct kvm *kvm, unsigned long hva)
126 {
127 	/* XXX could be more clever ;) */
128 	return 0;
129 }
130 
131 int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
132 {
133 	/* XXX could be more clever ;) */
134 	return 0;
135 }
136 
137 void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
138 {
139 	/* The page will get remapped properly on its next fault */
140 	kvm_unmap_hva(kvm, hva);
141 }
142 
143 /*****************************************/
144 
145 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
146 {
147 	ulong smsr = vcpu->arch.shared->msr;
148 
149 	/* Guest MSR values */
150 	smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE;
151 	/* Process MSR values */
152 	smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
153 	/* External providers the guest reserved */
154 	smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
155 	/* 64-bit Process MSR values */
156 #ifdef CONFIG_PPC_BOOK3S_64
157 	smsr |= MSR_ISF | MSR_HV;
158 #endif
159 	vcpu->arch.shadow_msr = smsr;
160 }
161 
162 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
163 {
164 	ulong old_msr = vcpu->arch.shared->msr;
165 
166 #ifdef EXIT_DEBUG
167 	printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
168 #endif
169 
170 	msr &= to_book3s(vcpu)->msr_mask;
171 	vcpu->arch.shared->msr = msr;
172 	kvmppc_recalc_shadow_msr(vcpu);
173 
174 	if (msr & MSR_POW) {
175 		if (!vcpu->arch.pending_exceptions) {
176 			kvm_vcpu_block(vcpu);
177 			clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
178 			vcpu->stat.halt_wakeup++;
179 
180 			/* Unset POW bit after we woke up */
181 			msr &= ~MSR_POW;
182 			vcpu->arch.shared->msr = msr;
183 		}
184 	}
185 
186 	if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
187 		   (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
188 		kvmppc_mmu_flush_segments(vcpu);
189 		kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
190 
191 		/* Preload magic page segment when in kernel mode */
192 		if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
193 			struct kvm_vcpu_arch *a = &vcpu->arch;
194 
195 			if (msr & MSR_DR)
196 				kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
197 			else
198 				kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
199 		}
200 	}
201 
202 	/*
203 	 * When switching from 32 to 64-bit, we may have a stale 32-bit
204 	 * magic page around, we need to flush it. Typically 32-bit magic
205 	 * page will be instanciated when calling into RTAS. Note: We
206 	 * assume that such transition only happens while in kernel mode,
207 	 * ie, we never transition from user 32-bit to kernel 64-bit with
208 	 * a 32-bit magic page around.
209 	 */
210 	if (vcpu->arch.magic_page_pa &&
211 	    !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
212 		/* going from RTAS to normal kernel code */
213 		kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
214 				     ~0xFFFUL);
215 	}
216 
217 	/* Preload FPU if it's enabled */
218 	if (vcpu->arch.shared->msr & MSR_FP)
219 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
220 }
221 
222 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
223 {
224 	u32 host_pvr;
225 
226 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
227 	vcpu->arch.pvr = pvr;
228 #ifdef CONFIG_PPC_BOOK3S_64
229 	if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
230 		kvmppc_mmu_book3s_64_init(vcpu);
231 		if (!to_book3s(vcpu)->hior_explicit)
232 			to_book3s(vcpu)->hior = 0xfff00000;
233 		to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
234 		vcpu->arch.cpu_type = KVM_CPU_3S_64;
235 	} else
236 #endif
237 	{
238 		kvmppc_mmu_book3s_32_init(vcpu);
239 		if (!to_book3s(vcpu)->hior_explicit)
240 			to_book3s(vcpu)->hior = 0;
241 		to_book3s(vcpu)->msr_mask = 0xffffffffULL;
242 		vcpu->arch.cpu_type = KVM_CPU_3S_32;
243 	}
244 
245 	kvmppc_sanity_check(vcpu);
246 
247 	/* If we are in hypervisor level on 970, we can tell the CPU to
248 	 * treat DCBZ as 32 bytes store */
249 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
250 	if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
251 	    !strcmp(cur_cpu_spec->platform, "ppc970"))
252 		vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
253 
254 	/* Cell performs badly if MSR_FEx are set. So let's hope nobody
255 	   really needs them in a VM on Cell and force disable them. */
256 	if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
257 		to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
258 
259 #ifdef CONFIG_PPC_BOOK3S_32
260 	/* 32 bit Book3S always has 32 byte dcbz */
261 	vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
262 #endif
263 
264 	/* On some CPUs we can execute paired single operations natively */
265 	asm ( "mfpvr %0" : "=r"(host_pvr));
266 	switch (host_pvr) {
267 	case 0x00080200:	/* lonestar 2.0 */
268 	case 0x00088202:	/* lonestar 2.2 */
269 	case 0x70000100:	/* gekko 1.0 */
270 	case 0x00080100:	/* gekko 2.0 */
271 	case 0x00083203:	/* gekko 2.3a */
272 	case 0x00083213:	/* gekko 2.3b */
273 	case 0x00083204:	/* gekko 2.4 */
274 	case 0x00083214:	/* gekko 2.4e (8SE) - retail HW2 */
275 	case 0x00087200:	/* broadway */
276 		vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
277 		/* Enable HID2.PSE - in case we need it later */
278 		mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
279 	}
280 }
281 
282 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
283  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
284  * emulate 32 bytes dcbz length.
285  *
286  * The Book3s_64 inventors also realized this case and implemented a special bit
287  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
288  *
289  * My approach here is to patch the dcbz instruction on executing pages.
290  */
291 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
292 {
293 	struct page *hpage;
294 	u64 hpage_offset;
295 	u32 *page;
296 	int i;
297 
298 	hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
299 	if (is_error_page(hpage))
300 		return;
301 
302 	hpage_offset = pte->raddr & ~PAGE_MASK;
303 	hpage_offset &= ~0xFFFULL;
304 	hpage_offset /= 4;
305 
306 	get_page(hpage);
307 	page = kmap_atomic(hpage);
308 
309 	/* patch dcbz into reserved instruction, so we trap */
310 	for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
311 		if ((page[i] & 0xff0007ff) == INS_DCBZ)
312 			page[i] &= 0xfffffff7;
313 
314 	kunmap_atomic(page);
315 	put_page(hpage);
316 }
317 
318 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
319 {
320 	ulong mp_pa = vcpu->arch.magic_page_pa;
321 
322 	if (!(vcpu->arch.shared->msr & MSR_SF))
323 		mp_pa = (uint32_t)mp_pa;
324 
325 	if (unlikely(mp_pa) &&
326 	    unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
327 		return 1;
328 	}
329 
330 	return kvm_is_visible_gfn(vcpu->kvm, gfn);
331 }
332 
333 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
334 			    ulong eaddr, int vec)
335 {
336 	bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
337 	int r = RESUME_GUEST;
338 	int relocated;
339 	int page_found = 0;
340 	struct kvmppc_pte pte;
341 	bool is_mmio = false;
342 	bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
343 	bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
344 	u64 vsid;
345 
346 	relocated = data ? dr : ir;
347 
348 	/* Resolve real address if translation turned on */
349 	if (relocated) {
350 		page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
351 	} else {
352 		pte.may_execute = true;
353 		pte.may_read = true;
354 		pte.may_write = true;
355 		pte.raddr = eaddr & KVM_PAM;
356 		pte.eaddr = eaddr;
357 		pte.vpage = eaddr >> 12;
358 	}
359 
360 	switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
361 	case 0:
362 		pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
363 		break;
364 	case MSR_DR:
365 	case MSR_IR:
366 		vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
367 
368 		if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
369 			pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
370 		else
371 			pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
372 		pte.vpage |= vsid;
373 
374 		if (vsid == -1)
375 			page_found = -EINVAL;
376 		break;
377 	}
378 
379 	if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
380 	   (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
381 		/*
382 		 * If we do the dcbz hack, we have to NX on every execution,
383 		 * so we can patch the executing code. This renders our guest
384 		 * NX-less.
385 		 */
386 		pte.may_execute = !data;
387 	}
388 
389 	if (page_found == -ENOENT) {
390 		/* Page not found in guest PTE entries */
391 		struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
392 		vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
393 		vcpu->arch.shared->dsisr = svcpu->fault_dsisr;
394 		vcpu->arch.shared->msr |=
395 			(svcpu->shadow_srr1 & 0x00000000f8000000ULL);
396 		svcpu_put(svcpu);
397 		kvmppc_book3s_queue_irqprio(vcpu, vec);
398 	} else if (page_found == -EPERM) {
399 		/* Storage protection */
400 		struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
401 		vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
402 		vcpu->arch.shared->dsisr = svcpu->fault_dsisr & ~DSISR_NOHPTE;
403 		vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
404 		vcpu->arch.shared->msr |=
405 			svcpu->shadow_srr1 & 0x00000000f8000000ULL;
406 		svcpu_put(svcpu);
407 		kvmppc_book3s_queue_irqprio(vcpu, vec);
408 	} else if (page_found == -EINVAL) {
409 		/* Page not found in guest SLB */
410 		vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
411 		kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
412 	} else if (!is_mmio &&
413 		   kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
414 		/* The guest's PTE is not mapped yet. Map on the host */
415 		kvmppc_mmu_map_page(vcpu, &pte);
416 		if (data)
417 			vcpu->stat.sp_storage++;
418 		else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
419 			(!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
420 			kvmppc_patch_dcbz(vcpu, &pte);
421 	} else {
422 		/* MMIO */
423 		vcpu->stat.mmio_exits++;
424 		vcpu->arch.paddr_accessed = pte.raddr;
425 		vcpu->arch.vaddr_accessed = pte.eaddr;
426 		r = kvmppc_emulate_mmio(run, vcpu);
427 		if ( r == RESUME_HOST_NV )
428 			r = RESUME_HOST;
429 	}
430 
431 	return r;
432 }
433 
434 static inline int get_fpr_index(int i)
435 {
436 	return i * TS_FPRWIDTH;
437 }
438 
439 /* Give up external provider (FPU, Altivec, VSX) */
440 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
441 {
442 	struct thread_struct *t = &current->thread;
443 	u64 *vcpu_fpr = vcpu->arch.fpr;
444 #ifdef CONFIG_VSX
445 	u64 *vcpu_vsx = vcpu->arch.vsr;
446 #endif
447 	u64 *thread_fpr = (u64*)t->fpr;
448 	int i;
449 
450 	/*
451 	 * VSX instructions can access FP and vector registers, so if
452 	 * we are giving up VSX, make sure we give up FP and VMX as well.
453 	 */
454 	if (msr & MSR_VSX)
455 		msr |= MSR_FP | MSR_VEC;
456 
457 	msr &= vcpu->arch.guest_owned_ext;
458 	if (!msr)
459 		return;
460 
461 #ifdef DEBUG_EXT
462 	printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
463 #endif
464 
465 	if (msr & MSR_FP) {
466 		/*
467 		 * Note that on CPUs with VSX, giveup_fpu stores
468 		 * both the traditional FP registers and the added VSX
469 		 * registers into thread.fpr[].
470 		 */
471 		giveup_fpu(current);
472 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
473 			vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
474 
475 		vcpu->arch.fpscr = t->fpscr.val;
476 
477 #ifdef CONFIG_VSX
478 		if (cpu_has_feature(CPU_FTR_VSX))
479 			for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr) / 2; i++)
480 				vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
481 #endif
482 	}
483 
484 #ifdef CONFIG_ALTIVEC
485 	if (msr & MSR_VEC) {
486 		giveup_altivec(current);
487 		memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
488 		vcpu->arch.vscr = t->vscr;
489 	}
490 #endif
491 
492 	vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
493 	kvmppc_recalc_shadow_msr(vcpu);
494 }
495 
496 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
497 {
498 	ulong srr0 = kvmppc_get_pc(vcpu);
499 	u32 last_inst = kvmppc_get_last_inst(vcpu);
500 	int ret;
501 
502 	ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
503 	if (ret == -ENOENT) {
504 		ulong msr = vcpu->arch.shared->msr;
505 
506 		msr = kvmppc_set_field(msr, 33, 33, 1);
507 		msr = kvmppc_set_field(msr, 34, 36, 0);
508 		vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
509 		kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
510 		return EMULATE_AGAIN;
511 	}
512 
513 	return EMULATE_DONE;
514 }
515 
516 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
517 {
518 
519 	/* Need to do paired single emulation? */
520 	if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
521 		return EMULATE_DONE;
522 
523 	/* Read out the instruction */
524 	if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
525 		/* Need to emulate */
526 		return EMULATE_FAIL;
527 
528 	return EMULATE_AGAIN;
529 }
530 
531 /* Handle external providers (FPU, Altivec, VSX) */
532 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
533 			     ulong msr)
534 {
535 	struct thread_struct *t = &current->thread;
536 	u64 *vcpu_fpr = vcpu->arch.fpr;
537 #ifdef CONFIG_VSX
538 	u64 *vcpu_vsx = vcpu->arch.vsr;
539 #endif
540 	u64 *thread_fpr = (u64*)t->fpr;
541 	int i;
542 
543 	/* When we have paired singles, we emulate in software */
544 	if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
545 		return RESUME_GUEST;
546 
547 	if (!(vcpu->arch.shared->msr & msr)) {
548 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
549 		return RESUME_GUEST;
550 	}
551 
552 	if (msr == MSR_VSX) {
553 		/* No VSX?  Give an illegal instruction interrupt */
554 #ifdef CONFIG_VSX
555 		if (!cpu_has_feature(CPU_FTR_VSX))
556 #endif
557 		{
558 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
559 			return RESUME_GUEST;
560 		}
561 
562 		/*
563 		 * We have to load up all the FP and VMX registers before
564 		 * we can let the guest use VSX instructions.
565 		 */
566 		msr = MSR_FP | MSR_VEC | MSR_VSX;
567 	}
568 
569 	/* See if we already own all the ext(s) needed */
570 	msr &= ~vcpu->arch.guest_owned_ext;
571 	if (!msr)
572 		return RESUME_GUEST;
573 
574 #ifdef DEBUG_EXT
575 	printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
576 #endif
577 
578 	current->thread.regs->msr |= msr;
579 
580 	if (msr & MSR_FP) {
581 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
582 			thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
583 #ifdef CONFIG_VSX
584 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr) / 2; i++)
585 			thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
586 #endif
587 		t->fpscr.val = vcpu->arch.fpscr;
588 		t->fpexc_mode = 0;
589 		kvmppc_load_up_fpu();
590 	}
591 
592 	if (msr & MSR_VEC) {
593 #ifdef CONFIG_ALTIVEC
594 		memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
595 		t->vscr = vcpu->arch.vscr;
596 		t->vrsave = -1;
597 		kvmppc_load_up_altivec();
598 #endif
599 	}
600 
601 	vcpu->arch.guest_owned_ext |= msr;
602 	kvmppc_recalc_shadow_msr(vcpu);
603 
604 	return RESUME_GUEST;
605 }
606 
607 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
608                        unsigned int exit_nr)
609 {
610 	int r = RESUME_HOST;
611 	int s;
612 
613 	vcpu->stat.sum_exits++;
614 
615 	run->exit_reason = KVM_EXIT_UNKNOWN;
616 	run->ready_for_interrupt_injection = 1;
617 
618 	/* We get here with MSR.EE=1 */
619 
620 	trace_kvm_exit(exit_nr, vcpu);
621 	kvm_guest_exit();
622 
623 	switch (exit_nr) {
624 	case BOOK3S_INTERRUPT_INST_STORAGE:
625 	{
626 		struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
627 		ulong shadow_srr1 = svcpu->shadow_srr1;
628 		vcpu->stat.pf_instruc++;
629 
630 #ifdef CONFIG_PPC_BOOK3S_32
631 		/* We set segments as unused segments when invalidating them. So
632 		 * treat the respective fault as segment fault. */
633 		if (svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT] == SR_INVALID) {
634 			kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
635 			r = RESUME_GUEST;
636 			svcpu_put(svcpu);
637 			break;
638 		}
639 #endif
640 		svcpu_put(svcpu);
641 
642 		/* only care about PTEG not found errors, but leave NX alone */
643 		if (shadow_srr1 & 0x40000000) {
644 			r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
645 			vcpu->stat.sp_instruc++;
646 		} else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
647 			  (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
648 			/*
649 			 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
650 			 *     so we can't use the NX bit inside the guest. Let's cross our fingers,
651 			 *     that no guest that needs the dcbz hack does NX.
652 			 */
653 			kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
654 			r = RESUME_GUEST;
655 		} else {
656 			vcpu->arch.shared->msr |= shadow_srr1 & 0x58000000;
657 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
658 			r = RESUME_GUEST;
659 		}
660 		break;
661 	}
662 	case BOOK3S_INTERRUPT_DATA_STORAGE:
663 	{
664 		ulong dar = kvmppc_get_fault_dar(vcpu);
665 		struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
666 		u32 fault_dsisr = svcpu->fault_dsisr;
667 		vcpu->stat.pf_storage++;
668 
669 #ifdef CONFIG_PPC_BOOK3S_32
670 		/* We set segments as unused segments when invalidating them. So
671 		 * treat the respective fault as segment fault. */
672 		if ((svcpu->sr[dar >> SID_SHIFT]) == SR_INVALID) {
673 			kvmppc_mmu_map_segment(vcpu, dar);
674 			r = RESUME_GUEST;
675 			svcpu_put(svcpu);
676 			break;
677 		}
678 #endif
679 		svcpu_put(svcpu);
680 
681 		/* The only case we need to handle is missing shadow PTEs */
682 		if (fault_dsisr & DSISR_NOHPTE) {
683 			r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
684 		} else {
685 			vcpu->arch.shared->dar = dar;
686 			vcpu->arch.shared->dsisr = fault_dsisr;
687 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
688 			r = RESUME_GUEST;
689 		}
690 		break;
691 	}
692 	case BOOK3S_INTERRUPT_DATA_SEGMENT:
693 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
694 			vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
695 			kvmppc_book3s_queue_irqprio(vcpu,
696 				BOOK3S_INTERRUPT_DATA_SEGMENT);
697 		}
698 		r = RESUME_GUEST;
699 		break;
700 	case BOOK3S_INTERRUPT_INST_SEGMENT:
701 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
702 			kvmppc_book3s_queue_irqprio(vcpu,
703 				BOOK3S_INTERRUPT_INST_SEGMENT);
704 		}
705 		r = RESUME_GUEST;
706 		break;
707 	/* We're good on these - the host merely wanted to get our attention */
708 	case BOOK3S_INTERRUPT_DECREMENTER:
709 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
710 		vcpu->stat.dec_exits++;
711 		r = RESUME_GUEST;
712 		break;
713 	case BOOK3S_INTERRUPT_EXTERNAL:
714 	case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
715 	case BOOK3S_INTERRUPT_EXTERNAL_HV:
716 		vcpu->stat.ext_intr_exits++;
717 		r = RESUME_GUEST;
718 		break;
719 	case BOOK3S_INTERRUPT_PERFMON:
720 		r = RESUME_GUEST;
721 		break;
722 	case BOOK3S_INTERRUPT_PROGRAM:
723 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
724 	{
725 		enum emulation_result er;
726 		struct kvmppc_book3s_shadow_vcpu *svcpu;
727 		ulong flags;
728 
729 program_interrupt:
730 		svcpu = svcpu_get(vcpu);
731 		flags = svcpu->shadow_srr1 & 0x1f0000ull;
732 		svcpu_put(svcpu);
733 
734 		if (vcpu->arch.shared->msr & MSR_PR) {
735 #ifdef EXIT_DEBUG
736 			printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
737 #endif
738 			if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
739 			    (INS_DCBZ & 0xfffffff7)) {
740 				kvmppc_core_queue_program(vcpu, flags);
741 				r = RESUME_GUEST;
742 				break;
743 			}
744 		}
745 
746 		vcpu->stat.emulated_inst_exits++;
747 		er = kvmppc_emulate_instruction(run, vcpu);
748 		switch (er) {
749 		case EMULATE_DONE:
750 			r = RESUME_GUEST_NV;
751 			break;
752 		case EMULATE_AGAIN:
753 			r = RESUME_GUEST;
754 			break;
755 		case EMULATE_FAIL:
756 			printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
757 			       __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
758 			kvmppc_core_queue_program(vcpu, flags);
759 			r = RESUME_GUEST;
760 			break;
761 		case EMULATE_DO_MMIO:
762 			run->exit_reason = KVM_EXIT_MMIO;
763 			r = RESUME_HOST_NV;
764 			break;
765 		case EMULATE_DO_PAPR:
766 			run->exit_reason = KVM_EXIT_PAPR_HCALL;
767 			vcpu->arch.hcall_needed = 1;
768 			r = RESUME_HOST_NV;
769 			break;
770 		default:
771 			BUG();
772 		}
773 		break;
774 	}
775 	case BOOK3S_INTERRUPT_SYSCALL:
776 		if (vcpu->arch.papr_enabled &&
777 		    (kvmppc_get_last_inst(vcpu) == 0x44000022) &&
778 		    !(vcpu->arch.shared->msr & MSR_PR)) {
779 			/* SC 1 papr hypercalls */
780 			ulong cmd = kvmppc_get_gpr(vcpu, 3);
781 			int i;
782 
783 #ifdef CONFIG_KVM_BOOK3S_64_PR
784 			if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
785 				r = RESUME_GUEST;
786 				break;
787 			}
788 #endif
789 
790 			run->papr_hcall.nr = cmd;
791 			for (i = 0; i < 9; ++i) {
792 				ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
793 				run->papr_hcall.args[i] = gpr;
794 			}
795 			run->exit_reason = KVM_EXIT_PAPR_HCALL;
796 			vcpu->arch.hcall_needed = 1;
797 			r = RESUME_HOST;
798 		} else if (vcpu->arch.osi_enabled &&
799 		    (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
800 		    (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
801 			/* MOL hypercalls */
802 			u64 *gprs = run->osi.gprs;
803 			int i;
804 
805 			run->exit_reason = KVM_EXIT_OSI;
806 			for (i = 0; i < 32; i++)
807 				gprs[i] = kvmppc_get_gpr(vcpu, i);
808 			vcpu->arch.osi_needed = 1;
809 			r = RESUME_HOST_NV;
810 		} else if (!(vcpu->arch.shared->msr & MSR_PR) &&
811 		    (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
812 			/* KVM PV hypercalls */
813 			kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
814 			r = RESUME_GUEST;
815 		} else {
816 			/* Guest syscalls */
817 			vcpu->stat.syscall_exits++;
818 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
819 			r = RESUME_GUEST;
820 		}
821 		break;
822 	case BOOK3S_INTERRUPT_FP_UNAVAIL:
823 	case BOOK3S_INTERRUPT_ALTIVEC:
824 	case BOOK3S_INTERRUPT_VSX:
825 	{
826 		int ext_msr = 0;
827 
828 		switch (exit_nr) {
829 		case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
830 		case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
831 		case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
832 		}
833 
834 		switch (kvmppc_check_ext(vcpu, exit_nr)) {
835 		case EMULATE_DONE:
836 			/* everything ok - let's enable the ext */
837 			r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
838 			break;
839 		case EMULATE_FAIL:
840 			/* we need to emulate this instruction */
841 			goto program_interrupt;
842 			break;
843 		default:
844 			/* nothing to worry about - go again */
845 			break;
846 		}
847 		break;
848 	}
849 	case BOOK3S_INTERRUPT_ALIGNMENT:
850 		if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
851 			vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
852 				kvmppc_get_last_inst(vcpu));
853 			vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
854 				kvmppc_get_last_inst(vcpu));
855 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
856 		}
857 		r = RESUME_GUEST;
858 		break;
859 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
860 	case BOOK3S_INTERRUPT_TRACE:
861 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
862 		r = RESUME_GUEST;
863 		break;
864 	default:
865 	{
866 		struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
867 		ulong shadow_srr1 = svcpu->shadow_srr1;
868 		svcpu_put(svcpu);
869 		/* Ugh - bork here! What did we get? */
870 		printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
871 			exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
872 		r = RESUME_HOST;
873 		BUG();
874 		break;
875 	}
876 	}
877 
878 	if (!(r & RESUME_HOST)) {
879 		/* To avoid clobbering exit_reason, only check for signals if
880 		 * we aren't already exiting to userspace for some other
881 		 * reason. */
882 
883 		/*
884 		 * Interrupts could be timers for the guest which we have to
885 		 * inject again, so let's postpone them until we're in the guest
886 		 * and if we really did time things so badly, then we just exit
887 		 * again due to a host external interrupt.
888 		 */
889 		local_irq_disable();
890 		s = kvmppc_prepare_to_enter(vcpu);
891 		if (s <= 0) {
892 			local_irq_enable();
893 			r = s;
894 		} else {
895 			kvmppc_lazy_ee_enable();
896 		}
897 	}
898 
899 	trace_kvm_book3s_reenter(r, vcpu);
900 
901 	return r;
902 }
903 
904 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
905                                   struct kvm_sregs *sregs)
906 {
907 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
908 	int i;
909 
910 	sregs->pvr = vcpu->arch.pvr;
911 
912 	sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
913 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
914 		for (i = 0; i < 64; i++) {
915 			sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
916 			sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
917 		}
918 	} else {
919 		for (i = 0; i < 16; i++)
920 			sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
921 
922 		for (i = 0; i < 8; i++) {
923 			sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
924 			sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
925 		}
926 	}
927 
928 	return 0;
929 }
930 
931 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
932                                   struct kvm_sregs *sregs)
933 {
934 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
935 	int i;
936 
937 	kvmppc_set_pvr(vcpu, sregs->pvr);
938 
939 	vcpu3s->sdr1 = sregs->u.s.sdr1;
940 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
941 		for (i = 0; i < 64; i++) {
942 			vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
943 						    sregs->u.s.ppc64.slb[i].slbe);
944 		}
945 	} else {
946 		for (i = 0; i < 16; i++) {
947 			vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
948 		}
949 		for (i = 0; i < 8; i++) {
950 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
951 				       (u32)sregs->u.s.ppc32.ibat[i]);
952 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
953 				       (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
954 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
955 				       (u32)sregs->u.s.ppc32.dbat[i]);
956 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
957 				       (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
958 		}
959 	}
960 
961 	/* Flush the MMU after messing with the segments */
962 	kvmppc_mmu_pte_flush(vcpu, 0, 0);
963 
964 	return 0;
965 }
966 
967 int kvmppc_get_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
968 {
969 	int r = 0;
970 
971 	switch (id) {
972 	case KVM_REG_PPC_HIOR:
973 		*val = get_reg_val(id, to_book3s(vcpu)->hior);
974 		break;
975 #ifdef CONFIG_VSX
976 	case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31: {
977 		long int i = id - KVM_REG_PPC_VSR0;
978 
979 		if (!cpu_has_feature(CPU_FTR_VSX)) {
980 			r = -ENXIO;
981 			break;
982 		}
983 		val->vsxval[0] = vcpu->arch.fpr[i];
984 		val->vsxval[1] = vcpu->arch.vsr[i];
985 		break;
986 	}
987 #endif /* CONFIG_VSX */
988 	default:
989 		r = -EINVAL;
990 		break;
991 	}
992 
993 	return r;
994 }
995 
996 int kvmppc_set_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
997 {
998 	int r = 0;
999 
1000 	switch (id) {
1001 	case KVM_REG_PPC_HIOR:
1002 		to_book3s(vcpu)->hior = set_reg_val(id, *val);
1003 		to_book3s(vcpu)->hior_explicit = true;
1004 		break;
1005 #ifdef CONFIG_VSX
1006 	case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31: {
1007 		long int i = id - KVM_REG_PPC_VSR0;
1008 
1009 		if (!cpu_has_feature(CPU_FTR_VSX)) {
1010 			r = -ENXIO;
1011 			break;
1012 		}
1013 		vcpu->arch.fpr[i] = val->vsxval[0];
1014 		vcpu->arch.vsr[i] = val->vsxval[1];
1015 		break;
1016 	}
1017 #endif /* CONFIG_VSX */
1018 	default:
1019 		r = -EINVAL;
1020 		break;
1021 	}
1022 
1023 	return r;
1024 }
1025 
1026 int kvmppc_core_check_processor_compat(void)
1027 {
1028 	return 0;
1029 }
1030 
1031 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
1032 {
1033 	struct kvmppc_vcpu_book3s *vcpu_book3s;
1034 	struct kvm_vcpu *vcpu;
1035 	int err = -ENOMEM;
1036 	unsigned long p;
1037 
1038 	vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1039 	if (!vcpu_book3s)
1040 		goto out;
1041 
1042 	vcpu_book3s->shadow_vcpu = (struct kvmppc_book3s_shadow_vcpu *)
1043 		kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
1044 	if (!vcpu_book3s->shadow_vcpu)
1045 		goto free_vcpu;
1046 
1047 	vcpu = &vcpu_book3s->vcpu;
1048 	err = kvm_vcpu_init(vcpu, kvm, id);
1049 	if (err)
1050 		goto free_shadow_vcpu;
1051 
1052 	p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1053 	/* the real shared page fills the last 4k of our page */
1054 	vcpu->arch.shared = (void*)(p + PAGE_SIZE - 4096);
1055 	if (!p)
1056 		goto uninit_vcpu;
1057 
1058 #ifdef CONFIG_PPC_BOOK3S_64
1059 	/* default to book3s_64 (970fx) */
1060 	vcpu->arch.pvr = 0x3C0301;
1061 #else
1062 	/* default to book3s_32 (750) */
1063 	vcpu->arch.pvr = 0x84202;
1064 #endif
1065 	kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
1066 	vcpu->arch.slb_nr = 64;
1067 
1068 	vcpu->arch.shadow_msr = MSR_USER64;
1069 
1070 	err = kvmppc_mmu_init(vcpu);
1071 	if (err < 0)
1072 		goto uninit_vcpu;
1073 
1074 	return vcpu;
1075 
1076 uninit_vcpu:
1077 	kvm_vcpu_uninit(vcpu);
1078 free_shadow_vcpu:
1079 	kfree(vcpu_book3s->shadow_vcpu);
1080 free_vcpu:
1081 	vfree(vcpu_book3s);
1082 out:
1083 	return ERR_PTR(err);
1084 }
1085 
1086 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
1087 {
1088 	struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1089 
1090 	free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1091 	kvm_vcpu_uninit(vcpu);
1092 	kfree(vcpu_book3s->shadow_vcpu);
1093 	vfree(vcpu_book3s);
1094 }
1095 
1096 int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1097 {
1098 	int ret;
1099 	double fpr[32][TS_FPRWIDTH];
1100 	unsigned int fpscr;
1101 	int fpexc_mode;
1102 #ifdef CONFIG_ALTIVEC
1103 	vector128 vr[32];
1104 	vector128 vscr;
1105 	unsigned long uninitialized_var(vrsave);
1106 	int used_vr;
1107 #endif
1108 #ifdef CONFIG_VSX
1109 	int used_vsr;
1110 #endif
1111 	ulong ext_msr;
1112 
1113 	/* Check if we can run the vcpu at all */
1114 	if (!vcpu->arch.sane) {
1115 		kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1116 		ret = -EINVAL;
1117 		goto out;
1118 	}
1119 
1120 	/*
1121 	 * Interrupts could be timers for the guest which we have to inject
1122 	 * again, so let's postpone them until we're in the guest and if we
1123 	 * really did time things so badly, then we just exit again due to
1124 	 * a host external interrupt.
1125 	 */
1126 	local_irq_disable();
1127 	ret = kvmppc_prepare_to_enter(vcpu);
1128 	if (ret <= 0) {
1129 		local_irq_enable();
1130 		goto out;
1131 	}
1132 
1133 	/* Save FPU state in stack */
1134 	if (current->thread.regs->msr & MSR_FP)
1135 		giveup_fpu(current);
1136 	memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
1137 	fpscr = current->thread.fpscr.val;
1138 	fpexc_mode = current->thread.fpexc_mode;
1139 
1140 #ifdef CONFIG_ALTIVEC
1141 	/* Save Altivec state in stack */
1142 	used_vr = current->thread.used_vr;
1143 	if (used_vr) {
1144 		if (current->thread.regs->msr & MSR_VEC)
1145 			giveup_altivec(current);
1146 		memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
1147 		vscr = current->thread.vscr;
1148 		vrsave = current->thread.vrsave;
1149 	}
1150 #endif
1151 
1152 #ifdef CONFIG_VSX
1153 	/* Save VSX state in stack */
1154 	used_vsr = current->thread.used_vsr;
1155 	if (used_vsr && (current->thread.regs->msr & MSR_VSX))
1156 		__giveup_vsx(current);
1157 #endif
1158 
1159 	/* Remember the MSR with disabled extensions */
1160 	ext_msr = current->thread.regs->msr;
1161 
1162 	/* Preload FPU if it's enabled */
1163 	if (vcpu->arch.shared->msr & MSR_FP)
1164 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1165 
1166 	kvmppc_lazy_ee_enable();
1167 
1168 	ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1169 
1170 	/* No need for kvm_guest_exit. It's done in handle_exit.
1171 	   We also get here with interrupts enabled. */
1172 
1173 	/* Make sure we save the guest FPU/Altivec/VSX state */
1174 	kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1175 
1176 	current->thread.regs->msr = ext_msr;
1177 
1178 	/* Restore FPU/VSX state from stack */
1179 	memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
1180 	current->thread.fpscr.val = fpscr;
1181 	current->thread.fpexc_mode = fpexc_mode;
1182 
1183 #ifdef CONFIG_ALTIVEC
1184 	/* Restore Altivec state from stack */
1185 	if (used_vr && current->thread.used_vr) {
1186 		memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
1187 		current->thread.vscr = vscr;
1188 		current->thread.vrsave = vrsave;
1189 	}
1190 	current->thread.used_vr = used_vr;
1191 #endif
1192 
1193 #ifdef CONFIG_VSX
1194 	current->thread.used_vsr = used_vsr;
1195 #endif
1196 
1197 out:
1198 	vcpu->mode = OUTSIDE_GUEST_MODE;
1199 	return ret;
1200 }
1201 
1202 /*
1203  * Get (and clear) the dirty memory log for a memory slot.
1204  */
1205 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1206 				      struct kvm_dirty_log *log)
1207 {
1208 	struct kvm_memory_slot *memslot;
1209 	struct kvm_vcpu *vcpu;
1210 	ulong ga, ga_end;
1211 	int is_dirty = 0;
1212 	int r;
1213 	unsigned long n;
1214 
1215 	mutex_lock(&kvm->slots_lock);
1216 
1217 	r = kvm_get_dirty_log(kvm, log, &is_dirty);
1218 	if (r)
1219 		goto out;
1220 
1221 	/* If nothing is dirty, don't bother messing with page tables. */
1222 	if (is_dirty) {
1223 		memslot = id_to_memslot(kvm->memslots, log->slot);
1224 
1225 		ga = memslot->base_gfn << PAGE_SHIFT;
1226 		ga_end = ga + (memslot->npages << PAGE_SHIFT);
1227 
1228 		kvm_for_each_vcpu(n, vcpu, kvm)
1229 			kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1230 
1231 		n = kvm_dirty_bitmap_bytes(memslot);
1232 		memset(memslot->dirty_bitmap, 0, n);
1233 	}
1234 
1235 	r = 0;
1236 out:
1237 	mutex_unlock(&kvm->slots_lock);
1238 	return r;
1239 }
1240 
1241 #ifdef CONFIG_PPC64
1242 int kvm_vm_ioctl_get_smmu_info(struct kvm *kvm, struct kvm_ppc_smmu_info *info)
1243 {
1244 	/* No flags */
1245 	info->flags = 0;
1246 
1247 	/* SLB is always 64 entries */
1248 	info->slb_size = 64;
1249 
1250 	/* Standard 4k base page size segment */
1251 	info->sps[0].page_shift = 12;
1252 	info->sps[0].slb_enc = 0;
1253 	info->sps[0].enc[0].page_shift = 12;
1254 	info->sps[0].enc[0].pte_enc = 0;
1255 
1256 	/* Standard 16M large page size segment */
1257 	info->sps[1].page_shift = 24;
1258 	info->sps[1].slb_enc = SLB_VSID_L;
1259 	info->sps[1].enc[0].page_shift = 24;
1260 	info->sps[1].enc[0].pte_enc = 0;
1261 
1262 	return 0;
1263 }
1264 #endif /* CONFIG_PPC64 */
1265 
1266 void kvmppc_core_free_memslot(struct kvm_memory_slot *free,
1267 			      struct kvm_memory_slot *dont)
1268 {
1269 }
1270 
1271 int kvmppc_core_create_memslot(struct kvm_memory_slot *slot,
1272 			       unsigned long npages)
1273 {
1274 	return 0;
1275 }
1276 
1277 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1278 				      struct kvm_memory_slot *memslot,
1279 				      struct kvm_userspace_memory_region *mem)
1280 {
1281 	return 0;
1282 }
1283 
1284 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1285 				struct kvm_userspace_memory_region *mem,
1286 				struct kvm_memory_slot old)
1287 {
1288 }
1289 
1290 void kvmppc_core_flush_memslot(struct kvm *kvm, struct kvm_memory_slot *memslot)
1291 {
1292 }
1293 
1294 static unsigned int kvm_global_user_count = 0;
1295 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1296 
1297 int kvmppc_core_init_vm(struct kvm *kvm)
1298 {
1299 #ifdef CONFIG_PPC64
1300 	INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
1301 #endif
1302 
1303 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1304 		spin_lock(&kvm_global_user_count_lock);
1305 		if (++kvm_global_user_count == 1)
1306 			pSeries_disable_reloc_on_exc();
1307 		spin_unlock(&kvm_global_user_count_lock);
1308 	}
1309 	return 0;
1310 }
1311 
1312 void kvmppc_core_destroy_vm(struct kvm *kvm)
1313 {
1314 #ifdef CONFIG_PPC64
1315 	WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1316 #endif
1317 
1318 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1319 		spin_lock(&kvm_global_user_count_lock);
1320 		BUG_ON(kvm_global_user_count == 0);
1321 		if (--kvm_global_user_count == 0)
1322 			pSeries_enable_reloc_on_exc();
1323 		spin_unlock(&kvm_global_user_count_lock);
1324 	}
1325 }
1326 
1327 static int kvmppc_book3s_init(void)
1328 {
1329 	int r;
1330 
1331 	r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), 0,
1332 		     THIS_MODULE);
1333 
1334 	if (r)
1335 		return r;
1336 
1337 	r = kvmppc_mmu_hpte_sysinit();
1338 
1339 	return r;
1340 }
1341 
1342 static void kvmppc_book3s_exit(void)
1343 {
1344 	kvmppc_mmu_hpte_sysexit();
1345 	kvm_exit();
1346 }
1347 
1348 module_init(kvmppc_book3s_init);
1349 module_exit(kvmppc_book3s_exit);
1350