xref: /openbmc/linux/arch/powerpc/kvm/book3s_pr.c (revision 77d84ff8)
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 #include <linux/module.h>
44 
45 #include "book3s.h"
46 
47 #define CREATE_TRACE_POINTS
48 #include "trace_pr.h"
49 
50 /* #define EXIT_DEBUG */
51 /* #define DEBUG_EXT */
52 
53 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
54 			     ulong msr);
55 
56 /* Some compatibility defines */
57 #ifdef CONFIG_PPC_BOOK3S_32
58 #define MSR_USER32 MSR_USER
59 #define MSR_USER64 MSR_USER
60 #define HW_PAGE_SIZE PAGE_SIZE
61 #endif
62 
63 static void kvmppc_core_vcpu_load_pr(struct kvm_vcpu *vcpu, int cpu)
64 {
65 #ifdef CONFIG_PPC_BOOK3S_64
66 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
67 	memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb));
68 	svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max;
69 	svcpu_put(svcpu);
70 #endif
71 	vcpu->cpu = smp_processor_id();
72 #ifdef CONFIG_PPC_BOOK3S_32
73 	current->thread.kvm_shadow_vcpu = vcpu->arch.shadow_vcpu;
74 #endif
75 }
76 
77 static void kvmppc_core_vcpu_put_pr(struct kvm_vcpu *vcpu)
78 {
79 #ifdef CONFIG_PPC_BOOK3S_64
80 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
81 	memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
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 /* Copy data needed by real-mode code from vcpu to shadow vcpu */
91 void kvmppc_copy_to_svcpu(struct kvmppc_book3s_shadow_vcpu *svcpu,
92 			  struct kvm_vcpu *vcpu)
93 {
94 	svcpu->gpr[0] = vcpu->arch.gpr[0];
95 	svcpu->gpr[1] = vcpu->arch.gpr[1];
96 	svcpu->gpr[2] = vcpu->arch.gpr[2];
97 	svcpu->gpr[3] = vcpu->arch.gpr[3];
98 	svcpu->gpr[4] = vcpu->arch.gpr[4];
99 	svcpu->gpr[5] = vcpu->arch.gpr[5];
100 	svcpu->gpr[6] = vcpu->arch.gpr[6];
101 	svcpu->gpr[7] = vcpu->arch.gpr[7];
102 	svcpu->gpr[8] = vcpu->arch.gpr[8];
103 	svcpu->gpr[9] = vcpu->arch.gpr[9];
104 	svcpu->gpr[10] = vcpu->arch.gpr[10];
105 	svcpu->gpr[11] = vcpu->arch.gpr[11];
106 	svcpu->gpr[12] = vcpu->arch.gpr[12];
107 	svcpu->gpr[13] = vcpu->arch.gpr[13];
108 	svcpu->cr  = vcpu->arch.cr;
109 	svcpu->xer = vcpu->arch.xer;
110 	svcpu->ctr = vcpu->arch.ctr;
111 	svcpu->lr  = vcpu->arch.lr;
112 	svcpu->pc  = vcpu->arch.pc;
113 }
114 
115 /* Copy data touched by real-mode code from shadow vcpu back to vcpu */
116 void kvmppc_copy_from_svcpu(struct kvm_vcpu *vcpu,
117 			    struct kvmppc_book3s_shadow_vcpu *svcpu)
118 {
119 	vcpu->arch.gpr[0] = svcpu->gpr[0];
120 	vcpu->arch.gpr[1] = svcpu->gpr[1];
121 	vcpu->arch.gpr[2] = svcpu->gpr[2];
122 	vcpu->arch.gpr[3] = svcpu->gpr[3];
123 	vcpu->arch.gpr[4] = svcpu->gpr[4];
124 	vcpu->arch.gpr[5] = svcpu->gpr[5];
125 	vcpu->arch.gpr[6] = svcpu->gpr[6];
126 	vcpu->arch.gpr[7] = svcpu->gpr[7];
127 	vcpu->arch.gpr[8] = svcpu->gpr[8];
128 	vcpu->arch.gpr[9] = svcpu->gpr[9];
129 	vcpu->arch.gpr[10] = svcpu->gpr[10];
130 	vcpu->arch.gpr[11] = svcpu->gpr[11];
131 	vcpu->arch.gpr[12] = svcpu->gpr[12];
132 	vcpu->arch.gpr[13] = svcpu->gpr[13];
133 	vcpu->arch.cr  = svcpu->cr;
134 	vcpu->arch.xer = svcpu->xer;
135 	vcpu->arch.ctr = svcpu->ctr;
136 	vcpu->arch.lr  = svcpu->lr;
137 	vcpu->arch.pc  = svcpu->pc;
138 	vcpu->arch.shadow_srr1 = svcpu->shadow_srr1;
139 	vcpu->arch.fault_dar   = svcpu->fault_dar;
140 	vcpu->arch.fault_dsisr = svcpu->fault_dsisr;
141 	vcpu->arch.last_inst   = svcpu->last_inst;
142 }
143 
144 static int kvmppc_core_check_requests_pr(struct kvm_vcpu *vcpu)
145 {
146 	int r = 1; /* Indicate we want to get back into the guest */
147 
148 	/* We misuse TLB_FLUSH to indicate that we want to clear
149 	   all shadow cache entries */
150 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
151 		kvmppc_mmu_pte_flush(vcpu, 0, 0);
152 
153 	return r;
154 }
155 
156 /************* MMU Notifiers *************/
157 static void do_kvm_unmap_hva(struct kvm *kvm, unsigned long start,
158 			     unsigned long end)
159 {
160 	long i;
161 	struct kvm_vcpu *vcpu;
162 	struct kvm_memslots *slots;
163 	struct kvm_memory_slot *memslot;
164 
165 	slots = kvm_memslots(kvm);
166 	kvm_for_each_memslot(memslot, slots) {
167 		unsigned long hva_start, hva_end;
168 		gfn_t gfn, gfn_end;
169 
170 		hva_start = max(start, memslot->userspace_addr);
171 		hva_end = min(end, memslot->userspace_addr +
172 					(memslot->npages << PAGE_SHIFT));
173 		if (hva_start >= hva_end)
174 			continue;
175 		/*
176 		 * {gfn(page) | page intersects with [hva_start, hva_end)} =
177 		 * {gfn, gfn+1, ..., gfn_end-1}.
178 		 */
179 		gfn = hva_to_gfn_memslot(hva_start, memslot);
180 		gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
181 		kvm_for_each_vcpu(i, vcpu, kvm)
182 			kvmppc_mmu_pte_pflush(vcpu, gfn << PAGE_SHIFT,
183 					      gfn_end << PAGE_SHIFT);
184 	}
185 }
186 
187 static int kvm_unmap_hva_pr(struct kvm *kvm, unsigned long hva)
188 {
189 	trace_kvm_unmap_hva(hva);
190 
191 	do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
192 
193 	return 0;
194 }
195 
196 static int kvm_unmap_hva_range_pr(struct kvm *kvm, unsigned long start,
197 				  unsigned long end)
198 {
199 	do_kvm_unmap_hva(kvm, start, end);
200 
201 	return 0;
202 }
203 
204 static int kvm_age_hva_pr(struct kvm *kvm, unsigned long hva)
205 {
206 	/* XXX could be more clever ;) */
207 	return 0;
208 }
209 
210 static int kvm_test_age_hva_pr(struct kvm *kvm, unsigned long hva)
211 {
212 	/* XXX could be more clever ;) */
213 	return 0;
214 }
215 
216 static void kvm_set_spte_hva_pr(struct kvm *kvm, unsigned long hva, pte_t pte)
217 {
218 	/* The page will get remapped properly on its next fault */
219 	do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
220 }
221 
222 /*****************************************/
223 
224 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
225 {
226 	ulong smsr = vcpu->arch.shared->msr;
227 
228 	/* Guest MSR values */
229 	smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE;
230 	/* Process MSR values */
231 	smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
232 	/* External providers the guest reserved */
233 	smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
234 	/* 64-bit Process MSR values */
235 #ifdef CONFIG_PPC_BOOK3S_64
236 	smsr |= MSR_ISF | MSR_HV;
237 #endif
238 	vcpu->arch.shadow_msr = smsr;
239 }
240 
241 static void kvmppc_set_msr_pr(struct kvm_vcpu *vcpu, u64 msr)
242 {
243 	ulong old_msr = vcpu->arch.shared->msr;
244 
245 #ifdef EXIT_DEBUG
246 	printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
247 #endif
248 
249 	msr &= to_book3s(vcpu)->msr_mask;
250 	vcpu->arch.shared->msr = msr;
251 	kvmppc_recalc_shadow_msr(vcpu);
252 
253 	if (msr & MSR_POW) {
254 		if (!vcpu->arch.pending_exceptions) {
255 			kvm_vcpu_block(vcpu);
256 			clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
257 			vcpu->stat.halt_wakeup++;
258 
259 			/* Unset POW bit after we woke up */
260 			msr &= ~MSR_POW;
261 			vcpu->arch.shared->msr = msr;
262 		}
263 	}
264 
265 	if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
266 		   (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
267 		kvmppc_mmu_flush_segments(vcpu);
268 		kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
269 
270 		/* Preload magic page segment when in kernel mode */
271 		if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
272 			struct kvm_vcpu_arch *a = &vcpu->arch;
273 
274 			if (msr & MSR_DR)
275 				kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
276 			else
277 				kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
278 		}
279 	}
280 
281 	/*
282 	 * When switching from 32 to 64-bit, we may have a stale 32-bit
283 	 * magic page around, we need to flush it. Typically 32-bit magic
284 	 * page will be instanciated when calling into RTAS. Note: We
285 	 * assume that such transition only happens while in kernel mode,
286 	 * ie, we never transition from user 32-bit to kernel 64-bit with
287 	 * a 32-bit magic page around.
288 	 */
289 	if (vcpu->arch.magic_page_pa &&
290 	    !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
291 		/* going from RTAS to normal kernel code */
292 		kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
293 				     ~0xFFFUL);
294 	}
295 
296 	/* Preload FPU if it's enabled */
297 	if (vcpu->arch.shared->msr & MSR_FP)
298 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
299 }
300 
301 void kvmppc_set_pvr_pr(struct kvm_vcpu *vcpu, u32 pvr)
302 {
303 	u32 host_pvr;
304 
305 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
306 	vcpu->arch.pvr = pvr;
307 #ifdef CONFIG_PPC_BOOK3S_64
308 	if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
309 		kvmppc_mmu_book3s_64_init(vcpu);
310 		if (!to_book3s(vcpu)->hior_explicit)
311 			to_book3s(vcpu)->hior = 0xfff00000;
312 		to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
313 		vcpu->arch.cpu_type = KVM_CPU_3S_64;
314 	} else
315 #endif
316 	{
317 		kvmppc_mmu_book3s_32_init(vcpu);
318 		if (!to_book3s(vcpu)->hior_explicit)
319 			to_book3s(vcpu)->hior = 0;
320 		to_book3s(vcpu)->msr_mask = 0xffffffffULL;
321 		vcpu->arch.cpu_type = KVM_CPU_3S_32;
322 	}
323 
324 	kvmppc_sanity_check(vcpu);
325 
326 	/* If we are in hypervisor level on 970, we can tell the CPU to
327 	 * treat DCBZ as 32 bytes store */
328 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
329 	if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
330 	    !strcmp(cur_cpu_spec->platform, "ppc970"))
331 		vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
332 
333 	/* Cell performs badly if MSR_FEx are set. So let's hope nobody
334 	   really needs them in a VM on Cell and force disable them. */
335 	if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
336 		to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
337 
338 	/*
339 	 * If they're asking for POWER6 or later, set the flag
340 	 * indicating that we can do multiple large page sizes
341 	 * and 1TB segments.
342 	 * Also set the flag that indicates that tlbie has the large
343 	 * page bit in the RB operand instead of the instruction.
344 	 */
345 	switch (PVR_VER(pvr)) {
346 	case PVR_POWER6:
347 	case PVR_POWER7:
348 	case PVR_POWER7p:
349 	case PVR_POWER8:
350 		vcpu->arch.hflags |= BOOK3S_HFLAG_MULTI_PGSIZE |
351 			BOOK3S_HFLAG_NEW_TLBIE;
352 		break;
353 	}
354 
355 #ifdef CONFIG_PPC_BOOK3S_32
356 	/* 32 bit Book3S always has 32 byte dcbz */
357 	vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
358 #endif
359 
360 	/* On some CPUs we can execute paired single operations natively */
361 	asm ( "mfpvr %0" : "=r"(host_pvr));
362 	switch (host_pvr) {
363 	case 0x00080200:	/* lonestar 2.0 */
364 	case 0x00088202:	/* lonestar 2.2 */
365 	case 0x70000100:	/* gekko 1.0 */
366 	case 0x00080100:	/* gekko 2.0 */
367 	case 0x00083203:	/* gekko 2.3a */
368 	case 0x00083213:	/* gekko 2.3b */
369 	case 0x00083204:	/* gekko 2.4 */
370 	case 0x00083214:	/* gekko 2.4e (8SE) - retail HW2 */
371 	case 0x00087200:	/* broadway */
372 		vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
373 		/* Enable HID2.PSE - in case we need it later */
374 		mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
375 	}
376 }
377 
378 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
379  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
380  * emulate 32 bytes dcbz length.
381  *
382  * The Book3s_64 inventors also realized this case and implemented a special bit
383  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
384  *
385  * My approach here is to patch the dcbz instruction on executing pages.
386  */
387 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
388 {
389 	struct page *hpage;
390 	u64 hpage_offset;
391 	u32 *page;
392 	int i;
393 
394 	hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
395 	if (is_error_page(hpage))
396 		return;
397 
398 	hpage_offset = pte->raddr & ~PAGE_MASK;
399 	hpage_offset &= ~0xFFFULL;
400 	hpage_offset /= 4;
401 
402 	get_page(hpage);
403 	page = kmap_atomic(hpage);
404 
405 	/* patch dcbz into reserved instruction, so we trap */
406 	for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
407 		if ((page[i] & 0xff0007ff) == INS_DCBZ)
408 			page[i] &= 0xfffffff7;
409 
410 	kunmap_atomic(page);
411 	put_page(hpage);
412 }
413 
414 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
415 {
416 	ulong mp_pa = vcpu->arch.magic_page_pa;
417 
418 	if (!(vcpu->arch.shared->msr & MSR_SF))
419 		mp_pa = (uint32_t)mp_pa;
420 
421 	if (unlikely(mp_pa) &&
422 	    unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
423 		return 1;
424 	}
425 
426 	return kvm_is_visible_gfn(vcpu->kvm, gfn);
427 }
428 
429 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
430 			    ulong eaddr, int vec)
431 {
432 	bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
433 	bool iswrite = false;
434 	int r = RESUME_GUEST;
435 	int relocated;
436 	int page_found = 0;
437 	struct kvmppc_pte pte;
438 	bool is_mmio = false;
439 	bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
440 	bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
441 	u64 vsid;
442 
443 	relocated = data ? dr : ir;
444 	if (data && (vcpu->arch.fault_dsisr & DSISR_ISSTORE))
445 		iswrite = true;
446 
447 	/* Resolve real address if translation turned on */
448 	if (relocated) {
449 		page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data, iswrite);
450 	} else {
451 		pte.may_execute = true;
452 		pte.may_read = true;
453 		pte.may_write = true;
454 		pte.raddr = eaddr & KVM_PAM;
455 		pte.eaddr = eaddr;
456 		pte.vpage = eaddr >> 12;
457 		pte.page_size = MMU_PAGE_64K;
458 	}
459 
460 	switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
461 	case 0:
462 		pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
463 		break;
464 	case MSR_DR:
465 	case MSR_IR:
466 		vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
467 
468 		if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
469 			pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
470 		else
471 			pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
472 		pte.vpage |= vsid;
473 
474 		if (vsid == -1)
475 			page_found = -EINVAL;
476 		break;
477 	}
478 
479 	if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
480 	   (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
481 		/*
482 		 * If we do the dcbz hack, we have to NX on every execution,
483 		 * so we can patch the executing code. This renders our guest
484 		 * NX-less.
485 		 */
486 		pte.may_execute = !data;
487 	}
488 
489 	if (page_found == -ENOENT) {
490 		/* Page not found in guest PTE entries */
491 		vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
492 		vcpu->arch.shared->dsisr = vcpu->arch.fault_dsisr;
493 		vcpu->arch.shared->msr |=
494 			vcpu->arch.shadow_srr1 & 0x00000000f8000000ULL;
495 		kvmppc_book3s_queue_irqprio(vcpu, vec);
496 	} else if (page_found == -EPERM) {
497 		/* Storage protection */
498 		vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
499 		vcpu->arch.shared->dsisr = vcpu->arch.fault_dsisr & ~DSISR_NOHPTE;
500 		vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
501 		vcpu->arch.shared->msr |=
502 			vcpu->arch.shadow_srr1 & 0x00000000f8000000ULL;
503 		kvmppc_book3s_queue_irqprio(vcpu, vec);
504 	} else if (page_found == -EINVAL) {
505 		/* Page not found in guest SLB */
506 		vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
507 		kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
508 	} else if (!is_mmio &&
509 		   kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
510 		if (data && !(vcpu->arch.fault_dsisr & DSISR_NOHPTE)) {
511 			/*
512 			 * There is already a host HPTE there, presumably
513 			 * a read-only one for a page the guest thinks
514 			 * is writable, so get rid of it first.
515 			 */
516 			kvmppc_mmu_unmap_page(vcpu, &pte);
517 		}
518 		/* The guest's PTE is not mapped yet. Map on the host */
519 		kvmppc_mmu_map_page(vcpu, &pte, iswrite);
520 		if (data)
521 			vcpu->stat.sp_storage++;
522 		else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
523 			 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
524 			kvmppc_patch_dcbz(vcpu, &pte);
525 	} else {
526 		/* MMIO */
527 		vcpu->stat.mmio_exits++;
528 		vcpu->arch.paddr_accessed = pte.raddr;
529 		vcpu->arch.vaddr_accessed = pte.eaddr;
530 		r = kvmppc_emulate_mmio(run, vcpu);
531 		if ( r == RESUME_HOST_NV )
532 			r = RESUME_HOST;
533 	}
534 
535 	return r;
536 }
537 
538 static inline int get_fpr_index(int i)
539 {
540 	return i * TS_FPRWIDTH;
541 }
542 
543 /* Give up external provider (FPU, Altivec, VSX) */
544 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
545 {
546 	struct thread_struct *t = &current->thread;
547 	u64 *vcpu_fpr = vcpu->arch.fpr;
548 #ifdef CONFIG_VSX
549 	u64 *vcpu_vsx = vcpu->arch.vsr;
550 #endif
551 	u64 *thread_fpr = &t->fp_state.fpr[0][0];
552 	int i;
553 
554 	/*
555 	 * VSX instructions can access FP and vector registers, so if
556 	 * we are giving up VSX, make sure we give up FP and VMX as well.
557 	 */
558 	if (msr & MSR_VSX)
559 		msr |= MSR_FP | MSR_VEC;
560 
561 	msr &= vcpu->arch.guest_owned_ext;
562 	if (!msr)
563 		return;
564 
565 #ifdef DEBUG_EXT
566 	printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
567 #endif
568 
569 	if (msr & MSR_FP) {
570 		/*
571 		 * Note that on CPUs with VSX, giveup_fpu stores
572 		 * both the traditional FP registers and the added VSX
573 		 * registers into thread.fp_state.fpr[].
574 		 */
575 		if (current->thread.regs->msr & MSR_FP)
576 			giveup_fpu(current);
577 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
578 			vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
579 
580 		vcpu->arch.fpscr = t->fp_state.fpscr;
581 
582 #ifdef CONFIG_VSX
583 		if (cpu_has_feature(CPU_FTR_VSX))
584 			for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr) / 2; i++)
585 				vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
586 #endif
587 	}
588 
589 #ifdef CONFIG_ALTIVEC
590 	if (msr & MSR_VEC) {
591 		if (current->thread.regs->msr & MSR_VEC)
592 			giveup_altivec(current);
593 		memcpy(vcpu->arch.vr, t->vr_state.vr, sizeof(vcpu->arch.vr));
594 		vcpu->arch.vscr = t->vr_state.vscr;
595 	}
596 #endif
597 
598 	vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
599 	kvmppc_recalc_shadow_msr(vcpu);
600 }
601 
602 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
603 {
604 	ulong srr0 = kvmppc_get_pc(vcpu);
605 	u32 last_inst = kvmppc_get_last_inst(vcpu);
606 	int ret;
607 
608 	ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
609 	if (ret == -ENOENT) {
610 		ulong msr = vcpu->arch.shared->msr;
611 
612 		msr = kvmppc_set_field(msr, 33, 33, 1);
613 		msr = kvmppc_set_field(msr, 34, 36, 0);
614 		vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
615 		kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
616 		return EMULATE_AGAIN;
617 	}
618 
619 	return EMULATE_DONE;
620 }
621 
622 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
623 {
624 
625 	/* Need to do paired single emulation? */
626 	if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
627 		return EMULATE_DONE;
628 
629 	/* Read out the instruction */
630 	if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
631 		/* Need to emulate */
632 		return EMULATE_FAIL;
633 
634 	return EMULATE_AGAIN;
635 }
636 
637 /* Handle external providers (FPU, Altivec, VSX) */
638 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
639 			     ulong msr)
640 {
641 	struct thread_struct *t = &current->thread;
642 	u64 *vcpu_fpr = vcpu->arch.fpr;
643 #ifdef CONFIG_VSX
644 	u64 *vcpu_vsx = vcpu->arch.vsr;
645 #endif
646 	u64 *thread_fpr = &t->fp_state.fpr[0][0];
647 	int i;
648 
649 	/* When we have paired singles, we emulate in software */
650 	if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
651 		return RESUME_GUEST;
652 
653 	if (!(vcpu->arch.shared->msr & msr)) {
654 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
655 		return RESUME_GUEST;
656 	}
657 
658 	if (msr == MSR_VSX) {
659 		/* No VSX?  Give an illegal instruction interrupt */
660 #ifdef CONFIG_VSX
661 		if (!cpu_has_feature(CPU_FTR_VSX))
662 #endif
663 		{
664 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
665 			return RESUME_GUEST;
666 		}
667 
668 		/*
669 		 * We have to load up all the FP and VMX registers before
670 		 * we can let the guest use VSX instructions.
671 		 */
672 		msr = MSR_FP | MSR_VEC | MSR_VSX;
673 	}
674 
675 	/* See if we already own all the ext(s) needed */
676 	msr &= ~vcpu->arch.guest_owned_ext;
677 	if (!msr)
678 		return RESUME_GUEST;
679 
680 #ifdef DEBUG_EXT
681 	printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
682 #endif
683 
684 	if (msr & MSR_FP) {
685 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
686 			thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
687 #ifdef CONFIG_VSX
688 		for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr) / 2; i++)
689 			thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
690 #endif
691 		t->fp_state.fpscr = vcpu->arch.fpscr;
692 		t->fpexc_mode = 0;
693 		kvmppc_load_up_fpu();
694 	}
695 
696 	if (msr & MSR_VEC) {
697 #ifdef CONFIG_ALTIVEC
698 		memcpy(t->vr_state.vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
699 		t->vr_state.vscr = vcpu->arch.vscr;
700 		t->vrsave = -1;
701 		kvmppc_load_up_altivec();
702 #endif
703 	}
704 
705 	current->thread.regs->msr |= msr;
706 	vcpu->arch.guest_owned_ext |= msr;
707 	kvmppc_recalc_shadow_msr(vcpu);
708 
709 	return RESUME_GUEST;
710 }
711 
712 /*
713  * Kernel code using FP or VMX could have flushed guest state to
714  * the thread_struct; if so, get it back now.
715  */
716 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
717 {
718 	unsigned long lost_ext;
719 
720 	lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
721 	if (!lost_ext)
722 		return;
723 
724 	if (lost_ext & MSR_FP)
725 		kvmppc_load_up_fpu();
726 #ifdef CONFIG_ALTIVEC
727 	if (lost_ext & MSR_VEC)
728 		kvmppc_load_up_altivec();
729 #endif
730 	current->thread.regs->msr |= lost_ext;
731 }
732 
733 int kvmppc_handle_exit_pr(struct kvm_run *run, struct kvm_vcpu *vcpu,
734 			  unsigned int exit_nr)
735 {
736 	int r = RESUME_HOST;
737 	int s;
738 
739 	vcpu->stat.sum_exits++;
740 
741 	run->exit_reason = KVM_EXIT_UNKNOWN;
742 	run->ready_for_interrupt_injection = 1;
743 
744 	/* We get here with MSR.EE=1 */
745 
746 	trace_kvm_exit(exit_nr, vcpu);
747 	kvm_guest_exit();
748 
749 	switch (exit_nr) {
750 	case BOOK3S_INTERRUPT_INST_STORAGE:
751 	{
752 		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
753 		vcpu->stat.pf_instruc++;
754 
755 #ifdef CONFIG_PPC_BOOK3S_32
756 		/* We set segments as unused segments when invalidating them. So
757 		 * treat the respective fault as segment fault. */
758 		{
759 			struct kvmppc_book3s_shadow_vcpu *svcpu;
760 			u32 sr;
761 
762 			svcpu = svcpu_get(vcpu);
763 			sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
764 			svcpu_put(svcpu);
765 			if (sr == SR_INVALID) {
766 				kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
767 				r = RESUME_GUEST;
768 				break;
769 			}
770 		}
771 #endif
772 
773 		/* only care about PTEG not found errors, but leave NX alone */
774 		if (shadow_srr1 & 0x40000000) {
775 			int idx = srcu_read_lock(&vcpu->kvm->srcu);
776 			r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
777 			srcu_read_unlock(&vcpu->kvm->srcu, idx);
778 			vcpu->stat.sp_instruc++;
779 		} else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
780 			  (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
781 			/*
782 			 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
783 			 *     so we can't use the NX bit inside the guest. Let's cross our fingers,
784 			 *     that no guest that needs the dcbz hack does NX.
785 			 */
786 			kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
787 			r = RESUME_GUEST;
788 		} else {
789 			vcpu->arch.shared->msr |= shadow_srr1 & 0x58000000;
790 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
791 			r = RESUME_GUEST;
792 		}
793 		break;
794 	}
795 	case BOOK3S_INTERRUPT_DATA_STORAGE:
796 	{
797 		ulong dar = kvmppc_get_fault_dar(vcpu);
798 		u32 fault_dsisr = vcpu->arch.fault_dsisr;
799 		vcpu->stat.pf_storage++;
800 
801 #ifdef CONFIG_PPC_BOOK3S_32
802 		/* We set segments as unused segments when invalidating them. So
803 		 * treat the respective fault as segment fault. */
804 		{
805 			struct kvmppc_book3s_shadow_vcpu *svcpu;
806 			u32 sr;
807 
808 			svcpu = svcpu_get(vcpu);
809 			sr = svcpu->sr[dar >> SID_SHIFT];
810 			svcpu_put(svcpu);
811 			if (sr == SR_INVALID) {
812 				kvmppc_mmu_map_segment(vcpu, dar);
813 				r = RESUME_GUEST;
814 				break;
815 			}
816 		}
817 #endif
818 
819 		/*
820 		 * We need to handle missing shadow PTEs, and
821 		 * protection faults due to us mapping a page read-only
822 		 * when the guest thinks it is writable.
823 		 */
824 		if (fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT)) {
825 			int idx = srcu_read_lock(&vcpu->kvm->srcu);
826 			r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
827 			srcu_read_unlock(&vcpu->kvm->srcu, idx);
828 		} else {
829 			vcpu->arch.shared->dar = dar;
830 			vcpu->arch.shared->dsisr = fault_dsisr;
831 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
832 			r = RESUME_GUEST;
833 		}
834 		break;
835 	}
836 	case BOOK3S_INTERRUPT_DATA_SEGMENT:
837 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
838 			vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
839 			kvmppc_book3s_queue_irqprio(vcpu,
840 				BOOK3S_INTERRUPT_DATA_SEGMENT);
841 		}
842 		r = RESUME_GUEST;
843 		break;
844 	case BOOK3S_INTERRUPT_INST_SEGMENT:
845 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
846 			kvmppc_book3s_queue_irqprio(vcpu,
847 				BOOK3S_INTERRUPT_INST_SEGMENT);
848 		}
849 		r = RESUME_GUEST;
850 		break;
851 	/* We're good on these - the host merely wanted to get our attention */
852 	case BOOK3S_INTERRUPT_DECREMENTER:
853 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
854 		vcpu->stat.dec_exits++;
855 		r = RESUME_GUEST;
856 		break;
857 	case BOOK3S_INTERRUPT_EXTERNAL:
858 	case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
859 	case BOOK3S_INTERRUPT_EXTERNAL_HV:
860 		vcpu->stat.ext_intr_exits++;
861 		r = RESUME_GUEST;
862 		break;
863 	case BOOK3S_INTERRUPT_PERFMON:
864 		r = RESUME_GUEST;
865 		break;
866 	case BOOK3S_INTERRUPT_PROGRAM:
867 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
868 	{
869 		enum emulation_result er;
870 		ulong flags;
871 
872 program_interrupt:
873 		flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
874 
875 		if (vcpu->arch.shared->msr & MSR_PR) {
876 #ifdef EXIT_DEBUG
877 			printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
878 #endif
879 			if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
880 			    (INS_DCBZ & 0xfffffff7)) {
881 				kvmppc_core_queue_program(vcpu, flags);
882 				r = RESUME_GUEST;
883 				break;
884 			}
885 		}
886 
887 		vcpu->stat.emulated_inst_exits++;
888 		er = kvmppc_emulate_instruction(run, vcpu);
889 		switch (er) {
890 		case EMULATE_DONE:
891 			r = RESUME_GUEST_NV;
892 			break;
893 		case EMULATE_AGAIN:
894 			r = RESUME_GUEST;
895 			break;
896 		case EMULATE_FAIL:
897 			printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
898 			       __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
899 			kvmppc_core_queue_program(vcpu, flags);
900 			r = RESUME_GUEST;
901 			break;
902 		case EMULATE_DO_MMIO:
903 			run->exit_reason = KVM_EXIT_MMIO;
904 			r = RESUME_HOST_NV;
905 			break;
906 		case EMULATE_EXIT_USER:
907 			r = RESUME_HOST_NV;
908 			break;
909 		default:
910 			BUG();
911 		}
912 		break;
913 	}
914 	case BOOK3S_INTERRUPT_SYSCALL:
915 		if (vcpu->arch.papr_enabled &&
916 		    (kvmppc_get_last_sc(vcpu) == 0x44000022) &&
917 		    !(vcpu->arch.shared->msr & MSR_PR)) {
918 			/* SC 1 papr hypercalls */
919 			ulong cmd = kvmppc_get_gpr(vcpu, 3);
920 			int i;
921 
922 #ifdef CONFIG_PPC_BOOK3S_64
923 			if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
924 				r = RESUME_GUEST;
925 				break;
926 			}
927 #endif
928 
929 			run->papr_hcall.nr = cmd;
930 			for (i = 0; i < 9; ++i) {
931 				ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
932 				run->papr_hcall.args[i] = gpr;
933 			}
934 			run->exit_reason = KVM_EXIT_PAPR_HCALL;
935 			vcpu->arch.hcall_needed = 1;
936 			r = RESUME_HOST;
937 		} else if (vcpu->arch.osi_enabled &&
938 		    (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
939 		    (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
940 			/* MOL hypercalls */
941 			u64 *gprs = run->osi.gprs;
942 			int i;
943 
944 			run->exit_reason = KVM_EXIT_OSI;
945 			for (i = 0; i < 32; i++)
946 				gprs[i] = kvmppc_get_gpr(vcpu, i);
947 			vcpu->arch.osi_needed = 1;
948 			r = RESUME_HOST_NV;
949 		} else if (!(vcpu->arch.shared->msr & MSR_PR) &&
950 		    (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
951 			/* KVM PV hypercalls */
952 			kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
953 			r = RESUME_GUEST;
954 		} else {
955 			/* Guest syscalls */
956 			vcpu->stat.syscall_exits++;
957 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
958 			r = RESUME_GUEST;
959 		}
960 		break;
961 	case BOOK3S_INTERRUPT_FP_UNAVAIL:
962 	case BOOK3S_INTERRUPT_ALTIVEC:
963 	case BOOK3S_INTERRUPT_VSX:
964 	{
965 		int ext_msr = 0;
966 
967 		switch (exit_nr) {
968 		case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
969 		case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
970 		case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
971 		}
972 
973 		switch (kvmppc_check_ext(vcpu, exit_nr)) {
974 		case EMULATE_DONE:
975 			/* everything ok - let's enable the ext */
976 			r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
977 			break;
978 		case EMULATE_FAIL:
979 			/* we need to emulate this instruction */
980 			goto program_interrupt;
981 			break;
982 		default:
983 			/* nothing to worry about - go again */
984 			break;
985 		}
986 		break;
987 	}
988 	case BOOK3S_INTERRUPT_ALIGNMENT:
989 		if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
990 			vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
991 				kvmppc_get_last_inst(vcpu));
992 			vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
993 				kvmppc_get_last_inst(vcpu));
994 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
995 		}
996 		r = RESUME_GUEST;
997 		break;
998 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
999 	case BOOK3S_INTERRUPT_TRACE:
1000 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1001 		r = RESUME_GUEST;
1002 		break;
1003 	default:
1004 	{
1005 		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1006 		/* Ugh - bork here! What did we get? */
1007 		printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1008 			exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
1009 		r = RESUME_HOST;
1010 		BUG();
1011 		break;
1012 	}
1013 	}
1014 
1015 	if (!(r & RESUME_HOST)) {
1016 		/* To avoid clobbering exit_reason, only check for signals if
1017 		 * we aren't already exiting to userspace for some other
1018 		 * reason. */
1019 
1020 		/*
1021 		 * Interrupts could be timers for the guest which we have to
1022 		 * inject again, so let's postpone them until we're in the guest
1023 		 * and if we really did time things so badly, then we just exit
1024 		 * again due to a host external interrupt.
1025 		 */
1026 		local_irq_disable();
1027 		s = kvmppc_prepare_to_enter(vcpu);
1028 		if (s <= 0) {
1029 			local_irq_enable();
1030 			r = s;
1031 		} else {
1032 			kvmppc_fix_ee_before_entry();
1033 		}
1034 		kvmppc_handle_lost_ext(vcpu);
1035 	}
1036 
1037 	trace_kvm_book3s_reenter(r, vcpu);
1038 
1039 	return r;
1040 }
1041 
1042 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu *vcpu,
1043 					    struct kvm_sregs *sregs)
1044 {
1045 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1046 	int i;
1047 
1048 	sregs->pvr = vcpu->arch.pvr;
1049 
1050 	sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1051 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1052 		for (i = 0; i < 64; i++) {
1053 			sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
1054 			sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1055 		}
1056 	} else {
1057 		for (i = 0; i < 16; i++)
1058 			sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
1059 
1060 		for (i = 0; i < 8; i++) {
1061 			sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1062 			sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1063 		}
1064 	}
1065 
1066 	return 0;
1067 }
1068 
1069 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu *vcpu,
1070 					    struct kvm_sregs *sregs)
1071 {
1072 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1073 	int i;
1074 
1075 	kvmppc_set_pvr_pr(vcpu, sregs->pvr);
1076 
1077 	vcpu3s->sdr1 = sregs->u.s.sdr1;
1078 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1079 		for (i = 0; i < 64; i++) {
1080 			vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1081 						    sregs->u.s.ppc64.slb[i].slbe);
1082 		}
1083 	} else {
1084 		for (i = 0; i < 16; i++) {
1085 			vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1086 		}
1087 		for (i = 0; i < 8; i++) {
1088 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1089 				       (u32)sregs->u.s.ppc32.ibat[i]);
1090 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1091 				       (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1092 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1093 				       (u32)sregs->u.s.ppc32.dbat[i]);
1094 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1095 				       (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1096 		}
1097 	}
1098 
1099 	/* Flush the MMU after messing with the segments */
1100 	kvmppc_mmu_pte_flush(vcpu, 0, 0);
1101 
1102 	return 0;
1103 }
1104 
1105 static int kvmppc_get_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1106 				 union kvmppc_one_reg *val)
1107 {
1108 	int r = 0;
1109 
1110 	switch (id) {
1111 	case KVM_REG_PPC_HIOR:
1112 		*val = get_reg_val(id, to_book3s(vcpu)->hior);
1113 		break;
1114 #ifdef CONFIG_VSX
1115 	case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31: {
1116 		long int i = id - KVM_REG_PPC_VSR0;
1117 
1118 		if (!cpu_has_feature(CPU_FTR_VSX)) {
1119 			r = -ENXIO;
1120 			break;
1121 		}
1122 		val->vsxval[0] = vcpu->arch.fpr[i];
1123 		val->vsxval[1] = vcpu->arch.vsr[i];
1124 		break;
1125 	}
1126 #endif /* CONFIG_VSX */
1127 	default:
1128 		r = -EINVAL;
1129 		break;
1130 	}
1131 
1132 	return r;
1133 }
1134 
1135 static int kvmppc_set_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1136 				 union kvmppc_one_reg *val)
1137 {
1138 	int r = 0;
1139 
1140 	switch (id) {
1141 	case KVM_REG_PPC_HIOR:
1142 		to_book3s(vcpu)->hior = set_reg_val(id, *val);
1143 		to_book3s(vcpu)->hior_explicit = true;
1144 		break;
1145 #ifdef CONFIG_VSX
1146 	case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31: {
1147 		long int i = id - KVM_REG_PPC_VSR0;
1148 
1149 		if (!cpu_has_feature(CPU_FTR_VSX)) {
1150 			r = -ENXIO;
1151 			break;
1152 		}
1153 		vcpu->arch.fpr[i] = val->vsxval[0];
1154 		vcpu->arch.vsr[i] = val->vsxval[1];
1155 		break;
1156 	}
1157 #endif /* CONFIG_VSX */
1158 	default:
1159 		r = -EINVAL;
1160 		break;
1161 	}
1162 
1163 	return r;
1164 }
1165 
1166 static struct kvm_vcpu *kvmppc_core_vcpu_create_pr(struct kvm *kvm,
1167 						   unsigned int id)
1168 {
1169 	struct kvmppc_vcpu_book3s *vcpu_book3s;
1170 	struct kvm_vcpu *vcpu;
1171 	int err = -ENOMEM;
1172 	unsigned long p;
1173 
1174 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1175 	if (!vcpu)
1176 		goto out;
1177 
1178 	vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1179 	if (!vcpu_book3s)
1180 		goto free_vcpu;
1181 	vcpu->arch.book3s = vcpu_book3s;
1182 
1183 #ifdef CONFIG_KVM_BOOK3S_32
1184 	vcpu->arch.shadow_vcpu =
1185 		kzalloc(sizeof(*vcpu->arch.shadow_vcpu), GFP_KERNEL);
1186 	if (!vcpu->arch.shadow_vcpu)
1187 		goto free_vcpu3s;
1188 #endif
1189 
1190 	err = kvm_vcpu_init(vcpu, kvm, id);
1191 	if (err)
1192 		goto free_shadow_vcpu;
1193 
1194 	err = -ENOMEM;
1195 	p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1196 	if (!p)
1197 		goto uninit_vcpu;
1198 	/* the real shared page fills the last 4k of our page */
1199 	vcpu->arch.shared = (void *)(p + PAGE_SIZE - 4096);
1200 
1201 #ifdef CONFIG_PPC_BOOK3S_64
1202 	/*
1203 	 * Default to the same as the host if we're on sufficiently
1204 	 * recent machine that we have 1TB segments;
1205 	 * otherwise default to PPC970FX.
1206 	 */
1207 	vcpu->arch.pvr = 0x3C0301;
1208 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1209 		vcpu->arch.pvr = mfspr(SPRN_PVR);
1210 #else
1211 	/* default to book3s_32 (750) */
1212 	vcpu->arch.pvr = 0x84202;
1213 #endif
1214 	kvmppc_set_pvr_pr(vcpu, vcpu->arch.pvr);
1215 	vcpu->arch.slb_nr = 64;
1216 
1217 	vcpu->arch.shadow_msr = MSR_USER64;
1218 
1219 	err = kvmppc_mmu_init(vcpu);
1220 	if (err < 0)
1221 		goto uninit_vcpu;
1222 
1223 	return vcpu;
1224 
1225 uninit_vcpu:
1226 	kvm_vcpu_uninit(vcpu);
1227 free_shadow_vcpu:
1228 #ifdef CONFIG_KVM_BOOK3S_32
1229 	kfree(vcpu->arch.shadow_vcpu);
1230 free_vcpu3s:
1231 #endif
1232 	vfree(vcpu_book3s);
1233 free_vcpu:
1234 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1235 out:
1236 	return ERR_PTR(err);
1237 }
1238 
1239 static void kvmppc_core_vcpu_free_pr(struct kvm_vcpu *vcpu)
1240 {
1241 	struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1242 
1243 	free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1244 	kvm_vcpu_uninit(vcpu);
1245 #ifdef CONFIG_KVM_BOOK3S_32
1246 	kfree(vcpu->arch.shadow_vcpu);
1247 #endif
1248 	vfree(vcpu_book3s);
1249 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1250 }
1251 
1252 static int kvmppc_vcpu_run_pr(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1253 {
1254 	int ret;
1255 	struct thread_fp_state fp;
1256 	int fpexc_mode;
1257 #ifdef CONFIG_ALTIVEC
1258 	struct thread_vr_state vr;
1259 	unsigned long uninitialized_var(vrsave);
1260 	int used_vr;
1261 #endif
1262 #ifdef CONFIG_VSX
1263 	int used_vsr;
1264 #endif
1265 	ulong ext_msr;
1266 
1267 	/* Check if we can run the vcpu at all */
1268 	if (!vcpu->arch.sane) {
1269 		kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1270 		ret = -EINVAL;
1271 		goto out;
1272 	}
1273 
1274 	/*
1275 	 * Interrupts could be timers for the guest which we have to inject
1276 	 * again, so let's postpone them until we're in the guest and if we
1277 	 * really did time things so badly, then we just exit again due to
1278 	 * a host external interrupt.
1279 	 */
1280 	local_irq_disable();
1281 	ret = kvmppc_prepare_to_enter(vcpu);
1282 	if (ret <= 0) {
1283 		local_irq_enable();
1284 		goto out;
1285 	}
1286 
1287 	/* Save FPU state in stack */
1288 	if (current->thread.regs->msr & MSR_FP)
1289 		giveup_fpu(current);
1290 	fp = current->thread.fp_state;
1291 	fpexc_mode = current->thread.fpexc_mode;
1292 
1293 #ifdef CONFIG_ALTIVEC
1294 	/* Save Altivec state in stack */
1295 	used_vr = current->thread.used_vr;
1296 	if (used_vr) {
1297 		if (current->thread.regs->msr & MSR_VEC)
1298 			giveup_altivec(current);
1299 		vr = current->thread.vr_state;
1300 		vrsave = current->thread.vrsave;
1301 	}
1302 #endif
1303 
1304 #ifdef CONFIG_VSX
1305 	/* Save VSX state in stack */
1306 	used_vsr = current->thread.used_vsr;
1307 	if (used_vsr && (current->thread.regs->msr & MSR_VSX))
1308 		__giveup_vsx(current);
1309 #endif
1310 
1311 	/* Remember the MSR with disabled extensions */
1312 	ext_msr = current->thread.regs->msr;
1313 
1314 	/* Preload FPU if it's enabled */
1315 	if (vcpu->arch.shared->msr & MSR_FP)
1316 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1317 
1318 	kvmppc_fix_ee_before_entry();
1319 
1320 	ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1321 
1322 	/* No need for kvm_guest_exit. It's done in handle_exit.
1323 	   We also get here with interrupts enabled. */
1324 
1325 	/* Make sure we save the guest FPU/Altivec/VSX state */
1326 	kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1327 
1328 	current->thread.regs->msr = ext_msr;
1329 
1330 	/* Restore FPU/VSX state from stack */
1331 	current->thread.fp_state = fp;
1332 	current->thread.fpexc_mode = fpexc_mode;
1333 
1334 #ifdef CONFIG_ALTIVEC
1335 	/* Restore Altivec state from stack */
1336 	if (used_vr && current->thread.used_vr) {
1337 		current->thread.vr_state = vr;
1338 		current->thread.vrsave = vrsave;
1339 	}
1340 	current->thread.used_vr = used_vr;
1341 #endif
1342 
1343 #ifdef CONFIG_VSX
1344 	current->thread.used_vsr = used_vsr;
1345 #endif
1346 
1347 out:
1348 	vcpu->mode = OUTSIDE_GUEST_MODE;
1349 	return ret;
1350 }
1351 
1352 /*
1353  * Get (and clear) the dirty memory log for a memory slot.
1354  */
1355 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm *kvm,
1356 					 struct kvm_dirty_log *log)
1357 {
1358 	struct kvm_memory_slot *memslot;
1359 	struct kvm_vcpu *vcpu;
1360 	ulong ga, ga_end;
1361 	int is_dirty = 0;
1362 	int r;
1363 	unsigned long n;
1364 
1365 	mutex_lock(&kvm->slots_lock);
1366 
1367 	r = kvm_get_dirty_log(kvm, log, &is_dirty);
1368 	if (r)
1369 		goto out;
1370 
1371 	/* If nothing is dirty, don't bother messing with page tables. */
1372 	if (is_dirty) {
1373 		memslot = id_to_memslot(kvm->memslots, log->slot);
1374 
1375 		ga = memslot->base_gfn << PAGE_SHIFT;
1376 		ga_end = ga + (memslot->npages << PAGE_SHIFT);
1377 
1378 		kvm_for_each_vcpu(n, vcpu, kvm)
1379 			kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1380 
1381 		n = kvm_dirty_bitmap_bytes(memslot);
1382 		memset(memslot->dirty_bitmap, 0, n);
1383 	}
1384 
1385 	r = 0;
1386 out:
1387 	mutex_unlock(&kvm->slots_lock);
1388 	return r;
1389 }
1390 
1391 static void kvmppc_core_flush_memslot_pr(struct kvm *kvm,
1392 					 struct kvm_memory_slot *memslot)
1393 {
1394 	return;
1395 }
1396 
1397 static int kvmppc_core_prepare_memory_region_pr(struct kvm *kvm,
1398 					struct kvm_memory_slot *memslot,
1399 					struct kvm_userspace_memory_region *mem)
1400 {
1401 	return 0;
1402 }
1403 
1404 static void kvmppc_core_commit_memory_region_pr(struct kvm *kvm,
1405 				struct kvm_userspace_memory_region *mem,
1406 				const struct kvm_memory_slot *old)
1407 {
1408 	return;
1409 }
1410 
1411 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot *free,
1412 					struct kvm_memory_slot *dont)
1413 {
1414 	return;
1415 }
1416 
1417 static int kvmppc_core_create_memslot_pr(struct kvm_memory_slot *slot,
1418 					 unsigned long npages)
1419 {
1420 	return 0;
1421 }
1422 
1423 
1424 #ifdef CONFIG_PPC64
1425 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1426 					 struct kvm_ppc_smmu_info *info)
1427 {
1428 	long int i;
1429 	struct kvm_vcpu *vcpu;
1430 
1431 	info->flags = 0;
1432 
1433 	/* SLB is always 64 entries */
1434 	info->slb_size = 64;
1435 
1436 	/* Standard 4k base page size segment */
1437 	info->sps[0].page_shift = 12;
1438 	info->sps[0].slb_enc = 0;
1439 	info->sps[0].enc[0].page_shift = 12;
1440 	info->sps[0].enc[0].pte_enc = 0;
1441 
1442 	/*
1443 	 * 64k large page size.
1444 	 * We only want to put this in if the CPUs we're emulating
1445 	 * support it, but unfortunately we don't have a vcpu easily
1446 	 * to hand here to test.  Just pick the first vcpu, and if
1447 	 * that doesn't exist yet, report the minimum capability,
1448 	 * i.e., no 64k pages.
1449 	 * 1T segment support goes along with 64k pages.
1450 	 */
1451 	i = 1;
1452 	vcpu = kvm_get_vcpu(kvm, 0);
1453 	if (vcpu && (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
1454 		info->flags = KVM_PPC_1T_SEGMENTS;
1455 		info->sps[i].page_shift = 16;
1456 		info->sps[i].slb_enc = SLB_VSID_L | SLB_VSID_LP_01;
1457 		info->sps[i].enc[0].page_shift = 16;
1458 		info->sps[i].enc[0].pte_enc = 1;
1459 		++i;
1460 	}
1461 
1462 	/* Standard 16M large page size segment */
1463 	info->sps[i].page_shift = 24;
1464 	info->sps[i].slb_enc = SLB_VSID_L;
1465 	info->sps[i].enc[0].page_shift = 24;
1466 	info->sps[i].enc[0].pte_enc = 0;
1467 
1468 	return 0;
1469 }
1470 #else
1471 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1472 					 struct kvm_ppc_smmu_info *info)
1473 {
1474 	/* We should not get called */
1475 	BUG();
1476 }
1477 #endif /* CONFIG_PPC64 */
1478 
1479 static unsigned int kvm_global_user_count = 0;
1480 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1481 
1482 static int kvmppc_core_init_vm_pr(struct kvm *kvm)
1483 {
1484 	mutex_init(&kvm->arch.hpt_mutex);
1485 
1486 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1487 		spin_lock(&kvm_global_user_count_lock);
1488 		if (++kvm_global_user_count == 1)
1489 			pSeries_disable_reloc_on_exc();
1490 		spin_unlock(&kvm_global_user_count_lock);
1491 	}
1492 	return 0;
1493 }
1494 
1495 static void kvmppc_core_destroy_vm_pr(struct kvm *kvm)
1496 {
1497 #ifdef CONFIG_PPC64
1498 	WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1499 #endif
1500 
1501 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1502 		spin_lock(&kvm_global_user_count_lock);
1503 		BUG_ON(kvm_global_user_count == 0);
1504 		if (--kvm_global_user_count == 0)
1505 			pSeries_enable_reloc_on_exc();
1506 		spin_unlock(&kvm_global_user_count_lock);
1507 	}
1508 }
1509 
1510 static int kvmppc_core_check_processor_compat_pr(void)
1511 {
1512 	/* we are always compatible */
1513 	return 0;
1514 }
1515 
1516 static long kvm_arch_vm_ioctl_pr(struct file *filp,
1517 				 unsigned int ioctl, unsigned long arg)
1518 {
1519 	return -ENOTTY;
1520 }
1521 
1522 static struct kvmppc_ops kvm_ops_pr = {
1523 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_pr,
1524 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_pr,
1525 	.get_one_reg = kvmppc_get_one_reg_pr,
1526 	.set_one_reg = kvmppc_set_one_reg_pr,
1527 	.vcpu_load   = kvmppc_core_vcpu_load_pr,
1528 	.vcpu_put    = kvmppc_core_vcpu_put_pr,
1529 	.set_msr     = kvmppc_set_msr_pr,
1530 	.vcpu_run    = kvmppc_vcpu_run_pr,
1531 	.vcpu_create = kvmppc_core_vcpu_create_pr,
1532 	.vcpu_free   = kvmppc_core_vcpu_free_pr,
1533 	.check_requests = kvmppc_core_check_requests_pr,
1534 	.get_dirty_log = kvm_vm_ioctl_get_dirty_log_pr,
1535 	.flush_memslot = kvmppc_core_flush_memslot_pr,
1536 	.prepare_memory_region = kvmppc_core_prepare_memory_region_pr,
1537 	.commit_memory_region = kvmppc_core_commit_memory_region_pr,
1538 	.unmap_hva = kvm_unmap_hva_pr,
1539 	.unmap_hva_range = kvm_unmap_hva_range_pr,
1540 	.age_hva  = kvm_age_hva_pr,
1541 	.test_age_hva = kvm_test_age_hva_pr,
1542 	.set_spte_hva = kvm_set_spte_hva_pr,
1543 	.mmu_destroy  = kvmppc_mmu_destroy_pr,
1544 	.free_memslot = kvmppc_core_free_memslot_pr,
1545 	.create_memslot = kvmppc_core_create_memslot_pr,
1546 	.init_vm = kvmppc_core_init_vm_pr,
1547 	.destroy_vm = kvmppc_core_destroy_vm_pr,
1548 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_pr,
1549 	.emulate_op = kvmppc_core_emulate_op_pr,
1550 	.emulate_mtspr = kvmppc_core_emulate_mtspr_pr,
1551 	.emulate_mfspr = kvmppc_core_emulate_mfspr_pr,
1552 	.fast_vcpu_kick = kvm_vcpu_kick,
1553 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_pr,
1554 };
1555 
1556 
1557 int kvmppc_book3s_init_pr(void)
1558 {
1559 	int r;
1560 
1561 	r = kvmppc_core_check_processor_compat_pr();
1562 	if (r < 0)
1563 		return r;
1564 
1565 	kvm_ops_pr.owner = THIS_MODULE;
1566 	kvmppc_pr_ops = &kvm_ops_pr;
1567 
1568 	r = kvmppc_mmu_hpte_sysinit();
1569 	return r;
1570 }
1571 
1572 void kvmppc_book3s_exit_pr(void)
1573 {
1574 	kvmppc_pr_ops = NULL;
1575 	kvmppc_mmu_hpte_sysexit();
1576 }
1577 
1578 /*
1579  * We only support separate modules for book3s 64
1580  */
1581 #ifdef CONFIG_PPC_BOOK3S_64
1582 
1583 module_init(kvmppc_book3s_init_pr);
1584 module_exit(kvmppc_book3s_exit_pr);
1585 
1586 MODULE_LICENSE("GPL");
1587 #endif
1588