xref: /openbmc/linux/arch/powerpc/kvm/book3s_pr.c (revision 37be287c)
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 #include <linux/miscdevice.h>
45 
46 #include "book3s.h"
47 
48 #define CREATE_TRACE_POINTS
49 #include "trace_pr.h"
50 
51 /* #define EXIT_DEBUG */
52 /* #define DEBUG_EXT */
53 
54 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
55 			     ulong msr);
56 
57 /* Some compatibility defines */
58 #ifdef CONFIG_PPC_BOOK3S_32
59 #define MSR_USER32 MSR_USER
60 #define MSR_USER64 MSR_USER
61 #define HW_PAGE_SIZE PAGE_SIZE
62 #endif
63 
64 static void kvmppc_core_vcpu_load_pr(struct kvm_vcpu *vcpu, int cpu)
65 {
66 #ifdef CONFIG_PPC_BOOK3S_64
67 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
68 	memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb));
69 	svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max;
70 	svcpu->in_use = 0;
71 	svcpu_put(svcpu);
72 #endif
73 	vcpu->cpu = smp_processor_id();
74 #ifdef CONFIG_PPC_BOOK3S_32
75 	current->thread.kvm_shadow_vcpu = vcpu->arch.shadow_vcpu;
76 #endif
77 }
78 
79 static void kvmppc_core_vcpu_put_pr(struct kvm_vcpu *vcpu)
80 {
81 #ifdef CONFIG_PPC_BOOK3S_64
82 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
83 	if (svcpu->in_use) {
84 		kvmppc_copy_from_svcpu(vcpu, svcpu);
85 	}
86 	memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
87 	to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max;
88 	svcpu_put(svcpu);
89 #endif
90 
91 	kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
92 	vcpu->cpu = -1;
93 }
94 
95 /* Copy data needed by real-mode code from vcpu to shadow vcpu */
96 void kvmppc_copy_to_svcpu(struct kvmppc_book3s_shadow_vcpu *svcpu,
97 			  struct kvm_vcpu *vcpu)
98 {
99 	svcpu->gpr[0] = vcpu->arch.gpr[0];
100 	svcpu->gpr[1] = vcpu->arch.gpr[1];
101 	svcpu->gpr[2] = vcpu->arch.gpr[2];
102 	svcpu->gpr[3] = vcpu->arch.gpr[3];
103 	svcpu->gpr[4] = vcpu->arch.gpr[4];
104 	svcpu->gpr[5] = vcpu->arch.gpr[5];
105 	svcpu->gpr[6] = vcpu->arch.gpr[6];
106 	svcpu->gpr[7] = vcpu->arch.gpr[7];
107 	svcpu->gpr[8] = vcpu->arch.gpr[8];
108 	svcpu->gpr[9] = vcpu->arch.gpr[9];
109 	svcpu->gpr[10] = vcpu->arch.gpr[10];
110 	svcpu->gpr[11] = vcpu->arch.gpr[11];
111 	svcpu->gpr[12] = vcpu->arch.gpr[12];
112 	svcpu->gpr[13] = vcpu->arch.gpr[13];
113 	svcpu->cr  = vcpu->arch.cr;
114 	svcpu->xer = vcpu->arch.xer;
115 	svcpu->ctr = vcpu->arch.ctr;
116 	svcpu->lr  = vcpu->arch.lr;
117 	svcpu->pc  = vcpu->arch.pc;
118 	svcpu->in_use = true;
119 }
120 
121 /* Copy data touched by real-mode code from shadow vcpu back to vcpu */
122 void kvmppc_copy_from_svcpu(struct kvm_vcpu *vcpu,
123 			    struct kvmppc_book3s_shadow_vcpu *svcpu)
124 {
125 	/*
126 	 * vcpu_put would just call us again because in_use hasn't
127 	 * been updated yet.
128 	 */
129 	preempt_disable();
130 
131 	/*
132 	 * Maybe we were already preempted and synced the svcpu from
133 	 * our preempt notifiers. Don't bother touching this svcpu then.
134 	 */
135 	if (!svcpu->in_use)
136 		goto out;
137 
138 	vcpu->arch.gpr[0] = svcpu->gpr[0];
139 	vcpu->arch.gpr[1] = svcpu->gpr[1];
140 	vcpu->arch.gpr[2] = svcpu->gpr[2];
141 	vcpu->arch.gpr[3] = svcpu->gpr[3];
142 	vcpu->arch.gpr[4] = svcpu->gpr[4];
143 	vcpu->arch.gpr[5] = svcpu->gpr[5];
144 	vcpu->arch.gpr[6] = svcpu->gpr[6];
145 	vcpu->arch.gpr[7] = svcpu->gpr[7];
146 	vcpu->arch.gpr[8] = svcpu->gpr[8];
147 	vcpu->arch.gpr[9] = svcpu->gpr[9];
148 	vcpu->arch.gpr[10] = svcpu->gpr[10];
149 	vcpu->arch.gpr[11] = svcpu->gpr[11];
150 	vcpu->arch.gpr[12] = svcpu->gpr[12];
151 	vcpu->arch.gpr[13] = svcpu->gpr[13];
152 	vcpu->arch.cr  = svcpu->cr;
153 	vcpu->arch.xer = svcpu->xer;
154 	vcpu->arch.ctr = svcpu->ctr;
155 	vcpu->arch.lr  = svcpu->lr;
156 	vcpu->arch.pc  = svcpu->pc;
157 	vcpu->arch.shadow_srr1 = svcpu->shadow_srr1;
158 	vcpu->arch.fault_dar   = svcpu->fault_dar;
159 	vcpu->arch.fault_dsisr = svcpu->fault_dsisr;
160 	vcpu->arch.last_inst   = svcpu->last_inst;
161 	svcpu->in_use = false;
162 
163 out:
164 	preempt_enable();
165 }
166 
167 static int kvmppc_core_check_requests_pr(struct kvm_vcpu *vcpu)
168 {
169 	int r = 1; /* Indicate we want to get back into the guest */
170 
171 	/* We misuse TLB_FLUSH to indicate that we want to clear
172 	   all shadow cache entries */
173 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
174 		kvmppc_mmu_pte_flush(vcpu, 0, 0);
175 
176 	return r;
177 }
178 
179 /************* MMU Notifiers *************/
180 static void do_kvm_unmap_hva(struct kvm *kvm, unsigned long start,
181 			     unsigned long end)
182 {
183 	long i;
184 	struct kvm_vcpu *vcpu;
185 	struct kvm_memslots *slots;
186 	struct kvm_memory_slot *memslot;
187 
188 	slots = kvm_memslots(kvm);
189 	kvm_for_each_memslot(memslot, slots) {
190 		unsigned long hva_start, hva_end;
191 		gfn_t gfn, gfn_end;
192 
193 		hva_start = max(start, memslot->userspace_addr);
194 		hva_end = min(end, memslot->userspace_addr +
195 					(memslot->npages << PAGE_SHIFT));
196 		if (hva_start >= hva_end)
197 			continue;
198 		/*
199 		 * {gfn(page) | page intersects with [hva_start, hva_end)} =
200 		 * {gfn, gfn+1, ..., gfn_end-1}.
201 		 */
202 		gfn = hva_to_gfn_memslot(hva_start, memslot);
203 		gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
204 		kvm_for_each_vcpu(i, vcpu, kvm)
205 			kvmppc_mmu_pte_pflush(vcpu, gfn << PAGE_SHIFT,
206 					      gfn_end << PAGE_SHIFT);
207 	}
208 }
209 
210 static int kvm_unmap_hva_pr(struct kvm *kvm, unsigned long hva)
211 {
212 	trace_kvm_unmap_hva(hva);
213 
214 	do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
215 
216 	return 0;
217 }
218 
219 static int kvm_unmap_hva_range_pr(struct kvm *kvm, unsigned long start,
220 				  unsigned long end)
221 {
222 	do_kvm_unmap_hva(kvm, start, end);
223 
224 	return 0;
225 }
226 
227 static int kvm_age_hva_pr(struct kvm *kvm, unsigned long hva)
228 {
229 	/* XXX could be more clever ;) */
230 	return 0;
231 }
232 
233 static int kvm_test_age_hva_pr(struct kvm *kvm, unsigned long hva)
234 {
235 	/* XXX could be more clever ;) */
236 	return 0;
237 }
238 
239 static void kvm_set_spte_hva_pr(struct kvm *kvm, unsigned long hva, pte_t pte)
240 {
241 	/* The page will get remapped properly on its next fault */
242 	do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
243 }
244 
245 /*****************************************/
246 
247 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
248 {
249 	ulong smsr = vcpu->arch.shared->msr;
250 
251 	/* Guest MSR values */
252 	smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE;
253 	/* Process MSR values */
254 	smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
255 	/* External providers the guest reserved */
256 	smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
257 	/* 64-bit Process MSR values */
258 #ifdef CONFIG_PPC_BOOK3S_64
259 	smsr |= MSR_ISF | MSR_HV;
260 #endif
261 	vcpu->arch.shadow_msr = smsr;
262 }
263 
264 static void kvmppc_set_msr_pr(struct kvm_vcpu *vcpu, u64 msr)
265 {
266 	ulong old_msr = vcpu->arch.shared->msr;
267 
268 #ifdef EXIT_DEBUG
269 	printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
270 #endif
271 
272 	msr &= to_book3s(vcpu)->msr_mask;
273 	vcpu->arch.shared->msr = msr;
274 	kvmppc_recalc_shadow_msr(vcpu);
275 
276 	if (msr & MSR_POW) {
277 		if (!vcpu->arch.pending_exceptions) {
278 			kvm_vcpu_block(vcpu);
279 			clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
280 			vcpu->stat.halt_wakeup++;
281 
282 			/* Unset POW bit after we woke up */
283 			msr &= ~MSR_POW;
284 			vcpu->arch.shared->msr = msr;
285 		}
286 	}
287 
288 	if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
289 		   (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
290 		kvmppc_mmu_flush_segments(vcpu);
291 		kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
292 
293 		/* Preload magic page segment when in kernel mode */
294 		if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
295 			struct kvm_vcpu_arch *a = &vcpu->arch;
296 
297 			if (msr & MSR_DR)
298 				kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
299 			else
300 				kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
301 		}
302 	}
303 
304 	/*
305 	 * When switching from 32 to 64-bit, we may have a stale 32-bit
306 	 * magic page around, we need to flush it. Typically 32-bit magic
307 	 * page will be instanciated when calling into RTAS. Note: We
308 	 * assume that such transition only happens while in kernel mode,
309 	 * ie, we never transition from user 32-bit to kernel 64-bit with
310 	 * a 32-bit magic page around.
311 	 */
312 	if (vcpu->arch.magic_page_pa &&
313 	    !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
314 		/* going from RTAS to normal kernel code */
315 		kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
316 				     ~0xFFFUL);
317 	}
318 
319 	/* Preload FPU if it's enabled */
320 	if (vcpu->arch.shared->msr & MSR_FP)
321 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
322 }
323 
324 void kvmppc_set_pvr_pr(struct kvm_vcpu *vcpu, u32 pvr)
325 {
326 	u32 host_pvr;
327 
328 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
329 	vcpu->arch.pvr = pvr;
330 #ifdef CONFIG_PPC_BOOK3S_64
331 	if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
332 		kvmppc_mmu_book3s_64_init(vcpu);
333 		if (!to_book3s(vcpu)->hior_explicit)
334 			to_book3s(vcpu)->hior = 0xfff00000;
335 		to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
336 		vcpu->arch.cpu_type = KVM_CPU_3S_64;
337 	} else
338 #endif
339 	{
340 		kvmppc_mmu_book3s_32_init(vcpu);
341 		if (!to_book3s(vcpu)->hior_explicit)
342 			to_book3s(vcpu)->hior = 0;
343 		to_book3s(vcpu)->msr_mask = 0xffffffffULL;
344 		vcpu->arch.cpu_type = KVM_CPU_3S_32;
345 	}
346 
347 	kvmppc_sanity_check(vcpu);
348 
349 	/* If we are in hypervisor level on 970, we can tell the CPU to
350 	 * treat DCBZ as 32 bytes store */
351 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
352 	if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
353 	    !strcmp(cur_cpu_spec->platform, "ppc970"))
354 		vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
355 
356 	/* Cell performs badly if MSR_FEx are set. So let's hope nobody
357 	   really needs them in a VM on Cell and force disable them. */
358 	if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
359 		to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
360 
361 	/*
362 	 * If they're asking for POWER6 or later, set the flag
363 	 * indicating that we can do multiple large page sizes
364 	 * and 1TB segments.
365 	 * Also set the flag that indicates that tlbie has the large
366 	 * page bit in the RB operand instead of the instruction.
367 	 */
368 	switch (PVR_VER(pvr)) {
369 	case PVR_POWER6:
370 	case PVR_POWER7:
371 	case PVR_POWER7p:
372 	case PVR_POWER8:
373 		vcpu->arch.hflags |= BOOK3S_HFLAG_MULTI_PGSIZE |
374 			BOOK3S_HFLAG_NEW_TLBIE;
375 		break;
376 	}
377 
378 #ifdef CONFIG_PPC_BOOK3S_32
379 	/* 32 bit Book3S always has 32 byte dcbz */
380 	vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
381 #endif
382 
383 	/* On some CPUs we can execute paired single operations natively */
384 	asm ( "mfpvr %0" : "=r"(host_pvr));
385 	switch (host_pvr) {
386 	case 0x00080200:	/* lonestar 2.0 */
387 	case 0x00088202:	/* lonestar 2.2 */
388 	case 0x70000100:	/* gekko 1.0 */
389 	case 0x00080100:	/* gekko 2.0 */
390 	case 0x00083203:	/* gekko 2.3a */
391 	case 0x00083213:	/* gekko 2.3b */
392 	case 0x00083204:	/* gekko 2.4 */
393 	case 0x00083214:	/* gekko 2.4e (8SE) - retail HW2 */
394 	case 0x00087200:	/* broadway */
395 		vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
396 		/* Enable HID2.PSE - in case we need it later */
397 		mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
398 	}
399 }
400 
401 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
402  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
403  * emulate 32 bytes dcbz length.
404  *
405  * The Book3s_64 inventors also realized this case and implemented a special bit
406  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
407  *
408  * My approach here is to patch the dcbz instruction on executing pages.
409  */
410 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
411 {
412 	struct page *hpage;
413 	u64 hpage_offset;
414 	u32 *page;
415 	int i;
416 
417 	hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
418 	if (is_error_page(hpage))
419 		return;
420 
421 	hpage_offset = pte->raddr & ~PAGE_MASK;
422 	hpage_offset &= ~0xFFFULL;
423 	hpage_offset /= 4;
424 
425 	get_page(hpage);
426 	page = kmap_atomic(hpage);
427 
428 	/* patch dcbz into reserved instruction, so we trap */
429 	for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
430 		if ((page[i] & 0xff0007ff) == INS_DCBZ)
431 			page[i] &= 0xfffffff7;
432 
433 	kunmap_atomic(page);
434 	put_page(hpage);
435 }
436 
437 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
438 {
439 	ulong mp_pa = vcpu->arch.magic_page_pa;
440 
441 	if (!(vcpu->arch.shared->msr & MSR_SF))
442 		mp_pa = (uint32_t)mp_pa;
443 
444 	if (unlikely(mp_pa) &&
445 	    unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
446 		return 1;
447 	}
448 
449 	return kvm_is_visible_gfn(vcpu->kvm, gfn);
450 }
451 
452 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
453 			    ulong eaddr, int vec)
454 {
455 	bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
456 	bool iswrite = false;
457 	int r = RESUME_GUEST;
458 	int relocated;
459 	int page_found = 0;
460 	struct kvmppc_pte pte;
461 	bool is_mmio = false;
462 	bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
463 	bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
464 	u64 vsid;
465 
466 	relocated = data ? dr : ir;
467 	if (data && (vcpu->arch.fault_dsisr & DSISR_ISSTORE))
468 		iswrite = true;
469 
470 	/* Resolve real address if translation turned on */
471 	if (relocated) {
472 		page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data, iswrite);
473 	} else {
474 		pte.may_execute = true;
475 		pte.may_read = true;
476 		pte.may_write = true;
477 		pte.raddr = eaddr & KVM_PAM;
478 		pte.eaddr = eaddr;
479 		pte.vpage = eaddr >> 12;
480 		pte.page_size = MMU_PAGE_64K;
481 	}
482 
483 	switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
484 	case 0:
485 		pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
486 		break;
487 	case MSR_DR:
488 	case MSR_IR:
489 		vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
490 
491 		if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
492 			pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
493 		else
494 			pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
495 		pte.vpage |= vsid;
496 
497 		if (vsid == -1)
498 			page_found = -EINVAL;
499 		break;
500 	}
501 
502 	if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
503 	   (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
504 		/*
505 		 * If we do the dcbz hack, we have to NX on every execution,
506 		 * so we can patch the executing code. This renders our guest
507 		 * NX-less.
508 		 */
509 		pte.may_execute = !data;
510 	}
511 
512 	if (page_found == -ENOENT) {
513 		/* Page not found in guest PTE entries */
514 		vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
515 		vcpu->arch.shared->dsisr = vcpu->arch.fault_dsisr;
516 		vcpu->arch.shared->msr |=
517 			vcpu->arch.shadow_srr1 & 0x00000000f8000000ULL;
518 		kvmppc_book3s_queue_irqprio(vcpu, vec);
519 	} else if (page_found == -EPERM) {
520 		/* Storage protection */
521 		vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
522 		vcpu->arch.shared->dsisr = vcpu->arch.fault_dsisr & ~DSISR_NOHPTE;
523 		vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
524 		vcpu->arch.shared->msr |=
525 			vcpu->arch.shadow_srr1 & 0x00000000f8000000ULL;
526 		kvmppc_book3s_queue_irqprio(vcpu, vec);
527 	} else if (page_found == -EINVAL) {
528 		/* Page not found in guest SLB */
529 		vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
530 		kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
531 	} else if (!is_mmio &&
532 		   kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
533 		if (data && !(vcpu->arch.fault_dsisr & DSISR_NOHPTE)) {
534 			/*
535 			 * There is already a host HPTE there, presumably
536 			 * a read-only one for a page the guest thinks
537 			 * is writable, so get rid of it first.
538 			 */
539 			kvmppc_mmu_unmap_page(vcpu, &pte);
540 		}
541 		/* The guest's PTE is not mapped yet. Map on the host */
542 		kvmppc_mmu_map_page(vcpu, &pte, iswrite);
543 		if (data)
544 			vcpu->stat.sp_storage++;
545 		else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
546 			 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
547 			kvmppc_patch_dcbz(vcpu, &pte);
548 	} else {
549 		/* MMIO */
550 		vcpu->stat.mmio_exits++;
551 		vcpu->arch.paddr_accessed = pte.raddr;
552 		vcpu->arch.vaddr_accessed = pte.eaddr;
553 		r = kvmppc_emulate_mmio(run, vcpu);
554 		if ( r == RESUME_HOST_NV )
555 			r = RESUME_HOST;
556 	}
557 
558 	return r;
559 }
560 
561 static inline int get_fpr_index(int i)
562 {
563 	return i * TS_FPRWIDTH;
564 }
565 
566 /* Give up external provider (FPU, Altivec, VSX) */
567 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
568 {
569 	struct thread_struct *t = &current->thread;
570 
571 	/*
572 	 * VSX instructions can access FP and vector registers, so if
573 	 * we are giving up VSX, make sure we give up FP and VMX as well.
574 	 */
575 	if (msr & MSR_VSX)
576 		msr |= MSR_FP | MSR_VEC;
577 
578 	msr &= vcpu->arch.guest_owned_ext;
579 	if (!msr)
580 		return;
581 
582 #ifdef DEBUG_EXT
583 	printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
584 #endif
585 
586 	if (msr & MSR_FP) {
587 		/*
588 		 * Note that on CPUs with VSX, giveup_fpu stores
589 		 * both the traditional FP registers and the added VSX
590 		 * registers into thread.fp_state.fpr[].
591 		 */
592 		if (t->regs->msr & MSR_FP)
593 			giveup_fpu(current);
594 		t->fp_save_area = NULL;
595 	}
596 
597 #ifdef CONFIG_ALTIVEC
598 	if (msr & MSR_VEC) {
599 		if (current->thread.regs->msr & MSR_VEC)
600 			giveup_altivec(current);
601 		t->vr_save_area = NULL;
602 	}
603 #endif
604 
605 	vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
606 	kvmppc_recalc_shadow_msr(vcpu);
607 }
608 
609 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
610 {
611 	ulong srr0 = kvmppc_get_pc(vcpu);
612 	u32 last_inst = kvmppc_get_last_inst(vcpu);
613 	int ret;
614 
615 	ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
616 	if (ret == -ENOENT) {
617 		ulong msr = vcpu->arch.shared->msr;
618 
619 		msr = kvmppc_set_field(msr, 33, 33, 1);
620 		msr = kvmppc_set_field(msr, 34, 36, 0);
621 		vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
622 		kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
623 		return EMULATE_AGAIN;
624 	}
625 
626 	return EMULATE_DONE;
627 }
628 
629 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
630 {
631 
632 	/* Need to do paired single emulation? */
633 	if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
634 		return EMULATE_DONE;
635 
636 	/* Read out the instruction */
637 	if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
638 		/* Need to emulate */
639 		return EMULATE_FAIL;
640 
641 	return EMULATE_AGAIN;
642 }
643 
644 /* Handle external providers (FPU, Altivec, VSX) */
645 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
646 			     ulong msr)
647 {
648 	struct thread_struct *t = &current->thread;
649 
650 	/* When we have paired singles, we emulate in software */
651 	if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
652 		return RESUME_GUEST;
653 
654 	if (!(vcpu->arch.shared->msr & msr)) {
655 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
656 		return RESUME_GUEST;
657 	}
658 
659 	if (msr == MSR_VSX) {
660 		/* No VSX?  Give an illegal instruction interrupt */
661 #ifdef CONFIG_VSX
662 		if (!cpu_has_feature(CPU_FTR_VSX))
663 #endif
664 		{
665 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
666 			return RESUME_GUEST;
667 		}
668 
669 		/*
670 		 * We have to load up all the FP and VMX registers before
671 		 * we can let the guest use VSX instructions.
672 		 */
673 		msr = MSR_FP | MSR_VEC | MSR_VSX;
674 	}
675 
676 	/* See if we already own all the ext(s) needed */
677 	msr &= ~vcpu->arch.guest_owned_ext;
678 	if (!msr)
679 		return RESUME_GUEST;
680 
681 #ifdef DEBUG_EXT
682 	printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
683 #endif
684 
685 	if (msr & MSR_FP) {
686 		enable_kernel_fp();
687 		load_fp_state(&vcpu->arch.fp);
688 		t->fp_save_area = &vcpu->arch.fp;
689 	}
690 
691 	if (msr & MSR_VEC) {
692 #ifdef CONFIG_ALTIVEC
693 		enable_kernel_altivec();
694 		load_vr_state(&vcpu->arch.vr);
695 		t->vr_save_area = &vcpu->arch.vr;
696 #endif
697 	}
698 
699 	t->regs->msr |= msr;
700 	vcpu->arch.guest_owned_ext |= msr;
701 	kvmppc_recalc_shadow_msr(vcpu);
702 
703 	return RESUME_GUEST;
704 }
705 
706 /*
707  * Kernel code using FP or VMX could have flushed guest state to
708  * the thread_struct; if so, get it back now.
709  */
710 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
711 {
712 	unsigned long lost_ext;
713 
714 	lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
715 	if (!lost_ext)
716 		return;
717 
718 	if (lost_ext & MSR_FP) {
719 		enable_kernel_fp();
720 		load_fp_state(&vcpu->arch.fp);
721 	}
722 #ifdef CONFIG_ALTIVEC
723 	if (lost_ext & MSR_VEC) {
724 		enable_kernel_altivec();
725 		load_vr_state(&vcpu->arch.vr);
726 	}
727 #endif
728 	current->thread.regs->msr |= lost_ext;
729 }
730 
731 int kvmppc_handle_exit_pr(struct kvm_run *run, struct kvm_vcpu *vcpu,
732 			  unsigned int exit_nr)
733 {
734 	int r = RESUME_HOST;
735 	int s;
736 
737 	vcpu->stat.sum_exits++;
738 
739 	run->exit_reason = KVM_EXIT_UNKNOWN;
740 	run->ready_for_interrupt_injection = 1;
741 
742 	/* We get here with MSR.EE=1 */
743 
744 	trace_kvm_exit(exit_nr, vcpu);
745 	kvm_guest_exit();
746 
747 	switch (exit_nr) {
748 	case BOOK3S_INTERRUPT_INST_STORAGE:
749 	{
750 		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
751 		vcpu->stat.pf_instruc++;
752 
753 #ifdef CONFIG_PPC_BOOK3S_32
754 		/* We set segments as unused segments when invalidating them. So
755 		 * treat the respective fault as segment fault. */
756 		{
757 			struct kvmppc_book3s_shadow_vcpu *svcpu;
758 			u32 sr;
759 
760 			svcpu = svcpu_get(vcpu);
761 			sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
762 			svcpu_put(svcpu);
763 			if (sr == SR_INVALID) {
764 				kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
765 				r = RESUME_GUEST;
766 				break;
767 			}
768 		}
769 #endif
770 
771 		/* only care about PTEG not found errors, but leave NX alone */
772 		if (shadow_srr1 & 0x40000000) {
773 			int idx = srcu_read_lock(&vcpu->kvm->srcu);
774 			r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
775 			srcu_read_unlock(&vcpu->kvm->srcu, idx);
776 			vcpu->stat.sp_instruc++;
777 		} else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
778 			  (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
779 			/*
780 			 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
781 			 *     so we can't use the NX bit inside the guest. Let's cross our fingers,
782 			 *     that no guest that needs the dcbz hack does NX.
783 			 */
784 			kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
785 			r = RESUME_GUEST;
786 		} else {
787 			vcpu->arch.shared->msr |= shadow_srr1 & 0x58000000;
788 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
789 			r = RESUME_GUEST;
790 		}
791 		break;
792 	}
793 	case BOOK3S_INTERRUPT_DATA_STORAGE:
794 	{
795 		ulong dar = kvmppc_get_fault_dar(vcpu);
796 		u32 fault_dsisr = vcpu->arch.fault_dsisr;
797 		vcpu->stat.pf_storage++;
798 
799 #ifdef CONFIG_PPC_BOOK3S_32
800 		/* We set segments as unused segments when invalidating them. So
801 		 * treat the respective fault as segment fault. */
802 		{
803 			struct kvmppc_book3s_shadow_vcpu *svcpu;
804 			u32 sr;
805 
806 			svcpu = svcpu_get(vcpu);
807 			sr = svcpu->sr[dar >> SID_SHIFT];
808 			svcpu_put(svcpu);
809 			if (sr == SR_INVALID) {
810 				kvmppc_mmu_map_segment(vcpu, dar);
811 				r = RESUME_GUEST;
812 				break;
813 			}
814 		}
815 #endif
816 
817 		/*
818 		 * We need to handle missing shadow PTEs, and
819 		 * protection faults due to us mapping a page read-only
820 		 * when the guest thinks it is writable.
821 		 */
822 		if (fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT)) {
823 			int idx = srcu_read_lock(&vcpu->kvm->srcu);
824 			r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
825 			srcu_read_unlock(&vcpu->kvm->srcu, idx);
826 		} else {
827 			vcpu->arch.shared->dar = dar;
828 			vcpu->arch.shared->dsisr = fault_dsisr;
829 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
830 			r = RESUME_GUEST;
831 		}
832 		break;
833 	}
834 	case BOOK3S_INTERRUPT_DATA_SEGMENT:
835 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
836 			vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
837 			kvmppc_book3s_queue_irqprio(vcpu,
838 				BOOK3S_INTERRUPT_DATA_SEGMENT);
839 		}
840 		r = RESUME_GUEST;
841 		break;
842 	case BOOK3S_INTERRUPT_INST_SEGMENT:
843 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
844 			kvmppc_book3s_queue_irqprio(vcpu,
845 				BOOK3S_INTERRUPT_INST_SEGMENT);
846 		}
847 		r = RESUME_GUEST;
848 		break;
849 	/* We're good on these - the host merely wanted to get our attention */
850 	case BOOK3S_INTERRUPT_DECREMENTER:
851 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
852 	case BOOK3S_INTERRUPT_DOORBELL:
853 		vcpu->stat.dec_exits++;
854 		r = RESUME_GUEST;
855 		break;
856 	case BOOK3S_INTERRUPT_EXTERNAL:
857 	case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
858 	case BOOK3S_INTERRUPT_EXTERNAL_HV:
859 		vcpu->stat.ext_intr_exits++;
860 		r = RESUME_GUEST;
861 		break;
862 	case BOOK3S_INTERRUPT_PERFMON:
863 		r = RESUME_GUEST;
864 		break;
865 	case BOOK3S_INTERRUPT_PROGRAM:
866 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
867 	{
868 		enum emulation_result er;
869 		ulong flags;
870 
871 program_interrupt:
872 		flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
873 
874 		if (vcpu->arch.shared->msr & MSR_PR) {
875 #ifdef EXIT_DEBUG
876 			printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
877 #endif
878 			if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
879 			    (INS_DCBZ & 0xfffffff7)) {
880 				kvmppc_core_queue_program(vcpu, flags);
881 				r = RESUME_GUEST;
882 				break;
883 			}
884 		}
885 
886 		vcpu->stat.emulated_inst_exits++;
887 		er = kvmppc_emulate_instruction(run, vcpu);
888 		switch (er) {
889 		case EMULATE_DONE:
890 			r = RESUME_GUEST_NV;
891 			break;
892 		case EMULATE_AGAIN:
893 			r = RESUME_GUEST;
894 			break;
895 		case EMULATE_FAIL:
896 			printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
897 			       __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
898 			kvmppc_core_queue_program(vcpu, flags);
899 			r = RESUME_GUEST;
900 			break;
901 		case EMULATE_DO_MMIO:
902 			run->exit_reason = KVM_EXIT_MMIO;
903 			r = RESUME_HOST_NV;
904 			break;
905 		case EMULATE_EXIT_USER:
906 			r = RESUME_HOST_NV;
907 			break;
908 		default:
909 			BUG();
910 		}
911 		break;
912 	}
913 	case BOOK3S_INTERRUPT_SYSCALL:
914 		if (vcpu->arch.papr_enabled &&
915 		    (kvmppc_get_last_sc(vcpu) == 0x44000022) &&
916 		    !(vcpu->arch.shared->msr & MSR_PR)) {
917 			/* SC 1 papr hypercalls */
918 			ulong cmd = kvmppc_get_gpr(vcpu, 3);
919 			int i;
920 
921 #ifdef CONFIG_PPC_BOOK3S_64
922 			if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
923 				r = RESUME_GUEST;
924 				break;
925 			}
926 #endif
927 
928 			run->papr_hcall.nr = cmd;
929 			for (i = 0; i < 9; ++i) {
930 				ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
931 				run->papr_hcall.args[i] = gpr;
932 			}
933 			run->exit_reason = KVM_EXIT_PAPR_HCALL;
934 			vcpu->arch.hcall_needed = 1;
935 			r = RESUME_HOST;
936 		} else if (vcpu->arch.osi_enabled &&
937 		    (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
938 		    (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
939 			/* MOL hypercalls */
940 			u64 *gprs = run->osi.gprs;
941 			int i;
942 
943 			run->exit_reason = KVM_EXIT_OSI;
944 			for (i = 0; i < 32; i++)
945 				gprs[i] = kvmppc_get_gpr(vcpu, i);
946 			vcpu->arch.osi_needed = 1;
947 			r = RESUME_HOST_NV;
948 		} else if (!(vcpu->arch.shared->msr & MSR_PR) &&
949 		    (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
950 			/* KVM PV hypercalls */
951 			kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
952 			r = RESUME_GUEST;
953 		} else {
954 			/* Guest syscalls */
955 			vcpu->stat.syscall_exits++;
956 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
957 			r = RESUME_GUEST;
958 		}
959 		break;
960 	case BOOK3S_INTERRUPT_FP_UNAVAIL:
961 	case BOOK3S_INTERRUPT_ALTIVEC:
962 	case BOOK3S_INTERRUPT_VSX:
963 	{
964 		int ext_msr = 0;
965 
966 		switch (exit_nr) {
967 		case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
968 		case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
969 		case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
970 		}
971 
972 		switch (kvmppc_check_ext(vcpu, exit_nr)) {
973 		case EMULATE_DONE:
974 			/* everything ok - let's enable the ext */
975 			r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
976 			break;
977 		case EMULATE_FAIL:
978 			/* we need to emulate this instruction */
979 			goto program_interrupt;
980 			break;
981 		default:
982 			/* nothing to worry about - go again */
983 			break;
984 		}
985 		break;
986 	}
987 	case BOOK3S_INTERRUPT_ALIGNMENT:
988 		if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
989 			vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
990 				kvmppc_get_last_inst(vcpu));
991 			vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
992 				kvmppc_get_last_inst(vcpu));
993 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
994 		}
995 		r = RESUME_GUEST;
996 		break;
997 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
998 	case BOOK3S_INTERRUPT_TRACE:
999 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1000 		r = RESUME_GUEST;
1001 		break;
1002 	default:
1003 	{
1004 		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1005 		/* Ugh - bork here! What did we get? */
1006 		printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1007 			exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
1008 		r = RESUME_HOST;
1009 		BUG();
1010 		break;
1011 	}
1012 	}
1013 
1014 	if (!(r & RESUME_HOST)) {
1015 		/* To avoid clobbering exit_reason, only check for signals if
1016 		 * we aren't already exiting to userspace for some other
1017 		 * reason. */
1018 
1019 		/*
1020 		 * Interrupts could be timers for the guest which we have to
1021 		 * inject again, so let's postpone them until we're in the guest
1022 		 * and if we really did time things so badly, then we just exit
1023 		 * again due to a host external interrupt.
1024 		 */
1025 		s = kvmppc_prepare_to_enter(vcpu);
1026 		if (s <= 0)
1027 			r = s;
1028 		else {
1029 			/* interrupts now hard-disabled */
1030 			kvmppc_fix_ee_before_entry();
1031 		}
1032 
1033 		kvmppc_handle_lost_ext(vcpu);
1034 	}
1035 
1036 	trace_kvm_book3s_reenter(r, vcpu);
1037 
1038 	return r;
1039 }
1040 
1041 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu *vcpu,
1042 					    struct kvm_sregs *sregs)
1043 {
1044 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1045 	int i;
1046 
1047 	sregs->pvr = vcpu->arch.pvr;
1048 
1049 	sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1050 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1051 		for (i = 0; i < 64; i++) {
1052 			sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
1053 			sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1054 		}
1055 	} else {
1056 		for (i = 0; i < 16; i++)
1057 			sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
1058 
1059 		for (i = 0; i < 8; i++) {
1060 			sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1061 			sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1062 		}
1063 	}
1064 
1065 	return 0;
1066 }
1067 
1068 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu *vcpu,
1069 					    struct kvm_sregs *sregs)
1070 {
1071 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1072 	int i;
1073 
1074 	kvmppc_set_pvr_pr(vcpu, sregs->pvr);
1075 
1076 	vcpu3s->sdr1 = sregs->u.s.sdr1;
1077 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1078 		for (i = 0; i < 64; i++) {
1079 			vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1080 						    sregs->u.s.ppc64.slb[i].slbe);
1081 		}
1082 	} else {
1083 		for (i = 0; i < 16; i++) {
1084 			vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1085 		}
1086 		for (i = 0; i < 8; i++) {
1087 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1088 				       (u32)sregs->u.s.ppc32.ibat[i]);
1089 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1090 				       (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1091 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1092 				       (u32)sregs->u.s.ppc32.dbat[i]);
1093 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1094 				       (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1095 		}
1096 	}
1097 
1098 	/* Flush the MMU after messing with the segments */
1099 	kvmppc_mmu_pte_flush(vcpu, 0, 0);
1100 
1101 	return 0;
1102 }
1103 
1104 static int kvmppc_get_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1105 				 union kvmppc_one_reg *val)
1106 {
1107 	int r = 0;
1108 
1109 	switch (id) {
1110 	case KVM_REG_PPC_HIOR:
1111 		*val = get_reg_val(id, to_book3s(vcpu)->hior);
1112 		break;
1113 	default:
1114 		r = -EINVAL;
1115 		break;
1116 	}
1117 
1118 	return r;
1119 }
1120 
1121 static int kvmppc_set_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1122 				 union kvmppc_one_reg *val)
1123 {
1124 	int r = 0;
1125 
1126 	switch (id) {
1127 	case KVM_REG_PPC_HIOR:
1128 		to_book3s(vcpu)->hior = set_reg_val(id, *val);
1129 		to_book3s(vcpu)->hior_explicit = true;
1130 		break;
1131 	default:
1132 		r = -EINVAL;
1133 		break;
1134 	}
1135 
1136 	return r;
1137 }
1138 
1139 static struct kvm_vcpu *kvmppc_core_vcpu_create_pr(struct kvm *kvm,
1140 						   unsigned int id)
1141 {
1142 	struct kvmppc_vcpu_book3s *vcpu_book3s;
1143 	struct kvm_vcpu *vcpu;
1144 	int err = -ENOMEM;
1145 	unsigned long p;
1146 
1147 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1148 	if (!vcpu)
1149 		goto out;
1150 
1151 	vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1152 	if (!vcpu_book3s)
1153 		goto free_vcpu;
1154 	vcpu->arch.book3s = vcpu_book3s;
1155 
1156 #ifdef CONFIG_KVM_BOOK3S_32
1157 	vcpu->arch.shadow_vcpu =
1158 		kzalloc(sizeof(*vcpu->arch.shadow_vcpu), GFP_KERNEL);
1159 	if (!vcpu->arch.shadow_vcpu)
1160 		goto free_vcpu3s;
1161 #endif
1162 
1163 	err = kvm_vcpu_init(vcpu, kvm, id);
1164 	if (err)
1165 		goto free_shadow_vcpu;
1166 
1167 	err = -ENOMEM;
1168 	p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1169 	if (!p)
1170 		goto uninit_vcpu;
1171 	/* the real shared page fills the last 4k of our page */
1172 	vcpu->arch.shared = (void *)(p + PAGE_SIZE - 4096);
1173 
1174 #ifdef CONFIG_PPC_BOOK3S_64
1175 	/*
1176 	 * Default to the same as the host if we're on sufficiently
1177 	 * recent machine that we have 1TB segments;
1178 	 * otherwise default to PPC970FX.
1179 	 */
1180 	vcpu->arch.pvr = 0x3C0301;
1181 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1182 		vcpu->arch.pvr = mfspr(SPRN_PVR);
1183 #else
1184 	/* default to book3s_32 (750) */
1185 	vcpu->arch.pvr = 0x84202;
1186 #endif
1187 	kvmppc_set_pvr_pr(vcpu, vcpu->arch.pvr);
1188 	vcpu->arch.slb_nr = 64;
1189 
1190 	vcpu->arch.shadow_msr = MSR_USER64;
1191 
1192 	err = kvmppc_mmu_init(vcpu);
1193 	if (err < 0)
1194 		goto uninit_vcpu;
1195 
1196 	return vcpu;
1197 
1198 uninit_vcpu:
1199 	kvm_vcpu_uninit(vcpu);
1200 free_shadow_vcpu:
1201 #ifdef CONFIG_KVM_BOOK3S_32
1202 	kfree(vcpu->arch.shadow_vcpu);
1203 free_vcpu3s:
1204 #endif
1205 	vfree(vcpu_book3s);
1206 free_vcpu:
1207 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1208 out:
1209 	return ERR_PTR(err);
1210 }
1211 
1212 static void kvmppc_core_vcpu_free_pr(struct kvm_vcpu *vcpu)
1213 {
1214 	struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1215 
1216 	free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1217 	kvm_vcpu_uninit(vcpu);
1218 #ifdef CONFIG_KVM_BOOK3S_32
1219 	kfree(vcpu->arch.shadow_vcpu);
1220 #endif
1221 	vfree(vcpu_book3s);
1222 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1223 }
1224 
1225 static int kvmppc_vcpu_run_pr(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1226 {
1227 	int ret;
1228 #ifdef CONFIG_ALTIVEC
1229 	unsigned long uninitialized_var(vrsave);
1230 #endif
1231 
1232 	/* Check if we can run the vcpu at all */
1233 	if (!vcpu->arch.sane) {
1234 		kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1235 		ret = -EINVAL;
1236 		goto out;
1237 	}
1238 
1239 	/*
1240 	 * Interrupts could be timers for the guest which we have to inject
1241 	 * again, so let's postpone them until we're in the guest and if we
1242 	 * really did time things so badly, then we just exit again due to
1243 	 * a host external interrupt.
1244 	 */
1245 	ret = kvmppc_prepare_to_enter(vcpu);
1246 	if (ret <= 0)
1247 		goto out;
1248 	/* interrupts now hard-disabled */
1249 
1250 	/* Save FPU state in thread_struct */
1251 	if (current->thread.regs->msr & MSR_FP)
1252 		giveup_fpu(current);
1253 
1254 #ifdef CONFIG_ALTIVEC
1255 	/* Save Altivec state in thread_struct */
1256 	if (current->thread.regs->msr & MSR_VEC)
1257 		giveup_altivec(current);
1258 #endif
1259 
1260 #ifdef CONFIG_VSX
1261 	/* Save VSX state in thread_struct */
1262 	if (current->thread.regs->msr & MSR_VSX)
1263 		__giveup_vsx(current);
1264 #endif
1265 
1266 	/* Preload FPU if it's enabled */
1267 	if (vcpu->arch.shared->msr & MSR_FP)
1268 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1269 
1270 	kvmppc_fix_ee_before_entry();
1271 
1272 	ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1273 
1274 	/* No need for kvm_guest_exit. It's done in handle_exit.
1275 	   We also get here with interrupts enabled. */
1276 
1277 	/* Make sure we save the guest FPU/Altivec/VSX state */
1278 	kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1279 
1280 out:
1281 	vcpu->mode = OUTSIDE_GUEST_MODE;
1282 	return ret;
1283 }
1284 
1285 /*
1286  * Get (and clear) the dirty memory log for a memory slot.
1287  */
1288 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm *kvm,
1289 					 struct kvm_dirty_log *log)
1290 {
1291 	struct kvm_memory_slot *memslot;
1292 	struct kvm_vcpu *vcpu;
1293 	ulong ga, ga_end;
1294 	int is_dirty = 0;
1295 	int r;
1296 	unsigned long n;
1297 
1298 	mutex_lock(&kvm->slots_lock);
1299 
1300 	r = kvm_get_dirty_log(kvm, log, &is_dirty);
1301 	if (r)
1302 		goto out;
1303 
1304 	/* If nothing is dirty, don't bother messing with page tables. */
1305 	if (is_dirty) {
1306 		memslot = id_to_memslot(kvm->memslots, log->slot);
1307 
1308 		ga = memslot->base_gfn << PAGE_SHIFT;
1309 		ga_end = ga + (memslot->npages << PAGE_SHIFT);
1310 
1311 		kvm_for_each_vcpu(n, vcpu, kvm)
1312 			kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1313 
1314 		n = kvm_dirty_bitmap_bytes(memslot);
1315 		memset(memslot->dirty_bitmap, 0, n);
1316 	}
1317 
1318 	r = 0;
1319 out:
1320 	mutex_unlock(&kvm->slots_lock);
1321 	return r;
1322 }
1323 
1324 static void kvmppc_core_flush_memslot_pr(struct kvm *kvm,
1325 					 struct kvm_memory_slot *memslot)
1326 {
1327 	return;
1328 }
1329 
1330 static int kvmppc_core_prepare_memory_region_pr(struct kvm *kvm,
1331 					struct kvm_memory_slot *memslot,
1332 					struct kvm_userspace_memory_region *mem)
1333 {
1334 	return 0;
1335 }
1336 
1337 static void kvmppc_core_commit_memory_region_pr(struct kvm *kvm,
1338 				struct kvm_userspace_memory_region *mem,
1339 				const struct kvm_memory_slot *old)
1340 {
1341 	return;
1342 }
1343 
1344 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot *free,
1345 					struct kvm_memory_slot *dont)
1346 {
1347 	return;
1348 }
1349 
1350 static int kvmppc_core_create_memslot_pr(struct kvm_memory_slot *slot,
1351 					 unsigned long npages)
1352 {
1353 	return 0;
1354 }
1355 
1356 
1357 #ifdef CONFIG_PPC64
1358 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1359 					 struct kvm_ppc_smmu_info *info)
1360 {
1361 	long int i;
1362 	struct kvm_vcpu *vcpu;
1363 
1364 	info->flags = 0;
1365 
1366 	/* SLB is always 64 entries */
1367 	info->slb_size = 64;
1368 
1369 	/* Standard 4k base page size segment */
1370 	info->sps[0].page_shift = 12;
1371 	info->sps[0].slb_enc = 0;
1372 	info->sps[0].enc[0].page_shift = 12;
1373 	info->sps[0].enc[0].pte_enc = 0;
1374 
1375 	/*
1376 	 * 64k large page size.
1377 	 * We only want to put this in if the CPUs we're emulating
1378 	 * support it, but unfortunately we don't have a vcpu easily
1379 	 * to hand here to test.  Just pick the first vcpu, and if
1380 	 * that doesn't exist yet, report the minimum capability,
1381 	 * i.e., no 64k pages.
1382 	 * 1T segment support goes along with 64k pages.
1383 	 */
1384 	i = 1;
1385 	vcpu = kvm_get_vcpu(kvm, 0);
1386 	if (vcpu && (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
1387 		info->flags = KVM_PPC_1T_SEGMENTS;
1388 		info->sps[i].page_shift = 16;
1389 		info->sps[i].slb_enc = SLB_VSID_L | SLB_VSID_LP_01;
1390 		info->sps[i].enc[0].page_shift = 16;
1391 		info->sps[i].enc[0].pte_enc = 1;
1392 		++i;
1393 	}
1394 
1395 	/* Standard 16M large page size segment */
1396 	info->sps[i].page_shift = 24;
1397 	info->sps[i].slb_enc = SLB_VSID_L;
1398 	info->sps[i].enc[0].page_shift = 24;
1399 	info->sps[i].enc[0].pte_enc = 0;
1400 
1401 	return 0;
1402 }
1403 #else
1404 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1405 					 struct kvm_ppc_smmu_info *info)
1406 {
1407 	/* We should not get called */
1408 	BUG();
1409 }
1410 #endif /* CONFIG_PPC64 */
1411 
1412 static unsigned int kvm_global_user_count = 0;
1413 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1414 
1415 static int kvmppc_core_init_vm_pr(struct kvm *kvm)
1416 {
1417 	mutex_init(&kvm->arch.hpt_mutex);
1418 
1419 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1420 		spin_lock(&kvm_global_user_count_lock);
1421 		if (++kvm_global_user_count == 1)
1422 			pSeries_disable_reloc_on_exc();
1423 		spin_unlock(&kvm_global_user_count_lock);
1424 	}
1425 	return 0;
1426 }
1427 
1428 static void kvmppc_core_destroy_vm_pr(struct kvm *kvm)
1429 {
1430 #ifdef CONFIG_PPC64
1431 	WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1432 #endif
1433 
1434 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1435 		spin_lock(&kvm_global_user_count_lock);
1436 		BUG_ON(kvm_global_user_count == 0);
1437 		if (--kvm_global_user_count == 0)
1438 			pSeries_enable_reloc_on_exc();
1439 		spin_unlock(&kvm_global_user_count_lock);
1440 	}
1441 }
1442 
1443 static int kvmppc_core_check_processor_compat_pr(void)
1444 {
1445 	/* we are always compatible */
1446 	return 0;
1447 }
1448 
1449 static long kvm_arch_vm_ioctl_pr(struct file *filp,
1450 				 unsigned int ioctl, unsigned long arg)
1451 {
1452 	return -ENOTTY;
1453 }
1454 
1455 static struct kvmppc_ops kvm_ops_pr = {
1456 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_pr,
1457 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_pr,
1458 	.get_one_reg = kvmppc_get_one_reg_pr,
1459 	.set_one_reg = kvmppc_set_one_reg_pr,
1460 	.vcpu_load   = kvmppc_core_vcpu_load_pr,
1461 	.vcpu_put    = kvmppc_core_vcpu_put_pr,
1462 	.set_msr     = kvmppc_set_msr_pr,
1463 	.vcpu_run    = kvmppc_vcpu_run_pr,
1464 	.vcpu_create = kvmppc_core_vcpu_create_pr,
1465 	.vcpu_free   = kvmppc_core_vcpu_free_pr,
1466 	.check_requests = kvmppc_core_check_requests_pr,
1467 	.get_dirty_log = kvm_vm_ioctl_get_dirty_log_pr,
1468 	.flush_memslot = kvmppc_core_flush_memslot_pr,
1469 	.prepare_memory_region = kvmppc_core_prepare_memory_region_pr,
1470 	.commit_memory_region = kvmppc_core_commit_memory_region_pr,
1471 	.unmap_hva = kvm_unmap_hva_pr,
1472 	.unmap_hva_range = kvm_unmap_hva_range_pr,
1473 	.age_hva  = kvm_age_hva_pr,
1474 	.test_age_hva = kvm_test_age_hva_pr,
1475 	.set_spte_hva = kvm_set_spte_hva_pr,
1476 	.mmu_destroy  = kvmppc_mmu_destroy_pr,
1477 	.free_memslot = kvmppc_core_free_memslot_pr,
1478 	.create_memslot = kvmppc_core_create_memslot_pr,
1479 	.init_vm = kvmppc_core_init_vm_pr,
1480 	.destroy_vm = kvmppc_core_destroy_vm_pr,
1481 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_pr,
1482 	.emulate_op = kvmppc_core_emulate_op_pr,
1483 	.emulate_mtspr = kvmppc_core_emulate_mtspr_pr,
1484 	.emulate_mfspr = kvmppc_core_emulate_mfspr_pr,
1485 	.fast_vcpu_kick = kvm_vcpu_kick,
1486 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_pr,
1487 };
1488 
1489 
1490 int kvmppc_book3s_init_pr(void)
1491 {
1492 	int r;
1493 
1494 	r = kvmppc_core_check_processor_compat_pr();
1495 	if (r < 0)
1496 		return r;
1497 
1498 	kvm_ops_pr.owner = THIS_MODULE;
1499 	kvmppc_pr_ops = &kvm_ops_pr;
1500 
1501 	r = kvmppc_mmu_hpte_sysinit();
1502 	return r;
1503 }
1504 
1505 void kvmppc_book3s_exit_pr(void)
1506 {
1507 	kvmppc_pr_ops = NULL;
1508 	kvmppc_mmu_hpte_sysexit();
1509 }
1510 
1511 /*
1512  * We only support separate modules for book3s 64
1513  */
1514 #ifdef CONFIG_PPC_BOOK3S_64
1515 
1516 module_init(kvmppc_book3s_init_pr);
1517 module_exit(kvmppc_book3s_exit_pr);
1518 
1519 MODULE_LICENSE("GPL");
1520 MODULE_ALIAS_MISCDEV(KVM_MINOR);
1521 MODULE_ALIAS("devname:kvm");
1522 #endif
1523