xref: /openbmc/linux/arch/powerpc/kvm/book3s_pr.c (revision 5d0e4d78)
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 <linux/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/setup.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 static void kvmppc_giveup_fac(struct kvm_vcpu *vcpu, ulong fac);
57 
58 /* Some compatibility defines */
59 #ifdef CONFIG_PPC_BOOK3S_32
60 #define MSR_USER32 MSR_USER
61 #define MSR_USER64 MSR_USER
62 #define HW_PAGE_SIZE PAGE_SIZE
63 #endif
64 
65 static bool kvmppc_is_split_real(struct kvm_vcpu *vcpu)
66 {
67 	ulong msr = kvmppc_get_msr(vcpu);
68 	return (msr & (MSR_IR|MSR_DR)) == MSR_DR;
69 }
70 
71 static void kvmppc_fixup_split_real(struct kvm_vcpu *vcpu)
72 {
73 	ulong msr = kvmppc_get_msr(vcpu);
74 	ulong pc = kvmppc_get_pc(vcpu);
75 
76 	/* We are in DR only split real mode */
77 	if ((msr & (MSR_IR|MSR_DR)) != MSR_DR)
78 		return;
79 
80 	/* We have not fixed up the guest already */
81 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK)
82 		return;
83 
84 	/* The code is in fixupable address space */
85 	if (pc & SPLIT_HACK_MASK)
86 		return;
87 
88 	vcpu->arch.hflags |= BOOK3S_HFLAG_SPLIT_HACK;
89 	kvmppc_set_pc(vcpu, pc | SPLIT_HACK_OFFS);
90 }
91 
92 void kvmppc_unfixup_split_real(struct kvm_vcpu *vcpu);
93 
94 static void kvmppc_core_vcpu_load_pr(struct kvm_vcpu *vcpu, int cpu)
95 {
96 #ifdef CONFIG_PPC_BOOK3S_64
97 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
98 	memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb));
99 	svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max;
100 	svcpu->in_use = 0;
101 	svcpu_put(svcpu);
102 #endif
103 
104 	/* Disable AIL if supported */
105 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
106 	    cpu_has_feature(CPU_FTR_ARCH_207S))
107 		mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) & ~LPCR_AIL);
108 
109 	vcpu->cpu = smp_processor_id();
110 #ifdef CONFIG_PPC_BOOK3S_32
111 	current->thread.kvm_shadow_vcpu = vcpu->arch.shadow_vcpu;
112 #endif
113 
114 	if (kvmppc_is_split_real(vcpu))
115 		kvmppc_fixup_split_real(vcpu);
116 }
117 
118 static void kvmppc_core_vcpu_put_pr(struct kvm_vcpu *vcpu)
119 {
120 #ifdef CONFIG_PPC_BOOK3S_64
121 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
122 	if (svcpu->in_use) {
123 		kvmppc_copy_from_svcpu(vcpu, svcpu);
124 	}
125 	memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
126 	to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max;
127 	svcpu_put(svcpu);
128 #endif
129 
130 	if (kvmppc_is_split_real(vcpu))
131 		kvmppc_unfixup_split_real(vcpu);
132 
133 	kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
134 	kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
135 
136 	/* Enable AIL if supported */
137 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
138 	    cpu_has_feature(CPU_FTR_ARCH_207S))
139 		mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_AIL_3);
140 
141 	vcpu->cpu = -1;
142 }
143 
144 /* Copy data needed by real-mode code from vcpu to shadow vcpu */
145 void kvmppc_copy_to_svcpu(struct kvmppc_book3s_shadow_vcpu *svcpu,
146 			  struct kvm_vcpu *vcpu)
147 {
148 	svcpu->gpr[0] = vcpu->arch.gpr[0];
149 	svcpu->gpr[1] = vcpu->arch.gpr[1];
150 	svcpu->gpr[2] = vcpu->arch.gpr[2];
151 	svcpu->gpr[3] = vcpu->arch.gpr[3];
152 	svcpu->gpr[4] = vcpu->arch.gpr[4];
153 	svcpu->gpr[5] = vcpu->arch.gpr[5];
154 	svcpu->gpr[6] = vcpu->arch.gpr[6];
155 	svcpu->gpr[7] = vcpu->arch.gpr[7];
156 	svcpu->gpr[8] = vcpu->arch.gpr[8];
157 	svcpu->gpr[9] = vcpu->arch.gpr[9];
158 	svcpu->gpr[10] = vcpu->arch.gpr[10];
159 	svcpu->gpr[11] = vcpu->arch.gpr[11];
160 	svcpu->gpr[12] = vcpu->arch.gpr[12];
161 	svcpu->gpr[13] = vcpu->arch.gpr[13];
162 	svcpu->cr  = vcpu->arch.cr;
163 	svcpu->xer = vcpu->arch.xer;
164 	svcpu->ctr = vcpu->arch.ctr;
165 	svcpu->lr  = vcpu->arch.lr;
166 	svcpu->pc  = vcpu->arch.pc;
167 #ifdef CONFIG_PPC_BOOK3S_64
168 	svcpu->shadow_fscr = vcpu->arch.shadow_fscr;
169 #endif
170 	/*
171 	 * Now also save the current time base value. We use this
172 	 * to find the guest purr and spurr value.
173 	 */
174 	vcpu->arch.entry_tb = get_tb();
175 	vcpu->arch.entry_vtb = get_vtb();
176 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
177 		vcpu->arch.entry_ic = mfspr(SPRN_IC);
178 	svcpu->in_use = true;
179 }
180 
181 /* Copy data touched by real-mode code from shadow vcpu back to vcpu */
182 void kvmppc_copy_from_svcpu(struct kvm_vcpu *vcpu,
183 			    struct kvmppc_book3s_shadow_vcpu *svcpu)
184 {
185 	/*
186 	 * vcpu_put would just call us again because in_use hasn't
187 	 * been updated yet.
188 	 */
189 	preempt_disable();
190 
191 	/*
192 	 * Maybe we were already preempted and synced the svcpu from
193 	 * our preempt notifiers. Don't bother touching this svcpu then.
194 	 */
195 	if (!svcpu->in_use)
196 		goto out;
197 
198 	vcpu->arch.gpr[0] = svcpu->gpr[0];
199 	vcpu->arch.gpr[1] = svcpu->gpr[1];
200 	vcpu->arch.gpr[2] = svcpu->gpr[2];
201 	vcpu->arch.gpr[3] = svcpu->gpr[3];
202 	vcpu->arch.gpr[4] = svcpu->gpr[4];
203 	vcpu->arch.gpr[5] = svcpu->gpr[5];
204 	vcpu->arch.gpr[6] = svcpu->gpr[6];
205 	vcpu->arch.gpr[7] = svcpu->gpr[7];
206 	vcpu->arch.gpr[8] = svcpu->gpr[8];
207 	vcpu->arch.gpr[9] = svcpu->gpr[9];
208 	vcpu->arch.gpr[10] = svcpu->gpr[10];
209 	vcpu->arch.gpr[11] = svcpu->gpr[11];
210 	vcpu->arch.gpr[12] = svcpu->gpr[12];
211 	vcpu->arch.gpr[13] = svcpu->gpr[13];
212 	vcpu->arch.cr  = svcpu->cr;
213 	vcpu->arch.xer = svcpu->xer;
214 	vcpu->arch.ctr = svcpu->ctr;
215 	vcpu->arch.lr  = svcpu->lr;
216 	vcpu->arch.pc  = svcpu->pc;
217 	vcpu->arch.shadow_srr1 = svcpu->shadow_srr1;
218 	vcpu->arch.fault_dar   = svcpu->fault_dar;
219 	vcpu->arch.fault_dsisr = svcpu->fault_dsisr;
220 	vcpu->arch.last_inst   = svcpu->last_inst;
221 #ifdef CONFIG_PPC_BOOK3S_64
222 	vcpu->arch.shadow_fscr = svcpu->shadow_fscr;
223 #endif
224 	/*
225 	 * Update purr and spurr using time base on exit.
226 	 */
227 	vcpu->arch.purr += get_tb() - vcpu->arch.entry_tb;
228 	vcpu->arch.spurr += get_tb() - vcpu->arch.entry_tb;
229 	to_book3s(vcpu)->vtb += get_vtb() - vcpu->arch.entry_vtb;
230 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
231 		vcpu->arch.ic += mfspr(SPRN_IC) - vcpu->arch.entry_ic;
232 	svcpu->in_use = false;
233 
234 out:
235 	preempt_enable();
236 }
237 
238 static int kvmppc_core_check_requests_pr(struct kvm_vcpu *vcpu)
239 {
240 	int r = 1; /* Indicate we want to get back into the guest */
241 
242 	/* We misuse TLB_FLUSH to indicate that we want to clear
243 	   all shadow cache entries */
244 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
245 		kvmppc_mmu_pte_flush(vcpu, 0, 0);
246 
247 	return r;
248 }
249 
250 /************* MMU Notifiers *************/
251 static void do_kvm_unmap_hva(struct kvm *kvm, unsigned long start,
252 			     unsigned long end)
253 {
254 	long i;
255 	struct kvm_vcpu *vcpu;
256 	struct kvm_memslots *slots;
257 	struct kvm_memory_slot *memslot;
258 
259 	slots = kvm_memslots(kvm);
260 	kvm_for_each_memslot(memslot, slots) {
261 		unsigned long hva_start, hva_end;
262 		gfn_t gfn, gfn_end;
263 
264 		hva_start = max(start, memslot->userspace_addr);
265 		hva_end = min(end, memslot->userspace_addr +
266 					(memslot->npages << PAGE_SHIFT));
267 		if (hva_start >= hva_end)
268 			continue;
269 		/*
270 		 * {gfn(page) | page intersects with [hva_start, hva_end)} =
271 		 * {gfn, gfn+1, ..., gfn_end-1}.
272 		 */
273 		gfn = hva_to_gfn_memslot(hva_start, memslot);
274 		gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
275 		kvm_for_each_vcpu(i, vcpu, kvm)
276 			kvmppc_mmu_pte_pflush(vcpu, gfn << PAGE_SHIFT,
277 					      gfn_end << PAGE_SHIFT);
278 	}
279 }
280 
281 static int kvm_unmap_hva_pr(struct kvm *kvm, unsigned long hva)
282 {
283 	trace_kvm_unmap_hva(hva);
284 
285 	do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
286 
287 	return 0;
288 }
289 
290 static int kvm_unmap_hva_range_pr(struct kvm *kvm, unsigned long start,
291 				  unsigned long end)
292 {
293 	do_kvm_unmap_hva(kvm, start, end);
294 
295 	return 0;
296 }
297 
298 static int kvm_age_hva_pr(struct kvm *kvm, unsigned long start,
299 			  unsigned long end)
300 {
301 	/* XXX could be more clever ;) */
302 	return 0;
303 }
304 
305 static int kvm_test_age_hva_pr(struct kvm *kvm, unsigned long hva)
306 {
307 	/* XXX could be more clever ;) */
308 	return 0;
309 }
310 
311 static void kvm_set_spte_hva_pr(struct kvm *kvm, unsigned long hva, pte_t pte)
312 {
313 	/* The page will get remapped properly on its next fault */
314 	do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
315 }
316 
317 /*****************************************/
318 
319 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
320 {
321 	ulong guest_msr = kvmppc_get_msr(vcpu);
322 	ulong smsr = guest_msr;
323 
324 	/* Guest MSR values */
325 	smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_LE;
326 	/* Process MSR values */
327 	smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
328 	/* External providers the guest reserved */
329 	smsr |= (guest_msr & vcpu->arch.guest_owned_ext);
330 	/* 64-bit Process MSR values */
331 #ifdef CONFIG_PPC_BOOK3S_64
332 	smsr |= MSR_ISF | MSR_HV;
333 #endif
334 	vcpu->arch.shadow_msr = smsr;
335 }
336 
337 static void kvmppc_set_msr_pr(struct kvm_vcpu *vcpu, u64 msr)
338 {
339 	ulong old_msr = kvmppc_get_msr(vcpu);
340 
341 #ifdef EXIT_DEBUG
342 	printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
343 #endif
344 
345 	msr &= to_book3s(vcpu)->msr_mask;
346 	kvmppc_set_msr_fast(vcpu, msr);
347 	kvmppc_recalc_shadow_msr(vcpu);
348 
349 	if (msr & MSR_POW) {
350 		if (!vcpu->arch.pending_exceptions) {
351 			kvm_vcpu_block(vcpu);
352 			kvm_clear_request(KVM_REQ_UNHALT, vcpu);
353 			vcpu->stat.halt_wakeup++;
354 
355 			/* Unset POW bit after we woke up */
356 			msr &= ~MSR_POW;
357 			kvmppc_set_msr_fast(vcpu, msr);
358 		}
359 	}
360 
361 	if (kvmppc_is_split_real(vcpu))
362 		kvmppc_fixup_split_real(vcpu);
363 	else
364 		kvmppc_unfixup_split_real(vcpu);
365 
366 	if ((kvmppc_get_msr(vcpu) & (MSR_PR|MSR_IR|MSR_DR)) !=
367 		   (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
368 		kvmppc_mmu_flush_segments(vcpu);
369 		kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
370 
371 		/* Preload magic page segment when in kernel mode */
372 		if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
373 			struct kvm_vcpu_arch *a = &vcpu->arch;
374 
375 			if (msr & MSR_DR)
376 				kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
377 			else
378 				kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
379 		}
380 	}
381 
382 	/*
383 	 * When switching from 32 to 64-bit, we may have a stale 32-bit
384 	 * magic page around, we need to flush it. Typically 32-bit magic
385 	 * page will be instanciated when calling into RTAS. Note: We
386 	 * assume that such transition only happens while in kernel mode,
387 	 * ie, we never transition from user 32-bit to kernel 64-bit with
388 	 * a 32-bit magic page around.
389 	 */
390 	if (vcpu->arch.magic_page_pa &&
391 	    !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
392 		/* going from RTAS to normal kernel code */
393 		kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
394 				     ~0xFFFUL);
395 	}
396 
397 	/* Preload FPU if it's enabled */
398 	if (kvmppc_get_msr(vcpu) & MSR_FP)
399 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
400 }
401 
402 void kvmppc_set_pvr_pr(struct kvm_vcpu *vcpu, u32 pvr)
403 {
404 	u32 host_pvr;
405 
406 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
407 	vcpu->arch.pvr = pvr;
408 #ifdef CONFIG_PPC_BOOK3S_64
409 	if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
410 		kvmppc_mmu_book3s_64_init(vcpu);
411 		if (!to_book3s(vcpu)->hior_explicit)
412 			to_book3s(vcpu)->hior = 0xfff00000;
413 		to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
414 		vcpu->arch.cpu_type = KVM_CPU_3S_64;
415 	} else
416 #endif
417 	{
418 		kvmppc_mmu_book3s_32_init(vcpu);
419 		if (!to_book3s(vcpu)->hior_explicit)
420 			to_book3s(vcpu)->hior = 0;
421 		to_book3s(vcpu)->msr_mask = 0xffffffffULL;
422 		vcpu->arch.cpu_type = KVM_CPU_3S_32;
423 	}
424 
425 	kvmppc_sanity_check(vcpu);
426 
427 	/* If we are in hypervisor level on 970, we can tell the CPU to
428 	 * treat DCBZ as 32 bytes store */
429 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
430 	if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
431 	    !strcmp(cur_cpu_spec->platform, "ppc970"))
432 		vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
433 
434 	/* Cell performs badly if MSR_FEx are set. So let's hope nobody
435 	   really needs them in a VM on Cell and force disable them. */
436 	if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
437 		to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
438 
439 	/*
440 	 * If they're asking for POWER6 or later, set the flag
441 	 * indicating that we can do multiple large page sizes
442 	 * and 1TB segments.
443 	 * Also set the flag that indicates that tlbie has the large
444 	 * page bit in the RB operand instead of the instruction.
445 	 */
446 	switch (PVR_VER(pvr)) {
447 	case PVR_POWER6:
448 	case PVR_POWER7:
449 	case PVR_POWER7p:
450 	case PVR_POWER8:
451 	case PVR_POWER8E:
452 	case PVR_POWER8NVL:
453 		vcpu->arch.hflags |= BOOK3S_HFLAG_MULTI_PGSIZE |
454 			BOOK3S_HFLAG_NEW_TLBIE;
455 		break;
456 	}
457 
458 #ifdef CONFIG_PPC_BOOK3S_32
459 	/* 32 bit Book3S always has 32 byte dcbz */
460 	vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
461 #endif
462 
463 	/* On some CPUs we can execute paired single operations natively */
464 	asm ( "mfpvr %0" : "=r"(host_pvr));
465 	switch (host_pvr) {
466 	case 0x00080200:	/* lonestar 2.0 */
467 	case 0x00088202:	/* lonestar 2.2 */
468 	case 0x70000100:	/* gekko 1.0 */
469 	case 0x00080100:	/* gekko 2.0 */
470 	case 0x00083203:	/* gekko 2.3a */
471 	case 0x00083213:	/* gekko 2.3b */
472 	case 0x00083204:	/* gekko 2.4 */
473 	case 0x00083214:	/* gekko 2.4e (8SE) - retail HW2 */
474 	case 0x00087200:	/* broadway */
475 		vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
476 		/* Enable HID2.PSE - in case we need it later */
477 		mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
478 	}
479 }
480 
481 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
482  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
483  * emulate 32 bytes dcbz length.
484  *
485  * The Book3s_64 inventors also realized this case and implemented a special bit
486  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
487  *
488  * My approach here is to patch the dcbz instruction on executing pages.
489  */
490 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
491 {
492 	struct page *hpage;
493 	u64 hpage_offset;
494 	u32 *page;
495 	int i;
496 
497 	hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
498 	if (is_error_page(hpage))
499 		return;
500 
501 	hpage_offset = pte->raddr & ~PAGE_MASK;
502 	hpage_offset &= ~0xFFFULL;
503 	hpage_offset /= 4;
504 
505 	get_page(hpage);
506 	page = kmap_atomic(hpage);
507 
508 	/* patch dcbz into reserved instruction, so we trap */
509 	for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
510 		if ((be32_to_cpu(page[i]) & 0xff0007ff) == INS_DCBZ)
511 			page[i] &= cpu_to_be32(0xfffffff7);
512 
513 	kunmap_atomic(page);
514 	put_page(hpage);
515 }
516 
517 static bool kvmppc_visible_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
518 {
519 	ulong mp_pa = vcpu->arch.magic_page_pa;
520 
521 	if (!(kvmppc_get_msr(vcpu) & MSR_SF))
522 		mp_pa = (uint32_t)mp_pa;
523 
524 	gpa &= ~0xFFFULL;
525 	if (unlikely(mp_pa) && unlikely((mp_pa & KVM_PAM) == (gpa & KVM_PAM))) {
526 		return true;
527 	}
528 
529 	return kvm_is_visible_gfn(vcpu->kvm, gpa >> PAGE_SHIFT);
530 }
531 
532 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
533 			    ulong eaddr, int vec)
534 {
535 	bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
536 	bool iswrite = false;
537 	int r = RESUME_GUEST;
538 	int relocated;
539 	int page_found = 0;
540 	struct kvmppc_pte pte = { 0 };
541 	bool dr = (kvmppc_get_msr(vcpu) & MSR_DR) ? true : false;
542 	bool ir = (kvmppc_get_msr(vcpu) & MSR_IR) ? true : false;
543 	u64 vsid;
544 
545 	relocated = data ? dr : ir;
546 	if (data && (vcpu->arch.fault_dsisr & DSISR_ISSTORE))
547 		iswrite = true;
548 
549 	/* Resolve real address if translation turned on */
550 	if (relocated) {
551 		page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data, iswrite);
552 	} else {
553 		pte.may_execute = true;
554 		pte.may_read = true;
555 		pte.may_write = true;
556 		pte.raddr = eaddr & KVM_PAM;
557 		pte.eaddr = eaddr;
558 		pte.vpage = eaddr >> 12;
559 		pte.page_size = MMU_PAGE_64K;
560 	}
561 
562 	switch (kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) {
563 	case 0:
564 		pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
565 		break;
566 	case MSR_DR:
567 		if (!data &&
568 		    (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK) &&
569 		    ((pte.raddr & SPLIT_HACK_MASK) == SPLIT_HACK_OFFS))
570 			pte.raddr &= ~SPLIT_HACK_MASK;
571 		/* fall through */
572 	case MSR_IR:
573 		vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
574 
575 		if ((kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) == MSR_DR)
576 			pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
577 		else
578 			pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
579 		pte.vpage |= vsid;
580 
581 		if (vsid == -1)
582 			page_found = -EINVAL;
583 		break;
584 	}
585 
586 	if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
587 	   (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
588 		/*
589 		 * If we do the dcbz hack, we have to NX on every execution,
590 		 * so we can patch the executing code. This renders our guest
591 		 * NX-less.
592 		 */
593 		pte.may_execute = !data;
594 	}
595 
596 	if (page_found == -ENOENT) {
597 		/* Page not found in guest PTE entries */
598 		u64 ssrr1 = vcpu->arch.shadow_srr1;
599 		u64 msr = kvmppc_get_msr(vcpu);
600 		kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
601 		kvmppc_set_dsisr(vcpu, vcpu->arch.fault_dsisr);
602 		kvmppc_set_msr_fast(vcpu, msr | (ssrr1 & 0xf8000000ULL));
603 		kvmppc_book3s_queue_irqprio(vcpu, vec);
604 	} else if (page_found == -EPERM) {
605 		/* Storage protection */
606 		u32 dsisr = vcpu->arch.fault_dsisr;
607 		u64 ssrr1 = vcpu->arch.shadow_srr1;
608 		u64 msr = kvmppc_get_msr(vcpu);
609 		kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
610 		dsisr = (dsisr & ~DSISR_NOHPTE) | DSISR_PROTFAULT;
611 		kvmppc_set_dsisr(vcpu, dsisr);
612 		kvmppc_set_msr_fast(vcpu, msr | (ssrr1 & 0xf8000000ULL));
613 		kvmppc_book3s_queue_irqprio(vcpu, vec);
614 	} else if (page_found == -EINVAL) {
615 		/* Page not found in guest SLB */
616 		kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
617 		kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
618 	} else if (kvmppc_visible_gpa(vcpu, pte.raddr)) {
619 		if (data && !(vcpu->arch.fault_dsisr & DSISR_NOHPTE)) {
620 			/*
621 			 * There is already a host HPTE there, presumably
622 			 * a read-only one for a page the guest thinks
623 			 * is writable, so get rid of it first.
624 			 */
625 			kvmppc_mmu_unmap_page(vcpu, &pte);
626 		}
627 		/* The guest's PTE is not mapped yet. Map on the host */
628 		if (kvmppc_mmu_map_page(vcpu, &pte, iswrite) == -EIO) {
629 			/* Exit KVM if mapping failed */
630 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
631 			return RESUME_HOST;
632 		}
633 		if (data)
634 			vcpu->stat.sp_storage++;
635 		else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
636 			 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
637 			kvmppc_patch_dcbz(vcpu, &pte);
638 	} else {
639 		/* MMIO */
640 		vcpu->stat.mmio_exits++;
641 		vcpu->arch.paddr_accessed = pte.raddr;
642 		vcpu->arch.vaddr_accessed = pte.eaddr;
643 		r = kvmppc_emulate_mmio(run, vcpu);
644 		if ( r == RESUME_HOST_NV )
645 			r = RESUME_HOST;
646 	}
647 
648 	return r;
649 }
650 
651 /* Give up external provider (FPU, Altivec, VSX) */
652 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
653 {
654 	struct thread_struct *t = &current->thread;
655 
656 	/*
657 	 * VSX instructions can access FP and vector registers, so if
658 	 * we are giving up VSX, make sure we give up FP and VMX as well.
659 	 */
660 	if (msr & MSR_VSX)
661 		msr |= MSR_FP | MSR_VEC;
662 
663 	msr &= vcpu->arch.guest_owned_ext;
664 	if (!msr)
665 		return;
666 
667 #ifdef DEBUG_EXT
668 	printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
669 #endif
670 
671 	if (msr & MSR_FP) {
672 		/*
673 		 * Note that on CPUs with VSX, giveup_fpu stores
674 		 * both the traditional FP registers and the added VSX
675 		 * registers into thread.fp_state.fpr[].
676 		 */
677 		if (t->regs->msr & MSR_FP)
678 			giveup_fpu(current);
679 		t->fp_save_area = NULL;
680 	}
681 
682 #ifdef CONFIG_ALTIVEC
683 	if (msr & MSR_VEC) {
684 		if (current->thread.regs->msr & MSR_VEC)
685 			giveup_altivec(current);
686 		t->vr_save_area = NULL;
687 	}
688 #endif
689 
690 	vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
691 	kvmppc_recalc_shadow_msr(vcpu);
692 }
693 
694 /* Give up facility (TAR / EBB / DSCR) */
695 static void kvmppc_giveup_fac(struct kvm_vcpu *vcpu, ulong fac)
696 {
697 #ifdef CONFIG_PPC_BOOK3S_64
698 	if (!(vcpu->arch.shadow_fscr & (1ULL << fac))) {
699 		/* Facility not available to the guest, ignore giveup request*/
700 		return;
701 	}
702 
703 	switch (fac) {
704 	case FSCR_TAR_LG:
705 		vcpu->arch.tar = mfspr(SPRN_TAR);
706 		mtspr(SPRN_TAR, current->thread.tar);
707 		vcpu->arch.shadow_fscr &= ~FSCR_TAR;
708 		break;
709 	}
710 #endif
711 }
712 
713 /* Handle external providers (FPU, Altivec, VSX) */
714 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
715 			     ulong msr)
716 {
717 	struct thread_struct *t = &current->thread;
718 
719 	/* When we have paired singles, we emulate in software */
720 	if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
721 		return RESUME_GUEST;
722 
723 	if (!(kvmppc_get_msr(vcpu) & msr)) {
724 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
725 		return RESUME_GUEST;
726 	}
727 
728 	if (msr == MSR_VSX) {
729 		/* No VSX?  Give an illegal instruction interrupt */
730 #ifdef CONFIG_VSX
731 		if (!cpu_has_feature(CPU_FTR_VSX))
732 #endif
733 		{
734 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
735 			return RESUME_GUEST;
736 		}
737 
738 		/*
739 		 * We have to load up all the FP and VMX registers before
740 		 * we can let the guest use VSX instructions.
741 		 */
742 		msr = MSR_FP | MSR_VEC | MSR_VSX;
743 	}
744 
745 	/* See if we already own all the ext(s) needed */
746 	msr &= ~vcpu->arch.guest_owned_ext;
747 	if (!msr)
748 		return RESUME_GUEST;
749 
750 #ifdef DEBUG_EXT
751 	printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
752 #endif
753 
754 	if (msr & MSR_FP) {
755 		preempt_disable();
756 		enable_kernel_fp();
757 		load_fp_state(&vcpu->arch.fp);
758 		disable_kernel_fp();
759 		t->fp_save_area = &vcpu->arch.fp;
760 		preempt_enable();
761 	}
762 
763 	if (msr & MSR_VEC) {
764 #ifdef CONFIG_ALTIVEC
765 		preempt_disable();
766 		enable_kernel_altivec();
767 		load_vr_state(&vcpu->arch.vr);
768 		disable_kernel_altivec();
769 		t->vr_save_area = &vcpu->arch.vr;
770 		preempt_enable();
771 #endif
772 	}
773 
774 	t->regs->msr |= msr;
775 	vcpu->arch.guest_owned_ext |= msr;
776 	kvmppc_recalc_shadow_msr(vcpu);
777 
778 	return RESUME_GUEST;
779 }
780 
781 /*
782  * Kernel code using FP or VMX could have flushed guest state to
783  * the thread_struct; if so, get it back now.
784  */
785 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
786 {
787 	unsigned long lost_ext;
788 
789 	lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
790 	if (!lost_ext)
791 		return;
792 
793 	if (lost_ext & MSR_FP) {
794 		preempt_disable();
795 		enable_kernel_fp();
796 		load_fp_state(&vcpu->arch.fp);
797 		disable_kernel_fp();
798 		preempt_enable();
799 	}
800 #ifdef CONFIG_ALTIVEC
801 	if (lost_ext & MSR_VEC) {
802 		preempt_disable();
803 		enable_kernel_altivec();
804 		load_vr_state(&vcpu->arch.vr);
805 		disable_kernel_altivec();
806 		preempt_enable();
807 	}
808 #endif
809 	current->thread.regs->msr |= lost_ext;
810 }
811 
812 #ifdef CONFIG_PPC_BOOK3S_64
813 
814 static void kvmppc_trigger_fac_interrupt(struct kvm_vcpu *vcpu, ulong fac)
815 {
816 	/* Inject the Interrupt Cause field and trigger a guest interrupt */
817 	vcpu->arch.fscr &= ~(0xffULL << 56);
818 	vcpu->arch.fscr |= (fac << 56);
819 	kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_FAC_UNAVAIL);
820 }
821 
822 static void kvmppc_emulate_fac(struct kvm_vcpu *vcpu, ulong fac)
823 {
824 	enum emulation_result er = EMULATE_FAIL;
825 
826 	if (!(kvmppc_get_msr(vcpu) & MSR_PR))
827 		er = kvmppc_emulate_instruction(vcpu->run, vcpu);
828 
829 	if ((er != EMULATE_DONE) && (er != EMULATE_AGAIN)) {
830 		/* Couldn't emulate, trigger interrupt in guest */
831 		kvmppc_trigger_fac_interrupt(vcpu, fac);
832 	}
833 }
834 
835 /* Enable facilities (TAR, EBB, DSCR) for the guest */
836 static int kvmppc_handle_fac(struct kvm_vcpu *vcpu, ulong fac)
837 {
838 	bool guest_fac_enabled;
839 	BUG_ON(!cpu_has_feature(CPU_FTR_ARCH_207S));
840 
841 	/*
842 	 * Not every facility is enabled by FSCR bits, check whether the
843 	 * guest has this facility enabled at all.
844 	 */
845 	switch (fac) {
846 	case FSCR_TAR_LG:
847 	case FSCR_EBB_LG:
848 		guest_fac_enabled = (vcpu->arch.fscr & (1ULL << fac));
849 		break;
850 	case FSCR_TM_LG:
851 		guest_fac_enabled = kvmppc_get_msr(vcpu) & MSR_TM;
852 		break;
853 	default:
854 		guest_fac_enabled = false;
855 		break;
856 	}
857 
858 	if (!guest_fac_enabled) {
859 		/* Facility not enabled by the guest */
860 		kvmppc_trigger_fac_interrupt(vcpu, fac);
861 		return RESUME_GUEST;
862 	}
863 
864 	switch (fac) {
865 	case FSCR_TAR_LG:
866 		/* TAR switching isn't lazy in Linux yet */
867 		current->thread.tar = mfspr(SPRN_TAR);
868 		mtspr(SPRN_TAR, vcpu->arch.tar);
869 		vcpu->arch.shadow_fscr |= FSCR_TAR;
870 		break;
871 	default:
872 		kvmppc_emulate_fac(vcpu, fac);
873 		break;
874 	}
875 
876 	return RESUME_GUEST;
877 }
878 
879 void kvmppc_set_fscr(struct kvm_vcpu *vcpu, u64 fscr)
880 {
881 	if ((vcpu->arch.fscr & FSCR_TAR) && !(fscr & FSCR_TAR)) {
882 		/* TAR got dropped, drop it in shadow too */
883 		kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
884 	}
885 	vcpu->arch.fscr = fscr;
886 }
887 #endif
888 
889 static void kvmppc_setup_debug(struct kvm_vcpu *vcpu)
890 {
891 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
892 		u64 msr = kvmppc_get_msr(vcpu);
893 
894 		kvmppc_set_msr(vcpu, msr | MSR_SE);
895 	}
896 }
897 
898 static void kvmppc_clear_debug(struct kvm_vcpu *vcpu)
899 {
900 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
901 		u64 msr = kvmppc_get_msr(vcpu);
902 
903 		kvmppc_set_msr(vcpu, msr & ~MSR_SE);
904 	}
905 }
906 
907 static int kvmppc_exit_pr_progint(struct kvm_run *run, struct kvm_vcpu *vcpu,
908 				  unsigned int exit_nr)
909 {
910 	enum emulation_result er;
911 	ulong flags;
912 	u32 last_inst;
913 	int emul, r;
914 
915 	/*
916 	 * shadow_srr1 only contains valid flags if we came here via a program
917 	 * exception. The other exceptions (emulation assist, FP unavailable,
918 	 * etc.) do not provide flags in SRR1, so use an illegal-instruction
919 	 * exception when injecting a program interrupt into the guest.
920 	 */
921 	if (exit_nr == BOOK3S_INTERRUPT_PROGRAM)
922 		flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
923 	else
924 		flags = SRR1_PROGILL;
925 
926 	emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
927 	if (emul != EMULATE_DONE)
928 		return RESUME_GUEST;
929 
930 	if (kvmppc_get_msr(vcpu) & MSR_PR) {
931 #ifdef EXIT_DEBUG
932 		pr_info("Userspace triggered 0x700 exception at\n 0x%lx (0x%x)\n",
933 			kvmppc_get_pc(vcpu), last_inst);
934 #endif
935 		if ((last_inst & 0xff0007ff) != (INS_DCBZ & 0xfffffff7)) {
936 			kvmppc_core_queue_program(vcpu, flags);
937 			return RESUME_GUEST;
938 		}
939 	}
940 
941 	vcpu->stat.emulated_inst_exits++;
942 	er = kvmppc_emulate_instruction(run, vcpu);
943 	switch (er) {
944 	case EMULATE_DONE:
945 		r = RESUME_GUEST_NV;
946 		break;
947 	case EMULATE_AGAIN:
948 		r = RESUME_GUEST;
949 		break;
950 	case EMULATE_FAIL:
951 		pr_crit("%s: emulation at %lx failed (%08x)\n",
952 			__func__, kvmppc_get_pc(vcpu), last_inst);
953 		kvmppc_core_queue_program(vcpu, flags);
954 		r = RESUME_GUEST;
955 		break;
956 	case EMULATE_DO_MMIO:
957 		run->exit_reason = KVM_EXIT_MMIO;
958 		r = RESUME_HOST_NV;
959 		break;
960 	case EMULATE_EXIT_USER:
961 		r = RESUME_HOST_NV;
962 		break;
963 	default:
964 		BUG();
965 	}
966 
967 	return r;
968 }
969 
970 int kvmppc_handle_exit_pr(struct kvm_run *run, struct kvm_vcpu *vcpu,
971 			  unsigned int exit_nr)
972 {
973 	int r = RESUME_HOST;
974 	int s;
975 
976 	vcpu->stat.sum_exits++;
977 
978 	run->exit_reason = KVM_EXIT_UNKNOWN;
979 	run->ready_for_interrupt_injection = 1;
980 
981 	/* We get here with MSR.EE=1 */
982 
983 	trace_kvm_exit(exit_nr, vcpu);
984 	guest_exit();
985 
986 	switch (exit_nr) {
987 	case BOOK3S_INTERRUPT_INST_STORAGE:
988 	{
989 		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
990 		vcpu->stat.pf_instruc++;
991 
992 		if (kvmppc_is_split_real(vcpu))
993 			kvmppc_fixup_split_real(vcpu);
994 
995 #ifdef CONFIG_PPC_BOOK3S_32
996 		/* We set segments as unused segments when invalidating them. So
997 		 * treat the respective fault as segment fault. */
998 		{
999 			struct kvmppc_book3s_shadow_vcpu *svcpu;
1000 			u32 sr;
1001 
1002 			svcpu = svcpu_get(vcpu);
1003 			sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
1004 			svcpu_put(svcpu);
1005 			if (sr == SR_INVALID) {
1006 				kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
1007 				r = RESUME_GUEST;
1008 				break;
1009 			}
1010 		}
1011 #endif
1012 
1013 		/* only care about PTEG not found errors, but leave NX alone */
1014 		if (shadow_srr1 & 0x40000000) {
1015 			int idx = srcu_read_lock(&vcpu->kvm->srcu);
1016 			r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
1017 			srcu_read_unlock(&vcpu->kvm->srcu, idx);
1018 			vcpu->stat.sp_instruc++;
1019 		} else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
1020 			  (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
1021 			/*
1022 			 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
1023 			 *     so we can't use the NX bit inside the guest. Let's cross our fingers,
1024 			 *     that no guest that needs the dcbz hack does NX.
1025 			 */
1026 			kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
1027 			r = RESUME_GUEST;
1028 		} else {
1029 			u64 msr = kvmppc_get_msr(vcpu);
1030 			msr |= shadow_srr1 & 0x58000000;
1031 			kvmppc_set_msr_fast(vcpu, msr);
1032 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1033 			r = RESUME_GUEST;
1034 		}
1035 		break;
1036 	}
1037 	case BOOK3S_INTERRUPT_DATA_STORAGE:
1038 	{
1039 		ulong dar = kvmppc_get_fault_dar(vcpu);
1040 		u32 fault_dsisr = vcpu->arch.fault_dsisr;
1041 		vcpu->stat.pf_storage++;
1042 
1043 #ifdef CONFIG_PPC_BOOK3S_32
1044 		/* We set segments as unused segments when invalidating them. So
1045 		 * treat the respective fault as segment fault. */
1046 		{
1047 			struct kvmppc_book3s_shadow_vcpu *svcpu;
1048 			u32 sr;
1049 
1050 			svcpu = svcpu_get(vcpu);
1051 			sr = svcpu->sr[dar >> SID_SHIFT];
1052 			svcpu_put(svcpu);
1053 			if (sr == SR_INVALID) {
1054 				kvmppc_mmu_map_segment(vcpu, dar);
1055 				r = RESUME_GUEST;
1056 				break;
1057 			}
1058 		}
1059 #endif
1060 
1061 		/*
1062 		 * We need to handle missing shadow PTEs, and
1063 		 * protection faults due to us mapping a page read-only
1064 		 * when the guest thinks it is writable.
1065 		 */
1066 		if (fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT)) {
1067 			int idx = srcu_read_lock(&vcpu->kvm->srcu);
1068 			r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
1069 			srcu_read_unlock(&vcpu->kvm->srcu, idx);
1070 		} else {
1071 			kvmppc_set_dar(vcpu, dar);
1072 			kvmppc_set_dsisr(vcpu, fault_dsisr);
1073 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1074 			r = RESUME_GUEST;
1075 		}
1076 		break;
1077 	}
1078 	case BOOK3S_INTERRUPT_DATA_SEGMENT:
1079 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
1080 			kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
1081 			kvmppc_book3s_queue_irqprio(vcpu,
1082 				BOOK3S_INTERRUPT_DATA_SEGMENT);
1083 		}
1084 		r = RESUME_GUEST;
1085 		break;
1086 	case BOOK3S_INTERRUPT_INST_SEGMENT:
1087 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
1088 			kvmppc_book3s_queue_irqprio(vcpu,
1089 				BOOK3S_INTERRUPT_INST_SEGMENT);
1090 		}
1091 		r = RESUME_GUEST;
1092 		break;
1093 	/* We're good on these - the host merely wanted to get our attention */
1094 	case BOOK3S_INTERRUPT_DECREMENTER:
1095 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1096 	case BOOK3S_INTERRUPT_DOORBELL:
1097 	case BOOK3S_INTERRUPT_H_DOORBELL:
1098 		vcpu->stat.dec_exits++;
1099 		r = RESUME_GUEST;
1100 		break;
1101 	case BOOK3S_INTERRUPT_EXTERNAL:
1102 	case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
1103 	case BOOK3S_INTERRUPT_EXTERNAL_HV:
1104 		vcpu->stat.ext_intr_exits++;
1105 		r = RESUME_GUEST;
1106 		break;
1107 	case BOOK3S_INTERRUPT_PERFMON:
1108 		r = RESUME_GUEST;
1109 		break;
1110 	case BOOK3S_INTERRUPT_PROGRAM:
1111 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1112 		r = kvmppc_exit_pr_progint(run, vcpu, exit_nr);
1113 		break;
1114 	case BOOK3S_INTERRUPT_SYSCALL:
1115 	{
1116 		u32 last_sc;
1117 		int emul;
1118 
1119 		/* Get last sc for papr */
1120 		if (vcpu->arch.papr_enabled) {
1121 			/* The sc instuction points SRR0 to the next inst */
1122 			emul = kvmppc_get_last_inst(vcpu, INST_SC, &last_sc);
1123 			if (emul != EMULATE_DONE) {
1124 				kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) - 4);
1125 				r = RESUME_GUEST;
1126 				break;
1127 			}
1128 		}
1129 
1130 		if (vcpu->arch.papr_enabled &&
1131 		    (last_sc == 0x44000022) &&
1132 		    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
1133 			/* SC 1 papr hypercalls */
1134 			ulong cmd = kvmppc_get_gpr(vcpu, 3);
1135 			int i;
1136 
1137 #ifdef CONFIG_PPC_BOOK3S_64
1138 			if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
1139 				r = RESUME_GUEST;
1140 				break;
1141 			}
1142 #endif
1143 
1144 			run->papr_hcall.nr = cmd;
1145 			for (i = 0; i < 9; ++i) {
1146 				ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
1147 				run->papr_hcall.args[i] = gpr;
1148 			}
1149 			run->exit_reason = KVM_EXIT_PAPR_HCALL;
1150 			vcpu->arch.hcall_needed = 1;
1151 			r = RESUME_HOST;
1152 		} else if (vcpu->arch.osi_enabled &&
1153 		    (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
1154 		    (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
1155 			/* MOL hypercalls */
1156 			u64 *gprs = run->osi.gprs;
1157 			int i;
1158 
1159 			run->exit_reason = KVM_EXIT_OSI;
1160 			for (i = 0; i < 32; i++)
1161 				gprs[i] = kvmppc_get_gpr(vcpu, i);
1162 			vcpu->arch.osi_needed = 1;
1163 			r = RESUME_HOST_NV;
1164 		} else if (!(kvmppc_get_msr(vcpu) & MSR_PR) &&
1165 		    (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
1166 			/* KVM PV hypercalls */
1167 			kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
1168 			r = RESUME_GUEST;
1169 		} else {
1170 			/* Guest syscalls */
1171 			vcpu->stat.syscall_exits++;
1172 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1173 			r = RESUME_GUEST;
1174 		}
1175 		break;
1176 	}
1177 	case BOOK3S_INTERRUPT_FP_UNAVAIL:
1178 	case BOOK3S_INTERRUPT_ALTIVEC:
1179 	case BOOK3S_INTERRUPT_VSX:
1180 	{
1181 		int ext_msr = 0;
1182 		int emul;
1183 		u32 last_inst;
1184 
1185 		if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE) {
1186 			/* Do paired single instruction emulation */
1187 			emul = kvmppc_get_last_inst(vcpu, INST_GENERIC,
1188 						    &last_inst);
1189 			if (emul == EMULATE_DONE)
1190 				r = kvmppc_exit_pr_progint(run, vcpu, exit_nr);
1191 			else
1192 				r = RESUME_GUEST;
1193 
1194 			break;
1195 		}
1196 
1197 		/* Enable external provider */
1198 		switch (exit_nr) {
1199 		case BOOK3S_INTERRUPT_FP_UNAVAIL:
1200 			ext_msr = MSR_FP;
1201 			break;
1202 
1203 		case BOOK3S_INTERRUPT_ALTIVEC:
1204 			ext_msr = MSR_VEC;
1205 			break;
1206 
1207 		case BOOK3S_INTERRUPT_VSX:
1208 			ext_msr = MSR_VSX;
1209 			break;
1210 		}
1211 
1212 		r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
1213 		break;
1214 	}
1215 	case BOOK3S_INTERRUPT_ALIGNMENT:
1216 	{
1217 		u32 last_inst;
1218 		int emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1219 
1220 		if (emul == EMULATE_DONE) {
1221 			u32 dsisr;
1222 			u64 dar;
1223 
1224 			dsisr = kvmppc_alignment_dsisr(vcpu, last_inst);
1225 			dar = kvmppc_alignment_dar(vcpu, last_inst);
1226 
1227 			kvmppc_set_dsisr(vcpu, dsisr);
1228 			kvmppc_set_dar(vcpu, dar);
1229 
1230 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1231 		}
1232 		r = RESUME_GUEST;
1233 		break;
1234 	}
1235 #ifdef CONFIG_PPC_BOOK3S_64
1236 	case BOOK3S_INTERRUPT_FAC_UNAVAIL:
1237 		kvmppc_handle_fac(vcpu, vcpu->arch.shadow_fscr >> 56);
1238 		r = RESUME_GUEST;
1239 		break;
1240 #endif
1241 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1242 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1243 		r = RESUME_GUEST;
1244 		break;
1245 	case BOOK3S_INTERRUPT_TRACE:
1246 		if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
1247 			run->exit_reason = KVM_EXIT_DEBUG;
1248 			r = RESUME_HOST;
1249 		} else {
1250 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1251 			r = RESUME_GUEST;
1252 		}
1253 		break;
1254 	default:
1255 	{
1256 		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1257 		/* Ugh - bork here! What did we get? */
1258 		printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1259 			exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
1260 		r = RESUME_HOST;
1261 		BUG();
1262 		break;
1263 	}
1264 	}
1265 
1266 	if (!(r & RESUME_HOST)) {
1267 		/* To avoid clobbering exit_reason, only check for signals if
1268 		 * we aren't already exiting to userspace for some other
1269 		 * reason. */
1270 
1271 		/*
1272 		 * Interrupts could be timers for the guest which we have to
1273 		 * inject again, so let's postpone them until we're in the guest
1274 		 * and if we really did time things so badly, then we just exit
1275 		 * again due to a host external interrupt.
1276 		 */
1277 		s = kvmppc_prepare_to_enter(vcpu);
1278 		if (s <= 0)
1279 			r = s;
1280 		else {
1281 			/* interrupts now hard-disabled */
1282 			kvmppc_fix_ee_before_entry();
1283 		}
1284 
1285 		kvmppc_handle_lost_ext(vcpu);
1286 	}
1287 
1288 	trace_kvm_book3s_reenter(r, vcpu);
1289 
1290 	return r;
1291 }
1292 
1293 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu *vcpu,
1294 					    struct kvm_sregs *sregs)
1295 {
1296 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1297 	int i;
1298 
1299 	sregs->pvr = vcpu->arch.pvr;
1300 
1301 	sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1302 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1303 		for (i = 0; i < 64; i++) {
1304 			sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
1305 			sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1306 		}
1307 	} else {
1308 		for (i = 0; i < 16; i++)
1309 			sregs->u.s.ppc32.sr[i] = kvmppc_get_sr(vcpu, i);
1310 
1311 		for (i = 0; i < 8; i++) {
1312 			sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1313 			sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1314 		}
1315 	}
1316 
1317 	return 0;
1318 }
1319 
1320 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu *vcpu,
1321 					    struct kvm_sregs *sregs)
1322 {
1323 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1324 	int i;
1325 
1326 	kvmppc_set_pvr_pr(vcpu, sregs->pvr);
1327 
1328 	vcpu3s->sdr1 = sregs->u.s.sdr1;
1329 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1330 		for (i = 0; i < 64; i++) {
1331 			vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1332 						    sregs->u.s.ppc64.slb[i].slbe);
1333 		}
1334 	} else {
1335 		for (i = 0; i < 16; i++) {
1336 			vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1337 		}
1338 		for (i = 0; i < 8; i++) {
1339 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1340 				       (u32)sregs->u.s.ppc32.ibat[i]);
1341 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1342 				       (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1343 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1344 				       (u32)sregs->u.s.ppc32.dbat[i]);
1345 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1346 				       (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1347 		}
1348 	}
1349 
1350 	/* Flush the MMU after messing with the segments */
1351 	kvmppc_mmu_pte_flush(vcpu, 0, 0);
1352 
1353 	return 0;
1354 }
1355 
1356 static int kvmppc_get_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1357 				 union kvmppc_one_reg *val)
1358 {
1359 	int r = 0;
1360 
1361 	switch (id) {
1362 	case KVM_REG_PPC_DEBUG_INST:
1363 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1364 		break;
1365 	case KVM_REG_PPC_HIOR:
1366 		*val = get_reg_val(id, to_book3s(vcpu)->hior);
1367 		break;
1368 	case KVM_REG_PPC_VTB:
1369 		*val = get_reg_val(id, to_book3s(vcpu)->vtb);
1370 		break;
1371 	case KVM_REG_PPC_LPCR:
1372 	case KVM_REG_PPC_LPCR_64:
1373 		/*
1374 		 * We are only interested in the LPCR_ILE bit
1375 		 */
1376 		if (vcpu->arch.intr_msr & MSR_LE)
1377 			*val = get_reg_val(id, LPCR_ILE);
1378 		else
1379 			*val = get_reg_val(id, 0);
1380 		break;
1381 	default:
1382 		r = -EINVAL;
1383 		break;
1384 	}
1385 
1386 	return r;
1387 }
1388 
1389 static void kvmppc_set_lpcr_pr(struct kvm_vcpu *vcpu, u64 new_lpcr)
1390 {
1391 	if (new_lpcr & LPCR_ILE)
1392 		vcpu->arch.intr_msr |= MSR_LE;
1393 	else
1394 		vcpu->arch.intr_msr &= ~MSR_LE;
1395 }
1396 
1397 static int kvmppc_set_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1398 				 union kvmppc_one_reg *val)
1399 {
1400 	int r = 0;
1401 
1402 	switch (id) {
1403 	case KVM_REG_PPC_HIOR:
1404 		to_book3s(vcpu)->hior = set_reg_val(id, *val);
1405 		to_book3s(vcpu)->hior_explicit = true;
1406 		break;
1407 	case KVM_REG_PPC_VTB:
1408 		to_book3s(vcpu)->vtb = set_reg_val(id, *val);
1409 		break;
1410 	case KVM_REG_PPC_LPCR:
1411 	case KVM_REG_PPC_LPCR_64:
1412 		kvmppc_set_lpcr_pr(vcpu, set_reg_val(id, *val));
1413 		break;
1414 	default:
1415 		r = -EINVAL;
1416 		break;
1417 	}
1418 
1419 	return r;
1420 }
1421 
1422 static struct kvm_vcpu *kvmppc_core_vcpu_create_pr(struct kvm *kvm,
1423 						   unsigned int id)
1424 {
1425 	struct kvmppc_vcpu_book3s *vcpu_book3s;
1426 	struct kvm_vcpu *vcpu;
1427 	int err = -ENOMEM;
1428 	unsigned long p;
1429 
1430 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1431 	if (!vcpu)
1432 		goto out;
1433 
1434 	vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1435 	if (!vcpu_book3s)
1436 		goto free_vcpu;
1437 	vcpu->arch.book3s = vcpu_book3s;
1438 
1439 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1440 	vcpu->arch.shadow_vcpu =
1441 		kzalloc(sizeof(*vcpu->arch.shadow_vcpu), GFP_KERNEL);
1442 	if (!vcpu->arch.shadow_vcpu)
1443 		goto free_vcpu3s;
1444 #endif
1445 
1446 	err = kvm_vcpu_init(vcpu, kvm, id);
1447 	if (err)
1448 		goto free_shadow_vcpu;
1449 
1450 	err = -ENOMEM;
1451 	p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1452 	if (!p)
1453 		goto uninit_vcpu;
1454 	vcpu->arch.shared = (void *)p;
1455 #ifdef CONFIG_PPC_BOOK3S_64
1456 	/* Always start the shared struct in native endian mode */
1457 #ifdef __BIG_ENDIAN__
1458         vcpu->arch.shared_big_endian = true;
1459 #else
1460         vcpu->arch.shared_big_endian = false;
1461 #endif
1462 
1463 	/*
1464 	 * Default to the same as the host if we're on sufficiently
1465 	 * recent machine that we have 1TB segments;
1466 	 * otherwise default to PPC970FX.
1467 	 */
1468 	vcpu->arch.pvr = 0x3C0301;
1469 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1470 		vcpu->arch.pvr = mfspr(SPRN_PVR);
1471 	vcpu->arch.intr_msr = MSR_SF;
1472 #else
1473 	/* default to book3s_32 (750) */
1474 	vcpu->arch.pvr = 0x84202;
1475 #endif
1476 	kvmppc_set_pvr_pr(vcpu, vcpu->arch.pvr);
1477 	vcpu->arch.slb_nr = 64;
1478 
1479 	vcpu->arch.shadow_msr = MSR_USER64 & ~MSR_LE;
1480 
1481 	err = kvmppc_mmu_init(vcpu);
1482 	if (err < 0)
1483 		goto uninit_vcpu;
1484 
1485 	return vcpu;
1486 
1487 uninit_vcpu:
1488 	kvm_vcpu_uninit(vcpu);
1489 free_shadow_vcpu:
1490 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1491 	kfree(vcpu->arch.shadow_vcpu);
1492 free_vcpu3s:
1493 #endif
1494 	vfree(vcpu_book3s);
1495 free_vcpu:
1496 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1497 out:
1498 	return ERR_PTR(err);
1499 }
1500 
1501 static void kvmppc_core_vcpu_free_pr(struct kvm_vcpu *vcpu)
1502 {
1503 	struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1504 
1505 	free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1506 	kvm_vcpu_uninit(vcpu);
1507 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1508 	kfree(vcpu->arch.shadow_vcpu);
1509 #endif
1510 	vfree(vcpu_book3s);
1511 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1512 }
1513 
1514 static int kvmppc_vcpu_run_pr(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1515 {
1516 	int ret;
1517 #ifdef CONFIG_ALTIVEC
1518 	unsigned long uninitialized_var(vrsave);
1519 #endif
1520 
1521 	/* Check if we can run the vcpu at all */
1522 	if (!vcpu->arch.sane) {
1523 		kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1524 		ret = -EINVAL;
1525 		goto out;
1526 	}
1527 
1528 	kvmppc_setup_debug(vcpu);
1529 
1530 	/*
1531 	 * Interrupts could be timers for the guest which we have to inject
1532 	 * again, so let's postpone them until we're in the guest and if we
1533 	 * really did time things so badly, then we just exit again due to
1534 	 * a host external interrupt.
1535 	 */
1536 	ret = kvmppc_prepare_to_enter(vcpu);
1537 	if (ret <= 0)
1538 		goto out;
1539 	/* interrupts now hard-disabled */
1540 
1541 	/* Save FPU, Altivec and VSX state */
1542 	giveup_all(current);
1543 
1544 	/* Preload FPU if it's enabled */
1545 	if (kvmppc_get_msr(vcpu) & MSR_FP)
1546 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1547 
1548 	kvmppc_fix_ee_before_entry();
1549 
1550 	ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1551 
1552 	kvmppc_clear_debug(vcpu);
1553 
1554 	/* No need for guest_exit. It's done in handle_exit.
1555 	   We also get here with interrupts enabled. */
1556 
1557 	/* Make sure we save the guest FPU/Altivec/VSX state */
1558 	kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1559 
1560 	/* Make sure we save the guest TAR/EBB/DSCR state */
1561 	kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
1562 
1563 out:
1564 	vcpu->mode = OUTSIDE_GUEST_MODE;
1565 	return ret;
1566 }
1567 
1568 /*
1569  * Get (and clear) the dirty memory log for a memory slot.
1570  */
1571 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm *kvm,
1572 					 struct kvm_dirty_log *log)
1573 {
1574 	struct kvm_memslots *slots;
1575 	struct kvm_memory_slot *memslot;
1576 	struct kvm_vcpu *vcpu;
1577 	ulong ga, ga_end;
1578 	int is_dirty = 0;
1579 	int r;
1580 	unsigned long n;
1581 
1582 	mutex_lock(&kvm->slots_lock);
1583 
1584 	r = kvm_get_dirty_log(kvm, log, &is_dirty);
1585 	if (r)
1586 		goto out;
1587 
1588 	/* If nothing is dirty, don't bother messing with page tables. */
1589 	if (is_dirty) {
1590 		slots = kvm_memslots(kvm);
1591 		memslot = id_to_memslot(slots, log->slot);
1592 
1593 		ga = memslot->base_gfn << PAGE_SHIFT;
1594 		ga_end = ga + (memslot->npages << PAGE_SHIFT);
1595 
1596 		kvm_for_each_vcpu(n, vcpu, kvm)
1597 			kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1598 
1599 		n = kvm_dirty_bitmap_bytes(memslot);
1600 		memset(memslot->dirty_bitmap, 0, n);
1601 	}
1602 
1603 	r = 0;
1604 out:
1605 	mutex_unlock(&kvm->slots_lock);
1606 	return r;
1607 }
1608 
1609 static void kvmppc_core_flush_memslot_pr(struct kvm *kvm,
1610 					 struct kvm_memory_slot *memslot)
1611 {
1612 	return;
1613 }
1614 
1615 static int kvmppc_core_prepare_memory_region_pr(struct kvm *kvm,
1616 					struct kvm_memory_slot *memslot,
1617 					const struct kvm_userspace_memory_region *mem)
1618 {
1619 	return 0;
1620 }
1621 
1622 static void kvmppc_core_commit_memory_region_pr(struct kvm *kvm,
1623 				const struct kvm_userspace_memory_region *mem,
1624 				const struct kvm_memory_slot *old,
1625 				const struct kvm_memory_slot *new)
1626 {
1627 	return;
1628 }
1629 
1630 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot *free,
1631 					struct kvm_memory_slot *dont)
1632 {
1633 	return;
1634 }
1635 
1636 static int kvmppc_core_create_memslot_pr(struct kvm_memory_slot *slot,
1637 					 unsigned long npages)
1638 {
1639 	return 0;
1640 }
1641 
1642 
1643 #ifdef CONFIG_PPC64
1644 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1645 					 struct kvm_ppc_smmu_info *info)
1646 {
1647 	long int i;
1648 	struct kvm_vcpu *vcpu;
1649 
1650 	info->flags = 0;
1651 
1652 	/* SLB is always 64 entries */
1653 	info->slb_size = 64;
1654 
1655 	/* Standard 4k base page size segment */
1656 	info->sps[0].page_shift = 12;
1657 	info->sps[0].slb_enc = 0;
1658 	info->sps[0].enc[0].page_shift = 12;
1659 	info->sps[0].enc[0].pte_enc = 0;
1660 
1661 	/*
1662 	 * 64k large page size.
1663 	 * We only want to put this in if the CPUs we're emulating
1664 	 * support it, but unfortunately we don't have a vcpu easily
1665 	 * to hand here to test.  Just pick the first vcpu, and if
1666 	 * that doesn't exist yet, report the minimum capability,
1667 	 * i.e., no 64k pages.
1668 	 * 1T segment support goes along with 64k pages.
1669 	 */
1670 	i = 1;
1671 	vcpu = kvm_get_vcpu(kvm, 0);
1672 	if (vcpu && (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
1673 		info->flags = KVM_PPC_1T_SEGMENTS;
1674 		info->sps[i].page_shift = 16;
1675 		info->sps[i].slb_enc = SLB_VSID_L | SLB_VSID_LP_01;
1676 		info->sps[i].enc[0].page_shift = 16;
1677 		info->sps[i].enc[0].pte_enc = 1;
1678 		++i;
1679 	}
1680 
1681 	/* Standard 16M large page size segment */
1682 	info->sps[i].page_shift = 24;
1683 	info->sps[i].slb_enc = SLB_VSID_L;
1684 	info->sps[i].enc[0].page_shift = 24;
1685 	info->sps[i].enc[0].pte_enc = 0;
1686 
1687 	return 0;
1688 }
1689 #else
1690 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1691 					 struct kvm_ppc_smmu_info *info)
1692 {
1693 	/* We should not get called */
1694 	BUG();
1695 }
1696 #endif /* CONFIG_PPC64 */
1697 
1698 static unsigned int kvm_global_user_count = 0;
1699 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1700 
1701 static int kvmppc_core_init_vm_pr(struct kvm *kvm)
1702 {
1703 	mutex_init(&kvm->arch.hpt_mutex);
1704 
1705 #ifdef CONFIG_PPC_BOOK3S_64
1706 	/* Start out with the default set of hcalls enabled */
1707 	kvmppc_pr_init_default_hcalls(kvm);
1708 #endif
1709 
1710 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1711 		spin_lock(&kvm_global_user_count_lock);
1712 		if (++kvm_global_user_count == 1)
1713 			pseries_disable_reloc_on_exc();
1714 		spin_unlock(&kvm_global_user_count_lock);
1715 	}
1716 	return 0;
1717 }
1718 
1719 static void kvmppc_core_destroy_vm_pr(struct kvm *kvm)
1720 {
1721 #ifdef CONFIG_PPC64
1722 	WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1723 #endif
1724 
1725 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1726 		spin_lock(&kvm_global_user_count_lock);
1727 		BUG_ON(kvm_global_user_count == 0);
1728 		if (--kvm_global_user_count == 0)
1729 			pseries_enable_reloc_on_exc();
1730 		spin_unlock(&kvm_global_user_count_lock);
1731 	}
1732 }
1733 
1734 static int kvmppc_core_check_processor_compat_pr(void)
1735 {
1736 	/*
1737 	 * Disable KVM for Power9 untill the required bits merged.
1738 	 */
1739 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1740 		return -EIO;
1741 	return 0;
1742 }
1743 
1744 static long kvm_arch_vm_ioctl_pr(struct file *filp,
1745 				 unsigned int ioctl, unsigned long arg)
1746 {
1747 	return -ENOTTY;
1748 }
1749 
1750 static struct kvmppc_ops kvm_ops_pr = {
1751 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_pr,
1752 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_pr,
1753 	.get_one_reg = kvmppc_get_one_reg_pr,
1754 	.set_one_reg = kvmppc_set_one_reg_pr,
1755 	.vcpu_load   = kvmppc_core_vcpu_load_pr,
1756 	.vcpu_put    = kvmppc_core_vcpu_put_pr,
1757 	.set_msr     = kvmppc_set_msr_pr,
1758 	.vcpu_run    = kvmppc_vcpu_run_pr,
1759 	.vcpu_create = kvmppc_core_vcpu_create_pr,
1760 	.vcpu_free   = kvmppc_core_vcpu_free_pr,
1761 	.check_requests = kvmppc_core_check_requests_pr,
1762 	.get_dirty_log = kvm_vm_ioctl_get_dirty_log_pr,
1763 	.flush_memslot = kvmppc_core_flush_memslot_pr,
1764 	.prepare_memory_region = kvmppc_core_prepare_memory_region_pr,
1765 	.commit_memory_region = kvmppc_core_commit_memory_region_pr,
1766 	.unmap_hva = kvm_unmap_hva_pr,
1767 	.unmap_hva_range = kvm_unmap_hva_range_pr,
1768 	.age_hva  = kvm_age_hva_pr,
1769 	.test_age_hva = kvm_test_age_hva_pr,
1770 	.set_spte_hva = kvm_set_spte_hva_pr,
1771 	.mmu_destroy  = kvmppc_mmu_destroy_pr,
1772 	.free_memslot = kvmppc_core_free_memslot_pr,
1773 	.create_memslot = kvmppc_core_create_memslot_pr,
1774 	.init_vm = kvmppc_core_init_vm_pr,
1775 	.destroy_vm = kvmppc_core_destroy_vm_pr,
1776 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_pr,
1777 	.emulate_op = kvmppc_core_emulate_op_pr,
1778 	.emulate_mtspr = kvmppc_core_emulate_mtspr_pr,
1779 	.emulate_mfspr = kvmppc_core_emulate_mfspr_pr,
1780 	.fast_vcpu_kick = kvm_vcpu_kick,
1781 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_pr,
1782 #ifdef CONFIG_PPC_BOOK3S_64
1783 	.hcall_implemented = kvmppc_hcall_impl_pr,
1784 #endif
1785 };
1786 
1787 
1788 int kvmppc_book3s_init_pr(void)
1789 {
1790 	int r;
1791 
1792 	r = kvmppc_core_check_processor_compat_pr();
1793 	if (r < 0)
1794 		return r;
1795 
1796 	kvm_ops_pr.owner = THIS_MODULE;
1797 	kvmppc_pr_ops = &kvm_ops_pr;
1798 
1799 	r = kvmppc_mmu_hpte_sysinit();
1800 	return r;
1801 }
1802 
1803 void kvmppc_book3s_exit_pr(void)
1804 {
1805 	kvmppc_pr_ops = NULL;
1806 	kvmppc_mmu_hpte_sysexit();
1807 }
1808 
1809 /*
1810  * We only support separate modules for book3s 64
1811  */
1812 #ifdef CONFIG_PPC_BOOK3S_64
1813 
1814 module_init(kvmppc_book3s_init_pr);
1815 module_exit(kvmppc_book3s_exit_pr);
1816 
1817 MODULE_LICENSE("GPL");
1818 MODULE_ALIAS_MISCDEV(KVM_MINOR);
1819 MODULE_ALIAS("devname:kvm");
1820 #endif
1821