xref: /openbmc/linux/arch/powerpc/kvm/book3s_pr.c (revision 82003e04)
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/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 			clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
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;
541 	bool is_mmio = false;
542 	bool dr = (kvmppc_get_msr(vcpu) & MSR_DR) ? true : false;
543 	bool ir = (kvmppc_get_msr(vcpu) & MSR_IR) ? true : false;
544 	u64 vsid;
545 
546 	relocated = data ? dr : ir;
547 	if (data && (vcpu->arch.fault_dsisr & DSISR_ISSTORE))
548 		iswrite = true;
549 
550 	/* Resolve real address if translation turned on */
551 	if (relocated) {
552 		page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data, iswrite);
553 	} else {
554 		pte.may_execute = true;
555 		pte.may_read = true;
556 		pte.may_write = true;
557 		pte.raddr = eaddr & KVM_PAM;
558 		pte.eaddr = eaddr;
559 		pte.vpage = eaddr >> 12;
560 		pte.page_size = MMU_PAGE_64K;
561 	}
562 
563 	switch (kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) {
564 	case 0:
565 		pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
566 		break;
567 	case MSR_DR:
568 		if (!data &&
569 		    (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK) &&
570 		    ((pte.raddr & SPLIT_HACK_MASK) == SPLIT_HACK_OFFS))
571 			pte.raddr &= ~SPLIT_HACK_MASK;
572 		/* fall through */
573 	case MSR_IR:
574 		vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
575 
576 		if ((kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) == MSR_DR)
577 			pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
578 		else
579 			pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
580 		pte.vpage |= vsid;
581 
582 		if (vsid == -1)
583 			page_found = -EINVAL;
584 		break;
585 	}
586 
587 	if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
588 	   (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
589 		/*
590 		 * If we do the dcbz hack, we have to NX on every execution,
591 		 * so we can patch the executing code. This renders our guest
592 		 * NX-less.
593 		 */
594 		pte.may_execute = !data;
595 	}
596 
597 	if (page_found == -ENOENT) {
598 		/* Page not found in guest PTE entries */
599 		u64 ssrr1 = vcpu->arch.shadow_srr1;
600 		u64 msr = kvmppc_get_msr(vcpu);
601 		kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
602 		kvmppc_set_dsisr(vcpu, vcpu->arch.fault_dsisr);
603 		kvmppc_set_msr_fast(vcpu, msr | (ssrr1 & 0xf8000000ULL));
604 		kvmppc_book3s_queue_irqprio(vcpu, vec);
605 	} else if (page_found == -EPERM) {
606 		/* Storage protection */
607 		u32 dsisr = vcpu->arch.fault_dsisr;
608 		u64 ssrr1 = vcpu->arch.shadow_srr1;
609 		u64 msr = kvmppc_get_msr(vcpu);
610 		kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
611 		dsisr = (dsisr & ~DSISR_NOHPTE) | DSISR_PROTFAULT;
612 		kvmppc_set_dsisr(vcpu, dsisr);
613 		kvmppc_set_msr_fast(vcpu, msr | (ssrr1 & 0xf8000000ULL));
614 		kvmppc_book3s_queue_irqprio(vcpu, vec);
615 	} else if (page_found == -EINVAL) {
616 		/* Page not found in guest SLB */
617 		kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
618 		kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
619 	} else if (!is_mmio &&
620 		   kvmppc_visible_gpa(vcpu, pte.raddr)) {
621 		if (data && !(vcpu->arch.fault_dsisr & DSISR_NOHPTE)) {
622 			/*
623 			 * There is already a host HPTE there, presumably
624 			 * a read-only one for a page the guest thinks
625 			 * is writable, so get rid of it first.
626 			 */
627 			kvmppc_mmu_unmap_page(vcpu, &pte);
628 		}
629 		/* The guest's PTE is not mapped yet. Map on the host */
630 		kvmppc_mmu_map_page(vcpu, &pte, iswrite);
631 		if (data)
632 			vcpu->stat.sp_storage++;
633 		else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
634 			 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
635 			kvmppc_patch_dcbz(vcpu, &pte);
636 	} else {
637 		/* MMIO */
638 		vcpu->stat.mmio_exits++;
639 		vcpu->arch.paddr_accessed = pte.raddr;
640 		vcpu->arch.vaddr_accessed = pte.eaddr;
641 		r = kvmppc_emulate_mmio(run, vcpu);
642 		if ( r == RESUME_HOST_NV )
643 			r = RESUME_HOST;
644 	}
645 
646 	return r;
647 }
648 
649 /* Give up external provider (FPU, Altivec, VSX) */
650 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
651 {
652 	struct thread_struct *t = &current->thread;
653 
654 	/*
655 	 * VSX instructions can access FP and vector registers, so if
656 	 * we are giving up VSX, make sure we give up FP and VMX as well.
657 	 */
658 	if (msr & MSR_VSX)
659 		msr |= MSR_FP | MSR_VEC;
660 
661 	msr &= vcpu->arch.guest_owned_ext;
662 	if (!msr)
663 		return;
664 
665 #ifdef DEBUG_EXT
666 	printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
667 #endif
668 
669 	if (msr & MSR_FP) {
670 		/*
671 		 * Note that on CPUs with VSX, giveup_fpu stores
672 		 * both the traditional FP registers and the added VSX
673 		 * registers into thread.fp_state.fpr[].
674 		 */
675 		if (t->regs->msr & MSR_FP)
676 			giveup_fpu(current);
677 		t->fp_save_area = NULL;
678 	}
679 
680 #ifdef CONFIG_ALTIVEC
681 	if (msr & MSR_VEC) {
682 		if (current->thread.regs->msr & MSR_VEC)
683 			giveup_altivec(current);
684 		t->vr_save_area = NULL;
685 	}
686 #endif
687 
688 	vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
689 	kvmppc_recalc_shadow_msr(vcpu);
690 }
691 
692 /* Give up facility (TAR / EBB / DSCR) */
693 static void kvmppc_giveup_fac(struct kvm_vcpu *vcpu, ulong fac)
694 {
695 #ifdef CONFIG_PPC_BOOK3S_64
696 	if (!(vcpu->arch.shadow_fscr & (1ULL << fac))) {
697 		/* Facility not available to the guest, ignore giveup request*/
698 		return;
699 	}
700 
701 	switch (fac) {
702 	case FSCR_TAR_LG:
703 		vcpu->arch.tar = mfspr(SPRN_TAR);
704 		mtspr(SPRN_TAR, current->thread.tar);
705 		vcpu->arch.shadow_fscr &= ~FSCR_TAR;
706 		break;
707 	}
708 #endif
709 }
710 
711 /* Handle external providers (FPU, Altivec, VSX) */
712 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
713 			     ulong msr)
714 {
715 	struct thread_struct *t = &current->thread;
716 
717 	/* When we have paired singles, we emulate in software */
718 	if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
719 		return RESUME_GUEST;
720 
721 	if (!(kvmppc_get_msr(vcpu) & msr)) {
722 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
723 		return RESUME_GUEST;
724 	}
725 
726 	if (msr == MSR_VSX) {
727 		/* No VSX?  Give an illegal instruction interrupt */
728 #ifdef CONFIG_VSX
729 		if (!cpu_has_feature(CPU_FTR_VSX))
730 #endif
731 		{
732 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
733 			return RESUME_GUEST;
734 		}
735 
736 		/*
737 		 * We have to load up all the FP and VMX registers before
738 		 * we can let the guest use VSX instructions.
739 		 */
740 		msr = MSR_FP | MSR_VEC | MSR_VSX;
741 	}
742 
743 	/* See if we already own all the ext(s) needed */
744 	msr &= ~vcpu->arch.guest_owned_ext;
745 	if (!msr)
746 		return RESUME_GUEST;
747 
748 #ifdef DEBUG_EXT
749 	printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
750 #endif
751 
752 	if (msr & MSR_FP) {
753 		preempt_disable();
754 		enable_kernel_fp();
755 		load_fp_state(&vcpu->arch.fp);
756 		disable_kernel_fp();
757 		t->fp_save_area = &vcpu->arch.fp;
758 		preempt_enable();
759 	}
760 
761 	if (msr & MSR_VEC) {
762 #ifdef CONFIG_ALTIVEC
763 		preempt_disable();
764 		enable_kernel_altivec();
765 		load_vr_state(&vcpu->arch.vr);
766 		disable_kernel_altivec();
767 		t->vr_save_area = &vcpu->arch.vr;
768 		preempt_enable();
769 #endif
770 	}
771 
772 	t->regs->msr |= msr;
773 	vcpu->arch.guest_owned_ext |= msr;
774 	kvmppc_recalc_shadow_msr(vcpu);
775 
776 	return RESUME_GUEST;
777 }
778 
779 /*
780  * Kernel code using FP or VMX could have flushed guest state to
781  * the thread_struct; if so, get it back now.
782  */
783 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
784 {
785 	unsigned long lost_ext;
786 
787 	lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
788 	if (!lost_ext)
789 		return;
790 
791 	if (lost_ext & MSR_FP) {
792 		preempt_disable();
793 		enable_kernel_fp();
794 		load_fp_state(&vcpu->arch.fp);
795 		disable_kernel_fp();
796 		preempt_enable();
797 	}
798 #ifdef CONFIG_ALTIVEC
799 	if (lost_ext & MSR_VEC) {
800 		preempt_disable();
801 		enable_kernel_altivec();
802 		load_vr_state(&vcpu->arch.vr);
803 		disable_kernel_altivec();
804 		preempt_enable();
805 	}
806 #endif
807 	current->thread.regs->msr |= lost_ext;
808 }
809 
810 #ifdef CONFIG_PPC_BOOK3S_64
811 
812 static void kvmppc_trigger_fac_interrupt(struct kvm_vcpu *vcpu, ulong fac)
813 {
814 	/* Inject the Interrupt Cause field and trigger a guest interrupt */
815 	vcpu->arch.fscr &= ~(0xffULL << 56);
816 	vcpu->arch.fscr |= (fac << 56);
817 	kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_FAC_UNAVAIL);
818 }
819 
820 static void kvmppc_emulate_fac(struct kvm_vcpu *vcpu, ulong fac)
821 {
822 	enum emulation_result er = EMULATE_FAIL;
823 
824 	if (!(kvmppc_get_msr(vcpu) & MSR_PR))
825 		er = kvmppc_emulate_instruction(vcpu->run, vcpu);
826 
827 	if ((er != EMULATE_DONE) && (er != EMULATE_AGAIN)) {
828 		/* Couldn't emulate, trigger interrupt in guest */
829 		kvmppc_trigger_fac_interrupt(vcpu, fac);
830 	}
831 }
832 
833 /* Enable facilities (TAR, EBB, DSCR) for the guest */
834 static int kvmppc_handle_fac(struct kvm_vcpu *vcpu, ulong fac)
835 {
836 	bool guest_fac_enabled;
837 	BUG_ON(!cpu_has_feature(CPU_FTR_ARCH_207S));
838 
839 	/*
840 	 * Not every facility is enabled by FSCR bits, check whether the
841 	 * guest has this facility enabled at all.
842 	 */
843 	switch (fac) {
844 	case FSCR_TAR_LG:
845 	case FSCR_EBB_LG:
846 		guest_fac_enabled = (vcpu->arch.fscr & (1ULL << fac));
847 		break;
848 	case FSCR_TM_LG:
849 		guest_fac_enabled = kvmppc_get_msr(vcpu) & MSR_TM;
850 		break;
851 	default:
852 		guest_fac_enabled = false;
853 		break;
854 	}
855 
856 	if (!guest_fac_enabled) {
857 		/* Facility not enabled by the guest */
858 		kvmppc_trigger_fac_interrupt(vcpu, fac);
859 		return RESUME_GUEST;
860 	}
861 
862 	switch (fac) {
863 	case FSCR_TAR_LG:
864 		/* TAR switching isn't lazy in Linux yet */
865 		current->thread.tar = mfspr(SPRN_TAR);
866 		mtspr(SPRN_TAR, vcpu->arch.tar);
867 		vcpu->arch.shadow_fscr |= FSCR_TAR;
868 		break;
869 	default:
870 		kvmppc_emulate_fac(vcpu, fac);
871 		break;
872 	}
873 
874 	return RESUME_GUEST;
875 }
876 
877 void kvmppc_set_fscr(struct kvm_vcpu *vcpu, u64 fscr)
878 {
879 	if ((vcpu->arch.fscr & FSCR_TAR) && !(fscr & FSCR_TAR)) {
880 		/* TAR got dropped, drop it in shadow too */
881 		kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
882 	}
883 	vcpu->arch.fscr = fscr;
884 }
885 #endif
886 
887 static void kvmppc_setup_debug(struct kvm_vcpu *vcpu)
888 {
889 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
890 		u64 msr = kvmppc_get_msr(vcpu);
891 
892 		kvmppc_set_msr(vcpu, msr | MSR_SE);
893 	}
894 }
895 
896 static void kvmppc_clear_debug(struct kvm_vcpu *vcpu)
897 {
898 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
899 		u64 msr = kvmppc_get_msr(vcpu);
900 
901 		kvmppc_set_msr(vcpu, msr & ~MSR_SE);
902 	}
903 }
904 
905 int kvmppc_handle_exit_pr(struct kvm_run *run, struct kvm_vcpu *vcpu,
906 			  unsigned int exit_nr)
907 {
908 	int r = RESUME_HOST;
909 	int s;
910 
911 	vcpu->stat.sum_exits++;
912 
913 	run->exit_reason = KVM_EXIT_UNKNOWN;
914 	run->ready_for_interrupt_injection = 1;
915 
916 	/* We get here with MSR.EE=1 */
917 
918 	trace_kvm_exit(exit_nr, vcpu);
919 	guest_exit();
920 
921 	switch (exit_nr) {
922 	case BOOK3S_INTERRUPT_INST_STORAGE:
923 	{
924 		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
925 		vcpu->stat.pf_instruc++;
926 
927 		if (kvmppc_is_split_real(vcpu))
928 			kvmppc_fixup_split_real(vcpu);
929 
930 #ifdef CONFIG_PPC_BOOK3S_32
931 		/* We set segments as unused segments when invalidating them. So
932 		 * treat the respective fault as segment fault. */
933 		{
934 			struct kvmppc_book3s_shadow_vcpu *svcpu;
935 			u32 sr;
936 
937 			svcpu = svcpu_get(vcpu);
938 			sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
939 			svcpu_put(svcpu);
940 			if (sr == SR_INVALID) {
941 				kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
942 				r = RESUME_GUEST;
943 				break;
944 			}
945 		}
946 #endif
947 
948 		/* only care about PTEG not found errors, but leave NX alone */
949 		if (shadow_srr1 & 0x40000000) {
950 			int idx = srcu_read_lock(&vcpu->kvm->srcu);
951 			r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
952 			srcu_read_unlock(&vcpu->kvm->srcu, idx);
953 			vcpu->stat.sp_instruc++;
954 		} else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
955 			  (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
956 			/*
957 			 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
958 			 *     so we can't use the NX bit inside the guest. Let's cross our fingers,
959 			 *     that no guest that needs the dcbz hack does NX.
960 			 */
961 			kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
962 			r = RESUME_GUEST;
963 		} else {
964 			u64 msr = kvmppc_get_msr(vcpu);
965 			msr |= shadow_srr1 & 0x58000000;
966 			kvmppc_set_msr_fast(vcpu, msr);
967 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
968 			r = RESUME_GUEST;
969 		}
970 		break;
971 	}
972 	case BOOK3S_INTERRUPT_DATA_STORAGE:
973 	{
974 		ulong dar = kvmppc_get_fault_dar(vcpu);
975 		u32 fault_dsisr = vcpu->arch.fault_dsisr;
976 		vcpu->stat.pf_storage++;
977 
978 #ifdef CONFIG_PPC_BOOK3S_32
979 		/* We set segments as unused segments when invalidating them. So
980 		 * treat the respective fault as segment fault. */
981 		{
982 			struct kvmppc_book3s_shadow_vcpu *svcpu;
983 			u32 sr;
984 
985 			svcpu = svcpu_get(vcpu);
986 			sr = svcpu->sr[dar >> SID_SHIFT];
987 			svcpu_put(svcpu);
988 			if (sr == SR_INVALID) {
989 				kvmppc_mmu_map_segment(vcpu, dar);
990 				r = RESUME_GUEST;
991 				break;
992 			}
993 		}
994 #endif
995 
996 		/*
997 		 * We need to handle missing shadow PTEs, and
998 		 * protection faults due to us mapping a page read-only
999 		 * when the guest thinks it is writable.
1000 		 */
1001 		if (fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT)) {
1002 			int idx = srcu_read_lock(&vcpu->kvm->srcu);
1003 			r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
1004 			srcu_read_unlock(&vcpu->kvm->srcu, idx);
1005 		} else {
1006 			kvmppc_set_dar(vcpu, dar);
1007 			kvmppc_set_dsisr(vcpu, fault_dsisr);
1008 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1009 			r = RESUME_GUEST;
1010 		}
1011 		break;
1012 	}
1013 	case BOOK3S_INTERRUPT_DATA_SEGMENT:
1014 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
1015 			kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
1016 			kvmppc_book3s_queue_irqprio(vcpu,
1017 				BOOK3S_INTERRUPT_DATA_SEGMENT);
1018 		}
1019 		r = RESUME_GUEST;
1020 		break;
1021 	case BOOK3S_INTERRUPT_INST_SEGMENT:
1022 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
1023 			kvmppc_book3s_queue_irqprio(vcpu,
1024 				BOOK3S_INTERRUPT_INST_SEGMENT);
1025 		}
1026 		r = RESUME_GUEST;
1027 		break;
1028 	/* We're good on these - the host merely wanted to get our attention */
1029 	case BOOK3S_INTERRUPT_DECREMENTER:
1030 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1031 	case BOOK3S_INTERRUPT_DOORBELL:
1032 	case BOOK3S_INTERRUPT_H_DOORBELL:
1033 		vcpu->stat.dec_exits++;
1034 		r = RESUME_GUEST;
1035 		break;
1036 	case BOOK3S_INTERRUPT_EXTERNAL:
1037 	case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
1038 	case BOOK3S_INTERRUPT_EXTERNAL_HV:
1039 		vcpu->stat.ext_intr_exits++;
1040 		r = RESUME_GUEST;
1041 		break;
1042 	case BOOK3S_INTERRUPT_PERFMON:
1043 		r = RESUME_GUEST;
1044 		break;
1045 	case BOOK3S_INTERRUPT_PROGRAM:
1046 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1047 	{
1048 		enum emulation_result er;
1049 		ulong flags;
1050 		u32 last_inst;
1051 		int emul;
1052 
1053 program_interrupt:
1054 		/*
1055 		 * shadow_srr1 only contains valid flags if we came here via
1056 		 * a program exception. The other exceptions (emulation assist,
1057 		 * FP unavailable, etc.) do not provide flags in SRR1, so use
1058 		 * an illegal-instruction exception when injecting a program
1059 		 * interrupt into the guest.
1060 		 */
1061 		if (exit_nr == BOOK3S_INTERRUPT_PROGRAM)
1062 			flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
1063 		else
1064 			flags = SRR1_PROGILL;
1065 
1066 		emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1067 		if (emul != EMULATE_DONE) {
1068 			r = RESUME_GUEST;
1069 			break;
1070 		}
1071 
1072 		if (kvmppc_get_msr(vcpu) & MSR_PR) {
1073 #ifdef EXIT_DEBUG
1074 			pr_info("Userspace triggered 0x700 exception at\n 0x%lx (0x%x)\n",
1075 				kvmppc_get_pc(vcpu), last_inst);
1076 #endif
1077 			if ((last_inst & 0xff0007ff) !=
1078 			    (INS_DCBZ & 0xfffffff7)) {
1079 				kvmppc_core_queue_program(vcpu, flags);
1080 				r = RESUME_GUEST;
1081 				break;
1082 			}
1083 		}
1084 
1085 		vcpu->stat.emulated_inst_exits++;
1086 		er = kvmppc_emulate_instruction(run, vcpu);
1087 		switch (er) {
1088 		case EMULATE_DONE:
1089 			r = RESUME_GUEST_NV;
1090 			break;
1091 		case EMULATE_AGAIN:
1092 			r = RESUME_GUEST;
1093 			break;
1094 		case EMULATE_FAIL:
1095 			printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
1096 			       __func__, kvmppc_get_pc(vcpu), last_inst);
1097 			kvmppc_core_queue_program(vcpu, flags);
1098 			r = RESUME_GUEST;
1099 			break;
1100 		case EMULATE_DO_MMIO:
1101 			run->exit_reason = KVM_EXIT_MMIO;
1102 			r = RESUME_HOST_NV;
1103 			break;
1104 		case EMULATE_EXIT_USER:
1105 			r = RESUME_HOST_NV;
1106 			break;
1107 		default:
1108 			BUG();
1109 		}
1110 		break;
1111 	}
1112 	case BOOK3S_INTERRUPT_SYSCALL:
1113 	{
1114 		u32 last_sc;
1115 		int emul;
1116 
1117 		/* Get last sc for papr */
1118 		if (vcpu->arch.papr_enabled) {
1119 			/* The sc instuction points SRR0 to the next inst */
1120 			emul = kvmppc_get_last_inst(vcpu, INST_SC, &last_sc);
1121 			if (emul != EMULATE_DONE) {
1122 				kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) - 4);
1123 				r = RESUME_GUEST;
1124 				break;
1125 			}
1126 		}
1127 
1128 		if (vcpu->arch.papr_enabled &&
1129 		    (last_sc == 0x44000022) &&
1130 		    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
1131 			/* SC 1 papr hypercalls */
1132 			ulong cmd = kvmppc_get_gpr(vcpu, 3);
1133 			int i;
1134 
1135 #ifdef CONFIG_PPC_BOOK3S_64
1136 			if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
1137 				r = RESUME_GUEST;
1138 				break;
1139 			}
1140 #endif
1141 
1142 			run->papr_hcall.nr = cmd;
1143 			for (i = 0; i < 9; ++i) {
1144 				ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
1145 				run->papr_hcall.args[i] = gpr;
1146 			}
1147 			run->exit_reason = KVM_EXIT_PAPR_HCALL;
1148 			vcpu->arch.hcall_needed = 1;
1149 			r = RESUME_HOST;
1150 		} else if (vcpu->arch.osi_enabled &&
1151 		    (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
1152 		    (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
1153 			/* MOL hypercalls */
1154 			u64 *gprs = run->osi.gprs;
1155 			int i;
1156 
1157 			run->exit_reason = KVM_EXIT_OSI;
1158 			for (i = 0; i < 32; i++)
1159 				gprs[i] = kvmppc_get_gpr(vcpu, i);
1160 			vcpu->arch.osi_needed = 1;
1161 			r = RESUME_HOST_NV;
1162 		} else if (!(kvmppc_get_msr(vcpu) & MSR_PR) &&
1163 		    (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
1164 			/* KVM PV hypercalls */
1165 			kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
1166 			r = RESUME_GUEST;
1167 		} else {
1168 			/* Guest syscalls */
1169 			vcpu->stat.syscall_exits++;
1170 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1171 			r = RESUME_GUEST;
1172 		}
1173 		break;
1174 	}
1175 	case BOOK3S_INTERRUPT_FP_UNAVAIL:
1176 	case BOOK3S_INTERRUPT_ALTIVEC:
1177 	case BOOK3S_INTERRUPT_VSX:
1178 	{
1179 		int ext_msr = 0;
1180 		int emul;
1181 		u32 last_inst;
1182 
1183 		if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE) {
1184 			/* Do paired single instruction emulation */
1185 			emul = kvmppc_get_last_inst(vcpu, INST_GENERIC,
1186 						    &last_inst);
1187 			if (emul == EMULATE_DONE)
1188 				goto program_interrupt;
1189 			else
1190 				r = RESUME_GUEST;
1191 
1192 			break;
1193 		}
1194 
1195 		/* Enable external provider */
1196 		switch (exit_nr) {
1197 		case BOOK3S_INTERRUPT_FP_UNAVAIL:
1198 			ext_msr = MSR_FP;
1199 			break;
1200 
1201 		case BOOK3S_INTERRUPT_ALTIVEC:
1202 			ext_msr = MSR_VEC;
1203 			break;
1204 
1205 		case BOOK3S_INTERRUPT_VSX:
1206 			ext_msr = MSR_VSX;
1207 			break;
1208 		}
1209 
1210 		r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
1211 		break;
1212 	}
1213 	case BOOK3S_INTERRUPT_ALIGNMENT:
1214 	{
1215 		u32 last_inst;
1216 		int emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1217 
1218 		if (emul == EMULATE_DONE) {
1219 			u32 dsisr;
1220 			u64 dar;
1221 
1222 			dsisr = kvmppc_alignment_dsisr(vcpu, last_inst);
1223 			dar = kvmppc_alignment_dar(vcpu, last_inst);
1224 
1225 			kvmppc_set_dsisr(vcpu, dsisr);
1226 			kvmppc_set_dar(vcpu, dar);
1227 
1228 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1229 		}
1230 		r = RESUME_GUEST;
1231 		break;
1232 	}
1233 #ifdef CONFIG_PPC_BOOK3S_64
1234 	case BOOK3S_INTERRUPT_FAC_UNAVAIL:
1235 		kvmppc_handle_fac(vcpu, vcpu->arch.shadow_fscr >> 56);
1236 		r = RESUME_GUEST;
1237 		break;
1238 #endif
1239 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1240 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1241 		r = RESUME_GUEST;
1242 		break;
1243 	case BOOK3S_INTERRUPT_TRACE:
1244 		if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
1245 			run->exit_reason = KVM_EXIT_DEBUG;
1246 			r = RESUME_HOST;
1247 		} else {
1248 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1249 			r = RESUME_GUEST;
1250 		}
1251 		break;
1252 	default:
1253 	{
1254 		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1255 		/* Ugh - bork here! What did we get? */
1256 		printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1257 			exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
1258 		r = RESUME_HOST;
1259 		BUG();
1260 		break;
1261 	}
1262 	}
1263 
1264 	if (!(r & RESUME_HOST)) {
1265 		/* To avoid clobbering exit_reason, only check for signals if
1266 		 * we aren't already exiting to userspace for some other
1267 		 * reason. */
1268 
1269 		/*
1270 		 * Interrupts could be timers for the guest which we have to
1271 		 * inject again, so let's postpone them until we're in the guest
1272 		 * and if we really did time things so badly, then we just exit
1273 		 * again due to a host external interrupt.
1274 		 */
1275 		s = kvmppc_prepare_to_enter(vcpu);
1276 		if (s <= 0)
1277 			r = s;
1278 		else {
1279 			/* interrupts now hard-disabled */
1280 			kvmppc_fix_ee_before_entry();
1281 		}
1282 
1283 		kvmppc_handle_lost_ext(vcpu);
1284 	}
1285 
1286 	trace_kvm_book3s_reenter(r, vcpu);
1287 
1288 	return r;
1289 }
1290 
1291 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu *vcpu,
1292 					    struct kvm_sregs *sregs)
1293 {
1294 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1295 	int i;
1296 
1297 	sregs->pvr = vcpu->arch.pvr;
1298 
1299 	sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1300 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1301 		for (i = 0; i < 64; i++) {
1302 			sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
1303 			sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1304 		}
1305 	} else {
1306 		for (i = 0; i < 16; i++)
1307 			sregs->u.s.ppc32.sr[i] = kvmppc_get_sr(vcpu, i);
1308 
1309 		for (i = 0; i < 8; i++) {
1310 			sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1311 			sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1312 		}
1313 	}
1314 
1315 	return 0;
1316 }
1317 
1318 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu *vcpu,
1319 					    struct kvm_sregs *sregs)
1320 {
1321 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1322 	int i;
1323 
1324 	kvmppc_set_pvr_pr(vcpu, sregs->pvr);
1325 
1326 	vcpu3s->sdr1 = sregs->u.s.sdr1;
1327 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1328 		for (i = 0; i < 64; i++) {
1329 			vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1330 						    sregs->u.s.ppc64.slb[i].slbe);
1331 		}
1332 	} else {
1333 		for (i = 0; i < 16; i++) {
1334 			vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1335 		}
1336 		for (i = 0; i < 8; i++) {
1337 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1338 				       (u32)sregs->u.s.ppc32.ibat[i]);
1339 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1340 				       (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1341 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1342 				       (u32)sregs->u.s.ppc32.dbat[i]);
1343 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1344 				       (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1345 		}
1346 	}
1347 
1348 	/* Flush the MMU after messing with the segments */
1349 	kvmppc_mmu_pte_flush(vcpu, 0, 0);
1350 
1351 	return 0;
1352 }
1353 
1354 static int kvmppc_get_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1355 				 union kvmppc_one_reg *val)
1356 {
1357 	int r = 0;
1358 
1359 	switch (id) {
1360 	case KVM_REG_PPC_DEBUG_INST:
1361 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1362 		break;
1363 	case KVM_REG_PPC_HIOR:
1364 		*val = get_reg_val(id, to_book3s(vcpu)->hior);
1365 		break;
1366 	case KVM_REG_PPC_VTB:
1367 		*val = get_reg_val(id, to_book3s(vcpu)->vtb);
1368 		break;
1369 	case KVM_REG_PPC_LPCR:
1370 	case KVM_REG_PPC_LPCR_64:
1371 		/*
1372 		 * We are only interested in the LPCR_ILE bit
1373 		 */
1374 		if (vcpu->arch.intr_msr & MSR_LE)
1375 			*val = get_reg_val(id, LPCR_ILE);
1376 		else
1377 			*val = get_reg_val(id, 0);
1378 		break;
1379 	default:
1380 		r = -EINVAL;
1381 		break;
1382 	}
1383 
1384 	return r;
1385 }
1386 
1387 static void kvmppc_set_lpcr_pr(struct kvm_vcpu *vcpu, u64 new_lpcr)
1388 {
1389 	if (new_lpcr & LPCR_ILE)
1390 		vcpu->arch.intr_msr |= MSR_LE;
1391 	else
1392 		vcpu->arch.intr_msr &= ~MSR_LE;
1393 }
1394 
1395 static int kvmppc_set_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1396 				 union kvmppc_one_reg *val)
1397 {
1398 	int r = 0;
1399 
1400 	switch (id) {
1401 	case KVM_REG_PPC_HIOR:
1402 		to_book3s(vcpu)->hior = set_reg_val(id, *val);
1403 		to_book3s(vcpu)->hior_explicit = true;
1404 		break;
1405 	case KVM_REG_PPC_VTB:
1406 		to_book3s(vcpu)->vtb = set_reg_val(id, *val);
1407 		break;
1408 	case KVM_REG_PPC_LPCR:
1409 	case KVM_REG_PPC_LPCR_64:
1410 		kvmppc_set_lpcr_pr(vcpu, set_reg_val(id, *val));
1411 		break;
1412 	default:
1413 		r = -EINVAL;
1414 		break;
1415 	}
1416 
1417 	return r;
1418 }
1419 
1420 static struct kvm_vcpu *kvmppc_core_vcpu_create_pr(struct kvm *kvm,
1421 						   unsigned int id)
1422 {
1423 	struct kvmppc_vcpu_book3s *vcpu_book3s;
1424 	struct kvm_vcpu *vcpu;
1425 	int err = -ENOMEM;
1426 	unsigned long p;
1427 
1428 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1429 	if (!vcpu)
1430 		goto out;
1431 
1432 	vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1433 	if (!vcpu_book3s)
1434 		goto free_vcpu;
1435 	vcpu->arch.book3s = vcpu_book3s;
1436 
1437 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1438 	vcpu->arch.shadow_vcpu =
1439 		kzalloc(sizeof(*vcpu->arch.shadow_vcpu), GFP_KERNEL);
1440 	if (!vcpu->arch.shadow_vcpu)
1441 		goto free_vcpu3s;
1442 #endif
1443 
1444 	err = kvm_vcpu_init(vcpu, kvm, id);
1445 	if (err)
1446 		goto free_shadow_vcpu;
1447 
1448 	err = -ENOMEM;
1449 	p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1450 	if (!p)
1451 		goto uninit_vcpu;
1452 	vcpu->arch.shared = (void *)p;
1453 #ifdef CONFIG_PPC_BOOK3S_64
1454 	/* Always start the shared struct in native endian mode */
1455 #ifdef __BIG_ENDIAN__
1456         vcpu->arch.shared_big_endian = true;
1457 #else
1458         vcpu->arch.shared_big_endian = false;
1459 #endif
1460 
1461 	/*
1462 	 * Default to the same as the host if we're on sufficiently
1463 	 * recent machine that we have 1TB segments;
1464 	 * otherwise default to PPC970FX.
1465 	 */
1466 	vcpu->arch.pvr = 0x3C0301;
1467 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1468 		vcpu->arch.pvr = mfspr(SPRN_PVR);
1469 	vcpu->arch.intr_msr = MSR_SF;
1470 #else
1471 	/* default to book3s_32 (750) */
1472 	vcpu->arch.pvr = 0x84202;
1473 #endif
1474 	kvmppc_set_pvr_pr(vcpu, vcpu->arch.pvr);
1475 	vcpu->arch.slb_nr = 64;
1476 
1477 	vcpu->arch.shadow_msr = MSR_USER64 & ~MSR_LE;
1478 
1479 	err = kvmppc_mmu_init(vcpu);
1480 	if (err < 0)
1481 		goto uninit_vcpu;
1482 
1483 	return vcpu;
1484 
1485 uninit_vcpu:
1486 	kvm_vcpu_uninit(vcpu);
1487 free_shadow_vcpu:
1488 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1489 	kfree(vcpu->arch.shadow_vcpu);
1490 free_vcpu3s:
1491 #endif
1492 	vfree(vcpu_book3s);
1493 free_vcpu:
1494 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1495 out:
1496 	return ERR_PTR(err);
1497 }
1498 
1499 static void kvmppc_core_vcpu_free_pr(struct kvm_vcpu *vcpu)
1500 {
1501 	struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1502 
1503 	free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1504 	kvm_vcpu_uninit(vcpu);
1505 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1506 	kfree(vcpu->arch.shadow_vcpu);
1507 #endif
1508 	vfree(vcpu_book3s);
1509 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1510 }
1511 
1512 static int kvmppc_vcpu_run_pr(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1513 {
1514 	int ret;
1515 #ifdef CONFIG_ALTIVEC
1516 	unsigned long uninitialized_var(vrsave);
1517 #endif
1518 
1519 	/* Check if we can run the vcpu at all */
1520 	if (!vcpu->arch.sane) {
1521 		kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1522 		ret = -EINVAL;
1523 		goto out;
1524 	}
1525 
1526 	kvmppc_setup_debug(vcpu);
1527 
1528 	/*
1529 	 * Interrupts could be timers for the guest which we have to inject
1530 	 * again, so let's postpone them until we're in the guest and if we
1531 	 * really did time things so badly, then we just exit again due to
1532 	 * a host external interrupt.
1533 	 */
1534 	ret = kvmppc_prepare_to_enter(vcpu);
1535 	if (ret <= 0)
1536 		goto out;
1537 	/* interrupts now hard-disabled */
1538 
1539 	/* Save FPU, Altivec and VSX state */
1540 	giveup_all(current);
1541 
1542 	/* Preload FPU if it's enabled */
1543 	if (kvmppc_get_msr(vcpu) & MSR_FP)
1544 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1545 
1546 	kvmppc_fix_ee_before_entry();
1547 
1548 	ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1549 
1550 	kvmppc_clear_debug(vcpu);
1551 
1552 	/* No need for guest_exit. It's done in handle_exit.
1553 	   We also get here with interrupts enabled. */
1554 
1555 	/* Make sure we save the guest FPU/Altivec/VSX state */
1556 	kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1557 
1558 	/* Make sure we save the guest TAR/EBB/DSCR state */
1559 	kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
1560 
1561 out:
1562 	vcpu->mode = OUTSIDE_GUEST_MODE;
1563 	return ret;
1564 }
1565 
1566 /*
1567  * Get (and clear) the dirty memory log for a memory slot.
1568  */
1569 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm *kvm,
1570 					 struct kvm_dirty_log *log)
1571 {
1572 	struct kvm_memslots *slots;
1573 	struct kvm_memory_slot *memslot;
1574 	struct kvm_vcpu *vcpu;
1575 	ulong ga, ga_end;
1576 	int is_dirty = 0;
1577 	int r;
1578 	unsigned long n;
1579 
1580 	mutex_lock(&kvm->slots_lock);
1581 
1582 	r = kvm_get_dirty_log(kvm, log, &is_dirty);
1583 	if (r)
1584 		goto out;
1585 
1586 	/* If nothing is dirty, don't bother messing with page tables. */
1587 	if (is_dirty) {
1588 		slots = kvm_memslots(kvm);
1589 		memslot = id_to_memslot(slots, log->slot);
1590 
1591 		ga = memslot->base_gfn << PAGE_SHIFT;
1592 		ga_end = ga + (memslot->npages << PAGE_SHIFT);
1593 
1594 		kvm_for_each_vcpu(n, vcpu, kvm)
1595 			kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1596 
1597 		n = kvm_dirty_bitmap_bytes(memslot);
1598 		memset(memslot->dirty_bitmap, 0, n);
1599 	}
1600 
1601 	r = 0;
1602 out:
1603 	mutex_unlock(&kvm->slots_lock);
1604 	return r;
1605 }
1606 
1607 static void kvmppc_core_flush_memslot_pr(struct kvm *kvm,
1608 					 struct kvm_memory_slot *memslot)
1609 {
1610 	return;
1611 }
1612 
1613 static int kvmppc_core_prepare_memory_region_pr(struct kvm *kvm,
1614 					struct kvm_memory_slot *memslot,
1615 					const struct kvm_userspace_memory_region *mem)
1616 {
1617 	return 0;
1618 }
1619 
1620 static void kvmppc_core_commit_memory_region_pr(struct kvm *kvm,
1621 				const struct kvm_userspace_memory_region *mem,
1622 				const struct kvm_memory_slot *old,
1623 				const struct kvm_memory_slot *new)
1624 {
1625 	return;
1626 }
1627 
1628 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot *free,
1629 					struct kvm_memory_slot *dont)
1630 {
1631 	return;
1632 }
1633 
1634 static int kvmppc_core_create_memslot_pr(struct kvm_memory_slot *slot,
1635 					 unsigned long npages)
1636 {
1637 	return 0;
1638 }
1639 
1640 
1641 #ifdef CONFIG_PPC64
1642 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1643 					 struct kvm_ppc_smmu_info *info)
1644 {
1645 	long int i;
1646 	struct kvm_vcpu *vcpu;
1647 
1648 	info->flags = 0;
1649 
1650 	/* SLB is always 64 entries */
1651 	info->slb_size = 64;
1652 
1653 	/* Standard 4k base page size segment */
1654 	info->sps[0].page_shift = 12;
1655 	info->sps[0].slb_enc = 0;
1656 	info->sps[0].enc[0].page_shift = 12;
1657 	info->sps[0].enc[0].pte_enc = 0;
1658 
1659 	/*
1660 	 * 64k large page size.
1661 	 * We only want to put this in if the CPUs we're emulating
1662 	 * support it, but unfortunately we don't have a vcpu easily
1663 	 * to hand here to test.  Just pick the first vcpu, and if
1664 	 * that doesn't exist yet, report the minimum capability,
1665 	 * i.e., no 64k pages.
1666 	 * 1T segment support goes along with 64k pages.
1667 	 */
1668 	i = 1;
1669 	vcpu = kvm_get_vcpu(kvm, 0);
1670 	if (vcpu && (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
1671 		info->flags = KVM_PPC_1T_SEGMENTS;
1672 		info->sps[i].page_shift = 16;
1673 		info->sps[i].slb_enc = SLB_VSID_L | SLB_VSID_LP_01;
1674 		info->sps[i].enc[0].page_shift = 16;
1675 		info->sps[i].enc[0].pte_enc = 1;
1676 		++i;
1677 	}
1678 
1679 	/* Standard 16M large page size segment */
1680 	info->sps[i].page_shift = 24;
1681 	info->sps[i].slb_enc = SLB_VSID_L;
1682 	info->sps[i].enc[0].page_shift = 24;
1683 	info->sps[i].enc[0].pte_enc = 0;
1684 
1685 	return 0;
1686 }
1687 #else
1688 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1689 					 struct kvm_ppc_smmu_info *info)
1690 {
1691 	/* We should not get called */
1692 	BUG();
1693 }
1694 #endif /* CONFIG_PPC64 */
1695 
1696 static unsigned int kvm_global_user_count = 0;
1697 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1698 
1699 static int kvmppc_core_init_vm_pr(struct kvm *kvm)
1700 {
1701 	mutex_init(&kvm->arch.hpt_mutex);
1702 
1703 #ifdef CONFIG_PPC_BOOK3S_64
1704 	/* Start out with the default set of hcalls enabled */
1705 	kvmppc_pr_init_default_hcalls(kvm);
1706 #endif
1707 
1708 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1709 		spin_lock(&kvm_global_user_count_lock);
1710 		if (++kvm_global_user_count == 1)
1711 			pseries_disable_reloc_on_exc();
1712 		spin_unlock(&kvm_global_user_count_lock);
1713 	}
1714 	return 0;
1715 }
1716 
1717 static void kvmppc_core_destroy_vm_pr(struct kvm *kvm)
1718 {
1719 #ifdef CONFIG_PPC64
1720 	WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1721 #endif
1722 
1723 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1724 		spin_lock(&kvm_global_user_count_lock);
1725 		BUG_ON(kvm_global_user_count == 0);
1726 		if (--kvm_global_user_count == 0)
1727 			pseries_enable_reloc_on_exc();
1728 		spin_unlock(&kvm_global_user_count_lock);
1729 	}
1730 }
1731 
1732 static int kvmppc_core_check_processor_compat_pr(void)
1733 {
1734 	/*
1735 	 * Disable KVM for Power9 untill the required bits merged.
1736 	 */
1737 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1738 		return -EIO;
1739 	return 0;
1740 }
1741 
1742 static long kvm_arch_vm_ioctl_pr(struct file *filp,
1743 				 unsigned int ioctl, unsigned long arg)
1744 {
1745 	return -ENOTTY;
1746 }
1747 
1748 static struct kvmppc_ops kvm_ops_pr = {
1749 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_pr,
1750 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_pr,
1751 	.get_one_reg = kvmppc_get_one_reg_pr,
1752 	.set_one_reg = kvmppc_set_one_reg_pr,
1753 	.vcpu_load   = kvmppc_core_vcpu_load_pr,
1754 	.vcpu_put    = kvmppc_core_vcpu_put_pr,
1755 	.set_msr     = kvmppc_set_msr_pr,
1756 	.vcpu_run    = kvmppc_vcpu_run_pr,
1757 	.vcpu_create = kvmppc_core_vcpu_create_pr,
1758 	.vcpu_free   = kvmppc_core_vcpu_free_pr,
1759 	.check_requests = kvmppc_core_check_requests_pr,
1760 	.get_dirty_log = kvm_vm_ioctl_get_dirty_log_pr,
1761 	.flush_memslot = kvmppc_core_flush_memslot_pr,
1762 	.prepare_memory_region = kvmppc_core_prepare_memory_region_pr,
1763 	.commit_memory_region = kvmppc_core_commit_memory_region_pr,
1764 	.unmap_hva = kvm_unmap_hva_pr,
1765 	.unmap_hva_range = kvm_unmap_hva_range_pr,
1766 	.age_hva  = kvm_age_hva_pr,
1767 	.test_age_hva = kvm_test_age_hva_pr,
1768 	.set_spte_hva = kvm_set_spte_hva_pr,
1769 	.mmu_destroy  = kvmppc_mmu_destroy_pr,
1770 	.free_memslot = kvmppc_core_free_memslot_pr,
1771 	.create_memslot = kvmppc_core_create_memslot_pr,
1772 	.init_vm = kvmppc_core_init_vm_pr,
1773 	.destroy_vm = kvmppc_core_destroy_vm_pr,
1774 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_pr,
1775 	.emulate_op = kvmppc_core_emulate_op_pr,
1776 	.emulate_mtspr = kvmppc_core_emulate_mtspr_pr,
1777 	.emulate_mfspr = kvmppc_core_emulate_mfspr_pr,
1778 	.fast_vcpu_kick = kvm_vcpu_kick,
1779 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_pr,
1780 #ifdef CONFIG_PPC_BOOK3S_64
1781 	.hcall_implemented = kvmppc_hcall_impl_pr,
1782 #endif
1783 };
1784 
1785 
1786 int kvmppc_book3s_init_pr(void)
1787 {
1788 	int r;
1789 
1790 	r = kvmppc_core_check_processor_compat_pr();
1791 	if (r < 0)
1792 		return r;
1793 
1794 	kvm_ops_pr.owner = THIS_MODULE;
1795 	kvmppc_pr_ops = &kvm_ops_pr;
1796 
1797 	r = kvmppc_mmu_hpte_sysinit();
1798 	return r;
1799 }
1800 
1801 void kvmppc_book3s_exit_pr(void)
1802 {
1803 	kvmppc_pr_ops = NULL;
1804 	kvmppc_mmu_hpte_sysexit();
1805 }
1806 
1807 /*
1808  * We only support separate modules for book3s 64
1809  */
1810 #ifdef CONFIG_PPC_BOOK3S_64
1811 
1812 module_init(kvmppc_book3s_init_pr);
1813 module_exit(kvmppc_book3s_exit_pr);
1814 
1815 MODULE_LICENSE("GPL");
1816 MODULE_ALIAS_MISCDEV(KVM_MINOR);
1817 MODULE_ALIAS("devname:kvm");
1818 #endif
1819