xref: /openbmc/linux/arch/powerpc/kvm/book3s_pr.c (revision 82806c25)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
4  *
5  * Authors:
6  *    Alexander Graf <agraf@suse.de>
7  *    Kevin Wolf <mail@kevin-wolf.de>
8  *    Paul Mackerras <paulus@samba.org>
9  *
10  * Description:
11  * Functions relating to running KVM on Book 3S processors where
12  * we don't have access to hypervisor mode, and we run the guest
13  * in problem state (user mode).
14  *
15  * This file is derived from arch/powerpc/kvm/44x.c,
16  * by Hollis Blanchard <hollisb@us.ibm.com>.
17  */
18 
19 #include <linux/kvm_host.h>
20 #include <linux/export.h>
21 #include <linux/err.h>
22 #include <linux/slab.h>
23 
24 #include <asm/reg.h>
25 #include <asm/cputable.h>
26 #include <asm/cacheflush.h>
27 #include <linux/uaccess.h>
28 #include <asm/interrupt.h>
29 #include <asm/io.h>
30 #include <asm/kvm_ppc.h>
31 #include <asm/kvm_book3s.h>
32 #include <asm/mmu_context.h>
33 #include <asm/switch_to.h>
34 #include <asm/firmware.h>
35 #include <asm/setup.h>
36 #include <linux/gfp.h>
37 #include <linux/sched.h>
38 #include <linux/vmalloc.h>
39 #include <linux/highmem.h>
40 #include <linux/module.h>
41 #include <linux/miscdevice.h>
42 #include <asm/asm-prototypes.h>
43 #include <asm/tm.h>
44 
45 #include "book3s.h"
46 
47 #define CREATE_TRACE_POINTS
48 #include "trace_pr.h"
49 
50 /* #define EXIT_DEBUG */
51 /* #define DEBUG_EXT */
52 
53 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
54 			     ulong msr);
55 #ifdef CONFIG_PPC_BOOK3S_64
56 static int kvmppc_handle_fac(struct kvm_vcpu *vcpu, ulong fac);
57 #endif
58 
59 /* Some compatibility defines */
60 #ifdef CONFIG_PPC_BOOK3S_32
61 #define MSR_USER32 MSR_USER
62 #define MSR_USER64 MSR_USER
63 #define HW_PAGE_SIZE PAGE_SIZE
64 #define HPTE_R_M   _PAGE_COHERENT
65 #endif
66 
67 static bool kvmppc_is_split_real(struct kvm_vcpu *vcpu)
68 {
69 	ulong msr = kvmppc_get_msr(vcpu);
70 	return (msr & (MSR_IR|MSR_DR)) == MSR_DR;
71 }
72 
73 static void kvmppc_fixup_split_real(struct kvm_vcpu *vcpu)
74 {
75 	ulong msr = kvmppc_get_msr(vcpu);
76 	ulong pc = kvmppc_get_pc(vcpu);
77 
78 	/* We are in DR only split real mode */
79 	if ((msr & (MSR_IR|MSR_DR)) != MSR_DR)
80 		return;
81 
82 	/* We have not fixed up the guest already */
83 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK)
84 		return;
85 
86 	/* The code is in fixupable address space */
87 	if (pc & SPLIT_HACK_MASK)
88 		return;
89 
90 	vcpu->arch.hflags |= BOOK3S_HFLAG_SPLIT_HACK;
91 	kvmppc_set_pc(vcpu, pc | SPLIT_HACK_OFFS);
92 }
93 
94 static void kvmppc_unfixup_split_real(struct kvm_vcpu *vcpu)
95 {
96 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK) {
97 		ulong pc = kvmppc_get_pc(vcpu);
98 		ulong lr = kvmppc_get_lr(vcpu);
99 		if ((pc & SPLIT_HACK_MASK) == SPLIT_HACK_OFFS)
100 			kvmppc_set_pc(vcpu, pc & ~SPLIT_HACK_MASK);
101 		if ((lr & SPLIT_HACK_MASK) == SPLIT_HACK_OFFS)
102 			kvmppc_set_lr(vcpu, lr & ~SPLIT_HACK_MASK);
103 		vcpu->arch.hflags &= ~BOOK3S_HFLAG_SPLIT_HACK;
104 	}
105 }
106 
107 static void kvmppc_inject_interrupt_pr(struct kvm_vcpu *vcpu, int vec, u64 srr1_flags)
108 {
109 	unsigned long msr, pc, new_msr, new_pc;
110 
111 	kvmppc_unfixup_split_real(vcpu);
112 
113 	msr = kvmppc_get_msr(vcpu);
114 	pc = kvmppc_get_pc(vcpu);
115 	new_msr = vcpu->arch.intr_msr;
116 	new_pc = to_book3s(vcpu)->hior + vec;
117 
118 #ifdef CONFIG_PPC_BOOK3S_64
119 	/* If transactional, change to suspend mode on IRQ delivery */
120 	if (MSR_TM_TRANSACTIONAL(msr))
121 		new_msr |= MSR_TS_S;
122 	else
123 		new_msr |= msr & MSR_TS_MASK;
124 #endif
125 
126 	kvmppc_set_srr0(vcpu, pc);
127 	kvmppc_set_srr1(vcpu, (msr & SRR1_MSR_BITS) | srr1_flags);
128 	kvmppc_set_pc(vcpu, new_pc);
129 	kvmppc_set_msr(vcpu, new_msr);
130 }
131 
132 static void kvmppc_core_vcpu_load_pr(struct kvm_vcpu *vcpu, int cpu)
133 {
134 #ifdef CONFIG_PPC_BOOK3S_64
135 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
136 	memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb));
137 	svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max;
138 	svcpu->in_use = 0;
139 	svcpu_put(svcpu);
140 
141 	/* Disable AIL if supported */
142 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
143 		if (cpu_has_feature(CPU_FTR_ARCH_207S))
144 			mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) & ~LPCR_AIL);
145 		if (cpu_has_feature(CPU_FTR_ARCH_300) && (current->thread.fscr & FSCR_SCV))
146 			mtspr(SPRN_FSCR, mfspr(SPRN_FSCR) & ~FSCR_SCV);
147 	}
148 #endif
149 
150 	vcpu->cpu = smp_processor_id();
151 #ifdef CONFIG_PPC_BOOK3S_32
152 	current->thread.kvm_shadow_vcpu = vcpu->arch.shadow_vcpu;
153 #endif
154 
155 	if (kvmppc_is_split_real(vcpu))
156 		kvmppc_fixup_split_real(vcpu);
157 
158 	kvmppc_restore_tm_pr(vcpu);
159 }
160 
161 static void kvmppc_core_vcpu_put_pr(struct kvm_vcpu *vcpu)
162 {
163 #ifdef CONFIG_PPC_BOOK3S_64
164 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
165 	if (svcpu->in_use) {
166 		kvmppc_copy_from_svcpu(vcpu);
167 	}
168 	memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
169 	to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max;
170 	svcpu_put(svcpu);
171 
172 	/* Enable AIL if supported */
173 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
174 		if (cpu_has_feature(CPU_FTR_ARCH_207S))
175 			mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_AIL_3);
176 		if (cpu_has_feature(CPU_FTR_ARCH_300) && (current->thread.fscr & FSCR_SCV))
177 			mtspr(SPRN_FSCR, mfspr(SPRN_FSCR) | FSCR_SCV);
178 	}
179 #endif
180 
181 	if (kvmppc_is_split_real(vcpu))
182 		kvmppc_unfixup_split_real(vcpu);
183 
184 	kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
185 	kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
186 	kvmppc_save_tm_pr(vcpu);
187 
188 	vcpu->cpu = -1;
189 }
190 
191 /* Copy data needed by real-mode code from vcpu to shadow vcpu */
192 void kvmppc_copy_to_svcpu(struct kvm_vcpu *vcpu)
193 {
194 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
195 
196 	svcpu->gpr[0] = vcpu->arch.regs.gpr[0];
197 	svcpu->gpr[1] = vcpu->arch.regs.gpr[1];
198 	svcpu->gpr[2] = vcpu->arch.regs.gpr[2];
199 	svcpu->gpr[3] = vcpu->arch.regs.gpr[3];
200 	svcpu->gpr[4] = vcpu->arch.regs.gpr[4];
201 	svcpu->gpr[5] = vcpu->arch.regs.gpr[5];
202 	svcpu->gpr[6] = vcpu->arch.regs.gpr[6];
203 	svcpu->gpr[7] = vcpu->arch.regs.gpr[7];
204 	svcpu->gpr[8] = vcpu->arch.regs.gpr[8];
205 	svcpu->gpr[9] = vcpu->arch.regs.gpr[9];
206 	svcpu->gpr[10] = vcpu->arch.regs.gpr[10];
207 	svcpu->gpr[11] = vcpu->arch.regs.gpr[11];
208 	svcpu->gpr[12] = vcpu->arch.regs.gpr[12];
209 	svcpu->gpr[13] = vcpu->arch.regs.gpr[13];
210 	svcpu->cr  = vcpu->arch.regs.ccr;
211 	svcpu->xer = vcpu->arch.regs.xer;
212 	svcpu->ctr = vcpu->arch.regs.ctr;
213 	svcpu->lr  = vcpu->arch.regs.link;
214 	svcpu->pc  = vcpu->arch.regs.nip;
215 #ifdef CONFIG_PPC_BOOK3S_64
216 	svcpu->shadow_fscr = vcpu->arch.shadow_fscr;
217 #endif
218 	/*
219 	 * Now also save the current time base value. We use this
220 	 * to find the guest purr and spurr value.
221 	 */
222 	vcpu->arch.entry_tb = get_tb();
223 	vcpu->arch.entry_vtb = get_vtb();
224 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
225 		vcpu->arch.entry_ic = mfspr(SPRN_IC);
226 	svcpu->in_use = true;
227 
228 	svcpu_put(svcpu);
229 }
230 
231 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
232 {
233 	ulong guest_msr = kvmppc_get_msr(vcpu);
234 	ulong smsr = guest_msr;
235 
236 	/* Guest MSR values */
237 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
238 	smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_LE |
239 		MSR_TM | MSR_TS_MASK;
240 #else
241 	smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_LE;
242 #endif
243 	/* Process MSR values */
244 	smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
245 	/* External providers the guest reserved */
246 	smsr |= (guest_msr & vcpu->arch.guest_owned_ext);
247 	/* 64-bit Process MSR values */
248 #ifdef CONFIG_PPC_BOOK3S_64
249 	smsr |= MSR_HV;
250 #endif
251 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
252 	/*
253 	 * in guest privileged state, we want to fail all TM transactions.
254 	 * So disable MSR TM bit so that all tbegin. will be able to be
255 	 * trapped into host.
256 	 */
257 	if (!(guest_msr & MSR_PR))
258 		smsr &= ~MSR_TM;
259 #endif
260 	vcpu->arch.shadow_msr = smsr;
261 }
262 
263 /* Copy data touched by real-mode code from shadow vcpu back to vcpu */
264 void kvmppc_copy_from_svcpu(struct kvm_vcpu *vcpu)
265 {
266 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
267 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
268 	ulong old_msr;
269 #endif
270 
271 	/*
272 	 * Maybe we were already preempted and synced the svcpu from
273 	 * our preempt notifiers. Don't bother touching this svcpu then.
274 	 */
275 	if (!svcpu->in_use)
276 		goto out;
277 
278 	vcpu->arch.regs.gpr[0] = svcpu->gpr[0];
279 	vcpu->arch.regs.gpr[1] = svcpu->gpr[1];
280 	vcpu->arch.regs.gpr[2] = svcpu->gpr[2];
281 	vcpu->arch.regs.gpr[3] = svcpu->gpr[3];
282 	vcpu->arch.regs.gpr[4] = svcpu->gpr[4];
283 	vcpu->arch.regs.gpr[5] = svcpu->gpr[5];
284 	vcpu->arch.regs.gpr[6] = svcpu->gpr[6];
285 	vcpu->arch.regs.gpr[7] = svcpu->gpr[7];
286 	vcpu->arch.regs.gpr[8] = svcpu->gpr[8];
287 	vcpu->arch.regs.gpr[9] = svcpu->gpr[9];
288 	vcpu->arch.regs.gpr[10] = svcpu->gpr[10];
289 	vcpu->arch.regs.gpr[11] = svcpu->gpr[11];
290 	vcpu->arch.regs.gpr[12] = svcpu->gpr[12];
291 	vcpu->arch.regs.gpr[13] = svcpu->gpr[13];
292 	vcpu->arch.regs.ccr  = svcpu->cr;
293 	vcpu->arch.regs.xer = svcpu->xer;
294 	vcpu->arch.regs.ctr = svcpu->ctr;
295 	vcpu->arch.regs.link  = svcpu->lr;
296 	vcpu->arch.regs.nip  = svcpu->pc;
297 	vcpu->arch.shadow_srr1 = svcpu->shadow_srr1;
298 	vcpu->arch.fault_dar   = svcpu->fault_dar;
299 	vcpu->arch.fault_dsisr = svcpu->fault_dsisr;
300 	vcpu->arch.last_inst   = svcpu->last_inst;
301 #ifdef CONFIG_PPC_BOOK3S_64
302 	vcpu->arch.shadow_fscr = svcpu->shadow_fscr;
303 #endif
304 	/*
305 	 * Update purr and spurr using time base on exit.
306 	 */
307 	vcpu->arch.purr += get_tb() - vcpu->arch.entry_tb;
308 	vcpu->arch.spurr += get_tb() - vcpu->arch.entry_tb;
309 	to_book3s(vcpu)->vtb += get_vtb() - vcpu->arch.entry_vtb;
310 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
311 		vcpu->arch.ic += mfspr(SPRN_IC) - vcpu->arch.entry_ic;
312 
313 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
314 	/*
315 	 * Unlike other MSR bits, MSR[TS]bits can be changed at guest without
316 	 * notifying host:
317 	 *  modified by unprivileged instructions like "tbegin"/"tend"/
318 	 * "tresume"/"tsuspend" in PR KVM guest.
319 	 *
320 	 * It is necessary to sync here to calculate a correct shadow_msr.
321 	 *
322 	 * privileged guest's tbegin will be failed at present. So we
323 	 * only take care of problem state guest.
324 	 */
325 	old_msr = kvmppc_get_msr(vcpu);
326 	if (unlikely((old_msr & MSR_PR) &&
327 		(vcpu->arch.shadow_srr1 & (MSR_TS_MASK)) !=
328 				(old_msr & (MSR_TS_MASK)))) {
329 		old_msr &= ~(MSR_TS_MASK);
330 		old_msr |= (vcpu->arch.shadow_srr1 & (MSR_TS_MASK));
331 		kvmppc_set_msr_fast(vcpu, old_msr);
332 		kvmppc_recalc_shadow_msr(vcpu);
333 	}
334 #endif
335 
336 	svcpu->in_use = false;
337 
338 out:
339 	svcpu_put(svcpu);
340 }
341 
342 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
343 void kvmppc_save_tm_sprs(struct kvm_vcpu *vcpu)
344 {
345 	tm_enable();
346 	vcpu->arch.tfhar = mfspr(SPRN_TFHAR);
347 	vcpu->arch.texasr = mfspr(SPRN_TEXASR);
348 	vcpu->arch.tfiar = mfspr(SPRN_TFIAR);
349 	tm_disable();
350 }
351 
352 void kvmppc_restore_tm_sprs(struct kvm_vcpu *vcpu)
353 {
354 	tm_enable();
355 	mtspr(SPRN_TFHAR, vcpu->arch.tfhar);
356 	mtspr(SPRN_TEXASR, vcpu->arch.texasr);
357 	mtspr(SPRN_TFIAR, vcpu->arch.tfiar);
358 	tm_disable();
359 }
360 
361 /* loadup math bits which is enabled at kvmppc_get_msr() but not enabled at
362  * hardware.
363  */
364 static void kvmppc_handle_lost_math_exts(struct kvm_vcpu *vcpu)
365 {
366 	ulong exit_nr;
367 	ulong ext_diff = (kvmppc_get_msr(vcpu) & ~vcpu->arch.guest_owned_ext) &
368 		(MSR_FP | MSR_VEC | MSR_VSX);
369 
370 	if (!ext_diff)
371 		return;
372 
373 	if (ext_diff == MSR_FP)
374 		exit_nr = BOOK3S_INTERRUPT_FP_UNAVAIL;
375 	else if (ext_diff == MSR_VEC)
376 		exit_nr = BOOK3S_INTERRUPT_ALTIVEC;
377 	else
378 		exit_nr = BOOK3S_INTERRUPT_VSX;
379 
380 	kvmppc_handle_ext(vcpu, exit_nr, ext_diff);
381 }
382 
383 void kvmppc_save_tm_pr(struct kvm_vcpu *vcpu)
384 {
385 	if (!(MSR_TM_ACTIVE(kvmppc_get_msr(vcpu)))) {
386 		kvmppc_save_tm_sprs(vcpu);
387 		return;
388 	}
389 
390 	kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
391 	kvmppc_giveup_ext(vcpu, MSR_VSX);
392 
393 	preempt_disable();
394 	_kvmppc_save_tm_pr(vcpu, mfmsr());
395 	preempt_enable();
396 }
397 
398 void kvmppc_restore_tm_pr(struct kvm_vcpu *vcpu)
399 {
400 	if (!MSR_TM_ACTIVE(kvmppc_get_msr(vcpu))) {
401 		kvmppc_restore_tm_sprs(vcpu);
402 		if (kvmppc_get_msr(vcpu) & MSR_TM) {
403 			kvmppc_handle_lost_math_exts(vcpu);
404 			if (vcpu->arch.fscr & FSCR_TAR)
405 				kvmppc_handle_fac(vcpu, FSCR_TAR_LG);
406 		}
407 		return;
408 	}
409 
410 	preempt_disable();
411 	_kvmppc_restore_tm_pr(vcpu, kvmppc_get_msr(vcpu));
412 	preempt_enable();
413 
414 	if (kvmppc_get_msr(vcpu) & MSR_TM) {
415 		kvmppc_handle_lost_math_exts(vcpu);
416 		if (vcpu->arch.fscr & FSCR_TAR)
417 			kvmppc_handle_fac(vcpu, FSCR_TAR_LG);
418 	}
419 }
420 #endif
421 
422 static int kvmppc_core_check_requests_pr(struct kvm_vcpu *vcpu)
423 {
424 	int r = 1; /* Indicate we want to get back into the guest */
425 
426 	/* We misuse TLB_FLUSH to indicate that we want to clear
427 	   all shadow cache entries */
428 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
429 		kvmppc_mmu_pte_flush(vcpu, 0, 0);
430 
431 	return r;
432 }
433 
434 /************* MMU Notifiers *************/
435 static bool do_kvm_unmap_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
436 {
437 	unsigned long i;
438 	struct kvm_vcpu *vcpu;
439 
440 	kvm_for_each_vcpu(i, vcpu, kvm)
441 		kvmppc_mmu_pte_pflush(vcpu, range->start << PAGE_SHIFT,
442 				      range->end << PAGE_SHIFT);
443 
444 	return false;
445 }
446 
447 static bool kvm_unmap_gfn_range_pr(struct kvm *kvm, struct kvm_gfn_range *range)
448 {
449 	return do_kvm_unmap_gfn(kvm, range);
450 }
451 
452 static bool kvm_age_gfn_pr(struct kvm *kvm, struct kvm_gfn_range *range)
453 {
454 	/* XXX could be more clever ;) */
455 	return false;
456 }
457 
458 static bool kvm_test_age_gfn_pr(struct kvm *kvm, struct kvm_gfn_range *range)
459 {
460 	/* XXX could be more clever ;) */
461 	return false;
462 }
463 
464 static bool kvm_set_spte_gfn_pr(struct kvm *kvm, struct kvm_gfn_range *range)
465 {
466 	/* The page will get remapped properly on its next fault */
467 	return do_kvm_unmap_gfn(kvm, range);
468 }
469 
470 /*****************************************/
471 
472 static void kvmppc_set_msr_pr(struct kvm_vcpu *vcpu, u64 msr)
473 {
474 	ulong old_msr;
475 
476 	/* For PAPR guest, make sure MSR reflects guest mode */
477 	if (vcpu->arch.papr_enabled)
478 		msr = (msr & ~MSR_HV) | MSR_ME;
479 
480 #ifdef EXIT_DEBUG
481 	printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
482 #endif
483 
484 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
485 	/* We should never target guest MSR to TS=10 && PR=0,
486 	 * since we always fail transaction for guest privilege
487 	 * state.
488 	 */
489 	if (!(msr & MSR_PR) && MSR_TM_TRANSACTIONAL(msr))
490 		kvmppc_emulate_tabort(vcpu,
491 			TM_CAUSE_KVM_FAC_UNAV | TM_CAUSE_PERSISTENT);
492 #endif
493 
494 	old_msr = kvmppc_get_msr(vcpu);
495 	msr &= to_book3s(vcpu)->msr_mask;
496 	kvmppc_set_msr_fast(vcpu, msr);
497 	kvmppc_recalc_shadow_msr(vcpu);
498 
499 	if (msr & MSR_POW) {
500 		if (!vcpu->arch.pending_exceptions) {
501 			kvm_vcpu_halt(vcpu);
502 			kvm_clear_request(KVM_REQ_UNHALT, vcpu);
503 			vcpu->stat.generic.halt_wakeup++;
504 
505 			/* Unset POW bit after we woke up */
506 			msr &= ~MSR_POW;
507 			kvmppc_set_msr_fast(vcpu, msr);
508 		}
509 	}
510 
511 	if (kvmppc_is_split_real(vcpu))
512 		kvmppc_fixup_split_real(vcpu);
513 	else
514 		kvmppc_unfixup_split_real(vcpu);
515 
516 	if ((kvmppc_get_msr(vcpu) & (MSR_PR|MSR_IR|MSR_DR)) !=
517 		   (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
518 		kvmppc_mmu_flush_segments(vcpu);
519 		kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
520 
521 		/* Preload magic page segment when in kernel mode */
522 		if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
523 			struct kvm_vcpu_arch *a = &vcpu->arch;
524 
525 			if (msr & MSR_DR)
526 				kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
527 			else
528 				kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
529 		}
530 	}
531 
532 	/*
533 	 * When switching from 32 to 64-bit, we may have a stale 32-bit
534 	 * magic page around, we need to flush it. Typically 32-bit magic
535 	 * page will be instantiated when calling into RTAS. Note: We
536 	 * assume that such transition only happens while in kernel mode,
537 	 * ie, we never transition from user 32-bit to kernel 64-bit with
538 	 * a 32-bit magic page around.
539 	 */
540 	if (vcpu->arch.magic_page_pa &&
541 	    !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
542 		/* going from RTAS to normal kernel code */
543 		kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
544 				     ~0xFFFUL);
545 	}
546 
547 	/* Preload FPU if it's enabled */
548 	if (kvmppc_get_msr(vcpu) & MSR_FP)
549 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
550 
551 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
552 	if (kvmppc_get_msr(vcpu) & MSR_TM)
553 		kvmppc_handle_lost_math_exts(vcpu);
554 #endif
555 }
556 
557 static void kvmppc_set_pvr_pr(struct kvm_vcpu *vcpu, u32 pvr)
558 {
559 	u32 host_pvr;
560 
561 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
562 	vcpu->arch.pvr = pvr;
563 #ifdef CONFIG_PPC_BOOK3S_64
564 	if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
565 		kvmppc_mmu_book3s_64_init(vcpu);
566 		if (!to_book3s(vcpu)->hior_explicit)
567 			to_book3s(vcpu)->hior = 0xfff00000;
568 		to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
569 		vcpu->arch.cpu_type = KVM_CPU_3S_64;
570 	} else
571 #endif
572 	{
573 		kvmppc_mmu_book3s_32_init(vcpu);
574 		if (!to_book3s(vcpu)->hior_explicit)
575 			to_book3s(vcpu)->hior = 0;
576 		to_book3s(vcpu)->msr_mask = 0xffffffffULL;
577 		vcpu->arch.cpu_type = KVM_CPU_3S_32;
578 	}
579 
580 	kvmppc_sanity_check(vcpu);
581 
582 	/* If we are in hypervisor level on 970, we can tell the CPU to
583 	 * treat DCBZ as 32 bytes store */
584 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
585 	if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
586 	    !strcmp(cur_cpu_spec->platform, "ppc970"))
587 		vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
588 
589 	/* Cell performs badly if MSR_FEx are set. So let's hope nobody
590 	   really needs them in a VM on Cell and force disable them. */
591 	if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
592 		to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
593 
594 	/*
595 	 * If they're asking for POWER6 or later, set the flag
596 	 * indicating that we can do multiple large page sizes
597 	 * and 1TB segments.
598 	 * Also set the flag that indicates that tlbie has the large
599 	 * page bit in the RB operand instead of the instruction.
600 	 */
601 	switch (PVR_VER(pvr)) {
602 	case PVR_POWER6:
603 	case PVR_POWER7:
604 	case PVR_POWER7p:
605 	case PVR_POWER8:
606 	case PVR_POWER8E:
607 	case PVR_POWER8NVL:
608 	case PVR_POWER9:
609 		vcpu->arch.hflags |= BOOK3S_HFLAG_MULTI_PGSIZE |
610 			BOOK3S_HFLAG_NEW_TLBIE;
611 		break;
612 	}
613 
614 #ifdef CONFIG_PPC_BOOK3S_32
615 	/* 32 bit Book3S always has 32 byte dcbz */
616 	vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
617 #endif
618 
619 	/* On some CPUs we can execute paired single operations natively */
620 	asm ( "mfpvr %0" : "=r"(host_pvr));
621 	switch (host_pvr) {
622 	case 0x00080200:	/* lonestar 2.0 */
623 	case 0x00088202:	/* lonestar 2.2 */
624 	case 0x70000100:	/* gekko 1.0 */
625 	case 0x00080100:	/* gekko 2.0 */
626 	case 0x00083203:	/* gekko 2.3a */
627 	case 0x00083213:	/* gekko 2.3b */
628 	case 0x00083204:	/* gekko 2.4 */
629 	case 0x00083214:	/* gekko 2.4e (8SE) - retail HW2 */
630 	case 0x00087200:	/* broadway */
631 		vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
632 		/* Enable HID2.PSE - in case we need it later */
633 		mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
634 	}
635 }
636 
637 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
638  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
639  * emulate 32 bytes dcbz length.
640  *
641  * The Book3s_64 inventors also realized this case and implemented a special bit
642  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
643  *
644  * My approach here is to patch the dcbz instruction on executing pages.
645  */
646 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
647 {
648 	struct page *hpage;
649 	u64 hpage_offset;
650 	u32 *page;
651 	int i;
652 
653 	hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
654 	if (is_error_page(hpage))
655 		return;
656 
657 	hpage_offset = pte->raddr & ~PAGE_MASK;
658 	hpage_offset &= ~0xFFFULL;
659 	hpage_offset /= 4;
660 
661 	get_page(hpage);
662 	page = kmap_atomic(hpage);
663 
664 	/* patch dcbz into reserved instruction, so we trap */
665 	for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
666 		if ((be32_to_cpu(page[i]) & 0xff0007ff) == INS_DCBZ)
667 			page[i] &= cpu_to_be32(0xfffffff7);
668 
669 	kunmap_atomic(page);
670 	put_page(hpage);
671 }
672 
673 static bool kvmppc_visible_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
674 {
675 	ulong mp_pa = vcpu->arch.magic_page_pa;
676 
677 	if (!(kvmppc_get_msr(vcpu) & MSR_SF))
678 		mp_pa = (uint32_t)mp_pa;
679 
680 	gpa &= ~0xFFFULL;
681 	if (unlikely(mp_pa) && unlikely((mp_pa & KVM_PAM) == (gpa & KVM_PAM))) {
682 		return true;
683 	}
684 
685 	return kvm_is_visible_gfn(vcpu->kvm, gpa >> PAGE_SHIFT);
686 }
687 
688 static int kvmppc_handle_pagefault(struct kvm_vcpu *vcpu,
689 			    ulong eaddr, int vec)
690 {
691 	bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
692 	bool iswrite = false;
693 	int r = RESUME_GUEST;
694 	int relocated;
695 	int page_found = 0;
696 	struct kvmppc_pte pte = { 0 };
697 	bool dr = (kvmppc_get_msr(vcpu) & MSR_DR) ? true : false;
698 	bool ir = (kvmppc_get_msr(vcpu) & MSR_IR) ? true : false;
699 	u64 vsid;
700 
701 	relocated = data ? dr : ir;
702 	if (data && (vcpu->arch.fault_dsisr & DSISR_ISSTORE))
703 		iswrite = true;
704 
705 	/* Resolve real address if translation turned on */
706 	if (relocated) {
707 		page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data, iswrite);
708 	} else {
709 		pte.may_execute = true;
710 		pte.may_read = true;
711 		pte.may_write = true;
712 		pte.raddr = eaddr & KVM_PAM;
713 		pte.eaddr = eaddr;
714 		pte.vpage = eaddr >> 12;
715 		pte.page_size = MMU_PAGE_64K;
716 		pte.wimg = HPTE_R_M;
717 	}
718 
719 	switch (kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) {
720 	case 0:
721 		pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
722 		break;
723 	case MSR_DR:
724 		if (!data &&
725 		    (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK) &&
726 		    ((pte.raddr & SPLIT_HACK_MASK) == SPLIT_HACK_OFFS))
727 			pte.raddr &= ~SPLIT_HACK_MASK;
728 		fallthrough;
729 	case MSR_IR:
730 		vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
731 
732 		if ((kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) == MSR_DR)
733 			pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
734 		else
735 			pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
736 		pte.vpage |= vsid;
737 
738 		if (vsid == -1)
739 			page_found = -EINVAL;
740 		break;
741 	}
742 
743 	if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
744 	   (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
745 		/*
746 		 * If we do the dcbz hack, we have to NX on every execution,
747 		 * so we can patch the executing code. This renders our guest
748 		 * NX-less.
749 		 */
750 		pte.may_execute = !data;
751 	}
752 
753 	if (page_found == -ENOENT || page_found == -EPERM) {
754 		/* Page not found in guest PTE entries, or protection fault */
755 		u64 flags;
756 
757 		if (page_found == -EPERM)
758 			flags = DSISR_PROTFAULT;
759 		else
760 			flags = DSISR_NOHPTE;
761 		if (data) {
762 			flags |= vcpu->arch.fault_dsisr & DSISR_ISSTORE;
763 			kvmppc_core_queue_data_storage(vcpu, eaddr, flags);
764 		} else {
765 			kvmppc_core_queue_inst_storage(vcpu, flags);
766 		}
767 	} else if (page_found == -EINVAL) {
768 		/* Page not found in guest SLB */
769 		kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
770 		kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
771 	} else if (kvmppc_visible_gpa(vcpu, pte.raddr)) {
772 		if (data && !(vcpu->arch.fault_dsisr & DSISR_NOHPTE)) {
773 			/*
774 			 * There is already a host HPTE there, presumably
775 			 * a read-only one for a page the guest thinks
776 			 * is writable, so get rid of it first.
777 			 */
778 			kvmppc_mmu_unmap_page(vcpu, &pte);
779 		}
780 		/* The guest's PTE is not mapped yet. Map on the host */
781 		if (kvmppc_mmu_map_page(vcpu, &pte, iswrite) == -EIO) {
782 			/* Exit KVM if mapping failed */
783 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
784 			return RESUME_HOST;
785 		}
786 		if (data)
787 			vcpu->stat.sp_storage++;
788 		else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
789 			 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
790 			kvmppc_patch_dcbz(vcpu, &pte);
791 	} else {
792 		/* MMIO */
793 		vcpu->stat.mmio_exits++;
794 		vcpu->arch.paddr_accessed = pte.raddr;
795 		vcpu->arch.vaddr_accessed = pte.eaddr;
796 		r = kvmppc_emulate_mmio(vcpu);
797 		if ( r == RESUME_HOST_NV )
798 			r = RESUME_HOST;
799 	}
800 
801 	return r;
802 }
803 
804 /* Give up external provider (FPU, Altivec, VSX) */
805 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
806 {
807 	struct thread_struct *t = &current->thread;
808 
809 	/*
810 	 * VSX instructions can access FP and vector registers, so if
811 	 * we are giving up VSX, make sure we give up FP and VMX as well.
812 	 */
813 	if (msr & MSR_VSX)
814 		msr |= MSR_FP | MSR_VEC;
815 
816 	msr &= vcpu->arch.guest_owned_ext;
817 	if (!msr)
818 		return;
819 
820 #ifdef DEBUG_EXT
821 	printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
822 #endif
823 
824 	if (msr & MSR_FP) {
825 		/*
826 		 * Note that on CPUs with VSX, giveup_fpu stores
827 		 * both the traditional FP registers and the added VSX
828 		 * registers into thread.fp_state.fpr[].
829 		 */
830 		if (t->regs->msr & MSR_FP)
831 			giveup_fpu(current);
832 		t->fp_save_area = NULL;
833 	}
834 
835 #ifdef CONFIG_ALTIVEC
836 	if (msr & MSR_VEC) {
837 		if (current->thread.regs->msr & MSR_VEC)
838 			giveup_altivec(current);
839 		t->vr_save_area = NULL;
840 	}
841 #endif
842 
843 	vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
844 	kvmppc_recalc_shadow_msr(vcpu);
845 }
846 
847 /* Give up facility (TAR / EBB / DSCR) */
848 void kvmppc_giveup_fac(struct kvm_vcpu *vcpu, ulong fac)
849 {
850 #ifdef CONFIG_PPC_BOOK3S_64
851 	if (!(vcpu->arch.shadow_fscr & (1ULL << fac))) {
852 		/* Facility not available to the guest, ignore giveup request*/
853 		return;
854 	}
855 
856 	switch (fac) {
857 	case FSCR_TAR_LG:
858 		vcpu->arch.tar = mfspr(SPRN_TAR);
859 		mtspr(SPRN_TAR, current->thread.tar);
860 		vcpu->arch.shadow_fscr &= ~FSCR_TAR;
861 		break;
862 	}
863 #endif
864 }
865 
866 /* Handle external providers (FPU, Altivec, VSX) */
867 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
868 			     ulong msr)
869 {
870 	struct thread_struct *t = &current->thread;
871 
872 	/* When we have paired singles, we emulate in software */
873 	if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
874 		return RESUME_GUEST;
875 
876 	if (!(kvmppc_get_msr(vcpu) & msr)) {
877 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
878 		return RESUME_GUEST;
879 	}
880 
881 	if (msr == MSR_VSX) {
882 		/* No VSX?  Give an illegal instruction interrupt */
883 #ifdef CONFIG_VSX
884 		if (!cpu_has_feature(CPU_FTR_VSX))
885 #endif
886 		{
887 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
888 			return RESUME_GUEST;
889 		}
890 
891 		/*
892 		 * We have to load up all the FP and VMX registers before
893 		 * we can let the guest use VSX instructions.
894 		 */
895 		msr = MSR_FP | MSR_VEC | MSR_VSX;
896 	}
897 
898 	/* See if we already own all the ext(s) needed */
899 	msr &= ~vcpu->arch.guest_owned_ext;
900 	if (!msr)
901 		return RESUME_GUEST;
902 
903 #ifdef DEBUG_EXT
904 	printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
905 #endif
906 
907 	if (msr & MSR_FP) {
908 		preempt_disable();
909 		enable_kernel_fp();
910 		load_fp_state(&vcpu->arch.fp);
911 		disable_kernel_fp();
912 		t->fp_save_area = &vcpu->arch.fp;
913 		preempt_enable();
914 	}
915 
916 	if (msr & MSR_VEC) {
917 #ifdef CONFIG_ALTIVEC
918 		preempt_disable();
919 		enable_kernel_altivec();
920 		load_vr_state(&vcpu->arch.vr);
921 		disable_kernel_altivec();
922 		t->vr_save_area = &vcpu->arch.vr;
923 		preempt_enable();
924 #endif
925 	}
926 
927 	t->regs->msr |= msr;
928 	vcpu->arch.guest_owned_ext |= msr;
929 	kvmppc_recalc_shadow_msr(vcpu);
930 
931 	return RESUME_GUEST;
932 }
933 
934 /*
935  * Kernel code using FP or VMX could have flushed guest state to
936  * the thread_struct; if so, get it back now.
937  */
938 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
939 {
940 	unsigned long lost_ext;
941 
942 	lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
943 	if (!lost_ext)
944 		return;
945 
946 	if (lost_ext & MSR_FP) {
947 		preempt_disable();
948 		enable_kernel_fp();
949 		load_fp_state(&vcpu->arch.fp);
950 		disable_kernel_fp();
951 		preempt_enable();
952 	}
953 #ifdef CONFIG_ALTIVEC
954 	if (lost_ext & MSR_VEC) {
955 		preempt_disable();
956 		enable_kernel_altivec();
957 		load_vr_state(&vcpu->arch.vr);
958 		disable_kernel_altivec();
959 		preempt_enable();
960 	}
961 #endif
962 	current->thread.regs->msr |= lost_ext;
963 }
964 
965 #ifdef CONFIG_PPC_BOOK3S_64
966 
967 void kvmppc_trigger_fac_interrupt(struct kvm_vcpu *vcpu, ulong fac)
968 {
969 	/* Inject the Interrupt Cause field and trigger a guest interrupt */
970 	vcpu->arch.fscr &= ~(0xffULL << 56);
971 	vcpu->arch.fscr |= (fac << 56);
972 	kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_FAC_UNAVAIL);
973 }
974 
975 static void kvmppc_emulate_fac(struct kvm_vcpu *vcpu, ulong fac)
976 {
977 	enum emulation_result er = EMULATE_FAIL;
978 
979 	if (!(kvmppc_get_msr(vcpu) & MSR_PR))
980 		er = kvmppc_emulate_instruction(vcpu);
981 
982 	if ((er != EMULATE_DONE) && (er != EMULATE_AGAIN)) {
983 		/* Couldn't emulate, trigger interrupt in guest */
984 		kvmppc_trigger_fac_interrupt(vcpu, fac);
985 	}
986 }
987 
988 /* Enable facilities (TAR, EBB, DSCR) for the guest */
989 static int kvmppc_handle_fac(struct kvm_vcpu *vcpu, ulong fac)
990 {
991 	bool guest_fac_enabled;
992 	BUG_ON(!cpu_has_feature(CPU_FTR_ARCH_207S));
993 
994 	/*
995 	 * Not every facility is enabled by FSCR bits, check whether the
996 	 * guest has this facility enabled at all.
997 	 */
998 	switch (fac) {
999 	case FSCR_TAR_LG:
1000 	case FSCR_EBB_LG:
1001 		guest_fac_enabled = (vcpu->arch.fscr & (1ULL << fac));
1002 		break;
1003 	case FSCR_TM_LG:
1004 		guest_fac_enabled = kvmppc_get_msr(vcpu) & MSR_TM;
1005 		break;
1006 	default:
1007 		guest_fac_enabled = false;
1008 		break;
1009 	}
1010 
1011 	if (!guest_fac_enabled) {
1012 		/* Facility not enabled by the guest */
1013 		kvmppc_trigger_fac_interrupt(vcpu, fac);
1014 		return RESUME_GUEST;
1015 	}
1016 
1017 	switch (fac) {
1018 	case FSCR_TAR_LG:
1019 		/* TAR switching isn't lazy in Linux yet */
1020 		current->thread.tar = mfspr(SPRN_TAR);
1021 		mtspr(SPRN_TAR, vcpu->arch.tar);
1022 		vcpu->arch.shadow_fscr |= FSCR_TAR;
1023 		break;
1024 	default:
1025 		kvmppc_emulate_fac(vcpu, fac);
1026 		break;
1027 	}
1028 
1029 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1030 	/* Since we disabled MSR_TM at privilege state, the mfspr instruction
1031 	 * for TM spr can trigger TM fac unavailable. In this case, the
1032 	 * emulation is handled by kvmppc_emulate_fac(), which invokes
1033 	 * kvmppc_emulate_mfspr() finally. But note the mfspr can include
1034 	 * RT for NV registers. So it need to restore those NV reg to reflect
1035 	 * the update.
1036 	 */
1037 	if ((fac == FSCR_TM_LG) && !(kvmppc_get_msr(vcpu) & MSR_PR))
1038 		return RESUME_GUEST_NV;
1039 #endif
1040 
1041 	return RESUME_GUEST;
1042 }
1043 
1044 void kvmppc_set_fscr(struct kvm_vcpu *vcpu, u64 fscr)
1045 {
1046 	if (fscr & FSCR_SCV)
1047 		fscr &= ~FSCR_SCV; /* SCV must not be enabled */
1048 	if ((vcpu->arch.fscr & FSCR_TAR) && !(fscr & FSCR_TAR)) {
1049 		/* TAR got dropped, drop it in shadow too */
1050 		kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
1051 	} else if (!(vcpu->arch.fscr & FSCR_TAR) && (fscr & FSCR_TAR)) {
1052 		vcpu->arch.fscr = fscr;
1053 		kvmppc_handle_fac(vcpu, FSCR_TAR_LG);
1054 		return;
1055 	}
1056 
1057 	vcpu->arch.fscr = fscr;
1058 }
1059 #endif
1060 
1061 static void kvmppc_setup_debug(struct kvm_vcpu *vcpu)
1062 {
1063 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
1064 		u64 msr = kvmppc_get_msr(vcpu);
1065 
1066 		kvmppc_set_msr(vcpu, msr | MSR_SE);
1067 	}
1068 }
1069 
1070 static void kvmppc_clear_debug(struct kvm_vcpu *vcpu)
1071 {
1072 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
1073 		u64 msr = kvmppc_get_msr(vcpu);
1074 
1075 		kvmppc_set_msr(vcpu, msr & ~MSR_SE);
1076 	}
1077 }
1078 
1079 static int kvmppc_exit_pr_progint(struct kvm_vcpu *vcpu, unsigned int exit_nr)
1080 {
1081 	enum emulation_result er;
1082 	ulong flags;
1083 	u32 last_inst;
1084 	int emul, r;
1085 
1086 	/*
1087 	 * shadow_srr1 only contains valid flags if we came here via a program
1088 	 * exception. The other exceptions (emulation assist, FP unavailable,
1089 	 * etc.) do not provide flags in SRR1, so use an illegal-instruction
1090 	 * exception when injecting a program interrupt into the guest.
1091 	 */
1092 	if (exit_nr == BOOK3S_INTERRUPT_PROGRAM)
1093 		flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
1094 	else
1095 		flags = SRR1_PROGILL;
1096 
1097 	emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1098 	if (emul != EMULATE_DONE)
1099 		return RESUME_GUEST;
1100 
1101 	if (kvmppc_get_msr(vcpu) & MSR_PR) {
1102 #ifdef EXIT_DEBUG
1103 		pr_info("Userspace triggered 0x700 exception at\n 0x%lx (0x%x)\n",
1104 			kvmppc_get_pc(vcpu), last_inst);
1105 #endif
1106 		if ((last_inst & 0xff0007ff) != (INS_DCBZ & 0xfffffff7)) {
1107 			kvmppc_core_queue_program(vcpu, flags);
1108 			return RESUME_GUEST;
1109 		}
1110 	}
1111 
1112 	vcpu->stat.emulated_inst_exits++;
1113 	er = kvmppc_emulate_instruction(vcpu);
1114 	switch (er) {
1115 	case EMULATE_DONE:
1116 		r = RESUME_GUEST_NV;
1117 		break;
1118 	case EMULATE_AGAIN:
1119 		r = RESUME_GUEST;
1120 		break;
1121 	case EMULATE_FAIL:
1122 		pr_crit("%s: emulation at %lx failed (%08x)\n",
1123 			__func__, kvmppc_get_pc(vcpu), last_inst);
1124 		kvmppc_core_queue_program(vcpu, flags);
1125 		r = RESUME_GUEST;
1126 		break;
1127 	case EMULATE_DO_MMIO:
1128 		vcpu->run->exit_reason = KVM_EXIT_MMIO;
1129 		r = RESUME_HOST_NV;
1130 		break;
1131 	case EMULATE_EXIT_USER:
1132 		r = RESUME_HOST_NV;
1133 		break;
1134 	default:
1135 		BUG();
1136 	}
1137 
1138 	return r;
1139 }
1140 
1141 int kvmppc_handle_exit_pr(struct kvm_vcpu *vcpu, unsigned int exit_nr)
1142 {
1143 	struct kvm_run *run = vcpu->run;
1144 	int r = RESUME_HOST;
1145 	int s;
1146 
1147 	vcpu->stat.sum_exits++;
1148 
1149 	run->exit_reason = KVM_EXIT_UNKNOWN;
1150 	run->ready_for_interrupt_injection = 1;
1151 
1152 	/* We get here with MSR.EE=1 */
1153 
1154 	trace_kvm_exit(exit_nr, vcpu);
1155 	guest_exit();
1156 
1157 	switch (exit_nr) {
1158 	case BOOK3S_INTERRUPT_INST_STORAGE:
1159 	{
1160 		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1161 		vcpu->stat.pf_instruc++;
1162 
1163 		if (kvmppc_is_split_real(vcpu))
1164 			kvmppc_fixup_split_real(vcpu);
1165 
1166 #ifdef CONFIG_PPC_BOOK3S_32
1167 		/* We set segments as unused segments when invalidating them. So
1168 		 * treat the respective fault as segment fault. */
1169 		{
1170 			struct kvmppc_book3s_shadow_vcpu *svcpu;
1171 			u32 sr;
1172 
1173 			svcpu = svcpu_get(vcpu);
1174 			sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
1175 			svcpu_put(svcpu);
1176 			if (sr == SR_INVALID) {
1177 				kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
1178 				r = RESUME_GUEST;
1179 				break;
1180 			}
1181 		}
1182 #endif
1183 
1184 		/* only care about PTEG not found errors, but leave NX alone */
1185 		if (shadow_srr1 & 0x40000000) {
1186 			int idx = srcu_read_lock(&vcpu->kvm->srcu);
1187 			r = kvmppc_handle_pagefault(vcpu, kvmppc_get_pc(vcpu), exit_nr);
1188 			srcu_read_unlock(&vcpu->kvm->srcu, idx);
1189 			vcpu->stat.sp_instruc++;
1190 		} else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
1191 			  (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
1192 			/*
1193 			 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
1194 			 *     so we can't use the NX bit inside the guest. Let's cross our fingers,
1195 			 *     that no guest that needs the dcbz hack does NX.
1196 			 */
1197 			kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
1198 			r = RESUME_GUEST;
1199 		} else {
1200 			kvmppc_core_queue_inst_storage(vcpu,
1201 						shadow_srr1 & 0x58000000);
1202 			r = RESUME_GUEST;
1203 		}
1204 		break;
1205 	}
1206 	case BOOK3S_INTERRUPT_DATA_STORAGE:
1207 	{
1208 		ulong dar = kvmppc_get_fault_dar(vcpu);
1209 		u32 fault_dsisr = vcpu->arch.fault_dsisr;
1210 		vcpu->stat.pf_storage++;
1211 
1212 #ifdef CONFIG_PPC_BOOK3S_32
1213 		/* We set segments as unused segments when invalidating them. So
1214 		 * treat the respective fault as segment fault. */
1215 		{
1216 			struct kvmppc_book3s_shadow_vcpu *svcpu;
1217 			u32 sr;
1218 
1219 			svcpu = svcpu_get(vcpu);
1220 			sr = svcpu->sr[dar >> SID_SHIFT];
1221 			svcpu_put(svcpu);
1222 			if (sr == SR_INVALID) {
1223 				kvmppc_mmu_map_segment(vcpu, dar);
1224 				r = RESUME_GUEST;
1225 				break;
1226 			}
1227 		}
1228 #endif
1229 
1230 		/*
1231 		 * We need to handle missing shadow PTEs, and
1232 		 * protection faults due to us mapping a page read-only
1233 		 * when the guest thinks it is writable.
1234 		 */
1235 		if (fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT)) {
1236 			int idx = srcu_read_lock(&vcpu->kvm->srcu);
1237 			r = kvmppc_handle_pagefault(vcpu, dar, exit_nr);
1238 			srcu_read_unlock(&vcpu->kvm->srcu, idx);
1239 		} else {
1240 			kvmppc_core_queue_data_storage(vcpu, dar, fault_dsisr);
1241 			r = RESUME_GUEST;
1242 		}
1243 		break;
1244 	}
1245 	case BOOK3S_INTERRUPT_DATA_SEGMENT:
1246 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
1247 			kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
1248 			kvmppc_book3s_queue_irqprio(vcpu,
1249 				BOOK3S_INTERRUPT_DATA_SEGMENT);
1250 		}
1251 		r = RESUME_GUEST;
1252 		break;
1253 	case BOOK3S_INTERRUPT_INST_SEGMENT:
1254 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
1255 			kvmppc_book3s_queue_irqprio(vcpu,
1256 				BOOK3S_INTERRUPT_INST_SEGMENT);
1257 		}
1258 		r = RESUME_GUEST;
1259 		break;
1260 	/* We're good on these - the host merely wanted to get our attention */
1261 	case BOOK3S_INTERRUPT_DECREMENTER:
1262 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1263 	case BOOK3S_INTERRUPT_DOORBELL:
1264 	case BOOK3S_INTERRUPT_H_DOORBELL:
1265 		vcpu->stat.dec_exits++;
1266 		r = RESUME_GUEST;
1267 		break;
1268 	case BOOK3S_INTERRUPT_EXTERNAL:
1269 	case BOOK3S_INTERRUPT_EXTERNAL_HV:
1270 	case BOOK3S_INTERRUPT_H_VIRT:
1271 		vcpu->stat.ext_intr_exits++;
1272 		r = RESUME_GUEST;
1273 		break;
1274 	case BOOK3S_INTERRUPT_HMI:
1275 	case BOOK3S_INTERRUPT_PERFMON:
1276 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1277 		r = RESUME_GUEST;
1278 		break;
1279 	case BOOK3S_INTERRUPT_PROGRAM:
1280 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1281 		r = kvmppc_exit_pr_progint(vcpu, exit_nr);
1282 		break;
1283 	case BOOK3S_INTERRUPT_SYSCALL:
1284 	{
1285 		u32 last_sc;
1286 		int emul;
1287 
1288 		/* Get last sc for papr */
1289 		if (vcpu->arch.papr_enabled) {
1290 			/* The sc instruction points SRR0 to the next inst */
1291 			emul = kvmppc_get_last_inst(vcpu, INST_SC, &last_sc);
1292 			if (emul != EMULATE_DONE) {
1293 				kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) - 4);
1294 				r = RESUME_GUEST;
1295 				break;
1296 			}
1297 		}
1298 
1299 		if (vcpu->arch.papr_enabled &&
1300 		    (last_sc == 0x44000022) &&
1301 		    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
1302 			/* SC 1 papr hypercalls */
1303 			ulong cmd = kvmppc_get_gpr(vcpu, 3);
1304 			int i;
1305 
1306 #ifdef CONFIG_PPC_BOOK3S_64
1307 			if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
1308 				r = RESUME_GUEST;
1309 				break;
1310 			}
1311 #endif
1312 
1313 			run->papr_hcall.nr = cmd;
1314 			for (i = 0; i < 9; ++i) {
1315 				ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
1316 				run->papr_hcall.args[i] = gpr;
1317 			}
1318 			run->exit_reason = KVM_EXIT_PAPR_HCALL;
1319 			vcpu->arch.hcall_needed = 1;
1320 			r = RESUME_HOST;
1321 		} else if (vcpu->arch.osi_enabled &&
1322 		    (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
1323 		    (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
1324 			/* MOL hypercalls */
1325 			u64 *gprs = run->osi.gprs;
1326 			int i;
1327 
1328 			run->exit_reason = KVM_EXIT_OSI;
1329 			for (i = 0; i < 32; i++)
1330 				gprs[i] = kvmppc_get_gpr(vcpu, i);
1331 			vcpu->arch.osi_needed = 1;
1332 			r = RESUME_HOST_NV;
1333 		} else if (!(kvmppc_get_msr(vcpu) & MSR_PR) &&
1334 		    (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
1335 			/* KVM PV hypercalls */
1336 			kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
1337 			r = RESUME_GUEST;
1338 		} else {
1339 			/* Guest syscalls */
1340 			vcpu->stat.syscall_exits++;
1341 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1342 			r = RESUME_GUEST;
1343 		}
1344 		break;
1345 	}
1346 	case BOOK3S_INTERRUPT_FP_UNAVAIL:
1347 	case BOOK3S_INTERRUPT_ALTIVEC:
1348 	case BOOK3S_INTERRUPT_VSX:
1349 	{
1350 		int ext_msr = 0;
1351 		int emul;
1352 		u32 last_inst;
1353 
1354 		if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE) {
1355 			/* Do paired single instruction emulation */
1356 			emul = kvmppc_get_last_inst(vcpu, INST_GENERIC,
1357 						    &last_inst);
1358 			if (emul == EMULATE_DONE)
1359 				r = kvmppc_exit_pr_progint(vcpu, exit_nr);
1360 			else
1361 				r = RESUME_GUEST;
1362 
1363 			break;
1364 		}
1365 
1366 		/* Enable external provider */
1367 		switch (exit_nr) {
1368 		case BOOK3S_INTERRUPT_FP_UNAVAIL:
1369 			ext_msr = MSR_FP;
1370 			break;
1371 
1372 		case BOOK3S_INTERRUPT_ALTIVEC:
1373 			ext_msr = MSR_VEC;
1374 			break;
1375 
1376 		case BOOK3S_INTERRUPT_VSX:
1377 			ext_msr = MSR_VSX;
1378 			break;
1379 		}
1380 
1381 		r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
1382 		break;
1383 	}
1384 	case BOOK3S_INTERRUPT_ALIGNMENT:
1385 	{
1386 		u32 last_inst;
1387 		int emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1388 
1389 		if (emul == EMULATE_DONE) {
1390 			u32 dsisr;
1391 			u64 dar;
1392 
1393 			dsisr = kvmppc_alignment_dsisr(vcpu, last_inst);
1394 			dar = kvmppc_alignment_dar(vcpu, last_inst);
1395 
1396 			kvmppc_set_dsisr(vcpu, dsisr);
1397 			kvmppc_set_dar(vcpu, dar);
1398 
1399 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1400 		}
1401 		r = RESUME_GUEST;
1402 		break;
1403 	}
1404 #ifdef CONFIG_PPC_BOOK3S_64
1405 	case BOOK3S_INTERRUPT_FAC_UNAVAIL:
1406 		r = kvmppc_handle_fac(vcpu, vcpu->arch.shadow_fscr >> 56);
1407 		break;
1408 #endif
1409 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1410 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1411 		r = RESUME_GUEST;
1412 		break;
1413 	case BOOK3S_INTERRUPT_TRACE:
1414 		if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
1415 			run->exit_reason = KVM_EXIT_DEBUG;
1416 			r = RESUME_HOST;
1417 		} else {
1418 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1419 			r = RESUME_GUEST;
1420 		}
1421 		break;
1422 	default:
1423 	{
1424 		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1425 		/* Ugh - bork here! What did we get? */
1426 		printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1427 			exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
1428 		r = RESUME_HOST;
1429 		BUG();
1430 		break;
1431 	}
1432 	}
1433 
1434 	if (!(r & RESUME_HOST)) {
1435 		/* To avoid clobbering exit_reason, only check for signals if
1436 		 * we aren't already exiting to userspace for some other
1437 		 * reason. */
1438 
1439 		/*
1440 		 * Interrupts could be timers for the guest which we have to
1441 		 * inject again, so let's postpone them until we're in the guest
1442 		 * and if we really did time things so badly, then we just exit
1443 		 * again due to a host external interrupt.
1444 		 */
1445 		s = kvmppc_prepare_to_enter(vcpu);
1446 		if (s <= 0)
1447 			r = s;
1448 		else {
1449 			/* interrupts now hard-disabled */
1450 			kvmppc_fix_ee_before_entry();
1451 		}
1452 
1453 		kvmppc_handle_lost_ext(vcpu);
1454 	}
1455 
1456 	trace_kvm_book3s_reenter(r, vcpu);
1457 
1458 	return r;
1459 }
1460 
1461 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu *vcpu,
1462 					    struct kvm_sregs *sregs)
1463 {
1464 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1465 	int i;
1466 
1467 	sregs->pvr = vcpu->arch.pvr;
1468 
1469 	sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1470 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1471 		for (i = 0; i < 64; i++) {
1472 			sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
1473 			sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1474 		}
1475 	} else {
1476 		for (i = 0; i < 16; i++)
1477 			sregs->u.s.ppc32.sr[i] = kvmppc_get_sr(vcpu, i);
1478 
1479 		for (i = 0; i < 8; i++) {
1480 			sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1481 			sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1482 		}
1483 	}
1484 
1485 	return 0;
1486 }
1487 
1488 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu *vcpu,
1489 					    struct kvm_sregs *sregs)
1490 {
1491 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1492 	int i;
1493 
1494 	kvmppc_set_pvr_pr(vcpu, sregs->pvr);
1495 
1496 	vcpu3s->sdr1 = sregs->u.s.sdr1;
1497 #ifdef CONFIG_PPC_BOOK3S_64
1498 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1499 		/* Flush all SLB entries */
1500 		vcpu->arch.mmu.slbmte(vcpu, 0, 0);
1501 		vcpu->arch.mmu.slbia(vcpu);
1502 
1503 		for (i = 0; i < 64; i++) {
1504 			u64 rb = sregs->u.s.ppc64.slb[i].slbe;
1505 			u64 rs = sregs->u.s.ppc64.slb[i].slbv;
1506 
1507 			if (rb & SLB_ESID_V)
1508 				vcpu->arch.mmu.slbmte(vcpu, rs, rb);
1509 		}
1510 	} else
1511 #endif
1512 	{
1513 		for (i = 0; i < 16; i++) {
1514 			vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1515 		}
1516 		for (i = 0; i < 8; i++) {
1517 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1518 				       (u32)sregs->u.s.ppc32.ibat[i]);
1519 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1520 				       (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1521 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1522 				       (u32)sregs->u.s.ppc32.dbat[i]);
1523 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1524 				       (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1525 		}
1526 	}
1527 
1528 	/* Flush the MMU after messing with the segments */
1529 	kvmppc_mmu_pte_flush(vcpu, 0, 0);
1530 
1531 	return 0;
1532 }
1533 
1534 static int kvmppc_get_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1535 				 union kvmppc_one_reg *val)
1536 {
1537 	int r = 0;
1538 
1539 	switch (id) {
1540 	case KVM_REG_PPC_DEBUG_INST:
1541 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1542 		break;
1543 	case KVM_REG_PPC_HIOR:
1544 		*val = get_reg_val(id, to_book3s(vcpu)->hior);
1545 		break;
1546 	case KVM_REG_PPC_VTB:
1547 		*val = get_reg_val(id, to_book3s(vcpu)->vtb);
1548 		break;
1549 	case KVM_REG_PPC_LPCR:
1550 	case KVM_REG_PPC_LPCR_64:
1551 		/*
1552 		 * We are only interested in the LPCR_ILE bit
1553 		 */
1554 		if (vcpu->arch.intr_msr & MSR_LE)
1555 			*val = get_reg_val(id, LPCR_ILE);
1556 		else
1557 			*val = get_reg_val(id, 0);
1558 		break;
1559 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1560 	case KVM_REG_PPC_TFHAR:
1561 		*val = get_reg_val(id, vcpu->arch.tfhar);
1562 		break;
1563 	case KVM_REG_PPC_TFIAR:
1564 		*val = get_reg_val(id, vcpu->arch.tfiar);
1565 		break;
1566 	case KVM_REG_PPC_TEXASR:
1567 		*val = get_reg_val(id, vcpu->arch.texasr);
1568 		break;
1569 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1570 		*val = get_reg_val(id,
1571 				vcpu->arch.gpr_tm[id-KVM_REG_PPC_TM_GPR0]);
1572 		break;
1573 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1574 	{
1575 		int i, j;
1576 
1577 		i = id - KVM_REG_PPC_TM_VSR0;
1578 		if (i < 32)
1579 			for (j = 0; j < TS_FPRWIDTH; j++)
1580 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1581 		else {
1582 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1583 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1584 			else
1585 				r = -ENXIO;
1586 		}
1587 		break;
1588 	}
1589 	case KVM_REG_PPC_TM_CR:
1590 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1591 		break;
1592 	case KVM_REG_PPC_TM_XER:
1593 		*val = get_reg_val(id, vcpu->arch.xer_tm);
1594 		break;
1595 	case KVM_REG_PPC_TM_LR:
1596 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1597 		break;
1598 	case KVM_REG_PPC_TM_CTR:
1599 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1600 		break;
1601 	case KVM_REG_PPC_TM_FPSCR:
1602 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1603 		break;
1604 	case KVM_REG_PPC_TM_AMR:
1605 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1606 		break;
1607 	case KVM_REG_PPC_TM_PPR:
1608 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1609 		break;
1610 	case KVM_REG_PPC_TM_VRSAVE:
1611 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1612 		break;
1613 	case KVM_REG_PPC_TM_VSCR:
1614 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1615 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1616 		else
1617 			r = -ENXIO;
1618 		break;
1619 	case KVM_REG_PPC_TM_DSCR:
1620 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1621 		break;
1622 	case KVM_REG_PPC_TM_TAR:
1623 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1624 		break;
1625 #endif
1626 	default:
1627 		r = -EINVAL;
1628 		break;
1629 	}
1630 
1631 	return r;
1632 }
1633 
1634 static void kvmppc_set_lpcr_pr(struct kvm_vcpu *vcpu, u64 new_lpcr)
1635 {
1636 	if (new_lpcr & LPCR_ILE)
1637 		vcpu->arch.intr_msr |= MSR_LE;
1638 	else
1639 		vcpu->arch.intr_msr &= ~MSR_LE;
1640 }
1641 
1642 static int kvmppc_set_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1643 				 union kvmppc_one_reg *val)
1644 {
1645 	int r = 0;
1646 
1647 	switch (id) {
1648 	case KVM_REG_PPC_HIOR:
1649 		to_book3s(vcpu)->hior = set_reg_val(id, *val);
1650 		to_book3s(vcpu)->hior_explicit = true;
1651 		break;
1652 	case KVM_REG_PPC_VTB:
1653 		to_book3s(vcpu)->vtb = set_reg_val(id, *val);
1654 		break;
1655 	case KVM_REG_PPC_LPCR:
1656 	case KVM_REG_PPC_LPCR_64:
1657 		kvmppc_set_lpcr_pr(vcpu, set_reg_val(id, *val));
1658 		break;
1659 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1660 	case KVM_REG_PPC_TFHAR:
1661 		vcpu->arch.tfhar = set_reg_val(id, *val);
1662 		break;
1663 	case KVM_REG_PPC_TFIAR:
1664 		vcpu->arch.tfiar = set_reg_val(id, *val);
1665 		break;
1666 	case KVM_REG_PPC_TEXASR:
1667 		vcpu->arch.texasr = set_reg_val(id, *val);
1668 		break;
1669 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1670 		vcpu->arch.gpr_tm[id - KVM_REG_PPC_TM_GPR0] =
1671 			set_reg_val(id, *val);
1672 		break;
1673 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1674 	{
1675 		int i, j;
1676 
1677 		i = id - KVM_REG_PPC_TM_VSR0;
1678 		if (i < 32)
1679 			for (j = 0; j < TS_FPRWIDTH; j++)
1680 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1681 		else
1682 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1683 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1684 			else
1685 				r = -ENXIO;
1686 		break;
1687 	}
1688 	case KVM_REG_PPC_TM_CR:
1689 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1690 		break;
1691 	case KVM_REG_PPC_TM_XER:
1692 		vcpu->arch.xer_tm = set_reg_val(id, *val);
1693 		break;
1694 	case KVM_REG_PPC_TM_LR:
1695 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1696 		break;
1697 	case KVM_REG_PPC_TM_CTR:
1698 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1699 		break;
1700 	case KVM_REG_PPC_TM_FPSCR:
1701 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1702 		break;
1703 	case KVM_REG_PPC_TM_AMR:
1704 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1705 		break;
1706 	case KVM_REG_PPC_TM_PPR:
1707 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1708 		break;
1709 	case KVM_REG_PPC_TM_VRSAVE:
1710 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1711 		break;
1712 	case KVM_REG_PPC_TM_VSCR:
1713 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1714 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1715 		else
1716 			r = -ENXIO;
1717 		break;
1718 	case KVM_REG_PPC_TM_DSCR:
1719 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1720 		break;
1721 	case KVM_REG_PPC_TM_TAR:
1722 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1723 		break;
1724 #endif
1725 	default:
1726 		r = -EINVAL;
1727 		break;
1728 	}
1729 
1730 	return r;
1731 }
1732 
1733 static int kvmppc_core_vcpu_create_pr(struct kvm_vcpu *vcpu)
1734 {
1735 	struct kvmppc_vcpu_book3s *vcpu_book3s;
1736 	unsigned long p;
1737 	int err;
1738 
1739 	err = -ENOMEM;
1740 
1741 	vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1742 	if (!vcpu_book3s)
1743 		goto out;
1744 	vcpu->arch.book3s = vcpu_book3s;
1745 
1746 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1747 	vcpu->arch.shadow_vcpu =
1748 		kzalloc(sizeof(*vcpu->arch.shadow_vcpu), GFP_KERNEL);
1749 	if (!vcpu->arch.shadow_vcpu)
1750 		goto free_vcpu3s;
1751 #endif
1752 
1753 	p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1754 	if (!p)
1755 		goto free_shadow_vcpu;
1756 	vcpu->arch.shared = (void *)p;
1757 #ifdef CONFIG_PPC_BOOK3S_64
1758 	/* Always start the shared struct in native endian mode */
1759 #ifdef __BIG_ENDIAN__
1760         vcpu->arch.shared_big_endian = true;
1761 #else
1762         vcpu->arch.shared_big_endian = false;
1763 #endif
1764 
1765 	/*
1766 	 * Default to the same as the host if we're on sufficiently
1767 	 * recent machine that we have 1TB segments;
1768 	 * otherwise default to PPC970FX.
1769 	 */
1770 	vcpu->arch.pvr = 0x3C0301;
1771 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1772 		vcpu->arch.pvr = mfspr(SPRN_PVR);
1773 	vcpu->arch.intr_msr = MSR_SF;
1774 #else
1775 	/* default to book3s_32 (750) */
1776 	vcpu->arch.pvr = 0x84202;
1777 	vcpu->arch.intr_msr = 0;
1778 #endif
1779 	kvmppc_set_pvr_pr(vcpu, vcpu->arch.pvr);
1780 	vcpu->arch.slb_nr = 64;
1781 
1782 	vcpu->arch.shadow_msr = MSR_USER64 & ~MSR_LE;
1783 
1784 	err = kvmppc_mmu_init_pr(vcpu);
1785 	if (err < 0)
1786 		goto free_shared_page;
1787 
1788 	return 0;
1789 
1790 free_shared_page:
1791 	free_page((unsigned long)vcpu->arch.shared);
1792 free_shadow_vcpu:
1793 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1794 	kfree(vcpu->arch.shadow_vcpu);
1795 free_vcpu3s:
1796 #endif
1797 	vfree(vcpu_book3s);
1798 out:
1799 	return err;
1800 }
1801 
1802 static void kvmppc_core_vcpu_free_pr(struct kvm_vcpu *vcpu)
1803 {
1804 	struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1805 
1806 	kvmppc_mmu_destroy_pr(vcpu);
1807 	free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1808 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1809 	kfree(vcpu->arch.shadow_vcpu);
1810 #endif
1811 	vfree(vcpu_book3s);
1812 }
1813 
1814 static int kvmppc_vcpu_run_pr(struct kvm_vcpu *vcpu)
1815 {
1816 	int ret;
1817 
1818 	/* Check if we can run the vcpu at all */
1819 	if (!vcpu->arch.sane) {
1820 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1821 		ret = -EINVAL;
1822 		goto out;
1823 	}
1824 
1825 	kvmppc_setup_debug(vcpu);
1826 
1827 	/*
1828 	 * Interrupts could be timers for the guest which we have to inject
1829 	 * again, so let's postpone them until we're in the guest and if we
1830 	 * really did time things so badly, then we just exit again due to
1831 	 * a host external interrupt.
1832 	 */
1833 	ret = kvmppc_prepare_to_enter(vcpu);
1834 	if (ret <= 0)
1835 		goto out;
1836 	/* interrupts now hard-disabled */
1837 
1838 	/* Save FPU, Altivec and VSX state */
1839 	giveup_all(current);
1840 
1841 	/* Preload FPU if it's enabled */
1842 	if (kvmppc_get_msr(vcpu) & MSR_FP)
1843 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1844 
1845 	kvmppc_fix_ee_before_entry();
1846 
1847 	ret = __kvmppc_vcpu_run(vcpu);
1848 
1849 	kvmppc_clear_debug(vcpu);
1850 
1851 	/* No need for guest_exit. It's done in handle_exit.
1852 	   We also get here with interrupts enabled. */
1853 
1854 	/* Make sure we save the guest FPU/Altivec/VSX state */
1855 	kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1856 
1857 	/* Make sure we save the guest TAR/EBB/DSCR state */
1858 	kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
1859 
1860 	srr_regs_clobbered();
1861 out:
1862 	vcpu->mode = OUTSIDE_GUEST_MODE;
1863 	return ret;
1864 }
1865 
1866 /*
1867  * Get (and clear) the dirty memory log for a memory slot.
1868  */
1869 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm *kvm,
1870 					 struct kvm_dirty_log *log)
1871 {
1872 	struct kvm_memory_slot *memslot;
1873 	struct kvm_vcpu *vcpu;
1874 	ulong ga, ga_end;
1875 	int is_dirty = 0;
1876 	int r;
1877 	unsigned long n;
1878 
1879 	mutex_lock(&kvm->slots_lock);
1880 
1881 	r = kvm_get_dirty_log(kvm, log, &is_dirty, &memslot);
1882 	if (r)
1883 		goto out;
1884 
1885 	/* If nothing is dirty, don't bother messing with page tables. */
1886 	if (is_dirty) {
1887 		ga = memslot->base_gfn << PAGE_SHIFT;
1888 		ga_end = ga + (memslot->npages << PAGE_SHIFT);
1889 
1890 		kvm_for_each_vcpu(n, vcpu, kvm)
1891 			kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1892 
1893 		n = kvm_dirty_bitmap_bytes(memslot);
1894 		memset(memslot->dirty_bitmap, 0, n);
1895 	}
1896 
1897 	r = 0;
1898 out:
1899 	mutex_unlock(&kvm->slots_lock);
1900 	return r;
1901 }
1902 
1903 static void kvmppc_core_flush_memslot_pr(struct kvm *kvm,
1904 					 struct kvm_memory_slot *memslot)
1905 {
1906 	return;
1907 }
1908 
1909 static int kvmppc_core_prepare_memory_region_pr(struct kvm *kvm,
1910 				const struct kvm_memory_slot *old,
1911 				struct kvm_memory_slot *new,
1912 				enum kvm_mr_change change)
1913 {
1914 	return 0;
1915 }
1916 
1917 static void kvmppc_core_commit_memory_region_pr(struct kvm *kvm,
1918 				struct kvm_memory_slot *old,
1919 				const struct kvm_memory_slot *new,
1920 				enum kvm_mr_change change)
1921 {
1922 	return;
1923 }
1924 
1925 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot *slot)
1926 {
1927 	return;
1928 }
1929 
1930 #ifdef CONFIG_PPC64
1931 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1932 					 struct kvm_ppc_smmu_info *info)
1933 {
1934 	long int i;
1935 	struct kvm_vcpu *vcpu;
1936 
1937 	info->flags = 0;
1938 
1939 	/* SLB is always 64 entries */
1940 	info->slb_size = 64;
1941 
1942 	/* Standard 4k base page size segment */
1943 	info->sps[0].page_shift = 12;
1944 	info->sps[0].slb_enc = 0;
1945 	info->sps[0].enc[0].page_shift = 12;
1946 	info->sps[0].enc[0].pte_enc = 0;
1947 
1948 	/*
1949 	 * 64k large page size.
1950 	 * We only want to put this in if the CPUs we're emulating
1951 	 * support it, but unfortunately we don't have a vcpu easily
1952 	 * to hand here to test.  Just pick the first vcpu, and if
1953 	 * that doesn't exist yet, report the minimum capability,
1954 	 * i.e., no 64k pages.
1955 	 * 1T segment support goes along with 64k pages.
1956 	 */
1957 	i = 1;
1958 	vcpu = kvm_get_vcpu(kvm, 0);
1959 	if (vcpu && (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
1960 		info->flags = KVM_PPC_1T_SEGMENTS;
1961 		info->sps[i].page_shift = 16;
1962 		info->sps[i].slb_enc = SLB_VSID_L | SLB_VSID_LP_01;
1963 		info->sps[i].enc[0].page_shift = 16;
1964 		info->sps[i].enc[0].pte_enc = 1;
1965 		++i;
1966 	}
1967 
1968 	/* Standard 16M large page size segment */
1969 	info->sps[i].page_shift = 24;
1970 	info->sps[i].slb_enc = SLB_VSID_L;
1971 	info->sps[i].enc[0].page_shift = 24;
1972 	info->sps[i].enc[0].pte_enc = 0;
1973 
1974 	return 0;
1975 }
1976 
1977 static int kvm_configure_mmu_pr(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
1978 {
1979 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
1980 		return -ENODEV;
1981 	/* Require flags and process table base and size to all be zero. */
1982 	if (cfg->flags || cfg->process_table)
1983 		return -EINVAL;
1984 	return 0;
1985 }
1986 
1987 #else
1988 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1989 					 struct kvm_ppc_smmu_info *info)
1990 {
1991 	/* We should not get called */
1992 	BUG();
1993 	return 0;
1994 }
1995 #endif /* CONFIG_PPC64 */
1996 
1997 static unsigned int kvm_global_user_count = 0;
1998 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1999 
2000 static int kvmppc_core_init_vm_pr(struct kvm *kvm)
2001 {
2002 	mutex_init(&kvm->arch.hpt_mutex);
2003 
2004 #ifdef CONFIG_PPC_BOOK3S_64
2005 	/* Start out with the default set of hcalls enabled */
2006 	kvmppc_pr_init_default_hcalls(kvm);
2007 #endif
2008 
2009 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
2010 		spin_lock(&kvm_global_user_count_lock);
2011 		if (++kvm_global_user_count == 1)
2012 			pseries_disable_reloc_on_exc();
2013 		spin_unlock(&kvm_global_user_count_lock);
2014 	}
2015 	return 0;
2016 }
2017 
2018 static void kvmppc_core_destroy_vm_pr(struct kvm *kvm)
2019 {
2020 #ifdef CONFIG_PPC64
2021 	WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
2022 #endif
2023 
2024 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
2025 		spin_lock(&kvm_global_user_count_lock);
2026 		BUG_ON(kvm_global_user_count == 0);
2027 		if (--kvm_global_user_count == 0)
2028 			pseries_enable_reloc_on_exc();
2029 		spin_unlock(&kvm_global_user_count_lock);
2030 	}
2031 }
2032 
2033 static int kvmppc_core_check_processor_compat_pr(void)
2034 {
2035 	/*
2036 	 * PR KVM can work on POWER9 inside a guest partition
2037 	 * running in HPT mode.  It can't work if we are using
2038 	 * radix translation (because radix provides no way for
2039 	 * a process to have unique translations in quadrant 3).
2040 	 */
2041 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
2042 		return -EIO;
2043 	return 0;
2044 }
2045 
2046 static long kvm_arch_vm_ioctl_pr(struct file *filp,
2047 				 unsigned int ioctl, unsigned long arg)
2048 {
2049 	return -ENOTTY;
2050 }
2051 
2052 static struct kvmppc_ops kvm_ops_pr = {
2053 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_pr,
2054 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_pr,
2055 	.get_one_reg = kvmppc_get_one_reg_pr,
2056 	.set_one_reg = kvmppc_set_one_reg_pr,
2057 	.vcpu_load   = kvmppc_core_vcpu_load_pr,
2058 	.vcpu_put    = kvmppc_core_vcpu_put_pr,
2059 	.inject_interrupt = kvmppc_inject_interrupt_pr,
2060 	.set_msr     = kvmppc_set_msr_pr,
2061 	.vcpu_run    = kvmppc_vcpu_run_pr,
2062 	.vcpu_create = kvmppc_core_vcpu_create_pr,
2063 	.vcpu_free   = kvmppc_core_vcpu_free_pr,
2064 	.check_requests = kvmppc_core_check_requests_pr,
2065 	.get_dirty_log = kvm_vm_ioctl_get_dirty_log_pr,
2066 	.flush_memslot = kvmppc_core_flush_memslot_pr,
2067 	.prepare_memory_region = kvmppc_core_prepare_memory_region_pr,
2068 	.commit_memory_region = kvmppc_core_commit_memory_region_pr,
2069 	.unmap_gfn_range = kvm_unmap_gfn_range_pr,
2070 	.age_gfn  = kvm_age_gfn_pr,
2071 	.test_age_gfn = kvm_test_age_gfn_pr,
2072 	.set_spte_gfn = kvm_set_spte_gfn_pr,
2073 	.free_memslot = kvmppc_core_free_memslot_pr,
2074 	.init_vm = kvmppc_core_init_vm_pr,
2075 	.destroy_vm = kvmppc_core_destroy_vm_pr,
2076 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_pr,
2077 	.emulate_op = kvmppc_core_emulate_op_pr,
2078 	.emulate_mtspr = kvmppc_core_emulate_mtspr_pr,
2079 	.emulate_mfspr = kvmppc_core_emulate_mfspr_pr,
2080 	.fast_vcpu_kick = kvm_vcpu_kick,
2081 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_pr,
2082 #ifdef CONFIG_PPC_BOOK3S_64
2083 	.hcall_implemented = kvmppc_hcall_impl_pr,
2084 	.configure_mmu = kvm_configure_mmu_pr,
2085 #endif
2086 	.giveup_ext = kvmppc_giveup_ext,
2087 };
2088 
2089 
2090 int kvmppc_book3s_init_pr(void)
2091 {
2092 	int r;
2093 
2094 	r = kvmppc_core_check_processor_compat_pr();
2095 	if (r < 0)
2096 		return r;
2097 
2098 	kvm_ops_pr.owner = THIS_MODULE;
2099 	kvmppc_pr_ops = &kvm_ops_pr;
2100 
2101 	r = kvmppc_mmu_hpte_sysinit();
2102 	return r;
2103 }
2104 
2105 void kvmppc_book3s_exit_pr(void)
2106 {
2107 	kvmppc_pr_ops = NULL;
2108 	kvmppc_mmu_hpte_sysexit();
2109 }
2110 
2111 /*
2112  * We only support separate modules for book3s 64
2113  */
2114 #ifdef CONFIG_PPC_BOOK3S_64
2115 
2116 module_init(kvmppc_book3s_init_pr);
2117 module_exit(kvmppc_book3s_exit_pr);
2118 
2119 MODULE_LICENSE("GPL");
2120 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2121 MODULE_ALIAS("devname:kvm");
2122 #endif
2123