xref: /openbmc/linux/arch/powerpc/kvm/book3s_pr.c (revision 35267cea)
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 #endif
141 
142 	/* Disable AIL if supported */
143 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
144 	    cpu_has_feature(CPU_FTR_ARCH_207S))
145 		mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) & ~LPCR_AIL);
146 
147 	vcpu->cpu = smp_processor_id();
148 #ifdef CONFIG_PPC_BOOK3S_32
149 	current->thread.kvm_shadow_vcpu = vcpu->arch.shadow_vcpu;
150 #endif
151 
152 	if (kvmppc_is_split_real(vcpu))
153 		kvmppc_fixup_split_real(vcpu);
154 
155 	kvmppc_restore_tm_pr(vcpu);
156 }
157 
158 static void kvmppc_core_vcpu_put_pr(struct kvm_vcpu *vcpu)
159 {
160 #ifdef CONFIG_PPC_BOOK3S_64
161 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
162 	if (svcpu->in_use) {
163 		kvmppc_copy_from_svcpu(vcpu);
164 	}
165 	memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
166 	to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max;
167 	svcpu_put(svcpu);
168 #endif
169 
170 	if (kvmppc_is_split_real(vcpu))
171 		kvmppc_unfixup_split_real(vcpu);
172 
173 	kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
174 	kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
175 	kvmppc_save_tm_pr(vcpu);
176 
177 	/* Enable AIL if supported */
178 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
179 	    cpu_has_feature(CPU_FTR_ARCH_207S))
180 		mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_AIL_3);
181 
182 	vcpu->cpu = -1;
183 }
184 
185 /* Copy data needed by real-mode code from vcpu to shadow vcpu */
186 void kvmppc_copy_to_svcpu(struct kvm_vcpu *vcpu)
187 {
188 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
189 
190 	svcpu->gpr[0] = vcpu->arch.regs.gpr[0];
191 	svcpu->gpr[1] = vcpu->arch.regs.gpr[1];
192 	svcpu->gpr[2] = vcpu->arch.regs.gpr[2];
193 	svcpu->gpr[3] = vcpu->arch.regs.gpr[3];
194 	svcpu->gpr[4] = vcpu->arch.regs.gpr[4];
195 	svcpu->gpr[5] = vcpu->arch.regs.gpr[5];
196 	svcpu->gpr[6] = vcpu->arch.regs.gpr[6];
197 	svcpu->gpr[7] = vcpu->arch.regs.gpr[7];
198 	svcpu->gpr[8] = vcpu->arch.regs.gpr[8];
199 	svcpu->gpr[9] = vcpu->arch.regs.gpr[9];
200 	svcpu->gpr[10] = vcpu->arch.regs.gpr[10];
201 	svcpu->gpr[11] = vcpu->arch.regs.gpr[11];
202 	svcpu->gpr[12] = vcpu->arch.regs.gpr[12];
203 	svcpu->gpr[13] = vcpu->arch.regs.gpr[13];
204 	svcpu->cr  = vcpu->arch.regs.ccr;
205 	svcpu->xer = vcpu->arch.regs.xer;
206 	svcpu->ctr = vcpu->arch.regs.ctr;
207 	svcpu->lr  = vcpu->arch.regs.link;
208 	svcpu->pc  = vcpu->arch.regs.nip;
209 #ifdef CONFIG_PPC_BOOK3S_64
210 	svcpu->shadow_fscr = vcpu->arch.shadow_fscr;
211 #endif
212 	/*
213 	 * Now also save the current time base value. We use this
214 	 * to find the guest purr and spurr value.
215 	 */
216 	vcpu->arch.entry_tb = get_tb();
217 	vcpu->arch.entry_vtb = get_vtb();
218 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
219 		vcpu->arch.entry_ic = mfspr(SPRN_IC);
220 	svcpu->in_use = true;
221 
222 	svcpu_put(svcpu);
223 }
224 
225 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
226 {
227 	ulong guest_msr = kvmppc_get_msr(vcpu);
228 	ulong smsr = guest_msr;
229 
230 	/* Guest MSR values */
231 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
232 	smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_LE |
233 		MSR_TM | MSR_TS_MASK;
234 #else
235 	smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_LE;
236 #endif
237 	/* Process MSR values */
238 	smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
239 	/* External providers the guest reserved */
240 	smsr |= (guest_msr & vcpu->arch.guest_owned_ext);
241 	/* 64-bit Process MSR values */
242 #ifdef CONFIG_PPC_BOOK3S_64
243 	smsr |= MSR_HV;
244 #endif
245 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
246 	/*
247 	 * in guest privileged state, we want to fail all TM transactions.
248 	 * So disable MSR TM bit so that all tbegin. will be able to be
249 	 * trapped into host.
250 	 */
251 	if (!(guest_msr & MSR_PR))
252 		smsr &= ~MSR_TM;
253 #endif
254 	vcpu->arch.shadow_msr = smsr;
255 }
256 
257 /* Copy data touched by real-mode code from shadow vcpu back to vcpu */
258 void kvmppc_copy_from_svcpu(struct kvm_vcpu *vcpu)
259 {
260 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
261 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
262 	ulong old_msr;
263 #endif
264 
265 	/*
266 	 * Maybe we were already preempted and synced the svcpu from
267 	 * our preempt notifiers. Don't bother touching this svcpu then.
268 	 */
269 	if (!svcpu->in_use)
270 		goto out;
271 
272 	vcpu->arch.regs.gpr[0] = svcpu->gpr[0];
273 	vcpu->arch.regs.gpr[1] = svcpu->gpr[1];
274 	vcpu->arch.regs.gpr[2] = svcpu->gpr[2];
275 	vcpu->arch.regs.gpr[3] = svcpu->gpr[3];
276 	vcpu->arch.regs.gpr[4] = svcpu->gpr[4];
277 	vcpu->arch.regs.gpr[5] = svcpu->gpr[5];
278 	vcpu->arch.regs.gpr[6] = svcpu->gpr[6];
279 	vcpu->arch.regs.gpr[7] = svcpu->gpr[7];
280 	vcpu->arch.regs.gpr[8] = svcpu->gpr[8];
281 	vcpu->arch.regs.gpr[9] = svcpu->gpr[9];
282 	vcpu->arch.regs.gpr[10] = svcpu->gpr[10];
283 	vcpu->arch.regs.gpr[11] = svcpu->gpr[11];
284 	vcpu->arch.regs.gpr[12] = svcpu->gpr[12];
285 	vcpu->arch.regs.gpr[13] = svcpu->gpr[13];
286 	vcpu->arch.regs.ccr  = svcpu->cr;
287 	vcpu->arch.regs.xer = svcpu->xer;
288 	vcpu->arch.regs.ctr = svcpu->ctr;
289 	vcpu->arch.regs.link  = svcpu->lr;
290 	vcpu->arch.regs.nip  = svcpu->pc;
291 	vcpu->arch.shadow_srr1 = svcpu->shadow_srr1;
292 	vcpu->arch.fault_dar   = svcpu->fault_dar;
293 	vcpu->arch.fault_dsisr = svcpu->fault_dsisr;
294 	vcpu->arch.last_inst   = svcpu->last_inst;
295 #ifdef CONFIG_PPC_BOOK3S_64
296 	vcpu->arch.shadow_fscr = svcpu->shadow_fscr;
297 #endif
298 	/*
299 	 * Update purr and spurr using time base on exit.
300 	 */
301 	vcpu->arch.purr += get_tb() - vcpu->arch.entry_tb;
302 	vcpu->arch.spurr += get_tb() - vcpu->arch.entry_tb;
303 	to_book3s(vcpu)->vtb += get_vtb() - vcpu->arch.entry_vtb;
304 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
305 		vcpu->arch.ic += mfspr(SPRN_IC) - vcpu->arch.entry_ic;
306 
307 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
308 	/*
309 	 * Unlike other MSR bits, MSR[TS]bits can be changed at guest without
310 	 * notifying host:
311 	 *  modified by unprivileged instructions like "tbegin"/"tend"/
312 	 * "tresume"/"tsuspend" in PR KVM guest.
313 	 *
314 	 * It is necessary to sync here to calculate a correct shadow_msr.
315 	 *
316 	 * privileged guest's tbegin will be failed at present. So we
317 	 * only take care of problem state guest.
318 	 */
319 	old_msr = kvmppc_get_msr(vcpu);
320 	if (unlikely((old_msr & MSR_PR) &&
321 		(vcpu->arch.shadow_srr1 & (MSR_TS_MASK)) !=
322 				(old_msr & (MSR_TS_MASK)))) {
323 		old_msr &= ~(MSR_TS_MASK);
324 		old_msr |= (vcpu->arch.shadow_srr1 & (MSR_TS_MASK));
325 		kvmppc_set_msr_fast(vcpu, old_msr);
326 		kvmppc_recalc_shadow_msr(vcpu);
327 	}
328 #endif
329 
330 	svcpu->in_use = false;
331 
332 out:
333 	svcpu_put(svcpu);
334 }
335 
336 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
337 void kvmppc_save_tm_sprs(struct kvm_vcpu *vcpu)
338 {
339 	tm_enable();
340 	vcpu->arch.tfhar = mfspr(SPRN_TFHAR);
341 	vcpu->arch.texasr = mfspr(SPRN_TEXASR);
342 	vcpu->arch.tfiar = mfspr(SPRN_TFIAR);
343 	tm_disable();
344 }
345 
346 void kvmppc_restore_tm_sprs(struct kvm_vcpu *vcpu)
347 {
348 	tm_enable();
349 	mtspr(SPRN_TFHAR, vcpu->arch.tfhar);
350 	mtspr(SPRN_TEXASR, vcpu->arch.texasr);
351 	mtspr(SPRN_TFIAR, vcpu->arch.tfiar);
352 	tm_disable();
353 }
354 
355 /* loadup math bits which is enabled at kvmppc_get_msr() but not enabled at
356  * hardware.
357  */
358 static void kvmppc_handle_lost_math_exts(struct kvm_vcpu *vcpu)
359 {
360 	ulong exit_nr;
361 	ulong ext_diff = (kvmppc_get_msr(vcpu) & ~vcpu->arch.guest_owned_ext) &
362 		(MSR_FP | MSR_VEC | MSR_VSX);
363 
364 	if (!ext_diff)
365 		return;
366 
367 	if (ext_diff == MSR_FP)
368 		exit_nr = BOOK3S_INTERRUPT_FP_UNAVAIL;
369 	else if (ext_diff == MSR_VEC)
370 		exit_nr = BOOK3S_INTERRUPT_ALTIVEC;
371 	else
372 		exit_nr = BOOK3S_INTERRUPT_VSX;
373 
374 	kvmppc_handle_ext(vcpu, exit_nr, ext_diff);
375 }
376 
377 void kvmppc_save_tm_pr(struct kvm_vcpu *vcpu)
378 {
379 	if (!(MSR_TM_ACTIVE(kvmppc_get_msr(vcpu)))) {
380 		kvmppc_save_tm_sprs(vcpu);
381 		return;
382 	}
383 
384 	kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
385 	kvmppc_giveup_ext(vcpu, MSR_VSX);
386 
387 	preempt_disable();
388 	_kvmppc_save_tm_pr(vcpu, mfmsr());
389 	preempt_enable();
390 }
391 
392 void kvmppc_restore_tm_pr(struct kvm_vcpu *vcpu)
393 {
394 	if (!MSR_TM_ACTIVE(kvmppc_get_msr(vcpu))) {
395 		kvmppc_restore_tm_sprs(vcpu);
396 		if (kvmppc_get_msr(vcpu) & MSR_TM) {
397 			kvmppc_handle_lost_math_exts(vcpu);
398 			if (vcpu->arch.fscr & FSCR_TAR)
399 				kvmppc_handle_fac(vcpu, FSCR_TAR_LG);
400 		}
401 		return;
402 	}
403 
404 	preempt_disable();
405 	_kvmppc_restore_tm_pr(vcpu, kvmppc_get_msr(vcpu));
406 	preempt_enable();
407 
408 	if (kvmppc_get_msr(vcpu) & MSR_TM) {
409 		kvmppc_handle_lost_math_exts(vcpu);
410 		if (vcpu->arch.fscr & FSCR_TAR)
411 			kvmppc_handle_fac(vcpu, FSCR_TAR_LG);
412 	}
413 }
414 #endif
415 
416 static int kvmppc_core_check_requests_pr(struct kvm_vcpu *vcpu)
417 {
418 	int r = 1; /* Indicate we want to get back into the guest */
419 
420 	/* We misuse TLB_FLUSH to indicate that we want to clear
421 	   all shadow cache entries */
422 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
423 		kvmppc_mmu_pte_flush(vcpu, 0, 0);
424 
425 	return r;
426 }
427 
428 /************* MMU Notifiers *************/
429 static bool do_kvm_unmap_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
430 {
431 	long i;
432 	struct kvm_vcpu *vcpu;
433 
434 	kvm_for_each_vcpu(i, vcpu, kvm)
435 		kvmppc_mmu_pte_pflush(vcpu, range->start << PAGE_SHIFT,
436 				      range->end << PAGE_SHIFT);
437 
438 	return false;
439 }
440 
441 static bool kvm_unmap_gfn_range_pr(struct kvm *kvm, struct kvm_gfn_range *range)
442 {
443 	return do_kvm_unmap_gfn(kvm, range);
444 }
445 
446 static bool kvm_age_gfn_pr(struct kvm *kvm, struct kvm_gfn_range *range)
447 {
448 	/* XXX could be more clever ;) */
449 	return false;
450 }
451 
452 static bool kvm_test_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_set_spte_gfn_pr(struct kvm *kvm, struct kvm_gfn_range *range)
459 {
460 	/* The page will get remapped properly on its next fault */
461 	return do_kvm_unmap_gfn(kvm, range);
462 }
463 
464 /*****************************************/
465 
466 static void kvmppc_set_msr_pr(struct kvm_vcpu *vcpu, u64 msr)
467 {
468 	ulong old_msr;
469 
470 	/* For PAPR guest, make sure MSR reflects guest mode */
471 	if (vcpu->arch.papr_enabled)
472 		msr = (msr & ~MSR_HV) | MSR_ME;
473 
474 #ifdef EXIT_DEBUG
475 	printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
476 #endif
477 
478 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
479 	/* We should never target guest MSR to TS=10 && PR=0,
480 	 * since we always fail transaction for guest privilege
481 	 * state.
482 	 */
483 	if (!(msr & MSR_PR) && MSR_TM_TRANSACTIONAL(msr))
484 		kvmppc_emulate_tabort(vcpu,
485 			TM_CAUSE_KVM_FAC_UNAV | TM_CAUSE_PERSISTENT);
486 #endif
487 
488 	old_msr = kvmppc_get_msr(vcpu);
489 	msr &= to_book3s(vcpu)->msr_mask;
490 	kvmppc_set_msr_fast(vcpu, msr);
491 	kvmppc_recalc_shadow_msr(vcpu);
492 
493 	if (msr & MSR_POW) {
494 		if (!vcpu->arch.pending_exceptions) {
495 			kvm_vcpu_block(vcpu);
496 			kvm_clear_request(KVM_REQ_UNHALT, vcpu);
497 			vcpu->stat.generic.halt_wakeup++;
498 
499 			/* Unset POW bit after we woke up */
500 			msr &= ~MSR_POW;
501 			kvmppc_set_msr_fast(vcpu, msr);
502 		}
503 	}
504 
505 	if (kvmppc_is_split_real(vcpu))
506 		kvmppc_fixup_split_real(vcpu);
507 	else
508 		kvmppc_unfixup_split_real(vcpu);
509 
510 	if ((kvmppc_get_msr(vcpu) & (MSR_PR|MSR_IR|MSR_DR)) !=
511 		   (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
512 		kvmppc_mmu_flush_segments(vcpu);
513 		kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
514 
515 		/* Preload magic page segment when in kernel mode */
516 		if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
517 			struct kvm_vcpu_arch *a = &vcpu->arch;
518 
519 			if (msr & MSR_DR)
520 				kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
521 			else
522 				kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
523 		}
524 	}
525 
526 	/*
527 	 * When switching from 32 to 64-bit, we may have a stale 32-bit
528 	 * magic page around, we need to flush it. Typically 32-bit magic
529 	 * page will be instantiated when calling into RTAS. Note: We
530 	 * assume that such transition only happens while in kernel mode,
531 	 * ie, we never transition from user 32-bit to kernel 64-bit with
532 	 * a 32-bit magic page around.
533 	 */
534 	if (vcpu->arch.magic_page_pa &&
535 	    !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
536 		/* going from RTAS to normal kernel code */
537 		kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
538 				     ~0xFFFUL);
539 	}
540 
541 	/* Preload FPU if it's enabled */
542 	if (kvmppc_get_msr(vcpu) & MSR_FP)
543 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
544 
545 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
546 	if (kvmppc_get_msr(vcpu) & MSR_TM)
547 		kvmppc_handle_lost_math_exts(vcpu);
548 #endif
549 }
550 
551 static void kvmppc_set_pvr_pr(struct kvm_vcpu *vcpu, u32 pvr)
552 {
553 	u32 host_pvr;
554 
555 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
556 	vcpu->arch.pvr = pvr;
557 #ifdef CONFIG_PPC_BOOK3S_64
558 	if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
559 		kvmppc_mmu_book3s_64_init(vcpu);
560 		if (!to_book3s(vcpu)->hior_explicit)
561 			to_book3s(vcpu)->hior = 0xfff00000;
562 		to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
563 		vcpu->arch.cpu_type = KVM_CPU_3S_64;
564 	} else
565 #endif
566 	{
567 		kvmppc_mmu_book3s_32_init(vcpu);
568 		if (!to_book3s(vcpu)->hior_explicit)
569 			to_book3s(vcpu)->hior = 0;
570 		to_book3s(vcpu)->msr_mask = 0xffffffffULL;
571 		vcpu->arch.cpu_type = KVM_CPU_3S_32;
572 	}
573 
574 	kvmppc_sanity_check(vcpu);
575 
576 	/* If we are in hypervisor level on 970, we can tell the CPU to
577 	 * treat DCBZ as 32 bytes store */
578 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
579 	if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
580 	    !strcmp(cur_cpu_spec->platform, "ppc970"))
581 		vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
582 
583 	/* Cell performs badly if MSR_FEx are set. So let's hope nobody
584 	   really needs them in a VM on Cell and force disable them. */
585 	if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
586 		to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
587 
588 	/*
589 	 * If they're asking for POWER6 or later, set the flag
590 	 * indicating that we can do multiple large page sizes
591 	 * and 1TB segments.
592 	 * Also set the flag that indicates that tlbie has the large
593 	 * page bit in the RB operand instead of the instruction.
594 	 */
595 	switch (PVR_VER(pvr)) {
596 	case PVR_POWER6:
597 	case PVR_POWER7:
598 	case PVR_POWER7p:
599 	case PVR_POWER8:
600 	case PVR_POWER8E:
601 	case PVR_POWER8NVL:
602 	case PVR_POWER9:
603 		vcpu->arch.hflags |= BOOK3S_HFLAG_MULTI_PGSIZE |
604 			BOOK3S_HFLAG_NEW_TLBIE;
605 		break;
606 	}
607 
608 #ifdef CONFIG_PPC_BOOK3S_32
609 	/* 32 bit Book3S always has 32 byte dcbz */
610 	vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
611 #endif
612 
613 	/* On some CPUs we can execute paired single operations natively */
614 	asm ( "mfpvr %0" : "=r"(host_pvr));
615 	switch (host_pvr) {
616 	case 0x00080200:	/* lonestar 2.0 */
617 	case 0x00088202:	/* lonestar 2.2 */
618 	case 0x70000100:	/* gekko 1.0 */
619 	case 0x00080100:	/* gekko 2.0 */
620 	case 0x00083203:	/* gekko 2.3a */
621 	case 0x00083213:	/* gekko 2.3b */
622 	case 0x00083204:	/* gekko 2.4 */
623 	case 0x00083214:	/* gekko 2.4e (8SE) - retail HW2 */
624 	case 0x00087200:	/* broadway */
625 		vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
626 		/* Enable HID2.PSE - in case we need it later */
627 		mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
628 	}
629 }
630 
631 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
632  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
633  * emulate 32 bytes dcbz length.
634  *
635  * The Book3s_64 inventors also realized this case and implemented a special bit
636  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
637  *
638  * My approach here is to patch the dcbz instruction on executing pages.
639  */
640 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
641 {
642 	struct page *hpage;
643 	u64 hpage_offset;
644 	u32 *page;
645 	int i;
646 
647 	hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
648 	if (is_error_page(hpage))
649 		return;
650 
651 	hpage_offset = pte->raddr & ~PAGE_MASK;
652 	hpage_offset &= ~0xFFFULL;
653 	hpage_offset /= 4;
654 
655 	get_page(hpage);
656 	page = kmap_atomic(hpage);
657 
658 	/* patch dcbz into reserved instruction, so we trap */
659 	for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
660 		if ((be32_to_cpu(page[i]) & 0xff0007ff) == INS_DCBZ)
661 			page[i] &= cpu_to_be32(0xfffffff7);
662 
663 	kunmap_atomic(page);
664 	put_page(hpage);
665 }
666 
667 static bool kvmppc_visible_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
668 {
669 	ulong mp_pa = vcpu->arch.magic_page_pa;
670 
671 	if (!(kvmppc_get_msr(vcpu) & MSR_SF))
672 		mp_pa = (uint32_t)mp_pa;
673 
674 	gpa &= ~0xFFFULL;
675 	if (unlikely(mp_pa) && unlikely((mp_pa & KVM_PAM) == (gpa & KVM_PAM))) {
676 		return true;
677 	}
678 
679 	return kvm_is_visible_gfn(vcpu->kvm, gpa >> PAGE_SHIFT);
680 }
681 
682 static int kvmppc_handle_pagefault(struct kvm_vcpu *vcpu,
683 			    ulong eaddr, int vec)
684 {
685 	bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
686 	bool iswrite = false;
687 	int r = RESUME_GUEST;
688 	int relocated;
689 	int page_found = 0;
690 	struct kvmppc_pte pte = { 0 };
691 	bool dr = (kvmppc_get_msr(vcpu) & MSR_DR) ? true : false;
692 	bool ir = (kvmppc_get_msr(vcpu) & MSR_IR) ? true : false;
693 	u64 vsid;
694 
695 	relocated = data ? dr : ir;
696 	if (data && (vcpu->arch.fault_dsisr & DSISR_ISSTORE))
697 		iswrite = true;
698 
699 	/* Resolve real address if translation turned on */
700 	if (relocated) {
701 		page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data, iswrite);
702 	} else {
703 		pte.may_execute = true;
704 		pte.may_read = true;
705 		pte.may_write = true;
706 		pte.raddr = eaddr & KVM_PAM;
707 		pte.eaddr = eaddr;
708 		pte.vpage = eaddr >> 12;
709 		pte.page_size = MMU_PAGE_64K;
710 		pte.wimg = HPTE_R_M;
711 	}
712 
713 	switch (kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) {
714 	case 0:
715 		pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
716 		break;
717 	case MSR_DR:
718 		if (!data &&
719 		    (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK) &&
720 		    ((pte.raddr & SPLIT_HACK_MASK) == SPLIT_HACK_OFFS))
721 			pte.raddr &= ~SPLIT_HACK_MASK;
722 		fallthrough;
723 	case MSR_IR:
724 		vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
725 
726 		if ((kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) == MSR_DR)
727 			pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
728 		else
729 			pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
730 		pte.vpage |= vsid;
731 
732 		if (vsid == -1)
733 			page_found = -EINVAL;
734 		break;
735 	}
736 
737 	if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
738 	   (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
739 		/*
740 		 * If we do the dcbz hack, we have to NX on every execution,
741 		 * so we can patch the executing code. This renders our guest
742 		 * NX-less.
743 		 */
744 		pte.may_execute = !data;
745 	}
746 
747 	if (page_found == -ENOENT || page_found == -EPERM) {
748 		/* Page not found in guest PTE entries, or protection fault */
749 		u64 flags;
750 
751 		if (page_found == -EPERM)
752 			flags = DSISR_PROTFAULT;
753 		else
754 			flags = DSISR_NOHPTE;
755 		if (data) {
756 			flags |= vcpu->arch.fault_dsisr & DSISR_ISSTORE;
757 			kvmppc_core_queue_data_storage(vcpu, eaddr, flags);
758 		} else {
759 			kvmppc_core_queue_inst_storage(vcpu, flags);
760 		}
761 	} else if (page_found == -EINVAL) {
762 		/* Page not found in guest SLB */
763 		kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
764 		kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
765 	} else if (kvmppc_visible_gpa(vcpu, pte.raddr)) {
766 		if (data && !(vcpu->arch.fault_dsisr & DSISR_NOHPTE)) {
767 			/*
768 			 * There is already a host HPTE there, presumably
769 			 * a read-only one for a page the guest thinks
770 			 * is writable, so get rid of it first.
771 			 */
772 			kvmppc_mmu_unmap_page(vcpu, &pte);
773 		}
774 		/* The guest's PTE is not mapped yet. Map on the host */
775 		if (kvmppc_mmu_map_page(vcpu, &pte, iswrite) == -EIO) {
776 			/* Exit KVM if mapping failed */
777 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
778 			return RESUME_HOST;
779 		}
780 		if (data)
781 			vcpu->stat.sp_storage++;
782 		else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
783 			 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
784 			kvmppc_patch_dcbz(vcpu, &pte);
785 	} else {
786 		/* MMIO */
787 		vcpu->stat.mmio_exits++;
788 		vcpu->arch.paddr_accessed = pte.raddr;
789 		vcpu->arch.vaddr_accessed = pte.eaddr;
790 		r = kvmppc_emulate_mmio(vcpu);
791 		if ( r == RESUME_HOST_NV )
792 			r = RESUME_HOST;
793 	}
794 
795 	return r;
796 }
797 
798 /* Give up external provider (FPU, Altivec, VSX) */
799 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
800 {
801 	struct thread_struct *t = &current->thread;
802 
803 	/*
804 	 * VSX instructions can access FP and vector registers, so if
805 	 * we are giving up VSX, make sure we give up FP and VMX as well.
806 	 */
807 	if (msr & MSR_VSX)
808 		msr |= MSR_FP | MSR_VEC;
809 
810 	msr &= vcpu->arch.guest_owned_ext;
811 	if (!msr)
812 		return;
813 
814 #ifdef DEBUG_EXT
815 	printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
816 #endif
817 
818 	if (msr & MSR_FP) {
819 		/*
820 		 * Note that on CPUs with VSX, giveup_fpu stores
821 		 * both the traditional FP registers and the added VSX
822 		 * registers into thread.fp_state.fpr[].
823 		 */
824 		if (t->regs->msr & MSR_FP)
825 			giveup_fpu(current);
826 		t->fp_save_area = NULL;
827 	}
828 
829 #ifdef CONFIG_ALTIVEC
830 	if (msr & MSR_VEC) {
831 		if (current->thread.regs->msr & MSR_VEC)
832 			giveup_altivec(current);
833 		t->vr_save_area = NULL;
834 	}
835 #endif
836 
837 	vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
838 	kvmppc_recalc_shadow_msr(vcpu);
839 }
840 
841 /* Give up facility (TAR / EBB / DSCR) */
842 void kvmppc_giveup_fac(struct kvm_vcpu *vcpu, ulong fac)
843 {
844 #ifdef CONFIG_PPC_BOOK3S_64
845 	if (!(vcpu->arch.shadow_fscr & (1ULL << fac))) {
846 		/* Facility not available to the guest, ignore giveup request*/
847 		return;
848 	}
849 
850 	switch (fac) {
851 	case FSCR_TAR_LG:
852 		vcpu->arch.tar = mfspr(SPRN_TAR);
853 		mtspr(SPRN_TAR, current->thread.tar);
854 		vcpu->arch.shadow_fscr &= ~FSCR_TAR;
855 		break;
856 	}
857 #endif
858 }
859 
860 /* Handle external providers (FPU, Altivec, VSX) */
861 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
862 			     ulong msr)
863 {
864 	struct thread_struct *t = &current->thread;
865 
866 	/* When we have paired singles, we emulate in software */
867 	if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
868 		return RESUME_GUEST;
869 
870 	if (!(kvmppc_get_msr(vcpu) & msr)) {
871 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
872 		return RESUME_GUEST;
873 	}
874 
875 	if (msr == MSR_VSX) {
876 		/* No VSX?  Give an illegal instruction interrupt */
877 #ifdef CONFIG_VSX
878 		if (!cpu_has_feature(CPU_FTR_VSX))
879 #endif
880 		{
881 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
882 			return RESUME_GUEST;
883 		}
884 
885 		/*
886 		 * We have to load up all the FP and VMX registers before
887 		 * we can let the guest use VSX instructions.
888 		 */
889 		msr = MSR_FP | MSR_VEC | MSR_VSX;
890 	}
891 
892 	/* See if we already own all the ext(s) needed */
893 	msr &= ~vcpu->arch.guest_owned_ext;
894 	if (!msr)
895 		return RESUME_GUEST;
896 
897 #ifdef DEBUG_EXT
898 	printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
899 #endif
900 
901 	if (msr & MSR_FP) {
902 		preempt_disable();
903 		enable_kernel_fp();
904 		load_fp_state(&vcpu->arch.fp);
905 		disable_kernel_fp();
906 		t->fp_save_area = &vcpu->arch.fp;
907 		preempt_enable();
908 	}
909 
910 	if (msr & MSR_VEC) {
911 #ifdef CONFIG_ALTIVEC
912 		preempt_disable();
913 		enable_kernel_altivec();
914 		load_vr_state(&vcpu->arch.vr);
915 		disable_kernel_altivec();
916 		t->vr_save_area = &vcpu->arch.vr;
917 		preempt_enable();
918 #endif
919 	}
920 
921 	t->regs->msr |= msr;
922 	vcpu->arch.guest_owned_ext |= msr;
923 	kvmppc_recalc_shadow_msr(vcpu);
924 
925 	return RESUME_GUEST;
926 }
927 
928 /*
929  * Kernel code using FP or VMX could have flushed guest state to
930  * the thread_struct; if so, get it back now.
931  */
932 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
933 {
934 	unsigned long lost_ext;
935 
936 	lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
937 	if (!lost_ext)
938 		return;
939 
940 	if (lost_ext & MSR_FP) {
941 		preempt_disable();
942 		enable_kernel_fp();
943 		load_fp_state(&vcpu->arch.fp);
944 		disable_kernel_fp();
945 		preempt_enable();
946 	}
947 #ifdef CONFIG_ALTIVEC
948 	if (lost_ext & MSR_VEC) {
949 		preempt_disable();
950 		enable_kernel_altivec();
951 		load_vr_state(&vcpu->arch.vr);
952 		disable_kernel_altivec();
953 		preempt_enable();
954 	}
955 #endif
956 	current->thread.regs->msr |= lost_ext;
957 }
958 
959 #ifdef CONFIG_PPC_BOOK3S_64
960 
961 void kvmppc_trigger_fac_interrupt(struct kvm_vcpu *vcpu, ulong fac)
962 {
963 	/* Inject the Interrupt Cause field and trigger a guest interrupt */
964 	vcpu->arch.fscr &= ~(0xffULL << 56);
965 	vcpu->arch.fscr |= (fac << 56);
966 	kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_FAC_UNAVAIL);
967 }
968 
969 static void kvmppc_emulate_fac(struct kvm_vcpu *vcpu, ulong fac)
970 {
971 	enum emulation_result er = EMULATE_FAIL;
972 
973 	if (!(kvmppc_get_msr(vcpu) & MSR_PR))
974 		er = kvmppc_emulate_instruction(vcpu);
975 
976 	if ((er != EMULATE_DONE) && (er != EMULATE_AGAIN)) {
977 		/* Couldn't emulate, trigger interrupt in guest */
978 		kvmppc_trigger_fac_interrupt(vcpu, fac);
979 	}
980 }
981 
982 /* Enable facilities (TAR, EBB, DSCR) for the guest */
983 static int kvmppc_handle_fac(struct kvm_vcpu *vcpu, ulong fac)
984 {
985 	bool guest_fac_enabled;
986 	BUG_ON(!cpu_has_feature(CPU_FTR_ARCH_207S));
987 
988 	/*
989 	 * Not every facility is enabled by FSCR bits, check whether the
990 	 * guest has this facility enabled at all.
991 	 */
992 	switch (fac) {
993 	case FSCR_TAR_LG:
994 	case FSCR_EBB_LG:
995 		guest_fac_enabled = (vcpu->arch.fscr & (1ULL << fac));
996 		break;
997 	case FSCR_TM_LG:
998 		guest_fac_enabled = kvmppc_get_msr(vcpu) & MSR_TM;
999 		break;
1000 	default:
1001 		guest_fac_enabled = false;
1002 		break;
1003 	}
1004 
1005 	if (!guest_fac_enabled) {
1006 		/* Facility not enabled by the guest */
1007 		kvmppc_trigger_fac_interrupt(vcpu, fac);
1008 		return RESUME_GUEST;
1009 	}
1010 
1011 	switch (fac) {
1012 	case FSCR_TAR_LG:
1013 		/* TAR switching isn't lazy in Linux yet */
1014 		current->thread.tar = mfspr(SPRN_TAR);
1015 		mtspr(SPRN_TAR, vcpu->arch.tar);
1016 		vcpu->arch.shadow_fscr |= FSCR_TAR;
1017 		break;
1018 	default:
1019 		kvmppc_emulate_fac(vcpu, fac);
1020 		break;
1021 	}
1022 
1023 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1024 	/* Since we disabled MSR_TM at privilege state, the mfspr instruction
1025 	 * for TM spr can trigger TM fac unavailable. In this case, the
1026 	 * emulation is handled by kvmppc_emulate_fac(), which invokes
1027 	 * kvmppc_emulate_mfspr() finally. But note the mfspr can include
1028 	 * RT for NV registers. So it need to restore those NV reg to reflect
1029 	 * the update.
1030 	 */
1031 	if ((fac == FSCR_TM_LG) && !(kvmppc_get_msr(vcpu) & MSR_PR))
1032 		return RESUME_GUEST_NV;
1033 #endif
1034 
1035 	return RESUME_GUEST;
1036 }
1037 
1038 void kvmppc_set_fscr(struct kvm_vcpu *vcpu, u64 fscr)
1039 {
1040 	if ((vcpu->arch.fscr & FSCR_TAR) && !(fscr & FSCR_TAR)) {
1041 		/* TAR got dropped, drop it in shadow too */
1042 		kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
1043 	} else if (!(vcpu->arch.fscr & FSCR_TAR) && (fscr & FSCR_TAR)) {
1044 		vcpu->arch.fscr = fscr;
1045 		kvmppc_handle_fac(vcpu, FSCR_TAR_LG);
1046 		return;
1047 	}
1048 
1049 	vcpu->arch.fscr = fscr;
1050 }
1051 #endif
1052 
1053 static void kvmppc_setup_debug(struct kvm_vcpu *vcpu)
1054 {
1055 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
1056 		u64 msr = kvmppc_get_msr(vcpu);
1057 
1058 		kvmppc_set_msr(vcpu, msr | MSR_SE);
1059 	}
1060 }
1061 
1062 static void kvmppc_clear_debug(struct kvm_vcpu *vcpu)
1063 {
1064 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
1065 		u64 msr = kvmppc_get_msr(vcpu);
1066 
1067 		kvmppc_set_msr(vcpu, msr & ~MSR_SE);
1068 	}
1069 }
1070 
1071 static int kvmppc_exit_pr_progint(struct kvm_vcpu *vcpu, unsigned int exit_nr)
1072 {
1073 	enum emulation_result er;
1074 	ulong flags;
1075 	u32 last_inst;
1076 	int emul, r;
1077 
1078 	/*
1079 	 * shadow_srr1 only contains valid flags if we came here via a program
1080 	 * exception. The other exceptions (emulation assist, FP unavailable,
1081 	 * etc.) do not provide flags in SRR1, so use an illegal-instruction
1082 	 * exception when injecting a program interrupt into the guest.
1083 	 */
1084 	if (exit_nr == BOOK3S_INTERRUPT_PROGRAM)
1085 		flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
1086 	else
1087 		flags = SRR1_PROGILL;
1088 
1089 	emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1090 	if (emul != EMULATE_DONE)
1091 		return RESUME_GUEST;
1092 
1093 	if (kvmppc_get_msr(vcpu) & MSR_PR) {
1094 #ifdef EXIT_DEBUG
1095 		pr_info("Userspace triggered 0x700 exception at\n 0x%lx (0x%x)\n",
1096 			kvmppc_get_pc(vcpu), last_inst);
1097 #endif
1098 		if ((last_inst & 0xff0007ff) != (INS_DCBZ & 0xfffffff7)) {
1099 			kvmppc_core_queue_program(vcpu, flags);
1100 			return RESUME_GUEST;
1101 		}
1102 	}
1103 
1104 	vcpu->stat.emulated_inst_exits++;
1105 	er = kvmppc_emulate_instruction(vcpu);
1106 	switch (er) {
1107 	case EMULATE_DONE:
1108 		r = RESUME_GUEST_NV;
1109 		break;
1110 	case EMULATE_AGAIN:
1111 		r = RESUME_GUEST;
1112 		break;
1113 	case EMULATE_FAIL:
1114 		pr_crit("%s: emulation at %lx failed (%08x)\n",
1115 			__func__, kvmppc_get_pc(vcpu), last_inst);
1116 		kvmppc_core_queue_program(vcpu, flags);
1117 		r = RESUME_GUEST;
1118 		break;
1119 	case EMULATE_DO_MMIO:
1120 		vcpu->run->exit_reason = KVM_EXIT_MMIO;
1121 		r = RESUME_HOST_NV;
1122 		break;
1123 	case EMULATE_EXIT_USER:
1124 		r = RESUME_HOST_NV;
1125 		break;
1126 	default:
1127 		BUG();
1128 	}
1129 
1130 	return r;
1131 }
1132 
1133 int kvmppc_handle_exit_pr(struct kvm_vcpu *vcpu, unsigned int exit_nr)
1134 {
1135 	struct kvm_run *run = vcpu->run;
1136 	int r = RESUME_HOST;
1137 	int s;
1138 
1139 	vcpu->stat.sum_exits++;
1140 
1141 	run->exit_reason = KVM_EXIT_UNKNOWN;
1142 	run->ready_for_interrupt_injection = 1;
1143 
1144 	/* We get here with MSR.EE=1 */
1145 
1146 	trace_kvm_exit(exit_nr, vcpu);
1147 	guest_exit();
1148 
1149 	switch (exit_nr) {
1150 	case BOOK3S_INTERRUPT_INST_STORAGE:
1151 	{
1152 		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1153 		vcpu->stat.pf_instruc++;
1154 
1155 		if (kvmppc_is_split_real(vcpu))
1156 			kvmppc_fixup_split_real(vcpu);
1157 
1158 #ifdef CONFIG_PPC_BOOK3S_32
1159 		/* We set segments as unused segments when invalidating them. So
1160 		 * treat the respective fault as segment fault. */
1161 		{
1162 			struct kvmppc_book3s_shadow_vcpu *svcpu;
1163 			u32 sr;
1164 
1165 			svcpu = svcpu_get(vcpu);
1166 			sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
1167 			svcpu_put(svcpu);
1168 			if (sr == SR_INVALID) {
1169 				kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
1170 				r = RESUME_GUEST;
1171 				break;
1172 			}
1173 		}
1174 #endif
1175 
1176 		/* only care about PTEG not found errors, but leave NX alone */
1177 		if (shadow_srr1 & 0x40000000) {
1178 			int idx = srcu_read_lock(&vcpu->kvm->srcu);
1179 			r = kvmppc_handle_pagefault(vcpu, kvmppc_get_pc(vcpu), exit_nr);
1180 			srcu_read_unlock(&vcpu->kvm->srcu, idx);
1181 			vcpu->stat.sp_instruc++;
1182 		} else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
1183 			  (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
1184 			/*
1185 			 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
1186 			 *     so we can't use the NX bit inside the guest. Let's cross our fingers,
1187 			 *     that no guest that needs the dcbz hack does NX.
1188 			 */
1189 			kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
1190 			r = RESUME_GUEST;
1191 		} else {
1192 			kvmppc_core_queue_inst_storage(vcpu,
1193 						shadow_srr1 & 0x58000000);
1194 			r = RESUME_GUEST;
1195 		}
1196 		break;
1197 	}
1198 	case BOOK3S_INTERRUPT_DATA_STORAGE:
1199 	{
1200 		ulong dar = kvmppc_get_fault_dar(vcpu);
1201 		u32 fault_dsisr = vcpu->arch.fault_dsisr;
1202 		vcpu->stat.pf_storage++;
1203 
1204 #ifdef CONFIG_PPC_BOOK3S_32
1205 		/* We set segments as unused segments when invalidating them. So
1206 		 * treat the respective fault as segment fault. */
1207 		{
1208 			struct kvmppc_book3s_shadow_vcpu *svcpu;
1209 			u32 sr;
1210 
1211 			svcpu = svcpu_get(vcpu);
1212 			sr = svcpu->sr[dar >> SID_SHIFT];
1213 			svcpu_put(svcpu);
1214 			if (sr == SR_INVALID) {
1215 				kvmppc_mmu_map_segment(vcpu, dar);
1216 				r = RESUME_GUEST;
1217 				break;
1218 			}
1219 		}
1220 #endif
1221 
1222 		/*
1223 		 * We need to handle missing shadow PTEs, and
1224 		 * protection faults due to us mapping a page read-only
1225 		 * when the guest thinks it is writable.
1226 		 */
1227 		if (fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT)) {
1228 			int idx = srcu_read_lock(&vcpu->kvm->srcu);
1229 			r = kvmppc_handle_pagefault(vcpu, dar, exit_nr);
1230 			srcu_read_unlock(&vcpu->kvm->srcu, idx);
1231 		} else {
1232 			kvmppc_core_queue_data_storage(vcpu, dar, fault_dsisr);
1233 			r = RESUME_GUEST;
1234 		}
1235 		break;
1236 	}
1237 	case BOOK3S_INTERRUPT_DATA_SEGMENT:
1238 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
1239 			kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
1240 			kvmppc_book3s_queue_irqprio(vcpu,
1241 				BOOK3S_INTERRUPT_DATA_SEGMENT);
1242 		}
1243 		r = RESUME_GUEST;
1244 		break;
1245 	case BOOK3S_INTERRUPT_INST_SEGMENT:
1246 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
1247 			kvmppc_book3s_queue_irqprio(vcpu,
1248 				BOOK3S_INTERRUPT_INST_SEGMENT);
1249 		}
1250 		r = RESUME_GUEST;
1251 		break;
1252 	/* We're good on these - the host merely wanted to get our attention */
1253 	case BOOK3S_INTERRUPT_DECREMENTER:
1254 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1255 	case BOOK3S_INTERRUPT_DOORBELL:
1256 	case BOOK3S_INTERRUPT_H_DOORBELL:
1257 		vcpu->stat.dec_exits++;
1258 		r = RESUME_GUEST;
1259 		break;
1260 	case BOOK3S_INTERRUPT_EXTERNAL:
1261 	case BOOK3S_INTERRUPT_EXTERNAL_HV:
1262 	case BOOK3S_INTERRUPT_H_VIRT:
1263 		vcpu->stat.ext_intr_exits++;
1264 		r = RESUME_GUEST;
1265 		break;
1266 	case BOOK3S_INTERRUPT_HMI:
1267 	case BOOK3S_INTERRUPT_PERFMON:
1268 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1269 		r = RESUME_GUEST;
1270 		break;
1271 	case BOOK3S_INTERRUPT_PROGRAM:
1272 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1273 		r = kvmppc_exit_pr_progint(vcpu, exit_nr);
1274 		break;
1275 	case BOOK3S_INTERRUPT_SYSCALL:
1276 	{
1277 		u32 last_sc;
1278 		int emul;
1279 
1280 		/* Get last sc for papr */
1281 		if (vcpu->arch.papr_enabled) {
1282 			/* The sc instuction points SRR0 to the next inst */
1283 			emul = kvmppc_get_last_inst(vcpu, INST_SC, &last_sc);
1284 			if (emul != EMULATE_DONE) {
1285 				kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) - 4);
1286 				r = RESUME_GUEST;
1287 				break;
1288 			}
1289 		}
1290 
1291 		if (vcpu->arch.papr_enabled &&
1292 		    (last_sc == 0x44000022) &&
1293 		    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
1294 			/* SC 1 papr hypercalls */
1295 			ulong cmd = kvmppc_get_gpr(vcpu, 3);
1296 			int i;
1297 
1298 #ifdef CONFIG_PPC_BOOK3S_64
1299 			if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
1300 				r = RESUME_GUEST;
1301 				break;
1302 			}
1303 #endif
1304 
1305 			run->papr_hcall.nr = cmd;
1306 			for (i = 0; i < 9; ++i) {
1307 				ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
1308 				run->papr_hcall.args[i] = gpr;
1309 			}
1310 			run->exit_reason = KVM_EXIT_PAPR_HCALL;
1311 			vcpu->arch.hcall_needed = 1;
1312 			r = RESUME_HOST;
1313 		} else if (vcpu->arch.osi_enabled &&
1314 		    (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
1315 		    (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
1316 			/* MOL hypercalls */
1317 			u64 *gprs = run->osi.gprs;
1318 			int i;
1319 
1320 			run->exit_reason = KVM_EXIT_OSI;
1321 			for (i = 0; i < 32; i++)
1322 				gprs[i] = kvmppc_get_gpr(vcpu, i);
1323 			vcpu->arch.osi_needed = 1;
1324 			r = RESUME_HOST_NV;
1325 		} else if (!(kvmppc_get_msr(vcpu) & MSR_PR) &&
1326 		    (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
1327 			/* KVM PV hypercalls */
1328 			kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
1329 			r = RESUME_GUEST;
1330 		} else {
1331 			/* Guest syscalls */
1332 			vcpu->stat.syscall_exits++;
1333 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1334 			r = RESUME_GUEST;
1335 		}
1336 		break;
1337 	}
1338 	case BOOK3S_INTERRUPT_FP_UNAVAIL:
1339 	case BOOK3S_INTERRUPT_ALTIVEC:
1340 	case BOOK3S_INTERRUPT_VSX:
1341 	{
1342 		int ext_msr = 0;
1343 		int emul;
1344 		u32 last_inst;
1345 
1346 		if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE) {
1347 			/* Do paired single instruction emulation */
1348 			emul = kvmppc_get_last_inst(vcpu, INST_GENERIC,
1349 						    &last_inst);
1350 			if (emul == EMULATE_DONE)
1351 				r = kvmppc_exit_pr_progint(vcpu, exit_nr);
1352 			else
1353 				r = RESUME_GUEST;
1354 
1355 			break;
1356 		}
1357 
1358 		/* Enable external provider */
1359 		switch (exit_nr) {
1360 		case BOOK3S_INTERRUPT_FP_UNAVAIL:
1361 			ext_msr = MSR_FP;
1362 			break;
1363 
1364 		case BOOK3S_INTERRUPT_ALTIVEC:
1365 			ext_msr = MSR_VEC;
1366 			break;
1367 
1368 		case BOOK3S_INTERRUPT_VSX:
1369 			ext_msr = MSR_VSX;
1370 			break;
1371 		}
1372 
1373 		r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
1374 		break;
1375 	}
1376 	case BOOK3S_INTERRUPT_ALIGNMENT:
1377 	{
1378 		u32 last_inst;
1379 		int emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1380 
1381 		if (emul == EMULATE_DONE) {
1382 			u32 dsisr;
1383 			u64 dar;
1384 
1385 			dsisr = kvmppc_alignment_dsisr(vcpu, last_inst);
1386 			dar = kvmppc_alignment_dar(vcpu, last_inst);
1387 
1388 			kvmppc_set_dsisr(vcpu, dsisr);
1389 			kvmppc_set_dar(vcpu, dar);
1390 
1391 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1392 		}
1393 		r = RESUME_GUEST;
1394 		break;
1395 	}
1396 #ifdef CONFIG_PPC_BOOK3S_64
1397 	case BOOK3S_INTERRUPT_FAC_UNAVAIL:
1398 		r = kvmppc_handle_fac(vcpu, vcpu->arch.shadow_fscr >> 56);
1399 		break;
1400 #endif
1401 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1402 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1403 		r = RESUME_GUEST;
1404 		break;
1405 	case BOOK3S_INTERRUPT_TRACE:
1406 		if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
1407 			run->exit_reason = KVM_EXIT_DEBUG;
1408 			r = RESUME_HOST;
1409 		} else {
1410 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1411 			r = RESUME_GUEST;
1412 		}
1413 		break;
1414 	default:
1415 	{
1416 		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1417 		/* Ugh - bork here! What did we get? */
1418 		printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1419 			exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
1420 		r = RESUME_HOST;
1421 		BUG();
1422 		break;
1423 	}
1424 	}
1425 
1426 	if (!(r & RESUME_HOST)) {
1427 		/* To avoid clobbering exit_reason, only check for signals if
1428 		 * we aren't already exiting to userspace for some other
1429 		 * reason. */
1430 
1431 		/*
1432 		 * Interrupts could be timers for the guest which we have to
1433 		 * inject again, so let's postpone them until we're in the guest
1434 		 * and if we really did time things so badly, then we just exit
1435 		 * again due to a host external interrupt.
1436 		 */
1437 		s = kvmppc_prepare_to_enter(vcpu);
1438 		if (s <= 0)
1439 			r = s;
1440 		else {
1441 			/* interrupts now hard-disabled */
1442 			kvmppc_fix_ee_before_entry();
1443 		}
1444 
1445 		kvmppc_handle_lost_ext(vcpu);
1446 	}
1447 
1448 	trace_kvm_book3s_reenter(r, vcpu);
1449 
1450 	return r;
1451 }
1452 
1453 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu *vcpu,
1454 					    struct kvm_sregs *sregs)
1455 {
1456 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1457 	int i;
1458 
1459 	sregs->pvr = vcpu->arch.pvr;
1460 
1461 	sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1462 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1463 		for (i = 0; i < 64; i++) {
1464 			sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
1465 			sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1466 		}
1467 	} else {
1468 		for (i = 0; i < 16; i++)
1469 			sregs->u.s.ppc32.sr[i] = kvmppc_get_sr(vcpu, i);
1470 
1471 		for (i = 0; i < 8; i++) {
1472 			sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1473 			sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1474 		}
1475 	}
1476 
1477 	return 0;
1478 }
1479 
1480 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu *vcpu,
1481 					    struct kvm_sregs *sregs)
1482 {
1483 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1484 	int i;
1485 
1486 	kvmppc_set_pvr_pr(vcpu, sregs->pvr);
1487 
1488 	vcpu3s->sdr1 = sregs->u.s.sdr1;
1489 #ifdef CONFIG_PPC_BOOK3S_64
1490 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1491 		/* Flush all SLB entries */
1492 		vcpu->arch.mmu.slbmte(vcpu, 0, 0);
1493 		vcpu->arch.mmu.slbia(vcpu);
1494 
1495 		for (i = 0; i < 64; i++) {
1496 			u64 rb = sregs->u.s.ppc64.slb[i].slbe;
1497 			u64 rs = sregs->u.s.ppc64.slb[i].slbv;
1498 
1499 			if (rb & SLB_ESID_V)
1500 				vcpu->arch.mmu.slbmte(vcpu, rs, rb);
1501 		}
1502 	} else
1503 #endif
1504 	{
1505 		for (i = 0; i < 16; i++) {
1506 			vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1507 		}
1508 		for (i = 0; i < 8; i++) {
1509 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1510 				       (u32)sregs->u.s.ppc32.ibat[i]);
1511 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1512 				       (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1513 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1514 				       (u32)sregs->u.s.ppc32.dbat[i]);
1515 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1516 				       (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1517 		}
1518 	}
1519 
1520 	/* Flush the MMU after messing with the segments */
1521 	kvmppc_mmu_pte_flush(vcpu, 0, 0);
1522 
1523 	return 0;
1524 }
1525 
1526 static int kvmppc_get_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1527 				 union kvmppc_one_reg *val)
1528 {
1529 	int r = 0;
1530 
1531 	switch (id) {
1532 	case KVM_REG_PPC_DEBUG_INST:
1533 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1534 		break;
1535 	case KVM_REG_PPC_HIOR:
1536 		*val = get_reg_val(id, to_book3s(vcpu)->hior);
1537 		break;
1538 	case KVM_REG_PPC_VTB:
1539 		*val = get_reg_val(id, to_book3s(vcpu)->vtb);
1540 		break;
1541 	case KVM_REG_PPC_LPCR:
1542 	case KVM_REG_PPC_LPCR_64:
1543 		/*
1544 		 * We are only interested in the LPCR_ILE bit
1545 		 */
1546 		if (vcpu->arch.intr_msr & MSR_LE)
1547 			*val = get_reg_val(id, LPCR_ILE);
1548 		else
1549 			*val = get_reg_val(id, 0);
1550 		break;
1551 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1552 	case KVM_REG_PPC_TFHAR:
1553 		*val = get_reg_val(id, vcpu->arch.tfhar);
1554 		break;
1555 	case KVM_REG_PPC_TFIAR:
1556 		*val = get_reg_val(id, vcpu->arch.tfiar);
1557 		break;
1558 	case KVM_REG_PPC_TEXASR:
1559 		*val = get_reg_val(id, vcpu->arch.texasr);
1560 		break;
1561 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1562 		*val = get_reg_val(id,
1563 				vcpu->arch.gpr_tm[id-KVM_REG_PPC_TM_GPR0]);
1564 		break;
1565 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1566 	{
1567 		int i, j;
1568 
1569 		i = id - KVM_REG_PPC_TM_VSR0;
1570 		if (i < 32)
1571 			for (j = 0; j < TS_FPRWIDTH; j++)
1572 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1573 		else {
1574 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1575 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1576 			else
1577 				r = -ENXIO;
1578 		}
1579 		break;
1580 	}
1581 	case KVM_REG_PPC_TM_CR:
1582 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1583 		break;
1584 	case KVM_REG_PPC_TM_XER:
1585 		*val = get_reg_val(id, vcpu->arch.xer_tm);
1586 		break;
1587 	case KVM_REG_PPC_TM_LR:
1588 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1589 		break;
1590 	case KVM_REG_PPC_TM_CTR:
1591 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1592 		break;
1593 	case KVM_REG_PPC_TM_FPSCR:
1594 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1595 		break;
1596 	case KVM_REG_PPC_TM_AMR:
1597 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1598 		break;
1599 	case KVM_REG_PPC_TM_PPR:
1600 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1601 		break;
1602 	case KVM_REG_PPC_TM_VRSAVE:
1603 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1604 		break;
1605 	case KVM_REG_PPC_TM_VSCR:
1606 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1607 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1608 		else
1609 			r = -ENXIO;
1610 		break;
1611 	case KVM_REG_PPC_TM_DSCR:
1612 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1613 		break;
1614 	case KVM_REG_PPC_TM_TAR:
1615 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1616 		break;
1617 #endif
1618 	default:
1619 		r = -EINVAL;
1620 		break;
1621 	}
1622 
1623 	return r;
1624 }
1625 
1626 static void kvmppc_set_lpcr_pr(struct kvm_vcpu *vcpu, u64 new_lpcr)
1627 {
1628 	if (new_lpcr & LPCR_ILE)
1629 		vcpu->arch.intr_msr |= MSR_LE;
1630 	else
1631 		vcpu->arch.intr_msr &= ~MSR_LE;
1632 }
1633 
1634 static int kvmppc_set_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1635 				 union kvmppc_one_reg *val)
1636 {
1637 	int r = 0;
1638 
1639 	switch (id) {
1640 	case KVM_REG_PPC_HIOR:
1641 		to_book3s(vcpu)->hior = set_reg_val(id, *val);
1642 		to_book3s(vcpu)->hior_explicit = true;
1643 		break;
1644 	case KVM_REG_PPC_VTB:
1645 		to_book3s(vcpu)->vtb = set_reg_val(id, *val);
1646 		break;
1647 	case KVM_REG_PPC_LPCR:
1648 	case KVM_REG_PPC_LPCR_64:
1649 		kvmppc_set_lpcr_pr(vcpu, set_reg_val(id, *val));
1650 		break;
1651 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1652 	case KVM_REG_PPC_TFHAR:
1653 		vcpu->arch.tfhar = set_reg_val(id, *val);
1654 		break;
1655 	case KVM_REG_PPC_TFIAR:
1656 		vcpu->arch.tfiar = set_reg_val(id, *val);
1657 		break;
1658 	case KVM_REG_PPC_TEXASR:
1659 		vcpu->arch.texasr = set_reg_val(id, *val);
1660 		break;
1661 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1662 		vcpu->arch.gpr_tm[id - KVM_REG_PPC_TM_GPR0] =
1663 			set_reg_val(id, *val);
1664 		break;
1665 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1666 	{
1667 		int i, j;
1668 
1669 		i = id - KVM_REG_PPC_TM_VSR0;
1670 		if (i < 32)
1671 			for (j = 0; j < TS_FPRWIDTH; j++)
1672 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1673 		else
1674 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1675 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1676 			else
1677 				r = -ENXIO;
1678 		break;
1679 	}
1680 	case KVM_REG_PPC_TM_CR:
1681 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1682 		break;
1683 	case KVM_REG_PPC_TM_XER:
1684 		vcpu->arch.xer_tm = set_reg_val(id, *val);
1685 		break;
1686 	case KVM_REG_PPC_TM_LR:
1687 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1688 		break;
1689 	case KVM_REG_PPC_TM_CTR:
1690 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1691 		break;
1692 	case KVM_REG_PPC_TM_FPSCR:
1693 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1694 		break;
1695 	case KVM_REG_PPC_TM_AMR:
1696 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1697 		break;
1698 	case KVM_REG_PPC_TM_PPR:
1699 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1700 		break;
1701 	case KVM_REG_PPC_TM_VRSAVE:
1702 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1703 		break;
1704 	case KVM_REG_PPC_TM_VSCR:
1705 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1706 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1707 		else
1708 			r = -ENXIO;
1709 		break;
1710 	case KVM_REG_PPC_TM_DSCR:
1711 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1712 		break;
1713 	case KVM_REG_PPC_TM_TAR:
1714 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1715 		break;
1716 #endif
1717 	default:
1718 		r = -EINVAL;
1719 		break;
1720 	}
1721 
1722 	return r;
1723 }
1724 
1725 static int kvmppc_core_vcpu_create_pr(struct kvm_vcpu *vcpu)
1726 {
1727 	struct kvmppc_vcpu_book3s *vcpu_book3s;
1728 	unsigned long p;
1729 	int err;
1730 
1731 	err = -ENOMEM;
1732 
1733 	vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1734 	if (!vcpu_book3s)
1735 		goto out;
1736 	vcpu->arch.book3s = vcpu_book3s;
1737 
1738 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1739 	vcpu->arch.shadow_vcpu =
1740 		kzalloc(sizeof(*vcpu->arch.shadow_vcpu), GFP_KERNEL);
1741 	if (!vcpu->arch.shadow_vcpu)
1742 		goto free_vcpu3s;
1743 #endif
1744 
1745 	p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1746 	if (!p)
1747 		goto free_shadow_vcpu;
1748 	vcpu->arch.shared = (void *)p;
1749 #ifdef CONFIG_PPC_BOOK3S_64
1750 	/* Always start the shared struct in native endian mode */
1751 #ifdef __BIG_ENDIAN__
1752         vcpu->arch.shared_big_endian = true;
1753 #else
1754         vcpu->arch.shared_big_endian = false;
1755 #endif
1756 
1757 	/*
1758 	 * Default to the same as the host if we're on sufficiently
1759 	 * recent machine that we have 1TB segments;
1760 	 * otherwise default to PPC970FX.
1761 	 */
1762 	vcpu->arch.pvr = 0x3C0301;
1763 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1764 		vcpu->arch.pvr = mfspr(SPRN_PVR);
1765 	vcpu->arch.intr_msr = MSR_SF;
1766 #else
1767 	/* default to book3s_32 (750) */
1768 	vcpu->arch.pvr = 0x84202;
1769 	vcpu->arch.intr_msr = 0;
1770 #endif
1771 	kvmppc_set_pvr_pr(vcpu, vcpu->arch.pvr);
1772 	vcpu->arch.slb_nr = 64;
1773 
1774 	vcpu->arch.shadow_msr = MSR_USER64 & ~MSR_LE;
1775 
1776 	err = kvmppc_mmu_init_pr(vcpu);
1777 	if (err < 0)
1778 		goto free_shared_page;
1779 
1780 	return 0;
1781 
1782 free_shared_page:
1783 	free_page((unsigned long)vcpu->arch.shared);
1784 free_shadow_vcpu:
1785 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1786 	kfree(vcpu->arch.shadow_vcpu);
1787 free_vcpu3s:
1788 #endif
1789 	vfree(vcpu_book3s);
1790 out:
1791 	return err;
1792 }
1793 
1794 static void kvmppc_core_vcpu_free_pr(struct kvm_vcpu *vcpu)
1795 {
1796 	struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1797 
1798 	kvmppc_mmu_destroy_pr(vcpu);
1799 	free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1800 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1801 	kfree(vcpu->arch.shadow_vcpu);
1802 #endif
1803 	vfree(vcpu_book3s);
1804 }
1805 
1806 static int kvmppc_vcpu_run_pr(struct kvm_vcpu *vcpu)
1807 {
1808 	int ret;
1809 
1810 	/* Check if we can run the vcpu at all */
1811 	if (!vcpu->arch.sane) {
1812 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1813 		ret = -EINVAL;
1814 		goto out;
1815 	}
1816 
1817 	kvmppc_setup_debug(vcpu);
1818 
1819 	/*
1820 	 * Interrupts could be timers for the guest which we have to inject
1821 	 * again, so let's postpone them until we're in the guest and if we
1822 	 * really did time things so badly, then we just exit again due to
1823 	 * a host external interrupt.
1824 	 */
1825 	ret = kvmppc_prepare_to_enter(vcpu);
1826 	if (ret <= 0)
1827 		goto out;
1828 	/* interrupts now hard-disabled */
1829 
1830 	/* Save FPU, Altivec and VSX state */
1831 	giveup_all(current);
1832 
1833 	/* Preload FPU if it's enabled */
1834 	if (kvmppc_get_msr(vcpu) & MSR_FP)
1835 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1836 
1837 	kvmppc_fix_ee_before_entry();
1838 
1839 	ret = __kvmppc_vcpu_run(vcpu);
1840 
1841 	kvmppc_clear_debug(vcpu);
1842 
1843 	/* No need for guest_exit. It's done in handle_exit.
1844 	   We also get here with interrupts enabled. */
1845 
1846 	/* Make sure we save the guest FPU/Altivec/VSX state */
1847 	kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1848 
1849 	/* Make sure we save the guest TAR/EBB/DSCR state */
1850 	kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
1851 
1852 	srr_regs_clobbered();
1853 out:
1854 	vcpu->mode = OUTSIDE_GUEST_MODE;
1855 	return ret;
1856 }
1857 
1858 /*
1859  * Get (and clear) the dirty memory log for a memory slot.
1860  */
1861 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm *kvm,
1862 					 struct kvm_dirty_log *log)
1863 {
1864 	struct kvm_memory_slot *memslot;
1865 	struct kvm_vcpu *vcpu;
1866 	ulong ga, ga_end;
1867 	int is_dirty = 0;
1868 	int r;
1869 	unsigned long n;
1870 
1871 	mutex_lock(&kvm->slots_lock);
1872 
1873 	r = kvm_get_dirty_log(kvm, log, &is_dirty, &memslot);
1874 	if (r)
1875 		goto out;
1876 
1877 	/* If nothing is dirty, don't bother messing with page tables. */
1878 	if (is_dirty) {
1879 		ga = memslot->base_gfn << PAGE_SHIFT;
1880 		ga_end = ga + (memslot->npages << PAGE_SHIFT);
1881 
1882 		kvm_for_each_vcpu(n, vcpu, kvm)
1883 			kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1884 
1885 		n = kvm_dirty_bitmap_bytes(memslot);
1886 		memset(memslot->dirty_bitmap, 0, n);
1887 	}
1888 
1889 	r = 0;
1890 out:
1891 	mutex_unlock(&kvm->slots_lock);
1892 	return r;
1893 }
1894 
1895 static void kvmppc_core_flush_memslot_pr(struct kvm *kvm,
1896 					 struct kvm_memory_slot *memslot)
1897 {
1898 	return;
1899 }
1900 
1901 static int kvmppc_core_prepare_memory_region_pr(struct kvm *kvm,
1902 					struct kvm_memory_slot *memslot,
1903 					const struct kvm_userspace_memory_region *mem,
1904 					enum kvm_mr_change change)
1905 {
1906 	return 0;
1907 }
1908 
1909 static void kvmppc_core_commit_memory_region_pr(struct kvm *kvm,
1910 				const struct kvm_userspace_memory_region *mem,
1911 				const struct kvm_memory_slot *old,
1912 				const struct kvm_memory_slot *new,
1913 				enum kvm_mr_change change)
1914 {
1915 	return;
1916 }
1917 
1918 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot *slot)
1919 {
1920 	return;
1921 }
1922 
1923 #ifdef CONFIG_PPC64
1924 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1925 					 struct kvm_ppc_smmu_info *info)
1926 {
1927 	long int i;
1928 	struct kvm_vcpu *vcpu;
1929 
1930 	info->flags = 0;
1931 
1932 	/* SLB is always 64 entries */
1933 	info->slb_size = 64;
1934 
1935 	/* Standard 4k base page size segment */
1936 	info->sps[0].page_shift = 12;
1937 	info->sps[0].slb_enc = 0;
1938 	info->sps[0].enc[0].page_shift = 12;
1939 	info->sps[0].enc[0].pte_enc = 0;
1940 
1941 	/*
1942 	 * 64k large page size.
1943 	 * We only want to put this in if the CPUs we're emulating
1944 	 * support it, but unfortunately we don't have a vcpu easily
1945 	 * to hand here to test.  Just pick the first vcpu, and if
1946 	 * that doesn't exist yet, report the minimum capability,
1947 	 * i.e., no 64k pages.
1948 	 * 1T segment support goes along with 64k pages.
1949 	 */
1950 	i = 1;
1951 	vcpu = kvm_get_vcpu(kvm, 0);
1952 	if (vcpu && (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
1953 		info->flags = KVM_PPC_1T_SEGMENTS;
1954 		info->sps[i].page_shift = 16;
1955 		info->sps[i].slb_enc = SLB_VSID_L | SLB_VSID_LP_01;
1956 		info->sps[i].enc[0].page_shift = 16;
1957 		info->sps[i].enc[0].pte_enc = 1;
1958 		++i;
1959 	}
1960 
1961 	/* Standard 16M large page size segment */
1962 	info->sps[i].page_shift = 24;
1963 	info->sps[i].slb_enc = SLB_VSID_L;
1964 	info->sps[i].enc[0].page_shift = 24;
1965 	info->sps[i].enc[0].pte_enc = 0;
1966 
1967 	return 0;
1968 }
1969 
1970 static int kvm_configure_mmu_pr(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
1971 {
1972 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
1973 		return -ENODEV;
1974 	/* Require flags and process table base and size to all be zero. */
1975 	if (cfg->flags || cfg->process_table)
1976 		return -EINVAL;
1977 	return 0;
1978 }
1979 
1980 #else
1981 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1982 					 struct kvm_ppc_smmu_info *info)
1983 {
1984 	/* We should not get called */
1985 	BUG();
1986 	return 0;
1987 }
1988 #endif /* CONFIG_PPC64 */
1989 
1990 static unsigned int kvm_global_user_count = 0;
1991 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1992 
1993 static int kvmppc_core_init_vm_pr(struct kvm *kvm)
1994 {
1995 	mutex_init(&kvm->arch.hpt_mutex);
1996 
1997 #ifdef CONFIG_PPC_BOOK3S_64
1998 	/* Start out with the default set of hcalls enabled */
1999 	kvmppc_pr_init_default_hcalls(kvm);
2000 #endif
2001 
2002 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
2003 		spin_lock(&kvm_global_user_count_lock);
2004 		if (++kvm_global_user_count == 1)
2005 			pseries_disable_reloc_on_exc();
2006 		spin_unlock(&kvm_global_user_count_lock);
2007 	}
2008 	return 0;
2009 }
2010 
2011 static void kvmppc_core_destroy_vm_pr(struct kvm *kvm)
2012 {
2013 #ifdef CONFIG_PPC64
2014 	WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
2015 #endif
2016 
2017 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
2018 		spin_lock(&kvm_global_user_count_lock);
2019 		BUG_ON(kvm_global_user_count == 0);
2020 		if (--kvm_global_user_count == 0)
2021 			pseries_enable_reloc_on_exc();
2022 		spin_unlock(&kvm_global_user_count_lock);
2023 	}
2024 }
2025 
2026 static int kvmppc_core_check_processor_compat_pr(void)
2027 {
2028 	/*
2029 	 * PR KVM can work on POWER9 inside a guest partition
2030 	 * running in HPT mode.  It can't work if we are using
2031 	 * radix translation (because radix provides no way for
2032 	 * a process to have unique translations in quadrant 3).
2033 	 */
2034 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
2035 		return -EIO;
2036 	return 0;
2037 }
2038 
2039 static long kvm_arch_vm_ioctl_pr(struct file *filp,
2040 				 unsigned int ioctl, unsigned long arg)
2041 {
2042 	return -ENOTTY;
2043 }
2044 
2045 static struct kvmppc_ops kvm_ops_pr = {
2046 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_pr,
2047 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_pr,
2048 	.get_one_reg = kvmppc_get_one_reg_pr,
2049 	.set_one_reg = kvmppc_set_one_reg_pr,
2050 	.vcpu_load   = kvmppc_core_vcpu_load_pr,
2051 	.vcpu_put    = kvmppc_core_vcpu_put_pr,
2052 	.inject_interrupt = kvmppc_inject_interrupt_pr,
2053 	.set_msr     = kvmppc_set_msr_pr,
2054 	.vcpu_run    = kvmppc_vcpu_run_pr,
2055 	.vcpu_create = kvmppc_core_vcpu_create_pr,
2056 	.vcpu_free   = kvmppc_core_vcpu_free_pr,
2057 	.check_requests = kvmppc_core_check_requests_pr,
2058 	.get_dirty_log = kvm_vm_ioctl_get_dirty_log_pr,
2059 	.flush_memslot = kvmppc_core_flush_memslot_pr,
2060 	.prepare_memory_region = kvmppc_core_prepare_memory_region_pr,
2061 	.commit_memory_region = kvmppc_core_commit_memory_region_pr,
2062 	.unmap_gfn_range = kvm_unmap_gfn_range_pr,
2063 	.age_gfn  = kvm_age_gfn_pr,
2064 	.test_age_gfn = kvm_test_age_gfn_pr,
2065 	.set_spte_gfn = kvm_set_spte_gfn_pr,
2066 	.free_memslot = kvmppc_core_free_memslot_pr,
2067 	.init_vm = kvmppc_core_init_vm_pr,
2068 	.destroy_vm = kvmppc_core_destroy_vm_pr,
2069 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_pr,
2070 	.emulate_op = kvmppc_core_emulate_op_pr,
2071 	.emulate_mtspr = kvmppc_core_emulate_mtspr_pr,
2072 	.emulate_mfspr = kvmppc_core_emulate_mfspr_pr,
2073 	.fast_vcpu_kick = kvm_vcpu_kick,
2074 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_pr,
2075 #ifdef CONFIG_PPC_BOOK3S_64
2076 	.hcall_implemented = kvmppc_hcall_impl_pr,
2077 	.configure_mmu = kvm_configure_mmu_pr,
2078 #endif
2079 	.giveup_ext = kvmppc_giveup_ext,
2080 };
2081 
2082 
2083 int kvmppc_book3s_init_pr(void)
2084 {
2085 	int r;
2086 
2087 	r = kvmppc_core_check_processor_compat_pr();
2088 	if (r < 0)
2089 		return r;
2090 
2091 	kvm_ops_pr.owner = THIS_MODULE;
2092 	kvmppc_pr_ops = &kvm_ops_pr;
2093 
2094 	r = kvmppc_mmu_hpte_sysinit();
2095 	return r;
2096 }
2097 
2098 void kvmppc_book3s_exit_pr(void)
2099 {
2100 	kvmppc_pr_ops = NULL;
2101 	kvmppc_mmu_hpte_sysexit();
2102 }
2103 
2104 /*
2105  * We only support separate modules for book3s 64
2106  */
2107 #ifdef CONFIG_PPC_BOOK3S_64
2108 
2109 module_init(kvmppc_book3s_init_pr);
2110 module_exit(kvmppc_book3s_exit_pr);
2111 
2112 MODULE_LICENSE("GPL");
2113 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2114 MODULE_ALIAS("devname:kvm");
2115 #endif
2116