xref: /openbmc/linux/arch/powerpc/kvm/book3s_pr.c (revision 5b394b2d)
1 /*
2  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
3  *
4  * Authors:
5  *    Alexander Graf <agraf@suse.de>
6  *    Kevin Wolf <mail@kevin-wolf.de>
7  *    Paul Mackerras <paulus@samba.org>
8  *
9  * Description:
10  * Functions relating to running KVM on Book 3S processors where
11  * we don't have access to hypervisor mode, and we run the guest
12  * in problem state (user mode).
13  *
14  * This file is derived from arch/powerpc/kvm/44x.c,
15  * by Hollis Blanchard <hollisb@us.ibm.com>.
16  *
17  * This program is free software; you can redistribute it and/or modify
18  * it under the terms of the GNU General Public License, version 2, as
19  * published by the Free Software Foundation.
20  */
21 
22 #include <linux/kvm_host.h>
23 #include <linux/export.h>
24 #include <linux/err.h>
25 #include <linux/slab.h>
26 
27 #include <asm/reg.h>
28 #include <asm/cputable.h>
29 #include <asm/cacheflush.h>
30 #include <linux/uaccess.h>
31 #include <asm/io.h>
32 #include <asm/kvm_ppc.h>
33 #include <asm/kvm_book3s.h>
34 #include <asm/mmu_context.h>
35 #include <asm/switch_to.h>
36 #include <asm/firmware.h>
37 #include <asm/setup.h>
38 #include <linux/gfp.h>
39 #include <linux/sched.h>
40 #include <linux/vmalloc.h>
41 #include <linux/highmem.h>
42 #include <linux/module.h>
43 #include <linux/miscdevice.h>
44 #include <asm/asm-prototypes.h>
45 #include <asm/tm.h>
46 
47 #include "book3s.h"
48 
49 #define CREATE_TRACE_POINTS
50 #include "trace_pr.h"
51 
52 /* #define EXIT_DEBUG */
53 /* #define DEBUG_EXT */
54 
55 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
56 			     ulong msr);
57 #ifdef CONFIG_PPC_BOOK3S_64
58 static int kvmppc_handle_fac(struct kvm_vcpu *vcpu, ulong fac);
59 #endif
60 
61 /* Some compatibility defines */
62 #ifdef CONFIG_PPC_BOOK3S_32
63 #define MSR_USER32 MSR_USER
64 #define MSR_USER64 MSR_USER
65 #define HW_PAGE_SIZE PAGE_SIZE
66 #define HPTE_R_M   _PAGE_COHERENT
67 #endif
68 
69 static bool kvmppc_is_split_real(struct kvm_vcpu *vcpu)
70 {
71 	ulong msr = kvmppc_get_msr(vcpu);
72 	return (msr & (MSR_IR|MSR_DR)) == MSR_DR;
73 }
74 
75 static void kvmppc_fixup_split_real(struct kvm_vcpu *vcpu)
76 {
77 	ulong msr = kvmppc_get_msr(vcpu);
78 	ulong pc = kvmppc_get_pc(vcpu);
79 
80 	/* We are in DR only split real mode */
81 	if ((msr & (MSR_IR|MSR_DR)) != MSR_DR)
82 		return;
83 
84 	/* We have not fixed up the guest already */
85 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK)
86 		return;
87 
88 	/* The code is in fixupable address space */
89 	if (pc & SPLIT_HACK_MASK)
90 		return;
91 
92 	vcpu->arch.hflags |= BOOK3S_HFLAG_SPLIT_HACK;
93 	kvmppc_set_pc(vcpu, pc | SPLIT_HACK_OFFS);
94 }
95 
96 void kvmppc_unfixup_split_real(struct kvm_vcpu *vcpu);
97 
98 static void kvmppc_core_vcpu_load_pr(struct kvm_vcpu *vcpu, int cpu)
99 {
100 #ifdef CONFIG_PPC_BOOK3S_64
101 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
102 	memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb));
103 	svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max;
104 	svcpu->in_use = 0;
105 	svcpu_put(svcpu);
106 #endif
107 
108 	/* Disable AIL if supported */
109 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
110 	    cpu_has_feature(CPU_FTR_ARCH_207S))
111 		mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) & ~LPCR_AIL);
112 
113 	vcpu->cpu = smp_processor_id();
114 #ifdef CONFIG_PPC_BOOK3S_32
115 	current->thread.kvm_shadow_vcpu = vcpu->arch.shadow_vcpu;
116 #endif
117 
118 	if (kvmppc_is_split_real(vcpu))
119 		kvmppc_fixup_split_real(vcpu);
120 
121 	kvmppc_restore_tm_pr(vcpu);
122 }
123 
124 static void kvmppc_core_vcpu_put_pr(struct kvm_vcpu *vcpu)
125 {
126 #ifdef CONFIG_PPC_BOOK3S_64
127 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
128 	if (svcpu->in_use) {
129 		kvmppc_copy_from_svcpu(vcpu);
130 	}
131 	memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
132 	to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max;
133 	svcpu_put(svcpu);
134 #endif
135 
136 	if (kvmppc_is_split_real(vcpu))
137 		kvmppc_unfixup_split_real(vcpu);
138 
139 	kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
140 	kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
141 	kvmppc_save_tm_pr(vcpu);
142 
143 	/* Enable AIL if supported */
144 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
145 	    cpu_has_feature(CPU_FTR_ARCH_207S))
146 		mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_AIL_3);
147 
148 	vcpu->cpu = -1;
149 }
150 
151 /* Copy data needed by real-mode code from vcpu to shadow vcpu */
152 void kvmppc_copy_to_svcpu(struct kvm_vcpu *vcpu)
153 {
154 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
155 
156 	svcpu->gpr[0] = vcpu->arch.regs.gpr[0];
157 	svcpu->gpr[1] = vcpu->arch.regs.gpr[1];
158 	svcpu->gpr[2] = vcpu->arch.regs.gpr[2];
159 	svcpu->gpr[3] = vcpu->arch.regs.gpr[3];
160 	svcpu->gpr[4] = vcpu->arch.regs.gpr[4];
161 	svcpu->gpr[5] = vcpu->arch.regs.gpr[5];
162 	svcpu->gpr[6] = vcpu->arch.regs.gpr[6];
163 	svcpu->gpr[7] = vcpu->arch.regs.gpr[7];
164 	svcpu->gpr[8] = vcpu->arch.regs.gpr[8];
165 	svcpu->gpr[9] = vcpu->arch.regs.gpr[9];
166 	svcpu->gpr[10] = vcpu->arch.regs.gpr[10];
167 	svcpu->gpr[11] = vcpu->arch.regs.gpr[11];
168 	svcpu->gpr[12] = vcpu->arch.regs.gpr[12];
169 	svcpu->gpr[13] = vcpu->arch.regs.gpr[13];
170 	svcpu->cr  = vcpu->arch.cr;
171 	svcpu->xer = vcpu->arch.regs.xer;
172 	svcpu->ctr = vcpu->arch.regs.ctr;
173 	svcpu->lr  = vcpu->arch.regs.link;
174 	svcpu->pc  = vcpu->arch.regs.nip;
175 #ifdef CONFIG_PPC_BOOK3S_64
176 	svcpu->shadow_fscr = vcpu->arch.shadow_fscr;
177 #endif
178 	/*
179 	 * Now also save the current time base value. We use this
180 	 * to find the guest purr and spurr value.
181 	 */
182 	vcpu->arch.entry_tb = get_tb();
183 	vcpu->arch.entry_vtb = get_vtb();
184 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
185 		vcpu->arch.entry_ic = mfspr(SPRN_IC);
186 	svcpu->in_use = true;
187 
188 	svcpu_put(svcpu);
189 }
190 
191 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
192 {
193 	ulong guest_msr = kvmppc_get_msr(vcpu);
194 	ulong smsr = guest_msr;
195 
196 	/* Guest MSR values */
197 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
198 	smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_LE |
199 		MSR_TM | MSR_TS_MASK;
200 #else
201 	smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_LE;
202 #endif
203 	/* Process MSR values */
204 	smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
205 	/* External providers the guest reserved */
206 	smsr |= (guest_msr & vcpu->arch.guest_owned_ext);
207 	/* 64-bit Process MSR values */
208 #ifdef CONFIG_PPC_BOOK3S_64
209 	smsr |= MSR_ISF | MSR_HV;
210 #endif
211 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
212 	/*
213 	 * in guest privileged state, we want to fail all TM transactions.
214 	 * So disable MSR TM bit so that all tbegin. will be able to be
215 	 * trapped into host.
216 	 */
217 	if (!(guest_msr & MSR_PR))
218 		smsr &= ~MSR_TM;
219 #endif
220 	vcpu->arch.shadow_msr = smsr;
221 }
222 
223 /* Copy data touched by real-mode code from shadow vcpu back to vcpu */
224 void kvmppc_copy_from_svcpu(struct kvm_vcpu *vcpu)
225 {
226 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
227 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
228 	ulong old_msr;
229 #endif
230 
231 	/*
232 	 * Maybe we were already preempted and synced the svcpu from
233 	 * our preempt notifiers. Don't bother touching this svcpu then.
234 	 */
235 	if (!svcpu->in_use)
236 		goto out;
237 
238 	vcpu->arch.regs.gpr[0] = svcpu->gpr[0];
239 	vcpu->arch.regs.gpr[1] = svcpu->gpr[1];
240 	vcpu->arch.regs.gpr[2] = svcpu->gpr[2];
241 	vcpu->arch.regs.gpr[3] = svcpu->gpr[3];
242 	vcpu->arch.regs.gpr[4] = svcpu->gpr[4];
243 	vcpu->arch.regs.gpr[5] = svcpu->gpr[5];
244 	vcpu->arch.regs.gpr[6] = svcpu->gpr[6];
245 	vcpu->arch.regs.gpr[7] = svcpu->gpr[7];
246 	vcpu->arch.regs.gpr[8] = svcpu->gpr[8];
247 	vcpu->arch.regs.gpr[9] = svcpu->gpr[9];
248 	vcpu->arch.regs.gpr[10] = svcpu->gpr[10];
249 	vcpu->arch.regs.gpr[11] = svcpu->gpr[11];
250 	vcpu->arch.regs.gpr[12] = svcpu->gpr[12];
251 	vcpu->arch.regs.gpr[13] = svcpu->gpr[13];
252 	vcpu->arch.cr  = svcpu->cr;
253 	vcpu->arch.regs.xer = svcpu->xer;
254 	vcpu->arch.regs.ctr = svcpu->ctr;
255 	vcpu->arch.regs.link  = svcpu->lr;
256 	vcpu->arch.regs.nip  = svcpu->pc;
257 	vcpu->arch.shadow_srr1 = svcpu->shadow_srr1;
258 	vcpu->arch.fault_dar   = svcpu->fault_dar;
259 	vcpu->arch.fault_dsisr = svcpu->fault_dsisr;
260 	vcpu->arch.last_inst   = svcpu->last_inst;
261 #ifdef CONFIG_PPC_BOOK3S_64
262 	vcpu->arch.shadow_fscr = svcpu->shadow_fscr;
263 #endif
264 	/*
265 	 * Update purr and spurr using time base on exit.
266 	 */
267 	vcpu->arch.purr += get_tb() - vcpu->arch.entry_tb;
268 	vcpu->arch.spurr += get_tb() - vcpu->arch.entry_tb;
269 	to_book3s(vcpu)->vtb += get_vtb() - vcpu->arch.entry_vtb;
270 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
271 		vcpu->arch.ic += mfspr(SPRN_IC) - vcpu->arch.entry_ic;
272 
273 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
274 	/*
275 	 * Unlike other MSR bits, MSR[TS]bits can be changed at guest without
276 	 * notifying host:
277 	 *  modified by unprivileged instructions like "tbegin"/"tend"/
278 	 * "tresume"/"tsuspend" in PR KVM guest.
279 	 *
280 	 * It is necessary to sync here to calculate a correct shadow_msr.
281 	 *
282 	 * privileged guest's tbegin will be failed at present. So we
283 	 * only take care of problem state guest.
284 	 */
285 	old_msr = kvmppc_get_msr(vcpu);
286 	if (unlikely((old_msr & MSR_PR) &&
287 		(vcpu->arch.shadow_srr1 & (MSR_TS_MASK)) !=
288 				(old_msr & (MSR_TS_MASK)))) {
289 		old_msr &= ~(MSR_TS_MASK);
290 		old_msr |= (vcpu->arch.shadow_srr1 & (MSR_TS_MASK));
291 		kvmppc_set_msr_fast(vcpu, old_msr);
292 		kvmppc_recalc_shadow_msr(vcpu);
293 	}
294 #endif
295 
296 	svcpu->in_use = false;
297 
298 out:
299 	svcpu_put(svcpu);
300 }
301 
302 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
303 void kvmppc_save_tm_sprs(struct kvm_vcpu *vcpu)
304 {
305 	tm_enable();
306 	vcpu->arch.tfhar = mfspr(SPRN_TFHAR);
307 	vcpu->arch.texasr = mfspr(SPRN_TEXASR);
308 	vcpu->arch.tfiar = mfspr(SPRN_TFIAR);
309 	tm_disable();
310 }
311 
312 void kvmppc_restore_tm_sprs(struct kvm_vcpu *vcpu)
313 {
314 	tm_enable();
315 	mtspr(SPRN_TFHAR, vcpu->arch.tfhar);
316 	mtspr(SPRN_TEXASR, vcpu->arch.texasr);
317 	mtspr(SPRN_TFIAR, vcpu->arch.tfiar);
318 	tm_disable();
319 }
320 
321 /* loadup math bits which is enabled at kvmppc_get_msr() but not enabled at
322  * hardware.
323  */
324 static void kvmppc_handle_lost_math_exts(struct kvm_vcpu *vcpu)
325 {
326 	ulong exit_nr;
327 	ulong ext_diff = (kvmppc_get_msr(vcpu) & ~vcpu->arch.guest_owned_ext) &
328 		(MSR_FP | MSR_VEC | MSR_VSX);
329 
330 	if (!ext_diff)
331 		return;
332 
333 	if (ext_diff == MSR_FP)
334 		exit_nr = BOOK3S_INTERRUPT_FP_UNAVAIL;
335 	else if (ext_diff == MSR_VEC)
336 		exit_nr = BOOK3S_INTERRUPT_ALTIVEC;
337 	else
338 		exit_nr = BOOK3S_INTERRUPT_VSX;
339 
340 	kvmppc_handle_ext(vcpu, exit_nr, ext_diff);
341 }
342 
343 void kvmppc_save_tm_pr(struct kvm_vcpu *vcpu)
344 {
345 	if (!(MSR_TM_ACTIVE(kvmppc_get_msr(vcpu)))) {
346 		kvmppc_save_tm_sprs(vcpu);
347 		return;
348 	}
349 
350 	kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
351 	kvmppc_giveup_ext(vcpu, MSR_VSX);
352 
353 	preempt_disable();
354 	_kvmppc_save_tm_pr(vcpu, mfmsr());
355 	preempt_enable();
356 }
357 
358 void kvmppc_restore_tm_pr(struct kvm_vcpu *vcpu)
359 {
360 	if (!MSR_TM_ACTIVE(kvmppc_get_msr(vcpu))) {
361 		kvmppc_restore_tm_sprs(vcpu);
362 		if (kvmppc_get_msr(vcpu) & MSR_TM) {
363 			kvmppc_handle_lost_math_exts(vcpu);
364 			if (vcpu->arch.fscr & FSCR_TAR)
365 				kvmppc_handle_fac(vcpu, FSCR_TAR_LG);
366 		}
367 		return;
368 	}
369 
370 	preempt_disable();
371 	_kvmppc_restore_tm_pr(vcpu, kvmppc_get_msr(vcpu));
372 	preempt_enable();
373 
374 	if (kvmppc_get_msr(vcpu) & MSR_TM) {
375 		kvmppc_handle_lost_math_exts(vcpu);
376 		if (vcpu->arch.fscr & FSCR_TAR)
377 			kvmppc_handle_fac(vcpu, FSCR_TAR_LG);
378 	}
379 }
380 #endif
381 
382 static int kvmppc_core_check_requests_pr(struct kvm_vcpu *vcpu)
383 {
384 	int r = 1; /* Indicate we want to get back into the guest */
385 
386 	/* We misuse TLB_FLUSH to indicate that we want to clear
387 	   all shadow cache entries */
388 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
389 		kvmppc_mmu_pte_flush(vcpu, 0, 0);
390 
391 	return r;
392 }
393 
394 /************* MMU Notifiers *************/
395 static void do_kvm_unmap_hva(struct kvm *kvm, unsigned long start,
396 			     unsigned long end)
397 {
398 	long i;
399 	struct kvm_vcpu *vcpu;
400 	struct kvm_memslots *slots;
401 	struct kvm_memory_slot *memslot;
402 
403 	slots = kvm_memslots(kvm);
404 	kvm_for_each_memslot(memslot, slots) {
405 		unsigned long hva_start, hva_end;
406 		gfn_t gfn, gfn_end;
407 
408 		hva_start = max(start, memslot->userspace_addr);
409 		hva_end = min(end, memslot->userspace_addr +
410 					(memslot->npages << PAGE_SHIFT));
411 		if (hva_start >= hva_end)
412 			continue;
413 		/*
414 		 * {gfn(page) | page intersects with [hva_start, hva_end)} =
415 		 * {gfn, gfn+1, ..., gfn_end-1}.
416 		 */
417 		gfn = hva_to_gfn_memslot(hva_start, memslot);
418 		gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
419 		kvm_for_each_vcpu(i, vcpu, kvm)
420 			kvmppc_mmu_pte_pflush(vcpu, gfn << PAGE_SHIFT,
421 					      gfn_end << PAGE_SHIFT);
422 	}
423 }
424 
425 static int kvm_unmap_hva_range_pr(struct kvm *kvm, unsigned long start,
426 				  unsigned long end)
427 {
428 	do_kvm_unmap_hva(kvm, start, end);
429 
430 	return 0;
431 }
432 
433 static int kvm_age_hva_pr(struct kvm *kvm, unsigned long start,
434 			  unsigned long end)
435 {
436 	/* XXX could be more clever ;) */
437 	return 0;
438 }
439 
440 static int kvm_test_age_hva_pr(struct kvm *kvm, unsigned long hva)
441 {
442 	/* XXX could be more clever ;) */
443 	return 0;
444 }
445 
446 static void kvm_set_spte_hva_pr(struct kvm *kvm, unsigned long hva, pte_t pte)
447 {
448 	/* The page will get remapped properly on its next fault */
449 	do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
450 }
451 
452 /*****************************************/
453 
454 static void kvmppc_set_msr_pr(struct kvm_vcpu *vcpu, u64 msr)
455 {
456 	ulong old_msr;
457 
458 	/* For PAPR guest, make sure MSR reflects guest mode */
459 	if (vcpu->arch.papr_enabled)
460 		msr = (msr & ~MSR_HV) | MSR_ME;
461 
462 #ifdef EXIT_DEBUG
463 	printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
464 #endif
465 
466 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
467 	/* We should never target guest MSR to TS=10 && PR=0,
468 	 * since we always fail transaction for guest privilege
469 	 * state.
470 	 */
471 	if (!(msr & MSR_PR) && MSR_TM_TRANSACTIONAL(msr))
472 		kvmppc_emulate_tabort(vcpu,
473 			TM_CAUSE_KVM_FAC_UNAV | TM_CAUSE_PERSISTENT);
474 #endif
475 
476 	old_msr = kvmppc_get_msr(vcpu);
477 	msr &= to_book3s(vcpu)->msr_mask;
478 	kvmppc_set_msr_fast(vcpu, msr);
479 	kvmppc_recalc_shadow_msr(vcpu);
480 
481 	if (msr & MSR_POW) {
482 		if (!vcpu->arch.pending_exceptions) {
483 			kvm_vcpu_block(vcpu);
484 			kvm_clear_request(KVM_REQ_UNHALT, vcpu);
485 			vcpu->stat.halt_wakeup++;
486 
487 			/* Unset POW bit after we woke up */
488 			msr &= ~MSR_POW;
489 			kvmppc_set_msr_fast(vcpu, msr);
490 		}
491 	}
492 
493 	if (kvmppc_is_split_real(vcpu))
494 		kvmppc_fixup_split_real(vcpu);
495 	else
496 		kvmppc_unfixup_split_real(vcpu);
497 
498 	if ((kvmppc_get_msr(vcpu) & (MSR_PR|MSR_IR|MSR_DR)) !=
499 		   (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
500 		kvmppc_mmu_flush_segments(vcpu);
501 		kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
502 
503 		/* Preload magic page segment when in kernel mode */
504 		if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
505 			struct kvm_vcpu_arch *a = &vcpu->arch;
506 
507 			if (msr & MSR_DR)
508 				kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
509 			else
510 				kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
511 		}
512 	}
513 
514 	/*
515 	 * When switching from 32 to 64-bit, we may have a stale 32-bit
516 	 * magic page around, we need to flush it. Typically 32-bit magic
517 	 * page will be instantiated when calling into RTAS. Note: We
518 	 * assume that such transition only happens while in kernel mode,
519 	 * ie, we never transition from user 32-bit to kernel 64-bit with
520 	 * a 32-bit magic page around.
521 	 */
522 	if (vcpu->arch.magic_page_pa &&
523 	    !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
524 		/* going from RTAS to normal kernel code */
525 		kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
526 				     ~0xFFFUL);
527 	}
528 
529 	/* Preload FPU if it's enabled */
530 	if (kvmppc_get_msr(vcpu) & MSR_FP)
531 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
532 
533 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
534 	if (kvmppc_get_msr(vcpu) & MSR_TM)
535 		kvmppc_handle_lost_math_exts(vcpu);
536 #endif
537 }
538 
539 void kvmppc_set_pvr_pr(struct kvm_vcpu *vcpu, u32 pvr)
540 {
541 	u32 host_pvr;
542 
543 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
544 	vcpu->arch.pvr = pvr;
545 #ifdef CONFIG_PPC_BOOK3S_64
546 	if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
547 		kvmppc_mmu_book3s_64_init(vcpu);
548 		if (!to_book3s(vcpu)->hior_explicit)
549 			to_book3s(vcpu)->hior = 0xfff00000;
550 		to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
551 		vcpu->arch.cpu_type = KVM_CPU_3S_64;
552 	} else
553 #endif
554 	{
555 		kvmppc_mmu_book3s_32_init(vcpu);
556 		if (!to_book3s(vcpu)->hior_explicit)
557 			to_book3s(vcpu)->hior = 0;
558 		to_book3s(vcpu)->msr_mask = 0xffffffffULL;
559 		vcpu->arch.cpu_type = KVM_CPU_3S_32;
560 	}
561 
562 	kvmppc_sanity_check(vcpu);
563 
564 	/* If we are in hypervisor level on 970, we can tell the CPU to
565 	 * treat DCBZ as 32 bytes store */
566 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
567 	if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
568 	    !strcmp(cur_cpu_spec->platform, "ppc970"))
569 		vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
570 
571 	/* Cell performs badly if MSR_FEx are set. So let's hope nobody
572 	   really needs them in a VM on Cell and force disable them. */
573 	if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
574 		to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
575 
576 	/*
577 	 * If they're asking for POWER6 or later, set the flag
578 	 * indicating that we can do multiple large page sizes
579 	 * and 1TB segments.
580 	 * Also set the flag that indicates that tlbie has the large
581 	 * page bit in the RB operand instead of the instruction.
582 	 */
583 	switch (PVR_VER(pvr)) {
584 	case PVR_POWER6:
585 	case PVR_POWER7:
586 	case PVR_POWER7p:
587 	case PVR_POWER8:
588 	case PVR_POWER8E:
589 	case PVR_POWER8NVL:
590 		vcpu->arch.hflags |= BOOK3S_HFLAG_MULTI_PGSIZE |
591 			BOOK3S_HFLAG_NEW_TLBIE;
592 		break;
593 	}
594 
595 #ifdef CONFIG_PPC_BOOK3S_32
596 	/* 32 bit Book3S always has 32 byte dcbz */
597 	vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
598 #endif
599 
600 	/* On some CPUs we can execute paired single operations natively */
601 	asm ( "mfpvr %0" : "=r"(host_pvr));
602 	switch (host_pvr) {
603 	case 0x00080200:	/* lonestar 2.0 */
604 	case 0x00088202:	/* lonestar 2.2 */
605 	case 0x70000100:	/* gekko 1.0 */
606 	case 0x00080100:	/* gekko 2.0 */
607 	case 0x00083203:	/* gekko 2.3a */
608 	case 0x00083213:	/* gekko 2.3b */
609 	case 0x00083204:	/* gekko 2.4 */
610 	case 0x00083214:	/* gekko 2.4e (8SE) - retail HW2 */
611 	case 0x00087200:	/* broadway */
612 		vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
613 		/* Enable HID2.PSE - in case we need it later */
614 		mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
615 	}
616 }
617 
618 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
619  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
620  * emulate 32 bytes dcbz length.
621  *
622  * The Book3s_64 inventors also realized this case and implemented a special bit
623  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
624  *
625  * My approach here is to patch the dcbz instruction on executing pages.
626  */
627 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
628 {
629 	struct page *hpage;
630 	u64 hpage_offset;
631 	u32 *page;
632 	int i;
633 
634 	hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
635 	if (is_error_page(hpage))
636 		return;
637 
638 	hpage_offset = pte->raddr & ~PAGE_MASK;
639 	hpage_offset &= ~0xFFFULL;
640 	hpage_offset /= 4;
641 
642 	get_page(hpage);
643 	page = kmap_atomic(hpage);
644 
645 	/* patch dcbz into reserved instruction, so we trap */
646 	for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
647 		if ((be32_to_cpu(page[i]) & 0xff0007ff) == INS_DCBZ)
648 			page[i] &= cpu_to_be32(0xfffffff7);
649 
650 	kunmap_atomic(page);
651 	put_page(hpage);
652 }
653 
654 static bool kvmppc_visible_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
655 {
656 	ulong mp_pa = vcpu->arch.magic_page_pa;
657 
658 	if (!(kvmppc_get_msr(vcpu) & MSR_SF))
659 		mp_pa = (uint32_t)mp_pa;
660 
661 	gpa &= ~0xFFFULL;
662 	if (unlikely(mp_pa) && unlikely((mp_pa & KVM_PAM) == (gpa & KVM_PAM))) {
663 		return true;
664 	}
665 
666 	return kvm_is_visible_gfn(vcpu->kvm, gpa >> PAGE_SHIFT);
667 }
668 
669 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
670 			    ulong eaddr, int vec)
671 {
672 	bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
673 	bool iswrite = false;
674 	int r = RESUME_GUEST;
675 	int relocated;
676 	int page_found = 0;
677 	struct kvmppc_pte pte = { 0 };
678 	bool dr = (kvmppc_get_msr(vcpu) & MSR_DR) ? true : false;
679 	bool ir = (kvmppc_get_msr(vcpu) & MSR_IR) ? true : false;
680 	u64 vsid;
681 
682 	relocated = data ? dr : ir;
683 	if (data && (vcpu->arch.fault_dsisr & DSISR_ISSTORE))
684 		iswrite = true;
685 
686 	/* Resolve real address if translation turned on */
687 	if (relocated) {
688 		page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data, iswrite);
689 	} else {
690 		pte.may_execute = true;
691 		pte.may_read = true;
692 		pte.may_write = true;
693 		pte.raddr = eaddr & KVM_PAM;
694 		pte.eaddr = eaddr;
695 		pte.vpage = eaddr >> 12;
696 		pte.page_size = MMU_PAGE_64K;
697 		pte.wimg = HPTE_R_M;
698 	}
699 
700 	switch (kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) {
701 	case 0:
702 		pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
703 		break;
704 	case MSR_DR:
705 		if (!data &&
706 		    (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK) &&
707 		    ((pte.raddr & SPLIT_HACK_MASK) == SPLIT_HACK_OFFS))
708 			pte.raddr &= ~SPLIT_HACK_MASK;
709 		/* fall through */
710 	case MSR_IR:
711 		vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
712 
713 		if ((kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) == MSR_DR)
714 			pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
715 		else
716 			pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
717 		pte.vpage |= vsid;
718 
719 		if (vsid == -1)
720 			page_found = -EINVAL;
721 		break;
722 	}
723 
724 	if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
725 	   (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
726 		/*
727 		 * If we do the dcbz hack, we have to NX on every execution,
728 		 * so we can patch the executing code. This renders our guest
729 		 * NX-less.
730 		 */
731 		pte.may_execute = !data;
732 	}
733 
734 	if (page_found == -ENOENT || page_found == -EPERM) {
735 		/* Page not found in guest PTE entries, or protection fault */
736 		u64 flags;
737 
738 		if (page_found == -EPERM)
739 			flags = DSISR_PROTFAULT;
740 		else
741 			flags = DSISR_NOHPTE;
742 		if (data) {
743 			flags |= vcpu->arch.fault_dsisr & DSISR_ISSTORE;
744 			kvmppc_core_queue_data_storage(vcpu, eaddr, flags);
745 		} else {
746 			kvmppc_core_queue_inst_storage(vcpu, flags);
747 		}
748 	} else if (page_found == -EINVAL) {
749 		/* Page not found in guest SLB */
750 		kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
751 		kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
752 	} else if (kvmppc_visible_gpa(vcpu, pte.raddr)) {
753 		if (data && !(vcpu->arch.fault_dsisr & DSISR_NOHPTE)) {
754 			/*
755 			 * There is already a host HPTE there, presumably
756 			 * a read-only one for a page the guest thinks
757 			 * is writable, so get rid of it first.
758 			 */
759 			kvmppc_mmu_unmap_page(vcpu, &pte);
760 		}
761 		/* The guest's PTE is not mapped yet. Map on the host */
762 		if (kvmppc_mmu_map_page(vcpu, &pte, iswrite) == -EIO) {
763 			/* Exit KVM if mapping failed */
764 			run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
765 			return RESUME_HOST;
766 		}
767 		if (data)
768 			vcpu->stat.sp_storage++;
769 		else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
770 			 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
771 			kvmppc_patch_dcbz(vcpu, &pte);
772 	} else {
773 		/* MMIO */
774 		vcpu->stat.mmio_exits++;
775 		vcpu->arch.paddr_accessed = pte.raddr;
776 		vcpu->arch.vaddr_accessed = pte.eaddr;
777 		r = kvmppc_emulate_mmio(run, vcpu);
778 		if ( r == RESUME_HOST_NV )
779 			r = RESUME_HOST;
780 	}
781 
782 	return r;
783 }
784 
785 /* Give up external provider (FPU, Altivec, VSX) */
786 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
787 {
788 	struct thread_struct *t = &current->thread;
789 
790 	/*
791 	 * VSX instructions can access FP and vector registers, so if
792 	 * we are giving up VSX, make sure we give up FP and VMX as well.
793 	 */
794 	if (msr & MSR_VSX)
795 		msr |= MSR_FP | MSR_VEC;
796 
797 	msr &= vcpu->arch.guest_owned_ext;
798 	if (!msr)
799 		return;
800 
801 #ifdef DEBUG_EXT
802 	printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
803 #endif
804 
805 	if (msr & MSR_FP) {
806 		/*
807 		 * Note that on CPUs with VSX, giveup_fpu stores
808 		 * both the traditional FP registers and the added VSX
809 		 * registers into thread.fp_state.fpr[].
810 		 */
811 		if (t->regs->msr & MSR_FP)
812 			giveup_fpu(current);
813 		t->fp_save_area = NULL;
814 	}
815 
816 #ifdef CONFIG_ALTIVEC
817 	if (msr & MSR_VEC) {
818 		if (current->thread.regs->msr & MSR_VEC)
819 			giveup_altivec(current);
820 		t->vr_save_area = NULL;
821 	}
822 #endif
823 
824 	vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
825 	kvmppc_recalc_shadow_msr(vcpu);
826 }
827 
828 /* Give up facility (TAR / EBB / DSCR) */
829 void kvmppc_giveup_fac(struct kvm_vcpu *vcpu, ulong fac)
830 {
831 #ifdef CONFIG_PPC_BOOK3S_64
832 	if (!(vcpu->arch.shadow_fscr & (1ULL << fac))) {
833 		/* Facility not available to the guest, ignore giveup request*/
834 		return;
835 	}
836 
837 	switch (fac) {
838 	case FSCR_TAR_LG:
839 		vcpu->arch.tar = mfspr(SPRN_TAR);
840 		mtspr(SPRN_TAR, current->thread.tar);
841 		vcpu->arch.shadow_fscr &= ~FSCR_TAR;
842 		break;
843 	}
844 #endif
845 }
846 
847 /* Handle external providers (FPU, Altivec, VSX) */
848 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
849 			     ulong msr)
850 {
851 	struct thread_struct *t = &current->thread;
852 
853 	/* When we have paired singles, we emulate in software */
854 	if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
855 		return RESUME_GUEST;
856 
857 	if (!(kvmppc_get_msr(vcpu) & msr)) {
858 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
859 		return RESUME_GUEST;
860 	}
861 
862 	if (msr == MSR_VSX) {
863 		/* No VSX?  Give an illegal instruction interrupt */
864 #ifdef CONFIG_VSX
865 		if (!cpu_has_feature(CPU_FTR_VSX))
866 #endif
867 		{
868 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
869 			return RESUME_GUEST;
870 		}
871 
872 		/*
873 		 * We have to load up all the FP and VMX registers before
874 		 * we can let the guest use VSX instructions.
875 		 */
876 		msr = MSR_FP | MSR_VEC | MSR_VSX;
877 	}
878 
879 	/* See if we already own all the ext(s) needed */
880 	msr &= ~vcpu->arch.guest_owned_ext;
881 	if (!msr)
882 		return RESUME_GUEST;
883 
884 #ifdef DEBUG_EXT
885 	printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
886 #endif
887 
888 	if (msr & MSR_FP) {
889 		preempt_disable();
890 		enable_kernel_fp();
891 		load_fp_state(&vcpu->arch.fp);
892 		disable_kernel_fp();
893 		t->fp_save_area = &vcpu->arch.fp;
894 		preempt_enable();
895 	}
896 
897 	if (msr & MSR_VEC) {
898 #ifdef CONFIG_ALTIVEC
899 		preempt_disable();
900 		enable_kernel_altivec();
901 		load_vr_state(&vcpu->arch.vr);
902 		disable_kernel_altivec();
903 		t->vr_save_area = &vcpu->arch.vr;
904 		preempt_enable();
905 #endif
906 	}
907 
908 	t->regs->msr |= msr;
909 	vcpu->arch.guest_owned_ext |= msr;
910 	kvmppc_recalc_shadow_msr(vcpu);
911 
912 	return RESUME_GUEST;
913 }
914 
915 /*
916  * Kernel code using FP or VMX could have flushed guest state to
917  * the thread_struct; if so, get it back now.
918  */
919 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
920 {
921 	unsigned long lost_ext;
922 
923 	lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
924 	if (!lost_ext)
925 		return;
926 
927 	if (lost_ext & MSR_FP) {
928 		preempt_disable();
929 		enable_kernel_fp();
930 		load_fp_state(&vcpu->arch.fp);
931 		disable_kernel_fp();
932 		preempt_enable();
933 	}
934 #ifdef CONFIG_ALTIVEC
935 	if (lost_ext & MSR_VEC) {
936 		preempt_disable();
937 		enable_kernel_altivec();
938 		load_vr_state(&vcpu->arch.vr);
939 		disable_kernel_altivec();
940 		preempt_enable();
941 	}
942 #endif
943 	current->thread.regs->msr |= lost_ext;
944 }
945 
946 #ifdef CONFIG_PPC_BOOK3S_64
947 
948 void kvmppc_trigger_fac_interrupt(struct kvm_vcpu *vcpu, ulong fac)
949 {
950 	/* Inject the Interrupt Cause field and trigger a guest interrupt */
951 	vcpu->arch.fscr &= ~(0xffULL << 56);
952 	vcpu->arch.fscr |= (fac << 56);
953 	kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_FAC_UNAVAIL);
954 }
955 
956 static void kvmppc_emulate_fac(struct kvm_vcpu *vcpu, ulong fac)
957 {
958 	enum emulation_result er = EMULATE_FAIL;
959 
960 	if (!(kvmppc_get_msr(vcpu) & MSR_PR))
961 		er = kvmppc_emulate_instruction(vcpu->run, vcpu);
962 
963 	if ((er != EMULATE_DONE) && (er != EMULATE_AGAIN)) {
964 		/* Couldn't emulate, trigger interrupt in guest */
965 		kvmppc_trigger_fac_interrupt(vcpu, fac);
966 	}
967 }
968 
969 /* Enable facilities (TAR, EBB, DSCR) for the guest */
970 static int kvmppc_handle_fac(struct kvm_vcpu *vcpu, ulong fac)
971 {
972 	bool guest_fac_enabled;
973 	BUG_ON(!cpu_has_feature(CPU_FTR_ARCH_207S));
974 
975 	/*
976 	 * Not every facility is enabled by FSCR bits, check whether the
977 	 * guest has this facility enabled at all.
978 	 */
979 	switch (fac) {
980 	case FSCR_TAR_LG:
981 	case FSCR_EBB_LG:
982 		guest_fac_enabled = (vcpu->arch.fscr & (1ULL << fac));
983 		break;
984 	case FSCR_TM_LG:
985 		guest_fac_enabled = kvmppc_get_msr(vcpu) & MSR_TM;
986 		break;
987 	default:
988 		guest_fac_enabled = false;
989 		break;
990 	}
991 
992 	if (!guest_fac_enabled) {
993 		/* Facility not enabled by the guest */
994 		kvmppc_trigger_fac_interrupt(vcpu, fac);
995 		return RESUME_GUEST;
996 	}
997 
998 	switch (fac) {
999 	case FSCR_TAR_LG:
1000 		/* TAR switching isn't lazy in Linux yet */
1001 		current->thread.tar = mfspr(SPRN_TAR);
1002 		mtspr(SPRN_TAR, vcpu->arch.tar);
1003 		vcpu->arch.shadow_fscr |= FSCR_TAR;
1004 		break;
1005 	default:
1006 		kvmppc_emulate_fac(vcpu, fac);
1007 		break;
1008 	}
1009 
1010 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1011 	/* Since we disabled MSR_TM at privilege state, the mfspr instruction
1012 	 * for TM spr can trigger TM fac unavailable. In this case, the
1013 	 * emulation is handled by kvmppc_emulate_fac(), which invokes
1014 	 * kvmppc_emulate_mfspr() finally. But note the mfspr can include
1015 	 * RT for NV registers. So it need to restore those NV reg to reflect
1016 	 * the update.
1017 	 */
1018 	if ((fac == FSCR_TM_LG) && !(kvmppc_get_msr(vcpu) & MSR_PR))
1019 		return RESUME_GUEST_NV;
1020 #endif
1021 
1022 	return RESUME_GUEST;
1023 }
1024 
1025 void kvmppc_set_fscr(struct kvm_vcpu *vcpu, u64 fscr)
1026 {
1027 	if ((vcpu->arch.fscr & FSCR_TAR) && !(fscr & FSCR_TAR)) {
1028 		/* TAR got dropped, drop it in shadow too */
1029 		kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
1030 	} else if (!(vcpu->arch.fscr & FSCR_TAR) && (fscr & FSCR_TAR)) {
1031 		vcpu->arch.fscr = fscr;
1032 		kvmppc_handle_fac(vcpu, FSCR_TAR_LG);
1033 		return;
1034 	}
1035 
1036 	vcpu->arch.fscr = fscr;
1037 }
1038 #endif
1039 
1040 static void kvmppc_setup_debug(struct kvm_vcpu *vcpu)
1041 {
1042 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
1043 		u64 msr = kvmppc_get_msr(vcpu);
1044 
1045 		kvmppc_set_msr(vcpu, msr | MSR_SE);
1046 	}
1047 }
1048 
1049 static void kvmppc_clear_debug(struct kvm_vcpu *vcpu)
1050 {
1051 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
1052 		u64 msr = kvmppc_get_msr(vcpu);
1053 
1054 		kvmppc_set_msr(vcpu, msr & ~MSR_SE);
1055 	}
1056 }
1057 
1058 static int kvmppc_exit_pr_progint(struct kvm_run *run, struct kvm_vcpu *vcpu,
1059 				  unsigned int exit_nr)
1060 {
1061 	enum emulation_result er;
1062 	ulong flags;
1063 	u32 last_inst;
1064 	int emul, r;
1065 
1066 	/*
1067 	 * shadow_srr1 only contains valid flags if we came here via a program
1068 	 * exception. The other exceptions (emulation assist, FP unavailable,
1069 	 * etc.) do not provide flags in SRR1, so use an illegal-instruction
1070 	 * exception when injecting a program interrupt into the guest.
1071 	 */
1072 	if (exit_nr == BOOK3S_INTERRUPT_PROGRAM)
1073 		flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
1074 	else
1075 		flags = SRR1_PROGILL;
1076 
1077 	emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1078 	if (emul != EMULATE_DONE)
1079 		return RESUME_GUEST;
1080 
1081 	if (kvmppc_get_msr(vcpu) & MSR_PR) {
1082 #ifdef EXIT_DEBUG
1083 		pr_info("Userspace triggered 0x700 exception at\n 0x%lx (0x%x)\n",
1084 			kvmppc_get_pc(vcpu), last_inst);
1085 #endif
1086 		if ((last_inst & 0xff0007ff) != (INS_DCBZ & 0xfffffff7)) {
1087 			kvmppc_core_queue_program(vcpu, flags);
1088 			return RESUME_GUEST;
1089 		}
1090 	}
1091 
1092 	vcpu->stat.emulated_inst_exits++;
1093 	er = kvmppc_emulate_instruction(run, vcpu);
1094 	switch (er) {
1095 	case EMULATE_DONE:
1096 		r = RESUME_GUEST_NV;
1097 		break;
1098 	case EMULATE_AGAIN:
1099 		r = RESUME_GUEST;
1100 		break;
1101 	case EMULATE_FAIL:
1102 		pr_crit("%s: emulation at %lx failed (%08x)\n",
1103 			__func__, kvmppc_get_pc(vcpu), last_inst);
1104 		kvmppc_core_queue_program(vcpu, flags);
1105 		r = RESUME_GUEST;
1106 		break;
1107 	case EMULATE_DO_MMIO:
1108 		run->exit_reason = KVM_EXIT_MMIO;
1109 		r = RESUME_HOST_NV;
1110 		break;
1111 	case EMULATE_EXIT_USER:
1112 		r = RESUME_HOST_NV;
1113 		break;
1114 	default:
1115 		BUG();
1116 	}
1117 
1118 	return r;
1119 }
1120 
1121 int kvmppc_handle_exit_pr(struct kvm_run *run, struct kvm_vcpu *vcpu,
1122 			  unsigned int exit_nr)
1123 {
1124 	int r = RESUME_HOST;
1125 	int s;
1126 
1127 	vcpu->stat.sum_exits++;
1128 
1129 	run->exit_reason = KVM_EXIT_UNKNOWN;
1130 	run->ready_for_interrupt_injection = 1;
1131 
1132 	/* We get here with MSR.EE=1 */
1133 
1134 	trace_kvm_exit(exit_nr, vcpu);
1135 	guest_exit();
1136 
1137 	switch (exit_nr) {
1138 	case BOOK3S_INTERRUPT_INST_STORAGE:
1139 	{
1140 		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1141 		vcpu->stat.pf_instruc++;
1142 
1143 		if (kvmppc_is_split_real(vcpu))
1144 			kvmppc_fixup_split_real(vcpu);
1145 
1146 #ifdef CONFIG_PPC_BOOK3S_32
1147 		/* We set segments as unused segments when invalidating them. So
1148 		 * treat the respective fault as segment fault. */
1149 		{
1150 			struct kvmppc_book3s_shadow_vcpu *svcpu;
1151 			u32 sr;
1152 
1153 			svcpu = svcpu_get(vcpu);
1154 			sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
1155 			svcpu_put(svcpu);
1156 			if (sr == SR_INVALID) {
1157 				kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
1158 				r = RESUME_GUEST;
1159 				break;
1160 			}
1161 		}
1162 #endif
1163 
1164 		/* only care about PTEG not found errors, but leave NX alone */
1165 		if (shadow_srr1 & 0x40000000) {
1166 			int idx = srcu_read_lock(&vcpu->kvm->srcu);
1167 			r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
1168 			srcu_read_unlock(&vcpu->kvm->srcu, idx);
1169 			vcpu->stat.sp_instruc++;
1170 		} else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
1171 			  (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
1172 			/*
1173 			 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
1174 			 *     so we can't use the NX bit inside the guest. Let's cross our fingers,
1175 			 *     that no guest that needs the dcbz hack does NX.
1176 			 */
1177 			kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
1178 			r = RESUME_GUEST;
1179 		} else {
1180 			kvmppc_core_queue_inst_storage(vcpu,
1181 						shadow_srr1 & 0x58000000);
1182 			r = RESUME_GUEST;
1183 		}
1184 		break;
1185 	}
1186 	case BOOK3S_INTERRUPT_DATA_STORAGE:
1187 	{
1188 		ulong dar = kvmppc_get_fault_dar(vcpu);
1189 		u32 fault_dsisr = vcpu->arch.fault_dsisr;
1190 		vcpu->stat.pf_storage++;
1191 
1192 #ifdef CONFIG_PPC_BOOK3S_32
1193 		/* We set segments as unused segments when invalidating them. So
1194 		 * treat the respective fault as segment fault. */
1195 		{
1196 			struct kvmppc_book3s_shadow_vcpu *svcpu;
1197 			u32 sr;
1198 
1199 			svcpu = svcpu_get(vcpu);
1200 			sr = svcpu->sr[dar >> SID_SHIFT];
1201 			svcpu_put(svcpu);
1202 			if (sr == SR_INVALID) {
1203 				kvmppc_mmu_map_segment(vcpu, dar);
1204 				r = RESUME_GUEST;
1205 				break;
1206 			}
1207 		}
1208 #endif
1209 
1210 		/*
1211 		 * We need to handle missing shadow PTEs, and
1212 		 * protection faults due to us mapping a page read-only
1213 		 * when the guest thinks it is writable.
1214 		 */
1215 		if (fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT)) {
1216 			int idx = srcu_read_lock(&vcpu->kvm->srcu);
1217 			r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
1218 			srcu_read_unlock(&vcpu->kvm->srcu, idx);
1219 		} else {
1220 			kvmppc_core_queue_data_storage(vcpu, dar, fault_dsisr);
1221 			r = RESUME_GUEST;
1222 		}
1223 		break;
1224 	}
1225 	case BOOK3S_INTERRUPT_DATA_SEGMENT:
1226 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
1227 			kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
1228 			kvmppc_book3s_queue_irqprio(vcpu,
1229 				BOOK3S_INTERRUPT_DATA_SEGMENT);
1230 		}
1231 		r = RESUME_GUEST;
1232 		break;
1233 	case BOOK3S_INTERRUPT_INST_SEGMENT:
1234 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
1235 			kvmppc_book3s_queue_irqprio(vcpu,
1236 				BOOK3S_INTERRUPT_INST_SEGMENT);
1237 		}
1238 		r = RESUME_GUEST;
1239 		break;
1240 	/* We're good on these - the host merely wanted to get our attention */
1241 	case BOOK3S_INTERRUPT_DECREMENTER:
1242 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1243 	case BOOK3S_INTERRUPT_DOORBELL:
1244 	case BOOK3S_INTERRUPT_H_DOORBELL:
1245 		vcpu->stat.dec_exits++;
1246 		r = RESUME_GUEST;
1247 		break;
1248 	case BOOK3S_INTERRUPT_EXTERNAL:
1249 	case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
1250 	case BOOK3S_INTERRUPT_EXTERNAL_HV:
1251 	case BOOK3S_INTERRUPT_H_VIRT:
1252 		vcpu->stat.ext_intr_exits++;
1253 		r = RESUME_GUEST;
1254 		break;
1255 	case BOOK3S_INTERRUPT_HMI:
1256 	case BOOK3S_INTERRUPT_PERFMON:
1257 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1258 		r = RESUME_GUEST;
1259 		break;
1260 	case BOOK3S_INTERRUPT_PROGRAM:
1261 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1262 		r = kvmppc_exit_pr_progint(run, vcpu, exit_nr);
1263 		break;
1264 	case BOOK3S_INTERRUPT_SYSCALL:
1265 	{
1266 		u32 last_sc;
1267 		int emul;
1268 
1269 		/* Get last sc for papr */
1270 		if (vcpu->arch.papr_enabled) {
1271 			/* The sc instuction points SRR0 to the next inst */
1272 			emul = kvmppc_get_last_inst(vcpu, INST_SC, &last_sc);
1273 			if (emul != EMULATE_DONE) {
1274 				kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) - 4);
1275 				r = RESUME_GUEST;
1276 				break;
1277 			}
1278 		}
1279 
1280 		if (vcpu->arch.papr_enabled &&
1281 		    (last_sc == 0x44000022) &&
1282 		    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
1283 			/* SC 1 papr hypercalls */
1284 			ulong cmd = kvmppc_get_gpr(vcpu, 3);
1285 			int i;
1286 
1287 #ifdef CONFIG_PPC_BOOK3S_64
1288 			if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
1289 				r = RESUME_GUEST;
1290 				break;
1291 			}
1292 #endif
1293 
1294 			run->papr_hcall.nr = cmd;
1295 			for (i = 0; i < 9; ++i) {
1296 				ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
1297 				run->papr_hcall.args[i] = gpr;
1298 			}
1299 			run->exit_reason = KVM_EXIT_PAPR_HCALL;
1300 			vcpu->arch.hcall_needed = 1;
1301 			r = RESUME_HOST;
1302 		} else if (vcpu->arch.osi_enabled &&
1303 		    (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
1304 		    (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
1305 			/* MOL hypercalls */
1306 			u64 *gprs = run->osi.gprs;
1307 			int i;
1308 
1309 			run->exit_reason = KVM_EXIT_OSI;
1310 			for (i = 0; i < 32; i++)
1311 				gprs[i] = kvmppc_get_gpr(vcpu, i);
1312 			vcpu->arch.osi_needed = 1;
1313 			r = RESUME_HOST_NV;
1314 		} else if (!(kvmppc_get_msr(vcpu) & MSR_PR) &&
1315 		    (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
1316 			/* KVM PV hypercalls */
1317 			kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
1318 			r = RESUME_GUEST;
1319 		} else {
1320 			/* Guest syscalls */
1321 			vcpu->stat.syscall_exits++;
1322 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1323 			r = RESUME_GUEST;
1324 		}
1325 		break;
1326 	}
1327 	case BOOK3S_INTERRUPT_FP_UNAVAIL:
1328 	case BOOK3S_INTERRUPT_ALTIVEC:
1329 	case BOOK3S_INTERRUPT_VSX:
1330 	{
1331 		int ext_msr = 0;
1332 		int emul;
1333 		u32 last_inst;
1334 
1335 		if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE) {
1336 			/* Do paired single instruction emulation */
1337 			emul = kvmppc_get_last_inst(vcpu, INST_GENERIC,
1338 						    &last_inst);
1339 			if (emul == EMULATE_DONE)
1340 				r = kvmppc_exit_pr_progint(run, vcpu, exit_nr);
1341 			else
1342 				r = RESUME_GUEST;
1343 
1344 			break;
1345 		}
1346 
1347 		/* Enable external provider */
1348 		switch (exit_nr) {
1349 		case BOOK3S_INTERRUPT_FP_UNAVAIL:
1350 			ext_msr = MSR_FP;
1351 			break;
1352 
1353 		case BOOK3S_INTERRUPT_ALTIVEC:
1354 			ext_msr = MSR_VEC;
1355 			break;
1356 
1357 		case BOOK3S_INTERRUPT_VSX:
1358 			ext_msr = MSR_VSX;
1359 			break;
1360 		}
1361 
1362 		r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
1363 		break;
1364 	}
1365 	case BOOK3S_INTERRUPT_ALIGNMENT:
1366 	{
1367 		u32 last_inst;
1368 		int emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1369 
1370 		if (emul == EMULATE_DONE) {
1371 			u32 dsisr;
1372 			u64 dar;
1373 
1374 			dsisr = kvmppc_alignment_dsisr(vcpu, last_inst);
1375 			dar = kvmppc_alignment_dar(vcpu, last_inst);
1376 
1377 			kvmppc_set_dsisr(vcpu, dsisr);
1378 			kvmppc_set_dar(vcpu, dar);
1379 
1380 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1381 		}
1382 		r = RESUME_GUEST;
1383 		break;
1384 	}
1385 #ifdef CONFIG_PPC_BOOK3S_64
1386 	case BOOK3S_INTERRUPT_FAC_UNAVAIL:
1387 		r = kvmppc_handle_fac(vcpu, vcpu->arch.shadow_fscr >> 56);
1388 		break;
1389 #endif
1390 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1391 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1392 		r = RESUME_GUEST;
1393 		break;
1394 	case BOOK3S_INTERRUPT_TRACE:
1395 		if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
1396 			run->exit_reason = KVM_EXIT_DEBUG;
1397 			r = RESUME_HOST;
1398 		} else {
1399 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1400 			r = RESUME_GUEST;
1401 		}
1402 		break;
1403 	default:
1404 	{
1405 		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1406 		/* Ugh - bork here! What did we get? */
1407 		printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1408 			exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
1409 		r = RESUME_HOST;
1410 		BUG();
1411 		break;
1412 	}
1413 	}
1414 
1415 	if (!(r & RESUME_HOST)) {
1416 		/* To avoid clobbering exit_reason, only check for signals if
1417 		 * we aren't already exiting to userspace for some other
1418 		 * reason. */
1419 
1420 		/*
1421 		 * Interrupts could be timers for the guest which we have to
1422 		 * inject again, so let's postpone them until we're in the guest
1423 		 * and if we really did time things so badly, then we just exit
1424 		 * again due to a host external interrupt.
1425 		 */
1426 		s = kvmppc_prepare_to_enter(vcpu);
1427 		if (s <= 0)
1428 			r = s;
1429 		else {
1430 			/* interrupts now hard-disabled */
1431 			kvmppc_fix_ee_before_entry();
1432 		}
1433 
1434 		kvmppc_handle_lost_ext(vcpu);
1435 	}
1436 
1437 	trace_kvm_book3s_reenter(r, vcpu);
1438 
1439 	return r;
1440 }
1441 
1442 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu *vcpu,
1443 					    struct kvm_sregs *sregs)
1444 {
1445 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1446 	int i;
1447 
1448 	sregs->pvr = vcpu->arch.pvr;
1449 
1450 	sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1451 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1452 		for (i = 0; i < 64; i++) {
1453 			sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
1454 			sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1455 		}
1456 	} else {
1457 		for (i = 0; i < 16; i++)
1458 			sregs->u.s.ppc32.sr[i] = kvmppc_get_sr(vcpu, i);
1459 
1460 		for (i = 0; i < 8; i++) {
1461 			sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1462 			sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1463 		}
1464 	}
1465 
1466 	return 0;
1467 }
1468 
1469 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu *vcpu,
1470 					    struct kvm_sregs *sregs)
1471 {
1472 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1473 	int i;
1474 
1475 	kvmppc_set_pvr_pr(vcpu, sregs->pvr);
1476 
1477 	vcpu3s->sdr1 = sregs->u.s.sdr1;
1478 #ifdef CONFIG_PPC_BOOK3S_64
1479 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1480 		/* Flush all SLB entries */
1481 		vcpu->arch.mmu.slbmte(vcpu, 0, 0);
1482 		vcpu->arch.mmu.slbia(vcpu);
1483 
1484 		for (i = 0; i < 64; i++) {
1485 			u64 rb = sregs->u.s.ppc64.slb[i].slbe;
1486 			u64 rs = sregs->u.s.ppc64.slb[i].slbv;
1487 
1488 			if (rb & SLB_ESID_V)
1489 				vcpu->arch.mmu.slbmte(vcpu, rs, rb);
1490 		}
1491 	} else
1492 #endif
1493 	{
1494 		for (i = 0; i < 16; i++) {
1495 			vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1496 		}
1497 		for (i = 0; i < 8; i++) {
1498 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1499 				       (u32)sregs->u.s.ppc32.ibat[i]);
1500 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1501 				       (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1502 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1503 				       (u32)sregs->u.s.ppc32.dbat[i]);
1504 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1505 				       (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1506 		}
1507 	}
1508 
1509 	/* Flush the MMU after messing with the segments */
1510 	kvmppc_mmu_pte_flush(vcpu, 0, 0);
1511 
1512 	return 0;
1513 }
1514 
1515 static int kvmppc_get_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1516 				 union kvmppc_one_reg *val)
1517 {
1518 	int r = 0;
1519 
1520 	switch (id) {
1521 	case KVM_REG_PPC_DEBUG_INST:
1522 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1523 		break;
1524 	case KVM_REG_PPC_HIOR:
1525 		*val = get_reg_val(id, to_book3s(vcpu)->hior);
1526 		break;
1527 	case KVM_REG_PPC_VTB:
1528 		*val = get_reg_val(id, to_book3s(vcpu)->vtb);
1529 		break;
1530 	case KVM_REG_PPC_LPCR:
1531 	case KVM_REG_PPC_LPCR_64:
1532 		/*
1533 		 * We are only interested in the LPCR_ILE bit
1534 		 */
1535 		if (vcpu->arch.intr_msr & MSR_LE)
1536 			*val = get_reg_val(id, LPCR_ILE);
1537 		else
1538 			*val = get_reg_val(id, 0);
1539 		break;
1540 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1541 	case KVM_REG_PPC_TFHAR:
1542 		*val = get_reg_val(id, vcpu->arch.tfhar);
1543 		break;
1544 	case KVM_REG_PPC_TFIAR:
1545 		*val = get_reg_val(id, vcpu->arch.tfiar);
1546 		break;
1547 	case KVM_REG_PPC_TEXASR:
1548 		*val = get_reg_val(id, vcpu->arch.texasr);
1549 		break;
1550 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1551 		*val = get_reg_val(id,
1552 				vcpu->arch.gpr_tm[id-KVM_REG_PPC_TM_GPR0]);
1553 		break;
1554 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1555 	{
1556 		int i, j;
1557 
1558 		i = id - KVM_REG_PPC_TM_VSR0;
1559 		if (i < 32)
1560 			for (j = 0; j < TS_FPRWIDTH; j++)
1561 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1562 		else {
1563 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1564 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1565 			else
1566 				r = -ENXIO;
1567 		}
1568 		break;
1569 	}
1570 	case KVM_REG_PPC_TM_CR:
1571 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1572 		break;
1573 	case KVM_REG_PPC_TM_XER:
1574 		*val = get_reg_val(id, vcpu->arch.xer_tm);
1575 		break;
1576 	case KVM_REG_PPC_TM_LR:
1577 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1578 		break;
1579 	case KVM_REG_PPC_TM_CTR:
1580 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1581 		break;
1582 	case KVM_REG_PPC_TM_FPSCR:
1583 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1584 		break;
1585 	case KVM_REG_PPC_TM_AMR:
1586 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1587 		break;
1588 	case KVM_REG_PPC_TM_PPR:
1589 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1590 		break;
1591 	case KVM_REG_PPC_TM_VRSAVE:
1592 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1593 		break;
1594 	case KVM_REG_PPC_TM_VSCR:
1595 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1596 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1597 		else
1598 			r = -ENXIO;
1599 		break;
1600 	case KVM_REG_PPC_TM_DSCR:
1601 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1602 		break;
1603 	case KVM_REG_PPC_TM_TAR:
1604 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1605 		break;
1606 #endif
1607 	default:
1608 		r = -EINVAL;
1609 		break;
1610 	}
1611 
1612 	return r;
1613 }
1614 
1615 static void kvmppc_set_lpcr_pr(struct kvm_vcpu *vcpu, u64 new_lpcr)
1616 {
1617 	if (new_lpcr & LPCR_ILE)
1618 		vcpu->arch.intr_msr |= MSR_LE;
1619 	else
1620 		vcpu->arch.intr_msr &= ~MSR_LE;
1621 }
1622 
1623 static int kvmppc_set_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1624 				 union kvmppc_one_reg *val)
1625 {
1626 	int r = 0;
1627 
1628 	switch (id) {
1629 	case KVM_REG_PPC_HIOR:
1630 		to_book3s(vcpu)->hior = set_reg_val(id, *val);
1631 		to_book3s(vcpu)->hior_explicit = true;
1632 		break;
1633 	case KVM_REG_PPC_VTB:
1634 		to_book3s(vcpu)->vtb = set_reg_val(id, *val);
1635 		break;
1636 	case KVM_REG_PPC_LPCR:
1637 	case KVM_REG_PPC_LPCR_64:
1638 		kvmppc_set_lpcr_pr(vcpu, set_reg_val(id, *val));
1639 		break;
1640 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1641 	case KVM_REG_PPC_TFHAR:
1642 		vcpu->arch.tfhar = set_reg_val(id, *val);
1643 		break;
1644 	case KVM_REG_PPC_TFIAR:
1645 		vcpu->arch.tfiar = set_reg_val(id, *val);
1646 		break;
1647 	case KVM_REG_PPC_TEXASR:
1648 		vcpu->arch.texasr = set_reg_val(id, *val);
1649 		break;
1650 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1651 		vcpu->arch.gpr_tm[id - KVM_REG_PPC_TM_GPR0] =
1652 			set_reg_val(id, *val);
1653 		break;
1654 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1655 	{
1656 		int i, j;
1657 
1658 		i = id - KVM_REG_PPC_TM_VSR0;
1659 		if (i < 32)
1660 			for (j = 0; j < TS_FPRWIDTH; j++)
1661 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1662 		else
1663 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1664 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1665 			else
1666 				r = -ENXIO;
1667 		break;
1668 	}
1669 	case KVM_REG_PPC_TM_CR:
1670 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1671 		break;
1672 	case KVM_REG_PPC_TM_XER:
1673 		vcpu->arch.xer_tm = set_reg_val(id, *val);
1674 		break;
1675 	case KVM_REG_PPC_TM_LR:
1676 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1677 		break;
1678 	case KVM_REG_PPC_TM_CTR:
1679 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1680 		break;
1681 	case KVM_REG_PPC_TM_FPSCR:
1682 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1683 		break;
1684 	case KVM_REG_PPC_TM_AMR:
1685 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1686 		break;
1687 	case KVM_REG_PPC_TM_PPR:
1688 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1689 		break;
1690 	case KVM_REG_PPC_TM_VRSAVE:
1691 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1692 		break;
1693 	case KVM_REG_PPC_TM_VSCR:
1694 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1695 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1696 		else
1697 			r = -ENXIO;
1698 		break;
1699 	case KVM_REG_PPC_TM_DSCR:
1700 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1701 		break;
1702 	case KVM_REG_PPC_TM_TAR:
1703 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1704 		break;
1705 #endif
1706 	default:
1707 		r = -EINVAL;
1708 		break;
1709 	}
1710 
1711 	return r;
1712 }
1713 
1714 static struct kvm_vcpu *kvmppc_core_vcpu_create_pr(struct kvm *kvm,
1715 						   unsigned int id)
1716 {
1717 	struct kvmppc_vcpu_book3s *vcpu_book3s;
1718 	struct kvm_vcpu *vcpu;
1719 	int err = -ENOMEM;
1720 	unsigned long p;
1721 
1722 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1723 	if (!vcpu)
1724 		goto out;
1725 
1726 	vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1727 	if (!vcpu_book3s)
1728 		goto free_vcpu;
1729 	vcpu->arch.book3s = vcpu_book3s;
1730 
1731 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1732 	vcpu->arch.shadow_vcpu =
1733 		kzalloc(sizeof(*vcpu->arch.shadow_vcpu), GFP_KERNEL);
1734 	if (!vcpu->arch.shadow_vcpu)
1735 		goto free_vcpu3s;
1736 #endif
1737 
1738 	err = kvm_vcpu_init(vcpu, kvm, id);
1739 	if (err)
1740 		goto free_shadow_vcpu;
1741 
1742 	err = -ENOMEM;
1743 	p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1744 	if (!p)
1745 		goto uninit_vcpu;
1746 	vcpu->arch.shared = (void *)p;
1747 #ifdef CONFIG_PPC_BOOK3S_64
1748 	/* Always start the shared struct in native endian mode */
1749 #ifdef __BIG_ENDIAN__
1750         vcpu->arch.shared_big_endian = true;
1751 #else
1752         vcpu->arch.shared_big_endian = false;
1753 #endif
1754 
1755 	/*
1756 	 * Default to the same as the host if we're on sufficiently
1757 	 * recent machine that we have 1TB segments;
1758 	 * otherwise default to PPC970FX.
1759 	 */
1760 	vcpu->arch.pvr = 0x3C0301;
1761 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1762 		vcpu->arch.pvr = mfspr(SPRN_PVR);
1763 	vcpu->arch.intr_msr = MSR_SF;
1764 #else
1765 	/* default to book3s_32 (750) */
1766 	vcpu->arch.pvr = 0x84202;
1767 #endif
1768 	kvmppc_set_pvr_pr(vcpu, vcpu->arch.pvr);
1769 	vcpu->arch.slb_nr = 64;
1770 
1771 	vcpu->arch.shadow_msr = MSR_USER64 & ~MSR_LE;
1772 
1773 	err = kvmppc_mmu_init(vcpu);
1774 	if (err < 0)
1775 		goto uninit_vcpu;
1776 
1777 	return vcpu;
1778 
1779 uninit_vcpu:
1780 	kvm_vcpu_uninit(vcpu);
1781 free_shadow_vcpu:
1782 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1783 	kfree(vcpu->arch.shadow_vcpu);
1784 free_vcpu3s:
1785 #endif
1786 	vfree(vcpu_book3s);
1787 free_vcpu:
1788 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1789 out:
1790 	return ERR_PTR(err);
1791 }
1792 
1793 static void kvmppc_core_vcpu_free_pr(struct kvm_vcpu *vcpu)
1794 {
1795 	struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1796 
1797 	free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1798 	kvm_vcpu_uninit(vcpu);
1799 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1800 	kfree(vcpu->arch.shadow_vcpu);
1801 #endif
1802 	vfree(vcpu_book3s);
1803 	kmem_cache_free(kvm_vcpu_cache, vcpu);
1804 }
1805 
1806 static int kvmppc_vcpu_run_pr(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1807 {
1808 	int ret;
1809 #ifdef CONFIG_ALTIVEC
1810 	unsigned long uninitialized_var(vrsave);
1811 #endif
1812 
1813 	/* Check if we can run the vcpu at all */
1814 	if (!vcpu->arch.sane) {
1815 		kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1816 		ret = -EINVAL;
1817 		goto out;
1818 	}
1819 
1820 	kvmppc_setup_debug(vcpu);
1821 
1822 	/*
1823 	 * Interrupts could be timers for the guest which we have to inject
1824 	 * again, so let's postpone them until we're in the guest and if we
1825 	 * really did time things so badly, then we just exit again due to
1826 	 * a host external interrupt.
1827 	 */
1828 	ret = kvmppc_prepare_to_enter(vcpu);
1829 	if (ret <= 0)
1830 		goto out;
1831 	/* interrupts now hard-disabled */
1832 
1833 	/* Save FPU, Altivec and VSX state */
1834 	giveup_all(current);
1835 
1836 	/* Preload FPU if it's enabled */
1837 	if (kvmppc_get_msr(vcpu) & MSR_FP)
1838 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1839 
1840 	kvmppc_fix_ee_before_entry();
1841 
1842 	ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1843 
1844 	kvmppc_clear_debug(vcpu);
1845 
1846 	/* No need for guest_exit. It's done in handle_exit.
1847 	   We also get here with interrupts enabled. */
1848 
1849 	/* Make sure we save the guest FPU/Altivec/VSX state */
1850 	kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1851 
1852 	/* Make sure we save the guest TAR/EBB/DSCR state */
1853 	kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
1854 
1855 out:
1856 	vcpu->mode = OUTSIDE_GUEST_MODE;
1857 	return ret;
1858 }
1859 
1860 /*
1861  * Get (and clear) the dirty memory log for a memory slot.
1862  */
1863 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm *kvm,
1864 					 struct kvm_dirty_log *log)
1865 {
1866 	struct kvm_memslots *slots;
1867 	struct kvm_memory_slot *memslot;
1868 	struct kvm_vcpu *vcpu;
1869 	ulong ga, ga_end;
1870 	int is_dirty = 0;
1871 	int r;
1872 	unsigned long n;
1873 
1874 	mutex_lock(&kvm->slots_lock);
1875 
1876 	r = kvm_get_dirty_log(kvm, log, &is_dirty);
1877 	if (r)
1878 		goto out;
1879 
1880 	/* If nothing is dirty, don't bother messing with page tables. */
1881 	if (is_dirty) {
1882 		slots = kvm_memslots(kvm);
1883 		memslot = id_to_memslot(slots, log->slot);
1884 
1885 		ga = memslot->base_gfn << PAGE_SHIFT;
1886 		ga_end = ga + (memslot->npages << PAGE_SHIFT);
1887 
1888 		kvm_for_each_vcpu(n, vcpu, kvm)
1889 			kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1890 
1891 		n = kvm_dirty_bitmap_bytes(memslot);
1892 		memset(memslot->dirty_bitmap, 0, n);
1893 	}
1894 
1895 	r = 0;
1896 out:
1897 	mutex_unlock(&kvm->slots_lock);
1898 	return r;
1899 }
1900 
1901 static void kvmppc_core_flush_memslot_pr(struct kvm *kvm,
1902 					 struct kvm_memory_slot *memslot)
1903 {
1904 	return;
1905 }
1906 
1907 static int kvmppc_core_prepare_memory_region_pr(struct kvm *kvm,
1908 					struct kvm_memory_slot *memslot,
1909 					const struct kvm_userspace_memory_region *mem)
1910 {
1911 	return 0;
1912 }
1913 
1914 static void kvmppc_core_commit_memory_region_pr(struct kvm *kvm,
1915 				const struct kvm_userspace_memory_region *mem,
1916 				const struct kvm_memory_slot *old,
1917 				const struct kvm_memory_slot *new)
1918 {
1919 	return;
1920 }
1921 
1922 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot *free,
1923 					struct kvm_memory_slot *dont)
1924 {
1925 	return;
1926 }
1927 
1928 static int kvmppc_core_create_memslot_pr(struct kvm_memory_slot *slot,
1929 					 unsigned long npages)
1930 {
1931 	return 0;
1932 }
1933 
1934 
1935 #ifdef CONFIG_PPC64
1936 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1937 					 struct kvm_ppc_smmu_info *info)
1938 {
1939 	long int i;
1940 	struct kvm_vcpu *vcpu;
1941 
1942 	info->flags = 0;
1943 
1944 	/* SLB is always 64 entries */
1945 	info->slb_size = 64;
1946 
1947 	/* Standard 4k base page size segment */
1948 	info->sps[0].page_shift = 12;
1949 	info->sps[0].slb_enc = 0;
1950 	info->sps[0].enc[0].page_shift = 12;
1951 	info->sps[0].enc[0].pte_enc = 0;
1952 
1953 	/*
1954 	 * 64k large page size.
1955 	 * We only want to put this in if the CPUs we're emulating
1956 	 * support it, but unfortunately we don't have a vcpu easily
1957 	 * to hand here to test.  Just pick the first vcpu, and if
1958 	 * that doesn't exist yet, report the minimum capability,
1959 	 * i.e., no 64k pages.
1960 	 * 1T segment support goes along with 64k pages.
1961 	 */
1962 	i = 1;
1963 	vcpu = kvm_get_vcpu(kvm, 0);
1964 	if (vcpu && (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
1965 		info->flags = KVM_PPC_1T_SEGMENTS;
1966 		info->sps[i].page_shift = 16;
1967 		info->sps[i].slb_enc = SLB_VSID_L | SLB_VSID_LP_01;
1968 		info->sps[i].enc[0].page_shift = 16;
1969 		info->sps[i].enc[0].pte_enc = 1;
1970 		++i;
1971 	}
1972 
1973 	/* Standard 16M large page size segment */
1974 	info->sps[i].page_shift = 24;
1975 	info->sps[i].slb_enc = SLB_VSID_L;
1976 	info->sps[i].enc[0].page_shift = 24;
1977 	info->sps[i].enc[0].pte_enc = 0;
1978 
1979 	return 0;
1980 }
1981 
1982 static int kvm_configure_mmu_pr(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
1983 {
1984 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
1985 		return -ENODEV;
1986 	/* Require flags and process table base and size to all be zero. */
1987 	if (cfg->flags || cfg->process_table)
1988 		return -EINVAL;
1989 	return 0;
1990 }
1991 
1992 #else
1993 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1994 					 struct kvm_ppc_smmu_info *info)
1995 {
1996 	/* We should not get called */
1997 	BUG();
1998 }
1999 #endif /* CONFIG_PPC64 */
2000 
2001 static unsigned int kvm_global_user_count = 0;
2002 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
2003 
2004 static int kvmppc_core_init_vm_pr(struct kvm *kvm)
2005 {
2006 	mutex_init(&kvm->arch.hpt_mutex);
2007 
2008 #ifdef CONFIG_PPC_BOOK3S_64
2009 	/* Start out with the default set of hcalls enabled */
2010 	kvmppc_pr_init_default_hcalls(kvm);
2011 #endif
2012 
2013 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
2014 		spin_lock(&kvm_global_user_count_lock);
2015 		if (++kvm_global_user_count == 1)
2016 			pseries_disable_reloc_on_exc();
2017 		spin_unlock(&kvm_global_user_count_lock);
2018 	}
2019 	return 0;
2020 }
2021 
2022 static void kvmppc_core_destroy_vm_pr(struct kvm *kvm)
2023 {
2024 #ifdef CONFIG_PPC64
2025 	WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
2026 #endif
2027 
2028 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
2029 		spin_lock(&kvm_global_user_count_lock);
2030 		BUG_ON(kvm_global_user_count == 0);
2031 		if (--kvm_global_user_count == 0)
2032 			pseries_enable_reloc_on_exc();
2033 		spin_unlock(&kvm_global_user_count_lock);
2034 	}
2035 }
2036 
2037 static int kvmppc_core_check_processor_compat_pr(void)
2038 {
2039 	/*
2040 	 * PR KVM can work on POWER9 inside a guest partition
2041 	 * running in HPT mode.  It can't work if we are using
2042 	 * radix translation (because radix provides no way for
2043 	 * a process to have unique translations in quadrant 3).
2044 	 */
2045 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
2046 		return -EIO;
2047 	return 0;
2048 }
2049 
2050 static long kvm_arch_vm_ioctl_pr(struct file *filp,
2051 				 unsigned int ioctl, unsigned long arg)
2052 {
2053 	return -ENOTTY;
2054 }
2055 
2056 static struct kvmppc_ops kvm_ops_pr = {
2057 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_pr,
2058 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_pr,
2059 	.get_one_reg = kvmppc_get_one_reg_pr,
2060 	.set_one_reg = kvmppc_set_one_reg_pr,
2061 	.vcpu_load   = kvmppc_core_vcpu_load_pr,
2062 	.vcpu_put    = kvmppc_core_vcpu_put_pr,
2063 	.set_msr     = kvmppc_set_msr_pr,
2064 	.vcpu_run    = kvmppc_vcpu_run_pr,
2065 	.vcpu_create = kvmppc_core_vcpu_create_pr,
2066 	.vcpu_free   = kvmppc_core_vcpu_free_pr,
2067 	.check_requests = kvmppc_core_check_requests_pr,
2068 	.get_dirty_log = kvm_vm_ioctl_get_dirty_log_pr,
2069 	.flush_memslot = kvmppc_core_flush_memslot_pr,
2070 	.prepare_memory_region = kvmppc_core_prepare_memory_region_pr,
2071 	.commit_memory_region = kvmppc_core_commit_memory_region_pr,
2072 	.unmap_hva_range = kvm_unmap_hva_range_pr,
2073 	.age_hva  = kvm_age_hva_pr,
2074 	.test_age_hva = kvm_test_age_hva_pr,
2075 	.set_spte_hva = kvm_set_spte_hva_pr,
2076 	.mmu_destroy  = kvmppc_mmu_destroy_pr,
2077 	.free_memslot = kvmppc_core_free_memslot_pr,
2078 	.create_memslot = kvmppc_core_create_memslot_pr,
2079 	.init_vm = kvmppc_core_init_vm_pr,
2080 	.destroy_vm = kvmppc_core_destroy_vm_pr,
2081 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_pr,
2082 	.emulate_op = kvmppc_core_emulate_op_pr,
2083 	.emulate_mtspr = kvmppc_core_emulate_mtspr_pr,
2084 	.emulate_mfspr = kvmppc_core_emulate_mfspr_pr,
2085 	.fast_vcpu_kick = kvm_vcpu_kick,
2086 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_pr,
2087 #ifdef CONFIG_PPC_BOOK3S_64
2088 	.hcall_implemented = kvmppc_hcall_impl_pr,
2089 	.configure_mmu = kvm_configure_mmu_pr,
2090 #endif
2091 	.giveup_ext = kvmppc_giveup_ext,
2092 };
2093 
2094 
2095 int kvmppc_book3s_init_pr(void)
2096 {
2097 	int r;
2098 
2099 	r = kvmppc_core_check_processor_compat_pr();
2100 	if (r < 0)
2101 		return r;
2102 
2103 	kvm_ops_pr.owner = THIS_MODULE;
2104 	kvmppc_pr_ops = &kvm_ops_pr;
2105 
2106 	r = kvmppc_mmu_hpte_sysinit();
2107 	return r;
2108 }
2109 
2110 void kvmppc_book3s_exit_pr(void)
2111 {
2112 	kvmppc_pr_ops = NULL;
2113 	kvmppc_mmu_hpte_sysexit();
2114 }
2115 
2116 /*
2117  * We only support separate modules for book3s 64
2118  */
2119 #ifdef CONFIG_PPC_BOOK3S_64
2120 
2121 module_init(kvmppc_book3s_init_pr);
2122 module_exit(kvmppc_book3s_exit_pr);
2123 
2124 MODULE_LICENSE("GPL");
2125 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2126 MODULE_ALIAS("devname:kvm");
2127 #endif
2128