xref: /openbmc/linux/arch/powerpc/kvm/book3s_pr.c (revision 877013bc)
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/io.h>
29 #include <asm/kvm_ppc.h>
30 #include <asm/kvm_book3s.h>
31 #include <asm/mmu_context.h>
32 #include <asm/switch_to.h>
33 #include <asm/firmware.h>
34 #include <asm/setup.h>
35 #include <linux/gfp.h>
36 #include <linux/sched.h>
37 #include <linux/vmalloc.h>
38 #include <linux/highmem.h>
39 #include <linux/module.h>
40 #include <linux/miscdevice.h>
41 #include <asm/asm-prototypes.h>
42 #include <asm/tm.h>
43 
44 #include "book3s.h"
45 
46 #define CREATE_TRACE_POINTS
47 #include "trace_pr.h"
48 
49 /* #define EXIT_DEBUG */
50 /* #define DEBUG_EXT */
51 
52 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
53 			     ulong msr);
54 #ifdef CONFIG_PPC_BOOK3S_64
55 static int kvmppc_handle_fac(struct kvm_vcpu *vcpu, ulong fac);
56 #endif
57 
58 /* Some compatibility defines */
59 #ifdef CONFIG_PPC_BOOK3S_32
60 #define MSR_USER32 MSR_USER
61 #define MSR_USER64 MSR_USER
62 #define HW_PAGE_SIZE PAGE_SIZE
63 #define HPTE_R_M   _PAGE_COHERENT
64 #endif
65 
66 static bool kvmppc_is_split_real(struct kvm_vcpu *vcpu)
67 {
68 	ulong msr = kvmppc_get_msr(vcpu);
69 	return (msr & (MSR_IR|MSR_DR)) == MSR_DR;
70 }
71 
72 static void kvmppc_fixup_split_real(struct kvm_vcpu *vcpu)
73 {
74 	ulong msr = kvmppc_get_msr(vcpu);
75 	ulong pc = kvmppc_get_pc(vcpu);
76 
77 	/* We are in DR only split real mode */
78 	if ((msr & (MSR_IR|MSR_DR)) != MSR_DR)
79 		return;
80 
81 	/* We have not fixed up the guest already */
82 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK)
83 		return;
84 
85 	/* The code is in fixupable address space */
86 	if (pc & SPLIT_HACK_MASK)
87 		return;
88 
89 	vcpu->arch.hflags |= BOOK3S_HFLAG_SPLIT_HACK;
90 	kvmppc_set_pc(vcpu, pc | SPLIT_HACK_OFFS);
91 }
92 
93 static void kvmppc_unfixup_split_real(struct kvm_vcpu *vcpu)
94 {
95 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK) {
96 		ulong pc = kvmppc_get_pc(vcpu);
97 		ulong lr = kvmppc_get_lr(vcpu);
98 		if ((pc & SPLIT_HACK_MASK) == SPLIT_HACK_OFFS)
99 			kvmppc_set_pc(vcpu, pc & ~SPLIT_HACK_MASK);
100 		if ((lr & SPLIT_HACK_MASK) == SPLIT_HACK_OFFS)
101 			kvmppc_set_lr(vcpu, lr & ~SPLIT_HACK_MASK);
102 		vcpu->arch.hflags &= ~BOOK3S_HFLAG_SPLIT_HACK;
103 	}
104 }
105 
106 static void kvmppc_inject_interrupt_pr(struct kvm_vcpu *vcpu, int vec, u64 srr1_flags)
107 {
108 	unsigned long msr, pc, new_msr, new_pc;
109 
110 	kvmppc_unfixup_split_real(vcpu);
111 
112 	msr = kvmppc_get_msr(vcpu);
113 	pc = kvmppc_get_pc(vcpu);
114 	new_msr = vcpu->arch.intr_msr;
115 	new_pc = to_book3s(vcpu)->hior + vec;
116 
117 #ifdef CONFIG_PPC_BOOK3S_64
118 	/* If transactional, change to suspend mode on IRQ delivery */
119 	if (MSR_TM_TRANSACTIONAL(msr))
120 		new_msr |= MSR_TS_S;
121 	else
122 		new_msr |= msr & MSR_TS_MASK;
123 #endif
124 
125 	kvmppc_set_srr0(vcpu, pc);
126 	kvmppc_set_srr1(vcpu, (msr & SRR1_MSR_BITS) | srr1_flags);
127 	kvmppc_set_pc(vcpu, new_pc);
128 	kvmppc_set_msr(vcpu, new_msr);
129 }
130 
131 static void kvmppc_core_vcpu_load_pr(struct kvm_vcpu *vcpu, int cpu)
132 {
133 #ifdef CONFIG_PPC_BOOK3S_64
134 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
135 	memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb));
136 	svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max;
137 	svcpu->in_use = 0;
138 	svcpu_put(svcpu);
139 #endif
140 
141 	/* Disable AIL if supported */
142 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
143 	    cpu_has_feature(CPU_FTR_ARCH_207S))
144 		mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) & ~LPCR_AIL);
145 
146 	vcpu->cpu = smp_processor_id();
147 #ifdef CONFIG_PPC_BOOK3S_32
148 	current->thread.kvm_shadow_vcpu = vcpu->arch.shadow_vcpu;
149 #endif
150 
151 	if (kvmppc_is_split_real(vcpu))
152 		kvmppc_fixup_split_real(vcpu);
153 
154 	kvmppc_restore_tm_pr(vcpu);
155 }
156 
157 static void kvmppc_core_vcpu_put_pr(struct kvm_vcpu *vcpu)
158 {
159 #ifdef CONFIG_PPC_BOOK3S_64
160 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
161 	if (svcpu->in_use) {
162 		kvmppc_copy_from_svcpu(vcpu);
163 	}
164 	memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
165 	to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max;
166 	svcpu_put(svcpu);
167 #endif
168 
169 	if (kvmppc_is_split_real(vcpu))
170 		kvmppc_unfixup_split_real(vcpu);
171 
172 	kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
173 	kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
174 	kvmppc_save_tm_pr(vcpu);
175 
176 	/* Enable AIL if supported */
177 	if (cpu_has_feature(CPU_FTR_HVMODE) &&
178 	    cpu_has_feature(CPU_FTR_ARCH_207S))
179 		mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_AIL_3);
180 
181 	vcpu->cpu = -1;
182 }
183 
184 /* Copy data needed by real-mode code from vcpu to shadow vcpu */
185 void kvmppc_copy_to_svcpu(struct kvm_vcpu *vcpu)
186 {
187 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
188 
189 	svcpu->gpr[0] = vcpu->arch.regs.gpr[0];
190 	svcpu->gpr[1] = vcpu->arch.regs.gpr[1];
191 	svcpu->gpr[2] = vcpu->arch.regs.gpr[2];
192 	svcpu->gpr[3] = vcpu->arch.regs.gpr[3];
193 	svcpu->gpr[4] = vcpu->arch.regs.gpr[4];
194 	svcpu->gpr[5] = vcpu->arch.regs.gpr[5];
195 	svcpu->gpr[6] = vcpu->arch.regs.gpr[6];
196 	svcpu->gpr[7] = vcpu->arch.regs.gpr[7];
197 	svcpu->gpr[8] = vcpu->arch.regs.gpr[8];
198 	svcpu->gpr[9] = vcpu->arch.regs.gpr[9];
199 	svcpu->gpr[10] = vcpu->arch.regs.gpr[10];
200 	svcpu->gpr[11] = vcpu->arch.regs.gpr[11];
201 	svcpu->gpr[12] = vcpu->arch.regs.gpr[12];
202 	svcpu->gpr[13] = vcpu->arch.regs.gpr[13];
203 	svcpu->cr  = vcpu->arch.regs.ccr;
204 	svcpu->xer = vcpu->arch.regs.xer;
205 	svcpu->ctr = vcpu->arch.regs.ctr;
206 	svcpu->lr  = vcpu->arch.regs.link;
207 	svcpu->pc  = vcpu->arch.regs.nip;
208 #ifdef CONFIG_PPC_BOOK3S_64
209 	svcpu->shadow_fscr = vcpu->arch.shadow_fscr;
210 #endif
211 	/*
212 	 * Now also save the current time base value. We use this
213 	 * to find the guest purr and spurr value.
214 	 */
215 	vcpu->arch.entry_tb = get_tb();
216 	vcpu->arch.entry_vtb = get_vtb();
217 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
218 		vcpu->arch.entry_ic = mfspr(SPRN_IC);
219 	svcpu->in_use = true;
220 
221 	svcpu_put(svcpu);
222 }
223 
224 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
225 {
226 	ulong guest_msr = kvmppc_get_msr(vcpu);
227 	ulong smsr = guest_msr;
228 
229 	/* Guest MSR values */
230 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
231 	smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_LE |
232 		MSR_TM | MSR_TS_MASK;
233 #else
234 	smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_LE;
235 #endif
236 	/* Process MSR values */
237 	smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
238 	/* External providers the guest reserved */
239 	smsr |= (guest_msr & vcpu->arch.guest_owned_ext);
240 	/* 64-bit Process MSR values */
241 #ifdef CONFIG_PPC_BOOK3S_64
242 	smsr |= MSR_HV;
243 #endif
244 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
245 	/*
246 	 * in guest privileged state, we want to fail all TM transactions.
247 	 * So disable MSR TM bit so that all tbegin. will be able to be
248 	 * trapped into host.
249 	 */
250 	if (!(guest_msr & MSR_PR))
251 		smsr &= ~MSR_TM;
252 #endif
253 	vcpu->arch.shadow_msr = smsr;
254 }
255 
256 /* Copy data touched by real-mode code from shadow vcpu back to vcpu */
257 void kvmppc_copy_from_svcpu(struct kvm_vcpu *vcpu)
258 {
259 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
260 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
261 	ulong old_msr;
262 #endif
263 
264 	/*
265 	 * Maybe we were already preempted and synced the svcpu from
266 	 * our preempt notifiers. Don't bother touching this svcpu then.
267 	 */
268 	if (!svcpu->in_use)
269 		goto out;
270 
271 	vcpu->arch.regs.gpr[0] = svcpu->gpr[0];
272 	vcpu->arch.regs.gpr[1] = svcpu->gpr[1];
273 	vcpu->arch.regs.gpr[2] = svcpu->gpr[2];
274 	vcpu->arch.regs.gpr[3] = svcpu->gpr[3];
275 	vcpu->arch.regs.gpr[4] = svcpu->gpr[4];
276 	vcpu->arch.regs.gpr[5] = svcpu->gpr[5];
277 	vcpu->arch.regs.gpr[6] = svcpu->gpr[6];
278 	vcpu->arch.regs.gpr[7] = svcpu->gpr[7];
279 	vcpu->arch.regs.gpr[8] = svcpu->gpr[8];
280 	vcpu->arch.regs.gpr[9] = svcpu->gpr[9];
281 	vcpu->arch.regs.gpr[10] = svcpu->gpr[10];
282 	vcpu->arch.regs.gpr[11] = svcpu->gpr[11];
283 	vcpu->arch.regs.gpr[12] = svcpu->gpr[12];
284 	vcpu->arch.regs.gpr[13] = svcpu->gpr[13];
285 	vcpu->arch.regs.ccr  = svcpu->cr;
286 	vcpu->arch.regs.xer = svcpu->xer;
287 	vcpu->arch.regs.ctr = svcpu->ctr;
288 	vcpu->arch.regs.link  = svcpu->lr;
289 	vcpu->arch.regs.nip  = svcpu->pc;
290 	vcpu->arch.shadow_srr1 = svcpu->shadow_srr1;
291 	vcpu->arch.fault_dar   = svcpu->fault_dar;
292 	vcpu->arch.fault_dsisr = svcpu->fault_dsisr;
293 	vcpu->arch.last_inst   = svcpu->last_inst;
294 #ifdef CONFIG_PPC_BOOK3S_64
295 	vcpu->arch.shadow_fscr = svcpu->shadow_fscr;
296 #endif
297 	/*
298 	 * Update purr and spurr using time base on exit.
299 	 */
300 	vcpu->arch.purr += get_tb() - vcpu->arch.entry_tb;
301 	vcpu->arch.spurr += get_tb() - vcpu->arch.entry_tb;
302 	to_book3s(vcpu)->vtb += get_vtb() - vcpu->arch.entry_vtb;
303 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
304 		vcpu->arch.ic += mfspr(SPRN_IC) - vcpu->arch.entry_ic;
305 
306 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
307 	/*
308 	 * Unlike other MSR bits, MSR[TS]bits can be changed at guest without
309 	 * notifying host:
310 	 *  modified by unprivileged instructions like "tbegin"/"tend"/
311 	 * "tresume"/"tsuspend" in PR KVM guest.
312 	 *
313 	 * It is necessary to sync here to calculate a correct shadow_msr.
314 	 *
315 	 * privileged guest's tbegin will be failed at present. So we
316 	 * only take care of problem state guest.
317 	 */
318 	old_msr = kvmppc_get_msr(vcpu);
319 	if (unlikely((old_msr & MSR_PR) &&
320 		(vcpu->arch.shadow_srr1 & (MSR_TS_MASK)) !=
321 				(old_msr & (MSR_TS_MASK)))) {
322 		old_msr &= ~(MSR_TS_MASK);
323 		old_msr |= (vcpu->arch.shadow_srr1 & (MSR_TS_MASK));
324 		kvmppc_set_msr_fast(vcpu, old_msr);
325 		kvmppc_recalc_shadow_msr(vcpu);
326 	}
327 #endif
328 
329 	svcpu->in_use = false;
330 
331 out:
332 	svcpu_put(svcpu);
333 }
334 
335 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
336 void kvmppc_save_tm_sprs(struct kvm_vcpu *vcpu)
337 {
338 	tm_enable();
339 	vcpu->arch.tfhar = mfspr(SPRN_TFHAR);
340 	vcpu->arch.texasr = mfspr(SPRN_TEXASR);
341 	vcpu->arch.tfiar = mfspr(SPRN_TFIAR);
342 	tm_disable();
343 }
344 
345 void kvmppc_restore_tm_sprs(struct kvm_vcpu *vcpu)
346 {
347 	tm_enable();
348 	mtspr(SPRN_TFHAR, vcpu->arch.tfhar);
349 	mtspr(SPRN_TEXASR, vcpu->arch.texasr);
350 	mtspr(SPRN_TFIAR, vcpu->arch.tfiar);
351 	tm_disable();
352 }
353 
354 /* loadup math bits which is enabled at kvmppc_get_msr() but not enabled at
355  * hardware.
356  */
357 static void kvmppc_handle_lost_math_exts(struct kvm_vcpu *vcpu)
358 {
359 	ulong exit_nr;
360 	ulong ext_diff = (kvmppc_get_msr(vcpu) & ~vcpu->arch.guest_owned_ext) &
361 		(MSR_FP | MSR_VEC | MSR_VSX);
362 
363 	if (!ext_diff)
364 		return;
365 
366 	if (ext_diff == MSR_FP)
367 		exit_nr = BOOK3S_INTERRUPT_FP_UNAVAIL;
368 	else if (ext_diff == MSR_VEC)
369 		exit_nr = BOOK3S_INTERRUPT_ALTIVEC;
370 	else
371 		exit_nr = BOOK3S_INTERRUPT_VSX;
372 
373 	kvmppc_handle_ext(vcpu, exit_nr, ext_diff);
374 }
375 
376 void kvmppc_save_tm_pr(struct kvm_vcpu *vcpu)
377 {
378 	if (!(MSR_TM_ACTIVE(kvmppc_get_msr(vcpu)))) {
379 		kvmppc_save_tm_sprs(vcpu);
380 		return;
381 	}
382 
383 	kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
384 	kvmppc_giveup_ext(vcpu, MSR_VSX);
385 
386 	preempt_disable();
387 	_kvmppc_save_tm_pr(vcpu, mfmsr());
388 	preempt_enable();
389 }
390 
391 void kvmppc_restore_tm_pr(struct kvm_vcpu *vcpu)
392 {
393 	if (!MSR_TM_ACTIVE(kvmppc_get_msr(vcpu))) {
394 		kvmppc_restore_tm_sprs(vcpu);
395 		if (kvmppc_get_msr(vcpu) & MSR_TM) {
396 			kvmppc_handle_lost_math_exts(vcpu);
397 			if (vcpu->arch.fscr & FSCR_TAR)
398 				kvmppc_handle_fac(vcpu, FSCR_TAR_LG);
399 		}
400 		return;
401 	}
402 
403 	preempt_disable();
404 	_kvmppc_restore_tm_pr(vcpu, kvmppc_get_msr(vcpu));
405 	preempt_enable();
406 
407 	if (kvmppc_get_msr(vcpu) & MSR_TM) {
408 		kvmppc_handle_lost_math_exts(vcpu);
409 		if (vcpu->arch.fscr & FSCR_TAR)
410 			kvmppc_handle_fac(vcpu, FSCR_TAR_LG);
411 	}
412 }
413 #endif
414 
415 static int kvmppc_core_check_requests_pr(struct kvm_vcpu *vcpu)
416 {
417 	int r = 1; /* Indicate we want to get back into the guest */
418 
419 	/* We misuse TLB_FLUSH to indicate that we want to clear
420 	   all shadow cache entries */
421 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
422 		kvmppc_mmu_pte_flush(vcpu, 0, 0);
423 
424 	return r;
425 }
426 
427 /************* MMU Notifiers *************/
428 static bool do_kvm_unmap_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
429 {
430 	long i;
431 	struct kvm_vcpu *vcpu;
432 
433 	kvm_for_each_vcpu(i, vcpu, kvm)
434 		kvmppc_mmu_pte_pflush(vcpu, range->start << PAGE_SHIFT,
435 				      range->end << PAGE_SHIFT);
436 
437 	return false;
438 }
439 
440 static bool kvm_unmap_gfn_range_pr(struct kvm *kvm, struct kvm_gfn_range *range)
441 {
442 	return do_kvm_unmap_gfn(kvm, range);
443 }
444 
445 static bool kvm_age_gfn_pr(struct kvm *kvm, struct kvm_gfn_range *range)
446 {
447 	/* XXX could be more clever ;) */
448 	return false;
449 }
450 
451 static bool kvm_test_age_gfn_pr(struct kvm *kvm, struct kvm_gfn_range *range)
452 {
453 	/* XXX could be more clever ;) */
454 	return false;
455 }
456 
457 static bool kvm_set_spte_gfn_pr(struct kvm *kvm, struct kvm_gfn_range *range)
458 {
459 	/* The page will get remapped properly on its next fault */
460 	return do_kvm_unmap_gfn(kvm, range);
461 }
462 
463 /*****************************************/
464 
465 static void kvmppc_set_msr_pr(struct kvm_vcpu *vcpu, u64 msr)
466 {
467 	ulong old_msr;
468 
469 	/* For PAPR guest, make sure MSR reflects guest mode */
470 	if (vcpu->arch.papr_enabled)
471 		msr = (msr & ~MSR_HV) | MSR_ME;
472 
473 #ifdef EXIT_DEBUG
474 	printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
475 #endif
476 
477 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
478 	/* We should never target guest MSR to TS=10 && PR=0,
479 	 * since we always fail transaction for guest privilege
480 	 * state.
481 	 */
482 	if (!(msr & MSR_PR) && MSR_TM_TRANSACTIONAL(msr))
483 		kvmppc_emulate_tabort(vcpu,
484 			TM_CAUSE_KVM_FAC_UNAV | TM_CAUSE_PERSISTENT);
485 #endif
486 
487 	old_msr = kvmppc_get_msr(vcpu);
488 	msr &= to_book3s(vcpu)->msr_mask;
489 	kvmppc_set_msr_fast(vcpu, msr);
490 	kvmppc_recalc_shadow_msr(vcpu);
491 
492 	if (msr & MSR_POW) {
493 		if (!vcpu->arch.pending_exceptions) {
494 			kvm_vcpu_block(vcpu);
495 			kvm_clear_request(KVM_REQ_UNHALT, vcpu);
496 			vcpu->stat.halt_wakeup++;
497 
498 			/* Unset POW bit after we woke up */
499 			msr &= ~MSR_POW;
500 			kvmppc_set_msr_fast(vcpu, msr);
501 		}
502 	}
503 
504 	if (kvmppc_is_split_real(vcpu))
505 		kvmppc_fixup_split_real(vcpu);
506 	else
507 		kvmppc_unfixup_split_real(vcpu);
508 
509 	if ((kvmppc_get_msr(vcpu) & (MSR_PR|MSR_IR|MSR_DR)) !=
510 		   (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
511 		kvmppc_mmu_flush_segments(vcpu);
512 		kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
513 
514 		/* Preload magic page segment when in kernel mode */
515 		if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
516 			struct kvm_vcpu_arch *a = &vcpu->arch;
517 
518 			if (msr & MSR_DR)
519 				kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
520 			else
521 				kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
522 		}
523 	}
524 
525 	/*
526 	 * When switching from 32 to 64-bit, we may have a stale 32-bit
527 	 * magic page around, we need to flush it. Typically 32-bit magic
528 	 * page will be instantiated when calling into RTAS. Note: We
529 	 * assume that such transition only happens while in kernel mode,
530 	 * ie, we never transition from user 32-bit to kernel 64-bit with
531 	 * a 32-bit magic page around.
532 	 */
533 	if (vcpu->arch.magic_page_pa &&
534 	    !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
535 		/* going from RTAS to normal kernel code */
536 		kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
537 				     ~0xFFFUL);
538 	}
539 
540 	/* Preload FPU if it's enabled */
541 	if (kvmppc_get_msr(vcpu) & MSR_FP)
542 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
543 
544 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
545 	if (kvmppc_get_msr(vcpu) & MSR_TM)
546 		kvmppc_handle_lost_math_exts(vcpu);
547 #endif
548 }
549 
550 static void kvmppc_set_pvr_pr(struct kvm_vcpu *vcpu, u32 pvr)
551 {
552 	u32 host_pvr;
553 
554 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
555 	vcpu->arch.pvr = pvr;
556 #ifdef CONFIG_PPC_BOOK3S_64
557 	if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
558 		kvmppc_mmu_book3s_64_init(vcpu);
559 		if (!to_book3s(vcpu)->hior_explicit)
560 			to_book3s(vcpu)->hior = 0xfff00000;
561 		to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
562 		vcpu->arch.cpu_type = KVM_CPU_3S_64;
563 	} else
564 #endif
565 	{
566 		kvmppc_mmu_book3s_32_init(vcpu);
567 		if (!to_book3s(vcpu)->hior_explicit)
568 			to_book3s(vcpu)->hior = 0;
569 		to_book3s(vcpu)->msr_mask = 0xffffffffULL;
570 		vcpu->arch.cpu_type = KVM_CPU_3S_32;
571 	}
572 
573 	kvmppc_sanity_check(vcpu);
574 
575 	/* If we are in hypervisor level on 970, we can tell the CPU to
576 	 * treat DCBZ as 32 bytes store */
577 	vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
578 	if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
579 	    !strcmp(cur_cpu_spec->platform, "ppc970"))
580 		vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
581 
582 	/* Cell performs badly if MSR_FEx are set. So let's hope nobody
583 	   really needs them in a VM on Cell and force disable them. */
584 	if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
585 		to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
586 
587 	/*
588 	 * If they're asking for POWER6 or later, set the flag
589 	 * indicating that we can do multiple large page sizes
590 	 * and 1TB segments.
591 	 * Also set the flag that indicates that tlbie has the large
592 	 * page bit in the RB operand instead of the instruction.
593 	 */
594 	switch (PVR_VER(pvr)) {
595 	case PVR_POWER6:
596 	case PVR_POWER7:
597 	case PVR_POWER7p:
598 	case PVR_POWER8:
599 	case PVR_POWER8E:
600 	case PVR_POWER8NVL:
601 	case PVR_POWER9:
602 		vcpu->arch.hflags |= BOOK3S_HFLAG_MULTI_PGSIZE |
603 			BOOK3S_HFLAG_NEW_TLBIE;
604 		break;
605 	}
606 
607 #ifdef CONFIG_PPC_BOOK3S_32
608 	/* 32 bit Book3S always has 32 byte dcbz */
609 	vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
610 #endif
611 
612 	/* On some CPUs we can execute paired single operations natively */
613 	asm ( "mfpvr %0" : "=r"(host_pvr));
614 	switch (host_pvr) {
615 	case 0x00080200:	/* lonestar 2.0 */
616 	case 0x00088202:	/* lonestar 2.2 */
617 	case 0x70000100:	/* gekko 1.0 */
618 	case 0x00080100:	/* gekko 2.0 */
619 	case 0x00083203:	/* gekko 2.3a */
620 	case 0x00083213:	/* gekko 2.3b */
621 	case 0x00083204:	/* gekko 2.4 */
622 	case 0x00083214:	/* gekko 2.4e (8SE) - retail HW2 */
623 	case 0x00087200:	/* broadway */
624 		vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
625 		/* Enable HID2.PSE - in case we need it later */
626 		mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
627 	}
628 }
629 
630 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
631  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
632  * emulate 32 bytes dcbz length.
633  *
634  * The Book3s_64 inventors also realized this case and implemented a special bit
635  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
636  *
637  * My approach here is to patch the dcbz instruction on executing pages.
638  */
639 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
640 {
641 	struct page *hpage;
642 	u64 hpage_offset;
643 	u32 *page;
644 	int i;
645 
646 	hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
647 	if (is_error_page(hpage))
648 		return;
649 
650 	hpage_offset = pte->raddr & ~PAGE_MASK;
651 	hpage_offset &= ~0xFFFULL;
652 	hpage_offset /= 4;
653 
654 	get_page(hpage);
655 	page = kmap_atomic(hpage);
656 
657 	/* patch dcbz into reserved instruction, so we trap */
658 	for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
659 		if ((be32_to_cpu(page[i]) & 0xff0007ff) == INS_DCBZ)
660 			page[i] &= cpu_to_be32(0xfffffff7);
661 
662 	kunmap_atomic(page);
663 	put_page(hpage);
664 }
665 
666 static bool kvmppc_visible_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
667 {
668 	ulong mp_pa = vcpu->arch.magic_page_pa;
669 
670 	if (!(kvmppc_get_msr(vcpu) & MSR_SF))
671 		mp_pa = (uint32_t)mp_pa;
672 
673 	gpa &= ~0xFFFULL;
674 	if (unlikely(mp_pa) && unlikely((mp_pa & KVM_PAM) == (gpa & KVM_PAM))) {
675 		return true;
676 	}
677 
678 	return kvm_is_visible_gfn(vcpu->kvm, gpa >> PAGE_SHIFT);
679 }
680 
681 static int kvmppc_handle_pagefault(struct kvm_vcpu *vcpu,
682 			    ulong eaddr, int vec)
683 {
684 	bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
685 	bool iswrite = false;
686 	int r = RESUME_GUEST;
687 	int relocated;
688 	int page_found = 0;
689 	struct kvmppc_pte pte = { 0 };
690 	bool dr = (kvmppc_get_msr(vcpu) & MSR_DR) ? true : false;
691 	bool ir = (kvmppc_get_msr(vcpu) & MSR_IR) ? true : false;
692 	u64 vsid;
693 
694 	relocated = data ? dr : ir;
695 	if (data && (vcpu->arch.fault_dsisr & DSISR_ISSTORE))
696 		iswrite = true;
697 
698 	/* Resolve real address if translation turned on */
699 	if (relocated) {
700 		page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data, iswrite);
701 	} else {
702 		pte.may_execute = true;
703 		pte.may_read = true;
704 		pte.may_write = true;
705 		pte.raddr = eaddr & KVM_PAM;
706 		pte.eaddr = eaddr;
707 		pte.vpage = eaddr >> 12;
708 		pte.page_size = MMU_PAGE_64K;
709 		pte.wimg = HPTE_R_M;
710 	}
711 
712 	switch (kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) {
713 	case 0:
714 		pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
715 		break;
716 	case MSR_DR:
717 		if (!data &&
718 		    (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK) &&
719 		    ((pte.raddr & SPLIT_HACK_MASK) == SPLIT_HACK_OFFS))
720 			pte.raddr &= ~SPLIT_HACK_MASK;
721 		fallthrough;
722 	case MSR_IR:
723 		vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
724 
725 		if ((kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) == MSR_DR)
726 			pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
727 		else
728 			pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
729 		pte.vpage |= vsid;
730 
731 		if (vsid == -1)
732 			page_found = -EINVAL;
733 		break;
734 	}
735 
736 	if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
737 	   (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
738 		/*
739 		 * If we do the dcbz hack, we have to NX on every execution,
740 		 * so we can patch the executing code. This renders our guest
741 		 * NX-less.
742 		 */
743 		pte.may_execute = !data;
744 	}
745 
746 	if (page_found == -ENOENT || page_found == -EPERM) {
747 		/* Page not found in guest PTE entries, or protection fault */
748 		u64 flags;
749 
750 		if (page_found == -EPERM)
751 			flags = DSISR_PROTFAULT;
752 		else
753 			flags = DSISR_NOHPTE;
754 		if (data) {
755 			flags |= vcpu->arch.fault_dsisr & DSISR_ISSTORE;
756 			kvmppc_core_queue_data_storage(vcpu, eaddr, flags);
757 		} else {
758 			kvmppc_core_queue_inst_storage(vcpu, flags);
759 		}
760 	} else if (page_found == -EINVAL) {
761 		/* Page not found in guest SLB */
762 		kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
763 		kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
764 	} else if (kvmppc_visible_gpa(vcpu, pte.raddr)) {
765 		if (data && !(vcpu->arch.fault_dsisr & DSISR_NOHPTE)) {
766 			/*
767 			 * There is already a host HPTE there, presumably
768 			 * a read-only one for a page the guest thinks
769 			 * is writable, so get rid of it first.
770 			 */
771 			kvmppc_mmu_unmap_page(vcpu, &pte);
772 		}
773 		/* The guest's PTE is not mapped yet. Map on the host */
774 		if (kvmppc_mmu_map_page(vcpu, &pte, iswrite) == -EIO) {
775 			/* Exit KVM if mapping failed */
776 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
777 			return RESUME_HOST;
778 		}
779 		if (data)
780 			vcpu->stat.sp_storage++;
781 		else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
782 			 (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
783 			kvmppc_patch_dcbz(vcpu, &pte);
784 	} else {
785 		/* MMIO */
786 		vcpu->stat.mmio_exits++;
787 		vcpu->arch.paddr_accessed = pte.raddr;
788 		vcpu->arch.vaddr_accessed = pte.eaddr;
789 		r = kvmppc_emulate_mmio(vcpu);
790 		if ( r == RESUME_HOST_NV )
791 			r = RESUME_HOST;
792 	}
793 
794 	return r;
795 }
796 
797 /* Give up external provider (FPU, Altivec, VSX) */
798 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
799 {
800 	struct thread_struct *t = &current->thread;
801 
802 	/*
803 	 * VSX instructions can access FP and vector registers, so if
804 	 * we are giving up VSX, make sure we give up FP and VMX as well.
805 	 */
806 	if (msr & MSR_VSX)
807 		msr |= MSR_FP | MSR_VEC;
808 
809 	msr &= vcpu->arch.guest_owned_ext;
810 	if (!msr)
811 		return;
812 
813 #ifdef DEBUG_EXT
814 	printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
815 #endif
816 
817 	if (msr & MSR_FP) {
818 		/*
819 		 * Note that on CPUs with VSX, giveup_fpu stores
820 		 * both the traditional FP registers and the added VSX
821 		 * registers into thread.fp_state.fpr[].
822 		 */
823 		if (t->regs->msr & MSR_FP)
824 			giveup_fpu(current);
825 		t->fp_save_area = NULL;
826 	}
827 
828 #ifdef CONFIG_ALTIVEC
829 	if (msr & MSR_VEC) {
830 		if (current->thread.regs->msr & MSR_VEC)
831 			giveup_altivec(current);
832 		t->vr_save_area = NULL;
833 	}
834 #endif
835 
836 	vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
837 	kvmppc_recalc_shadow_msr(vcpu);
838 }
839 
840 /* Give up facility (TAR / EBB / DSCR) */
841 void kvmppc_giveup_fac(struct kvm_vcpu *vcpu, ulong fac)
842 {
843 #ifdef CONFIG_PPC_BOOK3S_64
844 	if (!(vcpu->arch.shadow_fscr & (1ULL << fac))) {
845 		/* Facility not available to the guest, ignore giveup request*/
846 		return;
847 	}
848 
849 	switch (fac) {
850 	case FSCR_TAR_LG:
851 		vcpu->arch.tar = mfspr(SPRN_TAR);
852 		mtspr(SPRN_TAR, current->thread.tar);
853 		vcpu->arch.shadow_fscr &= ~FSCR_TAR;
854 		break;
855 	}
856 #endif
857 }
858 
859 /* Handle external providers (FPU, Altivec, VSX) */
860 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
861 			     ulong msr)
862 {
863 	struct thread_struct *t = &current->thread;
864 
865 	/* When we have paired singles, we emulate in software */
866 	if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
867 		return RESUME_GUEST;
868 
869 	if (!(kvmppc_get_msr(vcpu) & msr)) {
870 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
871 		return RESUME_GUEST;
872 	}
873 
874 	if (msr == MSR_VSX) {
875 		/* No VSX?  Give an illegal instruction interrupt */
876 #ifdef CONFIG_VSX
877 		if (!cpu_has_feature(CPU_FTR_VSX))
878 #endif
879 		{
880 			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
881 			return RESUME_GUEST;
882 		}
883 
884 		/*
885 		 * We have to load up all the FP and VMX registers before
886 		 * we can let the guest use VSX instructions.
887 		 */
888 		msr = MSR_FP | MSR_VEC | MSR_VSX;
889 	}
890 
891 	/* See if we already own all the ext(s) needed */
892 	msr &= ~vcpu->arch.guest_owned_ext;
893 	if (!msr)
894 		return RESUME_GUEST;
895 
896 #ifdef DEBUG_EXT
897 	printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
898 #endif
899 
900 	if (msr & MSR_FP) {
901 		preempt_disable();
902 		enable_kernel_fp();
903 		load_fp_state(&vcpu->arch.fp);
904 		disable_kernel_fp();
905 		t->fp_save_area = &vcpu->arch.fp;
906 		preempt_enable();
907 	}
908 
909 	if (msr & MSR_VEC) {
910 #ifdef CONFIG_ALTIVEC
911 		preempt_disable();
912 		enable_kernel_altivec();
913 		load_vr_state(&vcpu->arch.vr);
914 		disable_kernel_altivec();
915 		t->vr_save_area = &vcpu->arch.vr;
916 		preempt_enable();
917 #endif
918 	}
919 
920 	t->regs->msr |= msr;
921 	vcpu->arch.guest_owned_ext |= msr;
922 	kvmppc_recalc_shadow_msr(vcpu);
923 
924 	return RESUME_GUEST;
925 }
926 
927 /*
928  * Kernel code using FP or VMX could have flushed guest state to
929  * the thread_struct; if so, get it back now.
930  */
931 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
932 {
933 	unsigned long lost_ext;
934 
935 	lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
936 	if (!lost_ext)
937 		return;
938 
939 	if (lost_ext & MSR_FP) {
940 		preempt_disable();
941 		enable_kernel_fp();
942 		load_fp_state(&vcpu->arch.fp);
943 		disable_kernel_fp();
944 		preempt_enable();
945 	}
946 #ifdef CONFIG_ALTIVEC
947 	if (lost_ext & MSR_VEC) {
948 		preempt_disable();
949 		enable_kernel_altivec();
950 		load_vr_state(&vcpu->arch.vr);
951 		disable_kernel_altivec();
952 		preempt_enable();
953 	}
954 #endif
955 	current->thread.regs->msr |= lost_ext;
956 }
957 
958 #ifdef CONFIG_PPC_BOOK3S_64
959 
960 void kvmppc_trigger_fac_interrupt(struct kvm_vcpu *vcpu, ulong fac)
961 {
962 	/* Inject the Interrupt Cause field and trigger a guest interrupt */
963 	vcpu->arch.fscr &= ~(0xffULL << 56);
964 	vcpu->arch.fscr |= (fac << 56);
965 	kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_FAC_UNAVAIL);
966 }
967 
968 static void kvmppc_emulate_fac(struct kvm_vcpu *vcpu, ulong fac)
969 {
970 	enum emulation_result er = EMULATE_FAIL;
971 
972 	if (!(kvmppc_get_msr(vcpu) & MSR_PR))
973 		er = kvmppc_emulate_instruction(vcpu);
974 
975 	if ((er != EMULATE_DONE) && (er != EMULATE_AGAIN)) {
976 		/* Couldn't emulate, trigger interrupt in guest */
977 		kvmppc_trigger_fac_interrupt(vcpu, fac);
978 	}
979 }
980 
981 /* Enable facilities (TAR, EBB, DSCR) for the guest */
982 static int kvmppc_handle_fac(struct kvm_vcpu *vcpu, ulong fac)
983 {
984 	bool guest_fac_enabled;
985 	BUG_ON(!cpu_has_feature(CPU_FTR_ARCH_207S));
986 
987 	/*
988 	 * Not every facility is enabled by FSCR bits, check whether the
989 	 * guest has this facility enabled at all.
990 	 */
991 	switch (fac) {
992 	case FSCR_TAR_LG:
993 	case FSCR_EBB_LG:
994 		guest_fac_enabled = (vcpu->arch.fscr & (1ULL << fac));
995 		break;
996 	case FSCR_TM_LG:
997 		guest_fac_enabled = kvmppc_get_msr(vcpu) & MSR_TM;
998 		break;
999 	default:
1000 		guest_fac_enabled = false;
1001 		break;
1002 	}
1003 
1004 	if (!guest_fac_enabled) {
1005 		/* Facility not enabled by the guest */
1006 		kvmppc_trigger_fac_interrupt(vcpu, fac);
1007 		return RESUME_GUEST;
1008 	}
1009 
1010 	switch (fac) {
1011 	case FSCR_TAR_LG:
1012 		/* TAR switching isn't lazy in Linux yet */
1013 		current->thread.tar = mfspr(SPRN_TAR);
1014 		mtspr(SPRN_TAR, vcpu->arch.tar);
1015 		vcpu->arch.shadow_fscr |= FSCR_TAR;
1016 		break;
1017 	default:
1018 		kvmppc_emulate_fac(vcpu, fac);
1019 		break;
1020 	}
1021 
1022 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1023 	/* Since we disabled MSR_TM at privilege state, the mfspr instruction
1024 	 * for TM spr can trigger TM fac unavailable. In this case, the
1025 	 * emulation is handled by kvmppc_emulate_fac(), which invokes
1026 	 * kvmppc_emulate_mfspr() finally. But note the mfspr can include
1027 	 * RT for NV registers. So it need to restore those NV reg to reflect
1028 	 * the update.
1029 	 */
1030 	if ((fac == FSCR_TM_LG) && !(kvmppc_get_msr(vcpu) & MSR_PR))
1031 		return RESUME_GUEST_NV;
1032 #endif
1033 
1034 	return RESUME_GUEST;
1035 }
1036 
1037 void kvmppc_set_fscr(struct kvm_vcpu *vcpu, u64 fscr)
1038 {
1039 	if ((vcpu->arch.fscr & FSCR_TAR) && !(fscr & FSCR_TAR)) {
1040 		/* TAR got dropped, drop it in shadow too */
1041 		kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
1042 	} else if (!(vcpu->arch.fscr & FSCR_TAR) && (fscr & FSCR_TAR)) {
1043 		vcpu->arch.fscr = fscr;
1044 		kvmppc_handle_fac(vcpu, FSCR_TAR_LG);
1045 		return;
1046 	}
1047 
1048 	vcpu->arch.fscr = fscr;
1049 }
1050 #endif
1051 
1052 static void kvmppc_setup_debug(struct kvm_vcpu *vcpu)
1053 {
1054 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
1055 		u64 msr = kvmppc_get_msr(vcpu);
1056 
1057 		kvmppc_set_msr(vcpu, msr | MSR_SE);
1058 	}
1059 }
1060 
1061 static void kvmppc_clear_debug(struct kvm_vcpu *vcpu)
1062 {
1063 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
1064 		u64 msr = kvmppc_get_msr(vcpu);
1065 
1066 		kvmppc_set_msr(vcpu, msr & ~MSR_SE);
1067 	}
1068 }
1069 
1070 static int kvmppc_exit_pr_progint(struct kvm_vcpu *vcpu, unsigned int exit_nr)
1071 {
1072 	enum emulation_result er;
1073 	ulong flags;
1074 	u32 last_inst;
1075 	int emul, r;
1076 
1077 	/*
1078 	 * shadow_srr1 only contains valid flags if we came here via a program
1079 	 * exception. The other exceptions (emulation assist, FP unavailable,
1080 	 * etc.) do not provide flags in SRR1, so use an illegal-instruction
1081 	 * exception when injecting a program interrupt into the guest.
1082 	 */
1083 	if (exit_nr == BOOK3S_INTERRUPT_PROGRAM)
1084 		flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
1085 	else
1086 		flags = SRR1_PROGILL;
1087 
1088 	emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1089 	if (emul != EMULATE_DONE)
1090 		return RESUME_GUEST;
1091 
1092 	if (kvmppc_get_msr(vcpu) & MSR_PR) {
1093 #ifdef EXIT_DEBUG
1094 		pr_info("Userspace triggered 0x700 exception at\n 0x%lx (0x%x)\n",
1095 			kvmppc_get_pc(vcpu), last_inst);
1096 #endif
1097 		if ((last_inst & 0xff0007ff) != (INS_DCBZ & 0xfffffff7)) {
1098 			kvmppc_core_queue_program(vcpu, flags);
1099 			return RESUME_GUEST;
1100 		}
1101 	}
1102 
1103 	vcpu->stat.emulated_inst_exits++;
1104 	er = kvmppc_emulate_instruction(vcpu);
1105 	switch (er) {
1106 	case EMULATE_DONE:
1107 		r = RESUME_GUEST_NV;
1108 		break;
1109 	case EMULATE_AGAIN:
1110 		r = RESUME_GUEST;
1111 		break;
1112 	case EMULATE_FAIL:
1113 		pr_crit("%s: emulation at %lx failed (%08x)\n",
1114 			__func__, kvmppc_get_pc(vcpu), last_inst);
1115 		kvmppc_core_queue_program(vcpu, flags);
1116 		r = RESUME_GUEST;
1117 		break;
1118 	case EMULATE_DO_MMIO:
1119 		vcpu->run->exit_reason = KVM_EXIT_MMIO;
1120 		r = RESUME_HOST_NV;
1121 		break;
1122 	case EMULATE_EXIT_USER:
1123 		r = RESUME_HOST_NV;
1124 		break;
1125 	default:
1126 		BUG();
1127 	}
1128 
1129 	return r;
1130 }
1131 
1132 int kvmppc_handle_exit_pr(struct kvm_vcpu *vcpu, unsigned int exit_nr)
1133 {
1134 	struct kvm_run *run = vcpu->run;
1135 	int r = RESUME_HOST;
1136 	int s;
1137 
1138 	vcpu->stat.sum_exits++;
1139 
1140 	run->exit_reason = KVM_EXIT_UNKNOWN;
1141 	run->ready_for_interrupt_injection = 1;
1142 
1143 	/* We get here with MSR.EE=1 */
1144 
1145 	trace_kvm_exit(exit_nr, vcpu);
1146 	guest_exit();
1147 
1148 	switch (exit_nr) {
1149 	case BOOK3S_INTERRUPT_INST_STORAGE:
1150 	{
1151 		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1152 		vcpu->stat.pf_instruc++;
1153 
1154 		if (kvmppc_is_split_real(vcpu))
1155 			kvmppc_fixup_split_real(vcpu);
1156 
1157 #ifdef CONFIG_PPC_BOOK3S_32
1158 		/* We set segments as unused segments when invalidating them. So
1159 		 * treat the respective fault as segment fault. */
1160 		{
1161 			struct kvmppc_book3s_shadow_vcpu *svcpu;
1162 			u32 sr;
1163 
1164 			svcpu = svcpu_get(vcpu);
1165 			sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
1166 			svcpu_put(svcpu);
1167 			if (sr == SR_INVALID) {
1168 				kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
1169 				r = RESUME_GUEST;
1170 				break;
1171 			}
1172 		}
1173 #endif
1174 
1175 		/* only care about PTEG not found errors, but leave NX alone */
1176 		if (shadow_srr1 & 0x40000000) {
1177 			int idx = srcu_read_lock(&vcpu->kvm->srcu);
1178 			r = kvmppc_handle_pagefault(vcpu, kvmppc_get_pc(vcpu), exit_nr);
1179 			srcu_read_unlock(&vcpu->kvm->srcu, idx);
1180 			vcpu->stat.sp_instruc++;
1181 		} else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
1182 			  (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
1183 			/*
1184 			 * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
1185 			 *     so we can't use the NX bit inside the guest. Let's cross our fingers,
1186 			 *     that no guest that needs the dcbz hack does NX.
1187 			 */
1188 			kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
1189 			r = RESUME_GUEST;
1190 		} else {
1191 			kvmppc_core_queue_inst_storage(vcpu,
1192 						shadow_srr1 & 0x58000000);
1193 			r = RESUME_GUEST;
1194 		}
1195 		break;
1196 	}
1197 	case BOOK3S_INTERRUPT_DATA_STORAGE:
1198 	{
1199 		ulong dar = kvmppc_get_fault_dar(vcpu);
1200 		u32 fault_dsisr = vcpu->arch.fault_dsisr;
1201 		vcpu->stat.pf_storage++;
1202 
1203 #ifdef CONFIG_PPC_BOOK3S_32
1204 		/* We set segments as unused segments when invalidating them. So
1205 		 * treat the respective fault as segment fault. */
1206 		{
1207 			struct kvmppc_book3s_shadow_vcpu *svcpu;
1208 			u32 sr;
1209 
1210 			svcpu = svcpu_get(vcpu);
1211 			sr = svcpu->sr[dar >> SID_SHIFT];
1212 			svcpu_put(svcpu);
1213 			if (sr == SR_INVALID) {
1214 				kvmppc_mmu_map_segment(vcpu, dar);
1215 				r = RESUME_GUEST;
1216 				break;
1217 			}
1218 		}
1219 #endif
1220 
1221 		/*
1222 		 * We need to handle missing shadow PTEs, and
1223 		 * protection faults due to us mapping a page read-only
1224 		 * when the guest thinks it is writable.
1225 		 */
1226 		if (fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT)) {
1227 			int idx = srcu_read_lock(&vcpu->kvm->srcu);
1228 			r = kvmppc_handle_pagefault(vcpu, dar, exit_nr);
1229 			srcu_read_unlock(&vcpu->kvm->srcu, idx);
1230 		} else {
1231 			kvmppc_core_queue_data_storage(vcpu, dar, fault_dsisr);
1232 			r = RESUME_GUEST;
1233 		}
1234 		break;
1235 	}
1236 	case BOOK3S_INTERRUPT_DATA_SEGMENT:
1237 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
1238 			kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
1239 			kvmppc_book3s_queue_irqprio(vcpu,
1240 				BOOK3S_INTERRUPT_DATA_SEGMENT);
1241 		}
1242 		r = RESUME_GUEST;
1243 		break;
1244 	case BOOK3S_INTERRUPT_INST_SEGMENT:
1245 		if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
1246 			kvmppc_book3s_queue_irqprio(vcpu,
1247 				BOOK3S_INTERRUPT_INST_SEGMENT);
1248 		}
1249 		r = RESUME_GUEST;
1250 		break;
1251 	/* We're good on these - the host merely wanted to get our attention */
1252 	case BOOK3S_INTERRUPT_DECREMENTER:
1253 	case BOOK3S_INTERRUPT_HV_DECREMENTER:
1254 	case BOOK3S_INTERRUPT_DOORBELL:
1255 	case BOOK3S_INTERRUPT_H_DOORBELL:
1256 		vcpu->stat.dec_exits++;
1257 		r = RESUME_GUEST;
1258 		break;
1259 	case BOOK3S_INTERRUPT_EXTERNAL:
1260 	case BOOK3S_INTERRUPT_EXTERNAL_HV:
1261 	case BOOK3S_INTERRUPT_H_VIRT:
1262 		vcpu->stat.ext_intr_exits++;
1263 		r = RESUME_GUEST;
1264 		break;
1265 	case BOOK3S_INTERRUPT_HMI:
1266 	case BOOK3S_INTERRUPT_PERFMON:
1267 	case BOOK3S_INTERRUPT_SYSTEM_RESET:
1268 		r = RESUME_GUEST;
1269 		break;
1270 	case BOOK3S_INTERRUPT_PROGRAM:
1271 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1272 		r = kvmppc_exit_pr_progint(vcpu, exit_nr);
1273 		break;
1274 	case BOOK3S_INTERRUPT_SYSCALL:
1275 	{
1276 		u32 last_sc;
1277 		int emul;
1278 
1279 		/* Get last sc for papr */
1280 		if (vcpu->arch.papr_enabled) {
1281 			/* The sc instuction points SRR0 to the next inst */
1282 			emul = kvmppc_get_last_inst(vcpu, INST_SC, &last_sc);
1283 			if (emul != EMULATE_DONE) {
1284 				kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) - 4);
1285 				r = RESUME_GUEST;
1286 				break;
1287 			}
1288 		}
1289 
1290 		if (vcpu->arch.papr_enabled &&
1291 		    (last_sc == 0x44000022) &&
1292 		    !(kvmppc_get_msr(vcpu) & MSR_PR)) {
1293 			/* SC 1 papr hypercalls */
1294 			ulong cmd = kvmppc_get_gpr(vcpu, 3);
1295 			int i;
1296 
1297 #ifdef CONFIG_PPC_BOOK3S_64
1298 			if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
1299 				r = RESUME_GUEST;
1300 				break;
1301 			}
1302 #endif
1303 
1304 			run->papr_hcall.nr = cmd;
1305 			for (i = 0; i < 9; ++i) {
1306 				ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
1307 				run->papr_hcall.args[i] = gpr;
1308 			}
1309 			run->exit_reason = KVM_EXIT_PAPR_HCALL;
1310 			vcpu->arch.hcall_needed = 1;
1311 			r = RESUME_HOST;
1312 		} else if (vcpu->arch.osi_enabled &&
1313 		    (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
1314 		    (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
1315 			/* MOL hypercalls */
1316 			u64 *gprs = run->osi.gprs;
1317 			int i;
1318 
1319 			run->exit_reason = KVM_EXIT_OSI;
1320 			for (i = 0; i < 32; i++)
1321 				gprs[i] = kvmppc_get_gpr(vcpu, i);
1322 			vcpu->arch.osi_needed = 1;
1323 			r = RESUME_HOST_NV;
1324 		} else if (!(kvmppc_get_msr(vcpu) & MSR_PR) &&
1325 		    (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
1326 			/* KVM PV hypercalls */
1327 			kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
1328 			r = RESUME_GUEST;
1329 		} else {
1330 			/* Guest syscalls */
1331 			vcpu->stat.syscall_exits++;
1332 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1333 			r = RESUME_GUEST;
1334 		}
1335 		break;
1336 	}
1337 	case BOOK3S_INTERRUPT_FP_UNAVAIL:
1338 	case BOOK3S_INTERRUPT_ALTIVEC:
1339 	case BOOK3S_INTERRUPT_VSX:
1340 	{
1341 		int ext_msr = 0;
1342 		int emul;
1343 		u32 last_inst;
1344 
1345 		if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE) {
1346 			/* Do paired single instruction emulation */
1347 			emul = kvmppc_get_last_inst(vcpu, INST_GENERIC,
1348 						    &last_inst);
1349 			if (emul == EMULATE_DONE)
1350 				r = kvmppc_exit_pr_progint(vcpu, exit_nr);
1351 			else
1352 				r = RESUME_GUEST;
1353 
1354 			break;
1355 		}
1356 
1357 		/* Enable external provider */
1358 		switch (exit_nr) {
1359 		case BOOK3S_INTERRUPT_FP_UNAVAIL:
1360 			ext_msr = MSR_FP;
1361 			break;
1362 
1363 		case BOOK3S_INTERRUPT_ALTIVEC:
1364 			ext_msr = MSR_VEC;
1365 			break;
1366 
1367 		case BOOK3S_INTERRUPT_VSX:
1368 			ext_msr = MSR_VSX;
1369 			break;
1370 		}
1371 
1372 		r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
1373 		break;
1374 	}
1375 	case BOOK3S_INTERRUPT_ALIGNMENT:
1376 	{
1377 		u32 last_inst;
1378 		int emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1379 
1380 		if (emul == EMULATE_DONE) {
1381 			u32 dsisr;
1382 			u64 dar;
1383 
1384 			dsisr = kvmppc_alignment_dsisr(vcpu, last_inst);
1385 			dar = kvmppc_alignment_dar(vcpu, last_inst);
1386 
1387 			kvmppc_set_dsisr(vcpu, dsisr);
1388 			kvmppc_set_dar(vcpu, dar);
1389 
1390 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1391 		}
1392 		r = RESUME_GUEST;
1393 		break;
1394 	}
1395 #ifdef CONFIG_PPC_BOOK3S_64
1396 	case BOOK3S_INTERRUPT_FAC_UNAVAIL:
1397 		r = kvmppc_handle_fac(vcpu, vcpu->arch.shadow_fscr >> 56);
1398 		break;
1399 #endif
1400 	case BOOK3S_INTERRUPT_MACHINE_CHECK:
1401 		kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1402 		r = RESUME_GUEST;
1403 		break;
1404 	case BOOK3S_INTERRUPT_TRACE:
1405 		if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
1406 			run->exit_reason = KVM_EXIT_DEBUG;
1407 			r = RESUME_HOST;
1408 		} else {
1409 			kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1410 			r = RESUME_GUEST;
1411 		}
1412 		break;
1413 	default:
1414 	{
1415 		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1416 		/* Ugh - bork here! What did we get? */
1417 		printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1418 			exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
1419 		r = RESUME_HOST;
1420 		BUG();
1421 		break;
1422 	}
1423 	}
1424 
1425 	if (!(r & RESUME_HOST)) {
1426 		/* To avoid clobbering exit_reason, only check for signals if
1427 		 * we aren't already exiting to userspace for some other
1428 		 * reason. */
1429 
1430 		/*
1431 		 * Interrupts could be timers for the guest which we have to
1432 		 * inject again, so let's postpone them until we're in the guest
1433 		 * and if we really did time things so badly, then we just exit
1434 		 * again due to a host external interrupt.
1435 		 */
1436 		s = kvmppc_prepare_to_enter(vcpu);
1437 		if (s <= 0)
1438 			r = s;
1439 		else {
1440 			/* interrupts now hard-disabled */
1441 			kvmppc_fix_ee_before_entry();
1442 		}
1443 
1444 		kvmppc_handle_lost_ext(vcpu);
1445 	}
1446 
1447 	trace_kvm_book3s_reenter(r, vcpu);
1448 
1449 	return r;
1450 }
1451 
1452 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu *vcpu,
1453 					    struct kvm_sregs *sregs)
1454 {
1455 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1456 	int i;
1457 
1458 	sregs->pvr = vcpu->arch.pvr;
1459 
1460 	sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1461 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1462 		for (i = 0; i < 64; i++) {
1463 			sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
1464 			sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1465 		}
1466 	} else {
1467 		for (i = 0; i < 16; i++)
1468 			sregs->u.s.ppc32.sr[i] = kvmppc_get_sr(vcpu, i);
1469 
1470 		for (i = 0; i < 8; i++) {
1471 			sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1472 			sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1473 		}
1474 	}
1475 
1476 	return 0;
1477 }
1478 
1479 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu *vcpu,
1480 					    struct kvm_sregs *sregs)
1481 {
1482 	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1483 	int i;
1484 
1485 	kvmppc_set_pvr_pr(vcpu, sregs->pvr);
1486 
1487 	vcpu3s->sdr1 = sregs->u.s.sdr1;
1488 #ifdef CONFIG_PPC_BOOK3S_64
1489 	if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1490 		/* Flush all SLB entries */
1491 		vcpu->arch.mmu.slbmte(vcpu, 0, 0);
1492 		vcpu->arch.mmu.slbia(vcpu);
1493 
1494 		for (i = 0; i < 64; i++) {
1495 			u64 rb = sregs->u.s.ppc64.slb[i].slbe;
1496 			u64 rs = sregs->u.s.ppc64.slb[i].slbv;
1497 
1498 			if (rb & SLB_ESID_V)
1499 				vcpu->arch.mmu.slbmte(vcpu, rs, rb);
1500 		}
1501 	} else
1502 #endif
1503 	{
1504 		for (i = 0; i < 16; i++) {
1505 			vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1506 		}
1507 		for (i = 0; i < 8; i++) {
1508 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1509 				       (u32)sregs->u.s.ppc32.ibat[i]);
1510 			kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1511 				       (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1512 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1513 				       (u32)sregs->u.s.ppc32.dbat[i]);
1514 			kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1515 				       (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1516 		}
1517 	}
1518 
1519 	/* Flush the MMU after messing with the segments */
1520 	kvmppc_mmu_pte_flush(vcpu, 0, 0);
1521 
1522 	return 0;
1523 }
1524 
1525 static int kvmppc_get_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1526 				 union kvmppc_one_reg *val)
1527 {
1528 	int r = 0;
1529 
1530 	switch (id) {
1531 	case KVM_REG_PPC_DEBUG_INST:
1532 		*val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1533 		break;
1534 	case KVM_REG_PPC_HIOR:
1535 		*val = get_reg_val(id, to_book3s(vcpu)->hior);
1536 		break;
1537 	case KVM_REG_PPC_VTB:
1538 		*val = get_reg_val(id, to_book3s(vcpu)->vtb);
1539 		break;
1540 	case KVM_REG_PPC_LPCR:
1541 	case KVM_REG_PPC_LPCR_64:
1542 		/*
1543 		 * We are only interested in the LPCR_ILE bit
1544 		 */
1545 		if (vcpu->arch.intr_msr & MSR_LE)
1546 			*val = get_reg_val(id, LPCR_ILE);
1547 		else
1548 			*val = get_reg_val(id, 0);
1549 		break;
1550 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1551 	case KVM_REG_PPC_TFHAR:
1552 		*val = get_reg_val(id, vcpu->arch.tfhar);
1553 		break;
1554 	case KVM_REG_PPC_TFIAR:
1555 		*val = get_reg_val(id, vcpu->arch.tfiar);
1556 		break;
1557 	case KVM_REG_PPC_TEXASR:
1558 		*val = get_reg_val(id, vcpu->arch.texasr);
1559 		break;
1560 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1561 		*val = get_reg_val(id,
1562 				vcpu->arch.gpr_tm[id-KVM_REG_PPC_TM_GPR0]);
1563 		break;
1564 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1565 	{
1566 		int i, j;
1567 
1568 		i = id - KVM_REG_PPC_TM_VSR0;
1569 		if (i < 32)
1570 			for (j = 0; j < TS_FPRWIDTH; j++)
1571 				val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1572 		else {
1573 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1574 				val->vval = vcpu->arch.vr_tm.vr[i-32];
1575 			else
1576 				r = -ENXIO;
1577 		}
1578 		break;
1579 	}
1580 	case KVM_REG_PPC_TM_CR:
1581 		*val = get_reg_val(id, vcpu->arch.cr_tm);
1582 		break;
1583 	case KVM_REG_PPC_TM_XER:
1584 		*val = get_reg_val(id, vcpu->arch.xer_tm);
1585 		break;
1586 	case KVM_REG_PPC_TM_LR:
1587 		*val = get_reg_val(id, vcpu->arch.lr_tm);
1588 		break;
1589 	case KVM_REG_PPC_TM_CTR:
1590 		*val = get_reg_val(id, vcpu->arch.ctr_tm);
1591 		break;
1592 	case KVM_REG_PPC_TM_FPSCR:
1593 		*val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1594 		break;
1595 	case KVM_REG_PPC_TM_AMR:
1596 		*val = get_reg_val(id, vcpu->arch.amr_tm);
1597 		break;
1598 	case KVM_REG_PPC_TM_PPR:
1599 		*val = get_reg_val(id, vcpu->arch.ppr_tm);
1600 		break;
1601 	case KVM_REG_PPC_TM_VRSAVE:
1602 		*val = get_reg_val(id, vcpu->arch.vrsave_tm);
1603 		break;
1604 	case KVM_REG_PPC_TM_VSCR:
1605 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1606 			*val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1607 		else
1608 			r = -ENXIO;
1609 		break;
1610 	case KVM_REG_PPC_TM_DSCR:
1611 		*val = get_reg_val(id, vcpu->arch.dscr_tm);
1612 		break;
1613 	case KVM_REG_PPC_TM_TAR:
1614 		*val = get_reg_val(id, vcpu->arch.tar_tm);
1615 		break;
1616 #endif
1617 	default:
1618 		r = -EINVAL;
1619 		break;
1620 	}
1621 
1622 	return r;
1623 }
1624 
1625 static void kvmppc_set_lpcr_pr(struct kvm_vcpu *vcpu, u64 new_lpcr)
1626 {
1627 	if (new_lpcr & LPCR_ILE)
1628 		vcpu->arch.intr_msr |= MSR_LE;
1629 	else
1630 		vcpu->arch.intr_msr &= ~MSR_LE;
1631 }
1632 
1633 static int kvmppc_set_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1634 				 union kvmppc_one_reg *val)
1635 {
1636 	int r = 0;
1637 
1638 	switch (id) {
1639 	case KVM_REG_PPC_HIOR:
1640 		to_book3s(vcpu)->hior = set_reg_val(id, *val);
1641 		to_book3s(vcpu)->hior_explicit = true;
1642 		break;
1643 	case KVM_REG_PPC_VTB:
1644 		to_book3s(vcpu)->vtb = set_reg_val(id, *val);
1645 		break;
1646 	case KVM_REG_PPC_LPCR:
1647 	case KVM_REG_PPC_LPCR_64:
1648 		kvmppc_set_lpcr_pr(vcpu, set_reg_val(id, *val));
1649 		break;
1650 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1651 	case KVM_REG_PPC_TFHAR:
1652 		vcpu->arch.tfhar = set_reg_val(id, *val);
1653 		break;
1654 	case KVM_REG_PPC_TFIAR:
1655 		vcpu->arch.tfiar = set_reg_val(id, *val);
1656 		break;
1657 	case KVM_REG_PPC_TEXASR:
1658 		vcpu->arch.texasr = set_reg_val(id, *val);
1659 		break;
1660 	case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1661 		vcpu->arch.gpr_tm[id - KVM_REG_PPC_TM_GPR0] =
1662 			set_reg_val(id, *val);
1663 		break;
1664 	case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1665 	{
1666 		int i, j;
1667 
1668 		i = id - KVM_REG_PPC_TM_VSR0;
1669 		if (i < 32)
1670 			for (j = 0; j < TS_FPRWIDTH; j++)
1671 				vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1672 		else
1673 			if (cpu_has_feature(CPU_FTR_ALTIVEC))
1674 				vcpu->arch.vr_tm.vr[i-32] = val->vval;
1675 			else
1676 				r = -ENXIO;
1677 		break;
1678 	}
1679 	case KVM_REG_PPC_TM_CR:
1680 		vcpu->arch.cr_tm = set_reg_val(id, *val);
1681 		break;
1682 	case KVM_REG_PPC_TM_XER:
1683 		vcpu->arch.xer_tm = set_reg_val(id, *val);
1684 		break;
1685 	case KVM_REG_PPC_TM_LR:
1686 		vcpu->arch.lr_tm = set_reg_val(id, *val);
1687 		break;
1688 	case KVM_REG_PPC_TM_CTR:
1689 		vcpu->arch.ctr_tm = set_reg_val(id, *val);
1690 		break;
1691 	case KVM_REG_PPC_TM_FPSCR:
1692 		vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1693 		break;
1694 	case KVM_REG_PPC_TM_AMR:
1695 		vcpu->arch.amr_tm = set_reg_val(id, *val);
1696 		break;
1697 	case KVM_REG_PPC_TM_PPR:
1698 		vcpu->arch.ppr_tm = set_reg_val(id, *val);
1699 		break;
1700 	case KVM_REG_PPC_TM_VRSAVE:
1701 		vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1702 		break;
1703 	case KVM_REG_PPC_TM_VSCR:
1704 		if (cpu_has_feature(CPU_FTR_ALTIVEC))
1705 			vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1706 		else
1707 			r = -ENXIO;
1708 		break;
1709 	case KVM_REG_PPC_TM_DSCR:
1710 		vcpu->arch.dscr_tm = set_reg_val(id, *val);
1711 		break;
1712 	case KVM_REG_PPC_TM_TAR:
1713 		vcpu->arch.tar_tm = set_reg_val(id, *val);
1714 		break;
1715 #endif
1716 	default:
1717 		r = -EINVAL;
1718 		break;
1719 	}
1720 
1721 	return r;
1722 }
1723 
1724 static int kvmppc_core_vcpu_create_pr(struct kvm_vcpu *vcpu)
1725 {
1726 	struct kvmppc_vcpu_book3s *vcpu_book3s;
1727 	unsigned long p;
1728 	int err;
1729 
1730 	err = -ENOMEM;
1731 
1732 	vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1733 	if (!vcpu_book3s)
1734 		goto out;
1735 	vcpu->arch.book3s = vcpu_book3s;
1736 
1737 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1738 	vcpu->arch.shadow_vcpu =
1739 		kzalloc(sizeof(*vcpu->arch.shadow_vcpu), GFP_KERNEL);
1740 	if (!vcpu->arch.shadow_vcpu)
1741 		goto free_vcpu3s;
1742 #endif
1743 
1744 	p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1745 	if (!p)
1746 		goto free_shadow_vcpu;
1747 	vcpu->arch.shared = (void *)p;
1748 #ifdef CONFIG_PPC_BOOK3S_64
1749 	/* Always start the shared struct in native endian mode */
1750 #ifdef __BIG_ENDIAN__
1751         vcpu->arch.shared_big_endian = true;
1752 #else
1753         vcpu->arch.shared_big_endian = false;
1754 #endif
1755 
1756 	/*
1757 	 * Default to the same as the host if we're on sufficiently
1758 	 * recent machine that we have 1TB segments;
1759 	 * otherwise default to PPC970FX.
1760 	 */
1761 	vcpu->arch.pvr = 0x3C0301;
1762 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1763 		vcpu->arch.pvr = mfspr(SPRN_PVR);
1764 	vcpu->arch.intr_msr = MSR_SF;
1765 #else
1766 	/* default to book3s_32 (750) */
1767 	vcpu->arch.pvr = 0x84202;
1768 	vcpu->arch.intr_msr = 0;
1769 #endif
1770 	kvmppc_set_pvr_pr(vcpu, vcpu->arch.pvr);
1771 	vcpu->arch.slb_nr = 64;
1772 
1773 	vcpu->arch.shadow_msr = MSR_USER64 & ~MSR_LE;
1774 
1775 	err = kvmppc_mmu_init_pr(vcpu);
1776 	if (err < 0)
1777 		goto free_shared_page;
1778 
1779 	return 0;
1780 
1781 free_shared_page:
1782 	free_page((unsigned long)vcpu->arch.shared);
1783 free_shadow_vcpu:
1784 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1785 	kfree(vcpu->arch.shadow_vcpu);
1786 free_vcpu3s:
1787 #endif
1788 	vfree(vcpu_book3s);
1789 out:
1790 	return 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 	kvmppc_mmu_destroy_pr(vcpu);
1798 	free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1799 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1800 	kfree(vcpu->arch.shadow_vcpu);
1801 #endif
1802 	vfree(vcpu_book3s);
1803 }
1804 
1805 static int kvmppc_vcpu_run_pr(struct kvm_vcpu *vcpu)
1806 {
1807 	int ret;
1808 
1809 	/* Check if we can run the vcpu at all */
1810 	if (!vcpu->arch.sane) {
1811 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1812 		ret = -EINVAL;
1813 		goto out;
1814 	}
1815 
1816 	kvmppc_setup_debug(vcpu);
1817 
1818 	/*
1819 	 * Interrupts could be timers for the guest which we have to inject
1820 	 * again, so let's postpone them until we're in the guest and if we
1821 	 * really did time things so badly, then we just exit again due to
1822 	 * a host external interrupt.
1823 	 */
1824 	ret = kvmppc_prepare_to_enter(vcpu);
1825 	if (ret <= 0)
1826 		goto out;
1827 	/* interrupts now hard-disabled */
1828 
1829 	/* Save FPU, Altivec and VSX state */
1830 	giveup_all(current);
1831 
1832 	/* Preload FPU if it's enabled */
1833 	if (kvmppc_get_msr(vcpu) & MSR_FP)
1834 		kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1835 
1836 	kvmppc_fix_ee_before_entry();
1837 
1838 	ret = __kvmppc_vcpu_run(vcpu);
1839 
1840 	kvmppc_clear_debug(vcpu);
1841 
1842 	/* No need for guest_exit. It's done in handle_exit.
1843 	   We also get here with interrupts enabled. */
1844 
1845 	/* Make sure we save the guest FPU/Altivec/VSX state */
1846 	kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1847 
1848 	/* Make sure we save the guest TAR/EBB/DSCR state */
1849 	kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
1850 
1851 out:
1852 	vcpu->mode = OUTSIDE_GUEST_MODE;
1853 	return ret;
1854 }
1855 
1856 /*
1857  * Get (and clear) the dirty memory log for a memory slot.
1858  */
1859 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm *kvm,
1860 					 struct kvm_dirty_log *log)
1861 {
1862 	struct kvm_memory_slot *memslot;
1863 	struct kvm_vcpu *vcpu;
1864 	ulong ga, ga_end;
1865 	int is_dirty = 0;
1866 	int r;
1867 	unsigned long n;
1868 
1869 	mutex_lock(&kvm->slots_lock);
1870 
1871 	r = kvm_get_dirty_log(kvm, log, &is_dirty, &memslot);
1872 	if (r)
1873 		goto out;
1874 
1875 	/* If nothing is dirty, don't bother messing with page tables. */
1876 	if (is_dirty) {
1877 		ga = memslot->base_gfn << PAGE_SHIFT;
1878 		ga_end = ga + (memslot->npages << PAGE_SHIFT);
1879 
1880 		kvm_for_each_vcpu(n, vcpu, kvm)
1881 			kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1882 
1883 		n = kvm_dirty_bitmap_bytes(memslot);
1884 		memset(memslot->dirty_bitmap, 0, n);
1885 	}
1886 
1887 	r = 0;
1888 out:
1889 	mutex_unlock(&kvm->slots_lock);
1890 	return r;
1891 }
1892 
1893 static void kvmppc_core_flush_memslot_pr(struct kvm *kvm,
1894 					 struct kvm_memory_slot *memslot)
1895 {
1896 	return;
1897 }
1898 
1899 static int kvmppc_core_prepare_memory_region_pr(struct kvm *kvm,
1900 					struct kvm_memory_slot *memslot,
1901 					const struct kvm_userspace_memory_region *mem,
1902 					enum kvm_mr_change change)
1903 {
1904 	return 0;
1905 }
1906 
1907 static void kvmppc_core_commit_memory_region_pr(struct kvm *kvm,
1908 				const struct kvm_userspace_memory_region *mem,
1909 				const struct kvm_memory_slot *old,
1910 				const struct kvm_memory_slot *new,
1911 				enum kvm_mr_change change)
1912 {
1913 	return;
1914 }
1915 
1916 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot *slot)
1917 {
1918 	return;
1919 }
1920 
1921 #ifdef CONFIG_PPC64
1922 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1923 					 struct kvm_ppc_smmu_info *info)
1924 {
1925 	long int i;
1926 	struct kvm_vcpu *vcpu;
1927 
1928 	info->flags = 0;
1929 
1930 	/* SLB is always 64 entries */
1931 	info->slb_size = 64;
1932 
1933 	/* Standard 4k base page size segment */
1934 	info->sps[0].page_shift = 12;
1935 	info->sps[0].slb_enc = 0;
1936 	info->sps[0].enc[0].page_shift = 12;
1937 	info->sps[0].enc[0].pte_enc = 0;
1938 
1939 	/*
1940 	 * 64k large page size.
1941 	 * We only want to put this in if the CPUs we're emulating
1942 	 * support it, but unfortunately we don't have a vcpu easily
1943 	 * to hand here to test.  Just pick the first vcpu, and if
1944 	 * that doesn't exist yet, report the minimum capability,
1945 	 * i.e., no 64k pages.
1946 	 * 1T segment support goes along with 64k pages.
1947 	 */
1948 	i = 1;
1949 	vcpu = kvm_get_vcpu(kvm, 0);
1950 	if (vcpu && (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
1951 		info->flags = KVM_PPC_1T_SEGMENTS;
1952 		info->sps[i].page_shift = 16;
1953 		info->sps[i].slb_enc = SLB_VSID_L | SLB_VSID_LP_01;
1954 		info->sps[i].enc[0].page_shift = 16;
1955 		info->sps[i].enc[0].pte_enc = 1;
1956 		++i;
1957 	}
1958 
1959 	/* Standard 16M large page size segment */
1960 	info->sps[i].page_shift = 24;
1961 	info->sps[i].slb_enc = SLB_VSID_L;
1962 	info->sps[i].enc[0].page_shift = 24;
1963 	info->sps[i].enc[0].pte_enc = 0;
1964 
1965 	return 0;
1966 }
1967 
1968 static int kvm_configure_mmu_pr(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
1969 {
1970 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
1971 		return -ENODEV;
1972 	/* Require flags and process table base and size to all be zero. */
1973 	if (cfg->flags || cfg->process_table)
1974 		return -EINVAL;
1975 	return 0;
1976 }
1977 
1978 #else
1979 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1980 					 struct kvm_ppc_smmu_info *info)
1981 {
1982 	/* We should not get called */
1983 	BUG();
1984 	return 0;
1985 }
1986 #endif /* CONFIG_PPC64 */
1987 
1988 static unsigned int kvm_global_user_count = 0;
1989 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1990 
1991 static int kvmppc_core_init_vm_pr(struct kvm *kvm)
1992 {
1993 	mutex_init(&kvm->arch.hpt_mutex);
1994 
1995 #ifdef CONFIG_PPC_BOOK3S_64
1996 	/* Start out with the default set of hcalls enabled */
1997 	kvmppc_pr_init_default_hcalls(kvm);
1998 #endif
1999 
2000 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
2001 		spin_lock(&kvm_global_user_count_lock);
2002 		if (++kvm_global_user_count == 1)
2003 			pseries_disable_reloc_on_exc();
2004 		spin_unlock(&kvm_global_user_count_lock);
2005 	}
2006 	return 0;
2007 }
2008 
2009 static void kvmppc_core_destroy_vm_pr(struct kvm *kvm)
2010 {
2011 #ifdef CONFIG_PPC64
2012 	WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
2013 #endif
2014 
2015 	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
2016 		spin_lock(&kvm_global_user_count_lock);
2017 		BUG_ON(kvm_global_user_count == 0);
2018 		if (--kvm_global_user_count == 0)
2019 			pseries_enable_reloc_on_exc();
2020 		spin_unlock(&kvm_global_user_count_lock);
2021 	}
2022 }
2023 
2024 static int kvmppc_core_check_processor_compat_pr(void)
2025 {
2026 	/*
2027 	 * PR KVM can work on POWER9 inside a guest partition
2028 	 * running in HPT mode.  It can't work if we are using
2029 	 * radix translation (because radix provides no way for
2030 	 * a process to have unique translations in quadrant 3).
2031 	 */
2032 	if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
2033 		return -EIO;
2034 	return 0;
2035 }
2036 
2037 static long kvm_arch_vm_ioctl_pr(struct file *filp,
2038 				 unsigned int ioctl, unsigned long arg)
2039 {
2040 	return -ENOTTY;
2041 }
2042 
2043 static struct kvmppc_ops kvm_ops_pr = {
2044 	.get_sregs = kvm_arch_vcpu_ioctl_get_sregs_pr,
2045 	.set_sregs = kvm_arch_vcpu_ioctl_set_sregs_pr,
2046 	.get_one_reg = kvmppc_get_one_reg_pr,
2047 	.set_one_reg = kvmppc_set_one_reg_pr,
2048 	.vcpu_load   = kvmppc_core_vcpu_load_pr,
2049 	.vcpu_put    = kvmppc_core_vcpu_put_pr,
2050 	.inject_interrupt = kvmppc_inject_interrupt_pr,
2051 	.set_msr     = kvmppc_set_msr_pr,
2052 	.vcpu_run    = kvmppc_vcpu_run_pr,
2053 	.vcpu_create = kvmppc_core_vcpu_create_pr,
2054 	.vcpu_free   = kvmppc_core_vcpu_free_pr,
2055 	.check_requests = kvmppc_core_check_requests_pr,
2056 	.get_dirty_log = kvm_vm_ioctl_get_dirty_log_pr,
2057 	.flush_memslot = kvmppc_core_flush_memslot_pr,
2058 	.prepare_memory_region = kvmppc_core_prepare_memory_region_pr,
2059 	.commit_memory_region = kvmppc_core_commit_memory_region_pr,
2060 	.unmap_gfn_range = kvm_unmap_gfn_range_pr,
2061 	.age_gfn  = kvm_age_gfn_pr,
2062 	.test_age_gfn = kvm_test_age_gfn_pr,
2063 	.set_spte_gfn = kvm_set_spte_gfn_pr,
2064 	.free_memslot = kvmppc_core_free_memslot_pr,
2065 	.init_vm = kvmppc_core_init_vm_pr,
2066 	.destroy_vm = kvmppc_core_destroy_vm_pr,
2067 	.get_smmu_info = kvm_vm_ioctl_get_smmu_info_pr,
2068 	.emulate_op = kvmppc_core_emulate_op_pr,
2069 	.emulate_mtspr = kvmppc_core_emulate_mtspr_pr,
2070 	.emulate_mfspr = kvmppc_core_emulate_mfspr_pr,
2071 	.fast_vcpu_kick = kvm_vcpu_kick,
2072 	.arch_vm_ioctl  = kvm_arch_vm_ioctl_pr,
2073 #ifdef CONFIG_PPC_BOOK3S_64
2074 	.hcall_implemented = kvmppc_hcall_impl_pr,
2075 	.configure_mmu = kvm_configure_mmu_pr,
2076 #endif
2077 	.giveup_ext = kvmppc_giveup_ext,
2078 };
2079 
2080 
2081 int kvmppc_book3s_init_pr(void)
2082 {
2083 	int r;
2084 
2085 	r = kvmppc_core_check_processor_compat_pr();
2086 	if (r < 0)
2087 		return r;
2088 
2089 	kvm_ops_pr.owner = THIS_MODULE;
2090 	kvmppc_pr_ops = &kvm_ops_pr;
2091 
2092 	r = kvmppc_mmu_hpte_sysinit();
2093 	return r;
2094 }
2095 
2096 void kvmppc_book3s_exit_pr(void)
2097 {
2098 	kvmppc_pr_ops = NULL;
2099 	kvmppc_mmu_hpte_sysexit();
2100 }
2101 
2102 /*
2103  * We only support separate modules for book3s 64
2104  */
2105 #ifdef CONFIG_PPC_BOOK3S_64
2106 
2107 module_init(kvmppc_book3s_init_pr);
2108 module_exit(kvmppc_book3s_exit_pr);
2109 
2110 MODULE_LICENSE("GPL");
2111 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2112 MODULE_ALIAS("devname:kvm");
2113 #endif
2114