1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <linux/kernel.h>
3 #include <linux/kvm_host.h>
4 #include <asm/asm-prototypes.h>
5 #include <asm/dbell.h>
6 #include <asm/kvm_ppc.h>
7 #include <asm/pmc.h>
8 #include <asm/ppc-opcode.h>
9 
10 #include "book3s_hv.h"
11 
12 static void freeze_pmu(unsigned long mmcr0, unsigned long mmcra)
13 {
14 	if (!(mmcr0 & MMCR0_FC))
15 		goto do_freeze;
16 	if (mmcra & MMCRA_SAMPLE_ENABLE)
17 		goto do_freeze;
18 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
19 		if (!(mmcr0 & MMCR0_PMCCEXT))
20 			goto do_freeze;
21 		if (!(mmcra & MMCRA_BHRB_DISABLE))
22 			goto do_freeze;
23 	}
24 	return;
25 
26 do_freeze:
27 	mmcr0 = MMCR0_FC;
28 	mmcra = 0;
29 	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
30 		mmcr0 |= MMCR0_PMCCEXT;
31 		mmcra = MMCRA_BHRB_DISABLE;
32 	}
33 
34 	mtspr(SPRN_MMCR0, mmcr0);
35 	mtspr(SPRN_MMCRA, mmcra);
36 	isync();
37 }
38 
39 void switch_pmu_to_guest(struct kvm_vcpu *vcpu,
40 			 struct p9_host_os_sprs *host_os_sprs)
41 {
42 	struct lppaca *lp;
43 	int load_pmu = 1;
44 
45 	lp = vcpu->arch.vpa.pinned_addr;
46 	if (lp)
47 		load_pmu = lp->pmcregs_in_use;
48 
49 	/* Save host */
50 	if (ppc_get_pmu_inuse()) {
51 		/*
52 		 * It might be better to put PMU handling (at least for the
53 		 * host) in the perf subsystem because it knows more about what
54 		 * is being used.
55 		 */
56 
57 		/* POWER9, POWER10 do not implement HPMC or SPMC */
58 
59 		host_os_sprs->mmcr0 = mfspr(SPRN_MMCR0);
60 		host_os_sprs->mmcra = mfspr(SPRN_MMCRA);
61 
62 		freeze_pmu(host_os_sprs->mmcr0, host_os_sprs->mmcra);
63 
64 		host_os_sprs->pmc1 = mfspr(SPRN_PMC1);
65 		host_os_sprs->pmc2 = mfspr(SPRN_PMC2);
66 		host_os_sprs->pmc3 = mfspr(SPRN_PMC3);
67 		host_os_sprs->pmc4 = mfspr(SPRN_PMC4);
68 		host_os_sprs->pmc5 = mfspr(SPRN_PMC5);
69 		host_os_sprs->pmc6 = mfspr(SPRN_PMC6);
70 		host_os_sprs->mmcr1 = mfspr(SPRN_MMCR1);
71 		host_os_sprs->mmcr2 = mfspr(SPRN_MMCR2);
72 		host_os_sprs->sdar = mfspr(SPRN_SDAR);
73 		host_os_sprs->siar = mfspr(SPRN_SIAR);
74 		host_os_sprs->sier1 = mfspr(SPRN_SIER);
75 
76 		if (cpu_has_feature(CPU_FTR_ARCH_31)) {
77 			host_os_sprs->mmcr3 = mfspr(SPRN_MMCR3);
78 			host_os_sprs->sier2 = mfspr(SPRN_SIER2);
79 			host_os_sprs->sier3 = mfspr(SPRN_SIER3);
80 		}
81 	}
82 
83 #ifdef CONFIG_PPC_PSERIES
84 	/* After saving PMU, before loading guest PMU, flip pmcregs_in_use */
85 	if (kvmhv_on_pseries()) {
86 		barrier();
87 		get_lppaca()->pmcregs_in_use = load_pmu;
88 		barrier();
89 	}
90 #endif
91 
92 	/*
93 	 * Load guest. If the VPA said the PMCs are not in use but the guest
94 	 * tried to access them anyway, HFSCR[PM] will be set by the HFAC
95 	 * fault so we can make forward progress.
96 	 */
97 	if (load_pmu || (vcpu->arch.hfscr & HFSCR_PM)) {
98 		mtspr(SPRN_PMC1, vcpu->arch.pmc[0]);
99 		mtspr(SPRN_PMC2, vcpu->arch.pmc[1]);
100 		mtspr(SPRN_PMC3, vcpu->arch.pmc[2]);
101 		mtspr(SPRN_PMC4, vcpu->arch.pmc[3]);
102 		mtspr(SPRN_PMC5, vcpu->arch.pmc[4]);
103 		mtspr(SPRN_PMC6, vcpu->arch.pmc[5]);
104 		mtspr(SPRN_MMCR1, vcpu->arch.mmcr[1]);
105 		mtspr(SPRN_MMCR2, vcpu->arch.mmcr[2]);
106 		mtspr(SPRN_SDAR, vcpu->arch.sdar);
107 		mtspr(SPRN_SIAR, vcpu->arch.siar);
108 		mtspr(SPRN_SIER, vcpu->arch.sier[0]);
109 
110 		if (cpu_has_feature(CPU_FTR_ARCH_31)) {
111 			mtspr(SPRN_MMCR3, vcpu->arch.mmcr[3]);
112 			mtspr(SPRN_SIER2, vcpu->arch.sier[1]);
113 			mtspr(SPRN_SIER3, vcpu->arch.sier[2]);
114 		}
115 
116 		/* Set MMCRA then MMCR0 last */
117 		mtspr(SPRN_MMCRA, vcpu->arch.mmcra);
118 		mtspr(SPRN_MMCR0, vcpu->arch.mmcr[0]);
119 		/* No isync necessary because we're starting counters */
120 
121 		if (!vcpu->arch.nested &&
122 				(vcpu->arch.hfscr_permitted & HFSCR_PM))
123 			vcpu->arch.hfscr |= HFSCR_PM;
124 	}
125 }
126 EXPORT_SYMBOL_GPL(switch_pmu_to_guest);
127 
128 void switch_pmu_to_host(struct kvm_vcpu *vcpu,
129 			struct p9_host_os_sprs *host_os_sprs)
130 {
131 	struct lppaca *lp;
132 	int save_pmu = 1;
133 
134 	lp = vcpu->arch.vpa.pinned_addr;
135 	if (lp)
136 		save_pmu = lp->pmcregs_in_use;
137 	if (IS_ENABLED(CONFIG_KVM_BOOK3S_HV_NESTED_PMU_WORKAROUND)) {
138 		/*
139 		 * Save pmu if this guest is capable of running nested guests.
140 		 * This is option is for old L1s that do not set their
141 		 * lppaca->pmcregs_in_use properly when entering their L2.
142 		 */
143 		save_pmu |= nesting_enabled(vcpu->kvm);
144 	}
145 
146 	if (save_pmu) {
147 		vcpu->arch.mmcr[0] = mfspr(SPRN_MMCR0);
148 		vcpu->arch.mmcra = mfspr(SPRN_MMCRA);
149 
150 		freeze_pmu(vcpu->arch.mmcr[0], vcpu->arch.mmcra);
151 
152 		vcpu->arch.pmc[0] = mfspr(SPRN_PMC1);
153 		vcpu->arch.pmc[1] = mfspr(SPRN_PMC2);
154 		vcpu->arch.pmc[2] = mfspr(SPRN_PMC3);
155 		vcpu->arch.pmc[3] = mfspr(SPRN_PMC4);
156 		vcpu->arch.pmc[4] = mfspr(SPRN_PMC5);
157 		vcpu->arch.pmc[5] = mfspr(SPRN_PMC6);
158 		vcpu->arch.mmcr[1] = mfspr(SPRN_MMCR1);
159 		vcpu->arch.mmcr[2] = mfspr(SPRN_MMCR2);
160 		vcpu->arch.sdar = mfspr(SPRN_SDAR);
161 		vcpu->arch.siar = mfspr(SPRN_SIAR);
162 		vcpu->arch.sier[0] = mfspr(SPRN_SIER);
163 
164 		if (cpu_has_feature(CPU_FTR_ARCH_31)) {
165 			vcpu->arch.mmcr[3] = mfspr(SPRN_MMCR3);
166 			vcpu->arch.sier[1] = mfspr(SPRN_SIER2);
167 			vcpu->arch.sier[2] = mfspr(SPRN_SIER3);
168 		}
169 
170 	} else if (vcpu->arch.hfscr & HFSCR_PM) {
171 		/*
172 		 * The guest accessed PMC SPRs without specifying they should
173 		 * be preserved, or it cleared pmcregs_in_use after the last
174 		 * access. Just ensure they are frozen.
175 		 */
176 		freeze_pmu(mfspr(SPRN_MMCR0), mfspr(SPRN_MMCRA));
177 
178 		/*
179 		 * Demand-fault PMU register access in the guest.
180 		 *
181 		 * This is used to grab the guest's VPA pmcregs_in_use value
182 		 * and reflect it into the host's VPA in the case of a nested
183 		 * hypervisor.
184 		 *
185 		 * It also avoids having to zero-out SPRs after each guest
186 		 * exit to avoid side-channels when.
187 		 *
188 		 * This is cleared here when we exit the guest, so later HFSCR
189 		 * interrupt handling can add it back to run the guest with
190 		 * PM enabled next time.
191 		 */
192 		if (!vcpu->arch.nested)
193 			vcpu->arch.hfscr &= ~HFSCR_PM;
194 	} /* otherwise the PMU should still be frozen */
195 
196 #ifdef CONFIG_PPC_PSERIES
197 	if (kvmhv_on_pseries()) {
198 		barrier();
199 		get_lppaca()->pmcregs_in_use = ppc_get_pmu_inuse();
200 		barrier();
201 	}
202 #endif
203 
204 	if (ppc_get_pmu_inuse()) {
205 		mtspr(SPRN_PMC1, host_os_sprs->pmc1);
206 		mtspr(SPRN_PMC2, host_os_sprs->pmc2);
207 		mtspr(SPRN_PMC3, host_os_sprs->pmc3);
208 		mtspr(SPRN_PMC4, host_os_sprs->pmc4);
209 		mtspr(SPRN_PMC5, host_os_sprs->pmc5);
210 		mtspr(SPRN_PMC6, host_os_sprs->pmc6);
211 		mtspr(SPRN_MMCR1, host_os_sprs->mmcr1);
212 		mtspr(SPRN_MMCR2, host_os_sprs->mmcr2);
213 		mtspr(SPRN_SDAR, host_os_sprs->sdar);
214 		mtspr(SPRN_SIAR, host_os_sprs->siar);
215 		mtspr(SPRN_SIER, host_os_sprs->sier1);
216 
217 		if (cpu_has_feature(CPU_FTR_ARCH_31)) {
218 			mtspr(SPRN_MMCR3, host_os_sprs->mmcr3);
219 			mtspr(SPRN_SIER2, host_os_sprs->sier2);
220 			mtspr(SPRN_SIER3, host_os_sprs->sier3);
221 		}
222 
223 		/* Set MMCRA then MMCR0 last */
224 		mtspr(SPRN_MMCRA, host_os_sprs->mmcra);
225 		mtspr(SPRN_MMCR0, host_os_sprs->mmcr0);
226 		isync();
227 	}
228 }
229 EXPORT_SYMBOL_GPL(switch_pmu_to_host);
230 
231 static void load_spr_state(struct kvm_vcpu *vcpu,
232 				struct p9_host_os_sprs *host_os_sprs)
233 {
234 	/* TAR is very fast */
235 	mtspr(SPRN_TAR, vcpu->arch.tar);
236 
237 #ifdef CONFIG_ALTIVEC
238 	if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
239 	    current->thread.vrsave != vcpu->arch.vrsave)
240 		mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
241 #endif
242 
243 	if (vcpu->arch.hfscr & HFSCR_EBB) {
244 		if (current->thread.ebbhr != vcpu->arch.ebbhr)
245 			mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
246 		if (current->thread.ebbrr != vcpu->arch.ebbrr)
247 			mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
248 		if (current->thread.bescr != vcpu->arch.bescr)
249 			mtspr(SPRN_BESCR, vcpu->arch.bescr);
250 	}
251 
252 	if (cpu_has_feature(CPU_FTR_P9_TIDR) &&
253 			current->thread.tidr != vcpu->arch.tid)
254 		mtspr(SPRN_TIDR, vcpu->arch.tid);
255 	if (host_os_sprs->iamr != vcpu->arch.iamr)
256 		mtspr(SPRN_IAMR, vcpu->arch.iamr);
257 	if (host_os_sprs->amr != vcpu->arch.amr)
258 		mtspr(SPRN_AMR, vcpu->arch.amr);
259 	if (vcpu->arch.uamor != 0)
260 		mtspr(SPRN_UAMOR, vcpu->arch.uamor);
261 	if (current->thread.fscr != vcpu->arch.fscr)
262 		mtspr(SPRN_FSCR, vcpu->arch.fscr);
263 	if (current->thread.dscr != vcpu->arch.dscr)
264 		mtspr(SPRN_DSCR, vcpu->arch.dscr);
265 	if (vcpu->arch.pspb != 0)
266 		mtspr(SPRN_PSPB, vcpu->arch.pspb);
267 
268 	/*
269 	 * DAR, DSISR, and for nested HV, SPRGs must be set with MSR[RI]
270 	 * clear (or hstate set appropriately to catch those registers
271 	 * being clobbered if we take a MCE or SRESET), so those are done
272 	 * later.
273 	 */
274 
275 	if (!(vcpu->arch.ctrl & 1))
276 		mtspr(SPRN_CTRLT, 0);
277 }
278 
279 static void store_spr_state(struct kvm_vcpu *vcpu)
280 {
281 	vcpu->arch.tar = mfspr(SPRN_TAR);
282 
283 #ifdef CONFIG_ALTIVEC
284 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
285 		vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
286 #endif
287 
288 	if (vcpu->arch.hfscr & HFSCR_EBB) {
289 		vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
290 		vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
291 		vcpu->arch.bescr = mfspr(SPRN_BESCR);
292 	}
293 
294 	if (cpu_has_feature(CPU_FTR_P9_TIDR))
295 		vcpu->arch.tid = mfspr(SPRN_TIDR);
296 	vcpu->arch.iamr = mfspr(SPRN_IAMR);
297 	vcpu->arch.amr = mfspr(SPRN_AMR);
298 	vcpu->arch.uamor = mfspr(SPRN_UAMOR);
299 	vcpu->arch.fscr = mfspr(SPRN_FSCR);
300 	vcpu->arch.dscr = mfspr(SPRN_DSCR);
301 	vcpu->arch.pspb = mfspr(SPRN_PSPB);
302 
303 	vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
304 }
305 
306 /* Returns true if current MSR and/or guest MSR may have changed */
307 bool load_vcpu_state(struct kvm_vcpu *vcpu,
308 		     struct p9_host_os_sprs *host_os_sprs)
309 {
310 	bool ret = false;
311 
312 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
313 	if (cpu_has_feature(CPU_FTR_TM) ||
314 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) {
315 		unsigned long guest_msr = vcpu->arch.shregs.msr;
316 		if (MSR_TM_ACTIVE(guest_msr)) {
317 			kvmppc_restore_tm_hv(vcpu, guest_msr, true);
318 			ret = true;
319 		} else if (vcpu->arch.hfscr & HFSCR_TM) {
320 			mtspr(SPRN_TEXASR, vcpu->arch.texasr);
321 			mtspr(SPRN_TFHAR, vcpu->arch.tfhar);
322 			mtspr(SPRN_TFIAR, vcpu->arch.tfiar);
323 		}
324 	}
325 #endif
326 
327 	load_spr_state(vcpu, host_os_sprs);
328 
329 	load_fp_state(&vcpu->arch.fp);
330 #ifdef CONFIG_ALTIVEC
331 	load_vr_state(&vcpu->arch.vr);
332 #endif
333 
334 	return ret;
335 }
336 EXPORT_SYMBOL_GPL(load_vcpu_state);
337 
338 void store_vcpu_state(struct kvm_vcpu *vcpu)
339 {
340 	store_spr_state(vcpu);
341 
342 	store_fp_state(&vcpu->arch.fp);
343 #ifdef CONFIG_ALTIVEC
344 	store_vr_state(&vcpu->arch.vr);
345 #endif
346 
347 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
348 	if (cpu_has_feature(CPU_FTR_TM) ||
349 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) {
350 		unsigned long guest_msr = vcpu->arch.shregs.msr;
351 		if (MSR_TM_ACTIVE(guest_msr)) {
352 			kvmppc_save_tm_hv(vcpu, guest_msr, true);
353 		} else if (vcpu->arch.hfscr & HFSCR_TM) {
354 			vcpu->arch.texasr = mfspr(SPRN_TEXASR);
355 			vcpu->arch.tfhar = mfspr(SPRN_TFHAR);
356 			vcpu->arch.tfiar = mfspr(SPRN_TFIAR);
357 
358 			if (!vcpu->arch.nested) {
359 				vcpu->arch.load_tm++; /* see load_ebb comment */
360 				if (!vcpu->arch.load_tm)
361 					vcpu->arch.hfscr &= ~HFSCR_TM;
362 			}
363 		}
364 	}
365 #endif
366 }
367 EXPORT_SYMBOL_GPL(store_vcpu_state);
368 
369 void save_p9_host_os_sprs(struct p9_host_os_sprs *host_os_sprs)
370 {
371 	host_os_sprs->iamr = mfspr(SPRN_IAMR);
372 	host_os_sprs->amr = mfspr(SPRN_AMR);
373 }
374 EXPORT_SYMBOL_GPL(save_p9_host_os_sprs);
375 
376 /* vcpu guest regs must already be saved */
377 void restore_p9_host_os_sprs(struct kvm_vcpu *vcpu,
378 			     struct p9_host_os_sprs *host_os_sprs)
379 {
380 	/*
381 	 * current->thread.xxx registers must all be restored to host
382 	 * values before a potential context switch, otherwise the context
383 	 * switch itself will overwrite current->thread.xxx with the values
384 	 * from the guest SPRs.
385 	 */
386 
387 	mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
388 
389 	if (cpu_has_feature(CPU_FTR_P9_TIDR) &&
390 			current->thread.tidr != vcpu->arch.tid)
391 		mtspr(SPRN_TIDR, current->thread.tidr);
392 	if (host_os_sprs->iamr != vcpu->arch.iamr)
393 		mtspr(SPRN_IAMR, host_os_sprs->iamr);
394 	if (vcpu->arch.uamor != 0)
395 		mtspr(SPRN_UAMOR, 0);
396 	if (host_os_sprs->amr != vcpu->arch.amr)
397 		mtspr(SPRN_AMR, host_os_sprs->amr);
398 	if (current->thread.fscr != vcpu->arch.fscr)
399 		mtspr(SPRN_FSCR, current->thread.fscr);
400 	if (current->thread.dscr != vcpu->arch.dscr)
401 		mtspr(SPRN_DSCR, current->thread.dscr);
402 	if (vcpu->arch.pspb != 0)
403 		mtspr(SPRN_PSPB, 0);
404 
405 	/* Save guest CTRL register, set runlatch to 1 */
406 	if (!(vcpu->arch.ctrl & 1))
407 		mtspr(SPRN_CTRLT, 1);
408 
409 #ifdef CONFIG_ALTIVEC
410 	if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
411 	    vcpu->arch.vrsave != current->thread.vrsave)
412 		mtspr(SPRN_VRSAVE, current->thread.vrsave);
413 #endif
414 	if (vcpu->arch.hfscr & HFSCR_EBB) {
415 		if (vcpu->arch.bescr != current->thread.bescr)
416 			mtspr(SPRN_BESCR, current->thread.bescr);
417 		if (vcpu->arch.ebbhr != current->thread.ebbhr)
418 			mtspr(SPRN_EBBHR, current->thread.ebbhr);
419 		if (vcpu->arch.ebbrr != current->thread.ebbrr)
420 			mtspr(SPRN_EBBRR, current->thread.ebbrr);
421 
422 		if (!vcpu->arch.nested) {
423 			/*
424 			 * This is like load_fp in context switching, turn off
425 			 * the facility after it wraps the u8 to try avoiding
426 			 * saving and restoring the registers each partition
427 			 * switch.
428 			 */
429 			vcpu->arch.load_ebb++;
430 			if (!vcpu->arch.load_ebb)
431 				vcpu->arch.hfscr &= ~HFSCR_EBB;
432 		}
433 	}
434 
435 	if (vcpu->arch.tar != current->thread.tar)
436 		mtspr(SPRN_TAR, current->thread.tar);
437 }
438 EXPORT_SYMBOL_GPL(restore_p9_host_os_sprs);
439 
440 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
441 static void __start_timing(struct kvm_vcpu *vcpu, struct kvmhv_tb_accumulator *next)
442 {
443 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
444 	u64 tb = mftb() - vc->tb_offset_applied;
445 
446 	vcpu->arch.cur_activity = next;
447 	vcpu->arch.cur_tb_start = tb;
448 }
449 
450 static void __accumulate_time(struct kvm_vcpu *vcpu, struct kvmhv_tb_accumulator *next)
451 {
452 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
453 	struct kvmhv_tb_accumulator *curr;
454 	u64 tb = mftb() - vc->tb_offset_applied;
455 	u64 prev_tb;
456 	u64 delta;
457 	u64 seq;
458 
459 	curr = vcpu->arch.cur_activity;
460 	vcpu->arch.cur_activity = next;
461 	prev_tb = vcpu->arch.cur_tb_start;
462 	vcpu->arch.cur_tb_start = tb;
463 
464 	if (!curr)
465 		return;
466 
467 	delta = tb - prev_tb;
468 
469 	seq = curr->seqcount;
470 	curr->seqcount = seq + 1;
471 	smp_wmb();
472 	curr->tb_total += delta;
473 	if (seq == 0 || delta < curr->tb_min)
474 		curr->tb_min = delta;
475 	if (delta > curr->tb_max)
476 		curr->tb_max = delta;
477 	smp_wmb();
478 	curr->seqcount = seq + 2;
479 }
480 
481 #define start_timing(vcpu, next) __start_timing(vcpu, next)
482 #define end_timing(vcpu) __start_timing(vcpu, NULL)
483 #define accumulate_time(vcpu, next) __accumulate_time(vcpu, next)
484 #else
485 #define start_timing(vcpu, next) do {} while (0)
486 #define end_timing(vcpu) do {} while (0)
487 #define accumulate_time(vcpu, next) do {} while (0)
488 #endif
489 
490 static inline u64 mfslbv(unsigned int idx)
491 {
492 	u64 slbev;
493 
494 	asm volatile("slbmfev  %0,%1" : "=r" (slbev) : "r" (idx));
495 
496 	return slbev;
497 }
498 
499 static inline u64 mfslbe(unsigned int idx)
500 {
501 	u64 slbee;
502 
503 	asm volatile("slbmfee  %0,%1" : "=r" (slbee) : "r" (idx));
504 
505 	return slbee;
506 }
507 
508 static inline void mtslb(u64 slbee, u64 slbev)
509 {
510 	asm volatile("slbmte %0,%1" :: "r" (slbev), "r" (slbee));
511 }
512 
513 static inline void clear_slb_entry(unsigned int idx)
514 {
515 	mtslb(idx, 0);
516 }
517 
518 static inline void slb_clear_invalidate_partition(void)
519 {
520 	clear_slb_entry(0);
521 	asm volatile(PPC_SLBIA(6));
522 }
523 
524 /*
525  * Malicious or buggy radix guests may have inserted SLB entries
526  * (only 0..3 because radix always runs with UPRT=1), so these must
527  * be cleared here to avoid side-channels. slbmte is used rather
528  * than slbia, as it won't clear cached translations.
529  */
530 static void radix_clear_slb(void)
531 {
532 	int i;
533 
534 	for (i = 0; i < 4; i++)
535 		clear_slb_entry(i);
536 }
537 
538 static void switch_mmu_to_guest_radix(struct kvm *kvm, struct kvm_vcpu *vcpu, u64 lpcr)
539 {
540 	struct kvm_nested_guest *nested = vcpu->arch.nested;
541 	u32 lpid;
542 	u32 pid;
543 
544 	lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
545 	pid = vcpu->arch.pid;
546 
547 	/*
548 	 * Prior memory accesses to host PID Q3 must be completed before we
549 	 * start switching, and stores must be drained to avoid not-my-LPAR
550 	 * logic (see switch_mmu_to_host).
551 	 */
552 	asm volatile("hwsync" ::: "memory");
553 	isync();
554 	mtspr(SPRN_LPID, lpid);
555 	mtspr(SPRN_LPCR, lpcr);
556 	mtspr(SPRN_PID, pid);
557 	/*
558 	 * isync not required here because we are HRFID'ing to guest before
559 	 * any guest context access, which is context synchronising.
560 	 */
561 }
562 
563 static void switch_mmu_to_guest_hpt(struct kvm *kvm, struct kvm_vcpu *vcpu, u64 lpcr)
564 {
565 	u32 lpid;
566 	u32 pid;
567 	int i;
568 
569 	lpid = kvm->arch.lpid;
570 	pid = vcpu->arch.pid;
571 
572 	/*
573 	 * See switch_mmu_to_guest_radix. ptesync should not be required here
574 	 * even if the host is in HPT mode because speculative accesses would
575 	 * not cause RC updates (we are in real mode).
576 	 */
577 	asm volatile("hwsync" ::: "memory");
578 	isync();
579 	mtspr(SPRN_LPID, lpid);
580 	mtspr(SPRN_LPCR, lpcr);
581 	mtspr(SPRN_PID, pid);
582 
583 	for (i = 0; i < vcpu->arch.slb_max; i++)
584 		mtslb(vcpu->arch.slb[i].orige, vcpu->arch.slb[i].origv);
585 	/*
586 	 * isync not required here, see switch_mmu_to_guest_radix.
587 	 */
588 }
589 
590 static void switch_mmu_to_host(struct kvm *kvm, u32 pid)
591 {
592 	u32 lpid = kvm->arch.host_lpid;
593 	u64 lpcr = kvm->arch.host_lpcr;
594 
595 	/*
596 	 * The guest has exited, so guest MMU context is no longer being
597 	 * non-speculatively accessed, but a hwsync is needed before the
598 	 * mtLPIDR / mtPIDR switch, in order to ensure all stores are drained,
599 	 * so the not-my-LPAR tlbie logic does not overlook them.
600 	 */
601 	asm volatile("hwsync" ::: "memory");
602 	isync();
603 	mtspr(SPRN_PID, pid);
604 	mtspr(SPRN_LPID, lpid);
605 	mtspr(SPRN_LPCR, lpcr);
606 	/*
607 	 * isync is not required after the switch, because mtmsrd with L=0
608 	 * is performed after this switch, which is context synchronising.
609 	 */
610 
611 	if (!radix_enabled())
612 		slb_restore_bolted_realmode();
613 }
614 
615 static void save_clear_host_mmu(struct kvm *kvm)
616 {
617 	if (!radix_enabled()) {
618 		/*
619 		 * Hash host could save and restore host SLB entries to
620 		 * reduce SLB fault overheads of VM exits, but for now the
621 		 * existing code clears all entries and restores just the
622 		 * bolted ones when switching back to host.
623 		 */
624 		slb_clear_invalidate_partition();
625 	}
626 }
627 
628 static void save_clear_guest_mmu(struct kvm *kvm, struct kvm_vcpu *vcpu)
629 {
630 	if (kvm_is_radix(kvm)) {
631 		radix_clear_slb();
632 	} else {
633 		int i;
634 		int nr = 0;
635 
636 		/*
637 		 * This must run before switching to host (radix host can't
638 		 * access all SLBs).
639 		 */
640 		for (i = 0; i < vcpu->arch.slb_nr; i++) {
641 			u64 slbee, slbev;
642 
643 			slbee = mfslbe(i);
644 			if (slbee & SLB_ESID_V) {
645 				slbev = mfslbv(i);
646 				vcpu->arch.slb[nr].orige = slbee | i;
647 				vcpu->arch.slb[nr].origv = slbev;
648 				nr++;
649 			}
650 		}
651 		vcpu->arch.slb_max = nr;
652 		slb_clear_invalidate_partition();
653 	}
654 }
655 
656 static void flush_guest_tlb(struct kvm *kvm)
657 {
658 	unsigned long rb, set;
659 
660 	rb = PPC_BIT(52);	/* IS = 2 */
661 	if (kvm_is_radix(kvm)) {
662 		/* R=1 PRS=1 RIC=2 */
663 		asm volatile(PPC_TLBIEL(%0, %4, %3, %2, %1)
664 			     : : "r" (rb), "i" (1), "i" (1), "i" (2),
665 			       "r" (0) : "memory");
666 		for (set = 1; set < kvm->arch.tlb_sets; ++set) {
667 			rb += PPC_BIT(51);	/* increment set number */
668 			/* R=1 PRS=1 RIC=0 */
669 			asm volatile(PPC_TLBIEL(%0, %4, %3, %2, %1)
670 				     : : "r" (rb), "i" (1), "i" (1), "i" (0),
671 				       "r" (0) : "memory");
672 		}
673 		asm volatile("ptesync": : :"memory");
674 		// POWER9 congruence-class TLBIEL leaves ERAT. Flush it now.
675 		asm volatile(PPC_RADIX_INVALIDATE_ERAT_GUEST : : :"memory");
676 	} else {
677 		for (set = 0; set < kvm->arch.tlb_sets; ++set) {
678 			/* R=0 PRS=0 RIC=0 */
679 			asm volatile(PPC_TLBIEL(%0, %4, %3, %2, %1)
680 				     : : "r" (rb), "i" (0), "i" (0), "i" (0),
681 				       "r" (0) : "memory");
682 			rb += PPC_BIT(51);	/* increment set number */
683 		}
684 		asm volatile("ptesync": : :"memory");
685 		// POWER9 congruence-class TLBIEL leaves ERAT. Flush it now.
686 		asm volatile(PPC_ISA_3_0_INVALIDATE_ERAT : : :"memory");
687 	}
688 }
689 
690 static void check_need_tlb_flush(struct kvm *kvm, int pcpu,
691 				 struct kvm_nested_guest *nested)
692 {
693 	cpumask_t *need_tlb_flush;
694 	bool all_set = true;
695 	int i;
696 
697 	if (nested)
698 		need_tlb_flush = &nested->need_tlb_flush;
699 	else
700 		need_tlb_flush = &kvm->arch.need_tlb_flush;
701 
702 	if (likely(!cpumask_test_cpu(pcpu, need_tlb_flush)))
703 		return;
704 
705 	/*
706 	 * Individual threads can come in here, but the TLB is shared between
707 	 * the 4 threads in a core, hence invalidating on one thread
708 	 * invalidates for all, so only invalidate the first time (if all bits
709 	 * were set.  The others must still execute a ptesync.
710 	 *
711 	 * If a race occurs and two threads do the TLB flush, that is not a
712 	 * problem, just sub-optimal.
713 	 */
714 	for (i = cpu_first_tlb_thread_sibling(pcpu);
715 			i <= cpu_last_tlb_thread_sibling(pcpu);
716 			i += cpu_tlb_thread_sibling_step()) {
717 		if (!cpumask_test_cpu(i, need_tlb_flush)) {
718 			all_set = false;
719 			break;
720 		}
721 	}
722 	if (all_set)
723 		flush_guest_tlb(kvm);
724 	else
725 		asm volatile("ptesync" ::: "memory");
726 
727 	/* Clear the bit after the TLB flush */
728 	cpumask_clear_cpu(pcpu, need_tlb_flush);
729 }
730 
731 unsigned long kvmppc_msr_hard_disable_set_facilities(struct kvm_vcpu *vcpu, unsigned long msr)
732 {
733 	unsigned long msr_needed = 0;
734 
735 	msr &= ~MSR_EE;
736 
737 	/* MSR bits may have been cleared by context switch so must recheck */
738 	if (IS_ENABLED(CONFIG_PPC_FPU))
739 		msr_needed |= MSR_FP;
740 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
741 		msr_needed |= MSR_VEC;
742 	if (cpu_has_feature(CPU_FTR_VSX))
743 		msr_needed |= MSR_VSX;
744 	if ((cpu_has_feature(CPU_FTR_TM) ||
745 	    cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) &&
746 			(vcpu->arch.hfscr & HFSCR_TM))
747 		msr_needed |= MSR_TM;
748 
749 	/*
750 	 * This could be combined with MSR[RI] clearing, but that expands
751 	 * the unrecoverable window. It would be better to cover unrecoverable
752 	 * with KVM bad interrupt handling rather than use MSR[RI] at all.
753 	 *
754 	 * Much more difficult and less worthwhile to combine with IR/DR
755 	 * disable.
756 	 */
757 	if ((msr & msr_needed) != msr_needed) {
758 		msr |= msr_needed;
759 		__mtmsrd(msr, 0);
760 	} else {
761 		__hard_irq_disable();
762 	}
763 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
764 
765 	return msr;
766 }
767 EXPORT_SYMBOL_GPL(kvmppc_msr_hard_disable_set_facilities);
768 
769 int kvmhv_vcpu_entry_p9(struct kvm_vcpu *vcpu, u64 time_limit, unsigned long lpcr, u64 *tb)
770 {
771 	struct p9_host_os_sprs host_os_sprs;
772 	struct kvm *kvm = vcpu->kvm;
773 	struct kvm_nested_guest *nested = vcpu->arch.nested;
774 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
775 	s64 hdec, dec;
776 	u64 purr, spurr;
777 	u64 *exsave;
778 	int trap;
779 	unsigned long msr;
780 	unsigned long host_hfscr;
781 	unsigned long host_ciabr;
782 	unsigned long host_dawr0;
783 	unsigned long host_dawrx0;
784 	unsigned long host_psscr;
785 	unsigned long host_hpsscr;
786 	unsigned long host_pidr;
787 	unsigned long host_dawr1;
788 	unsigned long host_dawrx1;
789 	unsigned long dpdes;
790 
791 	hdec = time_limit - *tb;
792 	if (hdec < 0)
793 		return BOOK3S_INTERRUPT_HV_DECREMENTER;
794 
795 	WARN_ON_ONCE(vcpu->arch.shregs.msr & MSR_HV);
796 	WARN_ON_ONCE(!(vcpu->arch.shregs.msr & MSR_ME));
797 
798 	start_timing(vcpu, &vcpu->arch.rm_entry);
799 
800 	vcpu->arch.ceded = 0;
801 
802 	/* Save MSR for restore, with EE clear. */
803 	msr = mfmsr() & ~MSR_EE;
804 
805 	host_hfscr = mfspr(SPRN_HFSCR);
806 	host_ciabr = mfspr(SPRN_CIABR);
807 	host_psscr = mfspr(SPRN_PSSCR_PR);
808 	if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
809 		host_hpsscr = mfspr(SPRN_PSSCR);
810 	host_pidr = mfspr(SPRN_PID);
811 
812 	if (dawr_enabled()) {
813 		host_dawr0 = mfspr(SPRN_DAWR0);
814 		host_dawrx0 = mfspr(SPRN_DAWRX0);
815 		if (cpu_has_feature(CPU_FTR_DAWR1)) {
816 			host_dawr1 = mfspr(SPRN_DAWR1);
817 			host_dawrx1 = mfspr(SPRN_DAWRX1);
818 		}
819 	}
820 
821 	local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
822 	local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
823 
824 	save_p9_host_os_sprs(&host_os_sprs);
825 
826 	msr = kvmppc_msr_hard_disable_set_facilities(vcpu, msr);
827 	if (lazy_irq_pending()) {
828 		trap = 0;
829 		goto out;
830 	}
831 
832 	if (unlikely(load_vcpu_state(vcpu, &host_os_sprs)))
833 		msr = mfmsr(); /* MSR may have been updated */
834 
835 	if (vc->tb_offset) {
836 		u64 new_tb = *tb + vc->tb_offset;
837 		mtspr(SPRN_TBU40, new_tb);
838 		if ((mftb() & 0xffffff) < (new_tb & 0xffffff)) {
839 			new_tb += 0x1000000;
840 			mtspr(SPRN_TBU40, new_tb);
841 		}
842 		*tb = new_tb;
843 		vc->tb_offset_applied = vc->tb_offset;
844 	}
845 
846 	mtspr(SPRN_VTB, vc->vtb);
847 	mtspr(SPRN_PURR, vcpu->arch.purr);
848 	mtspr(SPRN_SPURR, vcpu->arch.spurr);
849 
850 	if (vc->pcr)
851 		mtspr(SPRN_PCR, vc->pcr | PCR_MASK);
852 	if (vcpu->arch.doorbell_request) {
853 		vcpu->arch.doorbell_request = 0;
854 		mtspr(SPRN_DPDES, 1);
855 	}
856 
857 	if (dawr_enabled()) {
858 		if (vcpu->arch.dawr0 != host_dawr0)
859 			mtspr(SPRN_DAWR0, vcpu->arch.dawr0);
860 		if (vcpu->arch.dawrx0 != host_dawrx0)
861 			mtspr(SPRN_DAWRX0, vcpu->arch.dawrx0);
862 		if (cpu_has_feature(CPU_FTR_DAWR1)) {
863 			if (vcpu->arch.dawr1 != host_dawr1)
864 				mtspr(SPRN_DAWR1, vcpu->arch.dawr1);
865 			if (vcpu->arch.dawrx1 != host_dawrx1)
866 				mtspr(SPRN_DAWRX1, vcpu->arch.dawrx1);
867 		}
868 	}
869 	if (vcpu->arch.ciabr != host_ciabr)
870 		mtspr(SPRN_CIABR, vcpu->arch.ciabr);
871 
872 
873 	if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) {
874 		mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
875 		      (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
876 	} else {
877 		if (vcpu->arch.psscr != host_psscr)
878 			mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
879 	}
880 
881 	mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
882 
883 	mtspr(SPRN_HSRR0, vcpu->arch.regs.nip);
884 	mtspr(SPRN_HSRR1, (vcpu->arch.shregs.msr & ~MSR_HV) | MSR_ME);
885 
886 	/*
887 	 * On POWER9 DD2.1 and below, sometimes on a Hypervisor Data Storage
888 	 * Interrupt (HDSI) the HDSISR is not be updated at all.
889 	 *
890 	 * To work around this we put a canary value into the HDSISR before
891 	 * returning to a guest and then check for this canary when we take a
892 	 * HDSI. If we find the canary on a HDSI, we know the hardware didn't
893 	 * update the HDSISR. In this case we return to the guest to retake the
894 	 * HDSI which should correctly update the HDSISR the second time HDSI
895 	 * entry.
896 	 *
897 	 * The "radix prefetch bug" test can be used to test for this bug, as
898 	 * it also exists fo DD2.1 and below.
899 	 */
900 	if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
901 		mtspr(SPRN_HDSISR, HDSISR_CANARY);
902 
903 	mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
904 	mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
905 	mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
906 	mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
907 
908 	/*
909 	 * It might be preferable to load_vcpu_state here, in order to get the
910 	 * GPR/FP register loads executing in parallel with the previous mtSPR
911 	 * instructions, but for now that can't be done because the TM handling
912 	 * in load_vcpu_state can change some SPRs and vcpu state (nip, msr).
913 	 * But TM could be split out if this would be a significant benefit.
914 	 */
915 
916 	/*
917 	 * MSR[RI] does not need to be cleared (and is not, for radix guests
918 	 * with no prefetch bug), because in_guest is set. If we take a SRESET
919 	 * or MCE with in_guest set but still in HV mode, then
920 	 * kvmppc_p9_bad_interrupt handles the interrupt, which effectively
921 	 * clears MSR[RI] and doesn't return.
922 	 */
923 	WRITE_ONCE(local_paca->kvm_hstate.in_guest, KVM_GUEST_MODE_HV_P9);
924 	barrier(); /* Open in_guest critical section */
925 
926 	/*
927 	 * Hash host, hash guest, or radix guest with prefetch bug, all have
928 	 * to disable the MMU before switching to guest MMU state.
929 	 */
930 	if (!radix_enabled() || !kvm_is_radix(kvm) ||
931 			cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
932 		__mtmsrd(msr & ~(MSR_IR|MSR_DR|MSR_RI), 0);
933 
934 	save_clear_host_mmu(kvm);
935 
936 	if (kvm_is_radix(kvm))
937 		switch_mmu_to_guest_radix(kvm, vcpu, lpcr);
938 	else
939 		switch_mmu_to_guest_hpt(kvm, vcpu, lpcr);
940 
941 	/* TLBIEL uses LPID=LPIDR, so run this after setting guest LPID */
942 	check_need_tlb_flush(kvm, vc->pcpu, nested);
943 
944 	/*
945 	 * P9 suppresses the HDEC exception when LPCR[HDICE] = 0,
946 	 * so set guest LPCR (with HDICE) before writing HDEC.
947 	 */
948 	mtspr(SPRN_HDEC, hdec);
949 
950 	mtspr(SPRN_DEC, vcpu->arch.dec_expires - *tb);
951 
952 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
953 tm_return_to_guest:
954 #endif
955 	mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
956 	mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
957 	mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
958 	mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
959 
960 	accumulate_time(vcpu, &vcpu->arch.guest_time);
961 
962 	switch_pmu_to_guest(vcpu, &host_os_sprs);
963 	kvmppc_p9_enter_guest(vcpu);
964 	switch_pmu_to_host(vcpu, &host_os_sprs);
965 
966 	accumulate_time(vcpu, &vcpu->arch.rm_intr);
967 
968 	/* XXX: Could get these from r11/12 and paca exsave instead */
969 	vcpu->arch.shregs.srr0 = mfspr(SPRN_SRR0);
970 	vcpu->arch.shregs.srr1 = mfspr(SPRN_SRR1);
971 	vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
972 	vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
973 
974 	/* 0x2 bit for HSRR is only used by PR and P7/8 HV paths, clear it */
975 	trap = local_paca->kvm_hstate.scratch0 & ~0x2;
976 
977 	if (likely(trap > BOOK3S_INTERRUPT_MACHINE_CHECK))
978 		exsave = local_paca->exgen;
979 	else if (trap == BOOK3S_INTERRUPT_SYSTEM_RESET)
980 		exsave = local_paca->exnmi;
981 	else /* trap == 0x200 */
982 		exsave = local_paca->exmc;
983 
984 	vcpu->arch.regs.gpr[1] = local_paca->kvm_hstate.scratch1;
985 	vcpu->arch.regs.gpr[3] = local_paca->kvm_hstate.scratch2;
986 
987 	/*
988 	 * After reading machine check regs (DAR, DSISR, SRR0/1) and hstate
989 	 * scratch (which we need to move into exsave to make re-entrant vs
990 	 * SRESET/MCE), register state is protected from reentrancy. However
991 	 * timebase, MMU, among other state is still set to guest, so don't
992 	 * enable MSR[RI] here. It gets enabled at the end, after in_guest
993 	 * is cleared.
994 	 *
995 	 * It is possible an NMI could come in here, which is why it is
996 	 * important to save the above state early so it can be debugged.
997 	 */
998 
999 	vcpu->arch.regs.gpr[9] = exsave[EX_R9/sizeof(u64)];
1000 	vcpu->arch.regs.gpr[10] = exsave[EX_R10/sizeof(u64)];
1001 	vcpu->arch.regs.gpr[11] = exsave[EX_R11/sizeof(u64)];
1002 	vcpu->arch.regs.gpr[12] = exsave[EX_R12/sizeof(u64)];
1003 	vcpu->arch.regs.gpr[13] = exsave[EX_R13/sizeof(u64)];
1004 	vcpu->arch.ppr = exsave[EX_PPR/sizeof(u64)];
1005 	vcpu->arch.cfar = exsave[EX_CFAR/sizeof(u64)];
1006 	vcpu->arch.regs.ctr = exsave[EX_CTR/sizeof(u64)];
1007 
1008 	vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
1009 
1010 	if (unlikely(trap == BOOK3S_INTERRUPT_MACHINE_CHECK)) {
1011 		vcpu->arch.fault_dar = exsave[EX_DAR/sizeof(u64)];
1012 		vcpu->arch.fault_dsisr = exsave[EX_DSISR/sizeof(u64)];
1013 		kvmppc_realmode_machine_check(vcpu);
1014 
1015 	} else if (unlikely(trap == BOOK3S_INTERRUPT_HMI)) {
1016 		kvmppc_p9_realmode_hmi_handler(vcpu);
1017 
1018 	} else if (trap == BOOK3S_INTERRUPT_H_EMUL_ASSIST) {
1019 		vcpu->arch.emul_inst = mfspr(SPRN_HEIR);
1020 
1021 	} else if (trap == BOOK3S_INTERRUPT_H_DATA_STORAGE) {
1022 		vcpu->arch.fault_dar = exsave[EX_DAR/sizeof(u64)];
1023 		vcpu->arch.fault_dsisr = exsave[EX_DSISR/sizeof(u64)];
1024 		vcpu->arch.fault_gpa = mfspr(SPRN_ASDR);
1025 
1026 	} else if (trap == BOOK3S_INTERRUPT_H_INST_STORAGE) {
1027 		vcpu->arch.fault_gpa = mfspr(SPRN_ASDR);
1028 
1029 	} else if (trap == BOOK3S_INTERRUPT_H_FAC_UNAVAIL) {
1030 		vcpu->arch.hfscr = mfspr(SPRN_HFSCR);
1031 
1032 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1033 	/*
1034 	 * Softpatch interrupt for transactional memory emulation cases
1035 	 * on POWER9 DD2.2.  This is early in the guest exit path - we
1036 	 * haven't saved registers or done a treclaim yet.
1037 	 */
1038 	} else if (trap == BOOK3S_INTERRUPT_HV_SOFTPATCH) {
1039 		vcpu->arch.emul_inst = mfspr(SPRN_HEIR);
1040 
1041 		/*
1042 		 * The cases we want to handle here are those where the guest
1043 		 * is in real suspend mode and is trying to transition to
1044 		 * transactional mode.
1045 		 */
1046 		if (!local_paca->kvm_hstate.fake_suspend &&
1047 				(vcpu->arch.shregs.msr & MSR_TS_S)) {
1048 			if (kvmhv_p9_tm_emulation_early(vcpu)) {
1049 				/*
1050 				 * Go straight back into the guest with the
1051 				 * new NIP/MSR as set by TM emulation.
1052 				 */
1053 				mtspr(SPRN_HSRR0, vcpu->arch.regs.nip);
1054 				mtspr(SPRN_HSRR1, vcpu->arch.shregs.msr);
1055 				goto tm_return_to_guest;
1056 			}
1057 		}
1058 #endif
1059 	}
1060 
1061 	accumulate_time(vcpu, &vcpu->arch.rm_exit);
1062 
1063 	/* Advance host PURR/SPURR by the amount used by guest */
1064 	purr = mfspr(SPRN_PURR);
1065 	spurr = mfspr(SPRN_SPURR);
1066 	local_paca->kvm_hstate.host_purr += purr - vcpu->arch.purr;
1067 	local_paca->kvm_hstate.host_spurr += spurr - vcpu->arch.spurr;
1068 	vcpu->arch.purr = purr;
1069 	vcpu->arch.spurr = spurr;
1070 
1071 	vcpu->arch.ic = mfspr(SPRN_IC);
1072 	vcpu->arch.pid = mfspr(SPRN_PID);
1073 	vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
1074 
1075 	vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
1076 	vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
1077 	vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
1078 	vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
1079 
1080 	dpdes = mfspr(SPRN_DPDES);
1081 	if (dpdes)
1082 		vcpu->arch.doorbell_request = 1;
1083 
1084 	vc->vtb = mfspr(SPRN_VTB);
1085 
1086 	dec = mfspr(SPRN_DEC);
1087 	if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
1088 		dec = (s32) dec;
1089 	*tb = mftb();
1090 	vcpu->arch.dec_expires = dec + *tb;
1091 
1092 	if (vc->tb_offset_applied) {
1093 		u64 new_tb = *tb - vc->tb_offset_applied;
1094 		mtspr(SPRN_TBU40, new_tb);
1095 		if ((mftb() & 0xffffff) < (new_tb & 0xffffff)) {
1096 			new_tb += 0x1000000;
1097 			mtspr(SPRN_TBU40, new_tb);
1098 		}
1099 		*tb = new_tb;
1100 		vc->tb_offset_applied = 0;
1101 	}
1102 
1103 	save_clear_guest_mmu(kvm, vcpu);
1104 	switch_mmu_to_host(kvm, host_pidr);
1105 
1106 	/*
1107 	 * Enable MSR here in order to have facilities enabled to save
1108 	 * guest registers. This enables MMU (if we were in realmode), so
1109 	 * only switch MMU on after the MMU is switched to host, to avoid
1110 	 * the P9_RADIX_PREFETCH_BUG or hash guest context.
1111 	 */
1112 	if (IS_ENABLED(CONFIG_PPC_TRANSACTIONAL_MEM) &&
1113 			vcpu->arch.shregs.msr & MSR_TS_MASK)
1114 		msr |= MSR_TS_S;
1115 	__mtmsrd(msr, 0);
1116 
1117 	store_vcpu_state(vcpu);
1118 
1119 	mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr);
1120 	mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr);
1121 
1122 	if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) {
1123 		/* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
1124 		mtspr(SPRN_PSSCR, host_hpsscr |
1125 		      (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
1126 	}
1127 
1128 	mtspr(SPRN_HFSCR, host_hfscr);
1129 	if (vcpu->arch.ciabr != host_ciabr)
1130 		mtspr(SPRN_CIABR, host_ciabr);
1131 
1132 	if (dawr_enabled()) {
1133 		if (vcpu->arch.dawr0 != host_dawr0)
1134 			mtspr(SPRN_DAWR0, host_dawr0);
1135 		if (vcpu->arch.dawrx0 != host_dawrx0)
1136 			mtspr(SPRN_DAWRX0, host_dawrx0);
1137 		if (cpu_has_feature(CPU_FTR_DAWR1)) {
1138 			if (vcpu->arch.dawr1 != host_dawr1)
1139 				mtspr(SPRN_DAWR1, host_dawr1);
1140 			if (vcpu->arch.dawrx1 != host_dawrx1)
1141 				mtspr(SPRN_DAWRX1, host_dawrx1);
1142 		}
1143 	}
1144 
1145 	if (dpdes)
1146 		mtspr(SPRN_DPDES, 0);
1147 	if (vc->pcr)
1148 		mtspr(SPRN_PCR, PCR_MASK);
1149 
1150 	/* HDEC must be at least as large as DEC, so decrementer_max fits */
1151 	mtspr(SPRN_HDEC, decrementer_max);
1152 
1153 	timer_rearm_host_dec(*tb);
1154 
1155 	restore_p9_host_os_sprs(vcpu, &host_os_sprs);
1156 
1157 	barrier(); /* Close in_guest critical section */
1158 	WRITE_ONCE(local_paca->kvm_hstate.in_guest, KVM_GUEST_MODE_NONE);
1159 	/* Interrupts are recoverable at this point */
1160 
1161 	/*
1162 	 * cp_abort is required if the processor supports local copy-paste
1163 	 * to clear the copy buffer that was under control of the guest.
1164 	 */
1165 	if (cpu_has_feature(CPU_FTR_ARCH_31))
1166 		asm volatile(PPC_CP_ABORT);
1167 
1168 out:
1169 	end_timing(vcpu);
1170 
1171 	return trap;
1172 }
1173 EXPORT_SYMBOL_GPL(kvmhv_vcpu_entry_p9);
1174