xref: /linux/arch/powerpc/kvm/book3s_hv_rmhandlers.S (revision 1f2367a39f17bd553a75e179a747f9b257bc9478)
1/*
2 * This program is free software; you can redistribute it and/or modify
3 * it under the terms of the GNU General Public License, version 2, as
4 * published by the Free Software Foundation.
5 *
6 * This program is distributed in the hope that it will be useful,
7 * but WITHOUT ANY WARRANTY; without even the implied warranty of
8 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9 * GNU General Public License for more details.
10 *
11 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
12 *
13 * Derived from book3s_rmhandlers.S and other files, which are:
14 *
15 * Copyright SUSE Linux Products GmbH 2009
16 *
17 * Authors: Alexander Graf <agraf@suse.de>
18 */
19
20#include <asm/ppc_asm.h>
21#include <asm/kvm_asm.h>
22#include <asm/reg.h>
23#include <asm/mmu.h>
24#include <asm/page.h>
25#include <asm/ptrace.h>
26#include <asm/hvcall.h>
27#include <asm/asm-offsets.h>
28#include <asm/exception-64s.h>
29#include <asm/kvm_book3s_asm.h>
30#include <asm/book3s/64/mmu-hash.h>
31#include <asm/export.h>
32#include <asm/tm.h>
33#include <asm/opal.h>
34#include <asm/xive-regs.h>
35#include <asm/thread_info.h>
36#include <asm/asm-compat.h>
37#include <asm/feature-fixups.h>
38
39/* Sign-extend HDEC if not on POWER9 */
40#define EXTEND_HDEC(reg)			\
41BEGIN_FTR_SECTION;				\
42	extsw	reg, reg;			\
43END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
44
45/* Values in HSTATE_NAPPING(r13) */
46#define NAPPING_CEDE	1
47#define NAPPING_NOVCPU	2
48
49/* Stack frame offsets for kvmppc_hv_entry */
50#define SFS			208
51#define STACK_SLOT_TRAP		(SFS-4)
52#define STACK_SLOT_SHORT_PATH	(SFS-8)
53#define STACK_SLOT_TID		(SFS-16)
54#define STACK_SLOT_PSSCR	(SFS-24)
55#define STACK_SLOT_PID		(SFS-32)
56#define STACK_SLOT_IAMR		(SFS-40)
57#define STACK_SLOT_CIABR	(SFS-48)
58#define STACK_SLOT_DAWR		(SFS-56)
59#define STACK_SLOT_DAWRX	(SFS-64)
60#define STACK_SLOT_HFSCR	(SFS-72)
61#define STACK_SLOT_AMR		(SFS-80)
62#define STACK_SLOT_UAMOR	(SFS-88)
63/* the following is used by the P9 short path */
64#define STACK_SLOT_NVGPRS	(SFS-152)	/* 18 gprs */
65
66/*
67 * Call kvmppc_hv_entry in real mode.
68 * Must be called with interrupts hard-disabled.
69 *
70 * Input Registers:
71 *
72 * LR = return address to continue at after eventually re-enabling MMU
73 */
74_GLOBAL_TOC(kvmppc_hv_entry_trampoline)
75	mflr	r0
76	std	r0, PPC_LR_STKOFF(r1)
77	stdu	r1, -112(r1)
78	mfmsr	r10
79	std	r10, HSTATE_HOST_MSR(r13)
80	LOAD_REG_ADDR(r5, kvmppc_call_hv_entry)
81	li	r0,MSR_RI
82	andc	r0,r10,r0
83	li	r6,MSR_IR | MSR_DR
84	andc	r6,r10,r6
85	mtmsrd	r0,1		/* clear RI in MSR */
86	mtsrr0	r5
87	mtsrr1	r6
88	RFI_TO_KERNEL
89
90kvmppc_call_hv_entry:
91BEGIN_FTR_SECTION
92	/* On P9, do LPCR setting, if necessary */
93	ld	r3, HSTATE_SPLIT_MODE(r13)
94	cmpdi	r3, 0
95	beq	46f
96	lwz	r4, KVM_SPLIT_DO_SET(r3)
97	cmpwi	r4, 0
98	beq	46f
99	bl	kvmhv_p9_set_lpcr
100	nop
10146:
102END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
103
104	ld	r4, HSTATE_KVM_VCPU(r13)
105	bl	kvmppc_hv_entry
106
107	/* Back from guest - restore host state and return to caller */
108
109BEGIN_FTR_SECTION
110	/* Restore host DABR and DABRX */
111	ld	r5,HSTATE_DABR(r13)
112	li	r6,7
113	mtspr	SPRN_DABR,r5
114	mtspr	SPRN_DABRX,r6
115END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
116
117	/* Restore SPRG3 */
118	ld	r3,PACA_SPRG_VDSO(r13)
119	mtspr	SPRN_SPRG_VDSO_WRITE,r3
120
121	/* Reload the host's PMU registers */
122	bl	kvmhv_load_host_pmu
123
124	/*
125	 * Reload DEC.  HDEC interrupts were disabled when
126	 * we reloaded the host's LPCR value.
127	 */
128	ld	r3, HSTATE_DECEXP(r13)
129	mftb	r4
130	subf	r4, r4, r3
131	mtspr	SPRN_DEC, r4
132
133	/* hwthread_req may have got set by cede or no vcpu, so clear it */
134	li	r0, 0
135	stb	r0, HSTATE_HWTHREAD_REQ(r13)
136
137	/*
138	 * For external interrupts we need to call the Linux
139	 * handler to process the interrupt. We do that by jumping
140	 * to absolute address 0x500 for external interrupts.
141	 * The [h]rfid at the end of the handler will return to
142	 * the book3s_hv_interrupts.S code. For other interrupts
143	 * we do the rfid to get back to the book3s_hv_interrupts.S
144	 * code here.
145	 */
146	ld	r8, 112+PPC_LR_STKOFF(r1)
147	addi	r1, r1, 112
148	ld	r7, HSTATE_HOST_MSR(r13)
149
150	/* Return the trap number on this thread as the return value */
151	mr	r3, r12
152
153	/*
154	 * If we came back from the guest via a relocation-on interrupt,
155	 * we will be in virtual mode at this point, which makes it a
156	 * little easier to get back to the caller.
157	 */
158	mfmsr	r0
159	andi.	r0, r0, MSR_IR		/* in real mode? */
160	bne	.Lvirt_return
161
162	/* RFI into the highmem handler */
163	mfmsr	r6
164	li	r0, MSR_RI
165	andc	r6, r6, r0
166	mtmsrd	r6, 1			/* Clear RI in MSR */
167	mtsrr0	r8
168	mtsrr1	r7
169	RFI_TO_KERNEL
170
171	/* Virtual-mode return */
172.Lvirt_return:
173	mtlr	r8
174	blr
175
176kvmppc_primary_no_guest:
177	/* We handle this much like a ceded vcpu */
178	/* put the HDEC into the DEC, since HDEC interrupts don't wake us */
179	/* HDEC may be larger than DEC for arch >= v3.00, but since the */
180	/* HDEC value came from DEC in the first place, it will fit */
181	mfspr	r3, SPRN_HDEC
182	mtspr	SPRN_DEC, r3
183	/*
184	 * Make sure the primary has finished the MMU switch.
185	 * We should never get here on a secondary thread, but
186	 * check it for robustness' sake.
187	 */
188	ld	r5, HSTATE_KVM_VCORE(r13)
18965:	lbz	r0, VCORE_IN_GUEST(r5)
190	cmpwi	r0, 0
191	beq	65b
192	/* Set LPCR. */
193	ld	r8,VCORE_LPCR(r5)
194	mtspr	SPRN_LPCR,r8
195	isync
196	/* set our bit in napping_threads */
197	ld	r5, HSTATE_KVM_VCORE(r13)
198	lbz	r7, HSTATE_PTID(r13)
199	li	r0, 1
200	sld	r0, r0, r7
201	addi	r6, r5, VCORE_NAPPING_THREADS
2021:	lwarx	r3, 0, r6
203	or	r3, r3, r0
204	stwcx.	r3, 0, r6
205	bne	1b
206	/* order napping_threads update vs testing entry_exit_map */
207	isync
208	li	r12, 0
209	lwz	r7, VCORE_ENTRY_EXIT(r5)
210	cmpwi	r7, 0x100
211	bge	kvm_novcpu_exit	/* another thread already exiting */
212	li	r3, NAPPING_NOVCPU
213	stb	r3, HSTATE_NAPPING(r13)
214
215	li	r3, 0		/* Don't wake on privileged (OS) doorbell */
216	b	kvm_do_nap
217
218/*
219 * kvm_novcpu_wakeup
220 *	Entered from kvm_start_guest if kvm_hstate.napping is set
221 *	to NAPPING_NOVCPU
222 *		r2 = kernel TOC
223 *		r13 = paca
224 */
225kvm_novcpu_wakeup:
226	ld	r1, HSTATE_HOST_R1(r13)
227	ld	r5, HSTATE_KVM_VCORE(r13)
228	li	r0, 0
229	stb	r0, HSTATE_NAPPING(r13)
230
231	/* check the wake reason */
232	bl	kvmppc_check_wake_reason
233
234	/*
235	 * Restore volatile registers since we could have called
236	 * a C routine in kvmppc_check_wake_reason.
237	 *	r5 = VCORE
238	 */
239	ld	r5, HSTATE_KVM_VCORE(r13)
240
241	/* see if any other thread is already exiting */
242	lwz	r0, VCORE_ENTRY_EXIT(r5)
243	cmpwi	r0, 0x100
244	bge	kvm_novcpu_exit
245
246	/* clear our bit in napping_threads */
247	lbz	r7, HSTATE_PTID(r13)
248	li	r0, 1
249	sld	r0, r0, r7
250	addi	r6, r5, VCORE_NAPPING_THREADS
2514:	lwarx	r7, 0, r6
252	andc	r7, r7, r0
253	stwcx.	r7, 0, r6
254	bne	4b
255
256	/* See if the wake reason means we need to exit */
257	cmpdi	r3, 0
258	bge	kvm_novcpu_exit
259
260	/* See if our timeslice has expired (HDEC is negative) */
261	mfspr	r0, SPRN_HDEC
262	EXTEND_HDEC(r0)
263	li	r12, BOOK3S_INTERRUPT_HV_DECREMENTER
264	cmpdi	r0, 0
265	blt	kvm_novcpu_exit
266
267	/* Got an IPI but other vcpus aren't yet exiting, must be a latecomer */
268	ld	r4, HSTATE_KVM_VCPU(r13)
269	cmpdi	r4, 0
270	beq	kvmppc_primary_no_guest
271
272#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
273	addi	r3, r4, VCPU_TB_RMENTRY
274	bl	kvmhv_start_timing
275#endif
276	b	kvmppc_got_guest
277
278kvm_novcpu_exit:
279#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
280	ld	r4, HSTATE_KVM_VCPU(r13)
281	cmpdi	r4, 0
282	beq	13f
283	addi	r3, r4, VCPU_TB_RMEXIT
284	bl	kvmhv_accumulate_time
285#endif
28613:	mr	r3, r12
287	stw	r12, STACK_SLOT_TRAP(r1)
288	bl	kvmhv_commence_exit
289	nop
290	b	kvmhv_switch_to_host
291
292/*
293 * We come in here when wakened from nap mode.
294 * Relocation is off and most register values are lost.
295 * r13 points to the PACA.
296 * r3 contains the SRR1 wakeup value, SRR1 is trashed.
297 */
298	.globl	kvm_start_guest
299kvm_start_guest:
300	/* Set runlatch bit the minute you wake up from nap */
301	mfspr	r0, SPRN_CTRLF
302	ori 	r0, r0, 1
303	mtspr	SPRN_CTRLT, r0
304
305	/*
306	 * Could avoid this and pass it through in r3. For now,
307	 * code expects it to be in SRR1.
308	 */
309	mtspr	SPRN_SRR1,r3
310
311	ld	r2,PACATOC(r13)
312
313	li	r0,0
314	stb	r0,PACA_FTRACE_ENABLED(r13)
315
316	li	r0,KVM_HWTHREAD_IN_KVM
317	stb	r0,HSTATE_HWTHREAD_STATE(r13)
318
319	/* NV GPR values from power7_idle() will no longer be valid */
320	li	r0,1
321	stb	r0,PACA_NAPSTATELOST(r13)
322
323	/* were we napping due to cede? */
324	lbz	r0,HSTATE_NAPPING(r13)
325	cmpwi	r0,NAPPING_CEDE
326	beq	kvm_end_cede
327	cmpwi	r0,NAPPING_NOVCPU
328	beq	kvm_novcpu_wakeup
329
330	ld	r1,PACAEMERGSP(r13)
331	subi	r1,r1,STACK_FRAME_OVERHEAD
332
333	/*
334	 * We weren't napping due to cede, so this must be a secondary
335	 * thread being woken up to run a guest, or being woken up due
336	 * to a stray IPI.  (Or due to some machine check or hypervisor
337	 * maintenance interrupt while the core is in KVM.)
338	 */
339
340	/* Check the wake reason in SRR1 to see why we got here */
341	bl	kvmppc_check_wake_reason
342	/*
343	 * kvmppc_check_wake_reason could invoke a C routine, but we
344	 * have no volatile registers to restore when we return.
345	 */
346
347	cmpdi	r3, 0
348	bge	kvm_no_guest
349
350	/* get vcore pointer, NULL if we have nothing to run */
351	ld	r5,HSTATE_KVM_VCORE(r13)
352	cmpdi	r5,0
353	/* if we have no vcore to run, go back to sleep */
354	beq	kvm_no_guest
355
356kvm_secondary_got_guest:
357
358	/* Set HSTATE_DSCR(r13) to something sensible */
359	ld	r6, PACA_DSCR_DEFAULT(r13)
360	std	r6, HSTATE_DSCR(r13)
361
362	/* On thread 0 of a subcore, set HDEC to max */
363	lbz	r4, HSTATE_PTID(r13)
364	cmpwi	r4, 0
365	bne	63f
366	LOAD_REG_ADDR(r6, decrementer_max)
367	ld	r6, 0(r6)
368	mtspr	SPRN_HDEC, r6
369	/* and set per-LPAR registers, if doing dynamic micro-threading */
370	ld	r6, HSTATE_SPLIT_MODE(r13)
371	cmpdi	r6, 0
372	beq	63f
373BEGIN_FTR_SECTION
374	ld	r0, KVM_SPLIT_RPR(r6)
375	mtspr	SPRN_RPR, r0
376	ld	r0, KVM_SPLIT_PMMAR(r6)
377	mtspr	SPRN_PMMAR, r0
378	ld	r0, KVM_SPLIT_LDBAR(r6)
379	mtspr	SPRN_LDBAR, r0
380	isync
381FTR_SECTION_ELSE
382	/* On P9 we use the split_info for coordinating LPCR changes */
383	lwz	r4, KVM_SPLIT_DO_SET(r6)
384	cmpwi	r4, 0
385	beq	1f
386	mr	r3, r6
387	bl	kvmhv_p9_set_lpcr
388	nop
3891:
390ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
39163:
392	/* Order load of vcpu after load of vcore */
393	lwsync
394	ld	r4, HSTATE_KVM_VCPU(r13)
395	bl	kvmppc_hv_entry
396
397	/* Back from the guest, go back to nap */
398	/* Clear our vcpu and vcore pointers so we don't come back in early */
399	li	r0, 0
400	std	r0, HSTATE_KVM_VCPU(r13)
401	/*
402	 * Once we clear HSTATE_KVM_VCORE(r13), the code in
403	 * kvmppc_run_core() is going to assume that all our vcpu
404	 * state is visible in memory.  This lwsync makes sure
405	 * that that is true.
406	 */
407	lwsync
408	std	r0, HSTATE_KVM_VCORE(r13)
409
410	/*
411	 * All secondaries exiting guest will fall through this path.
412	 * Before proceeding, just check for HMI interrupt and
413	 * invoke opal hmi handler. By now we are sure that the
414	 * primary thread on this core/subcore has already made partition
415	 * switch/TB resync and we are good to call opal hmi handler.
416	 */
417	cmpwi	r12, BOOK3S_INTERRUPT_HMI
418	bne	kvm_no_guest
419
420	li	r3,0			/* NULL argument */
421	bl	hmi_exception_realmode
422/*
423 * At this point we have finished executing in the guest.
424 * We need to wait for hwthread_req to become zero, since
425 * we may not turn on the MMU while hwthread_req is non-zero.
426 * While waiting we also need to check if we get given a vcpu to run.
427 */
428kvm_no_guest:
429	lbz	r3, HSTATE_HWTHREAD_REQ(r13)
430	cmpwi	r3, 0
431	bne	53f
432	HMT_MEDIUM
433	li	r0, KVM_HWTHREAD_IN_KERNEL
434	stb	r0, HSTATE_HWTHREAD_STATE(r13)
435	/* need to recheck hwthread_req after a barrier, to avoid race */
436	sync
437	lbz	r3, HSTATE_HWTHREAD_REQ(r13)
438	cmpwi	r3, 0
439	bne	54f
440/*
441 * We jump to pnv_wakeup_loss, which will return to the caller
442 * of power7_nap in the powernv cpu offline loop.  The value we
443 * put in r3 becomes the return value for power7_nap. pnv_wakeup_loss
444 * requires SRR1 in r12.
445 */
446	li	r3, LPCR_PECE0
447	mfspr	r4, SPRN_LPCR
448	rlwimi	r4, r3, 0, LPCR_PECE0 | LPCR_PECE1
449	mtspr	SPRN_LPCR, r4
450	li	r3, 0
451	mfspr	r12,SPRN_SRR1
452	b	pnv_wakeup_loss
453
45453:	HMT_LOW
455	ld	r5, HSTATE_KVM_VCORE(r13)
456	cmpdi	r5, 0
457	bne	60f
458	ld	r3, HSTATE_SPLIT_MODE(r13)
459	cmpdi	r3, 0
460	beq	kvm_no_guest
461	lwz	r0, KVM_SPLIT_DO_SET(r3)
462	cmpwi	r0, 0
463	bne	kvmhv_do_set
464	lwz	r0, KVM_SPLIT_DO_RESTORE(r3)
465	cmpwi	r0, 0
466	bne	kvmhv_do_restore
467	lbz	r0, KVM_SPLIT_DO_NAP(r3)
468	cmpwi	r0, 0
469	beq	kvm_no_guest
470	HMT_MEDIUM
471	b	kvm_unsplit_nap
47260:	HMT_MEDIUM
473	b	kvm_secondary_got_guest
474
47554:	li	r0, KVM_HWTHREAD_IN_KVM
476	stb	r0, HSTATE_HWTHREAD_STATE(r13)
477	b	kvm_no_guest
478
479kvmhv_do_set:
480	/* Set LPCR, LPIDR etc. on P9 */
481	HMT_MEDIUM
482	bl	kvmhv_p9_set_lpcr
483	nop
484	b	kvm_no_guest
485
486kvmhv_do_restore:
487	HMT_MEDIUM
488	bl	kvmhv_p9_restore_lpcr
489	nop
490	b	kvm_no_guest
491
492/*
493 * Here the primary thread is trying to return the core to
494 * whole-core mode, so we need to nap.
495 */
496kvm_unsplit_nap:
497	/*
498	 * When secondaries are napping in kvm_unsplit_nap() with
499	 * hwthread_req = 1, HMI goes ignored even though subcores are
500	 * already exited the guest. Hence HMI keeps waking up secondaries
501	 * from nap in a loop and secondaries always go back to nap since
502	 * no vcore is assigned to them. This makes impossible for primary
503	 * thread to get hold of secondary threads resulting into a soft
504	 * lockup in KVM path.
505	 *
506	 * Let us check if HMI is pending and handle it before we go to nap.
507	 */
508	cmpwi	r12, BOOK3S_INTERRUPT_HMI
509	bne	55f
510	li	r3, 0			/* NULL argument */
511	bl	hmi_exception_realmode
51255:
513	/*
514	 * Ensure that secondary doesn't nap when it has
515	 * its vcore pointer set.
516	 */
517	sync		/* matches smp_mb() before setting split_info.do_nap */
518	ld	r0, HSTATE_KVM_VCORE(r13)
519	cmpdi	r0, 0
520	bne	kvm_no_guest
521	/* clear any pending message */
522BEGIN_FTR_SECTION
523	lis	r6, (PPC_DBELL_SERVER << (63-36))@h
524	PPC_MSGCLR(6)
525END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
526	/* Set kvm_split_mode.napped[tid] = 1 */
527	ld	r3, HSTATE_SPLIT_MODE(r13)
528	li	r0, 1
529	lbz	r4, HSTATE_TID(r13)
530	addi	r4, r4, KVM_SPLIT_NAPPED
531	stbx	r0, r3, r4
532	/* Check the do_nap flag again after setting napped[] */
533	sync
534	lbz	r0, KVM_SPLIT_DO_NAP(r3)
535	cmpwi	r0, 0
536	beq	57f
537	li	r3, (LPCR_PECEDH | LPCR_PECE0) >> 4
538	mfspr	r5, SPRN_LPCR
539	rlwimi	r5, r3, 4, (LPCR_PECEDP | LPCR_PECEDH | LPCR_PECE0 | LPCR_PECE1)
540	b	kvm_nap_sequence
541
54257:	li	r0, 0
543	stbx	r0, r3, r4
544	b	kvm_no_guest
545
546/******************************************************************************
547 *                                                                            *
548 *                               Entry code                                   *
549 *                                                                            *
550 *****************************************************************************/
551
552.global kvmppc_hv_entry
553kvmppc_hv_entry:
554
555	/* Required state:
556	 *
557	 * R4 = vcpu pointer (or NULL)
558	 * MSR = ~IR|DR
559	 * R13 = PACA
560	 * R1 = host R1
561	 * R2 = TOC
562	 * all other volatile GPRS = free
563	 * Does not preserve non-volatile GPRs or CR fields
564	 */
565	mflr	r0
566	std	r0, PPC_LR_STKOFF(r1)
567	stdu	r1, -SFS(r1)
568
569	/* Save R1 in the PACA */
570	std	r1, HSTATE_HOST_R1(r13)
571
572	li	r6, KVM_GUEST_MODE_HOST_HV
573	stb	r6, HSTATE_IN_GUEST(r13)
574
575#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
576	/* Store initial timestamp */
577	cmpdi	r4, 0
578	beq	1f
579	addi	r3, r4, VCPU_TB_RMENTRY
580	bl	kvmhv_start_timing
5811:
582#endif
583
584	/* Use cr7 as an indication of radix mode */
585	ld	r5, HSTATE_KVM_VCORE(r13)
586	ld	r9, VCORE_KVM(r5)	/* pointer to struct kvm */
587	lbz	r0, KVM_RADIX(r9)
588	cmpwi	cr7, r0, 0
589
590	/*
591	 * POWER7/POWER8 host -> guest partition switch code.
592	 * We don't have to lock against concurrent tlbies,
593	 * but we do have to coordinate across hardware threads.
594	 */
595	/* Set bit in entry map iff exit map is zero. */
596	li	r7, 1
597	lbz	r6, HSTATE_PTID(r13)
598	sld	r7, r7, r6
599	addi	r8, r5, VCORE_ENTRY_EXIT
60021:	lwarx	r3, 0, r8
601	cmpwi	r3, 0x100		/* any threads starting to exit? */
602	bge	secondary_too_late	/* if so we're too late to the party */
603	or	r3, r3, r7
604	stwcx.	r3, 0, r8
605	bne	21b
606
607	/* Primary thread switches to guest partition. */
608	cmpwi	r6,0
609	bne	10f
610
611	/* Radix has already switched LPID and flushed core TLB */
612	bne	cr7, 22f
613
614	lwz	r7,KVM_LPID(r9)
615BEGIN_FTR_SECTION
616	ld	r6,KVM_SDR1(r9)
617	li	r0,LPID_RSVD		/* switch to reserved LPID */
618	mtspr	SPRN_LPID,r0
619	ptesync
620	mtspr	SPRN_SDR1,r6		/* switch to partition page table */
621END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
622	mtspr	SPRN_LPID,r7
623	isync
624
625	/* See if we need to flush the TLB. Hash has to be done in RM */
626	lhz	r6,PACAPACAINDEX(r13)	/* test_bit(cpu, need_tlb_flush) */
627BEGIN_FTR_SECTION
628	/*
629	 * On POWER9, individual threads can come in here, but the
630	 * TLB is shared between the 4 threads in a core, hence
631	 * invalidating on one thread invalidates for all.
632	 * Thus we make all 4 threads use the same bit here.
633	 */
634	clrrdi	r6,r6,2
635END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
636	clrldi	r7,r6,64-6		/* extract bit number (6 bits) */
637	srdi	r6,r6,6			/* doubleword number */
638	sldi	r6,r6,3			/* address offset */
639	add	r6,r6,r9
640	addi	r6,r6,KVM_NEED_FLUSH	/* dword in kvm->arch.need_tlb_flush */
641	li	r8,1
642	sld	r8,r8,r7
643	ld	r7,0(r6)
644	and.	r7,r7,r8
645	beq	22f
646	/* Flush the TLB of any entries for this LPID */
647	lwz	r0,KVM_TLB_SETS(r9)
648	mtctr	r0
649	li	r7,0x800		/* IS field = 0b10 */
650	ptesync
651	li	r0,0			/* RS for P9 version of tlbiel */
65228:	tlbiel	r7			/* On P9, rs=0, RIC=0, PRS=0, R=0 */
653	addi	r7,r7,0x1000
654	bdnz	28b
655	ptesync
65623:	ldarx	r7,0,r6			/* clear the bit after TLB flushed */
657	andc	r7,r7,r8
658	stdcx.	r7,0,r6
659	bne	23b
660
661	/* Add timebase offset onto timebase */
66222:	ld	r8,VCORE_TB_OFFSET(r5)
663	cmpdi	r8,0
664	beq	37f
665	std	r8, VCORE_TB_OFFSET_APPL(r5)
666	mftb	r6		/* current host timebase */
667	add	r8,r8,r6
668	mtspr	SPRN_TBU40,r8	/* update upper 40 bits */
669	mftb	r7		/* check if lower 24 bits overflowed */
670	clrldi	r6,r6,40
671	clrldi	r7,r7,40
672	cmpld	r7,r6
673	bge	37f
674	addis	r8,r8,0x100	/* if so, increment upper 40 bits */
675	mtspr	SPRN_TBU40,r8
676
677	/* Load guest PCR value to select appropriate compat mode */
67837:	ld	r7, VCORE_PCR(r5)
679	cmpdi	r7, 0
680	beq	38f
681	mtspr	SPRN_PCR, r7
68238:
683
684BEGIN_FTR_SECTION
685	/* DPDES and VTB are shared between threads */
686	ld	r8, VCORE_DPDES(r5)
687	ld	r7, VCORE_VTB(r5)
688	mtspr	SPRN_DPDES, r8
689	mtspr	SPRN_VTB, r7
690END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
691
692	/* Mark the subcore state as inside guest */
693	bl	kvmppc_subcore_enter_guest
694	nop
695	ld	r5, HSTATE_KVM_VCORE(r13)
696	ld	r4, HSTATE_KVM_VCPU(r13)
697	li	r0,1
698	stb	r0,VCORE_IN_GUEST(r5)	/* signal secondaries to continue */
699
700	/* Do we have a guest vcpu to run? */
70110:	cmpdi	r4, 0
702	beq	kvmppc_primary_no_guest
703kvmppc_got_guest:
704	/* Increment yield count if they have a VPA */
705	ld	r3, VCPU_VPA(r4)
706	cmpdi	r3, 0
707	beq	25f
708	li	r6, LPPACA_YIELDCOUNT
709	LWZX_BE	r5, r3, r6
710	addi	r5, r5, 1
711	STWX_BE	r5, r3, r6
712	li	r6, 1
713	stb	r6, VCPU_VPA_DIRTY(r4)
71425:
715
716	/* Save purr/spurr */
717	mfspr	r5,SPRN_PURR
718	mfspr	r6,SPRN_SPURR
719	std	r5,HSTATE_PURR(r13)
720	std	r6,HSTATE_SPURR(r13)
721	ld	r7,VCPU_PURR(r4)
722	ld	r8,VCPU_SPURR(r4)
723	mtspr	SPRN_PURR,r7
724	mtspr	SPRN_SPURR,r8
725
726	/* Save host values of some registers */
727BEGIN_FTR_SECTION
728	mfspr	r5, SPRN_TIDR
729	mfspr	r6, SPRN_PSSCR
730	mfspr	r7, SPRN_PID
731	std	r5, STACK_SLOT_TID(r1)
732	std	r6, STACK_SLOT_PSSCR(r1)
733	std	r7, STACK_SLOT_PID(r1)
734	mfspr	r5, SPRN_HFSCR
735	std	r5, STACK_SLOT_HFSCR(r1)
736END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
737BEGIN_FTR_SECTION
738	mfspr	r5, SPRN_CIABR
739	mfspr	r6, SPRN_DAWR
740	mfspr	r7, SPRN_DAWRX
741	mfspr	r8, SPRN_IAMR
742	std	r5, STACK_SLOT_CIABR(r1)
743	std	r6, STACK_SLOT_DAWR(r1)
744	std	r7, STACK_SLOT_DAWRX(r1)
745	std	r8, STACK_SLOT_IAMR(r1)
746END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
747
748	mfspr	r5, SPRN_AMR
749	std	r5, STACK_SLOT_AMR(r1)
750	mfspr	r6, SPRN_UAMOR
751	std	r6, STACK_SLOT_UAMOR(r1)
752
753BEGIN_FTR_SECTION
754	/* Set partition DABR */
755	/* Do this before re-enabling PMU to avoid P7 DABR corruption bug */
756	lwz	r5,VCPU_DABRX(r4)
757	ld	r6,VCPU_DABR(r4)
758	mtspr	SPRN_DABRX,r5
759	mtspr	SPRN_DABR,r6
760	isync
761END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
762
763#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
764/*
765 * Branch around the call if both CPU_FTR_TM and
766 * CPU_FTR_P9_TM_HV_ASSIST are off.
767 */
768BEGIN_FTR_SECTION
769	b	91f
770END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
771	/*
772	 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
773	 */
774	mr      r3, r4
775	ld      r4, VCPU_MSR(r3)
776	li	r5, 0			/* don't preserve non-vol regs */
777	bl	kvmppc_restore_tm_hv
778	nop
779	ld	r4, HSTATE_KVM_VCPU(r13)
78091:
781#endif
782
783	/* Load guest PMU registers; r4 = vcpu pointer here */
784	mr	r3, r4
785	bl	kvmhv_load_guest_pmu
786
787	/* Load up FP, VMX and VSX registers */
788	ld	r4, HSTATE_KVM_VCPU(r13)
789	bl	kvmppc_load_fp
790
791	ld	r14, VCPU_GPR(R14)(r4)
792	ld	r15, VCPU_GPR(R15)(r4)
793	ld	r16, VCPU_GPR(R16)(r4)
794	ld	r17, VCPU_GPR(R17)(r4)
795	ld	r18, VCPU_GPR(R18)(r4)
796	ld	r19, VCPU_GPR(R19)(r4)
797	ld	r20, VCPU_GPR(R20)(r4)
798	ld	r21, VCPU_GPR(R21)(r4)
799	ld	r22, VCPU_GPR(R22)(r4)
800	ld	r23, VCPU_GPR(R23)(r4)
801	ld	r24, VCPU_GPR(R24)(r4)
802	ld	r25, VCPU_GPR(R25)(r4)
803	ld	r26, VCPU_GPR(R26)(r4)
804	ld	r27, VCPU_GPR(R27)(r4)
805	ld	r28, VCPU_GPR(R28)(r4)
806	ld	r29, VCPU_GPR(R29)(r4)
807	ld	r30, VCPU_GPR(R30)(r4)
808	ld	r31, VCPU_GPR(R31)(r4)
809
810	/* Switch DSCR to guest value */
811	ld	r5, VCPU_DSCR(r4)
812	mtspr	SPRN_DSCR, r5
813
814BEGIN_FTR_SECTION
815	/* Skip next section on POWER7 */
816	b	8f
817END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
818	/* Load up POWER8-specific registers */
819	ld	r5, VCPU_IAMR(r4)
820	lwz	r6, VCPU_PSPB(r4)
821	ld	r7, VCPU_FSCR(r4)
822	mtspr	SPRN_IAMR, r5
823	mtspr	SPRN_PSPB, r6
824	mtspr	SPRN_FSCR, r7
825	ld	r5, VCPU_DAWR(r4)
826	ld	r6, VCPU_DAWRX(r4)
827	ld	r7, VCPU_CIABR(r4)
828	ld	r8, VCPU_TAR(r4)
829	/*
830	 * Handle broken DAWR case by not writing it. This means we
831	 * can still store the DAWR register for migration.
832	 */
833BEGIN_FTR_SECTION
834	mtspr	SPRN_DAWR, r5
835	mtspr	SPRN_DAWRX, r6
836END_FTR_SECTION_IFSET(CPU_FTR_DAWR)
837	mtspr	SPRN_CIABR, r7
838	mtspr	SPRN_TAR, r8
839	ld	r5, VCPU_IC(r4)
840	ld	r8, VCPU_EBBHR(r4)
841	mtspr	SPRN_IC, r5
842	mtspr	SPRN_EBBHR, r8
843	ld	r5, VCPU_EBBRR(r4)
844	ld	r6, VCPU_BESCR(r4)
845	lwz	r7, VCPU_GUEST_PID(r4)
846	ld	r8, VCPU_WORT(r4)
847	mtspr	SPRN_EBBRR, r5
848	mtspr	SPRN_BESCR, r6
849	mtspr	SPRN_PID, r7
850	mtspr	SPRN_WORT, r8
851BEGIN_FTR_SECTION
852	/* POWER8-only registers */
853	ld	r5, VCPU_TCSCR(r4)
854	ld	r6, VCPU_ACOP(r4)
855	ld	r7, VCPU_CSIGR(r4)
856	ld	r8, VCPU_TACR(r4)
857	mtspr	SPRN_TCSCR, r5
858	mtspr	SPRN_ACOP, r6
859	mtspr	SPRN_CSIGR, r7
860	mtspr	SPRN_TACR, r8
861	nop
862FTR_SECTION_ELSE
863	/* POWER9-only registers */
864	ld	r5, VCPU_TID(r4)
865	ld	r6, VCPU_PSSCR(r4)
866	lbz	r8, HSTATE_FAKE_SUSPEND(r13)
867	oris	r6, r6, PSSCR_EC@h	/* This makes stop trap to HV */
868	rldimi	r6, r8, PSSCR_FAKE_SUSPEND_LG, 63 - PSSCR_FAKE_SUSPEND_LG
869	ld	r7, VCPU_HFSCR(r4)
870	mtspr	SPRN_TIDR, r5
871	mtspr	SPRN_PSSCR, r6
872	mtspr	SPRN_HFSCR, r7
873ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
8748:
875
876	ld	r5, VCPU_SPRG0(r4)
877	ld	r6, VCPU_SPRG1(r4)
878	ld	r7, VCPU_SPRG2(r4)
879	ld	r8, VCPU_SPRG3(r4)
880	mtspr	SPRN_SPRG0, r5
881	mtspr	SPRN_SPRG1, r6
882	mtspr	SPRN_SPRG2, r7
883	mtspr	SPRN_SPRG3, r8
884
885	/* Load up DAR and DSISR */
886	ld	r5, VCPU_DAR(r4)
887	lwz	r6, VCPU_DSISR(r4)
888	mtspr	SPRN_DAR, r5
889	mtspr	SPRN_DSISR, r6
890
891	/* Restore AMR and UAMOR, set AMOR to all 1s */
892	ld	r5,VCPU_AMR(r4)
893	ld	r6,VCPU_UAMOR(r4)
894	li	r7,-1
895	mtspr	SPRN_AMR,r5
896	mtspr	SPRN_UAMOR,r6
897	mtspr	SPRN_AMOR,r7
898
899	/* Restore state of CTRL run bit; assume 1 on entry */
900	lwz	r5,VCPU_CTRL(r4)
901	andi.	r5,r5,1
902	bne	4f
903	mfspr	r6,SPRN_CTRLF
904	clrrdi	r6,r6,1
905	mtspr	SPRN_CTRLT,r6
9064:
907	/* Secondary threads wait for primary to have done partition switch */
908	ld	r5, HSTATE_KVM_VCORE(r13)
909	lbz	r6, HSTATE_PTID(r13)
910	cmpwi	r6, 0
911	beq	21f
912	lbz	r0, VCORE_IN_GUEST(r5)
913	cmpwi	r0, 0
914	bne	21f
915	HMT_LOW
91620:	lwz	r3, VCORE_ENTRY_EXIT(r5)
917	cmpwi	r3, 0x100
918	bge	no_switch_exit
919	lbz	r0, VCORE_IN_GUEST(r5)
920	cmpwi	r0, 0
921	beq	20b
922	HMT_MEDIUM
92321:
924	/* Set LPCR. */
925	ld	r8,VCORE_LPCR(r5)
926	mtspr	SPRN_LPCR,r8
927	isync
928
929	/*
930	 * Set the decrementer to the guest decrementer.
931	 */
932	ld	r8,VCPU_DEC_EXPIRES(r4)
933	/* r8 is a host timebase value here, convert to guest TB */
934	ld	r5,HSTATE_KVM_VCORE(r13)
935	ld	r6,VCORE_TB_OFFSET_APPL(r5)
936	add	r8,r8,r6
937	mftb	r7
938	subf	r3,r7,r8
939	mtspr	SPRN_DEC,r3
940
941	/* Check if HDEC expires soon */
942	mfspr	r3, SPRN_HDEC
943	EXTEND_HDEC(r3)
944	cmpdi	r3, 512		/* 1 microsecond */
945	blt	hdec_soon
946
947	/* For hash guest, clear out and reload the SLB */
948	ld	r6, VCPU_KVM(r4)
949	lbz	r0, KVM_RADIX(r6)
950	cmpwi	r0, 0
951	bne	9f
952	li	r6, 0
953	slbmte	r6, r6
954	slbia
955	ptesync
956
957	/* Load up guest SLB entries (N.B. slb_max will be 0 for radix) */
958	lwz	r5,VCPU_SLB_MAX(r4)
959	cmpwi	r5,0
960	beq	9f
961	mtctr	r5
962	addi	r6,r4,VCPU_SLB
9631:	ld	r8,VCPU_SLB_E(r6)
964	ld	r9,VCPU_SLB_V(r6)
965	slbmte	r9,r8
966	addi	r6,r6,VCPU_SLB_SIZE
967	bdnz	1b
9689:
969
970#ifdef CONFIG_KVM_XICS
971	/* We are entering the guest on that thread, push VCPU to XIVE */
972	ld	r10, HSTATE_XIVE_TIMA_PHYS(r13)
973	cmpldi	cr0, r10, 0
974	beq	no_xive
975	ld	r11, VCPU_XIVE_SAVED_STATE(r4)
976	li	r9, TM_QW1_OS
977	eieio
978	stdcix	r11,r9,r10
979	lwz	r11, VCPU_XIVE_CAM_WORD(r4)
980	li	r9, TM_QW1_OS + TM_WORD2
981	stwcix	r11,r9,r10
982	li	r9, 1
983	stb	r9, VCPU_XIVE_PUSHED(r4)
984	eieio
985
986	/*
987	 * We clear the irq_pending flag. There is a small chance of a
988	 * race vs. the escalation interrupt happening on another
989	 * processor setting it again, but the only consequence is to
990	 * cause a spurrious wakeup on the next H_CEDE which is not an
991	 * issue.
992	 */
993	li	r0,0
994	stb	r0, VCPU_IRQ_PENDING(r4)
995
996	/*
997	 * In single escalation mode, if the escalation interrupt is
998	 * on, we mask it.
999	 */
1000	lbz	r0, VCPU_XIVE_ESC_ON(r4)
1001	cmpwi	r0,0
1002	beq	1f
1003	ld	r10, VCPU_XIVE_ESC_RADDR(r4)
1004	li	r9, XIVE_ESB_SET_PQ_01
1005	ldcix	r0, r10, r9
1006	sync
1007
1008	/* We have a possible subtle race here: The escalation interrupt might
1009	 * have fired and be on its way to the host queue while we mask it,
1010	 * and if we unmask it early enough (re-cede right away), there is
1011	 * a theorical possibility that it fires again, thus landing in the
1012	 * target queue more than once which is a big no-no.
1013	 *
1014	 * Fortunately, solving this is rather easy. If the above load setting
1015	 * PQ to 01 returns a previous value where P is set, then we know the
1016	 * escalation interrupt is somewhere on its way to the host. In that
1017	 * case we simply don't clear the xive_esc_on flag below. It will be
1018	 * eventually cleared by the handler for the escalation interrupt.
1019	 *
1020	 * Then, when doing a cede, we check that flag again before re-enabling
1021	 * the escalation interrupt, and if set, we abort the cede.
1022	 */
1023	andi.	r0, r0, XIVE_ESB_VAL_P
1024	bne-	1f
1025
1026	/* Now P is 0, we can clear the flag */
1027	li	r0, 0
1028	stb	r0, VCPU_XIVE_ESC_ON(r4)
10291:
1030no_xive:
1031#endif /* CONFIG_KVM_XICS */
1032
1033	li	r0, 0
1034	stw	r0, STACK_SLOT_SHORT_PATH(r1)
1035
1036deliver_guest_interrupt:	/* r4 = vcpu, r13 = paca */
1037	/* Check if we can deliver an external or decrementer interrupt now */
1038	ld	r0, VCPU_PENDING_EXC(r4)
1039BEGIN_FTR_SECTION
1040	/* On POWER9, also check for emulated doorbell interrupt */
1041	lbz	r3, VCPU_DBELL_REQ(r4)
1042	or	r0, r0, r3
1043END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1044	cmpdi	r0, 0
1045	beq	71f
1046	mr	r3, r4
1047	bl	kvmppc_guest_entry_inject_int
1048	ld	r4, HSTATE_KVM_VCPU(r13)
104971:
1050	ld	r6, VCPU_SRR0(r4)
1051	ld	r7, VCPU_SRR1(r4)
1052	mtspr	SPRN_SRR0, r6
1053	mtspr	SPRN_SRR1, r7
1054
1055fast_guest_entry_c:
1056	ld	r10, VCPU_PC(r4)
1057	ld	r11, VCPU_MSR(r4)
1058	/* r11 = vcpu->arch.msr & ~MSR_HV */
1059	rldicl	r11, r11, 63 - MSR_HV_LG, 1
1060	rotldi	r11, r11, 1 + MSR_HV_LG
1061	ori	r11, r11, MSR_ME
1062
1063	ld	r6, VCPU_CTR(r4)
1064	ld	r7, VCPU_XER(r4)
1065	mtctr	r6
1066	mtxer	r7
1067
1068/*
1069 * Required state:
1070 * R4 = vcpu
1071 * R10: value for HSRR0
1072 * R11: value for HSRR1
1073 * R13 = PACA
1074 */
1075fast_guest_return:
1076	li	r0,0
1077	stb	r0,VCPU_CEDED(r4)	/* cancel cede */
1078	mtspr	SPRN_HSRR0,r10
1079	mtspr	SPRN_HSRR1,r11
1080
1081	/* Activate guest mode, so faults get handled by KVM */
1082	li	r9, KVM_GUEST_MODE_GUEST_HV
1083	stb	r9, HSTATE_IN_GUEST(r13)
1084
1085#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1086	/* Accumulate timing */
1087	addi	r3, r4, VCPU_TB_GUEST
1088	bl	kvmhv_accumulate_time
1089#endif
1090
1091	/* Enter guest */
1092
1093BEGIN_FTR_SECTION
1094	ld	r5, VCPU_CFAR(r4)
1095	mtspr	SPRN_CFAR, r5
1096END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
1097BEGIN_FTR_SECTION
1098	ld	r0, VCPU_PPR(r4)
1099END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1100
1101	ld	r5, VCPU_LR(r4)
1102	ld	r6, VCPU_CR(r4)
1103	mtlr	r5
1104	mtcr	r6
1105
1106	ld	r1, VCPU_GPR(R1)(r4)
1107	ld	r2, VCPU_GPR(R2)(r4)
1108	ld	r3, VCPU_GPR(R3)(r4)
1109	ld	r5, VCPU_GPR(R5)(r4)
1110	ld	r6, VCPU_GPR(R6)(r4)
1111	ld	r7, VCPU_GPR(R7)(r4)
1112	ld	r8, VCPU_GPR(R8)(r4)
1113	ld	r9, VCPU_GPR(R9)(r4)
1114	ld	r10, VCPU_GPR(R10)(r4)
1115	ld	r11, VCPU_GPR(R11)(r4)
1116	ld	r12, VCPU_GPR(R12)(r4)
1117	ld	r13, VCPU_GPR(R13)(r4)
1118
1119BEGIN_FTR_SECTION
1120	mtspr	SPRN_PPR, r0
1121END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1122
1123/* Move canary into DSISR to check for later */
1124BEGIN_FTR_SECTION
1125	li	r0, 0x7fff
1126	mtspr	SPRN_HDSISR, r0
1127END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1128
1129	ld	r0, VCPU_GPR(R0)(r4)
1130	ld	r4, VCPU_GPR(R4)(r4)
1131	HRFI_TO_GUEST
1132	b	.
1133
1134/*
1135 * Enter the guest on a P9 or later system where we have exactly
1136 * one vcpu per vcore and we don't need to go to real mode
1137 * (which implies that host and guest are both using radix MMU mode).
1138 * r3 = vcpu pointer
1139 * Most SPRs and all the VSRs have been loaded already.
1140 */
1141_GLOBAL(__kvmhv_vcpu_entry_p9)
1142EXPORT_SYMBOL_GPL(__kvmhv_vcpu_entry_p9)
1143	mflr	r0
1144	std	r0, PPC_LR_STKOFF(r1)
1145	stdu	r1, -SFS(r1)
1146
1147	li	r0, 1
1148	stw	r0, STACK_SLOT_SHORT_PATH(r1)
1149
1150	std	r3, HSTATE_KVM_VCPU(r13)
1151	mfcr	r4
1152	stw	r4, SFS+8(r1)
1153
1154	std	r1, HSTATE_HOST_R1(r13)
1155
1156	reg = 14
1157	.rept	18
1158	std	reg, STACK_SLOT_NVGPRS + ((reg - 14) * 8)(r1)
1159	reg = reg + 1
1160	.endr
1161
1162	reg = 14
1163	.rept	18
1164	ld	reg, __VCPU_GPR(reg)(r3)
1165	reg = reg + 1
1166	.endr
1167
1168	mfmsr	r10
1169	std	r10, HSTATE_HOST_MSR(r13)
1170
1171	mr	r4, r3
1172	b	fast_guest_entry_c
1173guest_exit_short_path:
1174
1175	li	r0, KVM_GUEST_MODE_NONE
1176	stb	r0, HSTATE_IN_GUEST(r13)
1177
1178	reg = 14
1179	.rept	18
1180	std	reg, __VCPU_GPR(reg)(r9)
1181	reg = reg + 1
1182	.endr
1183
1184	reg = 14
1185	.rept	18
1186	ld	reg, STACK_SLOT_NVGPRS + ((reg - 14) * 8)(r1)
1187	reg = reg + 1
1188	.endr
1189
1190	lwz	r4, SFS+8(r1)
1191	mtcr	r4
1192
1193	mr	r3, r12		/* trap number */
1194
1195	addi	r1, r1, SFS
1196	ld	r0, PPC_LR_STKOFF(r1)
1197	mtlr	r0
1198
1199	/* If we are in real mode, do a rfid to get back to the caller */
1200	mfmsr	r4
1201	andi.	r5, r4, MSR_IR
1202	bnelr
1203	rldicl	r5, r4, 64 - MSR_TS_S_LG, 62	/* extract TS field */
1204	mtspr	SPRN_SRR0, r0
1205	ld	r10, HSTATE_HOST_MSR(r13)
1206	rldimi	r10, r5, MSR_TS_S_LG, 63 - MSR_TS_T_LG
1207	mtspr	SPRN_SRR1, r10
1208	RFI_TO_KERNEL
1209	b	.
1210
1211secondary_too_late:
1212	li	r12, 0
1213	stw	r12, STACK_SLOT_TRAP(r1)
1214	cmpdi	r4, 0
1215	beq	11f
1216	stw	r12, VCPU_TRAP(r4)
1217#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1218	addi	r3, r4, VCPU_TB_RMEXIT
1219	bl	kvmhv_accumulate_time
1220#endif
122111:	b	kvmhv_switch_to_host
1222
1223no_switch_exit:
1224	HMT_MEDIUM
1225	li	r12, 0
1226	b	12f
1227hdec_soon:
1228	li	r12, BOOK3S_INTERRUPT_HV_DECREMENTER
122912:	stw	r12, VCPU_TRAP(r4)
1230	mr	r9, r4
1231#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1232	addi	r3, r4, VCPU_TB_RMEXIT
1233	bl	kvmhv_accumulate_time
1234#endif
1235	b	guest_bypass
1236
1237/******************************************************************************
1238 *                                                                            *
1239 *                               Exit code                                    *
1240 *                                                                            *
1241 *****************************************************************************/
1242
1243/*
1244 * We come here from the first-level interrupt handlers.
1245 */
1246	.globl	kvmppc_interrupt_hv
1247kvmppc_interrupt_hv:
1248	/*
1249	 * Register contents:
1250	 * R12		= (guest CR << 32) | interrupt vector
1251	 * R13		= PACA
1252	 * guest R12 saved in shadow VCPU SCRATCH0
1253	 * guest CTR saved in shadow VCPU SCRATCH1 if RELOCATABLE
1254	 * guest R13 saved in SPRN_SCRATCH0
1255	 */
1256	std	r9, HSTATE_SCRATCH2(r13)
1257	lbz	r9, HSTATE_IN_GUEST(r13)
1258	cmpwi	r9, KVM_GUEST_MODE_HOST_HV
1259	beq	kvmppc_bad_host_intr
1260#ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1261	cmpwi	r9, KVM_GUEST_MODE_GUEST
1262	ld	r9, HSTATE_SCRATCH2(r13)
1263	beq	kvmppc_interrupt_pr
1264#endif
1265	/* We're now back in the host but in guest MMU context */
1266	li	r9, KVM_GUEST_MODE_HOST_HV
1267	stb	r9, HSTATE_IN_GUEST(r13)
1268
1269	ld	r9, HSTATE_KVM_VCPU(r13)
1270
1271	/* Save registers */
1272
1273	std	r0, VCPU_GPR(R0)(r9)
1274	std	r1, VCPU_GPR(R1)(r9)
1275	std	r2, VCPU_GPR(R2)(r9)
1276	std	r3, VCPU_GPR(R3)(r9)
1277	std	r4, VCPU_GPR(R4)(r9)
1278	std	r5, VCPU_GPR(R5)(r9)
1279	std	r6, VCPU_GPR(R6)(r9)
1280	std	r7, VCPU_GPR(R7)(r9)
1281	std	r8, VCPU_GPR(R8)(r9)
1282	ld	r0, HSTATE_SCRATCH2(r13)
1283	std	r0, VCPU_GPR(R9)(r9)
1284	std	r10, VCPU_GPR(R10)(r9)
1285	std	r11, VCPU_GPR(R11)(r9)
1286	ld	r3, HSTATE_SCRATCH0(r13)
1287	std	r3, VCPU_GPR(R12)(r9)
1288	/* CR is in the high half of r12 */
1289	srdi	r4, r12, 32
1290	std	r4, VCPU_CR(r9)
1291BEGIN_FTR_SECTION
1292	ld	r3, HSTATE_CFAR(r13)
1293	std	r3, VCPU_CFAR(r9)
1294END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
1295BEGIN_FTR_SECTION
1296	ld	r4, HSTATE_PPR(r13)
1297	std	r4, VCPU_PPR(r9)
1298END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1299
1300	/* Restore R1/R2 so we can handle faults */
1301	ld	r1, HSTATE_HOST_R1(r13)
1302	ld	r2, PACATOC(r13)
1303
1304	mfspr	r10, SPRN_SRR0
1305	mfspr	r11, SPRN_SRR1
1306	std	r10, VCPU_SRR0(r9)
1307	std	r11, VCPU_SRR1(r9)
1308	/* trap is in the low half of r12, clear CR from the high half */
1309	clrldi	r12, r12, 32
1310	andi.	r0, r12, 2		/* need to read HSRR0/1? */
1311	beq	1f
1312	mfspr	r10, SPRN_HSRR0
1313	mfspr	r11, SPRN_HSRR1
1314	clrrdi	r12, r12, 2
13151:	std	r10, VCPU_PC(r9)
1316	std	r11, VCPU_MSR(r9)
1317
1318	GET_SCRATCH0(r3)
1319	mflr	r4
1320	std	r3, VCPU_GPR(R13)(r9)
1321	std	r4, VCPU_LR(r9)
1322
1323	stw	r12,VCPU_TRAP(r9)
1324
1325	/*
1326	 * Now that we have saved away SRR0/1 and HSRR0/1,
1327	 * interrupts are recoverable in principle, so set MSR_RI.
1328	 * This becomes important for relocation-on interrupts from
1329	 * the guest, which we can get in radix mode on POWER9.
1330	 */
1331	li	r0, MSR_RI
1332	mtmsrd	r0, 1
1333
1334#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1335	addi	r3, r9, VCPU_TB_RMINTR
1336	mr	r4, r9
1337	bl	kvmhv_accumulate_time
1338	ld	r5, VCPU_GPR(R5)(r9)
1339	ld	r6, VCPU_GPR(R6)(r9)
1340	ld	r7, VCPU_GPR(R7)(r9)
1341	ld	r8, VCPU_GPR(R8)(r9)
1342#endif
1343
1344	/* Save HEIR (HV emulation assist reg) in emul_inst
1345	   if this is an HEI (HV emulation interrupt, e40) */
1346	li	r3,KVM_INST_FETCH_FAILED
1347	stw	r3,VCPU_LAST_INST(r9)
1348	cmpwi	r12,BOOK3S_INTERRUPT_H_EMUL_ASSIST
1349	bne	11f
1350	mfspr	r3,SPRN_HEIR
135111:	stw	r3,VCPU_HEIR(r9)
1352
1353	/* these are volatile across C function calls */
1354#ifdef CONFIG_RELOCATABLE
1355	ld	r3, HSTATE_SCRATCH1(r13)
1356	mtctr	r3
1357#else
1358	mfctr	r3
1359#endif
1360	mfxer	r4
1361	std	r3, VCPU_CTR(r9)
1362	std	r4, VCPU_XER(r9)
1363
1364	/* Save more register state  */
1365	mfdar	r3
1366	mfdsisr	r4
1367	std	r3, VCPU_DAR(r9)
1368	stw	r4, VCPU_DSISR(r9)
1369
1370	/* If this is a page table miss then see if it's theirs or ours */
1371	cmpwi	r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
1372	beq	kvmppc_hdsi
1373	std	r3, VCPU_FAULT_DAR(r9)
1374	stw	r4, VCPU_FAULT_DSISR(r9)
1375	cmpwi	r12, BOOK3S_INTERRUPT_H_INST_STORAGE
1376	beq	kvmppc_hisi
1377
1378#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1379	/* For softpatch interrupt, go off and do TM instruction emulation */
1380	cmpwi	r12, BOOK3S_INTERRUPT_HV_SOFTPATCH
1381	beq	kvmppc_tm_emul
1382#endif
1383
1384	/* See if this is a leftover HDEC interrupt */
1385	cmpwi	r12,BOOK3S_INTERRUPT_HV_DECREMENTER
1386	bne	2f
1387	mfspr	r3,SPRN_HDEC
1388	EXTEND_HDEC(r3)
1389	cmpdi	r3,0
1390	mr	r4,r9
1391	bge	fast_guest_return
13922:
1393	/* See if this is an hcall we can handle in real mode */
1394	cmpwi	r12,BOOK3S_INTERRUPT_SYSCALL
1395	beq	hcall_try_real_mode
1396
1397	/* Hypervisor doorbell - exit only if host IPI flag set */
1398	cmpwi	r12, BOOK3S_INTERRUPT_H_DOORBELL
1399	bne	3f
1400BEGIN_FTR_SECTION
1401	PPC_MSGSYNC
1402	lwsync
1403	/* always exit if we're running a nested guest */
1404	ld	r0, VCPU_NESTED(r9)
1405	cmpdi	r0, 0
1406	bne	guest_exit_cont
1407END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1408	lbz	r0, HSTATE_HOST_IPI(r13)
1409	cmpwi	r0, 0
1410	beq	maybe_reenter_guest
1411	b	guest_exit_cont
14123:
1413	/* If it's a hypervisor facility unavailable interrupt, save HFSCR */
1414	cmpwi	r12, BOOK3S_INTERRUPT_H_FAC_UNAVAIL
1415	bne	14f
1416	mfspr	r3, SPRN_HFSCR
1417	std	r3, VCPU_HFSCR(r9)
1418	b	guest_exit_cont
141914:
1420	/* External interrupt ? */
1421	cmpwi	r12, BOOK3S_INTERRUPT_EXTERNAL
1422	beq	kvmppc_guest_external
1423	/* See if it is a machine check */
1424	cmpwi	r12, BOOK3S_INTERRUPT_MACHINE_CHECK
1425	beq	machine_check_realmode
1426	/* Or a hypervisor maintenance interrupt */
1427	cmpwi	r12, BOOK3S_INTERRUPT_HMI
1428	beq	hmi_realmode
1429
1430guest_exit_cont:		/* r9 = vcpu, r12 = trap, r13 = paca */
1431
1432#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1433	addi	r3, r9, VCPU_TB_RMEXIT
1434	mr	r4, r9
1435	bl	kvmhv_accumulate_time
1436#endif
1437#ifdef CONFIG_KVM_XICS
1438	/* We are exiting, pull the VP from the XIVE */
1439	lbz	r0, VCPU_XIVE_PUSHED(r9)
1440	cmpwi	cr0, r0, 0
1441	beq	1f
1442	li	r7, TM_SPC_PULL_OS_CTX
1443	li	r6, TM_QW1_OS
1444	mfmsr	r0
1445	andi.	r0, r0, MSR_DR		/* in real mode? */
1446	beq	2f
1447	ld	r10, HSTATE_XIVE_TIMA_VIRT(r13)
1448	cmpldi	cr0, r10, 0
1449	beq	1f
1450	/* First load to pull the context, we ignore the value */
1451	eieio
1452	lwzx	r11, r7, r10
1453	/* Second load to recover the context state (Words 0 and 1) */
1454	ldx	r11, r6, r10
1455	b	3f
14562:	ld	r10, HSTATE_XIVE_TIMA_PHYS(r13)
1457	cmpldi	cr0, r10, 0
1458	beq	1f
1459	/* First load to pull the context, we ignore the value */
1460	eieio
1461	lwzcix	r11, r7, r10
1462	/* Second load to recover the context state (Words 0 and 1) */
1463	ldcix	r11, r6, r10
14643:	std	r11, VCPU_XIVE_SAVED_STATE(r9)
1465	/* Fixup some of the state for the next load */
1466	li	r10, 0
1467	li	r0, 0xff
1468	stb	r10, VCPU_XIVE_PUSHED(r9)
1469	stb	r10, (VCPU_XIVE_SAVED_STATE+3)(r9)
1470	stb	r0, (VCPU_XIVE_SAVED_STATE+4)(r9)
1471	eieio
14721:
1473#endif /* CONFIG_KVM_XICS */
1474
1475	/* If we came in through the P9 short path, go back out to C now */
1476	lwz	r0, STACK_SLOT_SHORT_PATH(r1)
1477	cmpwi	r0, 0
1478	bne	guest_exit_short_path
1479
1480	/* For hash guest, read the guest SLB and save it away */
1481	ld	r5, VCPU_KVM(r9)
1482	lbz	r0, KVM_RADIX(r5)
1483	li	r5, 0
1484	cmpwi	r0, 0
1485	bne	3f			/* for radix, save 0 entries */
1486	lwz	r0,VCPU_SLB_NR(r9)	/* number of entries in SLB */
1487	mtctr	r0
1488	li	r6,0
1489	addi	r7,r9,VCPU_SLB
14901:	slbmfee	r8,r6
1491	andis.	r0,r8,SLB_ESID_V@h
1492	beq	2f
1493	add	r8,r8,r6		/* put index in */
1494	slbmfev	r3,r6
1495	std	r8,VCPU_SLB_E(r7)
1496	std	r3,VCPU_SLB_V(r7)
1497	addi	r7,r7,VCPU_SLB_SIZE
1498	addi	r5,r5,1
14992:	addi	r6,r6,1
1500	bdnz	1b
1501	/* Finally clear out the SLB */
1502	li	r0,0
1503	slbmte	r0,r0
1504	slbia
1505	ptesync
15063:	stw	r5,VCPU_SLB_MAX(r9)
1507
1508	/* load host SLB entries */
1509BEGIN_MMU_FTR_SECTION
1510	b	0f
1511END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
1512	ld	r8,PACA_SLBSHADOWPTR(r13)
1513
1514	.rept	SLB_NUM_BOLTED
1515	li	r3, SLBSHADOW_SAVEAREA
1516	LDX_BE	r5, r8, r3
1517	addi	r3, r3, 8
1518	LDX_BE	r6, r8, r3
1519	andis.	r7,r5,SLB_ESID_V@h
1520	beq	1f
1521	slbmte	r6,r5
15221:	addi	r8,r8,16
1523	.endr
15240:
1525
1526guest_bypass:
1527	stw	r12, STACK_SLOT_TRAP(r1)
1528
1529	/* Save DEC */
1530	/* Do this before kvmhv_commence_exit so we know TB is guest TB */
1531	ld	r3, HSTATE_KVM_VCORE(r13)
1532	mfspr	r5,SPRN_DEC
1533	mftb	r6
1534	/* On P9, if the guest has large decr enabled, don't sign extend */
1535BEGIN_FTR_SECTION
1536	ld	r4, VCORE_LPCR(r3)
1537	andis.	r4, r4, LPCR_LD@h
1538	bne	16f
1539END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1540	extsw	r5,r5
154116:	add	r5,r5,r6
1542	/* r5 is a guest timebase value here, convert to host TB */
1543	ld	r4,VCORE_TB_OFFSET_APPL(r3)
1544	subf	r5,r4,r5
1545	std	r5,VCPU_DEC_EXPIRES(r9)
1546
1547	/* Increment exit count, poke other threads to exit */
1548	mr 	r3, r12
1549	bl	kvmhv_commence_exit
1550	nop
1551	ld	r9, HSTATE_KVM_VCPU(r13)
1552
1553	/* Stop others sending VCPU interrupts to this physical CPU */
1554	li	r0, -1
1555	stw	r0, VCPU_CPU(r9)
1556	stw	r0, VCPU_THREAD_CPU(r9)
1557
1558	/* Save guest CTRL register, set runlatch to 1 */
1559	mfspr	r6,SPRN_CTRLF
1560	stw	r6,VCPU_CTRL(r9)
1561	andi.	r0,r6,1
1562	bne	4f
1563	ori	r6,r6,1
1564	mtspr	SPRN_CTRLT,r6
15654:
1566	/*
1567	 * Save the guest PURR/SPURR
1568	 */
1569	mfspr	r5,SPRN_PURR
1570	mfspr	r6,SPRN_SPURR
1571	ld	r7,VCPU_PURR(r9)
1572	ld	r8,VCPU_SPURR(r9)
1573	std	r5,VCPU_PURR(r9)
1574	std	r6,VCPU_SPURR(r9)
1575	subf	r5,r7,r5
1576	subf	r6,r8,r6
1577
1578	/*
1579	 * Restore host PURR/SPURR and add guest times
1580	 * so that the time in the guest gets accounted.
1581	 */
1582	ld	r3,HSTATE_PURR(r13)
1583	ld	r4,HSTATE_SPURR(r13)
1584	add	r3,r3,r5
1585	add	r4,r4,r6
1586	mtspr	SPRN_PURR,r3
1587	mtspr	SPRN_SPURR,r4
1588
1589BEGIN_FTR_SECTION
1590	b	8f
1591END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
1592	/* Save POWER8-specific registers */
1593	mfspr	r5, SPRN_IAMR
1594	mfspr	r6, SPRN_PSPB
1595	mfspr	r7, SPRN_FSCR
1596	std	r5, VCPU_IAMR(r9)
1597	stw	r6, VCPU_PSPB(r9)
1598	std	r7, VCPU_FSCR(r9)
1599	mfspr	r5, SPRN_IC
1600	mfspr	r7, SPRN_TAR
1601	std	r5, VCPU_IC(r9)
1602	std	r7, VCPU_TAR(r9)
1603	mfspr	r8, SPRN_EBBHR
1604	std	r8, VCPU_EBBHR(r9)
1605	mfspr	r5, SPRN_EBBRR
1606	mfspr	r6, SPRN_BESCR
1607	mfspr	r7, SPRN_PID
1608	mfspr	r8, SPRN_WORT
1609	std	r5, VCPU_EBBRR(r9)
1610	std	r6, VCPU_BESCR(r9)
1611	stw	r7, VCPU_GUEST_PID(r9)
1612	std	r8, VCPU_WORT(r9)
1613BEGIN_FTR_SECTION
1614	mfspr	r5, SPRN_TCSCR
1615	mfspr	r6, SPRN_ACOP
1616	mfspr	r7, SPRN_CSIGR
1617	mfspr	r8, SPRN_TACR
1618	std	r5, VCPU_TCSCR(r9)
1619	std	r6, VCPU_ACOP(r9)
1620	std	r7, VCPU_CSIGR(r9)
1621	std	r8, VCPU_TACR(r9)
1622FTR_SECTION_ELSE
1623	mfspr	r5, SPRN_TIDR
1624	mfspr	r6, SPRN_PSSCR
1625	std	r5, VCPU_TID(r9)
1626	rldicl	r6, r6, 4, 50		/* r6 &= PSSCR_GUEST_VIS */
1627	rotldi	r6, r6, 60
1628	std	r6, VCPU_PSSCR(r9)
1629	/* Restore host HFSCR value */
1630	ld	r7, STACK_SLOT_HFSCR(r1)
1631	mtspr	SPRN_HFSCR, r7
1632ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
1633	/*
1634	 * Restore various registers to 0, where non-zero values
1635	 * set by the guest could disrupt the host.
1636	 */
1637	li	r0, 0
1638	mtspr	SPRN_PSPB, r0
1639	mtspr	SPRN_WORT, r0
1640BEGIN_FTR_SECTION
1641	mtspr	SPRN_TCSCR, r0
1642	/* Set MMCRS to 1<<31 to freeze and disable the SPMC counters */
1643	li	r0, 1
1644	sldi	r0, r0, 31
1645	mtspr	SPRN_MMCRS, r0
1646END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1647
1648	/* Save and restore AMR, IAMR and UAMOR before turning on the MMU */
1649	ld	r8, STACK_SLOT_IAMR(r1)
1650	mtspr	SPRN_IAMR, r8
1651
16528:	/* Power7 jumps back in here */
1653	mfspr	r5,SPRN_AMR
1654	mfspr	r6,SPRN_UAMOR
1655	std	r5,VCPU_AMR(r9)
1656	std	r6,VCPU_UAMOR(r9)
1657	ld	r5,STACK_SLOT_AMR(r1)
1658	ld	r6,STACK_SLOT_UAMOR(r1)
1659	mtspr	SPRN_AMR, r5
1660	mtspr	SPRN_UAMOR, r6
1661
1662	/* Switch DSCR back to host value */
1663	mfspr	r8, SPRN_DSCR
1664	ld	r7, HSTATE_DSCR(r13)
1665	std	r8, VCPU_DSCR(r9)
1666	mtspr	SPRN_DSCR, r7
1667
1668	/* Save non-volatile GPRs */
1669	std	r14, VCPU_GPR(R14)(r9)
1670	std	r15, VCPU_GPR(R15)(r9)
1671	std	r16, VCPU_GPR(R16)(r9)
1672	std	r17, VCPU_GPR(R17)(r9)
1673	std	r18, VCPU_GPR(R18)(r9)
1674	std	r19, VCPU_GPR(R19)(r9)
1675	std	r20, VCPU_GPR(R20)(r9)
1676	std	r21, VCPU_GPR(R21)(r9)
1677	std	r22, VCPU_GPR(R22)(r9)
1678	std	r23, VCPU_GPR(R23)(r9)
1679	std	r24, VCPU_GPR(R24)(r9)
1680	std	r25, VCPU_GPR(R25)(r9)
1681	std	r26, VCPU_GPR(R26)(r9)
1682	std	r27, VCPU_GPR(R27)(r9)
1683	std	r28, VCPU_GPR(R28)(r9)
1684	std	r29, VCPU_GPR(R29)(r9)
1685	std	r30, VCPU_GPR(R30)(r9)
1686	std	r31, VCPU_GPR(R31)(r9)
1687
1688	/* Save SPRGs */
1689	mfspr	r3, SPRN_SPRG0
1690	mfspr	r4, SPRN_SPRG1
1691	mfspr	r5, SPRN_SPRG2
1692	mfspr	r6, SPRN_SPRG3
1693	std	r3, VCPU_SPRG0(r9)
1694	std	r4, VCPU_SPRG1(r9)
1695	std	r5, VCPU_SPRG2(r9)
1696	std	r6, VCPU_SPRG3(r9)
1697
1698	/* save FP state */
1699	mr	r3, r9
1700	bl	kvmppc_save_fp
1701
1702#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1703/*
1704 * Branch around the call if both CPU_FTR_TM and
1705 * CPU_FTR_P9_TM_HV_ASSIST are off.
1706 */
1707BEGIN_FTR_SECTION
1708	b	91f
1709END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
1710	/*
1711	 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
1712	 */
1713	mr      r3, r9
1714	ld      r4, VCPU_MSR(r3)
1715	li	r5, 0			/* don't preserve non-vol regs */
1716	bl	kvmppc_save_tm_hv
1717	nop
1718	ld	r9, HSTATE_KVM_VCPU(r13)
171991:
1720#endif
1721
1722	/* Increment yield count if they have a VPA */
1723	ld	r8, VCPU_VPA(r9)	/* do they have a VPA? */
1724	cmpdi	r8, 0
1725	beq	25f
1726	li	r4, LPPACA_YIELDCOUNT
1727	LWZX_BE	r3, r8, r4
1728	addi	r3, r3, 1
1729	STWX_BE	r3, r8, r4
1730	li	r3, 1
1731	stb	r3, VCPU_VPA_DIRTY(r9)
173225:
1733	/* Save PMU registers if requested */
1734	/* r8 and cr0.eq are live here */
1735	mr	r3, r9
1736	li	r4, 1
1737	beq	21f			/* if no VPA, save PMU stuff anyway */
1738	lbz	r4, LPPACA_PMCINUSE(r8)
173921:	bl	kvmhv_save_guest_pmu
1740	ld	r9, HSTATE_KVM_VCPU(r13)
1741
1742	/* Restore host values of some registers */
1743BEGIN_FTR_SECTION
1744	ld	r5, STACK_SLOT_CIABR(r1)
1745	ld	r6, STACK_SLOT_DAWR(r1)
1746	ld	r7, STACK_SLOT_DAWRX(r1)
1747	mtspr	SPRN_CIABR, r5
1748	/*
1749	 * If the DAWR doesn't work, it's ok to write these here as
1750	 * this value should always be zero
1751	*/
1752	mtspr	SPRN_DAWR, r6
1753	mtspr	SPRN_DAWRX, r7
1754END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1755BEGIN_FTR_SECTION
1756	ld	r5, STACK_SLOT_TID(r1)
1757	ld	r6, STACK_SLOT_PSSCR(r1)
1758	ld	r7, STACK_SLOT_PID(r1)
1759	mtspr	SPRN_TIDR, r5
1760	mtspr	SPRN_PSSCR, r6
1761	mtspr	SPRN_PID, r7
1762END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1763
1764#ifdef CONFIG_PPC_RADIX_MMU
1765	/*
1766	 * Are we running hash or radix ?
1767	 */
1768	ld	r5, VCPU_KVM(r9)
1769	lbz	r0, KVM_RADIX(r5)
1770	cmpwi	cr2, r0, 0
1771	beq	cr2, 2f
1772
1773	/*
1774	 * Radix: do eieio; tlbsync; ptesync sequence in case we
1775	 * interrupted the guest between a tlbie and a ptesync.
1776	 */
1777	eieio
1778	tlbsync
1779	ptesync
1780
1781	/* Radix: Handle the case where the guest used an illegal PID */
1782	LOAD_REG_ADDR(r4, mmu_base_pid)
1783	lwz	r3, VCPU_GUEST_PID(r9)
1784	lwz	r5, 0(r4)
1785	cmpw	cr0,r3,r5
1786	blt	2f
1787
1788	/*
1789	 * Illegal PID, the HW might have prefetched and cached in the TLB
1790	 * some translations for the  LPID 0 / guest PID combination which
1791	 * Linux doesn't know about, so we need to flush that PID out of
1792	 * the TLB. First we need to set LPIDR to 0 so tlbiel applies to
1793	 * the right context.
1794	*/
1795	li	r0,0
1796	mtspr	SPRN_LPID,r0
1797	isync
1798
1799	/* Then do a congruence class local flush */
1800	ld	r6,VCPU_KVM(r9)
1801	lwz	r0,KVM_TLB_SETS(r6)
1802	mtctr	r0
1803	li	r7,0x400		/* IS field = 0b01 */
1804	ptesync
1805	sldi	r0,r3,32		/* RS has PID */
18061:	PPC_TLBIEL(7,0,2,1,1)		/* RIC=2, PRS=1, R=1 */
1807	addi	r7,r7,0x1000
1808	bdnz	1b
1809	ptesync
1810
18112:
1812#endif /* CONFIG_PPC_RADIX_MMU */
1813
1814	/*
1815	 * POWER7/POWER8 guest -> host partition switch code.
1816	 * We don't have to lock against tlbies but we do
1817	 * have to coordinate the hardware threads.
1818	 * Here STACK_SLOT_TRAP(r1) contains the trap number.
1819	 */
1820kvmhv_switch_to_host:
1821	/* Secondary threads wait for primary to do partition switch */
1822	ld	r5,HSTATE_KVM_VCORE(r13)
1823	ld	r4,VCORE_KVM(r5)	/* pointer to struct kvm */
1824	lbz	r3,HSTATE_PTID(r13)
1825	cmpwi	r3,0
1826	beq	15f
1827	HMT_LOW
182813:	lbz	r3,VCORE_IN_GUEST(r5)
1829	cmpwi	r3,0
1830	bne	13b
1831	HMT_MEDIUM
1832	b	16f
1833
1834	/* Primary thread waits for all the secondaries to exit guest */
183515:	lwz	r3,VCORE_ENTRY_EXIT(r5)
1836	rlwinm	r0,r3,32-8,0xff
1837	clrldi	r3,r3,56
1838	cmpw	r3,r0
1839	bne	15b
1840	isync
1841
1842	/* Did we actually switch to the guest at all? */
1843	lbz	r6, VCORE_IN_GUEST(r5)
1844	cmpwi	r6, 0
1845	beq	19f
1846
1847	/* Primary thread switches back to host partition */
1848	lwz	r7,KVM_HOST_LPID(r4)
1849BEGIN_FTR_SECTION
1850	ld	r6,KVM_HOST_SDR1(r4)
1851	li	r8,LPID_RSVD		/* switch to reserved LPID */
1852	mtspr	SPRN_LPID,r8
1853	ptesync
1854	mtspr	SPRN_SDR1,r6		/* switch to host page table */
1855END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1856	mtspr	SPRN_LPID,r7
1857	isync
1858
1859BEGIN_FTR_SECTION
1860	/* DPDES and VTB are shared between threads */
1861	mfspr	r7, SPRN_DPDES
1862	mfspr	r8, SPRN_VTB
1863	std	r7, VCORE_DPDES(r5)
1864	std	r8, VCORE_VTB(r5)
1865	/* clear DPDES so we don't get guest doorbells in the host */
1866	li	r8, 0
1867	mtspr	SPRN_DPDES, r8
1868END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1869
1870	/* Subtract timebase offset from timebase */
1871	ld	r8, VCORE_TB_OFFSET_APPL(r5)
1872	cmpdi	r8,0
1873	beq	17f
1874	li	r0, 0
1875	std	r0, VCORE_TB_OFFSET_APPL(r5)
1876	mftb	r6			/* current guest timebase */
1877	subf	r8,r8,r6
1878	mtspr	SPRN_TBU40,r8		/* update upper 40 bits */
1879	mftb	r7			/* check if lower 24 bits overflowed */
1880	clrldi	r6,r6,40
1881	clrldi	r7,r7,40
1882	cmpld	r7,r6
1883	bge	17f
1884	addis	r8,r8,0x100		/* if so, increment upper 40 bits */
1885	mtspr	SPRN_TBU40,r8
1886
188717:
1888	/*
1889	 * If this is an HMI, we called kvmppc_realmode_hmi_handler
1890	 * above, which may or may not have already called
1891	 * kvmppc_subcore_exit_guest.  Fortunately, all that
1892	 * kvmppc_subcore_exit_guest does is clear a flag, so calling
1893	 * it again here is benign even if kvmppc_realmode_hmi_handler
1894	 * has already called it.
1895	 */
1896	bl	kvmppc_subcore_exit_guest
1897	nop
189830:	ld	r5,HSTATE_KVM_VCORE(r13)
1899	ld	r4,VCORE_KVM(r5)	/* pointer to struct kvm */
1900
1901	/* Reset PCR */
1902	ld	r0, VCORE_PCR(r5)
1903	cmpdi	r0, 0
1904	beq	18f
1905	li	r0, 0
1906	mtspr	SPRN_PCR, r0
190718:
1908	/* Signal secondary CPUs to continue */
1909	stb	r0,VCORE_IN_GUEST(r5)
191019:	lis	r8,0x7fff		/* MAX_INT@h */
1911	mtspr	SPRN_HDEC,r8
1912
191316:
1914BEGIN_FTR_SECTION
1915	/* On POWER9 with HPT-on-radix we need to wait for all other threads */
1916	ld	r3, HSTATE_SPLIT_MODE(r13)
1917	cmpdi	r3, 0
1918	beq	47f
1919	lwz	r8, KVM_SPLIT_DO_RESTORE(r3)
1920	cmpwi	r8, 0
1921	beq	47f
1922	bl	kvmhv_p9_restore_lpcr
1923	nop
1924	b	48f
192547:
1926END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1927	ld	r8,KVM_HOST_LPCR(r4)
1928	mtspr	SPRN_LPCR,r8
1929	isync
193048:
1931#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1932	/* Finish timing, if we have a vcpu */
1933	ld	r4, HSTATE_KVM_VCPU(r13)
1934	cmpdi	r4, 0
1935	li	r3, 0
1936	beq	2f
1937	bl	kvmhv_accumulate_time
19382:
1939#endif
1940	/* Unset guest mode */
1941	li	r0, KVM_GUEST_MODE_NONE
1942	stb	r0, HSTATE_IN_GUEST(r13)
1943
1944	lwz	r12, STACK_SLOT_TRAP(r1)	/* return trap # in r12 */
1945	ld	r0, SFS+PPC_LR_STKOFF(r1)
1946	addi	r1, r1, SFS
1947	mtlr	r0
1948	blr
1949
1950kvmppc_guest_external:
1951	/* External interrupt, first check for host_ipi. If this is
1952	 * set, we know the host wants us out so let's do it now
1953	 */
1954	bl	kvmppc_read_intr
1955
1956	/*
1957	 * Restore the active volatile registers after returning from
1958	 * a C function.
1959	 */
1960	ld	r9, HSTATE_KVM_VCPU(r13)
1961	li	r12, BOOK3S_INTERRUPT_EXTERNAL
1962
1963	/*
1964	 * kvmppc_read_intr return codes:
1965	 *
1966	 * Exit to host (r3 > 0)
1967	 *   1 An interrupt is pending that needs to be handled by the host
1968	 *     Exit guest and return to host by branching to guest_exit_cont
1969	 *
1970	 *   2 Passthrough that needs completion in the host
1971	 *     Exit guest and return to host by branching to guest_exit_cont
1972	 *     However, we also set r12 to BOOK3S_INTERRUPT_HV_RM_HARD
1973	 *     to indicate to the host to complete handling the interrupt
1974	 *
1975	 * Before returning to guest, we check if any CPU is heading out
1976	 * to the host and if so, we head out also. If no CPUs are heading
1977	 * check return values <= 0.
1978	 *
1979	 * Return to guest (r3 <= 0)
1980	 *  0 No external interrupt is pending
1981	 * -1 A guest wakeup IPI (which has now been cleared)
1982	 *    In either case, we return to guest to deliver any pending
1983	 *    guest interrupts.
1984	 *
1985	 * -2 A PCI passthrough external interrupt was handled
1986	 *    (interrupt was delivered directly to guest)
1987	 *    Return to guest to deliver any pending guest interrupts.
1988	 */
1989
1990	cmpdi	r3, 1
1991	ble	1f
1992
1993	/* Return code = 2 */
1994	li	r12, BOOK3S_INTERRUPT_HV_RM_HARD
1995	stw	r12, VCPU_TRAP(r9)
1996	b	guest_exit_cont
1997
19981:	/* Return code <= 1 */
1999	cmpdi	r3, 0
2000	bgt	guest_exit_cont
2001
2002	/* Return code <= 0 */
2003maybe_reenter_guest:
2004	ld	r5, HSTATE_KVM_VCORE(r13)
2005	lwz	r0, VCORE_ENTRY_EXIT(r5)
2006	cmpwi	r0, 0x100
2007	mr	r4, r9
2008	blt	deliver_guest_interrupt
2009	b	guest_exit_cont
2010
2011#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2012/*
2013 * Softpatch interrupt for transactional memory emulation cases
2014 * on POWER9 DD2.2.  This is early in the guest exit path - we
2015 * haven't saved registers or done a treclaim yet.
2016 */
2017kvmppc_tm_emul:
2018	/* Save instruction image in HEIR */
2019	mfspr	r3, SPRN_HEIR
2020	stw	r3, VCPU_HEIR(r9)
2021
2022	/*
2023	 * The cases we want to handle here are those where the guest
2024	 * is in real suspend mode and is trying to transition to
2025	 * transactional mode.
2026	 */
2027	lbz	r0, HSTATE_FAKE_SUSPEND(r13)
2028	cmpwi	r0, 0		/* keep exiting guest if in fake suspend */
2029	bne	guest_exit_cont
2030	rldicl	r3, r11, 64 - MSR_TS_S_LG, 62
2031	cmpwi	r3, 1		/* or if not in suspend state */
2032	bne	guest_exit_cont
2033
2034	/* Call C code to do the emulation */
2035	mr	r3, r9
2036	bl	kvmhv_p9_tm_emulation_early
2037	nop
2038	ld	r9, HSTATE_KVM_VCPU(r13)
2039	li	r12, BOOK3S_INTERRUPT_HV_SOFTPATCH
2040	cmpwi	r3, 0
2041	beq	guest_exit_cont		/* continue exiting if not handled */
2042	ld	r10, VCPU_PC(r9)
2043	ld	r11, VCPU_MSR(r9)
2044	b	fast_interrupt_c_return	/* go back to guest if handled */
2045#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
2046
2047/*
2048 * Check whether an HDSI is an HPTE not found fault or something else.
2049 * If it is an HPTE not found fault that is due to the guest accessing
2050 * a page that they have mapped but which we have paged out, then
2051 * we continue on with the guest exit path.  In all other cases,
2052 * reflect the HDSI to the guest as a DSI.
2053 */
2054kvmppc_hdsi:
2055	ld	r3, VCPU_KVM(r9)
2056	lbz	r0, KVM_RADIX(r3)
2057	mfspr	r4, SPRN_HDAR
2058	mfspr	r6, SPRN_HDSISR
2059BEGIN_FTR_SECTION
2060	/* Look for DSISR canary. If we find it, retry instruction */
2061	cmpdi	r6, 0x7fff
2062	beq	6f
2063END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2064	cmpwi	r0, 0
2065	bne	.Lradix_hdsi		/* on radix, just save DAR/DSISR/ASDR */
2066	/* HPTE not found fault or protection fault? */
2067	andis.	r0, r6, (DSISR_NOHPTE | DSISR_PROTFAULT)@h
2068	beq	1f			/* if not, send it to the guest */
2069	andi.	r0, r11, MSR_DR		/* data relocation enabled? */
2070	beq	3f
2071BEGIN_FTR_SECTION
2072	mfspr	r5, SPRN_ASDR		/* on POWER9, use ASDR to get VSID */
2073	b	4f
2074END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2075	clrrdi	r0, r4, 28
2076	PPC_SLBFEE_DOT(R5, R0)		/* if so, look up SLB */
2077	li	r0, BOOK3S_INTERRUPT_DATA_SEGMENT
2078	bne	7f			/* if no SLB entry found */
20794:	std	r4, VCPU_FAULT_DAR(r9)
2080	stw	r6, VCPU_FAULT_DSISR(r9)
2081
2082	/* Search the hash table. */
2083	mr	r3, r9			/* vcpu pointer */
2084	li	r7, 1			/* data fault */
2085	bl	kvmppc_hpte_hv_fault
2086	ld	r9, HSTATE_KVM_VCPU(r13)
2087	ld	r10, VCPU_PC(r9)
2088	ld	r11, VCPU_MSR(r9)
2089	li	r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
2090	cmpdi	r3, 0			/* retry the instruction */
2091	beq	6f
2092	cmpdi	r3, -1			/* handle in kernel mode */
2093	beq	guest_exit_cont
2094	cmpdi	r3, -2			/* MMIO emulation; need instr word */
2095	beq	2f
2096
2097	/* Synthesize a DSI (or DSegI) for the guest */
2098	ld	r4, VCPU_FAULT_DAR(r9)
2099	mr	r6, r3
21001:	li	r0, BOOK3S_INTERRUPT_DATA_STORAGE
2101	mtspr	SPRN_DSISR, r6
21027:	mtspr	SPRN_DAR, r4
2103	mtspr	SPRN_SRR0, r10
2104	mtspr	SPRN_SRR1, r11
2105	mr	r10, r0
2106	bl	kvmppc_msr_interrupt
2107fast_interrupt_c_return:
21086:	ld	r7, VCPU_CTR(r9)
2109	ld	r8, VCPU_XER(r9)
2110	mtctr	r7
2111	mtxer	r8
2112	mr	r4, r9
2113	b	fast_guest_return
2114
21153:	ld	r5, VCPU_KVM(r9)	/* not relocated, use VRMA */
2116	ld	r5, KVM_VRMA_SLB_V(r5)
2117	b	4b
2118
2119	/* If this is for emulated MMIO, load the instruction word */
21202:	li	r8, KVM_INST_FETCH_FAILED	/* In case lwz faults */
2121
2122	/* Set guest mode to 'jump over instruction' so if lwz faults
2123	 * we'll just continue at the next IP. */
2124	li	r0, KVM_GUEST_MODE_SKIP
2125	stb	r0, HSTATE_IN_GUEST(r13)
2126
2127	/* Do the access with MSR:DR enabled */
2128	mfmsr	r3
2129	ori	r4, r3, MSR_DR		/* Enable paging for data */
2130	mtmsrd	r4
2131	lwz	r8, 0(r10)
2132	mtmsrd	r3
2133
2134	/* Store the result */
2135	stw	r8, VCPU_LAST_INST(r9)
2136
2137	/* Unset guest mode. */
2138	li	r0, KVM_GUEST_MODE_HOST_HV
2139	stb	r0, HSTATE_IN_GUEST(r13)
2140	b	guest_exit_cont
2141
2142.Lradix_hdsi:
2143	std	r4, VCPU_FAULT_DAR(r9)
2144	stw	r6, VCPU_FAULT_DSISR(r9)
2145.Lradix_hisi:
2146	mfspr	r5, SPRN_ASDR
2147	std	r5, VCPU_FAULT_GPA(r9)
2148	b	guest_exit_cont
2149
2150/*
2151 * Similarly for an HISI, reflect it to the guest as an ISI unless
2152 * it is an HPTE not found fault for a page that we have paged out.
2153 */
2154kvmppc_hisi:
2155	ld	r3, VCPU_KVM(r9)
2156	lbz	r0, KVM_RADIX(r3)
2157	cmpwi	r0, 0
2158	bne	.Lradix_hisi		/* for radix, just save ASDR */
2159	andis.	r0, r11, SRR1_ISI_NOPT@h
2160	beq	1f
2161	andi.	r0, r11, MSR_IR		/* instruction relocation enabled? */
2162	beq	3f
2163BEGIN_FTR_SECTION
2164	mfspr	r5, SPRN_ASDR		/* on POWER9, use ASDR to get VSID */
2165	b	4f
2166END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2167	clrrdi	r0, r10, 28
2168	PPC_SLBFEE_DOT(R5, R0)		/* if so, look up SLB */
2169	li	r0, BOOK3S_INTERRUPT_INST_SEGMENT
2170	bne	7f			/* if no SLB entry found */
21714:
2172	/* Search the hash table. */
2173	mr	r3, r9			/* vcpu pointer */
2174	mr	r4, r10
2175	mr	r6, r11
2176	li	r7, 0			/* instruction fault */
2177	bl	kvmppc_hpte_hv_fault
2178	ld	r9, HSTATE_KVM_VCPU(r13)
2179	ld	r10, VCPU_PC(r9)
2180	ld	r11, VCPU_MSR(r9)
2181	li	r12, BOOK3S_INTERRUPT_H_INST_STORAGE
2182	cmpdi	r3, 0			/* retry the instruction */
2183	beq	fast_interrupt_c_return
2184	cmpdi	r3, -1			/* handle in kernel mode */
2185	beq	guest_exit_cont
2186
2187	/* Synthesize an ISI (or ISegI) for the guest */
2188	mr	r11, r3
21891:	li	r0, BOOK3S_INTERRUPT_INST_STORAGE
21907:	mtspr	SPRN_SRR0, r10
2191	mtspr	SPRN_SRR1, r11
2192	mr	r10, r0
2193	bl	kvmppc_msr_interrupt
2194	b	fast_interrupt_c_return
2195
21963:	ld	r6, VCPU_KVM(r9)	/* not relocated, use VRMA */
2197	ld	r5, KVM_VRMA_SLB_V(r6)
2198	b	4b
2199
2200/*
2201 * Try to handle an hcall in real mode.
2202 * Returns to the guest if we handle it, or continues on up to
2203 * the kernel if we can't (i.e. if we don't have a handler for
2204 * it, or if the handler returns H_TOO_HARD).
2205 *
2206 * r5 - r8 contain hcall args,
2207 * r9 = vcpu, r10 = pc, r11 = msr, r12 = trap, r13 = paca
2208 */
2209hcall_try_real_mode:
2210	ld	r3,VCPU_GPR(R3)(r9)
2211	andi.	r0,r11,MSR_PR
2212	/* sc 1 from userspace - reflect to guest syscall */
2213	bne	sc_1_fast_return
2214	/* sc 1 from nested guest - give it to L1 to handle */
2215	ld	r0, VCPU_NESTED(r9)
2216	cmpdi	r0, 0
2217	bne	guest_exit_cont
2218	clrrdi	r3,r3,2
2219	cmpldi	r3,hcall_real_table_end - hcall_real_table
2220	bge	guest_exit_cont
2221	/* See if this hcall is enabled for in-kernel handling */
2222	ld	r4, VCPU_KVM(r9)
2223	srdi	r0, r3, 8	/* r0 = (r3 / 4) >> 6 */
2224	sldi	r0, r0, 3	/* index into kvm->arch.enabled_hcalls[] */
2225	add	r4, r4, r0
2226	ld	r0, KVM_ENABLED_HCALLS(r4)
2227	rlwinm	r4, r3, 32-2, 0x3f	/* r4 = (r3 / 4) & 0x3f */
2228	srd	r0, r0, r4
2229	andi.	r0, r0, 1
2230	beq	guest_exit_cont
2231	/* Get pointer to handler, if any, and call it */
2232	LOAD_REG_ADDR(r4, hcall_real_table)
2233	lwax	r3,r3,r4
2234	cmpwi	r3,0
2235	beq	guest_exit_cont
2236	add	r12,r3,r4
2237	mtctr	r12
2238	mr	r3,r9		/* get vcpu pointer */
2239	ld	r4,VCPU_GPR(R4)(r9)
2240	bctrl
2241	cmpdi	r3,H_TOO_HARD
2242	beq	hcall_real_fallback
2243	ld	r4,HSTATE_KVM_VCPU(r13)
2244	std	r3,VCPU_GPR(R3)(r4)
2245	ld	r10,VCPU_PC(r4)
2246	ld	r11,VCPU_MSR(r4)
2247	b	fast_guest_return
2248
2249sc_1_fast_return:
2250	mtspr	SPRN_SRR0,r10
2251	mtspr	SPRN_SRR1,r11
2252	li	r10, BOOK3S_INTERRUPT_SYSCALL
2253	bl	kvmppc_msr_interrupt
2254	mr	r4,r9
2255	b	fast_guest_return
2256
2257	/* We've attempted a real mode hcall, but it's punted it back
2258	 * to userspace.  We need to restore some clobbered volatiles
2259	 * before resuming the pass-it-to-qemu path */
2260hcall_real_fallback:
2261	li	r12,BOOK3S_INTERRUPT_SYSCALL
2262	ld	r9, HSTATE_KVM_VCPU(r13)
2263
2264	b	guest_exit_cont
2265
2266	.globl	hcall_real_table
2267hcall_real_table:
2268	.long	0		/* 0 - unused */
2269	.long	DOTSYM(kvmppc_h_remove) - hcall_real_table
2270	.long	DOTSYM(kvmppc_h_enter) - hcall_real_table
2271	.long	DOTSYM(kvmppc_h_read) - hcall_real_table
2272	.long	DOTSYM(kvmppc_h_clear_mod) - hcall_real_table
2273	.long	DOTSYM(kvmppc_h_clear_ref) - hcall_real_table
2274	.long	DOTSYM(kvmppc_h_protect) - hcall_real_table
2275#ifdef CONFIG_SPAPR_TCE_IOMMU
2276	.long	DOTSYM(kvmppc_h_get_tce) - hcall_real_table
2277	.long	DOTSYM(kvmppc_rm_h_put_tce) - hcall_real_table
2278#else
2279	.long	0		/* 0x1c */
2280	.long	0		/* 0x20 */
2281#endif
2282	.long	0		/* 0x24 - H_SET_SPRG0 */
2283	.long	DOTSYM(kvmppc_h_set_dabr) - hcall_real_table
2284	.long	0		/* 0x2c */
2285	.long	0		/* 0x30 */
2286	.long	0		/* 0x34 */
2287	.long	0		/* 0x38 */
2288	.long	0		/* 0x3c */
2289	.long	0		/* 0x40 */
2290	.long	0		/* 0x44 */
2291	.long	0		/* 0x48 */
2292	.long	0		/* 0x4c */
2293	.long	0		/* 0x50 */
2294	.long	0		/* 0x54 */
2295	.long	0		/* 0x58 */
2296	.long	0		/* 0x5c */
2297	.long	0		/* 0x60 */
2298#ifdef CONFIG_KVM_XICS
2299	.long	DOTSYM(kvmppc_rm_h_eoi) - hcall_real_table
2300	.long	DOTSYM(kvmppc_rm_h_cppr) - hcall_real_table
2301	.long	DOTSYM(kvmppc_rm_h_ipi) - hcall_real_table
2302	.long	DOTSYM(kvmppc_rm_h_ipoll) - hcall_real_table
2303	.long	DOTSYM(kvmppc_rm_h_xirr) - hcall_real_table
2304#else
2305	.long	0		/* 0x64 - H_EOI */
2306	.long	0		/* 0x68 - H_CPPR */
2307	.long	0		/* 0x6c - H_IPI */
2308	.long	0		/* 0x70 - H_IPOLL */
2309	.long	0		/* 0x74 - H_XIRR */
2310#endif
2311	.long	0		/* 0x78 */
2312	.long	0		/* 0x7c */
2313	.long	0		/* 0x80 */
2314	.long	0		/* 0x84 */
2315	.long	0		/* 0x88 */
2316	.long	0		/* 0x8c */
2317	.long	0		/* 0x90 */
2318	.long	0		/* 0x94 */
2319	.long	0		/* 0x98 */
2320	.long	0		/* 0x9c */
2321	.long	0		/* 0xa0 */
2322	.long	0		/* 0xa4 */
2323	.long	0		/* 0xa8 */
2324	.long	0		/* 0xac */
2325	.long	0		/* 0xb0 */
2326	.long	0		/* 0xb4 */
2327	.long	0		/* 0xb8 */
2328	.long	0		/* 0xbc */
2329	.long	0		/* 0xc0 */
2330	.long	0		/* 0xc4 */
2331	.long	0		/* 0xc8 */
2332	.long	0		/* 0xcc */
2333	.long	0		/* 0xd0 */
2334	.long	0		/* 0xd4 */
2335	.long	0		/* 0xd8 */
2336	.long	0		/* 0xdc */
2337	.long	DOTSYM(kvmppc_h_cede) - hcall_real_table
2338	.long	DOTSYM(kvmppc_rm_h_confer) - hcall_real_table
2339	.long	0		/* 0xe8 */
2340	.long	0		/* 0xec */
2341	.long	0		/* 0xf0 */
2342	.long	0		/* 0xf4 */
2343	.long	0		/* 0xf8 */
2344	.long	0		/* 0xfc */
2345	.long	0		/* 0x100 */
2346	.long	0		/* 0x104 */
2347	.long	0		/* 0x108 */
2348	.long	0		/* 0x10c */
2349	.long	0		/* 0x110 */
2350	.long	0		/* 0x114 */
2351	.long	0		/* 0x118 */
2352	.long	0		/* 0x11c */
2353	.long	0		/* 0x120 */
2354	.long	DOTSYM(kvmppc_h_bulk_remove) - hcall_real_table
2355	.long	0		/* 0x128 */
2356	.long	0		/* 0x12c */
2357	.long	0		/* 0x130 */
2358	.long	DOTSYM(kvmppc_h_set_xdabr) - hcall_real_table
2359#ifdef CONFIG_SPAPR_TCE_IOMMU
2360	.long	DOTSYM(kvmppc_rm_h_stuff_tce) - hcall_real_table
2361	.long	DOTSYM(kvmppc_rm_h_put_tce_indirect) - hcall_real_table
2362#else
2363	.long	0		/* 0x138 */
2364	.long	0		/* 0x13c */
2365#endif
2366	.long	0		/* 0x140 */
2367	.long	0		/* 0x144 */
2368	.long	0		/* 0x148 */
2369	.long	0		/* 0x14c */
2370	.long	0		/* 0x150 */
2371	.long	0		/* 0x154 */
2372	.long	0		/* 0x158 */
2373	.long	0		/* 0x15c */
2374	.long	0		/* 0x160 */
2375	.long	0		/* 0x164 */
2376	.long	0		/* 0x168 */
2377	.long	0		/* 0x16c */
2378	.long	0		/* 0x170 */
2379	.long	0		/* 0x174 */
2380	.long	0		/* 0x178 */
2381	.long	0		/* 0x17c */
2382	.long	0		/* 0x180 */
2383	.long	0		/* 0x184 */
2384	.long	0		/* 0x188 */
2385	.long	0		/* 0x18c */
2386	.long	0		/* 0x190 */
2387	.long	0		/* 0x194 */
2388	.long	0		/* 0x198 */
2389	.long	0		/* 0x19c */
2390	.long	0		/* 0x1a0 */
2391	.long	0		/* 0x1a4 */
2392	.long	0		/* 0x1a8 */
2393	.long	0		/* 0x1ac */
2394	.long	0		/* 0x1b0 */
2395	.long	0		/* 0x1b4 */
2396	.long	0		/* 0x1b8 */
2397	.long	0		/* 0x1bc */
2398	.long	0		/* 0x1c0 */
2399	.long	0		/* 0x1c4 */
2400	.long	0		/* 0x1c8 */
2401	.long	0		/* 0x1cc */
2402	.long	0		/* 0x1d0 */
2403	.long	0		/* 0x1d4 */
2404	.long	0		/* 0x1d8 */
2405	.long	0		/* 0x1dc */
2406	.long	0		/* 0x1e0 */
2407	.long	0		/* 0x1e4 */
2408	.long	0		/* 0x1e8 */
2409	.long	0		/* 0x1ec */
2410	.long	0		/* 0x1f0 */
2411	.long	0		/* 0x1f4 */
2412	.long	0		/* 0x1f8 */
2413	.long	0		/* 0x1fc */
2414	.long	0		/* 0x200 */
2415	.long	0		/* 0x204 */
2416	.long	0		/* 0x208 */
2417	.long	0		/* 0x20c */
2418	.long	0		/* 0x210 */
2419	.long	0		/* 0x214 */
2420	.long	0		/* 0x218 */
2421	.long	0		/* 0x21c */
2422	.long	0		/* 0x220 */
2423	.long	0		/* 0x224 */
2424	.long	0		/* 0x228 */
2425	.long	0		/* 0x22c */
2426	.long	0		/* 0x230 */
2427	.long	0		/* 0x234 */
2428	.long	0		/* 0x238 */
2429	.long	0		/* 0x23c */
2430	.long	0		/* 0x240 */
2431	.long	0		/* 0x244 */
2432	.long	0		/* 0x248 */
2433	.long	0		/* 0x24c */
2434	.long	0		/* 0x250 */
2435	.long	0		/* 0x254 */
2436	.long	0		/* 0x258 */
2437	.long	0		/* 0x25c */
2438	.long	0		/* 0x260 */
2439	.long	0		/* 0x264 */
2440	.long	0		/* 0x268 */
2441	.long	0		/* 0x26c */
2442	.long	0		/* 0x270 */
2443	.long	0		/* 0x274 */
2444	.long	0		/* 0x278 */
2445	.long	0		/* 0x27c */
2446	.long	0		/* 0x280 */
2447	.long	0		/* 0x284 */
2448	.long	0		/* 0x288 */
2449	.long	0		/* 0x28c */
2450	.long	0		/* 0x290 */
2451	.long	0		/* 0x294 */
2452	.long	0		/* 0x298 */
2453	.long	0		/* 0x29c */
2454	.long	0		/* 0x2a0 */
2455	.long	0		/* 0x2a4 */
2456	.long	0		/* 0x2a8 */
2457	.long	0		/* 0x2ac */
2458	.long	0		/* 0x2b0 */
2459	.long	0		/* 0x2b4 */
2460	.long	0		/* 0x2b8 */
2461	.long	0		/* 0x2bc */
2462	.long	0		/* 0x2c0 */
2463	.long	0		/* 0x2c4 */
2464	.long	0		/* 0x2c8 */
2465	.long	0		/* 0x2cc */
2466	.long	0		/* 0x2d0 */
2467	.long	0		/* 0x2d4 */
2468	.long	0		/* 0x2d8 */
2469	.long	0		/* 0x2dc */
2470	.long	0		/* 0x2e0 */
2471	.long	0		/* 0x2e4 */
2472	.long	0		/* 0x2e8 */
2473	.long	0		/* 0x2ec */
2474	.long	0		/* 0x2f0 */
2475	.long	0		/* 0x2f4 */
2476	.long	0		/* 0x2f8 */
2477#ifdef CONFIG_KVM_XICS
2478	.long	DOTSYM(kvmppc_rm_h_xirr_x) - hcall_real_table
2479#else
2480	.long	0		/* 0x2fc - H_XIRR_X*/
2481#endif
2482	.long	DOTSYM(kvmppc_h_random) - hcall_real_table
2483	.globl	hcall_real_table_end
2484hcall_real_table_end:
2485
2486_GLOBAL(kvmppc_h_set_xdabr)
2487EXPORT_SYMBOL_GPL(kvmppc_h_set_xdabr)
2488	andi.	r0, r5, DABRX_USER | DABRX_KERNEL
2489	beq	6f
2490	li	r0, DABRX_USER | DABRX_KERNEL | DABRX_BTI
2491	andc.	r0, r5, r0
2492	beq	3f
24936:	li	r3, H_PARAMETER
2494	blr
2495
2496_GLOBAL(kvmppc_h_set_dabr)
2497EXPORT_SYMBOL_GPL(kvmppc_h_set_dabr)
2498	li	r5, DABRX_USER | DABRX_KERNEL
24993:
2500BEGIN_FTR_SECTION
2501	b	2f
2502END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2503	std	r4,VCPU_DABR(r3)
2504	stw	r5, VCPU_DABRX(r3)
2505	mtspr	SPRN_DABRX, r5
2506	/* Work around P7 bug where DABR can get corrupted on mtspr */
25071:	mtspr	SPRN_DABR,r4
2508	mfspr	r5, SPRN_DABR
2509	cmpd	r4, r5
2510	bne	1b
2511	isync
2512	li	r3,0
2513	blr
2514
25152:
2516BEGIN_FTR_SECTION
2517	/* POWER9 with disabled DAWR */
2518	li	r3, H_HARDWARE
2519	blr
2520END_FTR_SECTION_IFCLR(CPU_FTR_DAWR)
2521	/* Emulate H_SET_DABR/X on P8 for the sake of compat mode guests */
2522	rlwimi	r5, r4, 5, DAWRX_DR | DAWRX_DW
2523	rlwimi	r5, r4, 2, DAWRX_WT
2524	clrrdi	r4, r4, 3
2525	std	r4, VCPU_DAWR(r3)
2526	std	r5, VCPU_DAWRX(r3)
2527	mtspr	SPRN_DAWR, r4
2528	mtspr	SPRN_DAWRX, r5
2529	li	r3, 0
2530	blr
2531
2532_GLOBAL(kvmppc_h_cede)		/* r3 = vcpu pointer, r11 = msr, r13 = paca */
2533	ori	r11,r11,MSR_EE
2534	std	r11,VCPU_MSR(r3)
2535	li	r0,1
2536	stb	r0,VCPU_CEDED(r3)
2537	sync			/* order setting ceded vs. testing prodded */
2538	lbz	r5,VCPU_PRODDED(r3)
2539	cmpwi	r5,0
2540	bne	kvm_cede_prodded
2541	li	r12,0		/* set trap to 0 to say hcall is handled */
2542	stw	r12,VCPU_TRAP(r3)
2543	li	r0,H_SUCCESS
2544	std	r0,VCPU_GPR(R3)(r3)
2545
2546	/*
2547	 * Set our bit in the bitmask of napping threads unless all the
2548	 * other threads are already napping, in which case we send this
2549	 * up to the host.
2550	 */
2551	ld	r5,HSTATE_KVM_VCORE(r13)
2552	lbz	r6,HSTATE_PTID(r13)
2553	lwz	r8,VCORE_ENTRY_EXIT(r5)
2554	clrldi	r8,r8,56
2555	li	r0,1
2556	sld	r0,r0,r6
2557	addi	r6,r5,VCORE_NAPPING_THREADS
255831:	lwarx	r4,0,r6
2559	or	r4,r4,r0
2560	cmpw	r4,r8
2561	beq	kvm_cede_exit
2562	stwcx.	r4,0,r6
2563	bne	31b
2564	/* order napping_threads update vs testing entry_exit_map */
2565	isync
2566	li	r0,NAPPING_CEDE
2567	stb	r0,HSTATE_NAPPING(r13)
2568	lwz	r7,VCORE_ENTRY_EXIT(r5)
2569	cmpwi	r7,0x100
2570	bge	33f		/* another thread already exiting */
2571
2572/*
2573 * Although not specifically required by the architecture, POWER7
2574 * preserves the following registers in nap mode, even if an SMT mode
2575 * switch occurs: SLB entries, PURR, SPURR, AMOR, UAMOR, AMR, SPRG0-3,
2576 * DAR, DSISR, DABR, DABRX, DSCR, PMCx, MMCRx, SIAR, SDAR.
2577 */
2578	/* Save non-volatile GPRs */
2579	std	r14, VCPU_GPR(R14)(r3)
2580	std	r15, VCPU_GPR(R15)(r3)
2581	std	r16, VCPU_GPR(R16)(r3)
2582	std	r17, VCPU_GPR(R17)(r3)
2583	std	r18, VCPU_GPR(R18)(r3)
2584	std	r19, VCPU_GPR(R19)(r3)
2585	std	r20, VCPU_GPR(R20)(r3)
2586	std	r21, VCPU_GPR(R21)(r3)
2587	std	r22, VCPU_GPR(R22)(r3)
2588	std	r23, VCPU_GPR(R23)(r3)
2589	std	r24, VCPU_GPR(R24)(r3)
2590	std	r25, VCPU_GPR(R25)(r3)
2591	std	r26, VCPU_GPR(R26)(r3)
2592	std	r27, VCPU_GPR(R27)(r3)
2593	std	r28, VCPU_GPR(R28)(r3)
2594	std	r29, VCPU_GPR(R29)(r3)
2595	std	r30, VCPU_GPR(R30)(r3)
2596	std	r31, VCPU_GPR(R31)(r3)
2597
2598	/* save FP state */
2599	bl	kvmppc_save_fp
2600
2601#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2602/*
2603 * Branch around the call if both CPU_FTR_TM and
2604 * CPU_FTR_P9_TM_HV_ASSIST are off.
2605 */
2606BEGIN_FTR_SECTION
2607	b	91f
2608END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
2609	/*
2610	 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
2611	 */
2612	ld	r3, HSTATE_KVM_VCPU(r13)
2613	ld      r4, VCPU_MSR(r3)
2614	li	r5, 0			/* don't preserve non-vol regs */
2615	bl	kvmppc_save_tm_hv
2616	nop
261791:
2618#endif
2619
2620	/*
2621	 * Set DEC to the smaller of DEC and HDEC, so that we wake
2622	 * no later than the end of our timeslice (HDEC interrupts
2623	 * don't wake us from nap).
2624	 */
2625	mfspr	r3, SPRN_DEC
2626	mfspr	r4, SPRN_HDEC
2627	mftb	r5
2628BEGIN_FTR_SECTION
2629	/* On P9 check whether the guest has large decrementer mode enabled */
2630	ld	r6, HSTATE_KVM_VCORE(r13)
2631	ld	r6, VCORE_LPCR(r6)
2632	andis.	r6, r6, LPCR_LD@h
2633	bne	68f
2634END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2635	extsw	r3, r3
263668:	EXTEND_HDEC(r4)
2637	cmpd	r3, r4
2638	ble	67f
2639	mtspr	SPRN_DEC, r4
264067:
2641	/* save expiry time of guest decrementer */
2642	add	r3, r3, r5
2643	ld	r4, HSTATE_KVM_VCPU(r13)
2644	ld	r5, HSTATE_KVM_VCORE(r13)
2645	ld	r6, VCORE_TB_OFFSET_APPL(r5)
2646	subf	r3, r6, r3	/* convert to host TB value */
2647	std	r3, VCPU_DEC_EXPIRES(r4)
2648
2649#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2650	ld	r4, HSTATE_KVM_VCPU(r13)
2651	addi	r3, r4, VCPU_TB_CEDE
2652	bl	kvmhv_accumulate_time
2653#endif
2654
2655	lis	r3, LPCR_PECEDP@h	/* Do wake on privileged doorbell */
2656
2657	/*
2658	 * Take a nap until a decrementer or external or doobell interrupt
2659	 * occurs, with PECE1 and PECE0 set in LPCR.
2660	 * On POWER8, set PECEDH, and if we are ceding, also set PECEDP.
2661	 * Also clear the runlatch bit before napping.
2662	 */
2663kvm_do_nap:
2664	mfspr	r0, SPRN_CTRLF
2665	clrrdi	r0, r0, 1
2666	mtspr	SPRN_CTRLT, r0
2667
2668	li	r0,1
2669	stb	r0,HSTATE_HWTHREAD_REQ(r13)
2670	mfspr	r5,SPRN_LPCR
2671	ori	r5,r5,LPCR_PECE0 | LPCR_PECE1
2672BEGIN_FTR_SECTION
2673	ori	r5, r5, LPCR_PECEDH
2674	rlwimi	r5, r3, 0, LPCR_PECEDP
2675END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2676
2677kvm_nap_sequence:		/* desired LPCR value in r5 */
2678BEGIN_FTR_SECTION
2679	/*
2680	 * PSSCR bits:	exit criterion = 1 (wakeup based on LPCR at sreset)
2681	 *		enable state loss = 1 (allow SMT mode switch)
2682	 *		requested level = 0 (just stop dispatching)
2683	 */
2684	lis	r3, (PSSCR_EC | PSSCR_ESL)@h
2685	mtspr	SPRN_PSSCR, r3
2686	/* Set LPCR_PECE_HVEE bit to enable wakeup by HV interrupts */
2687	li	r4, LPCR_PECE_HVEE@higher
2688	sldi	r4, r4, 32
2689	or	r5, r5, r4
2690END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2691	mtspr	SPRN_LPCR,r5
2692	isync
2693	li	r0, 0
2694	std	r0, HSTATE_SCRATCH0(r13)
2695	ptesync
2696	ld	r0, HSTATE_SCRATCH0(r13)
26971:	cmpd	r0, r0
2698	bne	1b
2699BEGIN_FTR_SECTION
2700	nap
2701FTR_SECTION_ELSE
2702	PPC_STOP
2703ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
2704	b	.
2705
270633:	mr	r4, r3
2707	li	r3, 0
2708	li	r12, 0
2709	b	34f
2710
2711kvm_end_cede:
2712	/* get vcpu pointer */
2713	ld	r4, HSTATE_KVM_VCPU(r13)
2714
2715	/* Woken by external or decrementer interrupt */
2716	ld	r1, HSTATE_HOST_R1(r13)
2717
2718#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2719	addi	r3, r4, VCPU_TB_RMINTR
2720	bl	kvmhv_accumulate_time
2721#endif
2722
2723#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2724/*
2725 * Branch around the call if both CPU_FTR_TM and
2726 * CPU_FTR_P9_TM_HV_ASSIST are off.
2727 */
2728BEGIN_FTR_SECTION
2729	b	91f
2730END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
2731	/*
2732	 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
2733	 */
2734	mr      r3, r4
2735	ld      r4, VCPU_MSR(r3)
2736	li	r5, 0			/* don't preserve non-vol regs */
2737	bl	kvmppc_restore_tm_hv
2738	nop
2739	ld	r4, HSTATE_KVM_VCPU(r13)
274091:
2741#endif
2742
2743	/* load up FP state */
2744	bl	kvmppc_load_fp
2745
2746	/* Restore guest decrementer */
2747	ld	r3, VCPU_DEC_EXPIRES(r4)
2748	ld	r5, HSTATE_KVM_VCORE(r13)
2749	ld	r6, VCORE_TB_OFFSET_APPL(r5)
2750	add	r3, r3, r6	/* convert host TB to guest TB value */
2751	mftb	r7
2752	subf	r3, r7, r3
2753	mtspr	SPRN_DEC, r3
2754
2755	/* Load NV GPRS */
2756	ld	r14, VCPU_GPR(R14)(r4)
2757	ld	r15, VCPU_GPR(R15)(r4)
2758	ld	r16, VCPU_GPR(R16)(r4)
2759	ld	r17, VCPU_GPR(R17)(r4)
2760	ld	r18, VCPU_GPR(R18)(r4)
2761	ld	r19, VCPU_GPR(R19)(r4)
2762	ld	r20, VCPU_GPR(R20)(r4)
2763	ld	r21, VCPU_GPR(R21)(r4)
2764	ld	r22, VCPU_GPR(R22)(r4)
2765	ld	r23, VCPU_GPR(R23)(r4)
2766	ld	r24, VCPU_GPR(R24)(r4)
2767	ld	r25, VCPU_GPR(R25)(r4)
2768	ld	r26, VCPU_GPR(R26)(r4)
2769	ld	r27, VCPU_GPR(R27)(r4)
2770	ld	r28, VCPU_GPR(R28)(r4)
2771	ld	r29, VCPU_GPR(R29)(r4)
2772	ld	r30, VCPU_GPR(R30)(r4)
2773	ld	r31, VCPU_GPR(R31)(r4)
2774
2775	/* Check the wake reason in SRR1 to see why we got here */
2776	bl	kvmppc_check_wake_reason
2777
2778	/*
2779	 * Restore volatile registers since we could have called a
2780	 * C routine in kvmppc_check_wake_reason
2781	 *	r4 = VCPU
2782	 * r3 tells us whether we need to return to host or not
2783	 * WARNING: it gets checked further down:
2784	 * should not modify r3 until this check is done.
2785	 */
2786	ld	r4, HSTATE_KVM_VCPU(r13)
2787
2788	/* clear our bit in vcore->napping_threads */
278934:	ld	r5,HSTATE_KVM_VCORE(r13)
2790	lbz	r7,HSTATE_PTID(r13)
2791	li	r0,1
2792	sld	r0,r0,r7
2793	addi	r6,r5,VCORE_NAPPING_THREADS
279432:	lwarx	r7,0,r6
2795	andc	r7,r7,r0
2796	stwcx.	r7,0,r6
2797	bne	32b
2798	li	r0,0
2799	stb	r0,HSTATE_NAPPING(r13)
2800
2801	/* See if the wake reason saved in r3 means we need to exit */
2802	stw	r12, VCPU_TRAP(r4)
2803	mr	r9, r4
2804	cmpdi	r3, 0
2805	bgt	guest_exit_cont
2806	b	maybe_reenter_guest
2807
2808	/* cede when already previously prodded case */
2809kvm_cede_prodded:
2810	li	r0,0
2811	stb	r0,VCPU_PRODDED(r3)
2812	sync			/* order testing prodded vs. clearing ceded */
2813	stb	r0,VCPU_CEDED(r3)
2814	li	r3,H_SUCCESS
2815	blr
2816
2817	/* we've ceded but we want to give control to the host */
2818kvm_cede_exit:
2819	ld	r9, HSTATE_KVM_VCPU(r13)
2820#ifdef CONFIG_KVM_XICS
2821	/* Abort if we still have a pending escalation */
2822	lbz	r5, VCPU_XIVE_ESC_ON(r9)
2823	cmpwi	r5, 0
2824	beq	1f
2825	li	r0, 0
2826	stb	r0, VCPU_CEDED(r9)
28271:	/* Enable XIVE escalation */
2828	li	r5, XIVE_ESB_SET_PQ_00
2829	mfmsr	r0
2830	andi.	r0, r0, MSR_DR		/* in real mode? */
2831	beq	1f
2832	ld	r10, VCPU_XIVE_ESC_VADDR(r9)
2833	cmpdi	r10, 0
2834	beq	3f
2835	ldx	r0, r10, r5
2836	b	2f
28371:	ld	r10, VCPU_XIVE_ESC_RADDR(r9)
2838	cmpdi	r10, 0
2839	beq	3f
2840	ldcix	r0, r10, r5
28412:	sync
2842	li	r0, 1
2843	stb	r0, VCPU_XIVE_ESC_ON(r9)
2844#endif /* CONFIG_KVM_XICS */
28453:	b	guest_exit_cont
2846
2847	/* Try to do machine check recovery in real mode */
2848machine_check_realmode:
2849	mr	r3, r9		/* get vcpu pointer */
2850	bl	kvmppc_realmode_machine_check
2851	nop
2852	/* all machine checks go to virtual mode for further handling */
2853	ld	r9, HSTATE_KVM_VCPU(r13)
2854	li	r12, BOOK3S_INTERRUPT_MACHINE_CHECK
2855	b	guest_exit_cont
2856
2857/*
2858 * Call C code to handle a HMI in real mode.
2859 * Only the primary thread does the call, secondary threads are handled
2860 * by calling hmi_exception_realmode() after kvmppc_hv_entry returns.
2861 * r9 points to the vcpu on entry
2862 */
2863hmi_realmode:
2864	lbz	r0, HSTATE_PTID(r13)
2865	cmpwi	r0, 0
2866	bne	guest_exit_cont
2867	bl	kvmppc_realmode_hmi_handler
2868	ld	r9, HSTATE_KVM_VCPU(r13)
2869	li	r12, BOOK3S_INTERRUPT_HMI
2870	b	guest_exit_cont
2871
2872/*
2873 * Check the reason we woke from nap, and take appropriate action.
2874 * Returns (in r3):
2875 *	0 if nothing needs to be done
2876 *	1 if something happened that needs to be handled by the host
2877 *	-1 if there was a guest wakeup (IPI or msgsnd)
2878 *	-2 if we handled a PCI passthrough interrupt (returned by
2879 *		kvmppc_read_intr only)
2880 *
2881 * Also sets r12 to the interrupt vector for any interrupt that needs
2882 * to be handled now by the host (0x500 for external interrupt), or zero.
2883 * Modifies all volatile registers (since it may call a C function).
2884 * This routine calls kvmppc_read_intr, a C function, if an external
2885 * interrupt is pending.
2886 */
2887kvmppc_check_wake_reason:
2888	mfspr	r6, SPRN_SRR1
2889BEGIN_FTR_SECTION
2890	rlwinm	r6, r6, 45-31, 0xf	/* extract wake reason field (P8) */
2891FTR_SECTION_ELSE
2892	rlwinm	r6, r6, 45-31, 0xe	/* P7 wake reason field is 3 bits */
2893ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_207S)
2894	cmpwi	r6, 8			/* was it an external interrupt? */
2895	beq	7f			/* if so, see what it was */
2896	li	r3, 0
2897	li	r12, 0
2898	cmpwi	r6, 6			/* was it the decrementer? */
2899	beq	0f
2900BEGIN_FTR_SECTION
2901	cmpwi	r6, 5			/* privileged doorbell? */
2902	beq	0f
2903	cmpwi	r6, 3			/* hypervisor doorbell? */
2904	beq	3f
2905END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2906	cmpwi	r6, 0xa			/* Hypervisor maintenance ? */
2907	beq	4f
2908	li	r3, 1			/* anything else, return 1 */
29090:	blr
2910
2911	/* hypervisor doorbell */
29123:	li	r12, BOOK3S_INTERRUPT_H_DOORBELL
2913
2914	/*
2915	 * Clear the doorbell as we will invoke the handler
2916	 * explicitly in the guest exit path.
2917	 */
2918	lis	r6, (PPC_DBELL_SERVER << (63-36))@h
2919	PPC_MSGCLR(6)
2920	/* see if it's a host IPI */
2921	li	r3, 1
2922BEGIN_FTR_SECTION
2923	PPC_MSGSYNC
2924	lwsync
2925END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2926	lbz	r0, HSTATE_HOST_IPI(r13)
2927	cmpwi	r0, 0
2928	bnelr
2929	/* if not, return -1 */
2930	li	r3, -1
2931	blr
2932
2933	/* Woken up due to Hypervisor maintenance interrupt */
29344:	li	r12, BOOK3S_INTERRUPT_HMI
2935	li	r3, 1
2936	blr
2937
2938	/* external interrupt - create a stack frame so we can call C */
29397:	mflr	r0
2940	std	r0, PPC_LR_STKOFF(r1)
2941	stdu	r1, -PPC_MIN_STKFRM(r1)
2942	bl	kvmppc_read_intr
2943	nop
2944	li	r12, BOOK3S_INTERRUPT_EXTERNAL
2945	cmpdi	r3, 1
2946	ble	1f
2947
2948	/*
2949	 * Return code of 2 means PCI passthrough interrupt, but
2950	 * we need to return back to host to complete handling the
2951	 * interrupt. Trap reason is expected in r12 by guest
2952	 * exit code.
2953	 */
2954	li	r12, BOOK3S_INTERRUPT_HV_RM_HARD
29551:
2956	ld	r0, PPC_MIN_STKFRM+PPC_LR_STKOFF(r1)
2957	addi	r1, r1, PPC_MIN_STKFRM
2958	mtlr	r0
2959	blr
2960
2961/*
2962 * Save away FP, VMX and VSX registers.
2963 * r3 = vcpu pointer
2964 * N.B. r30 and r31 are volatile across this function,
2965 * thus it is not callable from C.
2966 */
2967kvmppc_save_fp:
2968	mflr	r30
2969	mr	r31,r3
2970	mfmsr	r5
2971	ori	r8,r5,MSR_FP
2972#ifdef CONFIG_ALTIVEC
2973BEGIN_FTR_SECTION
2974	oris	r8,r8,MSR_VEC@h
2975END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2976#endif
2977#ifdef CONFIG_VSX
2978BEGIN_FTR_SECTION
2979	oris	r8,r8,MSR_VSX@h
2980END_FTR_SECTION_IFSET(CPU_FTR_VSX)
2981#endif
2982	mtmsrd	r8
2983	addi	r3,r3,VCPU_FPRS
2984	bl	store_fp_state
2985#ifdef CONFIG_ALTIVEC
2986BEGIN_FTR_SECTION
2987	addi	r3,r31,VCPU_VRS
2988	bl	store_vr_state
2989END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2990#endif
2991	mfspr	r6,SPRN_VRSAVE
2992	stw	r6,VCPU_VRSAVE(r31)
2993	mtlr	r30
2994	blr
2995
2996/*
2997 * Load up FP, VMX and VSX registers
2998 * r4 = vcpu pointer
2999 * N.B. r30 and r31 are volatile across this function,
3000 * thus it is not callable from C.
3001 */
3002kvmppc_load_fp:
3003	mflr	r30
3004	mr	r31,r4
3005	mfmsr	r9
3006	ori	r8,r9,MSR_FP
3007#ifdef CONFIG_ALTIVEC
3008BEGIN_FTR_SECTION
3009	oris	r8,r8,MSR_VEC@h
3010END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3011#endif
3012#ifdef CONFIG_VSX
3013BEGIN_FTR_SECTION
3014	oris	r8,r8,MSR_VSX@h
3015END_FTR_SECTION_IFSET(CPU_FTR_VSX)
3016#endif
3017	mtmsrd	r8
3018	addi	r3,r4,VCPU_FPRS
3019	bl	load_fp_state
3020#ifdef CONFIG_ALTIVEC
3021BEGIN_FTR_SECTION
3022	addi	r3,r31,VCPU_VRS
3023	bl	load_vr_state
3024END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3025#endif
3026	lwz	r7,VCPU_VRSAVE(r31)
3027	mtspr	SPRN_VRSAVE,r7
3028	mtlr	r30
3029	mr	r4,r31
3030	blr
3031
3032#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3033/*
3034 * Save transactional state and TM-related registers.
3035 * Called with r3 pointing to the vcpu struct and r4 containing
3036 * the guest MSR value.
3037 * r5 is non-zero iff non-volatile register state needs to be maintained.
3038 * If r5 == 0, this can modify all checkpointed registers, but
3039 * restores r1 and r2 before exit.
3040 */
3041_GLOBAL_TOC(kvmppc_save_tm_hv)
3042EXPORT_SYMBOL_GPL(kvmppc_save_tm_hv)
3043	/* See if we need to handle fake suspend mode */
3044BEGIN_FTR_SECTION
3045	b	__kvmppc_save_tm
3046END_FTR_SECTION_IFCLR(CPU_FTR_P9_TM_HV_ASSIST)
3047
3048	lbz	r0, HSTATE_FAKE_SUSPEND(r13) /* Were we fake suspended? */
3049	cmpwi	r0, 0
3050	beq	__kvmppc_save_tm
3051
3052	/* The following code handles the fake_suspend = 1 case */
3053	mflr	r0
3054	std	r0, PPC_LR_STKOFF(r1)
3055	stdu	r1, -PPC_MIN_STKFRM(r1)
3056
3057	/* Turn on TM. */
3058	mfmsr	r8
3059	li	r0, 1
3060	rldimi	r8, r0, MSR_TM_LG, 63-MSR_TM_LG
3061	mtmsrd	r8
3062
3063	rldicl. r8, r8, 64 - MSR_TS_S_LG, 62 /* Did we actually hrfid? */
3064	beq	4f
3065BEGIN_FTR_SECTION
3066	bl	pnv_power9_force_smt4_catch
3067END_FTR_SECTION_IFSET(CPU_FTR_P9_TM_XER_SO_BUG)
3068	nop
3069
3070	/* We have to treclaim here because that's the only way to do S->N */
3071	li	r3, TM_CAUSE_KVM_RESCHED
3072	TRECLAIM(R3)
3073
3074	/*
3075	 * We were in fake suspend, so we are not going to save the
3076	 * register state as the guest checkpointed state (since
3077	 * we already have it), therefore we can now use any volatile GPR.
3078	 * In fact treclaim in fake suspend state doesn't modify
3079	 * any registers.
3080	 */
3081
3082BEGIN_FTR_SECTION
3083	bl	pnv_power9_force_smt4_release
3084END_FTR_SECTION_IFSET(CPU_FTR_P9_TM_XER_SO_BUG)
3085	nop
3086
30874:
3088	mfspr	r3, SPRN_PSSCR
3089	/* PSSCR_FAKE_SUSPEND is a write-only bit, but clear it anyway */
3090	li	r0, PSSCR_FAKE_SUSPEND
3091	andc	r3, r3, r0
3092	mtspr	SPRN_PSSCR, r3
3093
3094	/* Don't save TEXASR, use value from last exit in real suspend state */
3095	ld	r9, HSTATE_KVM_VCPU(r13)
3096	mfspr	r5, SPRN_TFHAR
3097	mfspr	r6, SPRN_TFIAR
3098	std	r5, VCPU_TFHAR(r9)
3099	std	r6, VCPU_TFIAR(r9)
3100
3101	addi	r1, r1, PPC_MIN_STKFRM
3102	ld	r0, PPC_LR_STKOFF(r1)
3103	mtlr	r0
3104	blr
3105
3106/*
3107 * Restore transactional state and TM-related registers.
3108 * Called with r3 pointing to the vcpu struct
3109 * and r4 containing the guest MSR value.
3110 * r5 is non-zero iff non-volatile register state needs to be maintained.
3111 * This potentially modifies all checkpointed registers.
3112 * It restores r1 and r2 from the PACA.
3113 */
3114_GLOBAL_TOC(kvmppc_restore_tm_hv)
3115EXPORT_SYMBOL_GPL(kvmppc_restore_tm_hv)
3116	/*
3117	 * If we are doing TM emulation for the guest on a POWER9 DD2,
3118	 * then we don't actually do a trechkpt -- we either set up
3119	 * fake-suspend mode, or emulate a TM rollback.
3120	 */
3121BEGIN_FTR_SECTION
3122	b	__kvmppc_restore_tm
3123END_FTR_SECTION_IFCLR(CPU_FTR_P9_TM_HV_ASSIST)
3124	mflr	r0
3125	std	r0, PPC_LR_STKOFF(r1)
3126
3127	li	r0, 0
3128	stb	r0, HSTATE_FAKE_SUSPEND(r13)
3129
3130	/* Turn on TM so we can restore TM SPRs */
3131	mfmsr	r5
3132	li	r0, 1
3133	rldimi	r5, r0, MSR_TM_LG, 63-MSR_TM_LG
3134	mtmsrd	r5
3135
3136	/*
3137	 * The user may change these outside of a transaction, so they must
3138	 * always be context switched.
3139	 */
3140	ld	r5, VCPU_TFHAR(r3)
3141	ld	r6, VCPU_TFIAR(r3)
3142	ld	r7, VCPU_TEXASR(r3)
3143	mtspr	SPRN_TFHAR, r5
3144	mtspr	SPRN_TFIAR, r6
3145	mtspr	SPRN_TEXASR, r7
3146
3147	rldicl. r5, r4, 64 - MSR_TS_S_LG, 62
3148	beqlr		/* TM not active in guest */
3149
3150	/* Make sure the failure summary is set */
3151	oris	r7, r7, (TEXASR_FS)@h
3152	mtspr	SPRN_TEXASR, r7
3153
3154	cmpwi	r5, 1		/* check for suspended state */
3155	bgt	10f
3156	stb	r5, HSTATE_FAKE_SUSPEND(r13)
3157	b	9f		/* and return */
315810:	stdu	r1, -PPC_MIN_STKFRM(r1)
3159	/* guest is in transactional state, so simulate rollback */
3160	bl	kvmhv_emulate_tm_rollback
3161	nop
3162	addi	r1, r1, PPC_MIN_STKFRM
31639:	ld	r0, PPC_LR_STKOFF(r1)
3164	mtlr	r0
3165	blr
3166#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
3167
3168/*
3169 * We come here if we get any exception or interrupt while we are
3170 * executing host real mode code while in guest MMU context.
3171 * r12 is (CR << 32) | vector
3172 * r13 points to our PACA
3173 * r12 is saved in HSTATE_SCRATCH0(r13)
3174 * ctr is saved in HSTATE_SCRATCH1(r13) if RELOCATABLE
3175 * r9 is saved in HSTATE_SCRATCH2(r13)
3176 * r13 is saved in HSPRG1
3177 * cfar is saved in HSTATE_CFAR(r13)
3178 * ppr is saved in HSTATE_PPR(r13)
3179 */
3180kvmppc_bad_host_intr:
3181	/*
3182	 * Switch to the emergency stack, but start half-way down in
3183	 * case we were already on it.
3184	 */
3185	mr	r9, r1
3186	std	r1, PACAR1(r13)
3187	ld	r1, PACAEMERGSP(r13)
3188	subi	r1, r1, THREAD_SIZE/2 + INT_FRAME_SIZE
3189	std	r9, 0(r1)
3190	std	r0, GPR0(r1)
3191	std	r9, GPR1(r1)
3192	std	r2, GPR2(r1)
3193	SAVE_4GPRS(3, r1)
3194	SAVE_2GPRS(7, r1)
3195	srdi	r0, r12, 32
3196	clrldi	r12, r12, 32
3197	std	r0, _CCR(r1)
3198	std	r12, _TRAP(r1)
3199	andi.	r0, r12, 2
3200	beq	1f
3201	mfspr	r3, SPRN_HSRR0
3202	mfspr	r4, SPRN_HSRR1
3203	mfspr	r5, SPRN_HDAR
3204	mfspr	r6, SPRN_HDSISR
3205	b	2f
32061:	mfspr	r3, SPRN_SRR0
3207	mfspr	r4, SPRN_SRR1
3208	mfspr	r5, SPRN_DAR
3209	mfspr	r6, SPRN_DSISR
32102:	std	r3, _NIP(r1)
3211	std	r4, _MSR(r1)
3212	std	r5, _DAR(r1)
3213	std	r6, _DSISR(r1)
3214	ld	r9, HSTATE_SCRATCH2(r13)
3215	ld	r12, HSTATE_SCRATCH0(r13)
3216	GET_SCRATCH0(r0)
3217	SAVE_4GPRS(9, r1)
3218	std	r0, GPR13(r1)
3219	SAVE_NVGPRS(r1)
3220	ld	r5, HSTATE_CFAR(r13)
3221	std	r5, ORIG_GPR3(r1)
3222	mflr	r3
3223#ifdef CONFIG_RELOCATABLE
3224	ld	r4, HSTATE_SCRATCH1(r13)
3225#else
3226	mfctr	r4
3227#endif
3228	mfxer	r5
3229	lbz	r6, PACAIRQSOFTMASK(r13)
3230	std	r3, _LINK(r1)
3231	std	r4, _CTR(r1)
3232	std	r5, _XER(r1)
3233	std	r6, SOFTE(r1)
3234	ld	r2, PACATOC(r13)
3235	LOAD_REG_IMMEDIATE(3, 0x7265677368657265)
3236	std	r3, STACK_FRAME_OVERHEAD-16(r1)
3237
3238	/*
3239	 * On POWER9 do a minimal restore of the MMU and call C code,
3240	 * which will print a message and panic.
3241	 * XXX On POWER7 and POWER8, we just spin here since we don't
3242	 * know what the other threads are doing (and we don't want to
3243	 * coordinate with them) - but at least we now have register state
3244	 * in memory that we might be able to look at from another CPU.
3245	 */
3246BEGIN_FTR_SECTION
3247	b	.
3248END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
3249	ld	r9, HSTATE_KVM_VCPU(r13)
3250	ld	r10, VCPU_KVM(r9)
3251
3252	li	r0, 0
3253	mtspr	SPRN_AMR, r0
3254	mtspr	SPRN_IAMR, r0
3255	mtspr	SPRN_CIABR, r0
3256	mtspr	SPRN_DAWRX, r0
3257
3258BEGIN_MMU_FTR_SECTION
3259	b	4f
3260END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
3261
3262	slbmte	r0, r0
3263	slbia
3264	ptesync
3265	ld	r8, PACA_SLBSHADOWPTR(r13)
3266	.rept	SLB_NUM_BOLTED
3267	li	r3, SLBSHADOW_SAVEAREA
3268	LDX_BE	r5, r8, r3
3269	addi	r3, r3, 8
3270	LDX_BE	r6, r8, r3
3271	andis.	r7, r5, SLB_ESID_V@h
3272	beq	3f
3273	slbmte	r6, r5
32743:	addi	r8, r8, 16
3275	.endr
3276
32774:	lwz	r7, KVM_HOST_LPID(r10)
3278	mtspr	SPRN_LPID, r7
3279	mtspr	SPRN_PID, r0
3280	ld	r8, KVM_HOST_LPCR(r10)
3281	mtspr	SPRN_LPCR, r8
3282	isync
3283	li	r0, KVM_GUEST_MODE_NONE
3284	stb	r0, HSTATE_IN_GUEST(r13)
3285
3286	/*
3287	 * Turn on the MMU and jump to C code
3288	 */
3289	bcl	20, 31, .+4
32905:	mflr	r3
3291	addi	r3, r3, 9f - 5b
3292	li	r4, -1
3293	rldimi	r3, r4, 62, 0	/* ensure 0xc000000000000000 bits are set */
3294	ld	r4, PACAKMSR(r13)
3295	mtspr	SPRN_SRR0, r3
3296	mtspr	SPRN_SRR1, r4
3297	RFI_TO_KERNEL
32989:	addi	r3, r1, STACK_FRAME_OVERHEAD
3299	bl	kvmppc_bad_interrupt
3300	b	9b
3301
3302/*
3303 * This mimics the MSR transition on IRQ delivery.  The new guest MSR is taken
3304 * from VCPU_INTR_MSR and is modified based on the required TM state changes.
3305 *   r11 has the guest MSR value (in/out)
3306 *   r9 has a vcpu pointer (in)
3307 *   r0 is used as a scratch register
3308 */
3309kvmppc_msr_interrupt:
3310	rldicl	r0, r11, 64 - MSR_TS_S_LG, 62
3311	cmpwi	r0, 2 /* Check if we are in transactional state..  */
3312	ld	r11, VCPU_INTR_MSR(r9)
3313	bne	1f
3314	/* ... if transactional, change to suspended */
3315	li	r0, 1
33161:	rldimi	r11, r0, MSR_TS_S_LG, 63 - MSR_TS_T_LG
3317	blr
3318
3319/*
3320 * Load up guest PMU state.  R3 points to the vcpu struct.
3321 */
3322_GLOBAL(kvmhv_load_guest_pmu)
3323EXPORT_SYMBOL_GPL(kvmhv_load_guest_pmu)
3324	mr	r4, r3
3325	mflr	r0
3326	li	r3, 1
3327	sldi	r3, r3, 31		/* MMCR0_FC (freeze counters) bit */
3328	mtspr	SPRN_MMCR0, r3		/* freeze all counters, disable ints */
3329	isync
3330BEGIN_FTR_SECTION
3331	ld	r3, VCPU_MMCR(r4)
3332	andi.	r5, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
3333	cmpwi	r5, MMCR0_PMAO
3334	beql	kvmppc_fix_pmao
3335END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
3336	lwz	r3, VCPU_PMC(r4)	/* always load up guest PMU registers */
3337	lwz	r5, VCPU_PMC + 4(r4)	/* to prevent information leak */
3338	lwz	r6, VCPU_PMC + 8(r4)
3339	lwz	r7, VCPU_PMC + 12(r4)
3340	lwz	r8, VCPU_PMC + 16(r4)
3341	lwz	r9, VCPU_PMC + 20(r4)
3342	mtspr	SPRN_PMC1, r3
3343	mtspr	SPRN_PMC2, r5
3344	mtspr	SPRN_PMC3, r6
3345	mtspr	SPRN_PMC4, r7
3346	mtspr	SPRN_PMC5, r8
3347	mtspr	SPRN_PMC6, r9
3348	ld	r3, VCPU_MMCR(r4)
3349	ld	r5, VCPU_MMCR + 8(r4)
3350	ld	r6, VCPU_MMCR + 16(r4)
3351	ld	r7, VCPU_SIAR(r4)
3352	ld	r8, VCPU_SDAR(r4)
3353	mtspr	SPRN_MMCR1, r5
3354	mtspr	SPRN_MMCRA, r6
3355	mtspr	SPRN_SIAR, r7
3356	mtspr	SPRN_SDAR, r8
3357BEGIN_FTR_SECTION
3358	ld	r5, VCPU_MMCR + 24(r4)
3359	ld	r6, VCPU_SIER(r4)
3360	mtspr	SPRN_MMCR2, r5
3361	mtspr	SPRN_SIER, r6
3362BEGIN_FTR_SECTION_NESTED(96)
3363	lwz	r7, VCPU_PMC + 24(r4)
3364	lwz	r8, VCPU_PMC + 28(r4)
3365	ld	r9, VCPU_MMCR + 32(r4)
3366	mtspr	SPRN_SPMC1, r7
3367	mtspr	SPRN_SPMC2, r8
3368	mtspr	SPRN_MMCRS, r9
3369END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
3370END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3371	mtspr	SPRN_MMCR0, r3
3372	isync
3373	mtlr	r0
3374	blr
3375
3376/*
3377 * Reload host PMU state saved in the PACA by kvmhv_save_host_pmu.
3378 */
3379_GLOBAL(kvmhv_load_host_pmu)
3380EXPORT_SYMBOL_GPL(kvmhv_load_host_pmu)
3381	mflr	r0
3382	lbz	r4, PACA_PMCINUSE(r13) /* is the host using the PMU? */
3383	cmpwi	r4, 0
3384	beq	23f			/* skip if not */
3385BEGIN_FTR_SECTION
3386	ld	r3, HSTATE_MMCR0(r13)
3387	andi.	r4, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
3388	cmpwi	r4, MMCR0_PMAO
3389	beql	kvmppc_fix_pmao
3390END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
3391	lwz	r3, HSTATE_PMC1(r13)
3392	lwz	r4, HSTATE_PMC2(r13)
3393	lwz	r5, HSTATE_PMC3(r13)
3394	lwz	r6, HSTATE_PMC4(r13)
3395	lwz	r8, HSTATE_PMC5(r13)
3396	lwz	r9, HSTATE_PMC6(r13)
3397	mtspr	SPRN_PMC1, r3
3398	mtspr	SPRN_PMC2, r4
3399	mtspr	SPRN_PMC3, r5
3400	mtspr	SPRN_PMC4, r6
3401	mtspr	SPRN_PMC5, r8
3402	mtspr	SPRN_PMC6, r9
3403	ld	r3, HSTATE_MMCR0(r13)
3404	ld	r4, HSTATE_MMCR1(r13)
3405	ld	r5, HSTATE_MMCRA(r13)
3406	ld	r6, HSTATE_SIAR(r13)
3407	ld	r7, HSTATE_SDAR(r13)
3408	mtspr	SPRN_MMCR1, r4
3409	mtspr	SPRN_MMCRA, r5
3410	mtspr	SPRN_SIAR, r6
3411	mtspr	SPRN_SDAR, r7
3412BEGIN_FTR_SECTION
3413	ld	r8, HSTATE_MMCR2(r13)
3414	ld	r9, HSTATE_SIER(r13)
3415	mtspr	SPRN_MMCR2, r8
3416	mtspr	SPRN_SIER, r9
3417END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3418	mtspr	SPRN_MMCR0, r3
3419	isync
3420	mtlr	r0
342123:	blr
3422
3423/*
3424 * Save guest PMU state into the vcpu struct.
3425 * r3 = vcpu, r4 = full save flag (PMU in use flag set in VPA)
3426 */
3427_GLOBAL(kvmhv_save_guest_pmu)
3428EXPORT_SYMBOL_GPL(kvmhv_save_guest_pmu)
3429	mr	r9, r3
3430	mr	r8, r4
3431BEGIN_FTR_SECTION
3432	/*
3433	 * POWER8 seems to have a hardware bug where setting
3434	 * MMCR0[PMAE] along with MMCR0[PMC1CE] and/or MMCR0[PMCjCE]
3435	 * when some counters are already negative doesn't seem
3436	 * to cause a performance monitor alert (and hence interrupt).
3437	 * The effect of this is that when saving the PMU state,
3438	 * if there is no PMU alert pending when we read MMCR0
3439	 * before freezing the counters, but one becomes pending
3440	 * before we read the counters, we lose it.
3441	 * To work around this, we need a way to freeze the counters
3442	 * before reading MMCR0.  Normally, freezing the counters
3443	 * is done by writing MMCR0 (to set MMCR0[FC]) which
3444	 * unavoidably writes MMCR0[PMA0] as well.  On POWER8,
3445	 * we can also freeze the counters using MMCR2, by writing
3446	 * 1s to all the counter freeze condition bits (there are
3447	 * 9 bits each for 6 counters).
3448	 */
3449	li	r3, -1			/* set all freeze bits */
3450	clrrdi	r3, r3, 10
3451	mfspr	r10, SPRN_MMCR2
3452	mtspr	SPRN_MMCR2, r3
3453	isync
3454END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3455	li	r3, 1
3456	sldi	r3, r3, 31		/* MMCR0_FC (freeze counters) bit */
3457	mfspr	r4, SPRN_MMCR0		/* save MMCR0 */
3458	mtspr	SPRN_MMCR0, r3		/* freeze all counters, disable ints */
3459	mfspr	r6, SPRN_MMCRA
3460	/* Clear MMCRA in order to disable SDAR updates */
3461	li	r7, 0
3462	mtspr	SPRN_MMCRA, r7
3463	isync
3464	cmpwi	r8, 0			/* did they ask for PMU stuff to be saved? */
3465	bne	21f
3466	std	r3, VCPU_MMCR(r9)	/* if not, set saved MMCR0 to FC */
3467	b	22f
346821:	mfspr	r5, SPRN_MMCR1
3469	mfspr	r7, SPRN_SIAR
3470	mfspr	r8, SPRN_SDAR
3471	std	r4, VCPU_MMCR(r9)
3472	std	r5, VCPU_MMCR + 8(r9)
3473	std	r6, VCPU_MMCR + 16(r9)
3474BEGIN_FTR_SECTION
3475	std	r10, VCPU_MMCR + 24(r9)
3476END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3477	std	r7, VCPU_SIAR(r9)
3478	std	r8, VCPU_SDAR(r9)
3479	mfspr	r3, SPRN_PMC1
3480	mfspr	r4, SPRN_PMC2
3481	mfspr	r5, SPRN_PMC3
3482	mfspr	r6, SPRN_PMC4
3483	mfspr	r7, SPRN_PMC5
3484	mfspr	r8, SPRN_PMC6
3485	stw	r3, VCPU_PMC(r9)
3486	stw	r4, VCPU_PMC + 4(r9)
3487	stw	r5, VCPU_PMC + 8(r9)
3488	stw	r6, VCPU_PMC + 12(r9)
3489	stw	r7, VCPU_PMC + 16(r9)
3490	stw	r8, VCPU_PMC + 20(r9)
3491BEGIN_FTR_SECTION
3492	mfspr	r5, SPRN_SIER
3493	std	r5, VCPU_SIER(r9)
3494BEGIN_FTR_SECTION_NESTED(96)
3495	mfspr	r6, SPRN_SPMC1
3496	mfspr	r7, SPRN_SPMC2
3497	mfspr	r8, SPRN_MMCRS
3498	stw	r6, VCPU_PMC + 24(r9)
3499	stw	r7, VCPU_PMC + 28(r9)
3500	std	r8, VCPU_MMCR + 32(r9)
3501	lis	r4, 0x8000
3502	mtspr	SPRN_MMCRS, r4
3503END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
3504END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
350522:	blr
3506
3507/*
3508 * This works around a hardware bug on POWER8E processors, where
3509 * writing a 1 to the MMCR0[PMAO] bit doesn't generate a
3510 * performance monitor interrupt.  Instead, when we need to have
3511 * an interrupt pending, we have to arrange for a counter to overflow.
3512 */
3513kvmppc_fix_pmao:
3514	li	r3, 0
3515	mtspr	SPRN_MMCR2, r3
3516	lis	r3, (MMCR0_PMXE | MMCR0_FCECE)@h
3517	ori	r3, r3, MMCR0_PMCjCE | MMCR0_C56RUN
3518	mtspr	SPRN_MMCR0, r3
3519	lis	r3, 0x7fff
3520	ori	r3, r3, 0xffff
3521	mtspr	SPRN_PMC6, r3
3522	isync
3523	blr
3524
3525#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
3526/*
3527 * Start timing an activity
3528 * r3 = pointer to time accumulation struct, r4 = vcpu
3529 */
3530kvmhv_start_timing:
3531	ld	r5, HSTATE_KVM_VCORE(r13)
3532	ld	r6, VCORE_TB_OFFSET_APPL(r5)
3533	mftb	r5
3534	subf	r5, r6, r5	/* subtract current timebase offset */
3535	std	r3, VCPU_CUR_ACTIVITY(r4)
3536	std	r5, VCPU_ACTIVITY_START(r4)
3537	blr
3538
3539/*
3540 * Accumulate time to one activity and start another.
3541 * r3 = pointer to new time accumulation struct, r4 = vcpu
3542 */
3543kvmhv_accumulate_time:
3544	ld	r5, HSTATE_KVM_VCORE(r13)
3545	ld	r8, VCORE_TB_OFFSET_APPL(r5)
3546	ld	r5, VCPU_CUR_ACTIVITY(r4)
3547	ld	r6, VCPU_ACTIVITY_START(r4)
3548	std	r3, VCPU_CUR_ACTIVITY(r4)
3549	mftb	r7
3550	subf	r7, r8, r7	/* subtract current timebase offset */
3551	std	r7, VCPU_ACTIVITY_START(r4)
3552	cmpdi	r5, 0
3553	beqlr
3554	subf	r3, r6, r7
3555	ld	r8, TAS_SEQCOUNT(r5)
3556	cmpdi	r8, 0
3557	addi	r8, r8, 1
3558	std	r8, TAS_SEQCOUNT(r5)
3559	lwsync
3560	ld	r7, TAS_TOTAL(r5)
3561	add	r7, r7, r3
3562	std	r7, TAS_TOTAL(r5)
3563	ld	r6, TAS_MIN(r5)
3564	ld	r7, TAS_MAX(r5)
3565	beq	3f
3566	cmpd	r3, r6
3567	bge	1f
35683:	std	r3, TAS_MIN(r5)
35691:	cmpd	r3, r7
3570	ble	2f
3571	std	r3, TAS_MAX(r5)
35722:	lwsync
3573	addi	r8, r8, 1
3574	std	r8, TAS_SEQCOUNT(r5)
3575	blr
3576#endif
3577