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#include <asm/cpuidle.h>
39
40/* Sign-extend HDEC if not on POWER9 */
41#define EXTEND_HDEC(reg)			\
42BEGIN_FTR_SECTION;				\
43	extsw	reg, reg;			\
44END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
45
46/* Values in HSTATE_NAPPING(r13) */
47#define NAPPING_CEDE	1
48#define NAPPING_NOVCPU	2
49#define NAPPING_UNSPLIT	3
50
51/* Stack frame offsets for kvmppc_hv_entry */
52#define SFS			208
53#define STACK_SLOT_TRAP		(SFS-4)
54#define STACK_SLOT_SHORT_PATH	(SFS-8)
55#define STACK_SLOT_TID		(SFS-16)
56#define STACK_SLOT_PSSCR	(SFS-24)
57#define STACK_SLOT_PID		(SFS-32)
58#define STACK_SLOT_IAMR		(SFS-40)
59#define STACK_SLOT_CIABR	(SFS-48)
60#define STACK_SLOT_DAWR		(SFS-56)
61#define STACK_SLOT_DAWRX	(SFS-64)
62#define STACK_SLOT_HFSCR	(SFS-72)
63#define STACK_SLOT_AMR		(SFS-80)
64#define STACK_SLOT_UAMOR	(SFS-88)
65/* the following is used by the P9 short path */
66#define STACK_SLOT_NVGPRS	(SFS-152)	/* 18 gprs */
67
68/*
69 * Call kvmppc_hv_entry in real mode.
70 * Must be called with interrupts hard-disabled.
71 *
72 * Input Registers:
73 *
74 * LR = return address to continue at after eventually re-enabling MMU
75 */
76_GLOBAL_TOC(kvmppc_hv_entry_trampoline)
77	mflr	r0
78	std	r0, PPC_LR_STKOFF(r1)
79	stdu	r1, -112(r1)
80	mfmsr	r10
81	std	r10, HSTATE_HOST_MSR(r13)
82	LOAD_REG_ADDR(r5, kvmppc_call_hv_entry)
83	li	r0,MSR_RI
84	andc	r0,r10,r0
85	li	r6,MSR_IR | MSR_DR
86	andc	r6,r10,r6
87	mtmsrd	r0,1		/* clear RI in MSR */
88	mtsrr0	r5
89	mtsrr1	r6
90	RFI_TO_KERNEL
91
92kvmppc_call_hv_entry:
93BEGIN_FTR_SECTION
94	/* On P9, do LPCR setting, if necessary */
95	ld	r3, HSTATE_SPLIT_MODE(r13)
96	cmpdi	r3, 0
97	beq	46f
98	lwz	r4, KVM_SPLIT_DO_SET(r3)
99	cmpwi	r4, 0
100	beq	46f
101	bl	kvmhv_p9_set_lpcr
102	nop
10346:
104END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
105
106	ld	r4, HSTATE_KVM_VCPU(r13)
107	bl	kvmppc_hv_entry
108
109	/* Back from guest - restore host state and return to caller */
110
111BEGIN_FTR_SECTION
112	/* Restore host DABR and DABRX */
113	ld	r5,HSTATE_DABR(r13)
114	li	r6,7
115	mtspr	SPRN_DABR,r5
116	mtspr	SPRN_DABRX,r6
117END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
118
119	/* Restore SPRG3 */
120	ld	r3,PACA_SPRG_VDSO(r13)
121	mtspr	SPRN_SPRG_VDSO_WRITE,r3
122
123	/* Reload the host's PMU registers */
124	bl	kvmhv_load_host_pmu
125
126	/*
127	 * Reload DEC.  HDEC interrupts were disabled when
128	 * we reloaded the host's LPCR value.
129	 */
130	ld	r3, HSTATE_DECEXP(r13)
131	mftb	r4
132	subf	r4, r4, r3
133	mtspr	SPRN_DEC, r4
134
135	/* hwthread_req may have got set by cede or no vcpu, so clear it */
136	li	r0, 0
137	stb	r0, HSTATE_HWTHREAD_REQ(r13)
138
139	/*
140	 * For external interrupts we need to call the Linux
141	 * handler to process the interrupt. We do that by jumping
142	 * to absolute address 0x500 for external interrupts.
143	 * The [h]rfid at the end of the handler will return to
144	 * the book3s_hv_interrupts.S code. For other interrupts
145	 * we do the rfid to get back to the book3s_hv_interrupts.S
146	 * code here.
147	 */
148	ld	r8, 112+PPC_LR_STKOFF(r1)
149	addi	r1, r1, 112
150	ld	r7, HSTATE_HOST_MSR(r13)
151
152	/* Return the trap number on this thread as the return value */
153	mr	r3, r12
154
155	/*
156	 * If we came back from the guest via a relocation-on interrupt,
157	 * we will be in virtual mode at this point, which makes it a
158	 * little easier to get back to the caller.
159	 */
160	mfmsr	r0
161	andi.	r0, r0, MSR_IR		/* in real mode? */
162	bne	.Lvirt_return
163
164	/* RFI into the highmem handler */
165	mfmsr	r6
166	li	r0, MSR_RI
167	andc	r6, r6, r0
168	mtmsrd	r6, 1			/* Clear RI in MSR */
169	mtsrr0	r8
170	mtsrr1	r7
171	RFI_TO_KERNEL
172
173	/* Virtual-mode return */
174.Lvirt_return:
175	mtlr	r8
176	blr
177
178kvmppc_primary_no_guest:
179	/* We handle this much like a ceded vcpu */
180	/* put the HDEC into the DEC, since HDEC interrupts don't wake us */
181	/* HDEC may be larger than DEC for arch >= v3.00, but since the */
182	/* HDEC value came from DEC in the first place, it will fit */
183	mfspr	r3, SPRN_HDEC
184	mtspr	SPRN_DEC, r3
185	/*
186	 * Make sure the primary has finished the MMU switch.
187	 * We should never get here on a secondary thread, but
188	 * check it for robustness' sake.
189	 */
190	ld	r5, HSTATE_KVM_VCORE(r13)
19165:	lbz	r0, VCORE_IN_GUEST(r5)
192	cmpwi	r0, 0
193	beq	65b
194	/* Set LPCR. */
195	ld	r8,VCORE_LPCR(r5)
196	mtspr	SPRN_LPCR,r8
197	isync
198	/* set our bit in napping_threads */
199	ld	r5, HSTATE_KVM_VCORE(r13)
200	lbz	r7, HSTATE_PTID(r13)
201	li	r0, 1
202	sld	r0, r0, r7
203	addi	r6, r5, VCORE_NAPPING_THREADS
2041:	lwarx	r3, 0, r6
205	or	r3, r3, r0
206	stwcx.	r3, 0, r6
207	bne	1b
208	/* order napping_threads update vs testing entry_exit_map */
209	isync
210	li	r12, 0
211	lwz	r7, VCORE_ENTRY_EXIT(r5)
212	cmpwi	r7, 0x100
213	bge	kvm_novcpu_exit	/* another thread already exiting */
214	li	r3, NAPPING_NOVCPU
215	stb	r3, HSTATE_NAPPING(r13)
216
217	li	r3, 0		/* Don't wake on privileged (OS) doorbell */
218	b	kvm_do_nap
219
220/*
221 * kvm_novcpu_wakeup
222 *	Entered from kvm_start_guest if kvm_hstate.napping is set
223 *	to NAPPING_NOVCPU
224 *		r2 = kernel TOC
225 *		r13 = paca
226 */
227kvm_novcpu_wakeup:
228	ld	r1, HSTATE_HOST_R1(r13)
229	ld	r5, HSTATE_KVM_VCORE(r13)
230	li	r0, 0
231	stb	r0, HSTATE_NAPPING(r13)
232
233	/* check the wake reason */
234	bl	kvmppc_check_wake_reason
235
236	/*
237	 * Restore volatile registers since we could have called
238	 * a C routine in kvmppc_check_wake_reason.
239	 *	r5 = VCORE
240	 */
241	ld	r5, HSTATE_KVM_VCORE(r13)
242
243	/* see if any other thread is already exiting */
244	lwz	r0, VCORE_ENTRY_EXIT(r5)
245	cmpwi	r0, 0x100
246	bge	kvm_novcpu_exit
247
248	/* clear our bit in napping_threads */
249	lbz	r7, HSTATE_PTID(r13)
250	li	r0, 1
251	sld	r0, r0, r7
252	addi	r6, r5, VCORE_NAPPING_THREADS
2534:	lwarx	r7, 0, r6
254	andc	r7, r7, r0
255	stwcx.	r7, 0, r6
256	bne	4b
257
258	/* See if the wake reason means we need to exit */
259	cmpdi	r3, 0
260	bge	kvm_novcpu_exit
261
262	/* See if our timeslice has expired (HDEC is negative) */
263	mfspr	r0, SPRN_HDEC
264	EXTEND_HDEC(r0)
265	li	r12, BOOK3S_INTERRUPT_HV_DECREMENTER
266	cmpdi	r0, 0
267	blt	kvm_novcpu_exit
268
269	/* Got an IPI but other vcpus aren't yet exiting, must be a latecomer */
270	ld	r4, HSTATE_KVM_VCPU(r13)
271	cmpdi	r4, 0
272	beq	kvmppc_primary_no_guest
273
274#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
275	addi	r3, r4, VCPU_TB_RMENTRY
276	bl	kvmhv_start_timing
277#endif
278	b	kvmppc_got_guest
279
280kvm_novcpu_exit:
281#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
282	ld	r4, HSTATE_KVM_VCPU(r13)
283	cmpdi	r4, 0
284	beq	13f
285	addi	r3, r4, VCPU_TB_RMEXIT
286	bl	kvmhv_accumulate_time
287#endif
28813:	mr	r3, r12
289	stw	r12, STACK_SLOT_TRAP(r1)
290	bl	kvmhv_commence_exit
291	nop
292	b	kvmhv_switch_to_host
293
294/*
295 * We come in here when wakened from Linux offline idle code.
296 * Relocation is off
297 * r3 contains the SRR1 wakeup value, SRR1 is trashed.
298 */
299_GLOBAL(idle_kvm_start_guest)
300	ld	r4,PACAEMERGSP(r13)
301	mfcr	r5
302	mflr	r0
303	std	r1,0(r4)
304	std	r5,8(r4)
305	std	r0,16(r4)
306	subi	r1,r4,STACK_FRAME_OVERHEAD
307	SAVE_NVGPRS(r1)
308
309	/*
310	 * Could avoid this and pass it through in r3. For now,
311	 * code expects it to be in SRR1.
312	 */
313	mtspr	SPRN_SRR1,r3
314
315	li	r0,0
316	stb	r0,PACA_FTRACE_ENABLED(r13)
317
318	li	r0,KVM_HWTHREAD_IN_KVM
319	stb	r0,HSTATE_HWTHREAD_STATE(r13)
320
321	/* kvm cede / napping does not come through here */
322	lbz	r0,HSTATE_NAPPING(r13)
323	twnei	r0,0
324
325	b	1f
326
327kvm_unsplit_wakeup:
328	li	r0, 0
329	stb	r0, HSTATE_NAPPING(r13)
330
3311:
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	/*
442	 * Jump to idle_return_gpr_loss, which returns to the
443	 * idle_kvm_start_guest caller.
444	 */
445	li	r3, LPCR_PECE0
446	mfspr	r4, SPRN_LPCR
447	rlwimi	r4, r3, 0, LPCR_PECE0 | LPCR_PECE1
448	mtspr	SPRN_LPCR, r4
449	/* set up r3 for return */
450	mfspr	r3,SPRN_SRR1
451	REST_NVGPRS(r1)
452	addi	r1, r1, STACK_FRAME_OVERHEAD
453	ld	r0, 16(r1)
454	ld	r5, 8(r1)
455	ld	r1, 0(r1)
456	mtlr	r0
457	mtcr	r5
458	blr
459
46053:	HMT_LOW
461	ld	r5, HSTATE_KVM_VCORE(r13)
462	cmpdi	r5, 0
463	bne	60f
464	ld	r3, HSTATE_SPLIT_MODE(r13)
465	cmpdi	r3, 0
466	beq	kvm_no_guest
467	lwz	r0, KVM_SPLIT_DO_SET(r3)
468	cmpwi	r0, 0
469	bne	kvmhv_do_set
470	lwz	r0, KVM_SPLIT_DO_RESTORE(r3)
471	cmpwi	r0, 0
472	bne	kvmhv_do_restore
473	lbz	r0, KVM_SPLIT_DO_NAP(r3)
474	cmpwi	r0, 0
475	beq	kvm_no_guest
476	HMT_MEDIUM
477	b	kvm_unsplit_nap
47860:	HMT_MEDIUM
479	b	kvm_secondary_got_guest
480
48154:	li	r0, KVM_HWTHREAD_IN_KVM
482	stb	r0, HSTATE_HWTHREAD_STATE(r13)
483	b	kvm_no_guest
484
485kvmhv_do_set:
486	/* Set LPCR, LPIDR etc. on P9 */
487	HMT_MEDIUM
488	bl	kvmhv_p9_set_lpcr
489	nop
490	b	kvm_no_guest
491
492kvmhv_do_restore:
493	HMT_MEDIUM
494	bl	kvmhv_p9_restore_lpcr
495	nop
496	b	kvm_no_guest
497
498/*
499 * Here the primary thread is trying to return the core to
500 * whole-core mode, so we need to nap.
501 */
502kvm_unsplit_nap:
503	/*
504	 * When secondaries are napping in kvm_unsplit_nap() with
505	 * hwthread_req = 1, HMI goes ignored even though subcores are
506	 * already exited the guest. Hence HMI keeps waking up secondaries
507	 * from nap in a loop and secondaries always go back to nap since
508	 * no vcore is assigned to them. This makes impossible for primary
509	 * thread to get hold of secondary threads resulting into a soft
510	 * lockup in KVM path.
511	 *
512	 * Let us check if HMI is pending and handle it before we go to nap.
513	 */
514	cmpwi	r12, BOOK3S_INTERRUPT_HMI
515	bne	55f
516	li	r3, 0			/* NULL argument */
517	bl	hmi_exception_realmode
51855:
519	/*
520	 * Ensure that secondary doesn't nap when it has
521	 * its vcore pointer set.
522	 */
523	sync		/* matches smp_mb() before setting split_info.do_nap */
524	ld	r0, HSTATE_KVM_VCORE(r13)
525	cmpdi	r0, 0
526	bne	kvm_no_guest
527	/* clear any pending message */
528BEGIN_FTR_SECTION
529	lis	r6, (PPC_DBELL_SERVER << (63-36))@h
530	PPC_MSGCLR(6)
531END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
532	/* Set kvm_split_mode.napped[tid] = 1 */
533	ld	r3, HSTATE_SPLIT_MODE(r13)
534	li	r0, 1
535	lbz	r4, HSTATE_TID(r13)
536	addi	r4, r4, KVM_SPLIT_NAPPED
537	stbx	r0, r3, r4
538	/* Check the do_nap flag again after setting napped[] */
539	sync
540	lbz	r0, KVM_SPLIT_DO_NAP(r3)
541	cmpwi	r0, 0
542	beq	57f
543	li	r3, NAPPING_UNSPLIT
544	stb	r3, HSTATE_NAPPING(r13)
545	li	r3, (LPCR_PECEDH | LPCR_PECE0) >> 4
546	mfspr	r5, SPRN_LPCR
547	rlwimi	r5, r3, 4, (LPCR_PECEDP | LPCR_PECEDH | LPCR_PECE0 | LPCR_PECE1)
548	b	kvm_nap_sequence
549
55057:	li	r0, 0
551	stbx	r0, r3, r4
552	b	kvm_no_guest
553
554/******************************************************************************
555 *                                                                            *
556 *                               Entry code                                   *
557 *                                                                            *
558 *****************************************************************************/
559
560.global kvmppc_hv_entry
561kvmppc_hv_entry:
562
563	/* Required state:
564	 *
565	 * R4 = vcpu pointer (or NULL)
566	 * MSR = ~IR|DR
567	 * R13 = PACA
568	 * R1 = host R1
569	 * R2 = TOC
570	 * all other volatile GPRS = free
571	 * Does not preserve non-volatile GPRs or CR fields
572	 */
573	mflr	r0
574	std	r0, PPC_LR_STKOFF(r1)
575	stdu	r1, -SFS(r1)
576
577	/* Save R1 in the PACA */
578	std	r1, HSTATE_HOST_R1(r13)
579
580	li	r6, KVM_GUEST_MODE_HOST_HV
581	stb	r6, HSTATE_IN_GUEST(r13)
582
583#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
584	/* Store initial timestamp */
585	cmpdi	r4, 0
586	beq	1f
587	addi	r3, r4, VCPU_TB_RMENTRY
588	bl	kvmhv_start_timing
5891:
590#endif
591
592	/* Use cr7 as an indication of radix mode */
593	ld	r5, HSTATE_KVM_VCORE(r13)
594	ld	r9, VCORE_KVM(r5)	/* pointer to struct kvm */
595	lbz	r0, KVM_RADIX(r9)
596	cmpwi	cr7, r0, 0
597
598	/*
599	 * POWER7/POWER8 host -> guest partition switch code.
600	 * We don't have to lock against concurrent tlbies,
601	 * but we do have to coordinate across hardware threads.
602	 */
603	/* Set bit in entry map iff exit map is zero. */
604	li	r7, 1
605	lbz	r6, HSTATE_PTID(r13)
606	sld	r7, r7, r6
607	addi	r8, r5, VCORE_ENTRY_EXIT
60821:	lwarx	r3, 0, r8
609	cmpwi	r3, 0x100		/* any threads starting to exit? */
610	bge	secondary_too_late	/* if so we're too late to the party */
611	or	r3, r3, r7
612	stwcx.	r3, 0, r8
613	bne	21b
614
615	/* Primary thread switches to guest partition. */
616	cmpwi	r6,0
617	bne	10f
618
619	/* Radix has already switched LPID and flushed core TLB */
620	bne	cr7, 22f
621
622	lwz	r7,KVM_LPID(r9)
623BEGIN_FTR_SECTION
624	ld	r6,KVM_SDR1(r9)
625	li	r0,LPID_RSVD		/* switch to reserved LPID */
626	mtspr	SPRN_LPID,r0
627	ptesync
628	mtspr	SPRN_SDR1,r6		/* switch to partition page table */
629END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
630	mtspr	SPRN_LPID,r7
631	isync
632
633	/* See if we need to flush the TLB. Hash has to be done in RM */
634	lhz	r6,PACAPACAINDEX(r13)	/* test_bit(cpu, need_tlb_flush) */
635BEGIN_FTR_SECTION
636	/*
637	 * On POWER9, individual threads can come in here, but the
638	 * TLB is shared between the 4 threads in a core, hence
639	 * invalidating on one thread invalidates for all.
640	 * Thus we make all 4 threads use the same bit here.
641	 */
642	clrrdi	r6,r6,2
643END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
644	clrldi	r7,r6,64-6		/* extract bit number (6 bits) */
645	srdi	r6,r6,6			/* doubleword number */
646	sldi	r6,r6,3			/* address offset */
647	add	r6,r6,r9
648	addi	r6,r6,KVM_NEED_FLUSH	/* dword in kvm->arch.need_tlb_flush */
649	li	r8,1
650	sld	r8,r8,r7
651	ld	r7,0(r6)
652	and.	r7,r7,r8
653	beq	22f
654	/* Flush the TLB of any entries for this LPID */
655	lwz	r0,KVM_TLB_SETS(r9)
656	mtctr	r0
657	li	r7,0x800		/* IS field = 0b10 */
658	ptesync
659	li	r0,0			/* RS for P9 version of tlbiel */
66028:	tlbiel	r7			/* On P9, rs=0, RIC=0, PRS=0, R=0 */
661	addi	r7,r7,0x1000
662	bdnz	28b
663	ptesync
66423:	ldarx	r7,0,r6			/* clear the bit after TLB flushed */
665	andc	r7,r7,r8
666	stdcx.	r7,0,r6
667	bne	23b
668
669	/* Add timebase offset onto timebase */
67022:	ld	r8,VCORE_TB_OFFSET(r5)
671	cmpdi	r8,0
672	beq	37f
673	std	r8, VCORE_TB_OFFSET_APPL(r5)
674	mftb	r6		/* current host timebase */
675	add	r8,r8,r6
676	mtspr	SPRN_TBU40,r8	/* update upper 40 bits */
677	mftb	r7		/* check if lower 24 bits overflowed */
678	clrldi	r6,r6,40
679	clrldi	r7,r7,40
680	cmpld	r7,r6
681	bge	37f
682	addis	r8,r8,0x100	/* if so, increment upper 40 bits */
683	mtspr	SPRN_TBU40,r8
684
685	/* Load guest PCR value to select appropriate compat mode */
68637:	ld	r7, VCORE_PCR(r5)
687	cmpdi	r7, 0
688	beq	38f
689	mtspr	SPRN_PCR, r7
69038:
691
692BEGIN_FTR_SECTION
693	/* DPDES and VTB are shared between threads */
694	ld	r8, VCORE_DPDES(r5)
695	ld	r7, VCORE_VTB(r5)
696	mtspr	SPRN_DPDES, r8
697	mtspr	SPRN_VTB, r7
698END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
699
700	/* Mark the subcore state as inside guest */
701	bl	kvmppc_subcore_enter_guest
702	nop
703	ld	r5, HSTATE_KVM_VCORE(r13)
704	ld	r4, HSTATE_KVM_VCPU(r13)
705	li	r0,1
706	stb	r0,VCORE_IN_GUEST(r5)	/* signal secondaries to continue */
707
708	/* Do we have a guest vcpu to run? */
70910:	cmpdi	r4, 0
710	beq	kvmppc_primary_no_guest
711kvmppc_got_guest:
712	/* Increment yield count if they have a VPA */
713	ld	r3, VCPU_VPA(r4)
714	cmpdi	r3, 0
715	beq	25f
716	li	r6, LPPACA_YIELDCOUNT
717	LWZX_BE	r5, r3, r6
718	addi	r5, r5, 1
719	STWX_BE	r5, r3, r6
720	li	r6, 1
721	stb	r6, VCPU_VPA_DIRTY(r4)
72225:
723
724	/* Save purr/spurr */
725	mfspr	r5,SPRN_PURR
726	mfspr	r6,SPRN_SPURR
727	std	r5,HSTATE_PURR(r13)
728	std	r6,HSTATE_SPURR(r13)
729	ld	r7,VCPU_PURR(r4)
730	ld	r8,VCPU_SPURR(r4)
731	mtspr	SPRN_PURR,r7
732	mtspr	SPRN_SPURR,r8
733
734	/* Save host values of some registers */
735BEGIN_FTR_SECTION
736	mfspr	r5, SPRN_TIDR
737	mfspr	r6, SPRN_PSSCR
738	mfspr	r7, SPRN_PID
739	std	r5, STACK_SLOT_TID(r1)
740	std	r6, STACK_SLOT_PSSCR(r1)
741	std	r7, STACK_SLOT_PID(r1)
742	mfspr	r5, SPRN_HFSCR
743	std	r5, STACK_SLOT_HFSCR(r1)
744END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
745BEGIN_FTR_SECTION
746	mfspr	r5, SPRN_CIABR
747	mfspr	r6, SPRN_DAWR
748	mfspr	r7, SPRN_DAWRX
749	mfspr	r8, SPRN_IAMR
750	std	r5, STACK_SLOT_CIABR(r1)
751	std	r6, STACK_SLOT_DAWR(r1)
752	std	r7, STACK_SLOT_DAWRX(r1)
753	std	r8, STACK_SLOT_IAMR(r1)
754END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
755
756	mfspr	r5, SPRN_AMR
757	std	r5, STACK_SLOT_AMR(r1)
758	mfspr	r6, SPRN_UAMOR
759	std	r6, STACK_SLOT_UAMOR(r1)
760
761BEGIN_FTR_SECTION
762	/* Set partition DABR */
763	/* Do this before re-enabling PMU to avoid P7 DABR corruption bug */
764	lwz	r5,VCPU_DABRX(r4)
765	ld	r6,VCPU_DABR(r4)
766	mtspr	SPRN_DABRX,r5
767	mtspr	SPRN_DABR,r6
768	isync
769END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
770
771#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
772/*
773 * Branch around the call if both CPU_FTR_TM and
774 * CPU_FTR_P9_TM_HV_ASSIST are off.
775 */
776BEGIN_FTR_SECTION
777	b	91f
778END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
779	/*
780	 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
781	 */
782	mr      r3, r4
783	ld      r4, VCPU_MSR(r3)
784	li	r5, 0			/* don't preserve non-vol regs */
785	bl	kvmppc_restore_tm_hv
786	nop
787	ld	r4, HSTATE_KVM_VCPU(r13)
78891:
789#endif
790
791	/* Load guest PMU registers; r4 = vcpu pointer here */
792	mr	r3, r4
793	bl	kvmhv_load_guest_pmu
794
795	/* Load up FP, VMX and VSX registers */
796	ld	r4, HSTATE_KVM_VCPU(r13)
797	bl	kvmppc_load_fp
798
799	ld	r14, VCPU_GPR(R14)(r4)
800	ld	r15, VCPU_GPR(R15)(r4)
801	ld	r16, VCPU_GPR(R16)(r4)
802	ld	r17, VCPU_GPR(R17)(r4)
803	ld	r18, VCPU_GPR(R18)(r4)
804	ld	r19, VCPU_GPR(R19)(r4)
805	ld	r20, VCPU_GPR(R20)(r4)
806	ld	r21, VCPU_GPR(R21)(r4)
807	ld	r22, VCPU_GPR(R22)(r4)
808	ld	r23, VCPU_GPR(R23)(r4)
809	ld	r24, VCPU_GPR(R24)(r4)
810	ld	r25, VCPU_GPR(R25)(r4)
811	ld	r26, VCPU_GPR(R26)(r4)
812	ld	r27, VCPU_GPR(R27)(r4)
813	ld	r28, VCPU_GPR(R28)(r4)
814	ld	r29, VCPU_GPR(R29)(r4)
815	ld	r30, VCPU_GPR(R30)(r4)
816	ld	r31, VCPU_GPR(R31)(r4)
817
818	/* Switch DSCR to guest value */
819	ld	r5, VCPU_DSCR(r4)
820	mtspr	SPRN_DSCR, r5
821
822BEGIN_FTR_SECTION
823	/* Skip next section on POWER7 */
824	b	8f
825END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
826	/* Load up POWER8-specific registers */
827	ld	r5, VCPU_IAMR(r4)
828	lwz	r6, VCPU_PSPB(r4)
829	ld	r7, VCPU_FSCR(r4)
830	mtspr	SPRN_IAMR, r5
831	mtspr	SPRN_PSPB, r6
832	mtspr	SPRN_FSCR, r7
833	/*
834	 * Handle broken DAWR case by not writing it. This means we
835	 * can still store the DAWR register for migration.
836	 */
837	LOAD_REG_ADDR(r5, dawr_force_enable)
838	lbz	r5, 0(r5)
839	cmpdi	r5, 0
840	beq	1f
841	ld	r5, VCPU_DAWR(r4)
842	ld	r6, VCPU_DAWRX(r4)
843	mtspr	SPRN_DAWR, r5
844	mtspr	SPRN_DAWRX, r6
8451:
846	ld	r7, VCPU_CIABR(r4)
847	ld	r8, VCPU_TAR(r4)
848	mtspr	SPRN_CIABR, r7
849	mtspr	SPRN_TAR, r8
850	ld	r5, VCPU_IC(r4)
851	ld	r8, VCPU_EBBHR(r4)
852	mtspr	SPRN_IC, r5
853	mtspr	SPRN_EBBHR, r8
854	ld	r5, VCPU_EBBRR(r4)
855	ld	r6, VCPU_BESCR(r4)
856	lwz	r7, VCPU_GUEST_PID(r4)
857	ld	r8, VCPU_WORT(r4)
858	mtspr	SPRN_EBBRR, r5
859	mtspr	SPRN_BESCR, r6
860	mtspr	SPRN_PID, r7
861	mtspr	SPRN_WORT, r8
862BEGIN_FTR_SECTION
863	/* POWER8-only registers */
864	ld	r5, VCPU_TCSCR(r4)
865	ld	r6, VCPU_ACOP(r4)
866	ld	r7, VCPU_CSIGR(r4)
867	ld	r8, VCPU_TACR(r4)
868	mtspr	SPRN_TCSCR, r5
869	mtspr	SPRN_ACOP, r6
870	mtspr	SPRN_CSIGR, r7
871	mtspr	SPRN_TACR, r8
872	nop
873FTR_SECTION_ELSE
874	/* POWER9-only registers */
875	ld	r5, VCPU_TID(r4)
876	ld	r6, VCPU_PSSCR(r4)
877	lbz	r8, HSTATE_FAKE_SUSPEND(r13)
878	oris	r6, r6, PSSCR_EC@h	/* This makes stop trap to HV */
879	rldimi	r6, r8, PSSCR_FAKE_SUSPEND_LG, 63 - PSSCR_FAKE_SUSPEND_LG
880	ld	r7, VCPU_HFSCR(r4)
881	mtspr	SPRN_TIDR, r5
882	mtspr	SPRN_PSSCR, r6
883	mtspr	SPRN_HFSCR, r7
884ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
8858:
886
887	ld	r5, VCPU_SPRG0(r4)
888	ld	r6, VCPU_SPRG1(r4)
889	ld	r7, VCPU_SPRG2(r4)
890	ld	r8, VCPU_SPRG3(r4)
891	mtspr	SPRN_SPRG0, r5
892	mtspr	SPRN_SPRG1, r6
893	mtspr	SPRN_SPRG2, r7
894	mtspr	SPRN_SPRG3, r8
895
896	/* Load up DAR and DSISR */
897	ld	r5, VCPU_DAR(r4)
898	lwz	r6, VCPU_DSISR(r4)
899	mtspr	SPRN_DAR, r5
900	mtspr	SPRN_DSISR, r6
901
902	/* Restore AMR and UAMOR, set AMOR to all 1s */
903	ld	r5,VCPU_AMR(r4)
904	ld	r6,VCPU_UAMOR(r4)
905	li	r7,-1
906	mtspr	SPRN_AMR,r5
907	mtspr	SPRN_UAMOR,r6
908	mtspr	SPRN_AMOR,r7
909
910	/* Restore state of CTRL run bit; assume 1 on entry */
911	lwz	r5,VCPU_CTRL(r4)
912	andi.	r5,r5,1
913	bne	4f
914	mfspr	r6,SPRN_CTRLF
915	clrrdi	r6,r6,1
916	mtspr	SPRN_CTRLT,r6
9174:
918	/* Secondary threads wait for primary to have done partition switch */
919	ld	r5, HSTATE_KVM_VCORE(r13)
920	lbz	r6, HSTATE_PTID(r13)
921	cmpwi	r6, 0
922	beq	21f
923	lbz	r0, VCORE_IN_GUEST(r5)
924	cmpwi	r0, 0
925	bne	21f
926	HMT_LOW
92720:	lwz	r3, VCORE_ENTRY_EXIT(r5)
928	cmpwi	r3, 0x100
929	bge	no_switch_exit
930	lbz	r0, VCORE_IN_GUEST(r5)
931	cmpwi	r0, 0
932	beq	20b
933	HMT_MEDIUM
93421:
935	/* Set LPCR. */
936	ld	r8,VCORE_LPCR(r5)
937	mtspr	SPRN_LPCR,r8
938	isync
939
940	/*
941	 * Set the decrementer to the guest decrementer.
942	 */
943	ld	r8,VCPU_DEC_EXPIRES(r4)
944	/* r8 is a host timebase value here, convert to guest TB */
945	ld	r5,HSTATE_KVM_VCORE(r13)
946	ld	r6,VCORE_TB_OFFSET_APPL(r5)
947	add	r8,r8,r6
948	mftb	r7
949	subf	r3,r7,r8
950	mtspr	SPRN_DEC,r3
951
952	/* Check if HDEC expires soon */
953	mfspr	r3, SPRN_HDEC
954	EXTEND_HDEC(r3)
955	cmpdi	r3, 512		/* 1 microsecond */
956	blt	hdec_soon
957
958	/* For hash guest, clear out and reload the SLB */
959	ld	r6, VCPU_KVM(r4)
960	lbz	r0, KVM_RADIX(r6)
961	cmpwi	r0, 0
962	bne	9f
963	li	r6, 0
964	slbmte	r6, r6
965	slbia
966	ptesync
967
968	/* Load up guest SLB entries (N.B. slb_max will be 0 for radix) */
969	lwz	r5,VCPU_SLB_MAX(r4)
970	cmpwi	r5,0
971	beq	9f
972	mtctr	r5
973	addi	r6,r4,VCPU_SLB
9741:	ld	r8,VCPU_SLB_E(r6)
975	ld	r9,VCPU_SLB_V(r6)
976	slbmte	r9,r8
977	addi	r6,r6,VCPU_SLB_SIZE
978	bdnz	1b
9799:
980
981#ifdef CONFIG_KVM_XICS
982	/* We are entering the guest on that thread, push VCPU to XIVE */
983	ld	r10, HSTATE_XIVE_TIMA_PHYS(r13)
984	cmpldi	cr0, r10, 0
985	beq	no_xive
986	ld	r11, VCPU_XIVE_SAVED_STATE(r4)
987	li	r9, TM_QW1_OS
988	eieio
989	stdcix	r11,r9,r10
990	lwz	r11, VCPU_XIVE_CAM_WORD(r4)
991	li	r9, TM_QW1_OS + TM_WORD2
992	stwcix	r11,r9,r10
993	li	r9, 1
994	stb	r9, VCPU_XIVE_PUSHED(r4)
995	eieio
996
997	/*
998	 * We clear the irq_pending flag. There is a small chance of a
999	 * race vs. the escalation interrupt happening on another
1000	 * processor setting it again, but the only consequence is to
1001	 * cause a spurrious wakeup on the next H_CEDE which is not an
1002	 * issue.
1003	 */
1004	li	r0,0
1005	stb	r0, VCPU_IRQ_PENDING(r4)
1006
1007	/*
1008	 * In single escalation mode, if the escalation interrupt is
1009	 * on, we mask it.
1010	 */
1011	lbz	r0, VCPU_XIVE_ESC_ON(r4)
1012	cmpwi	r0,0
1013	beq	1f
1014	ld	r10, VCPU_XIVE_ESC_RADDR(r4)
1015	li	r9, XIVE_ESB_SET_PQ_01
1016	ldcix	r0, r10, r9
1017	sync
1018
1019	/* We have a possible subtle race here: The escalation interrupt might
1020	 * have fired and be on its way to the host queue while we mask it,
1021	 * and if we unmask it early enough (re-cede right away), there is
1022	 * a theorical possibility that it fires again, thus landing in the
1023	 * target queue more than once which is a big no-no.
1024	 *
1025	 * Fortunately, solving this is rather easy. If the above load setting
1026	 * PQ to 01 returns a previous value where P is set, then we know the
1027	 * escalation interrupt is somewhere on its way to the host. In that
1028	 * case we simply don't clear the xive_esc_on flag below. It will be
1029	 * eventually cleared by the handler for the escalation interrupt.
1030	 *
1031	 * Then, when doing a cede, we check that flag again before re-enabling
1032	 * the escalation interrupt, and if set, we abort the cede.
1033	 */
1034	andi.	r0, r0, XIVE_ESB_VAL_P
1035	bne-	1f
1036
1037	/* Now P is 0, we can clear the flag */
1038	li	r0, 0
1039	stb	r0, VCPU_XIVE_ESC_ON(r4)
10401:
1041no_xive:
1042#endif /* CONFIG_KVM_XICS */
1043
1044	li	r0, 0
1045	stw	r0, STACK_SLOT_SHORT_PATH(r1)
1046
1047deliver_guest_interrupt:	/* r4 = vcpu, r13 = paca */
1048	/* Check if we can deliver an external or decrementer interrupt now */
1049	ld	r0, VCPU_PENDING_EXC(r4)
1050BEGIN_FTR_SECTION
1051	/* On POWER9, also check for emulated doorbell interrupt */
1052	lbz	r3, VCPU_DBELL_REQ(r4)
1053	or	r0, r0, r3
1054END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1055	cmpdi	r0, 0
1056	beq	71f
1057	mr	r3, r4
1058	bl	kvmppc_guest_entry_inject_int
1059	ld	r4, HSTATE_KVM_VCPU(r13)
106071:
1061	ld	r6, VCPU_SRR0(r4)
1062	ld	r7, VCPU_SRR1(r4)
1063	mtspr	SPRN_SRR0, r6
1064	mtspr	SPRN_SRR1, r7
1065
1066fast_guest_entry_c:
1067	ld	r10, VCPU_PC(r4)
1068	ld	r11, VCPU_MSR(r4)
1069	/* r11 = vcpu->arch.msr & ~MSR_HV */
1070	rldicl	r11, r11, 63 - MSR_HV_LG, 1
1071	rotldi	r11, r11, 1 + MSR_HV_LG
1072	ori	r11, r11, MSR_ME
1073
1074	ld	r6, VCPU_CTR(r4)
1075	ld	r7, VCPU_XER(r4)
1076	mtctr	r6
1077	mtxer	r7
1078
1079/*
1080 * Required state:
1081 * R4 = vcpu
1082 * R10: value for HSRR0
1083 * R11: value for HSRR1
1084 * R13 = PACA
1085 */
1086fast_guest_return:
1087	li	r0,0
1088	stb	r0,VCPU_CEDED(r4)	/* cancel cede */
1089	mtspr	SPRN_HSRR0,r10
1090	mtspr	SPRN_HSRR1,r11
1091
1092	/* Activate guest mode, so faults get handled by KVM */
1093	li	r9, KVM_GUEST_MODE_GUEST_HV
1094	stb	r9, HSTATE_IN_GUEST(r13)
1095
1096#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1097	/* Accumulate timing */
1098	addi	r3, r4, VCPU_TB_GUEST
1099	bl	kvmhv_accumulate_time
1100#endif
1101
1102	/* Enter guest */
1103
1104BEGIN_FTR_SECTION
1105	ld	r5, VCPU_CFAR(r4)
1106	mtspr	SPRN_CFAR, r5
1107END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
1108BEGIN_FTR_SECTION
1109	ld	r0, VCPU_PPR(r4)
1110END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1111
1112	ld	r5, VCPU_LR(r4)
1113	ld	r6, VCPU_CR(r4)
1114	mtlr	r5
1115	mtcr	r6
1116
1117	ld	r1, VCPU_GPR(R1)(r4)
1118	ld	r2, VCPU_GPR(R2)(r4)
1119	ld	r3, VCPU_GPR(R3)(r4)
1120	ld	r5, VCPU_GPR(R5)(r4)
1121	ld	r6, VCPU_GPR(R6)(r4)
1122	ld	r7, VCPU_GPR(R7)(r4)
1123	ld	r8, VCPU_GPR(R8)(r4)
1124	ld	r9, VCPU_GPR(R9)(r4)
1125	ld	r10, VCPU_GPR(R10)(r4)
1126	ld	r11, VCPU_GPR(R11)(r4)
1127	ld	r12, VCPU_GPR(R12)(r4)
1128	ld	r13, VCPU_GPR(R13)(r4)
1129
1130BEGIN_FTR_SECTION
1131	mtspr	SPRN_PPR, r0
1132END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1133
1134/* Move canary into DSISR to check for later */
1135BEGIN_FTR_SECTION
1136	li	r0, 0x7fff
1137	mtspr	SPRN_HDSISR, r0
1138END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1139
1140	ld	r0, VCPU_GPR(R0)(r4)
1141	ld	r4, VCPU_GPR(R4)(r4)
1142	HRFI_TO_GUEST
1143	b	.
1144
1145/*
1146 * Enter the guest on a P9 or later system where we have exactly
1147 * one vcpu per vcore and we don't need to go to real mode
1148 * (which implies that host and guest are both using radix MMU mode).
1149 * r3 = vcpu pointer
1150 * Most SPRs and all the VSRs have been loaded already.
1151 */
1152_GLOBAL(__kvmhv_vcpu_entry_p9)
1153EXPORT_SYMBOL_GPL(__kvmhv_vcpu_entry_p9)
1154	mflr	r0
1155	std	r0, PPC_LR_STKOFF(r1)
1156	stdu	r1, -SFS(r1)
1157
1158	li	r0, 1
1159	stw	r0, STACK_SLOT_SHORT_PATH(r1)
1160
1161	std	r3, HSTATE_KVM_VCPU(r13)
1162	mfcr	r4
1163	stw	r4, SFS+8(r1)
1164
1165	std	r1, HSTATE_HOST_R1(r13)
1166
1167	reg = 14
1168	.rept	18
1169	std	reg, STACK_SLOT_NVGPRS + ((reg - 14) * 8)(r1)
1170	reg = reg + 1
1171	.endr
1172
1173	reg = 14
1174	.rept	18
1175	ld	reg, __VCPU_GPR(reg)(r3)
1176	reg = reg + 1
1177	.endr
1178
1179	mfmsr	r10
1180	std	r10, HSTATE_HOST_MSR(r13)
1181
1182	mr	r4, r3
1183	b	fast_guest_entry_c
1184guest_exit_short_path:
1185
1186	li	r0, KVM_GUEST_MODE_NONE
1187	stb	r0, HSTATE_IN_GUEST(r13)
1188
1189	reg = 14
1190	.rept	18
1191	std	reg, __VCPU_GPR(reg)(r9)
1192	reg = reg + 1
1193	.endr
1194
1195	reg = 14
1196	.rept	18
1197	ld	reg, STACK_SLOT_NVGPRS + ((reg - 14) * 8)(r1)
1198	reg = reg + 1
1199	.endr
1200
1201	lwz	r4, SFS+8(r1)
1202	mtcr	r4
1203
1204	mr	r3, r12		/* trap number */
1205
1206	addi	r1, r1, SFS
1207	ld	r0, PPC_LR_STKOFF(r1)
1208	mtlr	r0
1209
1210	/* If we are in real mode, do a rfid to get back to the caller */
1211	mfmsr	r4
1212	andi.	r5, r4, MSR_IR
1213	bnelr
1214	rldicl	r5, r4, 64 - MSR_TS_S_LG, 62	/* extract TS field */
1215	mtspr	SPRN_SRR0, r0
1216	ld	r10, HSTATE_HOST_MSR(r13)
1217	rldimi	r10, r5, MSR_TS_S_LG, 63 - MSR_TS_T_LG
1218	mtspr	SPRN_SRR1, r10
1219	RFI_TO_KERNEL
1220	b	.
1221
1222secondary_too_late:
1223	li	r12, 0
1224	stw	r12, STACK_SLOT_TRAP(r1)
1225	cmpdi	r4, 0
1226	beq	11f
1227	stw	r12, VCPU_TRAP(r4)
1228#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1229	addi	r3, r4, VCPU_TB_RMEXIT
1230	bl	kvmhv_accumulate_time
1231#endif
123211:	b	kvmhv_switch_to_host
1233
1234no_switch_exit:
1235	HMT_MEDIUM
1236	li	r12, 0
1237	b	12f
1238hdec_soon:
1239	li	r12, BOOK3S_INTERRUPT_HV_DECREMENTER
124012:	stw	r12, VCPU_TRAP(r4)
1241	mr	r9, r4
1242#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1243	addi	r3, r4, VCPU_TB_RMEXIT
1244	bl	kvmhv_accumulate_time
1245#endif
1246	b	guest_bypass
1247
1248/******************************************************************************
1249 *                                                                            *
1250 *                               Exit code                                    *
1251 *                                                                            *
1252 *****************************************************************************/
1253
1254/*
1255 * We come here from the first-level interrupt handlers.
1256 */
1257	.globl	kvmppc_interrupt_hv
1258kvmppc_interrupt_hv:
1259	/*
1260	 * Register contents:
1261	 * R12		= (guest CR << 32) | interrupt vector
1262	 * R13		= PACA
1263	 * guest R12 saved in shadow VCPU SCRATCH0
1264	 * guest CTR saved in shadow VCPU SCRATCH1 if RELOCATABLE
1265	 * guest R13 saved in SPRN_SCRATCH0
1266	 */
1267	std	r9, HSTATE_SCRATCH2(r13)
1268	lbz	r9, HSTATE_IN_GUEST(r13)
1269	cmpwi	r9, KVM_GUEST_MODE_HOST_HV
1270	beq	kvmppc_bad_host_intr
1271#ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1272	cmpwi	r9, KVM_GUEST_MODE_GUEST
1273	ld	r9, HSTATE_SCRATCH2(r13)
1274	beq	kvmppc_interrupt_pr
1275#endif
1276	/* We're now back in the host but in guest MMU context */
1277	li	r9, KVM_GUEST_MODE_HOST_HV
1278	stb	r9, HSTATE_IN_GUEST(r13)
1279
1280	ld	r9, HSTATE_KVM_VCPU(r13)
1281
1282	/* Save registers */
1283
1284	std	r0, VCPU_GPR(R0)(r9)
1285	std	r1, VCPU_GPR(R1)(r9)
1286	std	r2, VCPU_GPR(R2)(r9)
1287	std	r3, VCPU_GPR(R3)(r9)
1288	std	r4, VCPU_GPR(R4)(r9)
1289	std	r5, VCPU_GPR(R5)(r9)
1290	std	r6, VCPU_GPR(R6)(r9)
1291	std	r7, VCPU_GPR(R7)(r9)
1292	std	r8, VCPU_GPR(R8)(r9)
1293	ld	r0, HSTATE_SCRATCH2(r13)
1294	std	r0, VCPU_GPR(R9)(r9)
1295	std	r10, VCPU_GPR(R10)(r9)
1296	std	r11, VCPU_GPR(R11)(r9)
1297	ld	r3, HSTATE_SCRATCH0(r13)
1298	std	r3, VCPU_GPR(R12)(r9)
1299	/* CR is in the high half of r12 */
1300	srdi	r4, r12, 32
1301	std	r4, VCPU_CR(r9)
1302BEGIN_FTR_SECTION
1303	ld	r3, HSTATE_CFAR(r13)
1304	std	r3, VCPU_CFAR(r9)
1305END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
1306BEGIN_FTR_SECTION
1307	ld	r4, HSTATE_PPR(r13)
1308	std	r4, VCPU_PPR(r9)
1309END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1310
1311	/* Restore R1/R2 so we can handle faults */
1312	ld	r1, HSTATE_HOST_R1(r13)
1313	ld	r2, PACATOC(r13)
1314
1315	mfspr	r10, SPRN_SRR0
1316	mfspr	r11, SPRN_SRR1
1317	std	r10, VCPU_SRR0(r9)
1318	std	r11, VCPU_SRR1(r9)
1319	/* trap is in the low half of r12, clear CR from the high half */
1320	clrldi	r12, r12, 32
1321	andi.	r0, r12, 2		/* need to read HSRR0/1? */
1322	beq	1f
1323	mfspr	r10, SPRN_HSRR0
1324	mfspr	r11, SPRN_HSRR1
1325	clrrdi	r12, r12, 2
13261:	std	r10, VCPU_PC(r9)
1327	std	r11, VCPU_MSR(r9)
1328
1329	GET_SCRATCH0(r3)
1330	mflr	r4
1331	std	r3, VCPU_GPR(R13)(r9)
1332	std	r4, VCPU_LR(r9)
1333
1334	stw	r12,VCPU_TRAP(r9)
1335
1336	/*
1337	 * Now that we have saved away SRR0/1 and HSRR0/1,
1338	 * interrupts are recoverable in principle, so set MSR_RI.
1339	 * This becomes important for relocation-on interrupts from
1340	 * the guest, which we can get in radix mode on POWER9.
1341	 */
1342	li	r0, MSR_RI
1343	mtmsrd	r0, 1
1344
1345#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1346	addi	r3, r9, VCPU_TB_RMINTR
1347	mr	r4, r9
1348	bl	kvmhv_accumulate_time
1349	ld	r5, VCPU_GPR(R5)(r9)
1350	ld	r6, VCPU_GPR(R6)(r9)
1351	ld	r7, VCPU_GPR(R7)(r9)
1352	ld	r8, VCPU_GPR(R8)(r9)
1353#endif
1354
1355	/* Save HEIR (HV emulation assist reg) in emul_inst
1356	   if this is an HEI (HV emulation interrupt, e40) */
1357	li	r3,KVM_INST_FETCH_FAILED
1358	stw	r3,VCPU_LAST_INST(r9)
1359	cmpwi	r12,BOOK3S_INTERRUPT_H_EMUL_ASSIST
1360	bne	11f
1361	mfspr	r3,SPRN_HEIR
136211:	stw	r3,VCPU_HEIR(r9)
1363
1364	/* these are volatile across C function calls */
1365#ifdef CONFIG_RELOCATABLE
1366	ld	r3, HSTATE_SCRATCH1(r13)
1367	mtctr	r3
1368#else
1369	mfctr	r3
1370#endif
1371	mfxer	r4
1372	std	r3, VCPU_CTR(r9)
1373	std	r4, VCPU_XER(r9)
1374
1375	/* Save more register state  */
1376	mfdar	r3
1377	mfdsisr	r4
1378	std	r3, VCPU_DAR(r9)
1379	stw	r4, VCPU_DSISR(r9)
1380
1381	/* If this is a page table miss then see if it's theirs or ours */
1382	cmpwi	r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
1383	beq	kvmppc_hdsi
1384	std	r3, VCPU_FAULT_DAR(r9)
1385	stw	r4, VCPU_FAULT_DSISR(r9)
1386	cmpwi	r12, BOOK3S_INTERRUPT_H_INST_STORAGE
1387	beq	kvmppc_hisi
1388
1389#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1390	/* For softpatch interrupt, go off and do TM instruction emulation */
1391	cmpwi	r12, BOOK3S_INTERRUPT_HV_SOFTPATCH
1392	beq	kvmppc_tm_emul
1393#endif
1394
1395	/* See if this is a leftover HDEC interrupt */
1396	cmpwi	r12,BOOK3S_INTERRUPT_HV_DECREMENTER
1397	bne	2f
1398	mfspr	r3,SPRN_HDEC
1399	EXTEND_HDEC(r3)
1400	cmpdi	r3,0
1401	mr	r4,r9
1402	bge	fast_guest_return
14032:
1404	/* See if this is an hcall we can handle in real mode */
1405	cmpwi	r12,BOOK3S_INTERRUPT_SYSCALL
1406	beq	hcall_try_real_mode
1407
1408	/* Hypervisor doorbell - exit only if host IPI flag set */
1409	cmpwi	r12, BOOK3S_INTERRUPT_H_DOORBELL
1410	bne	3f
1411BEGIN_FTR_SECTION
1412	PPC_MSGSYNC
1413	lwsync
1414	/* always exit if we're running a nested guest */
1415	ld	r0, VCPU_NESTED(r9)
1416	cmpdi	r0, 0
1417	bne	guest_exit_cont
1418END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1419	lbz	r0, HSTATE_HOST_IPI(r13)
1420	cmpwi	r0, 0
1421	beq	maybe_reenter_guest
1422	b	guest_exit_cont
14233:
1424	/* If it's a hypervisor facility unavailable interrupt, save HFSCR */
1425	cmpwi	r12, BOOK3S_INTERRUPT_H_FAC_UNAVAIL
1426	bne	14f
1427	mfspr	r3, SPRN_HFSCR
1428	std	r3, VCPU_HFSCR(r9)
1429	b	guest_exit_cont
143014:
1431	/* External interrupt ? */
1432	cmpwi	r12, BOOK3S_INTERRUPT_EXTERNAL
1433	beq	kvmppc_guest_external
1434	/* See if it is a machine check */
1435	cmpwi	r12, BOOK3S_INTERRUPT_MACHINE_CHECK
1436	beq	machine_check_realmode
1437	/* Or a hypervisor maintenance interrupt */
1438	cmpwi	r12, BOOK3S_INTERRUPT_HMI
1439	beq	hmi_realmode
1440
1441guest_exit_cont:		/* r9 = vcpu, r12 = trap, r13 = paca */
1442
1443#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1444	addi	r3, r9, VCPU_TB_RMEXIT
1445	mr	r4, r9
1446	bl	kvmhv_accumulate_time
1447#endif
1448#ifdef CONFIG_KVM_XICS
1449	/* We are exiting, pull the VP from the XIVE */
1450	lbz	r0, VCPU_XIVE_PUSHED(r9)
1451	cmpwi	cr0, r0, 0
1452	beq	1f
1453	li	r7, TM_SPC_PULL_OS_CTX
1454	li	r6, TM_QW1_OS
1455	mfmsr	r0
1456	andi.	r0, r0, MSR_DR		/* in real mode? */
1457	beq	2f
1458	ld	r10, HSTATE_XIVE_TIMA_VIRT(r13)
1459	cmpldi	cr0, r10, 0
1460	beq	1f
1461	/* First load to pull the context, we ignore the value */
1462	eieio
1463	lwzx	r11, r7, r10
1464	/* Second load to recover the context state (Words 0 and 1) */
1465	ldx	r11, r6, r10
1466	b	3f
14672:	ld	r10, HSTATE_XIVE_TIMA_PHYS(r13)
1468	cmpldi	cr0, r10, 0
1469	beq	1f
1470	/* First load to pull the context, we ignore the value */
1471	eieio
1472	lwzcix	r11, r7, r10
1473	/* Second load to recover the context state (Words 0 and 1) */
1474	ldcix	r11, r6, r10
14753:	std	r11, VCPU_XIVE_SAVED_STATE(r9)
1476	/* Fixup some of the state for the next load */
1477	li	r10, 0
1478	li	r0, 0xff
1479	stb	r10, VCPU_XIVE_PUSHED(r9)
1480	stb	r10, (VCPU_XIVE_SAVED_STATE+3)(r9)
1481	stb	r0, (VCPU_XIVE_SAVED_STATE+4)(r9)
1482	eieio
14831:
1484#endif /* CONFIG_KVM_XICS */
1485
1486	/* If we came in through the P9 short path, go back out to C now */
1487	lwz	r0, STACK_SLOT_SHORT_PATH(r1)
1488	cmpwi	r0, 0
1489	bne	guest_exit_short_path
1490
1491	/* For hash guest, read the guest SLB and save it away */
1492	ld	r5, VCPU_KVM(r9)
1493	lbz	r0, KVM_RADIX(r5)
1494	li	r5, 0
1495	cmpwi	r0, 0
1496	bne	3f			/* for radix, save 0 entries */
1497	lwz	r0,VCPU_SLB_NR(r9)	/* number of entries in SLB */
1498	mtctr	r0
1499	li	r6,0
1500	addi	r7,r9,VCPU_SLB
15011:	slbmfee	r8,r6
1502	andis.	r0,r8,SLB_ESID_V@h
1503	beq	2f
1504	add	r8,r8,r6		/* put index in */
1505	slbmfev	r3,r6
1506	std	r8,VCPU_SLB_E(r7)
1507	std	r3,VCPU_SLB_V(r7)
1508	addi	r7,r7,VCPU_SLB_SIZE
1509	addi	r5,r5,1
15102:	addi	r6,r6,1
1511	bdnz	1b
1512	/* Finally clear out the SLB */
1513	li	r0,0
1514	slbmte	r0,r0
1515	slbia
1516	ptesync
15173:	stw	r5,VCPU_SLB_MAX(r9)
1518
1519	/* load host SLB entries */
1520BEGIN_MMU_FTR_SECTION
1521	b	0f
1522END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
1523	ld	r8,PACA_SLBSHADOWPTR(r13)
1524
1525	.rept	SLB_NUM_BOLTED
1526	li	r3, SLBSHADOW_SAVEAREA
1527	LDX_BE	r5, r8, r3
1528	addi	r3, r3, 8
1529	LDX_BE	r6, r8, r3
1530	andis.	r7,r5,SLB_ESID_V@h
1531	beq	1f
1532	slbmte	r6,r5
15331:	addi	r8,r8,16
1534	.endr
15350:
1536
1537guest_bypass:
1538	stw	r12, STACK_SLOT_TRAP(r1)
1539
1540	/* Save DEC */
1541	/* Do this before kvmhv_commence_exit so we know TB is guest TB */
1542	ld	r3, HSTATE_KVM_VCORE(r13)
1543	mfspr	r5,SPRN_DEC
1544	mftb	r6
1545	/* On P9, if the guest has large decr enabled, don't sign extend */
1546BEGIN_FTR_SECTION
1547	ld	r4, VCORE_LPCR(r3)
1548	andis.	r4, r4, LPCR_LD@h
1549	bne	16f
1550END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1551	extsw	r5,r5
155216:	add	r5,r5,r6
1553	/* r5 is a guest timebase value here, convert to host TB */
1554	ld	r4,VCORE_TB_OFFSET_APPL(r3)
1555	subf	r5,r4,r5
1556	std	r5,VCPU_DEC_EXPIRES(r9)
1557
1558	/* Increment exit count, poke other threads to exit */
1559	mr 	r3, r12
1560	bl	kvmhv_commence_exit
1561	nop
1562	ld	r9, HSTATE_KVM_VCPU(r13)
1563
1564	/* Stop others sending VCPU interrupts to this physical CPU */
1565	li	r0, -1
1566	stw	r0, VCPU_CPU(r9)
1567	stw	r0, VCPU_THREAD_CPU(r9)
1568
1569	/* Save guest CTRL register, set runlatch to 1 */
1570	mfspr	r6,SPRN_CTRLF
1571	stw	r6,VCPU_CTRL(r9)
1572	andi.	r0,r6,1
1573	bne	4f
1574	ori	r6,r6,1
1575	mtspr	SPRN_CTRLT,r6
15764:
1577	/*
1578	 * Save the guest PURR/SPURR
1579	 */
1580	mfspr	r5,SPRN_PURR
1581	mfspr	r6,SPRN_SPURR
1582	ld	r7,VCPU_PURR(r9)
1583	ld	r8,VCPU_SPURR(r9)
1584	std	r5,VCPU_PURR(r9)
1585	std	r6,VCPU_SPURR(r9)
1586	subf	r5,r7,r5
1587	subf	r6,r8,r6
1588
1589	/*
1590	 * Restore host PURR/SPURR and add guest times
1591	 * so that the time in the guest gets accounted.
1592	 */
1593	ld	r3,HSTATE_PURR(r13)
1594	ld	r4,HSTATE_SPURR(r13)
1595	add	r3,r3,r5
1596	add	r4,r4,r6
1597	mtspr	SPRN_PURR,r3
1598	mtspr	SPRN_SPURR,r4
1599
1600BEGIN_FTR_SECTION
1601	b	8f
1602END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
1603	/* Save POWER8-specific registers */
1604	mfspr	r5, SPRN_IAMR
1605	mfspr	r6, SPRN_PSPB
1606	mfspr	r7, SPRN_FSCR
1607	std	r5, VCPU_IAMR(r9)
1608	stw	r6, VCPU_PSPB(r9)
1609	std	r7, VCPU_FSCR(r9)
1610	mfspr	r5, SPRN_IC
1611	mfspr	r7, SPRN_TAR
1612	std	r5, VCPU_IC(r9)
1613	std	r7, VCPU_TAR(r9)
1614	mfspr	r8, SPRN_EBBHR
1615	std	r8, VCPU_EBBHR(r9)
1616	mfspr	r5, SPRN_EBBRR
1617	mfspr	r6, SPRN_BESCR
1618	mfspr	r7, SPRN_PID
1619	mfspr	r8, SPRN_WORT
1620	std	r5, VCPU_EBBRR(r9)
1621	std	r6, VCPU_BESCR(r9)
1622	stw	r7, VCPU_GUEST_PID(r9)
1623	std	r8, VCPU_WORT(r9)
1624BEGIN_FTR_SECTION
1625	mfspr	r5, SPRN_TCSCR
1626	mfspr	r6, SPRN_ACOP
1627	mfspr	r7, SPRN_CSIGR
1628	mfspr	r8, SPRN_TACR
1629	std	r5, VCPU_TCSCR(r9)
1630	std	r6, VCPU_ACOP(r9)
1631	std	r7, VCPU_CSIGR(r9)
1632	std	r8, VCPU_TACR(r9)
1633FTR_SECTION_ELSE
1634	mfspr	r5, SPRN_TIDR
1635	mfspr	r6, SPRN_PSSCR
1636	std	r5, VCPU_TID(r9)
1637	rldicl	r6, r6, 4, 50		/* r6 &= PSSCR_GUEST_VIS */
1638	rotldi	r6, r6, 60
1639	std	r6, VCPU_PSSCR(r9)
1640	/* Restore host HFSCR value */
1641	ld	r7, STACK_SLOT_HFSCR(r1)
1642	mtspr	SPRN_HFSCR, r7
1643ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
1644	/*
1645	 * Restore various registers to 0, where non-zero values
1646	 * set by the guest could disrupt the host.
1647	 */
1648	li	r0, 0
1649	mtspr	SPRN_PSPB, r0
1650	mtspr	SPRN_WORT, r0
1651BEGIN_FTR_SECTION
1652	mtspr	SPRN_TCSCR, r0
1653	/* Set MMCRS to 1<<31 to freeze and disable the SPMC counters */
1654	li	r0, 1
1655	sldi	r0, r0, 31
1656	mtspr	SPRN_MMCRS, r0
1657END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1658
1659	/* Save and restore AMR, IAMR and UAMOR before turning on the MMU */
1660	ld	r8, STACK_SLOT_IAMR(r1)
1661	mtspr	SPRN_IAMR, r8
1662
16638:	/* Power7 jumps back in here */
1664	mfspr	r5,SPRN_AMR
1665	mfspr	r6,SPRN_UAMOR
1666	std	r5,VCPU_AMR(r9)
1667	std	r6,VCPU_UAMOR(r9)
1668	ld	r5,STACK_SLOT_AMR(r1)
1669	ld	r6,STACK_SLOT_UAMOR(r1)
1670	mtspr	SPRN_AMR, r5
1671	mtspr	SPRN_UAMOR, r6
1672
1673	/* Switch DSCR back to host value */
1674	mfspr	r8, SPRN_DSCR
1675	ld	r7, HSTATE_DSCR(r13)
1676	std	r8, VCPU_DSCR(r9)
1677	mtspr	SPRN_DSCR, r7
1678
1679	/* Save non-volatile GPRs */
1680	std	r14, VCPU_GPR(R14)(r9)
1681	std	r15, VCPU_GPR(R15)(r9)
1682	std	r16, VCPU_GPR(R16)(r9)
1683	std	r17, VCPU_GPR(R17)(r9)
1684	std	r18, VCPU_GPR(R18)(r9)
1685	std	r19, VCPU_GPR(R19)(r9)
1686	std	r20, VCPU_GPR(R20)(r9)
1687	std	r21, VCPU_GPR(R21)(r9)
1688	std	r22, VCPU_GPR(R22)(r9)
1689	std	r23, VCPU_GPR(R23)(r9)
1690	std	r24, VCPU_GPR(R24)(r9)
1691	std	r25, VCPU_GPR(R25)(r9)
1692	std	r26, VCPU_GPR(R26)(r9)
1693	std	r27, VCPU_GPR(R27)(r9)
1694	std	r28, VCPU_GPR(R28)(r9)
1695	std	r29, VCPU_GPR(R29)(r9)
1696	std	r30, VCPU_GPR(R30)(r9)
1697	std	r31, VCPU_GPR(R31)(r9)
1698
1699	/* Save SPRGs */
1700	mfspr	r3, SPRN_SPRG0
1701	mfspr	r4, SPRN_SPRG1
1702	mfspr	r5, SPRN_SPRG2
1703	mfspr	r6, SPRN_SPRG3
1704	std	r3, VCPU_SPRG0(r9)
1705	std	r4, VCPU_SPRG1(r9)
1706	std	r5, VCPU_SPRG2(r9)
1707	std	r6, VCPU_SPRG3(r9)
1708
1709	/* save FP state */
1710	mr	r3, r9
1711	bl	kvmppc_save_fp
1712
1713#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1714/*
1715 * Branch around the call if both CPU_FTR_TM and
1716 * CPU_FTR_P9_TM_HV_ASSIST are off.
1717 */
1718BEGIN_FTR_SECTION
1719	b	91f
1720END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
1721	/*
1722	 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
1723	 */
1724	mr      r3, r9
1725	ld      r4, VCPU_MSR(r3)
1726	li	r5, 0			/* don't preserve non-vol regs */
1727	bl	kvmppc_save_tm_hv
1728	nop
1729	ld	r9, HSTATE_KVM_VCPU(r13)
173091:
1731#endif
1732
1733	/* Increment yield count if they have a VPA */
1734	ld	r8, VCPU_VPA(r9)	/* do they have a VPA? */
1735	cmpdi	r8, 0
1736	beq	25f
1737	li	r4, LPPACA_YIELDCOUNT
1738	LWZX_BE	r3, r8, r4
1739	addi	r3, r3, 1
1740	STWX_BE	r3, r8, r4
1741	li	r3, 1
1742	stb	r3, VCPU_VPA_DIRTY(r9)
174325:
1744	/* Save PMU registers if requested */
1745	/* r8 and cr0.eq are live here */
1746	mr	r3, r9
1747	li	r4, 1
1748	beq	21f			/* if no VPA, save PMU stuff anyway */
1749	lbz	r4, LPPACA_PMCINUSE(r8)
175021:	bl	kvmhv_save_guest_pmu
1751	ld	r9, HSTATE_KVM_VCPU(r13)
1752
1753	/* Restore host values of some registers */
1754BEGIN_FTR_SECTION
1755	ld	r5, STACK_SLOT_CIABR(r1)
1756	ld	r6, STACK_SLOT_DAWR(r1)
1757	ld	r7, STACK_SLOT_DAWRX(r1)
1758	mtspr	SPRN_CIABR, r5
1759	/*
1760	 * If the DAWR doesn't work, it's ok to write these here as
1761	 * this value should always be zero
1762	*/
1763	mtspr	SPRN_DAWR, r6
1764	mtspr	SPRN_DAWRX, r7
1765END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1766BEGIN_FTR_SECTION
1767	ld	r5, STACK_SLOT_TID(r1)
1768	ld	r6, STACK_SLOT_PSSCR(r1)
1769	ld	r7, STACK_SLOT_PID(r1)
1770	mtspr	SPRN_TIDR, r5
1771	mtspr	SPRN_PSSCR, r6
1772	mtspr	SPRN_PID, r7
1773END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1774
1775#ifdef CONFIG_PPC_RADIX_MMU
1776	/*
1777	 * Are we running hash or radix ?
1778	 */
1779	ld	r5, VCPU_KVM(r9)
1780	lbz	r0, KVM_RADIX(r5)
1781	cmpwi	cr2, r0, 0
1782	beq	cr2, 2f
1783
1784	/*
1785	 * Radix: do eieio; tlbsync; ptesync sequence in case we
1786	 * interrupted the guest between a tlbie and a ptesync.
1787	 */
1788	eieio
1789	tlbsync
1790	ptesync
1791
1792	/* Radix: Handle the case where the guest used an illegal PID */
1793	LOAD_REG_ADDR(r4, mmu_base_pid)
1794	lwz	r3, VCPU_GUEST_PID(r9)
1795	lwz	r5, 0(r4)
1796	cmpw	cr0,r3,r5
1797	blt	2f
1798
1799	/*
1800	 * Illegal PID, the HW might have prefetched and cached in the TLB
1801	 * some translations for the  LPID 0 / guest PID combination which
1802	 * Linux doesn't know about, so we need to flush that PID out of
1803	 * the TLB. First we need to set LPIDR to 0 so tlbiel applies to
1804	 * the right context.
1805	*/
1806	li	r0,0
1807	mtspr	SPRN_LPID,r0
1808	isync
1809
1810	/* Then do a congruence class local flush */
1811	ld	r6,VCPU_KVM(r9)
1812	lwz	r0,KVM_TLB_SETS(r6)
1813	mtctr	r0
1814	li	r7,0x400		/* IS field = 0b01 */
1815	ptesync
1816	sldi	r0,r3,32		/* RS has PID */
18171:	PPC_TLBIEL(7,0,2,1,1)		/* RIC=2, PRS=1, R=1 */
1818	addi	r7,r7,0x1000
1819	bdnz	1b
1820	ptesync
1821
18222:
1823#endif /* CONFIG_PPC_RADIX_MMU */
1824
1825	/*
1826	 * POWER7/POWER8 guest -> host partition switch code.
1827	 * We don't have to lock against tlbies but we do
1828	 * have to coordinate the hardware threads.
1829	 * Here STACK_SLOT_TRAP(r1) contains the trap number.
1830	 */
1831kvmhv_switch_to_host:
1832	/* Secondary threads wait for primary to do partition switch */
1833	ld	r5,HSTATE_KVM_VCORE(r13)
1834	ld	r4,VCORE_KVM(r5)	/* pointer to struct kvm */
1835	lbz	r3,HSTATE_PTID(r13)
1836	cmpwi	r3,0
1837	beq	15f
1838	HMT_LOW
183913:	lbz	r3,VCORE_IN_GUEST(r5)
1840	cmpwi	r3,0
1841	bne	13b
1842	HMT_MEDIUM
1843	b	16f
1844
1845	/* Primary thread waits for all the secondaries to exit guest */
184615:	lwz	r3,VCORE_ENTRY_EXIT(r5)
1847	rlwinm	r0,r3,32-8,0xff
1848	clrldi	r3,r3,56
1849	cmpw	r3,r0
1850	bne	15b
1851	isync
1852
1853	/* Did we actually switch to the guest at all? */
1854	lbz	r6, VCORE_IN_GUEST(r5)
1855	cmpwi	r6, 0
1856	beq	19f
1857
1858	/* Primary thread switches back to host partition */
1859	lwz	r7,KVM_HOST_LPID(r4)
1860BEGIN_FTR_SECTION
1861	ld	r6,KVM_HOST_SDR1(r4)
1862	li	r8,LPID_RSVD		/* switch to reserved LPID */
1863	mtspr	SPRN_LPID,r8
1864	ptesync
1865	mtspr	SPRN_SDR1,r6		/* switch to host page table */
1866END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1867	mtspr	SPRN_LPID,r7
1868	isync
1869
1870BEGIN_FTR_SECTION
1871	/* DPDES and VTB are shared between threads */
1872	mfspr	r7, SPRN_DPDES
1873	mfspr	r8, SPRN_VTB
1874	std	r7, VCORE_DPDES(r5)
1875	std	r8, VCORE_VTB(r5)
1876	/* clear DPDES so we don't get guest doorbells in the host */
1877	li	r8, 0
1878	mtspr	SPRN_DPDES, r8
1879END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1880
1881	/* Subtract timebase offset from timebase */
1882	ld	r8, VCORE_TB_OFFSET_APPL(r5)
1883	cmpdi	r8,0
1884	beq	17f
1885	li	r0, 0
1886	std	r0, VCORE_TB_OFFSET_APPL(r5)
1887	mftb	r6			/* current guest timebase */
1888	subf	r8,r8,r6
1889	mtspr	SPRN_TBU40,r8		/* update upper 40 bits */
1890	mftb	r7			/* check if lower 24 bits overflowed */
1891	clrldi	r6,r6,40
1892	clrldi	r7,r7,40
1893	cmpld	r7,r6
1894	bge	17f
1895	addis	r8,r8,0x100		/* if so, increment upper 40 bits */
1896	mtspr	SPRN_TBU40,r8
1897
189817:
1899	/*
1900	 * If this is an HMI, we called kvmppc_realmode_hmi_handler
1901	 * above, which may or may not have already called
1902	 * kvmppc_subcore_exit_guest.  Fortunately, all that
1903	 * kvmppc_subcore_exit_guest does is clear a flag, so calling
1904	 * it again here is benign even if kvmppc_realmode_hmi_handler
1905	 * has already called it.
1906	 */
1907	bl	kvmppc_subcore_exit_guest
1908	nop
190930:	ld	r5,HSTATE_KVM_VCORE(r13)
1910	ld	r4,VCORE_KVM(r5)	/* pointer to struct kvm */
1911
1912	/* Reset PCR */
1913	ld	r0, VCORE_PCR(r5)
1914	cmpdi	r0, 0
1915	beq	18f
1916	li	r0, 0
1917	mtspr	SPRN_PCR, r0
191818:
1919	/* Signal secondary CPUs to continue */
1920	stb	r0,VCORE_IN_GUEST(r5)
192119:	lis	r8,0x7fff		/* MAX_INT@h */
1922	mtspr	SPRN_HDEC,r8
1923
192416:
1925BEGIN_FTR_SECTION
1926	/* On POWER9 with HPT-on-radix we need to wait for all other threads */
1927	ld	r3, HSTATE_SPLIT_MODE(r13)
1928	cmpdi	r3, 0
1929	beq	47f
1930	lwz	r8, KVM_SPLIT_DO_RESTORE(r3)
1931	cmpwi	r8, 0
1932	beq	47f
1933	bl	kvmhv_p9_restore_lpcr
1934	nop
1935	b	48f
193647:
1937END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1938	ld	r8,KVM_HOST_LPCR(r4)
1939	mtspr	SPRN_LPCR,r8
1940	isync
194148:
1942#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1943	/* Finish timing, if we have a vcpu */
1944	ld	r4, HSTATE_KVM_VCPU(r13)
1945	cmpdi	r4, 0
1946	li	r3, 0
1947	beq	2f
1948	bl	kvmhv_accumulate_time
19492:
1950#endif
1951	/* Unset guest mode */
1952	li	r0, KVM_GUEST_MODE_NONE
1953	stb	r0, HSTATE_IN_GUEST(r13)
1954
1955	lwz	r12, STACK_SLOT_TRAP(r1)	/* return trap # in r12 */
1956	ld	r0, SFS+PPC_LR_STKOFF(r1)
1957	addi	r1, r1, SFS
1958	mtlr	r0
1959	blr
1960
1961kvmppc_guest_external:
1962	/* External interrupt, first check for host_ipi. If this is
1963	 * set, we know the host wants us out so let's do it now
1964	 */
1965	bl	kvmppc_read_intr
1966
1967	/*
1968	 * Restore the active volatile registers after returning from
1969	 * a C function.
1970	 */
1971	ld	r9, HSTATE_KVM_VCPU(r13)
1972	li	r12, BOOK3S_INTERRUPT_EXTERNAL
1973
1974	/*
1975	 * kvmppc_read_intr return codes:
1976	 *
1977	 * Exit to host (r3 > 0)
1978	 *   1 An interrupt is pending that needs to be handled by the host
1979	 *     Exit guest and return to host by branching to guest_exit_cont
1980	 *
1981	 *   2 Passthrough that needs completion in the host
1982	 *     Exit guest and return to host by branching to guest_exit_cont
1983	 *     However, we also set r12 to BOOK3S_INTERRUPT_HV_RM_HARD
1984	 *     to indicate to the host to complete handling the interrupt
1985	 *
1986	 * Before returning to guest, we check if any CPU is heading out
1987	 * to the host and if so, we head out also. If no CPUs are heading
1988	 * check return values <= 0.
1989	 *
1990	 * Return to guest (r3 <= 0)
1991	 *  0 No external interrupt is pending
1992	 * -1 A guest wakeup IPI (which has now been cleared)
1993	 *    In either case, we return to guest to deliver any pending
1994	 *    guest interrupts.
1995	 *
1996	 * -2 A PCI passthrough external interrupt was handled
1997	 *    (interrupt was delivered directly to guest)
1998	 *    Return to guest to deliver any pending guest interrupts.
1999	 */
2000
2001	cmpdi	r3, 1
2002	ble	1f
2003
2004	/* Return code = 2 */
2005	li	r12, BOOK3S_INTERRUPT_HV_RM_HARD
2006	stw	r12, VCPU_TRAP(r9)
2007	b	guest_exit_cont
2008
20091:	/* Return code <= 1 */
2010	cmpdi	r3, 0
2011	bgt	guest_exit_cont
2012
2013	/* Return code <= 0 */
2014maybe_reenter_guest:
2015	ld	r5, HSTATE_KVM_VCORE(r13)
2016	lwz	r0, VCORE_ENTRY_EXIT(r5)
2017	cmpwi	r0, 0x100
2018	mr	r4, r9
2019	blt	deliver_guest_interrupt
2020	b	guest_exit_cont
2021
2022#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2023/*
2024 * Softpatch interrupt for transactional memory emulation cases
2025 * on POWER9 DD2.2.  This is early in the guest exit path - we
2026 * haven't saved registers or done a treclaim yet.
2027 */
2028kvmppc_tm_emul:
2029	/* Save instruction image in HEIR */
2030	mfspr	r3, SPRN_HEIR
2031	stw	r3, VCPU_HEIR(r9)
2032
2033	/*
2034	 * The cases we want to handle here are those where the guest
2035	 * is in real suspend mode and is trying to transition to
2036	 * transactional mode.
2037	 */
2038	lbz	r0, HSTATE_FAKE_SUSPEND(r13)
2039	cmpwi	r0, 0		/* keep exiting guest if in fake suspend */
2040	bne	guest_exit_cont
2041	rldicl	r3, r11, 64 - MSR_TS_S_LG, 62
2042	cmpwi	r3, 1		/* or if not in suspend state */
2043	bne	guest_exit_cont
2044
2045	/* Call C code to do the emulation */
2046	mr	r3, r9
2047	bl	kvmhv_p9_tm_emulation_early
2048	nop
2049	ld	r9, HSTATE_KVM_VCPU(r13)
2050	li	r12, BOOK3S_INTERRUPT_HV_SOFTPATCH
2051	cmpwi	r3, 0
2052	beq	guest_exit_cont		/* continue exiting if not handled */
2053	ld	r10, VCPU_PC(r9)
2054	ld	r11, VCPU_MSR(r9)
2055	b	fast_interrupt_c_return	/* go back to guest if handled */
2056#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
2057
2058/*
2059 * Check whether an HDSI is an HPTE not found fault or something else.
2060 * If it is an HPTE not found fault that is due to the guest accessing
2061 * a page that they have mapped but which we have paged out, then
2062 * we continue on with the guest exit path.  In all other cases,
2063 * reflect the HDSI to the guest as a DSI.
2064 */
2065kvmppc_hdsi:
2066	ld	r3, VCPU_KVM(r9)
2067	lbz	r0, KVM_RADIX(r3)
2068	mfspr	r4, SPRN_HDAR
2069	mfspr	r6, SPRN_HDSISR
2070BEGIN_FTR_SECTION
2071	/* Look for DSISR canary. If we find it, retry instruction */
2072	cmpdi	r6, 0x7fff
2073	beq	6f
2074END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2075	cmpwi	r0, 0
2076	bne	.Lradix_hdsi		/* on radix, just save DAR/DSISR/ASDR */
2077	/* HPTE not found fault or protection fault? */
2078	andis.	r0, r6, (DSISR_NOHPTE | DSISR_PROTFAULT)@h
2079	beq	1f			/* if not, send it to the guest */
2080	andi.	r0, r11, MSR_DR		/* data relocation enabled? */
2081	beq	3f
2082BEGIN_FTR_SECTION
2083	mfspr	r5, SPRN_ASDR		/* on POWER9, use ASDR to get VSID */
2084	b	4f
2085END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2086	clrrdi	r0, r4, 28
2087	PPC_SLBFEE_DOT(R5, R0)		/* if so, look up SLB */
2088	li	r0, BOOK3S_INTERRUPT_DATA_SEGMENT
2089	bne	7f			/* if no SLB entry found */
20904:	std	r4, VCPU_FAULT_DAR(r9)
2091	stw	r6, VCPU_FAULT_DSISR(r9)
2092
2093	/* Search the hash table. */
2094	mr	r3, r9			/* vcpu pointer */
2095	li	r7, 1			/* data fault */
2096	bl	kvmppc_hpte_hv_fault
2097	ld	r9, HSTATE_KVM_VCPU(r13)
2098	ld	r10, VCPU_PC(r9)
2099	ld	r11, VCPU_MSR(r9)
2100	li	r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
2101	cmpdi	r3, 0			/* retry the instruction */
2102	beq	6f
2103	cmpdi	r3, -1			/* handle in kernel mode */
2104	beq	guest_exit_cont
2105	cmpdi	r3, -2			/* MMIO emulation; need instr word */
2106	beq	2f
2107
2108	/* Synthesize a DSI (or DSegI) for the guest */
2109	ld	r4, VCPU_FAULT_DAR(r9)
2110	mr	r6, r3
21111:	li	r0, BOOK3S_INTERRUPT_DATA_STORAGE
2112	mtspr	SPRN_DSISR, r6
21137:	mtspr	SPRN_DAR, r4
2114	mtspr	SPRN_SRR0, r10
2115	mtspr	SPRN_SRR1, r11
2116	mr	r10, r0
2117	bl	kvmppc_msr_interrupt
2118fast_interrupt_c_return:
21196:	ld	r7, VCPU_CTR(r9)
2120	ld	r8, VCPU_XER(r9)
2121	mtctr	r7
2122	mtxer	r8
2123	mr	r4, r9
2124	b	fast_guest_return
2125
21263:	ld	r5, VCPU_KVM(r9)	/* not relocated, use VRMA */
2127	ld	r5, KVM_VRMA_SLB_V(r5)
2128	b	4b
2129
2130	/* If this is for emulated MMIO, load the instruction word */
21312:	li	r8, KVM_INST_FETCH_FAILED	/* In case lwz faults */
2132
2133	/* Set guest mode to 'jump over instruction' so if lwz faults
2134	 * we'll just continue at the next IP. */
2135	li	r0, KVM_GUEST_MODE_SKIP
2136	stb	r0, HSTATE_IN_GUEST(r13)
2137
2138	/* Do the access with MSR:DR enabled */
2139	mfmsr	r3
2140	ori	r4, r3, MSR_DR		/* Enable paging for data */
2141	mtmsrd	r4
2142	lwz	r8, 0(r10)
2143	mtmsrd	r3
2144
2145	/* Store the result */
2146	stw	r8, VCPU_LAST_INST(r9)
2147
2148	/* Unset guest mode. */
2149	li	r0, KVM_GUEST_MODE_HOST_HV
2150	stb	r0, HSTATE_IN_GUEST(r13)
2151	b	guest_exit_cont
2152
2153.Lradix_hdsi:
2154	std	r4, VCPU_FAULT_DAR(r9)
2155	stw	r6, VCPU_FAULT_DSISR(r9)
2156.Lradix_hisi:
2157	mfspr	r5, SPRN_ASDR
2158	std	r5, VCPU_FAULT_GPA(r9)
2159	b	guest_exit_cont
2160
2161/*
2162 * Similarly for an HISI, reflect it to the guest as an ISI unless
2163 * it is an HPTE not found fault for a page that we have paged out.
2164 */
2165kvmppc_hisi:
2166	ld	r3, VCPU_KVM(r9)
2167	lbz	r0, KVM_RADIX(r3)
2168	cmpwi	r0, 0
2169	bne	.Lradix_hisi		/* for radix, just save ASDR */
2170	andis.	r0, r11, SRR1_ISI_NOPT@h
2171	beq	1f
2172	andi.	r0, r11, MSR_IR		/* instruction relocation enabled? */
2173	beq	3f
2174BEGIN_FTR_SECTION
2175	mfspr	r5, SPRN_ASDR		/* on POWER9, use ASDR to get VSID */
2176	b	4f
2177END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2178	clrrdi	r0, r10, 28
2179	PPC_SLBFEE_DOT(R5, R0)		/* if so, look up SLB */
2180	li	r0, BOOK3S_INTERRUPT_INST_SEGMENT
2181	bne	7f			/* if no SLB entry found */
21824:
2183	/* Search the hash table. */
2184	mr	r3, r9			/* vcpu pointer */
2185	mr	r4, r10
2186	mr	r6, r11
2187	li	r7, 0			/* instruction fault */
2188	bl	kvmppc_hpte_hv_fault
2189	ld	r9, HSTATE_KVM_VCPU(r13)
2190	ld	r10, VCPU_PC(r9)
2191	ld	r11, VCPU_MSR(r9)
2192	li	r12, BOOK3S_INTERRUPT_H_INST_STORAGE
2193	cmpdi	r3, 0			/* retry the instruction */
2194	beq	fast_interrupt_c_return
2195	cmpdi	r3, -1			/* handle in kernel mode */
2196	beq	guest_exit_cont
2197
2198	/* Synthesize an ISI (or ISegI) for the guest */
2199	mr	r11, r3
22001:	li	r0, BOOK3S_INTERRUPT_INST_STORAGE
22017:	mtspr	SPRN_SRR0, r10
2202	mtspr	SPRN_SRR1, r11
2203	mr	r10, r0
2204	bl	kvmppc_msr_interrupt
2205	b	fast_interrupt_c_return
2206
22073:	ld	r6, VCPU_KVM(r9)	/* not relocated, use VRMA */
2208	ld	r5, KVM_VRMA_SLB_V(r6)
2209	b	4b
2210
2211/*
2212 * Try to handle an hcall in real mode.
2213 * Returns to the guest if we handle it, or continues on up to
2214 * the kernel if we can't (i.e. if we don't have a handler for
2215 * it, or if the handler returns H_TOO_HARD).
2216 *
2217 * r5 - r8 contain hcall args,
2218 * r9 = vcpu, r10 = pc, r11 = msr, r12 = trap, r13 = paca
2219 */
2220hcall_try_real_mode:
2221	ld	r3,VCPU_GPR(R3)(r9)
2222	andi.	r0,r11,MSR_PR
2223	/* sc 1 from userspace - reflect to guest syscall */
2224	bne	sc_1_fast_return
2225	/* sc 1 from nested guest - give it to L1 to handle */
2226	ld	r0, VCPU_NESTED(r9)
2227	cmpdi	r0, 0
2228	bne	guest_exit_cont
2229	clrrdi	r3,r3,2
2230	cmpldi	r3,hcall_real_table_end - hcall_real_table
2231	bge	guest_exit_cont
2232	/* See if this hcall is enabled for in-kernel handling */
2233	ld	r4, VCPU_KVM(r9)
2234	srdi	r0, r3, 8	/* r0 = (r3 / 4) >> 6 */
2235	sldi	r0, r0, 3	/* index into kvm->arch.enabled_hcalls[] */
2236	add	r4, r4, r0
2237	ld	r0, KVM_ENABLED_HCALLS(r4)
2238	rlwinm	r4, r3, 32-2, 0x3f	/* r4 = (r3 / 4) & 0x3f */
2239	srd	r0, r0, r4
2240	andi.	r0, r0, 1
2241	beq	guest_exit_cont
2242	/* Get pointer to handler, if any, and call it */
2243	LOAD_REG_ADDR(r4, hcall_real_table)
2244	lwax	r3,r3,r4
2245	cmpwi	r3,0
2246	beq	guest_exit_cont
2247	add	r12,r3,r4
2248	mtctr	r12
2249	mr	r3,r9		/* get vcpu pointer */
2250	ld	r4,VCPU_GPR(R4)(r9)
2251	bctrl
2252	cmpdi	r3,H_TOO_HARD
2253	beq	hcall_real_fallback
2254	ld	r4,HSTATE_KVM_VCPU(r13)
2255	std	r3,VCPU_GPR(R3)(r4)
2256	ld	r10,VCPU_PC(r4)
2257	ld	r11,VCPU_MSR(r4)
2258	b	fast_guest_return
2259
2260sc_1_fast_return:
2261	mtspr	SPRN_SRR0,r10
2262	mtspr	SPRN_SRR1,r11
2263	li	r10, BOOK3S_INTERRUPT_SYSCALL
2264	bl	kvmppc_msr_interrupt
2265	mr	r4,r9
2266	b	fast_guest_return
2267
2268	/* We've attempted a real mode hcall, but it's punted it back
2269	 * to userspace.  We need to restore some clobbered volatiles
2270	 * before resuming the pass-it-to-qemu path */
2271hcall_real_fallback:
2272	li	r12,BOOK3S_INTERRUPT_SYSCALL
2273	ld	r9, HSTATE_KVM_VCPU(r13)
2274
2275	b	guest_exit_cont
2276
2277	.globl	hcall_real_table
2278hcall_real_table:
2279	.long	0		/* 0 - unused */
2280	.long	DOTSYM(kvmppc_h_remove) - hcall_real_table
2281	.long	DOTSYM(kvmppc_h_enter) - hcall_real_table
2282	.long	DOTSYM(kvmppc_h_read) - hcall_real_table
2283	.long	DOTSYM(kvmppc_h_clear_mod) - hcall_real_table
2284	.long	DOTSYM(kvmppc_h_clear_ref) - hcall_real_table
2285	.long	DOTSYM(kvmppc_h_protect) - hcall_real_table
2286#ifdef CONFIG_SPAPR_TCE_IOMMU
2287	.long	DOTSYM(kvmppc_h_get_tce) - hcall_real_table
2288	.long	DOTSYM(kvmppc_rm_h_put_tce) - hcall_real_table
2289#else
2290	.long	0		/* 0x1c */
2291	.long	0		/* 0x20 */
2292#endif
2293	.long	0		/* 0x24 - H_SET_SPRG0 */
2294	.long	DOTSYM(kvmppc_h_set_dabr) - hcall_real_table
2295	.long	0		/* 0x2c */
2296	.long	0		/* 0x30 */
2297	.long	0		/* 0x34 */
2298	.long	0		/* 0x38 */
2299	.long	0		/* 0x3c */
2300	.long	0		/* 0x40 */
2301	.long	0		/* 0x44 */
2302	.long	0		/* 0x48 */
2303	.long	0		/* 0x4c */
2304	.long	0		/* 0x50 */
2305	.long	0		/* 0x54 */
2306	.long	0		/* 0x58 */
2307	.long	0		/* 0x5c */
2308	.long	0		/* 0x60 */
2309#ifdef CONFIG_KVM_XICS
2310	.long	DOTSYM(kvmppc_rm_h_eoi) - hcall_real_table
2311	.long	DOTSYM(kvmppc_rm_h_cppr) - hcall_real_table
2312	.long	DOTSYM(kvmppc_rm_h_ipi) - hcall_real_table
2313	.long	DOTSYM(kvmppc_rm_h_ipoll) - hcall_real_table
2314	.long	DOTSYM(kvmppc_rm_h_xirr) - hcall_real_table
2315#else
2316	.long	0		/* 0x64 - H_EOI */
2317	.long	0		/* 0x68 - H_CPPR */
2318	.long	0		/* 0x6c - H_IPI */
2319	.long	0		/* 0x70 - H_IPOLL */
2320	.long	0		/* 0x74 - H_XIRR */
2321#endif
2322	.long	0		/* 0x78 */
2323	.long	0		/* 0x7c */
2324	.long	0		/* 0x80 */
2325	.long	0		/* 0x84 */
2326	.long	0		/* 0x88 */
2327	.long	0		/* 0x8c */
2328	.long	0		/* 0x90 */
2329	.long	0		/* 0x94 */
2330	.long	0		/* 0x98 */
2331	.long	0		/* 0x9c */
2332	.long	0		/* 0xa0 */
2333	.long	0		/* 0xa4 */
2334	.long	0		/* 0xa8 */
2335	.long	0		/* 0xac */
2336	.long	0		/* 0xb0 */
2337	.long	0		/* 0xb4 */
2338	.long	0		/* 0xb8 */
2339	.long	0		/* 0xbc */
2340	.long	0		/* 0xc0 */
2341	.long	0		/* 0xc4 */
2342	.long	0		/* 0xc8 */
2343	.long	0		/* 0xcc */
2344	.long	0		/* 0xd0 */
2345	.long	0		/* 0xd4 */
2346	.long	0		/* 0xd8 */
2347	.long	0		/* 0xdc */
2348	.long	DOTSYM(kvmppc_h_cede) - hcall_real_table
2349	.long	DOTSYM(kvmppc_rm_h_confer) - hcall_real_table
2350	.long	0		/* 0xe8 */
2351	.long	0		/* 0xec */
2352	.long	0		/* 0xf0 */
2353	.long	0		/* 0xf4 */
2354	.long	0		/* 0xf8 */
2355	.long	0		/* 0xfc */
2356	.long	0		/* 0x100 */
2357	.long	0		/* 0x104 */
2358	.long	0		/* 0x108 */
2359	.long	0		/* 0x10c */
2360	.long	0		/* 0x110 */
2361	.long	0		/* 0x114 */
2362	.long	0		/* 0x118 */
2363	.long	0		/* 0x11c */
2364	.long	0		/* 0x120 */
2365	.long	DOTSYM(kvmppc_h_bulk_remove) - hcall_real_table
2366	.long	0		/* 0x128 */
2367	.long	0		/* 0x12c */
2368	.long	0		/* 0x130 */
2369	.long	DOTSYM(kvmppc_h_set_xdabr) - hcall_real_table
2370#ifdef CONFIG_SPAPR_TCE_IOMMU
2371	.long	DOTSYM(kvmppc_rm_h_stuff_tce) - hcall_real_table
2372	.long	DOTSYM(kvmppc_rm_h_put_tce_indirect) - hcall_real_table
2373#else
2374	.long	0		/* 0x138 */
2375	.long	0		/* 0x13c */
2376#endif
2377	.long	0		/* 0x140 */
2378	.long	0		/* 0x144 */
2379	.long	0		/* 0x148 */
2380	.long	0		/* 0x14c */
2381	.long	0		/* 0x150 */
2382	.long	0		/* 0x154 */
2383	.long	0		/* 0x158 */
2384	.long	0		/* 0x15c */
2385	.long	0		/* 0x160 */
2386	.long	0		/* 0x164 */
2387	.long	0		/* 0x168 */
2388	.long	0		/* 0x16c */
2389	.long	0		/* 0x170 */
2390	.long	0		/* 0x174 */
2391	.long	0		/* 0x178 */
2392	.long	0		/* 0x17c */
2393	.long	0		/* 0x180 */
2394	.long	0		/* 0x184 */
2395	.long	0		/* 0x188 */
2396	.long	0		/* 0x18c */
2397	.long	0		/* 0x190 */
2398	.long	0		/* 0x194 */
2399	.long	0		/* 0x198 */
2400	.long	0		/* 0x19c */
2401	.long	0		/* 0x1a0 */
2402	.long	0		/* 0x1a4 */
2403	.long	0		/* 0x1a8 */
2404	.long	0		/* 0x1ac */
2405	.long	0		/* 0x1b0 */
2406	.long	0		/* 0x1b4 */
2407	.long	0		/* 0x1b8 */
2408	.long	0		/* 0x1bc */
2409	.long	0		/* 0x1c0 */
2410	.long	0		/* 0x1c4 */
2411	.long	0		/* 0x1c8 */
2412	.long	0		/* 0x1cc */
2413	.long	0		/* 0x1d0 */
2414	.long	0		/* 0x1d4 */
2415	.long	0		/* 0x1d8 */
2416	.long	0		/* 0x1dc */
2417	.long	0		/* 0x1e0 */
2418	.long	0		/* 0x1e4 */
2419	.long	0		/* 0x1e8 */
2420	.long	0		/* 0x1ec */
2421	.long	0		/* 0x1f0 */
2422	.long	0		/* 0x1f4 */
2423	.long	0		/* 0x1f8 */
2424	.long	0		/* 0x1fc */
2425	.long	0		/* 0x200 */
2426	.long	0		/* 0x204 */
2427	.long	0		/* 0x208 */
2428	.long	0		/* 0x20c */
2429	.long	0		/* 0x210 */
2430	.long	0		/* 0x214 */
2431	.long	0		/* 0x218 */
2432	.long	0		/* 0x21c */
2433	.long	0		/* 0x220 */
2434	.long	0		/* 0x224 */
2435	.long	0		/* 0x228 */
2436	.long	0		/* 0x22c */
2437	.long	0		/* 0x230 */
2438	.long	0		/* 0x234 */
2439	.long	0		/* 0x238 */
2440	.long	0		/* 0x23c */
2441	.long	0		/* 0x240 */
2442	.long	0		/* 0x244 */
2443	.long	0		/* 0x248 */
2444	.long	0		/* 0x24c */
2445	.long	0		/* 0x250 */
2446	.long	0		/* 0x254 */
2447	.long	0		/* 0x258 */
2448	.long	0		/* 0x25c */
2449	.long	0		/* 0x260 */
2450	.long	0		/* 0x264 */
2451	.long	0		/* 0x268 */
2452	.long	0		/* 0x26c */
2453	.long	0		/* 0x270 */
2454	.long	0		/* 0x274 */
2455	.long	0		/* 0x278 */
2456	.long	0		/* 0x27c */
2457	.long	0		/* 0x280 */
2458	.long	0		/* 0x284 */
2459	.long	0		/* 0x288 */
2460	.long	0		/* 0x28c */
2461	.long	0		/* 0x290 */
2462	.long	0		/* 0x294 */
2463	.long	0		/* 0x298 */
2464	.long	0		/* 0x29c */
2465	.long	0		/* 0x2a0 */
2466	.long	0		/* 0x2a4 */
2467	.long	0		/* 0x2a8 */
2468	.long	0		/* 0x2ac */
2469	.long	0		/* 0x2b0 */
2470	.long	0		/* 0x2b4 */
2471	.long	0		/* 0x2b8 */
2472	.long	0		/* 0x2bc */
2473	.long	0		/* 0x2c0 */
2474	.long	0		/* 0x2c4 */
2475	.long	0		/* 0x2c8 */
2476	.long	0		/* 0x2cc */
2477	.long	0		/* 0x2d0 */
2478	.long	0		/* 0x2d4 */
2479	.long	0		/* 0x2d8 */
2480	.long	0		/* 0x2dc */
2481	.long	0		/* 0x2e0 */
2482	.long	0		/* 0x2e4 */
2483	.long	0		/* 0x2e8 */
2484	.long	0		/* 0x2ec */
2485	.long	0		/* 0x2f0 */
2486	.long	0		/* 0x2f4 */
2487	.long	0		/* 0x2f8 */
2488#ifdef CONFIG_KVM_XICS
2489	.long	DOTSYM(kvmppc_rm_h_xirr_x) - hcall_real_table
2490#else
2491	.long	0		/* 0x2fc - H_XIRR_X*/
2492#endif
2493	.long	DOTSYM(kvmppc_h_random) - hcall_real_table
2494	.globl	hcall_real_table_end
2495hcall_real_table_end:
2496
2497_GLOBAL(kvmppc_h_set_xdabr)
2498EXPORT_SYMBOL_GPL(kvmppc_h_set_xdabr)
2499	andi.	r0, r5, DABRX_USER | DABRX_KERNEL
2500	beq	6f
2501	li	r0, DABRX_USER | DABRX_KERNEL | DABRX_BTI
2502	andc.	r0, r5, r0
2503	beq	3f
25046:	li	r3, H_PARAMETER
2505	blr
2506
2507_GLOBAL(kvmppc_h_set_dabr)
2508EXPORT_SYMBOL_GPL(kvmppc_h_set_dabr)
2509	li	r5, DABRX_USER | DABRX_KERNEL
25103:
2511BEGIN_FTR_SECTION
2512	b	2f
2513END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2514	std	r4,VCPU_DABR(r3)
2515	stw	r5, VCPU_DABRX(r3)
2516	mtspr	SPRN_DABRX, r5
2517	/* Work around P7 bug where DABR can get corrupted on mtspr */
25181:	mtspr	SPRN_DABR,r4
2519	mfspr	r5, SPRN_DABR
2520	cmpd	r4, r5
2521	bne	1b
2522	isync
2523	li	r3,0
2524	blr
2525
25262:
2527	LOAD_REG_ADDR(r11, dawr_force_enable)
2528	lbz	r11, 0(r11)
2529	cmpdi	r11, 0
2530	li	r3, H_HARDWARE
2531	beqlr
2532	/* Emulate H_SET_DABR/X on P8 for the sake of compat mode guests */
2533	rlwimi	r5, r4, 5, DAWRX_DR | DAWRX_DW
2534	rlwimi	r5, r4, 2, DAWRX_WT
2535	clrrdi	r4, r4, 3
2536	std	r4, VCPU_DAWR(r3)
2537	std	r5, VCPU_DAWRX(r3)
2538	mtspr	SPRN_DAWR, r4
2539	mtspr	SPRN_DAWRX, r5
2540	li	r3, 0
2541	blr
2542
2543_GLOBAL(kvmppc_h_cede)		/* r3 = vcpu pointer, r11 = msr, r13 = paca */
2544	ori	r11,r11,MSR_EE
2545	std	r11,VCPU_MSR(r3)
2546	li	r0,1
2547	stb	r0,VCPU_CEDED(r3)
2548	sync			/* order setting ceded vs. testing prodded */
2549	lbz	r5,VCPU_PRODDED(r3)
2550	cmpwi	r5,0
2551	bne	kvm_cede_prodded
2552	li	r12,0		/* set trap to 0 to say hcall is handled */
2553	stw	r12,VCPU_TRAP(r3)
2554	li	r0,H_SUCCESS
2555	std	r0,VCPU_GPR(R3)(r3)
2556
2557	/*
2558	 * Set our bit in the bitmask of napping threads unless all the
2559	 * other threads are already napping, in which case we send this
2560	 * up to the host.
2561	 */
2562	ld	r5,HSTATE_KVM_VCORE(r13)
2563	lbz	r6,HSTATE_PTID(r13)
2564	lwz	r8,VCORE_ENTRY_EXIT(r5)
2565	clrldi	r8,r8,56
2566	li	r0,1
2567	sld	r0,r0,r6
2568	addi	r6,r5,VCORE_NAPPING_THREADS
256931:	lwarx	r4,0,r6
2570	or	r4,r4,r0
2571	cmpw	r4,r8
2572	beq	kvm_cede_exit
2573	stwcx.	r4,0,r6
2574	bne	31b
2575	/* order napping_threads update vs testing entry_exit_map */
2576	isync
2577	li	r0,NAPPING_CEDE
2578	stb	r0,HSTATE_NAPPING(r13)
2579	lwz	r7,VCORE_ENTRY_EXIT(r5)
2580	cmpwi	r7,0x100
2581	bge	33f		/* another thread already exiting */
2582
2583/*
2584 * Although not specifically required by the architecture, POWER7
2585 * preserves the following registers in nap mode, even if an SMT mode
2586 * switch occurs: SLB entries, PURR, SPURR, AMOR, UAMOR, AMR, SPRG0-3,
2587 * DAR, DSISR, DABR, DABRX, DSCR, PMCx, MMCRx, SIAR, SDAR.
2588 */
2589	/* Save non-volatile GPRs */
2590	std	r14, VCPU_GPR(R14)(r3)
2591	std	r15, VCPU_GPR(R15)(r3)
2592	std	r16, VCPU_GPR(R16)(r3)
2593	std	r17, VCPU_GPR(R17)(r3)
2594	std	r18, VCPU_GPR(R18)(r3)
2595	std	r19, VCPU_GPR(R19)(r3)
2596	std	r20, VCPU_GPR(R20)(r3)
2597	std	r21, VCPU_GPR(R21)(r3)
2598	std	r22, VCPU_GPR(R22)(r3)
2599	std	r23, VCPU_GPR(R23)(r3)
2600	std	r24, VCPU_GPR(R24)(r3)
2601	std	r25, VCPU_GPR(R25)(r3)
2602	std	r26, VCPU_GPR(R26)(r3)
2603	std	r27, VCPU_GPR(R27)(r3)
2604	std	r28, VCPU_GPR(R28)(r3)
2605	std	r29, VCPU_GPR(R29)(r3)
2606	std	r30, VCPU_GPR(R30)(r3)
2607	std	r31, VCPU_GPR(R31)(r3)
2608
2609	/* save FP state */
2610	bl	kvmppc_save_fp
2611
2612#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2613/*
2614 * Branch around the call if both CPU_FTR_TM and
2615 * CPU_FTR_P9_TM_HV_ASSIST are off.
2616 */
2617BEGIN_FTR_SECTION
2618	b	91f
2619END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
2620	/*
2621	 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
2622	 */
2623	ld	r3, HSTATE_KVM_VCPU(r13)
2624	ld      r4, VCPU_MSR(r3)
2625	li	r5, 0			/* don't preserve non-vol regs */
2626	bl	kvmppc_save_tm_hv
2627	nop
262891:
2629#endif
2630
2631	/*
2632	 * Set DEC to the smaller of DEC and HDEC, so that we wake
2633	 * no later than the end of our timeslice (HDEC interrupts
2634	 * don't wake us from nap).
2635	 */
2636	mfspr	r3, SPRN_DEC
2637	mfspr	r4, SPRN_HDEC
2638	mftb	r5
2639BEGIN_FTR_SECTION
2640	/* On P9 check whether the guest has large decrementer mode enabled */
2641	ld	r6, HSTATE_KVM_VCORE(r13)
2642	ld	r6, VCORE_LPCR(r6)
2643	andis.	r6, r6, LPCR_LD@h
2644	bne	68f
2645END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2646	extsw	r3, r3
264768:	EXTEND_HDEC(r4)
2648	cmpd	r3, r4
2649	ble	67f
2650	mtspr	SPRN_DEC, r4
265167:
2652	/* save expiry time of guest decrementer */
2653	add	r3, r3, r5
2654	ld	r4, HSTATE_KVM_VCPU(r13)
2655	ld	r5, HSTATE_KVM_VCORE(r13)
2656	ld	r6, VCORE_TB_OFFSET_APPL(r5)
2657	subf	r3, r6, r3	/* convert to host TB value */
2658	std	r3, VCPU_DEC_EXPIRES(r4)
2659
2660#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2661	ld	r4, HSTATE_KVM_VCPU(r13)
2662	addi	r3, r4, VCPU_TB_CEDE
2663	bl	kvmhv_accumulate_time
2664#endif
2665
2666	lis	r3, LPCR_PECEDP@h	/* Do wake on privileged doorbell */
2667
2668	/* Go back to host stack */
2669	ld	r1, HSTATE_HOST_R1(r13)
2670
2671	/*
2672	 * Take a nap until a decrementer or external or doobell interrupt
2673	 * occurs, with PECE1 and PECE0 set in LPCR.
2674	 * On POWER8, set PECEDH, and if we are ceding, also set PECEDP.
2675	 * Also clear the runlatch bit before napping.
2676	 */
2677kvm_do_nap:
2678	mfspr	r0, SPRN_CTRLF
2679	clrrdi	r0, r0, 1
2680	mtspr	SPRN_CTRLT, r0
2681
2682	li	r0,1
2683	stb	r0,HSTATE_HWTHREAD_REQ(r13)
2684	mfspr	r5,SPRN_LPCR
2685	ori	r5,r5,LPCR_PECE0 | LPCR_PECE1
2686BEGIN_FTR_SECTION
2687	ori	r5, r5, LPCR_PECEDH
2688	rlwimi	r5, r3, 0, LPCR_PECEDP
2689END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2690
2691kvm_nap_sequence:		/* desired LPCR value in r5 */
2692BEGIN_FTR_SECTION
2693	/*
2694	 * PSSCR bits:	exit criterion = 1 (wakeup based on LPCR at sreset)
2695	 *		enable state loss = 1 (allow SMT mode switch)
2696	 *		requested level = 0 (just stop dispatching)
2697	 */
2698	lis	r3, (PSSCR_EC | PSSCR_ESL)@h
2699	/* Set LPCR_PECE_HVEE bit to enable wakeup by HV interrupts */
2700	li	r4, LPCR_PECE_HVEE@higher
2701	sldi	r4, r4, 32
2702	or	r5, r5, r4
2703FTR_SECTION_ELSE
2704	li	r3, PNV_THREAD_NAP
2705ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_300)
2706	mtspr	SPRN_LPCR,r5
2707	isync
2708
2709BEGIN_FTR_SECTION
2710	bl	isa300_idle_stop_mayloss
2711FTR_SECTION_ELSE
2712	bl	isa206_idle_insn_mayloss
2713ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_300)
2714
2715	mfspr	r0, SPRN_CTRLF
2716	ori	r0, r0, 1
2717	mtspr	SPRN_CTRLT, r0
2718
2719	mtspr	SPRN_SRR1, r3
2720
2721	li	r0, 0
2722	stb	r0, PACA_FTRACE_ENABLED(r13)
2723
2724	li	r0, KVM_HWTHREAD_IN_KVM
2725	stb	r0, HSTATE_HWTHREAD_STATE(r13)
2726
2727	lbz	r0, HSTATE_NAPPING(r13)
2728	cmpwi	r0, NAPPING_CEDE
2729	beq	kvm_end_cede
2730	cmpwi	r0, NAPPING_NOVCPU
2731	beq	kvm_novcpu_wakeup
2732	cmpwi	r0, NAPPING_UNSPLIT
2733	beq	kvm_unsplit_wakeup
2734	twi	31,0,0 /* Nap state must not be zero */
2735
273633:	mr	r4, r3
2737	li	r3, 0
2738	li	r12, 0
2739	b	34f
2740
2741kvm_end_cede:
2742	/* Woken by external or decrementer interrupt */
2743
2744	/* get vcpu pointer */
2745	ld	r4, HSTATE_KVM_VCPU(r13)
2746
2747#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2748	addi	r3, r4, VCPU_TB_RMINTR
2749	bl	kvmhv_accumulate_time
2750#endif
2751
2752#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2753/*
2754 * Branch around the call if both CPU_FTR_TM and
2755 * CPU_FTR_P9_TM_HV_ASSIST are off.
2756 */
2757BEGIN_FTR_SECTION
2758	b	91f
2759END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
2760	/*
2761	 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
2762	 */
2763	mr      r3, r4
2764	ld      r4, VCPU_MSR(r3)
2765	li	r5, 0			/* don't preserve non-vol regs */
2766	bl	kvmppc_restore_tm_hv
2767	nop
2768	ld	r4, HSTATE_KVM_VCPU(r13)
276991:
2770#endif
2771
2772	/* load up FP state */
2773	bl	kvmppc_load_fp
2774
2775	/* Restore guest decrementer */
2776	ld	r3, VCPU_DEC_EXPIRES(r4)
2777	ld	r5, HSTATE_KVM_VCORE(r13)
2778	ld	r6, VCORE_TB_OFFSET_APPL(r5)
2779	add	r3, r3, r6	/* convert host TB to guest TB value */
2780	mftb	r7
2781	subf	r3, r7, r3
2782	mtspr	SPRN_DEC, r3
2783
2784	/* Load NV GPRS */
2785	ld	r14, VCPU_GPR(R14)(r4)
2786	ld	r15, VCPU_GPR(R15)(r4)
2787	ld	r16, VCPU_GPR(R16)(r4)
2788	ld	r17, VCPU_GPR(R17)(r4)
2789	ld	r18, VCPU_GPR(R18)(r4)
2790	ld	r19, VCPU_GPR(R19)(r4)
2791	ld	r20, VCPU_GPR(R20)(r4)
2792	ld	r21, VCPU_GPR(R21)(r4)
2793	ld	r22, VCPU_GPR(R22)(r4)
2794	ld	r23, VCPU_GPR(R23)(r4)
2795	ld	r24, VCPU_GPR(R24)(r4)
2796	ld	r25, VCPU_GPR(R25)(r4)
2797	ld	r26, VCPU_GPR(R26)(r4)
2798	ld	r27, VCPU_GPR(R27)(r4)
2799	ld	r28, VCPU_GPR(R28)(r4)
2800	ld	r29, VCPU_GPR(R29)(r4)
2801	ld	r30, VCPU_GPR(R30)(r4)
2802	ld	r31, VCPU_GPR(R31)(r4)
2803
2804	/* Check the wake reason in SRR1 to see why we got here */
2805	bl	kvmppc_check_wake_reason
2806
2807	/*
2808	 * Restore volatile registers since we could have called a
2809	 * C routine in kvmppc_check_wake_reason
2810	 *	r4 = VCPU
2811	 * r3 tells us whether we need to return to host or not
2812	 * WARNING: it gets checked further down:
2813	 * should not modify r3 until this check is done.
2814	 */
2815	ld	r4, HSTATE_KVM_VCPU(r13)
2816
2817	/* clear our bit in vcore->napping_threads */
281834:	ld	r5,HSTATE_KVM_VCORE(r13)
2819	lbz	r7,HSTATE_PTID(r13)
2820	li	r0,1
2821	sld	r0,r0,r7
2822	addi	r6,r5,VCORE_NAPPING_THREADS
282332:	lwarx	r7,0,r6
2824	andc	r7,r7,r0
2825	stwcx.	r7,0,r6
2826	bne	32b
2827	li	r0,0
2828	stb	r0,HSTATE_NAPPING(r13)
2829
2830	/* See if the wake reason saved in r3 means we need to exit */
2831	stw	r12, VCPU_TRAP(r4)
2832	mr	r9, r4
2833	cmpdi	r3, 0
2834	bgt	guest_exit_cont
2835	b	maybe_reenter_guest
2836
2837	/* cede when already previously prodded case */
2838kvm_cede_prodded:
2839	li	r0,0
2840	stb	r0,VCPU_PRODDED(r3)
2841	sync			/* order testing prodded vs. clearing ceded */
2842	stb	r0,VCPU_CEDED(r3)
2843	li	r3,H_SUCCESS
2844	blr
2845
2846	/* we've ceded but we want to give control to the host */
2847kvm_cede_exit:
2848	ld	r9, HSTATE_KVM_VCPU(r13)
2849#ifdef CONFIG_KVM_XICS
2850	/* Abort if we still have a pending escalation */
2851	lbz	r5, VCPU_XIVE_ESC_ON(r9)
2852	cmpwi	r5, 0
2853	beq	1f
2854	li	r0, 0
2855	stb	r0, VCPU_CEDED(r9)
28561:	/* Enable XIVE escalation */
2857	li	r5, XIVE_ESB_SET_PQ_00
2858	mfmsr	r0
2859	andi.	r0, r0, MSR_DR		/* in real mode? */
2860	beq	1f
2861	ld	r10, VCPU_XIVE_ESC_VADDR(r9)
2862	cmpdi	r10, 0
2863	beq	3f
2864	ldx	r0, r10, r5
2865	b	2f
28661:	ld	r10, VCPU_XIVE_ESC_RADDR(r9)
2867	cmpdi	r10, 0
2868	beq	3f
2869	ldcix	r0, r10, r5
28702:	sync
2871	li	r0, 1
2872	stb	r0, VCPU_XIVE_ESC_ON(r9)
2873#endif /* CONFIG_KVM_XICS */
28743:	b	guest_exit_cont
2875
2876	/* Try to do machine check recovery in real mode */
2877machine_check_realmode:
2878	mr	r3, r9		/* get vcpu pointer */
2879	bl	kvmppc_realmode_machine_check
2880	nop
2881	/* all machine checks go to virtual mode for further handling */
2882	ld	r9, HSTATE_KVM_VCPU(r13)
2883	li	r12, BOOK3S_INTERRUPT_MACHINE_CHECK
2884	b	guest_exit_cont
2885
2886/*
2887 * Call C code to handle a HMI in real mode.
2888 * Only the primary thread does the call, secondary threads are handled
2889 * by calling hmi_exception_realmode() after kvmppc_hv_entry returns.
2890 * r9 points to the vcpu on entry
2891 */
2892hmi_realmode:
2893	lbz	r0, HSTATE_PTID(r13)
2894	cmpwi	r0, 0
2895	bne	guest_exit_cont
2896	bl	kvmppc_realmode_hmi_handler
2897	ld	r9, HSTATE_KVM_VCPU(r13)
2898	li	r12, BOOK3S_INTERRUPT_HMI
2899	b	guest_exit_cont
2900
2901/*
2902 * Check the reason we woke from nap, and take appropriate action.
2903 * Returns (in r3):
2904 *	0 if nothing needs to be done
2905 *	1 if something happened that needs to be handled by the host
2906 *	-1 if there was a guest wakeup (IPI or msgsnd)
2907 *	-2 if we handled a PCI passthrough interrupt (returned by
2908 *		kvmppc_read_intr only)
2909 *
2910 * Also sets r12 to the interrupt vector for any interrupt that needs
2911 * to be handled now by the host (0x500 for external interrupt), or zero.
2912 * Modifies all volatile registers (since it may call a C function).
2913 * This routine calls kvmppc_read_intr, a C function, if an external
2914 * interrupt is pending.
2915 */
2916kvmppc_check_wake_reason:
2917	mfspr	r6, SPRN_SRR1
2918BEGIN_FTR_SECTION
2919	rlwinm	r6, r6, 45-31, 0xf	/* extract wake reason field (P8) */
2920FTR_SECTION_ELSE
2921	rlwinm	r6, r6, 45-31, 0xe	/* P7 wake reason field is 3 bits */
2922ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_207S)
2923	cmpwi	r6, 8			/* was it an external interrupt? */
2924	beq	7f			/* if so, see what it was */
2925	li	r3, 0
2926	li	r12, 0
2927	cmpwi	r6, 6			/* was it the decrementer? */
2928	beq	0f
2929BEGIN_FTR_SECTION
2930	cmpwi	r6, 5			/* privileged doorbell? */
2931	beq	0f
2932	cmpwi	r6, 3			/* hypervisor doorbell? */
2933	beq	3f
2934END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2935	cmpwi	r6, 0xa			/* Hypervisor maintenance ? */
2936	beq	4f
2937	li	r3, 1			/* anything else, return 1 */
29380:	blr
2939
2940	/* hypervisor doorbell */
29413:	li	r12, BOOK3S_INTERRUPT_H_DOORBELL
2942
2943	/*
2944	 * Clear the doorbell as we will invoke the handler
2945	 * explicitly in the guest exit path.
2946	 */
2947	lis	r6, (PPC_DBELL_SERVER << (63-36))@h
2948	PPC_MSGCLR(6)
2949	/* see if it's a host IPI */
2950	li	r3, 1
2951BEGIN_FTR_SECTION
2952	PPC_MSGSYNC
2953	lwsync
2954END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2955	lbz	r0, HSTATE_HOST_IPI(r13)
2956	cmpwi	r0, 0
2957	bnelr
2958	/* if not, return -1 */
2959	li	r3, -1
2960	blr
2961
2962	/* Woken up due to Hypervisor maintenance interrupt */
29634:	li	r12, BOOK3S_INTERRUPT_HMI
2964	li	r3, 1
2965	blr
2966
2967	/* external interrupt - create a stack frame so we can call C */
29687:	mflr	r0
2969	std	r0, PPC_LR_STKOFF(r1)
2970	stdu	r1, -PPC_MIN_STKFRM(r1)
2971	bl	kvmppc_read_intr
2972	nop
2973	li	r12, BOOK3S_INTERRUPT_EXTERNAL
2974	cmpdi	r3, 1
2975	ble	1f
2976
2977	/*
2978	 * Return code of 2 means PCI passthrough interrupt, but
2979	 * we need to return back to host to complete handling the
2980	 * interrupt. Trap reason is expected in r12 by guest
2981	 * exit code.
2982	 */
2983	li	r12, BOOK3S_INTERRUPT_HV_RM_HARD
29841:
2985	ld	r0, PPC_MIN_STKFRM+PPC_LR_STKOFF(r1)
2986	addi	r1, r1, PPC_MIN_STKFRM
2987	mtlr	r0
2988	blr
2989
2990/*
2991 * Save away FP, VMX and VSX registers.
2992 * r3 = vcpu pointer
2993 * N.B. r30 and r31 are volatile across this function,
2994 * thus it is not callable from C.
2995 */
2996kvmppc_save_fp:
2997	mflr	r30
2998	mr	r31,r3
2999	mfmsr	r5
3000	ori	r8,r5,MSR_FP
3001#ifdef CONFIG_ALTIVEC
3002BEGIN_FTR_SECTION
3003	oris	r8,r8,MSR_VEC@h
3004END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3005#endif
3006#ifdef CONFIG_VSX
3007BEGIN_FTR_SECTION
3008	oris	r8,r8,MSR_VSX@h
3009END_FTR_SECTION_IFSET(CPU_FTR_VSX)
3010#endif
3011	mtmsrd	r8
3012	addi	r3,r3,VCPU_FPRS
3013	bl	store_fp_state
3014#ifdef CONFIG_ALTIVEC
3015BEGIN_FTR_SECTION
3016	addi	r3,r31,VCPU_VRS
3017	bl	store_vr_state
3018END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3019#endif
3020	mfspr	r6,SPRN_VRSAVE
3021	stw	r6,VCPU_VRSAVE(r31)
3022	mtlr	r30
3023	blr
3024
3025/*
3026 * Load up FP, VMX and VSX registers
3027 * r4 = vcpu pointer
3028 * N.B. r30 and r31 are volatile across this function,
3029 * thus it is not callable from C.
3030 */
3031kvmppc_load_fp:
3032	mflr	r30
3033	mr	r31,r4
3034	mfmsr	r9
3035	ori	r8,r9,MSR_FP
3036#ifdef CONFIG_ALTIVEC
3037BEGIN_FTR_SECTION
3038	oris	r8,r8,MSR_VEC@h
3039END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3040#endif
3041#ifdef CONFIG_VSX
3042BEGIN_FTR_SECTION
3043	oris	r8,r8,MSR_VSX@h
3044END_FTR_SECTION_IFSET(CPU_FTR_VSX)
3045#endif
3046	mtmsrd	r8
3047	addi	r3,r4,VCPU_FPRS
3048	bl	load_fp_state
3049#ifdef CONFIG_ALTIVEC
3050BEGIN_FTR_SECTION
3051	addi	r3,r31,VCPU_VRS
3052	bl	load_vr_state
3053END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3054#endif
3055	lwz	r7,VCPU_VRSAVE(r31)
3056	mtspr	SPRN_VRSAVE,r7
3057	mtlr	r30
3058	mr	r4,r31
3059	blr
3060
3061#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3062/*
3063 * Save transactional state and TM-related registers.
3064 * Called with r3 pointing to the vcpu struct and r4 containing
3065 * the guest MSR value.
3066 * r5 is non-zero iff non-volatile register state needs to be maintained.
3067 * If r5 == 0, this can modify all checkpointed registers, but
3068 * restores r1 and r2 before exit.
3069 */
3070_GLOBAL_TOC(kvmppc_save_tm_hv)
3071EXPORT_SYMBOL_GPL(kvmppc_save_tm_hv)
3072	/* See if we need to handle fake suspend mode */
3073BEGIN_FTR_SECTION
3074	b	__kvmppc_save_tm
3075END_FTR_SECTION_IFCLR(CPU_FTR_P9_TM_HV_ASSIST)
3076
3077	lbz	r0, HSTATE_FAKE_SUSPEND(r13) /* Were we fake suspended? */
3078	cmpwi	r0, 0
3079	beq	__kvmppc_save_tm
3080
3081	/* The following code handles the fake_suspend = 1 case */
3082	mflr	r0
3083	std	r0, PPC_LR_STKOFF(r1)
3084	stdu	r1, -PPC_MIN_STKFRM(r1)
3085
3086	/* Turn on TM. */
3087	mfmsr	r8
3088	li	r0, 1
3089	rldimi	r8, r0, MSR_TM_LG, 63-MSR_TM_LG
3090	mtmsrd	r8
3091
3092	rldicl. r8, r8, 64 - MSR_TS_S_LG, 62 /* Did we actually hrfid? */
3093	beq	4f
3094BEGIN_FTR_SECTION
3095	bl	pnv_power9_force_smt4_catch
3096END_FTR_SECTION_IFSET(CPU_FTR_P9_TM_XER_SO_BUG)
3097	nop
3098
3099	/* We have to treclaim here because that's the only way to do S->N */
3100	li	r3, TM_CAUSE_KVM_RESCHED
3101	TRECLAIM(R3)
3102
3103	/*
3104	 * We were in fake suspend, so we are not going to save the
3105	 * register state as the guest checkpointed state (since
3106	 * we already have it), therefore we can now use any volatile GPR.
3107	 * In fact treclaim in fake suspend state doesn't modify
3108	 * any registers.
3109	 */
3110
3111BEGIN_FTR_SECTION
3112	bl	pnv_power9_force_smt4_release
3113END_FTR_SECTION_IFSET(CPU_FTR_P9_TM_XER_SO_BUG)
3114	nop
3115
31164:
3117	mfspr	r3, SPRN_PSSCR
3118	/* PSSCR_FAKE_SUSPEND is a write-only bit, but clear it anyway */
3119	li	r0, PSSCR_FAKE_SUSPEND
3120	andc	r3, r3, r0
3121	mtspr	SPRN_PSSCR, r3
3122
3123	/* Don't save TEXASR, use value from last exit in real suspend state */
3124	ld	r9, HSTATE_KVM_VCPU(r13)
3125	mfspr	r5, SPRN_TFHAR
3126	mfspr	r6, SPRN_TFIAR
3127	std	r5, VCPU_TFHAR(r9)
3128	std	r6, VCPU_TFIAR(r9)
3129
3130	addi	r1, r1, PPC_MIN_STKFRM
3131	ld	r0, PPC_LR_STKOFF(r1)
3132	mtlr	r0
3133	blr
3134
3135/*
3136 * Restore transactional state and TM-related registers.
3137 * Called with r3 pointing to the vcpu struct
3138 * and r4 containing the guest MSR value.
3139 * r5 is non-zero iff non-volatile register state needs to be maintained.
3140 * This potentially modifies all checkpointed registers.
3141 * It restores r1 and r2 from the PACA.
3142 */
3143_GLOBAL_TOC(kvmppc_restore_tm_hv)
3144EXPORT_SYMBOL_GPL(kvmppc_restore_tm_hv)
3145	/*
3146	 * If we are doing TM emulation for the guest on a POWER9 DD2,
3147	 * then we don't actually do a trechkpt -- we either set up
3148	 * fake-suspend mode, or emulate a TM rollback.
3149	 */
3150BEGIN_FTR_SECTION
3151	b	__kvmppc_restore_tm
3152END_FTR_SECTION_IFCLR(CPU_FTR_P9_TM_HV_ASSIST)
3153	mflr	r0
3154	std	r0, PPC_LR_STKOFF(r1)
3155
3156	li	r0, 0
3157	stb	r0, HSTATE_FAKE_SUSPEND(r13)
3158
3159	/* Turn on TM so we can restore TM SPRs */
3160	mfmsr	r5
3161	li	r0, 1
3162	rldimi	r5, r0, MSR_TM_LG, 63-MSR_TM_LG
3163	mtmsrd	r5
3164
3165	/*
3166	 * The user may change these outside of a transaction, so they must
3167	 * always be context switched.
3168	 */
3169	ld	r5, VCPU_TFHAR(r3)
3170	ld	r6, VCPU_TFIAR(r3)
3171	ld	r7, VCPU_TEXASR(r3)
3172	mtspr	SPRN_TFHAR, r5
3173	mtspr	SPRN_TFIAR, r6
3174	mtspr	SPRN_TEXASR, r7
3175
3176	rldicl. r5, r4, 64 - MSR_TS_S_LG, 62
3177	beqlr		/* TM not active in guest */
3178
3179	/* Make sure the failure summary is set */
3180	oris	r7, r7, (TEXASR_FS)@h
3181	mtspr	SPRN_TEXASR, r7
3182
3183	cmpwi	r5, 1		/* check for suspended state */
3184	bgt	10f
3185	stb	r5, HSTATE_FAKE_SUSPEND(r13)
3186	b	9f		/* and return */
318710:	stdu	r1, -PPC_MIN_STKFRM(r1)
3188	/* guest is in transactional state, so simulate rollback */
3189	bl	kvmhv_emulate_tm_rollback
3190	nop
3191	addi	r1, r1, PPC_MIN_STKFRM
31929:	ld	r0, PPC_LR_STKOFF(r1)
3193	mtlr	r0
3194	blr
3195#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
3196
3197/*
3198 * We come here if we get any exception or interrupt while we are
3199 * executing host real mode code while in guest MMU context.
3200 * r12 is (CR << 32) | vector
3201 * r13 points to our PACA
3202 * r12 is saved in HSTATE_SCRATCH0(r13)
3203 * ctr is saved in HSTATE_SCRATCH1(r13) if RELOCATABLE
3204 * r9 is saved in HSTATE_SCRATCH2(r13)
3205 * r13 is saved in HSPRG1
3206 * cfar is saved in HSTATE_CFAR(r13)
3207 * ppr is saved in HSTATE_PPR(r13)
3208 */
3209kvmppc_bad_host_intr:
3210	/*
3211	 * Switch to the emergency stack, but start half-way down in
3212	 * case we were already on it.
3213	 */
3214	mr	r9, r1
3215	std	r1, PACAR1(r13)
3216	ld	r1, PACAEMERGSP(r13)
3217	subi	r1, r1, THREAD_SIZE/2 + INT_FRAME_SIZE
3218	std	r9, 0(r1)
3219	std	r0, GPR0(r1)
3220	std	r9, GPR1(r1)
3221	std	r2, GPR2(r1)
3222	SAVE_4GPRS(3, r1)
3223	SAVE_2GPRS(7, r1)
3224	srdi	r0, r12, 32
3225	clrldi	r12, r12, 32
3226	std	r0, _CCR(r1)
3227	std	r12, _TRAP(r1)
3228	andi.	r0, r12, 2
3229	beq	1f
3230	mfspr	r3, SPRN_HSRR0
3231	mfspr	r4, SPRN_HSRR1
3232	mfspr	r5, SPRN_HDAR
3233	mfspr	r6, SPRN_HDSISR
3234	b	2f
32351:	mfspr	r3, SPRN_SRR0
3236	mfspr	r4, SPRN_SRR1
3237	mfspr	r5, SPRN_DAR
3238	mfspr	r6, SPRN_DSISR
32392:	std	r3, _NIP(r1)
3240	std	r4, _MSR(r1)
3241	std	r5, _DAR(r1)
3242	std	r6, _DSISR(r1)
3243	ld	r9, HSTATE_SCRATCH2(r13)
3244	ld	r12, HSTATE_SCRATCH0(r13)
3245	GET_SCRATCH0(r0)
3246	SAVE_4GPRS(9, r1)
3247	std	r0, GPR13(r1)
3248	SAVE_NVGPRS(r1)
3249	ld	r5, HSTATE_CFAR(r13)
3250	std	r5, ORIG_GPR3(r1)
3251	mflr	r3
3252#ifdef CONFIG_RELOCATABLE
3253	ld	r4, HSTATE_SCRATCH1(r13)
3254#else
3255	mfctr	r4
3256#endif
3257	mfxer	r5
3258	lbz	r6, PACAIRQSOFTMASK(r13)
3259	std	r3, _LINK(r1)
3260	std	r4, _CTR(r1)
3261	std	r5, _XER(r1)
3262	std	r6, SOFTE(r1)
3263	ld	r2, PACATOC(r13)
3264	LOAD_REG_IMMEDIATE(3, 0x7265677368657265)
3265	std	r3, STACK_FRAME_OVERHEAD-16(r1)
3266
3267	/*
3268	 * On POWER9 do a minimal restore of the MMU and call C code,
3269	 * which will print a message and panic.
3270	 * XXX On POWER7 and POWER8, we just spin here since we don't
3271	 * know what the other threads are doing (and we don't want to
3272	 * coordinate with them) - but at least we now have register state
3273	 * in memory that we might be able to look at from another CPU.
3274	 */
3275BEGIN_FTR_SECTION
3276	b	.
3277END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
3278	ld	r9, HSTATE_KVM_VCPU(r13)
3279	ld	r10, VCPU_KVM(r9)
3280
3281	li	r0, 0
3282	mtspr	SPRN_AMR, r0
3283	mtspr	SPRN_IAMR, r0
3284	mtspr	SPRN_CIABR, r0
3285	mtspr	SPRN_DAWRX, r0
3286
3287BEGIN_MMU_FTR_SECTION
3288	b	4f
3289END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
3290
3291	slbmte	r0, r0
3292	slbia
3293	ptesync
3294	ld	r8, PACA_SLBSHADOWPTR(r13)
3295	.rept	SLB_NUM_BOLTED
3296	li	r3, SLBSHADOW_SAVEAREA
3297	LDX_BE	r5, r8, r3
3298	addi	r3, r3, 8
3299	LDX_BE	r6, r8, r3
3300	andis.	r7, r5, SLB_ESID_V@h
3301	beq	3f
3302	slbmte	r6, r5
33033:	addi	r8, r8, 16
3304	.endr
3305
33064:	lwz	r7, KVM_HOST_LPID(r10)
3307	mtspr	SPRN_LPID, r7
3308	mtspr	SPRN_PID, r0
3309	ld	r8, KVM_HOST_LPCR(r10)
3310	mtspr	SPRN_LPCR, r8
3311	isync
3312	li	r0, KVM_GUEST_MODE_NONE
3313	stb	r0, HSTATE_IN_GUEST(r13)
3314
3315	/*
3316	 * Turn on the MMU and jump to C code
3317	 */
3318	bcl	20, 31, .+4
33195:	mflr	r3
3320	addi	r3, r3, 9f - 5b
3321	li	r4, -1
3322	rldimi	r3, r4, 62, 0	/* ensure 0xc000000000000000 bits are set */
3323	ld	r4, PACAKMSR(r13)
3324	mtspr	SPRN_SRR0, r3
3325	mtspr	SPRN_SRR1, r4
3326	RFI_TO_KERNEL
33279:	addi	r3, r1, STACK_FRAME_OVERHEAD
3328	bl	kvmppc_bad_interrupt
3329	b	9b
3330
3331/*
3332 * This mimics the MSR transition on IRQ delivery.  The new guest MSR is taken
3333 * from VCPU_INTR_MSR and is modified based on the required TM state changes.
3334 *   r11 has the guest MSR value (in/out)
3335 *   r9 has a vcpu pointer (in)
3336 *   r0 is used as a scratch register
3337 */
3338kvmppc_msr_interrupt:
3339	rldicl	r0, r11, 64 - MSR_TS_S_LG, 62
3340	cmpwi	r0, 2 /* Check if we are in transactional state..  */
3341	ld	r11, VCPU_INTR_MSR(r9)
3342	bne	1f
3343	/* ... if transactional, change to suspended */
3344	li	r0, 1
33451:	rldimi	r11, r0, MSR_TS_S_LG, 63 - MSR_TS_T_LG
3346	blr
3347
3348/*
3349 * Load up guest PMU state.  R3 points to the vcpu struct.
3350 */
3351_GLOBAL(kvmhv_load_guest_pmu)
3352EXPORT_SYMBOL_GPL(kvmhv_load_guest_pmu)
3353	mr	r4, r3
3354	mflr	r0
3355	li	r3, 1
3356	sldi	r3, r3, 31		/* MMCR0_FC (freeze counters) bit */
3357	mtspr	SPRN_MMCR0, r3		/* freeze all counters, disable ints */
3358	isync
3359BEGIN_FTR_SECTION
3360	ld	r3, VCPU_MMCR(r4)
3361	andi.	r5, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
3362	cmpwi	r5, MMCR0_PMAO
3363	beql	kvmppc_fix_pmao
3364END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
3365	lwz	r3, VCPU_PMC(r4)	/* always load up guest PMU registers */
3366	lwz	r5, VCPU_PMC + 4(r4)	/* to prevent information leak */
3367	lwz	r6, VCPU_PMC + 8(r4)
3368	lwz	r7, VCPU_PMC + 12(r4)
3369	lwz	r8, VCPU_PMC + 16(r4)
3370	lwz	r9, VCPU_PMC + 20(r4)
3371	mtspr	SPRN_PMC1, r3
3372	mtspr	SPRN_PMC2, r5
3373	mtspr	SPRN_PMC3, r6
3374	mtspr	SPRN_PMC4, r7
3375	mtspr	SPRN_PMC5, r8
3376	mtspr	SPRN_PMC6, r9
3377	ld	r3, VCPU_MMCR(r4)
3378	ld	r5, VCPU_MMCR + 8(r4)
3379	ld	r6, VCPU_MMCR + 16(r4)
3380	ld	r7, VCPU_SIAR(r4)
3381	ld	r8, VCPU_SDAR(r4)
3382	mtspr	SPRN_MMCR1, r5
3383	mtspr	SPRN_MMCRA, r6
3384	mtspr	SPRN_SIAR, r7
3385	mtspr	SPRN_SDAR, r8
3386BEGIN_FTR_SECTION
3387	ld	r5, VCPU_MMCR + 24(r4)
3388	ld	r6, VCPU_SIER(r4)
3389	mtspr	SPRN_MMCR2, r5
3390	mtspr	SPRN_SIER, r6
3391BEGIN_FTR_SECTION_NESTED(96)
3392	lwz	r7, VCPU_PMC + 24(r4)
3393	lwz	r8, VCPU_PMC + 28(r4)
3394	ld	r9, VCPU_MMCR + 32(r4)
3395	mtspr	SPRN_SPMC1, r7
3396	mtspr	SPRN_SPMC2, r8
3397	mtspr	SPRN_MMCRS, r9
3398END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
3399END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3400	mtspr	SPRN_MMCR0, r3
3401	isync
3402	mtlr	r0
3403	blr
3404
3405/*
3406 * Reload host PMU state saved in the PACA by kvmhv_save_host_pmu.
3407 */
3408_GLOBAL(kvmhv_load_host_pmu)
3409EXPORT_SYMBOL_GPL(kvmhv_load_host_pmu)
3410	mflr	r0
3411	lbz	r4, PACA_PMCINUSE(r13) /* is the host using the PMU? */
3412	cmpwi	r4, 0
3413	beq	23f			/* skip if not */
3414BEGIN_FTR_SECTION
3415	ld	r3, HSTATE_MMCR0(r13)
3416	andi.	r4, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
3417	cmpwi	r4, MMCR0_PMAO
3418	beql	kvmppc_fix_pmao
3419END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
3420	lwz	r3, HSTATE_PMC1(r13)
3421	lwz	r4, HSTATE_PMC2(r13)
3422	lwz	r5, HSTATE_PMC3(r13)
3423	lwz	r6, HSTATE_PMC4(r13)
3424	lwz	r8, HSTATE_PMC5(r13)
3425	lwz	r9, HSTATE_PMC6(r13)
3426	mtspr	SPRN_PMC1, r3
3427	mtspr	SPRN_PMC2, r4
3428	mtspr	SPRN_PMC3, r5
3429	mtspr	SPRN_PMC4, r6
3430	mtspr	SPRN_PMC5, r8
3431	mtspr	SPRN_PMC6, r9
3432	ld	r3, HSTATE_MMCR0(r13)
3433	ld	r4, HSTATE_MMCR1(r13)
3434	ld	r5, HSTATE_MMCRA(r13)
3435	ld	r6, HSTATE_SIAR(r13)
3436	ld	r7, HSTATE_SDAR(r13)
3437	mtspr	SPRN_MMCR1, r4
3438	mtspr	SPRN_MMCRA, r5
3439	mtspr	SPRN_SIAR, r6
3440	mtspr	SPRN_SDAR, r7
3441BEGIN_FTR_SECTION
3442	ld	r8, HSTATE_MMCR2(r13)
3443	ld	r9, HSTATE_SIER(r13)
3444	mtspr	SPRN_MMCR2, r8
3445	mtspr	SPRN_SIER, r9
3446END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3447	mtspr	SPRN_MMCR0, r3
3448	isync
3449	mtlr	r0
345023:	blr
3451
3452/*
3453 * Save guest PMU state into the vcpu struct.
3454 * r3 = vcpu, r4 = full save flag (PMU in use flag set in VPA)
3455 */
3456_GLOBAL(kvmhv_save_guest_pmu)
3457EXPORT_SYMBOL_GPL(kvmhv_save_guest_pmu)
3458	mr	r9, r3
3459	mr	r8, r4
3460BEGIN_FTR_SECTION
3461	/*
3462	 * POWER8 seems to have a hardware bug where setting
3463	 * MMCR0[PMAE] along with MMCR0[PMC1CE] and/or MMCR0[PMCjCE]
3464	 * when some counters are already negative doesn't seem
3465	 * to cause a performance monitor alert (and hence interrupt).
3466	 * The effect of this is that when saving the PMU state,
3467	 * if there is no PMU alert pending when we read MMCR0
3468	 * before freezing the counters, but one becomes pending
3469	 * before we read the counters, we lose it.
3470	 * To work around this, we need a way to freeze the counters
3471	 * before reading MMCR0.  Normally, freezing the counters
3472	 * is done by writing MMCR0 (to set MMCR0[FC]) which
3473	 * unavoidably writes MMCR0[PMA0] as well.  On POWER8,
3474	 * we can also freeze the counters using MMCR2, by writing
3475	 * 1s to all the counter freeze condition bits (there are
3476	 * 9 bits each for 6 counters).
3477	 */
3478	li	r3, -1			/* set all freeze bits */
3479	clrrdi	r3, r3, 10
3480	mfspr	r10, SPRN_MMCR2
3481	mtspr	SPRN_MMCR2, r3
3482	isync
3483END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3484	li	r3, 1
3485	sldi	r3, r3, 31		/* MMCR0_FC (freeze counters) bit */
3486	mfspr	r4, SPRN_MMCR0		/* save MMCR0 */
3487	mtspr	SPRN_MMCR0, r3		/* freeze all counters, disable ints */
3488	mfspr	r6, SPRN_MMCRA
3489	/* Clear MMCRA in order to disable SDAR updates */
3490	li	r7, 0
3491	mtspr	SPRN_MMCRA, r7
3492	isync
3493	cmpwi	r8, 0			/* did they ask for PMU stuff to be saved? */
3494	bne	21f
3495	std	r3, VCPU_MMCR(r9)	/* if not, set saved MMCR0 to FC */
3496	b	22f
349721:	mfspr	r5, SPRN_MMCR1
3498	mfspr	r7, SPRN_SIAR
3499	mfspr	r8, SPRN_SDAR
3500	std	r4, VCPU_MMCR(r9)
3501	std	r5, VCPU_MMCR + 8(r9)
3502	std	r6, VCPU_MMCR + 16(r9)
3503BEGIN_FTR_SECTION
3504	std	r10, VCPU_MMCR + 24(r9)
3505END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3506	std	r7, VCPU_SIAR(r9)
3507	std	r8, VCPU_SDAR(r9)
3508	mfspr	r3, SPRN_PMC1
3509	mfspr	r4, SPRN_PMC2
3510	mfspr	r5, SPRN_PMC3
3511	mfspr	r6, SPRN_PMC4
3512	mfspr	r7, SPRN_PMC5
3513	mfspr	r8, SPRN_PMC6
3514	stw	r3, VCPU_PMC(r9)
3515	stw	r4, VCPU_PMC + 4(r9)
3516	stw	r5, VCPU_PMC + 8(r9)
3517	stw	r6, VCPU_PMC + 12(r9)
3518	stw	r7, VCPU_PMC + 16(r9)
3519	stw	r8, VCPU_PMC + 20(r9)
3520BEGIN_FTR_SECTION
3521	mfspr	r5, SPRN_SIER
3522	std	r5, VCPU_SIER(r9)
3523BEGIN_FTR_SECTION_NESTED(96)
3524	mfspr	r6, SPRN_SPMC1
3525	mfspr	r7, SPRN_SPMC2
3526	mfspr	r8, SPRN_MMCRS
3527	stw	r6, VCPU_PMC + 24(r9)
3528	stw	r7, VCPU_PMC + 28(r9)
3529	std	r8, VCPU_MMCR + 32(r9)
3530	lis	r4, 0x8000
3531	mtspr	SPRN_MMCRS, r4
3532END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
3533END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
353422:	blr
3535
3536/*
3537 * This works around a hardware bug on POWER8E processors, where
3538 * writing a 1 to the MMCR0[PMAO] bit doesn't generate a
3539 * performance monitor interrupt.  Instead, when we need to have
3540 * an interrupt pending, we have to arrange for a counter to overflow.
3541 */
3542kvmppc_fix_pmao:
3543	li	r3, 0
3544	mtspr	SPRN_MMCR2, r3
3545	lis	r3, (MMCR0_PMXE | MMCR0_FCECE)@h
3546	ori	r3, r3, MMCR0_PMCjCE | MMCR0_C56RUN
3547	mtspr	SPRN_MMCR0, r3
3548	lis	r3, 0x7fff
3549	ori	r3, r3, 0xffff
3550	mtspr	SPRN_PMC6, r3
3551	isync
3552	blr
3553
3554#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
3555/*
3556 * Start timing an activity
3557 * r3 = pointer to time accumulation struct, r4 = vcpu
3558 */
3559kvmhv_start_timing:
3560	ld	r5, HSTATE_KVM_VCORE(r13)
3561	ld	r6, VCORE_TB_OFFSET_APPL(r5)
3562	mftb	r5
3563	subf	r5, r6, r5	/* subtract current timebase offset */
3564	std	r3, VCPU_CUR_ACTIVITY(r4)
3565	std	r5, VCPU_ACTIVITY_START(r4)
3566	blr
3567
3568/*
3569 * Accumulate time to one activity and start another.
3570 * r3 = pointer to new time accumulation struct, r4 = vcpu
3571 */
3572kvmhv_accumulate_time:
3573	ld	r5, HSTATE_KVM_VCORE(r13)
3574	ld	r8, VCORE_TB_OFFSET_APPL(r5)
3575	ld	r5, VCPU_CUR_ACTIVITY(r4)
3576	ld	r6, VCPU_ACTIVITY_START(r4)
3577	std	r3, VCPU_CUR_ACTIVITY(r4)
3578	mftb	r7
3579	subf	r7, r8, r7	/* subtract current timebase offset */
3580	std	r7, VCPU_ACTIVITY_START(r4)
3581	cmpdi	r5, 0
3582	beqlr
3583	subf	r3, r6, r7
3584	ld	r8, TAS_SEQCOUNT(r5)
3585	cmpdi	r8, 0
3586	addi	r8, r8, 1
3587	std	r8, TAS_SEQCOUNT(r5)
3588	lwsync
3589	ld	r7, TAS_TOTAL(r5)
3590	add	r7, r7, r3
3591	std	r7, TAS_TOTAL(r5)
3592	ld	r6, TAS_MIN(r5)
3593	ld	r7, TAS_MAX(r5)
3594	beq	3f
3595	cmpd	r3, r6
3596	bge	1f
35973:	std	r3, TAS_MIN(r5)
35981:	cmpd	r3, r7
3599	ble	2f
3600	std	r3, TAS_MAX(r5)
36012:	lwsync
3602	addi	r8, r8, 1
3603	std	r8, TAS_SEQCOUNT(r5)
3604	blr
3605#endif
3606