xref: /linux/arch/powerpc/kvm/book3s_hv_rmhandlers.S (revision 160b8e75932fd51a49607d32dbfa1d417977b79c)
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/tm.h>
32#include <asm/opal.h>
33#include <asm/xive-regs.h>
34#include <asm/thread_info.h>
35
36/* Sign-extend HDEC if not on POWER9 */
37#define EXTEND_HDEC(reg)			\
38BEGIN_FTR_SECTION;				\
39	extsw	reg, reg;			\
40END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
41
42#define VCPU_GPRS_TM(reg) (((reg) * ULONG_SIZE) + VCPU_GPR_TM)
43
44/* Values in HSTATE_NAPPING(r13) */
45#define NAPPING_CEDE	1
46#define NAPPING_NOVCPU	2
47
48/* Stack frame offsets for kvmppc_hv_entry */
49#define SFS			160
50#define STACK_SLOT_TRAP		(SFS-4)
51#define STACK_SLOT_TID		(SFS-16)
52#define STACK_SLOT_PSSCR	(SFS-24)
53#define STACK_SLOT_PID		(SFS-32)
54#define STACK_SLOT_IAMR		(SFS-40)
55#define STACK_SLOT_CIABR	(SFS-48)
56#define STACK_SLOT_DAWR		(SFS-56)
57#define STACK_SLOT_DAWRX	(SFS-64)
58#define STACK_SLOT_HFSCR	(SFS-72)
59
60/*
61 * Call kvmppc_hv_entry in real mode.
62 * Must be called with interrupts hard-disabled.
63 *
64 * Input Registers:
65 *
66 * LR = return address to continue at after eventually re-enabling MMU
67 */
68_GLOBAL_TOC(kvmppc_hv_entry_trampoline)
69	mflr	r0
70	std	r0, PPC_LR_STKOFF(r1)
71	stdu	r1, -112(r1)
72	mfmsr	r10
73	std	r10, HSTATE_HOST_MSR(r13)
74	LOAD_REG_ADDR(r5, kvmppc_call_hv_entry)
75	li	r0,MSR_RI
76	andc	r0,r10,r0
77	li	r6,MSR_IR | MSR_DR
78	andc	r6,r10,r6
79	mtmsrd	r0,1		/* clear RI in MSR */
80	mtsrr0	r5
81	mtsrr1	r6
82	RFI_TO_KERNEL
83
84kvmppc_call_hv_entry:
85BEGIN_FTR_SECTION
86	/* On P9, do LPCR setting, if necessary */
87	ld	r3, HSTATE_SPLIT_MODE(r13)
88	cmpdi	r3, 0
89	beq	46f
90	lwz	r4, KVM_SPLIT_DO_SET(r3)
91	cmpwi	r4, 0
92	beq	46f
93	bl	kvmhv_p9_set_lpcr
94	nop
9546:
96END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
97
98	ld	r4, HSTATE_KVM_VCPU(r13)
99	bl	kvmppc_hv_entry
100
101	/* Back from guest - restore host state and return to caller */
102
103BEGIN_FTR_SECTION
104	/* Restore host DABR and DABRX */
105	ld	r5,HSTATE_DABR(r13)
106	li	r6,7
107	mtspr	SPRN_DABR,r5
108	mtspr	SPRN_DABRX,r6
109END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
110
111	/* Restore SPRG3 */
112	ld	r3,PACA_SPRG_VDSO(r13)
113	mtspr	SPRN_SPRG_VDSO_WRITE,r3
114
115	/* Reload the host's PMU registers */
116	ld	r3, PACALPPACAPTR(r13)	/* is the host using the PMU? */
117	lbz	r4, LPPACA_PMCINUSE(r3)
118	cmpwi	r4, 0
119	beq	23f			/* skip if not */
120BEGIN_FTR_SECTION
121	ld	r3, HSTATE_MMCR0(r13)
122	andi.	r4, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
123	cmpwi	r4, MMCR0_PMAO
124	beql	kvmppc_fix_pmao
125END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
126	lwz	r3, HSTATE_PMC1(r13)
127	lwz	r4, HSTATE_PMC2(r13)
128	lwz	r5, HSTATE_PMC3(r13)
129	lwz	r6, HSTATE_PMC4(r13)
130	lwz	r8, HSTATE_PMC5(r13)
131	lwz	r9, HSTATE_PMC6(r13)
132	mtspr	SPRN_PMC1, r3
133	mtspr	SPRN_PMC2, r4
134	mtspr	SPRN_PMC3, r5
135	mtspr	SPRN_PMC4, r6
136	mtspr	SPRN_PMC5, r8
137	mtspr	SPRN_PMC6, r9
138	ld	r3, HSTATE_MMCR0(r13)
139	ld	r4, HSTATE_MMCR1(r13)
140	ld	r5, HSTATE_MMCRA(r13)
141	ld	r6, HSTATE_SIAR(r13)
142	ld	r7, HSTATE_SDAR(r13)
143	mtspr	SPRN_MMCR1, r4
144	mtspr	SPRN_MMCRA, r5
145	mtspr	SPRN_SIAR, r6
146	mtspr	SPRN_SDAR, r7
147BEGIN_FTR_SECTION
148	ld	r8, HSTATE_MMCR2(r13)
149	ld	r9, HSTATE_SIER(r13)
150	mtspr	SPRN_MMCR2, r8
151	mtspr	SPRN_SIER, r9
152END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
153	mtspr	SPRN_MMCR0, r3
154	isync
15523:
156
157	/*
158	 * Reload DEC.  HDEC interrupts were disabled when
159	 * we reloaded the host's LPCR value.
160	 */
161	ld	r3, HSTATE_DECEXP(r13)
162	mftb	r4
163	subf	r4, r4, r3
164	mtspr	SPRN_DEC, r4
165
166	/* hwthread_req may have got set by cede or no vcpu, so clear it */
167	li	r0, 0
168	stb	r0, HSTATE_HWTHREAD_REQ(r13)
169
170	/*
171	 * For external interrupts we need to call the Linux
172	 * handler to process the interrupt. We do that by jumping
173	 * to absolute address 0x500 for external interrupts.
174	 * The [h]rfid at the end of the handler will return to
175	 * the book3s_hv_interrupts.S code. For other interrupts
176	 * we do the rfid to get back to the book3s_hv_interrupts.S
177	 * code here.
178	 */
179	ld	r8, 112+PPC_LR_STKOFF(r1)
180	addi	r1, r1, 112
181	ld	r7, HSTATE_HOST_MSR(r13)
182
183	/* Return the trap number on this thread as the return value */
184	mr	r3, r12
185
186	/*
187	 * If we came back from the guest via a relocation-on interrupt,
188	 * we will be in virtual mode at this point, which makes it a
189	 * little easier to get back to the caller.
190	 */
191	mfmsr	r0
192	andi.	r0, r0, MSR_IR		/* in real mode? */
193	bne	.Lvirt_return
194
195	/* RFI into the highmem handler */
196	mfmsr	r6
197	li	r0, MSR_RI
198	andc	r6, r6, r0
199	mtmsrd	r6, 1			/* Clear RI in MSR */
200	mtsrr0	r8
201	mtsrr1	r7
202	RFI_TO_KERNEL
203
204	/* Virtual-mode return */
205.Lvirt_return:
206	mtlr	r8
207	blr
208
209kvmppc_primary_no_guest:
210	/* We handle this much like a ceded vcpu */
211	/* put the HDEC into the DEC, since HDEC interrupts don't wake us */
212	/* HDEC may be larger than DEC for arch >= v3.00, but since the */
213	/* HDEC value came from DEC in the first place, it will fit */
214	mfspr	r3, SPRN_HDEC
215	mtspr	SPRN_DEC, r3
216	/*
217	 * Make sure the primary has finished the MMU switch.
218	 * We should never get here on a secondary thread, but
219	 * check it for robustness' sake.
220	 */
221	ld	r5, HSTATE_KVM_VCORE(r13)
22265:	lbz	r0, VCORE_IN_GUEST(r5)
223	cmpwi	r0, 0
224	beq	65b
225	/* Set LPCR. */
226	ld	r8,VCORE_LPCR(r5)
227	mtspr	SPRN_LPCR,r8
228	isync
229	/* set our bit in napping_threads */
230	ld	r5, HSTATE_KVM_VCORE(r13)
231	lbz	r7, HSTATE_PTID(r13)
232	li	r0, 1
233	sld	r0, r0, r7
234	addi	r6, r5, VCORE_NAPPING_THREADS
2351:	lwarx	r3, 0, r6
236	or	r3, r3, r0
237	stwcx.	r3, 0, r6
238	bne	1b
239	/* order napping_threads update vs testing entry_exit_map */
240	isync
241	li	r12, 0
242	lwz	r7, VCORE_ENTRY_EXIT(r5)
243	cmpwi	r7, 0x100
244	bge	kvm_novcpu_exit	/* another thread already exiting */
245	li	r3, NAPPING_NOVCPU
246	stb	r3, HSTATE_NAPPING(r13)
247
248	li	r3, 0		/* Don't wake on privileged (OS) doorbell */
249	b	kvm_do_nap
250
251/*
252 * kvm_novcpu_wakeup
253 *	Entered from kvm_start_guest if kvm_hstate.napping is set
254 *	to NAPPING_NOVCPU
255 *		r2 = kernel TOC
256 *		r13 = paca
257 */
258kvm_novcpu_wakeup:
259	ld	r1, HSTATE_HOST_R1(r13)
260	ld	r5, HSTATE_KVM_VCORE(r13)
261	li	r0, 0
262	stb	r0, HSTATE_NAPPING(r13)
263
264	/* check the wake reason */
265	bl	kvmppc_check_wake_reason
266
267	/*
268	 * Restore volatile registers since we could have called
269	 * a C routine in kvmppc_check_wake_reason.
270	 *	r5 = VCORE
271	 */
272	ld	r5, HSTATE_KVM_VCORE(r13)
273
274	/* see if any other thread is already exiting */
275	lwz	r0, VCORE_ENTRY_EXIT(r5)
276	cmpwi	r0, 0x100
277	bge	kvm_novcpu_exit
278
279	/* clear our bit in napping_threads */
280	lbz	r7, HSTATE_PTID(r13)
281	li	r0, 1
282	sld	r0, r0, r7
283	addi	r6, r5, VCORE_NAPPING_THREADS
2844:	lwarx	r7, 0, r6
285	andc	r7, r7, r0
286	stwcx.	r7, 0, r6
287	bne	4b
288
289	/* See if the wake reason means we need to exit */
290	cmpdi	r3, 0
291	bge	kvm_novcpu_exit
292
293	/* See if our timeslice has expired (HDEC is negative) */
294	mfspr	r0, SPRN_HDEC
295	EXTEND_HDEC(r0)
296	li	r12, BOOK3S_INTERRUPT_HV_DECREMENTER
297	cmpdi	r0, 0
298	blt	kvm_novcpu_exit
299
300	/* Got an IPI but other vcpus aren't yet exiting, must be a latecomer */
301	ld	r4, HSTATE_KVM_VCPU(r13)
302	cmpdi	r4, 0
303	beq	kvmppc_primary_no_guest
304
305#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
306	addi	r3, r4, VCPU_TB_RMENTRY
307	bl	kvmhv_start_timing
308#endif
309	b	kvmppc_got_guest
310
311kvm_novcpu_exit:
312#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
313	ld	r4, HSTATE_KVM_VCPU(r13)
314	cmpdi	r4, 0
315	beq	13f
316	addi	r3, r4, VCPU_TB_RMEXIT
317	bl	kvmhv_accumulate_time
318#endif
31913:	mr	r3, r12
320	stw	r12, STACK_SLOT_TRAP(r1)
321	bl	kvmhv_commence_exit
322	nop
323	lwz	r12, STACK_SLOT_TRAP(r1)
324	b	kvmhv_switch_to_host
325
326/*
327 * We come in here when wakened from nap mode.
328 * Relocation is off and most register values are lost.
329 * r13 points to the PACA.
330 * r3 contains the SRR1 wakeup value, SRR1 is trashed.
331 */
332	.globl	kvm_start_guest
333kvm_start_guest:
334	/* Set runlatch bit the minute you wake up from nap */
335	mfspr	r0, SPRN_CTRLF
336	ori 	r0, r0, 1
337	mtspr	SPRN_CTRLT, r0
338
339	/*
340	 * Could avoid this and pass it through in r3. For now,
341	 * code expects it to be in SRR1.
342	 */
343	mtspr	SPRN_SRR1,r3
344
345	ld	r2,PACATOC(r13)
346
347	li	r0,KVM_HWTHREAD_IN_KVM
348	stb	r0,HSTATE_HWTHREAD_STATE(r13)
349
350	/* NV GPR values from power7_idle() will no longer be valid */
351	li	r0,1
352	stb	r0,PACA_NAPSTATELOST(r13)
353
354	/* were we napping due to cede? */
355	lbz	r0,HSTATE_NAPPING(r13)
356	cmpwi	r0,NAPPING_CEDE
357	beq	kvm_end_cede
358	cmpwi	r0,NAPPING_NOVCPU
359	beq	kvm_novcpu_wakeup
360
361	ld	r1,PACAEMERGSP(r13)
362	subi	r1,r1,STACK_FRAME_OVERHEAD
363
364	/*
365	 * We weren't napping due to cede, so this must be a secondary
366	 * thread being woken up to run a guest, or being woken up due
367	 * to a stray IPI.  (Or due to some machine check or hypervisor
368	 * maintenance interrupt while the core is in KVM.)
369	 */
370
371	/* Check the wake reason in SRR1 to see why we got here */
372	bl	kvmppc_check_wake_reason
373	/*
374	 * kvmppc_check_wake_reason could invoke a C routine, but we
375	 * have no volatile registers to restore when we return.
376	 */
377
378	cmpdi	r3, 0
379	bge	kvm_no_guest
380
381	/* get vcore pointer, NULL if we have nothing to run */
382	ld	r5,HSTATE_KVM_VCORE(r13)
383	cmpdi	r5,0
384	/* if we have no vcore to run, go back to sleep */
385	beq	kvm_no_guest
386
387kvm_secondary_got_guest:
388
389	/* Set HSTATE_DSCR(r13) to something sensible */
390	ld	r6, PACA_DSCR_DEFAULT(r13)
391	std	r6, HSTATE_DSCR(r13)
392
393	/* On thread 0 of a subcore, set HDEC to max */
394	lbz	r4, HSTATE_PTID(r13)
395	cmpwi	r4, 0
396	bne	63f
397	LOAD_REG_ADDR(r6, decrementer_max)
398	ld	r6, 0(r6)
399	mtspr	SPRN_HDEC, r6
400	/* and set per-LPAR registers, if doing dynamic micro-threading */
401	ld	r6, HSTATE_SPLIT_MODE(r13)
402	cmpdi	r6, 0
403	beq	63f
404BEGIN_FTR_SECTION
405	ld	r0, KVM_SPLIT_RPR(r6)
406	mtspr	SPRN_RPR, r0
407	ld	r0, KVM_SPLIT_PMMAR(r6)
408	mtspr	SPRN_PMMAR, r0
409	ld	r0, KVM_SPLIT_LDBAR(r6)
410	mtspr	SPRN_LDBAR, r0
411	isync
412FTR_SECTION_ELSE
413	/* On P9 we use the split_info for coordinating LPCR changes */
414	lwz	r4, KVM_SPLIT_DO_SET(r6)
415	cmpwi	r4, 0
416	beq	1f
417	mr	r3, r6
418	bl	kvmhv_p9_set_lpcr
419	nop
4201:
421ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
42263:
423	/* Order load of vcpu after load of vcore */
424	lwsync
425	ld	r4, HSTATE_KVM_VCPU(r13)
426	bl	kvmppc_hv_entry
427
428	/* Back from the guest, go back to nap */
429	/* Clear our vcpu and vcore pointers so we don't come back in early */
430	li	r0, 0
431	std	r0, HSTATE_KVM_VCPU(r13)
432	/*
433	 * Once we clear HSTATE_KVM_VCORE(r13), the code in
434	 * kvmppc_run_core() is going to assume that all our vcpu
435	 * state is visible in memory.  This lwsync makes sure
436	 * that that is true.
437	 */
438	lwsync
439	std	r0, HSTATE_KVM_VCORE(r13)
440
441	/*
442	 * All secondaries exiting guest will fall through this path.
443	 * Before proceeding, just check for HMI interrupt and
444	 * invoke opal hmi handler. By now we are sure that the
445	 * primary thread on this core/subcore has already made partition
446	 * switch/TB resync and we are good to call opal hmi handler.
447	 */
448	cmpwi	r12, BOOK3S_INTERRUPT_HMI
449	bne	kvm_no_guest
450
451	li	r3,0			/* NULL argument */
452	bl	hmi_exception_realmode
453/*
454 * At this point we have finished executing in the guest.
455 * We need to wait for hwthread_req to become zero, since
456 * we may not turn on the MMU while hwthread_req is non-zero.
457 * While waiting we also need to check if we get given a vcpu to run.
458 */
459kvm_no_guest:
460	lbz	r3, HSTATE_HWTHREAD_REQ(r13)
461	cmpwi	r3, 0
462	bne	53f
463	HMT_MEDIUM
464	li	r0, KVM_HWTHREAD_IN_KERNEL
465	stb	r0, HSTATE_HWTHREAD_STATE(r13)
466	/* need to recheck hwthread_req after a barrier, to avoid race */
467	sync
468	lbz	r3, HSTATE_HWTHREAD_REQ(r13)
469	cmpwi	r3, 0
470	bne	54f
471/*
472 * We jump to pnv_wakeup_loss, which will return to the caller
473 * of power7_nap in the powernv cpu offline loop.  The value we
474 * put in r3 becomes the return value for power7_nap. pnv_wakeup_loss
475 * requires SRR1 in r12.
476 */
477	li	r3, LPCR_PECE0
478	mfspr	r4, SPRN_LPCR
479	rlwimi	r4, r3, 0, LPCR_PECE0 | LPCR_PECE1
480	mtspr	SPRN_LPCR, r4
481	li	r3, 0
482	mfspr	r12,SPRN_SRR1
483	b	pnv_wakeup_loss
484
48553:	HMT_LOW
486	ld	r5, HSTATE_KVM_VCORE(r13)
487	cmpdi	r5, 0
488	bne	60f
489	ld	r3, HSTATE_SPLIT_MODE(r13)
490	cmpdi	r3, 0
491	beq	kvm_no_guest
492	lwz	r0, KVM_SPLIT_DO_SET(r3)
493	cmpwi	r0, 0
494	bne	kvmhv_do_set
495	lwz	r0, KVM_SPLIT_DO_RESTORE(r3)
496	cmpwi	r0, 0
497	bne	kvmhv_do_restore
498	lbz	r0, KVM_SPLIT_DO_NAP(r3)
499	cmpwi	r0, 0
500	beq	kvm_no_guest
501	HMT_MEDIUM
502	b	kvm_unsplit_nap
50360:	HMT_MEDIUM
504	b	kvm_secondary_got_guest
505
50654:	li	r0, KVM_HWTHREAD_IN_KVM
507	stb	r0, HSTATE_HWTHREAD_STATE(r13)
508	b	kvm_no_guest
509
510kvmhv_do_set:
511	/* Set LPCR, LPIDR etc. on P9 */
512	HMT_MEDIUM
513	bl	kvmhv_p9_set_lpcr
514	nop
515	b	kvm_no_guest
516
517kvmhv_do_restore:
518	HMT_MEDIUM
519	bl	kvmhv_p9_restore_lpcr
520	nop
521	b	kvm_no_guest
522
523/*
524 * Here the primary thread is trying to return the core to
525 * whole-core mode, so we need to nap.
526 */
527kvm_unsplit_nap:
528	/*
529	 * When secondaries are napping in kvm_unsplit_nap() with
530	 * hwthread_req = 1, HMI goes ignored even though subcores are
531	 * already exited the guest. Hence HMI keeps waking up secondaries
532	 * from nap in a loop and secondaries always go back to nap since
533	 * no vcore is assigned to them. This makes impossible for primary
534	 * thread to get hold of secondary threads resulting into a soft
535	 * lockup in KVM path.
536	 *
537	 * Let us check if HMI is pending and handle it before we go to nap.
538	 */
539	cmpwi	r12, BOOK3S_INTERRUPT_HMI
540	bne	55f
541	li	r3, 0			/* NULL argument */
542	bl	hmi_exception_realmode
54355:
544	/*
545	 * Ensure that secondary doesn't nap when it has
546	 * its vcore pointer set.
547	 */
548	sync		/* matches smp_mb() before setting split_info.do_nap */
549	ld	r0, HSTATE_KVM_VCORE(r13)
550	cmpdi	r0, 0
551	bne	kvm_no_guest
552	/* clear any pending message */
553BEGIN_FTR_SECTION
554	lis	r6, (PPC_DBELL_SERVER << (63-36))@h
555	PPC_MSGCLR(6)
556END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
557	/* Set kvm_split_mode.napped[tid] = 1 */
558	ld	r3, HSTATE_SPLIT_MODE(r13)
559	li	r0, 1
560	lbz	r4, HSTATE_TID(r13)
561	addi	r4, r4, KVM_SPLIT_NAPPED
562	stbx	r0, r3, r4
563	/* Check the do_nap flag again after setting napped[] */
564	sync
565	lbz	r0, KVM_SPLIT_DO_NAP(r3)
566	cmpwi	r0, 0
567	beq	57f
568	li	r3, (LPCR_PECEDH | LPCR_PECE0) >> 4
569	mfspr	r5, SPRN_LPCR
570	rlwimi	r5, r3, 4, (LPCR_PECEDP | LPCR_PECEDH | LPCR_PECE0 | LPCR_PECE1)
571	b	kvm_nap_sequence
572
57357:	li	r0, 0
574	stbx	r0, r3, r4
575	b	kvm_no_guest
576
577/******************************************************************************
578 *                                                                            *
579 *                               Entry code                                   *
580 *                                                                            *
581 *****************************************************************************/
582
583.global kvmppc_hv_entry
584kvmppc_hv_entry:
585
586	/* Required state:
587	 *
588	 * R4 = vcpu pointer (or NULL)
589	 * MSR = ~IR|DR
590	 * R13 = PACA
591	 * R1 = host R1
592	 * R2 = TOC
593	 * all other volatile GPRS = free
594	 * Does not preserve non-volatile GPRs or CR fields
595	 */
596	mflr	r0
597	std	r0, PPC_LR_STKOFF(r1)
598	stdu	r1, -SFS(r1)
599
600	/* Save R1 in the PACA */
601	std	r1, HSTATE_HOST_R1(r13)
602
603	li	r6, KVM_GUEST_MODE_HOST_HV
604	stb	r6, HSTATE_IN_GUEST(r13)
605
606#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
607	/* Store initial timestamp */
608	cmpdi	r4, 0
609	beq	1f
610	addi	r3, r4, VCPU_TB_RMENTRY
611	bl	kvmhv_start_timing
6121:
613#endif
614
615	/* Use cr7 as an indication of radix mode */
616	ld	r5, HSTATE_KVM_VCORE(r13)
617	ld	r9, VCORE_KVM(r5)	/* pointer to struct kvm */
618	lbz	r0, KVM_RADIX(r9)
619	cmpwi	cr7, r0, 0
620
621	/*
622	 * POWER7/POWER8 host -> guest partition switch code.
623	 * We don't have to lock against concurrent tlbies,
624	 * but we do have to coordinate across hardware threads.
625	 */
626	/* Set bit in entry map iff exit map is zero. */
627	li	r7, 1
628	lbz	r6, HSTATE_PTID(r13)
629	sld	r7, r7, r6
630	addi	r8, r5, VCORE_ENTRY_EXIT
63121:	lwarx	r3, 0, r8
632	cmpwi	r3, 0x100		/* any threads starting to exit? */
633	bge	secondary_too_late	/* if so we're too late to the party */
634	or	r3, r3, r7
635	stwcx.	r3, 0, r8
636	bne	21b
637
638	/* Primary thread switches to guest partition. */
639	cmpwi	r6,0
640	bne	10f
641	lwz	r7,KVM_LPID(r9)
642BEGIN_FTR_SECTION
643	ld	r6,KVM_SDR1(r9)
644	li	r0,LPID_RSVD		/* switch to reserved LPID */
645	mtspr	SPRN_LPID,r0
646	ptesync
647	mtspr	SPRN_SDR1,r6		/* switch to partition page table */
648END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
649	mtspr	SPRN_LPID,r7
650	isync
651
652	/* See if we need to flush the TLB */
653	lhz	r6,PACAPACAINDEX(r13)	/* test_bit(cpu, need_tlb_flush) */
654BEGIN_FTR_SECTION
655	/*
656	 * On POWER9, individual threads can come in here, but the
657	 * TLB is shared between the 4 threads in a core, hence
658	 * invalidating on one thread invalidates for all.
659	 * Thus we make all 4 threads use the same bit here.
660	 */
661	clrrdi	r6,r6,2
662END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
663	clrldi	r7,r6,64-6		/* extract bit number (6 bits) */
664	srdi	r6,r6,6			/* doubleword number */
665	sldi	r6,r6,3			/* address offset */
666	add	r6,r6,r9
667	addi	r6,r6,KVM_NEED_FLUSH	/* dword in kvm->arch.need_tlb_flush */
668	li	r8,1
669	sld	r8,r8,r7
670	ld	r7,0(r6)
671	and.	r7,r7,r8
672	beq	22f
673	/* Flush the TLB of any entries for this LPID */
674	lwz	r0,KVM_TLB_SETS(r9)
675	mtctr	r0
676	li	r7,0x800		/* IS field = 0b10 */
677	ptesync
678	li	r0,0			/* RS for P9 version of tlbiel */
679	bne	cr7, 29f
68028:	tlbiel	r7			/* On P9, rs=0, RIC=0, PRS=0, R=0 */
681	addi	r7,r7,0x1000
682	bdnz	28b
683	b	30f
68429:	PPC_TLBIEL(7,0,2,1,1)		/* for radix, RIC=2, PRS=1, R=1 */
685	addi	r7,r7,0x1000
686	bdnz	29b
68730:	ptesync
68823:	ldarx	r7,0,r6			/* clear the bit after TLB flushed */
689	andc	r7,r7,r8
690	stdcx.	r7,0,r6
691	bne	23b
692
693	/* Add timebase offset onto timebase */
69422:	ld	r8,VCORE_TB_OFFSET(r5)
695	cmpdi	r8,0
696	beq	37f
697	mftb	r6		/* current host timebase */
698	add	r8,r8,r6
699	mtspr	SPRN_TBU40,r8	/* update upper 40 bits */
700	mftb	r7		/* check if lower 24 bits overflowed */
701	clrldi	r6,r6,40
702	clrldi	r7,r7,40
703	cmpld	r7,r6
704	bge	37f
705	addis	r8,r8,0x100	/* if so, increment upper 40 bits */
706	mtspr	SPRN_TBU40,r8
707
708	/* Load guest PCR value to select appropriate compat mode */
70937:	ld	r7, VCORE_PCR(r5)
710	cmpdi	r7, 0
711	beq	38f
712	mtspr	SPRN_PCR, r7
71338:
714
715BEGIN_FTR_SECTION
716	/* DPDES and VTB are shared between threads */
717	ld	r8, VCORE_DPDES(r5)
718	ld	r7, VCORE_VTB(r5)
719	mtspr	SPRN_DPDES, r8
720	mtspr	SPRN_VTB, r7
721END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
722
723	/* Mark the subcore state as inside guest */
724	bl	kvmppc_subcore_enter_guest
725	nop
726	ld	r5, HSTATE_KVM_VCORE(r13)
727	ld	r4, HSTATE_KVM_VCPU(r13)
728	li	r0,1
729	stb	r0,VCORE_IN_GUEST(r5)	/* signal secondaries to continue */
730
731	/* Do we have a guest vcpu to run? */
73210:	cmpdi	r4, 0
733	beq	kvmppc_primary_no_guest
734kvmppc_got_guest:
735	/* Increment yield count if they have a VPA */
736	ld	r3, VCPU_VPA(r4)
737	cmpdi	r3, 0
738	beq	25f
739	li	r6, LPPACA_YIELDCOUNT
740	LWZX_BE	r5, r3, r6
741	addi	r5, r5, 1
742	STWX_BE	r5, r3, r6
743	li	r6, 1
744	stb	r6, VCPU_VPA_DIRTY(r4)
74525:
746
747	/* Save purr/spurr */
748	mfspr	r5,SPRN_PURR
749	mfspr	r6,SPRN_SPURR
750	std	r5,HSTATE_PURR(r13)
751	std	r6,HSTATE_SPURR(r13)
752	ld	r7,VCPU_PURR(r4)
753	ld	r8,VCPU_SPURR(r4)
754	mtspr	SPRN_PURR,r7
755	mtspr	SPRN_SPURR,r8
756
757	/* Save host values of some registers */
758BEGIN_FTR_SECTION
759	mfspr	r5, SPRN_TIDR
760	mfspr	r6, SPRN_PSSCR
761	mfspr	r7, SPRN_PID
762	mfspr	r8, SPRN_IAMR
763	std	r5, STACK_SLOT_TID(r1)
764	std	r6, STACK_SLOT_PSSCR(r1)
765	std	r7, STACK_SLOT_PID(r1)
766	std	r8, STACK_SLOT_IAMR(r1)
767	mfspr	r5, SPRN_HFSCR
768	std	r5, STACK_SLOT_HFSCR(r1)
769END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
770BEGIN_FTR_SECTION
771	mfspr	r5, SPRN_CIABR
772	mfspr	r6, SPRN_DAWR
773	mfspr	r7, SPRN_DAWRX
774	std	r5, STACK_SLOT_CIABR(r1)
775	std	r6, STACK_SLOT_DAWR(r1)
776	std	r7, STACK_SLOT_DAWRX(r1)
777END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
778
779BEGIN_FTR_SECTION
780	/* Set partition DABR */
781	/* Do this before re-enabling PMU to avoid P7 DABR corruption bug */
782	lwz	r5,VCPU_DABRX(r4)
783	ld	r6,VCPU_DABR(r4)
784	mtspr	SPRN_DABRX,r5
785	mtspr	SPRN_DABR,r6
786	isync
787END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
788
789#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
790BEGIN_FTR_SECTION
791	/*
792	 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS INCLUDING CR
793	 */
794	bl	kvmppc_restore_tm
795END_FTR_SECTION_IFSET(CPU_FTR_TM)
796#endif
797
798	/* Load guest PMU registers */
799	/* R4 is live here (vcpu pointer) */
800	li	r3, 1
801	sldi	r3, r3, 31		/* MMCR0_FC (freeze counters) bit */
802	mtspr	SPRN_MMCR0, r3		/* freeze all counters, disable ints */
803	isync
804BEGIN_FTR_SECTION
805	ld	r3, VCPU_MMCR(r4)
806	andi.	r5, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
807	cmpwi	r5, MMCR0_PMAO
808	beql	kvmppc_fix_pmao
809END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
810	lwz	r3, VCPU_PMC(r4)	/* always load up guest PMU registers */
811	lwz	r5, VCPU_PMC + 4(r4)	/* to prevent information leak */
812	lwz	r6, VCPU_PMC + 8(r4)
813	lwz	r7, VCPU_PMC + 12(r4)
814	lwz	r8, VCPU_PMC + 16(r4)
815	lwz	r9, VCPU_PMC + 20(r4)
816	mtspr	SPRN_PMC1, r3
817	mtspr	SPRN_PMC2, r5
818	mtspr	SPRN_PMC3, r6
819	mtspr	SPRN_PMC4, r7
820	mtspr	SPRN_PMC5, r8
821	mtspr	SPRN_PMC6, r9
822	ld	r3, VCPU_MMCR(r4)
823	ld	r5, VCPU_MMCR + 8(r4)
824	ld	r6, VCPU_MMCR + 16(r4)
825	ld	r7, VCPU_SIAR(r4)
826	ld	r8, VCPU_SDAR(r4)
827	mtspr	SPRN_MMCR1, r5
828	mtspr	SPRN_MMCRA, r6
829	mtspr	SPRN_SIAR, r7
830	mtspr	SPRN_SDAR, r8
831BEGIN_FTR_SECTION
832	ld	r5, VCPU_MMCR + 24(r4)
833	ld	r6, VCPU_SIER(r4)
834	mtspr	SPRN_MMCR2, r5
835	mtspr	SPRN_SIER, r6
836BEGIN_FTR_SECTION_NESTED(96)
837	lwz	r7, VCPU_PMC + 24(r4)
838	lwz	r8, VCPU_PMC + 28(r4)
839	ld	r9, VCPU_MMCR + 32(r4)
840	mtspr	SPRN_SPMC1, r7
841	mtspr	SPRN_SPMC2, r8
842	mtspr	SPRN_MMCRS, r9
843END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
844END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
845	mtspr	SPRN_MMCR0, r3
846	isync
847
848	/* Load up FP, VMX and VSX registers */
849	bl	kvmppc_load_fp
850
851	ld	r14, VCPU_GPR(R14)(r4)
852	ld	r15, VCPU_GPR(R15)(r4)
853	ld	r16, VCPU_GPR(R16)(r4)
854	ld	r17, VCPU_GPR(R17)(r4)
855	ld	r18, VCPU_GPR(R18)(r4)
856	ld	r19, VCPU_GPR(R19)(r4)
857	ld	r20, VCPU_GPR(R20)(r4)
858	ld	r21, VCPU_GPR(R21)(r4)
859	ld	r22, VCPU_GPR(R22)(r4)
860	ld	r23, VCPU_GPR(R23)(r4)
861	ld	r24, VCPU_GPR(R24)(r4)
862	ld	r25, VCPU_GPR(R25)(r4)
863	ld	r26, VCPU_GPR(R26)(r4)
864	ld	r27, VCPU_GPR(R27)(r4)
865	ld	r28, VCPU_GPR(R28)(r4)
866	ld	r29, VCPU_GPR(R29)(r4)
867	ld	r30, VCPU_GPR(R30)(r4)
868	ld	r31, VCPU_GPR(R31)(r4)
869
870	/* Switch DSCR to guest value */
871	ld	r5, VCPU_DSCR(r4)
872	mtspr	SPRN_DSCR, r5
873
874BEGIN_FTR_SECTION
875	/* Skip next section on POWER7 */
876	b	8f
877END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
878	/* Load up POWER8-specific registers */
879	ld	r5, VCPU_IAMR(r4)
880	lwz	r6, VCPU_PSPB(r4)
881	ld	r7, VCPU_FSCR(r4)
882	mtspr	SPRN_IAMR, r5
883	mtspr	SPRN_PSPB, r6
884	mtspr	SPRN_FSCR, r7
885	ld	r5, VCPU_DAWR(r4)
886	ld	r6, VCPU_DAWRX(r4)
887	ld	r7, VCPU_CIABR(r4)
888	ld	r8, VCPU_TAR(r4)
889	mtspr	SPRN_DAWR, r5
890	mtspr	SPRN_DAWRX, r6
891	mtspr	SPRN_CIABR, r7
892	mtspr	SPRN_TAR, r8
893	ld	r5, VCPU_IC(r4)
894	ld	r8, VCPU_EBBHR(r4)
895	mtspr	SPRN_IC, r5
896	mtspr	SPRN_EBBHR, r8
897	ld	r5, VCPU_EBBRR(r4)
898	ld	r6, VCPU_BESCR(r4)
899	lwz	r7, VCPU_GUEST_PID(r4)
900	ld	r8, VCPU_WORT(r4)
901	mtspr	SPRN_EBBRR, r5
902	mtspr	SPRN_BESCR, r6
903	mtspr	SPRN_PID, r7
904	mtspr	SPRN_WORT, r8
905BEGIN_FTR_SECTION
906	PPC_INVALIDATE_ERAT
907END_FTR_SECTION_IFSET(CPU_FTR_POWER9_DD1)
908BEGIN_FTR_SECTION
909	/* POWER8-only registers */
910	ld	r5, VCPU_TCSCR(r4)
911	ld	r6, VCPU_ACOP(r4)
912	ld	r7, VCPU_CSIGR(r4)
913	ld	r8, VCPU_TACR(r4)
914	mtspr	SPRN_TCSCR, r5
915	mtspr	SPRN_ACOP, r6
916	mtspr	SPRN_CSIGR, r7
917	mtspr	SPRN_TACR, r8
918FTR_SECTION_ELSE
919	/* POWER9-only registers */
920	ld	r5, VCPU_TID(r4)
921	ld	r6, VCPU_PSSCR(r4)
922	oris	r6, r6, PSSCR_EC@h	/* This makes stop trap to HV */
923	ld	r7, VCPU_HFSCR(r4)
924	mtspr	SPRN_TIDR, r5
925	mtspr	SPRN_PSSCR, r6
926	mtspr	SPRN_HFSCR, r7
927ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
9288:
929
930	/*
931	 * Set the decrementer to the guest decrementer.
932	 */
933	ld	r8,VCPU_DEC_EXPIRES(r4)
934	/* r8 is a host timebase value here, convert to guest TB */
935	ld	r5,HSTATE_KVM_VCORE(r13)
936	ld	r6,VCORE_TB_OFFSET(r5)
937	add	r8,r8,r6
938	mftb	r7
939	subf	r3,r7,r8
940	mtspr	SPRN_DEC,r3
941
942	ld	r5, VCPU_SPRG0(r4)
943	ld	r6, VCPU_SPRG1(r4)
944	ld	r7, VCPU_SPRG2(r4)
945	ld	r8, VCPU_SPRG3(r4)
946	mtspr	SPRN_SPRG0, r5
947	mtspr	SPRN_SPRG1, r6
948	mtspr	SPRN_SPRG2, r7
949	mtspr	SPRN_SPRG3, r8
950
951	/* Load up DAR and DSISR */
952	ld	r5, VCPU_DAR(r4)
953	lwz	r6, VCPU_DSISR(r4)
954	mtspr	SPRN_DAR, r5
955	mtspr	SPRN_DSISR, r6
956
957	/* Restore AMR and UAMOR, set AMOR to all 1s */
958	ld	r5,VCPU_AMR(r4)
959	ld	r6,VCPU_UAMOR(r4)
960	li	r7,-1
961	mtspr	SPRN_AMR,r5
962	mtspr	SPRN_UAMOR,r6
963	mtspr	SPRN_AMOR,r7
964
965	/* Restore state of CTRL run bit; assume 1 on entry */
966	lwz	r5,VCPU_CTRL(r4)
967	andi.	r5,r5,1
968	bne	4f
969	mfspr	r6,SPRN_CTRLF
970	clrrdi	r6,r6,1
971	mtspr	SPRN_CTRLT,r6
9724:
973	/* Secondary threads wait for primary to have done partition switch */
974	ld	r5, HSTATE_KVM_VCORE(r13)
975	lbz	r6, HSTATE_PTID(r13)
976	cmpwi	r6, 0
977	beq	21f
978	lbz	r0, VCORE_IN_GUEST(r5)
979	cmpwi	r0, 0
980	bne	21f
981	HMT_LOW
98220:	lwz	r3, VCORE_ENTRY_EXIT(r5)
983	cmpwi	r3, 0x100
984	bge	no_switch_exit
985	lbz	r0, VCORE_IN_GUEST(r5)
986	cmpwi	r0, 0
987	beq	20b
988	HMT_MEDIUM
98921:
990	/* Set LPCR. */
991	ld	r8,VCORE_LPCR(r5)
992	mtspr	SPRN_LPCR,r8
993	isync
994
995	/* Check if HDEC expires soon */
996	mfspr	r3, SPRN_HDEC
997	EXTEND_HDEC(r3)
998	cmpdi	r3, 512		/* 1 microsecond */
999	blt	hdec_soon
1000
1001	/* For hash guest, clear out and reload the SLB */
1002	ld	r6, VCPU_KVM(r4)
1003	lbz	r0, KVM_RADIX(r6)
1004	cmpwi	r0, 0
1005	bne	9f
1006	li	r6, 0
1007	slbmte	r6, r6
1008	slbia
1009	ptesync
1010
1011	/* Load up guest SLB entries (N.B. slb_max will be 0 for radix) */
1012	lwz	r5,VCPU_SLB_MAX(r4)
1013	cmpwi	r5,0
1014	beq	9f
1015	mtctr	r5
1016	addi	r6,r4,VCPU_SLB
10171:	ld	r8,VCPU_SLB_E(r6)
1018	ld	r9,VCPU_SLB_V(r6)
1019	slbmte	r9,r8
1020	addi	r6,r6,VCPU_SLB_SIZE
1021	bdnz	1b
10229:
1023
1024#ifdef CONFIG_KVM_XICS
1025	/* We are entering the guest on that thread, push VCPU to XIVE */
1026	ld	r10, HSTATE_XIVE_TIMA_PHYS(r13)
1027	cmpldi	cr0, r10, 0
1028	beq	no_xive
1029	ld	r11, VCPU_XIVE_SAVED_STATE(r4)
1030	li	r9, TM_QW1_OS
1031	eieio
1032	stdcix	r11,r9,r10
1033	lwz	r11, VCPU_XIVE_CAM_WORD(r4)
1034	li	r9, TM_QW1_OS + TM_WORD2
1035	stwcix	r11,r9,r10
1036	li	r9, 1
1037	stb	r9, VCPU_XIVE_PUSHED(r4)
1038	eieio
1039
1040	/*
1041	 * We clear the irq_pending flag. There is a small chance of a
1042	 * race vs. the escalation interrupt happening on another
1043	 * processor setting it again, but the only consequence is to
1044	 * cause a spurrious wakeup on the next H_CEDE which is not an
1045	 * issue.
1046	 */
1047	li	r0,0
1048	stb	r0, VCPU_IRQ_PENDING(r4)
1049
1050	/*
1051	 * In single escalation mode, if the escalation interrupt is
1052	 * on, we mask it.
1053	 */
1054	lbz	r0, VCPU_XIVE_ESC_ON(r4)
1055	cmpwi	r0,0
1056	beq	1f
1057	ld	r10, VCPU_XIVE_ESC_RADDR(r4)
1058	li	r9, XIVE_ESB_SET_PQ_01
1059	ldcix	r0, r10, r9
1060	sync
1061
1062	/* We have a possible subtle race here: The escalation interrupt might
1063	 * have fired and be on its way to the host queue while we mask it,
1064	 * and if we unmask it early enough (re-cede right away), there is
1065	 * a theorical possibility that it fires again, thus landing in the
1066	 * target queue more than once which is a big no-no.
1067	 *
1068	 * Fortunately, solving this is rather easy. If the above load setting
1069	 * PQ to 01 returns a previous value where P is set, then we know the
1070	 * escalation interrupt is somewhere on its way to the host. In that
1071	 * case we simply don't clear the xive_esc_on flag below. It will be
1072	 * eventually cleared by the handler for the escalation interrupt.
1073	 *
1074	 * Then, when doing a cede, we check that flag again before re-enabling
1075	 * the escalation interrupt, and if set, we abort the cede.
1076	 */
1077	andi.	r0, r0, XIVE_ESB_VAL_P
1078	bne-	1f
1079
1080	/* Now P is 0, we can clear the flag */
1081	li	r0, 0
1082	stb	r0, VCPU_XIVE_ESC_ON(r4)
10831:
1084no_xive:
1085#endif /* CONFIG_KVM_XICS */
1086
1087deliver_guest_interrupt:
1088	ld	r6, VCPU_CTR(r4)
1089	ld	r7, VCPU_XER(r4)
1090
1091	mtctr	r6
1092	mtxer	r7
1093
1094kvmppc_cede_reentry:		/* r4 = vcpu, r13 = paca */
1095	ld	r10, VCPU_PC(r4)
1096	ld	r11, VCPU_MSR(r4)
1097	ld	r6, VCPU_SRR0(r4)
1098	ld	r7, VCPU_SRR1(r4)
1099	mtspr	SPRN_SRR0, r6
1100	mtspr	SPRN_SRR1, r7
1101
1102	/* r11 = vcpu->arch.msr & ~MSR_HV */
1103	rldicl	r11, r11, 63 - MSR_HV_LG, 1
1104	rotldi	r11, r11, 1 + MSR_HV_LG
1105	ori	r11, r11, MSR_ME
1106
1107	/* Check if we can deliver an external or decrementer interrupt now */
1108	ld	r0, VCPU_PENDING_EXC(r4)
1109	rldicl	r0, r0, 64 - BOOK3S_IRQPRIO_EXTERNAL_LEVEL, 63
1110	cmpdi	cr1, r0, 0
1111	andi.	r8, r11, MSR_EE
1112	mfspr	r8, SPRN_LPCR
1113	/* Insert EXTERNAL_LEVEL bit into LPCR at the MER bit position */
1114	rldimi	r8, r0, LPCR_MER_SH, 63 - LPCR_MER_SH
1115	mtspr	SPRN_LPCR, r8
1116	isync
1117	beq	5f
1118	li	r0, BOOK3S_INTERRUPT_EXTERNAL
1119	bne	cr1, 12f
1120	mfspr	r0, SPRN_DEC
1121BEGIN_FTR_SECTION
1122	/* On POWER9 check whether the guest has large decrementer enabled */
1123	andis.	r8, r8, LPCR_LD@h
1124	bne	15f
1125END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1126	extsw	r0, r0
112715:	cmpdi	r0, 0
1128	li	r0, BOOK3S_INTERRUPT_DECREMENTER
1129	bge	5f
1130
113112:	mtspr	SPRN_SRR0, r10
1132	mr	r10,r0
1133	mtspr	SPRN_SRR1, r11
1134	mr	r9, r4
1135	bl	kvmppc_msr_interrupt
11365:
1137BEGIN_FTR_SECTION
1138	b	fast_guest_return
1139END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1140	/* On POWER9, check for pending doorbell requests */
1141	lbz	r0, VCPU_DBELL_REQ(r4)
1142	cmpwi	r0, 0
1143	beq	fast_guest_return
1144	ld	r5, HSTATE_KVM_VCORE(r13)
1145	/* Set DPDES register so the CPU will take a doorbell interrupt */
1146	li	r0, 1
1147	mtspr	SPRN_DPDES, r0
1148	std	r0, VCORE_DPDES(r5)
1149	/* Make sure other cpus see vcore->dpdes set before dbell req clear */
1150	lwsync
1151	/* Clear the pending doorbell request */
1152	li	r0, 0
1153	stb	r0, VCPU_DBELL_REQ(r4)
1154
1155/*
1156 * Required state:
1157 * R4 = vcpu
1158 * R10: value for HSRR0
1159 * R11: value for HSRR1
1160 * R13 = PACA
1161 */
1162fast_guest_return:
1163	li	r0,0
1164	stb	r0,VCPU_CEDED(r4)	/* cancel cede */
1165	mtspr	SPRN_HSRR0,r10
1166	mtspr	SPRN_HSRR1,r11
1167
1168	/* Activate guest mode, so faults get handled by KVM */
1169	li	r9, KVM_GUEST_MODE_GUEST_HV
1170	stb	r9, HSTATE_IN_GUEST(r13)
1171
1172#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1173	/* Accumulate timing */
1174	addi	r3, r4, VCPU_TB_GUEST
1175	bl	kvmhv_accumulate_time
1176#endif
1177
1178	/* Enter guest */
1179
1180BEGIN_FTR_SECTION
1181	ld	r5, VCPU_CFAR(r4)
1182	mtspr	SPRN_CFAR, r5
1183END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
1184BEGIN_FTR_SECTION
1185	ld	r0, VCPU_PPR(r4)
1186END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1187
1188	ld	r5, VCPU_LR(r4)
1189	lwz	r6, VCPU_CR(r4)
1190	mtlr	r5
1191	mtcr	r6
1192
1193	ld	r1, VCPU_GPR(R1)(r4)
1194	ld	r2, VCPU_GPR(R2)(r4)
1195	ld	r3, VCPU_GPR(R3)(r4)
1196	ld	r5, VCPU_GPR(R5)(r4)
1197	ld	r6, VCPU_GPR(R6)(r4)
1198	ld	r7, VCPU_GPR(R7)(r4)
1199	ld	r8, VCPU_GPR(R8)(r4)
1200	ld	r9, VCPU_GPR(R9)(r4)
1201	ld	r10, VCPU_GPR(R10)(r4)
1202	ld	r11, VCPU_GPR(R11)(r4)
1203	ld	r12, VCPU_GPR(R12)(r4)
1204	ld	r13, VCPU_GPR(R13)(r4)
1205
1206BEGIN_FTR_SECTION
1207	mtspr	SPRN_PPR, r0
1208END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1209
1210/* Move canary into DSISR to check for later */
1211BEGIN_FTR_SECTION
1212	li	r0, 0x7fff
1213	mtspr	SPRN_HDSISR, r0
1214END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1215
1216	ld	r0, VCPU_GPR(R0)(r4)
1217	ld	r4, VCPU_GPR(R4)(r4)
1218	HRFI_TO_GUEST
1219	b	.
1220
1221secondary_too_late:
1222	li	r12, 0
1223	cmpdi	r4, 0
1224	beq	11f
1225	stw	r12, VCPU_TRAP(r4)
1226#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1227	addi	r3, r4, VCPU_TB_RMEXIT
1228	bl	kvmhv_accumulate_time
1229#endif
123011:	b	kvmhv_switch_to_host
1231
1232no_switch_exit:
1233	HMT_MEDIUM
1234	li	r12, 0
1235	b	12f
1236hdec_soon:
1237	li	r12, BOOK3S_INTERRUPT_HV_DECREMENTER
123812:	stw	r12, VCPU_TRAP(r4)
1239	mr	r9, r4
1240#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1241	addi	r3, r4, VCPU_TB_RMEXIT
1242	bl	kvmhv_accumulate_time
1243#endif
1244	b	guest_bypass
1245
1246/******************************************************************************
1247 *                                                                            *
1248 *                               Exit code                                    *
1249 *                                                                            *
1250 *****************************************************************************/
1251
1252/*
1253 * We come here from the first-level interrupt handlers.
1254 */
1255	.globl	kvmppc_interrupt_hv
1256kvmppc_interrupt_hv:
1257	/*
1258	 * Register contents:
1259	 * R12		= (guest CR << 32) | interrupt vector
1260	 * R13		= PACA
1261	 * guest R12 saved in shadow VCPU SCRATCH0
1262	 * guest CTR saved in shadow VCPU SCRATCH1 if RELOCATABLE
1263	 * guest R13 saved in SPRN_SCRATCH0
1264	 */
1265	std	r9, HSTATE_SCRATCH2(r13)
1266	lbz	r9, HSTATE_IN_GUEST(r13)
1267	cmpwi	r9, KVM_GUEST_MODE_HOST_HV
1268	beq	kvmppc_bad_host_intr
1269#ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1270	cmpwi	r9, KVM_GUEST_MODE_GUEST
1271	ld	r9, HSTATE_SCRATCH2(r13)
1272	beq	kvmppc_interrupt_pr
1273#endif
1274	/* We're now back in the host but in guest MMU context */
1275	li	r9, KVM_GUEST_MODE_HOST_HV
1276	stb	r9, HSTATE_IN_GUEST(r13)
1277
1278	ld	r9, HSTATE_KVM_VCPU(r13)
1279
1280	/* Save registers */
1281
1282	std	r0, VCPU_GPR(R0)(r9)
1283	std	r1, VCPU_GPR(R1)(r9)
1284	std	r2, VCPU_GPR(R2)(r9)
1285	std	r3, VCPU_GPR(R3)(r9)
1286	std	r4, VCPU_GPR(R4)(r9)
1287	std	r5, VCPU_GPR(R5)(r9)
1288	std	r6, VCPU_GPR(R6)(r9)
1289	std	r7, VCPU_GPR(R7)(r9)
1290	std	r8, VCPU_GPR(R8)(r9)
1291	ld	r0, HSTATE_SCRATCH2(r13)
1292	std	r0, VCPU_GPR(R9)(r9)
1293	std	r10, VCPU_GPR(R10)(r9)
1294	std	r11, VCPU_GPR(R11)(r9)
1295	ld	r3, HSTATE_SCRATCH0(r13)
1296	std	r3, VCPU_GPR(R12)(r9)
1297	/* CR is in the high half of r12 */
1298	srdi	r4, r12, 32
1299	stw	r4, VCPU_CR(r9)
1300BEGIN_FTR_SECTION
1301	ld	r3, HSTATE_CFAR(r13)
1302	std	r3, VCPU_CFAR(r9)
1303END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
1304BEGIN_FTR_SECTION
1305	ld	r4, HSTATE_PPR(r13)
1306	std	r4, VCPU_PPR(r9)
1307END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1308
1309	/* Restore R1/R2 so we can handle faults */
1310	ld	r1, HSTATE_HOST_R1(r13)
1311	ld	r2, PACATOC(r13)
1312
1313	mfspr	r10, SPRN_SRR0
1314	mfspr	r11, SPRN_SRR1
1315	std	r10, VCPU_SRR0(r9)
1316	std	r11, VCPU_SRR1(r9)
1317	/* trap is in the low half of r12, clear CR from the high half */
1318	clrldi	r12, r12, 32
1319	andi.	r0, r12, 2		/* need to read HSRR0/1? */
1320	beq	1f
1321	mfspr	r10, SPRN_HSRR0
1322	mfspr	r11, SPRN_HSRR1
1323	clrrdi	r12, r12, 2
13241:	std	r10, VCPU_PC(r9)
1325	std	r11, VCPU_MSR(r9)
1326
1327	GET_SCRATCH0(r3)
1328	mflr	r4
1329	std	r3, VCPU_GPR(R13)(r9)
1330	std	r4, VCPU_LR(r9)
1331
1332	stw	r12,VCPU_TRAP(r9)
1333
1334	/*
1335	 * Now that we have saved away SRR0/1 and HSRR0/1,
1336	 * interrupts are recoverable in principle, so set MSR_RI.
1337	 * This becomes important for relocation-on interrupts from
1338	 * the guest, which we can get in radix mode on POWER9.
1339	 */
1340	li	r0, MSR_RI
1341	mtmsrd	r0, 1
1342
1343#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1344	addi	r3, r9, VCPU_TB_RMINTR
1345	mr	r4, r9
1346	bl	kvmhv_accumulate_time
1347	ld	r5, VCPU_GPR(R5)(r9)
1348	ld	r6, VCPU_GPR(R6)(r9)
1349	ld	r7, VCPU_GPR(R7)(r9)
1350	ld	r8, VCPU_GPR(R8)(r9)
1351#endif
1352
1353	/* Save HEIR (HV emulation assist reg) in emul_inst
1354	   if this is an HEI (HV emulation interrupt, e40) */
1355	li	r3,KVM_INST_FETCH_FAILED
1356	stw	r3,VCPU_LAST_INST(r9)
1357	cmpwi	r12,BOOK3S_INTERRUPT_H_EMUL_ASSIST
1358	bne	11f
1359	mfspr	r3,SPRN_HEIR
136011:	stw	r3,VCPU_HEIR(r9)
1361
1362	/* these are volatile across C function calls */
1363#ifdef CONFIG_RELOCATABLE
1364	ld	r3, HSTATE_SCRATCH1(r13)
1365	mtctr	r3
1366#else
1367	mfctr	r3
1368#endif
1369	mfxer	r4
1370	std	r3, VCPU_CTR(r9)
1371	std	r4, VCPU_XER(r9)
1372
1373	/* If this is a page table miss then see if it's theirs or ours */
1374	cmpwi	r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
1375	beq	kvmppc_hdsi
1376	cmpwi	r12, BOOK3S_INTERRUPT_H_INST_STORAGE
1377	beq	kvmppc_hisi
1378
1379	/* See if this is a leftover HDEC interrupt */
1380	cmpwi	r12,BOOK3S_INTERRUPT_HV_DECREMENTER
1381	bne	2f
1382	mfspr	r3,SPRN_HDEC
1383	EXTEND_HDEC(r3)
1384	cmpdi	r3,0
1385	mr	r4,r9
1386	bge	fast_guest_return
13872:
1388	/* See if this is an hcall we can handle in real mode */
1389	cmpwi	r12,BOOK3S_INTERRUPT_SYSCALL
1390	beq	hcall_try_real_mode
1391
1392	/* Hypervisor doorbell - exit only if host IPI flag set */
1393	cmpwi	r12, BOOK3S_INTERRUPT_H_DOORBELL
1394	bne	3f
1395BEGIN_FTR_SECTION
1396	PPC_MSGSYNC
1397	lwsync
1398END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1399	lbz	r0, HSTATE_HOST_IPI(r13)
1400	cmpwi	r0, 0
1401	beq	4f
1402	b	guest_exit_cont
14033:
1404	/* If it's a hypervisor facility unavailable interrupt, save HFSCR */
1405	cmpwi	r12, BOOK3S_INTERRUPT_H_FAC_UNAVAIL
1406	bne	14f
1407	mfspr	r3, SPRN_HFSCR
1408	std	r3, VCPU_HFSCR(r9)
1409	b	guest_exit_cont
141014:
1411	/* External interrupt ? */
1412	cmpwi	r12, BOOK3S_INTERRUPT_EXTERNAL
1413	bne+	guest_exit_cont
1414
1415	/* External interrupt, first check for host_ipi. If this is
1416	 * set, we know the host wants us out so let's do it now
1417	 */
1418	bl	kvmppc_read_intr
1419
1420	/*
1421	 * Restore the active volatile registers after returning from
1422	 * a C function.
1423	 */
1424	ld	r9, HSTATE_KVM_VCPU(r13)
1425	li	r12, BOOK3S_INTERRUPT_EXTERNAL
1426
1427	/*
1428	 * kvmppc_read_intr return codes:
1429	 *
1430	 * Exit to host (r3 > 0)
1431	 *   1 An interrupt is pending that needs to be handled by the host
1432	 *     Exit guest and return to host by branching to guest_exit_cont
1433	 *
1434	 *   2 Passthrough that needs completion in the host
1435	 *     Exit guest and return to host by branching to guest_exit_cont
1436	 *     However, we also set r12 to BOOK3S_INTERRUPT_HV_RM_HARD
1437	 *     to indicate to the host to complete handling the interrupt
1438	 *
1439	 * Before returning to guest, we check if any CPU is heading out
1440	 * to the host and if so, we head out also. If no CPUs are heading
1441	 * check return values <= 0.
1442	 *
1443	 * Return to guest (r3 <= 0)
1444	 *  0 No external interrupt is pending
1445	 * -1 A guest wakeup IPI (which has now been cleared)
1446	 *    In either case, we return to guest to deliver any pending
1447	 *    guest interrupts.
1448	 *
1449	 * -2 A PCI passthrough external interrupt was handled
1450	 *    (interrupt was delivered directly to guest)
1451	 *    Return to guest to deliver any pending guest interrupts.
1452	 */
1453
1454	cmpdi	r3, 1
1455	ble	1f
1456
1457	/* Return code = 2 */
1458	li	r12, BOOK3S_INTERRUPT_HV_RM_HARD
1459	stw	r12, VCPU_TRAP(r9)
1460	b	guest_exit_cont
1461
14621:	/* Return code <= 1 */
1463	cmpdi	r3, 0
1464	bgt	guest_exit_cont
1465
1466	/* Return code <= 0 */
14674:	ld	r5, HSTATE_KVM_VCORE(r13)
1468	lwz	r0, VCORE_ENTRY_EXIT(r5)
1469	cmpwi	r0, 0x100
1470	mr	r4, r9
1471	blt	deliver_guest_interrupt
1472
1473guest_exit_cont:		/* r9 = vcpu, r12 = trap, r13 = paca */
1474	/* Save more register state  */
1475	mfdar	r6
1476	mfdsisr	r7
1477	std	r6, VCPU_DAR(r9)
1478	stw	r7, VCPU_DSISR(r9)
1479	/* don't overwrite fault_dar/fault_dsisr if HDSI */
1480	cmpwi	r12,BOOK3S_INTERRUPT_H_DATA_STORAGE
1481	beq	mc_cont
1482	std	r6, VCPU_FAULT_DAR(r9)
1483	stw	r7, VCPU_FAULT_DSISR(r9)
1484
1485	/* See if it is a machine check */
1486	cmpwi	r12, BOOK3S_INTERRUPT_MACHINE_CHECK
1487	beq	machine_check_realmode
1488mc_cont:
1489#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1490	addi	r3, r9, VCPU_TB_RMEXIT
1491	mr	r4, r9
1492	bl	kvmhv_accumulate_time
1493#endif
1494#ifdef CONFIG_KVM_XICS
1495	/* We are exiting, pull the VP from the XIVE */
1496	lbz	r0, VCPU_XIVE_PUSHED(r9)
1497	cmpwi	cr0, r0, 0
1498	beq	1f
1499	li	r7, TM_SPC_PULL_OS_CTX
1500	li	r6, TM_QW1_OS
1501	mfmsr	r0
1502	andi.	r0, r0, MSR_DR		/* in real mode? */
1503	beq	2f
1504	ld	r10, HSTATE_XIVE_TIMA_VIRT(r13)
1505	cmpldi	cr0, r10, 0
1506	beq	1f
1507	/* First load to pull the context, we ignore the value */
1508	eieio
1509	lwzx	r11, r7, r10
1510	/* Second load to recover the context state (Words 0 and 1) */
1511	ldx	r11, r6, r10
1512	b	3f
15132:	ld	r10, HSTATE_XIVE_TIMA_PHYS(r13)
1514	cmpldi	cr0, r10, 0
1515	beq	1f
1516	/* First load to pull the context, we ignore the value */
1517	eieio
1518	lwzcix	r11, r7, r10
1519	/* Second load to recover the context state (Words 0 and 1) */
1520	ldcix	r11, r6, r10
15213:	std	r11, VCPU_XIVE_SAVED_STATE(r9)
1522	/* Fixup some of the state for the next load */
1523	li	r10, 0
1524	li	r0, 0xff
1525	stb	r10, VCPU_XIVE_PUSHED(r9)
1526	stb	r10, (VCPU_XIVE_SAVED_STATE+3)(r9)
1527	stb	r0, (VCPU_XIVE_SAVED_STATE+4)(r9)
1528	eieio
15291:
1530#endif /* CONFIG_KVM_XICS */
1531
1532	/* For hash guest, read the guest SLB and save it away */
1533	ld	r5, VCPU_KVM(r9)
1534	lbz	r0, KVM_RADIX(r5)
1535	li	r5, 0
1536	cmpwi	r0, 0
1537	bne	3f			/* for radix, save 0 entries */
1538	lwz	r0,VCPU_SLB_NR(r9)	/* number of entries in SLB */
1539	mtctr	r0
1540	li	r6,0
1541	addi	r7,r9,VCPU_SLB
15421:	slbmfee	r8,r6
1543	andis.	r0,r8,SLB_ESID_V@h
1544	beq	2f
1545	add	r8,r8,r6		/* put index in */
1546	slbmfev	r3,r6
1547	std	r8,VCPU_SLB_E(r7)
1548	std	r3,VCPU_SLB_V(r7)
1549	addi	r7,r7,VCPU_SLB_SIZE
1550	addi	r5,r5,1
15512:	addi	r6,r6,1
1552	bdnz	1b
1553	/* Finally clear out the SLB */
1554	li	r0,0
1555	slbmte	r0,r0
1556	slbia
1557	ptesync
15583:	stw	r5,VCPU_SLB_MAX(r9)
1559
1560guest_bypass:
1561	mr 	r3, r12
1562	/* Increment exit count, poke other threads to exit */
1563	bl	kvmhv_commence_exit
1564	nop
1565	ld	r9, HSTATE_KVM_VCPU(r13)
1566	lwz	r12, VCPU_TRAP(r9)
1567
1568	/* Stop others sending VCPU interrupts to this physical CPU */
1569	li	r0, -1
1570	stw	r0, VCPU_CPU(r9)
1571	stw	r0, VCPU_THREAD_CPU(r9)
1572
1573	/* Save guest CTRL register, set runlatch to 1 */
1574	mfspr	r6,SPRN_CTRLF
1575	stw	r6,VCPU_CTRL(r9)
1576	andi.	r0,r6,1
1577	bne	4f
1578	ori	r6,r6,1
1579	mtspr	SPRN_CTRLT,r6
15804:
1581	/*
1582	 * Save the guest PURR/SPURR
1583	 */
1584	mfspr	r5,SPRN_PURR
1585	mfspr	r6,SPRN_SPURR
1586	ld	r7,VCPU_PURR(r9)
1587	ld	r8,VCPU_SPURR(r9)
1588	std	r5,VCPU_PURR(r9)
1589	std	r6,VCPU_SPURR(r9)
1590	subf	r5,r7,r5
1591	subf	r6,r8,r6
1592
1593	/*
1594	 * Restore host PURR/SPURR and add guest times
1595	 * so that the time in the guest gets accounted.
1596	 */
1597	ld	r3,HSTATE_PURR(r13)
1598	ld	r4,HSTATE_SPURR(r13)
1599	add	r3,r3,r5
1600	add	r4,r4,r6
1601	mtspr	SPRN_PURR,r3
1602	mtspr	SPRN_SPURR,r4
1603
1604	/* Save DEC */
1605	ld	r3, HSTATE_KVM_VCORE(r13)
1606	mfspr	r5,SPRN_DEC
1607	mftb	r6
1608	/* On P9, if the guest has large decr enabled, don't sign extend */
1609BEGIN_FTR_SECTION
1610	ld	r4, VCORE_LPCR(r3)
1611	andis.	r4, r4, LPCR_LD@h
1612	bne	16f
1613END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1614	extsw	r5,r5
161516:	add	r5,r5,r6
1616	/* r5 is a guest timebase value here, convert to host TB */
1617	ld	r4,VCORE_TB_OFFSET(r3)
1618	subf	r5,r4,r5
1619	std	r5,VCPU_DEC_EXPIRES(r9)
1620
1621BEGIN_FTR_SECTION
1622	b	8f
1623END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
1624	/* Save POWER8-specific registers */
1625	mfspr	r5, SPRN_IAMR
1626	mfspr	r6, SPRN_PSPB
1627	mfspr	r7, SPRN_FSCR
1628	std	r5, VCPU_IAMR(r9)
1629	stw	r6, VCPU_PSPB(r9)
1630	std	r7, VCPU_FSCR(r9)
1631	mfspr	r5, SPRN_IC
1632	mfspr	r7, SPRN_TAR
1633	std	r5, VCPU_IC(r9)
1634	std	r7, VCPU_TAR(r9)
1635	mfspr	r8, SPRN_EBBHR
1636	std	r8, VCPU_EBBHR(r9)
1637	mfspr	r5, SPRN_EBBRR
1638	mfspr	r6, SPRN_BESCR
1639	mfspr	r7, SPRN_PID
1640	mfspr	r8, SPRN_WORT
1641	std	r5, VCPU_EBBRR(r9)
1642	std	r6, VCPU_BESCR(r9)
1643	stw	r7, VCPU_GUEST_PID(r9)
1644	std	r8, VCPU_WORT(r9)
1645BEGIN_FTR_SECTION
1646	mfspr	r5, SPRN_TCSCR
1647	mfspr	r6, SPRN_ACOP
1648	mfspr	r7, SPRN_CSIGR
1649	mfspr	r8, SPRN_TACR
1650	std	r5, VCPU_TCSCR(r9)
1651	std	r6, VCPU_ACOP(r9)
1652	std	r7, VCPU_CSIGR(r9)
1653	std	r8, VCPU_TACR(r9)
1654FTR_SECTION_ELSE
1655	mfspr	r5, SPRN_TIDR
1656	mfspr	r6, SPRN_PSSCR
1657	std	r5, VCPU_TID(r9)
1658	rldicl	r6, r6, 4, 50		/* r6 &= PSSCR_GUEST_VIS */
1659	rotldi	r6, r6, 60
1660	std	r6, VCPU_PSSCR(r9)
1661	/* Restore host HFSCR value */
1662	ld	r7, STACK_SLOT_HFSCR(r1)
1663	mtspr	SPRN_HFSCR, r7
1664ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
1665	/*
1666	 * Restore various registers to 0, where non-zero values
1667	 * set by the guest could disrupt the host.
1668	 */
1669	li	r0, 0
1670	mtspr	SPRN_PSPB, r0
1671	mtspr	SPRN_WORT, r0
1672BEGIN_FTR_SECTION
1673	mtspr	SPRN_IAMR, r0
1674	mtspr	SPRN_TCSCR, r0
1675	/* Set MMCRS to 1<<31 to freeze and disable the SPMC counters */
1676	li	r0, 1
1677	sldi	r0, r0, 31
1678	mtspr	SPRN_MMCRS, r0
1679END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
16808:
1681
1682	/* Save and reset AMR and UAMOR before turning on the MMU */
1683	mfspr	r5,SPRN_AMR
1684	mfspr	r6,SPRN_UAMOR
1685	std	r5,VCPU_AMR(r9)
1686	std	r6,VCPU_UAMOR(r9)
1687	li	r6,0
1688	mtspr	SPRN_AMR,r6
1689	mtspr	SPRN_UAMOR, r6
1690
1691	/* Switch DSCR back to host value */
1692	mfspr	r8, SPRN_DSCR
1693	ld	r7, HSTATE_DSCR(r13)
1694	std	r8, VCPU_DSCR(r9)
1695	mtspr	SPRN_DSCR, r7
1696
1697	/* Save non-volatile GPRs */
1698	std	r14, VCPU_GPR(R14)(r9)
1699	std	r15, VCPU_GPR(R15)(r9)
1700	std	r16, VCPU_GPR(R16)(r9)
1701	std	r17, VCPU_GPR(R17)(r9)
1702	std	r18, VCPU_GPR(R18)(r9)
1703	std	r19, VCPU_GPR(R19)(r9)
1704	std	r20, VCPU_GPR(R20)(r9)
1705	std	r21, VCPU_GPR(R21)(r9)
1706	std	r22, VCPU_GPR(R22)(r9)
1707	std	r23, VCPU_GPR(R23)(r9)
1708	std	r24, VCPU_GPR(R24)(r9)
1709	std	r25, VCPU_GPR(R25)(r9)
1710	std	r26, VCPU_GPR(R26)(r9)
1711	std	r27, VCPU_GPR(R27)(r9)
1712	std	r28, VCPU_GPR(R28)(r9)
1713	std	r29, VCPU_GPR(R29)(r9)
1714	std	r30, VCPU_GPR(R30)(r9)
1715	std	r31, VCPU_GPR(R31)(r9)
1716
1717	/* Save SPRGs */
1718	mfspr	r3, SPRN_SPRG0
1719	mfspr	r4, SPRN_SPRG1
1720	mfspr	r5, SPRN_SPRG2
1721	mfspr	r6, SPRN_SPRG3
1722	std	r3, VCPU_SPRG0(r9)
1723	std	r4, VCPU_SPRG1(r9)
1724	std	r5, VCPU_SPRG2(r9)
1725	std	r6, VCPU_SPRG3(r9)
1726
1727	/* save FP state */
1728	mr	r3, r9
1729	bl	kvmppc_save_fp
1730
1731#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1732BEGIN_FTR_SECTION
1733	/*
1734	 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS INCLUDING CR
1735	 */
1736	bl	kvmppc_save_tm
1737END_FTR_SECTION_IFSET(CPU_FTR_TM)
1738#endif
1739
1740	/* Increment yield count if they have a VPA */
1741	ld	r8, VCPU_VPA(r9)	/* do they have a VPA? */
1742	cmpdi	r8, 0
1743	beq	25f
1744	li	r4, LPPACA_YIELDCOUNT
1745	LWZX_BE	r3, r8, r4
1746	addi	r3, r3, 1
1747	STWX_BE	r3, r8, r4
1748	li	r3, 1
1749	stb	r3, VCPU_VPA_DIRTY(r9)
175025:
1751	/* Save PMU registers if requested */
1752	/* r8 and cr0.eq are live here */
1753BEGIN_FTR_SECTION
1754	/*
1755	 * POWER8 seems to have a hardware bug where setting
1756	 * MMCR0[PMAE] along with MMCR0[PMC1CE] and/or MMCR0[PMCjCE]
1757	 * when some counters are already negative doesn't seem
1758	 * to cause a performance monitor alert (and hence interrupt).
1759	 * The effect of this is that when saving the PMU state,
1760	 * if there is no PMU alert pending when we read MMCR0
1761	 * before freezing the counters, but one becomes pending
1762	 * before we read the counters, we lose it.
1763	 * To work around this, we need a way to freeze the counters
1764	 * before reading MMCR0.  Normally, freezing the counters
1765	 * is done by writing MMCR0 (to set MMCR0[FC]) which
1766	 * unavoidably writes MMCR0[PMA0] as well.  On POWER8,
1767	 * we can also freeze the counters using MMCR2, by writing
1768	 * 1s to all the counter freeze condition bits (there are
1769	 * 9 bits each for 6 counters).
1770	 */
1771	li	r3, -1			/* set all freeze bits */
1772	clrrdi	r3, r3, 10
1773	mfspr	r10, SPRN_MMCR2
1774	mtspr	SPRN_MMCR2, r3
1775	isync
1776END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1777	li	r3, 1
1778	sldi	r3, r3, 31		/* MMCR0_FC (freeze counters) bit */
1779	mfspr	r4, SPRN_MMCR0		/* save MMCR0 */
1780	mtspr	SPRN_MMCR0, r3		/* freeze all counters, disable ints */
1781	mfspr	r6, SPRN_MMCRA
1782	/* Clear MMCRA in order to disable SDAR updates */
1783	li	r7, 0
1784	mtspr	SPRN_MMCRA, r7
1785	isync
1786	beq	21f			/* if no VPA, save PMU stuff anyway */
1787	lbz	r7, LPPACA_PMCINUSE(r8)
1788	cmpwi	r7, 0			/* did they ask for PMU stuff to be saved? */
1789	bne	21f
1790	std	r3, VCPU_MMCR(r9)	/* if not, set saved MMCR0 to FC */
1791	b	22f
179221:	mfspr	r5, SPRN_MMCR1
1793	mfspr	r7, SPRN_SIAR
1794	mfspr	r8, SPRN_SDAR
1795	std	r4, VCPU_MMCR(r9)
1796	std	r5, VCPU_MMCR + 8(r9)
1797	std	r6, VCPU_MMCR + 16(r9)
1798BEGIN_FTR_SECTION
1799	std	r10, VCPU_MMCR + 24(r9)
1800END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1801	std	r7, VCPU_SIAR(r9)
1802	std	r8, VCPU_SDAR(r9)
1803	mfspr	r3, SPRN_PMC1
1804	mfspr	r4, SPRN_PMC2
1805	mfspr	r5, SPRN_PMC3
1806	mfspr	r6, SPRN_PMC4
1807	mfspr	r7, SPRN_PMC5
1808	mfspr	r8, SPRN_PMC6
1809	stw	r3, VCPU_PMC(r9)
1810	stw	r4, VCPU_PMC + 4(r9)
1811	stw	r5, VCPU_PMC + 8(r9)
1812	stw	r6, VCPU_PMC + 12(r9)
1813	stw	r7, VCPU_PMC + 16(r9)
1814	stw	r8, VCPU_PMC + 20(r9)
1815BEGIN_FTR_SECTION
1816	mfspr	r5, SPRN_SIER
1817	std	r5, VCPU_SIER(r9)
1818BEGIN_FTR_SECTION_NESTED(96)
1819	mfspr	r6, SPRN_SPMC1
1820	mfspr	r7, SPRN_SPMC2
1821	mfspr	r8, SPRN_MMCRS
1822	stw	r6, VCPU_PMC + 24(r9)
1823	stw	r7, VCPU_PMC + 28(r9)
1824	std	r8, VCPU_MMCR + 32(r9)
1825	lis	r4, 0x8000
1826	mtspr	SPRN_MMCRS, r4
1827END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
1828END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
182922:
1830
1831	/* Restore host values of some registers */
1832BEGIN_FTR_SECTION
1833	ld	r5, STACK_SLOT_CIABR(r1)
1834	ld	r6, STACK_SLOT_DAWR(r1)
1835	ld	r7, STACK_SLOT_DAWRX(r1)
1836	mtspr	SPRN_CIABR, r5
1837	mtspr	SPRN_DAWR, r6
1838	mtspr	SPRN_DAWRX, r7
1839END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1840BEGIN_FTR_SECTION
1841	ld	r5, STACK_SLOT_TID(r1)
1842	ld	r6, STACK_SLOT_PSSCR(r1)
1843	ld	r7, STACK_SLOT_PID(r1)
1844	ld	r8, STACK_SLOT_IAMR(r1)
1845	mtspr	SPRN_TIDR, r5
1846	mtspr	SPRN_PSSCR, r6
1847	mtspr	SPRN_PID, r7
1848	mtspr	SPRN_IAMR, r8
1849END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1850
1851#ifdef CONFIG_PPC_RADIX_MMU
1852	/*
1853	 * Are we running hash or radix ?
1854	 */
1855	ld	r5, VCPU_KVM(r9)
1856	lbz	r0, KVM_RADIX(r5)
1857	cmpwi	cr2, r0, 0
1858	beq	cr2, 4f
1859
1860	/* Radix: Handle the case where the guest used an illegal PID */
1861	LOAD_REG_ADDR(r4, mmu_base_pid)
1862	lwz	r3, VCPU_GUEST_PID(r9)
1863	lwz	r5, 0(r4)
1864	cmpw	cr0,r3,r5
1865	blt	2f
1866
1867	/*
1868	 * Illegal PID, the HW might have prefetched and cached in the TLB
1869	 * some translations for the  LPID 0 / guest PID combination which
1870	 * Linux doesn't know about, so we need to flush that PID out of
1871	 * the TLB. First we need to set LPIDR to 0 so tlbiel applies to
1872	 * the right context.
1873	*/
1874	li	r0,0
1875	mtspr	SPRN_LPID,r0
1876	isync
1877
1878	/* Then do a congruence class local flush */
1879	ld	r6,VCPU_KVM(r9)
1880	lwz	r0,KVM_TLB_SETS(r6)
1881	mtctr	r0
1882	li	r7,0x400		/* IS field = 0b01 */
1883	ptesync
1884	sldi	r0,r3,32		/* RS has PID */
18851:	PPC_TLBIEL(7,0,2,1,1)		/* RIC=2, PRS=1, R=1 */
1886	addi	r7,r7,0x1000
1887	bdnz	1b
1888	ptesync
1889
18902:	/* Flush the ERAT on radix P9 DD1 guest exit */
1891BEGIN_FTR_SECTION
1892	PPC_INVALIDATE_ERAT
1893END_FTR_SECTION_IFSET(CPU_FTR_POWER9_DD1)
18944:
1895#endif /* CONFIG_PPC_RADIX_MMU */
1896
1897	/*
1898	 * POWER7/POWER8 guest -> host partition switch code.
1899	 * We don't have to lock against tlbies but we do
1900	 * have to coordinate the hardware threads.
1901	 */
1902kvmhv_switch_to_host:
1903	/* Secondary threads wait for primary to do partition switch */
1904	ld	r5,HSTATE_KVM_VCORE(r13)
1905	ld	r4,VCORE_KVM(r5)	/* pointer to struct kvm */
1906	lbz	r3,HSTATE_PTID(r13)
1907	cmpwi	r3,0
1908	beq	15f
1909	HMT_LOW
191013:	lbz	r3,VCORE_IN_GUEST(r5)
1911	cmpwi	r3,0
1912	bne	13b
1913	HMT_MEDIUM
1914	b	16f
1915
1916	/* Primary thread waits for all the secondaries to exit guest */
191715:	lwz	r3,VCORE_ENTRY_EXIT(r5)
1918	rlwinm	r0,r3,32-8,0xff
1919	clrldi	r3,r3,56
1920	cmpw	r3,r0
1921	bne	15b
1922	isync
1923
1924	/* Did we actually switch to the guest at all? */
1925	lbz	r6, VCORE_IN_GUEST(r5)
1926	cmpwi	r6, 0
1927	beq	19f
1928
1929	/* Primary thread switches back to host partition */
1930	lwz	r7,KVM_HOST_LPID(r4)
1931BEGIN_FTR_SECTION
1932	ld	r6,KVM_HOST_SDR1(r4)
1933	li	r8,LPID_RSVD		/* switch to reserved LPID */
1934	mtspr	SPRN_LPID,r8
1935	ptesync
1936	mtspr	SPRN_SDR1,r6		/* switch to host page table */
1937END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1938	mtspr	SPRN_LPID,r7
1939	isync
1940
1941BEGIN_FTR_SECTION
1942	/* DPDES and VTB are shared between threads */
1943	mfspr	r7, SPRN_DPDES
1944	mfspr	r8, SPRN_VTB
1945	std	r7, VCORE_DPDES(r5)
1946	std	r8, VCORE_VTB(r5)
1947	/* clear DPDES so we don't get guest doorbells in the host */
1948	li	r8, 0
1949	mtspr	SPRN_DPDES, r8
1950END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1951
1952	/* If HMI, call kvmppc_realmode_hmi_handler() */
1953	cmpwi	r12, BOOK3S_INTERRUPT_HMI
1954	bne	27f
1955	bl	kvmppc_realmode_hmi_handler
1956	nop
1957	cmpdi	r3, 0
1958	li	r12, BOOK3S_INTERRUPT_HMI
1959	/*
1960	 * At this point kvmppc_realmode_hmi_handler may have resync-ed
1961	 * the TB, and if it has, we must not subtract the guest timebase
1962	 * offset from the timebase. So, skip it.
1963	 *
1964	 * Also, do not call kvmppc_subcore_exit_guest() because it has
1965	 * been invoked as part of kvmppc_realmode_hmi_handler().
1966	 */
1967	beq	30f
1968
196927:
1970	/* Subtract timebase offset from timebase */
1971	ld	r8,VCORE_TB_OFFSET(r5)
1972	cmpdi	r8,0
1973	beq	17f
1974	mftb	r6			/* current guest timebase */
1975	subf	r8,r8,r6
1976	mtspr	SPRN_TBU40,r8		/* update upper 40 bits */
1977	mftb	r7			/* check if lower 24 bits overflowed */
1978	clrldi	r6,r6,40
1979	clrldi	r7,r7,40
1980	cmpld	r7,r6
1981	bge	17f
1982	addis	r8,r8,0x100		/* if so, increment upper 40 bits */
1983	mtspr	SPRN_TBU40,r8
1984
198517:	bl	kvmppc_subcore_exit_guest
1986	nop
198730:	ld	r5,HSTATE_KVM_VCORE(r13)
1988	ld	r4,VCORE_KVM(r5)	/* pointer to struct kvm */
1989
1990	/* Reset PCR */
1991	ld	r0, VCORE_PCR(r5)
1992	cmpdi	r0, 0
1993	beq	18f
1994	li	r0, 0
1995	mtspr	SPRN_PCR, r0
199618:
1997	/* Signal secondary CPUs to continue */
1998	stb	r0,VCORE_IN_GUEST(r5)
199919:	lis	r8,0x7fff		/* MAX_INT@h */
2000	mtspr	SPRN_HDEC,r8
2001
200216:
2003BEGIN_FTR_SECTION
2004	/* On POWER9 with HPT-on-radix we need to wait for all other threads */
2005	ld	r3, HSTATE_SPLIT_MODE(r13)
2006	cmpdi	r3, 0
2007	beq	47f
2008	lwz	r8, KVM_SPLIT_DO_RESTORE(r3)
2009	cmpwi	r8, 0
2010	beq	47f
2011	stw	r12, STACK_SLOT_TRAP(r1)
2012	bl	kvmhv_p9_restore_lpcr
2013	nop
2014	lwz	r12, STACK_SLOT_TRAP(r1)
2015	b	48f
201647:
2017END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2018	ld	r8,KVM_HOST_LPCR(r4)
2019	mtspr	SPRN_LPCR,r8
2020	isync
202148:
2022	/* load host SLB entries */
2023BEGIN_MMU_FTR_SECTION
2024	b	0f
2025END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
2026	ld	r8,PACA_SLBSHADOWPTR(r13)
2027
2028	.rept	SLB_NUM_BOLTED
2029	li	r3, SLBSHADOW_SAVEAREA
2030	LDX_BE	r5, r8, r3
2031	addi	r3, r3, 8
2032	LDX_BE	r6, r8, r3
2033	andis.	r7,r5,SLB_ESID_V@h
2034	beq	1f
2035	slbmte	r6,r5
20361:	addi	r8,r8,16
2037	.endr
20380:
2039#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2040	/* Finish timing, if we have a vcpu */
2041	ld	r4, HSTATE_KVM_VCPU(r13)
2042	cmpdi	r4, 0
2043	li	r3, 0
2044	beq	2f
2045	bl	kvmhv_accumulate_time
20462:
2047#endif
2048	/* Unset guest mode */
2049	li	r0, KVM_GUEST_MODE_NONE
2050	stb	r0, HSTATE_IN_GUEST(r13)
2051
2052	ld	r0, SFS+PPC_LR_STKOFF(r1)
2053	addi	r1, r1, SFS
2054	mtlr	r0
2055	blr
2056
2057/*
2058 * Check whether an HDSI is an HPTE not found fault or something else.
2059 * If it is an HPTE not found fault that is due to the guest accessing
2060 * a page that they have mapped but which we have paged out, then
2061 * we continue on with the guest exit path.  In all other cases,
2062 * reflect the HDSI to the guest as a DSI.
2063 */
2064kvmppc_hdsi:
2065	ld	r3, VCPU_KVM(r9)
2066	lbz	r0, KVM_RADIX(r3)
2067	mfspr	r4, SPRN_HDAR
2068	mfspr	r6, SPRN_HDSISR
2069BEGIN_FTR_SECTION
2070	/* Look for DSISR canary. If we find it, retry instruction */
2071	cmpdi	r6, 0x7fff
2072	beq	6f
2073END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2074	cmpwi	r0, 0
2075	bne	.Lradix_hdsi		/* on radix, just save DAR/DSISR/ASDR */
2076	/* HPTE not found fault or protection fault? */
2077	andis.	r0, r6, (DSISR_NOHPTE | DSISR_PROTFAULT)@h
2078	beq	1f			/* if not, send it to the guest */
2079	andi.	r0, r11, MSR_DR		/* data relocation enabled? */
2080	beq	3f
2081BEGIN_FTR_SECTION
2082	mfspr	r5, SPRN_ASDR		/* on POWER9, use ASDR to get VSID */
2083	b	4f
2084END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2085	clrrdi	r0, r4, 28
2086	PPC_SLBFEE_DOT(R5, R0)		/* if so, look up SLB */
2087	li	r0, BOOK3S_INTERRUPT_DATA_SEGMENT
2088	bne	7f			/* if no SLB entry found */
20894:	std	r4, VCPU_FAULT_DAR(r9)
2090	stw	r6, VCPU_FAULT_DSISR(r9)
2091
2092	/* Search the hash table. */
2093	mr	r3, r9			/* vcpu pointer */
2094	li	r7, 1			/* data fault */
2095	bl	kvmppc_hpte_hv_fault
2096	ld	r9, HSTATE_KVM_VCPU(r13)
2097	ld	r10, VCPU_PC(r9)
2098	ld	r11, VCPU_MSR(r9)
2099	li	r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
2100	cmpdi	r3, 0			/* retry the instruction */
2101	beq	6f
2102	cmpdi	r3, -1			/* handle in kernel mode */
2103	beq	guest_exit_cont
2104	cmpdi	r3, -2			/* MMIO emulation; need instr word */
2105	beq	2f
2106
2107	/* Synthesize a DSI (or DSegI) for the guest */
2108	ld	r4, VCPU_FAULT_DAR(r9)
2109	mr	r6, r3
21101:	li	r0, BOOK3S_INTERRUPT_DATA_STORAGE
2111	mtspr	SPRN_DSISR, r6
21127:	mtspr	SPRN_DAR, r4
2113	mtspr	SPRN_SRR0, r10
2114	mtspr	SPRN_SRR1, r11
2115	mr	r10, r0
2116	bl	kvmppc_msr_interrupt
2117fast_interrupt_c_return:
21186:	ld	r7, VCPU_CTR(r9)
2119	ld	r8, VCPU_XER(r9)
2120	mtctr	r7
2121	mtxer	r8
2122	mr	r4, r9
2123	b	fast_guest_return
2124
21253:	ld	r5, VCPU_KVM(r9)	/* not relocated, use VRMA */
2126	ld	r5, KVM_VRMA_SLB_V(r5)
2127	b	4b
2128
2129	/* If this is for emulated MMIO, load the instruction word */
21302:	li	r8, KVM_INST_FETCH_FAILED	/* In case lwz faults */
2131
2132	/* Set guest mode to 'jump over instruction' so if lwz faults
2133	 * we'll just continue at the next IP. */
2134	li	r0, KVM_GUEST_MODE_SKIP
2135	stb	r0, HSTATE_IN_GUEST(r13)
2136
2137	/* Do the access with MSR:DR enabled */
2138	mfmsr	r3
2139	ori	r4, r3, MSR_DR		/* Enable paging for data */
2140	mtmsrd	r4
2141	lwz	r8, 0(r10)
2142	mtmsrd	r3
2143
2144	/* Store the result */
2145	stw	r8, VCPU_LAST_INST(r9)
2146
2147	/* Unset guest mode. */
2148	li	r0, KVM_GUEST_MODE_HOST_HV
2149	stb	r0, HSTATE_IN_GUEST(r13)
2150	b	guest_exit_cont
2151
2152.Lradix_hdsi:
2153	std	r4, VCPU_FAULT_DAR(r9)
2154	stw	r6, VCPU_FAULT_DSISR(r9)
2155.Lradix_hisi:
2156	mfspr	r5, SPRN_ASDR
2157	std	r5, VCPU_FAULT_GPA(r9)
2158	b	guest_exit_cont
2159
2160/*
2161 * Similarly for an HISI, reflect it to the guest as an ISI unless
2162 * it is an HPTE not found fault for a page that we have paged out.
2163 */
2164kvmppc_hisi:
2165	ld	r3, VCPU_KVM(r9)
2166	lbz	r0, KVM_RADIX(r3)
2167	cmpwi	r0, 0
2168	bne	.Lradix_hisi		/* for radix, just save ASDR */
2169	andis.	r0, r11, SRR1_ISI_NOPT@h
2170	beq	1f
2171	andi.	r0, r11, MSR_IR		/* instruction relocation enabled? */
2172	beq	3f
2173BEGIN_FTR_SECTION
2174	mfspr	r5, SPRN_ASDR		/* on POWER9, use ASDR to get VSID */
2175	b	4f
2176END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2177	clrrdi	r0, r10, 28
2178	PPC_SLBFEE_DOT(R5, R0)		/* if so, look up SLB */
2179	li	r0, BOOK3S_INTERRUPT_INST_SEGMENT
2180	bne	7f			/* if no SLB entry found */
21814:
2182	/* Search the hash table. */
2183	mr	r3, r9			/* vcpu pointer */
2184	mr	r4, r10
2185	mr	r6, r11
2186	li	r7, 0			/* instruction fault */
2187	bl	kvmppc_hpte_hv_fault
2188	ld	r9, HSTATE_KVM_VCPU(r13)
2189	ld	r10, VCPU_PC(r9)
2190	ld	r11, VCPU_MSR(r9)
2191	li	r12, BOOK3S_INTERRUPT_H_INST_STORAGE
2192	cmpdi	r3, 0			/* retry the instruction */
2193	beq	fast_interrupt_c_return
2194	cmpdi	r3, -1			/* handle in kernel mode */
2195	beq	guest_exit_cont
2196
2197	/* Synthesize an ISI (or ISegI) for the guest */
2198	mr	r11, r3
21991:	li	r0, BOOK3S_INTERRUPT_INST_STORAGE
22007:	mtspr	SPRN_SRR0, r10
2201	mtspr	SPRN_SRR1, r11
2202	mr	r10, r0
2203	bl	kvmppc_msr_interrupt
2204	b	fast_interrupt_c_return
2205
22063:	ld	r6, VCPU_KVM(r9)	/* not relocated, use VRMA */
2207	ld	r5, KVM_VRMA_SLB_V(r6)
2208	b	4b
2209
2210/*
2211 * Try to handle an hcall in real mode.
2212 * Returns to the guest if we handle it, or continues on up to
2213 * the kernel if we can't (i.e. if we don't have a handler for
2214 * it, or if the handler returns H_TOO_HARD).
2215 *
2216 * r5 - r8 contain hcall args,
2217 * r9 = vcpu, r10 = pc, r11 = msr, r12 = trap, r13 = paca
2218 */
2219hcall_try_real_mode:
2220	ld	r3,VCPU_GPR(R3)(r9)
2221	andi.	r0,r11,MSR_PR
2222	/* sc 1 from userspace - reflect to guest syscall */
2223	bne	sc_1_fast_return
2224	clrrdi	r3,r3,2
2225	cmpldi	r3,hcall_real_table_end - hcall_real_table
2226	bge	guest_exit_cont
2227	/* See if this hcall is enabled for in-kernel handling */
2228	ld	r4, VCPU_KVM(r9)
2229	srdi	r0, r3, 8	/* r0 = (r3 / 4) >> 6 */
2230	sldi	r0, r0, 3	/* index into kvm->arch.enabled_hcalls[] */
2231	add	r4, r4, r0
2232	ld	r0, KVM_ENABLED_HCALLS(r4)
2233	rlwinm	r4, r3, 32-2, 0x3f	/* r4 = (r3 / 4) & 0x3f */
2234	srd	r0, r0, r4
2235	andi.	r0, r0, 1
2236	beq	guest_exit_cont
2237	/* Get pointer to handler, if any, and call it */
2238	LOAD_REG_ADDR(r4, hcall_real_table)
2239	lwax	r3,r3,r4
2240	cmpwi	r3,0
2241	beq	guest_exit_cont
2242	add	r12,r3,r4
2243	mtctr	r12
2244	mr	r3,r9		/* get vcpu pointer */
2245	ld	r4,VCPU_GPR(R4)(r9)
2246	bctrl
2247	cmpdi	r3,H_TOO_HARD
2248	beq	hcall_real_fallback
2249	ld	r4,HSTATE_KVM_VCPU(r13)
2250	std	r3,VCPU_GPR(R3)(r4)
2251	ld	r10,VCPU_PC(r4)
2252	ld	r11,VCPU_MSR(r4)
2253	b	fast_guest_return
2254
2255sc_1_fast_return:
2256	mtspr	SPRN_SRR0,r10
2257	mtspr	SPRN_SRR1,r11
2258	li	r10, BOOK3S_INTERRUPT_SYSCALL
2259	bl	kvmppc_msr_interrupt
2260	mr	r4,r9
2261	b	fast_guest_return
2262
2263	/* We've attempted a real mode hcall, but it's punted it back
2264	 * to userspace.  We need to restore some clobbered volatiles
2265	 * before resuming the pass-it-to-qemu path */
2266hcall_real_fallback:
2267	li	r12,BOOK3S_INTERRUPT_SYSCALL
2268	ld	r9, HSTATE_KVM_VCPU(r13)
2269
2270	b	guest_exit_cont
2271
2272	.globl	hcall_real_table
2273hcall_real_table:
2274	.long	0		/* 0 - unused */
2275	.long	DOTSYM(kvmppc_h_remove) - hcall_real_table
2276	.long	DOTSYM(kvmppc_h_enter) - hcall_real_table
2277	.long	DOTSYM(kvmppc_h_read) - hcall_real_table
2278	.long	DOTSYM(kvmppc_h_clear_mod) - hcall_real_table
2279	.long	DOTSYM(kvmppc_h_clear_ref) - hcall_real_table
2280	.long	DOTSYM(kvmppc_h_protect) - hcall_real_table
2281	.long	DOTSYM(kvmppc_h_get_tce) - hcall_real_table
2282	.long	DOTSYM(kvmppc_rm_h_put_tce) - hcall_real_table
2283	.long	0		/* 0x24 - H_SET_SPRG0 */
2284	.long	DOTSYM(kvmppc_h_set_dabr) - hcall_real_table
2285	.long	0		/* 0x2c */
2286	.long	0		/* 0x30 */
2287	.long	0		/* 0x34 */
2288	.long	0		/* 0x38 */
2289	.long	0		/* 0x3c */
2290	.long	0		/* 0x40 */
2291	.long	0		/* 0x44 */
2292	.long	0		/* 0x48 */
2293	.long	0		/* 0x4c */
2294	.long	0		/* 0x50 */
2295	.long	0		/* 0x54 */
2296	.long	0		/* 0x58 */
2297	.long	0		/* 0x5c */
2298	.long	0		/* 0x60 */
2299#ifdef CONFIG_KVM_XICS
2300	.long	DOTSYM(kvmppc_rm_h_eoi) - hcall_real_table
2301	.long	DOTSYM(kvmppc_rm_h_cppr) - hcall_real_table
2302	.long	DOTSYM(kvmppc_rm_h_ipi) - hcall_real_table
2303	.long	DOTSYM(kvmppc_rm_h_ipoll) - hcall_real_table
2304	.long	DOTSYM(kvmppc_rm_h_xirr) - hcall_real_table
2305#else
2306	.long	0		/* 0x64 - H_EOI */
2307	.long	0		/* 0x68 - H_CPPR */
2308	.long	0		/* 0x6c - H_IPI */
2309	.long	0		/* 0x70 - H_IPOLL */
2310	.long	0		/* 0x74 - H_XIRR */
2311#endif
2312	.long	0		/* 0x78 */
2313	.long	0		/* 0x7c */
2314	.long	0		/* 0x80 */
2315	.long	0		/* 0x84 */
2316	.long	0		/* 0x88 */
2317	.long	0		/* 0x8c */
2318	.long	0		/* 0x90 */
2319	.long	0		/* 0x94 */
2320	.long	0		/* 0x98 */
2321	.long	0		/* 0x9c */
2322	.long	0		/* 0xa0 */
2323	.long	0		/* 0xa4 */
2324	.long	0		/* 0xa8 */
2325	.long	0		/* 0xac */
2326	.long	0		/* 0xb0 */
2327	.long	0		/* 0xb4 */
2328	.long	0		/* 0xb8 */
2329	.long	0		/* 0xbc */
2330	.long	0		/* 0xc0 */
2331	.long	0		/* 0xc4 */
2332	.long	0		/* 0xc8 */
2333	.long	0		/* 0xcc */
2334	.long	0		/* 0xd0 */
2335	.long	0		/* 0xd4 */
2336	.long	0		/* 0xd8 */
2337	.long	0		/* 0xdc */
2338	.long	DOTSYM(kvmppc_h_cede) - hcall_real_table
2339	.long	DOTSYM(kvmppc_rm_h_confer) - hcall_real_table
2340	.long	0		/* 0xe8 */
2341	.long	0		/* 0xec */
2342	.long	0		/* 0xf0 */
2343	.long	0		/* 0xf4 */
2344	.long	0		/* 0xf8 */
2345	.long	0		/* 0xfc */
2346	.long	0		/* 0x100 */
2347	.long	0		/* 0x104 */
2348	.long	0		/* 0x108 */
2349	.long	0		/* 0x10c */
2350	.long	0		/* 0x110 */
2351	.long	0		/* 0x114 */
2352	.long	0		/* 0x118 */
2353	.long	0		/* 0x11c */
2354	.long	0		/* 0x120 */
2355	.long	DOTSYM(kvmppc_h_bulk_remove) - hcall_real_table
2356	.long	0		/* 0x128 */
2357	.long	0		/* 0x12c */
2358	.long	0		/* 0x130 */
2359	.long	DOTSYM(kvmppc_h_set_xdabr) - hcall_real_table
2360	.long	DOTSYM(kvmppc_rm_h_stuff_tce) - hcall_real_table
2361	.long	DOTSYM(kvmppc_rm_h_put_tce_indirect) - hcall_real_table
2362	.long	0		/* 0x140 */
2363	.long	0		/* 0x144 */
2364	.long	0		/* 0x148 */
2365	.long	0		/* 0x14c */
2366	.long	0		/* 0x150 */
2367	.long	0		/* 0x154 */
2368	.long	0		/* 0x158 */
2369	.long	0		/* 0x15c */
2370	.long	0		/* 0x160 */
2371	.long	0		/* 0x164 */
2372	.long	0		/* 0x168 */
2373	.long	0		/* 0x16c */
2374	.long	0		/* 0x170 */
2375	.long	0		/* 0x174 */
2376	.long	0		/* 0x178 */
2377	.long	0		/* 0x17c */
2378	.long	0		/* 0x180 */
2379	.long	0		/* 0x184 */
2380	.long	0		/* 0x188 */
2381	.long	0		/* 0x18c */
2382	.long	0		/* 0x190 */
2383	.long	0		/* 0x194 */
2384	.long	0		/* 0x198 */
2385	.long	0		/* 0x19c */
2386	.long	0		/* 0x1a0 */
2387	.long	0		/* 0x1a4 */
2388	.long	0		/* 0x1a8 */
2389	.long	0		/* 0x1ac */
2390	.long	0		/* 0x1b0 */
2391	.long	0		/* 0x1b4 */
2392	.long	0		/* 0x1b8 */
2393	.long	0		/* 0x1bc */
2394	.long	0		/* 0x1c0 */
2395	.long	0		/* 0x1c4 */
2396	.long	0		/* 0x1c8 */
2397	.long	0		/* 0x1cc */
2398	.long	0		/* 0x1d0 */
2399	.long	0		/* 0x1d4 */
2400	.long	0		/* 0x1d8 */
2401	.long	0		/* 0x1dc */
2402	.long	0		/* 0x1e0 */
2403	.long	0		/* 0x1e4 */
2404	.long	0		/* 0x1e8 */
2405	.long	0		/* 0x1ec */
2406	.long	0		/* 0x1f0 */
2407	.long	0		/* 0x1f4 */
2408	.long	0		/* 0x1f8 */
2409	.long	0		/* 0x1fc */
2410	.long	0		/* 0x200 */
2411	.long	0		/* 0x204 */
2412	.long	0		/* 0x208 */
2413	.long	0		/* 0x20c */
2414	.long	0		/* 0x210 */
2415	.long	0		/* 0x214 */
2416	.long	0		/* 0x218 */
2417	.long	0		/* 0x21c */
2418	.long	0		/* 0x220 */
2419	.long	0		/* 0x224 */
2420	.long	0		/* 0x228 */
2421	.long	0		/* 0x22c */
2422	.long	0		/* 0x230 */
2423	.long	0		/* 0x234 */
2424	.long	0		/* 0x238 */
2425	.long	0		/* 0x23c */
2426	.long	0		/* 0x240 */
2427	.long	0		/* 0x244 */
2428	.long	0		/* 0x248 */
2429	.long	0		/* 0x24c */
2430	.long	0		/* 0x250 */
2431	.long	0		/* 0x254 */
2432	.long	0		/* 0x258 */
2433	.long	0		/* 0x25c */
2434	.long	0		/* 0x260 */
2435	.long	0		/* 0x264 */
2436	.long	0		/* 0x268 */
2437	.long	0		/* 0x26c */
2438	.long	0		/* 0x270 */
2439	.long	0		/* 0x274 */
2440	.long	0		/* 0x278 */
2441	.long	0		/* 0x27c */
2442	.long	0		/* 0x280 */
2443	.long	0		/* 0x284 */
2444	.long	0		/* 0x288 */
2445	.long	0		/* 0x28c */
2446	.long	0		/* 0x290 */
2447	.long	0		/* 0x294 */
2448	.long	0		/* 0x298 */
2449	.long	0		/* 0x29c */
2450	.long	0		/* 0x2a0 */
2451	.long	0		/* 0x2a4 */
2452	.long	0		/* 0x2a8 */
2453	.long	0		/* 0x2ac */
2454	.long	0		/* 0x2b0 */
2455	.long	0		/* 0x2b4 */
2456	.long	0		/* 0x2b8 */
2457	.long	0		/* 0x2bc */
2458	.long	0		/* 0x2c0 */
2459	.long	0		/* 0x2c4 */
2460	.long	0		/* 0x2c8 */
2461	.long	0		/* 0x2cc */
2462	.long	0		/* 0x2d0 */
2463	.long	0		/* 0x2d4 */
2464	.long	0		/* 0x2d8 */
2465	.long	0		/* 0x2dc */
2466	.long	0		/* 0x2e0 */
2467	.long	0		/* 0x2e4 */
2468	.long	0		/* 0x2e8 */
2469	.long	0		/* 0x2ec */
2470	.long	0		/* 0x2f0 */
2471	.long	0		/* 0x2f4 */
2472	.long	0		/* 0x2f8 */
2473#ifdef CONFIG_KVM_XICS
2474	.long	DOTSYM(kvmppc_rm_h_xirr_x) - hcall_real_table
2475#else
2476	.long	0		/* 0x2fc - H_XIRR_X*/
2477#endif
2478	.long	DOTSYM(kvmppc_h_random) - hcall_real_table
2479	.globl	hcall_real_table_end
2480hcall_real_table_end:
2481
2482_GLOBAL(kvmppc_h_set_xdabr)
2483	andi.	r0, r5, DABRX_USER | DABRX_KERNEL
2484	beq	6f
2485	li	r0, DABRX_USER | DABRX_KERNEL | DABRX_BTI
2486	andc.	r0, r5, r0
2487	beq	3f
24886:	li	r3, H_PARAMETER
2489	blr
2490
2491_GLOBAL(kvmppc_h_set_dabr)
2492	li	r5, DABRX_USER | DABRX_KERNEL
24933:
2494BEGIN_FTR_SECTION
2495	b	2f
2496END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2497	std	r4,VCPU_DABR(r3)
2498	stw	r5, VCPU_DABRX(r3)
2499	mtspr	SPRN_DABRX, r5
2500	/* Work around P7 bug where DABR can get corrupted on mtspr */
25011:	mtspr	SPRN_DABR,r4
2502	mfspr	r5, SPRN_DABR
2503	cmpd	r4, r5
2504	bne	1b
2505	isync
2506	li	r3,0
2507	blr
2508
2509	/* Emulate H_SET_DABR/X on P8 for the sake of compat mode guests */
25102:	rlwimi	r5, r4, 5, DAWRX_DR | DAWRX_DW
2511	rlwimi	r5, r4, 2, DAWRX_WT
2512	clrrdi	r4, r4, 3
2513	std	r4, VCPU_DAWR(r3)
2514	std	r5, VCPU_DAWRX(r3)
2515	mtspr	SPRN_DAWR, r4
2516	mtspr	SPRN_DAWRX, r5
2517	li	r3, 0
2518	blr
2519
2520_GLOBAL(kvmppc_h_cede)		/* r3 = vcpu pointer, r11 = msr, r13 = paca */
2521	ori	r11,r11,MSR_EE
2522	std	r11,VCPU_MSR(r3)
2523	li	r0,1
2524	stb	r0,VCPU_CEDED(r3)
2525	sync			/* order setting ceded vs. testing prodded */
2526	lbz	r5,VCPU_PRODDED(r3)
2527	cmpwi	r5,0
2528	bne	kvm_cede_prodded
2529	li	r12,0		/* set trap to 0 to say hcall is handled */
2530	stw	r12,VCPU_TRAP(r3)
2531	li	r0,H_SUCCESS
2532	std	r0,VCPU_GPR(R3)(r3)
2533
2534	/*
2535	 * Set our bit in the bitmask of napping threads unless all the
2536	 * other threads are already napping, in which case we send this
2537	 * up to the host.
2538	 */
2539	ld	r5,HSTATE_KVM_VCORE(r13)
2540	lbz	r6,HSTATE_PTID(r13)
2541	lwz	r8,VCORE_ENTRY_EXIT(r5)
2542	clrldi	r8,r8,56
2543	li	r0,1
2544	sld	r0,r0,r6
2545	addi	r6,r5,VCORE_NAPPING_THREADS
254631:	lwarx	r4,0,r6
2547	or	r4,r4,r0
2548	cmpw	r4,r8
2549	beq	kvm_cede_exit
2550	stwcx.	r4,0,r6
2551	bne	31b
2552	/* order napping_threads update vs testing entry_exit_map */
2553	isync
2554	li	r0,NAPPING_CEDE
2555	stb	r0,HSTATE_NAPPING(r13)
2556	lwz	r7,VCORE_ENTRY_EXIT(r5)
2557	cmpwi	r7,0x100
2558	bge	33f		/* another thread already exiting */
2559
2560/*
2561 * Although not specifically required by the architecture, POWER7
2562 * preserves the following registers in nap mode, even if an SMT mode
2563 * switch occurs: SLB entries, PURR, SPURR, AMOR, UAMOR, AMR, SPRG0-3,
2564 * DAR, DSISR, DABR, DABRX, DSCR, PMCx, MMCRx, SIAR, SDAR.
2565 */
2566	/* Save non-volatile GPRs */
2567	std	r14, VCPU_GPR(R14)(r3)
2568	std	r15, VCPU_GPR(R15)(r3)
2569	std	r16, VCPU_GPR(R16)(r3)
2570	std	r17, VCPU_GPR(R17)(r3)
2571	std	r18, VCPU_GPR(R18)(r3)
2572	std	r19, VCPU_GPR(R19)(r3)
2573	std	r20, VCPU_GPR(R20)(r3)
2574	std	r21, VCPU_GPR(R21)(r3)
2575	std	r22, VCPU_GPR(R22)(r3)
2576	std	r23, VCPU_GPR(R23)(r3)
2577	std	r24, VCPU_GPR(R24)(r3)
2578	std	r25, VCPU_GPR(R25)(r3)
2579	std	r26, VCPU_GPR(R26)(r3)
2580	std	r27, VCPU_GPR(R27)(r3)
2581	std	r28, VCPU_GPR(R28)(r3)
2582	std	r29, VCPU_GPR(R29)(r3)
2583	std	r30, VCPU_GPR(R30)(r3)
2584	std	r31, VCPU_GPR(R31)(r3)
2585
2586	/* save FP state */
2587	bl	kvmppc_save_fp
2588
2589#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2590BEGIN_FTR_SECTION
2591	/*
2592	 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS INCLUDING CR
2593	 */
2594	ld	r9, HSTATE_KVM_VCPU(r13)
2595	bl	kvmppc_save_tm
2596END_FTR_SECTION_IFSET(CPU_FTR_TM)
2597#endif
2598
2599	/*
2600	 * Set DEC to the smaller of DEC and HDEC, so that we wake
2601	 * no later than the end of our timeslice (HDEC interrupts
2602	 * don't wake us from nap).
2603	 */
2604	mfspr	r3, SPRN_DEC
2605	mfspr	r4, SPRN_HDEC
2606	mftb	r5
2607BEGIN_FTR_SECTION
2608	/* On P9 check whether the guest has large decrementer mode enabled */
2609	ld	r6, HSTATE_KVM_VCORE(r13)
2610	ld	r6, VCORE_LPCR(r6)
2611	andis.	r6, r6, LPCR_LD@h
2612	bne	68f
2613END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2614	extsw	r3, r3
261568:	EXTEND_HDEC(r4)
2616	cmpd	r3, r4
2617	ble	67f
2618	mtspr	SPRN_DEC, r4
261967:
2620	/* save expiry time of guest decrementer */
2621	add	r3, r3, r5
2622	ld	r4, HSTATE_KVM_VCPU(r13)
2623	ld	r5, HSTATE_KVM_VCORE(r13)
2624	ld	r6, VCORE_TB_OFFSET(r5)
2625	subf	r3, r6, r3	/* convert to host TB value */
2626	std	r3, VCPU_DEC_EXPIRES(r4)
2627
2628#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2629	ld	r4, HSTATE_KVM_VCPU(r13)
2630	addi	r3, r4, VCPU_TB_CEDE
2631	bl	kvmhv_accumulate_time
2632#endif
2633
2634	lis	r3, LPCR_PECEDP@h	/* Do wake on privileged doorbell */
2635
2636	/*
2637	 * Take a nap until a decrementer or external or doobell interrupt
2638	 * occurs, with PECE1 and PECE0 set in LPCR.
2639	 * On POWER8, set PECEDH, and if we are ceding, also set PECEDP.
2640	 * Also clear the runlatch bit before napping.
2641	 */
2642kvm_do_nap:
2643	mfspr	r0, SPRN_CTRLF
2644	clrrdi	r0, r0, 1
2645	mtspr	SPRN_CTRLT, r0
2646
2647	li	r0,1
2648	stb	r0,HSTATE_HWTHREAD_REQ(r13)
2649	mfspr	r5,SPRN_LPCR
2650	ori	r5,r5,LPCR_PECE0 | LPCR_PECE1
2651BEGIN_FTR_SECTION
2652	ori	r5, r5, LPCR_PECEDH
2653	rlwimi	r5, r3, 0, LPCR_PECEDP
2654END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2655
2656kvm_nap_sequence:		/* desired LPCR value in r5 */
2657BEGIN_FTR_SECTION
2658	/*
2659	 * PSSCR bits:	exit criterion = 1 (wakeup based on LPCR at sreset)
2660	 *		enable state loss = 1 (allow SMT mode switch)
2661	 *		requested level = 0 (just stop dispatching)
2662	 */
2663	lis	r3, (PSSCR_EC | PSSCR_ESL)@h
2664	mtspr	SPRN_PSSCR, r3
2665	/* Set LPCR_PECE_HVEE bit to enable wakeup by HV interrupts */
2666	li	r4, LPCR_PECE_HVEE@higher
2667	sldi	r4, r4, 32
2668	or	r5, r5, r4
2669END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2670	mtspr	SPRN_LPCR,r5
2671	isync
2672	li	r0, 0
2673	std	r0, HSTATE_SCRATCH0(r13)
2674	ptesync
2675	ld	r0, HSTATE_SCRATCH0(r13)
26761:	cmpd	r0, r0
2677	bne	1b
2678BEGIN_FTR_SECTION
2679	nap
2680FTR_SECTION_ELSE
2681	PPC_STOP
2682ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
2683	b	.
2684
268533:	mr	r4, r3
2686	li	r3, 0
2687	li	r12, 0
2688	b	34f
2689
2690kvm_end_cede:
2691	/* get vcpu pointer */
2692	ld	r4, HSTATE_KVM_VCPU(r13)
2693
2694	/* Woken by external or decrementer interrupt */
2695	ld	r1, HSTATE_HOST_R1(r13)
2696
2697#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2698	addi	r3, r4, VCPU_TB_RMINTR
2699	bl	kvmhv_accumulate_time
2700#endif
2701
2702#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2703BEGIN_FTR_SECTION
2704	/*
2705	 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS INCLUDING CR
2706	 */
2707	bl	kvmppc_restore_tm
2708END_FTR_SECTION_IFSET(CPU_FTR_TM)
2709#endif
2710
2711	/* load up FP state */
2712	bl	kvmppc_load_fp
2713
2714	/* Restore guest decrementer */
2715	ld	r3, VCPU_DEC_EXPIRES(r4)
2716	ld	r5, HSTATE_KVM_VCORE(r13)
2717	ld	r6, VCORE_TB_OFFSET(r5)
2718	add	r3, r3, r6	/* convert host TB to guest TB value */
2719	mftb	r7
2720	subf	r3, r7, r3
2721	mtspr	SPRN_DEC, r3
2722
2723	/* Load NV GPRS */
2724	ld	r14, VCPU_GPR(R14)(r4)
2725	ld	r15, VCPU_GPR(R15)(r4)
2726	ld	r16, VCPU_GPR(R16)(r4)
2727	ld	r17, VCPU_GPR(R17)(r4)
2728	ld	r18, VCPU_GPR(R18)(r4)
2729	ld	r19, VCPU_GPR(R19)(r4)
2730	ld	r20, VCPU_GPR(R20)(r4)
2731	ld	r21, VCPU_GPR(R21)(r4)
2732	ld	r22, VCPU_GPR(R22)(r4)
2733	ld	r23, VCPU_GPR(R23)(r4)
2734	ld	r24, VCPU_GPR(R24)(r4)
2735	ld	r25, VCPU_GPR(R25)(r4)
2736	ld	r26, VCPU_GPR(R26)(r4)
2737	ld	r27, VCPU_GPR(R27)(r4)
2738	ld	r28, VCPU_GPR(R28)(r4)
2739	ld	r29, VCPU_GPR(R29)(r4)
2740	ld	r30, VCPU_GPR(R30)(r4)
2741	ld	r31, VCPU_GPR(R31)(r4)
2742
2743	/* Check the wake reason in SRR1 to see why we got here */
2744	bl	kvmppc_check_wake_reason
2745
2746	/*
2747	 * Restore volatile registers since we could have called a
2748	 * C routine in kvmppc_check_wake_reason
2749	 *	r4 = VCPU
2750	 * r3 tells us whether we need to return to host or not
2751	 * WARNING: it gets checked further down:
2752	 * should not modify r3 until this check is done.
2753	 */
2754	ld	r4, HSTATE_KVM_VCPU(r13)
2755
2756	/* clear our bit in vcore->napping_threads */
275734:	ld	r5,HSTATE_KVM_VCORE(r13)
2758	lbz	r7,HSTATE_PTID(r13)
2759	li	r0,1
2760	sld	r0,r0,r7
2761	addi	r6,r5,VCORE_NAPPING_THREADS
276232:	lwarx	r7,0,r6
2763	andc	r7,r7,r0
2764	stwcx.	r7,0,r6
2765	bne	32b
2766	li	r0,0
2767	stb	r0,HSTATE_NAPPING(r13)
2768
2769	/* See if the wake reason saved in r3 means we need to exit */
2770	stw	r12, VCPU_TRAP(r4)
2771	mr	r9, r4
2772	cmpdi	r3, 0
2773	bgt	guest_exit_cont
2774
2775	/* see if any other thread is already exiting */
2776	lwz	r0,VCORE_ENTRY_EXIT(r5)
2777	cmpwi	r0,0x100
2778	bge	guest_exit_cont
2779
2780	b	kvmppc_cede_reentry	/* if not go back to guest */
2781
2782	/* cede when already previously prodded case */
2783kvm_cede_prodded:
2784	li	r0,0
2785	stb	r0,VCPU_PRODDED(r3)
2786	sync			/* order testing prodded vs. clearing ceded */
2787	stb	r0,VCPU_CEDED(r3)
2788	li	r3,H_SUCCESS
2789	blr
2790
2791	/* we've ceded but we want to give control to the host */
2792kvm_cede_exit:
2793	ld	r9, HSTATE_KVM_VCPU(r13)
2794#ifdef CONFIG_KVM_XICS
2795	/* Abort if we still have a pending escalation */
2796	lbz	r5, VCPU_XIVE_ESC_ON(r9)
2797	cmpwi	r5, 0
2798	beq	1f
2799	li	r0, 0
2800	stb	r0, VCPU_CEDED(r9)
28011:	/* Enable XIVE escalation */
2802	li	r5, XIVE_ESB_SET_PQ_00
2803	mfmsr	r0
2804	andi.	r0, r0, MSR_DR		/* in real mode? */
2805	beq	1f
2806	ld	r10, VCPU_XIVE_ESC_VADDR(r9)
2807	cmpdi	r10, 0
2808	beq	3f
2809	ldx	r0, r10, r5
2810	b	2f
28111:	ld	r10, VCPU_XIVE_ESC_RADDR(r9)
2812	cmpdi	r10, 0
2813	beq	3f
2814	ldcix	r0, r10, r5
28152:	sync
2816	li	r0, 1
2817	stb	r0, VCPU_XIVE_ESC_ON(r9)
2818#endif /* CONFIG_KVM_XICS */
28193:	b	guest_exit_cont
2820
2821	/* Try to handle a machine check in real mode */
2822machine_check_realmode:
2823	mr	r3, r9		/* get vcpu pointer */
2824	bl	kvmppc_realmode_machine_check
2825	nop
2826	ld	r9, HSTATE_KVM_VCPU(r13)
2827	li	r12, BOOK3S_INTERRUPT_MACHINE_CHECK
2828	/*
2829	 * For the guest that is FWNMI capable, deliver all the MCE errors
2830	 * (handled/unhandled) by exiting the guest with KVM_EXIT_NMI exit
2831	 * reason. This new approach injects machine check errors in guest
2832	 * address space to guest with additional information in the form
2833	 * of RTAS event, thus enabling guest kernel to suitably handle
2834	 * such errors.
2835	 *
2836	 * For the guest that is not FWNMI capable (old QEMU) fallback
2837	 * to old behaviour for backward compatibility:
2838	 * Deliver unhandled/fatal (e.g. UE) MCE errors to guest either
2839	 * through machine check interrupt (set HSRR0 to 0x200).
2840	 * For handled errors (no-fatal), just go back to guest execution
2841	 * with current HSRR0.
2842	 * if we receive machine check with MSR(RI=0) then deliver it to
2843	 * guest as machine check causing guest to crash.
2844	 */
2845	ld	r11, VCPU_MSR(r9)
2846	rldicl.	r0, r11, 64-MSR_HV_LG, 63 /* check if it happened in HV mode */
2847	bne	mc_cont			/* if so, exit to host */
2848	/* Check if guest is capable of handling NMI exit */
2849	ld	r10, VCPU_KVM(r9)
2850	lbz	r10, KVM_FWNMI(r10)
2851	cmpdi	r10, 1			/* FWNMI capable? */
2852	beq	mc_cont			/* if so, exit with KVM_EXIT_NMI. */
2853
2854	/* if not, fall through for backward compatibility. */
2855	andi.	r10, r11, MSR_RI	/* check for unrecoverable exception */
2856	beq	1f			/* Deliver a machine check to guest */
2857	ld	r10, VCPU_PC(r9)
2858	cmpdi	r3, 0		/* Did we handle MCE ? */
2859	bne	2f	/* Continue guest execution. */
2860	/* If not, deliver a machine check.  SRR0/1 are already set */
28611:	li	r10, BOOK3S_INTERRUPT_MACHINE_CHECK
2862	bl	kvmppc_msr_interrupt
28632:	b	fast_interrupt_c_return
2864
2865/*
2866 * Check the reason we woke from nap, and take appropriate action.
2867 * Returns (in r3):
2868 *	0 if nothing needs to be done
2869 *	1 if something happened that needs to be handled by the host
2870 *	-1 if there was a guest wakeup (IPI or msgsnd)
2871 *	-2 if we handled a PCI passthrough interrupt (returned by
2872 *		kvmppc_read_intr only)
2873 *
2874 * Also sets r12 to the interrupt vector for any interrupt that needs
2875 * to be handled now by the host (0x500 for external interrupt), or zero.
2876 * Modifies all volatile registers (since it may call a C function).
2877 * This routine calls kvmppc_read_intr, a C function, if an external
2878 * interrupt is pending.
2879 */
2880kvmppc_check_wake_reason:
2881	mfspr	r6, SPRN_SRR1
2882BEGIN_FTR_SECTION
2883	rlwinm	r6, r6, 45-31, 0xf	/* extract wake reason field (P8) */
2884FTR_SECTION_ELSE
2885	rlwinm	r6, r6, 45-31, 0xe	/* P7 wake reason field is 3 bits */
2886ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_207S)
2887	cmpwi	r6, 8			/* was it an external interrupt? */
2888	beq	7f			/* if so, see what it was */
2889	li	r3, 0
2890	li	r12, 0
2891	cmpwi	r6, 6			/* was it the decrementer? */
2892	beq	0f
2893BEGIN_FTR_SECTION
2894	cmpwi	r6, 5			/* privileged doorbell? */
2895	beq	0f
2896	cmpwi	r6, 3			/* hypervisor doorbell? */
2897	beq	3f
2898END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2899	cmpwi	r6, 0xa			/* Hypervisor maintenance ? */
2900	beq	4f
2901	li	r3, 1			/* anything else, return 1 */
29020:	blr
2903
2904	/* hypervisor doorbell */
29053:	li	r12, BOOK3S_INTERRUPT_H_DOORBELL
2906
2907	/*
2908	 * Clear the doorbell as we will invoke the handler
2909	 * explicitly in the guest exit path.
2910	 */
2911	lis	r6, (PPC_DBELL_SERVER << (63-36))@h
2912	PPC_MSGCLR(6)
2913	/* see if it's a host IPI */
2914	li	r3, 1
2915BEGIN_FTR_SECTION
2916	PPC_MSGSYNC
2917	lwsync
2918END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2919	lbz	r0, HSTATE_HOST_IPI(r13)
2920	cmpwi	r0, 0
2921	bnelr
2922	/* if not, return -1 */
2923	li	r3, -1
2924	blr
2925
2926	/* Woken up due to Hypervisor maintenance interrupt */
29274:	li	r12, BOOK3S_INTERRUPT_HMI
2928	li	r3, 1
2929	blr
2930
2931	/* external interrupt - create a stack frame so we can call C */
29327:	mflr	r0
2933	std	r0, PPC_LR_STKOFF(r1)
2934	stdu	r1, -PPC_MIN_STKFRM(r1)
2935	bl	kvmppc_read_intr
2936	nop
2937	li	r12, BOOK3S_INTERRUPT_EXTERNAL
2938	cmpdi	r3, 1
2939	ble	1f
2940
2941	/*
2942	 * Return code of 2 means PCI passthrough interrupt, but
2943	 * we need to return back to host to complete handling the
2944	 * interrupt. Trap reason is expected in r12 by guest
2945	 * exit code.
2946	 */
2947	li	r12, BOOK3S_INTERRUPT_HV_RM_HARD
29481:
2949	ld	r0, PPC_MIN_STKFRM+PPC_LR_STKOFF(r1)
2950	addi	r1, r1, PPC_MIN_STKFRM
2951	mtlr	r0
2952	blr
2953
2954/*
2955 * Save away FP, VMX and VSX registers.
2956 * r3 = vcpu pointer
2957 * N.B. r30 and r31 are volatile across this function,
2958 * thus it is not callable from C.
2959 */
2960kvmppc_save_fp:
2961	mflr	r30
2962	mr	r31,r3
2963	mfmsr	r5
2964	ori	r8,r5,MSR_FP
2965#ifdef CONFIG_ALTIVEC
2966BEGIN_FTR_SECTION
2967	oris	r8,r8,MSR_VEC@h
2968END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2969#endif
2970#ifdef CONFIG_VSX
2971BEGIN_FTR_SECTION
2972	oris	r8,r8,MSR_VSX@h
2973END_FTR_SECTION_IFSET(CPU_FTR_VSX)
2974#endif
2975	mtmsrd	r8
2976	addi	r3,r3,VCPU_FPRS
2977	bl	store_fp_state
2978#ifdef CONFIG_ALTIVEC
2979BEGIN_FTR_SECTION
2980	addi	r3,r31,VCPU_VRS
2981	bl	store_vr_state
2982END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2983#endif
2984	mfspr	r6,SPRN_VRSAVE
2985	stw	r6,VCPU_VRSAVE(r31)
2986	mtlr	r30
2987	blr
2988
2989/*
2990 * Load up FP, VMX and VSX registers
2991 * r4 = vcpu pointer
2992 * N.B. r30 and r31 are volatile across this function,
2993 * thus it is not callable from C.
2994 */
2995kvmppc_load_fp:
2996	mflr	r30
2997	mr	r31,r4
2998	mfmsr	r9
2999	ori	r8,r9,MSR_FP
3000#ifdef CONFIG_ALTIVEC
3001BEGIN_FTR_SECTION
3002	oris	r8,r8,MSR_VEC@h
3003END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3004#endif
3005#ifdef CONFIG_VSX
3006BEGIN_FTR_SECTION
3007	oris	r8,r8,MSR_VSX@h
3008END_FTR_SECTION_IFSET(CPU_FTR_VSX)
3009#endif
3010	mtmsrd	r8
3011	addi	r3,r4,VCPU_FPRS
3012	bl	load_fp_state
3013#ifdef CONFIG_ALTIVEC
3014BEGIN_FTR_SECTION
3015	addi	r3,r31,VCPU_VRS
3016	bl	load_vr_state
3017END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3018#endif
3019	lwz	r7,VCPU_VRSAVE(r31)
3020	mtspr	SPRN_VRSAVE,r7
3021	mtlr	r30
3022	mr	r4,r31
3023	blr
3024
3025#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3026/*
3027 * Save transactional state and TM-related registers.
3028 * Called with r9 pointing to the vcpu struct.
3029 * This can modify all checkpointed registers, but
3030 * restores r1, r2 and r9 (vcpu pointer) before exit.
3031 */
3032kvmppc_save_tm:
3033	mflr	r0
3034	std	r0, PPC_LR_STKOFF(r1)
3035
3036	/* Turn on TM. */
3037	mfmsr	r8
3038	li	r0, 1
3039	rldimi	r8, r0, MSR_TM_LG, 63-MSR_TM_LG
3040	mtmsrd	r8
3041
3042	ld	r5, VCPU_MSR(r9)
3043	rldicl. r5, r5, 64 - MSR_TS_S_LG, 62
3044	beq	1f	/* TM not active in guest. */
3045
3046	std	r1, HSTATE_HOST_R1(r13)
3047	li	r3, TM_CAUSE_KVM_RESCHED
3048
3049	/* Clear the MSR RI since r1, r13 are all going to be foobar. */
3050	li	r5, 0
3051	mtmsrd	r5, 1
3052
3053	/* All GPRs are volatile at this point. */
3054	TRECLAIM(R3)
3055
3056	/* Temporarily store r13 and r9 so we have some regs to play with */
3057	SET_SCRATCH0(r13)
3058	GET_PACA(r13)
3059	std	r9, PACATMSCRATCH(r13)
3060	ld	r9, HSTATE_KVM_VCPU(r13)
3061
3062	/* Get a few more GPRs free. */
3063	std	r29, VCPU_GPRS_TM(29)(r9)
3064	std	r30, VCPU_GPRS_TM(30)(r9)
3065	std	r31, VCPU_GPRS_TM(31)(r9)
3066
3067	/* Save away PPR and DSCR soon so don't run with user values. */
3068	mfspr	r31, SPRN_PPR
3069	HMT_MEDIUM
3070	mfspr	r30, SPRN_DSCR
3071	ld	r29, HSTATE_DSCR(r13)
3072	mtspr	SPRN_DSCR, r29
3073
3074	/* Save all but r9, r13 & r29-r31 */
3075	reg = 0
3076	.rept	29
3077	.if (reg != 9) && (reg != 13)
3078	std	reg, VCPU_GPRS_TM(reg)(r9)
3079	.endif
3080	reg = reg + 1
3081	.endr
3082	/* ... now save r13 */
3083	GET_SCRATCH0(r4)
3084	std	r4, VCPU_GPRS_TM(13)(r9)
3085	/* ... and save r9 */
3086	ld	r4, PACATMSCRATCH(r13)
3087	std	r4, VCPU_GPRS_TM(9)(r9)
3088
3089	/* Reload stack pointer and TOC. */
3090	ld	r1, HSTATE_HOST_R1(r13)
3091	ld	r2, PACATOC(r13)
3092
3093	/* Set MSR RI now we have r1 and r13 back. */
3094	li	r5, MSR_RI
3095	mtmsrd	r5, 1
3096
3097	/* Save away checkpinted SPRs. */
3098	std	r31, VCPU_PPR_TM(r9)
3099	std	r30, VCPU_DSCR_TM(r9)
3100	mflr	r5
3101	mfcr	r6
3102	mfctr	r7
3103	mfspr	r8, SPRN_AMR
3104	mfspr	r10, SPRN_TAR
3105	mfxer	r11
3106	std	r5, VCPU_LR_TM(r9)
3107	stw	r6, VCPU_CR_TM(r9)
3108	std	r7, VCPU_CTR_TM(r9)
3109	std	r8, VCPU_AMR_TM(r9)
3110	std	r10, VCPU_TAR_TM(r9)
3111	std	r11, VCPU_XER_TM(r9)
3112
3113	/* Restore r12 as trap number. */
3114	lwz	r12, VCPU_TRAP(r9)
3115
3116	/* Save FP/VSX. */
3117	addi	r3, r9, VCPU_FPRS_TM
3118	bl	store_fp_state
3119	addi	r3, r9, VCPU_VRS_TM
3120	bl	store_vr_state
3121	mfspr	r6, SPRN_VRSAVE
3122	stw	r6, VCPU_VRSAVE_TM(r9)
31231:
3124	/*
3125	 * We need to save these SPRs after the treclaim so that the software
3126	 * error code is recorded correctly in the TEXASR.  Also the user may
3127	 * change these outside of a transaction, so they must always be
3128	 * context switched.
3129	 */
3130	mfspr	r5, SPRN_TFHAR
3131	mfspr	r6, SPRN_TFIAR
3132	mfspr	r7, SPRN_TEXASR
3133	std	r5, VCPU_TFHAR(r9)
3134	std	r6, VCPU_TFIAR(r9)
3135	std	r7, VCPU_TEXASR(r9)
3136
3137	ld	r0, PPC_LR_STKOFF(r1)
3138	mtlr	r0
3139	blr
3140
3141/*
3142 * Restore transactional state and TM-related registers.
3143 * Called with r4 pointing to the vcpu struct.
3144 * This potentially modifies all checkpointed registers.
3145 * It restores r1, r2, r4 from the PACA.
3146 */
3147kvmppc_restore_tm:
3148	mflr	r0
3149	std	r0, PPC_LR_STKOFF(r1)
3150
3151	/* Turn on TM/FP/VSX/VMX so we can restore them. */
3152	mfmsr	r5
3153	li	r6, MSR_TM >> 32
3154	sldi	r6, r6, 32
3155	or	r5, r5, r6
3156	ori	r5, r5, MSR_FP
3157	oris	r5, r5, (MSR_VEC | MSR_VSX)@h
3158	mtmsrd	r5
3159
3160	/*
3161	 * The user may change these outside of a transaction, so they must
3162	 * always be context switched.
3163	 */
3164	ld	r5, VCPU_TFHAR(r4)
3165	ld	r6, VCPU_TFIAR(r4)
3166	ld	r7, VCPU_TEXASR(r4)
3167	mtspr	SPRN_TFHAR, r5
3168	mtspr	SPRN_TFIAR, r6
3169	mtspr	SPRN_TEXASR, r7
3170
3171	ld	r5, VCPU_MSR(r4)
3172	rldicl. r5, r5, 64 - MSR_TS_S_LG, 62
3173	beqlr		/* TM not active in guest */
3174	std	r1, HSTATE_HOST_R1(r13)
3175
3176	/* Make sure the failure summary is set, otherwise we'll program check
3177	 * when we trechkpt.  It's possible that this might have been not set
3178	 * on a kvmppc_set_one_reg() call but we shouldn't let this crash the
3179	 * host.
3180	 */
3181	oris	r7, r7, (TEXASR_FS)@h
3182	mtspr	SPRN_TEXASR, r7
3183
3184	/*
3185	 * We need to load up the checkpointed state for the guest.
3186	 * We need to do this early as it will blow away any GPRs, VSRs and
3187	 * some SPRs.
3188	 */
3189
3190	mr	r31, r4
3191	addi	r3, r31, VCPU_FPRS_TM
3192	bl	load_fp_state
3193	addi	r3, r31, VCPU_VRS_TM
3194	bl	load_vr_state
3195	mr	r4, r31
3196	lwz	r7, VCPU_VRSAVE_TM(r4)
3197	mtspr	SPRN_VRSAVE, r7
3198
3199	ld	r5, VCPU_LR_TM(r4)
3200	lwz	r6, VCPU_CR_TM(r4)
3201	ld	r7, VCPU_CTR_TM(r4)
3202	ld	r8, VCPU_AMR_TM(r4)
3203	ld	r9, VCPU_TAR_TM(r4)
3204	ld	r10, VCPU_XER_TM(r4)
3205	mtlr	r5
3206	mtcr	r6
3207	mtctr	r7
3208	mtspr	SPRN_AMR, r8
3209	mtspr	SPRN_TAR, r9
3210	mtxer	r10
3211
3212	/*
3213	 * Load up PPR and DSCR values but don't put them in the actual SPRs
3214	 * till the last moment to avoid running with userspace PPR and DSCR for
3215	 * too long.
3216	 */
3217	ld	r29, VCPU_DSCR_TM(r4)
3218	ld	r30, VCPU_PPR_TM(r4)
3219
3220	std	r2, PACATMSCRATCH(r13) /* Save TOC */
3221
3222	/* Clear the MSR RI since r1, r13 are all going to be foobar. */
3223	li	r5, 0
3224	mtmsrd	r5, 1
3225
3226	/* Load GPRs r0-r28 */
3227	reg = 0
3228	.rept	29
3229	ld	reg, VCPU_GPRS_TM(reg)(r31)
3230	reg = reg + 1
3231	.endr
3232
3233	mtspr	SPRN_DSCR, r29
3234	mtspr	SPRN_PPR, r30
3235
3236	/* Load final GPRs */
3237	ld	29, VCPU_GPRS_TM(29)(r31)
3238	ld	30, VCPU_GPRS_TM(30)(r31)
3239	ld	31, VCPU_GPRS_TM(31)(r31)
3240
3241	/* TM checkpointed state is now setup.  All GPRs are now volatile. */
3242	TRECHKPT
3243
3244	/* Now let's get back the state we need. */
3245	HMT_MEDIUM
3246	GET_PACA(r13)
3247	ld	r29, HSTATE_DSCR(r13)
3248	mtspr	SPRN_DSCR, r29
3249	ld	r4, HSTATE_KVM_VCPU(r13)
3250	ld	r1, HSTATE_HOST_R1(r13)
3251	ld	r2, PACATMSCRATCH(r13)
3252
3253	/* Set the MSR RI since we have our registers back. */
3254	li	r5, MSR_RI
3255	mtmsrd	r5, 1
3256
3257	ld	r0, PPC_LR_STKOFF(r1)
3258	mtlr	r0
3259	blr
3260#endif
3261
3262/*
3263 * We come here if we get any exception or interrupt while we are
3264 * executing host real mode code while in guest MMU context.
3265 * r12 is (CR << 32) | vector
3266 * r13 points to our PACA
3267 * r12 is saved in HSTATE_SCRATCH0(r13)
3268 * ctr is saved in HSTATE_SCRATCH1(r13) if RELOCATABLE
3269 * r9 is saved in HSTATE_SCRATCH2(r13)
3270 * r13 is saved in HSPRG1
3271 * cfar is saved in HSTATE_CFAR(r13)
3272 * ppr is saved in HSTATE_PPR(r13)
3273 */
3274kvmppc_bad_host_intr:
3275	/*
3276	 * Switch to the emergency stack, but start half-way down in
3277	 * case we were already on it.
3278	 */
3279	mr	r9, r1
3280	std	r1, PACAR1(r13)
3281	ld	r1, PACAEMERGSP(r13)
3282	subi	r1, r1, THREAD_SIZE/2 + INT_FRAME_SIZE
3283	std	r9, 0(r1)
3284	std	r0, GPR0(r1)
3285	std	r9, GPR1(r1)
3286	std	r2, GPR2(r1)
3287	SAVE_4GPRS(3, r1)
3288	SAVE_2GPRS(7, r1)
3289	srdi	r0, r12, 32
3290	clrldi	r12, r12, 32
3291	std	r0, _CCR(r1)
3292	std	r12, _TRAP(r1)
3293	andi.	r0, r12, 2
3294	beq	1f
3295	mfspr	r3, SPRN_HSRR0
3296	mfspr	r4, SPRN_HSRR1
3297	mfspr	r5, SPRN_HDAR
3298	mfspr	r6, SPRN_HDSISR
3299	b	2f
33001:	mfspr	r3, SPRN_SRR0
3301	mfspr	r4, SPRN_SRR1
3302	mfspr	r5, SPRN_DAR
3303	mfspr	r6, SPRN_DSISR
33042:	std	r3, _NIP(r1)
3305	std	r4, _MSR(r1)
3306	std	r5, _DAR(r1)
3307	std	r6, _DSISR(r1)
3308	ld	r9, HSTATE_SCRATCH2(r13)
3309	ld	r12, HSTATE_SCRATCH0(r13)
3310	GET_SCRATCH0(r0)
3311	SAVE_4GPRS(9, r1)
3312	std	r0, GPR13(r1)
3313	SAVE_NVGPRS(r1)
3314	ld	r5, HSTATE_CFAR(r13)
3315	std	r5, ORIG_GPR3(r1)
3316	mflr	r3
3317#ifdef CONFIG_RELOCATABLE
3318	ld	r4, HSTATE_SCRATCH1(r13)
3319#else
3320	mfctr	r4
3321#endif
3322	mfxer	r5
3323	lbz	r6, PACAIRQSOFTMASK(r13)
3324	std	r3, _LINK(r1)
3325	std	r4, _CTR(r1)
3326	std	r5, _XER(r1)
3327	std	r6, SOFTE(r1)
3328	ld	r2, PACATOC(r13)
3329	LOAD_REG_IMMEDIATE(3, 0x7265677368657265)
3330	std	r3, STACK_FRAME_OVERHEAD-16(r1)
3331
3332	/*
3333	 * On POWER9 do a minimal restore of the MMU and call C code,
3334	 * which will print a message and panic.
3335	 * XXX On POWER7 and POWER8, we just spin here since we don't
3336	 * know what the other threads are doing (and we don't want to
3337	 * coordinate with them) - but at least we now have register state
3338	 * in memory that we might be able to look at from another CPU.
3339	 */
3340BEGIN_FTR_SECTION
3341	b	.
3342END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
3343	ld	r9, HSTATE_KVM_VCPU(r13)
3344	ld	r10, VCPU_KVM(r9)
3345
3346	li	r0, 0
3347	mtspr	SPRN_AMR, r0
3348	mtspr	SPRN_IAMR, r0
3349	mtspr	SPRN_CIABR, r0
3350	mtspr	SPRN_DAWRX, r0
3351
3352	/* Flush the ERAT on radix P9 DD1 guest exit */
3353BEGIN_FTR_SECTION
3354	PPC_INVALIDATE_ERAT
3355END_FTR_SECTION_IFSET(CPU_FTR_POWER9_DD1)
3356
3357BEGIN_MMU_FTR_SECTION
3358	b	4f
3359END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
3360
3361	slbmte	r0, r0
3362	slbia
3363	ptesync
3364	ld	r8, PACA_SLBSHADOWPTR(r13)
3365	.rept	SLB_NUM_BOLTED
3366	li	r3, SLBSHADOW_SAVEAREA
3367	LDX_BE	r5, r8, r3
3368	addi	r3, r3, 8
3369	LDX_BE	r6, r8, r3
3370	andis.	r7, r5, SLB_ESID_V@h
3371	beq	3f
3372	slbmte	r6, r5
33733:	addi	r8, r8, 16
3374	.endr
3375
33764:	lwz	r7, KVM_HOST_LPID(r10)
3377	mtspr	SPRN_LPID, r7
3378	mtspr	SPRN_PID, r0
3379	ld	r8, KVM_HOST_LPCR(r10)
3380	mtspr	SPRN_LPCR, r8
3381	isync
3382	li	r0, KVM_GUEST_MODE_NONE
3383	stb	r0, HSTATE_IN_GUEST(r13)
3384
3385	/*
3386	 * Turn on the MMU and jump to C code
3387	 */
3388	bcl	20, 31, .+4
33895:	mflr	r3
3390	addi	r3, r3, 9f - 5b
3391	ld	r4, PACAKMSR(r13)
3392	mtspr	SPRN_SRR0, r3
3393	mtspr	SPRN_SRR1, r4
3394	RFI_TO_KERNEL
33959:	addi	r3, r1, STACK_FRAME_OVERHEAD
3396	bl	kvmppc_bad_interrupt
3397	b	9b
3398
3399/*
3400 * This mimics the MSR transition on IRQ delivery.  The new guest MSR is taken
3401 * from VCPU_INTR_MSR and is modified based on the required TM state changes.
3402 *   r11 has the guest MSR value (in/out)
3403 *   r9 has a vcpu pointer (in)
3404 *   r0 is used as a scratch register
3405 */
3406kvmppc_msr_interrupt:
3407	rldicl	r0, r11, 64 - MSR_TS_S_LG, 62
3408	cmpwi	r0, 2 /* Check if we are in transactional state..  */
3409	ld	r11, VCPU_INTR_MSR(r9)
3410	bne	1f
3411	/* ... if transactional, change to suspended */
3412	li	r0, 1
34131:	rldimi	r11, r0, MSR_TS_S_LG, 63 - MSR_TS_T_LG
3414	blr
3415
3416/*
3417 * This works around a hardware bug on POWER8E processors, where
3418 * writing a 1 to the MMCR0[PMAO] bit doesn't generate a
3419 * performance monitor interrupt.  Instead, when we need to have
3420 * an interrupt pending, we have to arrange for a counter to overflow.
3421 */
3422kvmppc_fix_pmao:
3423	li	r3, 0
3424	mtspr	SPRN_MMCR2, r3
3425	lis	r3, (MMCR0_PMXE | MMCR0_FCECE)@h
3426	ori	r3, r3, MMCR0_PMCjCE | MMCR0_C56RUN
3427	mtspr	SPRN_MMCR0, r3
3428	lis	r3, 0x7fff
3429	ori	r3, r3, 0xffff
3430	mtspr	SPRN_PMC6, r3
3431	isync
3432	blr
3433
3434#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
3435/*
3436 * Start timing an activity
3437 * r3 = pointer to time accumulation struct, r4 = vcpu
3438 */
3439kvmhv_start_timing:
3440	ld	r5, HSTATE_KVM_VCORE(r13)
3441	lbz	r6, VCORE_IN_GUEST(r5)
3442	cmpwi	r6, 0
3443	beq	5f				/* if in guest, need to */
3444	ld	r6, VCORE_TB_OFFSET(r5)		/* subtract timebase offset */
34455:	mftb	r5
3446	subf	r5, r6, r5
3447	std	r3, VCPU_CUR_ACTIVITY(r4)
3448	std	r5, VCPU_ACTIVITY_START(r4)
3449	blr
3450
3451/*
3452 * Accumulate time to one activity and start another.
3453 * r3 = pointer to new time accumulation struct, r4 = vcpu
3454 */
3455kvmhv_accumulate_time:
3456	ld	r5, HSTATE_KVM_VCORE(r13)
3457	lbz	r8, VCORE_IN_GUEST(r5)
3458	cmpwi	r8, 0
3459	beq	4f				/* if in guest, need to */
3460	ld	r8, VCORE_TB_OFFSET(r5)		/* subtract timebase offset */
34614:	ld	r5, VCPU_CUR_ACTIVITY(r4)
3462	ld	r6, VCPU_ACTIVITY_START(r4)
3463	std	r3, VCPU_CUR_ACTIVITY(r4)
3464	mftb	r7
3465	subf	r7, r8, r7
3466	std	r7, VCPU_ACTIVITY_START(r4)
3467	cmpdi	r5, 0
3468	beqlr
3469	subf	r3, r6, r7
3470	ld	r8, TAS_SEQCOUNT(r5)
3471	cmpdi	r8, 0
3472	addi	r8, r8, 1
3473	std	r8, TAS_SEQCOUNT(r5)
3474	lwsync
3475	ld	r7, TAS_TOTAL(r5)
3476	add	r7, r7, r3
3477	std	r7, TAS_TOTAL(r5)
3478	ld	r6, TAS_MIN(r5)
3479	ld	r7, TAS_MAX(r5)
3480	beq	3f
3481	cmpd	r3, r6
3482	bge	1f
34833:	std	r3, TAS_MIN(r5)
34841:	cmpd	r3, r7
3485	ble	2f
3486	std	r3, TAS_MAX(r5)
34872:	lwsync
3488	addi	r8, r8, 1
3489	std	r8, TAS_SEQCOUNT(r5)
3490	blr
3491#endif
3492