1/*
2 * This program is free software; you can redistribute it and/or modify
3 * it under the terms of the GNU General Public License, version 2, as
4 * published by the Free Software Foundation.
5 *
6 * This program is distributed in the hope that it will be useful,
7 * but WITHOUT ANY WARRANTY; without even the implied warranty of
8 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9 * GNU General Public License for more details.
10 *
11 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
12 *
13 * Derived from book3s_rmhandlers.S and other files, which are:
14 *
15 * Copyright SUSE Linux Products GmbH 2009
16 *
17 * Authors: Alexander Graf <agraf@suse.de>
18 */
19
20#include <asm/ppc_asm.h>
21#include <asm/kvm_asm.h>
22#include <asm/reg.h>
23#include <asm/mmu.h>
24#include <asm/page.h>
25#include <asm/ptrace.h>
26#include <asm/hvcall.h>
27#include <asm/asm-offsets.h>
28#include <asm/exception-64s.h>
29#include <asm/kvm_book3s_asm.h>
30#include <asm/book3s/64/mmu-hash.h>
31#include <asm/export.h>
32#include <asm/tm.h>
33#include <asm/opal.h>
34#include <asm/xive-regs.h>
35#include <asm/thread_info.h>
36#include <asm/asm-compat.h>
37#include <asm/feature-fixups.h>
38#include <asm/cpuidle.h>
39
40/* Sign-extend HDEC if not on POWER9 */
41#define EXTEND_HDEC(reg)			\
42BEGIN_FTR_SECTION;				\
43	extsw	reg, reg;			\
44END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
45
46/* Values in HSTATE_NAPPING(r13) */
47#define NAPPING_CEDE	1
48#define NAPPING_NOVCPU	2
49#define NAPPING_UNSPLIT	3
50
51/* Stack frame offsets for kvmppc_hv_entry */
52#define SFS			208
53#define STACK_SLOT_TRAP		(SFS-4)
54#define STACK_SLOT_SHORT_PATH	(SFS-8)
55#define STACK_SLOT_TID		(SFS-16)
56#define STACK_SLOT_PSSCR	(SFS-24)
57#define STACK_SLOT_PID		(SFS-32)
58#define STACK_SLOT_IAMR		(SFS-40)
59#define STACK_SLOT_CIABR	(SFS-48)
60#define STACK_SLOT_DAWR		(SFS-56)
61#define STACK_SLOT_DAWRX	(SFS-64)
62#define STACK_SLOT_HFSCR	(SFS-72)
63#define STACK_SLOT_AMR		(SFS-80)
64#define STACK_SLOT_UAMOR	(SFS-88)
65/* the following is used by the P9 short path */
66#define STACK_SLOT_NVGPRS	(SFS-152)	/* 18 gprs */
67
68/*
69 * Call kvmppc_hv_entry in real mode.
70 * Must be called with interrupts hard-disabled.
71 *
72 * Input Registers:
73 *
74 * LR = return address to continue at after eventually re-enabling MMU
75 */
76_GLOBAL_TOC(kvmppc_hv_entry_trampoline)
77	mflr	r0
78	std	r0, PPC_LR_STKOFF(r1)
79	stdu	r1, -112(r1)
80	mfmsr	r10
81	std	r10, HSTATE_HOST_MSR(r13)
82	LOAD_REG_ADDR(r5, kvmppc_call_hv_entry)
83	li	r0,MSR_RI
84	andc	r0,r10,r0
85	li	r6,MSR_IR | MSR_DR
86	andc	r6,r10,r6
87	mtmsrd	r0,1		/* clear RI in MSR */
88	mtsrr0	r5
89	mtsrr1	r6
90	RFI_TO_KERNEL
91
92kvmppc_call_hv_entry:
93BEGIN_FTR_SECTION
94	/* On P9, do LPCR setting, if necessary */
95	ld	r3, HSTATE_SPLIT_MODE(r13)
96	cmpdi	r3, 0
97	beq	46f
98	lwz	r4, KVM_SPLIT_DO_SET(r3)
99	cmpwi	r4, 0
100	beq	46f
101	bl	kvmhv_p9_set_lpcr
102	nop
10346:
104END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
105
106	ld	r4, HSTATE_KVM_VCPU(r13)
107	bl	kvmppc_hv_entry
108
109	/* Back from guest - restore host state and return to caller */
110
111BEGIN_FTR_SECTION
112	/* Restore host DABR and DABRX */
113	ld	r5,HSTATE_DABR(r13)
114	li	r6,7
115	mtspr	SPRN_DABR,r5
116	mtspr	SPRN_DABRX,r6
117END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
118
119	/* Restore SPRG3 */
120	ld	r3,PACA_SPRG_VDSO(r13)
121	mtspr	SPRN_SPRG_VDSO_WRITE,r3
122
123	/* Reload the host's PMU registers */
124	bl	kvmhv_load_host_pmu
125
126	/*
127	 * Reload DEC.  HDEC interrupts were disabled when
128	 * we reloaded the host's LPCR value.
129	 */
130	ld	r3, HSTATE_DECEXP(r13)
131	mftb	r4
132	subf	r4, r4, r3
133	mtspr	SPRN_DEC, r4
134
135	/* hwthread_req may have got set by cede or no vcpu, so clear it */
136	li	r0, 0
137	stb	r0, HSTATE_HWTHREAD_REQ(r13)
138
139	/*
140	 * For external interrupts we need to call the Linux
141	 * handler to process the interrupt. We do that by jumping
142	 * to absolute address 0x500 for external interrupts.
143	 * The [h]rfid at the end of the handler will return to
144	 * the book3s_hv_interrupts.S code. For other interrupts
145	 * we do the rfid to get back to the book3s_hv_interrupts.S
146	 * code here.
147	 */
148	ld	r8, 112+PPC_LR_STKOFF(r1)
149	addi	r1, r1, 112
150	ld	r7, HSTATE_HOST_MSR(r13)
151
152	/* Return the trap number on this thread as the return value */
153	mr	r3, r12
154
155	/*
156	 * If we came back from the guest via a relocation-on interrupt,
157	 * we will be in virtual mode at this point, which makes it a
158	 * little easier to get back to the caller.
159	 */
160	mfmsr	r0
161	andi.	r0, r0, MSR_IR		/* in real mode? */
162	bne	.Lvirt_return
163
164	/* RFI into the highmem handler */
165	mfmsr	r6
166	li	r0, MSR_RI
167	andc	r6, r6, r0
168	mtmsrd	r6, 1			/* Clear RI in MSR */
169	mtsrr0	r8
170	mtsrr1	r7
171	RFI_TO_KERNEL
172
173	/* Virtual-mode return */
174.Lvirt_return:
175	mtlr	r8
176	blr
177
178kvmppc_primary_no_guest:
179	/* We handle this much like a ceded vcpu */
180	/* put the HDEC into the DEC, since HDEC interrupts don't wake us */
181	/* HDEC may be larger than DEC for arch >= v3.00, but since the */
182	/* HDEC value came from DEC in the first place, it will fit */
183	mfspr	r3, SPRN_HDEC
184	mtspr	SPRN_DEC, r3
185	/*
186	 * Make sure the primary has finished the MMU switch.
187	 * We should never get here on a secondary thread, but
188	 * check it for robustness' sake.
189	 */
190	ld	r5, HSTATE_KVM_VCORE(r13)
19165:	lbz	r0, VCORE_IN_GUEST(r5)
192	cmpwi	r0, 0
193	beq	65b
194	/* Set LPCR. */
195	ld	r8,VCORE_LPCR(r5)
196	mtspr	SPRN_LPCR,r8
197	isync
198	/* set our bit in napping_threads */
199	ld	r5, HSTATE_KVM_VCORE(r13)
200	lbz	r7, HSTATE_PTID(r13)
201	li	r0, 1
202	sld	r0, r0, r7
203	addi	r6, r5, VCORE_NAPPING_THREADS
2041:	lwarx	r3, 0, r6
205	or	r3, r3, r0
206	stwcx.	r3, 0, r6
207	bne	1b
208	/* order napping_threads update vs testing entry_exit_map */
209	isync
210	li	r12, 0
211	lwz	r7, VCORE_ENTRY_EXIT(r5)
212	cmpwi	r7, 0x100
213	bge	kvm_novcpu_exit	/* another thread already exiting */
214	li	r3, NAPPING_NOVCPU
215	stb	r3, HSTATE_NAPPING(r13)
216
217	li	r3, 0		/* Don't wake on privileged (OS) doorbell */
218	b	kvm_do_nap
219
220/*
221 * kvm_novcpu_wakeup
222 *	Entered from kvm_start_guest if kvm_hstate.napping is set
223 *	to NAPPING_NOVCPU
224 *		r2 = kernel TOC
225 *		r13 = paca
226 */
227kvm_novcpu_wakeup:
228	ld	r1, HSTATE_HOST_R1(r13)
229	ld	r5, HSTATE_KVM_VCORE(r13)
230	li	r0, 0
231	stb	r0, HSTATE_NAPPING(r13)
232
233	/* check the wake reason */
234	bl	kvmppc_check_wake_reason
235
236	/*
237	 * Restore volatile registers since we could have called
238	 * a C routine in kvmppc_check_wake_reason.
239	 *	r5 = VCORE
240	 */
241	ld	r5, HSTATE_KVM_VCORE(r13)
242
243	/* see if any other thread is already exiting */
244	lwz	r0, VCORE_ENTRY_EXIT(r5)
245	cmpwi	r0, 0x100
246	bge	kvm_novcpu_exit
247
248	/* clear our bit in napping_threads */
249	lbz	r7, HSTATE_PTID(r13)
250	li	r0, 1
251	sld	r0, r0, r7
252	addi	r6, r5, VCORE_NAPPING_THREADS
2534:	lwarx	r7, 0, r6
254	andc	r7, r7, r0
255	stwcx.	r7, 0, r6
256	bne	4b
257
258	/* See if the wake reason means we need to exit */
259	cmpdi	r3, 0
260	bge	kvm_novcpu_exit
261
262	/* See if our timeslice has expired (HDEC is negative) */
263	mfspr	r0, SPRN_HDEC
264	EXTEND_HDEC(r0)
265	li	r12, BOOK3S_INTERRUPT_HV_DECREMENTER
266	cmpdi	r0, 0
267	blt	kvm_novcpu_exit
268
269	/* Got an IPI but other vcpus aren't yet exiting, must be a latecomer */
270	ld	r4, HSTATE_KVM_VCPU(r13)
271	cmpdi	r4, 0
272	beq	kvmppc_primary_no_guest
273
274#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
275	addi	r3, r4, VCPU_TB_RMENTRY
276	bl	kvmhv_start_timing
277#endif
278	b	kvmppc_got_guest
279
280kvm_novcpu_exit:
281#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
282	ld	r4, HSTATE_KVM_VCPU(r13)
283	cmpdi	r4, 0
284	beq	13f
285	addi	r3, r4, VCPU_TB_RMEXIT
286	bl	kvmhv_accumulate_time
287#endif
28813:	mr	r3, r12
289	stw	r12, STACK_SLOT_TRAP(r1)
290	bl	kvmhv_commence_exit
291	nop
292	b	kvmhv_switch_to_host
293
294/*
295 * We come in here when wakened from Linux offline idle code.
296 * Relocation is off
297 * r3 contains the SRR1 wakeup value, SRR1 is trashed.
298 */
299_GLOBAL(idle_kvm_start_guest)
300	ld	r4,PACAEMERGSP(r13)
301	mfcr	r5
302	mflr	r0
303	std	r1,0(r4)
304	std	r5,8(r4)
305	std	r0,16(r4)
306	subi	r1,r4,STACK_FRAME_OVERHEAD
307	SAVE_NVGPRS(r1)
308
309	/*
310	 * Could avoid this and pass it through in r3. For now,
311	 * code expects it to be in SRR1.
312	 */
313	mtspr	SPRN_SRR1,r3
314
315	li	r0,0
316	stb	r0,PACA_FTRACE_ENABLED(r13)
317
318	li	r0,KVM_HWTHREAD_IN_KVM
319	stb	r0,HSTATE_HWTHREAD_STATE(r13)
320
321	/* kvm cede / napping does not come through here */
322	lbz	r0,HSTATE_NAPPING(r13)
323	twnei	r0,0
324
325	b	1f
326
327kvm_unsplit_wakeup:
328	li	r0, 0
329	stb	r0, HSTATE_NAPPING(r13)
330
3311:
332
333	/*
334	 * We weren't napping due to cede, so this must be a secondary
335	 * thread being woken up to run a guest, or being woken up due
336	 * to a stray IPI.  (Or due to some machine check or hypervisor
337	 * maintenance interrupt while the core is in KVM.)
338	 */
339
340	/* Check the wake reason in SRR1 to see why we got here */
341	bl	kvmppc_check_wake_reason
342	/*
343	 * kvmppc_check_wake_reason could invoke a C routine, but we
344	 * have no volatile registers to restore when we return.
345	 */
346
347	cmpdi	r3, 0
348	bge	kvm_no_guest
349
350	/* get vcore pointer, NULL if we have nothing to run */
351	ld	r5,HSTATE_KVM_VCORE(r13)
352	cmpdi	r5,0
353	/* if we have no vcore to run, go back to sleep */
354	beq	kvm_no_guest
355
356kvm_secondary_got_guest:
357
358	/* Set HSTATE_DSCR(r13) to something sensible */
359	ld	r6, PACA_DSCR_DEFAULT(r13)
360	std	r6, HSTATE_DSCR(r13)
361
362	/* On thread 0 of a subcore, set HDEC to max */
363	lbz	r4, HSTATE_PTID(r13)
364	cmpwi	r4, 0
365	bne	63f
366	LOAD_REG_ADDR(r6, decrementer_max)
367	ld	r6, 0(r6)
368	mtspr	SPRN_HDEC, r6
369	/* and set per-LPAR registers, if doing dynamic micro-threading */
370	ld	r6, HSTATE_SPLIT_MODE(r13)
371	cmpdi	r6, 0
372	beq	63f
373BEGIN_FTR_SECTION
374	ld	r0, KVM_SPLIT_RPR(r6)
375	mtspr	SPRN_RPR, r0
376	ld	r0, KVM_SPLIT_PMMAR(r6)
377	mtspr	SPRN_PMMAR, r0
378	ld	r0, KVM_SPLIT_LDBAR(r6)
379	mtspr	SPRN_LDBAR, r0
380	isync
381FTR_SECTION_ELSE
382	/* On P9 we use the split_info for coordinating LPCR changes */
383	lwz	r4, KVM_SPLIT_DO_SET(r6)
384	cmpwi	r4, 0
385	beq	1f
386	mr	r3, r6
387	bl	kvmhv_p9_set_lpcr
388	nop
3891:
390ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
39163:
392	/* Order load of vcpu after load of vcore */
393	lwsync
394	ld	r4, HSTATE_KVM_VCPU(r13)
395	bl	kvmppc_hv_entry
396
397	/* Back from the guest, go back to nap */
398	/* Clear our vcpu and vcore pointers so we don't come back in early */
399	li	r0, 0
400	std	r0, HSTATE_KVM_VCPU(r13)
401	/*
402	 * Once we clear HSTATE_KVM_VCORE(r13), the code in
403	 * kvmppc_run_core() is going to assume that all our vcpu
404	 * state is visible in memory.  This lwsync makes sure
405	 * that that is true.
406	 */
407	lwsync
408	std	r0, HSTATE_KVM_VCORE(r13)
409
410	/*
411	 * All secondaries exiting guest will fall through this path.
412	 * Before proceeding, just check for HMI interrupt and
413	 * invoke opal hmi handler. By now we are sure that the
414	 * primary thread on this core/subcore has already made partition
415	 * switch/TB resync and we are good to call opal hmi handler.
416	 */
417	cmpwi	r12, BOOK3S_INTERRUPT_HMI
418	bne	kvm_no_guest
419
420	li	r3,0			/* NULL argument */
421	bl	hmi_exception_realmode
422/*
423 * At this point we have finished executing in the guest.
424 * We need to wait for hwthread_req to become zero, since
425 * we may not turn on the MMU while hwthread_req is non-zero.
426 * While waiting we also need to check if we get given a vcpu to run.
427 */
428kvm_no_guest:
429	lbz	r3, HSTATE_HWTHREAD_REQ(r13)
430	cmpwi	r3, 0
431	bne	53f
432	HMT_MEDIUM
433	li	r0, KVM_HWTHREAD_IN_KERNEL
434	stb	r0, HSTATE_HWTHREAD_STATE(r13)
435	/* need to recheck hwthread_req after a barrier, to avoid race */
436	sync
437	lbz	r3, HSTATE_HWTHREAD_REQ(r13)
438	cmpwi	r3, 0
439	bne	54f
440
441	/*
442	 * Jump to idle_return_gpr_loss, which returns to the
443	 * idle_kvm_start_guest caller.
444	 */
445	li	r3, LPCR_PECE0
446	mfspr	r4, SPRN_LPCR
447	rlwimi	r4, r3, 0, LPCR_PECE0 | LPCR_PECE1
448	mtspr	SPRN_LPCR, r4
449	/* set up r3 for return */
450	mfspr	r3,SPRN_SRR1
451	REST_NVGPRS(r1)
452	addi	r1, r1, STACK_FRAME_OVERHEAD
453	ld	r0, 16(r1)
454	ld	r5, 8(r1)
455	ld	r1, 0(r1)
456	mtlr	r0
457	mtcr	r5
458	blr
459
46053:	HMT_LOW
461	ld	r5, HSTATE_KVM_VCORE(r13)
462	cmpdi	r5, 0
463	bne	60f
464	ld	r3, HSTATE_SPLIT_MODE(r13)
465	cmpdi	r3, 0
466	beq	kvm_no_guest
467	lwz	r0, KVM_SPLIT_DO_SET(r3)
468	cmpwi	r0, 0
469	bne	kvmhv_do_set
470	lwz	r0, KVM_SPLIT_DO_RESTORE(r3)
471	cmpwi	r0, 0
472	bne	kvmhv_do_restore
473	lbz	r0, KVM_SPLIT_DO_NAP(r3)
474	cmpwi	r0, 0
475	beq	kvm_no_guest
476	HMT_MEDIUM
477	b	kvm_unsplit_nap
47860:	HMT_MEDIUM
479	b	kvm_secondary_got_guest
480
48154:	li	r0, KVM_HWTHREAD_IN_KVM
482	stb	r0, HSTATE_HWTHREAD_STATE(r13)
483	b	kvm_no_guest
484
485kvmhv_do_set:
486	/* Set LPCR, LPIDR etc. on P9 */
487	HMT_MEDIUM
488	bl	kvmhv_p9_set_lpcr
489	nop
490	b	kvm_no_guest
491
492kvmhv_do_restore:
493	HMT_MEDIUM
494	bl	kvmhv_p9_restore_lpcr
495	nop
496	b	kvm_no_guest
497
498/*
499 * Here the primary thread is trying to return the core to
500 * whole-core mode, so we need to nap.
501 */
502kvm_unsplit_nap:
503	/*
504	 * When secondaries are napping in kvm_unsplit_nap() with
505	 * hwthread_req = 1, HMI goes ignored even though subcores are
506	 * already exited the guest. Hence HMI keeps waking up secondaries
507	 * from nap in a loop and secondaries always go back to nap since
508	 * no vcore is assigned to them. This makes impossible for primary
509	 * thread to get hold of secondary threads resulting into a soft
510	 * lockup in KVM path.
511	 *
512	 * Let us check if HMI is pending and handle it before we go to nap.
513	 */
514	cmpwi	r12, BOOK3S_INTERRUPT_HMI
515	bne	55f
516	li	r3, 0			/* NULL argument */
517	bl	hmi_exception_realmode
51855:
519	/*
520	 * Ensure that secondary doesn't nap when it has
521	 * its vcore pointer set.
522	 */
523	sync		/* matches smp_mb() before setting split_info.do_nap */
524	ld	r0, HSTATE_KVM_VCORE(r13)
525	cmpdi	r0, 0
526	bne	kvm_no_guest
527	/* clear any pending message */
528BEGIN_FTR_SECTION
529	lis	r6, (PPC_DBELL_SERVER << (63-36))@h
530	PPC_MSGCLR(6)
531END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
532	/* Set kvm_split_mode.napped[tid] = 1 */
533	ld	r3, HSTATE_SPLIT_MODE(r13)
534	li	r0, 1
535	lbz	r4, HSTATE_TID(r13)
536	addi	r4, r4, KVM_SPLIT_NAPPED
537	stbx	r0, r3, r4
538	/* Check the do_nap flag again after setting napped[] */
539	sync
540	lbz	r0, KVM_SPLIT_DO_NAP(r3)
541	cmpwi	r0, 0
542	beq	57f
543	li	r3, NAPPING_UNSPLIT
544	stb	r3, HSTATE_NAPPING(r13)
545	li	r3, (LPCR_PECEDH | LPCR_PECE0) >> 4
546	mfspr	r5, SPRN_LPCR
547	rlwimi	r5, r3, 4, (LPCR_PECEDP | LPCR_PECEDH | LPCR_PECE0 | LPCR_PECE1)
548	b	kvm_nap_sequence
549
55057:	li	r0, 0
551	stbx	r0, r3, r4
552	b	kvm_no_guest
553
554/******************************************************************************
555 *                                                                            *
556 *                               Entry code                                   *
557 *                                                                            *
558 *****************************************************************************/
559
560.global kvmppc_hv_entry
561kvmppc_hv_entry:
562
563	/* Required state:
564	 *
565	 * R4 = vcpu pointer (or NULL)
566	 * MSR = ~IR|DR
567	 * R13 = PACA
568	 * R1 = host R1
569	 * R2 = TOC
570	 * all other volatile GPRS = free
571	 * Does not preserve non-volatile GPRs or CR fields
572	 */
573	mflr	r0
574	std	r0, PPC_LR_STKOFF(r1)
575	stdu	r1, -SFS(r1)
576
577	/* Save R1 in the PACA */
578	std	r1, HSTATE_HOST_R1(r13)
579
580	li	r6, KVM_GUEST_MODE_HOST_HV
581	stb	r6, HSTATE_IN_GUEST(r13)
582
583#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
584	/* Store initial timestamp */
585	cmpdi	r4, 0
586	beq	1f
587	addi	r3, r4, VCPU_TB_RMENTRY
588	bl	kvmhv_start_timing
5891:
590#endif
591
592	ld	r5, HSTATE_KVM_VCORE(r13)
593	ld	r9, VCORE_KVM(r5)	/* pointer to struct kvm */
594
595	/*
596	 * POWER7/POWER8 host -> guest partition switch code.
597	 * We don't have to lock against concurrent tlbies,
598	 * but we do have to coordinate across hardware threads.
599	 */
600	/* Set bit in entry map iff exit map is zero. */
601	li	r7, 1
602	lbz	r6, HSTATE_PTID(r13)
603	sld	r7, r7, r6
604	addi	r8, r5, VCORE_ENTRY_EXIT
60521:	lwarx	r3, 0, r8
606	cmpwi	r3, 0x100		/* any threads starting to exit? */
607	bge	secondary_too_late	/* if so we're too late to the party */
608	or	r3, r3, r7
609	stwcx.	r3, 0, r8
610	bne	21b
611
612	/* Primary thread switches to guest partition. */
613	cmpwi	r6,0
614	bne	10f
615
616	lwz	r7,KVM_LPID(r9)
617BEGIN_FTR_SECTION
618	ld	r6,KVM_SDR1(r9)
619	li	r0,LPID_RSVD		/* switch to reserved LPID */
620	mtspr	SPRN_LPID,r0
621	ptesync
622	mtspr	SPRN_SDR1,r6		/* switch to partition page table */
623END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
624	mtspr	SPRN_LPID,r7
625	isync
626
627	/* See if we need to flush the TLB. */
628	mr	r3, r9			/* kvm pointer */
629	lhz	r4, PACAPACAINDEX(r13)	/* physical cpu number */
630	li	r5, 0			/* nested vcpu pointer */
631	bl	kvmppc_check_need_tlb_flush
632	nop
633	ld	r5, HSTATE_KVM_VCORE(r13)
634
635	/* Add timebase offset onto timebase */
63622:	ld	r8,VCORE_TB_OFFSET(r5)
637	cmpdi	r8,0
638	beq	37f
639	std	r8, VCORE_TB_OFFSET_APPL(r5)
640	mftb	r6		/* current host timebase */
641	add	r8,r8,r6
642	mtspr	SPRN_TBU40,r8	/* update upper 40 bits */
643	mftb	r7		/* check if lower 24 bits overflowed */
644	clrldi	r6,r6,40
645	clrldi	r7,r7,40
646	cmpld	r7,r6
647	bge	37f
648	addis	r8,r8,0x100	/* if so, increment upper 40 bits */
649	mtspr	SPRN_TBU40,r8
650
651	/* Load guest PCR value to select appropriate compat mode */
65237:	ld	r7, VCORE_PCR(r5)
653	cmpdi	r7, 0
654	beq	38f
655	mtspr	SPRN_PCR, r7
65638:
657
658BEGIN_FTR_SECTION
659	/* DPDES and VTB are shared between threads */
660	ld	r8, VCORE_DPDES(r5)
661	ld	r7, VCORE_VTB(r5)
662	mtspr	SPRN_DPDES, r8
663	mtspr	SPRN_VTB, r7
664END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
665
666	/* Mark the subcore state as inside guest */
667	bl	kvmppc_subcore_enter_guest
668	nop
669	ld	r5, HSTATE_KVM_VCORE(r13)
670	ld	r4, HSTATE_KVM_VCPU(r13)
671	li	r0,1
672	stb	r0,VCORE_IN_GUEST(r5)	/* signal secondaries to continue */
673
674	/* Do we have a guest vcpu to run? */
67510:	cmpdi	r4, 0
676	beq	kvmppc_primary_no_guest
677kvmppc_got_guest:
678	/* Increment yield count if they have a VPA */
679	ld	r3, VCPU_VPA(r4)
680	cmpdi	r3, 0
681	beq	25f
682	li	r6, LPPACA_YIELDCOUNT
683	LWZX_BE	r5, r3, r6
684	addi	r5, r5, 1
685	STWX_BE	r5, r3, r6
686	li	r6, 1
687	stb	r6, VCPU_VPA_DIRTY(r4)
68825:
689
690	/* Save purr/spurr */
691	mfspr	r5,SPRN_PURR
692	mfspr	r6,SPRN_SPURR
693	std	r5,HSTATE_PURR(r13)
694	std	r6,HSTATE_SPURR(r13)
695	ld	r7,VCPU_PURR(r4)
696	ld	r8,VCPU_SPURR(r4)
697	mtspr	SPRN_PURR,r7
698	mtspr	SPRN_SPURR,r8
699
700	/* Save host values of some registers */
701BEGIN_FTR_SECTION
702	mfspr	r5, SPRN_TIDR
703	mfspr	r6, SPRN_PSSCR
704	mfspr	r7, SPRN_PID
705	std	r5, STACK_SLOT_TID(r1)
706	std	r6, STACK_SLOT_PSSCR(r1)
707	std	r7, STACK_SLOT_PID(r1)
708	mfspr	r5, SPRN_HFSCR
709	std	r5, STACK_SLOT_HFSCR(r1)
710END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
711BEGIN_FTR_SECTION
712	mfspr	r5, SPRN_CIABR
713	mfspr	r6, SPRN_DAWR
714	mfspr	r7, SPRN_DAWRX
715	mfspr	r8, SPRN_IAMR
716	std	r5, STACK_SLOT_CIABR(r1)
717	std	r6, STACK_SLOT_DAWR(r1)
718	std	r7, STACK_SLOT_DAWRX(r1)
719	std	r8, STACK_SLOT_IAMR(r1)
720END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
721
722	mfspr	r5, SPRN_AMR
723	std	r5, STACK_SLOT_AMR(r1)
724	mfspr	r6, SPRN_UAMOR
725	std	r6, STACK_SLOT_UAMOR(r1)
726
727BEGIN_FTR_SECTION
728	/* Set partition DABR */
729	/* Do this before re-enabling PMU to avoid P7 DABR corruption bug */
730	lwz	r5,VCPU_DABRX(r4)
731	ld	r6,VCPU_DABR(r4)
732	mtspr	SPRN_DABRX,r5
733	mtspr	SPRN_DABR,r6
734	isync
735END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
736
737#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
738/*
739 * Branch around the call if both CPU_FTR_TM and
740 * CPU_FTR_P9_TM_HV_ASSIST are off.
741 */
742BEGIN_FTR_SECTION
743	b	91f
744END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
745	/*
746	 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
747	 */
748	mr      r3, r4
749	ld      r4, VCPU_MSR(r3)
750	li	r5, 0			/* don't preserve non-vol regs */
751	bl	kvmppc_restore_tm_hv
752	nop
753	ld	r4, HSTATE_KVM_VCPU(r13)
75491:
755#endif
756
757	/* Load guest PMU registers; r4 = vcpu pointer here */
758	mr	r3, r4
759	bl	kvmhv_load_guest_pmu
760
761	/* Load up FP, VMX and VSX registers */
762	ld	r4, HSTATE_KVM_VCPU(r13)
763	bl	kvmppc_load_fp
764
765	ld	r14, VCPU_GPR(R14)(r4)
766	ld	r15, VCPU_GPR(R15)(r4)
767	ld	r16, VCPU_GPR(R16)(r4)
768	ld	r17, VCPU_GPR(R17)(r4)
769	ld	r18, VCPU_GPR(R18)(r4)
770	ld	r19, VCPU_GPR(R19)(r4)
771	ld	r20, VCPU_GPR(R20)(r4)
772	ld	r21, VCPU_GPR(R21)(r4)
773	ld	r22, VCPU_GPR(R22)(r4)
774	ld	r23, VCPU_GPR(R23)(r4)
775	ld	r24, VCPU_GPR(R24)(r4)
776	ld	r25, VCPU_GPR(R25)(r4)
777	ld	r26, VCPU_GPR(R26)(r4)
778	ld	r27, VCPU_GPR(R27)(r4)
779	ld	r28, VCPU_GPR(R28)(r4)
780	ld	r29, VCPU_GPR(R29)(r4)
781	ld	r30, VCPU_GPR(R30)(r4)
782	ld	r31, VCPU_GPR(R31)(r4)
783
784	/* Switch DSCR to guest value */
785	ld	r5, VCPU_DSCR(r4)
786	mtspr	SPRN_DSCR, r5
787
788BEGIN_FTR_SECTION
789	/* Skip next section on POWER7 */
790	b	8f
791END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
792	/* Load up POWER8-specific registers */
793	ld	r5, VCPU_IAMR(r4)
794	lwz	r6, VCPU_PSPB(r4)
795	ld	r7, VCPU_FSCR(r4)
796	mtspr	SPRN_IAMR, r5
797	mtspr	SPRN_PSPB, r6
798	mtspr	SPRN_FSCR, r7
799	/*
800	 * Handle broken DAWR case by not writing it. This means we
801	 * can still store the DAWR register for migration.
802	 */
803	LOAD_REG_ADDR(r5, dawr_force_enable)
804	lbz	r5, 0(r5)
805	cmpdi	r5, 0
806	beq	1f
807	ld	r5, VCPU_DAWR(r4)
808	ld	r6, VCPU_DAWRX(r4)
809	mtspr	SPRN_DAWR, r5
810	mtspr	SPRN_DAWRX, r6
8111:
812	ld	r7, VCPU_CIABR(r4)
813	ld	r8, VCPU_TAR(r4)
814	mtspr	SPRN_CIABR, r7
815	mtspr	SPRN_TAR, r8
816	ld	r5, VCPU_IC(r4)
817	ld	r8, VCPU_EBBHR(r4)
818	mtspr	SPRN_IC, r5
819	mtspr	SPRN_EBBHR, r8
820	ld	r5, VCPU_EBBRR(r4)
821	ld	r6, VCPU_BESCR(r4)
822	lwz	r7, VCPU_GUEST_PID(r4)
823	ld	r8, VCPU_WORT(r4)
824	mtspr	SPRN_EBBRR, r5
825	mtspr	SPRN_BESCR, r6
826	mtspr	SPRN_PID, r7
827	mtspr	SPRN_WORT, r8
828BEGIN_FTR_SECTION
829	/* POWER8-only registers */
830	ld	r5, VCPU_TCSCR(r4)
831	ld	r6, VCPU_ACOP(r4)
832	ld	r7, VCPU_CSIGR(r4)
833	ld	r8, VCPU_TACR(r4)
834	mtspr	SPRN_TCSCR, r5
835	mtspr	SPRN_ACOP, r6
836	mtspr	SPRN_CSIGR, r7
837	mtspr	SPRN_TACR, r8
838	nop
839FTR_SECTION_ELSE
840	/* POWER9-only registers */
841	ld	r5, VCPU_TID(r4)
842	ld	r6, VCPU_PSSCR(r4)
843	lbz	r8, HSTATE_FAKE_SUSPEND(r13)
844	oris	r6, r6, PSSCR_EC@h	/* This makes stop trap to HV */
845	rldimi	r6, r8, PSSCR_FAKE_SUSPEND_LG, 63 - PSSCR_FAKE_SUSPEND_LG
846	ld	r7, VCPU_HFSCR(r4)
847	mtspr	SPRN_TIDR, r5
848	mtspr	SPRN_PSSCR, r6
849	mtspr	SPRN_HFSCR, r7
850ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
8518:
852
853	ld	r5, VCPU_SPRG0(r4)
854	ld	r6, VCPU_SPRG1(r4)
855	ld	r7, VCPU_SPRG2(r4)
856	ld	r8, VCPU_SPRG3(r4)
857	mtspr	SPRN_SPRG0, r5
858	mtspr	SPRN_SPRG1, r6
859	mtspr	SPRN_SPRG2, r7
860	mtspr	SPRN_SPRG3, r8
861
862	/* Load up DAR and DSISR */
863	ld	r5, VCPU_DAR(r4)
864	lwz	r6, VCPU_DSISR(r4)
865	mtspr	SPRN_DAR, r5
866	mtspr	SPRN_DSISR, r6
867
868	/* Restore AMR and UAMOR, set AMOR to all 1s */
869	ld	r5,VCPU_AMR(r4)
870	ld	r6,VCPU_UAMOR(r4)
871	li	r7,-1
872	mtspr	SPRN_AMR,r5
873	mtspr	SPRN_UAMOR,r6
874	mtspr	SPRN_AMOR,r7
875
876	/* Restore state of CTRL run bit; assume 1 on entry */
877	lwz	r5,VCPU_CTRL(r4)
878	andi.	r5,r5,1
879	bne	4f
880	mfspr	r6,SPRN_CTRLF
881	clrrdi	r6,r6,1
882	mtspr	SPRN_CTRLT,r6
8834:
884	/* Secondary threads wait for primary to have done partition switch */
885	ld	r5, HSTATE_KVM_VCORE(r13)
886	lbz	r6, HSTATE_PTID(r13)
887	cmpwi	r6, 0
888	beq	21f
889	lbz	r0, VCORE_IN_GUEST(r5)
890	cmpwi	r0, 0
891	bne	21f
892	HMT_LOW
89320:	lwz	r3, VCORE_ENTRY_EXIT(r5)
894	cmpwi	r3, 0x100
895	bge	no_switch_exit
896	lbz	r0, VCORE_IN_GUEST(r5)
897	cmpwi	r0, 0
898	beq	20b
899	HMT_MEDIUM
90021:
901	/* Set LPCR. */
902	ld	r8,VCORE_LPCR(r5)
903	mtspr	SPRN_LPCR,r8
904	isync
905
906	/*
907	 * Set the decrementer to the guest decrementer.
908	 */
909	ld	r8,VCPU_DEC_EXPIRES(r4)
910	/* r8 is a host timebase value here, convert to guest TB */
911	ld	r5,HSTATE_KVM_VCORE(r13)
912	ld	r6,VCORE_TB_OFFSET_APPL(r5)
913	add	r8,r8,r6
914	mftb	r7
915	subf	r3,r7,r8
916	mtspr	SPRN_DEC,r3
917
918	/* Check if HDEC expires soon */
919	mfspr	r3, SPRN_HDEC
920	EXTEND_HDEC(r3)
921	cmpdi	r3, 512		/* 1 microsecond */
922	blt	hdec_soon
923
924	/* For hash guest, clear out and reload the SLB */
925	ld	r6, VCPU_KVM(r4)
926	lbz	r0, KVM_RADIX(r6)
927	cmpwi	r0, 0
928	bne	9f
929	li	r6, 0
930	slbmte	r6, r6
931	slbia
932	ptesync
933
934	/* Load up guest SLB entries (N.B. slb_max will be 0 for radix) */
935	lwz	r5,VCPU_SLB_MAX(r4)
936	cmpwi	r5,0
937	beq	9f
938	mtctr	r5
939	addi	r6,r4,VCPU_SLB
9401:	ld	r8,VCPU_SLB_E(r6)
941	ld	r9,VCPU_SLB_V(r6)
942	slbmte	r9,r8
943	addi	r6,r6,VCPU_SLB_SIZE
944	bdnz	1b
9459:
946
947#ifdef CONFIG_KVM_XICS
948	/* We are entering the guest on that thread, push VCPU to XIVE */
949	ld	r11, VCPU_XIVE_SAVED_STATE(r4)
950	li	r9, TM_QW1_OS
951	lwz	r8, VCPU_XIVE_CAM_WORD(r4)
952	li	r7, TM_QW1_OS + TM_WORD2
953	mfmsr	r0
954	andi.	r0, r0, MSR_DR		/* in real mode? */
955	beq	2f
956	ld	r10, HSTATE_XIVE_TIMA_VIRT(r13)
957	cmpldi	cr1, r10, 0
958	beq     cr1, no_xive
959	eieio
960	stdx	r11,r9,r10
961	stwx	r8,r7,r10
962	b	3f
9632:	ld	r10, HSTATE_XIVE_TIMA_PHYS(r13)
964	cmpldi	cr1, r10, 0
965	beq	cr1, no_xive
966	eieio
967	stdcix	r11,r9,r10
968	stwcix	r8,r7,r10
9693:	li	r9, 1
970	stb	r9, VCPU_XIVE_PUSHED(r4)
971	eieio
972
973	/*
974	 * We clear the irq_pending flag. There is a small chance of a
975	 * race vs. the escalation interrupt happening on another
976	 * processor setting it again, but the only consequence is to
977	 * cause a spurrious wakeup on the next H_CEDE which is not an
978	 * issue.
979	 */
980	li	r0,0
981	stb	r0, VCPU_IRQ_PENDING(r4)
982
983	/*
984	 * In single escalation mode, if the escalation interrupt is
985	 * on, we mask it.
986	 */
987	lbz	r0, VCPU_XIVE_ESC_ON(r4)
988	cmpwi	cr1, r0,0
989	beq	cr1, 1f
990	li	r9, XIVE_ESB_SET_PQ_01
991	beq	4f			/* in real mode? */
992	ld	r10, VCPU_XIVE_ESC_VADDR(r4)
993	ldx	r0, r10, r9
994	b	5f
9954:	ld	r10, VCPU_XIVE_ESC_RADDR(r4)
996	ldcix	r0, r10, r9
9975:	sync
998
999	/* We have a possible subtle race here: The escalation interrupt might
1000	 * have fired and be on its way to the host queue while we mask it,
1001	 * and if we unmask it early enough (re-cede right away), there is
1002	 * a theorical possibility that it fires again, thus landing in the
1003	 * target queue more than once which is a big no-no.
1004	 *
1005	 * Fortunately, solving this is rather easy. If the above load setting
1006	 * PQ to 01 returns a previous value where P is set, then we know the
1007	 * escalation interrupt is somewhere on its way to the host. In that
1008	 * case we simply don't clear the xive_esc_on flag below. It will be
1009	 * eventually cleared by the handler for the escalation interrupt.
1010	 *
1011	 * Then, when doing a cede, we check that flag again before re-enabling
1012	 * the escalation interrupt, and if set, we abort the cede.
1013	 */
1014	andi.	r0, r0, XIVE_ESB_VAL_P
1015	bne-	1f
1016
1017	/* Now P is 0, we can clear the flag */
1018	li	r0, 0
1019	stb	r0, VCPU_XIVE_ESC_ON(r4)
10201:
1021no_xive:
1022#endif /* CONFIG_KVM_XICS */
1023
1024	li	r0, 0
1025	stw	r0, STACK_SLOT_SHORT_PATH(r1)
1026
1027deliver_guest_interrupt:	/* r4 = vcpu, r13 = paca */
1028	/* Check if we can deliver an external or decrementer interrupt now */
1029	ld	r0, VCPU_PENDING_EXC(r4)
1030BEGIN_FTR_SECTION
1031	/* On POWER9, also check for emulated doorbell interrupt */
1032	lbz	r3, VCPU_DBELL_REQ(r4)
1033	or	r0, r0, r3
1034END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1035	cmpdi	r0, 0
1036	beq	71f
1037	mr	r3, r4
1038	bl	kvmppc_guest_entry_inject_int
1039	ld	r4, HSTATE_KVM_VCPU(r13)
104071:
1041	ld	r6, VCPU_SRR0(r4)
1042	ld	r7, VCPU_SRR1(r4)
1043	mtspr	SPRN_SRR0, r6
1044	mtspr	SPRN_SRR1, r7
1045
1046fast_guest_entry_c:
1047	ld	r10, VCPU_PC(r4)
1048	ld	r11, VCPU_MSR(r4)
1049	/* r11 = vcpu->arch.msr & ~MSR_HV */
1050	rldicl	r11, r11, 63 - MSR_HV_LG, 1
1051	rotldi	r11, r11, 1 + MSR_HV_LG
1052	ori	r11, r11, MSR_ME
1053
1054	ld	r6, VCPU_CTR(r4)
1055	ld	r7, VCPU_XER(r4)
1056	mtctr	r6
1057	mtxer	r7
1058
1059/*
1060 * Required state:
1061 * R4 = vcpu
1062 * R10: value for HSRR0
1063 * R11: value for HSRR1
1064 * R13 = PACA
1065 */
1066fast_guest_return:
1067	li	r0,0
1068	stb	r0,VCPU_CEDED(r4)	/* cancel cede */
1069	mtspr	SPRN_HSRR0,r10
1070	mtspr	SPRN_HSRR1,r11
1071
1072	/* Activate guest mode, so faults get handled by KVM */
1073	li	r9, KVM_GUEST_MODE_GUEST_HV
1074	stb	r9, HSTATE_IN_GUEST(r13)
1075
1076#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1077	/* Accumulate timing */
1078	addi	r3, r4, VCPU_TB_GUEST
1079	bl	kvmhv_accumulate_time
1080#endif
1081
1082	/* Enter guest */
1083
1084BEGIN_FTR_SECTION
1085	ld	r5, VCPU_CFAR(r4)
1086	mtspr	SPRN_CFAR, r5
1087END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
1088BEGIN_FTR_SECTION
1089	ld	r0, VCPU_PPR(r4)
1090END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1091
1092	ld	r5, VCPU_LR(r4)
1093	ld	r6, VCPU_CR(r4)
1094	mtlr	r5
1095	mtcr	r6
1096
1097	ld	r1, VCPU_GPR(R1)(r4)
1098	ld	r2, VCPU_GPR(R2)(r4)
1099	ld	r3, VCPU_GPR(R3)(r4)
1100	ld	r5, VCPU_GPR(R5)(r4)
1101	ld	r6, VCPU_GPR(R6)(r4)
1102	ld	r7, VCPU_GPR(R7)(r4)
1103	ld	r8, VCPU_GPR(R8)(r4)
1104	ld	r9, VCPU_GPR(R9)(r4)
1105	ld	r10, VCPU_GPR(R10)(r4)
1106	ld	r11, VCPU_GPR(R11)(r4)
1107	ld	r12, VCPU_GPR(R12)(r4)
1108	ld	r13, VCPU_GPR(R13)(r4)
1109
1110BEGIN_FTR_SECTION
1111	mtspr	SPRN_PPR, r0
1112END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1113
1114/* Move canary into DSISR to check for later */
1115BEGIN_FTR_SECTION
1116	li	r0, 0x7fff
1117	mtspr	SPRN_HDSISR, r0
1118END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1119
1120	ld	r0, VCPU_GPR(R0)(r4)
1121	ld	r4, VCPU_GPR(R4)(r4)
1122	HRFI_TO_GUEST
1123	b	.
1124
1125/*
1126 * Enter the guest on a P9 or later system where we have exactly
1127 * one vcpu per vcore and we don't need to go to real mode
1128 * (which implies that host and guest are both using radix MMU mode).
1129 * r3 = vcpu pointer
1130 * Most SPRs and all the VSRs have been loaded already.
1131 */
1132_GLOBAL(__kvmhv_vcpu_entry_p9)
1133EXPORT_SYMBOL_GPL(__kvmhv_vcpu_entry_p9)
1134	mflr	r0
1135	std	r0, PPC_LR_STKOFF(r1)
1136	stdu	r1, -SFS(r1)
1137
1138	li	r0, 1
1139	stw	r0, STACK_SLOT_SHORT_PATH(r1)
1140
1141	std	r3, HSTATE_KVM_VCPU(r13)
1142	mfcr	r4
1143	stw	r4, SFS+8(r1)
1144
1145	std	r1, HSTATE_HOST_R1(r13)
1146
1147	reg = 14
1148	.rept	18
1149	std	reg, STACK_SLOT_NVGPRS + ((reg - 14) * 8)(r1)
1150	reg = reg + 1
1151	.endr
1152
1153	reg = 14
1154	.rept	18
1155	ld	reg, __VCPU_GPR(reg)(r3)
1156	reg = reg + 1
1157	.endr
1158
1159	mfmsr	r10
1160	std	r10, HSTATE_HOST_MSR(r13)
1161
1162	mr	r4, r3
1163	b	fast_guest_entry_c
1164guest_exit_short_path:
1165
1166	li	r0, KVM_GUEST_MODE_NONE
1167	stb	r0, HSTATE_IN_GUEST(r13)
1168
1169	reg = 14
1170	.rept	18
1171	std	reg, __VCPU_GPR(reg)(r9)
1172	reg = reg + 1
1173	.endr
1174
1175	reg = 14
1176	.rept	18
1177	ld	reg, STACK_SLOT_NVGPRS + ((reg - 14) * 8)(r1)
1178	reg = reg + 1
1179	.endr
1180
1181	lwz	r4, SFS+8(r1)
1182	mtcr	r4
1183
1184	mr	r3, r12		/* trap number */
1185
1186	addi	r1, r1, SFS
1187	ld	r0, PPC_LR_STKOFF(r1)
1188	mtlr	r0
1189
1190	/* If we are in real mode, do a rfid to get back to the caller */
1191	mfmsr	r4
1192	andi.	r5, r4, MSR_IR
1193	bnelr
1194	rldicl	r5, r4, 64 - MSR_TS_S_LG, 62	/* extract TS field */
1195	mtspr	SPRN_SRR0, r0
1196	ld	r10, HSTATE_HOST_MSR(r13)
1197	rldimi	r10, r5, MSR_TS_S_LG, 63 - MSR_TS_T_LG
1198	mtspr	SPRN_SRR1, r10
1199	RFI_TO_KERNEL
1200	b	.
1201
1202secondary_too_late:
1203	li	r12, 0
1204	stw	r12, STACK_SLOT_TRAP(r1)
1205	cmpdi	r4, 0
1206	beq	11f
1207	stw	r12, VCPU_TRAP(r4)
1208#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1209	addi	r3, r4, VCPU_TB_RMEXIT
1210	bl	kvmhv_accumulate_time
1211#endif
121211:	b	kvmhv_switch_to_host
1213
1214no_switch_exit:
1215	HMT_MEDIUM
1216	li	r12, 0
1217	b	12f
1218hdec_soon:
1219	li	r12, BOOK3S_INTERRUPT_HV_DECREMENTER
122012:	stw	r12, VCPU_TRAP(r4)
1221	mr	r9, r4
1222#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1223	addi	r3, r4, VCPU_TB_RMEXIT
1224	bl	kvmhv_accumulate_time
1225#endif
1226	b	guest_bypass
1227
1228/******************************************************************************
1229 *                                                                            *
1230 *                               Exit code                                    *
1231 *                                                                            *
1232 *****************************************************************************/
1233
1234/*
1235 * We come here from the first-level interrupt handlers.
1236 */
1237	.globl	kvmppc_interrupt_hv
1238kvmppc_interrupt_hv:
1239	/*
1240	 * Register contents:
1241	 * R12		= (guest CR << 32) | interrupt vector
1242	 * R13		= PACA
1243	 * guest R12 saved in shadow VCPU SCRATCH0
1244	 * guest CTR saved in shadow VCPU SCRATCH1 if RELOCATABLE
1245	 * guest R13 saved in SPRN_SCRATCH0
1246	 */
1247	std	r9, HSTATE_SCRATCH2(r13)
1248	lbz	r9, HSTATE_IN_GUEST(r13)
1249	cmpwi	r9, KVM_GUEST_MODE_HOST_HV
1250	beq	kvmppc_bad_host_intr
1251#ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1252	cmpwi	r9, KVM_GUEST_MODE_GUEST
1253	ld	r9, HSTATE_SCRATCH2(r13)
1254	beq	kvmppc_interrupt_pr
1255#endif
1256	/* We're now back in the host but in guest MMU context */
1257	li	r9, KVM_GUEST_MODE_HOST_HV
1258	stb	r9, HSTATE_IN_GUEST(r13)
1259
1260	ld	r9, HSTATE_KVM_VCPU(r13)
1261
1262	/* Save registers */
1263
1264	std	r0, VCPU_GPR(R0)(r9)
1265	std	r1, VCPU_GPR(R1)(r9)
1266	std	r2, VCPU_GPR(R2)(r9)
1267	std	r3, VCPU_GPR(R3)(r9)
1268	std	r4, VCPU_GPR(R4)(r9)
1269	std	r5, VCPU_GPR(R5)(r9)
1270	std	r6, VCPU_GPR(R6)(r9)
1271	std	r7, VCPU_GPR(R7)(r9)
1272	std	r8, VCPU_GPR(R8)(r9)
1273	ld	r0, HSTATE_SCRATCH2(r13)
1274	std	r0, VCPU_GPR(R9)(r9)
1275	std	r10, VCPU_GPR(R10)(r9)
1276	std	r11, VCPU_GPR(R11)(r9)
1277	ld	r3, HSTATE_SCRATCH0(r13)
1278	std	r3, VCPU_GPR(R12)(r9)
1279	/* CR is in the high half of r12 */
1280	srdi	r4, r12, 32
1281	std	r4, VCPU_CR(r9)
1282BEGIN_FTR_SECTION
1283	ld	r3, HSTATE_CFAR(r13)
1284	std	r3, VCPU_CFAR(r9)
1285END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
1286BEGIN_FTR_SECTION
1287	ld	r4, HSTATE_PPR(r13)
1288	std	r4, VCPU_PPR(r9)
1289END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1290
1291	/* Restore R1/R2 so we can handle faults */
1292	ld	r1, HSTATE_HOST_R1(r13)
1293	ld	r2, PACATOC(r13)
1294
1295	mfspr	r10, SPRN_SRR0
1296	mfspr	r11, SPRN_SRR1
1297	std	r10, VCPU_SRR0(r9)
1298	std	r11, VCPU_SRR1(r9)
1299	/* trap is in the low half of r12, clear CR from the high half */
1300	clrldi	r12, r12, 32
1301	andi.	r0, r12, 2		/* need to read HSRR0/1? */
1302	beq	1f
1303	mfspr	r10, SPRN_HSRR0
1304	mfspr	r11, SPRN_HSRR1
1305	clrrdi	r12, r12, 2
13061:	std	r10, VCPU_PC(r9)
1307	std	r11, VCPU_MSR(r9)
1308
1309	GET_SCRATCH0(r3)
1310	mflr	r4
1311	std	r3, VCPU_GPR(R13)(r9)
1312	std	r4, VCPU_LR(r9)
1313
1314	stw	r12,VCPU_TRAP(r9)
1315
1316	/*
1317	 * Now that we have saved away SRR0/1 and HSRR0/1,
1318	 * interrupts are recoverable in principle, so set MSR_RI.
1319	 * This becomes important for relocation-on interrupts from
1320	 * the guest, which we can get in radix mode on POWER9.
1321	 */
1322	li	r0, MSR_RI
1323	mtmsrd	r0, 1
1324
1325#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1326	addi	r3, r9, VCPU_TB_RMINTR
1327	mr	r4, r9
1328	bl	kvmhv_accumulate_time
1329	ld	r5, VCPU_GPR(R5)(r9)
1330	ld	r6, VCPU_GPR(R6)(r9)
1331	ld	r7, VCPU_GPR(R7)(r9)
1332	ld	r8, VCPU_GPR(R8)(r9)
1333#endif
1334
1335	/* Save HEIR (HV emulation assist reg) in emul_inst
1336	   if this is an HEI (HV emulation interrupt, e40) */
1337	li	r3,KVM_INST_FETCH_FAILED
1338	stw	r3,VCPU_LAST_INST(r9)
1339	cmpwi	r12,BOOK3S_INTERRUPT_H_EMUL_ASSIST
1340	bne	11f
1341	mfspr	r3,SPRN_HEIR
134211:	stw	r3,VCPU_HEIR(r9)
1343
1344	/* these are volatile across C function calls */
1345#ifdef CONFIG_RELOCATABLE
1346	ld	r3, HSTATE_SCRATCH1(r13)
1347	mtctr	r3
1348#else
1349	mfctr	r3
1350#endif
1351	mfxer	r4
1352	std	r3, VCPU_CTR(r9)
1353	std	r4, VCPU_XER(r9)
1354
1355	/* Save more register state  */
1356	mfdar	r3
1357	mfdsisr	r4
1358	std	r3, VCPU_DAR(r9)
1359	stw	r4, VCPU_DSISR(r9)
1360
1361	/* If this is a page table miss then see if it's theirs or ours */
1362	cmpwi	r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
1363	beq	kvmppc_hdsi
1364	std	r3, VCPU_FAULT_DAR(r9)
1365	stw	r4, VCPU_FAULT_DSISR(r9)
1366	cmpwi	r12, BOOK3S_INTERRUPT_H_INST_STORAGE
1367	beq	kvmppc_hisi
1368
1369#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1370	/* For softpatch interrupt, go off and do TM instruction emulation */
1371	cmpwi	r12, BOOK3S_INTERRUPT_HV_SOFTPATCH
1372	beq	kvmppc_tm_emul
1373#endif
1374
1375	/* See if this is a leftover HDEC interrupt */
1376	cmpwi	r12,BOOK3S_INTERRUPT_HV_DECREMENTER
1377	bne	2f
1378	mfspr	r3,SPRN_HDEC
1379	EXTEND_HDEC(r3)
1380	cmpdi	r3,0
1381	mr	r4,r9
1382	bge	fast_guest_return
13832:
1384	/* See if this is an hcall we can handle in real mode */
1385	cmpwi	r12,BOOK3S_INTERRUPT_SYSCALL
1386	beq	hcall_try_real_mode
1387
1388	/* Hypervisor doorbell - exit only if host IPI flag set */
1389	cmpwi	r12, BOOK3S_INTERRUPT_H_DOORBELL
1390	bne	3f
1391BEGIN_FTR_SECTION
1392	PPC_MSGSYNC
1393	lwsync
1394	/* always exit if we're running a nested guest */
1395	ld	r0, VCPU_NESTED(r9)
1396	cmpdi	r0, 0
1397	bne	guest_exit_cont
1398END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1399	lbz	r0, HSTATE_HOST_IPI(r13)
1400	cmpwi	r0, 0
1401	beq	maybe_reenter_guest
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	beq	kvmppc_guest_external
1414	/* See if it is a machine check */
1415	cmpwi	r12, BOOK3S_INTERRUPT_MACHINE_CHECK
1416	beq	machine_check_realmode
1417	/* Or a hypervisor maintenance interrupt */
1418	cmpwi	r12, BOOK3S_INTERRUPT_HMI
1419	beq	hmi_realmode
1420
1421guest_exit_cont:		/* r9 = vcpu, r12 = trap, r13 = paca */
1422
1423#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1424	addi	r3, r9, VCPU_TB_RMEXIT
1425	mr	r4, r9
1426	bl	kvmhv_accumulate_time
1427#endif
1428#ifdef CONFIG_KVM_XICS
1429	/* We are exiting, pull the VP from the XIVE */
1430	lbz	r0, VCPU_XIVE_PUSHED(r9)
1431	cmpwi	cr0, r0, 0
1432	beq	1f
1433	li	r7, TM_SPC_PULL_OS_CTX
1434	li	r6, TM_QW1_OS
1435	mfmsr	r0
1436	andi.	r0, r0, MSR_DR		/* in real mode? */
1437	beq	2f
1438	ld	r10, HSTATE_XIVE_TIMA_VIRT(r13)
1439	cmpldi	cr0, r10, 0
1440	beq	1f
1441	/* First load to pull the context, we ignore the value */
1442	eieio
1443	lwzx	r11, r7, r10
1444	/* Second load to recover the context state (Words 0 and 1) */
1445	ldx	r11, r6, r10
1446	b	3f
14472:	ld	r10, HSTATE_XIVE_TIMA_PHYS(r13)
1448	cmpldi	cr0, r10, 0
1449	beq	1f
1450	/* First load to pull the context, we ignore the value */
1451	eieio
1452	lwzcix	r11, r7, r10
1453	/* Second load to recover the context state (Words 0 and 1) */
1454	ldcix	r11, r6, r10
14553:	std	r11, VCPU_XIVE_SAVED_STATE(r9)
1456	/* Fixup some of the state for the next load */
1457	li	r10, 0
1458	li	r0, 0xff
1459	stb	r10, VCPU_XIVE_PUSHED(r9)
1460	stb	r10, (VCPU_XIVE_SAVED_STATE+3)(r9)
1461	stb	r0, (VCPU_XIVE_SAVED_STATE+4)(r9)
1462	eieio
14631:
1464#endif /* CONFIG_KVM_XICS */
1465
1466	/* If we came in through the P9 short path, go back out to C now */
1467	lwz	r0, STACK_SLOT_SHORT_PATH(r1)
1468	cmpwi	r0, 0
1469	bne	guest_exit_short_path
1470
1471	/* For hash guest, read the guest SLB and save it away */
1472	ld	r5, VCPU_KVM(r9)
1473	lbz	r0, KVM_RADIX(r5)
1474	li	r5, 0
1475	cmpwi	r0, 0
1476	bne	3f			/* for radix, save 0 entries */
1477	lwz	r0,VCPU_SLB_NR(r9)	/* number of entries in SLB */
1478	mtctr	r0
1479	li	r6,0
1480	addi	r7,r9,VCPU_SLB
14811:	slbmfee	r8,r6
1482	andis.	r0,r8,SLB_ESID_V@h
1483	beq	2f
1484	add	r8,r8,r6		/* put index in */
1485	slbmfev	r3,r6
1486	std	r8,VCPU_SLB_E(r7)
1487	std	r3,VCPU_SLB_V(r7)
1488	addi	r7,r7,VCPU_SLB_SIZE
1489	addi	r5,r5,1
14902:	addi	r6,r6,1
1491	bdnz	1b
1492	/* Finally clear out the SLB */
1493	li	r0,0
1494	slbmte	r0,r0
1495	slbia
1496	ptesync
14973:	stw	r5,VCPU_SLB_MAX(r9)
1498
1499	/* load host SLB entries */
1500BEGIN_MMU_FTR_SECTION
1501	b	0f
1502END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
1503	ld	r8,PACA_SLBSHADOWPTR(r13)
1504
1505	.rept	SLB_NUM_BOLTED
1506	li	r3, SLBSHADOW_SAVEAREA
1507	LDX_BE	r5, r8, r3
1508	addi	r3, r3, 8
1509	LDX_BE	r6, r8, r3
1510	andis.	r7,r5,SLB_ESID_V@h
1511	beq	1f
1512	slbmte	r6,r5
15131:	addi	r8,r8,16
1514	.endr
15150:
1516
1517guest_bypass:
1518	stw	r12, STACK_SLOT_TRAP(r1)
1519
1520	/* Save DEC */
1521	/* Do this before kvmhv_commence_exit so we know TB is guest TB */
1522	ld	r3, HSTATE_KVM_VCORE(r13)
1523	mfspr	r5,SPRN_DEC
1524	mftb	r6
1525	/* On P9, if the guest has large decr enabled, don't sign extend */
1526BEGIN_FTR_SECTION
1527	ld	r4, VCORE_LPCR(r3)
1528	andis.	r4, r4, LPCR_LD@h
1529	bne	16f
1530END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1531	extsw	r5,r5
153216:	add	r5,r5,r6
1533	/* r5 is a guest timebase value here, convert to host TB */
1534	ld	r4,VCORE_TB_OFFSET_APPL(r3)
1535	subf	r5,r4,r5
1536	std	r5,VCPU_DEC_EXPIRES(r9)
1537
1538	/* Increment exit count, poke other threads to exit */
1539	mr 	r3, r12
1540	bl	kvmhv_commence_exit
1541	nop
1542	ld	r9, HSTATE_KVM_VCPU(r13)
1543
1544	/* Stop others sending VCPU interrupts to this physical CPU */
1545	li	r0, -1
1546	stw	r0, VCPU_CPU(r9)
1547	stw	r0, VCPU_THREAD_CPU(r9)
1548
1549	/* Save guest CTRL register, set runlatch to 1 */
1550	mfspr	r6,SPRN_CTRLF
1551	stw	r6,VCPU_CTRL(r9)
1552	andi.	r0,r6,1
1553	bne	4f
1554	ori	r6,r6,1
1555	mtspr	SPRN_CTRLT,r6
15564:
1557	/*
1558	 * Save the guest PURR/SPURR
1559	 */
1560	mfspr	r5,SPRN_PURR
1561	mfspr	r6,SPRN_SPURR
1562	ld	r7,VCPU_PURR(r9)
1563	ld	r8,VCPU_SPURR(r9)
1564	std	r5,VCPU_PURR(r9)
1565	std	r6,VCPU_SPURR(r9)
1566	subf	r5,r7,r5
1567	subf	r6,r8,r6
1568
1569	/*
1570	 * Restore host PURR/SPURR and add guest times
1571	 * so that the time in the guest gets accounted.
1572	 */
1573	ld	r3,HSTATE_PURR(r13)
1574	ld	r4,HSTATE_SPURR(r13)
1575	add	r3,r3,r5
1576	add	r4,r4,r6
1577	mtspr	SPRN_PURR,r3
1578	mtspr	SPRN_SPURR,r4
1579
1580BEGIN_FTR_SECTION
1581	b	8f
1582END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
1583	/* Save POWER8-specific registers */
1584	mfspr	r5, SPRN_IAMR
1585	mfspr	r6, SPRN_PSPB
1586	mfspr	r7, SPRN_FSCR
1587	std	r5, VCPU_IAMR(r9)
1588	stw	r6, VCPU_PSPB(r9)
1589	std	r7, VCPU_FSCR(r9)
1590	mfspr	r5, SPRN_IC
1591	mfspr	r7, SPRN_TAR
1592	std	r5, VCPU_IC(r9)
1593	std	r7, VCPU_TAR(r9)
1594	mfspr	r8, SPRN_EBBHR
1595	std	r8, VCPU_EBBHR(r9)
1596	mfspr	r5, SPRN_EBBRR
1597	mfspr	r6, SPRN_BESCR
1598	mfspr	r7, SPRN_PID
1599	mfspr	r8, SPRN_WORT
1600	std	r5, VCPU_EBBRR(r9)
1601	std	r6, VCPU_BESCR(r9)
1602	stw	r7, VCPU_GUEST_PID(r9)
1603	std	r8, VCPU_WORT(r9)
1604BEGIN_FTR_SECTION
1605	mfspr	r5, SPRN_TCSCR
1606	mfspr	r6, SPRN_ACOP
1607	mfspr	r7, SPRN_CSIGR
1608	mfspr	r8, SPRN_TACR
1609	std	r5, VCPU_TCSCR(r9)
1610	std	r6, VCPU_ACOP(r9)
1611	std	r7, VCPU_CSIGR(r9)
1612	std	r8, VCPU_TACR(r9)
1613FTR_SECTION_ELSE
1614	mfspr	r5, SPRN_TIDR
1615	mfspr	r6, SPRN_PSSCR
1616	std	r5, VCPU_TID(r9)
1617	rldicl	r6, r6, 4, 50		/* r6 &= PSSCR_GUEST_VIS */
1618	rotldi	r6, r6, 60
1619	std	r6, VCPU_PSSCR(r9)
1620	/* Restore host HFSCR value */
1621	ld	r7, STACK_SLOT_HFSCR(r1)
1622	mtspr	SPRN_HFSCR, r7
1623ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
1624	/*
1625	 * Restore various registers to 0, where non-zero values
1626	 * set by the guest could disrupt the host.
1627	 */
1628	li	r0, 0
1629	mtspr	SPRN_PSPB, r0
1630	mtspr	SPRN_WORT, r0
1631BEGIN_FTR_SECTION
1632	mtspr	SPRN_TCSCR, r0
1633	/* Set MMCRS to 1<<31 to freeze and disable the SPMC counters */
1634	li	r0, 1
1635	sldi	r0, r0, 31
1636	mtspr	SPRN_MMCRS, r0
1637END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1638
1639	/* Save and restore AMR, IAMR and UAMOR before turning on the MMU */
1640	ld	r8, STACK_SLOT_IAMR(r1)
1641	mtspr	SPRN_IAMR, r8
1642
16438:	/* Power7 jumps back in here */
1644	mfspr	r5,SPRN_AMR
1645	mfspr	r6,SPRN_UAMOR
1646	std	r5,VCPU_AMR(r9)
1647	std	r6,VCPU_UAMOR(r9)
1648	ld	r5,STACK_SLOT_AMR(r1)
1649	ld	r6,STACK_SLOT_UAMOR(r1)
1650	mtspr	SPRN_AMR, r5
1651	mtspr	SPRN_UAMOR, r6
1652
1653	/* Switch DSCR back to host value */
1654	mfspr	r8, SPRN_DSCR
1655	ld	r7, HSTATE_DSCR(r13)
1656	std	r8, VCPU_DSCR(r9)
1657	mtspr	SPRN_DSCR, r7
1658
1659	/* Save non-volatile GPRs */
1660	std	r14, VCPU_GPR(R14)(r9)
1661	std	r15, VCPU_GPR(R15)(r9)
1662	std	r16, VCPU_GPR(R16)(r9)
1663	std	r17, VCPU_GPR(R17)(r9)
1664	std	r18, VCPU_GPR(R18)(r9)
1665	std	r19, VCPU_GPR(R19)(r9)
1666	std	r20, VCPU_GPR(R20)(r9)
1667	std	r21, VCPU_GPR(R21)(r9)
1668	std	r22, VCPU_GPR(R22)(r9)
1669	std	r23, VCPU_GPR(R23)(r9)
1670	std	r24, VCPU_GPR(R24)(r9)
1671	std	r25, VCPU_GPR(R25)(r9)
1672	std	r26, VCPU_GPR(R26)(r9)
1673	std	r27, VCPU_GPR(R27)(r9)
1674	std	r28, VCPU_GPR(R28)(r9)
1675	std	r29, VCPU_GPR(R29)(r9)
1676	std	r30, VCPU_GPR(R30)(r9)
1677	std	r31, VCPU_GPR(R31)(r9)
1678
1679	/* Save SPRGs */
1680	mfspr	r3, SPRN_SPRG0
1681	mfspr	r4, SPRN_SPRG1
1682	mfspr	r5, SPRN_SPRG2
1683	mfspr	r6, SPRN_SPRG3
1684	std	r3, VCPU_SPRG0(r9)
1685	std	r4, VCPU_SPRG1(r9)
1686	std	r5, VCPU_SPRG2(r9)
1687	std	r6, VCPU_SPRG3(r9)
1688
1689	/* save FP state */
1690	mr	r3, r9
1691	bl	kvmppc_save_fp
1692
1693#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1694/*
1695 * Branch around the call if both CPU_FTR_TM and
1696 * CPU_FTR_P9_TM_HV_ASSIST are off.
1697 */
1698BEGIN_FTR_SECTION
1699	b	91f
1700END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
1701	/*
1702	 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
1703	 */
1704	mr      r3, r9
1705	ld      r4, VCPU_MSR(r3)
1706	li	r5, 0			/* don't preserve non-vol regs */
1707	bl	kvmppc_save_tm_hv
1708	nop
1709	ld	r9, HSTATE_KVM_VCPU(r13)
171091:
1711#endif
1712
1713	/* Increment yield count if they have a VPA */
1714	ld	r8, VCPU_VPA(r9)	/* do they have a VPA? */
1715	cmpdi	r8, 0
1716	beq	25f
1717	li	r4, LPPACA_YIELDCOUNT
1718	LWZX_BE	r3, r8, r4
1719	addi	r3, r3, 1
1720	STWX_BE	r3, r8, r4
1721	li	r3, 1
1722	stb	r3, VCPU_VPA_DIRTY(r9)
172325:
1724	/* Save PMU registers if requested */
1725	/* r8 and cr0.eq are live here */
1726	mr	r3, r9
1727	li	r4, 1
1728	beq	21f			/* if no VPA, save PMU stuff anyway */
1729	lbz	r4, LPPACA_PMCINUSE(r8)
173021:	bl	kvmhv_save_guest_pmu
1731	ld	r9, HSTATE_KVM_VCPU(r13)
1732
1733	/* Restore host values of some registers */
1734BEGIN_FTR_SECTION
1735	ld	r5, STACK_SLOT_CIABR(r1)
1736	ld	r6, STACK_SLOT_DAWR(r1)
1737	ld	r7, STACK_SLOT_DAWRX(r1)
1738	mtspr	SPRN_CIABR, r5
1739	/*
1740	 * If the DAWR doesn't work, it's ok to write these here as
1741	 * this value should always be zero
1742	*/
1743	mtspr	SPRN_DAWR, r6
1744	mtspr	SPRN_DAWRX, r7
1745END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1746BEGIN_FTR_SECTION
1747	ld	r5, STACK_SLOT_TID(r1)
1748	ld	r6, STACK_SLOT_PSSCR(r1)
1749	ld	r7, STACK_SLOT_PID(r1)
1750	mtspr	SPRN_TIDR, r5
1751	mtspr	SPRN_PSSCR, r6
1752	mtspr	SPRN_PID, r7
1753END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1754
1755#ifdef CONFIG_PPC_RADIX_MMU
1756	/*
1757	 * Are we running hash or radix ?
1758	 */
1759	ld	r5, VCPU_KVM(r9)
1760	lbz	r0, KVM_RADIX(r5)
1761	cmpwi	cr2, r0, 0
1762	beq	cr2, 2f
1763
1764	/*
1765	 * Radix: do eieio; tlbsync; ptesync sequence in case we
1766	 * interrupted the guest between a tlbie and a ptesync.
1767	 */
1768	eieio
1769	tlbsync
1770	ptesync
1771
1772	/* Radix: Handle the case where the guest used an illegal PID */
1773	LOAD_REG_ADDR(r4, mmu_base_pid)
1774	lwz	r3, VCPU_GUEST_PID(r9)
1775	lwz	r5, 0(r4)
1776	cmpw	cr0,r3,r5
1777	blt	2f
1778
1779	/*
1780	 * Illegal PID, the HW might have prefetched and cached in the TLB
1781	 * some translations for the  LPID 0 / guest PID combination which
1782	 * Linux doesn't know about, so we need to flush that PID out of
1783	 * the TLB. First we need to set LPIDR to 0 so tlbiel applies to
1784	 * the right context.
1785	*/
1786	li	r0,0
1787	mtspr	SPRN_LPID,r0
1788	isync
1789
1790	/* Then do a congruence class local flush */
1791	ld	r6,VCPU_KVM(r9)
1792	lwz	r0,KVM_TLB_SETS(r6)
1793	mtctr	r0
1794	li	r7,0x400		/* IS field = 0b01 */
1795	ptesync
1796	sldi	r0,r3,32		/* RS has PID */
17971:	PPC_TLBIEL(7,0,2,1,1)		/* RIC=2, PRS=1, R=1 */
1798	addi	r7,r7,0x1000
1799	bdnz	1b
1800	ptesync
1801
18022:
1803#endif /* CONFIG_PPC_RADIX_MMU */
1804
1805	/*
1806	 * POWER7/POWER8 guest -> host partition switch code.
1807	 * We don't have to lock against tlbies but we do
1808	 * have to coordinate the hardware threads.
1809	 * Here STACK_SLOT_TRAP(r1) contains the trap number.
1810	 */
1811kvmhv_switch_to_host:
1812	/* Secondary threads wait for primary to do partition switch */
1813	ld	r5,HSTATE_KVM_VCORE(r13)
1814	ld	r4,VCORE_KVM(r5)	/* pointer to struct kvm */
1815	lbz	r3,HSTATE_PTID(r13)
1816	cmpwi	r3,0
1817	beq	15f
1818	HMT_LOW
181913:	lbz	r3,VCORE_IN_GUEST(r5)
1820	cmpwi	r3,0
1821	bne	13b
1822	HMT_MEDIUM
1823	b	16f
1824
1825	/* Primary thread waits for all the secondaries to exit guest */
182615:	lwz	r3,VCORE_ENTRY_EXIT(r5)
1827	rlwinm	r0,r3,32-8,0xff
1828	clrldi	r3,r3,56
1829	cmpw	r3,r0
1830	bne	15b
1831	isync
1832
1833	/* Did we actually switch to the guest at all? */
1834	lbz	r6, VCORE_IN_GUEST(r5)
1835	cmpwi	r6, 0
1836	beq	19f
1837
1838	/* Primary thread switches back to host partition */
1839	lwz	r7,KVM_HOST_LPID(r4)
1840BEGIN_FTR_SECTION
1841	ld	r6,KVM_HOST_SDR1(r4)
1842	li	r8,LPID_RSVD		/* switch to reserved LPID */
1843	mtspr	SPRN_LPID,r8
1844	ptesync
1845	mtspr	SPRN_SDR1,r6		/* switch to host page table */
1846END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1847	mtspr	SPRN_LPID,r7
1848	isync
1849
1850BEGIN_FTR_SECTION
1851	/* DPDES and VTB are shared between threads */
1852	mfspr	r7, SPRN_DPDES
1853	mfspr	r8, SPRN_VTB
1854	std	r7, VCORE_DPDES(r5)
1855	std	r8, VCORE_VTB(r5)
1856	/* clear DPDES so we don't get guest doorbells in the host */
1857	li	r8, 0
1858	mtspr	SPRN_DPDES, r8
1859END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1860
1861	/* Subtract timebase offset from timebase */
1862	ld	r8, VCORE_TB_OFFSET_APPL(r5)
1863	cmpdi	r8,0
1864	beq	17f
1865	li	r0, 0
1866	std	r0, VCORE_TB_OFFSET_APPL(r5)
1867	mftb	r6			/* current guest timebase */
1868	subf	r8,r8,r6
1869	mtspr	SPRN_TBU40,r8		/* update upper 40 bits */
1870	mftb	r7			/* check if lower 24 bits overflowed */
1871	clrldi	r6,r6,40
1872	clrldi	r7,r7,40
1873	cmpld	r7,r6
1874	bge	17f
1875	addis	r8,r8,0x100		/* if so, increment upper 40 bits */
1876	mtspr	SPRN_TBU40,r8
1877
187817:
1879	/*
1880	 * If this is an HMI, we called kvmppc_realmode_hmi_handler
1881	 * above, which may or may not have already called
1882	 * kvmppc_subcore_exit_guest.  Fortunately, all that
1883	 * kvmppc_subcore_exit_guest does is clear a flag, so calling
1884	 * it again here is benign even if kvmppc_realmode_hmi_handler
1885	 * has already called it.
1886	 */
1887	bl	kvmppc_subcore_exit_guest
1888	nop
188930:	ld	r5,HSTATE_KVM_VCORE(r13)
1890	ld	r4,VCORE_KVM(r5)	/* pointer to struct kvm */
1891
1892	/* Reset PCR */
1893	ld	r0, VCORE_PCR(r5)
1894	cmpdi	r0, 0
1895	beq	18f
1896	li	r0, 0
1897	mtspr	SPRN_PCR, r0
189818:
1899	/* Signal secondary CPUs to continue */
1900	stb	r0,VCORE_IN_GUEST(r5)
190119:	lis	r8,0x7fff		/* MAX_INT@h */
1902	mtspr	SPRN_HDEC,r8
1903
190416:
1905BEGIN_FTR_SECTION
1906	/* On POWER9 with HPT-on-radix we need to wait for all other threads */
1907	ld	r3, HSTATE_SPLIT_MODE(r13)
1908	cmpdi	r3, 0
1909	beq	47f
1910	lwz	r8, KVM_SPLIT_DO_RESTORE(r3)
1911	cmpwi	r8, 0
1912	beq	47f
1913	bl	kvmhv_p9_restore_lpcr
1914	nop
1915	b	48f
191647:
1917END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1918	ld	r8,KVM_HOST_LPCR(r4)
1919	mtspr	SPRN_LPCR,r8
1920	isync
192148:
1922#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1923	/* Finish timing, if we have a vcpu */
1924	ld	r4, HSTATE_KVM_VCPU(r13)
1925	cmpdi	r4, 0
1926	li	r3, 0
1927	beq	2f
1928	bl	kvmhv_accumulate_time
19292:
1930#endif
1931	/* Unset guest mode */
1932	li	r0, KVM_GUEST_MODE_NONE
1933	stb	r0, HSTATE_IN_GUEST(r13)
1934
1935	lwz	r12, STACK_SLOT_TRAP(r1)	/* return trap # in r12 */
1936	ld	r0, SFS+PPC_LR_STKOFF(r1)
1937	addi	r1, r1, SFS
1938	mtlr	r0
1939	blr
1940
1941kvmppc_guest_external:
1942	/* External interrupt, first check for host_ipi. If this is
1943	 * set, we know the host wants us out so let's do it now
1944	 */
1945	bl	kvmppc_read_intr
1946
1947	/*
1948	 * Restore the active volatile registers after returning from
1949	 * a C function.
1950	 */
1951	ld	r9, HSTATE_KVM_VCPU(r13)
1952	li	r12, BOOK3S_INTERRUPT_EXTERNAL
1953
1954	/*
1955	 * kvmppc_read_intr return codes:
1956	 *
1957	 * Exit to host (r3 > 0)
1958	 *   1 An interrupt is pending that needs to be handled by the host
1959	 *     Exit guest and return to host by branching to guest_exit_cont
1960	 *
1961	 *   2 Passthrough that needs completion in the host
1962	 *     Exit guest and return to host by branching to guest_exit_cont
1963	 *     However, we also set r12 to BOOK3S_INTERRUPT_HV_RM_HARD
1964	 *     to indicate to the host to complete handling the interrupt
1965	 *
1966	 * Before returning to guest, we check if any CPU is heading out
1967	 * to the host and if so, we head out also. If no CPUs are heading
1968	 * check return values <= 0.
1969	 *
1970	 * Return to guest (r3 <= 0)
1971	 *  0 No external interrupt is pending
1972	 * -1 A guest wakeup IPI (which has now been cleared)
1973	 *    In either case, we return to guest to deliver any pending
1974	 *    guest interrupts.
1975	 *
1976	 * -2 A PCI passthrough external interrupt was handled
1977	 *    (interrupt was delivered directly to guest)
1978	 *    Return to guest to deliver any pending guest interrupts.
1979	 */
1980
1981	cmpdi	r3, 1
1982	ble	1f
1983
1984	/* Return code = 2 */
1985	li	r12, BOOK3S_INTERRUPT_HV_RM_HARD
1986	stw	r12, VCPU_TRAP(r9)
1987	b	guest_exit_cont
1988
19891:	/* Return code <= 1 */
1990	cmpdi	r3, 0
1991	bgt	guest_exit_cont
1992
1993	/* Return code <= 0 */
1994maybe_reenter_guest:
1995	ld	r5, HSTATE_KVM_VCORE(r13)
1996	lwz	r0, VCORE_ENTRY_EXIT(r5)
1997	cmpwi	r0, 0x100
1998	mr	r4, r9
1999	blt	deliver_guest_interrupt
2000	b	guest_exit_cont
2001
2002#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2003/*
2004 * Softpatch interrupt for transactional memory emulation cases
2005 * on POWER9 DD2.2.  This is early in the guest exit path - we
2006 * haven't saved registers or done a treclaim yet.
2007 */
2008kvmppc_tm_emul:
2009	/* Save instruction image in HEIR */
2010	mfspr	r3, SPRN_HEIR
2011	stw	r3, VCPU_HEIR(r9)
2012
2013	/*
2014	 * The cases we want to handle here are those where the guest
2015	 * is in real suspend mode and is trying to transition to
2016	 * transactional mode.
2017	 */
2018	lbz	r0, HSTATE_FAKE_SUSPEND(r13)
2019	cmpwi	r0, 0		/* keep exiting guest if in fake suspend */
2020	bne	guest_exit_cont
2021	rldicl	r3, r11, 64 - MSR_TS_S_LG, 62
2022	cmpwi	r3, 1		/* or if not in suspend state */
2023	bne	guest_exit_cont
2024
2025	/* Call C code to do the emulation */
2026	mr	r3, r9
2027	bl	kvmhv_p9_tm_emulation_early
2028	nop
2029	ld	r9, HSTATE_KVM_VCPU(r13)
2030	li	r12, BOOK3S_INTERRUPT_HV_SOFTPATCH
2031	cmpwi	r3, 0
2032	beq	guest_exit_cont		/* continue exiting if not handled */
2033	ld	r10, VCPU_PC(r9)
2034	ld	r11, VCPU_MSR(r9)
2035	b	fast_interrupt_c_return	/* go back to guest if handled */
2036#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
2037
2038/*
2039 * Check whether an HDSI is an HPTE not found fault or something else.
2040 * If it is an HPTE not found fault that is due to the guest accessing
2041 * a page that they have mapped but which we have paged out, then
2042 * we continue on with the guest exit path.  In all other cases,
2043 * reflect the HDSI to the guest as a DSI.
2044 */
2045kvmppc_hdsi:
2046	ld	r3, VCPU_KVM(r9)
2047	lbz	r0, KVM_RADIX(r3)
2048	mfspr	r4, SPRN_HDAR
2049	mfspr	r6, SPRN_HDSISR
2050BEGIN_FTR_SECTION
2051	/* Look for DSISR canary. If we find it, retry instruction */
2052	cmpdi	r6, 0x7fff
2053	beq	6f
2054END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2055	cmpwi	r0, 0
2056	bne	.Lradix_hdsi		/* on radix, just save DAR/DSISR/ASDR */
2057	/* HPTE not found fault or protection fault? */
2058	andis.	r0, r6, (DSISR_NOHPTE | DSISR_PROTFAULT)@h
2059	beq	1f			/* if not, send it to the guest */
2060	andi.	r0, r11, MSR_DR		/* data relocation enabled? */
2061	beq	3f
2062BEGIN_FTR_SECTION
2063	mfspr	r5, SPRN_ASDR		/* on POWER9, use ASDR to get VSID */
2064	b	4f
2065END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2066	clrrdi	r0, r4, 28
2067	PPC_SLBFEE_DOT(R5, R0)		/* if so, look up SLB */
2068	li	r0, BOOK3S_INTERRUPT_DATA_SEGMENT
2069	bne	7f			/* if no SLB entry found */
20704:	std	r4, VCPU_FAULT_DAR(r9)
2071	stw	r6, VCPU_FAULT_DSISR(r9)
2072
2073	/* Search the hash table. */
2074	mr	r3, r9			/* vcpu pointer */
2075	li	r7, 1			/* data fault */
2076	bl	kvmppc_hpte_hv_fault
2077	ld	r9, HSTATE_KVM_VCPU(r13)
2078	ld	r10, VCPU_PC(r9)
2079	ld	r11, VCPU_MSR(r9)
2080	li	r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
2081	cmpdi	r3, 0			/* retry the instruction */
2082	beq	6f
2083	cmpdi	r3, -1			/* handle in kernel mode */
2084	beq	guest_exit_cont
2085	cmpdi	r3, -2			/* MMIO emulation; need instr word */
2086	beq	2f
2087
2088	/* Synthesize a DSI (or DSegI) for the guest */
2089	ld	r4, VCPU_FAULT_DAR(r9)
2090	mr	r6, r3
20911:	li	r0, BOOK3S_INTERRUPT_DATA_STORAGE
2092	mtspr	SPRN_DSISR, r6
20937:	mtspr	SPRN_DAR, r4
2094	mtspr	SPRN_SRR0, r10
2095	mtspr	SPRN_SRR1, r11
2096	mr	r10, r0
2097	bl	kvmppc_msr_interrupt
2098fast_interrupt_c_return:
20996:	ld	r7, VCPU_CTR(r9)
2100	ld	r8, VCPU_XER(r9)
2101	mtctr	r7
2102	mtxer	r8
2103	mr	r4, r9
2104	b	fast_guest_return
2105
21063:	ld	r5, VCPU_KVM(r9)	/* not relocated, use VRMA */
2107	ld	r5, KVM_VRMA_SLB_V(r5)
2108	b	4b
2109
2110	/* If this is for emulated MMIO, load the instruction word */
21112:	li	r8, KVM_INST_FETCH_FAILED	/* In case lwz faults */
2112
2113	/* Set guest mode to 'jump over instruction' so if lwz faults
2114	 * we'll just continue at the next IP. */
2115	li	r0, KVM_GUEST_MODE_SKIP
2116	stb	r0, HSTATE_IN_GUEST(r13)
2117
2118	/* Do the access with MSR:DR enabled */
2119	mfmsr	r3
2120	ori	r4, r3, MSR_DR		/* Enable paging for data */
2121	mtmsrd	r4
2122	lwz	r8, 0(r10)
2123	mtmsrd	r3
2124
2125	/* Store the result */
2126	stw	r8, VCPU_LAST_INST(r9)
2127
2128	/* Unset guest mode. */
2129	li	r0, KVM_GUEST_MODE_HOST_HV
2130	stb	r0, HSTATE_IN_GUEST(r13)
2131	b	guest_exit_cont
2132
2133.Lradix_hdsi:
2134	std	r4, VCPU_FAULT_DAR(r9)
2135	stw	r6, VCPU_FAULT_DSISR(r9)
2136.Lradix_hisi:
2137	mfspr	r5, SPRN_ASDR
2138	std	r5, VCPU_FAULT_GPA(r9)
2139	b	guest_exit_cont
2140
2141/*
2142 * Similarly for an HISI, reflect it to the guest as an ISI unless
2143 * it is an HPTE not found fault for a page that we have paged out.
2144 */
2145kvmppc_hisi:
2146	ld	r3, VCPU_KVM(r9)
2147	lbz	r0, KVM_RADIX(r3)
2148	cmpwi	r0, 0
2149	bne	.Lradix_hisi		/* for radix, just save ASDR */
2150	andis.	r0, r11, SRR1_ISI_NOPT@h
2151	beq	1f
2152	andi.	r0, r11, MSR_IR		/* instruction relocation enabled? */
2153	beq	3f
2154BEGIN_FTR_SECTION
2155	mfspr	r5, SPRN_ASDR		/* on POWER9, use ASDR to get VSID */
2156	b	4f
2157END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2158	clrrdi	r0, r10, 28
2159	PPC_SLBFEE_DOT(R5, R0)		/* if so, look up SLB */
2160	li	r0, BOOK3S_INTERRUPT_INST_SEGMENT
2161	bne	7f			/* if no SLB entry found */
21624:
2163	/* Search the hash table. */
2164	mr	r3, r9			/* vcpu pointer */
2165	mr	r4, r10
2166	mr	r6, r11
2167	li	r7, 0			/* instruction fault */
2168	bl	kvmppc_hpte_hv_fault
2169	ld	r9, HSTATE_KVM_VCPU(r13)
2170	ld	r10, VCPU_PC(r9)
2171	ld	r11, VCPU_MSR(r9)
2172	li	r12, BOOK3S_INTERRUPT_H_INST_STORAGE
2173	cmpdi	r3, 0			/* retry the instruction */
2174	beq	fast_interrupt_c_return
2175	cmpdi	r3, -1			/* handle in kernel mode */
2176	beq	guest_exit_cont
2177
2178	/* Synthesize an ISI (or ISegI) for the guest */
2179	mr	r11, r3
21801:	li	r0, BOOK3S_INTERRUPT_INST_STORAGE
21817:	mtspr	SPRN_SRR0, r10
2182	mtspr	SPRN_SRR1, r11
2183	mr	r10, r0
2184	bl	kvmppc_msr_interrupt
2185	b	fast_interrupt_c_return
2186
21873:	ld	r6, VCPU_KVM(r9)	/* not relocated, use VRMA */
2188	ld	r5, KVM_VRMA_SLB_V(r6)
2189	b	4b
2190
2191/*
2192 * Try to handle an hcall in real mode.
2193 * Returns to the guest if we handle it, or continues on up to
2194 * the kernel if we can't (i.e. if we don't have a handler for
2195 * it, or if the handler returns H_TOO_HARD).
2196 *
2197 * r5 - r8 contain hcall args,
2198 * r9 = vcpu, r10 = pc, r11 = msr, r12 = trap, r13 = paca
2199 */
2200hcall_try_real_mode:
2201	ld	r3,VCPU_GPR(R3)(r9)
2202	andi.	r0,r11,MSR_PR
2203	/* sc 1 from userspace - reflect to guest syscall */
2204	bne	sc_1_fast_return
2205	/* sc 1 from nested guest - give it to L1 to handle */
2206	ld	r0, VCPU_NESTED(r9)
2207	cmpdi	r0, 0
2208	bne	guest_exit_cont
2209	clrrdi	r3,r3,2
2210	cmpldi	r3,hcall_real_table_end - hcall_real_table
2211	bge	guest_exit_cont
2212	/* See if this hcall is enabled for in-kernel handling */
2213	ld	r4, VCPU_KVM(r9)
2214	srdi	r0, r3, 8	/* r0 = (r3 / 4) >> 6 */
2215	sldi	r0, r0, 3	/* index into kvm->arch.enabled_hcalls[] */
2216	add	r4, r4, r0
2217	ld	r0, KVM_ENABLED_HCALLS(r4)
2218	rlwinm	r4, r3, 32-2, 0x3f	/* r4 = (r3 / 4) & 0x3f */
2219	srd	r0, r0, r4
2220	andi.	r0, r0, 1
2221	beq	guest_exit_cont
2222	/* Get pointer to handler, if any, and call it */
2223	LOAD_REG_ADDR(r4, hcall_real_table)
2224	lwax	r3,r3,r4
2225	cmpwi	r3,0
2226	beq	guest_exit_cont
2227	add	r12,r3,r4
2228	mtctr	r12
2229	mr	r3,r9		/* get vcpu pointer */
2230	ld	r4,VCPU_GPR(R4)(r9)
2231	bctrl
2232	cmpdi	r3,H_TOO_HARD
2233	beq	hcall_real_fallback
2234	ld	r4,HSTATE_KVM_VCPU(r13)
2235	std	r3,VCPU_GPR(R3)(r4)
2236	ld	r10,VCPU_PC(r4)
2237	ld	r11,VCPU_MSR(r4)
2238	b	fast_guest_return
2239
2240sc_1_fast_return:
2241	mtspr	SPRN_SRR0,r10
2242	mtspr	SPRN_SRR1,r11
2243	li	r10, BOOK3S_INTERRUPT_SYSCALL
2244	bl	kvmppc_msr_interrupt
2245	mr	r4,r9
2246	b	fast_guest_return
2247
2248	/* We've attempted a real mode hcall, but it's punted it back
2249	 * to userspace.  We need to restore some clobbered volatiles
2250	 * before resuming the pass-it-to-qemu path */
2251hcall_real_fallback:
2252	li	r12,BOOK3S_INTERRUPT_SYSCALL
2253	ld	r9, HSTATE_KVM_VCPU(r13)
2254
2255	b	guest_exit_cont
2256
2257	.globl	hcall_real_table
2258hcall_real_table:
2259	.long	0		/* 0 - unused */
2260	.long	DOTSYM(kvmppc_h_remove) - hcall_real_table
2261	.long	DOTSYM(kvmppc_h_enter) - hcall_real_table
2262	.long	DOTSYM(kvmppc_h_read) - hcall_real_table
2263	.long	DOTSYM(kvmppc_h_clear_mod) - hcall_real_table
2264	.long	DOTSYM(kvmppc_h_clear_ref) - hcall_real_table
2265	.long	DOTSYM(kvmppc_h_protect) - hcall_real_table
2266#ifdef CONFIG_SPAPR_TCE_IOMMU
2267	.long	DOTSYM(kvmppc_h_get_tce) - hcall_real_table
2268	.long	DOTSYM(kvmppc_rm_h_put_tce) - hcall_real_table
2269#else
2270	.long	0		/* 0x1c */
2271	.long	0		/* 0x20 */
2272#endif
2273	.long	0		/* 0x24 - H_SET_SPRG0 */
2274	.long	DOTSYM(kvmppc_h_set_dabr) - hcall_real_table
2275	.long	DOTSYM(kvmppc_rm_h_page_init) - hcall_real_table
2276	.long	0		/* 0x30 */
2277	.long	0		/* 0x34 */
2278	.long	0		/* 0x38 */
2279	.long	0		/* 0x3c */
2280	.long	0		/* 0x40 */
2281	.long	0		/* 0x44 */
2282	.long	0		/* 0x48 */
2283	.long	0		/* 0x4c */
2284	.long	0		/* 0x50 */
2285	.long	0		/* 0x54 */
2286	.long	0		/* 0x58 */
2287	.long	0		/* 0x5c */
2288	.long	0		/* 0x60 */
2289#ifdef CONFIG_KVM_XICS
2290	.long	DOTSYM(kvmppc_rm_h_eoi) - hcall_real_table
2291	.long	DOTSYM(kvmppc_rm_h_cppr) - hcall_real_table
2292	.long	DOTSYM(kvmppc_rm_h_ipi) - hcall_real_table
2293	.long	DOTSYM(kvmppc_rm_h_ipoll) - hcall_real_table
2294	.long	DOTSYM(kvmppc_rm_h_xirr) - hcall_real_table
2295#else
2296	.long	0		/* 0x64 - H_EOI */
2297	.long	0		/* 0x68 - H_CPPR */
2298	.long	0		/* 0x6c - H_IPI */
2299	.long	0		/* 0x70 - H_IPOLL */
2300	.long	0		/* 0x74 - H_XIRR */
2301#endif
2302	.long	0		/* 0x78 */
2303	.long	0		/* 0x7c */
2304	.long	0		/* 0x80 */
2305	.long	0		/* 0x84 */
2306	.long	0		/* 0x88 */
2307	.long	0		/* 0x8c */
2308	.long	0		/* 0x90 */
2309	.long	0		/* 0x94 */
2310	.long	0		/* 0x98 */
2311	.long	0		/* 0x9c */
2312	.long	0		/* 0xa0 */
2313	.long	0		/* 0xa4 */
2314	.long	0		/* 0xa8 */
2315	.long	0		/* 0xac */
2316	.long	0		/* 0xb0 */
2317	.long	0		/* 0xb4 */
2318	.long	0		/* 0xb8 */
2319	.long	0		/* 0xbc */
2320	.long	0		/* 0xc0 */
2321	.long	0		/* 0xc4 */
2322	.long	0		/* 0xc8 */
2323	.long	0		/* 0xcc */
2324	.long	0		/* 0xd0 */
2325	.long	0		/* 0xd4 */
2326	.long	0		/* 0xd8 */
2327	.long	0		/* 0xdc */
2328	.long	DOTSYM(kvmppc_h_cede) - hcall_real_table
2329	.long	DOTSYM(kvmppc_rm_h_confer) - hcall_real_table
2330	.long	0		/* 0xe8 */
2331	.long	0		/* 0xec */
2332	.long	0		/* 0xf0 */
2333	.long	0		/* 0xf4 */
2334	.long	0		/* 0xf8 */
2335	.long	0		/* 0xfc */
2336	.long	0		/* 0x100 */
2337	.long	0		/* 0x104 */
2338	.long	0		/* 0x108 */
2339	.long	0		/* 0x10c */
2340	.long	0		/* 0x110 */
2341	.long	0		/* 0x114 */
2342	.long	0		/* 0x118 */
2343	.long	0		/* 0x11c */
2344	.long	0		/* 0x120 */
2345	.long	DOTSYM(kvmppc_h_bulk_remove) - hcall_real_table
2346	.long	0		/* 0x128 */
2347	.long	0		/* 0x12c */
2348	.long	0		/* 0x130 */
2349	.long	DOTSYM(kvmppc_h_set_xdabr) - hcall_real_table
2350#ifdef CONFIG_SPAPR_TCE_IOMMU
2351	.long	DOTSYM(kvmppc_rm_h_stuff_tce) - hcall_real_table
2352	.long	DOTSYM(kvmppc_rm_h_put_tce_indirect) - hcall_real_table
2353#else
2354	.long	0		/* 0x138 */
2355	.long	0		/* 0x13c */
2356#endif
2357	.long	0		/* 0x140 */
2358	.long	0		/* 0x144 */
2359	.long	0		/* 0x148 */
2360	.long	0		/* 0x14c */
2361	.long	0		/* 0x150 */
2362	.long	0		/* 0x154 */
2363	.long	0		/* 0x158 */
2364	.long	0		/* 0x15c */
2365	.long	0		/* 0x160 */
2366	.long	0		/* 0x164 */
2367	.long	0		/* 0x168 */
2368	.long	0		/* 0x16c */
2369	.long	0		/* 0x170 */
2370	.long	0		/* 0x174 */
2371	.long	0		/* 0x178 */
2372	.long	0		/* 0x17c */
2373	.long	0		/* 0x180 */
2374	.long	0		/* 0x184 */
2375	.long	0		/* 0x188 */
2376	.long	0		/* 0x18c */
2377	.long	0		/* 0x190 */
2378	.long	0		/* 0x194 */
2379	.long	0		/* 0x198 */
2380	.long	0		/* 0x19c */
2381	.long	0		/* 0x1a0 */
2382	.long	0		/* 0x1a4 */
2383	.long	0		/* 0x1a8 */
2384	.long	0		/* 0x1ac */
2385	.long	0		/* 0x1b0 */
2386	.long	0		/* 0x1b4 */
2387	.long	0		/* 0x1b8 */
2388	.long	0		/* 0x1bc */
2389	.long	0		/* 0x1c0 */
2390	.long	0		/* 0x1c4 */
2391	.long	0		/* 0x1c8 */
2392	.long	0		/* 0x1cc */
2393	.long	0		/* 0x1d0 */
2394	.long	0		/* 0x1d4 */
2395	.long	0		/* 0x1d8 */
2396	.long	0		/* 0x1dc */
2397	.long	0		/* 0x1e0 */
2398	.long	0		/* 0x1e4 */
2399	.long	0		/* 0x1e8 */
2400	.long	0		/* 0x1ec */
2401	.long	0		/* 0x1f0 */
2402	.long	0		/* 0x1f4 */
2403	.long	0		/* 0x1f8 */
2404	.long	0		/* 0x1fc */
2405	.long	0		/* 0x200 */
2406	.long	0		/* 0x204 */
2407	.long	0		/* 0x208 */
2408	.long	0		/* 0x20c */
2409	.long	0		/* 0x210 */
2410	.long	0		/* 0x214 */
2411	.long	0		/* 0x218 */
2412	.long	0		/* 0x21c */
2413	.long	0		/* 0x220 */
2414	.long	0		/* 0x224 */
2415	.long	0		/* 0x228 */
2416	.long	0		/* 0x22c */
2417	.long	0		/* 0x230 */
2418	.long	0		/* 0x234 */
2419	.long	0		/* 0x238 */
2420	.long	0		/* 0x23c */
2421	.long	0		/* 0x240 */
2422	.long	0		/* 0x244 */
2423	.long	0		/* 0x248 */
2424	.long	0		/* 0x24c */
2425	.long	0		/* 0x250 */
2426	.long	0		/* 0x254 */
2427	.long	0		/* 0x258 */
2428	.long	0		/* 0x25c */
2429	.long	0		/* 0x260 */
2430	.long	0		/* 0x264 */
2431	.long	0		/* 0x268 */
2432	.long	0		/* 0x26c */
2433	.long	0		/* 0x270 */
2434	.long	0		/* 0x274 */
2435	.long	0		/* 0x278 */
2436	.long	0		/* 0x27c */
2437	.long	0		/* 0x280 */
2438	.long	0		/* 0x284 */
2439	.long	0		/* 0x288 */
2440	.long	0		/* 0x28c */
2441	.long	0		/* 0x290 */
2442	.long	0		/* 0x294 */
2443	.long	0		/* 0x298 */
2444	.long	0		/* 0x29c */
2445	.long	0		/* 0x2a0 */
2446	.long	0		/* 0x2a4 */
2447	.long	0		/* 0x2a8 */
2448	.long	0		/* 0x2ac */
2449	.long	0		/* 0x2b0 */
2450	.long	0		/* 0x2b4 */
2451	.long	0		/* 0x2b8 */
2452	.long	0		/* 0x2bc */
2453	.long	0		/* 0x2c0 */
2454	.long	0		/* 0x2c4 */
2455	.long	0		/* 0x2c8 */
2456	.long	0		/* 0x2cc */
2457	.long	0		/* 0x2d0 */
2458	.long	0		/* 0x2d4 */
2459	.long	0		/* 0x2d8 */
2460	.long	0		/* 0x2dc */
2461	.long	0		/* 0x2e0 */
2462	.long	0		/* 0x2e4 */
2463	.long	0		/* 0x2e8 */
2464	.long	0		/* 0x2ec */
2465	.long	0		/* 0x2f0 */
2466	.long	0		/* 0x2f4 */
2467	.long	0		/* 0x2f8 */
2468#ifdef CONFIG_KVM_XICS
2469	.long	DOTSYM(kvmppc_rm_h_xirr_x) - hcall_real_table
2470#else
2471	.long	0		/* 0x2fc - H_XIRR_X*/
2472#endif
2473	.long	DOTSYM(kvmppc_h_random) - hcall_real_table
2474	.globl	hcall_real_table_end
2475hcall_real_table_end:
2476
2477_GLOBAL(kvmppc_h_set_xdabr)
2478EXPORT_SYMBOL_GPL(kvmppc_h_set_xdabr)
2479	andi.	r0, r5, DABRX_USER | DABRX_KERNEL
2480	beq	6f
2481	li	r0, DABRX_USER | DABRX_KERNEL | DABRX_BTI
2482	andc.	r0, r5, r0
2483	beq	3f
24846:	li	r3, H_PARAMETER
2485	blr
2486
2487_GLOBAL(kvmppc_h_set_dabr)
2488EXPORT_SYMBOL_GPL(kvmppc_h_set_dabr)
2489	li	r5, DABRX_USER | DABRX_KERNEL
24903:
2491BEGIN_FTR_SECTION
2492	b	2f
2493END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2494	std	r4,VCPU_DABR(r3)
2495	stw	r5, VCPU_DABRX(r3)
2496	mtspr	SPRN_DABRX, r5
2497	/* Work around P7 bug where DABR can get corrupted on mtspr */
24981:	mtspr	SPRN_DABR,r4
2499	mfspr	r5, SPRN_DABR
2500	cmpd	r4, r5
2501	bne	1b
2502	isync
2503	li	r3,0
2504	blr
2505
25062:
2507	LOAD_REG_ADDR(r11, dawr_force_enable)
2508	lbz	r11, 0(r11)
2509	cmpdi	r11, 0
2510	li	r3, H_HARDWARE
2511	beqlr
2512	/* Emulate H_SET_DABR/X on P8 for the sake of compat mode guests */
2513	rlwimi	r5, r4, 5, DAWRX_DR | DAWRX_DW
2514	rlwimi	r5, r4, 2, DAWRX_WT
2515	clrrdi	r4, r4, 3
2516	std	r4, VCPU_DAWR(r3)
2517	std	r5, VCPU_DAWRX(r3)
2518	mtspr	SPRN_DAWR, r4
2519	mtspr	SPRN_DAWRX, r5
2520	li	r3, 0
2521	blr
2522
2523_GLOBAL(kvmppc_h_cede)		/* r3 = vcpu pointer, r11 = msr, r13 = paca */
2524	ori	r11,r11,MSR_EE
2525	std	r11,VCPU_MSR(r3)
2526	li	r0,1
2527	stb	r0,VCPU_CEDED(r3)
2528	sync			/* order setting ceded vs. testing prodded */
2529	lbz	r5,VCPU_PRODDED(r3)
2530	cmpwi	r5,0
2531	bne	kvm_cede_prodded
2532	li	r12,0		/* set trap to 0 to say hcall is handled */
2533	stw	r12,VCPU_TRAP(r3)
2534	li	r0,H_SUCCESS
2535	std	r0,VCPU_GPR(R3)(r3)
2536
2537	/*
2538	 * Set our bit in the bitmask of napping threads unless all the
2539	 * other threads are already napping, in which case we send this
2540	 * up to the host.
2541	 */
2542	ld	r5,HSTATE_KVM_VCORE(r13)
2543	lbz	r6,HSTATE_PTID(r13)
2544	lwz	r8,VCORE_ENTRY_EXIT(r5)
2545	clrldi	r8,r8,56
2546	li	r0,1
2547	sld	r0,r0,r6
2548	addi	r6,r5,VCORE_NAPPING_THREADS
254931:	lwarx	r4,0,r6
2550	or	r4,r4,r0
2551	cmpw	r4,r8
2552	beq	kvm_cede_exit
2553	stwcx.	r4,0,r6
2554	bne	31b
2555	/* order napping_threads update vs testing entry_exit_map */
2556	isync
2557	li	r0,NAPPING_CEDE
2558	stb	r0,HSTATE_NAPPING(r13)
2559	lwz	r7,VCORE_ENTRY_EXIT(r5)
2560	cmpwi	r7,0x100
2561	bge	33f		/* another thread already exiting */
2562
2563/*
2564 * Although not specifically required by the architecture, POWER7
2565 * preserves the following registers in nap mode, even if an SMT mode
2566 * switch occurs: SLB entries, PURR, SPURR, AMOR, UAMOR, AMR, SPRG0-3,
2567 * DAR, DSISR, DABR, DABRX, DSCR, PMCx, MMCRx, SIAR, SDAR.
2568 */
2569	/* Save non-volatile GPRs */
2570	std	r14, VCPU_GPR(R14)(r3)
2571	std	r15, VCPU_GPR(R15)(r3)
2572	std	r16, VCPU_GPR(R16)(r3)
2573	std	r17, VCPU_GPR(R17)(r3)
2574	std	r18, VCPU_GPR(R18)(r3)
2575	std	r19, VCPU_GPR(R19)(r3)
2576	std	r20, VCPU_GPR(R20)(r3)
2577	std	r21, VCPU_GPR(R21)(r3)
2578	std	r22, VCPU_GPR(R22)(r3)
2579	std	r23, VCPU_GPR(R23)(r3)
2580	std	r24, VCPU_GPR(R24)(r3)
2581	std	r25, VCPU_GPR(R25)(r3)
2582	std	r26, VCPU_GPR(R26)(r3)
2583	std	r27, VCPU_GPR(R27)(r3)
2584	std	r28, VCPU_GPR(R28)(r3)
2585	std	r29, VCPU_GPR(R29)(r3)
2586	std	r30, VCPU_GPR(R30)(r3)
2587	std	r31, VCPU_GPR(R31)(r3)
2588
2589	/* save FP state */
2590	bl	kvmppc_save_fp
2591
2592#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2593/*
2594 * Branch around the call if both CPU_FTR_TM and
2595 * CPU_FTR_P9_TM_HV_ASSIST are off.
2596 */
2597BEGIN_FTR_SECTION
2598	b	91f
2599END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
2600	/*
2601	 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
2602	 */
2603	ld	r3, HSTATE_KVM_VCPU(r13)
2604	ld      r4, VCPU_MSR(r3)
2605	li	r5, 0			/* don't preserve non-vol regs */
2606	bl	kvmppc_save_tm_hv
2607	nop
260891:
2609#endif
2610
2611	/*
2612	 * Set DEC to the smaller of DEC and HDEC, so that we wake
2613	 * no later than the end of our timeslice (HDEC interrupts
2614	 * don't wake us from nap).
2615	 */
2616	mfspr	r3, SPRN_DEC
2617	mfspr	r4, SPRN_HDEC
2618	mftb	r5
2619BEGIN_FTR_SECTION
2620	/* On P9 check whether the guest has large decrementer mode enabled */
2621	ld	r6, HSTATE_KVM_VCORE(r13)
2622	ld	r6, VCORE_LPCR(r6)
2623	andis.	r6, r6, LPCR_LD@h
2624	bne	68f
2625END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2626	extsw	r3, r3
262768:	EXTEND_HDEC(r4)
2628	cmpd	r3, r4
2629	ble	67f
2630	mtspr	SPRN_DEC, r4
263167:
2632	/* save expiry time of guest decrementer */
2633	add	r3, r3, r5
2634	ld	r4, HSTATE_KVM_VCPU(r13)
2635	ld	r5, HSTATE_KVM_VCORE(r13)
2636	ld	r6, VCORE_TB_OFFSET_APPL(r5)
2637	subf	r3, r6, r3	/* convert to host TB value */
2638	std	r3, VCPU_DEC_EXPIRES(r4)
2639
2640#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2641	ld	r4, HSTATE_KVM_VCPU(r13)
2642	addi	r3, r4, VCPU_TB_CEDE
2643	bl	kvmhv_accumulate_time
2644#endif
2645
2646	lis	r3, LPCR_PECEDP@h	/* Do wake on privileged doorbell */
2647
2648	/* Go back to host stack */
2649	ld	r1, HSTATE_HOST_R1(r13)
2650
2651	/*
2652	 * Take a nap until a decrementer or external or doobell interrupt
2653	 * occurs, with PECE1 and PECE0 set in LPCR.
2654	 * On POWER8, set PECEDH, and if we are ceding, also set PECEDP.
2655	 * Also clear the runlatch bit before napping.
2656	 */
2657kvm_do_nap:
2658	mfspr	r0, SPRN_CTRLF
2659	clrrdi	r0, r0, 1
2660	mtspr	SPRN_CTRLT, r0
2661
2662	li	r0,1
2663	stb	r0,HSTATE_HWTHREAD_REQ(r13)
2664	mfspr	r5,SPRN_LPCR
2665	ori	r5,r5,LPCR_PECE0 | LPCR_PECE1
2666BEGIN_FTR_SECTION
2667	ori	r5, r5, LPCR_PECEDH
2668	rlwimi	r5, r3, 0, LPCR_PECEDP
2669END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2670
2671kvm_nap_sequence:		/* desired LPCR value in r5 */
2672BEGIN_FTR_SECTION
2673	/*
2674	 * PSSCR bits:	exit criterion = 1 (wakeup based on LPCR at sreset)
2675	 *		enable state loss = 1 (allow SMT mode switch)
2676	 *		requested level = 0 (just stop dispatching)
2677	 */
2678	lis	r3, (PSSCR_EC | PSSCR_ESL)@h
2679	/* Set LPCR_PECE_HVEE bit to enable wakeup by HV interrupts */
2680	li	r4, LPCR_PECE_HVEE@higher
2681	sldi	r4, r4, 32
2682	or	r5, r5, r4
2683FTR_SECTION_ELSE
2684	li	r3, PNV_THREAD_NAP
2685ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_300)
2686	mtspr	SPRN_LPCR,r5
2687	isync
2688
2689BEGIN_FTR_SECTION
2690	bl	isa300_idle_stop_mayloss
2691FTR_SECTION_ELSE
2692	bl	isa206_idle_insn_mayloss
2693ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_300)
2694
2695	mfspr	r0, SPRN_CTRLF
2696	ori	r0, r0, 1
2697	mtspr	SPRN_CTRLT, r0
2698
2699	mtspr	SPRN_SRR1, r3
2700
2701	li	r0, 0
2702	stb	r0, PACA_FTRACE_ENABLED(r13)
2703
2704	li	r0, KVM_HWTHREAD_IN_KVM
2705	stb	r0, HSTATE_HWTHREAD_STATE(r13)
2706
2707	lbz	r0, HSTATE_NAPPING(r13)
2708	cmpwi	r0, NAPPING_CEDE
2709	beq	kvm_end_cede
2710	cmpwi	r0, NAPPING_NOVCPU
2711	beq	kvm_novcpu_wakeup
2712	cmpwi	r0, NAPPING_UNSPLIT
2713	beq	kvm_unsplit_wakeup
2714	twi	31,0,0 /* Nap state must not be zero */
2715
271633:	mr	r4, r3
2717	li	r3, 0
2718	li	r12, 0
2719	b	34f
2720
2721kvm_end_cede:
2722	/* Woken by external or decrementer interrupt */
2723
2724	/* get vcpu pointer */
2725	ld	r4, HSTATE_KVM_VCPU(r13)
2726
2727#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2728	addi	r3, r4, VCPU_TB_RMINTR
2729	bl	kvmhv_accumulate_time
2730#endif
2731
2732#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2733/*
2734 * Branch around the call if both CPU_FTR_TM and
2735 * CPU_FTR_P9_TM_HV_ASSIST are off.
2736 */
2737BEGIN_FTR_SECTION
2738	b	91f
2739END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
2740	/*
2741	 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
2742	 */
2743	mr      r3, r4
2744	ld      r4, VCPU_MSR(r3)
2745	li	r5, 0			/* don't preserve non-vol regs */
2746	bl	kvmppc_restore_tm_hv
2747	nop
2748	ld	r4, HSTATE_KVM_VCPU(r13)
274991:
2750#endif
2751
2752	/* load up FP state */
2753	bl	kvmppc_load_fp
2754
2755	/* Restore guest decrementer */
2756	ld	r3, VCPU_DEC_EXPIRES(r4)
2757	ld	r5, HSTATE_KVM_VCORE(r13)
2758	ld	r6, VCORE_TB_OFFSET_APPL(r5)
2759	add	r3, r3, r6	/* convert host TB to guest TB value */
2760	mftb	r7
2761	subf	r3, r7, r3
2762	mtspr	SPRN_DEC, r3
2763
2764	/* Load NV GPRS */
2765	ld	r14, VCPU_GPR(R14)(r4)
2766	ld	r15, VCPU_GPR(R15)(r4)
2767	ld	r16, VCPU_GPR(R16)(r4)
2768	ld	r17, VCPU_GPR(R17)(r4)
2769	ld	r18, VCPU_GPR(R18)(r4)
2770	ld	r19, VCPU_GPR(R19)(r4)
2771	ld	r20, VCPU_GPR(R20)(r4)
2772	ld	r21, VCPU_GPR(R21)(r4)
2773	ld	r22, VCPU_GPR(R22)(r4)
2774	ld	r23, VCPU_GPR(R23)(r4)
2775	ld	r24, VCPU_GPR(R24)(r4)
2776	ld	r25, VCPU_GPR(R25)(r4)
2777	ld	r26, VCPU_GPR(R26)(r4)
2778	ld	r27, VCPU_GPR(R27)(r4)
2779	ld	r28, VCPU_GPR(R28)(r4)
2780	ld	r29, VCPU_GPR(R29)(r4)
2781	ld	r30, VCPU_GPR(R30)(r4)
2782	ld	r31, VCPU_GPR(R31)(r4)
2783
2784	/* Check the wake reason in SRR1 to see why we got here */
2785	bl	kvmppc_check_wake_reason
2786
2787	/*
2788	 * Restore volatile registers since we could have called a
2789	 * C routine in kvmppc_check_wake_reason
2790	 *	r4 = VCPU
2791	 * r3 tells us whether we need to return to host or not
2792	 * WARNING: it gets checked further down:
2793	 * should not modify r3 until this check is done.
2794	 */
2795	ld	r4, HSTATE_KVM_VCPU(r13)
2796
2797	/* clear our bit in vcore->napping_threads */
279834:	ld	r5,HSTATE_KVM_VCORE(r13)
2799	lbz	r7,HSTATE_PTID(r13)
2800	li	r0,1
2801	sld	r0,r0,r7
2802	addi	r6,r5,VCORE_NAPPING_THREADS
280332:	lwarx	r7,0,r6
2804	andc	r7,r7,r0
2805	stwcx.	r7,0,r6
2806	bne	32b
2807	li	r0,0
2808	stb	r0,HSTATE_NAPPING(r13)
2809
2810	/* See if the wake reason saved in r3 means we need to exit */
2811	stw	r12, VCPU_TRAP(r4)
2812	mr	r9, r4
2813	cmpdi	r3, 0
2814	bgt	guest_exit_cont
2815	b	maybe_reenter_guest
2816
2817	/* cede when already previously prodded case */
2818kvm_cede_prodded:
2819	li	r0,0
2820	stb	r0,VCPU_PRODDED(r3)
2821	sync			/* order testing prodded vs. clearing ceded */
2822	stb	r0,VCPU_CEDED(r3)
2823	li	r3,H_SUCCESS
2824	blr
2825
2826	/* we've ceded but we want to give control to the host */
2827kvm_cede_exit:
2828	ld	r9, HSTATE_KVM_VCPU(r13)
2829#ifdef CONFIG_KVM_XICS
2830	/* Abort if we still have a pending escalation */
2831	lbz	r5, VCPU_XIVE_ESC_ON(r9)
2832	cmpwi	r5, 0
2833	beq	1f
2834	li	r0, 0
2835	stb	r0, VCPU_CEDED(r9)
28361:	/* Enable XIVE escalation */
2837	li	r5, XIVE_ESB_SET_PQ_00
2838	mfmsr	r0
2839	andi.	r0, r0, MSR_DR		/* in real mode? */
2840	beq	1f
2841	ld	r10, VCPU_XIVE_ESC_VADDR(r9)
2842	cmpdi	r10, 0
2843	beq	3f
2844	ldx	r0, r10, r5
2845	b	2f
28461:	ld	r10, VCPU_XIVE_ESC_RADDR(r9)
2847	cmpdi	r10, 0
2848	beq	3f
2849	ldcix	r0, r10, r5
28502:	sync
2851	li	r0, 1
2852	stb	r0, VCPU_XIVE_ESC_ON(r9)
2853#endif /* CONFIG_KVM_XICS */
28543:	b	guest_exit_cont
2855
2856	/* Try to do machine check recovery in real mode */
2857machine_check_realmode:
2858	mr	r3, r9		/* get vcpu pointer */
2859	bl	kvmppc_realmode_machine_check
2860	nop
2861	/* all machine checks go to virtual mode for further handling */
2862	ld	r9, HSTATE_KVM_VCPU(r13)
2863	li	r12, BOOK3S_INTERRUPT_MACHINE_CHECK
2864	b	guest_exit_cont
2865
2866/*
2867 * Call C code to handle a HMI in real mode.
2868 * Only the primary thread does the call, secondary threads are handled
2869 * by calling hmi_exception_realmode() after kvmppc_hv_entry returns.
2870 * r9 points to the vcpu on entry
2871 */
2872hmi_realmode:
2873	lbz	r0, HSTATE_PTID(r13)
2874	cmpwi	r0, 0
2875	bne	guest_exit_cont
2876	bl	kvmppc_realmode_hmi_handler
2877	ld	r9, HSTATE_KVM_VCPU(r13)
2878	li	r12, BOOK3S_INTERRUPT_HMI
2879	b	guest_exit_cont
2880
2881/*
2882 * Check the reason we woke from nap, and take appropriate action.
2883 * Returns (in r3):
2884 *	0 if nothing needs to be done
2885 *	1 if something happened that needs to be handled by the host
2886 *	-1 if there was a guest wakeup (IPI or msgsnd)
2887 *	-2 if we handled a PCI passthrough interrupt (returned by
2888 *		kvmppc_read_intr only)
2889 *
2890 * Also sets r12 to the interrupt vector for any interrupt that needs
2891 * to be handled now by the host (0x500 for external interrupt), or zero.
2892 * Modifies all volatile registers (since it may call a C function).
2893 * This routine calls kvmppc_read_intr, a C function, if an external
2894 * interrupt is pending.
2895 */
2896kvmppc_check_wake_reason:
2897	mfspr	r6, SPRN_SRR1
2898BEGIN_FTR_SECTION
2899	rlwinm	r6, r6, 45-31, 0xf	/* extract wake reason field (P8) */
2900FTR_SECTION_ELSE
2901	rlwinm	r6, r6, 45-31, 0xe	/* P7 wake reason field is 3 bits */
2902ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_207S)
2903	cmpwi	r6, 8			/* was it an external interrupt? */
2904	beq	7f			/* if so, see what it was */
2905	li	r3, 0
2906	li	r12, 0
2907	cmpwi	r6, 6			/* was it the decrementer? */
2908	beq	0f
2909BEGIN_FTR_SECTION
2910	cmpwi	r6, 5			/* privileged doorbell? */
2911	beq	0f
2912	cmpwi	r6, 3			/* hypervisor doorbell? */
2913	beq	3f
2914END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2915	cmpwi	r6, 0xa			/* Hypervisor maintenance ? */
2916	beq	4f
2917	li	r3, 1			/* anything else, return 1 */
29180:	blr
2919
2920	/* hypervisor doorbell */
29213:	li	r12, BOOK3S_INTERRUPT_H_DOORBELL
2922
2923	/*
2924	 * Clear the doorbell as we will invoke the handler
2925	 * explicitly in the guest exit path.
2926	 */
2927	lis	r6, (PPC_DBELL_SERVER << (63-36))@h
2928	PPC_MSGCLR(6)
2929	/* see if it's a host IPI */
2930	li	r3, 1
2931BEGIN_FTR_SECTION
2932	PPC_MSGSYNC
2933	lwsync
2934END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2935	lbz	r0, HSTATE_HOST_IPI(r13)
2936	cmpwi	r0, 0
2937	bnelr
2938	/* if not, return -1 */
2939	li	r3, -1
2940	blr
2941
2942	/* Woken up due to Hypervisor maintenance interrupt */
29434:	li	r12, BOOK3S_INTERRUPT_HMI
2944	li	r3, 1
2945	blr
2946
2947	/* external interrupt - create a stack frame so we can call C */
29487:	mflr	r0
2949	std	r0, PPC_LR_STKOFF(r1)
2950	stdu	r1, -PPC_MIN_STKFRM(r1)
2951	bl	kvmppc_read_intr
2952	nop
2953	li	r12, BOOK3S_INTERRUPT_EXTERNAL
2954	cmpdi	r3, 1
2955	ble	1f
2956
2957	/*
2958	 * Return code of 2 means PCI passthrough interrupt, but
2959	 * we need to return back to host to complete handling the
2960	 * interrupt. Trap reason is expected in r12 by guest
2961	 * exit code.
2962	 */
2963	li	r12, BOOK3S_INTERRUPT_HV_RM_HARD
29641:
2965	ld	r0, PPC_MIN_STKFRM+PPC_LR_STKOFF(r1)
2966	addi	r1, r1, PPC_MIN_STKFRM
2967	mtlr	r0
2968	blr
2969
2970/*
2971 * Save away FP, VMX and VSX registers.
2972 * r3 = vcpu pointer
2973 * N.B. r30 and r31 are volatile across this function,
2974 * thus it is not callable from C.
2975 */
2976kvmppc_save_fp:
2977	mflr	r30
2978	mr	r31,r3
2979	mfmsr	r5
2980	ori	r8,r5,MSR_FP
2981#ifdef CONFIG_ALTIVEC
2982BEGIN_FTR_SECTION
2983	oris	r8,r8,MSR_VEC@h
2984END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2985#endif
2986#ifdef CONFIG_VSX
2987BEGIN_FTR_SECTION
2988	oris	r8,r8,MSR_VSX@h
2989END_FTR_SECTION_IFSET(CPU_FTR_VSX)
2990#endif
2991	mtmsrd	r8
2992	addi	r3,r3,VCPU_FPRS
2993	bl	store_fp_state
2994#ifdef CONFIG_ALTIVEC
2995BEGIN_FTR_SECTION
2996	addi	r3,r31,VCPU_VRS
2997	bl	store_vr_state
2998END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2999#endif
3000	mfspr	r6,SPRN_VRSAVE
3001	stw	r6,VCPU_VRSAVE(r31)
3002	mtlr	r30
3003	blr
3004
3005/*
3006 * Load up FP, VMX and VSX registers
3007 * r4 = vcpu pointer
3008 * N.B. r30 and r31 are volatile across this function,
3009 * thus it is not callable from C.
3010 */
3011kvmppc_load_fp:
3012	mflr	r30
3013	mr	r31,r4
3014	mfmsr	r9
3015	ori	r8,r9,MSR_FP
3016#ifdef CONFIG_ALTIVEC
3017BEGIN_FTR_SECTION
3018	oris	r8,r8,MSR_VEC@h
3019END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3020#endif
3021#ifdef CONFIG_VSX
3022BEGIN_FTR_SECTION
3023	oris	r8,r8,MSR_VSX@h
3024END_FTR_SECTION_IFSET(CPU_FTR_VSX)
3025#endif
3026	mtmsrd	r8
3027	addi	r3,r4,VCPU_FPRS
3028	bl	load_fp_state
3029#ifdef CONFIG_ALTIVEC
3030BEGIN_FTR_SECTION
3031	addi	r3,r31,VCPU_VRS
3032	bl	load_vr_state
3033END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3034#endif
3035	lwz	r7,VCPU_VRSAVE(r31)
3036	mtspr	SPRN_VRSAVE,r7
3037	mtlr	r30
3038	mr	r4,r31
3039	blr
3040
3041#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3042/*
3043 * Save transactional state and TM-related registers.
3044 * Called with r3 pointing to the vcpu struct and r4 containing
3045 * the guest MSR value.
3046 * r5 is non-zero iff non-volatile register state needs to be maintained.
3047 * If r5 == 0, this can modify all checkpointed registers, but
3048 * restores r1 and r2 before exit.
3049 */
3050_GLOBAL_TOC(kvmppc_save_tm_hv)
3051EXPORT_SYMBOL_GPL(kvmppc_save_tm_hv)
3052	/* See if we need to handle fake suspend mode */
3053BEGIN_FTR_SECTION
3054	b	__kvmppc_save_tm
3055END_FTR_SECTION_IFCLR(CPU_FTR_P9_TM_HV_ASSIST)
3056
3057	lbz	r0, HSTATE_FAKE_SUSPEND(r13) /* Were we fake suspended? */
3058	cmpwi	r0, 0
3059	beq	__kvmppc_save_tm
3060
3061	/* The following code handles the fake_suspend = 1 case */
3062	mflr	r0
3063	std	r0, PPC_LR_STKOFF(r1)
3064	stdu	r1, -PPC_MIN_STKFRM(r1)
3065
3066	/* Turn on TM. */
3067	mfmsr	r8
3068	li	r0, 1
3069	rldimi	r8, r0, MSR_TM_LG, 63-MSR_TM_LG
3070	mtmsrd	r8
3071
3072	rldicl. r8, r8, 64 - MSR_TS_S_LG, 62 /* Did we actually hrfid? */
3073	beq	4f
3074BEGIN_FTR_SECTION
3075	bl	pnv_power9_force_smt4_catch
3076END_FTR_SECTION_IFSET(CPU_FTR_P9_TM_XER_SO_BUG)
3077	nop
3078
3079	/* We have to treclaim here because that's the only way to do S->N */
3080	li	r3, TM_CAUSE_KVM_RESCHED
3081	TRECLAIM(R3)
3082
3083	/*
3084	 * We were in fake suspend, so we are not going to save the
3085	 * register state as the guest checkpointed state (since
3086	 * we already have it), therefore we can now use any volatile GPR.
3087	 * In fact treclaim in fake suspend state doesn't modify
3088	 * any registers.
3089	 */
3090
3091BEGIN_FTR_SECTION
3092	bl	pnv_power9_force_smt4_release
3093END_FTR_SECTION_IFSET(CPU_FTR_P9_TM_XER_SO_BUG)
3094	nop
3095
30964:
3097	mfspr	r3, SPRN_PSSCR
3098	/* PSSCR_FAKE_SUSPEND is a write-only bit, but clear it anyway */
3099	li	r0, PSSCR_FAKE_SUSPEND
3100	andc	r3, r3, r0
3101	mtspr	SPRN_PSSCR, r3
3102
3103	/* Don't save TEXASR, use value from last exit in real suspend state */
3104	ld	r9, HSTATE_KVM_VCPU(r13)
3105	mfspr	r5, SPRN_TFHAR
3106	mfspr	r6, SPRN_TFIAR
3107	std	r5, VCPU_TFHAR(r9)
3108	std	r6, VCPU_TFIAR(r9)
3109
3110	addi	r1, r1, PPC_MIN_STKFRM
3111	ld	r0, PPC_LR_STKOFF(r1)
3112	mtlr	r0
3113	blr
3114
3115/*
3116 * Restore transactional state and TM-related registers.
3117 * Called with r3 pointing to the vcpu struct
3118 * and r4 containing the guest MSR value.
3119 * r5 is non-zero iff non-volatile register state needs to be maintained.
3120 * This potentially modifies all checkpointed registers.
3121 * It restores r1 and r2 from the PACA.
3122 */
3123_GLOBAL_TOC(kvmppc_restore_tm_hv)
3124EXPORT_SYMBOL_GPL(kvmppc_restore_tm_hv)
3125	/*
3126	 * If we are doing TM emulation for the guest on a POWER9 DD2,
3127	 * then we don't actually do a trechkpt -- we either set up
3128	 * fake-suspend mode, or emulate a TM rollback.
3129	 */
3130BEGIN_FTR_SECTION
3131	b	__kvmppc_restore_tm
3132END_FTR_SECTION_IFCLR(CPU_FTR_P9_TM_HV_ASSIST)
3133	mflr	r0
3134	std	r0, PPC_LR_STKOFF(r1)
3135
3136	li	r0, 0
3137	stb	r0, HSTATE_FAKE_SUSPEND(r13)
3138
3139	/* Turn on TM so we can restore TM SPRs */
3140	mfmsr	r5
3141	li	r0, 1
3142	rldimi	r5, r0, MSR_TM_LG, 63-MSR_TM_LG
3143	mtmsrd	r5
3144
3145	/*
3146	 * The user may change these outside of a transaction, so they must
3147	 * always be context switched.
3148	 */
3149	ld	r5, VCPU_TFHAR(r3)
3150	ld	r6, VCPU_TFIAR(r3)
3151	ld	r7, VCPU_TEXASR(r3)
3152	mtspr	SPRN_TFHAR, r5
3153	mtspr	SPRN_TFIAR, r6
3154	mtspr	SPRN_TEXASR, r7
3155
3156	rldicl. r5, r4, 64 - MSR_TS_S_LG, 62
3157	beqlr		/* TM not active in guest */
3158
3159	/* Make sure the failure summary is set */
3160	oris	r7, r7, (TEXASR_FS)@h
3161	mtspr	SPRN_TEXASR, r7
3162
3163	cmpwi	r5, 1		/* check for suspended state */
3164	bgt	10f
3165	stb	r5, HSTATE_FAKE_SUSPEND(r13)
3166	b	9f		/* and return */
316710:	stdu	r1, -PPC_MIN_STKFRM(r1)
3168	/* guest is in transactional state, so simulate rollback */
3169	bl	kvmhv_emulate_tm_rollback
3170	nop
3171	addi	r1, r1, PPC_MIN_STKFRM
31729:	ld	r0, PPC_LR_STKOFF(r1)
3173	mtlr	r0
3174	blr
3175#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
3176
3177/*
3178 * We come here if we get any exception or interrupt while we are
3179 * executing host real mode code while in guest MMU context.
3180 * r12 is (CR << 32) | vector
3181 * r13 points to our PACA
3182 * r12 is saved in HSTATE_SCRATCH0(r13)
3183 * ctr is saved in HSTATE_SCRATCH1(r13) if RELOCATABLE
3184 * r9 is saved in HSTATE_SCRATCH2(r13)
3185 * r13 is saved in HSPRG1
3186 * cfar is saved in HSTATE_CFAR(r13)
3187 * ppr is saved in HSTATE_PPR(r13)
3188 */
3189kvmppc_bad_host_intr:
3190	/*
3191	 * Switch to the emergency stack, but start half-way down in
3192	 * case we were already on it.
3193	 */
3194	mr	r9, r1
3195	std	r1, PACAR1(r13)
3196	ld	r1, PACAEMERGSP(r13)
3197	subi	r1, r1, THREAD_SIZE/2 + INT_FRAME_SIZE
3198	std	r9, 0(r1)
3199	std	r0, GPR0(r1)
3200	std	r9, GPR1(r1)
3201	std	r2, GPR2(r1)
3202	SAVE_4GPRS(3, r1)
3203	SAVE_2GPRS(7, r1)
3204	srdi	r0, r12, 32
3205	clrldi	r12, r12, 32
3206	std	r0, _CCR(r1)
3207	std	r12, _TRAP(r1)
3208	andi.	r0, r12, 2
3209	beq	1f
3210	mfspr	r3, SPRN_HSRR0
3211	mfspr	r4, SPRN_HSRR1
3212	mfspr	r5, SPRN_HDAR
3213	mfspr	r6, SPRN_HDSISR
3214	b	2f
32151:	mfspr	r3, SPRN_SRR0
3216	mfspr	r4, SPRN_SRR1
3217	mfspr	r5, SPRN_DAR
3218	mfspr	r6, SPRN_DSISR
32192:	std	r3, _NIP(r1)
3220	std	r4, _MSR(r1)
3221	std	r5, _DAR(r1)
3222	std	r6, _DSISR(r1)
3223	ld	r9, HSTATE_SCRATCH2(r13)
3224	ld	r12, HSTATE_SCRATCH0(r13)
3225	GET_SCRATCH0(r0)
3226	SAVE_4GPRS(9, r1)
3227	std	r0, GPR13(r1)
3228	SAVE_NVGPRS(r1)
3229	ld	r5, HSTATE_CFAR(r13)
3230	std	r5, ORIG_GPR3(r1)
3231	mflr	r3
3232#ifdef CONFIG_RELOCATABLE
3233	ld	r4, HSTATE_SCRATCH1(r13)
3234#else
3235	mfctr	r4
3236#endif
3237	mfxer	r5
3238	lbz	r6, PACAIRQSOFTMASK(r13)
3239	std	r3, _LINK(r1)
3240	std	r4, _CTR(r1)
3241	std	r5, _XER(r1)
3242	std	r6, SOFTE(r1)
3243	ld	r2, PACATOC(r13)
3244	LOAD_REG_IMMEDIATE(3, 0x7265677368657265)
3245	std	r3, STACK_FRAME_OVERHEAD-16(r1)
3246
3247	/*
3248	 * On POWER9 do a minimal restore of the MMU and call C code,
3249	 * which will print a message and panic.
3250	 * XXX On POWER7 and POWER8, we just spin here since we don't
3251	 * know what the other threads are doing (and we don't want to
3252	 * coordinate with them) - but at least we now have register state
3253	 * in memory that we might be able to look at from another CPU.
3254	 */
3255BEGIN_FTR_SECTION
3256	b	.
3257END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
3258	ld	r9, HSTATE_KVM_VCPU(r13)
3259	ld	r10, VCPU_KVM(r9)
3260
3261	li	r0, 0
3262	mtspr	SPRN_AMR, r0
3263	mtspr	SPRN_IAMR, r0
3264	mtspr	SPRN_CIABR, r0
3265	mtspr	SPRN_DAWRX, r0
3266
3267BEGIN_MMU_FTR_SECTION
3268	b	4f
3269END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
3270
3271	slbmte	r0, r0
3272	slbia
3273	ptesync
3274	ld	r8, PACA_SLBSHADOWPTR(r13)
3275	.rept	SLB_NUM_BOLTED
3276	li	r3, SLBSHADOW_SAVEAREA
3277	LDX_BE	r5, r8, r3
3278	addi	r3, r3, 8
3279	LDX_BE	r6, r8, r3
3280	andis.	r7, r5, SLB_ESID_V@h
3281	beq	3f
3282	slbmte	r6, r5
32833:	addi	r8, r8, 16
3284	.endr
3285
32864:	lwz	r7, KVM_HOST_LPID(r10)
3287	mtspr	SPRN_LPID, r7
3288	mtspr	SPRN_PID, r0
3289	ld	r8, KVM_HOST_LPCR(r10)
3290	mtspr	SPRN_LPCR, r8
3291	isync
3292	li	r0, KVM_GUEST_MODE_NONE
3293	stb	r0, HSTATE_IN_GUEST(r13)
3294
3295	/*
3296	 * Turn on the MMU and jump to C code
3297	 */
3298	bcl	20, 31, .+4
32995:	mflr	r3
3300	addi	r3, r3, 9f - 5b
3301	li	r4, -1
3302	rldimi	r3, r4, 62, 0	/* ensure 0xc000000000000000 bits are set */
3303	ld	r4, PACAKMSR(r13)
3304	mtspr	SPRN_SRR0, r3
3305	mtspr	SPRN_SRR1, r4
3306	RFI_TO_KERNEL
33079:	addi	r3, r1, STACK_FRAME_OVERHEAD
3308	bl	kvmppc_bad_interrupt
3309	b	9b
3310
3311/*
3312 * This mimics the MSR transition on IRQ delivery.  The new guest MSR is taken
3313 * from VCPU_INTR_MSR and is modified based on the required TM state changes.
3314 *   r11 has the guest MSR value (in/out)
3315 *   r9 has a vcpu pointer (in)
3316 *   r0 is used as a scratch register
3317 */
3318kvmppc_msr_interrupt:
3319	rldicl	r0, r11, 64 - MSR_TS_S_LG, 62
3320	cmpwi	r0, 2 /* Check if we are in transactional state..  */
3321	ld	r11, VCPU_INTR_MSR(r9)
3322	bne	1f
3323	/* ... if transactional, change to suspended */
3324	li	r0, 1
33251:	rldimi	r11, r0, MSR_TS_S_LG, 63 - MSR_TS_T_LG
3326	blr
3327
3328/*
3329 * Load up guest PMU state.  R3 points to the vcpu struct.
3330 */
3331_GLOBAL(kvmhv_load_guest_pmu)
3332EXPORT_SYMBOL_GPL(kvmhv_load_guest_pmu)
3333	mr	r4, r3
3334	mflr	r0
3335	li	r3, 1
3336	sldi	r3, r3, 31		/* MMCR0_FC (freeze counters) bit */
3337	mtspr	SPRN_MMCR0, r3		/* freeze all counters, disable ints */
3338	isync
3339BEGIN_FTR_SECTION
3340	ld	r3, VCPU_MMCR(r4)
3341	andi.	r5, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
3342	cmpwi	r5, MMCR0_PMAO
3343	beql	kvmppc_fix_pmao
3344END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
3345	lwz	r3, VCPU_PMC(r4)	/* always load up guest PMU registers */
3346	lwz	r5, VCPU_PMC + 4(r4)	/* to prevent information leak */
3347	lwz	r6, VCPU_PMC + 8(r4)
3348	lwz	r7, VCPU_PMC + 12(r4)
3349	lwz	r8, VCPU_PMC + 16(r4)
3350	lwz	r9, VCPU_PMC + 20(r4)
3351	mtspr	SPRN_PMC1, r3
3352	mtspr	SPRN_PMC2, r5
3353	mtspr	SPRN_PMC3, r6
3354	mtspr	SPRN_PMC4, r7
3355	mtspr	SPRN_PMC5, r8
3356	mtspr	SPRN_PMC6, r9
3357	ld	r3, VCPU_MMCR(r4)
3358	ld	r5, VCPU_MMCR + 8(r4)
3359	ld	r6, VCPU_MMCR + 16(r4)
3360	ld	r7, VCPU_SIAR(r4)
3361	ld	r8, VCPU_SDAR(r4)
3362	mtspr	SPRN_MMCR1, r5
3363	mtspr	SPRN_MMCRA, r6
3364	mtspr	SPRN_SIAR, r7
3365	mtspr	SPRN_SDAR, r8
3366BEGIN_FTR_SECTION
3367	ld	r5, VCPU_MMCR + 24(r4)
3368	ld	r6, VCPU_SIER(r4)
3369	mtspr	SPRN_MMCR2, r5
3370	mtspr	SPRN_SIER, r6
3371BEGIN_FTR_SECTION_NESTED(96)
3372	lwz	r7, VCPU_PMC + 24(r4)
3373	lwz	r8, VCPU_PMC + 28(r4)
3374	ld	r9, VCPU_MMCR + 32(r4)
3375	mtspr	SPRN_SPMC1, r7
3376	mtspr	SPRN_SPMC2, r8
3377	mtspr	SPRN_MMCRS, r9
3378END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
3379END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3380	mtspr	SPRN_MMCR0, r3
3381	isync
3382	mtlr	r0
3383	blr
3384
3385/*
3386 * Reload host PMU state saved in the PACA by kvmhv_save_host_pmu.
3387 */
3388_GLOBAL(kvmhv_load_host_pmu)
3389EXPORT_SYMBOL_GPL(kvmhv_load_host_pmu)
3390	mflr	r0
3391	lbz	r4, PACA_PMCINUSE(r13) /* is the host using the PMU? */
3392	cmpwi	r4, 0
3393	beq	23f			/* skip if not */
3394BEGIN_FTR_SECTION
3395	ld	r3, HSTATE_MMCR0(r13)
3396	andi.	r4, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
3397	cmpwi	r4, MMCR0_PMAO
3398	beql	kvmppc_fix_pmao
3399END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
3400	lwz	r3, HSTATE_PMC1(r13)
3401	lwz	r4, HSTATE_PMC2(r13)
3402	lwz	r5, HSTATE_PMC3(r13)
3403	lwz	r6, HSTATE_PMC4(r13)
3404	lwz	r8, HSTATE_PMC5(r13)
3405	lwz	r9, HSTATE_PMC6(r13)
3406	mtspr	SPRN_PMC1, r3
3407	mtspr	SPRN_PMC2, r4
3408	mtspr	SPRN_PMC3, r5
3409	mtspr	SPRN_PMC4, r6
3410	mtspr	SPRN_PMC5, r8
3411	mtspr	SPRN_PMC6, r9
3412	ld	r3, HSTATE_MMCR0(r13)
3413	ld	r4, HSTATE_MMCR1(r13)
3414	ld	r5, HSTATE_MMCRA(r13)
3415	ld	r6, HSTATE_SIAR(r13)
3416	ld	r7, HSTATE_SDAR(r13)
3417	mtspr	SPRN_MMCR1, r4
3418	mtspr	SPRN_MMCRA, r5
3419	mtspr	SPRN_SIAR, r6
3420	mtspr	SPRN_SDAR, r7
3421BEGIN_FTR_SECTION
3422	ld	r8, HSTATE_MMCR2(r13)
3423	ld	r9, HSTATE_SIER(r13)
3424	mtspr	SPRN_MMCR2, r8
3425	mtspr	SPRN_SIER, r9
3426END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3427	mtspr	SPRN_MMCR0, r3
3428	isync
3429	mtlr	r0
343023:	blr
3431
3432/*
3433 * Save guest PMU state into the vcpu struct.
3434 * r3 = vcpu, r4 = full save flag (PMU in use flag set in VPA)
3435 */
3436_GLOBAL(kvmhv_save_guest_pmu)
3437EXPORT_SYMBOL_GPL(kvmhv_save_guest_pmu)
3438	mr	r9, r3
3439	mr	r8, r4
3440BEGIN_FTR_SECTION
3441	/*
3442	 * POWER8 seems to have a hardware bug where setting
3443	 * MMCR0[PMAE] along with MMCR0[PMC1CE] and/or MMCR0[PMCjCE]
3444	 * when some counters are already negative doesn't seem
3445	 * to cause a performance monitor alert (and hence interrupt).
3446	 * The effect of this is that when saving the PMU state,
3447	 * if there is no PMU alert pending when we read MMCR0
3448	 * before freezing the counters, but one becomes pending
3449	 * before we read the counters, we lose it.
3450	 * To work around this, we need a way to freeze the counters
3451	 * before reading MMCR0.  Normally, freezing the counters
3452	 * is done by writing MMCR0 (to set MMCR0[FC]) which
3453	 * unavoidably writes MMCR0[PMA0] as well.  On POWER8,
3454	 * we can also freeze the counters using MMCR2, by writing
3455	 * 1s to all the counter freeze condition bits (there are
3456	 * 9 bits each for 6 counters).
3457	 */
3458	li	r3, -1			/* set all freeze bits */
3459	clrrdi	r3, r3, 10
3460	mfspr	r10, SPRN_MMCR2
3461	mtspr	SPRN_MMCR2, r3
3462	isync
3463END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3464	li	r3, 1
3465	sldi	r3, r3, 31		/* MMCR0_FC (freeze counters) bit */
3466	mfspr	r4, SPRN_MMCR0		/* save MMCR0 */
3467	mtspr	SPRN_MMCR0, r3		/* freeze all counters, disable ints */
3468	mfspr	r6, SPRN_MMCRA
3469	/* Clear MMCRA in order to disable SDAR updates */
3470	li	r7, 0
3471	mtspr	SPRN_MMCRA, r7
3472	isync
3473	cmpwi	r8, 0			/* did they ask for PMU stuff to be saved? */
3474	bne	21f
3475	std	r3, VCPU_MMCR(r9)	/* if not, set saved MMCR0 to FC */
3476	b	22f
347721:	mfspr	r5, SPRN_MMCR1
3478	mfspr	r7, SPRN_SIAR
3479	mfspr	r8, SPRN_SDAR
3480	std	r4, VCPU_MMCR(r9)
3481	std	r5, VCPU_MMCR + 8(r9)
3482	std	r6, VCPU_MMCR + 16(r9)
3483BEGIN_FTR_SECTION
3484	std	r10, VCPU_MMCR + 24(r9)
3485END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3486	std	r7, VCPU_SIAR(r9)
3487	std	r8, VCPU_SDAR(r9)
3488	mfspr	r3, SPRN_PMC1
3489	mfspr	r4, SPRN_PMC2
3490	mfspr	r5, SPRN_PMC3
3491	mfspr	r6, SPRN_PMC4
3492	mfspr	r7, SPRN_PMC5
3493	mfspr	r8, SPRN_PMC6
3494	stw	r3, VCPU_PMC(r9)
3495	stw	r4, VCPU_PMC + 4(r9)
3496	stw	r5, VCPU_PMC + 8(r9)
3497	stw	r6, VCPU_PMC + 12(r9)
3498	stw	r7, VCPU_PMC + 16(r9)
3499	stw	r8, VCPU_PMC + 20(r9)
3500BEGIN_FTR_SECTION
3501	mfspr	r5, SPRN_SIER
3502	std	r5, VCPU_SIER(r9)
3503BEGIN_FTR_SECTION_NESTED(96)
3504	mfspr	r6, SPRN_SPMC1
3505	mfspr	r7, SPRN_SPMC2
3506	mfspr	r8, SPRN_MMCRS
3507	stw	r6, VCPU_PMC + 24(r9)
3508	stw	r7, VCPU_PMC + 28(r9)
3509	std	r8, VCPU_MMCR + 32(r9)
3510	lis	r4, 0x8000
3511	mtspr	SPRN_MMCRS, r4
3512END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
3513END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
351422:	blr
3515
3516/*
3517 * This works around a hardware bug on POWER8E processors, where
3518 * writing a 1 to the MMCR0[PMAO] bit doesn't generate a
3519 * performance monitor interrupt.  Instead, when we need to have
3520 * an interrupt pending, we have to arrange for a counter to overflow.
3521 */
3522kvmppc_fix_pmao:
3523	li	r3, 0
3524	mtspr	SPRN_MMCR2, r3
3525	lis	r3, (MMCR0_PMXE | MMCR0_FCECE)@h
3526	ori	r3, r3, MMCR0_PMCjCE | MMCR0_C56RUN
3527	mtspr	SPRN_MMCR0, r3
3528	lis	r3, 0x7fff
3529	ori	r3, r3, 0xffff
3530	mtspr	SPRN_PMC6, r3
3531	isync
3532	blr
3533
3534#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
3535/*
3536 * Start timing an activity
3537 * r3 = pointer to time accumulation struct, r4 = vcpu
3538 */
3539kvmhv_start_timing:
3540	ld	r5, HSTATE_KVM_VCORE(r13)
3541	ld	r6, VCORE_TB_OFFSET_APPL(r5)
3542	mftb	r5
3543	subf	r5, r6, r5	/* subtract current timebase offset */
3544	std	r3, VCPU_CUR_ACTIVITY(r4)
3545	std	r5, VCPU_ACTIVITY_START(r4)
3546	blr
3547
3548/*
3549 * Accumulate time to one activity and start another.
3550 * r3 = pointer to new time accumulation struct, r4 = vcpu
3551 */
3552kvmhv_accumulate_time:
3553	ld	r5, HSTATE_KVM_VCORE(r13)
3554	ld	r8, VCORE_TB_OFFSET_APPL(r5)
3555	ld	r5, VCPU_CUR_ACTIVITY(r4)
3556	ld	r6, VCPU_ACTIVITY_START(r4)
3557	std	r3, VCPU_CUR_ACTIVITY(r4)
3558	mftb	r7
3559	subf	r7, r8, r7	/* subtract current timebase offset */
3560	std	r7, VCPU_ACTIVITY_START(r4)
3561	cmpdi	r5, 0
3562	beqlr
3563	subf	r3, r6, r7
3564	ld	r8, TAS_SEQCOUNT(r5)
3565	cmpdi	r8, 0
3566	addi	r8, r8, 1
3567	std	r8, TAS_SEQCOUNT(r5)
3568	lwsync
3569	ld	r7, TAS_TOTAL(r5)
3570	add	r7, r7, r3
3571	std	r7, TAS_TOTAL(r5)
3572	ld	r6, TAS_MIN(r5)
3573	ld	r7, TAS_MAX(r5)
3574	beq	3f
3575	cmpd	r3, r6
3576	bge	1f
35773:	std	r3, TAS_MIN(r5)
35781:	cmpd	r3, r7
3579	ble	2f
3580	std	r3, TAS_MAX(r5)
35812:	lwsync
3582	addi	r8, r8, 1
3583	std	r8, TAS_SEQCOUNT(r5)
3584	blr
3585#endif
3586