xref: /linux/arch/powerpc/kernel/idle_book3s.S (revision 2c86cd188f8a5631f3d75a1dea14d22df85189b4)
1/*
2 *  This file contains idle entry/exit functions for POWER7,
3 *  POWER8 and POWER9 CPUs.
4 *
5 *  This program is free software; you can redistribute it and/or
6 *  modify it under the terms of the GNU General Public License
7 *  as published by the Free Software Foundation; either version
8 *  2 of the License, or (at your option) any later version.
9 */
10
11#include <linux/threads.h>
12#include <asm/processor.h>
13#include <asm/page.h>
14#include <asm/cputable.h>
15#include <asm/thread_info.h>
16#include <asm/ppc_asm.h>
17#include <asm/asm-offsets.h>
18#include <asm/ppc-opcode.h>
19#include <asm/hw_irq.h>
20#include <asm/kvm_book3s_asm.h>
21#include <asm/opal.h>
22#include <asm/cpuidle.h>
23#include <asm/exception-64s.h>
24#include <asm/book3s/64/mmu-hash.h>
25#include <asm/mmu.h>
26#include <asm/asm-compat.h>
27#include <asm/feature-fixups.h>
28
29#undef DEBUG
30
31/*
32 * Use unused space in the interrupt stack to save and restore
33 * registers for winkle support.
34 */
35#define _MMCR0	GPR0
36#define _SDR1	GPR3
37#define _PTCR	GPR3
38#define _RPR	GPR4
39#define _SPURR	GPR5
40#define _PURR	GPR6
41#define _TSCR	GPR7
42#define _DSCR	GPR8
43#define _AMOR	GPR9
44#define _WORT	GPR10
45#define _WORC	GPR11
46#define _LPCR	GPR12
47
48#define PSSCR_EC_ESL_MASK_SHIFTED          (PSSCR_EC | PSSCR_ESL) >> 16
49
50	.text
51
52/*
53 * Used by threads before entering deep idle states. Saves SPRs
54 * in interrupt stack frame
55 */
56save_sprs_to_stack:
57	/*
58	 * Note all register i.e per-core, per-subcore or per-thread is saved
59	 * here since any thread in the core might wake up first
60	 */
61BEGIN_FTR_SECTION
62	/*
63	 * Note - SDR1 is dropped in Power ISA v3. Hence not restoring
64	 * SDR1 here
65	 */
66	mfspr	r3,SPRN_PTCR
67	std	r3,_PTCR(r1)
68	mfspr	r3,SPRN_LPCR
69	std	r3,_LPCR(r1)
70FTR_SECTION_ELSE
71	mfspr	r3,SPRN_SDR1
72	std	r3,_SDR1(r1)
73ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_300)
74	mfspr	r3,SPRN_RPR
75	std	r3,_RPR(r1)
76	mfspr	r3,SPRN_SPURR
77	std	r3,_SPURR(r1)
78	mfspr	r3,SPRN_PURR
79	std	r3,_PURR(r1)
80	mfspr	r3,SPRN_TSCR
81	std	r3,_TSCR(r1)
82	mfspr	r3,SPRN_DSCR
83	std	r3,_DSCR(r1)
84	mfspr	r3,SPRN_AMOR
85	std	r3,_AMOR(r1)
86	mfspr	r3,SPRN_WORT
87	std	r3,_WORT(r1)
88	mfspr	r3,SPRN_WORC
89	std	r3,_WORC(r1)
90/*
91 * On POWER9, there are idle states such as stop4, invoked via cpuidle,
92 * that lose hypervisor resources. In such cases, we need to save
93 * additional SPRs before entering those idle states so that they can
94 * be restored to their older values on wakeup from the idle state.
95 *
96 * On POWER8, the only such deep idle state is winkle which is used
97 * only in the context of CPU-Hotplug, where these additional SPRs are
98 * reinitiazed to a sane value. Hence there is no need to save/restore
99 * these SPRs.
100 */
101BEGIN_FTR_SECTION
102	blr
103END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
104
105power9_save_additional_sprs:
106	mfspr	r3, SPRN_PID
107	mfspr	r4, SPRN_LDBAR
108	std	r3, STOP_PID(r13)
109	std	r4, STOP_LDBAR(r13)
110
111	mfspr	r3, SPRN_FSCR
112	mfspr	r4, SPRN_HFSCR
113	std	r3, STOP_FSCR(r13)
114	std	r4, STOP_HFSCR(r13)
115
116	mfspr	r3, SPRN_MMCRA
117	mfspr	r4, SPRN_MMCR0
118	std	r3, STOP_MMCRA(r13)
119	std	r4, _MMCR0(r1)
120
121	mfspr	r3, SPRN_MMCR1
122	mfspr	r4, SPRN_MMCR2
123	std	r3, STOP_MMCR1(r13)
124	std	r4, STOP_MMCR2(r13)
125	blr
126
127power9_restore_additional_sprs:
128	ld	r3,_LPCR(r1)
129	ld	r4, STOP_PID(r13)
130	mtspr	SPRN_LPCR,r3
131	mtspr	SPRN_PID, r4
132
133	ld	r3, STOP_LDBAR(r13)
134	ld	r4, STOP_FSCR(r13)
135	mtspr	SPRN_LDBAR, r3
136	mtspr	SPRN_FSCR, r4
137
138	ld	r3, STOP_HFSCR(r13)
139	ld	r4, STOP_MMCRA(r13)
140	mtspr	SPRN_HFSCR, r3
141	mtspr	SPRN_MMCRA, r4
142
143	ld	r3, _MMCR0(r1)
144	ld	r4, STOP_MMCR1(r13)
145	mtspr	SPRN_MMCR0, r3
146	mtspr	SPRN_MMCR1, r4
147
148	ld	r3, STOP_MMCR2(r13)
149	mtspr	SPRN_MMCR2, r3
150	blr
151
152/*
153 * Used by threads when the lock bit of core_idle_state is set.
154 * Threads will spin in HMT_LOW until the lock bit is cleared.
155 * r14 - pointer to core_idle_state
156 * r15 - used to load contents of core_idle_state
157 * r9  - used as a temporary variable
158 */
159
160core_idle_lock_held:
161	HMT_LOW
1623:	lwz	r15,0(r14)
163	andis.	r15,r15,PNV_CORE_IDLE_LOCK_BIT@h
164	bne	3b
165	HMT_MEDIUM
166	lwarx	r15,0,r14
167	andis.	r9,r15,PNV_CORE_IDLE_LOCK_BIT@h
168	bne-	core_idle_lock_held
169	blr
170
171/*
172 * Pass requested state in r3:
173 *	r3 - PNV_THREAD_NAP/SLEEP/WINKLE in POWER8
174 *	   - Requested PSSCR value in POWER9
175 *
176 * Address of idle handler to branch to in realmode in r4
177 */
178pnv_powersave_common:
179	/* Use r3 to pass state nap/sleep/winkle */
180	/* NAP is a state loss, we create a regs frame on the
181	 * stack, fill it up with the state we care about and
182	 * stick a pointer to it in PACAR1. We really only
183	 * need to save PC, some CR bits and the NV GPRs,
184	 * but for now an interrupt frame will do.
185	 */
186	mtctr	r4
187
188	mflr	r0
189	std	r0,16(r1)
190	stdu	r1,-INT_FRAME_SIZE(r1)
191	std	r0,_LINK(r1)
192	std	r0,_NIP(r1)
193
194	/* We haven't lost state ... yet */
195	li	r0,0
196	stb	r0,PACA_NAPSTATELOST(r13)
197
198	/* Continue saving state */
199	SAVE_GPR(2, r1)
200	SAVE_NVGPRS(r1)
201	mfcr	r5
202	std	r5,_CCR(r1)
203	std	r1,PACAR1(r13)
204
205BEGIN_FTR_SECTION
206	/*
207	 * POWER9 does not require real mode to stop, and presently does not
208	 * set hwthread_state for KVM (threads don't share MMU context), so
209	 * we can remain in virtual mode for this.
210	 */
211	bctr
212END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
213	/*
214	 * POWER8
215	 * Go to real mode to do the nap, as required by the architecture.
216	 * Also, we need to be in real mode before setting hwthread_state,
217	 * because as soon as we do that, another thread can switch
218	 * the MMU context to the guest.
219	 */
220	LOAD_REG_IMMEDIATE(r7, MSR_IDLE)
221	mtmsrd	r7,0
222	bctr
223
224/*
225 * This is the sequence required to execute idle instructions, as
226 * specified in ISA v2.07 (and earlier). MSR[IR] and MSR[DR] must be 0.
227 */
228#define IDLE_STATE_ENTER_SEQ_NORET(IDLE_INST)			\
229	/* Magic NAP/SLEEP/WINKLE mode enter sequence */	\
230	std	r0,0(r1);					\
231	ptesync;						\
232	ld	r0,0(r1);					\
233236:	cmpd	cr0,r0,r0;					\
234	bne	236b;						\
235	IDLE_INST;
236
237
238	.globl pnv_enter_arch207_idle_mode
239pnv_enter_arch207_idle_mode:
240#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
241	/* Tell KVM we're entering idle */
242	li	r4,KVM_HWTHREAD_IN_IDLE
243	/******************************************************/
244	/*  N O T E   W E L L    ! ! !    N O T E   W E L L   */
245	/* The following store to HSTATE_HWTHREAD_STATE(r13)  */
246	/* MUST occur in real mode, i.e. with the MMU off,    */
247	/* and the MMU must stay off until we clear this flag */
248	/* and test HSTATE_HWTHREAD_REQ(r13) in               */
249	/* pnv_powersave_wakeup in this file.                 */
250	/* The reason is that another thread can switch the   */
251	/* MMU to a guest context whenever this flag is set   */
252	/* to KVM_HWTHREAD_IN_IDLE, and if the MMU was on,    */
253	/* that would potentially cause this thread to start  */
254	/* executing instructions from guest memory in        */
255	/* hypervisor mode, leading to a host crash or data   */
256	/* corruption, or worse.                              */
257	/******************************************************/
258	stb	r4,HSTATE_HWTHREAD_STATE(r13)
259#endif
260	stb	r3,PACA_THREAD_IDLE_STATE(r13)
261	cmpwi	cr3,r3,PNV_THREAD_SLEEP
262	bge	cr3,2f
263	IDLE_STATE_ENTER_SEQ_NORET(PPC_NAP)
264	/* No return */
2652:
266	/* Sleep or winkle */
267	lbz	r7,PACA_THREAD_MASK(r13)
268	ld	r14,PACA_CORE_IDLE_STATE_PTR(r13)
269	li	r5,0
270	beq	cr3,3f
271	lis	r5,PNV_CORE_IDLE_WINKLE_COUNT@h
2723:
273lwarx_loop1:
274	lwarx	r15,0,r14
275
276	andis.	r9,r15,PNV_CORE_IDLE_LOCK_BIT@h
277	bnel-	core_idle_lock_held
278
279	add	r15,r15,r5			/* Add if winkle */
280	andc	r15,r15,r7			/* Clear thread bit */
281
282	andi.	r9,r15,PNV_CORE_IDLE_THREAD_BITS
283
284/*
285 * If cr0 = 0, then current thread is the last thread of the core entering
286 * sleep. Last thread needs to execute the hardware bug workaround code if
287 * required by the platform.
288 * Make the workaround call unconditionally here. The below branch call is
289 * patched out when the idle states are discovered if the platform does not
290 * require it.
291 */
292.global pnv_fastsleep_workaround_at_entry
293pnv_fastsleep_workaround_at_entry:
294	beq	fastsleep_workaround_at_entry
295
296	stwcx.	r15,0,r14
297	bne-	lwarx_loop1
298	isync
299
300common_enter: /* common code for all the threads entering sleep or winkle */
301	bgt	cr3,enter_winkle
302	IDLE_STATE_ENTER_SEQ_NORET(PPC_SLEEP)
303
304fastsleep_workaround_at_entry:
305	oris	r15,r15,PNV_CORE_IDLE_LOCK_BIT@h
306	stwcx.	r15,0,r14
307	bne-	lwarx_loop1
308	isync
309
310	/* Fast sleep workaround */
311	li	r3,1
312	li	r4,1
313	bl	opal_config_cpu_idle_state
314
315	/* Unlock */
316	xoris	r15,r15,PNV_CORE_IDLE_LOCK_BIT@h
317	lwsync
318	stw	r15,0(r14)
319	b	common_enter
320
321enter_winkle:
322	bl	save_sprs_to_stack
323
324	IDLE_STATE_ENTER_SEQ_NORET(PPC_WINKLE)
325
326/*
327 * r3 - PSSCR value corresponding to the requested stop state.
328 */
329power_enter_stop:
330/*
331 * Check if we are executing the lite variant with ESL=EC=0
332 */
333	andis.   r4,r3,PSSCR_EC_ESL_MASK_SHIFTED
334	clrldi   r3,r3,60 /* r3 = Bits[60:63] = Requested Level (RL) */
335	bne	 .Lhandle_esl_ec_set
336	PPC_STOP
337	li	r3,0  /* Since we didn't lose state, return 0 */
338	std	r3, PACA_REQ_PSSCR(r13)
339
340	/*
341	 * pnv_wakeup_noloss() expects r12 to contain the SRR1 value so
342	 * it can determine if the wakeup reason is an HMI in
343	 * CHECK_HMI_INTERRUPT.
344	 *
345	 * However, when we wakeup with ESL=0, SRR1 will not contain the wakeup
346	 * reason, so there is no point setting r12 to SRR1.
347	 *
348	 * Further, we clear r12 here, so that we don't accidentally enter the
349	 * HMI in pnv_wakeup_noloss() if the value of r12[42:45] == WAKE_HMI.
350	 */
351	li	r12, 0
352	b 	pnv_wakeup_noloss
353
354.Lhandle_esl_ec_set:
355BEGIN_FTR_SECTION
356	/*
357	 * POWER9 DD2.0 or earlier can incorrectly set PMAO when waking up after
358	 * a state-loss idle. Saving and restoring MMCR0 over idle is a
359	 * workaround.
360	 */
361	mfspr	r4,SPRN_MMCR0
362	std	r4,_MMCR0(r1)
363END_FTR_SECTION_IFCLR(CPU_FTR_POWER9_DD2_1)
364
365/*
366 * Check if the requested state is a deep idle state.
367 */
368	LOAD_REG_ADDRBASE(r5,pnv_first_deep_stop_state)
369	ld	r4,ADDROFF(pnv_first_deep_stop_state)(r5)
370	cmpd	r3,r4
371	bge	.Lhandle_deep_stop
372	PPC_STOP	/* Does not return (system reset interrupt) */
373
374.Lhandle_deep_stop:
375/*
376 * Entering deep idle state.
377 * Clear thread bit in PACA_CORE_IDLE_STATE, save SPRs to
378 * stack and enter stop
379 */
380	lbz     r7,PACA_THREAD_MASK(r13)
381	ld      r14,PACA_CORE_IDLE_STATE_PTR(r13)
382
383lwarx_loop_stop:
384	lwarx   r15,0,r14
385	andis.	r9,r15,PNV_CORE_IDLE_LOCK_BIT@h
386	bnel-	core_idle_lock_held
387	andc    r15,r15,r7                      /* Clear thread bit */
388
389	stwcx.  r15,0,r14
390	bne-    lwarx_loop_stop
391	isync
392
393	bl	save_sprs_to_stack
394
395	PPC_STOP	/* Does not return (system reset interrupt) */
396
397/*
398 * Entered with MSR[EE]=0 and no soft-masked interrupts pending.
399 * r3 contains desired idle state (PNV_THREAD_NAP/SLEEP/WINKLE).
400 */
401_GLOBAL(power7_idle_insn)
402	/* Now check if user or arch enabled NAP mode */
403	LOAD_REG_ADDR(r4, pnv_enter_arch207_idle_mode)
404	b	pnv_powersave_common
405
406#define CHECK_HMI_INTERRUPT						\
407BEGIN_FTR_SECTION_NESTED(66);						\
408	rlwinm	r0,r12,45-31,0xf;  /* extract wake reason field (P8) */	\
409FTR_SECTION_ELSE_NESTED(66);						\
410	rlwinm	r0,r12,45-31,0xe;  /* P7 wake reason field is 3 bits */	\
411ALT_FTR_SECTION_END_NESTED_IFSET(CPU_FTR_ARCH_207S, 66);		\
412	cmpwi	r0,0xa;			/* Hypervisor maintenance ? */	\
413	bne+	20f;							\
414	/* Invoke opal call to handle hmi */				\
415	ld	r2,PACATOC(r13);					\
416	ld	r1,PACAR1(r13);						\
417	std	r3,ORIG_GPR3(r1);	/* Save original r3 */		\
418	li	r3,0;			/* NULL argument */		\
419	bl	hmi_exception_realmode;					\
420	nop;								\
421	ld	r3,ORIG_GPR3(r1);	/* Restore original r3 */	\
42220:	nop;
423
424/*
425 * Entered with MSR[EE]=0 and no soft-masked interrupts pending.
426 * r3 contains desired PSSCR register value.
427 *
428 * Offline (CPU unplug) case also must notify KVM that the CPU is
429 * idle.
430 */
431_GLOBAL(power9_offline_stop)
432#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
433	/*
434	 * Tell KVM we're entering idle.
435	 * This does not have to be done in real mode because the P9 MMU
436	 * is independent per-thread. Some steppings share radix/hash mode
437	 * between threads, but in that case KVM has a barrier sync in real
438	 * mode before and after switching between radix and hash.
439	 */
440	li	r4,KVM_HWTHREAD_IN_IDLE
441	stb	r4,HSTATE_HWTHREAD_STATE(r13)
442#endif
443	/* fall through */
444
445_GLOBAL(power9_idle_stop)
446	std	r3, PACA_REQ_PSSCR(r13)
447#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
448BEGIN_FTR_SECTION
449	sync
450	lwz	r5, PACA_DONT_STOP(r13)
451	cmpwi	r5, 0
452	bne	1f
453END_FTR_SECTION_IFSET(CPU_FTR_P9_TM_XER_SO_BUG)
454#endif
455	mtspr 	SPRN_PSSCR,r3
456	LOAD_REG_ADDR(r4,power_enter_stop)
457	b	pnv_powersave_common
458	/* No return */
459#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
4601:
461	/*
462	 * We get here when TM / thread reconfiguration bug workaround
463	 * code wants to get the CPU into SMT4 mode, and therefore
464	 * we are being asked not to stop.
465	 */
466	li	r3, 0
467	std	r3, PACA_REQ_PSSCR(r13)
468	blr		/* return 0 for wakeup cause / SRR1 value */
469#endif
470
471/*
472 * Called from machine check handler for powersave wakeups.
473 * Low level machine check processing has already been done. Now just
474 * go through the wake up path to get everything in order.
475 *
476 * r3 - The original SRR1 value.
477 * Original SRR[01] have been clobbered.
478 * MSR_RI is clear.
479 */
480.global pnv_powersave_wakeup_mce
481pnv_powersave_wakeup_mce:
482	/* Set cr3 for pnv_powersave_wakeup */
483	rlwinm	r11,r3,47-31,30,31
484	cmpwi	cr3,r11,2
485
486	/*
487	 * Now put the original SRR1 with SRR1_WAKEMCE_RESVD as the wake
488	 * reason into r12, which allows reuse of the system reset wakeup
489	 * code without being mistaken for another type of wakeup.
490	 */
491	oris	r12,r3,SRR1_WAKEMCE_RESVD@h
492
493	b	pnv_powersave_wakeup
494
495/*
496 * Called from reset vector for powersave wakeups.
497 * cr3 - set to gt if waking up with partial/complete hypervisor state loss
498 * r12 - SRR1
499 */
500.global pnv_powersave_wakeup
501pnv_powersave_wakeup:
502	ld	r2, PACATOC(r13)
503
504BEGIN_FTR_SECTION
505	bl	pnv_restore_hyp_resource_arch300
506FTR_SECTION_ELSE
507	bl	pnv_restore_hyp_resource_arch207
508ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_300)
509
510	li	r0,PNV_THREAD_RUNNING
511	stb	r0,PACA_THREAD_IDLE_STATE(r13)	/* Clear thread state */
512
513	mr	r3,r12
514
515#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
516	lbz	r0,HSTATE_HWTHREAD_STATE(r13)
517	cmpwi	r0,KVM_HWTHREAD_IN_KERNEL
518	beq	0f
519	li	r0,KVM_HWTHREAD_IN_KERNEL
520	stb	r0,HSTATE_HWTHREAD_STATE(r13)
521	/* Order setting hwthread_state vs. testing hwthread_req */
522	sync
5230:	lbz	r0,HSTATE_HWTHREAD_REQ(r13)
524	cmpwi	r0,0
525	beq	1f
526	b	kvm_start_guest
5271:
528#endif
529
530	/* Return SRR1 from power7_nap() */
531	blt	cr3,pnv_wakeup_noloss
532	b	pnv_wakeup_loss
533
534/*
535 * Check whether we have woken up with hypervisor state loss.
536 * If yes, restore hypervisor state and return back to link.
537 *
538 * cr3 - set to gt if waking up with partial/complete hypervisor state loss
539 */
540pnv_restore_hyp_resource_arch300:
541	/*
542	 * Workaround for POWER9, if we lost resources, the ERAT
543	 * might have been mixed up and needs flushing. We also need
544	 * to reload MMCR0 (see comment above). We also need to set
545	 * then clear bit 60 in MMCRA to ensure the PMU starts running.
546	 */
547	blt	cr3,1f
548BEGIN_FTR_SECTION
549	PPC_INVALIDATE_ERAT
550	ld	r1,PACAR1(r13)
551	ld	r4,_MMCR0(r1)
552	mtspr	SPRN_MMCR0,r4
553END_FTR_SECTION_IFCLR(CPU_FTR_POWER9_DD2_1)
554	mfspr	r4,SPRN_MMCRA
555	ori	r4,r4,(1 << (63-60))
556	mtspr	SPRN_MMCRA,r4
557	xori	r4,r4,(1 << (63-60))
558	mtspr	SPRN_MMCRA,r4
5591:
560	/*
561	 * POWER ISA 3. Use PSSCR to determine if we
562	 * are waking up from deep idle state
563	 */
564	LOAD_REG_ADDRBASE(r5,pnv_first_deep_stop_state)
565	ld	r4,ADDROFF(pnv_first_deep_stop_state)(r5)
566
567	/*
568	 * 0-3 bits correspond to Power-Saving Level Status
569	 * which indicates the idle state we are waking up from
570	 */
571	mfspr	r5, SPRN_PSSCR
572	rldicl  r5,r5,4,60
573	li	r0, 0		/* clear requested_psscr to say we're awake */
574	std	r0, PACA_REQ_PSSCR(r13)
575	cmpd	cr4,r5,r4
576	bge	cr4,pnv_wakeup_tb_loss /* returns to caller */
577
578	blr	/* Waking up without hypervisor state loss. */
579
580/* Same calling convention as arch300 */
581pnv_restore_hyp_resource_arch207:
582	/*
583	 * POWER ISA 2.07 or less.
584	 * Check if we slept with sleep or winkle.
585	 */
586	lbz	r4,PACA_THREAD_IDLE_STATE(r13)
587	cmpwi	cr2,r4,PNV_THREAD_NAP
588	bgt	cr2,pnv_wakeup_tb_loss	/* Either sleep or Winkle */
589
590	/*
591	 * We fall through here if PACA_THREAD_IDLE_STATE shows we are waking
592	 * up from nap. At this stage CR3 shouldn't contains 'gt' since that
593	 * indicates we are waking with hypervisor state loss from nap.
594	 */
595	bgt	cr3,.
596
597	blr	/* Waking up without hypervisor state loss */
598
599/*
600 * Called if waking up from idle state which can cause either partial or
601 * complete hyp state loss.
602 * In POWER8, called if waking up from fastsleep or winkle
603 * In POWER9, called if waking up from stop state >= pnv_first_deep_stop_state
604 *
605 * r13 - PACA
606 * cr3 - gt if waking up with partial/complete hypervisor state loss
607 *
608 * If ISA300:
609 * cr4 - gt or eq if waking up from complete hypervisor state loss.
610 *
611 * If ISA207:
612 * r4 - PACA_THREAD_IDLE_STATE
613 */
614pnv_wakeup_tb_loss:
615	ld	r1,PACAR1(r13)
616	/*
617	 * Before entering any idle state, the NVGPRs are saved in the stack.
618	 * If there was a state loss, or PACA_NAPSTATELOST was set, then the
619	 * NVGPRs are restored. If we are here, it is likely that state is lost,
620	 * but not guaranteed -- neither ISA207 nor ISA300 tests to reach
621	 * here are the same as the test to restore NVGPRS:
622	 * PACA_THREAD_IDLE_STATE test for ISA207, PSSCR test for ISA300,
623	 * and SRR1 test for restoring NVGPRs.
624	 *
625	 * We are about to clobber NVGPRs now, so set NAPSTATELOST to
626	 * guarantee they will always be restored. This might be tightened
627	 * with careful reading of specs (particularly for ISA300) but this
628	 * is already a slow wakeup path and it's simpler to be safe.
629	 */
630	li	r0,1
631	stb	r0,PACA_NAPSTATELOST(r13)
632
633	/*
634	 *
635	 * Save SRR1 and LR in NVGPRs as they might be clobbered in
636	 * opal_call() (called in CHECK_HMI_INTERRUPT). SRR1 is required
637	 * to determine the wakeup reason if we branch to kvm_start_guest. LR
638	 * is required to return back to reset vector after hypervisor state
639	 * restore is complete.
640	 */
641	mr	r19,r12
642	mr	r18,r4
643	mflr	r17
644BEGIN_FTR_SECTION
645	CHECK_HMI_INTERRUPT
646END_FTR_SECTION_IFSET(CPU_FTR_HVMODE)
647
648	ld	r14,PACA_CORE_IDLE_STATE_PTR(r13)
649	lbz	r7,PACA_THREAD_MASK(r13)
650
651	/*
652	 * Take the core lock to synchronize against other threads.
653	 *
654	 * Lock bit is set in one of the 2 cases-
655	 * a. In the sleep/winkle enter path, the last thread is executing
656	 * fastsleep workaround code.
657	 * b. In the wake up path, another thread is executing fastsleep
658	 * workaround undo code or resyncing timebase or restoring context
659	 * In either case loop until the lock bit is cleared.
660	 */
6611:
662	lwarx	r15,0,r14
663	andis.	r9,r15,PNV_CORE_IDLE_LOCK_BIT@h
664	bnel-	core_idle_lock_held
665	oris	r15,r15,PNV_CORE_IDLE_LOCK_BIT@h
666	stwcx.	r15,0,r14
667	bne-	1b
668	isync
669
670	andi.	r9,r15,PNV_CORE_IDLE_THREAD_BITS
671	cmpwi	cr2,r9,0
672
673	/*
674	 * At this stage
675	 * cr2 - eq if first thread to wakeup in core
676	 * cr3-  gt if waking up with partial/complete hypervisor state loss
677	 * ISA300:
678	 * cr4 - gt or eq if waking up from complete hypervisor state loss.
679	 */
680
681BEGIN_FTR_SECTION
682	/*
683	 * Were we in winkle?
684	 * If yes, check if all threads were in winkle, decrement our
685	 * winkle count, set all thread winkle bits if all were in winkle.
686	 * Check if our thread has a winkle bit set, and set cr4 accordingly
687	 * (to match ISA300, above). Pseudo-code for core idle state
688	 * transitions for ISA207 is as follows (everything happens atomically
689	 * due to store conditional and/or lock bit):
690	 *
691	 * nap_idle() { }
692	 * nap_wake() { }
693	 *
694	 * sleep_idle()
695	 * {
696	 *	core_idle_state &= ~thread_in_core
697	 * }
698	 *
699	 * sleep_wake()
700	 * {
701	 *     bool first_in_core, first_in_subcore;
702	 *
703	 *     first_in_core = (core_idle_state & IDLE_THREAD_BITS) == 0;
704	 *     first_in_subcore = (core_idle_state & SUBCORE_SIBLING_MASK) == 0;
705	 *
706	 *     core_idle_state |= thread_in_core;
707	 * }
708	 *
709	 * winkle_idle()
710	 * {
711	 *	core_idle_state &= ~thread_in_core;
712	 *	core_idle_state += 1 << WINKLE_COUNT_SHIFT;
713	 * }
714	 *
715	 * winkle_wake()
716	 * {
717	 *     bool first_in_core, first_in_subcore, winkle_state_lost;
718	 *
719	 *     first_in_core = (core_idle_state & IDLE_THREAD_BITS) == 0;
720	 *     first_in_subcore = (core_idle_state & SUBCORE_SIBLING_MASK) == 0;
721	 *
722	 *     core_idle_state |= thread_in_core;
723	 *
724	 *     if ((core_idle_state & WINKLE_MASK) == (8 << WINKLE_COUNT_SIHFT))
725	 *         core_idle_state |= THREAD_WINKLE_BITS;
726	 *     core_idle_state -= 1 << WINKLE_COUNT_SHIFT;
727	 *
728	 *     winkle_state_lost = core_idle_state &
729	 *				(thread_in_core << WINKLE_THREAD_SHIFT);
730	 *     core_idle_state &= ~(thread_in_core << WINKLE_THREAD_SHIFT);
731	 * }
732	 *
733	 */
734	cmpwi	r18,PNV_THREAD_WINKLE
735	bne	2f
736	andis.	r9,r15,PNV_CORE_IDLE_WINKLE_COUNT_ALL_BIT@h
737	subis	r15,r15,PNV_CORE_IDLE_WINKLE_COUNT@h
738	beq	2f
739	ori	r15,r15,PNV_CORE_IDLE_THREAD_WINKLE_BITS /* all were winkle */
7402:
741	/* Shift thread bit to winkle mask, then test if this thread is set,
742	 * and remove it from the winkle bits */
743	slwi	r8,r7,8
744	and	r8,r8,r15
745	andc	r15,r15,r8
746	cmpwi	cr4,r8,1 /* cr4 will be gt if our bit is set, lt if not */
747
748	lbz	r4,PACA_SUBCORE_SIBLING_MASK(r13)
749	and	r4,r4,r15
750	cmpwi	r4,0	/* Check if first in subcore */
751
752	or	r15,r15,r7		/* Set thread bit */
753	beq	first_thread_in_subcore
754END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
755
756	or	r15,r15,r7		/* Set thread bit */
757	beq	cr2,first_thread_in_core
758
759	/* Not first thread in core or subcore to wake up */
760	b	clear_lock
761
762first_thread_in_subcore:
763	/*
764	 * If waking up from sleep, subcore state is not lost. Hence
765	 * skip subcore state restore
766	 */
767	blt	cr4,subcore_state_restored
768
769	/* Restore per-subcore state */
770	ld      r4,_SDR1(r1)
771	mtspr   SPRN_SDR1,r4
772
773	ld      r4,_RPR(r1)
774	mtspr   SPRN_RPR,r4
775	ld	r4,_AMOR(r1)
776	mtspr	SPRN_AMOR,r4
777
778subcore_state_restored:
779	/*
780	 * Check if the thread is also the first thread in the core. If not,
781	 * skip to clear_lock.
782	 */
783	bne	cr2,clear_lock
784
785first_thread_in_core:
786
787	/*
788	 * First thread in the core waking up from any state which can cause
789	 * partial or complete hypervisor state loss. It needs to
790	 * call the fastsleep workaround code if the platform requires it.
791	 * Call it unconditionally here. The below branch instruction will
792	 * be patched out if the platform does not have fastsleep or does not
793	 * require the workaround. Patching will be performed during the
794	 * discovery of idle-states.
795	 */
796.global pnv_fastsleep_workaround_at_exit
797pnv_fastsleep_workaround_at_exit:
798	b	fastsleep_workaround_at_exit
799
800timebase_resync:
801	/*
802	 * Use cr3 which indicates that we are waking up with atleast partial
803	 * hypervisor state loss to determine if TIMEBASE RESYNC is needed.
804	 */
805	ble	cr3,.Ltb_resynced
806	/* Time base re-sync */
807	bl	opal_resync_timebase;
808	/*
809	 * If waking up from sleep (POWER8), per core state
810	 * is not lost, skip to clear_lock.
811	 */
812.Ltb_resynced:
813	blt	cr4,clear_lock
814
815	/*
816	 * First thread in the core to wake up and its waking up with
817	 * complete hypervisor state loss. Restore per core hypervisor
818	 * state.
819	 */
820BEGIN_FTR_SECTION
821	ld	r4,_PTCR(r1)
822	mtspr	SPRN_PTCR,r4
823	ld	r4,_RPR(r1)
824	mtspr	SPRN_RPR,r4
825	ld	r4,_AMOR(r1)
826	mtspr	SPRN_AMOR,r4
827END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
828
829	ld	r4,_TSCR(r1)
830	mtspr	SPRN_TSCR,r4
831	ld	r4,_WORC(r1)
832	mtspr	SPRN_WORC,r4
833
834clear_lock:
835	xoris	r15,r15,PNV_CORE_IDLE_LOCK_BIT@h
836	lwsync
837	stw	r15,0(r14)
838
839common_exit:
840	/*
841	 * Common to all threads.
842	 *
843	 * If waking up from sleep, hypervisor state is not lost. Hence
844	 * skip hypervisor state restore.
845	 */
846	blt	cr4,hypervisor_state_restored
847
848	/* Waking up from winkle */
849
850BEGIN_MMU_FTR_SECTION
851	b	no_segments
852END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
853	/* Restore SLB  from PACA */
854	ld	r8,PACA_SLBSHADOWPTR(r13)
855
856	.rept	SLB_NUM_BOLTED
857	li	r3, SLBSHADOW_SAVEAREA
858	LDX_BE	r5, r8, r3
859	addi	r3, r3, 8
860	LDX_BE	r6, r8, r3
861	andis.	r7,r5,SLB_ESID_V@h
862	beq	1f
863	slbmte	r6,r5
8641:	addi	r8,r8,16
865	.endr
866no_segments:
867
868	/* Restore per thread state */
869
870	ld	r4,_SPURR(r1)
871	mtspr	SPRN_SPURR,r4
872	ld	r4,_PURR(r1)
873	mtspr	SPRN_PURR,r4
874	ld	r4,_DSCR(r1)
875	mtspr	SPRN_DSCR,r4
876	ld	r4,_WORT(r1)
877	mtspr	SPRN_WORT,r4
878
879	/* Call cur_cpu_spec->cpu_restore() */
880	LOAD_REG_ADDR(r4, cur_cpu_spec)
881	ld	r4,0(r4)
882	ld	r12,CPU_SPEC_RESTORE(r4)
883#ifdef PPC64_ELF_ABI_v1
884	ld	r12,0(r12)
885#endif
886	mtctr	r12
887	bctrl
888
889/*
890 * On POWER9, we can come here on wakeup from a cpuidle stop state.
891 * Hence restore the additional SPRs to the saved value.
892 *
893 * On POWER8, we come here only on winkle. Since winkle is used
894 * only in the case of CPU-Hotplug, we don't need to restore
895 * the additional SPRs.
896 */
897BEGIN_FTR_SECTION
898	bl 	power9_restore_additional_sprs
899END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
900hypervisor_state_restored:
901
902	mr	r12,r19
903	mtlr	r17
904	blr		/* return to pnv_powersave_wakeup */
905
906fastsleep_workaround_at_exit:
907	li	r3,1
908	li	r4,0
909	bl	opal_config_cpu_idle_state
910	b	timebase_resync
911
912/*
913 * R3 here contains the value that will be returned to the caller
914 * of power7_nap.
915 * R12 contains SRR1 for CHECK_HMI_INTERRUPT.
916 */
917.global pnv_wakeup_loss
918pnv_wakeup_loss:
919	ld	r1,PACAR1(r13)
920BEGIN_FTR_SECTION
921	CHECK_HMI_INTERRUPT
922END_FTR_SECTION_IFSET(CPU_FTR_HVMODE)
923	REST_NVGPRS(r1)
924	REST_GPR(2, r1)
925	ld	r4,PACAKMSR(r13)
926	ld	r5,_LINK(r1)
927	ld	r6,_CCR(r1)
928	addi	r1,r1,INT_FRAME_SIZE
929	mtlr	r5
930	mtcr	r6
931	mtmsrd	r4
932	blr
933
934/*
935 * R3 here contains the value that will be returned to the caller
936 * of power7_nap.
937 * R12 contains SRR1 for CHECK_HMI_INTERRUPT.
938 */
939pnv_wakeup_noloss:
940	lbz	r0,PACA_NAPSTATELOST(r13)
941	cmpwi	r0,0
942	bne	pnv_wakeup_loss
943	ld	r1,PACAR1(r13)
944BEGIN_FTR_SECTION
945	CHECK_HMI_INTERRUPT
946END_FTR_SECTION_IFSET(CPU_FTR_HVMODE)
947	ld	r4,PACAKMSR(r13)
948	ld	r5,_NIP(r1)
949	ld	r6,_CCR(r1)
950	addi	r1,r1,INT_FRAME_SIZE
951	mtlr	r5
952	mtcr	r6
953	mtmsrd	r4
954	blr
955