xref: /openbmc/linux/arch/powerpc/kernel/idle_book3s.S (revision 19725496da5602b401eae389736ab00d1817e264)
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
27#undef DEBUG
28
29/*
30 * Use unused space in the interrupt stack to save and restore
31 * registers for winkle support.
32 */
33#define _MMCR0	GPR0
34#define _SDR1	GPR3
35#define _PTCR	GPR3
36#define _RPR	GPR4
37#define _SPURR	GPR5
38#define _PURR	GPR6
39#define _TSCR	GPR7
40#define _DSCR	GPR8
41#define _AMOR	GPR9
42#define _WORT	GPR10
43#define _WORC	GPR11
44#define _LPCR	GPR12
45
46#define PSSCR_EC_ESL_MASK_SHIFTED          (PSSCR_EC | PSSCR_ESL) >> 16
47
48	.text
49
50/*
51 * Used by threads before entering deep idle states. Saves SPRs
52 * in interrupt stack frame
53 */
54save_sprs_to_stack:
55	/*
56	 * Note all register i.e per-core, per-subcore or per-thread is saved
57	 * here since any thread in the core might wake up first
58	 */
59BEGIN_FTR_SECTION
60	/*
61	 * Note - SDR1 is dropped in Power ISA v3. Hence not restoring
62	 * SDR1 here
63	 */
64	mfspr	r3,SPRN_PTCR
65	std	r3,_PTCR(r1)
66	mfspr	r3,SPRN_LPCR
67	std	r3,_LPCR(r1)
68FTR_SECTION_ELSE
69	mfspr	r3,SPRN_SDR1
70	std	r3,_SDR1(r1)
71ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_300)
72	mfspr	r3,SPRN_RPR
73	std	r3,_RPR(r1)
74	mfspr	r3,SPRN_SPURR
75	std	r3,_SPURR(r1)
76	mfspr	r3,SPRN_PURR
77	std	r3,_PURR(r1)
78	mfspr	r3,SPRN_TSCR
79	std	r3,_TSCR(r1)
80	mfspr	r3,SPRN_DSCR
81	std	r3,_DSCR(r1)
82	mfspr	r3,SPRN_AMOR
83	std	r3,_AMOR(r1)
84	mfspr	r3,SPRN_WORT
85	std	r3,_WORT(r1)
86	mfspr	r3,SPRN_WORC
87	std	r3,_WORC(r1)
88/*
89 * On POWER9, there are idle states such as stop4, invoked via cpuidle,
90 * that lose hypervisor resources. In such cases, we need to save
91 * additional SPRs before entering those idle states so that they can
92 * be restored to their older values on wakeup from the idle state.
93 *
94 * On POWER8, the only such deep idle state is winkle which is used
95 * only in the context of CPU-Hotplug, where these additional SPRs are
96 * reinitiazed to a sane value. Hence there is no need to save/restore
97 * these SPRs.
98 */
99BEGIN_FTR_SECTION
100	blr
101END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
102
103power9_save_additional_sprs:
104	mfspr	r3, SPRN_PID
105	mfspr	r4, SPRN_LDBAR
106	std	r3, STOP_PID(r13)
107	std	r4, STOP_LDBAR(r13)
108
109	mfspr	r3, SPRN_FSCR
110	mfspr	r4, SPRN_HFSCR
111	std	r3, STOP_FSCR(r13)
112	std	r4, STOP_HFSCR(r13)
113
114	mfspr	r3, SPRN_MMCRA
115	mfspr	r4, SPRN_MMCR0
116	std	r3, STOP_MMCRA(r13)
117	std	r4, _MMCR0(r1)
118
119	mfspr	r3, SPRN_MMCR1
120	mfspr	r4, SPRN_MMCR2
121	std	r3, STOP_MMCR1(r13)
122	std	r4, STOP_MMCR2(r13)
123	blr
124
125power9_restore_additional_sprs:
126	ld	r3,_LPCR(r1)
127	ld	r4, STOP_PID(r13)
128	mtspr	SPRN_LPCR,r3
129	mtspr	SPRN_PID, r4
130
131	ld	r3, STOP_LDBAR(r13)
132	ld	r4, STOP_FSCR(r13)
133	mtspr	SPRN_LDBAR, r3
134	mtspr	SPRN_FSCR, r4
135
136	ld	r3, STOP_HFSCR(r13)
137	ld	r4, STOP_MMCRA(r13)
138	mtspr	SPRN_HFSCR, r3
139	mtspr	SPRN_MMCRA, r4
140
141	ld	r3, _MMCR0(r1)
142	ld	r4, STOP_MMCR1(r13)
143	mtspr	SPRN_MMCR0, r3
144	mtspr	SPRN_MMCR1, r4
145
146	ld	r3, STOP_MMCR2(r13)
147	ld	r4, PACA_SPRG_VDSO(r13)
148	mtspr	SPRN_MMCR2, r3
149	mtspr	SPRN_SPRG3, r4
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 * On waking up from stop 0,1,2 with ESL=1 on POWER9 DD1,
473 * HSPRG0 will be set to the HSPRG0 value of one of the
474 * threads in this core. Thus the value we have in r13
475 * may not be this thread's paca pointer.
476 *
477 * Fortunately, the TIR remains invariant. Since this thread's
478 * paca pointer is recorded in all its sibling's paca, we can
479 * correctly recover this thread's paca pointer if we
480 * know the index of this thread in the core.
481 *
482 * This index can be obtained from the TIR.
483 *
484 * i.e, thread's position in the core = TIR.
485 * If this value is i, then this thread's paca is
486 * paca->thread_sibling_pacas[i].
487 */
488power9_dd1_recover_paca:
489	mfspr	r4, SPRN_TIR
490	/*
491	 * Since each entry in thread_sibling_pacas is 8 bytes
492	 * we need to left-shift by 3 bits. Thus r4 = i * 8
493	 */
494	sldi	r4, r4, 3
495	/* Get &paca->thread_sibling_pacas[0] in r5 */
496	ld	r5, PACA_SIBLING_PACA_PTRS(r13)
497	/* Load paca->thread_sibling_pacas[i] into r13 */
498	ldx	r13, r4, r5
499	SET_PACA(r13)
500	/*
501	 * Indicate that we have lost NVGPR state
502	 * which needs to be restored from the stack.
503	 */
504	li	r3, 1
505	stb	r3,PACA_NAPSTATELOST(r13)
506	blr
507
508/*
509 * Called from machine check handler for powersave wakeups.
510 * Low level machine check processing has already been done. Now just
511 * go through the wake up path to get everything in order.
512 *
513 * r3 - The original SRR1 value.
514 * Original SRR[01] have been clobbered.
515 * MSR_RI is clear.
516 */
517.global pnv_powersave_wakeup_mce
518pnv_powersave_wakeup_mce:
519	/* Set cr3 for pnv_powersave_wakeup */
520	rlwinm	r11,r3,47-31,30,31
521	cmpwi	cr3,r11,2
522
523	/*
524	 * Now put the original SRR1 with SRR1_WAKEMCE_RESVD as the wake
525	 * reason into r12, which allows reuse of the system reset wakeup
526	 * code without being mistaken for another type of wakeup.
527	 */
528	oris	r12,r3,SRR1_WAKEMCE_RESVD@h
529
530	b	pnv_powersave_wakeup
531
532/*
533 * Called from reset vector for powersave wakeups.
534 * cr3 - set to gt if waking up with partial/complete hypervisor state loss
535 * r12 - SRR1
536 */
537.global pnv_powersave_wakeup
538pnv_powersave_wakeup:
539	ld	r2, PACATOC(r13)
540
541BEGIN_FTR_SECTION
542BEGIN_FTR_SECTION_NESTED(70)
543	bl	power9_dd1_recover_paca
544END_FTR_SECTION_NESTED_IFSET(CPU_FTR_POWER9_DD1, 70)
545	bl	pnv_restore_hyp_resource_arch300
546FTR_SECTION_ELSE
547	bl	pnv_restore_hyp_resource_arch207
548ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_300)
549
550	li	r0,PNV_THREAD_RUNNING
551	stb	r0,PACA_THREAD_IDLE_STATE(r13)	/* Clear thread state */
552
553	mr	r3,r12
554
555#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
556	lbz	r0,HSTATE_HWTHREAD_STATE(r13)
557	cmpwi	r0,KVM_HWTHREAD_IN_KERNEL
558	beq	0f
559	li	r0,KVM_HWTHREAD_IN_KERNEL
560	stb	r0,HSTATE_HWTHREAD_STATE(r13)
561	/* Order setting hwthread_state vs. testing hwthread_req */
562	sync
5630:	lbz	r0,HSTATE_HWTHREAD_REQ(r13)
564	cmpwi	r0,0
565	beq	1f
566	b	kvm_start_guest
5671:
568#endif
569
570	/* Return SRR1 from power7_nap() */
571	blt	cr3,pnv_wakeup_noloss
572	b	pnv_wakeup_loss
573
574/*
575 * Check whether we have woken up with hypervisor state loss.
576 * If yes, restore hypervisor state and return back to link.
577 *
578 * cr3 - set to gt if waking up with partial/complete hypervisor state loss
579 */
580pnv_restore_hyp_resource_arch300:
581	/*
582	 * Workaround for POWER9, if we lost resources, the ERAT
583	 * might have been mixed up and needs flushing. We also need
584	 * to reload MMCR0 (see comment above). We also need to set
585	 * then clear bit 60 in MMCRA to ensure the PMU starts running.
586	 */
587	blt	cr3,1f
588BEGIN_FTR_SECTION
589	PPC_INVALIDATE_ERAT
590	ld	r1,PACAR1(r13)
591	ld	r4,_MMCR0(r1)
592	mtspr	SPRN_MMCR0,r4
593END_FTR_SECTION_IFCLR(CPU_FTR_POWER9_DD2_1)
594	mfspr	r4,SPRN_MMCRA
595	ori	r4,r4,(1 << (63-60))
596	mtspr	SPRN_MMCRA,r4
597	xori	r4,r4,(1 << (63-60))
598	mtspr	SPRN_MMCRA,r4
5991:
600	/*
601	 * POWER ISA 3. Use PSSCR to determine if we
602	 * are waking up from deep idle state
603	 */
604	LOAD_REG_ADDRBASE(r5,pnv_first_deep_stop_state)
605	ld	r4,ADDROFF(pnv_first_deep_stop_state)(r5)
606
607BEGIN_FTR_SECTION_NESTED(71)
608	/*
609	 * Assume that we are waking up from the state
610	 * same as the Requested Level (RL) in the PSSCR
611	 * which are Bits 60-63
612	 */
613	ld	r5,PACA_REQ_PSSCR(r13)
614	rldicl  r5,r5,0,60
615FTR_SECTION_ELSE_NESTED(71)
616	/*
617	 * 0-3 bits correspond to Power-Saving Level Status
618	 * which indicates the idle state we are waking up from
619	 */
620	mfspr	r5, SPRN_PSSCR
621	rldicl  r5,r5,4,60
622ALT_FTR_SECTION_END_NESTED_IFSET(CPU_FTR_POWER9_DD1, 71)
623	li	r0, 0		/* clear requested_psscr to say we're awake */
624	std	r0, PACA_REQ_PSSCR(r13)
625	cmpd	cr4,r5,r4
626	bge	cr4,pnv_wakeup_tb_loss /* returns to caller */
627
628	blr	/* Waking up without hypervisor state loss. */
629
630/* Same calling convention as arch300 */
631pnv_restore_hyp_resource_arch207:
632	/*
633	 * POWER ISA 2.07 or less.
634	 * Check if we slept with sleep or winkle.
635	 */
636	lbz	r4,PACA_THREAD_IDLE_STATE(r13)
637	cmpwi	cr2,r4,PNV_THREAD_NAP
638	bgt	cr2,pnv_wakeup_tb_loss	/* Either sleep or Winkle */
639
640	/*
641	 * We fall through here if PACA_THREAD_IDLE_STATE shows we are waking
642	 * up from nap. At this stage CR3 shouldn't contains 'gt' since that
643	 * indicates we are waking with hypervisor state loss from nap.
644	 */
645	bgt	cr3,.
646
647	blr	/* Waking up without hypervisor state loss */
648
649/*
650 * Called if waking up from idle state which can cause either partial or
651 * complete hyp state loss.
652 * In POWER8, called if waking up from fastsleep or winkle
653 * In POWER9, called if waking up from stop state >= pnv_first_deep_stop_state
654 *
655 * r13 - PACA
656 * cr3 - gt if waking up with partial/complete hypervisor state loss
657 *
658 * If ISA300:
659 * cr4 - gt or eq if waking up from complete hypervisor state loss.
660 *
661 * If ISA207:
662 * r4 - PACA_THREAD_IDLE_STATE
663 */
664pnv_wakeup_tb_loss:
665	ld	r1,PACAR1(r13)
666	/*
667	 * Before entering any idle state, the NVGPRs are saved in the stack.
668	 * If there was a state loss, or PACA_NAPSTATELOST was set, then the
669	 * NVGPRs are restored. If we are here, it is likely that state is lost,
670	 * but not guaranteed -- neither ISA207 nor ISA300 tests to reach
671	 * here are the same as the test to restore NVGPRS:
672	 * PACA_THREAD_IDLE_STATE test for ISA207, PSSCR test for ISA300,
673	 * and SRR1 test for restoring NVGPRs.
674	 *
675	 * We are about to clobber NVGPRs now, so set NAPSTATELOST to
676	 * guarantee they will always be restored. This might be tightened
677	 * with careful reading of specs (particularly for ISA300) but this
678	 * is already a slow wakeup path and it's simpler to be safe.
679	 */
680	li	r0,1
681	stb	r0,PACA_NAPSTATELOST(r13)
682
683	/*
684	 *
685	 * Save SRR1 and LR in NVGPRs as they might be clobbered in
686	 * opal_call() (called in CHECK_HMI_INTERRUPT). SRR1 is required
687	 * to determine the wakeup reason if we branch to kvm_start_guest. LR
688	 * is required to return back to reset vector after hypervisor state
689	 * restore is complete.
690	 */
691	mr	r19,r12
692	mr	r18,r4
693	mflr	r17
694BEGIN_FTR_SECTION
695	CHECK_HMI_INTERRUPT
696END_FTR_SECTION_IFSET(CPU_FTR_HVMODE)
697
698	ld	r14,PACA_CORE_IDLE_STATE_PTR(r13)
699	lbz	r7,PACA_THREAD_MASK(r13)
700
701	/*
702	 * Take the core lock to synchronize against other threads.
703	 *
704	 * Lock bit is set in one of the 2 cases-
705	 * a. In the sleep/winkle enter path, the last thread is executing
706	 * fastsleep workaround code.
707	 * b. In the wake up path, another thread is executing fastsleep
708	 * workaround undo code or resyncing timebase or restoring context
709	 * In either case loop until the lock bit is cleared.
710	 */
7111:
712	lwarx	r15,0,r14
713	andis.	r9,r15,PNV_CORE_IDLE_LOCK_BIT@h
714	bnel-	core_idle_lock_held
715	oris	r15,r15,PNV_CORE_IDLE_LOCK_BIT@h
716	stwcx.	r15,0,r14
717	bne-	1b
718	isync
719
720	andi.	r9,r15,PNV_CORE_IDLE_THREAD_BITS
721	cmpwi	cr2,r9,0
722
723	/*
724	 * At this stage
725	 * cr2 - eq if first thread to wakeup in core
726	 * cr3-  gt if waking up with partial/complete hypervisor state loss
727	 * ISA300:
728	 * cr4 - gt or eq if waking up from complete hypervisor state loss.
729	 */
730
731BEGIN_FTR_SECTION
732	/*
733	 * Were we in winkle?
734	 * If yes, check if all threads were in winkle, decrement our
735	 * winkle count, set all thread winkle bits if all were in winkle.
736	 * Check if our thread has a winkle bit set, and set cr4 accordingly
737	 * (to match ISA300, above). Pseudo-code for core idle state
738	 * transitions for ISA207 is as follows (everything happens atomically
739	 * due to store conditional and/or lock bit):
740	 *
741	 * nap_idle() { }
742	 * nap_wake() { }
743	 *
744	 * sleep_idle()
745	 * {
746	 *	core_idle_state &= ~thread_in_core
747	 * }
748	 *
749	 * sleep_wake()
750	 * {
751	 *     bool first_in_core, first_in_subcore;
752	 *
753	 *     first_in_core = (core_idle_state & IDLE_THREAD_BITS) == 0;
754	 *     first_in_subcore = (core_idle_state & SUBCORE_SIBLING_MASK) == 0;
755	 *
756	 *     core_idle_state |= thread_in_core;
757	 * }
758	 *
759	 * winkle_idle()
760	 * {
761	 *	core_idle_state &= ~thread_in_core;
762	 *	core_idle_state += 1 << WINKLE_COUNT_SHIFT;
763	 * }
764	 *
765	 * winkle_wake()
766	 * {
767	 *     bool first_in_core, first_in_subcore, winkle_state_lost;
768	 *
769	 *     first_in_core = (core_idle_state & IDLE_THREAD_BITS) == 0;
770	 *     first_in_subcore = (core_idle_state & SUBCORE_SIBLING_MASK) == 0;
771	 *
772	 *     core_idle_state |= thread_in_core;
773	 *
774	 *     if ((core_idle_state & WINKLE_MASK) == (8 << WINKLE_COUNT_SIHFT))
775	 *         core_idle_state |= THREAD_WINKLE_BITS;
776	 *     core_idle_state -= 1 << WINKLE_COUNT_SHIFT;
777	 *
778	 *     winkle_state_lost = core_idle_state &
779	 *				(thread_in_core << WINKLE_THREAD_SHIFT);
780	 *     core_idle_state &= ~(thread_in_core << WINKLE_THREAD_SHIFT);
781	 * }
782	 *
783	 */
784	cmpwi	r18,PNV_THREAD_WINKLE
785	bne	2f
786	andis.	r9,r15,PNV_CORE_IDLE_WINKLE_COUNT_ALL_BIT@h
787	subis	r15,r15,PNV_CORE_IDLE_WINKLE_COUNT@h
788	beq	2f
789	ori	r15,r15,PNV_CORE_IDLE_THREAD_WINKLE_BITS /* all were winkle */
7902:
791	/* Shift thread bit to winkle mask, then test if this thread is set,
792	 * and remove it from the winkle bits */
793	slwi	r8,r7,8
794	and	r8,r8,r15
795	andc	r15,r15,r8
796	cmpwi	cr4,r8,1 /* cr4 will be gt if our bit is set, lt if not */
797
798	lbz	r4,PACA_SUBCORE_SIBLING_MASK(r13)
799	and	r4,r4,r15
800	cmpwi	r4,0	/* Check if first in subcore */
801
802	or	r15,r15,r7		/* Set thread bit */
803	beq	first_thread_in_subcore
804END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
805
806	or	r15,r15,r7		/* Set thread bit */
807	beq	cr2,first_thread_in_core
808
809	/* Not first thread in core or subcore to wake up */
810	b	clear_lock
811
812first_thread_in_subcore:
813	/*
814	 * If waking up from sleep, subcore state is not lost. Hence
815	 * skip subcore state restore
816	 */
817	blt	cr4,subcore_state_restored
818
819	/* Restore per-subcore state */
820	ld      r4,_SDR1(r1)
821	mtspr   SPRN_SDR1,r4
822
823	ld      r4,_RPR(r1)
824	mtspr   SPRN_RPR,r4
825	ld	r4,_AMOR(r1)
826	mtspr	SPRN_AMOR,r4
827
828subcore_state_restored:
829	/*
830	 * Check if the thread is also the first thread in the core. If not,
831	 * skip to clear_lock.
832	 */
833	bne	cr2,clear_lock
834
835first_thread_in_core:
836
837	/*
838	 * First thread in the core waking up from any state which can cause
839	 * partial or complete hypervisor state loss. It needs to
840	 * call the fastsleep workaround code if the platform requires it.
841	 * Call it unconditionally here. The below branch instruction will
842	 * be patched out if the platform does not have fastsleep or does not
843	 * require the workaround. Patching will be performed during the
844	 * discovery of idle-states.
845	 */
846.global pnv_fastsleep_workaround_at_exit
847pnv_fastsleep_workaround_at_exit:
848	b	fastsleep_workaround_at_exit
849
850timebase_resync:
851	/*
852	 * Use cr3 which indicates that we are waking up with atleast partial
853	 * hypervisor state loss to determine if TIMEBASE RESYNC is needed.
854	 */
855	ble	cr3,.Ltb_resynced
856	/* Time base re-sync */
857	bl	opal_resync_timebase;
858	/*
859	 * If waking up from sleep (POWER8), per core state
860	 * is not lost, skip to clear_lock.
861	 */
862.Ltb_resynced:
863	blt	cr4,clear_lock
864
865	/*
866	 * First thread in the core to wake up and its waking up with
867	 * complete hypervisor state loss. Restore per core hypervisor
868	 * state.
869	 */
870BEGIN_FTR_SECTION
871	ld	r4,_PTCR(r1)
872	mtspr	SPRN_PTCR,r4
873	ld	r4,_RPR(r1)
874	mtspr	SPRN_RPR,r4
875	ld	r4,_AMOR(r1)
876	mtspr	SPRN_AMOR,r4
877END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
878
879	ld	r4,_TSCR(r1)
880	mtspr	SPRN_TSCR,r4
881	ld	r4,_WORC(r1)
882	mtspr	SPRN_WORC,r4
883
884clear_lock:
885	xoris	r15,r15,PNV_CORE_IDLE_LOCK_BIT@h
886	lwsync
887	stw	r15,0(r14)
888
889common_exit:
890	/*
891	 * Common to all threads.
892	 *
893	 * If waking up from sleep, hypervisor state is not lost. Hence
894	 * skip hypervisor state restore.
895	 */
896	blt	cr4,hypervisor_state_restored
897
898	/* Waking up from winkle */
899
900BEGIN_MMU_FTR_SECTION
901	b	no_segments
902END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
903	/* Restore SLB  from PACA */
904	ld	r8,PACA_SLBSHADOWPTR(r13)
905
906	.rept	SLB_NUM_BOLTED
907	li	r3, SLBSHADOW_SAVEAREA
908	LDX_BE	r5, r8, r3
909	addi	r3, r3, 8
910	LDX_BE	r6, r8, r3
911	andis.	r7,r5,SLB_ESID_V@h
912	beq	1f
913	slbmte	r6,r5
9141:	addi	r8,r8,16
915	.endr
916no_segments:
917
918	/* Restore per thread state */
919
920	ld	r4,_SPURR(r1)
921	mtspr	SPRN_SPURR,r4
922	ld	r4,_PURR(r1)
923	mtspr	SPRN_PURR,r4
924	ld	r4,_DSCR(r1)
925	mtspr	SPRN_DSCR,r4
926	ld	r4,_WORT(r1)
927	mtspr	SPRN_WORT,r4
928
929	/* Call cur_cpu_spec->cpu_restore() */
930	LOAD_REG_ADDR(r4, cur_cpu_spec)
931	ld	r4,0(r4)
932	ld	r12,CPU_SPEC_RESTORE(r4)
933#ifdef PPC64_ELF_ABI_v1
934	ld	r12,0(r12)
935#endif
936	mtctr	r12
937	bctrl
938
939/*
940 * On POWER9, we can come here on wakeup from a cpuidle stop state.
941 * Hence restore the additional SPRs to the saved value.
942 *
943 * On POWER8, we come here only on winkle. Since winkle is used
944 * only in the case of CPU-Hotplug, we don't need to restore
945 * the additional SPRs.
946 */
947BEGIN_FTR_SECTION
948	bl 	power9_restore_additional_sprs
949END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
950hypervisor_state_restored:
951
952	mr	r12,r19
953	mtlr	r17
954	blr		/* return to pnv_powersave_wakeup */
955
956fastsleep_workaround_at_exit:
957	li	r3,1
958	li	r4,0
959	bl	opal_config_cpu_idle_state
960	b	timebase_resync
961
962/*
963 * R3 here contains the value that will be returned to the caller
964 * of power7_nap.
965 * R12 contains SRR1 for CHECK_HMI_INTERRUPT.
966 */
967.global pnv_wakeup_loss
968pnv_wakeup_loss:
969	ld	r1,PACAR1(r13)
970BEGIN_FTR_SECTION
971	CHECK_HMI_INTERRUPT
972END_FTR_SECTION_IFSET(CPU_FTR_HVMODE)
973	REST_NVGPRS(r1)
974	REST_GPR(2, r1)
975	ld	r4,PACAKMSR(r13)
976	ld	r5,_LINK(r1)
977	ld	r6,_CCR(r1)
978	addi	r1,r1,INT_FRAME_SIZE
979	mtlr	r5
980	mtcr	r6
981	mtmsrd	r4
982	blr
983
984/*
985 * R3 here contains the value that will be returned to the caller
986 * of power7_nap.
987 * R12 contains SRR1 for CHECK_HMI_INTERRUPT.
988 */
989pnv_wakeup_noloss:
990	lbz	r0,PACA_NAPSTATELOST(r13)
991	cmpwi	r0,0
992	bne	pnv_wakeup_loss
993	ld	r1,PACAR1(r13)
994BEGIN_FTR_SECTION
995	CHECK_HMI_INTERRUPT
996END_FTR_SECTION_IFSET(CPU_FTR_HVMODE)
997	ld	r4,PACAKMSR(r13)
998	ld	r5,_NIP(r1)
999	ld	r6,_CCR(r1)
1000	addi	r1,r1,INT_FRAME_SIZE
1001	mtlr	r5
1002	mtcr	r6
1003	mtmsrd	r4
1004	blr
1005