xref: /openbmc/linux/arch/powerpc/kernel/idle_book3s.S (revision b24413180f5600bcb3bb70fbed5cf186b60864bd)
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_MMCR1
116	std	r3, STOP_MMCRA(r13)
117	std	r4, STOP_MMCR1(r13)
118
119	mfspr	r3, SPRN_MMCR2
120	std	r3, STOP_MMCR2(r13)
121	blr
122
123power9_restore_additional_sprs:
124	ld	r3,_LPCR(r1)
125	ld	r4, STOP_PID(r13)
126	mtspr	SPRN_LPCR,r3
127	mtspr	SPRN_PID, r4
128
129	ld	r3, STOP_LDBAR(r13)
130	ld	r4, STOP_FSCR(r13)
131	mtspr	SPRN_LDBAR, r3
132	mtspr	SPRN_FSCR, r4
133
134	ld	r3, STOP_HFSCR(r13)
135	ld	r4, STOP_MMCRA(r13)
136	mtspr	SPRN_HFSCR, r3
137	mtspr	SPRN_MMCRA, r4
138	/* We have already restored PACA_MMCR0 */
139	ld	r3, STOP_MMCR1(r13)
140	ld	r4, STOP_MMCR2(r13)
141	mtspr	SPRN_MMCR1, r3
142	mtspr	SPRN_MMCR2, r4
143	blr
144
145/*
146 * Used by threads when the lock bit of core_idle_state is set.
147 * Threads will spin in HMT_LOW until the lock bit is cleared.
148 * r14 - pointer to core_idle_state
149 * r15 - used to load contents of core_idle_state
150 * r9  - used as a temporary variable
151 */
152
153core_idle_lock_held:
154	HMT_LOW
1553:	lwz	r15,0(r14)
156	andis.	r15,r15,PNV_CORE_IDLE_LOCK_BIT@h
157	bne	3b
158	HMT_MEDIUM
159	lwarx	r15,0,r14
160	andis.	r9,r15,PNV_CORE_IDLE_LOCK_BIT@h
161	bne-	core_idle_lock_held
162	blr
163
164/*
165 * Pass requested state in r3:
166 *	r3 - PNV_THREAD_NAP/SLEEP/WINKLE in POWER8
167 *	   - Requested PSSCR value in POWER9
168 *
169 * Address of idle handler to branch to in realmode in r4
170 */
171pnv_powersave_common:
172	/* Use r3 to pass state nap/sleep/winkle */
173	/* NAP is a state loss, we create a regs frame on the
174	 * stack, fill it up with the state we care about and
175	 * stick a pointer to it in PACAR1. We really only
176	 * need to save PC, some CR bits and the NV GPRs,
177	 * but for now an interrupt frame will do.
178	 */
179	mtctr	r4
180
181	mflr	r0
182	std	r0,16(r1)
183	stdu	r1,-INT_FRAME_SIZE(r1)
184	std	r0,_LINK(r1)
185	std	r0,_NIP(r1)
186
187	/* We haven't lost state ... yet */
188	li	r0,0
189	stb	r0,PACA_NAPSTATELOST(r13)
190
191	/* Continue saving state */
192	SAVE_GPR(2, r1)
193	SAVE_NVGPRS(r1)
194	mfcr	r5
195	std	r5,_CCR(r1)
196	std	r1,PACAR1(r13)
197
198BEGIN_FTR_SECTION
199	/*
200	 * POWER9 does not require real mode to stop, and presently does not
201	 * set hwthread_state for KVM (threads don't share MMU context), so
202	 * we can remain in virtual mode for this.
203	 */
204	bctr
205END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
206	/*
207	 * POWER8
208	 * Go to real mode to do the nap, as required by the architecture.
209	 * Also, we need to be in real mode before setting hwthread_state,
210	 * because as soon as we do that, another thread can switch
211	 * the MMU context to the guest.
212	 */
213	LOAD_REG_IMMEDIATE(r7, MSR_IDLE)
214	mtmsrd	r7,0
215	bctr
216
217/*
218 * This is the sequence required to execute idle instructions, as
219 * specified in ISA v2.07 (and earlier). MSR[IR] and MSR[DR] must be 0.
220 */
221#define IDLE_STATE_ENTER_SEQ_NORET(IDLE_INST)			\
222	/* Magic NAP/SLEEP/WINKLE mode enter sequence */	\
223	std	r0,0(r1);					\
224	ptesync;						\
225	ld	r0,0(r1);					\
226236:	cmpd	cr0,r0,r0;					\
227	bne	236b;						\
228	IDLE_INST;
229
230
231	.globl pnv_enter_arch207_idle_mode
232pnv_enter_arch207_idle_mode:
233#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
234	/* Tell KVM we're entering idle */
235	li	r4,KVM_HWTHREAD_IN_IDLE
236	/******************************************************/
237	/*  N O T E   W E L L    ! ! !    N O T E   W E L L   */
238	/* The following store to HSTATE_HWTHREAD_STATE(r13)  */
239	/* MUST occur in real mode, i.e. with the MMU off,    */
240	/* and the MMU must stay off until we clear this flag */
241	/* and test HSTATE_HWTHREAD_REQ(r13) in               */
242	/* pnv_powersave_wakeup in this file.                 */
243	/* The reason is that another thread can switch the   */
244	/* MMU to a guest context whenever this flag is set   */
245	/* to KVM_HWTHREAD_IN_IDLE, and if the MMU was on,    */
246	/* that would potentially cause this thread to start  */
247	/* executing instructions from guest memory in        */
248	/* hypervisor mode, leading to a host crash or data   */
249	/* corruption, or worse.                              */
250	/******************************************************/
251	stb	r4,HSTATE_HWTHREAD_STATE(r13)
252#endif
253	stb	r3,PACA_THREAD_IDLE_STATE(r13)
254	cmpwi	cr3,r3,PNV_THREAD_SLEEP
255	bge	cr3,2f
256	IDLE_STATE_ENTER_SEQ_NORET(PPC_NAP)
257	/* No return */
2582:
259	/* Sleep or winkle */
260	lbz	r7,PACA_THREAD_MASK(r13)
261	ld	r14,PACA_CORE_IDLE_STATE_PTR(r13)
262	li	r5,0
263	beq	cr3,3f
264	lis	r5,PNV_CORE_IDLE_WINKLE_COUNT@h
2653:
266lwarx_loop1:
267	lwarx	r15,0,r14
268
269	andis.	r9,r15,PNV_CORE_IDLE_LOCK_BIT@h
270	bnel-	core_idle_lock_held
271
272	add	r15,r15,r5			/* Add if winkle */
273	andc	r15,r15,r7			/* Clear thread bit */
274
275	andi.	r9,r15,PNV_CORE_IDLE_THREAD_BITS
276
277/*
278 * If cr0 = 0, then current thread is the last thread of the core entering
279 * sleep. Last thread needs to execute the hardware bug workaround code if
280 * required by the platform.
281 * Make the workaround call unconditionally here. The below branch call is
282 * patched out when the idle states are discovered if the platform does not
283 * require it.
284 */
285.global pnv_fastsleep_workaround_at_entry
286pnv_fastsleep_workaround_at_entry:
287	beq	fastsleep_workaround_at_entry
288
289	stwcx.	r15,0,r14
290	bne-	lwarx_loop1
291	isync
292
293common_enter: /* common code for all the threads entering sleep or winkle */
294	bgt	cr3,enter_winkle
295	IDLE_STATE_ENTER_SEQ_NORET(PPC_SLEEP)
296
297fastsleep_workaround_at_entry:
298	oris	r15,r15,PNV_CORE_IDLE_LOCK_BIT@h
299	stwcx.	r15,0,r14
300	bne-	lwarx_loop1
301	isync
302
303	/* Fast sleep workaround */
304	li	r3,1
305	li	r4,1
306	bl	opal_config_cpu_idle_state
307
308	/* Unlock */
309	xoris	r15,r15,PNV_CORE_IDLE_LOCK_BIT@h
310	lwsync
311	stw	r15,0(r14)
312	b	common_enter
313
314enter_winkle:
315	bl	save_sprs_to_stack
316
317	IDLE_STATE_ENTER_SEQ_NORET(PPC_WINKLE)
318
319/*
320 * r3 - PSSCR value corresponding to the requested stop state.
321 */
322#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
323power_enter_stop_kvm_rm:
324	/*
325	 * This is currently unused because POWER9 KVM does not have to
326	 * gather secondary threads into sibling mode, but the code is
327	 * here in case that function is required.
328	 *
329	 * Tell KVM we're entering idle.
330	 */
331	li	r4,KVM_HWTHREAD_IN_IDLE
332	/* DO THIS IN REAL MODE!  See comment above. */
333	stb	r4,HSTATE_HWTHREAD_STATE(r13)
334#endif
335power_enter_stop:
336/*
337 * Check if we are executing the lite variant with ESL=EC=0
338 */
339	andis.   r4,r3,PSSCR_EC_ESL_MASK_SHIFTED
340	clrldi   r3,r3,60 /* r3 = Bits[60:63] = Requested Level (RL) */
341	bne	 .Lhandle_esl_ec_set
342	PPC_STOP
343	li	r3,0  /* Since we didn't lose state, return 0 */
344
345	/*
346	 * pnv_wakeup_noloss() expects r12 to contain the SRR1 value so
347	 * it can determine if the wakeup reason is an HMI in
348	 * CHECK_HMI_INTERRUPT.
349	 *
350	 * However, when we wakeup with ESL=0, SRR1 will not contain the wakeup
351	 * reason, so there is no point setting r12 to SRR1.
352	 *
353	 * Further, we clear r12 here, so that we don't accidentally enter the
354	 * HMI in pnv_wakeup_noloss() if the value of r12[42:45] == WAKE_HMI.
355	 */
356	li	r12, 0
357	b 	pnv_wakeup_noloss
358
359.Lhandle_esl_ec_set:
360	/*
361	 * POWER9 DD2 can incorrectly set PMAO when waking up after a
362	 * state-loss idle. Saving and restoring MMCR0 over idle is a
363	 * workaround.
364	 */
365	mfspr	r4,SPRN_MMCR0
366	std	r4,_MMCR0(r1)
367
368/*
369 * Check if the requested state is a deep idle state.
370 */
371	LOAD_REG_ADDRBASE(r5,pnv_first_deep_stop_state)
372	ld	r4,ADDROFF(pnv_first_deep_stop_state)(r5)
373	cmpd	r3,r4
374	bge	.Lhandle_deep_stop
375	PPC_STOP	/* Does not return (system reset interrupt) */
376
377.Lhandle_deep_stop:
378/*
379 * Entering deep idle state.
380 * Clear thread bit in PACA_CORE_IDLE_STATE, save SPRs to
381 * stack and enter stop
382 */
383	lbz     r7,PACA_THREAD_MASK(r13)
384	ld      r14,PACA_CORE_IDLE_STATE_PTR(r13)
385
386lwarx_loop_stop:
387	lwarx   r15,0,r14
388	andis.	r9,r15,PNV_CORE_IDLE_LOCK_BIT@h
389	bnel-	core_idle_lock_held
390	andc    r15,r15,r7                      /* Clear thread bit */
391
392	stwcx.  r15,0,r14
393	bne-    lwarx_loop_stop
394	isync
395
396	bl	save_sprs_to_stack
397
398	PPC_STOP	/* Does not return (system reset interrupt) */
399
400/*
401 * Entered with MSR[EE]=0 and no soft-masked interrupts pending.
402 * r3 contains desired idle state (PNV_THREAD_NAP/SLEEP/WINKLE).
403 */
404_GLOBAL(power7_idle_insn)
405	/* Now check if user or arch enabled NAP mode */
406	LOAD_REG_ADDR(r4, pnv_enter_arch207_idle_mode)
407	b	pnv_powersave_common
408
409#define CHECK_HMI_INTERRUPT						\
410BEGIN_FTR_SECTION_NESTED(66);						\
411	rlwinm	r0,r12,45-31,0xf;  /* extract wake reason field (P8) */	\
412FTR_SECTION_ELSE_NESTED(66);						\
413	rlwinm	r0,r12,45-31,0xe;  /* P7 wake reason field is 3 bits */	\
414ALT_FTR_SECTION_END_NESTED_IFSET(CPU_FTR_ARCH_207S, 66);		\
415	cmpwi	r0,0xa;			/* Hypervisor maintenance ? */	\
416	bne+	20f;							\
417	/* Invoke opal call to handle hmi */				\
418	ld	r2,PACATOC(r13);					\
419	ld	r1,PACAR1(r13);						\
420	std	r3,ORIG_GPR3(r1);	/* Save original r3 */		\
421	li	r3,0;			/* NULL argument */		\
422	bl	hmi_exception_realmode;					\
423	nop;								\
424	ld	r3,ORIG_GPR3(r1);	/* Restore original r3 */	\
42520:	nop;
426
427/*
428 * Entered with MSR[EE]=0 and no soft-masked interrupts pending.
429 * r3 contains desired PSSCR register value.
430 */
431_GLOBAL(power9_idle_stop)
432	std	r3, PACA_REQ_PSSCR(r13)
433	mtspr 	SPRN_PSSCR,r3
434	LOAD_REG_ADDR(r4,power_enter_stop)
435	b	pnv_powersave_common
436	/* No return */
437
438/*
439 * On waking up from stop 0,1,2 with ESL=1 on POWER9 DD1,
440 * HSPRG0 will be set to the HSPRG0 value of one of the
441 * threads in this core. Thus the value we have in r13
442 * may not be this thread's paca pointer.
443 *
444 * Fortunately, the TIR remains invariant. Since this thread's
445 * paca pointer is recorded in all its sibling's paca, we can
446 * correctly recover this thread's paca pointer if we
447 * know the index of this thread in the core.
448 *
449 * This index can be obtained from the TIR.
450 *
451 * i.e, thread's position in the core = TIR.
452 * If this value is i, then this thread's paca is
453 * paca->thread_sibling_pacas[i].
454 */
455power9_dd1_recover_paca:
456	mfspr	r4, SPRN_TIR
457	/*
458	 * Since each entry in thread_sibling_pacas is 8 bytes
459	 * we need to left-shift by 3 bits. Thus r4 = i * 8
460	 */
461	sldi	r4, r4, 3
462	/* Get &paca->thread_sibling_pacas[0] in r5 */
463	ld	r5, PACA_SIBLING_PACA_PTRS(r13)
464	/* Load paca->thread_sibling_pacas[i] into r13 */
465	ldx	r13, r4, r5
466	SET_PACA(r13)
467	/*
468	 * Indicate that we have lost NVGPR state
469	 * which needs to be restored from the stack.
470	 */
471	li	r3, 1
472	stb	r3,PACA_NAPSTATELOST(r13)
473	blr
474
475/*
476 * Called from machine check handler for powersave wakeups.
477 * Low level machine check processing has already been done. Now just
478 * go through the wake up path to get everything in order.
479 *
480 * r3 - The original SRR1 value.
481 * Original SRR[01] have been clobbered.
482 * MSR_RI is clear.
483 */
484.global pnv_powersave_wakeup_mce
485pnv_powersave_wakeup_mce:
486	/* Set cr3 for pnv_powersave_wakeup */
487	rlwinm	r11,r3,47-31,30,31
488	cmpwi	cr3,r11,2
489
490	/*
491	 * Now put the original SRR1 with SRR1_WAKEMCE_RESVD as the wake
492	 * reason into r12, which allows reuse of the system reset wakeup
493	 * code without being mistaken for another type of wakeup.
494	 */
495	oris	r12,r3,SRR1_WAKEMCE_RESVD@h
496
497	b	pnv_powersave_wakeup
498
499#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
500kvm_start_guest_check:
501	li	r0,KVM_HWTHREAD_IN_KERNEL
502	stb	r0,HSTATE_HWTHREAD_STATE(r13)
503	/* Order setting hwthread_state vs. testing hwthread_req */
504	sync
505	lbz	r0,HSTATE_HWTHREAD_REQ(r13)
506	cmpwi	r0,0
507	beqlr
508	b	kvm_start_guest
509#endif
510
511/*
512 * Called from reset vector for powersave wakeups.
513 * cr3 - set to gt if waking up with partial/complete hypervisor state loss
514 * r12 - SRR1
515 */
516.global pnv_powersave_wakeup
517pnv_powersave_wakeup:
518	ld	r2, PACATOC(r13)
519
520BEGIN_FTR_SECTION
521BEGIN_FTR_SECTION_NESTED(70)
522	bl	power9_dd1_recover_paca
523END_FTR_SECTION_NESTED_IFSET(CPU_FTR_POWER9_DD1, 70)
524	bl	pnv_restore_hyp_resource_arch300
525FTR_SECTION_ELSE
526	bl	pnv_restore_hyp_resource_arch207
527ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_300)
528
529	li	r0,PNV_THREAD_RUNNING
530	stb	r0,PACA_THREAD_IDLE_STATE(r13)	/* Clear thread state */
531
532	mr	r3,r12
533
534#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
535BEGIN_FTR_SECTION
536	bl	kvm_start_guest_check
537END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
538#endif
539
540	/* Return SRR1 from power7_nap() */
541	blt	cr3,pnv_wakeup_noloss
542	b	pnv_wakeup_loss
543
544/*
545 * Check whether we have woken up with hypervisor state loss.
546 * If yes, restore hypervisor state and return back to link.
547 *
548 * cr3 - set to gt if waking up with partial/complete hypervisor state loss
549 */
550pnv_restore_hyp_resource_arch300:
551	/*
552	 * Workaround for POWER9, if we lost resources, the ERAT
553	 * might have been mixed up and needs flushing. We also need
554	 * to reload MMCR0 (see comment above). We also need to set
555	 * then clear bit 60 in MMCRA to ensure the PMU starts running.
556	 */
557	blt	cr3,1f
558	PPC_INVALIDATE_ERAT
559	ld	r1,PACAR1(r13)
560	mfspr	r4,SPRN_MMCRA
561	ori	r4,r4,(1 << (63-60))
562	mtspr	SPRN_MMCRA,r4
563	xori	r4,r4,(1 << (63-60))
564	mtspr	SPRN_MMCRA,r4
565	ld	r4,_MMCR0(r1)
566	mtspr	SPRN_MMCR0,r4
5671:
568	/*
569	 * POWER ISA 3. Use PSSCR to determine if we
570	 * are waking up from deep idle state
571	 */
572	LOAD_REG_ADDRBASE(r5,pnv_first_deep_stop_state)
573	ld	r4,ADDROFF(pnv_first_deep_stop_state)(r5)
574
575BEGIN_FTR_SECTION_NESTED(71)
576	/*
577	 * Assume that we are waking up from the state
578	 * same as the Requested Level (RL) in the PSSCR
579	 * which are Bits 60-63
580	 */
581	ld	r5,PACA_REQ_PSSCR(r13)
582	rldicl  r5,r5,0,60
583FTR_SECTION_ELSE_NESTED(71)
584	/*
585	 * 0-3 bits correspond to Power-Saving Level Status
586	 * which indicates the idle state we are waking up from
587	 */
588	mfspr	r5, SPRN_PSSCR
589	rldicl  r5,r5,4,60
590ALT_FTR_SECTION_END_NESTED_IFSET(CPU_FTR_POWER9_DD1, 71)
591	cmpd	cr4,r5,r4
592	bge	cr4,pnv_wakeup_tb_loss /* returns to caller */
593
594	blr	/* Waking up without hypervisor state loss. */
595
596/* Same calling convention as arch300 */
597pnv_restore_hyp_resource_arch207:
598	/*
599	 * POWER ISA 2.07 or less.
600	 * Check if we slept with sleep or winkle.
601	 */
602	lbz	r4,PACA_THREAD_IDLE_STATE(r13)
603	cmpwi	cr2,r4,PNV_THREAD_NAP
604	bgt	cr2,pnv_wakeup_tb_loss	/* Either sleep or Winkle */
605
606	/*
607	 * We fall through here if PACA_THREAD_IDLE_STATE shows we are waking
608	 * up from nap. At this stage CR3 shouldn't contains 'gt' since that
609	 * indicates we are waking with hypervisor state loss from nap.
610	 */
611	bgt	cr3,.
612
613	blr	/* Waking up without hypervisor state loss */
614
615/*
616 * Called if waking up from idle state which can cause either partial or
617 * complete hyp state loss.
618 * In POWER8, called if waking up from fastsleep or winkle
619 * In POWER9, called if waking up from stop state >= pnv_first_deep_stop_state
620 *
621 * r13 - PACA
622 * cr3 - gt if waking up with partial/complete hypervisor state loss
623 *
624 * If ISA300:
625 * cr4 - gt or eq if waking up from complete hypervisor state loss.
626 *
627 * If ISA207:
628 * r4 - PACA_THREAD_IDLE_STATE
629 */
630pnv_wakeup_tb_loss:
631	ld	r1,PACAR1(r13)
632	/*
633	 * Before entering any idle state, the NVGPRs are saved in the stack.
634	 * If there was a state loss, or PACA_NAPSTATELOST was set, then the
635	 * NVGPRs are restored. If we are here, it is likely that state is lost,
636	 * but not guaranteed -- neither ISA207 nor ISA300 tests to reach
637	 * here are the same as the test to restore NVGPRS:
638	 * PACA_THREAD_IDLE_STATE test for ISA207, PSSCR test for ISA300,
639	 * and SRR1 test for restoring NVGPRs.
640	 *
641	 * We are about to clobber NVGPRs now, so set NAPSTATELOST to
642	 * guarantee they will always be restored. This might be tightened
643	 * with careful reading of specs (particularly for ISA300) but this
644	 * is already a slow wakeup path and it's simpler to be safe.
645	 */
646	li	r0,1
647	stb	r0,PACA_NAPSTATELOST(r13)
648
649	/*
650	 *
651	 * Save SRR1 and LR in NVGPRs as they might be clobbered in
652	 * opal_call() (called in CHECK_HMI_INTERRUPT). SRR1 is required
653	 * to determine the wakeup reason if we branch to kvm_start_guest. LR
654	 * is required to return back to reset vector after hypervisor state
655	 * restore is complete.
656	 */
657	mr	r19,r12
658	mr	r18,r4
659	mflr	r17
660BEGIN_FTR_SECTION
661	CHECK_HMI_INTERRUPT
662END_FTR_SECTION_IFSET(CPU_FTR_HVMODE)
663
664	ld	r14,PACA_CORE_IDLE_STATE_PTR(r13)
665	lbz	r7,PACA_THREAD_MASK(r13)
666
667	/*
668	 * Take the core lock to synchronize against other threads.
669	 *
670	 * Lock bit is set in one of the 2 cases-
671	 * a. In the sleep/winkle enter path, the last thread is executing
672	 * fastsleep workaround code.
673	 * b. In the wake up path, another thread is executing fastsleep
674	 * workaround undo code or resyncing timebase or restoring context
675	 * In either case loop until the lock bit is cleared.
676	 */
6771:
678	lwarx	r15,0,r14
679	andis.	r9,r15,PNV_CORE_IDLE_LOCK_BIT@h
680	bnel-	core_idle_lock_held
681	oris	r15,r15,PNV_CORE_IDLE_LOCK_BIT@h
682	stwcx.	r15,0,r14
683	bne-	1b
684	isync
685
686	andi.	r9,r15,PNV_CORE_IDLE_THREAD_BITS
687	cmpwi	cr2,r9,0
688
689	/*
690	 * At this stage
691	 * cr2 - eq if first thread to wakeup in core
692	 * cr3-  gt if waking up with partial/complete hypervisor state loss
693	 * ISA300:
694	 * cr4 - gt or eq if waking up from complete hypervisor state loss.
695	 */
696
697BEGIN_FTR_SECTION
698	/*
699	 * Were we in winkle?
700	 * If yes, check if all threads were in winkle, decrement our
701	 * winkle count, set all thread winkle bits if all were in winkle.
702	 * Check if our thread has a winkle bit set, and set cr4 accordingly
703	 * (to match ISA300, above). Pseudo-code for core idle state
704	 * transitions for ISA207 is as follows (everything happens atomically
705	 * due to store conditional and/or lock bit):
706	 *
707	 * nap_idle() { }
708	 * nap_wake() { }
709	 *
710	 * sleep_idle()
711	 * {
712	 *	core_idle_state &= ~thread_in_core
713	 * }
714	 *
715	 * sleep_wake()
716	 * {
717	 *     bool first_in_core, first_in_subcore;
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	 *
725	 * winkle_idle()
726	 * {
727	 *	core_idle_state &= ~thread_in_core;
728	 *	core_idle_state += 1 << WINKLE_COUNT_SHIFT;
729	 * }
730	 *
731	 * winkle_wake()
732	 * {
733	 *     bool first_in_core, first_in_subcore, winkle_state_lost;
734	 *
735	 *     first_in_core = (core_idle_state & IDLE_THREAD_BITS) == 0;
736	 *     first_in_subcore = (core_idle_state & SUBCORE_SIBLING_MASK) == 0;
737	 *
738	 *     core_idle_state |= thread_in_core;
739	 *
740	 *     if ((core_idle_state & WINKLE_MASK) == (8 << WINKLE_COUNT_SIHFT))
741	 *         core_idle_state |= THREAD_WINKLE_BITS;
742	 *     core_idle_state -= 1 << WINKLE_COUNT_SHIFT;
743	 *
744	 *     winkle_state_lost = core_idle_state &
745	 *				(thread_in_core << WINKLE_THREAD_SHIFT);
746	 *     core_idle_state &= ~(thread_in_core << WINKLE_THREAD_SHIFT);
747	 * }
748	 *
749	 */
750	cmpwi	r18,PNV_THREAD_WINKLE
751	bne	2f
752	andis.	r9,r15,PNV_CORE_IDLE_WINKLE_COUNT_ALL_BIT@h
753	subis	r15,r15,PNV_CORE_IDLE_WINKLE_COUNT@h
754	beq	2f
755	ori	r15,r15,PNV_CORE_IDLE_THREAD_WINKLE_BITS /* all were winkle */
7562:
757	/* Shift thread bit to winkle mask, then test if this thread is set,
758	 * and remove it from the winkle bits */
759	slwi	r8,r7,8
760	and	r8,r8,r15
761	andc	r15,r15,r8
762	cmpwi	cr4,r8,1 /* cr4 will be gt if our bit is set, lt if not */
763
764	lbz	r4,PACA_SUBCORE_SIBLING_MASK(r13)
765	and	r4,r4,r15
766	cmpwi	r4,0	/* Check if first in subcore */
767
768	or	r15,r15,r7		/* Set thread bit */
769	beq	first_thread_in_subcore
770END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
771
772	or	r15,r15,r7		/* Set thread bit */
773	beq	cr2,first_thread_in_core
774
775	/* Not first thread in core or subcore to wake up */
776	b	clear_lock
777
778first_thread_in_subcore:
779	/*
780	 * If waking up from sleep, subcore state is not lost. Hence
781	 * skip subcore state restore
782	 */
783	blt	cr4,subcore_state_restored
784
785	/* Restore per-subcore state */
786	ld      r4,_SDR1(r1)
787	mtspr   SPRN_SDR1,r4
788
789	ld      r4,_RPR(r1)
790	mtspr   SPRN_RPR,r4
791	ld	r4,_AMOR(r1)
792	mtspr	SPRN_AMOR,r4
793
794subcore_state_restored:
795	/*
796	 * Check if the thread is also the first thread in the core. If not,
797	 * skip to clear_lock.
798	 */
799	bne	cr2,clear_lock
800
801first_thread_in_core:
802
803	/*
804	 * First thread in the core waking up from any state which can cause
805	 * partial or complete hypervisor state loss. It needs to
806	 * call the fastsleep workaround code if the platform requires it.
807	 * Call it unconditionally here. The below branch instruction will
808	 * be patched out if the platform does not have fastsleep or does not
809	 * require the workaround. Patching will be performed during the
810	 * discovery of idle-states.
811	 */
812.global pnv_fastsleep_workaround_at_exit
813pnv_fastsleep_workaround_at_exit:
814	b	fastsleep_workaround_at_exit
815
816timebase_resync:
817	/*
818	 * Use cr3 which indicates that we are waking up with atleast partial
819	 * hypervisor state loss to determine if TIMEBASE RESYNC is needed.
820	 */
821	ble	cr3,.Ltb_resynced
822	/* Time base re-sync */
823	bl	opal_resync_timebase;
824	/*
825	 * If waking up from sleep (POWER8), per core state
826	 * is not lost, skip to clear_lock.
827	 */
828.Ltb_resynced:
829	blt	cr4,clear_lock
830
831	/*
832	 * First thread in the core to wake up and its waking up with
833	 * complete hypervisor state loss. Restore per core hypervisor
834	 * state.
835	 */
836BEGIN_FTR_SECTION
837	ld	r4,_PTCR(r1)
838	mtspr	SPRN_PTCR,r4
839	ld	r4,_RPR(r1)
840	mtspr	SPRN_RPR,r4
841END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
842
843	ld	r4,_TSCR(r1)
844	mtspr	SPRN_TSCR,r4
845	ld	r4,_WORC(r1)
846	mtspr	SPRN_WORC,r4
847
848clear_lock:
849	xoris	r15,r15,PNV_CORE_IDLE_LOCK_BIT@h
850	lwsync
851	stw	r15,0(r14)
852
853common_exit:
854	/*
855	 * Common to all threads.
856	 *
857	 * If waking up from sleep, hypervisor state is not lost. Hence
858	 * skip hypervisor state restore.
859	 */
860	blt	cr4,hypervisor_state_restored
861
862	/* Waking up from winkle */
863
864BEGIN_MMU_FTR_SECTION
865	b	no_segments
866END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
867	/* Restore SLB  from PACA */
868	ld	r8,PACA_SLBSHADOWPTR(r13)
869
870	.rept	SLB_NUM_BOLTED
871	li	r3, SLBSHADOW_SAVEAREA
872	LDX_BE	r5, r8, r3
873	addi	r3, r3, 8
874	LDX_BE	r6, r8, r3
875	andis.	r7,r5,SLB_ESID_V@h
876	beq	1f
877	slbmte	r6,r5
8781:	addi	r8,r8,16
879	.endr
880no_segments:
881
882	/* Restore per thread state */
883
884	ld	r4,_SPURR(r1)
885	mtspr	SPRN_SPURR,r4
886	ld	r4,_PURR(r1)
887	mtspr	SPRN_PURR,r4
888	ld	r4,_DSCR(r1)
889	mtspr	SPRN_DSCR,r4
890	ld	r4,_WORT(r1)
891	mtspr	SPRN_WORT,r4
892
893	/* Call cur_cpu_spec->cpu_restore() */
894	LOAD_REG_ADDR(r4, cur_cpu_spec)
895	ld	r4,0(r4)
896	ld	r12,CPU_SPEC_RESTORE(r4)
897#ifdef PPC64_ELF_ABI_v1
898	ld	r12,0(r12)
899#endif
900	mtctr	r12
901	bctrl
902
903/*
904 * On POWER9, we can come here on wakeup from a cpuidle stop state.
905 * Hence restore the additional SPRs to the saved value.
906 *
907 * On POWER8, we come here only on winkle. Since winkle is used
908 * only in the case of CPU-Hotplug, we don't need to restore
909 * the additional SPRs.
910 */
911BEGIN_FTR_SECTION
912	bl 	power9_restore_additional_sprs
913END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
914hypervisor_state_restored:
915
916	mr	r12,r19
917	mtlr	r17
918	blr		/* return to pnv_powersave_wakeup */
919
920fastsleep_workaround_at_exit:
921	li	r3,1
922	li	r4,0
923	bl	opal_config_cpu_idle_state
924	b	timebase_resync
925
926/*
927 * R3 here contains the value that will be returned to the caller
928 * of power7_nap.
929 * R12 contains SRR1 for CHECK_HMI_INTERRUPT.
930 */
931.global pnv_wakeup_loss
932pnv_wakeup_loss:
933	ld	r1,PACAR1(r13)
934BEGIN_FTR_SECTION
935	CHECK_HMI_INTERRUPT
936END_FTR_SECTION_IFSET(CPU_FTR_HVMODE)
937	REST_NVGPRS(r1)
938	REST_GPR(2, r1)
939	ld	r4,PACAKMSR(r13)
940	ld	r5,_LINK(r1)
941	ld	r6,_CCR(r1)
942	addi	r1,r1,INT_FRAME_SIZE
943	mtlr	r5
944	mtcr	r6
945	mtmsrd	r4
946	blr
947
948/*
949 * R3 here contains the value that will be returned to the caller
950 * of power7_nap.
951 * R12 contains SRR1 for CHECK_HMI_INTERRUPT.
952 */
953pnv_wakeup_noloss:
954	lbz	r0,PACA_NAPSTATELOST(r13)
955	cmpwi	r0,0
956	bne	pnv_wakeup_loss
957	ld	r1,PACAR1(r13)
958BEGIN_FTR_SECTION
959	CHECK_HMI_INTERRUPT
960END_FTR_SECTION_IFSET(CPU_FTR_HVMODE)
961	ld	r4,PACAKMSR(r13)
962	ld	r5,_NIP(r1)
963	ld	r6,_CCR(r1)
964	addi	r1,r1,INT_FRAME_SIZE
965	mtlr	r5
966	mtcr	r6
967	mtmsrd	r4
968	blr
969