1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * PowerNV cpuidle code
4  *
5  * Copyright 2015 IBM Corp.
6  */
7 
8 #include <linux/types.h>
9 #include <linux/mm.h>
10 #include <linux/slab.h>
11 #include <linux/of.h>
12 #include <linux/device.h>
13 #include <linux/cpu.h>
14 
15 #include <asm/asm-prototypes.h>
16 #include <asm/firmware.h>
17 #include <asm/interrupt.h>
18 #include <asm/machdep.h>
19 #include <asm/opal.h>
20 #include <asm/cputhreads.h>
21 #include <asm/cpuidle.h>
22 #include <asm/code-patching.h>
23 #include <asm/smp.h>
24 #include <asm/runlatch.h>
25 #include <asm/dbell.h>
26 
27 #include "powernv.h"
28 #include "subcore.h"
29 
30 /* Power ISA 3.0 allows for stop states 0x0 - 0xF */
31 #define MAX_STOP_STATE	0xF
32 
33 #define P9_STOP_SPR_MSR 2000
34 #define P9_STOP_SPR_PSSCR      855
35 
36 static u32 supported_cpuidle_states;
37 struct pnv_idle_states_t *pnv_idle_states;
38 int nr_pnv_idle_states;
39 
40 /*
41  * The default stop state that will be used by ppc_md.power_save
42  * function on platforms that support stop instruction.
43  */
44 static u64 pnv_default_stop_val;
45 static u64 pnv_default_stop_mask;
46 static bool default_stop_found;
47 
48 /*
49  * First stop state levels when SPR and TB loss can occur.
50  */
51 static u64 pnv_first_tb_loss_level = MAX_STOP_STATE + 1;
52 static u64 deep_spr_loss_state = MAX_STOP_STATE + 1;
53 
54 /*
55  * psscr value and mask of the deepest stop idle state.
56  * Used when a cpu is offlined.
57  */
58 static u64 pnv_deepest_stop_psscr_val;
59 static u64 pnv_deepest_stop_psscr_mask;
60 static u64 pnv_deepest_stop_flag;
61 static bool deepest_stop_found;
62 
63 static unsigned long power7_offline_type;
64 
65 static int pnv_save_sprs_for_deep_states(void)
66 {
67 	int cpu;
68 	int rc;
69 
70 	/*
71 	 * hid0, hid1, hid4, hid5, hmeer and lpcr values are symmetric across
72 	 * all cpus at boot. Get these reg values of current cpu and use the
73 	 * same across all cpus.
74 	 */
75 	uint64_t lpcr_val	= mfspr(SPRN_LPCR);
76 	uint64_t hid0_val	= mfspr(SPRN_HID0);
77 	uint64_t hmeer_val	= mfspr(SPRN_HMEER);
78 	uint64_t msr_val = MSR_IDLE;
79 	uint64_t psscr_val = pnv_deepest_stop_psscr_val;
80 
81 	for_each_present_cpu(cpu) {
82 		uint64_t pir = get_hard_smp_processor_id(cpu);
83 		uint64_t hsprg0_val = (uint64_t)paca_ptrs[cpu];
84 
85 		rc = opal_slw_set_reg(pir, SPRN_HSPRG0, hsprg0_val);
86 		if (rc != 0)
87 			return rc;
88 
89 		rc = opal_slw_set_reg(pir, SPRN_LPCR, lpcr_val);
90 		if (rc != 0)
91 			return rc;
92 
93 		if (cpu_has_feature(CPU_FTR_ARCH_300)) {
94 			rc = opal_slw_set_reg(pir, P9_STOP_SPR_MSR, msr_val);
95 			if (rc)
96 				return rc;
97 
98 			rc = opal_slw_set_reg(pir,
99 					      P9_STOP_SPR_PSSCR, psscr_val);
100 
101 			if (rc)
102 				return rc;
103 		}
104 
105 		/* HIDs are per core registers */
106 		if (cpu_thread_in_core(cpu) == 0) {
107 
108 			rc = opal_slw_set_reg(pir, SPRN_HMEER, hmeer_val);
109 			if (rc != 0)
110 				return rc;
111 
112 			rc = opal_slw_set_reg(pir, SPRN_HID0, hid0_val);
113 			if (rc != 0)
114 				return rc;
115 
116 			/* Only p8 needs to set extra HID regiters */
117 			if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
118 				uint64_t hid1_val = mfspr(SPRN_HID1);
119 				uint64_t hid4_val = mfspr(SPRN_HID4);
120 				uint64_t hid5_val = mfspr(SPRN_HID5);
121 
122 				rc = opal_slw_set_reg(pir, SPRN_HID1, hid1_val);
123 				if (rc != 0)
124 					return rc;
125 
126 				rc = opal_slw_set_reg(pir, SPRN_HID4, hid4_val);
127 				if (rc != 0)
128 					return rc;
129 
130 				rc = opal_slw_set_reg(pir, SPRN_HID5, hid5_val);
131 				if (rc != 0)
132 					return rc;
133 			}
134 		}
135 	}
136 
137 	return 0;
138 }
139 
140 u32 pnv_get_supported_cpuidle_states(void)
141 {
142 	return supported_cpuidle_states;
143 }
144 EXPORT_SYMBOL_GPL(pnv_get_supported_cpuidle_states);
145 
146 static void pnv_fastsleep_workaround_apply(void *info)
147 
148 {
149 	int rc;
150 	int *err = info;
151 
152 	rc = opal_config_cpu_idle_state(OPAL_CONFIG_IDLE_FASTSLEEP,
153 					OPAL_CONFIG_IDLE_APPLY);
154 	if (rc)
155 		*err = 1;
156 }
157 
158 static bool power7_fastsleep_workaround_entry = true;
159 static bool power7_fastsleep_workaround_exit = true;
160 
161 /*
162  * Used to store fastsleep workaround state
163  * 0 - Workaround applied/undone at fastsleep entry/exit path (Default)
164  * 1 - Workaround applied once, never undone.
165  */
166 static u8 fastsleep_workaround_applyonce;
167 
168 static ssize_t show_fastsleep_workaround_applyonce(struct device *dev,
169 		struct device_attribute *attr, char *buf)
170 {
171 	return sprintf(buf, "%u\n", fastsleep_workaround_applyonce);
172 }
173 
174 static ssize_t store_fastsleep_workaround_applyonce(struct device *dev,
175 		struct device_attribute *attr, const char *buf,
176 		size_t count)
177 {
178 	cpumask_t primary_thread_mask;
179 	int err;
180 	u8 val;
181 
182 	if (kstrtou8(buf, 0, &val) || val != 1)
183 		return -EINVAL;
184 
185 	if (fastsleep_workaround_applyonce == 1)
186 		return count;
187 
188 	/*
189 	 * fastsleep_workaround_applyonce = 1 implies
190 	 * fastsleep workaround needs to be left in 'applied' state on all
191 	 * the cores. Do this by-
192 	 * 1. Disable the 'undo' workaround in fastsleep exit path
193 	 * 2. Sendi IPIs to all the cores which have at least one online thread
194 	 * 3. Disable the 'apply' workaround in fastsleep entry path
195 	 *
196 	 * There is no need to send ipi to cores which have all threads
197 	 * offlined, as last thread of the core entering fastsleep or deeper
198 	 * state would have applied workaround.
199 	 */
200 	power7_fastsleep_workaround_exit = false;
201 
202 	get_online_cpus();
203 	primary_thread_mask = cpu_online_cores_map();
204 	on_each_cpu_mask(&primary_thread_mask,
205 				pnv_fastsleep_workaround_apply,
206 				&err, 1);
207 	put_online_cpus();
208 	if (err) {
209 		pr_err("fastsleep_workaround_applyonce change failed while running pnv_fastsleep_workaround_apply");
210 		goto fail;
211 	}
212 
213 	power7_fastsleep_workaround_entry = false;
214 
215 	fastsleep_workaround_applyonce = 1;
216 
217 	return count;
218 fail:
219 	return -EIO;
220 }
221 
222 static DEVICE_ATTR(fastsleep_workaround_applyonce, 0600,
223 			show_fastsleep_workaround_applyonce,
224 			store_fastsleep_workaround_applyonce);
225 
226 static inline void atomic_start_thread_idle(void)
227 {
228 	int cpu = raw_smp_processor_id();
229 	int first = cpu_first_thread_sibling(cpu);
230 	int thread_nr = cpu_thread_in_core(cpu);
231 	unsigned long *state = &paca_ptrs[first]->idle_state;
232 
233 	clear_bit(thread_nr, state);
234 }
235 
236 static inline void atomic_stop_thread_idle(void)
237 {
238 	int cpu = raw_smp_processor_id();
239 	int first = cpu_first_thread_sibling(cpu);
240 	int thread_nr = cpu_thread_in_core(cpu);
241 	unsigned long *state = &paca_ptrs[first]->idle_state;
242 
243 	set_bit(thread_nr, state);
244 }
245 
246 static inline void atomic_lock_thread_idle(void)
247 {
248 	int cpu = raw_smp_processor_id();
249 	int first = cpu_first_thread_sibling(cpu);
250 	unsigned long *state = &paca_ptrs[first]->idle_state;
251 
252 	while (unlikely(test_and_set_bit_lock(NR_PNV_CORE_IDLE_LOCK_BIT, state)))
253 		barrier();
254 }
255 
256 static inline void atomic_unlock_and_stop_thread_idle(void)
257 {
258 	int cpu = raw_smp_processor_id();
259 	int first = cpu_first_thread_sibling(cpu);
260 	unsigned long thread = 1UL << cpu_thread_in_core(cpu);
261 	unsigned long *state = &paca_ptrs[first]->idle_state;
262 	u64 s = READ_ONCE(*state);
263 	u64 new, tmp;
264 
265 	BUG_ON(!(s & PNV_CORE_IDLE_LOCK_BIT));
266 	BUG_ON(s & thread);
267 
268 again:
269 	new = (s | thread) & ~PNV_CORE_IDLE_LOCK_BIT;
270 	tmp = cmpxchg(state, s, new);
271 	if (unlikely(tmp != s)) {
272 		s = tmp;
273 		goto again;
274 	}
275 }
276 
277 static inline void atomic_unlock_thread_idle(void)
278 {
279 	int cpu = raw_smp_processor_id();
280 	int first = cpu_first_thread_sibling(cpu);
281 	unsigned long *state = &paca_ptrs[first]->idle_state;
282 
283 	BUG_ON(!test_bit(NR_PNV_CORE_IDLE_LOCK_BIT, state));
284 	clear_bit_unlock(NR_PNV_CORE_IDLE_LOCK_BIT, state);
285 }
286 
287 /* P7 and P8 */
288 struct p7_sprs {
289 	/* per core */
290 	u64 tscr;
291 	u64 worc;
292 
293 	/* per subcore */
294 	u64 sdr1;
295 	u64 rpr;
296 
297 	/* per thread */
298 	u64 lpcr;
299 	u64 hfscr;
300 	u64 fscr;
301 	u64 purr;
302 	u64 spurr;
303 	u64 dscr;
304 	u64 wort;
305 
306 	/* per thread SPRs that get lost in shallow states */
307 	u64 amr;
308 	u64 iamr;
309 	u64 amor;
310 	u64 uamor;
311 };
312 
313 static unsigned long power7_idle_insn(unsigned long type)
314 {
315 	int cpu = raw_smp_processor_id();
316 	int first = cpu_first_thread_sibling(cpu);
317 	unsigned long *state = &paca_ptrs[first]->idle_state;
318 	unsigned long thread = 1UL << cpu_thread_in_core(cpu);
319 	unsigned long core_thread_mask = (1UL << threads_per_core) - 1;
320 	unsigned long srr1;
321 	bool full_winkle;
322 	struct p7_sprs sprs = {}; /* avoid false use-uninitialised */
323 	bool sprs_saved = false;
324 	int rc;
325 
326 	if (unlikely(type != PNV_THREAD_NAP)) {
327 		atomic_lock_thread_idle();
328 
329 		BUG_ON(!(*state & thread));
330 		*state &= ~thread;
331 
332 		if (power7_fastsleep_workaround_entry) {
333 			if ((*state & core_thread_mask) == 0) {
334 				rc = opal_config_cpu_idle_state(
335 						OPAL_CONFIG_IDLE_FASTSLEEP,
336 						OPAL_CONFIG_IDLE_APPLY);
337 				BUG_ON(rc);
338 			}
339 		}
340 
341 		if (type == PNV_THREAD_WINKLE) {
342 			sprs.tscr	= mfspr(SPRN_TSCR);
343 			sprs.worc	= mfspr(SPRN_WORC);
344 
345 			sprs.sdr1	= mfspr(SPRN_SDR1);
346 			sprs.rpr	= mfspr(SPRN_RPR);
347 
348 			sprs.lpcr	= mfspr(SPRN_LPCR);
349 			if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
350 				sprs.hfscr	= mfspr(SPRN_HFSCR);
351 				sprs.fscr	= mfspr(SPRN_FSCR);
352 			}
353 			sprs.purr	= mfspr(SPRN_PURR);
354 			sprs.spurr	= mfspr(SPRN_SPURR);
355 			sprs.dscr	= mfspr(SPRN_DSCR);
356 			sprs.wort	= mfspr(SPRN_WORT);
357 
358 			sprs_saved = true;
359 
360 			/*
361 			 * Increment winkle counter and set all winkle bits if
362 			 * all threads are winkling. This allows wakeup side to
363 			 * distinguish between fast sleep and winkle state
364 			 * loss. Fast sleep still has to resync the timebase so
365 			 * this may not be a really big win.
366 			 */
367 			*state += 1 << PNV_CORE_IDLE_WINKLE_COUNT_SHIFT;
368 			if ((*state & PNV_CORE_IDLE_WINKLE_COUNT_BITS)
369 					>> PNV_CORE_IDLE_WINKLE_COUNT_SHIFT
370 					== threads_per_core)
371 				*state |= PNV_CORE_IDLE_THREAD_WINKLE_BITS;
372 			WARN_ON((*state & PNV_CORE_IDLE_WINKLE_COUNT_BITS) == 0);
373 		}
374 
375 		atomic_unlock_thread_idle();
376 	}
377 
378 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
379 		sprs.amr	= mfspr(SPRN_AMR);
380 		sprs.iamr	= mfspr(SPRN_IAMR);
381 		sprs.amor	= mfspr(SPRN_AMOR);
382 		sprs.uamor	= mfspr(SPRN_UAMOR);
383 	}
384 
385 	local_paca->thread_idle_state = type;
386 	srr1 = isa206_idle_insn_mayloss(type);		/* go idle */
387 	local_paca->thread_idle_state = PNV_THREAD_RUNNING;
388 
389 	WARN_ON_ONCE(!srr1);
390 	WARN_ON_ONCE(mfmsr() & (MSR_IR|MSR_DR));
391 
392 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
393 		if ((srr1 & SRR1_WAKESTATE) != SRR1_WS_NOLOSS) {
394 			/*
395 			 * We don't need an isync after the mtsprs here because
396 			 * the upcoming mtmsrd is execution synchronizing.
397 			 */
398 			mtspr(SPRN_AMR,		sprs.amr);
399 			mtspr(SPRN_IAMR,	sprs.iamr);
400 			mtspr(SPRN_AMOR,	sprs.amor);
401 			mtspr(SPRN_UAMOR,	sprs.uamor);
402 		}
403 	}
404 
405 	if (unlikely((srr1 & SRR1_WAKEMASK_P8) == SRR1_WAKEHMI))
406 		hmi_exception_realmode(NULL);
407 
408 	if (likely((srr1 & SRR1_WAKESTATE) != SRR1_WS_HVLOSS)) {
409 		if (unlikely(type != PNV_THREAD_NAP)) {
410 			atomic_lock_thread_idle();
411 			if (type == PNV_THREAD_WINKLE) {
412 				WARN_ON((*state & PNV_CORE_IDLE_WINKLE_COUNT_BITS) == 0);
413 				*state -= 1 << PNV_CORE_IDLE_WINKLE_COUNT_SHIFT;
414 				*state &= ~(thread << PNV_CORE_IDLE_THREAD_WINKLE_BITS_SHIFT);
415 			}
416 			atomic_unlock_and_stop_thread_idle();
417 		}
418 		return srr1;
419 	}
420 
421 	/* HV state loss */
422 	BUG_ON(type == PNV_THREAD_NAP);
423 
424 	atomic_lock_thread_idle();
425 
426 	full_winkle = false;
427 	if (type == PNV_THREAD_WINKLE) {
428 		WARN_ON((*state & PNV_CORE_IDLE_WINKLE_COUNT_BITS) == 0);
429 		*state -= 1 << PNV_CORE_IDLE_WINKLE_COUNT_SHIFT;
430 		if (*state & (thread << PNV_CORE_IDLE_THREAD_WINKLE_BITS_SHIFT)) {
431 			*state &= ~(thread << PNV_CORE_IDLE_THREAD_WINKLE_BITS_SHIFT);
432 			full_winkle = true;
433 			BUG_ON(!sprs_saved);
434 		}
435 	}
436 
437 	WARN_ON(*state & thread);
438 
439 	if ((*state & core_thread_mask) != 0)
440 		goto core_woken;
441 
442 	/* Per-core SPRs */
443 	if (full_winkle) {
444 		mtspr(SPRN_TSCR,	sprs.tscr);
445 		mtspr(SPRN_WORC,	sprs.worc);
446 	}
447 
448 	if (power7_fastsleep_workaround_exit) {
449 		rc = opal_config_cpu_idle_state(OPAL_CONFIG_IDLE_FASTSLEEP,
450 						OPAL_CONFIG_IDLE_UNDO);
451 		BUG_ON(rc);
452 	}
453 
454 	/* TB */
455 	if (opal_resync_timebase() != OPAL_SUCCESS)
456 		BUG();
457 
458 core_woken:
459 	if (!full_winkle)
460 		goto subcore_woken;
461 
462 	if ((*state & local_paca->subcore_sibling_mask) != 0)
463 		goto subcore_woken;
464 
465 	/* Per-subcore SPRs */
466 	mtspr(SPRN_SDR1,	sprs.sdr1);
467 	mtspr(SPRN_RPR,		sprs.rpr);
468 
469 subcore_woken:
470 	/*
471 	 * isync after restoring shared SPRs and before unlocking. Unlock
472 	 * only contains hwsync which does not necessarily do the right
473 	 * thing for SPRs.
474 	 */
475 	isync();
476 	atomic_unlock_and_stop_thread_idle();
477 
478 	/* Fast sleep does not lose SPRs */
479 	if (!full_winkle)
480 		return srr1;
481 
482 	/* Per-thread SPRs */
483 	mtspr(SPRN_LPCR,	sprs.lpcr);
484 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
485 		mtspr(SPRN_HFSCR,	sprs.hfscr);
486 		mtspr(SPRN_FSCR,	sprs.fscr);
487 	}
488 	mtspr(SPRN_PURR,	sprs.purr);
489 	mtspr(SPRN_SPURR,	sprs.spurr);
490 	mtspr(SPRN_DSCR,	sprs.dscr);
491 	mtspr(SPRN_WORT,	sprs.wort);
492 
493 	mtspr(SPRN_SPRG3,	local_paca->sprg_vdso);
494 
495 	/*
496 	 * The SLB has to be restored here, but it sometimes still
497 	 * contains entries, so the __ variant must be used to prevent
498 	 * multi hits.
499 	 */
500 	__slb_restore_bolted_realmode();
501 
502 	return srr1;
503 }
504 
505 extern unsigned long idle_kvm_start_guest(unsigned long srr1);
506 
507 #ifdef CONFIG_HOTPLUG_CPU
508 static unsigned long power7_offline(void)
509 {
510 	unsigned long srr1;
511 
512 	mtmsr(MSR_IDLE);
513 
514 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
515 	/* Tell KVM we're entering idle. */
516 	/******************************************************/
517 	/*  N O T E   W E L L    ! ! !    N O T E   W E L L   */
518 	/* The following store to HSTATE_HWTHREAD_STATE(r13)  */
519 	/* MUST occur in real mode, i.e. with the MMU off,    */
520 	/* and the MMU must stay off until we clear this flag */
521 	/* and test HSTATE_HWTHREAD_REQ(r13) in               */
522 	/* pnv_powersave_wakeup in this file.                 */
523 	/* The reason is that another thread can switch the   */
524 	/* MMU to a guest context whenever this flag is set   */
525 	/* to KVM_HWTHREAD_IN_IDLE, and if the MMU was on,    */
526 	/* that would potentially cause this thread to start  */
527 	/* executing instructions from guest memory in        */
528 	/* hypervisor mode, leading to a host crash or data   */
529 	/* corruption, or worse.                              */
530 	/******************************************************/
531 	local_paca->kvm_hstate.hwthread_state = KVM_HWTHREAD_IN_IDLE;
532 #endif
533 
534 	__ppc64_runlatch_off();
535 	srr1 = power7_idle_insn(power7_offline_type);
536 	__ppc64_runlatch_on();
537 
538 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
539 	local_paca->kvm_hstate.hwthread_state = KVM_HWTHREAD_IN_KERNEL;
540 	/* Order setting hwthread_state vs. testing hwthread_req */
541 	smp_mb();
542 	if (local_paca->kvm_hstate.hwthread_req)
543 		srr1 = idle_kvm_start_guest(srr1);
544 #endif
545 
546 	mtmsr(MSR_KERNEL);
547 
548 	return srr1;
549 }
550 #endif
551 
552 void power7_idle_type(unsigned long type)
553 {
554 	unsigned long srr1;
555 
556 	if (!prep_irq_for_idle_irqsoff())
557 		return;
558 
559 	mtmsr(MSR_IDLE);
560 	__ppc64_runlatch_off();
561 	srr1 = power7_idle_insn(type);
562 	__ppc64_runlatch_on();
563 	mtmsr(MSR_KERNEL);
564 
565 	fini_irq_for_idle_irqsoff();
566 	irq_set_pending_from_srr1(srr1);
567 }
568 
569 static void power7_idle(void)
570 {
571 	if (!powersave_nap)
572 		return;
573 
574 	power7_idle_type(PNV_THREAD_NAP);
575 }
576 
577 struct p9_sprs {
578 	/* per core */
579 	u64 ptcr;
580 	u64 rpr;
581 	u64 tscr;
582 	u64 ldbar;
583 
584 	/* per thread */
585 	u64 lpcr;
586 	u64 hfscr;
587 	u64 fscr;
588 	u64 pid;
589 	u64 purr;
590 	u64 spurr;
591 	u64 dscr;
592 	u64 wort;
593 	u64 ciabr;
594 
595 	u64 mmcra;
596 	u32 mmcr0;
597 	u32 mmcr1;
598 	u64 mmcr2;
599 
600 	/* per thread SPRs that get lost in shallow states */
601 	u64 amr;
602 	u64 iamr;
603 	u64 amor;
604 	u64 uamor;
605 };
606 
607 static unsigned long power9_idle_stop(unsigned long psscr)
608 {
609 	int cpu = raw_smp_processor_id();
610 	int first = cpu_first_thread_sibling(cpu);
611 	unsigned long *state = &paca_ptrs[first]->idle_state;
612 	unsigned long core_thread_mask = (1UL << threads_per_core) - 1;
613 	unsigned long srr1;
614 	unsigned long pls;
615 	unsigned long mmcr0 = 0;
616 	unsigned long mmcra = 0;
617 	struct p9_sprs sprs = {}; /* avoid false used-uninitialised */
618 	bool sprs_saved = false;
619 
620 	if (!(psscr & (PSSCR_EC|PSSCR_ESL))) {
621 		/* EC=ESL=0 case */
622 
623 		/*
624 		 * Wake synchronously. SRESET via xscom may still cause
625 		 * a 0x100 powersave wakeup with SRR1 reason!
626 		 */
627 		srr1 = isa300_idle_stop_noloss(psscr);		/* go idle */
628 		if (likely(!srr1))
629 			return 0;
630 
631 		/*
632 		 * Registers not saved, can't recover!
633 		 * This would be a hardware bug
634 		 */
635 		BUG_ON((srr1 & SRR1_WAKESTATE) != SRR1_WS_NOLOSS);
636 
637 		goto out;
638 	}
639 
640 	/* EC=ESL=1 case */
641 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
642 	if (cpu_has_feature(CPU_FTR_P9_TM_XER_SO_BUG)) {
643 		local_paca->requested_psscr = psscr;
644 		/* order setting requested_psscr vs testing dont_stop */
645 		smp_mb();
646 		if (atomic_read(&local_paca->dont_stop)) {
647 			local_paca->requested_psscr = 0;
648 			return 0;
649 		}
650 	}
651 #endif
652 
653 	if (!cpu_has_feature(CPU_FTR_POWER9_DD2_1)) {
654 		 /*
655 		  * POWER9 DD2 can incorrectly set PMAO when waking up
656 		  * after a state-loss idle. Saving and restoring MMCR0
657 		  * over idle is a workaround.
658 		  */
659 		mmcr0		= mfspr(SPRN_MMCR0);
660 	}
661 
662 	if ((psscr & PSSCR_RL_MASK) >= deep_spr_loss_state) {
663 		sprs.lpcr	= mfspr(SPRN_LPCR);
664 		sprs.hfscr	= mfspr(SPRN_HFSCR);
665 		sprs.fscr	= mfspr(SPRN_FSCR);
666 		sprs.pid	= mfspr(SPRN_PID);
667 		sprs.purr	= mfspr(SPRN_PURR);
668 		sprs.spurr	= mfspr(SPRN_SPURR);
669 		sprs.dscr	= mfspr(SPRN_DSCR);
670 		sprs.wort	= mfspr(SPRN_WORT);
671 		sprs.ciabr	= mfspr(SPRN_CIABR);
672 
673 		sprs.mmcra	= mfspr(SPRN_MMCRA);
674 		sprs.mmcr0	= mfspr(SPRN_MMCR0);
675 		sprs.mmcr1	= mfspr(SPRN_MMCR1);
676 		sprs.mmcr2	= mfspr(SPRN_MMCR2);
677 
678 		sprs.ptcr	= mfspr(SPRN_PTCR);
679 		sprs.rpr	= mfspr(SPRN_RPR);
680 		sprs.tscr	= mfspr(SPRN_TSCR);
681 		if (!firmware_has_feature(FW_FEATURE_ULTRAVISOR))
682 			sprs.ldbar = mfspr(SPRN_LDBAR);
683 
684 		sprs_saved = true;
685 
686 		atomic_start_thread_idle();
687 	}
688 
689 	sprs.amr	= mfspr(SPRN_AMR);
690 	sprs.iamr	= mfspr(SPRN_IAMR);
691 	sprs.amor	= mfspr(SPRN_AMOR);
692 	sprs.uamor	= mfspr(SPRN_UAMOR);
693 
694 	srr1 = isa300_idle_stop_mayloss(psscr);		/* go idle */
695 
696 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
697 	local_paca->requested_psscr = 0;
698 #endif
699 
700 	psscr = mfspr(SPRN_PSSCR);
701 
702 	WARN_ON_ONCE(!srr1);
703 	WARN_ON_ONCE(mfmsr() & (MSR_IR|MSR_DR));
704 
705 	if ((srr1 & SRR1_WAKESTATE) != SRR1_WS_NOLOSS) {
706 		/*
707 		 * We don't need an isync after the mtsprs here because the
708 		 * upcoming mtmsrd is execution synchronizing.
709 		 */
710 		mtspr(SPRN_AMR,		sprs.amr);
711 		mtspr(SPRN_IAMR,	sprs.iamr);
712 		mtspr(SPRN_AMOR,	sprs.amor);
713 		mtspr(SPRN_UAMOR,	sprs.uamor);
714 
715 		/*
716 		 * Workaround for POWER9 DD2.0, if we lost resources, the ERAT
717 		 * might have been corrupted and needs flushing. We also need
718 		 * to reload MMCR0 (see mmcr0 comment above).
719 		 */
720 		if (!cpu_has_feature(CPU_FTR_POWER9_DD2_1)) {
721 			asm volatile(PPC_ISA_3_0_INVALIDATE_ERAT);
722 			mtspr(SPRN_MMCR0, mmcr0);
723 		}
724 
725 		/*
726 		 * DD2.2 and earlier need to set then clear bit 60 in MMCRA
727 		 * to ensure the PMU starts running.
728 		 */
729 		mmcra = mfspr(SPRN_MMCRA);
730 		mmcra |= PPC_BIT(60);
731 		mtspr(SPRN_MMCRA, mmcra);
732 		mmcra &= ~PPC_BIT(60);
733 		mtspr(SPRN_MMCRA, mmcra);
734 	}
735 
736 	if (unlikely((srr1 & SRR1_WAKEMASK_P8) == SRR1_WAKEHMI))
737 		hmi_exception_realmode(NULL);
738 
739 	/*
740 	 * On POWER9, SRR1 bits do not match exactly as expected.
741 	 * SRR1_WS_GPRLOSS (10b) can also result in SPR loss, so
742 	 * just always test PSSCR for SPR/TB state loss.
743 	 */
744 	pls = (psscr & PSSCR_PLS) >> PSSCR_PLS_SHIFT;
745 	if (likely(pls < deep_spr_loss_state)) {
746 		if (sprs_saved)
747 			atomic_stop_thread_idle();
748 		goto out;
749 	}
750 
751 	/* HV state loss */
752 	BUG_ON(!sprs_saved);
753 
754 	atomic_lock_thread_idle();
755 
756 	if ((*state & core_thread_mask) != 0)
757 		goto core_woken;
758 
759 	/* Per-core SPRs */
760 	mtspr(SPRN_PTCR,	sprs.ptcr);
761 	mtspr(SPRN_RPR,		sprs.rpr);
762 	mtspr(SPRN_TSCR,	sprs.tscr);
763 
764 	if (pls >= pnv_first_tb_loss_level) {
765 		/* TB loss */
766 		if (opal_resync_timebase() != OPAL_SUCCESS)
767 			BUG();
768 	}
769 
770 	/*
771 	 * isync after restoring shared SPRs and before unlocking. Unlock
772 	 * only contains hwsync which does not necessarily do the right
773 	 * thing for SPRs.
774 	 */
775 	isync();
776 
777 core_woken:
778 	atomic_unlock_and_stop_thread_idle();
779 
780 	/* Per-thread SPRs */
781 	mtspr(SPRN_LPCR,	sprs.lpcr);
782 	mtspr(SPRN_HFSCR,	sprs.hfscr);
783 	mtspr(SPRN_FSCR,	sprs.fscr);
784 	mtspr(SPRN_PID,		sprs.pid);
785 	mtspr(SPRN_PURR,	sprs.purr);
786 	mtspr(SPRN_SPURR,	sprs.spurr);
787 	mtspr(SPRN_DSCR,	sprs.dscr);
788 	mtspr(SPRN_WORT,	sprs.wort);
789 	mtspr(SPRN_CIABR,	sprs.ciabr);
790 
791 	mtspr(SPRN_MMCRA,	sprs.mmcra);
792 	mtspr(SPRN_MMCR0,	sprs.mmcr0);
793 	mtspr(SPRN_MMCR1,	sprs.mmcr1);
794 	mtspr(SPRN_MMCR2,	sprs.mmcr2);
795 	if (!firmware_has_feature(FW_FEATURE_ULTRAVISOR))
796 		mtspr(SPRN_LDBAR, sprs.ldbar);
797 
798 	mtspr(SPRN_SPRG3,	local_paca->sprg_vdso);
799 
800 	if (!radix_enabled())
801 		__slb_restore_bolted_realmode();
802 
803 out:
804 	mtmsr(MSR_KERNEL);
805 
806 	return srr1;
807 }
808 
809 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
810 /*
811  * This is used in working around bugs in thread reconfiguration
812  * on POWER9 (at least up to Nimbus DD2.2) relating to transactional
813  * memory and the way that XER[SO] is checkpointed.
814  * This function forces the core into SMT4 in order by asking
815  * all other threads not to stop, and sending a message to any
816  * that are in a stop state.
817  * Must be called with preemption disabled.
818  */
819 void pnv_power9_force_smt4_catch(void)
820 {
821 	int cpu, cpu0, thr;
822 	int awake_threads = 1;		/* this thread is awake */
823 	int poke_threads = 0;
824 	int need_awake = threads_per_core;
825 
826 	cpu = smp_processor_id();
827 	cpu0 = cpu & ~(threads_per_core - 1);
828 	for (thr = 0; thr < threads_per_core; ++thr) {
829 		if (cpu != cpu0 + thr)
830 			atomic_inc(&paca_ptrs[cpu0+thr]->dont_stop);
831 	}
832 	/* order setting dont_stop vs testing requested_psscr */
833 	smp_mb();
834 	for (thr = 0; thr < threads_per_core; ++thr) {
835 		if (!paca_ptrs[cpu0+thr]->requested_psscr)
836 			++awake_threads;
837 		else
838 			poke_threads |= (1 << thr);
839 	}
840 
841 	/* If at least 3 threads are awake, the core is in SMT4 already */
842 	if (awake_threads < need_awake) {
843 		/* We have to wake some threads; we'll use msgsnd */
844 		for (thr = 0; thr < threads_per_core; ++thr) {
845 			if (poke_threads & (1 << thr)) {
846 				ppc_msgsnd_sync();
847 				ppc_msgsnd(PPC_DBELL_MSGTYPE, 0,
848 					   paca_ptrs[cpu0+thr]->hw_cpu_id);
849 			}
850 		}
851 		/* now spin until at least 3 threads are awake */
852 		do {
853 			for (thr = 0; thr < threads_per_core; ++thr) {
854 				if ((poke_threads & (1 << thr)) &&
855 				    !paca_ptrs[cpu0+thr]->requested_psscr) {
856 					++awake_threads;
857 					poke_threads &= ~(1 << thr);
858 				}
859 			}
860 		} while (awake_threads < need_awake);
861 	}
862 }
863 EXPORT_SYMBOL_GPL(pnv_power9_force_smt4_catch);
864 
865 void pnv_power9_force_smt4_release(void)
866 {
867 	int cpu, cpu0, thr;
868 
869 	cpu = smp_processor_id();
870 	cpu0 = cpu & ~(threads_per_core - 1);
871 
872 	/* clear all the dont_stop flags */
873 	for (thr = 0; thr < threads_per_core; ++thr) {
874 		if (cpu != cpu0 + thr)
875 			atomic_dec(&paca_ptrs[cpu0+thr]->dont_stop);
876 	}
877 }
878 EXPORT_SYMBOL_GPL(pnv_power9_force_smt4_release);
879 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
880 
881 struct p10_sprs {
882 	/*
883 	 * SPRs that get lost in shallow states:
884 	 *
885 	 * P10 loses CR, LR, CTR, FPSCR, VSCR, XER, TAR, SPRG2, and HSPRG1
886 	 * isa300 idle routines restore CR, LR.
887 	 * CTR is volatile
888 	 * idle thread doesn't use FP or VEC
889 	 * kernel doesn't use TAR
890 	 * HSPRG1 is only live in HV interrupt entry
891 	 * SPRG2 is only live in KVM guests, KVM handles it.
892 	 */
893 };
894 
895 static unsigned long power10_idle_stop(unsigned long psscr)
896 {
897 	int cpu = raw_smp_processor_id();
898 	int first = cpu_first_thread_sibling(cpu);
899 	unsigned long *state = &paca_ptrs[first]->idle_state;
900 	unsigned long core_thread_mask = (1UL << threads_per_core) - 1;
901 	unsigned long srr1;
902 	unsigned long pls;
903 //	struct p10_sprs sprs = {}; /* avoid false used-uninitialised */
904 	bool sprs_saved = false;
905 
906 	if (!(psscr & (PSSCR_EC|PSSCR_ESL))) {
907 		/* EC=ESL=0 case */
908 
909 		/*
910 		 * Wake synchronously. SRESET via xscom may still cause
911 		 * a 0x100 powersave wakeup with SRR1 reason!
912 		 */
913 		srr1 = isa300_idle_stop_noloss(psscr);		/* go idle */
914 		if (likely(!srr1))
915 			return 0;
916 
917 		/*
918 		 * Registers not saved, can't recover!
919 		 * This would be a hardware bug
920 		 */
921 		BUG_ON((srr1 & SRR1_WAKESTATE) != SRR1_WS_NOLOSS);
922 
923 		goto out;
924 	}
925 
926 	/* EC=ESL=1 case */
927 	if ((psscr & PSSCR_RL_MASK) >= deep_spr_loss_state) {
928 		/* XXX: save SPRs for deep state loss here. */
929 
930 		sprs_saved = true;
931 
932 		atomic_start_thread_idle();
933 	}
934 
935 	srr1 = isa300_idle_stop_mayloss(psscr);		/* go idle */
936 
937 	psscr = mfspr(SPRN_PSSCR);
938 
939 	WARN_ON_ONCE(!srr1);
940 	WARN_ON_ONCE(mfmsr() & (MSR_IR|MSR_DR));
941 
942 	if (unlikely((srr1 & SRR1_WAKEMASK_P8) == SRR1_WAKEHMI))
943 		hmi_exception_realmode(NULL);
944 
945 	/*
946 	 * On POWER10, SRR1 bits do not match exactly as expected.
947 	 * SRR1_WS_GPRLOSS (10b) can also result in SPR loss, so
948 	 * just always test PSSCR for SPR/TB state loss.
949 	 */
950 	pls = (psscr & PSSCR_PLS) >> PSSCR_PLS_SHIFT;
951 	if (likely(pls < deep_spr_loss_state)) {
952 		if (sprs_saved)
953 			atomic_stop_thread_idle();
954 		goto out;
955 	}
956 
957 	/* HV state loss */
958 	BUG_ON(!sprs_saved);
959 
960 	atomic_lock_thread_idle();
961 
962 	if ((*state & core_thread_mask) != 0)
963 		goto core_woken;
964 
965 	/* XXX: restore per-core SPRs here */
966 
967 	if (pls >= pnv_first_tb_loss_level) {
968 		/* TB loss */
969 		if (opal_resync_timebase() != OPAL_SUCCESS)
970 			BUG();
971 	}
972 
973 	/*
974 	 * isync after restoring shared SPRs and before unlocking. Unlock
975 	 * only contains hwsync which does not necessarily do the right
976 	 * thing for SPRs.
977 	 */
978 	isync();
979 
980 core_woken:
981 	atomic_unlock_and_stop_thread_idle();
982 
983 	/* XXX: restore per-thread SPRs here */
984 
985 	if (!radix_enabled())
986 		__slb_restore_bolted_realmode();
987 
988 out:
989 	mtmsr(MSR_KERNEL);
990 
991 	return srr1;
992 }
993 
994 #ifdef CONFIG_HOTPLUG_CPU
995 static unsigned long arch300_offline_stop(unsigned long psscr)
996 {
997 	unsigned long srr1;
998 
999 	if (cpu_has_feature(CPU_FTR_ARCH_31))
1000 		srr1 = power10_idle_stop(psscr);
1001 	else
1002 		srr1 = power9_idle_stop(psscr);
1003 
1004 	return srr1;
1005 }
1006 #endif
1007 
1008 void arch300_idle_type(unsigned long stop_psscr_val,
1009 				      unsigned long stop_psscr_mask)
1010 {
1011 	unsigned long psscr;
1012 	unsigned long srr1;
1013 
1014 	if (!prep_irq_for_idle_irqsoff())
1015 		return;
1016 
1017 	psscr = mfspr(SPRN_PSSCR);
1018 	psscr = (psscr & ~stop_psscr_mask) | stop_psscr_val;
1019 
1020 	__ppc64_runlatch_off();
1021 	if (cpu_has_feature(CPU_FTR_ARCH_31))
1022 		srr1 = power10_idle_stop(psscr);
1023 	else
1024 		srr1 = power9_idle_stop(psscr);
1025 	__ppc64_runlatch_on();
1026 
1027 	fini_irq_for_idle_irqsoff();
1028 
1029 	irq_set_pending_from_srr1(srr1);
1030 }
1031 
1032 /*
1033  * Used for ppc_md.power_save which needs a function with no parameters
1034  */
1035 static void arch300_idle(void)
1036 {
1037 	arch300_idle_type(pnv_default_stop_val, pnv_default_stop_mask);
1038 }
1039 
1040 #ifdef CONFIG_HOTPLUG_CPU
1041 
1042 void pnv_program_cpu_hotplug_lpcr(unsigned int cpu, u64 lpcr_val)
1043 {
1044 	u64 pir = get_hard_smp_processor_id(cpu);
1045 
1046 	mtspr(SPRN_LPCR, lpcr_val);
1047 
1048 	/*
1049 	 * Program the LPCR via stop-api only if the deepest stop state
1050 	 * can lose hypervisor context.
1051 	 */
1052 	if (supported_cpuidle_states & OPAL_PM_LOSE_FULL_CONTEXT)
1053 		opal_slw_set_reg(pir, SPRN_LPCR, lpcr_val);
1054 }
1055 
1056 /*
1057  * pnv_cpu_offline: A function that puts the CPU into the deepest
1058  * available platform idle state on a CPU-Offline.
1059  * interrupts hard disabled and no lazy irq pending.
1060  */
1061 unsigned long pnv_cpu_offline(unsigned int cpu)
1062 {
1063 	unsigned long srr1;
1064 
1065 	__ppc64_runlatch_off();
1066 
1067 	if (cpu_has_feature(CPU_FTR_ARCH_300) && deepest_stop_found) {
1068 		unsigned long psscr;
1069 
1070 		psscr = mfspr(SPRN_PSSCR);
1071 		psscr = (psscr & ~pnv_deepest_stop_psscr_mask) |
1072 						pnv_deepest_stop_psscr_val;
1073 		srr1 = arch300_offline_stop(psscr);
1074 	} else if (cpu_has_feature(CPU_FTR_ARCH_206) && power7_offline_type) {
1075 		srr1 = power7_offline();
1076 	} else {
1077 		/* This is the fallback method. We emulate snooze */
1078 		while (!generic_check_cpu_restart(cpu)) {
1079 			HMT_low();
1080 			HMT_very_low();
1081 		}
1082 		srr1 = 0;
1083 		HMT_medium();
1084 	}
1085 
1086 	__ppc64_runlatch_on();
1087 
1088 	return srr1;
1089 }
1090 #endif
1091 
1092 /*
1093  * Power ISA 3.0 idle initialization.
1094  *
1095  * POWER ISA 3.0 defines a new SPR Processor stop Status and Control
1096  * Register (PSSCR) to control idle behavior.
1097  *
1098  * PSSCR layout:
1099  * ----------------------------------------------------------
1100  * | PLS | /// | SD | ESL | EC | PSLL | /// | TR | MTL | RL |
1101  * ----------------------------------------------------------
1102  * 0      4     41   42    43   44     48    54   56    60
1103  *
1104  * PSSCR key fields:
1105  *	Bits 0:3  - Power-Saving Level Status (PLS). This field indicates the
1106  *	lowest power-saving state the thread entered since stop instruction was
1107  *	last executed.
1108  *
1109  *	Bit 41 - Status Disable(SD)
1110  *	0 - Shows PLS entries
1111  *	1 - PLS entries are all 0
1112  *
1113  *	Bit 42 - Enable State Loss
1114  *	0 - No state is lost irrespective of other fields
1115  *	1 - Allows state loss
1116  *
1117  *	Bit 43 - Exit Criterion
1118  *	0 - Exit from power-save mode on any interrupt
1119  *	1 - Exit from power-save mode controlled by LPCR's PECE bits
1120  *
1121  *	Bits 44:47 - Power-Saving Level Limit
1122  *	This limits the power-saving level that can be entered into.
1123  *
1124  *	Bits 60:63 - Requested Level
1125  *	Used to specify which power-saving level must be entered on executing
1126  *	stop instruction
1127  */
1128 
1129 int validate_psscr_val_mask(u64 *psscr_val, u64 *psscr_mask, u32 flags)
1130 {
1131 	int err = 0;
1132 
1133 	/*
1134 	 * psscr_mask == 0xf indicates an older firmware.
1135 	 * Set remaining fields of psscr to the default values.
1136 	 * See NOTE above definition of PSSCR_HV_DEFAULT_VAL
1137 	 */
1138 	if (*psscr_mask == 0xf) {
1139 		*psscr_val = *psscr_val | PSSCR_HV_DEFAULT_VAL;
1140 		*psscr_mask = PSSCR_HV_DEFAULT_MASK;
1141 		return err;
1142 	}
1143 
1144 	/*
1145 	 * New firmware is expected to set the psscr_val bits correctly.
1146 	 * Validate that the following invariants are correctly maintained by
1147 	 * the new firmware.
1148 	 * - ESL bit value matches the EC bit value.
1149 	 * - ESL bit is set for all the deep stop states.
1150 	 */
1151 	if (GET_PSSCR_ESL(*psscr_val) != GET_PSSCR_EC(*psscr_val)) {
1152 		err = ERR_EC_ESL_MISMATCH;
1153 	} else if ((flags & OPAL_PM_LOSE_FULL_CONTEXT) &&
1154 		GET_PSSCR_ESL(*psscr_val) == 0) {
1155 		err = ERR_DEEP_STATE_ESL_MISMATCH;
1156 	}
1157 
1158 	return err;
1159 }
1160 
1161 /*
1162  * pnv_arch300_idle_init: Initializes the default idle state, first
1163  *                        deep idle state and deepest idle state on
1164  *                        ISA 3.0 CPUs.
1165  *
1166  * @np: /ibm,opal/power-mgt device node
1167  * @flags: cpu-idle-state-flags array
1168  * @dt_idle_states: Number of idle state entries
1169  * Returns 0 on success
1170  */
1171 static void __init pnv_arch300_idle_init(void)
1172 {
1173 	u64 max_residency_ns = 0;
1174 	int i;
1175 
1176 	/* stop is not really architected, we only have p9,p10 drivers */
1177 	if (!pvr_version_is(PVR_POWER10) && !pvr_version_is(PVR_POWER9))
1178 		return;
1179 
1180 	/*
1181 	 * pnv_deepest_stop_{val,mask} should be set to values corresponding to
1182 	 * the deepest stop state.
1183 	 *
1184 	 * pnv_default_stop_{val,mask} should be set to values corresponding to
1185 	 * the deepest loss-less (OPAL_PM_STOP_INST_FAST) stop state.
1186 	 */
1187 	pnv_first_tb_loss_level = MAX_STOP_STATE + 1;
1188 	deep_spr_loss_state = MAX_STOP_STATE + 1;
1189 	for (i = 0; i < nr_pnv_idle_states; i++) {
1190 		int err;
1191 		struct pnv_idle_states_t *state = &pnv_idle_states[i];
1192 		u64 psscr_rl = state->psscr_val & PSSCR_RL_MASK;
1193 
1194 		/* No deep loss driver implemented for POWER10 yet */
1195 		if (pvr_version_is(PVR_POWER10) &&
1196 				state->flags & (OPAL_PM_TIMEBASE_STOP|OPAL_PM_LOSE_FULL_CONTEXT))
1197 			continue;
1198 
1199 		if ((state->flags & OPAL_PM_TIMEBASE_STOP) &&
1200 		     (pnv_first_tb_loss_level > psscr_rl))
1201 			pnv_first_tb_loss_level = psscr_rl;
1202 
1203 		if ((state->flags & OPAL_PM_LOSE_FULL_CONTEXT) &&
1204 		     (deep_spr_loss_state > psscr_rl))
1205 			deep_spr_loss_state = psscr_rl;
1206 
1207 		/*
1208 		 * The idle code does not deal with TB loss occurring
1209 		 * in a shallower state than SPR loss, so force it to
1210 		 * behave like SPRs are lost if TB is lost. POWER9 would
1211 		 * never encouter this, but a POWER8 core would if it
1212 		 * implemented the stop instruction. So this is for forward
1213 		 * compatibility.
1214 		 */
1215 		if ((state->flags & OPAL_PM_TIMEBASE_STOP) &&
1216 		     (deep_spr_loss_state > psscr_rl))
1217 			deep_spr_loss_state = psscr_rl;
1218 
1219 		err = validate_psscr_val_mask(&state->psscr_val,
1220 					      &state->psscr_mask,
1221 					      state->flags);
1222 		if (err) {
1223 			report_invalid_psscr_val(state->psscr_val, err);
1224 			continue;
1225 		}
1226 
1227 		state->valid = true;
1228 
1229 		if (max_residency_ns < state->residency_ns) {
1230 			max_residency_ns = state->residency_ns;
1231 			pnv_deepest_stop_psscr_val = state->psscr_val;
1232 			pnv_deepest_stop_psscr_mask = state->psscr_mask;
1233 			pnv_deepest_stop_flag = state->flags;
1234 			deepest_stop_found = true;
1235 		}
1236 
1237 		if (!default_stop_found &&
1238 		    (state->flags & OPAL_PM_STOP_INST_FAST)) {
1239 			pnv_default_stop_val = state->psscr_val;
1240 			pnv_default_stop_mask = state->psscr_mask;
1241 			default_stop_found = true;
1242 			WARN_ON(state->flags & OPAL_PM_LOSE_FULL_CONTEXT);
1243 		}
1244 	}
1245 
1246 	if (unlikely(!default_stop_found)) {
1247 		pr_warn("cpuidle-powernv: No suitable default stop state found. Disabling platform idle.\n");
1248 	} else {
1249 		ppc_md.power_save = arch300_idle;
1250 		pr_info("cpuidle-powernv: Default stop: psscr = 0x%016llx,mask=0x%016llx\n",
1251 			pnv_default_stop_val, pnv_default_stop_mask);
1252 	}
1253 
1254 	if (unlikely(!deepest_stop_found)) {
1255 		pr_warn("cpuidle-powernv: No suitable stop state for CPU-Hotplug. Offlined CPUs will busy wait");
1256 	} else {
1257 		pr_info("cpuidle-powernv: Deepest stop: psscr = 0x%016llx,mask=0x%016llx\n",
1258 			pnv_deepest_stop_psscr_val,
1259 			pnv_deepest_stop_psscr_mask);
1260 	}
1261 
1262 	pr_info("cpuidle-powernv: First stop level that may lose SPRs = 0x%llx\n",
1263 		deep_spr_loss_state);
1264 
1265 	pr_info("cpuidle-powernv: First stop level that may lose timebase = 0x%llx\n",
1266 		pnv_first_tb_loss_level);
1267 }
1268 
1269 static void __init pnv_disable_deep_states(void)
1270 {
1271 	/*
1272 	 * The stop-api is unable to restore hypervisor
1273 	 * resources on wakeup from platform idle states which
1274 	 * lose full context. So disable such states.
1275 	 */
1276 	supported_cpuidle_states &= ~OPAL_PM_LOSE_FULL_CONTEXT;
1277 	pr_warn("cpuidle-powernv: Disabling idle states that lose full context\n");
1278 	pr_warn("cpuidle-powernv: Idle power-savings, CPU-Hotplug affected\n");
1279 
1280 	if (cpu_has_feature(CPU_FTR_ARCH_300) &&
1281 	    (pnv_deepest_stop_flag & OPAL_PM_LOSE_FULL_CONTEXT)) {
1282 		/*
1283 		 * Use the default stop state for CPU-Hotplug
1284 		 * if available.
1285 		 */
1286 		if (default_stop_found) {
1287 			pnv_deepest_stop_psscr_val = pnv_default_stop_val;
1288 			pnv_deepest_stop_psscr_mask = pnv_default_stop_mask;
1289 			pr_warn("cpuidle-powernv: Offlined CPUs will stop with psscr = 0x%016llx\n",
1290 				pnv_deepest_stop_psscr_val);
1291 		} else { /* Fallback to snooze loop for CPU-Hotplug */
1292 			deepest_stop_found = false;
1293 			pr_warn("cpuidle-powernv: Offlined CPUs will busy wait\n");
1294 		}
1295 	}
1296 }
1297 
1298 /*
1299  * Probe device tree for supported idle states
1300  */
1301 static void __init pnv_probe_idle_states(void)
1302 {
1303 	int i;
1304 
1305 	if (nr_pnv_idle_states < 0) {
1306 		pr_warn("cpuidle-powernv: no idle states found in the DT\n");
1307 		return;
1308 	}
1309 
1310 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1311 		pnv_arch300_idle_init();
1312 
1313 	for (i = 0; i < nr_pnv_idle_states; i++)
1314 		supported_cpuidle_states |= pnv_idle_states[i].flags;
1315 }
1316 
1317 /*
1318  * This function parses device-tree and populates all the information
1319  * into pnv_idle_states structure. It also sets up nr_pnv_idle_states
1320  * which is the number of cpuidle states discovered through device-tree.
1321  */
1322 
1323 static int pnv_parse_cpuidle_dt(void)
1324 {
1325 	struct device_node *np;
1326 	int nr_idle_states, i;
1327 	int rc = 0;
1328 	u32 *temp_u32;
1329 	u64 *temp_u64;
1330 	const char **temp_string;
1331 
1332 	np = of_find_node_by_path("/ibm,opal/power-mgt");
1333 	if (!np) {
1334 		pr_warn("opal: PowerMgmt Node not found\n");
1335 		return -ENODEV;
1336 	}
1337 	nr_idle_states = of_property_count_u32_elems(np,
1338 						"ibm,cpu-idle-state-flags");
1339 
1340 	pnv_idle_states = kcalloc(nr_idle_states, sizeof(*pnv_idle_states),
1341 				  GFP_KERNEL);
1342 	temp_u32 = kcalloc(nr_idle_states, sizeof(u32),  GFP_KERNEL);
1343 	temp_u64 = kcalloc(nr_idle_states, sizeof(u64),  GFP_KERNEL);
1344 	temp_string = kcalloc(nr_idle_states, sizeof(char *),  GFP_KERNEL);
1345 
1346 	if (!(pnv_idle_states && temp_u32 && temp_u64 && temp_string)) {
1347 		pr_err("Could not allocate memory for dt parsing\n");
1348 		rc = -ENOMEM;
1349 		goto out;
1350 	}
1351 
1352 	/* Read flags */
1353 	if (of_property_read_u32_array(np, "ibm,cpu-idle-state-flags",
1354 				       temp_u32, nr_idle_states)) {
1355 		pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-flags in DT\n");
1356 		rc = -EINVAL;
1357 		goto out;
1358 	}
1359 	for (i = 0; i < nr_idle_states; i++)
1360 		pnv_idle_states[i].flags = temp_u32[i];
1361 
1362 	/* Read latencies */
1363 	if (of_property_read_u32_array(np, "ibm,cpu-idle-state-latencies-ns",
1364 				       temp_u32, nr_idle_states)) {
1365 		pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-latencies-ns in DT\n");
1366 		rc = -EINVAL;
1367 		goto out;
1368 	}
1369 	for (i = 0; i < nr_idle_states; i++)
1370 		pnv_idle_states[i].latency_ns = temp_u32[i];
1371 
1372 	/* Read residencies */
1373 	if (of_property_read_u32_array(np, "ibm,cpu-idle-state-residency-ns",
1374 				       temp_u32, nr_idle_states)) {
1375 		pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-residency-ns in DT\n");
1376 		rc = -EINVAL;
1377 		goto out;
1378 	}
1379 	for (i = 0; i < nr_idle_states; i++)
1380 		pnv_idle_states[i].residency_ns = temp_u32[i];
1381 
1382 	/* For power9 and later */
1383 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1384 		/* Read pm_crtl_val */
1385 		if (of_property_read_u64_array(np, "ibm,cpu-idle-state-psscr",
1386 					       temp_u64, nr_idle_states)) {
1387 			pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-psscr in DT\n");
1388 			rc = -EINVAL;
1389 			goto out;
1390 		}
1391 		for (i = 0; i < nr_idle_states; i++)
1392 			pnv_idle_states[i].psscr_val = temp_u64[i];
1393 
1394 		/* Read pm_crtl_mask */
1395 		if (of_property_read_u64_array(np, "ibm,cpu-idle-state-psscr-mask",
1396 					       temp_u64, nr_idle_states)) {
1397 			pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-psscr-mask in DT\n");
1398 			rc = -EINVAL;
1399 			goto out;
1400 		}
1401 		for (i = 0; i < nr_idle_states; i++)
1402 			pnv_idle_states[i].psscr_mask = temp_u64[i];
1403 	}
1404 
1405 	/*
1406 	 * power8 specific properties ibm,cpu-idle-state-pmicr-mask and
1407 	 * ibm,cpu-idle-state-pmicr-val were never used and there is no
1408 	 * plan to use it in near future. Hence, not parsing these properties
1409 	 */
1410 
1411 	if (of_property_read_string_array(np, "ibm,cpu-idle-state-names",
1412 					  temp_string, nr_idle_states) < 0) {
1413 		pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-names in DT\n");
1414 		rc = -EINVAL;
1415 		goto out;
1416 	}
1417 	for (i = 0; i < nr_idle_states; i++)
1418 		strlcpy(pnv_idle_states[i].name, temp_string[i],
1419 			PNV_IDLE_NAME_LEN);
1420 	nr_pnv_idle_states = nr_idle_states;
1421 	rc = 0;
1422 out:
1423 	kfree(temp_u32);
1424 	kfree(temp_u64);
1425 	kfree(temp_string);
1426 	return rc;
1427 }
1428 
1429 static int __init pnv_init_idle_states(void)
1430 {
1431 	int cpu;
1432 	int rc = 0;
1433 
1434 	/* Set up PACA fields */
1435 	for_each_present_cpu(cpu) {
1436 		struct paca_struct *p = paca_ptrs[cpu];
1437 
1438 		p->idle_state = 0;
1439 		if (cpu == cpu_first_thread_sibling(cpu))
1440 			p->idle_state = (1 << threads_per_core) - 1;
1441 
1442 		if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1443 			/* P7/P8 nap */
1444 			p->thread_idle_state = PNV_THREAD_RUNNING;
1445 		} else if (pvr_version_is(PVR_POWER9)) {
1446 			/* P9 stop workarounds */
1447 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1448 			p->requested_psscr = 0;
1449 			atomic_set(&p->dont_stop, 0);
1450 #endif
1451 		}
1452 	}
1453 
1454 	/* In case we error out nr_pnv_idle_states will be zero */
1455 	nr_pnv_idle_states = 0;
1456 	supported_cpuidle_states = 0;
1457 
1458 	if (cpuidle_disable != IDLE_NO_OVERRIDE)
1459 		goto out;
1460 	rc = pnv_parse_cpuidle_dt();
1461 	if (rc)
1462 		return rc;
1463 	pnv_probe_idle_states();
1464 
1465 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1466 		if (!(supported_cpuidle_states & OPAL_PM_SLEEP_ENABLED_ER1)) {
1467 			power7_fastsleep_workaround_entry = false;
1468 			power7_fastsleep_workaround_exit = false;
1469 		} else {
1470 			/*
1471 			 * OPAL_PM_SLEEP_ENABLED_ER1 is set. It indicates that
1472 			 * workaround is needed to use fastsleep. Provide sysfs
1473 			 * control to choose how this workaround has to be
1474 			 * applied.
1475 			 */
1476 			device_create_file(cpu_subsys.dev_root,
1477 				&dev_attr_fastsleep_workaround_applyonce);
1478 		}
1479 
1480 		update_subcore_sibling_mask();
1481 
1482 		if (supported_cpuidle_states & OPAL_PM_NAP_ENABLED) {
1483 			ppc_md.power_save = power7_idle;
1484 			power7_offline_type = PNV_THREAD_NAP;
1485 		}
1486 
1487 		if ((supported_cpuidle_states & OPAL_PM_WINKLE_ENABLED) &&
1488 			   (supported_cpuidle_states & OPAL_PM_LOSE_FULL_CONTEXT))
1489 			power7_offline_type = PNV_THREAD_WINKLE;
1490 		else if ((supported_cpuidle_states & OPAL_PM_SLEEP_ENABLED) ||
1491 			   (supported_cpuidle_states & OPAL_PM_SLEEP_ENABLED_ER1))
1492 			power7_offline_type = PNV_THREAD_SLEEP;
1493 	}
1494 
1495 	if (supported_cpuidle_states & OPAL_PM_LOSE_FULL_CONTEXT) {
1496 		if (pnv_save_sprs_for_deep_states())
1497 			pnv_disable_deep_states();
1498 	}
1499 
1500 out:
1501 	return 0;
1502 }
1503 machine_subsys_initcall(powernv, pnv_init_idle_states);
1504