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/machdep.h>
18 #include <asm/opal.h>
19 #include <asm/cputhreads.h>
20 #include <asm/cpuidle.h>
21 #include <asm/code-patching.h>
22 #include <asm/smp.h>
23 #include <asm/runlatch.h>
24 #include <asm/dbell.h>
25 
26 #include "powernv.h"
27 #include "subcore.h"
28 
29 /* Power ISA 3.0 allows for stop states 0x0 - 0xF */
30 #define MAX_STOP_STATE	0xF
31 
32 #define P9_STOP_SPR_MSR 2000
33 #define P9_STOP_SPR_PSSCR      855
34 
35 static u32 supported_cpuidle_states;
36 struct pnv_idle_states_t *pnv_idle_states;
37 int nr_pnv_idle_states;
38 
39 /*
40  * The default stop state that will be used by ppc_md.power_save
41  * function on platforms that support stop instruction.
42  */
43 static u64 pnv_default_stop_val;
44 static u64 pnv_default_stop_mask;
45 static bool default_stop_found;
46 
47 /*
48  * First stop state levels when SPR and TB loss can occur.
49  */
50 static u64 pnv_first_tb_loss_level = MAX_STOP_STATE + 1;
51 static u64 deep_spr_loss_state = MAX_STOP_STATE + 1;
52 
53 /*
54  * psscr value and mask of the deepest stop idle state.
55  * Used when a cpu is offlined.
56  */
57 static u64 pnv_deepest_stop_psscr_val;
58 static u64 pnv_deepest_stop_psscr_mask;
59 static u64 pnv_deepest_stop_flag;
60 static bool deepest_stop_found;
61 
62 static unsigned long power7_offline_type;
63 
64 static int pnv_save_sprs_for_deep_states(void)
65 {
66 	int cpu;
67 	int rc;
68 
69 	/*
70 	 * hid0, hid1, hid4, hid5, hmeer and lpcr values are symmetric across
71 	 * all cpus at boot. Get these reg values of current cpu and use the
72 	 * same across all cpus.
73 	 */
74 	uint64_t lpcr_val	= mfspr(SPRN_LPCR);
75 	uint64_t hid0_val	= mfspr(SPRN_HID0);
76 	uint64_t hmeer_val	= mfspr(SPRN_HMEER);
77 	uint64_t msr_val = MSR_IDLE;
78 	uint64_t psscr_val = pnv_deepest_stop_psscr_val;
79 
80 	for_each_present_cpu(cpu) {
81 		uint64_t pir = get_hard_smp_processor_id(cpu);
82 		uint64_t hsprg0_val = (uint64_t)paca_ptrs[cpu];
83 
84 		rc = opal_slw_set_reg(pir, SPRN_HSPRG0, hsprg0_val);
85 		if (rc != 0)
86 			return rc;
87 
88 		rc = opal_slw_set_reg(pir, SPRN_LPCR, lpcr_val);
89 		if (rc != 0)
90 			return rc;
91 
92 		if (cpu_has_feature(CPU_FTR_ARCH_300)) {
93 			rc = opal_slw_set_reg(pir, P9_STOP_SPR_MSR, msr_val);
94 			if (rc)
95 				return rc;
96 
97 			rc = opal_slw_set_reg(pir,
98 					      P9_STOP_SPR_PSSCR, psscr_val);
99 
100 			if (rc)
101 				return rc;
102 		}
103 
104 		/* HIDs are per core registers */
105 		if (cpu_thread_in_core(cpu) == 0) {
106 
107 			rc = opal_slw_set_reg(pir, SPRN_HMEER, hmeer_val);
108 			if (rc != 0)
109 				return rc;
110 
111 			rc = opal_slw_set_reg(pir, SPRN_HID0, hid0_val);
112 			if (rc != 0)
113 				return rc;
114 
115 			/* Only p8 needs to set extra HID regiters */
116 			if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
117 				uint64_t hid1_val = mfspr(SPRN_HID1);
118 				uint64_t hid4_val = mfspr(SPRN_HID4);
119 				uint64_t hid5_val = mfspr(SPRN_HID5);
120 
121 				rc = opal_slw_set_reg(pir, SPRN_HID1, hid1_val);
122 				if (rc != 0)
123 					return rc;
124 
125 				rc = opal_slw_set_reg(pir, SPRN_HID4, hid4_val);
126 				if (rc != 0)
127 					return rc;
128 
129 				rc = opal_slw_set_reg(pir, SPRN_HID5, hid5_val);
130 				if (rc != 0)
131 					return rc;
132 			}
133 		}
134 	}
135 
136 	return 0;
137 }
138 
139 u32 pnv_get_supported_cpuidle_states(void)
140 {
141 	return supported_cpuidle_states;
142 }
143 EXPORT_SYMBOL_GPL(pnv_get_supported_cpuidle_states);
144 
145 static void pnv_fastsleep_workaround_apply(void *info)
146 
147 {
148 	int rc;
149 	int *err = info;
150 
151 	rc = opal_config_cpu_idle_state(OPAL_CONFIG_IDLE_FASTSLEEP,
152 					OPAL_CONFIG_IDLE_APPLY);
153 	if (rc)
154 		*err = 1;
155 }
156 
157 static bool power7_fastsleep_workaround_entry = true;
158 static bool power7_fastsleep_workaround_exit = true;
159 
160 /*
161  * Used to store fastsleep workaround state
162  * 0 - Workaround applied/undone at fastsleep entry/exit path (Default)
163  * 1 - Workaround applied once, never undone.
164  */
165 static u8 fastsleep_workaround_applyonce;
166 
167 static ssize_t show_fastsleep_workaround_applyonce(struct device *dev,
168 		struct device_attribute *attr, char *buf)
169 {
170 	return sprintf(buf, "%u\n", fastsleep_workaround_applyonce);
171 }
172 
173 static ssize_t store_fastsleep_workaround_applyonce(struct device *dev,
174 		struct device_attribute *attr, const char *buf,
175 		size_t count)
176 {
177 	cpumask_t primary_thread_mask;
178 	int err;
179 	u8 val;
180 
181 	if (kstrtou8(buf, 0, &val) || val != 1)
182 		return -EINVAL;
183 
184 	if (fastsleep_workaround_applyonce == 1)
185 		return count;
186 
187 	/*
188 	 * fastsleep_workaround_applyonce = 1 implies
189 	 * fastsleep workaround needs to be left in 'applied' state on all
190 	 * the cores. Do this by-
191 	 * 1. Disable the 'undo' workaround in fastsleep exit path
192 	 * 2. Sendi IPIs to all the cores which have at least one online thread
193 	 * 3. Disable the 'apply' workaround in fastsleep entry path
194 	 *
195 	 * There is no need to send ipi to cores which have all threads
196 	 * offlined, as last thread of the core entering fastsleep or deeper
197 	 * state would have applied workaround.
198 	 */
199 	power7_fastsleep_workaround_exit = false;
200 
201 	get_online_cpus();
202 	primary_thread_mask = cpu_online_cores_map();
203 	on_each_cpu_mask(&primary_thread_mask,
204 				pnv_fastsleep_workaround_apply,
205 				&err, 1);
206 	put_online_cpus();
207 	if (err) {
208 		pr_err("fastsleep_workaround_applyonce change failed while running pnv_fastsleep_workaround_apply");
209 		goto fail;
210 	}
211 
212 	power7_fastsleep_workaround_entry = false;
213 
214 	fastsleep_workaround_applyonce = 1;
215 
216 	return count;
217 fail:
218 	return -EIO;
219 }
220 
221 static DEVICE_ATTR(fastsleep_workaround_applyonce, 0600,
222 			show_fastsleep_workaround_applyonce,
223 			store_fastsleep_workaround_applyonce);
224 
225 static inline void atomic_start_thread_idle(void)
226 {
227 	int cpu = raw_smp_processor_id();
228 	int first = cpu_first_thread_sibling(cpu);
229 	int thread_nr = cpu_thread_in_core(cpu);
230 	unsigned long *state = &paca_ptrs[first]->idle_state;
231 
232 	clear_bit(thread_nr, state);
233 }
234 
235 static inline void atomic_stop_thread_idle(void)
236 {
237 	int cpu = raw_smp_processor_id();
238 	int first = cpu_first_thread_sibling(cpu);
239 	int thread_nr = cpu_thread_in_core(cpu);
240 	unsigned long *state = &paca_ptrs[first]->idle_state;
241 
242 	set_bit(thread_nr, state);
243 }
244 
245 static inline void atomic_lock_thread_idle(void)
246 {
247 	int cpu = raw_smp_processor_id();
248 	int first = cpu_first_thread_sibling(cpu);
249 	unsigned long *state = &paca_ptrs[first]->idle_state;
250 
251 	while (unlikely(test_and_set_bit_lock(NR_PNV_CORE_IDLE_LOCK_BIT, state)))
252 		barrier();
253 }
254 
255 static inline void atomic_unlock_and_stop_thread_idle(void)
256 {
257 	int cpu = raw_smp_processor_id();
258 	int first = cpu_first_thread_sibling(cpu);
259 	unsigned long thread = 1UL << cpu_thread_in_core(cpu);
260 	unsigned long *state = &paca_ptrs[first]->idle_state;
261 	u64 s = READ_ONCE(*state);
262 	u64 new, tmp;
263 
264 	BUG_ON(!(s & PNV_CORE_IDLE_LOCK_BIT));
265 	BUG_ON(s & thread);
266 
267 again:
268 	new = (s | thread) & ~PNV_CORE_IDLE_LOCK_BIT;
269 	tmp = cmpxchg(state, s, new);
270 	if (unlikely(tmp != s)) {
271 		s = tmp;
272 		goto again;
273 	}
274 }
275 
276 static inline void atomic_unlock_thread_idle(void)
277 {
278 	int cpu = raw_smp_processor_id();
279 	int first = cpu_first_thread_sibling(cpu);
280 	unsigned long *state = &paca_ptrs[first]->idle_state;
281 
282 	BUG_ON(!test_bit(NR_PNV_CORE_IDLE_LOCK_BIT, state));
283 	clear_bit_unlock(NR_PNV_CORE_IDLE_LOCK_BIT, state);
284 }
285 
286 /* P7 and P8 */
287 struct p7_sprs {
288 	/* per core */
289 	u64 tscr;
290 	u64 worc;
291 
292 	/* per subcore */
293 	u64 sdr1;
294 	u64 rpr;
295 
296 	/* per thread */
297 	u64 lpcr;
298 	u64 hfscr;
299 	u64 fscr;
300 	u64 purr;
301 	u64 spurr;
302 	u64 dscr;
303 	u64 wort;
304 
305 	/* per thread SPRs that get lost in shallow states */
306 	u64 amr;
307 	u64 iamr;
308 	u64 amor;
309 	u64 uamor;
310 };
311 
312 static unsigned long power7_idle_insn(unsigned long type)
313 {
314 	int cpu = raw_smp_processor_id();
315 	int first = cpu_first_thread_sibling(cpu);
316 	unsigned long *state = &paca_ptrs[first]->idle_state;
317 	unsigned long thread = 1UL << cpu_thread_in_core(cpu);
318 	unsigned long core_thread_mask = (1UL << threads_per_core) - 1;
319 	unsigned long srr1;
320 	bool full_winkle;
321 	struct p7_sprs sprs = {}; /* avoid false use-uninitialised */
322 	bool sprs_saved = false;
323 	int rc;
324 
325 	if (unlikely(type != PNV_THREAD_NAP)) {
326 		atomic_lock_thread_idle();
327 
328 		BUG_ON(!(*state & thread));
329 		*state &= ~thread;
330 
331 		if (power7_fastsleep_workaround_entry) {
332 			if ((*state & core_thread_mask) == 0) {
333 				rc = opal_config_cpu_idle_state(
334 						OPAL_CONFIG_IDLE_FASTSLEEP,
335 						OPAL_CONFIG_IDLE_APPLY);
336 				BUG_ON(rc);
337 			}
338 		}
339 
340 		if (type == PNV_THREAD_WINKLE) {
341 			sprs.tscr	= mfspr(SPRN_TSCR);
342 			sprs.worc	= mfspr(SPRN_WORC);
343 
344 			sprs.sdr1	= mfspr(SPRN_SDR1);
345 			sprs.rpr	= mfspr(SPRN_RPR);
346 
347 			sprs.lpcr	= mfspr(SPRN_LPCR);
348 			if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
349 				sprs.hfscr	= mfspr(SPRN_HFSCR);
350 				sprs.fscr	= mfspr(SPRN_FSCR);
351 			}
352 			sprs.purr	= mfspr(SPRN_PURR);
353 			sprs.spurr	= mfspr(SPRN_SPURR);
354 			sprs.dscr	= mfspr(SPRN_DSCR);
355 			sprs.wort	= mfspr(SPRN_WORT);
356 
357 			sprs_saved = true;
358 
359 			/*
360 			 * Increment winkle counter and set all winkle bits if
361 			 * all threads are winkling. This allows wakeup side to
362 			 * distinguish between fast sleep and winkle state
363 			 * loss. Fast sleep still has to resync the timebase so
364 			 * this may not be a really big win.
365 			 */
366 			*state += 1 << PNV_CORE_IDLE_WINKLE_COUNT_SHIFT;
367 			if ((*state & PNV_CORE_IDLE_WINKLE_COUNT_BITS)
368 					>> PNV_CORE_IDLE_WINKLE_COUNT_SHIFT
369 					== threads_per_core)
370 				*state |= PNV_CORE_IDLE_THREAD_WINKLE_BITS;
371 			WARN_ON((*state & PNV_CORE_IDLE_WINKLE_COUNT_BITS) == 0);
372 		}
373 
374 		atomic_unlock_thread_idle();
375 	}
376 
377 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
378 		sprs.amr	= mfspr(SPRN_AMR);
379 		sprs.iamr	= mfspr(SPRN_IAMR);
380 		sprs.amor	= mfspr(SPRN_AMOR);
381 		sprs.uamor	= mfspr(SPRN_UAMOR);
382 	}
383 
384 	local_paca->thread_idle_state = type;
385 	srr1 = isa206_idle_insn_mayloss(type);		/* go idle */
386 	local_paca->thread_idle_state = PNV_THREAD_RUNNING;
387 
388 	WARN_ON_ONCE(!srr1);
389 	WARN_ON_ONCE(mfmsr() & (MSR_IR|MSR_DR));
390 
391 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
392 		if ((srr1 & SRR1_WAKESTATE) != SRR1_WS_NOLOSS) {
393 			/*
394 			 * We don't need an isync after the mtsprs here because
395 			 * the upcoming mtmsrd is execution synchronizing.
396 			 */
397 			mtspr(SPRN_AMR,		sprs.amr);
398 			mtspr(SPRN_IAMR,	sprs.iamr);
399 			mtspr(SPRN_AMOR,	sprs.amor);
400 			mtspr(SPRN_UAMOR,	sprs.uamor);
401 		}
402 	}
403 
404 	if (unlikely((srr1 & SRR1_WAKEMASK_P8) == SRR1_WAKEHMI))
405 		hmi_exception_realmode(NULL);
406 
407 	if (likely((srr1 & SRR1_WAKESTATE) != SRR1_WS_HVLOSS)) {
408 		if (unlikely(type != PNV_THREAD_NAP)) {
409 			atomic_lock_thread_idle();
410 			if (type == PNV_THREAD_WINKLE) {
411 				WARN_ON((*state & PNV_CORE_IDLE_WINKLE_COUNT_BITS) == 0);
412 				*state -= 1 << PNV_CORE_IDLE_WINKLE_COUNT_SHIFT;
413 				*state &= ~(thread << PNV_CORE_IDLE_THREAD_WINKLE_BITS_SHIFT);
414 			}
415 			atomic_unlock_and_stop_thread_idle();
416 		}
417 		return srr1;
418 	}
419 
420 	/* HV state loss */
421 	BUG_ON(type == PNV_THREAD_NAP);
422 
423 	atomic_lock_thread_idle();
424 
425 	full_winkle = false;
426 	if (type == PNV_THREAD_WINKLE) {
427 		WARN_ON((*state & PNV_CORE_IDLE_WINKLE_COUNT_BITS) == 0);
428 		*state -= 1 << PNV_CORE_IDLE_WINKLE_COUNT_SHIFT;
429 		if (*state & (thread << PNV_CORE_IDLE_THREAD_WINKLE_BITS_SHIFT)) {
430 			*state &= ~(thread << PNV_CORE_IDLE_THREAD_WINKLE_BITS_SHIFT);
431 			full_winkle = true;
432 			BUG_ON(!sprs_saved);
433 		}
434 	}
435 
436 	WARN_ON(*state & thread);
437 
438 	if ((*state & core_thread_mask) != 0)
439 		goto core_woken;
440 
441 	/* Per-core SPRs */
442 	if (full_winkle) {
443 		mtspr(SPRN_TSCR,	sprs.tscr);
444 		mtspr(SPRN_WORC,	sprs.worc);
445 	}
446 
447 	if (power7_fastsleep_workaround_exit) {
448 		rc = opal_config_cpu_idle_state(OPAL_CONFIG_IDLE_FASTSLEEP,
449 						OPAL_CONFIG_IDLE_UNDO);
450 		BUG_ON(rc);
451 	}
452 
453 	/* TB */
454 	if (opal_resync_timebase() != OPAL_SUCCESS)
455 		BUG();
456 
457 core_woken:
458 	if (!full_winkle)
459 		goto subcore_woken;
460 
461 	if ((*state & local_paca->subcore_sibling_mask) != 0)
462 		goto subcore_woken;
463 
464 	/* Per-subcore SPRs */
465 	mtspr(SPRN_SDR1,	sprs.sdr1);
466 	mtspr(SPRN_RPR,		sprs.rpr);
467 
468 subcore_woken:
469 	/*
470 	 * isync after restoring shared SPRs and before unlocking. Unlock
471 	 * only contains hwsync which does not necessarily do the right
472 	 * thing for SPRs.
473 	 */
474 	isync();
475 	atomic_unlock_and_stop_thread_idle();
476 
477 	/* Fast sleep does not lose SPRs */
478 	if (!full_winkle)
479 		return srr1;
480 
481 	/* Per-thread SPRs */
482 	mtspr(SPRN_LPCR,	sprs.lpcr);
483 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
484 		mtspr(SPRN_HFSCR,	sprs.hfscr);
485 		mtspr(SPRN_FSCR,	sprs.fscr);
486 	}
487 	mtspr(SPRN_PURR,	sprs.purr);
488 	mtspr(SPRN_SPURR,	sprs.spurr);
489 	mtspr(SPRN_DSCR,	sprs.dscr);
490 	mtspr(SPRN_WORT,	sprs.wort);
491 
492 	mtspr(SPRN_SPRG3,	local_paca->sprg_vdso);
493 
494 	/*
495 	 * The SLB has to be restored here, but it sometimes still
496 	 * contains entries, so the __ variant must be used to prevent
497 	 * multi hits.
498 	 */
499 	__slb_restore_bolted_realmode();
500 
501 	return srr1;
502 }
503 
504 extern unsigned long idle_kvm_start_guest(unsigned long srr1);
505 
506 #ifdef CONFIG_HOTPLUG_CPU
507 static unsigned long power7_offline(void)
508 {
509 	unsigned long srr1;
510 
511 	mtmsr(MSR_IDLE);
512 
513 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
514 	/* Tell KVM we're entering idle. */
515 	/******************************************************/
516 	/*  N O T E   W E L L    ! ! !    N O T E   W E L L   */
517 	/* The following store to HSTATE_HWTHREAD_STATE(r13)  */
518 	/* MUST occur in real mode, i.e. with the MMU off,    */
519 	/* and the MMU must stay off until we clear this flag */
520 	/* and test HSTATE_HWTHREAD_REQ(r13) in               */
521 	/* pnv_powersave_wakeup in this file.                 */
522 	/* The reason is that another thread can switch the   */
523 	/* MMU to a guest context whenever this flag is set   */
524 	/* to KVM_HWTHREAD_IN_IDLE, and if the MMU was on,    */
525 	/* that would potentially cause this thread to start  */
526 	/* executing instructions from guest memory in        */
527 	/* hypervisor mode, leading to a host crash or data   */
528 	/* corruption, or worse.                              */
529 	/******************************************************/
530 	local_paca->kvm_hstate.hwthread_state = KVM_HWTHREAD_IN_IDLE;
531 #endif
532 
533 	__ppc64_runlatch_off();
534 	srr1 = power7_idle_insn(power7_offline_type);
535 	__ppc64_runlatch_on();
536 
537 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
538 	local_paca->kvm_hstate.hwthread_state = KVM_HWTHREAD_IN_KERNEL;
539 	/* Order setting hwthread_state vs. testing hwthread_req */
540 	smp_mb();
541 	if (local_paca->kvm_hstate.hwthread_req)
542 		srr1 = idle_kvm_start_guest(srr1);
543 #endif
544 
545 	mtmsr(MSR_KERNEL);
546 
547 	return srr1;
548 }
549 #endif
550 
551 void power7_idle_type(unsigned long type)
552 {
553 	unsigned long srr1;
554 
555 	if (!prep_irq_for_idle_irqsoff())
556 		return;
557 
558 	mtmsr(MSR_IDLE);
559 	__ppc64_runlatch_off();
560 	srr1 = power7_idle_insn(type);
561 	__ppc64_runlatch_on();
562 	mtmsr(MSR_KERNEL);
563 
564 	fini_irq_for_idle_irqsoff();
565 	irq_set_pending_from_srr1(srr1);
566 }
567 
568 void power7_idle(void)
569 {
570 	if (!powersave_nap)
571 		return;
572 
573 	power7_idle_type(PNV_THREAD_NAP);
574 }
575 
576 struct p9_sprs {
577 	/* per core */
578 	u64 ptcr;
579 	u64 rpr;
580 	u64 tscr;
581 	u64 ldbar;
582 
583 	/* per thread */
584 	u64 lpcr;
585 	u64 hfscr;
586 	u64 fscr;
587 	u64 pid;
588 	u64 purr;
589 	u64 spurr;
590 	u64 dscr;
591 	u64 wort;
592 
593 	u64 mmcra;
594 	u32 mmcr0;
595 	u32 mmcr1;
596 	u64 mmcr2;
597 
598 	/* per thread SPRs that get lost in shallow states */
599 	u64 amr;
600 	u64 iamr;
601 	u64 amor;
602 	u64 uamor;
603 };
604 
605 static unsigned long power9_idle_stop(unsigned long psscr, bool mmu_on)
606 {
607 	int cpu = raw_smp_processor_id();
608 	int first = cpu_first_thread_sibling(cpu);
609 	unsigned long *state = &paca_ptrs[first]->idle_state;
610 	unsigned long core_thread_mask = (1UL << threads_per_core) - 1;
611 	unsigned long srr1;
612 	unsigned long pls;
613 	unsigned long mmcr0 = 0;
614 	unsigned long mmcra = 0;
615 	struct p9_sprs sprs = {}; /* avoid false used-uninitialised */
616 	bool sprs_saved = false;
617 
618 	if (!(psscr & (PSSCR_EC|PSSCR_ESL))) {
619 		/* EC=ESL=0 case */
620 
621 		BUG_ON(!mmu_on);
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 (cpu_has_feature(CPU_FTR_ARCH_31)) {
663 		/*
664 		 * POWER10 uses MMCRA (BHRBRD) as BHRB disable bit.
665 		 * If the user hasn't asked for the BHRB to be
666 		 * written, the value of MMCRA[BHRBRD] is 1.
667 		 * On wakeup from stop, MMCRA[BHRBD] will be 0,
668 		 * since it is previleged resource and will be lost.
669 		 * Thus, if we do not save and restore the MMCRA[BHRBD],
670 		 * hardware will be needlessly writing to the BHRB
671 		 * in problem mode.
672 		 */
673 		mmcra		= mfspr(SPRN_MMCRA);
674 	}
675 
676 	if ((psscr & PSSCR_RL_MASK) >= deep_spr_loss_state) {
677 		sprs.lpcr	= mfspr(SPRN_LPCR);
678 		sprs.hfscr	= mfspr(SPRN_HFSCR);
679 		sprs.fscr	= mfspr(SPRN_FSCR);
680 		sprs.pid	= mfspr(SPRN_PID);
681 		sprs.purr	= mfspr(SPRN_PURR);
682 		sprs.spurr	= mfspr(SPRN_SPURR);
683 		sprs.dscr	= mfspr(SPRN_DSCR);
684 		sprs.wort	= mfspr(SPRN_WORT);
685 
686 		sprs.mmcra	= mfspr(SPRN_MMCRA);
687 		sprs.mmcr0	= mfspr(SPRN_MMCR0);
688 		sprs.mmcr1	= mfspr(SPRN_MMCR1);
689 		sprs.mmcr2	= mfspr(SPRN_MMCR2);
690 
691 		sprs.ptcr	= mfspr(SPRN_PTCR);
692 		sprs.rpr	= mfspr(SPRN_RPR);
693 		sprs.tscr	= mfspr(SPRN_TSCR);
694 		if (!firmware_has_feature(FW_FEATURE_ULTRAVISOR))
695 			sprs.ldbar = mfspr(SPRN_LDBAR);
696 
697 		sprs_saved = true;
698 
699 		atomic_start_thread_idle();
700 	}
701 
702 	sprs.amr	= mfspr(SPRN_AMR);
703 	sprs.iamr	= mfspr(SPRN_IAMR);
704 	sprs.amor	= mfspr(SPRN_AMOR);
705 	sprs.uamor	= mfspr(SPRN_UAMOR);
706 
707 	srr1 = isa300_idle_stop_mayloss(psscr);		/* go idle */
708 
709 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
710 	local_paca->requested_psscr = 0;
711 #endif
712 
713 	psscr = mfspr(SPRN_PSSCR);
714 
715 	WARN_ON_ONCE(!srr1);
716 	WARN_ON_ONCE(mfmsr() & (MSR_IR|MSR_DR));
717 
718 	if ((srr1 & SRR1_WAKESTATE) != SRR1_WS_NOLOSS) {
719 		/*
720 		 * We don't need an isync after the mtsprs here because the
721 		 * upcoming mtmsrd is execution synchronizing.
722 		 */
723 		mtspr(SPRN_AMR,		sprs.amr);
724 		mtspr(SPRN_IAMR,	sprs.iamr);
725 		mtspr(SPRN_AMOR,	sprs.amor);
726 		mtspr(SPRN_UAMOR,	sprs.uamor);
727 
728 		/*
729 		 * Workaround for POWER9 DD2.0, if we lost resources, the ERAT
730 		 * might have been corrupted and needs flushing. We also need
731 		 * to reload MMCR0 (see mmcr0 comment above).
732 		 */
733 		if (!cpu_has_feature(CPU_FTR_POWER9_DD2_1)) {
734 			asm volatile(PPC_ISA_3_0_INVALIDATE_ERAT);
735 			mtspr(SPRN_MMCR0, mmcr0);
736 		}
737 
738 		/* Reload MMCRA to restore BHRB disable bit for POWER10 */
739 		if (cpu_has_feature(CPU_FTR_ARCH_31))
740 			mtspr(SPRN_MMCRA, mmcra);
741 
742 		/*
743 		 * DD2.2 and earlier need to set then clear bit 60 in MMCRA
744 		 * to ensure the PMU starts running.
745 		 */
746 		mmcra = mfspr(SPRN_MMCRA);
747 		mmcra |= PPC_BIT(60);
748 		mtspr(SPRN_MMCRA, mmcra);
749 		mmcra &= ~PPC_BIT(60);
750 		mtspr(SPRN_MMCRA, mmcra);
751 	}
752 
753 	if (unlikely((srr1 & SRR1_WAKEMASK_P8) == SRR1_WAKEHMI))
754 		hmi_exception_realmode(NULL);
755 
756 	/*
757 	 * On POWER9, SRR1 bits do not match exactly as expected.
758 	 * SRR1_WS_GPRLOSS (10b) can also result in SPR loss, so
759 	 * just always test PSSCR for SPR/TB state loss.
760 	 */
761 	pls = (psscr & PSSCR_PLS) >> PSSCR_PLS_SHIFT;
762 	if (likely(pls < deep_spr_loss_state)) {
763 		if (sprs_saved)
764 			atomic_stop_thread_idle();
765 		goto out;
766 	}
767 
768 	/* HV state loss */
769 	BUG_ON(!sprs_saved);
770 
771 	atomic_lock_thread_idle();
772 
773 	if ((*state & core_thread_mask) != 0)
774 		goto core_woken;
775 
776 	/* Per-core SPRs */
777 	mtspr(SPRN_PTCR,	sprs.ptcr);
778 	mtspr(SPRN_RPR,		sprs.rpr);
779 	mtspr(SPRN_TSCR,	sprs.tscr);
780 
781 	if (pls >= pnv_first_tb_loss_level) {
782 		/* TB loss */
783 		if (opal_resync_timebase() != OPAL_SUCCESS)
784 			BUG();
785 	}
786 
787 	/*
788 	 * isync after restoring shared SPRs and before unlocking. Unlock
789 	 * only contains hwsync which does not necessarily do the right
790 	 * thing for SPRs.
791 	 */
792 	isync();
793 
794 core_woken:
795 	atomic_unlock_and_stop_thread_idle();
796 
797 	/* Per-thread SPRs */
798 	mtspr(SPRN_LPCR,	sprs.lpcr);
799 	mtspr(SPRN_HFSCR,	sprs.hfscr);
800 	mtspr(SPRN_FSCR,	sprs.fscr);
801 	mtspr(SPRN_PID,		sprs.pid);
802 	mtspr(SPRN_PURR,	sprs.purr);
803 	mtspr(SPRN_SPURR,	sprs.spurr);
804 	mtspr(SPRN_DSCR,	sprs.dscr);
805 	mtspr(SPRN_WORT,	sprs.wort);
806 
807 	mtspr(SPRN_MMCRA,	sprs.mmcra);
808 	mtspr(SPRN_MMCR0,	sprs.mmcr0);
809 	mtspr(SPRN_MMCR1,	sprs.mmcr1);
810 	mtspr(SPRN_MMCR2,	sprs.mmcr2);
811 	if (!firmware_has_feature(FW_FEATURE_ULTRAVISOR))
812 		mtspr(SPRN_LDBAR, sprs.ldbar);
813 
814 	mtspr(SPRN_SPRG3,	local_paca->sprg_vdso);
815 
816 	if (!radix_enabled())
817 		__slb_restore_bolted_realmode();
818 
819 out:
820 	if (mmu_on)
821 		mtmsr(MSR_KERNEL);
822 
823 	return srr1;
824 }
825 
826 #ifdef CONFIG_HOTPLUG_CPU
827 static unsigned long power9_offline_stop(unsigned long psscr)
828 {
829 	unsigned long srr1;
830 
831 #ifndef CONFIG_KVM_BOOK3S_HV_POSSIBLE
832 	__ppc64_runlatch_off();
833 	srr1 = power9_idle_stop(psscr, true);
834 	__ppc64_runlatch_on();
835 #else
836 	/*
837 	 * Tell KVM we're entering idle.
838 	 * This does not have to be done in real mode because the P9 MMU
839 	 * is independent per-thread. Some steppings share radix/hash mode
840 	 * between threads, but in that case KVM has a barrier sync in real
841 	 * mode before and after switching between radix and hash.
842 	 *
843 	 * kvm_start_guest must still be called in real mode though, hence
844 	 * the false argument.
845 	 */
846 	local_paca->kvm_hstate.hwthread_state = KVM_HWTHREAD_IN_IDLE;
847 
848 	__ppc64_runlatch_off();
849 	srr1 = power9_idle_stop(psscr, false);
850 	__ppc64_runlatch_on();
851 
852 	local_paca->kvm_hstate.hwthread_state = KVM_HWTHREAD_IN_KERNEL;
853 	/* Order setting hwthread_state vs. testing hwthread_req */
854 	smp_mb();
855 	if (local_paca->kvm_hstate.hwthread_req)
856 		srr1 = idle_kvm_start_guest(srr1);
857 	mtmsr(MSR_KERNEL);
858 #endif
859 
860 	return srr1;
861 }
862 #endif
863 
864 void power9_idle_type(unsigned long stop_psscr_val,
865 				      unsigned long stop_psscr_mask)
866 {
867 	unsigned long psscr;
868 	unsigned long srr1;
869 
870 	if (!prep_irq_for_idle_irqsoff())
871 		return;
872 
873 	psscr = mfspr(SPRN_PSSCR);
874 	psscr = (psscr & ~stop_psscr_mask) | stop_psscr_val;
875 
876 	__ppc64_runlatch_off();
877 	srr1 = power9_idle_stop(psscr, true);
878 	__ppc64_runlatch_on();
879 
880 	fini_irq_for_idle_irqsoff();
881 
882 	irq_set_pending_from_srr1(srr1);
883 }
884 
885 /*
886  * Used for ppc_md.power_save which needs a function with no parameters
887  */
888 void power9_idle(void)
889 {
890 	power9_idle_type(pnv_default_stop_val, pnv_default_stop_mask);
891 }
892 
893 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
894 /*
895  * This is used in working around bugs in thread reconfiguration
896  * on POWER9 (at least up to Nimbus DD2.2) relating to transactional
897  * memory and the way that XER[SO] is checkpointed.
898  * This function forces the core into SMT4 in order by asking
899  * all other threads not to stop, and sending a message to any
900  * that are in a stop state.
901  * Must be called with preemption disabled.
902  */
903 void pnv_power9_force_smt4_catch(void)
904 {
905 	int cpu, cpu0, thr;
906 	int awake_threads = 1;		/* this thread is awake */
907 	int poke_threads = 0;
908 	int need_awake = threads_per_core;
909 
910 	cpu = smp_processor_id();
911 	cpu0 = cpu & ~(threads_per_core - 1);
912 	for (thr = 0; thr < threads_per_core; ++thr) {
913 		if (cpu != cpu0 + thr)
914 			atomic_inc(&paca_ptrs[cpu0+thr]->dont_stop);
915 	}
916 	/* order setting dont_stop vs testing requested_psscr */
917 	smp_mb();
918 	for (thr = 0; thr < threads_per_core; ++thr) {
919 		if (!paca_ptrs[cpu0+thr]->requested_psscr)
920 			++awake_threads;
921 		else
922 			poke_threads |= (1 << thr);
923 	}
924 
925 	/* If at least 3 threads are awake, the core is in SMT4 already */
926 	if (awake_threads < need_awake) {
927 		/* We have to wake some threads; we'll use msgsnd */
928 		for (thr = 0; thr < threads_per_core; ++thr) {
929 			if (poke_threads & (1 << thr)) {
930 				ppc_msgsnd_sync();
931 				ppc_msgsnd(PPC_DBELL_MSGTYPE, 0,
932 					   paca_ptrs[cpu0+thr]->hw_cpu_id);
933 			}
934 		}
935 		/* now spin until at least 3 threads are awake */
936 		do {
937 			for (thr = 0; thr < threads_per_core; ++thr) {
938 				if ((poke_threads & (1 << thr)) &&
939 				    !paca_ptrs[cpu0+thr]->requested_psscr) {
940 					++awake_threads;
941 					poke_threads &= ~(1 << thr);
942 				}
943 			}
944 		} while (awake_threads < need_awake);
945 	}
946 }
947 EXPORT_SYMBOL_GPL(pnv_power9_force_smt4_catch);
948 
949 void pnv_power9_force_smt4_release(void)
950 {
951 	int cpu, cpu0, thr;
952 
953 	cpu = smp_processor_id();
954 	cpu0 = cpu & ~(threads_per_core - 1);
955 
956 	/* clear all the dont_stop flags */
957 	for (thr = 0; thr < threads_per_core; ++thr) {
958 		if (cpu != cpu0 + thr)
959 			atomic_dec(&paca_ptrs[cpu0+thr]->dont_stop);
960 	}
961 }
962 EXPORT_SYMBOL_GPL(pnv_power9_force_smt4_release);
963 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
964 
965 #ifdef CONFIG_HOTPLUG_CPU
966 
967 void pnv_program_cpu_hotplug_lpcr(unsigned int cpu, u64 lpcr_val)
968 {
969 	u64 pir = get_hard_smp_processor_id(cpu);
970 
971 	mtspr(SPRN_LPCR, lpcr_val);
972 
973 	/*
974 	 * Program the LPCR via stop-api only if the deepest stop state
975 	 * can lose hypervisor context.
976 	 */
977 	if (supported_cpuidle_states & OPAL_PM_LOSE_FULL_CONTEXT)
978 		opal_slw_set_reg(pir, SPRN_LPCR, lpcr_val);
979 }
980 
981 /*
982  * pnv_cpu_offline: A function that puts the CPU into the deepest
983  * available platform idle state on a CPU-Offline.
984  * interrupts hard disabled and no lazy irq pending.
985  */
986 unsigned long pnv_cpu_offline(unsigned int cpu)
987 {
988 	unsigned long srr1;
989 
990 	__ppc64_runlatch_off();
991 
992 	if (cpu_has_feature(CPU_FTR_ARCH_300) && deepest_stop_found) {
993 		unsigned long psscr;
994 
995 		psscr = mfspr(SPRN_PSSCR);
996 		psscr = (psscr & ~pnv_deepest_stop_psscr_mask) |
997 						pnv_deepest_stop_psscr_val;
998 		srr1 = power9_offline_stop(psscr);
999 	} else if (cpu_has_feature(CPU_FTR_ARCH_206) && power7_offline_type) {
1000 		srr1 = power7_offline();
1001 	} else {
1002 		/* This is the fallback method. We emulate snooze */
1003 		while (!generic_check_cpu_restart(cpu)) {
1004 			HMT_low();
1005 			HMT_very_low();
1006 		}
1007 		srr1 = 0;
1008 		HMT_medium();
1009 	}
1010 
1011 	__ppc64_runlatch_on();
1012 
1013 	return srr1;
1014 }
1015 #endif
1016 
1017 /*
1018  * Power ISA 3.0 idle initialization.
1019  *
1020  * POWER ISA 3.0 defines a new SPR Processor stop Status and Control
1021  * Register (PSSCR) to control idle behavior.
1022  *
1023  * PSSCR layout:
1024  * ----------------------------------------------------------
1025  * | PLS | /// | SD | ESL | EC | PSLL | /// | TR | MTL | RL |
1026  * ----------------------------------------------------------
1027  * 0      4     41   42    43   44     48    54   56    60
1028  *
1029  * PSSCR key fields:
1030  *	Bits 0:3  - Power-Saving Level Status (PLS). This field indicates the
1031  *	lowest power-saving state the thread entered since stop instruction was
1032  *	last executed.
1033  *
1034  *	Bit 41 - Status Disable(SD)
1035  *	0 - Shows PLS entries
1036  *	1 - PLS entries are all 0
1037  *
1038  *	Bit 42 - Enable State Loss
1039  *	0 - No state is lost irrespective of other fields
1040  *	1 - Allows state loss
1041  *
1042  *	Bit 43 - Exit Criterion
1043  *	0 - Exit from power-save mode on any interrupt
1044  *	1 - Exit from power-save mode controlled by LPCR's PECE bits
1045  *
1046  *	Bits 44:47 - Power-Saving Level Limit
1047  *	This limits the power-saving level that can be entered into.
1048  *
1049  *	Bits 60:63 - Requested Level
1050  *	Used to specify which power-saving level must be entered on executing
1051  *	stop instruction
1052  */
1053 
1054 int validate_psscr_val_mask(u64 *psscr_val, u64 *psscr_mask, u32 flags)
1055 {
1056 	int err = 0;
1057 
1058 	/*
1059 	 * psscr_mask == 0xf indicates an older firmware.
1060 	 * Set remaining fields of psscr to the default values.
1061 	 * See NOTE above definition of PSSCR_HV_DEFAULT_VAL
1062 	 */
1063 	if (*psscr_mask == 0xf) {
1064 		*psscr_val = *psscr_val | PSSCR_HV_DEFAULT_VAL;
1065 		*psscr_mask = PSSCR_HV_DEFAULT_MASK;
1066 		return err;
1067 	}
1068 
1069 	/*
1070 	 * New firmware is expected to set the psscr_val bits correctly.
1071 	 * Validate that the following invariants are correctly maintained by
1072 	 * the new firmware.
1073 	 * - ESL bit value matches the EC bit value.
1074 	 * - ESL bit is set for all the deep stop states.
1075 	 */
1076 	if (GET_PSSCR_ESL(*psscr_val) != GET_PSSCR_EC(*psscr_val)) {
1077 		err = ERR_EC_ESL_MISMATCH;
1078 	} else if ((flags & OPAL_PM_LOSE_FULL_CONTEXT) &&
1079 		GET_PSSCR_ESL(*psscr_val) == 0) {
1080 		err = ERR_DEEP_STATE_ESL_MISMATCH;
1081 	}
1082 
1083 	return err;
1084 }
1085 
1086 /*
1087  * pnv_arch300_idle_init: Initializes the default idle state, first
1088  *                        deep idle state and deepest idle state on
1089  *                        ISA 3.0 CPUs.
1090  *
1091  * @np: /ibm,opal/power-mgt device node
1092  * @flags: cpu-idle-state-flags array
1093  * @dt_idle_states: Number of idle state entries
1094  * Returns 0 on success
1095  */
1096 static void __init pnv_power9_idle_init(void)
1097 {
1098 	u64 max_residency_ns = 0;
1099 	int i;
1100 
1101 	/*
1102 	 * pnv_deepest_stop_{val,mask} should be set to values corresponding to
1103 	 * the deepest stop state.
1104 	 *
1105 	 * pnv_default_stop_{val,mask} should be set to values corresponding to
1106 	 * the deepest loss-less (OPAL_PM_STOP_INST_FAST) stop state.
1107 	 */
1108 	pnv_first_tb_loss_level = MAX_STOP_STATE + 1;
1109 	deep_spr_loss_state = MAX_STOP_STATE + 1;
1110 	for (i = 0; i < nr_pnv_idle_states; i++) {
1111 		int err;
1112 		struct pnv_idle_states_t *state = &pnv_idle_states[i];
1113 		u64 psscr_rl = state->psscr_val & PSSCR_RL_MASK;
1114 
1115 		if ((state->flags & OPAL_PM_TIMEBASE_STOP) &&
1116 		     (pnv_first_tb_loss_level > psscr_rl))
1117 			pnv_first_tb_loss_level = psscr_rl;
1118 
1119 		if ((state->flags & OPAL_PM_LOSE_FULL_CONTEXT) &&
1120 		     (deep_spr_loss_state > psscr_rl))
1121 			deep_spr_loss_state = psscr_rl;
1122 
1123 		/*
1124 		 * The idle code does not deal with TB loss occurring
1125 		 * in a shallower state than SPR loss, so force it to
1126 		 * behave like SPRs are lost if TB is lost. POWER9 would
1127 		 * never encouter this, but a POWER8 core would if it
1128 		 * implemented the stop instruction. So this is for forward
1129 		 * compatibility.
1130 		 */
1131 		if ((state->flags & OPAL_PM_TIMEBASE_STOP) &&
1132 		     (deep_spr_loss_state > psscr_rl))
1133 			deep_spr_loss_state = psscr_rl;
1134 
1135 		err = validate_psscr_val_mask(&state->psscr_val,
1136 					      &state->psscr_mask,
1137 					      state->flags);
1138 		if (err) {
1139 			report_invalid_psscr_val(state->psscr_val, err);
1140 			continue;
1141 		}
1142 
1143 		state->valid = true;
1144 
1145 		if (max_residency_ns < state->residency_ns) {
1146 			max_residency_ns = state->residency_ns;
1147 			pnv_deepest_stop_psscr_val = state->psscr_val;
1148 			pnv_deepest_stop_psscr_mask = state->psscr_mask;
1149 			pnv_deepest_stop_flag = state->flags;
1150 			deepest_stop_found = true;
1151 		}
1152 
1153 		if (!default_stop_found &&
1154 		    (state->flags & OPAL_PM_STOP_INST_FAST)) {
1155 			pnv_default_stop_val = state->psscr_val;
1156 			pnv_default_stop_mask = state->psscr_mask;
1157 			default_stop_found = true;
1158 			WARN_ON(state->flags & OPAL_PM_LOSE_FULL_CONTEXT);
1159 		}
1160 	}
1161 
1162 	if (unlikely(!default_stop_found)) {
1163 		pr_warn("cpuidle-powernv: No suitable default stop state found. Disabling platform idle.\n");
1164 	} else {
1165 		ppc_md.power_save = power9_idle;
1166 		pr_info("cpuidle-powernv: Default stop: psscr = 0x%016llx,mask=0x%016llx\n",
1167 			pnv_default_stop_val, pnv_default_stop_mask);
1168 	}
1169 
1170 	if (unlikely(!deepest_stop_found)) {
1171 		pr_warn("cpuidle-powernv: No suitable stop state for CPU-Hotplug. Offlined CPUs will busy wait");
1172 	} else {
1173 		pr_info("cpuidle-powernv: Deepest stop: psscr = 0x%016llx,mask=0x%016llx\n",
1174 			pnv_deepest_stop_psscr_val,
1175 			pnv_deepest_stop_psscr_mask);
1176 	}
1177 
1178 	pr_info("cpuidle-powernv: First stop level that may lose SPRs = 0x%llx\n",
1179 		deep_spr_loss_state);
1180 
1181 	pr_info("cpuidle-powernv: First stop level that may lose timebase = 0x%llx\n",
1182 		pnv_first_tb_loss_level);
1183 }
1184 
1185 static void __init pnv_disable_deep_states(void)
1186 {
1187 	/*
1188 	 * The stop-api is unable to restore hypervisor
1189 	 * resources on wakeup from platform idle states which
1190 	 * lose full context. So disable such states.
1191 	 */
1192 	supported_cpuidle_states &= ~OPAL_PM_LOSE_FULL_CONTEXT;
1193 	pr_warn("cpuidle-powernv: Disabling idle states that lose full context\n");
1194 	pr_warn("cpuidle-powernv: Idle power-savings, CPU-Hotplug affected\n");
1195 
1196 	if (cpu_has_feature(CPU_FTR_ARCH_300) &&
1197 	    (pnv_deepest_stop_flag & OPAL_PM_LOSE_FULL_CONTEXT)) {
1198 		/*
1199 		 * Use the default stop state for CPU-Hotplug
1200 		 * if available.
1201 		 */
1202 		if (default_stop_found) {
1203 			pnv_deepest_stop_psscr_val = pnv_default_stop_val;
1204 			pnv_deepest_stop_psscr_mask = pnv_default_stop_mask;
1205 			pr_warn("cpuidle-powernv: Offlined CPUs will stop with psscr = 0x%016llx\n",
1206 				pnv_deepest_stop_psscr_val);
1207 		} else { /* Fallback to snooze loop for CPU-Hotplug */
1208 			deepest_stop_found = false;
1209 			pr_warn("cpuidle-powernv: Offlined CPUs will busy wait\n");
1210 		}
1211 	}
1212 }
1213 
1214 /*
1215  * Probe device tree for supported idle states
1216  */
1217 static void __init pnv_probe_idle_states(void)
1218 {
1219 	int i;
1220 
1221 	if (nr_pnv_idle_states < 0) {
1222 		pr_warn("cpuidle-powernv: no idle states found in the DT\n");
1223 		return;
1224 	}
1225 
1226 	if (pvr_version_is(PVR_POWER9))
1227 		pnv_power9_idle_init();
1228 
1229 	for (i = 0; i < nr_pnv_idle_states; i++)
1230 		supported_cpuidle_states |= pnv_idle_states[i].flags;
1231 }
1232 
1233 /*
1234  * This function parses device-tree and populates all the information
1235  * into pnv_idle_states structure. It also sets up nr_pnv_idle_states
1236  * which is the number of cpuidle states discovered through device-tree.
1237  */
1238 
1239 static int pnv_parse_cpuidle_dt(void)
1240 {
1241 	struct device_node *np;
1242 	int nr_idle_states, i;
1243 	int rc = 0;
1244 	u32 *temp_u32;
1245 	u64 *temp_u64;
1246 	const char **temp_string;
1247 
1248 	np = of_find_node_by_path("/ibm,opal/power-mgt");
1249 	if (!np) {
1250 		pr_warn("opal: PowerMgmt Node not found\n");
1251 		return -ENODEV;
1252 	}
1253 	nr_idle_states = of_property_count_u32_elems(np,
1254 						"ibm,cpu-idle-state-flags");
1255 
1256 	pnv_idle_states = kcalloc(nr_idle_states, sizeof(*pnv_idle_states),
1257 				  GFP_KERNEL);
1258 	temp_u32 = kcalloc(nr_idle_states, sizeof(u32),  GFP_KERNEL);
1259 	temp_u64 = kcalloc(nr_idle_states, sizeof(u64),  GFP_KERNEL);
1260 	temp_string = kcalloc(nr_idle_states, sizeof(char *),  GFP_KERNEL);
1261 
1262 	if (!(pnv_idle_states && temp_u32 && temp_u64 && temp_string)) {
1263 		pr_err("Could not allocate memory for dt parsing\n");
1264 		rc = -ENOMEM;
1265 		goto out;
1266 	}
1267 
1268 	/* Read flags */
1269 	if (of_property_read_u32_array(np, "ibm,cpu-idle-state-flags",
1270 				       temp_u32, nr_idle_states)) {
1271 		pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-flags in DT\n");
1272 		rc = -EINVAL;
1273 		goto out;
1274 	}
1275 	for (i = 0; i < nr_idle_states; i++)
1276 		pnv_idle_states[i].flags = temp_u32[i];
1277 
1278 	/* Read latencies */
1279 	if (of_property_read_u32_array(np, "ibm,cpu-idle-state-latencies-ns",
1280 				       temp_u32, nr_idle_states)) {
1281 		pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-latencies-ns in DT\n");
1282 		rc = -EINVAL;
1283 		goto out;
1284 	}
1285 	for (i = 0; i < nr_idle_states; i++)
1286 		pnv_idle_states[i].latency_ns = temp_u32[i];
1287 
1288 	/* Read residencies */
1289 	if (of_property_read_u32_array(np, "ibm,cpu-idle-state-residency-ns",
1290 				       temp_u32, nr_idle_states)) {
1291 		pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-residency-ns in DT\n");
1292 		rc = -EINVAL;
1293 		goto out;
1294 	}
1295 	for (i = 0; i < nr_idle_states; i++)
1296 		pnv_idle_states[i].residency_ns = temp_u32[i];
1297 
1298 	/* For power9 */
1299 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1300 		/* Read pm_crtl_val */
1301 		if (of_property_read_u64_array(np, "ibm,cpu-idle-state-psscr",
1302 					       temp_u64, nr_idle_states)) {
1303 			pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-psscr in DT\n");
1304 			rc = -EINVAL;
1305 			goto out;
1306 		}
1307 		for (i = 0; i < nr_idle_states; i++)
1308 			pnv_idle_states[i].psscr_val = temp_u64[i];
1309 
1310 		/* Read pm_crtl_mask */
1311 		if (of_property_read_u64_array(np, "ibm,cpu-idle-state-psscr-mask",
1312 					       temp_u64, nr_idle_states)) {
1313 			pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-psscr-mask in DT\n");
1314 			rc = -EINVAL;
1315 			goto out;
1316 		}
1317 		for (i = 0; i < nr_idle_states; i++)
1318 			pnv_idle_states[i].psscr_mask = temp_u64[i];
1319 	}
1320 
1321 	/*
1322 	 * power8 specific properties ibm,cpu-idle-state-pmicr-mask and
1323 	 * ibm,cpu-idle-state-pmicr-val were never used and there is no
1324 	 * plan to use it in near future. Hence, not parsing these properties
1325 	 */
1326 
1327 	if (of_property_read_string_array(np, "ibm,cpu-idle-state-names",
1328 					  temp_string, nr_idle_states) < 0) {
1329 		pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-names in DT\n");
1330 		rc = -EINVAL;
1331 		goto out;
1332 	}
1333 	for (i = 0; i < nr_idle_states; i++)
1334 		strlcpy(pnv_idle_states[i].name, temp_string[i],
1335 			PNV_IDLE_NAME_LEN);
1336 	nr_pnv_idle_states = nr_idle_states;
1337 	rc = 0;
1338 out:
1339 	kfree(temp_u32);
1340 	kfree(temp_u64);
1341 	kfree(temp_string);
1342 	return rc;
1343 }
1344 
1345 static int __init pnv_init_idle_states(void)
1346 {
1347 	int cpu;
1348 	int rc = 0;
1349 
1350 	/* Set up PACA fields */
1351 	for_each_present_cpu(cpu) {
1352 		struct paca_struct *p = paca_ptrs[cpu];
1353 
1354 		p->idle_state = 0;
1355 		if (cpu == cpu_first_thread_sibling(cpu))
1356 			p->idle_state = (1 << threads_per_core) - 1;
1357 
1358 		if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1359 			/* P7/P8 nap */
1360 			p->thread_idle_state = PNV_THREAD_RUNNING;
1361 		} else {
1362 			/* P9 stop */
1363 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1364 			p->requested_psscr = 0;
1365 			atomic_set(&p->dont_stop, 0);
1366 #endif
1367 		}
1368 	}
1369 
1370 	/* In case we error out nr_pnv_idle_states will be zero */
1371 	nr_pnv_idle_states = 0;
1372 	supported_cpuidle_states = 0;
1373 
1374 	if (cpuidle_disable != IDLE_NO_OVERRIDE)
1375 		goto out;
1376 	rc = pnv_parse_cpuidle_dt();
1377 	if (rc)
1378 		return rc;
1379 	pnv_probe_idle_states();
1380 
1381 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1382 		if (!(supported_cpuidle_states & OPAL_PM_SLEEP_ENABLED_ER1)) {
1383 			power7_fastsleep_workaround_entry = false;
1384 			power7_fastsleep_workaround_exit = false;
1385 		} else {
1386 			/*
1387 			 * OPAL_PM_SLEEP_ENABLED_ER1 is set. It indicates that
1388 			 * workaround is needed to use fastsleep. Provide sysfs
1389 			 * control to choose how this workaround has to be
1390 			 * applied.
1391 			 */
1392 			device_create_file(cpu_subsys.dev_root,
1393 				&dev_attr_fastsleep_workaround_applyonce);
1394 		}
1395 
1396 		update_subcore_sibling_mask();
1397 
1398 		if (supported_cpuidle_states & OPAL_PM_NAP_ENABLED) {
1399 			ppc_md.power_save = power7_idle;
1400 			power7_offline_type = PNV_THREAD_NAP;
1401 		}
1402 
1403 		if ((supported_cpuidle_states & OPAL_PM_WINKLE_ENABLED) &&
1404 			   (supported_cpuidle_states & OPAL_PM_LOSE_FULL_CONTEXT))
1405 			power7_offline_type = PNV_THREAD_WINKLE;
1406 		else if ((supported_cpuidle_states & OPAL_PM_SLEEP_ENABLED) ||
1407 			   (supported_cpuidle_states & OPAL_PM_SLEEP_ENABLED_ER1))
1408 			power7_offline_type = PNV_THREAD_SLEEP;
1409 	}
1410 
1411 	if (supported_cpuidle_states & OPAL_PM_LOSE_FULL_CONTEXT) {
1412 		if (pnv_save_sprs_for_deep_states())
1413 			pnv_disable_deep_states();
1414 	}
1415 
1416 out:
1417 	return 0;
1418 }
1419 machine_subsys_initcall(powernv, pnv_init_idle_states);
1420