1 /*
2  * PowerNV cpuidle code
3  *
4  * Copyright 2015 IBM Corp.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version
9  * 2 of the License, or (at your option) any later version.
10  */
11 
12 #include <linux/types.h>
13 #include <linux/mm.h>
14 #include <linux/slab.h>
15 #include <linux/of.h>
16 #include <linux/device.h>
17 #include <linux/cpu.h>
18 
19 #include <asm/firmware.h>
20 #include <asm/machdep.h>
21 #include <asm/opal.h>
22 #include <asm/cputhreads.h>
23 #include <asm/cpuidle.h>
24 #include <asm/code-patching.h>
25 #include <asm/smp.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 static u32 supported_cpuidle_states;
34 
35 static int pnv_save_sprs_for_deep_states(void)
36 {
37 	int cpu;
38 	int rc;
39 
40 	/*
41 	 * hid0, hid1, hid4, hid5, hmeer and lpcr values are symmetric across
42 	 * all cpus at boot. Get these reg values of current cpu and use the
43 	 * same across all cpus.
44 	 */
45 	uint64_t lpcr_val = mfspr(SPRN_LPCR) & ~(u64)LPCR_PECE1;
46 	uint64_t hid0_val = mfspr(SPRN_HID0);
47 	uint64_t hid1_val = mfspr(SPRN_HID1);
48 	uint64_t hid4_val = mfspr(SPRN_HID4);
49 	uint64_t hid5_val = mfspr(SPRN_HID5);
50 	uint64_t hmeer_val = mfspr(SPRN_HMEER);
51 
52 	for_each_possible_cpu(cpu) {
53 		uint64_t pir = get_hard_smp_processor_id(cpu);
54 		uint64_t hsprg0_val = (uint64_t)&paca[cpu];
55 
56 		if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
57 			/*
58 			 * HSPRG0 is used to store the cpu's pointer to paca.
59 			 * Hence last 3 bits are guaranteed to be 0. Program
60 			 * slw to restore HSPRG0 with 63rd bit set, so that
61 			 * when a thread wakes up at 0x100 we can use this bit
62 			 * to distinguish between fastsleep and deep winkle.
63 			 * This is not necessary with stop/psscr since PLS
64 			 * field of psscr indicates which state we are waking
65 			 * up from.
66 			 */
67 			hsprg0_val |= 1;
68 		}
69 		rc = opal_slw_set_reg(pir, SPRN_HSPRG0, hsprg0_val);
70 		if (rc != 0)
71 			return rc;
72 
73 		rc = opal_slw_set_reg(pir, SPRN_LPCR, lpcr_val);
74 		if (rc != 0)
75 			return rc;
76 
77 		/* HIDs are per core registers */
78 		if (cpu_thread_in_core(cpu) == 0) {
79 
80 			rc = opal_slw_set_reg(pir, SPRN_HMEER, hmeer_val);
81 			if (rc != 0)
82 				return rc;
83 
84 			rc = opal_slw_set_reg(pir, SPRN_HID0, hid0_val);
85 			if (rc != 0)
86 				return rc;
87 
88 			rc = opal_slw_set_reg(pir, SPRN_HID1, hid1_val);
89 			if (rc != 0)
90 				return rc;
91 
92 			rc = opal_slw_set_reg(pir, SPRN_HID4, hid4_val);
93 			if (rc != 0)
94 				return rc;
95 
96 			rc = opal_slw_set_reg(pir, SPRN_HID5, hid5_val);
97 			if (rc != 0)
98 				return rc;
99 		}
100 	}
101 
102 	return 0;
103 }
104 
105 static void pnv_alloc_idle_core_states(void)
106 {
107 	int i, j;
108 	int nr_cores = cpu_nr_cores();
109 	u32 *core_idle_state;
110 
111 	/*
112 	 * core_idle_state - First 8 bits track the idle state of each thread
113 	 * of the core. The 8th bit is the lock bit. Initially all thread bits
114 	 * are set. They are cleared when the thread enters deep idle state
115 	 * like sleep and winkle. Initially the lock bit is cleared.
116 	 * The lock bit has 2 purposes
117 	 * a. While the first thread is restoring core state, it prevents
118 	 * other threads in the core from switching to process context.
119 	 * b. While the last thread in the core is saving the core state, it
120 	 * prevents a different thread from waking up.
121 	 */
122 	for (i = 0; i < nr_cores; i++) {
123 		int first_cpu = i * threads_per_core;
124 		int node = cpu_to_node(first_cpu);
125 
126 		core_idle_state = kmalloc_node(sizeof(u32), GFP_KERNEL, node);
127 		*core_idle_state = PNV_CORE_IDLE_THREAD_BITS;
128 
129 		for (j = 0; j < threads_per_core; j++) {
130 			int cpu = first_cpu + j;
131 
132 			paca[cpu].core_idle_state_ptr = core_idle_state;
133 			paca[cpu].thread_idle_state = PNV_THREAD_RUNNING;
134 			paca[cpu].thread_mask = 1 << j;
135 		}
136 	}
137 
138 	update_subcore_sibling_mask();
139 
140 	if (supported_cpuidle_states & OPAL_PM_LOSE_FULL_CONTEXT)
141 		pnv_save_sprs_for_deep_states();
142 }
143 
144 u32 pnv_get_supported_cpuidle_states(void)
145 {
146 	return supported_cpuidle_states;
147 }
148 EXPORT_SYMBOL_GPL(pnv_get_supported_cpuidle_states);
149 
150 
151 static void pnv_fastsleep_workaround_apply(void *info)
152 
153 {
154 	int rc;
155 	int *err = info;
156 
157 	rc = opal_config_cpu_idle_state(OPAL_CONFIG_IDLE_FASTSLEEP,
158 					OPAL_CONFIG_IDLE_APPLY);
159 	if (rc)
160 		*err = 1;
161 }
162 
163 /*
164  * Used to store fastsleep workaround state
165  * 0 - Workaround applied/undone at fastsleep entry/exit path (Default)
166  * 1 - Workaround applied once, never undone.
167  */
168 static u8 fastsleep_workaround_applyonce;
169 
170 static ssize_t show_fastsleep_workaround_applyonce(struct device *dev,
171 		struct device_attribute *attr, char *buf)
172 {
173 	return sprintf(buf, "%u\n", fastsleep_workaround_applyonce);
174 }
175 
176 static ssize_t store_fastsleep_workaround_applyonce(struct device *dev,
177 		struct device_attribute *attr, const char *buf,
178 		size_t count)
179 {
180 	cpumask_t primary_thread_mask;
181 	int err;
182 	u8 val;
183 
184 	if (kstrtou8(buf, 0, &val) || val != 1)
185 		return -EINVAL;
186 
187 	if (fastsleep_workaround_applyonce == 1)
188 		return count;
189 
190 	/*
191 	 * fastsleep_workaround_applyonce = 1 implies
192 	 * fastsleep workaround needs to be left in 'applied' state on all
193 	 * the cores. Do this by-
194 	 * 1. Patching out the call to 'undo' workaround in fastsleep exit path
195 	 * 2. Sending ipi to all the cores which have at least one online thread
196 	 * 3. Patching out the call to 'apply' workaround in fastsleep entry
197 	 * path
198 	 * There is no need to send ipi to cores which have all threads
199 	 * offlined, as last thread of the core entering fastsleep or deeper
200 	 * state would have applied workaround.
201 	 */
202 	err = patch_instruction(
203 		(unsigned int *)pnv_fastsleep_workaround_at_exit,
204 		PPC_INST_NOP);
205 	if (err) {
206 		pr_err("fastsleep_workaround_applyonce change failed while patching pnv_fastsleep_workaround_at_exit");
207 		goto fail;
208 	}
209 
210 	get_online_cpus();
211 	primary_thread_mask = cpu_online_cores_map();
212 	on_each_cpu_mask(&primary_thread_mask,
213 				pnv_fastsleep_workaround_apply,
214 				&err, 1);
215 	put_online_cpus();
216 	if (err) {
217 		pr_err("fastsleep_workaround_applyonce change failed while running pnv_fastsleep_workaround_apply");
218 		goto fail;
219 	}
220 
221 	err = patch_instruction(
222 		(unsigned int *)pnv_fastsleep_workaround_at_entry,
223 		PPC_INST_NOP);
224 	if (err) {
225 		pr_err("fastsleep_workaround_applyonce change failed while patching pnv_fastsleep_workaround_at_entry");
226 		goto fail;
227 	}
228 
229 	fastsleep_workaround_applyonce = 1;
230 
231 	return count;
232 fail:
233 	return -EIO;
234 }
235 
236 static DEVICE_ATTR(fastsleep_workaround_applyonce, 0600,
237 			show_fastsleep_workaround_applyonce,
238 			store_fastsleep_workaround_applyonce);
239 
240 /*
241  * The default stop state that will be used by ppc_md.power_save
242  * function on platforms that support stop instruction.
243  */
244 u64 pnv_default_stop_val;
245 u64 pnv_default_stop_mask;
246 
247 /*
248  * Used for ppc_md.power_save which needs a function with no parameters
249  */
250 static void power9_idle(void)
251 {
252 	power9_idle_stop(pnv_default_stop_val, pnv_default_stop_mask);
253 }
254 
255 /*
256  * First deep stop state. Used to figure out when to save/restore
257  * hypervisor context.
258  */
259 u64 pnv_first_deep_stop_state = MAX_STOP_STATE;
260 
261 /*
262  * psscr value and mask of the deepest stop idle state.
263  * Used when a cpu is offlined.
264  */
265 u64 pnv_deepest_stop_psscr_val;
266 u64 pnv_deepest_stop_psscr_mask;
267 
268 /*
269  * Power ISA 3.0 idle initialization.
270  *
271  * POWER ISA 3.0 defines a new SPR Processor stop Status and Control
272  * Register (PSSCR) to control idle behavior.
273  *
274  * PSSCR layout:
275  * ----------------------------------------------------------
276  * | PLS | /// | SD | ESL | EC | PSLL | /// | TR | MTL | RL |
277  * ----------------------------------------------------------
278  * 0      4     41   42    43   44     48    54   56    60
279  *
280  * PSSCR key fields:
281  *	Bits 0:3  - Power-Saving Level Status (PLS). This field indicates the
282  *	lowest power-saving state the thread entered since stop instruction was
283  *	last executed.
284  *
285  *	Bit 41 - Status Disable(SD)
286  *	0 - Shows PLS entries
287  *	1 - PLS entries are all 0
288  *
289  *	Bit 42 - Enable State Loss
290  *	0 - No state is lost irrespective of other fields
291  *	1 - Allows state loss
292  *
293  *	Bit 43 - Exit Criterion
294  *	0 - Exit from power-save mode on any interrupt
295  *	1 - Exit from power-save mode controlled by LPCR's PECE bits
296  *
297  *	Bits 44:47 - Power-Saving Level Limit
298  *	This limits the power-saving level that can be entered into.
299  *
300  *	Bits 60:63 - Requested Level
301  *	Used to specify which power-saving level must be entered on executing
302  *	stop instruction
303  */
304 
305 int validate_psscr_val_mask(u64 *psscr_val, u64 *psscr_mask, u32 flags)
306 {
307 	int err = 0;
308 
309 	/*
310 	 * psscr_mask == 0xf indicates an older firmware.
311 	 * Set remaining fields of psscr to the default values.
312 	 * See NOTE above definition of PSSCR_HV_DEFAULT_VAL
313 	 */
314 	if (*psscr_mask == 0xf) {
315 		*psscr_val = *psscr_val | PSSCR_HV_DEFAULT_VAL;
316 		*psscr_mask = PSSCR_HV_DEFAULT_MASK;
317 		return err;
318 	}
319 
320 	/*
321 	 * New firmware is expected to set the psscr_val bits correctly.
322 	 * Validate that the following invariants are correctly maintained by
323 	 * the new firmware.
324 	 * - ESL bit value matches the EC bit value.
325 	 * - ESL bit is set for all the deep stop states.
326 	 */
327 	if (GET_PSSCR_ESL(*psscr_val) != GET_PSSCR_EC(*psscr_val)) {
328 		err = ERR_EC_ESL_MISMATCH;
329 	} else if ((flags & OPAL_PM_LOSE_FULL_CONTEXT) &&
330 		GET_PSSCR_ESL(*psscr_val) == 0) {
331 		err = ERR_DEEP_STATE_ESL_MISMATCH;
332 	}
333 
334 	return err;
335 }
336 
337 /*
338  * pnv_arch300_idle_init: Initializes the default idle state, first
339  *                        deep idle state and deepest idle state on
340  *                        ISA 3.0 CPUs.
341  *
342  * @np: /ibm,opal/power-mgt device node
343  * @flags: cpu-idle-state-flags array
344  * @dt_idle_states: Number of idle state entries
345  * Returns 0 on success
346  */
347 static int __init pnv_power9_idle_init(struct device_node *np, u32 *flags,
348 					int dt_idle_states)
349 {
350 	u64 *psscr_val = NULL;
351 	u64 *psscr_mask = NULL;
352 	u32 *residency_ns = NULL;
353 	u64 max_residency_ns = 0;
354 	int rc = 0, i;
355 	bool default_stop_found = false, deepest_stop_found = false;
356 
357 	psscr_val = kcalloc(dt_idle_states, sizeof(*psscr_val), GFP_KERNEL);
358 	psscr_mask = kcalloc(dt_idle_states, sizeof(*psscr_mask), GFP_KERNEL);
359 	residency_ns = kcalloc(dt_idle_states, sizeof(*residency_ns),
360 			       GFP_KERNEL);
361 
362 	if (!psscr_val || !psscr_mask || !residency_ns) {
363 		rc = -1;
364 		goto out;
365 	}
366 
367 	if (of_property_read_u64_array(np,
368 		"ibm,cpu-idle-state-psscr",
369 		psscr_val, dt_idle_states)) {
370 		pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-psscr in DT\n");
371 		rc = -1;
372 		goto out;
373 	}
374 
375 	if (of_property_read_u64_array(np,
376 				       "ibm,cpu-idle-state-psscr-mask",
377 				       psscr_mask, dt_idle_states)) {
378 		pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-psscr-mask in DT\n");
379 		rc = -1;
380 		goto out;
381 	}
382 
383 	if (of_property_read_u32_array(np,
384 				       "ibm,cpu-idle-state-residency-ns",
385 					residency_ns, dt_idle_states)) {
386 		pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-residency-ns in DT\n");
387 		rc = -1;
388 		goto out;
389 	}
390 
391 	/*
392 	 * Set pnv_first_deep_stop_state, pnv_deepest_stop_psscr_{val,mask},
393 	 * and the pnv_default_stop_{val,mask}.
394 	 *
395 	 * pnv_first_deep_stop_state should be set to the first stop
396 	 * level to cause hypervisor state loss.
397 	 *
398 	 * pnv_deepest_stop_{val,mask} should be set to values corresponding to
399 	 * the deepest stop state.
400 	 *
401 	 * pnv_default_stop_{val,mask} should be set to values corresponding to
402 	 * the shallowest (OPAL_PM_STOP_INST_FAST) loss-less stop state.
403 	 */
404 	pnv_first_deep_stop_state = MAX_STOP_STATE;
405 	for (i = 0; i < dt_idle_states; i++) {
406 		int err;
407 		u64 psscr_rl = psscr_val[i] & PSSCR_RL_MASK;
408 
409 		if ((flags[i] & OPAL_PM_LOSE_FULL_CONTEXT) &&
410 		     (pnv_first_deep_stop_state > psscr_rl))
411 			pnv_first_deep_stop_state = psscr_rl;
412 
413 		err = validate_psscr_val_mask(&psscr_val[i], &psscr_mask[i],
414 					      flags[i]);
415 		if (err) {
416 			report_invalid_psscr_val(psscr_val[i], err);
417 			continue;
418 		}
419 
420 		if (max_residency_ns < residency_ns[i]) {
421 			max_residency_ns = residency_ns[i];
422 			pnv_deepest_stop_psscr_val = psscr_val[i];
423 			pnv_deepest_stop_psscr_mask = psscr_mask[i];
424 			deepest_stop_found = true;
425 		}
426 
427 		if (!default_stop_found &&
428 		    (flags[i] & OPAL_PM_STOP_INST_FAST)) {
429 			pnv_default_stop_val = psscr_val[i];
430 			pnv_default_stop_mask = psscr_mask[i];
431 			default_stop_found = true;
432 		}
433 	}
434 
435 	if (!default_stop_found) {
436 		pnv_default_stop_val = PSSCR_HV_DEFAULT_VAL;
437 		pnv_default_stop_mask = PSSCR_HV_DEFAULT_MASK;
438 		pr_warn("Setting default stop psscr val=0x%016llx,mask=0x%016llx\n",
439 			pnv_default_stop_val, pnv_default_stop_mask);
440 	}
441 
442 	if (!deepest_stop_found) {
443 		pnv_deepest_stop_psscr_val = PSSCR_HV_DEFAULT_VAL;
444 		pnv_deepest_stop_psscr_mask = PSSCR_HV_DEFAULT_MASK;
445 		pr_warn("Setting default stop psscr val=0x%016llx,mask=0x%016llx\n",
446 			pnv_deepest_stop_psscr_val,
447 			pnv_deepest_stop_psscr_mask);
448 	}
449 
450 out:
451 	kfree(psscr_val);
452 	kfree(psscr_mask);
453 	kfree(residency_ns);
454 	return rc;
455 }
456 
457 /*
458  * Probe device tree for supported idle states
459  */
460 static void __init pnv_probe_idle_states(void)
461 {
462 	struct device_node *np;
463 	int dt_idle_states;
464 	u32 *flags = NULL;
465 	int i;
466 
467 	np = of_find_node_by_path("/ibm,opal/power-mgt");
468 	if (!np) {
469 		pr_warn("opal: PowerMgmt Node not found\n");
470 		goto out;
471 	}
472 	dt_idle_states = of_property_count_u32_elems(np,
473 			"ibm,cpu-idle-state-flags");
474 	if (dt_idle_states < 0) {
475 		pr_warn("cpuidle-powernv: no idle states found in the DT\n");
476 		goto out;
477 	}
478 
479 	flags = kcalloc(dt_idle_states, sizeof(*flags),  GFP_KERNEL);
480 
481 	if (of_property_read_u32_array(np,
482 			"ibm,cpu-idle-state-flags", flags, dt_idle_states)) {
483 		pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-flags in DT\n");
484 		goto out;
485 	}
486 
487 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
488 		if (pnv_power9_idle_init(np, flags, dt_idle_states))
489 			goto out;
490 	}
491 
492 	for (i = 0; i < dt_idle_states; i++)
493 		supported_cpuidle_states |= flags[i];
494 
495 out:
496 	kfree(flags);
497 }
498 static int __init pnv_init_idle_states(void)
499 {
500 
501 	supported_cpuidle_states = 0;
502 
503 	if (cpuidle_disable != IDLE_NO_OVERRIDE)
504 		goto out;
505 
506 	pnv_probe_idle_states();
507 
508 	if (!(supported_cpuidle_states & OPAL_PM_SLEEP_ENABLED_ER1)) {
509 		patch_instruction(
510 			(unsigned int *)pnv_fastsleep_workaround_at_entry,
511 			PPC_INST_NOP);
512 		patch_instruction(
513 			(unsigned int *)pnv_fastsleep_workaround_at_exit,
514 			PPC_INST_NOP);
515 	} else {
516 		/*
517 		 * OPAL_PM_SLEEP_ENABLED_ER1 is set. It indicates that
518 		 * workaround is needed to use fastsleep. Provide sysfs
519 		 * control to choose how this workaround has to be applied.
520 		 */
521 		device_create_file(cpu_subsys.dev_root,
522 				&dev_attr_fastsleep_workaround_applyonce);
523 	}
524 
525 	pnv_alloc_idle_core_states();
526 
527 	if (supported_cpuidle_states & OPAL_PM_NAP_ENABLED)
528 		ppc_md.power_save = power7_idle;
529 	else if (supported_cpuidle_states & OPAL_PM_STOP_INST_FAST)
530 		ppc_md.power_save = power9_idle;
531 
532 out:
533 	return 0;
534 }
535 machine_subsys_initcall(powernv, pnv_init_idle_states);
536