1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  cpuidle-powernv - idle state cpuidle driver.
4  *  Adapted from drivers/cpuidle/cpuidle-pseries
5  *
6  */
7 
8 #include <linux/kernel.h>
9 #include <linux/module.h>
10 #include <linux/init.h>
11 #include <linux/moduleparam.h>
12 #include <linux/cpuidle.h>
13 #include <linux/cpu.h>
14 #include <linux/notifier.h>
15 #include <linux/clockchips.h>
16 #include <linux/of.h>
17 #include <linux/slab.h>
18 
19 #include <asm/machdep.h>
20 #include <asm/firmware.h>
21 #include <asm/opal.h>
22 #include <asm/runlatch.h>
23 #include <asm/cpuidle.h>
24 
25 /*
26  * Expose only those Hardware idle states via the cpuidle framework
27  * that have latency value below POWERNV_THRESHOLD_LATENCY_NS.
28  */
29 #define POWERNV_THRESHOLD_LATENCY_NS 200000
30 
31 static struct cpuidle_driver powernv_idle_driver = {
32 	.name             = "powernv_idle",
33 	.owner            = THIS_MODULE,
34 };
35 
36 static int max_idle_state __read_mostly;
37 static struct cpuidle_state *cpuidle_state_table __read_mostly;
38 
39 struct stop_psscr_table {
40 	u64 val;
41 	u64 mask;
42 };
43 
44 static struct stop_psscr_table stop_psscr_table[CPUIDLE_STATE_MAX] __read_mostly;
45 
46 static u64 default_snooze_timeout __read_mostly;
47 static bool snooze_timeout_en __read_mostly;
48 
49 static u64 get_snooze_timeout(struct cpuidle_device *dev,
50 			      struct cpuidle_driver *drv,
51 			      int index)
52 {
53 	int i;
54 
55 	if (unlikely(!snooze_timeout_en))
56 		return default_snooze_timeout;
57 
58 	for (i = index + 1; i < drv->state_count; i++) {
59 		struct cpuidle_state *s = &drv->states[i];
60 		struct cpuidle_state_usage *su = &dev->states_usage[i];
61 
62 		if (s->disabled || su->disable)
63 			continue;
64 
65 		return s->target_residency * tb_ticks_per_usec;
66 	}
67 
68 	return default_snooze_timeout;
69 }
70 
71 static int snooze_loop(struct cpuidle_device *dev,
72 			struct cpuidle_driver *drv,
73 			int index)
74 {
75 	u64 snooze_exit_time;
76 
77 	set_thread_flag(TIF_POLLING_NRFLAG);
78 
79 	local_irq_enable();
80 
81 	snooze_exit_time = get_tb() + get_snooze_timeout(dev, drv, index);
82 	ppc64_runlatch_off();
83 	HMT_very_low();
84 	while (!need_resched()) {
85 		if (likely(snooze_timeout_en) && get_tb() > snooze_exit_time) {
86 			/*
87 			 * Task has not woken up but we are exiting the polling
88 			 * loop anyway. Require a barrier after polling is
89 			 * cleared to order subsequent test of need_resched().
90 			 */
91 			clear_thread_flag(TIF_POLLING_NRFLAG);
92 			smp_mb();
93 			break;
94 		}
95 	}
96 
97 	HMT_medium();
98 	ppc64_runlatch_on();
99 	clear_thread_flag(TIF_POLLING_NRFLAG);
100 
101 	local_irq_disable();
102 
103 	return index;
104 }
105 
106 static int nap_loop(struct cpuidle_device *dev,
107 			struct cpuidle_driver *drv,
108 			int index)
109 {
110 	power7_idle_type(PNV_THREAD_NAP);
111 
112 	return index;
113 }
114 
115 /* Register for fastsleep only in oneshot mode of broadcast */
116 #ifdef CONFIG_TICK_ONESHOT
117 static int fastsleep_loop(struct cpuidle_device *dev,
118 				struct cpuidle_driver *drv,
119 				int index)
120 {
121 	unsigned long old_lpcr = mfspr(SPRN_LPCR);
122 	unsigned long new_lpcr;
123 
124 	if (unlikely(system_state < SYSTEM_RUNNING))
125 		return index;
126 
127 	new_lpcr = old_lpcr;
128 	/* Do not exit powersave upon decrementer as we've setup the timer
129 	 * offload.
130 	 */
131 	new_lpcr &= ~LPCR_PECE1;
132 
133 	mtspr(SPRN_LPCR, new_lpcr);
134 
135 	power7_idle_type(PNV_THREAD_SLEEP);
136 
137 	mtspr(SPRN_LPCR, old_lpcr);
138 
139 	return index;
140 }
141 #endif
142 
143 static int stop_loop(struct cpuidle_device *dev,
144 		     struct cpuidle_driver *drv,
145 		     int index)
146 {
147 	power9_idle_type(stop_psscr_table[index].val,
148 			 stop_psscr_table[index].mask);
149 	return index;
150 }
151 
152 /*
153  * States for dedicated partition case.
154  */
155 static struct cpuidle_state powernv_states[CPUIDLE_STATE_MAX] = {
156 	{ /* Snooze */
157 		.name = "snooze",
158 		.desc = "snooze",
159 		.exit_latency = 0,
160 		.target_residency = 0,
161 		.enter = snooze_loop },
162 };
163 
164 static int powernv_cpuidle_cpu_online(unsigned int cpu)
165 {
166 	struct cpuidle_device *dev = per_cpu(cpuidle_devices, cpu);
167 
168 	if (dev && cpuidle_get_driver()) {
169 		cpuidle_pause_and_lock();
170 		cpuidle_enable_device(dev);
171 		cpuidle_resume_and_unlock();
172 	}
173 	return 0;
174 }
175 
176 static int powernv_cpuidle_cpu_dead(unsigned int cpu)
177 {
178 	struct cpuidle_device *dev = per_cpu(cpuidle_devices, cpu);
179 
180 	if (dev && cpuidle_get_driver()) {
181 		cpuidle_pause_and_lock();
182 		cpuidle_disable_device(dev);
183 		cpuidle_resume_and_unlock();
184 	}
185 	return 0;
186 }
187 
188 /*
189  * powernv_cpuidle_driver_init()
190  */
191 static int powernv_cpuidle_driver_init(void)
192 {
193 	int idle_state;
194 	struct cpuidle_driver *drv = &powernv_idle_driver;
195 
196 	drv->state_count = 0;
197 
198 	for (idle_state = 0; idle_state < max_idle_state; ++idle_state) {
199 		/* Is the state not enabled? */
200 		if (cpuidle_state_table[idle_state].enter == NULL)
201 			continue;
202 
203 		drv->states[drv->state_count] =	/* structure copy */
204 			cpuidle_state_table[idle_state];
205 
206 		drv->state_count += 1;
207 	}
208 
209 	/*
210 	 * On the PowerNV platform cpu_present may be less than cpu_possible in
211 	 * cases when firmware detects the CPU, but it is not available to the
212 	 * OS.  If CONFIG_HOTPLUG_CPU=n, then such CPUs are not hotplugable at
213 	 * run time and hence cpu_devices are not created for those CPUs by the
214 	 * generic topology_init().
215 	 *
216 	 * drv->cpumask defaults to cpu_possible_mask in
217 	 * __cpuidle_driver_init().  This breaks cpuidle on PowerNV where
218 	 * cpu_devices are not created for CPUs in cpu_possible_mask that
219 	 * cannot be hot-added later at run time.
220 	 *
221 	 * Trying cpuidle_register_device() on a CPU without a cpu_device is
222 	 * incorrect, so pass a correct CPU mask to the generic cpuidle driver.
223 	 */
224 
225 	drv->cpumask = (struct cpumask *)cpu_present_mask;
226 
227 	return 0;
228 }
229 
230 static inline void add_powernv_state(int index, const char *name,
231 				     unsigned int flags,
232 				     int (*idle_fn)(struct cpuidle_device *,
233 						    struct cpuidle_driver *,
234 						    int),
235 				     unsigned int target_residency,
236 				     unsigned int exit_latency,
237 				     u64 psscr_val, u64 psscr_mask)
238 {
239 	strlcpy(powernv_states[index].name, name, CPUIDLE_NAME_LEN);
240 	strlcpy(powernv_states[index].desc, name, CPUIDLE_NAME_LEN);
241 	powernv_states[index].flags = flags;
242 	powernv_states[index].target_residency = target_residency;
243 	powernv_states[index].exit_latency = exit_latency;
244 	powernv_states[index].enter = idle_fn;
245 	stop_psscr_table[index].val = psscr_val;
246 	stop_psscr_table[index].mask = psscr_mask;
247 }
248 
249 /*
250  * Returns 0 if prop1_len == prop2_len. Else returns -1
251  */
252 static inline int validate_dt_prop_sizes(const char *prop1, int prop1_len,
253 					 const char *prop2, int prop2_len)
254 {
255 	if (prop1_len == prop2_len)
256 		return 0;
257 
258 	pr_warn("cpuidle-powernv: array sizes don't match for %s and %s\n",
259 		prop1, prop2);
260 	return -1;
261 }
262 
263 extern u32 pnv_get_supported_cpuidle_states(void);
264 static int powernv_add_idle_states(void)
265 {
266 	struct device_node *power_mgt;
267 	int nr_idle_states = 1; /* Snooze */
268 	int dt_idle_states, count;
269 	u32 latency_ns[CPUIDLE_STATE_MAX];
270 	u32 residency_ns[CPUIDLE_STATE_MAX];
271 	u32 flags[CPUIDLE_STATE_MAX];
272 	u64 psscr_val[CPUIDLE_STATE_MAX];
273 	u64 psscr_mask[CPUIDLE_STATE_MAX];
274 	const char *names[CPUIDLE_STATE_MAX];
275 	u32 has_stop_states = 0;
276 	int i, rc;
277 	u32 supported_flags = pnv_get_supported_cpuidle_states();
278 
279 
280 	/* Currently we have snooze statically defined */
281 
282 	power_mgt = of_find_node_by_path("/ibm,opal/power-mgt");
283 	if (!power_mgt) {
284 		pr_warn("opal: PowerMgmt Node not found\n");
285 		goto out;
286 	}
287 
288 	/* Read values of any property to determine the num of idle states */
289 	dt_idle_states = of_property_count_u32_elems(power_mgt, "ibm,cpu-idle-state-flags");
290 	if (dt_idle_states < 0) {
291 		pr_warn("cpuidle-powernv: no idle states found in the DT\n");
292 		goto out;
293 	}
294 
295 	count = of_property_count_u32_elems(power_mgt,
296 					    "ibm,cpu-idle-state-latencies-ns");
297 
298 	if (validate_dt_prop_sizes("ibm,cpu-idle-state-flags", dt_idle_states,
299 				   "ibm,cpu-idle-state-latencies-ns",
300 				   count) != 0)
301 		goto out;
302 
303 	count = of_property_count_strings(power_mgt,
304 					  "ibm,cpu-idle-state-names");
305 	if (validate_dt_prop_sizes("ibm,cpu-idle-state-flags", dt_idle_states,
306 				   "ibm,cpu-idle-state-names",
307 				   count) != 0)
308 		goto out;
309 
310 	/*
311 	 * Since snooze is used as first idle state, max idle states allowed is
312 	 * CPUIDLE_STATE_MAX -1
313 	 */
314 	if (dt_idle_states > CPUIDLE_STATE_MAX - 1) {
315 		pr_warn("cpuidle-powernv: discovered idle states more than allowed");
316 		dt_idle_states = CPUIDLE_STATE_MAX - 1;
317 	}
318 
319 	if (of_property_read_u32_array(power_mgt,
320 			"ibm,cpu-idle-state-flags", flags, dt_idle_states)) {
321 		pr_warn("cpuidle-powernv : missing ibm,cpu-idle-state-flags in DT\n");
322 		goto out;
323 	}
324 
325 	if (of_property_read_u32_array(power_mgt,
326 		"ibm,cpu-idle-state-latencies-ns", latency_ns,
327 		dt_idle_states)) {
328 		pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-latencies-ns in DT\n");
329 		goto out;
330 	}
331 	if (of_property_read_string_array(power_mgt,
332 		"ibm,cpu-idle-state-names", names, dt_idle_states) < 0) {
333 		pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-names in DT\n");
334 		goto out;
335 	}
336 
337 	/*
338 	 * If the idle states use stop instruction, probe for psscr values
339 	 * and psscr mask which are necessary to specify required stop level.
340 	 */
341 	has_stop_states = (flags[0] &
342 			   (OPAL_PM_STOP_INST_FAST | OPAL_PM_STOP_INST_DEEP));
343 	if (has_stop_states) {
344 		count = of_property_count_u64_elems(power_mgt,
345 						    "ibm,cpu-idle-state-psscr");
346 		if (validate_dt_prop_sizes("ibm,cpu-idle-state-flags",
347 					   dt_idle_states,
348 					   "ibm,cpu-idle-state-psscr",
349 					   count) != 0)
350 			goto out;
351 
352 		count = of_property_count_u64_elems(power_mgt,
353 						    "ibm,cpu-idle-state-psscr-mask");
354 		if (validate_dt_prop_sizes("ibm,cpu-idle-state-flags",
355 					   dt_idle_states,
356 					   "ibm,cpu-idle-state-psscr-mask",
357 					   count) != 0)
358 			goto out;
359 
360 		if (of_property_read_u64_array(power_mgt,
361 		    "ibm,cpu-idle-state-psscr", psscr_val, dt_idle_states)) {
362 			pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-psscr in DT\n");
363 			goto out;
364 		}
365 
366 		if (of_property_read_u64_array(power_mgt,
367 					       "ibm,cpu-idle-state-psscr-mask",
368 						psscr_mask, dt_idle_states)) {
369 			pr_warn("cpuidle-powernv:Missing ibm,cpu-idle-state-psscr-mask in DT\n");
370 			goto out;
371 		}
372 	}
373 
374 	count = of_property_count_u32_elems(power_mgt,
375 					    "ibm,cpu-idle-state-residency-ns");
376 
377 	if (count < 0) {
378 		rc = count;
379 	} else if (validate_dt_prop_sizes("ibm,cpu-idle-state-flags",
380 					  dt_idle_states,
381 					  "ibm,cpu-idle-state-residency-ns",
382 					  count) != 0) {
383 		goto out;
384 	} else {
385 		rc = of_property_read_u32_array(power_mgt,
386 						"ibm,cpu-idle-state-residency-ns",
387 						residency_ns, dt_idle_states);
388 	}
389 
390 	for (i = 0; i < dt_idle_states; i++) {
391 		unsigned int exit_latency, target_residency;
392 		bool stops_timebase = false;
393 
394 		/*
395 		 * Skip the platform idle state whose flag isn't in
396 		 * the supported_cpuidle_states flag mask.
397 		 */
398 		if ((flags[i] & supported_flags) != flags[i])
399 			continue;
400 		/*
401 		 * If an idle state has exit latency beyond
402 		 * POWERNV_THRESHOLD_LATENCY_NS then don't use it
403 		 * in cpu-idle.
404 		 */
405 		if (latency_ns[i] > POWERNV_THRESHOLD_LATENCY_NS)
406 			continue;
407 		/*
408 		 * Firmware passes residency and latency values in ns.
409 		 * cpuidle expects it in us.
410 		 */
411 		exit_latency = DIV_ROUND_UP(latency_ns[i], 1000);
412 		if (!rc)
413 			target_residency = DIV_ROUND_UP(residency_ns[i], 1000);
414 		else
415 			target_residency = 0;
416 
417 		if (has_stop_states) {
418 			int err = validate_psscr_val_mask(&psscr_val[i],
419 							  &psscr_mask[i],
420 							  flags[i]);
421 			if (err) {
422 				report_invalid_psscr_val(psscr_val[i], err);
423 				continue;
424 			}
425 		}
426 
427 		if (flags[i] & OPAL_PM_TIMEBASE_STOP)
428 			stops_timebase = true;
429 
430 		/*
431 		 * For nap and fastsleep, use default target_residency
432 		 * values if f/w does not expose it.
433 		 */
434 		if (flags[i] & OPAL_PM_NAP_ENABLED) {
435 			if (!rc)
436 				target_residency = 100;
437 			/* Add NAP state */
438 			add_powernv_state(nr_idle_states, "Nap",
439 					  CPUIDLE_FLAG_NONE, nap_loop,
440 					  target_residency, exit_latency, 0, 0);
441 		} else if (has_stop_states && !stops_timebase) {
442 			add_powernv_state(nr_idle_states, names[i],
443 					  CPUIDLE_FLAG_NONE, stop_loop,
444 					  target_residency, exit_latency,
445 					  psscr_val[i], psscr_mask[i]);
446 		}
447 
448 		/*
449 		 * All cpuidle states with CPUIDLE_FLAG_TIMER_STOP set must come
450 		 * within this config dependency check.
451 		 */
452 #ifdef CONFIG_TICK_ONESHOT
453 		else if (flags[i] & OPAL_PM_SLEEP_ENABLED ||
454 			 flags[i] & OPAL_PM_SLEEP_ENABLED_ER1) {
455 			if (!rc)
456 				target_residency = 300000;
457 			/* Add FASTSLEEP state */
458 			add_powernv_state(nr_idle_states, "FastSleep",
459 					  CPUIDLE_FLAG_TIMER_STOP,
460 					  fastsleep_loop,
461 					  target_residency, exit_latency, 0, 0);
462 		} else if (has_stop_states && stops_timebase) {
463 			add_powernv_state(nr_idle_states, names[i],
464 					  CPUIDLE_FLAG_TIMER_STOP, stop_loop,
465 					  target_residency, exit_latency,
466 					  psscr_val[i], psscr_mask[i]);
467 		}
468 #endif
469 		else
470 			continue;
471 		nr_idle_states++;
472 	}
473 out:
474 	return nr_idle_states;
475 }
476 
477 /*
478  * powernv_idle_probe()
479  * Choose state table for shared versus dedicated partition
480  */
481 static int powernv_idle_probe(void)
482 {
483 	if (cpuidle_disable != IDLE_NO_OVERRIDE)
484 		return -ENODEV;
485 
486 	if (firmware_has_feature(FW_FEATURE_OPAL)) {
487 		cpuidle_state_table = powernv_states;
488 		/* Device tree can indicate more idle states */
489 		max_idle_state = powernv_add_idle_states();
490 		default_snooze_timeout = TICK_USEC * tb_ticks_per_usec;
491 		if (max_idle_state > 1)
492 			snooze_timeout_en = true;
493  	} else
494  		return -ENODEV;
495 
496 	return 0;
497 }
498 
499 static int __init powernv_processor_idle_init(void)
500 {
501 	int retval;
502 
503 	retval = powernv_idle_probe();
504 	if (retval)
505 		return retval;
506 
507 	powernv_cpuidle_driver_init();
508 	retval = cpuidle_register(&powernv_idle_driver, NULL);
509 	if (retval) {
510 		printk(KERN_DEBUG "Registration of powernv driver failed.\n");
511 		return retval;
512 	}
513 
514 	retval = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
515 					   "cpuidle/powernv:online",
516 					   powernv_cpuidle_cpu_online, NULL);
517 	WARN_ON(retval < 0);
518 	retval = cpuhp_setup_state_nocalls(CPUHP_CPUIDLE_DEAD,
519 					   "cpuidle/powernv:dead", NULL,
520 					   powernv_cpuidle_cpu_dead);
521 	WARN_ON(retval < 0);
522 	printk(KERN_DEBUG "powernv_idle_driver registered\n");
523 	return 0;
524 }
525 
526 device_initcall(powernv_processor_idle_init);
527