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