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 extern u32 pnv_get_supported_cpuidle_states(void); 239 static int powernv_add_idle_states(void) 240 { 241 struct device_node *power_mgt; 242 int nr_idle_states = 1; /* Snooze */ 243 int dt_idle_states, count; 244 u32 latency_ns[CPUIDLE_STATE_MAX]; 245 u32 residency_ns[CPUIDLE_STATE_MAX]; 246 u32 flags[CPUIDLE_STATE_MAX]; 247 u64 psscr_val[CPUIDLE_STATE_MAX]; 248 u64 psscr_mask[CPUIDLE_STATE_MAX]; 249 const char *names[CPUIDLE_STATE_MAX]; 250 u32 has_stop_states = 0; 251 int i, rc; 252 u32 supported_flags = pnv_get_supported_cpuidle_states(); 253 254 255 /* Currently we have snooze statically defined */ 256 257 power_mgt = of_find_node_by_path("/ibm,opal/power-mgt"); 258 if (!power_mgt) { 259 pr_warn("opal: PowerMgmt Node not found\n"); 260 goto out; 261 } 262 263 /* Read values of any property to determine the num of idle states */ 264 dt_idle_states = of_property_count_u32_elems(power_mgt, "ibm,cpu-idle-state-flags"); 265 if (dt_idle_states < 0) { 266 pr_warn("cpuidle-powernv: no idle states found in the DT\n"); 267 goto out; 268 } 269 270 count = of_property_count_u32_elems(power_mgt, 271 "ibm,cpu-idle-state-latencies-ns"); 272 273 if (validate_dt_prop_sizes("ibm,cpu-idle-state-flags", dt_idle_states, 274 "ibm,cpu-idle-state-latencies-ns", 275 count) != 0) 276 goto out; 277 278 count = of_property_count_strings(power_mgt, 279 "ibm,cpu-idle-state-names"); 280 if (validate_dt_prop_sizes("ibm,cpu-idle-state-flags", dt_idle_states, 281 "ibm,cpu-idle-state-names", 282 count) != 0) 283 goto out; 284 285 /* 286 * Since snooze is used as first idle state, max idle states allowed is 287 * CPUIDLE_STATE_MAX -1 288 */ 289 if (dt_idle_states > CPUIDLE_STATE_MAX - 1) { 290 pr_warn("cpuidle-powernv: discovered idle states more than allowed"); 291 dt_idle_states = CPUIDLE_STATE_MAX - 1; 292 } 293 294 if (of_property_read_u32_array(power_mgt, 295 "ibm,cpu-idle-state-flags", flags, dt_idle_states)) { 296 pr_warn("cpuidle-powernv : missing ibm,cpu-idle-state-flags in DT\n"); 297 goto out; 298 } 299 300 if (of_property_read_u32_array(power_mgt, 301 "ibm,cpu-idle-state-latencies-ns", latency_ns, 302 dt_idle_states)) { 303 pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-latencies-ns in DT\n"); 304 goto out; 305 } 306 if (of_property_read_string_array(power_mgt, 307 "ibm,cpu-idle-state-names", names, dt_idle_states) < 0) { 308 pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-names in DT\n"); 309 goto out; 310 } 311 312 /* 313 * If the idle states use stop instruction, probe for psscr values 314 * and psscr mask which are necessary to specify required stop level. 315 */ 316 has_stop_states = (flags[0] & 317 (OPAL_PM_STOP_INST_FAST | OPAL_PM_STOP_INST_DEEP)); 318 if (has_stop_states) { 319 count = of_property_count_u64_elems(power_mgt, 320 "ibm,cpu-idle-state-psscr"); 321 if (validate_dt_prop_sizes("ibm,cpu-idle-state-flags", 322 dt_idle_states, 323 "ibm,cpu-idle-state-psscr", 324 count) != 0) 325 goto out; 326 327 count = of_property_count_u64_elems(power_mgt, 328 "ibm,cpu-idle-state-psscr-mask"); 329 if (validate_dt_prop_sizes("ibm,cpu-idle-state-flags", 330 dt_idle_states, 331 "ibm,cpu-idle-state-psscr-mask", 332 count) != 0) 333 goto out; 334 335 if (of_property_read_u64_array(power_mgt, 336 "ibm,cpu-idle-state-psscr", psscr_val, dt_idle_states)) { 337 pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-psscr in DT\n"); 338 goto out; 339 } 340 341 if (of_property_read_u64_array(power_mgt, 342 "ibm,cpu-idle-state-psscr-mask", 343 psscr_mask, dt_idle_states)) { 344 pr_warn("cpuidle-powernv:Missing ibm,cpu-idle-state-psscr-mask in DT\n"); 345 goto out; 346 } 347 } 348 349 count = of_property_count_u32_elems(power_mgt, 350 "ibm,cpu-idle-state-residency-ns"); 351 352 if (count < 0) { 353 rc = count; 354 } else if (validate_dt_prop_sizes("ibm,cpu-idle-state-flags", 355 dt_idle_states, 356 "ibm,cpu-idle-state-residency-ns", 357 count) != 0) { 358 goto out; 359 } else { 360 rc = of_property_read_u32_array(power_mgt, 361 "ibm,cpu-idle-state-residency-ns", 362 residency_ns, dt_idle_states); 363 } 364 365 for (i = 0; i < dt_idle_states; i++) { 366 unsigned int exit_latency, target_residency; 367 bool stops_timebase = false; 368 369 /* 370 * Skip the platform idle state whose flag isn't in 371 * the supported_cpuidle_states flag mask. 372 */ 373 if ((flags[i] & supported_flags) != flags[i]) 374 continue; 375 /* 376 * If an idle state has exit latency beyond 377 * POWERNV_THRESHOLD_LATENCY_NS then don't use it 378 * in cpu-idle. 379 */ 380 if (latency_ns[i] > POWERNV_THRESHOLD_LATENCY_NS) 381 continue; 382 /* 383 * Firmware passes residency and latency values in ns. 384 * cpuidle expects it in us. 385 */ 386 exit_latency = latency_ns[i] / 1000; 387 if (!rc) 388 target_residency = residency_ns[i] / 1000; 389 else 390 target_residency = 0; 391 392 if (has_stop_states) { 393 int err = validate_psscr_val_mask(&psscr_val[i], 394 &psscr_mask[i], 395 flags[i]); 396 if (err) { 397 report_invalid_psscr_val(psscr_val[i], err); 398 continue; 399 } 400 } 401 402 if (flags[i] & OPAL_PM_TIMEBASE_STOP) 403 stops_timebase = true; 404 405 /* 406 * For nap and fastsleep, use default target_residency 407 * values if f/w does not expose it. 408 */ 409 if (flags[i] & OPAL_PM_NAP_ENABLED) { 410 if (!rc) 411 target_residency = 100; 412 /* Add NAP state */ 413 add_powernv_state(nr_idle_states, "Nap", 414 CPUIDLE_FLAG_NONE, nap_loop, 415 target_residency, exit_latency, 0, 0); 416 } else if (has_stop_states && !stops_timebase) { 417 add_powernv_state(nr_idle_states, names[i], 418 CPUIDLE_FLAG_NONE, stop_loop, 419 target_residency, exit_latency, 420 psscr_val[i], psscr_mask[i]); 421 } 422 423 /* 424 * All cpuidle states with CPUIDLE_FLAG_TIMER_STOP set must come 425 * within this config dependency check. 426 */ 427 #ifdef CONFIG_TICK_ONESHOT 428 else if (flags[i] & OPAL_PM_SLEEP_ENABLED || 429 flags[i] & OPAL_PM_SLEEP_ENABLED_ER1) { 430 if (!rc) 431 target_residency = 300000; 432 /* Add FASTSLEEP state */ 433 add_powernv_state(nr_idle_states, "FastSleep", 434 CPUIDLE_FLAG_TIMER_STOP, 435 fastsleep_loop, 436 target_residency, exit_latency, 0, 0); 437 } else if (has_stop_states && stops_timebase) { 438 add_powernv_state(nr_idle_states, names[i], 439 CPUIDLE_FLAG_TIMER_STOP, stop_loop, 440 target_residency, exit_latency, 441 psscr_val[i], psscr_mask[i]); 442 } 443 #endif 444 else 445 continue; 446 nr_idle_states++; 447 } 448 out: 449 return nr_idle_states; 450 } 451 452 /* 453 * powernv_idle_probe() 454 * Choose state table for shared versus dedicated partition 455 */ 456 static int powernv_idle_probe(void) 457 { 458 if (cpuidle_disable != IDLE_NO_OVERRIDE) 459 return -ENODEV; 460 461 if (firmware_has_feature(FW_FEATURE_OPAL)) { 462 cpuidle_state_table = powernv_states; 463 /* Device tree can indicate more idle states */ 464 max_idle_state = powernv_add_idle_states(); 465 if (max_idle_state > 1) { 466 snooze_timeout_en = true; 467 snooze_timeout = powernv_states[1].target_residency * 468 tb_ticks_per_usec; 469 } 470 } else 471 return -ENODEV; 472 473 return 0; 474 } 475 476 static int __init powernv_processor_idle_init(void) 477 { 478 int retval; 479 480 retval = powernv_idle_probe(); 481 if (retval) 482 return retval; 483 484 powernv_cpuidle_driver_init(); 485 retval = cpuidle_register(&powernv_idle_driver, NULL); 486 if (retval) { 487 printk(KERN_DEBUG "Registration of powernv driver failed.\n"); 488 return retval; 489 } 490 491 retval = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, 492 "cpuidle/powernv:online", 493 powernv_cpuidle_cpu_online, NULL); 494 WARN_ON(retval < 0); 495 retval = cpuhp_setup_state_nocalls(CPUHP_CPUIDLE_DEAD, 496 "cpuidle/powernv:dead", NULL, 497 powernv_cpuidle_cpu_dead); 498 WARN_ON(retval < 0); 499 printk(KERN_DEBUG "powernv_idle_driver registered\n"); 500 return 0; 501 } 502 503 device_initcall(powernv_processor_idle_init); 504