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