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