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 23 #define MAX_POWERNV_IDLE_STATES 8 24 25 struct cpuidle_driver powernv_idle_driver = { 26 .name = "powernv_idle", 27 .owner = THIS_MODULE, 28 }; 29 30 static int max_idle_state; 31 static struct cpuidle_state *cpuidle_state_table; 32 static u64 snooze_timeout; 33 static bool snooze_timeout_en; 34 35 static int snooze_loop(struct cpuidle_device *dev, 36 struct cpuidle_driver *drv, 37 int index) 38 { 39 u64 snooze_exit_time; 40 41 local_irq_enable(); 42 set_thread_flag(TIF_POLLING_NRFLAG); 43 44 snooze_exit_time = get_tb() + snooze_timeout; 45 ppc64_runlatch_off(); 46 while (!need_resched()) { 47 HMT_low(); 48 HMT_very_low(); 49 if (snooze_timeout_en && get_tb() > snooze_exit_time) 50 break; 51 } 52 53 HMT_medium(); 54 ppc64_runlatch_on(); 55 clear_thread_flag(TIF_POLLING_NRFLAG); 56 smp_mb(); 57 return index; 58 } 59 60 static int nap_loop(struct cpuidle_device *dev, 61 struct cpuidle_driver *drv, 62 int index) 63 { 64 ppc64_runlatch_off(); 65 power7_idle(); 66 ppc64_runlatch_on(); 67 return index; 68 } 69 70 static int fastsleep_loop(struct cpuidle_device *dev, 71 struct cpuidle_driver *drv, 72 int index) 73 { 74 unsigned long old_lpcr = mfspr(SPRN_LPCR); 75 unsigned long new_lpcr; 76 77 if (unlikely(system_state < SYSTEM_RUNNING)) 78 return index; 79 80 new_lpcr = old_lpcr; 81 /* Do not exit powersave upon decrementer as we've setup the timer 82 * offload. 83 */ 84 new_lpcr &= ~LPCR_PECE1; 85 86 mtspr(SPRN_LPCR, new_lpcr); 87 power7_sleep(); 88 89 mtspr(SPRN_LPCR, old_lpcr); 90 91 return index; 92 } 93 94 /* 95 * States for dedicated partition case. 96 */ 97 static struct cpuidle_state powernv_states[MAX_POWERNV_IDLE_STATES] = { 98 { /* Snooze */ 99 .name = "snooze", 100 .desc = "snooze", 101 .exit_latency = 0, 102 .target_residency = 0, 103 .enter = &snooze_loop }, 104 }; 105 106 static int powernv_cpuidle_add_cpu_notifier(struct notifier_block *n, 107 unsigned long action, void *hcpu) 108 { 109 int hotcpu = (unsigned long)hcpu; 110 struct cpuidle_device *dev = 111 per_cpu(cpuidle_devices, hotcpu); 112 113 if (dev && cpuidle_get_driver()) { 114 switch (action) { 115 case CPU_ONLINE: 116 case CPU_ONLINE_FROZEN: 117 cpuidle_pause_and_lock(); 118 cpuidle_enable_device(dev); 119 cpuidle_resume_and_unlock(); 120 break; 121 122 case CPU_DEAD: 123 case CPU_DEAD_FROZEN: 124 cpuidle_pause_and_lock(); 125 cpuidle_disable_device(dev); 126 cpuidle_resume_and_unlock(); 127 break; 128 129 default: 130 return NOTIFY_DONE; 131 } 132 } 133 return NOTIFY_OK; 134 } 135 136 static struct notifier_block setup_hotplug_notifier = { 137 .notifier_call = powernv_cpuidle_add_cpu_notifier, 138 }; 139 140 /* 141 * powernv_cpuidle_driver_init() 142 */ 143 static int powernv_cpuidle_driver_init(void) 144 { 145 int idle_state; 146 struct cpuidle_driver *drv = &powernv_idle_driver; 147 148 drv->state_count = 0; 149 150 for (idle_state = 0; idle_state < max_idle_state; ++idle_state) { 151 /* Is the state not enabled? */ 152 if (cpuidle_state_table[idle_state].enter == NULL) 153 continue; 154 155 drv->states[drv->state_count] = /* structure copy */ 156 cpuidle_state_table[idle_state]; 157 158 drv->state_count += 1; 159 } 160 161 return 0; 162 } 163 164 static int powernv_add_idle_states(void) 165 { 166 struct device_node *power_mgt; 167 int nr_idle_states = 1; /* Snooze */ 168 int dt_idle_states; 169 u32 *latency_ns, *residency_ns, *flags; 170 int i, rc; 171 172 /* Currently we have snooze statically defined */ 173 174 power_mgt = of_find_node_by_path("/ibm,opal/power-mgt"); 175 if (!power_mgt) { 176 pr_warn("opal: PowerMgmt Node not found\n"); 177 goto out; 178 } 179 180 /* Read values of any property to determine the num of idle states */ 181 dt_idle_states = of_property_count_u32_elems(power_mgt, "ibm,cpu-idle-state-flags"); 182 if (dt_idle_states < 0) { 183 pr_warn("cpuidle-powernv: no idle states found in the DT\n"); 184 goto out; 185 } 186 187 flags = kzalloc(sizeof(*flags) * dt_idle_states, GFP_KERNEL); 188 if (of_property_read_u32_array(power_mgt, 189 "ibm,cpu-idle-state-flags", flags, dt_idle_states)) { 190 pr_warn("cpuidle-powernv : missing ibm,cpu-idle-state-flags in DT\n"); 191 goto out_free_flags; 192 } 193 194 latency_ns = kzalloc(sizeof(*latency_ns) * dt_idle_states, GFP_KERNEL); 195 rc = of_property_read_u32_array(power_mgt, 196 "ibm,cpu-idle-state-latencies-ns", latency_ns, dt_idle_states); 197 if (rc) { 198 pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-latencies-ns in DT\n"); 199 goto out_free_latency; 200 } 201 202 residency_ns = kzalloc(sizeof(*residency_ns) * dt_idle_states, GFP_KERNEL); 203 rc = of_property_read_u32_array(power_mgt, 204 "ibm,cpu-idle-state-residency-ns", residency_ns, dt_idle_states); 205 206 for (i = 0; i < dt_idle_states; i++) { 207 208 /* 209 * Cpuidle accepts exit_latency and target_residency in us. 210 * Use default target_residency values if f/w does not expose it. 211 */ 212 if (flags[i] & OPAL_PM_NAP_ENABLED) { 213 /* Add NAP state */ 214 strcpy(powernv_states[nr_idle_states].name, "Nap"); 215 strcpy(powernv_states[nr_idle_states].desc, "Nap"); 216 powernv_states[nr_idle_states].flags = 0; 217 powernv_states[nr_idle_states].target_residency = 100; 218 powernv_states[nr_idle_states].enter = &nap_loop; 219 } else if (flags[i] & OPAL_PM_SLEEP_ENABLED || 220 flags[i] & OPAL_PM_SLEEP_ENABLED_ER1) { 221 /* Add FASTSLEEP state */ 222 strcpy(powernv_states[nr_idle_states].name, "FastSleep"); 223 strcpy(powernv_states[nr_idle_states].desc, "FastSleep"); 224 powernv_states[nr_idle_states].flags = CPUIDLE_FLAG_TIMER_STOP; 225 powernv_states[nr_idle_states].target_residency = 300000; 226 powernv_states[nr_idle_states].enter = &fastsleep_loop; 227 } 228 229 powernv_states[nr_idle_states].exit_latency = 230 ((unsigned int)latency_ns[i]) / 1000; 231 232 if (!rc) { 233 powernv_states[nr_idle_states].target_residency = 234 ((unsigned int)residency_ns[i]) / 1000; 235 } 236 237 nr_idle_states++; 238 } 239 240 kfree(residency_ns); 241 out_free_latency: 242 kfree(latency_ns); 243 out_free_flags: 244 kfree(flags); 245 out: 246 return nr_idle_states; 247 } 248 249 /* 250 * powernv_idle_probe() 251 * Choose state table for shared versus dedicated partition 252 */ 253 static int powernv_idle_probe(void) 254 { 255 if (cpuidle_disable != IDLE_NO_OVERRIDE) 256 return -ENODEV; 257 258 if (firmware_has_feature(FW_FEATURE_OPALv3)) { 259 cpuidle_state_table = powernv_states; 260 /* Device tree can indicate more idle states */ 261 max_idle_state = powernv_add_idle_states(); 262 if (max_idle_state > 1) { 263 snooze_timeout_en = true; 264 snooze_timeout = powernv_states[1].target_residency * 265 tb_ticks_per_usec; 266 } 267 } else 268 return -ENODEV; 269 270 return 0; 271 } 272 273 static int __init powernv_processor_idle_init(void) 274 { 275 int retval; 276 277 retval = powernv_idle_probe(); 278 if (retval) 279 return retval; 280 281 powernv_cpuidle_driver_init(); 282 retval = cpuidle_register(&powernv_idle_driver, NULL); 283 if (retval) { 284 printk(KERN_DEBUG "Registration of powernv driver failed.\n"); 285 return retval; 286 } 287 288 register_cpu_notifier(&setup_hotplug_notifier); 289 printk(KERN_DEBUG "powernv_idle_driver registered\n"); 290 return 0; 291 } 292 293 device_initcall(powernv_processor_idle_init); 294