arch/x86/kernel/process.c

   1 #include <linux/errno.h>
   2 #include <linux/kernel.h>
   3 #include <linux/mm.h>
   4 #include <asm/idle.h>
   5 #include <linux/smp.h>
   6 #include <linux/slab.h>
   7 #include <linux/sched.h>
   8 #include <linux/module.h>
   9 #include <linux/pm.h>
  10 #include <linux/clockchips.h>
  11 #include <trace/power.h>
  12 #include <asm/system.h>
  13 #include <asm/apic.h>
  14
  15 unsigned long idle_halt;
  16 EXPORT_SYMBOL(idle_halt);
  17 unsigned long idle_nomwait;
  18 EXPORT_SYMBOL(idle_nomwait);
  19
  20 struct kmem_cache *task_xstate_cachep;
  21
  22 DEFINE_TRACE(power_start);
  23 DEFINE_TRACE(power_end);
  24
  25 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
  26 {
  27         *dst = *src;
  28         if (src->thread.xstate) {
  29                 dst->thread.xstate = kmem_cache_alloc(task_xstate_cachep,
  30                                                       GFP_KERNEL);
  31                 if (!dst->thread.xstate)
  32                         return -ENOMEM;
  33                 WARN_ON((unsigned long)dst->thread.xstate & 15);
  34                 memcpy(dst->thread.xstate, src->thread.xstate, xstate_size);
  35         }
  36         return 0;
  37 }
  38
  39 void free_thread_xstate(struct task_struct *tsk)
  40 {
  41         if (tsk->thread.xstate) {
  42                 kmem_cache_free(task_xstate_cachep, tsk->thread.xstate);
  43                 tsk->thread.xstate = NULL;
  44         }
  45 }
  46
  47 void free_thread_info(struct thread_info *ti)
  48 {
  49         free_thread_xstate(ti->task);
  50         free_pages((unsigned long)ti, get_order(THREAD_SIZE));
  51 }
  52
  53 void arch_task_cache_init(void)
  54 {
  55         task_xstate_cachep =
  56                 kmem_cache_create("task_xstate", xstate_size,
  57                                   __alignof__(union thread_xstate),
  58                                   SLAB_PANIC, NULL);
  59 }
  60
  61 /*
  62  * Idle related variables and functions
  63  */
  64 unsigned long boot_option_idle_override = 0;
  65 EXPORT_SYMBOL(boot_option_idle_override);
  66
  67 /*
  68  * Powermanagement idle function, if any..
  69  */
  70 void (*pm_idle)(void);
  71 EXPORT_SYMBOL(pm_idle);
  72
  73 #ifdef CONFIG_X86_32
  74 /*
  75  * This halt magic was a workaround for ancient floppy DMA
  76  * wreckage. It should be safe to remove.
  77  */
  78 static int hlt_counter;
  79 void disable_hlt(void)
  80 {
  81         hlt_counter++;
  82 }
  83 EXPORT_SYMBOL(disable_hlt);
  84
  85 void enable_hlt(void)
  86 {
  87         hlt_counter--;
  88 }
  89 EXPORT_SYMBOL(enable_hlt);
  90
  91 static inline int hlt_use_halt(void)
  92 {
  93         return (!hlt_counter && boot_cpu_data.hlt_works_ok);
  94 }
  95 #else
  96 static inline int hlt_use_halt(void)
  97 {
  98         return 1;
  99 }
 100 #endif
 101
 102 /*
 103  * We use this if we don't have any better
 104  * idle routine..
 105  */
 106 void default_idle(void)
 107 {
 108         if (hlt_use_halt()) {
 109                 struct power_trace it;
 110
 111                 trace_power_start(&it, POWER_CSTATE, 1);
 112                 current_thread_info()->status &= ~TS_POLLING;
 113                 /*
 114                  * TS_POLLING-cleared state must be visible before we
 115                  * test NEED_RESCHED:
 116                  */
 117                 smp_mb();
 118
 119                 if (!need_resched())
 120                         safe_halt();    /* enables interrupts racelessly */
 121                 else
 122                         local_irq_enable();
 123                 current_thread_info()->status |= TS_POLLING;
 124                 trace_power_end(&it);
 125         } else {
 126                 local_irq_enable();
 127                 /* loop is done by the caller */
 128                 cpu_relax();
 129         }
 130 }
 131 #ifdef CONFIG_APM_MODULE
 132 EXPORT_SYMBOL(default_idle);
 133 #endif
 134
 135 void stop_this_cpu(void *dummy)
 136 {
 137         local_irq_disable();
 138         /*
 139          * Remove this CPU:
 140          */
 141         cpu_clear(smp_processor_id(), cpu_online_map);
 142         disable_local_APIC();
 143
 144         for (;;) {
 145                 if (hlt_works(smp_processor_id()))
 146                         halt();
 147         }
 148 }
 149
 150 static void do_nothing(void *unused)
 151 {
 152 }
 153
 154 /*
 155  * cpu_idle_wait - Used to ensure that all the CPUs discard old value of
 156  * pm_idle and update to new pm_idle value. Required while changing pm_idle
 157  * handler on SMP systems.
 158  *
 159  * Caller must have changed pm_idle to the new value before the call. Old
 160  * pm_idle value will not be used by any CPU after the return of this function.
 161  */
 162 void cpu_idle_wait(void)
 163 {
 164         smp_mb();
 165         /* kick all the CPUs so that they exit out of pm_idle */
 166         smp_call_function(do_nothing, NULL, 1);
 167 }
 168 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 169
 170 /*
 171  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
 172  * which can obviate IPI to trigger checking of need_resched.
 173  * We execute MONITOR against need_resched and enter optimized wait state
 174  * through MWAIT. Whenever someone changes need_resched, we would be woken
 175  * up from MWAIT (without an IPI).
 176  *
 177  * New with Core Duo processors, MWAIT can take some hints based on CPU
 178  * capability.
 179  */
 180 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
 181 {
 182         struct power_trace it;
 183
 184         trace_power_start(&it, POWER_CSTATE, (ax>>4)+1);
 185         if (!need_resched()) {
 186                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 187                 smp_mb();
 188                 if (!need_resched())
 189                         __mwait(ax, cx);
 190         }
 191         trace_power_end(&it);
 192 }
 193
 194 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
 195 static void mwait_idle(void)
 196 {
 197         struct power_trace it;
 198         if (!need_resched()) {
 199                 trace_power_start(&it, POWER_CSTATE, 1);
 200                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 201                 smp_mb();
 202                 if (!need_resched())
 203                         __sti_mwait(0, 0);
 204                 else
 205                         local_irq_enable();
 206                 trace_power_end(&it);
 207         } else
 208                 local_irq_enable();
 209 }
 210
 211 /*
 212  * On SMP it's slightly faster (but much more power-consuming!)
 213  * to poll the ->work.need_resched flag instead of waiting for the
 214  * cross-CPU IPI to arrive. Use this option with caution.
 215  */
 216 static void poll_idle(void)
 217 {
 218         struct power_trace it;
 219
 220         trace_power_start(&it, POWER_CSTATE, 0);
 221         local_irq_enable();
 222         while (!need_resched())
 223                 cpu_relax();
 224         trace_power_end(&it);
 225 }
 226
 227 /*
 228  * mwait selection logic:
 229  *
 230  * It depends on the CPU. For AMD CPUs that support MWAIT this is
 231  * wrong. Family 0x10 and 0x11 CPUs will enter C1 on HLT. Powersavings
 232  * then depend on a clock divisor and current Pstate of the core. If
 233  * all cores of a processor are in halt state (C1) the processor can
 234  * enter the C1E (C1 enhanced) state. If mwait is used this will never
 235  * happen.
 236  *
 237  * idle=mwait overrides this decision and forces the usage of mwait.
 238  */
 239 static int __cpuinitdata force_mwait;
 240
 241 #define MWAIT_INFO                      0x05
 242 #define MWAIT_ECX_EXTENDED_INFO         0x01
 243 #define MWAIT_EDX_C1                    0xf0
 244
 245 static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c)
 246 {
 247         u32 eax, ebx, ecx, edx;
 248
 249         if (force_mwait)
 250                 return 1;
 251
 252         if (c->cpuid_level < MWAIT_INFO)
 253                 return 0;
 254
 255         cpuid(MWAIT_INFO, &eax, &ebx, &ecx, &edx);
 256         /* Check, whether EDX has extended info about MWAIT */
 257         if (!(ecx & MWAIT_ECX_EXTENDED_INFO))
 258                 return 1;
 259
 260         /*
 261          * edx enumeratios MONITOR/MWAIT extensions. Check, whether
 262          * C1  supports MWAIT
 263          */
 264         return (edx & MWAIT_EDX_C1);
 265 }
 266
 267 /*
 268  * Check for AMD CPUs, which have potentially C1E support
 269  */
 270 static int __cpuinit check_c1e_idle(const struct cpuinfo_x86 *c)
 271 {
 272         if (c->x86_vendor != X86_VENDOR_AMD)
 273                 return 0;
 274
 275         if (c->x86 < 0x0F)
 276                 return 0;
 277
 278         /* Family 0x0f models < rev F do not have C1E */
 279         if (c->x86 == 0x0f && c->x86_model < 0x40)
 280                 return 0;
 281
 282         return 1;
 283 }
 284
 285 static cpumask_t c1e_mask = CPU_MASK_NONE;
 286 static int c1e_detected;
 287
 288 void c1e_remove_cpu(int cpu)
 289 {
 290         cpu_clear(cpu, c1e_mask);
 291 }
 292
 293 /*
 294  * C1E aware idle routine. We check for C1E active in the interrupt
 295  * pending message MSR. If we detect C1E, then we handle it the same
 296  * way as C3 power states (local apic timer and TSC stop)
 297  */
 298 static void c1e_idle(void)
 299 {
 300         if (need_resched())
 301                 return;
 302
 303         if (!c1e_detected) {
 304                 u32 lo, hi;
 305
 306                 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
 307                 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
 308                         c1e_detected = 1;
 309                         if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
 310                                 mark_tsc_unstable("TSC halt in AMD C1E");
 311                         printk(KERN_INFO "System has AMD C1E enabled\n");
 312                         set_cpu_cap(&boot_cpu_data, X86_FEATURE_AMDC1E);
 313                 }
 314         }
 315
 316         if (c1e_detected) {
 317                 int cpu = smp_processor_id();
 318
 319                 if (!cpu_isset(cpu, c1e_mask)) {
 320                         cpu_set(cpu, c1e_mask);
 321                         /*
 322                          * Force broadcast so ACPI can not interfere. Needs
 323                          * to run with interrupts enabled as it uses
 324                          * smp_function_call.
 325                          */
 326                         local_irq_enable();
 327                         clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
 328                                            &cpu);
 329                         printk(KERN_INFO "Switch to broadcast mode on CPU%d\n",
 330                                cpu);
 331                         local_irq_disable();
 332                 }
 333                 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
 334
 335                 default_idle();
 336
 337                 /*
 338                  * The switch back from broadcast mode needs to be
 339                  * called with interrupts disabled.
 340                  */
 341                  local_irq_disable();
 342                  clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
 343                  local_irq_enable();
 344         } else
 345                 default_idle();
 346 }
 347
 348 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
 349 {
 350 #ifdef CONFIG_X86_SMP
 351         if (pm_idle == poll_idle && smp_num_siblings > 1) {
 352                 printk(KERN_WARNING "WARNING: polling idle and HT enabled,"
 353                         " performance may degrade.\n");
 354         }
 355 #endif
 356         if (pm_idle)
 357                 return;
 358
 359         if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
 360                 /*
 361                  * One CPU supports mwait => All CPUs supports mwait
 362                  */
 363                 printk(KERN_INFO "using mwait in idle threads.\n");
 364                 pm_idle = mwait_idle;
 365         } else if (check_c1e_idle(c)) {
 366                 printk(KERN_INFO "using C1E aware idle routine\n");
 367                 pm_idle = c1e_idle;
 368         } else
 369                 pm_idle = default_idle;
 370 }
 371
 372 static int __init idle_setup(char *str)
 373 {
 374         if (!str)
 375                 return -EINVAL;
 376
 377         if (!strcmp(str, "poll")) {
 378                 printk("using polling idle threads.\n");
 379                 pm_idle = poll_idle;
 380         } else if (!strcmp(str, "mwait"))
 381                 force_mwait = 1;
 382         else if (!strcmp(str, "halt")) {
 383                 /*
 384                  * When the boot option of idle=halt is added, halt is
 385                  * forced to be used for CPU idle. In such case CPU C2/C3
 386                  * won't be used again.
 387                  * To continue to load the CPU idle driver, don't touch
 388                  * the boot_option_idle_override.
 389                  */
 390                 pm_idle = default_idle;
 391                 idle_halt = 1;
 392                 return 0;
 393         } else if (!strcmp(str, "nomwait")) {
 394                 /*
 395                  * If the boot option of "idle=nomwait" is added,
 396                  * it means that mwait will be disabled for CPU C2/C3
 397                  * states. In such case it won't touch the variable
 398                  * of boot_option_idle_override.
 399                  */
 400                 idle_nomwait = 1;
 401                 return 0;
 402         } else
 403                 return -1;
 404
 405         boot_option_idle_override = 1;
 406         return 0;
 407 }
 408 early_param("idle", idle_setup);
 409