| 1 | #include <linux/errno.h> |
| 2 | #include <linux/kernel.h> |
| 3 | #include <linux/mm.h> |
| 4 | #include <linux/smp.h> |
| 5 | #include <linux/slab.h> |
| 6 | #include <linux/sched.h> |
| 7 | #include <linux/module.h> |
| 8 | #include <linux/pm.h> |
| 9 | |
| 10 | struct kmem_cache *task_xstate_cachep; |
| 11 | |
| 12 | int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src) |
| 13 | { |
| 14 | *dst = *src; |
| 15 | if (src->thread.xstate) { |
| 16 | dst->thread.xstate = kmem_cache_alloc(task_xstate_cachep, |
| 17 | GFP_KERNEL); |
| 18 | if (!dst->thread.xstate) |
| 19 | return -ENOMEM; |
| 20 | WARN_ON((unsigned long)dst->thread.xstate & 15); |
| 21 | memcpy(dst->thread.xstate, src->thread.xstate, xstate_size); |
| 22 | } |
| 23 | return 0; |
| 24 | } |
| 25 | |
| 26 | void free_thread_xstate(struct task_struct *tsk) |
| 27 | { |
| 28 | if (tsk->thread.xstate) { |
| 29 | kmem_cache_free(task_xstate_cachep, tsk->thread.xstate); |
| 30 | tsk->thread.xstate = NULL; |
| 31 | } |
| 32 | } |
| 33 | |
| 34 | void free_thread_info(struct thread_info *ti) |
| 35 | { |
| 36 | free_thread_xstate(ti->task); |
| 37 | free_pages((unsigned long)ti, get_order(THREAD_SIZE)); |
| 38 | } |
| 39 | |
| 40 | void arch_task_cache_init(void) |
| 41 | { |
| 42 | task_xstate_cachep = |
| 43 | kmem_cache_create("task_xstate", xstate_size, |
| 44 | __alignof__(union thread_xstate), |
| 45 | SLAB_PANIC, NULL); |
| 46 | } |
| 47 | |
| 48 | static void do_nothing(void *unused) |
| 49 | { |
| 50 | } |
| 51 | |
| 52 | /* |
| 53 | * cpu_idle_wait - Used to ensure that all the CPUs discard old value of |
| 54 | * pm_idle and update to new pm_idle value. Required while changing pm_idle |
| 55 | * handler on SMP systems. |
| 56 | * |
| 57 | * Caller must have changed pm_idle to the new value before the call. Old |
| 58 | * pm_idle value will not be used by any CPU after the return of this function. |
| 59 | */ |
| 60 | void cpu_idle_wait(void) |
| 61 | { |
| 62 | smp_mb(); |
| 63 | /* kick all the CPUs so that they exit out of pm_idle */ |
| 64 | smp_call_function(do_nothing, NULL, 0, 1); |
| 65 | } |
| 66 | EXPORT_SYMBOL_GPL(cpu_idle_wait); |
| 67 | |
| 68 | /* |
| 69 | * This uses new MONITOR/MWAIT instructions on P4 processors with PNI, |
| 70 | * which can obviate IPI to trigger checking of need_resched. |
| 71 | * We execute MONITOR against need_resched and enter optimized wait state |
| 72 | * through MWAIT. Whenever someone changes need_resched, we would be woken |
| 73 | * up from MWAIT (without an IPI). |
| 74 | * |
| 75 | * New with Core Duo processors, MWAIT can take some hints based on CPU |
| 76 | * capability. |
| 77 | */ |
| 78 | void mwait_idle_with_hints(unsigned long ax, unsigned long cx) |
| 79 | { |
| 80 | if (!need_resched()) { |
| 81 | __monitor((void *)¤t_thread_info()->flags, 0, 0); |
| 82 | smp_mb(); |
| 83 | if (!need_resched()) |
| 84 | __mwait(ax, cx); |
| 85 | } |
| 86 | } |
| 87 | |
| 88 | /* Default MONITOR/MWAIT with no hints, used for default C1 state */ |
| 89 | static void mwait_idle(void) |
| 90 | { |
| 91 | if (!need_resched()) { |
| 92 | __monitor((void *)¤t_thread_info()->flags, 0, 0); |
| 93 | smp_mb(); |
| 94 | if (!need_resched()) |
| 95 | __sti_mwait(0, 0); |
| 96 | else |
| 97 | local_irq_enable(); |
| 98 | } else |
| 99 | local_irq_enable(); |
| 100 | } |
| 101 | |
| 102 | /* |
| 103 | * On SMP it's slightly faster (but much more power-consuming!) |
| 104 | * to poll the ->work.need_resched flag instead of waiting for the |
| 105 | * cross-CPU IPI to arrive. Use this option with caution. |
| 106 | */ |
| 107 | static void poll_idle(void) |
| 108 | { |
| 109 | local_irq_enable(); |
| 110 | cpu_relax(); |
| 111 | } |
| 112 | |
| 113 | /* |
| 114 | * mwait selection logic: |
| 115 | * |
| 116 | * It depends on the CPU. For AMD CPUs that support MWAIT this is |
| 117 | * wrong. Family 0x10 and 0x11 CPUs will enter C1 on HLT. Powersavings |
| 118 | * then depend on a clock divisor and current Pstate of the core. If |
| 119 | * all cores of a processor are in halt state (C1) the processor can |
| 120 | * enter the C1E (C1 enhanced) state. If mwait is used this will never |
| 121 | * happen. |
| 122 | * |
| 123 | * idle=mwait overrides this decision and forces the usage of mwait. |
| 124 | */ |
| 125 | |
| 126 | #define MWAIT_INFO 0x05 |
| 127 | #define MWAIT_ECX_EXTENDED_INFO 0x01 |
| 128 | #define MWAIT_EDX_C1 0xf0 |
| 129 | |
| 130 | static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c) |
| 131 | { |
| 132 | u32 eax, ebx, ecx, edx; |
| 133 | |
| 134 | if (force_mwait) |
| 135 | return 1; |
| 136 | |
| 137 | if (c->cpuid_level < MWAIT_INFO) |
| 138 | return 0; |
| 139 | |
| 140 | cpuid(MWAIT_INFO, &eax, &ebx, &ecx, &edx); |
| 141 | /* Check, whether EDX has extended info about MWAIT */ |
| 142 | if (!(ecx & MWAIT_ECX_EXTENDED_INFO)) |
| 143 | return 1; |
| 144 | |
| 145 | /* |
| 146 | * edx enumeratios MONITOR/MWAIT extensions. Check, whether |
| 147 | * C1 supports MWAIT |
| 148 | */ |
| 149 | return (edx & MWAIT_EDX_C1); |
| 150 | } |
| 151 | |
| 152 | void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c) |
| 153 | { |
| 154 | #ifdef CONFIG_X86_SMP |
| 155 | if (pm_idle == poll_idle && smp_num_siblings > 1) { |
| 156 | printk(KERN_WARNING "WARNING: polling idle and HT enabled," |
| 157 | " performance may degrade.\n"); |
| 158 | } |
| 159 | #endif |
| 160 | if (pm_idle) |
| 161 | return; |
| 162 | |
| 163 | if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) { |
| 164 | /* |
| 165 | * One CPU supports mwait => All CPUs supports mwait |
| 166 | */ |
| 167 | printk(KERN_INFO "using mwait in idle threads.\n"); |
| 168 | pm_idle = mwait_idle; |
| 169 | } else |
| 170 | pm_idle = default_idle; |
| 171 | } |
| 172 | |
| 173 | static int __init idle_setup(char *str) |
| 174 | { |
| 175 | if (!strcmp(str, "poll")) { |
| 176 | printk("using polling idle threads.\n"); |
| 177 | pm_idle = poll_idle; |
| 178 | } else if (!strcmp(str, "mwait")) |
| 179 | force_mwait = 1; |
| 180 | else |
| 181 | return -1; |
| 182 | |
| 183 | boot_option_idle_override = 1; |
| 184 | return 0; |
| 185 | } |
| 186 | early_param("idle", idle_setup); |
| 187 | |