Merge commit 'v2.6.26' into x86/core
[pandora-kernel.git] / arch / x86 / kernel / process.c
index 67e9b4a..4061d63 100644 (file)
@@ -6,6 +6,7 @@
 #include <linux/sched.h>
 #include <linux/module.h>
 #include <linux/pm.h>
+#include <linux/clockchips.h>
 
 struct kmem_cache *task_xstate_cachep;
 
@@ -45,6 +46,76 @@ void arch_task_cache_init(void)
                                  SLAB_PANIC, NULL);
 }
 
+/*
+ * Idle related variables and functions
+ */
+unsigned long boot_option_idle_override = 0;
+EXPORT_SYMBOL(boot_option_idle_override);
+
+/*
+ * Powermanagement idle function, if any..
+ */
+void (*pm_idle)(void);
+EXPORT_SYMBOL(pm_idle);
+
+#ifdef CONFIG_X86_32
+/*
+ * This halt magic was a workaround for ancient floppy DMA
+ * wreckage. It should be safe to remove.
+ */
+static int hlt_counter;
+void disable_hlt(void)
+{
+       hlt_counter++;
+}
+EXPORT_SYMBOL(disable_hlt);
+
+void enable_hlt(void)
+{
+       hlt_counter--;
+}
+EXPORT_SYMBOL(enable_hlt);
+
+static inline int hlt_use_halt(void)
+{
+       return (!hlt_counter && boot_cpu_data.hlt_works_ok);
+}
+#else
+static inline int hlt_use_halt(void)
+{
+       return 1;
+}
+#endif
+
+/*
+ * We use this if we don't have any better
+ * idle routine..
+ */
+void default_idle(void)
+{
+       if (hlt_use_halt()) {
+               current_thread_info()->status &= ~TS_POLLING;
+               /*
+                * TS_POLLING-cleared state must be visible before we
+                * test NEED_RESCHED:
+                */
+               smp_mb();
+
+               if (!need_resched())
+                       safe_halt();    /* enables interrupts racelessly */
+               else
+                       local_irq_enable();
+               current_thread_info()->status |= TS_POLLING;
+       } else {
+               local_irq_enable();
+               /* loop is done by the caller */
+               cpu_relax();
+       }
+}
+#ifdef CONFIG_APM_MODULE
+EXPORT_SYMBOL(default_idle);
+#endif
+
 static void do_nothing(void *unused)
 {
 }
@@ -99,49 +170,152 @@ static void mwait_idle(void)
                local_irq_enable();
 }
 
+/*
+ * On SMP it's slightly faster (but much more power-consuming!)
+ * to poll the ->work.need_resched flag instead of waiting for the
+ * cross-CPU IPI to arrive. Use this option with caution.
+ */
+static void poll_idle(void)
+{
+       local_irq_enable();
+       cpu_relax();
+}
+
+/*
+ * mwait selection logic:
+ *
+ * It depends on the CPU. For AMD CPUs that support MWAIT this is
+ * wrong. Family 0x10 and 0x11 CPUs will enter C1 on HLT. Powersavings
+ * then depend on a clock divisor and current Pstate of the core. If
+ * all cores of a processor are in halt state (C1) the processor can
+ * enter the C1E (C1 enhanced) state. If mwait is used this will never
+ * happen.
+ *
+ * idle=mwait overrides this decision and forces the usage of mwait.
+ */
+
+#define MWAIT_INFO                     0x05
+#define MWAIT_ECX_EXTENDED_INFO                0x01
+#define MWAIT_EDX_C1                   0xf0
 
 static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c)
 {
+       u32 eax, ebx, ecx, edx;
+
        if (force_mwait)
                return 1;
-       /* Any C1 states supported? */
-       return c->cpuid_level >= 5 && ((cpuid_edx(5) >> 4) & 0xf) > 0;
+
+       if (c->cpuid_level < MWAIT_INFO)
+               return 0;
+
+       cpuid(MWAIT_INFO, &eax, &ebx, &ecx, &edx);
+       /* Check, whether EDX has extended info about MWAIT */
+       if (!(ecx & MWAIT_ECX_EXTENDED_INFO))
+               return 1;
+
+       /*
+        * edx enumeratios MONITOR/MWAIT extensions. Check, whether
+        * C1  supports MWAIT
+        */
+       return (edx & MWAIT_EDX_C1);
 }
 
 /*
- * On SMP it's slightly faster (but much more power-consuming!)
- * to poll the ->work.need_resched flag instead of waiting for the
- * cross-CPU IPI to arrive. Use this option with caution.
+ * Check for AMD CPUs, which have potentially C1E support
  */
-static void poll_idle(void)
+static int __cpuinit check_c1e_idle(const struct cpuinfo_x86 *c)
 {
-       local_irq_enable();
-       cpu_relax();
+       if (c->x86_vendor != X86_VENDOR_AMD)
+               return 0;
+
+       if (c->x86 < 0x0F)
+               return 0;
+
+       /* Family 0x0f models < rev F do not have C1E */
+       if (c->x86 == 0x0f && c->x86_model < 0x40)
+               return 0;
+
+       return 1;
 }
 
-void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
+/*
+ * C1E aware idle routine. We check for C1E active in the interrupt
+ * pending message MSR. If we detect C1E, then we handle it the same
+ * way as C3 power states (local apic timer and TSC stop)
+ */
+static void c1e_idle(void)
 {
-       static int selected;
+       static cpumask_t c1e_mask = CPU_MASK_NONE;
+       static int c1e_detected;
 
-       if (selected)
+       if (need_resched())
                return;
+
+       if (!c1e_detected) {
+               u32 lo, hi;
+
+               rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
+               if (lo & K8_INTP_C1E_ACTIVE_MASK) {
+                       c1e_detected = 1;
+                       mark_tsc_unstable("TSC halt in C1E");
+                       printk(KERN_INFO "System has C1E enabled\n");
+               }
+       }
+
+       if (c1e_detected) {
+               int cpu = smp_processor_id();
+
+               if (!cpu_isset(cpu, c1e_mask)) {
+                       cpu_set(cpu, c1e_mask);
+                       /*
+                        * Force broadcast so ACPI can not interfere. Needs
+                        * to run with interrupts enabled as it uses
+                        * smp_function_call.
+                        */
+                       local_irq_enable();
+                       clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
+                                          &cpu);
+                       printk(KERN_INFO "Switch to broadcast mode on CPU%d\n",
+                              cpu);
+                       local_irq_disable();
+               }
+               clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
+
+               default_idle();
+
+               /*
+                * The switch back from broadcast mode needs to be
+                * called with interrupts disabled.
+                */
+                local_irq_disable();
+                clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
+                local_irq_enable();
+       } else
+               default_idle();
+}
+
+void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
+{
 #ifdef CONFIG_X86_SMP
        if (pm_idle == poll_idle && smp_num_siblings > 1) {
                printk(KERN_WARNING "WARNING: polling idle and HT enabled,"
                        " performance may degrade.\n");
        }
 #endif
+       if (pm_idle)
+               return;
+
        if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
                /*
-                * Skip, if setup has overridden idle.
                 * One CPU supports mwait => All CPUs supports mwait
                 */
-               if (!pm_idle) {
-                       printk(KERN_INFO "using mwait in idle threads.\n");
-                       pm_idle = mwait_idle;
-               }
-       }
-       selected = 1;
+               printk(KERN_INFO "using mwait in idle threads.\n");
+               pm_idle = mwait_idle;
+       } else if (check_c1e_idle(c)) {
+               printk(KERN_INFO "using C1E aware idle routine\n");
+               pm_idle = c1e_idle;
+       } else
+               pm_idle = default_idle;
 }
 
 static int __init idle_setup(char *str)