]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/i386/machine_routines.c
xnu-3247.1.106.tar.gz
[apple/xnu.git] / osfmk / i386 / machine_routines.c
index 22eae015943d27947ef6bd4f438510ba4da5d3b3..a90d68178a80f6dfc3ea6372a097d9778cba0aff 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2000-2010 Apple Inc. All rights reserved.
+ * Copyright (c) 2000-2012 Apple Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
 #include <kern/cpu_data.h>
 #include <kern/cpu_number.h>
 #include <kern/thread.h>
+#include <kern/thread_call.h>
+#include <prng/random.h>
 #include <i386/machine_cpu.h>
 #include <i386/lapic.h>
-#include <i386/lock.h>
+#include <i386/bit_routines.h>
 #include <i386/mp_events.h>
 #include <i386/pmCPU.h>
 #include <i386/trap.h>
 #include <i386/pmap.h>
 #include <i386/pmap_internal.h>
 #include <i386/misc_protos.h>
+#include <kern/timer_queue.h>
+#if KPC
+#include <kern/kpc.h>
+#endif
+#include <architecture/i386/pio.h>
 
 #if DEBUG
 #define DBG(x...)      kprintf("DBG: " x)
@@ -61,6 +68,7 @@ extern void   wakeup(void *);
 static int max_cpus_initialized = 0;
 
 unsigned int   LockTimeOut;
+unsigned int   TLBTimeOut;
 unsigned int   LockTimeOutTSC;
 unsigned int   MutexSpin;
 uint64_t       LastDebuggerEntryAllowance;
@@ -70,6 +78,12 @@ extern uint64_t panic_restart_timeout;
 
 boolean_t virtualized = FALSE;
 
+decl_simple_lock_data(static,  ml_timer_evaluation_slock);
+uint32_t ml_timer_eager_evaluations;
+uint64_t ml_timer_eager_evaluation_max;
+static boolean_t ml_timer_evaluation_in_progress = FALSE;
+
+
 #define MAX_CPUS_SET    0x1
 #define MAX_CPUS_WAIT   0x2
 
@@ -139,14 +153,17 @@ ml_static_mfree(
                        }
                        pmap_remove(kernel_pmap, vaddr_cur, vaddr_cur+PAGE_SIZE);
                        assert(pmap_valid_page(ppn));
-
                        if (IS_MANAGED_PAGE(ppn)) {
                                vm_page_create(ppn,(ppn+1));
-                               vm_page_wire_count--;
                                freed_pages++;
                        }
                }
        }
+       vm_page_lockspin_queues();
+       vm_page_wire_count -= freed_pages;
+       vm_page_wire_count_initial -= freed_pages;
+       vm_page_unlock_queues();
+
 #if    DEBUG   
        kprintf("ml_static_mfree: Released 0x%x pages at VA %p, size:0x%llx, last ppn: 0x%x\n", freed_pages, (void *)vaddr, (uint64_t)size, ppn);
 #endif
@@ -200,6 +217,38 @@ vm_size_t ml_nofault_copy(
        return nbytes;
 }
 
+/*
+ *     Routine:        ml_validate_nofault
+ *     Function: Validate that ths address range has a valid translations
+ *                     in the kernel pmap.  If translations are present, they are
+ *                     assumed to be wired; i.e. no attempt is made to guarantee
+ *                     that the translation persist after the check.
+ *  Returns: TRUE if the range is mapped and will not cause a fault,
+ *                     FALSE otherwise.
+ */
+
+boolean_t ml_validate_nofault(
+       vm_offset_t virtsrc, vm_size_t size)
+{
+       addr64_t cur_phys_src;
+       uint32_t count;
+
+       while (size > 0) {
+               if (!(cur_phys_src = kvtophys(virtsrc)))
+                       return FALSE;
+               if (!pmap_valid_page(i386_btop(cur_phys_src)))
+                       return FALSE;
+               count = (uint32_t)(PAGE_SIZE - (cur_phys_src & PAGE_MASK));
+               if (count > size)
+                       count = (uint32_t)size;
+
+               virtsrc += count;
+               size -= count;
+       }
+
+       return TRUE;
+}
+
 /* Interrupt handling */
 
 /* Initialize Interrupts */
@@ -226,13 +275,15 @@ boolean_t ml_set_interrupts_enabled(boolean_t enable)
        
        __asm__ volatile("pushf; pop    %0" :  "=r" (flags));
 
+       assert(get_interrupt_level() ? (enable == FALSE) : TRUE);
+
        istate = ((flags & EFL_IF) != 0);
 
        if (enable) {
                __asm__ volatile("sti;nop");
 
                if ((get_preemption_level() == 0) && (*ast_pending() & AST_URGENT))
-                       __asm__ volatile ("int $0xff");
+                       __asm__ volatile ("int %0" :: "N" (T_PREEMPT));
        }
        else {
                if (istate)
@@ -248,26 +299,40 @@ boolean_t ml_at_interrupt_context(void)
        return get_interrupt_level() != 0;
 }
 
+void ml_get_power_state(boolean_t *icp, boolean_t *pidlep) {
+       *icp = (get_interrupt_level() != 0);
+       /* These will be technically inaccurate for interrupts that occur
+        * successively within a single "idle exit" event, but shouldn't
+        * matter statistically.
+        */
+       *pidlep = (current_cpu_datap()->lcpu.package->num_idle == topoParms.nLThreadsPerPackage);
+}
+
 /* Generate a fake interrupt */
 void ml_cause_interrupt(void)
 {
        panic("ml_cause_interrupt not defined yet on Intel");
 }
 
+/*
+ * TODO: transition users of this to kernel_thread_start_priority
+ * ml_thread_policy is an unsupported KPI
+ */
 void ml_thread_policy(
        thread_t thread,
 __unused       unsigned policy_id,
        unsigned policy_info)
 {
        if (policy_info & MACHINE_NETWORK_WORKLOOP) {
-               spl_t           s = splsched();
+               thread_precedence_policy_data_t info;
+               __assert_only kern_return_t kret;
 
-               thread_lock(thread);
+               info.importance = 1;
 
-               set_priority(thread, thread->priority + 1);
-
-               thread_unlock(thread);
-               splx(s);
+               kret = thread_policy_set_internal(thread, THREAD_PRECEDENCE_POLICY,
+                                                               (thread_policy_t)&info,
+                                                               THREAD_PRECEDENCE_POLICY_COUNT);
+               assert(kret == KERN_SUCCESS);
        }
 }
 
@@ -299,6 +364,20 @@ machine_signal_idle(
        cpu_interrupt(processor->cpu_id);
 }
 
+void
+machine_signal_idle_deferred(
+       __unused processor_t processor)
+{
+       panic("Unimplemented");
+}
+
+void
+machine_signal_idle_cancel(
+       __unused processor_t processor)
+{
+       panic("Unimplemented");
+}
+
 static kern_return_t
 register_cpu(
         uint32_t        lapic_id,
@@ -331,6 +410,11 @@ register_cpu(
        if (this_cpu_datap->cpu_chud == NULL)
                goto failed;
 
+#if KPC
+       if (kpc_register_cpu(this_cpu_datap) != TRUE)
+               goto failed;
+#endif
+
        if (!boot_cpu) {
                cpu_thread_alloc(this_cpu_datap->cpu_number);
                if (this_cpu_datap->lcpu.core == NULL)
@@ -363,6 +447,13 @@ failed:
 #endif
        chudxnu_cpu_free(this_cpu_datap->cpu_chud);
        console_cpu_free(this_cpu_datap->cpu_console_buf);
+#if KPC
+       kpc_counterbuf_free(this_cpu_datap->cpu_kpc_buf[0]);
+       kpc_counterbuf_free(this_cpu_datap->cpu_kpc_buf[1]);
+       kpc_counterbuf_free(this_cpu_datap->cpu_kpc_shadow);
+       kpc_counterbuf_free(this_cpu_datap->cpu_kpc_reload);
+#endif
+
        return KERN_FAILURE;
 }
 
@@ -412,6 +503,12 @@ ml_processor_register(
     /* fix the CPU id */
     this_cpu_datap->cpu_id = cpu_id;
 
+    /* allocate and initialize other per-cpu structures */
+    if (!boot_cpu) {
+       mp_cpus_call_cpu_init(cpunum);
+       prng_cpu_init(cpunum);
+    }
+
     /* output arg */
     *processor_out = this_cpu_datap->cpu_processor;
 
@@ -525,7 +622,11 @@ ml_init_lock_timeout(void)
 {
        uint64_t        abstime;
        uint32_t        mtxspin;
+#if DEVELOPMENT || DEBUG
        uint64_t        default_timeout_ns = NSEC_PER_SEC>>2;
+#else
+       uint64_t        default_timeout_ns = NSEC_PER_SEC>>1;
+#endif
        uint32_t        slto;
        uint32_t        prt;
 
@@ -537,6 +638,26 @@ ml_init_lock_timeout(void)
        LockTimeOut = (uint32_t) abstime;
        LockTimeOutTSC = (uint32_t) tmrCvt(abstime, tscFCvtn2t);
 
+       /*
+        * TLBTimeOut dictates the TLB flush timeout period. It defaults to
+        * LockTimeOut but can be overriden separately. In particular, a
+        * zero value inhibits the timeout-panic and cuts a trace evnt instead
+        * - see pmap_flush_tlbs().
+        */
+       if (PE_parse_boot_argn("tlbto_us", &slto, sizeof (slto))) {
+               default_timeout_ns = slto * NSEC_PER_USEC;
+               nanoseconds_to_absolutetime(default_timeout_ns, &abstime);
+               TLBTimeOut = (uint32_t) abstime;
+       } else {
+               TLBTimeOut = LockTimeOut;
+       }
+
+       if (PE_parse_boot_argn("phyreadmaxus", &slto, sizeof (slto))) {
+               default_timeout_ns = slto * NSEC_PER_USEC;
+               nanoseconds_to_absolutetime(default_timeout_ns, &abstime);
+               reportphyreaddelayabs = abstime;
+       }
+
        if (PE_parse_boot_argn("mtxspin", &mtxspin, sizeof (mtxspin))) {
                if (mtxspin > USEC_PER_SEC>>4)
                        mtxspin =  USEC_PER_SEC>>4;
@@ -551,16 +672,18 @@ ml_init_lock_timeout(void)
                nanoseconds_to_absolutetime(prt * NSEC_PER_SEC, &panic_restart_timeout);
        virtualized = ((cpuid_features() & CPUID_FEATURE_VMM) != 0);
        interrupt_latency_tracker_setup();
+       simple_lock_init(&ml_timer_evaluation_slock, 0);
 }
 
 /*
  * Threshold above which we should attempt to block
  * instead of spinning for clock_delay_until().
  */
+
 void
-ml_init_delay_spin_threshold(void)
+ml_init_delay_spin_threshold(int threshold_us)
 {
-       nanoseconds_to_absolutetime(10ULL * NSEC_PER_USEC, &delay_spin_threshold);
+       nanoseconds_to_absolutetime(threshold_us * NSEC_PER_USEC, &delay_spin_threshold);
 }
 
 boolean_t
@@ -570,7 +693,7 @@ ml_delay_should_spin(uint64_t interval)
 }
 
 /*
- * This is called from the machine-independent routine cpu_up()
+ * This is called from the machine-independent layer
  * to perform machine-dependent info updates. Defer to cpu_thread_init().
  */
 void
@@ -580,12 +703,14 @@ ml_cpu_up(void)
 }
 
 /*
- * This is called from the machine-independent routine cpu_down()
+ * This is called from the machine-independent layer
  * to perform machine-dependent info updates.
  */
 void
 ml_cpu_down(void)
 {
+       i386_deactivate_cpu();
+
        return;
 }
 
@@ -636,17 +761,7 @@ void ml_cpu_set_ldt(int selector)
            current_cpu_datap()->cpu_ldt == KERNEL_LDT)
                return;
 
-#if defined(__i386__)
-       /*
-        * If 64bit this requires a mode switch (and back). 
-        */
-       if (cpu_mode_is64bit())
-               ml_64bit_lldt(selector);
-       else
-               lldt(selector);
-#else
        lldt(selector);
-#endif
        current_cpu_datap()->cpu_ldt = selector;
 }
 
@@ -697,5 +812,64 @@ kernel_preempt_check(void)
 }
 
 boolean_t machine_timeout_suspended(void) {
-       return (virtualized || pmap_tlb_flush_timeout || spinlock_timed_out || panic_active() || mp_recent_debugger_activity());
+       return (virtualized || pmap_tlb_flush_timeout || spinlock_timed_out || panic_active() || mp_recent_debugger_activity() || ml_recent_wake());
+}
+
+/* Eagerly evaluate all pending timer and thread callouts
+ */
+void ml_timer_evaluate(void) {
+       KERNEL_DEBUG_CONSTANT(DECR_TIMER_RESCAN|DBG_FUNC_START, 0, 0, 0, 0, 0);
+
+       uint64_t te_end, te_start = mach_absolute_time();
+       simple_lock(&ml_timer_evaluation_slock);
+       ml_timer_evaluation_in_progress = TRUE;
+       thread_call_delayed_timer_rescan_all();
+       mp_cpus_call(CPUMASK_ALL, ASYNC, timer_queue_expire_rescan, NULL);
+       ml_timer_evaluation_in_progress = FALSE;
+       ml_timer_eager_evaluations++;
+       te_end = mach_absolute_time();
+       ml_timer_eager_evaluation_max = MAX(ml_timer_eager_evaluation_max, (te_end - te_start));
+       simple_unlock(&ml_timer_evaluation_slock);
+
+       KERNEL_DEBUG_CONSTANT(DECR_TIMER_RESCAN|DBG_FUNC_END, 0, 0, 0, 0, 0);
+}
+
+boolean_t
+ml_timer_forced_evaluation(void) {
+       return ml_timer_evaluation_in_progress;
+}
+
+/* 32-bit right-rotate n bits */
+static inline uint32_t ror32(uint32_t val, const unsigned int n)
+{      
+       __asm__ volatile("rorl %%cl,%0" : "=r" (val) : "0" (val), "c" (n));
+       return val;
+}
+
+void
+ml_entropy_collect(void)
+{
+       uint32_t        tsc_lo, tsc_hi;
+       uint32_t        *ep;
+
+       assert(cpu_number() == master_cpu);
+
+       /* update buffer pointer cyclically */
+       if (EntropyData.index_ptr - EntropyData.buffer == ENTROPY_BUFFER_SIZE)
+               ep = EntropyData.index_ptr = EntropyData.buffer;
+       else
+               ep = EntropyData.index_ptr++;
+
+       rdtsc_nofence(tsc_lo, tsc_hi);
+       *ep = ror32(*ep, 9) ^ tsc_lo;
+}
+
+void
+ml_gpu_stat_update(uint64_t gpu_ns_delta) {
+       current_thread()->machine.thread_gpu_ns += gpu_ns_delta;
+}
+
+uint64_t
+ml_gpu_stat(thread_t t) {
+       return t->machine.thread_gpu_ns;
 }