]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/i386/cpu_data.h
xnu-4903.221.2.tar.gz
[apple/xnu.git] / osfmk / i386 / cpu_data.h
index e4ba6fd33b09e679660d56c63c914d7bd7a19c4a..4201068f4d7be6506c9125121493e2e789bcc6ae 100644 (file)
@@ -1,16 +1,19 @@
 /*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2012 Apple Inc. All rights reserved.
  *
- * @APPLE_LICENSE_HEADER_START@
- * 
- * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in
- * compliance with the License. Please obtain a copy of the License at
- * http://www.opensource.apple.com/apsl/ and read it before using this
- * file.
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
+ * 
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
  * 
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
@@ -20,7 +23,7 @@
  * Please see the License for the specific language governing rights and
  * limitations under the License.
  * 
- * @APPLE_LICENSE_HEADER_END@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /*
  * @OSF_COPYRIGHT@
 #ifndef        I386_CPU_DATA
 #define I386_CPU_DATA
 
-#include <cpus.h>
 #include <mach_assert.h>
 
-#if    defined(__GNUC__)
-
 #include <kern/assert.h>
 #include <kern/kern_types.h>
+#include <kern/queue.h>
+#include <kern/processor.h>
+#include <kern/pms.h>
+#include <pexpert/pexpert.h>
+#include <mach/i386/thread_status.h>
+#include <mach/i386/vm_param.h>
+#include <i386/locks.h>
+#include <i386/rtclock_protos.h>
+#include <i386/pmCPU.h>
+#include <i386/cpu_topology.h>
+#include <i386/seg.h>
+
+#if CONFIG_VMX
+#include <i386/vmx/vmx_cpu.h>
+#endif
+
+#if MONOTONIC
+#include <machine/monotonic.h>
+#endif /* MONOTONIC */
+
+#include <machine/pal_routines.h>
+
+/*
+ * Data structures referenced (anonymously) from per-cpu data:
+ */
+struct cpu_cons_buffer;
+struct cpu_desc_table;
+struct mca_state;
+struct prngContext;
+
+/*
+ * Data structures embedded in per-cpu data:
+ */
+typedef struct rtclock_timer {
+       mpqueue_head_t          queue;
+       uint64_t                deadline;
+       uint64_t                when_set;
+       boolean_t               has_expired;
+} rtclock_timer_t;
+
+typedef struct {
+       /* The 'u' suffixed fields store the double-mapped descriptor addresses */
+       struct x86_64_tss       *cdi_ktssu;
+       struct x86_64_tss       *cdi_ktssb;
+       x86_64_desc_register_t  cdi_gdtu;
+       x86_64_desc_register_t  cdi_gdtb;
+       x86_64_desc_register_t  cdi_idtu;
+       x86_64_desc_register_t  cdi_idtb;
+       struct fake_descriptor  *cdi_ldtu;
+       struct fake_descriptor  *cdi_ldtb;
+       vm_offset_t             cdi_sstku;
+       vm_offset_t             cdi_sstkb;
+} cpu_desc_index_t;
+
+typedef enum {
+       TASK_MAP_32BIT,                 /* 32-bit user, compatibility mode */ 
+       TASK_MAP_64BIT,                 /* 64-bit user thread, shared space */ 
+} task_map_t;
+
+
+/*
+ * This structure is used on entry into the (uber-)kernel on syscall from
+ * a 64-bit user. It contains the address of the machine state save area
+ * for the current thread and a temporary place to save the user's rsp
+ * before loading this address into rsp.
+ */
+typedef struct {
+       addr64_t        cu_isf;         /* thread->pcb->iss.isf */
+       uint64_t        cu_tmp;         /* temporary scratch */ 
+       addr64_t        cu_user_gs_base;
+} cpu_uber_t;
+
+typedef        uint16_t        pcid_t;
+typedef        uint8_t         pcid_ref_t;
+
+#define CPU_RTIME_BINS (12)
+#define CPU_ITIME_BINS (CPU_RTIME_BINS)
+
+#define MAXPLFRAMES (16)
+typedef struct {
+       boolean_t pltype;
+       int plevel;
+       uint64_t plbt[MAXPLFRAMES];
+} plrecord_t;
 
-#if 0
-#ifndef        __OPTIMIZE__
-#define extern static
+/*
+ * Per-cpu data.
+ *
+ * Each processor has a per-cpu data area which is dereferenced through the
+ * current_cpu_datap() macro. For speed, the %gs segment is based here, and
+ * using this, inlines provides single-instruction access to frequently used
+ * members - such as get_cpu_number()/cpu_number(), and get_active_thread()/
+ * current_thread(). 
+ * 
+ * Cpu data owned by another processor can be accessed using the
+ * cpu_datap(cpu_number) macro which uses the cpu_data_ptr[] array of per-cpu
+ * pointers.
+ */
+typedef struct {
+       pcid_t                  cpu_pcid_free_hint;
+#define        PMAP_PCID_MAX_PCID      (0x800)
+       pcid_ref_t              cpu_pcid_refcounts[PMAP_PCID_MAX_PCID];
+       pmap_t                  cpu_pcid_last_pmap_dispatched[PMAP_PCID_MAX_PCID];
+} pcid_cdata_t;
+
+typedef struct cpu_data
+{
+       struct pal_cpu_data     cpu_pal_data;           /* PAL-specific data */
+#define                                cpu_pd cpu_pal_data     /* convenience alias */
+       struct cpu_data         *cpu_this;              /* pointer to myself */
+       thread_t                cpu_active_thread;
+       thread_t                cpu_nthread;
+       volatile int            cpu_preemption_level;
+       int                     cpu_number;             /* Logical CPU */
+       void                    *cpu_int_state;         /* interrupt state */
+       vm_offset_t             cpu_active_stack;       /* kernel stack base */
+       vm_offset_t             cpu_kernel_stack;       /* kernel stack top */
+       vm_offset_t             cpu_int_stack_top;
+       int                     cpu_interrupt_level;
+       volatile int            cpu_signals;            /* IPI events */
+       volatile int            cpu_prior_signals;      /* Last set of events,
+                                                        * debugging
+                                                        */
+       ast_t                   cpu_pending_ast;
+       volatile int            cpu_running;
+#if !MONOTONIC
+       boolean_t               cpu_fixed_pmcs_enabled;
+#endif /* !MONOTONIC */
+       rtclock_timer_t         rtclock_timer;
+       uint64_t                quantum_timer_deadline;
+       volatile addr64_t       cpu_active_cr3 __attribute((aligned(64)));
+       union {
+               volatile uint32_t cpu_tlb_invalid;
+               struct {
+                       volatile uint16_t cpu_tlb_invalid_local;
+                       volatile uint16_t cpu_tlb_invalid_global;
+               };
+       };
+       volatile task_map_t     cpu_task_map;
+       volatile addr64_t       cpu_task_cr3;
+       addr64_t                cpu_kernel_cr3;
+       volatile addr64_t       cpu_ucr3;
+       boolean_t               cpu_pagezero_mapped;
+       cpu_uber_t              cpu_uber;
+/* Double-mapped per-CPU exception stack address */
+       uintptr_t               cd_estack;
+       int                     cpu_xstate;
+/* Address of shadowed, partially mirrored CPU data structures located
+ * in the double mapped PML4
+ */
+       void                    *cd_shadow;
+       struct processor        *cpu_processor;
+#if NCOPY_WINDOWS > 0
+       struct cpu_pmap         *cpu_pmap;
 #endif
+       struct real_descriptor  *cpu_ldtp;
+       struct cpu_desc_table   *cpu_desc_tablep;
+       cpu_desc_index_t        cpu_desc_index;
+       int                     cpu_ldt;
+#if NCOPY_WINDOWS > 0
+       vm_offset_t             cpu_copywindow_base;
+       uint64_t                *cpu_copywindow_pdp;
+
+       vm_offset_t             cpu_physwindow_base;
+       uint64_t                *cpu_physwindow_ptep;
 #endif
 
-extern cpu_data_t      cpu_data[NCPUS];  
+#define HWINTCNT_SIZE 256
+       uint32_t                cpu_hwIntCnt[HWINTCNT_SIZE];    /* Interrupt counts */
+       uint64_t                cpu_hwIntpexits[HWINTCNT_SIZE];
+       uint64_t                cpu_dr7; /* debug control register */
+       uint64_t                cpu_int_event_time;     /* intr entry/exit time */
+       pal_rtc_nanotime_t      *cpu_nanotime;          /* Nanotime info */
+#if KPC
+       /* double-buffered performance counter data */
+       uint64_t                *cpu_kpc_buf[2];
+       /* PMC shadow and reload value buffers */
+       uint64_t                *cpu_kpc_shadow;
+       uint64_t                *cpu_kpc_reload;
+#endif
+#if MONOTONIC
+       struct mt_cpu cpu_monotonic;
+#endif /* MONOTONIC */
+       uint32_t                cpu_pmap_pcid_enabled;
+       pcid_t                  cpu_active_pcid;
+       pcid_t                  cpu_last_pcid;
+       pcid_t                  cpu_kernel_pcid;
+       volatile pcid_ref_t     *cpu_pmap_pcid_coherentp;
+       volatile pcid_ref_t     *cpu_pmap_pcid_coherentp_kernel;
+       pcid_cdata_t            *cpu_pcid_data;
+#ifdef PCID_STATS
+       uint64_t                cpu_pmap_pcid_flushes;
+       uint64_t                cpu_pmap_pcid_preserves;
+#endif
+       uint64_t                cpu_aperf;
+       uint64_t                cpu_mperf;
+       uint64_t                cpu_c3res;
+       uint64_t                cpu_c6res;
+       uint64_t                cpu_c7res;
+       uint64_t                cpu_itime_total;
+       uint64_t                cpu_rtime_total;
+       uint64_t                cpu_ixtime;
+       uint64_t                cpu_idle_exits;
+       uint64_t                cpu_rtimes[CPU_RTIME_BINS];
+       uint64_t                cpu_itimes[CPU_ITIME_BINS];
+#if !MONOTONIC
+       uint64_t                cpu_cur_insns;
+       uint64_t                cpu_cur_ucc;
+       uint64_t                cpu_cur_urc;
+#endif /* !MONOTONIC */
+       uint64_t                cpu_gpmcs[4];
+       uint64_t                cpu_max_observed_int_latency;
+       int                     cpu_max_observed_int_latency_vector;
+       volatile boolean_t      cpu_NMI_acknowledged;
+       uint64_t                debugger_entry_time;
+       uint64_t                debugger_ipi_time;
+       /* A separate nested interrupt stack flag, to account
+        * for non-nested interrupts arriving while on the interrupt stack
+        * Currently only occurs when AICPM enables interrupts on the
+        * interrupt stack during processor offlining.
+        */
+       uint32_t                cpu_nested_istack;
+       uint32_t                cpu_nested_istack_events;
+       x86_saved_state64_t     *cpu_fatal_trap_state;
+       x86_saved_state64_t     *cpu_post_fatal_trap_state;
+#if CONFIG_VMX
+       vmx_cpu_t               cpu_vmx;                /* wonderful world of virtualization */
+#endif
+#if CONFIG_MCA
+       struct mca_state        *cpu_mca_state;         /* State at MC fault */
+#endif
+       int                     cpu_type;
+       int                     cpu_subtype;
+       int                     cpu_threadtype;
+       boolean_t               cpu_iflag;
+       boolean_t               cpu_boot_complete;
+       int                     cpu_hibernate;
+#define MAX_PREEMPTION_RECORDS (8)
+#if    DEVELOPMENT || DEBUG
+       int                     cpu_plri;
+       plrecord_t              plrecords[MAX_PREEMPTION_RECORDS];
+#endif
+       void                    *cpu_console_buf;
+       struct x86_lcpu         lcpu;
+       int                     cpu_phys_number;        /* Physical CPU */
+       cpu_id_t                cpu_id;                 /* Platform Expert */
+#if DEBUG
+       uint64_t                cpu_entry_cr3;
+       uint64_t                cpu_exit_cr3;
+       uint64_t                cpu_pcid_last_cr3;
+#endif
+       boolean_t               cpu_rendezvous_in_progress;
+} cpu_data_t;
+
+extern cpu_data_t      *cpu_data_ptr[];  
+
+/* Macro to generate inline bodies to retrieve per-cpu data fields. */
+#if defined(__clang__)
+#define GS_RELATIVE volatile __attribute__((address_space(256)))
+#ifndef offsetof
+#define offsetof(TYPE,MEMBER) __builtin_offsetof(TYPE,MEMBER)
+#endif
 
-#define        get_cpu_data()  &cpu_data[cpu_number()]
+#define CPU_DATA_GET(member,type)                                                                              \
+       cpu_data_t GS_RELATIVE *cpu_data =                                                      \
+               (cpu_data_t GS_RELATIVE *)0UL;                                                                  \
+       type ret;                                                                                                                       \
+       ret = cpu_data->member;                                                                                         \
+       return ret;
+
+#define CPU_DATA_GET_INDEX(member,index,type)                                                  \
+       cpu_data_t GS_RELATIVE *cpu_data =                                                      \
+               (cpu_data_t GS_RELATIVE *)0UL;                                                                  \
+       type ret;                                                                                                                       \
+       ret = cpu_data->member[index];                                                                          \
+       return ret;
+
+#define CPU_DATA_SET(member,value)                                                                             \
+       cpu_data_t GS_RELATIVE *cpu_data =                                                      \
+               (cpu_data_t GS_RELATIVE *)0UL;                                                                  \
+       cpu_data->member = value;
+
+#define CPU_DATA_XCHG(member,value,type)                                                               \
+       cpu_data_t GS_RELATIVE *cpu_data =                                                      \
+               (cpu_data_t GS_RELATIVE *)0UL;                                                                  \
+       type ret;                                                                                                                       \
+       ret = cpu_data->member;                                                                                         \
+       cpu_data->member = value;                                                                                       \
+       return ret;
+
+#else /* !defined(__clang__) */
+
+#ifndef offsetof
+#define offsetof(TYPE,MEMBER) ((size_t) &((TYPE *)0)->MEMBER)
+#endif /* offsetof */
+#define CPU_DATA_GET(member,type)                                      \
+       type ret;                                                       \
+       __asm__ volatile ("mov %%gs:%P1,%0"                             \
+               : "=r" (ret)                                            \
+               : "i" (offsetof(cpu_data_t,member)));                   \
+       return ret;
+
+#define CPU_DATA_GET_INDEX(member,index,type)  \
+       type ret;                                                       \
+       __asm__ volatile ("mov %%gs:(%1),%0"                            \
+               : "=r" (ret)                                            \
+               : "r" (offsetof(cpu_data_t,member[index])));                    \
+       return ret;
+
+#define CPU_DATA_SET(member,value)                                     \
+       __asm__ volatile ("mov %0,%%gs:%P1"                             \
+               :                                                       \
+               : "r" (value), "i" (offsetof(cpu_data_t,member)));
+
+#define CPU_DATA_XCHG(member,value,type)                               \
+       type ret;                                                       \
+       __asm__ volatile ("xchg %0,%%gs:%P1"                            \
+               : "=r" (ret)                                            \
+               : "i" (offsetof(cpu_data_t,member)), "0" (value));      \
+       return ret;
+
+#endif /* !defined(__clang__) */
 
 /*
  * Everyone within the osfmk part of the kernel can use the fast
  * inline versions of these routines.  Everyone outside, must call
  * the real thing,
  */
-extern thread_t        __inline__ current_thread_fast(void);
-extern thread_t __inline__ current_thread_fast(void)
-{
-       register thread_t       ct;
-       register int            idx = (int)&((cpu_data_t *)0)->active_thread;
 
-       __asm__ volatile ("     movl %%gs:(%1),%0" : "=r" (ct) : "r" (idx));
 
-       return (ct);
+/*
+ * The "volatile" flavor of current_thread() is intended for use by
+ * scheduler code which may need to update the thread pointer in the
+ * course of a context switch.  Any call to current_thread() made
+ * prior to the thread pointer update should be safe to optimize away
+ * as it should be consistent with that thread's state to the extent
+ * the compiler can reason about it.  Likewise, the context switch
+ * path will eventually result in an arbitrary branch to the new
+ * thread's pc, about which the compiler won't be able to reason.
+ * Thus any compile-time optimization of current_thread() calls made
+ * within the new thread should be safely encapsulated in its
+ * register/stack state.  The volatile form therefore exists to cover
+ * the window between the thread pointer update and the branch to
+ * the new pc.
+ */
+static inline thread_t
+get_active_thread_volatile(void)
+{
+       CPU_DATA_GET(cpu_active_thread,thread_t)
+}
+
+static inline __pure2 thread_t
+get_active_thread(void)
+{
+       CPU_DATA_GET(cpu_active_thread,thread_t)
 }
 
-#define current_thread()       current_thread_fast()
+#define current_thread_fast()          get_active_thread()
+#define current_thread_volatile()      get_active_thread_volatile()
+#define current_thread()               current_thread_fast()
 
-extern int     __inline__      get_preemption_level(void);
-extern void    __inline__      disable_preemption(void);
-extern void    __inline__      enable_preemption(void);
-extern void    __inline__      enable_preemption_no_check(void);
-extern void    __inline__      mp_disable_preemption(void);
-extern void    __inline__      mp_enable_preemption(void);
-extern void    __inline__      mp_enable_preemption_no_check(void);
-extern int     __inline__      get_simple_lock_count(void);
-extern int     __inline__      get_interrupt_level(void);
+#define cpu_mode_is64bit()             TRUE
 
-extern int __inline__          get_preemption_level(void)
+static inline int
+get_preemption_level(void)
+{
+       CPU_DATA_GET(cpu_preemption_level,int)
+}
+static inline int
+get_interrupt_level(void)
 {
-       register int    idx = (int)&((cpu_data_t *)0)->preemption_level;
-       register int    pl;
+       CPU_DATA_GET(cpu_interrupt_level,int)
+}
+static inline int
+get_cpu_number(void)
+{
+       CPU_DATA_GET(cpu_number,int)
+}
+static inline int
+get_cpu_phys_number(void)
+{
+       CPU_DATA_GET(cpu_phys_number,int)
+}
 
-       __asm__ volatile ("     movl %%gs:(%1),%0" : "=r" (pl) : "r" (idx));
+static inline cpu_data_t *
+current_cpu_datap(void) {
+       CPU_DATA_GET(cpu_this, cpu_data_t *);
+}
 
-       return (pl);
+/*
+ * Facility to diagnose preemption-level imbalances, which are otherwise
+ * challenging to debug. On each operation that enables or disables preemption,
+ * we record a backtrace into a per-CPU ring buffer, along with the current
+ * preemption level and operation type. Thus, if an imbalance is observed,
+ * one can examine these per-CPU records to determine which codepath failed
+ * to re-enable preemption, enabled premption without a corresponding
+ * disablement etc. The backtracer determines which stack is currently active,
+ * and uses that to perform bounds checks on unterminated stacks.
+ * To enable, sysctl -w machdep.pltrace=1 on DEVELOPMENT or DEBUG kernels (DRK '15)
+ * The bounds check currently doesn't account for non-default thread stack sizes.
+ */
+#if DEVELOPMENT || DEBUG
+static inline void pltrace_bt(uint64_t *rets, int maxframes, uint64_t stacklo, uint64_t stackhi) {
+       uint64_t *cfp = (uint64_t *) __builtin_frame_address(0);
+       int plbtf;
+
+       assert(stacklo !=0  && stackhi !=0);
+
+       for (plbtf = 0; plbtf < maxframes; plbtf++) {
+               if (((uint64_t)cfp == 0) || (((uint64_t)cfp < stacklo) || ((uint64_t)cfp > stackhi))) {
+                       rets[plbtf] = 0;
+                       continue;
+               }
+               rets[plbtf] = *(cfp + 1);
+               cfp = (uint64_t *) (*cfp);
+       }
 }
 
-extern void __inline__         disable_preemption(void)
-{
-#if    MACH_ASSERT
-       extern void _disable_preemption(void);
 
-       _disable_preemption();
-#else  /* MACH_ASSERT */
-       register int    idx = (int)&((cpu_data_t *)0)->preemption_level;
+extern uint32_t                low_intstack[];         /* bottom */
+extern uint32_t                low_eintstack[];        /* top */
+extern char            mp_slave_stack[PAGE_SIZE];
+
+static inline void pltrace_internal(boolean_t enable) {
+       cpu_data_t *cdata = current_cpu_datap();
+       int cpli = cdata->cpu_preemption_level;
+       int cplrecord = cdata->cpu_plri;
+       uint64_t kstackb, kstackt, *plbts;
+
+       assert(cpli >= 0);
+
+       cdata->plrecords[cplrecord].pltype = enable;
+       cdata->plrecords[cplrecord].plevel = cpli;
+
+       plbts = &cdata->plrecords[cplrecord].plbt[0];
+
+       cplrecord++;
+
+       if (cplrecord >= MAX_PREEMPTION_RECORDS) {
+               cplrecord = 0;
+       }
+
+       cdata->cpu_plri = cplrecord;
+       /* Obtain the 'current' program counter, initial backtrace
+        * element. This will also indicate if we were unable to
+        * trace further up the stack for some reason
+        */
+       __asm__ volatile("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:"
+           : "=m" (plbts[0])
+           :
+           : "rax");
+
+
+       thread_t cplthread = cdata->cpu_active_thread;
+       if (cplthread) {
+               uintptr_t csp;
+               __asm__ __volatile__ ("movq %%rsp, %0": "=r" (csp):);
+               /* Determine which stack we're on to populate stack bounds.
+                * We don't need to trace across stack boundaries for this
+                * routine.
+                */
+               kstackb = cdata->cpu_active_stack;
+               kstackt = kstackb + KERNEL_STACK_SIZE;
+               if (csp < kstackb || csp > kstackt) {
+                       kstackt = cdata->cpu_kernel_stack;
+                       kstackb = kstackb - KERNEL_STACK_SIZE;
+                       if (csp < kstackb || csp > kstackt) {
+                               kstackt = cdata->cpu_int_stack_top;
+                               kstackb = kstackt - INTSTACK_SIZE;
+                               if (csp < kstackb || csp > kstackt) {
+                                       kstackt = (uintptr_t)low_eintstack;
+                                       kstackb = (uintptr_t)low_eintstack - INTSTACK_SIZE;
+                                       if (csp < kstackb || csp > kstackt) {
+                                               kstackb = (uintptr_t) mp_slave_stack;
+                                               kstackt = (uintptr_t) mp_slave_stack + PAGE_SIZE;
+                                       }
+                               }
+                       }
+               }
+
+               if (kstackb) {
+                       pltrace_bt(&plbts[1], MAXPLFRAMES - 1, kstackb, kstackt);
+               }
+       }
+}
 
-       __asm__ volatile ("     incl %%gs:(%0)" : : "r" (idx));
-#endif /* MACH_ASSERT */
+extern int plctrace_enabled;
+#endif /* DEVELOPMENT || DEBUG */
+
+static inline void pltrace(boolean_t plenable) {
+#if DEVELOPMENT || DEBUG
+       if (__improbable(plctrace_enabled != 0)) {
+               pltrace_internal(plenable);
+       }
+#else
+       (void)plenable;
+#endif
 }
 
-extern void __inline__         enable_preemption(void)
-{
-#if    MACH_ASSERT
-       extern void _enable_preemption(void);
+static inline void
+disable_preemption_internal(void) {
+       assert(get_preemption_level() >= 0);
+
+#if defined(__clang__)
+       cpu_data_t GS_RELATIVE *cpu_data = (cpu_data_t GS_RELATIVE *)0UL;
+       cpu_data->cpu_preemption_level++;
+#else
+       __asm__ volatile ("incl %%gs:%P0"
+           :
+           : "i" (offsetof(cpu_data_t, cpu_preemption_level)));
+#endif
+       pltrace(FALSE);
+}
 
+static inline void
+enable_preemption_internal(void) {
        assert(get_preemption_level() > 0);
-       _enable_preemption();
-#else  /* MACH_ASSERT */
-       extern void             kernel_preempt_check (void);
-       register int    idx = (int)&((cpu_data_t *)0)->preemption_level;
-       register void (*kpc)(void)=     kernel_preempt_check;
-
-       __asm__ volatile ("decl %%gs:(%0); jne 1f; \
-                       call %1; 1:"
+       pltrace(TRUE);
+#if defined(__clang__)
+       cpu_data_t GS_RELATIVE *cpu_data = (cpu_data_t GS_RELATIVE *)0UL;
+       if (0 == --cpu_data->cpu_preemption_level)
+               kernel_preempt_check();
+#else
+       __asm__ volatile ("decl %%gs:%P0                \n\t"
+                         "jne 1f                       \n\t"
+                         "call _kernel_preempt_check   \n\t"
+                         "1:"
                        : /* no outputs */
-                       : "r" (idx), "r" (kpc)
-                       : "%eax", "%ecx", "%edx", "cc", "memory");
-#endif /* MACH_ASSERT */
+                       : "i" (offsetof(cpu_data_t, cpu_preemption_level))
+                       : "eax", "ecx", "edx", "cc", "memory");
+#endif
 }
 
-extern void __inline__         enable_preemption_no_check(void)
+static inline void
+enable_preemption_no_check(void)
 {
-#if    MACH_ASSERT
-       extern void _enable_preemption_no_check(void);
-
        assert(get_preemption_level() > 0);
-       _enable_preemption_no_check();
-#else  /* MACH_ASSERT */
-       register int    idx = (int)&((cpu_data_t *)0)->preemption_level;
 
-       __asm__ volatile ("decl %%gs:(%0)"
+       pltrace(TRUE);
+#if defined(__clang__)
+       cpu_data_t GS_RELATIVE *cpu_data = (cpu_data_t GS_RELATIVE *)0UL;
+       cpu_data->cpu_preemption_level--;
+#else
+       __asm__ volatile ("decl %%gs:%P0"
                        : /* no outputs */
-                       : "r" (idx)
+                       : "i" (offsetof(cpu_data_t, cpu_preemption_level))
                        : "cc", "memory");
-#endif /* MACH_ASSERT */
+#endif
 }
 
-extern void __inline__         mp_disable_preemption(void)
-{
-#if    NCPUS > 1
-       disable_preemption();
-#endif /* NCPUS > 1 */
+static inline void
+_enable_preemption_no_check(void) {
+       enable_preemption_no_check();
 }
 
-extern void __inline__         mp_enable_preemption(void)
+static inline void
+mp_disable_preemption(void)
 {
-#if    NCPUS > 1
-       enable_preemption();
-#endif /* NCPUS > 1 */
+       disable_preemption_internal();
 }
 
-extern void __inline__         mp_enable_preemption_no_check(void)
+static inline void
+_mp_disable_preemption(void)
 {
-#if    NCPUS > 1
-       enable_preemption_no_check();
-#endif /* NCPUS > 1 */
+       disable_preemption_internal();
 }
 
-extern int __inline__          get_simple_lock_count(void)
+static inline void
+mp_enable_preemption(void)
 {
-       register int    idx = (int)&((cpu_data_t *)0)->simple_lock_count;
-       register int    pl;
+       enable_preemption_internal();
+}
 
-       __asm__ volatile ("     movl %%gs:(%1),%0" : "=r" (pl) : "r" (idx));
+static inline void
+_mp_enable_preemption(void) {
+       enable_preemption_internal();
+}
 
-       return (pl);
+static inline void
+mp_enable_preemption_no_check(void) {
+       enable_preemption_no_check();
 }
 
-extern int __inline__          get_interrupt_level(void)
-{
-       register int    idx = (int)&((cpu_data_t *)0)->interrupt_level;
-       register int    pl;
+static inline void
+_mp_enable_preemption_no_check(void) {
+       enable_preemption_no_check();
+}
 
-       __asm__ volatile ("     movl %%gs:(%1),%0" : "=r" (pl) : "r" (idx));
+#ifdef XNU_KERNEL_PRIVATE
+#define disable_preemption() disable_preemption_internal()
+#define enable_preemption() enable_preemption_internal()
+#define MACHINE_PREEMPTION_MACROS (1)
+#endif
 
-       return (pl);
+static inline cpu_data_t *
+cpu_datap(int cpu) {
+       return cpu_data_ptr[cpu];
 }
 
-#if 0
-#ifndef        __OPTIMIZE__
-#undef         extern 
-#endif
-#endif
+static inline int
+cpu_is_running(int cpu) {
+       return ((cpu_datap(cpu) != NULL) && (cpu_datap(cpu)->cpu_running));
+}
 
-#else  /* !defined(__GNUC__) */
+#ifdef MACH_KERNEL_PRIVATE
+static inline cpu_data_t *
+cpu_shadowp(int cpu) {
+       return cpu_data_ptr[cpu]->cd_shadow;
+}
 
-#endif /* defined(__GNUC__) */
+#endif
+extern cpu_data_t *cpu_data_alloc(boolean_t is_boot_cpu);
+extern void cpu_data_realloc(void);
 
 #endif /* I386_CPU_DATA */