percpu.h

Go to the documentation of this file.

/*
 *  COPYRIGHT (c) 1989-2011.
 *  On-Line Applications Research Corporation (OAR).
 *
 *  Copyright (c) 2012, 2018 embedded brains GmbH
 *
 *  The license and distribution terms for this file may be
 *  found in the file LICENSE in this distribution or at
 *  http://www.rtems.org/license/LICENSE.
 */

#ifndef _RTEMS_PERCPU_H
#define _RTEMS_PERCPU_H

#include <rtems/score/cpuimpl.h>

#if defined( ASM )
  #include <rtems/asm.h>
#else
  #include <rtems/score/assert.h>
  #include <rtems/score/chain.h>
  #include <rtems/score/isrlock.h>
  #include <rtems/score/smp.h>
  #include <rtems/score/smplock.h>
  #include <rtems/score/timestamp.h>
  #include <rtems/score/watchdog.h>
#endif

#ifdef __cplusplus
extern "C" {
#endif

#if defined(RTEMS_SMP)
  #if defined(RTEMS_PROFILING)
    #define PER_CPU_CONTROL_SIZE_APPROX ( 512 + CPU_INTERRUPT_FRAME_SIZE )
  #elif defined(RTEMS_DEBUG) || CPU_SIZEOF_POINTER > 4
    #define PER_CPU_CONTROL_SIZE_APPROX ( 256 + CPU_INTERRUPT_FRAME_SIZE )
  #else
    #define PER_CPU_CONTROL_SIZE_APPROX ( 128 + CPU_INTERRUPT_FRAME_SIZE )
  #endif

  /*
   * This ensures that on SMP configurations the individual per-CPU controls
   * are on different cache lines to prevent false sharing.  This define can be
   * used in assembler code to easily get the per-CPU control for a particular
   * processor.
   */
  #if PER_CPU_CONTROL_SIZE_APPROX > 1024
    #define PER_CPU_CONTROL_SIZE_LOG2 11
  #elif PER_CPU_CONTROL_SIZE_APPROX > 512
    #define PER_CPU_CONTROL_SIZE_LOG2 10
  #elif PER_CPU_CONTROL_SIZE_APPROX > 256
    #define PER_CPU_CONTROL_SIZE_LOG2 9
  #elif PER_CPU_CONTROL_SIZE_APPROX > 128
    #define PER_CPU_CONTROL_SIZE_LOG2 8
  #else
    #define PER_CPU_CONTROL_SIZE_LOG2 7
  #endif

  #define PER_CPU_CONTROL_SIZE ( 1 << PER_CPU_CONTROL_SIZE_LOG2 )
#endif

#if !defined( ASM )

struct Record_Control;

struct _Thread_Control;

struct Scheduler_Context;

#if defined( RTEMS_SMP )

typedef enum {
  PER_CPU_STATE_INITIAL,

  PER_CPU_STATE_READY_TO_START_MULTITASKING,

  PER_CPU_STATE_REQUEST_START_MULTITASKING,

  PER_CPU_STATE_UP,

  PER_CPU_STATE_SHUTDOWN
} Per_CPU_State;

#endif /* defined( RTEMS_SMP ) */

typedef struct {
#if defined( RTEMS_PROFILING )

  CPU_Counter_ticks thread_dispatch_disabled_instant;

  CPU_Counter_ticks max_thread_dispatch_disabled_time;

  CPU_Counter_ticks max_interrupt_time;

  CPU_Counter_ticks max_interrupt_delay;

  uint64_t thread_dispatch_disabled_count;

  uint64_t total_thread_dispatch_disabled_time;

  uint64_t interrupt_count;

  uint64_t total_interrupt_time;
#endif /* defined( RTEMS_PROFILING ) */
} Per_CPU_Stats;

typedef enum {
  PER_CPU_WATCHDOG_TICKS,

  PER_CPU_WATCHDOG_REALTIME,

  PER_CPU_WATCHDOG_MONOTONIC,

  PER_CPU_WATCHDOG_COUNT
} Per_CPU_Watchdog_index;

typedef struct Per_CPU_Control {
  #if CPU_PER_CPU_CONTROL_SIZE > 0

    CPU_Per_CPU_control cpu_per_cpu;
  #endif

  void *interrupt_stack_low;

  void *interrupt_stack_high;

  uint32_t isr_nest_level;

  uint32_t isr_dispatch_disable;

  volatile uint32_t thread_dispatch_disable_level;

  volatile bool dispatch_necessary;

  /*
   * Ensure that the executing member is at least 4-byte aligned, see
   * PER_CPU_OFFSET_EXECUTING.  This is necessary on CPU ports with relaxed
   * alignment restrictions, e.g. type alignment is less than the type size.
   */
  bool reserved_for_executing_alignment[ 3 ];

  struct _Thread_Control *executing;

  struct _Thread_Control *heir;

#if defined(RTEMS_SMP)
  CPU_Interrupt_frame Interrupt_frame;
#endif

  Timestamp_Control cpu_usage_timestamp;

  struct {
    ISR_LOCK_MEMBER( Lock )

    
    uint64_t ticks;

    Watchdog_Header Header[ PER_CPU_WATCHDOG_COUNT ];
  } Watchdog;

  #if defined( RTEMS_SMP )

    SMP_ticket_lock_Control Lock;

    #if defined( RTEMS_PROFILING )

      SMP_lock_Stats Lock_stats;

      SMP_lock_Stats_context Lock_stats_context;
    #endif

    Chain_Control Threads_in_need_for_help;

    Atomic_Ulong message;

    struct {
      const struct _Scheduler_Control *control;

      const struct Scheduler_Context *context;

      struct _Thread_Control *idle_if_online_and_unused;
    } Scheduler;

    char *data;

    Per_CPU_State state;

    Atomic_Uintptr before_multitasking_action;

    bool online;

    bool boot;
  #endif

  struct Record_Control *record;

  Per_CPU_Stats Stats;
} Per_CPU_Control;

#if defined( RTEMS_SMP )
typedef struct {
  Per_CPU_Control per_cpu;
  char unused_space_for_cache_line_alignment
    [ PER_CPU_CONTROL_SIZE - sizeof( Per_CPU_Control ) ];
} Per_CPU_Control_envelope;
#else
typedef struct {
  Per_CPU_Control per_cpu;
} Per_CPU_Control_envelope;
#endif

extern Per_CPU_Control_envelope _Per_CPU_Information[] CPU_STRUCTURE_ALIGNMENT;

#if defined( RTEMS_SMP )
#define _Per_CPU_Acquire( cpu ) \
  _SMP_ticket_lock_Acquire( \
    &( cpu )->Lock, \
    &( cpu )->Lock_stats, \
    &( cpu )->Lock_stats_context \
  )
#else
#define _Per_CPU_Acquire( cpu ) \
  do { \
    (void) ( cpu ); \
  } while ( 0 )
#endif

#if defined( RTEMS_SMP )
#define _Per_CPU_Release( cpu ) \
  _SMP_ticket_lock_Release( \
    &( cpu )->Lock, \
    &( cpu )->Lock_stats_context \
  )
#else
#define _Per_CPU_Release( cpu ) \
  do { \
    (void) ( cpu ); \
  } while ( 0 )
#endif

#if defined( RTEMS_SMP )
#define _Per_CPU_ISR_disable_and_acquire( cpu, isr_cookie ) \
  do { \
    _ISR_Local_disable( isr_cookie ); \
    _Per_CPU_Acquire( cpu ); \
  } while ( 0 )
#else
#define _Per_CPU_ISR_disable_and_acquire( cpu, isr_cookie ) \
  do { \
    _ISR_Local_disable( isr_cookie ); \
    (void) ( cpu ); \
  } while ( 0 )
#endif

#if defined( RTEMS_SMP )
#define _Per_CPU_Release_and_ISR_enable( cpu, isr_cookie ) \
  do { \
    _Per_CPU_Release( cpu ); \
    _ISR_Local_enable( isr_cookie ); \
  } while ( 0 )
#else
#define _Per_CPU_Release_and_ISR_enable( cpu, isr_cookie ) \
  do { \
    (void) ( cpu ); \
    _ISR_Local_enable( isr_cookie ); \
  } while ( 0 )
#endif

#if defined( RTEMS_SMP )
#define _Per_CPU_Acquire_all( isr_cookie ) \
  do { \
    uint32_t ncpus = _SMP_Get_processor_count(); \
    uint32_t cpu; \
    _ISR_Local_disable( isr_cookie ); \
    for ( cpu = 0 ; cpu < ncpus ; ++cpu ) { \
      _Per_CPU_Acquire( _Per_CPU_Get_by_index( cpu ) ); \
    } \
  } while ( 0 )
#else
#define _Per_CPU_Acquire_all( isr_cookie ) \
  _ISR_Local_disable( isr_cookie )
#endif

#if defined( RTEMS_SMP )
#define _Per_CPU_Release_all( isr_cookie ) \
  do { \
    uint32_t ncpus = _SMP_Get_processor_count(); \
    uint32_t cpu; \
    for ( cpu = 0 ; cpu < ncpus ; ++cpu ) { \
      _Per_CPU_Release( _Per_CPU_Get_by_index( cpu ) ); \
    } \
    _ISR_Local_enable( isr_cookie ); \
  } while ( 0 )
#else
#define _Per_CPU_Release_all( isr_cookie ) \
  _ISR_Local_enable( isr_cookie )
#endif

/*
 * If we get the current processor index in a context which allows thread
 * dispatching, then we may already run on another processor right after the
 * read instruction.  There are very few cases in which this makes sense (here
 * we can use _Per_CPU_Get_snapshot()).  All other places must use
 * _Per_CPU_Get() so that we can add checks for RTEMS_DEBUG.
 */
#if defined( _CPU_Get_current_per_CPU_control )
  #define _Per_CPU_Get_snapshot() _CPU_Get_current_per_CPU_control()
#else
  #define _Per_CPU_Get_snapshot() \
    ( &_Per_CPU_Information[ _SMP_Get_current_processor() ].per_cpu )
#endif

#if defined( RTEMS_SMP )
static inline Per_CPU_Control *_Per_CPU_Get( void )
{
  Per_CPU_Control *cpu_self = _Per_CPU_Get_snapshot();

  _Assert(
    cpu_self->thread_dispatch_disable_level != 0 || _ISR_Get_level() != 0
  );

  return cpu_self;
}
#else
#define _Per_CPU_Get() _Per_CPU_Get_snapshot()
#endif

static inline Per_CPU_Control *_Per_CPU_Get_by_index( uint32_t index )
{
  return &_Per_CPU_Information[ index ].per_cpu;
}

static inline uint32_t _Per_CPU_Get_index( const Per_CPU_Control *cpu )
{
  const Per_CPU_Control_envelope *per_cpu_envelope =
    ( const Per_CPU_Control_envelope * ) cpu;

  return ( uint32_t ) ( per_cpu_envelope - &_Per_CPU_Information[ 0 ] );
}

static inline struct _Thread_Control *_Per_CPU_Get_executing(
  const Per_CPU_Control *cpu
)
{
  return cpu->executing;
}

static inline bool _Per_CPU_Is_processor_online(
  const Per_CPU_Control *cpu
)
{
#if defined( RTEMS_SMP )
  return cpu->online;
#else
  (void) cpu;

  return true;
#endif
}

static inline bool _Per_CPU_Is_boot_processor(
  const Per_CPU_Control *cpu
)
{
#if defined( RTEMS_SMP )
  return cpu->boot;
#else
  (void) cpu;

  return true;
#endif
}

#if defined( RTEMS_SMP )

void _Per_CPU_Initialize(void);

void _Per_CPU_State_change(
  Per_CPU_Control *cpu,
  Per_CPU_State new_state
);

bool _Per_CPU_State_wait_for_non_initial_state(
  uint32_t cpu_index,
  uint32_t timeout_in_ns
);

#endif /* defined( RTEMS_SMP ) */

/*
 * On a non SMP system, the _SMP_Get_current_processor() is defined to 0.
 * Thus when built for non-SMP, there should be no performance penalty.
 */
#define _Thread_Dispatch_disable_level \
  _Per_CPU_Get()->thread_dispatch_disable_level
#define _Thread_Heir \
  _Per_CPU_Get()->heir

#if defined(_CPU_Get_thread_executing)
#define _Thread_Executing \
  _CPU_Get_thread_executing()
#else
#define _Thread_Executing \
  _Per_CPU_Get_executing( _Per_CPU_Get() )
#endif

#define _ISR_Nest_level \
  _Per_CPU_Get()->isr_nest_level
#define _CPU_Interrupt_stack_low \
  _Per_CPU_Get()->interrupt_stack_low
#define _CPU_Interrupt_stack_high \
  _Per_CPU_Get()->interrupt_stack_high
#define _Thread_Dispatch_necessary \
  _Per_CPU_Get()->dispatch_necessary

RTEMS_INLINE_ROUTINE struct _Thread_Control *_Thread_Get_executing( void )
{
  struct _Thread_Control *executing;

  #if defined(RTEMS_SMP) && !defined(_CPU_Get_thread_executing)
    ISR_Level level;

    _ISR_Local_disable( level );
  #endif

  executing = _Thread_Executing;

  #if defined(RTEMS_SMP) && !defined(_CPU_Get_thread_executing)
    _ISR_Local_enable( level );
  #endif

  return executing;
}

#endif /* !defined( ASM ) */

#if defined( ASM ) || defined( _RTEMS_PERCPU_DEFINE_OFFSETS )

#define PER_CPU_INTERRUPT_STACK_LOW \
  CPU_PER_CPU_CONTROL_SIZE
#define PER_CPU_INTERRUPT_STACK_HIGH \
  PER_CPU_INTERRUPT_STACK_LOW + CPU_SIZEOF_POINTER

#define INTERRUPT_STACK_LOW \
  (SYM(_Per_CPU_Information) + PER_CPU_INTERRUPT_STACK_LOW)
#define INTERRUPT_STACK_HIGH \
  (SYM(_Per_CPU_Information) + PER_CPU_INTERRUPT_STACK_HIGH)

/*
 *  These are the offsets of the required elements in the per CPU table.
 */
#define PER_CPU_ISR_NEST_LEVEL \
  PER_CPU_INTERRUPT_STACK_HIGH + CPU_SIZEOF_POINTER
#define PER_CPU_ISR_DISPATCH_DISABLE \
  PER_CPU_ISR_NEST_LEVEL + 4
#define PER_CPU_THREAD_DISPATCH_DISABLE_LEVEL \
  PER_CPU_ISR_DISPATCH_DISABLE + 4
#define PER_CPU_DISPATCH_NEEDED \
  PER_CPU_THREAD_DISPATCH_DISABLE_LEVEL + 4
#define PER_CPU_OFFSET_EXECUTING \
  PER_CPU_DISPATCH_NEEDED + 4
#define PER_CPU_OFFSET_HEIR \
  PER_CPU_OFFSET_EXECUTING + CPU_SIZEOF_POINTER
#if defined(RTEMS_SMP)
#define PER_CPU_INTERRUPT_FRAME_AREA \
  PER_CPU_OFFSET_HEIR + CPU_SIZEOF_POINTER
#endif

#define THREAD_DISPATCH_DISABLE_LEVEL \
  (SYM(_Per_CPU_Information) + PER_CPU_THREAD_DISPATCH_DISABLE_LEVEL)
#define ISR_NEST_LEVEL \
  (SYM(_Per_CPU_Information) + PER_CPU_ISR_NEST_LEVEL)
#define DISPATCH_NEEDED \
  (SYM(_Per_CPU_Information) + PER_CPU_DISPATCH_NEEDED)

#endif /* defined( ASM ) || defined( _RTEMS_PERCPU_DEFINE_OFFSETS ) */

#ifdef __cplusplus
}
#endif

#endif
/* end of include file */