lwip/doc/sys_arch.txt

sys_arch interface for lwIP

Author: Adam Dunkels
        Simon Goldschmidt

The operating system emulation layer provides a common interface
between the lwIP code and the underlying operating system kernel. The
general idea is that porting lwIP to new architectures requires only
small changes to a few header files and a new sys_arch
implementation. It is also possible to do a sys_arch implementation
that does not rely on any underlying operating system.

The sys_arch provides semaphores, mailboxes and mutexes to lwIP. For the full
lwIP functionality, multiple threads support can be implemented in the
sys_arch, but this is not required for the basic lwIP
functionality. Timer scheduling is implemented in lwIP, but can be implemented
by the sys_arch port (LWIP_TIMERS_CUSTOM==1).

In addition to the source file providing the functionality of sys_arch,
the OS emulation layer must provide several header files defining
macros used throughout lwip.  The files required and the macros they
must define are listed below the sys_arch description.

Semaphores can be either counting or binary - lwIP works with both
kinds. Mailboxes should be implemented as a queue which allows multiple messages
to be posted (implementing as a rendez-vous point where only one message can be
posted at a time can have a highly negative impact on performance). A message
in a mailbox is just a pointer, nothing more. 

Semaphores are represented by the type "sys_sem_t" which is typedef'd
in the sys_arch.h file. Mailboxes are equivalently represented by the
type "sys_mbox_t". Mutexes are represented by the type "sys_mutex_t".
lwIP does not place any restrictions on how these types are represented
internally.

Since lwIP 1.4.0, semaphore, mutexes and mailbox functions are prototyped in a way that
allows both using pointers or actual OS structures to be used. This way, memory
required for such types can be either allocated in place (globally or on the
stack) or on the heap (allocated internally in the "*_new()" functions).

The following functions must be implemented by the sys_arch:

- void sys_init(void)

  Is called to initialize the sys_arch layer.

- err_t sys_sem_new(sys_sem_t *sem, u8_t count)

  Creates a new semaphore. The semaphore is allocated to the memory that 'sem'
  points to (which can be both a pointer or the actual OS structure).
  The "count" argument specifies the initial state of the semaphore (which is
  either 0 or 1).
  If the semaphore has been created, ERR_OK should be returned. Returning any
  other error will provide a hint what went wrong, but except for assertions,
  no real error handling is implemented.

- void sys_sem_free(sys_sem_t *sem)

  Deallocates a semaphore.

- void sys_sem_signal(sys_sem_t *sem)

  Signals a semaphore.

- u32_t sys_arch_sem_wait(sys_sem_t *sem, u32_t timeout)

  Blocks the thread while waiting for the semaphore to be signaled. If the
  "timeout" argument is non-zero, the thread should only be blocked for the
  specified time (measured in milliseconds). If the "timeout" argument is zero,
  the thread should be blocked until the semaphore is signalled.

  The return value is SYS_ARCH_TIMEOUT if the semaphore wasn't signaled within
  the specified time or any other value if it was signaled (with or without
  waiting).

  Notice that lwIP implements a function with a similar name,
  sys_sem_wait(), that uses the sys_arch_sem_wait() function.

- int sys_sem_valid(sys_sem_t *sem)

  Returns 1 if the semaphore is valid, 0 if it is not valid.
  When using pointers, a simple way is to check the pointer for != NULL.
  When directly using OS structures, implementing this may be more complex.
  This may also be a define, in which case the function is not prototyped.

- void sys_sem_set_invalid(sys_sem_t *sem)

  Invalidate a semaphore so that sys_sem_valid() returns 0.
  ATTENTION: This does NOT mean that the semaphore shall be deallocated:
  sys_sem_free() is always called before calling this function!
  This may also be a define, in which case the function is not prototyped.

- void sys_mutex_new(sys_mutex_t *mutex)

  Creates a new mutex. The mutex is allocated to the memory that 'mutex'
  points to (which can be both a pointer or the actual OS structure).
  If the mutex has been created, ERR_OK should be returned. Returning any
  other error will provide a hint what went wrong, but except for assertions,
  no real error handling is implemented.

- void sys_mutex_free(sys_mutex_t *mutex)

  Deallocates a mutex.

- void sys_mutex_lock(sys_mutex_t *mutex)
  
  Blocks the thread until the mutex can be grabbed.

- void sys_mutex_unlock(sys_mutex_t *mutex)

  Releases the mutex previously locked through 'sys_mutex_lock()'.

- void sys_mutex_valid(sys_mutex_t *mutex)

  Returns 1 if the mutes is valid, 0 if it is not valid.
  When using pointers, a simple way is to check the pointer for != NULL.
  When directly using OS structures, implementing this may be more complex.
  This may also be a define, in which case the function is not prototyped.

- void sys_mutex_set_invalid(sys_mutex_t *mutex)

  Invalidate a mutex so that sys_mutex_valid() returns 0.
  ATTENTION: This does NOT mean that the mutex shall be deallocated:
  sys_mutex_free() is always called before calling this function!
  This may also be a define, in which case the function is not prototyped.

- err_t sys_mbox_new(sys_mbox_t *mbox, int size)

  Creates an empty mailbox for maximum "size" elements. Elements stored
  in mailboxes are pointers. You have to define macros "_MBOX_SIZE"
  in your lwipopts.h, or ignore this parameter in your implementation
  and use a default size.
  If the mailbox has been created, ERR_OK should be returned. Returning any
  other error will provide a hint what went wrong, but except for assertions,
  no real error handling is implemented.

- void sys_mbox_free(sys_mbox_t *mbox)

  Deallocates a mailbox. If there are messages still present in the
  mailbox when the mailbox is deallocated, it is an indication of a
  programming error in lwIP and the developer should be notified.

- void sys_mbox_post(sys_mbox_t *mbox, void *msg)

  Posts the "msg" to the mailbox. This function have to block until
  the "msg" is really posted.

- err_t sys_mbox_trypost(sys_mbox_t *mbox, void *msg)

  Try to post the "msg" to the mailbox. Returns ERR_MEM if this one
  is full, else, ERR_OK if the "msg" is posted.

- u32_t sys_arch_mbox_fetch(sys_mbox_t *mbox, void **msg, u32_t timeout)

  Blocks the thread until a message arrives in the mailbox, but does
  not block the thread longer than "timeout" milliseconds (similar to
  the sys_arch_sem_wait() function). If "timeout" is 0, the thread should
  be blocked until a message arrives. The "msg" argument is a result
  parameter that is set by the function (i.e., by doing "*msg =
  ptr"). The "msg" parameter maybe NULL to indicate that the message
  should be dropped.

  The return values are the same as for the sys_arch_sem_wait() function:
  SYS_ARCH_TIMEOUT if there was a timeout, any other value if a messages
  is received.

  Note that a function with a similar name, sys_mbox_fetch(), is
  implemented by lwIP. 

- u32_t sys_arch_mbox_tryfetch(sys_mbox_t *mbox, void **msg)

  This is similar to sys_arch_mbox_fetch, however if a message is not
  present in the mailbox, it immediately returns with the code
  SYS_MBOX_EMPTY. On success 0 is returned.

  To allow for efficient implementations, this can be defined as a
  function-like macro in sys_arch.h instead of a normal function. For
  example, a naive implementation could be:
    #define sys_arch_mbox_tryfetch(mbox,msg) \
      sys_arch_mbox_fetch(mbox,msg,1)
  although this would introduce unnecessary delays.

- int sys_mbox_valid(sys_mbox_t *mbox)

  Returns 1 if the mailbox is valid, 0 if it is not valid.
  When using pointers, a simple way is to check the pointer for != NULL.
  When directly using OS structures, implementing this may be more complex.
  This may also be a define, in which case the function is not prototyped.

- void sys_mbox_set_invalid(sys_mbox_t *mbox)

  Invalidate a mailbox so that sys_mbox_valid() returns 0.
  ATTENTION: This does NOT mean that the mailbox shall be deallocated:
  sys_mbox_free() is always called before calling this function!
  This may also be a define, in which case the function is not prototyped.

If threads are supported by the underlying operating system and if
such functionality is needed in lwIP, the following function will have
to be implemented as well:

- sys_thread_t sys_thread_new(char *name, void (* thread)(void *arg), void *arg, int stacksize, int prio)

  Starts a new thread named "name" with priority "prio" that will begin its
  execution in the function "thread()". The "arg" argument will be passed as an
  argument to the thread() function. The stack size to used for this thread is
  the "stacksize" parameter. The id of the new thread is returned. Both the id
  and the priority are system dependent.

When lwIP is used from more than one context (e.g. from multiple threads OR from
main-loop and from interrupts), the SYS_LIGHTWEIGHT_PROT protection SHOULD be enabled!

- sys_prot_t sys_arch_protect(void)

  This optional function does a "fast" critical region protection. This function
  is only called during very short critical regions. An embedded system which
  supports ISR-based drivers might want to implement this function by disabling
  interrupts. Task-based systems might want to implement this by using a mutex
  or disabling tasking. This function should support recursive calls from the
  same task or interrupt. In other words, sys_arch_protect() could be called
  while already protected.

  The return value is opaque to lwip and passed to the sys_arch_unprotect() call
  matching the sys_arch_protect() call at the same nesting level. This value
  might be used to restore the status. However implementations may depend on
  every call to sys_arch_protect() having a matching call to sys_arch_unprotect()
  and thus can use a nesting count or a recursive mutex.

  sys_arch_protect() is only required if your port is supporting an operating
  system.

- void sys_arch_unprotect(sys_prot_t pval)

  This optional function does a "fast" exit of critical region protection
  nesting level. The value passed in pval is the opaque value returned the
  respective call to sys_arch_protect(). See the documentation for
  sys_arch_protect() for more information. This function is only required if
  your port is supporting an operating system.

For some configurations, you also need:

- u32_t sys_now(void)

  This optional function returns the current time in milliseconds (don't care
  for wraparound, this is only used for time diffs).
  Not implementing this function means you cannot use some modules (e.g. TCP
  timestamps, internal timeouts for NO_SYS==1).


Note:

Be careful with using mem_malloc() in sys_arch. When malloc() refers to
mem_malloc() you can run into a circular function call problem. In mem.c
mem_init() tries to allcate a semaphore using mem_malloc, which of course
can't be performed when sys_arch uses mem_malloc.

-------------------------------------------------------------------------------
Additional files required for the "OS support" emulation layer:
-------------------------------------------------------------------------------

cc.h       - Architecture environment, some compiler specific, some
             environment specific (probably should move env stuff 
             to sys_arch.h.)

  Typedefs for the types used by lwip -
    u8_t, s8_t, u16_t, s16_t, u32_t, s32_t, mem_ptr_t

  Compiler hints for packing lwip's structures -
    PACK_STRUCT_FIELD(x)
    PACK_STRUCT_STRUCT
    PACK_STRUCT_BEGIN
    PACK_STRUCT_END

  Platform specific diagnostic output -
    LWIP_PLATFORM_DIAG(x)    - non-fatal, print a message.
    LWIP_PLATFORM_ASSERT(x)  - fatal, print message and abandon execution.
    Portability defines for printf formatters:
    U16_F, S16_F, X16_F, U32_F, S32_F, X32_F, SZT_F

  "lightweight" synchronization mechanisms -
    SYS_ARCH_DECL_PROTECT(x) - declare a protection state variable.
    SYS_ARCH_PROTECT(x)      - enter protection mode.
    SYS_ARCH_UNPROTECT(x)    - leave protection mode.

  If the compiler does not provide memset() this file must include a
  definition of it, or include a file which defines it.

  This file must either include a system-local <errno.h> which defines
  the standard *nix error codes (or define LWIP_ERRNO_INCLUDE to that file name),
  or it should #define LWIP_PROVIDE_ERRNO to make lwip/arch.h define the codes
  which are used throughout.


perf.h     - Architecture specific performance measurement.
  Measurement calls made throughout lwip, these can be defined to nothing.
    PERF_START               - start measuring something.
    PERF_STOP(x)             - stop measuring something, and record the result.

sys_arch.h - Tied to sys_arch.c

  Arch dependent types for the following objects:
    sys_sem_t, sys_mbox_t, sys_thread_t,
  And, optionally:
    sys_prot_t

  Defines to set vars of sys_mbox_t and sys_sem_t to NULL.
    SYS_MBOX_NULL NULL
    SYS_SEM_NULL NULL