/* newlib_syscalls.c - newlib syscalls for ESP8266 * * Part of esp-open-rtos * Copyright (C) 2105 Superhouse Automation Pty Ltd * BSD Licensed as described in the file LICENSE */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * The file descriptor index space is allocated in blocks. The first block of 3 * is for newlib I/O the stdin stdout and stderr. The next block of * MEMP_NUM_NETCONN is allocated for lwip sockets, and the remainer to file * system descriptors. The newlib default FD_SETSIZE is 64. */ #if LWIP_SOCKET_OFFSET < 3 #error Expecting a LWIP_SOCKET_OFFSET >= 3, to allow room for the standard I/O descriptors. #endif #define FILE_DESCRIPTOR_OFFSET (LWIP_SOCKET_OFFSET + MEMP_NUM_NETCONN) #if FILE_DESCRIPTOR_OFFSET > FD_SETSIZE #error Too many lwip sockets for the FD_SETSIZE. #endif extern void *xPortSupervisorStackPointer; IRAM void *_sbrk_r (struct _reent *r, ptrdiff_t incr) { extern char _heap_start; /* linker script defined */ static char * heap_end; char * prev_heap_end; if (heap_end == NULL) heap_end = &_heap_start; prev_heap_end = heap_end; intptr_t sp = (intptr_t)xPortSupervisorStackPointer; if(sp == 0) /* scheduler not started */ SP(sp); if ((intptr_t)heap_end + incr >= sp) { r->_errno = ENOMEM; return (caddr_t)-1; } heap_end += incr; return (caddr_t) prev_heap_end; } /* Insert a disjoint region into the nano malloc pool. Create a malloc chunk, * filling the size as newlib nano malloc expects, and then free it. */ void nano_malloc_insert_chunk(void *start, size_t size) { *(uint32_t *)start = size; free(start + sizeof(size_t)); } /* syscall implementation for stdio write to UART */ __attribute__((weak)) ssize_t _write_stdout_r(struct _reent *r, int fd, const void *ptr, size_t len ) { for(int i = 0; i < len; i++) { /* Auto convert CR to CRLF, ignore other LFs (compatible with Espressif SDK behaviour) */ if(((char *)ptr)[i] == '\r') continue; if(((char *)ptr)[i] == '\n') uart_putc(0, '\r'); uart_putc(0, ((char *)ptr)[i]); } return len; } static _WriteFunction *current_stdout_write_r = &_write_stdout_r; void set_write_stdout(_WriteFunction *f) { if (f != NULL) { current_stdout_write_r = f; } else { current_stdout_write_r = &_write_stdout_r; } } _WriteFunction *get_write_stdout() { return current_stdout_write_r; } /* default implementation, replace in a filesystem */ __attribute__((weak)) ssize_t _write_filesystem_r(struct _reent *r, int fd, const void *ptr, size_t len) { r->_errno = EBADF; return -1; } __attribute__((weak)) ssize_t _write_r(struct _reent *r, int fd, const void *ptr, size_t len) { if (fd >= FILE_DESCRIPTOR_OFFSET) { return _write_filesystem_r(r, fd, ptr, len); } if (fd >= LWIP_SOCKET_OFFSET) { return lwip_write(fd, ptr, len); } if (fd == r->_stdout->_file) { return current_stdout_write_r(r, fd, ptr, len); } r->_errno = EBADF; return -1; } /* syscall implementation for stdio read from UART */ __attribute__((weak)) ssize_t _read_stdin_r(struct _reent *r, int fd, void *ptr, size_t len) { int ch, i; uart_rxfifo_wait(0, 1); for(i = 0; i < len; i++) { ch = uart_getc_nowait(0); if (ch < 0) break; ((char *)ptr)[i] = ch; } return i; } /* default implementation, replace in a filesystem */ __attribute__((weak)) ssize_t _read_filesystem_r( struct _reent *r, int fd, void *ptr, size_t len ) { r->_errno = EBADF; return -1; } __attribute__((weak)) ssize_t _read_r( struct _reent *r, int fd, void *ptr, size_t len ) { if (fd >= FILE_DESCRIPTOR_OFFSET) { return _read_filesystem_r(r, fd, ptr, len); } if (fd >= LWIP_SOCKET_OFFSET) { return lwip_read(fd, ptr, len); } if (fd == r->_stdin->_file) { return _read_stdin_r(r, fd, ptr, len); } r->_errno = EBADF; return -1; } /* default implementation, replace in a filesystem */ __attribute__((weak)) int _close_filesystem_r(struct _reent *r, int fd) { r->_errno = EBADF; return -1; } __attribute__((weak)) int _close_r(struct _reent *r, int fd) { if (fd >= FILE_DESCRIPTOR_OFFSET) { return _close_filesystem_r(r, fd); } if (fd >= LWIP_SOCKET_OFFSET) { return lwip_close(fd); } r->_errno = EBADF; return -1; } /* Stub syscall implementations follow, to allow compiling newlib functions that pull these in via various codepaths */ __attribute__((weak, alias("syscall_returns_enosys"))) int _open_r(struct _reent *r, const char *pathname, int flags, int mode); __attribute__((weak, alias("syscall_returns_enosys"))) int _unlink_r(struct _reent *r, const char *path); __attribute__((weak, alias("syscall_returns_enosys"))) int _fstat_r(struct _reent *r, int fd, struct stat *buf); __attribute__((weak, alias("syscall_returns_enosys"))) int _stat_r(struct _reent *r, const char *pathname, struct stat *buf); __attribute__((weak, alias("syscall_returns_enosys"))) off_t _lseek_r(struct _reent *r, int fd, off_t offset, int whence); __attribute__((weak, alias("_gettimeofday_r"))) int _gettimeofday_r (struct _reent *ptr, struct timeval *ptimeval, void *ptimezone) { ptimeval->tv_sec = 0; ptimeval->tv_usec = 0; errno = ENOSYS; return -1; } /* Generic stub for any newlib syscall that fails with errno ENOSYS ("Function not implemented") and a return value equivalent to (int)-1. */ static int syscall_returns_enosys(struct _reent *r) { r->_errno=ENOSYS; return -1; } int getentropy(void *ptr, size_t n) { hwrand_fill(ptr, n); return 0; } void _arc4random_getentropy_fail(void) { } void _exit(int status) { while(1); } /* * Newlib lock implementation. Some newlib locks are statically allocated, but * can not be statically initialized so are set to NULL and initialized at * startup. The malloc lock is used before it can be initialized so there are * runtime checks on the functions that use it early. */ static int locks_initialized = 0; extern _lock_t __arc4random_mutex; extern _lock_t __at_quick_exit_mutex; //extern _lock_t __dd_hash_mutex; extern _lock_t __tz_mutex; extern _lock_t __atexit_recursive_mutex; extern _lock_t __env_recursive_mutex; extern _lock_t __malloc_recursive_mutex; extern _lock_t __sfp_recursive_mutex; extern _lock_t __sinit_recursive_mutex; void init_newlib_locks() { #if 0 /* Used a separate mutex for each lock. * Each mutex uses about 96 bytes which adds up. */ _lock_init(&__arc4random_mutex); _lock_init(&__at_quick_exit_mutex); //_lock_init(&__dd_hash_mutex); _lock_init(&__tz_mutex); _lock_init_recursive(&__atexit_recursive_mutex); _lock_init_recursive(&__env_recursive_mutex); _lock_init_recursive(&__malloc_recursive_mutex); _lock_init_recursive(&__sfp_recursive_mutex); _lock_init_recursive(&__sinit_recursive_mutex); #else /* Reuse one mutex and one recursive mutex for this set, reducing memory * usage. Newlib will still allocate other locks dynamically and some of * those need to be separate such as the file lock where a thread might * block with them held. */ _lock_init(&__arc4random_mutex); __at_quick_exit_mutex = __arc4random_mutex; //__dd_hash_mutex = __arc4random_mutex; __tz_mutex = __arc4random_mutex; _lock_init_recursive(&__atexit_recursive_mutex); __env_recursive_mutex = __atexit_recursive_mutex; __malloc_recursive_mutex = __atexit_recursive_mutex; __sfp_recursive_mutex = __atexit_recursive_mutex; __sinit_recursive_mutex = __atexit_recursive_mutex; #endif locks_initialized = 1; } void _lock_init(_lock_t *lock) { *lock = (_lock_t)xSemaphoreCreateMutex(); } void _lock_init_recursive(_lock_t *lock) { *lock = (_lock_t)xSemaphoreCreateRecursiveMutex(); } void _lock_close(_lock_t *lock) { vSemaphoreDelete((QueueHandle_t)*lock); *lock = 0; } void _lock_close_recursive(_lock_t *lock) { vSemaphoreDelete((QueueHandle_t)*lock); *lock = 0; } void _lock_acquire(_lock_t *lock) { xSemaphoreTake((QueueHandle_t)*lock, portMAX_DELAY); } void _lock_acquire_recursive(_lock_t *lock) { if (locks_initialized) { if (sdk_NMIIrqIsOn) { uart_putc(0, ':'); return; } xSemaphoreTakeRecursive((QueueHandle_t)*lock, portMAX_DELAY); } } int _lock_try_acquire(_lock_t *lock) { return xSemaphoreTake((QueueHandle_t)*lock, 0); } int _lock_try_acquire_recursive(_lock_t *lock) { return xSemaphoreTakeRecursive((QueueHandle_t)*lock, 0); } void _lock_release(_lock_t *lock) { xSemaphoreGive((QueueHandle_t)*lock); } void _lock_release_recursive(_lock_t *lock) { if (locks_initialized) { if (sdk_NMIIrqIsOn) { return; } xSemaphoreGiveRecursive((QueueHandle_t)*lock); } }