diff options
Diffstat (limited to 'erts/emulator/sys/unix')
-rw-r--r-- | erts/emulator/sys/unix/erl_unix_sys.h | 181 | ||||
-rw-r--r-- | erts/emulator/sys/unix/sys.c | 494 | ||||
-rw-r--r-- | erts/emulator/sys/unix/sys_float.c | 4 | ||||
-rw-r--r-- | erts/emulator/sys/unix/sys_time.c | 751 |
4 files changed, 1245 insertions, 185 deletions
diff --git a/erts/emulator/sys/unix/erl_unix_sys.h b/erts/emulator/sys/unix/erl_unix_sys.h index dda50c486c..94adcc00c8 100644 --- a/erts/emulator/sys/unix/erl_unix_sys.h +++ b/erts/emulator/sys/unix/erl_unix_sys.h @@ -45,7 +45,7 @@ #include <fcntl.h> #include "erl_errno.h" #include <signal.h> - +#include <setjmp.h> #if HAVE_SYS_SOCKETIO_H # include <sys/socketio.h> @@ -114,11 +114,6 @@ /* * Make sure that MAXPATHLEN is defined. */ -#ifdef GETHRTIME_WITH_CLOCK_GETTIME -#undef HAVE_GETHRTIME -#define HAVE_GETHRTIME 1 -#endif - #ifndef MAXPATHLEN # ifdef PATH_MAX # define MAXPATHLEN PATH_MAX @@ -160,35 +155,136 @@ typedef struct timeval SysTimeval; typedef struct tms SysTimes; -extern int erts_ticks_per_sec; - -#define SYS_CLK_TCK (erts_ticks_per_sec) +#define SYS_CLK_TCK (erts_sys_time_data__.r.o.ticks_per_sec) #define sys_times(Arg) times(Arg) -#define ERTS_WRAP_SYS_TIMES 1 -extern int erts_ticks_per_sec_wrap; -#define SYS_CLK_TCK_WRAP (erts_ticks_per_sec_wrap) -extern clock_t sys_times_wrap(void); +#if SIZEOF_LONG == 8 +typedef long ErtsMonotonicTime; +typedef long ErtsSysHrTime; +#elif SIZEOF_LONG_LONG == 8 +typedef long long ErtsMonotonicTime; +typedef long long ErtsSysHrTime; +#else +#error No signed 64-bit type found... +#endif + +typedef ErtsMonotonicTime ErtsSystemTime; + +#define ERTS_MONOTONIC_TIME_MIN (((ErtsMonotonicTime) 1) << 63) +#define ERTS_MONOTONIC_TIME_MAX (~ERTS_MONOTONIC_TIME_MIN) + +/* + * OS monotonic time and OS system time + */ + +#undef ERTS_OS_TIMES_INLINE_FUNC_PTR_CALL__ + +#if defined(OS_SYSTEM_TIME_USING_CLOCK_GETTIME) +# if defined(__linux__) +# define ERTS_OS_TIMES_INLINE_FUNC_PTR_CALL__ 1 +# endif +#endif + +ErtsSystemTime erts_os_system_time(void); + +#undef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT +#undef ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT +#undef ERTS_OS_MONOTONIC_INLINE_FUNC_PTR_CALL__ +#undef ERTS_HAVE_CORRECTED_OS_MONOTONIC + +#if defined(OS_MONOTONIC_TIME_USING_CLOCK_GETTIME) +# define ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT 1 +# define ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT (1000*1000*1000) +# if defined(__linux__) +# define ERTS_HAVE_CORRECTED_OS_MONOTONIC 1 +# define ERTS_OS_MONOTONIC_INLINE_FUNC_PTR_CALL__ 1 +# endif +#elif defined(OS_MONOTONIC_TIME_USING_MACH_CLOCK_GET_TIME) +# define ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT 1 +# define ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT (1000*1000*1000) +#elif defined(OS_MONOTONIC_TIME_USING_GETHRTIME) +# define ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT 1 +# define ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT (1000*1000*1000) +#elif defined(OS_MONOTONIC_TIME_USING_TIMES) +# define ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT 1 +/* Time unit determined at runtime... */ +# define ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT 0 +#else /* No OS monotonic available... */ +# define ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT (1000*1000) +#endif + +/* + * erts_sys_hrtime() is the highest resolution + * time function found. Time unit is nano-seconds. + * It may or may not be monotonic. + */ +ErtsSysHrTime erts_sys_hrtime(void); + +struct erts_sys_time_read_only_data__ { +#ifdef ERTS_OS_MONOTONIC_INLINE_FUNC_PTR_CALL__ + ErtsMonotonicTime (*os_monotonic_time)(void); +#endif +#ifdef ERTS_OS_TIMES_INLINE_FUNC_PTR_CALL__ + void (*os_times)(ErtsMonotonicTime *, ErtsSystemTime *); +#endif + int ticks_per_sec; +}; + +typedef struct { + union { + struct erts_sys_time_read_only_data__ o; + char align__[(((sizeof(struct erts_sys_time_read_only_data__) - 1) + / ASSUMED_CACHE_LINE_SIZE) + 1) + * ASSUMED_CACHE_LINE_SIZE]; + } r; +} ErtsSysTimeData__; -#ifdef HAVE_GETHRTIME -#ifdef GETHRTIME_WITH_CLOCK_GETTIME -typedef long long SysHrTime; +extern ErtsSysTimeData__ erts_sys_time_data__; -extern SysHrTime sys_gethrtime(void); -#define sys_init_hrtime() /* Nothing */ +#ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT -#else /* Real gethrtime (Solaris) */ +#ifdef ERTS_OS_MONOTONIC_INLINE_FUNC_PTR_CALL__ +ERTS_GLB_INLINE +#endif +ErtsMonotonicTime erts_os_monotonic_time(void); -typedef hrtime_t SysHrTime; +#ifdef ERTS_OS_TIMES_INLINE_FUNC_PTR_CALL__ +ERTS_GLB_INLINE +#endif +void erts_os_times(ErtsMonotonicTime *, ErtsSystemTime *); -#define sys_gethrtime() gethrtime() -#define sys_init_hrtime() /* Nothing */ +#if ERTS_GLB_INLINE_INCL_FUNC_DEF + +#ifdef ERTS_OS_MONOTONIC_INLINE_FUNC_PTR_CALL__ + +ERTS_GLB_INLINE ErtsMonotonicTime +erts_os_monotonic_time(void) +{ + return (*erts_sys_time_data__.r.o.os_monotonic_time)(); +} + +#endif /* ERTS_OS_MONOTONIC_INLINE_FUNC_PTR_CALL__ */ + +#ifdef ERTS_OS_TIMES_INLINE_FUNC_PTR_CALL__ + +ERTS_GLB_INLINE void +erts_os_times(ErtsMonotonicTime *mtimep, ErtsSystemTime *stimep) +{ + return (*erts_sys_time_data__.r.o.os_times)(mtimep, stimep); +} -#endif /* GETHRTIME_WITH_CLOCK_GETTIME */ -#endif /* HAVE_GETHRTIME */ +#endif /* ERTS_OS_TIMES_INLINE_FUNC_PTR_CALL__ */ -#if (defined(HAVE_GETHRVTIME) || defined(HAVE_CLOCK_GETTIME)) +#endif /* ERTS_GLB_INLINE_INCL_FUNC_DEF */ + +#endif /* ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT */ + +/* + * + */ + +#if (defined(HAVE_GETHRVTIME) || defined(HAVE_CLOCK_GETTIME_CPU_TIME)) typedef long long SysCpuTime; typedef struct timespec SysTimespec; @@ -200,7 +296,7 @@ typedef struct timespec SysTimespec; int sys_start_hrvtime(void); int sys_stop_hrvtime(void); -#elif defined(HAVE_CLOCK_GETTIME) +#elif defined(HAVE_CLOCK_GETTIME_CPU_TIME) #define sys_clock_gettime(cid,tp) clock_gettime((cid),&(tp)) #define sys_get_proc_cputime(t,tp) sys_clock_gettime(CLOCK_PROCESS_CPUTIME_ID,(tp)) @@ -211,13 +307,8 @@ int sys_stop_hrvtime(void); #define SYS_CLOCK_RESOLUTION 1 /* These are defined in sys.c */ -#if defined(SIG_SIGSET) /* Old SysV */ -RETSIGTYPE (*sys_sigset())(); -#elif defined(SIG_SIGNAL) /* Old BSD */ -RETSIGTYPE (*sys_sigset())(); -#else -RETSIGTYPE (*sys_sigset(int, RETSIGTYPE (*func)(int)))(int); -#endif +typedef void (*SIGFUNC)(int); +extern SIGFUNC sys_signal(int, SIGFUNC); extern void sys_sigrelease(int); extern void sys_sigblock(int); extern void sys_stop_cat(void); @@ -356,4 +447,28 @@ extern int exit_async(void); #define ERTS_EXIT_AFTER_DUMP _exit +#if !defined(__APPLE__) && !defined(__MACH__) +/* Some OS X versions do not allow (ab)using signal handlers like this */ +#define ERTS_HAVE_TRY_CATCH 1 + +/* We try to simulate a try catch in C with the help of signal handlers. + * Only use this as a very last resort, as it is not very portable and + * quite unstable. It is also not thread safe, so make sure that only + * one thread can call this at a time! + */ +extern void erts_sys_sigsegv_handler(int); +extern jmp_buf erts_sys_sigsegv_jmp; +#define ERTS_SYS_TRY_CATCH(EXPR,CATCH) \ + do { \ + SIGFUNC prev_handler = sys_signal(SIGSEGV, \ + erts_sys_sigsegv_handler); \ + if (!setjmp(erts_sys_sigsegv_jmp)) { \ + EXPR; \ + } else { \ + CATCH; \ + } \ + sys_signal(SIGSEGV,prev_handler); \ + } while(0) +#endif + #endif /* #ifndef _ERL_UNIX_SYS_H */ diff --git a/erts/emulator/sys/unix/sys.c b/erts/emulator/sys/unix/sys.c index cd87b320e2..7d52650a70 100644 --- a/erts/emulator/sys/unix/sys.c +++ b/erts/emulator/sys/unix/sys.c @@ -34,6 +34,7 @@ #include <termios.h> #include <ctype.h> #include <sys/utsname.h> +#include <sys/select.h> #ifdef ISC32 #include <sys/bsdtypes.h> @@ -85,14 +86,24 @@ static erts_smp_rwmtx_t environ_rwmtx; #define DISABLE_VFORK 0 #endif +#if defined IOV_MAX +#define MAXIOV IOV_MAX +#elif defined UIO_MAXIOV +#define MAXIOV UIO_MAXIOV +#else +#define MAXIOV 16 +#endif + #ifdef USE_THREADS # ifdef ENABLE_CHILD_WAITER_THREAD # define CHLDWTHR ENABLE_CHILD_WAITER_THREAD # else # define CHLDWTHR 0 # endif +# define FDBLOCK 1 #else # define CHLDWTHR 0 +# define FDBLOCK 0 #endif /* * [OTP-3906] @@ -121,6 +132,15 @@ struct ErtsSysReportExit_ { #endif }; +/* Used by the fd driver iff the fd could not be set to non-blocking */ +typedef struct ErtsSysBlocking_ { + ErlDrvPDL pdl; + int res; + int err; + unsigned int pkey; +} ErtsSysBlocking; + + /* This data is shared by these drivers - initialized by spawn_init() */ static struct driver_data { ErlDrvPort port_num; @@ -129,6 +149,8 @@ static struct driver_data { int pid; int alive; int status; + int terminating; + ErtsSysBlocking *blocking; } *driver_data; /* indexed by fd */ static ErtsSysReportExit *report_exit_list; @@ -202,8 +224,14 @@ static erts_smp_atomic_t sys_misc_mem_sz; #if defined(ERTS_SMP) static void smp_sig_notify(char c); static int sig_notify_fds[2] = {-1, -1}; + +static int sig_suspend_fds[2] = {-1, -1}; +#define ERTS_SYS_SUSPEND_SIGNAL SIGUSR2 + #endif +jmp_buf erts_sys_sigsegv_jmp; + #if CHLDWTHR || defined(ERTS_SMP) erts_mtx_t chld_stat_mtx; #endif @@ -280,7 +308,7 @@ struct { int (*event)(ErlDrvPort, ErlDrvEvent, ErlDrvEventData); void (*check_io_as_interrupt)(void); void (*check_io_interrupt)(int); - void (*check_io_interrupt_tmd)(int, erts_short_time_t); + void (*check_io_interrupt_tmd)(int, ErtsMonotonicTime); void (*check_io)(int); Uint (*size)(void); Eterm (*info)(void *); @@ -386,9 +414,9 @@ erts_sys_schedule_interrupt(int set) #ifdef ERTS_SMP void -erts_sys_schedule_interrupt_timed(int set, erts_short_time_t msec) +erts_sys_schedule_interrupt_timed(int set, ErtsMonotonicTime timeout_time) { - ERTS_CHK_IO_INTR_TMD(set, msec); + ERTS_CHK_IO_INTR_TMD(set, timeout_time); } #endif @@ -504,11 +532,14 @@ thr_create_prepare_child(void *vtcdp) void erts_sys_pre_init(void) { +#ifdef USE_THREADS + erts_thr_init_data_t eid = ERTS_THR_INIT_DATA_DEF_INITER; +#endif + erts_printf_add_cr_to_stdout = 1; erts_printf_add_cr_to_stderr = 1; + #ifdef USE_THREADS - { - erts_thr_init_data_t eid = ERTS_THR_INIT_DATA_DEF_INITER; eid.thread_create_child_func = thr_create_prepare_child; /* Before creation in parent */ @@ -531,6 +562,12 @@ erts_sys_pre_init(void) erts_lcnt_init(); #endif +#endif /* USE_THREADS */ + + erts_init_sys_time_sup(); + +#ifdef USE_THREADS + #if CHLDWTHR || defined(ERTS_SMP) erts_mtx_init(&chld_stat_mtx, "child_status"); #endif @@ -541,7 +578,7 @@ erts_sys_pre_init(void) erts_cnd_init(&chld_stat_cnd); children_alive = 0; #endif - } + #ifdef ERTS_SMP erts_smp_atomic32_init_nob(&erts_break_requested, 0); erts_smp_atomic32_init_nob(&erts_got_sigusr1, 0); @@ -554,7 +591,9 @@ erts_sys_pre_init(void) #if !CHLDWTHR && !defined(ERTS_SMP) children_died = 0; #endif + #endif /* USE_THREADS */ + erts_smp_atomic_init_nob(&sys_misc_mem_sz, 0); { @@ -640,39 +679,7 @@ erl_sys_init(void) /* signal handling */ -#ifdef SIG_SIGSET /* Old SysV */ -RETSIGTYPE (*sys_sigset(sig, func))() -int sig; -RETSIGTYPE (*func)(); -{ - return(sigset(sig, func)); -} -void sys_sigblock(int sig) -{ - sighold(sig); -} -void sys_sigrelease(int sig) -{ - sigrelse(sig); -} -#else /* !SIG_SIGSET */ -#ifdef SIG_SIGNAL /* Old BSD */ -RETSIGTYPE (*sys_sigset(sig, func))(int, int) -int sig; -RETSIGTYPE (*func)(); -{ - return(signal(sig, func)); -} -sys_sigblock(int sig) -{ - sigblock(sig); -} -sys_sigrelease(int sig) -{ - sigsetmask(sigblock(0) & ~sigmask(sig)); -} -#else /* !SIG_SIGNAL */ /* The True Way - POSIX!:-) */ -RETSIGTYPE (*sys_sigset(int sig, RETSIGTYPE (*func)(int)))(int) +SIGFUNC sys_signal(int sig, SIGFUNC func) { struct sigaction act, oact; @@ -705,23 +712,35 @@ void sys_sigrelease(int sig) sigaddset(&mask, sig); sigprocmask(SIG_UNBLOCK, &mask, (sigset_t *)NULL); } -#endif /* !SIG_SIGNAL */ -#endif /* !SIG_SIGSET */ -#if (0) /* not used? -- gordon */ -static void (*break_func)(); -static RETSIGTYPE break_handler(int sig) -{ -#ifdef QNX - /* Turn off SIGCHLD during break processing */ - sys_sigblock(SIGCHLD); -#endif - (*break_func)(); -#ifdef QNX - sys_sigrelease(SIGCHLD); -#endif +void erts_sys_sigsegv_handler(int signo) { + if (signo == SIGSEGV) { + longjmp(erts_sys_sigsegv_jmp, 1); + } +} + +/* + * Function returns 1 if we can read from all values in between + * start and stop. + */ +int +erts_sys_is_area_readable(char *start, char *stop) { + int fds[2]; + if (!pipe(fds)) { + /* We let write try to figure out if the pointers are readable */ + int res = write(fds[1], start, (char*)stop - (char*)start); + if (res == -1) { + close(fds[0]); + close(fds[1]); + return 0; + } + close(fds[0]); + close(fds[1]); + return 1; + } + return 0; + } -#endif /* 0 */ static ERTS_INLINE int prepare_crash_dump(int secs) @@ -849,9 +868,23 @@ sigusr1_exit(void) #ifdef ETHR_UNUSABLE_SIGUSRX #warning "Unusable SIGUSR1 & SIGUSR2. Disabling use of these signals" -#endif -#ifndef ETHR_UNUSABLE_SIGUSRX +#else + +#ifdef ERTS_SMP +void +sys_thr_suspend(erts_tid_t tid) { + erts_thr_kill(tid, ERTS_SYS_SUSPEND_SIGNAL); +} + +void +sys_thr_resume(erts_tid_t tid) { + int i = 0, res; + do { + res = write(sig_suspend_fds[1],&i,sizeof(i)); + } while (res < 0 && errno == EAGAIN); +} +#endif #if (defined(SIG_SIGSET) || defined(SIG_SIGNAL)) static RETSIGTYPE user_signal1(void) @@ -866,20 +899,20 @@ static RETSIGTYPE user_signal1(int signum) #endif } -#ifdef QUANTIFY +#ifdef ERTS_SMP #if (defined(SIG_SIGSET) || defined(SIG_SIGNAL)) -static RETSIGTYPE user_signal2(void) +static RETSIGTYPE suspend_signal(void) #else -static RETSIGTYPE user_signal2(int signum) +static RETSIGTYPE suspend_signal(int signum) #endif { -#ifdef ERTS_SMP - smp_sig_notify('2'); -#else - quantify_save_data(); -#endif + int res; + int buf[1]; + do { + res = read(sig_suspend_fds[0], buf, sizeof(int)); + } while (res < 0 && errno == EINTR); } -#endif +#endif /* #ifdef ERTS_SMP */ #endif /* #ifndef ETHR_UNUSABLE_SIGUSRX */ @@ -904,9 +937,9 @@ static RETSIGTYPE do_quit(int signum) /* Disable break */ void erts_set_ignore_break(void) { - sys_sigset(SIGINT, SIG_IGN); - sys_sigset(SIGQUIT, SIG_IGN); - sys_sigset(SIGTSTP, SIG_IGN); + sys_signal(SIGINT, SIG_IGN); + sys_signal(SIGQUIT, SIG_IGN); + sys_signal(SIGTSTP, SIG_IGN); } /* Don't use ctrl-c for break handler but let it be @@ -929,14 +962,14 @@ void erts_replace_intr(void) { void init_break_handler(void) { - sys_sigset(SIGINT, request_break); + sys_signal(SIGINT, request_break); #ifndef ETHR_UNUSABLE_SIGUSRX - sys_sigset(SIGUSR1, user_signal1); -#ifdef QUANTIFY - sys_sigset(SIGUSR2, user_signal2); -#endif + sys_signal(SIGUSR1, user_signal1); +#ifdef ERTS_SMP + sys_signal(ERTS_SYS_SUSPEND_SIGNAL, suspend_signal); +#endif /* #ifdef ERTS_SMP */ #endif /* #ifndef ETHR_UNUSABLE_SIGUSRX */ - sys_sigset(SIGQUIT, do_quit); + sys_signal(SIGQUIT, do_quit); } int sys_max_files(void) @@ -953,8 +986,13 @@ static void block_signals(void) sys_sigblock(SIGINT); #ifndef ETHR_UNUSABLE_SIGUSRX sys_sigblock(SIGUSR1); +#endif /* #ifndef ETHR_UNUSABLE_SIGUSRX */ +#endif /* #ifndef ERTS_SMP */ + +#if defined(ERTS_SMP) && !defined(ETHR_UNUSABLE_SIGUSRX) + sys_sigblock(ERTS_SYS_SUSPEND_SIGNAL); #endif -#endif + } static void unblock_signals(void) @@ -968,26 +1006,13 @@ static void unblock_signals(void) #ifndef ETHR_UNUSABLE_SIGUSRX sys_sigrelease(SIGUSR1); #endif /* #ifndef ETHR_UNUSABLE_SIGUSRX */ +#endif /* #ifndef ERTS_SMP */ + +#if defined(ERTS_SMP) && !defined(ETHR_UNUSABLE_SIGUSRX) + sys_sigrelease(ERTS_SYS_SUSPEND_SIGNAL); #endif -} -/************************** Time stuff **************************/ -#ifdef HAVE_GETHRTIME -#ifdef GETHRTIME_WITH_CLOCK_GETTIME -SysHrTime sys_gethrtime(void) -{ - struct timespec ts; - long long result; - if (clock_gettime(CLOCK_MONOTONIC,&ts) != 0) { - erl_exit(1,"Fatal, could not get clock_monotonic value!, " - "errno = %d\n", errno); - } - result = ((long long) ts.tv_sec) * 1000000000LL + - ((long long) ts.tv_nsec); - return (SysHrTime) result; } -#endif -#endif /************************** OS info *******************************/ @@ -1094,11 +1119,16 @@ void fini_getenv_state(GETENV_STATE *state) /* Driver interfaces */ static ErlDrvData spawn_start(ErlDrvPort, char*, SysDriverOpts*); static ErlDrvData fd_start(ErlDrvPort, char*, SysDriverOpts*); +#if FDBLOCK +static void fd_async(void *); +static void fd_ready_async(ErlDrvData drv_data, ErlDrvThreadData thread_data); +#endif static ErlDrvSSizeT fd_control(ErlDrvData, unsigned int, char *, ErlDrvSizeT, char **, ErlDrvSizeT); static ErlDrvData vanilla_start(ErlDrvPort, char*, SysDriverOpts*); static int spawn_init(void); static void fd_stop(ErlDrvData); +static void fd_flush(ErlDrvData); static void stop(ErlDrvData); static void ready_input(ErlDrvData, ErlDrvEvent); static void ready_output(ErlDrvData, ErlDrvEvent); @@ -1143,8 +1173,12 @@ struct erl_drv_entry fd_driver_entry = { fd_control, NULL, outputv, - NULL, /* ready_async */ - NULL, /* flush */ +#if FDBLOCK + fd_ready_async, /* ready_async */ +#else + NULL, +#endif + fd_flush, /* flush */ NULL, /* call */ NULL, /* event */ ERL_DRV_EXTENDED_MARKER, @@ -1198,13 +1232,28 @@ static RETSIGTYPE onchld(int signum) #endif } +static int set_blocking_data(struct driver_data *dd) { + + dd->blocking = erts_alloc(ERTS_ALC_T_SYS_BLOCKING, sizeof(ErtsSysBlocking)); + + erts_smp_atomic_add_nob(&sys_misc_mem_sz, sizeof(ErtsSysBlocking)); + + dd->blocking->pdl = driver_pdl_create(dd->port_num); + dd->blocking->res = 0; + dd->blocking->err = 0; + dd->blocking->pkey = driver_async_port_key(dd->port_num); + + return 1; +} + static int set_driver_data(ErlDrvPort port_num, int ifd, int ofd, int packet_bytes, int read_write, int exit_status, - int pid) + int pid, + int is_blocking) { Port *prt; ErtsSysReportExit *report_exit; @@ -1236,8 +1285,13 @@ static int set_driver_data(ErlDrvPort port_num, driver_data[ifd].pid = pid; driver_data[ifd].alive = 1; driver_data[ifd].status = 0; + driver_data[ifd].terminating = 0; + driver_data[ifd].blocking = NULL; if (read_write & DO_WRITE) { driver_data[ifd].ofd = ofd; + if (is_blocking && FDBLOCK) + if (!set_blocking_data(driver_data+ifd)) + return -1; if (ifd != ofd) driver_data[ofd] = driver_data[ifd]; /* structure copy */ } else { /* DO_READ only */ @@ -1253,6 +1307,11 @@ static int set_driver_data(ErlDrvPort port_num, driver_data[ofd].pid = pid; driver_data[ofd].alive = 1; driver_data[ofd].status = 0; + driver_data[ofd].terminating = 0; + driver_data[ofd].blocking = NULL; + if (is_blocking && FDBLOCK) + if (!set_blocking_data(driver_data+ofd)) + return -1; return(ofd); } } @@ -1262,11 +1321,13 @@ static int spawn_init() int i; #if CHLDWTHR erts_thr_opts_t thr_opts = ERTS_THR_OPTS_DEFAULT_INITER; + thr_opts.detached = 0; thr_opts.suggested_stack_size = 0; /* Smallest possible */ + thr_opts.name = "child_waiter"; #endif - sys_sigset(SIGPIPE, SIG_IGN); /* Ignore - we'll handle the write failure */ + sys_signal(SIGPIPE, SIG_IGN); /* Ignore - we'll handle the write failure */ driver_data = (struct driver_data *) erts_alloc(ERTS_ALC_T_DRV_TAB, max_files * sizeof(struct driver_data)); erts_smp_atomic_add_nob(&sys_misc_mem_sz, @@ -1279,7 +1340,7 @@ static int spawn_init() sys_sigblock(SIGCHLD); #endif - sys_sigset(SIGCHLD, onchld); /* Reap children */ + sys_signal(SIGCHLD, onchld); /* Reap children */ #if CHLDWTHR erts_thr_create(&child_waiter_tid, child_waiter, NULL, &thr_opts); @@ -1745,7 +1806,7 @@ static ErlDrvData spawn_start(ErlDrvPort port_num, char* name, SysDriverOpts* op } res = set_driver_data(port_num, ifd[0], ofd[1], opts->packet_bytes, - opts->read_write, opts->exit_status, pid); + opts->read_write, opts->exit_status, pid, 0); /* Don't unblock SIGCHLD until now, since the call above must first complete putting away the info about our new subprocess. */ unblock_signals(); @@ -1830,6 +1891,7 @@ static ErlDrvData fd_start(ErlDrvPort port_num, char* name, SysDriverOpts* opts) { ErlDrvData res; + int non_blocking = 0; if (((opts->read_write & DO_READ) && opts->ifd >= max_files) || ((opts->read_write & DO_WRITE) && opts->ofd >= max_files)) @@ -1902,6 +1964,20 @@ static ErlDrvData fd_start(ErlDrvPort port_num, char* name, * case - it can be called with any old pre-existing file descriptors, * the relations between which (if they're even two) we can only guess * at - still, we try our best... + * + * Added note OTP 18: Some systems seem to use stdout/stderr to log data + * using unix pipes, so we cannot allow the system to block on a write. + * Therefore we use an async thread to write the data to fd's that could + * not be set to non-blocking. When no async threads are available we + * fall back on the old behaviour. + * + * Also the guarantee about what is delivered to the OS has changed. + * Pre 18 the fd driver did no flushing of data before terminating. + * Now it does. This is because we want to be able to guarantee that things + * such as escripts and friends really have outputted all data before + * terminating. This could potentially block the termination of the system + * for a very long time, but if the user wants to terminate fast she should + * use erlang:halt with flush=false. */ if (opts->read_write & DO_READ) { @@ -1924,6 +2000,7 @@ static ErlDrvData fd_start(ErlDrvPort port_num, char* name, imagine a scenario where setting non-blocking mode here would cause problems - go ahead and do it. */ + non_blocking = 1; SET_NONBLOCKING(opts->ofd); } else { /* output fd is a tty, input fd isn't */ @@ -1966,6 +2043,7 @@ static ErlDrvData fd_start(ErlDrvPort port_num, char* name, (nfd = open(tty, O_WRONLY)) != -1) { dup2(nfd, opts->ofd); close(nfd); + non_blocking = 1; SET_NONBLOCKING(opts->ofd); } } @@ -1974,8 +2052,9 @@ static ErlDrvData fd_start(ErlDrvPort port_num, char* name, } CHLD_STAT_LOCK; res = (ErlDrvData)(long)set_driver_data(port_num, opts->ifd, opts->ofd, - opts->packet_bytes, - opts->read_write, 0, -1); + opts->packet_bytes, + opts->read_write, 0, -1, + !non_blocking); CHLD_STAT_UNLOCK; return res; } @@ -2001,14 +2080,30 @@ static void nbio_stop_fd(ErlDrvPort prt, int fd) SET_BLOCKING(fd); } -static void fd_stop(ErlDrvData fd) /* Does not close the fds */ +static void fd_stop(ErlDrvData ev) /* Does not close the fds */ { int ofd; + int fd = (int)(long)ev; + ErlDrvPort prt = driver_data[fd].port_num; - nbio_stop_fd(driver_data[(int)(long)fd].port_num, (int)(long)fd); - ofd = driver_data[(int)(long)fd].ofd; - if (ofd != (int)(long)fd && ofd != -1) - nbio_stop_fd(driver_data[(int)(long)fd].port_num, (int)(long)ofd); +#if FDBLOCK + if (driver_data[fd].blocking) { + erts_free(ERTS_ALC_T_SYS_BLOCKING,driver_data[fd].blocking); + driver_data[fd].blocking = NULL; + erts_smp_atomic_add_nob(&sys_misc_mem_sz, -1*sizeof(ErtsSysBlocking)); + } +#endif + + nbio_stop_fd(prt, fd); + ofd = driver_data[fd].ofd; + if (ofd != fd && ofd != -1) + nbio_stop_fd(prt, ofd); +} + +static void fd_flush(ErlDrvData fd) +{ + if (!driver_data[(int)(long)fd].terminating) + driver_data[(int)(long)fd].terminating = 1; } static ErlDrvData vanilla_start(ErlDrvPort port_num, char* name, @@ -2031,8 +2126,8 @@ static ErlDrvData vanilla_start(ErlDrvPort port_num, char* name, CHLD_STAT_LOCK; res = (ErlDrvData)(long)set_driver_data(port_num, fd, fd, - opts->packet_bytes, - opts->read_write, 0, -1); + opts->packet_bytes, + opts->read_write, 0, -1, 0); CHLD_STAT_UNLOCK; return res; } @@ -2069,6 +2164,7 @@ static void stop(ErlDrvData fd) } } +/* used by fd_driver */ static void outputv(ErlDrvData e, ErlIOVec* ev) { int fd = (int)(long)e; @@ -2094,12 +2190,21 @@ static void outputv(ErlDrvData e, ErlIOVec* ev) ev->iov[0].iov_base = lbp; ev->iov[0].iov_len = pb; ev->size += pb; + + if (driver_data[fd].blocking && FDBLOCK) + driver_pdl_lock(driver_data[fd].blocking->pdl); + if ((sz = driver_sizeq(ix)) > 0) { driver_enqv(ix, ev, 0); + + if (driver_data[fd].blocking && FDBLOCK) + driver_pdl_unlock(driver_data[fd].blocking->pdl); + if (sz + ev->size >= (1 << 13)) set_busy_port(ix, 1); } - else { + else if (!driver_data[fd].blocking || !FDBLOCK) { + /* We try to write directly if the fd in non-blocking */ int vsize = ev->vsize > MAX_VSIZE ? MAX_VSIZE : ev->vsize; n = writev(ofd, (const void *) (ev->iov), vsize); @@ -2115,10 +2220,22 @@ static void outputv(ErlDrvData e, ErlIOVec* ev) driver_enqv(ix, ev, n); /* n is the skip value */ driver_select(ix, ofd, ERL_DRV_WRITE|ERL_DRV_USE, 1); } +#if FDBLOCK + else { + if (ev->size != 0) { + driver_enqv(ix, ev, 0); + driver_pdl_unlock(driver_data[fd].blocking->pdl); + driver_async(ix, &driver_data[fd].blocking->pkey, + fd_async, driver_data+fd, NULL); + } else { + driver_pdl_unlock(driver_data[fd].blocking->pdl); + } + } +#endif /* return 0;*/ } - +/* Used by spawn_driver and vanilla driver */ static void output(ErlDrvData e, char* buf, ErlDrvSizeT len) { int fd = (int)(long)e; @@ -2181,6 +2298,23 @@ static int port_inp_failure(ErlDrvPort port_num, int ready_fd, int res) ASSERT(res <= 0); (void) driver_select(port_num, ready_fd, ERL_DRV_READ|ERL_DRV_WRITE, 0); clear_fd_data(ready_fd); + + if (driver_data[ready_fd].blocking && FDBLOCK) { + driver_pdl_lock(driver_data[ready_fd].blocking->pdl); + if (driver_sizeq(driver_data[ready_fd].port_num) > 0) { + driver_pdl_unlock(driver_data[ready_fd].blocking->pdl); + /* We have stuff in the output queue, so we just + set the state to terminating and wait for fd_async_ready + to terminate the port */ + if (res == 0) + driver_data[ready_fd].terminating = 2; + else + driver_data[ready_fd].terminating = -err; + return 0; + } + driver_pdl_unlock(driver_data[ready_fd].blocking->pdl); + } + if (res == 0) { if (driver_data[ready_fd].report_exit) { CHLD_STAT_LOCK; @@ -2231,6 +2365,7 @@ static void ready_input(ErlDrvData e, ErlDrvEvent ready_fd) port_num = driver_data[fd].port_num; packet_bytes = driver_data[fd].packet_bytes; + if (packet_bytes == 0) { byte *read_buf = (byte *) erts_alloc(ERTS_ALC_T_SYS_READ_BUF, ERTS_SYS_READ_BUF_SZ); @@ -2354,6 +2489,8 @@ static void ready_output(ErlDrvData e, ErlDrvEvent ready_fd) if ((iv = (struct iovec*) driver_peekq(ix, &vsize)) == NULL) { driver_select(ix, ready_fd, ERL_DRV_WRITE, 0); + if (driver_data[fd].terminating) + driver_failure_atom(driver_data[fd].port_num,"normal"); return; /* 0; */ } vsize = vsize > MAX_VSIZE ? MAX_VSIZE : vsize; @@ -2379,6 +2516,78 @@ static void stop_select(ErlDrvEvent fd, void* _) close((int)fd); } +#if FDBLOCK + +static void +fd_async(void *async_data) +{ + int res; + struct driver_data *dd = (struct driver_data*)async_data; + SysIOVec *iov0; + SysIOVec *iov; + int iovlen; + int err; + /* much of this code is stolen from efile_drv:invoke_writev */ + driver_pdl_lock(dd->blocking->pdl); + iov0 = driver_peekq(dd->port_num, &iovlen); + iovlen = iovlen < MAXIOV ? iovlen : MAXIOV; + iov = erts_alloc_fnf(ERTS_ALC_T_SYS_WRITE_BUF, + sizeof(SysIOVec)*iovlen); + if (!iov) { + res = -1; + err = ENOMEM; + driver_pdl_unlock(dd->blocking->pdl); + } else { + memcpy(iov,iov0,iovlen*sizeof(SysIOVec)); + driver_pdl_unlock(dd->blocking->pdl); + + res = writev(dd->ofd, iov, iovlen); + err = errno; + + erts_free(ERTS_ALC_T_SYS_WRITE_BUF, iov); + } + dd->blocking->res = res; + dd->blocking->err = err; +} + +void fd_ready_async(ErlDrvData drv_data, + ErlDrvThreadData thread_data) { + struct driver_data *dd = (struct driver_data *)thread_data; + ErlDrvPort port_num = dd->port_num; + + ASSERT(dd->blocking); + ASSERT(dd == (driver_data + (int)(long)drv_data)); + + if (dd->blocking->res > 0) { + driver_pdl_lock(dd->blocking->pdl); + if (driver_deq(port_num, dd->blocking->res) == 0) { + driver_pdl_unlock(dd->blocking->pdl); + set_busy_port(port_num, 0); + if (dd->terminating) { + /* The port is has been ordered to terminate + from either fd_flush or port_inp_failure */ + if (dd->terminating == 1) + driver_failure_atom(port_num, "normal"); + else if (dd->terminating == 2) + driver_failure_eof(port_num); + else if (dd->terminating < 0) + driver_failure_posix(port_num, -dd->terminating); + return; /* -1; */ + } + } else { + driver_pdl_unlock(dd->blocking->pdl); + /* still data left to write in queue */ + driver_async(port_num, &dd->blocking->pkey, fd_async, dd, NULL); + return /* 0; */; + } + } else if (dd->blocking->res < 0) { + driver_failure_posix(port_num, dd->blocking->err); + return; /* -1; */ + } + return; /* 0; */ +} + +#endif void erts_do_break_handling(void) { @@ -2648,18 +2857,30 @@ void sys_preload_end(Preload* p) /* Nothing */ } -/* Read a key from console (?) */ - +/* Read a key from console, used by break.c + Here we assume that all schedulers are stopped so that erl_poll + does not interfere with the select below. +*/ int sys_get_key(fd) int fd; { - int c; + int c, ret; unsigned char rbuf[64]; + fd_set fds; fflush(stdout); /* Flush query ??? */ - if ((c = read(fd,rbuf,64)) <= 0) { - return c; + FD_ZERO(&fds); + FD_SET(fd,&fds); + + ret = select(fd+1, &fds, NULL, NULL, NULL); + + if (ret == 1) { + do { + c = read(fd,rbuf,64); + } while (c < 0 && errno == EAGAIN); + if (c <= 0) + return c; } return rbuf[0]; @@ -2983,13 +3204,6 @@ signal_dispatcher_thread_func(void *unused) case '1': /* SIGUSR1 */ sigusr1_exit(); break; -#ifdef QUANTIFY - case '2': /* SIGUSR2 */ - quantify_save_data(); /* Might take a substantial amount of - time, but this is a test/debug - build */ - break; -#endif default: erl_exit(ERTS_ABORT_EXIT, "signal-dispatcher thread received unknown " @@ -3007,6 +3221,7 @@ init_smp_sig_notify(void) { erts_smp_thr_opts_t thr_opts = ERTS_SMP_THR_OPTS_DEFAULT_INITER; thr_opts.detached = 1; + thr_opts.name = "sys_sig_dispatcher"; if (pipe(sig_notify_fds) < 0) { erl_exit(ERTS_ABORT_EXIT, @@ -3021,6 +3236,17 @@ init_smp_sig_notify(void) NULL, &thr_opts); } + +static void +init_smp_sig_suspend(void) { + if (pipe(sig_suspend_fds) < 0) { + erl_exit(ERTS_ABORT_EXIT, + "Failed to create sig_suspend pipe: %s (%d)\n", + erl_errno_id(errno), + errno); + } +} + #ifdef __DARWIN__ int erts_darwin_main_thread_pipe[2]; @@ -3048,9 +3274,11 @@ erts_sys_main_thread(void) #endif smp_sig_notify(0); /* Notify initialized */ - while (1) { - /* Wait for a signal to arrive... */ + + /* Wait for a signal to arrive... */ + #ifdef __DARWIN__ + while (1) { /* * The wx driver needs to be able to steal the main thread for Cocoa to * work properly. @@ -3065,12 +3293,24 @@ erts_sys_main_thread(void) void* (*func)(void*); void* arg; void *resp; - read(erts_darwin_main_thread_pipe[0],&func,sizeof(void* (*)(void*))); - read(erts_darwin_main_thread_pipe[0],&arg, sizeof(void*)); + res = read(erts_darwin_main_thread_pipe[0],&func,sizeof(void* (*)(void*))); + if (res != sizeof(void* (*)(void*))) + break; + res = read(erts_darwin_main_thread_pipe[0],&arg,sizeof(void*)); + if (res != sizeof(void*)) + break; resp = (*func)(arg); write(erts_darwin_main_thread_result_pipe[1],&resp,sizeof(void *)); } -#else + + if (res == -1 && errno != EINTR) + break; + } + /* Something broke with the main thread pipe, so we ignore it for now. + Most probably erts has closed this pipe and is about to exit. */ +#endif /* #ifdef __DARWIN__ */ + + while (1) { #ifdef DEBUG int res = #else @@ -3079,7 +3319,6 @@ erts_sys_main_thread(void) select(0, NULL, NULL, NULL, NULL); ASSERT(res < 0); ASSERT(errno == EINTR); -#endif } } @@ -3171,6 +3410,7 @@ erl_sys_args(int* argc, char** argv) #ifdef ERTS_SMP init_smp_sig_notify(); + init_smp_sig_suspend(); #endif /* Handled arguments have been marked with NULL. Slide arguments diff --git a/erts/emulator/sys/unix/sys_float.c b/erts/emulator/sys/unix/sys_float.c index cafeab547e..2ffa649767 100644 --- a/erts/emulator/sys/unix/sys_float.c +++ b/erts/emulator/sys/unix/sys_float.c @@ -32,7 +32,7 @@ void erts_sys_init_float(void) { # ifdef SIGFPE - sys_sigset(SIGFPE, SIG_IGN); /* Ignore so we can test for NaN and Inf */ + sys_signal(SIGFPE, SIG_IGN); /* Ignore so we can test for NaN and Inf */ # endif } @@ -667,7 +667,7 @@ static void fpe_sig_handler(int sig) static void erts_thread_catch_fp_exceptions(void) { - sys_sigset(SIGFPE, fpe_sig_handler); + sys_signal(SIGFPE, fpe_sig_handler); unmask_fpe(); } diff --git a/erts/emulator/sys/unix/sys_time.c b/erts/emulator/sys/unix/sys_time.c index fcce54a2c4..d535457977 100644 --- a/erts/emulator/sys/unix/sys_time.c +++ b/erts/emulator/sys/unix/sys_time.c @@ -33,6 +33,21 @@ #include "sys.h" #include "global.h" +#include "erl_os_monotonic_time_extender.h" + +#undef ERTS_HAVE_ERTS_OS_TIMES_IMPL__ +#undef ERTS_HAVE_ERTS_SYS_HRTIME_IMPL__ + +#if defined(OS_MONOTONIC_TIME_USING_MACH_CLOCK_GET_TIME) \ + || defined(OS_SYSTEM_TIME_USING_MACH_CLOCK_GET_TIME) +# include <mach/clock.h> +# include <mach/mach.h> +# ifdef HAVE_CLOCK_GET_ATTRIBUTES +# define ERTS_HAVE_MACH_CLOCK_GETRES +static Sint64 +mach_clock_getres(clock_id_t clkid, char *clkid_str); +# endif +#endif #ifdef NO_SYSCONF # define TICKS_PER_SEC() HZ @@ -53,9 +68,23 @@ /******************* Routines for time measurement *********************/ -int erts_ticks_per_sec = 0; /* Will be SYS_CLK_TCK in erl_unix_sys.h */ -int erts_ticks_per_sec_wrap = 0; /* Will be SYS_CLK_TCK_WRAP */ -static int ticks_bsr = 0; /* Shift wrapped tick value this much to the right */ +#undef ERTS_SYS_TIME_INTERNAL_STATE_WRITE_FREQ__ +#undef ERTS_SYS_TIME_INTERNAL_STATE_READ_ONLY__ +#undef ERTS_SYS_TIME_INTERNAL_STATE_READ_MOSTLY__ + +#if defined(OS_MONOTONIC_TIME_USING_TIMES) + +static Uint32 +get_tick_count(void) +{ + struct tms unused; + return (Uint32) times(&unused); +} + +#define ERTS_SYS_TIME_INTERNAL_STATE_READ_ONLY__ +#define ERTS_SYS_TIME_INTERNAL_STATE_READ_MOSTLY__ + +#endif /* * init timers, chose a tick length, and return it. @@ -63,37 +92,713 @@ static int ticks_bsr = 0; /* Shift wrapped tick value this much to the right */ * does almost everything. Other platforms have to * emulate Unix in this sense. */ -int sys_init_time(void) + +ErtsSysTimeData__ erts_sys_time_data__ erts_align_attribute(ERTS_CACHE_LINE_SIZE); + +#if defined(__linux__) && defined(OS_MONOTONIC_TIME_USING_CLOCK_GETTIME) + +#define ERTS_SYS_TIME_INTERNAL_STATE_WRITE_FREQ__ + +static ErtsMonotonicTime clock_gettime_monotonic_raw(void); +static ErtsMonotonicTime clock_gettime_monotonic_verified(void); +#if defined(OS_SYSTEM_TIME_USING_CLOCK_GETTIME) +static void clock_gettime_times_raw(ErtsMonotonicTime *, ErtsSystemTime *); +static void clock_gettime_times_verified(ErtsMonotonicTime *, ErtsSystemTime *); +#endif + +#endif /* defined(__linux__) && defined(OS_MONOTONIC_TIME_USING_CLOCK_GETTIME) */ + +#ifdef ERTS_SYS_TIME_INTERNAL_STATE_READ_ONLY__ +struct sys_time_internal_state_read_only__ { +#if defined(OS_MONOTONIC_TIME_USING_TIMES) + int times_shift; +#endif +}; +#endif + +#ifdef ERTS_SYS_TIME_INTERNAL_STATE_READ_MOSTLY__ +struct sys_time_internal_state_read_mostly__ { +#if defined(OS_MONOTONIC_TIME_USING_TIMES) + ErtsOsMonotonicTimeExtendState os_mtime_xtnd; +#endif +}; +#endif + +#ifdef ERTS_SYS_TIME_INTERNAL_STATE_WRITE_FREQ__ +struct sys_time_internal_state_write_freq__ { + erts_smp_mtx_t mtx; +#if defined(__linux__) && defined(OS_MONOTONIC_TIME_USING_CLOCK_GETTIME) + ErtsMonotonicTime last_delivered; +#endif +}; +#endif + +#if defined(ERTS_SYS_TIME_INTERNAL_STATE_READ_ONLY__) \ + || defined(ERTS_SYS_TIME_INTERNAL_STATE_WRITE_FREQ__) +static struct { +#ifdef ERTS_SYS_TIME_INTERNAL_STATE_READ_ONLY__ + union { + struct sys_time_internal_state_read_only__ o; + char align__[(((sizeof(struct sys_time_internal_state_read_only__) - 1) + / ASSUMED_CACHE_LINE_SIZE) + 1) + * ASSUMED_CACHE_LINE_SIZE]; + } r; +#endif +#ifdef ERTS_SYS_TIME_INTERNAL_STATE_READ_MOSTLY__ + union { + struct sys_time_internal_state_read_mostly__ m; + char align__[(((sizeof(struct sys_time_internal_state_read_mostly__) - 1) + / ASSUMED_CACHE_LINE_SIZE) + 1) + * ASSUMED_CACHE_LINE_SIZE]; + } wr; +#endif +#ifdef ERTS_SYS_TIME_INTERNAL_STATE_WRITE_FREQ__ + union { + struct sys_time_internal_state_write_freq__ f; + char align__[(((sizeof(struct sys_time_internal_state_write_freq__) - 1) + / ASSUMED_CACHE_LINE_SIZE) + 1) + * ASSUMED_CACHE_LINE_SIZE]; + } w; +#endif +} internal_state erts_align_attribute(ERTS_CACHE_LINE_SIZE); +#endif + +void +sys_init_time(ErtsSysInitTimeResult *init_resp) { +#if !defined(ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT) + + init_resp->have_os_monotonic_time = 0; + +#else /* defined(ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT) */ + + int major, minor, build, vsn; + + init_resp->os_monotonic_time_info.resolution = (Uint64) 1000*1000*1000; +#if defined(HAVE_CLOCK_GETRES) && defined(MONOTONIC_CLOCK_ID) + { + struct timespec ts; + if (clock_getres(MONOTONIC_CLOCK_ID, &ts) == 0) { + if (ts.tv_sec == 0 && ts.tv_nsec != 0) + init_resp->os_monotonic_time_info.resolution /= ts.tv_nsec; + else if (ts.tv_sec >= 1) + init_resp->os_monotonic_time_info.resolution = 1; + } + } +#elif defined(ERTS_HAVE_MACH_CLOCK_GETRES) && defined(MONOTONIC_CLOCK_ID) + init_resp->os_monotonic_time_info.resolution + = mach_clock_getres(MONOTONIC_CLOCK_ID, MONOTONIC_CLOCK_ID_STR); +#endif + +#ifdef MONOTONIC_CLOCK_ID_STR + init_resp->os_monotonic_time_info.clock_id = MONOTONIC_CLOCK_ID_STR; +#else + init_resp->os_monotonic_time_info.clock_id = NULL; +#endif + + init_resp->os_monotonic_time_info.locked_use = 0; + +#if defined(OS_MONOTONIC_TIME_USING_CLOCK_GETTIME) + init_resp->os_monotonic_time_info.func = "clock_gettime"; +#elif defined(OS_MONOTONIC_TIME_USING_MACH_CLOCK_GET_TIME) + init_resp->os_monotonic_time_info.func = "clock_get_time"; +#elif defined(OS_MONOTONIC_TIME_USING_GETHRTIME) + init_resp->os_monotonic_time_info.func = "gethrtime"; +#elif defined(OS_MONOTONIC_TIME_USING_TIMES) + init_resp->os_monotonic_time_info.func = "times"; +#else +# error Unknown erts_os_monotonic_time() implementation +#endif + + init_resp->have_os_monotonic_time = 1; + + os_version(&major, &minor, &build); + + vsn = ERTS_MK_VSN_INT(major, minor, build); + + +#if defined(__linux__) && defined(OS_MONOTONIC_TIME_USING_CLOCK_GETTIME) + if (vsn >= ERTS_MK_VSN_INT(2, 6, 33)) { + erts_sys_time_data__.r.o.os_monotonic_time = + clock_gettime_monotonic_raw; +#if defined(OS_SYSTEM_TIME_USING_CLOCK_GETTIME) + erts_sys_time_data__.r.o.os_times = + clock_gettime_times_raw; +#endif + } + else { + /* + * Linux versions prior to 2.6.33 have a + * known bug that sometimes cause monotonic + * time to take small steps backwards. + */ + erts_sys_time_data__.r.o.os_monotonic_time = + clock_gettime_monotonic_verified; +#if defined(OS_SYSTEM_TIME_USING_CLOCK_GETTIME) + erts_sys_time_data__.r.o.os_times = + clock_gettime_times_verified; +#endif + erts_smp_mtx_init(&internal_state.w.f.mtx, + "os_monotonic_time"); + internal_state.w.f.last_delivered + = clock_gettime_monotonic_raw(); + init_resp->os_monotonic_time_info.locked_use = 1; + } +#else /* !(defined(__linux__) && defined(OS_MONOTONIC_TIME_USING_CLOCK_GETTIME)) */ + { + char flavor[1024]; + + os_flavor(flavor, sizeof(flavor)); + + if (sys_strcmp(flavor, "sunos") == 0) { + /* + * Don't trust hrtime on multi processors + * on SunOS prior to SunOS 5.8 + */ + if (vsn < ERTS_MK_VSN_INT(5, 8, 0)) { +#if defined(HAVE_SYSCONF) && defined(_SC_NPROCESSORS_CONF) + if (sysconf(_SC_NPROCESSORS_CONF) > 1) +#endif + init_resp->have_os_monotonic_time = 0; + } + } + } +#endif /* !(defined(__linux__) && defined(OS_MONOTONIC_TIME_USING_CLOCK_GETTIME)) */ + +#endif /* defined(ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT) */ + +#ifdef ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT + init_resp->os_monotonic_time_unit = ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT; +#endif + init_resp->sys_clock_resolution = SYS_CLOCK_RESOLUTION; + /* - * This (erts_ticks_per_sec) is only for times() (CLK_TCK), - * the resolution is always one millisecond.. + * This (erts_sys_time_data__.r.o.ticks_per_sec) is only for + * times() (CLK_TCK), the resolution is always one millisecond.. */ - if ((erts_ticks_per_sec = TICKS_PER_SEC()) < 0) - erl_exit(1, "Can't get clock ticks/sec\n"); - if (erts_ticks_per_sec >= 1000) { - /* Workaround for beta linux kernels, need to be done in runtime - to make erlang run on both 2.4 and 2.5 kernels. In the future, - the kernel ticks might as - well be used as a high res timer instead, but that's for when the - majority uses kernels with HZ == 1024 */ - ticks_bsr = 3; - } else { - ticks_bsr = 0; + if ((erts_sys_time_data__.r.o.ticks_per_sec = TICKS_PER_SEC()) < 0) + erl_exit(ERTS_ABORT_EXIT, "Can't get clock ticks/sec\n"); + +#if defined(OS_MONOTONIC_TIME_USING_TIMES) +#if ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT +# error Time unit is supposed to be determined at runtime... +#endif + { + ErtsMonotonicTime resolution = erts_sys_time_data__.r.o.ticks_per_sec; + ErtsMonotonicTime time_unit = resolution; + int shift = 0; + + while (time_unit < 1000*1000) { + time_unit <<= 1; + shift++; + } + + init_resp->os_monotonic_time_info.resolution = resolution; + init_resp->os_monotonic_time_unit = time_unit; + init_resp->os_monotonic_time_info.extended = 1; + internal_state.r.o.times_shift = shift; + + erts_init_os_monotonic_time_extender(&internal_state.wr.m.os_mtime_xtnd, + get_tick_count, + (1 << 29) / resolution); } - erts_ticks_per_sec_wrap = (erts_ticks_per_sec >> ticks_bsr); - return SYS_CLOCK_RESOLUTION; +#endif /* defined(OS_MONOTONIC_TIME_USING_TIMES) */ + +#ifdef WALL_CLOCK_ID_STR + init_resp->os_system_time_info.clock_id = WALL_CLOCK_ID_STR; +#else + init_resp->os_system_time_info.clock_id = NULL; +#endif + + init_resp->os_system_time_info.locked_use = 0; + init_resp->os_system_time_info.resolution = (Uint64) 1000*1000*1000; +#if defined(HAVE_CLOCK_GETRES) && defined(WALL_CLOCK_ID) + { + struct timespec ts; + if (clock_getres(WALL_CLOCK_ID, &ts) == 0) { + if (ts.tv_sec == 0 && ts.tv_nsec != 0) + init_resp->os_system_time_info.resolution /= ts.tv_nsec; + else if (ts.tv_sec >= 1) + init_resp->os_system_time_info.resolution = 1; + } + } +#elif defined(ERTS_HAVE_MACH_CLOCK_GETRES) && defined(WALL_CLOCK_ID) + init_resp->os_system_time_info.resolution + = mach_clock_getres(WALL_CLOCK_ID, WALL_CLOCK_ID_STR); +#endif + +#if defined(OS_SYSTEM_TIME_USING_CLOCK_GETTIME) + init_resp->os_system_time_info.func = "clock_gettime"; +#elif defined(OS_SYSTEM_TIME_USING_MACH_CLOCK_GET_TIME) + init_resp->os_system_time_info.func = "clock_get_time"; +#elif defined(OS_SYSTEM_TIME_GETTIMEOFDAY) + init_resp->os_system_time_info.func = "gettimeofday"; + init_resp->os_system_time_info.resolution = 1000*1000; + init_resp->os_system_time_info.clock_id = NULL; +#else +# error Missing erts_os_system_time() implementation +#endif + +} + +void +erts_late_sys_init_time(void) +{ +#if defined(OS_MONOTONIC_TIME_USING_TIMES) + erts_late_init_os_monotonic_time_extender(&internal_state.wr.m.os_mtime_xtnd); +#endif } -clock_t sys_times_wrap(void) +static ERTS_INLINE ErtsSystemTime +adj_stime_time_unit(ErtsSystemTime stime, Uint32 res) { - SysTimes dummy; - clock_t result = (sys_times(&dummy) >> ticks_bsr); - return result; + if (res == ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT) + return stime; + if (res == (Uint32) 1000*1000*1000 + && ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT == 1000*1000) + return stime/1000; + if (res == (Uint32) 1000*1000 + && ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT == 1000*1000*1000) + return stime*1000; + return ((ErtsSystemTime) + erts_time_unit_conversion(stime, + (Uint32) res, + (Uint32) ERTS_MONOTONIC_TIME_UNIT)); } +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *\ + * POSIX clock_gettime() * +\* */ +#if defined(OS_MONOTONIC_TIME_USING_CLOCK_GETTIME) \ + || defined(OS_SYSTEM_TIME_USING_CLOCK_GETTIME) + +static ERTS_INLINE ErtsMonotonicTime +timespec2montime(struct timespec *ts) +{ + ErtsMonotonicTime time; + time = (ErtsMonotonicTime) ts->tv_sec; + time *= (ErtsMonotonicTime) 1000*1000*1000; + time += (ErtsMonotonicTime) ts->tv_nsec; + return time; +} + +static ERTS_INLINE ErtsMonotonicTime +posix_clock_gettime(clockid_t id, char *name) +{ + struct timespec ts; + + if (clock_gettime(id, &ts) != 0) { + int err = errno; + char *errstr = err ? strerror(err) : "unknown"; + erl_exit(ERTS_ABORT_EXIT, + "clock_gettime(%s, _) failed: %s (%d)\n", + name, errstr, err); + } + return timespec2montime(&ts); +} + +#endif /* defined(OS_MONOTONIC_TIME_USING_CLOCK_GETTIME) \ + || defined(OS_SYSTEM_TIME_USING_CLOCK_GETTIME) */ + +#if defined(OS_SYSTEM_TIME_USING_CLOCK_GETTIME) + +ErtsSystemTime +erts_os_system_time(void) +{ + ErtsSystemTime stime; + + stime = (ErtsSystemTime) posix_clock_gettime(WALL_CLOCK_ID, + WALL_CLOCK_ID_STR); + return adj_stime_time_unit(stime, (Uint32) 1000*1000*1000); +} + +#endif /* defined(OS_SYSTEM_TIME_USING_CLOCK_GETTIME) */ + +#if defined(OS_MONOTONIC_TIME_USING_CLOCK_GETTIME) + +#if defined(OS_SYSTEM_TIME_USING_CLOCK_GETTIME) + +#define ERTS_HAVE_ERTS_OS_TIMES_IMPL__ + +static ERTS_INLINE void +posix_clock_gettime_times(ErtsMonotonicTime *mtimep, + ErtsSystemTime *stimep) +{ + struct timespec mts, sts; + int mres, sres, merr, serr; + + mres = clock_gettime(MONOTONIC_CLOCK_ID, &mts); + merr = errno; + sres = clock_gettime(WALL_CLOCK_ID, &sts); + serr = errno; + + if (mres != 0) { + char *errstr = merr ? strerror(merr) : "unknown"; + erl_exit(ERTS_ABORT_EXIT, + "clock_gettime(%s, _) failed: %s (%d)\n", + MONOTONIC_CLOCK_ID_STR, errstr, merr); + } + if (sres != 0) { + char *errstr = serr ? strerror(serr) : "unknown"; + erl_exit(ERTS_ABORT_EXIT, + "clock_gettime(%s, _) failed: %s (%d)\n", + WALL_CLOCK_ID_STR, errstr, serr); + } + + *mtimep = timespec2montime(&mts); + *stimep = (ErtsSystemTime) timespec2montime(&sts); +} + +#endif /* defined(OS_SYSTEM_TIME_USING_CLOCK_GETTIME) */ + +#if defined(__linux__) + +static ErtsMonotonicTime clock_gettime_monotonic_verified(void) +{ + ErtsMonotonicTime mtime = posix_clock_gettime(MONOTONIC_CLOCK_ID, + MONOTONIC_CLOCK_ID_STR); + + erts_smp_mtx_lock(&internal_state.w.f.mtx); + if (mtime < internal_state.w.f.last_delivered) + mtime = internal_state.w.f.last_delivered; + else + internal_state.w.f.last_delivered = mtime; + erts_smp_mtx_unlock(&internal_state.w.f.mtx); + + return mtime; +} + +#if defined(OS_SYSTEM_TIME_USING_CLOCK_GETTIME) + +static void clock_gettime_times_verified(ErtsMonotonicTime *mtimep, + ErtsSystemTime *stimep) +{ + posix_clock_gettime_times(mtimep, stimep); + + erts_smp_mtx_lock(&internal_state.w.f.mtx); + if (*mtimep < internal_state.w.f.last_delivered) + *mtimep = internal_state.w.f.last_delivered; + else + internal_state.w.f.last_delivered = *mtimep; + erts_smp_mtx_unlock(&internal_state.w.f.mtx); +} + +#endif /* defined(OS_SYSTEM_TIME_USING_CLOCK_GETTIME) */ + +static ErtsMonotonicTime clock_gettime_monotonic_raw(void) +{ + return posix_clock_gettime(MONOTONIC_CLOCK_ID, + MONOTONIC_CLOCK_ID_STR); +} + +#if defined(OS_SYSTEM_TIME_USING_CLOCK_GETTIME) + +static void clock_gettime_times_raw(ErtsMonotonicTime *mtimep, + ErtsSystemTime *stimep) +{ + posix_clock_gettime_times(mtimep, stimep); +} + +#endif /* defined(OS_SYSTEM_TIME_USING_CLOCK_GETTIME) */ + +#else /* !defined(__linux__) */ + +ErtsMonotonicTime erts_os_monotonic_time(void) +{ + return posix_clock_gettime(MONOTONIC_CLOCK_ID, + MONOTONIC_CLOCK_ID_STR); +} + +#if defined(OS_SYSTEM_TIME_USING_CLOCK_GETTIME) + +void erts_os_times(ErtsMonotonicTime *mtimep, ErtsSystemTime *stimep) +{ + posix_clock_gettime_times(mtimep, stimep); +} + +#endif /* defined(OS_SYSTEM_TIME_USING_CLOCK_GETTIME) */ + +#endif /* !defined(__linux__) */ + +#define ERTS_HAVE_ERTS_SYS_HRTIME_IMPL__ + +ErtsSysHrTime +erts_sys_hrtime(void) +{ + return (ErtsSysHrTime) posix_clock_gettime(MONOTONIC_CLOCK_ID, + MONOTONIC_CLOCK_ID_STR); +} + +#endif /* defined(OS_MONOTONIC_TIME_USING_CLOCK_GETTIME) */ + +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *\ + * MACH clock_get_time() * +\* */ + +#if defined(OS_MONOTONIC_TIME_USING_MACH_CLOCK_GET_TIME) \ + || defined(OS_SYSTEM_TIME_USING_MACH_CLOCK_GET_TIME) + +#ifdef ERTS_HAVE_MACH_CLOCK_GETRES + +static Sint64 +mach_clock_getres(clock_id_t clkid, char *clkid_str) +{ + mach_port_t task; + host_name_port_t host; + natural_t attr[1]; + kern_return_t kret; + clock_serv_t clk_srv; + mach_msg_type_number_t cnt; + + host = mach_host_self(); + kret = host_get_clock_service(host, clkid, &clk_srv); + if (kret != KERN_SUCCESS) { + erl_exit(ERTS_ABORT_EXIT, + "host_get_clock_service(_, %s, _) failed\n", + clkid_str); + } + + cnt = sizeof(attr); + kret = clock_get_attributes(clk_srv, CLOCK_GET_TIME_RES, (clock_attr_t) attr, &cnt); + if (kret != KERN_SUCCESS) { + erl_exit(ERTS_ABORT_EXIT, + "clock_get_attributes(%s, _) failed\n", + clkid_str); + } + task = mach_task_self(); + mach_port_deallocate(task, host); + mach_port_deallocate(task, clk_srv); + + return (Sint64) attr[0]; +} + +#endif /* ERTS_HAVE_MACH_CLOCK_GETRES */ + +static ERTS_INLINE Sint64 +mach_clock_gettime(clock_id_t clkid, char *clkid_str) +{ + Sint64 time; + mach_port_t task; + host_name_port_t host; + kern_return_t kret; + clock_serv_t clk_srv; + mach_timespec_t time_spec; + + host = mach_host_self(); + kret = host_get_clock_service(host, clkid, &clk_srv); + if (kret != KERN_SUCCESS) { + erl_exit(ERTS_ABORT_EXIT, + "host_get_clock_service(_, %s, _) failed\n", + clkid_str); + } + errno = 0; + kret = clock_get_time(clk_srv, &time_spec); + if (kret != KERN_SUCCESS) { + erl_exit(ERTS_ABORT_EXIT, + "clock_get_time(%s, _) failed\n", + clkid_str); + } + task = mach_task_self(); + mach_port_deallocate(task, host); + mach_port_deallocate(task, clk_srv); + + time = (Sint64) time_spec.tv_sec; + time *= (Sint64) 1000*1000*1000; + time += (Sint64) time_spec.tv_nsec; + return time; +} + +#endif /* defined(OS_MONOTONIC_TIME_USING_MACH_CLOCK_GET_TIME) \ + || defined(OS_SYSTEM_TIME_USING_MACH_CLOCK_GET_TIME) */ + +#if defined(OS_SYSTEM_TIME_USING_MACH_CLOCK_GET_TIME) + +#define ERTS_HAVE_ERTS_OS_TIMES_IMPL__ + +ErtsSystemTime +erts_os_system_time(void) +{ + ErtsSystemTime stime; + stime = (ErtsSystemTime) mach_clock_gettime(WALL_CLOCK_ID, + WALL_CLOCK_ID_STR); + return adj_stime_time_unit(stime, (Uint32) 1000*1000*1000); +} + +#endif /* defined(OS_SYSTEM_TIME_USING_MACH_CLOCK_GET_TIME) */ + +#if defined(OS_MONOTONIC_TIME_USING_MACH_CLOCK_GET_TIME) + +ErtsMonotonicTime +erts_os_monotonic_time(void) +{ + return (ErtsMonotonicTime) mach_clock_gettime(MONOTONIC_CLOCK_ID, + MONOTONIC_CLOCK_ID_STR); +} + +#define ERTS_HAVE_ERTS_SYS_HRTIME_IMPL__ + +ErtsSysHrTime +erts_sys_hrtime(void) +{ + return (ErtsMonotonicTime) mach_clock_gettime(MONOTONIC_CLOCK_ID, + MONOTONIC_CLOCK_ID_STR); +} + +#if defined(OS_SYSTEM_TIME_USING_MACH_CLOCK_GET_TIME) + +#define ERTS_HAVE_ERTS_OS_TIMES_IMPL__ + +void +erts_os_times(ErtsMonotonicTime *mtimep, ErtsSystemTime *stimep) +{ + ErtsMonotonicTime mtime; + ErtsSystemTime stime; + mach_port_t task; + host_name_port_t host; + kern_return_t mkret, skret; + clock_serv_t mclk_srv, sclk_srv; + mach_timespec_t mon_time_spec, sys_time_spec; + + host = mach_host_self(); + mkret = host_get_clock_service(host, MONOTONIC_CLOCK_ID, &mclk_srv); + skret = host_get_clock_service(host, WALL_CLOCK_ID, &sclk_srv); + if (mkret != KERN_SUCCESS) { + erl_exit(ERTS_ABORT_EXIT, + "host_get_clock_service(_, %s, _) failed\n", + MONOTONIC_CLOCK_ID); + } + if (skret != KERN_SUCCESS) { + erl_exit(ERTS_ABORT_EXIT, + "host_get_clock_service(_, %s, _) failed\n", + WALL_CLOCK_ID); + } + mkret = clock_get_time(mclk_srv, &mon_time_spec); + skret = clock_get_time(sclk_srv, &sys_time_spec); + if (mkret != KERN_SUCCESS) { + erl_exit(ERTS_ABORT_EXIT, + "clock_get_time(%s, _) failed\n", + MONOTONIC_CLOCK_ID); + } + if (skret != KERN_SUCCESS) { + erl_exit(ERTS_ABORT_EXIT, + "clock_get_time(%s, _) failed\n", + WALL_CLOCK_ID); + } + task = mach_task_self(); + mach_port_deallocate(task, host); + mach_port_deallocate(task, mclk_srv); + mach_port_deallocate(task, sclk_srv); + + mtime = (ErtsMonotonicTime) mon_time_spec.tv_sec; + mtime *= (ErtsMonotonicTime) 1000*1000*1000; + mtime += (ErtsMonotonicTime) mon_time_spec.tv_nsec; + stime = (ErtsSystemTime) sys_time_spec.tv_sec; + stime *= (ErtsSystemTime) 1000*1000*1000; + stime += (ErtsSystemTime) sys_time_spec.tv_nsec; + *mtimep = mtime; + *stimep = stime; +} + +#endif /* defined(OS_SYSTEM_TIME_USING_MACH_CLOCK_GET_TIME) */ + +#endif /* defined(OS_MONOTONIC_TIME_USING_MACH_CLOCK_GET_TIME) */ + +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *\ + * Solaris gethrtime() - OS monotonic time * +\* */ + +#if defined(OS_MONOTONIC_TIME_USING_GETHRTIME) + +ErtsMonotonicTime erts_os_monotonic_time(void) +{ + return (ErtsMonotonicTime) gethrtime(); +} + +#define ERTS_HAVE_ERTS_SYS_HRTIME_IMPL__ + +ErtsSysHrTime +erts_sys_hrtime(void) +{ + return (ErtsSysHrTime) gethrtime(); +} + +#endif /* defined(OS_MONOTONIC_TIME_USING_GETHRTIME) */ + +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *\ + * gettimeofday() - OS system time * +\* */ + +#if defined(OS_SYSTEM_TIME_GETTIMEOFDAY) + +ErtsSystemTime +erts_os_system_time(void) +{ + ErtsSystemTime stime; + struct timeval tv; + + if (gettimeofday(&tv, NULL) != 0) { + int err = errno; + char *errstr = err ? strerror(err) : "unknown"; + erl_exit(ERTS_ABORT_EXIT, + "gettimeofday(_, NULL) failed: %s (%d)\n", + errstr, err); + } + + stime = (ErtsSystemTime) tv.tv_sec; + stime *= (ErtsSystemTime) 1000*1000; + stime += (ErtsSystemTime) tv.tv_usec; + + return adj_stime_time_unit(stime, (Uint32) 1000*1000); +} + +#endif /* defined(OS_SYSTEM_TIME_GETTIMEOFDAY) */ + +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *\ + * times() - OS monotonic time * +\* */ + +#if defined(OS_MONOTONIC_TIME_USING_TIMES) + +ErtsMonotonicTime +erts_os_monotonic_time(void) +{ + Uint32 ticks = get_tick_count(); + ERTS_CHK_EXTEND_OS_MONOTONIC_TIME(&internal_state.wr.m.os_mtime_xtnd, + ticks); + return ERTS_EXTEND_OS_MONOTONIC_TIME(&internal_state.wr.m.os_mtime_xtnd, + ticks) << internal_state.r.o.times_shift; +} + +#endif + +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *\ + * Fallbacks * +\* */ + +#ifndef ERTS_HAVE_ERTS_SYS_HRTIME_IMPL__ + +ErtsSysHrTime +erts_sys_hrtime(void) +{ + return (ErtsSysHrTime) ERTS_MONOTONIC_TO_NSEC(erts_os_system_time()); +} + +#endif + +#if !defined(ERTS_HAVE_ERTS_OS_TIMES_IMPL__) \ + && defined(ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT) + +void +erts_os_times(ErtsMonotonicTime *mtimep, ErtsSystemTime *stimep) +{ + *mtimep = erts_os_monotonic_time(); + *stimep = erts_os_system_time(); +} + +#endif +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ #ifdef HAVE_GETHRVTIME_PROCFS_IOCTL |