| Message ID | 1322516429-8471-1-git-send-email-weisserm@arcor.de |
|---|---|
| State | Superseded |
| Delegated to: | Mike Frysinger |
| Headers | show |
On Monday 28 November 2011 16:40:29 Matthias Weisser wrote:
> Making sleep command work
i like the idea, but using clock_xxx funcs makes me sad (generally requires a
"newerish" glibc and -lrt). my only alternative proposal is to use usec's
with gettimeofday ... maybe that's good enough ? although we'll probably need
to add nsec support eventually anyways for the newer timer code ...
feel like wiring up timer_get_us() too ?
-mike
Am 28.11.2011 23:57, schrieb Mike Frysinger: > On Monday 28 November 2011 16:40:29 Matthias Weisser wrote: >> Making sleep command work > > i like the idea, but using clock_xxx funcs makes me sad (generally requires a > "newerish" glibc and -lrt). my only alternative proposal is to use usec's > with gettimeofday ... maybe that's good enough ? although we'll probably need > to add nsec support eventually anyways for the newer timer code ... I would like to stay with clock_gettime(CLOCK_MONOTONIC, ...) as the value isn't subjected to change when wall time is changed. Still there are chances to get wrong (non monotonic) values when NTP adjusts frequencies. Maybe some ifdef magic can be used to even use CLOCK_MONOTONIC_RAW if it is available and fall back to gettimeofday in case of older systems. The only thing would be then -lrt on the linker command line which has to be sorted out somehow. > feel like wiring up timer_get_us() too ? Will do in a v2. Matthias
On Tuesday 29 November 2011 04:35:06 Matthias Weißer wrote: > Am 28.11.2011 23:57, schrieb Mike Frysinger: > > On Monday 28 November 2011 16:40:29 Matthias Weisser wrote: > >> Making sleep command work > > > > i like the idea, but using clock_xxx funcs makes me sad (generally > > requires a "newerish" glibc and -lrt). my only alternative proposal is > > to use usec's with gettimeofday ... maybe that's good enough ? although > > we'll probably need to add nsec support eventually anyways for the newer > > timer code ... > > I would like to stay with clock_gettime(CLOCK_MONOTONIC, ...) as the > value isn't subjected to change when wall time is changed. Still there > are chances to get wrong (non monotonic) values when NTP adjusts > frequencies. Maybe some ifdef magic can be used to even use > CLOCK_MONOTONIC_RAW if it is available and fall back to gettimeofday in > case of older systems. The only thing would be then -lrt on the linker > command line which has to be sorted out somehow. i think atm, we're focusing on sandbox working on Linux/glibc hosts. we're not *too* worried about portability atm. we can worry about that once sandbox is more than just a PoC toy :). -mike
diff --git a/arch/sandbox/config.mk b/arch/sandbox/config.mk index ab33026..2ec1bb7 100644 --- a/arch/sandbox/config.mk +++ b/arch/sandbox/config.mk @@ -18,3 +18,4 @@ # MA 02111-1307 USA PLATFORM_CPPFLAGS += -DCONFIG_SANDBOX -D__SANDBOX__ +PLATFORM_LIBS += -lrt diff --git a/arch/sandbox/cpu/cpu.c b/arch/sandbox/cpu/cpu.c index c7bf8a9..a15d3c2 100644 --- a/arch/sandbox/cpu/cpu.c +++ b/arch/sandbox/cpu/cpu.c @@ -34,7 +34,7 @@ int do_reset(cmd_tbl_t *cmdtp, int flag, int argc, char * const argv[]) /* delay x useconds */ void __udelay(unsigned long usec) { - /* Ignore this for now */ + os_usleep(usec); } unsigned long timer_get_us(void) diff --git a/arch/sandbox/cpu/os.c b/arch/sandbox/cpu/os.c index b7c3bf5..d3357db 100644 --- a/arch/sandbox/cpu/os.c +++ b/arch/sandbox/cpu/os.c @@ -23,6 +23,7 @@ #include <stdlib.h> #include <termios.h> #include <unistd.h> +#include <time.h> #include <sys/types.h> #include <sys/stat.h> #include <sys/mman.h> @@ -94,3 +95,17 @@ void *os_malloc(size_t length) return mmap(NULL, length, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); } + +void os_usleep(unsigned long usec) +{ + usleep(usec); +} + +u64 os_get_nsec(void) +{ + struct timespec tp; + + clock_gettime(CLOCK_MONOTONIC, &tp); + + return tp.tv_sec * 1000000000ULL + tp.tv_nsec; +} diff --git a/board/sandbox/sandbox/sandbox.c b/board/sandbox/sandbox/sandbox.c index 95a5045..f376c74 100644 --- a/board/sandbox/sandbox/sandbox.c +++ b/board/sandbox/sandbox/sandbox.c @@ -21,6 +21,8 @@ #include <common.h> +#include <os.h> + /* * Pointer to initial global data area * @@ -34,7 +36,7 @@ void flush_cache(unsigned long start, unsigned long size) ulong get_timer(ulong base) { - return 0; + return (os_get_nsec() / 1000000) - base; } int timer_init(void) diff --git a/include/os.h b/include/os.h index fd4120c..f3af4f0 100644 --- a/include/os.h +++ b/include/os.h @@ -84,3 +84,17 @@ void os_tty_raw(int fd); * \return Pointer to length bytes or NULL on error */ void *os_malloc(size_t length); + +/** + * Access to the usleep function of the os + * + * \param usec Time to sleep in micro seconds + */ +void os_usleep(unsigned long usec); + +/** + * Gets a monotonic increasing number of nano seconds from the OS + * + * \return A monotonic increasing time scaled in nano seconds + */ +u64 os_get_nsec(void);
Making sleep command work Signed-off-by: Matthias Weisser <weisserm@arcor.de> --- arch/sandbox/config.mk | 1 + arch/sandbox/cpu/cpu.c | 2 +- arch/sandbox/cpu/os.c | 15 +++++++++++++++ board/sandbox/sandbox/sandbox.c | 4 +++- include/os.h | 14 ++++++++++++++ 5 files changed, 34 insertions(+), 2 deletions(-)