On Tue, Dec 5, 2017 at 1:58 AM, Laurent Pinchart firstname.lastname@example.org wrote:
On Tuesday, 5 December 2017 02:37:27 EET Laurent Pinchart wrote:
On Monday, 27 November 2017 15:19:54 EET Arnd Bergmann wrote:
uvc_video_get_ts() returns a 'struct timespec', but all its users really want a nanoseconds variable anyway.
Changing the deprecated ktime_get_ts/ktime_get_real_ts to ktime_get and ktime_get_real simplifies the code noticeably, while keeping the resulting numbers unchanged.
Signed-off-by: Arnd Bergmann email@example.com
drivers/media/usb/uvc/uvc_video.c | 37 ++++++++++++--------------------- drivers/media/usb/uvc/uvcvideo.h | 2 +- 2 files changed, 13 insertions(+), 26 deletions(-)
- struct timespec ts;
- u64 timestamp;
uvc_trace(UVC_TRACE_CLOCK, "%s: SOF %u.%06llu y %llu ts %llu " "buf ts %llu (x1 %u/%u/%u x2 %u/%u/%u y1 %u y2 %u)\n", stream->dev->name, sof >> 16, div_u64(((u64)sof & 0xffff) * 1000000LLU, 65536),
y, timespec_to_ns(&ts), vbuf->vb2_buf.timestamp,
y, timestamp, vbuf->vb2_buf.timestamp, x1, first->host_sof, first->dev_sof, x2, last->host_sof, last->dev_sof, y1, y2);
As you've done lots of work moving code away from timespec I figured out I would ask, what is the preferred way to print a ktime in secs.nsecs format ? Should I use ktime_to_timespec and print ts.tv_sec and ts.tv_nsec, or is there a better way ?
We had patches under discussion to add a special printk format string that would pretty-print a date, but that never got merged. Usually there is a tradeoff between runtime to convert the nanoseconds into a different format and how useful you want it to be. ktime_to_timespec() can be a bit slow on some architectures, since it has to do a 64-bit division, but then again the sprintf logic also needs to do that. If the output isn't on a time-critical path, you can use time64_to_tm and print it in years/months/days/hours/ minutes/seconds, but depending on where it gets printed, that may not be easier to interpret than the seconds/nanoseconds or pure nanoseconds.