Re: RTC subsystem and fractions of seconds

[Philippe De Muyter <phdm@macqel.be>]

On Sun, 7 Jan 2007, Hugh Dickins wrote:
> Author: Matt Mackall <mpm@selenic.com>
> 
>     [PATCH] RTC: Remove RTC UIP synchronization on x86
>     
>     Reading the CMOS clock on x86 and some other arches currently takes up to one
>     second because it synchronizes with the CMOS second tick-over.  This delay
>     shows up at boot time as well a resume time.

That is true if kernel's idea of an RTC's precision is 1 second.  With better
RTC's (e.g. m41t81) this delay can be reduced to 0.01 second, which is
acceptable IMHO in the boot phase.  That needs however changes in the kernel
interface to RTC's :

    - read_time should take a new parameter *nsec (or struct rtc_time should
    contain a new nsec field, so that the RTC can report its complete time
    information to the kernel.

    - struct rtc_device (or rtc_class_ops with another name) should contain
    a new field giving the RTC's resolution in nsecs, if we want to avoid the
    loop, but do not want our 0.01 second precision be destructed by always
    adding 0.5 second in rtc_hctosys
    
Philippe