libpayload: add gettimeofday()

[Jordan Crouse]

Add a simple gettimeofday() implementation.  If you find my math laughable,
then please fix it. :)

We don't have an OS underneath constantly correcting the time math, so this
will drift away from RTC fairly quickly.  Its good for small intervals
of time, but if you need to figure out what time it is, use the
rtc_read_clock() function (hm - mabye I should rename that localtime?)

Also, add in a NVRAM function to make sure we get a stable date value.

Please review and comment.

[Uwe Hermann]

On Wed, Apr 16, 2008 at 05:44:45PM -0600, Jordan Crouse wrote:
> libpayload:  Add gettimeofday() and friends
>
> Add a gettimeofday() implementation - it works pretty well, but it
> drifts a little bit so its not very suitable for keeping time.  It
> works best to track changes in time over small periods of time.
>
> Signed-off-by: Jordan Crouse 
> Index: libpayload/drivers/nvram.c
> ===================================================================
> --- libpayload.orig/drivers/nvram.c	2008-04-18 04:16:35.000000000 -0600
> +++ libpayload/drivers/nvram.c	2008-04-18 04:20:20.000000000 -0600
> @@ -93,3 +93,42 @@
>    outb(addr, rtc_port);
>    outb(val, rtc_port + 1);
>  }
>
> +/**
> Return 1 if the NVRAM is currently updating
> */

... "and zero otherwise."

Also, please add a "@return" item for doxygen.
> +int nvram_updating(void)
> +{
>   return (nvram_read(NVRAM_RTC_FREQ_SELECT) & NVRAM_RTC_UIP) ? 1 : 0;
>
> +}
>
> +/**
> Get the current time and date from the RTC
> @param time A pointer to a broken-down time structure
> */
>
> +void rtc_read_clock(struct tm *time)
> +{
> memset(time, 0, sizeof(*time));
> while(nvram_updating());
> time->tm_mon = bcd2dec(nvram_read(NVRAM_RTC_MONTH)) - 1;
> time->tm_sec = bcd2dec(nvram_read(NVRAM_RTC_SECONDS));
> time->tm_min = bcd2dec(nvram_read(NVRAM_RTC_MINUTES));
> time->tm_mday = bcd2dec(nvram_read(NVRAM_RTC_DAY));
> time->tm_hour = bcd2dec(nvram_read(NVRAM_RTC_HOURS));
> /* Instead of finding the century register,
>   we just make an assumption that if the year value is
>   less then 80, then it is 2000+
> */

I'd add a FIXME or TODO here :)
> time->tm_year = bcd2dec(nvram_read(NVRAM_RTC_YEAR));
> if (time->tm_year < 80)
> time->tm_year += 100;
>
> +}
> Index: libpayload/include/arch/rdtsc.h
>
> --- libpayload.orig/include/arch/rdtsc.h	2008-04-18 04:16:34.000000000 -0600
> +++ libpayload/include/arch/rdtsc.h	2008-04-18 04:17:09.000000000 -0600
> @@ -30,9 +30,9 @@
>  #ifndef _ARCH_RDTSC_H
>  #define _ARCH_RDTSC_H
>
> -static inline unsigned long long rdtsc(void)
> +static inline u64 rdtsc(void)
>  {
> unsigned long long val;
> u64 val;

Yep.

Can the inline be removed as well or is it required here?
> +static unsigned int day_of_year(int mon, int day, int year)
> +{
> static u8 mdays[12] = { 31, 28, 31, 30, 31, 30,
> 		31, 31, 30, 31, 30, 31 };
> int ret = 0, i;

Maybe this?

int i, ret = 0;

Looks a bit less confusing. Or only "int ret, i" and put the
"ret = 0" in the for-loop:

for(i = 0, ret = 0; i < mon; i++)
> f(r(i = 0; i < mon; i++) {
> ret += mdays[i];
> if (i == 1 && (year % 4))
> 	ret++;
> }
> return (ret + day);
>
> +}
>
> +static void gtod_init(void)

Nonovious abbreviation, use gettimeofday_init() for better readability
please. Ditto for other "gtod" occurences.

A test on real hardware is pending, I'll report back.

Uwe.

[Jordan Crouse]

On 21/04/08 21:02 +0200, Uwe Hermann wrote:
> > /* Instead of finding the century register,
> >   we just make an assumption that if the year value is
> >   less then 80, then it is 2000+
> > */
>
> I'd add a FIXME or TODO here :)

*is* it a TODO?  The century register is implementation dependent.  
This is not a 5 9s OS here, and I would pefer simple over complex
#ifdefs.  Or not?
> > @@ -30,9 +30,9 @@
> >  #ifndef _ARCH_RDTSC_H
> >  #define _ARCH_RDTSC_H
> >
> > -static inline unsigned long long rdtsc(void)
> > +static inline u64 rdtsc(void)
> >  {
> > unsigned long long val;
> > u64 val;
>
> Yep.
>
> Can the inline be removed as well or is it required here?

No - for modern compilers, we should go ahead and remove inlines
everywhere.

Jordan

[Jordan Crouse]

Changes made - awaiting your ACK.

On 21/04/08 21:02 +0200, Uwe Hermann wrote:
> On Wed, Apr 16, 2008 at 05:44:45PM -0600, Jordan Crouse wrote:
> > libpayload:  Add gettimeofday() and friends
> >
> > Add a gettimeofday() implementation - it works pretty well, but it
> > drifts a little bit so its not very suitable for keeping time.  It
> > works best to track changes in time over small periods of time.
> >
> > Signed-off-by: Jordan Crouse 
> > Index: libpayload/drivers/nvram.c
> > ===================================================================
> > --- libpayload.orig/drivers/nvram.c	2008-04-18 04:16:35.000000000 -0600
> > +++ libpayload/drivers/nvram.c	2008-04-18 04:20:20.000000000 -0600
> > @@ -93,3 +93,42 @@
> >    outb(addr, rtc_port);
> >    outb(val, rtc_port + 1);
> >  }
> >
> > +/**
> > Return 1 if the NVRAM is currently updating
> > */
>
> ... "and zero otherwise."
>
> Also, please add a "@return" item for doxygen.
> > +int nvram_updating(void)
> > +{
> >   return (nvram_read(NVRAM_RTC_FREQ_SELECT) & NVRAM_RTC_UIP) ? 1 : 0;
> >
> > +}
> >
> > +/**
> > Get the current time and date from the RTC
> > @param time A pointer to a broken-down time structure
> > */
> >
> > +void rtc_read_clock(struct tm *time)
> > +{
> > memset(time, 0, sizeof(*time));
> > while(nvram_updating());
> > time->tm_mon = bcd2dec(nvram_read(NVRAM_RTC_MONTH)) - 1;
> > time->tm_sec = bcd2dec(nvram_read(NVRAM_RTC_SECONDS));
> > time->tm_min = bcd2dec(nvram_read(NVRAM_RTC_MINUTES));
> > time->tm_mday = bcd2dec(nvram_read(NVRAM_RTC_DAY));
> > time->tm_hour = bcd2dec(nvram_read(NVRAM_RTC_HOURS));
> > /* Instead of finding the century register,
> >   we just make an assumption that if the year value is
> >   less then 80, then it is 2000+
> > */
>
> I'd add a FIXME or TODO here :)
> > time->tm_year = bcd2dec(nvram_read(NVRAM_RTC_YEAR));
> > if (time->tm_year < 80)
> > time->tm_year += 100;
> >
> > +}
> > Index: libpayload/include/arch/rdtsc.h
> >
> > --- libpayload.orig/include/arch/rdtsc.h	2008-04-18 04:16:34.000000000 -0600
> > +++ libpayload/include/arch/rdtsc.h	2008-04-18 04:17:09.000000000 -0600
> > @@ -30,9 +30,9 @@
> >  #ifndef _ARCH_RDTSC_H
> >  #define _ARCH_RDTSC_H
> >
> > -static inline unsigned long long rdtsc(void)
> > +static inline u64 rdtsc(void)
> >  {
> > unsigned long long val;
> > u64 val;
>
> Yep.
>
> Can the inline be removed as well or is it required here?
> > +static unsigned int day_of_year(int mon, int day, int year)
> > +{
> > static u8 mdays[12] = { 31, 28, 31, 30, 31, 30,
> > 		31, 31, 30, 31, 30, 31 };
> > int ret = 0, i;
>
> Maybe this?
>
> int i, ret = 0;
>
> Looks a bit less confusing. Or only "int ret, i" and put the
> "ret = 0" in the for-loop:
>
> for(i = 0, ret = 0; i < mon; i++)
> > f(r(i = 0; i < mon; i++) {
> > ret += mdays[i];
> > if (i == 1 && (year % 4))
> > 	ret++;
> > }
> > return (ret + day);
> >
> > +}
> >
> > +static void gtod_init(void)
>
> Nonovious abbreviation, use gettimeofday_init() for better readability
> please. Ditto for other "gtod" occurences.
>
> A test on real hardware is pending, I'll report back.
> Uwe.
>
> http://www.hermann-uwe.de  | http://www.holsham-traders.de
> http://www.crazy-hacks.org | http://www.unmaintained-free-software.org

[Uwe Hermann]

On Tue, Apr 22, 2008 at 09:43:15AM -0600, Jordan Crouse wrote:
> Changes made - awaiting your ACK.

Acked-by: Uwe Hermann

Uwe.

[Jordan Crouse]

On 25/04/08 22:48 +0200, Uwe Hermann wrote:
> On Tue, Apr 22, 2008 at 09:43:15AM -0600, Jordan Crouse wrote:
> > Changes made - awaiting your ACK.
>
> Acked-by: Uwe Hermann

r3272.  Thanks.