Skip to contentSkip to conversation

RTC subsystem and fractions of seconds

DISCUSSION David Brownell 3 participants 10 messages Jan 6, 2007 – Jan 8, 2007

lore
Share & tools
lore
#0 lore raw
Quoted 2 lines

Those rtc's actually have a 1/100th of second
register. Should the generic rtc interface not support that?

Are you implying a new userspace API, or just an in-kernel update?

Either way, that raises the question of what other features should be included. What sub-second precision? Multiple alarms? Ways to manage output clocks? Sub-HZ periodic alarms?

#1 lore raw
Quoted 4 lines · David Brownell
Quoted 2 lines

Those rtc's actually have a 1/100th of second
register. Should the generic rtc interface not support that?

Are you implying a new userspace API, or just an in-kernel update?

My only concern at the moment is initializing linux's timeofday from the rtc quickly and with a good precision. The way it is done currently in drivers/rtc/hctosys.c is 0.5 sec off. We could obtain a much better precision by looping there until the next change (next second for old clocks, next 0.01 second for m41t81, maybe even better for other ones).

Quoted 4 lines

Either way, that raises the question of what other features should be included. What sub-second precision? Multiple alarms? Ways to manage output clocks? Sub-HZ periodic alarms?

I cannot answer that, but others may have other needs.

Philippe

#2 lore raw
Quoted 8 lines · Philippe De Muyter
Quoted 4 lines · David Brownell
Quoted 2 lines

Those rtc's actually have a 1/100th of second
register. Should the generic rtc interface not support that?

Are you implying a new userspace API, or just an in-kernel update?

My only concern at the moment is initializing linux's timeofday from the rtc quickly and with a good precision.

There will necessarily be a bit of fuzz there since it can take time to get that RTC's mutex, and the task setting that time can be preempted. Plus, there can also be delays at the I2C or SPI transaction level.

Quoted 4 lines

The way it is done currently
in drivers/rtc/hctosys.c is 0.5 sec off. We could obtain a much better precision by looping there until the next change (next second for old clocks, next 0.01 second for m41t81, maybe even better for other ones).

Hmm ... "looping" fights against "quickly"; as would "wait for next update IRQ" (on RTCs that support that). But it would improve precision, at least in the sense of having the system clock and that RTC spending less time with the lowest "seconds" digit disagreeing.

This is something you could write a patch for, n'est-ce pas?

- Dave

#3 lore raw
Quoted 7 lines · David Brownell
Quoted 4 lines · Philippe De Muyter

The way it is done currently
in drivers/rtc/hctosys.c is 0.5 sec off. We could obtain a much better precision by looping there until the next change (next second for old clocks, next 0.01 second for m41t81, maybe even better for other ones).

Quoted 6 lines

Hmm ... "looping" fights against "quickly"; as would "wait for next update IRQ" (on RTCs that support that). But it would improve precision, at least in the sense of having the system clock and that RTC spending less time with the lowest "seconds" digit disagreeing.

This is something you could write a patch for, n'est-ce pas?

If you're thinking of going that way, it'd be courteous to CC Matt, who devoted some effort to removing just that loop in 2.6.17 ;)

Hugh

#4 lore raw
Quoted 8 lines
Quoted 6 lines

Hmm ... "looping" fights against "quickly"; as would "wait for next update IRQ" (on RTCs that support that). But it would improve precision, at least in the sense of having the system clock and that RTC spending less time with the lowest "seconds" digit disagreeing.

This is something you could write a patch for, n'est-ce pas?

If you're thinking of going that way,

Philippe seemed to be ...

Quoted 2 lines

it'd be courteous to CC Matt,
who devoted some effort to removing just that loop in 2.6.17 ;)

Hmm ... "git whatchanged drivers/rtc/hctosys.c" shows no such change. So I can't find any record of such a change or its rationale.

- dave

#5 lore raw
Quoted 9 lines · David Brownell
Quoted 8 lines
Quoted 6 lines

Hmm ... "looping" fights against "quickly"; as would "wait for next update IRQ" (on RTCs that support that). But it would improve precision, at least in the sense of having the system clock and that RTC spending less time with the lowest "seconds" digit disagreeing.

This is something you could write a patch for, n'est-ce pas?

If you're thinking of going that way,

Quoted 7 lines

Philippe seemed to be ...

Quoted 2 lines

it'd be courteous to CC Matt,
who devoted some effort to removing just that loop in 2.6.17 ;)

Hmm ... "git whatchanged drivers/rtc/hctosys.c" shows no such change. So I can't find any record of such a change or its rationale.

Perhaps I'm muddled - I was thinking of this and its associates:

commit 63732c2f37093d63102d53e70866cf87bf0c0479

Author: Matt Mackall <[email protected]>
Date:   Tue Mar 28 01:55:58 2006 -0800

[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.

This is the currently the most substantial boot time delay for machines that are working towards instant-on capability. Also, a quick back of the envelope calculation (.5sec * 2M users * 1 boot a day * 10 years) suggests it has cost Linux users in the neighborhood of a million man-hours.

An earlier thread on this topic is here:

http://groups.google.com/group/linux.kernel/browse_frm/thread/8a24255215ff6151/2aa97e66a977653d?hl=en&lr=&ie=UTF-8&rnum=1&prev=/groups%3Fhl%3Den%26lr%3D%26ie%3DUTF-8%26selm%3D1To2R-2S7-11%40gated-at.bofh.it#2aa97e66a977653d

..from which the consensus seems to be that it's no longer desirable.

In my view, there are basically four cases to consider:

1) networked, need precise walltime: use NTP
2) networked, don't need precise walltime: use NTP anyway
3) not networked, don't need sub-second precision walltime: don't care
4) not networked, need sub-second precision walltime:
get a network or a radio time source because RTC isn't good enough anyway

So this patch series simply removes the synchronization in favor of a simple seqlock-like approach using the seconds value.

Note that for purposes of timer accuracy on wakeup, this patch will cause us to fire timers up to one second late. But as the current timer resume code will already sync once (or more!), it's no worse for short timers.

Cc Andi Kleen <[email protected]>
Cc "David S. Miller" <[email protected]>
Cc Benjamin Herrenschmidt <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Russell King <[email protected]>
Cc Ralf Baechle <[email protected]>
Cc Paul Mundt <[email protected]>
Cc Kazumoto Kojima <[email protected]>
Cc Alessandro Zummo <[email protected]>
include/asm-i386/mach-default/mach_time.h +2 -14
index b749aa4..ff03cf0 100644
@@ -82,21 +82,8 @@ static inline int mach_set_rtc_mmss(unsi
 static inline unsigned long mach_get_cmos_time(void)
 {
 	unsigned int year, mon, day, hour, min, sec;
-	int i;
 
-	/* The Linux interpretation of the CMOS clock register contents:
-	 * When the Update-In-Progress (UIP) flag goes from 1 to 0, the
-	 * RTC registers show the second which has precisely just started.
-	 * Let's hope other operating systems interpret the RTC the same way.
-	 */
-	/* read RTC exactly on falling edge of update flag */
-	for (i = 0 ; i < 1000000 ; i++)	/* may take up to 1 second... */
-		if (CMOS_READ(RTC_FREQ_SELECT) & RTC_UIP)
-			break;
-	for (i = 0 ; i < 1000000 ; i++)	/* must try at least 2.228 ms */
-		if (!(CMOS_READ(RTC_FREQ_SELECT) & RTC_UIP))
-			break;
-	do { /* Isn't this overkill ? UIP above should guarantee consistency */
+	do {
 		sec = CMOS_READ(RTC_SECONDS);
 		min = CMOS_READ(RTC_MINUTES);
 		hour = CMOS_READ(RTC_HOURS);
@@ -104,6 +91,7 @@ static inline unsigned long mach_get_cmo
 		mon = CMOS_READ(RTC_MONTH);
 		year = CMOS_READ(RTC_YEAR);
 	} while (sec != CMOS_READ(RTC_SECONDS));
+
 	if (!(CMOS_READ(RTC_CONTROL) & RTC_DM_BINARY) || RTC_ALWAYS_BCD)
 	  {
 	    BCD_TO_BIN(sec);
#6 lore raw
In reply to David Brownell
Quoted 20 lines · David Brownell
Quoted 8 lines · Philippe De Muyter
Quoted 4 lines · David Brownell
Quoted 2 lines

Those rtc's actually have a 1/100th of second
register. Should the generic rtc interface not support that?

Are you implying a new userspace API, or just an in-kernel update?

My only concern at the moment is initializing linux's timeofday from the rtc quickly and with a good precision.

There will necessarily be a bit of fuzz there since it can take time to get that RTC's mutex, and the task setting that time can be preempted. Plus, there can also be delays at the I2C or SPI transaction level.

Quoted 4 lines

The way it is done currently
in drivers/rtc/hctosys.c is 0.5 sec off. We could obtain a much better precision by looping there until the next change (next second for old clocks, next 0.01 second for m41t81, maybe even better for other ones).

Quoted 6 lines

Hmm ... "looping" fights against "quickly"; as would "wait for next update IRQ" (on RTCs that support that). But it would improve precision, at least in the sense of having the system clock and that RTC spending less time with the lowest "seconds" digit disagreeing.

This is something you could write a patch for, n'est-ce pas?

That would require changing the interface provided by the rtc class, but I'll look at it.

Quoted 4 lines

- Dave

#7 lore raw
In reply to David Brownell
Quoted 8 lines · David Brownell
Quoted 2 lines

Those rtc's actually have a 1/100th of second
register. Should the generic rtc interface not support that?

Are you implying a new userspace API, or just an in-kernel update?

Either way, that raises the question of what other features should be included. What sub-second precision? Multiple alarms? Ways to manage output clocks? Sub-HZ periodic alarms?

One usefull addition for my needs and with a m41t81 is the support of the calibration of the rtc. However this can perhaps be hidden in the .set_mmss function.

Philippe

#8 lore raw
Quoted 6 lines · Philippe De Muyter
Quoted 5 lines · David Brownell
Quoted 2 lines

Those rtc's actually have a 1/100th of second
register. Should the generic rtc interface not support that?

Are you implying a new userspace API, or just an in-kernel update?

Quoted 7 lines
Quoted 3 lines

Either way, that raises the question of what other features should be included. What sub-second precision? Multiple alarms? Ways to manage output clocks? Sub-HZ periodic alarms?

One usefull addition for my needs and with a m41t81 is the support of the calibration of the rtc. However this can perhaps be hidden in the .set_mmss function.

Doesn't seem like an set_mmss() mechanism at all. Some drivers give sysfs access to an oscillator "trim" mechanism. What tools do you have which track how far off that crystal is?

- Dave

#9 lore raw
Quoted 7 lines · David Brownell
Quoted 3 lines

One usefull addition for my needs and with a m41t81 is the support of the calibration of the rtc. However this can perhaps be hidden in the .set_mmss function.

Doesn't seem like an set_mmss() mechanism at all. Some drivers give sysfs access to an oscillator "trim" mechanism. What tools do you have which track how far off that crystal is?

ntp, either from network or from a radio-clock

and using set_mmss for that purpose seems not so disgusting, because :

- writing to the clock (the expected work of set_mmss) must be disabled when we observe the clock drift
- set_mmss is always called with a very good precision, so one could use the time given to set_mmss to compare it to the time given by the RTC over a long enough period of time (more than 11 minutes) to compute the RTC's drift and the necessary adjustment.

Of course, if we want something that is not based on ntp, then we may not overload set_mmss.

Philippe