Skip to contentSkip to conversation

do_posix_clock_monotonic_gettime() returns negative nsec

PATCH Herbert Poetzl 4 participants 8 messages Dec 3, 2004

lore
Share & tools
lore
#0 lore raw

Hi Folks!

recent kernels (tested 2.6.10-rc2 and 2.6.10-rc2-bk15) produce funny output in /proc/uptime like this:

	# cat /proc/uptime
	  12.4294967218 9.05
	# cat /proc/uptime
	  13.4294967251 10.33
	# cat /proc/uptime
	  14.4294967295 11.73

a short investigation of the issue, ended at do_posix_clock_monotonic_gettime() which can (and often does) return negative nsec values (within one second), so while the actual 'time' returned is correct, some parts of the kernel assume that those part is within the range (0 - NSEC_PER_SEC)

        len = sprintf(page,"%lu.%02lu %lu.%02lu\n",
                        (unsigned long) uptime.tv_sec,
                        (uptime.tv_nsec / (NSEC_PER_SEC / 100)),

as the function itself corrects overflows, it would make sense to me to correct underflows too, for example with the following patch:

kernel/posix-timers.c renamed from kernel/posix-timers.c.orig +4 -1
@@ -1208,7 +1208,10 @@ int do_posix_clock_monotonic_gettime(str
 	tp->tv_sec += wall_to_mono.tv_sec;
 	tp->tv_nsec += wall_to_mono.tv_nsec;
 
-	if ((tp->tv_nsec - NSEC_PER_SEC) > 0) {
+	if (tp->tv_nsec < 0) {
+		tp->tv_nsec += NSEC_PER_SEC;
+		tp->tv_sec--;
+	} else if ((tp->tv_nsec - NSEC_PER_SEC) > 0) {
 		tp->tv_nsec -= NSEC_PER_SEC;
 		tp->tv_sec++;
 	}

best,
Herbert

#1 lore raw
Quoted 29 lines · Herbert Poetzl

recent kernels (tested 2.6.10-rc2 and 2.6.10-rc2-bk15) produce funny output in /proc/uptime like this:

	# cat /proc/uptime
	  12.4294967218 9.05
	# cat /proc/uptime
	  13.4294967251 10.33
	# cat /proc/uptime
	  14.4294967295 11.73

a short investigation of the issue, ended at do_posix_clock_monotonic_gettime() which can (and often does) return negative nsec values (within one second), so while the actual 'time' returned is correct, some parts of the kernel assume that those part is within the range (0 - NSEC_PER_SEC)

        len = sprintf(page,"%lu.%02lu %lu.%02lu\n",
                        (unsigned long) uptime.tv_sec,
                        (uptime.tv_nsec / (NSEC_PER_SEC / 100)),

as the function itself corrects overflows, it would make sense to me to correct underflows too, for example with the following patch:

kernel/posix-timers.c renamed from kernel/posix-timers.c.orig +0 -0
@@ -1208,7 +1208,10 @@ int do_posix_clock_monotonic_gettime(str
 	tp->tv_sec += wall_to_mono.tv_sec;

kernel/posix-timers.c int do_posix_clock_monotonic_gettime(str · line 1217

Quoted 10 lines
kernel/posix-timers.c renamed from kernel/posix-timers.c.orig +4 -1
 	tp->tv_nsec += wall_to_mono.tv_nsec;
 
-	if ((tp->tv_nsec - NSEC_PER_SEC) > 0) {
+	if (tp->tv_nsec < 0) {
+		tp->tv_nsec += NSEC_PER_SEC;
+		tp->tv_sec--;
+	} else if ((tp->tv_nsec - NSEC_PER_SEC) > 0) {
 		tp->tv_nsec -= NSEC_PER_SEC;
 		tp->tv_sec++;
 	}

Sounds like its a good fix to me.

George: You have any comment?

thanks
-john

#2 lore raw
Quoted 43 lines · john stultz
Quoted 27 lines · Herbert Poetzl

recent kernels (tested 2.6.10-rc2 and 2.6.10-rc2-bk15) produce funny output in /proc/uptime like this:

	# cat /proc/uptime
	  12.4294967218 9.05
	# cat /proc/uptime
	  13.4294967251 10.33
	# cat /proc/uptime
	  14.4294967295 11.73

a short investigation of the issue, ended at
do_posix_clock_monotonic_gettime() which can (and often does) return negative nsec values (within one second), so while the actual 'time' returned
is correct, some parts of the kernel assume that
those part is within the range (0 - NSEC_PER_SEC)

       len = sprintf(page,"%lu.%02lu %lu.%02lu\n",
                       (unsigned long) uptime.tv_sec,
                       (uptime.tv_nsec / (NSEC_PER_SEC / 100)),

as the function itself corrects overflows, it would
make sense to me to correct underflows too, for example with the following patch:

kernel/posix-timers.c renamed from kernel/posix-timers.c.orig +0 -0
@@ -1208,7 +1208,10 @@ int do_posix_clock_monotonic_gettime(str
 	tp->tv_sec += wall_to_mono.tv_sec;

kernel/posix-timers.c int do_posix_clock_monotonic_gettime(str · line 1217

Quoted 10 lines
kernel/posix-timers.c renamed from kernel/posix-timers.c.orig +4 -1
 	tp->tv_nsec += wall_to_mono.tv_nsec;
 
-	if ((tp->tv_nsec - NSEC_PER_SEC) > 0) {
+	if (tp->tv_nsec < 0) {
+		tp->tv_nsec += NSEC_PER_SEC;
+		tp->tv_sec--;
+	} else if ((tp->tv_nsec - NSEC_PER_SEC) > 0) {
 		tp->tv_nsec -= NSEC_PER_SEC;
 		tp->tv_sec++;
 	}

Sounds like its a good fix to me.

Quoted 1 line

George: You have any comment?

Two, in fact. First, the result here is the sum of wall_to_monotonic and getnstimeofday(). If nsec < 0, one or more of these must be also. Both of these values are SUPPOSED to be normalized.

Second, I would rather see:

	set_normalized_timespec(tp, tp->tv_sec + wall_to_mono.tv_sec, 		
				tp->tv_nsec + wall_to_mono.tv_nsec);

Still, doing this paves over the first issue....

Quoted 5 lines

thanks
-john

Signature · 2 lines
George Anzinger   [email protected]
High-res-timers:  http://sourceforge.net/projects/high-res-timers/
#3 lore raw
Quoted 6 lines · George Anzinger
Quoted 2 lines

George: You have any comment?

Two, in fact. First, the result here is the sum of wall_to_monotonic and getnstimeofday(). If nsec < 0, one or more of these must be also. Both of these values are SUPPOSED to be normalized.

As Herbert points out, hpet_time_init() and time_init() and who knows what else do:

	wall_to_monotonic.tv_sec = -xtime.tv_sec;
	xtime.tv_nsec = (INITIAL_JIFFIES % HZ) * (NSEC_PER_SEC / HZ);
	wall_to_monotonic.tv_nsec = -xtime.tv_nsec;

And this:

* The current time
* wall_to_monotonic is what we need to add to xtime (or xtime corrected
* for sub jiffie times) to get to monotonic time. Monotonic is pegged at zero
* at zero at system boot time, so wall_to_monotonic will be negative,
* however, we will ALWAYS keep the tv_nsec part positive so we can use
* the usual normalization.

So I assume the bug is that time_init() is failing to normalise wall_to_monotonic?

#4 lore raw
In reply to Herbert Poetzl
Quoted 33 lines · Herbert Poetzl

Hi Folks!

recent kernels (tested 2.6.10-rc2 and 2.6.10-rc2-bk15) produce funny output in /proc/uptime like this:

	# cat /proc/uptime
	  12.4294967218 9.05
	# cat /proc/uptime
	  13.4294967251 10.33
	# cat /proc/uptime
	  14.4294967295 11.73

a short investigation of the issue, ended at do_posix_clock_monotonic_gettime() which can (and often does) return negative nsec values (within one second), so while the actual 'time' returned is correct, some parts of the kernel assume that those part is within the range (0 - NSEC_PER_SEC)

        len = sprintf(page,"%lu.%02lu %lu.%02lu\n",
                        (unsigned long) uptime.tv_sec,
                        (uptime.tv_nsec / (NSEC_PER_SEC / 100)),

as the function itself corrects overflows, it would make sense to me to correct underflows too, for example with the following patch:

kernel/posix-timers.c renamed from kernel/posix-timers.c.orig +0 -0
@@ -1208,7 +1208,10 @@ int do_posix_clock_monotonic_gettime(str
 	tp->tv_sec += wall_to_mono.tv_sec;

kernel/posix-timers.c int do_posix_clock_monotonic_gettime(str · line 1217

Quoted 10 lines
kernel/posix-timers.c renamed from kernel/posix-timers.c.orig +4 -1
 	tp->tv_nsec += wall_to_mono.tv_nsec;
 
-	if ((tp->tv_nsec - NSEC_PER_SEC) > 0) {
+	if (tp->tv_nsec < 0) {
+		tp->tv_nsec += NSEC_PER_SEC;
+		tp->tv_sec--;
+	} else if ((tp->tv_nsec - NSEC_PER_SEC) > 0) {
 		tp->tv_nsec -= NSEC_PER_SEC;
 		tp->tv_sec++;
 	}

Doesn't this imply that do_posix_clock_monotonic_gettime_parts() is returning a negative tv_nsec?

If so, that would point back at getnstimeofday(). What is your setting of CONFIG_TIME_INTERPOLATION?

#5 lore raw
Quoted 36 lines · Andrew Morton
Quoted 35 lines · Herbert Poetzl

Hi Folks!

recent kernels (tested 2.6.10-rc2 and 2.6.10-rc2-bk15) produce funny output in /proc/uptime like this:

	# cat /proc/uptime
	  12.4294967218 9.05
	# cat /proc/uptime
	  13.4294967251 10.33
	# cat /proc/uptime
	  14.4294967295 11.73

a short investigation of the issue, ended at do_posix_clock_monotonic_gettime() which can (and often does) return negative nsec values (within one second), so while the actual 'time' returned is correct, some parts of the kernel assume that those part is within the range (0 - NSEC_PER_SEC)

        len = sprintf(page,"%lu.%02lu %lu.%02lu\n",
                        (unsigned long) uptime.tv_sec,
                        (uptime.tv_nsec / (NSEC_PER_SEC / 100)),

as the function itself corrects overflows, it would make sense to me to correct underflows too, for example with the following patch:

kernel/posix-timers.c renamed from kernel/posix-timers.c.orig +0 -0
@@ -1208,7 +1208,10 @@ int do_posix_clock_monotonic_gettime(str
 	tp->tv_sec += wall_to_mono.tv_sec;
 	tp->tv_nsec += wall_to_mono.tv_nsec;
 
Quoted 11 lines

kernel/posix-timers.c int do_posix_clock_monotonic_gettime(str · line 1217

Quoted 8 lines
kernel/posix-timers.c renamed from kernel/posix-timers.c.orig +4 -1
-	if ((tp->tv_nsec - NSEC_PER_SEC) > 0) {
+	if (tp->tv_nsec < 0) {
+		tp->tv_nsec += NSEC_PER_SEC;
+		tp->tv_sec--;
+	} else if ((tp->tv_nsec - NSEC_PER_SEC) > 0) {
 		tp->tv_nsec -= NSEC_PER_SEC;
 		tp->tv_sec++;
 	}

Doesn't this imply that do_posix_clock_monotonic_gettime_parts() is returning a negative tv_nsec?

nope, not necessarily, because after that ...

        tp->tv_sec += wall_to_mono.tv_sec;
        tp->tv_nsec += wall_to_mono.tv_nsec;

might add a negative value, which explains the
underflow ...

and if you look closer:

	xtime.tv_sec = get_cmos_time();
        wall_to_monotonic.tv_sec = -xtime.tv_sec;
        xtime.tv_nsec = (INITIAL_JIFFIES % HZ) * (NSEC_PER_SEC / HZ);
        wall_to_monotonic.tv_nsec = -xtime.tv_nsec;

#define INITIAL_JIFFIES ((unsigned long)(unsigned int) (-300*HZ))

which might need a fix too ... but that's a
different story ...

Quoted 2 lines

If so, that would point back at getnstimeofday(). What is your setting of CONFIG_TIME_INTERPOLATION?

# grep TIME .config
# CONFIG_HPET_TIMER is not set
# CONFIG_HANGCHECK_TIMER is not set

best,
Herbert

#6 lore raw
Quoted 57 lines · Herbert Poetzl
Quoted 34 lines · Andrew Morton
Quoted 32 lines · Herbert Poetzl

Hi Folks!

recent kernels (tested 2.6.10-rc2 and 2.6.10-rc2-bk15) produce funny output in /proc/uptime like this:

	# cat /proc/uptime
	  12.4294967218 9.05
	# cat /proc/uptime
	  13.4294967251 10.33
	# cat /proc/uptime
	  14.4294967295 11.73

a short investigation of the issue, ended at
do_posix_clock_monotonic_gettime() which can (and often does) return negative nsec values (within one second), so while the actual 'time' returned
is correct, some parts of the kernel assume that
those part is within the range (0 - NSEC_PER_SEC)

       len = sprintf(page,"%lu.%02lu %lu.%02lu\n",
                       (unsigned long) uptime.tv_sec,
                       (uptime.tv_nsec / (NSEC_PER_SEC / 100)),

as the function itself corrects overflows, it would
make sense to me to correct underflows too, for example with the following patch:

kernel/posix-timers.c renamed from kernel/posix-timers.c.orig +0 -0
@@ -1208,7 +1208,10 @@ int do_posix_clock_monotonic_gettime(str
 	tp->tv_sec += wall_to_mono.tv_sec;
 	tp->tv_nsec += wall_to_mono.tv_nsec;
 
Quoted 11 lines

kernel/posix-timers.c int do_posix_clock_monotonic_gettime(str · line 1217

Quoted 8 lines
kernel/posix-timers.c renamed from kernel/posix-timers.c.orig +4 -1
-	if ((tp->tv_nsec - NSEC_PER_SEC) > 0) {
+	if (tp->tv_nsec < 0) {
+		tp->tv_nsec += NSEC_PER_SEC;
+		tp->tv_sec--;
+	} else if ((tp->tv_nsec - NSEC_PER_SEC) > 0) {
 		tp->tv_nsec -= NSEC_PER_SEC;
 		tp->tv_sec++;
 	}

Doesn't this imply that do_posix_clock_monotonic_gettime_parts() is returning a negative tv_nsec?

nope, not necessarily, because after that ...

        tp->tv_sec += wall_to_mono.tv_sec;
        tp->tv_nsec += wall_to_mono.tv_nsec;

might add a negative value, which explains the
underflow ...

Quoted 6 lines

and if you look closer:

	xtime.tv_sec = get_cmos_time();
        wall_to_monotonic.tv_sec = -xtime.tv_sec;
        xtime.tv_nsec = (INITIAL_JIFFIES % HZ) * (NSEC_PER_SEC / HZ);
        wall_to_monotonic.tv_nsec = -xtime.tv_nsec;

Yep, that IS the problem. It should be normalized here, I.e.

	set_normalized_timespec(wall_to_monotonic,
		wall_to_monotonic.tv_sec - xtime.tv_sec,
		wall_to_monotonic.tv_nsec - xtime.tv_nsec);

with the obvious delets :)

Still, this should be corrected by the first settimeofday, which most systems do on the way up, or is that just those who use NTP?

Signature · 2 lines
George Anzinger   [email protected]
High-res-timers:  http://sourceforge.net/projects/high-res-timers/
#7 lore raw
Quoted 66 lines · George Anzinger
Quoted 58 lines · Herbert Poetzl
Quoted 36 lines · Andrew Morton
Quoted 34 lines · Herbert Poetzl

Hi Folks!

recent kernels (tested 2.6.10-rc2 and 2.6.10-rc2-bk15) produce funny output in /proc/uptime like this:

	# cat /proc/uptime
	  12.4294967218 9.05
	# cat /proc/uptime
	  13.4294967251 10.33
	# cat /proc/uptime
	  14.4294967295 11.73

a short investigation of the issue, ended at do_posix_clock_monotonic_gettime() which can (and often does) return negative nsec values (within one second), so while the actual 'time' returned is correct, some parts of the kernel assume that those part is within the range (0 - NSEC_PER_SEC)

      len = sprintf(page,"%lu.%02lu %lu.%02lu\n",
                      (unsigned long) uptime.tv_sec,
                      (uptime.tv_nsec / (NSEC_PER_SEC / 100)),

as the function itself corrects overflows, it would make sense to me to correct underflows too, for example with the following patch:

kernel/posix-timers.c renamed from kernel/posix-timers.c.orig +0 -0
@@ -1208,7 +1208,10 @@ int do_posix_clock_monotonic_gettime(str
 	tp->tv_sec += wall_to_mono.tv_sec;
 	tp->tv_nsec += wall_to_mono.tv_nsec;
 
Quoted 11 lines

kernel/posix-timers.c int do_posix_clock_monotonic_gettime(str · line 1217

Quoted 8 lines
kernel/posix-timers.c renamed from kernel/posix-timers.c.orig +4 -1
-	if ((tp->tv_nsec - NSEC_PER_SEC) > 0) {
+	if (tp->tv_nsec < 0) {
+		tp->tv_nsec += NSEC_PER_SEC;
+		tp->tv_sec--;
+	} else if ((tp->tv_nsec - NSEC_PER_SEC) > 0) {
 		tp->tv_nsec -= NSEC_PER_SEC;
 		tp->tv_sec++;
 	}

Doesn't this imply that do_posix_clock_monotonic_gettime_parts() is returning a negative tv_nsec?

nope, not necessarily, because after that ...

       tp->tv_sec += wall_to_mono.tv_sec;
       tp->tv_nsec += wall_to_mono.tv_nsec;

might add a negative value, which explains the
underflow ...

Quoted 6 lines

and if you look closer:

	xtime.tv_sec = get_cmos_time();
       wall_to_monotonic.tv_sec = -xtime.tv_sec;
       xtime.tv_nsec = (INITIAL_JIFFIES % HZ) * (NSEC_PER_SEC / HZ);
       wall_to_monotonic.tv_nsec = -xtime.tv_nsec;
Quoted 5 lines

Yep, that IS the problem. It should be normalized here, I.e.

	set_normalized_timespec(wall_to_monotonic,
		wall_to_monotonic.tv_sec - xtime.tv_sec,
		wall_to_monotonic.tv_nsec - xtime.tv_nsec);

with the obvious delets :)

agreed, especially as almost all archs do it
properly (jsut i386 and m32r miss that)

Quoted 2 lines

Still, this should be corrected by the first settimeofday, which most systems do on the way up, or is that just those who use NTP?

correct, but not all use settimeofday() ...

here is a better patch:

Fix the non-normalized wall_to_monotonic for i386 and m32r (The other archs seem to get it right)

arch/i386/kernel/time.c +4 -4
diff -NurpP --minimal linux-2.6.10-rc2-bk15/arch/i386/kernel/time.c linux-2.6.10-rc2-bk15.2/arch/i386/kernel/time.c
@@ -381,9 +381,9 @@ extern void (*late_time_init)(void);
 void __init hpet_time_init(void)
 {
 	xtime.tv_sec = get_cmos_time();
-	wall_to_monotonic.tv_sec = -xtime.tv_sec;
 	xtime.tv_nsec = (INITIAL_JIFFIES % HZ) * (NSEC_PER_SEC / HZ);
-	wall_to_monotonic.tv_nsec = -xtime.tv_nsec;
+	set_normalized_timespec(&wall_to_monotonic,
+		-xtime.tv_sec, -xtime.tv_nsec);
 
 	if (hpet_enable() >= 0) {
 		printk("Using HPET for base-timer\n");
@@ -409,9 +409,9 @@ void __init time_init(void)
 	}
 #endif
 	xtime.tv_sec = get_cmos_time();
-	wall_to_monotonic.tv_sec = -xtime.tv_sec;
 	xtime.tv_nsec = (INITIAL_JIFFIES % HZ) * (NSEC_PER_SEC / HZ);
-	wall_to_monotonic.tv_nsec = -xtime.tv_nsec;
+	set_normalized_timespec(&wall_to_monotonic,
+		-xtime.tv_sec, -xtime.tv_nsec);
 
 	cur_timer = select_timer();
 	printk(KERN_INFO "Using %s for high-res timesource\n",cur_timer->name);
arch/m32r/kernel/time.c +2 -2
diff -NurpP --minimal linux-2.6.10-rc2-bk15/arch/m32r/kernel/time.c linux-2.6.10-rc2-bk15.2/arch/m32r/kernel/time.c
@@ -275,8 +275,8 @@ void __init time_init(void)
 
 	xtime.tv_sec = mktime(year, mon, day, hour, min, sec);
 	xtime.tv_nsec = (INITIAL_JIFFIES % HZ) * (NSEC_PER_SEC / HZ);
-	wall_to_monotonic.tv_sec = -xtime.tv_sec;
-	wall_to_monotonic.tv_nsec = -xtime.tv_nsec;
+	set_normalized_timespec(&wall_to_monotonic,
+		-xtime.tv_sec, -xtime.tv_nsec);
 
 #if defined(CONFIG_CHIP_M32102) || defined(CONFIG_CHIP_XNUX2) \
 	|| defined(CONFIG_CHIP_VDEC2) || defined(CONFIG_CHIP_M32700) \

best,
Herbert

Quoted 9 lines
Signature · 8 lines
George Anzinger   [email protected]
High-res-timers:  http://sourceforge.net/projects/high-res-timers/

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/