Skip to contentSkip to conversation

compat: Avoid double rounding in times()

PATCH Applied Hungyu Lin 3 participants 3 messages Oct 20 Stable branch

lore
Share & tools
lore

Hungyu Lin

author
#0 lore raw

PATCH 1/1 compat: Avoid double rounding in times()

do_sys_times() already converts CPU time to USER_HZ ticks. The compat path converts each value back to jiffies before converting it to COMPAT_USER_HZ. This round trip can lose precision even when the two userspace rates match.

With HZ=250 and USER_HZ=COMPAT_USER_HZ=100, one user tick becomes two jiffies and then zero compat ticks. Every odd tick count loses one tick, so any of the four tms fields can underreport CPU time by an additional 10 ms.

Return the value directly when the native and compat units match. Keep the existing conversion for differing rates.

Fixes f06febc96ba8 ("timers: fix itimer/many thread hang")
---
 kernel/sys.c | 3 +++
 1 file changed, 3 insertions(+)
kernel/sys.c +3 -0
index 35b538ba843c..755bfa5ccd92 100644
@@ -1077,6 +1077,9 @@ SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
 #ifdef CONFIG_COMPAT
 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
 {
+	if (USER_HZ == COMPAT_USER_HZ)
+		return x;
+
 	return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
 }
 

base-commit: f47d223627f80cb7a1eb14c585fe6cb35de5e984

Signature · 1 line
2.43.0
#1 lore raw
Quoted 5 lines · On Tue, 20 Oct 2026 21:01:57 +0000 Hungyu Lin <[email protected]> wrote:

do_sys_times() already converts CPU time to USER_HZ ticks. The compat path converts each value back to jiffies before converting it to COMPAT_USER_HZ. This round trip can lose precision even when the two userspace rates match.

Quoted 7 lines

With HZ=250 and USER_HZ=COMPAT_USER_HZ=100, one user tick becomes two jiffies and then zero compat ticks. Every odd tick count loses one tick, so any of the four tms fields can underreport CPU time by an additional 10 ms.

Return the value directly when the native and compat units match. Keep the existing conversion for differing rates.

The fix looks correct. The compat times() path unnecessarily converts CPU time from USER_HZ ticks to jiffies and back, losing precision.

I had $LLM model the userspace-visible results for tick counts 0 through 10,000. With HZ=250 and USER_HZ=COMPAT_USER_HZ=100, 5,000 values underreport CPU time by 10 ms. The patch eliminates these errors without changing the differing-rate case.

Thanks.

#2 lore raw
In reply to Hungyu Lin
Quoted 5 lines · Hungyu Lin

do_sys_times() already converts CPU time to USER_HZ ticks. The compat path converts each value back to jiffies before converting it to COMPAT_USER_HZ. This round trip can lose precision even when the two userspace rates match.

Quoted 7 lines

With HZ=250 and USER_HZ=COMPAT_USER_HZ=100, one user tick becomes two jiffies and then zero compat ticks. Every odd tick count loses one tick, so any of the four tms fields can underreport CPU time by an additional 10 ms.

Return the value directly when the native and compat units match. Keep the existing conversion for differing rates.

Good catch!

Quoted 1 line
Fixes f06febc96ba8 ("timers: fix itimer/many thread hang")

That one apparently just moved the code into a separate function, but the bug was already there before 2.6.12:

+               tmp.tms_utime = compat_jiffies_to_clock_t(cputime_to_jiffies(utime));
+               tmp.tms_stime = compat_jiffies_to_clock_t(cputime_to_jiffies(stime));
+               tmp.tms_cutime = compat_jiffies_to_clock_t(cputime_to_jiffies(cutime));
+               tmp.tms_cstime = compat_jiffies_to_clock_t(cputime_to_jiffies(cstime));

kernel/sys.c SYSCALL_DEFINE1(times, struct tms __user *, tbuf) · line 1083

Quoted 7 lines

#ifdef CONFIG_COMPAT
static compat_clock_t clock_t_to_compat_clock_t(clock_t x) {

kernel/sys.c +3 -0
+	if (USER_HZ == COMPAT_USER_HZ)
+		return x;
+
 	return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));

I don't think there are any architectures left that define USER_HZ and COMPAT_USER_HZ differently. This was true on ia64 with x86 compat, but that was dropped in v2.6.33 back in 2010.

Arnd