linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH] Avoid overflows in kernel/time.c
@ 2007-11-30  0:19 H. Peter Anvin
  2007-11-30  1:54 ` Andrew Morton
                   ` (3 more replies)
  0 siblings, 4 replies; 27+ messages in thread
From: H. Peter Anvin @ 2007-11-30  0:19 UTC (permalink / raw)
  To: Andrew Morton; +Cc: Linux Kernel Mailing List, H. Peter Anvin

When the conversion factor between jiffies and milli- or microseconds
is not a single multiply or divide, as for the case of HZ == 300, we
currently do a multiply followed by a divide.  The intervening
result, however, is subject to overflows, especially since the
fraction is not simplified (for HZ == 300, we multiply by 300 and
divide by 1000).

This is exposed to the user when passing a large timeout to poll(),
for example.

This patch replaces the multiply-divide with a reciprocal
multiplication on 32-bit platforms.  When the input is an unsigned
long, there is no portable way to do this on 64-bit platforms there is
no portable way to do this since it requires a 128-bit intermediate
result (which gcc does support on 64-bit platforms but may generate
libgcc calls, e.g. on 64-bit s390), but since the output is a 32-bit
integer in the cases affected, just simplify the multiply-divide
(*3/10 instead of *300/1000).

The reciprocal multiply used can have off-by-one errors in the upper
half of the valid output range.  This could be avoided at the expense
of having to deal with a potential 65-bit intermediate result.  Since
the intent is to avoid overflow problems and most of the other time
conversions are only semiexact, the off-by-one errors were considered
an acceptable tradeoff.

NOTE: This patch uses a bc(1) script to compute the appropriate
constants.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
---
 kernel/Makefile     |    8 +++
 kernel/time.c       |   29 +++++++++---
 kernel/timeconst.bc |  123 +++++++++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 152 insertions(+), 8 deletions(-)
 create mode 100644 kernel/timeconst.bc

diff --git a/kernel/Makefile b/kernel/Makefile
index dfa9695..f136d18 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -80,3 +80,11 @@ quiet_cmd_ikconfiggz = IKCFG   $@
 targets += config_data.h
 $(obj)/config_data.h: $(obj)/config_data.gz FORCE
 	$(call if_changed,ikconfiggz)
+
+$(obj)/time.o: $(obj)/timeconst.h
+
+quiet_cmd_timeconst  = BC      $@
+      cmd_timeconst = (echo $(CONFIG_HZ) | bc -q $<) > $@
+targets += timeconst.h
+$(obj)/timeconst.h: $(src)/timeconst.bc $(wildcard include/config/hz.h) FORCE
+	$(call if_changed,timeconst)
diff --git a/kernel/time.c b/kernel/time.c
index 09d3c45..8e790b5 100644
--- a/kernel/time.c
+++ b/kernel/time.c
@@ -39,6 +39,8 @@
 #include <asm/uaccess.h>
 #include <asm/unistd.h>
 
+#include "timeconst.h"
+
 /*
  * The timezone where the local system is located.  Used as a default by some
  * programs who obtain this value by using gettimeofday.
@@ -93,7 +95,8 @@ asmlinkage long sys_stime(time_t __user *tptr)
 
 #endif /* __ARCH_WANT_SYS_TIME */
 
-asmlinkage long sys_gettimeofday(struct timeval __user *tv, struct timezone __user *tz)
+asmlinkage long sys_gettimeofday(struct timeval __user *tv,
+				 struct timezone __user *tz)
 {
 	if (likely(tv != NULL)) {
 		struct timeval ktv;
@@ -118,7 +121,7 @@ asmlinkage long sys_gettimeofday(struct timeval __user *tv, struct timezone __us
  * hard to make the program warp the clock precisely n hours)  or
  * compile in the timezone information into the kernel.  Bad, bad....
  *
- *              				- TYT, 1992-01-01
+ *						- TYT, 1992-01-01
  *
  * The best thing to do is to keep the CMOS clock in universal time (UTC)
  * as real UNIX machines always do it. This avoids all headaches about
@@ -239,7 +242,11 @@ unsigned int inline jiffies_to_msecs(const unsigned long j)
 #elif HZ > MSEC_PER_SEC && !(HZ % MSEC_PER_SEC)
 	return (j + (HZ / MSEC_PER_SEC) - 1)/(HZ / MSEC_PER_SEC);
 #else
-	return (j * MSEC_PER_SEC) / HZ;
+# if BITS_PER_LONG == 32
+	return ((u64)HZ_TO_MSEC_MUL32 * j) >> HZ_TO_MSEC_SHR32;
+# else
+	return (j * HZ_TO_MSEC_NUM) / HZ_TO_MSEC_DEN;
+# endif
 #endif
 }
 EXPORT_SYMBOL(jiffies_to_msecs);
@@ -251,7 +258,11 @@ unsigned int inline jiffies_to_usecs(const unsigned long j)
 #elif HZ > USEC_PER_SEC && !(HZ % USEC_PER_SEC)
 	return (j + (HZ / USEC_PER_SEC) - 1)/(HZ / USEC_PER_SEC);
 #else
-	return (j * USEC_PER_SEC) / HZ;
+# if BITS_PER_LONG == 32
+	return ((u64)HZ_TO_USEC_MUL32 * j) >> HZ_TO_USEC_SHR32;
+# else
+	return (j * HZ_TO_USEC_NUM) / HZ_TO_USEC_DEN;
+# endif
 #endif
 }
 EXPORT_SYMBOL(jiffies_to_usecs);
@@ -351,7 +362,7 @@ EXPORT_SYMBOL(mktime);
  * normalize to the timespec storage format
  *
  * Note: The tv_nsec part is always in the range of
- * 	0 <= tv_nsec < NSEC_PER_SEC
+ *	0 <= tv_nsec < NSEC_PER_SEC
  * For negative values only the tv_sec field is negative !
  */
 void set_normalized_timespec(struct timespec *ts, time_t sec, long nsec)
@@ -452,12 +463,13 @@ unsigned long msecs_to_jiffies(const unsigned int m)
 	/*
 	 * Generic case - multiply, round and divide. But first
 	 * check that if we are doing a net multiplication, that
-	 * we wouldnt overflow:
+	 * we wouldn't overflow:
 	 */
 	if (HZ > MSEC_PER_SEC && m > jiffies_to_msecs(MAX_JIFFY_OFFSET))
 		return MAX_JIFFY_OFFSET;
 
-	return (m * HZ + MSEC_PER_SEC - 1) / MSEC_PER_SEC;
+	return ((u64)MSEC_TO_HZ_MUL32 * m + MSEC_TO_HZ_ADJ32)
+		>> MSEC_TO_HZ_SHR32;
 #endif
 }
 EXPORT_SYMBOL(msecs_to_jiffies);
@@ -471,7 +483,8 @@ unsigned long usecs_to_jiffies(const unsigned int u)
 #elif HZ > USEC_PER_SEC && !(HZ % USEC_PER_SEC)
 	return u * (HZ / USEC_PER_SEC);
 #else
-	return (u * HZ + USEC_PER_SEC - 1) / USEC_PER_SEC;
+	return ((u64)USEC_TO_HZ_MUL32 * m + USEC_TO_HZ_ADJ32)
+		>> USEC_TO_HZ_SHR32;
 #endif
 }
 EXPORT_SYMBOL(usecs_to_jiffies);
diff --git a/kernel/timeconst.bc b/kernel/timeconst.bc
new file mode 100644
index 0000000..79e291f
--- /dev/null
+++ b/kernel/timeconst.bc
@@ -0,0 +1,123 @@
+hz=read()
+scale=0
+
+define gcd(a,b) {
+	auto t;
+	while (b) {
+		t = b;
+		b = a % b;
+		a = t;
+	}
+	return a;
+}
+
+/* Division by reciprocal multiplication. */
+define fmul(b,n,d) {
+       return (2^b*n+d-1)/d;
+}
+
+/* Adjustment factor when a ceiling value is used.  Use as:
+   (imul * n) + (fmulxx * n + fadjxx) >> xx) */
+define fadj(b,n,d) {
+	auto v;
+	d = d/gcd(n,d);
+	v = 2^b*(d-1)/d;
+	return v;
+}
+
+/* Compute the appropriate mul/adj values as well as a shift count,
+   which brings the mul value into the range 2^b-1 <= x < 2^b.  Such
+   a shift value will be correct in the signed integer range and off
+   by at most one in the upper half of the unsigned range. */
+define fmuls(b,n,d) {
+	auto s, m;
+	for (s = 0; 1; s++) {
+		m = fmul(s,n,d);
+		if (m >= 2^(b-1))
+			return s;
+	}
+	return 0;
+}
+
+print "/* Automatically generated from timeconst.bc */\n"
+print "/* Time conversion constants for HZ == ", hz, " */\n"
+print "\n"
+
+print "#ifndef KERNEL_TIMECONST_H\n"
+print "#define KERNEL_TIMECONST_H\n\n"
+
+s=fmuls(32,1000,hz)
+obase=16
+print "#define HZ_TO_MSEC_MUL32 0x", fmul(s,1000,hz), "\n"
+print "#define HZ_TO_MSEC_ADJ32 0x", fadj(s,1000,hz), "\n"
+obase=10
+print "#define HZ_TO_MSEC_SHR32 ", s, "\n"
+
+s=fmuls(64,1000,hz)
+obase=16
+print "#define HZ_TO_MSEC_MUL64 0x", fmul(s,1000,hz), "\n"
+print "#define HZ_TO_MSEC_ADJ64 0x", fadj(s,1000,hz), "\n"
+obase=10
+print "#define HZ_TO_MSEC_SHR64 ", s, "\n"
+
+s=fmuls(32,hz,1000)
+obase=16
+print "#define MSEC_TO_HZ_MUL32 0x", fmul(s,hz,1000), "\n"
+print "#define MSEC_TO_HZ_ADJ32 0x", fadj(s,hz,1000), "\n"
+obase=10
+print "#define MSEC_TO_HZ_SHR32 ", s, "\n"
+
+s=fmuls(64,hz,1000)
+obase=16
+print "#define MSEC_TO_HZ_MUL64 0x", fmul(s,hz,1000), "\n"
+print "#define MSEC_TO_HZ_ADJ64 0x", fadj(s,hz,1000), "\n"
+obase=10
+print "#define MSEC_TO_HZ_SHR64 ", s, "\n"
+
+obase=10
+cd=gcd(hz,1000)
+print "#define HZ_TO_MSEC_NUM ", 1000/cd, "\n"
+print "#define HZ_TO_MSEC_DEN ", hz/cd, "\n"
+print "#define MSEC_TO_HZ_NUM ", hz/cd, "\n"
+print "#define MSEC_TO_HZ_DEN ", 1000/cd, "\n"
+print "\n"
+
+s=fmuls(32,1000000,hz)
+obase=16
+print "#define HZ_TO_USEC_MUL32 0x", fmul(s,1000000,hz), "\n"
+print "#define HZ_TO_USEC_ADJ32 0x", fadj(s,1000000,hz), "\n"
+obase=10
+print "#define HZ_TO_USEC_SHR32 ", s, "\n"
+
+s=fmuls(64,1000000,hz)
+obase=16
+print "#define HZ_TO_USEC_MUL64 0x", fmul(s,1000000,hz), "\n"
+print "#define HZ_TO_USEC_ADJ64 0x", fadj(s,1000000,hz), "\n"
+obase=10
+print "#define HZ_TO_USEC_SHR64 ", s, "\n"
+
+s=fmuls(32,hz,1000000)
+obase=16
+print "#define USEC_TO_HZ_MUL32 0x", fmul(s,hz,1000000), "\n"
+print "#define USEC_TO_HZ_ADJ32 0x", fadj(s,hz,1000000), "\n"
+obase=10
+print "#define USEC_TO_HZ_SHR32 ", s, "\n"
+
+s=fmuls(64,hz,1000000)
+obase=16
+print "#define USEC_TO_HZ_MUL64 0x", fmul(s,hz,1000000), "\n"
+print "#define USEC_TO_HZ_ADJ64 0x", fadj(s,hz,1000000), "\n"
+obase=10
+print "#define USEC_TO_HZ_SHR64 ", s, "\n"
+
+obase=10
+cd=gcd(hz,1000000)
+print "#define HZ_TO_USEC_NUM ", 1000000/cd, "\n"
+print "#define HZ_TO_USEC_DEN ", hz/cd, "\n"
+print "#define USEC_TO_HZ_NUM ", hz/cd, "\n"
+print "#define USEC_TO_HZ_DEN ", 1000000/cd, "\n"
+print "\n"
+
+print "#endif\n"
+
+halt
-- 
1.5.3.4


^ permalink raw reply related	[flat|nested] 27+ messages in thread

end of thread, other threads:[~2007-12-10 22:05 UTC | newest]

Thread overview: 27+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2007-11-30  0:19 [PATCH] Avoid overflows in kernel/time.c H. Peter Anvin
2007-11-30  1:54 ` Andrew Morton
2007-11-30  3:01   ` H. Peter Anvin
2007-11-30  3:27   ` [PATCH] Documentation/Changes -> Documentation/Requirements H. Peter Anvin
2007-11-30  3:32   ` [PATCH] Documentation/Changes -> Documentation/Requirements (resend without truncated comment text) H. Peter Anvin
2007-11-30  7:16     ` Jarek Poplawski
2007-11-30 17:40       ` H. Peter Anvin
2007-11-30 17:47       ` [PATCH] Documentation/Changes -> Documentation/Requirements H. Peter Anvin
2007-11-30 18:09         ` Robert P. J. Day
2007-11-30 18:20           ` H. Peter Anvin
2007-11-30  1:59 ` [PATCH] Avoid overflows in kernel/time.c Chris Snook
2007-11-30  3:04   ` H. Peter Anvin
2007-11-30  3:40     ` Arjan van de Ven
2007-11-30  3:54       ` H. Peter Anvin
2007-12-02 18:37         ` Pavel Machek
2007-12-03 14:53         ` Jan Engelhardt
2007-12-10 16:37           ` H. Peter Anvin
2007-12-01  0:33 ` Adrian Bunk
2007-12-01  4:19   ` H. Peter Anvin
2007-12-01 13:20   ` Alan Cox
2007-12-01 13:33     ` Alan Cox
2007-12-02  1:53     ` H. Peter Anvin
2007-12-07  0:22   ` Jeremy Fitzhardinge
2007-12-04 11:29 ` Andrew Morton
2007-12-10 16:46   ` H. Peter Anvin
2007-12-10 18:59   ` H. Peter Anvin
2007-12-10 22:04     ` Andrew Morton

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).