- div_long_long_rem(): add a generic version in asm-generic/div64.h; - add div_long_long_rem() implemented in terms of do_div() for those architectures that define their own assembly optimized version; - asm-arm/div64.h:do_div(): add missing parenthesis for macro args - kill dupes of div_long_long_rem() defined on-the-fly in linux/time.h and kernel/posix-timers.c; - make div_long_long_rem() arguments consistent with do_div() and its current usage (it's u64/u32 -> u32/u32). - add casts where a signed integer was being passed for the remainder; Applies to 2.6.0-test1. Tested on i386 and m68knommu. Sorry if it breaks any other architecture (it shouldn't, but...). diff -Nrup linux-2.6.0-test1.orig/include/asm-generic/div64.h linux-2.6.0-test1/include/asm-generic/div64.h --- linux-2.6.0-test1.orig/include/asm-generic/div64.h 2003-07-14 05:37:32.000000000 +0200 +++ linux-2.6.0-test1/include/asm-generic/div64.h 2003-07-15 04:56:08.000000000 +0200 @@ -4,6 +4,7 @@ * Copyright (C) 2003 Bernardo Innocenti * Based on former asm-ppc/div64.h and asm-m68knommu/div64.h * + * * The semantics of do_div() are: * * uint32_t do_div(uint64_t *n, uint32_t base) @@ -15,6 +16,17 @@ * * NOTE: macro parameter n is evaluated multiple times, * beware of side effects! + * + * + * Semantics for div_long_long_rem(): + * + * uint32_t div_long_long_rem(uint64_t n, uint32_t base, uint32_t *rem) + * { + * *rem = n % base; + * return n / base; + * } + * + * NOTE: this will do an exception if result overflows. */ #include @@ -30,6 +42,13 @@ __rem; \ }) +extern inline uint32_t +div_long_long_rem(uint64_t n, uint32_t base, uint32_t *rem) +{ + *rem = n % base; + return n / base; +} + #elif BITS_PER_LONG == 32 extern uint32_t __div64_32(uint64_t *dividend, uint32_t divisor); @@ -49,6 +68,18 @@ extern uint32_t __div64_32(uint64_t *div __rem; \ }) +extern inline uint32_t +div_long_long_rem(uint64_t n, uint32_t base, uint32_t *rem) +{ + if (likely((n >> 32) == 0)) { + *rem = ((uint32_t)n) % base; + return ((uint32_t)n) / base; + } else { + *rem = __div64_32(&n, base); + return n; + } +} + #else /* BITS_PER_LONG == ?? */ # error do_div() does not yet support the C64 @@ -56,3 +87,4 @@ extern uint32_t __div64_32(uint64_t *div #endif /* BITS_PER_LONG */ #endif /* _ASM_GENERIC_DIV64_H */ + diff -Nrup linux-2.6.0-test1.orig/include/asm-arm/div64.h linux-2.6.0-test1/include/asm-arm/div64.h --- linux-2.6.0-test1.orig/include/asm-arm/div64.h 2003-07-14 05:33:50.000000000 +0200 +++ linux-2.6.0-test1/include/asm-arm/div64.h 2003-07-15 05:23:42.000000000 +0200 @@ -1,18 +1,32 @@ #ifndef __ASM_ARM_DIV64 #define __ASM_ARM_DIV64 +#include + /* We're not 64-bit, but... */ #define do_div(n,base) \ ({ \ - register int __res asm("r2") = base; \ - register unsigned long long __n asm("r0") = n; \ + register int __res asm("r2") = (base); \ + register unsigned long long __n asm("r0") = (n); \ asm("bl do_div64" \ : "=r" (__n), "=r" (__res) \ : "0" (__n), "1" (__res) \ : "r3", "ip", "lr", "cc"); \ - n = __n; \ + (n) = __n; \ __res; \ }) +/* + * See for specifications. + * + * FIXME: performance could be improved by writing specific asm + */ +extern inline uint32_t +div_long_long_rem(uint64_t n, uint32_t base, uint32_t *rem) +{ + *rem = do_div(n, base); + return n; +} + #endif diff -Nrup linux-2.6.0-test1.orig/include/asm-i386/div64.h linux-2.6.0-test1/include/asm-i386/div64.h --- linux-2.6.0-test1.orig/include/asm-i386/div64.h 2003-07-14 05:28:55.000000000 +0200 +++ linux-2.6.0-test1/include/asm-i386/div64.h 2003-07-15 06:02:42.000000000 +0200 @@ -1,6 +1,12 @@ #ifndef __I386_DIV64 #define __I386_DIV64 +#include + +/* See asm-generic/div64.h for semantics of + * do_div() and div_long_long_rem() + */ + #define do_div(n,base) ({ \ unsigned long __upper, __low, __high, __mod; \ asm("":"=a" (__low), "=d" (__high):"A" (n)); \ @@ -14,22 +20,14 @@ __mod; \ }) -/* - * (long)X = ((long long)divs) / (long)div - * (long)rem = ((long long)divs) % (long)div - * - * Warning, this will do an exception if X overflows. - */ -#define div_long_long_rem(a,b,c) div_ll_X_l_rem(a,b,c) - -extern inline long -div_ll_X_l_rem(long long divs, long div, long *rem) +extern inline uint32_t +div_long_long_rem(uint64_t divs, uint32_t div, uint32_t *rem) { - long dum2; - __asm__("divl %2":"=a"(dum2), "=d"(*rem) - : "rm"(div), "A"(divs)); - - return dum2; + uint32_t result; + __asm__("divl %2":"=a"(result), "=d"(*rem) + : "rm"(div), "A"(divs)); + return result; } + #endif diff -Nrup linux-2.6.0-test1.orig/include/asm-m68k/div64.h linux-2.6.0-test1/include/asm-m68k/div64.h --- linux-2.6.0-test1.orig/include/asm-m68k/div64.h 2003-07-14 05:37:59.000000000 +0200 +++ linux-2.6.0-test1/include/asm-m68k/div64.h 2003-07-15 05:23:53.000000000 +0200 @@ -1,6 +1,8 @@ #ifndef _M68K_DIV64_H #define _M68K_DIV64_H +#include + /* n = n / base; return rem; */ #define do_div(n, base) ({ \ @@ -23,4 +25,16 @@ __rem; \ }) +/* + * See for specifications. + * + * FIXME: performance could be improved by writing specific asm + */ +extern inline uint32_t +div_long_long_rem(uint64_t n, uint32_t base, uint32_t *rem) +{ + *rem = do_div(n, base); + return n; +} + #endif /* _M68K_DIV64_H */ diff -Nrup linux-2.6.0-test1.orig/include/asm-mips/div64.h linux-2.6.0-test1/include/asm-mips/div64.h --- linux-2.6.0-test1.orig/include/asm-mips/div64.h 2003-07-14 05:36:33.000000000 +0200 +++ linux-2.6.0-test1/include/asm-mips/div64.h 2003-07-15 05:24:16.000000000 +0200 @@ -8,6 +8,8 @@ #ifndef _ASM_DIV64_H #define _ASM_DIV64_H +#include + /* * No traps on overflows for any of these... */ @@ -73,4 +75,16 @@ (n) = __quot; \ __mod; }) +/* + * See for specifications. + * + * FIXME: performance could be improved by writing specific asm + */ +extern inline uint32_t +div_long_long_rem(uint64_t n, uint32_t base, uint32_t *rem) +{ + *rem = do_div(n, base); + return n; +} + #endif /* _ASM_DIV64_H */ diff -Nrup linux-2.6.0-test1.orig/include/asm-s390/div64.h linux-2.6.0-test1/include/asm-s390/div64.h --- linux-2.6.0-test1.orig/include/asm-s390/div64.h 2003-07-14 05:35:52.000000000 +0200 +++ linux-2.6.0-test1/include/asm-s390/div64.h 2003-07-15 05:24:28.000000000 +0200 @@ -3,6 +3,8 @@ #ifndef __s390x__ +#include + /* for do_div "base" needs to be smaller than 2^31-1 */ #define do_div(n, base) ({ \ unsigned long long __n = (n); \ @@ -42,6 +44,18 @@ __r; \ }) +/* + * See for specifications. + * + * FIXME: performance could be improved by writing specific asm + */ +extern inline uint32_t +div_long_long_rem(uint64_t n, uint32_t base, uint32_t *rem) +{ + *rem = do_div(n, base); + return n; +} + #else /* __s390x__ */ #include #endif /* __s390x__ */ diff -Nrup linux-2.6.0-test1.orig/include/linux/time.h linux-2.6.0-test1/include/linux/time.h --- linux-2.6.0-test1.orig/include/linux/time.h 2003-07-14 05:35:16.000000000 +0200 +++ linux-2.6.0-test1/include/linux/time.h 2003-07-15 05:20:26.000000000 +0200 @@ -28,14 +28,6 @@ struct timezone { #include #include #include -#ifndef div_long_long_rem - -#define div_long_long_rem(dividend,divisor,remainder) ({ \ - u64 result = dividend; \ - *remainder = do_div(result,divisor); \ - result; }) - -#endif /* * Have the 32 bit jiffies value wrap 5 minutes after boot @@ -122,7 +114,7 @@ jiffies_to_timespec(unsigned long jiffie * one divide. */ u64 nsec = (u64)jiffies * TICK_NSEC; - value->tv_sec = div_long_long_rem(nsec, NSEC_PER_SEC, &value->tv_nsec); + value->tv_sec = div_long_long_rem(nsec, NSEC_PER_SEC, (uint32_t *)&value->tv_nsec); } /* Same for "timeval" */ @@ -150,7 +142,7 @@ jiffies_to_timeval(unsigned long jiffies * one divide. */ u64 nsec = (u64)jiffies * TICK_NSEC; - value->tv_sec = div_long_long_rem(nsec, NSEC_PER_SEC, &value->tv_usec); + value->tv_sec = div_long_long_rem(nsec, NSEC_PER_SEC, (uint32_t *)&value->tv_usec); value->tv_usec /= NSEC_PER_USEC; } diff -Nrup linux-2.6.0-test1.orig/kernel/posix-timers.c linux-2.6.0-test1/kernel/posix-timers.c --- linux-2.6.0-test1.orig/kernel/posix-timers.c 2003-07-14 05:34:32.000000000 +0200 +++ linux-2.6.0-test1/kernel/posix-timers.c 2003-07-15 05:21:24.000000000 +0200 @@ -23,16 +23,8 @@ #include #include #include - -#ifndef div_long_long_rem #include -#define div_long_long_rem(dividend,divisor,remainder) ({ \ - u64 result = dividend; \ - *remainder = do_div(result,divisor); \ - result; }) - -#endif #define CLOCK_REALTIME_RES TICK_NSEC // In nano seconds. static inline u64 mpy_l_X_l_ll(unsigned long mpy1,unsigned long mpy2) @@ -1226,7 +1218,7 @@ do_clock_nanosleep(clockid_t which_clock left *= TICK_NSEC; tsave->tv_sec = div_long_long_rem(left, NSEC_PER_SEC, - &tsave->tv_nsec); + (uint32_t *)&tsave->tv_nsec); restart_block->fn = clock_nanosleep_restart; restart_block->arg0 = which_clock; restart_block->arg1 = (unsigned long)tsave;