1*4882a593Smuzhiyun /* SPDX-License-Identifier: GPL-2.0 */
2*4882a593Smuzhiyun #ifndef _ASM_X86_DIV64_H
3*4882a593Smuzhiyun #define _ASM_X86_DIV64_H
4*4882a593Smuzhiyun
5*4882a593Smuzhiyun #ifdef CONFIG_X86_32
6*4882a593Smuzhiyun
7*4882a593Smuzhiyun #include <linux/types.h>
8*4882a593Smuzhiyun #include <linux/log2.h>
9*4882a593Smuzhiyun
10*4882a593Smuzhiyun /*
11*4882a593Smuzhiyun * do_div() is NOT a C function. It wants to return
12*4882a593Smuzhiyun * two values (the quotient and the remainder), but
13*4882a593Smuzhiyun * since that doesn't work very well in C, what it
14*4882a593Smuzhiyun * does is:
15*4882a593Smuzhiyun *
16*4882a593Smuzhiyun * - modifies the 64-bit dividend _in_place_
17*4882a593Smuzhiyun * - returns the 32-bit remainder
18*4882a593Smuzhiyun *
19*4882a593Smuzhiyun * This ends up being the most efficient "calling
20*4882a593Smuzhiyun * convention" on x86.
21*4882a593Smuzhiyun */
22*4882a593Smuzhiyun #define do_div(n, base) \
23*4882a593Smuzhiyun ({ \
24*4882a593Smuzhiyun unsigned long __upper, __low, __high, __mod, __base; \
25*4882a593Smuzhiyun __base = (base); \
26*4882a593Smuzhiyun if (__builtin_constant_p(__base) && is_power_of_2(__base)) { \
27*4882a593Smuzhiyun __mod = n & (__base - 1); \
28*4882a593Smuzhiyun n >>= ilog2(__base); \
29*4882a593Smuzhiyun } else { \
30*4882a593Smuzhiyun asm("" : "=a" (__low), "=d" (__high) : "A" (n));\
31*4882a593Smuzhiyun __upper = __high; \
32*4882a593Smuzhiyun if (__high) { \
33*4882a593Smuzhiyun __upper = __high % (__base); \
34*4882a593Smuzhiyun __high = __high / (__base); \
35*4882a593Smuzhiyun } \
36*4882a593Smuzhiyun asm("divl %2" : "=a" (__low), "=d" (__mod) \
37*4882a593Smuzhiyun : "rm" (__base), "0" (__low), "1" (__upper)); \
38*4882a593Smuzhiyun asm("" : "=A" (n) : "a" (__low), "d" (__high)); \
39*4882a593Smuzhiyun } \
40*4882a593Smuzhiyun __mod; \
41*4882a593Smuzhiyun })
42*4882a593Smuzhiyun
div_u64_rem(u64 dividend,u32 divisor,u32 * remainder)43*4882a593Smuzhiyun static inline u64 div_u64_rem(u64 dividend, u32 divisor, u32 *remainder)
44*4882a593Smuzhiyun {
45*4882a593Smuzhiyun union {
46*4882a593Smuzhiyun u64 v64;
47*4882a593Smuzhiyun u32 v32[2];
48*4882a593Smuzhiyun } d = { dividend };
49*4882a593Smuzhiyun u32 upper;
50*4882a593Smuzhiyun
51*4882a593Smuzhiyun upper = d.v32[1];
52*4882a593Smuzhiyun d.v32[1] = 0;
53*4882a593Smuzhiyun if (upper >= divisor) {
54*4882a593Smuzhiyun d.v32[1] = upper / divisor;
55*4882a593Smuzhiyun upper %= divisor;
56*4882a593Smuzhiyun }
57*4882a593Smuzhiyun asm ("divl %2" : "=a" (d.v32[0]), "=d" (*remainder) :
58*4882a593Smuzhiyun "rm" (divisor), "0" (d.v32[0]), "1" (upper));
59*4882a593Smuzhiyun return d.v64;
60*4882a593Smuzhiyun }
61*4882a593Smuzhiyun #define div_u64_rem div_u64_rem
62*4882a593Smuzhiyun
mul_u32_u32(u32 a,u32 b)63*4882a593Smuzhiyun static inline u64 mul_u32_u32(u32 a, u32 b)
64*4882a593Smuzhiyun {
65*4882a593Smuzhiyun u32 high, low;
66*4882a593Smuzhiyun
67*4882a593Smuzhiyun asm ("mull %[b]" : "=a" (low), "=d" (high)
68*4882a593Smuzhiyun : [a] "a" (a), [b] "rm" (b) );
69*4882a593Smuzhiyun
70*4882a593Smuzhiyun return low | ((u64)high) << 32;
71*4882a593Smuzhiyun }
72*4882a593Smuzhiyun #define mul_u32_u32 mul_u32_u32
73*4882a593Smuzhiyun
74*4882a593Smuzhiyun #else
75*4882a593Smuzhiyun # include <asm-generic/div64.h>
76*4882a593Smuzhiyun
77*4882a593Smuzhiyun /*
78*4882a593Smuzhiyun * Will generate an #DE when the result doesn't fit u64, could fix with an
79*4882a593Smuzhiyun * __ex_table[] entry when it becomes an issue.
80*4882a593Smuzhiyun */
mul_u64_u64_div_u64(u64 a,u64 mul,u64 div)81*4882a593Smuzhiyun static inline u64 mul_u64_u64_div_u64(u64 a, u64 mul, u64 div)
82*4882a593Smuzhiyun {
83*4882a593Smuzhiyun u64 q;
84*4882a593Smuzhiyun
85*4882a593Smuzhiyun asm ("mulq %2; divq %3" : "=a" (q)
86*4882a593Smuzhiyun : "a" (a), "rm" (mul), "rm" (div)
87*4882a593Smuzhiyun : "rdx");
88*4882a593Smuzhiyun
89*4882a593Smuzhiyun return q;
90*4882a593Smuzhiyun }
91*4882a593Smuzhiyun #define mul_u64_u64_div_u64 mul_u64_u64_div_u64
92*4882a593Smuzhiyun
mul_u64_u32_div(u64 a,u32 mul,u32 div)93*4882a593Smuzhiyun static inline u64 mul_u64_u32_div(u64 a, u32 mul, u32 div)
94*4882a593Smuzhiyun {
95*4882a593Smuzhiyun return mul_u64_u64_div_u64(a, mul, div);
96*4882a593Smuzhiyun }
97*4882a593Smuzhiyun #define mul_u64_u32_div mul_u64_u32_div
98*4882a593Smuzhiyun
99*4882a593Smuzhiyun #endif /* CONFIG_X86_32 */
100*4882a593Smuzhiyun
101*4882a593Smuzhiyun #endif /* _ASM_X86_DIV64_H */
102