1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright (C) 2007-2010 Lawrence Livermore National Security, LLC.
4 * Copyright (C) 2007 The Regents of the University of California.
5 * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
6 * Written by Brian Behlendorf <behlendorf1@llnl.gov>.
7 * UCRL-CODE-235197
8 *
9 * This file is part of the SPL, Solaris Porting Layer.
10 *
11 * The SPL is free software; you can redistribute it and/or modify it
12 * under the terms of the GNU General Public License as published by the
13 * Free Software Foundation; either version 2 of the License, or (at your
14 * option) any later version.
15 *
16 * The SPL is distributed in the hope that it will be useful, but WITHOUT
17 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
18 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
19 * for more details.
20 *
21 * You should have received a copy of the GNU General Public License along
22 * with the SPL. If not, see <http://www.gnu.org/licenses/>.
23 *
24 * Solaris Porting Layer (SPL) Generic Implementation.
25 */
26
27 #include <sys/isa_defs.h>
28 #include <sys/sysmacros.h>
29
30 /*
31 * 64-bit math support for 32-bit platforms. Compilers will generatee
32 * references to the functions here if required.
33 */
34
35 #if BITS_PER_LONG == 32
36
37 /*
38 * Support 64/64 => 64 division on a 32-bit platform. While the kernel
39 * provides a div64_u64() function for this we do not use it because the
40 * implementation is flawed. There are cases which return incorrect
41 * results as late as linux-2.6.35. Until this is fixed upstream the
42 * spl must provide its own implementation.
43 *
44 * This implementation is a slightly modified version of the algorithm
45 * proposed by the book 'Hacker's Delight'. The original source can be
46 * found here and is available for use without restriction.
47 *
48 * http://www.hackersdelight.org/HDcode/newCode/divDouble.c
49 */
50
51 /*
52 * Calculate number of leading of zeros for a 64-bit value.
53 */
54 static int
nlz64(uint64_t x)55 nlz64(uint64_t x)
56 {
57 register int n = 0;
58
59 if (x == 0)
60 return (64);
61
62 if (x <= 0x00000000FFFFFFFFULL) { n = n + 32; x = x << 32; }
63 if (x <= 0x0000FFFFFFFFFFFFULL) { n = n + 16; x = x << 16; }
64 if (x <= 0x00FFFFFFFFFFFFFFULL) { n = n + 8; x = x << 8; }
65 if (x <= 0x0FFFFFFFFFFFFFFFULL) { n = n + 4; x = x << 4; }
66 if (x <= 0x3FFFFFFFFFFFFFFFULL) { n = n + 2; x = x << 2; }
67 if (x <= 0x7FFFFFFFFFFFFFFFULL) { n = n + 1; }
68
69 return (n);
70 }
71
72 /*
73 * Newer kernels have a div_u64() function but we define our own
74 * to simplify portability between kernel versions.
75 */
76 static inline uint64_t
__div_u64(uint64_t u,uint32_t v)77 __div_u64(uint64_t u, uint32_t v)
78 {
79 (void) do_div(u, v);
80 return (u);
81 }
82
83 /*
84 * Implementation of 64-bit unsigned division for 32-bit machines.
85 *
86 * First the procedure takes care of the case in which the divisor is a
87 * 32-bit quantity. There are two subcases: (1) If the left half of the
88 * dividend is less than the divisor, one execution of do_div() is all that
89 * is required (overflow is not possible). (2) Otherwise it does two
90 * divisions, using the grade school method.
91 */
92 uint64_t
__udivdi3(uint64_t u,uint64_t v)93 __udivdi3(uint64_t u, uint64_t v)
94 {
95 uint64_t u0, u1, v1, q0, q1, k;
96 int n;
97
98 if (v >> 32 == 0) { // If v < 2**32:
99 if (u >> 32 < v) { // If u/v cannot overflow,
100 return (__div_u64(u, v)); // just do one division.
101 } else { // If u/v would overflow:
102 u1 = u >> 32; // Break u into two halves.
103 u0 = u & 0xFFFFFFFF;
104 q1 = __div_u64(u1, v); // First quotient digit.
105 k = u1 - q1 * v; // First remainder, < v.
106 u0 += (k << 32);
107 q0 = __div_u64(u0, v); // Seconds quotient digit.
108 return ((q1 << 32) + q0);
109 }
110 } else { // If v >= 2**32:
111 n = nlz64(v); // 0 <= n <= 31.
112 v1 = (v << n) >> 32; // Normalize divisor, MSB is 1.
113 u1 = u >> 1; // To ensure no overflow.
114 q1 = __div_u64(u1, v1); // Get quotient from
115 q0 = (q1 << n) >> 31; // Undo normalization and
116 // division of u by 2.
117 if (q0 != 0) // Make q0 correct or
118 q0 = q0 - 1; // too small by 1.
119 if ((u - q0 * v) >= v)
120 q0 = q0 + 1; // Now q0 is correct.
121
122 return (q0);
123 }
124 }
125 EXPORT_SYMBOL(__udivdi3);
126
127 #ifndef abs64
128 /* CSTYLED */
129 #define abs64(x) ({ uint64_t t = (x) >> 63; ((x) ^ t) - t; })
130 #endif
131
132 /*
133 * Implementation of 64-bit signed division for 32-bit machines.
134 */
135 int64_t
__divdi3(int64_t u,int64_t v)136 __divdi3(int64_t u, int64_t v)
137 {
138 int64_t q, t;
139 q = __udivdi3(abs64(u), abs64(v));
140 t = (u ^ v) >> 63; // If u, v have different
141 return ((q ^ t) - t); // signs, negate q.
142 }
143 EXPORT_SYMBOL(__divdi3);
144
145 /*
146 * Implementation of 64-bit unsigned modulo for 32-bit machines.
147 */
148 uint64_t
__umoddi3(uint64_t dividend,uint64_t divisor)149 __umoddi3(uint64_t dividend, uint64_t divisor)
150 {
151 return (dividend - (divisor * __udivdi3(dividend, divisor)));
152 }
153 EXPORT_SYMBOL(__umoddi3);
154
155 /* 64-bit signed modulo for 32-bit machines. */
156 int64_t
__moddi3(int64_t n,int64_t d)157 __moddi3(int64_t n, int64_t d)
158 {
159 int64_t q;
160 boolean_t nn = B_FALSE;
161
162 if (n < 0) {
163 nn = B_TRUE;
164 n = -n;
165 }
166 if (d < 0)
167 d = -d;
168
169 q = __umoddi3(n, d);
170
171 return (nn ? -q : q);
172 }
173 EXPORT_SYMBOL(__moddi3);
174
175 /*
176 * Implementation of 64-bit unsigned division/modulo for 32-bit machines.
177 */
178 uint64_t
__udivmoddi4(uint64_t n,uint64_t d,uint64_t * r)179 __udivmoddi4(uint64_t n, uint64_t d, uint64_t *r)
180 {
181 uint64_t q = __udivdi3(n, d);
182 if (r)
183 *r = n - d * q;
184 return (q);
185 }
186 EXPORT_SYMBOL(__udivmoddi4);
187
188 /*
189 * Implementation of 64-bit signed division/modulo for 32-bit machines.
190 */
191 int64_t
__divmoddi4(int64_t n,int64_t d,int64_t * r)192 __divmoddi4(int64_t n, int64_t d, int64_t *r)
193 {
194 int64_t q, rr;
195 boolean_t nn = B_FALSE;
196 boolean_t nd = B_FALSE;
197 if (n < 0) {
198 nn = B_TRUE;
199 n = -n;
200 }
201 if (d < 0) {
202 nd = B_TRUE;
203 d = -d;
204 }
205
206 q = __udivmoddi4(n, d, (uint64_t *)&rr);
207
208 if (nn != nd)
209 q = -q;
210 if (nn)
211 rr = -rr;
212 if (r)
213 *r = rr;
214 return (q);
215 }
216 EXPORT_SYMBOL(__divmoddi4);
217
218 #if defined(__arm) || defined(__arm__)
219 /*
220 * Implementation of 64-bit (un)signed division for 32-bit arm machines.
221 *
222 * Run-time ABI for the ARM Architecture (page 20). A pair of (unsigned)
223 * long longs is returned in {{r0, r1}, {r2,r3}}, the quotient in {r0, r1},
224 * and the remainder in {r2, r3}. The return type is specifically left
225 * set to 'void' to ensure the compiler does not overwrite these registers
226 * during the return. All results are in registers as per ABI
227 */
228 void
__aeabi_uldivmod(uint64_t u,uint64_t v)229 __aeabi_uldivmod(uint64_t u, uint64_t v)
230 {
231 uint64_t res;
232 uint64_t mod;
233
234 res = __udivdi3(u, v);
235 mod = __umoddi3(u, v);
236 {
237 register uint32_t r0 asm("r0") = (res & 0xFFFFFFFF);
238 register uint32_t r1 asm("r1") = (res >> 32);
239 register uint32_t r2 asm("r2") = (mod & 0xFFFFFFFF);
240 register uint32_t r3 asm("r3") = (mod >> 32);
241
242 asm volatile(""
243 : "+r"(r0), "+r"(r1), "+r"(r2), "+r"(r3) /* output */
244 : "r"(r0), "r"(r1), "r"(r2), "r"(r3)); /* input */
245
246 return; /* r0; */
247 }
248 }
249 EXPORT_SYMBOL(__aeabi_uldivmod);
250
251 void
__aeabi_ldivmod(int64_t u,int64_t v)252 __aeabi_ldivmod(int64_t u, int64_t v)
253 {
254 int64_t res;
255 uint64_t mod;
256
257 res = __divdi3(u, v);
258 mod = __umoddi3(u, v);
259 {
260 register uint32_t r0 asm("r0") = (res & 0xFFFFFFFF);
261 register uint32_t r1 asm("r1") = (res >> 32);
262 register uint32_t r2 asm("r2") = (mod & 0xFFFFFFFF);
263 register uint32_t r3 asm("r3") = (mod >> 32);
264
265 asm volatile(""
266 : "+r"(r0), "+r"(r1), "+r"(r2), "+r"(r3) /* output */
267 : "r"(r0), "r"(r1), "r"(r2), "r"(r3)); /* input */
268
269 return; /* r0; */
270 }
271 }
272 EXPORT_SYMBOL(__aeabi_ldivmod);
273 #endif /* __arm || __arm__ */
274
275 #endif /* BITS_PER_LONG */
276