]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blob - arch/riscv/lib/delay.c
Merge remote-tracking branches 'spi/fix/armada', 'spi/fix/atmel', 'spi/fix/doc',...
[mirror_ubuntu-hirsute-kernel.git] / arch / riscv / lib / delay.c
1 /*
2 * Copyright (C) 2012 Regents of the University of California
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation, version 2.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
12 */
13
14 #include <linux/delay.h>
15 #include <linux/param.h>
16 #include <linux/timex.h>
17 #include <linux/export.h>
18
19 /*
20 * This is copies from arch/arm/include/asm/delay.h
21 *
22 * Loop (or tick) based delay:
23 *
24 * loops = loops_per_jiffy * jiffies_per_sec * delay_us / us_per_sec
25 *
26 * where:
27 *
28 * jiffies_per_sec = HZ
29 * us_per_sec = 1000000
30 *
31 * Therefore the constant part is HZ / 1000000 which is a small
32 * fractional number. To make this usable with integer math, we
33 * scale up this constant by 2^31, perform the actual multiplication,
34 * and scale the result back down by 2^31 with a simple shift:
35 *
36 * loops = (loops_per_jiffy * delay_us * UDELAY_MULT) >> 31
37 *
38 * where:
39 *
40 * UDELAY_MULT = 2^31 * HZ / 1000000
41 * = (2^31 / 1000000) * HZ
42 * = 2147.483648 * HZ
43 * = 2147 * HZ + 483648 * HZ / 1000000
44 *
45 * 31 is the biggest scale shift value that won't overflow 32 bits for
46 * delay_us * UDELAY_MULT assuming HZ <= 1000 and delay_us <= 2000.
47 */
48 #define MAX_UDELAY_US 2000
49 #define MAX_UDELAY_HZ 1000
50 #define UDELAY_MULT (2147UL * HZ + 483648UL * HZ / 1000000UL)
51 #define UDELAY_SHIFT 31
52
53 #if HZ > MAX_UDELAY_HZ
54 #error "HZ > MAX_UDELAY_HZ"
55 #endif
56
57 /*
58 * RISC-V supports both UDELAY and NDELAY. This is largely the same as above,
59 * but with different constants. I added 10 bits to the shift to get this, but
60 * the result is that I need a 64-bit multiply, which is slow on 32-bit
61 * platforms.
62 *
63 * NDELAY_MULT = 2^41 * HZ / 1000000000
64 * = (2^41 / 1000000000) * HZ
65 * = 2199.02325555 * HZ
66 * = 2199 * HZ + 23255550 * HZ / 1000000000
67 *
68 * The maximum here is to avoid 64-bit overflow, but it isn't checked as it
69 * won't happen.
70 */
71 #define MAX_NDELAY_NS (1ULL << 42)
72 #define MAX_NDELAY_HZ MAX_UDELAY_HZ
73 #define NDELAY_MULT ((unsigned long long)(2199ULL * HZ + 23255550ULL * HZ / 1000000000ULL))
74 #define NDELAY_SHIFT 41
75
76 #if HZ > MAX_NDELAY_HZ
77 #error "HZ > MAX_NDELAY_HZ"
78 #endif
79
80 void __delay(unsigned long cycles)
81 {
82 u64 t0 = get_cycles();
83
84 while ((unsigned long)(get_cycles() - t0) < cycles)
85 cpu_relax();
86 }
87 EXPORT_SYMBOL(__delay);
88
89 void udelay(unsigned long usecs)
90 {
91 unsigned long ucycles = usecs * lpj_fine * UDELAY_MULT;
92
93 if (unlikely(usecs > MAX_UDELAY_US)) {
94 __delay((u64)usecs * riscv_timebase / 1000000ULL);
95 return;
96 }
97
98 __delay(ucycles >> UDELAY_SHIFT);
99 }
100 EXPORT_SYMBOL(udelay);
101
102 void ndelay(unsigned long nsecs)
103 {
104 /*
105 * This doesn't bother checking for overflow, as it won't happen (it's
106 * an hour) of delay.
107 */
108 unsigned long long ncycles = nsecs * lpj_fine * NDELAY_MULT;
109 __delay(ncycles >> NDELAY_SHIFT);
110 }
111 EXPORT_SYMBOL(ndelay);