1*4882a593Smuzhiyun /* SPDX-License-Identifier: GPL-2.0 */
2*4882a593Smuzhiyun /*
3*4882a593Smuzhiyun * Copyright (C) 2008 Michal Simek
4*4882a593Smuzhiyun * Copyright (C) 2007 John Williams
5*4882a593Smuzhiyun * Copyright (C) 2006 Atmark Techno, Inc.
6*4882a593Smuzhiyun */
7*4882a593Smuzhiyun
8*4882a593Smuzhiyun #ifndef _ASM_MICROBLAZE_DELAY_H
9*4882a593Smuzhiyun #define _ASM_MICROBLAZE_DELAY_H
10*4882a593Smuzhiyun
11*4882a593Smuzhiyun #include <linux/param.h>
12*4882a593Smuzhiyun
__delay(unsigned long loops)13*4882a593Smuzhiyun static inline void __delay(unsigned long loops)
14*4882a593Smuzhiyun {
15*4882a593Smuzhiyun asm volatile ("# __delay \n\t" \
16*4882a593Smuzhiyun "1: addi %0, %0, -1\t\n" \
17*4882a593Smuzhiyun "bneid %0, 1b \t\n" \
18*4882a593Smuzhiyun "nop \t\n"
19*4882a593Smuzhiyun : "=r" (loops)
20*4882a593Smuzhiyun : "0" (loops));
21*4882a593Smuzhiyun }
22*4882a593Smuzhiyun
23*4882a593Smuzhiyun /*
24*4882a593Smuzhiyun * Note that 19 * 226 == 4294 ==~ 2^32 / 10^6, so
25*4882a593Smuzhiyun * loops = (4294 * usecs * loops_per_jiffy * HZ) / 2^32.
26*4882a593Smuzhiyun *
27*4882a593Smuzhiyun * The mul instruction gives us loops = (a * b) / 2^32.
28*4882a593Smuzhiyun * We choose a = usecs * 19 * HZ and b = loops_per_jiffy * 226
29*4882a593Smuzhiyun * because this lets us support a wide range of HZ and
30*4882a593Smuzhiyun * loops_per_jiffy values without either a or b overflowing 2^32.
31*4882a593Smuzhiyun * Thus we need usecs * HZ <= (2^32 - 1) / 19 = 226050910 and
32*4882a593Smuzhiyun * loops_per_jiffy <= (2^32 - 1) / 226 = 19004280
33*4882a593Smuzhiyun * (which corresponds to ~3800 bogomips at HZ = 100).
34*4882a593Smuzhiyun * -- paulus
35*4882a593Smuzhiyun */
36*4882a593Smuzhiyun #define __MAX_UDELAY (226050910UL/HZ) /* maximum udelay argument */
37*4882a593Smuzhiyun #define __MAX_NDELAY (4294967295UL/HZ) /* maximum ndelay argument */
38*4882a593Smuzhiyun
39*4882a593Smuzhiyun extern unsigned long loops_per_jiffy;
40*4882a593Smuzhiyun
__udelay(unsigned int x)41*4882a593Smuzhiyun static inline void __udelay(unsigned int x)
42*4882a593Smuzhiyun {
43*4882a593Smuzhiyun
44*4882a593Smuzhiyun unsigned long long tmp =
45*4882a593Smuzhiyun (unsigned long long)x * (unsigned long long)loops_per_jiffy \
46*4882a593Smuzhiyun * 226LL;
47*4882a593Smuzhiyun unsigned loops = tmp >> 32;
48*4882a593Smuzhiyun
49*4882a593Smuzhiyun /*
50*4882a593Smuzhiyun __asm__("mulxuu %0,%1,%2" : "=r" (loops) :
51*4882a593Smuzhiyun "r" (x), "r" (loops_per_jiffy * 226));
52*4882a593Smuzhiyun */
53*4882a593Smuzhiyun __delay(loops);
54*4882a593Smuzhiyun }
55*4882a593Smuzhiyun
56*4882a593Smuzhiyun extern void __bad_udelay(void); /* deliberately undefined */
57*4882a593Smuzhiyun extern void __bad_ndelay(void); /* deliberately undefined */
58*4882a593Smuzhiyun
59*4882a593Smuzhiyun #define udelay(n) \
60*4882a593Smuzhiyun ({ \
61*4882a593Smuzhiyun if (__builtin_constant_p(n)) { \
62*4882a593Smuzhiyun if ((n) / __MAX_UDELAY >= 1) \
63*4882a593Smuzhiyun __bad_udelay(); \
64*4882a593Smuzhiyun else \
65*4882a593Smuzhiyun __udelay((n) * (19 * HZ)); \
66*4882a593Smuzhiyun } else { \
67*4882a593Smuzhiyun __udelay((n) * (19 * HZ)); \
68*4882a593Smuzhiyun } \
69*4882a593Smuzhiyun })
70*4882a593Smuzhiyun
71*4882a593Smuzhiyun #define ndelay(n) \
72*4882a593Smuzhiyun ({ \
73*4882a593Smuzhiyun if (__builtin_constant_p(n)) { \
74*4882a593Smuzhiyun if ((n) / __MAX_NDELAY >= 1) \
75*4882a593Smuzhiyun __bad_ndelay(); \
76*4882a593Smuzhiyun else \
77*4882a593Smuzhiyun __udelay((n) * HZ); \
78*4882a593Smuzhiyun } else { \
79*4882a593Smuzhiyun __udelay((n) * HZ); \
80*4882a593Smuzhiyun } \
81*4882a593Smuzhiyun })
82*4882a593Smuzhiyun
83*4882a593Smuzhiyun #define muldiv(a, b, c) (((a)*(b))/(c))
84*4882a593Smuzhiyun
85*4882a593Smuzhiyun #endif /* _ASM_MICROBLAZE_DELAY_H */
86