aboutsummaryrefslogtreecommitdiffstats
path: root/arch/m68k/include
diff options
context:
space:
mode:
Diffstat (limited to 'arch/m68k/include')
-rw-r--r--arch/m68k/include/asm/delay.h97
-rw-r--r--arch/m68k/include/asm/delay_mm.h57
-rw-r--r--arch/m68k/include/asm/delay_no.h76
3 files changed, 94 insertions, 136 deletions
diff --git a/arch/m68k/include/asm/delay.h b/arch/m68k/include/asm/delay.h
index d2598e3dd7b2..9c09becfd4c9 100644
--- a/arch/m68k/include/asm/delay.h
+++ b/arch/m68k/include/asm/delay.h
@@ -1,5 +1,96 @@
1#ifdef __uClinux__ 1#ifndef _M68K_DELAY_H
2#include "delay_no.h" 2#define _M68K_DELAY_H
3
4#include <asm/param.h>
5
6/*
7 * Copyright (C) 1994 Hamish Macdonald
8 * Copyright (C) 2004 Greg Ungerer <gerg@uclinux.com>
9 *
10 * Delay routines, using a pre-computed "loops_per_jiffy" value.
11 */
12
13#if defined(CONFIG_COLDFIRE)
14/*
15 * The ColdFire runs the delay loop at significantly different speeds
16 * depending upon long word alignment or not. We'll pad it to
17 * long word alignment which is the faster version.
18 * The 0x4a8e is of course a 'tstl %fp' instruction. This is better
19 * than using a NOP (0x4e71) instruction because it executes in one
20 * cycle not three and doesn't allow for an arbitrary delay waiting
21 * for bus cycles to finish. Also fp/a6 isn't likely to cause a
22 * stall waiting for the register to become valid if such is added
23 * to the coldfire at some stage.
24 */
25#define DELAY_ALIGN ".balignw 4, 0x4a8e\n\t"
3#else 26#else
4#include "delay_mm.h" 27/*
28 * No instruction alignment required for other m68k types.
29 */
30#define DELAY_ALIGN
5#endif 31#endif
32
33static inline void __delay(unsigned long loops)
34{
35 __asm__ __volatile__ (
36 DELAY_ALIGN
37 "1: subql #1,%0\n\t"
38 "jcc 1b"
39 : "=d" (loops)
40 : "0" (loops));
41}
42
43extern void __bad_udelay(void);
44
45
46#if defined(CONFIG_M68000) || defined(CONFIG_COLDFIRE)
47/*
48 * The simpler m68k and ColdFire processors do not have a 32*32->64
49 * multiply instruction. So we need to handle them a little differently.
50 * We use a bit of shifting and a single 32*32->32 multiply to get close.
51 * This is a macro so that the const version can factor out the first
52 * multiply and shift.
53 */
54#define HZSCALE (268435456 / (1000000 / HZ))
55
56#define __const_udelay(u) \
57 __delay(((((u) * HZSCALE) >> 11) * (loops_per_jiffy >> 11)) >> 6)
58
59#else
60
61static inline void __xdelay(unsigned long xloops)
62{
63 unsigned long tmp;
64
65 __asm__ ("mulul %2,%0:%1"
66 : "=d" (xloops), "=d" (tmp)
67 : "d" (xloops), "1" (loops_per_jiffy));
68 __delay(xloops * HZ);
69}
70
71/*
72 * The definition of __const_udelay is specifically made a macro so that
73 * the const factor (4295 = 2**32 / 1000000) can be optimized out when
74 * the delay is a const.
75 */
76#define __const_udelay(n) (__xdelay((n) * 4295))
77
78#endif
79
80static inline void __udelay(unsigned long usecs)
81{
82 __const_udelay(usecs);
83}
84
85/*
86 * Use only for very small delays ( < 1 msec). Should probably use a
87 * lookup table, really, as the multiplications take much too long with
88 * short delays. This is a "reasonable" implementation, though (and the
89 * first constant multiplications gets optimized away if the delay is
90 * a constant)
91 */
92#define udelay(n) (__builtin_constant_p(n) ? \
93 ((n) > 20000 ? __bad_udelay() : __const_udelay(n)) : __udelay(n))
94
95
96#endif /* defined(_M68K_DELAY_H) */
diff --git a/arch/m68k/include/asm/delay_mm.h b/arch/m68k/include/asm/delay_mm.h
deleted file mode 100644
index 5ed92851bc6f..000000000000
--- a/arch/m68k/include/asm/delay_mm.h
+++ /dev/null
@@ -1,57 +0,0 @@
1#ifndef _M68K_DELAY_H
2#define _M68K_DELAY_H
3
4#include <asm/param.h>
5
6/*
7 * Copyright (C) 1994 Hamish Macdonald
8 *
9 * Delay routines, using a pre-computed "loops_per_jiffy" value.
10 */
11
12static inline void __delay(unsigned long loops)
13{
14 __asm__ __volatile__ ("1: subql #1,%0; jcc 1b"
15 : "=d" (loops) : "0" (loops));
16}
17
18extern void __bad_udelay(void);
19
20/*
21 * Use only for very small delays ( < 1 msec). Should probably use a
22 * lookup table, really, as the multiplications take much too long with
23 * short delays. This is a "reasonable" implementation, though (and the
24 * first constant multiplications gets optimized away if the delay is
25 * a constant)
26 */
27static inline void __const_udelay(unsigned long xloops)
28{
29 unsigned long tmp;
30
31 __asm__ ("mulul %2,%0:%1"
32 : "=d" (xloops), "=d" (tmp)
33 : "d" (xloops), "1" (loops_per_jiffy));
34 __delay(xloops * HZ);
35}
36
37static inline void __udelay(unsigned long usecs)
38{
39 __const_udelay(usecs * 4295); /* 2**32 / 1000000 */
40}
41
42#define udelay(n) (__builtin_constant_p(n) ? \
43 ((n) > 20000 ? __bad_udelay() : __const_udelay((n) * 4295)) : \
44 __udelay(n))
45
46static inline unsigned long muldiv(unsigned long a, unsigned long b,
47 unsigned long c)
48{
49 unsigned long tmp;
50
51 __asm__ ("mulul %2,%0:%1; divul %3,%0:%1"
52 : "=d" (tmp), "=d" (a)
53 : "d" (b), "d" (c), "1" (a));
54 return a;
55}
56
57#endif /* defined(_M68K_DELAY_H) */
diff --git a/arch/m68k/include/asm/delay_no.h b/arch/m68k/include/asm/delay_no.h
deleted file mode 100644
index c3a0edc90f21..000000000000
--- a/arch/m68k/include/asm/delay_no.h
+++ /dev/null
@@ -1,76 +0,0 @@
1#ifndef _M68KNOMMU_DELAY_H
2#define _M68KNOMMU_DELAY_H
3
4/*
5 * Copyright (C) 1994 Hamish Macdonald
6 * Copyright (C) 2004 Greg Ungerer <gerg@snapgear.com>
7 */
8
9#include <asm/param.h>
10
11static inline void __delay(unsigned long loops)
12{
13#if defined(CONFIG_COLDFIRE)
14 /* The coldfire runs this loop at significantly different speeds
15 * depending upon long word alignment or not. We'll pad it to
16 * long word alignment which is the faster version.
17 * The 0x4a8e is of course a 'tstl %fp' instruction. This is better
18 * than using a NOP (0x4e71) instruction because it executes in one
19 * cycle not three and doesn't allow for an arbitrary delay waiting
20 * for bus cycles to finish. Also fp/a6 isn't likely to cause a
21 * stall waiting for the register to become valid if such is added
22 * to the coldfire at some stage.
23 */
24 __asm__ __volatile__ ( ".balignw 4, 0x4a8e\n\t"
25 "1: subql #1, %0\n\t"
26 "jcc 1b"
27 : "=d" (loops) : "0" (loops));
28#else
29 __asm__ __volatile__ ( "1: subql #1, %0\n\t"
30 "jcc 1b"
31 : "=d" (loops) : "0" (loops));
32#endif
33}
34
35/*
36 * Ideally we use a 32*32->64 multiply to calculate the number of
37 * loop iterations, but the older standard 68k and ColdFire do not
38 * have this instruction. So for them we have a clsoe approximation
39 * loop using 32*32->32 multiplies only. This calculation based on
40 * the ARM version of delay.
41 *
42 * We want to implement:
43 *
44 * loops = (usecs * 0x10c6 * HZ * loops_per_jiffy) / 2^32
45 */
46
47#define HZSCALE (268435456 / (1000000/HZ))
48
49extern unsigned long loops_per_jiffy;
50
51static inline void _udelay(unsigned long usecs)
52{
53#if defined(CONFIG_M68328) || defined(CONFIG_M68EZ328) || \
54 defined(CONFIG_M68VZ328) || defined(CONFIG_M68360) || \
55 defined(CONFIG_COLDFIRE)
56 __delay((((usecs * HZSCALE) >> 11) * (loops_per_jiffy >> 11)) >> 6);
57#else
58 unsigned long tmp;
59
60 usecs *= 4295; /* 2**32 / 1000000 */
61 __asm__ ("mulul %2,%0:%1"
62 : "=d" (usecs), "=d" (tmp)
63 : "d" (usecs), "1" (loops_per_jiffy*HZ));
64 __delay(usecs);
65#endif
66}
67
68/*
69 * Moved the udelay() function into library code, no longer inlined.
70 * I had to change the algorithm because we are overflowing now on
71 * the faster ColdFire parts. The code is a little bigger, so it makes
72 * sense to library it.
73 */
74extern void udelay(unsigned long usecs);
75
76#endif /* defined(_M68KNOMMU_DELAY_H) */
generated by cgit v1.2.2 (git 2.25.0) at 2025-02-11 06:19:23 -0500