diff options
author | Martin Schwidefsky <schwidefsky@de.ibm.com> | 2006-09-28 10:55:39 -0400 |
---|---|---|
committer | Martin Schwidefsky <schwidefsky@de.ibm.com> | 2006-09-28 10:55:39 -0400 |
commit | d9f7a745d55527d0d41684b22506a86c4381f7f1 (patch) | |
tree | ea8870ef06c3723ad59b78aac97bfe8152894c72 /arch | |
parent | 1fce518e8e7de62597c823d6d795cafc694e7910 (diff) |
[S390] __div64_32 for 31 bit.
The clocksource infrastructure introduced with commit
ad596171ed635c51a9eef829187af100cbf8dcf7 broke 31 bit s390.
The reason is that the do_div() primitive for 31 bit always
had a restriction: it could only divide an unsigned 64 bit
integer by an unsigned 31 bit integer. The clocksource code
now uses do_div() with a base value that has the most
significant bit set. The result is that clock->cycle_interval
has a funny value which causes the linux time to jump around
like mad.
The solution is "obvious": implement a proper __div64_32
function for 31 bit s390.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Diffstat (limited to 'arch')
-rw-r--r-- | arch/s390/Kconfig | 4 | ||||
-rw-r--r-- | arch/s390/lib/Makefile | 1 | ||||
-rw-r--r-- | arch/s390/lib/div64.c | 151 |
3 files changed, 156 insertions, 0 deletions
diff --git a/arch/s390/Kconfig b/arch/s390/Kconfig index b216ca659cdf..b6b42f9f0d52 100644 --- a/arch/s390/Kconfig +++ b/arch/s390/Kconfig | |||
@@ -51,6 +51,10 @@ config 64BIT | |||
51 | Select this option if you have a 64 bit IBM zSeries machine | 51 | Select this option if you have a 64 bit IBM zSeries machine |
52 | and want to use the 64 bit addressing mode. | 52 | and want to use the 64 bit addressing mode. |
53 | 53 | ||
54 | config 32BIT | ||
55 | bool | ||
56 | default y if !64BIT | ||
57 | |||
54 | config SMP | 58 | config SMP |
55 | bool "Symmetric multi-processing support" | 59 | bool "Symmetric multi-processing support" |
56 | ---help--- | 60 | ---help--- |
diff --git a/arch/s390/lib/Makefile b/arch/s390/lib/Makefile index c42ffedfdb49..b0cfa6c4883d 100644 --- a/arch/s390/lib/Makefile +++ b/arch/s390/lib/Makefile | |||
@@ -5,5 +5,6 @@ | |||
5 | EXTRA_AFLAGS := -traditional | 5 | EXTRA_AFLAGS := -traditional |
6 | 6 | ||
7 | lib-y += delay.o string.o uaccess_std.o | 7 | lib-y += delay.o string.o uaccess_std.o |
8 | lib-$(CONFIG_32BIT) += div64.o | ||
8 | lib-$(CONFIG_64BIT) += uaccess_mvcos.o | 9 | lib-$(CONFIG_64BIT) += uaccess_mvcos.o |
9 | lib-$(CONFIG_SMP) += spinlock.o | 10 | lib-$(CONFIG_SMP) += spinlock.o |
diff --git a/arch/s390/lib/div64.c b/arch/s390/lib/div64.c new file mode 100644 index 000000000000..0481f3424a13 --- /dev/null +++ b/arch/s390/lib/div64.c | |||
@@ -0,0 +1,151 @@ | |||
1 | /* | ||
2 | * arch/s390/lib/div64.c | ||
3 | * | ||
4 | * __div64_32 implementation for 31 bit. | ||
5 | * | ||
6 | * Copyright (C) IBM Corp. 2006 | ||
7 | * Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com), | ||
8 | */ | ||
9 | |||
10 | #include <linux/types.h> | ||
11 | #include <linux/module.h> | ||
12 | |||
13 | #ifdef CONFIG_MARCH_G5 | ||
14 | |||
15 | /* | ||
16 | * Function to divide an unsigned 64 bit integer by an unsigned | ||
17 | * 31 bit integer using signed 64/32 bit division. | ||
18 | */ | ||
19 | static uint32_t __div64_31(uint64_t *n, uint32_t base) | ||
20 | { | ||
21 | register uint32_t reg2 asm("2"); | ||
22 | register uint32_t reg3 asm("3"); | ||
23 | uint32_t *words = (uint32_t *) n; | ||
24 | uint32_t tmp; | ||
25 | |||
26 | /* Special case base==1, remainder = 0, quotient = n */ | ||
27 | if (base == 1) | ||
28 | return 0; | ||
29 | /* | ||
30 | * Special case base==0 will cause a fixed point divide exception | ||
31 | * on the dr instruction and may not happen anyway. For the | ||
32 | * following calculation we can assume base > 1. The first | ||
33 | * signed 64 / 32 bit division with an upper half of 0 will | ||
34 | * give the correct upper half of the 64 bit quotient. | ||
35 | */ | ||
36 | reg2 = 0UL; | ||
37 | reg3 = words[0]; | ||
38 | asm volatile( | ||
39 | " dr %0,%2\n" | ||
40 | : "+d" (reg2), "+d" (reg3) : "d" (base) : "cc" ); | ||
41 | words[0] = reg3; | ||
42 | reg3 = words[1]; | ||
43 | /* | ||
44 | * To get the lower half of the 64 bit quotient and the 32 bit | ||
45 | * remainder we have to use a little trick. Since we only have | ||
46 | * a signed division the quotient can get too big. To avoid this | ||
47 | * the 64 bit dividend is halved, then the signed division will | ||
48 | * work. Afterwards the quotient and the remainder are doubled. | ||
49 | * If the last bit of the dividend has been one the remainder | ||
50 | * is increased by one then checked against the base. If the | ||
51 | * remainder has overflown subtract base and increase the | ||
52 | * quotient. Simple, no ? | ||
53 | */ | ||
54 | asm volatile( | ||
55 | " nr %2,%1\n" | ||
56 | " srdl %0,1\n" | ||
57 | " dr %0,%3\n" | ||
58 | " alr %0,%0\n" | ||
59 | " alr %1,%1\n" | ||
60 | " alr %0,%2\n" | ||
61 | " clr %0,%3\n" | ||
62 | " jl 0f\n" | ||
63 | " slr %0,%3\n" | ||
64 | " alr %1,%2\n" | ||
65 | "0:\n" | ||
66 | : "+d" (reg2), "+d" (reg3), "=d" (tmp) | ||
67 | : "d" (base), "2" (1UL) : "cc" ); | ||
68 | words[1] = reg3; | ||
69 | return reg2; | ||
70 | } | ||
71 | |||
72 | /* | ||
73 | * Function to divide an unsigned 64 bit integer by an unsigned | ||
74 | * 32 bit integer using the unsigned 64/31 bit division. | ||
75 | */ | ||
76 | uint32_t __div64_32(uint64_t *n, uint32_t base) | ||
77 | { | ||
78 | uint32_t r; | ||
79 | |||
80 | /* | ||
81 | * If the most significant bit of base is set, divide n by | ||
82 | * (base/2). That allows to use 64/31 bit division and gives a | ||
83 | * good approximation of the result: n = (base/2)*q + r. The | ||
84 | * result needs to be corrected with two simple transformations. | ||
85 | * If base is already < 2^31-1 __div64_31 can be used directly. | ||
86 | */ | ||
87 | r = __div64_31(n, ((signed) base < 0) ? (base/2) : base); | ||
88 | if ((signed) base < 0) { | ||
89 | uint64_t q = *n; | ||
90 | /* | ||
91 | * First transformation: | ||
92 | * n = (base/2)*q + r | ||
93 | * = ((base/2)*2)*(q/2) + ((q&1) ? (base/2) : 0) + r | ||
94 | * Since r < (base/2), r + (base/2) < base. | ||
95 | * With q1 = (q/2) and r1 = r + ((q&1) ? (base/2) : 0) | ||
96 | * n = ((base/2)*2)*q1 + r1 with r1 < base. | ||
97 | */ | ||
98 | if (q & 1) | ||
99 | r += base/2; | ||
100 | q >>= 1; | ||
101 | /* | ||
102 | * Second transformation. ((base/2)*2) could have lost the | ||
103 | * last bit. | ||
104 | * n = ((base/2)*2)*q1 + r1 | ||
105 | * = base*q1 - ((base&1) ? q1 : 0) + r1 | ||
106 | */ | ||
107 | if (base & 1) { | ||
108 | int64_t rx = r - q; | ||
109 | /* | ||
110 | * base is >= 2^31. The worst case for the while | ||
111 | * loop is n=2^64-1 base=2^31+1. That gives a | ||
112 | * maximum for q=(2^64-1)/2^31 = 0x1ffffffff. Since | ||
113 | * base >= 2^31 the loop is finished after a maximum | ||
114 | * of three iterations. | ||
115 | */ | ||
116 | while (rx < 0) { | ||
117 | rx += base; | ||
118 | q--; | ||
119 | } | ||
120 | r = rx; | ||
121 | } | ||
122 | *n = q; | ||
123 | } | ||
124 | return r; | ||
125 | } | ||
126 | |||
127 | #else /* MARCH_G5 */ | ||
128 | |||
129 | uint32_t __div64_32(uint64_t *n, uint32_t base) | ||
130 | { | ||
131 | register uint32_t reg2 asm("2"); | ||
132 | register uint32_t reg3 asm("3"); | ||
133 | uint32_t *words = (uint32_t *) n; | ||
134 | |||
135 | reg2 = 0UL; | ||
136 | reg3 = words[0]; | ||
137 | asm volatile( | ||
138 | " dlr %0,%2\n" | ||
139 | : "+d" (reg2), "+d" (reg3) : "d" (base) : "cc" ); | ||
140 | words[0] = reg3; | ||
141 | reg3 = words[1]; | ||
142 | asm volatile( | ||
143 | " dlr %0,%2\n" | ||
144 | : "+d" (reg2), "+d" (reg3) : "d" (base) : "cc" ); | ||
145 | words[1] = reg3; | ||
146 | return reg2; | ||
147 | } | ||
148 | |||
149 | #endif /* MARCH_G5 */ | ||
150 | |||
151 | EXPORT_SYMBOL(__div64_32); | ||