Root/
1 | /* |
2 | * Extend a 32-bit counter to 63 bits |
3 | * |
4 | * Author: Nicolas Pitre |
5 | * Created: December 3, 2006 |
6 | * Copyright: MontaVista Software, Inc. |
7 | * |
8 | * This program is free software; you can redistribute it and/or modify |
9 | * it under the terms of the GNU General Public License version 2 |
10 | * as published by the Free Software Foundation. |
11 | */ |
12 | |
13 | #ifndef __LINUX_CNT32_TO_63_H__ |
14 | #define __LINUX_CNT32_TO_63_H__ |
15 | |
16 | #include <linux/compiler.h> |
17 | #include <linux/types.h> |
18 | #include <asm/byteorder.h> |
19 | |
20 | /* this is used only to give gcc a clue about good code generation */ |
21 | union cnt32_to_63 { |
22 | struct { |
23 | #if defined(__LITTLE_ENDIAN) |
24 | u32 lo, hi; |
25 | #elif defined(__BIG_ENDIAN) |
26 | u32 hi, lo; |
27 | #endif |
28 | }; |
29 | u64 val; |
30 | }; |
31 | |
32 | |
33 | /** |
34 | * cnt32_to_63 - Expand a 32-bit counter to a 63-bit counter |
35 | * @cnt_lo: The low part of the counter |
36 | * |
37 | * Many hardware clock counters are only 32 bits wide and therefore have |
38 | * a relatively short period making wrap-arounds rather frequent. This |
39 | * is a problem when implementing sched_clock() for example, where a 64-bit |
40 | * non-wrapping monotonic value is expected to be returned. |
41 | * |
42 | * To overcome that limitation, let's extend a 32-bit counter to 63 bits |
43 | * in a completely lock free fashion. Bits 0 to 31 of the clock are provided |
44 | * by the hardware while bits 32 to 62 are stored in memory. The top bit in |
45 | * memory is used to synchronize with the hardware clock half-period. When |
46 | * the top bit of both counters (hardware and in memory) differ then the |
47 | * memory is updated with a new value, incrementing it when the hardware |
48 | * counter wraps around. |
49 | * |
50 | * Because a word store in memory is atomic then the incremented value will |
51 | * always be in synch with the top bit indicating to any potential concurrent |
52 | * reader if the value in memory is up to date or not with regards to the |
53 | * needed increment. And any race in updating the value in memory is harmless |
54 | * as the same value would simply be stored more than once. |
55 | * |
56 | * The restrictions for the algorithm to work properly are: |
57 | * |
58 | * 1) this code must be called at least once per each half period of the |
59 | * 32-bit counter; |
60 | * |
61 | * 2) this code must not be preempted for a duration longer than the |
62 | * 32-bit counter half period minus the longest period between two |
63 | * calls to this code; |
64 | * |
65 | * Those requirements ensure proper update to the state bit in memory. |
66 | * This is usually not a problem in practice, but if it is then a kernel |
67 | * timer should be scheduled to manage for this code to be executed often |
68 | * enough. |
69 | * |
70 | * And finally: |
71 | * |
72 | * 3) the cnt_lo argument must be seen as a globally incrementing value, |
73 | * meaning that it should be a direct reference to the counter data which |
74 | * can be evaluated according to a specific ordering within the macro, |
75 | * and not the result of a previous evaluation stored in a variable. |
76 | * |
77 | * For example, this is wrong: |
78 | * |
79 | * u32 partial = get_hw_count(); |
80 | * u64 full = cnt32_to_63(partial); |
81 | * return full; |
82 | * |
83 | * This is fine: |
84 | * |
85 | * u64 full = cnt32_to_63(get_hw_count()); |
86 | * return full; |
87 | * |
88 | * Note that the top bit (bit 63) in the returned value should be considered |
89 | * as garbage. It is not cleared here because callers are likely to use a |
90 | * multiplier on the returned value which can get rid of the top bit |
91 | * implicitly by making the multiplier even, therefore saving on a runtime |
92 | * clear-bit instruction. Otherwise caller must remember to clear the top |
93 | * bit explicitly. |
94 | */ |
95 | #define cnt32_to_63(cnt_lo) \ |
96 | ({ \ |
97 | static u32 __m_cnt_hi; \ |
98 | union cnt32_to_63 __x; \ |
99 | __x.hi = __m_cnt_hi; \ |
100 | smp_rmb(); \ |
101 | __x.lo = (cnt_lo); \ |
102 | if (unlikely((s32)(__x.hi ^ __x.lo) < 0)) \ |
103 | __m_cnt_hi = __x.hi = (__x.hi ^ 0x80000000) + (__x.hi >> 31); \ |
104 | __x.val; \ |
105 | }) |
106 | |
107 | #endif |
108 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9