1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
|
// IBM_PROLOG_BEGIN_TAG
// This is an automatically generated prolog.
//
// $Source: src/lib/sync.C $
//
// IBM CONFIDENTIAL
//
// COPYRIGHT International Business Machines Corp. 2011
//
// p1
//
// Object Code Only (OCO) source materials
// Licensed Internal Code Source Materials
// IBM HostBoot Licensed Internal Code
//
// The source code for this program is not published or other-
// wise divested of its trade secrets, irrespective of what has
// been deposited with the U.S. Copyright Office.
//
// Origin: 30
//
// IBM_PROLOG_END
#include <arch/ppc.H>
#include <sys/sync.h>
#include <sys/syscall.h>
#include <assert.h>
using namespace Systemcalls;
//-----------------------------------------------------------------------------
uint64_t futex_wait(uint64_t * i_addr, uint64_t i_val)
{
return (uint64_t) _syscall2(FUTEX_WAIT,i_addr, (void *)i_val);
}
//-----------------------------------------------------------------------------
uint64_t futex_wake(uint64_t * i_addr, uint64_t i_count)
{
return (uint64_t) _syscall2(FUTEX_WAKE, i_addr, (void *)i_count);
}
//-----------------------------------------------------------------------------
void barrier_init (barrier_t * o_barrier, uint64_t i_count)
{
mutex_init(&(o_barrier->iv_mutex));
o_barrier->iv_event = 0;
o_barrier->iv_count = i_count;
o_barrier->iv_missing = i_count;
return;
}
//-----------------------------------------------------------------------------
void barrier_destroy (barrier_t * i_barrier)
{
crit_assert(i_barrier->iv_missing == i_barrier->iv_count);
return;
}
//-----------------------------------------------------------------------------
void barrier_wait (barrier_t * i_barrier)
{
mutex_lock(&(i_barrier->iv_mutex));
--(i_barrier->iv_missing);
if(i_barrier->iv_missing > 0)
{
uint64_t l_event = i_barrier->iv_event;
mutex_unlock(&(i_barrier->iv_mutex));
do
{
futex_wait(&(i_barrier->iv_event), l_event);
} while (i_barrier->iv_event == l_event);
}
else
{
++(i_barrier->iv_event);
i_barrier->iv_missing = i_barrier->iv_count;
// Wake em all up
futex_wake(&(i_barrier->iv_event), UINT64_MAX);
mutex_unlock(&(i_barrier->iv_mutex));
}
return;
}
//-----------------------------------------------------------------------------
void mutex_init(mutex_t * o_mutex)
{
o_mutex->iv_val = 0;
return;
}
//-----------------------------------------------------------------------------
void mutex_destroy(mutex_t * i_mutex)
{
i_mutex->iv_val = ~0;
return;
}
//-----------------------------------------------------------------------------
void mutex_lock(mutex_t * i_mutex)
{
// Weak consistency notes:
// Since this is a lock, we do not need to ensure that all writes
// are globally visible prior to execution (lwsync) but we do need
// to ensure that all instructions finish completion prior to
// leaving (isync). Both __sync_val_compare_and_swap and
// __sync_lock_test_and_set have an implied isync.
uint64_t l_count = __sync_val_compare_and_swap(&(i_mutex->iv_val),0,1);
if(unlikely(l_count != 0))
{
if (likely(l_count != 2))
{
l_count = __sync_lock_test_and_set(&(i_mutex->iv_val), 2);
}
while( l_count != 0 )
{
futex_wait( &(i_mutex->iv_val), 2);
l_count = __sync_lock_test_and_set(&(i_mutex->iv_val),2);
// if more than one thread gets out - one continues while
// the rest get blocked again.
}
}
return;
}
//-----------------------------------------------------------------------------
void mutex_unlock(mutex_t * i_mutex)
{
// Weak consistency notes:
// Since this is an unlock we need to ensure that all writes are
// globally visible prior to execution (lwsync). The
// __sync_fetch_and_sub has an implied lwsync. If we need to
// release a task, due to a contended mutex, the write to iv_val
// and futex_wake pair will appear globally ordered due to the
// context synchronizing nature of the 'sc' instruction.
uint64_t l_count = __sync_fetch_and_sub(&(i_mutex->iv_val),1);
if(unlikely(2 <= l_count))
{
i_mutex->iv_val = 0;
futex_wake(&(i_mutex->iv_val), 1); // wake one thread
}
return;
}
|
