blob: 2242bdd4370eb19851ff5a31540a02c65fa26498 (
plain) (
blame)
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
|
/*
* Count register synchronisation.
*
* All CPUs will have their count registers synchronised to the CPU0 next time
* value. This can cause a small timewarp for CPU0. All other CPU's should
* not have done anything significant (but they may have had interrupts
* enabled briefly - prom_smp_finish() should not be responsible for enabling
* interrupts...)
*/
#include <linux/kernel.h>
#include <linux/irqflags.h>
#include <linux/cpumask.h>
#include <asm/r4k-timer.h>
#include <linux/atomic.h>
#include <asm/barrier.h>
#include <asm/mipsregs.h>
static atomic_t count_start_flag = ATOMIC_INIT(0);
static atomic_t count_count_start = ATOMIC_INIT(0);
static atomic_t count_count_stop = ATOMIC_INIT(0);
static atomic_t count_reference = ATOMIC_INIT(0);
#define COUNTON 100
#define NR_LOOPS 5
void synchronise_count_master(int cpu)
{
int i;
unsigned long flags;
unsigned int initcount;
printk(KERN_INFO "Synchronize counters for CPU %u: ", cpu);
local_irq_save(flags);
/*
* Notify the slaves that it's time to start
*/
atomic_set(&count_reference, read_c0_count());
atomic_set(&count_start_flag, cpu);
smp_wmb();
/* Count will be initialised to current timer for all CPU's */
initcount = read_c0_count();
/*
* We loop a few times to get a primed instruction cache,
* then the last pass is more or less synchronised and
* the master and slaves each set their cycle counters to a known
* value all at once. This reduces the chance of having random offsets
* between the processors, and guarantees that the maximum
* delay between the cycle counters is never bigger than
* the latency of information-passing (cachelines) between
* two CPUs.
*/
for (i = 0; i < NR_LOOPS; i++) {
/* slaves loop on '!= 2' */
while (atomic_read(&count_count_start) != 1)
mb();
atomic_set(&count_count_stop, 0);
smp_wmb();
/* this lets the slaves write their count register */
atomic_inc(&count_count_start);
/*
* Everyone initialises count in the last loop:
*/
if (i == NR_LOOPS-1)
write_c0_count(initcount);
/*
* Wait for all slaves to leave the synchronization point:
*/
while (atomic_read(&count_count_stop) != 1)
mb();
atomic_set(&count_count_start, 0);
smp_wmb();
atomic_inc(&count_count_stop);
}
/* Arrange for an interrupt in a short while */
write_c0_compare(read_c0_count() + COUNTON);
atomic_set(&count_start_flag, 0);
local_irq_restore(flags);
/*
* i386 code reported the skew here, but the
* count registers were almost certainly out of sync
* so no point in alarming people
*/
printk("done.\n");
}
void synchronise_count_slave(int cpu)
{
int i;
unsigned int initcount;
/*
* Not every cpu is online at the time this gets called,
* so we first wait for the master to say everyone is ready
*/
while (atomic_read(&count_start_flag) != cpu)
mb();
/* Count will be initialised to next expire for all CPU's */
initcount = atomic_read(&count_reference);
for (i = 0; i < NR_LOOPS; i++) {
atomic_inc(&count_count_start);
while (atomic_read(&count_count_start) != 2)
mb();
/*
* Everyone initialises count in the last loop:
*/
if (i == NR_LOOPS-1)
write_c0_count(initcount);
atomic_inc(&count_count_stop);
while (atomic_read(&count_count_stop) != 2)
mb();
}
/* Arrange for an interrupt in a short while */
write_c0_compare(read_c0_count() + COUNTON);
}
#undef NR_LOOPS
|
