summaryrefslogtreecommitdiff
path: root/arch/powerpc/oprofile/cell/spu_profiler.c
blob: 4e099e556645ad082de4c24e58ef07e108bf0947 (plain)
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
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
/*
 * Cell Broadband Engine OProfile Support
 *
 * (C) Copyright IBM Corporation 2006
 *
 * Authors: Maynard Johnson <maynardj@us.ibm.com>
 *	    Carl Love <carll@us.ibm.com>
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * as published by the Free Software Foundation; either version
 * 2 of the License, or (at your option) any later version.
 */

#include <linux/hrtimer.h>
#include <linux/smp.h>
#include <linux/slab.h>
#include <asm/cell-pmu.h>
#include <asm/time.h>
#include "pr_util.h"

#define SCALE_SHIFT 14

static u32 *samples;

/* spu_prof_running is a flag used to indicate if spu profiling is enabled
 * or not.  It is set by the routines start_spu_profiling_cycles() and
 * start_spu_profiling_events().  The flag is cleared by the routines
 * stop_spu_profiling_cycles() and stop_spu_profiling_events().  These
 * routines are called via global_start() and global_stop() which are called in
 * op_powerpc_start() and op_powerpc_stop().  These routines are called once
 * per system as a result of the user starting/stopping oprofile.  Hence, only
 * one CPU per user at a time will be changing  the value of spu_prof_running.
 * In general, OProfile does not protect against multiple users trying to run
 * OProfile at a time.
 */
int spu_prof_running;
static unsigned int profiling_interval;

#define NUM_SPU_BITS_TRBUF 16
#define SPUS_PER_TB_ENTRY   4

#define SPU_PC_MASK	     0xFFFF

DEFINE_SPINLOCK(oprof_spu_smpl_arry_lck);
static unsigned long oprof_spu_smpl_arry_lck_flags;

void set_spu_profiling_frequency(unsigned int freq_khz, unsigned int cycles_reset)
{
	unsigned long ns_per_cyc;

	if (!freq_khz)
		freq_khz = ppc_proc_freq/1000;

	/* To calculate a timeout in nanoseconds, the basic
	 * formula is ns = cycles_reset * (NSEC_PER_SEC / cpu frequency).
	 * To avoid floating point math, we use the scale math
	 * technique as described in linux/jiffies.h.  We use
	 * a scale factor of SCALE_SHIFT, which provides 4 decimal places
	 * of precision.  This is close enough for the purpose at hand.
	 *
	 * The value of the timeout should be small enough that the hw
	 * trace buffer will not get more than about 1/3 full for the
	 * maximum user specified (the LFSR value) hw sampling frequency.
	 * This is to ensure the trace buffer will never fill even if the
	 * kernel thread scheduling varies under a heavy system load.
	 */

	ns_per_cyc = (USEC_PER_SEC << SCALE_SHIFT)/freq_khz;
	profiling_interval = (ns_per_cyc * cycles_reset) >> SCALE_SHIFT;

}

/*
 * Extract SPU PC from trace buffer entry
 */
static void spu_pc_extract(int cpu, int entry)
{
	/* the trace buffer is 128 bits */
	u64 trace_buffer[2];
	u64 spu_mask;
	int spu;

	spu_mask = SPU_PC_MASK;

	/* Each SPU PC is 16 bits; hence, four spus in each of
	 * the two 64-bit buffer entries that make up the
	 * 128-bit trace_buffer entry.	Process two 64-bit values
	 * simultaneously.
	 * trace[0] SPU PC contents are: 0 1 2 3
	 * trace[1] SPU PC contents are: 4 5 6 7
	 */

	cbe_read_trace_buffer(cpu, trace_buffer);

	for (spu = SPUS_PER_TB_ENTRY-1; spu >= 0; spu--) {
		/* spu PC trace entry is upper 16 bits of the
		 * 18 bit SPU program counter
		 */
		samples[spu * TRACE_ARRAY_SIZE + entry]
			= (spu_mask & trace_buffer[0]) << 2;
		samples[(spu + SPUS_PER_TB_ENTRY) * TRACE_ARRAY_SIZE + entry]
			= (spu_mask & trace_buffer[1]) << 2;

		trace_buffer[0] = trace_buffer[0] >> NUM_SPU_BITS_TRBUF;
		trace_buffer[1] = trace_buffer[1] >> NUM_SPU_BITS_TRBUF;
	}
}

static int cell_spu_pc_collection(int cpu)
{
	u32 trace_addr;
	int entry;

	/* process the collected SPU PC for the node */

	entry = 0;

	trace_addr = cbe_read_pm(cpu, trace_address);
	while (!(trace_addr & CBE_PM_TRACE_BUF_EMPTY)) {
		/* there is data in the trace buffer to process */
		spu_pc_extract(cpu, entry);

		entry++;

		if (entry >= TRACE_ARRAY_SIZE)
			/* spu_samples is full */
			break;

		trace_addr = cbe_read_pm(cpu, trace_address);
	}

	return entry;
}


static enum hrtimer_restart profile_spus(struct hrtimer *timer)
{
	ktime_t kt;
	int cpu, node, k, num_samples, spu_num;

	if (!spu_prof_running)
		goto stop;

	for_each_online_cpu(cpu) {
		if (cbe_get_hw_thread_id(cpu))
			continue;

		node = cbe_cpu_to_node(cpu);

		/* There should only be one kernel thread at a time processing
		 * the samples.	 In the very unlikely case that the processing
		 * is taking a very long time and multiple kernel threads are
		 * started to process the samples.  Make sure only one kernel
		 * thread is working on the samples array at a time.  The
		 * sample array must be loaded and then processed for a given
		 * cpu.	 The sample array is not per cpu.
		 */
		spin_lock_irqsave(&oprof_spu_smpl_arry_lck,
				  oprof_spu_smpl_arry_lck_flags);
		num_samples = cell_spu_pc_collection(cpu);

		if (num_samples == 0) {
			spin_unlock_irqrestore(&oprof_spu_smpl_arry_lck,
					       oprof_spu_smpl_arry_lck_flags);
			continue;
		}

		for (k = 0; k < SPUS_PER_NODE; k++) {
			spu_num = k + (node * SPUS_PER_NODE);
			spu_sync_buffer(spu_num,
					samples + (k * TRACE_ARRAY_SIZE),
					num_samples);
		}

		spin_unlock_irqrestore(&oprof_spu_smpl_arry_lck,
				       oprof_spu_smpl_arry_lck_flags);

	}
	smp_wmb();	/* insure spu event buffer updates are written */
			/* don't want events intermingled... */

	kt = profiling_interval;
	if (!spu_prof_running)
		goto stop;
	hrtimer_forward(timer, timer->base->get_time(), kt);
	return HRTIMER_RESTART;

 stop:
	printk(KERN_INFO "SPU_PROF: spu-prof timer ending\n");
	return HRTIMER_NORESTART;
}

static struct hrtimer timer;
/*
 * Entry point for SPU cycle profiling.
 * NOTE:  SPU profiling is done system-wide, not per-CPU.
 *
 * cycles_reset is the count value specified by the user when
 * setting up OProfile to count SPU_CYCLES.
 */
int start_spu_profiling_cycles(unsigned int cycles_reset)
{
	ktime_t kt;

	pr_debug("timer resolution: %lu\n", TICK_NSEC);
	kt = profiling_interval;
	hrtimer_init(&timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	hrtimer_set_expires(&timer, kt);
	timer.function = profile_spus;

	/* Allocate arrays for collecting SPU PC samples */
	samples = kcalloc(SPUS_PER_NODE * TRACE_ARRAY_SIZE, sizeof(u32),
			  GFP_KERNEL);

	if (!samples)
		return -ENOMEM;

	spu_prof_running = 1;
	hrtimer_start(&timer, kt, HRTIMER_MODE_REL);
	schedule_delayed_work(&spu_work, DEFAULT_TIMER_EXPIRE);

	return 0;
}

/*
 * Entry point for SPU event profiling.
 * NOTE:  SPU profiling is done system-wide, not per-CPU.
 *
 * cycles_reset is the count value specified by the user when
 * setting up OProfile to count SPU_CYCLES.
 */
void start_spu_profiling_events(void)
{
	spu_prof_running = 1;
	schedule_delayed_work(&spu_work, DEFAULT_TIMER_EXPIRE);

	return;
}

void stop_spu_profiling_cycles(void)
{
	spu_prof_running = 0;
	hrtimer_cancel(&timer);
	kfree(samples);
	pr_debug("SPU_PROF: stop_spu_profiling_cycles issued\n");
}

void stop_spu_profiling_events(void)
{
	spu_prof_running = 0;
}