diff options
Diffstat (limited to 'arch/sh/oprofile/op_model_sh7750.c')
-rw-r--r-- | arch/sh/oprofile/op_model_sh7750.c | 281 |
1 files changed, 281 insertions, 0 deletions
diff --git a/arch/sh/oprofile/op_model_sh7750.c b/arch/sh/oprofile/op_model_sh7750.c new file mode 100644 index 000000000000..5ec9ddcc4b0b --- /dev/null +++ b/arch/sh/oprofile/op_model_sh7750.c | |||
@@ -0,0 +1,281 @@ | |||
1 | /* | ||
2 | * arch/sh/oprofile/op_model_sh7750.c | ||
3 | * | ||
4 | * OProfile support for SH7750/SH7750S Performance Counters | ||
5 | * | ||
6 | * Copyright (C) 2003, 2004 Paul Mundt | ||
7 | * | ||
8 | * This file is subject to the terms and conditions of the GNU General Public | ||
9 | * License. See the file "COPYING" in the main directory of this archive | ||
10 | * for more details. | ||
11 | */ | ||
12 | #include <linux/kernel.h> | ||
13 | #include <linux/oprofile.h> | ||
14 | #include <linux/profile.h> | ||
15 | #include <linux/init.h> | ||
16 | #include <linux/errno.h> | ||
17 | #include <linux/interrupt.h> | ||
18 | #include <linux/fs.h> | ||
19 | #include <linux/notifier.h> | ||
20 | #include <asm/uaccess.h> | ||
21 | #include <asm/io.h> | ||
22 | |||
23 | #define PM_CR_BASE 0xff000084 /* 16-bit */ | ||
24 | #define PM_CTR_BASE 0xff100004 /* 32-bit */ | ||
25 | |||
26 | #define PMCR1 (PM_CR_BASE + 0x00) | ||
27 | #define PMCR2 (PM_CR_BASE + 0x04) | ||
28 | #define PMCTR1H (PM_CTR_BASE + 0x00) | ||
29 | #define PMCTR1L (PM_CTR_BASE + 0x04) | ||
30 | #define PMCTR2H (PM_CTR_BASE + 0x08) | ||
31 | #define PMCTR2L (PM_CTR_BASE + 0x0c) | ||
32 | |||
33 | #define PMCR_PMM_MASK 0x0000003f | ||
34 | |||
35 | #define PMCR_CLKF 0x00000100 | ||
36 | #define PMCR_PMCLR 0x00002000 | ||
37 | #define PMCR_PMST 0x00004000 | ||
38 | #define PMCR_PMEN 0x00008000 | ||
39 | |||
40 | #define PMCR_ENABLE (PMCR_PMST | PMCR_PMEN) | ||
41 | |||
42 | /* | ||
43 | * SH7750/SH7750S have 2 perf counters | ||
44 | */ | ||
45 | #define NR_CNTRS 2 | ||
46 | |||
47 | extern const char *get_cpu_subtype(void); | ||
48 | |||
49 | struct op_counter_config { | ||
50 | unsigned long enabled; | ||
51 | unsigned long event; | ||
52 | unsigned long count; | ||
53 | |||
54 | /* Dummy values for userspace tool compliance */ | ||
55 | unsigned long kernel; | ||
56 | unsigned long user; | ||
57 | unsigned long unit_mask; | ||
58 | }; | ||
59 | |||
60 | static struct op_counter_config ctr[NR_CNTRS]; | ||
61 | |||
62 | /* | ||
63 | * There are a number of events supported by each counter (33 in total). | ||
64 | * Since we have 2 counters, each counter will take the event code as it | ||
65 | * corresponds to the PMCR PMM setting. Each counter can be configured | ||
66 | * independently. | ||
67 | * | ||
68 | * Event Code Description | ||
69 | * ---------- ----------- | ||
70 | * | ||
71 | * 0x01 Operand read access | ||
72 | * 0x02 Operand write access | ||
73 | * 0x03 UTLB miss | ||
74 | * 0x04 Operand cache read miss | ||
75 | * 0x05 Operand cache write miss | ||
76 | * 0x06 Instruction fetch (w/ cache) | ||
77 | * 0x07 Instruction TLB miss | ||
78 | * 0x08 Instruction cache miss | ||
79 | * 0x09 All operand accesses | ||
80 | * 0x0a All instruction accesses | ||
81 | * 0x0b OC RAM operand access | ||
82 | * 0x0d On-chip I/O space access | ||
83 | * 0x0e Operand access (r/w) | ||
84 | * 0x0f Operand cache miss (r/w) | ||
85 | * 0x10 Branch instruction | ||
86 | * 0x11 Branch taken | ||
87 | * 0x12 BSR/BSRF/JSR | ||
88 | * 0x13 Instruction execution | ||
89 | * 0x14 Instruction execution in parallel | ||
90 | * 0x15 FPU Instruction execution | ||
91 | * 0x16 Interrupt | ||
92 | * 0x17 NMI | ||
93 | * 0x18 trapa instruction execution | ||
94 | * 0x19 UBCA match | ||
95 | * 0x1a UBCB match | ||
96 | * 0x21 Instruction cache fill | ||
97 | * 0x22 Operand cache fill | ||
98 | * 0x23 Elapsed time | ||
99 | * 0x24 Pipeline freeze by I-cache miss | ||
100 | * 0x25 Pipeline freeze by D-cache miss | ||
101 | * 0x27 Pipeline freeze by branch instruction | ||
102 | * 0x28 Pipeline freeze by CPU register | ||
103 | * 0x29 Pipeline freeze by FPU | ||
104 | * | ||
105 | * Unfortunately we don't have a native exception or interrupt for counter | ||
106 | * overflow (although since these counters can run for 16.3 days without | ||
107 | * overflowing, it's not really necessary). | ||
108 | * | ||
109 | * OProfile on the other hand likes to have samples taken periodically, so | ||
110 | * for now we just piggyback the timer interrupt to get the expected | ||
111 | * behavior. | ||
112 | */ | ||
113 | |||
114 | static int sh7750_timer_notify(struct notifier_block *self, | ||
115 | unsigned long val, void *regs) | ||
116 | { | ||
117 | oprofile_add_sample((struct pt_regs *)regs, 0); | ||
118 | return 0; | ||
119 | } | ||
120 | |||
121 | static struct notifier_block sh7750_timer_notifier = { | ||
122 | .notifier_call = sh7750_timer_notify, | ||
123 | }; | ||
124 | |||
125 | static u64 sh7750_read_counter(int counter) | ||
126 | { | ||
127 | u32 hi, lo; | ||
128 | |||
129 | hi = (counter == 0) ? ctrl_inl(PMCTR1H) : ctrl_inl(PMCTR2H); | ||
130 | lo = (counter == 0) ? ctrl_inl(PMCTR1L) : ctrl_inl(PMCTR2L); | ||
131 | |||
132 | return (u64)((u64)(hi & 0xffff) << 32) | lo; | ||
133 | } | ||
134 | |||
135 | /* | ||
136 | * Files will be in a path like: | ||
137 | * | ||
138 | * /<oprofilefs mount point>/<counter number>/<file> | ||
139 | * | ||
140 | * So when dealing with <file>, we look to the parent dentry for the counter | ||
141 | * number. | ||
142 | */ | ||
143 | static inline int to_counter(struct file *file) | ||
144 | { | ||
145 | const unsigned char *name = file->f_dentry->d_parent->d_name.name; | ||
146 | |||
147 | return (int)simple_strtol(name, NULL, 10); | ||
148 | } | ||
149 | |||
150 | /* | ||
151 | * XXX: We have 48-bit counters, so we're probably going to want something | ||
152 | * more along the lines of oprofilefs_ullong_to_user().. Truncating to | ||
153 | * unsigned long works fine for now though, as long as we don't attempt to | ||
154 | * profile for too horribly long. | ||
155 | */ | ||
156 | static ssize_t sh7750_read_count(struct file *file, char __user *buf, | ||
157 | size_t count, loff_t *ppos) | ||
158 | { | ||
159 | int counter = to_counter(file); | ||
160 | u64 val = sh7750_read_counter(counter); | ||
161 | |||
162 | return oprofilefs_ulong_to_user((unsigned long)val, buf, count, ppos); | ||
163 | } | ||
164 | |||
165 | static ssize_t sh7750_write_count(struct file *file, const char __user *buf, | ||
166 | size_t count, loff_t *ppos) | ||
167 | { | ||
168 | int counter = to_counter(file); | ||
169 | unsigned long val; | ||
170 | |||
171 | if (oprofilefs_ulong_from_user(&val, buf, count)) | ||
172 | return -EFAULT; | ||
173 | |||
174 | /* | ||
175 | * Any write will clear the counter, although only 0 should be | ||
176 | * written for this purpose, as we do not support setting the | ||
177 | * counter to an arbitrary value. | ||
178 | */ | ||
179 | WARN_ON(val != 0); | ||
180 | |||
181 | if (counter == 0) { | ||
182 | ctrl_outw(ctrl_inw(PMCR1) | PMCR_PMCLR, PMCR1); | ||
183 | } else { | ||
184 | ctrl_outw(ctrl_inw(PMCR2) | PMCR_PMCLR, PMCR2); | ||
185 | } | ||
186 | |||
187 | return count; | ||
188 | } | ||
189 | |||
190 | static struct file_operations count_fops = { | ||
191 | .read = sh7750_read_count, | ||
192 | .write = sh7750_write_count, | ||
193 | }; | ||
194 | |||
195 | static int sh7750_perf_counter_create_files(struct super_block *sb, struct dentry *root) | ||
196 | { | ||
197 | int i; | ||
198 | |||
199 | for (i = 0; i < NR_CNTRS; i++) { | ||
200 | struct dentry *dir; | ||
201 | char buf[3]; | ||
202 | |||
203 | snprintf(buf, sizeof(buf), "%d", i); | ||
204 | dir = oprofilefs_mkdir(sb, root, buf); | ||
205 | |||
206 | oprofilefs_create_ulong(sb, dir, "enabled", &ctr[i].enabled); | ||
207 | oprofilefs_create_ulong(sb, dir, "event", &ctr[i].event); | ||
208 | oprofilefs_create_file(sb, dir, "count", &count_fops); | ||
209 | |||
210 | /* Dummy entries */ | ||
211 | oprofilefs_create_ulong(sb, dir, "kernel", &ctr[i].kernel); | ||
212 | oprofilefs_create_ulong(sb, dir, "user", &ctr[i].user); | ||
213 | oprofilefs_create_ulong(sb, dir, "unit_mask", &ctr[i].unit_mask); | ||
214 | } | ||
215 | |||
216 | return 0; | ||
217 | } | ||
218 | |||
219 | static int sh7750_perf_counter_start(void) | ||
220 | { | ||
221 | u16 pmcr; | ||
222 | |||
223 | /* Enable counter 1 */ | ||
224 | if (ctr[0].enabled) { | ||
225 | pmcr = ctrl_inw(PMCR1); | ||
226 | WARN_ON(pmcr & PMCR_PMEN); | ||
227 | |||
228 | pmcr &= ~PMCR_PMM_MASK; | ||
229 | pmcr |= ctr[0].event; | ||
230 | ctrl_outw(pmcr | PMCR_ENABLE, PMCR1); | ||
231 | } | ||
232 | |||
233 | /* Enable counter 2 */ | ||
234 | if (ctr[1].enabled) { | ||
235 | pmcr = ctrl_inw(PMCR2); | ||
236 | WARN_ON(pmcr & PMCR_PMEN); | ||
237 | |||
238 | pmcr &= ~PMCR_PMM_MASK; | ||
239 | pmcr |= ctr[1].event; | ||
240 | ctrl_outw(pmcr | PMCR_ENABLE, PMCR2); | ||
241 | } | ||
242 | |||
243 | return register_profile_notifier(&sh7750_timer_notifier); | ||
244 | } | ||
245 | |||
246 | static void sh7750_perf_counter_stop(void) | ||
247 | { | ||
248 | ctrl_outw(ctrl_inw(PMCR1) & ~PMCR_PMEN, PMCR1); | ||
249 | ctrl_outw(ctrl_inw(PMCR2) & ~PMCR_PMEN, PMCR2); | ||
250 | |||
251 | unregister_profile_notifier(&sh7750_timer_notifier); | ||
252 | } | ||
253 | |||
254 | static struct oprofile_operations sh7750_perf_counter_ops = { | ||
255 | .create_files = sh7750_perf_counter_create_files, | ||
256 | .start = sh7750_perf_counter_start, | ||
257 | .stop = sh7750_perf_counter_stop, | ||
258 | }; | ||
259 | |||
260 | int __init oprofile_arch_init(struct oprofile_operations **ops) | ||
261 | { | ||
262 | if (!(cpu_data->flags & CPU_HAS_PERF_COUNTER)) | ||
263 | return -ENODEV; | ||
264 | |||
265 | sh7750_perf_counter_ops.cpu_type = (char *)get_cpu_subtype(); | ||
266 | *ops = &sh7750_perf_counter_ops; | ||
267 | |||
268 | printk(KERN_INFO "oprofile: using SH-4 (%s) performance monitoring.\n", | ||
269 | sh7750_perf_counter_ops.cpu_type); | ||
270 | |||
271 | /* Clear the counters */ | ||
272 | ctrl_outw(ctrl_inw(PMCR1) | PMCR_PMCLR, PMCR1); | ||
273 | ctrl_outw(ctrl_inw(PMCR2) | PMCR_PMCLR, PMCR2); | ||
274 | |||
275 | return 0; | ||
276 | } | ||
277 | |||
278 | void oprofile_arch_exit(void) | ||
279 | { | ||
280 | } | ||
281 | |||