Commit | Line | Data |
---|---|---|
54176736 RB |
1 | /* |
2 | * This file is subject to the terms and conditions of the GNU General Public | |
3 | * License. See the file "COPYING" in the main directory of this archive | |
4 | * for more details. | |
5 | * | |
937a8015 | 6 | * Copyright (C) 2004, 05, 06 by Ralf Baechle |
54176736 RB |
7 | * Copyright (C) 2005 by MIPS Technologies, Inc. |
8 | */ | |
5e2862eb | 9 | #include <linux/cpumask.h> |
54176736 RB |
10 | #include <linux/oprofile.h> |
11 | #include <linux/interrupt.h> | |
12 | #include <linux/smp.h> | |
937a8015 | 13 | #include <asm/irq_regs.h> |
54176736 RB |
14 | |
15 | #include "op_impl.h" | |
16 | ||
92c7b62f RB |
17 | #define M_PERFCTL_EXL (1UL << 0) |
18 | #define M_PERFCTL_KERNEL (1UL << 1) | |
19 | #define M_PERFCTL_SUPERVISOR (1UL << 2) | |
20 | #define M_PERFCTL_USER (1UL << 3) | |
21 | #define M_PERFCTL_INTERRUPT_ENABLE (1UL << 4) | |
39a51109 | 22 | #define M_PERFCTL_EVENT(event) (((event) & 0x3ff) << 5) |
92c7b62f RB |
23 | #define M_PERFCTL_VPEID(vpe) ((vpe) << 16) |
24 | #define M_PERFCTL_MT_EN(filter) ((filter) << 20) | |
25 | #define M_TC_EN_ALL M_PERFCTL_MT_EN(0) | |
26 | #define M_TC_EN_VPE M_PERFCTL_MT_EN(1) | |
27 | #define M_TC_EN_TC M_PERFCTL_MT_EN(2) | |
28 | #define M_PERFCTL_TCID(tcid) ((tcid) << 22) | |
29 | #define M_PERFCTL_WIDE (1UL << 30) | |
30 | #define M_PERFCTL_MORE (1UL << 31) | |
31 | ||
32 | #define M_COUNTER_OVERFLOW (1UL << 31) | |
33 | ||
46684734 DV |
34 | static int (*save_perf_irq)(void); |
35 | ||
92c7b62f | 36 | #ifdef CONFIG_MIPS_MT_SMP |
39b8d525 RB |
37 | static int cpu_has_mipsmt_pertccounters; |
38 | #define WHAT (M_TC_EN_VPE | \ | |
39 | M_PERFCTL_VPEID(cpu_data[smp_processor_id()].vpe_id)) | |
40 | #define vpe_id() (cpu_has_mipsmt_pertccounters ? \ | |
41 | 0 : cpu_data[smp_processor_id()].vpe_id) | |
5e2862eb RB |
42 | |
43 | /* | |
44 | * The number of bits to shift to convert between counters per core and | |
45 | * counters per VPE. There is no reasonable interface atm to obtain the | |
46 | * number of VPEs used by Linux and in the 34K this number is fixed to two | |
47 | * anyways so we hardcore a few things here for the moment. The way it's | |
48 | * done here will ensure that oprofile VSMP kernel will run right on a lesser | |
49 | * core like a 24K also or with maxcpus=1. | |
50 | */ | |
51 | static inline unsigned int vpe_shift(void) | |
52 | { | |
53 | if (num_possible_cpus() > 1) | |
54 | return 1; | |
55 | ||
56 | return 0; | |
57 | } | |
58 | ||
92c7b62f | 59 | #else |
5e2862eb | 60 | |
be609f35 | 61 | #define WHAT 0 |
6f4c5bde | 62 | #define vpe_id() 0 |
5e2862eb RB |
63 | |
64 | static inline unsigned int vpe_shift(void) | |
65 | { | |
66 | return 0; | |
67 | } | |
68 | ||
92c7b62f | 69 | #endif |
54176736 | 70 | |
5e2862eb RB |
71 | static inline unsigned int counters_total_to_per_cpu(unsigned int counters) |
72 | { | |
73 | return counters >> vpe_shift(); | |
74 | } | |
75 | ||
76 | static inline unsigned int counters_per_cpu_to_total(unsigned int counters) | |
77 | { | |
78 | return counters << vpe_shift(); | |
79 | } | |
80 | ||
92c7b62f RB |
81 | #define __define_perf_accessors(r, n, np) \ |
82 | \ | |
83 | static inline unsigned int r_c0_ ## r ## n(void) \ | |
84 | { \ | |
be609f35 | 85 | unsigned int cpu = vpe_id(); \ |
92c7b62f RB |
86 | \ |
87 | switch (cpu) { \ | |
88 | case 0: \ | |
89 | return read_c0_ ## r ## n(); \ | |
90 | case 1: \ | |
91 | return read_c0_ ## r ## np(); \ | |
92 | default: \ | |
93 | BUG(); \ | |
94 | } \ | |
30f244ae | 95 | return 0; \ |
92c7b62f RB |
96 | } \ |
97 | \ | |
98 | static inline void w_c0_ ## r ## n(unsigned int value) \ | |
99 | { \ | |
be609f35 | 100 | unsigned int cpu = vpe_id(); \ |
92c7b62f RB |
101 | \ |
102 | switch (cpu) { \ | |
103 | case 0: \ | |
104 | write_c0_ ## r ## n(value); \ | |
105 | return; \ | |
106 | case 1: \ | |
107 | write_c0_ ## r ## np(value); \ | |
108 | return; \ | |
109 | default: \ | |
110 | BUG(); \ | |
111 | } \ | |
30f244ae | 112 | return; \ |
92c7b62f RB |
113 | } \ |
114 | ||
115 | __define_perf_accessors(perfcntr, 0, 2) | |
116 | __define_perf_accessors(perfcntr, 1, 3) | |
795a2258 CD |
117 | __define_perf_accessors(perfcntr, 2, 0) |
118 | __define_perf_accessors(perfcntr, 3, 1) | |
92c7b62f RB |
119 | |
120 | __define_perf_accessors(perfctrl, 0, 2) | |
121 | __define_perf_accessors(perfctrl, 1, 3) | |
795a2258 CD |
122 | __define_perf_accessors(perfctrl, 2, 0) |
123 | __define_perf_accessors(perfctrl, 3, 1) | |
54176736 | 124 | |
1acf1ca7 | 125 | struct op_mips_model op_model_mipsxx_ops; |
54176736 RB |
126 | |
127 | static struct mipsxx_register_config { | |
128 | unsigned int control[4]; | |
129 | unsigned int counter[4]; | |
130 | } reg; | |
131 | ||
132 | /* Compute all of the registers in preparation for enabling profiling. */ | |
133 | ||
134 | static void mipsxx_reg_setup(struct op_counter_config *ctr) | |
135 | { | |
1acf1ca7 | 136 | unsigned int counters = op_model_mipsxx_ops.num_counters; |
54176736 RB |
137 | int i; |
138 | ||
139 | /* Compute the performance counter control word. */ | |
54176736 RB |
140 | for (i = 0; i < counters; i++) { |
141 | reg.control[i] = 0; | |
142 | reg.counter[i] = 0; | |
143 | ||
144 | if (!ctr[i].enabled) | |
145 | continue; | |
146 | ||
147 | reg.control[i] = M_PERFCTL_EVENT(ctr[i].event) | | |
148 | M_PERFCTL_INTERRUPT_ENABLE; | |
149 | if (ctr[i].kernel) | |
150 | reg.control[i] |= M_PERFCTL_KERNEL; | |
151 | if (ctr[i].user) | |
152 | reg.control[i] |= M_PERFCTL_USER; | |
153 | if (ctr[i].exl) | |
154 | reg.control[i] |= M_PERFCTL_EXL; | |
155 | reg.counter[i] = 0x80000000 - ctr[i].count; | |
156 | } | |
157 | } | |
158 | ||
159 | /* Program all of the registers in preparation for enabling profiling. */ | |
160 | ||
49a89efb | 161 | static void mipsxx_cpu_setup(void *args) |
54176736 | 162 | { |
1acf1ca7 | 163 | unsigned int counters = op_model_mipsxx_ops.num_counters; |
54176736 RB |
164 | |
165 | switch (counters) { | |
166 | case 4: | |
92c7b62f RB |
167 | w_c0_perfctrl3(0); |
168 | w_c0_perfcntr3(reg.counter[3]); | |
54176736 | 169 | case 3: |
92c7b62f RB |
170 | w_c0_perfctrl2(0); |
171 | w_c0_perfcntr2(reg.counter[2]); | |
54176736 | 172 | case 2: |
92c7b62f RB |
173 | w_c0_perfctrl1(0); |
174 | w_c0_perfcntr1(reg.counter[1]); | |
54176736 | 175 | case 1: |
92c7b62f RB |
176 | w_c0_perfctrl0(0); |
177 | w_c0_perfcntr0(reg.counter[0]); | |
54176736 RB |
178 | } |
179 | } | |
180 | ||
181 | /* Start all counters on current CPU */ | |
182 | static void mipsxx_cpu_start(void *args) | |
183 | { | |
1acf1ca7 | 184 | unsigned int counters = op_model_mipsxx_ops.num_counters; |
54176736 RB |
185 | |
186 | switch (counters) { | |
187 | case 4: | |
92c7b62f | 188 | w_c0_perfctrl3(WHAT | reg.control[3]); |
54176736 | 189 | case 3: |
92c7b62f | 190 | w_c0_perfctrl2(WHAT | reg.control[2]); |
54176736 | 191 | case 2: |
92c7b62f | 192 | w_c0_perfctrl1(WHAT | reg.control[1]); |
54176736 | 193 | case 1: |
92c7b62f | 194 | w_c0_perfctrl0(WHAT | reg.control[0]); |
54176736 RB |
195 | } |
196 | } | |
197 | ||
198 | /* Stop all counters on current CPU */ | |
199 | static void mipsxx_cpu_stop(void *args) | |
200 | { | |
1acf1ca7 | 201 | unsigned int counters = op_model_mipsxx_ops.num_counters; |
54176736 RB |
202 | |
203 | switch (counters) { | |
204 | case 4: | |
92c7b62f | 205 | w_c0_perfctrl3(0); |
54176736 | 206 | case 3: |
92c7b62f | 207 | w_c0_perfctrl2(0); |
54176736 | 208 | case 2: |
92c7b62f | 209 | w_c0_perfctrl1(0); |
54176736 | 210 | case 1: |
92c7b62f | 211 | w_c0_perfctrl0(0); |
54176736 RB |
212 | } |
213 | } | |
214 | ||
937a8015 | 215 | static int mipsxx_perfcount_handler(void) |
54176736 | 216 | { |
1acf1ca7 | 217 | unsigned int counters = op_model_mipsxx_ops.num_counters; |
54176736 RB |
218 | unsigned int control; |
219 | unsigned int counter; | |
ffe9ee47 CD |
220 | int handled = IRQ_NONE; |
221 | ||
222 | if (cpu_has_mips_r2 && !(read_c0_cause() & (1 << 26))) | |
223 | return handled; | |
54176736 RB |
224 | |
225 | switch (counters) { | |
226 | #define HANDLE_COUNTER(n) \ | |
227 | case n + 1: \ | |
92c7b62f RB |
228 | control = r_c0_perfctrl ## n(); \ |
229 | counter = r_c0_perfcntr ## n(); \ | |
54176736 RB |
230 | if ((control & M_PERFCTL_INTERRUPT_ENABLE) && \ |
231 | (counter & M_COUNTER_OVERFLOW)) { \ | |
937a8015 | 232 | oprofile_add_sample(get_irq_regs(), n); \ |
92c7b62f | 233 | w_c0_perfcntr ## n(reg.counter[n]); \ |
ffe9ee47 | 234 | handled = IRQ_HANDLED; \ |
54176736 RB |
235 | } |
236 | HANDLE_COUNTER(3) | |
237 | HANDLE_COUNTER(2) | |
238 | HANDLE_COUNTER(1) | |
239 | HANDLE_COUNTER(0) | |
240 | } | |
ba339c03 RB |
241 | |
242 | return handled; | |
54176736 RB |
243 | } |
244 | ||
245 | #define M_CONFIG1_PC (1 << 4) | |
246 | ||
92c7b62f | 247 | static inline int __n_counters(void) |
54176736 RB |
248 | { |
249 | if (!(read_c0_config1() & M_CONFIG1_PC)) | |
250 | return 0; | |
39b8d525 | 251 | if (!(read_c0_perfctrl0() & M_PERFCTL_MORE)) |
54176736 | 252 | return 1; |
39b8d525 | 253 | if (!(read_c0_perfctrl1() & M_PERFCTL_MORE)) |
54176736 | 254 | return 2; |
39b8d525 | 255 | if (!(read_c0_perfctrl2() & M_PERFCTL_MORE)) |
54176736 RB |
256 | return 3; |
257 | ||
258 | return 4; | |
259 | } | |
260 | ||
92c7b62f RB |
261 | static inline int n_counters(void) |
262 | { | |
714cfe78 RB |
263 | int counters; |
264 | ||
10cc3529 | 265 | switch (current_cpu_type()) { |
714cfe78 RB |
266 | case CPU_R10000: |
267 | counters = 2; | |
148171b2 | 268 | break; |
714cfe78 RB |
269 | |
270 | case CPU_R12000: | |
271 | case CPU_R14000: | |
272 | counters = 4; | |
148171b2 | 273 | break; |
714cfe78 RB |
274 | |
275 | default: | |
276 | counters = __n_counters(); | |
277 | } | |
92c7b62f | 278 | |
92c7b62f RB |
279 | return counters; |
280 | } | |
281 | ||
39b8d525 | 282 | static void reset_counters(void *arg) |
54176736 | 283 | { |
005ca9a3 | 284 | int counters = (int)(long)arg; |
54176736 RB |
285 | switch (counters) { |
286 | case 4: | |
92c7b62f RB |
287 | w_c0_perfctrl3(0); |
288 | w_c0_perfcntr3(0); | |
54176736 | 289 | case 3: |
92c7b62f RB |
290 | w_c0_perfctrl2(0); |
291 | w_c0_perfcntr2(0); | |
54176736 | 292 | case 2: |
92c7b62f RB |
293 | w_c0_perfctrl1(0); |
294 | w_c0_perfcntr1(0); | |
54176736 | 295 | case 1: |
92c7b62f RB |
296 | w_c0_perfctrl0(0); |
297 | w_c0_perfcntr0(0); | |
54176736 RB |
298 | } |
299 | } | |
300 | ||
3572a2c3 FF |
301 | static irqreturn_t mipsxx_perfcount_int(int irq, void *dev_id) |
302 | { | |
303 | return mipsxx_perfcount_handler(); | |
304 | } | |
305 | ||
54176736 RB |
306 | static int __init mipsxx_init(void) |
307 | { | |
308 | int counters; | |
309 | ||
310 | counters = n_counters(); | |
9efeae9a RB |
311 | if (counters == 0) { |
312 | printk(KERN_ERR "Oprofile: CPU has no performance counters\n"); | |
54176736 | 313 | return -ENODEV; |
9efeae9a | 314 | } |
54176736 | 315 | |
39b8d525 RB |
316 | #ifdef CONFIG_MIPS_MT_SMP |
317 | cpu_has_mipsmt_pertccounters = read_c0_config7() & (1<<19); | |
318 | if (!cpu_has_mipsmt_pertccounters) | |
319 | counters = counters_total_to_per_cpu(counters); | |
320 | #endif | |
f6f88e9b | 321 | on_each_cpu(reset_counters, (void *)(long)counters, 1); |
795a2258 | 322 | |
1acf1ca7 | 323 | op_model_mipsxx_ops.num_counters = counters; |
10cc3529 | 324 | switch (current_cpu_type()) { |
113c62d9 SH |
325 | case CPU_M14KC: |
326 | op_model_mipsxx_ops.cpu_type = "mips/M14Kc"; | |
327 | break; | |
328 | ||
2065988e | 329 | case CPU_20KC: |
1acf1ca7 | 330 | op_model_mipsxx_ops.cpu_type = "mips/20K"; |
2065988e RB |
331 | break; |
332 | ||
54176736 | 333 | case CPU_24K: |
1acf1ca7 | 334 | op_model_mipsxx_ops.cpu_type = "mips/24K"; |
54176736 RB |
335 | break; |
336 | ||
2065988e | 337 | case CPU_25KF: |
1acf1ca7 | 338 | op_model_mipsxx_ops.cpu_type = "mips/25K"; |
2065988e RB |
339 | break; |
340 | ||
39b8d525 RB |
341 | case CPU_1004K: |
342 | #if 0 | |
343 | /* FIXME: report as 34K for now */ | |
344 | op_model_mipsxx_ops.cpu_type = "mips/1004K"; | |
345 | break; | |
346 | #endif | |
347 | ||
fcfd980c | 348 | case CPU_34K: |
1acf1ca7 | 349 | op_model_mipsxx_ops.cpu_type = "mips/34K"; |
fcfd980c | 350 | break; |
c620953c CD |
351 | |
352 | case CPU_74K: | |
1acf1ca7 | 353 | op_model_mipsxx_ops.cpu_type = "mips/74K"; |
c620953c | 354 | break; |
fcfd980c | 355 | |
2065988e | 356 | case CPU_5KC: |
1acf1ca7 | 357 | op_model_mipsxx_ops.cpu_type = "mips/5K"; |
2065988e RB |
358 | break; |
359 | ||
714cfe78 RB |
360 | case CPU_R10000: |
361 | if ((current_cpu_data.processor_id & 0xff) == 0x20) | |
362 | op_model_mipsxx_ops.cpu_type = "mips/r10000-v2.x"; | |
363 | else | |
364 | op_model_mipsxx_ops.cpu_type = "mips/r10000"; | |
365 | break; | |
366 | ||
367 | case CPU_R12000: | |
368 | case CPU_R14000: | |
369 | op_model_mipsxx_ops.cpu_type = "mips/r12000"; | |
370 | break; | |
371 | ||
c03bc121 MM |
372 | case CPU_SB1: |
373 | case CPU_SB1A: | |
1acf1ca7 | 374 | op_model_mipsxx_ops.cpu_type = "mips/sb1"; |
c03bc121 MM |
375 | break; |
376 | ||
54176736 RB |
377 | default: |
378 | printk(KERN_ERR "Profiling unsupported for this CPU\n"); | |
379 | ||
380 | return -ENODEV; | |
381 | } | |
382 | ||
46684734 | 383 | save_perf_irq = perf_irq; |
54176736 RB |
384 | perf_irq = mipsxx_perfcount_handler; |
385 | ||
3572a2c3 FF |
386 | if ((cp0_perfcount_irq >= 0) && (cp0_compare_irq != cp0_perfcount_irq)) |
387 | return request_irq(cp0_perfcount_irq, mipsxx_perfcount_int, | |
388 | 0, "Perfcounter", save_perf_irq); | |
389 | ||
54176736 RB |
390 | return 0; |
391 | } | |
392 | ||
393 | static void mipsxx_exit(void) | |
394 | { | |
795a2258 | 395 | int counters = op_model_mipsxx_ops.num_counters; |
5e2862eb | 396 | |
3572a2c3 FF |
397 | if ((cp0_perfcount_irq >= 0) && (cp0_compare_irq != cp0_perfcount_irq)) |
398 | free_irq(cp0_perfcount_irq, save_perf_irq); | |
399 | ||
5e2862eb | 400 | counters = counters_per_cpu_to_total(counters); |
f6f88e9b | 401 | on_each_cpu(reset_counters, (void *)(long)counters, 1); |
54176736 | 402 | |
46684734 | 403 | perf_irq = save_perf_irq; |
54176736 RB |
404 | } |
405 | ||
1acf1ca7 | 406 | struct op_mips_model op_model_mipsxx_ops = { |
54176736 RB |
407 | .reg_setup = mipsxx_reg_setup, |
408 | .cpu_setup = mipsxx_cpu_setup, | |
409 | .init = mipsxx_init, | |
410 | .exit = mipsxx_exit, | |
411 | .cpu_start = mipsxx_cpu_start, | |
412 | .cpu_stop = mipsxx_cpu_stop, | |
413 | }; |