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> |
a669efc4 | 14 | #include <asm/time.h> |
54176736 RB |
15 | |
16 | #include "op_impl.h" | |
17 | ||
2654294b JH |
18 | #define M_PERFCTL_EVENT(event) (((event) << MIPS_PERFCTRL_EVENT_S) & \ |
19 | MIPS_PERFCTRL_EVENT) | |
20 | #define M_PERFCTL_VPEID(vpe) ((vpe) << MIPS_PERFCTRL_VPEID_S) | |
92c7b62f | 21 | |
70342287 | 22 | #define M_COUNTER_OVERFLOW (1UL << 31) |
92c7b62f | 23 | |
46684734 | 24 | static int (*save_perf_irq)(void); |
a669efc4 | 25 | static int perfcount_irq; |
46684734 | 26 | |
c783390a MB |
27 | /* |
28 | * XLR has only one set of counters per core. Designate the | |
29 | * first hardware thread in the core for setup and init. | |
30 | * Skip CPUs with non-zero hardware thread id (4 hwt per core) | |
31 | */ | |
83a18415 | 32 | #if defined(CONFIG_CPU_XLR) && defined(CONFIG_SMP) |
c783390a MB |
33 | #define oprofile_skip_cpu(c) ((cpu_logical_map(c) & 0x3) != 0) |
34 | #else | |
35 | #define oprofile_skip_cpu(c) 0 | |
36 | #endif | |
37 | ||
92c7b62f | 38 | #ifdef CONFIG_MIPS_MT_SMP |
2654294b | 39 | #define WHAT (MIPS_PERFCTRL_MT_EN_VPE | \ |
f875a832 | 40 | M_PERFCTL_VPEID(cpu_vpe_id(¤t_cpu_data))) |
39b8d525 | 41 | #define vpe_id() (cpu_has_mipsmt_pertccounters ? \ |
f875a832 | 42 | 0 : cpu_vpe_id(¤t_cpu_data)) |
5e2862eb RB |
43 | |
44 | /* | |
45 | * The number of bits to shift to convert between counters per core and | |
46 | * counters per VPE. There is no reasonable interface atm to obtain the | |
47 | * number of VPEs used by Linux and in the 34K this number is fixed to two | |
48 | * anyways so we hardcore a few things here for the moment. The way it's | |
49 | * done here will ensure that oprofile VSMP kernel will run right on a lesser | |
50 | * core like a 24K also or with maxcpus=1. | |
51 | */ | |
52 | static inline unsigned int vpe_shift(void) | |
53 | { | |
54 | if (num_possible_cpus() > 1) | |
55 | return 1; | |
56 | ||
57 | return 0; | |
58 | } | |
59 | ||
92c7b62f | 60 | #else |
5e2862eb | 61 | |
be609f35 | 62 | #define WHAT 0 |
6f4c5bde | 63 | #define vpe_id() 0 |
5e2862eb RB |
64 | |
65 | static inline unsigned int vpe_shift(void) | |
66 | { | |
67 | return 0; | |
68 | } | |
69 | ||
92c7b62f | 70 | #endif |
54176736 | 71 | |
5e2862eb RB |
72 | static inline unsigned int counters_total_to_per_cpu(unsigned int counters) |
73 | { | |
74 | return counters >> vpe_shift(); | |
75 | } | |
76 | ||
77 | static inline unsigned int counters_per_cpu_to_total(unsigned int counters) | |
78 | { | |
79 | return counters << vpe_shift(); | |
80 | } | |
81 | ||
92c7b62f RB |
82 | #define __define_perf_accessors(r, n, np) \ |
83 | \ | |
84 | static inline unsigned int r_c0_ ## r ## n(void) \ | |
85 | { \ | |
be609f35 | 86 | unsigned int cpu = vpe_id(); \ |
92c7b62f RB |
87 | \ |
88 | switch (cpu) { \ | |
89 | case 0: \ | |
90 | return read_c0_ ## r ## n(); \ | |
91 | case 1: \ | |
92 | return read_c0_ ## r ## np(); \ | |
93 | default: \ | |
94 | BUG(); \ | |
95 | } \ | |
30f244ae | 96 | return 0; \ |
92c7b62f RB |
97 | } \ |
98 | \ | |
99 | static inline void w_c0_ ## r ## n(unsigned int value) \ | |
100 | { \ | |
be609f35 | 101 | unsigned int cpu = vpe_id(); \ |
92c7b62f RB |
102 | \ |
103 | switch (cpu) { \ | |
104 | case 0: \ | |
105 | write_c0_ ## r ## n(value); \ | |
106 | return; \ | |
107 | case 1: \ | |
108 | write_c0_ ## r ## np(value); \ | |
109 | return; \ | |
110 | default: \ | |
111 | BUG(); \ | |
112 | } \ | |
30f244ae | 113 | return; \ |
92c7b62f RB |
114 | } \ |
115 | ||
116 | __define_perf_accessors(perfcntr, 0, 2) | |
117 | __define_perf_accessors(perfcntr, 1, 3) | |
795a2258 CD |
118 | __define_perf_accessors(perfcntr, 2, 0) |
119 | __define_perf_accessors(perfcntr, 3, 1) | |
92c7b62f RB |
120 | |
121 | __define_perf_accessors(perfctrl, 0, 2) | |
122 | __define_perf_accessors(perfctrl, 1, 3) | |
795a2258 CD |
123 | __define_perf_accessors(perfctrl, 2, 0) |
124 | __define_perf_accessors(perfctrl, 3, 1) | |
54176736 | 125 | |
1acf1ca7 | 126 | struct op_mips_model op_model_mipsxx_ops; |
54176736 RB |
127 | |
128 | static struct mipsxx_register_config { | |
129 | unsigned int control[4]; | |
130 | unsigned int counter[4]; | |
131 | } reg; | |
132 | ||
70342287 | 133 | /* Compute all of the registers in preparation for enabling profiling. */ |
54176736 RB |
134 | |
135 | static void mipsxx_reg_setup(struct op_counter_config *ctr) | |
136 | { | |
1acf1ca7 | 137 | unsigned int counters = op_model_mipsxx_ops.num_counters; |
54176736 RB |
138 | int i; |
139 | ||
140 | /* Compute the performance counter control word. */ | |
54176736 RB |
141 | for (i = 0; i < counters; i++) { |
142 | reg.control[i] = 0; | |
143 | reg.counter[i] = 0; | |
144 | ||
145 | if (!ctr[i].enabled) | |
146 | continue; | |
147 | ||
148 | reg.control[i] = M_PERFCTL_EVENT(ctr[i].event) | | |
2654294b | 149 | MIPS_PERFCTRL_IE; |
54176736 | 150 | if (ctr[i].kernel) |
2654294b | 151 | reg.control[i] |= MIPS_PERFCTRL_K; |
54176736 | 152 | if (ctr[i].user) |
2654294b | 153 | reg.control[i] |= MIPS_PERFCTRL_U; |
54176736 | 154 | if (ctr[i].exl) |
2654294b | 155 | reg.control[i] |= MIPS_PERFCTRL_EXL; |
cf5b2d23 | 156 | if (boot_cpu_type() == CPU_XLR) |
2654294b | 157 | reg.control[i] |= XLR_PERFCTRL_ALLTHREADS; |
54176736 RB |
158 | reg.counter[i] = 0x80000000 - ctr[i].count; |
159 | } | |
160 | } | |
161 | ||
70342287 | 162 | /* Program all of the registers in preparation for enabling profiling. */ |
54176736 | 163 | |
49a89efb | 164 | static void mipsxx_cpu_setup(void *args) |
54176736 | 165 | { |
1acf1ca7 | 166 | unsigned int counters = op_model_mipsxx_ops.num_counters; |
54176736 | 167 | |
c783390a MB |
168 | if (oprofile_skip_cpu(smp_processor_id())) |
169 | return; | |
170 | ||
54176736 RB |
171 | switch (counters) { |
172 | case 4: | |
92c7b62f RB |
173 | w_c0_perfctrl3(0); |
174 | w_c0_perfcntr3(reg.counter[3]); | |
2d291e6c | 175 | /* fall through */ |
54176736 | 176 | case 3: |
92c7b62f RB |
177 | w_c0_perfctrl2(0); |
178 | w_c0_perfcntr2(reg.counter[2]); | |
2d291e6c | 179 | /* fall through */ |
54176736 | 180 | case 2: |
92c7b62f RB |
181 | w_c0_perfctrl1(0); |
182 | w_c0_perfcntr1(reg.counter[1]); | |
2d291e6c | 183 | /* fall through */ |
54176736 | 184 | case 1: |
92c7b62f RB |
185 | w_c0_perfctrl0(0); |
186 | w_c0_perfcntr0(reg.counter[0]); | |
54176736 RB |
187 | } |
188 | } | |
189 | ||
190 | /* Start all counters on current CPU */ | |
191 | static void mipsxx_cpu_start(void *args) | |
192 | { | |
1acf1ca7 | 193 | unsigned int counters = op_model_mipsxx_ops.num_counters; |
54176736 | 194 | |
c783390a MB |
195 | if (oprofile_skip_cpu(smp_processor_id())) |
196 | return; | |
197 | ||
54176736 RB |
198 | switch (counters) { |
199 | case 4: | |
92c7b62f | 200 | w_c0_perfctrl3(WHAT | reg.control[3]); |
2d291e6c | 201 | /* fall through */ |
54176736 | 202 | case 3: |
92c7b62f | 203 | w_c0_perfctrl2(WHAT | reg.control[2]); |
2d291e6c | 204 | /* fall through */ |
54176736 | 205 | case 2: |
92c7b62f | 206 | w_c0_perfctrl1(WHAT | reg.control[1]); |
2d291e6c | 207 | /* fall through */ |
54176736 | 208 | case 1: |
92c7b62f | 209 | w_c0_perfctrl0(WHAT | reg.control[0]); |
54176736 RB |
210 | } |
211 | } | |
212 | ||
213 | /* Stop all counters on current CPU */ | |
214 | static void mipsxx_cpu_stop(void *args) | |
215 | { | |
1acf1ca7 | 216 | unsigned int counters = op_model_mipsxx_ops.num_counters; |
54176736 | 217 | |
c783390a MB |
218 | if (oprofile_skip_cpu(smp_processor_id())) |
219 | return; | |
220 | ||
54176736 RB |
221 | switch (counters) { |
222 | case 4: | |
92c7b62f | 223 | w_c0_perfctrl3(0); |
2d291e6c | 224 | /* fall through */ |
54176736 | 225 | case 3: |
92c7b62f | 226 | w_c0_perfctrl2(0); |
2d291e6c | 227 | /* fall through */ |
54176736 | 228 | case 2: |
92c7b62f | 229 | w_c0_perfctrl1(0); |
2d291e6c | 230 | /* fall through */ |
54176736 | 231 | case 1: |
92c7b62f | 232 | w_c0_perfctrl0(0); |
54176736 RB |
233 | } |
234 | } | |
235 | ||
937a8015 | 236 | static int mipsxx_perfcount_handler(void) |
54176736 | 237 | { |
1acf1ca7 | 238 | unsigned int counters = op_model_mipsxx_ops.num_counters; |
54176736 RB |
239 | unsigned int control; |
240 | unsigned int counter; | |
ffe9ee47 CD |
241 | int handled = IRQ_NONE; |
242 | ||
3ba5040a | 243 | if (cpu_has_mips_r2 && !(read_c0_cause() & CAUSEF_PCI)) |
ffe9ee47 | 244 | return handled; |
54176736 RB |
245 | |
246 | switch (counters) { | |
247 | #define HANDLE_COUNTER(n) \ | |
2d291e6c | 248 | /* fall through */ \ |
54176736 | 249 | case n + 1: \ |
92c7b62f RB |
250 | control = r_c0_perfctrl ## n(); \ |
251 | counter = r_c0_perfcntr ## n(); \ | |
2654294b | 252 | if ((control & MIPS_PERFCTRL_IE) && \ |
54176736 | 253 | (counter & M_COUNTER_OVERFLOW)) { \ |
937a8015 | 254 | oprofile_add_sample(get_irq_regs(), n); \ |
92c7b62f | 255 | w_c0_perfcntr ## n(reg.counter[n]); \ |
ffe9ee47 | 256 | handled = IRQ_HANDLED; \ |
54176736 RB |
257 | } |
258 | HANDLE_COUNTER(3) | |
259 | HANDLE_COUNTER(2) | |
260 | HANDLE_COUNTER(1) | |
261 | HANDLE_COUNTER(0) | |
262 | } | |
ba339c03 RB |
263 | |
264 | return handled; | |
54176736 RB |
265 | } |
266 | ||
92c7b62f | 267 | static inline int __n_counters(void) |
54176736 | 268 | { |
30228c40 | 269 | if (!cpu_has_perf) |
54176736 | 270 | return 0; |
2654294b | 271 | if (!(read_c0_perfctrl0() & MIPS_PERFCTRL_M)) |
54176736 | 272 | return 1; |
2654294b | 273 | if (!(read_c0_perfctrl1() & MIPS_PERFCTRL_M)) |
54176736 | 274 | return 2; |
2654294b | 275 | if (!(read_c0_perfctrl2() & MIPS_PERFCTRL_M)) |
54176736 RB |
276 | return 3; |
277 | ||
278 | return 4; | |
279 | } | |
280 | ||
92c7b62f RB |
281 | static inline int n_counters(void) |
282 | { | |
714cfe78 RB |
283 | int counters; |
284 | ||
10cc3529 | 285 | switch (current_cpu_type()) { |
714cfe78 RB |
286 | case CPU_R10000: |
287 | counters = 2; | |
148171b2 | 288 | break; |
714cfe78 RB |
289 | |
290 | case CPU_R12000: | |
291 | case CPU_R14000: | |
30577391 | 292 | case CPU_R16000: |
714cfe78 | 293 | counters = 4; |
148171b2 | 294 | break; |
714cfe78 RB |
295 | |
296 | default: | |
297 | counters = __n_counters(); | |
298 | } | |
92c7b62f | 299 | |
92c7b62f RB |
300 | return counters; |
301 | } | |
302 | ||
39b8d525 | 303 | static void reset_counters(void *arg) |
54176736 | 304 | { |
005ca9a3 | 305 | int counters = (int)(long)arg; |
54176736 RB |
306 | switch (counters) { |
307 | case 4: | |
92c7b62f RB |
308 | w_c0_perfctrl3(0); |
309 | w_c0_perfcntr3(0); | |
2d291e6c | 310 | /* fall through */ |
54176736 | 311 | case 3: |
92c7b62f RB |
312 | w_c0_perfctrl2(0); |
313 | w_c0_perfcntr2(0); | |
2d291e6c | 314 | /* fall through */ |
54176736 | 315 | case 2: |
92c7b62f RB |
316 | w_c0_perfctrl1(0); |
317 | w_c0_perfcntr1(0); | |
2d291e6c | 318 | /* fall through */ |
54176736 | 319 | case 1: |
92c7b62f RB |
320 | w_c0_perfctrl0(0); |
321 | w_c0_perfcntr0(0); | |
54176736 RB |
322 | } |
323 | } | |
324 | ||
3572a2c3 FF |
325 | static irqreturn_t mipsxx_perfcount_int(int irq, void *dev_id) |
326 | { | |
327 | return mipsxx_perfcount_handler(); | |
328 | } | |
329 | ||
54176736 RB |
330 | static int __init mipsxx_init(void) |
331 | { | |
332 | int counters; | |
333 | ||
334 | counters = n_counters(); | |
9efeae9a RB |
335 | if (counters == 0) { |
336 | printk(KERN_ERR "Oprofile: CPU has no performance counters\n"); | |
54176736 | 337 | return -ENODEV; |
9efeae9a | 338 | } |
54176736 | 339 | |
39b8d525 | 340 | #ifdef CONFIG_MIPS_MT_SMP |
39b8d525 RB |
341 | if (!cpu_has_mipsmt_pertccounters) |
342 | counters = counters_total_to_per_cpu(counters); | |
343 | #endif | |
f6f88e9b | 344 | on_each_cpu(reset_counters, (void *)(long)counters, 1); |
795a2258 | 345 | |
1acf1ca7 | 346 | op_model_mipsxx_ops.num_counters = counters; |
10cc3529 | 347 | switch (current_cpu_type()) { |
113c62d9 SH |
348 | case CPU_M14KC: |
349 | op_model_mipsxx_ops.cpu_type = "mips/M14Kc"; | |
350 | break; | |
351 | ||
f8fa4811 SH |
352 | case CPU_M14KEC: |
353 | op_model_mipsxx_ops.cpu_type = "mips/M14KEc"; | |
354 | break; | |
355 | ||
2065988e | 356 | case CPU_20KC: |
1acf1ca7 | 357 | op_model_mipsxx_ops.cpu_type = "mips/20K"; |
2065988e RB |
358 | break; |
359 | ||
54176736 | 360 | case CPU_24K: |
1acf1ca7 | 361 | op_model_mipsxx_ops.cpu_type = "mips/24K"; |
54176736 RB |
362 | break; |
363 | ||
2065988e | 364 | case CPU_25KF: |
1acf1ca7 | 365 | op_model_mipsxx_ops.cpu_type = "mips/25K"; |
2065988e RB |
366 | break; |
367 | ||
39b8d525 | 368 | case CPU_1004K: |
fcfd980c | 369 | case CPU_34K: |
1acf1ca7 | 370 | op_model_mipsxx_ops.cpu_type = "mips/34K"; |
fcfd980c | 371 | break; |
c620953c | 372 | |
442e14a2 | 373 | case CPU_1074K: |
c620953c | 374 | case CPU_74K: |
1acf1ca7 | 375 | op_model_mipsxx_ops.cpu_type = "mips/74K"; |
c620953c | 376 | break; |
fcfd980c | 377 | |
26ab96df LY |
378 | case CPU_INTERAPTIV: |
379 | op_model_mipsxx_ops.cpu_type = "mips/interAptiv"; | |
380 | break; | |
381 | ||
708ac4b8 LY |
382 | case CPU_PROAPTIV: |
383 | op_model_mipsxx_ops.cpu_type = "mips/proAptiv"; | |
384 | break; | |
385 | ||
8c7f6ba3 JH |
386 | case CPU_P5600: |
387 | op_model_mipsxx_ops.cpu_type = "mips/P5600"; | |
388 | break; | |
389 | ||
4e88a862 MC |
390 | case CPU_I6400: |
391 | op_model_mipsxx_ops.cpu_type = "mips/I6400"; | |
392 | break; | |
393 | ||
f36c4720 LY |
394 | case CPU_M5150: |
395 | op_model_mipsxx_ops.cpu_type = "mips/M5150"; | |
396 | break; | |
397 | ||
2065988e | 398 | case CPU_5KC: |
1acf1ca7 | 399 | op_model_mipsxx_ops.cpu_type = "mips/5K"; |
2065988e RB |
400 | break; |
401 | ||
714cfe78 RB |
402 | case CPU_R10000: |
403 | if ((current_cpu_data.processor_id & 0xff) == 0x20) | |
404 | op_model_mipsxx_ops.cpu_type = "mips/r10000-v2.x"; | |
405 | else | |
406 | op_model_mipsxx_ops.cpu_type = "mips/r10000"; | |
407 | break; | |
408 | ||
409 | case CPU_R12000: | |
410 | case CPU_R14000: | |
411 | op_model_mipsxx_ops.cpu_type = "mips/r12000"; | |
412 | break; | |
413 | ||
30577391 JK |
414 | case CPU_R16000: |
415 | op_model_mipsxx_ops.cpu_type = "mips/r16000"; | |
416 | break; | |
417 | ||
c03bc121 MM |
418 | case CPU_SB1: |
419 | case CPU_SB1A: | |
1acf1ca7 | 420 | op_model_mipsxx_ops.cpu_type = "mips/sb1"; |
c03bc121 MM |
421 | break; |
422 | ||
b2afb64c | 423 | case CPU_LOONGSON32: |
2fa36399 KC |
424 | op_model_mipsxx_ops.cpu_type = "mips/loongson1"; |
425 | break; | |
426 | ||
c783390a MB |
427 | case CPU_XLR: |
428 | op_model_mipsxx_ops.cpu_type = "mips/xlr"; | |
429 | break; | |
430 | ||
54176736 RB |
431 | default: |
432 | printk(KERN_ERR "Profiling unsupported for this CPU\n"); | |
433 | ||
434 | return -ENODEV; | |
435 | } | |
436 | ||
46684734 | 437 | save_perf_irq = perf_irq; |
54176736 RB |
438 | perf_irq = mipsxx_perfcount_handler; |
439 | ||
a669efc4 AB |
440 | if (get_c0_perfcount_int) |
441 | perfcount_irq = get_c0_perfcount_int(); | |
7eca5b14 | 442 | else if (cp0_perfcount_irq >= 0) |
a669efc4 AB |
443 | perfcount_irq = MIPS_CPU_IRQ_BASE + cp0_perfcount_irq; |
444 | else | |
445 | perfcount_irq = -1; | |
446 | ||
447 | if (perfcount_irq >= 0) | |
448 | return request_irq(perfcount_irq, mipsxx_perfcount_int, | |
369a93bb JH |
449 | IRQF_PERCPU | IRQF_NOBALANCING | |
450 | IRQF_NO_THREAD | IRQF_NO_SUSPEND | | |
451 | IRQF_SHARED, | |
452 | "Perfcounter", save_perf_irq); | |
3572a2c3 | 453 | |
54176736 RB |
454 | return 0; |
455 | } | |
456 | ||
457 | static void mipsxx_exit(void) | |
458 | { | |
795a2258 | 459 | int counters = op_model_mipsxx_ops.num_counters; |
5e2862eb | 460 | |
a669efc4 AB |
461 | if (perfcount_irq >= 0) |
462 | free_irq(perfcount_irq, save_perf_irq); | |
3572a2c3 | 463 | |
5e2862eb | 464 | counters = counters_per_cpu_to_total(counters); |
f6f88e9b | 465 | on_each_cpu(reset_counters, (void *)(long)counters, 1); |
54176736 | 466 | |
46684734 | 467 | perf_irq = save_perf_irq; |
54176736 RB |
468 | } |
469 | ||
1acf1ca7 | 470 | struct op_mips_model op_model_mipsxx_ops = { |
54176736 RB |
471 | .reg_setup = mipsxx_reg_setup, |
472 | .cpu_setup = mipsxx_cpu_setup, | |
473 | .init = mipsxx_init, | |
474 | .exit = mipsxx_exit, | |
475 | .cpu_start = mipsxx_cpu_start, | |
476 | .cpu_stop = mipsxx_cpu_stop, | |
477 | }; |