]>
Commit | Line | Data |
---|---|---|
cdd6c482 | 1 | /* Performance event support for sparc64. |
59abbd1e | 2 | * |
4f6dbe4a | 3 | * Copyright (C) 2009, 2010 David S. Miller <davem@davemloft.net> |
59abbd1e | 4 | * |
cdd6c482 | 5 | * This code is based almost entirely upon the x86 perf event |
59abbd1e DM |
6 | * code, which is: |
7 | * | |
8 | * Copyright (C) 2008 Thomas Gleixner <tglx@linutronix.de> | |
9 | * Copyright (C) 2008-2009 Red Hat, Inc., Ingo Molnar | |
10 | * Copyright (C) 2009 Jaswinder Singh Rajput | |
11 | * Copyright (C) 2009 Advanced Micro Devices, Inc., Robert Richter | |
12 | * Copyright (C) 2008-2009 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com> | |
13 | */ | |
14 | ||
cdd6c482 | 15 | #include <linux/perf_event.h> |
59abbd1e | 16 | #include <linux/kprobes.h> |
667f0cee | 17 | #include <linux/ftrace.h> |
59abbd1e DM |
18 | #include <linux/kernel.h> |
19 | #include <linux/kdebug.h> | |
20 | #include <linux/mutex.h> | |
21 | ||
4f6dbe4a | 22 | #include <asm/stacktrace.h> |
59abbd1e | 23 | #include <asm/cpudata.h> |
4f6dbe4a | 24 | #include <asm/uaccess.h> |
60063497 | 25 | #include <linux/atomic.h> |
59abbd1e DM |
26 | #include <asm/nmi.h> |
27 | #include <asm/pcr.h> | |
d550bbd4 | 28 | #include <asm/cacheflush.h> |
59abbd1e | 29 | |
cb1b8209 | 30 | #include "kernel.h" |
4f6dbe4a DM |
31 | #include "kstack.h" |
32 | ||
59abbd1e DM |
33 | /* Sparc64 chips have two performance counters, 32-bits each, with |
34 | * overflow interrupts generated on transition from 0xffffffff to 0. | |
35 | * The counters are accessed in one go using a 64-bit register. | |
36 | * | |
37 | * Both counters are controlled using a single control register. The | |
38 | * only way to stop all sampling is to clear all of the context (user, | |
39 | * supervisor, hypervisor) sampling enable bits. But these bits apply | |
40 | * to both counters, thus the two counters can't be enabled/disabled | |
41 | * individually. | |
42 | * | |
43 | * The control register has two event fields, one for each of the two | |
44 | * counters. It's thus nearly impossible to have one counter going | |
45 | * while keeping the other one stopped. Therefore it is possible to | |
46 | * get overflow interrupts for counters not currently "in use" and | |
47 | * that condition must be checked in the overflow interrupt handler. | |
48 | * | |
49 | * So we use a hack, in that we program inactive counters with the | |
50 | * "sw_count0" and "sw_count1" events. These count how many times | |
51 | * the instruction "sethi %hi(0xfc000), %g0" is executed. It's an | |
52 | * unusual way to encode a NOP and therefore will not trigger in | |
53 | * normal code. | |
54 | */ | |
55 | ||
cdd6c482 | 56 | #define MAX_HWEVENTS 2 |
59abbd1e DM |
57 | #define MAX_PERIOD ((1UL << 32) - 1) |
58 | ||
59 | #define PIC_UPPER_INDEX 0 | |
60 | #define PIC_LOWER_INDEX 1 | |
e7bef6b0 | 61 | #define PIC_NO_INDEX -1 |
59abbd1e | 62 | |
cdd6c482 | 63 | struct cpu_hw_events { |
e7bef6b0 DM |
64 | /* Number of events currently scheduled onto this cpu. |
65 | * This tells how many entries in the arrays below | |
66 | * are valid. | |
67 | */ | |
68 | int n_events; | |
69 | ||
70 | /* Number of new events added since the last hw_perf_disable(). | |
71 | * This works because the perf event layer always adds new | |
72 | * events inside of a perf_{disable,enable}() sequence. | |
73 | */ | |
74 | int n_added; | |
75 | ||
76 | /* Array of events current scheduled on this cpu. */ | |
77 | struct perf_event *event[MAX_HWEVENTS]; | |
78 | ||
79 | /* Array of encoded longs, specifying the %pcr register | |
80 | * encoding and the mask of PIC counters this even can | |
81 | * be scheduled on. See perf_event_encode() et al. | |
82 | */ | |
83 | unsigned long events[MAX_HWEVENTS]; | |
84 | ||
85 | /* The current counter index assigned to an event. When the | |
86 | * event hasn't been programmed into the cpu yet, this will | |
87 | * hold PIC_NO_INDEX. The event->hw.idx value tells us where | |
88 | * we ought to schedule the event. | |
89 | */ | |
90 | int current_idx[MAX_HWEVENTS]; | |
91 | ||
92 | /* Software copy of %pcr register on this cpu. */ | |
d1751388 | 93 | u64 pcr; |
e7bef6b0 DM |
94 | |
95 | /* Enabled/disable state. */ | |
d1751388 | 96 | int enabled; |
a13c3afd LM |
97 | |
98 | unsigned int group_flag; | |
59abbd1e | 99 | }; |
cdd6c482 | 100 | DEFINE_PER_CPU(struct cpu_hw_events, cpu_hw_events) = { .enabled = 1, }; |
59abbd1e | 101 | |
e7bef6b0 DM |
102 | /* An event map describes the characteristics of a performance |
103 | * counter event. In particular it gives the encoding as well as | |
104 | * a mask telling which counters the event can be measured on. | |
105 | */ | |
59abbd1e DM |
106 | struct perf_event_map { |
107 | u16 encoding; | |
108 | u8 pic_mask; | |
109 | #define PIC_NONE 0x00 | |
110 | #define PIC_UPPER 0x01 | |
111 | #define PIC_LOWER 0x02 | |
112 | }; | |
113 | ||
e7bef6b0 | 114 | /* Encode a perf_event_map entry into a long. */ |
a72a8a5f DM |
115 | static unsigned long perf_event_encode(const struct perf_event_map *pmap) |
116 | { | |
117 | return ((unsigned long) pmap->encoding << 16) | pmap->pic_mask; | |
118 | } | |
119 | ||
e7bef6b0 DM |
120 | static u8 perf_event_get_msk(unsigned long val) |
121 | { | |
122 | return val & 0xff; | |
123 | } | |
124 | ||
125 | static u64 perf_event_get_enc(unsigned long val) | |
a72a8a5f | 126 | { |
e7bef6b0 | 127 | return val >> 16; |
a72a8a5f DM |
128 | } |
129 | ||
2ce4da2e DM |
130 | #define C(x) PERF_COUNT_HW_CACHE_##x |
131 | ||
132 | #define CACHE_OP_UNSUPPORTED 0xfffe | |
133 | #define CACHE_OP_NONSENSE 0xffff | |
134 | ||
135 | typedef struct perf_event_map cache_map_t | |
136 | [PERF_COUNT_HW_CACHE_MAX] | |
137 | [PERF_COUNT_HW_CACHE_OP_MAX] | |
138 | [PERF_COUNT_HW_CACHE_RESULT_MAX]; | |
139 | ||
59abbd1e DM |
140 | struct sparc_pmu { |
141 | const struct perf_event_map *(*event_map)(int); | |
2ce4da2e | 142 | const cache_map_t *cache_map; |
59abbd1e DM |
143 | int max_events; |
144 | int upper_shift; | |
145 | int lower_shift; | |
146 | int event_mask; | |
91b9286d | 147 | int hv_bit; |
496c07e3 | 148 | int irq_bit; |
660d1376 DM |
149 | int upper_nop; |
150 | int lower_nop; | |
59abbd1e DM |
151 | }; |
152 | ||
28e8f9be | 153 | static const struct perf_event_map ultra3_perfmon_event_map[] = { |
59abbd1e DM |
154 | [PERF_COUNT_HW_CPU_CYCLES] = { 0x0000, PIC_UPPER | PIC_LOWER }, |
155 | [PERF_COUNT_HW_INSTRUCTIONS] = { 0x0001, PIC_UPPER | PIC_LOWER }, | |
156 | [PERF_COUNT_HW_CACHE_REFERENCES] = { 0x0009, PIC_LOWER }, | |
157 | [PERF_COUNT_HW_CACHE_MISSES] = { 0x0009, PIC_UPPER }, | |
158 | }; | |
159 | ||
28e8f9be | 160 | static const struct perf_event_map *ultra3_event_map(int event_id) |
59abbd1e | 161 | { |
28e8f9be | 162 | return &ultra3_perfmon_event_map[event_id]; |
59abbd1e DM |
163 | } |
164 | ||
28e8f9be | 165 | static const cache_map_t ultra3_cache_map = { |
2ce4da2e DM |
166 | [C(L1D)] = { |
167 | [C(OP_READ)] = { | |
168 | [C(RESULT_ACCESS)] = { 0x09, PIC_LOWER, }, | |
169 | [C(RESULT_MISS)] = { 0x09, PIC_UPPER, }, | |
170 | }, | |
171 | [C(OP_WRITE)] = { | |
172 | [C(RESULT_ACCESS)] = { 0x0a, PIC_LOWER }, | |
173 | [C(RESULT_MISS)] = { 0x0a, PIC_UPPER }, | |
174 | }, | |
175 | [C(OP_PREFETCH)] = { | |
176 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
177 | [C(RESULT_MISS)] = { CACHE_OP_UNSUPPORTED }, | |
178 | }, | |
179 | }, | |
180 | [C(L1I)] = { | |
181 | [C(OP_READ)] = { | |
182 | [C(RESULT_ACCESS)] = { 0x09, PIC_LOWER, }, | |
183 | [C(RESULT_MISS)] = { 0x09, PIC_UPPER, }, | |
184 | }, | |
185 | [ C(OP_WRITE) ] = { | |
186 | [ C(RESULT_ACCESS) ] = { CACHE_OP_NONSENSE }, | |
187 | [ C(RESULT_MISS) ] = { CACHE_OP_NONSENSE }, | |
188 | }, | |
189 | [ C(OP_PREFETCH) ] = { | |
190 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
191 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
192 | }, | |
193 | }, | |
194 | [C(LL)] = { | |
195 | [C(OP_READ)] = { | |
196 | [C(RESULT_ACCESS)] = { 0x0c, PIC_LOWER, }, | |
197 | [C(RESULT_MISS)] = { 0x0c, PIC_UPPER, }, | |
198 | }, | |
199 | [C(OP_WRITE)] = { | |
200 | [C(RESULT_ACCESS)] = { 0x0c, PIC_LOWER }, | |
201 | [C(RESULT_MISS)] = { 0x0c, PIC_UPPER }, | |
202 | }, | |
203 | [C(OP_PREFETCH)] = { | |
204 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
205 | [C(RESULT_MISS)] = { CACHE_OP_UNSUPPORTED }, | |
206 | }, | |
207 | }, | |
208 | [C(DTLB)] = { | |
209 | [C(OP_READ)] = { | |
210 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
211 | [C(RESULT_MISS)] = { 0x12, PIC_UPPER, }, | |
212 | }, | |
213 | [ C(OP_WRITE) ] = { | |
214 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
215 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
216 | }, | |
217 | [ C(OP_PREFETCH) ] = { | |
218 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
219 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
220 | }, | |
221 | }, | |
222 | [C(ITLB)] = { | |
223 | [C(OP_READ)] = { | |
224 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
225 | [C(RESULT_MISS)] = { 0x11, PIC_UPPER, }, | |
226 | }, | |
227 | [ C(OP_WRITE) ] = { | |
228 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
229 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
230 | }, | |
231 | [ C(OP_PREFETCH) ] = { | |
232 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
233 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
234 | }, | |
235 | }, | |
236 | [C(BPU)] = { | |
237 | [C(OP_READ)] = { | |
238 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
239 | [C(RESULT_MISS)] = { CACHE_OP_UNSUPPORTED }, | |
240 | }, | |
241 | [ C(OP_WRITE) ] = { | |
242 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
243 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
244 | }, | |
245 | [ C(OP_PREFETCH) ] = { | |
246 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
247 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
248 | }, | |
249 | }, | |
89d6c0b5 PZ |
250 | [C(NODE)] = { |
251 | [C(OP_READ)] = { | |
252 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
253 | [C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
254 | }, | |
255 | [ C(OP_WRITE) ] = { | |
256 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
257 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
258 | }, | |
259 | [ C(OP_PREFETCH) ] = { | |
260 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
261 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
262 | }, | |
263 | }, | |
2ce4da2e DM |
264 | }; |
265 | ||
28e8f9be DM |
266 | static const struct sparc_pmu ultra3_pmu = { |
267 | .event_map = ultra3_event_map, | |
268 | .cache_map = &ultra3_cache_map, | |
269 | .max_events = ARRAY_SIZE(ultra3_perfmon_event_map), | |
59abbd1e DM |
270 | .upper_shift = 11, |
271 | .lower_shift = 4, | |
272 | .event_mask = 0x3f, | |
660d1376 DM |
273 | .upper_nop = 0x1c, |
274 | .lower_nop = 0x14, | |
59abbd1e DM |
275 | }; |
276 | ||
7eebda60 DM |
277 | /* Niagara1 is very limited. The upper PIC is hard-locked to count |
278 | * only instructions, so it is free running which creates all kinds of | |
6e804251 | 279 | * problems. Some hardware designs make one wonder if the creator |
7eebda60 DM |
280 | * even looked at how this stuff gets used by software. |
281 | */ | |
282 | static const struct perf_event_map niagara1_perfmon_event_map[] = { | |
283 | [PERF_COUNT_HW_CPU_CYCLES] = { 0x00, PIC_UPPER }, | |
284 | [PERF_COUNT_HW_INSTRUCTIONS] = { 0x00, PIC_UPPER }, | |
285 | [PERF_COUNT_HW_CACHE_REFERENCES] = { 0, PIC_NONE }, | |
286 | [PERF_COUNT_HW_CACHE_MISSES] = { 0x03, PIC_LOWER }, | |
287 | }; | |
288 | ||
289 | static const struct perf_event_map *niagara1_event_map(int event_id) | |
290 | { | |
291 | return &niagara1_perfmon_event_map[event_id]; | |
292 | } | |
293 | ||
294 | static const cache_map_t niagara1_cache_map = { | |
295 | [C(L1D)] = { | |
296 | [C(OP_READ)] = { | |
297 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
298 | [C(RESULT_MISS)] = { 0x03, PIC_LOWER, }, | |
299 | }, | |
300 | [C(OP_WRITE)] = { | |
301 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
302 | [C(RESULT_MISS)] = { 0x03, PIC_LOWER, }, | |
303 | }, | |
304 | [C(OP_PREFETCH)] = { | |
305 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
306 | [C(RESULT_MISS)] = { CACHE_OP_UNSUPPORTED }, | |
307 | }, | |
308 | }, | |
309 | [C(L1I)] = { | |
310 | [C(OP_READ)] = { | |
311 | [C(RESULT_ACCESS)] = { 0x00, PIC_UPPER }, | |
312 | [C(RESULT_MISS)] = { 0x02, PIC_LOWER, }, | |
313 | }, | |
314 | [ C(OP_WRITE) ] = { | |
315 | [ C(RESULT_ACCESS) ] = { CACHE_OP_NONSENSE }, | |
316 | [ C(RESULT_MISS) ] = { CACHE_OP_NONSENSE }, | |
317 | }, | |
318 | [ C(OP_PREFETCH) ] = { | |
319 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
320 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
321 | }, | |
322 | }, | |
323 | [C(LL)] = { | |
324 | [C(OP_READ)] = { | |
325 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
326 | [C(RESULT_MISS)] = { 0x07, PIC_LOWER, }, | |
327 | }, | |
328 | [C(OP_WRITE)] = { | |
329 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
330 | [C(RESULT_MISS)] = { 0x07, PIC_LOWER, }, | |
331 | }, | |
332 | [C(OP_PREFETCH)] = { | |
333 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
334 | [C(RESULT_MISS)] = { CACHE_OP_UNSUPPORTED }, | |
335 | }, | |
336 | }, | |
337 | [C(DTLB)] = { | |
338 | [C(OP_READ)] = { | |
339 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
340 | [C(RESULT_MISS)] = { 0x05, PIC_LOWER, }, | |
341 | }, | |
342 | [ C(OP_WRITE) ] = { | |
343 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
344 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
345 | }, | |
346 | [ C(OP_PREFETCH) ] = { | |
347 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
348 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
349 | }, | |
350 | }, | |
351 | [C(ITLB)] = { | |
352 | [C(OP_READ)] = { | |
353 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
354 | [C(RESULT_MISS)] = { 0x04, PIC_LOWER, }, | |
355 | }, | |
356 | [ C(OP_WRITE) ] = { | |
357 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
358 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
359 | }, | |
360 | [ C(OP_PREFETCH) ] = { | |
361 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
362 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
363 | }, | |
364 | }, | |
365 | [C(BPU)] = { | |
366 | [C(OP_READ)] = { | |
367 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
368 | [C(RESULT_MISS)] = { CACHE_OP_UNSUPPORTED }, | |
369 | }, | |
370 | [ C(OP_WRITE) ] = { | |
371 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
372 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
373 | }, | |
374 | [ C(OP_PREFETCH) ] = { | |
375 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
376 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
377 | }, | |
378 | }, | |
89d6c0b5 PZ |
379 | [C(NODE)] = { |
380 | [C(OP_READ)] = { | |
381 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
382 | [C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
383 | }, | |
384 | [ C(OP_WRITE) ] = { | |
385 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
386 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
387 | }, | |
388 | [ C(OP_PREFETCH) ] = { | |
389 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
390 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
391 | }, | |
392 | }, | |
7eebda60 DM |
393 | }; |
394 | ||
395 | static const struct sparc_pmu niagara1_pmu = { | |
396 | .event_map = niagara1_event_map, | |
397 | .cache_map = &niagara1_cache_map, | |
398 | .max_events = ARRAY_SIZE(niagara1_perfmon_event_map), | |
399 | .upper_shift = 0, | |
400 | .lower_shift = 4, | |
401 | .event_mask = 0x7, | |
402 | .upper_nop = 0x0, | |
403 | .lower_nop = 0x0, | |
404 | }; | |
405 | ||
b73d8847 DM |
406 | static const struct perf_event_map niagara2_perfmon_event_map[] = { |
407 | [PERF_COUNT_HW_CPU_CYCLES] = { 0x02ff, PIC_UPPER | PIC_LOWER }, | |
408 | [PERF_COUNT_HW_INSTRUCTIONS] = { 0x02ff, PIC_UPPER | PIC_LOWER }, | |
409 | [PERF_COUNT_HW_CACHE_REFERENCES] = { 0x0208, PIC_UPPER | PIC_LOWER }, | |
410 | [PERF_COUNT_HW_CACHE_MISSES] = { 0x0302, PIC_UPPER | PIC_LOWER }, | |
411 | [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = { 0x0201, PIC_UPPER | PIC_LOWER }, | |
412 | [PERF_COUNT_HW_BRANCH_MISSES] = { 0x0202, PIC_UPPER | PIC_LOWER }, | |
413 | }; | |
414 | ||
cdd6c482 | 415 | static const struct perf_event_map *niagara2_event_map(int event_id) |
b73d8847 | 416 | { |
cdd6c482 | 417 | return &niagara2_perfmon_event_map[event_id]; |
b73d8847 DM |
418 | } |
419 | ||
d0b86480 DM |
420 | static const cache_map_t niagara2_cache_map = { |
421 | [C(L1D)] = { | |
422 | [C(OP_READ)] = { | |
423 | [C(RESULT_ACCESS)] = { 0x0208, PIC_UPPER | PIC_LOWER, }, | |
424 | [C(RESULT_MISS)] = { 0x0302, PIC_UPPER | PIC_LOWER, }, | |
425 | }, | |
426 | [C(OP_WRITE)] = { | |
427 | [C(RESULT_ACCESS)] = { 0x0210, PIC_UPPER | PIC_LOWER, }, | |
428 | [C(RESULT_MISS)] = { 0x0302, PIC_UPPER | PIC_LOWER, }, | |
429 | }, | |
430 | [C(OP_PREFETCH)] = { | |
431 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
432 | [C(RESULT_MISS)] = { CACHE_OP_UNSUPPORTED }, | |
433 | }, | |
434 | }, | |
435 | [C(L1I)] = { | |
436 | [C(OP_READ)] = { | |
437 | [C(RESULT_ACCESS)] = { 0x02ff, PIC_UPPER | PIC_LOWER, }, | |
438 | [C(RESULT_MISS)] = { 0x0301, PIC_UPPER | PIC_LOWER, }, | |
439 | }, | |
440 | [ C(OP_WRITE) ] = { | |
441 | [ C(RESULT_ACCESS) ] = { CACHE_OP_NONSENSE }, | |
442 | [ C(RESULT_MISS) ] = { CACHE_OP_NONSENSE }, | |
443 | }, | |
444 | [ C(OP_PREFETCH) ] = { | |
445 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
446 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
447 | }, | |
448 | }, | |
449 | [C(LL)] = { | |
450 | [C(OP_READ)] = { | |
451 | [C(RESULT_ACCESS)] = { 0x0208, PIC_UPPER | PIC_LOWER, }, | |
452 | [C(RESULT_MISS)] = { 0x0330, PIC_UPPER | PIC_LOWER, }, | |
453 | }, | |
454 | [C(OP_WRITE)] = { | |
455 | [C(RESULT_ACCESS)] = { 0x0210, PIC_UPPER | PIC_LOWER, }, | |
456 | [C(RESULT_MISS)] = { 0x0320, PIC_UPPER | PIC_LOWER, }, | |
457 | }, | |
458 | [C(OP_PREFETCH)] = { | |
459 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
460 | [C(RESULT_MISS)] = { CACHE_OP_UNSUPPORTED }, | |
461 | }, | |
462 | }, | |
463 | [C(DTLB)] = { | |
464 | [C(OP_READ)] = { | |
465 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
466 | [C(RESULT_MISS)] = { 0x0b08, PIC_UPPER | PIC_LOWER, }, | |
467 | }, | |
468 | [ C(OP_WRITE) ] = { | |
469 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
470 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
471 | }, | |
472 | [ C(OP_PREFETCH) ] = { | |
473 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
474 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
475 | }, | |
476 | }, | |
477 | [C(ITLB)] = { | |
478 | [C(OP_READ)] = { | |
479 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
480 | [C(RESULT_MISS)] = { 0xb04, PIC_UPPER | PIC_LOWER, }, | |
481 | }, | |
482 | [ C(OP_WRITE) ] = { | |
483 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
484 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
485 | }, | |
486 | [ C(OP_PREFETCH) ] = { | |
487 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
488 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
489 | }, | |
490 | }, | |
491 | [C(BPU)] = { | |
492 | [C(OP_READ)] = { | |
493 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
494 | [C(RESULT_MISS)] = { CACHE_OP_UNSUPPORTED }, | |
495 | }, | |
496 | [ C(OP_WRITE) ] = { | |
497 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
89d6c0b5 PZ |
498 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, |
499 | }, | |
500 | [ C(OP_PREFETCH) ] = { | |
501 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
502 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
503 | }, | |
504 | }, | |
505 | [C(NODE)] = { | |
506 | [C(OP_READ)] = { | |
507 | [C(RESULT_ACCESS)] = { CACHE_OP_UNSUPPORTED }, | |
508 | [C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
509 | }, | |
510 | [ C(OP_WRITE) ] = { | |
511 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
d0b86480 DM |
512 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, |
513 | }, | |
514 | [ C(OP_PREFETCH) ] = { | |
515 | [ C(RESULT_ACCESS) ] = { CACHE_OP_UNSUPPORTED }, | |
516 | [ C(RESULT_MISS) ] = { CACHE_OP_UNSUPPORTED }, | |
517 | }, | |
518 | }, | |
519 | }; | |
520 | ||
b73d8847 DM |
521 | static const struct sparc_pmu niagara2_pmu = { |
522 | .event_map = niagara2_event_map, | |
d0b86480 | 523 | .cache_map = &niagara2_cache_map, |
b73d8847 DM |
524 | .max_events = ARRAY_SIZE(niagara2_perfmon_event_map), |
525 | .upper_shift = 19, | |
526 | .lower_shift = 6, | |
527 | .event_mask = 0xfff, | |
528 | .hv_bit = 0x8, | |
de23cf3c | 529 | .irq_bit = 0x30, |
b73d8847 DM |
530 | .upper_nop = 0x220, |
531 | .lower_nop = 0x220, | |
532 | }; | |
533 | ||
59abbd1e DM |
534 | static const struct sparc_pmu *sparc_pmu __read_mostly; |
535 | ||
cdd6c482 | 536 | static u64 event_encoding(u64 event_id, int idx) |
59abbd1e DM |
537 | { |
538 | if (idx == PIC_UPPER_INDEX) | |
cdd6c482 | 539 | event_id <<= sparc_pmu->upper_shift; |
59abbd1e | 540 | else |
cdd6c482 IM |
541 | event_id <<= sparc_pmu->lower_shift; |
542 | return event_id; | |
59abbd1e DM |
543 | } |
544 | ||
545 | static u64 mask_for_index(int idx) | |
546 | { | |
547 | return event_encoding(sparc_pmu->event_mask, idx); | |
548 | } | |
549 | ||
550 | static u64 nop_for_index(int idx) | |
551 | { | |
552 | return event_encoding(idx == PIC_UPPER_INDEX ? | |
660d1376 DM |
553 | sparc_pmu->upper_nop : |
554 | sparc_pmu->lower_nop, idx); | |
59abbd1e DM |
555 | } |
556 | ||
d1751388 | 557 | static inline void sparc_pmu_enable_event(struct cpu_hw_events *cpuc, struct hw_perf_event *hwc, int idx) |
59abbd1e DM |
558 | { |
559 | u64 val, mask = mask_for_index(idx); | |
560 | ||
d1751388 DM |
561 | val = cpuc->pcr; |
562 | val &= ~mask; | |
563 | val |= hwc->config; | |
564 | cpuc->pcr = val; | |
565 | ||
09d053c7 | 566 | pcr_ops->write_pcr(0, cpuc->pcr); |
59abbd1e DM |
567 | } |
568 | ||
d1751388 | 569 | static inline void sparc_pmu_disable_event(struct cpu_hw_events *cpuc, struct hw_perf_event *hwc, int idx) |
59abbd1e DM |
570 | { |
571 | u64 mask = mask_for_index(idx); | |
572 | u64 nop = nop_for_index(idx); | |
d1751388 | 573 | u64 val; |
59abbd1e | 574 | |
d1751388 DM |
575 | val = cpuc->pcr; |
576 | val &= ~mask; | |
577 | val |= nop; | |
578 | cpuc->pcr = val; | |
579 | ||
09d053c7 | 580 | pcr_ops->write_pcr(0, cpuc->pcr); |
59abbd1e DM |
581 | } |
582 | ||
59abbd1e DM |
583 | static u32 read_pmc(int idx) |
584 | { | |
585 | u64 val; | |
586 | ||
09d053c7 | 587 | val = pcr_ops->read_pic(0); |
59abbd1e DM |
588 | if (idx == PIC_UPPER_INDEX) |
589 | val >>= 32; | |
590 | ||
591 | return val & 0xffffffff; | |
592 | } | |
593 | ||
594 | static void write_pmc(int idx, u64 val) | |
595 | { | |
596 | u64 shift, mask, pic; | |
597 | ||
598 | shift = 0; | |
599 | if (idx == PIC_UPPER_INDEX) | |
600 | shift = 32; | |
601 | ||
602 | mask = ((u64) 0xffffffff) << shift; | |
603 | val <<= shift; | |
604 | ||
09d053c7 | 605 | pic = pcr_ops->read_pic(0); |
59abbd1e DM |
606 | pic &= ~mask; |
607 | pic |= val; | |
09d053c7 | 608 | pcr_ops->write_pic(0, pic); |
59abbd1e DM |
609 | } |
610 | ||
e7bef6b0 DM |
611 | static u64 sparc_perf_event_update(struct perf_event *event, |
612 | struct hw_perf_event *hwc, int idx) | |
613 | { | |
614 | int shift = 64 - 32; | |
615 | u64 prev_raw_count, new_raw_count; | |
616 | s64 delta; | |
617 | ||
618 | again: | |
e7850595 | 619 | prev_raw_count = local64_read(&hwc->prev_count); |
e7bef6b0 DM |
620 | new_raw_count = read_pmc(idx); |
621 | ||
e7850595 | 622 | if (local64_cmpxchg(&hwc->prev_count, prev_raw_count, |
e7bef6b0 DM |
623 | new_raw_count) != prev_raw_count) |
624 | goto again; | |
625 | ||
626 | delta = (new_raw_count << shift) - (prev_raw_count << shift); | |
627 | delta >>= shift; | |
628 | ||
e7850595 PZ |
629 | local64_add(delta, &event->count); |
630 | local64_sub(delta, &hwc->period_left); | |
e7bef6b0 DM |
631 | |
632 | return new_raw_count; | |
633 | } | |
634 | ||
cdd6c482 | 635 | static int sparc_perf_event_set_period(struct perf_event *event, |
d29862f0 | 636 | struct hw_perf_event *hwc, int idx) |
59abbd1e | 637 | { |
e7850595 | 638 | s64 left = local64_read(&hwc->period_left); |
59abbd1e DM |
639 | s64 period = hwc->sample_period; |
640 | int ret = 0; | |
641 | ||
642 | if (unlikely(left <= -period)) { | |
643 | left = period; | |
e7850595 | 644 | local64_set(&hwc->period_left, left); |
59abbd1e DM |
645 | hwc->last_period = period; |
646 | ret = 1; | |
647 | } | |
648 | ||
649 | if (unlikely(left <= 0)) { | |
650 | left += period; | |
e7850595 | 651 | local64_set(&hwc->period_left, left); |
59abbd1e DM |
652 | hwc->last_period = period; |
653 | ret = 1; | |
654 | } | |
655 | if (left > MAX_PERIOD) | |
656 | left = MAX_PERIOD; | |
657 | ||
e7850595 | 658 | local64_set(&hwc->prev_count, (u64)-left); |
59abbd1e DM |
659 | |
660 | write_pmc(idx, (u64)(-left) & 0xffffffff); | |
661 | ||
cdd6c482 | 662 | perf_event_update_userpage(event); |
59abbd1e DM |
663 | |
664 | return ret; | |
665 | } | |
666 | ||
e7bef6b0 DM |
667 | /* If performance event entries have been added, move existing |
668 | * events around (if necessary) and then assign new entries to | |
669 | * counters. | |
670 | */ | |
671 | static u64 maybe_change_configuration(struct cpu_hw_events *cpuc, u64 pcr) | |
59abbd1e | 672 | { |
e7bef6b0 | 673 | int i; |
59abbd1e | 674 | |
e7bef6b0 DM |
675 | if (!cpuc->n_added) |
676 | goto out; | |
59abbd1e | 677 | |
e7bef6b0 DM |
678 | /* Read in the counters which are moving. */ |
679 | for (i = 0; i < cpuc->n_events; i++) { | |
680 | struct perf_event *cp = cpuc->event[i]; | |
59abbd1e | 681 | |
e7bef6b0 DM |
682 | if (cpuc->current_idx[i] != PIC_NO_INDEX && |
683 | cpuc->current_idx[i] != cp->hw.idx) { | |
684 | sparc_perf_event_update(cp, &cp->hw, | |
685 | cpuc->current_idx[i]); | |
686 | cpuc->current_idx[i] = PIC_NO_INDEX; | |
687 | } | |
688 | } | |
59abbd1e | 689 | |
e7bef6b0 DM |
690 | /* Assign to counters all unassigned events. */ |
691 | for (i = 0; i < cpuc->n_events; i++) { | |
692 | struct perf_event *cp = cpuc->event[i]; | |
693 | struct hw_perf_event *hwc = &cp->hw; | |
694 | int idx = hwc->idx; | |
695 | u64 enc; | |
696 | ||
697 | if (cpuc->current_idx[i] != PIC_NO_INDEX) | |
698 | continue; | |
699 | ||
700 | sparc_perf_event_set_period(cp, hwc, idx); | |
701 | cpuc->current_idx[i] = idx; | |
702 | ||
703 | enc = perf_event_get_enc(cpuc->events[i]); | |
b7d45c3f | 704 | pcr &= ~mask_for_index(idx); |
a4eaf7f1 PZ |
705 | if (hwc->state & PERF_HES_STOPPED) |
706 | pcr |= nop_for_index(idx); | |
707 | else | |
708 | pcr |= event_encoding(enc, idx); | |
e7bef6b0 DM |
709 | } |
710 | out: | |
711 | return pcr; | |
59abbd1e DM |
712 | } |
713 | ||
a4eaf7f1 | 714 | static void sparc_pmu_enable(struct pmu *pmu) |
59abbd1e | 715 | { |
e7bef6b0 DM |
716 | struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); |
717 | u64 pcr; | |
59abbd1e | 718 | |
e7bef6b0 DM |
719 | if (cpuc->enabled) |
720 | return; | |
59abbd1e | 721 | |
e7bef6b0 DM |
722 | cpuc->enabled = 1; |
723 | barrier(); | |
59abbd1e | 724 | |
e7bef6b0 DM |
725 | pcr = cpuc->pcr; |
726 | if (!cpuc->n_events) { | |
727 | pcr = 0; | |
728 | } else { | |
729 | pcr = maybe_change_configuration(cpuc, pcr); | |
59abbd1e | 730 | |
e7bef6b0 DM |
731 | /* We require that all of the events have the same |
732 | * configuration, so just fetch the settings from the | |
733 | * first entry. | |
734 | */ | |
735 | cpuc->pcr = pcr | cpuc->event[0]->hw.config_base; | |
736 | } | |
59abbd1e | 737 | |
09d053c7 | 738 | pcr_ops->write_pcr(0, cpuc->pcr); |
e7bef6b0 DM |
739 | } |
740 | ||
a4eaf7f1 | 741 | static void sparc_pmu_disable(struct pmu *pmu) |
e7bef6b0 DM |
742 | { |
743 | struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); | |
744 | u64 val; | |
745 | ||
746 | if (!cpuc->enabled) | |
747 | return; | |
748 | ||
749 | cpuc->enabled = 0; | |
750 | cpuc->n_added = 0; | |
751 | ||
752 | val = cpuc->pcr; | |
753 | val &= ~(PCR_UTRACE | PCR_STRACE | | |
754 | sparc_pmu->hv_bit | sparc_pmu->irq_bit); | |
755 | cpuc->pcr = val; | |
756 | ||
09d053c7 | 757 | pcr_ops->write_pcr(0, cpuc->pcr); |
59abbd1e DM |
758 | } |
759 | ||
a4eaf7f1 PZ |
760 | static int active_event_index(struct cpu_hw_events *cpuc, |
761 | struct perf_event *event) | |
762 | { | |
763 | int i; | |
764 | ||
765 | for (i = 0; i < cpuc->n_events; i++) { | |
766 | if (cpuc->event[i] == event) | |
767 | break; | |
768 | } | |
769 | BUG_ON(i == cpuc->n_events); | |
770 | return cpuc->current_idx[i]; | |
771 | } | |
772 | ||
773 | static void sparc_pmu_start(struct perf_event *event, int flags) | |
774 | { | |
775 | struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); | |
776 | int idx = active_event_index(cpuc, event); | |
777 | ||
778 | if (flags & PERF_EF_RELOAD) { | |
779 | WARN_ON_ONCE(!(event->hw.state & PERF_HES_UPTODATE)); | |
780 | sparc_perf_event_set_period(event, &event->hw, idx); | |
781 | } | |
782 | ||
783 | event->hw.state = 0; | |
784 | ||
785 | sparc_pmu_enable_event(cpuc, &event->hw, idx); | |
786 | } | |
787 | ||
788 | static void sparc_pmu_stop(struct perf_event *event, int flags) | |
789 | { | |
790 | struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); | |
791 | int idx = active_event_index(cpuc, event); | |
792 | ||
793 | if (!(event->hw.state & PERF_HES_STOPPED)) { | |
794 | sparc_pmu_disable_event(cpuc, &event->hw, idx); | |
795 | event->hw.state |= PERF_HES_STOPPED; | |
796 | } | |
797 | ||
798 | if (!(event->hw.state & PERF_HES_UPTODATE) && (flags & PERF_EF_UPDATE)) { | |
799 | sparc_perf_event_update(event, &event->hw, idx); | |
800 | event->hw.state |= PERF_HES_UPTODATE; | |
801 | } | |
802 | } | |
803 | ||
804 | static void sparc_pmu_del(struct perf_event *event, int _flags) | |
59abbd1e | 805 | { |
cdd6c482 | 806 | struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); |
e7bef6b0 DM |
807 | unsigned long flags; |
808 | int i; | |
59abbd1e | 809 | |
e7bef6b0 | 810 | local_irq_save(flags); |
33696fc0 | 811 | perf_pmu_disable(event->pmu); |
e7bef6b0 DM |
812 | |
813 | for (i = 0; i < cpuc->n_events; i++) { | |
814 | if (event == cpuc->event[i]) { | |
a4eaf7f1 PZ |
815 | /* Absorb the final count and turn off the |
816 | * event. | |
817 | */ | |
818 | sparc_pmu_stop(event, PERF_EF_UPDATE); | |
e7bef6b0 DM |
819 | |
820 | /* Shift remaining entries down into | |
821 | * the existing slot. | |
822 | */ | |
823 | while (++i < cpuc->n_events) { | |
824 | cpuc->event[i - 1] = cpuc->event[i]; | |
825 | cpuc->events[i - 1] = cpuc->events[i]; | |
826 | cpuc->current_idx[i - 1] = | |
827 | cpuc->current_idx[i]; | |
828 | } | |
829 | ||
e7bef6b0 | 830 | perf_event_update_userpage(event); |
59abbd1e | 831 | |
e7bef6b0 DM |
832 | cpuc->n_events--; |
833 | break; | |
834 | } | |
835 | } | |
59abbd1e | 836 | |
33696fc0 | 837 | perf_pmu_enable(event->pmu); |
e7bef6b0 DM |
838 | local_irq_restore(flags); |
839 | } | |
840 | ||
cdd6c482 | 841 | static void sparc_pmu_read(struct perf_event *event) |
59abbd1e | 842 | { |
e7bef6b0 DM |
843 | struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); |
844 | int idx = active_event_index(cpuc, event); | |
cdd6c482 | 845 | struct hw_perf_event *hwc = &event->hw; |
d1751388 | 846 | |
e7bef6b0 | 847 | sparc_perf_event_update(event, hwc, idx); |
59abbd1e DM |
848 | } |
849 | ||
cdd6c482 | 850 | static atomic_t active_events = ATOMIC_INIT(0); |
59abbd1e DM |
851 | static DEFINE_MUTEX(pmc_grab_mutex); |
852 | ||
d1751388 DM |
853 | static void perf_stop_nmi_watchdog(void *unused) |
854 | { | |
855 | struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); | |
856 | ||
857 | stop_nmi_watchdog(NULL); | |
09d053c7 | 858 | cpuc->pcr = pcr_ops->read_pcr(0); |
d1751388 DM |
859 | } |
860 | ||
cdd6c482 | 861 | void perf_event_grab_pmc(void) |
59abbd1e | 862 | { |
cdd6c482 | 863 | if (atomic_inc_not_zero(&active_events)) |
59abbd1e DM |
864 | return; |
865 | ||
866 | mutex_lock(&pmc_grab_mutex); | |
cdd6c482 | 867 | if (atomic_read(&active_events) == 0) { |
59abbd1e | 868 | if (atomic_read(&nmi_active) > 0) { |
d1751388 | 869 | on_each_cpu(perf_stop_nmi_watchdog, NULL, 1); |
59abbd1e DM |
870 | BUG_ON(atomic_read(&nmi_active) != 0); |
871 | } | |
cdd6c482 | 872 | atomic_inc(&active_events); |
59abbd1e DM |
873 | } |
874 | mutex_unlock(&pmc_grab_mutex); | |
875 | } | |
876 | ||
cdd6c482 | 877 | void perf_event_release_pmc(void) |
59abbd1e | 878 | { |
cdd6c482 | 879 | if (atomic_dec_and_mutex_lock(&active_events, &pmc_grab_mutex)) { |
59abbd1e DM |
880 | if (atomic_read(&nmi_active) == 0) |
881 | on_each_cpu(start_nmi_watchdog, NULL, 1); | |
882 | mutex_unlock(&pmc_grab_mutex); | |
883 | } | |
884 | } | |
885 | ||
2ce4da2e DM |
886 | static const struct perf_event_map *sparc_map_cache_event(u64 config) |
887 | { | |
888 | unsigned int cache_type, cache_op, cache_result; | |
889 | const struct perf_event_map *pmap; | |
890 | ||
891 | if (!sparc_pmu->cache_map) | |
892 | return ERR_PTR(-ENOENT); | |
893 | ||
894 | cache_type = (config >> 0) & 0xff; | |
895 | if (cache_type >= PERF_COUNT_HW_CACHE_MAX) | |
896 | return ERR_PTR(-EINVAL); | |
897 | ||
898 | cache_op = (config >> 8) & 0xff; | |
899 | if (cache_op >= PERF_COUNT_HW_CACHE_OP_MAX) | |
900 | return ERR_PTR(-EINVAL); | |
901 | ||
902 | cache_result = (config >> 16) & 0xff; | |
903 | if (cache_result >= PERF_COUNT_HW_CACHE_RESULT_MAX) | |
904 | return ERR_PTR(-EINVAL); | |
905 | ||
906 | pmap = &((*sparc_pmu->cache_map)[cache_type][cache_op][cache_result]); | |
907 | ||
908 | if (pmap->encoding == CACHE_OP_UNSUPPORTED) | |
909 | return ERR_PTR(-ENOENT); | |
910 | ||
911 | if (pmap->encoding == CACHE_OP_NONSENSE) | |
912 | return ERR_PTR(-EINVAL); | |
913 | ||
914 | return pmap; | |
915 | } | |
916 | ||
cdd6c482 | 917 | static void hw_perf_event_destroy(struct perf_event *event) |
59abbd1e | 918 | { |
cdd6c482 | 919 | perf_event_release_pmc(); |
59abbd1e DM |
920 | } |
921 | ||
a72a8a5f DM |
922 | /* Make sure all events can be scheduled into the hardware at |
923 | * the same time. This is simplified by the fact that we only | |
924 | * need to support 2 simultaneous HW events. | |
e7bef6b0 DM |
925 | * |
926 | * As a side effect, the evts[]->hw.idx values will be assigned | |
927 | * on success. These are pending indexes. When the events are | |
928 | * actually programmed into the chip, these values will propagate | |
929 | * to the per-cpu cpuc->current_idx[] slots, see the code in | |
930 | * maybe_change_configuration() for details. | |
a72a8a5f | 931 | */ |
e7bef6b0 DM |
932 | static int sparc_check_constraints(struct perf_event **evts, |
933 | unsigned long *events, int n_ev) | |
a72a8a5f | 934 | { |
e7bef6b0 DM |
935 | u8 msk0 = 0, msk1 = 0; |
936 | int idx0 = 0; | |
937 | ||
938 | /* This case is possible when we are invoked from | |
939 | * hw_perf_group_sched_in(). | |
940 | */ | |
941 | if (!n_ev) | |
942 | return 0; | |
943 | ||
15ac9a39 | 944 | if (n_ev > MAX_HWEVENTS) |
e7bef6b0 DM |
945 | return -1; |
946 | ||
947 | msk0 = perf_event_get_msk(events[0]); | |
948 | if (n_ev == 1) { | |
949 | if (msk0 & PIC_LOWER) | |
950 | idx0 = 1; | |
951 | goto success; | |
952 | } | |
953 | BUG_ON(n_ev != 2); | |
954 | msk1 = perf_event_get_msk(events[1]); | |
955 | ||
956 | /* If both events can go on any counter, OK. */ | |
957 | if (msk0 == (PIC_UPPER | PIC_LOWER) && | |
958 | msk1 == (PIC_UPPER | PIC_LOWER)) | |
959 | goto success; | |
960 | ||
961 | /* If one event is limited to a specific counter, | |
962 | * and the other can go on both, OK. | |
963 | */ | |
964 | if ((msk0 == PIC_UPPER || msk0 == PIC_LOWER) && | |
965 | msk1 == (PIC_UPPER | PIC_LOWER)) { | |
966 | if (msk0 & PIC_LOWER) | |
967 | idx0 = 1; | |
968 | goto success; | |
a72a8a5f DM |
969 | } |
970 | ||
e7bef6b0 DM |
971 | if ((msk1 == PIC_UPPER || msk1 == PIC_LOWER) && |
972 | msk0 == (PIC_UPPER | PIC_LOWER)) { | |
973 | if (msk1 & PIC_UPPER) | |
974 | idx0 = 1; | |
975 | goto success; | |
976 | } | |
977 | ||
978 | /* If the events are fixed to different counters, OK. */ | |
979 | if ((msk0 == PIC_UPPER && msk1 == PIC_LOWER) || | |
980 | (msk0 == PIC_LOWER && msk1 == PIC_UPPER)) { | |
981 | if (msk0 & PIC_LOWER) | |
982 | idx0 = 1; | |
983 | goto success; | |
984 | } | |
985 | ||
986 | /* Otherwise, there is a conflict. */ | |
a72a8a5f | 987 | return -1; |
e7bef6b0 DM |
988 | |
989 | success: | |
990 | evts[0]->hw.idx = idx0; | |
991 | if (n_ev == 2) | |
992 | evts[1]->hw.idx = idx0 ^ 1; | |
993 | return 0; | |
a72a8a5f DM |
994 | } |
995 | ||
01552f76 DM |
996 | static int check_excludes(struct perf_event **evts, int n_prev, int n_new) |
997 | { | |
998 | int eu = 0, ek = 0, eh = 0; | |
999 | struct perf_event *event; | |
1000 | int i, n, first; | |
1001 | ||
1002 | n = n_prev + n_new; | |
1003 | if (n <= 1) | |
1004 | return 0; | |
1005 | ||
1006 | first = 1; | |
1007 | for (i = 0; i < n; i++) { | |
1008 | event = evts[i]; | |
1009 | if (first) { | |
1010 | eu = event->attr.exclude_user; | |
1011 | ek = event->attr.exclude_kernel; | |
1012 | eh = event->attr.exclude_hv; | |
1013 | first = 0; | |
1014 | } else if (event->attr.exclude_user != eu || | |
1015 | event->attr.exclude_kernel != ek || | |
1016 | event->attr.exclude_hv != eh) { | |
1017 | return -EAGAIN; | |
1018 | } | |
1019 | } | |
1020 | ||
1021 | return 0; | |
1022 | } | |
1023 | ||
1024 | static int collect_events(struct perf_event *group, int max_count, | |
e7bef6b0 DM |
1025 | struct perf_event *evts[], unsigned long *events, |
1026 | int *current_idx) | |
01552f76 DM |
1027 | { |
1028 | struct perf_event *event; | |
1029 | int n = 0; | |
1030 | ||
1031 | if (!is_software_event(group)) { | |
1032 | if (n >= max_count) | |
1033 | return -1; | |
1034 | evts[n] = group; | |
e7bef6b0 DM |
1035 | events[n] = group->hw.event_base; |
1036 | current_idx[n++] = PIC_NO_INDEX; | |
01552f76 DM |
1037 | } |
1038 | list_for_each_entry(event, &group->sibling_list, group_entry) { | |
1039 | if (!is_software_event(event) && | |
1040 | event->state != PERF_EVENT_STATE_OFF) { | |
1041 | if (n >= max_count) | |
1042 | return -1; | |
1043 | evts[n] = event; | |
e7bef6b0 DM |
1044 | events[n] = event->hw.event_base; |
1045 | current_idx[n++] = PIC_NO_INDEX; | |
01552f76 DM |
1046 | } |
1047 | } | |
1048 | return n; | |
1049 | } | |
1050 | ||
a4eaf7f1 | 1051 | static int sparc_pmu_add(struct perf_event *event, int ef_flags) |
e7bef6b0 DM |
1052 | { |
1053 | struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); | |
1054 | int n0, ret = -EAGAIN; | |
1055 | unsigned long flags; | |
1056 | ||
1057 | local_irq_save(flags); | |
33696fc0 | 1058 | perf_pmu_disable(event->pmu); |
e7bef6b0 DM |
1059 | |
1060 | n0 = cpuc->n_events; | |
15ac9a39 | 1061 | if (n0 >= MAX_HWEVENTS) |
e7bef6b0 DM |
1062 | goto out; |
1063 | ||
1064 | cpuc->event[n0] = event; | |
1065 | cpuc->events[n0] = event->hw.event_base; | |
1066 | cpuc->current_idx[n0] = PIC_NO_INDEX; | |
1067 | ||
a4eaf7f1 PZ |
1068 | event->hw.state = PERF_HES_UPTODATE; |
1069 | if (!(ef_flags & PERF_EF_START)) | |
1070 | event->hw.state |= PERF_HES_STOPPED; | |
1071 | ||
a13c3afd LM |
1072 | /* |
1073 | * If group events scheduling transaction was started, | |
25985edc | 1074 | * skip the schedulability test here, it will be performed |
a13c3afd LM |
1075 | * at commit time(->commit_txn) as a whole |
1076 | */ | |
8d2cacbb | 1077 | if (cpuc->group_flag & PERF_EVENT_TXN) |
a13c3afd LM |
1078 | goto nocheck; |
1079 | ||
e7bef6b0 DM |
1080 | if (check_excludes(cpuc->event, n0, 1)) |
1081 | goto out; | |
1082 | if (sparc_check_constraints(cpuc->event, cpuc->events, n0 + 1)) | |
1083 | goto out; | |
1084 | ||
a13c3afd | 1085 | nocheck: |
e7bef6b0 DM |
1086 | cpuc->n_events++; |
1087 | cpuc->n_added++; | |
1088 | ||
1089 | ret = 0; | |
1090 | out: | |
33696fc0 | 1091 | perf_pmu_enable(event->pmu); |
e7bef6b0 DM |
1092 | local_irq_restore(flags); |
1093 | return ret; | |
1094 | } | |
1095 | ||
b0a873eb | 1096 | static int sparc_pmu_event_init(struct perf_event *event) |
59abbd1e | 1097 | { |
cdd6c482 | 1098 | struct perf_event_attr *attr = &event->attr; |
01552f76 | 1099 | struct perf_event *evts[MAX_HWEVENTS]; |
cdd6c482 | 1100 | struct hw_perf_event *hwc = &event->hw; |
a72a8a5f | 1101 | unsigned long events[MAX_HWEVENTS]; |
e7bef6b0 | 1102 | int current_idx_dmy[MAX_HWEVENTS]; |
59abbd1e | 1103 | const struct perf_event_map *pmap; |
01552f76 | 1104 | int n; |
59abbd1e DM |
1105 | |
1106 | if (atomic_read(&nmi_active) < 0) | |
1107 | return -ENODEV; | |
1108 | ||
2481c5fa SE |
1109 | /* does not support taken branch sampling */ |
1110 | if (has_branch_stack(event)) | |
1111 | return -EOPNOTSUPP; | |
1112 | ||
b0a873eb PZ |
1113 | switch (attr->type) { |
1114 | case PERF_TYPE_HARDWARE: | |
2ce4da2e DM |
1115 | if (attr->config >= sparc_pmu->max_events) |
1116 | return -EINVAL; | |
1117 | pmap = sparc_pmu->event_map(attr->config); | |
b0a873eb PZ |
1118 | break; |
1119 | ||
1120 | case PERF_TYPE_HW_CACHE: | |
2ce4da2e DM |
1121 | pmap = sparc_map_cache_event(attr->config); |
1122 | if (IS_ERR(pmap)) | |
1123 | return PTR_ERR(pmap); | |
b0a873eb PZ |
1124 | break; |
1125 | ||
1126 | case PERF_TYPE_RAW: | |
d0303d71 IM |
1127 | pmap = NULL; |
1128 | break; | |
59abbd1e | 1129 | |
b0a873eb PZ |
1130 | default: |
1131 | return -ENOENT; | |
1132 | ||
1133 | } | |
1134 | ||
b343ae51 DM |
1135 | if (pmap) { |
1136 | hwc->event_base = perf_event_encode(pmap); | |
1137 | } else { | |
d0303d71 IM |
1138 | /* |
1139 | * User gives us "(encoding << 16) | pic_mask" for | |
b343ae51 DM |
1140 | * PERF_TYPE_RAW events. |
1141 | */ | |
1142 | hwc->event_base = attr->config; | |
1143 | } | |
1144 | ||
e7bef6b0 | 1145 | /* We save the enable bits in the config_base. */ |
496c07e3 | 1146 | hwc->config_base = sparc_pmu->irq_bit; |
59abbd1e DM |
1147 | if (!attr->exclude_user) |
1148 | hwc->config_base |= PCR_UTRACE; | |
1149 | if (!attr->exclude_kernel) | |
1150 | hwc->config_base |= PCR_STRACE; | |
91b9286d DM |
1151 | if (!attr->exclude_hv) |
1152 | hwc->config_base |= sparc_pmu->hv_bit; | |
59abbd1e | 1153 | |
01552f76 DM |
1154 | n = 0; |
1155 | if (event->group_leader != event) { | |
1156 | n = collect_events(event->group_leader, | |
15ac9a39 | 1157 | MAX_HWEVENTS - 1, |
e7bef6b0 | 1158 | evts, events, current_idx_dmy); |
01552f76 DM |
1159 | if (n < 0) |
1160 | return -EINVAL; | |
1161 | } | |
a72a8a5f | 1162 | events[n] = hwc->event_base; |
01552f76 DM |
1163 | evts[n] = event; |
1164 | ||
1165 | if (check_excludes(evts, n, 1)) | |
1166 | return -EINVAL; | |
1167 | ||
e7bef6b0 | 1168 | if (sparc_check_constraints(evts, events, n + 1)) |
a72a8a5f DM |
1169 | return -EINVAL; |
1170 | ||
e7bef6b0 DM |
1171 | hwc->idx = PIC_NO_INDEX; |
1172 | ||
01552f76 DM |
1173 | /* Try to do all error checking before this point, as unwinding |
1174 | * state after grabbing the PMC is difficult. | |
1175 | */ | |
1176 | perf_event_grab_pmc(); | |
1177 | event->destroy = hw_perf_event_destroy; | |
1178 | ||
59abbd1e DM |
1179 | if (!hwc->sample_period) { |
1180 | hwc->sample_period = MAX_PERIOD; | |
1181 | hwc->last_period = hwc->sample_period; | |
e7850595 | 1182 | local64_set(&hwc->period_left, hwc->sample_period); |
59abbd1e DM |
1183 | } |
1184 | ||
59abbd1e DM |
1185 | return 0; |
1186 | } | |
1187 | ||
a13c3afd LM |
1188 | /* |
1189 | * Start group events scheduling transaction | |
1190 | * Set the flag to make pmu::enable() not perform the | |
1191 | * schedulability test, it will be performed at commit time | |
1192 | */ | |
51b0fe39 | 1193 | static void sparc_pmu_start_txn(struct pmu *pmu) |
a13c3afd LM |
1194 | { |
1195 | struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events); | |
1196 | ||
33696fc0 | 1197 | perf_pmu_disable(pmu); |
8d2cacbb | 1198 | cpuhw->group_flag |= PERF_EVENT_TXN; |
a13c3afd LM |
1199 | } |
1200 | ||
1201 | /* | |
1202 | * Stop group events scheduling transaction | |
1203 | * Clear the flag and pmu::enable() will perform the | |
1204 | * schedulability test. | |
1205 | */ | |
51b0fe39 | 1206 | static void sparc_pmu_cancel_txn(struct pmu *pmu) |
a13c3afd LM |
1207 | { |
1208 | struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events); | |
1209 | ||
8d2cacbb | 1210 | cpuhw->group_flag &= ~PERF_EVENT_TXN; |
33696fc0 | 1211 | perf_pmu_enable(pmu); |
a13c3afd LM |
1212 | } |
1213 | ||
1214 | /* | |
1215 | * Commit group events scheduling transaction | |
1216 | * Perform the group schedulability test as a whole | |
1217 | * Return 0 if success | |
1218 | */ | |
51b0fe39 | 1219 | static int sparc_pmu_commit_txn(struct pmu *pmu) |
a13c3afd LM |
1220 | { |
1221 | struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); | |
1222 | int n; | |
1223 | ||
1224 | if (!sparc_pmu) | |
1225 | return -EINVAL; | |
1226 | ||
1227 | cpuc = &__get_cpu_var(cpu_hw_events); | |
1228 | n = cpuc->n_events; | |
1229 | if (check_excludes(cpuc->event, 0, n)) | |
1230 | return -EINVAL; | |
1231 | if (sparc_check_constraints(cpuc->event, cpuc->events, n)) | |
1232 | return -EAGAIN; | |
1233 | ||
8d2cacbb | 1234 | cpuc->group_flag &= ~PERF_EVENT_TXN; |
33696fc0 | 1235 | perf_pmu_enable(pmu); |
a13c3afd LM |
1236 | return 0; |
1237 | } | |
1238 | ||
51b0fe39 | 1239 | static struct pmu pmu = { |
a4eaf7f1 PZ |
1240 | .pmu_enable = sparc_pmu_enable, |
1241 | .pmu_disable = sparc_pmu_disable, | |
b0a873eb | 1242 | .event_init = sparc_pmu_event_init, |
a4eaf7f1 PZ |
1243 | .add = sparc_pmu_add, |
1244 | .del = sparc_pmu_del, | |
1245 | .start = sparc_pmu_start, | |
1246 | .stop = sparc_pmu_stop, | |
59abbd1e | 1247 | .read = sparc_pmu_read, |
a13c3afd LM |
1248 | .start_txn = sparc_pmu_start_txn, |
1249 | .cancel_txn = sparc_pmu_cancel_txn, | |
1250 | .commit_txn = sparc_pmu_commit_txn, | |
59abbd1e DM |
1251 | }; |
1252 | ||
cdd6c482 | 1253 | void perf_event_print_debug(void) |
59abbd1e DM |
1254 | { |
1255 | unsigned long flags; | |
1256 | u64 pcr, pic; | |
1257 | int cpu; | |
1258 | ||
1259 | if (!sparc_pmu) | |
1260 | return; | |
1261 | ||
1262 | local_irq_save(flags); | |
1263 | ||
1264 | cpu = smp_processor_id(); | |
1265 | ||
09d053c7 DM |
1266 | pcr = pcr_ops->read_pcr(0); |
1267 | pic = pcr_ops->read_pic(0); | |
59abbd1e DM |
1268 | |
1269 | pr_info("\n"); | |
1270 | pr_info("CPU#%d: PCR[%016llx] PIC[%016llx]\n", | |
1271 | cpu, pcr, pic); | |
1272 | ||
1273 | local_irq_restore(flags); | |
1274 | } | |
1275 | ||
cdd6c482 | 1276 | static int __kprobes perf_event_nmi_handler(struct notifier_block *self, |
d29862f0 | 1277 | unsigned long cmd, void *__args) |
59abbd1e DM |
1278 | { |
1279 | struct die_args *args = __args; | |
1280 | struct perf_sample_data data; | |
cdd6c482 | 1281 | struct cpu_hw_events *cpuc; |
59abbd1e | 1282 | struct pt_regs *regs; |
e7bef6b0 | 1283 | int i; |
59abbd1e | 1284 | |
cdd6c482 | 1285 | if (!atomic_read(&active_events)) |
59abbd1e DM |
1286 | return NOTIFY_DONE; |
1287 | ||
1288 | switch (cmd) { | |
1289 | case DIE_NMI: | |
1290 | break; | |
1291 | ||
1292 | default: | |
1293 | return NOTIFY_DONE; | |
1294 | } | |
1295 | ||
1296 | regs = args->regs; | |
1297 | ||
cdd6c482 | 1298 | cpuc = &__get_cpu_var(cpu_hw_events); |
e04ed38d DM |
1299 | |
1300 | /* If the PMU has the TOE IRQ enable bits, we need to do a | |
1301 | * dummy write to the %pcr to clear the overflow bits and thus | |
1302 | * the interrupt. | |
1303 | * | |
1304 | * Do this before we peek at the counters to determine | |
1305 | * overflow so we don't lose any events. | |
1306 | */ | |
1307 | if (sparc_pmu->irq_bit) | |
09d053c7 | 1308 | pcr_ops->write_pcr(0, cpuc->pcr); |
e04ed38d | 1309 | |
e7bef6b0 DM |
1310 | for (i = 0; i < cpuc->n_events; i++) { |
1311 | struct perf_event *event = cpuc->event[i]; | |
1312 | int idx = cpuc->current_idx[i]; | |
cdd6c482 | 1313 | struct hw_perf_event *hwc; |
59abbd1e DM |
1314 | u64 val; |
1315 | ||
cdd6c482 IM |
1316 | hwc = &event->hw; |
1317 | val = sparc_perf_event_update(event, hwc, idx); | |
59abbd1e DM |
1318 | if (val & (1ULL << 31)) |
1319 | continue; | |
1320 | ||
fd0d000b | 1321 | perf_sample_data_init(&data, 0, hwc->last_period); |
cdd6c482 | 1322 | if (!sparc_perf_event_set_period(event, hwc, idx)) |
59abbd1e DM |
1323 | continue; |
1324 | ||
a8b0ca17 | 1325 | if (perf_event_overflow(event, &data, regs)) |
a4eaf7f1 | 1326 | sparc_pmu_stop(event, 0); |
59abbd1e DM |
1327 | } |
1328 | ||
1329 | return NOTIFY_STOP; | |
1330 | } | |
1331 | ||
cdd6c482 IM |
1332 | static __read_mostly struct notifier_block perf_event_nmi_notifier = { |
1333 | .notifier_call = perf_event_nmi_handler, | |
59abbd1e DM |
1334 | }; |
1335 | ||
1336 | static bool __init supported_pmu(void) | |
1337 | { | |
28e8f9be DM |
1338 | if (!strcmp(sparc_pmu_type, "ultra3") || |
1339 | !strcmp(sparc_pmu_type, "ultra3+") || | |
1340 | !strcmp(sparc_pmu_type, "ultra3i") || | |
1341 | !strcmp(sparc_pmu_type, "ultra4+")) { | |
1342 | sparc_pmu = &ultra3_pmu; | |
59abbd1e DM |
1343 | return true; |
1344 | } | |
7eebda60 DM |
1345 | if (!strcmp(sparc_pmu_type, "niagara")) { |
1346 | sparc_pmu = &niagara1_pmu; | |
1347 | return true; | |
1348 | } | |
4ba991d3 DM |
1349 | if (!strcmp(sparc_pmu_type, "niagara2") || |
1350 | !strcmp(sparc_pmu_type, "niagara3")) { | |
b73d8847 DM |
1351 | sparc_pmu = &niagara2_pmu; |
1352 | return true; | |
1353 | } | |
59abbd1e DM |
1354 | return false; |
1355 | } | |
1356 | ||
004417a6 | 1357 | int __init init_hw_perf_events(void) |
59abbd1e | 1358 | { |
cdd6c482 | 1359 | pr_info("Performance events: "); |
59abbd1e DM |
1360 | |
1361 | if (!supported_pmu()) { | |
1362 | pr_cont("No support for PMU type '%s'\n", sparc_pmu_type); | |
004417a6 | 1363 | return 0; |
59abbd1e DM |
1364 | } |
1365 | ||
1366 | pr_cont("Supported PMU type is '%s'\n", sparc_pmu_type); | |
1367 | ||
2e80a82a | 1368 | perf_pmu_register(&pmu, "cpu", PERF_TYPE_RAW); |
cdd6c482 | 1369 | register_die_notifier(&perf_event_nmi_notifier); |
004417a6 PZ |
1370 | |
1371 | return 0; | |
59abbd1e | 1372 | } |
efc70d24 | 1373 | early_initcall(init_hw_perf_events); |
4f6dbe4a | 1374 | |
56962b44 FW |
1375 | void perf_callchain_kernel(struct perf_callchain_entry *entry, |
1376 | struct pt_regs *regs) | |
4f6dbe4a DM |
1377 | { |
1378 | unsigned long ksp, fp; | |
667f0cee DM |
1379 | #ifdef CONFIG_FUNCTION_GRAPH_TRACER |
1380 | int graph = 0; | |
1381 | #endif | |
4f6dbe4a | 1382 | |
56962b44 FW |
1383 | stack_trace_flush(); |
1384 | ||
70791ce9 | 1385 | perf_callchain_store(entry, regs->tpc); |
4f6dbe4a DM |
1386 | |
1387 | ksp = regs->u_regs[UREG_I6]; | |
1388 | fp = ksp + STACK_BIAS; | |
1389 | do { | |
1390 | struct sparc_stackf *sf; | |
1391 | struct pt_regs *regs; | |
1392 | unsigned long pc; | |
1393 | ||
1394 | if (!kstack_valid(current_thread_info(), fp)) | |
1395 | break; | |
1396 | ||
1397 | sf = (struct sparc_stackf *) fp; | |
1398 | regs = (struct pt_regs *) (sf + 1); | |
1399 | ||
1400 | if (kstack_is_trap_frame(current_thread_info(), regs)) { | |
1401 | if (user_mode(regs)) | |
1402 | break; | |
1403 | pc = regs->tpc; | |
1404 | fp = regs->u_regs[UREG_I6] + STACK_BIAS; | |
1405 | } else { | |
1406 | pc = sf->callers_pc; | |
1407 | fp = (unsigned long)sf->fp + STACK_BIAS; | |
1408 | } | |
70791ce9 | 1409 | perf_callchain_store(entry, pc); |
667f0cee DM |
1410 | #ifdef CONFIG_FUNCTION_GRAPH_TRACER |
1411 | if ((pc + 8UL) == (unsigned long) &return_to_handler) { | |
1412 | int index = current->curr_ret_stack; | |
1413 | if (current->ret_stack && index >= graph) { | |
1414 | pc = current->ret_stack[index - graph].ret; | |
70791ce9 | 1415 | perf_callchain_store(entry, pc); |
667f0cee DM |
1416 | graph++; |
1417 | } | |
1418 | } | |
1419 | #endif | |
4f6dbe4a DM |
1420 | } while (entry->nr < PERF_MAX_STACK_DEPTH); |
1421 | } | |
1422 | ||
56962b44 FW |
1423 | static void perf_callchain_user_64(struct perf_callchain_entry *entry, |
1424 | struct pt_regs *regs) | |
4f6dbe4a DM |
1425 | { |
1426 | unsigned long ufp; | |
1427 | ||
70791ce9 | 1428 | perf_callchain_store(entry, regs->tpc); |
4f6dbe4a DM |
1429 | |
1430 | ufp = regs->u_regs[UREG_I6] + STACK_BIAS; | |
1431 | do { | |
1432 | struct sparc_stackf *usf, sf; | |
1433 | unsigned long pc; | |
1434 | ||
1435 | usf = (struct sparc_stackf *) ufp; | |
1436 | if (__copy_from_user_inatomic(&sf, usf, sizeof(sf))) | |
1437 | break; | |
1438 | ||
1439 | pc = sf.callers_pc; | |
1440 | ufp = (unsigned long)sf.fp + STACK_BIAS; | |
70791ce9 | 1441 | perf_callchain_store(entry, pc); |
4f6dbe4a DM |
1442 | } while (entry->nr < PERF_MAX_STACK_DEPTH); |
1443 | } | |
1444 | ||
56962b44 FW |
1445 | static void perf_callchain_user_32(struct perf_callchain_entry *entry, |
1446 | struct pt_regs *regs) | |
4f6dbe4a DM |
1447 | { |
1448 | unsigned long ufp; | |
1449 | ||
70791ce9 | 1450 | perf_callchain_store(entry, regs->tpc); |
4f6dbe4a | 1451 | |
9e8307ec | 1452 | ufp = regs->u_regs[UREG_I6] & 0xffffffffUL; |
4f6dbe4a DM |
1453 | do { |
1454 | struct sparc_stackf32 *usf, sf; | |
1455 | unsigned long pc; | |
1456 | ||
1457 | usf = (struct sparc_stackf32 *) ufp; | |
1458 | if (__copy_from_user_inatomic(&sf, usf, sizeof(sf))) | |
1459 | break; | |
1460 | ||
1461 | pc = sf.callers_pc; | |
1462 | ufp = (unsigned long)sf.fp; | |
70791ce9 | 1463 | perf_callchain_store(entry, pc); |
4f6dbe4a DM |
1464 | } while (entry->nr < PERF_MAX_STACK_DEPTH); |
1465 | } | |
1466 | ||
56962b44 FW |
1467 | void |
1468 | perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs) | |
4f6dbe4a | 1469 | { |
56962b44 FW |
1470 | flushw_user(); |
1471 | if (test_thread_flag(TIF_32BIT)) | |
1472 | perf_callchain_user_32(entry, regs); | |
1473 | else | |
1474 | perf_callchain_user_64(entry, regs); | |
4f6dbe4a | 1475 | } |