Commit | Line | Data |
---|---|---|
eee3af4a MM |
1 | /* |
2 | * Debug Store support | |
3 | * | |
4 | * This provides a low-level interface to the hardware's Debug Store | |
5 | * feature that is used for last branch recording (LBR) and | |
6 | * precise-event based sampling (PEBS). | |
7 | * | |
8 | * Different architectures use a different DS layout/pointer size. | |
9 | * The below functions therefore work on a void*. | |
10 | * | |
11 | * | |
12 | * Since there is no user for PEBS, yet, only LBR (or branch | |
13 | * trace store, BTS) is supported. | |
14 | * | |
15 | * | |
16 | * Copyright (C) 2007 Intel Corporation. | |
17 | * Markus Metzger <markus.t.metzger@intel.com>, Dec 2007 | |
18 | */ | |
19 | ||
20 | #include <asm/ds.h> | |
21 | ||
22 | #include <linux/errno.h> | |
23 | #include <linux/string.h> | |
24 | #include <linux/slab.h> | |
25 | ||
26 | ||
27 | /* | |
28 | * Debug Store (DS) save area configuration (see Intel64 and IA32 | |
29 | * Architectures Software Developer's Manual, section 18.5) | |
30 | * | |
31 | * The DS configuration consists of the following fields; different | |
32 | * architetures vary in the size of those fields. | |
33 | * - double-word aligned base linear address of the BTS buffer | |
34 | * - write pointer into the BTS buffer | |
35 | * - end linear address of the BTS buffer (one byte beyond the end of | |
36 | * the buffer) | |
37 | * - interrupt pointer into BTS buffer | |
38 | * (interrupt occurs when write pointer passes interrupt pointer) | |
39 | * - double-word aligned base linear address of the PEBS buffer | |
40 | * - write pointer into the PEBS buffer | |
41 | * - end linear address of the PEBS buffer (one byte beyond the end of | |
42 | * the buffer) | |
43 | * - interrupt pointer into PEBS buffer | |
44 | * (interrupt occurs when write pointer passes interrupt pointer) | |
45 | * - value to which counter is reset following counter overflow | |
46 | * | |
47 | * On later architectures, the last branch recording hardware uses | |
48 | * 64bit pointers even in 32bit mode. | |
49 | * | |
50 | * | |
51 | * Branch Trace Store (BTS) records store information about control | |
52 | * flow changes. They at least provide the following information: | |
53 | * - source linear address | |
54 | * - destination linear address | |
55 | * | |
56 | * Netburst supported a predicated bit that had been dropped in later | |
57 | * architectures. We do not suppor it. | |
58 | * | |
59 | * | |
60 | * In order to abstract from the actual DS and BTS layout, we describe | |
61 | * the access to the relevant fields. | |
62 | * Thanks to Andi Kleen for proposing this design. | |
63 | * | |
64 | * The implementation, however, is not as general as it might seem. In | |
65 | * order to stay somewhat simple and efficient, we assume an | |
66 | * underlying unsigned type (mostly a pointer type) and we expect the | |
67 | * field to be at least as big as that type. | |
68 | */ | |
69 | ||
70 | /* | |
71 | * A special from_ip address to indicate that the BTS record is an | |
72 | * info record that needs to be interpreted or skipped. | |
73 | */ | |
74 | #define BTS_ESCAPE_ADDRESS (-1) | |
75 | ||
76 | /* | |
77 | * A field access descriptor | |
78 | */ | |
79 | struct access_desc { | |
80 | unsigned char offset; | |
81 | unsigned char size; | |
82 | }; | |
83 | ||
84 | /* | |
85 | * The configuration for a particular DS/BTS hardware implementation. | |
86 | */ | |
87 | struct ds_configuration { | |
88 | /* the DS configuration */ | |
89 | unsigned char sizeof_ds; | |
90 | struct access_desc bts_buffer_base; | |
91 | struct access_desc bts_index; | |
92 | struct access_desc bts_absolute_maximum; | |
93 | struct access_desc bts_interrupt_threshold; | |
94 | /* the BTS configuration */ | |
95 | unsigned char sizeof_bts; | |
96 | struct access_desc from_ip; | |
97 | struct access_desc to_ip; | |
98 | /* BTS variants used to store additional information like | |
99 | timestamps */ | |
100 | struct access_desc info_type; | |
101 | struct access_desc info_data; | |
102 | unsigned long debugctl_mask; | |
103 | }; | |
104 | ||
105 | /* | |
106 | * The global configuration used by the below accessor functions | |
107 | */ | |
108 | static struct ds_configuration ds_cfg; | |
109 | ||
110 | /* | |
111 | * Accessor functions for some DS and BTS fields using the above | |
112 | * global ptrace_bts_cfg. | |
113 | */ | |
e6ae5d95 | 114 | static inline unsigned long get_bts_buffer_base(char *base) |
eee3af4a | 115 | { |
e6ae5d95 | 116 | return *(unsigned long *)(base + ds_cfg.bts_buffer_base.offset); |
eee3af4a | 117 | } |
e6ae5d95 | 118 | static inline void set_bts_buffer_base(char *base, unsigned long value) |
eee3af4a | 119 | { |
e6ae5d95 | 120 | (*(unsigned long *)(base + ds_cfg.bts_buffer_base.offset)) = value; |
eee3af4a | 121 | } |
e6ae5d95 | 122 | static inline unsigned long get_bts_index(char *base) |
eee3af4a | 123 | { |
e6ae5d95 | 124 | return *(unsigned long *)(base + ds_cfg.bts_index.offset); |
eee3af4a | 125 | } |
e6ae5d95 | 126 | static inline void set_bts_index(char *base, unsigned long value) |
eee3af4a | 127 | { |
e6ae5d95 | 128 | (*(unsigned long *)(base + ds_cfg.bts_index.offset)) = value; |
eee3af4a | 129 | } |
e6ae5d95 | 130 | static inline unsigned long get_bts_absolute_maximum(char *base) |
eee3af4a | 131 | { |
e6ae5d95 | 132 | return *(unsigned long *)(base + ds_cfg.bts_absolute_maximum.offset); |
eee3af4a | 133 | } |
e6ae5d95 | 134 | static inline void set_bts_absolute_maximum(char *base, unsigned long value) |
eee3af4a | 135 | { |
e6ae5d95 | 136 | (*(unsigned long *)(base + ds_cfg.bts_absolute_maximum.offset)) = value; |
eee3af4a | 137 | } |
e6ae5d95 | 138 | static inline unsigned long get_bts_interrupt_threshold(char *base) |
eee3af4a | 139 | { |
e6ae5d95 | 140 | return *(unsigned long *)(base + ds_cfg.bts_interrupt_threshold.offset); |
eee3af4a | 141 | } |
e6ae5d95 | 142 | static inline void set_bts_interrupt_threshold(char *base, unsigned long value) |
eee3af4a | 143 | { |
e6ae5d95 | 144 | (*(unsigned long *)(base + ds_cfg.bts_interrupt_threshold.offset)) = value; |
eee3af4a | 145 | } |
e6ae5d95 | 146 | static inline unsigned long get_from_ip(char *base) |
eee3af4a | 147 | { |
e6ae5d95 | 148 | return *(unsigned long *)(base + ds_cfg.from_ip.offset); |
eee3af4a | 149 | } |
e6ae5d95 | 150 | static inline void set_from_ip(char *base, unsigned long value) |
eee3af4a | 151 | { |
e6ae5d95 | 152 | (*(unsigned long *)(base + ds_cfg.from_ip.offset)) = value; |
eee3af4a | 153 | } |
e6ae5d95 | 154 | static inline unsigned long get_to_ip(char *base) |
eee3af4a | 155 | { |
e6ae5d95 | 156 | return *(unsigned long *)(base + ds_cfg.to_ip.offset); |
eee3af4a | 157 | } |
e6ae5d95 | 158 | static inline void set_to_ip(char *base, unsigned long value) |
eee3af4a | 159 | { |
e6ae5d95 | 160 | (*(unsigned long *)(base + ds_cfg.to_ip.offset)) = value; |
eee3af4a MM |
161 | } |
162 | static inline unsigned char get_info_type(char *base) | |
163 | { | |
164 | return *(unsigned char *)(base + ds_cfg.info_type.offset); | |
165 | } | |
166 | static inline void set_info_type(char *base, unsigned char value) | |
167 | { | |
168 | (*(unsigned char *)(base + ds_cfg.info_type.offset)) = value; | |
169 | } | |
3c68904f | 170 | static inline unsigned long get_info_data(char *base) |
eee3af4a | 171 | { |
3c68904f | 172 | return *(unsigned long *)(base + ds_cfg.info_data.offset); |
eee3af4a | 173 | } |
3c68904f | 174 | static inline void set_info_data(char *base, unsigned long value) |
eee3af4a | 175 | { |
3c68904f | 176 | (*(unsigned long *)(base + ds_cfg.info_data.offset)) = value; |
eee3af4a MM |
177 | } |
178 | ||
179 | ||
a95d67f8 | 180 | int ds_allocate(void **dsp, size_t bts_size_in_bytes) |
eee3af4a | 181 | { |
a95d67f8 | 182 | size_t bts_size_in_records; |
e6ae5d95 | 183 | unsigned long bts; |
a95d67f8 | 184 | void *ds; |
eee3af4a MM |
185 | |
186 | if (!ds_cfg.sizeof_ds || !ds_cfg.sizeof_bts) | |
187 | return -EOPNOTSUPP; | |
188 | ||
a95d67f8 | 189 | if (bts_size_in_bytes < 0) |
eee3af4a MM |
190 | return -EINVAL; |
191 | ||
a95d67f8 MM |
192 | bts_size_in_records = |
193 | bts_size_in_bytes / ds_cfg.sizeof_bts; | |
eee3af4a MM |
194 | bts_size_in_bytes = |
195 | bts_size_in_records * ds_cfg.sizeof_bts; | |
196 | ||
197 | if (bts_size_in_bytes <= 0) | |
198 | return -EINVAL; | |
199 | ||
e6ae5d95 | 200 | bts = (unsigned long)kzalloc(bts_size_in_bytes, GFP_KERNEL); |
eee3af4a MM |
201 | |
202 | if (!bts) | |
203 | return -ENOMEM; | |
204 | ||
205 | ds = kzalloc(ds_cfg.sizeof_ds, GFP_KERNEL); | |
206 | ||
207 | if (!ds) { | |
e6ae5d95 | 208 | kfree((void *)bts); |
eee3af4a MM |
209 | return -ENOMEM; |
210 | } | |
211 | ||
212 | set_bts_buffer_base(ds, bts); | |
213 | set_bts_index(ds, bts); | |
214 | set_bts_absolute_maximum(ds, bts + bts_size_in_bytes); | |
215 | set_bts_interrupt_threshold(ds, bts + bts_size_in_bytes + 1); | |
216 | ||
217 | *dsp = ds; | |
218 | return 0; | |
219 | } | |
220 | ||
221 | int ds_free(void **dsp) | |
222 | { | |
223 | if (*dsp) | |
e6ae5d95 | 224 | kfree((void *)get_bts_buffer_base(*dsp)); |
eee3af4a MM |
225 | kfree(*dsp); |
226 | *dsp = 0; | |
227 | ||
228 | return 0; | |
229 | } | |
230 | ||
231 | int ds_get_bts_size(void *ds) | |
232 | { | |
e6ae5d95 | 233 | int size_in_bytes; |
eee3af4a MM |
234 | |
235 | if (!ds_cfg.sizeof_ds || !ds_cfg.sizeof_bts) | |
236 | return -EOPNOTSUPP; | |
237 | ||
a95d67f8 MM |
238 | if (!ds) |
239 | return 0; | |
240 | ||
eee3af4a MM |
241 | size_in_bytes = |
242 | get_bts_absolute_maximum(ds) - | |
243 | get_bts_buffer_base(ds); | |
a95d67f8 MM |
244 | return size_in_bytes; |
245 | } | |
246 | ||
247 | int ds_get_bts_end(void *ds) | |
248 | { | |
e6ae5d95 | 249 | int size_in_bytes = ds_get_bts_size(ds); |
a95d67f8 MM |
250 | |
251 | if (size_in_bytes <= 0) | |
252 | return size_in_bytes; | |
eee3af4a MM |
253 | |
254 | return size_in_bytes / ds_cfg.sizeof_bts; | |
255 | } | |
256 | ||
257 | int ds_get_bts_index(void *ds) | |
258 | { | |
e6ae5d95 | 259 | int index_offset_in_bytes; |
eee3af4a MM |
260 | |
261 | if (!ds_cfg.sizeof_ds || !ds_cfg.sizeof_bts) | |
262 | return -EOPNOTSUPP; | |
263 | ||
264 | index_offset_in_bytes = | |
265 | get_bts_index(ds) - | |
266 | get_bts_buffer_base(ds); | |
267 | ||
268 | return index_offset_in_bytes / ds_cfg.sizeof_bts; | |
269 | } | |
270 | ||
a95d67f8 MM |
271 | int ds_set_overflow(void *ds, int method) |
272 | { | |
273 | switch (method) { | |
274 | case DS_O_SIGNAL: | |
275 | return -EOPNOTSUPP; | |
276 | case DS_O_WRAP: | |
277 | return 0; | |
278 | default: | |
279 | return -EINVAL; | |
280 | } | |
281 | } | |
282 | ||
283 | int ds_get_overflow(void *ds) | |
284 | { | |
285 | return DS_O_WRAP; | |
286 | } | |
287 | ||
288 | int ds_clear(void *ds) | |
289 | { | |
290 | int bts_size = ds_get_bts_size(ds); | |
e6ae5d95 | 291 | unsigned long bts_base; |
a95d67f8 MM |
292 | |
293 | if (bts_size <= 0) | |
294 | return bts_size; | |
295 | ||
296 | bts_base = get_bts_buffer_base(ds); | |
e6ae5d95 | 297 | memset((void *)bts_base, 0, bts_size); |
a95d67f8 MM |
298 | |
299 | set_bts_index(ds, bts_base); | |
300 | return 0; | |
301 | } | |
302 | ||
e6ae5d95 | 303 | int ds_read_bts(void *ds, int index, struct bts_struct *out) |
eee3af4a MM |
304 | { |
305 | void *bts; | |
306 | ||
307 | if (!ds_cfg.sizeof_ds || !ds_cfg.sizeof_bts) | |
308 | return -EOPNOTSUPP; | |
309 | ||
310 | if (index < 0) | |
311 | return -EINVAL; | |
312 | ||
313 | if (index >= ds_get_bts_size(ds)) | |
314 | return -EINVAL; | |
315 | ||
e6ae5d95 | 316 | bts = (void *)(get_bts_buffer_base(ds) + (index * ds_cfg.sizeof_bts)); |
eee3af4a MM |
317 | |
318 | memset(out, 0, sizeof(*out)); | |
319 | if (get_from_ip(bts) == BTS_ESCAPE_ADDRESS) { | |
3c68904f MM |
320 | out->qualifier = get_info_type(bts); |
321 | out->variant.jiffies = get_info_data(bts); | |
eee3af4a MM |
322 | } else { |
323 | out->qualifier = BTS_BRANCH; | |
324 | out->variant.lbr.from_ip = get_from_ip(bts); | |
325 | out->variant.lbr.to_ip = get_to_ip(bts); | |
326 | } | |
327 | ||
e6ae5d95 | 328 | return sizeof(*out);; |
eee3af4a MM |
329 | } |
330 | ||
331 | int ds_write_bts(void *ds, const struct bts_struct *in) | |
332 | { | |
e6ae5d95 | 333 | unsigned long bts; |
eee3af4a MM |
334 | |
335 | if (!ds_cfg.sizeof_ds || !ds_cfg.sizeof_bts) | |
336 | return -EOPNOTSUPP; | |
337 | ||
338 | if (ds_get_bts_size(ds) <= 0) | |
339 | return -ENXIO; | |
340 | ||
341 | bts = get_bts_index(ds); | |
342 | ||
e6ae5d95 | 343 | memset((void *)bts, 0, ds_cfg.sizeof_bts); |
eee3af4a MM |
344 | switch (in->qualifier) { |
345 | case BTS_INVALID: | |
346 | break; | |
347 | ||
348 | case BTS_BRANCH: | |
e6ae5d95 MM |
349 | set_from_ip((void *)bts, in->variant.lbr.from_ip); |
350 | set_to_ip((void *)bts, in->variant.lbr.to_ip); | |
eee3af4a MM |
351 | break; |
352 | ||
353 | case BTS_TASK_ARRIVES: | |
354 | case BTS_TASK_DEPARTS: | |
e6ae5d95 MM |
355 | set_from_ip((void *)bts, BTS_ESCAPE_ADDRESS); |
356 | set_info_type((void *)bts, in->qualifier); | |
357 | set_info_data((void *)bts, in->variant.jiffies); | |
eee3af4a MM |
358 | break; |
359 | ||
360 | default: | |
361 | return -EINVAL; | |
362 | } | |
363 | ||
e6ae5d95 | 364 | bts = bts + ds_cfg.sizeof_bts; |
eee3af4a MM |
365 | if (bts >= get_bts_absolute_maximum(ds)) |
366 | bts = get_bts_buffer_base(ds); | |
367 | set_bts_index(ds, bts); | |
368 | ||
e6ae5d95 | 369 | return ds_cfg.sizeof_bts; |
eee3af4a MM |
370 | } |
371 | ||
372 | unsigned long ds_debugctl_mask(void) | |
373 | { | |
374 | return ds_cfg.debugctl_mask; | |
375 | } | |
376 | ||
377 | #ifdef __i386__ | |
378 | static const struct ds_configuration ds_cfg_netburst = { | |
379 | .sizeof_ds = 9 * 4, | |
380 | .bts_buffer_base = { 0, 4 }, | |
381 | .bts_index = { 4, 4 }, | |
382 | .bts_absolute_maximum = { 8, 4 }, | |
383 | .bts_interrupt_threshold = { 12, 4 }, | |
384 | .sizeof_bts = 3 * 4, | |
385 | .from_ip = { 0, 4 }, | |
386 | .to_ip = { 4, 4 }, | |
387 | .info_type = { 4, 1 }, | |
3c68904f | 388 | .info_data = { 8, 4 }, |
eee3af4a MM |
389 | .debugctl_mask = (1<<2)|(1<<3) |
390 | }; | |
391 | ||
392 | static const struct ds_configuration ds_cfg_pentium_m = { | |
393 | .sizeof_ds = 9 * 4, | |
394 | .bts_buffer_base = { 0, 4 }, | |
395 | .bts_index = { 4, 4 }, | |
396 | .bts_absolute_maximum = { 8, 4 }, | |
397 | .bts_interrupt_threshold = { 12, 4 }, | |
398 | .sizeof_bts = 3 * 4, | |
399 | .from_ip = { 0, 4 }, | |
400 | .to_ip = { 4, 4 }, | |
401 | .info_type = { 4, 1 }, | |
3c68904f | 402 | .info_data = { 8, 4 }, |
eee3af4a MM |
403 | .debugctl_mask = (1<<6)|(1<<7) |
404 | }; | |
405 | #endif /* _i386_ */ | |
406 | ||
407 | static const struct ds_configuration ds_cfg_core2 = { | |
408 | .sizeof_ds = 9 * 8, | |
409 | .bts_buffer_base = { 0, 8 }, | |
410 | .bts_index = { 8, 8 }, | |
411 | .bts_absolute_maximum = { 16, 8 }, | |
412 | .bts_interrupt_threshold = { 24, 8 }, | |
413 | .sizeof_bts = 3 * 8, | |
414 | .from_ip = { 0, 8 }, | |
415 | .to_ip = { 8, 8 }, | |
416 | .info_type = { 8, 1 }, | |
3c68904f | 417 | .info_data = { 16, 8 }, |
eee3af4a MM |
418 | .debugctl_mask = (1<<6)|(1<<7)|(1<<9) |
419 | }; | |
420 | ||
421 | static inline void | |
422 | ds_configure(const struct ds_configuration *cfg) | |
423 | { | |
424 | ds_cfg = *cfg; | |
425 | } | |
426 | ||
427 | void __cpuinit ds_init_intel(struct cpuinfo_x86 *c) | |
428 | { | |
429 | switch (c->x86) { | |
430 | case 0x6: | |
431 | switch (c->x86_model) { | |
432 | #ifdef __i386__ | |
433 | case 0xD: | |
434 | case 0xE: /* Pentium M */ | |
435 | ds_configure(&ds_cfg_pentium_m); | |
436 | break; | |
437 | #endif /* _i386_ */ | |
438 | case 0xF: /* Core2 */ | |
439 | ds_configure(&ds_cfg_core2); | |
440 | break; | |
441 | default: | |
442 | /* sorry, don't know about them */ | |
443 | break; | |
444 | } | |
445 | break; | |
446 | case 0xF: | |
447 | switch (c->x86_model) { | |
448 | #ifdef __i386__ | |
449 | case 0x0: | |
450 | case 0x1: | |
451 | case 0x2: /* Netburst */ | |
452 | ds_configure(&ds_cfg_netburst); | |
453 | break; | |
454 | #endif /* _i386_ */ | |
455 | default: | |
456 | /* sorry, don't know about them */ | |
457 | break; | |
458 | } | |
459 | break; | |
460 | default: | |
461 | /* sorry, don't know about them */ | |
462 | break; | |
463 | } | |
464 | } |