xref: /openbmc/linux/tools/perf/arch/x86/util/intel-bts.c (revision 1eaf496e)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * intel-bts.c: Intel Processor Trace support
4  * Copyright (c) 2013-2015, Intel Corporation.
5  */
6 
7 #include <errno.h>
8 #include <linux/kernel.h>
9 #include <linux/types.h>
10 #include <linux/bitops.h>
11 #include <linux/log2.h>
12 #include <linux/zalloc.h>
13 
14 #include "../../../util/cpumap.h"
15 #include "../../../util/event.h"
16 #include "../../../util/evsel.h"
17 #include "../../../util/evlist.h"
18 #include "../../../util/mmap.h"
19 #include "../../../util/session.h"
20 #include "../../../util/pmus.h"
21 #include "../../../util/debug.h"
22 #include "../../../util/record.h"
23 #include "../../../util/tsc.h"
24 #include "../../../util/auxtrace.h"
25 #include "../../../util/intel-bts.h"
26 #include <internal/lib.h> // page_size
27 
28 #define KiB(x) ((x) * 1024)
29 #define MiB(x) ((x) * 1024 * 1024)
30 #define KiB_MASK(x) (KiB(x) - 1)
31 #define MiB_MASK(x) (MiB(x) - 1)
32 
33 struct intel_bts_snapshot_ref {
34 	void	*ref_buf;
35 	size_t	ref_offset;
36 	bool	wrapped;
37 };
38 
39 struct intel_bts_recording {
40 	struct auxtrace_record		itr;
41 	struct perf_pmu			*intel_bts_pmu;
42 	struct evlist		*evlist;
43 	bool				snapshot_mode;
44 	size_t				snapshot_size;
45 	int				snapshot_ref_cnt;
46 	struct intel_bts_snapshot_ref	*snapshot_refs;
47 };
48 
49 struct branch {
50 	u64 from;
51 	u64 to;
52 	u64 misc;
53 };
54 
55 static size_t
intel_bts_info_priv_size(struct auxtrace_record * itr __maybe_unused,struct evlist * evlist __maybe_unused)56 intel_bts_info_priv_size(struct auxtrace_record *itr __maybe_unused,
57 			 struct evlist *evlist __maybe_unused)
58 {
59 	return INTEL_BTS_AUXTRACE_PRIV_SIZE;
60 }
61 
intel_bts_info_fill(struct auxtrace_record * itr,struct perf_session * session,struct perf_record_auxtrace_info * auxtrace_info,size_t priv_size)62 static int intel_bts_info_fill(struct auxtrace_record *itr,
63 			       struct perf_session *session,
64 			       struct perf_record_auxtrace_info *auxtrace_info,
65 			       size_t priv_size)
66 {
67 	struct intel_bts_recording *btsr =
68 			container_of(itr, struct intel_bts_recording, itr);
69 	struct perf_pmu *intel_bts_pmu = btsr->intel_bts_pmu;
70 	struct perf_event_mmap_page *pc;
71 	struct perf_tsc_conversion tc = { .time_mult = 0, };
72 	bool cap_user_time_zero = false;
73 	int err;
74 
75 	if (priv_size != INTEL_BTS_AUXTRACE_PRIV_SIZE)
76 		return -EINVAL;
77 
78 	if (!session->evlist->core.nr_mmaps)
79 		return -EINVAL;
80 
81 	pc = session->evlist->mmap[0].core.base;
82 	if (pc) {
83 		err = perf_read_tsc_conversion(pc, &tc);
84 		if (err) {
85 			if (err != -EOPNOTSUPP)
86 				return err;
87 		} else {
88 			cap_user_time_zero = tc.time_mult != 0;
89 		}
90 		if (!cap_user_time_zero)
91 			ui__warning("Intel BTS: TSC not available\n");
92 	}
93 
94 	auxtrace_info->type = PERF_AUXTRACE_INTEL_BTS;
95 	auxtrace_info->priv[INTEL_BTS_PMU_TYPE] = intel_bts_pmu->type;
96 	auxtrace_info->priv[INTEL_BTS_TIME_SHIFT] = tc.time_shift;
97 	auxtrace_info->priv[INTEL_BTS_TIME_MULT] = tc.time_mult;
98 	auxtrace_info->priv[INTEL_BTS_TIME_ZERO] = tc.time_zero;
99 	auxtrace_info->priv[INTEL_BTS_CAP_USER_TIME_ZERO] = cap_user_time_zero;
100 	auxtrace_info->priv[INTEL_BTS_SNAPSHOT_MODE] = btsr->snapshot_mode;
101 
102 	return 0;
103 }
104 
intel_bts_recording_options(struct auxtrace_record * itr,struct evlist * evlist,struct record_opts * opts)105 static int intel_bts_recording_options(struct auxtrace_record *itr,
106 				       struct evlist *evlist,
107 				       struct record_opts *opts)
108 {
109 	struct intel_bts_recording *btsr =
110 			container_of(itr, struct intel_bts_recording, itr);
111 	struct perf_pmu *intel_bts_pmu = btsr->intel_bts_pmu;
112 	struct evsel *evsel, *intel_bts_evsel = NULL;
113 	const struct perf_cpu_map *cpus = evlist->core.user_requested_cpus;
114 	bool privileged = perf_event_paranoid_check(-1);
115 
116 	if (opts->auxtrace_sample_mode) {
117 		pr_err("Intel BTS does not support AUX area sampling\n");
118 		return -EINVAL;
119 	}
120 
121 	btsr->evlist = evlist;
122 	btsr->snapshot_mode = opts->auxtrace_snapshot_mode;
123 
124 	evlist__for_each_entry(evlist, evsel) {
125 		if (evsel->core.attr.type == intel_bts_pmu->type) {
126 			if (intel_bts_evsel) {
127 				pr_err("There may be only one " INTEL_BTS_PMU_NAME " event\n");
128 				return -EINVAL;
129 			}
130 			evsel->core.attr.freq = 0;
131 			evsel->core.attr.sample_period = 1;
132 			evsel->needs_auxtrace_mmap = true;
133 			intel_bts_evsel = evsel;
134 			opts->full_auxtrace = true;
135 		}
136 	}
137 
138 	if (opts->auxtrace_snapshot_mode && !opts->full_auxtrace) {
139 		pr_err("Snapshot mode (-S option) requires " INTEL_BTS_PMU_NAME " PMU event (-e " INTEL_BTS_PMU_NAME ")\n");
140 		return -EINVAL;
141 	}
142 
143 	if (!opts->full_auxtrace)
144 		return 0;
145 
146 	if (opts->full_auxtrace && !perf_cpu_map__empty(cpus)) {
147 		pr_err(INTEL_BTS_PMU_NAME " does not support per-cpu recording\n");
148 		return -EINVAL;
149 	}
150 
151 	/* Set default sizes for snapshot mode */
152 	if (opts->auxtrace_snapshot_mode) {
153 		if (!opts->auxtrace_snapshot_size && !opts->auxtrace_mmap_pages) {
154 			if (privileged) {
155 				opts->auxtrace_mmap_pages = MiB(4) / page_size;
156 			} else {
157 				opts->auxtrace_mmap_pages = KiB(128) / page_size;
158 				if (opts->mmap_pages == UINT_MAX)
159 					opts->mmap_pages = KiB(256) / page_size;
160 			}
161 		} else if (!opts->auxtrace_mmap_pages && !privileged &&
162 			   opts->mmap_pages == UINT_MAX) {
163 			opts->mmap_pages = KiB(256) / page_size;
164 		}
165 		if (!opts->auxtrace_snapshot_size)
166 			opts->auxtrace_snapshot_size =
167 				opts->auxtrace_mmap_pages * (size_t)page_size;
168 		if (!opts->auxtrace_mmap_pages) {
169 			size_t sz = opts->auxtrace_snapshot_size;
170 
171 			sz = round_up(sz, page_size) / page_size;
172 			opts->auxtrace_mmap_pages = roundup_pow_of_two(sz);
173 		}
174 		if (opts->auxtrace_snapshot_size >
175 				opts->auxtrace_mmap_pages * (size_t)page_size) {
176 			pr_err("Snapshot size %zu must not be greater than AUX area tracing mmap size %zu\n",
177 			       opts->auxtrace_snapshot_size,
178 			       opts->auxtrace_mmap_pages * (size_t)page_size);
179 			return -EINVAL;
180 		}
181 		if (!opts->auxtrace_snapshot_size || !opts->auxtrace_mmap_pages) {
182 			pr_err("Failed to calculate default snapshot size and/or AUX area tracing mmap pages\n");
183 			return -EINVAL;
184 		}
185 		pr_debug2("Intel BTS snapshot size: %zu\n",
186 			  opts->auxtrace_snapshot_size);
187 	}
188 
189 	/* Set default sizes for full trace mode */
190 	if (opts->full_auxtrace && !opts->auxtrace_mmap_pages) {
191 		if (privileged) {
192 			opts->auxtrace_mmap_pages = MiB(4) / page_size;
193 		} else {
194 			opts->auxtrace_mmap_pages = KiB(128) / page_size;
195 			if (opts->mmap_pages == UINT_MAX)
196 				opts->mmap_pages = KiB(256) / page_size;
197 		}
198 	}
199 
200 	/* Validate auxtrace_mmap_pages */
201 	if (opts->auxtrace_mmap_pages) {
202 		size_t sz = opts->auxtrace_mmap_pages * (size_t)page_size;
203 		size_t min_sz;
204 
205 		if (opts->auxtrace_snapshot_mode)
206 			min_sz = KiB(4);
207 		else
208 			min_sz = KiB(8);
209 
210 		if (sz < min_sz || !is_power_of_2(sz)) {
211 			pr_err("Invalid mmap size for Intel BTS: must be at least %zuKiB and a power of 2\n",
212 			       min_sz / 1024);
213 			return -EINVAL;
214 		}
215 	}
216 
217 	if (intel_bts_evsel) {
218 		/*
219 		 * To obtain the auxtrace buffer file descriptor, the auxtrace event
220 		 * must come first.
221 		 */
222 		evlist__to_front(evlist, intel_bts_evsel);
223 		/*
224 		 * In the case of per-cpu mmaps, we need the CPU on the
225 		 * AUX event.
226 		 */
227 		if (!perf_cpu_map__empty(cpus))
228 			evsel__set_sample_bit(intel_bts_evsel, CPU);
229 	}
230 
231 	/* Add dummy event to keep tracking */
232 	if (opts->full_auxtrace) {
233 		struct evsel *tracking_evsel;
234 		int err;
235 
236 		err = parse_event(evlist, "dummy:u");
237 		if (err)
238 			return err;
239 
240 		tracking_evsel = evlist__last(evlist);
241 
242 		evlist__set_tracking_event(evlist, tracking_evsel);
243 
244 		tracking_evsel->core.attr.freq = 0;
245 		tracking_evsel->core.attr.sample_period = 1;
246 	}
247 
248 	return 0;
249 }
250 
intel_bts_parse_snapshot_options(struct auxtrace_record * itr,struct record_opts * opts,const char * str)251 static int intel_bts_parse_snapshot_options(struct auxtrace_record *itr,
252 					    struct record_opts *opts,
253 					    const char *str)
254 {
255 	struct intel_bts_recording *btsr =
256 			container_of(itr, struct intel_bts_recording, itr);
257 	unsigned long long snapshot_size = 0;
258 	char *endptr;
259 
260 	if (str) {
261 		snapshot_size = strtoull(str, &endptr, 0);
262 		if (*endptr || snapshot_size > SIZE_MAX)
263 			return -1;
264 	}
265 
266 	opts->auxtrace_snapshot_mode = true;
267 	opts->auxtrace_snapshot_size = snapshot_size;
268 
269 	btsr->snapshot_size = snapshot_size;
270 
271 	return 0;
272 }
273 
intel_bts_reference(struct auxtrace_record * itr __maybe_unused)274 static u64 intel_bts_reference(struct auxtrace_record *itr __maybe_unused)
275 {
276 	return rdtsc();
277 }
278 
intel_bts_alloc_snapshot_refs(struct intel_bts_recording * btsr,int idx)279 static int intel_bts_alloc_snapshot_refs(struct intel_bts_recording *btsr,
280 					 int idx)
281 {
282 	const size_t sz = sizeof(struct intel_bts_snapshot_ref);
283 	int cnt = btsr->snapshot_ref_cnt, new_cnt = cnt * 2;
284 	struct intel_bts_snapshot_ref *refs;
285 
286 	if (!new_cnt)
287 		new_cnt = 16;
288 
289 	while (new_cnt <= idx)
290 		new_cnt *= 2;
291 
292 	refs = calloc(new_cnt, sz);
293 	if (!refs)
294 		return -ENOMEM;
295 
296 	memcpy(refs, btsr->snapshot_refs, cnt * sz);
297 
298 	btsr->snapshot_refs = refs;
299 	btsr->snapshot_ref_cnt = new_cnt;
300 
301 	return 0;
302 }
303 
intel_bts_free_snapshot_refs(struct intel_bts_recording * btsr)304 static void intel_bts_free_snapshot_refs(struct intel_bts_recording *btsr)
305 {
306 	int i;
307 
308 	for (i = 0; i < btsr->snapshot_ref_cnt; i++)
309 		zfree(&btsr->snapshot_refs[i].ref_buf);
310 	zfree(&btsr->snapshot_refs);
311 }
312 
intel_bts_recording_free(struct auxtrace_record * itr)313 static void intel_bts_recording_free(struct auxtrace_record *itr)
314 {
315 	struct intel_bts_recording *btsr =
316 			container_of(itr, struct intel_bts_recording, itr);
317 
318 	intel_bts_free_snapshot_refs(btsr);
319 	free(btsr);
320 }
321 
intel_bts_snapshot_start(struct auxtrace_record * itr)322 static int intel_bts_snapshot_start(struct auxtrace_record *itr)
323 {
324 	struct intel_bts_recording *btsr =
325 			container_of(itr, struct intel_bts_recording, itr);
326 	struct evsel *evsel;
327 
328 	evlist__for_each_entry(btsr->evlist, evsel) {
329 		if (evsel->core.attr.type == btsr->intel_bts_pmu->type)
330 			return evsel__disable(evsel);
331 	}
332 	return -EINVAL;
333 }
334 
intel_bts_snapshot_finish(struct auxtrace_record * itr)335 static int intel_bts_snapshot_finish(struct auxtrace_record *itr)
336 {
337 	struct intel_bts_recording *btsr =
338 			container_of(itr, struct intel_bts_recording, itr);
339 	struct evsel *evsel;
340 
341 	evlist__for_each_entry(btsr->evlist, evsel) {
342 		if (evsel->core.attr.type == btsr->intel_bts_pmu->type)
343 			return evsel__enable(evsel);
344 	}
345 	return -EINVAL;
346 }
347 
intel_bts_first_wrap(u64 * data,size_t buf_size)348 static bool intel_bts_first_wrap(u64 *data, size_t buf_size)
349 {
350 	int i, a, b;
351 
352 	b = buf_size >> 3;
353 	a = b - 512;
354 	if (a < 0)
355 		a = 0;
356 
357 	for (i = a; i < b; i++) {
358 		if (data[i])
359 			return true;
360 	}
361 
362 	return false;
363 }
364 
intel_bts_find_snapshot(struct auxtrace_record * itr,int idx,struct auxtrace_mmap * mm,unsigned char * data,u64 * head,u64 * old)365 static int intel_bts_find_snapshot(struct auxtrace_record *itr, int idx,
366 				   struct auxtrace_mmap *mm, unsigned char *data,
367 				   u64 *head, u64 *old)
368 {
369 	struct intel_bts_recording *btsr =
370 			container_of(itr, struct intel_bts_recording, itr);
371 	bool wrapped;
372 	int err;
373 
374 	pr_debug3("%s: mmap index %d old head %zu new head %zu\n",
375 		  __func__, idx, (size_t)*old, (size_t)*head);
376 
377 	if (idx >= btsr->snapshot_ref_cnt) {
378 		err = intel_bts_alloc_snapshot_refs(btsr, idx);
379 		if (err)
380 			goto out_err;
381 	}
382 
383 	wrapped = btsr->snapshot_refs[idx].wrapped;
384 	if (!wrapped && intel_bts_first_wrap((u64 *)data, mm->len)) {
385 		btsr->snapshot_refs[idx].wrapped = true;
386 		wrapped = true;
387 	}
388 
389 	/*
390 	 * In full trace mode 'head' continually increases.  However in snapshot
391 	 * mode 'head' is an offset within the buffer.  Here 'old' and 'head'
392 	 * are adjusted to match the full trace case which expects that 'old' is
393 	 * always less than 'head'.
394 	 */
395 	if (wrapped) {
396 		*old = *head;
397 		*head += mm->len;
398 	} else {
399 		if (mm->mask)
400 			*old &= mm->mask;
401 		else
402 			*old %= mm->len;
403 		if (*old > *head)
404 			*head += mm->len;
405 	}
406 
407 	pr_debug3("%s: wrap-around %sdetected, adjusted old head %zu adjusted new head %zu\n",
408 		  __func__, wrapped ? "" : "not ", (size_t)*old, (size_t)*head);
409 
410 	return 0;
411 
412 out_err:
413 	pr_err("%s: failed, error %d\n", __func__, err);
414 	return err;
415 }
416 
intel_bts_recording_init(int * err)417 struct auxtrace_record *intel_bts_recording_init(int *err)
418 {
419 	struct perf_pmu *intel_bts_pmu = perf_pmus__find(INTEL_BTS_PMU_NAME);
420 	struct intel_bts_recording *btsr;
421 
422 	if (!intel_bts_pmu)
423 		return NULL;
424 
425 	if (setenv("JITDUMP_USE_ARCH_TIMESTAMP", "1", 1)) {
426 		*err = -errno;
427 		return NULL;
428 	}
429 
430 	btsr = zalloc(sizeof(struct intel_bts_recording));
431 	if (!btsr) {
432 		*err = -ENOMEM;
433 		return NULL;
434 	}
435 
436 	btsr->intel_bts_pmu = intel_bts_pmu;
437 	btsr->itr.pmu = intel_bts_pmu;
438 	btsr->itr.recording_options = intel_bts_recording_options;
439 	btsr->itr.info_priv_size = intel_bts_info_priv_size;
440 	btsr->itr.info_fill = intel_bts_info_fill;
441 	btsr->itr.free = intel_bts_recording_free;
442 	btsr->itr.snapshot_start = intel_bts_snapshot_start;
443 	btsr->itr.snapshot_finish = intel_bts_snapshot_finish;
444 	btsr->itr.find_snapshot = intel_bts_find_snapshot;
445 	btsr->itr.parse_snapshot_options = intel_bts_parse_snapshot_options;
446 	btsr->itr.reference = intel_bts_reference;
447 	btsr->itr.read_finish = auxtrace_record__read_finish;
448 	btsr->itr.alignment = sizeof(struct branch);
449 	return &btsr->itr;
450 }
451