1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * intel_pt_decoder.c: Intel Processor Trace support
4 * Copyright (c) 2013-2014, Intel Corporation.
5 */
6
7 #ifndef _GNU_SOURCE
8 #define _GNU_SOURCE
9 #endif
10 #include <stdlib.h>
11 #include <stdbool.h>
12 #include <string.h>
13 #include <errno.h>
14 #include <stdint.h>
15 #include <inttypes.h>
16 #include <linux/compiler.h>
17 #include <linux/string.h>
18 #include <linux/zalloc.h>
19
20 #include "../auxtrace.h"
21
22 #include "intel-pt-insn-decoder.h"
23 #include "intel-pt-pkt-decoder.h"
24 #include "intel-pt-decoder.h"
25 #include "intel-pt-log.h"
26
27 #define BITULL(x) (1ULL << (x))
28
29 /* IA32_RTIT_CTL MSR bits */
30 #define INTEL_PT_CYC_ENABLE BITULL(1)
31 #define INTEL_PT_CYC_THRESHOLD (BITULL(22) | BITULL(21) | BITULL(20) | BITULL(19))
32 #define INTEL_PT_CYC_THRESHOLD_SHIFT 19
33
34 #define INTEL_PT_BLK_SIZE 1024
35
36 #define BIT63 (((uint64_t)1 << 63))
37
38 #define INTEL_PT_RETURN 1
39
40 /* Maximum number of loops with no packets consumed i.e. stuck in a loop */
41 #define INTEL_PT_MAX_LOOPS 10000
42
43 struct intel_pt_blk {
44 struct intel_pt_blk *prev;
45 uint64_t ip[INTEL_PT_BLK_SIZE];
46 };
47
48 struct intel_pt_stack {
49 struct intel_pt_blk *blk;
50 struct intel_pt_blk *spare;
51 int pos;
52 };
53
54 enum intel_pt_pkt_state {
55 INTEL_PT_STATE_NO_PSB,
56 INTEL_PT_STATE_NO_IP,
57 INTEL_PT_STATE_ERR_RESYNC,
58 INTEL_PT_STATE_IN_SYNC,
59 INTEL_PT_STATE_TNT_CONT,
60 INTEL_PT_STATE_TNT,
61 INTEL_PT_STATE_TIP,
62 INTEL_PT_STATE_TIP_PGD,
63 INTEL_PT_STATE_FUP,
64 INTEL_PT_STATE_FUP_NO_TIP,
65 INTEL_PT_STATE_RESAMPLE,
66 };
67
intel_pt_sample_time(enum intel_pt_pkt_state pkt_state)68 static inline bool intel_pt_sample_time(enum intel_pt_pkt_state pkt_state)
69 {
70 switch (pkt_state) {
71 case INTEL_PT_STATE_NO_PSB:
72 case INTEL_PT_STATE_NO_IP:
73 case INTEL_PT_STATE_ERR_RESYNC:
74 case INTEL_PT_STATE_IN_SYNC:
75 case INTEL_PT_STATE_TNT_CONT:
76 case INTEL_PT_STATE_RESAMPLE:
77 return true;
78 case INTEL_PT_STATE_TNT:
79 case INTEL_PT_STATE_TIP:
80 case INTEL_PT_STATE_TIP_PGD:
81 case INTEL_PT_STATE_FUP:
82 case INTEL_PT_STATE_FUP_NO_TIP:
83 return false;
84 default:
85 return true;
86 };
87 }
88
89 #ifdef INTEL_PT_STRICT
90 #define INTEL_PT_STATE_ERR1 INTEL_PT_STATE_NO_PSB
91 #define INTEL_PT_STATE_ERR2 INTEL_PT_STATE_NO_PSB
92 #define INTEL_PT_STATE_ERR3 INTEL_PT_STATE_NO_PSB
93 #define INTEL_PT_STATE_ERR4 INTEL_PT_STATE_NO_PSB
94 #else
95 #define INTEL_PT_STATE_ERR1 (decoder->pkt_state)
96 #define INTEL_PT_STATE_ERR2 INTEL_PT_STATE_NO_IP
97 #define INTEL_PT_STATE_ERR3 INTEL_PT_STATE_ERR_RESYNC
98 #define INTEL_PT_STATE_ERR4 INTEL_PT_STATE_IN_SYNC
99 #endif
100
101 struct intel_pt_decoder {
102 int (*get_trace)(struct intel_pt_buffer *buffer, void *data);
103 int (*walk_insn)(struct intel_pt_insn *intel_pt_insn,
104 uint64_t *insn_cnt_ptr, uint64_t *ip, uint64_t to_ip,
105 uint64_t max_insn_cnt, void *data);
106 bool (*pgd_ip)(uint64_t ip, void *data);
107 int (*lookahead)(void *data, intel_pt_lookahead_cb_t cb, void *cb_data);
108 void *data;
109 struct intel_pt_state state;
110 const unsigned char *buf;
111 size_t len;
112 bool return_compression;
113 bool branch_enable;
114 bool mtc_insn;
115 bool pge;
116 bool have_tma;
117 bool have_cyc;
118 bool fixup_last_mtc;
119 bool have_last_ip;
120 bool in_psb;
121 bool hop;
122 bool hop_psb_fup;
123 bool leap;
124 enum intel_pt_param_flags flags;
125 uint64_t pos;
126 uint64_t last_ip;
127 uint64_t ip;
128 uint64_t cr3;
129 uint64_t timestamp;
130 uint64_t tsc_timestamp;
131 uint64_t ref_timestamp;
132 uint64_t buf_timestamp;
133 uint64_t sample_timestamp;
134 uint64_t ret_addr;
135 uint64_t ctc_timestamp;
136 uint64_t ctc_delta;
137 uint64_t cycle_cnt;
138 uint64_t cyc_ref_timestamp;
139 uint32_t last_mtc;
140 uint32_t tsc_ctc_ratio_n;
141 uint32_t tsc_ctc_ratio_d;
142 uint32_t tsc_ctc_mult;
143 uint32_t tsc_slip;
144 uint32_t ctc_rem_mask;
145 int mtc_shift;
146 struct intel_pt_stack stack;
147 enum intel_pt_pkt_state pkt_state;
148 enum intel_pt_pkt_ctx pkt_ctx;
149 enum intel_pt_pkt_ctx prev_pkt_ctx;
150 enum intel_pt_blk_type blk_type;
151 int blk_type_pos;
152 struct intel_pt_pkt packet;
153 struct intel_pt_pkt tnt;
154 int pkt_step;
155 int pkt_len;
156 int last_packet_type;
157 unsigned int cbr;
158 unsigned int cbr_seen;
159 unsigned int max_non_turbo_ratio;
160 double max_non_turbo_ratio_fp;
161 double cbr_cyc_to_tsc;
162 double calc_cyc_to_tsc;
163 bool have_calc_cyc_to_tsc;
164 int exec_mode;
165 unsigned int insn_bytes;
166 uint64_t period;
167 enum intel_pt_period_type period_type;
168 uint64_t tot_insn_cnt;
169 uint64_t period_insn_cnt;
170 uint64_t period_mask;
171 uint64_t period_ticks;
172 uint64_t last_masked_timestamp;
173 uint64_t tot_cyc_cnt;
174 uint64_t sample_tot_cyc_cnt;
175 uint64_t base_cyc_cnt;
176 uint64_t cyc_cnt_timestamp;
177 uint64_t ctl;
178 uint64_t cyc_threshold;
179 double tsc_to_cyc;
180 bool continuous_period;
181 bool overflow;
182 bool set_fup_tx_flags;
183 bool set_fup_ptw;
184 bool set_fup_mwait;
185 bool set_fup_pwre;
186 bool set_fup_exstop;
187 bool set_fup_bep;
188 bool sample_cyc;
189 unsigned int fup_tx_flags;
190 unsigned int tx_flags;
191 uint64_t fup_ptw_payload;
192 uint64_t fup_mwait_payload;
193 uint64_t fup_pwre_payload;
194 uint64_t cbr_payload;
195 uint64_t timestamp_insn_cnt;
196 uint64_t sample_insn_cnt;
197 uint64_t stuck_ip;
198 int no_progress;
199 int stuck_ip_prd;
200 int stuck_ip_cnt;
201 const unsigned char *next_buf;
202 size_t next_len;
203 unsigned char temp_buf[INTEL_PT_PKT_MAX_SZ];
204 };
205
intel_pt_lower_power_of_2(uint64_t x)206 static uint64_t intel_pt_lower_power_of_2(uint64_t x)
207 {
208 int i;
209
210 for (i = 0; x != 1; i++)
211 x >>= 1;
212
213 return x << i;
214 }
215
intel_pt_cyc_threshold(uint64_t ctl)216 static uint64_t intel_pt_cyc_threshold(uint64_t ctl)
217 {
218 if (!(ctl & INTEL_PT_CYC_ENABLE))
219 return 0;
220
221 return (ctl & INTEL_PT_CYC_THRESHOLD) >> INTEL_PT_CYC_THRESHOLD_SHIFT;
222 }
223
intel_pt_setup_period(struct intel_pt_decoder * decoder)224 static void intel_pt_setup_period(struct intel_pt_decoder *decoder)
225 {
226 if (decoder->period_type == INTEL_PT_PERIOD_TICKS) {
227 uint64_t period;
228
229 period = intel_pt_lower_power_of_2(decoder->period);
230 decoder->period_mask = ~(period - 1);
231 decoder->period_ticks = period;
232 }
233 }
234
multdiv(uint64_t t,uint32_t n,uint32_t d)235 static uint64_t multdiv(uint64_t t, uint32_t n, uint32_t d)
236 {
237 if (!d)
238 return 0;
239 return (t / d) * n + ((t % d) * n) / d;
240 }
241
intel_pt_decoder_new(struct intel_pt_params * params)242 struct intel_pt_decoder *intel_pt_decoder_new(struct intel_pt_params *params)
243 {
244 struct intel_pt_decoder *decoder;
245
246 if (!params->get_trace || !params->walk_insn)
247 return NULL;
248
249 decoder = zalloc(sizeof(struct intel_pt_decoder));
250 if (!decoder)
251 return NULL;
252
253 decoder->get_trace = params->get_trace;
254 decoder->walk_insn = params->walk_insn;
255 decoder->pgd_ip = params->pgd_ip;
256 decoder->lookahead = params->lookahead;
257 decoder->data = params->data;
258 decoder->return_compression = params->return_compression;
259 decoder->branch_enable = params->branch_enable;
260 decoder->hop = params->quick >= 1;
261 decoder->leap = params->quick >= 2;
262
263 decoder->flags = params->flags;
264
265 decoder->ctl = params->ctl;
266 decoder->period = params->period;
267 decoder->period_type = params->period_type;
268
269 decoder->max_non_turbo_ratio = params->max_non_turbo_ratio;
270 decoder->max_non_turbo_ratio_fp = params->max_non_turbo_ratio;
271
272 decoder->cyc_threshold = intel_pt_cyc_threshold(decoder->ctl);
273
274 intel_pt_setup_period(decoder);
275
276 decoder->mtc_shift = params->mtc_period;
277 decoder->ctc_rem_mask = (1 << decoder->mtc_shift) - 1;
278
279 decoder->tsc_ctc_ratio_n = params->tsc_ctc_ratio_n;
280 decoder->tsc_ctc_ratio_d = params->tsc_ctc_ratio_d;
281
282 if (!decoder->tsc_ctc_ratio_n)
283 decoder->tsc_ctc_ratio_d = 0;
284
285 if (decoder->tsc_ctc_ratio_d) {
286 if (!(decoder->tsc_ctc_ratio_n % decoder->tsc_ctc_ratio_d))
287 decoder->tsc_ctc_mult = decoder->tsc_ctc_ratio_n /
288 decoder->tsc_ctc_ratio_d;
289 }
290
291 /*
292 * A TSC packet can slip past MTC packets so that the timestamp appears
293 * to go backwards. One estimate is that can be up to about 40 CPU
294 * cycles, which is certainly less than 0x1000 TSC ticks, but accept
295 * slippage an order of magnitude more to be on the safe side.
296 */
297 decoder->tsc_slip = 0x10000;
298
299 intel_pt_log("timestamp: mtc_shift %u\n", decoder->mtc_shift);
300 intel_pt_log("timestamp: tsc_ctc_ratio_n %u\n", decoder->tsc_ctc_ratio_n);
301 intel_pt_log("timestamp: tsc_ctc_ratio_d %u\n", decoder->tsc_ctc_ratio_d);
302 intel_pt_log("timestamp: tsc_ctc_mult %u\n", decoder->tsc_ctc_mult);
303 intel_pt_log("timestamp: tsc_slip %#x\n", decoder->tsc_slip);
304
305 if (decoder->hop)
306 intel_pt_log("Hop mode: decoding FUP and TIPs, but not TNT\n");
307
308 return decoder;
309 }
310
intel_pt_pop_blk(struct intel_pt_stack * stack)311 static void intel_pt_pop_blk(struct intel_pt_stack *stack)
312 {
313 struct intel_pt_blk *blk = stack->blk;
314
315 stack->blk = blk->prev;
316 if (!stack->spare)
317 stack->spare = blk;
318 else
319 free(blk);
320 }
321
intel_pt_pop(struct intel_pt_stack * stack)322 static uint64_t intel_pt_pop(struct intel_pt_stack *stack)
323 {
324 if (!stack->pos) {
325 if (!stack->blk)
326 return 0;
327 intel_pt_pop_blk(stack);
328 if (!stack->blk)
329 return 0;
330 stack->pos = INTEL_PT_BLK_SIZE;
331 }
332 return stack->blk->ip[--stack->pos];
333 }
334
intel_pt_alloc_blk(struct intel_pt_stack * stack)335 static int intel_pt_alloc_blk(struct intel_pt_stack *stack)
336 {
337 struct intel_pt_blk *blk;
338
339 if (stack->spare) {
340 blk = stack->spare;
341 stack->spare = NULL;
342 } else {
343 blk = malloc(sizeof(struct intel_pt_blk));
344 if (!blk)
345 return -ENOMEM;
346 }
347
348 blk->prev = stack->blk;
349 stack->blk = blk;
350 stack->pos = 0;
351 return 0;
352 }
353
intel_pt_push(struct intel_pt_stack * stack,uint64_t ip)354 static int intel_pt_push(struct intel_pt_stack *stack, uint64_t ip)
355 {
356 int err;
357
358 if (!stack->blk || stack->pos == INTEL_PT_BLK_SIZE) {
359 err = intel_pt_alloc_blk(stack);
360 if (err)
361 return err;
362 }
363
364 stack->blk->ip[stack->pos++] = ip;
365 return 0;
366 }
367
intel_pt_clear_stack(struct intel_pt_stack * stack)368 static void intel_pt_clear_stack(struct intel_pt_stack *stack)
369 {
370 while (stack->blk)
371 intel_pt_pop_blk(stack);
372 stack->pos = 0;
373 }
374
intel_pt_free_stack(struct intel_pt_stack * stack)375 static void intel_pt_free_stack(struct intel_pt_stack *stack)
376 {
377 intel_pt_clear_stack(stack);
378 zfree(&stack->blk);
379 zfree(&stack->spare);
380 }
381
intel_pt_decoder_free(struct intel_pt_decoder * decoder)382 void intel_pt_decoder_free(struct intel_pt_decoder *decoder)
383 {
384 intel_pt_free_stack(&decoder->stack);
385 free(decoder);
386 }
387
intel_pt_ext_err(int code)388 static int intel_pt_ext_err(int code)
389 {
390 switch (code) {
391 case -ENOMEM:
392 return INTEL_PT_ERR_NOMEM;
393 case -ENOSYS:
394 return INTEL_PT_ERR_INTERN;
395 case -EBADMSG:
396 return INTEL_PT_ERR_BADPKT;
397 case -ENODATA:
398 return INTEL_PT_ERR_NODATA;
399 case -EILSEQ:
400 return INTEL_PT_ERR_NOINSN;
401 case -ENOENT:
402 return INTEL_PT_ERR_MISMAT;
403 case -EOVERFLOW:
404 return INTEL_PT_ERR_OVR;
405 case -ENOSPC:
406 return INTEL_PT_ERR_LOST;
407 case -ELOOP:
408 return INTEL_PT_ERR_NELOOP;
409 default:
410 return INTEL_PT_ERR_UNK;
411 }
412 }
413
414 static const char *intel_pt_err_msgs[] = {
415 [INTEL_PT_ERR_NOMEM] = "Memory allocation failed",
416 [INTEL_PT_ERR_INTERN] = "Internal error",
417 [INTEL_PT_ERR_BADPKT] = "Bad packet",
418 [INTEL_PT_ERR_NODATA] = "No more data",
419 [INTEL_PT_ERR_NOINSN] = "Failed to get instruction",
420 [INTEL_PT_ERR_MISMAT] = "Trace doesn't match instruction",
421 [INTEL_PT_ERR_OVR] = "Overflow packet",
422 [INTEL_PT_ERR_LOST] = "Lost trace data",
423 [INTEL_PT_ERR_UNK] = "Unknown error!",
424 [INTEL_PT_ERR_NELOOP] = "Never-ending loop",
425 };
426
intel_pt__strerror(int code,char * buf,size_t buflen)427 int intel_pt__strerror(int code, char *buf, size_t buflen)
428 {
429 if (code < 1 || code >= INTEL_PT_ERR_MAX)
430 code = INTEL_PT_ERR_UNK;
431 strlcpy(buf, intel_pt_err_msgs[code], buflen);
432 return 0;
433 }
434
intel_pt_calc_ip(const struct intel_pt_pkt * packet,uint64_t last_ip)435 static uint64_t intel_pt_calc_ip(const struct intel_pt_pkt *packet,
436 uint64_t last_ip)
437 {
438 uint64_t ip;
439
440 switch (packet->count) {
441 case 1:
442 ip = (last_ip & (uint64_t)0xffffffffffff0000ULL) |
443 packet->payload;
444 break;
445 case 2:
446 ip = (last_ip & (uint64_t)0xffffffff00000000ULL) |
447 packet->payload;
448 break;
449 case 3:
450 ip = packet->payload;
451 /* Sign-extend 6-byte ip */
452 if (ip & (uint64_t)0x800000000000ULL)
453 ip |= (uint64_t)0xffff000000000000ULL;
454 break;
455 case 4:
456 ip = (last_ip & (uint64_t)0xffff000000000000ULL) |
457 packet->payload;
458 break;
459 case 6:
460 ip = packet->payload;
461 break;
462 default:
463 return 0;
464 }
465
466 return ip;
467 }
468
intel_pt_set_last_ip(struct intel_pt_decoder * decoder)469 static inline void intel_pt_set_last_ip(struct intel_pt_decoder *decoder)
470 {
471 decoder->last_ip = intel_pt_calc_ip(&decoder->packet, decoder->last_ip);
472 decoder->have_last_ip = true;
473 }
474
intel_pt_set_ip(struct intel_pt_decoder * decoder)475 static inline void intel_pt_set_ip(struct intel_pt_decoder *decoder)
476 {
477 intel_pt_set_last_ip(decoder);
478 decoder->ip = decoder->last_ip;
479 }
480
intel_pt_decoder_log_packet(struct intel_pt_decoder * decoder)481 static void intel_pt_decoder_log_packet(struct intel_pt_decoder *decoder)
482 {
483 intel_pt_log_packet(&decoder->packet, decoder->pkt_len, decoder->pos,
484 decoder->buf);
485 }
486
intel_pt_bug(struct intel_pt_decoder * decoder)487 static int intel_pt_bug(struct intel_pt_decoder *decoder)
488 {
489 intel_pt_log("ERROR: Internal error\n");
490 decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
491 return -ENOSYS;
492 }
493
intel_pt_clear_tx_flags(struct intel_pt_decoder * decoder)494 static inline void intel_pt_clear_tx_flags(struct intel_pt_decoder *decoder)
495 {
496 decoder->tx_flags = 0;
497 }
498
intel_pt_update_in_tx(struct intel_pt_decoder * decoder)499 static inline void intel_pt_update_in_tx(struct intel_pt_decoder *decoder)
500 {
501 decoder->tx_flags = decoder->packet.payload & INTEL_PT_IN_TX;
502 }
503
intel_pt_bad_packet(struct intel_pt_decoder * decoder)504 static int intel_pt_bad_packet(struct intel_pt_decoder *decoder)
505 {
506 intel_pt_clear_tx_flags(decoder);
507 decoder->have_tma = false;
508 decoder->pkt_len = 1;
509 decoder->pkt_step = 1;
510 intel_pt_decoder_log_packet(decoder);
511 if (decoder->pkt_state != INTEL_PT_STATE_NO_PSB) {
512 intel_pt_log("ERROR: Bad packet\n");
513 decoder->pkt_state = INTEL_PT_STATE_ERR1;
514 }
515 return -EBADMSG;
516 }
517
intel_pt_update_sample_time(struct intel_pt_decoder * decoder)518 static inline void intel_pt_update_sample_time(struct intel_pt_decoder *decoder)
519 {
520 decoder->sample_timestamp = decoder->timestamp;
521 decoder->sample_insn_cnt = decoder->timestamp_insn_cnt;
522 }
523
intel_pt_reposition(struct intel_pt_decoder * decoder)524 static void intel_pt_reposition(struct intel_pt_decoder *decoder)
525 {
526 decoder->ip = 0;
527 decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
528 decoder->timestamp = 0;
529 decoder->have_tma = false;
530 }
531
intel_pt_get_data(struct intel_pt_decoder * decoder,bool reposition)532 static int intel_pt_get_data(struct intel_pt_decoder *decoder, bool reposition)
533 {
534 struct intel_pt_buffer buffer = { .buf = 0, };
535 int ret;
536
537 decoder->pkt_step = 0;
538
539 intel_pt_log("Getting more data\n");
540 ret = decoder->get_trace(&buffer, decoder->data);
541 if (ret)
542 return ret;
543 decoder->buf = buffer.buf;
544 decoder->len = buffer.len;
545 if (!decoder->len) {
546 intel_pt_log("No more data\n");
547 return -ENODATA;
548 }
549 decoder->buf_timestamp = buffer.ref_timestamp;
550 if (!buffer.consecutive || reposition) {
551 intel_pt_reposition(decoder);
552 decoder->ref_timestamp = buffer.ref_timestamp;
553 decoder->state.trace_nr = buffer.trace_nr;
554 intel_pt_log("Reference timestamp 0x%" PRIx64 "\n",
555 decoder->ref_timestamp);
556 return -ENOLINK;
557 }
558
559 return 0;
560 }
561
intel_pt_get_next_data(struct intel_pt_decoder * decoder,bool reposition)562 static int intel_pt_get_next_data(struct intel_pt_decoder *decoder,
563 bool reposition)
564 {
565 if (!decoder->next_buf)
566 return intel_pt_get_data(decoder, reposition);
567
568 decoder->buf = decoder->next_buf;
569 decoder->len = decoder->next_len;
570 decoder->next_buf = 0;
571 decoder->next_len = 0;
572 return 0;
573 }
574
intel_pt_get_split_packet(struct intel_pt_decoder * decoder)575 static int intel_pt_get_split_packet(struct intel_pt_decoder *decoder)
576 {
577 unsigned char *buf = decoder->temp_buf;
578 size_t old_len, len, n;
579 int ret;
580
581 old_len = decoder->len;
582 len = decoder->len;
583 memcpy(buf, decoder->buf, len);
584
585 ret = intel_pt_get_data(decoder, false);
586 if (ret) {
587 decoder->pos += old_len;
588 return ret < 0 ? ret : -EINVAL;
589 }
590
591 n = INTEL_PT_PKT_MAX_SZ - len;
592 if (n > decoder->len)
593 n = decoder->len;
594 memcpy(buf + len, decoder->buf, n);
595 len += n;
596
597 decoder->prev_pkt_ctx = decoder->pkt_ctx;
598 ret = intel_pt_get_packet(buf, len, &decoder->packet, &decoder->pkt_ctx);
599 if (ret < (int)old_len) {
600 decoder->next_buf = decoder->buf;
601 decoder->next_len = decoder->len;
602 decoder->buf = buf;
603 decoder->len = old_len;
604 return intel_pt_bad_packet(decoder);
605 }
606
607 decoder->next_buf = decoder->buf + (ret - old_len);
608 decoder->next_len = decoder->len - (ret - old_len);
609
610 decoder->buf = buf;
611 decoder->len = ret;
612
613 return ret;
614 }
615
616 struct intel_pt_pkt_info {
617 struct intel_pt_decoder *decoder;
618 struct intel_pt_pkt packet;
619 uint64_t pos;
620 int pkt_len;
621 int last_packet_type;
622 void *data;
623 };
624
625 typedef int (*intel_pt_pkt_cb_t)(struct intel_pt_pkt_info *pkt_info);
626
627 /* Lookahead packets in current buffer */
intel_pt_pkt_lookahead(struct intel_pt_decoder * decoder,intel_pt_pkt_cb_t cb,void * data)628 static int intel_pt_pkt_lookahead(struct intel_pt_decoder *decoder,
629 intel_pt_pkt_cb_t cb, void *data)
630 {
631 struct intel_pt_pkt_info pkt_info;
632 const unsigned char *buf = decoder->buf;
633 enum intel_pt_pkt_ctx pkt_ctx = decoder->pkt_ctx;
634 size_t len = decoder->len;
635 int ret;
636
637 pkt_info.decoder = decoder;
638 pkt_info.pos = decoder->pos;
639 pkt_info.pkt_len = decoder->pkt_step;
640 pkt_info.last_packet_type = decoder->last_packet_type;
641 pkt_info.data = data;
642
643 while (1) {
644 do {
645 pkt_info.pos += pkt_info.pkt_len;
646 buf += pkt_info.pkt_len;
647 len -= pkt_info.pkt_len;
648
649 if (!len)
650 return INTEL_PT_NEED_MORE_BYTES;
651
652 ret = intel_pt_get_packet(buf, len, &pkt_info.packet,
653 &pkt_ctx);
654 if (!ret)
655 return INTEL_PT_NEED_MORE_BYTES;
656 if (ret < 0)
657 return ret;
658
659 pkt_info.pkt_len = ret;
660 } while (pkt_info.packet.type == INTEL_PT_PAD);
661
662 ret = cb(&pkt_info);
663 if (ret)
664 return 0;
665
666 pkt_info.last_packet_type = pkt_info.packet.type;
667 }
668 }
669
670 struct intel_pt_calc_cyc_to_tsc_info {
671 uint64_t cycle_cnt;
672 unsigned int cbr;
673 uint32_t last_mtc;
674 uint64_t ctc_timestamp;
675 uint64_t ctc_delta;
676 uint64_t tsc_timestamp;
677 uint64_t timestamp;
678 bool have_tma;
679 bool fixup_last_mtc;
680 bool from_mtc;
681 double cbr_cyc_to_tsc;
682 };
683
684 /*
685 * MTC provides a 8-bit slice of CTC but the TMA packet only provides the lower
686 * 16 bits of CTC. If mtc_shift > 8 then some of the MTC bits are not in the CTC
687 * provided by the TMA packet. Fix-up the last_mtc calculated from the TMA
688 * packet by copying the missing bits from the current MTC assuming the least
689 * difference between the two, and that the current MTC comes after last_mtc.
690 */
intel_pt_fixup_last_mtc(uint32_t mtc,int mtc_shift,uint32_t * last_mtc)691 static void intel_pt_fixup_last_mtc(uint32_t mtc, int mtc_shift,
692 uint32_t *last_mtc)
693 {
694 uint32_t first_missing_bit = 1U << (16 - mtc_shift);
695 uint32_t mask = ~(first_missing_bit - 1);
696
697 *last_mtc |= mtc & mask;
698 if (*last_mtc >= mtc) {
699 *last_mtc -= first_missing_bit;
700 *last_mtc &= 0xff;
701 }
702 }
703
intel_pt_calc_cyc_cb(struct intel_pt_pkt_info * pkt_info)704 static int intel_pt_calc_cyc_cb(struct intel_pt_pkt_info *pkt_info)
705 {
706 struct intel_pt_decoder *decoder = pkt_info->decoder;
707 struct intel_pt_calc_cyc_to_tsc_info *data = pkt_info->data;
708 uint64_t timestamp;
709 double cyc_to_tsc;
710 unsigned int cbr;
711 uint32_t mtc, mtc_delta, ctc, fc, ctc_rem;
712
713 switch (pkt_info->packet.type) {
714 case INTEL_PT_TNT:
715 case INTEL_PT_TIP_PGE:
716 case INTEL_PT_TIP:
717 case INTEL_PT_FUP:
718 case INTEL_PT_PSB:
719 case INTEL_PT_PIP:
720 case INTEL_PT_MODE_EXEC:
721 case INTEL_PT_MODE_TSX:
722 case INTEL_PT_PSBEND:
723 case INTEL_PT_PAD:
724 case INTEL_PT_VMCS:
725 case INTEL_PT_MNT:
726 case INTEL_PT_PTWRITE:
727 case INTEL_PT_PTWRITE_IP:
728 case INTEL_PT_BBP:
729 case INTEL_PT_BIP:
730 case INTEL_PT_BEP:
731 case INTEL_PT_BEP_IP:
732 return 0;
733
734 case INTEL_PT_MTC:
735 if (!data->have_tma)
736 return 0;
737
738 mtc = pkt_info->packet.payload;
739 if (decoder->mtc_shift > 8 && data->fixup_last_mtc) {
740 data->fixup_last_mtc = false;
741 intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
742 &data->last_mtc);
743 }
744 if (mtc > data->last_mtc)
745 mtc_delta = mtc - data->last_mtc;
746 else
747 mtc_delta = mtc + 256 - data->last_mtc;
748 data->ctc_delta += mtc_delta << decoder->mtc_shift;
749 data->last_mtc = mtc;
750
751 if (decoder->tsc_ctc_mult) {
752 timestamp = data->ctc_timestamp +
753 data->ctc_delta * decoder->tsc_ctc_mult;
754 } else {
755 timestamp = data->ctc_timestamp +
756 multdiv(data->ctc_delta,
757 decoder->tsc_ctc_ratio_n,
758 decoder->tsc_ctc_ratio_d);
759 }
760
761 if (timestamp < data->timestamp)
762 return 1;
763
764 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
765 data->timestamp = timestamp;
766 return 0;
767 }
768
769 break;
770
771 case INTEL_PT_TSC:
772 /*
773 * For now, do not support using TSC packets - refer
774 * intel_pt_calc_cyc_to_tsc().
775 */
776 if (data->from_mtc)
777 return 1;
778 timestamp = pkt_info->packet.payload |
779 (data->timestamp & (0xffULL << 56));
780 if (data->from_mtc && timestamp < data->timestamp &&
781 data->timestamp - timestamp < decoder->tsc_slip)
782 return 1;
783 if (timestamp < data->timestamp)
784 timestamp += (1ULL << 56);
785 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
786 if (data->from_mtc)
787 return 1;
788 data->tsc_timestamp = timestamp;
789 data->timestamp = timestamp;
790 return 0;
791 }
792 break;
793
794 case INTEL_PT_TMA:
795 if (data->from_mtc)
796 return 1;
797
798 if (!decoder->tsc_ctc_ratio_d)
799 return 0;
800
801 ctc = pkt_info->packet.payload;
802 fc = pkt_info->packet.count;
803 ctc_rem = ctc & decoder->ctc_rem_mask;
804
805 data->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
806
807 data->ctc_timestamp = data->tsc_timestamp - fc;
808 if (decoder->tsc_ctc_mult) {
809 data->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
810 } else {
811 data->ctc_timestamp -=
812 multdiv(ctc_rem, decoder->tsc_ctc_ratio_n,
813 decoder->tsc_ctc_ratio_d);
814 }
815
816 data->ctc_delta = 0;
817 data->have_tma = true;
818 data->fixup_last_mtc = true;
819
820 return 0;
821
822 case INTEL_PT_CYC:
823 data->cycle_cnt += pkt_info->packet.payload;
824 return 0;
825
826 case INTEL_PT_CBR:
827 cbr = pkt_info->packet.payload;
828 if (data->cbr && data->cbr != cbr)
829 return 1;
830 data->cbr = cbr;
831 data->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
832 return 0;
833
834 case INTEL_PT_TIP_PGD:
835 case INTEL_PT_TRACESTOP:
836 case INTEL_PT_EXSTOP:
837 case INTEL_PT_EXSTOP_IP:
838 case INTEL_PT_MWAIT:
839 case INTEL_PT_PWRE:
840 case INTEL_PT_PWRX:
841 case INTEL_PT_OVF:
842 case INTEL_PT_BAD: /* Does not happen */
843 default:
844 return 1;
845 }
846
847 if (!data->cbr && decoder->cbr) {
848 data->cbr = decoder->cbr;
849 data->cbr_cyc_to_tsc = decoder->cbr_cyc_to_tsc;
850 }
851
852 if (!data->cycle_cnt)
853 return 1;
854
855 cyc_to_tsc = (double)(timestamp - decoder->timestamp) / data->cycle_cnt;
856
857 if (data->cbr && cyc_to_tsc > data->cbr_cyc_to_tsc &&
858 cyc_to_tsc / data->cbr_cyc_to_tsc > 1.25) {
859 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle too big (c.f. CBR-based value %g), pos " x64_fmt "\n",
860 cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
861 return 1;
862 }
863
864 decoder->calc_cyc_to_tsc = cyc_to_tsc;
865 decoder->have_calc_cyc_to_tsc = true;
866
867 if (data->cbr) {
868 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. CBR-based value %g, pos " x64_fmt "\n",
869 cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
870 } else {
871 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. unknown CBR-based value, pos " x64_fmt "\n",
872 cyc_to_tsc, pkt_info->pos);
873 }
874
875 return 1;
876 }
877
intel_pt_calc_cyc_to_tsc(struct intel_pt_decoder * decoder,bool from_mtc)878 static void intel_pt_calc_cyc_to_tsc(struct intel_pt_decoder *decoder,
879 bool from_mtc)
880 {
881 struct intel_pt_calc_cyc_to_tsc_info data = {
882 .cycle_cnt = 0,
883 .cbr = 0,
884 .last_mtc = decoder->last_mtc,
885 .ctc_timestamp = decoder->ctc_timestamp,
886 .ctc_delta = decoder->ctc_delta,
887 .tsc_timestamp = decoder->tsc_timestamp,
888 .timestamp = decoder->timestamp,
889 .have_tma = decoder->have_tma,
890 .fixup_last_mtc = decoder->fixup_last_mtc,
891 .from_mtc = from_mtc,
892 .cbr_cyc_to_tsc = 0,
893 };
894
895 /*
896 * For now, do not support using TSC packets for at least the reasons:
897 * 1) timing might have stopped
898 * 2) TSC packets within PSB+ can slip against CYC packets
899 */
900 if (!from_mtc)
901 return;
902
903 intel_pt_pkt_lookahead(decoder, intel_pt_calc_cyc_cb, &data);
904 }
905
intel_pt_get_next_packet(struct intel_pt_decoder * decoder)906 static int intel_pt_get_next_packet(struct intel_pt_decoder *decoder)
907 {
908 int ret;
909
910 decoder->last_packet_type = decoder->packet.type;
911
912 do {
913 decoder->pos += decoder->pkt_step;
914 decoder->buf += decoder->pkt_step;
915 decoder->len -= decoder->pkt_step;
916
917 if (!decoder->len) {
918 ret = intel_pt_get_next_data(decoder, false);
919 if (ret)
920 return ret;
921 }
922
923 decoder->prev_pkt_ctx = decoder->pkt_ctx;
924 ret = intel_pt_get_packet(decoder->buf, decoder->len,
925 &decoder->packet, &decoder->pkt_ctx);
926 if (ret == INTEL_PT_NEED_MORE_BYTES && BITS_PER_LONG == 32 &&
927 decoder->len < INTEL_PT_PKT_MAX_SZ && !decoder->next_buf) {
928 ret = intel_pt_get_split_packet(decoder);
929 if (ret < 0)
930 return ret;
931 }
932 if (ret <= 0)
933 return intel_pt_bad_packet(decoder);
934
935 decoder->pkt_len = ret;
936 decoder->pkt_step = ret;
937 intel_pt_decoder_log_packet(decoder);
938 } while (decoder->packet.type == INTEL_PT_PAD);
939
940 return 0;
941 }
942
intel_pt_next_period(struct intel_pt_decoder * decoder)943 static uint64_t intel_pt_next_period(struct intel_pt_decoder *decoder)
944 {
945 uint64_t timestamp, masked_timestamp;
946
947 timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
948 masked_timestamp = timestamp & decoder->period_mask;
949 if (decoder->continuous_period) {
950 if (masked_timestamp > decoder->last_masked_timestamp)
951 return 1;
952 } else {
953 timestamp += 1;
954 masked_timestamp = timestamp & decoder->period_mask;
955 if (masked_timestamp > decoder->last_masked_timestamp) {
956 decoder->last_masked_timestamp = masked_timestamp;
957 decoder->continuous_period = true;
958 }
959 }
960
961 if (masked_timestamp < decoder->last_masked_timestamp)
962 return decoder->period_ticks;
963
964 return decoder->period_ticks - (timestamp - masked_timestamp);
965 }
966
intel_pt_next_sample(struct intel_pt_decoder * decoder)967 static uint64_t intel_pt_next_sample(struct intel_pt_decoder *decoder)
968 {
969 switch (decoder->period_type) {
970 case INTEL_PT_PERIOD_INSTRUCTIONS:
971 return decoder->period - decoder->period_insn_cnt;
972 case INTEL_PT_PERIOD_TICKS:
973 return intel_pt_next_period(decoder);
974 case INTEL_PT_PERIOD_NONE:
975 case INTEL_PT_PERIOD_MTC:
976 default:
977 return 0;
978 }
979 }
980
intel_pt_sample_insn(struct intel_pt_decoder * decoder)981 static void intel_pt_sample_insn(struct intel_pt_decoder *decoder)
982 {
983 uint64_t timestamp, masked_timestamp;
984
985 switch (decoder->period_type) {
986 case INTEL_PT_PERIOD_INSTRUCTIONS:
987 decoder->period_insn_cnt = 0;
988 break;
989 case INTEL_PT_PERIOD_TICKS:
990 timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
991 masked_timestamp = timestamp & decoder->period_mask;
992 if (masked_timestamp > decoder->last_masked_timestamp)
993 decoder->last_masked_timestamp = masked_timestamp;
994 else
995 decoder->last_masked_timestamp += decoder->period_ticks;
996 break;
997 case INTEL_PT_PERIOD_NONE:
998 case INTEL_PT_PERIOD_MTC:
999 default:
1000 break;
1001 }
1002
1003 decoder->state.type |= INTEL_PT_INSTRUCTION;
1004 }
1005
intel_pt_walk_insn(struct intel_pt_decoder * decoder,struct intel_pt_insn * intel_pt_insn,uint64_t ip)1006 static int intel_pt_walk_insn(struct intel_pt_decoder *decoder,
1007 struct intel_pt_insn *intel_pt_insn, uint64_t ip)
1008 {
1009 uint64_t max_insn_cnt, insn_cnt = 0;
1010 int err;
1011
1012 if (!decoder->mtc_insn)
1013 decoder->mtc_insn = true;
1014
1015 max_insn_cnt = intel_pt_next_sample(decoder);
1016
1017 err = decoder->walk_insn(intel_pt_insn, &insn_cnt, &decoder->ip, ip,
1018 max_insn_cnt, decoder->data);
1019
1020 decoder->tot_insn_cnt += insn_cnt;
1021 decoder->timestamp_insn_cnt += insn_cnt;
1022 decoder->sample_insn_cnt += insn_cnt;
1023 decoder->period_insn_cnt += insn_cnt;
1024
1025 if (err) {
1026 decoder->no_progress = 0;
1027 decoder->pkt_state = INTEL_PT_STATE_ERR2;
1028 intel_pt_log_at("ERROR: Failed to get instruction",
1029 decoder->ip);
1030 if (err == -ENOENT)
1031 return -ENOLINK;
1032 return -EILSEQ;
1033 }
1034
1035 if (ip && decoder->ip == ip) {
1036 err = -EAGAIN;
1037 goto out;
1038 }
1039
1040 if (max_insn_cnt && insn_cnt >= max_insn_cnt)
1041 intel_pt_sample_insn(decoder);
1042
1043 if (intel_pt_insn->branch == INTEL_PT_BR_NO_BRANCH) {
1044 decoder->state.type = INTEL_PT_INSTRUCTION;
1045 decoder->state.from_ip = decoder->ip;
1046 decoder->state.to_ip = 0;
1047 decoder->ip += intel_pt_insn->length;
1048 err = INTEL_PT_RETURN;
1049 goto out;
1050 }
1051
1052 if (intel_pt_insn->op == INTEL_PT_OP_CALL) {
1053 /* Zero-length calls are excluded */
1054 if (intel_pt_insn->branch != INTEL_PT_BR_UNCONDITIONAL ||
1055 intel_pt_insn->rel) {
1056 err = intel_pt_push(&decoder->stack, decoder->ip +
1057 intel_pt_insn->length);
1058 if (err)
1059 goto out;
1060 }
1061 } else if (intel_pt_insn->op == INTEL_PT_OP_RET) {
1062 decoder->ret_addr = intel_pt_pop(&decoder->stack);
1063 }
1064
1065 if (intel_pt_insn->branch == INTEL_PT_BR_UNCONDITIONAL) {
1066 int cnt = decoder->no_progress++;
1067
1068 decoder->state.from_ip = decoder->ip;
1069 decoder->ip += intel_pt_insn->length +
1070 intel_pt_insn->rel;
1071 decoder->state.to_ip = decoder->ip;
1072 err = INTEL_PT_RETURN;
1073
1074 /*
1075 * Check for being stuck in a loop. This can happen if a
1076 * decoder error results in the decoder erroneously setting the
1077 * ip to an address that is itself in an infinite loop that
1078 * consumes no packets. When that happens, there must be an
1079 * unconditional branch.
1080 */
1081 if (cnt) {
1082 if (cnt == 1) {
1083 decoder->stuck_ip = decoder->state.to_ip;
1084 decoder->stuck_ip_prd = 1;
1085 decoder->stuck_ip_cnt = 1;
1086 } else if (cnt > INTEL_PT_MAX_LOOPS ||
1087 decoder->state.to_ip == decoder->stuck_ip) {
1088 intel_pt_log_at("ERROR: Never-ending loop",
1089 decoder->state.to_ip);
1090 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1091 err = -ELOOP;
1092 goto out;
1093 } else if (!--decoder->stuck_ip_cnt) {
1094 decoder->stuck_ip_prd += 1;
1095 decoder->stuck_ip_cnt = decoder->stuck_ip_prd;
1096 decoder->stuck_ip = decoder->state.to_ip;
1097 }
1098 }
1099 goto out_no_progress;
1100 }
1101 out:
1102 decoder->no_progress = 0;
1103 out_no_progress:
1104 decoder->state.insn_op = intel_pt_insn->op;
1105 decoder->state.insn_len = intel_pt_insn->length;
1106 memcpy(decoder->state.insn, intel_pt_insn->buf,
1107 INTEL_PT_INSN_BUF_SZ);
1108
1109 if (decoder->tx_flags & INTEL_PT_IN_TX)
1110 decoder->state.flags |= INTEL_PT_IN_TX;
1111
1112 return err;
1113 }
1114
intel_pt_fup_event(struct intel_pt_decoder * decoder)1115 static bool intel_pt_fup_event(struct intel_pt_decoder *decoder)
1116 {
1117 bool ret = false;
1118
1119 if (decoder->set_fup_tx_flags) {
1120 decoder->set_fup_tx_flags = false;
1121 decoder->tx_flags = decoder->fup_tx_flags;
1122 decoder->state.type = INTEL_PT_TRANSACTION;
1123 if (decoder->fup_tx_flags & INTEL_PT_ABORT_TX)
1124 decoder->state.type |= INTEL_PT_BRANCH;
1125 decoder->state.from_ip = decoder->ip;
1126 decoder->state.to_ip = 0;
1127 decoder->state.flags = decoder->fup_tx_flags;
1128 return true;
1129 }
1130 if (decoder->set_fup_ptw) {
1131 decoder->set_fup_ptw = false;
1132 decoder->state.type = INTEL_PT_PTW;
1133 decoder->state.flags |= INTEL_PT_FUP_IP;
1134 decoder->state.from_ip = decoder->ip;
1135 decoder->state.to_ip = 0;
1136 decoder->state.ptw_payload = decoder->fup_ptw_payload;
1137 return true;
1138 }
1139 if (decoder->set_fup_mwait) {
1140 decoder->set_fup_mwait = false;
1141 decoder->state.type = INTEL_PT_MWAIT_OP;
1142 decoder->state.from_ip = decoder->ip;
1143 decoder->state.to_ip = 0;
1144 decoder->state.mwait_payload = decoder->fup_mwait_payload;
1145 ret = true;
1146 }
1147 if (decoder->set_fup_pwre) {
1148 decoder->set_fup_pwre = false;
1149 decoder->state.type |= INTEL_PT_PWR_ENTRY;
1150 decoder->state.type &= ~INTEL_PT_BRANCH;
1151 decoder->state.from_ip = decoder->ip;
1152 decoder->state.to_ip = 0;
1153 decoder->state.pwre_payload = decoder->fup_pwre_payload;
1154 ret = true;
1155 }
1156 if (decoder->set_fup_exstop) {
1157 decoder->set_fup_exstop = false;
1158 decoder->state.type |= INTEL_PT_EX_STOP;
1159 decoder->state.type &= ~INTEL_PT_BRANCH;
1160 decoder->state.flags |= INTEL_PT_FUP_IP;
1161 decoder->state.from_ip = decoder->ip;
1162 decoder->state.to_ip = 0;
1163 ret = true;
1164 }
1165 if (decoder->set_fup_bep) {
1166 decoder->set_fup_bep = false;
1167 decoder->state.type |= INTEL_PT_BLK_ITEMS;
1168 decoder->state.type &= ~INTEL_PT_BRANCH;
1169 decoder->state.from_ip = decoder->ip;
1170 decoder->state.to_ip = 0;
1171 ret = true;
1172 }
1173 return ret;
1174 }
1175
intel_pt_fup_with_nlip(struct intel_pt_decoder * decoder,struct intel_pt_insn * intel_pt_insn,uint64_t ip,int err)1176 static inline bool intel_pt_fup_with_nlip(struct intel_pt_decoder *decoder,
1177 struct intel_pt_insn *intel_pt_insn,
1178 uint64_t ip, int err)
1179 {
1180 return decoder->flags & INTEL_PT_FUP_WITH_NLIP && !err &&
1181 intel_pt_insn->branch == INTEL_PT_BR_INDIRECT &&
1182 ip == decoder->ip + intel_pt_insn->length;
1183 }
1184
intel_pt_walk_fup(struct intel_pt_decoder * decoder)1185 static int intel_pt_walk_fup(struct intel_pt_decoder *decoder)
1186 {
1187 struct intel_pt_insn intel_pt_insn;
1188 uint64_t ip;
1189 int err;
1190
1191 ip = decoder->last_ip;
1192
1193 while (1) {
1194 err = intel_pt_walk_insn(decoder, &intel_pt_insn, ip);
1195 if (err == INTEL_PT_RETURN)
1196 return 0;
1197 if (err == -EAGAIN ||
1198 intel_pt_fup_with_nlip(decoder, &intel_pt_insn, ip, err)) {
1199 bool no_tip = decoder->pkt_state != INTEL_PT_STATE_FUP;
1200
1201 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1202 if (intel_pt_fup_event(decoder) && no_tip)
1203 return 0;
1204 return -EAGAIN;
1205 }
1206 decoder->set_fup_tx_flags = false;
1207 if (err)
1208 return err;
1209
1210 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1211 intel_pt_log_at("ERROR: Unexpected indirect branch",
1212 decoder->ip);
1213 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1214 return -ENOENT;
1215 }
1216
1217 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1218 intel_pt_log_at("ERROR: Unexpected conditional branch",
1219 decoder->ip);
1220 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1221 return -ENOENT;
1222 }
1223
1224 intel_pt_bug(decoder);
1225 }
1226 }
1227
intel_pt_walk_tip(struct intel_pt_decoder * decoder)1228 static int intel_pt_walk_tip(struct intel_pt_decoder *decoder)
1229 {
1230 struct intel_pt_insn intel_pt_insn;
1231 int err;
1232
1233 err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1234 if (err == INTEL_PT_RETURN &&
1235 decoder->pgd_ip &&
1236 decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1237 (decoder->state.type & INTEL_PT_BRANCH) &&
1238 decoder->pgd_ip(decoder->state.to_ip, decoder->data)) {
1239 /* Unconditional branch leaving filter region */
1240 decoder->no_progress = 0;
1241 decoder->pge = false;
1242 decoder->continuous_period = false;
1243 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1244 decoder->state.type |= INTEL_PT_TRACE_END;
1245 return 0;
1246 }
1247 if (err == INTEL_PT_RETURN)
1248 return 0;
1249 if (err)
1250 return err;
1251
1252 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1253 if (decoder->pkt_state == INTEL_PT_STATE_TIP_PGD) {
1254 decoder->pge = false;
1255 decoder->continuous_period = false;
1256 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1257 decoder->state.from_ip = decoder->ip;
1258 if (decoder->packet.count == 0) {
1259 decoder->state.to_ip = 0;
1260 } else {
1261 decoder->state.to_ip = decoder->last_ip;
1262 decoder->ip = decoder->last_ip;
1263 }
1264 decoder->state.type |= INTEL_PT_TRACE_END;
1265 } else {
1266 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1267 decoder->state.from_ip = decoder->ip;
1268 if (decoder->packet.count == 0) {
1269 decoder->state.to_ip = 0;
1270 } else {
1271 decoder->state.to_ip = decoder->last_ip;
1272 decoder->ip = decoder->last_ip;
1273 }
1274 }
1275 return 0;
1276 }
1277
1278 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1279 uint64_t to_ip = decoder->ip + intel_pt_insn.length +
1280 intel_pt_insn.rel;
1281
1282 if (decoder->pgd_ip &&
1283 decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1284 decoder->pgd_ip(to_ip, decoder->data)) {
1285 /* Conditional branch leaving filter region */
1286 decoder->pge = false;
1287 decoder->continuous_period = false;
1288 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1289 decoder->ip = to_ip;
1290 decoder->state.from_ip = decoder->ip;
1291 decoder->state.to_ip = to_ip;
1292 decoder->state.type |= INTEL_PT_TRACE_END;
1293 return 0;
1294 }
1295 intel_pt_log_at("ERROR: Conditional branch when expecting indirect branch",
1296 decoder->ip);
1297 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1298 return -ENOENT;
1299 }
1300
1301 return intel_pt_bug(decoder);
1302 }
1303
intel_pt_walk_tnt(struct intel_pt_decoder * decoder)1304 static int intel_pt_walk_tnt(struct intel_pt_decoder *decoder)
1305 {
1306 struct intel_pt_insn intel_pt_insn;
1307 int err;
1308
1309 while (1) {
1310 err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1311 if (err == INTEL_PT_RETURN)
1312 return 0;
1313 if (err)
1314 return err;
1315
1316 if (intel_pt_insn.op == INTEL_PT_OP_RET) {
1317 if (!decoder->return_compression) {
1318 intel_pt_log_at("ERROR: RET when expecting conditional branch",
1319 decoder->ip);
1320 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1321 return -ENOENT;
1322 }
1323 if (!decoder->ret_addr) {
1324 intel_pt_log_at("ERROR: Bad RET compression (stack empty)",
1325 decoder->ip);
1326 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1327 return -ENOENT;
1328 }
1329 if (!(decoder->tnt.payload & BIT63)) {
1330 intel_pt_log_at("ERROR: Bad RET compression (TNT=N)",
1331 decoder->ip);
1332 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1333 return -ENOENT;
1334 }
1335 decoder->tnt.count -= 1;
1336 if (decoder->tnt.count)
1337 decoder->pkt_state = INTEL_PT_STATE_TNT_CONT;
1338 else
1339 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1340 decoder->tnt.payload <<= 1;
1341 decoder->state.from_ip = decoder->ip;
1342 decoder->ip = decoder->ret_addr;
1343 decoder->state.to_ip = decoder->ip;
1344 return 0;
1345 }
1346
1347 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1348 /* Handle deferred TIPs */
1349 err = intel_pt_get_next_packet(decoder);
1350 if (err)
1351 return err;
1352 if (decoder->packet.type != INTEL_PT_TIP ||
1353 decoder->packet.count == 0) {
1354 intel_pt_log_at("ERROR: Missing deferred TIP for indirect branch",
1355 decoder->ip);
1356 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1357 decoder->pkt_step = 0;
1358 return -ENOENT;
1359 }
1360 intel_pt_set_last_ip(decoder);
1361 decoder->state.from_ip = decoder->ip;
1362 decoder->state.to_ip = decoder->last_ip;
1363 decoder->ip = decoder->last_ip;
1364 return 0;
1365 }
1366
1367 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1368 decoder->tnt.count -= 1;
1369 if (decoder->tnt.count)
1370 decoder->pkt_state = INTEL_PT_STATE_TNT_CONT;
1371 else
1372 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1373 if (decoder->tnt.payload & BIT63) {
1374 decoder->tnt.payload <<= 1;
1375 decoder->state.from_ip = decoder->ip;
1376 decoder->ip += intel_pt_insn.length +
1377 intel_pt_insn.rel;
1378 decoder->state.to_ip = decoder->ip;
1379 return 0;
1380 }
1381 /* Instruction sample for a non-taken branch */
1382 if (decoder->state.type & INTEL_PT_INSTRUCTION) {
1383 decoder->tnt.payload <<= 1;
1384 decoder->state.type = INTEL_PT_INSTRUCTION;
1385 decoder->state.from_ip = decoder->ip;
1386 decoder->state.to_ip = 0;
1387 decoder->ip += intel_pt_insn.length;
1388 return 0;
1389 }
1390 decoder->sample_cyc = false;
1391 decoder->ip += intel_pt_insn.length;
1392 if (!decoder->tnt.count) {
1393 intel_pt_update_sample_time(decoder);
1394 return -EAGAIN;
1395 }
1396 decoder->tnt.payload <<= 1;
1397 continue;
1398 }
1399
1400 return intel_pt_bug(decoder);
1401 }
1402 }
1403
intel_pt_mode_tsx(struct intel_pt_decoder * decoder,bool * no_tip)1404 static int intel_pt_mode_tsx(struct intel_pt_decoder *decoder, bool *no_tip)
1405 {
1406 unsigned int fup_tx_flags;
1407 int err;
1408
1409 fup_tx_flags = decoder->packet.payload &
1410 (INTEL_PT_IN_TX | INTEL_PT_ABORT_TX);
1411 err = intel_pt_get_next_packet(decoder);
1412 if (err)
1413 return err;
1414 if (decoder->packet.type == INTEL_PT_FUP) {
1415 decoder->fup_tx_flags = fup_tx_flags;
1416 decoder->set_fup_tx_flags = true;
1417 if (!(decoder->fup_tx_flags & INTEL_PT_ABORT_TX))
1418 *no_tip = true;
1419 } else {
1420 intel_pt_log_at("ERROR: Missing FUP after MODE.TSX",
1421 decoder->pos);
1422 intel_pt_update_in_tx(decoder);
1423 }
1424 return 0;
1425 }
1426
intel_pt_8b_tsc(uint64_t timestamp,uint64_t ref_timestamp)1427 static uint64_t intel_pt_8b_tsc(uint64_t timestamp, uint64_t ref_timestamp)
1428 {
1429 timestamp |= (ref_timestamp & (0xffULL << 56));
1430
1431 if (timestamp < ref_timestamp) {
1432 if (ref_timestamp - timestamp > (1ULL << 55))
1433 timestamp += (1ULL << 56);
1434 } else {
1435 if (timestamp - ref_timestamp > (1ULL << 55))
1436 timestamp -= (1ULL << 56);
1437 }
1438
1439 return timestamp;
1440 }
1441
intel_pt_calc_tsc_timestamp(struct intel_pt_decoder * decoder)1442 static void intel_pt_calc_tsc_timestamp(struct intel_pt_decoder *decoder)
1443 {
1444 uint64_t timestamp;
1445
1446 decoder->have_tma = false;
1447
1448 if (decoder->ref_timestamp) {
1449 timestamp = intel_pt_8b_tsc(decoder->packet.payload,
1450 decoder->ref_timestamp);
1451 decoder->tsc_timestamp = timestamp;
1452 decoder->timestamp = timestamp;
1453 decoder->ref_timestamp = 0;
1454 decoder->timestamp_insn_cnt = 0;
1455 } else if (decoder->timestamp) {
1456 timestamp = decoder->packet.payload |
1457 (decoder->timestamp & (0xffULL << 56));
1458 decoder->tsc_timestamp = timestamp;
1459 if (timestamp < decoder->timestamp &&
1460 decoder->timestamp - timestamp < decoder->tsc_slip) {
1461 intel_pt_log_to("Suppressing backwards timestamp",
1462 timestamp);
1463 timestamp = decoder->timestamp;
1464 }
1465 if (timestamp < decoder->timestamp) {
1466 intel_pt_log_to("Wraparound timestamp", timestamp);
1467 timestamp += (1ULL << 56);
1468 decoder->tsc_timestamp = timestamp;
1469 }
1470 decoder->timestamp = timestamp;
1471 decoder->timestamp_insn_cnt = 0;
1472 }
1473
1474 if (decoder->last_packet_type == INTEL_PT_CYC) {
1475 decoder->cyc_ref_timestamp = decoder->timestamp;
1476 decoder->cycle_cnt = 0;
1477 decoder->have_calc_cyc_to_tsc = false;
1478 intel_pt_calc_cyc_to_tsc(decoder, false);
1479 }
1480
1481 intel_pt_log_to("Setting timestamp", decoder->timestamp);
1482 }
1483
intel_pt_overflow(struct intel_pt_decoder * decoder)1484 static int intel_pt_overflow(struct intel_pt_decoder *decoder)
1485 {
1486 intel_pt_log("ERROR: Buffer overflow\n");
1487 intel_pt_clear_tx_flags(decoder);
1488 decoder->timestamp_insn_cnt = 0;
1489 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1490 decoder->overflow = true;
1491 return -EOVERFLOW;
1492 }
1493
intel_pt_mtc_cyc_cnt_pge(struct intel_pt_decoder * decoder)1494 static inline void intel_pt_mtc_cyc_cnt_pge(struct intel_pt_decoder *decoder)
1495 {
1496 if (decoder->have_cyc)
1497 return;
1498
1499 decoder->cyc_cnt_timestamp = decoder->timestamp;
1500 decoder->base_cyc_cnt = decoder->tot_cyc_cnt;
1501 }
1502
intel_pt_mtc_cyc_cnt_cbr(struct intel_pt_decoder * decoder)1503 static inline void intel_pt_mtc_cyc_cnt_cbr(struct intel_pt_decoder *decoder)
1504 {
1505 decoder->tsc_to_cyc = decoder->cbr / decoder->max_non_turbo_ratio_fp;
1506
1507 if (decoder->pge)
1508 intel_pt_mtc_cyc_cnt_pge(decoder);
1509 }
1510
intel_pt_mtc_cyc_cnt_upd(struct intel_pt_decoder * decoder)1511 static inline void intel_pt_mtc_cyc_cnt_upd(struct intel_pt_decoder *decoder)
1512 {
1513 uint64_t tot_cyc_cnt, tsc_delta;
1514
1515 if (decoder->have_cyc)
1516 return;
1517
1518 decoder->sample_cyc = true;
1519
1520 if (!decoder->pge || decoder->timestamp <= decoder->cyc_cnt_timestamp)
1521 return;
1522
1523 tsc_delta = decoder->timestamp - decoder->cyc_cnt_timestamp;
1524 tot_cyc_cnt = tsc_delta * decoder->tsc_to_cyc + decoder->base_cyc_cnt;
1525
1526 if (tot_cyc_cnt > decoder->tot_cyc_cnt)
1527 decoder->tot_cyc_cnt = tot_cyc_cnt;
1528 }
1529
intel_pt_calc_tma(struct intel_pt_decoder * decoder)1530 static void intel_pt_calc_tma(struct intel_pt_decoder *decoder)
1531 {
1532 uint32_t ctc = decoder->packet.payload;
1533 uint32_t fc = decoder->packet.count;
1534 uint32_t ctc_rem = ctc & decoder->ctc_rem_mask;
1535
1536 if (!decoder->tsc_ctc_ratio_d)
1537 return;
1538
1539 if (decoder->pge && !decoder->in_psb)
1540 intel_pt_mtc_cyc_cnt_pge(decoder);
1541 else
1542 intel_pt_mtc_cyc_cnt_upd(decoder);
1543
1544 decoder->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
1545 decoder->ctc_timestamp = decoder->tsc_timestamp - fc;
1546 if (decoder->tsc_ctc_mult) {
1547 decoder->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
1548 } else {
1549 decoder->ctc_timestamp -= multdiv(ctc_rem,
1550 decoder->tsc_ctc_ratio_n,
1551 decoder->tsc_ctc_ratio_d);
1552 }
1553 decoder->ctc_delta = 0;
1554 decoder->have_tma = true;
1555 decoder->fixup_last_mtc = true;
1556 intel_pt_log("CTC timestamp " x64_fmt " last MTC %#x CTC rem %#x\n",
1557 decoder->ctc_timestamp, decoder->last_mtc, ctc_rem);
1558 }
1559
intel_pt_calc_mtc_timestamp(struct intel_pt_decoder * decoder)1560 static void intel_pt_calc_mtc_timestamp(struct intel_pt_decoder *decoder)
1561 {
1562 uint64_t timestamp;
1563 uint32_t mtc, mtc_delta;
1564
1565 if (!decoder->have_tma)
1566 return;
1567
1568 mtc = decoder->packet.payload;
1569
1570 if (decoder->mtc_shift > 8 && decoder->fixup_last_mtc) {
1571 decoder->fixup_last_mtc = false;
1572 intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
1573 &decoder->last_mtc);
1574 }
1575
1576 if (mtc > decoder->last_mtc)
1577 mtc_delta = mtc - decoder->last_mtc;
1578 else
1579 mtc_delta = mtc + 256 - decoder->last_mtc;
1580
1581 decoder->ctc_delta += mtc_delta << decoder->mtc_shift;
1582
1583 if (decoder->tsc_ctc_mult) {
1584 timestamp = decoder->ctc_timestamp +
1585 decoder->ctc_delta * decoder->tsc_ctc_mult;
1586 } else {
1587 timestamp = decoder->ctc_timestamp +
1588 multdiv(decoder->ctc_delta,
1589 decoder->tsc_ctc_ratio_n,
1590 decoder->tsc_ctc_ratio_d);
1591 }
1592
1593 if (timestamp < decoder->timestamp)
1594 intel_pt_log("Suppressing MTC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1595 timestamp, decoder->timestamp);
1596 else
1597 decoder->timestamp = timestamp;
1598
1599 intel_pt_mtc_cyc_cnt_upd(decoder);
1600
1601 decoder->timestamp_insn_cnt = 0;
1602 decoder->last_mtc = mtc;
1603
1604 if (decoder->last_packet_type == INTEL_PT_CYC) {
1605 decoder->cyc_ref_timestamp = decoder->timestamp;
1606 decoder->cycle_cnt = 0;
1607 decoder->have_calc_cyc_to_tsc = false;
1608 intel_pt_calc_cyc_to_tsc(decoder, true);
1609 }
1610
1611 intel_pt_log_to("Setting timestamp", decoder->timestamp);
1612 }
1613
intel_pt_calc_cbr(struct intel_pt_decoder * decoder)1614 static void intel_pt_calc_cbr(struct intel_pt_decoder *decoder)
1615 {
1616 unsigned int cbr = decoder->packet.payload & 0xff;
1617
1618 decoder->cbr_payload = decoder->packet.payload;
1619
1620 if (decoder->cbr == cbr)
1621 return;
1622
1623 decoder->cbr = cbr;
1624 decoder->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
1625
1626 intel_pt_mtc_cyc_cnt_cbr(decoder);
1627 }
1628
intel_pt_calc_cyc_timestamp(struct intel_pt_decoder * decoder)1629 static void intel_pt_calc_cyc_timestamp(struct intel_pt_decoder *decoder)
1630 {
1631 uint64_t timestamp = decoder->cyc_ref_timestamp;
1632
1633 decoder->have_cyc = true;
1634
1635 decoder->cycle_cnt += decoder->packet.payload;
1636 if (decoder->pge)
1637 decoder->tot_cyc_cnt += decoder->packet.payload;
1638 decoder->sample_cyc = true;
1639
1640 if (!decoder->cyc_ref_timestamp)
1641 return;
1642
1643 if (decoder->have_calc_cyc_to_tsc)
1644 timestamp += decoder->cycle_cnt * decoder->calc_cyc_to_tsc;
1645 else if (decoder->cbr)
1646 timestamp += decoder->cycle_cnt * decoder->cbr_cyc_to_tsc;
1647 else
1648 return;
1649
1650 if (timestamp < decoder->timestamp)
1651 intel_pt_log("Suppressing CYC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1652 timestamp, decoder->timestamp);
1653 else
1654 decoder->timestamp = timestamp;
1655
1656 decoder->timestamp_insn_cnt = 0;
1657
1658 intel_pt_log_to("Setting timestamp", decoder->timestamp);
1659 }
1660
intel_pt_bbp(struct intel_pt_decoder * decoder)1661 static void intel_pt_bbp(struct intel_pt_decoder *decoder)
1662 {
1663 if (decoder->prev_pkt_ctx == INTEL_PT_NO_CTX) {
1664 memset(decoder->state.items.mask, 0, sizeof(decoder->state.items.mask));
1665 decoder->state.items.is_32_bit = false;
1666 }
1667 decoder->blk_type = decoder->packet.payload;
1668 decoder->blk_type_pos = intel_pt_blk_type_pos(decoder->blk_type);
1669 if (decoder->blk_type == INTEL_PT_GP_REGS)
1670 decoder->state.items.is_32_bit = decoder->packet.count;
1671 if (decoder->blk_type_pos < 0) {
1672 intel_pt_log("WARNING: Unknown block type %u\n",
1673 decoder->blk_type);
1674 } else if (decoder->state.items.mask[decoder->blk_type_pos]) {
1675 intel_pt_log("WARNING: Duplicate block type %u\n",
1676 decoder->blk_type);
1677 }
1678 }
1679
intel_pt_bip(struct intel_pt_decoder * decoder)1680 static void intel_pt_bip(struct intel_pt_decoder *decoder)
1681 {
1682 uint32_t id = decoder->packet.count;
1683 uint32_t bit = 1 << id;
1684 int pos = decoder->blk_type_pos;
1685
1686 if (pos < 0 || id >= INTEL_PT_BLK_ITEM_ID_CNT) {
1687 intel_pt_log("WARNING: Unknown block item %u type %d\n",
1688 id, decoder->blk_type);
1689 return;
1690 }
1691
1692 if (decoder->state.items.mask[pos] & bit) {
1693 intel_pt_log("WARNING: Duplicate block item %u type %d\n",
1694 id, decoder->blk_type);
1695 }
1696
1697 decoder->state.items.mask[pos] |= bit;
1698 decoder->state.items.val[pos][id] = decoder->packet.payload;
1699 }
1700
1701 /* Walk PSB+ packets when already in sync. */
intel_pt_walk_psbend(struct intel_pt_decoder * decoder)1702 static int intel_pt_walk_psbend(struct intel_pt_decoder *decoder)
1703 {
1704 int err;
1705
1706 decoder->in_psb = true;
1707
1708 while (1) {
1709 err = intel_pt_get_next_packet(decoder);
1710 if (err)
1711 goto out;
1712
1713 switch (decoder->packet.type) {
1714 case INTEL_PT_PSBEND:
1715 err = 0;
1716 goto out;
1717
1718 case INTEL_PT_TIP_PGD:
1719 case INTEL_PT_TIP_PGE:
1720 case INTEL_PT_TIP:
1721 case INTEL_PT_TNT:
1722 case INTEL_PT_TRACESTOP:
1723 case INTEL_PT_BAD:
1724 case INTEL_PT_PSB:
1725 case INTEL_PT_PTWRITE:
1726 case INTEL_PT_PTWRITE_IP:
1727 case INTEL_PT_EXSTOP:
1728 case INTEL_PT_EXSTOP_IP:
1729 case INTEL_PT_MWAIT:
1730 case INTEL_PT_PWRE:
1731 case INTEL_PT_PWRX:
1732 case INTEL_PT_BBP:
1733 case INTEL_PT_BIP:
1734 case INTEL_PT_BEP:
1735 case INTEL_PT_BEP_IP:
1736 decoder->have_tma = false;
1737 intel_pt_log("ERROR: Unexpected packet\n");
1738 err = -EAGAIN;
1739 goto out;
1740
1741 case INTEL_PT_OVF:
1742 err = intel_pt_overflow(decoder);
1743 goto out;
1744
1745 case INTEL_PT_TSC:
1746 intel_pt_calc_tsc_timestamp(decoder);
1747 break;
1748
1749 case INTEL_PT_TMA:
1750 intel_pt_calc_tma(decoder);
1751 break;
1752
1753 case INTEL_PT_CBR:
1754 intel_pt_calc_cbr(decoder);
1755 break;
1756
1757 case INTEL_PT_MODE_EXEC:
1758 decoder->exec_mode = decoder->packet.payload;
1759 break;
1760
1761 case INTEL_PT_PIP:
1762 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1763 break;
1764
1765 case INTEL_PT_FUP:
1766 decoder->pge = true;
1767 if (decoder->packet.count) {
1768 intel_pt_set_last_ip(decoder);
1769 if (decoder->hop) {
1770 /* Act on FUP at PSBEND */
1771 decoder->ip = decoder->last_ip;
1772 decoder->hop_psb_fup = true;
1773 }
1774 }
1775 break;
1776
1777 case INTEL_PT_MODE_TSX:
1778 intel_pt_update_in_tx(decoder);
1779 break;
1780
1781 case INTEL_PT_MTC:
1782 intel_pt_calc_mtc_timestamp(decoder);
1783 if (decoder->period_type == INTEL_PT_PERIOD_MTC)
1784 decoder->state.type |= INTEL_PT_INSTRUCTION;
1785 break;
1786
1787 case INTEL_PT_CYC:
1788 intel_pt_calc_cyc_timestamp(decoder);
1789 break;
1790
1791 case INTEL_PT_VMCS:
1792 case INTEL_PT_MNT:
1793 case INTEL_PT_PAD:
1794 default:
1795 break;
1796 }
1797 }
1798 out:
1799 decoder->in_psb = false;
1800
1801 return err;
1802 }
1803
intel_pt_walk_fup_tip(struct intel_pt_decoder * decoder)1804 static int intel_pt_walk_fup_tip(struct intel_pt_decoder *decoder)
1805 {
1806 int err;
1807
1808 if (decoder->tx_flags & INTEL_PT_ABORT_TX) {
1809 decoder->tx_flags = 0;
1810 decoder->state.flags &= ~INTEL_PT_IN_TX;
1811 decoder->state.flags |= INTEL_PT_ABORT_TX;
1812 } else {
1813 decoder->state.flags |= INTEL_PT_ASYNC;
1814 }
1815
1816 while (1) {
1817 err = intel_pt_get_next_packet(decoder);
1818 if (err)
1819 return err;
1820
1821 switch (decoder->packet.type) {
1822 case INTEL_PT_TNT:
1823 case INTEL_PT_FUP:
1824 case INTEL_PT_TRACESTOP:
1825 case INTEL_PT_PSB:
1826 case INTEL_PT_TSC:
1827 case INTEL_PT_TMA:
1828 case INTEL_PT_MODE_TSX:
1829 case INTEL_PT_BAD:
1830 case INTEL_PT_PSBEND:
1831 case INTEL_PT_PTWRITE:
1832 case INTEL_PT_PTWRITE_IP:
1833 case INTEL_PT_EXSTOP:
1834 case INTEL_PT_EXSTOP_IP:
1835 case INTEL_PT_MWAIT:
1836 case INTEL_PT_PWRE:
1837 case INTEL_PT_PWRX:
1838 case INTEL_PT_BBP:
1839 case INTEL_PT_BIP:
1840 case INTEL_PT_BEP:
1841 case INTEL_PT_BEP_IP:
1842 intel_pt_log("ERROR: Missing TIP after FUP\n");
1843 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1844 decoder->pkt_step = 0;
1845 return -ENOENT;
1846
1847 case INTEL_PT_CBR:
1848 intel_pt_calc_cbr(decoder);
1849 break;
1850
1851 case INTEL_PT_OVF:
1852 return intel_pt_overflow(decoder);
1853
1854 case INTEL_PT_TIP_PGD:
1855 decoder->state.from_ip = decoder->ip;
1856 if (decoder->packet.count == 0) {
1857 decoder->state.to_ip = 0;
1858 } else {
1859 intel_pt_set_ip(decoder);
1860 decoder->state.to_ip = decoder->ip;
1861 }
1862 decoder->pge = false;
1863 decoder->continuous_period = false;
1864 decoder->state.type |= INTEL_PT_TRACE_END;
1865 return 0;
1866
1867 case INTEL_PT_TIP_PGE:
1868 decoder->pge = true;
1869 intel_pt_log("Omitting PGE ip " x64_fmt "\n",
1870 decoder->ip);
1871 decoder->state.from_ip = 0;
1872 if (decoder->packet.count == 0) {
1873 decoder->state.to_ip = 0;
1874 } else {
1875 intel_pt_set_ip(decoder);
1876 decoder->state.to_ip = decoder->ip;
1877 }
1878 decoder->state.type |= INTEL_PT_TRACE_BEGIN;
1879 intel_pt_mtc_cyc_cnt_pge(decoder);
1880 return 0;
1881
1882 case INTEL_PT_TIP:
1883 decoder->state.from_ip = decoder->ip;
1884 if (decoder->packet.count == 0) {
1885 decoder->state.to_ip = 0;
1886 } else {
1887 intel_pt_set_ip(decoder);
1888 decoder->state.to_ip = decoder->ip;
1889 }
1890 return 0;
1891
1892 case INTEL_PT_PIP:
1893 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1894 break;
1895
1896 case INTEL_PT_MTC:
1897 intel_pt_calc_mtc_timestamp(decoder);
1898 if (decoder->period_type == INTEL_PT_PERIOD_MTC)
1899 decoder->state.type |= INTEL_PT_INSTRUCTION;
1900 break;
1901
1902 case INTEL_PT_CYC:
1903 intel_pt_calc_cyc_timestamp(decoder);
1904 break;
1905
1906 case INTEL_PT_MODE_EXEC:
1907 decoder->exec_mode = decoder->packet.payload;
1908 break;
1909
1910 case INTEL_PT_VMCS:
1911 case INTEL_PT_MNT:
1912 case INTEL_PT_PAD:
1913 break;
1914
1915 default:
1916 return intel_pt_bug(decoder);
1917 }
1918 }
1919 }
1920
intel_pt_resample(struct intel_pt_decoder * decoder)1921 static int intel_pt_resample(struct intel_pt_decoder *decoder)
1922 {
1923 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1924 decoder->state.type = INTEL_PT_INSTRUCTION;
1925 decoder->state.from_ip = decoder->ip;
1926 decoder->state.to_ip = 0;
1927 return 0;
1928 }
1929
1930 #define HOP_PROCESS 0
1931 #define HOP_IGNORE 1
1932 #define HOP_RETURN 2
1933 #define HOP_AGAIN 3
1934
1935 static int intel_pt_scan_for_psb(struct intel_pt_decoder *decoder);
1936
1937 /* Hop mode: Ignore TNT, do not walk code, but get ip from FUPs and TIPs */
intel_pt_hop_trace(struct intel_pt_decoder * decoder,bool * no_tip,int * err)1938 static int intel_pt_hop_trace(struct intel_pt_decoder *decoder, bool *no_tip, int *err)
1939 {
1940 /* Leap from PSB to PSB, getting ip from FUP within PSB+ */
1941 if (decoder->leap && !decoder->in_psb && decoder->packet.type != INTEL_PT_PSB) {
1942 *err = intel_pt_scan_for_psb(decoder);
1943 if (*err)
1944 return HOP_RETURN;
1945 }
1946
1947 switch (decoder->packet.type) {
1948 case INTEL_PT_TNT:
1949 return HOP_IGNORE;
1950
1951 case INTEL_PT_TIP_PGD:
1952 if (!decoder->packet.count)
1953 return HOP_IGNORE;
1954 intel_pt_set_ip(decoder);
1955 decoder->state.type |= INTEL_PT_TRACE_END;
1956 decoder->state.from_ip = 0;
1957 decoder->state.to_ip = decoder->ip;
1958 return HOP_RETURN;
1959
1960 case INTEL_PT_TIP:
1961 if (!decoder->packet.count)
1962 return HOP_IGNORE;
1963 intel_pt_set_ip(decoder);
1964 decoder->state.type = INTEL_PT_INSTRUCTION;
1965 decoder->state.from_ip = decoder->ip;
1966 decoder->state.to_ip = 0;
1967 return HOP_RETURN;
1968
1969 case INTEL_PT_FUP:
1970 if (!decoder->packet.count)
1971 return HOP_IGNORE;
1972 intel_pt_set_ip(decoder);
1973 if (intel_pt_fup_event(decoder))
1974 return HOP_RETURN;
1975 if (!decoder->branch_enable)
1976 *no_tip = true;
1977 if (*no_tip) {
1978 decoder->state.type = INTEL_PT_INSTRUCTION;
1979 decoder->state.from_ip = decoder->ip;
1980 decoder->state.to_ip = 0;
1981 return HOP_RETURN;
1982 }
1983 *err = intel_pt_walk_fup_tip(decoder);
1984 if (!*err)
1985 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
1986 return HOP_RETURN;
1987
1988 case INTEL_PT_PSB:
1989 decoder->last_ip = 0;
1990 decoder->have_last_ip = true;
1991 decoder->hop_psb_fup = false;
1992 *err = intel_pt_walk_psbend(decoder);
1993 if (*err == -EAGAIN)
1994 return HOP_AGAIN;
1995 if (*err)
1996 return HOP_RETURN;
1997 if (decoder->hop_psb_fup) {
1998 decoder->hop_psb_fup = false;
1999 decoder->state.type = INTEL_PT_INSTRUCTION;
2000 decoder->state.from_ip = decoder->ip;
2001 decoder->state.to_ip = 0;
2002 return HOP_RETURN;
2003 }
2004 if (decoder->cbr != decoder->cbr_seen) {
2005 decoder->state.type = 0;
2006 return HOP_RETURN;
2007 }
2008 return HOP_IGNORE;
2009
2010 case INTEL_PT_BAD:
2011 case INTEL_PT_PAD:
2012 case INTEL_PT_TIP_PGE:
2013 case INTEL_PT_TSC:
2014 case INTEL_PT_TMA:
2015 case INTEL_PT_MODE_EXEC:
2016 case INTEL_PT_MODE_TSX:
2017 case INTEL_PT_MTC:
2018 case INTEL_PT_CYC:
2019 case INTEL_PT_VMCS:
2020 case INTEL_PT_PSBEND:
2021 case INTEL_PT_CBR:
2022 case INTEL_PT_TRACESTOP:
2023 case INTEL_PT_PIP:
2024 case INTEL_PT_OVF:
2025 case INTEL_PT_MNT:
2026 case INTEL_PT_PTWRITE:
2027 case INTEL_PT_PTWRITE_IP:
2028 case INTEL_PT_EXSTOP:
2029 case INTEL_PT_EXSTOP_IP:
2030 case INTEL_PT_MWAIT:
2031 case INTEL_PT_PWRE:
2032 case INTEL_PT_PWRX:
2033 case INTEL_PT_BBP:
2034 case INTEL_PT_BIP:
2035 case INTEL_PT_BEP:
2036 case INTEL_PT_BEP_IP:
2037 default:
2038 return HOP_PROCESS;
2039 }
2040 }
2041
intel_pt_walk_trace(struct intel_pt_decoder * decoder)2042 static int intel_pt_walk_trace(struct intel_pt_decoder *decoder)
2043 {
2044 int last_packet_type = INTEL_PT_PAD;
2045 bool no_tip = false;
2046 int err;
2047
2048 while (1) {
2049 err = intel_pt_get_next_packet(decoder);
2050 if (err)
2051 return err;
2052 next:
2053 if (decoder->cyc_threshold) {
2054 if (decoder->sample_cyc && last_packet_type != INTEL_PT_CYC)
2055 decoder->sample_cyc = false;
2056 last_packet_type = decoder->packet.type;
2057 }
2058
2059 if (decoder->hop) {
2060 switch (intel_pt_hop_trace(decoder, &no_tip, &err)) {
2061 case HOP_IGNORE:
2062 continue;
2063 case HOP_RETURN:
2064 return err;
2065 case HOP_AGAIN:
2066 goto next;
2067 default:
2068 break;
2069 }
2070 }
2071
2072 switch (decoder->packet.type) {
2073 case INTEL_PT_TNT:
2074 if (!decoder->packet.count)
2075 break;
2076 decoder->tnt = decoder->packet;
2077 decoder->pkt_state = INTEL_PT_STATE_TNT;
2078 err = intel_pt_walk_tnt(decoder);
2079 if (err == -EAGAIN)
2080 break;
2081 return err;
2082
2083 case INTEL_PT_TIP_PGD:
2084 if (decoder->packet.count != 0)
2085 intel_pt_set_last_ip(decoder);
2086 decoder->pkt_state = INTEL_PT_STATE_TIP_PGD;
2087 return intel_pt_walk_tip(decoder);
2088
2089 case INTEL_PT_TIP_PGE: {
2090 decoder->pge = true;
2091 intel_pt_mtc_cyc_cnt_pge(decoder);
2092 if (decoder->packet.count == 0) {
2093 intel_pt_log_at("Skipping zero TIP.PGE",
2094 decoder->pos);
2095 break;
2096 }
2097 intel_pt_set_ip(decoder);
2098 decoder->state.from_ip = 0;
2099 decoder->state.to_ip = decoder->ip;
2100 decoder->state.type |= INTEL_PT_TRACE_BEGIN;
2101 /*
2102 * In hop mode, resample to get the to_ip as an
2103 * "instruction" sample.
2104 */
2105 if (decoder->hop)
2106 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
2107 return 0;
2108 }
2109
2110 case INTEL_PT_OVF:
2111 return intel_pt_overflow(decoder);
2112
2113 case INTEL_PT_TIP:
2114 if (decoder->packet.count != 0)
2115 intel_pt_set_last_ip(decoder);
2116 decoder->pkt_state = INTEL_PT_STATE_TIP;
2117 return intel_pt_walk_tip(decoder);
2118
2119 case INTEL_PT_FUP:
2120 if (decoder->packet.count == 0) {
2121 intel_pt_log_at("Skipping zero FUP",
2122 decoder->pos);
2123 no_tip = false;
2124 break;
2125 }
2126 intel_pt_set_last_ip(decoder);
2127 if (!decoder->branch_enable) {
2128 decoder->ip = decoder->last_ip;
2129 if (intel_pt_fup_event(decoder))
2130 return 0;
2131 no_tip = false;
2132 break;
2133 }
2134 if (decoder->set_fup_mwait)
2135 no_tip = true;
2136 if (no_tip)
2137 decoder->pkt_state = INTEL_PT_STATE_FUP_NO_TIP;
2138 else
2139 decoder->pkt_state = INTEL_PT_STATE_FUP;
2140 err = intel_pt_walk_fup(decoder);
2141 if (err != -EAGAIN)
2142 return err;
2143 if (no_tip) {
2144 no_tip = false;
2145 break;
2146 }
2147 return intel_pt_walk_fup_tip(decoder);
2148
2149 case INTEL_PT_TRACESTOP:
2150 decoder->pge = false;
2151 decoder->continuous_period = false;
2152 intel_pt_clear_tx_flags(decoder);
2153 decoder->have_tma = false;
2154 break;
2155
2156 case INTEL_PT_PSB:
2157 decoder->last_ip = 0;
2158 decoder->have_last_ip = true;
2159 intel_pt_clear_stack(&decoder->stack);
2160 err = intel_pt_walk_psbend(decoder);
2161 if (err == -EAGAIN)
2162 goto next;
2163 if (err)
2164 return err;
2165 /*
2166 * PSB+ CBR will not have changed but cater for the
2167 * possibility of another CBR change that gets caught up
2168 * in the PSB+.
2169 */
2170 if (decoder->cbr != decoder->cbr_seen) {
2171 decoder->state.type = 0;
2172 return 0;
2173 }
2174 break;
2175
2176 case INTEL_PT_PIP:
2177 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
2178 break;
2179
2180 case INTEL_PT_MTC:
2181 intel_pt_calc_mtc_timestamp(decoder);
2182 if (decoder->period_type != INTEL_PT_PERIOD_MTC)
2183 break;
2184 /*
2185 * Ensure that there has been an instruction since the
2186 * last MTC.
2187 */
2188 if (!decoder->mtc_insn)
2189 break;
2190 decoder->mtc_insn = false;
2191 /* Ensure that there is a timestamp */
2192 if (!decoder->timestamp)
2193 break;
2194 decoder->state.type = INTEL_PT_INSTRUCTION;
2195 decoder->state.from_ip = decoder->ip;
2196 decoder->state.to_ip = 0;
2197 decoder->mtc_insn = false;
2198 return 0;
2199
2200 case INTEL_PT_TSC:
2201 intel_pt_calc_tsc_timestamp(decoder);
2202 break;
2203
2204 case INTEL_PT_TMA:
2205 intel_pt_calc_tma(decoder);
2206 break;
2207
2208 case INTEL_PT_CYC:
2209 intel_pt_calc_cyc_timestamp(decoder);
2210 break;
2211
2212 case INTEL_PT_CBR:
2213 intel_pt_calc_cbr(decoder);
2214 if (decoder->cbr != decoder->cbr_seen) {
2215 decoder->state.type = 0;
2216 return 0;
2217 }
2218 break;
2219
2220 case INTEL_PT_MODE_EXEC:
2221 decoder->exec_mode = decoder->packet.payload;
2222 break;
2223
2224 case INTEL_PT_MODE_TSX:
2225 /* MODE_TSX need not be followed by FUP */
2226 if (!decoder->pge || decoder->in_psb) {
2227 intel_pt_update_in_tx(decoder);
2228 break;
2229 }
2230 err = intel_pt_mode_tsx(decoder, &no_tip);
2231 if (err)
2232 return err;
2233 goto next;
2234
2235 case INTEL_PT_BAD: /* Does not happen */
2236 return intel_pt_bug(decoder);
2237
2238 case INTEL_PT_PSBEND:
2239 case INTEL_PT_VMCS:
2240 case INTEL_PT_MNT:
2241 case INTEL_PT_PAD:
2242 break;
2243
2244 case INTEL_PT_PTWRITE_IP:
2245 decoder->fup_ptw_payload = decoder->packet.payload;
2246 err = intel_pt_get_next_packet(decoder);
2247 if (err)
2248 return err;
2249 if (decoder->packet.type == INTEL_PT_FUP) {
2250 decoder->set_fup_ptw = true;
2251 no_tip = true;
2252 } else {
2253 intel_pt_log_at("ERROR: Missing FUP after PTWRITE",
2254 decoder->pos);
2255 }
2256 goto next;
2257
2258 case INTEL_PT_PTWRITE:
2259 decoder->state.type = INTEL_PT_PTW;
2260 decoder->state.from_ip = decoder->ip;
2261 decoder->state.to_ip = 0;
2262 decoder->state.ptw_payload = decoder->packet.payload;
2263 return 0;
2264
2265 case INTEL_PT_MWAIT:
2266 decoder->fup_mwait_payload = decoder->packet.payload;
2267 decoder->set_fup_mwait = true;
2268 break;
2269
2270 case INTEL_PT_PWRE:
2271 if (decoder->set_fup_mwait) {
2272 decoder->fup_pwre_payload =
2273 decoder->packet.payload;
2274 decoder->set_fup_pwre = true;
2275 break;
2276 }
2277 decoder->state.type = INTEL_PT_PWR_ENTRY;
2278 decoder->state.from_ip = decoder->ip;
2279 decoder->state.to_ip = 0;
2280 decoder->state.pwrx_payload = decoder->packet.payload;
2281 return 0;
2282
2283 case INTEL_PT_EXSTOP_IP:
2284 err = intel_pt_get_next_packet(decoder);
2285 if (err)
2286 return err;
2287 if (decoder->packet.type == INTEL_PT_FUP) {
2288 decoder->set_fup_exstop = true;
2289 no_tip = true;
2290 } else {
2291 intel_pt_log_at("ERROR: Missing FUP after EXSTOP",
2292 decoder->pos);
2293 }
2294 goto next;
2295
2296 case INTEL_PT_EXSTOP:
2297 decoder->state.type = INTEL_PT_EX_STOP;
2298 decoder->state.from_ip = decoder->ip;
2299 decoder->state.to_ip = 0;
2300 return 0;
2301
2302 case INTEL_PT_PWRX:
2303 decoder->state.type = INTEL_PT_PWR_EXIT;
2304 decoder->state.from_ip = decoder->ip;
2305 decoder->state.to_ip = 0;
2306 decoder->state.pwrx_payload = decoder->packet.payload;
2307 return 0;
2308
2309 case INTEL_PT_BBP:
2310 intel_pt_bbp(decoder);
2311 break;
2312
2313 case INTEL_PT_BIP:
2314 intel_pt_bip(decoder);
2315 break;
2316
2317 case INTEL_PT_BEP:
2318 decoder->state.type = INTEL_PT_BLK_ITEMS;
2319 decoder->state.from_ip = decoder->ip;
2320 decoder->state.to_ip = 0;
2321 return 0;
2322
2323 case INTEL_PT_BEP_IP:
2324 err = intel_pt_get_next_packet(decoder);
2325 if (err)
2326 return err;
2327 if (decoder->packet.type == INTEL_PT_FUP) {
2328 decoder->set_fup_bep = true;
2329 no_tip = true;
2330 } else {
2331 intel_pt_log_at("ERROR: Missing FUP after BEP",
2332 decoder->pos);
2333 }
2334 goto next;
2335
2336 default:
2337 return intel_pt_bug(decoder);
2338 }
2339 }
2340 }
2341
intel_pt_have_ip(struct intel_pt_decoder * decoder)2342 static inline bool intel_pt_have_ip(struct intel_pt_decoder *decoder)
2343 {
2344 return decoder->packet.count &&
2345 (decoder->have_last_ip || decoder->packet.count == 3 ||
2346 decoder->packet.count == 6);
2347 }
2348
2349 /* Walk PSB+ packets to get in sync. */
intel_pt_walk_psb(struct intel_pt_decoder * decoder)2350 static int intel_pt_walk_psb(struct intel_pt_decoder *decoder)
2351 {
2352 int err;
2353
2354 decoder->in_psb = true;
2355
2356 while (1) {
2357 err = intel_pt_get_next_packet(decoder);
2358 if (err)
2359 goto out;
2360
2361 switch (decoder->packet.type) {
2362 case INTEL_PT_TIP_PGD:
2363 decoder->continuous_period = false;
2364 __fallthrough;
2365 case INTEL_PT_TIP_PGE:
2366 case INTEL_PT_TIP:
2367 case INTEL_PT_PTWRITE:
2368 case INTEL_PT_PTWRITE_IP:
2369 case INTEL_PT_EXSTOP:
2370 case INTEL_PT_EXSTOP_IP:
2371 case INTEL_PT_MWAIT:
2372 case INTEL_PT_PWRE:
2373 case INTEL_PT_PWRX:
2374 case INTEL_PT_BBP:
2375 case INTEL_PT_BIP:
2376 case INTEL_PT_BEP:
2377 case INTEL_PT_BEP_IP:
2378 intel_pt_log("ERROR: Unexpected packet\n");
2379 err = -ENOENT;
2380 goto out;
2381
2382 case INTEL_PT_FUP:
2383 decoder->pge = true;
2384 if (intel_pt_have_ip(decoder)) {
2385 uint64_t current_ip = decoder->ip;
2386
2387 intel_pt_set_ip(decoder);
2388 if (current_ip)
2389 intel_pt_log_to("Setting IP",
2390 decoder->ip);
2391 }
2392 break;
2393
2394 case INTEL_PT_MTC:
2395 intel_pt_calc_mtc_timestamp(decoder);
2396 break;
2397
2398 case INTEL_PT_TSC:
2399 intel_pt_calc_tsc_timestamp(decoder);
2400 break;
2401
2402 case INTEL_PT_TMA:
2403 intel_pt_calc_tma(decoder);
2404 break;
2405
2406 case INTEL_PT_CYC:
2407 intel_pt_calc_cyc_timestamp(decoder);
2408 break;
2409
2410 case INTEL_PT_CBR:
2411 intel_pt_calc_cbr(decoder);
2412 break;
2413
2414 case INTEL_PT_PIP:
2415 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
2416 break;
2417
2418 case INTEL_PT_MODE_EXEC:
2419 decoder->exec_mode = decoder->packet.payload;
2420 break;
2421
2422 case INTEL_PT_MODE_TSX:
2423 intel_pt_update_in_tx(decoder);
2424 break;
2425
2426 case INTEL_PT_TRACESTOP:
2427 decoder->pge = false;
2428 decoder->continuous_period = false;
2429 intel_pt_clear_tx_flags(decoder);
2430 __fallthrough;
2431
2432 case INTEL_PT_TNT:
2433 decoder->have_tma = false;
2434 intel_pt_log("ERROR: Unexpected packet\n");
2435 if (decoder->ip)
2436 decoder->pkt_state = INTEL_PT_STATE_ERR4;
2437 else
2438 decoder->pkt_state = INTEL_PT_STATE_ERR3;
2439 err = -ENOENT;
2440 goto out;
2441
2442 case INTEL_PT_BAD: /* Does not happen */
2443 err = intel_pt_bug(decoder);
2444 goto out;
2445
2446 case INTEL_PT_OVF:
2447 err = intel_pt_overflow(decoder);
2448 goto out;
2449
2450 case INTEL_PT_PSBEND:
2451 err = 0;
2452 goto out;
2453
2454 case INTEL_PT_PSB:
2455 case INTEL_PT_VMCS:
2456 case INTEL_PT_MNT:
2457 case INTEL_PT_PAD:
2458 default:
2459 break;
2460 }
2461 }
2462 out:
2463 decoder->in_psb = false;
2464
2465 return err;
2466 }
2467
intel_pt_walk_to_ip(struct intel_pt_decoder * decoder)2468 static int intel_pt_walk_to_ip(struct intel_pt_decoder *decoder)
2469 {
2470 int err;
2471
2472 while (1) {
2473 err = intel_pt_get_next_packet(decoder);
2474 if (err)
2475 return err;
2476
2477 switch (decoder->packet.type) {
2478 case INTEL_PT_TIP_PGD:
2479 decoder->continuous_period = false;
2480 decoder->pge = false;
2481 if (intel_pt_have_ip(decoder))
2482 intel_pt_set_ip(decoder);
2483 if (!decoder->ip)
2484 break;
2485 decoder->state.type |= INTEL_PT_TRACE_END;
2486 return 0;
2487
2488 case INTEL_PT_TIP_PGE:
2489 decoder->pge = true;
2490 intel_pt_mtc_cyc_cnt_pge(decoder);
2491 if (intel_pt_have_ip(decoder))
2492 intel_pt_set_ip(decoder);
2493 if (!decoder->ip)
2494 break;
2495 decoder->state.type |= INTEL_PT_TRACE_BEGIN;
2496 return 0;
2497
2498 case INTEL_PT_TIP:
2499 decoder->pge = true;
2500 if (intel_pt_have_ip(decoder))
2501 intel_pt_set_ip(decoder);
2502 if (!decoder->ip)
2503 break;
2504 return 0;
2505
2506 case INTEL_PT_FUP:
2507 if (intel_pt_have_ip(decoder))
2508 intel_pt_set_ip(decoder);
2509 if (decoder->ip)
2510 return 0;
2511 break;
2512
2513 case INTEL_PT_MTC:
2514 intel_pt_calc_mtc_timestamp(decoder);
2515 break;
2516
2517 case INTEL_PT_TSC:
2518 intel_pt_calc_tsc_timestamp(decoder);
2519 break;
2520
2521 case INTEL_PT_TMA:
2522 intel_pt_calc_tma(decoder);
2523 break;
2524
2525 case INTEL_PT_CYC:
2526 intel_pt_calc_cyc_timestamp(decoder);
2527 break;
2528
2529 case INTEL_PT_CBR:
2530 intel_pt_calc_cbr(decoder);
2531 break;
2532
2533 case INTEL_PT_PIP:
2534 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
2535 break;
2536
2537 case INTEL_PT_MODE_EXEC:
2538 decoder->exec_mode = decoder->packet.payload;
2539 break;
2540
2541 case INTEL_PT_MODE_TSX:
2542 intel_pt_update_in_tx(decoder);
2543 break;
2544
2545 case INTEL_PT_OVF:
2546 return intel_pt_overflow(decoder);
2547
2548 case INTEL_PT_BAD: /* Does not happen */
2549 return intel_pt_bug(decoder);
2550
2551 case INTEL_PT_TRACESTOP:
2552 decoder->pge = false;
2553 decoder->continuous_period = false;
2554 intel_pt_clear_tx_flags(decoder);
2555 decoder->have_tma = false;
2556 break;
2557
2558 case INTEL_PT_PSB:
2559 decoder->last_ip = 0;
2560 decoder->have_last_ip = true;
2561 intel_pt_clear_stack(&decoder->stack);
2562 err = intel_pt_walk_psb(decoder);
2563 if (err)
2564 return err;
2565 if (decoder->ip) {
2566 /* Do not have a sample */
2567 decoder->state.type = 0;
2568 return 0;
2569 }
2570 break;
2571
2572 case INTEL_PT_TNT:
2573 case INTEL_PT_PSBEND:
2574 case INTEL_PT_VMCS:
2575 case INTEL_PT_MNT:
2576 case INTEL_PT_PAD:
2577 case INTEL_PT_PTWRITE:
2578 case INTEL_PT_PTWRITE_IP:
2579 case INTEL_PT_EXSTOP:
2580 case INTEL_PT_EXSTOP_IP:
2581 case INTEL_PT_MWAIT:
2582 case INTEL_PT_PWRE:
2583 case INTEL_PT_PWRX:
2584 case INTEL_PT_BBP:
2585 case INTEL_PT_BIP:
2586 case INTEL_PT_BEP:
2587 case INTEL_PT_BEP_IP:
2588 default:
2589 break;
2590 }
2591 }
2592 }
2593
intel_pt_sync_ip(struct intel_pt_decoder * decoder)2594 static int intel_pt_sync_ip(struct intel_pt_decoder *decoder)
2595 {
2596 int err;
2597
2598 decoder->set_fup_tx_flags = false;
2599 decoder->set_fup_ptw = false;
2600 decoder->set_fup_mwait = false;
2601 decoder->set_fup_pwre = false;
2602 decoder->set_fup_exstop = false;
2603 decoder->set_fup_bep = false;
2604
2605 if (!decoder->branch_enable) {
2606 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2607 decoder->overflow = false;
2608 decoder->state.type = 0; /* Do not have a sample */
2609 return 0;
2610 }
2611
2612 intel_pt_log("Scanning for full IP\n");
2613 err = intel_pt_walk_to_ip(decoder);
2614 if (err)
2615 return err;
2616
2617 /* In hop mode, resample to get the to_ip as an "instruction" sample */
2618 if (decoder->hop)
2619 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
2620 else
2621 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2622 decoder->overflow = false;
2623
2624 decoder->state.from_ip = 0;
2625 decoder->state.to_ip = decoder->ip;
2626 intel_pt_log_to("Setting IP", decoder->ip);
2627
2628 return 0;
2629 }
2630
intel_pt_part_psb(struct intel_pt_decoder * decoder)2631 static int intel_pt_part_psb(struct intel_pt_decoder *decoder)
2632 {
2633 const unsigned char *end = decoder->buf + decoder->len;
2634 size_t i;
2635
2636 for (i = INTEL_PT_PSB_LEN - 1; i; i--) {
2637 if (i > decoder->len)
2638 continue;
2639 if (!memcmp(end - i, INTEL_PT_PSB_STR, i))
2640 return i;
2641 }
2642 return 0;
2643 }
2644
intel_pt_rest_psb(struct intel_pt_decoder * decoder,int part_psb)2645 static int intel_pt_rest_psb(struct intel_pt_decoder *decoder, int part_psb)
2646 {
2647 size_t rest_psb = INTEL_PT_PSB_LEN - part_psb;
2648 const char *psb = INTEL_PT_PSB_STR;
2649
2650 if (rest_psb > decoder->len ||
2651 memcmp(decoder->buf, psb + part_psb, rest_psb))
2652 return 0;
2653
2654 return rest_psb;
2655 }
2656
intel_pt_get_split_psb(struct intel_pt_decoder * decoder,int part_psb)2657 static int intel_pt_get_split_psb(struct intel_pt_decoder *decoder,
2658 int part_psb)
2659 {
2660 int rest_psb, ret;
2661
2662 decoder->pos += decoder->len;
2663 decoder->len = 0;
2664
2665 ret = intel_pt_get_next_data(decoder, false);
2666 if (ret)
2667 return ret;
2668
2669 rest_psb = intel_pt_rest_psb(decoder, part_psb);
2670 if (!rest_psb)
2671 return 0;
2672
2673 decoder->pos -= part_psb;
2674 decoder->next_buf = decoder->buf + rest_psb;
2675 decoder->next_len = decoder->len - rest_psb;
2676 memcpy(decoder->temp_buf, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2677 decoder->buf = decoder->temp_buf;
2678 decoder->len = INTEL_PT_PSB_LEN;
2679
2680 return 0;
2681 }
2682
intel_pt_scan_for_psb(struct intel_pt_decoder * decoder)2683 static int intel_pt_scan_for_psb(struct intel_pt_decoder *decoder)
2684 {
2685 unsigned char *next;
2686 int ret;
2687
2688 intel_pt_log("Scanning for PSB\n");
2689 while (1) {
2690 if (!decoder->len) {
2691 ret = intel_pt_get_next_data(decoder, false);
2692 if (ret)
2693 return ret;
2694 }
2695
2696 next = memmem(decoder->buf, decoder->len, INTEL_PT_PSB_STR,
2697 INTEL_PT_PSB_LEN);
2698 if (!next) {
2699 int part_psb;
2700
2701 part_psb = intel_pt_part_psb(decoder);
2702 if (part_psb) {
2703 ret = intel_pt_get_split_psb(decoder, part_psb);
2704 if (ret)
2705 return ret;
2706 } else {
2707 decoder->pos += decoder->len;
2708 decoder->len = 0;
2709 }
2710 continue;
2711 }
2712
2713 decoder->pkt_step = next - decoder->buf;
2714 return intel_pt_get_next_packet(decoder);
2715 }
2716 }
2717
intel_pt_sync(struct intel_pt_decoder * decoder)2718 static int intel_pt_sync(struct intel_pt_decoder *decoder)
2719 {
2720 int err;
2721
2722 decoder->pge = false;
2723 decoder->continuous_period = false;
2724 decoder->have_last_ip = false;
2725 decoder->last_ip = 0;
2726 decoder->ip = 0;
2727 intel_pt_clear_stack(&decoder->stack);
2728
2729 leap:
2730 err = intel_pt_scan_for_psb(decoder);
2731 if (err)
2732 return err;
2733
2734 decoder->have_last_ip = true;
2735 decoder->pkt_state = INTEL_PT_STATE_NO_IP;
2736
2737 err = intel_pt_walk_psb(decoder);
2738 if (err)
2739 return err;
2740
2741 if (decoder->ip) {
2742 decoder->state.type = 0; /* Do not have a sample */
2743 /*
2744 * In hop mode, resample to get the PSB FUP ip as an
2745 * "instruction" sample.
2746 */
2747 if (decoder->hop)
2748 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
2749 else
2750 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2751 } else if (decoder->leap) {
2752 /*
2753 * In leap mode, only PSB+ is decoded, so keeping leaping to the
2754 * next PSB until there is an ip.
2755 */
2756 goto leap;
2757 } else {
2758 return intel_pt_sync_ip(decoder);
2759 }
2760
2761 return 0;
2762 }
2763
intel_pt_est_timestamp(struct intel_pt_decoder * decoder)2764 static uint64_t intel_pt_est_timestamp(struct intel_pt_decoder *decoder)
2765 {
2766 uint64_t est = decoder->sample_insn_cnt << 1;
2767
2768 if (!decoder->cbr || !decoder->max_non_turbo_ratio)
2769 goto out;
2770
2771 est *= decoder->max_non_turbo_ratio;
2772 est /= decoder->cbr;
2773 out:
2774 return decoder->sample_timestamp + est;
2775 }
2776
intel_pt_decode(struct intel_pt_decoder * decoder)2777 const struct intel_pt_state *intel_pt_decode(struct intel_pt_decoder *decoder)
2778 {
2779 int err;
2780
2781 do {
2782 decoder->state.type = INTEL_PT_BRANCH;
2783 decoder->state.flags = 0;
2784
2785 switch (decoder->pkt_state) {
2786 case INTEL_PT_STATE_NO_PSB:
2787 err = intel_pt_sync(decoder);
2788 break;
2789 case INTEL_PT_STATE_NO_IP:
2790 decoder->have_last_ip = false;
2791 decoder->last_ip = 0;
2792 decoder->ip = 0;
2793 __fallthrough;
2794 case INTEL_PT_STATE_ERR_RESYNC:
2795 err = intel_pt_sync_ip(decoder);
2796 break;
2797 case INTEL_PT_STATE_IN_SYNC:
2798 err = intel_pt_walk_trace(decoder);
2799 break;
2800 case INTEL_PT_STATE_TNT:
2801 case INTEL_PT_STATE_TNT_CONT:
2802 err = intel_pt_walk_tnt(decoder);
2803 if (err == -EAGAIN)
2804 err = intel_pt_walk_trace(decoder);
2805 break;
2806 case INTEL_PT_STATE_TIP:
2807 case INTEL_PT_STATE_TIP_PGD:
2808 err = intel_pt_walk_tip(decoder);
2809 break;
2810 case INTEL_PT_STATE_FUP:
2811 err = intel_pt_walk_fup(decoder);
2812 if (err == -EAGAIN)
2813 err = intel_pt_walk_fup_tip(decoder);
2814 break;
2815 case INTEL_PT_STATE_FUP_NO_TIP:
2816 err = intel_pt_walk_fup(decoder);
2817 if (err == -EAGAIN)
2818 err = intel_pt_walk_trace(decoder);
2819 break;
2820 case INTEL_PT_STATE_RESAMPLE:
2821 err = intel_pt_resample(decoder);
2822 break;
2823 default:
2824 err = intel_pt_bug(decoder);
2825 break;
2826 }
2827 } while (err == -ENOLINK);
2828
2829 if (err) {
2830 decoder->state.err = intel_pt_ext_err(err);
2831 decoder->state.from_ip = decoder->ip;
2832 intel_pt_update_sample_time(decoder);
2833 decoder->sample_tot_cyc_cnt = decoder->tot_cyc_cnt;
2834 } else {
2835 decoder->state.err = 0;
2836 if (decoder->cbr != decoder->cbr_seen) {
2837 decoder->cbr_seen = decoder->cbr;
2838 if (!decoder->state.type) {
2839 decoder->state.from_ip = decoder->ip;
2840 decoder->state.to_ip = 0;
2841 }
2842 decoder->state.type |= INTEL_PT_CBR_CHG;
2843 decoder->state.cbr_payload = decoder->cbr_payload;
2844 decoder->state.cbr = decoder->cbr;
2845 }
2846 if (intel_pt_sample_time(decoder->pkt_state)) {
2847 intel_pt_update_sample_time(decoder);
2848 if (decoder->sample_cyc) {
2849 decoder->sample_tot_cyc_cnt = decoder->tot_cyc_cnt;
2850 decoder->state.flags |= INTEL_PT_SAMPLE_IPC;
2851 decoder->sample_cyc = false;
2852 }
2853 }
2854 /*
2855 * When using only TSC/MTC to compute cycles, IPC can be
2856 * sampled as soon as the cycle count changes.
2857 */
2858 if (!decoder->have_cyc)
2859 decoder->state.flags |= INTEL_PT_SAMPLE_IPC;
2860 }
2861
2862 decoder->state.timestamp = decoder->sample_timestamp;
2863 decoder->state.est_timestamp = intel_pt_est_timestamp(decoder);
2864 decoder->state.cr3 = decoder->cr3;
2865 decoder->state.tot_insn_cnt = decoder->tot_insn_cnt;
2866 decoder->state.tot_cyc_cnt = decoder->sample_tot_cyc_cnt;
2867
2868 return &decoder->state;
2869 }
2870
2871 /**
2872 * intel_pt_next_psb - move buffer pointer to the start of the next PSB packet.
2873 * @buf: pointer to buffer pointer
2874 * @len: size of buffer
2875 *
2876 * Updates the buffer pointer to point to the start of the next PSB packet if
2877 * there is one, otherwise the buffer pointer is unchanged. If @buf is updated,
2878 * @len is adjusted accordingly.
2879 *
2880 * Return: %true if a PSB packet is found, %false otherwise.
2881 */
intel_pt_next_psb(unsigned char ** buf,size_t * len)2882 static bool intel_pt_next_psb(unsigned char **buf, size_t *len)
2883 {
2884 unsigned char *next;
2885
2886 next = memmem(*buf, *len, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2887 if (next) {
2888 *len -= next - *buf;
2889 *buf = next;
2890 return true;
2891 }
2892 return false;
2893 }
2894
2895 /**
2896 * intel_pt_step_psb - move buffer pointer to the start of the following PSB
2897 * packet.
2898 * @buf: pointer to buffer pointer
2899 * @len: size of buffer
2900 *
2901 * Updates the buffer pointer to point to the start of the following PSB packet
2902 * (skipping the PSB at @buf itself) if there is one, otherwise the buffer
2903 * pointer is unchanged. If @buf is updated, @len is adjusted accordingly.
2904 *
2905 * Return: %true if a PSB packet is found, %false otherwise.
2906 */
intel_pt_step_psb(unsigned char ** buf,size_t * len)2907 static bool intel_pt_step_psb(unsigned char **buf, size_t *len)
2908 {
2909 unsigned char *next;
2910
2911 if (!*len)
2912 return false;
2913
2914 next = memmem(*buf + 1, *len - 1, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2915 if (next) {
2916 *len -= next - *buf;
2917 *buf = next;
2918 return true;
2919 }
2920 return false;
2921 }
2922
2923 /**
2924 * intel_pt_last_psb - find the last PSB packet in a buffer.
2925 * @buf: buffer
2926 * @len: size of buffer
2927 *
2928 * This function finds the last PSB in a buffer.
2929 *
2930 * Return: A pointer to the last PSB in @buf if found, %NULL otherwise.
2931 */
intel_pt_last_psb(unsigned char * buf,size_t len)2932 static unsigned char *intel_pt_last_psb(unsigned char *buf, size_t len)
2933 {
2934 const char *n = INTEL_PT_PSB_STR;
2935 unsigned char *p;
2936 size_t k;
2937
2938 if (len < INTEL_PT_PSB_LEN)
2939 return NULL;
2940
2941 k = len - INTEL_PT_PSB_LEN + 1;
2942 while (1) {
2943 p = memrchr(buf, n[0], k);
2944 if (!p)
2945 return NULL;
2946 if (!memcmp(p + 1, n + 1, INTEL_PT_PSB_LEN - 1))
2947 return p;
2948 k = p - buf;
2949 if (!k)
2950 return NULL;
2951 }
2952 }
2953
2954 /**
2955 * intel_pt_next_tsc - find and return next TSC.
2956 * @buf: buffer
2957 * @len: size of buffer
2958 * @tsc: TSC value returned
2959 * @rem: returns remaining size when TSC is found
2960 *
2961 * Find a TSC packet in @buf and return the TSC value. This function assumes
2962 * that @buf starts at a PSB and that PSB+ will contain TSC and so stops if a
2963 * PSBEND packet is found.
2964 *
2965 * Return: %true if TSC is found, false otherwise.
2966 */
intel_pt_next_tsc(unsigned char * buf,size_t len,uint64_t * tsc,size_t * rem)2967 static bool intel_pt_next_tsc(unsigned char *buf, size_t len, uint64_t *tsc,
2968 size_t *rem)
2969 {
2970 enum intel_pt_pkt_ctx ctx = INTEL_PT_NO_CTX;
2971 struct intel_pt_pkt packet;
2972 int ret;
2973
2974 while (len) {
2975 ret = intel_pt_get_packet(buf, len, &packet, &ctx);
2976 if (ret <= 0)
2977 return false;
2978 if (packet.type == INTEL_PT_TSC) {
2979 *tsc = packet.payload;
2980 *rem = len;
2981 return true;
2982 }
2983 if (packet.type == INTEL_PT_PSBEND)
2984 return false;
2985 buf += ret;
2986 len -= ret;
2987 }
2988 return false;
2989 }
2990
2991 /**
2992 * intel_pt_tsc_cmp - compare 7-byte TSCs.
2993 * @tsc1: first TSC to compare
2994 * @tsc2: second TSC to compare
2995 *
2996 * This function compares 7-byte TSC values allowing for the possibility that
2997 * TSC wrapped around. Generally it is not possible to know if TSC has wrapped
2998 * around so for that purpose this function assumes the absolute difference is
2999 * less than half the maximum difference.
3000 *
3001 * Return: %-1 if @tsc1 is before @tsc2, %0 if @tsc1 == @tsc2, %1 if @tsc1 is
3002 * after @tsc2.
3003 */
intel_pt_tsc_cmp(uint64_t tsc1,uint64_t tsc2)3004 static int intel_pt_tsc_cmp(uint64_t tsc1, uint64_t tsc2)
3005 {
3006 const uint64_t halfway = (1ULL << 55);
3007
3008 if (tsc1 == tsc2)
3009 return 0;
3010
3011 if (tsc1 < tsc2) {
3012 if (tsc2 - tsc1 < halfway)
3013 return -1;
3014 else
3015 return 1;
3016 } else {
3017 if (tsc1 - tsc2 < halfway)
3018 return 1;
3019 else
3020 return -1;
3021 }
3022 }
3023
3024 #define MAX_PADDING (PERF_AUXTRACE_RECORD_ALIGNMENT - 1)
3025
3026 /**
3027 * adj_for_padding - adjust overlap to account for padding.
3028 * @buf_b: second buffer
3029 * @buf_a: first buffer
3030 * @len_a: size of first buffer
3031 *
3032 * @buf_a might have up to 7 bytes of padding appended. Adjust the overlap
3033 * accordingly.
3034 *
3035 * Return: A pointer into @buf_b from where non-overlapped data starts
3036 */
adj_for_padding(unsigned char * buf_b,unsigned char * buf_a,size_t len_a)3037 static unsigned char *adj_for_padding(unsigned char *buf_b,
3038 unsigned char *buf_a, size_t len_a)
3039 {
3040 unsigned char *p = buf_b - MAX_PADDING;
3041 unsigned char *q = buf_a + len_a - MAX_PADDING;
3042 int i;
3043
3044 for (i = MAX_PADDING; i; i--, p++, q++) {
3045 if (*p != *q)
3046 break;
3047 }
3048
3049 return p;
3050 }
3051
3052 /**
3053 * intel_pt_find_overlap_tsc - determine start of non-overlapped trace data
3054 * using TSC.
3055 * @buf_a: first buffer
3056 * @len_a: size of first buffer
3057 * @buf_b: second buffer
3058 * @len_b: size of second buffer
3059 * @consecutive: returns true if there is data in buf_b that is consecutive
3060 * to buf_a
3061 *
3062 * If the trace contains TSC we can look at the last TSC of @buf_a and the
3063 * first TSC of @buf_b in order to determine if the buffers overlap, and then
3064 * walk forward in @buf_b until a later TSC is found. A precondition is that
3065 * @buf_a and @buf_b are positioned at a PSB.
3066 *
3067 * Return: A pointer into @buf_b from where non-overlapped data starts, or
3068 * @buf_b + @len_b if there is no non-overlapped data.
3069 */
intel_pt_find_overlap_tsc(unsigned char * buf_a,size_t len_a,unsigned char * buf_b,size_t len_b,bool * consecutive)3070 static unsigned char *intel_pt_find_overlap_tsc(unsigned char *buf_a,
3071 size_t len_a,
3072 unsigned char *buf_b,
3073 size_t len_b, bool *consecutive)
3074 {
3075 uint64_t tsc_a, tsc_b;
3076 unsigned char *p;
3077 size_t len, rem_a, rem_b;
3078
3079 p = intel_pt_last_psb(buf_a, len_a);
3080 if (!p)
3081 return buf_b; /* No PSB in buf_a => no overlap */
3082
3083 len = len_a - (p - buf_a);
3084 if (!intel_pt_next_tsc(p, len, &tsc_a, &rem_a)) {
3085 /* The last PSB+ in buf_a is incomplete, so go back one more */
3086 len_a -= len;
3087 p = intel_pt_last_psb(buf_a, len_a);
3088 if (!p)
3089 return buf_b; /* No full PSB+ => assume no overlap */
3090 len = len_a - (p - buf_a);
3091 if (!intel_pt_next_tsc(p, len, &tsc_a, &rem_a))
3092 return buf_b; /* No TSC in buf_a => assume no overlap */
3093 }
3094
3095 while (1) {
3096 /* Ignore PSB+ with no TSC */
3097 if (intel_pt_next_tsc(buf_b, len_b, &tsc_b, &rem_b)) {
3098 int cmp = intel_pt_tsc_cmp(tsc_a, tsc_b);
3099
3100 /* Same TSC, so buffers are consecutive */
3101 if (!cmp && rem_b >= rem_a) {
3102 unsigned char *start;
3103
3104 *consecutive = true;
3105 start = buf_b + len_b - (rem_b - rem_a);
3106 return adj_for_padding(start, buf_a, len_a);
3107 }
3108 if (cmp < 0)
3109 return buf_b; /* tsc_a < tsc_b => no overlap */
3110 }
3111
3112 if (!intel_pt_step_psb(&buf_b, &len_b))
3113 return buf_b + len_b; /* No PSB in buf_b => no data */
3114 }
3115 }
3116
3117 /**
3118 * intel_pt_find_overlap - determine start of non-overlapped trace data.
3119 * @buf_a: first buffer
3120 * @len_a: size of first buffer
3121 * @buf_b: second buffer
3122 * @len_b: size of second buffer
3123 * @have_tsc: can use TSC packets to detect overlap
3124 * @consecutive: returns true if there is data in buf_b that is consecutive
3125 * to buf_a
3126 *
3127 * When trace samples or snapshots are recorded there is the possibility that
3128 * the data overlaps. Note that, for the purposes of decoding, data is only
3129 * useful if it begins with a PSB packet.
3130 *
3131 * Return: A pointer into @buf_b from where non-overlapped data starts, or
3132 * @buf_b + @len_b if there is no non-overlapped data.
3133 */
intel_pt_find_overlap(unsigned char * buf_a,size_t len_a,unsigned char * buf_b,size_t len_b,bool have_tsc,bool * consecutive)3134 unsigned char *intel_pt_find_overlap(unsigned char *buf_a, size_t len_a,
3135 unsigned char *buf_b, size_t len_b,
3136 bool have_tsc, bool *consecutive)
3137 {
3138 unsigned char *found;
3139
3140 /* Buffer 'b' must start at PSB so throw away everything before that */
3141 if (!intel_pt_next_psb(&buf_b, &len_b))
3142 return buf_b + len_b; /* No PSB */
3143
3144 if (!intel_pt_next_psb(&buf_a, &len_a))
3145 return buf_b; /* No overlap */
3146
3147 if (have_tsc) {
3148 found = intel_pt_find_overlap_tsc(buf_a, len_a, buf_b, len_b,
3149 consecutive);
3150 if (found)
3151 return found;
3152 }
3153
3154 /*
3155 * Buffer 'b' cannot end within buffer 'a' so, for comparison purposes,
3156 * we can ignore the first part of buffer 'a'.
3157 */
3158 while (len_b < len_a) {
3159 if (!intel_pt_step_psb(&buf_a, &len_a))
3160 return buf_b; /* No overlap */
3161 }
3162
3163 /* Now len_b >= len_a */
3164 while (1) {
3165 /* Potential overlap so check the bytes */
3166 found = memmem(buf_a, len_a, buf_b, len_a);
3167 if (found) {
3168 *consecutive = true;
3169 return adj_for_padding(buf_b + len_a, buf_a, len_a);
3170 }
3171
3172 /* Try again at next PSB in buffer 'a' */
3173 if (!intel_pt_step_psb(&buf_a, &len_a))
3174 return buf_b; /* No overlap */
3175 }
3176 }
3177
3178 /**
3179 * struct fast_forward_data - data used by intel_pt_ff_cb().
3180 * @timestamp: timestamp to fast forward towards
3181 * @buf_timestamp: buffer timestamp of last buffer with trace data earlier than
3182 * the fast forward timestamp.
3183 */
3184 struct fast_forward_data {
3185 uint64_t timestamp;
3186 uint64_t buf_timestamp;
3187 };
3188
3189 /**
3190 * intel_pt_ff_cb - fast forward lookahead callback.
3191 * @buffer: Intel PT trace buffer
3192 * @data: opaque pointer to fast forward data (struct fast_forward_data)
3193 *
3194 * Determine if @buffer trace is past the fast forward timestamp.
3195 *
3196 * Return: 1 (stop lookahead) if @buffer trace is past the fast forward
3197 * timestamp, and 0 otherwise.
3198 */
intel_pt_ff_cb(struct intel_pt_buffer * buffer,void * data)3199 static int intel_pt_ff_cb(struct intel_pt_buffer *buffer, void *data)
3200 {
3201 struct fast_forward_data *d = data;
3202 unsigned char *buf;
3203 uint64_t tsc;
3204 size_t rem;
3205 size_t len;
3206
3207 buf = (unsigned char *)buffer->buf;
3208 len = buffer->len;
3209
3210 if (!intel_pt_next_psb(&buf, &len) ||
3211 !intel_pt_next_tsc(buf, len, &tsc, &rem))
3212 return 0;
3213
3214 tsc = intel_pt_8b_tsc(tsc, buffer->ref_timestamp);
3215
3216 intel_pt_log("Buffer 1st timestamp " x64_fmt " ref timestamp " x64_fmt "\n",
3217 tsc, buffer->ref_timestamp);
3218
3219 /*
3220 * If the buffer contains a timestamp earlier that the fast forward
3221 * timestamp, then record it, else stop.
3222 */
3223 if (tsc < d->timestamp)
3224 d->buf_timestamp = buffer->ref_timestamp;
3225 else
3226 return 1;
3227
3228 return 0;
3229 }
3230
3231 /**
3232 * intel_pt_fast_forward - reposition decoder forwards.
3233 * @decoder: Intel PT decoder
3234 * @timestamp: timestamp to fast forward towards
3235 *
3236 * Reposition decoder at the last PSB with a timestamp earlier than @timestamp.
3237 *
3238 * Return: 0 on success or negative error code on failure.
3239 */
intel_pt_fast_forward(struct intel_pt_decoder * decoder,uint64_t timestamp)3240 int intel_pt_fast_forward(struct intel_pt_decoder *decoder, uint64_t timestamp)
3241 {
3242 struct fast_forward_data d = { .timestamp = timestamp };
3243 unsigned char *buf;
3244 size_t len;
3245 int err;
3246
3247 intel_pt_log("Fast forward towards timestamp " x64_fmt "\n", timestamp);
3248
3249 /* Find buffer timestamp of buffer to fast forward to */
3250 err = decoder->lookahead(decoder->data, intel_pt_ff_cb, &d);
3251 if (err < 0)
3252 return err;
3253
3254 /* Walk to buffer with same buffer timestamp */
3255 if (d.buf_timestamp) {
3256 do {
3257 decoder->pos += decoder->len;
3258 decoder->len = 0;
3259 err = intel_pt_get_next_data(decoder, true);
3260 /* -ENOLINK means non-consecutive trace */
3261 if (err && err != -ENOLINK)
3262 return err;
3263 } while (decoder->buf_timestamp != d.buf_timestamp);
3264 }
3265
3266 if (!decoder->buf)
3267 return 0;
3268
3269 buf = (unsigned char *)decoder->buf;
3270 len = decoder->len;
3271
3272 if (!intel_pt_next_psb(&buf, &len))
3273 return 0;
3274
3275 /*
3276 * Walk PSBs while the PSB timestamp is less than the fast forward
3277 * timestamp.
3278 */
3279 do {
3280 uint64_t tsc;
3281 size_t rem;
3282
3283 if (!intel_pt_next_tsc(buf, len, &tsc, &rem))
3284 break;
3285 tsc = intel_pt_8b_tsc(tsc, decoder->buf_timestamp);
3286 /*
3287 * A TSC packet can slip past MTC packets but, after fast
3288 * forward, decoding starts at the TSC timestamp. That means
3289 * the timestamps may not be exactly the same as the timestamps
3290 * that would have been decoded without fast forward.
3291 */
3292 if (tsc < timestamp) {
3293 intel_pt_log("Fast forward to next PSB timestamp " x64_fmt "\n", tsc);
3294 decoder->pos += decoder->len - len;
3295 decoder->buf = buf;
3296 decoder->len = len;
3297 intel_pt_reposition(decoder);
3298 } else {
3299 break;
3300 }
3301 } while (intel_pt_step_psb(&buf, &len));
3302
3303 return 0;
3304 }
3305