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 SEVEN_BYTES 0xffffffffffffffULL
39
40 #define NO_VMCS 0xffffffffffULL
41
42 #define INTEL_PT_RETURN 1
43
44 /*
45 * Default maximum number of loops with no packets consumed i.e. stuck in a
46 * loop.
47 */
48 #define INTEL_PT_MAX_LOOPS 100000
49
50 struct intel_pt_blk {
51 struct intel_pt_blk *prev;
52 uint64_t ip[INTEL_PT_BLK_SIZE];
53 };
54
55 struct intel_pt_stack {
56 struct intel_pt_blk *blk;
57 struct intel_pt_blk *spare;
58 int pos;
59 };
60
61 enum intel_pt_p_once {
62 INTEL_PT_PRT_ONCE_UNK_VMCS,
63 INTEL_PT_PRT_ONCE_ERANGE,
64 };
65
66 enum intel_pt_pkt_state {
67 INTEL_PT_STATE_NO_PSB,
68 INTEL_PT_STATE_NO_IP,
69 INTEL_PT_STATE_ERR_RESYNC,
70 INTEL_PT_STATE_IN_SYNC,
71 INTEL_PT_STATE_TNT_CONT,
72 INTEL_PT_STATE_TNT,
73 INTEL_PT_STATE_TIP,
74 INTEL_PT_STATE_TIP_PGD,
75 INTEL_PT_STATE_FUP,
76 INTEL_PT_STATE_FUP_NO_TIP,
77 INTEL_PT_STATE_FUP_IN_PSB,
78 INTEL_PT_STATE_RESAMPLE,
79 INTEL_PT_STATE_VM_TIME_CORRELATION,
80 };
81
intel_pt_sample_time(enum intel_pt_pkt_state pkt_state)82 static inline bool intel_pt_sample_time(enum intel_pt_pkt_state pkt_state)
83 {
84 switch (pkt_state) {
85 case INTEL_PT_STATE_NO_PSB:
86 case INTEL_PT_STATE_NO_IP:
87 case INTEL_PT_STATE_ERR_RESYNC:
88 case INTEL_PT_STATE_IN_SYNC:
89 case INTEL_PT_STATE_TNT_CONT:
90 case INTEL_PT_STATE_RESAMPLE:
91 case INTEL_PT_STATE_VM_TIME_CORRELATION:
92 return true;
93 case INTEL_PT_STATE_TNT:
94 case INTEL_PT_STATE_TIP:
95 case INTEL_PT_STATE_TIP_PGD:
96 case INTEL_PT_STATE_FUP:
97 case INTEL_PT_STATE_FUP_NO_TIP:
98 case INTEL_PT_STATE_FUP_IN_PSB:
99 return false;
100 default:
101 return true;
102 };
103 }
104
105 #ifdef INTEL_PT_STRICT
106 #define INTEL_PT_STATE_ERR1 INTEL_PT_STATE_NO_PSB
107 #define INTEL_PT_STATE_ERR2 INTEL_PT_STATE_NO_PSB
108 #define INTEL_PT_STATE_ERR3 INTEL_PT_STATE_NO_PSB
109 #define INTEL_PT_STATE_ERR4 INTEL_PT_STATE_NO_PSB
110 #else
111 #define INTEL_PT_STATE_ERR1 (decoder->pkt_state)
112 #define INTEL_PT_STATE_ERR2 INTEL_PT_STATE_NO_IP
113 #define INTEL_PT_STATE_ERR3 INTEL_PT_STATE_ERR_RESYNC
114 #define INTEL_PT_STATE_ERR4 INTEL_PT_STATE_IN_SYNC
115 #endif
116
117 struct intel_pt_decoder {
118 int (*get_trace)(struct intel_pt_buffer *buffer, void *data);
119 int (*walk_insn)(struct intel_pt_insn *intel_pt_insn,
120 uint64_t *insn_cnt_ptr, uint64_t *ip, uint64_t to_ip,
121 uint64_t max_insn_cnt, void *data);
122 bool (*pgd_ip)(uint64_t ip, void *data);
123 int (*lookahead)(void *data, intel_pt_lookahead_cb_t cb, void *cb_data);
124 struct intel_pt_vmcs_info *(*findnew_vmcs_info)(void *data, uint64_t vmcs);
125 void *data;
126 struct intel_pt_state state;
127 const unsigned char *buf;
128 size_t len;
129 bool return_compression;
130 bool branch_enable;
131 bool mtc_insn;
132 bool pge;
133 bool have_tma;
134 bool have_cyc;
135 bool fixup_last_mtc;
136 bool have_last_ip;
137 bool in_psb;
138 bool hop;
139 bool leap;
140 bool emulated_ptwrite;
141 bool vm_time_correlation;
142 bool vm_tm_corr_dry_run;
143 bool vm_tm_corr_reliable;
144 bool vm_tm_corr_same_buf;
145 bool vm_tm_corr_continuous;
146 bool nr;
147 bool next_nr;
148 bool iflag;
149 bool next_iflag;
150 enum intel_pt_param_flags flags;
151 uint64_t pos;
152 uint64_t last_ip;
153 uint64_t ip;
154 uint64_t pip_payload;
155 uint64_t timestamp;
156 uint64_t tsc_timestamp;
157 uint64_t ref_timestamp;
158 uint64_t buf_timestamp;
159 uint64_t sample_timestamp;
160 uint64_t ret_addr;
161 uint64_t ctc_timestamp;
162 uint64_t ctc_delta;
163 uint64_t cycle_cnt;
164 uint64_t cyc_ref_timestamp;
165 uint64_t first_timestamp;
166 uint64_t last_reliable_timestamp;
167 uint64_t vmcs;
168 uint64_t print_once;
169 uint64_t last_ctc;
170 uint32_t last_mtc;
171 uint32_t tsc_ctc_ratio_n;
172 uint32_t tsc_ctc_ratio_d;
173 uint32_t tsc_ctc_mult;
174 uint32_t tsc_slip;
175 uint32_t ctc_rem_mask;
176 int mtc_shift;
177 struct intel_pt_stack stack;
178 enum intel_pt_pkt_state pkt_state;
179 enum intel_pt_pkt_ctx pkt_ctx;
180 enum intel_pt_pkt_ctx prev_pkt_ctx;
181 enum intel_pt_blk_type blk_type;
182 int blk_type_pos;
183 struct intel_pt_pkt packet;
184 struct intel_pt_pkt tnt;
185 int pkt_step;
186 int pkt_len;
187 int last_packet_type;
188 unsigned int cbr;
189 unsigned int cbr_seen;
190 unsigned int max_non_turbo_ratio;
191 double max_non_turbo_ratio_fp;
192 double cbr_cyc_to_tsc;
193 double calc_cyc_to_tsc;
194 bool have_calc_cyc_to_tsc;
195 int exec_mode;
196 unsigned int insn_bytes;
197 uint64_t period;
198 enum intel_pt_period_type period_type;
199 uint64_t tot_insn_cnt;
200 uint64_t period_insn_cnt;
201 uint64_t period_mask;
202 uint64_t period_ticks;
203 uint64_t last_masked_timestamp;
204 uint64_t tot_cyc_cnt;
205 uint64_t sample_tot_cyc_cnt;
206 uint64_t base_cyc_cnt;
207 uint64_t cyc_cnt_timestamp;
208 uint64_t ctl;
209 uint64_t cyc_threshold;
210 double tsc_to_cyc;
211 bool continuous_period;
212 bool overflow;
213 bool set_fup_tx_flags;
214 bool set_fup_ptw;
215 bool set_fup_mwait;
216 bool set_fup_pwre;
217 bool set_fup_exstop;
218 bool set_fup_bep;
219 bool set_fup_cfe_ip;
220 bool set_fup_cfe;
221 bool set_fup_mode_exec;
222 bool sample_cyc;
223 unsigned int fup_tx_flags;
224 unsigned int tx_flags;
225 uint64_t fup_ptw_payload;
226 uint64_t fup_mwait_payload;
227 uint64_t fup_pwre_payload;
228 uint64_t cbr_payload;
229 uint64_t timestamp_insn_cnt;
230 uint64_t sample_insn_cnt;
231 uint64_t stuck_ip;
232 struct intel_pt_pkt fup_cfe_pkt;
233 int max_loops;
234 int no_progress;
235 int stuck_ip_prd;
236 int stuck_ip_cnt;
237 uint64_t psb_ip;
238 const unsigned char *next_buf;
239 size_t next_len;
240 unsigned char temp_buf[INTEL_PT_PKT_MAX_SZ];
241 int evd_cnt;
242 struct intel_pt_evd evd[INTEL_PT_MAX_EVDS];
243 };
244
intel_pt_lower_power_of_2(uint64_t x)245 static uint64_t intel_pt_lower_power_of_2(uint64_t x)
246 {
247 int i;
248
249 for (i = 0; x != 1; i++)
250 x >>= 1;
251
252 return x << i;
253 }
254
255 __printf(1, 2)
p_log(const char * fmt,...)256 static void p_log(const char *fmt, ...)
257 {
258 char buf[512];
259 va_list args;
260
261 va_start(args, fmt);
262 vsnprintf(buf, sizeof(buf), fmt, args);
263 va_end(args);
264
265 fprintf(stderr, "%s\n", buf);
266 intel_pt_log("%s\n", buf);
267 }
268
intel_pt_print_once(struct intel_pt_decoder * decoder,enum intel_pt_p_once id)269 static bool intel_pt_print_once(struct intel_pt_decoder *decoder,
270 enum intel_pt_p_once id)
271 {
272 uint64_t bit = 1ULL << id;
273
274 if (decoder->print_once & bit)
275 return false;
276 decoder->print_once |= bit;
277 return true;
278 }
279
intel_pt_cyc_threshold(uint64_t ctl)280 static uint64_t intel_pt_cyc_threshold(uint64_t ctl)
281 {
282 if (!(ctl & INTEL_PT_CYC_ENABLE))
283 return 0;
284
285 return (ctl & INTEL_PT_CYC_THRESHOLD) >> INTEL_PT_CYC_THRESHOLD_SHIFT;
286 }
287
intel_pt_setup_period(struct intel_pt_decoder * decoder)288 static void intel_pt_setup_period(struct intel_pt_decoder *decoder)
289 {
290 if (decoder->period_type == INTEL_PT_PERIOD_TICKS) {
291 uint64_t period;
292
293 period = intel_pt_lower_power_of_2(decoder->period);
294 decoder->period_mask = ~(period - 1);
295 decoder->period_ticks = period;
296 }
297 }
298
multdiv(uint64_t t,uint32_t n,uint32_t d)299 static uint64_t multdiv(uint64_t t, uint32_t n, uint32_t d)
300 {
301 if (!d)
302 return 0;
303 return (t / d) * n + ((t % d) * n) / d;
304 }
305
intel_pt_decoder_new(struct intel_pt_params * params)306 struct intel_pt_decoder *intel_pt_decoder_new(struct intel_pt_params *params)
307 {
308 struct intel_pt_decoder *decoder;
309
310 if (!params->get_trace || !params->walk_insn)
311 return NULL;
312
313 decoder = zalloc(sizeof(struct intel_pt_decoder));
314 if (!decoder)
315 return NULL;
316
317 decoder->get_trace = params->get_trace;
318 decoder->walk_insn = params->walk_insn;
319 decoder->pgd_ip = params->pgd_ip;
320 decoder->lookahead = params->lookahead;
321 decoder->findnew_vmcs_info = params->findnew_vmcs_info;
322 decoder->data = params->data;
323 decoder->return_compression = params->return_compression;
324 decoder->branch_enable = params->branch_enable;
325 decoder->hop = params->quick >= 1;
326 decoder->leap = params->quick >= 2;
327 decoder->vm_time_correlation = params->vm_time_correlation;
328 decoder->vm_tm_corr_dry_run = params->vm_tm_corr_dry_run;
329 decoder->first_timestamp = params->first_timestamp;
330 decoder->last_reliable_timestamp = params->first_timestamp;
331 decoder->max_loops = params->max_loops ? params->max_loops : INTEL_PT_MAX_LOOPS;
332
333 decoder->flags = params->flags;
334
335 decoder->ctl = params->ctl;
336 decoder->period = params->period;
337 decoder->period_type = params->period_type;
338
339 decoder->max_non_turbo_ratio = params->max_non_turbo_ratio;
340 decoder->max_non_turbo_ratio_fp = params->max_non_turbo_ratio;
341
342 decoder->cyc_threshold = intel_pt_cyc_threshold(decoder->ctl);
343
344 intel_pt_setup_period(decoder);
345
346 decoder->mtc_shift = params->mtc_period;
347 decoder->ctc_rem_mask = (1 << decoder->mtc_shift) - 1;
348
349 decoder->tsc_ctc_ratio_n = params->tsc_ctc_ratio_n;
350 decoder->tsc_ctc_ratio_d = params->tsc_ctc_ratio_d;
351
352 if (!decoder->tsc_ctc_ratio_n)
353 decoder->tsc_ctc_ratio_d = 0;
354
355 if (decoder->tsc_ctc_ratio_d) {
356 if (!(decoder->tsc_ctc_ratio_n % decoder->tsc_ctc_ratio_d))
357 decoder->tsc_ctc_mult = decoder->tsc_ctc_ratio_n /
358 decoder->tsc_ctc_ratio_d;
359 }
360
361 /*
362 * A TSC packet can slip past MTC packets so that the timestamp appears
363 * to go backwards. One estimate is that can be up to about 40 CPU
364 * cycles, which is certainly less than 0x1000 TSC ticks, but accept
365 * slippage an order of magnitude more to be on the safe side.
366 */
367 decoder->tsc_slip = 0x10000;
368
369 intel_pt_log("timestamp: mtc_shift %u\n", decoder->mtc_shift);
370 intel_pt_log("timestamp: tsc_ctc_ratio_n %u\n", decoder->tsc_ctc_ratio_n);
371 intel_pt_log("timestamp: tsc_ctc_ratio_d %u\n", decoder->tsc_ctc_ratio_d);
372 intel_pt_log("timestamp: tsc_ctc_mult %u\n", decoder->tsc_ctc_mult);
373 intel_pt_log("timestamp: tsc_slip %#x\n", decoder->tsc_slip);
374
375 if (decoder->hop)
376 intel_pt_log("Hop mode: decoding FUP and TIPs, but not TNT\n");
377
378 return decoder;
379 }
380
intel_pt_set_first_timestamp(struct intel_pt_decoder * decoder,uint64_t first_timestamp)381 void intel_pt_set_first_timestamp(struct intel_pt_decoder *decoder,
382 uint64_t first_timestamp)
383 {
384 decoder->first_timestamp = first_timestamp;
385 }
386
intel_pt_pop_blk(struct intel_pt_stack * stack)387 static void intel_pt_pop_blk(struct intel_pt_stack *stack)
388 {
389 struct intel_pt_blk *blk = stack->blk;
390
391 stack->blk = blk->prev;
392 if (!stack->spare)
393 stack->spare = blk;
394 else
395 free(blk);
396 }
397
intel_pt_pop(struct intel_pt_stack * stack)398 static uint64_t intel_pt_pop(struct intel_pt_stack *stack)
399 {
400 if (!stack->pos) {
401 if (!stack->blk)
402 return 0;
403 intel_pt_pop_blk(stack);
404 if (!stack->blk)
405 return 0;
406 stack->pos = INTEL_PT_BLK_SIZE;
407 }
408 return stack->blk->ip[--stack->pos];
409 }
410
intel_pt_alloc_blk(struct intel_pt_stack * stack)411 static int intel_pt_alloc_blk(struct intel_pt_stack *stack)
412 {
413 struct intel_pt_blk *blk;
414
415 if (stack->spare) {
416 blk = stack->spare;
417 stack->spare = NULL;
418 } else {
419 blk = malloc(sizeof(struct intel_pt_blk));
420 if (!blk)
421 return -ENOMEM;
422 }
423
424 blk->prev = stack->blk;
425 stack->blk = blk;
426 stack->pos = 0;
427 return 0;
428 }
429
intel_pt_push(struct intel_pt_stack * stack,uint64_t ip)430 static int intel_pt_push(struct intel_pt_stack *stack, uint64_t ip)
431 {
432 int err;
433
434 if (!stack->blk || stack->pos == INTEL_PT_BLK_SIZE) {
435 err = intel_pt_alloc_blk(stack);
436 if (err)
437 return err;
438 }
439
440 stack->blk->ip[stack->pos++] = ip;
441 return 0;
442 }
443
intel_pt_clear_stack(struct intel_pt_stack * stack)444 static void intel_pt_clear_stack(struct intel_pt_stack *stack)
445 {
446 while (stack->blk)
447 intel_pt_pop_blk(stack);
448 stack->pos = 0;
449 }
450
intel_pt_free_stack(struct intel_pt_stack * stack)451 static void intel_pt_free_stack(struct intel_pt_stack *stack)
452 {
453 intel_pt_clear_stack(stack);
454 zfree(&stack->blk);
455 zfree(&stack->spare);
456 }
457
intel_pt_decoder_free(struct intel_pt_decoder * decoder)458 void intel_pt_decoder_free(struct intel_pt_decoder *decoder)
459 {
460 intel_pt_free_stack(&decoder->stack);
461 free(decoder);
462 }
463
intel_pt_ext_err(int code)464 static int intel_pt_ext_err(int code)
465 {
466 switch (code) {
467 case -ENOMEM:
468 return INTEL_PT_ERR_NOMEM;
469 case -ENOSYS:
470 return INTEL_PT_ERR_INTERN;
471 case -EBADMSG:
472 return INTEL_PT_ERR_BADPKT;
473 case -ENODATA:
474 return INTEL_PT_ERR_NODATA;
475 case -EILSEQ:
476 return INTEL_PT_ERR_NOINSN;
477 case -ENOENT:
478 return INTEL_PT_ERR_MISMAT;
479 case -EOVERFLOW:
480 return INTEL_PT_ERR_OVR;
481 case -ENOSPC:
482 return INTEL_PT_ERR_LOST;
483 case -ELOOP:
484 return INTEL_PT_ERR_NELOOP;
485 case -ECONNRESET:
486 return INTEL_PT_ERR_EPTW;
487 default:
488 return INTEL_PT_ERR_UNK;
489 }
490 }
491
492 static const char *intel_pt_err_msgs[] = {
493 [INTEL_PT_ERR_NOMEM] = "Memory allocation failed",
494 [INTEL_PT_ERR_INTERN] = "Internal error",
495 [INTEL_PT_ERR_BADPKT] = "Bad packet",
496 [INTEL_PT_ERR_NODATA] = "No more data",
497 [INTEL_PT_ERR_NOINSN] = "Failed to get instruction",
498 [INTEL_PT_ERR_MISMAT] = "Trace doesn't match instruction",
499 [INTEL_PT_ERR_OVR] = "Overflow packet",
500 [INTEL_PT_ERR_LOST] = "Lost trace data",
501 [INTEL_PT_ERR_UNK] = "Unknown error!",
502 [INTEL_PT_ERR_NELOOP] = "Never-ending loop (refer perf config intel-pt.max-loops)",
503 [INTEL_PT_ERR_EPTW] = "Broken emulated ptwrite",
504 };
505
intel_pt__strerror(int code,char * buf,size_t buflen)506 int intel_pt__strerror(int code, char *buf, size_t buflen)
507 {
508 if (code < 1 || code >= INTEL_PT_ERR_MAX)
509 code = INTEL_PT_ERR_UNK;
510 strlcpy(buf, intel_pt_err_msgs[code], buflen);
511 return 0;
512 }
513
intel_pt_calc_ip(const struct intel_pt_pkt * packet,uint64_t last_ip)514 static uint64_t intel_pt_calc_ip(const struct intel_pt_pkt *packet,
515 uint64_t last_ip)
516 {
517 uint64_t ip;
518
519 switch (packet->count) {
520 case 1:
521 ip = (last_ip & (uint64_t)0xffffffffffff0000ULL) |
522 packet->payload;
523 break;
524 case 2:
525 ip = (last_ip & (uint64_t)0xffffffff00000000ULL) |
526 packet->payload;
527 break;
528 case 3:
529 ip = packet->payload;
530 /* Sign-extend 6-byte ip */
531 if (ip & (uint64_t)0x800000000000ULL)
532 ip |= (uint64_t)0xffff000000000000ULL;
533 break;
534 case 4:
535 ip = (last_ip & (uint64_t)0xffff000000000000ULL) |
536 packet->payload;
537 break;
538 case 6:
539 ip = packet->payload;
540 break;
541 default:
542 return 0;
543 }
544
545 return ip;
546 }
547
intel_pt_set_last_ip(struct intel_pt_decoder * decoder)548 static inline void intel_pt_set_last_ip(struct intel_pt_decoder *decoder)
549 {
550 decoder->last_ip = intel_pt_calc_ip(&decoder->packet, decoder->last_ip);
551 decoder->have_last_ip = true;
552 }
553
intel_pt_set_ip(struct intel_pt_decoder * decoder)554 static inline void intel_pt_set_ip(struct intel_pt_decoder *decoder)
555 {
556 intel_pt_set_last_ip(decoder);
557 decoder->ip = decoder->last_ip;
558 }
559
intel_pt_decoder_log_packet(struct intel_pt_decoder * decoder)560 static void intel_pt_decoder_log_packet(struct intel_pt_decoder *decoder)
561 {
562 intel_pt_log_packet(&decoder->packet, decoder->pkt_len, decoder->pos,
563 decoder->buf);
564 }
565
intel_pt_bug(struct intel_pt_decoder * decoder)566 static int intel_pt_bug(struct intel_pt_decoder *decoder)
567 {
568 intel_pt_log("ERROR: Internal error\n");
569 decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
570 return -ENOSYS;
571 }
572
intel_pt_clear_tx_flags(struct intel_pt_decoder * decoder)573 static inline void intel_pt_clear_tx_flags(struct intel_pt_decoder *decoder)
574 {
575 decoder->tx_flags = 0;
576 }
577
intel_pt_update_in_tx(struct intel_pt_decoder * decoder)578 static inline void intel_pt_update_in_tx(struct intel_pt_decoder *decoder)
579 {
580 decoder->tx_flags = decoder->packet.payload & INTEL_PT_IN_TX;
581 }
582
intel_pt_update_pip(struct intel_pt_decoder * decoder)583 static inline void intel_pt_update_pip(struct intel_pt_decoder *decoder)
584 {
585 decoder->pip_payload = decoder->packet.payload;
586 }
587
intel_pt_update_nr(struct intel_pt_decoder * decoder)588 static inline void intel_pt_update_nr(struct intel_pt_decoder *decoder)
589 {
590 decoder->next_nr = decoder->pip_payload & 1;
591 }
592
intel_pt_set_nr(struct intel_pt_decoder * decoder)593 static inline void intel_pt_set_nr(struct intel_pt_decoder *decoder)
594 {
595 decoder->nr = decoder->pip_payload & 1;
596 decoder->next_nr = decoder->nr;
597 }
598
intel_pt_set_pip(struct intel_pt_decoder * decoder)599 static inline void intel_pt_set_pip(struct intel_pt_decoder *decoder)
600 {
601 intel_pt_update_pip(decoder);
602 intel_pt_set_nr(decoder);
603 }
604
intel_pt_bad_packet(struct intel_pt_decoder * decoder)605 static int intel_pt_bad_packet(struct intel_pt_decoder *decoder)
606 {
607 intel_pt_clear_tx_flags(decoder);
608 decoder->have_tma = false;
609 decoder->pkt_len = 1;
610 decoder->pkt_step = 1;
611 intel_pt_decoder_log_packet(decoder);
612 if (decoder->pkt_state != INTEL_PT_STATE_NO_PSB) {
613 intel_pt_log("ERROR: Bad packet\n");
614 decoder->pkt_state = INTEL_PT_STATE_ERR1;
615 }
616 return -EBADMSG;
617 }
618
intel_pt_update_sample_time(struct intel_pt_decoder * decoder)619 static inline void intel_pt_update_sample_time(struct intel_pt_decoder *decoder)
620 {
621 decoder->sample_timestamp = decoder->timestamp;
622 decoder->sample_insn_cnt = decoder->timestamp_insn_cnt;
623 decoder->state.cycles = decoder->tot_cyc_cnt;
624 }
625
intel_pt_reposition(struct intel_pt_decoder * decoder)626 static void intel_pt_reposition(struct intel_pt_decoder *decoder)
627 {
628 decoder->ip = 0;
629 decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
630 decoder->timestamp = 0;
631 decoder->have_tma = false;
632 }
633
intel_pt_get_data(struct intel_pt_decoder * decoder,bool reposition)634 static int intel_pt_get_data(struct intel_pt_decoder *decoder, bool reposition)
635 {
636 struct intel_pt_buffer buffer = { .buf = 0, };
637 int ret;
638
639 decoder->pkt_step = 0;
640
641 intel_pt_log("Getting more data\n");
642 ret = decoder->get_trace(&buffer, decoder->data);
643 if (ret)
644 return ret;
645 decoder->buf = buffer.buf;
646 decoder->len = buffer.len;
647 if (!decoder->len) {
648 intel_pt_log("No more data\n");
649 return -ENODATA;
650 }
651 decoder->buf_timestamp = buffer.ref_timestamp;
652 if (!buffer.consecutive || reposition) {
653 intel_pt_reposition(decoder);
654 decoder->ref_timestamp = buffer.ref_timestamp;
655 decoder->state.trace_nr = buffer.trace_nr;
656 decoder->vm_tm_corr_same_buf = false;
657 intel_pt_log("Reference timestamp 0x%" PRIx64 "\n",
658 decoder->ref_timestamp);
659 return -ENOLINK;
660 }
661
662 return 0;
663 }
664
intel_pt_get_next_data(struct intel_pt_decoder * decoder,bool reposition)665 static int intel_pt_get_next_data(struct intel_pt_decoder *decoder,
666 bool reposition)
667 {
668 if (!decoder->next_buf)
669 return intel_pt_get_data(decoder, reposition);
670
671 decoder->buf = decoder->next_buf;
672 decoder->len = decoder->next_len;
673 decoder->next_buf = 0;
674 decoder->next_len = 0;
675 return 0;
676 }
677
intel_pt_get_split_packet(struct intel_pt_decoder * decoder)678 static int intel_pt_get_split_packet(struct intel_pt_decoder *decoder)
679 {
680 unsigned char *buf = decoder->temp_buf;
681 size_t old_len, len, n;
682 int ret;
683
684 old_len = decoder->len;
685 len = decoder->len;
686 memcpy(buf, decoder->buf, len);
687
688 ret = intel_pt_get_data(decoder, false);
689 if (ret) {
690 decoder->pos += old_len;
691 return ret < 0 ? ret : -EINVAL;
692 }
693
694 n = INTEL_PT_PKT_MAX_SZ - len;
695 if (n > decoder->len)
696 n = decoder->len;
697 memcpy(buf + len, decoder->buf, n);
698 len += n;
699
700 decoder->prev_pkt_ctx = decoder->pkt_ctx;
701 ret = intel_pt_get_packet(buf, len, &decoder->packet, &decoder->pkt_ctx);
702 if (ret < (int)old_len) {
703 decoder->next_buf = decoder->buf;
704 decoder->next_len = decoder->len;
705 decoder->buf = buf;
706 decoder->len = old_len;
707 return intel_pt_bad_packet(decoder);
708 }
709
710 decoder->next_buf = decoder->buf + (ret - old_len);
711 decoder->next_len = decoder->len - (ret - old_len);
712
713 decoder->buf = buf;
714 decoder->len = ret;
715
716 return ret;
717 }
718
719 struct intel_pt_pkt_info {
720 struct intel_pt_decoder *decoder;
721 struct intel_pt_pkt packet;
722 uint64_t pos;
723 int pkt_len;
724 int last_packet_type;
725 void *data;
726 };
727
728 typedef int (*intel_pt_pkt_cb_t)(struct intel_pt_pkt_info *pkt_info);
729
730 /* Lookahead packets in current buffer */
intel_pt_pkt_lookahead(struct intel_pt_decoder * decoder,intel_pt_pkt_cb_t cb,void * data)731 static int intel_pt_pkt_lookahead(struct intel_pt_decoder *decoder,
732 intel_pt_pkt_cb_t cb, void *data)
733 {
734 struct intel_pt_pkt_info pkt_info;
735 const unsigned char *buf = decoder->buf;
736 enum intel_pt_pkt_ctx pkt_ctx = decoder->pkt_ctx;
737 size_t len = decoder->len;
738 int ret;
739
740 pkt_info.decoder = decoder;
741 pkt_info.pos = decoder->pos;
742 pkt_info.pkt_len = decoder->pkt_step;
743 pkt_info.last_packet_type = decoder->last_packet_type;
744 pkt_info.data = data;
745
746 while (1) {
747 do {
748 pkt_info.pos += pkt_info.pkt_len;
749 buf += pkt_info.pkt_len;
750 len -= pkt_info.pkt_len;
751
752 if (!len)
753 return INTEL_PT_NEED_MORE_BYTES;
754
755 ret = intel_pt_get_packet(buf, len, &pkt_info.packet,
756 &pkt_ctx);
757 if (!ret)
758 return INTEL_PT_NEED_MORE_BYTES;
759 if (ret < 0)
760 return ret;
761
762 pkt_info.pkt_len = ret;
763 } while (pkt_info.packet.type == INTEL_PT_PAD);
764
765 ret = cb(&pkt_info);
766 if (ret)
767 return 0;
768
769 pkt_info.last_packet_type = pkt_info.packet.type;
770 }
771 }
772
773 struct intel_pt_calc_cyc_to_tsc_info {
774 uint64_t cycle_cnt;
775 unsigned int cbr;
776 uint32_t last_mtc;
777 uint64_t ctc_timestamp;
778 uint64_t ctc_delta;
779 uint64_t tsc_timestamp;
780 uint64_t timestamp;
781 bool have_tma;
782 bool fixup_last_mtc;
783 bool from_mtc;
784 double cbr_cyc_to_tsc;
785 };
786
787 /*
788 * MTC provides a 8-bit slice of CTC but the TMA packet only provides the lower
789 * 16 bits of CTC. If mtc_shift > 8 then some of the MTC bits are not in the CTC
790 * provided by the TMA packet. Fix-up the last_mtc calculated from the TMA
791 * packet by copying the missing bits from the current MTC assuming the least
792 * difference between the two, and that the current MTC comes after last_mtc.
793 */
intel_pt_fixup_last_mtc(uint32_t mtc,int mtc_shift,uint32_t * last_mtc)794 static void intel_pt_fixup_last_mtc(uint32_t mtc, int mtc_shift,
795 uint32_t *last_mtc)
796 {
797 uint32_t first_missing_bit = 1U << (16 - mtc_shift);
798 uint32_t mask = ~(first_missing_bit - 1);
799
800 *last_mtc |= mtc & mask;
801 if (*last_mtc >= mtc) {
802 *last_mtc -= first_missing_bit;
803 *last_mtc &= 0xff;
804 }
805 }
806
intel_pt_calc_cyc_cb(struct intel_pt_pkt_info * pkt_info)807 static int intel_pt_calc_cyc_cb(struct intel_pt_pkt_info *pkt_info)
808 {
809 struct intel_pt_decoder *decoder = pkt_info->decoder;
810 struct intel_pt_calc_cyc_to_tsc_info *data = pkt_info->data;
811 uint64_t timestamp;
812 double cyc_to_tsc;
813 unsigned int cbr;
814 uint32_t mtc, mtc_delta, ctc, fc, ctc_rem;
815
816 switch (pkt_info->packet.type) {
817 case INTEL_PT_TNT:
818 case INTEL_PT_TIP_PGE:
819 case INTEL_PT_TIP:
820 case INTEL_PT_FUP:
821 case INTEL_PT_PSB:
822 case INTEL_PT_PIP:
823 case INTEL_PT_MODE_EXEC:
824 case INTEL_PT_MODE_TSX:
825 case INTEL_PT_PSBEND:
826 case INTEL_PT_PAD:
827 case INTEL_PT_VMCS:
828 case INTEL_PT_MNT:
829 case INTEL_PT_PTWRITE:
830 case INTEL_PT_PTWRITE_IP:
831 case INTEL_PT_BBP:
832 case INTEL_PT_BIP:
833 case INTEL_PT_BEP:
834 case INTEL_PT_BEP_IP:
835 case INTEL_PT_CFE:
836 case INTEL_PT_CFE_IP:
837 case INTEL_PT_EVD:
838 return 0;
839
840 case INTEL_PT_MTC:
841 if (!data->have_tma)
842 return 0;
843
844 mtc = pkt_info->packet.payload;
845 if (decoder->mtc_shift > 8 && data->fixup_last_mtc) {
846 data->fixup_last_mtc = false;
847 intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
848 &data->last_mtc);
849 }
850 if (mtc > data->last_mtc)
851 mtc_delta = mtc - data->last_mtc;
852 else
853 mtc_delta = mtc + 256 - data->last_mtc;
854 data->ctc_delta += mtc_delta << decoder->mtc_shift;
855 data->last_mtc = mtc;
856
857 if (decoder->tsc_ctc_mult) {
858 timestamp = data->ctc_timestamp +
859 data->ctc_delta * decoder->tsc_ctc_mult;
860 } else {
861 timestamp = data->ctc_timestamp +
862 multdiv(data->ctc_delta,
863 decoder->tsc_ctc_ratio_n,
864 decoder->tsc_ctc_ratio_d);
865 }
866
867 if (timestamp < data->timestamp)
868 return 1;
869
870 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
871 data->timestamp = timestamp;
872 return 0;
873 }
874
875 break;
876
877 case INTEL_PT_TSC:
878 /*
879 * For now, do not support using TSC packets - refer
880 * intel_pt_calc_cyc_to_tsc().
881 */
882 if (data->from_mtc)
883 return 1;
884 timestamp = pkt_info->packet.payload |
885 (data->timestamp & (0xffULL << 56));
886 if (data->from_mtc && timestamp < data->timestamp &&
887 data->timestamp - timestamp < decoder->tsc_slip)
888 return 1;
889 if (timestamp < data->timestamp)
890 timestamp += (1ULL << 56);
891 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
892 if (data->from_mtc)
893 return 1;
894 data->tsc_timestamp = timestamp;
895 data->timestamp = timestamp;
896 return 0;
897 }
898 break;
899
900 case INTEL_PT_TMA:
901 if (data->from_mtc)
902 return 1;
903
904 if (!decoder->tsc_ctc_ratio_d)
905 return 0;
906
907 ctc = pkt_info->packet.payload;
908 fc = pkt_info->packet.count;
909 ctc_rem = ctc & decoder->ctc_rem_mask;
910
911 data->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
912
913 data->ctc_timestamp = data->tsc_timestamp - fc;
914 if (decoder->tsc_ctc_mult) {
915 data->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
916 } else {
917 data->ctc_timestamp -=
918 multdiv(ctc_rem, decoder->tsc_ctc_ratio_n,
919 decoder->tsc_ctc_ratio_d);
920 }
921
922 data->ctc_delta = 0;
923 data->have_tma = true;
924 data->fixup_last_mtc = true;
925
926 return 0;
927
928 case INTEL_PT_CYC:
929 data->cycle_cnt += pkt_info->packet.payload;
930 return 0;
931
932 case INTEL_PT_CBR:
933 cbr = pkt_info->packet.payload;
934 if (data->cbr && data->cbr != cbr)
935 return 1;
936 data->cbr = cbr;
937 data->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
938 return 0;
939
940 case INTEL_PT_TIP_PGD:
941 case INTEL_PT_TRACESTOP:
942 case INTEL_PT_EXSTOP:
943 case INTEL_PT_EXSTOP_IP:
944 case INTEL_PT_MWAIT:
945 case INTEL_PT_PWRE:
946 case INTEL_PT_PWRX:
947 case INTEL_PT_OVF:
948 case INTEL_PT_BAD: /* Does not happen */
949 default:
950 return 1;
951 }
952
953 if (!data->cbr && decoder->cbr) {
954 data->cbr = decoder->cbr;
955 data->cbr_cyc_to_tsc = decoder->cbr_cyc_to_tsc;
956 }
957
958 if (!data->cycle_cnt)
959 return 1;
960
961 cyc_to_tsc = (double)(timestamp - decoder->timestamp) / data->cycle_cnt;
962
963 if (data->cbr && cyc_to_tsc > data->cbr_cyc_to_tsc &&
964 cyc_to_tsc / data->cbr_cyc_to_tsc > 1.25) {
965 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle too big (c.f. CBR-based value %g), pos " x64_fmt "\n",
966 cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
967 return 1;
968 }
969
970 decoder->calc_cyc_to_tsc = cyc_to_tsc;
971 decoder->have_calc_cyc_to_tsc = true;
972
973 if (data->cbr) {
974 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. CBR-based value %g, pos " x64_fmt "\n",
975 cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
976 } else {
977 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. unknown CBR-based value, pos " x64_fmt "\n",
978 cyc_to_tsc, pkt_info->pos);
979 }
980
981 return 1;
982 }
983
intel_pt_calc_cyc_to_tsc(struct intel_pt_decoder * decoder,bool from_mtc)984 static void intel_pt_calc_cyc_to_tsc(struct intel_pt_decoder *decoder,
985 bool from_mtc)
986 {
987 struct intel_pt_calc_cyc_to_tsc_info data = {
988 .cycle_cnt = 0,
989 .cbr = 0,
990 .last_mtc = decoder->last_mtc,
991 .ctc_timestamp = decoder->ctc_timestamp,
992 .ctc_delta = decoder->ctc_delta,
993 .tsc_timestamp = decoder->tsc_timestamp,
994 .timestamp = decoder->timestamp,
995 .have_tma = decoder->have_tma,
996 .fixup_last_mtc = decoder->fixup_last_mtc,
997 .from_mtc = from_mtc,
998 .cbr_cyc_to_tsc = 0,
999 };
1000
1001 /*
1002 * For now, do not support using TSC packets for at least the reasons:
1003 * 1) timing might have stopped
1004 * 2) TSC packets within PSB+ can slip against CYC packets
1005 */
1006 if (!from_mtc)
1007 return;
1008
1009 intel_pt_pkt_lookahead(decoder, intel_pt_calc_cyc_cb, &data);
1010 }
1011
intel_pt_get_next_packet(struct intel_pt_decoder * decoder)1012 static int intel_pt_get_next_packet(struct intel_pt_decoder *decoder)
1013 {
1014 int ret;
1015
1016 decoder->last_packet_type = decoder->packet.type;
1017
1018 do {
1019 decoder->pos += decoder->pkt_step;
1020 decoder->buf += decoder->pkt_step;
1021 decoder->len -= decoder->pkt_step;
1022
1023 if (!decoder->len) {
1024 ret = intel_pt_get_next_data(decoder, false);
1025 if (ret)
1026 return ret;
1027 }
1028
1029 decoder->prev_pkt_ctx = decoder->pkt_ctx;
1030 ret = intel_pt_get_packet(decoder->buf, decoder->len,
1031 &decoder->packet, &decoder->pkt_ctx);
1032 if (ret == INTEL_PT_NEED_MORE_BYTES && BITS_PER_LONG == 32 &&
1033 decoder->len < INTEL_PT_PKT_MAX_SZ && !decoder->next_buf) {
1034 ret = intel_pt_get_split_packet(decoder);
1035 if (ret < 0)
1036 return ret;
1037 }
1038 if (ret <= 0)
1039 return intel_pt_bad_packet(decoder);
1040
1041 decoder->pkt_len = ret;
1042 decoder->pkt_step = ret;
1043 intel_pt_decoder_log_packet(decoder);
1044 } while (decoder->packet.type == INTEL_PT_PAD);
1045
1046 return 0;
1047 }
1048
intel_pt_next_period(struct intel_pt_decoder * decoder)1049 static uint64_t intel_pt_next_period(struct intel_pt_decoder *decoder)
1050 {
1051 uint64_t timestamp, masked_timestamp;
1052
1053 timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
1054 masked_timestamp = timestamp & decoder->period_mask;
1055 if (decoder->continuous_period) {
1056 if (masked_timestamp > decoder->last_masked_timestamp)
1057 return 1;
1058 } else {
1059 timestamp += 1;
1060 masked_timestamp = timestamp & decoder->period_mask;
1061 if (masked_timestamp > decoder->last_masked_timestamp) {
1062 decoder->last_masked_timestamp = masked_timestamp;
1063 decoder->continuous_period = true;
1064 }
1065 }
1066
1067 if (masked_timestamp < decoder->last_masked_timestamp)
1068 return decoder->period_ticks;
1069
1070 return decoder->period_ticks - (timestamp - masked_timestamp);
1071 }
1072
intel_pt_next_sample(struct intel_pt_decoder * decoder)1073 static uint64_t intel_pt_next_sample(struct intel_pt_decoder *decoder)
1074 {
1075 switch (decoder->period_type) {
1076 case INTEL_PT_PERIOD_INSTRUCTIONS:
1077 return decoder->period - decoder->period_insn_cnt;
1078 case INTEL_PT_PERIOD_TICKS:
1079 return intel_pt_next_period(decoder);
1080 case INTEL_PT_PERIOD_NONE:
1081 case INTEL_PT_PERIOD_MTC:
1082 default:
1083 return 0;
1084 }
1085 }
1086
intel_pt_sample_insn(struct intel_pt_decoder * decoder)1087 static void intel_pt_sample_insn(struct intel_pt_decoder *decoder)
1088 {
1089 uint64_t timestamp, masked_timestamp;
1090
1091 switch (decoder->period_type) {
1092 case INTEL_PT_PERIOD_INSTRUCTIONS:
1093 decoder->period_insn_cnt = 0;
1094 break;
1095 case INTEL_PT_PERIOD_TICKS:
1096 timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
1097 masked_timestamp = timestamp & decoder->period_mask;
1098 if (masked_timestamp > decoder->last_masked_timestamp)
1099 decoder->last_masked_timestamp = masked_timestamp;
1100 else
1101 decoder->last_masked_timestamp += decoder->period_ticks;
1102 break;
1103 case INTEL_PT_PERIOD_NONE:
1104 case INTEL_PT_PERIOD_MTC:
1105 default:
1106 break;
1107 }
1108
1109 decoder->state.type |= INTEL_PT_INSTRUCTION;
1110 }
1111
1112 /*
1113 * Sample FUP instruction at the same time as reporting the FUP event, so the
1114 * instruction sample gets the same flags as the FUP event.
1115 */
intel_pt_sample_fup_insn(struct intel_pt_decoder * decoder)1116 static void intel_pt_sample_fup_insn(struct intel_pt_decoder *decoder)
1117 {
1118 struct intel_pt_insn intel_pt_insn;
1119 uint64_t max_insn_cnt, insn_cnt = 0;
1120 int err;
1121
1122 decoder->state.insn_op = INTEL_PT_OP_OTHER;
1123 decoder->state.insn_len = 0;
1124
1125 if (!decoder->branch_enable || !decoder->pge || decoder->hop ||
1126 decoder->ip != decoder->last_ip)
1127 return;
1128
1129 if (!decoder->mtc_insn)
1130 decoder->mtc_insn = true;
1131
1132 max_insn_cnt = intel_pt_next_sample(decoder);
1133 if (max_insn_cnt != 1)
1134 return;
1135
1136 err = decoder->walk_insn(&intel_pt_insn, &insn_cnt, &decoder->ip,
1137 0, max_insn_cnt, decoder->data);
1138 /* Ignore error, it will be reported next walk anyway */
1139 if (err)
1140 return;
1141
1142 if (intel_pt_insn.branch != INTEL_PT_BR_NO_BRANCH) {
1143 intel_pt_log_at("ERROR: Unexpected branch at FUP instruction", decoder->ip);
1144 return;
1145 }
1146
1147 decoder->tot_insn_cnt += insn_cnt;
1148 decoder->timestamp_insn_cnt += insn_cnt;
1149 decoder->sample_insn_cnt += insn_cnt;
1150 decoder->period_insn_cnt += insn_cnt;
1151
1152 intel_pt_sample_insn(decoder);
1153
1154 decoder->state.type |= INTEL_PT_INSTRUCTION;
1155 decoder->ip += intel_pt_insn.length;
1156 }
1157
intel_pt_walk_insn(struct intel_pt_decoder * decoder,struct intel_pt_insn * intel_pt_insn,uint64_t ip)1158 static int intel_pt_walk_insn(struct intel_pt_decoder *decoder,
1159 struct intel_pt_insn *intel_pt_insn, uint64_t ip)
1160 {
1161 uint64_t max_insn_cnt, insn_cnt = 0;
1162 int err;
1163
1164 if (!decoder->mtc_insn)
1165 decoder->mtc_insn = true;
1166
1167 max_insn_cnt = intel_pt_next_sample(decoder);
1168
1169 err = decoder->walk_insn(intel_pt_insn, &insn_cnt, &decoder->ip, ip,
1170 max_insn_cnt, decoder->data);
1171
1172 decoder->tot_insn_cnt += insn_cnt;
1173 decoder->timestamp_insn_cnt += insn_cnt;
1174 decoder->sample_insn_cnt += insn_cnt;
1175 decoder->period_insn_cnt += insn_cnt;
1176
1177 if (err) {
1178 decoder->no_progress = 0;
1179 decoder->pkt_state = INTEL_PT_STATE_ERR2;
1180 intel_pt_log_at("ERROR: Failed to get instruction",
1181 decoder->ip);
1182 if (err == -ENOENT)
1183 return -ENOLINK;
1184 return -EILSEQ;
1185 }
1186
1187 if (ip && decoder->ip == ip) {
1188 err = -EAGAIN;
1189 goto out;
1190 }
1191
1192 if (max_insn_cnt && insn_cnt >= max_insn_cnt)
1193 intel_pt_sample_insn(decoder);
1194
1195 if (intel_pt_insn->branch == INTEL_PT_BR_NO_BRANCH) {
1196 decoder->state.type = INTEL_PT_INSTRUCTION;
1197 decoder->state.from_ip = decoder->ip;
1198 decoder->state.to_ip = 0;
1199 decoder->ip += intel_pt_insn->length;
1200 err = INTEL_PT_RETURN;
1201 goto out;
1202 }
1203
1204 if (intel_pt_insn->op == INTEL_PT_OP_CALL) {
1205 /* Zero-length calls are excluded */
1206 if (intel_pt_insn->branch != INTEL_PT_BR_UNCONDITIONAL ||
1207 intel_pt_insn->rel) {
1208 err = intel_pt_push(&decoder->stack, decoder->ip +
1209 intel_pt_insn->length);
1210 if (err)
1211 goto out;
1212 }
1213 } else if (intel_pt_insn->op == INTEL_PT_OP_RET) {
1214 decoder->ret_addr = intel_pt_pop(&decoder->stack);
1215 }
1216
1217 if (intel_pt_insn->branch == INTEL_PT_BR_UNCONDITIONAL) {
1218 int cnt = decoder->no_progress++;
1219
1220 decoder->state.from_ip = decoder->ip;
1221 decoder->ip += intel_pt_insn->length +
1222 intel_pt_insn->rel;
1223 decoder->state.to_ip = decoder->ip;
1224 err = INTEL_PT_RETURN;
1225
1226 /*
1227 * Check for being stuck in a loop. This can happen if a
1228 * decoder error results in the decoder erroneously setting the
1229 * ip to an address that is itself in an infinite loop that
1230 * consumes no packets. When that happens, there must be an
1231 * unconditional branch.
1232 */
1233 if (cnt) {
1234 if (cnt == 1) {
1235 decoder->stuck_ip = decoder->state.to_ip;
1236 decoder->stuck_ip_prd = 1;
1237 decoder->stuck_ip_cnt = 1;
1238 } else if (cnt > decoder->max_loops ||
1239 decoder->state.to_ip == decoder->stuck_ip) {
1240 intel_pt_log_at("ERROR: Never-ending loop",
1241 decoder->state.to_ip);
1242 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1243 err = -ELOOP;
1244 goto out;
1245 } else if (!--decoder->stuck_ip_cnt) {
1246 decoder->stuck_ip_prd += 1;
1247 decoder->stuck_ip_cnt = decoder->stuck_ip_prd;
1248 decoder->stuck_ip = decoder->state.to_ip;
1249 }
1250 }
1251 goto out_no_progress;
1252 }
1253 out:
1254 decoder->no_progress = 0;
1255 out_no_progress:
1256 decoder->state.insn_op = intel_pt_insn->op;
1257 decoder->state.insn_len = intel_pt_insn->length;
1258 memcpy(decoder->state.insn, intel_pt_insn->buf,
1259 INTEL_PT_INSN_BUF_SZ);
1260
1261 if (decoder->tx_flags & INTEL_PT_IN_TX)
1262 decoder->state.flags |= INTEL_PT_IN_TX;
1263
1264 return err;
1265 }
1266
intel_pt_mode_exec_status(struct intel_pt_decoder * decoder)1267 static void intel_pt_mode_exec_status(struct intel_pt_decoder *decoder)
1268 {
1269 bool iflag = decoder->packet.count & INTEL_PT_IFLAG;
1270
1271 decoder->exec_mode = decoder->packet.payload;
1272 decoder->iflag = iflag;
1273 decoder->next_iflag = iflag;
1274 decoder->state.from_iflag = iflag;
1275 decoder->state.to_iflag = iflag;
1276 }
1277
intel_pt_mode_exec(struct intel_pt_decoder * decoder)1278 static void intel_pt_mode_exec(struct intel_pt_decoder *decoder)
1279 {
1280 bool iflag = decoder->packet.count & INTEL_PT_IFLAG;
1281
1282 decoder->exec_mode = decoder->packet.payload;
1283 decoder->next_iflag = iflag;
1284 }
1285
intel_pt_sample_iflag(struct intel_pt_decoder * decoder)1286 static void intel_pt_sample_iflag(struct intel_pt_decoder *decoder)
1287 {
1288 decoder->state.type |= INTEL_PT_IFLAG_CHG;
1289 decoder->state.from_iflag = decoder->iflag;
1290 decoder->state.to_iflag = decoder->next_iflag;
1291 decoder->iflag = decoder->next_iflag;
1292 }
1293
intel_pt_sample_iflag_chg(struct intel_pt_decoder * decoder)1294 static void intel_pt_sample_iflag_chg(struct intel_pt_decoder *decoder)
1295 {
1296 if (decoder->iflag != decoder->next_iflag)
1297 intel_pt_sample_iflag(decoder);
1298 }
1299
intel_pt_clear_fup_event(struct intel_pt_decoder * decoder)1300 static void intel_pt_clear_fup_event(struct intel_pt_decoder *decoder)
1301 {
1302 decoder->set_fup_tx_flags = false;
1303 decoder->set_fup_ptw = false;
1304 decoder->set_fup_mwait = false;
1305 decoder->set_fup_pwre = false;
1306 decoder->set_fup_exstop = false;
1307 decoder->set_fup_bep = false;
1308 decoder->set_fup_cfe_ip = false;
1309 decoder->set_fup_cfe = false;
1310 decoder->evd_cnt = 0;
1311 decoder->set_fup_mode_exec = false;
1312 decoder->iflag = decoder->next_iflag;
1313 }
1314
intel_pt_fup_event(struct intel_pt_decoder * decoder,bool no_tip)1315 static bool intel_pt_fup_event(struct intel_pt_decoder *decoder, bool no_tip)
1316 {
1317 enum intel_pt_sample_type type = decoder->state.type;
1318 bool sample_fup_insn = false;
1319 bool ret = false;
1320
1321 decoder->state.type &= ~INTEL_PT_BRANCH;
1322
1323 if (decoder->set_fup_cfe_ip || decoder->set_fup_cfe) {
1324 bool ip = decoder->set_fup_cfe_ip;
1325
1326 decoder->set_fup_cfe_ip = false;
1327 decoder->set_fup_cfe = false;
1328 decoder->state.type |= INTEL_PT_EVT;
1329 if (!ip && decoder->pge)
1330 decoder->state.type |= INTEL_PT_BRANCH;
1331 decoder->state.cfe_type = decoder->fup_cfe_pkt.count;
1332 decoder->state.cfe_vector = decoder->fup_cfe_pkt.payload;
1333 decoder->state.evd_cnt = decoder->evd_cnt;
1334 decoder->state.evd = decoder->evd;
1335 decoder->evd_cnt = 0;
1336 if (ip || decoder->pge)
1337 decoder->state.flags |= INTEL_PT_FUP_IP;
1338 ret = true;
1339 }
1340 if (decoder->set_fup_mode_exec) {
1341 decoder->set_fup_mode_exec = false;
1342 intel_pt_sample_iflag(decoder);
1343 sample_fup_insn = no_tip;
1344 ret = true;
1345 }
1346 if (decoder->set_fup_tx_flags) {
1347 decoder->set_fup_tx_flags = false;
1348 decoder->tx_flags = decoder->fup_tx_flags;
1349 decoder->state.type |= INTEL_PT_TRANSACTION;
1350 if (decoder->fup_tx_flags & INTEL_PT_ABORT_TX)
1351 decoder->state.type |= INTEL_PT_BRANCH;
1352 decoder->state.flags = decoder->fup_tx_flags;
1353 ret = true;
1354 }
1355 if (decoder->set_fup_ptw) {
1356 decoder->set_fup_ptw = false;
1357 decoder->state.type |= INTEL_PT_PTW;
1358 decoder->state.flags |= INTEL_PT_FUP_IP;
1359 decoder->state.ptw_payload = decoder->fup_ptw_payload;
1360 ret = true;
1361 }
1362 if (decoder->set_fup_mwait) {
1363 decoder->set_fup_mwait = false;
1364 decoder->state.type |= INTEL_PT_MWAIT_OP;
1365 decoder->state.mwait_payload = decoder->fup_mwait_payload;
1366 ret = true;
1367 }
1368 if (decoder->set_fup_pwre) {
1369 decoder->set_fup_pwre = false;
1370 decoder->state.type |= INTEL_PT_PWR_ENTRY;
1371 decoder->state.pwre_payload = decoder->fup_pwre_payload;
1372 ret = true;
1373 }
1374 if (decoder->set_fup_exstop) {
1375 decoder->set_fup_exstop = false;
1376 decoder->state.type |= INTEL_PT_EX_STOP;
1377 decoder->state.flags |= INTEL_PT_FUP_IP;
1378 ret = true;
1379 }
1380 if (decoder->set_fup_bep) {
1381 decoder->set_fup_bep = false;
1382 decoder->state.type |= INTEL_PT_BLK_ITEMS;
1383 ret = true;
1384 }
1385 if (decoder->overflow) {
1386 decoder->overflow = false;
1387 if (!ret && !decoder->pge) {
1388 if (decoder->hop) {
1389 decoder->state.type = 0;
1390 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
1391 }
1392 decoder->pge = true;
1393 decoder->state.type |= INTEL_PT_BRANCH | INTEL_PT_TRACE_BEGIN;
1394 decoder->state.from_ip = 0;
1395 decoder->state.to_ip = decoder->ip;
1396 return true;
1397 }
1398 }
1399 if (ret) {
1400 decoder->state.from_ip = decoder->ip;
1401 decoder->state.to_ip = 0;
1402 if (sample_fup_insn)
1403 intel_pt_sample_fup_insn(decoder);
1404 } else {
1405 decoder->state.type = type;
1406 }
1407 return ret;
1408 }
1409
intel_pt_fup_with_nlip(struct intel_pt_decoder * decoder,struct intel_pt_insn * intel_pt_insn,uint64_t ip,int err)1410 static inline bool intel_pt_fup_with_nlip(struct intel_pt_decoder *decoder,
1411 struct intel_pt_insn *intel_pt_insn,
1412 uint64_t ip, int err)
1413 {
1414 return decoder->flags & INTEL_PT_FUP_WITH_NLIP && !err &&
1415 intel_pt_insn->branch == INTEL_PT_BR_INDIRECT &&
1416 ip == decoder->ip + intel_pt_insn->length;
1417 }
1418
intel_pt_walk_fup(struct intel_pt_decoder * decoder)1419 static int intel_pt_walk_fup(struct intel_pt_decoder *decoder)
1420 {
1421 struct intel_pt_insn intel_pt_insn;
1422 uint64_t ip;
1423 int err;
1424
1425 ip = decoder->last_ip;
1426
1427 while (1) {
1428 err = intel_pt_walk_insn(decoder, &intel_pt_insn, ip);
1429 if (err == INTEL_PT_RETURN)
1430 return 0;
1431 if (err == -EAGAIN ||
1432 intel_pt_fup_with_nlip(decoder, &intel_pt_insn, ip, err)) {
1433 bool no_tip = decoder->pkt_state != INTEL_PT_STATE_FUP;
1434
1435 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1436 if (intel_pt_fup_event(decoder, no_tip) && no_tip)
1437 return 0;
1438 return -EAGAIN;
1439 }
1440 decoder->set_fup_tx_flags = false;
1441 if (err)
1442 return err;
1443
1444 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1445 intel_pt_log_at("ERROR: Unexpected indirect branch",
1446 decoder->ip);
1447 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1448 return -ENOENT;
1449 }
1450
1451 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1452 intel_pt_log_at("ERROR: Unexpected conditional branch",
1453 decoder->ip);
1454 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1455 return -ENOENT;
1456 }
1457
1458 intel_pt_bug(decoder);
1459 }
1460 }
1461
intel_pt_walk_tip(struct intel_pt_decoder * decoder)1462 static int intel_pt_walk_tip(struct intel_pt_decoder *decoder)
1463 {
1464 struct intel_pt_insn intel_pt_insn;
1465 int err;
1466
1467 err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1468 if (err == INTEL_PT_RETURN &&
1469 decoder->pgd_ip &&
1470 decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1471 (decoder->state.type & INTEL_PT_BRANCH) &&
1472 decoder->pgd_ip(decoder->state.to_ip, decoder->data)) {
1473 /* Unconditional branch leaving filter region */
1474 decoder->no_progress = 0;
1475 decoder->pge = false;
1476 decoder->continuous_period = false;
1477 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1478 decoder->state.type |= INTEL_PT_TRACE_END;
1479 intel_pt_update_nr(decoder);
1480 return 0;
1481 }
1482 if (err == INTEL_PT_RETURN)
1483 return 0;
1484 if (err)
1485 return err;
1486
1487 intel_pt_update_nr(decoder);
1488 intel_pt_sample_iflag_chg(decoder);
1489
1490 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1491 if (decoder->pkt_state == INTEL_PT_STATE_TIP_PGD) {
1492 decoder->pge = false;
1493 decoder->continuous_period = false;
1494 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1495 decoder->state.from_ip = decoder->ip;
1496 if (decoder->packet.count == 0) {
1497 decoder->state.to_ip = 0;
1498 } else {
1499 decoder->state.to_ip = decoder->last_ip;
1500 decoder->ip = decoder->last_ip;
1501 }
1502 decoder->state.type |= INTEL_PT_TRACE_END;
1503 } else {
1504 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1505 decoder->state.from_ip = decoder->ip;
1506 if (decoder->packet.count == 0) {
1507 decoder->state.to_ip = 0;
1508 } else {
1509 decoder->state.to_ip = decoder->last_ip;
1510 decoder->ip = decoder->last_ip;
1511 }
1512 }
1513 return 0;
1514 }
1515
1516 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1517 uint64_t to_ip = decoder->ip + intel_pt_insn.length +
1518 intel_pt_insn.rel;
1519
1520 if (decoder->pgd_ip &&
1521 decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1522 decoder->pgd_ip(to_ip, decoder->data)) {
1523 /* Conditional branch leaving filter region */
1524 decoder->pge = false;
1525 decoder->continuous_period = false;
1526 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1527 decoder->ip = to_ip;
1528 decoder->state.from_ip = decoder->ip;
1529 decoder->state.to_ip = to_ip;
1530 decoder->state.type |= INTEL_PT_TRACE_END;
1531 return 0;
1532 }
1533 intel_pt_log_at("ERROR: Conditional branch when expecting indirect branch",
1534 decoder->ip);
1535 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1536 return -ENOENT;
1537 }
1538
1539 return intel_pt_bug(decoder);
1540 }
1541
1542 struct eptw_data {
1543 int bit_countdown;
1544 uint64_t payload;
1545 };
1546
intel_pt_eptw_lookahead_cb(struct intel_pt_pkt_info * pkt_info)1547 static int intel_pt_eptw_lookahead_cb(struct intel_pt_pkt_info *pkt_info)
1548 {
1549 struct eptw_data *data = pkt_info->data;
1550 int nr_bits;
1551
1552 switch (pkt_info->packet.type) {
1553 case INTEL_PT_PAD:
1554 case INTEL_PT_MNT:
1555 case INTEL_PT_MODE_EXEC:
1556 case INTEL_PT_MODE_TSX:
1557 case INTEL_PT_MTC:
1558 case INTEL_PT_FUP:
1559 case INTEL_PT_CYC:
1560 case INTEL_PT_CBR:
1561 case INTEL_PT_TSC:
1562 case INTEL_PT_TMA:
1563 case INTEL_PT_PIP:
1564 case INTEL_PT_VMCS:
1565 case INTEL_PT_PSB:
1566 case INTEL_PT_PSBEND:
1567 case INTEL_PT_PTWRITE:
1568 case INTEL_PT_PTWRITE_IP:
1569 case INTEL_PT_EXSTOP:
1570 case INTEL_PT_EXSTOP_IP:
1571 case INTEL_PT_MWAIT:
1572 case INTEL_PT_PWRE:
1573 case INTEL_PT_PWRX:
1574 case INTEL_PT_BBP:
1575 case INTEL_PT_BIP:
1576 case INTEL_PT_BEP:
1577 case INTEL_PT_BEP_IP:
1578 case INTEL_PT_CFE:
1579 case INTEL_PT_CFE_IP:
1580 case INTEL_PT_EVD:
1581 break;
1582
1583 case INTEL_PT_TNT:
1584 nr_bits = data->bit_countdown;
1585 if (nr_bits > pkt_info->packet.count)
1586 nr_bits = pkt_info->packet.count;
1587 data->payload <<= nr_bits;
1588 data->payload |= pkt_info->packet.payload >> (64 - nr_bits);
1589 data->bit_countdown -= nr_bits;
1590 return !data->bit_countdown;
1591
1592 case INTEL_PT_TIP_PGE:
1593 case INTEL_PT_TIP_PGD:
1594 case INTEL_PT_TIP:
1595 case INTEL_PT_BAD:
1596 case INTEL_PT_OVF:
1597 case INTEL_PT_TRACESTOP:
1598 default:
1599 return 1;
1600 }
1601
1602 return 0;
1603 }
1604
intel_pt_emulated_ptwrite(struct intel_pt_decoder * decoder)1605 static int intel_pt_emulated_ptwrite(struct intel_pt_decoder *decoder)
1606 {
1607 int n = 64 - decoder->tnt.count;
1608 struct eptw_data data = {
1609 .bit_countdown = n,
1610 .payload = decoder->tnt.payload >> n,
1611 };
1612
1613 decoder->emulated_ptwrite = false;
1614 intel_pt_log("Emulated ptwrite detected\n");
1615
1616 intel_pt_pkt_lookahead(decoder, intel_pt_eptw_lookahead_cb, &data);
1617 if (data.bit_countdown)
1618 return -ECONNRESET;
1619
1620 decoder->state.type = INTEL_PT_PTW;
1621 decoder->state.from_ip = decoder->ip;
1622 decoder->state.to_ip = 0;
1623 decoder->state.ptw_payload = data.payload;
1624 return 0;
1625 }
1626
intel_pt_walk_tnt(struct intel_pt_decoder * decoder)1627 static int intel_pt_walk_tnt(struct intel_pt_decoder *decoder)
1628 {
1629 struct intel_pt_insn intel_pt_insn;
1630 int err;
1631
1632 while (1) {
1633 if (decoder->emulated_ptwrite)
1634 return intel_pt_emulated_ptwrite(decoder);
1635 err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1636 if (err == INTEL_PT_RETURN) {
1637 decoder->emulated_ptwrite = intel_pt_insn.emulated_ptwrite;
1638 return 0;
1639 }
1640 if (err) {
1641 decoder->emulated_ptwrite = false;
1642 return err;
1643 }
1644
1645 if (intel_pt_insn.op == INTEL_PT_OP_RET) {
1646 if (!decoder->return_compression) {
1647 intel_pt_log_at("ERROR: RET when expecting conditional branch",
1648 decoder->ip);
1649 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1650 return -ENOENT;
1651 }
1652 if (!decoder->ret_addr) {
1653 intel_pt_log_at("ERROR: Bad RET compression (stack empty)",
1654 decoder->ip);
1655 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1656 return -ENOENT;
1657 }
1658 if (!(decoder->tnt.payload & BIT63)) {
1659 intel_pt_log_at("ERROR: Bad RET compression (TNT=N)",
1660 decoder->ip);
1661 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1662 return -ENOENT;
1663 }
1664 decoder->tnt.count -= 1;
1665 if (decoder->tnt.count)
1666 decoder->pkt_state = INTEL_PT_STATE_TNT_CONT;
1667 else
1668 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1669 decoder->tnt.payload <<= 1;
1670 decoder->state.from_ip = decoder->ip;
1671 decoder->ip = decoder->ret_addr;
1672 decoder->state.to_ip = decoder->ip;
1673 return 0;
1674 }
1675
1676 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1677 /* Handle deferred TIPs */
1678 err = intel_pt_get_next_packet(decoder);
1679 if (err)
1680 return err;
1681 if (decoder->packet.type != INTEL_PT_TIP ||
1682 decoder->packet.count == 0) {
1683 intel_pt_log_at("ERROR: Missing deferred TIP for indirect branch",
1684 decoder->ip);
1685 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1686 decoder->pkt_step = 0;
1687 return -ENOENT;
1688 }
1689 intel_pt_set_last_ip(decoder);
1690 decoder->state.from_ip = decoder->ip;
1691 decoder->state.to_ip = decoder->last_ip;
1692 decoder->ip = decoder->last_ip;
1693 intel_pt_update_nr(decoder);
1694 intel_pt_sample_iflag_chg(decoder);
1695 return 0;
1696 }
1697
1698 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1699 decoder->tnt.count -= 1;
1700 if (decoder->tnt.count)
1701 decoder->pkt_state = INTEL_PT_STATE_TNT_CONT;
1702 else
1703 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1704 if (decoder->tnt.payload & BIT63) {
1705 decoder->tnt.payload <<= 1;
1706 decoder->state.from_ip = decoder->ip;
1707 decoder->ip += intel_pt_insn.length +
1708 intel_pt_insn.rel;
1709 decoder->state.to_ip = decoder->ip;
1710 return 0;
1711 }
1712 /* Instruction sample for a non-taken branch */
1713 if (decoder->state.type & INTEL_PT_INSTRUCTION) {
1714 decoder->tnt.payload <<= 1;
1715 decoder->state.type = INTEL_PT_INSTRUCTION;
1716 decoder->state.from_ip = decoder->ip;
1717 decoder->state.to_ip = 0;
1718 decoder->ip += intel_pt_insn.length;
1719 return 0;
1720 }
1721 decoder->sample_cyc = false;
1722 decoder->ip += intel_pt_insn.length;
1723 if (!decoder->tnt.count) {
1724 intel_pt_update_sample_time(decoder);
1725 return -EAGAIN;
1726 }
1727 decoder->tnt.payload <<= 1;
1728 continue;
1729 }
1730
1731 return intel_pt_bug(decoder);
1732 }
1733 }
1734
intel_pt_mode_tsx(struct intel_pt_decoder * decoder,bool * no_tip)1735 static int intel_pt_mode_tsx(struct intel_pt_decoder *decoder, bool *no_tip)
1736 {
1737 unsigned int fup_tx_flags;
1738 int err;
1739
1740 fup_tx_flags = decoder->packet.payload &
1741 (INTEL_PT_IN_TX | INTEL_PT_ABORT_TX);
1742 err = intel_pt_get_next_packet(decoder);
1743 if (err)
1744 return err;
1745 if (decoder->packet.type == INTEL_PT_FUP) {
1746 decoder->fup_tx_flags = fup_tx_flags;
1747 decoder->set_fup_tx_flags = true;
1748 if (!(decoder->fup_tx_flags & INTEL_PT_ABORT_TX))
1749 *no_tip = true;
1750 } else {
1751 intel_pt_log_at("ERROR: Missing FUP after MODE.TSX",
1752 decoder->pos);
1753 intel_pt_update_in_tx(decoder);
1754 }
1755 return 0;
1756 }
1757
intel_pt_evd(struct intel_pt_decoder * decoder)1758 static int intel_pt_evd(struct intel_pt_decoder *decoder)
1759 {
1760 if (decoder->evd_cnt >= INTEL_PT_MAX_EVDS) {
1761 intel_pt_log_at("ERROR: Too many EVD packets", decoder->pos);
1762 return -ENOSYS;
1763 }
1764 decoder->evd[decoder->evd_cnt++] = (struct intel_pt_evd){
1765 .type = decoder->packet.count,
1766 .payload = decoder->packet.payload,
1767 };
1768 return 0;
1769 }
1770
intel_pt_8b_tsc(uint64_t timestamp,uint64_t ref_timestamp)1771 static uint64_t intel_pt_8b_tsc(uint64_t timestamp, uint64_t ref_timestamp)
1772 {
1773 timestamp |= (ref_timestamp & (0xffULL << 56));
1774
1775 if (timestamp < ref_timestamp) {
1776 if (ref_timestamp - timestamp > (1ULL << 55))
1777 timestamp += (1ULL << 56);
1778 } else {
1779 if (timestamp - ref_timestamp > (1ULL << 55))
1780 timestamp -= (1ULL << 56);
1781 }
1782
1783 return timestamp;
1784 }
1785
1786 /* For use only when decoder->vm_time_correlation is true */
intel_pt_time_in_range(struct intel_pt_decoder * decoder,uint64_t timestamp)1787 static bool intel_pt_time_in_range(struct intel_pt_decoder *decoder,
1788 uint64_t timestamp)
1789 {
1790 uint64_t max_timestamp = decoder->buf_timestamp;
1791
1792 if (!max_timestamp) {
1793 max_timestamp = decoder->last_reliable_timestamp +
1794 0x400000000ULL;
1795 }
1796 return timestamp >= decoder->last_reliable_timestamp &&
1797 timestamp < decoder->buf_timestamp;
1798 }
1799
intel_pt_calc_tsc_timestamp(struct intel_pt_decoder * decoder)1800 static void intel_pt_calc_tsc_timestamp(struct intel_pt_decoder *decoder)
1801 {
1802 uint64_t timestamp;
1803 bool bad = false;
1804
1805 decoder->have_tma = false;
1806
1807 if (decoder->ref_timestamp) {
1808 timestamp = intel_pt_8b_tsc(decoder->packet.payload,
1809 decoder->ref_timestamp);
1810 decoder->tsc_timestamp = timestamp;
1811 decoder->timestamp = timestamp;
1812 decoder->ref_timestamp = 0;
1813 decoder->timestamp_insn_cnt = 0;
1814 } else if (decoder->timestamp) {
1815 timestamp = decoder->packet.payload |
1816 (decoder->timestamp & (0xffULL << 56));
1817 decoder->tsc_timestamp = timestamp;
1818 if (timestamp < decoder->timestamp &&
1819 decoder->timestamp - timestamp < decoder->tsc_slip) {
1820 intel_pt_log_to("Suppressing backwards timestamp",
1821 timestamp);
1822 timestamp = decoder->timestamp;
1823 }
1824 if (timestamp < decoder->timestamp) {
1825 if (!decoder->buf_timestamp ||
1826 (timestamp + (1ULL << 56) < decoder->buf_timestamp)) {
1827 intel_pt_log_to("Wraparound timestamp", timestamp);
1828 timestamp += (1ULL << 56);
1829 decoder->tsc_timestamp = timestamp;
1830 } else {
1831 intel_pt_log_to("Suppressing bad timestamp", timestamp);
1832 timestamp = decoder->timestamp;
1833 bad = true;
1834 }
1835 }
1836 if (decoder->vm_time_correlation &&
1837 (bad || !intel_pt_time_in_range(decoder, timestamp)) &&
1838 intel_pt_print_once(decoder, INTEL_PT_PRT_ONCE_ERANGE))
1839 p_log("Timestamp out of range");
1840 decoder->timestamp = timestamp;
1841 decoder->timestamp_insn_cnt = 0;
1842 }
1843
1844 if (decoder->last_packet_type == INTEL_PT_CYC) {
1845 decoder->cyc_ref_timestamp = decoder->timestamp;
1846 decoder->cycle_cnt = 0;
1847 decoder->have_calc_cyc_to_tsc = false;
1848 intel_pt_calc_cyc_to_tsc(decoder, false);
1849 }
1850
1851 intel_pt_log_to("Setting timestamp", decoder->timestamp);
1852 }
1853
intel_pt_overflow(struct intel_pt_decoder * decoder)1854 static int intel_pt_overflow(struct intel_pt_decoder *decoder)
1855 {
1856 intel_pt_log("ERROR: Buffer overflow\n");
1857 intel_pt_clear_tx_flags(decoder);
1858 intel_pt_set_nr(decoder);
1859 decoder->timestamp_insn_cnt = 0;
1860 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1861 decoder->state.from_ip = decoder->ip;
1862 decoder->ip = 0;
1863 decoder->pge = false;
1864 intel_pt_clear_fup_event(decoder);
1865 decoder->overflow = true;
1866 return -EOVERFLOW;
1867 }
1868
intel_pt_mtc_cyc_cnt_pge(struct intel_pt_decoder * decoder)1869 static inline void intel_pt_mtc_cyc_cnt_pge(struct intel_pt_decoder *decoder)
1870 {
1871 if (decoder->have_cyc)
1872 return;
1873
1874 decoder->cyc_cnt_timestamp = decoder->timestamp;
1875 decoder->base_cyc_cnt = decoder->tot_cyc_cnt;
1876 }
1877
intel_pt_mtc_cyc_cnt_cbr(struct intel_pt_decoder * decoder)1878 static inline void intel_pt_mtc_cyc_cnt_cbr(struct intel_pt_decoder *decoder)
1879 {
1880 decoder->tsc_to_cyc = decoder->cbr / decoder->max_non_turbo_ratio_fp;
1881
1882 if (decoder->pge)
1883 intel_pt_mtc_cyc_cnt_pge(decoder);
1884 }
1885
intel_pt_mtc_cyc_cnt_upd(struct intel_pt_decoder * decoder)1886 static inline void intel_pt_mtc_cyc_cnt_upd(struct intel_pt_decoder *decoder)
1887 {
1888 uint64_t tot_cyc_cnt, tsc_delta;
1889
1890 if (decoder->have_cyc)
1891 return;
1892
1893 decoder->sample_cyc = true;
1894
1895 if (!decoder->pge || decoder->timestamp <= decoder->cyc_cnt_timestamp)
1896 return;
1897
1898 tsc_delta = decoder->timestamp - decoder->cyc_cnt_timestamp;
1899 tot_cyc_cnt = tsc_delta * decoder->tsc_to_cyc + decoder->base_cyc_cnt;
1900
1901 if (tot_cyc_cnt > decoder->tot_cyc_cnt)
1902 decoder->tot_cyc_cnt = tot_cyc_cnt;
1903 }
1904
intel_pt_calc_tma(struct intel_pt_decoder * decoder)1905 static void intel_pt_calc_tma(struct intel_pt_decoder *decoder)
1906 {
1907 uint32_t ctc = decoder->packet.payload;
1908 uint32_t fc = decoder->packet.count;
1909 uint32_t ctc_rem = ctc & decoder->ctc_rem_mask;
1910
1911 if (!decoder->tsc_ctc_ratio_d)
1912 return;
1913
1914 if (decoder->pge && !decoder->in_psb)
1915 intel_pt_mtc_cyc_cnt_pge(decoder);
1916 else
1917 intel_pt_mtc_cyc_cnt_upd(decoder);
1918
1919 decoder->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
1920 decoder->last_ctc = ctc - ctc_rem;
1921 decoder->ctc_timestamp = decoder->tsc_timestamp - fc;
1922 if (decoder->tsc_ctc_mult) {
1923 decoder->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
1924 } else {
1925 decoder->ctc_timestamp -= multdiv(ctc_rem,
1926 decoder->tsc_ctc_ratio_n,
1927 decoder->tsc_ctc_ratio_d);
1928 }
1929 decoder->ctc_delta = 0;
1930 decoder->have_tma = true;
1931 decoder->fixup_last_mtc = true;
1932 intel_pt_log("CTC timestamp " x64_fmt " last MTC %#x CTC rem %#x\n",
1933 decoder->ctc_timestamp, decoder->last_mtc, ctc_rem);
1934 }
1935
intel_pt_calc_mtc_timestamp(struct intel_pt_decoder * decoder)1936 static void intel_pt_calc_mtc_timestamp(struct intel_pt_decoder *decoder)
1937 {
1938 uint64_t timestamp;
1939 uint32_t mtc, mtc_delta;
1940
1941 if (!decoder->have_tma)
1942 return;
1943
1944 mtc = decoder->packet.payload;
1945
1946 if (decoder->mtc_shift > 8 && decoder->fixup_last_mtc) {
1947 decoder->fixup_last_mtc = false;
1948 intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
1949 &decoder->last_mtc);
1950 }
1951
1952 if (mtc > decoder->last_mtc)
1953 mtc_delta = mtc - decoder->last_mtc;
1954 else
1955 mtc_delta = mtc + 256 - decoder->last_mtc;
1956
1957 decoder->ctc_delta += mtc_delta << decoder->mtc_shift;
1958
1959 if (decoder->tsc_ctc_mult) {
1960 timestamp = decoder->ctc_timestamp +
1961 decoder->ctc_delta * decoder->tsc_ctc_mult;
1962 } else {
1963 timestamp = decoder->ctc_timestamp +
1964 multdiv(decoder->ctc_delta,
1965 decoder->tsc_ctc_ratio_n,
1966 decoder->tsc_ctc_ratio_d);
1967 }
1968
1969 if (timestamp < decoder->timestamp)
1970 intel_pt_log("Suppressing MTC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1971 timestamp, decoder->timestamp);
1972 else
1973 decoder->timestamp = timestamp;
1974
1975 intel_pt_mtc_cyc_cnt_upd(decoder);
1976
1977 decoder->timestamp_insn_cnt = 0;
1978 decoder->last_mtc = mtc;
1979
1980 if (decoder->last_packet_type == INTEL_PT_CYC) {
1981 decoder->cyc_ref_timestamp = decoder->timestamp;
1982 decoder->cycle_cnt = 0;
1983 decoder->have_calc_cyc_to_tsc = false;
1984 intel_pt_calc_cyc_to_tsc(decoder, true);
1985 }
1986
1987 intel_pt_log_to("Setting timestamp", decoder->timestamp);
1988 }
1989
intel_pt_calc_cbr(struct intel_pt_decoder * decoder)1990 static void intel_pt_calc_cbr(struct intel_pt_decoder *decoder)
1991 {
1992 unsigned int cbr = decoder->packet.payload & 0xff;
1993
1994 decoder->cbr_payload = decoder->packet.payload;
1995
1996 if (decoder->cbr == cbr)
1997 return;
1998
1999 decoder->cbr = cbr;
2000 decoder->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
2001
2002 intel_pt_mtc_cyc_cnt_cbr(decoder);
2003 }
2004
intel_pt_calc_cyc_timestamp(struct intel_pt_decoder * decoder)2005 static void intel_pt_calc_cyc_timestamp(struct intel_pt_decoder *decoder)
2006 {
2007 uint64_t timestamp = decoder->cyc_ref_timestamp;
2008
2009 decoder->have_cyc = true;
2010
2011 decoder->cycle_cnt += decoder->packet.payload;
2012 if (decoder->pge)
2013 decoder->tot_cyc_cnt += decoder->packet.payload;
2014 decoder->sample_cyc = true;
2015
2016 if (!decoder->cyc_ref_timestamp)
2017 return;
2018
2019 if (decoder->have_calc_cyc_to_tsc)
2020 timestamp += decoder->cycle_cnt * decoder->calc_cyc_to_tsc;
2021 else if (decoder->cbr)
2022 timestamp += decoder->cycle_cnt * decoder->cbr_cyc_to_tsc;
2023 else
2024 return;
2025
2026 if (timestamp < decoder->timestamp)
2027 intel_pt_log("Suppressing CYC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
2028 timestamp, decoder->timestamp);
2029 else
2030 decoder->timestamp = timestamp;
2031
2032 decoder->timestamp_insn_cnt = 0;
2033
2034 intel_pt_log_to("Setting timestamp", decoder->timestamp);
2035 }
2036
intel_pt_bbp(struct intel_pt_decoder * decoder)2037 static void intel_pt_bbp(struct intel_pt_decoder *decoder)
2038 {
2039 if (decoder->prev_pkt_ctx == INTEL_PT_NO_CTX) {
2040 memset(decoder->state.items.mask, 0, sizeof(decoder->state.items.mask));
2041 decoder->state.items.is_32_bit = false;
2042 }
2043 decoder->blk_type = decoder->packet.payload;
2044 decoder->blk_type_pos = intel_pt_blk_type_pos(decoder->blk_type);
2045 if (decoder->blk_type == INTEL_PT_GP_REGS)
2046 decoder->state.items.is_32_bit = decoder->packet.count;
2047 if (decoder->blk_type_pos < 0) {
2048 intel_pt_log("WARNING: Unknown block type %u\n",
2049 decoder->blk_type);
2050 } else if (decoder->state.items.mask[decoder->blk_type_pos]) {
2051 intel_pt_log("WARNING: Duplicate block type %u\n",
2052 decoder->blk_type);
2053 }
2054 }
2055
intel_pt_bip(struct intel_pt_decoder * decoder)2056 static void intel_pt_bip(struct intel_pt_decoder *decoder)
2057 {
2058 uint32_t id = decoder->packet.count;
2059 uint32_t bit = 1 << id;
2060 int pos = decoder->blk_type_pos;
2061
2062 if (pos < 0 || id >= INTEL_PT_BLK_ITEM_ID_CNT) {
2063 intel_pt_log("WARNING: Unknown block item %u type %d\n",
2064 id, decoder->blk_type);
2065 return;
2066 }
2067
2068 if (decoder->state.items.mask[pos] & bit) {
2069 intel_pt_log("WARNING: Duplicate block item %u type %d\n",
2070 id, decoder->blk_type);
2071 }
2072
2073 decoder->state.items.mask[pos] |= bit;
2074 decoder->state.items.val[pos][id] = decoder->packet.payload;
2075 }
2076
2077 /* Walk PSB+ packets when already in sync. */
intel_pt_walk_psbend(struct intel_pt_decoder * decoder)2078 static int intel_pt_walk_psbend(struct intel_pt_decoder *decoder)
2079 {
2080 int err;
2081
2082 decoder->in_psb = true;
2083
2084 while (1) {
2085 err = intel_pt_get_next_packet(decoder);
2086 if (err)
2087 goto out;
2088
2089 switch (decoder->packet.type) {
2090 case INTEL_PT_PSBEND:
2091 err = 0;
2092 goto out;
2093
2094 case INTEL_PT_TIP_PGD:
2095 case INTEL_PT_TIP_PGE:
2096 case INTEL_PT_TIP:
2097 case INTEL_PT_TNT:
2098 case INTEL_PT_TRACESTOP:
2099 case INTEL_PT_BAD:
2100 case INTEL_PT_PSB:
2101 case INTEL_PT_PTWRITE:
2102 case INTEL_PT_PTWRITE_IP:
2103 case INTEL_PT_EXSTOP:
2104 case INTEL_PT_EXSTOP_IP:
2105 case INTEL_PT_MWAIT:
2106 case INTEL_PT_PWRE:
2107 case INTEL_PT_PWRX:
2108 case INTEL_PT_BBP:
2109 case INTEL_PT_BIP:
2110 case INTEL_PT_BEP:
2111 case INTEL_PT_BEP_IP:
2112 case INTEL_PT_CFE:
2113 case INTEL_PT_CFE_IP:
2114 case INTEL_PT_EVD:
2115 decoder->have_tma = false;
2116 intel_pt_log("ERROR: Unexpected packet\n");
2117 err = -EAGAIN;
2118 goto out;
2119
2120 case INTEL_PT_OVF:
2121 err = intel_pt_overflow(decoder);
2122 goto out;
2123
2124 case INTEL_PT_TSC:
2125 intel_pt_calc_tsc_timestamp(decoder);
2126 break;
2127
2128 case INTEL_PT_TMA:
2129 intel_pt_calc_tma(decoder);
2130 break;
2131
2132 case INTEL_PT_CBR:
2133 intel_pt_calc_cbr(decoder);
2134 break;
2135
2136 case INTEL_PT_MODE_EXEC:
2137 intel_pt_mode_exec_status(decoder);
2138 break;
2139
2140 case INTEL_PT_PIP:
2141 intel_pt_set_pip(decoder);
2142 break;
2143
2144 case INTEL_PT_FUP:
2145 decoder->pge = true;
2146 if (decoder->packet.count) {
2147 intel_pt_set_last_ip(decoder);
2148 decoder->psb_ip = decoder->last_ip;
2149 }
2150 break;
2151
2152 case INTEL_PT_MODE_TSX:
2153 intel_pt_update_in_tx(decoder);
2154 break;
2155
2156 case INTEL_PT_MTC:
2157 intel_pt_calc_mtc_timestamp(decoder);
2158 if (decoder->period_type == INTEL_PT_PERIOD_MTC)
2159 decoder->state.type |= INTEL_PT_INSTRUCTION;
2160 break;
2161
2162 case INTEL_PT_CYC:
2163 intel_pt_calc_cyc_timestamp(decoder);
2164 break;
2165
2166 case INTEL_PT_VMCS:
2167 case INTEL_PT_MNT:
2168 case INTEL_PT_PAD:
2169 default:
2170 break;
2171 }
2172 }
2173 out:
2174 decoder->in_psb = false;
2175
2176 return err;
2177 }
2178
intel_pt_walk_fup_tip(struct intel_pt_decoder * decoder)2179 static int intel_pt_walk_fup_tip(struct intel_pt_decoder *decoder)
2180 {
2181 int err;
2182
2183 if (decoder->tx_flags & INTEL_PT_ABORT_TX) {
2184 decoder->tx_flags = 0;
2185 decoder->state.flags &= ~INTEL_PT_IN_TX;
2186 decoder->state.flags |= INTEL_PT_ABORT_TX;
2187 } else {
2188 decoder->state.flags |= INTEL_PT_ASYNC;
2189 }
2190
2191 while (1) {
2192 err = intel_pt_get_next_packet(decoder);
2193 if (err)
2194 return err;
2195
2196 switch (decoder->packet.type) {
2197 case INTEL_PT_TNT:
2198 case INTEL_PT_FUP:
2199 case INTEL_PT_TRACESTOP:
2200 case INTEL_PT_PSB:
2201 case INTEL_PT_TSC:
2202 case INTEL_PT_TMA:
2203 case INTEL_PT_MODE_TSX:
2204 case INTEL_PT_BAD:
2205 case INTEL_PT_PSBEND:
2206 case INTEL_PT_PTWRITE:
2207 case INTEL_PT_PTWRITE_IP:
2208 case INTEL_PT_EXSTOP:
2209 case INTEL_PT_EXSTOP_IP:
2210 case INTEL_PT_MWAIT:
2211 case INTEL_PT_PWRE:
2212 case INTEL_PT_PWRX:
2213 case INTEL_PT_BBP:
2214 case INTEL_PT_BIP:
2215 case INTEL_PT_BEP:
2216 case INTEL_PT_BEP_IP:
2217 case INTEL_PT_CFE:
2218 case INTEL_PT_CFE_IP:
2219 case INTEL_PT_EVD:
2220 intel_pt_log("ERROR: Missing TIP after FUP\n");
2221 decoder->pkt_state = INTEL_PT_STATE_ERR3;
2222 decoder->pkt_step = 0;
2223 return -ENOENT;
2224
2225 case INTEL_PT_CBR:
2226 intel_pt_calc_cbr(decoder);
2227 break;
2228
2229 case INTEL_PT_OVF:
2230 return intel_pt_overflow(decoder);
2231
2232 case INTEL_PT_TIP_PGD:
2233 decoder->state.from_ip = decoder->ip;
2234 if (decoder->packet.count == 0) {
2235 decoder->state.to_ip = 0;
2236 } else {
2237 intel_pt_set_ip(decoder);
2238 decoder->state.to_ip = decoder->ip;
2239 }
2240 decoder->pge = false;
2241 decoder->continuous_period = false;
2242 decoder->state.type |= INTEL_PT_TRACE_END;
2243 intel_pt_update_nr(decoder);
2244 return 0;
2245
2246 case INTEL_PT_TIP_PGE:
2247 decoder->pge = true;
2248 intel_pt_log("Omitting PGE ip " x64_fmt "\n",
2249 decoder->ip);
2250 decoder->state.from_ip = 0;
2251 if (decoder->packet.count == 0) {
2252 decoder->state.to_ip = 0;
2253 } else {
2254 intel_pt_set_ip(decoder);
2255 decoder->state.to_ip = decoder->ip;
2256 }
2257 decoder->state.type |= INTEL_PT_TRACE_BEGIN;
2258 intel_pt_mtc_cyc_cnt_pge(decoder);
2259 intel_pt_set_nr(decoder);
2260 return 0;
2261
2262 case INTEL_PT_TIP:
2263 decoder->state.from_ip = decoder->ip;
2264 if (decoder->packet.count == 0) {
2265 decoder->state.to_ip = 0;
2266 } else {
2267 intel_pt_set_ip(decoder);
2268 decoder->state.to_ip = decoder->ip;
2269 }
2270 intel_pt_update_nr(decoder);
2271 intel_pt_sample_iflag_chg(decoder);
2272 return 0;
2273
2274 case INTEL_PT_PIP:
2275 intel_pt_update_pip(decoder);
2276 break;
2277
2278 case INTEL_PT_MTC:
2279 intel_pt_calc_mtc_timestamp(decoder);
2280 if (decoder->period_type == INTEL_PT_PERIOD_MTC)
2281 decoder->state.type |= INTEL_PT_INSTRUCTION;
2282 break;
2283
2284 case INTEL_PT_CYC:
2285 intel_pt_calc_cyc_timestamp(decoder);
2286 break;
2287
2288 case INTEL_PT_MODE_EXEC:
2289 intel_pt_mode_exec(decoder);
2290 break;
2291
2292 case INTEL_PT_VMCS:
2293 case INTEL_PT_MNT:
2294 case INTEL_PT_PAD:
2295 break;
2296
2297 default:
2298 return intel_pt_bug(decoder);
2299 }
2300 }
2301 }
2302
intel_pt_resample(struct intel_pt_decoder * decoder)2303 static int intel_pt_resample(struct intel_pt_decoder *decoder)
2304 {
2305 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2306 decoder->state.type = INTEL_PT_INSTRUCTION;
2307 decoder->state.from_ip = decoder->ip;
2308 decoder->state.to_ip = 0;
2309 return 0;
2310 }
2311
2312 struct intel_pt_vm_tsc_info {
2313 struct intel_pt_pkt pip_packet;
2314 struct intel_pt_pkt vmcs_packet;
2315 struct intel_pt_pkt tma_packet;
2316 bool tsc, pip, vmcs, tma, psbend;
2317 uint64_t ctc_delta;
2318 uint64_t last_ctc;
2319 int max_lookahead;
2320 };
2321
2322 /* Lookahead and get the PIP, VMCS and TMA packets from PSB+ */
intel_pt_vm_psb_lookahead_cb(struct intel_pt_pkt_info * pkt_info)2323 static int intel_pt_vm_psb_lookahead_cb(struct intel_pt_pkt_info *pkt_info)
2324 {
2325 struct intel_pt_vm_tsc_info *data = pkt_info->data;
2326
2327 switch (pkt_info->packet.type) {
2328 case INTEL_PT_PAD:
2329 case INTEL_PT_MNT:
2330 case INTEL_PT_MODE_EXEC:
2331 case INTEL_PT_MODE_TSX:
2332 case INTEL_PT_MTC:
2333 case INTEL_PT_FUP:
2334 case INTEL_PT_CYC:
2335 case INTEL_PT_CBR:
2336 break;
2337
2338 case INTEL_PT_TSC:
2339 data->tsc = true;
2340 break;
2341
2342 case INTEL_PT_TMA:
2343 data->tma_packet = pkt_info->packet;
2344 data->tma = true;
2345 break;
2346
2347 case INTEL_PT_PIP:
2348 data->pip_packet = pkt_info->packet;
2349 data->pip = true;
2350 break;
2351
2352 case INTEL_PT_VMCS:
2353 data->vmcs_packet = pkt_info->packet;
2354 data->vmcs = true;
2355 break;
2356
2357 case INTEL_PT_PSBEND:
2358 data->psbend = true;
2359 return 1;
2360
2361 case INTEL_PT_TIP_PGE:
2362 case INTEL_PT_PTWRITE:
2363 case INTEL_PT_PTWRITE_IP:
2364 case INTEL_PT_EXSTOP:
2365 case INTEL_PT_EXSTOP_IP:
2366 case INTEL_PT_MWAIT:
2367 case INTEL_PT_PWRE:
2368 case INTEL_PT_PWRX:
2369 case INTEL_PT_BBP:
2370 case INTEL_PT_BIP:
2371 case INTEL_PT_BEP:
2372 case INTEL_PT_BEP_IP:
2373 case INTEL_PT_OVF:
2374 case INTEL_PT_BAD:
2375 case INTEL_PT_TNT:
2376 case INTEL_PT_TIP_PGD:
2377 case INTEL_PT_TIP:
2378 case INTEL_PT_PSB:
2379 case INTEL_PT_TRACESTOP:
2380 case INTEL_PT_CFE:
2381 case INTEL_PT_CFE_IP:
2382 case INTEL_PT_EVD:
2383 default:
2384 return 1;
2385 }
2386
2387 return 0;
2388 }
2389
2390 struct intel_pt_ovf_fup_info {
2391 int max_lookahead;
2392 bool found;
2393 };
2394
2395 /* Lookahead to detect a FUP packet after OVF */
intel_pt_ovf_fup_lookahead_cb(struct intel_pt_pkt_info * pkt_info)2396 static int intel_pt_ovf_fup_lookahead_cb(struct intel_pt_pkt_info *pkt_info)
2397 {
2398 struct intel_pt_ovf_fup_info *data = pkt_info->data;
2399
2400 if (pkt_info->packet.type == INTEL_PT_CYC ||
2401 pkt_info->packet.type == INTEL_PT_MTC ||
2402 pkt_info->packet.type == INTEL_PT_TSC)
2403 return !--(data->max_lookahead);
2404 data->found = pkt_info->packet.type == INTEL_PT_FUP;
2405 return 1;
2406 }
2407
intel_pt_ovf_fup_lookahead(struct intel_pt_decoder * decoder)2408 static bool intel_pt_ovf_fup_lookahead(struct intel_pt_decoder *decoder)
2409 {
2410 struct intel_pt_ovf_fup_info data = {
2411 .max_lookahead = 16,
2412 .found = false,
2413 };
2414
2415 intel_pt_pkt_lookahead(decoder, intel_pt_ovf_fup_lookahead_cb, &data);
2416 return data.found;
2417 }
2418
2419 /* Lookahead and get the TMA packet after TSC */
intel_pt_tma_lookahead_cb(struct intel_pt_pkt_info * pkt_info)2420 static int intel_pt_tma_lookahead_cb(struct intel_pt_pkt_info *pkt_info)
2421 {
2422 struct intel_pt_vm_tsc_info *data = pkt_info->data;
2423
2424 if (pkt_info->packet.type == INTEL_PT_CYC ||
2425 pkt_info->packet.type == INTEL_PT_MTC)
2426 return !--(data->max_lookahead);
2427
2428 if (pkt_info->packet.type == INTEL_PT_TMA) {
2429 data->tma_packet = pkt_info->packet;
2430 data->tma = true;
2431 }
2432 return 1;
2433 }
2434
intel_pt_ctc_to_tsc(struct intel_pt_decoder * decoder,uint64_t ctc)2435 static uint64_t intel_pt_ctc_to_tsc(struct intel_pt_decoder *decoder, uint64_t ctc)
2436 {
2437 if (decoder->tsc_ctc_mult)
2438 return ctc * decoder->tsc_ctc_mult;
2439 else
2440 return multdiv(ctc, decoder->tsc_ctc_ratio_n, decoder->tsc_ctc_ratio_d);
2441 }
2442
intel_pt_calc_expected_tsc(struct intel_pt_decoder * decoder,uint32_t ctc,uint32_t fc,uint64_t last_ctc_timestamp,uint64_t ctc_delta,uint32_t last_ctc)2443 static uint64_t intel_pt_calc_expected_tsc(struct intel_pt_decoder *decoder,
2444 uint32_t ctc,
2445 uint32_t fc,
2446 uint64_t last_ctc_timestamp,
2447 uint64_t ctc_delta,
2448 uint32_t last_ctc)
2449 {
2450 /* Number of CTC ticks from last_ctc_timestamp to last_mtc */
2451 uint64_t last_mtc_ctc = last_ctc + ctc_delta;
2452 /*
2453 * Number of CTC ticks from there until current TMA packet. We would
2454 * expect last_mtc_ctc to be before ctc, but the TSC packet can slip
2455 * past an MTC, so a sign-extended value is used.
2456 */
2457 uint64_t delta = (int16_t)((uint16_t)ctc - (uint16_t)last_mtc_ctc);
2458 /* Total CTC ticks from last_ctc_timestamp to current TMA packet */
2459 uint64_t new_ctc_delta = ctc_delta + delta;
2460 uint64_t expected_tsc;
2461
2462 /*
2463 * Convert CTC ticks to TSC ticks, add the starting point
2464 * (last_ctc_timestamp) and the fast counter from the TMA packet.
2465 */
2466 expected_tsc = last_ctc_timestamp + intel_pt_ctc_to_tsc(decoder, new_ctc_delta) + fc;
2467
2468 if (intel_pt_enable_logging) {
2469 intel_pt_log_x64(last_mtc_ctc);
2470 intel_pt_log_x32(last_ctc);
2471 intel_pt_log_x64(ctc_delta);
2472 intel_pt_log_x64(delta);
2473 intel_pt_log_x32(ctc);
2474 intel_pt_log_x64(new_ctc_delta);
2475 intel_pt_log_x64(last_ctc_timestamp);
2476 intel_pt_log_x32(fc);
2477 intel_pt_log_x64(intel_pt_ctc_to_tsc(decoder, new_ctc_delta));
2478 intel_pt_log_x64(expected_tsc);
2479 }
2480
2481 return expected_tsc;
2482 }
2483
intel_pt_expected_tsc(struct intel_pt_decoder * decoder,struct intel_pt_vm_tsc_info * data)2484 static uint64_t intel_pt_expected_tsc(struct intel_pt_decoder *decoder,
2485 struct intel_pt_vm_tsc_info *data)
2486 {
2487 uint32_t ctc = data->tma_packet.payload;
2488 uint32_t fc = data->tma_packet.count;
2489
2490 return intel_pt_calc_expected_tsc(decoder, ctc, fc,
2491 decoder->ctc_timestamp,
2492 data->ctc_delta, data->last_ctc);
2493 }
2494
intel_pt_translate_vm_tsc(struct intel_pt_decoder * decoder,struct intel_pt_vmcs_info * vmcs_info)2495 static void intel_pt_translate_vm_tsc(struct intel_pt_decoder *decoder,
2496 struct intel_pt_vmcs_info *vmcs_info)
2497 {
2498 uint64_t payload = decoder->packet.payload;
2499
2500 /* VMX adds the TSC Offset, so subtract to get host TSC */
2501 decoder->packet.payload -= vmcs_info->tsc_offset;
2502 /* TSC packet has only 7 bytes */
2503 decoder->packet.payload &= SEVEN_BYTES;
2504
2505 /*
2506 * The buffer is mmapped from the data file, so this also updates the
2507 * data file.
2508 */
2509 if (!decoder->vm_tm_corr_dry_run)
2510 memcpy((void *)decoder->buf + 1, &decoder->packet.payload, 7);
2511
2512 intel_pt_log("Translated VM TSC %#" PRIx64 " -> %#" PRIx64
2513 " VMCS %#" PRIx64 " TSC Offset %#" PRIx64 "\n",
2514 payload, decoder->packet.payload, vmcs_info->vmcs,
2515 vmcs_info->tsc_offset);
2516 }
2517
intel_pt_translate_vm_tsc_offset(struct intel_pt_decoder * decoder,uint64_t tsc_offset)2518 static void intel_pt_translate_vm_tsc_offset(struct intel_pt_decoder *decoder,
2519 uint64_t tsc_offset)
2520 {
2521 struct intel_pt_vmcs_info vmcs_info = {
2522 .vmcs = NO_VMCS,
2523 .tsc_offset = tsc_offset
2524 };
2525
2526 intel_pt_translate_vm_tsc(decoder, &vmcs_info);
2527 }
2528
in_vm(uint64_t pip_payload)2529 static inline bool in_vm(uint64_t pip_payload)
2530 {
2531 return pip_payload & 1;
2532 }
2533
pip_in_vm(struct intel_pt_pkt * pip_packet)2534 static inline bool pip_in_vm(struct intel_pt_pkt *pip_packet)
2535 {
2536 return pip_packet->payload & 1;
2537 }
2538
intel_pt_print_vmcs_info(struct intel_pt_vmcs_info * vmcs_info)2539 static void intel_pt_print_vmcs_info(struct intel_pt_vmcs_info *vmcs_info)
2540 {
2541 p_log("VMCS: %#" PRIx64 " TSC Offset %#" PRIx64,
2542 vmcs_info->vmcs, vmcs_info->tsc_offset);
2543 }
2544
intel_pt_vm_tm_corr_psb(struct intel_pt_decoder * decoder,struct intel_pt_vm_tsc_info * data)2545 static void intel_pt_vm_tm_corr_psb(struct intel_pt_decoder *decoder,
2546 struct intel_pt_vm_tsc_info *data)
2547 {
2548 memset(data, 0, sizeof(*data));
2549 data->ctc_delta = decoder->ctc_delta;
2550 data->last_ctc = decoder->last_ctc;
2551 intel_pt_pkt_lookahead(decoder, intel_pt_vm_psb_lookahead_cb, data);
2552 if (data->tsc && !data->psbend)
2553 p_log("ERROR: PSB without PSBEND");
2554 decoder->in_psb = data->psbend;
2555 }
2556
intel_pt_vm_tm_corr_first_tsc(struct intel_pt_decoder * decoder,struct intel_pt_vm_tsc_info * data,struct intel_pt_vmcs_info * vmcs_info,uint64_t host_tsc)2557 static void intel_pt_vm_tm_corr_first_tsc(struct intel_pt_decoder *decoder,
2558 struct intel_pt_vm_tsc_info *data,
2559 struct intel_pt_vmcs_info *vmcs_info,
2560 uint64_t host_tsc)
2561 {
2562 if (!decoder->in_psb) {
2563 /* Can't happen */
2564 p_log("ERROR: First TSC is not in PSB+");
2565 }
2566
2567 if (data->pip) {
2568 if (pip_in_vm(&data->pip_packet)) { /* Guest */
2569 if (vmcs_info && vmcs_info->tsc_offset) {
2570 intel_pt_translate_vm_tsc(decoder, vmcs_info);
2571 decoder->vm_tm_corr_reliable = true;
2572 } else {
2573 p_log("ERROR: First TSC, unknown TSC Offset");
2574 }
2575 } else { /* Host */
2576 decoder->vm_tm_corr_reliable = true;
2577 }
2578 } else { /* Host or Guest */
2579 decoder->vm_tm_corr_reliable = false;
2580 if (intel_pt_time_in_range(decoder, host_tsc)) {
2581 /* Assume Host */
2582 } else {
2583 /* Assume Guest */
2584 if (vmcs_info && vmcs_info->tsc_offset)
2585 intel_pt_translate_vm_tsc(decoder, vmcs_info);
2586 else
2587 p_log("ERROR: First TSC, no PIP, unknown TSC Offset");
2588 }
2589 }
2590 }
2591
intel_pt_vm_tm_corr_tsc(struct intel_pt_decoder * decoder,struct intel_pt_vm_tsc_info * data)2592 static void intel_pt_vm_tm_corr_tsc(struct intel_pt_decoder *decoder,
2593 struct intel_pt_vm_tsc_info *data)
2594 {
2595 struct intel_pt_vmcs_info *vmcs_info;
2596 uint64_t tsc_offset = 0;
2597 uint64_t vmcs;
2598 bool reliable = true;
2599 uint64_t expected_tsc;
2600 uint64_t host_tsc;
2601 uint64_t ref_timestamp;
2602
2603 bool assign = false;
2604 bool assign_reliable = false;
2605
2606 /* Already have 'data' for the in_psb case */
2607 if (!decoder->in_psb) {
2608 memset(data, 0, sizeof(*data));
2609 data->ctc_delta = decoder->ctc_delta;
2610 data->last_ctc = decoder->last_ctc;
2611 data->max_lookahead = 16;
2612 intel_pt_pkt_lookahead(decoder, intel_pt_tma_lookahead_cb, data);
2613 if (decoder->pge) {
2614 data->pip = true;
2615 data->pip_packet.payload = decoder->pip_payload;
2616 }
2617 }
2618
2619 /* Calculations depend on having TMA packets */
2620 if (!data->tma) {
2621 p_log("ERROR: TSC without TMA");
2622 return;
2623 }
2624
2625 vmcs = data->vmcs ? data->vmcs_packet.payload : decoder->vmcs;
2626 if (vmcs == NO_VMCS)
2627 vmcs = 0;
2628
2629 vmcs_info = decoder->findnew_vmcs_info(decoder->data, vmcs);
2630
2631 ref_timestamp = decoder->timestamp ? decoder->timestamp : decoder->buf_timestamp;
2632 host_tsc = intel_pt_8b_tsc(decoder->packet.payload, ref_timestamp);
2633
2634 if (!decoder->ctc_timestamp) {
2635 intel_pt_vm_tm_corr_first_tsc(decoder, data, vmcs_info, host_tsc);
2636 return;
2637 }
2638
2639 expected_tsc = intel_pt_expected_tsc(decoder, data);
2640
2641 tsc_offset = host_tsc - expected_tsc;
2642
2643 /* Determine if TSC is from Host or Guest */
2644 if (data->pip) {
2645 if (pip_in_vm(&data->pip_packet)) { /* Guest */
2646 if (!vmcs_info) {
2647 /* PIP NR=1 without VMCS cannot happen */
2648 p_log("ERROR: Missing VMCS");
2649 intel_pt_translate_vm_tsc_offset(decoder, tsc_offset);
2650 decoder->vm_tm_corr_reliable = false;
2651 return;
2652 }
2653 } else { /* Host */
2654 decoder->last_reliable_timestamp = host_tsc;
2655 decoder->vm_tm_corr_reliable = true;
2656 return;
2657 }
2658 } else { /* Host or Guest */
2659 reliable = false; /* Host/Guest is a guess, so not reliable */
2660 if (decoder->in_psb) {
2661 if (!tsc_offset)
2662 return; /* Zero TSC Offset, assume Host */
2663 /*
2664 * TSC packet has only 7 bytes of TSC. We have no
2665 * information about the Guest's 8th byte, but it
2666 * doesn't matter because we only need 7 bytes.
2667 * Here, since the 8th byte is unreliable and
2668 * irrelevant, compare only 7 byes.
2669 */
2670 if (vmcs_info &&
2671 (tsc_offset & SEVEN_BYTES) ==
2672 (vmcs_info->tsc_offset & SEVEN_BYTES)) {
2673 /* Same TSC Offset as last VMCS, assume Guest */
2674 goto guest;
2675 }
2676 }
2677 /*
2678 * Check if the host_tsc is within the expected range.
2679 * Note, we could narrow the range more by looking ahead for
2680 * the next host TSC in the same buffer, but we don't bother to
2681 * do that because this is probably good enough.
2682 */
2683 if (host_tsc >= expected_tsc && intel_pt_time_in_range(decoder, host_tsc)) {
2684 /* Within expected range for Host TSC, assume Host */
2685 decoder->vm_tm_corr_reliable = false;
2686 return;
2687 }
2688 }
2689
2690 guest: /* Assuming Guest */
2691
2692 /* Determine whether to assign TSC Offset */
2693 if (vmcs_info && vmcs_info->vmcs) {
2694 if (vmcs_info->tsc_offset && vmcs_info->reliable) {
2695 assign = false;
2696 } else if (decoder->in_psb && data->pip && decoder->vm_tm_corr_reliable &&
2697 decoder->vm_tm_corr_continuous && decoder->vm_tm_corr_same_buf) {
2698 /* Continuous tracing, TSC in a PSB is not a time loss */
2699 assign = true;
2700 assign_reliable = true;
2701 } else if (decoder->in_psb && data->pip && decoder->vm_tm_corr_same_buf) {
2702 /*
2703 * Unlikely to be a time loss TSC in a PSB which is not
2704 * at the start of a buffer.
2705 */
2706 assign = true;
2707 assign_reliable = false;
2708 }
2709 }
2710
2711 /* Record VMCS TSC Offset */
2712 if (assign && (vmcs_info->tsc_offset != tsc_offset ||
2713 vmcs_info->reliable != assign_reliable)) {
2714 bool print = vmcs_info->tsc_offset != tsc_offset;
2715
2716 vmcs_info->tsc_offset = tsc_offset;
2717 vmcs_info->reliable = assign_reliable;
2718 if (print)
2719 intel_pt_print_vmcs_info(vmcs_info);
2720 }
2721
2722 /* Determine what TSC Offset to use */
2723 if (vmcs_info && vmcs_info->tsc_offset) {
2724 if (!vmcs_info->reliable)
2725 reliable = false;
2726 intel_pt_translate_vm_tsc(decoder, vmcs_info);
2727 } else {
2728 reliable = false;
2729 if (vmcs_info) {
2730 if (!vmcs_info->error_printed) {
2731 p_log("ERROR: Unknown TSC Offset for VMCS %#" PRIx64,
2732 vmcs_info->vmcs);
2733 vmcs_info->error_printed = true;
2734 }
2735 } else {
2736 if (intel_pt_print_once(decoder, INTEL_PT_PRT_ONCE_UNK_VMCS))
2737 p_log("ERROR: Unknown VMCS");
2738 }
2739 intel_pt_translate_vm_tsc_offset(decoder, tsc_offset);
2740 }
2741
2742 decoder->vm_tm_corr_reliable = reliable;
2743 }
2744
intel_pt_vm_tm_corr_pebs_tsc(struct intel_pt_decoder * decoder)2745 static void intel_pt_vm_tm_corr_pebs_tsc(struct intel_pt_decoder *decoder)
2746 {
2747 uint64_t host_tsc = decoder->packet.payload;
2748 uint64_t guest_tsc = decoder->packet.payload;
2749 struct intel_pt_vmcs_info *vmcs_info;
2750 uint64_t vmcs;
2751
2752 vmcs = decoder->vmcs;
2753 if (vmcs == NO_VMCS)
2754 vmcs = 0;
2755
2756 vmcs_info = decoder->findnew_vmcs_info(decoder->data, vmcs);
2757
2758 if (decoder->pge) {
2759 if (in_vm(decoder->pip_payload)) { /* Guest */
2760 if (!vmcs_info) {
2761 /* PIP NR=1 without VMCS cannot happen */
2762 p_log("ERROR: Missing VMCS");
2763 }
2764 } else { /* Host */
2765 return;
2766 }
2767 } else { /* Host or Guest */
2768 if (intel_pt_time_in_range(decoder, host_tsc)) {
2769 /* Within expected range for Host TSC, assume Host */
2770 return;
2771 }
2772 }
2773
2774 if (vmcs_info) {
2775 /* Translate Guest TSC to Host TSC */
2776 host_tsc = ((guest_tsc & SEVEN_BYTES) - vmcs_info->tsc_offset) & SEVEN_BYTES;
2777 host_tsc = intel_pt_8b_tsc(host_tsc, decoder->timestamp);
2778 intel_pt_log("Translated VM TSC %#" PRIx64 " -> %#" PRIx64
2779 " VMCS %#" PRIx64 " TSC Offset %#" PRIx64 "\n",
2780 guest_tsc, host_tsc, vmcs_info->vmcs,
2781 vmcs_info->tsc_offset);
2782 if (!intel_pt_time_in_range(decoder, host_tsc) &&
2783 intel_pt_print_once(decoder, INTEL_PT_PRT_ONCE_ERANGE))
2784 p_log("Timestamp out of range");
2785 } else {
2786 if (intel_pt_print_once(decoder, INTEL_PT_PRT_ONCE_UNK_VMCS))
2787 p_log("ERROR: Unknown VMCS");
2788 host_tsc = decoder->timestamp;
2789 }
2790
2791 decoder->packet.payload = host_tsc;
2792
2793 if (!decoder->vm_tm_corr_dry_run)
2794 memcpy((void *)decoder->buf + 1, &host_tsc, 8);
2795 }
2796
intel_pt_vm_time_correlation(struct intel_pt_decoder * decoder)2797 static int intel_pt_vm_time_correlation(struct intel_pt_decoder *decoder)
2798 {
2799 struct intel_pt_vm_tsc_info data = { .psbend = false };
2800 bool pge;
2801 int err;
2802
2803 if (decoder->in_psb)
2804 intel_pt_vm_tm_corr_psb(decoder, &data);
2805
2806 while (1) {
2807 err = intel_pt_get_next_packet(decoder);
2808 if (err == -ENOLINK)
2809 continue;
2810 if (err)
2811 break;
2812
2813 switch (decoder->packet.type) {
2814 case INTEL_PT_TIP_PGD:
2815 decoder->pge = false;
2816 decoder->vm_tm_corr_continuous = false;
2817 break;
2818
2819 case INTEL_PT_TNT:
2820 case INTEL_PT_TIP:
2821 case INTEL_PT_TIP_PGE:
2822 decoder->pge = true;
2823 break;
2824
2825 case INTEL_PT_OVF:
2826 decoder->in_psb = false;
2827 pge = decoder->pge;
2828 decoder->pge = intel_pt_ovf_fup_lookahead(decoder);
2829 if (pge != decoder->pge)
2830 intel_pt_log("Surprising PGE change in OVF!");
2831 if (!decoder->pge)
2832 decoder->vm_tm_corr_continuous = false;
2833 break;
2834
2835 case INTEL_PT_FUP:
2836 if (decoder->in_psb)
2837 decoder->pge = true;
2838 break;
2839
2840 case INTEL_PT_TRACESTOP:
2841 decoder->pge = false;
2842 decoder->vm_tm_corr_continuous = false;
2843 decoder->have_tma = false;
2844 break;
2845
2846 case INTEL_PT_PSB:
2847 intel_pt_vm_tm_corr_psb(decoder, &data);
2848 break;
2849
2850 case INTEL_PT_PIP:
2851 decoder->pip_payload = decoder->packet.payload;
2852 break;
2853
2854 case INTEL_PT_MTC:
2855 intel_pt_calc_mtc_timestamp(decoder);
2856 break;
2857
2858 case INTEL_PT_TSC:
2859 intel_pt_vm_tm_corr_tsc(decoder, &data);
2860 intel_pt_calc_tsc_timestamp(decoder);
2861 decoder->vm_tm_corr_same_buf = true;
2862 decoder->vm_tm_corr_continuous = decoder->pge;
2863 break;
2864
2865 case INTEL_PT_TMA:
2866 intel_pt_calc_tma(decoder);
2867 break;
2868
2869 case INTEL_PT_CYC:
2870 intel_pt_calc_cyc_timestamp(decoder);
2871 break;
2872
2873 case INTEL_PT_CBR:
2874 intel_pt_calc_cbr(decoder);
2875 break;
2876
2877 case INTEL_PT_PSBEND:
2878 decoder->in_psb = false;
2879 data.psbend = false;
2880 break;
2881
2882 case INTEL_PT_VMCS:
2883 if (decoder->packet.payload != NO_VMCS)
2884 decoder->vmcs = decoder->packet.payload;
2885 break;
2886
2887 case INTEL_PT_BBP:
2888 decoder->blk_type = decoder->packet.payload;
2889 break;
2890
2891 case INTEL_PT_BIP:
2892 if (decoder->blk_type == INTEL_PT_PEBS_BASIC &&
2893 decoder->packet.count == 2)
2894 intel_pt_vm_tm_corr_pebs_tsc(decoder);
2895 break;
2896
2897 case INTEL_PT_BEP:
2898 case INTEL_PT_BEP_IP:
2899 decoder->blk_type = 0;
2900 break;
2901
2902 case INTEL_PT_CFE:
2903 case INTEL_PT_CFE_IP:
2904 case INTEL_PT_EVD:
2905 case INTEL_PT_MODE_EXEC:
2906 case INTEL_PT_MODE_TSX:
2907 case INTEL_PT_MNT:
2908 case INTEL_PT_PAD:
2909 case INTEL_PT_PTWRITE_IP:
2910 case INTEL_PT_PTWRITE:
2911 case INTEL_PT_MWAIT:
2912 case INTEL_PT_PWRE:
2913 case INTEL_PT_EXSTOP_IP:
2914 case INTEL_PT_EXSTOP:
2915 case INTEL_PT_PWRX:
2916 case INTEL_PT_BAD: /* Does not happen */
2917 default:
2918 break;
2919 }
2920 }
2921
2922 return err;
2923 }
2924
2925 #define HOP_PROCESS 0
2926 #define HOP_IGNORE 1
2927 #define HOP_RETURN 2
2928 #define HOP_AGAIN 3
2929
2930 static int intel_pt_scan_for_psb(struct intel_pt_decoder *decoder);
2931
2932 /* 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)2933 static int intel_pt_hop_trace(struct intel_pt_decoder *decoder, bool *no_tip, int *err)
2934 {
2935 *err = 0;
2936
2937 /* Leap from PSB to PSB, getting ip from FUP within PSB+ */
2938 if (decoder->leap && !decoder->in_psb && decoder->packet.type != INTEL_PT_PSB) {
2939 *err = intel_pt_scan_for_psb(decoder);
2940 if (*err)
2941 return HOP_RETURN;
2942 }
2943
2944 switch (decoder->packet.type) {
2945 case INTEL_PT_TNT:
2946 return HOP_IGNORE;
2947
2948 case INTEL_PT_TIP_PGD:
2949 decoder->pge = false;
2950 if (!decoder->packet.count) {
2951 intel_pt_set_nr(decoder);
2952 return HOP_IGNORE;
2953 }
2954 intel_pt_set_ip(decoder);
2955 decoder->state.type |= INTEL_PT_TRACE_END;
2956 decoder->state.from_ip = 0;
2957 decoder->state.to_ip = decoder->ip;
2958 intel_pt_update_nr(decoder);
2959 return HOP_RETURN;
2960
2961 case INTEL_PT_TIP:
2962 if (!decoder->packet.count) {
2963 intel_pt_set_nr(decoder);
2964 return HOP_IGNORE;
2965 }
2966 intel_pt_set_ip(decoder);
2967 decoder->state.type = INTEL_PT_INSTRUCTION;
2968 decoder->state.from_ip = decoder->ip;
2969 decoder->state.to_ip = 0;
2970 intel_pt_update_nr(decoder);
2971 intel_pt_sample_iflag_chg(decoder);
2972 return HOP_RETURN;
2973
2974 case INTEL_PT_FUP:
2975 if (!decoder->packet.count)
2976 return HOP_IGNORE;
2977 intel_pt_set_ip(decoder);
2978 if (decoder->set_fup_mwait || decoder->set_fup_pwre)
2979 *no_tip = true;
2980 if (!decoder->branch_enable || !decoder->pge)
2981 *no_tip = true;
2982 if (*no_tip) {
2983 decoder->state.type = INTEL_PT_INSTRUCTION;
2984 decoder->state.from_ip = decoder->ip;
2985 decoder->state.to_ip = 0;
2986 intel_pt_fup_event(decoder, *no_tip);
2987 return HOP_RETURN;
2988 }
2989 intel_pt_fup_event(decoder, *no_tip);
2990 decoder->state.type |= INTEL_PT_INSTRUCTION | INTEL_PT_BRANCH;
2991 *err = intel_pt_walk_fup_tip(decoder);
2992 if (!*err && decoder->state.to_ip)
2993 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
2994 return HOP_RETURN;
2995
2996 case INTEL_PT_PSB:
2997 decoder->state.psb_offset = decoder->pos;
2998 decoder->psb_ip = 0;
2999 decoder->last_ip = 0;
3000 decoder->have_last_ip = true;
3001 *err = intel_pt_walk_psbend(decoder);
3002 if (*err == -EAGAIN)
3003 return HOP_AGAIN;
3004 if (*err)
3005 return HOP_RETURN;
3006 decoder->state.type = INTEL_PT_PSB_EVT;
3007 if (decoder->psb_ip) {
3008 decoder->state.type |= INTEL_PT_INSTRUCTION;
3009 decoder->ip = decoder->psb_ip;
3010 }
3011 decoder->state.from_ip = decoder->psb_ip;
3012 decoder->state.to_ip = 0;
3013 return HOP_RETURN;
3014
3015 case INTEL_PT_BAD:
3016 case INTEL_PT_PAD:
3017 case INTEL_PT_TIP_PGE:
3018 case INTEL_PT_TSC:
3019 case INTEL_PT_TMA:
3020 case INTEL_PT_MODE_EXEC:
3021 case INTEL_PT_MODE_TSX:
3022 case INTEL_PT_MTC:
3023 case INTEL_PT_CYC:
3024 case INTEL_PT_VMCS:
3025 case INTEL_PT_PSBEND:
3026 case INTEL_PT_CBR:
3027 case INTEL_PT_TRACESTOP:
3028 case INTEL_PT_PIP:
3029 case INTEL_PT_OVF:
3030 case INTEL_PT_MNT:
3031 case INTEL_PT_PTWRITE:
3032 case INTEL_PT_PTWRITE_IP:
3033 case INTEL_PT_EXSTOP:
3034 case INTEL_PT_EXSTOP_IP:
3035 case INTEL_PT_MWAIT:
3036 case INTEL_PT_PWRE:
3037 case INTEL_PT_PWRX:
3038 case INTEL_PT_BBP:
3039 case INTEL_PT_BIP:
3040 case INTEL_PT_BEP:
3041 case INTEL_PT_BEP_IP:
3042 case INTEL_PT_CFE:
3043 case INTEL_PT_CFE_IP:
3044 case INTEL_PT_EVD:
3045 default:
3046 return HOP_PROCESS;
3047 }
3048 }
3049
3050 struct intel_pt_psb_info {
3051 struct intel_pt_pkt fup_packet;
3052 bool fup;
3053 int after_psbend;
3054 };
3055
3056 /* Lookahead and get the FUP packet from PSB+ */
intel_pt_psb_lookahead_cb(struct intel_pt_pkt_info * pkt_info)3057 static int intel_pt_psb_lookahead_cb(struct intel_pt_pkt_info *pkt_info)
3058 {
3059 struct intel_pt_psb_info *data = pkt_info->data;
3060
3061 switch (pkt_info->packet.type) {
3062 case INTEL_PT_PAD:
3063 case INTEL_PT_MNT:
3064 case INTEL_PT_TSC:
3065 case INTEL_PT_TMA:
3066 case INTEL_PT_MODE_EXEC:
3067 case INTEL_PT_MODE_TSX:
3068 case INTEL_PT_MTC:
3069 case INTEL_PT_CYC:
3070 case INTEL_PT_VMCS:
3071 case INTEL_PT_CBR:
3072 case INTEL_PT_PIP:
3073 if (data->after_psbend) {
3074 data->after_psbend -= 1;
3075 if (!data->after_psbend)
3076 return 1;
3077 }
3078 break;
3079
3080 case INTEL_PT_FUP:
3081 if (data->after_psbend)
3082 return 1;
3083 if (data->fup || pkt_info->packet.count == 0)
3084 return 1;
3085 data->fup_packet = pkt_info->packet;
3086 data->fup = true;
3087 break;
3088
3089 case INTEL_PT_PSBEND:
3090 if (!data->fup)
3091 return 1;
3092 /* Keep going to check for a TIP.PGE */
3093 data->after_psbend = 6;
3094 break;
3095
3096 case INTEL_PT_TIP_PGE:
3097 /* Ignore FUP in PSB+ if followed by TIP.PGE */
3098 if (data->after_psbend)
3099 data->fup = false;
3100 return 1;
3101
3102 case INTEL_PT_PTWRITE:
3103 case INTEL_PT_PTWRITE_IP:
3104 case INTEL_PT_EXSTOP:
3105 case INTEL_PT_EXSTOP_IP:
3106 case INTEL_PT_MWAIT:
3107 case INTEL_PT_PWRE:
3108 case INTEL_PT_PWRX:
3109 case INTEL_PT_BBP:
3110 case INTEL_PT_BIP:
3111 case INTEL_PT_BEP:
3112 case INTEL_PT_BEP_IP:
3113 case INTEL_PT_CFE:
3114 case INTEL_PT_CFE_IP:
3115 case INTEL_PT_EVD:
3116 if (data->after_psbend) {
3117 data->after_psbend -= 1;
3118 if (!data->after_psbend)
3119 return 1;
3120 break;
3121 }
3122 return 1;
3123
3124 case INTEL_PT_OVF:
3125 case INTEL_PT_BAD:
3126 case INTEL_PT_TNT:
3127 case INTEL_PT_TIP_PGD:
3128 case INTEL_PT_TIP:
3129 case INTEL_PT_PSB:
3130 case INTEL_PT_TRACESTOP:
3131 default:
3132 return 1;
3133 }
3134
3135 return 0;
3136 }
3137
intel_pt_psb(struct intel_pt_decoder * decoder)3138 static int intel_pt_psb(struct intel_pt_decoder *decoder)
3139 {
3140 int err;
3141
3142 decoder->last_ip = 0;
3143 decoder->psb_ip = 0;
3144 decoder->have_last_ip = true;
3145 intel_pt_clear_stack(&decoder->stack);
3146 err = intel_pt_walk_psbend(decoder);
3147 if (err)
3148 return err;
3149 decoder->state.type = INTEL_PT_PSB_EVT;
3150 decoder->state.from_ip = decoder->psb_ip;
3151 decoder->state.to_ip = 0;
3152 return 0;
3153 }
3154
intel_pt_fup_in_psb(struct intel_pt_decoder * decoder)3155 static int intel_pt_fup_in_psb(struct intel_pt_decoder *decoder)
3156 {
3157 int err;
3158
3159 if (decoder->ip != decoder->last_ip) {
3160 err = intel_pt_walk_fup(decoder);
3161 if (!err || err != -EAGAIN)
3162 return err;
3163 }
3164
3165 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
3166 err = intel_pt_psb(decoder);
3167 if (err) {
3168 decoder->pkt_state = INTEL_PT_STATE_ERR3;
3169 return -ENOENT;
3170 }
3171
3172 return 0;
3173 }
3174
intel_pt_psb_with_fup(struct intel_pt_decoder * decoder,int * err)3175 static bool intel_pt_psb_with_fup(struct intel_pt_decoder *decoder, int *err)
3176 {
3177 struct intel_pt_psb_info data = { .fup = false };
3178
3179 if (!decoder->branch_enable)
3180 return false;
3181
3182 intel_pt_pkt_lookahead(decoder, intel_pt_psb_lookahead_cb, &data);
3183 if (!data.fup)
3184 return false;
3185
3186 decoder->packet = data.fup_packet;
3187 intel_pt_set_last_ip(decoder);
3188 decoder->pkt_state = INTEL_PT_STATE_FUP_IN_PSB;
3189
3190 *err = intel_pt_fup_in_psb(decoder);
3191
3192 return true;
3193 }
3194
intel_pt_walk_trace(struct intel_pt_decoder * decoder)3195 static int intel_pt_walk_trace(struct intel_pt_decoder *decoder)
3196 {
3197 int last_packet_type = INTEL_PT_PAD;
3198 bool no_tip = false;
3199 int err;
3200
3201 while (1) {
3202 err = intel_pt_get_next_packet(decoder);
3203 if (err)
3204 return err;
3205 next:
3206 err = 0;
3207 if (decoder->cyc_threshold) {
3208 if (decoder->sample_cyc && last_packet_type != INTEL_PT_CYC)
3209 decoder->sample_cyc = false;
3210 last_packet_type = decoder->packet.type;
3211 }
3212
3213 if (decoder->hop) {
3214 switch (intel_pt_hop_trace(decoder, &no_tip, &err)) {
3215 case HOP_IGNORE:
3216 continue;
3217 case HOP_RETURN:
3218 return err;
3219 case HOP_AGAIN:
3220 goto next;
3221 default:
3222 break;
3223 }
3224 }
3225
3226 switch (decoder->packet.type) {
3227 case INTEL_PT_TNT:
3228 if (!decoder->packet.count)
3229 break;
3230 decoder->tnt = decoder->packet;
3231 decoder->pkt_state = INTEL_PT_STATE_TNT;
3232 err = intel_pt_walk_tnt(decoder);
3233 if (err == -EAGAIN)
3234 break;
3235 return err;
3236
3237 case INTEL_PT_TIP_PGD:
3238 if (decoder->packet.count != 0)
3239 intel_pt_set_last_ip(decoder);
3240 decoder->pkt_state = INTEL_PT_STATE_TIP_PGD;
3241 return intel_pt_walk_tip(decoder);
3242
3243 case INTEL_PT_TIP_PGE: {
3244 decoder->pge = true;
3245 decoder->overflow = false;
3246 intel_pt_mtc_cyc_cnt_pge(decoder);
3247 intel_pt_set_nr(decoder);
3248 if (decoder->packet.count == 0) {
3249 intel_pt_log_at("Skipping zero TIP.PGE",
3250 decoder->pos);
3251 break;
3252 }
3253 intel_pt_sample_iflag_chg(decoder);
3254 intel_pt_set_ip(decoder);
3255 decoder->state.from_ip = 0;
3256 decoder->state.to_ip = decoder->ip;
3257 decoder->state.type |= INTEL_PT_TRACE_BEGIN;
3258 /*
3259 * In hop mode, resample to get the to_ip as an
3260 * "instruction" sample.
3261 */
3262 if (decoder->hop)
3263 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
3264 return 0;
3265 }
3266
3267 case INTEL_PT_OVF:
3268 return intel_pt_overflow(decoder);
3269
3270 case INTEL_PT_TIP:
3271 if (decoder->packet.count != 0)
3272 intel_pt_set_last_ip(decoder);
3273 decoder->pkt_state = INTEL_PT_STATE_TIP;
3274 return intel_pt_walk_tip(decoder);
3275
3276 case INTEL_PT_FUP:
3277 if (decoder->packet.count == 0) {
3278 intel_pt_log_at("Skipping zero FUP",
3279 decoder->pos);
3280 no_tip = false;
3281 break;
3282 }
3283 intel_pt_set_last_ip(decoder);
3284 if (!decoder->branch_enable || !decoder->pge) {
3285 decoder->ip = decoder->last_ip;
3286 if (intel_pt_fup_event(decoder, no_tip))
3287 return 0;
3288 no_tip = false;
3289 break;
3290 }
3291 if (decoder->set_fup_mwait)
3292 no_tip = true;
3293 if (no_tip)
3294 decoder->pkt_state = INTEL_PT_STATE_FUP_NO_TIP;
3295 else
3296 decoder->pkt_state = INTEL_PT_STATE_FUP;
3297 err = intel_pt_walk_fup(decoder);
3298 if (err != -EAGAIN)
3299 return err;
3300 if (no_tip) {
3301 no_tip = false;
3302 break;
3303 }
3304 return intel_pt_walk_fup_tip(decoder);
3305
3306 case INTEL_PT_TRACESTOP:
3307 decoder->pge = false;
3308 decoder->continuous_period = false;
3309 intel_pt_clear_tx_flags(decoder);
3310 decoder->have_tma = false;
3311 break;
3312
3313 case INTEL_PT_PSB:
3314 decoder->state.psb_offset = decoder->pos;
3315 decoder->psb_ip = 0;
3316 if (intel_pt_psb_with_fup(decoder, &err))
3317 return err;
3318 err = intel_pt_psb(decoder);
3319 if (err == -EAGAIN)
3320 goto next;
3321 return err;
3322
3323 case INTEL_PT_PIP:
3324 intel_pt_update_pip(decoder);
3325 break;
3326
3327 case INTEL_PT_MTC:
3328 intel_pt_calc_mtc_timestamp(decoder);
3329 if (decoder->period_type != INTEL_PT_PERIOD_MTC)
3330 break;
3331 /*
3332 * Ensure that there has been an instruction since the
3333 * last MTC.
3334 */
3335 if (!decoder->mtc_insn)
3336 break;
3337 decoder->mtc_insn = false;
3338 /* Ensure that there is a timestamp */
3339 if (!decoder->timestamp)
3340 break;
3341 decoder->state.type = INTEL_PT_INSTRUCTION;
3342 decoder->state.from_ip = decoder->ip;
3343 decoder->state.to_ip = 0;
3344 decoder->mtc_insn = false;
3345 return 0;
3346
3347 case INTEL_PT_TSC:
3348 intel_pt_calc_tsc_timestamp(decoder);
3349 break;
3350
3351 case INTEL_PT_TMA:
3352 intel_pt_calc_tma(decoder);
3353 break;
3354
3355 case INTEL_PT_CYC:
3356 intel_pt_calc_cyc_timestamp(decoder);
3357 break;
3358
3359 case INTEL_PT_CBR:
3360 intel_pt_calc_cbr(decoder);
3361 if (decoder->cbr != decoder->cbr_seen) {
3362 decoder->state.type = 0;
3363 return 0;
3364 }
3365 break;
3366
3367 case INTEL_PT_MODE_EXEC:
3368 intel_pt_mode_exec(decoder);
3369 err = intel_pt_get_next_packet(decoder);
3370 if (err)
3371 return err;
3372 if (decoder->packet.type == INTEL_PT_FUP) {
3373 decoder->set_fup_mode_exec = true;
3374 no_tip = true;
3375 }
3376 goto next;
3377
3378 case INTEL_PT_MODE_TSX:
3379 /* MODE_TSX need not be followed by FUP */
3380 if (!decoder->pge || decoder->in_psb) {
3381 intel_pt_update_in_tx(decoder);
3382 break;
3383 }
3384 err = intel_pt_mode_tsx(decoder, &no_tip);
3385 if (err)
3386 return err;
3387 goto next;
3388
3389 case INTEL_PT_BAD: /* Does not happen */
3390 return intel_pt_bug(decoder);
3391
3392 case INTEL_PT_PSBEND:
3393 case INTEL_PT_VMCS:
3394 case INTEL_PT_MNT:
3395 case INTEL_PT_PAD:
3396 break;
3397
3398 case INTEL_PT_PTWRITE_IP:
3399 decoder->fup_ptw_payload = decoder->packet.payload;
3400 err = intel_pt_get_next_packet(decoder);
3401 if (err)
3402 return err;
3403 if (decoder->packet.type == INTEL_PT_FUP) {
3404 decoder->set_fup_ptw = true;
3405 no_tip = true;
3406 } else {
3407 intel_pt_log_at("ERROR: Missing FUP after PTWRITE",
3408 decoder->pos);
3409 }
3410 goto next;
3411
3412 case INTEL_PT_PTWRITE:
3413 decoder->state.type = INTEL_PT_PTW;
3414 decoder->state.from_ip = decoder->ip;
3415 decoder->state.to_ip = 0;
3416 decoder->state.ptw_payload = decoder->packet.payload;
3417 return 0;
3418
3419 case INTEL_PT_MWAIT:
3420 decoder->fup_mwait_payload = decoder->packet.payload;
3421 decoder->set_fup_mwait = true;
3422 break;
3423
3424 case INTEL_PT_PWRE:
3425 if (decoder->set_fup_mwait) {
3426 decoder->fup_pwre_payload =
3427 decoder->packet.payload;
3428 decoder->set_fup_pwre = true;
3429 break;
3430 }
3431 decoder->state.type = INTEL_PT_PWR_ENTRY;
3432 decoder->state.from_ip = decoder->ip;
3433 decoder->state.to_ip = 0;
3434 decoder->state.pwrx_payload = decoder->packet.payload;
3435 return 0;
3436
3437 case INTEL_PT_EXSTOP_IP:
3438 err = intel_pt_get_next_packet(decoder);
3439 if (err)
3440 return err;
3441 if (decoder->packet.type == INTEL_PT_FUP) {
3442 decoder->set_fup_exstop = true;
3443 no_tip = true;
3444 } else {
3445 intel_pt_log_at("ERROR: Missing FUP after EXSTOP",
3446 decoder->pos);
3447 }
3448 goto next;
3449
3450 case INTEL_PT_EXSTOP:
3451 decoder->state.type = INTEL_PT_EX_STOP;
3452 decoder->state.from_ip = decoder->ip;
3453 decoder->state.to_ip = 0;
3454 return 0;
3455
3456 case INTEL_PT_PWRX:
3457 decoder->state.type = INTEL_PT_PWR_EXIT;
3458 decoder->state.from_ip = decoder->ip;
3459 decoder->state.to_ip = 0;
3460 decoder->state.pwrx_payload = decoder->packet.payload;
3461 return 0;
3462
3463 case INTEL_PT_BBP:
3464 intel_pt_bbp(decoder);
3465 break;
3466
3467 case INTEL_PT_BIP:
3468 intel_pt_bip(decoder);
3469 break;
3470
3471 case INTEL_PT_BEP:
3472 decoder->state.type = INTEL_PT_BLK_ITEMS;
3473 decoder->state.from_ip = decoder->ip;
3474 decoder->state.to_ip = 0;
3475 return 0;
3476
3477 case INTEL_PT_BEP_IP:
3478 err = intel_pt_get_next_packet(decoder);
3479 if (err)
3480 return err;
3481 if (decoder->packet.type == INTEL_PT_FUP) {
3482 decoder->set_fup_bep = true;
3483 no_tip = true;
3484 } else {
3485 intel_pt_log_at("ERROR: Missing FUP after BEP",
3486 decoder->pos);
3487 }
3488 goto next;
3489
3490 case INTEL_PT_CFE:
3491 decoder->fup_cfe_pkt = decoder->packet;
3492 decoder->set_fup_cfe = true;
3493 if (!decoder->pge) {
3494 intel_pt_fup_event(decoder, true);
3495 return 0;
3496 }
3497 break;
3498
3499 case INTEL_PT_CFE_IP:
3500 decoder->fup_cfe_pkt = decoder->packet;
3501 err = intel_pt_get_next_packet(decoder);
3502 if (err)
3503 return err;
3504 if (decoder->packet.type == INTEL_PT_FUP) {
3505 decoder->set_fup_cfe_ip = true;
3506 no_tip = true;
3507 } else {
3508 intel_pt_log_at("ERROR: Missing FUP after CFE",
3509 decoder->pos);
3510 }
3511 goto next;
3512
3513 case INTEL_PT_EVD:
3514 err = intel_pt_evd(decoder);
3515 if (err)
3516 return err;
3517 break;
3518
3519 default:
3520 return intel_pt_bug(decoder);
3521 }
3522 }
3523 }
3524
intel_pt_have_ip(struct intel_pt_decoder * decoder)3525 static inline bool intel_pt_have_ip(struct intel_pt_decoder *decoder)
3526 {
3527 return decoder->packet.count &&
3528 (decoder->have_last_ip || decoder->packet.count == 3 ||
3529 decoder->packet.count == 6);
3530 }
3531
3532 /* Walk PSB+ packets to get in sync. */
intel_pt_walk_psb(struct intel_pt_decoder * decoder)3533 static int intel_pt_walk_psb(struct intel_pt_decoder *decoder)
3534 {
3535 int err;
3536
3537 decoder->in_psb = true;
3538
3539 while (1) {
3540 err = intel_pt_get_next_packet(decoder);
3541 if (err)
3542 goto out;
3543
3544 switch (decoder->packet.type) {
3545 case INTEL_PT_TIP_PGD:
3546 decoder->continuous_period = false;
3547 __fallthrough;
3548 case INTEL_PT_TIP_PGE:
3549 case INTEL_PT_TIP:
3550 case INTEL_PT_PTWRITE:
3551 case INTEL_PT_PTWRITE_IP:
3552 case INTEL_PT_EXSTOP:
3553 case INTEL_PT_EXSTOP_IP:
3554 case INTEL_PT_MWAIT:
3555 case INTEL_PT_PWRE:
3556 case INTEL_PT_PWRX:
3557 case INTEL_PT_BBP:
3558 case INTEL_PT_BIP:
3559 case INTEL_PT_BEP:
3560 case INTEL_PT_BEP_IP:
3561 case INTEL_PT_CFE:
3562 case INTEL_PT_CFE_IP:
3563 case INTEL_PT_EVD:
3564 intel_pt_log("ERROR: Unexpected packet\n");
3565 err = -ENOENT;
3566 goto out;
3567
3568 case INTEL_PT_FUP:
3569 decoder->pge = true;
3570 if (intel_pt_have_ip(decoder)) {
3571 uint64_t current_ip = decoder->ip;
3572
3573 intel_pt_set_ip(decoder);
3574 decoder->psb_ip = decoder->ip;
3575 if (current_ip)
3576 intel_pt_log_to("Setting IP",
3577 decoder->ip);
3578 }
3579 break;
3580
3581 case INTEL_PT_MTC:
3582 intel_pt_calc_mtc_timestamp(decoder);
3583 break;
3584
3585 case INTEL_PT_TSC:
3586 intel_pt_calc_tsc_timestamp(decoder);
3587 break;
3588
3589 case INTEL_PT_TMA:
3590 intel_pt_calc_tma(decoder);
3591 break;
3592
3593 case INTEL_PT_CYC:
3594 intel_pt_calc_cyc_timestamp(decoder);
3595 break;
3596
3597 case INTEL_PT_CBR:
3598 intel_pt_calc_cbr(decoder);
3599 break;
3600
3601 case INTEL_PT_PIP:
3602 intel_pt_set_pip(decoder);
3603 break;
3604
3605 case INTEL_PT_MODE_EXEC:
3606 intel_pt_mode_exec_status(decoder);
3607 break;
3608
3609 case INTEL_PT_MODE_TSX:
3610 intel_pt_update_in_tx(decoder);
3611 break;
3612
3613 case INTEL_PT_TRACESTOP:
3614 decoder->pge = false;
3615 decoder->continuous_period = false;
3616 intel_pt_clear_tx_flags(decoder);
3617 __fallthrough;
3618
3619 case INTEL_PT_TNT:
3620 decoder->have_tma = false;
3621 intel_pt_log("ERROR: Unexpected packet\n");
3622 if (decoder->ip)
3623 decoder->pkt_state = INTEL_PT_STATE_ERR4;
3624 else
3625 decoder->pkt_state = INTEL_PT_STATE_ERR3;
3626 err = -ENOENT;
3627 goto out;
3628
3629 case INTEL_PT_BAD: /* Does not happen */
3630 err = intel_pt_bug(decoder);
3631 goto out;
3632
3633 case INTEL_PT_OVF:
3634 err = intel_pt_overflow(decoder);
3635 goto out;
3636
3637 case INTEL_PT_PSBEND:
3638 err = 0;
3639 goto out;
3640
3641 case INTEL_PT_PSB:
3642 case INTEL_PT_VMCS:
3643 case INTEL_PT_MNT:
3644 case INTEL_PT_PAD:
3645 default:
3646 break;
3647 }
3648 }
3649 out:
3650 decoder->in_psb = false;
3651
3652 return err;
3653 }
3654
intel_pt_walk_to_ip(struct intel_pt_decoder * decoder)3655 static int intel_pt_walk_to_ip(struct intel_pt_decoder *decoder)
3656 {
3657 int err;
3658
3659 while (1) {
3660 err = intel_pt_get_next_packet(decoder);
3661 if (err)
3662 return err;
3663
3664 switch (decoder->packet.type) {
3665 case INTEL_PT_TIP_PGD:
3666 decoder->continuous_period = false;
3667 decoder->pge = false;
3668 if (intel_pt_have_ip(decoder))
3669 intel_pt_set_ip(decoder);
3670 if (!decoder->ip)
3671 break;
3672 decoder->state.type |= INTEL_PT_TRACE_END;
3673 return 0;
3674
3675 case INTEL_PT_TIP_PGE:
3676 decoder->pge = true;
3677 intel_pt_mtc_cyc_cnt_pge(decoder);
3678 if (intel_pt_have_ip(decoder))
3679 intel_pt_set_ip(decoder);
3680 if (!decoder->ip)
3681 break;
3682 decoder->state.type |= INTEL_PT_TRACE_BEGIN;
3683 return 0;
3684
3685 case INTEL_PT_TIP:
3686 decoder->pge = true;
3687 if (intel_pt_have_ip(decoder))
3688 intel_pt_set_ip(decoder);
3689 if (!decoder->ip)
3690 break;
3691 return 0;
3692
3693 case INTEL_PT_FUP:
3694 if (intel_pt_have_ip(decoder))
3695 intel_pt_set_ip(decoder);
3696 if (decoder->ip)
3697 return 0;
3698 break;
3699
3700 case INTEL_PT_MTC:
3701 intel_pt_calc_mtc_timestamp(decoder);
3702 break;
3703
3704 case INTEL_PT_TSC:
3705 intel_pt_calc_tsc_timestamp(decoder);
3706 break;
3707
3708 case INTEL_PT_TMA:
3709 intel_pt_calc_tma(decoder);
3710 break;
3711
3712 case INTEL_PT_CYC:
3713 intel_pt_calc_cyc_timestamp(decoder);
3714 break;
3715
3716 case INTEL_PT_CBR:
3717 intel_pt_calc_cbr(decoder);
3718 break;
3719
3720 case INTEL_PT_PIP:
3721 intel_pt_set_pip(decoder);
3722 break;
3723
3724 case INTEL_PT_MODE_EXEC:
3725 intel_pt_mode_exec_status(decoder);
3726 break;
3727
3728 case INTEL_PT_MODE_TSX:
3729 intel_pt_update_in_tx(decoder);
3730 break;
3731
3732 case INTEL_PT_OVF:
3733 return intel_pt_overflow(decoder);
3734
3735 case INTEL_PT_BAD: /* Does not happen */
3736 return intel_pt_bug(decoder);
3737
3738 case INTEL_PT_TRACESTOP:
3739 decoder->pge = false;
3740 decoder->continuous_period = false;
3741 intel_pt_clear_tx_flags(decoder);
3742 decoder->have_tma = false;
3743 break;
3744
3745 case INTEL_PT_PSB:
3746 decoder->state.psb_offset = decoder->pos;
3747 decoder->psb_ip = 0;
3748 decoder->last_ip = 0;
3749 decoder->have_last_ip = true;
3750 intel_pt_clear_stack(&decoder->stack);
3751 err = intel_pt_walk_psb(decoder);
3752 if (err)
3753 return err;
3754 decoder->state.type = INTEL_PT_PSB_EVT;
3755 decoder->state.from_ip = decoder->psb_ip;
3756 decoder->state.to_ip = 0;
3757 return 0;
3758
3759 case INTEL_PT_TNT:
3760 case INTEL_PT_PSBEND:
3761 case INTEL_PT_VMCS:
3762 case INTEL_PT_MNT:
3763 case INTEL_PT_PAD:
3764 case INTEL_PT_PTWRITE:
3765 case INTEL_PT_PTWRITE_IP:
3766 case INTEL_PT_EXSTOP:
3767 case INTEL_PT_EXSTOP_IP:
3768 case INTEL_PT_MWAIT:
3769 case INTEL_PT_PWRE:
3770 case INTEL_PT_PWRX:
3771 case INTEL_PT_BBP:
3772 case INTEL_PT_BIP:
3773 case INTEL_PT_BEP:
3774 case INTEL_PT_BEP_IP:
3775 case INTEL_PT_CFE:
3776 case INTEL_PT_CFE_IP:
3777 case INTEL_PT_EVD:
3778 default:
3779 break;
3780 }
3781 }
3782 }
3783
intel_pt_sync_ip(struct intel_pt_decoder * decoder)3784 static int intel_pt_sync_ip(struct intel_pt_decoder *decoder)
3785 {
3786 int err;
3787
3788 intel_pt_clear_fup_event(decoder);
3789 decoder->overflow = false;
3790
3791 if (!decoder->branch_enable) {
3792 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
3793 decoder->state.type = 0; /* Do not have a sample */
3794 return 0;
3795 }
3796
3797 intel_pt_log("Scanning for full IP\n");
3798 err = intel_pt_walk_to_ip(decoder);
3799 if (err || ((decoder->state.type & INTEL_PT_PSB_EVT) && !decoder->ip))
3800 return err;
3801
3802 /* In hop mode, resample to get the to_ip as an "instruction" sample */
3803 if (decoder->hop)
3804 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
3805 else
3806 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
3807
3808 decoder->state.from_ip = 0;
3809 decoder->state.to_ip = decoder->ip;
3810 intel_pt_log_to("Setting IP", decoder->ip);
3811
3812 return 0;
3813 }
3814
intel_pt_part_psb(struct intel_pt_decoder * decoder)3815 static int intel_pt_part_psb(struct intel_pt_decoder *decoder)
3816 {
3817 const unsigned char *end = decoder->buf + decoder->len;
3818 size_t i;
3819
3820 for (i = INTEL_PT_PSB_LEN - 1; i; i--) {
3821 if (i > decoder->len)
3822 continue;
3823 if (!memcmp(end - i, INTEL_PT_PSB_STR, i))
3824 return i;
3825 }
3826 return 0;
3827 }
3828
intel_pt_rest_psb(struct intel_pt_decoder * decoder,int part_psb)3829 static int intel_pt_rest_psb(struct intel_pt_decoder *decoder, int part_psb)
3830 {
3831 size_t rest_psb = INTEL_PT_PSB_LEN - part_psb;
3832 const char *psb = INTEL_PT_PSB_STR;
3833
3834 if (rest_psb > decoder->len ||
3835 memcmp(decoder->buf, psb + part_psb, rest_psb))
3836 return 0;
3837
3838 return rest_psb;
3839 }
3840
intel_pt_get_split_psb(struct intel_pt_decoder * decoder,int part_psb)3841 static int intel_pt_get_split_psb(struct intel_pt_decoder *decoder,
3842 int part_psb)
3843 {
3844 int rest_psb, ret;
3845
3846 decoder->pos += decoder->len;
3847 decoder->len = 0;
3848
3849 ret = intel_pt_get_next_data(decoder, false);
3850 if (ret)
3851 return ret;
3852
3853 rest_psb = intel_pt_rest_psb(decoder, part_psb);
3854 if (!rest_psb)
3855 return 0;
3856
3857 decoder->pos -= part_psb;
3858 decoder->next_buf = decoder->buf + rest_psb;
3859 decoder->next_len = decoder->len - rest_psb;
3860 memcpy(decoder->temp_buf, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
3861 decoder->buf = decoder->temp_buf;
3862 decoder->len = INTEL_PT_PSB_LEN;
3863
3864 return 0;
3865 }
3866
intel_pt_scan_for_psb(struct intel_pt_decoder * decoder)3867 static int intel_pt_scan_for_psb(struct intel_pt_decoder *decoder)
3868 {
3869 unsigned char *next;
3870 int ret;
3871
3872 intel_pt_log("Scanning for PSB\n");
3873 while (1) {
3874 if (!decoder->len) {
3875 ret = intel_pt_get_next_data(decoder, false);
3876 if (ret)
3877 return ret;
3878 }
3879
3880 next = memmem(decoder->buf, decoder->len, INTEL_PT_PSB_STR,
3881 INTEL_PT_PSB_LEN);
3882 if (!next) {
3883 int part_psb;
3884
3885 part_psb = intel_pt_part_psb(decoder);
3886 if (part_psb) {
3887 ret = intel_pt_get_split_psb(decoder, part_psb);
3888 if (ret)
3889 return ret;
3890 } else {
3891 decoder->pos += decoder->len;
3892 decoder->len = 0;
3893 }
3894 continue;
3895 }
3896
3897 decoder->pkt_step = next - decoder->buf;
3898 return intel_pt_get_next_packet(decoder);
3899 }
3900 }
3901
intel_pt_sync(struct intel_pt_decoder * decoder)3902 static int intel_pt_sync(struct intel_pt_decoder *decoder)
3903 {
3904 int err;
3905
3906 decoder->pge = false;
3907 decoder->continuous_period = false;
3908 decoder->have_last_ip = false;
3909 decoder->last_ip = 0;
3910 decoder->psb_ip = 0;
3911 decoder->ip = 0;
3912 intel_pt_clear_stack(&decoder->stack);
3913
3914 err = intel_pt_scan_for_psb(decoder);
3915 if (err)
3916 return err;
3917
3918 if (decoder->vm_time_correlation) {
3919 decoder->in_psb = true;
3920 if (!decoder->timestamp)
3921 decoder->timestamp = 1;
3922 decoder->state.type = 0;
3923 decoder->pkt_state = INTEL_PT_STATE_VM_TIME_CORRELATION;
3924 return 0;
3925 }
3926
3927 decoder->have_last_ip = true;
3928 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
3929
3930 err = intel_pt_walk_psb(decoder);
3931 if (err)
3932 return err;
3933
3934 decoder->state.type = INTEL_PT_PSB_EVT; /* Only PSB sample */
3935 decoder->state.from_ip = decoder->psb_ip;
3936 decoder->state.to_ip = 0;
3937
3938 if (decoder->ip) {
3939 /*
3940 * In hop mode, resample to get the PSB FUP ip as an
3941 * "instruction" sample.
3942 */
3943 if (decoder->hop)
3944 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
3945 else
3946 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
3947 }
3948
3949 return 0;
3950 }
3951
intel_pt_est_timestamp(struct intel_pt_decoder * decoder)3952 static uint64_t intel_pt_est_timestamp(struct intel_pt_decoder *decoder)
3953 {
3954 uint64_t est = decoder->sample_insn_cnt << 1;
3955
3956 if (!decoder->cbr || !decoder->max_non_turbo_ratio)
3957 goto out;
3958
3959 est *= decoder->max_non_turbo_ratio;
3960 est /= decoder->cbr;
3961 out:
3962 return decoder->sample_timestamp + est;
3963 }
3964
intel_pt_decode(struct intel_pt_decoder * decoder)3965 const struct intel_pt_state *intel_pt_decode(struct intel_pt_decoder *decoder)
3966 {
3967 int err;
3968
3969 do {
3970 decoder->state.type = INTEL_PT_BRANCH;
3971 decoder->state.flags = 0;
3972
3973 switch (decoder->pkt_state) {
3974 case INTEL_PT_STATE_NO_PSB:
3975 err = intel_pt_sync(decoder);
3976 break;
3977 case INTEL_PT_STATE_NO_IP:
3978 decoder->have_last_ip = false;
3979 decoder->last_ip = 0;
3980 decoder->ip = 0;
3981 __fallthrough;
3982 case INTEL_PT_STATE_ERR_RESYNC:
3983 err = intel_pt_sync_ip(decoder);
3984 break;
3985 case INTEL_PT_STATE_IN_SYNC:
3986 err = intel_pt_walk_trace(decoder);
3987 break;
3988 case INTEL_PT_STATE_TNT:
3989 case INTEL_PT_STATE_TNT_CONT:
3990 err = intel_pt_walk_tnt(decoder);
3991 if (err == -EAGAIN)
3992 err = intel_pt_walk_trace(decoder);
3993 break;
3994 case INTEL_PT_STATE_TIP:
3995 case INTEL_PT_STATE_TIP_PGD:
3996 err = intel_pt_walk_tip(decoder);
3997 break;
3998 case INTEL_PT_STATE_FUP:
3999 err = intel_pt_walk_fup(decoder);
4000 if (err == -EAGAIN)
4001 err = intel_pt_walk_fup_tip(decoder);
4002 break;
4003 case INTEL_PT_STATE_FUP_NO_TIP:
4004 err = intel_pt_walk_fup(decoder);
4005 if (err == -EAGAIN)
4006 err = intel_pt_walk_trace(decoder);
4007 break;
4008 case INTEL_PT_STATE_FUP_IN_PSB:
4009 err = intel_pt_fup_in_psb(decoder);
4010 break;
4011 case INTEL_PT_STATE_RESAMPLE:
4012 err = intel_pt_resample(decoder);
4013 break;
4014 case INTEL_PT_STATE_VM_TIME_CORRELATION:
4015 err = intel_pt_vm_time_correlation(decoder);
4016 break;
4017 default:
4018 err = intel_pt_bug(decoder);
4019 break;
4020 }
4021 } while (err == -ENOLINK);
4022
4023 if (err) {
4024 decoder->state.err = intel_pt_ext_err(err);
4025 if (err != -EOVERFLOW)
4026 decoder->state.from_ip = decoder->ip;
4027 intel_pt_update_sample_time(decoder);
4028 decoder->sample_tot_cyc_cnt = decoder->tot_cyc_cnt;
4029 intel_pt_set_nr(decoder);
4030 } else {
4031 decoder->state.err = 0;
4032 if (decoder->cbr != decoder->cbr_seen) {
4033 decoder->cbr_seen = decoder->cbr;
4034 if (!decoder->state.type) {
4035 decoder->state.from_ip = decoder->ip;
4036 decoder->state.to_ip = 0;
4037 }
4038 decoder->state.type |= INTEL_PT_CBR_CHG;
4039 decoder->state.cbr_payload = decoder->cbr_payload;
4040 decoder->state.cbr = decoder->cbr;
4041 }
4042 if (intel_pt_sample_time(decoder->pkt_state)) {
4043 intel_pt_update_sample_time(decoder);
4044 if (decoder->sample_cyc) {
4045 decoder->sample_tot_cyc_cnt = decoder->tot_cyc_cnt;
4046 decoder->state.flags |= INTEL_PT_SAMPLE_IPC;
4047 decoder->sample_cyc = false;
4048 }
4049 }
4050 /*
4051 * When using only TSC/MTC to compute cycles, IPC can be
4052 * sampled as soon as the cycle count changes.
4053 */
4054 if (!decoder->have_cyc)
4055 decoder->state.flags |= INTEL_PT_SAMPLE_IPC;
4056 }
4057
4058 /* Let PSB event always have TSC timestamp */
4059 if ((decoder->state.type & INTEL_PT_PSB_EVT) && decoder->tsc_timestamp)
4060 decoder->sample_timestamp = decoder->tsc_timestamp;
4061
4062 decoder->state.from_nr = decoder->nr;
4063 decoder->state.to_nr = decoder->next_nr;
4064 decoder->nr = decoder->next_nr;
4065
4066 decoder->state.timestamp = decoder->sample_timestamp;
4067 decoder->state.est_timestamp = intel_pt_est_timestamp(decoder);
4068 decoder->state.tot_insn_cnt = decoder->tot_insn_cnt;
4069 decoder->state.tot_cyc_cnt = decoder->sample_tot_cyc_cnt;
4070
4071 return &decoder->state;
4072 }
4073
4074 /**
4075 * intel_pt_next_psb - move buffer pointer to the start of the next PSB packet.
4076 * @buf: pointer to buffer pointer
4077 * @len: size of buffer
4078 *
4079 * Updates the buffer pointer to point to the start of the next PSB packet if
4080 * there is one, otherwise the buffer pointer is unchanged. If @buf is updated,
4081 * @len is adjusted accordingly.
4082 *
4083 * Return: %true if a PSB packet is found, %false otherwise.
4084 */
intel_pt_next_psb(unsigned char ** buf,size_t * len)4085 static bool intel_pt_next_psb(unsigned char **buf, size_t *len)
4086 {
4087 unsigned char *next;
4088
4089 next = memmem(*buf, *len, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
4090 if (next) {
4091 *len -= next - *buf;
4092 *buf = next;
4093 return true;
4094 }
4095 return false;
4096 }
4097
4098 /**
4099 * intel_pt_step_psb - move buffer pointer to the start of the following PSB
4100 * packet.
4101 * @buf: pointer to buffer pointer
4102 * @len: size of buffer
4103 *
4104 * Updates the buffer pointer to point to the start of the following PSB packet
4105 * (skipping the PSB at @buf itself) if there is one, otherwise the buffer
4106 * pointer is unchanged. If @buf is updated, @len is adjusted accordingly.
4107 *
4108 * Return: %true if a PSB packet is found, %false otherwise.
4109 */
intel_pt_step_psb(unsigned char ** buf,size_t * len)4110 static bool intel_pt_step_psb(unsigned char **buf, size_t *len)
4111 {
4112 unsigned char *next;
4113
4114 if (!*len)
4115 return false;
4116
4117 next = memmem(*buf + 1, *len - 1, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
4118 if (next) {
4119 *len -= next - *buf;
4120 *buf = next;
4121 return true;
4122 }
4123 return false;
4124 }
4125
4126 /**
4127 * intel_pt_last_psb - find the last PSB packet in a buffer.
4128 * @buf: buffer
4129 * @len: size of buffer
4130 *
4131 * This function finds the last PSB in a buffer.
4132 *
4133 * Return: A pointer to the last PSB in @buf if found, %NULL otherwise.
4134 */
intel_pt_last_psb(unsigned char * buf,size_t len)4135 static unsigned char *intel_pt_last_psb(unsigned char *buf, size_t len)
4136 {
4137 const char *n = INTEL_PT_PSB_STR;
4138 unsigned char *p;
4139 size_t k;
4140
4141 if (len < INTEL_PT_PSB_LEN)
4142 return NULL;
4143
4144 k = len - INTEL_PT_PSB_LEN + 1;
4145 while (1) {
4146 p = memrchr(buf, n[0], k);
4147 if (!p)
4148 return NULL;
4149 if (!memcmp(p + 1, n + 1, INTEL_PT_PSB_LEN - 1))
4150 return p;
4151 k = p - buf;
4152 if (!k)
4153 return NULL;
4154 }
4155 }
4156
4157 /**
4158 * intel_pt_next_tsc - find and return next TSC.
4159 * @buf: buffer
4160 * @len: size of buffer
4161 * @tsc: TSC value returned
4162 * @rem: returns remaining size when TSC is found
4163 *
4164 * Find a TSC packet in @buf and return the TSC value. This function assumes
4165 * that @buf starts at a PSB and that PSB+ will contain TSC and so stops if a
4166 * PSBEND packet is found.
4167 *
4168 * Return: %true if TSC is found, false otherwise.
4169 */
intel_pt_next_tsc(unsigned char * buf,size_t len,uint64_t * tsc,size_t * rem)4170 static bool intel_pt_next_tsc(unsigned char *buf, size_t len, uint64_t *tsc,
4171 size_t *rem)
4172 {
4173 enum intel_pt_pkt_ctx ctx = INTEL_PT_NO_CTX;
4174 struct intel_pt_pkt packet;
4175 int ret;
4176
4177 while (len) {
4178 ret = intel_pt_get_packet(buf, len, &packet, &ctx);
4179 if (ret <= 0)
4180 return false;
4181 if (packet.type == INTEL_PT_TSC) {
4182 *tsc = packet.payload;
4183 *rem = len;
4184 return true;
4185 }
4186 if (packet.type == INTEL_PT_PSBEND)
4187 return false;
4188 buf += ret;
4189 len -= ret;
4190 }
4191 return false;
4192 }
4193
4194 /**
4195 * intel_pt_tsc_cmp - compare 7-byte TSCs.
4196 * @tsc1: first TSC to compare
4197 * @tsc2: second TSC to compare
4198 *
4199 * This function compares 7-byte TSC values allowing for the possibility that
4200 * TSC wrapped around. Generally it is not possible to know if TSC has wrapped
4201 * around so for that purpose this function assumes the absolute difference is
4202 * less than half the maximum difference.
4203 *
4204 * Return: %-1 if @tsc1 is before @tsc2, %0 if @tsc1 == @tsc2, %1 if @tsc1 is
4205 * after @tsc2.
4206 */
intel_pt_tsc_cmp(uint64_t tsc1,uint64_t tsc2)4207 static int intel_pt_tsc_cmp(uint64_t tsc1, uint64_t tsc2)
4208 {
4209 const uint64_t halfway = (1ULL << 55);
4210
4211 if (tsc1 == tsc2)
4212 return 0;
4213
4214 if (tsc1 < tsc2) {
4215 if (tsc2 - tsc1 < halfway)
4216 return -1;
4217 else
4218 return 1;
4219 } else {
4220 if (tsc1 - tsc2 < halfway)
4221 return 1;
4222 else
4223 return -1;
4224 }
4225 }
4226
4227 #define MAX_PADDING (PERF_AUXTRACE_RECORD_ALIGNMENT - 1)
4228
4229 /**
4230 * adj_for_padding - adjust overlap to account for padding.
4231 * @buf_b: second buffer
4232 * @buf_a: first buffer
4233 * @len_a: size of first buffer
4234 *
4235 * @buf_a might have up to 7 bytes of padding appended. Adjust the overlap
4236 * accordingly.
4237 *
4238 * Return: A pointer into @buf_b from where non-overlapped data starts
4239 */
adj_for_padding(unsigned char * buf_b,unsigned char * buf_a,size_t len_a)4240 static unsigned char *adj_for_padding(unsigned char *buf_b,
4241 unsigned char *buf_a, size_t len_a)
4242 {
4243 unsigned char *p = buf_b - MAX_PADDING;
4244 unsigned char *q = buf_a + len_a - MAX_PADDING;
4245 int i;
4246
4247 for (i = MAX_PADDING; i; i--, p++, q++) {
4248 if (*p != *q)
4249 break;
4250 }
4251
4252 return p;
4253 }
4254
4255 /**
4256 * intel_pt_find_overlap_tsc - determine start of non-overlapped trace data
4257 * using TSC.
4258 * @buf_a: first buffer
4259 * @len_a: size of first buffer
4260 * @buf_b: second buffer
4261 * @len_b: size of second buffer
4262 * @consecutive: returns true if there is data in buf_b that is consecutive
4263 * to buf_a
4264 * @ooo_tsc: out-of-order TSC due to VM TSC offset / scaling
4265 *
4266 * If the trace contains TSC we can look at the last TSC of @buf_a and the
4267 * first TSC of @buf_b in order to determine if the buffers overlap, and then
4268 * walk forward in @buf_b until a later TSC is found. A precondition is that
4269 * @buf_a and @buf_b are positioned at a PSB.
4270 *
4271 * Return: A pointer into @buf_b from where non-overlapped data starts, or
4272 * @buf_b + @len_b if there is no non-overlapped data.
4273 */
intel_pt_find_overlap_tsc(unsigned char * buf_a,size_t len_a,unsigned char * buf_b,size_t len_b,bool * consecutive,bool ooo_tsc)4274 static unsigned char *intel_pt_find_overlap_tsc(unsigned char *buf_a,
4275 size_t len_a,
4276 unsigned char *buf_b,
4277 size_t len_b, bool *consecutive,
4278 bool ooo_tsc)
4279 {
4280 uint64_t tsc_a, tsc_b;
4281 unsigned char *p;
4282 size_t len, rem_a, rem_b;
4283
4284 p = intel_pt_last_psb(buf_a, len_a);
4285 if (!p)
4286 return buf_b; /* No PSB in buf_a => no overlap */
4287
4288 len = len_a - (p - buf_a);
4289 if (!intel_pt_next_tsc(p, len, &tsc_a, &rem_a)) {
4290 /* The last PSB+ in buf_a is incomplete, so go back one more */
4291 len_a -= len;
4292 p = intel_pt_last_psb(buf_a, len_a);
4293 if (!p)
4294 return buf_b; /* No full PSB+ => assume no overlap */
4295 len = len_a - (p - buf_a);
4296 if (!intel_pt_next_tsc(p, len, &tsc_a, &rem_a))
4297 return buf_b; /* No TSC in buf_a => assume no overlap */
4298 }
4299
4300 while (1) {
4301 /* Ignore PSB+ with no TSC */
4302 if (intel_pt_next_tsc(buf_b, len_b, &tsc_b, &rem_b)) {
4303 int cmp = intel_pt_tsc_cmp(tsc_a, tsc_b);
4304
4305 /* Same TSC, so buffers are consecutive */
4306 if (!cmp && rem_b >= rem_a) {
4307 unsigned char *start;
4308
4309 *consecutive = true;
4310 start = buf_b + len_b - (rem_b - rem_a);
4311 return adj_for_padding(start, buf_a, len_a);
4312 }
4313 if (cmp < 0 && !ooo_tsc)
4314 return buf_b; /* tsc_a < tsc_b => no overlap */
4315 }
4316
4317 if (!intel_pt_step_psb(&buf_b, &len_b))
4318 return buf_b + len_b; /* No PSB in buf_b => no data */
4319 }
4320 }
4321
4322 /**
4323 * intel_pt_find_overlap - determine start of non-overlapped trace data.
4324 * @buf_a: first buffer
4325 * @len_a: size of first buffer
4326 * @buf_b: second buffer
4327 * @len_b: size of second buffer
4328 * @have_tsc: can use TSC packets to detect overlap
4329 * @consecutive: returns true if there is data in buf_b that is consecutive
4330 * to buf_a
4331 * @ooo_tsc: out-of-order TSC due to VM TSC offset / scaling
4332 *
4333 * When trace samples or snapshots are recorded there is the possibility that
4334 * the data overlaps. Note that, for the purposes of decoding, data is only
4335 * useful if it begins with a PSB packet.
4336 *
4337 * Return: A pointer into @buf_b from where non-overlapped data starts, or
4338 * @buf_b + @len_b if there is no non-overlapped data.
4339 */
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,bool ooo_tsc)4340 unsigned char *intel_pt_find_overlap(unsigned char *buf_a, size_t len_a,
4341 unsigned char *buf_b, size_t len_b,
4342 bool have_tsc, bool *consecutive,
4343 bool ooo_tsc)
4344 {
4345 unsigned char *found;
4346
4347 /* Buffer 'b' must start at PSB so throw away everything before that */
4348 if (!intel_pt_next_psb(&buf_b, &len_b))
4349 return buf_b + len_b; /* No PSB */
4350
4351 if (!intel_pt_next_psb(&buf_a, &len_a))
4352 return buf_b; /* No overlap */
4353
4354 if (have_tsc) {
4355 found = intel_pt_find_overlap_tsc(buf_a, len_a, buf_b, len_b,
4356 consecutive, ooo_tsc);
4357 if (found)
4358 return found;
4359 }
4360
4361 /*
4362 * Buffer 'b' cannot end within buffer 'a' so, for comparison purposes,
4363 * we can ignore the first part of buffer 'a'.
4364 */
4365 while (len_b < len_a) {
4366 if (!intel_pt_step_psb(&buf_a, &len_a))
4367 return buf_b; /* No overlap */
4368 }
4369
4370 /* Now len_b >= len_a */
4371 while (1) {
4372 /* Potential overlap so check the bytes */
4373 found = memmem(buf_a, len_a, buf_b, len_a);
4374 if (found) {
4375 *consecutive = true;
4376 return adj_for_padding(buf_b + len_a, buf_a, len_a);
4377 }
4378
4379 /* Try again at next PSB in buffer 'a' */
4380 if (!intel_pt_step_psb(&buf_a, &len_a))
4381 return buf_b; /* No overlap */
4382 }
4383 }
4384
4385 /**
4386 * struct fast_forward_data - data used by intel_pt_ff_cb().
4387 * @timestamp: timestamp to fast forward towards
4388 * @buf_timestamp: buffer timestamp of last buffer with trace data earlier than
4389 * the fast forward timestamp.
4390 */
4391 struct fast_forward_data {
4392 uint64_t timestamp;
4393 uint64_t buf_timestamp;
4394 };
4395
4396 /**
4397 * intel_pt_ff_cb - fast forward lookahead callback.
4398 * @buffer: Intel PT trace buffer
4399 * @data: opaque pointer to fast forward data (struct fast_forward_data)
4400 *
4401 * Determine if @buffer trace is past the fast forward timestamp.
4402 *
4403 * Return: 1 (stop lookahead) if @buffer trace is past the fast forward
4404 * timestamp, and 0 otherwise.
4405 */
intel_pt_ff_cb(struct intel_pt_buffer * buffer,void * data)4406 static int intel_pt_ff_cb(struct intel_pt_buffer *buffer, void *data)
4407 {
4408 struct fast_forward_data *d = data;
4409 unsigned char *buf;
4410 uint64_t tsc;
4411 size_t rem;
4412 size_t len;
4413
4414 buf = (unsigned char *)buffer->buf;
4415 len = buffer->len;
4416
4417 if (!intel_pt_next_psb(&buf, &len) ||
4418 !intel_pt_next_tsc(buf, len, &tsc, &rem))
4419 return 0;
4420
4421 tsc = intel_pt_8b_tsc(tsc, buffer->ref_timestamp);
4422
4423 intel_pt_log("Buffer 1st timestamp " x64_fmt " ref timestamp " x64_fmt "\n",
4424 tsc, buffer->ref_timestamp);
4425
4426 /*
4427 * If the buffer contains a timestamp earlier that the fast forward
4428 * timestamp, then record it, else stop.
4429 */
4430 if (tsc < d->timestamp)
4431 d->buf_timestamp = buffer->ref_timestamp;
4432 else
4433 return 1;
4434
4435 return 0;
4436 }
4437
4438 /**
4439 * intel_pt_fast_forward - reposition decoder forwards.
4440 * @decoder: Intel PT decoder
4441 * @timestamp: timestamp to fast forward towards
4442 *
4443 * Reposition decoder at the last PSB with a timestamp earlier than @timestamp.
4444 *
4445 * Return: 0 on success or negative error code on failure.
4446 */
intel_pt_fast_forward(struct intel_pt_decoder * decoder,uint64_t timestamp)4447 int intel_pt_fast_forward(struct intel_pt_decoder *decoder, uint64_t timestamp)
4448 {
4449 struct fast_forward_data d = { .timestamp = timestamp };
4450 unsigned char *buf;
4451 size_t len;
4452 int err;
4453
4454 intel_pt_log("Fast forward towards timestamp " x64_fmt "\n", timestamp);
4455
4456 /* Find buffer timestamp of buffer to fast forward to */
4457 err = decoder->lookahead(decoder->data, intel_pt_ff_cb, &d);
4458 if (err < 0)
4459 return err;
4460
4461 /* Walk to buffer with same buffer timestamp */
4462 if (d.buf_timestamp) {
4463 do {
4464 decoder->pos += decoder->len;
4465 decoder->len = 0;
4466 err = intel_pt_get_next_data(decoder, true);
4467 /* -ENOLINK means non-consecutive trace */
4468 if (err && err != -ENOLINK)
4469 return err;
4470 } while (decoder->buf_timestamp != d.buf_timestamp);
4471 }
4472
4473 if (!decoder->buf)
4474 return 0;
4475
4476 buf = (unsigned char *)decoder->buf;
4477 len = decoder->len;
4478
4479 if (!intel_pt_next_psb(&buf, &len))
4480 return 0;
4481
4482 /*
4483 * Walk PSBs while the PSB timestamp is less than the fast forward
4484 * timestamp.
4485 */
4486 do {
4487 uint64_t tsc;
4488 size_t rem;
4489
4490 if (!intel_pt_next_tsc(buf, len, &tsc, &rem))
4491 break;
4492 tsc = intel_pt_8b_tsc(tsc, decoder->buf_timestamp);
4493 /*
4494 * A TSC packet can slip past MTC packets but, after fast
4495 * forward, decoding starts at the TSC timestamp. That means
4496 * the timestamps may not be exactly the same as the timestamps
4497 * that would have been decoded without fast forward.
4498 */
4499 if (tsc < timestamp) {
4500 intel_pt_log("Fast forward to next PSB timestamp " x64_fmt "\n", tsc);
4501 decoder->pos += decoder->len - len;
4502 decoder->buf = buf;
4503 decoder->len = len;
4504 intel_pt_reposition(decoder);
4505 } else {
4506 break;
4507 }
4508 } while (intel_pt_step_psb(&buf, &len));
4509
4510 return 0;
4511 }
4512