2 * intel_pt_decoder.c: Intel Processor Trace support
3 * Copyright (c) 2013-2014, Intel Corporation.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
25 #include <linux/compiler.h>
29 #include "../auxtrace.h"
31 #include "intel-pt-insn-decoder.h"
32 #include "intel-pt-pkt-decoder.h"
33 #include "intel-pt-decoder.h"
34 #include "intel-pt-log.h"
36 #define INTEL_PT_BLK_SIZE 1024
38 #define BIT63 (((uint64_t)1 << 63))
40 #define INTEL_PT_RETURN 1
42 /* Maximum number of loops with no packets consumed i.e. stuck in a loop */
43 #define INTEL_PT_MAX_LOOPS 10000
46 struct intel_pt_blk *prev;
47 uint64_t ip[INTEL_PT_BLK_SIZE];
50 struct intel_pt_stack {
51 struct intel_pt_blk *blk;
52 struct intel_pt_blk *spare;
56 enum intel_pt_pkt_state {
57 INTEL_PT_STATE_NO_PSB,
59 INTEL_PT_STATE_ERR_RESYNC,
60 INTEL_PT_STATE_IN_SYNC,
63 INTEL_PT_STATE_TIP_PGD,
65 INTEL_PT_STATE_FUP_NO_TIP,
68 static inline bool intel_pt_sample_time(enum intel_pt_pkt_state pkt_state)
71 case INTEL_PT_STATE_NO_PSB:
72 case INTEL_PT_STATE_NO_IP:
73 case INTEL_PT_STATE_ERR_RESYNC:
74 case INTEL_PT_STATE_IN_SYNC:
75 case INTEL_PT_STATE_TNT:
77 case INTEL_PT_STATE_TIP:
78 case INTEL_PT_STATE_TIP_PGD:
79 case INTEL_PT_STATE_FUP:
80 case INTEL_PT_STATE_FUP_NO_TIP:
87 #ifdef INTEL_PT_STRICT
88 #define INTEL_PT_STATE_ERR1 INTEL_PT_STATE_NO_PSB
89 #define INTEL_PT_STATE_ERR2 INTEL_PT_STATE_NO_PSB
90 #define INTEL_PT_STATE_ERR3 INTEL_PT_STATE_NO_PSB
91 #define INTEL_PT_STATE_ERR4 INTEL_PT_STATE_NO_PSB
93 #define INTEL_PT_STATE_ERR1 (decoder->pkt_state)
94 #define INTEL_PT_STATE_ERR2 INTEL_PT_STATE_NO_IP
95 #define INTEL_PT_STATE_ERR3 INTEL_PT_STATE_ERR_RESYNC
96 #define INTEL_PT_STATE_ERR4 INTEL_PT_STATE_IN_SYNC
99 struct intel_pt_decoder {
100 int (*get_trace)(struct intel_pt_buffer *buffer, void *data);
101 int (*walk_insn)(struct intel_pt_insn *intel_pt_insn,
102 uint64_t *insn_cnt_ptr, uint64_t *ip, uint64_t to_ip,
103 uint64_t max_insn_cnt, void *data);
104 bool (*pgd_ip)(uint64_t ip, void *data);
106 struct intel_pt_state state;
107 const unsigned char *buf;
109 bool return_compression;
117 enum intel_pt_param_flags flags;
123 uint64_t tsc_timestamp;
124 uint64_t ref_timestamp;
125 uint64_t sample_timestamp;
127 uint64_t ctc_timestamp;
130 uint64_t cyc_ref_timestamp;
132 uint32_t tsc_ctc_ratio_n;
133 uint32_t tsc_ctc_ratio_d;
134 uint32_t tsc_ctc_mult;
136 uint32_t ctc_rem_mask;
138 struct intel_pt_stack stack;
139 enum intel_pt_pkt_state pkt_state;
140 struct intel_pt_pkt packet;
141 struct intel_pt_pkt tnt;
144 int last_packet_type;
146 unsigned int cbr_seen;
147 unsigned int max_non_turbo_ratio;
148 double max_non_turbo_ratio_fp;
149 double cbr_cyc_to_tsc;
150 double calc_cyc_to_tsc;
151 bool have_calc_cyc_to_tsc;
153 unsigned int insn_bytes;
155 enum intel_pt_period_type period_type;
156 uint64_t tot_insn_cnt;
157 uint64_t period_insn_cnt;
158 uint64_t period_mask;
159 uint64_t period_ticks;
160 uint64_t last_masked_timestamp;
161 bool continuous_period;
163 bool set_fup_tx_flags;
168 unsigned int fup_tx_flags;
169 unsigned int tx_flags;
170 uint64_t fup_ptw_payload;
171 uint64_t fup_mwait_payload;
172 uint64_t fup_pwre_payload;
173 uint64_t cbr_payload;
174 uint64_t timestamp_insn_cnt;
175 uint64_t sample_insn_cnt;
180 const unsigned char *next_buf;
182 unsigned char temp_buf[INTEL_PT_PKT_MAX_SZ];
185 static uint64_t intel_pt_lower_power_of_2(uint64_t x)
189 for (i = 0; x != 1; i++)
195 static void intel_pt_setup_period(struct intel_pt_decoder *decoder)
197 if (decoder->period_type == INTEL_PT_PERIOD_TICKS) {
200 period = intel_pt_lower_power_of_2(decoder->period);
201 decoder->period_mask = ~(period - 1);
202 decoder->period_ticks = period;
206 static uint64_t multdiv(uint64_t t, uint32_t n, uint32_t d)
210 return (t / d) * n + ((t % d) * n) / d;
213 struct intel_pt_decoder *intel_pt_decoder_new(struct intel_pt_params *params)
215 struct intel_pt_decoder *decoder;
217 if (!params->get_trace || !params->walk_insn)
220 decoder = zalloc(sizeof(struct intel_pt_decoder));
224 decoder->get_trace = params->get_trace;
225 decoder->walk_insn = params->walk_insn;
226 decoder->pgd_ip = params->pgd_ip;
227 decoder->data = params->data;
228 decoder->return_compression = params->return_compression;
229 decoder->branch_enable = params->branch_enable;
231 decoder->flags = params->flags;
233 decoder->period = params->period;
234 decoder->period_type = params->period_type;
236 decoder->max_non_turbo_ratio = params->max_non_turbo_ratio;
237 decoder->max_non_turbo_ratio_fp = params->max_non_turbo_ratio;
239 intel_pt_setup_period(decoder);
241 decoder->mtc_shift = params->mtc_period;
242 decoder->ctc_rem_mask = (1 << decoder->mtc_shift) - 1;
244 decoder->tsc_ctc_ratio_n = params->tsc_ctc_ratio_n;
245 decoder->tsc_ctc_ratio_d = params->tsc_ctc_ratio_d;
247 if (!decoder->tsc_ctc_ratio_n)
248 decoder->tsc_ctc_ratio_d = 0;
250 if (decoder->tsc_ctc_ratio_d) {
251 if (!(decoder->tsc_ctc_ratio_n % decoder->tsc_ctc_ratio_d))
252 decoder->tsc_ctc_mult = decoder->tsc_ctc_ratio_n /
253 decoder->tsc_ctc_ratio_d;
257 * A TSC packet can slip past MTC packets so that the timestamp appears
258 * to go backwards. One estimate is that can be up to about 40 CPU
259 * cycles, which is certainly less than 0x1000 TSC ticks, but accept
260 * slippage an order of magnitude more to be on the safe side.
262 decoder->tsc_slip = 0x10000;
264 intel_pt_log("timestamp: mtc_shift %u\n", decoder->mtc_shift);
265 intel_pt_log("timestamp: tsc_ctc_ratio_n %u\n", decoder->tsc_ctc_ratio_n);
266 intel_pt_log("timestamp: tsc_ctc_ratio_d %u\n", decoder->tsc_ctc_ratio_d);
267 intel_pt_log("timestamp: tsc_ctc_mult %u\n", decoder->tsc_ctc_mult);
268 intel_pt_log("timestamp: tsc_slip %#x\n", decoder->tsc_slip);
273 static void intel_pt_pop_blk(struct intel_pt_stack *stack)
275 struct intel_pt_blk *blk = stack->blk;
277 stack->blk = blk->prev;
284 static uint64_t intel_pt_pop(struct intel_pt_stack *stack)
289 intel_pt_pop_blk(stack);
292 stack->pos = INTEL_PT_BLK_SIZE;
294 return stack->blk->ip[--stack->pos];
297 static int intel_pt_alloc_blk(struct intel_pt_stack *stack)
299 struct intel_pt_blk *blk;
305 blk = malloc(sizeof(struct intel_pt_blk));
310 blk->prev = stack->blk;
316 static int intel_pt_push(struct intel_pt_stack *stack, uint64_t ip)
320 if (!stack->blk || stack->pos == INTEL_PT_BLK_SIZE) {
321 err = intel_pt_alloc_blk(stack);
326 stack->blk->ip[stack->pos++] = ip;
330 static void intel_pt_clear_stack(struct intel_pt_stack *stack)
333 intel_pt_pop_blk(stack);
337 static void intel_pt_free_stack(struct intel_pt_stack *stack)
339 intel_pt_clear_stack(stack);
341 zfree(&stack->spare);
344 void intel_pt_decoder_free(struct intel_pt_decoder *decoder)
346 intel_pt_free_stack(&decoder->stack);
350 static int intel_pt_ext_err(int code)
354 return INTEL_PT_ERR_NOMEM;
356 return INTEL_PT_ERR_INTERN;
358 return INTEL_PT_ERR_BADPKT;
360 return INTEL_PT_ERR_NODATA;
362 return INTEL_PT_ERR_NOINSN;
364 return INTEL_PT_ERR_MISMAT;
366 return INTEL_PT_ERR_OVR;
368 return INTEL_PT_ERR_LOST;
370 return INTEL_PT_ERR_NELOOP;
372 return INTEL_PT_ERR_UNK;
376 static const char *intel_pt_err_msgs[] = {
377 [INTEL_PT_ERR_NOMEM] = "Memory allocation failed",
378 [INTEL_PT_ERR_INTERN] = "Internal error",
379 [INTEL_PT_ERR_BADPKT] = "Bad packet",
380 [INTEL_PT_ERR_NODATA] = "No more data",
381 [INTEL_PT_ERR_NOINSN] = "Failed to get instruction",
382 [INTEL_PT_ERR_MISMAT] = "Trace doesn't match instruction",
383 [INTEL_PT_ERR_OVR] = "Overflow packet",
384 [INTEL_PT_ERR_LOST] = "Lost trace data",
385 [INTEL_PT_ERR_UNK] = "Unknown error!",
386 [INTEL_PT_ERR_NELOOP] = "Never-ending loop",
389 int intel_pt__strerror(int code, char *buf, size_t buflen)
391 if (code < 1 || code >= INTEL_PT_ERR_MAX)
392 code = INTEL_PT_ERR_UNK;
393 strlcpy(buf, intel_pt_err_msgs[code], buflen);
397 static uint64_t intel_pt_calc_ip(const struct intel_pt_pkt *packet,
402 switch (packet->count) {
404 ip = (last_ip & (uint64_t)0xffffffffffff0000ULL) |
408 ip = (last_ip & (uint64_t)0xffffffff00000000ULL) |
412 ip = packet->payload;
413 /* Sign-extend 6-byte ip */
414 if (ip & (uint64_t)0x800000000000ULL)
415 ip |= (uint64_t)0xffff000000000000ULL;
418 ip = (last_ip & (uint64_t)0xffff000000000000ULL) |
422 ip = packet->payload;
431 static inline void intel_pt_set_last_ip(struct intel_pt_decoder *decoder)
433 decoder->last_ip = intel_pt_calc_ip(&decoder->packet, decoder->last_ip);
434 decoder->have_last_ip = true;
437 static inline void intel_pt_set_ip(struct intel_pt_decoder *decoder)
439 intel_pt_set_last_ip(decoder);
440 decoder->ip = decoder->last_ip;
443 static void intel_pt_decoder_log_packet(struct intel_pt_decoder *decoder)
445 intel_pt_log_packet(&decoder->packet, decoder->pkt_len, decoder->pos,
449 static int intel_pt_bug(struct intel_pt_decoder *decoder)
451 intel_pt_log("ERROR: Internal error\n");
452 decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
456 static inline void intel_pt_clear_tx_flags(struct intel_pt_decoder *decoder)
458 decoder->tx_flags = 0;
461 static inline void intel_pt_update_in_tx(struct intel_pt_decoder *decoder)
463 decoder->tx_flags = decoder->packet.payload & INTEL_PT_IN_TX;
466 static int intel_pt_bad_packet(struct intel_pt_decoder *decoder)
468 intel_pt_clear_tx_flags(decoder);
469 decoder->have_tma = false;
470 decoder->pkt_len = 1;
471 decoder->pkt_step = 1;
472 intel_pt_decoder_log_packet(decoder);
473 if (decoder->pkt_state != INTEL_PT_STATE_NO_PSB) {
474 intel_pt_log("ERROR: Bad packet\n");
475 decoder->pkt_state = INTEL_PT_STATE_ERR1;
480 static int intel_pt_get_data(struct intel_pt_decoder *decoder)
482 struct intel_pt_buffer buffer = { .buf = 0, };
485 decoder->pkt_step = 0;
487 intel_pt_log("Getting more data\n");
488 ret = decoder->get_trace(&buffer, decoder->data);
491 decoder->buf = buffer.buf;
492 decoder->len = buffer.len;
494 intel_pt_log("No more data\n");
497 if (!buffer.consecutive) {
499 decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
500 decoder->ref_timestamp = buffer.ref_timestamp;
501 decoder->timestamp = 0;
502 decoder->have_tma = false;
503 decoder->state.trace_nr = buffer.trace_nr;
504 intel_pt_log("Reference timestamp 0x%" PRIx64 "\n",
505 decoder->ref_timestamp);
512 static int intel_pt_get_next_data(struct intel_pt_decoder *decoder)
514 if (!decoder->next_buf)
515 return intel_pt_get_data(decoder);
517 decoder->buf = decoder->next_buf;
518 decoder->len = decoder->next_len;
519 decoder->next_buf = 0;
520 decoder->next_len = 0;
524 static int intel_pt_get_split_packet(struct intel_pt_decoder *decoder)
526 unsigned char *buf = decoder->temp_buf;
527 size_t old_len, len, n;
530 old_len = decoder->len;
532 memcpy(buf, decoder->buf, len);
534 ret = intel_pt_get_data(decoder);
536 decoder->pos += old_len;
537 return ret < 0 ? ret : -EINVAL;
540 n = INTEL_PT_PKT_MAX_SZ - len;
541 if (n > decoder->len)
543 memcpy(buf + len, decoder->buf, n);
546 ret = intel_pt_get_packet(buf, len, &decoder->packet);
547 if (ret < (int)old_len) {
548 decoder->next_buf = decoder->buf;
549 decoder->next_len = decoder->len;
551 decoder->len = old_len;
552 return intel_pt_bad_packet(decoder);
555 decoder->next_buf = decoder->buf + (ret - old_len);
556 decoder->next_len = decoder->len - (ret - old_len);
564 struct intel_pt_pkt_info {
565 struct intel_pt_decoder *decoder;
566 struct intel_pt_pkt packet;
569 int last_packet_type;
573 typedef int (*intel_pt_pkt_cb_t)(struct intel_pt_pkt_info *pkt_info);
575 /* Lookahead packets in current buffer */
576 static int intel_pt_pkt_lookahead(struct intel_pt_decoder *decoder,
577 intel_pt_pkt_cb_t cb, void *data)
579 struct intel_pt_pkt_info pkt_info;
580 const unsigned char *buf = decoder->buf;
581 size_t len = decoder->len;
584 pkt_info.decoder = decoder;
585 pkt_info.pos = decoder->pos;
586 pkt_info.pkt_len = decoder->pkt_step;
587 pkt_info.last_packet_type = decoder->last_packet_type;
588 pkt_info.data = data;
592 pkt_info.pos += pkt_info.pkt_len;
593 buf += pkt_info.pkt_len;
594 len -= pkt_info.pkt_len;
597 return INTEL_PT_NEED_MORE_BYTES;
599 ret = intel_pt_get_packet(buf, len, &pkt_info.packet);
601 return INTEL_PT_NEED_MORE_BYTES;
605 pkt_info.pkt_len = ret;
606 } while (pkt_info.packet.type == INTEL_PT_PAD);
612 pkt_info.last_packet_type = pkt_info.packet.type;
616 struct intel_pt_calc_cyc_to_tsc_info {
620 uint64_t ctc_timestamp;
622 uint64_t tsc_timestamp;
627 double cbr_cyc_to_tsc;
631 * MTC provides a 8-bit slice of CTC but the TMA packet only provides the lower
632 * 16 bits of CTC. If mtc_shift > 8 then some of the MTC bits are not in the CTC
633 * provided by the TMA packet. Fix-up the last_mtc calculated from the TMA
634 * packet by copying the missing bits from the current MTC assuming the least
635 * difference between the two, and that the current MTC comes after last_mtc.
637 static void intel_pt_fixup_last_mtc(uint32_t mtc, int mtc_shift,
640 uint32_t first_missing_bit = 1U << (16 - mtc_shift);
641 uint32_t mask = ~(first_missing_bit - 1);
643 *last_mtc |= mtc & mask;
644 if (*last_mtc >= mtc) {
645 *last_mtc -= first_missing_bit;
650 static int intel_pt_calc_cyc_cb(struct intel_pt_pkt_info *pkt_info)
652 struct intel_pt_decoder *decoder = pkt_info->decoder;
653 struct intel_pt_calc_cyc_to_tsc_info *data = pkt_info->data;
657 uint32_t mtc, mtc_delta, ctc, fc, ctc_rem;
659 switch (pkt_info->packet.type) {
661 case INTEL_PT_TIP_PGE:
666 case INTEL_PT_MODE_EXEC:
667 case INTEL_PT_MODE_TSX:
668 case INTEL_PT_PSBEND:
672 case INTEL_PT_PTWRITE:
673 case INTEL_PT_PTWRITE_IP:
680 mtc = pkt_info->packet.payload;
681 if (decoder->mtc_shift > 8 && data->fixup_last_mtc) {
682 data->fixup_last_mtc = false;
683 intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
686 if (mtc > data->last_mtc)
687 mtc_delta = mtc - data->last_mtc;
689 mtc_delta = mtc + 256 - data->last_mtc;
690 data->ctc_delta += mtc_delta << decoder->mtc_shift;
691 data->last_mtc = mtc;
693 if (decoder->tsc_ctc_mult) {
694 timestamp = data->ctc_timestamp +
695 data->ctc_delta * decoder->tsc_ctc_mult;
697 timestamp = data->ctc_timestamp +
698 multdiv(data->ctc_delta,
699 decoder->tsc_ctc_ratio_n,
700 decoder->tsc_ctc_ratio_d);
703 if (timestamp < data->timestamp)
706 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
707 data->timestamp = timestamp;
715 * For now, do not support using TSC packets - refer
716 * intel_pt_calc_cyc_to_tsc().
720 timestamp = pkt_info->packet.payload |
721 (data->timestamp & (0xffULL << 56));
722 if (data->from_mtc && timestamp < data->timestamp &&
723 data->timestamp - timestamp < decoder->tsc_slip)
725 if (timestamp < data->timestamp)
726 timestamp += (1ULL << 56);
727 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
730 data->tsc_timestamp = timestamp;
731 data->timestamp = timestamp;
740 if (!decoder->tsc_ctc_ratio_d)
743 ctc = pkt_info->packet.payload;
744 fc = pkt_info->packet.count;
745 ctc_rem = ctc & decoder->ctc_rem_mask;
747 data->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
749 data->ctc_timestamp = data->tsc_timestamp - fc;
750 if (decoder->tsc_ctc_mult) {
751 data->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
753 data->ctc_timestamp -=
754 multdiv(ctc_rem, decoder->tsc_ctc_ratio_n,
755 decoder->tsc_ctc_ratio_d);
759 data->have_tma = true;
760 data->fixup_last_mtc = true;
765 data->cycle_cnt += pkt_info->packet.payload;
769 cbr = pkt_info->packet.payload;
770 if (data->cbr && data->cbr != cbr)
773 data->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
776 case INTEL_PT_TIP_PGD:
777 case INTEL_PT_TRACESTOP:
778 case INTEL_PT_EXSTOP:
779 case INTEL_PT_EXSTOP_IP:
784 case INTEL_PT_BAD: /* Does not happen */
789 if (!data->cbr && decoder->cbr) {
790 data->cbr = decoder->cbr;
791 data->cbr_cyc_to_tsc = decoder->cbr_cyc_to_tsc;
794 if (!data->cycle_cnt)
797 cyc_to_tsc = (double)(timestamp - decoder->timestamp) / data->cycle_cnt;
799 if (data->cbr && cyc_to_tsc > data->cbr_cyc_to_tsc &&
800 cyc_to_tsc / data->cbr_cyc_to_tsc > 1.25) {
801 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle too big (c.f. CBR-based value %g), pos " x64_fmt "\n",
802 cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
806 decoder->calc_cyc_to_tsc = cyc_to_tsc;
807 decoder->have_calc_cyc_to_tsc = true;
810 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. CBR-based value %g, pos " x64_fmt "\n",
811 cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
813 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. unknown CBR-based value, pos " x64_fmt "\n",
814 cyc_to_tsc, pkt_info->pos);
820 static void intel_pt_calc_cyc_to_tsc(struct intel_pt_decoder *decoder,
823 struct intel_pt_calc_cyc_to_tsc_info data = {
826 .last_mtc = decoder->last_mtc,
827 .ctc_timestamp = decoder->ctc_timestamp,
828 .ctc_delta = decoder->ctc_delta,
829 .tsc_timestamp = decoder->tsc_timestamp,
830 .timestamp = decoder->timestamp,
831 .have_tma = decoder->have_tma,
832 .fixup_last_mtc = decoder->fixup_last_mtc,
833 .from_mtc = from_mtc,
838 * For now, do not support using TSC packets for at least the reasons:
839 * 1) timing might have stopped
840 * 2) TSC packets within PSB+ can slip against CYC packets
845 intel_pt_pkt_lookahead(decoder, intel_pt_calc_cyc_cb, &data);
848 static int intel_pt_get_next_packet(struct intel_pt_decoder *decoder)
852 decoder->last_packet_type = decoder->packet.type;
855 decoder->pos += decoder->pkt_step;
856 decoder->buf += decoder->pkt_step;
857 decoder->len -= decoder->pkt_step;
860 ret = intel_pt_get_next_data(decoder);
865 ret = intel_pt_get_packet(decoder->buf, decoder->len,
867 if (ret == INTEL_PT_NEED_MORE_BYTES && BITS_PER_LONG == 32 &&
868 decoder->len < INTEL_PT_PKT_MAX_SZ && !decoder->next_buf) {
869 ret = intel_pt_get_split_packet(decoder);
874 return intel_pt_bad_packet(decoder);
876 decoder->pkt_len = ret;
877 decoder->pkt_step = ret;
878 intel_pt_decoder_log_packet(decoder);
879 } while (decoder->packet.type == INTEL_PT_PAD);
884 static uint64_t intel_pt_next_period(struct intel_pt_decoder *decoder)
886 uint64_t timestamp, masked_timestamp;
888 timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
889 masked_timestamp = timestamp & decoder->period_mask;
890 if (decoder->continuous_period) {
891 if (masked_timestamp != decoder->last_masked_timestamp)
895 masked_timestamp = timestamp & decoder->period_mask;
896 if (masked_timestamp != decoder->last_masked_timestamp) {
897 decoder->last_masked_timestamp = masked_timestamp;
898 decoder->continuous_period = true;
901 return decoder->period_ticks - (timestamp - masked_timestamp);
904 static uint64_t intel_pt_next_sample(struct intel_pt_decoder *decoder)
906 switch (decoder->period_type) {
907 case INTEL_PT_PERIOD_INSTRUCTIONS:
908 return decoder->period - decoder->period_insn_cnt;
909 case INTEL_PT_PERIOD_TICKS:
910 return intel_pt_next_period(decoder);
911 case INTEL_PT_PERIOD_NONE:
912 case INTEL_PT_PERIOD_MTC:
918 static void intel_pt_sample_insn(struct intel_pt_decoder *decoder)
920 uint64_t timestamp, masked_timestamp;
922 switch (decoder->period_type) {
923 case INTEL_PT_PERIOD_INSTRUCTIONS:
924 decoder->period_insn_cnt = 0;
926 case INTEL_PT_PERIOD_TICKS:
927 timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
928 masked_timestamp = timestamp & decoder->period_mask;
929 decoder->last_masked_timestamp = masked_timestamp;
931 case INTEL_PT_PERIOD_NONE:
932 case INTEL_PT_PERIOD_MTC:
937 decoder->state.type |= INTEL_PT_INSTRUCTION;
940 static int intel_pt_walk_insn(struct intel_pt_decoder *decoder,
941 struct intel_pt_insn *intel_pt_insn, uint64_t ip)
943 uint64_t max_insn_cnt, insn_cnt = 0;
946 if (!decoder->mtc_insn)
947 decoder->mtc_insn = true;
949 max_insn_cnt = intel_pt_next_sample(decoder);
951 err = decoder->walk_insn(intel_pt_insn, &insn_cnt, &decoder->ip, ip,
952 max_insn_cnt, decoder->data);
954 decoder->tot_insn_cnt += insn_cnt;
955 decoder->timestamp_insn_cnt += insn_cnt;
956 decoder->sample_insn_cnt += insn_cnt;
957 decoder->period_insn_cnt += insn_cnt;
960 decoder->no_progress = 0;
961 decoder->pkt_state = INTEL_PT_STATE_ERR2;
962 intel_pt_log_at("ERROR: Failed to get instruction",
969 if (ip && decoder->ip == ip) {
974 if (max_insn_cnt && insn_cnt >= max_insn_cnt)
975 intel_pt_sample_insn(decoder);
977 if (intel_pt_insn->branch == INTEL_PT_BR_NO_BRANCH) {
978 decoder->state.type = INTEL_PT_INSTRUCTION;
979 decoder->state.from_ip = decoder->ip;
980 decoder->state.to_ip = 0;
981 decoder->ip += intel_pt_insn->length;
982 err = INTEL_PT_RETURN;
986 if (intel_pt_insn->op == INTEL_PT_OP_CALL) {
987 /* Zero-length calls are excluded */
988 if (intel_pt_insn->branch != INTEL_PT_BR_UNCONDITIONAL ||
989 intel_pt_insn->rel) {
990 err = intel_pt_push(&decoder->stack, decoder->ip +
991 intel_pt_insn->length);
995 } else if (intel_pt_insn->op == INTEL_PT_OP_RET) {
996 decoder->ret_addr = intel_pt_pop(&decoder->stack);
999 if (intel_pt_insn->branch == INTEL_PT_BR_UNCONDITIONAL) {
1000 int cnt = decoder->no_progress++;
1002 decoder->state.from_ip = decoder->ip;
1003 decoder->ip += intel_pt_insn->length +
1005 decoder->state.to_ip = decoder->ip;
1006 err = INTEL_PT_RETURN;
1009 * Check for being stuck in a loop. This can happen if a
1010 * decoder error results in the decoder erroneously setting the
1011 * ip to an address that is itself in an infinite loop that
1012 * consumes no packets. When that happens, there must be an
1013 * unconditional branch.
1017 decoder->stuck_ip = decoder->state.to_ip;
1018 decoder->stuck_ip_prd = 1;
1019 decoder->stuck_ip_cnt = 1;
1020 } else if (cnt > INTEL_PT_MAX_LOOPS ||
1021 decoder->state.to_ip == decoder->stuck_ip) {
1022 intel_pt_log_at("ERROR: Never-ending loop",
1023 decoder->state.to_ip);
1024 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1027 } else if (!--decoder->stuck_ip_cnt) {
1028 decoder->stuck_ip_prd += 1;
1029 decoder->stuck_ip_cnt = decoder->stuck_ip_prd;
1030 decoder->stuck_ip = decoder->state.to_ip;
1033 goto out_no_progress;
1036 decoder->no_progress = 0;
1038 decoder->state.insn_op = intel_pt_insn->op;
1039 decoder->state.insn_len = intel_pt_insn->length;
1040 memcpy(decoder->state.insn, intel_pt_insn->buf,
1041 INTEL_PT_INSN_BUF_SZ);
1043 if (decoder->tx_flags & INTEL_PT_IN_TX)
1044 decoder->state.flags |= INTEL_PT_IN_TX;
1049 static bool intel_pt_fup_event(struct intel_pt_decoder *decoder)
1053 if (decoder->set_fup_tx_flags) {
1054 decoder->set_fup_tx_flags = false;
1055 decoder->tx_flags = decoder->fup_tx_flags;
1056 decoder->state.type = INTEL_PT_TRANSACTION;
1057 decoder->state.from_ip = decoder->ip;
1058 decoder->state.to_ip = 0;
1059 decoder->state.flags = decoder->fup_tx_flags;
1062 if (decoder->set_fup_ptw) {
1063 decoder->set_fup_ptw = false;
1064 decoder->state.type = INTEL_PT_PTW;
1065 decoder->state.flags |= INTEL_PT_FUP_IP;
1066 decoder->state.from_ip = decoder->ip;
1067 decoder->state.to_ip = 0;
1068 decoder->state.ptw_payload = decoder->fup_ptw_payload;
1071 if (decoder->set_fup_mwait) {
1072 decoder->set_fup_mwait = false;
1073 decoder->state.type = INTEL_PT_MWAIT_OP;
1074 decoder->state.from_ip = decoder->ip;
1075 decoder->state.to_ip = 0;
1076 decoder->state.mwait_payload = decoder->fup_mwait_payload;
1079 if (decoder->set_fup_pwre) {
1080 decoder->set_fup_pwre = false;
1081 decoder->state.type |= INTEL_PT_PWR_ENTRY;
1082 decoder->state.type &= ~INTEL_PT_BRANCH;
1083 decoder->state.from_ip = decoder->ip;
1084 decoder->state.to_ip = 0;
1085 decoder->state.pwre_payload = decoder->fup_pwre_payload;
1088 if (decoder->set_fup_exstop) {
1089 decoder->set_fup_exstop = false;
1090 decoder->state.type |= INTEL_PT_EX_STOP;
1091 decoder->state.type &= ~INTEL_PT_BRANCH;
1092 decoder->state.flags |= INTEL_PT_FUP_IP;
1093 decoder->state.from_ip = decoder->ip;
1094 decoder->state.to_ip = 0;
1100 static inline bool intel_pt_fup_with_nlip(struct intel_pt_decoder *decoder,
1101 struct intel_pt_insn *intel_pt_insn,
1102 uint64_t ip, int err)
1104 return decoder->flags & INTEL_PT_FUP_WITH_NLIP && !err &&
1105 intel_pt_insn->branch == INTEL_PT_BR_INDIRECT &&
1106 ip == decoder->ip + intel_pt_insn->length;
1109 static int intel_pt_walk_fup(struct intel_pt_decoder *decoder)
1111 struct intel_pt_insn intel_pt_insn;
1115 ip = decoder->last_ip;
1118 err = intel_pt_walk_insn(decoder, &intel_pt_insn, ip);
1119 if (err == INTEL_PT_RETURN)
1121 if (err == -EAGAIN ||
1122 intel_pt_fup_with_nlip(decoder, &intel_pt_insn, ip, err)) {
1123 if (intel_pt_fup_event(decoder))
1127 decoder->set_fup_tx_flags = false;
1131 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1132 intel_pt_log_at("ERROR: Unexpected indirect branch",
1134 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1138 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1139 intel_pt_log_at("ERROR: Unexpected conditional branch",
1141 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1145 intel_pt_bug(decoder);
1149 static int intel_pt_walk_tip(struct intel_pt_decoder *decoder)
1151 struct intel_pt_insn intel_pt_insn;
1154 err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1155 if (err == INTEL_PT_RETURN &&
1157 decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1158 (decoder->state.type & INTEL_PT_BRANCH) &&
1159 decoder->pgd_ip(decoder->state.to_ip, decoder->data)) {
1160 /* Unconditional branch leaving filter region */
1161 decoder->no_progress = 0;
1162 decoder->pge = false;
1163 decoder->continuous_period = false;
1164 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1165 decoder->state.type |= INTEL_PT_TRACE_END;
1168 if (err == INTEL_PT_RETURN)
1173 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1174 if (decoder->pkt_state == INTEL_PT_STATE_TIP_PGD) {
1175 decoder->pge = false;
1176 decoder->continuous_period = false;
1177 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1178 decoder->state.from_ip = decoder->ip;
1179 if (decoder->packet.count == 0) {
1180 decoder->state.to_ip = 0;
1182 decoder->state.to_ip = decoder->last_ip;
1183 decoder->ip = decoder->last_ip;
1185 decoder->state.type |= INTEL_PT_TRACE_END;
1187 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1188 decoder->state.from_ip = decoder->ip;
1189 if (decoder->packet.count == 0) {
1190 decoder->state.to_ip = 0;
1192 decoder->state.to_ip = decoder->last_ip;
1193 decoder->ip = decoder->last_ip;
1199 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1200 uint64_t to_ip = decoder->ip + intel_pt_insn.length +
1203 if (decoder->pgd_ip &&
1204 decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1205 decoder->pgd_ip(to_ip, decoder->data)) {
1206 /* Conditional branch leaving filter region */
1207 decoder->pge = false;
1208 decoder->continuous_period = false;
1209 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1210 decoder->ip = to_ip;
1211 decoder->state.from_ip = decoder->ip;
1212 decoder->state.to_ip = to_ip;
1213 decoder->state.type |= INTEL_PT_TRACE_END;
1216 intel_pt_log_at("ERROR: Conditional branch when expecting indirect branch",
1218 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1222 return intel_pt_bug(decoder);
1225 static int intel_pt_walk_tnt(struct intel_pt_decoder *decoder)
1227 struct intel_pt_insn intel_pt_insn;
1231 err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1232 if (err == INTEL_PT_RETURN)
1237 if (intel_pt_insn.op == INTEL_PT_OP_RET) {
1238 if (!decoder->return_compression) {
1239 intel_pt_log_at("ERROR: RET when expecting conditional branch",
1241 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1244 if (!decoder->ret_addr) {
1245 intel_pt_log_at("ERROR: Bad RET compression (stack empty)",
1247 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1250 if (!(decoder->tnt.payload & BIT63)) {
1251 intel_pt_log_at("ERROR: Bad RET compression (TNT=N)",
1253 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1256 decoder->tnt.count -= 1;
1257 if (!decoder->tnt.count)
1258 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1259 decoder->tnt.payload <<= 1;
1260 decoder->state.from_ip = decoder->ip;
1261 decoder->ip = decoder->ret_addr;
1262 decoder->state.to_ip = decoder->ip;
1266 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1267 /* Handle deferred TIPs */
1268 err = intel_pt_get_next_packet(decoder);
1271 if (decoder->packet.type != INTEL_PT_TIP ||
1272 decoder->packet.count == 0) {
1273 intel_pt_log_at("ERROR: Missing deferred TIP for indirect branch",
1275 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1276 decoder->pkt_step = 0;
1279 intel_pt_set_last_ip(decoder);
1280 decoder->state.from_ip = decoder->ip;
1281 decoder->state.to_ip = decoder->last_ip;
1282 decoder->ip = decoder->last_ip;
1286 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1287 decoder->tnt.count -= 1;
1288 if (!decoder->tnt.count)
1289 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1290 if (decoder->tnt.payload & BIT63) {
1291 decoder->tnt.payload <<= 1;
1292 decoder->state.from_ip = decoder->ip;
1293 decoder->ip += intel_pt_insn.length +
1295 decoder->state.to_ip = decoder->ip;
1298 /* Instruction sample for a non-taken branch */
1299 if (decoder->state.type & INTEL_PT_INSTRUCTION) {
1300 decoder->tnt.payload <<= 1;
1301 decoder->state.type = INTEL_PT_INSTRUCTION;
1302 decoder->state.from_ip = decoder->ip;
1303 decoder->state.to_ip = 0;
1304 decoder->ip += intel_pt_insn.length;
1307 decoder->ip += intel_pt_insn.length;
1308 if (!decoder->tnt.count)
1310 decoder->tnt.payload <<= 1;
1314 return intel_pt_bug(decoder);
1318 static int intel_pt_mode_tsx(struct intel_pt_decoder *decoder, bool *no_tip)
1320 unsigned int fup_tx_flags;
1323 fup_tx_flags = decoder->packet.payload &
1324 (INTEL_PT_IN_TX | INTEL_PT_ABORT_TX);
1325 err = intel_pt_get_next_packet(decoder);
1328 if (decoder->packet.type == INTEL_PT_FUP) {
1329 decoder->fup_tx_flags = fup_tx_flags;
1330 decoder->set_fup_tx_flags = true;
1331 if (!(decoder->fup_tx_flags & INTEL_PT_ABORT_TX))
1334 intel_pt_log_at("ERROR: Missing FUP after MODE.TSX",
1336 intel_pt_update_in_tx(decoder);
1341 static void intel_pt_calc_tsc_timestamp(struct intel_pt_decoder *decoder)
1345 decoder->have_tma = false;
1347 if (decoder->ref_timestamp) {
1348 timestamp = decoder->packet.payload |
1349 (decoder->ref_timestamp & (0xffULL << 56));
1350 if (timestamp < decoder->ref_timestamp) {
1351 if (decoder->ref_timestamp - timestamp > (1ULL << 55))
1352 timestamp += (1ULL << 56);
1354 if (timestamp - decoder->ref_timestamp > (1ULL << 55))
1355 timestamp -= (1ULL << 56);
1357 decoder->tsc_timestamp = timestamp;
1358 decoder->timestamp = timestamp;
1359 decoder->ref_timestamp = 0;
1360 decoder->timestamp_insn_cnt = 0;
1361 } else if (decoder->timestamp) {
1362 timestamp = decoder->packet.payload |
1363 (decoder->timestamp & (0xffULL << 56));
1364 decoder->tsc_timestamp = timestamp;
1365 if (timestamp < decoder->timestamp &&
1366 decoder->timestamp - timestamp < decoder->tsc_slip) {
1367 intel_pt_log_to("Suppressing backwards timestamp",
1369 timestamp = decoder->timestamp;
1371 if (timestamp < decoder->timestamp) {
1372 intel_pt_log_to("Wraparound timestamp", timestamp);
1373 timestamp += (1ULL << 56);
1374 decoder->tsc_timestamp = timestamp;
1376 decoder->timestamp = timestamp;
1377 decoder->timestamp_insn_cnt = 0;
1380 if (decoder->last_packet_type == INTEL_PT_CYC) {
1381 decoder->cyc_ref_timestamp = decoder->timestamp;
1382 decoder->cycle_cnt = 0;
1383 decoder->have_calc_cyc_to_tsc = false;
1384 intel_pt_calc_cyc_to_tsc(decoder, false);
1387 intel_pt_log_to("Setting timestamp", decoder->timestamp);
1390 static int intel_pt_overflow(struct intel_pt_decoder *decoder)
1392 intel_pt_log("ERROR: Buffer overflow\n");
1393 intel_pt_clear_tx_flags(decoder);
1394 decoder->timestamp_insn_cnt = 0;
1395 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1396 decoder->overflow = true;
1400 static void intel_pt_calc_tma(struct intel_pt_decoder *decoder)
1402 uint32_t ctc = decoder->packet.payload;
1403 uint32_t fc = decoder->packet.count;
1404 uint32_t ctc_rem = ctc & decoder->ctc_rem_mask;
1406 if (!decoder->tsc_ctc_ratio_d)
1409 decoder->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
1410 decoder->ctc_timestamp = decoder->tsc_timestamp - fc;
1411 if (decoder->tsc_ctc_mult) {
1412 decoder->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
1414 decoder->ctc_timestamp -= multdiv(ctc_rem,
1415 decoder->tsc_ctc_ratio_n,
1416 decoder->tsc_ctc_ratio_d);
1418 decoder->ctc_delta = 0;
1419 decoder->have_tma = true;
1420 decoder->fixup_last_mtc = true;
1421 intel_pt_log("CTC timestamp " x64_fmt " last MTC %#x CTC rem %#x\n",
1422 decoder->ctc_timestamp, decoder->last_mtc, ctc_rem);
1425 static void intel_pt_calc_mtc_timestamp(struct intel_pt_decoder *decoder)
1428 uint32_t mtc, mtc_delta;
1430 if (!decoder->have_tma)
1433 mtc = decoder->packet.payload;
1435 if (decoder->mtc_shift > 8 && decoder->fixup_last_mtc) {
1436 decoder->fixup_last_mtc = false;
1437 intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
1438 &decoder->last_mtc);
1441 if (mtc > decoder->last_mtc)
1442 mtc_delta = mtc - decoder->last_mtc;
1444 mtc_delta = mtc + 256 - decoder->last_mtc;
1446 decoder->ctc_delta += mtc_delta << decoder->mtc_shift;
1448 if (decoder->tsc_ctc_mult) {
1449 timestamp = decoder->ctc_timestamp +
1450 decoder->ctc_delta * decoder->tsc_ctc_mult;
1452 timestamp = decoder->ctc_timestamp +
1453 multdiv(decoder->ctc_delta,
1454 decoder->tsc_ctc_ratio_n,
1455 decoder->tsc_ctc_ratio_d);
1458 if (timestamp < decoder->timestamp)
1459 intel_pt_log("Suppressing MTC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1460 timestamp, decoder->timestamp);
1462 decoder->timestamp = timestamp;
1464 decoder->timestamp_insn_cnt = 0;
1465 decoder->last_mtc = mtc;
1467 if (decoder->last_packet_type == INTEL_PT_CYC) {
1468 decoder->cyc_ref_timestamp = decoder->timestamp;
1469 decoder->cycle_cnt = 0;
1470 decoder->have_calc_cyc_to_tsc = false;
1471 intel_pt_calc_cyc_to_tsc(decoder, true);
1474 intel_pt_log_to("Setting timestamp", decoder->timestamp);
1477 static void intel_pt_calc_cbr(struct intel_pt_decoder *decoder)
1479 unsigned int cbr = decoder->packet.payload & 0xff;
1481 decoder->cbr_payload = decoder->packet.payload;
1483 if (decoder->cbr == cbr)
1487 decoder->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
1490 static void intel_pt_calc_cyc_timestamp(struct intel_pt_decoder *decoder)
1492 uint64_t timestamp = decoder->cyc_ref_timestamp;
1494 decoder->have_cyc = true;
1496 decoder->cycle_cnt += decoder->packet.payload;
1498 if (!decoder->cyc_ref_timestamp)
1501 if (decoder->have_calc_cyc_to_tsc)
1502 timestamp += decoder->cycle_cnt * decoder->calc_cyc_to_tsc;
1503 else if (decoder->cbr)
1504 timestamp += decoder->cycle_cnt * decoder->cbr_cyc_to_tsc;
1508 if (timestamp < decoder->timestamp)
1509 intel_pt_log("Suppressing CYC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1510 timestamp, decoder->timestamp);
1512 decoder->timestamp = timestamp;
1514 decoder->timestamp_insn_cnt = 0;
1516 intel_pt_log_to("Setting timestamp", decoder->timestamp);
1519 /* Walk PSB+ packets when already in sync. */
1520 static int intel_pt_walk_psbend(struct intel_pt_decoder *decoder)
1525 err = intel_pt_get_next_packet(decoder);
1529 switch (decoder->packet.type) {
1530 case INTEL_PT_PSBEND:
1533 case INTEL_PT_TIP_PGD:
1534 case INTEL_PT_TIP_PGE:
1537 case INTEL_PT_TRACESTOP:
1540 case INTEL_PT_PTWRITE:
1541 case INTEL_PT_PTWRITE_IP:
1542 case INTEL_PT_EXSTOP:
1543 case INTEL_PT_EXSTOP_IP:
1544 case INTEL_PT_MWAIT:
1547 decoder->have_tma = false;
1548 intel_pt_log("ERROR: Unexpected packet\n");
1552 return intel_pt_overflow(decoder);
1555 intel_pt_calc_tsc_timestamp(decoder);
1559 intel_pt_calc_tma(decoder);
1563 intel_pt_calc_cbr(decoder);
1566 case INTEL_PT_MODE_EXEC:
1567 decoder->exec_mode = decoder->packet.payload;
1571 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1575 decoder->pge = true;
1576 if (decoder->packet.count)
1577 intel_pt_set_last_ip(decoder);
1580 case INTEL_PT_MODE_TSX:
1581 intel_pt_update_in_tx(decoder);
1585 intel_pt_calc_mtc_timestamp(decoder);
1586 if (decoder->period_type == INTEL_PT_PERIOD_MTC)
1587 decoder->state.type |= INTEL_PT_INSTRUCTION;
1600 static int intel_pt_walk_fup_tip(struct intel_pt_decoder *decoder)
1604 if (decoder->tx_flags & INTEL_PT_ABORT_TX) {
1605 decoder->tx_flags = 0;
1606 decoder->state.flags &= ~INTEL_PT_IN_TX;
1607 decoder->state.flags |= INTEL_PT_ABORT_TX;
1609 decoder->state.flags |= INTEL_PT_ASYNC;
1613 err = intel_pt_get_next_packet(decoder);
1617 switch (decoder->packet.type) {
1620 case INTEL_PT_TRACESTOP:
1624 case INTEL_PT_MODE_TSX:
1626 case INTEL_PT_PSBEND:
1627 case INTEL_PT_PTWRITE:
1628 case INTEL_PT_PTWRITE_IP:
1629 case INTEL_PT_EXSTOP:
1630 case INTEL_PT_EXSTOP_IP:
1631 case INTEL_PT_MWAIT:
1634 intel_pt_log("ERROR: Missing TIP after FUP\n");
1635 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1636 decoder->pkt_step = 0;
1640 intel_pt_calc_cbr(decoder);
1644 return intel_pt_overflow(decoder);
1646 case INTEL_PT_TIP_PGD:
1647 decoder->state.from_ip = decoder->ip;
1648 if (decoder->packet.count == 0) {
1649 decoder->state.to_ip = 0;
1651 intel_pt_set_ip(decoder);
1652 decoder->state.to_ip = decoder->ip;
1654 decoder->pge = false;
1655 decoder->continuous_period = false;
1656 decoder->state.type |= INTEL_PT_TRACE_END;
1659 case INTEL_PT_TIP_PGE:
1660 decoder->pge = true;
1661 intel_pt_log("Omitting PGE ip " x64_fmt "\n",
1663 decoder->state.from_ip = 0;
1664 if (decoder->packet.count == 0) {
1665 decoder->state.to_ip = 0;
1667 intel_pt_set_ip(decoder);
1668 decoder->state.to_ip = decoder->ip;
1670 decoder->state.type |= INTEL_PT_TRACE_BEGIN;
1674 decoder->state.from_ip = decoder->ip;
1675 if (decoder->packet.count == 0) {
1676 decoder->state.to_ip = 0;
1678 intel_pt_set_ip(decoder);
1679 decoder->state.to_ip = decoder->ip;
1684 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1688 intel_pt_calc_mtc_timestamp(decoder);
1689 if (decoder->period_type == INTEL_PT_PERIOD_MTC)
1690 decoder->state.type |= INTEL_PT_INSTRUCTION;
1694 intel_pt_calc_cyc_timestamp(decoder);
1697 case INTEL_PT_MODE_EXEC:
1698 decoder->exec_mode = decoder->packet.payload;
1707 return intel_pt_bug(decoder);
1712 static int intel_pt_walk_trace(struct intel_pt_decoder *decoder)
1714 bool no_tip = false;
1718 err = intel_pt_get_next_packet(decoder);
1722 switch (decoder->packet.type) {
1724 if (!decoder->packet.count)
1726 decoder->tnt = decoder->packet;
1727 decoder->pkt_state = INTEL_PT_STATE_TNT;
1728 err = intel_pt_walk_tnt(decoder);
1733 case INTEL_PT_TIP_PGD:
1734 if (decoder->packet.count != 0)
1735 intel_pt_set_last_ip(decoder);
1736 decoder->pkt_state = INTEL_PT_STATE_TIP_PGD;
1737 return intel_pt_walk_tip(decoder);
1739 case INTEL_PT_TIP_PGE: {
1740 decoder->pge = true;
1741 if (decoder->packet.count == 0) {
1742 intel_pt_log_at("Skipping zero TIP.PGE",
1746 intel_pt_set_ip(decoder);
1747 decoder->state.from_ip = 0;
1748 decoder->state.to_ip = decoder->ip;
1749 decoder->state.type |= INTEL_PT_TRACE_BEGIN;
1754 return intel_pt_overflow(decoder);
1757 if (decoder->packet.count != 0)
1758 intel_pt_set_last_ip(decoder);
1759 decoder->pkt_state = INTEL_PT_STATE_TIP;
1760 return intel_pt_walk_tip(decoder);
1763 if (decoder->packet.count == 0) {
1764 intel_pt_log_at("Skipping zero FUP",
1769 intel_pt_set_last_ip(decoder);
1770 if (!decoder->branch_enable) {
1771 decoder->ip = decoder->last_ip;
1772 if (intel_pt_fup_event(decoder))
1777 if (decoder->set_fup_mwait)
1779 err = intel_pt_walk_fup(decoder);
1780 if (err != -EAGAIN) {
1784 decoder->pkt_state =
1785 INTEL_PT_STATE_FUP_NO_TIP;
1787 decoder->pkt_state = INTEL_PT_STATE_FUP;
1794 return intel_pt_walk_fup_tip(decoder);
1796 case INTEL_PT_TRACESTOP:
1797 decoder->pge = false;
1798 decoder->continuous_period = false;
1799 intel_pt_clear_tx_flags(decoder);
1800 decoder->have_tma = false;
1804 decoder->last_ip = 0;
1805 decoder->have_last_ip = true;
1806 intel_pt_clear_stack(&decoder->stack);
1807 err = intel_pt_walk_psbend(decoder);
1815 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1819 intel_pt_calc_mtc_timestamp(decoder);
1820 if (decoder->period_type != INTEL_PT_PERIOD_MTC)
1823 * Ensure that there has been an instruction since the
1826 if (!decoder->mtc_insn)
1828 decoder->mtc_insn = false;
1829 /* Ensure that there is a timestamp */
1830 if (!decoder->timestamp)
1832 decoder->state.type = INTEL_PT_INSTRUCTION;
1833 decoder->state.from_ip = decoder->ip;
1834 decoder->state.to_ip = 0;
1835 decoder->mtc_insn = false;
1839 intel_pt_calc_tsc_timestamp(decoder);
1843 intel_pt_calc_tma(decoder);
1847 intel_pt_calc_cyc_timestamp(decoder);
1851 intel_pt_calc_cbr(decoder);
1852 if (!decoder->branch_enable &&
1853 decoder->cbr != decoder->cbr_seen) {
1854 decoder->cbr_seen = decoder->cbr;
1855 decoder->state.type = INTEL_PT_CBR_CHG;
1856 decoder->state.from_ip = decoder->ip;
1857 decoder->state.to_ip = 0;
1858 decoder->state.cbr_payload =
1859 decoder->packet.payload;
1864 case INTEL_PT_MODE_EXEC:
1865 decoder->exec_mode = decoder->packet.payload;
1868 case INTEL_PT_MODE_TSX:
1869 /* MODE_TSX need not be followed by FUP */
1870 if (!decoder->pge) {
1871 intel_pt_update_in_tx(decoder);
1874 err = intel_pt_mode_tsx(decoder, &no_tip);
1879 case INTEL_PT_BAD: /* Does not happen */
1880 return intel_pt_bug(decoder);
1882 case INTEL_PT_PSBEND:
1888 case INTEL_PT_PTWRITE_IP:
1889 decoder->fup_ptw_payload = decoder->packet.payload;
1890 err = intel_pt_get_next_packet(decoder);
1893 if (decoder->packet.type == INTEL_PT_FUP) {
1894 decoder->set_fup_ptw = true;
1897 intel_pt_log_at("ERROR: Missing FUP after PTWRITE",
1902 case INTEL_PT_PTWRITE:
1903 decoder->state.type = INTEL_PT_PTW;
1904 decoder->state.from_ip = decoder->ip;
1905 decoder->state.to_ip = 0;
1906 decoder->state.ptw_payload = decoder->packet.payload;
1909 case INTEL_PT_MWAIT:
1910 decoder->fup_mwait_payload = decoder->packet.payload;
1911 decoder->set_fup_mwait = true;
1915 if (decoder->set_fup_mwait) {
1916 decoder->fup_pwre_payload =
1917 decoder->packet.payload;
1918 decoder->set_fup_pwre = true;
1921 decoder->state.type = INTEL_PT_PWR_ENTRY;
1922 decoder->state.from_ip = decoder->ip;
1923 decoder->state.to_ip = 0;
1924 decoder->state.pwrx_payload = decoder->packet.payload;
1927 case INTEL_PT_EXSTOP_IP:
1928 err = intel_pt_get_next_packet(decoder);
1931 if (decoder->packet.type == INTEL_PT_FUP) {
1932 decoder->set_fup_exstop = true;
1935 intel_pt_log_at("ERROR: Missing FUP after EXSTOP",
1940 case INTEL_PT_EXSTOP:
1941 decoder->state.type = INTEL_PT_EX_STOP;
1942 decoder->state.from_ip = decoder->ip;
1943 decoder->state.to_ip = 0;
1947 decoder->state.type = INTEL_PT_PWR_EXIT;
1948 decoder->state.from_ip = decoder->ip;
1949 decoder->state.to_ip = 0;
1950 decoder->state.pwrx_payload = decoder->packet.payload;
1954 return intel_pt_bug(decoder);
1959 static inline bool intel_pt_have_ip(struct intel_pt_decoder *decoder)
1961 return decoder->packet.count &&
1962 (decoder->have_last_ip || decoder->packet.count == 3 ||
1963 decoder->packet.count == 6);
1966 /* Walk PSB+ packets to get in sync. */
1967 static int intel_pt_walk_psb(struct intel_pt_decoder *decoder)
1972 err = intel_pt_get_next_packet(decoder);
1976 switch (decoder->packet.type) {
1977 case INTEL_PT_TIP_PGD:
1978 decoder->continuous_period = false;
1980 case INTEL_PT_TIP_PGE:
1982 case INTEL_PT_PTWRITE:
1983 case INTEL_PT_PTWRITE_IP:
1984 case INTEL_PT_EXSTOP:
1985 case INTEL_PT_EXSTOP_IP:
1986 case INTEL_PT_MWAIT:
1989 intel_pt_log("ERROR: Unexpected packet\n");
1993 decoder->pge = true;
1994 if (intel_pt_have_ip(decoder)) {
1995 uint64_t current_ip = decoder->ip;
1997 intel_pt_set_ip(decoder);
1999 intel_pt_log_to("Setting IP",
2005 intel_pt_calc_mtc_timestamp(decoder);
2009 intel_pt_calc_tsc_timestamp(decoder);
2013 intel_pt_calc_tma(decoder);
2017 intel_pt_calc_cyc_timestamp(decoder);
2021 intel_pt_calc_cbr(decoder);
2025 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
2028 case INTEL_PT_MODE_EXEC:
2029 decoder->exec_mode = decoder->packet.payload;
2032 case INTEL_PT_MODE_TSX:
2033 intel_pt_update_in_tx(decoder);
2036 case INTEL_PT_TRACESTOP:
2037 decoder->pge = false;
2038 decoder->continuous_period = false;
2039 intel_pt_clear_tx_flags(decoder);
2043 decoder->have_tma = false;
2044 intel_pt_log("ERROR: Unexpected packet\n");
2046 decoder->pkt_state = INTEL_PT_STATE_ERR4;
2048 decoder->pkt_state = INTEL_PT_STATE_ERR3;
2051 case INTEL_PT_BAD: /* Does not happen */
2052 return intel_pt_bug(decoder);
2055 return intel_pt_overflow(decoder);
2057 case INTEL_PT_PSBEND:
2070 static int intel_pt_walk_to_ip(struct intel_pt_decoder *decoder)
2075 err = intel_pt_get_next_packet(decoder);
2079 switch (decoder->packet.type) {
2080 case INTEL_PT_TIP_PGD:
2081 decoder->continuous_period = false;
2083 case INTEL_PT_TIP_PGE:
2085 decoder->pge = decoder->packet.type != INTEL_PT_TIP_PGD;
2086 if (intel_pt_have_ip(decoder))
2087 intel_pt_set_ip(decoder);
2090 if (decoder->packet.type == INTEL_PT_TIP_PGE)
2091 decoder->state.type |= INTEL_PT_TRACE_BEGIN;
2092 if (decoder->packet.type == INTEL_PT_TIP_PGD)
2093 decoder->state.type |= INTEL_PT_TRACE_END;
2097 if (intel_pt_have_ip(decoder))
2098 intel_pt_set_ip(decoder);
2104 intel_pt_calc_mtc_timestamp(decoder);
2108 intel_pt_calc_tsc_timestamp(decoder);
2112 intel_pt_calc_tma(decoder);
2116 intel_pt_calc_cyc_timestamp(decoder);
2120 intel_pt_calc_cbr(decoder);
2124 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
2127 case INTEL_PT_MODE_EXEC:
2128 decoder->exec_mode = decoder->packet.payload;
2131 case INTEL_PT_MODE_TSX:
2132 intel_pt_update_in_tx(decoder);
2136 return intel_pt_overflow(decoder);
2138 case INTEL_PT_BAD: /* Does not happen */
2139 return intel_pt_bug(decoder);
2141 case INTEL_PT_TRACESTOP:
2142 decoder->pge = false;
2143 decoder->continuous_period = false;
2144 intel_pt_clear_tx_flags(decoder);
2145 decoder->have_tma = false;
2149 decoder->last_ip = 0;
2150 decoder->have_last_ip = true;
2151 intel_pt_clear_stack(&decoder->stack);
2152 err = intel_pt_walk_psb(decoder);
2156 /* Do not have a sample */
2157 decoder->state.type = 0;
2163 case INTEL_PT_PSBEND:
2167 case INTEL_PT_PTWRITE:
2168 case INTEL_PT_PTWRITE_IP:
2169 case INTEL_PT_EXSTOP:
2170 case INTEL_PT_EXSTOP_IP:
2171 case INTEL_PT_MWAIT:
2180 static int intel_pt_sync_ip(struct intel_pt_decoder *decoder)
2184 decoder->set_fup_tx_flags = false;
2185 decoder->set_fup_ptw = false;
2186 decoder->set_fup_mwait = false;
2187 decoder->set_fup_pwre = false;
2188 decoder->set_fup_exstop = false;
2190 if (!decoder->branch_enable) {
2191 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2192 decoder->overflow = false;
2193 decoder->state.type = 0; /* Do not have a sample */
2197 intel_pt_log("Scanning for full IP\n");
2198 err = intel_pt_walk_to_ip(decoder);
2202 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2203 decoder->overflow = false;
2205 decoder->state.from_ip = 0;
2206 decoder->state.to_ip = decoder->ip;
2207 intel_pt_log_to("Setting IP", decoder->ip);
2212 static int intel_pt_part_psb(struct intel_pt_decoder *decoder)
2214 const unsigned char *end = decoder->buf + decoder->len;
2217 for (i = INTEL_PT_PSB_LEN - 1; i; i--) {
2218 if (i > decoder->len)
2220 if (!memcmp(end - i, INTEL_PT_PSB_STR, i))
2226 static int intel_pt_rest_psb(struct intel_pt_decoder *decoder, int part_psb)
2228 size_t rest_psb = INTEL_PT_PSB_LEN - part_psb;
2229 const char *psb = INTEL_PT_PSB_STR;
2231 if (rest_psb > decoder->len ||
2232 memcmp(decoder->buf, psb + part_psb, rest_psb))
2238 static int intel_pt_get_split_psb(struct intel_pt_decoder *decoder,
2243 decoder->pos += decoder->len;
2246 ret = intel_pt_get_next_data(decoder);
2250 rest_psb = intel_pt_rest_psb(decoder, part_psb);
2254 decoder->pos -= part_psb;
2255 decoder->next_buf = decoder->buf + rest_psb;
2256 decoder->next_len = decoder->len - rest_psb;
2257 memcpy(decoder->temp_buf, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2258 decoder->buf = decoder->temp_buf;
2259 decoder->len = INTEL_PT_PSB_LEN;
2264 static int intel_pt_scan_for_psb(struct intel_pt_decoder *decoder)
2266 unsigned char *next;
2269 intel_pt_log("Scanning for PSB\n");
2271 if (!decoder->len) {
2272 ret = intel_pt_get_next_data(decoder);
2277 next = memmem(decoder->buf, decoder->len, INTEL_PT_PSB_STR,
2282 part_psb = intel_pt_part_psb(decoder);
2284 ret = intel_pt_get_split_psb(decoder, part_psb);
2288 decoder->pos += decoder->len;
2294 decoder->pkt_step = next - decoder->buf;
2295 return intel_pt_get_next_packet(decoder);
2299 static int intel_pt_sync(struct intel_pt_decoder *decoder)
2303 decoder->pge = false;
2304 decoder->continuous_period = false;
2305 decoder->have_last_ip = false;
2306 decoder->last_ip = 0;
2308 intel_pt_clear_stack(&decoder->stack);
2310 err = intel_pt_scan_for_psb(decoder);
2314 decoder->have_last_ip = true;
2315 decoder->pkt_state = INTEL_PT_STATE_NO_IP;
2317 err = intel_pt_walk_psb(decoder);
2322 decoder->state.type = 0; /* Do not have a sample */
2323 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2325 return intel_pt_sync_ip(decoder);
2331 static uint64_t intel_pt_est_timestamp(struct intel_pt_decoder *decoder)
2333 uint64_t est = decoder->sample_insn_cnt << 1;
2335 if (!decoder->cbr || !decoder->max_non_turbo_ratio)
2338 est *= decoder->max_non_turbo_ratio;
2339 est /= decoder->cbr;
2341 return decoder->sample_timestamp + est;
2344 const struct intel_pt_state *intel_pt_decode(struct intel_pt_decoder *decoder)
2349 decoder->state.type = INTEL_PT_BRANCH;
2350 decoder->state.flags = 0;
2352 switch (decoder->pkt_state) {
2353 case INTEL_PT_STATE_NO_PSB:
2354 err = intel_pt_sync(decoder);
2356 case INTEL_PT_STATE_NO_IP:
2357 decoder->have_last_ip = false;
2358 decoder->last_ip = 0;
2361 case INTEL_PT_STATE_ERR_RESYNC:
2362 err = intel_pt_sync_ip(decoder);
2364 case INTEL_PT_STATE_IN_SYNC:
2365 err = intel_pt_walk_trace(decoder);
2367 case INTEL_PT_STATE_TNT:
2368 err = intel_pt_walk_tnt(decoder);
2370 err = intel_pt_walk_trace(decoder);
2372 case INTEL_PT_STATE_TIP:
2373 case INTEL_PT_STATE_TIP_PGD:
2374 err = intel_pt_walk_tip(decoder);
2376 case INTEL_PT_STATE_FUP:
2377 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2378 err = intel_pt_walk_fup(decoder);
2380 err = intel_pt_walk_fup_tip(decoder);
2382 decoder->pkt_state = INTEL_PT_STATE_FUP;
2384 case INTEL_PT_STATE_FUP_NO_TIP:
2385 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2386 err = intel_pt_walk_fup(decoder);
2388 err = intel_pt_walk_trace(decoder);
2391 err = intel_pt_bug(decoder);
2394 } while (err == -ENOLINK);
2397 decoder->state.err = intel_pt_ext_err(err);
2398 decoder->state.from_ip = decoder->ip;
2399 decoder->sample_timestamp = decoder->timestamp;
2400 decoder->sample_insn_cnt = decoder->timestamp_insn_cnt;
2402 decoder->state.err = 0;
2403 if (decoder->cbr != decoder->cbr_seen && decoder->state.type) {
2404 decoder->cbr_seen = decoder->cbr;
2405 decoder->state.type |= INTEL_PT_CBR_CHG;
2406 decoder->state.cbr_payload = decoder->cbr_payload;
2408 if (intel_pt_sample_time(decoder->pkt_state)) {
2409 decoder->sample_timestamp = decoder->timestamp;
2410 decoder->sample_insn_cnt = decoder->timestamp_insn_cnt;
2414 decoder->state.timestamp = decoder->sample_timestamp;
2415 decoder->state.est_timestamp = intel_pt_est_timestamp(decoder);
2416 decoder->state.cr3 = decoder->cr3;
2417 decoder->state.tot_insn_cnt = decoder->tot_insn_cnt;
2419 return &decoder->state;
2423 * intel_pt_next_psb - move buffer pointer to the start of the next PSB packet.
2424 * @buf: pointer to buffer pointer
2425 * @len: size of buffer
2427 * Updates the buffer pointer to point to the start of the next PSB packet if
2428 * there is one, otherwise the buffer pointer is unchanged. If @buf is updated,
2429 * @len is adjusted accordingly.
2431 * Return: %true if a PSB packet is found, %false otherwise.
2433 static bool intel_pt_next_psb(unsigned char **buf, size_t *len)
2435 unsigned char *next;
2437 next = memmem(*buf, *len, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2439 *len -= next - *buf;
2447 * intel_pt_step_psb - move buffer pointer to the start of the following PSB
2449 * @buf: pointer to buffer pointer
2450 * @len: size of buffer
2452 * Updates the buffer pointer to point to the start of the following PSB packet
2453 * (skipping the PSB at @buf itself) if there is one, otherwise the buffer
2454 * pointer is unchanged. If @buf is updated, @len is adjusted accordingly.
2456 * Return: %true if a PSB packet is found, %false otherwise.
2458 static bool intel_pt_step_psb(unsigned char **buf, size_t *len)
2460 unsigned char *next;
2465 next = memmem(*buf + 1, *len - 1, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2467 *len -= next - *buf;
2475 * intel_pt_last_psb - find the last PSB packet in a buffer.
2477 * @len: size of buffer
2479 * This function finds the last PSB in a buffer.
2481 * Return: A pointer to the last PSB in @buf if found, %NULL otherwise.
2483 static unsigned char *intel_pt_last_psb(unsigned char *buf, size_t len)
2485 const char *n = INTEL_PT_PSB_STR;
2489 if (len < INTEL_PT_PSB_LEN)
2492 k = len - INTEL_PT_PSB_LEN + 1;
2494 p = memrchr(buf, n[0], k);
2497 if (!memcmp(p + 1, n + 1, INTEL_PT_PSB_LEN - 1))
2506 * intel_pt_next_tsc - find and return next TSC.
2508 * @len: size of buffer
2509 * @tsc: TSC value returned
2510 * @rem: returns remaining size when TSC is found
2512 * Find a TSC packet in @buf and return the TSC value. This function assumes
2513 * that @buf starts at a PSB and that PSB+ will contain TSC and so stops if a
2514 * PSBEND packet is found.
2516 * Return: %true if TSC is found, false otherwise.
2518 static bool intel_pt_next_tsc(unsigned char *buf, size_t len, uint64_t *tsc,
2521 struct intel_pt_pkt packet;
2525 ret = intel_pt_get_packet(buf, len, &packet);
2528 if (packet.type == INTEL_PT_TSC) {
2529 *tsc = packet.payload;
2533 if (packet.type == INTEL_PT_PSBEND)
2542 * intel_pt_tsc_cmp - compare 7-byte TSCs.
2543 * @tsc1: first TSC to compare
2544 * @tsc2: second TSC to compare
2546 * This function compares 7-byte TSC values allowing for the possibility that
2547 * TSC wrapped around. Generally it is not possible to know if TSC has wrapped
2548 * around so for that purpose this function assumes the absolute difference is
2549 * less than half the maximum difference.
2551 * Return: %-1 if @tsc1 is before @tsc2, %0 if @tsc1 == @tsc2, %1 if @tsc1 is
2554 static int intel_pt_tsc_cmp(uint64_t tsc1, uint64_t tsc2)
2556 const uint64_t halfway = (1ULL << 55);
2562 if (tsc2 - tsc1 < halfway)
2567 if (tsc1 - tsc2 < halfway)
2574 #define MAX_PADDING (PERF_AUXTRACE_RECORD_ALIGNMENT - 1)
2577 * adj_for_padding - adjust overlap to account for padding.
2578 * @buf_b: second buffer
2579 * @buf_a: first buffer
2580 * @len_a: size of first buffer
2582 * @buf_a might have up to 7 bytes of padding appended. Adjust the overlap
2585 * Return: A pointer into @buf_b from where non-overlapped data starts
2587 static unsigned char *adj_for_padding(unsigned char *buf_b,
2588 unsigned char *buf_a, size_t len_a)
2590 unsigned char *p = buf_b - MAX_PADDING;
2591 unsigned char *q = buf_a + len_a - MAX_PADDING;
2594 for (i = MAX_PADDING; i; i--, p++, q++) {
2603 * intel_pt_find_overlap_tsc - determine start of non-overlapped trace data
2605 * @buf_a: first buffer
2606 * @len_a: size of first buffer
2607 * @buf_b: second buffer
2608 * @len_b: size of second buffer
2609 * @consecutive: returns true if there is data in buf_b that is consecutive
2612 * If the trace contains TSC we can look at the last TSC of @buf_a and the
2613 * first TSC of @buf_b in order to determine if the buffers overlap, and then
2614 * walk forward in @buf_b until a later TSC is found. A precondition is that
2615 * @buf_a and @buf_b are positioned at a PSB.
2617 * Return: A pointer into @buf_b from where non-overlapped data starts, or
2618 * @buf_b + @len_b if there is no non-overlapped data.
2620 static unsigned char *intel_pt_find_overlap_tsc(unsigned char *buf_a,
2622 unsigned char *buf_b,
2623 size_t len_b, bool *consecutive)
2625 uint64_t tsc_a, tsc_b;
2627 size_t len, rem_a, rem_b;
2629 p = intel_pt_last_psb(buf_a, len_a);
2631 return buf_b; /* No PSB in buf_a => no overlap */
2633 len = len_a - (p - buf_a);
2634 if (!intel_pt_next_tsc(p, len, &tsc_a, &rem_a)) {
2635 /* The last PSB+ in buf_a is incomplete, so go back one more */
2637 p = intel_pt_last_psb(buf_a, len_a);
2639 return buf_b; /* No full PSB+ => assume no overlap */
2640 len = len_a - (p - buf_a);
2641 if (!intel_pt_next_tsc(p, len, &tsc_a, &rem_a))
2642 return buf_b; /* No TSC in buf_a => assume no overlap */
2646 /* Ignore PSB+ with no TSC */
2647 if (intel_pt_next_tsc(buf_b, len_b, &tsc_b, &rem_b)) {
2648 int cmp = intel_pt_tsc_cmp(tsc_a, tsc_b);
2650 /* Same TSC, so buffers are consecutive */
2651 if (!cmp && rem_b >= rem_a) {
2652 unsigned char *start;
2654 *consecutive = true;
2655 start = buf_b + len_b - (rem_b - rem_a);
2656 return adj_for_padding(start, buf_a, len_a);
2659 return buf_b; /* tsc_a < tsc_b => no overlap */
2662 if (!intel_pt_step_psb(&buf_b, &len_b))
2663 return buf_b + len_b; /* No PSB in buf_b => no data */
2668 * intel_pt_find_overlap - determine start of non-overlapped trace data.
2669 * @buf_a: first buffer
2670 * @len_a: size of first buffer
2671 * @buf_b: second buffer
2672 * @len_b: size of second buffer
2673 * @have_tsc: can use TSC packets to detect overlap
2674 * @consecutive: returns true if there is data in buf_b that is consecutive
2677 * When trace samples or snapshots are recorded there is the possibility that
2678 * the data overlaps. Note that, for the purposes of decoding, data is only
2679 * useful if it begins with a PSB packet.
2681 * Return: A pointer into @buf_b from where non-overlapped data starts, or
2682 * @buf_b + @len_b if there is no non-overlapped data.
2684 unsigned char *intel_pt_find_overlap(unsigned char *buf_a, size_t len_a,
2685 unsigned char *buf_b, size_t len_b,
2686 bool have_tsc, bool *consecutive)
2688 unsigned char *found;
2690 /* Buffer 'b' must start at PSB so throw away everything before that */
2691 if (!intel_pt_next_psb(&buf_b, &len_b))
2692 return buf_b + len_b; /* No PSB */
2694 if (!intel_pt_next_psb(&buf_a, &len_a))
2695 return buf_b; /* No overlap */
2698 found = intel_pt_find_overlap_tsc(buf_a, len_a, buf_b, len_b,
2705 * Buffer 'b' cannot end within buffer 'a' so, for comparison purposes,
2706 * we can ignore the first part of buffer 'a'.
2708 while (len_b < len_a) {
2709 if (!intel_pt_step_psb(&buf_a, &len_a))
2710 return buf_b; /* No overlap */
2713 /* Now len_b >= len_a */
2715 /* Potential overlap so check the bytes */
2716 found = memmem(buf_a, len_a, buf_b, len_a);
2718 *consecutive = true;
2719 return adj_for_padding(buf_b + len_a, buf_a, len_a);
2722 /* Try again at next PSB in buffer 'a' */
2723 if (!intel_pt_step_psb(&buf_a, &len_a))
2724 return buf_b; /* No overlap */